xref: /illumos-gate/usr/src/uts/common/io/igb/igb_main.c (revision c211fc479225fa54805cf480633bf6689ca9a2db)
1 /*
2  * CDDL HEADER START
3  *
4  * Copyright(c) 2007-2009 Intel Corporation. All rights reserved.
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at:
10  *	http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When using or redistributing this file, you may do so under the
15  * License only. No other modification of this header is permitted.
16  *
17  * If applicable, add the following below this CDDL HEADER, with the
18  * fields enclosed by brackets "[]" replaced with your own identifying
19  * information: Portions Copyright [yyyy] [name of copyright owner]
20  *
21  * CDDL HEADER END
22  */
23 
24 /*
25  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
26  * Use is subject to license terms of the CDDL.
27  */
28 
29 #include "igb_sw.h"
30 
31 static char ident[] = "Intel 1Gb Ethernet";
32 static char igb_version[] = "igb 1.1.5";
33 
34 /*
35  * Local function protoypes
36  */
37 static int igb_register_mac(igb_t *);
38 static int igb_identify_hardware(igb_t *);
39 static int igb_regs_map(igb_t *);
40 static void igb_init_properties(igb_t *);
41 static int igb_init_driver_settings(igb_t *);
42 static void igb_init_locks(igb_t *);
43 static void igb_destroy_locks(igb_t *);
44 static int igb_init_mac_address(igb_t *);
45 static int igb_init(igb_t *);
46 static int igb_init_adapter(igb_t *);
47 static void igb_stop_adapter(igb_t *);
48 static int igb_reset(igb_t *);
49 static void igb_tx_clean(igb_t *);
50 static boolean_t igb_tx_drain(igb_t *);
51 static boolean_t igb_rx_drain(igb_t *);
52 static int igb_alloc_rings(igb_t *);
53 static void igb_free_rings(igb_t *);
54 static void igb_setup_rings(igb_t *);
55 static void igb_setup_rx(igb_t *);
56 static void igb_setup_tx(igb_t *);
57 static void igb_setup_rx_ring(igb_rx_ring_t *);
58 static void igb_setup_tx_ring(igb_tx_ring_t *);
59 static void igb_setup_rss(igb_t *);
60 static void igb_setup_mac_rss_classify(igb_t *);
61 static void igb_setup_mac_classify(igb_t *);
62 static void igb_init_unicst(igb_t *);
63 static void igb_setup_multicst(igb_t *);
64 static void igb_get_phy_state(igb_t *);
65 static void igb_get_conf(igb_t *);
66 static int igb_get_prop(igb_t *, char *, int, int, int);
67 static boolean_t igb_is_link_up(igb_t *);
68 static boolean_t igb_link_check(igb_t *);
69 static void igb_local_timer(void *);
70 static void igb_arm_watchdog_timer(igb_t *);
71 static void igb_start_watchdog_timer(igb_t *);
72 static void igb_restart_watchdog_timer(igb_t *);
73 static void igb_stop_watchdog_timer(igb_t *);
74 static void igb_disable_adapter_interrupts(igb_t *);
75 static void igb_enable_adapter_interrupts_82575(igb_t *);
76 static void igb_enable_adapter_interrupts_82576(igb_t *);
77 static boolean_t is_valid_mac_addr(uint8_t *);
78 static boolean_t igb_stall_check(igb_t *);
79 static boolean_t igb_set_loopback_mode(igb_t *, uint32_t);
80 static void igb_set_external_loopback(igb_t *);
81 static void igb_set_internal_mac_loopback(igb_t *);
82 static void igb_set_internal_phy_loopback(igb_t *);
83 static void igb_set_internal_serdes_loopback(igb_t *);
84 static boolean_t igb_find_mac_address(igb_t *);
85 static int igb_alloc_intrs(igb_t *);
86 static int igb_alloc_intr_handles(igb_t *, int);
87 static int igb_add_intr_handlers(igb_t *);
88 static void igb_rem_intr_handlers(igb_t *);
89 static void igb_rem_intrs(igb_t *);
90 static int igb_enable_intrs(igb_t *);
91 static int igb_disable_intrs(igb_t *);
92 static void igb_setup_msix_82575(igb_t *);
93 static void igb_setup_msix_82576(igb_t *);
94 static uint_t igb_intr_legacy(void *, void *);
95 static uint_t igb_intr_msi(void *, void *);
96 static uint_t igb_intr_rx(void *, void *);
97 static uint_t igb_intr_tx(void *, void *);
98 static uint_t igb_intr_tx_other(void *, void *);
99 static void igb_intr_rx_work(igb_rx_ring_t *);
100 static void igb_intr_tx_work(igb_tx_ring_t *);
101 static void igb_intr_link_work(igb_t *);
102 static void igb_get_driver_control(struct e1000_hw *);
103 static void igb_release_driver_control(struct e1000_hw *);
104 
105 static int igb_attach(dev_info_t *, ddi_attach_cmd_t);
106 static int igb_detach(dev_info_t *, ddi_detach_cmd_t);
107 static int igb_resume(dev_info_t *);
108 static int igb_suspend(dev_info_t *);
109 static int igb_quiesce(dev_info_t *);
110 static void igb_unconfigure(dev_info_t *, igb_t *);
111 static int igb_fm_error_cb(dev_info_t *, ddi_fm_error_t *,
112     const void *);
113 static void igb_fm_init(igb_t *);
114 static void igb_fm_fini(igb_t *);
115 
116 
117 static struct cb_ops igb_cb_ops = {
118 	nulldev,		/* cb_open */
119 	nulldev,		/* cb_close */
120 	nodev,			/* cb_strategy */
121 	nodev,			/* cb_print */
122 	nodev,			/* cb_dump */
123 	nodev,			/* cb_read */
124 	nodev,			/* cb_write */
125 	nodev,			/* cb_ioctl */
126 	nodev,			/* cb_devmap */
127 	nodev,			/* cb_mmap */
128 	nodev,			/* cb_segmap */
129 	nochpoll,		/* cb_chpoll */
130 	ddi_prop_op,		/* cb_prop_op */
131 	NULL,			/* cb_stream */
132 	D_MP | D_HOTPLUG,	/* cb_flag */
133 	CB_REV,			/* cb_rev */
134 	nodev,			/* cb_aread */
135 	nodev			/* cb_awrite */
136 };
137 
138 static struct dev_ops igb_dev_ops = {
139 	DEVO_REV,		/* devo_rev */
140 	0,			/* devo_refcnt */
141 	NULL,			/* devo_getinfo */
142 	nulldev,		/* devo_identify */
143 	nulldev,		/* devo_probe */
144 	igb_attach,		/* devo_attach */
145 	igb_detach,		/* devo_detach */
146 	nodev,			/* devo_reset */
147 	&igb_cb_ops,		/* devo_cb_ops */
148 	NULL,			/* devo_bus_ops */
149 	ddi_power,		/* devo_power */
150 	igb_quiesce,	/* devo_quiesce */
151 };
152 
153 static struct modldrv igb_modldrv = {
154 	&mod_driverops,		/* Type of module.  This one is a driver */
155 	ident,			/* Discription string */
156 	&igb_dev_ops,		/* driver ops */
157 };
158 
159 static struct modlinkage igb_modlinkage = {
160 	MODREV_1, &igb_modldrv, NULL
161 };
162 
163 /* Access attributes for register mapping */
164 ddi_device_acc_attr_t igb_regs_acc_attr = {
165 	DDI_DEVICE_ATTR_V0,
166 	DDI_STRUCTURE_LE_ACC,
167 	DDI_STRICTORDER_ACC,
168 	DDI_FLAGERR_ACC
169 };
170 
171 #define	IGB_M_CALLBACK_FLAGS	(MC_IOCTL | MC_GETCAPAB)
172 
173 static mac_callbacks_t igb_m_callbacks = {
174 	IGB_M_CALLBACK_FLAGS,
175 	igb_m_stat,
176 	igb_m_start,
177 	igb_m_stop,
178 	igb_m_promisc,
179 	igb_m_multicst,
180 	NULL,
181 	NULL,
182 	igb_m_ioctl,
183 	igb_m_getcapab
184 };
185 
186 /*
187  * Initialize capabilities of each supported adapter type
188  */
189 static adapter_info_t igb_82575_cap = {
190 	/* limits */
191 	4,		/* maximum number of rx queues */
192 	1,		/* minimum number of rx queues */
193 	4,		/* default number of rx queues */
194 	4,		/* maximum number of tx queues */
195 	1,		/* minimum number of tx queues */
196 	4,		/* default number of tx queues */
197 	65535,		/* maximum interrupt throttle rate */
198 	0,		/* minimum interrupt throttle rate */
199 	200,		/* default interrupt throttle rate */
200 
201 	/* function pointers */
202 	igb_enable_adapter_interrupts_82575,
203 	igb_setup_msix_82575,
204 
205 	/* capabilities */
206 	(IGB_FLAG_HAS_DCA |	/* capability flags */
207 	IGB_FLAG_VMDQ_POOL),
208 
209 	0xffc00000		/* mask for RXDCTL register */
210 };
211 
212 static adapter_info_t igb_82576_cap = {
213 	/* limits */
214 	16,		/* maximum number of rx queues */
215 	1,		/* minimum number of rx queues */
216 	4,		/* default number of rx queues */
217 	16,		/* maximum number of tx queues */
218 	1,		/* minimum number of tx queues */
219 	4,		/* default number of tx queues */
220 	65535,		/* maximum interrupt throttle rate */
221 	0,		/* minimum interrupt throttle rate */
222 	200,		/* default interrupt throttle rate */
223 
224 	/* function pointers */
225 	igb_enable_adapter_interrupts_82576,
226 	igb_setup_msix_82576,
227 
228 	/* capabilities */
229 	(IGB_FLAG_HAS_DCA |	/* capability flags */
230 	IGB_FLAG_VMDQ_POOL |
231 	IGB_FLAG_NEED_CTX_IDX),
232 
233 	0xffe00000		/* mask for RXDCTL register */
234 };
235 
236 /*
237  * Module Initialization Functions
238  */
239 
240 int
241 _init(void)
242 {
243 	int status;
244 
245 	mac_init_ops(&igb_dev_ops, MODULE_NAME);
246 
247 	status = mod_install(&igb_modlinkage);
248 
249 	if (status != DDI_SUCCESS) {
250 		mac_fini_ops(&igb_dev_ops);
251 	}
252 
253 	return (status);
254 }
255 
256 int
257 _fini(void)
258 {
259 	int status;
260 
261 	status = mod_remove(&igb_modlinkage);
262 
263 	if (status == DDI_SUCCESS) {
264 		mac_fini_ops(&igb_dev_ops);
265 	}
266 
267 	return (status);
268 
269 }
270 
271 int
272 _info(struct modinfo *modinfop)
273 {
274 	int status;
275 
276 	status = mod_info(&igb_modlinkage, modinfop);
277 
278 	return (status);
279 }
280 
281 /*
282  * igb_attach - driver attach
283  *
284  * This function is the device specific initialization entry
285  * point. This entry point is required and must be written.
286  * The DDI_ATTACH command must be provided in the attach entry
287  * point. When attach() is called with cmd set to DDI_ATTACH,
288  * all normal kernel services (such as kmem_alloc(9F)) are
289  * available for use by the driver.
290  *
291  * The attach() function will be called once for each instance
292  * of  the  device  on  the  system with cmd set to DDI_ATTACH.
293  * Until attach() succeeds, the only driver entry points which
294  * may be called are open(9E) and getinfo(9E).
295  */
296 static int
297 igb_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
298 {
299 	igb_t *igb;
300 	struct igb_osdep *osdep;
301 	struct e1000_hw *hw;
302 	int instance;
303 
304 	/*
305 	 * Check the command and perform corresponding operations
306 	 */
307 	switch (cmd) {
308 	default:
309 		return (DDI_FAILURE);
310 
311 	case DDI_RESUME:
312 		return (igb_resume(devinfo));
313 
314 	case DDI_ATTACH:
315 		break;
316 	}
317 
318 	/* Get the device instance */
319 	instance = ddi_get_instance(devinfo);
320 
321 	/* Allocate memory for the instance data structure */
322 	igb = kmem_zalloc(sizeof (igb_t), KM_SLEEP);
323 
324 	igb->dip = devinfo;
325 	igb->instance = instance;
326 
327 	hw = &igb->hw;
328 	osdep = &igb->osdep;
329 	hw->back = osdep;
330 	osdep->igb = igb;
331 
332 	/* Attach the instance pointer to the dev_info data structure */
333 	ddi_set_driver_private(devinfo, igb);
334 
335 
336 	/* Initialize for fma support */
337 	igb->fm_capabilities = igb_get_prop(igb, "fm-capable",
338 	    0, 0x0f,
339 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
340 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
341 	igb_fm_init(igb);
342 	igb->attach_progress |= ATTACH_PROGRESS_FMINIT;
343 
344 	/*
345 	 * Map PCI config space registers
346 	 */
347 	if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
348 		igb_error(igb, "Failed to map PCI configurations");
349 		goto attach_fail;
350 	}
351 	igb->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
352 
353 	/*
354 	 * Identify the chipset family
355 	 */
356 	if (igb_identify_hardware(igb) != IGB_SUCCESS) {
357 		igb_error(igb, "Failed to identify hardware");
358 		goto attach_fail;
359 	}
360 
361 	/*
362 	 * Map device registers
363 	 */
364 	if (igb_regs_map(igb) != IGB_SUCCESS) {
365 		igb_error(igb, "Failed to map device registers");
366 		goto attach_fail;
367 	}
368 	igb->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
369 
370 	/*
371 	 * Initialize driver parameters
372 	 */
373 	igb_init_properties(igb);
374 	igb->attach_progress |= ATTACH_PROGRESS_PROPS;
375 
376 	/*
377 	 * Allocate interrupts
378 	 */
379 	if (igb_alloc_intrs(igb) != IGB_SUCCESS) {
380 		igb_error(igb, "Failed to allocate interrupts");
381 		goto attach_fail;
382 	}
383 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
384 
385 	/*
386 	 * Allocate rx/tx rings based on the ring numbers.
387 	 * The actual numbers of rx/tx rings are decided by the number of
388 	 * allocated interrupt vectors, so we should allocate the rings after
389 	 * interrupts are allocated.
390 	 */
391 	if (igb_alloc_rings(igb) != IGB_SUCCESS) {
392 		igb_error(igb, "Failed to allocate rx/tx rings or groups");
393 		goto attach_fail;
394 	}
395 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_RINGS;
396 
397 	/*
398 	 * Add interrupt handlers
399 	 */
400 	if (igb_add_intr_handlers(igb) != IGB_SUCCESS) {
401 		igb_error(igb, "Failed to add interrupt handlers");
402 		goto attach_fail;
403 	}
404 	igb->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
405 
406 	/*
407 	 * Initialize driver parameters
408 	 */
409 	if (igb_init_driver_settings(igb) != IGB_SUCCESS) {
410 		igb_error(igb, "Failed to initialize driver settings");
411 		goto attach_fail;
412 	}
413 
414 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK) {
415 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
416 		goto attach_fail;
417 	}
418 
419 	/*
420 	 * Initialize mutexes for this device.
421 	 * Do this before enabling the interrupt handler and
422 	 * register the softint to avoid the condition where
423 	 * interrupt handler can try using uninitialized mutex
424 	 */
425 	igb_init_locks(igb);
426 	igb->attach_progress |= ATTACH_PROGRESS_LOCKS;
427 
428 	/*
429 	 * Allocate DMA resources
430 	 */
431 	if (igb_alloc_dma(igb) != IGB_SUCCESS) {
432 		igb_error(igb, "Failed to allocate DMA resources");
433 		goto attach_fail;
434 	}
435 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_DMA;
436 
437 	/*
438 	 * Initialize the adapter and setup the rx/tx rings
439 	 */
440 	if (igb_init(igb) != IGB_SUCCESS) {
441 		igb_error(igb, "Failed to initialize adapter");
442 		goto attach_fail;
443 	}
444 	igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
445 
446 	/*
447 	 * Initialize statistics
448 	 */
449 	if (igb_init_stats(igb) != IGB_SUCCESS) {
450 		igb_error(igb, "Failed to initialize statistics");
451 		goto attach_fail;
452 	}
453 	igb->attach_progress |= ATTACH_PROGRESS_STATS;
454 
455 	/*
456 	 * Initialize NDD parameters
457 	 */
458 	if (igb_nd_init(igb) != IGB_SUCCESS) {
459 		igb_error(igb, "Failed to initialize ndd");
460 		goto attach_fail;
461 	}
462 	igb->attach_progress |= ATTACH_PROGRESS_NDD;
463 
464 	/*
465 	 * Register the driver to the MAC
466 	 */
467 	if (igb_register_mac(igb) != IGB_SUCCESS) {
468 		igb_error(igb, "Failed to register MAC");
469 		goto attach_fail;
470 	}
471 	igb->attach_progress |= ATTACH_PROGRESS_MAC;
472 
473 	/*
474 	 * Now that mutex locks are initialized, and the chip is also
475 	 * initialized, enable interrupts.
476 	 */
477 	if (igb_enable_intrs(igb) != IGB_SUCCESS) {
478 		igb_error(igb, "Failed to enable DDI interrupts");
479 		goto attach_fail;
480 	}
481 	igb->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
482 
483 	igb_log(igb, "%s", igb_version);
484 	igb->igb_state |= IGB_INITIALIZED;
485 
486 	return (DDI_SUCCESS);
487 
488 attach_fail:
489 	igb_unconfigure(devinfo, igb);
490 	return (DDI_FAILURE);
491 }
492 
493 /*
494  * igb_detach - driver detach
495  *
496  * The detach() function is the complement of the attach routine.
497  * If cmd is set to DDI_DETACH, detach() is used to remove  the
498  * state  associated  with  a  given  instance of a device node
499  * prior to the removal of that instance from the system.
500  *
501  * The detach() function will be called once for each  instance
502  * of the device for which there has been a successful attach()
503  * once there are no longer  any  opens  on  the  device.
504  *
505  * Interrupts routine are disabled, All memory allocated by this
506  * driver are freed.
507  */
508 static int
509 igb_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
510 {
511 	igb_t *igb;
512 
513 	/*
514 	 * Check detach command
515 	 */
516 	switch (cmd) {
517 	default:
518 		return (DDI_FAILURE);
519 
520 	case DDI_SUSPEND:
521 		return (igb_suspend(devinfo));
522 
523 	case DDI_DETACH:
524 		break;
525 	}
526 
527 
528 	/*
529 	 * Get the pointer to the driver private data structure
530 	 */
531 	igb = (igb_t *)ddi_get_driver_private(devinfo);
532 	if (igb == NULL)
533 		return (DDI_FAILURE);
534 
535 	/*
536 	 * Unregister MAC. If failed, we have to fail the detach
537 	 */
538 	if (mac_unregister(igb->mac_hdl) != 0) {
539 		igb_error(igb, "Failed to unregister MAC");
540 		return (DDI_FAILURE);
541 	}
542 	igb->attach_progress &= ~ATTACH_PROGRESS_MAC;
543 
544 	/*
545 	 * If the device is still running, it needs to be stopped first.
546 	 * This check is necessary because under some specific circumstances,
547 	 * the detach routine can be called without stopping the interface
548 	 * first.
549 	 */
550 	mutex_enter(&igb->gen_lock);
551 	if (igb->igb_state & IGB_STARTED) {
552 		igb->igb_state &= ~IGB_STARTED;
553 		igb_stop(igb);
554 		mutex_exit(&igb->gen_lock);
555 		/* Disable and stop the watchdog timer */
556 		igb_disable_watchdog_timer(igb);
557 	} else
558 		mutex_exit(&igb->gen_lock);
559 
560 	/*
561 	 * Check if there are still rx buffers held by the upper layer.
562 	 * If so, fail the detach.
563 	 */
564 	if (!igb_rx_drain(igb))
565 		return (DDI_FAILURE);
566 
567 	/*
568 	 * Do the remaining unconfigure routines
569 	 */
570 	igb_unconfigure(devinfo, igb);
571 
572 	return (DDI_SUCCESS);
573 }
574 
575 /*
576  * quiesce(9E) entry point.
577  *
578  * This function is called when the system is single-threaded at high
579  * PIL with preemption disabled. Therefore, this function must not be
580  * blocked.
581  *
582  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
583  * DDI_FAILURE indicates an error condition and should almost never happen.
584  */
585 static int
586 igb_quiesce(dev_info_t *devinfo)
587 {
588 	igb_t *igb;
589 	struct e1000_hw *hw;
590 
591 	igb = (igb_t *)ddi_get_driver_private(devinfo);
592 
593 	if (igb == NULL)
594 		return (DDI_FAILURE);
595 
596 	hw = &igb->hw;
597 
598 	/*
599 	 * Disable the adapter interrupts
600 	 */
601 	igb_disable_adapter_interrupts(igb);
602 
603 	/* Tell firmware driver is no longer in control */
604 	igb_release_driver_control(hw);
605 
606 	/*
607 	 * Reset the chipset
608 	 */
609 	(void) e1000_reset_hw(hw);
610 
611 	/*
612 	 * Reset PHY if possible
613 	 */
614 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
615 		(void) e1000_phy_hw_reset(hw);
616 
617 	return (DDI_SUCCESS);
618 }
619 
620 /*
621  * igb_unconfigure - release all resources held by this instance
622  */
623 static void
624 igb_unconfigure(dev_info_t *devinfo, igb_t *igb)
625 {
626 	/*
627 	 * Disable interrupt
628 	 */
629 	if (igb->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
630 		(void) igb_disable_intrs(igb);
631 	}
632 
633 	/*
634 	 * Unregister MAC
635 	 */
636 	if (igb->attach_progress & ATTACH_PROGRESS_MAC) {
637 		(void) mac_unregister(igb->mac_hdl);
638 	}
639 
640 	/*
641 	 * Free ndd parameters
642 	 */
643 	if (igb->attach_progress & ATTACH_PROGRESS_NDD) {
644 		igb_nd_cleanup(igb);
645 	}
646 
647 	/*
648 	 * Free statistics
649 	 */
650 	if (igb->attach_progress & ATTACH_PROGRESS_STATS) {
651 		kstat_delete((kstat_t *)igb->igb_ks);
652 	}
653 
654 	/*
655 	 * Remove interrupt handlers
656 	 */
657 	if (igb->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
658 		igb_rem_intr_handlers(igb);
659 	}
660 
661 	/*
662 	 * Remove interrupts
663 	 */
664 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_INTR) {
665 		igb_rem_intrs(igb);
666 	}
667 
668 	/*
669 	 * Remove driver properties
670 	 */
671 	if (igb->attach_progress & ATTACH_PROGRESS_PROPS) {
672 		(void) ddi_prop_remove_all(devinfo);
673 	}
674 
675 	/*
676 	 * Release the DMA resources of rx/tx rings
677 	 */
678 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_DMA) {
679 		igb_free_dma(igb);
680 	}
681 
682 	/*
683 	 * Stop the adapter
684 	 */
685 	if (igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) {
686 		mutex_enter(&igb->gen_lock);
687 		igb_stop_adapter(igb);
688 		mutex_exit(&igb->gen_lock);
689 		if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
690 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_UNAFFECTED);
691 	}
692 
693 	/*
694 	 * Free register handle
695 	 */
696 	if (igb->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
697 		if (igb->osdep.reg_handle != NULL)
698 			ddi_regs_map_free(&igb->osdep.reg_handle);
699 	}
700 
701 	/*
702 	 * Free PCI config handle
703 	 */
704 	if (igb->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
705 		if (igb->osdep.cfg_handle != NULL)
706 			pci_config_teardown(&igb->osdep.cfg_handle);
707 	}
708 
709 	/*
710 	 * Free locks
711 	 */
712 	if (igb->attach_progress & ATTACH_PROGRESS_LOCKS) {
713 		igb_destroy_locks(igb);
714 	}
715 
716 	/*
717 	 * Free the rx/tx rings
718 	 */
719 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
720 		igb_free_rings(igb);
721 	}
722 
723 	/*
724 	 * Remove FMA
725 	 */
726 	if (igb->attach_progress & ATTACH_PROGRESS_FMINIT) {
727 		igb_fm_fini(igb);
728 	}
729 
730 	/*
731 	 * Free the driver data structure
732 	 */
733 	kmem_free(igb, sizeof (igb_t));
734 
735 	ddi_set_driver_private(devinfo, NULL);
736 }
737 
738 /*
739  * igb_register_mac - Register the driver and its function pointers with
740  * the GLD interface
741  */
742 static int
743 igb_register_mac(igb_t *igb)
744 {
745 	struct e1000_hw *hw = &igb->hw;
746 	mac_register_t *mac;
747 	int status;
748 
749 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
750 		return (IGB_FAILURE);
751 
752 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
753 	mac->m_driver = igb;
754 	mac->m_dip = igb->dip;
755 	mac->m_src_addr = hw->mac.addr;
756 	mac->m_callbacks = &igb_m_callbacks;
757 	mac->m_min_sdu = 0;
758 	mac->m_max_sdu = igb->max_frame_size -
759 	    sizeof (struct ether_vlan_header) - ETHERFCSL;
760 	mac->m_margin = VLAN_TAGSZ;
761 	mac->m_v12n = MAC_VIRT_LEVEL1;
762 
763 	status = mac_register(mac, &igb->mac_hdl);
764 
765 	mac_free(mac);
766 
767 	return ((status == 0) ? IGB_SUCCESS : IGB_FAILURE);
768 }
769 
770 /*
771  * igb_identify_hardware - Identify the type of the chipset
772  */
773 static int
774 igb_identify_hardware(igb_t *igb)
775 {
776 	struct e1000_hw *hw = &igb->hw;
777 	struct igb_osdep *osdep = &igb->osdep;
778 
779 	/*
780 	 * Get the device id
781 	 */
782 	hw->vendor_id =
783 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
784 	hw->device_id =
785 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
786 	hw->revision_id =
787 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
788 	hw->subsystem_device_id =
789 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
790 	hw->subsystem_vendor_id =
791 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
792 
793 	/*
794 	 * Set the mac type of the adapter based on the device id
795 	 */
796 	if (e1000_set_mac_type(hw) != E1000_SUCCESS) {
797 		return (IGB_FAILURE);
798 	}
799 
800 	/*
801 	 * Install adapter capabilities based on mac type
802 	 */
803 	switch (hw->mac.type) {
804 	case e1000_82575:
805 		igb->capab = &igb_82575_cap;
806 		break;
807 	case e1000_82576:
808 		igb->capab = &igb_82576_cap;
809 		break;
810 	default:
811 		return (IGB_FAILURE);
812 	}
813 
814 	return (IGB_SUCCESS);
815 }
816 
817 /*
818  * igb_regs_map - Map the device registers
819  */
820 static int
821 igb_regs_map(igb_t *igb)
822 {
823 	dev_info_t *devinfo = igb->dip;
824 	struct e1000_hw *hw = &igb->hw;
825 	struct igb_osdep *osdep = &igb->osdep;
826 	off_t mem_size;
827 
828 	/*
829 	 * First get the size of device registers to be mapped.
830 	 */
831 	if (ddi_dev_regsize(devinfo, IGB_ADAPTER_REGSET, &mem_size) !=
832 	    DDI_SUCCESS) {
833 		return (IGB_FAILURE);
834 	}
835 
836 	/*
837 	 * Call ddi_regs_map_setup() to map registers
838 	 */
839 	if ((ddi_regs_map_setup(devinfo, IGB_ADAPTER_REGSET,
840 	    (caddr_t *)&hw->hw_addr, 0,
841 	    mem_size, &igb_regs_acc_attr,
842 	    &osdep->reg_handle)) != DDI_SUCCESS) {
843 		return (IGB_FAILURE);
844 	}
845 
846 	return (IGB_SUCCESS);
847 }
848 
849 /*
850  * igb_init_properties - Initialize driver properties
851  */
852 static void
853 igb_init_properties(igb_t *igb)
854 {
855 	/*
856 	 * Get conf file properties, including link settings
857 	 * jumbo frames, ring number, descriptor number, etc.
858 	 */
859 	igb_get_conf(igb);
860 }
861 
862 /*
863  * igb_init_driver_settings - Initialize driver settings
864  *
865  * The settings include hardware function pointers, bus information,
866  * rx/tx rings settings, link state, and any other parameters that
867  * need to be setup during driver initialization.
868  */
869 static int
870 igb_init_driver_settings(igb_t *igb)
871 {
872 	struct e1000_hw *hw = &igb->hw;
873 	igb_rx_ring_t *rx_ring;
874 	igb_tx_ring_t *tx_ring;
875 	uint32_t rx_size;
876 	uint32_t tx_size;
877 	int i;
878 
879 	/*
880 	 * Initialize chipset specific hardware function pointers
881 	 */
882 	if (e1000_setup_init_funcs(hw, B_TRUE) != E1000_SUCCESS) {
883 		return (IGB_FAILURE);
884 	}
885 
886 	/*
887 	 * Get bus information
888 	 */
889 	if (e1000_get_bus_info(hw) != E1000_SUCCESS) {
890 		return (IGB_FAILURE);
891 	}
892 
893 	/*
894 	 * Set rx buffer size
895 	 * The IP header alignment room is counted in the calculation.
896 	 * The rx buffer size is in unit of 1K that is required by the
897 	 * chipset hardware.
898 	 */
899 	rx_size = igb->max_frame_size + IPHDR_ALIGN_ROOM;
900 	igb->rx_buf_size = ((rx_size >> 10) +
901 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
902 
903 	/*
904 	 * Set tx buffer size
905 	 */
906 	tx_size = igb->max_frame_size;
907 	igb->tx_buf_size = ((tx_size >> 10) +
908 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
909 
910 	/*
911 	 * Initialize rx/tx rings parameters
912 	 */
913 	for (i = 0; i < igb->num_rx_rings; i++) {
914 		rx_ring = &igb->rx_rings[i];
915 		rx_ring->index = i;
916 		rx_ring->igb = igb;
917 
918 		rx_ring->ring_size = igb->rx_ring_size;
919 		rx_ring->free_list_size = igb->rx_ring_size;
920 		rx_ring->copy_thresh = igb->rx_copy_thresh;
921 		rx_ring->limit_per_intr = igb->rx_limit_per_intr;
922 	}
923 
924 	for (i = 0; i < igb->num_tx_rings; i++) {
925 		tx_ring = &igb->tx_rings[i];
926 		tx_ring->index = i;
927 		tx_ring->igb = igb;
928 		if (igb->tx_head_wb_enable)
929 			tx_ring->tx_recycle = igb_tx_recycle_head_wb;
930 		else
931 			tx_ring->tx_recycle = igb_tx_recycle_legacy;
932 
933 		tx_ring->ring_size = igb->tx_ring_size;
934 		tx_ring->free_list_size = igb->tx_ring_size +
935 		    (igb->tx_ring_size >> 1);
936 		tx_ring->copy_thresh = igb->tx_copy_thresh;
937 		tx_ring->recycle_thresh = igb->tx_recycle_thresh;
938 		tx_ring->overload_thresh = igb->tx_overload_thresh;
939 		tx_ring->resched_thresh = igb->tx_resched_thresh;
940 	}
941 
942 	/*
943 	 * Initialize values of interrupt throttling rates
944 	 */
945 	for (i = 1; i < MAX_NUM_EITR; i++)
946 		igb->intr_throttling[i] = igb->intr_throttling[0];
947 
948 	/*
949 	 * The initial link state should be "unknown"
950 	 */
951 	igb->link_state = LINK_STATE_UNKNOWN;
952 
953 	return (IGB_SUCCESS);
954 }
955 
956 /*
957  * igb_init_locks - Initialize locks
958  */
959 static void
960 igb_init_locks(igb_t *igb)
961 {
962 	igb_rx_ring_t *rx_ring;
963 	igb_tx_ring_t *tx_ring;
964 	int i;
965 
966 	for (i = 0; i < igb->num_rx_rings; i++) {
967 		rx_ring = &igb->rx_rings[i];
968 		mutex_init(&rx_ring->rx_lock, NULL,
969 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
970 		mutex_init(&rx_ring->recycle_lock, NULL,
971 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
972 	}
973 
974 	for (i = 0; i < igb->num_tx_rings; i++) {
975 		tx_ring = &igb->tx_rings[i];
976 		mutex_init(&tx_ring->tx_lock, NULL,
977 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
978 		mutex_init(&tx_ring->recycle_lock, NULL,
979 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
980 		mutex_init(&tx_ring->tcb_head_lock, NULL,
981 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
982 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
983 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
984 	}
985 
986 	mutex_init(&igb->gen_lock, NULL,
987 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
988 
989 	mutex_init(&igb->watchdog_lock, NULL,
990 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
991 }
992 
993 /*
994  * igb_destroy_locks - Destroy locks
995  */
996 static void
997 igb_destroy_locks(igb_t *igb)
998 {
999 	igb_rx_ring_t *rx_ring;
1000 	igb_tx_ring_t *tx_ring;
1001 	int i;
1002 
1003 	for (i = 0; i < igb->num_rx_rings; i++) {
1004 		rx_ring = &igb->rx_rings[i];
1005 		mutex_destroy(&rx_ring->rx_lock);
1006 		mutex_destroy(&rx_ring->recycle_lock);
1007 	}
1008 
1009 	for (i = 0; i < igb->num_tx_rings; i++) {
1010 		tx_ring = &igb->tx_rings[i];
1011 		mutex_destroy(&tx_ring->tx_lock);
1012 		mutex_destroy(&tx_ring->recycle_lock);
1013 		mutex_destroy(&tx_ring->tcb_head_lock);
1014 		mutex_destroy(&tx_ring->tcb_tail_lock);
1015 	}
1016 
1017 	mutex_destroy(&igb->gen_lock);
1018 	mutex_destroy(&igb->watchdog_lock);
1019 }
1020 
1021 static int
1022 igb_resume(dev_info_t *devinfo)
1023 {
1024 	igb_t *igb;
1025 
1026 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1027 	if (igb == NULL)
1028 		return (DDI_FAILURE);
1029 
1030 	mutex_enter(&igb->gen_lock);
1031 
1032 	if (igb->igb_state & IGB_STARTED) {
1033 		if (igb_start(igb) != IGB_SUCCESS) {
1034 			mutex_exit(&igb->gen_lock);
1035 			return (DDI_FAILURE);
1036 		}
1037 
1038 		/*
1039 		 * Enable and start the watchdog timer
1040 		 */
1041 		igb_enable_watchdog_timer(igb);
1042 	}
1043 
1044 	igb->igb_state &= ~IGB_SUSPENDED;
1045 
1046 	mutex_exit(&igb->gen_lock);
1047 
1048 	return (DDI_SUCCESS);
1049 }
1050 
1051 static int
1052 igb_suspend(dev_info_t *devinfo)
1053 {
1054 	igb_t *igb;
1055 
1056 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1057 	if (igb == NULL)
1058 		return (DDI_FAILURE);
1059 
1060 	mutex_enter(&igb->gen_lock);
1061 
1062 	igb->igb_state |= IGB_SUSPENDED;
1063 
1064 	if (!(igb->igb_state & IGB_STARTED)) {
1065 		mutex_exit(&igb->gen_lock);
1066 		return (DDI_SUCCESS);
1067 	}
1068 
1069 	igb_stop(igb);
1070 
1071 	mutex_exit(&igb->gen_lock);
1072 
1073 	/*
1074 	 * Disable and stop the watchdog timer
1075 	 */
1076 	igb_disable_watchdog_timer(igb);
1077 
1078 	return (DDI_SUCCESS);
1079 }
1080 
1081 static int
1082 igb_init(igb_t *igb)
1083 {
1084 	int i;
1085 
1086 	mutex_enter(&igb->gen_lock);
1087 
1088 	/*
1089 	 * Initilize the adapter
1090 	 */
1091 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1092 		mutex_exit(&igb->gen_lock);
1093 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1094 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1095 		return (IGB_FAILURE);
1096 	}
1097 
1098 	/*
1099 	 * Setup the rx/tx rings
1100 	 */
1101 	for (i = 0; i < igb->num_rx_rings; i++)
1102 		mutex_enter(&igb->rx_rings[i].rx_lock);
1103 	for (i = 0; i < igb->num_tx_rings; i++)
1104 		mutex_enter(&igb->tx_rings[i].tx_lock);
1105 
1106 	igb_setup_rings(igb);
1107 
1108 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1109 		mutex_exit(&igb->tx_rings[i].tx_lock);
1110 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1111 		mutex_exit(&igb->rx_rings[i].rx_lock);
1112 
1113 	mutex_exit(&igb->gen_lock);
1114 
1115 	return (IGB_SUCCESS);
1116 }
1117 
1118 /*
1119  * igb_init_mac_address - Initialize the default MAC address
1120  *
1121  * On success, the MAC address is entered in the igb->hw.mac.addr
1122  * and hw->mac.perm_addr fields and the adapter's RAR(0) receive
1123  * address register.
1124  *
1125  * Important side effects:
1126  * 1. adapter is reset - this is required to put it in a known state.
1127  * 2. all of non-volatile memory (NVM) is read & checksummed - NVM is where
1128  * MAC address and all default settings are stored, so a valid checksum
1129  * is required.
1130  */
1131 static int
1132 igb_init_mac_address(igb_t *igb)
1133 {
1134 	struct e1000_hw *hw = &igb->hw;
1135 
1136 	ASSERT(mutex_owned(&igb->gen_lock));
1137 
1138 	/*
1139 	 * Reset chipset to put the hardware in a known state
1140 	 * before we try to get MAC address from NVM.
1141 	 */
1142 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1143 		igb_error(igb, "Adapter reset failed.");
1144 		goto init_mac_fail;
1145 	}
1146 
1147 	/*
1148 	 * NVM validation
1149 	 */
1150 	if (e1000_validate_nvm_checksum(hw) < 0) {
1151 		/*
1152 		 * Some PCI-E parts fail the first check due to
1153 		 * the link being in sleep state.  Call it again,
1154 		 * if it fails a second time its a real issue.
1155 		 */
1156 		if (e1000_validate_nvm_checksum(hw) < 0) {
1157 			igb_error(igb,
1158 			    "Invalid NVM checksum. Please contact "
1159 			    "the vendor to update the NVM.");
1160 			goto init_mac_fail;
1161 		}
1162 	}
1163 
1164 	/*
1165 	 * Get the mac address
1166 	 * This function should handle SPARC case correctly.
1167 	 */
1168 	if (!igb_find_mac_address(igb)) {
1169 		igb_error(igb, "Failed to get the mac address");
1170 		goto init_mac_fail;
1171 	}
1172 
1173 	/* Validate mac address */
1174 	if (!is_valid_mac_addr(hw->mac.addr)) {
1175 		igb_error(igb, "Invalid mac address");
1176 		goto init_mac_fail;
1177 	}
1178 
1179 	return (IGB_SUCCESS);
1180 
1181 init_mac_fail:
1182 	return (IGB_FAILURE);
1183 }
1184 
1185 /*
1186  * igb_init_adapter - Initialize the adapter
1187  */
1188 static int
1189 igb_init_adapter(igb_t *igb)
1190 {
1191 	struct e1000_hw *hw = &igb->hw;
1192 	uint32_t pba;
1193 	uint32_t high_water;
1194 	int i;
1195 
1196 	ASSERT(mutex_owned(&igb->gen_lock));
1197 
1198 	/*
1199 	 * In order to obtain the default MAC address, this will reset the
1200 	 * adapter and validate the NVM that the address and many other
1201 	 * default settings come from.
1202 	 */
1203 	if (igb_init_mac_address(igb) != IGB_SUCCESS) {
1204 		igb_error(igb, "Failed to initialize MAC address");
1205 		goto init_adapter_fail;
1206 	}
1207 
1208 	/*
1209 	 * Setup flow control
1210 	 *
1211 	 * These parameters set thresholds for the adapter's generation(Tx)
1212 	 * and response(Rx) to Ethernet PAUSE frames.  These are just threshold
1213 	 * settings.  Flow control is enabled or disabled in the configuration
1214 	 * file.
1215 	 * High-water mark is set down from the top of the rx fifo (not
1216 	 * sensitive to max_frame_size) and low-water is set just below
1217 	 * high-water mark.
1218 	 * The high water mark must be low enough to fit one full frame above
1219 	 * it in the rx FIFO.  Should be the lower of:
1220 	 * 90% of the Rx FIFO size, or the full Rx FIFO size minus one full
1221 	 * frame.
1222 	 */
1223 	/*
1224 	 * The default setting of PBA is correct for 82575 and other supported
1225 	 * adapters do not have the E1000_PBA register, so PBA value is only
1226 	 * used for calculation here and is never written to the adapter.
1227 	 */
1228 	if (hw->mac.type == e1000_82575) {
1229 		pba = E1000_PBA_34K;
1230 	} else {
1231 		pba = E1000_PBA_64K;
1232 	}
1233 
1234 	high_water = min(((pba << 10) * 9 / 10),
1235 	    ((pba << 10) - igb->max_frame_size));
1236 
1237 	if (hw->mac.type == e1000_82575) {
1238 		/* 8-byte granularity */
1239 		hw->fc.high_water = high_water & 0xFFF8;
1240 		hw->fc.low_water = hw->fc.high_water - 8;
1241 	} else {
1242 		/* 16-byte granularity */
1243 		hw->fc.high_water = high_water & 0xFFF0;
1244 		hw->fc.low_water = hw->fc.high_water - 16;
1245 	}
1246 
1247 	hw->fc.pause_time = E1000_FC_PAUSE_TIME;
1248 	hw->fc.send_xon = B_TRUE;
1249 
1250 	e1000_validate_mdi_setting(hw);
1251 
1252 	/*
1253 	 * Reset the chipset hardware the second time to put PBA settings
1254 	 * into effect.
1255 	 */
1256 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1257 		igb_error(igb, "Second reset failed");
1258 		goto init_adapter_fail;
1259 	}
1260 
1261 	/*
1262 	 * Don't wait for auto-negotiation to complete
1263 	 */
1264 	hw->phy.autoneg_wait_to_complete = B_FALSE;
1265 
1266 	/*
1267 	 * Copper options
1268 	 */
1269 	if (hw->phy.media_type == e1000_media_type_copper) {
1270 		hw->phy.mdix = 0;	/* AUTO_ALL_MODES */
1271 		hw->phy.disable_polarity_correction = B_FALSE;
1272 		hw->phy.ms_type = e1000_ms_hw_default; /* E1000_MASTER_SLAVE */
1273 	}
1274 
1275 	/*
1276 	 * Initialize link settings
1277 	 */
1278 	(void) igb_setup_link(igb, B_FALSE);
1279 
1280 	/*
1281 	 * Configure/Initialize hardware
1282 	 */
1283 	if (e1000_init_hw(hw) != E1000_SUCCESS) {
1284 		igb_error(igb, "Failed to initialize hardware");
1285 		goto init_adapter_fail;
1286 	}
1287 
1288 	/*
1289 	 * Disable wakeup control by default
1290 	 */
1291 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1292 
1293 	/*
1294 	 * Record phy info in hw struct
1295 	 */
1296 	(void) e1000_get_phy_info(hw);
1297 
1298 	/*
1299 	 * Make sure driver has control
1300 	 */
1301 	igb_get_driver_control(hw);
1302 
1303 	/*
1304 	 * Restore LED settings to the default from EEPROM
1305 	 * to meet the standard for Sun platforms.
1306 	 */
1307 	(void) e1000_cleanup_led(hw);
1308 
1309 	/*
1310 	 * Setup MSI-X interrupts
1311 	 */
1312 	if (igb->intr_type == DDI_INTR_TYPE_MSIX)
1313 		igb->capab->setup_msix(igb);
1314 
1315 	/*
1316 	 * Initialize unicast addresses.
1317 	 */
1318 	igb_init_unicst(igb);
1319 
1320 	/*
1321 	 * Setup and initialize the mctable structures.
1322 	 */
1323 	igb_setup_multicst(igb);
1324 
1325 	/*
1326 	 * Set interrupt throttling rate
1327 	 */
1328 	for (i = 0; i < igb->intr_cnt; i++)
1329 		E1000_WRITE_REG(hw, E1000_EITR(i), igb->intr_throttling[i]);
1330 
1331 	/*
1332 	 * Save the state of the phy
1333 	 */
1334 	igb_get_phy_state(igb);
1335 
1336 	return (IGB_SUCCESS);
1337 
1338 init_adapter_fail:
1339 	/*
1340 	 * Reset PHY if possible
1341 	 */
1342 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
1343 		(void) e1000_phy_hw_reset(hw);
1344 
1345 	return (IGB_FAILURE);
1346 }
1347 
1348 /*
1349  * igb_stop_adapter - Stop the adapter
1350  */
1351 static void
1352 igb_stop_adapter(igb_t *igb)
1353 {
1354 	struct e1000_hw *hw = &igb->hw;
1355 
1356 	ASSERT(mutex_owned(&igb->gen_lock));
1357 
1358 	/* Tell firmware driver is no longer in control */
1359 	igb_release_driver_control(hw);
1360 
1361 	/*
1362 	 * Reset the chipset
1363 	 */
1364 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1365 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1366 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1367 	}
1368 
1369 	/*
1370 	 * e1000_phy_hw_reset is not needed here, MAC reset above is sufficient
1371 	 */
1372 }
1373 
1374 /*
1375  * igb_reset - Reset the chipset and restart the driver.
1376  *
1377  * It involves stopping and re-starting the chipset,
1378  * and re-configuring the rx/tx rings.
1379  */
1380 static int
1381 igb_reset(igb_t *igb)
1382 {
1383 	int i;
1384 
1385 	mutex_enter(&igb->gen_lock);
1386 
1387 	ASSERT(igb->igb_state & IGB_STARTED);
1388 
1389 	/*
1390 	 * Disable the adapter interrupts to stop any rx/tx activities
1391 	 * before draining pending data and resetting hardware.
1392 	 */
1393 	igb_disable_adapter_interrupts(igb);
1394 
1395 	/*
1396 	 * Drain the pending transmit packets
1397 	 */
1398 	(void) igb_tx_drain(igb);
1399 
1400 	for (i = 0; i < igb->num_rx_rings; i++)
1401 		mutex_enter(&igb->rx_rings[i].rx_lock);
1402 	for (i = 0; i < igb->num_tx_rings; i++)
1403 		mutex_enter(&igb->tx_rings[i].tx_lock);
1404 
1405 	/*
1406 	 * Stop the adapter
1407 	 */
1408 	igb_stop_adapter(igb);
1409 
1410 	/*
1411 	 * Clean the pending tx data/resources
1412 	 */
1413 	igb_tx_clean(igb);
1414 
1415 	/*
1416 	 * Start the adapter
1417 	 */
1418 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1419 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1420 		goto reset_failure;
1421 	}
1422 
1423 	/*
1424 	 * Setup the rx/tx rings
1425 	 */
1426 	igb_setup_rings(igb);
1427 
1428 	/*
1429 	 * Enable adapter interrupts
1430 	 * The interrupts must be enabled after the driver state is START
1431 	 */
1432 	igb->capab->enable_intr(igb);
1433 
1434 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1435 		goto reset_failure;
1436 
1437 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1438 		goto reset_failure;
1439 
1440 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1441 		mutex_exit(&igb->tx_rings[i].tx_lock);
1442 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1443 		mutex_exit(&igb->rx_rings[i].rx_lock);
1444 
1445 	mutex_exit(&igb->gen_lock);
1446 
1447 	return (IGB_SUCCESS);
1448 
1449 reset_failure:
1450 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1451 		mutex_exit(&igb->tx_rings[i].tx_lock);
1452 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1453 		mutex_exit(&igb->rx_rings[i].rx_lock);
1454 
1455 	mutex_exit(&igb->gen_lock);
1456 
1457 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1458 
1459 	return (IGB_FAILURE);
1460 }
1461 
1462 /*
1463  * igb_tx_clean - Clean the pending transmit packets and DMA resources
1464  */
1465 static void
1466 igb_tx_clean(igb_t *igb)
1467 {
1468 	igb_tx_ring_t *tx_ring;
1469 	tx_control_block_t *tcb;
1470 	link_list_t pending_list;
1471 	uint32_t desc_num;
1472 	int i, j;
1473 
1474 	LINK_LIST_INIT(&pending_list);
1475 
1476 	for (i = 0; i < igb->num_tx_rings; i++) {
1477 		tx_ring = &igb->tx_rings[i];
1478 
1479 		mutex_enter(&tx_ring->recycle_lock);
1480 
1481 		/*
1482 		 * Clean the pending tx data - the pending packets in the
1483 		 * work_list that have no chances to be transmitted again.
1484 		 *
1485 		 * We must ensure the chipset is stopped or the link is down
1486 		 * before cleaning the transmit packets.
1487 		 */
1488 		desc_num = 0;
1489 		for (j = 0; j < tx_ring->ring_size; j++) {
1490 			tcb = tx_ring->work_list[j];
1491 			if (tcb != NULL) {
1492 				desc_num += tcb->desc_num;
1493 
1494 				tx_ring->work_list[j] = NULL;
1495 
1496 				igb_free_tcb(tcb);
1497 
1498 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1499 			}
1500 		}
1501 
1502 		if (desc_num > 0) {
1503 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1504 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1505 
1506 			/*
1507 			 * Reset the head and tail pointers of the tbd ring;
1508 			 * Reset the head write-back if it is enabled.
1509 			 */
1510 			tx_ring->tbd_head = 0;
1511 			tx_ring->tbd_tail = 0;
1512 			if (igb->tx_head_wb_enable)
1513 				*tx_ring->tbd_head_wb = 0;
1514 
1515 			E1000_WRITE_REG(&igb->hw, E1000_TDH(tx_ring->index), 0);
1516 			E1000_WRITE_REG(&igb->hw, E1000_TDT(tx_ring->index), 0);
1517 		}
1518 
1519 		mutex_exit(&tx_ring->recycle_lock);
1520 
1521 		/*
1522 		 * Add the tx control blocks in the pending list to
1523 		 * the free list.
1524 		 */
1525 		igb_put_free_list(tx_ring, &pending_list);
1526 	}
1527 }
1528 
1529 /*
1530  * igb_tx_drain - Drain the tx rings to allow pending packets to be transmitted
1531  */
1532 static boolean_t
1533 igb_tx_drain(igb_t *igb)
1534 {
1535 	igb_tx_ring_t *tx_ring;
1536 	boolean_t done;
1537 	int i, j;
1538 
1539 	/*
1540 	 * Wait for a specific time to allow pending tx packets
1541 	 * to be transmitted.
1542 	 *
1543 	 * Check the counter tbd_free to see if transmission is done.
1544 	 * No lock protection is needed here.
1545 	 *
1546 	 * Return B_TRUE if all pending packets have been transmitted;
1547 	 * Otherwise return B_FALSE;
1548 	 */
1549 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1550 
1551 		done = B_TRUE;
1552 		for (j = 0; j < igb->num_tx_rings; j++) {
1553 			tx_ring = &igb->tx_rings[j];
1554 			done = done &&
1555 			    (tx_ring->tbd_free == tx_ring->ring_size);
1556 		}
1557 
1558 		if (done)
1559 			break;
1560 
1561 		msec_delay(1);
1562 	}
1563 
1564 	return (done);
1565 }
1566 
1567 /*
1568  * igb_rx_drain - Wait for all rx buffers to be released by upper layer
1569  */
1570 static boolean_t
1571 igb_rx_drain(igb_t *igb)
1572 {
1573 	igb_rx_ring_t *rx_ring;
1574 	boolean_t done;
1575 	int i, j;
1576 
1577 	/*
1578 	 * Polling the rx free list to check if those rx buffers held by
1579 	 * the upper layer are released.
1580 	 *
1581 	 * Check the counter rcb_free to see if all pending buffers are
1582 	 * released. No lock protection is needed here.
1583 	 *
1584 	 * Return B_TRUE if all pending buffers have been released;
1585 	 * Otherwise return B_FALSE;
1586 	 */
1587 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1588 
1589 		done = B_TRUE;
1590 		for (j = 0; j < igb->num_rx_rings; j++) {
1591 			rx_ring = &igb->rx_rings[j];
1592 			done = done &&
1593 			    (rx_ring->rcb_free == rx_ring->free_list_size);
1594 		}
1595 
1596 		if (done)
1597 			break;
1598 
1599 		msec_delay(1);
1600 	}
1601 
1602 	return (done);
1603 }
1604 
1605 /*
1606  * igb_start - Start the driver/chipset
1607  */
1608 int
1609 igb_start(igb_t *igb)
1610 {
1611 	int i;
1612 
1613 	ASSERT(mutex_owned(&igb->gen_lock));
1614 
1615 	for (i = 0; i < igb->num_rx_rings; i++)
1616 		mutex_enter(&igb->rx_rings[i].rx_lock);
1617 	for (i = 0; i < igb->num_tx_rings; i++)
1618 		mutex_enter(&igb->tx_rings[i].tx_lock);
1619 
1620 	/*
1621 	 * Start the adapter
1622 	 */
1623 	if ((igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) == 0) {
1624 		if (igb_init_adapter(igb) != IGB_SUCCESS) {
1625 			igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1626 			goto start_failure;
1627 		}
1628 		igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
1629 
1630 		/*
1631 		 * Setup the rx/tx rings
1632 		 */
1633 		igb_setup_rings(igb);
1634 	}
1635 
1636 	/*
1637 	 * Enable adapter interrupts
1638 	 * The interrupts must be enabled after the driver state is START
1639 	 */
1640 	igb->capab->enable_intr(igb);
1641 
1642 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1643 		goto start_failure;
1644 
1645 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1646 		goto start_failure;
1647 
1648 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1649 		mutex_exit(&igb->tx_rings[i].tx_lock);
1650 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1651 		mutex_exit(&igb->rx_rings[i].rx_lock);
1652 
1653 	return (IGB_SUCCESS);
1654 
1655 start_failure:
1656 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1657 		mutex_exit(&igb->tx_rings[i].tx_lock);
1658 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1659 		mutex_exit(&igb->rx_rings[i].rx_lock);
1660 
1661 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1662 
1663 	return (IGB_FAILURE);
1664 }
1665 
1666 /*
1667  * igb_stop - Stop the driver/chipset
1668  */
1669 void
1670 igb_stop(igb_t *igb)
1671 {
1672 	int i;
1673 
1674 	ASSERT(mutex_owned(&igb->gen_lock));
1675 
1676 	igb->attach_progress &= ~ATTACH_PROGRESS_INIT_ADAPTER;
1677 
1678 	/*
1679 	 * Disable the adapter interrupts
1680 	 */
1681 	igb_disable_adapter_interrupts(igb);
1682 
1683 	/*
1684 	 * Drain the pending tx packets
1685 	 */
1686 	(void) igb_tx_drain(igb);
1687 
1688 	for (i = 0; i < igb->num_rx_rings; i++)
1689 		mutex_enter(&igb->rx_rings[i].rx_lock);
1690 	for (i = 0; i < igb->num_tx_rings; i++)
1691 		mutex_enter(&igb->tx_rings[i].tx_lock);
1692 
1693 	/*
1694 	 * Stop the adapter
1695 	 */
1696 	igb_stop_adapter(igb);
1697 
1698 	/*
1699 	 * Clean the pending tx data/resources
1700 	 */
1701 	igb_tx_clean(igb);
1702 
1703 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1704 		mutex_exit(&igb->tx_rings[i].tx_lock);
1705 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1706 		mutex_exit(&igb->rx_rings[i].rx_lock);
1707 
1708 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1709 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1710 }
1711 
1712 /*
1713  * igb_alloc_rings - Allocate memory space for rx/tx rings
1714  */
1715 static int
1716 igb_alloc_rings(igb_t *igb)
1717 {
1718 	/*
1719 	 * Allocate memory space for rx rings
1720 	 */
1721 	igb->rx_rings = kmem_zalloc(
1722 	    sizeof (igb_rx_ring_t) * igb->num_rx_rings,
1723 	    KM_NOSLEEP);
1724 
1725 	if (igb->rx_rings == NULL) {
1726 		return (IGB_FAILURE);
1727 	}
1728 
1729 	/*
1730 	 * Allocate memory space for tx rings
1731 	 */
1732 	igb->tx_rings = kmem_zalloc(
1733 	    sizeof (igb_tx_ring_t) * igb->num_tx_rings,
1734 	    KM_NOSLEEP);
1735 
1736 	if (igb->tx_rings == NULL) {
1737 		kmem_free(igb->rx_rings,
1738 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1739 		igb->rx_rings = NULL;
1740 		return (IGB_FAILURE);
1741 	}
1742 
1743 	/*
1744 	 * Allocate memory space for rx ring groups
1745 	 */
1746 	igb->rx_groups = kmem_zalloc(
1747 	    sizeof (igb_rx_group_t) * igb->num_rx_groups,
1748 	    KM_NOSLEEP);
1749 
1750 	if (igb->rx_groups == NULL) {
1751 		kmem_free(igb->rx_rings,
1752 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1753 		kmem_free(igb->tx_rings,
1754 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1755 		igb->rx_rings = NULL;
1756 		igb->tx_rings = NULL;
1757 		return (IGB_FAILURE);
1758 	}
1759 
1760 	return (IGB_SUCCESS);
1761 }
1762 
1763 /*
1764  * igb_free_rings - Free the memory space of rx/tx rings.
1765  */
1766 static void
1767 igb_free_rings(igb_t *igb)
1768 {
1769 	if (igb->rx_rings != NULL) {
1770 		kmem_free(igb->rx_rings,
1771 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1772 		igb->rx_rings = NULL;
1773 	}
1774 
1775 	if (igb->tx_rings != NULL) {
1776 		kmem_free(igb->tx_rings,
1777 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1778 		igb->tx_rings = NULL;
1779 	}
1780 
1781 	if (igb->rx_groups != NULL) {
1782 		kmem_free(igb->rx_groups,
1783 		    sizeof (igb_rx_group_t) * igb->num_rx_groups);
1784 		igb->rx_groups = NULL;
1785 	}
1786 }
1787 
1788 /*
1789  * igb_setup_rings - Setup rx/tx rings
1790  */
1791 static void
1792 igb_setup_rings(igb_t *igb)
1793 {
1794 	/*
1795 	 * Setup the rx/tx rings, including the following:
1796 	 *
1797 	 * 1. Setup the descriptor ring and the control block buffers;
1798 	 * 2. Initialize necessary registers for receive/transmit;
1799 	 * 3. Initialize software pointers/parameters for receive/transmit;
1800 	 */
1801 	igb_setup_rx(igb);
1802 
1803 	igb_setup_tx(igb);
1804 
1805 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1806 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1807 }
1808 
1809 static void
1810 igb_setup_rx_ring(igb_rx_ring_t *rx_ring)
1811 {
1812 	igb_t *igb = rx_ring->igb;
1813 	struct e1000_hw *hw = &igb->hw;
1814 	rx_control_block_t *rcb;
1815 	union e1000_adv_rx_desc	*rbd;
1816 	uint32_t size;
1817 	uint32_t buf_low;
1818 	uint32_t buf_high;
1819 	uint32_t rxdctl;
1820 	int i;
1821 
1822 	ASSERT(mutex_owned(&rx_ring->rx_lock));
1823 	ASSERT(mutex_owned(&igb->gen_lock));
1824 
1825 	/*
1826 	 * Initialize descriptor ring with buffer addresses
1827 	 */
1828 	for (i = 0; i < igb->rx_ring_size; i++) {
1829 		rcb = rx_ring->work_list[i];
1830 		rbd = &rx_ring->rbd_ring[i];
1831 
1832 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
1833 		rbd->read.hdr_addr = NULL;
1834 	}
1835 
1836 	/*
1837 	 * Initialize the base address registers
1838 	 */
1839 	buf_low = (uint32_t)rx_ring->rbd_area.dma_address;
1840 	buf_high = (uint32_t)(rx_ring->rbd_area.dma_address >> 32);
1841 	E1000_WRITE_REG(hw, E1000_RDBAH(rx_ring->index), buf_high);
1842 	E1000_WRITE_REG(hw, E1000_RDBAL(rx_ring->index), buf_low);
1843 
1844 	/*
1845 	 * Initialize the length register
1846 	 */
1847 	size = rx_ring->ring_size * sizeof (union e1000_adv_rx_desc);
1848 	E1000_WRITE_REG(hw, E1000_RDLEN(rx_ring->index), size);
1849 
1850 	/*
1851 	 * Initialize buffer size & descriptor type
1852 	 */
1853 	E1000_WRITE_REG(hw, E1000_SRRCTL(rx_ring->index),
1854 	    ((igb->rx_buf_size >> E1000_SRRCTL_BSIZEPKT_SHIFT) |
1855 	    E1000_SRRCTL_DESCTYPE_ADV_ONEBUF));
1856 
1857 	/*
1858 	 * Setup the Receive Descriptor Control Register (RXDCTL)
1859 	 */
1860 	rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(rx_ring->index));
1861 	rxdctl &= igb->capab->rxdctl_mask;
1862 	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1863 	rxdctl |= 16;		/* pthresh */
1864 	rxdctl |= 8 << 8;	/* hthresh */
1865 	rxdctl |= 1 << 16;	/* wthresh */
1866 	E1000_WRITE_REG(hw, E1000_RXDCTL(rx_ring->index), rxdctl);
1867 
1868 	rx_ring->rbd_next = 0;
1869 
1870 	/*
1871 	 * Note: Considering the case that the chipset is being reset
1872 	 * and there are still some buffers held by the upper layer,
1873 	 * we should not reset the values of rcb_head, rcb_tail and
1874 	 * rcb_free;
1875 	 */
1876 	if (igb->igb_state == IGB_UNKNOWN) {
1877 		rx_ring->rcb_head = 0;
1878 		rx_ring->rcb_tail = 0;
1879 		rx_ring->rcb_free = rx_ring->free_list_size;
1880 	}
1881 }
1882 
1883 static void
1884 igb_setup_rx(igb_t *igb)
1885 {
1886 	igb_rx_ring_t *rx_ring;
1887 	igb_rx_group_t *rx_group;
1888 	struct e1000_hw *hw = &igb->hw;
1889 	uint32_t rctl, rxcsum;
1890 	uint32_t ring_per_group;
1891 	int i;
1892 
1893 	/*
1894 	 * Setup the Receive Control Register (RCTL), and enable the
1895 	 * receiver. The initial configuration is to: enable the receiver,
1896 	 * accept broadcasts, discard bad packets, accept long packets,
1897 	 * disable VLAN filter checking, and set receive buffer size to
1898 	 * 2k.  For 82575, also set the receive descriptor minimum
1899 	 * threshold size to 1/2 the ring.
1900 	 */
1901 	rctl = E1000_READ_REG(hw, E1000_RCTL);
1902 
1903 	/*
1904 	 * Clear the field used for wakeup control.  This driver doesn't do
1905 	 * wakeup but leave this here for completeness.
1906 	 */
1907 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1908 
1909 	switch (hw->mac.type) {
1910 	case e1000_82575:
1911 		rctl |= (E1000_RCTL_EN |	/* Enable Receive Unit */
1912 		    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
1913 		    E1000_RCTL_LPE |		/* Large Packet Enable */
1914 						/* Multicast filter offset */
1915 		    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
1916 		    E1000_RCTL_RDMTS_HALF |	/* rx descriptor threshold */
1917 		    E1000_RCTL_SECRC);		/* Strip Ethernet CRC */
1918 		break;
1919 
1920 	case e1000_82576:
1921 		rctl |= (E1000_RCTL_EN |	/* Enable Receive Unit */
1922 		    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
1923 		    E1000_RCTL_LPE |		/* Large Packet Enable */
1924 						/* Multicast filter offset */
1925 		    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
1926 		    E1000_RCTL_SECRC);		/* Strip Ethernet CRC */
1927 		break;
1928 
1929 	default:
1930 		igb_log(igb, "unsupported MAC type: %d", hw->mac.type);
1931 		return;	/* should never come here; this will cause rx failure */
1932 	}
1933 
1934 	for (i = 0; i < igb->num_rx_groups; i++) {
1935 		rx_group = &igb->rx_groups[i];
1936 		rx_group->index = i;
1937 		rx_group->igb = igb;
1938 	}
1939 
1940 	/*
1941 	 * Set up all rx descriptor rings - must be called before receive unit
1942 	 * enabled.
1943 	 */
1944 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
1945 	for (i = 0; i < igb->num_rx_rings; i++) {
1946 		rx_ring = &igb->rx_rings[i];
1947 		igb_setup_rx_ring(rx_ring);
1948 
1949 		/*
1950 		 * Map a ring to a group by assigning a group index
1951 		 */
1952 		rx_ring->group_index = i / ring_per_group;
1953 	}
1954 
1955 	/*
1956 	 * Setup the Rx Long Packet Max Length register
1957 	 */
1958 	E1000_WRITE_REG(hw, E1000_RLPML, igb->max_frame_size);
1959 
1960 	/*
1961 	 * Hardware checksum settings
1962 	 */
1963 	if (igb->rx_hcksum_enable) {
1964 		rxcsum =
1965 		    E1000_RXCSUM_TUOFL |	/* TCP/UDP checksum */
1966 		    E1000_RXCSUM_IPOFL;		/* IP checksum */
1967 
1968 		E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
1969 	}
1970 
1971 	/*
1972 	 * Setup classify and RSS for multiple receive queues
1973 	 */
1974 	switch (igb->vmdq_mode) {
1975 	case E1000_VMDQ_OFF:
1976 		/*
1977 		 * One ring group, only RSS is needed when more than
1978 		 * one ring enabled.
1979 		 */
1980 		if (igb->num_rx_rings > 1)
1981 			igb_setup_rss(igb);
1982 		break;
1983 	case E1000_VMDQ_MAC:
1984 		/*
1985 		 * Multiple groups, each group has one ring,
1986 		 * only the MAC classification is needed.
1987 		 */
1988 		igb_setup_mac_classify(igb);
1989 		break;
1990 	case E1000_VMDQ_MAC_RSS:
1991 		/*
1992 		 * Multiple groups and multiple rings, both
1993 		 * MAC classification and RSS are needed.
1994 		 */
1995 		igb_setup_mac_rss_classify(igb);
1996 		break;
1997 	}
1998 
1999 	/*
2000 	 * Enable the receive unit - must be done after all
2001 	 * the rx setup above.
2002 	 */
2003 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2004 
2005 	/*
2006 	 * Initialize all adapter ring head & tail pointers - must
2007 	 * be done after receive unit is enabled
2008 	 */
2009 	for (i = 0; i < igb->num_rx_rings; i++) {
2010 		rx_ring = &igb->rx_rings[i];
2011 		E1000_WRITE_REG(hw, E1000_RDH(i), 0);
2012 		E1000_WRITE_REG(hw, E1000_RDT(i), rx_ring->ring_size - 1);
2013 	}
2014 
2015 	/*
2016 	 * 82575 with manageability enabled needs a special flush to make
2017 	 * sure the fifos start clean.
2018 	 */
2019 	if ((hw->mac.type == e1000_82575) &&
2020 	    (E1000_READ_REG(hw, E1000_MANC) & E1000_MANC_RCV_TCO_EN)) {
2021 		e1000_rx_fifo_flush_82575(hw);
2022 	}
2023 }
2024 
2025 static void
2026 igb_setup_tx_ring(igb_tx_ring_t *tx_ring)
2027 {
2028 	igb_t *igb = tx_ring->igb;
2029 	struct e1000_hw *hw = &igb->hw;
2030 	uint32_t size;
2031 	uint32_t buf_low;
2032 	uint32_t buf_high;
2033 	uint32_t reg_val;
2034 
2035 	ASSERT(mutex_owned(&tx_ring->tx_lock));
2036 	ASSERT(mutex_owned(&igb->gen_lock));
2037 
2038 
2039 	/*
2040 	 * Initialize the length register
2041 	 */
2042 	size = tx_ring->ring_size * sizeof (union e1000_adv_tx_desc);
2043 	E1000_WRITE_REG(hw, E1000_TDLEN(tx_ring->index), size);
2044 
2045 	/*
2046 	 * Initialize the base address registers
2047 	 */
2048 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
2049 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
2050 	E1000_WRITE_REG(hw, E1000_TDBAL(tx_ring->index), buf_low);
2051 	E1000_WRITE_REG(hw, E1000_TDBAH(tx_ring->index), buf_high);
2052 
2053 	/*
2054 	 * Setup head & tail pointers
2055 	 */
2056 	E1000_WRITE_REG(hw, E1000_TDH(tx_ring->index), 0);
2057 	E1000_WRITE_REG(hw, E1000_TDT(tx_ring->index), 0);
2058 
2059 	/*
2060 	 * Setup head write-back
2061 	 */
2062 	if (igb->tx_head_wb_enable) {
2063 		/*
2064 		 * The memory of the head write-back is allocated using
2065 		 * the extra tbd beyond the tail of the tbd ring.
2066 		 */
2067 		tx_ring->tbd_head_wb = (uint32_t *)
2068 		    ((uintptr_t)tx_ring->tbd_area.address + size);
2069 		*tx_ring->tbd_head_wb = 0;
2070 
2071 		buf_low = (uint32_t)
2072 		    (tx_ring->tbd_area.dma_address + size);
2073 		buf_high = (uint32_t)
2074 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
2075 
2076 		/* Set the head write-back enable bit */
2077 		buf_low |= E1000_TX_HEAD_WB_ENABLE;
2078 
2079 		E1000_WRITE_REG(hw, E1000_TDWBAL(tx_ring->index), buf_low);
2080 		E1000_WRITE_REG(hw, E1000_TDWBAH(tx_ring->index), buf_high);
2081 
2082 		/*
2083 		 * Turn off relaxed ordering for head write back or it will
2084 		 * cause problems with the tx recycling
2085 		 */
2086 		reg_val = E1000_READ_REG(hw,
2087 		    E1000_DCA_TXCTRL(tx_ring->index));
2088 		reg_val &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
2089 		E1000_WRITE_REG(hw,
2090 		    E1000_DCA_TXCTRL(tx_ring->index), reg_val);
2091 	} else {
2092 		tx_ring->tbd_head_wb = NULL;
2093 	}
2094 
2095 	tx_ring->tbd_head = 0;
2096 	tx_ring->tbd_tail = 0;
2097 	tx_ring->tbd_free = tx_ring->ring_size;
2098 
2099 	/*
2100 	 * Note: for the case that the chipset is being reset, we should not
2101 	 * reset the values of tcb_head, tcb_tail. And considering there might
2102 	 * still be some packets kept in the pending_list, we should not assert
2103 	 * (tcb_free == free_list_size) here.
2104 	 */
2105 	if (igb->igb_state == IGB_UNKNOWN) {
2106 		tx_ring->tcb_head = 0;
2107 		tx_ring->tcb_tail = 0;
2108 		tx_ring->tcb_free = tx_ring->free_list_size;
2109 	}
2110 
2111 	/*
2112 	 * Initialize hardware checksum offload settings
2113 	 */
2114 	tx_ring->hcksum_context.hcksum_flags = 0;
2115 	tx_ring->hcksum_context.ip_hdr_len = 0;
2116 	tx_ring->hcksum_context.mac_hdr_len = 0;
2117 	tx_ring->hcksum_context.l4_proto = 0;
2118 
2119 	/*
2120 	 * Enable TXDCTL per queue
2121 	 */
2122 	reg_val = E1000_READ_REG(hw, E1000_TXDCTL(tx_ring->index));
2123 	reg_val |= E1000_TXDCTL_QUEUE_ENABLE;
2124 	E1000_WRITE_REG(hw, E1000_TXDCTL(tx_ring->index), reg_val);
2125 }
2126 
2127 static void
2128 igb_setup_tx(igb_t *igb)
2129 {
2130 	igb_tx_ring_t *tx_ring;
2131 	struct e1000_hw *hw = &igb->hw;
2132 	uint32_t reg_val;
2133 	int i;
2134 
2135 	for (i = 0; i < igb->num_tx_rings; i++) {
2136 		tx_ring = &igb->tx_rings[i];
2137 		igb_setup_tx_ring(tx_ring);
2138 	}
2139 
2140 	/*
2141 	 * Setup the Transmit Control Register (TCTL)
2142 	 */
2143 	reg_val = E1000_READ_REG(hw, E1000_TCTL);
2144 	reg_val &= ~E1000_TCTL_CT;
2145 	reg_val |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2146 	    (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2147 
2148 	/* Enable transmits */
2149 	reg_val |= E1000_TCTL_EN;
2150 
2151 	E1000_WRITE_REG(hw, E1000_TCTL, reg_val);
2152 }
2153 
2154 /*
2155  * igb_setup_rss - Setup receive-side scaling feature
2156  */
2157 static void
2158 igb_setup_rss(igb_t *igb)
2159 {
2160 	struct e1000_hw *hw = &igb->hw;
2161 	uint32_t i, mrqc, rxcsum;
2162 	int shift = 0;
2163 	uint32_t random;
2164 	union e1000_reta {
2165 		uint32_t	dword;
2166 		uint8_t		bytes[4];
2167 	} reta;
2168 
2169 	/* Setup the Redirection Table */
2170 	if (hw->mac.type == e1000_82576) {
2171 		shift = 0;
2172 	} else if (hw->mac.type == e1000_82575) {
2173 		shift = 6;
2174 	}
2175 	for (i = 0; i < (32 * 4); i++) {
2176 		reta.bytes[i & 3] = (i % igb->num_rx_rings) << shift;
2177 		if ((i & 3) == 3) {
2178 			E1000_WRITE_REG(hw,
2179 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2180 		}
2181 	}
2182 
2183 	/* Fill out hash function seeds */
2184 	for (i = 0; i < 10; i++) {
2185 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2186 		    sizeof (uint32_t));
2187 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2188 	}
2189 
2190 	/* Setup the Multiple Receive Queue Control register */
2191 	mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2192 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2193 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2194 	    E1000_MRQC_RSS_FIELD_IPV6 |
2195 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2196 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2197 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2198 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2199 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2200 
2201 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2202 
2203 	/*
2204 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2205 	 *
2206 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2207 	 * checksum offloading provided by the 82575 chipset besides the IP
2208 	 * header checksum offloading and the TCP/UDP checksum offloading.
2209 	 * The Packet Checksum is by default computed over the entire packet
2210 	 * from the first byte of the DA through the last byte of the CRC,
2211 	 * including the Ethernet and IP headers.
2212 	 *
2213 	 * It is a hardware limitation that Packet Checksum is mutually
2214 	 * exclusive with RSS.
2215 	 */
2216 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2217 	rxcsum |= E1000_RXCSUM_PCSD;
2218 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2219 }
2220 
2221 /*
2222  * igb_setup_mac_rss_classify - Setup MAC classification and rss
2223  */
2224 static void
2225 igb_setup_mac_rss_classify(igb_t *igb)
2226 {
2227 	struct e1000_hw *hw = &igb->hw;
2228 	uint32_t i, mrqc, vmdctl, rxcsum;
2229 	uint32_t ring_per_group;
2230 	int shift_group0, shift_group1;
2231 	uint32_t random;
2232 	union e1000_reta {
2233 		uint32_t	dword;
2234 		uint8_t		bytes[4];
2235 	} reta;
2236 
2237 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2238 
2239 	/* Setup the Redirection Table, it is shared between two groups */
2240 	shift_group0 = 2;
2241 	shift_group1 = 6;
2242 	for (i = 0; i < (32 * 4); i++) {
2243 		reta.bytes[i & 3] = ((i % ring_per_group) << shift_group0) |
2244 		    ((ring_per_group + (i % ring_per_group)) << shift_group1);
2245 		if ((i & 3) == 3) {
2246 			E1000_WRITE_REG(hw,
2247 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2248 		}
2249 	}
2250 
2251 	/* Fill out hash function seeds */
2252 	for (i = 0; i < 10; i++) {
2253 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2254 		    sizeof (uint32_t));
2255 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2256 	}
2257 
2258 	/*
2259 	 * Setup the Multiple Receive Queue Control register,
2260 	 * enable VMDq based on packet destination MAC address and RSS.
2261 	 */
2262 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_RSS_GROUP;
2263 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2264 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2265 	    E1000_MRQC_RSS_FIELD_IPV6 |
2266 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2267 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2268 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2269 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2270 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2271 
2272 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2273 
2274 
2275 	/* Define the default group and default queues */
2276 	vmdctl = E1000_VMDQ_MAC_GROUP_DEFAULT_QUEUE;
2277 	E1000_WRITE_REG(hw, E1000_VT_CTL, vmdctl);
2278 
2279 	/*
2280 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2281 	 *
2282 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2283 	 * checksum offloading provided by the 82575 chipset besides the IP
2284 	 * header checksum offloading and the TCP/UDP checksum offloading.
2285 	 * The Packet Checksum is by default computed over the entire packet
2286 	 * from the first byte of the DA through the last byte of the CRC,
2287 	 * including the Ethernet and IP headers.
2288 	 *
2289 	 * It is a hardware limitation that Packet Checksum is mutually
2290 	 * exclusive with RSS.
2291 	 */
2292 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2293 	rxcsum |= E1000_RXCSUM_PCSD;
2294 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2295 }
2296 
2297 /*
2298  * igb_setup_mac_classify - Setup MAC classification feature
2299  */
2300 static void
2301 igb_setup_mac_classify(igb_t *igb)
2302 {
2303 	struct e1000_hw *hw = &igb->hw;
2304 	uint32_t mrqc, rxcsum;
2305 
2306 	/*
2307 	 * Setup the Multiple Receive Queue Control register,
2308 	 * enable VMDq based on packet destination MAC address.
2309 	 */
2310 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_GROUP;
2311 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2312 
2313 	/*
2314 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2315 	 *
2316 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2317 	 * checksum offloading provided by the 82575 chipset besides the IP
2318 	 * header checksum offloading and the TCP/UDP checksum offloading.
2319 	 * The Packet Checksum is by default computed over the entire packet
2320 	 * from the first byte of the DA through the last byte of the CRC,
2321 	 * including the Ethernet and IP headers.
2322 	 *
2323 	 * It is a hardware limitation that Packet Checksum is mutually
2324 	 * exclusive with RSS.
2325 	 */
2326 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2327 	rxcsum |= E1000_RXCSUM_PCSD;
2328 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2329 
2330 }
2331 
2332 /*
2333  * igb_init_unicst - Initialize the unicast addresses
2334  */
2335 static void
2336 igb_init_unicst(igb_t *igb)
2337 {
2338 	struct e1000_hw *hw = &igb->hw;
2339 	int slot;
2340 
2341 	/*
2342 	 * Here we should consider two situations:
2343 	 *
2344 	 * 1. Chipset is initialized the first time
2345 	 *    Initialize the multiple unicast addresses, and
2346 	 *    save the default MAC address.
2347 	 *
2348 	 * 2. Chipset is reset
2349 	 *    Recover the multiple unicast addresses from the
2350 	 *    software data structure to the RAR registers.
2351 	 */
2352 
2353 	/*
2354 	 * Clear the default MAC address in the RAR0 rgister,
2355 	 * which is loaded from EEPROM when system boot or chipreset,
2356 	 * this will cause the conficts with add_mac/rem_mac entry
2357 	 * points when VMDq is enabled. For this reason, the RAR0
2358 	 * must be cleared for both cases mentioned above.
2359 	 */
2360 	e1000_rar_clear(hw, 0);
2361 
2362 	if (!igb->unicst_init) {
2363 
2364 		/* Initialize the multiple unicast addresses */
2365 		igb->unicst_total = MAX_NUM_UNICAST_ADDRESSES;
2366 		igb->unicst_avail = igb->unicst_total;
2367 
2368 		for (slot = 0; slot < igb->unicst_total; slot++)
2369 			igb->unicst_addr[slot].mac.set = 0;
2370 
2371 		igb->unicst_init = B_TRUE;
2372 	} else {
2373 		/* Re-configure the RAR registers */
2374 		for (slot = 0; slot < igb->unicst_total; slot++) {
2375 			e1000_rar_set_vmdq(hw, igb->unicst_addr[slot].mac.addr,
2376 			    slot, igb->vmdq_mode,
2377 			    igb->unicst_addr[slot].mac.group_index);
2378 		}
2379 	}
2380 
2381 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2382 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2383 }
2384 
2385 /*
2386  * igb_unicst_find - Find the slot for the specified unicast address
2387  */
2388 int
2389 igb_unicst_find(igb_t *igb, const uint8_t *mac_addr)
2390 {
2391 	int slot;
2392 
2393 	ASSERT(mutex_owned(&igb->gen_lock));
2394 
2395 	for (slot = 0; slot < igb->unicst_total; slot++) {
2396 		if (bcmp(igb->unicst_addr[slot].mac.addr,
2397 		    mac_addr, ETHERADDRL) == 0)
2398 			return (slot);
2399 	}
2400 
2401 	return (-1);
2402 }
2403 
2404 /*
2405  * igb_unicst_set - Set the unicast address to the specified slot
2406  */
2407 int
2408 igb_unicst_set(igb_t *igb, const uint8_t *mac_addr,
2409     int slot)
2410 {
2411 	struct e1000_hw *hw = &igb->hw;
2412 
2413 	ASSERT(mutex_owned(&igb->gen_lock));
2414 
2415 	/*
2416 	 * Save the unicast address in the software data structure
2417 	 */
2418 	bcopy(mac_addr, igb->unicst_addr[slot].mac.addr, ETHERADDRL);
2419 
2420 	/*
2421 	 * Set the unicast address to the RAR register
2422 	 */
2423 	e1000_rar_set(hw, (uint8_t *)mac_addr, slot);
2424 
2425 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2426 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2427 		return (EIO);
2428 	}
2429 
2430 	return (0);
2431 }
2432 
2433 /*
2434  * igb_multicst_add - Add a multicst address
2435  */
2436 int
2437 igb_multicst_add(igb_t *igb, const uint8_t *multiaddr)
2438 {
2439 	ASSERT(mutex_owned(&igb->gen_lock));
2440 
2441 	if ((multiaddr[0] & 01) == 0) {
2442 		return (EINVAL);
2443 	}
2444 
2445 	if (igb->mcast_count >= MAX_NUM_MULTICAST_ADDRESSES) {
2446 		return (ENOENT);
2447 	}
2448 
2449 	bcopy(multiaddr,
2450 	    &igb->mcast_table[igb->mcast_count], ETHERADDRL);
2451 	igb->mcast_count++;
2452 
2453 	/*
2454 	 * Update the multicast table in the hardware
2455 	 */
2456 	igb_setup_multicst(igb);
2457 
2458 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2459 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2460 		return (EIO);
2461 	}
2462 
2463 	return (0);
2464 }
2465 
2466 /*
2467  * igb_multicst_remove - Remove a multicst address
2468  */
2469 int
2470 igb_multicst_remove(igb_t *igb, const uint8_t *multiaddr)
2471 {
2472 	int i;
2473 
2474 	ASSERT(mutex_owned(&igb->gen_lock));
2475 
2476 	for (i = 0; i < igb->mcast_count; i++) {
2477 		if (bcmp(multiaddr, &igb->mcast_table[i],
2478 		    ETHERADDRL) == 0) {
2479 			for (i++; i < igb->mcast_count; i++) {
2480 				igb->mcast_table[i - 1] =
2481 				    igb->mcast_table[i];
2482 			}
2483 			igb->mcast_count--;
2484 			break;
2485 		}
2486 	}
2487 
2488 	/*
2489 	 * Update the multicast table in the hardware
2490 	 */
2491 	igb_setup_multicst(igb);
2492 
2493 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2494 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2495 		return (EIO);
2496 	}
2497 
2498 	return (0);
2499 }
2500 
2501 /*
2502  * igb_setup_multicast - setup multicast data structures
2503  *
2504  * This routine initializes all of the multicast related structures
2505  * and save them in the hardware registers.
2506  */
2507 static void
2508 igb_setup_multicst(igb_t *igb)
2509 {
2510 	uint8_t *mc_addr_list;
2511 	uint32_t mc_addr_count;
2512 	struct e1000_hw *hw = &igb->hw;
2513 
2514 	ASSERT(mutex_owned(&igb->gen_lock));
2515 
2516 	ASSERT(igb->mcast_count <= MAX_NUM_MULTICAST_ADDRESSES);
2517 
2518 	mc_addr_list = (uint8_t *)igb->mcast_table;
2519 	mc_addr_count = igb->mcast_count;
2520 
2521 	/*
2522 	 * Update the multicase addresses to the MTA registers
2523 	 */
2524 	e1000_update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2525 	    igb->unicst_total, hw->mac.rar_entry_count);
2526 }
2527 
2528 /*
2529  * igb_get_conf - Get driver configurations set in driver.conf
2530  *
2531  * This routine gets user-configured values out of the configuration
2532  * file igb.conf.
2533  *
2534  * For each configurable value, there is a minimum, a maximum, and a
2535  * default.
2536  * If user does not configure a value, use the default.
2537  * If user configures below the minimum, use the minumum.
2538  * If user configures above the maximum, use the maxumum.
2539  */
2540 static void
2541 igb_get_conf(igb_t *igb)
2542 {
2543 	struct e1000_hw *hw = &igb->hw;
2544 	uint32_t default_mtu;
2545 	uint32_t flow_control;
2546 	uint32_t ring_per_group;
2547 	int i;
2548 
2549 	/*
2550 	 * igb driver supports the following user configurations:
2551 	 *
2552 	 * Link configurations:
2553 	 *    adv_autoneg_cap
2554 	 *    adv_1000fdx_cap
2555 	 *    adv_100fdx_cap
2556 	 *    adv_100hdx_cap
2557 	 *    adv_10fdx_cap
2558 	 *    adv_10hdx_cap
2559 	 * Note: 1000hdx is not supported.
2560 	 *
2561 	 * Jumbo frame configuration:
2562 	 *    default_mtu
2563 	 *
2564 	 * Ethernet flow control configuration:
2565 	 *    flow_control
2566 	 *
2567 	 * Multiple rings configurations:
2568 	 *    tx_queue_number
2569 	 *    tx_ring_size
2570 	 *    rx_queue_number
2571 	 *    rx_ring_size
2572 	 *
2573 	 * Call igb_get_prop() to get the value for a specific
2574 	 * configuration parameter.
2575 	 */
2576 
2577 	/*
2578 	 * Link configurations
2579 	 */
2580 	igb->param_adv_autoneg_cap = igb_get_prop(igb,
2581 	    PROP_ADV_AUTONEG_CAP, 0, 1, 1);
2582 	igb->param_adv_1000fdx_cap = igb_get_prop(igb,
2583 	    PROP_ADV_1000FDX_CAP, 0, 1, 1);
2584 	igb->param_adv_100fdx_cap = igb_get_prop(igb,
2585 	    PROP_ADV_100FDX_CAP, 0, 1, 1);
2586 	igb->param_adv_100hdx_cap = igb_get_prop(igb,
2587 	    PROP_ADV_100HDX_CAP, 0, 1, 1);
2588 	igb->param_adv_10fdx_cap = igb_get_prop(igb,
2589 	    PROP_ADV_10FDX_CAP, 0, 1, 1);
2590 	igb->param_adv_10hdx_cap = igb_get_prop(igb,
2591 	    PROP_ADV_10HDX_CAP, 0, 1, 1);
2592 
2593 	/*
2594 	 * Jumbo frame configurations
2595 	 */
2596 	default_mtu = igb_get_prop(igb, PROP_DEFAULT_MTU,
2597 	    MIN_MTU, MAX_MTU, DEFAULT_MTU);
2598 
2599 	igb->max_frame_size = default_mtu +
2600 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
2601 
2602 	/*
2603 	 * Ethernet flow control configuration
2604 	 */
2605 	flow_control = igb_get_prop(igb, PROP_FLOW_CONTROL,
2606 	    e1000_fc_none, 4, e1000_fc_full);
2607 	if (flow_control == 4)
2608 		flow_control = e1000_fc_default;
2609 
2610 	hw->fc.requested_mode = flow_control;
2611 
2612 	/*
2613 	 * Multiple rings configurations
2614 	 */
2615 	igb->tx_ring_size = igb_get_prop(igb, PROP_TX_RING_SIZE,
2616 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
2617 	igb->rx_ring_size = igb_get_prop(igb, PROP_RX_RING_SIZE,
2618 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
2619 
2620 	igb->mr_enable = igb_get_prop(igb, PROP_MR_ENABLE, 0, 1, 1);
2621 	igb->num_rx_groups = igb_get_prop(igb, PROP_RX_GROUP_NUM,
2622 	    MIN_RX_GROUP_NUM, MAX_RX_GROUP_NUM, DEFAULT_RX_GROUP_NUM);
2623 	/*
2624 	 * Currently we do not support VMDq for 82576.
2625 	 * If it is e1000_82576, set num_rx_groups to 1.
2626 	 */
2627 	if (hw->mac.type == e1000_82576)
2628 		igb->num_rx_groups = 1;
2629 
2630 	if (igb->mr_enable) {
2631 		igb->num_tx_rings = igb->capab->def_tx_que_num;
2632 		igb->num_rx_rings = igb->capab->def_rx_que_num;
2633 	} else {
2634 		igb->num_tx_rings = 1;
2635 		igb->num_rx_rings = 1;
2636 
2637 		if (igb->num_rx_groups > 1) {
2638 			igb_error(igb,
2639 			    "Invalid rx groups number. Please enable multiple "
2640 			    "rings first");
2641 			igb->num_rx_groups = 1;
2642 		}
2643 	}
2644 
2645 	/*
2646 	 * Check the divisibility between rx rings and rx groups.
2647 	 */
2648 	for (i = igb->num_rx_groups; i > 0; i--) {
2649 		if ((igb->num_rx_rings % i) == 0)
2650 			break;
2651 	}
2652 	if (i != igb->num_rx_groups) {
2653 		igb_error(igb,
2654 		    "Invalid rx groups number. Downgrade the rx group "
2655 		    "number to %d.", i);
2656 		igb->num_rx_groups = i;
2657 	}
2658 
2659 	/*
2660 	 * Get the ring number per group.
2661 	 */
2662 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2663 
2664 	if (igb->num_rx_groups == 1) {
2665 		/*
2666 		 * One rx ring group, the rx ring number is num_rx_rings.
2667 		 */
2668 		igb->vmdq_mode = E1000_VMDQ_OFF;
2669 	} else if (ring_per_group == 1) {
2670 		/*
2671 		 * Multiple rx groups, each group has one rx ring.
2672 		 */
2673 		igb->vmdq_mode = E1000_VMDQ_MAC;
2674 	} else {
2675 		/*
2676 		 * Multiple groups and multiple rings.
2677 		 */
2678 		igb->vmdq_mode = E1000_VMDQ_MAC_RSS;
2679 	}
2680 
2681 	/*
2682 	 * Tunable used to force an interrupt type. The only use is
2683 	 * for testing of the lesser interrupt types.
2684 	 * 0 = don't force interrupt type
2685 	 * 1 = force interrupt type MSIX
2686 	 * 2 = force interrupt type MSI
2687 	 * 3 = force interrupt type Legacy
2688 	 */
2689 	igb->intr_force = igb_get_prop(igb, PROP_INTR_FORCE,
2690 	    IGB_INTR_NONE, IGB_INTR_LEGACY, IGB_INTR_NONE);
2691 
2692 	igb->tx_hcksum_enable = igb_get_prop(igb, PROP_TX_HCKSUM_ENABLE,
2693 	    0, 1, 1);
2694 	igb->rx_hcksum_enable = igb_get_prop(igb, PROP_RX_HCKSUM_ENABLE,
2695 	    0, 1, 1);
2696 	igb->lso_enable = igb_get_prop(igb, PROP_LSO_ENABLE,
2697 	    0, 1, 0);
2698 	igb->tx_head_wb_enable = igb_get_prop(igb, PROP_TX_HEAD_WB_ENABLE,
2699 	    0, 1, 1);
2700 
2701 	igb->tx_copy_thresh = igb_get_prop(igb, PROP_TX_COPY_THRESHOLD,
2702 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
2703 	    DEFAULT_TX_COPY_THRESHOLD);
2704 	igb->tx_recycle_thresh = igb_get_prop(igb, PROP_TX_RECYCLE_THRESHOLD,
2705 	    MIN_TX_RECYCLE_THRESHOLD, MAX_TX_RECYCLE_THRESHOLD,
2706 	    DEFAULT_TX_RECYCLE_THRESHOLD);
2707 	igb->tx_overload_thresh = igb_get_prop(igb, PROP_TX_OVERLOAD_THRESHOLD,
2708 	    MIN_TX_OVERLOAD_THRESHOLD, MAX_TX_OVERLOAD_THRESHOLD,
2709 	    DEFAULT_TX_OVERLOAD_THRESHOLD);
2710 	igb->tx_resched_thresh = igb_get_prop(igb, PROP_TX_RESCHED_THRESHOLD,
2711 	    MIN_TX_RESCHED_THRESHOLD, MAX_TX_RESCHED_THRESHOLD,
2712 	    DEFAULT_TX_RESCHED_THRESHOLD);
2713 
2714 	igb->rx_copy_thresh = igb_get_prop(igb, PROP_RX_COPY_THRESHOLD,
2715 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
2716 	    DEFAULT_RX_COPY_THRESHOLD);
2717 	igb->rx_limit_per_intr = igb_get_prop(igb, PROP_RX_LIMIT_PER_INTR,
2718 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
2719 	    DEFAULT_RX_LIMIT_PER_INTR);
2720 
2721 	igb->intr_throttling[0] = igb_get_prop(igb, PROP_INTR_THROTTLING,
2722 	    igb->capab->min_intr_throttle,
2723 	    igb->capab->max_intr_throttle,
2724 	    igb->capab->def_intr_throttle);
2725 }
2726 
2727 /*
2728  * igb_get_prop - Get a property value out of the configuration file igb.conf
2729  *
2730  * Caller provides the name of the property, a default value, a minimum
2731  * value, and a maximum value.
2732  *
2733  * Return configured value of the property, with default, minimum and
2734  * maximum properly applied.
2735  */
2736 static int
2737 igb_get_prop(igb_t *igb,
2738     char *propname,	/* name of the property */
2739     int minval,		/* minimum acceptable value */
2740     int maxval,		/* maximim acceptable value */
2741     int defval)		/* default value */
2742 {
2743 	int value;
2744 
2745 	/*
2746 	 * Call ddi_prop_get_int() to read the conf settings
2747 	 */
2748 	value = ddi_prop_get_int(DDI_DEV_T_ANY, igb->dip,
2749 	    DDI_PROP_DONTPASS, propname, defval);
2750 
2751 	if (value > maxval)
2752 		value = maxval;
2753 
2754 	if (value < minval)
2755 		value = minval;
2756 
2757 	return (value);
2758 }
2759 
2760 /*
2761  * igb_setup_link - Using the link properties to setup the link
2762  */
2763 int
2764 igb_setup_link(igb_t *igb, boolean_t setup_hw)
2765 {
2766 	struct e1000_mac_info *mac;
2767 	struct e1000_phy_info *phy;
2768 	boolean_t invalid;
2769 
2770 	mac = &igb->hw.mac;
2771 	phy = &igb->hw.phy;
2772 	invalid = B_FALSE;
2773 
2774 	if (igb->param_adv_autoneg_cap == 1) {
2775 		mac->autoneg = B_TRUE;
2776 		phy->autoneg_advertised = 0;
2777 
2778 		/*
2779 		 * 1000hdx is not supported for autonegotiation
2780 		 */
2781 		if (igb->param_adv_1000fdx_cap == 1)
2782 			phy->autoneg_advertised |= ADVERTISE_1000_FULL;
2783 
2784 		if (igb->param_adv_100fdx_cap == 1)
2785 			phy->autoneg_advertised |= ADVERTISE_100_FULL;
2786 
2787 		if (igb->param_adv_100hdx_cap == 1)
2788 			phy->autoneg_advertised |= ADVERTISE_100_HALF;
2789 
2790 		if (igb->param_adv_10fdx_cap == 1)
2791 			phy->autoneg_advertised |= ADVERTISE_10_FULL;
2792 
2793 		if (igb->param_adv_10hdx_cap == 1)
2794 			phy->autoneg_advertised |= ADVERTISE_10_HALF;
2795 
2796 		if (phy->autoneg_advertised == 0)
2797 			invalid = B_TRUE;
2798 	} else {
2799 		mac->autoneg = B_FALSE;
2800 
2801 		/*
2802 		 * 1000fdx and 1000hdx are not supported for forced link
2803 		 */
2804 		if (igb->param_adv_100fdx_cap == 1)
2805 			mac->forced_speed_duplex = ADVERTISE_100_FULL;
2806 		else if (igb->param_adv_100hdx_cap == 1)
2807 			mac->forced_speed_duplex = ADVERTISE_100_HALF;
2808 		else if (igb->param_adv_10fdx_cap == 1)
2809 			mac->forced_speed_duplex = ADVERTISE_10_FULL;
2810 		else if (igb->param_adv_10hdx_cap == 1)
2811 			mac->forced_speed_duplex = ADVERTISE_10_HALF;
2812 		else
2813 			invalid = B_TRUE;
2814 	}
2815 
2816 	if (invalid) {
2817 		igb_notice(igb, "Invalid link settings. Setup link to "
2818 		    "autonegotiation with full link capabilities.");
2819 		mac->autoneg = B_TRUE;
2820 		phy->autoneg_advertised = ADVERTISE_1000_FULL |
2821 		    ADVERTISE_100_FULL | ADVERTISE_100_HALF |
2822 		    ADVERTISE_10_FULL | ADVERTISE_10_HALF;
2823 	}
2824 
2825 	if (setup_hw) {
2826 		if (e1000_setup_link(&igb->hw) != E1000_SUCCESS)
2827 			return (IGB_FAILURE);
2828 	}
2829 
2830 	return (IGB_SUCCESS);
2831 }
2832 
2833 
2834 /*
2835  * igb_is_link_up - Check if the link is up
2836  */
2837 static boolean_t
2838 igb_is_link_up(igb_t *igb)
2839 {
2840 	struct e1000_hw *hw = &igb->hw;
2841 	boolean_t link_up = B_FALSE;
2842 
2843 	ASSERT(mutex_owned(&igb->gen_lock));
2844 
2845 	/*
2846 	 * get_link_status is set in the interrupt handler on link-status-change
2847 	 * or rx sequence error interrupt.  get_link_status will stay
2848 	 * false until the e1000_check_for_link establishes link only
2849 	 * for copper adapters.
2850 	 */
2851 	switch (hw->phy.media_type) {
2852 	case e1000_media_type_copper:
2853 		if (hw->mac.get_link_status) {
2854 			(void) e1000_check_for_link(hw);
2855 			link_up = !hw->mac.get_link_status;
2856 		} else {
2857 			link_up = B_TRUE;
2858 		}
2859 		break;
2860 	case e1000_media_type_fiber:
2861 		(void) e1000_check_for_link(hw);
2862 		link_up = (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU);
2863 		break;
2864 	case e1000_media_type_internal_serdes:
2865 		(void) e1000_check_for_link(hw);
2866 		link_up = hw->mac.serdes_has_link;
2867 		break;
2868 	}
2869 
2870 	return (link_up);
2871 }
2872 
2873 /*
2874  * igb_link_check - Link status processing
2875  */
2876 static boolean_t
2877 igb_link_check(igb_t *igb)
2878 {
2879 	struct e1000_hw *hw = &igb->hw;
2880 	uint16_t speed = 0, duplex = 0;
2881 	boolean_t link_changed = B_FALSE;
2882 
2883 	ASSERT(mutex_owned(&igb->gen_lock));
2884 
2885 	if (igb_is_link_up(igb)) {
2886 		/*
2887 		 * The Link is up, check whether it was marked as down earlier
2888 		 */
2889 		if (igb->link_state != LINK_STATE_UP) {
2890 			(void) e1000_get_speed_and_duplex(hw, &speed, &duplex);
2891 			igb->link_speed = speed;
2892 			igb->link_duplex = duplex;
2893 			igb->link_state = LINK_STATE_UP;
2894 			igb->link_down_timeout = 0;
2895 			link_changed = B_TRUE;
2896 		}
2897 	} else {
2898 		if (igb->link_state != LINK_STATE_DOWN) {
2899 			igb->link_speed = 0;
2900 			igb->link_duplex = 0;
2901 			igb->link_state = LINK_STATE_DOWN;
2902 			link_changed = B_TRUE;
2903 		}
2904 
2905 		if (igb->igb_state & IGB_STARTED) {
2906 			if (igb->link_down_timeout < MAX_LINK_DOWN_TIMEOUT) {
2907 				igb->link_down_timeout++;
2908 			} else if (igb->link_down_timeout ==
2909 			    MAX_LINK_DOWN_TIMEOUT) {
2910 				igb_tx_clean(igb);
2911 				igb->link_down_timeout++;
2912 			}
2913 		}
2914 	}
2915 
2916 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2917 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2918 
2919 	return (link_changed);
2920 }
2921 
2922 /*
2923  * igb_local_timer - driver watchdog function
2924  *
2925  * This function will handle the transmit stall check, link status check and
2926  * other routines.
2927  */
2928 static void
2929 igb_local_timer(void *arg)
2930 {
2931 	igb_t *igb = (igb_t *)arg;
2932 	boolean_t link_changed = B_FALSE;
2933 
2934 	if (igb_stall_check(igb)) {
2935 		igb_fm_ereport(igb, DDI_FM_DEVICE_STALL);
2936 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
2937 		igb->reset_count++;
2938 		if (igb_reset(igb) == IGB_SUCCESS)
2939 			ddi_fm_service_impact(igb->dip,
2940 			    DDI_SERVICE_RESTORED);
2941 	}
2942 
2943 	mutex_enter(&igb->gen_lock);
2944 	if (!(igb->igb_state & IGB_SUSPENDED) && (igb->igb_state & IGB_STARTED))
2945 		link_changed = igb_link_check(igb);
2946 	mutex_exit(&igb->gen_lock);
2947 
2948 	if (link_changed)
2949 		mac_link_update(igb->mac_hdl, igb->link_state);
2950 
2951 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2952 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2953 
2954 	igb_restart_watchdog_timer(igb);
2955 }
2956 
2957 /*
2958  * igb_stall_check - check for transmit stall
2959  *
2960  * This function checks if the adapter is stalled (in transmit).
2961  *
2962  * It is called each time the watchdog timeout is invoked.
2963  * If the transmit descriptor reclaim continuously fails,
2964  * the watchdog value will increment by 1. If the watchdog
2965  * value exceeds the threshold, the igb is assumed to
2966  * have stalled and need to be reset.
2967  */
2968 static boolean_t
2969 igb_stall_check(igb_t *igb)
2970 {
2971 	igb_tx_ring_t *tx_ring;
2972 	boolean_t result;
2973 	int i;
2974 
2975 	if (igb->link_state != LINK_STATE_UP)
2976 		return (B_FALSE);
2977 
2978 	/*
2979 	 * If any tx ring is stalled, we'll reset the chipset
2980 	 */
2981 	result = B_FALSE;
2982 	for (i = 0; i < igb->num_tx_rings; i++) {
2983 		tx_ring = &igb->tx_rings[i];
2984 
2985 		if (tx_ring->recycle_fail > 0)
2986 			tx_ring->stall_watchdog++;
2987 		else
2988 			tx_ring->stall_watchdog = 0;
2989 
2990 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
2991 			result = B_TRUE;
2992 			break;
2993 		}
2994 	}
2995 
2996 	if (result) {
2997 		tx_ring->stall_watchdog = 0;
2998 		tx_ring->recycle_fail = 0;
2999 	}
3000 
3001 	return (result);
3002 }
3003 
3004 
3005 /*
3006  * is_valid_mac_addr - Check if the mac address is valid
3007  */
3008 static boolean_t
3009 is_valid_mac_addr(uint8_t *mac_addr)
3010 {
3011 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
3012 	const uint8_t addr_test2[6] =
3013 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
3014 
3015 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
3016 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
3017 		return (B_FALSE);
3018 
3019 	return (B_TRUE);
3020 }
3021 
3022 static boolean_t
3023 igb_find_mac_address(igb_t *igb)
3024 {
3025 	struct e1000_hw *hw = &igb->hw;
3026 #ifdef __sparc
3027 	uchar_t *bytes;
3028 	struct ether_addr sysaddr;
3029 	uint_t nelts;
3030 	int err;
3031 	boolean_t found = B_FALSE;
3032 
3033 	/*
3034 	 * The "vendor's factory-set address" may already have
3035 	 * been extracted from the chip, but if the property
3036 	 * "local-mac-address" is set we use that instead.
3037 	 *
3038 	 * We check whether it looks like an array of 6
3039 	 * bytes (which it should, if OBP set it).  If we can't
3040 	 * make sense of it this way, we'll ignore it.
3041 	 */
3042 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3043 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
3044 	if (err == DDI_PROP_SUCCESS) {
3045 		if (nelts == ETHERADDRL) {
3046 			while (nelts--)
3047 				hw->mac.addr[nelts] = bytes[nelts];
3048 			found = B_TRUE;
3049 		}
3050 		ddi_prop_free(bytes);
3051 	}
3052 
3053 	/*
3054 	 * Look up the OBP property "local-mac-address?". If the user has set
3055 	 * 'local-mac-address? = false', use "the system address" instead.
3056 	 */
3057 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip, 0,
3058 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
3059 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
3060 			if (localetheraddr(NULL, &sysaddr) != 0) {
3061 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
3062 				found = B_TRUE;
3063 			}
3064 		}
3065 		ddi_prop_free(bytes);
3066 	}
3067 
3068 	/*
3069 	 * Finally(!), if there's a valid "mac-address" property (created
3070 	 * if we netbooted from this interface), we must use this instead
3071 	 * of any of the above to ensure that the NFS/install server doesn't
3072 	 * get confused by the address changing as Solaris takes over!
3073 	 */
3074 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3075 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
3076 	if (err == DDI_PROP_SUCCESS) {
3077 		if (nelts == ETHERADDRL) {
3078 			while (nelts--)
3079 				hw->mac.addr[nelts] = bytes[nelts];
3080 			found = B_TRUE;
3081 		}
3082 		ddi_prop_free(bytes);
3083 	}
3084 
3085 	if (found) {
3086 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
3087 		return (B_TRUE);
3088 	}
3089 #endif
3090 
3091 	/*
3092 	 * Read the device MAC address from the EEPROM
3093 	 */
3094 	if (e1000_read_mac_addr(hw) != E1000_SUCCESS)
3095 		return (B_FALSE);
3096 
3097 	return (B_TRUE);
3098 }
3099 
3100 #pragma inline(igb_arm_watchdog_timer)
3101 
3102 static void
3103 igb_arm_watchdog_timer(igb_t *igb)
3104 {
3105 	/*
3106 	 * Fire a watchdog timer
3107 	 */
3108 	igb->watchdog_tid =
3109 	    timeout(igb_local_timer,
3110 	    (void *)igb, 1 * drv_usectohz(1000000));
3111 
3112 }
3113 
3114 /*
3115  * igb_enable_watchdog_timer - Enable and start the driver watchdog timer
3116  */
3117 void
3118 igb_enable_watchdog_timer(igb_t *igb)
3119 {
3120 	mutex_enter(&igb->watchdog_lock);
3121 
3122 	if (!igb->watchdog_enable) {
3123 		igb->watchdog_enable = B_TRUE;
3124 		igb->watchdog_start = B_TRUE;
3125 		igb_arm_watchdog_timer(igb);
3126 	}
3127 
3128 	mutex_exit(&igb->watchdog_lock);
3129 
3130 }
3131 
3132 /*
3133  * igb_disable_watchdog_timer - Disable and stop the driver watchdog timer
3134  */
3135 void
3136 igb_disable_watchdog_timer(igb_t *igb)
3137 {
3138 	timeout_id_t tid;
3139 
3140 	mutex_enter(&igb->watchdog_lock);
3141 
3142 	igb->watchdog_enable = B_FALSE;
3143 	igb->watchdog_start = B_FALSE;
3144 	tid = igb->watchdog_tid;
3145 	igb->watchdog_tid = 0;
3146 
3147 	mutex_exit(&igb->watchdog_lock);
3148 
3149 	if (tid != 0)
3150 		(void) untimeout(tid);
3151 
3152 }
3153 
3154 /*
3155  * igb_start_watchdog_timer - Start the driver watchdog timer
3156  */
3157 static void
3158 igb_start_watchdog_timer(igb_t *igb)
3159 {
3160 	mutex_enter(&igb->watchdog_lock);
3161 
3162 	if (igb->watchdog_enable) {
3163 		if (!igb->watchdog_start) {
3164 			igb->watchdog_start = B_TRUE;
3165 			igb_arm_watchdog_timer(igb);
3166 		}
3167 	}
3168 
3169 	mutex_exit(&igb->watchdog_lock);
3170 }
3171 
3172 /*
3173  * igb_restart_watchdog_timer - Restart the driver watchdog timer
3174  */
3175 static void
3176 igb_restart_watchdog_timer(igb_t *igb)
3177 {
3178 	mutex_enter(&igb->watchdog_lock);
3179 
3180 	if (igb->watchdog_start)
3181 		igb_arm_watchdog_timer(igb);
3182 
3183 	mutex_exit(&igb->watchdog_lock);
3184 }
3185 
3186 /*
3187  * igb_stop_watchdog_timer - Stop the driver watchdog timer
3188  */
3189 static void
3190 igb_stop_watchdog_timer(igb_t *igb)
3191 {
3192 	timeout_id_t tid;
3193 
3194 	mutex_enter(&igb->watchdog_lock);
3195 
3196 	igb->watchdog_start = B_FALSE;
3197 	tid = igb->watchdog_tid;
3198 	igb->watchdog_tid = 0;
3199 
3200 	mutex_exit(&igb->watchdog_lock);
3201 
3202 	if (tid != 0)
3203 		(void) untimeout(tid);
3204 }
3205 
3206 /*
3207  * igb_disable_adapter_interrupts - Clear/disable all hardware interrupts
3208  */
3209 static void
3210 igb_disable_adapter_interrupts(igb_t *igb)
3211 {
3212 	struct e1000_hw *hw = &igb->hw;
3213 
3214 	/*
3215 	 * Set the IMC register to mask all the interrupts,
3216 	 * including the tx interrupts.
3217 	 */
3218 	E1000_WRITE_REG(hw, E1000_IMC, ~0);
3219 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3220 
3221 	/*
3222 	 * Additional disabling for MSI-X
3223 	 */
3224 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3225 		E1000_WRITE_REG(hw, E1000_EIMC, ~0);
3226 		E1000_WRITE_REG(hw, E1000_EIAC, 0);
3227 		E1000_WRITE_REG(hw, E1000_EIAM, 0);
3228 	}
3229 
3230 	E1000_WRITE_FLUSH(hw);
3231 }
3232 
3233 /*
3234  * igb_enable_adapter_interrupts_82576 - Enable NIC interrupts for 82576
3235  */
3236 static void
3237 igb_enable_adapter_interrupts_82576(igb_t *igb)
3238 {
3239 	struct e1000_hw *hw = &igb->hw;
3240 
3241 	/* Clear any pending interrupts */
3242 	(void) E1000_READ_REG(hw, E1000_ICR);
3243 
3244 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3245 
3246 		/* Interrupt enabling for MSI-X */
3247 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3248 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3249 		igb->ims_mask = E1000_IMS_LSC;
3250 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3251 	} else {
3252 		/* Interrupt enabling for MSI and legacy */
3253 		E1000_WRITE_REG(hw, E1000_IVAR0, E1000_IVAR_VALID);
3254 		igb->ims_mask = IMS_ENABLE_MASK | E1000_IMS_TXQE;
3255 		E1000_WRITE_REG(hw, E1000_IMS,
3256 		    (IMS_ENABLE_MASK | E1000_IMS_TXQE));
3257 	}
3258 
3259 	/* Disable auto-mask for ICR interrupt bits */
3260 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3261 
3262 	E1000_WRITE_FLUSH(hw);
3263 }
3264 
3265 /*
3266  * igb_enable_adapter_interrupts_82575 - Enable NIC interrupts for 82575
3267  */
3268 static void
3269 igb_enable_adapter_interrupts_82575(igb_t *igb)
3270 {
3271 	struct e1000_hw *hw = &igb->hw;
3272 	uint32_t reg;
3273 
3274 	/* Clear any pending interrupts */
3275 	(void) E1000_READ_REG(hw, E1000_ICR);
3276 
3277 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3278 		/* Interrupt enabling for MSI-X */
3279 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3280 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3281 		igb->ims_mask = E1000_IMS_LSC;
3282 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3283 
3284 		/* Enable MSI-X PBA support */
3285 		reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
3286 		reg |= E1000_CTRL_EXT_PBA_CLR;
3287 
3288 		/* Non-selective interrupt clear-on-read */
3289 		reg |= E1000_CTRL_EXT_IRCA;	/* Called NSICR in the EAS */
3290 
3291 		E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
3292 	} else {
3293 		/* Interrupt enabling for MSI and legacy */
3294 		igb->ims_mask = IMS_ENABLE_MASK;
3295 		E1000_WRITE_REG(hw, E1000_IMS, IMS_ENABLE_MASK);
3296 	}
3297 
3298 	E1000_WRITE_FLUSH(hw);
3299 }
3300 
3301 /*
3302  * Loopback Support
3303  */
3304 static lb_property_t lb_normal =
3305 	{ normal,	"normal",	IGB_LB_NONE		};
3306 static lb_property_t lb_external =
3307 	{ external,	"External",	IGB_LB_EXTERNAL		};
3308 static lb_property_t lb_mac =
3309 	{ internal,	"MAC",		IGB_LB_INTERNAL_MAC	};
3310 static lb_property_t lb_phy =
3311 	{ internal,	"PHY",		IGB_LB_INTERNAL_PHY	};
3312 static lb_property_t lb_serdes =
3313 	{ internal,	"SerDes",	IGB_LB_INTERNAL_SERDES	};
3314 
3315 enum ioc_reply
3316 igb_loopback_ioctl(igb_t *igb, struct iocblk *iocp, mblk_t *mp)
3317 {
3318 	lb_info_sz_t *lbsp;
3319 	lb_property_t *lbpp;
3320 	struct e1000_hw *hw;
3321 	uint32_t *lbmp;
3322 	uint32_t size;
3323 	uint32_t value;
3324 
3325 	hw = &igb->hw;
3326 
3327 	if (mp->b_cont == NULL)
3328 		return (IOC_INVAL);
3329 
3330 	switch (iocp->ioc_cmd) {
3331 	default:
3332 		return (IOC_INVAL);
3333 
3334 	case LB_GET_INFO_SIZE:
3335 		size = sizeof (lb_info_sz_t);
3336 		if (iocp->ioc_count != size)
3337 			return (IOC_INVAL);
3338 
3339 		value = sizeof (lb_normal);
3340 		value += sizeof (lb_mac);
3341 		if (hw->phy.media_type == e1000_media_type_copper)
3342 			value += sizeof (lb_phy);
3343 		else
3344 			value += sizeof (lb_serdes);
3345 		value += sizeof (lb_external);
3346 
3347 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
3348 		*lbsp = value;
3349 		break;
3350 
3351 	case LB_GET_INFO:
3352 		value = sizeof (lb_normal);
3353 		value += sizeof (lb_mac);
3354 		if (hw->phy.media_type == e1000_media_type_copper)
3355 			value += sizeof (lb_phy);
3356 		else
3357 			value += sizeof (lb_serdes);
3358 		value += sizeof (lb_external);
3359 
3360 		size = value;
3361 		if (iocp->ioc_count != size)
3362 			return (IOC_INVAL);
3363 
3364 		value = 0;
3365 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
3366 
3367 		lbpp[value++] = lb_normal;
3368 		lbpp[value++] = lb_mac;
3369 		if (hw->phy.media_type == e1000_media_type_copper)
3370 			lbpp[value++] = lb_phy;
3371 		else
3372 			lbpp[value++] = lb_serdes;
3373 		lbpp[value++] = lb_external;
3374 		break;
3375 
3376 	case LB_GET_MODE:
3377 		size = sizeof (uint32_t);
3378 		if (iocp->ioc_count != size)
3379 			return (IOC_INVAL);
3380 
3381 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3382 		*lbmp = igb->loopback_mode;
3383 		break;
3384 
3385 	case LB_SET_MODE:
3386 		size = 0;
3387 		if (iocp->ioc_count != sizeof (uint32_t))
3388 			return (IOC_INVAL);
3389 
3390 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3391 		if (!igb_set_loopback_mode(igb, *lbmp))
3392 			return (IOC_INVAL);
3393 		break;
3394 	}
3395 
3396 	iocp->ioc_count = size;
3397 	iocp->ioc_error = 0;
3398 
3399 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3400 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3401 		return (IOC_INVAL);
3402 	}
3403 
3404 	return (IOC_REPLY);
3405 }
3406 
3407 /*
3408  * igb_set_loopback_mode - Setup loopback based on the loopback mode
3409  */
3410 static boolean_t
3411 igb_set_loopback_mode(igb_t *igb, uint32_t mode)
3412 {
3413 	struct e1000_hw *hw;
3414 
3415 	if (mode == igb->loopback_mode)
3416 		return (B_TRUE);
3417 
3418 	hw = &igb->hw;
3419 
3420 	igb->loopback_mode = mode;
3421 
3422 	if (mode == IGB_LB_NONE) {
3423 		/* Reset the chip */
3424 		hw->phy.autoneg_wait_to_complete = B_TRUE;
3425 		(void) igb_reset(igb);
3426 		hw->phy.autoneg_wait_to_complete = B_FALSE;
3427 		return (B_TRUE);
3428 	}
3429 
3430 	mutex_enter(&igb->gen_lock);
3431 
3432 	switch (mode) {
3433 	default:
3434 		mutex_exit(&igb->gen_lock);
3435 		return (B_FALSE);
3436 
3437 	case IGB_LB_EXTERNAL:
3438 		igb_set_external_loopback(igb);
3439 		break;
3440 
3441 	case IGB_LB_INTERNAL_MAC:
3442 		igb_set_internal_mac_loopback(igb);
3443 		break;
3444 
3445 	case IGB_LB_INTERNAL_PHY:
3446 		igb_set_internal_phy_loopback(igb);
3447 		break;
3448 
3449 	case IGB_LB_INTERNAL_SERDES:
3450 		igb_set_internal_serdes_loopback(igb);
3451 		break;
3452 	}
3453 
3454 	mutex_exit(&igb->gen_lock);
3455 
3456 	return (B_TRUE);
3457 }
3458 
3459 /*
3460  * igb_set_external_loopback - Set the external loopback mode
3461  */
3462 static void
3463 igb_set_external_loopback(igb_t *igb)
3464 {
3465 	struct e1000_hw *hw;
3466 
3467 	hw = &igb->hw;
3468 
3469 	/* Set phy to known state */
3470 	(void) e1000_phy_hw_reset(hw);
3471 
3472 	(void) e1000_write_phy_reg(hw, 0x0, 0x0140);
3473 	(void) e1000_write_phy_reg(hw, 0x9, 0x1b00);
3474 	(void) e1000_write_phy_reg(hw, 0x12, 0x1610);
3475 	(void) e1000_write_phy_reg(hw, 0x1f37, 0x3f1c);
3476 }
3477 
3478 /*
3479  * igb_set_internal_mac_loopback - Set the internal MAC loopback mode
3480  */
3481 static void
3482 igb_set_internal_mac_loopback(igb_t *igb)
3483 {
3484 	struct e1000_hw *hw;
3485 	uint32_t ctrl;
3486 	uint32_t rctl;
3487 	uint32_t ctrl_ext;
3488 	uint16_t phy_ctrl;
3489 	uint16_t phy_status;
3490 
3491 	hw = &igb->hw;
3492 
3493 	(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
3494 	phy_ctrl &= ~MII_CR_AUTO_NEG_EN;
3495 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3496 
3497 	(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
3498 
3499 	/* Set link mode to PHY (00b) in the Extended Control register */
3500 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3501 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3502 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3503 
3504 	/* Set the Device Control register */
3505 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3506 	if (!(phy_status & MII_SR_LINK_STATUS))
3507 		ctrl |= E1000_CTRL_ILOS; /* Set ILOS when the link is down */
3508 	ctrl &= ~E1000_CTRL_SPD_SEL;	/* Clear the speed sel bits */
3509 	ctrl |= (E1000_CTRL_SLU |	/* Force link up */
3510 	    E1000_CTRL_FRCSPD |		/* Force speed */
3511 	    E1000_CTRL_FRCDPX |		/* Force duplex */
3512 	    E1000_CTRL_SPD_1000 |	/* Force speed to 1000 */
3513 	    E1000_CTRL_FD);		/* Force full duplex */
3514 
3515 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3516 
3517 	/* Set the Receive Control register */
3518 	rctl = E1000_READ_REG(hw, E1000_RCTL);
3519 	rctl &= ~E1000_RCTL_LBM_TCVR;
3520 	rctl |= E1000_RCTL_LBM_MAC;
3521 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
3522 }
3523 
3524 /*
3525  * igb_set_internal_phy_loopback - Set the internal PHY loopback mode
3526  */
3527 static void
3528 igb_set_internal_phy_loopback(igb_t *igb)
3529 {
3530 	struct e1000_hw *hw;
3531 	uint32_t ctrl_ext;
3532 	uint16_t phy_ctrl;
3533 	uint16_t phy_pconf;
3534 
3535 	hw = &igb->hw;
3536 
3537 	/* Set link mode to PHY (00b) in the Extended Control register */
3538 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3539 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3540 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3541 
3542 	/*
3543 	 * Set PHY control register (0x4140):
3544 	 *    Set full duplex mode
3545 	 *    Set loopback bit
3546 	 *    Clear auto-neg enable bit
3547 	 *    Set PHY speed
3548 	 */
3549 	phy_ctrl = MII_CR_FULL_DUPLEX | MII_CR_SPEED_1000 | MII_CR_LOOPBACK;
3550 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3551 
3552 	/* Set the link disable bit in the Port Configuration register */
3553 	(void) e1000_read_phy_reg(hw, 0x10, &phy_pconf);
3554 	phy_pconf |= (uint16_t)1 << 14;
3555 	(void) e1000_write_phy_reg(hw, 0x10, phy_pconf);
3556 }
3557 
3558 /*
3559  * igb_set_internal_serdes_loopback - Set the internal SerDes loopback mode
3560  */
3561 static void
3562 igb_set_internal_serdes_loopback(igb_t *igb)
3563 {
3564 	struct e1000_hw *hw;
3565 	uint32_t ctrl_ext;
3566 	uint32_t ctrl;
3567 	uint32_t pcs_lctl;
3568 	uint32_t connsw;
3569 
3570 	hw = &igb->hw;
3571 
3572 	/* Set link mode to SerDes (11b) in the Extended Control register */
3573 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3574 	ctrl_ext |= E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
3575 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3576 
3577 	/* Configure the SerDes to loopback */
3578 	E1000_WRITE_REG(hw, E1000_SCTL, 0x410);
3579 
3580 	/* Set Device Control register */
3581 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3582 	ctrl |= (E1000_CTRL_FD |	/* Force full duplex */
3583 	    E1000_CTRL_SLU);		/* Force link up */
3584 	ctrl &= ~(E1000_CTRL_RFCE |	/* Disable receive flow control */
3585 	    E1000_CTRL_TFCE |		/* Disable transmit flow control */
3586 	    E1000_CTRL_LRST);		/* Clear link reset */
3587 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3588 
3589 	/* Set PCS Link Control register */
3590 	pcs_lctl = E1000_READ_REG(hw, E1000_PCS_LCTL);
3591 	pcs_lctl |= (E1000_PCS_LCTL_FORCE_LINK |
3592 	    E1000_PCS_LCTL_FSD |
3593 	    E1000_PCS_LCTL_FDV_FULL |
3594 	    E1000_PCS_LCTL_FLV_LINK_UP);
3595 	pcs_lctl &= ~E1000_PCS_LCTL_AN_ENABLE;
3596 	E1000_WRITE_REG(hw, E1000_PCS_LCTL, pcs_lctl);
3597 
3598 	/* Set the Copper/Fiber Switch Control - CONNSW register */
3599 	connsw = E1000_READ_REG(hw, E1000_CONNSW);
3600 	connsw &= ~E1000_CONNSW_ENRGSRC;
3601 	E1000_WRITE_REG(hw, E1000_CONNSW, connsw);
3602 }
3603 
3604 #pragma inline(igb_intr_rx_work)
3605 /*
3606  * igb_intr_rx_work - rx processing of ISR
3607  */
3608 static void
3609 igb_intr_rx_work(igb_rx_ring_t *rx_ring)
3610 {
3611 	mblk_t *mp;
3612 
3613 	mutex_enter(&rx_ring->rx_lock);
3614 	mp = igb_rx(rx_ring, IGB_NO_POLL);
3615 	mutex_exit(&rx_ring->rx_lock);
3616 
3617 	if (mp != NULL)
3618 		mac_rx_ring(rx_ring->igb->mac_hdl, rx_ring->ring_handle, mp,
3619 		    rx_ring->ring_gen_num);
3620 }
3621 
3622 #pragma inline(igb_intr_tx_work)
3623 /*
3624  * igb_intr_tx_work - tx processing of ISR
3625  */
3626 static void
3627 igb_intr_tx_work(igb_tx_ring_t *tx_ring)
3628 {
3629 	/* Recycle the tx descriptors */
3630 	tx_ring->tx_recycle(tx_ring);
3631 
3632 	/* Schedule the re-transmit */
3633 	if (tx_ring->reschedule &&
3634 	    (tx_ring->tbd_free >= tx_ring->resched_thresh)) {
3635 		tx_ring->reschedule = B_FALSE;
3636 		mac_tx_ring_update(tx_ring->igb->mac_hdl, tx_ring->ring_handle);
3637 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3638 	}
3639 }
3640 
3641 #pragma inline(igb_intr_link_work)
3642 /*
3643  * igb_intr_link_work - link-status-change processing of ISR
3644  */
3645 static void
3646 igb_intr_link_work(igb_t *igb)
3647 {
3648 	boolean_t link_changed;
3649 
3650 	igb_stop_watchdog_timer(igb);
3651 
3652 	mutex_enter(&igb->gen_lock);
3653 
3654 	/*
3655 	 * Because we got a link-status-change interrupt, force
3656 	 * e1000_check_for_link() to look at phy
3657 	 */
3658 	igb->hw.mac.get_link_status = B_TRUE;
3659 
3660 	/* igb_link_check takes care of link status change */
3661 	link_changed = igb_link_check(igb);
3662 
3663 	/* Get new phy state */
3664 	igb_get_phy_state(igb);
3665 
3666 	mutex_exit(&igb->gen_lock);
3667 
3668 	if (link_changed)
3669 		mac_link_update(igb->mac_hdl, igb->link_state);
3670 
3671 	igb_start_watchdog_timer(igb);
3672 }
3673 
3674 /*
3675  * igb_intr_legacy - Interrupt handler for legacy interrupts
3676  */
3677 static uint_t
3678 igb_intr_legacy(void *arg1, void *arg2)
3679 {
3680 	igb_t *igb = (igb_t *)arg1;
3681 	igb_tx_ring_t *tx_ring;
3682 	uint32_t icr;
3683 	mblk_t *mp;
3684 	boolean_t tx_reschedule;
3685 	boolean_t link_changed;
3686 	uint_t result;
3687 
3688 	_NOTE(ARGUNUSED(arg2));
3689 
3690 	mutex_enter(&igb->gen_lock);
3691 
3692 	if (igb->igb_state & IGB_SUSPENDED) {
3693 		mutex_exit(&igb->gen_lock);
3694 		return (DDI_INTR_UNCLAIMED);
3695 	}
3696 
3697 	mp = NULL;
3698 	tx_reschedule = B_FALSE;
3699 	link_changed = B_FALSE;
3700 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3701 
3702 	if (icr & E1000_ICR_INT_ASSERTED) {
3703 		/*
3704 		 * E1000_ICR_INT_ASSERTED bit was set:
3705 		 * Read(Clear) the ICR, claim this interrupt,
3706 		 * look for work to do.
3707 		 */
3708 		ASSERT(igb->num_rx_rings == 1);
3709 		ASSERT(igb->num_tx_rings == 1);
3710 
3711 		/* Make sure all interrupt causes cleared */
3712 		(void) E1000_READ_REG(&igb->hw, E1000_EICR);
3713 
3714 		if (icr & E1000_ICR_RXT0) {
3715 			mp = igb_rx(&igb->rx_rings[0], IGB_NO_POLL);
3716 		}
3717 
3718 		if (icr & E1000_ICR_TXDW) {
3719 			tx_ring = &igb->tx_rings[0];
3720 
3721 			/* Recycle the tx descriptors */
3722 			tx_ring->tx_recycle(tx_ring);
3723 
3724 			/* Schedule the re-transmit */
3725 			tx_reschedule = (tx_ring->reschedule &&
3726 			    (tx_ring->tbd_free >= tx_ring->resched_thresh));
3727 		}
3728 
3729 		if (icr & E1000_ICR_LSC) {
3730 			/*
3731 			 * Because we got a link-status-change interrupt, force
3732 			 * e1000_check_for_link() to look at phy
3733 			 */
3734 			igb->hw.mac.get_link_status = B_TRUE;
3735 
3736 			/* igb_link_check takes care of link status change */
3737 			link_changed = igb_link_check(igb);
3738 
3739 			/* Get new phy state */
3740 			igb_get_phy_state(igb);
3741 		}
3742 
3743 		result = DDI_INTR_CLAIMED;
3744 	} else {
3745 		/*
3746 		 * E1000_ICR_INT_ASSERTED bit was not set:
3747 		 * Don't claim this interrupt.
3748 		 */
3749 		result = DDI_INTR_UNCLAIMED;
3750 	}
3751 
3752 	mutex_exit(&igb->gen_lock);
3753 
3754 	/*
3755 	 * Do the following work outside of the gen_lock
3756 	 */
3757 	if (mp != NULL)
3758 		mac_rx(igb->mac_hdl, NULL, mp);
3759 
3760 	if (tx_reschedule)  {
3761 		tx_ring->reschedule = B_FALSE;
3762 		mac_tx_ring_update(igb->mac_hdl, tx_ring->ring_handle);
3763 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3764 	}
3765 
3766 	if (link_changed)
3767 		mac_link_update(igb->mac_hdl, igb->link_state);
3768 
3769 	return (result);
3770 }
3771 
3772 /*
3773  * igb_intr_msi - Interrupt handler for MSI
3774  */
3775 static uint_t
3776 igb_intr_msi(void *arg1, void *arg2)
3777 {
3778 	igb_t *igb = (igb_t *)arg1;
3779 	uint32_t icr;
3780 
3781 	_NOTE(ARGUNUSED(arg2));
3782 
3783 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3784 
3785 	/* Make sure all interrupt causes cleared */
3786 	(void) E1000_READ_REG(&igb->hw, E1000_EICR);
3787 
3788 	/*
3789 	 * For MSI interrupt, we have only one vector,
3790 	 * so we have only one rx ring and one tx ring enabled.
3791 	 */
3792 	ASSERT(igb->num_rx_rings == 1);
3793 	ASSERT(igb->num_tx_rings == 1);
3794 
3795 	if (icr & E1000_ICR_RXT0) {
3796 		igb_intr_rx_work(&igb->rx_rings[0]);
3797 	}
3798 
3799 	if (icr & E1000_ICR_TXDW) {
3800 		igb_intr_tx_work(&igb->tx_rings[0]);
3801 	}
3802 
3803 	if (icr & E1000_ICR_LSC) {
3804 		igb_intr_link_work(igb);
3805 	}
3806 
3807 	return (DDI_INTR_CLAIMED);
3808 }
3809 
3810 /*
3811  * igb_intr_rx - Interrupt handler for rx
3812  */
3813 static uint_t
3814 igb_intr_rx(void *arg1, void *arg2)
3815 {
3816 	igb_rx_ring_t *rx_ring = (igb_rx_ring_t *)arg1;
3817 
3818 	_NOTE(ARGUNUSED(arg2));
3819 
3820 	/*
3821 	 * Only used via MSI-X vector so don't check cause bits
3822 	 * and only clean the given ring.
3823 	 */
3824 	igb_intr_rx_work(rx_ring);
3825 
3826 	return (DDI_INTR_CLAIMED);
3827 }
3828 
3829 /*
3830  * igb_intr_tx - Interrupt handler for tx
3831  */
3832 static uint_t
3833 igb_intr_tx(void *arg1, void *arg2)
3834 {
3835 	igb_tx_ring_t *tx_ring = (igb_tx_ring_t *)arg1;
3836 
3837 	_NOTE(ARGUNUSED(arg2));
3838 
3839 	/*
3840 	 * Only used via MSI-X vector so don't check cause bits
3841 	 * and only clean the given ring.
3842 	 */
3843 	igb_intr_tx_work(tx_ring);
3844 
3845 	return (DDI_INTR_CLAIMED);
3846 }
3847 
3848 /*
3849  * igb_intr_tx_other - Interrupt handler for both tx and other
3850  *
3851  */
3852 static uint_t
3853 igb_intr_tx_other(void *arg1, void *arg2)
3854 {
3855 	igb_t *igb = (igb_t *)arg1;
3856 	uint32_t icr;
3857 
3858 	_NOTE(ARGUNUSED(arg2));
3859 
3860 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3861 
3862 	/*
3863 	 * Look for tx reclaiming work first. Remember, in the
3864 	 * case of only interrupt sharing, only one tx ring is
3865 	 * used
3866 	 */
3867 	igb_intr_tx_work(&igb->tx_rings[0]);
3868 
3869 	/*
3870 	 * Check for "other" causes.
3871 	 */
3872 	if (icr & E1000_ICR_LSC) {
3873 		igb_intr_link_work(igb);
3874 	}
3875 
3876 	/*
3877 	 * The DOUTSYNC bit indicates a tx packet dropped because
3878 	 * DMA engine gets "out of sync". There isn't a real fix
3879 	 * for this. The Intel recommendation is to count the number
3880 	 * of occurrences so user can detect when it is happening.
3881 	 * The issue is non-fatal and there's no recovery action
3882 	 * available.
3883 	 */
3884 	if (icr & E1000_ICR_DOUTSYNC) {
3885 		IGB_STAT(igb->dout_sync);
3886 	}
3887 
3888 	return (DDI_INTR_CLAIMED);
3889 }
3890 
3891 /*
3892  * igb_alloc_intrs - Allocate interrupts for the driver
3893  *
3894  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
3895  * if not successful, try Legacy.
3896  * igb->intr_force can be used to force sequence to start with
3897  * any of the 3 types.
3898  * If MSI-X is not used, number of tx/rx rings is forced to 1.
3899  */
3900 static int
3901 igb_alloc_intrs(igb_t *igb)
3902 {
3903 	dev_info_t *devinfo;
3904 	int intr_types;
3905 	int rc;
3906 
3907 	devinfo = igb->dip;
3908 
3909 	/* Get supported interrupt types */
3910 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
3911 
3912 	if (rc != DDI_SUCCESS) {
3913 		igb_log(igb,
3914 		    "Get supported interrupt types failed: %d", rc);
3915 		return (IGB_FAILURE);
3916 	}
3917 	IGB_DEBUGLOG_1(igb, "Supported interrupt types: %x", intr_types);
3918 
3919 	igb->intr_type = 0;
3920 
3921 	/* Install MSI-X interrupts */
3922 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
3923 	    (igb->intr_force <= IGB_INTR_MSIX)) {
3924 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSIX);
3925 
3926 		if (rc == IGB_SUCCESS)
3927 			return (IGB_SUCCESS);
3928 
3929 		igb_log(igb,
3930 		    "Allocate MSI-X failed, trying MSI interrupts...");
3931 	}
3932 
3933 	/* MSI-X not used, force rings to 1 */
3934 	igb->num_rx_rings = 1;
3935 	igb->num_tx_rings = 1;
3936 	igb_log(igb,
3937 	    "MSI-X not used, force rx and tx queue number to 1");
3938 
3939 	/* Install MSI interrupts */
3940 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
3941 	    (igb->intr_force <= IGB_INTR_MSI)) {
3942 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSI);
3943 
3944 		if (rc == IGB_SUCCESS)
3945 			return (IGB_SUCCESS);
3946 
3947 		igb_log(igb,
3948 		    "Allocate MSI failed, trying Legacy interrupts...");
3949 	}
3950 
3951 	/* Install legacy interrupts */
3952 	if (intr_types & DDI_INTR_TYPE_FIXED) {
3953 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_FIXED);
3954 
3955 		if (rc == IGB_SUCCESS)
3956 			return (IGB_SUCCESS);
3957 
3958 		igb_log(igb,
3959 		    "Allocate Legacy interrupts failed");
3960 	}
3961 
3962 	/* If none of the 3 types succeeded, return failure */
3963 	return (IGB_FAILURE);
3964 }
3965 
3966 /*
3967  * igb_alloc_intr_handles - Allocate interrupt handles.
3968  *
3969  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
3970  * if fewer than 2 handles are available, return failure.
3971  * Upon success, this sets the number of Rx rings to a number that
3972  * matches the handles available for Rx interrupts.
3973  */
3974 static int
3975 igb_alloc_intr_handles(igb_t *igb, int intr_type)
3976 {
3977 	dev_info_t *devinfo;
3978 	int orig, request, count, avail, actual;
3979 	int diff, minimum;
3980 	int rc;
3981 
3982 	devinfo = igb->dip;
3983 
3984 	switch (intr_type) {
3985 	case DDI_INTR_TYPE_FIXED:
3986 		request = 1;	/* Request 1 legacy interrupt handle */
3987 		minimum = 1;
3988 		IGB_DEBUGLOG_0(igb, "interrupt type: legacy");
3989 		break;
3990 
3991 	case DDI_INTR_TYPE_MSI:
3992 		request = 1;	/* Request 1 MSI interrupt handle */
3993 		minimum = 1;
3994 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI");
3995 		break;
3996 
3997 	case DDI_INTR_TYPE_MSIX:
3998 		/*
3999 		 * Number of vectors for the adapter is
4000 		 * # rx rings + # tx rings
4001 		 * One of tx vectors is for tx & other
4002 		 */
4003 		request = igb->num_rx_rings + igb->num_tx_rings;
4004 		orig = request;
4005 		minimum = 2;
4006 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI-X");
4007 		break;
4008 
4009 	default:
4010 		igb_log(igb,
4011 		    "invalid call to igb_alloc_intr_handles(): %d\n",
4012 		    intr_type);
4013 		return (IGB_FAILURE);
4014 	}
4015 	IGB_DEBUGLOG_2(igb, "interrupt handles requested: %d  minimum: %d",
4016 	    request, minimum);
4017 
4018 	/*
4019 	 * Get number of supported interrupts
4020 	 */
4021 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
4022 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
4023 		igb_log(igb,
4024 		    "Get supported interrupt number failed. "
4025 		    "Return: %d, count: %d", rc, count);
4026 		return (IGB_FAILURE);
4027 	}
4028 	IGB_DEBUGLOG_1(igb, "interrupts supported: %d", count);
4029 
4030 	/*
4031 	 * Get number of available interrupts
4032 	 */
4033 	rc = ddi_intr_get_navail(devinfo, intr_type, &avail);
4034 	if ((rc != DDI_SUCCESS) || (avail < minimum)) {
4035 		igb_log(igb,
4036 		    "Get available interrupt number failed. "
4037 		    "Return: %d, available: %d", rc, avail);
4038 		return (IGB_FAILURE);
4039 	}
4040 	IGB_DEBUGLOG_1(igb, "interrupts available: %d", avail);
4041 
4042 	if (avail < request) {
4043 		igb_log(igb, "Request %d handles, %d available",
4044 		    request, avail);
4045 		request = avail;
4046 	}
4047 
4048 	actual = 0;
4049 	igb->intr_cnt = 0;
4050 
4051 	/*
4052 	 * Allocate an array of interrupt handles
4053 	 */
4054 	igb->intr_size = request * sizeof (ddi_intr_handle_t);
4055 	igb->htable = kmem_alloc(igb->intr_size, KM_SLEEP);
4056 
4057 	rc = ddi_intr_alloc(devinfo, igb->htable, intr_type, 0,
4058 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
4059 	if (rc != DDI_SUCCESS) {
4060 		igb_log(igb, "Allocate interrupts failed. "
4061 		    "return: %d, request: %d, actual: %d",
4062 		    rc, request, actual);
4063 		goto alloc_handle_fail;
4064 	}
4065 	IGB_DEBUGLOG_1(igb, "interrupts actually allocated: %d", actual);
4066 
4067 	igb->intr_cnt = actual;
4068 
4069 	if (actual < minimum) {
4070 		igb_log(igb, "Insufficient interrupt handles allocated: %d",
4071 		    actual);
4072 		goto alloc_handle_fail;
4073 	}
4074 
4075 	/*
4076 	 * For MSI-X, actual might force us to reduce number of tx & rx rings
4077 	 */
4078 	if ((intr_type == DDI_INTR_TYPE_MSIX) && (orig > actual)) {
4079 		diff = orig - actual;
4080 		if (diff < igb->num_tx_rings) {
4081 			igb_log(igb,
4082 			    "MSI-X vectors force Tx queue number to %d",
4083 			    igb->num_tx_rings - diff);
4084 			igb->num_tx_rings -= diff;
4085 		} else {
4086 			igb_log(igb,
4087 			    "MSI-X vectors force Tx queue number to 1");
4088 			igb->num_tx_rings = 1;
4089 
4090 			igb_log(igb,
4091 			    "MSI-X vectors force Rx queue number to %d",
4092 			    actual - 1);
4093 			igb->num_rx_rings = actual - 1;
4094 		}
4095 	}
4096 
4097 	/*
4098 	 * Get priority for first vector, assume remaining are all the same
4099 	 */
4100 	rc = ddi_intr_get_pri(igb->htable[0], &igb->intr_pri);
4101 	if (rc != DDI_SUCCESS) {
4102 		igb_log(igb,
4103 		    "Get interrupt priority failed: %d", rc);
4104 		goto alloc_handle_fail;
4105 	}
4106 
4107 	rc = ddi_intr_get_cap(igb->htable[0], &igb->intr_cap);
4108 	if (rc != DDI_SUCCESS) {
4109 		igb_log(igb,
4110 		    "Get interrupt cap failed: %d", rc);
4111 		goto alloc_handle_fail;
4112 	}
4113 
4114 	igb->intr_type = intr_type;
4115 
4116 	return (IGB_SUCCESS);
4117 
4118 alloc_handle_fail:
4119 	igb_rem_intrs(igb);
4120 
4121 	return (IGB_FAILURE);
4122 }
4123 
4124 /*
4125  * igb_add_intr_handlers - Add interrupt handlers based on the interrupt type
4126  *
4127  * Before adding the interrupt handlers, the interrupt vectors have
4128  * been allocated, and the rx/tx rings have also been allocated.
4129  */
4130 static int
4131 igb_add_intr_handlers(igb_t *igb)
4132 {
4133 	igb_rx_ring_t *rx_ring;
4134 	igb_tx_ring_t *tx_ring;
4135 	int vector;
4136 	int rc;
4137 	int i;
4138 
4139 	vector = 0;
4140 
4141 	switch (igb->intr_type) {
4142 	case DDI_INTR_TYPE_MSIX:
4143 		/* Add interrupt handler for tx + other */
4144 		tx_ring = &igb->tx_rings[0];
4145 		rc = ddi_intr_add_handler(igb->htable[vector],
4146 		    (ddi_intr_handler_t *)igb_intr_tx_other,
4147 		    (void *)igb, NULL);
4148 
4149 		if (rc != DDI_SUCCESS) {
4150 			igb_log(igb,
4151 			    "Add tx/other interrupt handler failed: %d", rc);
4152 			return (IGB_FAILURE);
4153 		}
4154 		tx_ring->intr_vector = vector;
4155 		vector++;
4156 
4157 		/* Add interrupt handler for each rx ring */
4158 		for (i = 0; i < igb->num_rx_rings; i++) {
4159 			rx_ring = &igb->rx_rings[i];
4160 
4161 			rc = ddi_intr_add_handler(igb->htable[vector],
4162 			    (ddi_intr_handler_t *)igb_intr_rx,
4163 			    (void *)rx_ring, NULL);
4164 
4165 			if (rc != DDI_SUCCESS) {
4166 				igb_log(igb,
4167 				    "Add rx interrupt handler failed. "
4168 				    "return: %d, rx ring: %d", rc, i);
4169 				for (vector--; vector >= 0; vector--) {
4170 					(void) ddi_intr_remove_handler(
4171 					    igb->htable[vector]);
4172 				}
4173 				return (IGB_FAILURE);
4174 			}
4175 
4176 			rx_ring->intr_vector = vector;
4177 
4178 			vector++;
4179 		}
4180 
4181 		/* Add interrupt handler for each tx ring from 2nd ring */
4182 		for (i = 1; i < igb->num_tx_rings; i++) {
4183 			tx_ring = &igb->tx_rings[i];
4184 
4185 			rc = ddi_intr_add_handler(igb->htable[vector],
4186 			    (ddi_intr_handler_t *)igb_intr_tx,
4187 			    (void *)tx_ring, NULL);
4188 
4189 			if (rc != DDI_SUCCESS) {
4190 				igb_log(igb,
4191 				    "Add tx interrupt handler failed. "
4192 				    "return: %d, tx ring: %d", rc, i);
4193 				for (vector--; vector >= 0; vector--) {
4194 					(void) ddi_intr_remove_handler(
4195 					    igb->htable[vector]);
4196 				}
4197 				return (IGB_FAILURE);
4198 			}
4199 
4200 			tx_ring->intr_vector = vector;
4201 
4202 			vector++;
4203 		}
4204 
4205 		break;
4206 
4207 	case DDI_INTR_TYPE_MSI:
4208 		/* Add interrupt handlers for the only vector */
4209 		rc = ddi_intr_add_handler(igb->htable[vector],
4210 		    (ddi_intr_handler_t *)igb_intr_msi,
4211 		    (void *)igb, NULL);
4212 
4213 		if (rc != DDI_SUCCESS) {
4214 			igb_log(igb,
4215 			    "Add MSI interrupt handler failed: %d", rc);
4216 			return (IGB_FAILURE);
4217 		}
4218 
4219 		rx_ring = &igb->rx_rings[0];
4220 		rx_ring->intr_vector = vector;
4221 
4222 		vector++;
4223 		break;
4224 
4225 	case DDI_INTR_TYPE_FIXED:
4226 		/* Add interrupt handlers for the only vector */
4227 		rc = ddi_intr_add_handler(igb->htable[vector],
4228 		    (ddi_intr_handler_t *)igb_intr_legacy,
4229 		    (void *)igb, NULL);
4230 
4231 		if (rc != DDI_SUCCESS) {
4232 			igb_log(igb,
4233 			    "Add legacy interrupt handler failed: %d", rc);
4234 			return (IGB_FAILURE);
4235 		}
4236 
4237 		rx_ring = &igb->rx_rings[0];
4238 		rx_ring->intr_vector = vector;
4239 
4240 		vector++;
4241 		break;
4242 
4243 	default:
4244 		return (IGB_FAILURE);
4245 	}
4246 
4247 	ASSERT(vector == igb->intr_cnt);
4248 
4249 	return (IGB_SUCCESS);
4250 }
4251 
4252 /*
4253  * igb_setup_msix_82575 - setup 82575 adapter to use MSI-X interrupts
4254  *
4255  * For each vector enabled on the adapter, Set the MSIXBM register accordingly
4256  */
4257 static void
4258 igb_setup_msix_82575(igb_t *igb)
4259 {
4260 	uint32_t eims = 0;
4261 	int i, vector;
4262 	struct e1000_hw *hw = &igb->hw;
4263 
4264 	/*
4265 	 * Set vector for tx ring 0 and other causes.
4266 	 * NOTE assumption that it is vector 0.
4267 	 */
4268 	vector = 0;
4269 
4270 	igb->eims_mask = E1000_EICR_TX_QUEUE0 | E1000_EICR_OTHER;
4271 	E1000_WRITE_REG(hw, E1000_MSIXBM(vector), igb->eims_mask);
4272 	vector++;
4273 
4274 	for (i = 0; i < igb->num_rx_rings; i++) {
4275 		/*
4276 		 * Set vector for each rx ring
4277 		 */
4278 		eims = (E1000_EICR_RX_QUEUE0 << i);
4279 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4280 
4281 		/*
4282 		 * Accumulate bits to enable in
4283 		 * igb_enable_adapter_interrupts_82575()
4284 		 */
4285 		igb->eims_mask |= eims;
4286 
4287 		vector++;
4288 	}
4289 
4290 	for (i = 1; i < igb->num_tx_rings; i++) {
4291 		/*
4292 		 * Set vector for each tx ring from 2nd tx ring
4293 		 */
4294 		eims = (E1000_EICR_TX_QUEUE0 << i);
4295 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4296 
4297 		/*
4298 		 * Accumulate bits to enable in
4299 		 * igb_enable_adapter_interrupts_82575()
4300 		 */
4301 		igb->eims_mask |= eims;
4302 
4303 		vector++;
4304 	}
4305 
4306 	ASSERT(vector == igb->intr_cnt);
4307 
4308 	/*
4309 	 * Disable IAM for ICR interrupt bits
4310 	 */
4311 	E1000_WRITE_REG(hw, E1000_IAM, 0);
4312 	E1000_WRITE_FLUSH(hw);
4313 }
4314 
4315 /*
4316  * igb_setup_msix_82576 - setup 82576 adapter to use MSI-X interrupts
4317  *
4318  * 82576 uses a table based method for assigning vectors.  Each queue has a
4319  * single entry in the table to which we write a vector number along with a
4320  * "valid" bit.  The entry is a single byte in a 4-byte register.  Vectors
4321  * take a different position in the 4-byte register depending on whether
4322  * they are numbered above or below 8.
4323  */
4324 static void
4325 igb_setup_msix_82576(igb_t *igb)
4326 {
4327 	struct e1000_hw *hw = &igb->hw;
4328 	uint32_t ivar, index, vector;
4329 	int i;
4330 
4331 	/* must enable msi-x capability before IVAR settings */
4332 	E1000_WRITE_REG(hw, E1000_GPIE,
4333 	    (E1000_GPIE_MSIX_MODE | E1000_GPIE_PBA | E1000_GPIE_NSICR));
4334 
4335 	/*
4336 	 * Set vector for tx ring 0 and other causes.
4337 	 * NOTE assumption that it is vector 0.
4338 	 * This is also interdependent with installation of interrupt service
4339 	 * routines in igb_add_intr_handlers().
4340 	 */
4341 
4342 	/* assign "other" causes to vector 0 */
4343 	vector = 0;
4344 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4345 	E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4346 
4347 	/* assign tx ring 0 to vector 0 */
4348 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4349 	E1000_WRITE_REG(hw, E1000_IVAR0, ivar);
4350 
4351 	/* prepare to enable tx & other interrupt causes */
4352 	igb->eims_mask = (1 << vector);
4353 
4354 	vector ++;
4355 	for (i = 0; i < igb->num_rx_rings; i++) {
4356 		/*
4357 		 * Set vector for each rx ring
4358 		 */
4359 		index = (i & 0x7);
4360 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4361 
4362 		if (i < 8) {
4363 			/* vector goes into low byte of register */
4364 			ivar = ivar & 0xFFFFFF00;
4365 			ivar |= (vector | E1000_IVAR_VALID);
4366 		} else {
4367 			/* vector goes into third byte of register */
4368 			ivar = ivar & 0xFF00FFFF;
4369 			ivar |= ((vector | E1000_IVAR_VALID) << 16);
4370 		}
4371 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4372 
4373 		/* Accumulate interrupt-cause bits to enable */
4374 		igb->eims_mask |= (1 << vector);
4375 
4376 		vector ++;
4377 	}
4378 
4379 	for (i = 1; i < igb->num_tx_rings; i++) {
4380 		/*
4381 		 * Set vector for each tx ring from 2nd tx ring.
4382 		 * Note assumption that tx vectors numericall follow rx vectors.
4383 		 */
4384 		index = (i & 0x7);
4385 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4386 
4387 		if (i < 8) {
4388 			/* vector goes into second byte of register */
4389 			ivar = ivar & 0xFFFF00FF;
4390 			ivar |= ((vector | E1000_IVAR_VALID) << 8);
4391 		} else {
4392 			/* vector goes into fourth byte of register */
4393 			ivar = ivar & 0x00FFFFFF;
4394 			ivar |= (vector | E1000_IVAR_VALID) << 24;
4395 		}
4396 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4397 
4398 		/* Accumulate interrupt-cause bits to enable */
4399 		igb->eims_mask |= (1 << vector);
4400 
4401 		vector ++;
4402 	}
4403 
4404 	ASSERT(vector == igb->intr_cnt);
4405 }
4406 
4407 /*
4408  * igb_rem_intr_handlers - remove the interrupt handlers
4409  */
4410 static void
4411 igb_rem_intr_handlers(igb_t *igb)
4412 {
4413 	int i;
4414 	int rc;
4415 
4416 	for (i = 0; i < igb->intr_cnt; i++) {
4417 		rc = ddi_intr_remove_handler(igb->htable[i]);
4418 		if (rc != DDI_SUCCESS) {
4419 			IGB_DEBUGLOG_1(igb,
4420 			    "Remove intr handler failed: %d", rc);
4421 		}
4422 	}
4423 }
4424 
4425 /*
4426  * igb_rem_intrs - remove the allocated interrupts
4427  */
4428 static void
4429 igb_rem_intrs(igb_t *igb)
4430 {
4431 	int i;
4432 	int rc;
4433 
4434 	for (i = 0; i < igb->intr_cnt; i++) {
4435 		rc = ddi_intr_free(igb->htable[i]);
4436 		if (rc != DDI_SUCCESS) {
4437 			IGB_DEBUGLOG_1(igb,
4438 			    "Free intr failed: %d", rc);
4439 		}
4440 	}
4441 
4442 	kmem_free(igb->htable, igb->intr_size);
4443 	igb->htable = NULL;
4444 }
4445 
4446 /*
4447  * igb_enable_intrs - enable all the ddi interrupts
4448  */
4449 static int
4450 igb_enable_intrs(igb_t *igb)
4451 {
4452 	int i;
4453 	int rc;
4454 
4455 	/* Enable interrupts */
4456 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4457 		/* Call ddi_intr_block_enable() for MSI */
4458 		rc = ddi_intr_block_enable(igb->htable, igb->intr_cnt);
4459 		if (rc != DDI_SUCCESS) {
4460 			igb_log(igb,
4461 			    "Enable block intr failed: %d", rc);
4462 			return (IGB_FAILURE);
4463 		}
4464 	} else {
4465 		/* Call ddi_intr_enable() for Legacy/MSI non block enable */
4466 		for (i = 0; i < igb->intr_cnt; i++) {
4467 			rc = ddi_intr_enable(igb->htable[i]);
4468 			if (rc != DDI_SUCCESS) {
4469 				igb_log(igb,
4470 				    "Enable intr failed: %d", rc);
4471 				return (IGB_FAILURE);
4472 			}
4473 		}
4474 	}
4475 
4476 	return (IGB_SUCCESS);
4477 }
4478 
4479 /*
4480  * igb_disable_intrs - disable all the ddi interrupts
4481  */
4482 static int
4483 igb_disable_intrs(igb_t *igb)
4484 {
4485 	int i;
4486 	int rc;
4487 
4488 	/* Disable all interrupts */
4489 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4490 		rc = ddi_intr_block_disable(igb->htable, igb->intr_cnt);
4491 		if (rc != DDI_SUCCESS) {
4492 			igb_log(igb,
4493 			    "Disable block intr failed: %d", rc);
4494 			return (IGB_FAILURE);
4495 		}
4496 	} else {
4497 		for (i = 0; i < igb->intr_cnt; i++) {
4498 			rc = ddi_intr_disable(igb->htable[i]);
4499 			if (rc != DDI_SUCCESS) {
4500 				igb_log(igb,
4501 				    "Disable intr failed: %d", rc);
4502 				return (IGB_FAILURE);
4503 			}
4504 		}
4505 	}
4506 
4507 	return (IGB_SUCCESS);
4508 }
4509 
4510 /*
4511  * igb_get_phy_state - Get and save the parameters read from PHY registers
4512  */
4513 static void
4514 igb_get_phy_state(igb_t *igb)
4515 {
4516 	struct e1000_hw *hw = &igb->hw;
4517 	uint16_t phy_ctrl;
4518 	uint16_t phy_status;
4519 	uint16_t phy_an_adv;
4520 	uint16_t phy_an_exp;
4521 	uint16_t phy_ext_status;
4522 	uint16_t phy_1000t_ctrl;
4523 	uint16_t phy_1000t_status;
4524 	uint16_t phy_lp_able;
4525 
4526 	ASSERT(mutex_owned(&igb->gen_lock));
4527 
4528 	(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
4529 	(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
4530 	(void) e1000_read_phy_reg(hw, PHY_AUTONEG_ADV, &phy_an_adv);
4531 	(void) e1000_read_phy_reg(hw, PHY_AUTONEG_EXP, &phy_an_exp);
4532 	(void) e1000_read_phy_reg(hw, PHY_EXT_STATUS, &phy_ext_status);
4533 	(void) e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_1000t_ctrl);
4534 	(void) e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_1000t_status);
4535 	(void) e1000_read_phy_reg(hw, PHY_LP_ABILITY, &phy_lp_able);
4536 
4537 	igb->param_autoneg_cap =
4538 	    (phy_status & MII_SR_AUTONEG_CAPS) ? 1 : 0;
4539 	igb->param_pause_cap =
4540 	    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4541 	igb->param_asym_pause_cap =
4542 	    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4543 	igb->param_1000fdx_cap = ((phy_ext_status & IEEE_ESR_1000T_FD_CAPS) ||
4544 	    (phy_ext_status & IEEE_ESR_1000X_FD_CAPS)) ? 1 : 0;
4545 	igb->param_1000hdx_cap = ((phy_ext_status & IEEE_ESR_1000T_HD_CAPS) ||
4546 	    (phy_ext_status & IEEE_ESR_1000X_HD_CAPS)) ? 1 : 0;
4547 	igb->param_100t4_cap =
4548 	    (phy_status & MII_SR_100T4_CAPS) ? 1 : 0;
4549 	igb->param_100fdx_cap = ((phy_status & MII_SR_100X_FD_CAPS) ||
4550 	    (phy_status & MII_SR_100T2_FD_CAPS)) ? 1 : 0;
4551 	igb->param_100hdx_cap = ((phy_status & MII_SR_100X_HD_CAPS) ||
4552 	    (phy_status & MII_SR_100T2_HD_CAPS)) ? 1 : 0;
4553 	igb->param_10fdx_cap =
4554 	    (phy_status & MII_SR_10T_FD_CAPS) ? 1 : 0;
4555 	igb->param_10hdx_cap =
4556 	    (phy_status & MII_SR_10T_HD_CAPS) ? 1 : 0;
4557 	igb->param_rem_fault =
4558 	    (phy_status & MII_SR_REMOTE_FAULT) ? 1 : 0;
4559 
4560 	igb->param_adv_autoneg_cap = hw->mac.autoneg;
4561 	igb->param_adv_pause_cap =
4562 	    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4563 	igb->param_adv_asym_pause_cap =
4564 	    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4565 	igb->param_adv_1000hdx_cap =
4566 	    (phy_1000t_ctrl & CR_1000T_HD_CAPS) ? 1 : 0;
4567 	igb->param_adv_100t4_cap =
4568 	    (phy_an_adv & NWAY_AR_100T4_CAPS) ? 1 : 0;
4569 	igb->param_adv_rem_fault =
4570 	    (phy_an_adv & NWAY_AR_REMOTE_FAULT) ? 1 : 0;
4571 	if (igb->param_adv_autoneg_cap == 1) {
4572 		igb->param_adv_1000fdx_cap =
4573 		    (phy_1000t_ctrl & CR_1000T_FD_CAPS) ? 1 : 0;
4574 		igb->param_adv_100fdx_cap =
4575 		    (phy_an_adv & NWAY_AR_100TX_FD_CAPS) ? 1 : 0;
4576 		igb->param_adv_100hdx_cap =
4577 		    (phy_an_adv & NWAY_AR_100TX_HD_CAPS) ? 1 : 0;
4578 		igb->param_adv_10fdx_cap =
4579 		    (phy_an_adv & NWAY_AR_10T_FD_CAPS) ? 1 : 0;
4580 		igb->param_adv_10hdx_cap =
4581 		    (phy_an_adv & NWAY_AR_10T_HD_CAPS) ? 1 : 0;
4582 	}
4583 
4584 	igb->param_lp_autoneg_cap =
4585 	    (phy_an_exp & NWAY_ER_LP_NWAY_CAPS) ? 1 : 0;
4586 	igb->param_lp_pause_cap =
4587 	    (phy_lp_able & NWAY_LPAR_PAUSE) ? 1 : 0;
4588 	igb->param_lp_asym_pause_cap =
4589 	    (phy_lp_able & NWAY_LPAR_ASM_DIR) ? 1 : 0;
4590 	igb->param_lp_1000fdx_cap =
4591 	    (phy_1000t_status & SR_1000T_LP_FD_CAPS) ? 1 : 0;
4592 	igb->param_lp_1000hdx_cap =
4593 	    (phy_1000t_status & SR_1000T_LP_HD_CAPS) ? 1 : 0;
4594 	igb->param_lp_100t4_cap =
4595 	    (phy_lp_able & NWAY_LPAR_100T4_CAPS) ? 1 : 0;
4596 	igb->param_lp_100fdx_cap =
4597 	    (phy_lp_able & NWAY_LPAR_100TX_FD_CAPS) ? 1 : 0;
4598 	igb->param_lp_100hdx_cap =
4599 	    (phy_lp_able & NWAY_LPAR_100TX_HD_CAPS) ? 1 : 0;
4600 	igb->param_lp_10fdx_cap =
4601 	    (phy_lp_able & NWAY_LPAR_10T_FD_CAPS) ? 1 : 0;
4602 	igb->param_lp_10hdx_cap =
4603 	    (phy_lp_able & NWAY_LPAR_10T_HD_CAPS) ? 1 : 0;
4604 	igb->param_lp_rem_fault =
4605 	    (phy_lp_able & NWAY_LPAR_REMOTE_FAULT) ? 1 : 0;
4606 }
4607 
4608 /*
4609  * igb_get_driver_control
4610  */
4611 static void
4612 igb_get_driver_control(struct e1000_hw *hw)
4613 {
4614 	uint32_t ctrl_ext;
4615 
4616 	/* Notify firmware that driver is in control of device */
4617 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
4618 	ctrl_ext |= E1000_CTRL_EXT_DRV_LOAD;
4619 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
4620 }
4621 
4622 /*
4623  * igb_release_driver_control
4624  */
4625 static void
4626 igb_release_driver_control(struct e1000_hw *hw)
4627 {
4628 	uint32_t ctrl_ext;
4629 
4630 	/* Notify firmware that driver is no longer in control of device */
4631 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
4632 	ctrl_ext &= ~E1000_CTRL_EXT_DRV_LOAD;
4633 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
4634 }
4635 
4636 /*
4637  * igb_atomic_reserve - Atomic decrease operation
4638  */
4639 int
4640 igb_atomic_reserve(uint32_t *count_p, uint32_t n)
4641 {
4642 	uint32_t oldval;
4643 	uint32_t newval;
4644 
4645 	/* ATOMICALLY */
4646 	do {
4647 		oldval = *count_p;
4648 		if (oldval < n)
4649 			return (-1);
4650 		newval = oldval - n;
4651 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
4652 
4653 	return (newval);
4654 }
4655 
4656 /*
4657  * FMA support
4658  */
4659 
4660 int
4661 igb_check_acc_handle(ddi_acc_handle_t handle)
4662 {
4663 	ddi_fm_error_t de;
4664 
4665 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
4666 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
4667 	return (de.fme_status);
4668 }
4669 
4670 int
4671 igb_check_dma_handle(ddi_dma_handle_t handle)
4672 {
4673 	ddi_fm_error_t de;
4674 
4675 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
4676 	return (de.fme_status);
4677 }
4678 
4679 /*
4680  * The IO fault service error handling callback function
4681  */
4682 /*ARGSUSED*/
4683 static int
4684 igb_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
4685 {
4686 	/*
4687 	 * as the driver can always deal with an error in any dma or
4688 	 * access handle, we can just return the fme_status value.
4689 	 */
4690 	pci_ereport_post(dip, err, NULL);
4691 	return (err->fme_status);
4692 }
4693 
4694 static void
4695 igb_fm_init(igb_t *igb)
4696 {
4697 	ddi_iblock_cookie_t iblk;
4698 	int fma_acc_flag, fma_dma_flag;
4699 
4700 	/* Only register with IO Fault Services if we have some capability */
4701 	if (igb->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
4702 		igb_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
4703 		fma_acc_flag = 1;
4704 	} else {
4705 		igb_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
4706 		fma_acc_flag = 0;
4707 	}
4708 
4709 	if (igb->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
4710 		fma_dma_flag = 1;
4711 	} else {
4712 		fma_dma_flag = 0;
4713 	}
4714 
4715 	(void) igb_set_fma_flags(fma_acc_flag, fma_dma_flag);
4716 
4717 	if (igb->fm_capabilities) {
4718 
4719 		/* Register capabilities with IO Fault Services */
4720 		ddi_fm_init(igb->dip, &igb->fm_capabilities, &iblk);
4721 
4722 		/*
4723 		 * Initialize pci ereport capabilities if ereport capable
4724 		 */
4725 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
4726 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4727 			pci_ereport_setup(igb->dip);
4728 
4729 		/*
4730 		 * Register error callback if error callback capable
4731 		 */
4732 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4733 			ddi_fm_handler_register(igb->dip,
4734 			    igb_fm_error_cb, (void*) igb);
4735 	}
4736 }
4737 
4738 static void
4739 igb_fm_fini(igb_t *igb)
4740 {
4741 	/* Only unregister FMA capabilities if we registered some */
4742 	if (igb->fm_capabilities) {
4743 
4744 		/*
4745 		 * Release any resources allocated by pci_ereport_setup()
4746 		 */
4747 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
4748 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4749 			pci_ereport_teardown(igb->dip);
4750 
4751 		/*
4752 		 * Un-register error callback if error callback capable
4753 		 */
4754 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4755 			ddi_fm_handler_unregister(igb->dip);
4756 
4757 		/* Unregister from IO Fault Services */
4758 		ddi_fm_fini(igb->dip);
4759 	}
4760 }
4761 
4762 void
4763 igb_fm_ereport(igb_t *igb, char *detail)
4764 {
4765 	uint64_t ena;
4766 	char buf[FM_MAX_CLASS];
4767 
4768 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
4769 	ena = fm_ena_generate(0, FM_ENA_FMT1);
4770 	if (DDI_FM_EREPORT_CAP(igb->fm_capabilities)) {
4771 		ddi_fm_ereport_post(igb->dip, buf, ena, DDI_NOSLEEP,
4772 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
4773 	}
4774 }
4775