xref: /illumos-gate/usr/src/uts/common/io/igb/igb_main.c (revision 4eaa471005973e11a6110b69fe990530b3b95a38)
1 /*
2  * CDDL HEADER START
3  *
4  * Copyright(c) 2007-2009 Intel Corporation. All rights reserved.
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at:
10  *	http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When using or redistributing this file, you may do so under the
15  * License only. No other modification of this header is permitted.
16  *
17  * If applicable, add the following below this CDDL HEADER, with the
18  * fields enclosed by brackets "[]" replaced with your own identifying
19  * information: Portions Copyright [yyyy] [name of copyright owner]
20  *
21  * CDDL HEADER END
22  */
23 
24 /*
25  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
26  * Use is subject to license terms.
27  */
28 
29 #include "igb_sw.h"
30 
31 static char ident[] = "Intel 1Gb Ethernet";
32 static char igb_version[] = "igb 1.1.8";
33 
34 /*
35  * Local function protoypes
36  */
37 static int igb_register_mac(igb_t *);
38 static int igb_identify_hardware(igb_t *);
39 static int igb_regs_map(igb_t *);
40 static void igb_init_properties(igb_t *);
41 static int igb_init_driver_settings(igb_t *);
42 static void igb_init_locks(igb_t *);
43 static void igb_destroy_locks(igb_t *);
44 static int igb_init_mac_address(igb_t *);
45 static int igb_init(igb_t *);
46 static int igb_init_adapter(igb_t *);
47 static void igb_stop_adapter(igb_t *);
48 static int igb_reset(igb_t *);
49 static void igb_tx_clean(igb_t *);
50 static boolean_t igb_tx_drain(igb_t *);
51 static boolean_t igb_rx_drain(igb_t *);
52 static int igb_alloc_rings(igb_t *);
53 static void igb_free_rings(igb_t *);
54 static void igb_setup_rings(igb_t *);
55 static void igb_setup_rx(igb_t *);
56 static void igb_setup_tx(igb_t *);
57 static void igb_setup_rx_ring(igb_rx_ring_t *);
58 static void igb_setup_tx_ring(igb_tx_ring_t *);
59 static void igb_setup_rss(igb_t *);
60 static void igb_setup_mac_rss_classify(igb_t *);
61 static void igb_setup_mac_classify(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_82575(igb_t *);
76 static void igb_enable_adapter_interrupts_82576(igb_t *);
77 static boolean_t is_valid_mac_addr(uint8_t *);
78 static boolean_t igb_stall_check(igb_t *);
79 static boolean_t igb_set_loopback_mode(igb_t *, uint32_t);
80 static void igb_set_external_loopback(igb_t *);
81 static void igb_set_internal_mac_loopback(igb_t *);
82 static void igb_set_internal_phy_loopback(igb_t *);
83 static void igb_set_internal_serdes_loopback(igb_t *);
84 static boolean_t igb_find_mac_address(igb_t *);
85 static int igb_alloc_intrs(igb_t *);
86 static int igb_alloc_intr_handles(igb_t *, int);
87 static int igb_add_intr_handlers(igb_t *);
88 static void igb_rem_intr_handlers(igb_t *);
89 static void igb_rem_intrs(igb_t *);
90 static int igb_enable_intrs(igb_t *);
91 static int igb_disable_intrs(igb_t *);
92 static void igb_setup_msix_82575(igb_t *);
93 static void igb_setup_msix_82576(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(void *, void *);
98 static uint_t igb_intr_tx_other(void *, void *);
99 static void igb_intr_rx_work(igb_rx_ring_t *);
100 static void igb_intr_tx_work(igb_tx_ring_t *);
101 static void igb_intr_link_work(igb_t *);
102 static void igb_get_driver_control(struct e1000_hw *);
103 static void igb_release_driver_control(struct e1000_hw *);
104 
105 static int igb_attach(dev_info_t *, ddi_attach_cmd_t);
106 static int igb_detach(dev_info_t *, ddi_detach_cmd_t);
107 static int igb_resume(dev_info_t *);
108 static int igb_suspend(dev_info_t *);
109 static int igb_quiesce(dev_info_t *);
110 static void igb_unconfigure(dev_info_t *, igb_t *);
111 static int igb_fm_error_cb(dev_info_t *, ddi_fm_error_t *,
112     const void *);
113 static void igb_fm_init(igb_t *);
114 static void igb_fm_fini(igb_t *);
115 static void igb_release_multicast(igb_t *);
116 
117 static struct cb_ops igb_cb_ops = {
118 	nulldev,		/* cb_open */
119 	nulldev,		/* cb_close */
120 	nodev,			/* cb_strategy */
121 	nodev,			/* cb_print */
122 	nodev,			/* cb_dump */
123 	nodev,			/* cb_read */
124 	nodev,			/* cb_write */
125 	nodev,			/* cb_ioctl */
126 	nodev,			/* cb_devmap */
127 	nodev,			/* cb_mmap */
128 	nodev,			/* cb_segmap */
129 	nochpoll,		/* cb_chpoll */
130 	ddi_prop_op,		/* cb_prop_op */
131 	NULL,			/* cb_stream */
132 	D_MP | D_HOTPLUG,	/* cb_flag */
133 	CB_REV,			/* cb_rev */
134 	nodev,			/* cb_aread */
135 	nodev			/* cb_awrite */
136 };
137 
138 static struct dev_ops igb_dev_ops = {
139 	DEVO_REV,		/* devo_rev */
140 	0,			/* devo_refcnt */
141 	NULL,			/* devo_getinfo */
142 	nulldev,		/* devo_identify */
143 	nulldev,		/* devo_probe */
144 	igb_attach,		/* devo_attach */
145 	igb_detach,		/* devo_detach */
146 	nodev,			/* devo_reset */
147 	&igb_cb_ops,		/* devo_cb_ops */
148 	NULL,			/* devo_bus_ops */
149 	ddi_power,		/* devo_power */
150 	igb_quiesce,	/* devo_quiesce */
151 };
152 
153 static struct modldrv igb_modldrv = {
154 	&mod_driverops,		/* Type of module.  This one is a driver */
155 	ident,			/* Discription string */
156 	&igb_dev_ops,		/* driver ops */
157 };
158 
159 static struct modlinkage igb_modlinkage = {
160 	MODREV_1, &igb_modldrv, NULL
161 };
162 
163 /* Access attributes for register mapping */
164 ddi_device_acc_attr_t igb_regs_acc_attr = {
165 	DDI_DEVICE_ATTR_V0,
166 	DDI_STRUCTURE_LE_ACC,
167 	DDI_STRICTORDER_ACC,
168 	DDI_FLAGERR_ACC
169 };
170 
171 #define	IGB_M_CALLBACK_FLAGS	(MC_IOCTL | MC_GETCAPAB)
172 
173 static mac_callbacks_t igb_m_callbacks = {
174 	IGB_M_CALLBACK_FLAGS,
175 	igb_m_stat,
176 	igb_m_start,
177 	igb_m_stop,
178 	igb_m_promisc,
179 	igb_m_multicst,
180 	NULL,
181 	NULL,
182 	igb_m_ioctl,
183 	igb_m_getcapab
184 };
185 
186 /*
187  * Initialize capabilities of each supported adapter type
188  */
189 static adapter_info_t igb_82575_cap = {
190 	/* limits */
191 	4,		/* maximum number of rx queues */
192 	1,		/* minimum number of rx queues */
193 	4,		/* default number of rx queues */
194 	4,		/* maximum number of tx queues */
195 	1,		/* minimum number of tx queues */
196 	4,		/* default number of tx queues */
197 	65535,		/* maximum interrupt throttle rate */
198 	0,		/* minimum interrupt throttle rate */
199 	200,		/* default interrupt throttle rate */
200 
201 	/* function pointers */
202 	igb_enable_adapter_interrupts_82575,
203 	igb_setup_msix_82575,
204 
205 	/* capabilities */
206 	(IGB_FLAG_HAS_DCA |	/* capability flags */
207 	IGB_FLAG_VMDQ_POOL),
208 
209 	0xffc00000		/* mask for RXDCTL register */
210 };
211 
212 static adapter_info_t igb_82576_cap = {
213 	/* limits */
214 	16,		/* maximum number of rx queues */
215 	1,		/* minimum number of rx queues */
216 	4,		/* default number of rx queues */
217 	16,		/* maximum number of tx queues */
218 	1,		/* minimum number of tx queues */
219 	4,		/* default number of tx queues */
220 	65535,		/* maximum interrupt throttle rate */
221 	0,		/* minimum interrupt throttle rate */
222 	200,		/* default interrupt throttle rate */
223 
224 	/* function pointers */
225 	igb_enable_adapter_interrupts_82576,
226 	igb_setup_msix_82576,
227 
228 	/* capabilities */
229 	(IGB_FLAG_HAS_DCA |	/* capability flags */
230 	IGB_FLAG_VMDQ_POOL |
231 	IGB_FLAG_NEED_CTX_IDX),
232 
233 	0xffe00000		/* mask for RXDCTL register */
234 };
235 
236 /*
237  * Module Initialization Functions
238  */
239 
240 int
241 _init(void)
242 {
243 	int status;
244 
245 	mac_init_ops(&igb_dev_ops, MODULE_NAME);
246 
247 	status = mod_install(&igb_modlinkage);
248 
249 	if (status != DDI_SUCCESS) {
250 		mac_fini_ops(&igb_dev_ops);
251 	}
252 
253 	return (status);
254 }
255 
256 int
257 _fini(void)
258 {
259 	int status;
260 
261 	status = mod_remove(&igb_modlinkage);
262 
263 	if (status == DDI_SUCCESS) {
264 		mac_fini_ops(&igb_dev_ops);
265 	}
266 
267 	return (status);
268 
269 }
270 
271 int
272 _info(struct modinfo *modinfop)
273 {
274 	int status;
275 
276 	status = mod_info(&igb_modlinkage, modinfop);
277 
278 	return (status);
279 }
280 
281 /*
282  * igb_attach - driver attach
283  *
284  * This function is the device specific initialization entry
285  * point. This entry point is required and must be written.
286  * The DDI_ATTACH command must be provided in the attach entry
287  * point. When attach() is called with cmd set to DDI_ATTACH,
288  * all normal kernel services (such as kmem_alloc(9F)) are
289  * available for use by the driver.
290  *
291  * The attach() function will be called once for each instance
292  * of  the  device  on  the  system with cmd set to DDI_ATTACH.
293  * Until attach() succeeds, the only driver entry points which
294  * may be called are open(9E) and getinfo(9E).
295  */
296 static int
297 igb_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
298 {
299 	igb_t *igb;
300 	struct igb_osdep *osdep;
301 	struct e1000_hw *hw;
302 	int instance;
303 
304 	/*
305 	 * Check the command and perform corresponding operations
306 	 */
307 	switch (cmd) {
308 	default:
309 		return (DDI_FAILURE);
310 
311 	case DDI_RESUME:
312 		return (igb_resume(devinfo));
313 
314 	case DDI_ATTACH:
315 		break;
316 	}
317 
318 	/* Get the device instance */
319 	instance = ddi_get_instance(devinfo);
320 
321 	/* Allocate memory for the instance data structure */
322 	igb = kmem_zalloc(sizeof (igb_t), KM_SLEEP);
323 
324 	igb->dip = devinfo;
325 	igb->instance = instance;
326 
327 	hw = &igb->hw;
328 	osdep = &igb->osdep;
329 	hw->back = osdep;
330 	osdep->igb = igb;
331 
332 	/* Attach the instance pointer to the dev_info data structure */
333 	ddi_set_driver_private(devinfo, igb);
334 
335 
336 	/* Initialize for fma support */
337 	igb->fm_capabilities = igb_get_prop(igb, "fm-capable",
338 	    0, 0x0f,
339 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
340 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
341 	igb_fm_init(igb);
342 	igb->attach_progress |= ATTACH_PROGRESS_FMINIT;
343 
344 	/*
345 	 * Map PCI config space registers
346 	 */
347 	if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
348 		igb_error(igb, "Failed to map PCI configurations");
349 		goto attach_fail;
350 	}
351 	igb->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
352 
353 	/*
354 	 * Identify the chipset family
355 	 */
356 	if (igb_identify_hardware(igb) != IGB_SUCCESS) {
357 		igb_error(igb, "Failed to identify hardware");
358 		goto attach_fail;
359 	}
360 
361 	/*
362 	 * Map device registers
363 	 */
364 	if (igb_regs_map(igb) != IGB_SUCCESS) {
365 		igb_error(igb, "Failed to map device registers");
366 		goto attach_fail;
367 	}
368 	igb->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
369 
370 	/*
371 	 * Initialize driver parameters
372 	 */
373 	igb_init_properties(igb);
374 	igb->attach_progress |= ATTACH_PROGRESS_PROPS;
375 
376 	/*
377 	 * Allocate interrupts
378 	 */
379 	if (igb_alloc_intrs(igb) != IGB_SUCCESS) {
380 		igb_error(igb, "Failed to allocate interrupts");
381 		goto attach_fail;
382 	}
383 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
384 
385 	/*
386 	 * Allocate rx/tx rings based on the ring numbers.
387 	 * The actual numbers of rx/tx rings are decided by the number of
388 	 * allocated interrupt vectors, so we should allocate the rings after
389 	 * interrupts are allocated.
390 	 */
391 	if (igb_alloc_rings(igb) != IGB_SUCCESS) {
392 		igb_error(igb, "Failed to allocate rx/tx rings or groups");
393 		goto attach_fail;
394 	}
395 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_RINGS;
396 
397 	/*
398 	 * Add interrupt handlers
399 	 */
400 	if (igb_add_intr_handlers(igb) != IGB_SUCCESS) {
401 		igb_error(igb, "Failed to add interrupt handlers");
402 		goto attach_fail;
403 	}
404 	igb->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
405 
406 	/*
407 	 * Initialize driver parameters
408 	 */
409 	if (igb_init_driver_settings(igb) != IGB_SUCCESS) {
410 		igb_error(igb, "Failed to initialize driver settings");
411 		goto attach_fail;
412 	}
413 
414 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK) {
415 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
416 		goto attach_fail;
417 	}
418 
419 	/*
420 	 * Initialize mutexes for this device.
421 	 * Do this before enabling the interrupt handler and
422 	 * register the softint to avoid the condition where
423 	 * interrupt handler can try using uninitialized mutex
424 	 */
425 	igb_init_locks(igb);
426 	igb->attach_progress |= ATTACH_PROGRESS_LOCKS;
427 
428 	/*
429 	 * Allocate DMA resources
430 	 */
431 	if (igb_alloc_dma(igb) != IGB_SUCCESS) {
432 		igb_error(igb, "Failed to allocate DMA resources");
433 		goto attach_fail;
434 	}
435 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_DMA;
436 
437 	/*
438 	 * Initialize the adapter and setup the rx/tx rings
439 	 */
440 	if (igb_init(igb) != IGB_SUCCESS) {
441 		igb_error(igb, "Failed to initialize adapter");
442 		goto attach_fail;
443 	}
444 	igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
445 
446 	/*
447 	 * Initialize statistics
448 	 */
449 	if (igb_init_stats(igb) != IGB_SUCCESS) {
450 		igb_error(igb, "Failed to initialize statistics");
451 		goto attach_fail;
452 	}
453 	igb->attach_progress |= ATTACH_PROGRESS_STATS;
454 
455 	/*
456 	 * Initialize NDD parameters
457 	 */
458 	if (igb_nd_init(igb) != IGB_SUCCESS) {
459 		igb_error(igb, "Failed to initialize ndd");
460 		goto attach_fail;
461 	}
462 	igb->attach_progress |= ATTACH_PROGRESS_NDD;
463 
464 	/*
465 	 * Register the driver to the MAC
466 	 */
467 	if (igb_register_mac(igb) != IGB_SUCCESS) {
468 		igb_error(igb, "Failed to register MAC");
469 		goto attach_fail;
470 	}
471 	igb->attach_progress |= ATTACH_PROGRESS_MAC;
472 
473 	/*
474 	 * Now that mutex locks are initialized, and the chip is also
475 	 * initialized, enable interrupts.
476 	 */
477 	if (igb_enable_intrs(igb) != IGB_SUCCESS) {
478 		igb_error(igb, "Failed to enable DDI interrupts");
479 		goto attach_fail;
480 	}
481 	igb->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
482 
483 	igb_log(igb, "%s", igb_version);
484 	igb->igb_state |= IGB_INITIALIZED;
485 
486 	return (DDI_SUCCESS);
487 
488 attach_fail:
489 	igb_unconfigure(devinfo, igb);
490 	return (DDI_FAILURE);
491 }
492 
493 /*
494  * igb_detach - driver detach
495  *
496  * The detach() function is the complement of the attach routine.
497  * If cmd is set to DDI_DETACH, detach() is used to remove  the
498  * state  associated  with  a  given  instance of a device node
499  * prior to the removal of that instance from the system.
500  *
501  * The detach() function will be called once for each  instance
502  * of the device for which there has been a successful attach()
503  * once there are no longer  any  opens  on  the  device.
504  *
505  * Interrupts routine are disabled, All memory allocated by this
506  * driver are freed.
507  */
508 static int
509 igb_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
510 {
511 	igb_t *igb;
512 
513 	/*
514 	 * Check detach command
515 	 */
516 	switch (cmd) {
517 	default:
518 		return (DDI_FAILURE);
519 
520 	case DDI_SUSPEND:
521 		return (igb_suspend(devinfo));
522 
523 	case DDI_DETACH:
524 		break;
525 	}
526 
527 
528 	/*
529 	 * Get the pointer to the driver private data structure
530 	 */
531 	igb = (igb_t *)ddi_get_driver_private(devinfo);
532 	if (igb == NULL)
533 		return (DDI_FAILURE);
534 
535 	/*
536 	 * Unregister MAC. If failed, we have to fail the detach
537 	 */
538 	if (mac_unregister(igb->mac_hdl) != 0) {
539 		igb_error(igb, "Failed to unregister MAC");
540 		return (DDI_FAILURE);
541 	}
542 	igb->attach_progress &= ~ATTACH_PROGRESS_MAC;
543 
544 	/*
545 	 * If the device is still running, it needs to be stopped first.
546 	 * This check is necessary because under some specific circumstances,
547 	 * the detach routine can be called without stopping the interface
548 	 * first.
549 	 */
550 	mutex_enter(&igb->gen_lock);
551 	if (igb->igb_state & IGB_STARTED) {
552 		igb->igb_state &= ~IGB_STARTED;
553 		igb_stop(igb);
554 		mutex_exit(&igb->gen_lock);
555 		/* Disable and stop the watchdog timer */
556 		igb_disable_watchdog_timer(igb);
557 	} else
558 		mutex_exit(&igb->gen_lock);
559 
560 	/*
561 	 * Check if there are still rx buffers held by the upper layer.
562 	 * If so, fail the detach.
563 	 */
564 	if (!igb_rx_drain(igb))
565 		return (DDI_FAILURE);
566 
567 	/*
568 	 * Do the remaining unconfigure routines
569 	 */
570 	igb_unconfigure(devinfo, igb);
571 
572 	return (DDI_SUCCESS);
573 }
574 
575 /*
576  * quiesce(9E) entry point.
577  *
578  * This function is called when the system is single-threaded at high
579  * PIL with preemption disabled. Therefore, this function must not be
580  * blocked.
581  *
582  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
583  * DDI_FAILURE indicates an error condition and should almost never happen.
584  */
585 static int
586 igb_quiesce(dev_info_t *devinfo)
587 {
588 	igb_t *igb;
589 	struct e1000_hw *hw;
590 
591 	igb = (igb_t *)ddi_get_driver_private(devinfo);
592 
593 	if (igb == NULL)
594 		return (DDI_FAILURE);
595 
596 	hw = &igb->hw;
597 
598 	/*
599 	 * Disable the adapter interrupts
600 	 */
601 	igb_disable_adapter_interrupts(igb);
602 
603 	/* Tell firmware driver is no longer in control */
604 	igb_release_driver_control(hw);
605 
606 	/*
607 	 * Reset the chipset
608 	 */
609 	(void) e1000_reset_hw(hw);
610 
611 	/*
612 	 * Reset PHY if possible
613 	 */
614 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
615 		(void) e1000_phy_hw_reset(hw);
616 
617 	return (DDI_SUCCESS);
618 }
619 
620 /*
621  * igb_unconfigure - release all resources held by this instance
622  */
623 static void
624 igb_unconfigure(dev_info_t *devinfo, igb_t *igb)
625 {
626 	/*
627 	 * Disable interrupt
628 	 */
629 	if (igb->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
630 		(void) igb_disable_intrs(igb);
631 	}
632 
633 	/*
634 	 * Unregister MAC
635 	 */
636 	if (igb->attach_progress & ATTACH_PROGRESS_MAC) {
637 		(void) mac_unregister(igb->mac_hdl);
638 	}
639 
640 	/*
641 	 * Free ndd parameters
642 	 */
643 	if (igb->attach_progress & ATTACH_PROGRESS_NDD) {
644 		igb_nd_cleanup(igb);
645 	}
646 
647 	/*
648 	 * Free statistics
649 	 */
650 	if (igb->attach_progress & ATTACH_PROGRESS_STATS) {
651 		kstat_delete((kstat_t *)igb->igb_ks);
652 	}
653 
654 	/*
655 	 * Remove interrupt handlers
656 	 */
657 	if (igb->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
658 		igb_rem_intr_handlers(igb);
659 	}
660 
661 	/*
662 	 * Remove interrupts
663 	 */
664 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_INTR) {
665 		igb_rem_intrs(igb);
666 	}
667 
668 	/*
669 	 * Remove driver properties
670 	 */
671 	if (igb->attach_progress & ATTACH_PROGRESS_PROPS) {
672 		(void) ddi_prop_remove_all(devinfo);
673 	}
674 
675 	/*
676 	 * Release the DMA resources of rx/tx rings
677 	 */
678 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_DMA) {
679 		igb_free_dma(igb);
680 	}
681 
682 	/*
683 	 * Stop the adapter
684 	 */
685 	if (igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) {
686 		mutex_enter(&igb->gen_lock);
687 		igb_stop_adapter(igb);
688 		mutex_exit(&igb->gen_lock);
689 		if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
690 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_UNAFFECTED);
691 	}
692 
693 	/*
694 	 * Free multicast table
695 	 */
696 	igb_release_multicast(igb);
697 
698 	/*
699 	 * Free register handle
700 	 */
701 	if (igb->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
702 		if (igb->osdep.reg_handle != NULL)
703 			ddi_regs_map_free(&igb->osdep.reg_handle);
704 	}
705 
706 	/*
707 	 * Free PCI config handle
708 	 */
709 	if (igb->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
710 		if (igb->osdep.cfg_handle != NULL)
711 			pci_config_teardown(&igb->osdep.cfg_handle);
712 	}
713 
714 	/*
715 	 * Free locks
716 	 */
717 	if (igb->attach_progress & ATTACH_PROGRESS_LOCKS) {
718 		igb_destroy_locks(igb);
719 	}
720 
721 	/*
722 	 * Free the rx/tx rings
723 	 */
724 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
725 		igb_free_rings(igb);
726 	}
727 
728 	/*
729 	 * Remove FMA
730 	 */
731 	if (igb->attach_progress & ATTACH_PROGRESS_FMINIT) {
732 		igb_fm_fini(igb);
733 	}
734 
735 	/*
736 	 * Free the driver data structure
737 	 */
738 	kmem_free(igb, sizeof (igb_t));
739 
740 	ddi_set_driver_private(devinfo, NULL);
741 }
742 
743 /*
744  * igb_register_mac - Register the driver and its function pointers with
745  * the GLD interface
746  */
747 static int
748 igb_register_mac(igb_t *igb)
749 {
750 	struct e1000_hw *hw = &igb->hw;
751 	mac_register_t *mac;
752 	int status;
753 
754 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
755 		return (IGB_FAILURE);
756 
757 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
758 	mac->m_driver = igb;
759 	mac->m_dip = igb->dip;
760 	mac->m_src_addr = hw->mac.addr;
761 	mac->m_callbacks = &igb_m_callbacks;
762 	mac->m_min_sdu = 0;
763 	mac->m_max_sdu = igb->max_frame_size -
764 	    sizeof (struct ether_vlan_header) - ETHERFCSL;
765 	mac->m_margin = VLAN_TAGSZ;
766 	mac->m_v12n = MAC_VIRT_LEVEL1;
767 
768 	status = mac_register(mac, &igb->mac_hdl);
769 
770 	mac_free(mac);
771 
772 	return ((status == 0) ? IGB_SUCCESS : IGB_FAILURE);
773 }
774 
775 /*
776  * igb_identify_hardware - Identify the type of the chipset
777  */
778 static int
779 igb_identify_hardware(igb_t *igb)
780 {
781 	struct e1000_hw *hw = &igb->hw;
782 	struct igb_osdep *osdep = &igb->osdep;
783 
784 	/*
785 	 * Get the device id
786 	 */
787 	hw->vendor_id =
788 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
789 	hw->device_id =
790 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
791 	hw->revision_id =
792 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
793 	hw->subsystem_device_id =
794 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
795 	hw->subsystem_vendor_id =
796 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
797 
798 	/*
799 	 * Set the mac type of the adapter based on the device id
800 	 */
801 	if (e1000_set_mac_type(hw) != E1000_SUCCESS) {
802 		return (IGB_FAILURE);
803 	}
804 
805 	/*
806 	 * Install adapter capabilities based on mac type
807 	 */
808 	switch (hw->mac.type) {
809 	case e1000_82575:
810 		igb->capab = &igb_82575_cap;
811 		break;
812 	case e1000_82576:
813 		igb->capab = &igb_82576_cap;
814 		break;
815 	default:
816 		return (IGB_FAILURE);
817 	}
818 
819 	return (IGB_SUCCESS);
820 }
821 
822 /*
823  * igb_regs_map - Map the device registers
824  */
825 static int
826 igb_regs_map(igb_t *igb)
827 {
828 	dev_info_t *devinfo = igb->dip;
829 	struct e1000_hw *hw = &igb->hw;
830 	struct igb_osdep *osdep = &igb->osdep;
831 	off_t mem_size;
832 
833 	/*
834 	 * First get the size of device registers to be mapped.
835 	 */
836 	if (ddi_dev_regsize(devinfo, IGB_ADAPTER_REGSET, &mem_size) !=
837 	    DDI_SUCCESS) {
838 		return (IGB_FAILURE);
839 	}
840 
841 	/*
842 	 * Call ddi_regs_map_setup() to map registers
843 	 */
844 	if ((ddi_regs_map_setup(devinfo, IGB_ADAPTER_REGSET,
845 	    (caddr_t *)&hw->hw_addr, 0,
846 	    mem_size, &igb_regs_acc_attr,
847 	    &osdep->reg_handle)) != DDI_SUCCESS) {
848 		return (IGB_FAILURE);
849 	}
850 
851 	return (IGB_SUCCESS);
852 }
853 
854 /*
855  * igb_init_properties - Initialize driver properties
856  */
857 static void
858 igb_init_properties(igb_t *igb)
859 {
860 	/*
861 	 * Get conf file properties, including link settings
862 	 * jumbo frames, ring number, descriptor number, etc.
863 	 */
864 	igb_get_conf(igb);
865 }
866 
867 /*
868  * igb_init_driver_settings - Initialize driver settings
869  *
870  * The settings include hardware function pointers, bus information,
871  * rx/tx rings settings, link state, and any other parameters that
872  * need to be setup during driver initialization.
873  */
874 static int
875 igb_init_driver_settings(igb_t *igb)
876 {
877 	struct e1000_hw *hw = &igb->hw;
878 	igb_rx_ring_t *rx_ring;
879 	igb_tx_ring_t *tx_ring;
880 	uint32_t rx_size;
881 	uint32_t tx_size;
882 	int i;
883 
884 	/*
885 	 * Initialize chipset specific hardware function pointers
886 	 */
887 	if (e1000_setup_init_funcs(hw, B_TRUE) != E1000_SUCCESS) {
888 		return (IGB_FAILURE);
889 	}
890 
891 	/*
892 	 * Get bus information
893 	 */
894 	if (e1000_get_bus_info(hw) != E1000_SUCCESS) {
895 		return (IGB_FAILURE);
896 	}
897 
898 	/*
899 	 * Get the system page size
900 	 */
901 	igb->page_size = ddi_ptob(igb->dip, (ulong_t)1);
902 
903 	/*
904 	 * Set rx buffer size
905 	 * The IP header alignment room is counted in the calculation.
906 	 * The rx buffer size is in unit of 1K that is required by the
907 	 * chipset hardware.
908 	 */
909 	rx_size = igb->max_frame_size + IPHDR_ALIGN_ROOM;
910 	igb->rx_buf_size = ((rx_size >> 10) +
911 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
912 
913 	/*
914 	 * Set tx buffer size
915 	 */
916 	tx_size = igb->max_frame_size;
917 	igb->tx_buf_size = ((tx_size >> 10) +
918 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
919 
920 	/*
921 	 * Initialize rx/tx rings parameters
922 	 */
923 	for (i = 0; i < igb->num_rx_rings; i++) {
924 		rx_ring = &igb->rx_rings[i];
925 		rx_ring->index = i;
926 		rx_ring->igb = igb;
927 
928 		rx_ring->ring_size = igb->rx_ring_size;
929 		rx_ring->free_list_size = igb->rx_ring_size;
930 		rx_ring->copy_thresh = igb->rx_copy_thresh;
931 		rx_ring->limit_per_intr = igb->rx_limit_per_intr;
932 	}
933 
934 	for (i = 0; i < igb->num_tx_rings; i++) {
935 		tx_ring = &igb->tx_rings[i];
936 		tx_ring->index = i;
937 		tx_ring->igb = igb;
938 		if (igb->tx_head_wb_enable)
939 			tx_ring->tx_recycle = igb_tx_recycle_head_wb;
940 		else
941 			tx_ring->tx_recycle = igb_tx_recycle_legacy;
942 
943 		tx_ring->ring_size = igb->tx_ring_size;
944 		tx_ring->free_list_size = igb->tx_ring_size +
945 		    (igb->tx_ring_size >> 1);
946 		tx_ring->copy_thresh = igb->tx_copy_thresh;
947 		tx_ring->recycle_thresh = igb->tx_recycle_thresh;
948 		tx_ring->overload_thresh = igb->tx_overload_thresh;
949 		tx_ring->resched_thresh = igb->tx_resched_thresh;
950 	}
951 
952 	/*
953 	 * Initialize values of interrupt throttling rates
954 	 */
955 	for (i = 1; i < MAX_NUM_EITR; i++)
956 		igb->intr_throttling[i] = igb->intr_throttling[0];
957 
958 	/*
959 	 * The initial link state should be "unknown"
960 	 */
961 	igb->link_state = LINK_STATE_UNKNOWN;
962 
963 	return (IGB_SUCCESS);
964 }
965 
966 /*
967  * igb_init_locks - Initialize locks
968  */
969 static void
970 igb_init_locks(igb_t *igb)
971 {
972 	igb_rx_ring_t *rx_ring;
973 	igb_tx_ring_t *tx_ring;
974 	int i;
975 
976 	for (i = 0; i < igb->num_rx_rings; i++) {
977 		rx_ring = &igb->rx_rings[i];
978 		mutex_init(&rx_ring->rx_lock, NULL,
979 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
980 		mutex_init(&rx_ring->recycle_lock, NULL,
981 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
982 	}
983 
984 	for (i = 0; i < igb->num_tx_rings; i++) {
985 		tx_ring = &igb->tx_rings[i];
986 		mutex_init(&tx_ring->tx_lock, NULL,
987 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
988 		mutex_init(&tx_ring->recycle_lock, NULL,
989 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
990 		mutex_init(&tx_ring->tcb_head_lock, NULL,
991 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
992 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
993 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
994 	}
995 
996 	mutex_init(&igb->gen_lock, NULL,
997 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
998 
999 	mutex_init(&igb->watchdog_lock, NULL,
1000 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1001 }
1002 
1003 /*
1004  * igb_destroy_locks - Destroy locks
1005  */
1006 static void
1007 igb_destroy_locks(igb_t *igb)
1008 {
1009 	igb_rx_ring_t *rx_ring;
1010 	igb_tx_ring_t *tx_ring;
1011 	int i;
1012 
1013 	for (i = 0; i < igb->num_rx_rings; i++) {
1014 		rx_ring = &igb->rx_rings[i];
1015 		mutex_destroy(&rx_ring->rx_lock);
1016 		mutex_destroy(&rx_ring->recycle_lock);
1017 	}
1018 
1019 	for (i = 0; i < igb->num_tx_rings; i++) {
1020 		tx_ring = &igb->tx_rings[i];
1021 		mutex_destroy(&tx_ring->tx_lock);
1022 		mutex_destroy(&tx_ring->recycle_lock);
1023 		mutex_destroy(&tx_ring->tcb_head_lock);
1024 		mutex_destroy(&tx_ring->tcb_tail_lock);
1025 	}
1026 
1027 	mutex_destroy(&igb->gen_lock);
1028 	mutex_destroy(&igb->watchdog_lock);
1029 }
1030 
1031 static int
1032 igb_resume(dev_info_t *devinfo)
1033 {
1034 	igb_t *igb;
1035 
1036 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1037 	if (igb == NULL)
1038 		return (DDI_FAILURE);
1039 
1040 	mutex_enter(&igb->gen_lock);
1041 
1042 	if (igb->igb_state & IGB_STARTED) {
1043 		if (igb_start(igb) != IGB_SUCCESS) {
1044 			mutex_exit(&igb->gen_lock);
1045 			return (DDI_FAILURE);
1046 		}
1047 
1048 		/*
1049 		 * Enable and start the watchdog timer
1050 		 */
1051 		igb_enable_watchdog_timer(igb);
1052 	}
1053 
1054 	igb->igb_state &= ~IGB_SUSPENDED;
1055 
1056 	mutex_exit(&igb->gen_lock);
1057 
1058 	return (DDI_SUCCESS);
1059 }
1060 
1061 static int
1062 igb_suspend(dev_info_t *devinfo)
1063 {
1064 	igb_t *igb;
1065 
1066 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1067 	if (igb == NULL)
1068 		return (DDI_FAILURE);
1069 
1070 	mutex_enter(&igb->gen_lock);
1071 
1072 	igb->igb_state |= IGB_SUSPENDED;
1073 
1074 	if (!(igb->igb_state & IGB_STARTED)) {
1075 		mutex_exit(&igb->gen_lock);
1076 		return (DDI_SUCCESS);
1077 	}
1078 
1079 	igb_stop(igb);
1080 
1081 	mutex_exit(&igb->gen_lock);
1082 
1083 	/*
1084 	 * Disable and stop the watchdog timer
1085 	 */
1086 	igb_disable_watchdog_timer(igb);
1087 
1088 	return (DDI_SUCCESS);
1089 }
1090 
1091 static int
1092 igb_init(igb_t *igb)
1093 {
1094 	int i;
1095 
1096 	mutex_enter(&igb->gen_lock);
1097 
1098 	/*
1099 	 * Initilize the adapter
1100 	 */
1101 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1102 		mutex_exit(&igb->gen_lock);
1103 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1104 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1105 		return (IGB_FAILURE);
1106 	}
1107 
1108 	/*
1109 	 * Setup the rx/tx rings
1110 	 */
1111 	for (i = 0; i < igb->num_rx_rings; i++)
1112 		mutex_enter(&igb->rx_rings[i].rx_lock);
1113 	for (i = 0; i < igb->num_tx_rings; i++)
1114 		mutex_enter(&igb->tx_rings[i].tx_lock);
1115 
1116 	igb_setup_rings(igb);
1117 
1118 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1119 		mutex_exit(&igb->tx_rings[i].tx_lock);
1120 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1121 		mutex_exit(&igb->rx_rings[i].rx_lock);
1122 
1123 	mutex_exit(&igb->gen_lock);
1124 
1125 	return (IGB_SUCCESS);
1126 }
1127 
1128 /*
1129  * igb_init_mac_address - Initialize the default MAC address
1130  *
1131  * On success, the MAC address is entered in the igb->hw.mac.addr
1132  * and hw->mac.perm_addr fields and the adapter's RAR(0) receive
1133  * address register.
1134  *
1135  * Important side effects:
1136  * 1. adapter is reset - this is required to put it in a known state.
1137  * 2. all of non-volatile memory (NVM) is read & checksummed - NVM is where
1138  * MAC address and all default settings are stored, so a valid checksum
1139  * is required.
1140  */
1141 static int
1142 igb_init_mac_address(igb_t *igb)
1143 {
1144 	struct e1000_hw *hw = &igb->hw;
1145 
1146 	ASSERT(mutex_owned(&igb->gen_lock));
1147 
1148 	/*
1149 	 * Reset chipset to put the hardware in a known state
1150 	 * before we try to get MAC address from NVM.
1151 	 */
1152 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1153 		igb_error(igb, "Adapter reset failed.");
1154 		goto init_mac_fail;
1155 	}
1156 
1157 	/*
1158 	 * NVM validation
1159 	 */
1160 	if (e1000_validate_nvm_checksum(hw) < 0) {
1161 		/*
1162 		 * Some PCI-E parts fail the first check due to
1163 		 * the link being in sleep state.  Call it again,
1164 		 * if it fails a second time its a real issue.
1165 		 */
1166 		if (e1000_validate_nvm_checksum(hw) < 0) {
1167 			igb_error(igb,
1168 			    "Invalid NVM checksum. Please contact "
1169 			    "the vendor to update the NVM.");
1170 			goto init_mac_fail;
1171 		}
1172 	}
1173 
1174 	/*
1175 	 * Get the mac address
1176 	 * This function should handle SPARC case correctly.
1177 	 */
1178 	if (!igb_find_mac_address(igb)) {
1179 		igb_error(igb, "Failed to get the mac address");
1180 		goto init_mac_fail;
1181 	}
1182 
1183 	/* Validate mac address */
1184 	if (!is_valid_mac_addr(hw->mac.addr)) {
1185 		igb_error(igb, "Invalid mac address");
1186 		goto init_mac_fail;
1187 	}
1188 
1189 	return (IGB_SUCCESS);
1190 
1191 init_mac_fail:
1192 	return (IGB_FAILURE);
1193 }
1194 
1195 /*
1196  * igb_init_adapter - Initialize the adapter
1197  */
1198 static int
1199 igb_init_adapter(igb_t *igb)
1200 {
1201 	struct e1000_hw *hw = &igb->hw;
1202 	uint32_t pba;
1203 	uint32_t high_water;
1204 	int i;
1205 
1206 	ASSERT(mutex_owned(&igb->gen_lock));
1207 
1208 	/*
1209 	 * In order to obtain the default MAC address, this will reset the
1210 	 * adapter and validate the NVM that the address and many other
1211 	 * default settings come from.
1212 	 */
1213 	if (igb_init_mac_address(igb) != IGB_SUCCESS) {
1214 		igb_error(igb, "Failed to initialize MAC address");
1215 		goto init_adapter_fail;
1216 	}
1217 
1218 	/*
1219 	 * Setup flow control
1220 	 *
1221 	 * These parameters set thresholds for the adapter's generation(Tx)
1222 	 * and response(Rx) to Ethernet PAUSE frames.  These are just threshold
1223 	 * settings.  Flow control is enabled or disabled in the configuration
1224 	 * file.
1225 	 * High-water mark is set down from the top of the rx fifo (not
1226 	 * sensitive to max_frame_size) and low-water is set just below
1227 	 * high-water mark.
1228 	 * The high water mark must be low enough to fit one full frame above
1229 	 * it in the rx FIFO.  Should be the lower of:
1230 	 * 90% of the Rx FIFO size, or the full Rx FIFO size minus one full
1231 	 * frame.
1232 	 */
1233 	/*
1234 	 * The default setting of PBA is correct for 82575 and other supported
1235 	 * adapters do not have the E1000_PBA register, so PBA value is only
1236 	 * used for calculation here and is never written to the adapter.
1237 	 */
1238 	if (hw->mac.type == e1000_82575) {
1239 		pba = E1000_PBA_34K;
1240 	} else {
1241 		pba = E1000_PBA_64K;
1242 	}
1243 
1244 	high_water = min(((pba << 10) * 9 / 10),
1245 	    ((pba << 10) - igb->max_frame_size));
1246 
1247 	if (hw->mac.type == e1000_82575) {
1248 		/* 8-byte granularity */
1249 		hw->fc.high_water = high_water & 0xFFF8;
1250 		hw->fc.low_water = hw->fc.high_water - 8;
1251 	} else {
1252 		/* 16-byte granularity */
1253 		hw->fc.high_water = high_water & 0xFFF0;
1254 		hw->fc.low_water = hw->fc.high_water - 16;
1255 	}
1256 
1257 	hw->fc.pause_time = E1000_FC_PAUSE_TIME;
1258 	hw->fc.send_xon = B_TRUE;
1259 
1260 	e1000_validate_mdi_setting(hw);
1261 
1262 	/*
1263 	 * Reset the chipset hardware the second time to put PBA settings
1264 	 * into effect.
1265 	 */
1266 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1267 		igb_error(igb, "Second reset failed");
1268 		goto init_adapter_fail;
1269 	}
1270 
1271 	/*
1272 	 * Don't wait for auto-negotiation to complete
1273 	 */
1274 	hw->phy.autoneg_wait_to_complete = B_FALSE;
1275 
1276 	/*
1277 	 * Copper options
1278 	 */
1279 	if (hw->phy.media_type == e1000_media_type_copper) {
1280 		hw->phy.mdix = 0;	/* AUTO_ALL_MODES */
1281 		hw->phy.disable_polarity_correction = B_FALSE;
1282 		hw->phy.ms_type = e1000_ms_hw_default; /* E1000_MASTER_SLAVE */
1283 	}
1284 
1285 	/*
1286 	 * Initialize link settings
1287 	 */
1288 	(void) igb_setup_link(igb, B_FALSE);
1289 
1290 	/*
1291 	 * Configure/Initialize hardware
1292 	 */
1293 	if (e1000_init_hw(hw) != E1000_SUCCESS) {
1294 		igb_error(igb, "Failed to initialize hardware");
1295 		goto init_adapter_fail;
1296 	}
1297 
1298 	/*
1299 	 * Disable wakeup control by default
1300 	 */
1301 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1302 
1303 	/*
1304 	 * Record phy info in hw struct
1305 	 */
1306 	(void) e1000_get_phy_info(hw);
1307 
1308 	/*
1309 	 * Make sure driver has control
1310 	 */
1311 	igb_get_driver_control(hw);
1312 
1313 	/*
1314 	 * Restore LED settings to the default from EEPROM
1315 	 * to meet the standard for Sun platforms.
1316 	 */
1317 	(void) e1000_cleanup_led(hw);
1318 
1319 	/*
1320 	 * Setup MSI-X interrupts
1321 	 */
1322 	if (igb->intr_type == DDI_INTR_TYPE_MSIX)
1323 		igb->capab->setup_msix(igb);
1324 
1325 	/*
1326 	 * Initialize unicast addresses.
1327 	 */
1328 	igb_init_unicst(igb);
1329 
1330 	/*
1331 	 * Setup and initialize the mctable structures.
1332 	 */
1333 	igb_setup_multicst(igb);
1334 
1335 	/*
1336 	 * Set interrupt throttling rate
1337 	 */
1338 	for (i = 0; i < igb->intr_cnt; i++)
1339 		E1000_WRITE_REG(hw, E1000_EITR(i), igb->intr_throttling[i]);
1340 
1341 	/*
1342 	 * Save the state of the phy
1343 	 */
1344 	igb_get_phy_state(igb);
1345 
1346 	return (IGB_SUCCESS);
1347 
1348 init_adapter_fail:
1349 	/*
1350 	 * Reset PHY if possible
1351 	 */
1352 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
1353 		(void) e1000_phy_hw_reset(hw);
1354 
1355 	return (IGB_FAILURE);
1356 }
1357 
1358 /*
1359  * igb_stop_adapter - Stop the adapter
1360  */
1361 static void
1362 igb_stop_adapter(igb_t *igb)
1363 {
1364 	struct e1000_hw *hw = &igb->hw;
1365 
1366 	ASSERT(mutex_owned(&igb->gen_lock));
1367 
1368 	/* Tell firmware driver is no longer in control */
1369 	igb_release_driver_control(hw);
1370 
1371 	/*
1372 	 * Reset the chipset
1373 	 */
1374 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1375 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1376 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1377 	}
1378 
1379 	/*
1380 	 * e1000_phy_hw_reset is not needed here, MAC reset above is sufficient
1381 	 */
1382 }
1383 
1384 /*
1385  * igb_reset - Reset the chipset and restart the driver.
1386  *
1387  * It involves stopping and re-starting the chipset,
1388  * and re-configuring the rx/tx rings.
1389  */
1390 static int
1391 igb_reset(igb_t *igb)
1392 {
1393 	int i;
1394 
1395 	mutex_enter(&igb->gen_lock);
1396 
1397 	ASSERT(igb->igb_state & IGB_STARTED);
1398 
1399 	/*
1400 	 * Disable the adapter interrupts to stop any rx/tx activities
1401 	 * before draining pending data and resetting hardware.
1402 	 */
1403 	igb_disable_adapter_interrupts(igb);
1404 
1405 	/*
1406 	 * Drain the pending transmit packets
1407 	 */
1408 	(void) igb_tx_drain(igb);
1409 
1410 	for (i = 0; i < igb->num_rx_rings; i++)
1411 		mutex_enter(&igb->rx_rings[i].rx_lock);
1412 	for (i = 0; i < igb->num_tx_rings; i++)
1413 		mutex_enter(&igb->tx_rings[i].tx_lock);
1414 
1415 	/*
1416 	 * Stop the adapter
1417 	 */
1418 	igb_stop_adapter(igb);
1419 
1420 	/*
1421 	 * Clean the pending tx data/resources
1422 	 */
1423 	igb_tx_clean(igb);
1424 
1425 	/*
1426 	 * Start the adapter
1427 	 */
1428 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1429 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1430 		goto reset_failure;
1431 	}
1432 
1433 	/*
1434 	 * Setup the rx/tx rings
1435 	 */
1436 	igb_setup_rings(igb);
1437 
1438 	/*
1439 	 * Enable adapter interrupts
1440 	 * The interrupts must be enabled after the driver state is START
1441 	 */
1442 	igb->capab->enable_intr(igb);
1443 
1444 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1445 		goto reset_failure;
1446 
1447 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1448 		goto reset_failure;
1449 
1450 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1451 		mutex_exit(&igb->tx_rings[i].tx_lock);
1452 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1453 		mutex_exit(&igb->rx_rings[i].rx_lock);
1454 
1455 	mutex_exit(&igb->gen_lock);
1456 
1457 	return (IGB_SUCCESS);
1458 
1459 reset_failure:
1460 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1461 		mutex_exit(&igb->tx_rings[i].tx_lock);
1462 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1463 		mutex_exit(&igb->rx_rings[i].rx_lock);
1464 
1465 	mutex_exit(&igb->gen_lock);
1466 
1467 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1468 
1469 	return (IGB_FAILURE);
1470 }
1471 
1472 /*
1473  * igb_tx_clean - Clean the pending transmit packets and DMA resources
1474  */
1475 static void
1476 igb_tx_clean(igb_t *igb)
1477 {
1478 	igb_tx_ring_t *tx_ring;
1479 	tx_control_block_t *tcb;
1480 	link_list_t pending_list;
1481 	uint32_t desc_num;
1482 	int i, j;
1483 
1484 	LINK_LIST_INIT(&pending_list);
1485 
1486 	for (i = 0; i < igb->num_tx_rings; i++) {
1487 		tx_ring = &igb->tx_rings[i];
1488 
1489 		mutex_enter(&tx_ring->recycle_lock);
1490 
1491 		/*
1492 		 * Clean the pending tx data - the pending packets in the
1493 		 * work_list that have no chances to be transmitted again.
1494 		 *
1495 		 * We must ensure the chipset is stopped or the link is down
1496 		 * before cleaning the transmit packets.
1497 		 */
1498 		desc_num = 0;
1499 		for (j = 0; j < tx_ring->ring_size; j++) {
1500 			tcb = tx_ring->work_list[j];
1501 			if (tcb != NULL) {
1502 				desc_num += tcb->desc_num;
1503 
1504 				tx_ring->work_list[j] = NULL;
1505 
1506 				igb_free_tcb(tcb);
1507 
1508 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1509 			}
1510 		}
1511 
1512 		if (desc_num > 0) {
1513 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1514 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1515 
1516 			/*
1517 			 * Reset the head and tail pointers of the tbd ring;
1518 			 * Reset the head write-back if it is enabled.
1519 			 */
1520 			tx_ring->tbd_head = 0;
1521 			tx_ring->tbd_tail = 0;
1522 			if (igb->tx_head_wb_enable)
1523 				*tx_ring->tbd_head_wb = 0;
1524 
1525 			E1000_WRITE_REG(&igb->hw, E1000_TDH(tx_ring->index), 0);
1526 			E1000_WRITE_REG(&igb->hw, E1000_TDT(tx_ring->index), 0);
1527 		}
1528 
1529 		mutex_exit(&tx_ring->recycle_lock);
1530 
1531 		/*
1532 		 * Add the tx control blocks in the pending list to
1533 		 * the free list.
1534 		 */
1535 		igb_put_free_list(tx_ring, &pending_list);
1536 	}
1537 }
1538 
1539 /*
1540  * igb_tx_drain - Drain the tx rings to allow pending packets to be transmitted
1541  */
1542 static boolean_t
1543 igb_tx_drain(igb_t *igb)
1544 {
1545 	igb_tx_ring_t *tx_ring;
1546 	boolean_t done;
1547 	int i, j;
1548 
1549 	/*
1550 	 * Wait for a specific time to allow pending tx packets
1551 	 * to be transmitted.
1552 	 *
1553 	 * Check the counter tbd_free to see if transmission is done.
1554 	 * No lock protection is needed here.
1555 	 *
1556 	 * Return B_TRUE if all pending packets have been transmitted;
1557 	 * Otherwise return B_FALSE;
1558 	 */
1559 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1560 
1561 		done = B_TRUE;
1562 		for (j = 0; j < igb->num_tx_rings; j++) {
1563 			tx_ring = &igb->tx_rings[j];
1564 			done = done &&
1565 			    (tx_ring->tbd_free == tx_ring->ring_size);
1566 		}
1567 
1568 		if (done)
1569 			break;
1570 
1571 		msec_delay(1);
1572 	}
1573 
1574 	return (done);
1575 }
1576 
1577 /*
1578  * igb_rx_drain - Wait for all rx buffers to be released by upper layer
1579  */
1580 static boolean_t
1581 igb_rx_drain(igb_t *igb)
1582 {
1583 	igb_rx_ring_t *rx_ring;
1584 	boolean_t done;
1585 	int i, j;
1586 
1587 	/*
1588 	 * Polling the rx free list to check if those rx buffers held by
1589 	 * the upper layer are released.
1590 	 *
1591 	 * Check the counter rcb_free to see if all pending buffers are
1592 	 * released. No lock protection is needed here.
1593 	 *
1594 	 * Return B_TRUE if all pending buffers have been released;
1595 	 * Otherwise return B_FALSE;
1596 	 */
1597 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1598 
1599 		done = B_TRUE;
1600 		for (j = 0; j < igb->num_rx_rings; j++) {
1601 			rx_ring = &igb->rx_rings[j];
1602 			done = done &&
1603 			    (rx_ring->rcb_free == rx_ring->free_list_size);
1604 		}
1605 
1606 		if (done)
1607 			break;
1608 
1609 		msec_delay(1);
1610 	}
1611 
1612 	return (done);
1613 }
1614 
1615 /*
1616  * igb_start - Start the driver/chipset
1617  */
1618 int
1619 igb_start(igb_t *igb)
1620 {
1621 	int i;
1622 
1623 	ASSERT(mutex_owned(&igb->gen_lock));
1624 
1625 	for (i = 0; i < igb->num_rx_rings; i++)
1626 		mutex_enter(&igb->rx_rings[i].rx_lock);
1627 	for (i = 0; i < igb->num_tx_rings; i++)
1628 		mutex_enter(&igb->tx_rings[i].tx_lock);
1629 
1630 	/*
1631 	 * Start the adapter
1632 	 */
1633 	if ((igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) == 0) {
1634 		if (igb_init_adapter(igb) != IGB_SUCCESS) {
1635 			igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1636 			goto start_failure;
1637 		}
1638 		igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
1639 
1640 		/*
1641 		 * Setup the rx/tx rings
1642 		 */
1643 		igb_setup_rings(igb);
1644 	}
1645 
1646 	/*
1647 	 * Enable adapter interrupts
1648 	 * The interrupts must be enabled after the driver state is START
1649 	 */
1650 	igb->capab->enable_intr(igb);
1651 
1652 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1653 		goto start_failure;
1654 
1655 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1656 		goto start_failure;
1657 
1658 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1659 		mutex_exit(&igb->tx_rings[i].tx_lock);
1660 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1661 		mutex_exit(&igb->rx_rings[i].rx_lock);
1662 
1663 	return (IGB_SUCCESS);
1664 
1665 start_failure:
1666 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1667 		mutex_exit(&igb->tx_rings[i].tx_lock);
1668 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1669 		mutex_exit(&igb->rx_rings[i].rx_lock);
1670 
1671 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1672 
1673 	return (IGB_FAILURE);
1674 }
1675 
1676 /*
1677  * igb_stop - Stop the driver/chipset
1678  */
1679 void
1680 igb_stop(igb_t *igb)
1681 {
1682 	int i;
1683 
1684 	ASSERT(mutex_owned(&igb->gen_lock));
1685 
1686 	igb->attach_progress &= ~ATTACH_PROGRESS_INIT_ADAPTER;
1687 
1688 	/*
1689 	 * Disable the adapter interrupts
1690 	 */
1691 	igb_disable_adapter_interrupts(igb);
1692 
1693 	/*
1694 	 * Drain the pending tx packets
1695 	 */
1696 	(void) igb_tx_drain(igb);
1697 
1698 	for (i = 0; i < igb->num_rx_rings; i++)
1699 		mutex_enter(&igb->rx_rings[i].rx_lock);
1700 	for (i = 0; i < igb->num_tx_rings; i++)
1701 		mutex_enter(&igb->tx_rings[i].tx_lock);
1702 
1703 	/*
1704 	 * Stop the adapter
1705 	 */
1706 	igb_stop_adapter(igb);
1707 
1708 	/*
1709 	 * Clean the pending tx data/resources
1710 	 */
1711 	igb_tx_clean(igb);
1712 
1713 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1714 		mutex_exit(&igb->tx_rings[i].tx_lock);
1715 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1716 		mutex_exit(&igb->rx_rings[i].rx_lock);
1717 
1718 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1719 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1720 }
1721 
1722 /*
1723  * igb_alloc_rings - Allocate memory space for rx/tx rings
1724  */
1725 static int
1726 igb_alloc_rings(igb_t *igb)
1727 {
1728 	/*
1729 	 * Allocate memory space for rx rings
1730 	 */
1731 	igb->rx_rings = kmem_zalloc(
1732 	    sizeof (igb_rx_ring_t) * igb->num_rx_rings,
1733 	    KM_NOSLEEP);
1734 
1735 	if (igb->rx_rings == NULL) {
1736 		return (IGB_FAILURE);
1737 	}
1738 
1739 	/*
1740 	 * Allocate memory space for tx rings
1741 	 */
1742 	igb->tx_rings = kmem_zalloc(
1743 	    sizeof (igb_tx_ring_t) * igb->num_tx_rings,
1744 	    KM_NOSLEEP);
1745 
1746 	if (igb->tx_rings == NULL) {
1747 		kmem_free(igb->rx_rings,
1748 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1749 		igb->rx_rings = NULL;
1750 		return (IGB_FAILURE);
1751 	}
1752 
1753 	/*
1754 	 * Allocate memory space for rx ring groups
1755 	 */
1756 	igb->rx_groups = kmem_zalloc(
1757 	    sizeof (igb_rx_group_t) * igb->num_rx_groups,
1758 	    KM_NOSLEEP);
1759 
1760 	if (igb->rx_groups == NULL) {
1761 		kmem_free(igb->rx_rings,
1762 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1763 		kmem_free(igb->tx_rings,
1764 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1765 		igb->rx_rings = NULL;
1766 		igb->tx_rings = NULL;
1767 		return (IGB_FAILURE);
1768 	}
1769 
1770 	return (IGB_SUCCESS);
1771 }
1772 
1773 /*
1774  * igb_free_rings - Free the memory space of rx/tx rings.
1775  */
1776 static void
1777 igb_free_rings(igb_t *igb)
1778 {
1779 	if (igb->rx_rings != NULL) {
1780 		kmem_free(igb->rx_rings,
1781 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1782 		igb->rx_rings = NULL;
1783 	}
1784 
1785 	if (igb->tx_rings != NULL) {
1786 		kmem_free(igb->tx_rings,
1787 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1788 		igb->tx_rings = NULL;
1789 	}
1790 
1791 	if (igb->rx_groups != NULL) {
1792 		kmem_free(igb->rx_groups,
1793 		    sizeof (igb_rx_group_t) * igb->num_rx_groups);
1794 		igb->rx_groups = NULL;
1795 	}
1796 }
1797 
1798 /*
1799  * igb_setup_rings - Setup rx/tx rings
1800  */
1801 static void
1802 igb_setup_rings(igb_t *igb)
1803 {
1804 	/*
1805 	 * Setup the rx/tx rings, including the following:
1806 	 *
1807 	 * 1. Setup the descriptor ring and the control block buffers;
1808 	 * 2. Initialize necessary registers for receive/transmit;
1809 	 * 3. Initialize software pointers/parameters for receive/transmit;
1810 	 */
1811 	igb_setup_rx(igb);
1812 
1813 	igb_setup_tx(igb);
1814 
1815 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1816 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1817 }
1818 
1819 static void
1820 igb_setup_rx_ring(igb_rx_ring_t *rx_ring)
1821 {
1822 	igb_t *igb = rx_ring->igb;
1823 	struct e1000_hw *hw = &igb->hw;
1824 	rx_control_block_t *rcb;
1825 	union e1000_adv_rx_desc	*rbd;
1826 	uint32_t size;
1827 	uint32_t buf_low;
1828 	uint32_t buf_high;
1829 	uint32_t rxdctl;
1830 	int i;
1831 
1832 	ASSERT(mutex_owned(&rx_ring->rx_lock));
1833 	ASSERT(mutex_owned(&igb->gen_lock));
1834 
1835 	/*
1836 	 * Initialize descriptor ring with buffer addresses
1837 	 */
1838 	for (i = 0; i < igb->rx_ring_size; i++) {
1839 		rcb = rx_ring->work_list[i];
1840 		rbd = &rx_ring->rbd_ring[i];
1841 
1842 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
1843 		rbd->read.hdr_addr = NULL;
1844 	}
1845 
1846 	/*
1847 	 * Initialize the base address registers
1848 	 */
1849 	buf_low = (uint32_t)rx_ring->rbd_area.dma_address;
1850 	buf_high = (uint32_t)(rx_ring->rbd_area.dma_address >> 32);
1851 	E1000_WRITE_REG(hw, E1000_RDBAH(rx_ring->index), buf_high);
1852 	E1000_WRITE_REG(hw, E1000_RDBAL(rx_ring->index), buf_low);
1853 
1854 	/*
1855 	 * Initialize the length register
1856 	 */
1857 	size = rx_ring->ring_size * sizeof (union e1000_adv_rx_desc);
1858 	E1000_WRITE_REG(hw, E1000_RDLEN(rx_ring->index), size);
1859 
1860 	/*
1861 	 * Initialize buffer size & descriptor type
1862 	 */
1863 	E1000_WRITE_REG(hw, E1000_SRRCTL(rx_ring->index),
1864 	    ((igb->rx_buf_size >> E1000_SRRCTL_BSIZEPKT_SHIFT) |
1865 	    E1000_SRRCTL_DESCTYPE_ADV_ONEBUF));
1866 
1867 	/*
1868 	 * Setup the Receive Descriptor Control Register (RXDCTL)
1869 	 */
1870 	rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(rx_ring->index));
1871 	rxdctl &= igb->capab->rxdctl_mask;
1872 	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1873 	rxdctl |= 16;		/* pthresh */
1874 	rxdctl |= 8 << 8;	/* hthresh */
1875 	rxdctl |= 1 << 16;	/* wthresh */
1876 	E1000_WRITE_REG(hw, E1000_RXDCTL(rx_ring->index), rxdctl);
1877 
1878 	rx_ring->rbd_next = 0;
1879 
1880 	/*
1881 	 * Note: Considering the case that the chipset is being reset
1882 	 * and there are still some buffers held by the upper layer,
1883 	 * we should not reset the values of rcb_head, rcb_tail and
1884 	 * rcb_free;
1885 	 */
1886 	if (igb->igb_state == IGB_UNKNOWN) {
1887 		rx_ring->rcb_head = 0;
1888 		rx_ring->rcb_tail = 0;
1889 		rx_ring->rcb_free = rx_ring->free_list_size;
1890 	}
1891 }
1892 
1893 static void
1894 igb_setup_rx(igb_t *igb)
1895 {
1896 	igb_rx_ring_t *rx_ring;
1897 	igb_rx_group_t *rx_group;
1898 	struct e1000_hw *hw = &igb->hw;
1899 	uint32_t rctl, rxcsum;
1900 	uint32_t ring_per_group;
1901 	int i;
1902 
1903 	/*
1904 	 * Setup the Receive Control Register (RCTL), and enable the
1905 	 * receiver. The initial configuration is to: enable the receiver,
1906 	 * accept broadcasts, discard bad packets, accept long packets,
1907 	 * disable VLAN filter checking, and set receive buffer size to
1908 	 * 2k.  For 82575, also set the receive descriptor minimum
1909 	 * threshold size to 1/2 the ring.
1910 	 */
1911 	rctl = E1000_READ_REG(hw, E1000_RCTL);
1912 
1913 	/*
1914 	 * Clear the field used for wakeup control.  This driver doesn't do
1915 	 * wakeup but leave this here for completeness.
1916 	 */
1917 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1918 
1919 	switch (hw->mac.type) {
1920 	case e1000_82575:
1921 		rctl |= (E1000_RCTL_EN |	/* Enable Receive Unit */
1922 		    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
1923 		    E1000_RCTL_LPE |		/* Large Packet Enable */
1924 						/* Multicast filter offset */
1925 		    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
1926 		    E1000_RCTL_RDMTS_HALF |	/* rx descriptor threshold */
1927 		    E1000_RCTL_SECRC);		/* Strip Ethernet CRC */
1928 		break;
1929 
1930 	case e1000_82576:
1931 		rctl |= (E1000_RCTL_EN |	/* Enable Receive Unit */
1932 		    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
1933 		    E1000_RCTL_LPE |		/* Large Packet Enable */
1934 						/* Multicast filter offset */
1935 		    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
1936 		    E1000_RCTL_SECRC);		/* Strip Ethernet CRC */
1937 		break;
1938 
1939 	default:
1940 		igb_log(igb, "unsupported MAC type: %d", hw->mac.type);
1941 		return;	/* should never come here; this will cause rx failure */
1942 	}
1943 
1944 	for (i = 0; i < igb->num_rx_groups; i++) {
1945 		rx_group = &igb->rx_groups[i];
1946 		rx_group->index = i;
1947 		rx_group->igb = igb;
1948 	}
1949 
1950 	/*
1951 	 * Set up all rx descriptor rings - must be called before receive unit
1952 	 * enabled.
1953 	 */
1954 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
1955 	for (i = 0; i < igb->num_rx_rings; i++) {
1956 		rx_ring = &igb->rx_rings[i];
1957 		igb_setup_rx_ring(rx_ring);
1958 
1959 		/*
1960 		 * Map a ring to a group by assigning a group index
1961 		 */
1962 		rx_ring->group_index = i / ring_per_group;
1963 	}
1964 
1965 	/*
1966 	 * Setup the Rx Long Packet Max Length register
1967 	 */
1968 	E1000_WRITE_REG(hw, E1000_RLPML, igb->max_frame_size);
1969 
1970 	/*
1971 	 * Hardware checksum settings
1972 	 */
1973 	if (igb->rx_hcksum_enable) {
1974 		rxcsum =
1975 		    E1000_RXCSUM_TUOFL |	/* TCP/UDP checksum */
1976 		    E1000_RXCSUM_IPOFL;		/* IP checksum */
1977 
1978 		E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
1979 	}
1980 
1981 	/*
1982 	 * Setup classify and RSS for multiple receive queues
1983 	 */
1984 	switch (igb->vmdq_mode) {
1985 	case E1000_VMDQ_OFF:
1986 		/*
1987 		 * One ring group, only RSS is needed when more than
1988 		 * one ring enabled.
1989 		 */
1990 		if (igb->num_rx_rings > 1)
1991 			igb_setup_rss(igb);
1992 		break;
1993 	case E1000_VMDQ_MAC:
1994 		/*
1995 		 * Multiple groups, each group has one ring,
1996 		 * only the MAC classification is needed.
1997 		 */
1998 		igb_setup_mac_classify(igb);
1999 		break;
2000 	case E1000_VMDQ_MAC_RSS:
2001 		/*
2002 		 * Multiple groups and multiple rings, both
2003 		 * MAC classification and RSS are needed.
2004 		 */
2005 		igb_setup_mac_rss_classify(igb);
2006 		break;
2007 	}
2008 
2009 	/*
2010 	 * Enable the receive unit - must be done after all
2011 	 * the rx setup above.
2012 	 */
2013 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2014 
2015 	/*
2016 	 * Initialize all adapter ring head & tail pointers - must
2017 	 * be done after receive unit is enabled
2018 	 */
2019 	for (i = 0; i < igb->num_rx_rings; i++) {
2020 		rx_ring = &igb->rx_rings[i];
2021 		E1000_WRITE_REG(hw, E1000_RDH(i), 0);
2022 		E1000_WRITE_REG(hw, E1000_RDT(i), rx_ring->ring_size - 1);
2023 	}
2024 
2025 	/*
2026 	 * 82575 with manageability enabled needs a special flush to make
2027 	 * sure the fifos start clean.
2028 	 */
2029 	if ((hw->mac.type == e1000_82575) &&
2030 	    (E1000_READ_REG(hw, E1000_MANC) & E1000_MANC_RCV_TCO_EN)) {
2031 		e1000_rx_fifo_flush_82575(hw);
2032 	}
2033 }
2034 
2035 static void
2036 igb_setup_tx_ring(igb_tx_ring_t *tx_ring)
2037 {
2038 	igb_t *igb = tx_ring->igb;
2039 	struct e1000_hw *hw = &igb->hw;
2040 	uint32_t size;
2041 	uint32_t buf_low;
2042 	uint32_t buf_high;
2043 	uint32_t reg_val;
2044 
2045 	ASSERT(mutex_owned(&tx_ring->tx_lock));
2046 	ASSERT(mutex_owned(&igb->gen_lock));
2047 
2048 
2049 	/*
2050 	 * Initialize the length register
2051 	 */
2052 	size = tx_ring->ring_size * sizeof (union e1000_adv_tx_desc);
2053 	E1000_WRITE_REG(hw, E1000_TDLEN(tx_ring->index), size);
2054 
2055 	/*
2056 	 * Initialize the base address registers
2057 	 */
2058 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
2059 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
2060 	E1000_WRITE_REG(hw, E1000_TDBAL(tx_ring->index), buf_low);
2061 	E1000_WRITE_REG(hw, E1000_TDBAH(tx_ring->index), buf_high);
2062 
2063 	/*
2064 	 * Setup head & tail pointers
2065 	 */
2066 	E1000_WRITE_REG(hw, E1000_TDH(tx_ring->index), 0);
2067 	E1000_WRITE_REG(hw, E1000_TDT(tx_ring->index), 0);
2068 
2069 	/*
2070 	 * Setup head write-back
2071 	 */
2072 	if (igb->tx_head_wb_enable) {
2073 		/*
2074 		 * The memory of the head write-back is allocated using
2075 		 * the extra tbd beyond the tail of the tbd ring.
2076 		 */
2077 		tx_ring->tbd_head_wb = (uint32_t *)
2078 		    ((uintptr_t)tx_ring->tbd_area.address + size);
2079 		*tx_ring->tbd_head_wb = 0;
2080 
2081 		buf_low = (uint32_t)
2082 		    (tx_ring->tbd_area.dma_address + size);
2083 		buf_high = (uint32_t)
2084 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
2085 
2086 		/* Set the head write-back enable bit */
2087 		buf_low |= E1000_TX_HEAD_WB_ENABLE;
2088 
2089 		E1000_WRITE_REG(hw, E1000_TDWBAL(tx_ring->index), buf_low);
2090 		E1000_WRITE_REG(hw, E1000_TDWBAH(tx_ring->index), buf_high);
2091 
2092 		/*
2093 		 * Turn off relaxed ordering for head write back or it will
2094 		 * cause problems with the tx recycling
2095 		 */
2096 		reg_val = E1000_READ_REG(hw,
2097 		    E1000_DCA_TXCTRL(tx_ring->index));
2098 		reg_val &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
2099 		E1000_WRITE_REG(hw,
2100 		    E1000_DCA_TXCTRL(tx_ring->index), reg_val);
2101 	} else {
2102 		tx_ring->tbd_head_wb = NULL;
2103 	}
2104 
2105 	tx_ring->tbd_head = 0;
2106 	tx_ring->tbd_tail = 0;
2107 	tx_ring->tbd_free = tx_ring->ring_size;
2108 
2109 	/*
2110 	 * Note: for the case that the chipset is being reset, we should not
2111 	 * reset the values of tcb_head, tcb_tail. And considering there might
2112 	 * still be some packets kept in the pending_list, we should not assert
2113 	 * (tcb_free == free_list_size) here.
2114 	 */
2115 	if (igb->igb_state == IGB_UNKNOWN) {
2116 		tx_ring->tcb_head = 0;
2117 		tx_ring->tcb_tail = 0;
2118 		tx_ring->tcb_free = tx_ring->free_list_size;
2119 	}
2120 
2121 	/*
2122 	 * Enable TXDCTL per queue
2123 	 */
2124 	reg_val = E1000_READ_REG(hw, E1000_TXDCTL(tx_ring->index));
2125 	reg_val |= E1000_TXDCTL_QUEUE_ENABLE;
2126 	E1000_WRITE_REG(hw, E1000_TXDCTL(tx_ring->index), reg_val);
2127 
2128 	/*
2129 	 * Initialize hardware checksum offload settings
2130 	 */
2131 	bzero(&tx_ring->tx_context, sizeof (tx_context_t));
2132 }
2133 
2134 static void
2135 igb_setup_tx(igb_t *igb)
2136 {
2137 	igb_tx_ring_t *tx_ring;
2138 	struct e1000_hw *hw = &igb->hw;
2139 	uint32_t reg_val;
2140 	int i;
2141 
2142 	for (i = 0; i < igb->num_tx_rings; i++) {
2143 		tx_ring = &igb->tx_rings[i];
2144 		igb_setup_tx_ring(tx_ring);
2145 	}
2146 
2147 	/*
2148 	 * Setup the Transmit Control Register (TCTL)
2149 	 */
2150 	reg_val = E1000_READ_REG(hw, E1000_TCTL);
2151 	reg_val &= ~E1000_TCTL_CT;
2152 	reg_val |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2153 	    (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2154 
2155 	/* Enable transmits */
2156 	reg_val |= E1000_TCTL_EN;
2157 
2158 	E1000_WRITE_REG(hw, E1000_TCTL, reg_val);
2159 }
2160 
2161 /*
2162  * igb_setup_rss - Setup receive-side scaling feature
2163  */
2164 static void
2165 igb_setup_rss(igb_t *igb)
2166 {
2167 	struct e1000_hw *hw = &igb->hw;
2168 	uint32_t i, mrqc, rxcsum;
2169 	int shift = 0;
2170 	uint32_t random;
2171 	union e1000_reta {
2172 		uint32_t	dword;
2173 		uint8_t		bytes[4];
2174 	} reta;
2175 
2176 	/* Setup the Redirection Table */
2177 	if (hw->mac.type == e1000_82576) {
2178 		shift = 0;
2179 	} else if (hw->mac.type == e1000_82575) {
2180 		shift = 6;
2181 	}
2182 	for (i = 0; i < (32 * 4); i++) {
2183 		reta.bytes[i & 3] = (i % igb->num_rx_rings) << shift;
2184 		if ((i & 3) == 3) {
2185 			E1000_WRITE_REG(hw,
2186 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2187 		}
2188 	}
2189 
2190 	/* Fill out hash function seeds */
2191 	for (i = 0; i < 10; i++) {
2192 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2193 		    sizeof (uint32_t));
2194 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2195 	}
2196 
2197 	/* Setup the Multiple Receive Queue Control register */
2198 	mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2199 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2200 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2201 	    E1000_MRQC_RSS_FIELD_IPV6 |
2202 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2203 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2204 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2205 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2206 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2207 
2208 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2209 
2210 	/*
2211 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2212 	 *
2213 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2214 	 * checksum offloading provided by the 82575 chipset besides the IP
2215 	 * header checksum offloading and the TCP/UDP checksum offloading.
2216 	 * The Packet Checksum is by default computed over the entire packet
2217 	 * from the first byte of the DA through the last byte of the CRC,
2218 	 * including the Ethernet and IP headers.
2219 	 *
2220 	 * It is a hardware limitation that Packet Checksum is mutually
2221 	 * exclusive with RSS.
2222 	 */
2223 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2224 	rxcsum |= E1000_RXCSUM_PCSD;
2225 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2226 }
2227 
2228 /*
2229  * igb_setup_mac_rss_classify - Setup MAC classification and rss
2230  */
2231 static void
2232 igb_setup_mac_rss_classify(igb_t *igb)
2233 {
2234 	struct e1000_hw *hw = &igb->hw;
2235 	uint32_t i, mrqc, vmdctl, rxcsum;
2236 	uint32_t ring_per_group;
2237 	int shift_group0, shift_group1;
2238 	uint32_t random;
2239 	union e1000_reta {
2240 		uint32_t	dword;
2241 		uint8_t		bytes[4];
2242 	} reta;
2243 
2244 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2245 
2246 	/* Setup the Redirection Table, it is shared between two groups */
2247 	shift_group0 = 2;
2248 	shift_group1 = 6;
2249 	for (i = 0; i < (32 * 4); i++) {
2250 		reta.bytes[i & 3] = ((i % ring_per_group) << shift_group0) |
2251 		    ((ring_per_group + (i % ring_per_group)) << shift_group1);
2252 		if ((i & 3) == 3) {
2253 			E1000_WRITE_REG(hw,
2254 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2255 		}
2256 	}
2257 
2258 	/* Fill out hash function seeds */
2259 	for (i = 0; i < 10; i++) {
2260 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2261 		    sizeof (uint32_t));
2262 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2263 	}
2264 
2265 	/*
2266 	 * Setup the Multiple Receive Queue Control register,
2267 	 * enable VMDq based on packet destination MAC address and RSS.
2268 	 */
2269 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_RSS_GROUP;
2270 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2271 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2272 	    E1000_MRQC_RSS_FIELD_IPV6 |
2273 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2274 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2275 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2276 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2277 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2278 
2279 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2280 
2281 
2282 	/* Define the default group and default queues */
2283 	vmdctl = E1000_VMDQ_MAC_GROUP_DEFAULT_QUEUE;
2284 	E1000_WRITE_REG(hw, E1000_VT_CTL, vmdctl);
2285 
2286 	/*
2287 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2288 	 *
2289 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2290 	 * checksum offloading provided by the 82575 chipset besides the IP
2291 	 * header checksum offloading and the TCP/UDP checksum offloading.
2292 	 * The Packet Checksum is by default computed over the entire packet
2293 	 * from the first byte of the DA through the last byte of the CRC,
2294 	 * including the Ethernet and IP headers.
2295 	 *
2296 	 * It is a hardware limitation that Packet Checksum is mutually
2297 	 * exclusive with RSS.
2298 	 */
2299 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2300 	rxcsum |= E1000_RXCSUM_PCSD;
2301 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2302 }
2303 
2304 /*
2305  * igb_setup_mac_classify - Setup MAC classification feature
2306  */
2307 static void
2308 igb_setup_mac_classify(igb_t *igb)
2309 {
2310 	struct e1000_hw *hw = &igb->hw;
2311 	uint32_t mrqc, rxcsum;
2312 
2313 	/*
2314 	 * Setup the Multiple Receive Queue Control register,
2315 	 * enable VMDq based on packet destination MAC address.
2316 	 */
2317 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_GROUP;
2318 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2319 
2320 	/*
2321 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2322 	 *
2323 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2324 	 * checksum offloading provided by the 82575 chipset besides the IP
2325 	 * header checksum offloading and the TCP/UDP checksum offloading.
2326 	 * The Packet Checksum is by default computed over the entire packet
2327 	 * from the first byte of the DA through the last byte of the CRC,
2328 	 * including the Ethernet and IP headers.
2329 	 *
2330 	 * It is a hardware limitation that Packet Checksum is mutually
2331 	 * exclusive with RSS.
2332 	 */
2333 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2334 	rxcsum |= E1000_RXCSUM_PCSD;
2335 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2336 
2337 }
2338 
2339 /*
2340  * igb_init_unicst - Initialize the unicast addresses
2341  */
2342 static void
2343 igb_init_unicst(igb_t *igb)
2344 {
2345 	struct e1000_hw *hw = &igb->hw;
2346 	int slot;
2347 
2348 	/*
2349 	 * Here we should consider two situations:
2350 	 *
2351 	 * 1. Chipset is initialized the first time
2352 	 *    Initialize the multiple unicast addresses, and
2353 	 *    save the default MAC address.
2354 	 *
2355 	 * 2. Chipset is reset
2356 	 *    Recover the multiple unicast addresses from the
2357 	 *    software data structure to the RAR registers.
2358 	 */
2359 
2360 	/*
2361 	 * Clear the default MAC address in the RAR0 rgister,
2362 	 * which is loaded from EEPROM when system boot or chipreset,
2363 	 * this will cause the conficts with add_mac/rem_mac entry
2364 	 * points when VMDq is enabled. For this reason, the RAR0
2365 	 * must be cleared for both cases mentioned above.
2366 	 */
2367 	e1000_rar_clear(hw, 0);
2368 
2369 	if (!igb->unicst_init) {
2370 
2371 		/* Initialize the multiple unicast addresses */
2372 		igb->unicst_total = MAX_NUM_UNICAST_ADDRESSES;
2373 		igb->unicst_avail = igb->unicst_total;
2374 
2375 		for (slot = 0; slot < igb->unicst_total; slot++)
2376 			igb->unicst_addr[slot].mac.set = 0;
2377 
2378 		igb->unicst_init = B_TRUE;
2379 	} else {
2380 		/* Re-configure the RAR registers */
2381 		for (slot = 0; slot < igb->unicst_total; slot++) {
2382 			e1000_rar_set_vmdq(hw, igb->unicst_addr[slot].mac.addr,
2383 			    slot, igb->vmdq_mode,
2384 			    igb->unicst_addr[slot].mac.group_index);
2385 		}
2386 	}
2387 
2388 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2389 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2390 }
2391 
2392 /*
2393  * igb_unicst_find - Find the slot for the specified unicast address
2394  */
2395 int
2396 igb_unicst_find(igb_t *igb, const uint8_t *mac_addr)
2397 {
2398 	int slot;
2399 
2400 	ASSERT(mutex_owned(&igb->gen_lock));
2401 
2402 	for (slot = 0; slot < igb->unicst_total; slot++) {
2403 		if (bcmp(igb->unicst_addr[slot].mac.addr,
2404 		    mac_addr, ETHERADDRL) == 0)
2405 			return (slot);
2406 	}
2407 
2408 	return (-1);
2409 }
2410 
2411 /*
2412  * igb_unicst_set - Set the unicast address to the specified slot
2413  */
2414 int
2415 igb_unicst_set(igb_t *igb, const uint8_t *mac_addr,
2416     int slot)
2417 {
2418 	struct e1000_hw *hw = &igb->hw;
2419 
2420 	ASSERT(mutex_owned(&igb->gen_lock));
2421 
2422 	/*
2423 	 * Save the unicast address in the software data structure
2424 	 */
2425 	bcopy(mac_addr, igb->unicst_addr[slot].mac.addr, ETHERADDRL);
2426 
2427 	/*
2428 	 * Set the unicast address to the RAR register
2429 	 */
2430 	e1000_rar_set(hw, (uint8_t *)mac_addr, slot);
2431 
2432 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2433 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2434 		return (EIO);
2435 	}
2436 
2437 	return (0);
2438 }
2439 
2440 /*
2441  * igb_multicst_add - Add a multicst address
2442  */
2443 int
2444 igb_multicst_add(igb_t *igb, const uint8_t *multiaddr)
2445 {
2446 	struct ether_addr *new_table;
2447 	size_t new_len;
2448 	size_t old_len;
2449 
2450 	ASSERT(mutex_owned(&igb->gen_lock));
2451 
2452 	if ((multiaddr[0] & 01) == 0) {
2453 		igb_error(igb, "Illegal multicast address");
2454 		return (EINVAL);
2455 	}
2456 
2457 	if (igb->mcast_count >= igb->mcast_max_num) {
2458 		igb_error(igb, "Adapter requested more than %d mcast addresses",
2459 		    igb->mcast_max_num);
2460 		return (ENOENT);
2461 	}
2462 
2463 	if (igb->mcast_count == igb->mcast_alloc_count) {
2464 		old_len = igb->mcast_alloc_count *
2465 		    sizeof (struct ether_addr);
2466 		new_len = (igb->mcast_alloc_count + MCAST_ALLOC_COUNT) *
2467 		    sizeof (struct ether_addr);
2468 
2469 		new_table = kmem_alloc(new_len, KM_NOSLEEP);
2470 		if (new_table == NULL) {
2471 			igb_error(igb,
2472 			    "Not enough memory to alloc mcast table");
2473 			return (ENOMEM);
2474 		}
2475 
2476 		if (igb->mcast_table != NULL) {
2477 			bcopy(igb->mcast_table, new_table, old_len);
2478 			kmem_free(igb->mcast_table, old_len);
2479 		}
2480 		igb->mcast_alloc_count += MCAST_ALLOC_COUNT;
2481 		igb->mcast_table = new_table;
2482 	}
2483 
2484 	bcopy(multiaddr,
2485 	    &igb->mcast_table[igb->mcast_count], ETHERADDRL);
2486 	igb->mcast_count++;
2487 
2488 	/*
2489 	 * Update the multicast table in the hardware
2490 	 */
2491 	igb_setup_multicst(igb);
2492 
2493 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2494 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2495 		return (EIO);
2496 	}
2497 
2498 	return (0);
2499 }
2500 
2501 /*
2502  * igb_multicst_remove - Remove a multicst address
2503  */
2504 int
2505 igb_multicst_remove(igb_t *igb, const uint8_t *multiaddr)
2506 {
2507 	struct ether_addr *new_table;
2508 	size_t new_len;
2509 	size_t old_len;
2510 	int i;
2511 
2512 	ASSERT(mutex_owned(&igb->gen_lock));
2513 
2514 	for (i = 0; i < igb->mcast_count; i++) {
2515 		if (bcmp(multiaddr, &igb->mcast_table[i],
2516 		    ETHERADDRL) == 0) {
2517 			for (i++; i < igb->mcast_count; i++) {
2518 				igb->mcast_table[i - 1] =
2519 				    igb->mcast_table[i];
2520 			}
2521 			igb->mcast_count--;
2522 			break;
2523 		}
2524 	}
2525 
2526 	if ((igb->mcast_alloc_count - igb->mcast_count) >
2527 	    MCAST_ALLOC_COUNT) {
2528 		old_len = igb->mcast_alloc_count *
2529 		    sizeof (struct ether_addr);
2530 		new_len = (igb->mcast_alloc_count - MCAST_ALLOC_COUNT) *
2531 		    sizeof (struct ether_addr);
2532 
2533 		new_table = kmem_alloc(new_len, KM_NOSLEEP);
2534 		if (new_table != NULL) {
2535 			bcopy(igb->mcast_table, new_table, new_len);
2536 			kmem_free(igb->mcast_table, old_len);
2537 			igb->mcast_alloc_count -= MCAST_ALLOC_COUNT;
2538 			igb->mcast_table = new_table;
2539 		}
2540 	}
2541 
2542 	/*
2543 	 * Update the multicast table in the hardware
2544 	 */
2545 	igb_setup_multicst(igb);
2546 
2547 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2548 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2549 		return (EIO);
2550 	}
2551 
2552 	return (0);
2553 }
2554 
2555 static void
2556 igb_release_multicast(igb_t *igb)
2557 {
2558 	if (igb->mcast_table != NULL) {
2559 		kmem_free(igb->mcast_table,
2560 		    igb->mcast_alloc_count * sizeof (struct ether_addr));
2561 		igb->mcast_table = NULL;
2562 	}
2563 }
2564 
2565 /*
2566  * igb_setup_multicast - setup multicast data structures
2567  *
2568  * This routine initializes all of the multicast related structures
2569  * and save them in the hardware registers.
2570  */
2571 static void
2572 igb_setup_multicst(igb_t *igb)
2573 {
2574 	uint8_t *mc_addr_list;
2575 	uint32_t mc_addr_count;
2576 	struct e1000_hw *hw = &igb->hw;
2577 
2578 	ASSERT(mutex_owned(&igb->gen_lock));
2579 	ASSERT(igb->mcast_count <= igb->mcast_max_num);
2580 
2581 	mc_addr_list = (uint8_t *)igb->mcast_table;
2582 	mc_addr_count = igb->mcast_count;
2583 
2584 	/*
2585 	 * Update the multicase addresses to the MTA registers
2586 	 */
2587 	e1000_update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2588 }
2589 
2590 /*
2591  * igb_get_conf - Get driver configurations set in driver.conf
2592  *
2593  * This routine gets user-configured values out of the configuration
2594  * file igb.conf.
2595  *
2596  * For each configurable value, there is a minimum, a maximum, and a
2597  * default.
2598  * If user does not configure a value, use the default.
2599  * If user configures below the minimum, use the minumum.
2600  * If user configures above the maximum, use the maxumum.
2601  */
2602 static void
2603 igb_get_conf(igb_t *igb)
2604 {
2605 	struct e1000_hw *hw = &igb->hw;
2606 	uint32_t default_mtu;
2607 	uint32_t flow_control;
2608 	uint32_t ring_per_group;
2609 	int i;
2610 
2611 	/*
2612 	 * igb driver supports the following user configurations:
2613 	 *
2614 	 * Link configurations:
2615 	 *    adv_autoneg_cap
2616 	 *    adv_1000fdx_cap
2617 	 *    adv_100fdx_cap
2618 	 *    adv_100hdx_cap
2619 	 *    adv_10fdx_cap
2620 	 *    adv_10hdx_cap
2621 	 * Note: 1000hdx is not supported.
2622 	 *
2623 	 * Jumbo frame configuration:
2624 	 *    default_mtu
2625 	 *
2626 	 * Ethernet flow control configuration:
2627 	 *    flow_control
2628 	 *
2629 	 * Multiple rings configurations:
2630 	 *    tx_queue_number
2631 	 *    tx_ring_size
2632 	 *    rx_queue_number
2633 	 *    rx_ring_size
2634 	 *
2635 	 * Call igb_get_prop() to get the value for a specific
2636 	 * configuration parameter.
2637 	 */
2638 
2639 	/*
2640 	 * Link configurations
2641 	 */
2642 	igb->param_adv_autoneg_cap = igb_get_prop(igb,
2643 	    PROP_ADV_AUTONEG_CAP, 0, 1, 1);
2644 	igb->param_adv_1000fdx_cap = igb_get_prop(igb,
2645 	    PROP_ADV_1000FDX_CAP, 0, 1, 1);
2646 	igb->param_adv_100fdx_cap = igb_get_prop(igb,
2647 	    PROP_ADV_100FDX_CAP, 0, 1, 1);
2648 	igb->param_adv_100hdx_cap = igb_get_prop(igb,
2649 	    PROP_ADV_100HDX_CAP, 0, 1, 1);
2650 	igb->param_adv_10fdx_cap = igb_get_prop(igb,
2651 	    PROP_ADV_10FDX_CAP, 0, 1, 1);
2652 	igb->param_adv_10hdx_cap = igb_get_prop(igb,
2653 	    PROP_ADV_10HDX_CAP, 0, 1, 1);
2654 
2655 	/*
2656 	 * Jumbo frame configurations
2657 	 */
2658 	default_mtu = igb_get_prop(igb, PROP_DEFAULT_MTU,
2659 	    MIN_MTU, MAX_MTU, DEFAULT_MTU);
2660 
2661 	igb->max_frame_size = default_mtu +
2662 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
2663 
2664 	/*
2665 	 * Ethernet flow control configuration
2666 	 */
2667 	flow_control = igb_get_prop(igb, PROP_FLOW_CONTROL,
2668 	    e1000_fc_none, 4, e1000_fc_full);
2669 	if (flow_control == 4)
2670 		flow_control = e1000_fc_default;
2671 
2672 	hw->fc.requested_mode = flow_control;
2673 
2674 	/*
2675 	 * Multiple rings configurations
2676 	 */
2677 	igb->tx_ring_size = igb_get_prop(igb, PROP_TX_RING_SIZE,
2678 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
2679 	igb->rx_ring_size = igb_get_prop(igb, PROP_RX_RING_SIZE,
2680 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
2681 
2682 	igb->mr_enable = igb_get_prop(igb, PROP_MR_ENABLE, 0, 1, 0);
2683 	igb->num_rx_groups = igb_get_prop(igb, PROP_RX_GROUP_NUM,
2684 	    MIN_RX_GROUP_NUM, MAX_RX_GROUP_NUM, DEFAULT_RX_GROUP_NUM);
2685 	/*
2686 	 * Currently we do not support VMDq for 82576.
2687 	 * If it is e1000_82576, set num_rx_groups to 1.
2688 	 */
2689 	if (hw->mac.type == e1000_82576)
2690 		igb->num_rx_groups = 1;
2691 
2692 	if (igb->mr_enable) {
2693 		igb->num_tx_rings = igb->capab->def_tx_que_num;
2694 		igb->num_rx_rings = igb->capab->def_rx_que_num;
2695 	} else {
2696 		igb->num_tx_rings = 1;
2697 		igb->num_rx_rings = 1;
2698 
2699 		if (igb->num_rx_groups > 1) {
2700 			igb_error(igb,
2701 			    "Invalid rx groups number. Please enable multiple "
2702 			    "rings first");
2703 			igb->num_rx_groups = 1;
2704 		}
2705 	}
2706 
2707 	/*
2708 	 * Check the divisibility between rx rings and rx groups.
2709 	 */
2710 	for (i = igb->num_rx_groups; i > 0; i--) {
2711 		if ((igb->num_rx_rings % i) == 0)
2712 			break;
2713 	}
2714 	if (i != igb->num_rx_groups) {
2715 		igb_error(igb,
2716 		    "Invalid rx groups number. Downgrade the rx group "
2717 		    "number to %d.", i);
2718 		igb->num_rx_groups = i;
2719 	}
2720 
2721 	/*
2722 	 * Get the ring number per group.
2723 	 */
2724 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2725 
2726 	if (igb->num_rx_groups == 1) {
2727 		/*
2728 		 * One rx ring group, the rx ring number is num_rx_rings.
2729 		 */
2730 		igb->vmdq_mode = E1000_VMDQ_OFF;
2731 	} else if (ring_per_group == 1) {
2732 		/*
2733 		 * Multiple rx groups, each group has one rx ring.
2734 		 */
2735 		igb->vmdq_mode = E1000_VMDQ_MAC;
2736 	} else {
2737 		/*
2738 		 * Multiple groups and multiple rings.
2739 		 */
2740 		igb->vmdq_mode = E1000_VMDQ_MAC_RSS;
2741 	}
2742 
2743 	/*
2744 	 * Tunable used to force an interrupt type. The only use is
2745 	 * for testing of the lesser interrupt types.
2746 	 * 0 = don't force interrupt type
2747 	 * 1 = force interrupt type MSIX
2748 	 * 2 = force interrupt type MSI
2749 	 * 3 = force interrupt type Legacy
2750 	 */
2751 	igb->intr_force = igb_get_prop(igb, PROP_INTR_FORCE,
2752 	    IGB_INTR_NONE, IGB_INTR_LEGACY, IGB_INTR_NONE);
2753 
2754 	igb->tx_hcksum_enable = igb_get_prop(igb, PROP_TX_HCKSUM_ENABLE,
2755 	    0, 1, 1);
2756 	igb->rx_hcksum_enable = igb_get_prop(igb, PROP_RX_HCKSUM_ENABLE,
2757 	    0, 1, 1);
2758 	igb->lso_enable = igb_get_prop(igb, PROP_LSO_ENABLE,
2759 	    0, 1, 1);
2760 	igb->tx_head_wb_enable = igb_get_prop(igb, PROP_TX_HEAD_WB_ENABLE,
2761 	    0, 1, 1);
2762 
2763 	/*
2764 	 * igb LSO needs the tx h/w checksum support.
2765 	 * Here LSO will be disabled if tx h/w checksum has been disabled.
2766 	 */
2767 	if (igb->tx_hcksum_enable == B_FALSE)
2768 		igb->lso_enable = B_FALSE;
2769 
2770 	igb->tx_copy_thresh = igb_get_prop(igb, PROP_TX_COPY_THRESHOLD,
2771 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
2772 	    DEFAULT_TX_COPY_THRESHOLD);
2773 	igb->tx_recycle_thresh = igb_get_prop(igb, PROP_TX_RECYCLE_THRESHOLD,
2774 	    MIN_TX_RECYCLE_THRESHOLD, MAX_TX_RECYCLE_THRESHOLD,
2775 	    DEFAULT_TX_RECYCLE_THRESHOLD);
2776 	igb->tx_overload_thresh = igb_get_prop(igb, PROP_TX_OVERLOAD_THRESHOLD,
2777 	    MIN_TX_OVERLOAD_THRESHOLD, MAX_TX_OVERLOAD_THRESHOLD,
2778 	    DEFAULT_TX_OVERLOAD_THRESHOLD);
2779 	igb->tx_resched_thresh = igb_get_prop(igb, PROP_TX_RESCHED_THRESHOLD,
2780 	    MIN_TX_RESCHED_THRESHOLD, MAX_TX_RESCHED_THRESHOLD,
2781 	    DEFAULT_TX_RESCHED_THRESHOLD);
2782 
2783 	igb->rx_copy_thresh = igb_get_prop(igb, PROP_RX_COPY_THRESHOLD,
2784 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
2785 	    DEFAULT_RX_COPY_THRESHOLD);
2786 	igb->rx_limit_per_intr = igb_get_prop(igb, PROP_RX_LIMIT_PER_INTR,
2787 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
2788 	    DEFAULT_RX_LIMIT_PER_INTR);
2789 
2790 	igb->intr_throttling[0] = igb_get_prop(igb, PROP_INTR_THROTTLING,
2791 	    igb->capab->min_intr_throttle,
2792 	    igb->capab->max_intr_throttle,
2793 	    igb->capab->def_intr_throttle);
2794 
2795 	/*
2796 	 * Max number of multicast addresses
2797 	 */
2798 	igb->mcast_max_num =
2799 	    igb_get_prop(igb, PROP_MCAST_MAX_NUM,
2800 	    MIN_MCAST_NUM, MAX_MCAST_NUM, DEFAULT_MCAST_NUM);
2801 }
2802 
2803 /*
2804  * igb_get_prop - Get a property value out of the configuration file igb.conf
2805  *
2806  * Caller provides the name of the property, a default value, a minimum
2807  * value, and a maximum value.
2808  *
2809  * Return configured value of the property, with default, minimum and
2810  * maximum properly applied.
2811  */
2812 static int
2813 igb_get_prop(igb_t *igb,
2814     char *propname,	/* name of the property */
2815     int minval,		/* minimum acceptable value */
2816     int maxval,		/* maximim acceptable value */
2817     int defval)		/* default value */
2818 {
2819 	int value;
2820 
2821 	/*
2822 	 * Call ddi_prop_get_int() to read the conf settings
2823 	 */
2824 	value = ddi_prop_get_int(DDI_DEV_T_ANY, igb->dip,
2825 	    DDI_PROP_DONTPASS, propname, defval);
2826 
2827 	if (value > maxval)
2828 		value = maxval;
2829 
2830 	if (value < minval)
2831 		value = minval;
2832 
2833 	return (value);
2834 }
2835 
2836 /*
2837  * igb_setup_link - Using the link properties to setup the link
2838  */
2839 int
2840 igb_setup_link(igb_t *igb, boolean_t setup_hw)
2841 {
2842 	struct e1000_mac_info *mac;
2843 	struct e1000_phy_info *phy;
2844 	boolean_t invalid;
2845 
2846 	mac = &igb->hw.mac;
2847 	phy = &igb->hw.phy;
2848 	invalid = B_FALSE;
2849 
2850 	if (igb->param_adv_autoneg_cap == 1) {
2851 		mac->autoneg = B_TRUE;
2852 		phy->autoneg_advertised = 0;
2853 
2854 		/*
2855 		 * 1000hdx is not supported for autonegotiation
2856 		 */
2857 		if (igb->param_adv_1000fdx_cap == 1)
2858 			phy->autoneg_advertised |= ADVERTISE_1000_FULL;
2859 
2860 		if (igb->param_adv_100fdx_cap == 1)
2861 			phy->autoneg_advertised |= ADVERTISE_100_FULL;
2862 
2863 		if (igb->param_adv_100hdx_cap == 1)
2864 			phy->autoneg_advertised |= ADVERTISE_100_HALF;
2865 
2866 		if (igb->param_adv_10fdx_cap == 1)
2867 			phy->autoneg_advertised |= ADVERTISE_10_FULL;
2868 
2869 		if (igb->param_adv_10hdx_cap == 1)
2870 			phy->autoneg_advertised |= ADVERTISE_10_HALF;
2871 
2872 		if (phy->autoneg_advertised == 0)
2873 			invalid = B_TRUE;
2874 	} else {
2875 		mac->autoneg = B_FALSE;
2876 
2877 		/*
2878 		 * 1000fdx and 1000hdx are not supported for forced link
2879 		 */
2880 		if (igb->param_adv_100fdx_cap == 1)
2881 			mac->forced_speed_duplex = ADVERTISE_100_FULL;
2882 		else if (igb->param_adv_100hdx_cap == 1)
2883 			mac->forced_speed_duplex = ADVERTISE_100_HALF;
2884 		else if (igb->param_adv_10fdx_cap == 1)
2885 			mac->forced_speed_duplex = ADVERTISE_10_FULL;
2886 		else if (igb->param_adv_10hdx_cap == 1)
2887 			mac->forced_speed_duplex = ADVERTISE_10_HALF;
2888 		else
2889 			invalid = B_TRUE;
2890 	}
2891 
2892 	if (invalid) {
2893 		igb_notice(igb, "Invalid link settings. Setup link to "
2894 		    "autonegotiation with full link capabilities.");
2895 		mac->autoneg = B_TRUE;
2896 		phy->autoneg_advertised = ADVERTISE_1000_FULL |
2897 		    ADVERTISE_100_FULL | ADVERTISE_100_HALF |
2898 		    ADVERTISE_10_FULL | ADVERTISE_10_HALF;
2899 	}
2900 
2901 	if (setup_hw) {
2902 		if (e1000_setup_link(&igb->hw) != E1000_SUCCESS)
2903 			return (IGB_FAILURE);
2904 	}
2905 
2906 	return (IGB_SUCCESS);
2907 }
2908 
2909 
2910 /*
2911  * igb_is_link_up - Check if the link is up
2912  */
2913 static boolean_t
2914 igb_is_link_up(igb_t *igb)
2915 {
2916 	struct e1000_hw *hw = &igb->hw;
2917 	boolean_t link_up = B_FALSE;
2918 
2919 	ASSERT(mutex_owned(&igb->gen_lock));
2920 
2921 	/*
2922 	 * get_link_status is set in the interrupt handler on link-status-change
2923 	 * or rx sequence error interrupt.  get_link_status will stay
2924 	 * false until the e1000_check_for_link establishes link only
2925 	 * for copper adapters.
2926 	 */
2927 	switch (hw->phy.media_type) {
2928 	case e1000_media_type_copper:
2929 		if (hw->mac.get_link_status) {
2930 			(void) e1000_check_for_link(hw);
2931 			link_up = !hw->mac.get_link_status;
2932 		} else {
2933 			link_up = B_TRUE;
2934 		}
2935 		break;
2936 	case e1000_media_type_fiber:
2937 		(void) e1000_check_for_link(hw);
2938 		link_up = (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU);
2939 		break;
2940 	case e1000_media_type_internal_serdes:
2941 		(void) e1000_check_for_link(hw);
2942 		link_up = hw->mac.serdes_has_link;
2943 		break;
2944 	}
2945 
2946 	return (link_up);
2947 }
2948 
2949 /*
2950  * igb_link_check - Link status processing
2951  */
2952 static boolean_t
2953 igb_link_check(igb_t *igb)
2954 {
2955 	struct e1000_hw *hw = &igb->hw;
2956 	uint16_t speed = 0, duplex = 0;
2957 	boolean_t link_changed = B_FALSE;
2958 
2959 	ASSERT(mutex_owned(&igb->gen_lock));
2960 
2961 	if (igb_is_link_up(igb)) {
2962 		/*
2963 		 * The Link is up, check whether it was marked as down earlier
2964 		 */
2965 		if (igb->link_state != LINK_STATE_UP) {
2966 			(void) e1000_get_speed_and_duplex(hw, &speed, &duplex);
2967 			igb->link_speed = speed;
2968 			igb->link_duplex = duplex;
2969 			igb->link_state = LINK_STATE_UP;
2970 			igb->link_down_timeout = 0;
2971 			link_changed = B_TRUE;
2972 		}
2973 	} else {
2974 		if (igb->link_state != LINK_STATE_DOWN) {
2975 			igb->link_speed = 0;
2976 			igb->link_duplex = 0;
2977 			igb->link_state = LINK_STATE_DOWN;
2978 			link_changed = B_TRUE;
2979 		}
2980 
2981 		if (igb->igb_state & IGB_STARTED) {
2982 			if (igb->link_down_timeout < MAX_LINK_DOWN_TIMEOUT) {
2983 				igb->link_down_timeout++;
2984 			} else if (igb->link_down_timeout ==
2985 			    MAX_LINK_DOWN_TIMEOUT) {
2986 				igb_tx_clean(igb);
2987 				igb->link_down_timeout++;
2988 			}
2989 		}
2990 	}
2991 
2992 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2993 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2994 
2995 	return (link_changed);
2996 }
2997 
2998 /*
2999  * igb_local_timer - driver watchdog function
3000  *
3001  * This function will handle the transmit stall check, link status check and
3002  * other routines.
3003  */
3004 static void
3005 igb_local_timer(void *arg)
3006 {
3007 	igb_t *igb = (igb_t *)arg;
3008 	boolean_t link_changed = B_FALSE;
3009 
3010 	if (igb_stall_check(igb)) {
3011 		igb_fm_ereport(igb, DDI_FM_DEVICE_STALL);
3012 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
3013 		igb->reset_count++;
3014 		if (igb_reset(igb) == IGB_SUCCESS)
3015 			ddi_fm_service_impact(igb->dip,
3016 			    DDI_SERVICE_RESTORED);
3017 	}
3018 
3019 	mutex_enter(&igb->gen_lock);
3020 	if (!(igb->igb_state & IGB_SUSPENDED) && (igb->igb_state & IGB_STARTED))
3021 		link_changed = igb_link_check(igb);
3022 	mutex_exit(&igb->gen_lock);
3023 
3024 	if (link_changed)
3025 		mac_link_update(igb->mac_hdl, igb->link_state);
3026 
3027 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
3028 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3029 
3030 	igb_restart_watchdog_timer(igb);
3031 }
3032 
3033 /*
3034  * igb_stall_check - check for transmit stall
3035  *
3036  * This function checks if the adapter is stalled (in transmit).
3037  *
3038  * It is called each time the watchdog timeout is invoked.
3039  * If the transmit descriptor reclaim continuously fails,
3040  * the watchdog value will increment by 1. If the watchdog
3041  * value exceeds the threshold, the igb is assumed to
3042  * have stalled and need to be reset.
3043  */
3044 static boolean_t
3045 igb_stall_check(igb_t *igb)
3046 {
3047 	igb_tx_ring_t *tx_ring;
3048 	boolean_t result;
3049 	int i;
3050 
3051 	if (igb->link_state != LINK_STATE_UP)
3052 		return (B_FALSE);
3053 
3054 	/*
3055 	 * If any tx ring is stalled, we'll reset the chipset
3056 	 */
3057 	result = B_FALSE;
3058 	for (i = 0; i < igb->num_tx_rings; i++) {
3059 		tx_ring = &igb->tx_rings[i];
3060 
3061 		if (tx_ring->recycle_fail > 0)
3062 			tx_ring->stall_watchdog++;
3063 		else
3064 			tx_ring->stall_watchdog = 0;
3065 
3066 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
3067 			result = B_TRUE;
3068 			break;
3069 		}
3070 	}
3071 
3072 	if (result) {
3073 		tx_ring->stall_watchdog = 0;
3074 		tx_ring->recycle_fail = 0;
3075 	}
3076 
3077 	return (result);
3078 }
3079 
3080 
3081 /*
3082  * is_valid_mac_addr - Check if the mac address is valid
3083  */
3084 static boolean_t
3085 is_valid_mac_addr(uint8_t *mac_addr)
3086 {
3087 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
3088 	const uint8_t addr_test2[6] =
3089 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
3090 
3091 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
3092 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
3093 		return (B_FALSE);
3094 
3095 	return (B_TRUE);
3096 }
3097 
3098 static boolean_t
3099 igb_find_mac_address(igb_t *igb)
3100 {
3101 	struct e1000_hw *hw = &igb->hw;
3102 #ifdef __sparc
3103 	uchar_t *bytes;
3104 	struct ether_addr sysaddr;
3105 	uint_t nelts;
3106 	int err;
3107 	boolean_t found = B_FALSE;
3108 
3109 	/*
3110 	 * The "vendor's factory-set address" may already have
3111 	 * been extracted from the chip, but if the property
3112 	 * "local-mac-address" is set we use that instead.
3113 	 *
3114 	 * We check whether it looks like an array of 6
3115 	 * bytes (which it should, if OBP set it).  If we can't
3116 	 * make sense of it this way, we'll ignore it.
3117 	 */
3118 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3119 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
3120 	if (err == DDI_PROP_SUCCESS) {
3121 		if (nelts == ETHERADDRL) {
3122 			while (nelts--)
3123 				hw->mac.addr[nelts] = bytes[nelts];
3124 			found = B_TRUE;
3125 		}
3126 		ddi_prop_free(bytes);
3127 	}
3128 
3129 	/*
3130 	 * Look up the OBP property "local-mac-address?". If the user has set
3131 	 * 'local-mac-address? = false', use "the system address" instead.
3132 	 */
3133 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip, 0,
3134 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
3135 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
3136 			if (localetheraddr(NULL, &sysaddr) != 0) {
3137 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
3138 				found = B_TRUE;
3139 			}
3140 		}
3141 		ddi_prop_free(bytes);
3142 	}
3143 
3144 	/*
3145 	 * Finally(!), if there's a valid "mac-address" property (created
3146 	 * if we netbooted from this interface), we must use this instead
3147 	 * of any of the above to ensure that the NFS/install server doesn't
3148 	 * get confused by the address changing as Solaris takes over!
3149 	 */
3150 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3151 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
3152 	if (err == DDI_PROP_SUCCESS) {
3153 		if (nelts == ETHERADDRL) {
3154 			while (nelts--)
3155 				hw->mac.addr[nelts] = bytes[nelts];
3156 			found = B_TRUE;
3157 		}
3158 		ddi_prop_free(bytes);
3159 	}
3160 
3161 	if (found) {
3162 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
3163 		return (B_TRUE);
3164 	}
3165 #endif
3166 
3167 	/*
3168 	 * Read the device MAC address from the EEPROM
3169 	 */
3170 	if (e1000_read_mac_addr(hw) != E1000_SUCCESS)
3171 		return (B_FALSE);
3172 
3173 	return (B_TRUE);
3174 }
3175 
3176 #pragma inline(igb_arm_watchdog_timer)
3177 
3178 static void
3179 igb_arm_watchdog_timer(igb_t *igb)
3180 {
3181 	/*
3182 	 * Fire a watchdog timer
3183 	 */
3184 	igb->watchdog_tid =
3185 	    timeout(igb_local_timer,
3186 	    (void *)igb, 1 * drv_usectohz(1000000));
3187 
3188 }
3189 
3190 /*
3191  * igb_enable_watchdog_timer - Enable and start the driver watchdog timer
3192  */
3193 void
3194 igb_enable_watchdog_timer(igb_t *igb)
3195 {
3196 	mutex_enter(&igb->watchdog_lock);
3197 
3198 	if (!igb->watchdog_enable) {
3199 		igb->watchdog_enable = B_TRUE;
3200 		igb->watchdog_start = B_TRUE;
3201 		igb_arm_watchdog_timer(igb);
3202 	}
3203 
3204 	mutex_exit(&igb->watchdog_lock);
3205 
3206 }
3207 
3208 /*
3209  * igb_disable_watchdog_timer - Disable and stop the driver watchdog timer
3210  */
3211 void
3212 igb_disable_watchdog_timer(igb_t *igb)
3213 {
3214 	timeout_id_t tid;
3215 
3216 	mutex_enter(&igb->watchdog_lock);
3217 
3218 	igb->watchdog_enable = B_FALSE;
3219 	igb->watchdog_start = B_FALSE;
3220 	tid = igb->watchdog_tid;
3221 	igb->watchdog_tid = 0;
3222 
3223 	mutex_exit(&igb->watchdog_lock);
3224 
3225 	if (tid != 0)
3226 		(void) untimeout(tid);
3227 
3228 }
3229 
3230 /*
3231  * igb_start_watchdog_timer - Start the driver watchdog timer
3232  */
3233 static void
3234 igb_start_watchdog_timer(igb_t *igb)
3235 {
3236 	mutex_enter(&igb->watchdog_lock);
3237 
3238 	if (igb->watchdog_enable) {
3239 		if (!igb->watchdog_start) {
3240 			igb->watchdog_start = B_TRUE;
3241 			igb_arm_watchdog_timer(igb);
3242 		}
3243 	}
3244 
3245 	mutex_exit(&igb->watchdog_lock);
3246 }
3247 
3248 /*
3249  * igb_restart_watchdog_timer - Restart the driver watchdog timer
3250  */
3251 static void
3252 igb_restart_watchdog_timer(igb_t *igb)
3253 {
3254 	mutex_enter(&igb->watchdog_lock);
3255 
3256 	if (igb->watchdog_start)
3257 		igb_arm_watchdog_timer(igb);
3258 
3259 	mutex_exit(&igb->watchdog_lock);
3260 }
3261 
3262 /*
3263  * igb_stop_watchdog_timer - Stop the driver watchdog timer
3264  */
3265 static void
3266 igb_stop_watchdog_timer(igb_t *igb)
3267 {
3268 	timeout_id_t tid;
3269 
3270 	mutex_enter(&igb->watchdog_lock);
3271 
3272 	igb->watchdog_start = B_FALSE;
3273 	tid = igb->watchdog_tid;
3274 	igb->watchdog_tid = 0;
3275 
3276 	mutex_exit(&igb->watchdog_lock);
3277 
3278 	if (tid != 0)
3279 		(void) untimeout(tid);
3280 }
3281 
3282 /*
3283  * igb_disable_adapter_interrupts - Clear/disable all hardware interrupts
3284  */
3285 static void
3286 igb_disable_adapter_interrupts(igb_t *igb)
3287 {
3288 	struct e1000_hw *hw = &igb->hw;
3289 
3290 	/*
3291 	 * Set the IMC register to mask all the interrupts,
3292 	 * including the tx interrupts.
3293 	 */
3294 	E1000_WRITE_REG(hw, E1000_IMC, ~0);
3295 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3296 
3297 	/*
3298 	 * Additional disabling for MSI-X
3299 	 */
3300 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3301 		E1000_WRITE_REG(hw, E1000_EIMC, ~0);
3302 		E1000_WRITE_REG(hw, E1000_EIAC, 0);
3303 		E1000_WRITE_REG(hw, E1000_EIAM, 0);
3304 	}
3305 
3306 	E1000_WRITE_FLUSH(hw);
3307 }
3308 
3309 /*
3310  * igb_enable_adapter_interrupts_82576 - Enable NIC interrupts for 82576
3311  */
3312 static void
3313 igb_enable_adapter_interrupts_82576(igb_t *igb)
3314 {
3315 	struct e1000_hw *hw = &igb->hw;
3316 
3317 	/* Clear any pending interrupts */
3318 	(void) E1000_READ_REG(hw, E1000_ICR);
3319 
3320 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3321 
3322 		/* Interrupt enabling for MSI-X */
3323 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3324 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3325 		igb->ims_mask = E1000_IMS_LSC;
3326 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3327 	} else {
3328 		/* Interrupt enabling for MSI and legacy */
3329 		E1000_WRITE_REG(hw, E1000_IVAR0, E1000_IVAR_VALID);
3330 		igb->ims_mask = IMS_ENABLE_MASK | E1000_IMS_TXQE;
3331 		E1000_WRITE_REG(hw, E1000_IMS,
3332 		    (IMS_ENABLE_MASK | E1000_IMS_TXQE));
3333 	}
3334 
3335 	/* Disable auto-mask for ICR interrupt bits */
3336 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3337 
3338 	E1000_WRITE_FLUSH(hw);
3339 }
3340 
3341 /*
3342  * igb_enable_adapter_interrupts_82575 - Enable NIC interrupts for 82575
3343  */
3344 static void
3345 igb_enable_adapter_interrupts_82575(igb_t *igb)
3346 {
3347 	struct e1000_hw *hw = &igb->hw;
3348 	uint32_t reg;
3349 
3350 	/* Clear any pending interrupts */
3351 	(void) E1000_READ_REG(hw, E1000_ICR);
3352 
3353 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3354 		/* Interrupt enabling for MSI-X */
3355 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3356 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3357 		igb->ims_mask = E1000_IMS_LSC;
3358 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3359 
3360 		/* Enable MSI-X PBA support */
3361 		reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
3362 		reg |= E1000_CTRL_EXT_PBA_CLR;
3363 
3364 		/* Non-selective interrupt clear-on-read */
3365 		reg |= E1000_CTRL_EXT_IRCA;	/* Called NSICR in the EAS */
3366 
3367 		E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
3368 	} else {
3369 		/* Interrupt enabling for MSI and legacy */
3370 		igb->ims_mask = IMS_ENABLE_MASK;
3371 		E1000_WRITE_REG(hw, E1000_IMS, IMS_ENABLE_MASK);
3372 	}
3373 
3374 	E1000_WRITE_FLUSH(hw);
3375 }
3376 
3377 /*
3378  * Loopback Support
3379  */
3380 static lb_property_t lb_normal =
3381 	{ normal,	"normal",	IGB_LB_NONE		};
3382 static lb_property_t lb_external =
3383 	{ external,	"External",	IGB_LB_EXTERNAL		};
3384 static lb_property_t lb_mac =
3385 	{ internal,	"MAC",		IGB_LB_INTERNAL_MAC	};
3386 static lb_property_t lb_phy =
3387 	{ internal,	"PHY",		IGB_LB_INTERNAL_PHY	};
3388 static lb_property_t lb_serdes =
3389 	{ internal,	"SerDes",	IGB_LB_INTERNAL_SERDES	};
3390 
3391 enum ioc_reply
3392 igb_loopback_ioctl(igb_t *igb, struct iocblk *iocp, mblk_t *mp)
3393 {
3394 	lb_info_sz_t *lbsp;
3395 	lb_property_t *lbpp;
3396 	struct e1000_hw *hw;
3397 	uint32_t *lbmp;
3398 	uint32_t size;
3399 	uint32_t value;
3400 
3401 	hw = &igb->hw;
3402 
3403 	if (mp->b_cont == NULL)
3404 		return (IOC_INVAL);
3405 
3406 	switch (iocp->ioc_cmd) {
3407 	default:
3408 		return (IOC_INVAL);
3409 
3410 	case LB_GET_INFO_SIZE:
3411 		size = sizeof (lb_info_sz_t);
3412 		if (iocp->ioc_count != size)
3413 			return (IOC_INVAL);
3414 
3415 		value = sizeof (lb_normal);
3416 		value += sizeof (lb_mac);
3417 		if (hw->phy.media_type == e1000_media_type_copper)
3418 			value += sizeof (lb_phy);
3419 		else
3420 			value += sizeof (lb_serdes);
3421 		value += sizeof (lb_external);
3422 
3423 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
3424 		*lbsp = value;
3425 		break;
3426 
3427 	case LB_GET_INFO:
3428 		value = sizeof (lb_normal);
3429 		value += sizeof (lb_mac);
3430 		if (hw->phy.media_type == e1000_media_type_copper)
3431 			value += sizeof (lb_phy);
3432 		else
3433 			value += sizeof (lb_serdes);
3434 		value += sizeof (lb_external);
3435 
3436 		size = value;
3437 		if (iocp->ioc_count != size)
3438 			return (IOC_INVAL);
3439 
3440 		value = 0;
3441 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
3442 
3443 		lbpp[value++] = lb_normal;
3444 		lbpp[value++] = lb_mac;
3445 		if (hw->phy.media_type == e1000_media_type_copper)
3446 			lbpp[value++] = lb_phy;
3447 		else
3448 			lbpp[value++] = lb_serdes;
3449 		lbpp[value++] = lb_external;
3450 		break;
3451 
3452 	case LB_GET_MODE:
3453 		size = sizeof (uint32_t);
3454 		if (iocp->ioc_count != size)
3455 			return (IOC_INVAL);
3456 
3457 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3458 		*lbmp = igb->loopback_mode;
3459 		break;
3460 
3461 	case LB_SET_MODE:
3462 		size = 0;
3463 		if (iocp->ioc_count != sizeof (uint32_t))
3464 			return (IOC_INVAL);
3465 
3466 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3467 		if (!igb_set_loopback_mode(igb, *lbmp))
3468 			return (IOC_INVAL);
3469 		break;
3470 	}
3471 
3472 	iocp->ioc_count = size;
3473 	iocp->ioc_error = 0;
3474 
3475 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3476 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3477 		return (IOC_INVAL);
3478 	}
3479 
3480 	return (IOC_REPLY);
3481 }
3482 
3483 /*
3484  * igb_set_loopback_mode - Setup loopback based on the loopback mode
3485  */
3486 static boolean_t
3487 igb_set_loopback_mode(igb_t *igb, uint32_t mode)
3488 {
3489 	struct e1000_hw *hw;
3490 
3491 	if (mode == igb->loopback_mode)
3492 		return (B_TRUE);
3493 
3494 	hw = &igb->hw;
3495 
3496 	igb->loopback_mode = mode;
3497 
3498 	if (mode == IGB_LB_NONE) {
3499 		/* Reset the chip */
3500 		hw->phy.autoneg_wait_to_complete = B_TRUE;
3501 		(void) igb_reset(igb);
3502 		hw->phy.autoneg_wait_to_complete = B_FALSE;
3503 		return (B_TRUE);
3504 	}
3505 
3506 	mutex_enter(&igb->gen_lock);
3507 
3508 	switch (mode) {
3509 	default:
3510 		mutex_exit(&igb->gen_lock);
3511 		return (B_FALSE);
3512 
3513 	case IGB_LB_EXTERNAL:
3514 		igb_set_external_loopback(igb);
3515 		break;
3516 
3517 	case IGB_LB_INTERNAL_MAC:
3518 		igb_set_internal_mac_loopback(igb);
3519 		break;
3520 
3521 	case IGB_LB_INTERNAL_PHY:
3522 		igb_set_internal_phy_loopback(igb);
3523 		break;
3524 
3525 	case IGB_LB_INTERNAL_SERDES:
3526 		igb_set_internal_serdes_loopback(igb);
3527 		break;
3528 	}
3529 
3530 	mutex_exit(&igb->gen_lock);
3531 
3532 	return (B_TRUE);
3533 }
3534 
3535 /*
3536  * igb_set_external_loopback - Set the external loopback mode
3537  */
3538 static void
3539 igb_set_external_loopback(igb_t *igb)
3540 {
3541 	struct e1000_hw *hw;
3542 
3543 	hw = &igb->hw;
3544 
3545 	/* Set phy to known state */
3546 	(void) e1000_phy_hw_reset(hw);
3547 
3548 	(void) e1000_write_phy_reg(hw, 0x0, 0x0140);
3549 	(void) e1000_write_phy_reg(hw, 0x9, 0x1b00);
3550 	(void) e1000_write_phy_reg(hw, 0x12, 0x1610);
3551 	(void) e1000_write_phy_reg(hw, 0x1f37, 0x3f1c);
3552 }
3553 
3554 /*
3555  * igb_set_internal_mac_loopback - Set the internal MAC loopback mode
3556  */
3557 static void
3558 igb_set_internal_mac_loopback(igb_t *igb)
3559 {
3560 	struct e1000_hw *hw;
3561 	uint32_t ctrl;
3562 	uint32_t rctl;
3563 	uint32_t ctrl_ext;
3564 	uint16_t phy_ctrl;
3565 	uint16_t phy_status;
3566 
3567 	hw = &igb->hw;
3568 
3569 	(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
3570 	phy_ctrl &= ~MII_CR_AUTO_NEG_EN;
3571 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3572 
3573 	(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
3574 
3575 	/* Set link mode to PHY (00b) in the Extended Control register */
3576 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3577 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3578 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3579 
3580 	/* Set the Device Control register */
3581 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3582 	if (!(phy_status & MII_SR_LINK_STATUS))
3583 		ctrl |= E1000_CTRL_ILOS; /* Set ILOS when the link is down */
3584 	ctrl &= ~E1000_CTRL_SPD_SEL;	/* Clear the speed sel bits */
3585 	ctrl |= (E1000_CTRL_SLU |	/* Force link up */
3586 	    E1000_CTRL_FRCSPD |		/* Force speed */
3587 	    E1000_CTRL_FRCDPX |		/* Force duplex */
3588 	    E1000_CTRL_SPD_1000 |	/* Force speed to 1000 */
3589 	    E1000_CTRL_FD);		/* Force full duplex */
3590 
3591 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3592 
3593 	/* Set the Receive Control register */
3594 	rctl = E1000_READ_REG(hw, E1000_RCTL);
3595 	rctl &= ~E1000_RCTL_LBM_TCVR;
3596 	rctl |= E1000_RCTL_LBM_MAC;
3597 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
3598 }
3599 
3600 /*
3601  * igb_set_internal_phy_loopback - Set the internal PHY loopback mode
3602  */
3603 static void
3604 igb_set_internal_phy_loopback(igb_t *igb)
3605 {
3606 	struct e1000_hw *hw;
3607 	uint32_t ctrl_ext;
3608 	uint16_t phy_ctrl;
3609 	uint16_t phy_pconf;
3610 
3611 	hw = &igb->hw;
3612 
3613 	/* Set link mode to PHY (00b) in the Extended Control register */
3614 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3615 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3616 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3617 
3618 	/*
3619 	 * Set PHY control register (0x4140):
3620 	 *    Set full duplex mode
3621 	 *    Set loopback bit
3622 	 *    Clear auto-neg enable bit
3623 	 *    Set PHY speed
3624 	 */
3625 	phy_ctrl = MII_CR_FULL_DUPLEX | MII_CR_SPEED_1000 | MII_CR_LOOPBACK;
3626 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3627 
3628 	/* Set the link disable bit in the Port Configuration register */
3629 	(void) e1000_read_phy_reg(hw, 0x10, &phy_pconf);
3630 	phy_pconf |= (uint16_t)1 << 14;
3631 	(void) e1000_write_phy_reg(hw, 0x10, phy_pconf);
3632 }
3633 
3634 /*
3635  * igb_set_internal_serdes_loopback - Set the internal SerDes loopback mode
3636  */
3637 static void
3638 igb_set_internal_serdes_loopback(igb_t *igb)
3639 {
3640 	struct e1000_hw *hw;
3641 	uint32_t ctrl_ext;
3642 	uint32_t ctrl;
3643 	uint32_t pcs_lctl;
3644 	uint32_t connsw;
3645 
3646 	hw = &igb->hw;
3647 
3648 	/* Set link mode to SerDes (11b) in the Extended Control register */
3649 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3650 	ctrl_ext |= E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
3651 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3652 
3653 	/* Configure the SerDes to loopback */
3654 	E1000_WRITE_REG(hw, E1000_SCTL, 0x410);
3655 
3656 	/* Set Device Control register */
3657 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3658 	ctrl |= (E1000_CTRL_FD |	/* Force full duplex */
3659 	    E1000_CTRL_SLU);		/* Force link up */
3660 	ctrl &= ~(E1000_CTRL_RFCE |	/* Disable receive flow control */
3661 	    E1000_CTRL_TFCE |		/* Disable transmit flow control */
3662 	    E1000_CTRL_LRST);		/* Clear link reset */
3663 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3664 
3665 	/* Set PCS Link Control register */
3666 	pcs_lctl = E1000_READ_REG(hw, E1000_PCS_LCTL);
3667 	pcs_lctl |= (E1000_PCS_LCTL_FORCE_LINK |
3668 	    E1000_PCS_LCTL_FSD |
3669 	    E1000_PCS_LCTL_FDV_FULL |
3670 	    E1000_PCS_LCTL_FLV_LINK_UP);
3671 	pcs_lctl &= ~E1000_PCS_LCTL_AN_ENABLE;
3672 	E1000_WRITE_REG(hw, E1000_PCS_LCTL, pcs_lctl);
3673 
3674 	/* Set the Copper/Fiber Switch Control - CONNSW register */
3675 	connsw = E1000_READ_REG(hw, E1000_CONNSW);
3676 	connsw &= ~E1000_CONNSW_ENRGSRC;
3677 	E1000_WRITE_REG(hw, E1000_CONNSW, connsw);
3678 }
3679 
3680 #pragma inline(igb_intr_rx_work)
3681 /*
3682  * igb_intr_rx_work - rx processing of ISR
3683  */
3684 static void
3685 igb_intr_rx_work(igb_rx_ring_t *rx_ring)
3686 {
3687 	mblk_t *mp;
3688 
3689 	mutex_enter(&rx_ring->rx_lock);
3690 	mp = igb_rx(rx_ring, IGB_NO_POLL);
3691 	mutex_exit(&rx_ring->rx_lock);
3692 
3693 	if (mp != NULL)
3694 		mac_rx_ring(rx_ring->igb->mac_hdl, rx_ring->ring_handle, mp,
3695 		    rx_ring->ring_gen_num);
3696 }
3697 
3698 #pragma inline(igb_intr_tx_work)
3699 /*
3700  * igb_intr_tx_work - tx processing of ISR
3701  */
3702 static void
3703 igb_intr_tx_work(igb_tx_ring_t *tx_ring)
3704 {
3705 	/* Recycle the tx descriptors */
3706 	tx_ring->tx_recycle(tx_ring);
3707 
3708 	/* Schedule the re-transmit */
3709 	if (tx_ring->reschedule &&
3710 	    (tx_ring->tbd_free >= tx_ring->resched_thresh)) {
3711 		tx_ring->reschedule = B_FALSE;
3712 		mac_tx_ring_update(tx_ring->igb->mac_hdl, tx_ring->ring_handle);
3713 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3714 	}
3715 }
3716 
3717 #pragma inline(igb_intr_link_work)
3718 /*
3719  * igb_intr_link_work - link-status-change processing of ISR
3720  */
3721 static void
3722 igb_intr_link_work(igb_t *igb)
3723 {
3724 	boolean_t link_changed;
3725 
3726 	igb_stop_watchdog_timer(igb);
3727 
3728 	mutex_enter(&igb->gen_lock);
3729 
3730 	/*
3731 	 * Because we got a link-status-change interrupt, force
3732 	 * e1000_check_for_link() to look at phy
3733 	 */
3734 	igb->hw.mac.get_link_status = B_TRUE;
3735 
3736 	/* igb_link_check takes care of link status change */
3737 	link_changed = igb_link_check(igb);
3738 
3739 	/* Get new phy state */
3740 	igb_get_phy_state(igb);
3741 
3742 	mutex_exit(&igb->gen_lock);
3743 
3744 	if (link_changed)
3745 		mac_link_update(igb->mac_hdl, igb->link_state);
3746 
3747 	igb_start_watchdog_timer(igb);
3748 }
3749 
3750 /*
3751  * igb_intr_legacy - Interrupt handler for legacy interrupts
3752  */
3753 static uint_t
3754 igb_intr_legacy(void *arg1, void *arg2)
3755 {
3756 	igb_t *igb = (igb_t *)arg1;
3757 	igb_tx_ring_t *tx_ring;
3758 	uint32_t icr;
3759 	mblk_t *mp;
3760 	boolean_t tx_reschedule;
3761 	boolean_t link_changed;
3762 	uint_t result;
3763 
3764 	_NOTE(ARGUNUSED(arg2));
3765 
3766 	mutex_enter(&igb->gen_lock);
3767 
3768 	if (igb->igb_state & IGB_SUSPENDED) {
3769 		mutex_exit(&igb->gen_lock);
3770 		return (DDI_INTR_UNCLAIMED);
3771 	}
3772 
3773 	mp = NULL;
3774 	tx_reschedule = B_FALSE;
3775 	link_changed = B_FALSE;
3776 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3777 
3778 	if (icr & E1000_ICR_INT_ASSERTED) {
3779 		/*
3780 		 * E1000_ICR_INT_ASSERTED bit was set:
3781 		 * Read(Clear) the ICR, claim this interrupt,
3782 		 * look for work to do.
3783 		 */
3784 		ASSERT(igb->num_rx_rings == 1);
3785 		ASSERT(igb->num_tx_rings == 1);
3786 
3787 		/* Make sure all interrupt causes cleared */
3788 		(void) E1000_READ_REG(&igb->hw, E1000_EICR);
3789 
3790 		if (icr & E1000_ICR_RXT0) {
3791 			mp = igb_rx(&igb->rx_rings[0], IGB_NO_POLL);
3792 		}
3793 
3794 		if (icr & E1000_ICR_TXDW) {
3795 			tx_ring = &igb->tx_rings[0];
3796 
3797 			/* Recycle the tx descriptors */
3798 			tx_ring->tx_recycle(tx_ring);
3799 
3800 			/* Schedule the re-transmit */
3801 			tx_reschedule = (tx_ring->reschedule &&
3802 			    (tx_ring->tbd_free >= tx_ring->resched_thresh));
3803 		}
3804 
3805 		if (icr & E1000_ICR_LSC) {
3806 			/*
3807 			 * Because we got a link-status-change interrupt, force
3808 			 * e1000_check_for_link() to look at phy
3809 			 */
3810 			igb->hw.mac.get_link_status = B_TRUE;
3811 
3812 			/* igb_link_check takes care of link status change */
3813 			link_changed = igb_link_check(igb);
3814 
3815 			/* Get new phy state */
3816 			igb_get_phy_state(igb);
3817 		}
3818 
3819 		result = DDI_INTR_CLAIMED;
3820 	} else {
3821 		/*
3822 		 * E1000_ICR_INT_ASSERTED bit was not set:
3823 		 * Don't claim this interrupt.
3824 		 */
3825 		result = DDI_INTR_UNCLAIMED;
3826 	}
3827 
3828 	mutex_exit(&igb->gen_lock);
3829 
3830 	/*
3831 	 * Do the following work outside of the gen_lock
3832 	 */
3833 	if (mp != NULL)
3834 		mac_rx(igb->mac_hdl, NULL, mp);
3835 
3836 	if (tx_reschedule)  {
3837 		tx_ring->reschedule = B_FALSE;
3838 		mac_tx_ring_update(igb->mac_hdl, tx_ring->ring_handle);
3839 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3840 	}
3841 
3842 	if (link_changed)
3843 		mac_link_update(igb->mac_hdl, igb->link_state);
3844 
3845 	return (result);
3846 }
3847 
3848 /*
3849  * igb_intr_msi - Interrupt handler for MSI
3850  */
3851 static uint_t
3852 igb_intr_msi(void *arg1, void *arg2)
3853 {
3854 	igb_t *igb = (igb_t *)arg1;
3855 	uint32_t icr;
3856 
3857 	_NOTE(ARGUNUSED(arg2));
3858 
3859 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3860 
3861 	/* Make sure all interrupt causes cleared */
3862 	(void) E1000_READ_REG(&igb->hw, E1000_EICR);
3863 
3864 	/*
3865 	 * For MSI interrupt, we have only one vector,
3866 	 * so we have only one rx ring and one tx ring enabled.
3867 	 */
3868 	ASSERT(igb->num_rx_rings == 1);
3869 	ASSERT(igb->num_tx_rings == 1);
3870 
3871 	if (icr & E1000_ICR_RXT0) {
3872 		igb_intr_rx_work(&igb->rx_rings[0]);
3873 	}
3874 
3875 	if (icr & E1000_ICR_TXDW) {
3876 		igb_intr_tx_work(&igb->tx_rings[0]);
3877 	}
3878 
3879 	if (icr & E1000_ICR_LSC) {
3880 		igb_intr_link_work(igb);
3881 	}
3882 
3883 	return (DDI_INTR_CLAIMED);
3884 }
3885 
3886 /*
3887  * igb_intr_rx - Interrupt handler for rx
3888  */
3889 static uint_t
3890 igb_intr_rx(void *arg1, void *arg2)
3891 {
3892 	igb_rx_ring_t *rx_ring = (igb_rx_ring_t *)arg1;
3893 
3894 	_NOTE(ARGUNUSED(arg2));
3895 
3896 	/*
3897 	 * Only used via MSI-X vector so don't check cause bits
3898 	 * and only clean the given ring.
3899 	 */
3900 	igb_intr_rx_work(rx_ring);
3901 
3902 	return (DDI_INTR_CLAIMED);
3903 }
3904 
3905 /*
3906  * igb_intr_tx - Interrupt handler for tx
3907  */
3908 static uint_t
3909 igb_intr_tx(void *arg1, void *arg2)
3910 {
3911 	igb_tx_ring_t *tx_ring = (igb_tx_ring_t *)arg1;
3912 
3913 	_NOTE(ARGUNUSED(arg2));
3914 
3915 	/*
3916 	 * Only used via MSI-X vector so don't check cause bits
3917 	 * and only clean the given ring.
3918 	 */
3919 	igb_intr_tx_work(tx_ring);
3920 
3921 	return (DDI_INTR_CLAIMED);
3922 }
3923 
3924 /*
3925  * igb_intr_tx_other - Interrupt handler for both tx and other
3926  *
3927  */
3928 static uint_t
3929 igb_intr_tx_other(void *arg1, void *arg2)
3930 {
3931 	igb_t *igb = (igb_t *)arg1;
3932 	uint32_t icr;
3933 
3934 	_NOTE(ARGUNUSED(arg2));
3935 
3936 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3937 
3938 	/*
3939 	 * Look for tx reclaiming work first. Remember, in the
3940 	 * case of only interrupt sharing, only one tx ring is
3941 	 * used
3942 	 */
3943 	igb_intr_tx_work(&igb->tx_rings[0]);
3944 
3945 	/*
3946 	 * Check for "other" causes.
3947 	 */
3948 	if (icr & E1000_ICR_LSC) {
3949 		igb_intr_link_work(igb);
3950 	}
3951 
3952 	/*
3953 	 * The DOUTSYNC bit indicates a tx packet dropped because
3954 	 * DMA engine gets "out of sync". There isn't a real fix
3955 	 * for this. The Intel recommendation is to count the number
3956 	 * of occurrences so user can detect when it is happening.
3957 	 * The issue is non-fatal and there's no recovery action
3958 	 * available.
3959 	 */
3960 	if (icr & E1000_ICR_DOUTSYNC) {
3961 		IGB_STAT(igb->dout_sync);
3962 	}
3963 
3964 	return (DDI_INTR_CLAIMED);
3965 }
3966 
3967 /*
3968  * igb_alloc_intrs - Allocate interrupts for the driver
3969  *
3970  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
3971  * if not successful, try Legacy.
3972  * igb->intr_force can be used to force sequence to start with
3973  * any of the 3 types.
3974  * If MSI-X is not used, number of tx/rx rings is forced to 1.
3975  */
3976 static int
3977 igb_alloc_intrs(igb_t *igb)
3978 {
3979 	dev_info_t *devinfo;
3980 	int intr_types;
3981 	int rc;
3982 
3983 	devinfo = igb->dip;
3984 
3985 	/* Get supported interrupt types */
3986 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
3987 
3988 	if (rc != DDI_SUCCESS) {
3989 		igb_log(igb,
3990 		    "Get supported interrupt types failed: %d", rc);
3991 		return (IGB_FAILURE);
3992 	}
3993 	IGB_DEBUGLOG_1(igb, "Supported interrupt types: %x", intr_types);
3994 
3995 	igb->intr_type = 0;
3996 
3997 	/* Install MSI-X interrupts */
3998 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
3999 	    (igb->intr_force <= IGB_INTR_MSIX)) {
4000 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSIX);
4001 
4002 		if (rc == IGB_SUCCESS)
4003 			return (IGB_SUCCESS);
4004 
4005 		igb_log(igb,
4006 		    "Allocate MSI-X failed, trying MSI interrupts...");
4007 	}
4008 
4009 	/* MSI-X not used, force rings to 1 */
4010 	igb->num_rx_rings = 1;
4011 	igb->num_tx_rings = 1;
4012 	igb_log(igb,
4013 	    "MSI-X not used, force rx and tx queue number to 1");
4014 
4015 	/* Install MSI interrupts */
4016 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
4017 	    (igb->intr_force <= IGB_INTR_MSI)) {
4018 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSI);
4019 
4020 		if (rc == IGB_SUCCESS)
4021 			return (IGB_SUCCESS);
4022 
4023 		igb_log(igb,
4024 		    "Allocate MSI failed, trying Legacy interrupts...");
4025 	}
4026 
4027 	/* Install legacy interrupts */
4028 	if (intr_types & DDI_INTR_TYPE_FIXED) {
4029 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_FIXED);
4030 
4031 		if (rc == IGB_SUCCESS)
4032 			return (IGB_SUCCESS);
4033 
4034 		igb_log(igb,
4035 		    "Allocate Legacy interrupts failed");
4036 	}
4037 
4038 	/* If none of the 3 types succeeded, return failure */
4039 	return (IGB_FAILURE);
4040 }
4041 
4042 /*
4043  * igb_alloc_intr_handles - Allocate interrupt handles.
4044  *
4045  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
4046  * if fewer than 2 handles are available, return failure.
4047  * Upon success, this sets the number of Rx rings to a number that
4048  * matches the handles available for Rx interrupts.
4049  */
4050 static int
4051 igb_alloc_intr_handles(igb_t *igb, int intr_type)
4052 {
4053 	dev_info_t *devinfo;
4054 	int orig, request, count, avail, actual;
4055 	int diff, minimum;
4056 	int rc;
4057 
4058 	devinfo = igb->dip;
4059 
4060 	switch (intr_type) {
4061 	case DDI_INTR_TYPE_FIXED:
4062 		request = 1;	/* Request 1 legacy interrupt handle */
4063 		minimum = 1;
4064 		IGB_DEBUGLOG_0(igb, "interrupt type: legacy");
4065 		break;
4066 
4067 	case DDI_INTR_TYPE_MSI:
4068 		request = 1;	/* Request 1 MSI interrupt handle */
4069 		minimum = 1;
4070 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI");
4071 		break;
4072 
4073 	case DDI_INTR_TYPE_MSIX:
4074 		/*
4075 		 * Number of vectors for the adapter is
4076 		 * # rx rings + # tx rings
4077 		 * One of tx vectors is for tx & other
4078 		 */
4079 		request = igb->num_rx_rings + igb->num_tx_rings;
4080 		orig = request;
4081 		minimum = 2;
4082 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI-X");
4083 		break;
4084 
4085 	default:
4086 		igb_log(igb,
4087 		    "invalid call to igb_alloc_intr_handles(): %d\n",
4088 		    intr_type);
4089 		return (IGB_FAILURE);
4090 	}
4091 	IGB_DEBUGLOG_2(igb, "interrupt handles requested: %d  minimum: %d",
4092 	    request, minimum);
4093 
4094 	/*
4095 	 * Get number of supported interrupts
4096 	 */
4097 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
4098 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
4099 		igb_log(igb,
4100 		    "Get supported interrupt number failed. "
4101 		    "Return: %d, count: %d", rc, count);
4102 		return (IGB_FAILURE);
4103 	}
4104 	IGB_DEBUGLOG_1(igb, "interrupts supported: %d", count);
4105 
4106 	/*
4107 	 * Get number of available interrupts
4108 	 */
4109 	rc = ddi_intr_get_navail(devinfo, intr_type, &avail);
4110 	if ((rc != DDI_SUCCESS) || (avail < minimum)) {
4111 		igb_log(igb,
4112 		    "Get available interrupt number failed. "
4113 		    "Return: %d, available: %d", rc, avail);
4114 		return (IGB_FAILURE);
4115 	}
4116 	IGB_DEBUGLOG_1(igb, "interrupts available: %d", avail);
4117 
4118 	if (avail < request) {
4119 		igb_log(igb, "Request %d handles, %d available",
4120 		    request, avail);
4121 		request = avail;
4122 	}
4123 
4124 	actual = 0;
4125 	igb->intr_cnt = 0;
4126 
4127 	/*
4128 	 * Allocate an array of interrupt handles
4129 	 */
4130 	igb->intr_size = request * sizeof (ddi_intr_handle_t);
4131 	igb->htable = kmem_alloc(igb->intr_size, KM_SLEEP);
4132 
4133 	rc = ddi_intr_alloc(devinfo, igb->htable, intr_type, 0,
4134 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
4135 	if (rc != DDI_SUCCESS) {
4136 		igb_log(igb, "Allocate interrupts failed. "
4137 		    "return: %d, request: %d, actual: %d",
4138 		    rc, request, actual);
4139 		goto alloc_handle_fail;
4140 	}
4141 	IGB_DEBUGLOG_1(igb, "interrupts actually allocated: %d", actual);
4142 
4143 	igb->intr_cnt = actual;
4144 
4145 	if (actual < minimum) {
4146 		igb_log(igb, "Insufficient interrupt handles allocated: %d",
4147 		    actual);
4148 		goto alloc_handle_fail;
4149 	}
4150 
4151 	/*
4152 	 * For MSI-X, actual might force us to reduce number of tx & rx rings
4153 	 */
4154 	if ((intr_type == DDI_INTR_TYPE_MSIX) && (orig > actual)) {
4155 		diff = orig - actual;
4156 		if (diff < igb->num_tx_rings) {
4157 			igb_log(igb,
4158 			    "MSI-X vectors force Tx queue number to %d",
4159 			    igb->num_tx_rings - diff);
4160 			igb->num_tx_rings -= diff;
4161 		} else {
4162 			igb_log(igb,
4163 			    "MSI-X vectors force Tx queue number to 1");
4164 			igb->num_tx_rings = 1;
4165 
4166 			igb_log(igb,
4167 			    "MSI-X vectors force Rx queue number to %d",
4168 			    actual - 1);
4169 			igb->num_rx_rings = actual - 1;
4170 		}
4171 	}
4172 
4173 	/*
4174 	 * Get priority for first vector, assume remaining are all the same
4175 	 */
4176 	rc = ddi_intr_get_pri(igb->htable[0], &igb->intr_pri);
4177 	if (rc != DDI_SUCCESS) {
4178 		igb_log(igb,
4179 		    "Get interrupt priority failed: %d", rc);
4180 		goto alloc_handle_fail;
4181 	}
4182 
4183 	rc = ddi_intr_get_cap(igb->htable[0], &igb->intr_cap);
4184 	if (rc != DDI_SUCCESS) {
4185 		igb_log(igb,
4186 		    "Get interrupt cap failed: %d", rc);
4187 		goto alloc_handle_fail;
4188 	}
4189 
4190 	igb->intr_type = intr_type;
4191 
4192 	return (IGB_SUCCESS);
4193 
4194 alloc_handle_fail:
4195 	igb_rem_intrs(igb);
4196 
4197 	return (IGB_FAILURE);
4198 }
4199 
4200 /*
4201  * igb_add_intr_handlers - Add interrupt handlers based on the interrupt type
4202  *
4203  * Before adding the interrupt handlers, the interrupt vectors have
4204  * been allocated, and the rx/tx rings have also been allocated.
4205  */
4206 static int
4207 igb_add_intr_handlers(igb_t *igb)
4208 {
4209 	igb_rx_ring_t *rx_ring;
4210 	igb_tx_ring_t *tx_ring;
4211 	int vector;
4212 	int rc;
4213 	int i;
4214 
4215 	vector = 0;
4216 
4217 	switch (igb->intr_type) {
4218 	case DDI_INTR_TYPE_MSIX:
4219 		/* Add interrupt handler for tx + other */
4220 		tx_ring = &igb->tx_rings[0];
4221 		rc = ddi_intr_add_handler(igb->htable[vector],
4222 		    (ddi_intr_handler_t *)igb_intr_tx_other,
4223 		    (void *)igb, NULL);
4224 
4225 		if (rc != DDI_SUCCESS) {
4226 			igb_log(igb,
4227 			    "Add tx/other interrupt handler failed: %d", rc);
4228 			return (IGB_FAILURE);
4229 		}
4230 		tx_ring->intr_vector = vector;
4231 		vector++;
4232 
4233 		/* Add interrupt handler for each rx ring */
4234 		for (i = 0; i < igb->num_rx_rings; i++) {
4235 			rx_ring = &igb->rx_rings[i];
4236 
4237 			rc = ddi_intr_add_handler(igb->htable[vector],
4238 			    (ddi_intr_handler_t *)igb_intr_rx,
4239 			    (void *)rx_ring, NULL);
4240 
4241 			if (rc != DDI_SUCCESS) {
4242 				igb_log(igb,
4243 				    "Add rx interrupt handler failed. "
4244 				    "return: %d, rx ring: %d", rc, i);
4245 				for (vector--; vector >= 0; vector--) {
4246 					(void) ddi_intr_remove_handler(
4247 					    igb->htable[vector]);
4248 				}
4249 				return (IGB_FAILURE);
4250 			}
4251 
4252 			rx_ring->intr_vector = vector;
4253 
4254 			vector++;
4255 		}
4256 
4257 		/* Add interrupt handler for each tx ring from 2nd ring */
4258 		for (i = 1; i < igb->num_tx_rings; i++) {
4259 			tx_ring = &igb->tx_rings[i];
4260 
4261 			rc = ddi_intr_add_handler(igb->htable[vector],
4262 			    (ddi_intr_handler_t *)igb_intr_tx,
4263 			    (void *)tx_ring, NULL);
4264 
4265 			if (rc != DDI_SUCCESS) {
4266 				igb_log(igb,
4267 				    "Add tx interrupt handler failed. "
4268 				    "return: %d, tx ring: %d", rc, i);
4269 				for (vector--; vector >= 0; vector--) {
4270 					(void) ddi_intr_remove_handler(
4271 					    igb->htable[vector]);
4272 				}
4273 				return (IGB_FAILURE);
4274 			}
4275 
4276 			tx_ring->intr_vector = vector;
4277 
4278 			vector++;
4279 		}
4280 
4281 		break;
4282 
4283 	case DDI_INTR_TYPE_MSI:
4284 		/* Add interrupt handlers for the only vector */
4285 		rc = ddi_intr_add_handler(igb->htable[vector],
4286 		    (ddi_intr_handler_t *)igb_intr_msi,
4287 		    (void *)igb, NULL);
4288 
4289 		if (rc != DDI_SUCCESS) {
4290 			igb_log(igb,
4291 			    "Add MSI interrupt handler failed: %d", rc);
4292 			return (IGB_FAILURE);
4293 		}
4294 
4295 		rx_ring = &igb->rx_rings[0];
4296 		rx_ring->intr_vector = vector;
4297 
4298 		vector++;
4299 		break;
4300 
4301 	case DDI_INTR_TYPE_FIXED:
4302 		/* Add interrupt handlers for the only vector */
4303 		rc = ddi_intr_add_handler(igb->htable[vector],
4304 		    (ddi_intr_handler_t *)igb_intr_legacy,
4305 		    (void *)igb, NULL);
4306 
4307 		if (rc != DDI_SUCCESS) {
4308 			igb_log(igb,
4309 			    "Add legacy interrupt handler failed: %d", rc);
4310 			return (IGB_FAILURE);
4311 		}
4312 
4313 		rx_ring = &igb->rx_rings[0];
4314 		rx_ring->intr_vector = vector;
4315 
4316 		vector++;
4317 		break;
4318 
4319 	default:
4320 		return (IGB_FAILURE);
4321 	}
4322 
4323 	ASSERT(vector == igb->intr_cnt);
4324 
4325 	return (IGB_SUCCESS);
4326 }
4327 
4328 /*
4329  * igb_setup_msix_82575 - setup 82575 adapter to use MSI-X interrupts
4330  *
4331  * For each vector enabled on the adapter, Set the MSIXBM register accordingly
4332  */
4333 static void
4334 igb_setup_msix_82575(igb_t *igb)
4335 {
4336 	uint32_t eims = 0;
4337 	int i, vector;
4338 	struct e1000_hw *hw = &igb->hw;
4339 
4340 	/*
4341 	 * Set vector for tx ring 0 and other causes.
4342 	 * NOTE assumption that it is vector 0.
4343 	 */
4344 	vector = 0;
4345 
4346 	igb->eims_mask = E1000_EICR_TX_QUEUE0 | E1000_EICR_OTHER;
4347 	E1000_WRITE_REG(hw, E1000_MSIXBM(vector), igb->eims_mask);
4348 	vector++;
4349 
4350 	for (i = 0; i < igb->num_rx_rings; i++) {
4351 		/*
4352 		 * Set vector for each rx ring
4353 		 */
4354 		eims = (E1000_EICR_RX_QUEUE0 << i);
4355 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4356 
4357 		/*
4358 		 * Accumulate bits to enable in
4359 		 * igb_enable_adapter_interrupts_82575()
4360 		 */
4361 		igb->eims_mask |= eims;
4362 
4363 		vector++;
4364 	}
4365 
4366 	for (i = 1; i < igb->num_tx_rings; i++) {
4367 		/*
4368 		 * Set vector for each tx ring from 2nd tx ring
4369 		 */
4370 		eims = (E1000_EICR_TX_QUEUE0 << i);
4371 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4372 
4373 		/*
4374 		 * Accumulate bits to enable in
4375 		 * igb_enable_adapter_interrupts_82575()
4376 		 */
4377 		igb->eims_mask |= eims;
4378 
4379 		vector++;
4380 	}
4381 
4382 	ASSERT(vector == igb->intr_cnt);
4383 
4384 	/*
4385 	 * Disable IAM for ICR interrupt bits
4386 	 */
4387 	E1000_WRITE_REG(hw, E1000_IAM, 0);
4388 	E1000_WRITE_FLUSH(hw);
4389 }
4390 
4391 /*
4392  * igb_setup_msix_82576 - setup 82576 adapter to use MSI-X interrupts
4393  *
4394  * 82576 uses a table based method for assigning vectors.  Each queue has a
4395  * single entry in the table to which we write a vector number along with a
4396  * "valid" bit.  The entry is a single byte in a 4-byte register.  Vectors
4397  * take a different position in the 4-byte register depending on whether
4398  * they are numbered above or below 8.
4399  */
4400 static void
4401 igb_setup_msix_82576(igb_t *igb)
4402 {
4403 	struct e1000_hw *hw = &igb->hw;
4404 	uint32_t ivar, index, vector;
4405 	int i;
4406 
4407 	/* must enable msi-x capability before IVAR settings */
4408 	E1000_WRITE_REG(hw, E1000_GPIE,
4409 	    (E1000_GPIE_MSIX_MODE | E1000_GPIE_PBA | E1000_GPIE_NSICR));
4410 
4411 	/*
4412 	 * Set vector for tx ring 0 and other causes.
4413 	 * NOTE assumption that it is vector 0.
4414 	 * This is also interdependent with installation of interrupt service
4415 	 * routines in igb_add_intr_handlers().
4416 	 */
4417 
4418 	/* assign "other" causes to vector 0 */
4419 	vector = 0;
4420 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4421 	E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4422 
4423 	/* assign tx ring 0 to vector 0 */
4424 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4425 	E1000_WRITE_REG(hw, E1000_IVAR0, ivar);
4426 
4427 	/* prepare to enable tx & other interrupt causes */
4428 	igb->eims_mask = (1 << vector);
4429 
4430 	vector ++;
4431 	for (i = 0; i < igb->num_rx_rings; i++) {
4432 		/*
4433 		 * Set vector for each rx ring
4434 		 */
4435 		index = (i & 0x7);
4436 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4437 
4438 		if (i < 8) {
4439 			/* vector goes into low byte of register */
4440 			ivar = ivar & 0xFFFFFF00;
4441 			ivar |= (vector | E1000_IVAR_VALID);
4442 		} else {
4443 			/* vector goes into third byte of register */
4444 			ivar = ivar & 0xFF00FFFF;
4445 			ivar |= ((vector | E1000_IVAR_VALID) << 16);
4446 		}
4447 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4448 
4449 		/* Accumulate interrupt-cause bits to enable */
4450 		igb->eims_mask |= (1 << vector);
4451 
4452 		vector ++;
4453 	}
4454 
4455 	for (i = 1; i < igb->num_tx_rings; i++) {
4456 		/*
4457 		 * Set vector for each tx ring from 2nd tx ring.
4458 		 * Note assumption that tx vectors numericall follow rx vectors.
4459 		 */
4460 		index = (i & 0x7);
4461 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4462 
4463 		if (i < 8) {
4464 			/* vector goes into second byte of register */
4465 			ivar = ivar & 0xFFFF00FF;
4466 			ivar |= ((vector | E1000_IVAR_VALID) << 8);
4467 		} else {
4468 			/* vector goes into fourth byte of register */
4469 			ivar = ivar & 0x00FFFFFF;
4470 			ivar |= (vector | E1000_IVAR_VALID) << 24;
4471 		}
4472 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4473 
4474 		/* Accumulate interrupt-cause bits to enable */
4475 		igb->eims_mask |= (1 << vector);
4476 
4477 		vector ++;
4478 	}
4479 
4480 	ASSERT(vector == igb->intr_cnt);
4481 }
4482 
4483 /*
4484  * igb_rem_intr_handlers - remove the interrupt handlers
4485  */
4486 static void
4487 igb_rem_intr_handlers(igb_t *igb)
4488 {
4489 	int i;
4490 	int rc;
4491 
4492 	for (i = 0; i < igb->intr_cnt; i++) {
4493 		rc = ddi_intr_remove_handler(igb->htable[i]);
4494 		if (rc != DDI_SUCCESS) {
4495 			IGB_DEBUGLOG_1(igb,
4496 			    "Remove intr handler failed: %d", rc);
4497 		}
4498 	}
4499 }
4500 
4501 /*
4502  * igb_rem_intrs - remove the allocated interrupts
4503  */
4504 static void
4505 igb_rem_intrs(igb_t *igb)
4506 {
4507 	int i;
4508 	int rc;
4509 
4510 	for (i = 0; i < igb->intr_cnt; i++) {
4511 		rc = ddi_intr_free(igb->htable[i]);
4512 		if (rc != DDI_SUCCESS) {
4513 			IGB_DEBUGLOG_1(igb,
4514 			    "Free intr failed: %d", rc);
4515 		}
4516 	}
4517 
4518 	kmem_free(igb->htable, igb->intr_size);
4519 	igb->htable = NULL;
4520 }
4521 
4522 /*
4523  * igb_enable_intrs - enable all the ddi interrupts
4524  */
4525 static int
4526 igb_enable_intrs(igb_t *igb)
4527 {
4528 	int i;
4529 	int rc;
4530 
4531 	/* Enable interrupts */
4532 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4533 		/* Call ddi_intr_block_enable() for MSI */
4534 		rc = ddi_intr_block_enable(igb->htable, igb->intr_cnt);
4535 		if (rc != DDI_SUCCESS) {
4536 			igb_log(igb,
4537 			    "Enable block intr failed: %d", rc);
4538 			return (IGB_FAILURE);
4539 		}
4540 	} else {
4541 		/* Call ddi_intr_enable() for Legacy/MSI non block enable */
4542 		for (i = 0; i < igb->intr_cnt; i++) {
4543 			rc = ddi_intr_enable(igb->htable[i]);
4544 			if (rc != DDI_SUCCESS) {
4545 				igb_log(igb,
4546 				    "Enable intr failed: %d", rc);
4547 				return (IGB_FAILURE);
4548 			}
4549 		}
4550 	}
4551 
4552 	return (IGB_SUCCESS);
4553 }
4554 
4555 /*
4556  * igb_disable_intrs - disable all the ddi interrupts
4557  */
4558 static int
4559 igb_disable_intrs(igb_t *igb)
4560 {
4561 	int i;
4562 	int rc;
4563 
4564 	/* Disable all interrupts */
4565 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4566 		rc = ddi_intr_block_disable(igb->htable, igb->intr_cnt);
4567 		if (rc != DDI_SUCCESS) {
4568 			igb_log(igb,
4569 			    "Disable block intr failed: %d", rc);
4570 			return (IGB_FAILURE);
4571 		}
4572 	} else {
4573 		for (i = 0; i < igb->intr_cnt; i++) {
4574 			rc = ddi_intr_disable(igb->htable[i]);
4575 			if (rc != DDI_SUCCESS) {
4576 				igb_log(igb,
4577 				    "Disable intr failed: %d", rc);
4578 				return (IGB_FAILURE);
4579 			}
4580 		}
4581 	}
4582 
4583 	return (IGB_SUCCESS);
4584 }
4585 
4586 /*
4587  * igb_get_phy_state - Get and save the parameters read from PHY registers
4588  */
4589 static void
4590 igb_get_phy_state(igb_t *igb)
4591 {
4592 	struct e1000_hw *hw = &igb->hw;
4593 	uint16_t phy_ctrl;
4594 	uint16_t phy_status;
4595 	uint16_t phy_an_adv;
4596 	uint16_t phy_an_exp;
4597 	uint16_t phy_ext_status;
4598 	uint16_t phy_1000t_ctrl;
4599 	uint16_t phy_1000t_status;
4600 	uint16_t phy_lp_able;
4601 
4602 	ASSERT(mutex_owned(&igb->gen_lock));
4603 
4604 	(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
4605 	(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
4606 	(void) e1000_read_phy_reg(hw, PHY_AUTONEG_ADV, &phy_an_adv);
4607 	(void) e1000_read_phy_reg(hw, PHY_AUTONEG_EXP, &phy_an_exp);
4608 	(void) e1000_read_phy_reg(hw, PHY_EXT_STATUS, &phy_ext_status);
4609 	(void) e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_1000t_ctrl);
4610 	(void) e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_1000t_status);
4611 	(void) e1000_read_phy_reg(hw, PHY_LP_ABILITY, &phy_lp_able);
4612 
4613 	igb->param_autoneg_cap =
4614 	    (phy_status & MII_SR_AUTONEG_CAPS) ? 1 : 0;
4615 	igb->param_pause_cap =
4616 	    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4617 	igb->param_asym_pause_cap =
4618 	    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4619 	igb->param_1000fdx_cap = ((phy_ext_status & IEEE_ESR_1000T_FD_CAPS) ||
4620 	    (phy_ext_status & IEEE_ESR_1000X_FD_CAPS)) ? 1 : 0;
4621 	igb->param_1000hdx_cap = ((phy_ext_status & IEEE_ESR_1000T_HD_CAPS) ||
4622 	    (phy_ext_status & IEEE_ESR_1000X_HD_CAPS)) ? 1 : 0;
4623 	igb->param_100t4_cap =
4624 	    (phy_status & MII_SR_100T4_CAPS) ? 1 : 0;
4625 	igb->param_100fdx_cap = ((phy_status & MII_SR_100X_FD_CAPS) ||
4626 	    (phy_status & MII_SR_100T2_FD_CAPS)) ? 1 : 0;
4627 	igb->param_100hdx_cap = ((phy_status & MII_SR_100X_HD_CAPS) ||
4628 	    (phy_status & MII_SR_100T2_HD_CAPS)) ? 1 : 0;
4629 	igb->param_10fdx_cap =
4630 	    (phy_status & MII_SR_10T_FD_CAPS) ? 1 : 0;
4631 	igb->param_10hdx_cap =
4632 	    (phy_status & MII_SR_10T_HD_CAPS) ? 1 : 0;
4633 	igb->param_rem_fault =
4634 	    (phy_status & MII_SR_REMOTE_FAULT) ? 1 : 0;
4635 
4636 	igb->param_adv_autoneg_cap = hw->mac.autoneg;
4637 	igb->param_adv_pause_cap =
4638 	    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4639 	igb->param_adv_asym_pause_cap =
4640 	    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4641 	igb->param_adv_1000hdx_cap =
4642 	    (phy_1000t_ctrl & CR_1000T_HD_CAPS) ? 1 : 0;
4643 	igb->param_adv_100t4_cap =
4644 	    (phy_an_adv & NWAY_AR_100T4_CAPS) ? 1 : 0;
4645 	igb->param_adv_rem_fault =
4646 	    (phy_an_adv & NWAY_AR_REMOTE_FAULT) ? 1 : 0;
4647 	if (igb->param_adv_autoneg_cap == 1) {
4648 		igb->param_adv_1000fdx_cap =
4649 		    (phy_1000t_ctrl & CR_1000T_FD_CAPS) ? 1 : 0;
4650 		igb->param_adv_100fdx_cap =
4651 		    (phy_an_adv & NWAY_AR_100TX_FD_CAPS) ? 1 : 0;
4652 		igb->param_adv_100hdx_cap =
4653 		    (phy_an_adv & NWAY_AR_100TX_HD_CAPS) ? 1 : 0;
4654 		igb->param_adv_10fdx_cap =
4655 		    (phy_an_adv & NWAY_AR_10T_FD_CAPS) ? 1 : 0;
4656 		igb->param_adv_10hdx_cap =
4657 		    (phy_an_adv & NWAY_AR_10T_HD_CAPS) ? 1 : 0;
4658 	}
4659 
4660 	igb->param_lp_autoneg_cap =
4661 	    (phy_an_exp & NWAY_ER_LP_NWAY_CAPS) ? 1 : 0;
4662 	igb->param_lp_pause_cap =
4663 	    (phy_lp_able & NWAY_LPAR_PAUSE) ? 1 : 0;
4664 	igb->param_lp_asym_pause_cap =
4665 	    (phy_lp_able & NWAY_LPAR_ASM_DIR) ? 1 : 0;
4666 	igb->param_lp_1000fdx_cap =
4667 	    (phy_1000t_status & SR_1000T_LP_FD_CAPS) ? 1 : 0;
4668 	igb->param_lp_1000hdx_cap =
4669 	    (phy_1000t_status & SR_1000T_LP_HD_CAPS) ? 1 : 0;
4670 	igb->param_lp_100t4_cap =
4671 	    (phy_lp_able & NWAY_LPAR_100T4_CAPS) ? 1 : 0;
4672 	igb->param_lp_100fdx_cap =
4673 	    (phy_lp_able & NWAY_LPAR_100TX_FD_CAPS) ? 1 : 0;
4674 	igb->param_lp_100hdx_cap =
4675 	    (phy_lp_able & NWAY_LPAR_100TX_HD_CAPS) ? 1 : 0;
4676 	igb->param_lp_10fdx_cap =
4677 	    (phy_lp_able & NWAY_LPAR_10T_FD_CAPS) ? 1 : 0;
4678 	igb->param_lp_10hdx_cap =
4679 	    (phy_lp_able & NWAY_LPAR_10T_HD_CAPS) ? 1 : 0;
4680 	igb->param_lp_rem_fault =
4681 	    (phy_lp_able & NWAY_LPAR_REMOTE_FAULT) ? 1 : 0;
4682 }
4683 
4684 /*
4685  * igb_get_driver_control
4686  */
4687 static void
4688 igb_get_driver_control(struct e1000_hw *hw)
4689 {
4690 	uint32_t ctrl_ext;
4691 
4692 	/* Notify firmware that driver is in control of device */
4693 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
4694 	ctrl_ext |= E1000_CTRL_EXT_DRV_LOAD;
4695 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
4696 }
4697 
4698 /*
4699  * igb_release_driver_control
4700  */
4701 static void
4702 igb_release_driver_control(struct e1000_hw *hw)
4703 {
4704 	uint32_t ctrl_ext;
4705 
4706 	/* Notify firmware that driver is no longer in control of device */
4707 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
4708 	ctrl_ext &= ~E1000_CTRL_EXT_DRV_LOAD;
4709 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
4710 }
4711 
4712 /*
4713  * igb_atomic_reserve - Atomic decrease operation
4714  */
4715 int
4716 igb_atomic_reserve(uint32_t *count_p, uint32_t n)
4717 {
4718 	uint32_t oldval;
4719 	uint32_t newval;
4720 
4721 	/* ATOMICALLY */
4722 	do {
4723 		oldval = *count_p;
4724 		if (oldval < n)
4725 			return (-1);
4726 		newval = oldval - n;
4727 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
4728 
4729 	return (newval);
4730 }
4731 
4732 /*
4733  * FMA support
4734  */
4735 
4736 int
4737 igb_check_acc_handle(ddi_acc_handle_t handle)
4738 {
4739 	ddi_fm_error_t de;
4740 
4741 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
4742 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
4743 	return (de.fme_status);
4744 }
4745 
4746 int
4747 igb_check_dma_handle(ddi_dma_handle_t handle)
4748 {
4749 	ddi_fm_error_t de;
4750 
4751 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
4752 	return (de.fme_status);
4753 }
4754 
4755 /*
4756  * The IO fault service error handling callback function
4757  */
4758 /*ARGSUSED*/
4759 static int
4760 igb_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
4761 {
4762 	/*
4763 	 * as the driver can always deal with an error in any dma or
4764 	 * access handle, we can just return the fme_status value.
4765 	 */
4766 	pci_ereport_post(dip, err, NULL);
4767 	return (err->fme_status);
4768 }
4769 
4770 static void
4771 igb_fm_init(igb_t *igb)
4772 {
4773 	ddi_iblock_cookie_t iblk;
4774 	int fma_acc_flag, fma_dma_flag;
4775 
4776 	/* Only register with IO Fault Services if we have some capability */
4777 	if (igb->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
4778 		igb_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
4779 		fma_acc_flag = 1;
4780 	} else {
4781 		igb_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
4782 		fma_acc_flag = 0;
4783 	}
4784 
4785 	if (igb->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
4786 		fma_dma_flag = 1;
4787 	} else {
4788 		fma_dma_flag = 0;
4789 	}
4790 
4791 	(void) igb_set_fma_flags(fma_acc_flag, fma_dma_flag);
4792 
4793 	if (igb->fm_capabilities) {
4794 
4795 		/* Register capabilities with IO Fault Services */
4796 		ddi_fm_init(igb->dip, &igb->fm_capabilities, &iblk);
4797 
4798 		/*
4799 		 * Initialize pci ereport capabilities if ereport capable
4800 		 */
4801 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
4802 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4803 			pci_ereport_setup(igb->dip);
4804 
4805 		/*
4806 		 * Register error callback if error callback capable
4807 		 */
4808 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4809 			ddi_fm_handler_register(igb->dip,
4810 			    igb_fm_error_cb, (void*) igb);
4811 	}
4812 }
4813 
4814 static void
4815 igb_fm_fini(igb_t *igb)
4816 {
4817 	/* Only unregister FMA capabilities if we registered some */
4818 	if (igb->fm_capabilities) {
4819 
4820 		/*
4821 		 * Release any resources allocated by pci_ereport_setup()
4822 		 */
4823 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
4824 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4825 			pci_ereport_teardown(igb->dip);
4826 
4827 		/*
4828 		 * Un-register error callback if error callback capable
4829 		 */
4830 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4831 			ddi_fm_handler_unregister(igb->dip);
4832 
4833 		/* Unregister from IO Fault Services */
4834 		ddi_fm_fini(igb->dip);
4835 	}
4836 }
4837 
4838 void
4839 igb_fm_ereport(igb_t *igb, char *detail)
4840 {
4841 	uint64_t ena;
4842 	char buf[FM_MAX_CLASS];
4843 
4844 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
4845 	ena = fm_ena_generate(0, FM_ENA_FMT1);
4846 	if (DDI_FM_EREPORT_CAP(igb->fm_capabilities)) {
4847 		ddi_fm_ereport_post(igb->dip, buf, ena, DDI_NOSLEEP,
4848 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
4849 	}
4850 }
4851