xref: /illumos-gate/usr/src/uts/common/io/igb/igb_main.c (revision eb9a1df2aeb866bf1de4494433b6d7e5fa07b3ae)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2007-2012 Intel Corporation. All rights reserved.
24  */
25 
26 /*
27  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Copyright 2013, Nexenta Systems, Inc. All rights reserved.
29  * Copyright 2016 Joyent, Inc.
30  */
31 
32 #include "igb_sw.h"
33 
34 static char ident[] = "Intel 1Gb Ethernet";
35 static char igb_version[] = "igb 2.3.8-ish";
36 
37 /*
38  * Local function protoypes
39  */
40 static int igb_register_mac(igb_t *);
41 static int igb_identify_hardware(igb_t *);
42 static int igb_regs_map(igb_t *);
43 static void igb_init_properties(igb_t *);
44 static int igb_init_driver_settings(igb_t *);
45 static void igb_init_locks(igb_t *);
46 static void igb_destroy_locks(igb_t *);
47 static int igb_init_mac_address(igb_t *);
48 static int igb_init(igb_t *);
49 static int igb_init_adapter(igb_t *);
50 static void igb_stop_adapter(igb_t *);
51 static int igb_reset(igb_t *);
52 static void igb_tx_clean(igb_t *);
53 static boolean_t igb_tx_drain(igb_t *);
54 static boolean_t igb_rx_drain(igb_t *);
55 static int igb_alloc_rings(igb_t *);
56 static int igb_alloc_rx_data(igb_t *);
57 static void igb_free_rx_data(igb_t *);
58 static void igb_free_rings(igb_t *);
59 static void igb_setup_rings(igb_t *);
60 static void igb_setup_rx(igb_t *);
61 static void igb_setup_tx(igb_t *);
62 static void igb_setup_rx_ring(igb_rx_ring_t *);
63 static void igb_setup_tx_ring(igb_tx_ring_t *);
64 static void igb_setup_rss(igb_t *);
65 static void igb_setup_mac_rss_classify(igb_t *);
66 static void igb_setup_mac_classify(igb_t *);
67 static void igb_init_unicst(igb_t *);
68 static void igb_setup_multicst(igb_t *);
69 static void igb_get_phy_state(igb_t *);
70 static void igb_param_sync(igb_t *);
71 static void igb_get_conf(igb_t *);
72 static int igb_get_prop(igb_t *, char *, int, int, int);
73 static boolean_t igb_is_link_up(igb_t *);
74 static boolean_t igb_link_check(igb_t *);
75 static void igb_local_timer(void *);
76 static void igb_link_timer(void *);
77 static void igb_arm_watchdog_timer(igb_t *);
78 static void igb_start_watchdog_timer(igb_t *);
79 static void igb_restart_watchdog_timer(igb_t *);
80 static void igb_stop_watchdog_timer(igb_t *);
81 static void igb_start_link_timer(igb_t *);
82 static void igb_stop_link_timer(igb_t *);
83 static void igb_disable_adapter_interrupts(igb_t *);
84 static void igb_enable_adapter_interrupts_82575(igb_t *);
85 static void igb_enable_adapter_interrupts_82576(igb_t *);
86 static void igb_enable_adapter_interrupts_82580(igb_t *);
87 static boolean_t is_valid_mac_addr(uint8_t *);
88 static boolean_t igb_stall_check(igb_t *);
89 static boolean_t igb_set_loopback_mode(igb_t *, uint32_t);
90 static void igb_set_external_loopback(igb_t *);
91 static void igb_set_internal_phy_loopback(igb_t *);
92 static void igb_set_internal_serdes_loopback(igb_t *);
93 static boolean_t igb_find_mac_address(igb_t *);
94 static int igb_alloc_intrs(igb_t *);
95 static int igb_alloc_intr_handles(igb_t *, int);
96 static int igb_add_intr_handlers(igb_t *);
97 static void igb_rem_intr_handlers(igb_t *);
98 static void igb_rem_intrs(igb_t *);
99 static int igb_enable_intrs(igb_t *);
100 static int igb_disable_intrs(igb_t *);
101 static void igb_setup_msix_82575(igb_t *);
102 static void igb_setup_msix_82576(igb_t *);
103 static void igb_setup_msix_82580(igb_t *);
104 static uint_t igb_intr_legacy(void *, void *);
105 static uint_t igb_intr_msi(void *, void *);
106 static uint_t igb_intr_rx(void *, void *);
107 static uint_t igb_intr_tx(void *, void *);
108 static uint_t igb_intr_tx_other(void *, void *);
109 static void igb_intr_rx_work(igb_rx_ring_t *);
110 static void igb_intr_tx_work(igb_tx_ring_t *);
111 static void igb_intr_link_work(igb_t *);
112 static void igb_get_driver_control(struct e1000_hw *);
113 static void igb_release_driver_control(struct e1000_hw *);
114 
115 static int igb_attach(dev_info_t *, ddi_attach_cmd_t);
116 static int igb_detach(dev_info_t *, ddi_detach_cmd_t);
117 static int igb_resume(dev_info_t *);
118 static int igb_suspend(dev_info_t *);
119 static int igb_quiesce(dev_info_t *);
120 static void igb_unconfigure(dev_info_t *, igb_t *);
121 static int igb_fm_error_cb(dev_info_t *, ddi_fm_error_t *,
122     const void *);
123 static void igb_fm_init(igb_t *);
124 static void igb_fm_fini(igb_t *);
125 static void igb_release_multicast(igb_t *);
126 
127 char *igb_priv_props[] = {
128 	"_eee_support",
129 	"_tx_copy_thresh",
130 	"_tx_recycle_thresh",
131 	"_tx_overload_thresh",
132 	"_tx_resched_thresh",
133 	"_rx_copy_thresh",
134 	"_rx_limit_per_intr",
135 	"_intr_throttling",
136 	"_adv_pause_cap",
137 	"_adv_asym_pause_cap",
138 	NULL
139 };
140 
141 static struct cb_ops igb_cb_ops = {
142 	nulldev,		/* cb_open */
143 	nulldev,		/* cb_close */
144 	nodev,			/* cb_strategy */
145 	nodev,			/* cb_print */
146 	nodev,			/* cb_dump */
147 	nodev,			/* cb_read */
148 	nodev,			/* cb_write */
149 	nodev,			/* cb_ioctl */
150 	nodev,			/* cb_devmap */
151 	nodev,			/* cb_mmap */
152 	nodev,			/* cb_segmap */
153 	nochpoll,		/* cb_chpoll */
154 	ddi_prop_op,		/* cb_prop_op */
155 	NULL,			/* cb_stream */
156 	D_MP | D_HOTPLUG,	/* cb_flag */
157 	CB_REV,			/* cb_rev */
158 	nodev,			/* cb_aread */
159 	nodev			/* cb_awrite */
160 };
161 
162 static struct dev_ops igb_dev_ops = {
163 	DEVO_REV,		/* devo_rev */
164 	0,			/* devo_refcnt */
165 	NULL,			/* devo_getinfo */
166 	nulldev,		/* devo_identify */
167 	nulldev,		/* devo_probe */
168 	igb_attach,		/* devo_attach */
169 	igb_detach,		/* devo_detach */
170 	nodev,			/* devo_reset */
171 	&igb_cb_ops,		/* devo_cb_ops */
172 	NULL,			/* devo_bus_ops */
173 	ddi_power,		/* devo_power */
174 	igb_quiesce,	/* devo_quiesce */
175 };
176 
177 static struct modldrv igb_modldrv = {
178 	&mod_driverops,		/* Type of module.  This one is a driver */
179 	ident,			/* Discription string */
180 	&igb_dev_ops,		/* driver ops */
181 };
182 
183 static struct modlinkage igb_modlinkage = {
184 	MODREV_1, &igb_modldrv, NULL
185 };
186 
187 /* Access attributes for register mapping */
188 ddi_device_acc_attr_t igb_regs_acc_attr = {
189 	DDI_DEVICE_ATTR_V1,
190 	DDI_STRUCTURE_LE_ACC,
191 	DDI_STRICTORDER_ACC,
192 	DDI_FLAGERR_ACC
193 };
194 
195 #define	IGB_M_CALLBACK_FLAGS \
196 	(MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP | MC_PROPINFO)
197 
198 static mac_callbacks_t igb_m_callbacks = {
199 	IGB_M_CALLBACK_FLAGS,
200 	igb_m_stat,
201 	igb_m_start,
202 	igb_m_stop,
203 	igb_m_promisc,
204 	igb_m_multicst,
205 	NULL,
206 	NULL,
207 	NULL,
208 	igb_m_ioctl,
209 	igb_m_getcapab,
210 	NULL,
211 	NULL,
212 	igb_m_setprop,
213 	igb_m_getprop,
214 	igb_m_propinfo
215 };
216 
217 /*
218  * Initialize capabilities of each supported adapter type
219  */
220 static adapter_info_t igb_82575_cap = {
221 	/* limits */
222 	4,		/* maximum number of rx queues */
223 	1,		/* minimum number of rx queues */
224 	4,		/* default number of rx queues */
225 	4,		/* maximum number of tx queues */
226 	1,		/* minimum number of tx queues */
227 	4,		/* default number of tx queues */
228 	65535,		/* maximum interrupt throttle rate */
229 	0,		/* minimum interrupt throttle rate */
230 	200,		/* default interrupt throttle rate */
231 
232 	/* function pointers */
233 	igb_enable_adapter_interrupts_82575,
234 	igb_setup_msix_82575,
235 
236 	/* capabilities */
237 	(IGB_FLAG_HAS_DCA |	/* capability flags */
238 	IGB_FLAG_VMDQ_POOL),
239 
240 	0xffc00000		/* mask for RXDCTL register */
241 };
242 
243 static adapter_info_t igb_82576_cap = {
244 	/* limits */
245 	16,		/* maximum number of rx queues */
246 	1,		/* minimum number of rx queues */
247 	4,		/* default number of rx queues */
248 	16,		/* maximum number of tx queues */
249 	1,		/* minimum number of tx queues */
250 	4,		/* default number of tx queues */
251 	65535,		/* maximum interrupt throttle rate */
252 	0,		/* minimum interrupt throttle rate */
253 	200,		/* default interrupt throttle rate */
254 
255 	/* function pointers */
256 	igb_enable_adapter_interrupts_82576,
257 	igb_setup_msix_82576,
258 
259 	/* capabilities */
260 	(IGB_FLAG_HAS_DCA |	/* capability flags */
261 	IGB_FLAG_VMDQ_POOL |
262 	IGB_FLAG_NEED_CTX_IDX),
263 
264 	0xffe00000		/* mask for RXDCTL register */
265 };
266 
267 static adapter_info_t igb_82580_cap = {
268 	/* limits */
269 	8,		/* maximum number of rx queues */
270 	1,		/* minimum number of rx queues */
271 	4,		/* default number of rx queues */
272 	8,		/* maximum number of tx queues */
273 	1,		/* minimum number of tx queues */
274 	4,		/* default number of tx queues */
275 	65535,		/* maximum interrupt throttle rate */
276 	0,		/* minimum interrupt throttle rate */
277 	200,		/* default interrupt throttle rate */
278 
279 	/* function pointers */
280 	igb_enable_adapter_interrupts_82580,
281 	igb_setup_msix_82580,
282 
283 	/* capabilities */
284 	(IGB_FLAG_HAS_DCA |	/* capability flags */
285 	IGB_FLAG_VMDQ_POOL |
286 	IGB_FLAG_NEED_CTX_IDX),
287 
288 	0xffe00000		/* mask for RXDCTL register */
289 };
290 
291 static adapter_info_t igb_i350_cap = {
292 	/* limits */
293 	8,		/* maximum number of rx queues */
294 	1,		/* minimum number of rx queues */
295 	4,		/* default number of rx queues */
296 	8,		/* maximum number of tx queues */
297 	1,		/* minimum number of tx queues */
298 	4,		/* default number of tx queues */
299 	65535,		/* maximum interrupt throttle rate */
300 	0,		/* minimum interrupt throttle rate */
301 	200,		/* default interrupt throttle rate */
302 
303 	/* function pointers */
304 	igb_enable_adapter_interrupts_82580,
305 	igb_setup_msix_82580,
306 
307 	/* capabilities */
308 	(IGB_FLAG_HAS_DCA |	/* capability flags */
309 	IGB_FLAG_VMDQ_POOL |
310 	IGB_FLAG_NEED_CTX_IDX),
311 
312 	0xffe00000		/* mask for RXDCTL register */
313 };
314 
315 static adapter_info_t igb_i210_cap = {
316 	/* limits */
317 	4,		/* maximum number of rx queues */
318 	1,		/* minimum number of rx queues */
319 	4,		/* default number of rx queues */
320 	4,		/* maximum number of tx queues */
321 	1,		/* minimum number of tx queues */
322 	4,		/* default number of tx queues */
323 	65535,		/* maximum interrupt throttle rate */
324 	0,		/* minimum interrupt throttle rate */
325 	200,		/* default interrupt throttle rate */
326 
327 	/* function pointers */
328 	igb_enable_adapter_interrupts_82580,
329 	igb_setup_msix_82580,
330 
331 	/* capabilities */
332 	(IGB_FLAG_HAS_DCA |	/* capability flags */
333 	IGB_FLAG_VMDQ_POOL |
334 	IGB_FLAG_NEED_CTX_IDX),
335 
336 	0xfff00000		/* mask for RXDCTL register */
337 };
338 
339 static adapter_info_t igb_i354_cap = {
340 	/* limits */
341 	8,		/* maximum number of rx queues */
342 	1,		/* minimum number of rx queues */
343 	4,		/* default number of rx queues */
344 	8,		/* maximum number of tx queues */
345 	1,		/* minimum number of tx queues */
346 	4,		/* default number of tx queues */
347 	65535,		/* maximum interrupt throttle rate */
348 	0,		/* minimum interrupt throttle rate */
349 	200,		/* default interrupt throttle rate */
350 
351 	/* function pointers */
352 	igb_enable_adapter_interrupts_82580,
353 	igb_setup_msix_82580,
354 
355 	/* capabilities */
356 	(IGB_FLAG_HAS_DCA |	/* capability flags */
357 	IGB_FLAG_VMDQ_POOL |
358 	IGB_FLAG_NEED_CTX_IDX),
359 
360 	0xfff00000		/* mask for RXDCTL register */
361 };
362 
363 /*
364  * Module Initialization Functions
365  */
366 
367 int
368 _init(void)
369 {
370 	int status;
371 
372 	mac_init_ops(&igb_dev_ops, MODULE_NAME);
373 
374 	status = mod_install(&igb_modlinkage);
375 
376 	if (status != DDI_SUCCESS) {
377 		mac_fini_ops(&igb_dev_ops);
378 	}
379 
380 	return (status);
381 }
382 
383 int
384 _fini(void)
385 {
386 	int status;
387 
388 	status = mod_remove(&igb_modlinkage);
389 
390 	if (status == DDI_SUCCESS) {
391 		mac_fini_ops(&igb_dev_ops);
392 	}
393 
394 	return (status);
395 
396 }
397 
398 int
399 _info(struct modinfo *modinfop)
400 {
401 	int status;
402 
403 	status = mod_info(&igb_modlinkage, modinfop);
404 
405 	return (status);
406 }
407 
408 /*
409  * igb_attach - driver attach
410  *
411  * This function is the device specific initialization entry
412  * point. This entry point is required and must be written.
413  * The DDI_ATTACH command must be provided in the attach entry
414  * point. When attach() is called with cmd set to DDI_ATTACH,
415  * all normal kernel services (such as kmem_alloc(9F)) are
416  * available for use by the driver.
417  *
418  * The attach() function will be called once for each instance
419  * of  the  device  on  the  system with cmd set to DDI_ATTACH.
420  * Until attach() succeeds, the only driver entry points which
421  * may be called are open(9E) and getinfo(9E).
422  */
423 static int
424 igb_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
425 {
426 	igb_t *igb;
427 	struct igb_osdep *osdep;
428 	struct e1000_hw *hw;
429 	int instance;
430 
431 	/*
432 	 * Check the command and perform corresponding operations
433 	 */
434 	switch (cmd) {
435 	default:
436 		return (DDI_FAILURE);
437 
438 	case DDI_RESUME:
439 		return (igb_resume(devinfo));
440 
441 	case DDI_ATTACH:
442 		break;
443 	}
444 
445 	/* Get the device instance */
446 	instance = ddi_get_instance(devinfo);
447 
448 	/* Allocate memory for the instance data structure */
449 	igb = kmem_zalloc(sizeof (igb_t), KM_SLEEP);
450 
451 	igb->dip = devinfo;
452 	igb->instance = instance;
453 
454 	hw = &igb->hw;
455 	osdep = &igb->osdep;
456 	hw->back = osdep;
457 	osdep->igb = igb;
458 
459 	/* Attach the instance pointer to the dev_info data structure */
460 	ddi_set_driver_private(devinfo, igb);
461 
462 
463 	/* Initialize for fma support */
464 	igb->fm_capabilities = igb_get_prop(igb, "fm-capable",
465 	    0, 0x0f,
466 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
467 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
468 	igb_fm_init(igb);
469 	igb->attach_progress |= ATTACH_PROGRESS_FMINIT;
470 
471 	/*
472 	 * Map PCI config space registers
473 	 */
474 	if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
475 		igb_log(igb, IGB_LOG_ERROR, "Failed to map PCI configurations");
476 		goto attach_fail;
477 	}
478 	igb->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
479 
480 	/*
481 	 * Identify the chipset family
482 	 */
483 	if (igb_identify_hardware(igb) != IGB_SUCCESS) {
484 		igb_log(igb, IGB_LOG_ERROR, "Failed to identify hardware");
485 		goto attach_fail;
486 	}
487 
488 	/*
489 	 * Map device registers
490 	 */
491 	if (igb_regs_map(igb) != IGB_SUCCESS) {
492 		igb_log(igb, IGB_LOG_ERROR, "Failed to map device registers");
493 		goto attach_fail;
494 	}
495 	igb->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
496 
497 	/*
498 	 * Initialize driver parameters
499 	 */
500 	igb_init_properties(igb);
501 	igb->attach_progress |= ATTACH_PROGRESS_PROPS;
502 
503 	/*
504 	 * Allocate interrupts
505 	 */
506 	if (igb_alloc_intrs(igb) != IGB_SUCCESS) {
507 		igb_log(igb, IGB_LOG_ERROR, "Failed to allocate interrupts");
508 		goto attach_fail;
509 	}
510 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
511 
512 	/*
513 	 * Allocate rx/tx rings based on the ring numbers.
514 	 * The actual numbers of rx/tx rings are decided by the number of
515 	 * allocated interrupt vectors, so we should allocate the rings after
516 	 * interrupts are allocated.
517 	 */
518 	if (igb_alloc_rings(igb) != IGB_SUCCESS) {
519 		igb_log(igb, IGB_LOG_ERROR,
520 		    "Failed to allocate rx/tx rings or groups");
521 		goto attach_fail;
522 	}
523 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_RINGS;
524 
525 	/*
526 	 * Add interrupt handlers
527 	 */
528 	if (igb_add_intr_handlers(igb) != IGB_SUCCESS) {
529 		igb_log(igb, IGB_LOG_ERROR, "Failed to add interrupt handlers");
530 		goto attach_fail;
531 	}
532 	igb->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
533 
534 	/*
535 	 * Initialize driver parameters
536 	 */
537 	if (igb_init_driver_settings(igb) != IGB_SUCCESS) {
538 		igb_log(igb, IGB_LOG_ERROR,
539 		    "Failed to initialize driver settings");
540 		goto attach_fail;
541 	}
542 
543 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK) {
544 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
545 		goto attach_fail;
546 	}
547 
548 	/*
549 	 * Initialize mutexes for this device.
550 	 * Do this before enabling the interrupt handler and
551 	 * register the softint to avoid the condition where
552 	 * interrupt handler can try using uninitialized mutex
553 	 */
554 	igb_init_locks(igb);
555 	igb->attach_progress |= ATTACH_PROGRESS_LOCKS;
556 
557 	/*
558 	 * Initialize the adapter
559 	 */
560 	if (igb_init(igb) != IGB_SUCCESS) {
561 		igb_log(igb, IGB_LOG_ERROR, "Failed to initialize adapter");
562 		goto attach_fail;
563 	}
564 	igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
565 
566 	/*
567 	 * Initialize statistics
568 	 */
569 	if (igb_init_stats(igb) != IGB_SUCCESS) {
570 		igb_log(igb, IGB_LOG_ERROR, "Failed to initialize statistics");
571 		goto attach_fail;
572 	}
573 	igb->attach_progress |= ATTACH_PROGRESS_STATS;
574 
575 	/*
576 	 * Register the driver to the MAC
577 	 */
578 	if (igb_register_mac(igb) != IGB_SUCCESS) {
579 		igb_log(igb, IGB_LOG_ERROR, "Failed to register MAC");
580 		goto attach_fail;
581 	}
582 	igb->attach_progress |= ATTACH_PROGRESS_MAC;
583 
584 	/*
585 	 * Now that mutex locks are initialized, and the chip is also
586 	 * initialized, enable interrupts.
587 	 */
588 	if (igb_enable_intrs(igb) != IGB_SUCCESS) {
589 		igb_log(igb, IGB_LOG_ERROR, "Failed to enable DDI interrupts");
590 		goto attach_fail;
591 	}
592 	igb->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
593 
594 	igb_log(igb, IGB_LOG_INFO, "%s", igb_version);
595 	atomic_or_32(&igb->igb_state, IGB_INITIALIZED);
596 
597 	/*
598 	 * Newer models have Energy Efficient Ethernet, let's disable this by
599 	 * default.
600 	 */
601 	if (igb->hw.mac.type == e1000_i350)
602 		(void) e1000_set_eee_i350(&igb->hw, B_FALSE, B_FALSE);
603 	else if (igb->hw.mac.type == e1000_i354)
604 		(void) e1000_set_eee_i354(&igb->hw, B_FALSE, B_FALSE);
605 
606 	return (DDI_SUCCESS);
607 
608 attach_fail:
609 	igb_unconfigure(devinfo, igb);
610 	return (DDI_FAILURE);
611 }
612 
613 /*
614  * igb_detach - driver detach
615  *
616  * The detach() function is the complement of the attach routine.
617  * If cmd is set to DDI_DETACH, detach() is used to remove  the
618  * state  associated  with  a  given  instance of a device node
619  * prior to the removal of that instance from the system.
620  *
621  * The detach() function will be called once for each  instance
622  * of the device for which there has been a successful attach()
623  * once there are no longer  any  opens  on  the  device.
624  *
625  * Interrupts routine are disabled, All memory allocated by this
626  * driver are freed.
627  */
628 static int
629 igb_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
630 {
631 	igb_t *igb;
632 
633 	/*
634 	 * Check detach command
635 	 */
636 	switch (cmd) {
637 	default:
638 		return (DDI_FAILURE);
639 
640 	case DDI_SUSPEND:
641 		return (igb_suspend(devinfo));
642 
643 	case DDI_DETACH:
644 		break;
645 	}
646 
647 
648 	/*
649 	 * Get the pointer to the driver private data structure
650 	 */
651 	igb = (igb_t *)ddi_get_driver_private(devinfo);
652 	if (igb == NULL)
653 		return (DDI_FAILURE);
654 
655 	/*
656 	 * Unregister MAC. If failed, we have to fail the detach
657 	 */
658 	if (mac_unregister(igb->mac_hdl) != 0) {
659 		igb_log(igb, IGB_LOG_ERROR, "Failed to unregister MAC");
660 		return (DDI_FAILURE);
661 	}
662 	igb->attach_progress &= ~ATTACH_PROGRESS_MAC;
663 
664 	/*
665 	 * If the device is still running, it needs to be stopped first.
666 	 * This check is necessary because under some specific circumstances,
667 	 * the detach routine can be called without stopping the interface
668 	 * first.
669 	 */
670 	mutex_enter(&igb->gen_lock);
671 	if (igb->igb_state & IGB_STARTED) {
672 		atomic_and_32(&igb->igb_state, ~IGB_STARTED);
673 		igb_stop(igb, B_TRUE);
674 		mutex_exit(&igb->gen_lock);
675 		/* Disable and stop the watchdog timer */
676 		igb_disable_watchdog_timer(igb);
677 	} else
678 		mutex_exit(&igb->gen_lock);
679 
680 	/*
681 	 * Check if there are still rx buffers held by the upper layer.
682 	 * If so, fail the detach.
683 	 */
684 	if (!igb_rx_drain(igb))
685 		return (DDI_FAILURE);
686 
687 	/*
688 	 * Do the remaining unconfigure routines
689 	 */
690 	igb_unconfigure(devinfo, igb);
691 
692 	return (DDI_SUCCESS);
693 }
694 
695 /*
696  * quiesce(9E) entry point.
697  *
698  * This function is called when the system is single-threaded at high
699  * PIL with preemption disabled. Therefore, this function must not be
700  * blocked.
701  *
702  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
703  * DDI_FAILURE indicates an error condition and should almost never happen.
704  */
705 static int
706 igb_quiesce(dev_info_t *devinfo)
707 {
708 	igb_t *igb;
709 	struct e1000_hw *hw;
710 
711 	igb = (igb_t *)ddi_get_driver_private(devinfo);
712 
713 	if (igb == NULL)
714 		return (DDI_FAILURE);
715 
716 	hw = &igb->hw;
717 
718 	/*
719 	 * Disable the adapter interrupts
720 	 */
721 	igb_disable_adapter_interrupts(igb);
722 
723 	/* Tell firmware driver is no longer in control */
724 	igb_release_driver_control(hw);
725 
726 	/*
727 	 * Reset the chipset
728 	 */
729 	(void) e1000_reset_hw(hw);
730 
731 	/*
732 	 * Reset PHY if possible
733 	 */
734 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
735 		(void) e1000_phy_hw_reset(hw);
736 
737 	return (DDI_SUCCESS);
738 }
739 
740 /*
741  * igb_unconfigure - release all resources held by this instance
742  */
743 static void
744 igb_unconfigure(dev_info_t *devinfo, igb_t *igb)
745 {
746 	/*
747 	 * Disable interrupt
748 	 */
749 	if (igb->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
750 		(void) igb_disable_intrs(igb);
751 	}
752 
753 	/*
754 	 * Unregister MAC
755 	 */
756 	if (igb->attach_progress & ATTACH_PROGRESS_MAC) {
757 		(void) mac_unregister(igb->mac_hdl);
758 	}
759 
760 	/*
761 	 * Free statistics
762 	 */
763 	if (igb->attach_progress & ATTACH_PROGRESS_STATS) {
764 		kstat_delete((kstat_t *)igb->igb_ks);
765 	}
766 
767 	/*
768 	 * Remove interrupt handlers
769 	 */
770 	if (igb->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
771 		igb_rem_intr_handlers(igb);
772 	}
773 
774 	/*
775 	 * Remove interrupts
776 	 */
777 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_INTR) {
778 		igb_rem_intrs(igb);
779 	}
780 
781 	/*
782 	 * Remove driver properties
783 	 */
784 	if (igb->attach_progress & ATTACH_PROGRESS_PROPS) {
785 		(void) ddi_prop_remove_all(devinfo);
786 	}
787 
788 	/*
789 	 * Stop the adapter
790 	 */
791 	if (igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) {
792 		mutex_enter(&igb->gen_lock);
793 		igb_stop_adapter(igb);
794 		mutex_exit(&igb->gen_lock);
795 		if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
796 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_UNAFFECTED);
797 	}
798 
799 	/*
800 	 * Free multicast table
801 	 */
802 	igb_release_multicast(igb);
803 
804 	/*
805 	 * Free register handle
806 	 */
807 	if (igb->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
808 		if (igb->osdep.reg_handle != NULL)
809 			ddi_regs_map_free(&igb->osdep.reg_handle);
810 	}
811 
812 	/*
813 	 * Free PCI config handle
814 	 */
815 	if (igb->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
816 		if (igb->osdep.cfg_handle != NULL)
817 			pci_config_teardown(&igb->osdep.cfg_handle);
818 	}
819 
820 	/*
821 	 * Free locks
822 	 */
823 	if (igb->attach_progress & ATTACH_PROGRESS_LOCKS) {
824 		igb_destroy_locks(igb);
825 	}
826 
827 	/*
828 	 * Free the rx/tx rings
829 	 */
830 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
831 		igb_free_rings(igb);
832 	}
833 
834 	/*
835 	 * Remove FMA
836 	 */
837 	if (igb->attach_progress & ATTACH_PROGRESS_FMINIT) {
838 		igb_fm_fini(igb);
839 	}
840 
841 	/*
842 	 * Free the driver data structure
843 	 */
844 	kmem_free(igb, sizeof (igb_t));
845 
846 	ddi_set_driver_private(devinfo, NULL);
847 }
848 
849 /*
850  * igb_register_mac - Register the driver and its function pointers with
851  * the GLD interface
852  */
853 static int
854 igb_register_mac(igb_t *igb)
855 {
856 	struct e1000_hw *hw = &igb->hw;
857 	mac_register_t *mac;
858 	int status;
859 
860 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
861 		return (IGB_FAILURE);
862 
863 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
864 	mac->m_driver = igb;
865 	mac->m_dip = igb->dip;
866 	mac->m_src_addr = hw->mac.addr;
867 	mac->m_callbacks = &igb_m_callbacks;
868 	mac->m_min_sdu = 0;
869 	mac->m_max_sdu = igb->max_frame_size -
870 	    sizeof (struct ether_vlan_header) - ETHERFCSL;
871 	mac->m_margin = VLAN_TAGSZ;
872 	mac->m_priv_props = igb_priv_props;
873 	mac->m_v12n = MAC_VIRT_LEVEL1;
874 
875 	status = mac_register(mac, &igb->mac_hdl);
876 
877 	mac_free(mac);
878 
879 	return ((status == 0) ? IGB_SUCCESS : IGB_FAILURE);
880 }
881 
882 /*
883  * igb_identify_hardware - Identify the type of the chipset
884  */
885 static int
886 igb_identify_hardware(igb_t *igb)
887 {
888 	struct e1000_hw *hw = &igb->hw;
889 	struct igb_osdep *osdep = &igb->osdep;
890 
891 	/*
892 	 * Get the device id
893 	 */
894 	hw->vendor_id =
895 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
896 	hw->device_id =
897 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
898 	hw->revision_id =
899 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
900 	hw->subsystem_device_id =
901 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
902 	hw->subsystem_vendor_id =
903 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
904 
905 	/*
906 	 * Set the mac type of the adapter based on the device id
907 	 */
908 	if (e1000_set_mac_type(hw) != E1000_SUCCESS) {
909 		return (IGB_FAILURE);
910 	}
911 
912 	/*
913 	 * Install adapter capabilities based on mac type
914 	 */
915 	switch (hw->mac.type) {
916 	case e1000_82575:
917 		igb->capab = &igb_82575_cap;
918 		break;
919 	case e1000_82576:
920 		igb->capab = &igb_82576_cap;
921 		break;
922 	case e1000_82580:
923 		igb->capab = &igb_82580_cap;
924 		break;
925 	case e1000_i350:
926 		igb->capab = &igb_i350_cap;
927 		break;
928 	case e1000_i210:
929 	case e1000_i211:
930 		igb->capab = &igb_i210_cap;
931 		break;
932 	case e1000_i354:
933 		igb->capab = &igb_i354_cap;
934 		break;
935 	default:
936 		return (IGB_FAILURE);
937 	}
938 
939 	return (IGB_SUCCESS);
940 }
941 
942 /*
943  * igb_regs_map - Map the device registers
944  */
945 static int
946 igb_regs_map(igb_t *igb)
947 {
948 	dev_info_t *devinfo = igb->dip;
949 	struct e1000_hw *hw = &igb->hw;
950 	struct igb_osdep *osdep = &igb->osdep;
951 	off_t mem_size;
952 
953 	/*
954 	 * First get the size of device registers to be mapped.
955 	 */
956 	if (ddi_dev_regsize(devinfo, IGB_ADAPTER_REGSET, &mem_size) !=
957 	    DDI_SUCCESS) {
958 		return (IGB_FAILURE);
959 	}
960 
961 	/*
962 	 * Call ddi_regs_map_setup() to map registers
963 	 */
964 	if ((ddi_regs_map_setup(devinfo, IGB_ADAPTER_REGSET,
965 	    (caddr_t *)&hw->hw_addr, 0,
966 	    mem_size, &igb_regs_acc_attr,
967 	    &osdep->reg_handle)) != DDI_SUCCESS) {
968 		return (IGB_FAILURE);
969 	}
970 
971 	return (IGB_SUCCESS);
972 }
973 
974 /*
975  * igb_init_properties - Initialize driver properties
976  */
977 static void
978 igb_init_properties(igb_t *igb)
979 {
980 	/*
981 	 * Get conf file properties, including link settings
982 	 * jumbo frames, ring number, descriptor number, etc.
983 	 */
984 	igb_get_conf(igb);
985 }
986 
987 /*
988  * igb_init_driver_settings - Initialize driver settings
989  *
990  * The settings include hardware function pointers, bus information,
991  * rx/tx rings settings, link state, and any other parameters that
992  * need to be setup during driver initialization.
993  */
994 static int
995 igb_init_driver_settings(igb_t *igb)
996 {
997 	struct e1000_hw *hw = &igb->hw;
998 	igb_rx_ring_t *rx_ring;
999 	igb_tx_ring_t *tx_ring;
1000 	uint32_t rx_size;
1001 	uint32_t tx_size;
1002 	int i;
1003 
1004 	/*
1005 	 * Initialize chipset specific hardware function pointers
1006 	 */
1007 	if (e1000_setup_init_funcs(hw, B_TRUE) != E1000_SUCCESS) {
1008 		return (IGB_FAILURE);
1009 	}
1010 
1011 	/*
1012 	 * Get bus information
1013 	 */
1014 	if (e1000_get_bus_info(hw) != E1000_SUCCESS) {
1015 		return (IGB_FAILURE);
1016 	}
1017 
1018 	/*
1019 	 * Get the system page size
1020 	 */
1021 	igb->page_size = ddi_ptob(igb->dip, (ulong_t)1);
1022 
1023 	/*
1024 	 * Set rx buffer size
1025 	 * The IP header alignment room is counted in the calculation.
1026 	 * The rx buffer size is in unit of 1K that is required by the
1027 	 * chipset hardware.
1028 	 */
1029 	rx_size = igb->max_frame_size + IPHDR_ALIGN_ROOM;
1030 	igb->rx_buf_size = ((rx_size >> 10) +
1031 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
1032 
1033 	/*
1034 	 * Set tx buffer size
1035 	 */
1036 	tx_size = igb->max_frame_size;
1037 	igb->tx_buf_size = ((tx_size >> 10) +
1038 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
1039 
1040 	/*
1041 	 * Initialize rx/tx rings parameters
1042 	 */
1043 	for (i = 0; i < igb->num_rx_rings; i++) {
1044 		rx_ring = &igb->rx_rings[i];
1045 		rx_ring->index = i;
1046 		rx_ring->igb = igb;
1047 	}
1048 
1049 	for (i = 0; i < igb->num_tx_rings; i++) {
1050 		tx_ring = &igb->tx_rings[i];
1051 		tx_ring->index = i;
1052 		tx_ring->igb = igb;
1053 		if (igb->tx_head_wb_enable)
1054 			tx_ring->tx_recycle = igb_tx_recycle_head_wb;
1055 		else
1056 			tx_ring->tx_recycle = igb_tx_recycle_legacy;
1057 
1058 		tx_ring->ring_size = igb->tx_ring_size;
1059 		tx_ring->free_list_size = igb->tx_ring_size +
1060 		    (igb->tx_ring_size >> 1);
1061 	}
1062 
1063 	/*
1064 	 * Initialize values of interrupt throttling rates
1065 	 */
1066 	for (i = 1; i < MAX_NUM_EITR; i++)
1067 		igb->intr_throttling[i] = igb->intr_throttling[0];
1068 
1069 	/*
1070 	 * The initial link state should be "unknown"
1071 	 */
1072 	igb->link_state = LINK_STATE_UNKNOWN;
1073 
1074 	return (IGB_SUCCESS);
1075 }
1076 
1077 /*
1078  * igb_init_locks - Initialize locks
1079  */
1080 static void
1081 igb_init_locks(igb_t *igb)
1082 {
1083 	igb_rx_ring_t *rx_ring;
1084 	igb_tx_ring_t *tx_ring;
1085 	int i;
1086 
1087 	for (i = 0; i < igb->num_rx_rings; i++) {
1088 		rx_ring = &igb->rx_rings[i];
1089 		mutex_init(&rx_ring->rx_lock, NULL,
1090 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1091 	}
1092 
1093 	for (i = 0; i < igb->num_tx_rings; i++) {
1094 		tx_ring = &igb->tx_rings[i];
1095 		mutex_init(&tx_ring->tx_lock, NULL,
1096 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1097 		mutex_init(&tx_ring->recycle_lock, NULL,
1098 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1099 		mutex_init(&tx_ring->tcb_head_lock, NULL,
1100 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1101 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
1102 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1103 	}
1104 
1105 	mutex_init(&igb->gen_lock, NULL,
1106 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1107 
1108 	mutex_init(&igb->watchdog_lock, NULL,
1109 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1110 
1111 	mutex_init(&igb->link_lock, NULL,
1112 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1113 }
1114 
1115 /*
1116  * igb_destroy_locks - Destroy locks
1117  */
1118 static void
1119 igb_destroy_locks(igb_t *igb)
1120 {
1121 	igb_rx_ring_t *rx_ring;
1122 	igb_tx_ring_t *tx_ring;
1123 	int i;
1124 
1125 	for (i = 0; i < igb->num_rx_rings; i++) {
1126 		rx_ring = &igb->rx_rings[i];
1127 		mutex_destroy(&rx_ring->rx_lock);
1128 	}
1129 
1130 	for (i = 0; i < igb->num_tx_rings; i++) {
1131 		tx_ring = &igb->tx_rings[i];
1132 		mutex_destroy(&tx_ring->tx_lock);
1133 		mutex_destroy(&tx_ring->recycle_lock);
1134 		mutex_destroy(&tx_ring->tcb_head_lock);
1135 		mutex_destroy(&tx_ring->tcb_tail_lock);
1136 	}
1137 
1138 	mutex_destroy(&igb->gen_lock);
1139 	mutex_destroy(&igb->watchdog_lock);
1140 	mutex_destroy(&igb->link_lock);
1141 }
1142 
1143 static int
1144 igb_resume(dev_info_t *devinfo)
1145 {
1146 	igb_t *igb;
1147 
1148 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1149 	if (igb == NULL)
1150 		return (DDI_FAILURE);
1151 
1152 	mutex_enter(&igb->gen_lock);
1153 
1154 	/*
1155 	 * Enable interrupts
1156 	 */
1157 	if (igb->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
1158 		if (igb_enable_intrs(igb) != IGB_SUCCESS) {
1159 			igb_log(igb, IGB_LOG_ERROR,
1160 			    "Failed to enable DDI interrupts");
1161 			mutex_exit(&igb->gen_lock);
1162 			return (DDI_FAILURE);
1163 		}
1164 	}
1165 
1166 	if (igb->igb_state & IGB_STARTED) {
1167 		if (igb_start(igb, B_FALSE) != IGB_SUCCESS) {
1168 			mutex_exit(&igb->gen_lock);
1169 			return (DDI_FAILURE);
1170 		}
1171 
1172 		/*
1173 		 * Enable and start the watchdog timer
1174 		 */
1175 		igb_enable_watchdog_timer(igb);
1176 	}
1177 
1178 	atomic_and_32(&igb->igb_state, ~IGB_SUSPENDED);
1179 
1180 	mutex_exit(&igb->gen_lock);
1181 
1182 	return (DDI_SUCCESS);
1183 }
1184 
1185 static int
1186 igb_suspend(dev_info_t *devinfo)
1187 {
1188 	igb_t *igb;
1189 
1190 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1191 	if (igb == NULL)
1192 		return (DDI_FAILURE);
1193 
1194 	mutex_enter(&igb->gen_lock);
1195 
1196 	atomic_or_32(&igb->igb_state, IGB_SUSPENDED);
1197 
1198 	/*
1199 	 * Disable interrupts
1200 	 */
1201 	if (igb->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
1202 		(void) igb_disable_intrs(igb);
1203 	}
1204 
1205 	if (!(igb->igb_state & IGB_STARTED)) {
1206 		mutex_exit(&igb->gen_lock);
1207 		return (DDI_SUCCESS);
1208 	}
1209 
1210 	igb_stop(igb, B_FALSE);
1211 
1212 	mutex_exit(&igb->gen_lock);
1213 
1214 	/*
1215 	 * Disable and stop the watchdog timer
1216 	 */
1217 	igb_disable_watchdog_timer(igb);
1218 
1219 	return (DDI_SUCCESS);
1220 }
1221 
1222 static int
1223 igb_init(igb_t *igb)
1224 {
1225 	mutex_enter(&igb->gen_lock);
1226 
1227 	/*
1228 	 * Initilize the adapter
1229 	 */
1230 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1231 		mutex_exit(&igb->gen_lock);
1232 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1233 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1234 		return (IGB_FAILURE);
1235 	}
1236 
1237 	mutex_exit(&igb->gen_lock);
1238 
1239 	return (IGB_SUCCESS);
1240 }
1241 
1242 /*
1243  * igb_init_mac_address - Initialize the default MAC address
1244  *
1245  * On success, the MAC address is entered in the igb->hw.mac.addr
1246  * and hw->mac.perm_addr fields and the adapter's RAR(0) receive
1247  * address register.
1248  *
1249  * Important side effects:
1250  * 1. adapter is reset - this is required to put it in a known state.
1251  * 2. all of non-volatile memory (NVM) is read & checksummed - NVM is where
1252  * MAC address and all default settings are stored, so a valid checksum
1253  * is required.
1254  */
1255 static int
1256 igb_init_mac_address(igb_t *igb)
1257 {
1258 	struct e1000_hw *hw = &igb->hw;
1259 
1260 	ASSERT(mutex_owned(&igb->gen_lock));
1261 
1262 	/*
1263 	 * Reset chipset to put the hardware in a known state
1264 	 * before we try to get MAC address from NVM.
1265 	 */
1266 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1267 		igb_log(igb, IGB_LOG_ERROR, "Adapter reset failed.");
1268 		goto init_mac_fail;
1269 	}
1270 
1271 	/*
1272 	 * NVM validation
1273 	 */
1274 	if (((igb->hw.mac.type != e1000_i210) &&
1275 	    (igb->hw.mac.type != e1000_i211)) &&
1276 	    (e1000_validate_nvm_checksum(hw) < 0)) {
1277 		/*
1278 		 * Some PCI-E parts fail the first check due to
1279 		 * the link being in sleep state.  Call it again,
1280 		 * if it fails a second time its a real issue.
1281 		 */
1282 		if (e1000_validate_nvm_checksum(hw) < 0) {
1283 			igb_log(igb, IGB_LOG_ERROR,
1284 			    "Invalid NVM checksum. Please contact "
1285 			    "the vendor to update the NVM.");
1286 			goto init_mac_fail;
1287 		}
1288 	}
1289 
1290 	/*
1291 	 * Get the mac address
1292 	 * This function should handle SPARC case correctly.
1293 	 */
1294 	if (!igb_find_mac_address(igb)) {
1295 		igb_log(igb, IGB_LOG_ERROR, "Failed to get the mac address");
1296 		goto init_mac_fail;
1297 	}
1298 
1299 	/* Validate mac address */
1300 	if (!is_valid_mac_addr(hw->mac.addr)) {
1301 		igb_log(igb, IGB_LOG_ERROR, "Invalid mac address");
1302 		goto init_mac_fail;
1303 	}
1304 
1305 	return (IGB_SUCCESS);
1306 
1307 init_mac_fail:
1308 	return (IGB_FAILURE);
1309 }
1310 
1311 /*
1312  * igb_init_adapter - Initialize the adapter
1313  */
1314 static int
1315 igb_init_adapter(igb_t *igb)
1316 {
1317 	struct e1000_hw *hw = &igb->hw;
1318 	uint32_t pba;
1319 	int oemid[2];
1320 	uint16_t nvmword;
1321 	uint32_t hwm;
1322 	uint32_t default_mtu;
1323 	u8 pbanum[E1000_PBANUM_LENGTH];
1324 	char eepromver[5];	/* f.ff */
1325 	int i;
1326 
1327 	ASSERT(mutex_owned(&igb->gen_lock));
1328 
1329 	/*
1330 	 * In order to obtain the default MAC address, this will reset the
1331 	 * adapter and validate the NVM that the address and many other
1332 	 * default settings come from.
1333 	 */
1334 	if (igb_init_mac_address(igb) != IGB_SUCCESS) {
1335 		igb_log(igb, IGB_LOG_ERROR, "Failed to initialize MAC address");
1336 		goto init_adapter_fail;
1337 	}
1338 
1339 	/*
1340 	 * Packet Buffer Allocation (PBA)
1341 	 * Writing PBA sets the receive portion of the buffer
1342 	 * the remainder is used for the transmit buffer.
1343 	 */
1344 	switch (hw->mac.type) {
1345 	case e1000_82575:
1346 		pba = E1000_PBA_32K;
1347 		break;
1348 	case e1000_82576:
1349 		pba = E1000_READ_REG(hw, E1000_RXPBS);
1350 		pba &= E1000_RXPBS_SIZE_MASK_82576;
1351 		break;
1352 	case e1000_82580:
1353 	case e1000_i350:
1354 	case e1000_i354:
1355 		pba = E1000_READ_REG(hw, E1000_RXPBS);
1356 		pba = e1000_rxpbs_adjust_82580(pba);
1357 		break;
1358 	case e1000_i210:
1359 	case e1000_i211:
1360 		pba = E1000_PBA_34K;
1361 	default:
1362 		break;
1363 	}
1364 
1365 	/* Special needs in case of Jumbo frames */
1366 	default_mtu = igb_get_prop(igb, PROP_DEFAULT_MTU,
1367 	    MIN_MTU, MAX_MTU, DEFAULT_MTU);
1368 	if ((hw->mac.type == e1000_82575) && (default_mtu > ETHERMTU)) {
1369 		u32 tx_space, min_tx, min_rx;
1370 		pba = E1000_READ_REG(hw, E1000_PBA);
1371 		tx_space = pba >> 16;
1372 		pba &= 0xffff;
1373 		min_tx = (igb->max_frame_size +
1374 		    sizeof (struct e1000_tx_desc) - ETHERNET_FCS_SIZE) * 2;
1375 		min_tx = roundup(min_tx, 1024);
1376 		min_tx >>= 10;
1377 		min_rx = igb->max_frame_size;
1378 		min_rx = roundup(min_rx, 1024);
1379 		min_rx >>= 10;
1380 		if (tx_space < min_tx &&
1381 		    ((min_tx - tx_space) < pba)) {
1382 			pba = pba - (min_tx - tx_space);
1383 			/*
1384 			 * if short on rx space, rx wins
1385 			 * and must trump tx adjustment
1386 			 */
1387 			if (pba < min_rx)
1388 				pba = min_rx;
1389 		}
1390 		E1000_WRITE_REG(hw, E1000_PBA, pba);
1391 	}
1392 
1393 	DEBUGOUT1("igb_init: pba=%dK", pba);
1394 
1395 	/*
1396 	 * These parameters control the automatic generation (Tx) and
1397 	 * response (Rx) to Ethernet PAUSE frames.
1398 	 * - High water mark should allow for at least two frames to be
1399 	 *   received after sending an XOFF.
1400 	 * - Low water mark works best when it is very near the high water mark.
1401 	 *   This allows the receiver to restart by sending XON when it has
1402 	 *   drained a bit.
1403 	 */
1404 	hwm = min(((pba << 10) * 9 / 10),
1405 	    ((pba << 10) - 2 * igb->max_frame_size));
1406 
1407 	if (hw->mac.type < e1000_82576) {
1408 		hw->fc.high_water = hwm & 0xFFF8;  /* 8-byte granularity */
1409 		hw->fc.low_water = hw->fc.high_water - 8;
1410 	} else {
1411 		hw->fc.high_water = hwm & 0xFFF0;  /* 16-byte granularity */
1412 		hw->fc.low_water = hw->fc.high_water - 16;
1413 	}
1414 
1415 	hw->fc.pause_time = E1000_FC_PAUSE_TIME;
1416 	hw->fc.send_xon = B_TRUE;
1417 
1418 	(void) e1000_validate_mdi_setting(hw);
1419 
1420 	/*
1421 	 * Reset the chipset hardware the second time to put PBA settings
1422 	 * into effect.
1423 	 */
1424 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1425 		igb_log(igb, IGB_LOG_ERROR, "Second reset failed");
1426 		goto init_adapter_fail;
1427 	}
1428 
1429 	/*
1430 	 * Don't wait for auto-negotiation to complete
1431 	 */
1432 	hw->phy.autoneg_wait_to_complete = B_FALSE;
1433 
1434 	/*
1435 	 * Copper options
1436 	 */
1437 	if (hw->phy.media_type == e1000_media_type_copper) {
1438 		hw->phy.mdix = 0;	/* AUTO_ALL_MODES */
1439 		hw->phy.disable_polarity_correction = B_FALSE;
1440 		hw->phy.ms_type = e1000_ms_hw_default; /* E1000_MASTER_SLAVE */
1441 	}
1442 
1443 	/*
1444 	 * Initialize link settings
1445 	 */
1446 	(void) igb_setup_link(igb, B_FALSE);
1447 
1448 	/*
1449 	 * Configure/Initialize hardware
1450 	 */
1451 	if (e1000_init_hw(hw) != E1000_SUCCESS) {
1452 		igb_log(igb, IGB_LOG_ERROR, "Failed to initialize hardware");
1453 		goto init_adapter_fail;
1454 	}
1455 
1456 	/*
1457 	 *  Start the link setup timer
1458 	 */
1459 	igb_start_link_timer(igb);
1460 
1461 	/*
1462 	 * Disable wakeup control by default
1463 	 */
1464 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1465 
1466 	/*
1467 	 * Record phy info in hw struct
1468 	 */
1469 	(void) e1000_get_phy_info(hw);
1470 
1471 	/*
1472 	 * Make sure driver has control
1473 	 */
1474 	igb_get_driver_control(hw);
1475 
1476 	/*
1477 	 * Restore LED settings to the default from EEPROM
1478 	 * to meet the standard for Sun platforms.
1479 	 */
1480 	(void) e1000_cleanup_led(hw);
1481 
1482 	/*
1483 	 * Setup MSI-X interrupts
1484 	 */
1485 	if (igb->intr_type == DDI_INTR_TYPE_MSIX)
1486 		igb->capab->setup_msix(igb);
1487 
1488 	/*
1489 	 * Initialize unicast addresses.
1490 	 */
1491 	igb_init_unicst(igb);
1492 
1493 	/*
1494 	 * Setup and initialize the mctable structures.
1495 	 */
1496 	igb_setup_multicst(igb);
1497 
1498 	/*
1499 	 * Set interrupt throttling rate
1500 	 */
1501 	for (i = 0; i < igb->intr_cnt; i++)
1502 		E1000_WRITE_REG(hw, E1000_EITR(i), igb->intr_throttling[i]);
1503 
1504 	/*
1505 	 * Read identifying information and place in devinfo.
1506 	 */
1507 	nvmword = 0xffff;
1508 	(void) e1000_read_nvm(&igb->hw, NVM_OEM_OFFSET_0, 1, &nvmword);
1509 	oemid[0] = (int)nvmword;
1510 	(void) e1000_read_nvm(&igb->hw, NVM_OEM_OFFSET_1, 1, &nvmword);
1511 	oemid[1] = (int)nvmword;
1512 	(void) ddi_prop_update_int_array(DDI_DEV_T_NONE, igb->dip,
1513 	    "oem-identifier", oemid, 2);
1514 
1515 	pbanum[0] = '\0';
1516 	(void) e1000_read_pba_string(&igb->hw, pbanum, sizeof (pbanum));
1517 	if (*pbanum != '\0') {
1518 		(void) ddi_prop_update_string(DDI_DEV_T_NONE, igb->dip,
1519 		    "printed-board-assembly", (char *)pbanum);
1520 	}
1521 
1522 	nvmword = 0xffff;
1523 	(void) e1000_read_nvm(&igb->hw, NVM_VERSION, 1, &nvmword);
1524 	if ((nvmword & 0xf00) == 0) {
1525 		(void) snprintf(eepromver, sizeof (eepromver), "%x.%x",
1526 		    (nvmword & 0xf000) >> 12, (nvmword & 0xff));
1527 		(void) ddi_prop_update_string(DDI_DEV_T_NONE, igb->dip,
1528 		    "nvm-version", eepromver);
1529 	}
1530 
1531 	/*
1532 	 * Save the state of the phy
1533 	 */
1534 	igb_get_phy_state(igb);
1535 
1536 	igb_param_sync(igb);
1537 
1538 	return (IGB_SUCCESS);
1539 
1540 init_adapter_fail:
1541 	/*
1542 	 * Reset PHY if possible
1543 	 */
1544 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
1545 		(void) e1000_phy_hw_reset(hw);
1546 
1547 	return (IGB_FAILURE);
1548 }
1549 
1550 /*
1551  * igb_stop_adapter - Stop the adapter
1552  */
1553 static void
1554 igb_stop_adapter(igb_t *igb)
1555 {
1556 	struct e1000_hw *hw = &igb->hw;
1557 
1558 	ASSERT(mutex_owned(&igb->gen_lock));
1559 
1560 	/* Stop the link setup timer */
1561 	igb_stop_link_timer(igb);
1562 
1563 	/* Tell firmware driver is no longer in control */
1564 	igb_release_driver_control(hw);
1565 
1566 	/*
1567 	 * Reset the chipset
1568 	 */
1569 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1570 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1571 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1572 	}
1573 
1574 	/*
1575 	 * e1000_phy_hw_reset is not needed here, MAC reset above is sufficient
1576 	 */
1577 }
1578 
1579 /*
1580  * igb_reset - Reset the chipset and restart the driver.
1581  *
1582  * It involves stopping and re-starting the chipset,
1583  * and re-configuring the rx/tx rings.
1584  */
1585 static int
1586 igb_reset(igb_t *igb)
1587 {
1588 	int i;
1589 
1590 	mutex_enter(&igb->gen_lock);
1591 
1592 	ASSERT(igb->igb_state & IGB_STARTED);
1593 	atomic_and_32(&igb->igb_state, ~IGB_STARTED);
1594 
1595 	/*
1596 	 * Disable the adapter interrupts to stop any rx/tx activities
1597 	 * before draining pending data and resetting hardware.
1598 	 */
1599 	igb_disable_adapter_interrupts(igb);
1600 
1601 	/*
1602 	 * Drain the pending transmit packets
1603 	 */
1604 	(void) igb_tx_drain(igb);
1605 
1606 	for (i = 0; i < igb->num_rx_rings; i++)
1607 		mutex_enter(&igb->rx_rings[i].rx_lock);
1608 	for (i = 0; i < igb->num_tx_rings; i++)
1609 		mutex_enter(&igb->tx_rings[i].tx_lock);
1610 
1611 	/*
1612 	 * Stop the adapter
1613 	 */
1614 	igb_stop_adapter(igb);
1615 
1616 	/*
1617 	 * Clean the pending tx data/resources
1618 	 */
1619 	igb_tx_clean(igb);
1620 
1621 	/*
1622 	 * Start the adapter
1623 	 */
1624 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1625 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1626 		goto reset_failure;
1627 	}
1628 
1629 	/*
1630 	 * Setup the rx/tx rings
1631 	 */
1632 	igb->tx_ring_init = B_FALSE;
1633 	igb_setup_rings(igb);
1634 
1635 	atomic_and_32(&igb->igb_state, ~(IGB_ERROR | IGB_STALL));
1636 
1637 	/*
1638 	 * Enable adapter interrupts
1639 	 * The interrupts must be enabled after the driver state is START
1640 	 */
1641 	igb->capab->enable_intr(igb);
1642 
1643 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1644 		goto reset_failure;
1645 
1646 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1647 		goto reset_failure;
1648 
1649 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1650 		mutex_exit(&igb->tx_rings[i].tx_lock);
1651 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1652 		mutex_exit(&igb->rx_rings[i].rx_lock);
1653 
1654 	atomic_or_32(&igb->igb_state, IGB_STARTED);
1655 
1656 	mutex_exit(&igb->gen_lock);
1657 
1658 	return (IGB_SUCCESS);
1659 
1660 reset_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 	mutex_exit(&igb->gen_lock);
1667 
1668 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1669 
1670 	return (IGB_FAILURE);
1671 }
1672 
1673 /*
1674  * igb_tx_clean - Clean the pending transmit packets and DMA resources
1675  */
1676 static void
1677 igb_tx_clean(igb_t *igb)
1678 {
1679 	igb_tx_ring_t *tx_ring;
1680 	tx_control_block_t *tcb;
1681 	link_list_t pending_list;
1682 	uint32_t desc_num;
1683 	int i, j;
1684 
1685 	LINK_LIST_INIT(&pending_list);
1686 
1687 	for (i = 0; i < igb->num_tx_rings; i++) {
1688 		tx_ring = &igb->tx_rings[i];
1689 
1690 		mutex_enter(&tx_ring->recycle_lock);
1691 
1692 		/*
1693 		 * Clean the pending tx data - the pending packets in the
1694 		 * work_list that have no chances to be transmitted again.
1695 		 *
1696 		 * We must ensure the chipset is stopped or the link is down
1697 		 * before cleaning the transmit packets.
1698 		 */
1699 		desc_num = 0;
1700 		for (j = 0; j < tx_ring->ring_size; j++) {
1701 			tcb = tx_ring->work_list[j];
1702 			if (tcb != NULL) {
1703 				desc_num += tcb->desc_num;
1704 
1705 				tx_ring->work_list[j] = NULL;
1706 
1707 				igb_free_tcb(tcb);
1708 
1709 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1710 			}
1711 		}
1712 
1713 		if (desc_num > 0) {
1714 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1715 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1716 
1717 			/*
1718 			 * Reset the head and tail pointers of the tbd ring;
1719 			 * Reset the head write-back if it is enabled.
1720 			 */
1721 			tx_ring->tbd_head = 0;
1722 			tx_ring->tbd_tail = 0;
1723 			if (igb->tx_head_wb_enable)
1724 				*tx_ring->tbd_head_wb = 0;
1725 
1726 			E1000_WRITE_REG(&igb->hw, E1000_TDH(tx_ring->index), 0);
1727 			E1000_WRITE_REG(&igb->hw, E1000_TDT(tx_ring->index), 0);
1728 		}
1729 
1730 		mutex_exit(&tx_ring->recycle_lock);
1731 
1732 		/*
1733 		 * Add the tx control blocks in the pending list to
1734 		 * the free list.
1735 		 */
1736 		igb_put_free_list(tx_ring, &pending_list);
1737 	}
1738 }
1739 
1740 /*
1741  * igb_tx_drain - Drain the tx rings to allow pending packets to be transmitted
1742  */
1743 static boolean_t
1744 igb_tx_drain(igb_t *igb)
1745 {
1746 	igb_tx_ring_t *tx_ring;
1747 	boolean_t done;
1748 	int i, j;
1749 
1750 	/*
1751 	 * Wait for a specific time to allow pending tx packets
1752 	 * to be transmitted.
1753 	 *
1754 	 * Check the counter tbd_free to see if transmission is done.
1755 	 * No lock protection is needed here.
1756 	 *
1757 	 * Return B_TRUE if all pending packets have been transmitted;
1758 	 * Otherwise return B_FALSE;
1759 	 */
1760 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1761 
1762 		done = B_TRUE;
1763 		for (j = 0; j < igb->num_tx_rings; j++) {
1764 			tx_ring = &igb->tx_rings[j];
1765 			done = done &&
1766 			    (tx_ring->tbd_free == tx_ring->ring_size);
1767 		}
1768 
1769 		if (done)
1770 			break;
1771 
1772 		msec_delay(1);
1773 	}
1774 
1775 	return (done);
1776 }
1777 
1778 /*
1779  * igb_rx_drain - Wait for all rx buffers to be released by upper layer
1780  */
1781 static boolean_t
1782 igb_rx_drain(igb_t *igb)
1783 {
1784 	boolean_t done;
1785 	int i;
1786 
1787 	/*
1788 	 * Polling the rx free list to check if those rx buffers held by
1789 	 * the upper layer are released.
1790 	 *
1791 	 * Check the counter rcb_free to see if all pending buffers are
1792 	 * released. No lock protection is needed here.
1793 	 *
1794 	 * Return B_TRUE if all pending buffers have been released;
1795 	 * Otherwise return B_FALSE;
1796 	 */
1797 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1798 		done = (igb->rcb_pending == 0);
1799 
1800 		if (done)
1801 			break;
1802 
1803 		msec_delay(1);
1804 	}
1805 
1806 	return (done);
1807 }
1808 
1809 /*
1810  * igb_start - Start the driver/chipset
1811  */
1812 int
1813 igb_start(igb_t *igb, boolean_t alloc_buffer)
1814 {
1815 	int i;
1816 
1817 	ASSERT(mutex_owned(&igb->gen_lock));
1818 
1819 	if (alloc_buffer) {
1820 		if (igb_alloc_rx_data(igb) != IGB_SUCCESS) {
1821 			igb_log(igb, IGB_LOG_ERROR,
1822 			    "Failed to allocate software receive rings");
1823 			return (IGB_FAILURE);
1824 		}
1825 
1826 		/* Allocate buffers for all the rx/tx rings */
1827 		if (igb_alloc_dma(igb) != IGB_SUCCESS) {
1828 			igb_log(igb, IGB_LOG_ERROR,
1829 			    "Failed to allocate DMA resource");
1830 			return (IGB_FAILURE);
1831 		}
1832 
1833 		igb->tx_ring_init = B_TRUE;
1834 	} else {
1835 		igb->tx_ring_init = B_FALSE;
1836 	}
1837 
1838 	for (i = 0; i < igb->num_rx_rings; i++)
1839 		mutex_enter(&igb->rx_rings[i].rx_lock);
1840 	for (i = 0; i < igb->num_tx_rings; i++)
1841 		mutex_enter(&igb->tx_rings[i].tx_lock);
1842 
1843 	/*
1844 	 * Start the adapter
1845 	 */
1846 	if ((igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) == 0) {
1847 		if (igb_init_adapter(igb) != IGB_SUCCESS) {
1848 			igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1849 			goto start_failure;
1850 		}
1851 		igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
1852 	}
1853 
1854 	/*
1855 	 * Setup the rx/tx rings
1856 	 */
1857 	igb_setup_rings(igb);
1858 
1859 	/*
1860 	 * Enable adapter interrupts
1861 	 * The interrupts must be enabled after the driver state is START
1862 	 */
1863 	igb->capab->enable_intr(igb);
1864 
1865 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1866 		goto start_failure;
1867 
1868 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1869 		goto start_failure;
1870 
1871 	if (igb->hw.mac.type == e1000_i350)
1872 		(void) e1000_set_eee_i350(&igb->hw, B_FALSE, B_FALSE);
1873 	else if (igb->hw.mac.type == e1000_i354)
1874 		(void) e1000_set_eee_i354(&igb->hw, B_FALSE, B_FALSE);
1875 
1876 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1877 		mutex_exit(&igb->tx_rings[i].tx_lock);
1878 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1879 		mutex_exit(&igb->rx_rings[i].rx_lock);
1880 
1881 	return (IGB_SUCCESS);
1882 
1883 start_failure:
1884 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1885 		mutex_exit(&igb->tx_rings[i].tx_lock);
1886 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1887 		mutex_exit(&igb->rx_rings[i].rx_lock);
1888 
1889 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1890 
1891 	return (IGB_FAILURE);
1892 }
1893 
1894 /*
1895  * igb_stop - Stop the driver/chipset
1896  */
1897 void
1898 igb_stop(igb_t *igb, boolean_t free_buffer)
1899 {
1900 	int i;
1901 
1902 	ASSERT(mutex_owned(&igb->gen_lock));
1903 
1904 	igb->attach_progress &= ~ATTACH_PROGRESS_INIT_ADAPTER;
1905 
1906 	/*
1907 	 * Disable the adapter interrupts
1908 	 */
1909 	igb_disable_adapter_interrupts(igb);
1910 
1911 	/*
1912 	 * Drain the pending tx packets
1913 	 */
1914 	(void) igb_tx_drain(igb);
1915 
1916 	for (i = 0; i < igb->num_rx_rings; i++)
1917 		mutex_enter(&igb->rx_rings[i].rx_lock);
1918 	for (i = 0; i < igb->num_tx_rings; i++)
1919 		mutex_enter(&igb->tx_rings[i].tx_lock);
1920 
1921 	/*
1922 	 * Stop the adapter
1923 	 */
1924 	igb_stop_adapter(igb);
1925 
1926 	/*
1927 	 * Clean the pending tx data/resources
1928 	 */
1929 	igb_tx_clean(igb);
1930 
1931 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1932 		mutex_exit(&igb->tx_rings[i].tx_lock);
1933 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1934 		mutex_exit(&igb->rx_rings[i].rx_lock);
1935 
1936 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1937 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1938 
1939 	if (igb->link_state == LINK_STATE_UP) {
1940 		igb->link_state = LINK_STATE_UNKNOWN;
1941 		mac_link_update(igb->mac_hdl, igb->link_state);
1942 	}
1943 
1944 	if (free_buffer) {
1945 		/*
1946 		 * Release the DMA/memory resources of rx/tx rings
1947 		 */
1948 		igb_free_dma(igb);
1949 		igb_free_rx_data(igb);
1950 	}
1951 }
1952 
1953 /*
1954  * igb_alloc_rings - Allocate memory space for rx/tx rings
1955  */
1956 static int
1957 igb_alloc_rings(igb_t *igb)
1958 {
1959 	/*
1960 	 * Allocate memory space for rx rings
1961 	 */
1962 	igb->rx_rings = kmem_zalloc(
1963 	    sizeof (igb_rx_ring_t) * igb->num_rx_rings,
1964 	    KM_NOSLEEP);
1965 
1966 	if (igb->rx_rings == NULL) {
1967 		return (IGB_FAILURE);
1968 	}
1969 
1970 	/*
1971 	 * Allocate memory space for tx rings
1972 	 */
1973 	igb->tx_rings = kmem_zalloc(
1974 	    sizeof (igb_tx_ring_t) * igb->num_tx_rings,
1975 	    KM_NOSLEEP);
1976 
1977 	if (igb->tx_rings == NULL) {
1978 		kmem_free(igb->rx_rings,
1979 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1980 		igb->rx_rings = NULL;
1981 		return (IGB_FAILURE);
1982 	}
1983 
1984 	/*
1985 	 * Allocate memory space for rx ring groups
1986 	 */
1987 	igb->rx_groups = kmem_zalloc(
1988 	    sizeof (igb_rx_group_t) * igb->num_rx_groups,
1989 	    KM_NOSLEEP);
1990 
1991 	if (igb->rx_groups == NULL) {
1992 		kmem_free(igb->rx_rings,
1993 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1994 		kmem_free(igb->tx_rings,
1995 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1996 		igb->rx_rings = NULL;
1997 		igb->tx_rings = NULL;
1998 		return (IGB_FAILURE);
1999 	}
2000 
2001 	return (IGB_SUCCESS);
2002 }
2003 
2004 /*
2005  * igb_free_rings - Free the memory space of rx/tx rings.
2006  */
2007 static void
2008 igb_free_rings(igb_t *igb)
2009 {
2010 	if (igb->rx_rings != NULL) {
2011 		kmem_free(igb->rx_rings,
2012 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
2013 		igb->rx_rings = NULL;
2014 	}
2015 
2016 	if (igb->tx_rings != NULL) {
2017 		kmem_free(igb->tx_rings,
2018 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
2019 		igb->tx_rings = NULL;
2020 	}
2021 
2022 	if (igb->rx_groups != NULL) {
2023 		kmem_free(igb->rx_groups,
2024 		    sizeof (igb_rx_group_t) * igb->num_rx_groups);
2025 		igb->rx_groups = NULL;
2026 	}
2027 }
2028 
2029 static int
2030 igb_alloc_rx_data(igb_t *igb)
2031 {
2032 	igb_rx_ring_t *rx_ring;
2033 	int i;
2034 
2035 	for (i = 0; i < igb->num_rx_rings; i++) {
2036 		rx_ring = &igb->rx_rings[i];
2037 		if (igb_alloc_rx_ring_data(rx_ring) != IGB_SUCCESS)
2038 			goto alloc_rx_rings_failure;
2039 	}
2040 	return (IGB_SUCCESS);
2041 
2042 alloc_rx_rings_failure:
2043 	igb_free_rx_data(igb);
2044 	return (IGB_FAILURE);
2045 }
2046 
2047 static void
2048 igb_free_rx_data(igb_t *igb)
2049 {
2050 	igb_rx_ring_t *rx_ring;
2051 	igb_rx_data_t *rx_data;
2052 	int i;
2053 
2054 	for (i = 0; i < igb->num_rx_rings; i++) {
2055 		rx_ring = &igb->rx_rings[i];
2056 
2057 		mutex_enter(&igb->rx_pending_lock);
2058 		rx_data = rx_ring->rx_data;
2059 
2060 		if (rx_data != NULL) {
2061 			rx_data->flag |= IGB_RX_STOPPED;
2062 
2063 			if (rx_data->rcb_pending == 0) {
2064 				igb_free_rx_ring_data(rx_data);
2065 				rx_ring->rx_data = NULL;
2066 			}
2067 		}
2068 
2069 		mutex_exit(&igb->rx_pending_lock);
2070 	}
2071 }
2072 
2073 /*
2074  * igb_setup_rings - Setup rx/tx rings
2075  */
2076 static void
2077 igb_setup_rings(igb_t *igb)
2078 {
2079 	/*
2080 	 * Setup the rx/tx rings, including the following:
2081 	 *
2082 	 * 1. Setup the descriptor ring and the control block buffers;
2083 	 * 2. Initialize necessary registers for receive/transmit;
2084 	 * 3. Initialize software pointers/parameters for receive/transmit;
2085 	 */
2086 	igb_setup_rx(igb);
2087 
2088 	igb_setup_tx(igb);
2089 }
2090 
2091 static void
2092 igb_setup_rx_ring(igb_rx_ring_t *rx_ring)
2093 {
2094 	igb_t *igb = rx_ring->igb;
2095 	igb_rx_data_t *rx_data = rx_ring->rx_data;
2096 	struct e1000_hw *hw = &igb->hw;
2097 	rx_control_block_t *rcb;
2098 	union e1000_adv_rx_desc	*rbd;
2099 	uint32_t size;
2100 	uint32_t buf_low;
2101 	uint32_t buf_high;
2102 	uint32_t rxdctl;
2103 	int i;
2104 
2105 	ASSERT(mutex_owned(&rx_ring->rx_lock));
2106 	ASSERT(mutex_owned(&igb->gen_lock));
2107 
2108 	/*
2109 	 * Initialize descriptor ring with buffer addresses
2110 	 */
2111 	for (i = 0; i < igb->rx_ring_size; i++) {
2112 		rcb = rx_data->work_list[i];
2113 		rbd = &rx_data->rbd_ring[i];
2114 
2115 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
2116 		rbd->read.hdr_addr = 0;
2117 	}
2118 
2119 	/*
2120 	 * Initialize the base address registers
2121 	 */
2122 	buf_low = (uint32_t)rx_data->rbd_area.dma_address;
2123 	buf_high = (uint32_t)(rx_data->rbd_area.dma_address >> 32);
2124 	E1000_WRITE_REG(hw, E1000_RDBAH(rx_ring->index), buf_high);
2125 	E1000_WRITE_REG(hw, E1000_RDBAL(rx_ring->index), buf_low);
2126 
2127 	/*
2128 	 * Initialize the length register
2129 	 */
2130 	size = rx_data->ring_size * sizeof (union e1000_adv_rx_desc);
2131 	E1000_WRITE_REG(hw, E1000_RDLEN(rx_ring->index), size);
2132 
2133 	/*
2134 	 * Initialize buffer size & descriptor type
2135 	 */
2136 	E1000_WRITE_REG(hw, E1000_SRRCTL(rx_ring->index),
2137 	    ((igb->rx_buf_size >> E1000_SRRCTL_BSIZEPKT_SHIFT) |
2138 	    E1000_SRRCTL_DESCTYPE_ADV_ONEBUF));
2139 
2140 	/*
2141 	 * Setup the Receive Descriptor Control Register (RXDCTL)
2142 	 */
2143 	rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(rx_ring->index));
2144 	rxdctl &= igb->capab->rxdctl_mask;
2145 	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2146 	rxdctl |= 16;		/* pthresh */
2147 	rxdctl |= 8 << 8;	/* hthresh */
2148 	rxdctl |= 1 << 16;	/* wthresh */
2149 	E1000_WRITE_REG(hw, E1000_RXDCTL(rx_ring->index), rxdctl);
2150 
2151 	rx_data->rbd_next = 0;
2152 }
2153 
2154 static void
2155 igb_setup_rx(igb_t *igb)
2156 {
2157 	igb_rx_ring_t *rx_ring;
2158 	igb_rx_data_t *rx_data;
2159 	igb_rx_group_t *rx_group;
2160 	struct e1000_hw *hw = &igb->hw;
2161 	uint32_t rctl, rxcsum;
2162 	uint32_t ring_per_group;
2163 	int i;
2164 
2165 	/*
2166 	 * Setup the Receive Control Register (RCTL), and enable the
2167 	 * receiver. The initial configuration is to: enable the receiver,
2168 	 * accept broadcasts, discard bad packets, accept long packets,
2169 	 * disable VLAN filter checking, and set receive buffer size to
2170 	 * 2k.  For 82575, also set the receive descriptor minimum
2171 	 * threshold size to 1/2 the ring.
2172 	 */
2173 	rctl = E1000_READ_REG(hw, E1000_RCTL);
2174 
2175 	/*
2176 	 * Clear the field used for wakeup control.  This driver doesn't do
2177 	 * wakeup but leave this here for completeness.
2178 	 */
2179 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2180 	rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2181 
2182 	rctl |= (E1000_RCTL_EN |	/* Enable Receive Unit */
2183 	    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
2184 	    E1000_RCTL_LPE |		/* Large Packet Enable */
2185 					/* Multicast filter offset */
2186 	    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
2187 	    E1000_RCTL_RDMTS_HALF |	/* rx descriptor threshold */
2188 	    E1000_RCTL_SECRC);		/* Strip Ethernet CRC */
2189 
2190 	for (i = 0; i < igb->num_rx_groups; i++) {
2191 		rx_group = &igb->rx_groups[i];
2192 		rx_group->index = i;
2193 		rx_group->igb = igb;
2194 	}
2195 
2196 	/*
2197 	 * Set up all rx descriptor rings - must be called before receive unit
2198 	 * enabled.
2199 	 */
2200 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2201 	for (i = 0; i < igb->num_rx_rings; i++) {
2202 		rx_ring = &igb->rx_rings[i];
2203 		igb_setup_rx_ring(rx_ring);
2204 
2205 		/*
2206 		 * Map a ring to a group by assigning a group index
2207 		 */
2208 		rx_ring->group_index = i / ring_per_group;
2209 	}
2210 
2211 	/*
2212 	 * Setup the Rx Long Packet Max Length register
2213 	 */
2214 	E1000_WRITE_REG(hw, E1000_RLPML, igb->max_frame_size);
2215 
2216 	/*
2217 	 * Hardware checksum settings
2218 	 */
2219 	if (igb->rx_hcksum_enable) {
2220 		rxcsum =
2221 		    E1000_RXCSUM_TUOFL |	/* TCP/UDP checksum */
2222 		    E1000_RXCSUM_IPOFL;		/* IP checksum */
2223 
2224 		E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2225 	}
2226 
2227 	/*
2228 	 * Setup classify and RSS for multiple receive queues
2229 	 */
2230 	switch (igb->vmdq_mode) {
2231 	case E1000_VMDQ_OFF:
2232 		/*
2233 		 * One ring group, only RSS is needed when more than
2234 		 * one ring enabled.
2235 		 */
2236 		if (igb->num_rx_rings > 1)
2237 			igb_setup_rss(igb);
2238 		break;
2239 	case E1000_VMDQ_MAC:
2240 		/*
2241 		 * Multiple groups, each group has one ring,
2242 		 * only the MAC classification is needed.
2243 		 */
2244 		igb_setup_mac_classify(igb);
2245 		break;
2246 	case E1000_VMDQ_MAC_RSS:
2247 		/*
2248 		 * Multiple groups and multiple rings, both
2249 		 * MAC classification and RSS are needed.
2250 		 */
2251 		igb_setup_mac_rss_classify(igb);
2252 		break;
2253 	}
2254 
2255 	/*
2256 	 * Enable the receive unit - must be done after all
2257 	 * the rx setup above.
2258 	 */
2259 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2260 
2261 	/*
2262 	 * Initialize all adapter ring head & tail pointers - must
2263 	 * be done after receive unit is enabled
2264 	 */
2265 	for (i = 0; i < igb->num_rx_rings; i++) {
2266 		rx_ring = &igb->rx_rings[i];
2267 		rx_data = rx_ring->rx_data;
2268 		E1000_WRITE_REG(hw, E1000_RDH(i), 0);
2269 		E1000_WRITE_REG(hw, E1000_RDT(i), rx_data->ring_size - 1);
2270 	}
2271 
2272 	/*
2273 	 * 82575 with manageability enabled needs a special flush to make
2274 	 * sure the fifos start clean.
2275 	 */
2276 	if ((hw->mac.type == e1000_82575) &&
2277 	    (E1000_READ_REG(hw, E1000_MANC) & E1000_MANC_RCV_TCO_EN)) {
2278 		e1000_rx_fifo_flush_82575(hw);
2279 	}
2280 }
2281 
2282 static void
2283 igb_setup_tx_ring(igb_tx_ring_t *tx_ring)
2284 {
2285 	igb_t *igb = tx_ring->igb;
2286 	struct e1000_hw *hw = &igb->hw;
2287 	uint32_t size;
2288 	uint32_t buf_low;
2289 	uint32_t buf_high;
2290 	uint32_t reg_val;
2291 
2292 	ASSERT(mutex_owned(&tx_ring->tx_lock));
2293 	ASSERT(mutex_owned(&igb->gen_lock));
2294 
2295 
2296 	/*
2297 	 * Initialize the length register
2298 	 */
2299 	size = tx_ring->ring_size * sizeof (union e1000_adv_tx_desc);
2300 	E1000_WRITE_REG(hw, E1000_TDLEN(tx_ring->index), size);
2301 
2302 	/*
2303 	 * Initialize the base address registers
2304 	 */
2305 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
2306 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
2307 	E1000_WRITE_REG(hw, E1000_TDBAL(tx_ring->index), buf_low);
2308 	E1000_WRITE_REG(hw, E1000_TDBAH(tx_ring->index), buf_high);
2309 
2310 	/*
2311 	 * Setup head & tail pointers
2312 	 */
2313 	E1000_WRITE_REG(hw, E1000_TDH(tx_ring->index), 0);
2314 	E1000_WRITE_REG(hw, E1000_TDT(tx_ring->index), 0);
2315 
2316 	/*
2317 	 * Setup head write-back
2318 	 */
2319 	if (igb->tx_head_wb_enable) {
2320 		/*
2321 		 * The memory of the head write-back is allocated using
2322 		 * the extra tbd beyond the tail of the tbd ring.
2323 		 */
2324 		tx_ring->tbd_head_wb = (uint32_t *)
2325 		    ((uintptr_t)tx_ring->tbd_area.address + size);
2326 		*tx_ring->tbd_head_wb = 0;
2327 
2328 		buf_low = (uint32_t)
2329 		    (tx_ring->tbd_area.dma_address + size);
2330 		buf_high = (uint32_t)
2331 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
2332 
2333 		/* Set the head write-back enable bit */
2334 		buf_low |= E1000_TX_HEAD_WB_ENABLE;
2335 
2336 		E1000_WRITE_REG(hw, E1000_TDWBAL(tx_ring->index), buf_low);
2337 		E1000_WRITE_REG(hw, E1000_TDWBAH(tx_ring->index), buf_high);
2338 
2339 		/*
2340 		 * Turn off relaxed ordering for head write back or it will
2341 		 * cause problems with the tx recycling
2342 		 */
2343 		reg_val = E1000_READ_REG(hw,
2344 		    E1000_DCA_TXCTRL(tx_ring->index));
2345 		reg_val &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
2346 		E1000_WRITE_REG(hw,
2347 		    E1000_DCA_TXCTRL(tx_ring->index), reg_val);
2348 	} else {
2349 		tx_ring->tbd_head_wb = NULL;
2350 	}
2351 
2352 	tx_ring->tbd_head = 0;
2353 	tx_ring->tbd_tail = 0;
2354 	tx_ring->tbd_free = tx_ring->ring_size;
2355 
2356 	if (igb->tx_ring_init == B_TRUE) {
2357 		tx_ring->tcb_head = 0;
2358 		tx_ring->tcb_tail = 0;
2359 		tx_ring->tcb_free = tx_ring->free_list_size;
2360 	}
2361 
2362 	/*
2363 	 * Enable TXDCTL per queue
2364 	 */
2365 	reg_val = E1000_READ_REG(hw, E1000_TXDCTL(tx_ring->index));
2366 	reg_val |= E1000_TXDCTL_QUEUE_ENABLE;
2367 	E1000_WRITE_REG(hw, E1000_TXDCTL(tx_ring->index), reg_val);
2368 
2369 	/*
2370 	 * Initialize hardware checksum offload settings
2371 	 */
2372 	bzero(&tx_ring->tx_context, sizeof (tx_context_t));
2373 }
2374 
2375 static void
2376 igb_setup_tx(igb_t *igb)
2377 {
2378 	igb_tx_ring_t *tx_ring;
2379 	struct e1000_hw *hw = &igb->hw;
2380 	uint32_t reg_val;
2381 	int i;
2382 
2383 	for (i = 0; i < igb->num_tx_rings; i++) {
2384 		tx_ring = &igb->tx_rings[i];
2385 		igb_setup_tx_ring(tx_ring);
2386 	}
2387 
2388 	/*
2389 	 * Setup the Transmit Control Register (TCTL)
2390 	 */
2391 	reg_val = E1000_READ_REG(hw, E1000_TCTL);
2392 	reg_val &= ~E1000_TCTL_CT;
2393 	reg_val |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2394 	    (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2395 
2396 	/* Enable transmits */
2397 	reg_val |= E1000_TCTL_EN;
2398 
2399 	E1000_WRITE_REG(hw, E1000_TCTL, reg_val);
2400 }
2401 
2402 /*
2403  * igb_setup_rss - Setup receive-side scaling feature
2404  */
2405 static void
2406 igb_setup_rss(igb_t *igb)
2407 {
2408 	struct e1000_hw *hw = &igb->hw;
2409 	uint32_t i, mrqc, rxcsum;
2410 	int shift = 0;
2411 	uint32_t random;
2412 	union e1000_reta {
2413 		uint32_t	dword;
2414 		uint8_t		bytes[4];
2415 	} reta;
2416 
2417 	/* Setup the Redirection Table */
2418 	if (hw->mac.type == e1000_82576) {
2419 		shift = 3;
2420 	} else if (hw->mac.type == e1000_82575) {
2421 		shift = 6;
2422 	}
2423 	for (i = 0; i < (32 * 4); i++) {
2424 		reta.bytes[i & 3] = (i % igb->num_rx_rings) << shift;
2425 		if ((i & 3) == 3) {
2426 			E1000_WRITE_REG(hw,
2427 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2428 		}
2429 	}
2430 
2431 	/* Fill out hash function seeds */
2432 	for (i = 0; i < 10; i++) {
2433 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2434 		    sizeof (uint32_t));
2435 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2436 	}
2437 
2438 	/* Setup the Multiple Receive Queue Control register */
2439 	mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2440 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2441 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2442 	    E1000_MRQC_RSS_FIELD_IPV6 |
2443 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2444 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2445 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2446 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2447 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2448 
2449 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2450 
2451 	/*
2452 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2453 	 *
2454 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2455 	 * checksum offloading provided by the 82575 chipset besides the IP
2456 	 * header checksum offloading and the TCP/UDP checksum offloading.
2457 	 * The Packet Checksum is by default computed over the entire packet
2458 	 * from the first byte of the DA through the last byte of the CRC,
2459 	 * including the Ethernet and IP headers.
2460 	 *
2461 	 * It is a hardware limitation that Packet Checksum is mutually
2462 	 * exclusive with RSS.
2463 	 */
2464 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2465 	rxcsum |= E1000_RXCSUM_PCSD;
2466 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2467 }
2468 
2469 /*
2470  * igb_setup_mac_rss_classify - Setup MAC classification and rss
2471  */
2472 static void
2473 igb_setup_mac_rss_classify(igb_t *igb)
2474 {
2475 	struct e1000_hw *hw = &igb->hw;
2476 	uint32_t i, mrqc, vmdctl, rxcsum;
2477 	uint32_t ring_per_group;
2478 	int shift_group0, shift_group1;
2479 	uint32_t random;
2480 	union e1000_reta {
2481 		uint32_t	dword;
2482 		uint8_t		bytes[4];
2483 	} reta;
2484 
2485 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2486 
2487 	/* Setup the Redirection Table, it is shared between two groups */
2488 	shift_group0 = 2;
2489 	shift_group1 = 6;
2490 	for (i = 0; i < (32 * 4); i++) {
2491 		reta.bytes[i & 3] = ((i % ring_per_group) << shift_group0) |
2492 		    ((ring_per_group + (i % ring_per_group)) << shift_group1);
2493 		if ((i & 3) == 3) {
2494 			E1000_WRITE_REG(hw,
2495 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2496 		}
2497 	}
2498 
2499 	/* Fill out hash function seeds */
2500 	for (i = 0; i < 10; i++) {
2501 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2502 		    sizeof (uint32_t));
2503 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2504 	}
2505 
2506 	/*
2507 	 * Setup the Multiple Receive Queue Control register,
2508 	 * enable VMDq based on packet destination MAC address and RSS.
2509 	 */
2510 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_RSS_GROUP;
2511 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2512 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2513 	    E1000_MRQC_RSS_FIELD_IPV6 |
2514 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2515 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2516 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2517 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2518 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2519 
2520 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2521 
2522 
2523 	/* Define the default group and default queues */
2524 	vmdctl = E1000_VMDQ_MAC_GROUP_DEFAULT_QUEUE;
2525 	E1000_WRITE_REG(hw, E1000_VT_CTL, vmdctl);
2526 
2527 	/*
2528 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2529 	 *
2530 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2531 	 * checksum offloading provided by the 82575 chipset besides the IP
2532 	 * header checksum offloading and the TCP/UDP checksum offloading.
2533 	 * The Packet Checksum is by default computed over the entire packet
2534 	 * from the first byte of the DA through the last byte of the CRC,
2535 	 * including the Ethernet and IP headers.
2536 	 *
2537 	 * It is a hardware limitation that Packet Checksum is mutually
2538 	 * exclusive with RSS.
2539 	 */
2540 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2541 	rxcsum |= E1000_RXCSUM_PCSD;
2542 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2543 }
2544 
2545 /*
2546  * igb_setup_mac_classify - Setup MAC classification feature
2547  */
2548 static void
2549 igb_setup_mac_classify(igb_t *igb)
2550 {
2551 	struct e1000_hw *hw = &igb->hw;
2552 	uint32_t mrqc, rxcsum;
2553 
2554 	/*
2555 	 * Setup the Multiple Receive Queue Control register,
2556 	 * enable VMDq based on packet destination MAC address.
2557 	 */
2558 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_GROUP;
2559 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2560 
2561 	/*
2562 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2563 	 *
2564 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2565 	 * checksum offloading provided by the 82575 chipset besides the IP
2566 	 * header checksum offloading and the TCP/UDP checksum offloading.
2567 	 * The Packet Checksum is by default computed over the entire packet
2568 	 * from the first byte of the DA through the last byte of the CRC,
2569 	 * including the Ethernet and IP headers.
2570 	 *
2571 	 * It is a hardware limitation that Packet Checksum is mutually
2572 	 * exclusive with RSS.
2573 	 */
2574 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2575 	rxcsum |= E1000_RXCSUM_PCSD;
2576 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2577 
2578 }
2579 
2580 /*
2581  * igb_init_unicst - Initialize the unicast addresses
2582  */
2583 static void
2584 igb_init_unicst(igb_t *igb)
2585 {
2586 	struct e1000_hw *hw = &igb->hw;
2587 	int slot;
2588 
2589 	/*
2590 	 * Here we should consider two situations:
2591 	 *
2592 	 * 1. Chipset is initialized the first time
2593 	 *    Initialize the multiple unicast addresses, and
2594 	 *    save the default MAC address.
2595 	 *
2596 	 * 2. Chipset is reset
2597 	 *    Recover the multiple unicast addresses from the
2598 	 *    software data structure to the RAR registers.
2599 	 */
2600 
2601 	/*
2602 	 * Clear the default MAC address in the RAR0 rgister,
2603 	 * which is loaded from EEPROM when system boot or chipreset,
2604 	 * this will cause the conficts with add_mac/rem_mac entry
2605 	 * points when VMDq is enabled. For this reason, the RAR0
2606 	 * must be cleared for both cases mentioned above.
2607 	 */
2608 	e1000_rar_clear(hw, 0);
2609 
2610 	if (!igb->unicst_init) {
2611 
2612 		/* Initialize the multiple unicast addresses */
2613 		igb->unicst_total = MAX_NUM_UNICAST_ADDRESSES;
2614 		igb->unicst_avail = igb->unicst_total;
2615 
2616 		for (slot = 0; slot < igb->unicst_total; slot++)
2617 			igb->unicst_addr[slot].mac.set = 0;
2618 
2619 		igb->unicst_init = B_TRUE;
2620 	} else {
2621 		/* Re-configure the RAR registers */
2622 		for (slot = 0; slot < igb->unicst_total; slot++) {
2623 			(void) e1000_rar_set_vmdq(hw,
2624 			    igb->unicst_addr[slot].mac.addr,
2625 			    slot, igb->vmdq_mode,
2626 			    igb->unicst_addr[slot].mac.group_index);
2627 		}
2628 	}
2629 }
2630 
2631 /*
2632  * igb_unicst_find - Find the slot for the specified unicast address
2633  */
2634 int
2635 igb_unicst_find(igb_t *igb, const uint8_t *mac_addr)
2636 {
2637 	int slot;
2638 
2639 	ASSERT(mutex_owned(&igb->gen_lock));
2640 
2641 	for (slot = 0; slot < igb->unicst_total; slot++) {
2642 		if (bcmp(igb->unicst_addr[slot].mac.addr,
2643 		    mac_addr, ETHERADDRL) == 0)
2644 			return (slot);
2645 	}
2646 
2647 	return (-1);
2648 }
2649 
2650 /*
2651  * igb_unicst_set - Set the unicast address to the specified slot
2652  */
2653 int
2654 igb_unicst_set(igb_t *igb, const uint8_t *mac_addr,
2655     int slot)
2656 {
2657 	struct e1000_hw *hw = &igb->hw;
2658 
2659 	ASSERT(mutex_owned(&igb->gen_lock));
2660 
2661 	/*
2662 	 * Save the unicast address in the software data structure
2663 	 */
2664 	bcopy(mac_addr, igb->unicst_addr[slot].mac.addr, ETHERADDRL);
2665 
2666 	/*
2667 	 * Set the unicast address to the RAR register
2668 	 */
2669 	(void) e1000_rar_set(hw, (uint8_t *)mac_addr, slot);
2670 
2671 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2672 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2673 		return (EIO);
2674 	}
2675 
2676 	return (0);
2677 }
2678 
2679 /*
2680  * igb_multicst_add - Add a multicst address
2681  */
2682 int
2683 igb_multicst_add(igb_t *igb, const uint8_t *multiaddr)
2684 {
2685 	struct ether_addr *new_table;
2686 	size_t new_len;
2687 	size_t old_len;
2688 
2689 	ASSERT(mutex_owned(&igb->gen_lock));
2690 
2691 	if ((multiaddr[0] & 01) == 0) {
2692 		igb_log(igb, IGB_LOG_ERROR, "Illegal multicast address");
2693 		return (EINVAL);
2694 	}
2695 
2696 	if (igb->mcast_count >= igb->mcast_max_num) {
2697 		igb_log(igb, IGB_LOG_ERROR,
2698 		    "Adapter requested more than %d mcast addresses",
2699 		    igb->mcast_max_num);
2700 		return (ENOENT);
2701 	}
2702 
2703 	if (igb->mcast_count == igb->mcast_alloc_count) {
2704 		old_len = igb->mcast_alloc_count *
2705 		    sizeof (struct ether_addr);
2706 		new_len = (igb->mcast_alloc_count + MCAST_ALLOC_COUNT) *
2707 		    sizeof (struct ether_addr);
2708 
2709 		new_table = kmem_alloc(new_len, KM_NOSLEEP);
2710 		if (new_table == NULL) {
2711 			igb_log(igb, IGB_LOG_ERROR,
2712 			    "Not enough memory to alloc mcast table");
2713 			return (ENOMEM);
2714 		}
2715 
2716 		if (igb->mcast_table != NULL) {
2717 			bcopy(igb->mcast_table, new_table, old_len);
2718 			kmem_free(igb->mcast_table, old_len);
2719 		}
2720 		igb->mcast_alloc_count += MCAST_ALLOC_COUNT;
2721 		igb->mcast_table = new_table;
2722 	}
2723 
2724 	bcopy(multiaddr,
2725 	    &igb->mcast_table[igb->mcast_count], ETHERADDRL);
2726 	igb->mcast_count++;
2727 
2728 	/*
2729 	 * Update the multicast table in the hardware
2730 	 */
2731 	igb_setup_multicst(igb);
2732 
2733 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2734 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2735 		return (EIO);
2736 	}
2737 
2738 	return (0);
2739 }
2740 
2741 /*
2742  * igb_multicst_remove - Remove a multicst address
2743  */
2744 int
2745 igb_multicst_remove(igb_t *igb, const uint8_t *multiaddr)
2746 {
2747 	struct ether_addr *new_table;
2748 	size_t new_len;
2749 	size_t old_len;
2750 	int i;
2751 
2752 	ASSERT(mutex_owned(&igb->gen_lock));
2753 
2754 	for (i = 0; i < igb->mcast_count; i++) {
2755 		if (bcmp(multiaddr, &igb->mcast_table[i],
2756 		    ETHERADDRL) == 0) {
2757 			for (i++; i < igb->mcast_count; i++) {
2758 				igb->mcast_table[i - 1] =
2759 				    igb->mcast_table[i];
2760 			}
2761 			igb->mcast_count--;
2762 			break;
2763 		}
2764 	}
2765 
2766 	if ((igb->mcast_alloc_count - igb->mcast_count) >
2767 	    MCAST_ALLOC_COUNT) {
2768 		old_len = igb->mcast_alloc_count *
2769 		    sizeof (struct ether_addr);
2770 		new_len = (igb->mcast_alloc_count - MCAST_ALLOC_COUNT) *
2771 		    sizeof (struct ether_addr);
2772 
2773 		new_table = kmem_alloc(new_len, KM_NOSLEEP);
2774 		if (new_table != NULL) {
2775 			bcopy(igb->mcast_table, new_table, new_len);
2776 			kmem_free(igb->mcast_table, old_len);
2777 			igb->mcast_alloc_count -= MCAST_ALLOC_COUNT;
2778 			igb->mcast_table = new_table;
2779 		}
2780 	}
2781 
2782 	/*
2783 	 * Update the multicast table in the hardware
2784 	 */
2785 	igb_setup_multicst(igb);
2786 
2787 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2788 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2789 		return (EIO);
2790 	}
2791 
2792 	return (0);
2793 }
2794 
2795 static void
2796 igb_release_multicast(igb_t *igb)
2797 {
2798 	if (igb->mcast_table != NULL) {
2799 		kmem_free(igb->mcast_table,
2800 		    igb->mcast_alloc_count * sizeof (struct ether_addr));
2801 		igb->mcast_table = NULL;
2802 	}
2803 }
2804 
2805 /*
2806  * igb_setup_multicast - setup multicast data structures
2807  *
2808  * This routine initializes all of the multicast related structures
2809  * and save them in the hardware registers.
2810  */
2811 static void
2812 igb_setup_multicst(igb_t *igb)
2813 {
2814 	uint8_t *mc_addr_list;
2815 	uint32_t mc_addr_count;
2816 	struct e1000_hw *hw = &igb->hw;
2817 
2818 	ASSERT(mutex_owned(&igb->gen_lock));
2819 	ASSERT(igb->mcast_count <= igb->mcast_max_num);
2820 
2821 	mc_addr_list = (uint8_t *)igb->mcast_table;
2822 	mc_addr_count = igb->mcast_count;
2823 
2824 	/*
2825 	 * Update the multicase addresses to the MTA registers
2826 	 */
2827 	e1000_update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2828 }
2829 
2830 /*
2831  * igb_get_conf - Get driver configurations set in driver.conf
2832  *
2833  * This routine gets user-configured values out of the configuration
2834  * file igb.conf.
2835  *
2836  * For each configurable value, there is a minimum, a maximum, and a
2837  * default.
2838  * If user does not configure a value, use the default.
2839  * If user configures below the minimum, use the minumum.
2840  * If user configures above the maximum, use the maxumum.
2841  */
2842 static void
2843 igb_get_conf(igb_t *igb)
2844 {
2845 	struct e1000_hw *hw = &igb->hw;
2846 	uint32_t default_mtu;
2847 	uint32_t flow_control;
2848 	uint32_t ring_per_group;
2849 	int i;
2850 
2851 	/*
2852 	 * igb driver supports the following user configurations:
2853 	 *
2854 	 * Link configurations:
2855 	 *    adv_autoneg_cap
2856 	 *    adv_1000fdx_cap
2857 	 *    adv_100fdx_cap
2858 	 *    adv_100hdx_cap
2859 	 *    adv_10fdx_cap
2860 	 *    adv_10hdx_cap
2861 	 * Note: 1000hdx is not supported.
2862 	 *
2863 	 * Jumbo frame configuration:
2864 	 *    default_mtu
2865 	 *
2866 	 * Ethernet flow control configuration:
2867 	 *    flow_control
2868 	 *
2869 	 * Multiple rings configurations:
2870 	 *    tx_queue_number
2871 	 *    tx_ring_size
2872 	 *    rx_queue_number
2873 	 *    rx_ring_size
2874 	 *
2875 	 * Call igb_get_prop() to get the value for a specific
2876 	 * configuration parameter.
2877 	 */
2878 
2879 	/*
2880 	 * Link configurations
2881 	 */
2882 	igb->param_adv_autoneg_cap = igb_get_prop(igb,
2883 	    PROP_ADV_AUTONEG_CAP, 0, 1, 1);
2884 	igb->param_adv_1000fdx_cap = igb_get_prop(igb,
2885 	    PROP_ADV_1000FDX_CAP, 0, 1, 1);
2886 	igb->param_adv_100fdx_cap = igb_get_prop(igb,
2887 	    PROP_ADV_100FDX_CAP, 0, 1, 1);
2888 	igb->param_adv_100hdx_cap = igb_get_prop(igb,
2889 	    PROP_ADV_100HDX_CAP, 0, 1, 1);
2890 	igb->param_adv_10fdx_cap = igb_get_prop(igb,
2891 	    PROP_ADV_10FDX_CAP, 0, 1, 1);
2892 	igb->param_adv_10hdx_cap = igb_get_prop(igb,
2893 	    PROP_ADV_10HDX_CAP, 0, 1, 1);
2894 
2895 	/*
2896 	 * Jumbo frame configurations
2897 	 */
2898 	default_mtu = igb_get_prop(igb, PROP_DEFAULT_MTU,
2899 	    MIN_MTU, MAX_MTU, DEFAULT_MTU);
2900 
2901 	igb->max_frame_size = default_mtu +
2902 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
2903 
2904 	/*
2905 	 * Ethernet flow control configuration
2906 	 */
2907 	flow_control = igb_get_prop(igb, PROP_FLOW_CONTROL,
2908 	    e1000_fc_none, 4, e1000_fc_full);
2909 	if (flow_control == 4)
2910 		flow_control = e1000_fc_default;
2911 
2912 	hw->fc.requested_mode = flow_control;
2913 
2914 	/*
2915 	 * Multiple rings configurations
2916 	 */
2917 	igb->tx_ring_size = igb_get_prop(igb, PROP_TX_RING_SIZE,
2918 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
2919 	igb->rx_ring_size = igb_get_prop(igb, PROP_RX_RING_SIZE,
2920 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
2921 
2922 	igb->mr_enable = igb_get_prop(igb, PROP_MR_ENABLE, 0, 1, 0);
2923 	igb->num_rx_groups = igb_get_prop(igb, PROP_RX_GROUP_NUM,
2924 	    MIN_RX_GROUP_NUM, MAX_RX_GROUP_NUM, DEFAULT_RX_GROUP_NUM);
2925 	/*
2926 	 * Currently we do not support VMDq for 82576 and 82580.
2927 	 * If it is e1000_82576, set num_rx_groups to 1.
2928 	 */
2929 	if (hw->mac.type >= e1000_82576)
2930 		igb->num_rx_groups = 1;
2931 
2932 	if (igb->mr_enable) {
2933 		igb->num_tx_rings = igb->capab->def_tx_que_num;
2934 		igb->num_rx_rings = igb->capab->def_rx_que_num;
2935 	} else {
2936 		igb->num_tx_rings = 1;
2937 		igb->num_rx_rings = 1;
2938 
2939 		if (igb->num_rx_groups > 1) {
2940 			igb_log(igb, IGB_LOG_ERROR,
2941 			    "Invalid rx groups number. Please enable multiple "
2942 			    "rings first");
2943 			igb->num_rx_groups = 1;
2944 		}
2945 	}
2946 
2947 	/*
2948 	 * Check the divisibility between rx rings and rx groups.
2949 	 */
2950 	for (i = igb->num_rx_groups; i > 0; i--) {
2951 		if ((igb->num_rx_rings % i) == 0)
2952 			break;
2953 	}
2954 	if (i != igb->num_rx_groups) {
2955 		igb_log(igb, IGB_LOG_ERROR,
2956 		    "Invalid rx groups number. Downgrade the rx group "
2957 		    "number to %d.", i);
2958 		igb->num_rx_groups = i;
2959 	}
2960 
2961 	/*
2962 	 * Get the ring number per group.
2963 	 */
2964 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2965 
2966 	if (igb->num_rx_groups == 1) {
2967 		/*
2968 		 * One rx ring group, the rx ring number is num_rx_rings.
2969 		 */
2970 		igb->vmdq_mode = E1000_VMDQ_OFF;
2971 	} else if (ring_per_group == 1) {
2972 		/*
2973 		 * Multiple rx groups, each group has one rx ring.
2974 		 */
2975 		igb->vmdq_mode = E1000_VMDQ_MAC;
2976 	} else {
2977 		/*
2978 		 * Multiple groups and multiple rings.
2979 		 */
2980 		igb->vmdq_mode = E1000_VMDQ_MAC_RSS;
2981 	}
2982 
2983 	/*
2984 	 * Tunable used to force an interrupt type. The only use is
2985 	 * for testing of the lesser interrupt types.
2986 	 * 0 = don't force interrupt type
2987 	 * 1 = force interrupt type MSIX
2988 	 * 2 = force interrupt type MSI
2989 	 * 3 = force interrupt type Legacy
2990 	 */
2991 	igb->intr_force = igb_get_prop(igb, PROP_INTR_FORCE,
2992 	    IGB_INTR_NONE, IGB_INTR_LEGACY, IGB_INTR_NONE);
2993 
2994 	igb->tx_hcksum_enable = igb_get_prop(igb, PROP_TX_HCKSUM_ENABLE,
2995 	    0, 1, 1);
2996 	igb->rx_hcksum_enable = igb_get_prop(igb, PROP_RX_HCKSUM_ENABLE,
2997 	    0, 1, 1);
2998 	igb->lso_enable = igb_get_prop(igb, PROP_LSO_ENABLE,
2999 	    0, 1, 1);
3000 	igb->tx_head_wb_enable = igb_get_prop(igb, PROP_TX_HEAD_WB_ENABLE,
3001 	    0, 1, 1);
3002 
3003 	/*
3004 	 * igb LSO needs the tx h/w checksum support.
3005 	 * Here LSO will be disabled if tx h/w checksum has been disabled.
3006 	 */
3007 	if (igb->tx_hcksum_enable == B_FALSE)
3008 		igb->lso_enable = B_FALSE;
3009 
3010 	igb->tx_copy_thresh = igb_get_prop(igb, PROP_TX_COPY_THRESHOLD,
3011 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
3012 	    DEFAULT_TX_COPY_THRESHOLD);
3013 	igb->tx_recycle_thresh = igb_get_prop(igb, PROP_TX_RECYCLE_THRESHOLD,
3014 	    MIN_TX_RECYCLE_THRESHOLD, MAX_TX_RECYCLE_THRESHOLD,
3015 	    DEFAULT_TX_RECYCLE_THRESHOLD);
3016 	igb->tx_overload_thresh = igb_get_prop(igb, PROP_TX_OVERLOAD_THRESHOLD,
3017 	    MIN_TX_OVERLOAD_THRESHOLD, MAX_TX_OVERLOAD_THRESHOLD,
3018 	    DEFAULT_TX_OVERLOAD_THRESHOLD);
3019 	igb->tx_resched_thresh = igb_get_prop(igb, PROP_TX_RESCHED_THRESHOLD,
3020 	    MIN_TX_RESCHED_THRESHOLD,
3021 	    MIN(igb->tx_ring_size, MAX_TX_RESCHED_THRESHOLD),
3022 	    igb->tx_ring_size > DEFAULT_TX_RESCHED_THRESHOLD ?
3023 	    DEFAULT_TX_RESCHED_THRESHOLD : DEFAULT_TX_RESCHED_THRESHOLD_LOW);
3024 
3025 	igb->rx_copy_thresh = igb_get_prop(igb, PROP_RX_COPY_THRESHOLD,
3026 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
3027 	    DEFAULT_RX_COPY_THRESHOLD);
3028 	igb->rx_limit_per_intr = igb_get_prop(igb, PROP_RX_LIMIT_PER_INTR,
3029 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
3030 	    DEFAULT_RX_LIMIT_PER_INTR);
3031 
3032 	igb->intr_throttling[0] = igb_get_prop(igb, PROP_INTR_THROTTLING,
3033 	    igb->capab->min_intr_throttle,
3034 	    igb->capab->max_intr_throttle,
3035 	    igb->capab->def_intr_throttle);
3036 
3037 	/*
3038 	 * Max number of multicast addresses
3039 	 */
3040 	igb->mcast_max_num =
3041 	    igb_get_prop(igb, PROP_MCAST_MAX_NUM,
3042 	    MIN_MCAST_NUM, MAX_MCAST_NUM, DEFAULT_MCAST_NUM);
3043 }
3044 
3045 /*
3046  * igb_get_prop - Get a property value out of the configuration file igb.conf
3047  *
3048  * Caller provides the name of the property, a default value, a minimum
3049  * value, and a maximum value.
3050  *
3051  * Return configured value of the property, with default, minimum and
3052  * maximum properly applied.
3053  */
3054 static int
3055 igb_get_prop(igb_t *igb,
3056     char *propname,	/* name of the property */
3057     int minval,		/* minimum acceptable value */
3058     int maxval,		/* maximim acceptable value */
3059     int defval)		/* default value */
3060 {
3061 	int value;
3062 
3063 	/*
3064 	 * Call ddi_prop_get_int() to read the conf settings
3065 	 */
3066 	value = ddi_prop_get_int(DDI_DEV_T_ANY, igb->dip,
3067 	    DDI_PROP_DONTPASS, propname, defval);
3068 
3069 	if (value > maxval)
3070 		value = maxval;
3071 
3072 	if (value < minval)
3073 		value = minval;
3074 
3075 	return (value);
3076 }
3077 
3078 /*
3079  * igb_setup_link - Using the link properties to setup the link
3080  */
3081 int
3082 igb_setup_link(igb_t *igb, boolean_t setup_hw)
3083 {
3084 	struct e1000_mac_info *mac;
3085 	struct e1000_phy_info *phy;
3086 	boolean_t invalid;
3087 
3088 	mac = &igb->hw.mac;
3089 	phy = &igb->hw.phy;
3090 	invalid = B_FALSE;
3091 
3092 	if (igb->param_adv_autoneg_cap == 1) {
3093 		mac->autoneg = B_TRUE;
3094 		phy->autoneg_advertised = 0;
3095 
3096 		/*
3097 		 * 1000hdx is not supported for autonegotiation
3098 		 */
3099 		if (igb->param_adv_1000fdx_cap == 1)
3100 			phy->autoneg_advertised |= ADVERTISE_1000_FULL;
3101 
3102 		if (igb->param_adv_100fdx_cap == 1)
3103 			phy->autoneg_advertised |= ADVERTISE_100_FULL;
3104 
3105 		if (igb->param_adv_100hdx_cap == 1)
3106 			phy->autoneg_advertised |= ADVERTISE_100_HALF;
3107 
3108 		if (igb->param_adv_10fdx_cap == 1)
3109 			phy->autoneg_advertised |= ADVERTISE_10_FULL;
3110 
3111 		if (igb->param_adv_10hdx_cap == 1)
3112 			phy->autoneg_advertised |= ADVERTISE_10_HALF;
3113 
3114 		if (phy->autoneg_advertised == 0)
3115 			invalid = B_TRUE;
3116 	} else {
3117 		mac->autoneg = B_FALSE;
3118 
3119 		/*
3120 		 * 1000fdx and 1000hdx are not supported for forced link
3121 		 */
3122 		if (igb->param_adv_100fdx_cap == 1)
3123 			mac->forced_speed_duplex = ADVERTISE_100_FULL;
3124 		else if (igb->param_adv_100hdx_cap == 1)
3125 			mac->forced_speed_duplex = ADVERTISE_100_HALF;
3126 		else if (igb->param_adv_10fdx_cap == 1)
3127 			mac->forced_speed_duplex = ADVERTISE_10_FULL;
3128 		else if (igb->param_adv_10hdx_cap == 1)
3129 			mac->forced_speed_duplex = ADVERTISE_10_HALF;
3130 		else
3131 			invalid = B_TRUE;
3132 	}
3133 
3134 	if (invalid) {
3135 		igb_log(igb, IGB_LOG_INFO, "Invalid link settings. Setup "
3136 		    "link to autonegotiation with full link capabilities.");
3137 		mac->autoneg = B_TRUE;
3138 		phy->autoneg_advertised = ADVERTISE_1000_FULL |
3139 		    ADVERTISE_100_FULL | ADVERTISE_100_HALF |
3140 		    ADVERTISE_10_FULL | ADVERTISE_10_HALF;
3141 	}
3142 
3143 	if (setup_hw) {
3144 		if (e1000_setup_link(&igb->hw) != E1000_SUCCESS)
3145 			return (IGB_FAILURE);
3146 	}
3147 
3148 	return (IGB_SUCCESS);
3149 }
3150 
3151 
3152 /*
3153  * igb_is_link_up - Check if the link is up
3154  */
3155 static boolean_t
3156 igb_is_link_up(igb_t *igb)
3157 {
3158 	struct e1000_hw *hw = &igb->hw;
3159 	boolean_t link_up = B_FALSE;
3160 
3161 	ASSERT(mutex_owned(&igb->gen_lock));
3162 
3163 	/*
3164 	 * get_link_status is set in the interrupt handler on link-status-change
3165 	 * or rx sequence error interrupt.  get_link_status will stay
3166 	 * false until the e1000_check_for_link establishes link only
3167 	 * for copper adapters.
3168 	 */
3169 	switch (hw->phy.media_type) {
3170 	case e1000_media_type_copper:
3171 		if (hw->mac.get_link_status) {
3172 			(void) e1000_check_for_link(hw);
3173 			link_up = !hw->mac.get_link_status;
3174 		} else {
3175 			link_up = B_TRUE;
3176 		}
3177 		break;
3178 	case e1000_media_type_fiber:
3179 		(void) e1000_check_for_link(hw);
3180 		link_up = (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU);
3181 		break;
3182 	case e1000_media_type_internal_serdes:
3183 		(void) e1000_check_for_link(hw);
3184 		link_up = hw->mac.serdes_has_link;
3185 		break;
3186 	}
3187 
3188 	return (link_up);
3189 }
3190 
3191 /*
3192  * igb_link_check - Link status processing
3193  */
3194 static boolean_t
3195 igb_link_check(igb_t *igb)
3196 {
3197 	struct e1000_hw *hw = &igb->hw;
3198 	uint16_t speed = 0, duplex = 0;
3199 	boolean_t link_changed = B_FALSE;
3200 
3201 	ASSERT(mutex_owned(&igb->gen_lock));
3202 
3203 	if (igb_is_link_up(igb)) {
3204 		/*
3205 		 * The Link is up, check whether it was marked as down earlier
3206 		 */
3207 		if (igb->link_state != LINK_STATE_UP) {
3208 			(void) e1000_get_speed_and_duplex(hw, &speed, &duplex);
3209 			igb->link_speed = speed;
3210 			igb->link_duplex = duplex;
3211 			igb->link_state = LINK_STATE_UP;
3212 			link_changed = B_TRUE;
3213 			if (!igb->link_complete)
3214 				igb_stop_link_timer(igb);
3215 		}
3216 	} else if (igb->link_complete) {
3217 		if (igb->link_state != LINK_STATE_DOWN) {
3218 			igb->link_speed = 0;
3219 			igb->link_duplex = 0;
3220 			igb->link_state = LINK_STATE_DOWN;
3221 			link_changed = B_TRUE;
3222 		}
3223 	}
3224 
3225 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3226 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3227 		return (B_FALSE);
3228 	}
3229 
3230 	return (link_changed);
3231 }
3232 
3233 /*
3234  * igb_local_timer - driver watchdog function
3235  *
3236  * This function will handle the hardware stall check, link status
3237  * check and other routines.
3238  */
3239 static void
3240 igb_local_timer(void *arg)
3241 {
3242 	igb_t *igb = (igb_t *)arg;
3243 	boolean_t link_changed = B_FALSE;
3244 
3245 	if (igb->igb_state & IGB_ERROR) {
3246 		igb->reset_count++;
3247 		if (igb_reset(igb) == IGB_SUCCESS)
3248 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_RESTORED);
3249 
3250 		igb_restart_watchdog_timer(igb);
3251 		return;
3252 	}
3253 
3254 	if (igb_stall_check(igb) || (igb->igb_state & IGB_STALL)) {
3255 		igb_fm_ereport(igb, DDI_FM_DEVICE_STALL);
3256 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
3257 		igb->reset_count++;
3258 		if (igb_reset(igb) == IGB_SUCCESS)
3259 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_RESTORED);
3260 
3261 		igb_restart_watchdog_timer(igb);
3262 		return;
3263 	}
3264 
3265 	mutex_enter(&igb->gen_lock);
3266 	if (!(igb->igb_state & IGB_SUSPENDED) && (igb->igb_state & IGB_STARTED))
3267 		link_changed = igb_link_check(igb);
3268 	mutex_exit(&igb->gen_lock);
3269 
3270 	if (link_changed)
3271 		mac_link_update(igb->mac_hdl, igb->link_state);
3272 
3273 	igb_restart_watchdog_timer(igb);
3274 }
3275 
3276 /*
3277  * igb_link_timer - link setup timer function
3278  *
3279  * It is called when the timer for link setup is expired, which indicates
3280  * the completion of the link setup. The link state will not be updated
3281  * until the link setup is completed. And the link state will not be sent
3282  * to the upper layer through mac_link_update() in this function. It will
3283  * be updated in the local timer routine or the interrupts service routine
3284  * after the interface is started (plumbed).
3285  */
3286 static void
3287 igb_link_timer(void *arg)
3288 {
3289 	igb_t *igb = (igb_t *)arg;
3290 
3291 	mutex_enter(&igb->link_lock);
3292 	igb->link_complete = B_TRUE;
3293 	igb->link_tid = 0;
3294 	mutex_exit(&igb->link_lock);
3295 }
3296 /*
3297  * igb_stall_check - check for transmit stall
3298  *
3299  * This function checks if the adapter is stalled (in transmit).
3300  *
3301  * It is called each time the watchdog timeout is invoked.
3302  * If the transmit descriptor reclaim continuously fails,
3303  * the watchdog value will increment by 1. If the watchdog
3304  * value exceeds the threshold, the igb is assumed to
3305  * have stalled and need to be reset.
3306  */
3307 static boolean_t
3308 igb_stall_check(igb_t *igb)
3309 {
3310 	igb_tx_ring_t *tx_ring;
3311 	struct e1000_hw *hw = &igb->hw;
3312 	boolean_t result;
3313 	int i;
3314 
3315 	if (igb->link_state != LINK_STATE_UP)
3316 		return (B_FALSE);
3317 
3318 	/*
3319 	 * If any tx ring is stalled, we'll reset the chipset
3320 	 */
3321 	result = B_FALSE;
3322 	for (i = 0; i < igb->num_tx_rings; i++) {
3323 		tx_ring = &igb->tx_rings[i];
3324 
3325 		if (tx_ring->recycle_fail > 0)
3326 			tx_ring->stall_watchdog++;
3327 		else
3328 			tx_ring->stall_watchdog = 0;
3329 
3330 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
3331 			result = B_TRUE;
3332 			if (hw->mac.type == e1000_82580) {
3333 				hw->dev_spec._82575.global_device_reset
3334 				    = B_TRUE;
3335 			}
3336 			break;
3337 		}
3338 	}
3339 
3340 	if (result) {
3341 		tx_ring->stall_watchdog = 0;
3342 		tx_ring->recycle_fail = 0;
3343 	}
3344 
3345 	return (result);
3346 }
3347 
3348 
3349 /*
3350  * is_valid_mac_addr - Check if the mac address is valid
3351  */
3352 static boolean_t
3353 is_valid_mac_addr(uint8_t *mac_addr)
3354 {
3355 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
3356 	const uint8_t addr_test2[6] =
3357 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
3358 
3359 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
3360 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
3361 		return (B_FALSE);
3362 
3363 	return (B_TRUE);
3364 }
3365 
3366 static boolean_t
3367 igb_find_mac_address(igb_t *igb)
3368 {
3369 	struct e1000_hw *hw = &igb->hw;
3370 #ifdef __sparc
3371 	uchar_t *bytes;
3372 	struct ether_addr sysaddr;
3373 	uint_t nelts;
3374 	int err;
3375 	boolean_t found = B_FALSE;
3376 
3377 	/*
3378 	 * The "vendor's factory-set address" may already have
3379 	 * been extracted from the chip, but if the property
3380 	 * "local-mac-address" is set we use that instead.
3381 	 *
3382 	 * We check whether it looks like an array of 6
3383 	 * bytes (which it should, if OBP set it).  If we can't
3384 	 * make sense of it this way, we'll ignore it.
3385 	 */
3386 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3387 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
3388 	if (err == DDI_PROP_SUCCESS) {
3389 		if (nelts == ETHERADDRL) {
3390 			while (nelts--)
3391 				hw->mac.addr[nelts] = bytes[nelts];
3392 			found = B_TRUE;
3393 		}
3394 		ddi_prop_free(bytes);
3395 	}
3396 
3397 	/*
3398 	 * Look up the OBP property "local-mac-address?". If the user has set
3399 	 * 'local-mac-address? = false', use "the system address" instead.
3400 	 */
3401 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip, 0,
3402 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
3403 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
3404 			if (localetheraddr(NULL, &sysaddr) != 0) {
3405 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
3406 				found = B_TRUE;
3407 			}
3408 		}
3409 		ddi_prop_free(bytes);
3410 	}
3411 
3412 	/*
3413 	 * Finally(!), if there's a valid "mac-address" property (created
3414 	 * if we netbooted from this interface), we must use this instead
3415 	 * of any of the above to ensure that the NFS/install server doesn't
3416 	 * get confused by the address changing as Solaris takes over!
3417 	 */
3418 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3419 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
3420 	if (err == DDI_PROP_SUCCESS) {
3421 		if (nelts == ETHERADDRL) {
3422 			while (nelts--)
3423 				hw->mac.addr[nelts] = bytes[nelts];
3424 			found = B_TRUE;
3425 		}
3426 		ddi_prop_free(bytes);
3427 	}
3428 
3429 	if (found) {
3430 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
3431 		return (B_TRUE);
3432 	}
3433 #endif
3434 
3435 	/*
3436 	 * Read the device MAC address from the EEPROM
3437 	 */
3438 	if (e1000_read_mac_addr(hw) != E1000_SUCCESS)
3439 		return (B_FALSE);
3440 
3441 	return (B_TRUE);
3442 }
3443 
3444 #pragma inline(igb_arm_watchdog_timer)
3445 
3446 static void
3447 igb_arm_watchdog_timer(igb_t *igb)
3448 {
3449 	/*
3450 	 * Fire a watchdog timer
3451 	 */
3452 	igb->watchdog_tid =
3453 	    timeout(igb_local_timer,
3454 	    (void *)igb, 1 * drv_usectohz(1000000));
3455 
3456 }
3457 
3458 /*
3459  * igb_enable_watchdog_timer - Enable and start the driver watchdog timer
3460  */
3461 void
3462 igb_enable_watchdog_timer(igb_t *igb)
3463 {
3464 	mutex_enter(&igb->watchdog_lock);
3465 
3466 	if (!igb->watchdog_enable) {
3467 		igb->watchdog_enable = B_TRUE;
3468 		igb->watchdog_start = B_TRUE;
3469 		igb_arm_watchdog_timer(igb);
3470 	}
3471 
3472 	mutex_exit(&igb->watchdog_lock);
3473 
3474 }
3475 
3476 /*
3477  * igb_disable_watchdog_timer - Disable and stop the driver watchdog timer
3478  */
3479 void
3480 igb_disable_watchdog_timer(igb_t *igb)
3481 {
3482 	timeout_id_t tid;
3483 
3484 	mutex_enter(&igb->watchdog_lock);
3485 
3486 	igb->watchdog_enable = B_FALSE;
3487 	igb->watchdog_start = B_FALSE;
3488 	tid = igb->watchdog_tid;
3489 	igb->watchdog_tid = 0;
3490 
3491 	mutex_exit(&igb->watchdog_lock);
3492 
3493 	if (tid != 0)
3494 		(void) untimeout(tid);
3495 
3496 }
3497 
3498 /*
3499  * igb_start_watchdog_timer - Start the driver watchdog timer
3500  */
3501 static void
3502 igb_start_watchdog_timer(igb_t *igb)
3503 {
3504 	mutex_enter(&igb->watchdog_lock);
3505 
3506 	if (igb->watchdog_enable) {
3507 		if (!igb->watchdog_start) {
3508 			igb->watchdog_start = B_TRUE;
3509 			igb_arm_watchdog_timer(igb);
3510 		}
3511 	}
3512 
3513 	mutex_exit(&igb->watchdog_lock);
3514 }
3515 
3516 /*
3517  * igb_restart_watchdog_timer - Restart the driver watchdog timer
3518  */
3519 static void
3520 igb_restart_watchdog_timer(igb_t *igb)
3521 {
3522 	mutex_enter(&igb->watchdog_lock);
3523 
3524 	if (igb->watchdog_start)
3525 		igb_arm_watchdog_timer(igb);
3526 
3527 	mutex_exit(&igb->watchdog_lock);
3528 }
3529 
3530 /*
3531  * igb_stop_watchdog_timer - Stop the driver watchdog timer
3532  */
3533 static void
3534 igb_stop_watchdog_timer(igb_t *igb)
3535 {
3536 	timeout_id_t tid;
3537 
3538 	mutex_enter(&igb->watchdog_lock);
3539 
3540 	igb->watchdog_start = B_FALSE;
3541 	tid = igb->watchdog_tid;
3542 	igb->watchdog_tid = 0;
3543 
3544 	mutex_exit(&igb->watchdog_lock);
3545 
3546 	if (tid != 0)
3547 		(void) untimeout(tid);
3548 }
3549 
3550 /*
3551  * igb_start_link_timer - Start the link setup timer
3552  */
3553 static void
3554 igb_start_link_timer(struct igb *igb)
3555 {
3556 	struct e1000_hw *hw = &igb->hw;
3557 	clock_t link_timeout;
3558 
3559 	if (hw->mac.autoneg)
3560 		link_timeout = PHY_AUTO_NEG_LIMIT *
3561 		    drv_usectohz(100000);
3562 	else
3563 		link_timeout = PHY_FORCE_LIMIT * drv_usectohz(100000);
3564 
3565 	mutex_enter(&igb->link_lock);
3566 	if (hw->phy.autoneg_wait_to_complete) {
3567 		igb->link_complete = B_TRUE;
3568 	} else {
3569 		igb->link_complete = B_FALSE;
3570 		igb->link_tid = timeout(igb_link_timer, (void *)igb,
3571 		    link_timeout);
3572 	}
3573 	mutex_exit(&igb->link_lock);
3574 }
3575 
3576 /*
3577  * igb_stop_link_timer - Stop the link setup timer
3578  */
3579 static void
3580 igb_stop_link_timer(struct igb *igb)
3581 {
3582 	timeout_id_t tid;
3583 
3584 	mutex_enter(&igb->link_lock);
3585 	igb->link_complete = B_TRUE;
3586 	tid = igb->link_tid;
3587 	igb->link_tid = 0;
3588 	mutex_exit(&igb->link_lock);
3589 
3590 	if (tid != 0)
3591 		(void) untimeout(tid);
3592 }
3593 
3594 /*
3595  * igb_disable_adapter_interrupts - Clear/disable all hardware interrupts
3596  */
3597 static void
3598 igb_disable_adapter_interrupts(igb_t *igb)
3599 {
3600 	struct e1000_hw *hw = &igb->hw;
3601 
3602 	/*
3603 	 * Set the IMC register to mask all the interrupts,
3604 	 * including the tx interrupts.
3605 	 */
3606 	E1000_WRITE_REG(hw, E1000_IMC, ~0);
3607 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3608 
3609 	/*
3610 	 * Additional disabling for MSI-X
3611 	 */
3612 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3613 		E1000_WRITE_REG(hw, E1000_EIMC, ~0);
3614 		E1000_WRITE_REG(hw, E1000_EIAC, 0);
3615 		E1000_WRITE_REG(hw, E1000_EIAM, 0);
3616 	}
3617 
3618 	E1000_WRITE_FLUSH(hw);
3619 }
3620 
3621 /*
3622  * igb_enable_adapter_interrupts_82580 - Enable NIC interrupts for 82580
3623  */
3624 static void
3625 igb_enable_adapter_interrupts_82580(igb_t *igb)
3626 {
3627 	struct e1000_hw *hw = &igb->hw;
3628 
3629 	/* Clear any pending interrupts */
3630 	(void) E1000_READ_REG(hw, E1000_ICR);
3631 	igb->ims_mask |= E1000_IMS_DRSTA;
3632 
3633 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3634 
3635 		/* Interrupt enabling for MSI-X */
3636 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3637 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3638 		igb->ims_mask = (E1000_IMS_LSC | E1000_IMS_DRSTA);
3639 		E1000_WRITE_REG(hw, E1000_IMS, igb->ims_mask);
3640 	} else { /* Interrupt enabling for MSI and legacy */
3641 		E1000_WRITE_REG(hw, E1000_IVAR0, E1000_IVAR_VALID);
3642 		igb->ims_mask = IMS_ENABLE_MASK | E1000_IMS_TXQE;
3643 		igb->ims_mask |= E1000_IMS_DRSTA;
3644 		E1000_WRITE_REG(hw, E1000_IMS, igb->ims_mask);
3645 	}
3646 
3647 	/* Disable auto-mask for ICR interrupt bits */
3648 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3649 
3650 	E1000_WRITE_FLUSH(hw);
3651 }
3652 
3653 /*
3654  * igb_enable_adapter_interrupts_82576 - Enable NIC interrupts for 82576
3655  */
3656 static void
3657 igb_enable_adapter_interrupts_82576(igb_t *igb)
3658 {
3659 	struct e1000_hw *hw = &igb->hw;
3660 
3661 	/* Clear any pending interrupts */
3662 	(void) E1000_READ_REG(hw, E1000_ICR);
3663 
3664 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3665 
3666 		/* Interrupt enabling for MSI-X */
3667 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3668 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3669 		igb->ims_mask = E1000_IMS_LSC;
3670 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3671 	} else {
3672 		/* Interrupt enabling for MSI and legacy */
3673 		E1000_WRITE_REG(hw, E1000_IVAR0, E1000_IVAR_VALID);
3674 		igb->ims_mask = IMS_ENABLE_MASK | E1000_IMS_TXQE;
3675 		E1000_WRITE_REG(hw, E1000_IMS,
3676 		    (IMS_ENABLE_MASK | E1000_IMS_TXQE));
3677 	}
3678 
3679 	/* Disable auto-mask for ICR interrupt bits */
3680 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3681 
3682 	E1000_WRITE_FLUSH(hw);
3683 }
3684 
3685 /*
3686  * igb_enable_adapter_interrupts_82575 - Enable NIC interrupts for 82575
3687  */
3688 static void
3689 igb_enable_adapter_interrupts_82575(igb_t *igb)
3690 {
3691 	struct e1000_hw *hw = &igb->hw;
3692 	uint32_t reg;
3693 
3694 	/* Clear any pending interrupts */
3695 	(void) E1000_READ_REG(hw, E1000_ICR);
3696 
3697 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3698 		/* Interrupt enabling for MSI-X */
3699 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3700 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3701 		igb->ims_mask = E1000_IMS_LSC;
3702 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3703 
3704 		/* Enable MSI-X PBA support */
3705 		reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
3706 		reg |= E1000_CTRL_EXT_PBA_CLR;
3707 
3708 		/* Non-selective interrupt clear-on-read */
3709 		reg |= E1000_CTRL_EXT_IRCA;	/* Called NSICR in the EAS */
3710 
3711 		E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
3712 	} else {
3713 		/* Interrupt enabling for MSI and legacy */
3714 		igb->ims_mask = IMS_ENABLE_MASK;
3715 		E1000_WRITE_REG(hw, E1000_IMS, IMS_ENABLE_MASK);
3716 	}
3717 
3718 	E1000_WRITE_FLUSH(hw);
3719 }
3720 
3721 /*
3722  * Loopback Support
3723  */
3724 static lb_property_t lb_normal =
3725 	{ normal,	"normal",	IGB_LB_NONE		};
3726 static lb_property_t lb_external =
3727 	{ external,	"External",	IGB_LB_EXTERNAL		};
3728 static lb_property_t lb_phy =
3729 	{ internal,	"PHY",		IGB_LB_INTERNAL_PHY	};
3730 static lb_property_t lb_serdes =
3731 	{ internal,	"SerDes",	IGB_LB_INTERNAL_SERDES	};
3732 
3733 enum ioc_reply
3734 igb_loopback_ioctl(igb_t *igb, struct iocblk *iocp, mblk_t *mp)
3735 {
3736 	lb_info_sz_t *lbsp;
3737 	lb_property_t *lbpp;
3738 	struct e1000_hw *hw;
3739 	uint32_t *lbmp;
3740 	uint32_t size;
3741 	uint32_t value;
3742 
3743 	hw = &igb->hw;
3744 
3745 	if (mp->b_cont == NULL)
3746 		return (IOC_INVAL);
3747 
3748 	switch (iocp->ioc_cmd) {
3749 	default:
3750 		return (IOC_INVAL);
3751 
3752 	case LB_GET_INFO_SIZE:
3753 		size = sizeof (lb_info_sz_t);
3754 		if (iocp->ioc_count != size)
3755 			return (IOC_INVAL);
3756 
3757 		value = sizeof (lb_normal);
3758 		if (hw->phy.media_type == e1000_media_type_copper)
3759 			value += sizeof (lb_phy);
3760 		else
3761 			value += sizeof (lb_serdes);
3762 		value += sizeof (lb_external);
3763 
3764 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
3765 		*lbsp = value;
3766 		break;
3767 
3768 	case LB_GET_INFO:
3769 		value = sizeof (lb_normal);
3770 		if (hw->phy.media_type == e1000_media_type_copper)
3771 			value += sizeof (lb_phy);
3772 		else
3773 			value += sizeof (lb_serdes);
3774 		value += sizeof (lb_external);
3775 
3776 		size = value;
3777 		if (iocp->ioc_count != size)
3778 			return (IOC_INVAL);
3779 
3780 		value = 0;
3781 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
3782 
3783 		lbpp[value++] = lb_normal;
3784 		if (hw->phy.media_type == e1000_media_type_copper)
3785 			lbpp[value++] = lb_phy;
3786 		else
3787 			lbpp[value++] = lb_serdes;
3788 		lbpp[value++] = lb_external;
3789 		break;
3790 
3791 	case LB_GET_MODE:
3792 		size = sizeof (uint32_t);
3793 		if (iocp->ioc_count != size)
3794 			return (IOC_INVAL);
3795 
3796 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3797 		*lbmp = igb->loopback_mode;
3798 		break;
3799 
3800 	case LB_SET_MODE:
3801 		size = 0;
3802 		if (iocp->ioc_count != sizeof (uint32_t))
3803 			return (IOC_INVAL);
3804 
3805 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3806 		if (!igb_set_loopback_mode(igb, *lbmp))
3807 			return (IOC_INVAL);
3808 		break;
3809 	}
3810 
3811 	iocp->ioc_count = size;
3812 	iocp->ioc_error = 0;
3813 
3814 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3815 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3816 		return (IOC_INVAL);
3817 	}
3818 
3819 	return (IOC_REPLY);
3820 }
3821 
3822 /*
3823  * igb_set_loopback_mode - Setup loopback based on the loopback mode
3824  */
3825 static boolean_t
3826 igb_set_loopback_mode(igb_t *igb, uint32_t mode)
3827 {
3828 	struct e1000_hw *hw;
3829 	int i;
3830 
3831 	if (mode == igb->loopback_mode)
3832 		return (B_TRUE);
3833 
3834 	hw = &igb->hw;
3835 
3836 	igb->loopback_mode = mode;
3837 
3838 	if (mode == IGB_LB_NONE) {
3839 		/* Reset the chip */
3840 		hw->phy.autoneg_wait_to_complete = B_TRUE;
3841 		(void) igb_reset(igb);
3842 		hw->phy.autoneg_wait_to_complete = B_FALSE;
3843 		return (B_TRUE);
3844 	}
3845 
3846 	mutex_enter(&igb->gen_lock);
3847 
3848 	switch (mode) {
3849 	default:
3850 		mutex_exit(&igb->gen_lock);
3851 		return (B_FALSE);
3852 
3853 	case IGB_LB_EXTERNAL:
3854 		igb_set_external_loopback(igb);
3855 		break;
3856 
3857 	case IGB_LB_INTERNAL_PHY:
3858 		igb_set_internal_phy_loopback(igb);
3859 		break;
3860 
3861 	case IGB_LB_INTERNAL_SERDES:
3862 		igb_set_internal_serdes_loopback(igb);
3863 		break;
3864 	}
3865 
3866 	mutex_exit(&igb->gen_lock);
3867 
3868 	/*
3869 	 * When external loopback is set, wait up to 1000ms to get the link up.
3870 	 * According to test, 1000ms can work and it's an experimental value.
3871 	 */
3872 	if (mode == IGB_LB_EXTERNAL) {
3873 		for (i = 0; i <= 10; i++) {
3874 			mutex_enter(&igb->gen_lock);
3875 			(void) igb_link_check(igb);
3876 			mutex_exit(&igb->gen_lock);
3877 
3878 			if (igb->link_state == LINK_STATE_UP)
3879 				break;
3880 
3881 			msec_delay(100);
3882 		}
3883 
3884 		if (igb->link_state != LINK_STATE_UP) {
3885 			/*
3886 			 * Does not support external loopback.
3887 			 * Reset driver to loopback none.
3888 			 */
3889 			igb->loopback_mode = IGB_LB_NONE;
3890 
3891 			/* Reset the chip */
3892 			hw->phy.autoneg_wait_to_complete = B_TRUE;
3893 			(void) igb_reset(igb);
3894 			hw->phy.autoneg_wait_to_complete = B_FALSE;
3895 
3896 			igb_log(igb, IGB_LOG_INFO, "Set external loopback "
3897 			    "failed, reset to loopback none.");
3898 
3899 			return (B_FALSE);
3900 		}
3901 	}
3902 
3903 	return (B_TRUE);
3904 }
3905 
3906 /*
3907  * igb_set_external_loopback - Set the external loopback mode
3908  */
3909 static void
3910 igb_set_external_loopback(igb_t *igb)
3911 {
3912 	struct e1000_hw *hw;
3913 	uint32_t ctrl_ext;
3914 
3915 	hw = &igb->hw;
3916 
3917 	/* Set link mode to PHY (00b) in the Extended Control register */
3918 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3919 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3920 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3921 
3922 	(void) e1000_write_phy_reg(hw, 0x0, 0x0140);
3923 	(void) e1000_write_phy_reg(hw, 0x9, 0x1a00);
3924 	(void) e1000_write_phy_reg(hw, 0x12, 0x1610);
3925 	(void) e1000_write_phy_reg(hw, 0x1f37, 0x3f1c);
3926 }
3927 
3928 /*
3929  * igb_set_internal_phy_loopback - Set the internal PHY loopback mode
3930  */
3931 static void
3932 igb_set_internal_phy_loopback(igb_t *igb)
3933 {
3934 	struct e1000_hw *hw;
3935 	uint32_t ctrl_ext;
3936 	uint16_t phy_ctrl;
3937 	uint16_t phy_pconf;
3938 
3939 	hw = &igb->hw;
3940 
3941 	/* Set link mode to PHY (00b) in the Extended Control register */
3942 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3943 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3944 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3945 
3946 	/*
3947 	 * Set PHY control register (0x4140):
3948 	 *    Set full duplex mode
3949 	 *    Set loopback bit
3950 	 *    Clear auto-neg enable bit
3951 	 *    Set PHY speed
3952 	 */
3953 	phy_ctrl = MII_CR_FULL_DUPLEX | MII_CR_SPEED_1000 | MII_CR_LOOPBACK;
3954 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3955 
3956 	/* Set the link disable bit in the Port Configuration register */
3957 	(void) e1000_read_phy_reg(hw, 0x10, &phy_pconf);
3958 	phy_pconf |= (uint16_t)1 << 14;
3959 	(void) e1000_write_phy_reg(hw, 0x10, phy_pconf);
3960 }
3961 
3962 /*
3963  * igb_set_internal_serdes_loopback - Set the internal SerDes loopback mode
3964  */
3965 static void
3966 igb_set_internal_serdes_loopback(igb_t *igb)
3967 {
3968 	struct e1000_hw *hw;
3969 	uint32_t ctrl_ext;
3970 	uint32_t ctrl;
3971 	uint32_t pcs_lctl;
3972 	uint32_t connsw;
3973 
3974 	hw = &igb->hw;
3975 
3976 	/* Set link mode to SerDes (11b) in the Extended Control register */
3977 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3978 	ctrl_ext |= E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
3979 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3980 
3981 	/* Configure the SerDes to loopback */
3982 	E1000_WRITE_REG(hw, E1000_SCTL, 0x410);
3983 
3984 	/* Set Device Control register */
3985 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3986 	ctrl |= (E1000_CTRL_FD |	/* Force full duplex */
3987 	    E1000_CTRL_SLU);		/* Force link up */
3988 	ctrl &= ~(E1000_CTRL_RFCE |	/* Disable receive flow control */
3989 	    E1000_CTRL_TFCE |		/* Disable transmit flow control */
3990 	    E1000_CTRL_LRST);		/* Clear link reset */
3991 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3992 
3993 	/* Set PCS Link Control register */
3994 	pcs_lctl = E1000_READ_REG(hw, E1000_PCS_LCTL);
3995 	pcs_lctl |= (E1000_PCS_LCTL_FORCE_LINK |
3996 	    E1000_PCS_LCTL_FSD |
3997 	    E1000_PCS_LCTL_FDV_FULL |
3998 	    E1000_PCS_LCTL_FLV_LINK_UP);
3999 	pcs_lctl &= ~E1000_PCS_LCTL_AN_ENABLE;
4000 	E1000_WRITE_REG(hw, E1000_PCS_LCTL, pcs_lctl);
4001 
4002 	/* Set the Copper/Fiber Switch Control - CONNSW register */
4003 	connsw = E1000_READ_REG(hw, E1000_CONNSW);
4004 	connsw &= ~E1000_CONNSW_ENRGSRC;
4005 	E1000_WRITE_REG(hw, E1000_CONNSW, connsw);
4006 }
4007 
4008 #pragma inline(igb_intr_rx_work)
4009 /*
4010  * igb_intr_rx_work - rx processing of ISR
4011  */
4012 static void
4013 igb_intr_rx_work(igb_rx_ring_t *rx_ring)
4014 {
4015 	mblk_t *mp;
4016 
4017 	mutex_enter(&rx_ring->rx_lock);
4018 	mp = igb_rx(rx_ring, IGB_NO_POLL);
4019 	mutex_exit(&rx_ring->rx_lock);
4020 
4021 	if (mp != NULL)
4022 		mac_rx_ring(rx_ring->igb->mac_hdl, rx_ring->ring_handle, mp,
4023 		    rx_ring->ring_gen_num);
4024 }
4025 
4026 #pragma inline(igb_intr_tx_work)
4027 /*
4028  * igb_intr_tx_work - tx processing of ISR
4029  */
4030 static void
4031 igb_intr_tx_work(igb_tx_ring_t *tx_ring)
4032 {
4033 	igb_t *igb = tx_ring->igb;
4034 
4035 	/* Recycle the tx descriptors */
4036 	tx_ring->tx_recycle(tx_ring);
4037 
4038 	/* Schedule the re-transmit */
4039 	if (tx_ring->reschedule &&
4040 	    (tx_ring->tbd_free >= igb->tx_resched_thresh)) {
4041 		tx_ring->reschedule = B_FALSE;
4042 		mac_tx_ring_update(tx_ring->igb->mac_hdl, tx_ring->ring_handle);
4043 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
4044 	}
4045 }
4046 
4047 #pragma inline(igb_intr_link_work)
4048 /*
4049  * igb_intr_link_work - link-status-change processing of ISR
4050  */
4051 static void
4052 igb_intr_link_work(igb_t *igb)
4053 {
4054 	boolean_t link_changed;
4055 
4056 	igb_stop_watchdog_timer(igb);
4057 
4058 	mutex_enter(&igb->gen_lock);
4059 
4060 	/*
4061 	 * Because we got a link-status-change interrupt, force
4062 	 * e1000_check_for_link() to look at phy
4063 	 */
4064 	igb->hw.mac.get_link_status = B_TRUE;
4065 
4066 	/* igb_link_check takes care of link status change */
4067 	link_changed = igb_link_check(igb);
4068 
4069 	/* Get new phy state */
4070 	igb_get_phy_state(igb);
4071 
4072 	mutex_exit(&igb->gen_lock);
4073 
4074 	if (link_changed)
4075 		mac_link_update(igb->mac_hdl, igb->link_state);
4076 
4077 	igb_start_watchdog_timer(igb);
4078 }
4079 
4080 /*
4081  * igb_intr_legacy - Interrupt handler for legacy interrupts
4082  */
4083 static uint_t
4084 igb_intr_legacy(void *arg1, void *arg2)
4085 {
4086 	igb_t *igb = (igb_t *)arg1;
4087 	igb_tx_ring_t *tx_ring;
4088 	uint32_t icr;
4089 	mblk_t *mp;
4090 	boolean_t tx_reschedule;
4091 	boolean_t link_changed;
4092 	uint_t result;
4093 
4094 	_NOTE(ARGUNUSED(arg2));
4095 
4096 	mutex_enter(&igb->gen_lock);
4097 
4098 	if (igb->igb_state & IGB_SUSPENDED) {
4099 		mutex_exit(&igb->gen_lock);
4100 		return (DDI_INTR_UNCLAIMED);
4101 	}
4102 
4103 	mp = NULL;
4104 	tx_reschedule = B_FALSE;
4105 	link_changed = B_FALSE;
4106 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
4107 
4108 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
4109 		mutex_exit(&igb->gen_lock);
4110 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
4111 		atomic_or_32(&igb->igb_state, IGB_ERROR);
4112 		return (DDI_INTR_UNCLAIMED);
4113 	}
4114 
4115 	if (icr & E1000_ICR_INT_ASSERTED) {
4116 		/*
4117 		 * E1000_ICR_INT_ASSERTED bit was set:
4118 		 * Read(Clear) the ICR, claim this interrupt,
4119 		 * look for work to do.
4120 		 */
4121 		ASSERT(igb->num_rx_rings == 1);
4122 		ASSERT(igb->num_tx_rings == 1);
4123 
4124 		/* Make sure all interrupt causes cleared */
4125 		(void) E1000_READ_REG(&igb->hw, E1000_EICR);
4126 
4127 		if (icr & E1000_ICR_RXT0) {
4128 			mp = igb_rx(&igb->rx_rings[0], IGB_NO_POLL);
4129 		}
4130 
4131 		if (icr & E1000_ICR_TXDW) {
4132 			tx_ring = &igb->tx_rings[0];
4133 
4134 			/* Recycle the tx descriptors */
4135 			tx_ring->tx_recycle(tx_ring);
4136 
4137 			/* Schedule the re-transmit */
4138 			tx_reschedule = (tx_ring->reschedule &&
4139 			    (tx_ring->tbd_free >= igb->tx_resched_thresh));
4140 		}
4141 
4142 		if (icr & E1000_ICR_LSC) {
4143 			/*
4144 			 * Because we got a link-status-change interrupt, force
4145 			 * e1000_check_for_link() to look at phy
4146 			 */
4147 			igb->hw.mac.get_link_status = B_TRUE;
4148 
4149 			/* igb_link_check takes care of link status change */
4150 			link_changed = igb_link_check(igb);
4151 
4152 			/* Get new phy state */
4153 			igb_get_phy_state(igb);
4154 		}
4155 
4156 		if (icr & E1000_ICR_DRSTA) {
4157 			/* 82580 Full Device Reset needed */
4158 			atomic_or_32(&igb->igb_state, IGB_STALL);
4159 		}
4160 
4161 		result = DDI_INTR_CLAIMED;
4162 	} else {
4163 		/*
4164 		 * E1000_ICR_INT_ASSERTED bit was not set:
4165 		 * Don't claim this interrupt.
4166 		 */
4167 		result = DDI_INTR_UNCLAIMED;
4168 	}
4169 
4170 	mutex_exit(&igb->gen_lock);
4171 
4172 	/*
4173 	 * Do the following work outside of the gen_lock
4174 	 */
4175 	if (mp != NULL)
4176 		mac_rx(igb->mac_hdl, NULL, mp);
4177 
4178 	if (tx_reschedule)  {
4179 		tx_ring->reschedule = B_FALSE;
4180 		mac_tx_ring_update(igb->mac_hdl, tx_ring->ring_handle);
4181 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
4182 	}
4183 
4184 	if (link_changed)
4185 		mac_link_update(igb->mac_hdl, igb->link_state);
4186 
4187 	return (result);
4188 }
4189 
4190 /*
4191  * igb_intr_msi - Interrupt handler for MSI
4192  */
4193 static uint_t
4194 igb_intr_msi(void *arg1, void *arg2)
4195 {
4196 	igb_t *igb = (igb_t *)arg1;
4197 	uint32_t icr;
4198 
4199 	_NOTE(ARGUNUSED(arg2));
4200 
4201 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
4202 
4203 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
4204 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
4205 		atomic_or_32(&igb->igb_state, IGB_ERROR);
4206 		return (DDI_INTR_CLAIMED);
4207 	}
4208 
4209 	/* Make sure all interrupt causes cleared */
4210 	(void) E1000_READ_REG(&igb->hw, E1000_EICR);
4211 
4212 	/*
4213 	 * For MSI interrupt, we have only one vector,
4214 	 * so we have only one rx ring and one tx ring enabled.
4215 	 */
4216 	ASSERT(igb->num_rx_rings == 1);
4217 	ASSERT(igb->num_tx_rings == 1);
4218 
4219 	if (icr & E1000_ICR_RXT0) {
4220 		igb_intr_rx_work(&igb->rx_rings[0]);
4221 	}
4222 
4223 	if (icr & E1000_ICR_TXDW) {
4224 		igb_intr_tx_work(&igb->tx_rings[0]);
4225 	}
4226 
4227 	if (icr & E1000_ICR_LSC) {
4228 		igb_intr_link_work(igb);
4229 	}
4230 
4231 	if (icr & E1000_ICR_DRSTA) {
4232 		/* 82580 Full Device Reset needed */
4233 		atomic_or_32(&igb->igb_state, IGB_STALL);
4234 	}
4235 
4236 	return (DDI_INTR_CLAIMED);
4237 }
4238 
4239 /*
4240  * igb_intr_rx - Interrupt handler for rx
4241  */
4242 static uint_t
4243 igb_intr_rx(void *arg1, void *arg2)
4244 {
4245 	igb_rx_ring_t *rx_ring = (igb_rx_ring_t *)arg1;
4246 
4247 	_NOTE(ARGUNUSED(arg2));
4248 
4249 	/*
4250 	 * Only used via MSI-X vector so don't check cause bits
4251 	 * and only clean the given ring.
4252 	 */
4253 	igb_intr_rx_work(rx_ring);
4254 
4255 	return (DDI_INTR_CLAIMED);
4256 }
4257 
4258 /*
4259  * igb_intr_tx - Interrupt handler for tx
4260  */
4261 static uint_t
4262 igb_intr_tx(void *arg1, void *arg2)
4263 {
4264 	igb_tx_ring_t *tx_ring = (igb_tx_ring_t *)arg1;
4265 
4266 	_NOTE(ARGUNUSED(arg2));
4267 
4268 	/*
4269 	 * Only used via MSI-X vector so don't check cause bits
4270 	 * and only clean the given ring.
4271 	 */
4272 	igb_intr_tx_work(tx_ring);
4273 
4274 	return (DDI_INTR_CLAIMED);
4275 }
4276 
4277 /*
4278  * igb_intr_tx_other - Interrupt handler for both tx and other
4279  *
4280  */
4281 static uint_t
4282 igb_intr_tx_other(void *arg1, void *arg2)
4283 {
4284 	igb_t *igb = (igb_t *)arg1;
4285 	uint32_t icr;
4286 
4287 	_NOTE(ARGUNUSED(arg2));
4288 
4289 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
4290 
4291 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
4292 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
4293 		atomic_or_32(&igb->igb_state, IGB_ERROR);
4294 		return (DDI_INTR_CLAIMED);
4295 	}
4296 
4297 	/*
4298 	 * Look for tx reclaiming work first. Remember, in the
4299 	 * case of only interrupt sharing, only one tx ring is
4300 	 * used
4301 	 */
4302 	igb_intr_tx_work(&igb->tx_rings[0]);
4303 
4304 	/*
4305 	 * Check for "other" causes.
4306 	 */
4307 	if (icr & E1000_ICR_LSC) {
4308 		igb_intr_link_work(igb);
4309 	}
4310 
4311 	/*
4312 	 * The DOUTSYNC bit indicates a tx packet dropped because
4313 	 * DMA engine gets "out of sync". There isn't a real fix
4314 	 * for this. The Intel recommendation is to count the number
4315 	 * of occurrences so user can detect when it is happening.
4316 	 * The issue is non-fatal and there's no recovery action
4317 	 * available.
4318 	 */
4319 	if (icr & E1000_ICR_DOUTSYNC) {
4320 		IGB_STAT(igb->dout_sync);
4321 	}
4322 
4323 	if (icr & E1000_ICR_DRSTA) {
4324 		/* 82580 Full Device Reset needed */
4325 		atomic_or_32(&igb->igb_state, IGB_STALL);
4326 	}
4327 
4328 	return (DDI_INTR_CLAIMED);
4329 }
4330 
4331 /*
4332  * igb_alloc_intrs - Allocate interrupts for the driver
4333  *
4334  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
4335  * if not successful, try Legacy.
4336  * igb->intr_force can be used to force sequence to start with
4337  * any of the 3 types.
4338  * If MSI-X is not used, number of tx/rx rings is forced to 1.
4339  */
4340 static int
4341 igb_alloc_intrs(igb_t *igb)
4342 {
4343 	dev_info_t *devinfo;
4344 	int intr_types;
4345 	int rc;
4346 
4347 	devinfo = igb->dip;
4348 
4349 	/* Get supported interrupt types */
4350 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
4351 
4352 	if (rc != DDI_SUCCESS) {
4353 		igb_log(igb, IGB_LOG_ERROR,
4354 		    "Get supported interrupt types failed: %d", rc);
4355 		return (IGB_FAILURE);
4356 	}
4357 	igb_log(igb, IGB_LOG_INFO, "Supported interrupt types: %x",
4358 	    intr_types);
4359 
4360 	igb->intr_type = 0;
4361 
4362 	/* Install MSI-X interrupts */
4363 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
4364 	    (igb->intr_force <= IGB_INTR_MSIX)) {
4365 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSIX);
4366 
4367 		if (rc == IGB_SUCCESS)
4368 			return (IGB_SUCCESS);
4369 
4370 		igb_log(igb, IGB_LOG_INFO,
4371 		    "Allocate MSI-X failed, trying MSI interrupts...");
4372 	}
4373 
4374 	/* MSI-X not used, force rings to 1 */
4375 	igb->num_rx_rings = 1;
4376 	igb->num_tx_rings = 1;
4377 	igb_log(igb, IGB_LOG_INFO,
4378 	    "MSI-X not used, force rx and tx queue number to 1");
4379 
4380 	/* Install MSI interrupts */
4381 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
4382 	    (igb->intr_force <= IGB_INTR_MSI)) {
4383 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSI);
4384 
4385 		if (rc == IGB_SUCCESS)
4386 			return (IGB_SUCCESS);
4387 
4388 		igb_log(igb, IGB_LOG_INFO,
4389 		    "Allocate MSI failed, trying Legacy interrupts...");
4390 	}
4391 
4392 	/* Install legacy interrupts */
4393 	if (intr_types & DDI_INTR_TYPE_FIXED) {
4394 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_FIXED);
4395 
4396 		if (rc == IGB_SUCCESS)
4397 			return (IGB_SUCCESS);
4398 
4399 		igb_log(igb, IGB_LOG_INFO,
4400 		    "Allocate Legacy interrupts failed");
4401 	}
4402 
4403 	/* If none of the 3 types succeeded, return failure */
4404 	return (IGB_FAILURE);
4405 }
4406 
4407 /*
4408  * igb_alloc_intr_handles - Allocate interrupt handles.
4409  *
4410  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
4411  * if fewer than 2 handles are available, return failure.
4412  * Upon success, this sets the number of Rx rings to a number that
4413  * matches the handles available for Rx interrupts.
4414  */
4415 static int
4416 igb_alloc_intr_handles(igb_t *igb, int intr_type)
4417 {
4418 	dev_info_t *devinfo;
4419 	int orig, request, count, avail, actual;
4420 	int diff, minimum;
4421 	int rc;
4422 
4423 	devinfo = igb->dip;
4424 
4425 	switch (intr_type) {
4426 	case DDI_INTR_TYPE_FIXED:
4427 		request = 1;	/* Request 1 legacy interrupt handle */
4428 		minimum = 1;
4429 		igb_log(igb, IGB_LOG_INFO, "interrupt type: legacy");
4430 		break;
4431 
4432 	case DDI_INTR_TYPE_MSI:
4433 		request = 1;	/* Request 1 MSI interrupt handle */
4434 		minimum = 1;
4435 		igb_log(igb, IGB_LOG_INFO, "interrupt type: MSI");
4436 		break;
4437 
4438 	case DDI_INTR_TYPE_MSIX:
4439 		/*
4440 		 * Number of vectors for the adapter is
4441 		 * # rx rings + # tx rings
4442 		 * One of tx vectors is for tx & other
4443 		 */
4444 		request = igb->num_rx_rings + igb->num_tx_rings;
4445 		orig = request;
4446 		minimum = 2;
4447 		igb_log(igb, IGB_LOG_INFO, "interrupt type: MSI-X");
4448 		break;
4449 
4450 	default:
4451 		igb_log(igb, IGB_LOG_INFO,
4452 		    "invalid call to igb_alloc_intr_handles(): %d\n",
4453 		    intr_type);
4454 		return (IGB_FAILURE);
4455 	}
4456 	igb_log(igb, IGB_LOG_INFO,
4457 	    "interrupt handles requested: %d  minimum: %d",
4458 	    request, minimum);
4459 
4460 	/*
4461 	 * Get number of supported interrupts
4462 	 */
4463 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
4464 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
4465 		igb_log(igb, IGB_LOG_INFO,
4466 		    "Get supported interrupt number failed. "
4467 		    "Return: %d, count: %d", rc, count);
4468 		return (IGB_FAILURE);
4469 	}
4470 	igb_log(igb, IGB_LOG_INFO, "interrupts supported: %d", count);
4471 
4472 	/*
4473 	 * Get number of available interrupts
4474 	 */
4475 	rc = ddi_intr_get_navail(devinfo, intr_type, &avail);
4476 	if ((rc != DDI_SUCCESS) || (avail < minimum)) {
4477 		igb_log(igb, IGB_LOG_INFO,
4478 		    "Get available interrupt number failed. "
4479 		    "Return: %d, available: %d", rc, avail);
4480 		return (IGB_FAILURE);
4481 	}
4482 	igb_log(igb, IGB_LOG_INFO, "interrupts available: %d", avail);
4483 
4484 	if (avail < request) {
4485 		igb_log(igb, IGB_LOG_INFO,
4486 		    "Request %d handles, %d available",
4487 		    request, avail);
4488 		request = avail;
4489 	}
4490 
4491 	actual = 0;
4492 	igb->intr_cnt = 0;
4493 
4494 	/*
4495 	 * Allocate an array of interrupt handles
4496 	 */
4497 	igb->intr_size = request * sizeof (ddi_intr_handle_t);
4498 	igb->htable = kmem_alloc(igb->intr_size, KM_SLEEP);
4499 
4500 	rc = ddi_intr_alloc(devinfo, igb->htable, intr_type, 0,
4501 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
4502 	if (rc != DDI_SUCCESS) {
4503 		igb_log(igb, IGB_LOG_INFO, "Allocate interrupts failed. "
4504 		    "return: %d, request: %d, actual: %d",
4505 		    rc, request, actual);
4506 		goto alloc_handle_fail;
4507 	}
4508 	igb_log(igb, IGB_LOG_INFO, "interrupts actually allocated: %d", actual);
4509 
4510 	igb->intr_cnt = actual;
4511 
4512 	if (actual < minimum) {
4513 		igb_log(igb, IGB_LOG_INFO,
4514 		    "Insufficient interrupt handles allocated: %d",
4515 		    actual);
4516 		goto alloc_handle_fail;
4517 	}
4518 
4519 	/*
4520 	 * For MSI-X, actual might force us to reduce number of tx & rx rings
4521 	 */
4522 	if ((intr_type == DDI_INTR_TYPE_MSIX) && (orig > actual)) {
4523 		diff = orig - actual;
4524 		if (diff < igb->num_tx_rings) {
4525 			igb_log(igb, IGB_LOG_INFO,
4526 			    "MSI-X vectors force Tx queue number to %d",
4527 			    igb->num_tx_rings - diff);
4528 			igb->num_tx_rings -= diff;
4529 		} else {
4530 			igb_log(igb, IGB_LOG_INFO,
4531 			    "MSI-X vectors force Tx queue number to 1");
4532 			igb->num_tx_rings = 1;
4533 
4534 			igb_log(igb, IGB_LOG_INFO,
4535 			    "MSI-X vectors force Rx queue number to %d",
4536 			    actual - 1);
4537 			igb->num_rx_rings = actual - 1;
4538 		}
4539 	}
4540 
4541 	/*
4542 	 * Get priority for first vector, assume remaining are all the same
4543 	 */
4544 	rc = ddi_intr_get_pri(igb->htable[0], &igb->intr_pri);
4545 	if (rc != DDI_SUCCESS) {
4546 		igb_log(igb, IGB_LOG_INFO,
4547 		    "Get interrupt priority failed: %d", rc);
4548 		goto alloc_handle_fail;
4549 	}
4550 
4551 	rc = ddi_intr_get_cap(igb->htable[0], &igb->intr_cap);
4552 	if (rc != DDI_SUCCESS) {
4553 		igb_log(igb, IGB_LOG_INFO,
4554 		    "Get interrupt cap failed: %d", rc);
4555 		goto alloc_handle_fail;
4556 	}
4557 
4558 	igb->intr_type = intr_type;
4559 
4560 	return (IGB_SUCCESS);
4561 
4562 alloc_handle_fail:
4563 	igb_rem_intrs(igb);
4564 
4565 	return (IGB_FAILURE);
4566 }
4567 
4568 /*
4569  * igb_add_intr_handlers - Add interrupt handlers based on the interrupt type
4570  *
4571  * Before adding the interrupt handlers, the interrupt vectors have
4572  * been allocated, and the rx/tx rings have also been allocated.
4573  */
4574 static int
4575 igb_add_intr_handlers(igb_t *igb)
4576 {
4577 	igb_rx_ring_t *rx_ring;
4578 	igb_tx_ring_t *tx_ring;
4579 	int vector;
4580 	int rc;
4581 	int i;
4582 
4583 	vector = 0;
4584 
4585 	switch (igb->intr_type) {
4586 	case DDI_INTR_TYPE_MSIX:
4587 		/* Add interrupt handler for tx + other */
4588 		tx_ring = &igb->tx_rings[0];
4589 		rc = ddi_intr_add_handler(igb->htable[vector],
4590 		    (ddi_intr_handler_t *)igb_intr_tx_other,
4591 		    (void *)igb, NULL);
4592 
4593 		if (rc != DDI_SUCCESS) {
4594 			igb_log(igb, IGB_LOG_INFO,
4595 			    "Add tx/other interrupt handler failed: %d", rc);
4596 			return (IGB_FAILURE);
4597 		}
4598 		tx_ring->intr_vector = vector;
4599 		vector++;
4600 
4601 		/* Add interrupt handler for each rx ring */
4602 		for (i = 0; i < igb->num_rx_rings; i++) {
4603 			rx_ring = &igb->rx_rings[i];
4604 
4605 			rc = ddi_intr_add_handler(igb->htable[vector],
4606 			    (ddi_intr_handler_t *)igb_intr_rx,
4607 			    (void *)rx_ring, NULL);
4608 
4609 			if (rc != DDI_SUCCESS) {
4610 				igb_log(igb, IGB_LOG_INFO,
4611 				    "Add rx interrupt handler failed. "
4612 				    "return: %d, rx ring: %d", rc, i);
4613 				for (vector--; vector >= 0; vector--) {
4614 					(void) ddi_intr_remove_handler(
4615 					    igb->htable[vector]);
4616 				}
4617 				return (IGB_FAILURE);
4618 			}
4619 
4620 			rx_ring->intr_vector = vector;
4621 
4622 			vector++;
4623 		}
4624 
4625 		/* Add interrupt handler for each tx ring from 2nd ring */
4626 		for (i = 1; i < igb->num_tx_rings; i++) {
4627 			tx_ring = &igb->tx_rings[i];
4628 
4629 			rc = ddi_intr_add_handler(igb->htable[vector],
4630 			    (ddi_intr_handler_t *)igb_intr_tx,
4631 			    (void *)tx_ring, NULL);
4632 
4633 			if (rc != DDI_SUCCESS) {
4634 				igb_log(igb, IGB_LOG_INFO,
4635 				    "Add tx interrupt handler failed. "
4636 				    "return: %d, tx ring: %d", rc, i);
4637 				for (vector--; vector >= 0; vector--) {
4638 					(void) ddi_intr_remove_handler(
4639 					    igb->htable[vector]);
4640 				}
4641 				return (IGB_FAILURE);
4642 			}
4643 
4644 			tx_ring->intr_vector = vector;
4645 
4646 			vector++;
4647 		}
4648 
4649 		break;
4650 
4651 	case DDI_INTR_TYPE_MSI:
4652 		/* Add interrupt handlers for the only vector */
4653 		rc = ddi_intr_add_handler(igb->htable[vector],
4654 		    (ddi_intr_handler_t *)igb_intr_msi,
4655 		    (void *)igb, NULL);
4656 
4657 		if (rc != DDI_SUCCESS) {
4658 			igb_log(igb, IGB_LOG_INFO,
4659 			    "Add MSI interrupt handler failed: %d", rc);
4660 			return (IGB_FAILURE);
4661 		}
4662 
4663 		rx_ring = &igb->rx_rings[0];
4664 		rx_ring->intr_vector = vector;
4665 
4666 		vector++;
4667 		break;
4668 
4669 	case DDI_INTR_TYPE_FIXED:
4670 		/* Add interrupt handlers for the only vector */
4671 		rc = ddi_intr_add_handler(igb->htable[vector],
4672 		    (ddi_intr_handler_t *)igb_intr_legacy,
4673 		    (void *)igb, NULL);
4674 
4675 		if (rc != DDI_SUCCESS) {
4676 			igb_log(igb, IGB_LOG_INFO,
4677 			    "Add legacy interrupt handler failed: %d", rc);
4678 			return (IGB_FAILURE);
4679 		}
4680 
4681 		rx_ring = &igb->rx_rings[0];
4682 		rx_ring->intr_vector = vector;
4683 
4684 		vector++;
4685 		break;
4686 
4687 	default:
4688 		return (IGB_FAILURE);
4689 	}
4690 
4691 	ASSERT(vector == igb->intr_cnt);
4692 
4693 	return (IGB_SUCCESS);
4694 }
4695 
4696 /*
4697  * igb_setup_msix_82575 - setup 82575 adapter to use MSI-X interrupts
4698  *
4699  * For each vector enabled on the adapter, Set the MSIXBM register accordingly
4700  */
4701 static void
4702 igb_setup_msix_82575(igb_t *igb)
4703 {
4704 	uint32_t eims = 0;
4705 	int i, vector;
4706 	struct e1000_hw *hw = &igb->hw;
4707 
4708 	/*
4709 	 * Set vector for tx ring 0 and other causes.
4710 	 * NOTE assumption that it is vector 0.
4711 	 */
4712 	vector = 0;
4713 
4714 	igb->eims_mask = E1000_EICR_TX_QUEUE0 | E1000_EICR_OTHER;
4715 	E1000_WRITE_REG(hw, E1000_MSIXBM(vector), igb->eims_mask);
4716 	vector++;
4717 
4718 	for (i = 0; i < igb->num_rx_rings; i++) {
4719 		/*
4720 		 * Set vector for each rx ring
4721 		 */
4722 		eims = (E1000_EICR_RX_QUEUE0 << i);
4723 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4724 
4725 		/*
4726 		 * Accumulate bits to enable in
4727 		 * igb_enable_adapter_interrupts_82575()
4728 		 */
4729 		igb->eims_mask |= eims;
4730 
4731 		vector++;
4732 	}
4733 
4734 	for (i = 1; i < igb->num_tx_rings; i++) {
4735 		/*
4736 		 * Set vector for each tx ring from 2nd tx ring
4737 		 */
4738 		eims = (E1000_EICR_TX_QUEUE0 << i);
4739 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4740 
4741 		/*
4742 		 * Accumulate bits to enable in
4743 		 * igb_enable_adapter_interrupts_82575()
4744 		 */
4745 		igb->eims_mask |= eims;
4746 
4747 		vector++;
4748 	}
4749 
4750 	ASSERT(vector == igb->intr_cnt);
4751 
4752 	/*
4753 	 * Disable IAM for ICR interrupt bits
4754 	 */
4755 	E1000_WRITE_REG(hw, E1000_IAM, 0);
4756 	E1000_WRITE_FLUSH(hw);
4757 }
4758 
4759 /*
4760  * igb_setup_msix_82576 - setup 82576 adapter to use MSI-X interrupts
4761  *
4762  * 82576 uses a table based method for assigning vectors.  Each queue has a
4763  * single entry in the table to which we write a vector number along with a
4764  * "valid" bit.  The entry is a single byte in a 4-byte register.  Vectors
4765  * take a different position in the 4-byte register depending on whether
4766  * they are numbered above or below 8.
4767  */
4768 static void
4769 igb_setup_msix_82576(igb_t *igb)
4770 {
4771 	struct e1000_hw *hw = &igb->hw;
4772 	uint32_t ivar, index, vector;
4773 	int i;
4774 
4775 	/* must enable msi-x capability before IVAR settings */
4776 	E1000_WRITE_REG(hw, E1000_GPIE,
4777 	    (E1000_GPIE_MSIX_MODE | E1000_GPIE_PBA | E1000_GPIE_NSICR));
4778 
4779 	/*
4780 	 * Set vector for tx ring 0 and other causes.
4781 	 * NOTE assumption that it is vector 0.
4782 	 * This is also interdependent with installation of interrupt service
4783 	 * routines in igb_add_intr_handlers().
4784 	 */
4785 
4786 	/* assign "other" causes to vector 0 */
4787 	vector = 0;
4788 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4789 	E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4790 
4791 	/* assign tx ring 0 to vector 0 */
4792 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4793 	E1000_WRITE_REG(hw, E1000_IVAR0, ivar);
4794 
4795 	/* prepare to enable tx & other interrupt causes */
4796 	igb->eims_mask = (1 << vector);
4797 
4798 	vector ++;
4799 	for (i = 0; i < igb->num_rx_rings; i++) {
4800 		/*
4801 		 * Set vector for each rx ring
4802 		 */
4803 		index = (i & 0x7);
4804 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4805 
4806 		if (i < 8) {
4807 			/* vector goes into low byte of register */
4808 			ivar = ivar & 0xFFFFFF00;
4809 			ivar |= (vector | E1000_IVAR_VALID);
4810 		} else {
4811 			/* vector goes into third byte of register */
4812 			ivar = ivar & 0xFF00FFFF;
4813 			ivar |= ((vector | E1000_IVAR_VALID) << 16);
4814 		}
4815 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4816 
4817 		/* Accumulate interrupt-cause bits to enable */
4818 		igb->eims_mask |= (1 << vector);
4819 
4820 		vector ++;
4821 	}
4822 
4823 	for (i = 1; i < igb->num_tx_rings; i++) {
4824 		/*
4825 		 * Set vector for each tx ring from 2nd tx ring.
4826 		 * Note assumption that tx vectors numericall follow rx vectors.
4827 		 */
4828 		index = (i & 0x7);
4829 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4830 
4831 		if (i < 8) {
4832 			/* vector goes into second byte of register */
4833 			ivar = ivar & 0xFFFF00FF;
4834 			ivar |= ((vector | E1000_IVAR_VALID) << 8);
4835 		} else {
4836 			/* vector goes into fourth byte of register */
4837 			ivar = ivar & 0x00FFFFFF;
4838 			ivar |= (vector | E1000_IVAR_VALID) << 24;
4839 		}
4840 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4841 
4842 		/* Accumulate interrupt-cause bits to enable */
4843 		igb->eims_mask |= (1 << vector);
4844 
4845 		vector ++;
4846 	}
4847 
4848 	ASSERT(vector == igb->intr_cnt);
4849 }
4850 
4851 /*
4852  * igb_setup_msix_82580 - setup 82580 adapter to use MSI-X interrupts
4853  *
4854  * 82580 uses same table approach at 82576 but has fewer entries.  Each
4855  * queue has a single entry in the table to which we write a vector number
4856  * along with a "valid" bit.  Vectors take a different position in the
4857  * register depending on * whether * they are numbered above or below 4.
4858  */
4859 static void
4860 igb_setup_msix_82580(igb_t *igb)
4861 {
4862 	struct e1000_hw *hw = &igb->hw;
4863 	uint32_t ivar, index, vector;
4864 	int i;
4865 
4866 	/* must enable msi-x capability before IVAR settings */
4867 	E1000_WRITE_REG(hw, E1000_GPIE, (E1000_GPIE_MSIX_MODE |
4868 	    E1000_GPIE_PBA | E1000_GPIE_NSICR | E1000_GPIE_EIAME));
4869 	/*
4870 	 * Set vector for tx ring 0 and other causes.
4871 	 * NOTE assumption that it is vector 0.
4872 	 * This is also interdependent with installation of interrupt service
4873 	 * routines in igb_add_intr_handlers().
4874 	 */
4875 
4876 	/* assign "other" causes to vector 0 */
4877 	vector = 0;
4878 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4879 	E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4880 
4881 	/* assign tx ring 0 to vector 0 */
4882 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4883 	E1000_WRITE_REG(hw, E1000_IVAR0, ivar);
4884 
4885 	/* prepare to enable tx & other interrupt causes */
4886 	igb->eims_mask = (1 << vector);
4887 
4888 	vector ++;
4889 
4890 	for (i = 0; i < igb->num_rx_rings; i++) {
4891 		/*
4892 		 * Set vector for each rx ring
4893 		 */
4894 		index = (i >> 1);
4895 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4896 
4897 		if (i & 1) {
4898 			/* vector goes into third byte of register */
4899 			ivar = ivar & 0xFF00FFFF;
4900 			ivar |= ((vector | E1000_IVAR_VALID) << 16);
4901 		} else {
4902 			/* vector goes into low byte of register */
4903 			ivar = ivar & 0xFFFFFF00;
4904 			ivar |= (vector | E1000_IVAR_VALID);
4905 		}
4906 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4907 
4908 		/* Accumulate interrupt-cause bits to enable */
4909 		igb->eims_mask |= (1 << vector);
4910 
4911 		vector ++;
4912 	}
4913 
4914 	for (i = 1; i < igb->num_tx_rings; i++) {
4915 		/*
4916 		 * Set vector for each tx ring from 2nd tx ring.
4917 		 * Note assumption that tx vectors numericall follow rx vectors.
4918 		 */
4919 		index = (i >> 1);
4920 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4921 
4922 		if (i & 1) {
4923 			/* vector goes into high byte of register */
4924 			ivar = ivar & 0x00FFFFFF;
4925 			ivar |= ((vector | E1000_IVAR_VALID) << 24);
4926 		} else {
4927 			/* vector goes into second byte of register */
4928 			ivar = ivar & 0xFFFF00FF;
4929 			ivar |= (vector | E1000_IVAR_VALID) << 8;
4930 		}
4931 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4932 
4933 		/* Accumulate interrupt-cause bits to enable */
4934 		igb->eims_mask |= (1 << vector);
4935 
4936 		vector ++;
4937 	}
4938 	ASSERT(vector == igb->intr_cnt);
4939 }
4940 
4941 /*
4942  * igb_rem_intr_handlers - remove the interrupt handlers
4943  */
4944 static void
4945 igb_rem_intr_handlers(igb_t *igb)
4946 {
4947 	int i;
4948 	int rc;
4949 
4950 	for (i = 0; i < igb->intr_cnt; i++) {
4951 		rc = ddi_intr_remove_handler(igb->htable[i]);
4952 		if (rc != DDI_SUCCESS) {
4953 			igb_log(igb, IGB_LOG_INFO,
4954 			    "Remove intr handler failed: %d", rc);
4955 		}
4956 	}
4957 }
4958 
4959 /*
4960  * igb_rem_intrs - remove the allocated interrupts
4961  */
4962 static void
4963 igb_rem_intrs(igb_t *igb)
4964 {
4965 	int i;
4966 	int rc;
4967 
4968 	for (i = 0; i < igb->intr_cnt; i++) {
4969 		rc = ddi_intr_free(igb->htable[i]);
4970 		if (rc != DDI_SUCCESS) {
4971 			igb_log(igb, IGB_LOG_INFO,
4972 			    "Free intr failed: %d", rc);
4973 		}
4974 	}
4975 
4976 	kmem_free(igb->htable, igb->intr_size);
4977 	igb->htable = NULL;
4978 }
4979 
4980 /*
4981  * igb_enable_intrs - enable all the ddi interrupts
4982  */
4983 static int
4984 igb_enable_intrs(igb_t *igb)
4985 {
4986 	int i;
4987 	int rc;
4988 
4989 	/* Enable interrupts */
4990 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4991 		/* Call ddi_intr_block_enable() for MSI */
4992 		rc = ddi_intr_block_enable(igb->htable, igb->intr_cnt);
4993 		if (rc != DDI_SUCCESS) {
4994 			igb_log(igb, IGB_LOG_ERROR,
4995 			    "Enable block intr failed: %d", rc);
4996 			return (IGB_FAILURE);
4997 		}
4998 	} else {
4999 		/* Call ddi_intr_enable() for Legacy/MSI non block enable */
5000 		for (i = 0; i < igb->intr_cnt; i++) {
5001 			rc = ddi_intr_enable(igb->htable[i]);
5002 			if (rc != DDI_SUCCESS) {
5003 				igb_log(igb, IGB_LOG_ERROR,
5004 				    "Enable intr failed: %d", rc);
5005 				return (IGB_FAILURE);
5006 			}
5007 		}
5008 	}
5009 
5010 	return (IGB_SUCCESS);
5011 }
5012 
5013 /*
5014  * igb_disable_intrs - disable all the ddi interrupts
5015  */
5016 static int
5017 igb_disable_intrs(igb_t *igb)
5018 {
5019 	int i;
5020 	int rc;
5021 
5022 	/* Disable all interrupts */
5023 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
5024 		rc = ddi_intr_block_disable(igb->htable, igb->intr_cnt);
5025 		if (rc != DDI_SUCCESS) {
5026 			igb_log(igb, IGB_LOG_ERROR,
5027 			    "Disable block intr failed: %d", rc);
5028 			return (IGB_FAILURE);
5029 		}
5030 	} else {
5031 		for (i = 0; i < igb->intr_cnt; i++) {
5032 			rc = ddi_intr_disable(igb->htable[i]);
5033 			if (rc != DDI_SUCCESS) {
5034 				igb_log(igb, IGB_LOG_ERROR,
5035 				    "Disable intr failed: %d", rc);
5036 				return (IGB_FAILURE);
5037 			}
5038 		}
5039 	}
5040 
5041 	return (IGB_SUCCESS);
5042 }
5043 
5044 /*
5045  * igb_get_phy_state - Get and save the parameters read from PHY registers
5046  */
5047 static void
5048 igb_get_phy_state(igb_t *igb)
5049 {
5050 	struct e1000_hw *hw = &igb->hw;
5051 	uint16_t phy_ctrl;
5052 	uint16_t phy_status;
5053 	uint16_t phy_an_adv;
5054 	uint16_t phy_an_exp;
5055 	uint16_t phy_ext_status;
5056 	uint16_t phy_1000t_ctrl;
5057 	uint16_t phy_1000t_status;
5058 	uint16_t phy_lp_able;
5059 
5060 	ASSERT(mutex_owned(&igb->gen_lock));
5061 
5062 	if (hw->phy.media_type == e1000_media_type_copper) {
5063 		(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
5064 		(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
5065 		(void) e1000_read_phy_reg(hw, PHY_AUTONEG_ADV, &phy_an_adv);
5066 		(void) e1000_read_phy_reg(hw, PHY_AUTONEG_EXP, &phy_an_exp);
5067 		(void) e1000_read_phy_reg(hw, PHY_EXT_STATUS, &phy_ext_status);
5068 		(void) e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_1000t_ctrl);
5069 		(void) e1000_read_phy_reg(hw,
5070 		    PHY_1000T_STATUS, &phy_1000t_status);
5071 		(void) e1000_read_phy_reg(hw, PHY_LP_ABILITY, &phy_lp_able);
5072 
5073 		igb->param_autoneg_cap =
5074 		    (phy_status & MII_SR_AUTONEG_CAPS) ? 1 : 0;
5075 		igb->param_pause_cap =
5076 		    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
5077 		igb->param_asym_pause_cap =
5078 		    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
5079 		igb->param_1000fdx_cap =
5080 		    ((phy_ext_status & IEEE_ESR_1000T_FD_CAPS) ||
5081 		    (phy_ext_status & IEEE_ESR_1000X_FD_CAPS)) ? 1 : 0;
5082 		igb->param_1000hdx_cap =
5083 		    ((phy_ext_status & IEEE_ESR_1000T_HD_CAPS) ||
5084 		    (phy_ext_status & IEEE_ESR_1000X_HD_CAPS)) ? 1 : 0;
5085 		igb->param_100t4_cap =
5086 		    (phy_status & MII_SR_100T4_CAPS) ? 1 : 0;
5087 		igb->param_100fdx_cap = ((phy_status & MII_SR_100X_FD_CAPS) ||
5088 		    (phy_status & MII_SR_100T2_FD_CAPS)) ? 1 : 0;
5089 		igb->param_100hdx_cap = ((phy_status & MII_SR_100X_HD_CAPS) ||
5090 		    (phy_status & MII_SR_100T2_HD_CAPS)) ? 1 : 0;
5091 		igb->param_10fdx_cap =
5092 		    (phy_status & MII_SR_10T_FD_CAPS) ? 1 : 0;
5093 		igb->param_10hdx_cap =
5094 		    (phy_status & MII_SR_10T_HD_CAPS) ? 1 : 0;
5095 		igb->param_rem_fault =
5096 		    (phy_status & MII_SR_REMOTE_FAULT) ? 1 : 0;
5097 
5098 		igb->param_adv_autoneg_cap = hw->mac.autoneg;
5099 		igb->param_adv_pause_cap =
5100 		    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
5101 		igb->param_adv_asym_pause_cap =
5102 		    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
5103 		igb->param_adv_1000hdx_cap =
5104 		    (phy_1000t_ctrl & CR_1000T_HD_CAPS) ? 1 : 0;
5105 		igb->param_adv_100t4_cap =
5106 		    (phy_an_adv & NWAY_AR_100T4_CAPS) ? 1 : 0;
5107 		igb->param_adv_rem_fault =
5108 		    (phy_an_adv & NWAY_AR_REMOTE_FAULT) ? 1 : 0;
5109 		if (igb->param_adv_autoneg_cap == 1) {
5110 			igb->param_adv_1000fdx_cap =
5111 			    (phy_1000t_ctrl & CR_1000T_FD_CAPS) ? 1 : 0;
5112 			igb->param_adv_100fdx_cap =
5113 			    (phy_an_adv & NWAY_AR_100TX_FD_CAPS) ? 1 : 0;
5114 			igb->param_adv_100hdx_cap =
5115 			    (phy_an_adv & NWAY_AR_100TX_HD_CAPS) ? 1 : 0;
5116 			igb->param_adv_10fdx_cap =
5117 			    (phy_an_adv & NWAY_AR_10T_FD_CAPS) ? 1 : 0;
5118 			igb->param_adv_10hdx_cap =
5119 			    (phy_an_adv & NWAY_AR_10T_HD_CAPS) ? 1 : 0;
5120 		}
5121 
5122 		igb->param_lp_autoneg_cap =
5123 		    (phy_an_exp & NWAY_ER_LP_NWAY_CAPS) ? 1 : 0;
5124 		igb->param_lp_pause_cap =
5125 		    (phy_lp_able & NWAY_LPAR_PAUSE) ? 1 : 0;
5126 		igb->param_lp_asym_pause_cap =
5127 		    (phy_lp_able & NWAY_LPAR_ASM_DIR) ? 1 : 0;
5128 		igb->param_lp_1000fdx_cap =
5129 		    (phy_1000t_status & SR_1000T_LP_FD_CAPS) ? 1 : 0;
5130 		igb->param_lp_1000hdx_cap =
5131 		    (phy_1000t_status & SR_1000T_LP_HD_CAPS) ? 1 : 0;
5132 		igb->param_lp_100t4_cap =
5133 		    (phy_lp_able & NWAY_LPAR_100T4_CAPS) ? 1 : 0;
5134 		igb->param_lp_100fdx_cap =
5135 		    (phy_lp_able & NWAY_LPAR_100TX_FD_CAPS) ? 1 : 0;
5136 		igb->param_lp_100hdx_cap =
5137 		    (phy_lp_able & NWAY_LPAR_100TX_HD_CAPS) ? 1 : 0;
5138 		igb->param_lp_10fdx_cap =
5139 		    (phy_lp_able & NWAY_LPAR_10T_FD_CAPS) ? 1 : 0;
5140 		igb->param_lp_10hdx_cap =
5141 		    (phy_lp_able & NWAY_LPAR_10T_HD_CAPS) ? 1 : 0;
5142 		igb->param_lp_rem_fault =
5143 		    (phy_lp_able & NWAY_LPAR_REMOTE_FAULT) ? 1 : 0;
5144 	} else {
5145 		/*
5146 		 * 1Gig Fiber adapter only offers 1Gig Full Duplex.
5147 		 */
5148 		igb->param_autoneg_cap = 0;
5149 		igb->param_pause_cap = 1;
5150 		igb->param_asym_pause_cap = 1;
5151 		igb->param_1000fdx_cap = 1;
5152 		igb->param_1000hdx_cap = 0;
5153 		igb->param_100t4_cap = 0;
5154 		igb->param_100fdx_cap = 0;
5155 		igb->param_100hdx_cap = 0;
5156 		igb->param_10fdx_cap = 0;
5157 		igb->param_10hdx_cap = 0;
5158 
5159 		igb->param_adv_autoneg_cap = 0;
5160 		igb->param_adv_pause_cap = 1;
5161 		igb->param_adv_asym_pause_cap = 1;
5162 		igb->param_adv_1000fdx_cap = 1;
5163 		igb->param_adv_1000hdx_cap = 0;
5164 		igb->param_adv_100t4_cap = 0;
5165 		igb->param_adv_100fdx_cap = 0;
5166 		igb->param_adv_100hdx_cap = 0;
5167 		igb->param_adv_10fdx_cap = 0;
5168 		igb->param_adv_10hdx_cap = 0;
5169 
5170 		igb->param_lp_autoneg_cap = 0;
5171 		igb->param_lp_pause_cap = 0;
5172 		igb->param_lp_asym_pause_cap = 0;
5173 		igb->param_lp_1000fdx_cap = 0;
5174 		igb->param_lp_1000hdx_cap = 0;
5175 		igb->param_lp_100t4_cap = 0;
5176 		igb->param_lp_100fdx_cap = 0;
5177 		igb->param_lp_100hdx_cap = 0;
5178 		igb->param_lp_10fdx_cap = 0;
5179 		igb->param_lp_10hdx_cap = 0;
5180 		igb->param_lp_rem_fault = 0;
5181 	}
5182 }
5183 
5184 /*
5185  * synchronize the adv* and en* parameters.
5186  *
5187  * See comments in <sys/dld.h> for details of the *_en_*
5188  * parameters. The usage of ndd for setting adv parameters will
5189  * synchronize all the en parameters with the e1000g parameters,
5190  * implicitly disabling any settings made via dladm.
5191  */
5192 static void
5193 igb_param_sync(igb_t *igb)
5194 {
5195 	igb->param_en_1000fdx_cap = igb->param_adv_1000fdx_cap;
5196 	igb->param_en_1000hdx_cap = igb->param_adv_1000hdx_cap;
5197 	igb->param_en_100t4_cap = igb->param_adv_100t4_cap;
5198 	igb->param_en_100fdx_cap = igb->param_adv_100fdx_cap;
5199 	igb->param_en_100hdx_cap = igb->param_adv_100hdx_cap;
5200 	igb->param_en_10fdx_cap = igb->param_adv_10fdx_cap;
5201 	igb->param_en_10hdx_cap = igb->param_adv_10hdx_cap;
5202 }
5203 
5204 /*
5205  * igb_get_driver_control
5206  */
5207 static void
5208 igb_get_driver_control(struct e1000_hw *hw)
5209 {
5210 	uint32_t ctrl_ext;
5211 
5212 	/* Notify firmware that driver is in control of device */
5213 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
5214 	ctrl_ext |= E1000_CTRL_EXT_DRV_LOAD;
5215 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
5216 }
5217 
5218 /*
5219  * igb_release_driver_control
5220  */
5221 static void
5222 igb_release_driver_control(struct e1000_hw *hw)
5223 {
5224 	uint32_t ctrl_ext;
5225 
5226 	/* Notify firmware that driver is no longer in control of device */
5227 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
5228 	ctrl_ext &= ~E1000_CTRL_EXT_DRV_LOAD;
5229 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
5230 }
5231 
5232 /*
5233  * igb_atomic_reserve - Atomic decrease operation
5234  */
5235 int
5236 igb_atomic_reserve(uint32_t *count_p, uint32_t n)
5237 {
5238 	uint32_t oldval;
5239 	uint32_t newval;
5240 
5241 	/* ATOMICALLY */
5242 	do {
5243 		oldval = *count_p;
5244 		if (oldval < n)
5245 			return (-1);
5246 		newval = oldval - n;
5247 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
5248 
5249 	return (newval);
5250 }
5251 
5252 /*
5253  * FMA support
5254  */
5255 
5256 int
5257 igb_check_acc_handle(ddi_acc_handle_t handle)
5258 {
5259 	ddi_fm_error_t de;
5260 
5261 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
5262 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
5263 	return (de.fme_status);
5264 }
5265 
5266 int
5267 igb_check_dma_handle(ddi_dma_handle_t handle)
5268 {
5269 	ddi_fm_error_t de;
5270 
5271 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
5272 	return (de.fme_status);
5273 }
5274 
5275 /*
5276  * The IO fault service error handling callback function
5277  */
5278 /*ARGSUSED*/
5279 static int
5280 igb_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
5281 {
5282 	/*
5283 	 * as the driver can always deal with an error in any dma or
5284 	 * access handle, we can just return the fme_status value.
5285 	 */
5286 	pci_ereport_post(dip, err, NULL);
5287 	return (err->fme_status);
5288 }
5289 
5290 static void
5291 igb_fm_init(igb_t *igb)
5292 {
5293 	ddi_iblock_cookie_t iblk;
5294 	int fma_dma_flag;
5295 
5296 	/* Only register with IO Fault Services if we have some capability */
5297 	if (igb->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
5298 		igb_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
5299 	} else {
5300 		igb_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
5301 	}
5302 
5303 	if (igb->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
5304 		fma_dma_flag = 1;
5305 	} else {
5306 		fma_dma_flag = 0;
5307 	}
5308 
5309 	(void) igb_set_fma_flags(fma_dma_flag);
5310 
5311 	if (igb->fm_capabilities) {
5312 
5313 		/* Register capabilities with IO Fault Services */
5314 		ddi_fm_init(igb->dip, &igb->fm_capabilities, &iblk);
5315 
5316 		/*
5317 		 * Initialize pci ereport capabilities if ereport capable
5318 		 */
5319 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
5320 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5321 			pci_ereport_setup(igb->dip);
5322 
5323 		/*
5324 		 * Register error callback if error callback capable
5325 		 */
5326 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5327 			ddi_fm_handler_register(igb->dip,
5328 			    igb_fm_error_cb, (void*) igb);
5329 	}
5330 }
5331 
5332 static void
5333 igb_fm_fini(igb_t *igb)
5334 {
5335 	/* Only unregister FMA capabilities if we registered some */
5336 	if (igb->fm_capabilities) {
5337 
5338 		/*
5339 		 * Release any resources allocated by pci_ereport_setup()
5340 		 */
5341 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
5342 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5343 			pci_ereport_teardown(igb->dip);
5344 
5345 		/*
5346 		 * Un-register error callback if error callback capable
5347 		 */
5348 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5349 			ddi_fm_handler_unregister(igb->dip);
5350 
5351 		/* Unregister from IO Fault Services */
5352 		ddi_fm_fini(igb->dip);
5353 	}
5354 }
5355 
5356 void
5357 igb_fm_ereport(igb_t *igb, char *detail)
5358 {
5359 	uint64_t ena;
5360 	char buf[FM_MAX_CLASS];
5361 
5362 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
5363 	ena = fm_ena_generate(0, FM_ENA_FMT1);
5364 	if (DDI_FM_EREPORT_CAP(igb->fm_capabilities)) {
5365 		ddi_fm_ereport_post(igb->dip, buf, ena, DDI_NOSLEEP,
5366 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
5367 	}
5368 }
5369