xref: /titanic_50/usr/src/uts/common/io/igb/igb_main.c (revision 5547f1d8c68fa119522b859bbf38315dc773e696)
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 1.1.6";
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 
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 register handle
695 	 */
696 	if (igb->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
697 		if (igb->osdep.reg_handle != NULL)
698 			ddi_regs_map_free(&igb->osdep.reg_handle);
699 	}
700 
701 	/*
702 	 * Free PCI config handle
703 	 */
704 	if (igb->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
705 		if (igb->osdep.cfg_handle != NULL)
706 			pci_config_teardown(&igb->osdep.cfg_handle);
707 	}
708 
709 	/*
710 	 * Free locks
711 	 */
712 	if (igb->attach_progress & ATTACH_PROGRESS_LOCKS) {
713 		igb_destroy_locks(igb);
714 	}
715 
716 	/*
717 	 * Free the rx/tx rings
718 	 */
719 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
720 		igb_free_rings(igb);
721 	}
722 
723 	/*
724 	 * Remove FMA
725 	 */
726 	if (igb->attach_progress & ATTACH_PROGRESS_FMINIT) {
727 		igb_fm_fini(igb);
728 	}
729 
730 	/*
731 	 * Free the driver data structure
732 	 */
733 	kmem_free(igb, sizeof (igb_t));
734 
735 	ddi_set_driver_private(devinfo, NULL);
736 }
737 
738 /*
739  * igb_register_mac - Register the driver and its function pointers with
740  * the GLD interface
741  */
742 static int
743 igb_register_mac(igb_t *igb)
744 {
745 	struct e1000_hw *hw = &igb->hw;
746 	mac_register_t *mac;
747 	int status;
748 
749 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
750 		return (IGB_FAILURE);
751 
752 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
753 	mac->m_driver = igb;
754 	mac->m_dip = igb->dip;
755 	mac->m_src_addr = hw->mac.addr;
756 	mac->m_callbacks = &igb_m_callbacks;
757 	mac->m_min_sdu = 0;
758 	mac->m_max_sdu = igb->max_frame_size -
759 	    sizeof (struct ether_vlan_header) - ETHERFCSL;
760 	mac->m_margin = VLAN_TAGSZ;
761 	mac->m_v12n = MAC_VIRT_LEVEL1;
762 
763 	status = mac_register(mac, &igb->mac_hdl);
764 
765 	mac_free(mac);
766 
767 	return ((status == 0) ? IGB_SUCCESS : IGB_FAILURE);
768 }
769 
770 /*
771  * igb_identify_hardware - Identify the type of the chipset
772  */
773 static int
774 igb_identify_hardware(igb_t *igb)
775 {
776 	struct e1000_hw *hw = &igb->hw;
777 	struct igb_osdep *osdep = &igb->osdep;
778 
779 	/*
780 	 * Get the device id
781 	 */
782 	hw->vendor_id =
783 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
784 	hw->device_id =
785 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
786 	hw->revision_id =
787 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
788 	hw->subsystem_device_id =
789 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
790 	hw->subsystem_vendor_id =
791 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
792 
793 	/*
794 	 * Set the mac type of the adapter based on the device id
795 	 */
796 	if (e1000_set_mac_type(hw) != E1000_SUCCESS) {
797 		return (IGB_FAILURE);
798 	}
799 
800 	/*
801 	 * Install adapter capabilities based on mac type
802 	 */
803 	switch (hw->mac.type) {
804 	case e1000_82575:
805 		igb->capab = &igb_82575_cap;
806 		break;
807 	case e1000_82576:
808 		igb->capab = &igb_82576_cap;
809 		break;
810 	default:
811 		return (IGB_FAILURE);
812 	}
813 
814 	return (IGB_SUCCESS);
815 }
816 
817 /*
818  * igb_regs_map - Map the device registers
819  */
820 static int
821 igb_regs_map(igb_t *igb)
822 {
823 	dev_info_t *devinfo = igb->dip;
824 	struct e1000_hw *hw = &igb->hw;
825 	struct igb_osdep *osdep = &igb->osdep;
826 	off_t mem_size;
827 
828 	/*
829 	 * First get the size of device registers to be mapped.
830 	 */
831 	if (ddi_dev_regsize(devinfo, IGB_ADAPTER_REGSET, &mem_size) !=
832 	    DDI_SUCCESS) {
833 		return (IGB_FAILURE);
834 	}
835 
836 	/*
837 	 * Call ddi_regs_map_setup() to map registers
838 	 */
839 	if ((ddi_regs_map_setup(devinfo, IGB_ADAPTER_REGSET,
840 	    (caddr_t *)&hw->hw_addr, 0,
841 	    mem_size, &igb_regs_acc_attr,
842 	    &osdep->reg_handle)) != DDI_SUCCESS) {
843 		return (IGB_FAILURE);
844 	}
845 
846 	return (IGB_SUCCESS);
847 }
848 
849 /*
850  * igb_init_properties - Initialize driver properties
851  */
852 static void
853 igb_init_properties(igb_t *igb)
854 {
855 	/*
856 	 * Get conf file properties, including link settings
857 	 * jumbo frames, ring number, descriptor number, etc.
858 	 */
859 	igb_get_conf(igb);
860 }
861 
862 /*
863  * igb_init_driver_settings - Initialize driver settings
864  *
865  * The settings include hardware function pointers, bus information,
866  * rx/tx rings settings, link state, and any other parameters that
867  * need to be setup during driver initialization.
868  */
869 static int
870 igb_init_driver_settings(igb_t *igb)
871 {
872 	struct e1000_hw *hw = &igb->hw;
873 	igb_rx_ring_t *rx_ring;
874 	igb_tx_ring_t *tx_ring;
875 	uint32_t rx_size;
876 	uint32_t tx_size;
877 	int i;
878 
879 	/*
880 	 * Initialize chipset specific hardware function pointers
881 	 */
882 	if (e1000_setup_init_funcs(hw, B_TRUE) != E1000_SUCCESS) {
883 		return (IGB_FAILURE);
884 	}
885 
886 	/*
887 	 * Get bus information
888 	 */
889 	if (e1000_get_bus_info(hw) != E1000_SUCCESS) {
890 		return (IGB_FAILURE);
891 	}
892 
893 	/*
894 	 * Get the system page size
895 	 */
896 	igb->page_size = ddi_ptob(igb->dip, (ulong_t)1);
897 
898 	/*
899 	 * Set rx buffer size
900 	 * The IP header alignment room is counted in the calculation.
901 	 * The rx buffer size is in unit of 1K that is required by the
902 	 * chipset hardware.
903 	 */
904 	rx_size = igb->max_frame_size + IPHDR_ALIGN_ROOM;
905 	igb->rx_buf_size = ((rx_size >> 10) +
906 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
907 
908 	/*
909 	 * Set tx buffer size
910 	 */
911 	tx_size = igb->max_frame_size;
912 	igb->tx_buf_size = ((tx_size >> 10) +
913 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
914 
915 	/*
916 	 * Initialize rx/tx rings parameters
917 	 */
918 	for (i = 0; i < igb->num_rx_rings; i++) {
919 		rx_ring = &igb->rx_rings[i];
920 		rx_ring->index = i;
921 		rx_ring->igb = igb;
922 
923 		rx_ring->ring_size = igb->rx_ring_size;
924 		rx_ring->free_list_size = igb->rx_ring_size;
925 		rx_ring->copy_thresh = igb->rx_copy_thresh;
926 		rx_ring->limit_per_intr = igb->rx_limit_per_intr;
927 	}
928 
929 	for (i = 0; i < igb->num_tx_rings; i++) {
930 		tx_ring = &igb->tx_rings[i];
931 		tx_ring->index = i;
932 		tx_ring->igb = igb;
933 		if (igb->tx_head_wb_enable)
934 			tx_ring->tx_recycle = igb_tx_recycle_head_wb;
935 		else
936 			tx_ring->tx_recycle = igb_tx_recycle_legacy;
937 
938 		tx_ring->ring_size = igb->tx_ring_size;
939 		tx_ring->free_list_size = igb->tx_ring_size +
940 		    (igb->tx_ring_size >> 1);
941 		tx_ring->copy_thresh = igb->tx_copy_thresh;
942 		tx_ring->recycle_thresh = igb->tx_recycle_thresh;
943 		tx_ring->overload_thresh = igb->tx_overload_thresh;
944 		tx_ring->resched_thresh = igb->tx_resched_thresh;
945 	}
946 
947 	/*
948 	 * Initialize values of interrupt throttling rates
949 	 */
950 	for (i = 1; i < MAX_NUM_EITR; i++)
951 		igb->intr_throttling[i] = igb->intr_throttling[0];
952 
953 	/*
954 	 * The initial link state should be "unknown"
955 	 */
956 	igb->link_state = LINK_STATE_UNKNOWN;
957 
958 	return (IGB_SUCCESS);
959 }
960 
961 /*
962  * igb_init_locks - Initialize locks
963  */
964 static void
965 igb_init_locks(igb_t *igb)
966 {
967 	igb_rx_ring_t *rx_ring;
968 	igb_tx_ring_t *tx_ring;
969 	int i;
970 
971 	for (i = 0; i < igb->num_rx_rings; i++) {
972 		rx_ring = &igb->rx_rings[i];
973 		mutex_init(&rx_ring->rx_lock, NULL,
974 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
975 		mutex_init(&rx_ring->recycle_lock, NULL,
976 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
977 	}
978 
979 	for (i = 0; i < igb->num_tx_rings; i++) {
980 		tx_ring = &igb->tx_rings[i];
981 		mutex_init(&tx_ring->tx_lock, NULL,
982 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
983 		mutex_init(&tx_ring->recycle_lock, NULL,
984 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
985 		mutex_init(&tx_ring->tcb_head_lock, NULL,
986 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
987 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
988 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
989 	}
990 
991 	mutex_init(&igb->gen_lock, NULL,
992 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
993 
994 	mutex_init(&igb->watchdog_lock, NULL,
995 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
996 }
997 
998 /*
999  * igb_destroy_locks - Destroy locks
1000  */
1001 static void
1002 igb_destroy_locks(igb_t *igb)
1003 {
1004 	igb_rx_ring_t *rx_ring;
1005 	igb_tx_ring_t *tx_ring;
1006 	int i;
1007 
1008 	for (i = 0; i < igb->num_rx_rings; i++) {
1009 		rx_ring = &igb->rx_rings[i];
1010 		mutex_destroy(&rx_ring->rx_lock);
1011 		mutex_destroy(&rx_ring->recycle_lock);
1012 	}
1013 
1014 	for (i = 0; i < igb->num_tx_rings; i++) {
1015 		tx_ring = &igb->tx_rings[i];
1016 		mutex_destroy(&tx_ring->tx_lock);
1017 		mutex_destroy(&tx_ring->recycle_lock);
1018 		mutex_destroy(&tx_ring->tcb_head_lock);
1019 		mutex_destroy(&tx_ring->tcb_tail_lock);
1020 	}
1021 
1022 	mutex_destroy(&igb->gen_lock);
1023 	mutex_destroy(&igb->watchdog_lock);
1024 }
1025 
1026 static int
1027 igb_resume(dev_info_t *devinfo)
1028 {
1029 	igb_t *igb;
1030 
1031 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1032 	if (igb == NULL)
1033 		return (DDI_FAILURE);
1034 
1035 	mutex_enter(&igb->gen_lock);
1036 
1037 	if (igb->igb_state & IGB_STARTED) {
1038 		if (igb_start(igb) != IGB_SUCCESS) {
1039 			mutex_exit(&igb->gen_lock);
1040 			return (DDI_FAILURE);
1041 		}
1042 
1043 		/*
1044 		 * Enable and start the watchdog timer
1045 		 */
1046 		igb_enable_watchdog_timer(igb);
1047 	}
1048 
1049 	igb->igb_state &= ~IGB_SUSPENDED;
1050 
1051 	mutex_exit(&igb->gen_lock);
1052 
1053 	return (DDI_SUCCESS);
1054 }
1055 
1056 static int
1057 igb_suspend(dev_info_t *devinfo)
1058 {
1059 	igb_t *igb;
1060 
1061 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1062 	if (igb == NULL)
1063 		return (DDI_FAILURE);
1064 
1065 	mutex_enter(&igb->gen_lock);
1066 
1067 	igb->igb_state |= IGB_SUSPENDED;
1068 
1069 	if (!(igb->igb_state & IGB_STARTED)) {
1070 		mutex_exit(&igb->gen_lock);
1071 		return (DDI_SUCCESS);
1072 	}
1073 
1074 	igb_stop(igb);
1075 
1076 	mutex_exit(&igb->gen_lock);
1077 
1078 	/*
1079 	 * Disable and stop the watchdog timer
1080 	 */
1081 	igb_disable_watchdog_timer(igb);
1082 
1083 	return (DDI_SUCCESS);
1084 }
1085 
1086 static int
1087 igb_init(igb_t *igb)
1088 {
1089 	int i;
1090 
1091 	mutex_enter(&igb->gen_lock);
1092 
1093 	/*
1094 	 * Initilize the adapter
1095 	 */
1096 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1097 		mutex_exit(&igb->gen_lock);
1098 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1099 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1100 		return (IGB_FAILURE);
1101 	}
1102 
1103 	/*
1104 	 * Setup the rx/tx rings
1105 	 */
1106 	for (i = 0; i < igb->num_rx_rings; i++)
1107 		mutex_enter(&igb->rx_rings[i].rx_lock);
1108 	for (i = 0; i < igb->num_tx_rings; i++)
1109 		mutex_enter(&igb->tx_rings[i].tx_lock);
1110 
1111 	igb_setup_rings(igb);
1112 
1113 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1114 		mutex_exit(&igb->tx_rings[i].tx_lock);
1115 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1116 		mutex_exit(&igb->rx_rings[i].rx_lock);
1117 
1118 	mutex_exit(&igb->gen_lock);
1119 
1120 	return (IGB_SUCCESS);
1121 }
1122 
1123 /*
1124  * igb_init_mac_address - Initialize the default MAC address
1125  *
1126  * On success, the MAC address is entered in the igb->hw.mac.addr
1127  * and hw->mac.perm_addr fields and the adapter's RAR(0) receive
1128  * address register.
1129  *
1130  * Important side effects:
1131  * 1. adapter is reset - this is required to put it in a known state.
1132  * 2. all of non-volatile memory (NVM) is read & checksummed - NVM is where
1133  * MAC address and all default settings are stored, so a valid checksum
1134  * is required.
1135  */
1136 static int
1137 igb_init_mac_address(igb_t *igb)
1138 {
1139 	struct e1000_hw *hw = &igb->hw;
1140 
1141 	ASSERT(mutex_owned(&igb->gen_lock));
1142 
1143 	/*
1144 	 * Reset chipset to put the hardware in a known state
1145 	 * before we try to get MAC address from NVM.
1146 	 */
1147 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1148 		igb_error(igb, "Adapter reset failed.");
1149 		goto init_mac_fail;
1150 	}
1151 
1152 	/*
1153 	 * NVM validation
1154 	 */
1155 	if (e1000_validate_nvm_checksum(hw) < 0) {
1156 		/*
1157 		 * Some PCI-E parts fail the first check due to
1158 		 * the link being in sleep state.  Call it again,
1159 		 * if it fails a second time its a real issue.
1160 		 */
1161 		if (e1000_validate_nvm_checksum(hw) < 0) {
1162 			igb_error(igb,
1163 			    "Invalid NVM checksum. Please contact "
1164 			    "the vendor to update the NVM.");
1165 			goto init_mac_fail;
1166 		}
1167 	}
1168 
1169 	/*
1170 	 * Get the mac address
1171 	 * This function should handle SPARC case correctly.
1172 	 */
1173 	if (!igb_find_mac_address(igb)) {
1174 		igb_error(igb, "Failed to get the mac address");
1175 		goto init_mac_fail;
1176 	}
1177 
1178 	/* Validate mac address */
1179 	if (!is_valid_mac_addr(hw->mac.addr)) {
1180 		igb_error(igb, "Invalid mac address");
1181 		goto init_mac_fail;
1182 	}
1183 
1184 	return (IGB_SUCCESS);
1185 
1186 init_mac_fail:
1187 	return (IGB_FAILURE);
1188 }
1189 
1190 /*
1191  * igb_init_adapter - Initialize the adapter
1192  */
1193 static int
1194 igb_init_adapter(igb_t *igb)
1195 {
1196 	struct e1000_hw *hw = &igb->hw;
1197 	uint32_t pba;
1198 	uint32_t high_water;
1199 	int i;
1200 
1201 	ASSERT(mutex_owned(&igb->gen_lock));
1202 
1203 	/*
1204 	 * In order to obtain the default MAC address, this will reset the
1205 	 * adapter and validate the NVM that the address and many other
1206 	 * default settings come from.
1207 	 */
1208 	if (igb_init_mac_address(igb) != IGB_SUCCESS) {
1209 		igb_error(igb, "Failed to initialize MAC address");
1210 		goto init_adapter_fail;
1211 	}
1212 
1213 	/*
1214 	 * Setup flow control
1215 	 *
1216 	 * These parameters set thresholds for the adapter's generation(Tx)
1217 	 * and response(Rx) to Ethernet PAUSE frames.  These are just threshold
1218 	 * settings.  Flow control is enabled or disabled in the configuration
1219 	 * file.
1220 	 * High-water mark is set down from the top of the rx fifo (not
1221 	 * sensitive to max_frame_size) and low-water is set just below
1222 	 * high-water mark.
1223 	 * The high water mark must be low enough to fit one full frame above
1224 	 * it in the rx FIFO.  Should be the lower of:
1225 	 * 90% of the Rx FIFO size, or the full Rx FIFO size minus one full
1226 	 * frame.
1227 	 */
1228 	/*
1229 	 * The default setting of PBA is correct for 82575 and other supported
1230 	 * adapters do not have the E1000_PBA register, so PBA value is only
1231 	 * used for calculation here and is never written to the adapter.
1232 	 */
1233 	if (hw->mac.type == e1000_82575) {
1234 		pba = E1000_PBA_34K;
1235 	} else {
1236 		pba = E1000_PBA_64K;
1237 	}
1238 
1239 	high_water = min(((pba << 10) * 9 / 10),
1240 	    ((pba << 10) - igb->max_frame_size));
1241 
1242 	if (hw->mac.type == e1000_82575) {
1243 		/* 8-byte granularity */
1244 		hw->fc.high_water = high_water & 0xFFF8;
1245 		hw->fc.low_water = hw->fc.high_water - 8;
1246 	} else {
1247 		/* 16-byte granularity */
1248 		hw->fc.high_water = high_water & 0xFFF0;
1249 		hw->fc.low_water = hw->fc.high_water - 16;
1250 	}
1251 
1252 	hw->fc.pause_time = E1000_FC_PAUSE_TIME;
1253 	hw->fc.send_xon = B_TRUE;
1254 
1255 	e1000_validate_mdi_setting(hw);
1256 
1257 	/*
1258 	 * Reset the chipset hardware the second time to put PBA settings
1259 	 * into effect.
1260 	 */
1261 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1262 		igb_error(igb, "Second reset failed");
1263 		goto init_adapter_fail;
1264 	}
1265 
1266 	/*
1267 	 * Don't wait for auto-negotiation to complete
1268 	 */
1269 	hw->phy.autoneg_wait_to_complete = B_FALSE;
1270 
1271 	/*
1272 	 * Copper options
1273 	 */
1274 	if (hw->phy.media_type == e1000_media_type_copper) {
1275 		hw->phy.mdix = 0;	/* AUTO_ALL_MODES */
1276 		hw->phy.disable_polarity_correction = B_FALSE;
1277 		hw->phy.ms_type = e1000_ms_hw_default; /* E1000_MASTER_SLAVE */
1278 	}
1279 
1280 	/*
1281 	 * Initialize link settings
1282 	 */
1283 	(void) igb_setup_link(igb, B_FALSE);
1284 
1285 	/*
1286 	 * Configure/Initialize hardware
1287 	 */
1288 	if (e1000_init_hw(hw) != E1000_SUCCESS) {
1289 		igb_error(igb, "Failed to initialize hardware");
1290 		goto init_adapter_fail;
1291 	}
1292 
1293 	/*
1294 	 * Disable wakeup control by default
1295 	 */
1296 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1297 
1298 	/*
1299 	 * Record phy info in hw struct
1300 	 */
1301 	(void) e1000_get_phy_info(hw);
1302 
1303 	/*
1304 	 * Make sure driver has control
1305 	 */
1306 	igb_get_driver_control(hw);
1307 
1308 	/*
1309 	 * Restore LED settings to the default from EEPROM
1310 	 * to meet the standard for Sun platforms.
1311 	 */
1312 	(void) e1000_cleanup_led(hw);
1313 
1314 	/*
1315 	 * Setup MSI-X interrupts
1316 	 */
1317 	if (igb->intr_type == DDI_INTR_TYPE_MSIX)
1318 		igb->capab->setup_msix(igb);
1319 
1320 	/*
1321 	 * Initialize unicast addresses.
1322 	 */
1323 	igb_init_unicst(igb);
1324 
1325 	/*
1326 	 * Setup and initialize the mctable structures.
1327 	 */
1328 	igb_setup_multicst(igb);
1329 
1330 	/*
1331 	 * Set interrupt throttling rate
1332 	 */
1333 	for (i = 0; i < igb->intr_cnt; i++)
1334 		E1000_WRITE_REG(hw, E1000_EITR(i), igb->intr_throttling[i]);
1335 
1336 	/*
1337 	 * Save the state of the phy
1338 	 */
1339 	igb_get_phy_state(igb);
1340 
1341 	return (IGB_SUCCESS);
1342 
1343 init_adapter_fail:
1344 	/*
1345 	 * Reset PHY if possible
1346 	 */
1347 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
1348 		(void) e1000_phy_hw_reset(hw);
1349 
1350 	return (IGB_FAILURE);
1351 }
1352 
1353 /*
1354  * igb_stop_adapter - Stop the adapter
1355  */
1356 static void
1357 igb_stop_adapter(igb_t *igb)
1358 {
1359 	struct e1000_hw *hw = &igb->hw;
1360 
1361 	ASSERT(mutex_owned(&igb->gen_lock));
1362 
1363 	/* Tell firmware driver is no longer in control */
1364 	igb_release_driver_control(hw);
1365 
1366 	/*
1367 	 * Reset the chipset
1368 	 */
1369 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1370 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1371 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1372 	}
1373 
1374 	/*
1375 	 * e1000_phy_hw_reset is not needed here, MAC reset above is sufficient
1376 	 */
1377 }
1378 
1379 /*
1380  * igb_reset - Reset the chipset and restart the driver.
1381  *
1382  * It involves stopping and re-starting the chipset,
1383  * and re-configuring the rx/tx rings.
1384  */
1385 static int
1386 igb_reset(igb_t *igb)
1387 {
1388 	int i;
1389 
1390 	mutex_enter(&igb->gen_lock);
1391 
1392 	ASSERT(igb->igb_state & IGB_STARTED);
1393 
1394 	/*
1395 	 * Disable the adapter interrupts to stop any rx/tx activities
1396 	 * before draining pending data and resetting hardware.
1397 	 */
1398 	igb_disable_adapter_interrupts(igb);
1399 
1400 	/*
1401 	 * Drain the pending transmit packets
1402 	 */
1403 	(void) igb_tx_drain(igb);
1404 
1405 	for (i = 0; i < igb->num_rx_rings; i++)
1406 		mutex_enter(&igb->rx_rings[i].rx_lock);
1407 	for (i = 0; i < igb->num_tx_rings; i++)
1408 		mutex_enter(&igb->tx_rings[i].tx_lock);
1409 
1410 	/*
1411 	 * Stop the adapter
1412 	 */
1413 	igb_stop_adapter(igb);
1414 
1415 	/*
1416 	 * Clean the pending tx data/resources
1417 	 */
1418 	igb_tx_clean(igb);
1419 
1420 	/*
1421 	 * Start the adapter
1422 	 */
1423 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1424 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1425 		goto reset_failure;
1426 	}
1427 
1428 	/*
1429 	 * Setup the rx/tx rings
1430 	 */
1431 	igb_setup_rings(igb);
1432 
1433 	/*
1434 	 * Enable adapter interrupts
1435 	 * The interrupts must be enabled after the driver state is START
1436 	 */
1437 	igb->capab->enable_intr(igb);
1438 
1439 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1440 		goto reset_failure;
1441 
1442 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1443 		goto reset_failure;
1444 
1445 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1446 		mutex_exit(&igb->tx_rings[i].tx_lock);
1447 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1448 		mutex_exit(&igb->rx_rings[i].rx_lock);
1449 
1450 	mutex_exit(&igb->gen_lock);
1451 
1452 	return (IGB_SUCCESS);
1453 
1454 reset_failure:
1455 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1456 		mutex_exit(&igb->tx_rings[i].tx_lock);
1457 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1458 		mutex_exit(&igb->rx_rings[i].rx_lock);
1459 
1460 	mutex_exit(&igb->gen_lock);
1461 
1462 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1463 
1464 	return (IGB_FAILURE);
1465 }
1466 
1467 /*
1468  * igb_tx_clean - Clean the pending transmit packets and DMA resources
1469  */
1470 static void
1471 igb_tx_clean(igb_t *igb)
1472 {
1473 	igb_tx_ring_t *tx_ring;
1474 	tx_control_block_t *tcb;
1475 	link_list_t pending_list;
1476 	uint32_t desc_num;
1477 	int i, j;
1478 
1479 	LINK_LIST_INIT(&pending_list);
1480 
1481 	for (i = 0; i < igb->num_tx_rings; i++) {
1482 		tx_ring = &igb->tx_rings[i];
1483 
1484 		mutex_enter(&tx_ring->recycle_lock);
1485 
1486 		/*
1487 		 * Clean the pending tx data - the pending packets in the
1488 		 * work_list that have no chances to be transmitted again.
1489 		 *
1490 		 * We must ensure the chipset is stopped or the link is down
1491 		 * before cleaning the transmit packets.
1492 		 */
1493 		desc_num = 0;
1494 		for (j = 0; j < tx_ring->ring_size; j++) {
1495 			tcb = tx_ring->work_list[j];
1496 			if (tcb != NULL) {
1497 				desc_num += tcb->desc_num;
1498 
1499 				tx_ring->work_list[j] = NULL;
1500 
1501 				igb_free_tcb(tcb);
1502 
1503 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1504 			}
1505 		}
1506 
1507 		if (desc_num > 0) {
1508 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1509 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1510 
1511 			/*
1512 			 * Reset the head and tail pointers of the tbd ring;
1513 			 * Reset the head write-back if it is enabled.
1514 			 */
1515 			tx_ring->tbd_head = 0;
1516 			tx_ring->tbd_tail = 0;
1517 			if (igb->tx_head_wb_enable)
1518 				*tx_ring->tbd_head_wb = 0;
1519 
1520 			E1000_WRITE_REG(&igb->hw, E1000_TDH(tx_ring->index), 0);
1521 			E1000_WRITE_REG(&igb->hw, E1000_TDT(tx_ring->index), 0);
1522 		}
1523 
1524 		mutex_exit(&tx_ring->recycle_lock);
1525 
1526 		/*
1527 		 * Add the tx control blocks in the pending list to
1528 		 * the free list.
1529 		 */
1530 		igb_put_free_list(tx_ring, &pending_list);
1531 	}
1532 }
1533 
1534 /*
1535  * igb_tx_drain - Drain the tx rings to allow pending packets to be transmitted
1536  */
1537 static boolean_t
1538 igb_tx_drain(igb_t *igb)
1539 {
1540 	igb_tx_ring_t *tx_ring;
1541 	boolean_t done;
1542 	int i, j;
1543 
1544 	/*
1545 	 * Wait for a specific time to allow pending tx packets
1546 	 * to be transmitted.
1547 	 *
1548 	 * Check the counter tbd_free to see if transmission is done.
1549 	 * No lock protection is needed here.
1550 	 *
1551 	 * Return B_TRUE if all pending packets have been transmitted;
1552 	 * Otherwise return B_FALSE;
1553 	 */
1554 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1555 
1556 		done = B_TRUE;
1557 		for (j = 0; j < igb->num_tx_rings; j++) {
1558 			tx_ring = &igb->tx_rings[j];
1559 			done = done &&
1560 			    (tx_ring->tbd_free == tx_ring->ring_size);
1561 		}
1562 
1563 		if (done)
1564 			break;
1565 
1566 		msec_delay(1);
1567 	}
1568 
1569 	return (done);
1570 }
1571 
1572 /*
1573  * igb_rx_drain - Wait for all rx buffers to be released by upper layer
1574  */
1575 static boolean_t
1576 igb_rx_drain(igb_t *igb)
1577 {
1578 	igb_rx_ring_t *rx_ring;
1579 	boolean_t done;
1580 	int i, j;
1581 
1582 	/*
1583 	 * Polling the rx free list to check if those rx buffers held by
1584 	 * the upper layer are released.
1585 	 *
1586 	 * Check the counter rcb_free to see if all pending buffers are
1587 	 * released. No lock protection is needed here.
1588 	 *
1589 	 * Return B_TRUE if all pending buffers have been released;
1590 	 * Otherwise return B_FALSE;
1591 	 */
1592 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1593 
1594 		done = B_TRUE;
1595 		for (j = 0; j < igb->num_rx_rings; j++) {
1596 			rx_ring = &igb->rx_rings[j];
1597 			done = done &&
1598 			    (rx_ring->rcb_free == rx_ring->free_list_size);
1599 		}
1600 
1601 		if (done)
1602 			break;
1603 
1604 		msec_delay(1);
1605 	}
1606 
1607 	return (done);
1608 }
1609 
1610 /*
1611  * igb_start - Start the driver/chipset
1612  */
1613 int
1614 igb_start(igb_t *igb)
1615 {
1616 	int i;
1617 
1618 	ASSERT(mutex_owned(&igb->gen_lock));
1619 
1620 	for (i = 0; i < igb->num_rx_rings; i++)
1621 		mutex_enter(&igb->rx_rings[i].rx_lock);
1622 	for (i = 0; i < igb->num_tx_rings; i++)
1623 		mutex_enter(&igb->tx_rings[i].tx_lock);
1624 
1625 	/*
1626 	 * Start the adapter
1627 	 */
1628 	if ((igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) == 0) {
1629 		if (igb_init_adapter(igb) != IGB_SUCCESS) {
1630 			igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1631 			goto start_failure;
1632 		}
1633 		igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
1634 
1635 		/*
1636 		 * Setup the rx/tx rings
1637 		 */
1638 		igb_setup_rings(igb);
1639 	}
1640 
1641 	/*
1642 	 * Enable adapter interrupts
1643 	 * The interrupts must be enabled after the driver state is START
1644 	 */
1645 	igb->capab->enable_intr(igb);
1646 
1647 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1648 		goto start_failure;
1649 
1650 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1651 		goto start_failure;
1652 
1653 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1654 		mutex_exit(&igb->tx_rings[i].tx_lock);
1655 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1656 		mutex_exit(&igb->rx_rings[i].rx_lock);
1657 
1658 	return (IGB_SUCCESS);
1659 
1660 start_failure:
1661 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1662 		mutex_exit(&igb->tx_rings[i].tx_lock);
1663 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1664 		mutex_exit(&igb->rx_rings[i].rx_lock);
1665 
1666 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1667 
1668 	return (IGB_FAILURE);
1669 }
1670 
1671 /*
1672  * igb_stop - Stop the driver/chipset
1673  */
1674 void
1675 igb_stop(igb_t *igb)
1676 {
1677 	int i;
1678 
1679 	ASSERT(mutex_owned(&igb->gen_lock));
1680 
1681 	igb->attach_progress &= ~ATTACH_PROGRESS_INIT_ADAPTER;
1682 
1683 	/*
1684 	 * Disable the adapter interrupts
1685 	 */
1686 	igb_disable_adapter_interrupts(igb);
1687 
1688 	/*
1689 	 * Drain the pending tx packets
1690 	 */
1691 	(void) igb_tx_drain(igb);
1692 
1693 	for (i = 0; i < igb->num_rx_rings; i++)
1694 		mutex_enter(&igb->rx_rings[i].rx_lock);
1695 	for (i = 0; i < igb->num_tx_rings; i++)
1696 		mutex_enter(&igb->tx_rings[i].tx_lock);
1697 
1698 	/*
1699 	 * Stop the adapter
1700 	 */
1701 	igb_stop_adapter(igb);
1702 
1703 	/*
1704 	 * Clean the pending tx data/resources
1705 	 */
1706 	igb_tx_clean(igb);
1707 
1708 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1709 		mutex_exit(&igb->tx_rings[i].tx_lock);
1710 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1711 		mutex_exit(&igb->rx_rings[i].rx_lock);
1712 
1713 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1714 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1715 }
1716 
1717 /*
1718  * igb_alloc_rings - Allocate memory space for rx/tx rings
1719  */
1720 static int
1721 igb_alloc_rings(igb_t *igb)
1722 {
1723 	/*
1724 	 * Allocate memory space for rx rings
1725 	 */
1726 	igb->rx_rings = kmem_zalloc(
1727 	    sizeof (igb_rx_ring_t) * igb->num_rx_rings,
1728 	    KM_NOSLEEP);
1729 
1730 	if (igb->rx_rings == NULL) {
1731 		return (IGB_FAILURE);
1732 	}
1733 
1734 	/*
1735 	 * Allocate memory space for tx rings
1736 	 */
1737 	igb->tx_rings = kmem_zalloc(
1738 	    sizeof (igb_tx_ring_t) * igb->num_tx_rings,
1739 	    KM_NOSLEEP);
1740 
1741 	if (igb->tx_rings == NULL) {
1742 		kmem_free(igb->rx_rings,
1743 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1744 		igb->rx_rings = NULL;
1745 		return (IGB_FAILURE);
1746 	}
1747 
1748 	/*
1749 	 * Allocate memory space for rx ring groups
1750 	 */
1751 	igb->rx_groups = kmem_zalloc(
1752 	    sizeof (igb_rx_group_t) * igb->num_rx_groups,
1753 	    KM_NOSLEEP);
1754 
1755 	if (igb->rx_groups == NULL) {
1756 		kmem_free(igb->rx_rings,
1757 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1758 		kmem_free(igb->tx_rings,
1759 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1760 		igb->rx_rings = NULL;
1761 		igb->tx_rings = NULL;
1762 		return (IGB_FAILURE);
1763 	}
1764 
1765 	return (IGB_SUCCESS);
1766 }
1767 
1768 /*
1769  * igb_free_rings - Free the memory space of rx/tx rings.
1770  */
1771 static void
1772 igb_free_rings(igb_t *igb)
1773 {
1774 	if (igb->rx_rings != NULL) {
1775 		kmem_free(igb->rx_rings,
1776 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1777 		igb->rx_rings = NULL;
1778 	}
1779 
1780 	if (igb->tx_rings != NULL) {
1781 		kmem_free(igb->tx_rings,
1782 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1783 		igb->tx_rings = NULL;
1784 	}
1785 
1786 	if (igb->rx_groups != NULL) {
1787 		kmem_free(igb->rx_groups,
1788 		    sizeof (igb_rx_group_t) * igb->num_rx_groups);
1789 		igb->rx_groups = NULL;
1790 	}
1791 }
1792 
1793 /*
1794  * igb_setup_rings - Setup rx/tx rings
1795  */
1796 static void
1797 igb_setup_rings(igb_t *igb)
1798 {
1799 	/*
1800 	 * Setup the rx/tx rings, including the following:
1801 	 *
1802 	 * 1. Setup the descriptor ring and the control block buffers;
1803 	 * 2. Initialize necessary registers for receive/transmit;
1804 	 * 3. Initialize software pointers/parameters for receive/transmit;
1805 	 */
1806 	igb_setup_rx(igb);
1807 
1808 	igb_setup_tx(igb);
1809 
1810 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1811 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1812 }
1813 
1814 static void
1815 igb_setup_rx_ring(igb_rx_ring_t *rx_ring)
1816 {
1817 	igb_t *igb = rx_ring->igb;
1818 	struct e1000_hw *hw = &igb->hw;
1819 	rx_control_block_t *rcb;
1820 	union e1000_adv_rx_desc	*rbd;
1821 	uint32_t size;
1822 	uint32_t buf_low;
1823 	uint32_t buf_high;
1824 	uint32_t rxdctl;
1825 	int i;
1826 
1827 	ASSERT(mutex_owned(&rx_ring->rx_lock));
1828 	ASSERT(mutex_owned(&igb->gen_lock));
1829 
1830 	/*
1831 	 * Initialize descriptor ring with buffer addresses
1832 	 */
1833 	for (i = 0; i < igb->rx_ring_size; i++) {
1834 		rcb = rx_ring->work_list[i];
1835 		rbd = &rx_ring->rbd_ring[i];
1836 
1837 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
1838 		rbd->read.hdr_addr = NULL;
1839 	}
1840 
1841 	/*
1842 	 * Initialize the base address registers
1843 	 */
1844 	buf_low = (uint32_t)rx_ring->rbd_area.dma_address;
1845 	buf_high = (uint32_t)(rx_ring->rbd_area.dma_address >> 32);
1846 	E1000_WRITE_REG(hw, E1000_RDBAH(rx_ring->index), buf_high);
1847 	E1000_WRITE_REG(hw, E1000_RDBAL(rx_ring->index), buf_low);
1848 
1849 	/*
1850 	 * Initialize the length register
1851 	 */
1852 	size = rx_ring->ring_size * sizeof (union e1000_adv_rx_desc);
1853 	E1000_WRITE_REG(hw, E1000_RDLEN(rx_ring->index), size);
1854 
1855 	/*
1856 	 * Initialize buffer size & descriptor type
1857 	 */
1858 	E1000_WRITE_REG(hw, E1000_SRRCTL(rx_ring->index),
1859 	    ((igb->rx_buf_size >> E1000_SRRCTL_BSIZEPKT_SHIFT) |
1860 	    E1000_SRRCTL_DESCTYPE_ADV_ONEBUF));
1861 
1862 	/*
1863 	 * Setup the Receive Descriptor Control Register (RXDCTL)
1864 	 */
1865 	rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(rx_ring->index));
1866 	rxdctl &= igb->capab->rxdctl_mask;
1867 	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1868 	rxdctl |= 16;		/* pthresh */
1869 	rxdctl |= 8 << 8;	/* hthresh */
1870 	rxdctl |= 1 << 16;	/* wthresh */
1871 	E1000_WRITE_REG(hw, E1000_RXDCTL(rx_ring->index), rxdctl);
1872 
1873 	rx_ring->rbd_next = 0;
1874 
1875 	/*
1876 	 * Note: Considering the case that the chipset is being reset
1877 	 * and there are still some buffers held by the upper layer,
1878 	 * we should not reset the values of rcb_head, rcb_tail and
1879 	 * rcb_free;
1880 	 */
1881 	if (igb->igb_state == IGB_UNKNOWN) {
1882 		rx_ring->rcb_head = 0;
1883 		rx_ring->rcb_tail = 0;
1884 		rx_ring->rcb_free = rx_ring->free_list_size;
1885 	}
1886 }
1887 
1888 static void
1889 igb_setup_rx(igb_t *igb)
1890 {
1891 	igb_rx_ring_t *rx_ring;
1892 	igb_rx_group_t *rx_group;
1893 	struct e1000_hw *hw = &igb->hw;
1894 	uint32_t rctl, rxcsum;
1895 	uint32_t ring_per_group;
1896 	int i;
1897 
1898 	/*
1899 	 * Setup the Receive Control Register (RCTL), and enable the
1900 	 * receiver. The initial configuration is to: enable the receiver,
1901 	 * accept broadcasts, discard bad packets, accept long packets,
1902 	 * disable VLAN filter checking, and set receive buffer size to
1903 	 * 2k.  For 82575, also set the receive descriptor minimum
1904 	 * threshold size to 1/2 the ring.
1905 	 */
1906 	rctl = E1000_READ_REG(hw, E1000_RCTL);
1907 
1908 	/*
1909 	 * Clear the field used for wakeup control.  This driver doesn't do
1910 	 * wakeup but leave this here for completeness.
1911 	 */
1912 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1913 
1914 	switch (hw->mac.type) {
1915 	case e1000_82575:
1916 		rctl |= (E1000_RCTL_EN |	/* Enable Receive Unit */
1917 		    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
1918 		    E1000_RCTL_LPE |		/* Large Packet Enable */
1919 						/* Multicast filter offset */
1920 		    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
1921 		    E1000_RCTL_RDMTS_HALF |	/* rx descriptor threshold */
1922 		    E1000_RCTL_SECRC);		/* Strip Ethernet CRC */
1923 		break;
1924 
1925 	case e1000_82576:
1926 		rctl |= (E1000_RCTL_EN |	/* Enable Receive Unit */
1927 		    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
1928 		    E1000_RCTL_LPE |		/* Large Packet Enable */
1929 						/* Multicast filter offset */
1930 		    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
1931 		    E1000_RCTL_SECRC);		/* Strip Ethernet CRC */
1932 		break;
1933 
1934 	default:
1935 		igb_log(igb, "unsupported MAC type: %d", hw->mac.type);
1936 		return;	/* should never come here; this will cause rx failure */
1937 	}
1938 
1939 	for (i = 0; i < igb->num_rx_groups; i++) {
1940 		rx_group = &igb->rx_groups[i];
1941 		rx_group->index = i;
1942 		rx_group->igb = igb;
1943 	}
1944 
1945 	/*
1946 	 * Set up all rx descriptor rings - must be called before receive unit
1947 	 * enabled.
1948 	 */
1949 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
1950 	for (i = 0; i < igb->num_rx_rings; i++) {
1951 		rx_ring = &igb->rx_rings[i];
1952 		igb_setup_rx_ring(rx_ring);
1953 
1954 		/*
1955 		 * Map a ring to a group by assigning a group index
1956 		 */
1957 		rx_ring->group_index = i / ring_per_group;
1958 	}
1959 
1960 	/*
1961 	 * Setup the Rx Long Packet Max Length register
1962 	 */
1963 	E1000_WRITE_REG(hw, E1000_RLPML, igb->max_frame_size);
1964 
1965 	/*
1966 	 * Hardware checksum settings
1967 	 */
1968 	if (igb->rx_hcksum_enable) {
1969 		rxcsum =
1970 		    E1000_RXCSUM_TUOFL |	/* TCP/UDP checksum */
1971 		    E1000_RXCSUM_IPOFL;		/* IP checksum */
1972 
1973 		E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
1974 	}
1975 
1976 	/*
1977 	 * Setup classify and RSS for multiple receive queues
1978 	 */
1979 	switch (igb->vmdq_mode) {
1980 	case E1000_VMDQ_OFF:
1981 		/*
1982 		 * One ring group, only RSS is needed when more than
1983 		 * one ring enabled.
1984 		 */
1985 		if (igb->num_rx_rings > 1)
1986 			igb_setup_rss(igb);
1987 		break;
1988 	case E1000_VMDQ_MAC:
1989 		/*
1990 		 * Multiple groups, each group has one ring,
1991 		 * only the MAC classification is needed.
1992 		 */
1993 		igb_setup_mac_classify(igb);
1994 		break;
1995 	case E1000_VMDQ_MAC_RSS:
1996 		/*
1997 		 * Multiple groups and multiple rings, both
1998 		 * MAC classification and RSS are needed.
1999 		 */
2000 		igb_setup_mac_rss_classify(igb);
2001 		break;
2002 	}
2003 
2004 	/*
2005 	 * Enable the receive unit - must be done after all
2006 	 * the rx setup above.
2007 	 */
2008 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2009 
2010 	/*
2011 	 * Initialize all adapter ring head & tail pointers - must
2012 	 * be done after receive unit is enabled
2013 	 */
2014 	for (i = 0; i < igb->num_rx_rings; i++) {
2015 		rx_ring = &igb->rx_rings[i];
2016 		E1000_WRITE_REG(hw, E1000_RDH(i), 0);
2017 		E1000_WRITE_REG(hw, E1000_RDT(i), rx_ring->ring_size - 1);
2018 	}
2019 
2020 	/*
2021 	 * 82575 with manageability enabled needs a special flush to make
2022 	 * sure the fifos start clean.
2023 	 */
2024 	if ((hw->mac.type == e1000_82575) &&
2025 	    (E1000_READ_REG(hw, E1000_MANC) & E1000_MANC_RCV_TCO_EN)) {
2026 		e1000_rx_fifo_flush_82575(hw);
2027 	}
2028 }
2029 
2030 static void
2031 igb_setup_tx_ring(igb_tx_ring_t *tx_ring)
2032 {
2033 	igb_t *igb = tx_ring->igb;
2034 	struct e1000_hw *hw = &igb->hw;
2035 	uint32_t size;
2036 	uint32_t buf_low;
2037 	uint32_t buf_high;
2038 	uint32_t reg_val;
2039 
2040 	ASSERT(mutex_owned(&tx_ring->tx_lock));
2041 	ASSERT(mutex_owned(&igb->gen_lock));
2042 
2043 
2044 	/*
2045 	 * Initialize the length register
2046 	 */
2047 	size = tx_ring->ring_size * sizeof (union e1000_adv_tx_desc);
2048 	E1000_WRITE_REG(hw, E1000_TDLEN(tx_ring->index), size);
2049 
2050 	/*
2051 	 * Initialize the base address registers
2052 	 */
2053 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
2054 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
2055 	E1000_WRITE_REG(hw, E1000_TDBAL(tx_ring->index), buf_low);
2056 	E1000_WRITE_REG(hw, E1000_TDBAH(tx_ring->index), buf_high);
2057 
2058 	/*
2059 	 * Setup head & tail pointers
2060 	 */
2061 	E1000_WRITE_REG(hw, E1000_TDH(tx_ring->index), 0);
2062 	E1000_WRITE_REG(hw, E1000_TDT(tx_ring->index), 0);
2063 
2064 	/*
2065 	 * Setup head write-back
2066 	 */
2067 	if (igb->tx_head_wb_enable) {
2068 		/*
2069 		 * The memory of the head write-back is allocated using
2070 		 * the extra tbd beyond the tail of the tbd ring.
2071 		 */
2072 		tx_ring->tbd_head_wb = (uint32_t *)
2073 		    ((uintptr_t)tx_ring->tbd_area.address + size);
2074 		*tx_ring->tbd_head_wb = 0;
2075 
2076 		buf_low = (uint32_t)
2077 		    (tx_ring->tbd_area.dma_address + size);
2078 		buf_high = (uint32_t)
2079 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
2080 
2081 		/* Set the head write-back enable bit */
2082 		buf_low |= E1000_TX_HEAD_WB_ENABLE;
2083 
2084 		E1000_WRITE_REG(hw, E1000_TDWBAL(tx_ring->index), buf_low);
2085 		E1000_WRITE_REG(hw, E1000_TDWBAH(tx_ring->index), buf_high);
2086 
2087 		/*
2088 		 * Turn off relaxed ordering for head write back or it will
2089 		 * cause problems with the tx recycling
2090 		 */
2091 		reg_val = E1000_READ_REG(hw,
2092 		    E1000_DCA_TXCTRL(tx_ring->index));
2093 		reg_val &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
2094 		E1000_WRITE_REG(hw,
2095 		    E1000_DCA_TXCTRL(tx_ring->index), reg_val);
2096 	} else {
2097 		tx_ring->tbd_head_wb = NULL;
2098 	}
2099 
2100 	tx_ring->tbd_head = 0;
2101 	tx_ring->tbd_tail = 0;
2102 	tx_ring->tbd_free = tx_ring->ring_size;
2103 
2104 	/*
2105 	 * Note: for the case that the chipset is being reset, we should not
2106 	 * reset the values of tcb_head, tcb_tail. And considering there might
2107 	 * still be some packets kept in the pending_list, we should not assert
2108 	 * (tcb_free == free_list_size) here.
2109 	 */
2110 	if (igb->igb_state == IGB_UNKNOWN) {
2111 		tx_ring->tcb_head = 0;
2112 		tx_ring->tcb_tail = 0;
2113 		tx_ring->tcb_free = tx_ring->free_list_size;
2114 	}
2115 
2116 	/*
2117 	 * Enable TXDCTL per queue
2118 	 */
2119 	reg_val = E1000_READ_REG(hw, E1000_TXDCTL(tx_ring->index));
2120 	reg_val |= E1000_TXDCTL_QUEUE_ENABLE;
2121 	E1000_WRITE_REG(hw, E1000_TXDCTL(tx_ring->index), reg_val);
2122 
2123 	/*
2124 	 * Initialize hardware checksum offload settings
2125 	 */
2126 	bzero(&tx_ring->tx_context, sizeof (tx_context_t));
2127 }
2128 
2129 static void
2130 igb_setup_tx(igb_t *igb)
2131 {
2132 	igb_tx_ring_t *tx_ring;
2133 	struct e1000_hw *hw = &igb->hw;
2134 	uint32_t reg_val;
2135 	int i;
2136 
2137 	for (i = 0; i < igb->num_tx_rings; i++) {
2138 		tx_ring = &igb->tx_rings[i];
2139 		igb_setup_tx_ring(tx_ring);
2140 	}
2141 
2142 	/*
2143 	 * Setup the Transmit Control Register (TCTL)
2144 	 */
2145 	reg_val = E1000_READ_REG(hw, E1000_TCTL);
2146 	reg_val &= ~E1000_TCTL_CT;
2147 	reg_val |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2148 	    (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2149 
2150 	/* Enable transmits */
2151 	reg_val |= E1000_TCTL_EN;
2152 
2153 	E1000_WRITE_REG(hw, E1000_TCTL, reg_val);
2154 }
2155 
2156 /*
2157  * igb_setup_rss - Setup receive-side scaling feature
2158  */
2159 static void
2160 igb_setup_rss(igb_t *igb)
2161 {
2162 	struct e1000_hw *hw = &igb->hw;
2163 	uint32_t i, mrqc, rxcsum;
2164 	int shift = 0;
2165 	uint32_t random;
2166 	union e1000_reta {
2167 		uint32_t	dword;
2168 		uint8_t		bytes[4];
2169 	} reta;
2170 
2171 	/* Setup the Redirection Table */
2172 	if (hw->mac.type == e1000_82576) {
2173 		shift = 0;
2174 	} else if (hw->mac.type == e1000_82575) {
2175 		shift = 6;
2176 	}
2177 	for (i = 0; i < (32 * 4); i++) {
2178 		reta.bytes[i & 3] = (i % igb->num_rx_rings) << shift;
2179 		if ((i & 3) == 3) {
2180 			E1000_WRITE_REG(hw,
2181 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2182 		}
2183 	}
2184 
2185 	/* Fill out hash function seeds */
2186 	for (i = 0; i < 10; i++) {
2187 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2188 		    sizeof (uint32_t));
2189 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2190 	}
2191 
2192 	/* Setup the Multiple Receive Queue Control register */
2193 	mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2194 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2195 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2196 	    E1000_MRQC_RSS_FIELD_IPV6 |
2197 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2198 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2199 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2200 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2201 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2202 
2203 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2204 
2205 	/*
2206 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2207 	 *
2208 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2209 	 * checksum offloading provided by the 82575 chipset besides the IP
2210 	 * header checksum offloading and the TCP/UDP checksum offloading.
2211 	 * The Packet Checksum is by default computed over the entire packet
2212 	 * from the first byte of the DA through the last byte of the CRC,
2213 	 * including the Ethernet and IP headers.
2214 	 *
2215 	 * It is a hardware limitation that Packet Checksum is mutually
2216 	 * exclusive with RSS.
2217 	 */
2218 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2219 	rxcsum |= E1000_RXCSUM_PCSD;
2220 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2221 }
2222 
2223 /*
2224  * igb_setup_mac_rss_classify - Setup MAC classification and rss
2225  */
2226 static void
2227 igb_setup_mac_rss_classify(igb_t *igb)
2228 {
2229 	struct e1000_hw *hw = &igb->hw;
2230 	uint32_t i, mrqc, vmdctl, rxcsum;
2231 	uint32_t ring_per_group;
2232 	int shift_group0, shift_group1;
2233 	uint32_t random;
2234 	union e1000_reta {
2235 		uint32_t	dword;
2236 		uint8_t		bytes[4];
2237 	} reta;
2238 
2239 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2240 
2241 	/* Setup the Redirection Table, it is shared between two groups */
2242 	shift_group0 = 2;
2243 	shift_group1 = 6;
2244 	for (i = 0; i < (32 * 4); i++) {
2245 		reta.bytes[i & 3] = ((i % ring_per_group) << shift_group0) |
2246 		    ((ring_per_group + (i % ring_per_group)) << shift_group1);
2247 		if ((i & 3) == 3) {
2248 			E1000_WRITE_REG(hw,
2249 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2250 		}
2251 	}
2252 
2253 	/* Fill out hash function seeds */
2254 	for (i = 0; i < 10; i++) {
2255 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2256 		    sizeof (uint32_t));
2257 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2258 	}
2259 
2260 	/*
2261 	 * Setup the Multiple Receive Queue Control register,
2262 	 * enable VMDq based on packet destination MAC address and RSS.
2263 	 */
2264 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_RSS_GROUP;
2265 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2266 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2267 	    E1000_MRQC_RSS_FIELD_IPV6 |
2268 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2269 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2270 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2271 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2272 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2273 
2274 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2275 
2276 
2277 	/* Define the default group and default queues */
2278 	vmdctl = E1000_VMDQ_MAC_GROUP_DEFAULT_QUEUE;
2279 	E1000_WRITE_REG(hw, E1000_VT_CTL, vmdctl);
2280 
2281 	/*
2282 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2283 	 *
2284 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2285 	 * checksum offloading provided by the 82575 chipset besides the IP
2286 	 * header checksum offloading and the TCP/UDP checksum offloading.
2287 	 * The Packet Checksum is by default computed over the entire packet
2288 	 * from the first byte of the DA through the last byte of the CRC,
2289 	 * including the Ethernet and IP headers.
2290 	 *
2291 	 * It is a hardware limitation that Packet Checksum is mutually
2292 	 * exclusive with RSS.
2293 	 */
2294 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2295 	rxcsum |= E1000_RXCSUM_PCSD;
2296 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2297 }
2298 
2299 /*
2300  * igb_setup_mac_classify - Setup MAC classification feature
2301  */
2302 static void
2303 igb_setup_mac_classify(igb_t *igb)
2304 {
2305 	struct e1000_hw *hw = &igb->hw;
2306 	uint32_t mrqc, rxcsum;
2307 
2308 	/*
2309 	 * Setup the Multiple Receive Queue Control register,
2310 	 * enable VMDq based on packet destination MAC address.
2311 	 */
2312 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_GROUP;
2313 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2314 
2315 	/*
2316 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2317 	 *
2318 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2319 	 * checksum offloading provided by the 82575 chipset besides the IP
2320 	 * header checksum offloading and the TCP/UDP checksum offloading.
2321 	 * The Packet Checksum is by default computed over the entire packet
2322 	 * from the first byte of the DA through the last byte of the CRC,
2323 	 * including the Ethernet and IP headers.
2324 	 *
2325 	 * It is a hardware limitation that Packet Checksum is mutually
2326 	 * exclusive with RSS.
2327 	 */
2328 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2329 	rxcsum |= E1000_RXCSUM_PCSD;
2330 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2331 
2332 }
2333 
2334 /*
2335  * igb_init_unicst - Initialize the unicast addresses
2336  */
2337 static void
2338 igb_init_unicst(igb_t *igb)
2339 {
2340 	struct e1000_hw *hw = &igb->hw;
2341 	int slot;
2342 
2343 	/*
2344 	 * Here we should consider two situations:
2345 	 *
2346 	 * 1. Chipset is initialized the first time
2347 	 *    Initialize the multiple unicast addresses, and
2348 	 *    save the default MAC address.
2349 	 *
2350 	 * 2. Chipset is reset
2351 	 *    Recover the multiple unicast addresses from the
2352 	 *    software data structure to the RAR registers.
2353 	 */
2354 
2355 	/*
2356 	 * Clear the default MAC address in the RAR0 rgister,
2357 	 * which is loaded from EEPROM when system boot or chipreset,
2358 	 * this will cause the conficts with add_mac/rem_mac entry
2359 	 * points when VMDq is enabled. For this reason, the RAR0
2360 	 * must be cleared for both cases mentioned above.
2361 	 */
2362 	e1000_rar_clear(hw, 0);
2363 
2364 	if (!igb->unicst_init) {
2365 
2366 		/* Initialize the multiple unicast addresses */
2367 		igb->unicst_total = MAX_NUM_UNICAST_ADDRESSES;
2368 		igb->unicst_avail = igb->unicst_total;
2369 
2370 		for (slot = 0; slot < igb->unicst_total; slot++)
2371 			igb->unicst_addr[slot].mac.set = 0;
2372 
2373 		igb->unicst_init = B_TRUE;
2374 	} else {
2375 		/* Re-configure the RAR registers */
2376 		for (slot = 0; slot < igb->unicst_total; slot++) {
2377 			e1000_rar_set_vmdq(hw, igb->unicst_addr[slot].mac.addr,
2378 			    slot, igb->vmdq_mode,
2379 			    igb->unicst_addr[slot].mac.group_index);
2380 		}
2381 	}
2382 
2383 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2384 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2385 }
2386 
2387 /*
2388  * igb_unicst_find - Find the slot for the specified unicast address
2389  */
2390 int
2391 igb_unicst_find(igb_t *igb, const uint8_t *mac_addr)
2392 {
2393 	int slot;
2394 
2395 	ASSERT(mutex_owned(&igb->gen_lock));
2396 
2397 	for (slot = 0; slot < igb->unicst_total; slot++) {
2398 		if (bcmp(igb->unicst_addr[slot].mac.addr,
2399 		    mac_addr, ETHERADDRL) == 0)
2400 			return (slot);
2401 	}
2402 
2403 	return (-1);
2404 }
2405 
2406 /*
2407  * igb_unicst_set - Set the unicast address to the specified slot
2408  */
2409 int
2410 igb_unicst_set(igb_t *igb, const uint8_t *mac_addr,
2411     int slot)
2412 {
2413 	struct e1000_hw *hw = &igb->hw;
2414 
2415 	ASSERT(mutex_owned(&igb->gen_lock));
2416 
2417 	/*
2418 	 * Save the unicast address in the software data structure
2419 	 */
2420 	bcopy(mac_addr, igb->unicst_addr[slot].mac.addr, ETHERADDRL);
2421 
2422 	/*
2423 	 * Set the unicast address to the RAR register
2424 	 */
2425 	e1000_rar_set(hw, (uint8_t *)mac_addr, slot);
2426 
2427 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2428 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2429 		return (EIO);
2430 	}
2431 
2432 	return (0);
2433 }
2434 
2435 /*
2436  * igb_multicst_add - Add a multicst address
2437  */
2438 int
2439 igb_multicst_add(igb_t *igb, const uint8_t *multiaddr)
2440 {
2441 	ASSERT(mutex_owned(&igb->gen_lock));
2442 
2443 	if ((multiaddr[0] & 01) == 0) {
2444 		return (EINVAL);
2445 	}
2446 
2447 	if (igb->mcast_count >= MAX_NUM_MULTICAST_ADDRESSES) {
2448 		return (ENOENT);
2449 	}
2450 
2451 	bcopy(multiaddr,
2452 	    &igb->mcast_table[igb->mcast_count], ETHERADDRL);
2453 	igb->mcast_count++;
2454 
2455 	/*
2456 	 * Update the multicast table in the hardware
2457 	 */
2458 	igb_setup_multicst(igb);
2459 
2460 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2461 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2462 		return (EIO);
2463 	}
2464 
2465 	return (0);
2466 }
2467 
2468 /*
2469  * igb_multicst_remove - Remove a multicst address
2470  */
2471 int
2472 igb_multicst_remove(igb_t *igb, const uint8_t *multiaddr)
2473 {
2474 	int i;
2475 
2476 	ASSERT(mutex_owned(&igb->gen_lock));
2477 
2478 	for (i = 0; i < igb->mcast_count; i++) {
2479 		if (bcmp(multiaddr, &igb->mcast_table[i],
2480 		    ETHERADDRL) == 0) {
2481 			for (i++; i < igb->mcast_count; i++) {
2482 				igb->mcast_table[i - 1] =
2483 				    igb->mcast_table[i];
2484 			}
2485 			igb->mcast_count--;
2486 			break;
2487 		}
2488 	}
2489 
2490 	/*
2491 	 * Update the multicast table in the hardware
2492 	 */
2493 	igb_setup_multicst(igb);
2494 
2495 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2496 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2497 		return (EIO);
2498 	}
2499 
2500 	return (0);
2501 }
2502 
2503 /*
2504  * igb_setup_multicast - setup multicast data structures
2505  *
2506  * This routine initializes all of the multicast related structures
2507  * and save them in the hardware registers.
2508  */
2509 static void
2510 igb_setup_multicst(igb_t *igb)
2511 {
2512 	uint8_t *mc_addr_list;
2513 	uint32_t mc_addr_count;
2514 	struct e1000_hw *hw = &igb->hw;
2515 
2516 	ASSERT(mutex_owned(&igb->gen_lock));
2517 
2518 	ASSERT(igb->mcast_count <= MAX_NUM_MULTICAST_ADDRESSES);
2519 
2520 	mc_addr_list = (uint8_t *)igb->mcast_table;
2521 	mc_addr_count = igb->mcast_count;
2522 
2523 	/*
2524 	 * Update the multicase addresses to the MTA registers
2525 	 */
2526 	e1000_update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2527 	    igb->unicst_total, hw->mac.rar_entry_count);
2528 }
2529 
2530 /*
2531  * igb_get_conf - Get driver configurations set in driver.conf
2532  *
2533  * This routine gets user-configured values out of the configuration
2534  * file igb.conf.
2535  *
2536  * For each configurable value, there is a minimum, a maximum, and a
2537  * default.
2538  * If user does not configure a value, use the default.
2539  * If user configures below the minimum, use the minumum.
2540  * If user configures above the maximum, use the maxumum.
2541  */
2542 static void
2543 igb_get_conf(igb_t *igb)
2544 {
2545 	struct e1000_hw *hw = &igb->hw;
2546 	uint32_t default_mtu;
2547 	uint32_t flow_control;
2548 	uint32_t ring_per_group;
2549 	int i;
2550 
2551 	/*
2552 	 * igb driver supports the following user configurations:
2553 	 *
2554 	 * Link configurations:
2555 	 *    adv_autoneg_cap
2556 	 *    adv_1000fdx_cap
2557 	 *    adv_100fdx_cap
2558 	 *    adv_100hdx_cap
2559 	 *    adv_10fdx_cap
2560 	 *    adv_10hdx_cap
2561 	 * Note: 1000hdx is not supported.
2562 	 *
2563 	 * Jumbo frame configuration:
2564 	 *    default_mtu
2565 	 *
2566 	 * Ethernet flow control configuration:
2567 	 *    flow_control
2568 	 *
2569 	 * Multiple rings configurations:
2570 	 *    tx_queue_number
2571 	 *    tx_ring_size
2572 	 *    rx_queue_number
2573 	 *    rx_ring_size
2574 	 *
2575 	 * Call igb_get_prop() to get the value for a specific
2576 	 * configuration parameter.
2577 	 */
2578 
2579 	/*
2580 	 * Link configurations
2581 	 */
2582 	igb->param_adv_autoneg_cap = igb_get_prop(igb,
2583 	    PROP_ADV_AUTONEG_CAP, 0, 1, 1);
2584 	igb->param_adv_1000fdx_cap = igb_get_prop(igb,
2585 	    PROP_ADV_1000FDX_CAP, 0, 1, 1);
2586 	igb->param_adv_100fdx_cap = igb_get_prop(igb,
2587 	    PROP_ADV_100FDX_CAP, 0, 1, 1);
2588 	igb->param_adv_100hdx_cap = igb_get_prop(igb,
2589 	    PROP_ADV_100HDX_CAP, 0, 1, 1);
2590 	igb->param_adv_10fdx_cap = igb_get_prop(igb,
2591 	    PROP_ADV_10FDX_CAP, 0, 1, 1);
2592 	igb->param_adv_10hdx_cap = igb_get_prop(igb,
2593 	    PROP_ADV_10HDX_CAP, 0, 1, 1);
2594 
2595 	/*
2596 	 * Jumbo frame configurations
2597 	 */
2598 	default_mtu = igb_get_prop(igb, PROP_DEFAULT_MTU,
2599 	    MIN_MTU, MAX_MTU, DEFAULT_MTU);
2600 
2601 	igb->max_frame_size = default_mtu +
2602 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
2603 
2604 	/*
2605 	 * Ethernet flow control configuration
2606 	 */
2607 	flow_control = igb_get_prop(igb, PROP_FLOW_CONTROL,
2608 	    e1000_fc_none, 4, e1000_fc_full);
2609 	if (flow_control == 4)
2610 		flow_control = e1000_fc_default;
2611 
2612 	hw->fc.requested_mode = flow_control;
2613 
2614 	/*
2615 	 * Multiple rings configurations
2616 	 */
2617 	igb->tx_ring_size = igb_get_prop(igb, PROP_TX_RING_SIZE,
2618 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
2619 	igb->rx_ring_size = igb_get_prop(igb, PROP_RX_RING_SIZE,
2620 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
2621 
2622 	igb->mr_enable = igb_get_prop(igb, PROP_MR_ENABLE, 0, 1, 1);
2623 	igb->num_rx_groups = igb_get_prop(igb, PROP_RX_GROUP_NUM,
2624 	    MIN_RX_GROUP_NUM, MAX_RX_GROUP_NUM, DEFAULT_RX_GROUP_NUM);
2625 	/*
2626 	 * Currently we do not support VMDq for 82576.
2627 	 * If it is e1000_82576, set num_rx_groups to 1.
2628 	 */
2629 	if (hw->mac.type == e1000_82576)
2630 		igb->num_rx_groups = 1;
2631 
2632 	if (igb->mr_enable) {
2633 		igb->num_tx_rings = igb->capab->def_tx_que_num;
2634 		igb->num_rx_rings = igb->capab->def_rx_que_num;
2635 	} else {
2636 		igb->num_tx_rings = 1;
2637 		igb->num_rx_rings = 1;
2638 
2639 		if (igb->num_rx_groups > 1) {
2640 			igb_error(igb,
2641 			    "Invalid rx groups number. Please enable multiple "
2642 			    "rings first");
2643 			igb->num_rx_groups = 1;
2644 		}
2645 	}
2646 
2647 	/*
2648 	 * Check the divisibility between rx rings and rx groups.
2649 	 */
2650 	for (i = igb->num_rx_groups; i > 0; i--) {
2651 		if ((igb->num_rx_rings % i) == 0)
2652 			break;
2653 	}
2654 	if (i != igb->num_rx_groups) {
2655 		igb_error(igb,
2656 		    "Invalid rx groups number. Downgrade the rx group "
2657 		    "number to %d.", i);
2658 		igb->num_rx_groups = i;
2659 	}
2660 
2661 	/*
2662 	 * Get the ring number per group.
2663 	 */
2664 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2665 
2666 	if (igb->num_rx_groups == 1) {
2667 		/*
2668 		 * One rx ring group, the rx ring number is num_rx_rings.
2669 		 */
2670 		igb->vmdq_mode = E1000_VMDQ_OFF;
2671 	} else if (ring_per_group == 1) {
2672 		/*
2673 		 * Multiple rx groups, each group has one rx ring.
2674 		 */
2675 		igb->vmdq_mode = E1000_VMDQ_MAC;
2676 	} else {
2677 		/*
2678 		 * Multiple groups and multiple rings.
2679 		 */
2680 		igb->vmdq_mode = E1000_VMDQ_MAC_RSS;
2681 	}
2682 
2683 	/*
2684 	 * Tunable used to force an interrupt type. The only use is
2685 	 * for testing of the lesser interrupt types.
2686 	 * 0 = don't force interrupt type
2687 	 * 1 = force interrupt type MSIX
2688 	 * 2 = force interrupt type MSI
2689 	 * 3 = force interrupt type Legacy
2690 	 */
2691 	igb->intr_force = igb_get_prop(igb, PROP_INTR_FORCE,
2692 	    IGB_INTR_NONE, IGB_INTR_LEGACY, IGB_INTR_NONE);
2693 
2694 	igb->tx_hcksum_enable = igb_get_prop(igb, PROP_TX_HCKSUM_ENABLE,
2695 	    0, 1, 1);
2696 	igb->rx_hcksum_enable = igb_get_prop(igb, PROP_RX_HCKSUM_ENABLE,
2697 	    0, 1, 1);
2698 	igb->lso_enable = igb_get_prop(igb, PROP_LSO_ENABLE,
2699 	    0, 1, 1);
2700 	igb->tx_head_wb_enable = igb_get_prop(igb, PROP_TX_HEAD_WB_ENABLE,
2701 	    0, 1, 1);
2702 
2703 	/*
2704 	 * igb LSO needs the tx h/w checksum support.
2705 	 * Here LSO will be disabled if tx h/w checksum has been disabled.
2706 	 */
2707 	if (igb->tx_hcksum_enable == B_FALSE)
2708 		igb->lso_enable = B_FALSE;
2709 
2710 	igb->tx_copy_thresh = igb_get_prop(igb, PROP_TX_COPY_THRESHOLD,
2711 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
2712 	    DEFAULT_TX_COPY_THRESHOLD);
2713 	igb->tx_recycle_thresh = igb_get_prop(igb, PROP_TX_RECYCLE_THRESHOLD,
2714 	    MIN_TX_RECYCLE_THRESHOLD, MAX_TX_RECYCLE_THRESHOLD,
2715 	    DEFAULT_TX_RECYCLE_THRESHOLD);
2716 	igb->tx_overload_thresh = igb_get_prop(igb, PROP_TX_OVERLOAD_THRESHOLD,
2717 	    MIN_TX_OVERLOAD_THRESHOLD, MAX_TX_OVERLOAD_THRESHOLD,
2718 	    DEFAULT_TX_OVERLOAD_THRESHOLD);
2719 	igb->tx_resched_thresh = igb_get_prop(igb, PROP_TX_RESCHED_THRESHOLD,
2720 	    MIN_TX_RESCHED_THRESHOLD, MAX_TX_RESCHED_THRESHOLD,
2721 	    DEFAULT_TX_RESCHED_THRESHOLD);
2722 
2723 	igb->rx_copy_thresh = igb_get_prop(igb, PROP_RX_COPY_THRESHOLD,
2724 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
2725 	    DEFAULT_RX_COPY_THRESHOLD);
2726 	igb->rx_limit_per_intr = igb_get_prop(igb, PROP_RX_LIMIT_PER_INTR,
2727 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
2728 	    DEFAULT_RX_LIMIT_PER_INTR);
2729 
2730 	igb->intr_throttling[0] = igb_get_prop(igb, PROP_INTR_THROTTLING,
2731 	    igb->capab->min_intr_throttle,
2732 	    igb->capab->max_intr_throttle,
2733 	    igb->capab->def_intr_throttle);
2734 }
2735 
2736 /*
2737  * igb_get_prop - Get a property value out of the configuration file igb.conf
2738  *
2739  * Caller provides the name of the property, a default value, a minimum
2740  * value, and a maximum value.
2741  *
2742  * Return configured value of the property, with default, minimum and
2743  * maximum properly applied.
2744  */
2745 static int
2746 igb_get_prop(igb_t *igb,
2747     char *propname,	/* name of the property */
2748     int minval,		/* minimum acceptable value */
2749     int maxval,		/* maximim acceptable value */
2750     int defval)		/* default value */
2751 {
2752 	int value;
2753 
2754 	/*
2755 	 * Call ddi_prop_get_int() to read the conf settings
2756 	 */
2757 	value = ddi_prop_get_int(DDI_DEV_T_ANY, igb->dip,
2758 	    DDI_PROP_DONTPASS, propname, defval);
2759 
2760 	if (value > maxval)
2761 		value = maxval;
2762 
2763 	if (value < minval)
2764 		value = minval;
2765 
2766 	return (value);
2767 }
2768 
2769 /*
2770  * igb_setup_link - Using the link properties to setup the link
2771  */
2772 int
2773 igb_setup_link(igb_t *igb, boolean_t setup_hw)
2774 {
2775 	struct e1000_mac_info *mac;
2776 	struct e1000_phy_info *phy;
2777 	boolean_t invalid;
2778 
2779 	mac = &igb->hw.mac;
2780 	phy = &igb->hw.phy;
2781 	invalid = B_FALSE;
2782 
2783 	if (igb->param_adv_autoneg_cap == 1) {
2784 		mac->autoneg = B_TRUE;
2785 		phy->autoneg_advertised = 0;
2786 
2787 		/*
2788 		 * 1000hdx is not supported for autonegotiation
2789 		 */
2790 		if (igb->param_adv_1000fdx_cap == 1)
2791 			phy->autoneg_advertised |= ADVERTISE_1000_FULL;
2792 
2793 		if (igb->param_adv_100fdx_cap == 1)
2794 			phy->autoneg_advertised |= ADVERTISE_100_FULL;
2795 
2796 		if (igb->param_adv_100hdx_cap == 1)
2797 			phy->autoneg_advertised |= ADVERTISE_100_HALF;
2798 
2799 		if (igb->param_adv_10fdx_cap == 1)
2800 			phy->autoneg_advertised |= ADVERTISE_10_FULL;
2801 
2802 		if (igb->param_adv_10hdx_cap == 1)
2803 			phy->autoneg_advertised |= ADVERTISE_10_HALF;
2804 
2805 		if (phy->autoneg_advertised == 0)
2806 			invalid = B_TRUE;
2807 	} else {
2808 		mac->autoneg = B_FALSE;
2809 
2810 		/*
2811 		 * 1000fdx and 1000hdx are not supported for forced link
2812 		 */
2813 		if (igb->param_adv_100fdx_cap == 1)
2814 			mac->forced_speed_duplex = ADVERTISE_100_FULL;
2815 		else if (igb->param_adv_100hdx_cap == 1)
2816 			mac->forced_speed_duplex = ADVERTISE_100_HALF;
2817 		else if (igb->param_adv_10fdx_cap == 1)
2818 			mac->forced_speed_duplex = ADVERTISE_10_FULL;
2819 		else if (igb->param_adv_10hdx_cap == 1)
2820 			mac->forced_speed_duplex = ADVERTISE_10_HALF;
2821 		else
2822 			invalid = B_TRUE;
2823 	}
2824 
2825 	if (invalid) {
2826 		igb_notice(igb, "Invalid link settings. Setup link to "
2827 		    "autonegotiation with full link capabilities.");
2828 		mac->autoneg = B_TRUE;
2829 		phy->autoneg_advertised = ADVERTISE_1000_FULL |
2830 		    ADVERTISE_100_FULL | ADVERTISE_100_HALF |
2831 		    ADVERTISE_10_FULL | ADVERTISE_10_HALF;
2832 	}
2833 
2834 	if (setup_hw) {
2835 		if (e1000_setup_link(&igb->hw) != E1000_SUCCESS)
2836 			return (IGB_FAILURE);
2837 	}
2838 
2839 	return (IGB_SUCCESS);
2840 }
2841 
2842 
2843 /*
2844  * igb_is_link_up - Check if the link is up
2845  */
2846 static boolean_t
2847 igb_is_link_up(igb_t *igb)
2848 {
2849 	struct e1000_hw *hw = &igb->hw;
2850 	boolean_t link_up = B_FALSE;
2851 
2852 	ASSERT(mutex_owned(&igb->gen_lock));
2853 
2854 	/*
2855 	 * get_link_status is set in the interrupt handler on link-status-change
2856 	 * or rx sequence error interrupt.  get_link_status will stay
2857 	 * false until the e1000_check_for_link establishes link only
2858 	 * for copper adapters.
2859 	 */
2860 	switch (hw->phy.media_type) {
2861 	case e1000_media_type_copper:
2862 		if (hw->mac.get_link_status) {
2863 			(void) e1000_check_for_link(hw);
2864 			link_up = !hw->mac.get_link_status;
2865 		} else {
2866 			link_up = B_TRUE;
2867 		}
2868 		break;
2869 	case e1000_media_type_fiber:
2870 		(void) e1000_check_for_link(hw);
2871 		link_up = (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU);
2872 		break;
2873 	case e1000_media_type_internal_serdes:
2874 		(void) e1000_check_for_link(hw);
2875 		link_up = hw->mac.serdes_has_link;
2876 		break;
2877 	}
2878 
2879 	return (link_up);
2880 }
2881 
2882 /*
2883  * igb_link_check - Link status processing
2884  */
2885 static boolean_t
2886 igb_link_check(igb_t *igb)
2887 {
2888 	struct e1000_hw *hw = &igb->hw;
2889 	uint16_t speed = 0, duplex = 0;
2890 	boolean_t link_changed = B_FALSE;
2891 
2892 	ASSERT(mutex_owned(&igb->gen_lock));
2893 
2894 	if (igb_is_link_up(igb)) {
2895 		/*
2896 		 * The Link is up, check whether it was marked as down earlier
2897 		 */
2898 		if (igb->link_state != LINK_STATE_UP) {
2899 			(void) e1000_get_speed_and_duplex(hw, &speed, &duplex);
2900 			igb->link_speed = speed;
2901 			igb->link_duplex = duplex;
2902 			igb->link_state = LINK_STATE_UP;
2903 			igb->link_down_timeout = 0;
2904 			link_changed = B_TRUE;
2905 		}
2906 	} else {
2907 		if (igb->link_state != LINK_STATE_DOWN) {
2908 			igb->link_speed = 0;
2909 			igb->link_duplex = 0;
2910 			igb->link_state = LINK_STATE_DOWN;
2911 			link_changed = B_TRUE;
2912 		}
2913 
2914 		if (igb->igb_state & IGB_STARTED) {
2915 			if (igb->link_down_timeout < MAX_LINK_DOWN_TIMEOUT) {
2916 				igb->link_down_timeout++;
2917 			} else if (igb->link_down_timeout ==
2918 			    MAX_LINK_DOWN_TIMEOUT) {
2919 				igb_tx_clean(igb);
2920 				igb->link_down_timeout++;
2921 			}
2922 		}
2923 	}
2924 
2925 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2926 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2927 
2928 	return (link_changed);
2929 }
2930 
2931 /*
2932  * igb_local_timer - driver watchdog function
2933  *
2934  * This function will handle the transmit stall check, link status check and
2935  * other routines.
2936  */
2937 static void
2938 igb_local_timer(void *arg)
2939 {
2940 	igb_t *igb = (igb_t *)arg;
2941 	boolean_t link_changed = B_FALSE;
2942 
2943 	if (igb_stall_check(igb)) {
2944 		igb_fm_ereport(igb, DDI_FM_DEVICE_STALL);
2945 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
2946 		igb->reset_count++;
2947 		if (igb_reset(igb) == IGB_SUCCESS)
2948 			ddi_fm_service_impact(igb->dip,
2949 			    DDI_SERVICE_RESTORED);
2950 	}
2951 
2952 	mutex_enter(&igb->gen_lock);
2953 	if (!(igb->igb_state & IGB_SUSPENDED) && (igb->igb_state & IGB_STARTED))
2954 		link_changed = igb_link_check(igb);
2955 	mutex_exit(&igb->gen_lock);
2956 
2957 	if (link_changed)
2958 		mac_link_update(igb->mac_hdl, igb->link_state);
2959 
2960 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2961 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2962 
2963 	igb_restart_watchdog_timer(igb);
2964 }
2965 
2966 /*
2967  * igb_stall_check - check for transmit stall
2968  *
2969  * This function checks if the adapter is stalled (in transmit).
2970  *
2971  * It is called each time the watchdog timeout is invoked.
2972  * If the transmit descriptor reclaim continuously fails,
2973  * the watchdog value will increment by 1. If the watchdog
2974  * value exceeds the threshold, the igb is assumed to
2975  * have stalled and need to be reset.
2976  */
2977 static boolean_t
2978 igb_stall_check(igb_t *igb)
2979 {
2980 	igb_tx_ring_t *tx_ring;
2981 	boolean_t result;
2982 	int i;
2983 
2984 	if (igb->link_state != LINK_STATE_UP)
2985 		return (B_FALSE);
2986 
2987 	/*
2988 	 * If any tx ring is stalled, we'll reset the chipset
2989 	 */
2990 	result = B_FALSE;
2991 	for (i = 0; i < igb->num_tx_rings; i++) {
2992 		tx_ring = &igb->tx_rings[i];
2993 
2994 		if (tx_ring->recycle_fail > 0)
2995 			tx_ring->stall_watchdog++;
2996 		else
2997 			tx_ring->stall_watchdog = 0;
2998 
2999 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
3000 			result = B_TRUE;
3001 			break;
3002 		}
3003 	}
3004 
3005 	if (result) {
3006 		tx_ring->stall_watchdog = 0;
3007 		tx_ring->recycle_fail = 0;
3008 	}
3009 
3010 	return (result);
3011 }
3012 
3013 
3014 /*
3015  * is_valid_mac_addr - Check if the mac address is valid
3016  */
3017 static boolean_t
3018 is_valid_mac_addr(uint8_t *mac_addr)
3019 {
3020 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
3021 	const uint8_t addr_test2[6] =
3022 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
3023 
3024 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
3025 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
3026 		return (B_FALSE);
3027 
3028 	return (B_TRUE);
3029 }
3030 
3031 static boolean_t
3032 igb_find_mac_address(igb_t *igb)
3033 {
3034 	struct e1000_hw *hw = &igb->hw;
3035 #ifdef __sparc
3036 	uchar_t *bytes;
3037 	struct ether_addr sysaddr;
3038 	uint_t nelts;
3039 	int err;
3040 	boolean_t found = B_FALSE;
3041 
3042 	/*
3043 	 * The "vendor's factory-set address" may already have
3044 	 * been extracted from the chip, but if the property
3045 	 * "local-mac-address" is set we use that instead.
3046 	 *
3047 	 * We check whether it looks like an array of 6
3048 	 * bytes (which it should, if OBP set it).  If we can't
3049 	 * make sense of it this way, we'll ignore it.
3050 	 */
3051 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3052 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
3053 	if (err == DDI_PROP_SUCCESS) {
3054 		if (nelts == ETHERADDRL) {
3055 			while (nelts--)
3056 				hw->mac.addr[nelts] = bytes[nelts];
3057 			found = B_TRUE;
3058 		}
3059 		ddi_prop_free(bytes);
3060 	}
3061 
3062 	/*
3063 	 * Look up the OBP property "local-mac-address?". If the user has set
3064 	 * 'local-mac-address? = false', use "the system address" instead.
3065 	 */
3066 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip, 0,
3067 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
3068 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
3069 			if (localetheraddr(NULL, &sysaddr) != 0) {
3070 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
3071 				found = B_TRUE;
3072 			}
3073 		}
3074 		ddi_prop_free(bytes);
3075 	}
3076 
3077 	/*
3078 	 * Finally(!), if there's a valid "mac-address" property (created
3079 	 * if we netbooted from this interface), we must use this instead
3080 	 * of any of the above to ensure that the NFS/install server doesn't
3081 	 * get confused by the address changing as Solaris takes over!
3082 	 */
3083 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3084 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
3085 	if (err == DDI_PROP_SUCCESS) {
3086 		if (nelts == ETHERADDRL) {
3087 			while (nelts--)
3088 				hw->mac.addr[nelts] = bytes[nelts];
3089 			found = B_TRUE;
3090 		}
3091 		ddi_prop_free(bytes);
3092 	}
3093 
3094 	if (found) {
3095 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
3096 		return (B_TRUE);
3097 	}
3098 #endif
3099 
3100 	/*
3101 	 * Read the device MAC address from the EEPROM
3102 	 */
3103 	if (e1000_read_mac_addr(hw) != E1000_SUCCESS)
3104 		return (B_FALSE);
3105 
3106 	return (B_TRUE);
3107 }
3108 
3109 #pragma inline(igb_arm_watchdog_timer)
3110 
3111 static void
3112 igb_arm_watchdog_timer(igb_t *igb)
3113 {
3114 	/*
3115 	 * Fire a watchdog timer
3116 	 */
3117 	igb->watchdog_tid =
3118 	    timeout(igb_local_timer,
3119 	    (void *)igb, 1 * drv_usectohz(1000000));
3120 
3121 }
3122 
3123 /*
3124  * igb_enable_watchdog_timer - Enable and start the driver watchdog timer
3125  */
3126 void
3127 igb_enable_watchdog_timer(igb_t *igb)
3128 {
3129 	mutex_enter(&igb->watchdog_lock);
3130 
3131 	if (!igb->watchdog_enable) {
3132 		igb->watchdog_enable = B_TRUE;
3133 		igb->watchdog_start = B_TRUE;
3134 		igb_arm_watchdog_timer(igb);
3135 	}
3136 
3137 	mutex_exit(&igb->watchdog_lock);
3138 
3139 }
3140 
3141 /*
3142  * igb_disable_watchdog_timer - Disable and stop the driver watchdog timer
3143  */
3144 void
3145 igb_disable_watchdog_timer(igb_t *igb)
3146 {
3147 	timeout_id_t tid;
3148 
3149 	mutex_enter(&igb->watchdog_lock);
3150 
3151 	igb->watchdog_enable = B_FALSE;
3152 	igb->watchdog_start = B_FALSE;
3153 	tid = igb->watchdog_tid;
3154 	igb->watchdog_tid = 0;
3155 
3156 	mutex_exit(&igb->watchdog_lock);
3157 
3158 	if (tid != 0)
3159 		(void) untimeout(tid);
3160 
3161 }
3162 
3163 /*
3164  * igb_start_watchdog_timer - Start the driver watchdog timer
3165  */
3166 static void
3167 igb_start_watchdog_timer(igb_t *igb)
3168 {
3169 	mutex_enter(&igb->watchdog_lock);
3170 
3171 	if (igb->watchdog_enable) {
3172 		if (!igb->watchdog_start) {
3173 			igb->watchdog_start = B_TRUE;
3174 			igb_arm_watchdog_timer(igb);
3175 		}
3176 	}
3177 
3178 	mutex_exit(&igb->watchdog_lock);
3179 }
3180 
3181 /*
3182  * igb_restart_watchdog_timer - Restart the driver watchdog timer
3183  */
3184 static void
3185 igb_restart_watchdog_timer(igb_t *igb)
3186 {
3187 	mutex_enter(&igb->watchdog_lock);
3188 
3189 	if (igb->watchdog_start)
3190 		igb_arm_watchdog_timer(igb);
3191 
3192 	mutex_exit(&igb->watchdog_lock);
3193 }
3194 
3195 /*
3196  * igb_stop_watchdog_timer - Stop the driver watchdog timer
3197  */
3198 static void
3199 igb_stop_watchdog_timer(igb_t *igb)
3200 {
3201 	timeout_id_t tid;
3202 
3203 	mutex_enter(&igb->watchdog_lock);
3204 
3205 	igb->watchdog_start = B_FALSE;
3206 	tid = igb->watchdog_tid;
3207 	igb->watchdog_tid = 0;
3208 
3209 	mutex_exit(&igb->watchdog_lock);
3210 
3211 	if (tid != 0)
3212 		(void) untimeout(tid);
3213 }
3214 
3215 /*
3216  * igb_disable_adapter_interrupts - Clear/disable all hardware interrupts
3217  */
3218 static void
3219 igb_disable_adapter_interrupts(igb_t *igb)
3220 {
3221 	struct e1000_hw *hw = &igb->hw;
3222 
3223 	/*
3224 	 * Set the IMC register to mask all the interrupts,
3225 	 * including the tx interrupts.
3226 	 */
3227 	E1000_WRITE_REG(hw, E1000_IMC, ~0);
3228 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3229 
3230 	/*
3231 	 * Additional disabling for MSI-X
3232 	 */
3233 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3234 		E1000_WRITE_REG(hw, E1000_EIMC, ~0);
3235 		E1000_WRITE_REG(hw, E1000_EIAC, 0);
3236 		E1000_WRITE_REG(hw, E1000_EIAM, 0);
3237 	}
3238 
3239 	E1000_WRITE_FLUSH(hw);
3240 }
3241 
3242 /*
3243  * igb_enable_adapter_interrupts_82576 - Enable NIC interrupts for 82576
3244  */
3245 static void
3246 igb_enable_adapter_interrupts_82576(igb_t *igb)
3247 {
3248 	struct e1000_hw *hw = &igb->hw;
3249 
3250 	/* Clear any pending interrupts */
3251 	(void) E1000_READ_REG(hw, E1000_ICR);
3252 
3253 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3254 
3255 		/* Interrupt enabling for MSI-X */
3256 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3257 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3258 		igb->ims_mask = E1000_IMS_LSC;
3259 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3260 	} else {
3261 		/* Interrupt enabling for MSI and legacy */
3262 		E1000_WRITE_REG(hw, E1000_IVAR0, E1000_IVAR_VALID);
3263 		igb->ims_mask = IMS_ENABLE_MASK | E1000_IMS_TXQE;
3264 		E1000_WRITE_REG(hw, E1000_IMS,
3265 		    (IMS_ENABLE_MASK | E1000_IMS_TXQE));
3266 	}
3267 
3268 	/* Disable auto-mask for ICR interrupt bits */
3269 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3270 
3271 	E1000_WRITE_FLUSH(hw);
3272 }
3273 
3274 /*
3275  * igb_enable_adapter_interrupts_82575 - Enable NIC interrupts for 82575
3276  */
3277 static void
3278 igb_enable_adapter_interrupts_82575(igb_t *igb)
3279 {
3280 	struct e1000_hw *hw = &igb->hw;
3281 	uint32_t reg;
3282 
3283 	/* Clear any pending interrupts */
3284 	(void) E1000_READ_REG(hw, E1000_ICR);
3285 
3286 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3287 		/* Interrupt enabling for MSI-X */
3288 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3289 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3290 		igb->ims_mask = E1000_IMS_LSC;
3291 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3292 
3293 		/* Enable MSI-X PBA support */
3294 		reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
3295 		reg |= E1000_CTRL_EXT_PBA_CLR;
3296 
3297 		/* Non-selective interrupt clear-on-read */
3298 		reg |= E1000_CTRL_EXT_IRCA;	/* Called NSICR in the EAS */
3299 
3300 		E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
3301 	} else {
3302 		/* Interrupt enabling for MSI and legacy */
3303 		igb->ims_mask = IMS_ENABLE_MASK;
3304 		E1000_WRITE_REG(hw, E1000_IMS, IMS_ENABLE_MASK);
3305 	}
3306 
3307 	E1000_WRITE_FLUSH(hw);
3308 }
3309 
3310 /*
3311  * Loopback Support
3312  */
3313 static lb_property_t lb_normal =
3314 	{ normal,	"normal",	IGB_LB_NONE		};
3315 static lb_property_t lb_external =
3316 	{ external,	"External",	IGB_LB_EXTERNAL		};
3317 static lb_property_t lb_mac =
3318 	{ internal,	"MAC",		IGB_LB_INTERNAL_MAC	};
3319 static lb_property_t lb_phy =
3320 	{ internal,	"PHY",		IGB_LB_INTERNAL_PHY	};
3321 static lb_property_t lb_serdes =
3322 	{ internal,	"SerDes",	IGB_LB_INTERNAL_SERDES	};
3323 
3324 enum ioc_reply
3325 igb_loopback_ioctl(igb_t *igb, struct iocblk *iocp, mblk_t *mp)
3326 {
3327 	lb_info_sz_t *lbsp;
3328 	lb_property_t *lbpp;
3329 	struct e1000_hw *hw;
3330 	uint32_t *lbmp;
3331 	uint32_t size;
3332 	uint32_t value;
3333 
3334 	hw = &igb->hw;
3335 
3336 	if (mp->b_cont == NULL)
3337 		return (IOC_INVAL);
3338 
3339 	switch (iocp->ioc_cmd) {
3340 	default:
3341 		return (IOC_INVAL);
3342 
3343 	case LB_GET_INFO_SIZE:
3344 		size = sizeof (lb_info_sz_t);
3345 		if (iocp->ioc_count != size)
3346 			return (IOC_INVAL);
3347 
3348 		value = sizeof (lb_normal);
3349 		value += sizeof (lb_mac);
3350 		if (hw->phy.media_type == e1000_media_type_copper)
3351 			value += sizeof (lb_phy);
3352 		else
3353 			value += sizeof (lb_serdes);
3354 		value += sizeof (lb_external);
3355 
3356 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
3357 		*lbsp = value;
3358 		break;
3359 
3360 	case LB_GET_INFO:
3361 		value = sizeof (lb_normal);
3362 		value += sizeof (lb_mac);
3363 		if (hw->phy.media_type == e1000_media_type_copper)
3364 			value += sizeof (lb_phy);
3365 		else
3366 			value += sizeof (lb_serdes);
3367 		value += sizeof (lb_external);
3368 
3369 		size = value;
3370 		if (iocp->ioc_count != size)
3371 			return (IOC_INVAL);
3372 
3373 		value = 0;
3374 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
3375 
3376 		lbpp[value++] = lb_normal;
3377 		lbpp[value++] = lb_mac;
3378 		if (hw->phy.media_type == e1000_media_type_copper)
3379 			lbpp[value++] = lb_phy;
3380 		else
3381 			lbpp[value++] = lb_serdes;
3382 		lbpp[value++] = lb_external;
3383 		break;
3384 
3385 	case LB_GET_MODE:
3386 		size = sizeof (uint32_t);
3387 		if (iocp->ioc_count != size)
3388 			return (IOC_INVAL);
3389 
3390 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3391 		*lbmp = igb->loopback_mode;
3392 		break;
3393 
3394 	case LB_SET_MODE:
3395 		size = 0;
3396 		if (iocp->ioc_count != sizeof (uint32_t))
3397 			return (IOC_INVAL);
3398 
3399 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3400 		if (!igb_set_loopback_mode(igb, *lbmp))
3401 			return (IOC_INVAL);
3402 		break;
3403 	}
3404 
3405 	iocp->ioc_count = size;
3406 	iocp->ioc_error = 0;
3407 
3408 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3409 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3410 		return (IOC_INVAL);
3411 	}
3412 
3413 	return (IOC_REPLY);
3414 }
3415 
3416 /*
3417  * igb_set_loopback_mode - Setup loopback based on the loopback mode
3418  */
3419 static boolean_t
3420 igb_set_loopback_mode(igb_t *igb, uint32_t mode)
3421 {
3422 	struct e1000_hw *hw;
3423 
3424 	if (mode == igb->loopback_mode)
3425 		return (B_TRUE);
3426 
3427 	hw = &igb->hw;
3428 
3429 	igb->loopback_mode = mode;
3430 
3431 	if (mode == IGB_LB_NONE) {
3432 		/* Reset the chip */
3433 		hw->phy.autoneg_wait_to_complete = B_TRUE;
3434 		(void) igb_reset(igb);
3435 		hw->phy.autoneg_wait_to_complete = B_FALSE;
3436 		return (B_TRUE);
3437 	}
3438 
3439 	mutex_enter(&igb->gen_lock);
3440 
3441 	switch (mode) {
3442 	default:
3443 		mutex_exit(&igb->gen_lock);
3444 		return (B_FALSE);
3445 
3446 	case IGB_LB_EXTERNAL:
3447 		igb_set_external_loopback(igb);
3448 		break;
3449 
3450 	case IGB_LB_INTERNAL_MAC:
3451 		igb_set_internal_mac_loopback(igb);
3452 		break;
3453 
3454 	case IGB_LB_INTERNAL_PHY:
3455 		igb_set_internal_phy_loopback(igb);
3456 		break;
3457 
3458 	case IGB_LB_INTERNAL_SERDES:
3459 		igb_set_internal_serdes_loopback(igb);
3460 		break;
3461 	}
3462 
3463 	mutex_exit(&igb->gen_lock);
3464 
3465 	return (B_TRUE);
3466 }
3467 
3468 /*
3469  * igb_set_external_loopback - Set the external loopback mode
3470  */
3471 static void
3472 igb_set_external_loopback(igb_t *igb)
3473 {
3474 	struct e1000_hw *hw;
3475 
3476 	hw = &igb->hw;
3477 
3478 	/* Set phy to known state */
3479 	(void) e1000_phy_hw_reset(hw);
3480 
3481 	(void) e1000_write_phy_reg(hw, 0x0, 0x0140);
3482 	(void) e1000_write_phy_reg(hw, 0x9, 0x1b00);
3483 	(void) e1000_write_phy_reg(hw, 0x12, 0x1610);
3484 	(void) e1000_write_phy_reg(hw, 0x1f37, 0x3f1c);
3485 }
3486 
3487 /*
3488  * igb_set_internal_mac_loopback - Set the internal MAC loopback mode
3489  */
3490 static void
3491 igb_set_internal_mac_loopback(igb_t *igb)
3492 {
3493 	struct e1000_hw *hw;
3494 	uint32_t ctrl;
3495 	uint32_t rctl;
3496 	uint32_t ctrl_ext;
3497 	uint16_t phy_ctrl;
3498 	uint16_t phy_status;
3499 
3500 	hw = &igb->hw;
3501 
3502 	(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
3503 	phy_ctrl &= ~MII_CR_AUTO_NEG_EN;
3504 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3505 
3506 	(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
3507 
3508 	/* Set link mode to PHY (00b) in the Extended Control register */
3509 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3510 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3511 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3512 
3513 	/* Set the Device Control register */
3514 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3515 	if (!(phy_status & MII_SR_LINK_STATUS))
3516 		ctrl |= E1000_CTRL_ILOS; /* Set ILOS when the link is down */
3517 	ctrl &= ~E1000_CTRL_SPD_SEL;	/* Clear the speed sel bits */
3518 	ctrl |= (E1000_CTRL_SLU |	/* Force link up */
3519 	    E1000_CTRL_FRCSPD |		/* Force speed */
3520 	    E1000_CTRL_FRCDPX |		/* Force duplex */
3521 	    E1000_CTRL_SPD_1000 |	/* Force speed to 1000 */
3522 	    E1000_CTRL_FD);		/* Force full duplex */
3523 
3524 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3525 
3526 	/* Set the Receive Control register */
3527 	rctl = E1000_READ_REG(hw, E1000_RCTL);
3528 	rctl &= ~E1000_RCTL_LBM_TCVR;
3529 	rctl |= E1000_RCTL_LBM_MAC;
3530 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
3531 }
3532 
3533 /*
3534  * igb_set_internal_phy_loopback - Set the internal PHY loopback mode
3535  */
3536 static void
3537 igb_set_internal_phy_loopback(igb_t *igb)
3538 {
3539 	struct e1000_hw *hw;
3540 	uint32_t ctrl_ext;
3541 	uint16_t phy_ctrl;
3542 	uint16_t phy_pconf;
3543 
3544 	hw = &igb->hw;
3545 
3546 	/* Set link mode to PHY (00b) in the Extended Control register */
3547 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3548 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3549 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3550 
3551 	/*
3552 	 * Set PHY control register (0x4140):
3553 	 *    Set full duplex mode
3554 	 *    Set loopback bit
3555 	 *    Clear auto-neg enable bit
3556 	 *    Set PHY speed
3557 	 */
3558 	phy_ctrl = MII_CR_FULL_DUPLEX | MII_CR_SPEED_1000 | MII_CR_LOOPBACK;
3559 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3560 
3561 	/* Set the link disable bit in the Port Configuration register */
3562 	(void) e1000_read_phy_reg(hw, 0x10, &phy_pconf);
3563 	phy_pconf |= (uint16_t)1 << 14;
3564 	(void) e1000_write_phy_reg(hw, 0x10, phy_pconf);
3565 }
3566 
3567 /*
3568  * igb_set_internal_serdes_loopback - Set the internal SerDes loopback mode
3569  */
3570 static void
3571 igb_set_internal_serdes_loopback(igb_t *igb)
3572 {
3573 	struct e1000_hw *hw;
3574 	uint32_t ctrl_ext;
3575 	uint32_t ctrl;
3576 	uint32_t pcs_lctl;
3577 	uint32_t connsw;
3578 
3579 	hw = &igb->hw;
3580 
3581 	/* Set link mode to SerDes (11b) in the Extended Control register */
3582 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3583 	ctrl_ext |= E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
3584 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3585 
3586 	/* Configure the SerDes to loopback */
3587 	E1000_WRITE_REG(hw, E1000_SCTL, 0x410);
3588 
3589 	/* Set Device Control register */
3590 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3591 	ctrl |= (E1000_CTRL_FD |	/* Force full duplex */
3592 	    E1000_CTRL_SLU);		/* Force link up */
3593 	ctrl &= ~(E1000_CTRL_RFCE |	/* Disable receive flow control */
3594 	    E1000_CTRL_TFCE |		/* Disable transmit flow control */
3595 	    E1000_CTRL_LRST);		/* Clear link reset */
3596 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3597 
3598 	/* Set PCS Link Control register */
3599 	pcs_lctl = E1000_READ_REG(hw, E1000_PCS_LCTL);
3600 	pcs_lctl |= (E1000_PCS_LCTL_FORCE_LINK |
3601 	    E1000_PCS_LCTL_FSD |
3602 	    E1000_PCS_LCTL_FDV_FULL |
3603 	    E1000_PCS_LCTL_FLV_LINK_UP);
3604 	pcs_lctl &= ~E1000_PCS_LCTL_AN_ENABLE;
3605 	E1000_WRITE_REG(hw, E1000_PCS_LCTL, pcs_lctl);
3606 
3607 	/* Set the Copper/Fiber Switch Control - CONNSW register */
3608 	connsw = E1000_READ_REG(hw, E1000_CONNSW);
3609 	connsw &= ~E1000_CONNSW_ENRGSRC;
3610 	E1000_WRITE_REG(hw, E1000_CONNSW, connsw);
3611 }
3612 
3613 #pragma inline(igb_intr_rx_work)
3614 /*
3615  * igb_intr_rx_work - rx processing of ISR
3616  */
3617 static void
3618 igb_intr_rx_work(igb_rx_ring_t *rx_ring)
3619 {
3620 	mblk_t *mp;
3621 
3622 	mutex_enter(&rx_ring->rx_lock);
3623 	mp = igb_rx(rx_ring, IGB_NO_POLL);
3624 	mutex_exit(&rx_ring->rx_lock);
3625 
3626 	if (mp != NULL)
3627 		mac_rx_ring(rx_ring->igb->mac_hdl, rx_ring->ring_handle, mp,
3628 		    rx_ring->ring_gen_num);
3629 }
3630 
3631 #pragma inline(igb_intr_tx_work)
3632 /*
3633  * igb_intr_tx_work - tx processing of ISR
3634  */
3635 static void
3636 igb_intr_tx_work(igb_tx_ring_t *tx_ring)
3637 {
3638 	/* Recycle the tx descriptors */
3639 	tx_ring->tx_recycle(tx_ring);
3640 
3641 	/* Schedule the re-transmit */
3642 	if (tx_ring->reschedule &&
3643 	    (tx_ring->tbd_free >= tx_ring->resched_thresh)) {
3644 		tx_ring->reschedule = B_FALSE;
3645 		mac_tx_ring_update(tx_ring->igb->mac_hdl, tx_ring->ring_handle);
3646 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3647 	}
3648 }
3649 
3650 #pragma inline(igb_intr_link_work)
3651 /*
3652  * igb_intr_link_work - link-status-change processing of ISR
3653  */
3654 static void
3655 igb_intr_link_work(igb_t *igb)
3656 {
3657 	boolean_t link_changed;
3658 
3659 	igb_stop_watchdog_timer(igb);
3660 
3661 	mutex_enter(&igb->gen_lock);
3662 
3663 	/*
3664 	 * Because we got a link-status-change interrupt, force
3665 	 * e1000_check_for_link() to look at phy
3666 	 */
3667 	igb->hw.mac.get_link_status = B_TRUE;
3668 
3669 	/* igb_link_check takes care of link status change */
3670 	link_changed = igb_link_check(igb);
3671 
3672 	/* Get new phy state */
3673 	igb_get_phy_state(igb);
3674 
3675 	mutex_exit(&igb->gen_lock);
3676 
3677 	if (link_changed)
3678 		mac_link_update(igb->mac_hdl, igb->link_state);
3679 
3680 	igb_start_watchdog_timer(igb);
3681 }
3682 
3683 /*
3684  * igb_intr_legacy - Interrupt handler for legacy interrupts
3685  */
3686 static uint_t
3687 igb_intr_legacy(void *arg1, void *arg2)
3688 {
3689 	igb_t *igb = (igb_t *)arg1;
3690 	igb_tx_ring_t *tx_ring;
3691 	uint32_t icr;
3692 	mblk_t *mp;
3693 	boolean_t tx_reschedule;
3694 	boolean_t link_changed;
3695 	uint_t result;
3696 
3697 	_NOTE(ARGUNUSED(arg2));
3698 
3699 	mutex_enter(&igb->gen_lock);
3700 
3701 	if (igb->igb_state & IGB_SUSPENDED) {
3702 		mutex_exit(&igb->gen_lock);
3703 		return (DDI_INTR_UNCLAIMED);
3704 	}
3705 
3706 	mp = NULL;
3707 	tx_reschedule = B_FALSE;
3708 	link_changed = B_FALSE;
3709 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3710 
3711 	if (icr & E1000_ICR_INT_ASSERTED) {
3712 		/*
3713 		 * E1000_ICR_INT_ASSERTED bit was set:
3714 		 * Read(Clear) the ICR, claim this interrupt,
3715 		 * look for work to do.
3716 		 */
3717 		ASSERT(igb->num_rx_rings == 1);
3718 		ASSERT(igb->num_tx_rings == 1);
3719 
3720 		/* Make sure all interrupt causes cleared */
3721 		(void) E1000_READ_REG(&igb->hw, E1000_EICR);
3722 
3723 		if (icr & E1000_ICR_RXT0) {
3724 			mp = igb_rx(&igb->rx_rings[0], IGB_NO_POLL);
3725 		}
3726 
3727 		if (icr & E1000_ICR_TXDW) {
3728 			tx_ring = &igb->tx_rings[0];
3729 
3730 			/* Recycle the tx descriptors */
3731 			tx_ring->tx_recycle(tx_ring);
3732 
3733 			/* Schedule the re-transmit */
3734 			tx_reschedule = (tx_ring->reschedule &&
3735 			    (tx_ring->tbd_free >= tx_ring->resched_thresh));
3736 		}
3737 
3738 		if (icr & E1000_ICR_LSC) {
3739 			/*
3740 			 * Because we got a link-status-change interrupt, force
3741 			 * e1000_check_for_link() to look at phy
3742 			 */
3743 			igb->hw.mac.get_link_status = B_TRUE;
3744 
3745 			/* igb_link_check takes care of link status change */
3746 			link_changed = igb_link_check(igb);
3747 
3748 			/* Get new phy state */
3749 			igb_get_phy_state(igb);
3750 		}
3751 
3752 		result = DDI_INTR_CLAIMED;
3753 	} else {
3754 		/*
3755 		 * E1000_ICR_INT_ASSERTED bit was not set:
3756 		 * Don't claim this interrupt.
3757 		 */
3758 		result = DDI_INTR_UNCLAIMED;
3759 	}
3760 
3761 	mutex_exit(&igb->gen_lock);
3762 
3763 	/*
3764 	 * Do the following work outside of the gen_lock
3765 	 */
3766 	if (mp != NULL)
3767 		mac_rx(igb->mac_hdl, NULL, mp);
3768 
3769 	if (tx_reschedule)  {
3770 		tx_ring->reschedule = B_FALSE;
3771 		mac_tx_ring_update(igb->mac_hdl, tx_ring->ring_handle);
3772 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3773 	}
3774 
3775 	if (link_changed)
3776 		mac_link_update(igb->mac_hdl, igb->link_state);
3777 
3778 	return (result);
3779 }
3780 
3781 /*
3782  * igb_intr_msi - Interrupt handler for MSI
3783  */
3784 static uint_t
3785 igb_intr_msi(void *arg1, void *arg2)
3786 {
3787 	igb_t *igb = (igb_t *)arg1;
3788 	uint32_t icr;
3789 
3790 	_NOTE(ARGUNUSED(arg2));
3791 
3792 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3793 
3794 	/* Make sure all interrupt causes cleared */
3795 	(void) E1000_READ_REG(&igb->hw, E1000_EICR);
3796 
3797 	/*
3798 	 * For MSI interrupt, we have only one vector,
3799 	 * so we have only one rx ring and one tx ring enabled.
3800 	 */
3801 	ASSERT(igb->num_rx_rings == 1);
3802 	ASSERT(igb->num_tx_rings == 1);
3803 
3804 	if (icr & E1000_ICR_RXT0) {
3805 		igb_intr_rx_work(&igb->rx_rings[0]);
3806 	}
3807 
3808 	if (icr & E1000_ICR_TXDW) {
3809 		igb_intr_tx_work(&igb->tx_rings[0]);
3810 	}
3811 
3812 	if (icr & E1000_ICR_LSC) {
3813 		igb_intr_link_work(igb);
3814 	}
3815 
3816 	return (DDI_INTR_CLAIMED);
3817 }
3818 
3819 /*
3820  * igb_intr_rx - Interrupt handler for rx
3821  */
3822 static uint_t
3823 igb_intr_rx(void *arg1, void *arg2)
3824 {
3825 	igb_rx_ring_t *rx_ring = (igb_rx_ring_t *)arg1;
3826 
3827 	_NOTE(ARGUNUSED(arg2));
3828 
3829 	/*
3830 	 * Only used via MSI-X vector so don't check cause bits
3831 	 * and only clean the given ring.
3832 	 */
3833 	igb_intr_rx_work(rx_ring);
3834 
3835 	return (DDI_INTR_CLAIMED);
3836 }
3837 
3838 /*
3839  * igb_intr_tx - Interrupt handler for tx
3840  */
3841 static uint_t
3842 igb_intr_tx(void *arg1, void *arg2)
3843 {
3844 	igb_tx_ring_t *tx_ring = (igb_tx_ring_t *)arg1;
3845 
3846 	_NOTE(ARGUNUSED(arg2));
3847 
3848 	/*
3849 	 * Only used via MSI-X vector so don't check cause bits
3850 	 * and only clean the given ring.
3851 	 */
3852 	igb_intr_tx_work(tx_ring);
3853 
3854 	return (DDI_INTR_CLAIMED);
3855 }
3856 
3857 /*
3858  * igb_intr_tx_other - Interrupt handler for both tx and other
3859  *
3860  */
3861 static uint_t
3862 igb_intr_tx_other(void *arg1, void *arg2)
3863 {
3864 	igb_t *igb = (igb_t *)arg1;
3865 	uint32_t icr;
3866 
3867 	_NOTE(ARGUNUSED(arg2));
3868 
3869 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3870 
3871 	/*
3872 	 * Look for tx reclaiming work first. Remember, in the
3873 	 * case of only interrupt sharing, only one tx ring is
3874 	 * used
3875 	 */
3876 	igb_intr_tx_work(&igb->tx_rings[0]);
3877 
3878 	/*
3879 	 * Check for "other" causes.
3880 	 */
3881 	if (icr & E1000_ICR_LSC) {
3882 		igb_intr_link_work(igb);
3883 	}
3884 
3885 	/*
3886 	 * The DOUTSYNC bit indicates a tx packet dropped because
3887 	 * DMA engine gets "out of sync". There isn't a real fix
3888 	 * for this. The Intel recommendation is to count the number
3889 	 * of occurrences so user can detect when it is happening.
3890 	 * The issue is non-fatal and there's no recovery action
3891 	 * available.
3892 	 */
3893 	if (icr & E1000_ICR_DOUTSYNC) {
3894 		IGB_STAT(igb->dout_sync);
3895 	}
3896 
3897 	return (DDI_INTR_CLAIMED);
3898 }
3899 
3900 /*
3901  * igb_alloc_intrs - Allocate interrupts for the driver
3902  *
3903  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
3904  * if not successful, try Legacy.
3905  * igb->intr_force can be used to force sequence to start with
3906  * any of the 3 types.
3907  * If MSI-X is not used, number of tx/rx rings is forced to 1.
3908  */
3909 static int
3910 igb_alloc_intrs(igb_t *igb)
3911 {
3912 	dev_info_t *devinfo;
3913 	int intr_types;
3914 	int rc;
3915 
3916 	devinfo = igb->dip;
3917 
3918 	/* Get supported interrupt types */
3919 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
3920 
3921 	if (rc != DDI_SUCCESS) {
3922 		igb_log(igb,
3923 		    "Get supported interrupt types failed: %d", rc);
3924 		return (IGB_FAILURE);
3925 	}
3926 	IGB_DEBUGLOG_1(igb, "Supported interrupt types: %x", intr_types);
3927 
3928 	igb->intr_type = 0;
3929 
3930 	/* Install MSI-X interrupts */
3931 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
3932 	    (igb->intr_force <= IGB_INTR_MSIX)) {
3933 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSIX);
3934 
3935 		if (rc == IGB_SUCCESS)
3936 			return (IGB_SUCCESS);
3937 
3938 		igb_log(igb,
3939 		    "Allocate MSI-X failed, trying MSI interrupts...");
3940 	}
3941 
3942 	/* MSI-X not used, force rings to 1 */
3943 	igb->num_rx_rings = 1;
3944 	igb->num_tx_rings = 1;
3945 	igb_log(igb,
3946 	    "MSI-X not used, force rx and tx queue number to 1");
3947 
3948 	/* Install MSI interrupts */
3949 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
3950 	    (igb->intr_force <= IGB_INTR_MSI)) {
3951 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSI);
3952 
3953 		if (rc == IGB_SUCCESS)
3954 			return (IGB_SUCCESS);
3955 
3956 		igb_log(igb,
3957 		    "Allocate MSI failed, trying Legacy interrupts...");
3958 	}
3959 
3960 	/* Install legacy interrupts */
3961 	if (intr_types & DDI_INTR_TYPE_FIXED) {
3962 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_FIXED);
3963 
3964 		if (rc == IGB_SUCCESS)
3965 			return (IGB_SUCCESS);
3966 
3967 		igb_log(igb,
3968 		    "Allocate Legacy interrupts failed");
3969 	}
3970 
3971 	/* If none of the 3 types succeeded, return failure */
3972 	return (IGB_FAILURE);
3973 }
3974 
3975 /*
3976  * igb_alloc_intr_handles - Allocate interrupt handles.
3977  *
3978  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
3979  * if fewer than 2 handles are available, return failure.
3980  * Upon success, this sets the number of Rx rings to a number that
3981  * matches the handles available for Rx interrupts.
3982  */
3983 static int
3984 igb_alloc_intr_handles(igb_t *igb, int intr_type)
3985 {
3986 	dev_info_t *devinfo;
3987 	int orig, request, count, avail, actual;
3988 	int diff, minimum;
3989 	int rc;
3990 
3991 	devinfo = igb->dip;
3992 
3993 	switch (intr_type) {
3994 	case DDI_INTR_TYPE_FIXED:
3995 		request = 1;	/* Request 1 legacy interrupt handle */
3996 		minimum = 1;
3997 		IGB_DEBUGLOG_0(igb, "interrupt type: legacy");
3998 		break;
3999 
4000 	case DDI_INTR_TYPE_MSI:
4001 		request = 1;	/* Request 1 MSI interrupt handle */
4002 		minimum = 1;
4003 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI");
4004 		break;
4005 
4006 	case DDI_INTR_TYPE_MSIX:
4007 		/*
4008 		 * Number of vectors for the adapter is
4009 		 * # rx rings + # tx rings
4010 		 * One of tx vectors is for tx & other
4011 		 */
4012 		request = igb->num_rx_rings + igb->num_tx_rings;
4013 		orig = request;
4014 		minimum = 2;
4015 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI-X");
4016 		break;
4017 
4018 	default:
4019 		igb_log(igb,
4020 		    "invalid call to igb_alloc_intr_handles(): %d\n",
4021 		    intr_type);
4022 		return (IGB_FAILURE);
4023 	}
4024 	IGB_DEBUGLOG_2(igb, "interrupt handles requested: %d  minimum: %d",
4025 	    request, minimum);
4026 
4027 	/*
4028 	 * Get number of supported interrupts
4029 	 */
4030 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
4031 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
4032 		igb_log(igb,
4033 		    "Get supported interrupt number failed. "
4034 		    "Return: %d, count: %d", rc, count);
4035 		return (IGB_FAILURE);
4036 	}
4037 	IGB_DEBUGLOG_1(igb, "interrupts supported: %d", count);
4038 
4039 	/*
4040 	 * Get number of available interrupts
4041 	 */
4042 	rc = ddi_intr_get_navail(devinfo, intr_type, &avail);
4043 	if ((rc != DDI_SUCCESS) || (avail < minimum)) {
4044 		igb_log(igb,
4045 		    "Get available interrupt number failed. "
4046 		    "Return: %d, available: %d", rc, avail);
4047 		return (IGB_FAILURE);
4048 	}
4049 	IGB_DEBUGLOG_1(igb, "interrupts available: %d", avail);
4050 
4051 	if (avail < request) {
4052 		igb_log(igb, "Request %d handles, %d available",
4053 		    request, avail);
4054 		request = avail;
4055 	}
4056 
4057 	actual = 0;
4058 	igb->intr_cnt = 0;
4059 
4060 	/*
4061 	 * Allocate an array of interrupt handles
4062 	 */
4063 	igb->intr_size = request * sizeof (ddi_intr_handle_t);
4064 	igb->htable = kmem_alloc(igb->intr_size, KM_SLEEP);
4065 
4066 	rc = ddi_intr_alloc(devinfo, igb->htable, intr_type, 0,
4067 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
4068 	if (rc != DDI_SUCCESS) {
4069 		igb_log(igb, "Allocate interrupts failed. "
4070 		    "return: %d, request: %d, actual: %d",
4071 		    rc, request, actual);
4072 		goto alloc_handle_fail;
4073 	}
4074 	IGB_DEBUGLOG_1(igb, "interrupts actually allocated: %d", actual);
4075 
4076 	igb->intr_cnt = actual;
4077 
4078 	if (actual < minimum) {
4079 		igb_log(igb, "Insufficient interrupt handles allocated: %d",
4080 		    actual);
4081 		goto alloc_handle_fail;
4082 	}
4083 
4084 	/*
4085 	 * For MSI-X, actual might force us to reduce number of tx & rx rings
4086 	 */
4087 	if ((intr_type == DDI_INTR_TYPE_MSIX) && (orig > actual)) {
4088 		diff = orig - actual;
4089 		if (diff < igb->num_tx_rings) {
4090 			igb_log(igb,
4091 			    "MSI-X vectors force Tx queue number to %d",
4092 			    igb->num_tx_rings - diff);
4093 			igb->num_tx_rings -= diff;
4094 		} else {
4095 			igb_log(igb,
4096 			    "MSI-X vectors force Tx queue number to 1");
4097 			igb->num_tx_rings = 1;
4098 
4099 			igb_log(igb,
4100 			    "MSI-X vectors force Rx queue number to %d",
4101 			    actual - 1);
4102 			igb->num_rx_rings = actual - 1;
4103 		}
4104 	}
4105 
4106 	/*
4107 	 * Get priority for first vector, assume remaining are all the same
4108 	 */
4109 	rc = ddi_intr_get_pri(igb->htable[0], &igb->intr_pri);
4110 	if (rc != DDI_SUCCESS) {
4111 		igb_log(igb,
4112 		    "Get interrupt priority failed: %d", rc);
4113 		goto alloc_handle_fail;
4114 	}
4115 
4116 	rc = ddi_intr_get_cap(igb->htable[0], &igb->intr_cap);
4117 	if (rc != DDI_SUCCESS) {
4118 		igb_log(igb,
4119 		    "Get interrupt cap failed: %d", rc);
4120 		goto alloc_handle_fail;
4121 	}
4122 
4123 	igb->intr_type = intr_type;
4124 
4125 	return (IGB_SUCCESS);
4126 
4127 alloc_handle_fail:
4128 	igb_rem_intrs(igb);
4129 
4130 	return (IGB_FAILURE);
4131 }
4132 
4133 /*
4134  * igb_add_intr_handlers - Add interrupt handlers based on the interrupt type
4135  *
4136  * Before adding the interrupt handlers, the interrupt vectors have
4137  * been allocated, and the rx/tx rings have also been allocated.
4138  */
4139 static int
4140 igb_add_intr_handlers(igb_t *igb)
4141 {
4142 	igb_rx_ring_t *rx_ring;
4143 	igb_tx_ring_t *tx_ring;
4144 	int vector;
4145 	int rc;
4146 	int i;
4147 
4148 	vector = 0;
4149 
4150 	switch (igb->intr_type) {
4151 	case DDI_INTR_TYPE_MSIX:
4152 		/* Add interrupt handler for tx + other */
4153 		tx_ring = &igb->tx_rings[0];
4154 		rc = ddi_intr_add_handler(igb->htable[vector],
4155 		    (ddi_intr_handler_t *)igb_intr_tx_other,
4156 		    (void *)igb, NULL);
4157 
4158 		if (rc != DDI_SUCCESS) {
4159 			igb_log(igb,
4160 			    "Add tx/other interrupt handler failed: %d", rc);
4161 			return (IGB_FAILURE);
4162 		}
4163 		tx_ring->intr_vector = vector;
4164 		vector++;
4165 
4166 		/* Add interrupt handler for each rx ring */
4167 		for (i = 0; i < igb->num_rx_rings; i++) {
4168 			rx_ring = &igb->rx_rings[i];
4169 
4170 			rc = ddi_intr_add_handler(igb->htable[vector],
4171 			    (ddi_intr_handler_t *)igb_intr_rx,
4172 			    (void *)rx_ring, NULL);
4173 
4174 			if (rc != DDI_SUCCESS) {
4175 				igb_log(igb,
4176 				    "Add rx interrupt handler failed. "
4177 				    "return: %d, rx ring: %d", rc, i);
4178 				for (vector--; vector >= 0; vector--) {
4179 					(void) ddi_intr_remove_handler(
4180 					    igb->htable[vector]);
4181 				}
4182 				return (IGB_FAILURE);
4183 			}
4184 
4185 			rx_ring->intr_vector = vector;
4186 
4187 			vector++;
4188 		}
4189 
4190 		/* Add interrupt handler for each tx ring from 2nd ring */
4191 		for (i = 1; i < igb->num_tx_rings; i++) {
4192 			tx_ring = &igb->tx_rings[i];
4193 
4194 			rc = ddi_intr_add_handler(igb->htable[vector],
4195 			    (ddi_intr_handler_t *)igb_intr_tx,
4196 			    (void *)tx_ring, NULL);
4197 
4198 			if (rc != DDI_SUCCESS) {
4199 				igb_log(igb,
4200 				    "Add tx interrupt handler failed. "
4201 				    "return: %d, tx ring: %d", rc, i);
4202 				for (vector--; vector >= 0; vector--) {
4203 					(void) ddi_intr_remove_handler(
4204 					    igb->htable[vector]);
4205 				}
4206 				return (IGB_FAILURE);
4207 			}
4208 
4209 			tx_ring->intr_vector = vector;
4210 
4211 			vector++;
4212 		}
4213 
4214 		break;
4215 
4216 	case DDI_INTR_TYPE_MSI:
4217 		/* Add interrupt handlers for the only vector */
4218 		rc = ddi_intr_add_handler(igb->htable[vector],
4219 		    (ddi_intr_handler_t *)igb_intr_msi,
4220 		    (void *)igb, NULL);
4221 
4222 		if (rc != DDI_SUCCESS) {
4223 			igb_log(igb,
4224 			    "Add MSI interrupt handler failed: %d", rc);
4225 			return (IGB_FAILURE);
4226 		}
4227 
4228 		rx_ring = &igb->rx_rings[0];
4229 		rx_ring->intr_vector = vector;
4230 
4231 		vector++;
4232 		break;
4233 
4234 	case DDI_INTR_TYPE_FIXED:
4235 		/* Add interrupt handlers for the only vector */
4236 		rc = ddi_intr_add_handler(igb->htable[vector],
4237 		    (ddi_intr_handler_t *)igb_intr_legacy,
4238 		    (void *)igb, NULL);
4239 
4240 		if (rc != DDI_SUCCESS) {
4241 			igb_log(igb,
4242 			    "Add legacy interrupt handler failed: %d", rc);
4243 			return (IGB_FAILURE);
4244 		}
4245 
4246 		rx_ring = &igb->rx_rings[0];
4247 		rx_ring->intr_vector = vector;
4248 
4249 		vector++;
4250 		break;
4251 
4252 	default:
4253 		return (IGB_FAILURE);
4254 	}
4255 
4256 	ASSERT(vector == igb->intr_cnt);
4257 
4258 	return (IGB_SUCCESS);
4259 }
4260 
4261 /*
4262  * igb_setup_msix_82575 - setup 82575 adapter to use MSI-X interrupts
4263  *
4264  * For each vector enabled on the adapter, Set the MSIXBM register accordingly
4265  */
4266 static void
4267 igb_setup_msix_82575(igb_t *igb)
4268 {
4269 	uint32_t eims = 0;
4270 	int i, vector;
4271 	struct e1000_hw *hw = &igb->hw;
4272 
4273 	/*
4274 	 * Set vector for tx ring 0 and other causes.
4275 	 * NOTE assumption that it is vector 0.
4276 	 */
4277 	vector = 0;
4278 
4279 	igb->eims_mask = E1000_EICR_TX_QUEUE0 | E1000_EICR_OTHER;
4280 	E1000_WRITE_REG(hw, E1000_MSIXBM(vector), igb->eims_mask);
4281 	vector++;
4282 
4283 	for (i = 0; i < igb->num_rx_rings; i++) {
4284 		/*
4285 		 * Set vector for each rx ring
4286 		 */
4287 		eims = (E1000_EICR_RX_QUEUE0 << i);
4288 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4289 
4290 		/*
4291 		 * Accumulate bits to enable in
4292 		 * igb_enable_adapter_interrupts_82575()
4293 		 */
4294 		igb->eims_mask |= eims;
4295 
4296 		vector++;
4297 	}
4298 
4299 	for (i = 1; i < igb->num_tx_rings; i++) {
4300 		/*
4301 		 * Set vector for each tx ring from 2nd tx ring
4302 		 */
4303 		eims = (E1000_EICR_TX_QUEUE0 << i);
4304 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4305 
4306 		/*
4307 		 * Accumulate bits to enable in
4308 		 * igb_enable_adapter_interrupts_82575()
4309 		 */
4310 		igb->eims_mask |= eims;
4311 
4312 		vector++;
4313 	}
4314 
4315 	ASSERT(vector == igb->intr_cnt);
4316 
4317 	/*
4318 	 * Disable IAM for ICR interrupt bits
4319 	 */
4320 	E1000_WRITE_REG(hw, E1000_IAM, 0);
4321 	E1000_WRITE_FLUSH(hw);
4322 }
4323 
4324 /*
4325  * igb_setup_msix_82576 - setup 82576 adapter to use MSI-X interrupts
4326  *
4327  * 82576 uses a table based method for assigning vectors.  Each queue has a
4328  * single entry in the table to which we write a vector number along with a
4329  * "valid" bit.  The entry is a single byte in a 4-byte register.  Vectors
4330  * take a different position in the 4-byte register depending on whether
4331  * they are numbered above or below 8.
4332  */
4333 static void
4334 igb_setup_msix_82576(igb_t *igb)
4335 {
4336 	struct e1000_hw *hw = &igb->hw;
4337 	uint32_t ivar, index, vector;
4338 	int i;
4339 
4340 	/* must enable msi-x capability before IVAR settings */
4341 	E1000_WRITE_REG(hw, E1000_GPIE,
4342 	    (E1000_GPIE_MSIX_MODE | E1000_GPIE_PBA | E1000_GPIE_NSICR));
4343 
4344 	/*
4345 	 * Set vector for tx ring 0 and other causes.
4346 	 * NOTE assumption that it is vector 0.
4347 	 * This is also interdependent with installation of interrupt service
4348 	 * routines in igb_add_intr_handlers().
4349 	 */
4350 
4351 	/* assign "other" causes to vector 0 */
4352 	vector = 0;
4353 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4354 	E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4355 
4356 	/* assign tx ring 0 to vector 0 */
4357 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4358 	E1000_WRITE_REG(hw, E1000_IVAR0, ivar);
4359 
4360 	/* prepare to enable tx & other interrupt causes */
4361 	igb->eims_mask = (1 << vector);
4362 
4363 	vector ++;
4364 	for (i = 0; i < igb->num_rx_rings; i++) {
4365 		/*
4366 		 * Set vector for each rx ring
4367 		 */
4368 		index = (i & 0x7);
4369 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4370 
4371 		if (i < 8) {
4372 			/* vector goes into low byte of register */
4373 			ivar = ivar & 0xFFFFFF00;
4374 			ivar |= (vector | E1000_IVAR_VALID);
4375 		} else {
4376 			/* vector goes into third byte of register */
4377 			ivar = ivar & 0xFF00FFFF;
4378 			ivar |= ((vector | E1000_IVAR_VALID) << 16);
4379 		}
4380 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4381 
4382 		/* Accumulate interrupt-cause bits to enable */
4383 		igb->eims_mask |= (1 << vector);
4384 
4385 		vector ++;
4386 	}
4387 
4388 	for (i = 1; i < igb->num_tx_rings; i++) {
4389 		/*
4390 		 * Set vector for each tx ring from 2nd tx ring.
4391 		 * Note assumption that tx vectors numericall follow rx vectors.
4392 		 */
4393 		index = (i & 0x7);
4394 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4395 
4396 		if (i < 8) {
4397 			/* vector goes into second byte of register */
4398 			ivar = ivar & 0xFFFF00FF;
4399 			ivar |= ((vector | E1000_IVAR_VALID) << 8);
4400 		} else {
4401 			/* vector goes into fourth byte of register */
4402 			ivar = ivar & 0x00FFFFFF;
4403 			ivar |= (vector | E1000_IVAR_VALID) << 24;
4404 		}
4405 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4406 
4407 		/* Accumulate interrupt-cause bits to enable */
4408 		igb->eims_mask |= (1 << vector);
4409 
4410 		vector ++;
4411 	}
4412 
4413 	ASSERT(vector == igb->intr_cnt);
4414 }
4415 
4416 /*
4417  * igb_rem_intr_handlers - remove the interrupt handlers
4418  */
4419 static void
4420 igb_rem_intr_handlers(igb_t *igb)
4421 {
4422 	int i;
4423 	int rc;
4424 
4425 	for (i = 0; i < igb->intr_cnt; i++) {
4426 		rc = ddi_intr_remove_handler(igb->htable[i]);
4427 		if (rc != DDI_SUCCESS) {
4428 			IGB_DEBUGLOG_1(igb,
4429 			    "Remove intr handler failed: %d", rc);
4430 		}
4431 	}
4432 }
4433 
4434 /*
4435  * igb_rem_intrs - remove the allocated interrupts
4436  */
4437 static void
4438 igb_rem_intrs(igb_t *igb)
4439 {
4440 	int i;
4441 	int rc;
4442 
4443 	for (i = 0; i < igb->intr_cnt; i++) {
4444 		rc = ddi_intr_free(igb->htable[i]);
4445 		if (rc != DDI_SUCCESS) {
4446 			IGB_DEBUGLOG_1(igb,
4447 			    "Free intr failed: %d", rc);
4448 		}
4449 	}
4450 
4451 	kmem_free(igb->htable, igb->intr_size);
4452 	igb->htable = NULL;
4453 }
4454 
4455 /*
4456  * igb_enable_intrs - enable all the ddi interrupts
4457  */
4458 static int
4459 igb_enable_intrs(igb_t *igb)
4460 {
4461 	int i;
4462 	int rc;
4463 
4464 	/* Enable interrupts */
4465 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4466 		/* Call ddi_intr_block_enable() for MSI */
4467 		rc = ddi_intr_block_enable(igb->htable, igb->intr_cnt);
4468 		if (rc != DDI_SUCCESS) {
4469 			igb_log(igb,
4470 			    "Enable block intr failed: %d", rc);
4471 			return (IGB_FAILURE);
4472 		}
4473 	} else {
4474 		/* Call ddi_intr_enable() for Legacy/MSI non block enable */
4475 		for (i = 0; i < igb->intr_cnt; i++) {
4476 			rc = ddi_intr_enable(igb->htable[i]);
4477 			if (rc != DDI_SUCCESS) {
4478 				igb_log(igb,
4479 				    "Enable intr failed: %d", rc);
4480 				return (IGB_FAILURE);
4481 			}
4482 		}
4483 	}
4484 
4485 	return (IGB_SUCCESS);
4486 }
4487 
4488 /*
4489  * igb_disable_intrs - disable all the ddi interrupts
4490  */
4491 static int
4492 igb_disable_intrs(igb_t *igb)
4493 {
4494 	int i;
4495 	int rc;
4496 
4497 	/* Disable all interrupts */
4498 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4499 		rc = ddi_intr_block_disable(igb->htable, igb->intr_cnt);
4500 		if (rc != DDI_SUCCESS) {
4501 			igb_log(igb,
4502 			    "Disable block intr failed: %d", rc);
4503 			return (IGB_FAILURE);
4504 		}
4505 	} else {
4506 		for (i = 0; i < igb->intr_cnt; i++) {
4507 			rc = ddi_intr_disable(igb->htable[i]);
4508 			if (rc != DDI_SUCCESS) {
4509 				igb_log(igb,
4510 				    "Disable intr failed: %d", rc);
4511 				return (IGB_FAILURE);
4512 			}
4513 		}
4514 	}
4515 
4516 	return (IGB_SUCCESS);
4517 }
4518 
4519 /*
4520  * igb_get_phy_state - Get and save the parameters read from PHY registers
4521  */
4522 static void
4523 igb_get_phy_state(igb_t *igb)
4524 {
4525 	struct e1000_hw *hw = &igb->hw;
4526 	uint16_t phy_ctrl;
4527 	uint16_t phy_status;
4528 	uint16_t phy_an_adv;
4529 	uint16_t phy_an_exp;
4530 	uint16_t phy_ext_status;
4531 	uint16_t phy_1000t_ctrl;
4532 	uint16_t phy_1000t_status;
4533 	uint16_t phy_lp_able;
4534 
4535 	ASSERT(mutex_owned(&igb->gen_lock));
4536 
4537 	(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
4538 	(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
4539 	(void) e1000_read_phy_reg(hw, PHY_AUTONEG_ADV, &phy_an_adv);
4540 	(void) e1000_read_phy_reg(hw, PHY_AUTONEG_EXP, &phy_an_exp);
4541 	(void) e1000_read_phy_reg(hw, PHY_EXT_STATUS, &phy_ext_status);
4542 	(void) e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_1000t_ctrl);
4543 	(void) e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_1000t_status);
4544 	(void) e1000_read_phy_reg(hw, PHY_LP_ABILITY, &phy_lp_able);
4545 
4546 	igb->param_autoneg_cap =
4547 	    (phy_status & MII_SR_AUTONEG_CAPS) ? 1 : 0;
4548 	igb->param_pause_cap =
4549 	    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4550 	igb->param_asym_pause_cap =
4551 	    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4552 	igb->param_1000fdx_cap = ((phy_ext_status & IEEE_ESR_1000T_FD_CAPS) ||
4553 	    (phy_ext_status & IEEE_ESR_1000X_FD_CAPS)) ? 1 : 0;
4554 	igb->param_1000hdx_cap = ((phy_ext_status & IEEE_ESR_1000T_HD_CAPS) ||
4555 	    (phy_ext_status & IEEE_ESR_1000X_HD_CAPS)) ? 1 : 0;
4556 	igb->param_100t4_cap =
4557 	    (phy_status & MII_SR_100T4_CAPS) ? 1 : 0;
4558 	igb->param_100fdx_cap = ((phy_status & MII_SR_100X_FD_CAPS) ||
4559 	    (phy_status & MII_SR_100T2_FD_CAPS)) ? 1 : 0;
4560 	igb->param_100hdx_cap = ((phy_status & MII_SR_100X_HD_CAPS) ||
4561 	    (phy_status & MII_SR_100T2_HD_CAPS)) ? 1 : 0;
4562 	igb->param_10fdx_cap =
4563 	    (phy_status & MII_SR_10T_FD_CAPS) ? 1 : 0;
4564 	igb->param_10hdx_cap =
4565 	    (phy_status & MII_SR_10T_HD_CAPS) ? 1 : 0;
4566 	igb->param_rem_fault =
4567 	    (phy_status & MII_SR_REMOTE_FAULT) ? 1 : 0;
4568 
4569 	igb->param_adv_autoneg_cap = hw->mac.autoneg;
4570 	igb->param_adv_pause_cap =
4571 	    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4572 	igb->param_adv_asym_pause_cap =
4573 	    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4574 	igb->param_adv_1000hdx_cap =
4575 	    (phy_1000t_ctrl & CR_1000T_HD_CAPS) ? 1 : 0;
4576 	igb->param_adv_100t4_cap =
4577 	    (phy_an_adv & NWAY_AR_100T4_CAPS) ? 1 : 0;
4578 	igb->param_adv_rem_fault =
4579 	    (phy_an_adv & NWAY_AR_REMOTE_FAULT) ? 1 : 0;
4580 	if (igb->param_adv_autoneg_cap == 1) {
4581 		igb->param_adv_1000fdx_cap =
4582 		    (phy_1000t_ctrl & CR_1000T_FD_CAPS) ? 1 : 0;
4583 		igb->param_adv_100fdx_cap =
4584 		    (phy_an_adv & NWAY_AR_100TX_FD_CAPS) ? 1 : 0;
4585 		igb->param_adv_100hdx_cap =
4586 		    (phy_an_adv & NWAY_AR_100TX_HD_CAPS) ? 1 : 0;
4587 		igb->param_adv_10fdx_cap =
4588 		    (phy_an_adv & NWAY_AR_10T_FD_CAPS) ? 1 : 0;
4589 		igb->param_adv_10hdx_cap =
4590 		    (phy_an_adv & NWAY_AR_10T_HD_CAPS) ? 1 : 0;
4591 	}
4592 
4593 	igb->param_lp_autoneg_cap =
4594 	    (phy_an_exp & NWAY_ER_LP_NWAY_CAPS) ? 1 : 0;
4595 	igb->param_lp_pause_cap =
4596 	    (phy_lp_able & NWAY_LPAR_PAUSE) ? 1 : 0;
4597 	igb->param_lp_asym_pause_cap =
4598 	    (phy_lp_able & NWAY_LPAR_ASM_DIR) ? 1 : 0;
4599 	igb->param_lp_1000fdx_cap =
4600 	    (phy_1000t_status & SR_1000T_LP_FD_CAPS) ? 1 : 0;
4601 	igb->param_lp_1000hdx_cap =
4602 	    (phy_1000t_status & SR_1000T_LP_HD_CAPS) ? 1 : 0;
4603 	igb->param_lp_100t4_cap =
4604 	    (phy_lp_able & NWAY_LPAR_100T4_CAPS) ? 1 : 0;
4605 	igb->param_lp_100fdx_cap =
4606 	    (phy_lp_able & NWAY_LPAR_100TX_FD_CAPS) ? 1 : 0;
4607 	igb->param_lp_100hdx_cap =
4608 	    (phy_lp_able & NWAY_LPAR_100TX_HD_CAPS) ? 1 : 0;
4609 	igb->param_lp_10fdx_cap =
4610 	    (phy_lp_able & NWAY_LPAR_10T_FD_CAPS) ? 1 : 0;
4611 	igb->param_lp_10hdx_cap =
4612 	    (phy_lp_able & NWAY_LPAR_10T_HD_CAPS) ? 1 : 0;
4613 	igb->param_lp_rem_fault =
4614 	    (phy_lp_able & NWAY_LPAR_REMOTE_FAULT) ? 1 : 0;
4615 }
4616 
4617 /*
4618  * igb_get_driver_control
4619  */
4620 static void
4621 igb_get_driver_control(struct e1000_hw *hw)
4622 {
4623 	uint32_t ctrl_ext;
4624 
4625 	/* Notify firmware that driver is in control of device */
4626 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
4627 	ctrl_ext |= E1000_CTRL_EXT_DRV_LOAD;
4628 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
4629 }
4630 
4631 /*
4632  * igb_release_driver_control
4633  */
4634 static void
4635 igb_release_driver_control(struct e1000_hw *hw)
4636 {
4637 	uint32_t ctrl_ext;
4638 
4639 	/* Notify firmware that driver is no longer in control of device */
4640 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
4641 	ctrl_ext &= ~E1000_CTRL_EXT_DRV_LOAD;
4642 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
4643 }
4644 
4645 /*
4646  * igb_atomic_reserve - Atomic decrease operation
4647  */
4648 int
4649 igb_atomic_reserve(uint32_t *count_p, uint32_t n)
4650 {
4651 	uint32_t oldval;
4652 	uint32_t newval;
4653 
4654 	/* ATOMICALLY */
4655 	do {
4656 		oldval = *count_p;
4657 		if (oldval < n)
4658 			return (-1);
4659 		newval = oldval - n;
4660 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
4661 
4662 	return (newval);
4663 }
4664 
4665 /*
4666  * FMA support
4667  */
4668 
4669 int
4670 igb_check_acc_handle(ddi_acc_handle_t handle)
4671 {
4672 	ddi_fm_error_t de;
4673 
4674 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
4675 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
4676 	return (de.fme_status);
4677 }
4678 
4679 int
4680 igb_check_dma_handle(ddi_dma_handle_t handle)
4681 {
4682 	ddi_fm_error_t de;
4683 
4684 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
4685 	return (de.fme_status);
4686 }
4687 
4688 /*
4689  * The IO fault service error handling callback function
4690  */
4691 /*ARGSUSED*/
4692 static int
4693 igb_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
4694 {
4695 	/*
4696 	 * as the driver can always deal with an error in any dma or
4697 	 * access handle, we can just return the fme_status value.
4698 	 */
4699 	pci_ereport_post(dip, err, NULL);
4700 	return (err->fme_status);
4701 }
4702 
4703 static void
4704 igb_fm_init(igb_t *igb)
4705 {
4706 	ddi_iblock_cookie_t iblk;
4707 	int fma_acc_flag, fma_dma_flag;
4708 
4709 	/* Only register with IO Fault Services if we have some capability */
4710 	if (igb->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
4711 		igb_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
4712 		fma_acc_flag = 1;
4713 	} else {
4714 		igb_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
4715 		fma_acc_flag = 0;
4716 	}
4717 
4718 	if (igb->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
4719 		fma_dma_flag = 1;
4720 	} else {
4721 		fma_dma_flag = 0;
4722 	}
4723 
4724 	(void) igb_set_fma_flags(fma_acc_flag, fma_dma_flag);
4725 
4726 	if (igb->fm_capabilities) {
4727 
4728 		/* Register capabilities with IO Fault Services */
4729 		ddi_fm_init(igb->dip, &igb->fm_capabilities, &iblk);
4730 
4731 		/*
4732 		 * Initialize pci ereport capabilities if ereport capable
4733 		 */
4734 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
4735 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4736 			pci_ereport_setup(igb->dip);
4737 
4738 		/*
4739 		 * Register error callback if error callback capable
4740 		 */
4741 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4742 			ddi_fm_handler_register(igb->dip,
4743 			    igb_fm_error_cb, (void*) igb);
4744 	}
4745 }
4746 
4747 static void
4748 igb_fm_fini(igb_t *igb)
4749 {
4750 	/* Only unregister FMA capabilities if we registered some */
4751 	if (igb->fm_capabilities) {
4752 
4753 		/*
4754 		 * Release any resources allocated by pci_ereport_setup()
4755 		 */
4756 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
4757 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4758 			pci_ereport_teardown(igb->dip);
4759 
4760 		/*
4761 		 * Un-register error callback if error callback capable
4762 		 */
4763 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4764 			ddi_fm_handler_unregister(igb->dip);
4765 
4766 		/* Unregister from IO Fault Services */
4767 		ddi_fm_fini(igb->dip);
4768 	}
4769 }
4770 
4771 void
4772 igb_fm_ereport(igb_t *igb, char *detail)
4773 {
4774 	uint64_t ena;
4775 	char buf[FM_MAX_CLASS];
4776 
4777 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
4778 	ena = fm_ena_generate(0, FM_ENA_FMT1);
4779 	if (DDI_FM_EREPORT_CAP(igb->fm_capabilities)) {
4780 		ddi_fm_ereport_post(igb->dip, buf, ena, DDI_NOSLEEP,
4781 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
4782 	}
4783 }
4784