xref: /titanic_52/usr/src/uts/common/io/igb/igb_main.c (revision c2785286330b58810071fc24f6ca69f401682dff)
1 /*
2  * CDDL HEADER START
3  *
4  * Copyright(c) 2007-2008 Intel Corporation. All rights reserved.
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at:
10  *	http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When using or redistributing this file, you may do so under the
15  * License only. No other modification of this header is permitted.
16  *
17  * If applicable, add the following below this CDDL HEADER, with the
18  * fields enclosed by brackets "[]" replaced with your own identifying
19  * information: Portions Copyright [yyyy] [name of copyright owner]
20  *
21  * CDDL HEADER END
22  */
23 
24 /*
25  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
26  * Use is subject to license terms of the CDDL.
27  */
28 
29 #pragma ident	"%Z%%M%	%I%	%E% SMI"
30 
31 #include "igb_sw.h"
32 
33 static char ident[] = "Intel 1Gb Ethernet 1.1.2";
34 
35 /*
36  * Local function protoypes
37  */
38 static int igb_register_mac(igb_t *);
39 static int igb_identify_hardware(igb_t *);
40 static int igb_regs_map(igb_t *);
41 static void igb_init_properties(igb_t *);
42 static int igb_init_driver_settings(igb_t *);
43 static void igb_init_locks(igb_t *);
44 static void igb_destroy_locks(igb_t *);
45 static int igb_init(igb_t *);
46 static int igb_chip_start(igb_t *);
47 static void igb_chip_stop(igb_t *);
48 static int igb_reset(igb_t *);
49 static void igb_tx_clean(igb_t *);
50 static boolean_t igb_tx_drain(igb_t *);
51 static boolean_t igb_rx_drain(igb_t *);
52 static int igb_alloc_rings(igb_t *);
53 static int igb_init_rings(igb_t *);
54 static void igb_free_rings(igb_t *);
55 static void igb_fini_rings(igb_t *);
56 static void igb_setup_rings(igb_t *);
57 static void igb_setup_rx(igb_t *);
58 static void igb_setup_tx(igb_t *);
59 static void igb_setup_rx_ring(igb_rx_ring_t *);
60 static void igb_setup_tx_ring(igb_tx_ring_t *);
61 static void igb_setup_rss(igb_t *);
62 static void igb_init_unicst(igb_t *);
63 static void igb_setup_multicst(igb_t *);
64 static void igb_get_phy_state(igb_t *);
65 static void igb_get_conf(igb_t *);
66 static int igb_get_prop(igb_t *, char *, int, int, int);
67 static boolean_t igb_is_link_up(igb_t *);
68 static boolean_t igb_link_check(igb_t *);
69 static void igb_local_timer(void *);
70 static void igb_arm_watchdog_timer(igb_t *);
71 static void igb_start_watchdog_timer(igb_t *);
72 static void igb_restart_watchdog_timer(igb_t *);
73 static void igb_stop_watchdog_timer(igb_t *);
74 static void igb_disable_adapter_interrupts(igb_t *);
75 static void igb_enable_adapter_interrupts(igb_t *);
76 static boolean_t is_valid_mac_addr(uint8_t *);
77 static boolean_t igb_stall_check(igb_t *);
78 static boolean_t igb_set_loopback_mode(igb_t *, uint32_t);
79 static void igb_set_external_loopback(igb_t *);
80 static void igb_set_internal_mac_loopback(igb_t *);
81 static void igb_set_internal_phy_loopback(igb_t *);
82 static void igb_set_internal_serdes_loopback(igb_t *);
83 static boolean_t igb_find_mac_address(igb_t *);
84 static int igb_alloc_intrs(igb_t *);
85 static int igb_alloc_intrs_msix(igb_t *);
86 static int igb_alloc_intrs_msi(igb_t *);
87 static int igb_alloc_intrs_legacy(igb_t *);
88 static int igb_add_intr_handlers(igb_t *);
89 static void igb_rem_intr_handlers(igb_t *);
90 static void igb_rem_intrs(igb_t *);
91 static int igb_enable_intrs(igb_t *);
92 static int igb_disable_intrs(igb_t *);
93 static void igb_setup_adapter_msix(igb_t *);
94 static uint_t igb_intr_legacy(void *, void *);
95 static uint_t igb_intr_msi(void *, void *);
96 static uint_t igb_intr_rx(void *, void *);
97 static uint_t igb_intr_tx_other(void *, void *);
98 static void igb_intr_rx_work(igb_rx_ring_t *);
99 static void igb_intr_tx_work(igb_tx_ring_t *);
100 static void igb_intr_other_work(igb_t *);
101 static void igb_get_driver_control(struct e1000_hw *);
102 static void igb_release_driver_control(struct e1000_hw *);
103 
104 static int igb_attach(dev_info_t *, ddi_attach_cmd_t);
105 static int igb_detach(dev_info_t *, ddi_detach_cmd_t);
106 static int igb_resume(dev_info_t *);
107 static int igb_suspend(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 };
149 
150 static struct modldrv igb_modldrv = {
151 	&mod_driverops,		/* Type of module.  This one is a driver */
152 	ident,			/* Discription string */
153 	&igb_dev_ops,		/* driver ops */
154 };
155 
156 static struct modlinkage igb_modlinkage = {
157 	MODREV_1, &igb_modldrv, NULL
158 };
159 
160 /* Access attributes for register mapping */
161 ddi_device_acc_attr_t igb_regs_acc_attr = {
162 	DDI_DEVICE_ATTR_V0,
163 	DDI_STRUCTURE_LE_ACC,
164 	DDI_STRICTORDER_ACC,
165 	DDI_FLAGERR_ACC
166 };
167 
168 #define	IGB_M_CALLBACK_FLAGS	(MC_IOCTL | MC_GETCAPAB)
169 
170 static mac_callbacks_t igb_m_callbacks = {
171 	IGB_M_CALLBACK_FLAGS,
172 	igb_m_stat,
173 	igb_m_start,
174 	igb_m_stop,
175 	igb_m_promisc,
176 	igb_m_multicst,
177 	igb_m_unicst,
178 	igb_m_tx,
179 	NULL,
180 	igb_m_ioctl,
181 	igb_m_getcapab
182 };
183 
184 
185 /*
186  * Module Initialization Functions
187  */
188 
189 int
190 _init(void)
191 {
192 	int status;
193 
194 	mac_init_ops(&igb_dev_ops, MODULE_NAME);
195 
196 	status = mod_install(&igb_modlinkage);
197 
198 	if (status != DDI_SUCCESS) {
199 		mac_fini_ops(&igb_dev_ops);
200 	}
201 
202 	return (status);
203 }
204 
205 int
206 _fini(void)
207 {
208 	int status;
209 
210 	status = mod_remove(&igb_modlinkage);
211 
212 	if (status == DDI_SUCCESS) {
213 		mac_fini_ops(&igb_dev_ops);
214 	}
215 
216 	return (status);
217 
218 }
219 
220 int
221 _info(struct modinfo *modinfop)
222 {
223 	int status;
224 
225 	status = mod_info(&igb_modlinkage, modinfop);
226 
227 	return (status);
228 }
229 
230 /*
231  * igb_attach - driver attach
232  *
233  * This function is the device specific initialization entry
234  * point. This entry point is required and must be written.
235  * The DDI_ATTACH command must be provided in the attach entry
236  * point. When attach() is called with cmd set to DDI_ATTACH,
237  * all normal kernel services (such as kmem_alloc(9F)) are
238  * available for use by the driver.
239  *
240  * The attach() function will be called once for each instance
241  * of  the  device  on  the  system with cmd set to DDI_ATTACH.
242  * Until attach() succeeds, the only driver entry points which
243  * may be called are open(9E) and getinfo(9E).
244  */
245 static int
246 igb_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
247 {
248 	igb_t *igb;
249 	struct igb_osdep *osdep;
250 	struct e1000_hw *hw;
251 	int instance;
252 
253 	/*
254 	 * Check the command and perform corresponding operations
255 	 */
256 	switch (cmd) {
257 	default:
258 		return (DDI_FAILURE);
259 
260 	case DDI_RESUME:
261 		return (igb_resume(devinfo));
262 
263 	case DDI_ATTACH:
264 		break;
265 	}
266 
267 	/* Get the device instance */
268 	instance = ddi_get_instance(devinfo);
269 
270 	/* Allocate memory for the instance data structure */
271 	igb = kmem_zalloc(sizeof (igb_t), KM_SLEEP);
272 
273 	igb->dip = devinfo;
274 	igb->instance = instance;
275 
276 	hw = &igb->hw;
277 	osdep = &igb->osdep;
278 	hw->back = osdep;
279 	osdep->igb = igb;
280 
281 	/* Attach the instance pointer to the dev_info data structure */
282 	ddi_set_driver_private(devinfo, igb);
283 
284 
285 	/* Initialize for fma support */
286 	igb->fm_capabilities = igb_get_prop(igb, "fm-capable",
287 	    0, 0x0f,
288 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
289 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
290 	igb_fm_init(igb);
291 	igb->attach_progress |= ATTACH_PROGRESS_FMINIT;
292 
293 	/*
294 	 * Map PCI config space registers
295 	 */
296 	if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
297 		igb_error(igb, "Failed to map PCI configurations");
298 		goto attach_fail;
299 	}
300 	igb->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
301 
302 	/*
303 	 * Identify the chipset family
304 	 */
305 	if (igb_identify_hardware(igb) != IGB_SUCCESS) {
306 		igb_error(igb, "Failed to identify hardware");
307 		goto attach_fail;
308 	}
309 
310 	/*
311 	 * Map device registers
312 	 */
313 	if (igb_regs_map(igb) != IGB_SUCCESS) {
314 		igb_error(igb, "Failed to map device registers");
315 		goto attach_fail;
316 	}
317 	igb->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
318 
319 	/*
320 	 * Initialize driver parameters
321 	 */
322 	igb_init_properties(igb);
323 	igb->attach_progress |= ATTACH_PROGRESS_PROPS;
324 
325 	/*
326 	 * Allocate interrupts
327 	 */
328 	if (igb_alloc_intrs(igb) != IGB_SUCCESS) {
329 		igb_error(igb, "Failed to allocate interrupts");
330 		goto attach_fail;
331 	}
332 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
333 
334 	/*
335 	 * Allocate rx/tx rings based on the ring numbers.
336 	 * The actual numbers of rx/tx rings are decided by the number of
337 	 * allocated interrupt vectors, so we should allocate the rings after
338 	 * interrupts are allocated.
339 	 */
340 	if (igb_alloc_rings(igb) != IGB_SUCCESS) {
341 		igb_error(igb, "Failed to allocate rx and tx rings");
342 		goto attach_fail;
343 	}
344 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_RINGS;
345 
346 	/*
347 	 * Add interrupt handlers
348 	 */
349 	if (igb_add_intr_handlers(igb) != IGB_SUCCESS) {
350 		igb_error(igb, "Failed to add interrupt handlers");
351 		goto attach_fail;
352 	}
353 	igb->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
354 
355 	/*
356 	 * Initialize driver parameters
357 	 */
358 	if (igb_init_driver_settings(igb) != IGB_SUCCESS) {
359 		igb_error(igb, "Failed to initialize driver settings");
360 		goto attach_fail;
361 	}
362 
363 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK) {
364 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
365 		goto attach_fail;
366 	}
367 
368 	/*
369 	 * Initialize mutexes for this device.
370 	 * Do this before enabling the interrupt handler and
371 	 * register the softint to avoid the condition where
372 	 * interrupt handler can try using uninitialized mutex
373 	 */
374 	igb_init_locks(igb);
375 	igb->attach_progress |= ATTACH_PROGRESS_LOCKS;
376 
377 	/*
378 	 * Initialize chipset hardware
379 	 */
380 	if (igb_init(igb) != IGB_SUCCESS) {
381 		igb_error(igb, "Failed to initialize adapter");
382 		goto attach_fail;
383 	}
384 	igb->attach_progress |= ATTACH_PROGRESS_INIT;
385 
386 	/*
387 	 * Initialize DMA and hardware settings for rx/tx rings
388 	 */
389 	if (igb_init_rings(igb) != IGB_SUCCESS) {
390 		igb_error(igb, "Failed to initialize rings");
391 		goto attach_fail;
392 	}
393 	igb->attach_progress |= ATTACH_PROGRESS_INIT_RINGS;
394 
395 	/*
396 	 * Initialize statistics
397 	 */
398 	if (igb_init_stats(igb) != IGB_SUCCESS) {
399 		igb_error(igb, "Failed to initialize statistics");
400 		goto attach_fail;
401 	}
402 	igb->attach_progress |= ATTACH_PROGRESS_STATS;
403 
404 	/*
405 	 * Initialize NDD parameters
406 	 */
407 	if (igb_nd_init(igb) != IGB_SUCCESS) {
408 		igb_error(igb, "Failed to initialize ndd");
409 		goto attach_fail;
410 	}
411 	igb->attach_progress |= ATTACH_PROGRESS_NDD;
412 
413 	/*
414 	 * Register the driver to the MAC
415 	 */
416 	if (igb_register_mac(igb) != IGB_SUCCESS) {
417 		igb_error(igb, "Failed to register MAC");
418 		goto attach_fail;
419 	}
420 	igb->attach_progress |= ATTACH_PROGRESS_MAC;
421 
422 	/*
423 	 * Now that mutex locks are initialized, and the chip is also
424 	 * initialized, enable interrupts.
425 	 */
426 	if (igb_enable_intrs(igb) != IGB_SUCCESS) {
427 		igb_error(igb, "Failed to enable DDI interrupts");
428 		goto attach_fail;
429 	}
430 	igb->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
431 
432 	igb->igb_state |= IGB_INITIALIZED;
433 
434 	return (DDI_SUCCESS);
435 
436 attach_fail:
437 	igb_unconfigure(devinfo, igb);
438 	return (DDI_FAILURE);
439 }
440 
441 /*
442  * igb_detach - driver detach
443  *
444  * The detach() function is the complement of the attach routine.
445  * If cmd is set to DDI_DETACH, detach() is used to remove  the
446  * state  associated  with  a  given  instance of a device node
447  * prior to the removal of that instance from the system.
448  *
449  * The detach() function will be called once for each  instance
450  * of the device for which there has been a successful attach()
451  * once there are no longer  any  opens  on  the  device.
452  *
453  * Interrupts routine are disabled, All memory allocated by this
454  * driver are freed.
455  */
456 static int
457 igb_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
458 {
459 	igb_t *igb;
460 
461 	/*
462 	 * Check detach command
463 	 */
464 	switch (cmd) {
465 	default:
466 		return (DDI_FAILURE);
467 
468 	case DDI_SUSPEND:
469 		return (igb_suspend(devinfo));
470 
471 	case DDI_DETACH:
472 		break;
473 	}
474 
475 
476 	/*
477 	 * Get the pointer to the driver private data structure
478 	 */
479 	igb = (igb_t *)ddi_get_driver_private(devinfo);
480 	if (igb == NULL)
481 		return (DDI_FAILURE);
482 
483 	/*
484 	 * Unregister MAC. If failed, we have to fail the detach
485 	 */
486 	if (mac_unregister(igb->mac_hdl) != 0) {
487 		igb_error(igb, "Failed to unregister MAC");
488 		return (DDI_FAILURE);
489 	}
490 	igb->attach_progress &= ~ATTACH_PROGRESS_MAC;
491 
492 	/*
493 	 * If the device is still running, it needs to be stopped first.
494 	 * This check is necessary because under some specific circumstances,
495 	 * the detach routine can be called without stopping the interface
496 	 * first.
497 	 */
498 	mutex_enter(&igb->gen_lock);
499 	if (igb->igb_state & IGB_STARTED) {
500 		igb->igb_state &= ~IGB_STARTED;
501 		igb_stop(igb);
502 		mutex_exit(&igb->gen_lock);
503 		/* Disable and stop the watchdog timer */
504 		igb_disable_watchdog_timer(igb);
505 	} else
506 		mutex_exit(&igb->gen_lock);
507 
508 	/*
509 	 * Check if there are still rx buffers held by the upper layer.
510 	 * If so, fail the detach.
511 	 */
512 	if (!igb_rx_drain(igb))
513 		return (DDI_FAILURE);
514 
515 	/*
516 	 * Do the remaining unconfigure routines
517 	 */
518 	igb_unconfigure(devinfo, igb);
519 
520 	return (DDI_SUCCESS);
521 }
522 
523 static void
524 igb_unconfigure(dev_info_t *devinfo, igb_t *igb)
525 {
526 	/*
527 	 * Disable interrupt
528 	 */
529 	if (igb->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
530 		(void) igb_disable_intrs(igb);
531 	}
532 
533 	/*
534 	 * Unregister MAC
535 	 */
536 	if (igb->attach_progress & ATTACH_PROGRESS_MAC) {
537 		(void) mac_unregister(igb->mac_hdl);
538 	}
539 
540 	/*
541 	 * Free ndd parameters
542 	 */
543 	if (igb->attach_progress & ATTACH_PROGRESS_NDD) {
544 		igb_nd_cleanup(igb);
545 	}
546 
547 	/*
548 	 * Free statistics
549 	 */
550 	if (igb->attach_progress & ATTACH_PROGRESS_STATS) {
551 		kstat_delete((kstat_t *)igb->igb_ks);
552 	}
553 
554 	/*
555 	 * Remove interrupt handlers
556 	 */
557 	if (igb->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
558 		igb_rem_intr_handlers(igb);
559 	}
560 
561 	/*
562 	 * Remove interrupts
563 	 */
564 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_INTR) {
565 		igb_rem_intrs(igb);
566 	}
567 
568 	/*
569 	 * Remove driver properties
570 	 */
571 	if (igb->attach_progress & ATTACH_PROGRESS_PROPS) {
572 		(void) ddi_prop_remove_all(devinfo);
573 	}
574 
575 	/*
576 	 * Release the DMA resources of rx/tx rings
577 	 */
578 	if (igb->attach_progress & ATTACH_PROGRESS_INIT_RINGS) {
579 		igb_fini_rings(igb);
580 	}
581 
582 	/*
583 	 * Stop the chipset
584 	 */
585 	if (igb->attach_progress & ATTACH_PROGRESS_INIT) {
586 		mutex_enter(&igb->gen_lock);
587 		igb_chip_stop(igb);
588 		mutex_exit(&igb->gen_lock);
589 		if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
590 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_UNAFFECTED);
591 	}
592 
593 	/*
594 	 * Free register handle
595 	 */
596 	if (igb->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
597 		if (igb->osdep.reg_handle != NULL)
598 			ddi_regs_map_free(&igb->osdep.reg_handle);
599 	}
600 
601 	/*
602 	 * Free PCI config handle
603 	 */
604 	if (igb->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
605 		if (igb->osdep.cfg_handle != NULL)
606 			pci_config_teardown(&igb->osdep.cfg_handle);
607 	}
608 
609 	/*
610 	 * Free locks
611 	 */
612 	if (igb->attach_progress & ATTACH_PROGRESS_LOCKS) {
613 		igb_destroy_locks(igb);
614 	}
615 
616 	/*
617 	 * Free the rx/tx rings
618 	 */
619 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
620 		igb_free_rings(igb);
621 	}
622 
623 	/*
624 	 * Remove FMA
625 	 */
626 	if (igb->attach_progress & ATTACH_PROGRESS_FMINIT) {
627 		igb_fm_fini(igb);
628 	}
629 
630 	/*
631 	 * Free device specific structure
632 	 */
633 	e1000_remove_device(&igb->hw);
634 
635 	/*
636 	 * Free the driver data structure
637 	 */
638 	kmem_free(igb, sizeof (igb_t));
639 
640 	ddi_set_driver_private(devinfo, NULL);
641 }
642 
643 /*
644  * igb_register_mac - Register the driver and its function pointers with
645  * the GLD interface
646  */
647 static int
648 igb_register_mac(igb_t *igb)
649 {
650 	struct e1000_hw *hw = &igb->hw;
651 	mac_register_t *mac;
652 	int status;
653 
654 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
655 		return (IGB_FAILURE);
656 
657 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
658 	mac->m_driver = igb;
659 	mac->m_dip = igb->dip;
660 	mac->m_src_addr = hw->mac.addr;
661 	mac->m_callbacks = &igb_m_callbacks;
662 	mac->m_min_sdu = 0;
663 	mac->m_max_sdu = igb->max_frame_size -
664 	    sizeof (struct ether_vlan_header) - ETHERFCSL;
665 	mac->m_margin = VLAN_TAGSZ;
666 
667 	status = mac_register(mac, &igb->mac_hdl);
668 
669 	mac_free(mac);
670 
671 	return ((status == 0) ? IGB_SUCCESS : IGB_FAILURE);
672 }
673 
674 /*
675  * igb_identify_hardware - Identify the type of the chipset
676  */
677 static int
678 igb_identify_hardware(igb_t *igb)
679 {
680 	struct e1000_hw *hw = &igb->hw;
681 	struct igb_osdep *osdep = &igb->osdep;
682 
683 	/*
684 	 * Get the device id
685 	 */
686 	hw->vendor_id =
687 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
688 	hw->device_id =
689 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
690 	hw->revision_id =
691 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
692 	hw->subsystem_device_id =
693 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
694 	hw->subsystem_vendor_id =
695 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
696 
697 	/*
698 	 * Set the mac type of the adapter based on the device id
699 	 */
700 	if (e1000_set_mac_type(hw) != E1000_SUCCESS) {
701 		return (IGB_FAILURE);
702 	}
703 
704 	return (IGB_SUCCESS);
705 }
706 
707 /*
708  * igb_regs_map - Map the device registers
709  */
710 static int
711 igb_regs_map(igb_t *igb)
712 {
713 	dev_info_t *devinfo = igb->dip;
714 	struct e1000_hw *hw = &igb->hw;
715 	struct igb_osdep *osdep = &igb->osdep;
716 	off_t mem_size;
717 
718 	/*
719 	 * First get the size of device registers to be mapped.
720 	 */
721 	if (ddi_dev_regsize(devinfo, 1, &mem_size) != DDI_SUCCESS) {
722 		return (IGB_FAILURE);
723 	}
724 
725 	/*
726 	 * Call ddi_regs_map_setup() to map registers
727 	 */
728 	if ((ddi_regs_map_setup(devinfo, 1,
729 	    (caddr_t *)&hw->hw_addr, 0,
730 	    mem_size, &igb_regs_acc_attr,
731 	    &osdep->reg_handle)) != DDI_SUCCESS) {
732 		return (IGB_FAILURE);
733 	}
734 
735 	return (IGB_SUCCESS);
736 }
737 
738 /*
739  * igb_init_properties - Initialize driver properties
740  */
741 static void
742 igb_init_properties(igb_t *igb)
743 {
744 	/*
745 	 * Get conf file properties, including link settings
746 	 * jumbo frames, ring number, descriptor number, etc.
747 	 */
748 	igb_get_conf(igb);
749 }
750 
751 /*
752  * igb_init_driver_settings - Initialize driver settings
753  *
754  * The settings include hardware function pointers, bus information,
755  * rx/tx rings settings, link state, and any other parameters that
756  * need to be setup during driver initialization.
757  */
758 static int
759 igb_init_driver_settings(igb_t *igb)
760 {
761 	struct e1000_hw *hw = &igb->hw;
762 	igb_rx_ring_t *rx_ring;
763 	igb_tx_ring_t *tx_ring;
764 	uint32_t rx_size;
765 	uint32_t tx_size;
766 	int i;
767 
768 	/*
769 	 * Initialize chipset specific hardware function pointers
770 	 */
771 	if (e1000_setup_init_funcs(hw, B_TRUE) != E1000_SUCCESS) {
772 		return (IGB_FAILURE);
773 	}
774 
775 	/*
776 	 * Get bus information
777 	 */
778 	if (e1000_get_bus_info(hw) != E1000_SUCCESS) {
779 		return (IGB_FAILURE);
780 	}
781 
782 	/*
783 	 * Set rx buffer size
784 	 * The IP header alignment room is counted in the calculation.
785 	 * The rx buffer size is in unit of 1K that is required by the
786 	 * chipset hardware.
787 	 */
788 	rx_size = igb->max_frame_size + IPHDR_ALIGN_ROOM;
789 	igb->rx_buf_size = ((rx_size >> 10) +
790 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
791 
792 	/*
793 	 * Set tx buffer size
794 	 */
795 	tx_size = igb->max_frame_size;
796 	igb->tx_buf_size = ((tx_size >> 10) +
797 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
798 
799 	/*
800 	 * Initialize rx/tx rings parameters
801 	 */
802 	for (i = 0; i < igb->num_rx_rings; i++) {
803 		rx_ring = &igb->rx_rings[i];
804 		rx_ring->index = i;
805 		rx_ring->igb = igb;
806 
807 		rx_ring->ring_size = igb->rx_ring_size;
808 		rx_ring->free_list_size = igb->rx_ring_size;
809 		rx_ring->copy_thresh = igb->rx_copy_thresh;
810 		rx_ring->limit_per_intr = igb->rx_limit_per_intr;
811 	}
812 
813 	for (i = 0; i < igb->num_tx_rings; i++) {
814 		tx_ring = &igb->tx_rings[i];
815 		tx_ring->index = i;
816 		tx_ring->igb = igb;
817 		if (igb->tx_head_wb_enable)
818 			tx_ring->tx_recycle = igb_tx_recycle_head_wb;
819 		else
820 			tx_ring->tx_recycle = igb_tx_recycle_legacy;
821 
822 		tx_ring->ring_size = igb->tx_ring_size;
823 		tx_ring->free_list_size = igb->tx_ring_size +
824 		    (igb->tx_ring_size >> 1);
825 		tx_ring->copy_thresh = igb->tx_copy_thresh;
826 		tx_ring->recycle_thresh = igb->tx_recycle_thresh;
827 		tx_ring->overload_thresh = igb->tx_overload_thresh;
828 		tx_ring->resched_thresh = igb->tx_resched_thresh;
829 	}
830 
831 	/*
832 	 * Initialize values of interrupt throttling rate
833 	 */
834 	for (i = 1; i < MAX_NUM_EITR; i++)
835 		igb->intr_throttling[i] = igb->intr_throttling[0];
836 
837 	/*
838 	 * The initial link state should be "unknown"
839 	 */
840 	igb->link_state = LINK_STATE_UNKNOWN;
841 
842 	return (IGB_SUCCESS);
843 }
844 
845 /*
846  * igb_init_locks - Initialize locks
847  */
848 static void
849 igb_init_locks(igb_t *igb)
850 {
851 	igb_rx_ring_t *rx_ring;
852 	igb_tx_ring_t *tx_ring;
853 	int i;
854 
855 	for (i = 0; i < igb->num_rx_rings; i++) {
856 		rx_ring = &igb->rx_rings[i];
857 		mutex_init(&rx_ring->rx_lock, NULL,
858 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
859 		mutex_init(&rx_ring->recycle_lock, NULL,
860 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
861 	}
862 
863 	for (i = 0; i < igb->num_tx_rings; i++) {
864 		tx_ring = &igb->tx_rings[i];
865 		mutex_init(&tx_ring->tx_lock, NULL,
866 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
867 		mutex_init(&tx_ring->recycle_lock, NULL,
868 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
869 		mutex_init(&tx_ring->tcb_head_lock, NULL,
870 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
871 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
872 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
873 	}
874 
875 	mutex_init(&igb->gen_lock, NULL,
876 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
877 
878 	mutex_init(&igb->watchdog_lock, NULL,
879 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
880 }
881 
882 /*
883  * igb_destroy_locks - Destroy locks
884  */
885 static void
886 igb_destroy_locks(igb_t *igb)
887 {
888 	igb_rx_ring_t *rx_ring;
889 	igb_tx_ring_t *tx_ring;
890 	int i;
891 
892 	for (i = 0; i < igb->num_rx_rings; i++) {
893 		rx_ring = &igb->rx_rings[i];
894 		mutex_destroy(&rx_ring->rx_lock);
895 		mutex_destroy(&rx_ring->recycle_lock);
896 	}
897 
898 	for (i = 0; i < igb->num_tx_rings; i++) {
899 		tx_ring = &igb->tx_rings[i];
900 		mutex_destroy(&tx_ring->tx_lock);
901 		mutex_destroy(&tx_ring->recycle_lock);
902 		mutex_destroy(&tx_ring->tcb_head_lock);
903 		mutex_destroy(&tx_ring->tcb_tail_lock);
904 	}
905 
906 	mutex_destroy(&igb->gen_lock);
907 	mutex_destroy(&igb->watchdog_lock);
908 }
909 
910 static int
911 igb_resume(dev_info_t *devinfo)
912 {
913 	igb_t *igb;
914 
915 	igb = (igb_t *)ddi_get_driver_private(devinfo);
916 	if (igb == NULL)
917 		return (DDI_FAILURE);
918 
919 	mutex_enter(&igb->gen_lock);
920 
921 	if (igb->igb_state & IGB_STARTED) {
922 		if (igb_start(igb) != IGB_SUCCESS) {
923 			mutex_exit(&igb->gen_lock);
924 			return (DDI_FAILURE);
925 		}
926 
927 		/*
928 		 * Enable and start the watchdog timer
929 		 */
930 		igb_enable_watchdog_timer(igb);
931 	}
932 
933 	igb->igb_state &= ~IGB_SUSPENDED;
934 
935 	mutex_exit(&igb->gen_lock);
936 
937 	return (DDI_SUCCESS);
938 }
939 
940 static int
941 igb_suspend(dev_info_t *devinfo)
942 {
943 	igb_t *igb;
944 
945 	igb = (igb_t *)ddi_get_driver_private(devinfo);
946 	if (igb == NULL)
947 		return (DDI_FAILURE);
948 
949 	mutex_enter(&igb->gen_lock);
950 
951 	igb->igb_state |= IGB_SUSPENDED;
952 
953 	igb_stop(igb);
954 
955 	mutex_exit(&igb->gen_lock);
956 
957 	/*
958 	 * Disable and stop the watchdog timer
959 	 */
960 	igb_disable_watchdog_timer(igb);
961 
962 	return (DDI_SUCCESS);
963 }
964 
965 /*
966  * igb_init - Initialize the device
967  */
968 static int
969 igb_init(igb_t *igb)
970 {
971 	struct e1000_hw *hw = &igb->hw;
972 	uint32_t pba;
973 	uint32_t high_water;
974 
975 	mutex_enter(&igb->gen_lock);
976 
977 	/*
978 	 * Reset chipset to put the hardware in a known state
979 	 * before we try to do anything with the eeprom
980 	 */
981 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
982 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
983 		goto init_fail;
984 	}
985 
986 	/*
987 	 * NVM validation
988 	 */
989 	if (e1000_validate_nvm_checksum(hw) < 0) {
990 		/*
991 		 * Some PCI-E parts fail the first check due to
992 		 * the link being in sleep state.  Call it again,
993 		 * if it fails a second time its a real issue.
994 		 */
995 		if (e1000_validate_nvm_checksum(hw) < 0) {
996 			igb_error(igb,
997 			    "Invalid NVM checksum. Please contact "
998 			    "the vendor to update the NVM.");
999 			igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1000 			goto init_fail;
1001 		}
1002 	}
1003 
1004 	/*
1005 	 * Set the FIFO size
1006 	 */
1007 	pba = E1000_PBA_32K;	/* 32K for Rx, 16K for Tx */
1008 	E1000_WRITE_REG(hw, E1000_PBA, pba);
1009 
1010 	/*
1011 	 * Setup flow control
1012 	 *
1013 	 * These parameters set thresholds for the adapter's generation(Tx)
1014 	 * and response(Rx) to Ethernet PAUSE frames.  These are just threshold
1015 	 * settings.  Flow control is enabled or disabled in the configuration
1016 	 * file.
1017 	 * High-water mark is set down from the top of the rx fifo (not
1018 	 * sensitive to max_frame_size) and low-water is set just below
1019 	 * high-water mark.
1020 	 * The high water mark must be low enough to fit one full frame above
1021 	 * it in the rx FIFO.  Should be the lower of:
1022 	 * 90% of the Rx FIFO size, or the full Rx FIFO size minus one full
1023 	 * frame.
1024 	 */
1025 	high_water = min(((pba << 10) * 9 / 10),
1026 	    ((pba << 10) - igb->max_frame_size));
1027 
1028 	hw->fc.high_water = high_water & 0xFFF8;
1029 	hw->fc.low_water = hw->fc.high_water - 8;
1030 	hw->fc.pause_time = E1000_FC_PAUSE_TIME;
1031 	hw->fc.send_xon = B_TRUE;
1032 
1033 	/*
1034 	 * Reset the chipset hardware the second time to validate
1035 	 * the PBA setting.
1036 	 */
1037 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1038 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1039 		goto init_fail;
1040 	}
1041 
1042 	/*
1043 	 * Don't wait for auto-negotiation to complete
1044 	 */
1045 	hw->phy.autoneg_wait_to_complete = B_FALSE;
1046 
1047 	/*
1048 	 * Copper options
1049 	 */
1050 	if (hw->phy.media_type == e1000_media_type_copper) {
1051 		hw->phy.mdix = 0;	/* AUTO_ALL_MODES */
1052 		hw->phy.disable_polarity_correction = B_FALSE;
1053 		hw->phy.ms_type = e1000_ms_hw_default; /* E1000_MASTER_SLAVE */
1054 	}
1055 
1056 	/*
1057 	 * Initialize link settings
1058 	 */
1059 	(void) igb_setup_link(igb, B_FALSE);
1060 
1061 	/*
1062 	 * Initialize the chipset hardware
1063 	 */
1064 	if (igb_chip_start(igb) != IGB_SUCCESS) {
1065 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1066 		goto init_fail;
1067 	}
1068 
1069 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK) {
1070 		goto init_fail;
1071 	}
1072 
1073 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
1074 		goto init_fail;
1075 	}
1076 
1077 	mutex_exit(&igb->gen_lock);
1078 	return (IGB_SUCCESS);
1079 
1080 init_fail:
1081 	/*
1082 	 * Reset PHY if possible
1083 	 */
1084 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
1085 		(void) e1000_phy_hw_reset(hw);
1086 
1087 	mutex_exit(&igb->gen_lock);
1088 
1089 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1090 
1091 	return (IGB_FAILURE);
1092 }
1093 
1094 /*
1095  * igb_init_rings - Allocate DMA resources for all rx/tx rings and
1096  * initialize relevant hardware settings.
1097  */
1098 static int
1099 igb_init_rings(igb_t *igb)
1100 {
1101 	int i;
1102 
1103 	/*
1104 	 * Allocate buffers for all the rx/tx rings
1105 	 */
1106 	if (igb_alloc_dma(igb) != IGB_SUCCESS)
1107 		return (IGB_FAILURE);
1108 
1109 	/*
1110 	 * Setup the rx/tx rings
1111 	 */
1112 	mutex_enter(&igb->gen_lock);
1113 
1114 	for (i = 0; i < igb->num_rx_rings; i++)
1115 		mutex_enter(&igb->rx_rings[i].rx_lock);
1116 	for (i = 0; i < igb->num_tx_rings; i++)
1117 		mutex_enter(&igb->tx_rings[i].tx_lock);
1118 
1119 	igb_setup_rings(igb);
1120 
1121 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1122 		mutex_exit(&igb->tx_rings[i].tx_lock);
1123 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1124 		mutex_exit(&igb->rx_rings[i].rx_lock);
1125 
1126 	mutex_exit(&igb->gen_lock);
1127 
1128 	return (IGB_SUCCESS);
1129 }
1130 
1131 /*
1132  * igb_fini_rings - Release DMA resources of all rx/tx rings
1133  */
1134 static void
1135 igb_fini_rings(igb_t *igb)
1136 {
1137 	/*
1138 	 * Release the DMA/memory resources of rx/tx rings
1139 	 */
1140 	igb_free_dma(igb);
1141 }
1142 
1143 /*
1144  * igb_chip_start - Initialize and start the chipset hardware
1145  */
1146 static int
1147 igb_chip_start(igb_t *igb)
1148 {
1149 	struct e1000_hw *hw = &igb->hw;
1150 	int i;
1151 
1152 	ASSERT(mutex_owned(&igb->gen_lock));
1153 
1154 	/*
1155 	 * Get the mac address
1156 	 * This function should handle SPARC case correctly.
1157 	 */
1158 	if (!igb_find_mac_address(igb)) {
1159 		igb_error(igb, "Failed to get the mac address");
1160 		return (IGB_FAILURE);
1161 	}
1162 
1163 	/* Validate mac address */
1164 	if (!is_valid_mac_addr(hw->mac.addr)) {
1165 		igb_error(igb, "Invalid mac address");
1166 		return (IGB_FAILURE);
1167 	}
1168 
1169 	/* Disable wakeup control by default */
1170 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1171 
1172 	/*
1173 	 * Configure/Initialize hardware
1174 	 */
1175 	if (e1000_init_hw(hw) != E1000_SUCCESS) {
1176 		igb_error(igb, "Failed to initialize hardware");
1177 		return (IGB_FAILURE);
1178 	}
1179 
1180 	/*
1181 	 * Make sure driver has control
1182 	 */
1183 	igb_get_driver_control(hw);
1184 
1185 	/*
1186 	 * Setup MSI-X interrupts
1187 	 */
1188 	if (igb->intr_type == DDI_INTR_TYPE_MSIX)
1189 		igb_setup_adapter_msix(igb);
1190 
1191 	/*
1192 	 * Initialize unicast addresses.
1193 	 */
1194 	igb_init_unicst(igb);
1195 
1196 	/*
1197 	 * Setup and initialize the mctable structures.
1198 	 */
1199 	igb_setup_multicst(igb);
1200 
1201 	/*
1202 	 * Set interrupt throttling rate
1203 	 */
1204 	for (i = 0; i < igb->intr_cnt; i++)
1205 		E1000_WRITE_REG(hw, E1000_EITR(i), igb->intr_throttling[i]);
1206 
1207 	/* Enable PCI-E master */
1208 	if (hw->bus.type == e1000_bus_type_pci_express) {
1209 		e1000_enable_pciex_master(hw);
1210 	}
1211 
1212 	/*
1213 	 * Save the state of the phy
1214 	 */
1215 	igb_get_phy_state(igb);
1216 
1217 	return (IGB_SUCCESS);
1218 }
1219 
1220 /*
1221  * igb_chip_stop - Stop the chipset hardware
1222  */
1223 static void
1224 igb_chip_stop(igb_t *igb)
1225 {
1226 	struct e1000_hw *hw = &igb->hw;
1227 
1228 	ASSERT(mutex_owned(&igb->gen_lock));
1229 
1230 	/* Tell firmware driver is no longer in control */
1231 	igb_release_driver_control(hw);
1232 
1233 	/*
1234 	 * Reset the chipset
1235 	 */
1236 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1237 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1238 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1239 	}
1240 
1241 	/*
1242 	 * Reset PHY if possible
1243 	 */
1244 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
1245 		(void) e1000_phy_hw_reset(hw);
1246 }
1247 
1248 /*
1249  * igb_reset - Reset the chipset and restart the driver.
1250  *
1251  * It involves stopping and re-starting the chipset,
1252  * and re-configuring the rx/tx rings.
1253  */
1254 static int
1255 igb_reset(igb_t *igb)
1256 {
1257 	int i;
1258 
1259 	mutex_enter(&igb->gen_lock);
1260 
1261 	ASSERT(igb->igb_state & IGB_STARTED);
1262 
1263 	/*
1264 	 * Disable the adapter interrupts to stop any rx/tx activities
1265 	 * before draining pending data and resetting hardware.
1266 	 */
1267 	igb_disable_adapter_interrupts(igb);
1268 
1269 	/*
1270 	 * Drain the pending transmit packets
1271 	 */
1272 	(void) igb_tx_drain(igb);
1273 
1274 	for (i = 0; i < igb->num_rx_rings; i++)
1275 		mutex_enter(&igb->rx_rings[i].rx_lock);
1276 	for (i = 0; i < igb->num_tx_rings; i++)
1277 		mutex_enter(&igb->tx_rings[i].tx_lock);
1278 
1279 	/*
1280 	 * Stop the chipset hardware
1281 	 */
1282 	igb_chip_stop(igb);
1283 
1284 	/*
1285 	 * Clean the pending tx data/resources
1286 	 */
1287 	igb_tx_clean(igb);
1288 
1289 	/*
1290 	 * Start the chipset hardware
1291 	 */
1292 	if (igb_chip_start(igb) != IGB_SUCCESS) {
1293 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1294 		goto reset_failure;
1295 	}
1296 
1297 	/*
1298 	 * Setup the rx/tx rings
1299 	 */
1300 	igb_setup_rings(igb);
1301 
1302 	/*
1303 	 * Enable adapter interrupts
1304 	 * The interrupts must be enabled after the driver state is START
1305 	 */
1306 	igb_enable_adapter_interrupts(igb);
1307 
1308 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1309 		goto reset_failure;
1310 
1311 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1312 		goto reset_failure;
1313 
1314 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1315 		mutex_exit(&igb->tx_rings[i].tx_lock);
1316 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1317 		mutex_exit(&igb->rx_rings[i].rx_lock);
1318 
1319 	mutex_exit(&igb->gen_lock);
1320 
1321 	return (IGB_SUCCESS);
1322 
1323 reset_failure:
1324 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1325 		mutex_exit(&igb->tx_rings[i].tx_lock);
1326 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1327 		mutex_exit(&igb->rx_rings[i].rx_lock);
1328 
1329 	mutex_exit(&igb->gen_lock);
1330 
1331 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1332 
1333 	return (IGB_FAILURE);
1334 }
1335 
1336 /*
1337  * igb_tx_clean - Clean the pending transmit packets and DMA resources
1338  */
1339 static void
1340 igb_tx_clean(igb_t *igb)
1341 {
1342 	igb_tx_ring_t *tx_ring;
1343 	tx_control_block_t *tcb;
1344 	link_list_t pending_list;
1345 	uint32_t desc_num;
1346 	int i, j;
1347 
1348 	LINK_LIST_INIT(&pending_list);
1349 
1350 	for (i = 0; i < igb->num_tx_rings; i++) {
1351 		tx_ring = &igb->tx_rings[i];
1352 
1353 		mutex_enter(&tx_ring->recycle_lock);
1354 
1355 		/*
1356 		 * Clean the pending tx data - the pending packets in the
1357 		 * work_list that have no chances to be transmitted again.
1358 		 *
1359 		 * We must ensure the chipset is stopped or the link is down
1360 		 * before cleaning the transmit packets.
1361 		 */
1362 		desc_num = 0;
1363 		for (j = 0; j < tx_ring->ring_size; j++) {
1364 			tcb = tx_ring->work_list[j];
1365 			if (tcb != NULL) {
1366 				desc_num += tcb->desc_num;
1367 
1368 				tx_ring->work_list[j] = NULL;
1369 
1370 				igb_free_tcb(tcb);
1371 
1372 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1373 			}
1374 		}
1375 
1376 		if (desc_num > 0) {
1377 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1378 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1379 
1380 			/*
1381 			 * Reset the head and tail pointers of the tbd ring
1382 			 */
1383 			tx_ring->tbd_head = 0;
1384 			tx_ring->tbd_tail = 0;
1385 
1386 			E1000_WRITE_REG(&igb->hw, E1000_TDH(tx_ring->index), 0);
1387 			E1000_WRITE_REG(&igb->hw, E1000_TDT(tx_ring->index), 0);
1388 		}
1389 
1390 		mutex_exit(&tx_ring->recycle_lock);
1391 
1392 		/*
1393 		 * Add the tx control blocks in the pending list to
1394 		 * the free list.
1395 		 */
1396 		igb_put_free_list(tx_ring, &pending_list);
1397 	}
1398 }
1399 
1400 /*
1401  * igb_tx_drain - Drain the tx rings to allow pending packets to be transmitted
1402  */
1403 static boolean_t
1404 igb_tx_drain(igb_t *igb)
1405 {
1406 	igb_tx_ring_t *tx_ring;
1407 	boolean_t done;
1408 	int i, j;
1409 
1410 	/*
1411 	 * Wait for a specific time to allow pending tx packets
1412 	 * to be transmitted.
1413 	 *
1414 	 * Check the counter tbd_free to see if transmission is done.
1415 	 * No lock protection is needed here.
1416 	 *
1417 	 * Return B_TRUE if all pending packets have been transmitted;
1418 	 * Otherwise return B_FALSE;
1419 	 */
1420 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1421 
1422 		done = B_TRUE;
1423 		for (j = 0; j < igb->num_tx_rings; j++) {
1424 			tx_ring = &igb->tx_rings[j];
1425 			done = done &&
1426 			    (tx_ring->tbd_free == tx_ring->ring_size);
1427 		}
1428 
1429 		if (done)
1430 			break;
1431 
1432 		msec_delay(1);
1433 	}
1434 
1435 	return (done);
1436 }
1437 
1438 /*
1439  * igb_rx_drain - Wait for all rx buffers to be released by upper layer
1440  */
1441 static boolean_t
1442 igb_rx_drain(igb_t *igb)
1443 {
1444 	igb_rx_ring_t *rx_ring;
1445 	boolean_t done;
1446 	int i, j;
1447 
1448 	/*
1449 	 * Polling the rx free list to check if those rx buffers held by
1450 	 * the upper layer are released.
1451 	 *
1452 	 * Check the counter rcb_free to see if all pending buffers are
1453 	 * released. No lock protection is needed here.
1454 	 *
1455 	 * Return B_TRUE if all pending buffers have been released;
1456 	 * Otherwise return B_FALSE;
1457 	 */
1458 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1459 
1460 		done = B_TRUE;
1461 		for (j = 0; j < igb->num_rx_rings; j++) {
1462 			rx_ring = &igb->rx_rings[j];
1463 			done = done &&
1464 			    (rx_ring->rcb_free == rx_ring->free_list_size);
1465 		}
1466 
1467 		if (done)
1468 			break;
1469 
1470 		msec_delay(1);
1471 	}
1472 
1473 	return (done);
1474 }
1475 
1476 /*
1477  * igb_start - Start the driver/chipset
1478  */
1479 int
1480 igb_start(igb_t *igb)
1481 {
1482 	int i;
1483 
1484 	ASSERT(mutex_owned(&igb->gen_lock));
1485 
1486 	for (i = 0; i < igb->num_rx_rings; i++)
1487 		mutex_enter(&igb->rx_rings[i].rx_lock);
1488 	for (i = 0; i < igb->num_tx_rings; i++)
1489 		mutex_enter(&igb->tx_rings[i].tx_lock);
1490 
1491 	/*
1492 	 * Start the chipset hardware
1493 	 */
1494 	if (igb_chip_start(igb) != IGB_SUCCESS) {
1495 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1496 		goto start_failure;
1497 	}
1498 
1499 	/*
1500 	 * Setup the rx/tx rings
1501 	 */
1502 	igb_setup_rings(igb);
1503 
1504 	/*
1505 	 * Enable adapter interrupts
1506 	 * The interrupts must be enabled after the driver state is START
1507 	 */
1508 	igb_enable_adapter_interrupts(igb);
1509 
1510 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1511 		goto start_failure;
1512 
1513 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1514 		goto start_failure;
1515 
1516 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1517 		mutex_exit(&igb->tx_rings[i].tx_lock);
1518 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1519 		mutex_exit(&igb->rx_rings[i].rx_lock);
1520 
1521 	return (IGB_SUCCESS);
1522 
1523 start_failure:
1524 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1525 		mutex_exit(&igb->tx_rings[i].tx_lock);
1526 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1527 		mutex_exit(&igb->rx_rings[i].rx_lock);
1528 
1529 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1530 
1531 	return (IGB_FAILURE);
1532 }
1533 
1534 /*
1535  * igb_stop - Stop the driver/chipset
1536  */
1537 void
1538 igb_stop(igb_t *igb)
1539 {
1540 	int i;
1541 
1542 	ASSERT(mutex_owned(&igb->gen_lock));
1543 
1544 	/*
1545 	 * Disable the adapter interrupts
1546 	 */
1547 	igb_disable_adapter_interrupts(igb);
1548 
1549 	/*
1550 	 * Drain the pending tx packets
1551 	 */
1552 	(void) igb_tx_drain(igb);
1553 
1554 	for (i = 0; i < igb->num_rx_rings; i++)
1555 		mutex_enter(&igb->rx_rings[i].rx_lock);
1556 	for (i = 0; i < igb->num_tx_rings; i++)
1557 		mutex_enter(&igb->tx_rings[i].tx_lock);
1558 
1559 	/*
1560 	 * Stop the chipset hardware
1561 	 */
1562 	igb_chip_stop(igb);
1563 
1564 	/*
1565 	 * Clean the pending tx data/resources
1566 	 */
1567 	igb_tx_clean(igb);
1568 
1569 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1570 		mutex_exit(&igb->tx_rings[i].tx_lock);
1571 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1572 		mutex_exit(&igb->rx_rings[i].rx_lock);
1573 
1574 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1575 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1576 }
1577 
1578 /*
1579  * igb_alloc_rings - Allocate memory space for rx/tx rings
1580  */
1581 static int
1582 igb_alloc_rings(igb_t *igb)
1583 {
1584 	/*
1585 	 * Allocate memory space for rx rings
1586 	 */
1587 	igb->rx_rings = kmem_zalloc(
1588 	    sizeof (igb_rx_ring_t) * igb->num_rx_rings,
1589 	    KM_NOSLEEP);
1590 
1591 	if (igb->rx_rings == NULL) {
1592 		return (IGB_FAILURE);
1593 	}
1594 
1595 	/*
1596 	 * Allocate memory space for tx rings
1597 	 */
1598 	igb->tx_rings = kmem_zalloc(
1599 	    sizeof (igb_tx_ring_t) * igb->num_tx_rings,
1600 	    KM_NOSLEEP);
1601 
1602 	if (igb->tx_rings == NULL) {
1603 		kmem_free(igb->rx_rings,
1604 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1605 		igb->rx_rings = NULL;
1606 		return (IGB_FAILURE);
1607 	}
1608 
1609 	return (IGB_SUCCESS);
1610 }
1611 
1612 /*
1613  * igb_free_rings - Free the memory space of rx/tx rings.
1614  */
1615 static void
1616 igb_free_rings(igb_t *igb)
1617 {
1618 	if (igb->rx_rings != NULL) {
1619 		kmem_free(igb->rx_rings,
1620 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1621 		igb->rx_rings = NULL;
1622 	}
1623 
1624 	if (igb->tx_rings != NULL) {
1625 		kmem_free(igb->tx_rings,
1626 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1627 		igb->tx_rings = NULL;
1628 	}
1629 }
1630 
1631 /*
1632  * igb_setup_rings - Setup rx/tx rings
1633  */
1634 static void
1635 igb_setup_rings(igb_t *igb)
1636 {
1637 	/*
1638 	 * Setup the rx/tx rings, including the following:
1639 	 *
1640 	 * 1. Setup the descriptor ring and the control block buffers;
1641 	 * 2. Initialize necessary registers for receive/transmit;
1642 	 * 3. Initialize software pointers/parameters for receive/transmit;
1643 	 */
1644 	igb_setup_rx(igb);
1645 
1646 	igb_setup_tx(igb);
1647 
1648 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1649 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1650 }
1651 
1652 static void
1653 igb_setup_rx_ring(igb_rx_ring_t *rx_ring)
1654 {
1655 	igb_t *igb = rx_ring->igb;
1656 	struct e1000_hw *hw = &igb->hw;
1657 	rx_control_block_t *rcb;
1658 	union e1000_adv_rx_desc	*rbd;
1659 	uint32_t size;
1660 	uint32_t buf_low;
1661 	uint32_t buf_high;
1662 	uint32_t reg_val;
1663 	int i;
1664 
1665 	ASSERT(mutex_owned(&rx_ring->rx_lock));
1666 	ASSERT(mutex_owned(&igb->gen_lock));
1667 
1668 	for (i = 0; i < igb->rx_ring_size; i++) {
1669 		rcb = rx_ring->work_list[i];
1670 		rbd = &rx_ring->rbd_ring[i];
1671 
1672 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
1673 		rbd->read.hdr_addr = NULL;
1674 	}
1675 
1676 	/*
1677 	 * Initialize the length register
1678 	 */
1679 	size = rx_ring->ring_size * sizeof (union e1000_adv_rx_desc);
1680 	E1000_WRITE_REG(hw, E1000_RDLEN(rx_ring->index), size);
1681 
1682 	/*
1683 	 * Initialize the base address registers
1684 	 */
1685 	buf_low = (uint32_t)rx_ring->rbd_area.dma_address;
1686 	buf_high = (uint32_t)(rx_ring->rbd_area.dma_address >> 32);
1687 	E1000_WRITE_REG(hw, E1000_RDBAH(rx_ring->index), buf_high);
1688 	E1000_WRITE_REG(hw, E1000_RDBAL(rx_ring->index), buf_low);
1689 
1690 	/*
1691 	 * Setup head & tail pointers
1692 	 */
1693 	E1000_WRITE_REG(hw, E1000_RDT(rx_ring->index), rx_ring->ring_size - 1);
1694 	E1000_WRITE_REG(hw, E1000_RDH(rx_ring->index), 0);
1695 
1696 	rx_ring->rbd_next = 0;
1697 
1698 	/*
1699 	 * Note: Considering the case that the chipset is being reset
1700 	 * and there are still some buffers held by the upper layer,
1701 	 * we should not reset the values of rcb_head, rcb_tail and
1702 	 * rcb_free;
1703 	 */
1704 	if (igb->igb_state == IGB_UNKNOWN) {
1705 		rx_ring->rcb_head = 0;
1706 		rx_ring->rcb_tail = 0;
1707 		rx_ring->rcb_free = rx_ring->free_list_size;
1708 	}
1709 
1710 	/*
1711 	 * Setup the Receive Descriptor Control Register (RXDCTL)
1712 	 */
1713 	reg_val = E1000_READ_REG(hw, E1000_RXDCTL(rx_ring->index));
1714 	reg_val |= E1000_RXDCTL_QUEUE_ENABLE;
1715 	reg_val &= 0xFFF00000;
1716 	reg_val |= 16;		/* pthresh */
1717 	reg_val |= 8 << 8;	/* hthresh */
1718 	reg_val |= 1 << 16;	/* wthresh */
1719 	E1000_WRITE_REG(hw, E1000_RXDCTL(rx_ring->index), reg_val);
1720 
1721 	/*
1722 	 * Setup the Split and Replication Receive Control Register.
1723 	 * Set the rx buffer size and the advanced descriptor type.
1724 	 */
1725 	reg_val = (igb->rx_buf_size >> E1000_SRRCTL_BSIZEPKT_SHIFT) |
1726 	    E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1727 
1728 	E1000_WRITE_REG(hw, E1000_SRRCTL(rx_ring->index), reg_val);
1729 }
1730 
1731 static void
1732 igb_setup_rx(igb_t *igb)
1733 {
1734 	igb_rx_ring_t *rx_ring;
1735 	struct e1000_hw *hw = &igb->hw;
1736 	uint32_t reg_val;
1737 	int i;
1738 
1739 	/*
1740 	 * Setup the Receive Control Register (RCTL), and ENABLE the
1741 	 * receiver. The initial configuration is to: Enable the receiver,
1742 	 * accept broadcasts, discard bad packets (and long packets),
1743 	 * disable VLAN filter checking, set the receive descriptor
1744 	 * minimum threshold size to 1/2, and the receive buffer size to
1745 	 * 2k.
1746 	 */
1747 	reg_val = E1000_RCTL_EN |	/* Enable Receive Unit */
1748 	    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
1749 	    E1000_RCTL_LPE |		/* Large Packet Enable bit */
1750 	    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
1751 	    E1000_RCTL_RDMTS_HALF |
1752 	    E1000_RCTL_SECRC |		/* Strip Ethernet CRC */
1753 	    E1000_RCTL_LBM_NO;		/* Loopback Mode = none */
1754 
1755 	E1000_WRITE_REG(hw, E1000_RCTL, reg_val);
1756 
1757 	/*
1758 	 * igb_setup_rx_ring must be called after configuring RCTL
1759 	 */
1760 	for (i = 0; i < igb->num_rx_rings; i++) {
1761 		rx_ring = &igb->rx_rings[i];
1762 		igb_setup_rx_ring(rx_ring);
1763 	}
1764 
1765 	/*
1766 	 * Setup the Rx Long Packet Max Length register
1767 	 */
1768 	E1000_WRITE_REG(hw, E1000_RLPML, igb->max_frame_size);
1769 
1770 	/*
1771 	 * Hardware checksum settings
1772 	 */
1773 	if (igb->rx_hcksum_enable) {
1774 		reg_val =
1775 		    E1000_RXCSUM_TUOFL |	/* TCP/UDP checksum */
1776 		    E1000_RXCSUM_IPOFL;		/* IP checksum */
1777 
1778 		E1000_WRITE_REG(hw, E1000_RXCSUM, reg_val);
1779 	}
1780 
1781 	/*
1782 	 * Setup RSS for multiple receive queues
1783 	 */
1784 	if (igb->num_rx_rings > 1)
1785 		igb_setup_rss(igb);
1786 }
1787 
1788 static void
1789 igb_setup_tx_ring(igb_tx_ring_t *tx_ring)
1790 {
1791 	igb_t *igb = tx_ring->igb;
1792 	struct e1000_hw *hw = &igb->hw;
1793 	uint32_t size;
1794 	uint32_t buf_low;
1795 	uint32_t buf_high;
1796 	uint32_t reg_val;
1797 
1798 	ASSERT(mutex_owned(&tx_ring->tx_lock));
1799 	ASSERT(mutex_owned(&igb->gen_lock));
1800 
1801 	/*
1802 	 * Initialize the length register
1803 	 */
1804 	size = tx_ring->ring_size * sizeof (union e1000_adv_tx_desc);
1805 	E1000_WRITE_REG(hw, E1000_TDLEN(tx_ring->index), size);
1806 
1807 	/*
1808 	 * Initialize the base address registers
1809 	 */
1810 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
1811 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
1812 	E1000_WRITE_REG(hw, E1000_TDBAL(tx_ring->index), buf_low);
1813 	E1000_WRITE_REG(hw, E1000_TDBAH(tx_ring->index), buf_high);
1814 
1815 	/*
1816 	 * Setup head & tail pointers
1817 	 */
1818 	E1000_WRITE_REG(hw, E1000_TDH(tx_ring->index), 0);
1819 	E1000_WRITE_REG(hw, E1000_TDT(tx_ring->index), 0);
1820 
1821 	/*
1822 	 * Setup head write-back
1823 	 */
1824 	if (igb->tx_head_wb_enable) {
1825 		/*
1826 		 * The memory of the head write-back is allocated using
1827 		 * the extra tbd beyond the tail of the tbd ring.
1828 		 */
1829 		tx_ring->tbd_head_wb = (uint32_t *)
1830 		    ((uintptr_t)tx_ring->tbd_area.address + size);
1831 
1832 		buf_low = (uint32_t)
1833 		    (tx_ring->tbd_area.dma_address + size);
1834 		buf_high = (uint32_t)
1835 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
1836 
1837 		/* Set the head write-back enable bit */
1838 		buf_low |= E1000_TX_HEAD_WB_ENABLE;
1839 
1840 		E1000_WRITE_REG(hw, E1000_TDWBAL(tx_ring->index), buf_low);
1841 		E1000_WRITE_REG(hw, E1000_TDWBAH(tx_ring->index), buf_high);
1842 
1843 		/*
1844 		 * Turn off relaxed ordering for head write back or it will
1845 		 * cause problems with the tx recycling
1846 		 */
1847 		reg_val = E1000_READ_REG(hw,
1848 		    E1000_DCA_TXCTRL(tx_ring->index));
1849 		reg_val &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1850 		E1000_WRITE_REG(hw,
1851 		    E1000_DCA_TXCTRL(tx_ring->index), reg_val);
1852 	} else {
1853 		tx_ring->tbd_head_wb = NULL;
1854 	}
1855 
1856 	tx_ring->tbd_head = 0;
1857 	tx_ring->tbd_tail = 0;
1858 	tx_ring->tbd_free = tx_ring->ring_size;
1859 
1860 	/*
1861 	 * Note: Considering the case that the chipset is being reset,
1862 	 * and there are still some buffers held by the upper layer,
1863 	 * we should not reset the values of tcb_head, tcb_tail.
1864 	 */
1865 	if (igb->igb_state == IGB_UNKNOWN) {
1866 		tx_ring->tcb_head = 0;
1867 		tx_ring->tcb_tail = 0;
1868 		tx_ring->tcb_free = tx_ring->free_list_size;
1869 	} else {
1870 		ASSERT(tx_ring->tcb_free == tx_ring->free_list_size);
1871 	}
1872 
1873 	/*
1874 	 * Initialize hardware checksum offload settings
1875 	 */
1876 	tx_ring->hcksum_context.hcksum_flags = 0;
1877 	tx_ring->hcksum_context.ip_hdr_len = 0;
1878 	tx_ring->hcksum_context.mac_hdr_len = 0;
1879 	tx_ring->hcksum_context.l4_proto = 0;
1880 }
1881 
1882 static void
1883 igb_setup_tx(igb_t *igb)
1884 {
1885 	igb_tx_ring_t *tx_ring;
1886 	struct e1000_hw *hw = &igb->hw;
1887 	uint32_t reg_val;
1888 	int i;
1889 
1890 	for (i = 0; i < igb->num_tx_rings; i++) {
1891 		tx_ring = &igb->tx_rings[i];
1892 		igb_setup_tx_ring(tx_ring);
1893 	}
1894 
1895 	/*
1896 	 * Setup the Transmit Control Register (TCTL)
1897 	 */
1898 	reg_val = E1000_TCTL_PSP | E1000_TCTL_EN |
1899 	    (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT) |
1900 	    (E1000_COLLISION_DISTANCE << E1000_COLD_SHIFT) |
1901 	    E1000_TCTL_RTLC;
1902 
1903 	/* Enable the MULR bit */
1904 	if (hw->bus.type == e1000_bus_type_pci_express)
1905 		reg_val |= E1000_TCTL_MULR;
1906 
1907 	E1000_WRITE_REG(hw, E1000_TCTL, reg_val);
1908 
1909 	/*
1910 	 * Set the default values for the Tx Inter Packet Gap timer
1911 	 */
1912 	if (hw->phy.media_type == e1000_media_type_fiber)
1913 		reg_val = DEFAULT_82543_TIPG_IPGT_FIBER;
1914 	else
1915 		reg_val = DEFAULT_82543_TIPG_IPGT_COPPER;
1916 	reg_val |=
1917 	    DEFAULT_82543_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT;
1918 	reg_val |=
1919 	    DEFAULT_82543_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
1920 
1921 	E1000_WRITE_REG(hw, E1000_TIPG, reg_val);
1922 }
1923 
1924 /*
1925  * igb_setup_rss - Setup receive-side scaling feature
1926  */
1927 static void
1928 igb_setup_rss(igb_t *igb)
1929 {
1930 	struct e1000_hw *hw = &igb->hw;
1931 	uint32_t i, mrqc, rxcsum;
1932 	int shift;
1933 	uint32_t random;
1934 	union e1000_reta {
1935 		uint32_t	dword;
1936 		uint8_t		bytes[4];
1937 	} reta;
1938 
1939 	/* Setup the Redirection Table */
1940 	shift = 6;
1941 	for (i = 0; i < (32 * 4); i++) {
1942 		reta.bytes[i & 3] = (i % igb->num_rx_rings) << shift;
1943 		if ((i & 3) == 3) {
1944 			E1000_WRITE_REG(hw,
1945 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
1946 		}
1947 	}
1948 
1949 	/* Fill out hash function seeds */
1950 	for (i = 0; i < 10; i++) {
1951 		(void) random_get_pseudo_bytes((uint8_t *)&random,
1952 		    sizeof (uint32_t));
1953 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
1954 	}
1955 
1956 	/* Setup the Multiple Receive Queue Control register */
1957 	mrqc = E1000_MRQC_ENABLE_RSS_4Q;
1958 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
1959 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
1960 	    E1000_MRQC_RSS_FIELD_IPV6 |
1961 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
1962 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
1963 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
1964 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
1965 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
1966 
1967 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
1968 
1969 	/*
1970 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
1971 	 *
1972 	 * The Packet Checksum is not ethernet CRC. It is another kind of
1973 	 * checksum offloading provided by the 82575 chipset besides the IP
1974 	 * header checksum offloading and the TCP/UDP checksum offloading.
1975 	 * The Packet Checksum is by default computed over the entire packet
1976 	 * from the first byte of the DA through the last byte of the CRC,
1977 	 * including the Ethernet and IP headers.
1978 	 *
1979 	 * It is a hardware limitation that Packet Checksum is mutually
1980 	 * exclusive with RSS.
1981 	 */
1982 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
1983 	rxcsum |= E1000_RXCSUM_PCSD;
1984 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
1985 }
1986 
1987 /*
1988  * igb_init_unicst - Initialize the unicast addresses
1989  */
1990 static void
1991 igb_init_unicst(igb_t *igb)
1992 {
1993 	struct e1000_hw *hw = &igb->hw;
1994 	int slot;
1995 
1996 	/*
1997 	 * Here we should consider two situations:
1998 	 *
1999 	 * 1. Chipset is initialized the first time
2000 	 *    Initialize the multiple unicast addresses, and
2001 	 *    save the default mac address.
2002 	 *
2003 	 * 2. Chipset is reset
2004 	 *    Recover the multiple unicast addresses from the
2005 	 *    software data structure to the RAR registers.
2006 	 */
2007 	if (!igb->unicst_init) {
2008 		/* Initialize the multiple unicast addresses */
2009 		igb->unicst_total = MAX_NUM_UNICAST_ADDRESSES;
2010 
2011 		igb->unicst_avail = igb->unicst_total - 1;
2012 
2013 		/* Store the default mac address */
2014 		e1000_rar_set(hw, hw->mac.addr, 0);
2015 
2016 		bcopy(hw->mac.addr, igb->unicst_addr[0].mac.addr,
2017 		    ETHERADDRL);
2018 		igb->unicst_addr[0].mac.set = 1;
2019 
2020 		for (slot = 1; slot < igb->unicst_total; slot++)
2021 			igb->unicst_addr[slot].mac.set = 0;
2022 
2023 		igb->unicst_init = B_TRUE;
2024 	} else {
2025 		/* Recover the default mac address */
2026 		bcopy(igb->unicst_addr[0].mac.addr, hw->mac.addr,
2027 		    ETHERADDRL);
2028 
2029 		/* Store the default mac address */
2030 		e1000_rar_set(hw, hw->mac.addr, 0);
2031 
2032 		/* Re-configure the RAR registers */
2033 		for (slot = 1; slot < igb->unicst_total; slot++)
2034 			e1000_rar_set(hw,
2035 			    igb->unicst_addr[slot].mac.addr, slot);
2036 	}
2037 
2038 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2039 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2040 }
2041 
2042 /*
2043  * igb_unicst_set - Set the unicast address to the specified slot
2044  */
2045 int
2046 igb_unicst_set(igb_t *igb, const uint8_t *mac_addr,
2047     mac_addr_slot_t slot)
2048 {
2049 	struct e1000_hw *hw = &igb->hw;
2050 
2051 	ASSERT(mutex_owned(&igb->gen_lock));
2052 
2053 	/*
2054 	 * Save the unicast address in the software data structure
2055 	 */
2056 	bcopy(mac_addr, igb->unicst_addr[slot].mac.addr, ETHERADDRL);
2057 
2058 	/*
2059 	 * Set the unicast address to the RAR register
2060 	 */
2061 	e1000_rar_set(hw, (uint8_t *)mac_addr, slot);
2062 
2063 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2064 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2065 		return (EIO);
2066 	}
2067 
2068 	return (0);
2069 }
2070 
2071 /*
2072  * igb_multicst_add - Add a multicst address
2073  */
2074 int
2075 igb_multicst_add(igb_t *igb, const uint8_t *multiaddr)
2076 {
2077 	ASSERT(mutex_owned(&igb->gen_lock));
2078 
2079 	if ((multiaddr[0] & 01) == 0) {
2080 		return (EINVAL);
2081 	}
2082 
2083 	if (igb->mcast_count >= MAX_NUM_MULTICAST_ADDRESSES) {
2084 		return (ENOENT);
2085 	}
2086 
2087 	bcopy(multiaddr,
2088 	    &igb->mcast_table[igb->mcast_count], ETHERADDRL);
2089 	igb->mcast_count++;
2090 
2091 	/*
2092 	 * Update the multicast table in the hardware
2093 	 */
2094 	igb_setup_multicst(igb);
2095 
2096 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2097 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2098 		return (EIO);
2099 	}
2100 
2101 	return (0);
2102 }
2103 
2104 /*
2105  * igb_multicst_remove - Remove a multicst address
2106  */
2107 int
2108 igb_multicst_remove(igb_t *igb, const uint8_t *multiaddr)
2109 {
2110 	int i;
2111 
2112 	ASSERT(mutex_owned(&igb->gen_lock));
2113 
2114 	for (i = 0; i < igb->mcast_count; i++) {
2115 		if (bcmp(multiaddr, &igb->mcast_table[i],
2116 		    ETHERADDRL) == 0) {
2117 			for (i++; i < igb->mcast_count; i++) {
2118 				igb->mcast_table[i - 1] =
2119 				    igb->mcast_table[i];
2120 			}
2121 			igb->mcast_count--;
2122 			break;
2123 		}
2124 	}
2125 
2126 	/*
2127 	 * Update the multicast table in the hardware
2128 	 */
2129 	igb_setup_multicst(igb);
2130 
2131 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2132 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2133 		return (EIO);
2134 	}
2135 
2136 	return (0);
2137 }
2138 
2139 /*
2140  * igb_setup_multicast - setup multicast data structures
2141  *
2142  * This routine initializes all of the multicast related structures
2143  * and save them in the hardware registers.
2144  */
2145 static void
2146 igb_setup_multicst(igb_t *igb)
2147 {
2148 	uint8_t *mc_addr_list;
2149 	uint32_t mc_addr_count;
2150 	struct e1000_hw *hw = &igb->hw;
2151 
2152 	ASSERT(mutex_owned(&igb->gen_lock));
2153 
2154 	ASSERT(igb->mcast_count <= MAX_NUM_MULTICAST_ADDRESSES);
2155 
2156 	mc_addr_list = (uint8_t *)igb->mcast_table;
2157 	mc_addr_count = igb->mcast_count;
2158 
2159 	/*
2160 	 * Update the multicase addresses to the MTA registers
2161 	 */
2162 	e1000_update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2163 	    igb->unicst_total, hw->mac.rar_entry_count);
2164 }
2165 
2166 /*
2167  * igb_get_conf - Get driver configurations set in driver.conf
2168  *
2169  * This routine gets user-configured values out of the configuration
2170  * file igb.conf.
2171  *
2172  * For each configurable value, there is a minimum, a maximum, and a
2173  * default.
2174  * If user does not configure a value, use the default.
2175  * If user configures below the minimum, use the minumum.
2176  * If user configures above the maximum, use the maxumum.
2177  */
2178 static void
2179 igb_get_conf(igb_t *igb)
2180 {
2181 	struct e1000_hw *hw = &igb->hw;
2182 	uint32_t default_mtu;
2183 	uint32_t flow_control;
2184 
2185 	/*
2186 	 * igb driver supports the following user configurations:
2187 	 *
2188 	 * Link configurations:
2189 	 *    adv_autoneg_cap
2190 	 *    adv_1000fdx_cap
2191 	 *    adv_100fdx_cap
2192 	 *    adv_100hdx_cap
2193 	 *    adv_10fdx_cap
2194 	 *    adv_10hdx_cap
2195 	 * Note: 1000hdx is not supported.
2196 	 *
2197 	 * Jumbo frame configuration:
2198 	 *    default_mtu
2199 	 *
2200 	 * Ethernet flow control configuration:
2201 	 *    flow_control
2202 	 *
2203 	 * Multiple rings configurations:
2204 	 *    tx_queue_number
2205 	 *    tx_ring_size
2206 	 *    rx_queue_number
2207 	 *    rx_ring_size
2208 	 *
2209 	 * Call igb_get_prop() to get the value for a specific
2210 	 * configuration parameter.
2211 	 */
2212 
2213 	/*
2214 	 * Link configurations
2215 	 */
2216 	igb->param_adv_autoneg_cap = igb_get_prop(igb,
2217 	    PROP_ADV_AUTONEG_CAP, 0, 1, 1);
2218 	igb->param_adv_1000fdx_cap = igb_get_prop(igb,
2219 	    PROP_ADV_1000FDX_CAP, 0, 1, 1);
2220 	igb->param_adv_100fdx_cap = igb_get_prop(igb,
2221 	    PROP_ADV_100FDX_CAP, 0, 1, 1);
2222 	igb->param_adv_100hdx_cap = igb_get_prop(igb,
2223 	    PROP_ADV_100HDX_CAP, 0, 1, 1);
2224 	igb->param_adv_10fdx_cap = igb_get_prop(igb,
2225 	    PROP_ADV_10FDX_CAP, 0, 1, 1);
2226 	igb->param_adv_10hdx_cap = igb_get_prop(igb,
2227 	    PROP_ADV_10HDX_CAP, 0, 1, 1);
2228 
2229 	/*
2230 	 * Jumbo frame configurations
2231 	 */
2232 	default_mtu = igb_get_prop(igb, PROP_DEFAULT_MTU,
2233 	    MIN_MTU, MAX_MTU, DEFAULT_MTU);
2234 
2235 	igb->max_frame_size = default_mtu +
2236 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
2237 
2238 	/*
2239 	 * Ethernet flow control configuration
2240 	 */
2241 	flow_control = igb_get_prop(igb, PROP_FLOW_CONTROL,
2242 	    e1000_fc_none, 4, e1000_fc_full);
2243 	if (flow_control == 4)
2244 		flow_control = e1000_fc_default;
2245 
2246 	hw->fc.type = flow_control;
2247 
2248 	/*
2249 	 * Multiple rings configurations
2250 	 */
2251 	igb->num_tx_rings = igb_get_prop(igb, PROP_TX_QUEUE_NUM,
2252 	    MIN_TX_QUEUE_NUM, MAX_TX_QUEUE_NUM, DEFAULT_TX_QUEUE_NUM);
2253 	igb->tx_ring_size = igb_get_prop(igb, PROP_TX_RING_SIZE,
2254 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
2255 
2256 	igb->num_rx_rings = igb_get_prop(igb, PROP_RX_QUEUE_NUM,
2257 	    MIN_RX_QUEUE_NUM, MAX_RX_QUEUE_NUM, DEFAULT_RX_QUEUE_NUM);
2258 	igb->rx_ring_size = igb_get_prop(igb, PROP_RX_RING_SIZE,
2259 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
2260 
2261 	/*
2262 	 * Tunable used to force an interrupt type. The only use is
2263 	 * for testing of the lesser interrupt types.
2264 	 * 0 = don't force interrupt type
2265 	 * 1 = force interrupt type MSIX
2266 	 * 2 = force interrupt type MSI
2267 	 * 3 = force interrupt type Legacy
2268 	 */
2269 	igb->intr_force = igb_get_prop(igb, PROP_INTR_FORCE,
2270 	    IGB_INTR_NONE, IGB_INTR_LEGACY, IGB_INTR_NONE);
2271 
2272 	igb->tx_hcksum_enable = igb_get_prop(igb, PROP_TX_HCKSUM_ENABLE,
2273 	    0, 1, 1);
2274 	igb->rx_hcksum_enable = igb_get_prop(igb, PROP_RX_HCKSUM_ENABLE,
2275 	    0, 1, 1);
2276 	igb->lso_enable = igb_get_prop(igb, PROP_LSO_ENABLE,
2277 	    0, 1, 0);
2278 	igb->tx_head_wb_enable = igb_get_prop(igb, PROP_TX_HEAD_WB_ENABLE,
2279 	    0, 1, 1);
2280 
2281 	igb->tx_copy_thresh = igb_get_prop(igb, PROP_TX_COPY_THRESHOLD,
2282 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
2283 	    DEFAULT_TX_COPY_THRESHOLD);
2284 	igb->tx_recycle_thresh = igb_get_prop(igb, PROP_TX_RECYCLE_THRESHOLD,
2285 	    MIN_TX_RECYCLE_THRESHOLD, MAX_TX_RECYCLE_THRESHOLD,
2286 	    DEFAULT_TX_RECYCLE_THRESHOLD);
2287 	igb->tx_overload_thresh = igb_get_prop(igb, PROP_TX_OVERLOAD_THRESHOLD,
2288 	    MIN_TX_OVERLOAD_THRESHOLD, MAX_TX_OVERLOAD_THRESHOLD,
2289 	    DEFAULT_TX_OVERLOAD_THRESHOLD);
2290 	igb->tx_resched_thresh = igb_get_prop(igb, PROP_TX_RESCHED_THRESHOLD,
2291 	    MIN_TX_RESCHED_THRESHOLD, MAX_TX_RESCHED_THRESHOLD,
2292 	    DEFAULT_TX_RESCHED_THRESHOLD);
2293 
2294 	igb->rx_copy_thresh = igb_get_prop(igb, PROP_RX_COPY_THRESHOLD,
2295 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
2296 	    DEFAULT_RX_COPY_THRESHOLD);
2297 	igb->rx_limit_per_intr = igb_get_prop(igb, PROP_RX_LIMIT_PER_INTR,
2298 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
2299 	    DEFAULT_RX_LIMIT_PER_INTR);
2300 
2301 	igb->intr_throttling[0] = igb_get_prop(igb, PROP_INTR_THROTTLING,
2302 	    MIN_INTR_THROTTLING, MAX_INTR_THROTTLING,
2303 	    DEFAULT_INTR_THROTTLING);
2304 }
2305 
2306 /*
2307  * igb_get_prop - Get a property value out of the configuration file igb.conf
2308  *
2309  * Caller provides the name of the property, a default value, a minimum
2310  * value, and a maximum value.
2311  *
2312  * Return configured value of the property, with default, minimum and
2313  * maximum properly applied.
2314  */
2315 static int
2316 igb_get_prop(igb_t *igb,
2317     char *propname,	/* name of the property */
2318     int minval,		/* minimum acceptable value */
2319     int maxval,		/* maximim acceptable value */
2320     int defval)		/* default value */
2321 {
2322 	int value;
2323 
2324 	/*
2325 	 * Call ddi_prop_get_int() to read the conf settings
2326 	 */
2327 	value = ddi_prop_get_int(DDI_DEV_T_ANY, igb->dip,
2328 	    DDI_PROP_DONTPASS, propname, defval);
2329 
2330 	if (value > maxval)
2331 		value = maxval;
2332 
2333 	if (value < minval)
2334 		value = minval;
2335 
2336 	return (value);
2337 }
2338 
2339 /*
2340  * igb_setup_link - Using the link properties to setup the link
2341  */
2342 int
2343 igb_setup_link(igb_t *igb, boolean_t setup_hw)
2344 {
2345 	struct e1000_mac_info *mac;
2346 	struct e1000_phy_info *phy;
2347 	boolean_t invalid;
2348 
2349 	mac = &igb->hw.mac;
2350 	phy = &igb->hw.phy;
2351 	invalid = B_FALSE;
2352 
2353 	if (igb->param_adv_autoneg_cap == 1) {
2354 		mac->autoneg = B_TRUE;
2355 		phy->autoneg_advertised = 0;
2356 
2357 		/*
2358 		 * 1000hdx is not supported for autonegotiation
2359 		 */
2360 		if (igb->param_adv_1000fdx_cap == 1)
2361 			phy->autoneg_advertised |= ADVERTISE_1000_FULL;
2362 
2363 		if (igb->param_adv_100fdx_cap == 1)
2364 			phy->autoneg_advertised |= ADVERTISE_100_FULL;
2365 
2366 		if (igb->param_adv_100hdx_cap == 1)
2367 			phy->autoneg_advertised |= ADVERTISE_100_HALF;
2368 
2369 		if (igb->param_adv_10fdx_cap == 1)
2370 			phy->autoneg_advertised |= ADVERTISE_10_FULL;
2371 
2372 		if (igb->param_adv_10hdx_cap == 1)
2373 			phy->autoneg_advertised |= ADVERTISE_10_HALF;
2374 
2375 		if (phy->autoneg_advertised == 0)
2376 			invalid = B_TRUE;
2377 	} else {
2378 		mac->autoneg = B_FALSE;
2379 
2380 		/*
2381 		 * 1000fdx and 1000hdx are not supported for forced link
2382 		 */
2383 		if (igb->param_adv_100fdx_cap == 1)
2384 			mac->forced_speed_duplex = ADVERTISE_100_FULL;
2385 		else if (igb->param_adv_100hdx_cap == 1)
2386 			mac->forced_speed_duplex = ADVERTISE_100_HALF;
2387 		else if (igb->param_adv_10fdx_cap == 1)
2388 			mac->forced_speed_duplex = ADVERTISE_10_FULL;
2389 		else if (igb->param_adv_10hdx_cap == 1)
2390 			mac->forced_speed_duplex = ADVERTISE_10_HALF;
2391 		else
2392 			invalid = B_TRUE;
2393 	}
2394 
2395 	if (invalid) {
2396 		igb_notice(igb, "Invalid link settings. Setup link to "
2397 		    "autonegotiation with full link capabilities.");
2398 		mac->autoneg = B_TRUE;
2399 		phy->autoneg_advertised = ADVERTISE_1000_FULL |
2400 		    ADVERTISE_100_FULL | ADVERTISE_100_HALF |
2401 		    ADVERTISE_10_FULL | ADVERTISE_10_HALF;
2402 	}
2403 
2404 	if (setup_hw) {
2405 		if (e1000_setup_link(&igb->hw) != E1000_SUCCESS)
2406 			return (IGB_FAILURE);
2407 	}
2408 
2409 	return (IGB_SUCCESS);
2410 }
2411 
2412 
2413 /*
2414  * igb_is_link_up - Check if the link is up
2415  */
2416 static boolean_t
2417 igb_is_link_up(igb_t *igb)
2418 {
2419 	struct e1000_hw *hw = &igb->hw;
2420 	boolean_t link_up;
2421 
2422 	ASSERT(mutex_owned(&igb->gen_lock));
2423 
2424 	(void) e1000_check_for_link(hw);
2425 
2426 	if ((E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU) ||
2427 	    ((hw->phy.media_type == e1000_media_type_internal_serdes) &&
2428 	    (hw->mac.serdes_has_link))) {
2429 		link_up = B_TRUE;
2430 	} else {
2431 		link_up = B_FALSE;
2432 	}
2433 
2434 	return (link_up);
2435 }
2436 
2437 /*
2438  * igb_link_check - Link status processing
2439  */
2440 static boolean_t
2441 igb_link_check(igb_t *igb)
2442 {
2443 	struct e1000_hw *hw = &igb->hw;
2444 	uint16_t speed = 0, duplex = 0;
2445 	boolean_t link_changed = B_FALSE;
2446 
2447 	ASSERT(mutex_owned(&igb->gen_lock));
2448 
2449 	if (igb_is_link_up(igb)) {
2450 		/*
2451 		 * The Link is up, check whether it was marked as down earlier
2452 		 */
2453 		if (igb->link_state != LINK_STATE_UP) {
2454 			(void) e1000_get_speed_and_duplex(hw, &speed, &duplex);
2455 			igb->link_speed = speed;
2456 			igb->link_duplex = duplex;
2457 			igb->link_state = LINK_STATE_UP;
2458 			igb->link_down_timeout = 0;
2459 			link_changed = B_TRUE;
2460 		}
2461 	} else {
2462 		if (igb->link_state != LINK_STATE_DOWN) {
2463 			igb->link_speed = 0;
2464 			igb->link_duplex = 0;
2465 			igb->link_state = LINK_STATE_DOWN;
2466 			link_changed = B_TRUE;
2467 		}
2468 
2469 		if (igb->igb_state & IGB_STARTED) {
2470 			if (igb->link_down_timeout < MAX_LINK_DOWN_TIMEOUT) {
2471 				igb->link_down_timeout++;
2472 			} else if (igb->link_down_timeout ==
2473 			    MAX_LINK_DOWN_TIMEOUT) {
2474 				igb_tx_clean(igb);
2475 				igb->link_down_timeout++;
2476 			}
2477 		}
2478 	}
2479 
2480 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2481 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2482 
2483 	return (link_changed);
2484 }
2485 
2486 /*
2487  * igb_local_timer - driver watchdog function
2488  *
2489  * This function will handle the transmit stall check, link status check and
2490  * other routines.
2491  */
2492 static void
2493 igb_local_timer(void *arg)
2494 {
2495 	igb_t *igb = (igb_t *)arg;
2496 	struct e1000_hw *hw = &igb->hw;
2497 	boolean_t link_changed;
2498 
2499 	if (igb_stall_check(igb)) {
2500 		igb_fm_ereport(igb, DDI_FM_DEVICE_STALL);
2501 		igb->reset_count++;
2502 		if (igb_reset(igb) == IGB_SUCCESS)
2503 			ddi_fm_service_impact(igb->dip,
2504 			    DDI_SERVICE_RESTORED);
2505 	}
2506 
2507 	mutex_enter(&igb->gen_lock);
2508 	link_changed = igb_link_check(igb);
2509 	mutex_exit(&igb->gen_lock);
2510 
2511 	if (link_changed)
2512 		mac_link_update(igb->mac_hdl, igb->link_state);
2513 
2514 	/*
2515 	 * Set Timer Interrupts
2516 	 */
2517 	if (igb->intr_type != DDI_INTR_TYPE_MSIX)
2518 		E1000_WRITE_REG(hw, E1000_ICS, E1000_IMS_RXT0);
2519 
2520 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2521 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2522 
2523 	igb_restart_watchdog_timer(igb);
2524 }
2525 
2526 /*
2527  * igb_stall_check - check for transmit stall
2528  *
2529  * This function checks if the adapter is stalled (in transmit).
2530  *
2531  * It is called each time the watchdog timeout is invoked.
2532  * If the transmit descriptor reclaim continuously fails,
2533  * the watchdog value will increment by 1. If the watchdog
2534  * value exceeds the threshold, the igb is assumed to
2535  * have stalled and need to be reset.
2536  */
2537 static boolean_t
2538 igb_stall_check(igb_t *igb)
2539 {
2540 	igb_tx_ring_t *tx_ring;
2541 	boolean_t result;
2542 	int i;
2543 
2544 	if (igb->link_state != LINK_STATE_UP)
2545 		return (B_FALSE);
2546 
2547 	/*
2548 	 * If any tx ring is stalled, we'll reset the chipset
2549 	 */
2550 	result = B_FALSE;
2551 	for (i = 0; i < igb->num_tx_rings; i++) {
2552 		tx_ring = &igb->tx_rings[i];
2553 
2554 		if (tx_ring->recycle_fail > 0)
2555 			tx_ring->stall_watchdog++;
2556 		else
2557 			tx_ring->stall_watchdog = 0;
2558 
2559 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
2560 			result = B_TRUE;
2561 			break;
2562 		}
2563 	}
2564 
2565 	if (result) {
2566 		tx_ring->stall_watchdog = 0;
2567 		tx_ring->recycle_fail = 0;
2568 	}
2569 
2570 	return (result);
2571 }
2572 
2573 
2574 /*
2575  * is_valid_mac_addr - Check if the mac address is valid
2576  */
2577 static boolean_t
2578 is_valid_mac_addr(uint8_t *mac_addr)
2579 {
2580 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
2581 	const uint8_t addr_test2[6] =
2582 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
2583 
2584 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
2585 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
2586 		return (B_FALSE);
2587 
2588 	return (B_TRUE);
2589 }
2590 
2591 static boolean_t
2592 igb_find_mac_address(igb_t *igb)
2593 {
2594 	struct e1000_hw *hw = &igb->hw;
2595 #ifdef __sparc
2596 	uchar_t *bytes;
2597 	struct ether_addr sysaddr;
2598 	uint_t nelts;
2599 	int err;
2600 	boolean_t found = B_FALSE;
2601 
2602 	/*
2603 	 * The "vendor's factory-set address" may already have
2604 	 * been extracted from the chip, but if the property
2605 	 * "local-mac-address" is set we use that instead.
2606 	 *
2607 	 * We check whether it looks like an array of 6
2608 	 * bytes (which it should, if OBP set it).  If we can't
2609 	 * make sense of it this way, we'll ignore it.
2610 	 */
2611 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
2612 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
2613 	if (err == DDI_PROP_SUCCESS) {
2614 		if (nelts == ETHERADDRL) {
2615 			while (nelts--)
2616 				hw->mac.addr[nelts] = bytes[nelts];
2617 			found = B_TRUE;
2618 		}
2619 		ddi_prop_free(bytes);
2620 	}
2621 
2622 	/*
2623 	 * Look up the OBP property "local-mac-address?". If the user has set
2624 	 * 'local-mac-address? = false', use "the system address" instead.
2625 	 */
2626 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip, 0,
2627 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
2628 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
2629 			if (localetheraddr(NULL, &sysaddr) != 0) {
2630 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
2631 				found = B_TRUE;
2632 			}
2633 		}
2634 		ddi_prop_free(bytes);
2635 	}
2636 
2637 	/*
2638 	 * Finally(!), if there's a valid "mac-address" property (created
2639 	 * if we netbooted from this interface), we must use this instead
2640 	 * of any of the above to ensure that the NFS/install server doesn't
2641 	 * get confused by the address changing as Solaris takes over!
2642 	 */
2643 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
2644 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
2645 	if (err == DDI_PROP_SUCCESS) {
2646 		if (nelts == ETHERADDRL) {
2647 			while (nelts--)
2648 				hw->mac.addr[nelts] = bytes[nelts];
2649 			found = B_TRUE;
2650 		}
2651 		ddi_prop_free(bytes);
2652 	}
2653 
2654 	if (found) {
2655 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
2656 		return (B_TRUE);
2657 	}
2658 #endif
2659 
2660 	/*
2661 	 * Read the device MAC address from the EEPROM
2662 	 */
2663 	if (e1000_read_mac_addr(hw) != E1000_SUCCESS)
2664 		return (B_FALSE);
2665 
2666 	return (B_TRUE);
2667 }
2668 
2669 #pragma inline(igb_arm_watchdog_timer)
2670 
2671 static void
2672 igb_arm_watchdog_timer(igb_t *igb)
2673 {
2674 	/*
2675 	 * Fire a watchdog timer
2676 	 */
2677 	igb->watchdog_tid =
2678 	    timeout(igb_local_timer,
2679 	    (void *)igb, 1 * drv_usectohz(1000000));
2680 
2681 }
2682 
2683 /*
2684  * igb_enable_watchdog_timer - Enable and start the driver watchdog timer
2685  */
2686 void
2687 igb_enable_watchdog_timer(igb_t *igb)
2688 {
2689 	mutex_enter(&igb->watchdog_lock);
2690 
2691 	if (!igb->watchdog_enable) {
2692 		igb->watchdog_enable = B_TRUE;
2693 		igb->watchdog_start = B_TRUE;
2694 		igb_arm_watchdog_timer(igb);
2695 	}
2696 
2697 	mutex_exit(&igb->watchdog_lock);
2698 
2699 }
2700 
2701 /*
2702  * igb_disable_watchdog_timer - Disable and stop the driver watchdog timer
2703  */
2704 void
2705 igb_disable_watchdog_timer(igb_t *igb)
2706 {
2707 	timeout_id_t tid;
2708 
2709 	mutex_enter(&igb->watchdog_lock);
2710 
2711 	igb->watchdog_enable = B_FALSE;
2712 	igb->watchdog_start = B_FALSE;
2713 	tid = igb->watchdog_tid;
2714 	igb->watchdog_tid = 0;
2715 
2716 	mutex_exit(&igb->watchdog_lock);
2717 
2718 	if (tid != 0)
2719 		(void) untimeout(tid);
2720 
2721 }
2722 
2723 /*
2724  * igb_start_watchdog_timer - Start the driver watchdog timer
2725  */
2726 static void
2727 igb_start_watchdog_timer(igb_t *igb)
2728 {
2729 	mutex_enter(&igb->watchdog_lock);
2730 
2731 	if (igb->watchdog_enable) {
2732 		if (!igb->watchdog_start) {
2733 			igb->watchdog_start = B_TRUE;
2734 			igb_arm_watchdog_timer(igb);
2735 		}
2736 	}
2737 
2738 	mutex_exit(&igb->watchdog_lock);
2739 }
2740 
2741 /*
2742  * igb_restart_watchdog_timer - Restart the driver watchdog timer
2743  */
2744 static void
2745 igb_restart_watchdog_timer(igb_t *igb)
2746 {
2747 	mutex_enter(&igb->watchdog_lock);
2748 
2749 	if (igb->watchdog_start)
2750 		igb_arm_watchdog_timer(igb);
2751 
2752 	mutex_exit(&igb->watchdog_lock);
2753 }
2754 
2755 /*
2756  * igb_stop_watchdog_timer - Stop the driver watchdog timer
2757  */
2758 static void
2759 igb_stop_watchdog_timer(igb_t *igb)
2760 {
2761 	timeout_id_t tid;
2762 
2763 	mutex_enter(&igb->watchdog_lock);
2764 
2765 	igb->watchdog_start = B_FALSE;
2766 	tid = igb->watchdog_tid;
2767 	igb->watchdog_tid = 0;
2768 
2769 	mutex_exit(&igb->watchdog_lock);
2770 
2771 	if (tid != 0)
2772 		(void) untimeout(tid);
2773 }
2774 
2775 /*
2776  * igb_disable_adapter_interrupts - Clear/disable all hardware interrupts
2777  */
2778 static void
2779 igb_disable_adapter_interrupts(igb_t *igb)
2780 {
2781 	struct e1000_hw *hw = &igb->hw;
2782 
2783 	/*
2784 	 * Set the IMC register to mask all the interrupts,
2785 	 * including the tx interrupts.
2786 	 */
2787 	E1000_WRITE_REG(hw, E1000_IMC, 0xffffffff);
2788 
2789 	/*
2790 	 * Additional disabling for MSI-X
2791 	 */
2792 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
2793 		E1000_WRITE_REG(hw, E1000_EIMC, 0xffffffff);
2794 		E1000_WRITE_REG(hw, E1000_EIAC, 0x0);
2795 	}
2796 
2797 	E1000_WRITE_FLUSH(hw);
2798 }
2799 
2800 /*
2801  * igb_enable_adapter_interrupts - Mask/enable all hardware interrupts
2802  */
2803 static void
2804 igb_enable_adapter_interrupts(igb_t *igb)
2805 {
2806 	struct e1000_hw *hw = &igb->hw;
2807 	uint32_t reg;
2808 
2809 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
2810 		/* Interrupt enabling for MSI-X */
2811 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
2812 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
2813 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
2814 
2815 		/* Enable MSI-X PBA support */
2816 		reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
2817 		reg |= E1000_CTRL_EXT_PBA_CLR;
2818 
2819 		/* Non-selective interrupt clear-on-read */
2820 		reg |= E1000_CTRL_EXT_IRCA;	/* Called NSICR in the EAS */
2821 
2822 		E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
2823 	} else {
2824 		/* Interrupt enabling for MSI and legacy */
2825 		E1000_WRITE_REG(hw, E1000_IMS, IMS_ENABLE_MASK);
2826 	}
2827 
2828 	E1000_WRITE_FLUSH(hw);
2829 }
2830 
2831 /*
2832  * Loopback Support
2833  */
2834 static lb_property_t lb_normal =
2835 	{ normal,	"normal",	IGB_LB_NONE		};
2836 static lb_property_t lb_external =
2837 	{ external,	"External",	IGB_LB_EXTERNAL		};
2838 static lb_property_t lb_mac =
2839 	{ internal,	"MAC",		IGB_LB_INTERNAL_MAC	};
2840 static lb_property_t lb_phy =
2841 	{ internal,	"PHY",		IGB_LB_INTERNAL_PHY	};
2842 static lb_property_t lb_serdes =
2843 	{ internal,	"SerDes",	IGB_LB_INTERNAL_SERDES	};
2844 
2845 enum ioc_reply
2846 igb_loopback_ioctl(igb_t *igb, struct iocblk *iocp, mblk_t *mp)
2847 {
2848 	lb_info_sz_t *lbsp;
2849 	lb_property_t *lbpp;
2850 	struct e1000_hw *hw;
2851 	uint32_t *lbmp;
2852 	uint32_t size;
2853 	uint32_t value;
2854 
2855 	hw = &igb->hw;
2856 
2857 	if (mp->b_cont == NULL)
2858 		return (IOC_INVAL);
2859 
2860 	switch (iocp->ioc_cmd) {
2861 	default:
2862 		return (IOC_INVAL);
2863 
2864 	case LB_GET_INFO_SIZE:
2865 		size = sizeof (lb_info_sz_t);
2866 		if (iocp->ioc_count != size)
2867 			return (IOC_INVAL);
2868 
2869 		value = sizeof (lb_normal);
2870 		value += sizeof (lb_mac);
2871 		if (hw->phy.media_type == e1000_media_type_copper)
2872 			value += sizeof (lb_phy);
2873 		else
2874 			value += sizeof (lb_serdes);
2875 		value += sizeof (lb_external);
2876 
2877 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
2878 		*lbsp = value;
2879 		break;
2880 
2881 	case LB_GET_INFO:
2882 		value = sizeof (lb_normal);
2883 		value += sizeof (lb_mac);
2884 		if (hw->phy.media_type == e1000_media_type_copper)
2885 			value += sizeof (lb_phy);
2886 		else
2887 			value += sizeof (lb_serdes);
2888 		value += sizeof (lb_external);
2889 
2890 		size = value;
2891 		if (iocp->ioc_count != size)
2892 			return (IOC_INVAL);
2893 
2894 		value = 0;
2895 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
2896 
2897 		lbpp[value++] = lb_normal;
2898 		lbpp[value++] = lb_mac;
2899 		if (hw->phy.media_type == e1000_media_type_copper)
2900 			lbpp[value++] = lb_phy;
2901 		else
2902 			lbpp[value++] = lb_serdes;
2903 		lbpp[value++] = lb_external;
2904 		break;
2905 
2906 	case LB_GET_MODE:
2907 		size = sizeof (uint32_t);
2908 		if (iocp->ioc_count != size)
2909 			return (IOC_INVAL);
2910 
2911 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
2912 		*lbmp = igb->loopback_mode;
2913 		break;
2914 
2915 	case LB_SET_MODE:
2916 		size = 0;
2917 		if (iocp->ioc_count != sizeof (uint32_t))
2918 			return (IOC_INVAL);
2919 
2920 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
2921 		if (!igb_set_loopback_mode(igb, *lbmp))
2922 			return (IOC_INVAL);
2923 		break;
2924 	}
2925 
2926 	iocp->ioc_count = size;
2927 	iocp->ioc_error = 0;
2928 
2929 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2930 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2931 		return (IOC_INVAL);
2932 	}
2933 
2934 	return (IOC_REPLY);
2935 }
2936 
2937 /*
2938  * igb_set_loopback_mode - Setup loopback based on the loopback mode
2939  */
2940 static boolean_t
2941 igb_set_loopback_mode(igb_t *igb, uint32_t mode)
2942 {
2943 	struct e1000_hw *hw;
2944 
2945 	if (mode == igb->loopback_mode)
2946 		return (B_TRUE);
2947 
2948 	hw = &igb->hw;
2949 
2950 	igb->loopback_mode = mode;
2951 
2952 	if (mode == IGB_LB_NONE) {
2953 		/* Reset the chip */
2954 		hw->phy.autoneg_wait_to_complete = B_TRUE;
2955 		(void) igb_reset(igb);
2956 		hw->phy.autoneg_wait_to_complete = B_FALSE;
2957 		return (B_TRUE);
2958 	}
2959 
2960 	mutex_enter(&igb->gen_lock);
2961 
2962 	switch (mode) {
2963 	default:
2964 		mutex_exit(&igb->gen_lock);
2965 		return (B_FALSE);
2966 
2967 	case IGB_LB_EXTERNAL:
2968 		igb_set_external_loopback(igb);
2969 		break;
2970 
2971 	case IGB_LB_INTERNAL_MAC:
2972 		igb_set_internal_mac_loopback(igb);
2973 		break;
2974 
2975 	case IGB_LB_INTERNAL_PHY:
2976 		igb_set_internal_phy_loopback(igb);
2977 		break;
2978 
2979 	case IGB_LB_INTERNAL_SERDES:
2980 		igb_set_internal_serdes_loopback(igb);
2981 		break;
2982 	}
2983 
2984 	mutex_exit(&igb->gen_lock);
2985 
2986 	return (B_TRUE);
2987 }
2988 
2989 /*
2990  * igb_set_external_loopback - Set the external loopback mode
2991  */
2992 static void
2993 igb_set_external_loopback(igb_t *igb)
2994 {
2995 	struct e1000_hw *hw;
2996 
2997 	hw = &igb->hw;
2998 
2999 	/* Set phy to known state */
3000 	(void) e1000_phy_hw_reset(hw);
3001 
3002 	(void) e1000_write_phy_reg(hw, 0x0, 0x0140);
3003 	(void) e1000_write_phy_reg(hw, 0x9, 0x1b00);
3004 	(void) e1000_write_phy_reg(hw, 0x12, 0x1610);
3005 	(void) e1000_write_phy_reg(hw, 0x1f37, 0x3f1c);
3006 }
3007 
3008 /*
3009  * igb_set_internal_mac_loopback - Set the internal MAC loopback mode
3010  */
3011 static void
3012 igb_set_internal_mac_loopback(igb_t *igb)
3013 {
3014 	struct e1000_hw *hw;
3015 	uint32_t ctrl;
3016 	uint32_t rctl;
3017 
3018 	hw = &igb->hw;
3019 
3020 	/* Set the Receive Control register */
3021 	rctl = E1000_READ_REG(hw, E1000_RCTL);
3022 	rctl &= ~E1000_RCTL_LBM_TCVR;
3023 	rctl |= E1000_RCTL_LBM_MAC;
3024 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
3025 
3026 	/* Set the Device Control register */
3027 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3028 	ctrl &= ~E1000_CTRL_SPD_SEL;	/* Clear the speed sel bits */
3029 	ctrl |= (E1000_CTRL_SLU |	/* Force link up */
3030 	    E1000_CTRL_FRCSPD |		/* Force speed */
3031 	    E1000_CTRL_FRCDPX |		/* Force duplex */
3032 	    E1000_CTRL_SPD_1000 |	/* Force speed to 1000 */
3033 	    E1000_CTRL_FD);		/* Force full duplex */
3034 	ctrl &= ~E1000_CTRL_ILOS;	/* Clear ILOS when there's a link */
3035 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3036 }
3037 
3038 /*
3039  * igb_set_internal_phy_loopback - Set the internal PHY loopback mode
3040  */
3041 static void
3042 igb_set_internal_phy_loopback(igb_t *igb)
3043 {
3044 	struct e1000_hw *hw;
3045 	uint32_t ctrl_ext;
3046 	uint16_t phy_ctrl;
3047 	uint16_t phy_pconf;
3048 
3049 	hw = &igb->hw;
3050 
3051 	/* Set link mode to PHY (00b) in the Extended Control register */
3052 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3053 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3054 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3055 
3056 	/*
3057 	 * Set PHY control register (0x4140):
3058 	 *    Set full duplex mode
3059 	 *    Set loopback bit
3060 	 *    Clear auto-neg enable bit
3061 	 *    Set PHY speed
3062 	 */
3063 	phy_ctrl = MII_CR_FULL_DUPLEX | MII_CR_SPEED_1000 | MII_CR_LOOPBACK;
3064 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3065 
3066 	/* Set the link disable bit in the Port Configuration register */
3067 	(void) e1000_read_phy_reg(hw, 0x10, &phy_pconf);
3068 	phy_pconf |= (uint16_t)1 << 14;
3069 	(void) e1000_write_phy_reg(hw, 0x10, phy_pconf);
3070 }
3071 
3072 /*
3073  * igb_set_internal_serdes_loopback - Set the internal SerDes loopback mode
3074  */
3075 static void
3076 igb_set_internal_serdes_loopback(igb_t *igb)
3077 {
3078 	struct e1000_hw *hw;
3079 	uint32_t ctrl_ext;
3080 	uint32_t ctrl;
3081 	uint32_t pcs_lctl;
3082 	uint32_t connsw;
3083 
3084 	hw = &igb->hw;
3085 
3086 	/* Set link mode to SerDes (11b) in the Extended Control register */
3087 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3088 	ctrl_ext |= E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
3089 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3090 
3091 	/* Configure the SerDes to loopback */
3092 	E1000_WRITE_REG(hw, E1000_SCTL, 0x410);
3093 
3094 	/* Set Device Control register */
3095 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3096 	ctrl |= (E1000_CTRL_FD |	/* Force full duplex */
3097 	    E1000_CTRL_SLU);		/* Force link up */
3098 	ctrl &= ~(E1000_CTRL_RFCE |	/* Disable receive flow control */
3099 	    E1000_CTRL_TFCE |		/* Disable transmit flow control */
3100 	    E1000_CTRL_LRST);		/* Clear link reset */
3101 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3102 
3103 	/* Set PCS Link Control register */
3104 	pcs_lctl = E1000_READ_REG(hw, E1000_PCS_LCTL);
3105 	pcs_lctl |= (E1000_PCS_LCTL_FORCE_LINK |
3106 	    E1000_PCS_LCTL_FSD |
3107 	    E1000_PCS_LCTL_FDV_FULL |
3108 	    E1000_PCS_LCTL_FLV_LINK_UP);
3109 	pcs_lctl &= ~E1000_PCS_LCTL_AN_ENABLE;
3110 	E1000_WRITE_REG(hw, E1000_PCS_LCTL, pcs_lctl);
3111 
3112 	/* Set the Copper/Fiber Switch Control - CONNSW register */
3113 	connsw = E1000_READ_REG(hw, E1000_CONNSW);
3114 	connsw &= ~E1000_CONNSW_ENRGSRC;
3115 	E1000_WRITE_REG(hw, E1000_CONNSW, connsw);
3116 }
3117 
3118 #pragma inline(igb_intr_rx_work)
3119 /*
3120  * igb_intr_rx_work - rx processing of ISR
3121  */
3122 static void
3123 igb_intr_rx_work(igb_rx_ring_t *rx_ring)
3124 {
3125 	mblk_t *mp;
3126 
3127 	mutex_enter(&rx_ring->rx_lock);
3128 	mp = igb_rx(rx_ring);
3129 	mutex_exit(&rx_ring->rx_lock);
3130 
3131 	if (mp != NULL)
3132 		mac_rx(rx_ring->igb->mac_hdl, NULL, mp);
3133 }
3134 
3135 #pragma inline(igb_intr_tx_work)
3136 /*
3137  * igb_intr_tx_work - tx processing of ISR
3138  */
3139 static void
3140 igb_intr_tx_work(igb_tx_ring_t *tx_ring)
3141 {
3142 	/* Recycle the tx descriptors */
3143 	tx_ring->tx_recycle(tx_ring);
3144 
3145 	/* Schedule the re-transmit */
3146 	if (tx_ring->reschedule &&
3147 	    (tx_ring->tbd_free >= tx_ring->resched_thresh)) {
3148 		tx_ring->reschedule = B_FALSE;
3149 		mac_tx_update(tx_ring->igb->mac_hdl);
3150 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3151 	}
3152 }
3153 
3154 #pragma inline(igb_intr_other_work)
3155 /*
3156  * igb_intr_other_work - other processing of ISR
3157  */
3158 static void
3159 igb_intr_other_work(igb_t *igb)
3160 {
3161 	boolean_t link_changed;
3162 
3163 	igb_stop_watchdog_timer(igb);
3164 
3165 	mutex_enter(&igb->gen_lock);
3166 
3167 	/*
3168 	 * Because we got a link-status-change interrupt, force
3169 	 * e1000_check_for_link() to look at phy
3170 	 */
3171 	igb->hw.mac.get_link_status = B_TRUE;
3172 
3173 	/* igb_link_check takes care of link status change */
3174 	link_changed = igb_link_check(igb);
3175 
3176 	/* Get new phy state */
3177 	igb_get_phy_state(igb);
3178 
3179 	mutex_exit(&igb->gen_lock);
3180 
3181 	if (link_changed)
3182 		mac_link_update(igb->mac_hdl, igb->link_state);
3183 
3184 	igb_start_watchdog_timer(igb);
3185 }
3186 
3187 /*
3188  * igb_intr_legacy - Interrupt handler for legacy interrupts
3189  */
3190 static uint_t
3191 igb_intr_legacy(void *arg1, void *arg2)
3192 {
3193 	igb_t *igb = (igb_t *)arg1;
3194 	igb_tx_ring_t *tx_ring;
3195 	uint32_t icr;
3196 	mblk_t *mp;
3197 	boolean_t tx_reschedule;
3198 	boolean_t link_changed;
3199 	uint_t result;
3200 
3201 	_NOTE(ARGUNUSED(arg2));
3202 
3203 	mutex_enter(&igb->gen_lock);
3204 
3205 	if (igb->igb_state & IGB_SUSPENDED) {
3206 		mutex_exit(&igb->gen_lock);
3207 		return (DDI_INTR_UNCLAIMED);
3208 	}
3209 
3210 	mp = NULL;
3211 	tx_reschedule = B_FALSE;
3212 	link_changed = B_FALSE;
3213 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3214 
3215 	if (icr & E1000_ICR_INT_ASSERTED) {
3216 		/*
3217 		 * E1000_ICR_INT_ASSERTED bit was set:
3218 		 * Read(Clear) the ICR, claim this interrupt,
3219 		 * look for work to do.
3220 		 */
3221 		ASSERT(igb->num_rx_rings == 1);
3222 		ASSERT(igb->num_tx_rings == 1);
3223 
3224 		if (icr & E1000_ICR_RXT0) {
3225 			mp = igb_rx(&igb->rx_rings[0]);
3226 		}
3227 
3228 		if (icr & E1000_ICR_TXDW) {
3229 			tx_ring = &igb->tx_rings[0];
3230 
3231 			/* Recycle the tx descriptors */
3232 			tx_ring->tx_recycle(tx_ring);
3233 
3234 			/* Schedule the re-transmit */
3235 			tx_reschedule = (tx_ring->reschedule &&
3236 			    (tx_ring->tbd_free >= tx_ring->resched_thresh));
3237 		}
3238 
3239 		if (icr & E1000_ICR_LSC) {
3240 			/*
3241 			 * Because we got a link-status-change interrupt, force
3242 			 * e1000_check_for_link() to look at phy
3243 			 */
3244 			igb->hw.mac.get_link_status = B_TRUE;
3245 
3246 			/* igb_link_check takes care of link status change */
3247 			link_changed = igb_link_check(igb);
3248 
3249 			/* Get new phy state */
3250 			igb_get_phy_state(igb);
3251 		}
3252 
3253 		result = DDI_INTR_CLAIMED;
3254 	} else {
3255 		/*
3256 		 * E1000_ICR_INT_ASSERTED bit was not set:
3257 		 * Don't claim this interrupt.
3258 		 */
3259 		result = DDI_INTR_UNCLAIMED;
3260 	}
3261 
3262 	mutex_exit(&igb->gen_lock);
3263 
3264 	/*
3265 	 * Do the following work outside of the gen_lock
3266 	 */
3267 	if (mp != NULL)
3268 		mac_rx(igb->mac_hdl, NULL, mp);
3269 
3270 	if (tx_reschedule)  {
3271 		tx_ring->reschedule = B_FALSE;
3272 		mac_tx_update(igb->mac_hdl);
3273 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3274 	}
3275 
3276 	if (link_changed)
3277 		mac_link_update(igb->mac_hdl, igb->link_state);
3278 
3279 	return (result);
3280 }
3281 
3282 /*
3283  * igb_intr_msi - Interrupt handler for MSI
3284  */
3285 static uint_t
3286 igb_intr_msi(void *arg1, void *arg2)
3287 {
3288 	igb_t *igb = (igb_t *)arg1;
3289 	uint32_t icr;
3290 
3291 	_NOTE(ARGUNUSED(arg2));
3292 
3293 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3294 
3295 	/*
3296 	 * For MSI interrupt, we have only one vector,
3297 	 * so we have only one rx ring and one tx ring enabled.
3298 	 */
3299 	ASSERT(igb->num_rx_rings == 1);
3300 	ASSERT(igb->num_tx_rings == 1);
3301 
3302 	if (icr & E1000_ICR_RXT0) {
3303 		igb_intr_rx_work(&igb->rx_rings[0]);
3304 	}
3305 
3306 	if (icr & E1000_ICR_TXDW) {
3307 		igb_intr_tx_work(&igb->tx_rings[0]);
3308 	}
3309 
3310 	if (icr & E1000_ICR_LSC) {
3311 		igb_intr_other_work(igb);
3312 	}
3313 
3314 	return (DDI_INTR_CLAIMED);
3315 }
3316 
3317 /*
3318  * igb_intr_rx - Interrupt handler for rx
3319  */
3320 static uint_t
3321 igb_intr_rx(void *arg1, void *arg2)
3322 {
3323 	igb_rx_ring_t *rx_ring = (igb_rx_ring_t *)arg1;
3324 
3325 	_NOTE(ARGUNUSED(arg2));
3326 
3327 	/*
3328 	 * Only used via MSI-X vector so don't check cause bits
3329 	 * and only clean the given ring.
3330 	 */
3331 	igb_intr_rx_work(rx_ring);
3332 
3333 	return (DDI_INTR_CLAIMED);
3334 }
3335 
3336 /*
3337  * igb_intr_tx_other - Interrupt handler for both tx and other
3338  *
3339  * Always look for Tx cleanup work.  Only look for other work if the right
3340  * bits are set in the Interrupt Cause Register.
3341  */
3342 static uint_t
3343 igb_intr_tx_other(void *arg1, void *arg2)
3344 {
3345 	igb_t *igb = (igb_t *)arg1;
3346 	uint32_t icr;
3347 
3348 	_NOTE(ARGUNUSED(arg2));
3349 
3350 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3351 
3352 	/*
3353 	 * Always look for Tx cleanup work.  We don't have separate
3354 	 * transmit vectors, so we have only one tx ring enabled.
3355 	 */
3356 	ASSERT(igb->num_tx_rings == 1);
3357 	igb_intr_tx_work(&igb->tx_rings[0]);
3358 
3359 	/*
3360 	 * Check for "other" causes.
3361 	 */
3362 	if (icr & E1000_ICR_LSC) {
3363 		igb_intr_other_work(igb);
3364 	}
3365 
3366 	return (DDI_INTR_CLAIMED);
3367 }
3368 
3369 /*
3370  * igb_alloc_intrs - Allocate interrupts for the driver
3371  *
3372  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
3373  * if not successful, try Legacy.
3374  * igb->intr_force can be used to force sequence to start with
3375  * any of the 3 types.
3376  * If MSI-X is not used, number of tx/rx rings is forced to 1.
3377  */
3378 static int
3379 igb_alloc_intrs(igb_t *igb)
3380 {
3381 	dev_info_t *devinfo;
3382 	int intr_types;
3383 	int rc;
3384 
3385 	devinfo = igb->dip;
3386 
3387 	/* Get supported interrupt types */
3388 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
3389 
3390 	if (rc != DDI_SUCCESS) {
3391 		igb_log(igb,
3392 		    "Get supported interrupt types failed: %d", rc);
3393 		return (IGB_FAILURE);
3394 	}
3395 	IGB_DEBUGLOG_1(igb, "Supported interrupt types: %x", intr_types);
3396 
3397 	igb->intr_type = 0;
3398 
3399 	/* Install MSI-X interrupts */
3400 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
3401 	    (igb->intr_force <= IGB_INTR_MSIX)) {
3402 		rc = igb_alloc_intrs_msix(igb);
3403 
3404 		if (rc == IGB_SUCCESS)
3405 			return (IGB_SUCCESS);
3406 
3407 		igb_log(igb,
3408 		    "Allocate MSI-X failed, trying MSI interrupts...");
3409 	}
3410 
3411 	/* MSI-X not used, force rings to 1 */
3412 	igb->num_rx_rings = 1;
3413 	igb->num_tx_rings = 1;
3414 	igb_log(igb,
3415 	    "MSI-X not used, force rx and tx queue number to 1");
3416 
3417 	/* Install MSI interrupts */
3418 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
3419 	    (igb->intr_force <= IGB_INTR_MSI)) {
3420 		rc = igb_alloc_intrs_msi(igb);
3421 
3422 		if (rc == IGB_SUCCESS)
3423 			return (IGB_SUCCESS);
3424 
3425 		igb_log(igb,
3426 		    "Allocate MSI failed, trying Legacy interrupts...");
3427 	}
3428 
3429 	/* Install legacy interrupts */
3430 	if (intr_types & DDI_INTR_TYPE_FIXED) {
3431 		rc = igb_alloc_intrs_legacy(igb);
3432 
3433 		if (rc == IGB_SUCCESS)
3434 			return (IGB_SUCCESS);
3435 
3436 		igb_log(igb,
3437 		    "Allocate Legacy interrupts failed");
3438 	}
3439 
3440 	/* If none of the 3 types succeeded, return failure */
3441 	return (IGB_FAILURE);
3442 }
3443 
3444 /*
3445  * igb_alloc_intrs_msix - Allocate the MSIX interrupts
3446  *
3447  * If fewer than 2 vectors are available, return failure.
3448  * Upon success, this sets the number of Rx rings to a number that
3449  * matches the vectors available for Rx interrupts.
3450  */
3451 static int
3452 igb_alloc_intrs_msix(igb_t *igb)
3453 {
3454 	dev_info_t *devinfo;
3455 	int request, count, avail, actual;
3456 	int rx_rings;
3457 	int rc;
3458 
3459 	devinfo = igb->dip;
3460 
3461 	/*
3462 	 * Currently only 1 tx ring is supported. More tx rings
3463 	 * will be supported with future enhancement.
3464 	 */
3465 	if (igb->num_tx_rings > 1) {
3466 		igb->num_tx_rings = 1;
3467 		igb_log(igb,
3468 		    "Use only 1 MSI-X vector for tx, "
3469 		    "force tx queue number to 1");
3470 	}
3471 
3472 	/*
3473 	 * Best number of vectors for the adapter is
3474 	 * # rx rings + # tx rings + 1 for other
3475 	 * But currently we only support number of vectors of
3476 	 * # rx rings + 1 for tx & other
3477 	 */
3478 	request = igb->num_rx_rings + 1;
3479 	IGB_DEBUGLOG_1(igb, "MSI-X interrupts requested: %d", request);
3480 
3481 	/* Get number of supported interrupts */
3482 	rc = ddi_intr_get_nintrs(devinfo, DDI_INTR_TYPE_MSIX, &count);
3483 	if ((rc != DDI_SUCCESS) || (count == 0)) {
3484 		igb_log(igb,
3485 		    "Get interrupt number failed. Return: %d, count: %d",
3486 		    rc, count);
3487 		return (IGB_FAILURE);
3488 	}
3489 	IGB_DEBUGLOG_1(igb, "MSI-X interrupts supported: %d", count);
3490 
3491 	/* Get number of available interrupts */
3492 	rc = ddi_intr_get_navail(devinfo, DDI_INTR_TYPE_MSIX, &avail);
3493 	if ((rc != DDI_SUCCESS) || (avail == 0)) {
3494 		igb_log(igb,
3495 		    "Get interrupt available number failed. "
3496 		    "Return: %d, available: %d", rc, avail);
3497 		return (IGB_FAILURE);
3498 	}
3499 	IGB_DEBUGLOG_1(igb, "MSI-X interrupts available: %d", avail);
3500 
3501 	if (avail < request) {
3502 		igb_log(igb,
3503 		    "Request %d MSI-X vectors, %d available",
3504 		    request, avail);
3505 		request = avail;
3506 	}
3507 
3508 	actual = 0;
3509 	igb->intr_cnt = 0;
3510 
3511 	/* Allocate an array of interrupt handles */
3512 	igb->intr_size = request * sizeof (ddi_intr_handle_t);
3513 	igb->htable = kmem_alloc(igb->intr_size, KM_SLEEP);
3514 
3515 	/* Call ddi_intr_alloc() */
3516 	rc = ddi_intr_alloc(devinfo, igb->htable, DDI_INTR_TYPE_MSIX, 0,
3517 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
3518 	if (rc != DDI_SUCCESS) {
3519 		igb_log(igb, "Allocate MSI-X interrupts failed. "
3520 		    "return: %d, request: %d, actual: %d",
3521 		    rc, request, actual);
3522 		goto alloc_msix_fail;
3523 	}
3524 	IGB_DEBUGLOG_1(igb, "MSI-X interrupts actually allocated: %d", actual);
3525 
3526 	igb->intr_cnt = actual;
3527 
3528 	/*
3529 	 * Now we know the actual number of vectors.  Here we assume that
3530 	 * tx and other will share 1 vector and all remaining (must be at
3531 	 * least 1 remaining) will be used for rx.
3532 	 */
3533 	if (actual < 2) {
3534 		igb_log(igb, "Insufficient MSI-X interrupts available: %d",
3535 		    actual);
3536 		goto alloc_msix_fail;
3537 	}
3538 
3539 	rx_rings = actual - 1;
3540 	if (rx_rings < igb->num_rx_rings) {
3541 		igb_log(igb, "MSI-X vectors force Rx queue number to %d",
3542 		    rx_rings);
3543 		igb->num_rx_rings = rx_rings;
3544 	}
3545 
3546 	/* Get priority for first vector, assume remaining are all the same */
3547 	rc = ddi_intr_get_pri(igb->htable[0], &igb->intr_pri);
3548 	if (rc != DDI_SUCCESS) {
3549 		igb_log(igb,
3550 		    "Get interrupt priority failed: %d", rc);
3551 		goto alloc_msix_fail;
3552 	}
3553 
3554 	rc = ddi_intr_get_cap(igb->htable[0], &igb->intr_cap);
3555 	if (rc != DDI_SUCCESS) {
3556 		igb_log(igb,
3557 		    "Get interrupt cap failed: %d", rc);
3558 		goto alloc_msix_fail;
3559 	}
3560 
3561 	igb->intr_type = DDI_INTR_TYPE_MSIX;
3562 
3563 	return (IGB_SUCCESS);
3564 
3565 alloc_msix_fail:
3566 	igb_rem_intrs(igb);
3567 
3568 	return (IGB_FAILURE);
3569 }
3570 
3571 /*
3572  * igb_alloc_intrs_msi - Allocate the MSI interrupts
3573  */
3574 static int
3575 igb_alloc_intrs_msi(igb_t *igb)
3576 {
3577 	dev_info_t *devinfo;
3578 	int request, count, avail, actual;
3579 	int rc;
3580 
3581 	devinfo = igb->dip;
3582 
3583 	/* Request 1 MSI interrupt vector */
3584 	request = 1;
3585 	IGB_DEBUGLOG_1(igb, "MSI interrupts requested: %d", request);
3586 
3587 	/* Get number of supported interrupts */
3588 	rc = ddi_intr_get_nintrs(devinfo, DDI_INTR_TYPE_MSI, &count);
3589 	if ((rc != DDI_SUCCESS) || (count == 0)) {
3590 		igb_log(igb,
3591 		    "Get MSI supported number failed. Return: %d, count: %d",
3592 		    rc, count);
3593 		return (IGB_FAILURE);
3594 	}
3595 	IGB_DEBUGLOG_1(igb, "MSI interrupts supported: %d", count);
3596 
3597 	/* Get number of available interrupts */
3598 	rc = ddi_intr_get_navail(devinfo, DDI_INTR_TYPE_MSI, &avail);
3599 	if ((rc != DDI_SUCCESS) || (avail == 0)) {
3600 		igb_log(igb,
3601 		    "Get MSI available number failed. "
3602 		    "Return: %d, available: %d", rc, avail);
3603 		return (IGB_FAILURE);
3604 	}
3605 	IGB_DEBUGLOG_1(igb, "MSI interrupts available: %d", avail);
3606 
3607 	actual = 0;
3608 	igb->intr_cnt = 0;
3609 
3610 	/* Allocate an array of interrupt handles */
3611 	igb->intr_size = request * sizeof (ddi_intr_handle_t);
3612 	igb->htable = kmem_alloc(igb->intr_size, KM_SLEEP);
3613 
3614 	/* Call ddi_intr_alloc() */
3615 	rc = ddi_intr_alloc(devinfo, igb->htable, DDI_INTR_TYPE_MSI, 0,
3616 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
3617 	if ((rc != DDI_SUCCESS) || (actual == 0)) {
3618 		igb_log(igb,
3619 		    "Allocate MSI interrupts failed: %d", rc);
3620 		goto alloc_msi_fail;
3621 	}
3622 
3623 	ASSERT(actual == 1);
3624 	igb->intr_cnt = actual;
3625 
3626 	/* Get priority for first msi, assume remaining are all the same */
3627 	rc = ddi_intr_get_pri(igb->htable[0], &igb->intr_pri);
3628 	if (rc != DDI_SUCCESS) {
3629 		igb_log(igb,
3630 		    "Get interrupt priority failed: %d", rc);
3631 		goto alloc_msi_fail;
3632 	}
3633 
3634 	rc = ddi_intr_get_cap(igb->htable[0], &igb->intr_cap);
3635 	if (rc != DDI_SUCCESS) {
3636 		igb_log(igb,
3637 		    "Get interrupt cap failed: %d\n", rc);
3638 		goto alloc_msi_fail;
3639 
3640 	}
3641 
3642 	igb->intr_type = DDI_INTR_TYPE_MSI;
3643 
3644 	return (IGB_SUCCESS);
3645 
3646 alloc_msi_fail:
3647 	igb_rem_intrs(igb);
3648 
3649 	return (IGB_FAILURE);
3650 }
3651 
3652 /*
3653  * igb_alloc_intrs_legacy - Allocate the Legacy interrupts
3654  */
3655 static int
3656 igb_alloc_intrs_legacy(igb_t *igb)
3657 {
3658 	dev_info_t *devinfo;
3659 	int request, count, avail, actual;
3660 	int rc;
3661 
3662 	devinfo = igb->dip;
3663 
3664 	/* Request 1 Legacy interrupt vector */
3665 	request = 1;
3666 	IGB_DEBUGLOG_1(igb, "Legacy interrupts requested: %d", request);
3667 
3668 	/* Get number of supported interrupts */
3669 	rc = ddi_intr_get_nintrs(devinfo, DDI_INTR_TYPE_FIXED, &count);
3670 	if ((rc != DDI_SUCCESS) || (count == 0)) {
3671 		igb_log(igb,
3672 		    "Get Legacy supported number failed. Return: %d, count: %d",
3673 		    rc, count);
3674 		return (IGB_FAILURE);
3675 	}
3676 	IGB_DEBUGLOG_1(igb, "Legacy interrupts supported: %d", count);
3677 
3678 	/* Get number of available interrupts */
3679 	rc = ddi_intr_get_navail(devinfo, DDI_INTR_TYPE_FIXED, &avail);
3680 	if ((rc != DDI_SUCCESS) || (avail == 0)) {
3681 		igb_log(igb,
3682 		    "Get Legacy available number failed. "
3683 		    "Return: %d, available: %d", rc, avail);
3684 		return (IGB_FAILURE);
3685 	}
3686 	IGB_DEBUGLOG_1(igb, "Legacy interrupts available: %d", avail);
3687 
3688 	actual = 0;
3689 	igb->intr_cnt = 0;
3690 
3691 	/* Allocate an array of interrupt handles */
3692 	igb->intr_size = request * sizeof (ddi_intr_handle_t);
3693 	igb->htable = kmem_alloc(igb->intr_size, KM_SLEEP);
3694 
3695 	/* Call ddi_intr_alloc() */
3696 	rc = ddi_intr_alloc(devinfo, igb->htable, DDI_INTR_TYPE_FIXED, 0,
3697 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
3698 	if ((rc != DDI_SUCCESS) || (actual == 0)) {
3699 		igb_log(igb,
3700 		    "Allocate Legacy interrupts failed: %d", rc);
3701 		goto alloc_legacy_fail;
3702 	}
3703 
3704 	ASSERT(actual == 1);
3705 	igb->intr_cnt = actual;
3706 
3707 	/* Get priority for first msi, assume remaining are all the same */
3708 	rc = ddi_intr_get_pri(igb->htable[0], &igb->intr_pri);
3709 	if (rc != DDI_SUCCESS) {
3710 		igb_log(igb,
3711 		    "Get interrupt priority failed: %d", rc);
3712 		goto alloc_legacy_fail;
3713 	}
3714 
3715 	rc = ddi_intr_get_cap(igb->htable[0], &igb->intr_cap);
3716 	if (rc != DDI_SUCCESS) {
3717 		igb_log(igb,
3718 		    "Get interrupt cap failed: %d\n", rc);
3719 		goto alloc_legacy_fail;
3720 	}
3721 
3722 	igb->intr_type = DDI_INTR_TYPE_FIXED;
3723 
3724 	return (IGB_SUCCESS);
3725 
3726 alloc_legacy_fail:
3727 	igb_rem_intrs(igb);
3728 
3729 	return (IGB_FAILURE);
3730 }
3731 
3732 /*
3733  * igb_add_intr_handlers - Add interrupt handlers based on the interrupt type
3734  *
3735  * Before adding the interrupt handlers, the interrupt vectors have
3736  * been allocated, and the rx/tx rings have also been allocated.
3737  */
3738 static int
3739 igb_add_intr_handlers(igb_t *igb)
3740 {
3741 	igb_rx_ring_t *rx_ring;
3742 	int vector;
3743 	int rc;
3744 	int i;
3745 
3746 	vector = 0;
3747 
3748 	switch (igb->intr_type) {
3749 	case DDI_INTR_TYPE_MSIX:
3750 		/* Add interrupt handler for tx + other */
3751 		rc = ddi_intr_add_handler(igb->htable[vector],
3752 		    (ddi_intr_handler_t *)igb_intr_tx_other,
3753 		    (void *)igb, NULL);
3754 		if (rc != DDI_SUCCESS) {
3755 			igb_log(igb,
3756 			    "Add tx/other interrupt handler failed: %d", rc);
3757 			return (IGB_FAILURE);
3758 		}
3759 		vector++;
3760 
3761 		/* Add interrupt handler for each rx ring */
3762 		for (i = 0; i < igb->num_rx_rings; i++) {
3763 			rx_ring = &igb->rx_rings[i];
3764 
3765 			rc = ddi_intr_add_handler(igb->htable[vector],
3766 			    (ddi_intr_handler_t *)igb_intr_rx,
3767 			    (void *)rx_ring, NULL);
3768 
3769 			if (rc != DDI_SUCCESS) {
3770 				igb_log(igb,
3771 				    "Add rx interrupt handler failed. "
3772 				    "return: %d, rx ring: %d", rc, i);
3773 				for (vector--; vector >= 0; vector--) {
3774 					(void) ddi_intr_remove_handler(
3775 					    igb->htable[vector]);
3776 				}
3777 				return (IGB_FAILURE);
3778 			}
3779 
3780 			rx_ring->intr_vector = vector;
3781 
3782 			vector++;
3783 		}
3784 		break;
3785 
3786 	case DDI_INTR_TYPE_MSI:
3787 		/* Add interrupt handlers for the only vector */
3788 		rc = ddi_intr_add_handler(igb->htable[vector],
3789 		    (ddi_intr_handler_t *)igb_intr_msi,
3790 		    (void *)igb, NULL);
3791 
3792 		if (rc != DDI_SUCCESS) {
3793 			igb_log(igb,
3794 			    "Add MSI interrupt handler failed: %d", rc);
3795 			return (IGB_FAILURE);
3796 		}
3797 
3798 		rx_ring = &igb->rx_rings[0];
3799 		rx_ring->intr_vector = vector;
3800 
3801 		vector++;
3802 		break;
3803 
3804 	case DDI_INTR_TYPE_FIXED:
3805 		/* Add interrupt handlers for the only vector */
3806 		rc = ddi_intr_add_handler(igb->htable[vector],
3807 		    (ddi_intr_handler_t *)igb_intr_legacy,
3808 		    (void *)igb, NULL);
3809 
3810 		if (rc != DDI_SUCCESS) {
3811 			igb_log(igb,
3812 			    "Add legacy interrupt handler failed: %d", rc);
3813 			return (IGB_FAILURE);
3814 		}
3815 
3816 		rx_ring = &igb->rx_rings[0];
3817 		rx_ring->intr_vector = vector;
3818 
3819 		vector++;
3820 		break;
3821 
3822 	default:
3823 		return (IGB_FAILURE);
3824 	}
3825 
3826 	ASSERT(vector == igb->intr_cnt);
3827 
3828 	return (IGB_SUCCESS);
3829 }
3830 
3831 /*
3832  * igb_setup_adapter_msix - setup the adapter to use MSI-X interrupts
3833  *
3834  * For each vector enabled on the adapter, Set the MSIXBM register accordingly
3835  */
3836 static void
3837 igb_setup_adapter_msix(igb_t *igb)
3838 {
3839 	uint32_t eims = 0;
3840 	int i, vector;
3841 	struct e1000_hw *hw = &igb->hw;
3842 
3843 	/*
3844 	 * Set vector for Tx + Other causes
3845 	 * NOTE assumption that there is only one of these and it is vector 0
3846 	 */
3847 	vector = 0;
3848 	igb->eims_mask = E1000_EICR_TX_QUEUE0 | E1000_EICR_OTHER;
3849 	E1000_WRITE_REG(hw, E1000_MSIXBM(vector), igb->eims_mask);
3850 
3851 	vector++;
3852 	for (i = 0; i < igb->num_rx_rings; i++) {
3853 		/*
3854 		 * Set vector for each rx ring
3855 		 */
3856 		eims = (E1000_EICR_RX_QUEUE0 << i);
3857 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
3858 
3859 		/*
3860 		 * Accumulate bits to enable in igb_enable_adapter_interrupts()
3861 		 */
3862 		igb->eims_mask |= eims;
3863 
3864 		vector++;
3865 	}
3866 
3867 	ASSERT(vector == igb->intr_cnt);
3868 
3869 	/*
3870 	 * Disable IAM for ICR interrupt bits
3871 	 */
3872 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3873 	E1000_WRITE_FLUSH(hw);
3874 }
3875 
3876 /*
3877  * igb_rem_intr_handlers - remove the interrupt handlers
3878  */
3879 static void
3880 igb_rem_intr_handlers(igb_t *igb)
3881 {
3882 	int i;
3883 	int rc;
3884 
3885 	for (i = 0; i < igb->intr_cnt; i++) {
3886 		rc = ddi_intr_remove_handler(igb->htable[i]);
3887 		if (rc != DDI_SUCCESS) {
3888 			IGB_DEBUGLOG_1(igb,
3889 			    "Remove intr handler failed: %d", rc);
3890 		}
3891 	}
3892 }
3893 
3894 /*
3895  * igb_rem_intrs - remove the allocated interrupts
3896  */
3897 static void
3898 igb_rem_intrs(igb_t *igb)
3899 {
3900 	int i;
3901 	int rc;
3902 
3903 	for (i = 0; i < igb->intr_cnt; i++) {
3904 		rc = ddi_intr_free(igb->htable[i]);
3905 		if (rc != DDI_SUCCESS) {
3906 			IGB_DEBUGLOG_1(igb,
3907 			    "Free intr failed: %d", rc);
3908 		}
3909 	}
3910 
3911 	kmem_free(igb->htable, igb->intr_size);
3912 	igb->htable = NULL;
3913 }
3914 
3915 /*
3916  * igb_enable_intrs - enable all the ddi interrupts
3917  */
3918 static int
3919 igb_enable_intrs(igb_t *igb)
3920 {
3921 	int i;
3922 	int rc;
3923 
3924 	/* Enable interrupts */
3925 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
3926 		/* Call ddi_intr_block_enable() for MSI */
3927 		rc = ddi_intr_block_enable(igb->htable, igb->intr_cnt);
3928 		if (rc != DDI_SUCCESS) {
3929 			igb_log(igb,
3930 			    "Enable block intr failed: %d", rc);
3931 			return (IGB_FAILURE);
3932 		}
3933 	} else {
3934 		/* Call ddi_intr_enable() for Legacy/MSI non block enable */
3935 		for (i = 0; i < igb->intr_cnt; i++) {
3936 			rc = ddi_intr_enable(igb->htable[i]);
3937 			if (rc != DDI_SUCCESS) {
3938 				igb_log(igb,
3939 				    "Enable intr failed: %d", rc);
3940 				return (IGB_FAILURE);
3941 			}
3942 		}
3943 	}
3944 
3945 	return (IGB_SUCCESS);
3946 }
3947 
3948 /*
3949  * igb_disable_intrs - disable all the ddi interrupts
3950  */
3951 static int
3952 igb_disable_intrs(igb_t *igb)
3953 {
3954 	int i;
3955 	int rc;
3956 
3957 	/* Disable all interrupts */
3958 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
3959 		rc = ddi_intr_block_disable(igb->htable, igb->intr_cnt);
3960 		if (rc != DDI_SUCCESS) {
3961 			igb_log(igb,
3962 			    "Disable block intr failed: %d", rc);
3963 			return (IGB_FAILURE);
3964 		}
3965 	} else {
3966 		for (i = 0; i < igb->intr_cnt; i++) {
3967 			rc = ddi_intr_disable(igb->htable[i]);
3968 			if (rc != DDI_SUCCESS) {
3969 				igb_log(igb,
3970 				    "Disable intr failed: %d", rc);
3971 				return (IGB_FAILURE);
3972 			}
3973 		}
3974 	}
3975 
3976 	return (IGB_SUCCESS);
3977 }
3978 
3979 /*
3980  * igb_get_phy_state - Get and save the parameters read from PHY registers
3981  */
3982 static void
3983 igb_get_phy_state(igb_t *igb)
3984 {
3985 	struct e1000_hw *hw = &igb->hw;
3986 	uint16_t phy_ctrl;
3987 	uint16_t phy_status;
3988 	uint16_t phy_an_adv;
3989 	uint16_t phy_an_exp;
3990 	uint16_t phy_ext_status;
3991 	uint16_t phy_1000t_ctrl;
3992 	uint16_t phy_1000t_status;
3993 	uint16_t phy_lp_able;
3994 
3995 	ASSERT(mutex_owned(&igb->gen_lock));
3996 
3997 	(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
3998 	(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
3999 	(void) e1000_read_phy_reg(hw, PHY_AUTONEG_ADV, &phy_an_adv);
4000 	(void) e1000_read_phy_reg(hw, PHY_AUTONEG_EXP, &phy_an_exp);
4001 	(void) e1000_read_phy_reg(hw, PHY_EXT_STATUS, &phy_ext_status);
4002 	(void) e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_1000t_ctrl);
4003 	(void) e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_1000t_status);
4004 	(void) e1000_read_phy_reg(hw, PHY_LP_ABILITY, &phy_lp_able);
4005 
4006 	igb->param_autoneg_cap =
4007 	    (phy_status & MII_SR_AUTONEG_CAPS) ? 1 : 0;
4008 	igb->param_pause_cap =
4009 	    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4010 	igb->param_asym_pause_cap =
4011 	    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4012 	igb->param_1000fdx_cap = ((phy_ext_status & IEEE_ESR_1000T_FD_CAPS) ||
4013 	    (phy_ext_status & IEEE_ESR_1000X_FD_CAPS)) ? 1 : 0;
4014 	igb->param_1000hdx_cap = ((phy_ext_status & IEEE_ESR_1000T_HD_CAPS) ||
4015 	    (phy_ext_status & IEEE_ESR_1000X_HD_CAPS)) ? 1 : 0;
4016 	igb->param_100t4_cap =
4017 	    (phy_status & MII_SR_100T4_CAPS) ? 1 : 0;
4018 	igb->param_100fdx_cap = ((phy_status & MII_SR_100X_FD_CAPS) ||
4019 	    (phy_status & MII_SR_100T2_FD_CAPS)) ? 1 : 0;
4020 	igb->param_100hdx_cap = ((phy_status & MII_SR_100X_HD_CAPS) ||
4021 	    (phy_status & MII_SR_100T2_HD_CAPS)) ? 1 : 0;
4022 	igb->param_10fdx_cap =
4023 	    (phy_status & MII_SR_10T_FD_CAPS) ? 1 : 0;
4024 	igb->param_10hdx_cap =
4025 	    (phy_status & MII_SR_10T_HD_CAPS) ? 1 : 0;
4026 	igb->param_rem_fault =
4027 	    (phy_status & MII_SR_REMOTE_FAULT) ? 1 : 0;
4028 
4029 	igb->param_adv_autoneg_cap = hw->mac.autoneg;
4030 	igb->param_adv_pause_cap =
4031 	    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4032 	igb->param_adv_asym_pause_cap =
4033 	    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4034 	igb->param_adv_1000hdx_cap =
4035 	    (phy_1000t_ctrl & CR_1000T_HD_CAPS) ? 1 : 0;
4036 	igb->param_adv_100t4_cap =
4037 	    (phy_an_adv & NWAY_AR_100T4_CAPS) ? 1 : 0;
4038 	igb->param_adv_rem_fault =
4039 	    (phy_an_adv & NWAY_AR_REMOTE_FAULT) ? 1 : 0;
4040 	if (igb->param_adv_autoneg_cap == 1) {
4041 		igb->param_adv_1000fdx_cap =
4042 		    (phy_1000t_ctrl & CR_1000T_FD_CAPS) ? 1 : 0;
4043 		igb->param_adv_100fdx_cap =
4044 		    (phy_an_adv & NWAY_AR_100TX_FD_CAPS) ? 1 : 0;
4045 		igb->param_adv_100hdx_cap =
4046 		    (phy_an_adv & NWAY_AR_100TX_HD_CAPS) ? 1 : 0;
4047 		igb->param_adv_10fdx_cap =
4048 		    (phy_an_adv & NWAY_AR_10T_FD_CAPS) ? 1 : 0;
4049 		igb->param_adv_10hdx_cap =
4050 		    (phy_an_adv & NWAY_AR_10T_HD_CAPS) ? 1 : 0;
4051 	}
4052 
4053 	igb->param_lp_autoneg_cap =
4054 	    (phy_an_exp & NWAY_ER_LP_NWAY_CAPS) ? 1 : 0;
4055 	igb->param_lp_pause_cap =
4056 	    (phy_lp_able & NWAY_LPAR_PAUSE) ? 1 : 0;
4057 	igb->param_lp_asym_pause_cap =
4058 	    (phy_lp_able & NWAY_LPAR_ASM_DIR) ? 1 : 0;
4059 	igb->param_lp_1000fdx_cap =
4060 	    (phy_1000t_status & SR_1000T_LP_FD_CAPS) ? 1 : 0;
4061 	igb->param_lp_1000hdx_cap =
4062 	    (phy_1000t_status & SR_1000T_LP_HD_CAPS) ? 1 : 0;
4063 	igb->param_lp_100t4_cap =
4064 	    (phy_lp_able & NWAY_LPAR_100T4_CAPS) ? 1 : 0;
4065 	igb->param_lp_100fdx_cap =
4066 	    (phy_lp_able & NWAY_LPAR_100TX_FD_CAPS) ? 1 : 0;
4067 	igb->param_lp_100hdx_cap =
4068 	    (phy_lp_able & NWAY_LPAR_100TX_HD_CAPS) ? 1 : 0;
4069 	igb->param_lp_10fdx_cap =
4070 	    (phy_lp_able & NWAY_LPAR_10T_FD_CAPS) ? 1 : 0;
4071 	igb->param_lp_10hdx_cap =
4072 	    (phy_lp_able & NWAY_LPAR_10T_HD_CAPS) ? 1 : 0;
4073 	igb->param_lp_rem_fault =
4074 	    (phy_lp_able & NWAY_LPAR_REMOTE_FAULT) ? 1 : 0;
4075 }
4076 
4077 /*
4078  * igb_get_driver_control
4079  */
4080 static void
4081 igb_get_driver_control(struct e1000_hw *hw)
4082 {
4083 	uint32_t ctrl_ext;
4084 
4085 	/* Notify firmware that driver is in control of device */
4086 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
4087 	ctrl_ext |= E1000_CTRL_EXT_DRV_LOAD;
4088 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
4089 }
4090 
4091 /*
4092  * igb_release_driver_control
4093  */
4094 static void
4095 igb_release_driver_control(struct e1000_hw *hw)
4096 {
4097 	uint32_t ctrl_ext;
4098 
4099 	/* Notify firmware that driver is no longer in control of device */
4100 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
4101 	ctrl_ext &= ~E1000_CTRL_EXT_DRV_LOAD;
4102 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
4103 }
4104 
4105 /*
4106  * igb_atomic_reserve - Atomic decrease operation
4107  */
4108 int
4109 igb_atomic_reserve(uint32_t *count_p, uint32_t n)
4110 {
4111 	uint32_t oldval;
4112 	uint32_t newval;
4113 
4114 	/* ATOMICALLY */
4115 	do {
4116 		oldval = *count_p;
4117 		if (oldval < n)
4118 			return (-1);
4119 		newval = oldval - n;
4120 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
4121 
4122 	return (newval);
4123 }
4124 
4125 /*
4126  * FMA support
4127  */
4128 
4129 int
4130 igb_check_acc_handle(ddi_acc_handle_t handle)
4131 {
4132 	ddi_fm_error_t de;
4133 
4134 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
4135 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
4136 	return (de.fme_status);
4137 }
4138 
4139 int
4140 igb_check_dma_handle(ddi_dma_handle_t handle)
4141 {
4142 	ddi_fm_error_t de;
4143 
4144 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
4145 	return (de.fme_status);
4146 }
4147 
4148 /*
4149  * The IO fault service error handling callback function
4150  */
4151 /*ARGSUSED*/
4152 static int
4153 igb_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
4154 {
4155 	/*
4156 	 * as the driver can always deal with an error in any dma or
4157 	 * access handle, we can just return the fme_status value.
4158 	 */
4159 	pci_ereport_post(dip, err, NULL);
4160 	return (err->fme_status);
4161 }
4162 
4163 static void
4164 igb_fm_init(igb_t *igb)
4165 {
4166 	ddi_iblock_cookie_t iblk;
4167 	int fma_acc_flag, fma_dma_flag;
4168 
4169 	/* Only register with IO Fault Services if we have some capability */
4170 	if (igb->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
4171 		igb_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
4172 		fma_acc_flag = 1;
4173 	} else {
4174 		igb_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
4175 		fma_acc_flag = 0;
4176 	}
4177 
4178 	if (igb->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
4179 		fma_dma_flag = 1;
4180 	} else {
4181 		fma_dma_flag = 0;
4182 	}
4183 
4184 	(void) igb_set_fma_flags(fma_acc_flag, fma_dma_flag);
4185 
4186 	if (igb->fm_capabilities) {
4187 
4188 		/* Register capabilities with IO Fault Services */
4189 		ddi_fm_init(igb->dip, &igb->fm_capabilities, &iblk);
4190 
4191 		/*
4192 		 * Initialize pci ereport capabilities if ereport capable
4193 		 */
4194 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
4195 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4196 			pci_ereport_setup(igb->dip);
4197 
4198 		/*
4199 		 * Register error callback if error callback capable
4200 		 */
4201 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4202 			ddi_fm_handler_register(igb->dip,
4203 			    igb_fm_error_cb, (void*) igb);
4204 	}
4205 }
4206 
4207 static void
4208 igb_fm_fini(igb_t *igb)
4209 {
4210 	/* Only unregister FMA capabilities if we registered some */
4211 	if (igb->fm_capabilities) {
4212 
4213 		/*
4214 		 * Release any resources allocated by pci_ereport_setup()
4215 		 */
4216 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
4217 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4218 			pci_ereport_teardown(igb->dip);
4219 
4220 		/*
4221 		 * Un-register error callback if error callback capable
4222 		 */
4223 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4224 			ddi_fm_handler_unregister(igb->dip);
4225 
4226 		/* Unregister from IO Fault Services */
4227 		ddi_fm_fini(igb->dip);
4228 	}
4229 }
4230 
4231 void
4232 igb_fm_ereport(igb_t *igb, char *detail)
4233 {
4234 	uint64_t ena;
4235 	char buf[FM_MAX_CLASS];
4236 
4237 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
4238 	ena = fm_ena_generate(0, FM_ENA_FMT1);
4239 	if (DDI_FM_EREPORT_CAP(igb->fm_capabilities)) {
4240 		ddi_fm_ereport_post(igb->dip, buf, ena, DDI_NOSLEEP,
4241 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
4242 	}
4243 }
4244