xref: /illumos-gate/usr/src/uts/common/io/ixgbe/ixgbe_main.c (revision 8a2b682e57a046b828f37bcde1776f131ef4629f)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright(c) 2007-2010 Intel Corporation. All rights reserved.
24  */
25 
26 /*
27  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Copyright 2019 Joyent, Inc.
29  * Copyright 2012 Nexenta Systems, Inc. All rights reserved.
30  * Copyright (c) 2013 Saso Kiselkov. All rights reserved.
31  * Copyright (c) 2013 OSN Online Service Nuernberg GmbH. All rights reserved.
32  * Copyright 2016 OmniTI Computer Consulting, Inc. All rights reserved.
33  */
34 
35 #include "ixgbe_sw.h"
36 
37 static char ixgbe_ident[] = "Intel 10Gb Ethernet";
38 
39 /*
40  * Local function protoypes
41  */
42 static int ixgbe_register_mac(ixgbe_t *);
43 static int ixgbe_identify_hardware(ixgbe_t *);
44 static int ixgbe_regs_map(ixgbe_t *);
45 static void ixgbe_init_properties(ixgbe_t *);
46 static int ixgbe_init_driver_settings(ixgbe_t *);
47 static void ixgbe_init_locks(ixgbe_t *);
48 static void ixgbe_destroy_locks(ixgbe_t *);
49 static int ixgbe_init(ixgbe_t *);
50 static int ixgbe_chip_start(ixgbe_t *);
51 static void ixgbe_chip_stop(ixgbe_t *);
52 static int ixgbe_reset(ixgbe_t *);
53 static void ixgbe_tx_clean(ixgbe_t *);
54 static boolean_t ixgbe_tx_drain(ixgbe_t *);
55 static boolean_t ixgbe_rx_drain(ixgbe_t *);
56 static int ixgbe_alloc_rings(ixgbe_t *);
57 static void ixgbe_free_rings(ixgbe_t *);
58 static int ixgbe_alloc_rx_data(ixgbe_t *);
59 static void ixgbe_free_rx_data(ixgbe_t *);
60 static void ixgbe_setup_rings(ixgbe_t *);
61 static void ixgbe_setup_rx(ixgbe_t *);
62 static void ixgbe_setup_tx(ixgbe_t *);
63 static void ixgbe_setup_rx_ring(ixgbe_rx_ring_t *);
64 static void ixgbe_setup_tx_ring(ixgbe_tx_ring_t *);
65 static void ixgbe_setup_rss(ixgbe_t *);
66 static void ixgbe_setup_vmdq(ixgbe_t *);
67 static void ixgbe_setup_vmdq_rss(ixgbe_t *);
68 static void ixgbe_setup_rss_table(ixgbe_t *);
69 static void ixgbe_init_unicst(ixgbe_t *);
70 static int ixgbe_unicst_find(ixgbe_t *, const uint8_t *);
71 static void ixgbe_setup_multicst(ixgbe_t *);
72 static void ixgbe_get_hw_state(ixgbe_t *);
73 static void ixgbe_setup_vmdq_rss_conf(ixgbe_t *ixgbe);
74 static void ixgbe_get_conf(ixgbe_t *);
75 static void ixgbe_init_params(ixgbe_t *);
76 static int ixgbe_get_prop(ixgbe_t *, char *, int, int, int);
77 static void ixgbe_driver_link_check(ixgbe_t *);
78 static void ixgbe_sfp_check(void *);
79 static void ixgbe_overtemp_check(void *);
80 static void ixgbe_phy_check(void *);
81 static void ixgbe_link_timer(void *);
82 static void ixgbe_local_timer(void *);
83 static void ixgbe_arm_watchdog_timer(ixgbe_t *);
84 static void ixgbe_restart_watchdog_timer(ixgbe_t *);
85 static void ixgbe_disable_adapter_interrupts(ixgbe_t *);
86 static void ixgbe_enable_adapter_interrupts(ixgbe_t *);
87 static boolean_t is_valid_mac_addr(uint8_t *);
88 static boolean_t ixgbe_stall_check(ixgbe_t *);
89 static boolean_t ixgbe_set_loopback_mode(ixgbe_t *, uint32_t);
90 static void ixgbe_set_internal_mac_loopback(ixgbe_t *);
91 static boolean_t ixgbe_find_mac_address(ixgbe_t *);
92 static int ixgbe_alloc_intrs(ixgbe_t *);
93 static int ixgbe_alloc_intr_handles(ixgbe_t *, int);
94 static int ixgbe_add_intr_handlers(ixgbe_t *);
95 static void ixgbe_map_rxring_to_vector(ixgbe_t *, int, int);
96 static void ixgbe_map_txring_to_vector(ixgbe_t *, int, int);
97 static void ixgbe_setup_ivar(ixgbe_t *, uint16_t, uint8_t, int8_t);
98 static void ixgbe_enable_ivar(ixgbe_t *, uint16_t, int8_t);
99 static void ixgbe_disable_ivar(ixgbe_t *, uint16_t, int8_t);
100 static uint32_t ixgbe_get_hw_rx_index(ixgbe_t *ixgbe, uint32_t sw_rx_index);
101 static int ixgbe_map_intrs_to_vectors(ixgbe_t *);
102 static void ixgbe_setup_adapter_vector(ixgbe_t *);
103 static void ixgbe_rem_intr_handlers(ixgbe_t *);
104 static void ixgbe_rem_intrs(ixgbe_t *);
105 static int ixgbe_enable_intrs(ixgbe_t *);
106 static int ixgbe_disable_intrs(ixgbe_t *);
107 static uint_t ixgbe_intr_legacy(void *, void *);
108 static uint_t ixgbe_intr_msi(void *, void *);
109 static uint_t ixgbe_intr_msix(void *, void *);
110 static void ixgbe_intr_rx_work(ixgbe_rx_ring_t *);
111 static void ixgbe_intr_tx_work(ixgbe_tx_ring_t *);
112 static void ixgbe_intr_other_work(ixgbe_t *, uint32_t);
113 static void ixgbe_get_driver_control(struct ixgbe_hw *);
114 static int ixgbe_addmac(void *, const uint8_t *);
115 static int ixgbe_remmac(void *, const uint8_t *);
116 static void ixgbe_release_driver_control(struct ixgbe_hw *);
117 
118 static int ixgbe_attach(dev_info_t *, ddi_attach_cmd_t);
119 static int ixgbe_detach(dev_info_t *, ddi_detach_cmd_t);
120 static int ixgbe_resume(dev_info_t *);
121 static int ixgbe_suspend(dev_info_t *);
122 static int ixgbe_quiesce(dev_info_t *);
123 static void ixgbe_unconfigure(dev_info_t *, ixgbe_t *);
124 static uint8_t *ixgbe_mc_table_itr(struct ixgbe_hw *, uint8_t **, uint32_t *);
125 static int ixgbe_cbfunc(dev_info_t *, ddi_cb_action_t, void *, void *, void *);
126 static int ixgbe_intr_cb_register(ixgbe_t *);
127 static int ixgbe_intr_adjust(ixgbe_t *, ddi_cb_action_t, int);
128 
129 static int ixgbe_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err,
130     const void *impl_data);
131 static void ixgbe_fm_init(ixgbe_t *);
132 static void ixgbe_fm_fini(ixgbe_t *);
133 
134 char *ixgbe_priv_props[] = {
135 	"_tx_copy_thresh",
136 	"_tx_recycle_thresh",
137 	"_tx_overload_thresh",
138 	"_tx_resched_thresh",
139 	"_rx_copy_thresh",
140 	"_rx_limit_per_intr",
141 	"_intr_throttling",
142 	"_adv_pause_cap",
143 	"_adv_asym_pause_cap",
144 	NULL
145 };
146 
147 #define	IXGBE_MAX_PRIV_PROPS \
148 	(sizeof (ixgbe_priv_props) / sizeof (mac_priv_prop_t))
149 
150 static struct cb_ops ixgbe_cb_ops = {
151 	nulldev,		/* cb_open */
152 	nulldev,		/* cb_close */
153 	nodev,			/* cb_strategy */
154 	nodev,			/* cb_print */
155 	nodev,			/* cb_dump */
156 	nodev,			/* cb_read */
157 	nodev,			/* cb_write */
158 	nodev,			/* cb_ioctl */
159 	nodev,			/* cb_devmap */
160 	nodev,			/* cb_mmap */
161 	nodev,			/* cb_segmap */
162 	nochpoll,		/* cb_chpoll */
163 	ddi_prop_op,		/* cb_prop_op */
164 	NULL,			/* cb_stream */
165 	D_MP | D_HOTPLUG,	/* cb_flag */
166 	CB_REV,			/* cb_rev */
167 	nodev,			/* cb_aread */
168 	nodev			/* cb_awrite */
169 };
170 
171 static struct dev_ops ixgbe_dev_ops = {
172 	DEVO_REV,		/* devo_rev */
173 	0,			/* devo_refcnt */
174 	NULL,			/* devo_getinfo */
175 	nulldev,		/* devo_identify */
176 	nulldev,		/* devo_probe */
177 	ixgbe_attach,		/* devo_attach */
178 	ixgbe_detach,		/* devo_detach */
179 	nodev,			/* devo_reset */
180 	&ixgbe_cb_ops,		/* devo_cb_ops */
181 	NULL,			/* devo_bus_ops */
182 	ddi_power,		/* devo_power */
183 	ixgbe_quiesce,		/* devo_quiesce */
184 };
185 
186 static struct modldrv ixgbe_modldrv = {
187 	&mod_driverops,		/* Type of module.  This one is a driver */
188 	ixgbe_ident,		/* Discription string */
189 	&ixgbe_dev_ops		/* driver ops */
190 };
191 
192 static struct modlinkage ixgbe_modlinkage = {
193 	MODREV_1, &ixgbe_modldrv, NULL
194 };
195 
196 /*
197  * Access attributes for register mapping
198  */
199 ddi_device_acc_attr_t ixgbe_regs_acc_attr = {
200 	DDI_DEVICE_ATTR_V1,
201 	DDI_STRUCTURE_LE_ACC,
202 	DDI_STRICTORDER_ACC,
203 	DDI_FLAGERR_ACC
204 };
205 
206 /*
207  * Loopback property
208  */
209 static lb_property_t lb_normal = {
210 	normal,	"normal", IXGBE_LB_NONE
211 };
212 
213 static lb_property_t lb_mac = {
214 	internal, "MAC", IXGBE_LB_INTERNAL_MAC
215 };
216 
217 static lb_property_t lb_external = {
218 	external, "External", IXGBE_LB_EXTERNAL
219 };
220 
221 #define	IXGBE_M_CALLBACK_FLAGS \
222 	(MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP | MC_PROPINFO)
223 
224 static mac_callbacks_t ixgbe_m_callbacks = {
225 	IXGBE_M_CALLBACK_FLAGS,
226 	ixgbe_m_stat,
227 	ixgbe_m_start,
228 	ixgbe_m_stop,
229 	ixgbe_m_promisc,
230 	ixgbe_m_multicst,
231 	NULL,
232 	NULL,
233 	NULL,
234 	ixgbe_m_ioctl,
235 	ixgbe_m_getcapab,
236 	NULL,
237 	NULL,
238 	ixgbe_m_setprop,
239 	ixgbe_m_getprop,
240 	ixgbe_m_propinfo
241 };
242 
243 /*
244  * Initialize capabilities of each supported adapter type
245  */
246 static adapter_info_t ixgbe_82598eb_cap = {
247 	64,		/* maximum number of rx queues */
248 	1,		/* minimum number of rx queues */
249 	64,		/* default number of rx queues */
250 	16,		/* maximum number of rx groups */
251 	1,		/* minimum number of rx groups */
252 	1,		/* default number of rx groups */
253 	32,		/* maximum number of tx queues */
254 	1,		/* minimum number of tx queues */
255 	8,		/* default number of tx queues */
256 	16366,		/* maximum MTU size */
257 	0xFFFF,		/* maximum interrupt throttle rate */
258 	0,		/* minimum interrupt throttle rate */
259 	200,		/* default interrupt throttle rate */
260 	18,		/* maximum total msix vectors */
261 	16,		/* maximum number of ring vectors */
262 	2,		/* maximum number of other vectors */
263 	IXGBE_EICR_LSC,	/* "other" interrupt types handled */
264 	0,		/* "other" interrupt types enable mask */
265 	(IXGBE_FLAG_DCA_CAPABLE	/* capability flags */
266 	| IXGBE_FLAG_RSS_CAPABLE
267 	| IXGBE_FLAG_VMDQ_CAPABLE)
268 };
269 
270 static adapter_info_t ixgbe_82599eb_cap = {
271 	128,		/* maximum number of rx queues */
272 	1,		/* minimum number of rx queues */
273 	128,		/* default number of rx queues */
274 	64,		/* maximum number of rx groups */
275 	1,		/* minimum number of rx groups */
276 	1,		/* default number of rx groups */
277 	128,		/* maximum number of tx queues */
278 	1,		/* minimum number of tx queues */
279 	8,		/* default number of tx queues */
280 	15500,		/* maximum MTU size */
281 	0xFF8,		/* maximum interrupt throttle rate */
282 	0,		/* minimum interrupt throttle rate */
283 	200,		/* default interrupt throttle rate */
284 	64,		/* maximum total msix vectors */
285 	16,		/* maximum number of ring vectors */
286 	2,		/* maximum number of other vectors */
287 	(IXGBE_EICR_LSC
288 	| IXGBE_EICR_GPI_SDP1
289 	| IXGBE_EICR_GPI_SDP2), /* "other" interrupt types handled */
290 
291 	(IXGBE_SDP1_GPIEN
292 	| IXGBE_SDP2_GPIEN), /* "other" interrupt types enable mask */
293 
294 	(IXGBE_FLAG_DCA_CAPABLE
295 	| IXGBE_FLAG_RSS_CAPABLE
296 	| IXGBE_FLAG_VMDQ_CAPABLE
297 	| IXGBE_FLAG_RSC_CAPABLE
298 	| IXGBE_FLAG_SFP_PLUG_CAPABLE) /* capability flags */
299 };
300 
301 static adapter_info_t ixgbe_X540_cap = {
302 	128,		/* maximum number of rx queues */
303 	1,		/* minimum number of rx queues */
304 	128,		/* default number of rx queues */
305 	64,		/* maximum number of rx groups */
306 	1,		/* minimum number of rx groups */
307 	1,		/* default number of rx groups */
308 	128,		/* maximum number of tx queues */
309 	1,		/* minimum number of tx queues */
310 	8,		/* default number of tx queues */
311 	15500,		/* maximum MTU size */
312 	0xFF8,		/* maximum interrupt throttle rate */
313 	0,		/* minimum interrupt throttle rate */
314 	200,		/* default interrupt throttle rate */
315 	64,		/* maximum total msix vectors */
316 	16,		/* maximum number of ring vectors */
317 	2,		/* maximum number of other vectors */
318 	(IXGBE_EICR_LSC
319 	| IXGBE_EICR_GPI_SDP1_X540
320 	| IXGBE_EICR_GPI_SDP2_X540), /* "other" interrupt types handled */
321 
322 	(IXGBE_SDP1_GPIEN_X540
323 	| IXGBE_SDP2_GPIEN_X540), /* "other" interrupt types enable mask */
324 
325 	(IXGBE_FLAG_DCA_CAPABLE
326 	| IXGBE_FLAG_RSS_CAPABLE
327 	| IXGBE_FLAG_VMDQ_CAPABLE
328 	| IXGBE_FLAG_RSC_CAPABLE) /* capability flags */
329 };
330 
331 static adapter_info_t ixgbe_X550_cap = {
332 	128,		/* maximum number of rx queues */
333 	1,		/* minimum number of rx queues */
334 	128,		/* default number of rx queues */
335 	64,		/* maximum number of rx groups */
336 	1,		/* minimum number of rx groups */
337 	1,		/* default number of rx groups */
338 	128,		/* maximum number of tx queues */
339 	1,		/* minimum number of tx queues */
340 	8,		/* default number of tx queues */
341 	15500,		/* maximum MTU size */
342 	0xFF8,		/* maximum interrupt throttle rate */
343 	0,		/* minimum interrupt throttle rate */
344 	0x200,		/* default interrupt throttle rate */
345 	64,		/* maximum total msix vectors */
346 	16,		/* maximum number of ring vectors */
347 	2,		/* maximum number of other vectors */
348 	IXGBE_EICR_LSC,	/* "other" interrupt types handled */
349 	0,		/* "other" interrupt types enable mask */
350 	(IXGBE_FLAG_RSS_CAPABLE
351 	| IXGBE_FLAG_VMDQ_CAPABLE
352 	| IXGBE_FLAG_RSC_CAPABLE) /* capability flags */
353 };
354 
355 /*
356  * Module Initialization Functions.
357  */
358 
359 int
360 _init(void)
361 {
362 	int status;
363 
364 	mac_init_ops(&ixgbe_dev_ops, MODULE_NAME);
365 
366 	status = mod_install(&ixgbe_modlinkage);
367 
368 	if (status != DDI_SUCCESS) {
369 		mac_fini_ops(&ixgbe_dev_ops);
370 	}
371 
372 	return (status);
373 }
374 
375 int
376 _fini(void)
377 {
378 	int status;
379 
380 	status = mod_remove(&ixgbe_modlinkage);
381 
382 	if (status == DDI_SUCCESS) {
383 		mac_fini_ops(&ixgbe_dev_ops);
384 	}
385 
386 	return (status);
387 }
388 
389 int
390 _info(struct modinfo *modinfop)
391 {
392 	int status;
393 
394 	status = mod_info(&ixgbe_modlinkage, modinfop);
395 
396 	return (status);
397 }
398 
399 /*
400  * ixgbe_attach - Driver attach.
401  *
402  * This function is the device specific initialization entry
403  * point. This entry point is required and must be written.
404  * The DDI_ATTACH command must be provided in the attach entry
405  * point. When attach() is called with cmd set to DDI_ATTACH,
406  * all normal kernel services (such as kmem_alloc(9F)) are
407  * available for use by the driver.
408  *
409  * The attach() function will be called once for each instance
410  * of  the  device  on  the  system with cmd set to DDI_ATTACH.
411  * Until attach() succeeds, the only driver entry points which
412  * may be called are open(9E) and getinfo(9E).
413  */
414 static int
415 ixgbe_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
416 {
417 	ixgbe_t *ixgbe;
418 	struct ixgbe_osdep *osdep;
419 	struct ixgbe_hw *hw;
420 	int instance;
421 	char taskqname[32];
422 
423 	/*
424 	 * Check the command and perform corresponding operations
425 	 */
426 	switch (cmd) {
427 	default:
428 		return (DDI_FAILURE);
429 
430 	case DDI_RESUME:
431 		return (ixgbe_resume(devinfo));
432 
433 	case DDI_ATTACH:
434 		break;
435 	}
436 
437 	/* Get the device instance */
438 	instance = ddi_get_instance(devinfo);
439 
440 	/* Allocate memory for the instance data structure */
441 	ixgbe = kmem_zalloc(sizeof (ixgbe_t), KM_SLEEP);
442 
443 	ixgbe->dip = devinfo;
444 	ixgbe->instance = instance;
445 
446 	hw = &ixgbe->hw;
447 	osdep = &ixgbe->osdep;
448 	hw->back = osdep;
449 	osdep->ixgbe = ixgbe;
450 
451 	/* Attach the instance pointer to the dev_info data structure */
452 	ddi_set_driver_private(devinfo, ixgbe);
453 
454 	/*
455 	 * Initialize for FMA support
456 	 */
457 	ixgbe->fm_capabilities = ixgbe_get_prop(ixgbe, PROP_FM_CAPABLE,
458 	    0, 0x0f, DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
459 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
460 	ixgbe_fm_init(ixgbe);
461 	ixgbe->attach_progress |= ATTACH_PROGRESS_FM_INIT;
462 
463 	/*
464 	 * Map PCI config space registers
465 	 */
466 	if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
467 		ixgbe_error(ixgbe, "Failed to map PCI configurations");
468 		goto attach_fail;
469 	}
470 	ixgbe->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
471 
472 	/*
473 	 * Identify the chipset family
474 	 */
475 	if (ixgbe_identify_hardware(ixgbe) != IXGBE_SUCCESS) {
476 		ixgbe_error(ixgbe, "Failed to identify hardware");
477 		goto attach_fail;
478 	}
479 
480 	/*
481 	 * Map device registers
482 	 */
483 	if (ixgbe_regs_map(ixgbe) != IXGBE_SUCCESS) {
484 		ixgbe_error(ixgbe, "Failed to map device registers");
485 		goto attach_fail;
486 	}
487 	ixgbe->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
488 
489 	/*
490 	 * Initialize driver parameters
491 	 */
492 	ixgbe_init_properties(ixgbe);
493 	ixgbe->attach_progress |= ATTACH_PROGRESS_PROPS;
494 
495 	/*
496 	 * Register interrupt callback
497 	 */
498 	if (ixgbe_intr_cb_register(ixgbe) != IXGBE_SUCCESS) {
499 		ixgbe_error(ixgbe, "Failed to register interrupt callback");
500 		goto attach_fail;
501 	}
502 
503 	/*
504 	 * Allocate interrupts
505 	 */
506 	if (ixgbe_alloc_intrs(ixgbe) != IXGBE_SUCCESS) {
507 		ixgbe_error(ixgbe, "Failed to allocate interrupts");
508 		goto attach_fail;
509 	}
510 	ixgbe->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
511 
512 	/*
513 	 * Allocate rx/tx rings based on the ring numbers.
514 	 * The actual numbers of rx/tx rings are decided by the number of
515 	 * allocated interrupt vectors, so we should allocate the rings after
516 	 * interrupts are allocated.
517 	 */
518 	if (ixgbe_alloc_rings(ixgbe) != IXGBE_SUCCESS) {
519 		ixgbe_error(ixgbe, "Failed to allocate rx and tx rings");
520 		goto attach_fail;
521 	}
522 	ixgbe->attach_progress |= ATTACH_PROGRESS_ALLOC_RINGS;
523 
524 	/*
525 	 * Map rings to interrupt vectors
526 	 */
527 	if (ixgbe_map_intrs_to_vectors(ixgbe) != IXGBE_SUCCESS) {
528 		ixgbe_error(ixgbe, "Failed to map interrupts to vectors");
529 		goto attach_fail;
530 	}
531 
532 	/*
533 	 * Add interrupt handlers
534 	 */
535 	if (ixgbe_add_intr_handlers(ixgbe) != IXGBE_SUCCESS) {
536 		ixgbe_error(ixgbe, "Failed to add interrupt handlers");
537 		goto attach_fail;
538 	}
539 	ixgbe->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
540 
541 	/*
542 	 * Create a taskq for sfp-change
543 	 */
544 	(void) sprintf(taskqname, "ixgbe%d_sfp_taskq", instance);
545 	if ((ixgbe->sfp_taskq = ddi_taskq_create(devinfo, taskqname,
546 	    1, TASKQ_DEFAULTPRI, 0)) == NULL) {
547 		ixgbe_error(ixgbe, "sfp_taskq create failed");
548 		goto attach_fail;
549 	}
550 	ixgbe->attach_progress |= ATTACH_PROGRESS_SFP_TASKQ;
551 
552 	/*
553 	 * Create a taskq for over-temp
554 	 */
555 	(void) sprintf(taskqname, "ixgbe%d_overtemp_taskq", instance);
556 	if ((ixgbe->overtemp_taskq = ddi_taskq_create(devinfo, taskqname,
557 	    1, TASKQ_DEFAULTPRI, 0)) == NULL) {
558 		ixgbe_error(ixgbe, "overtemp_taskq create failed");
559 		goto attach_fail;
560 	}
561 	ixgbe->attach_progress |= ATTACH_PROGRESS_OVERTEMP_TASKQ;
562 
563 	/*
564 	 * Create a taskq for processing external PHY interrupts
565 	 */
566 	(void) sprintf(taskqname, "ixgbe%d_phy_taskq", instance);
567 	if ((ixgbe->phy_taskq = ddi_taskq_create(devinfo, taskqname,
568 	    1, TASKQ_DEFAULTPRI, 0)) == NULL) {
569 		ixgbe_error(ixgbe, "phy_taskq create failed");
570 		goto attach_fail;
571 	}
572 	ixgbe->attach_progress |= ATTACH_PROGRESS_PHY_TASKQ;
573 
574 	/*
575 	 * Initialize driver parameters
576 	 */
577 	if (ixgbe_init_driver_settings(ixgbe) != IXGBE_SUCCESS) {
578 		ixgbe_error(ixgbe, "Failed to initialize driver settings");
579 		goto attach_fail;
580 	}
581 
582 	/*
583 	 * Initialize mutexes for this device.
584 	 * Do this before enabling the interrupt handler and
585 	 * register the softint to avoid the condition where
586 	 * interrupt handler can try using uninitialized mutex.
587 	 */
588 	ixgbe_init_locks(ixgbe);
589 	ixgbe->attach_progress |= ATTACH_PROGRESS_LOCKS;
590 
591 	/*
592 	 * Initialize chipset hardware
593 	 */
594 	if (ixgbe_init(ixgbe) != IXGBE_SUCCESS) {
595 		ixgbe_error(ixgbe, "Failed to initialize adapter");
596 		goto attach_fail;
597 	}
598 	ixgbe->link_check_complete = B_FALSE;
599 	ixgbe->link_check_hrtime = gethrtime() +
600 	    (IXGBE_LINK_UP_TIME * 100000000ULL);
601 	ixgbe->attach_progress |= ATTACH_PROGRESS_INIT;
602 
603 	if (ixgbe_check_acc_handle(ixgbe->osdep.cfg_handle) != DDI_FM_OK) {
604 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
605 		goto attach_fail;
606 	}
607 
608 	/*
609 	 * Initialize adapter capabilities
610 	 */
611 	ixgbe_init_params(ixgbe);
612 
613 	/*
614 	 * Initialize statistics
615 	 */
616 	if (ixgbe_init_stats(ixgbe) != IXGBE_SUCCESS) {
617 		ixgbe_error(ixgbe, "Failed to initialize statistics");
618 		goto attach_fail;
619 	}
620 	ixgbe->attach_progress |= ATTACH_PROGRESS_STATS;
621 
622 	/*
623 	 * Register the driver to the MAC
624 	 */
625 	if (ixgbe_register_mac(ixgbe) != IXGBE_SUCCESS) {
626 		ixgbe_error(ixgbe, "Failed to register MAC");
627 		goto attach_fail;
628 	}
629 	mac_link_update(ixgbe->mac_hdl, LINK_STATE_UNKNOWN);
630 	ixgbe->attach_progress |= ATTACH_PROGRESS_MAC;
631 
632 	ixgbe->periodic_id = ddi_periodic_add(ixgbe_link_timer, ixgbe,
633 	    IXGBE_CYCLIC_PERIOD, DDI_IPL_0);
634 	if (ixgbe->periodic_id == 0) {
635 		ixgbe_error(ixgbe, "Failed to add the link check timer");
636 		goto attach_fail;
637 	}
638 	ixgbe->attach_progress |= ATTACH_PROGRESS_LINK_TIMER;
639 
640 	/*
641 	 * Now that mutex locks are initialized, and the chip is also
642 	 * initialized, enable interrupts.
643 	 */
644 	if (ixgbe_enable_intrs(ixgbe) != IXGBE_SUCCESS) {
645 		ixgbe_error(ixgbe, "Failed to enable DDI interrupts");
646 		goto attach_fail;
647 	}
648 	ixgbe->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
649 
650 	ixgbe_log(ixgbe, "%s", ixgbe_ident);
651 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_INITIALIZED);
652 
653 	return (DDI_SUCCESS);
654 
655 attach_fail:
656 	ixgbe_unconfigure(devinfo, ixgbe);
657 	return (DDI_FAILURE);
658 }
659 
660 /*
661  * ixgbe_detach - Driver detach.
662  *
663  * The detach() function is the complement of the attach routine.
664  * If cmd is set to DDI_DETACH, detach() is used to remove  the
665  * state  associated  with  a  given  instance of a device node
666  * prior to the removal of that instance from the system.
667  *
668  * The detach() function will be called once for each  instance
669  * of the device for which there has been a successful attach()
670  * once there are no longer  any  opens  on  the  device.
671  *
672  * Interrupts routine are disabled, All memory allocated by this
673  * driver are freed.
674  */
675 static int
676 ixgbe_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
677 {
678 	ixgbe_t *ixgbe;
679 
680 	/*
681 	 * Check detach command
682 	 */
683 	switch (cmd) {
684 	default:
685 		return (DDI_FAILURE);
686 
687 	case DDI_SUSPEND:
688 		return (ixgbe_suspend(devinfo));
689 
690 	case DDI_DETACH:
691 		break;
692 	}
693 
694 	/*
695 	 * Get the pointer to the driver private data structure
696 	 */
697 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
698 	if (ixgbe == NULL)
699 		return (DDI_FAILURE);
700 
701 	/*
702 	 * If the device is still running, it needs to be stopped first.
703 	 * This check is necessary because under some specific circumstances,
704 	 * the detach routine can be called without stopping the interface
705 	 * first.
706 	 */
707 	if (ixgbe->ixgbe_state & IXGBE_STARTED) {
708 		atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_STARTED);
709 		mutex_enter(&ixgbe->gen_lock);
710 		ixgbe_stop(ixgbe, B_TRUE);
711 		mutex_exit(&ixgbe->gen_lock);
712 		/* Disable and stop the watchdog timer */
713 		ixgbe_disable_watchdog_timer(ixgbe);
714 	}
715 
716 	/*
717 	 * Check if there are still rx buffers held by the upper layer.
718 	 * If so, fail the detach.
719 	 */
720 	if (!ixgbe_rx_drain(ixgbe))
721 		return (DDI_FAILURE);
722 
723 	/*
724 	 * Do the remaining unconfigure routines
725 	 */
726 	ixgbe_unconfigure(devinfo, ixgbe);
727 
728 	return (DDI_SUCCESS);
729 }
730 
731 /*
732  * quiesce(9E) entry point.
733  *
734  * This function is called when the system is single-threaded at high
735  * PIL with preemption disabled. Therefore, this function must not be
736  * blocked.
737  *
738  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
739  * DDI_FAILURE indicates an error condition and should almost never happen.
740  */
741 static int
742 ixgbe_quiesce(dev_info_t *devinfo)
743 {
744 	ixgbe_t *ixgbe;
745 	struct ixgbe_hw *hw;
746 
747 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
748 
749 	if (ixgbe == NULL)
750 		return (DDI_FAILURE);
751 
752 	hw = &ixgbe->hw;
753 
754 	/*
755 	 * Disable the adapter interrupts
756 	 */
757 	ixgbe_disable_adapter_interrupts(ixgbe);
758 
759 	/*
760 	 * Tell firmware driver is no longer in control
761 	 */
762 	ixgbe_release_driver_control(hw);
763 
764 	/*
765 	 * Reset the chipset
766 	 */
767 	(void) ixgbe_reset_hw(hw);
768 
769 	/*
770 	 * Reset PHY
771 	 */
772 	(void) ixgbe_reset_phy(hw);
773 
774 	return (DDI_SUCCESS);
775 }
776 
777 static void
778 ixgbe_unconfigure(dev_info_t *devinfo, ixgbe_t *ixgbe)
779 {
780 	/*
781 	 * Disable interrupt
782 	 */
783 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
784 		(void) ixgbe_disable_intrs(ixgbe);
785 	}
786 
787 	/*
788 	 * remove the link check timer
789 	 */
790 	if (ixgbe->attach_progress & ATTACH_PROGRESS_LINK_TIMER) {
791 		if (ixgbe->periodic_id != NULL) {
792 			ddi_periodic_delete(ixgbe->periodic_id);
793 			ixgbe->periodic_id = NULL;
794 		}
795 	}
796 
797 	/*
798 	 * Unregister MAC
799 	 */
800 	if (ixgbe->attach_progress & ATTACH_PROGRESS_MAC) {
801 		(void) mac_unregister(ixgbe->mac_hdl);
802 	}
803 
804 	/*
805 	 * Free statistics
806 	 */
807 	if (ixgbe->attach_progress & ATTACH_PROGRESS_STATS) {
808 		kstat_delete((kstat_t *)ixgbe->ixgbe_ks);
809 	}
810 
811 	/*
812 	 * Remove interrupt handlers
813 	 */
814 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
815 		ixgbe_rem_intr_handlers(ixgbe);
816 	}
817 
818 	/*
819 	 * Remove taskq for sfp-status-change
820 	 */
821 	if (ixgbe->attach_progress & ATTACH_PROGRESS_SFP_TASKQ) {
822 		ddi_taskq_destroy(ixgbe->sfp_taskq);
823 	}
824 
825 	/*
826 	 * Remove taskq for over-temp
827 	 */
828 	if (ixgbe->attach_progress & ATTACH_PROGRESS_OVERTEMP_TASKQ) {
829 		ddi_taskq_destroy(ixgbe->overtemp_taskq);
830 	}
831 
832 	/*
833 	 * Remove taskq for external PHYs
834 	 */
835 	if (ixgbe->attach_progress & ATTACH_PROGRESS_PHY_TASKQ) {
836 		ddi_taskq_destroy(ixgbe->phy_taskq);
837 	}
838 
839 	/*
840 	 * Remove interrupts
841 	 */
842 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ALLOC_INTR) {
843 		ixgbe_rem_intrs(ixgbe);
844 	}
845 
846 	/*
847 	 * Unregister interrupt callback handler
848 	 */
849 	if (ixgbe->cb_hdl != NULL) {
850 		(void) ddi_cb_unregister(ixgbe->cb_hdl);
851 	}
852 
853 	/*
854 	 * Remove driver properties
855 	 */
856 	if (ixgbe->attach_progress & ATTACH_PROGRESS_PROPS) {
857 		(void) ddi_prop_remove_all(devinfo);
858 	}
859 
860 	/*
861 	 * Stop the chipset
862 	 */
863 	if (ixgbe->attach_progress & ATTACH_PROGRESS_INIT) {
864 		mutex_enter(&ixgbe->gen_lock);
865 		ixgbe_chip_stop(ixgbe);
866 		mutex_exit(&ixgbe->gen_lock);
867 	}
868 
869 	/*
870 	 * Free register handle
871 	 */
872 	if (ixgbe->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
873 		if (ixgbe->osdep.reg_handle != NULL)
874 			ddi_regs_map_free(&ixgbe->osdep.reg_handle);
875 	}
876 
877 	/*
878 	 * Free PCI config handle
879 	 */
880 	if (ixgbe->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
881 		if (ixgbe->osdep.cfg_handle != NULL)
882 			pci_config_teardown(&ixgbe->osdep.cfg_handle);
883 	}
884 
885 	/*
886 	 * Free locks
887 	 */
888 	if (ixgbe->attach_progress & ATTACH_PROGRESS_LOCKS) {
889 		ixgbe_destroy_locks(ixgbe);
890 	}
891 
892 	/*
893 	 * Free the rx/tx rings
894 	 */
895 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
896 		ixgbe_free_rings(ixgbe);
897 	}
898 
899 	/*
900 	 * Unregister FMA capabilities
901 	 */
902 	if (ixgbe->attach_progress & ATTACH_PROGRESS_FM_INIT) {
903 		ixgbe_fm_fini(ixgbe);
904 	}
905 
906 	/*
907 	 * Free the driver data structure
908 	 */
909 	kmem_free(ixgbe, sizeof (ixgbe_t));
910 
911 	ddi_set_driver_private(devinfo, NULL);
912 }
913 
914 /*
915  * ixgbe_register_mac - Register the driver and its function pointers with
916  * the GLD interface.
917  */
918 static int
919 ixgbe_register_mac(ixgbe_t *ixgbe)
920 {
921 	struct ixgbe_hw *hw = &ixgbe->hw;
922 	mac_register_t *mac;
923 	int status;
924 
925 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
926 		return (IXGBE_FAILURE);
927 
928 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
929 	mac->m_driver = ixgbe;
930 	mac->m_dip = ixgbe->dip;
931 	mac->m_src_addr = hw->mac.addr;
932 	mac->m_callbacks = &ixgbe_m_callbacks;
933 	mac->m_min_sdu = 0;
934 	mac->m_max_sdu = ixgbe->default_mtu;
935 	mac->m_margin = VLAN_TAGSZ;
936 	mac->m_priv_props = ixgbe_priv_props;
937 	mac->m_v12n = MAC_VIRT_LEVEL1;
938 
939 	status = mac_register(mac, &ixgbe->mac_hdl);
940 
941 	mac_free(mac);
942 
943 	return ((status == 0) ? IXGBE_SUCCESS : IXGBE_FAILURE);
944 }
945 
946 /*
947  * ixgbe_identify_hardware - Identify the type of the chipset.
948  */
949 static int
950 ixgbe_identify_hardware(ixgbe_t *ixgbe)
951 {
952 	struct ixgbe_hw *hw = &ixgbe->hw;
953 	struct ixgbe_osdep *osdep = &ixgbe->osdep;
954 
955 	/*
956 	 * Get the device id
957 	 */
958 	hw->vendor_id =
959 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
960 	hw->device_id =
961 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
962 	hw->revision_id =
963 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
964 	hw->subsystem_device_id =
965 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
966 	hw->subsystem_vendor_id =
967 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
968 
969 	/*
970 	 * Set the mac type of the adapter based on the device id
971 	 */
972 	if (ixgbe_set_mac_type(hw) != IXGBE_SUCCESS) {
973 		return (IXGBE_FAILURE);
974 	}
975 
976 	/*
977 	 * Install adapter capabilities
978 	 */
979 	switch (hw->mac.type) {
980 	case ixgbe_mac_82598EB:
981 		IXGBE_DEBUGLOG_0(ixgbe, "identify 82598 adapter\n");
982 		ixgbe->capab = &ixgbe_82598eb_cap;
983 
984 		if (ixgbe_get_media_type(hw) == ixgbe_media_type_copper) {
985 			ixgbe->capab->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
986 			ixgbe->capab->other_intr |= IXGBE_EICR_GPI_SDP1;
987 			ixgbe->capab->other_gpie |= IXGBE_SDP1_GPIEN;
988 		}
989 		break;
990 
991 	case ixgbe_mac_82599EB:
992 		IXGBE_DEBUGLOG_0(ixgbe, "identify 82599 adapter\n");
993 		ixgbe->capab = &ixgbe_82599eb_cap;
994 
995 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM) {
996 			ixgbe->capab->flags |= IXGBE_FLAG_TEMP_SENSOR_CAPABLE;
997 			ixgbe->capab->other_intr |= IXGBE_EICR_GPI_SDP0;
998 			ixgbe->capab->other_gpie |= IXGBE_SDP0_GPIEN;
999 		}
1000 		break;
1001 
1002 	case ixgbe_mac_X540:
1003 		IXGBE_DEBUGLOG_0(ixgbe, "identify X540 adapter\n");
1004 		ixgbe->capab = &ixgbe_X540_cap;
1005 		/*
1006 		 * For now, X540 is all set in its capab structure.
1007 		 * As other X540 variants show up, things can change here.
1008 		 */
1009 		break;
1010 
1011 	case ixgbe_mac_X550:
1012 	case ixgbe_mac_X550EM_x:
1013 	case ixgbe_mac_X550EM_a:
1014 		IXGBE_DEBUGLOG_0(ixgbe, "identify X550 adapter\n");
1015 		ixgbe->capab = &ixgbe_X550_cap;
1016 
1017 		if (hw->device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
1018 		    hw->device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
1019 		    hw->device_id == IXGBE_DEV_ID_X550EM_A_SFP_N ||
1020 		    hw->device_id == IXGBE_DEV_ID_X550EM_A_QSFP ||
1021 		    hw->device_id == IXGBE_DEV_ID_X550EM_A_QSFP_N) {
1022 			ixgbe->capab->flags |= IXGBE_FLAG_SFP_PLUG_CAPABLE;
1023 		}
1024 
1025 		/*
1026 		 * Link detection on X552 SFP+ and X552/X557-AT
1027 		 */
1028 		if (hw->device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
1029 		    hw->device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
1030 		    hw->device_id == IXGBE_DEV_ID_X550EM_A_SFP_N ||
1031 		    hw->device_id == IXGBE_DEV_ID_X550EM_X_10G_T) {
1032 			ixgbe->capab->other_intr |=
1033 			    IXGBE_EIMS_GPI_SDP0_BY_MAC(hw);
1034 		}
1035 		if (hw->phy.type == ixgbe_phy_x550em_ext_t) {
1036 			ixgbe->capab->other_gpie |= IXGBE_SDP0_GPIEN_X540;
1037 		}
1038 		break;
1039 
1040 	default:
1041 		IXGBE_DEBUGLOG_1(ixgbe,
1042 		    "adapter not supported in ixgbe_identify_hardware(): %d\n",
1043 		    hw->mac.type);
1044 		return (IXGBE_FAILURE);
1045 	}
1046 
1047 	return (IXGBE_SUCCESS);
1048 }
1049 
1050 /*
1051  * ixgbe_regs_map - Map the device registers.
1052  *
1053  */
1054 static int
1055 ixgbe_regs_map(ixgbe_t *ixgbe)
1056 {
1057 	dev_info_t *devinfo = ixgbe->dip;
1058 	struct ixgbe_hw *hw = &ixgbe->hw;
1059 	struct ixgbe_osdep *osdep = &ixgbe->osdep;
1060 	off_t mem_size;
1061 
1062 	/*
1063 	 * First get the size of device registers to be mapped.
1064 	 */
1065 	if (ddi_dev_regsize(devinfo, IXGBE_ADAPTER_REGSET, &mem_size)
1066 	    != DDI_SUCCESS) {
1067 		return (IXGBE_FAILURE);
1068 	}
1069 
1070 	/*
1071 	 * Call ddi_regs_map_setup() to map registers
1072 	 */
1073 	if ((ddi_regs_map_setup(devinfo, IXGBE_ADAPTER_REGSET,
1074 	    (caddr_t *)&hw->hw_addr, 0,
1075 	    mem_size, &ixgbe_regs_acc_attr,
1076 	    &osdep->reg_handle)) != DDI_SUCCESS) {
1077 		return (IXGBE_FAILURE);
1078 	}
1079 
1080 	return (IXGBE_SUCCESS);
1081 }
1082 
1083 /*
1084  * ixgbe_init_properties - Initialize driver properties.
1085  */
1086 static void
1087 ixgbe_init_properties(ixgbe_t *ixgbe)
1088 {
1089 	/*
1090 	 * Get conf file properties, including link settings
1091 	 * jumbo frames, ring number, descriptor number, etc.
1092 	 */
1093 	ixgbe_get_conf(ixgbe);
1094 }
1095 
1096 /*
1097  * ixgbe_init_driver_settings - Initialize driver settings.
1098  *
1099  * The settings include hardware function pointers, bus information,
1100  * rx/tx rings settings, link state, and any other parameters that
1101  * need to be setup during driver initialization.
1102  */
1103 static int
1104 ixgbe_init_driver_settings(ixgbe_t *ixgbe)
1105 {
1106 	struct ixgbe_hw *hw = &ixgbe->hw;
1107 	dev_info_t *devinfo = ixgbe->dip;
1108 	ixgbe_rx_ring_t *rx_ring;
1109 	ixgbe_rx_group_t *rx_group;
1110 	ixgbe_tx_ring_t *tx_ring;
1111 	uint32_t rx_size;
1112 	uint32_t tx_size;
1113 	uint32_t ring_per_group;
1114 	int i;
1115 
1116 	/*
1117 	 * Initialize chipset specific hardware function pointers
1118 	 */
1119 	if (ixgbe_init_shared_code(hw) != IXGBE_SUCCESS) {
1120 		return (IXGBE_FAILURE);
1121 	}
1122 
1123 	/*
1124 	 * Get the system page size
1125 	 */
1126 	ixgbe->sys_page_size = ddi_ptob(devinfo, (ulong_t)1);
1127 
1128 	/*
1129 	 * Set rx buffer size
1130 	 *
1131 	 * The IP header alignment room is counted in the calculation.
1132 	 * The rx buffer size is in unit of 1K that is required by the
1133 	 * chipset hardware.
1134 	 */
1135 	rx_size = ixgbe->max_frame_size + IPHDR_ALIGN_ROOM;
1136 	ixgbe->rx_buf_size = ((rx_size >> 10) +
1137 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
1138 
1139 	/*
1140 	 * Set tx buffer size
1141 	 */
1142 	tx_size = ixgbe->max_frame_size;
1143 	ixgbe->tx_buf_size = ((tx_size >> 10) +
1144 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
1145 
1146 	/*
1147 	 * Initialize rx/tx rings/groups parameters
1148 	 */
1149 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
1150 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1151 		rx_ring = &ixgbe->rx_rings[i];
1152 		rx_ring->index = i;
1153 		rx_ring->ixgbe = ixgbe;
1154 		rx_ring->group_index = i / ring_per_group;
1155 		rx_ring->hw_index = ixgbe_get_hw_rx_index(ixgbe, i);
1156 	}
1157 
1158 	for (i = 0; i < ixgbe->num_rx_groups; i++) {
1159 		rx_group = &ixgbe->rx_groups[i];
1160 		rx_group->index = i;
1161 		rx_group->ixgbe = ixgbe;
1162 	}
1163 
1164 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1165 		tx_ring = &ixgbe->tx_rings[i];
1166 		tx_ring->index = i;
1167 		tx_ring->ixgbe = ixgbe;
1168 		if (ixgbe->tx_head_wb_enable)
1169 			tx_ring->tx_recycle = ixgbe_tx_recycle_head_wb;
1170 		else
1171 			tx_ring->tx_recycle = ixgbe_tx_recycle_legacy;
1172 
1173 		tx_ring->ring_size = ixgbe->tx_ring_size;
1174 		tx_ring->free_list_size = ixgbe->tx_ring_size +
1175 		    (ixgbe->tx_ring_size >> 1);
1176 	}
1177 
1178 	/*
1179 	 * Initialize values of interrupt throttling rate
1180 	 */
1181 	for (i = 1; i < MAX_INTR_VECTOR; i++)
1182 		ixgbe->intr_throttling[i] = ixgbe->intr_throttling[0];
1183 
1184 	/*
1185 	 * The initial link state should be "unknown"
1186 	 */
1187 	ixgbe->link_state = LINK_STATE_UNKNOWN;
1188 
1189 	return (IXGBE_SUCCESS);
1190 }
1191 
1192 /*
1193  * ixgbe_init_locks - Initialize locks.
1194  */
1195 static void
1196 ixgbe_init_locks(ixgbe_t *ixgbe)
1197 {
1198 	ixgbe_rx_ring_t *rx_ring;
1199 	ixgbe_tx_ring_t *tx_ring;
1200 	int i;
1201 
1202 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1203 		rx_ring = &ixgbe->rx_rings[i];
1204 		mutex_init(&rx_ring->rx_lock, NULL,
1205 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1206 	}
1207 
1208 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1209 		tx_ring = &ixgbe->tx_rings[i];
1210 		mutex_init(&tx_ring->tx_lock, NULL,
1211 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1212 		mutex_init(&tx_ring->recycle_lock, NULL,
1213 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1214 		mutex_init(&tx_ring->tcb_head_lock, NULL,
1215 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1216 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
1217 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1218 	}
1219 
1220 	mutex_init(&ixgbe->gen_lock, NULL,
1221 	    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1222 
1223 	mutex_init(&ixgbe->watchdog_lock, NULL,
1224 	    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1225 }
1226 
1227 /*
1228  * ixgbe_destroy_locks - Destroy locks.
1229  */
1230 static void
1231 ixgbe_destroy_locks(ixgbe_t *ixgbe)
1232 {
1233 	ixgbe_rx_ring_t *rx_ring;
1234 	ixgbe_tx_ring_t *tx_ring;
1235 	int i;
1236 
1237 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1238 		rx_ring = &ixgbe->rx_rings[i];
1239 		mutex_destroy(&rx_ring->rx_lock);
1240 	}
1241 
1242 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1243 		tx_ring = &ixgbe->tx_rings[i];
1244 		mutex_destroy(&tx_ring->tx_lock);
1245 		mutex_destroy(&tx_ring->recycle_lock);
1246 		mutex_destroy(&tx_ring->tcb_head_lock);
1247 		mutex_destroy(&tx_ring->tcb_tail_lock);
1248 	}
1249 
1250 	mutex_destroy(&ixgbe->gen_lock);
1251 	mutex_destroy(&ixgbe->watchdog_lock);
1252 }
1253 
1254 /*
1255  * We need to try and determine which LED index in hardware corresponds to the
1256  * link/activity LED. This is the one that'll be overwritten when we perform
1257  * GLDv3 LED activity.
1258  */
1259 static void
1260 ixgbe_led_init(ixgbe_t *ixgbe)
1261 {
1262 	uint32_t reg, i;
1263 	struct ixgbe_hw *hw = &ixgbe->hw;
1264 
1265 	reg = IXGBE_READ_REG(hw, IXGBE_LEDCTL);
1266 	for (i = 0; i < 4; i++) {
1267 		if (((reg >> IXGBE_LED_MODE_SHIFT(i)) &
1268 		    IXGBE_LED_MODE_MASK_BASE) == IXGBE_LED_LINK_ACTIVE) {
1269 			ixgbe->ixgbe_led_index = i;
1270 			return;
1271 		}
1272 	}
1273 
1274 	/*
1275 	 * If we couldn't determine this, we use the default for various MACs
1276 	 * based on information Intel has inserted into other drivers over the
1277 	 * years.
1278 	 */
1279 	switch (hw->mac.type) {
1280 	case ixgbe_mac_X550EM_a:
1281 		ixgbe->ixgbe_led_index = 0;
1282 		break;
1283 	case ixgbe_mac_X550EM_x:
1284 		ixgbe->ixgbe_led_index = 1;
1285 		break;
1286 	default:
1287 		ixgbe->ixgbe_led_index = 2;
1288 		break;
1289 	}
1290 }
1291 
1292 static int
1293 ixgbe_resume(dev_info_t *devinfo)
1294 {
1295 	ixgbe_t *ixgbe;
1296 	int i;
1297 
1298 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
1299 	if (ixgbe == NULL)
1300 		return (DDI_FAILURE);
1301 
1302 	mutex_enter(&ixgbe->gen_lock);
1303 
1304 	if (ixgbe->ixgbe_state & IXGBE_STARTED) {
1305 		if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
1306 			mutex_exit(&ixgbe->gen_lock);
1307 			return (DDI_FAILURE);
1308 		}
1309 
1310 		/*
1311 		 * Enable and start the watchdog timer
1312 		 */
1313 		ixgbe_enable_watchdog_timer(ixgbe);
1314 	}
1315 
1316 	atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_SUSPENDED);
1317 
1318 	if (ixgbe->ixgbe_state & IXGBE_STARTED) {
1319 		for (i = 0; i < ixgbe->num_tx_rings; i++) {
1320 			mac_tx_ring_update(ixgbe->mac_hdl,
1321 			    ixgbe->tx_rings[i].ring_handle);
1322 		}
1323 	}
1324 
1325 	mutex_exit(&ixgbe->gen_lock);
1326 
1327 	return (DDI_SUCCESS);
1328 }
1329 
1330 static int
1331 ixgbe_suspend(dev_info_t *devinfo)
1332 {
1333 	ixgbe_t *ixgbe;
1334 
1335 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
1336 	if (ixgbe == NULL)
1337 		return (DDI_FAILURE);
1338 
1339 	mutex_enter(&ixgbe->gen_lock);
1340 
1341 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_SUSPENDED);
1342 	if (!(ixgbe->ixgbe_state & IXGBE_STARTED)) {
1343 		mutex_exit(&ixgbe->gen_lock);
1344 		return (DDI_SUCCESS);
1345 	}
1346 	ixgbe_stop(ixgbe, B_FALSE);
1347 
1348 	mutex_exit(&ixgbe->gen_lock);
1349 
1350 	/*
1351 	 * Disable and stop the watchdog timer
1352 	 */
1353 	ixgbe_disable_watchdog_timer(ixgbe);
1354 
1355 	return (DDI_SUCCESS);
1356 }
1357 
1358 /*
1359  * ixgbe_init - Initialize the device.
1360  */
1361 static int
1362 ixgbe_init(ixgbe_t *ixgbe)
1363 {
1364 	struct ixgbe_hw *hw = &ixgbe->hw;
1365 	u8 pbanum[IXGBE_PBANUM_LENGTH];
1366 	int rv;
1367 
1368 	mutex_enter(&ixgbe->gen_lock);
1369 
1370 	/*
1371 	 * Configure/Initialize hardware
1372 	 */
1373 	rv = ixgbe_init_hw(hw);
1374 	if (rv != IXGBE_SUCCESS) {
1375 		switch (rv) {
1376 
1377 		/*
1378 		 * The first three errors are not prohibitive to us progressing
1379 		 * further, and are maily advisory in nature. In the case of a
1380 		 * SFP module not being present or not deemed supported by the
1381 		 * common code, we adivse the operator of this fact but carry on
1382 		 * instead of failing hard, as SFPs can be inserted or replaced
1383 		 * while the driver is running. In the case of a unknown error,
1384 		 * we fail-hard, logging the reason and emitting a FMA event.
1385 		 */
1386 		case IXGBE_ERR_EEPROM_VERSION:
1387 			ixgbe_error(ixgbe,
1388 			    "This Intel 10Gb Ethernet device is pre-release and"
1389 			    " contains outdated firmware. Please contact your"
1390 			    " hardware vendor for a replacement.");
1391 			break;
1392 		case IXGBE_ERR_SFP_NOT_PRESENT:
1393 			ixgbe_error(ixgbe,
1394 			    "No SFP+ module detected on this interface. Please "
1395 			    "install a supported SFP+ module for this "
1396 			    "interface to become operational.");
1397 			break;
1398 		case IXGBE_ERR_SFP_NOT_SUPPORTED:
1399 			ixgbe_error(ixgbe,
1400 			    "Unsupported SFP+ module detected. Please replace "
1401 			    "it with a supported SFP+ module per Intel "
1402 			    "documentation, or bypass this check with "
1403 			    "allow_unsupported_sfp=1 in ixgbe.conf.");
1404 			break;
1405 		default:
1406 			ixgbe_error(ixgbe,
1407 			    "Failed to initialize hardware. ixgbe_init_hw "
1408 			    "returned %d", rv);
1409 			ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1410 			goto init_fail;
1411 		}
1412 	}
1413 
1414 	/*
1415 	 * Need to init eeprom before validating the checksum.
1416 	 */
1417 	if (ixgbe_init_eeprom_params(hw) < 0) {
1418 		ixgbe_error(ixgbe,
1419 		    "Unable to intitialize the eeprom interface.");
1420 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1421 		goto init_fail;
1422 	}
1423 
1424 	/*
1425 	 * NVM validation
1426 	 */
1427 	if (ixgbe_validate_eeprom_checksum(hw, NULL) < 0) {
1428 		/*
1429 		 * Some PCI-E parts fail the first check due to
1430 		 * the link being in sleep state.  Call it again,
1431 		 * if it fails a second time it's a real issue.
1432 		 */
1433 		if (ixgbe_validate_eeprom_checksum(hw, NULL) < 0) {
1434 			ixgbe_error(ixgbe,
1435 			    "Invalid NVM checksum. Please contact "
1436 			    "the vendor to update the NVM.");
1437 			ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1438 			goto init_fail;
1439 		}
1440 	}
1441 
1442 	/*
1443 	 * Setup default flow control thresholds - enable/disable
1444 	 * & flow control type is controlled by ixgbe.conf
1445 	 */
1446 	hw->fc.high_water[0] = DEFAULT_FCRTH;
1447 	hw->fc.low_water[0] = DEFAULT_FCRTL;
1448 	hw->fc.pause_time = DEFAULT_FCPAUSE;
1449 	hw->fc.send_xon = B_TRUE;
1450 
1451 	/*
1452 	 * Initialize flow control
1453 	 */
1454 	(void) ixgbe_start_hw(hw);
1455 
1456 	/*
1457 	 * Initialize link settings
1458 	 */
1459 	(void) ixgbe_driver_setup_link(ixgbe, B_FALSE);
1460 
1461 	/*
1462 	 * Initialize the chipset hardware
1463 	 */
1464 	if (ixgbe_chip_start(ixgbe) != IXGBE_SUCCESS) {
1465 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1466 		goto init_fail;
1467 	}
1468 
1469 	/*
1470 	 * Read identifying information and place in devinfo.
1471 	 */
1472 	pbanum[0] = '\0';
1473 	(void) ixgbe_read_pba_string(hw, pbanum, sizeof (pbanum));
1474 	if (*pbanum != '\0') {
1475 		(void) ddi_prop_update_string(DDI_DEV_T_NONE, ixgbe->dip,
1476 		    "printed-board-assembly", (char *)pbanum);
1477 	}
1478 
1479 	/*
1480 	 * Determine LED index.
1481 	 */
1482 	ixgbe_led_init(ixgbe);
1483 
1484 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1485 		goto init_fail;
1486 	}
1487 
1488 	mutex_exit(&ixgbe->gen_lock);
1489 	return (IXGBE_SUCCESS);
1490 
1491 init_fail:
1492 	/*
1493 	 * Reset PHY
1494 	 */
1495 	(void) ixgbe_reset_phy(hw);
1496 
1497 	mutex_exit(&ixgbe->gen_lock);
1498 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1499 	return (IXGBE_FAILURE);
1500 }
1501 
1502 /*
1503  * ixgbe_chip_start - Initialize and start the chipset hardware.
1504  */
1505 static int
1506 ixgbe_chip_start(ixgbe_t *ixgbe)
1507 {
1508 	struct ixgbe_hw *hw = &ixgbe->hw;
1509 	int i;
1510 
1511 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1512 
1513 	/*
1514 	 * Get the mac address
1515 	 * This function should handle SPARC case correctly.
1516 	 */
1517 	if (!ixgbe_find_mac_address(ixgbe)) {
1518 		ixgbe_error(ixgbe, "Failed to get the mac address");
1519 		return (IXGBE_FAILURE);
1520 	}
1521 
1522 	/*
1523 	 * Validate the mac address
1524 	 */
1525 	(void) ixgbe_init_rx_addrs(hw);
1526 	if (!is_valid_mac_addr(hw->mac.addr)) {
1527 		ixgbe_error(ixgbe, "Invalid mac address");
1528 		return (IXGBE_FAILURE);
1529 	}
1530 
1531 	/*
1532 	 * Re-enable relaxed ordering for performance.  It is disabled
1533 	 * by default in the hardware init.
1534 	 */
1535 	if (ixgbe->relax_order_enable == B_TRUE)
1536 		ixgbe_enable_relaxed_ordering(hw);
1537 
1538 	/*
1539 	 * Setup adapter interrupt vectors
1540 	 */
1541 	ixgbe_setup_adapter_vector(ixgbe);
1542 
1543 	/*
1544 	 * Initialize unicast addresses.
1545 	 */
1546 	ixgbe_init_unicst(ixgbe);
1547 
1548 	/*
1549 	 * Setup and initialize the mctable structures.
1550 	 */
1551 	ixgbe_setup_multicst(ixgbe);
1552 
1553 	/*
1554 	 * Set interrupt throttling rate
1555 	 */
1556 	for (i = 0; i < ixgbe->intr_cnt; i++) {
1557 		IXGBE_WRITE_REG(hw, IXGBE_EITR(i), ixgbe->intr_throttling[i]);
1558 	}
1559 
1560 	/*
1561 	 * Disable Wake-on-LAN
1562 	 */
1563 	IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
1564 
1565 	/*
1566 	 * Some adapters offer Energy Efficient Ethernet (EEE) support.
1567 	 * Due to issues with EEE in e1000g/igb, we disable this by default
1568 	 * as a precautionary measure.
1569 	 *
1570 	 * Currently, this is present on a number of the X550 family parts.
1571 	 */
1572 	(void) ixgbe_setup_eee(hw, B_FALSE);
1573 
1574 	/*
1575 	 * Turn on any present SFP Tx laser
1576 	 */
1577 	ixgbe_enable_tx_laser(hw);
1578 
1579 	/*
1580 	 * Power on the PHY
1581 	 */
1582 	(void) ixgbe_set_phy_power(hw, B_TRUE);
1583 
1584 	/*
1585 	 * Save the state of the PHY
1586 	 */
1587 	ixgbe_get_hw_state(ixgbe);
1588 
1589 	/*
1590 	 * Make sure driver has control
1591 	 */
1592 	ixgbe_get_driver_control(hw);
1593 
1594 	return (IXGBE_SUCCESS);
1595 }
1596 
1597 /*
1598  * ixgbe_chip_stop - Stop the chipset hardware
1599  */
1600 static void
1601 ixgbe_chip_stop(ixgbe_t *ixgbe)
1602 {
1603 	struct ixgbe_hw *hw = &ixgbe->hw;
1604 	int rv;
1605 
1606 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1607 
1608 	/*
1609 	 * Stop interupt generation and disable Tx unit
1610 	 */
1611 	hw->adapter_stopped = B_FALSE;
1612 	(void) ixgbe_stop_adapter(hw);
1613 
1614 	/*
1615 	 * Reset the chipset
1616 	 */
1617 	(void) ixgbe_reset_hw(hw);
1618 
1619 	/*
1620 	 * Reset PHY
1621 	 */
1622 	(void) ixgbe_reset_phy(hw);
1623 
1624 	/*
1625 	 * Enter LPLU (Low Power, Link Up) mode, if available. Avoid resetting
1626 	 * the PHY while doing so. Else, just power down the PHY.
1627 	 */
1628 	if (hw->phy.ops.enter_lplu != NULL) {
1629 		hw->phy.reset_disable = B_TRUE;
1630 		rv = hw->phy.ops.enter_lplu(hw);
1631 		if (rv != IXGBE_SUCCESS)
1632 			ixgbe_error(ixgbe, "Error while entering LPLU: %d", rv);
1633 		hw->phy.reset_disable = B_FALSE;
1634 	} else {
1635 		(void) ixgbe_set_phy_power(hw, B_FALSE);
1636 	}
1637 
1638 	/*
1639 	 * Turn off any present SFP Tx laser
1640 	 * Expected for health and safety reasons
1641 	 */
1642 	ixgbe_disable_tx_laser(hw);
1643 
1644 	/*
1645 	 * Tell firmware driver is no longer in control
1646 	 */
1647 	ixgbe_release_driver_control(hw);
1648 
1649 }
1650 
1651 /*
1652  * ixgbe_reset - Reset the chipset and re-start the driver.
1653  *
1654  * It involves stopping and re-starting the chipset,
1655  * and re-configuring the rx/tx rings.
1656  */
1657 static int
1658 ixgbe_reset(ixgbe_t *ixgbe)
1659 {
1660 	int i;
1661 
1662 	/*
1663 	 * Disable and stop the watchdog timer
1664 	 */
1665 	ixgbe_disable_watchdog_timer(ixgbe);
1666 
1667 	mutex_enter(&ixgbe->gen_lock);
1668 
1669 	ASSERT(ixgbe->ixgbe_state & IXGBE_STARTED);
1670 	atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_STARTED);
1671 
1672 	ixgbe_stop(ixgbe, B_FALSE);
1673 
1674 	if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
1675 		mutex_exit(&ixgbe->gen_lock);
1676 		return (IXGBE_FAILURE);
1677 	}
1678 
1679 	/*
1680 	 * After resetting, need to recheck the link status.
1681 	 */
1682 	ixgbe->link_check_complete = B_FALSE;
1683 	ixgbe->link_check_hrtime = gethrtime() +
1684 	    (IXGBE_LINK_UP_TIME * 100000000ULL);
1685 
1686 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_STARTED);
1687 
1688 	if (!(ixgbe->ixgbe_state & IXGBE_SUSPENDED)) {
1689 		for (i = 0; i < ixgbe->num_tx_rings; i++) {
1690 			mac_tx_ring_update(ixgbe->mac_hdl,
1691 			    ixgbe->tx_rings[i].ring_handle);
1692 		}
1693 	}
1694 
1695 	mutex_exit(&ixgbe->gen_lock);
1696 
1697 	/*
1698 	 * Enable and start the watchdog timer
1699 	 */
1700 	ixgbe_enable_watchdog_timer(ixgbe);
1701 
1702 	return (IXGBE_SUCCESS);
1703 }
1704 
1705 /*
1706  * ixgbe_tx_clean - Clean the pending transmit packets and DMA resources.
1707  */
1708 static void
1709 ixgbe_tx_clean(ixgbe_t *ixgbe)
1710 {
1711 	ixgbe_tx_ring_t *tx_ring;
1712 	tx_control_block_t *tcb;
1713 	link_list_t pending_list;
1714 	uint32_t desc_num;
1715 	int i, j;
1716 
1717 	LINK_LIST_INIT(&pending_list);
1718 
1719 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1720 		tx_ring = &ixgbe->tx_rings[i];
1721 
1722 		mutex_enter(&tx_ring->recycle_lock);
1723 
1724 		/*
1725 		 * Clean the pending tx data - the pending packets in the
1726 		 * work_list that have no chances to be transmitted again.
1727 		 *
1728 		 * We must ensure the chipset is stopped or the link is down
1729 		 * before cleaning the transmit packets.
1730 		 */
1731 		desc_num = 0;
1732 		for (j = 0; j < tx_ring->ring_size; j++) {
1733 			tcb = tx_ring->work_list[j];
1734 			if (tcb != NULL) {
1735 				desc_num += tcb->desc_num;
1736 
1737 				tx_ring->work_list[j] = NULL;
1738 
1739 				ixgbe_free_tcb(tcb);
1740 
1741 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1742 			}
1743 		}
1744 
1745 		if (desc_num > 0) {
1746 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1747 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1748 
1749 			/*
1750 			 * Reset the head and tail pointers of the tbd ring;
1751 			 * Reset the writeback head if it's enable.
1752 			 */
1753 			tx_ring->tbd_head = 0;
1754 			tx_ring->tbd_tail = 0;
1755 			if (ixgbe->tx_head_wb_enable)
1756 				*tx_ring->tbd_head_wb = 0;
1757 
1758 			IXGBE_WRITE_REG(&ixgbe->hw,
1759 			    IXGBE_TDH(tx_ring->index), 0);
1760 			IXGBE_WRITE_REG(&ixgbe->hw,
1761 			    IXGBE_TDT(tx_ring->index), 0);
1762 		}
1763 
1764 		mutex_exit(&tx_ring->recycle_lock);
1765 
1766 		/*
1767 		 * Add the tx control blocks in the pending list to
1768 		 * the free list.
1769 		 */
1770 		ixgbe_put_free_list(tx_ring, &pending_list);
1771 	}
1772 }
1773 
1774 /*
1775  * ixgbe_tx_drain - Drain the tx rings to allow pending packets to be
1776  * transmitted.
1777  */
1778 static boolean_t
1779 ixgbe_tx_drain(ixgbe_t *ixgbe)
1780 {
1781 	ixgbe_tx_ring_t *tx_ring;
1782 	boolean_t done;
1783 	int i, j;
1784 
1785 	/*
1786 	 * Wait for a specific time to allow pending tx packets
1787 	 * to be transmitted.
1788 	 *
1789 	 * Check the counter tbd_free to see if transmission is done.
1790 	 * No lock protection is needed here.
1791 	 *
1792 	 * Return B_TRUE if all pending packets have been transmitted;
1793 	 * Otherwise return B_FALSE;
1794 	 */
1795 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1796 
1797 		done = B_TRUE;
1798 		for (j = 0; j < ixgbe->num_tx_rings; j++) {
1799 			tx_ring = &ixgbe->tx_rings[j];
1800 			done = done &&
1801 			    (tx_ring->tbd_free == tx_ring->ring_size);
1802 		}
1803 
1804 		if (done)
1805 			break;
1806 
1807 		msec_delay(1);
1808 	}
1809 
1810 	return (done);
1811 }
1812 
1813 /*
1814  * ixgbe_rx_drain - Wait for all rx buffers to be released by upper layer.
1815  */
1816 static boolean_t
1817 ixgbe_rx_drain(ixgbe_t *ixgbe)
1818 {
1819 	boolean_t done = B_TRUE;
1820 	int i;
1821 
1822 	/*
1823 	 * Polling the rx free list to check if those rx buffers held by
1824 	 * the upper layer are released.
1825 	 *
1826 	 * Check the counter rcb_free to see if all pending buffers are
1827 	 * released. No lock protection is needed here.
1828 	 *
1829 	 * Return B_TRUE if all pending buffers have been released;
1830 	 * Otherwise return B_FALSE;
1831 	 */
1832 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1833 		done = (ixgbe->rcb_pending == 0);
1834 
1835 		if (done)
1836 			break;
1837 
1838 		msec_delay(1);
1839 	}
1840 
1841 	return (done);
1842 }
1843 
1844 /*
1845  * ixgbe_start - Start the driver/chipset.
1846  */
1847 int
1848 ixgbe_start(ixgbe_t *ixgbe, boolean_t alloc_buffer)
1849 {
1850 	struct ixgbe_hw *hw = &ixgbe->hw;
1851 	int i;
1852 
1853 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1854 
1855 	if (alloc_buffer) {
1856 		if (ixgbe_alloc_rx_data(ixgbe) != IXGBE_SUCCESS) {
1857 			ixgbe_error(ixgbe,
1858 			    "Failed to allocate software receive rings");
1859 			return (IXGBE_FAILURE);
1860 		}
1861 
1862 		/* Allocate buffers for all the rx/tx rings */
1863 		if (ixgbe_alloc_dma(ixgbe) != IXGBE_SUCCESS) {
1864 			ixgbe_error(ixgbe, "Failed to allocate DMA resource");
1865 			return (IXGBE_FAILURE);
1866 		}
1867 
1868 		ixgbe->tx_ring_init = B_TRUE;
1869 	} else {
1870 		ixgbe->tx_ring_init = B_FALSE;
1871 	}
1872 
1873 	for (i = 0; i < ixgbe->num_rx_rings; i++)
1874 		mutex_enter(&ixgbe->rx_rings[i].rx_lock);
1875 	for (i = 0; i < ixgbe->num_tx_rings; i++)
1876 		mutex_enter(&ixgbe->tx_rings[i].tx_lock);
1877 
1878 	/*
1879 	 * Start the chipset hardware
1880 	 */
1881 	if (ixgbe_chip_start(ixgbe) != IXGBE_SUCCESS) {
1882 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1883 		goto start_failure;
1884 	}
1885 
1886 	/*
1887 	 * Configure link now for X550
1888 	 *
1889 	 * X550 possesses a LPLU (Low-Power Link Up) mode which keeps the
1890 	 * resting state of the adapter at a 1Gb FDX speed. Prior to the X550,
1891 	 * the resting state of the link would be the maximum speed that
1892 	 * autonegotiation will allow (usually 10Gb, infrastructure allowing)
1893 	 * so we never bothered with explicitly setting the link to 10Gb as it
1894 	 * would already be at that state on driver attach. With X550, we must
1895 	 * trigger a re-negotiation of the link in order to switch from a LPLU
1896 	 * 1Gb link to 10Gb (cable and link partner permitting.)
1897 	 */
1898 	if (hw->mac.type == ixgbe_mac_X550 ||
1899 	    hw->mac.type == ixgbe_mac_X550EM_a ||
1900 	    hw->mac.type == ixgbe_mac_X550EM_x) {
1901 		(void) ixgbe_driver_setup_link(ixgbe, B_TRUE);
1902 		ixgbe_get_hw_state(ixgbe);
1903 	}
1904 
1905 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1906 		goto start_failure;
1907 	}
1908 
1909 	/*
1910 	 * Setup the rx/tx rings
1911 	 */
1912 	ixgbe_setup_rings(ixgbe);
1913 
1914 	/*
1915 	 * ixgbe_start() will be called when resetting, however if reset
1916 	 * happens, we need to clear the ERROR, STALL and OVERTEMP flags
1917 	 * before enabling the interrupts.
1918 	 */
1919 	atomic_and_32(&ixgbe->ixgbe_state, ~(IXGBE_ERROR
1920 	    | IXGBE_STALL| IXGBE_OVERTEMP));
1921 
1922 	/*
1923 	 * Enable adapter interrupts
1924 	 * The interrupts must be enabled after the driver state is START
1925 	 */
1926 	ixgbe_enable_adapter_interrupts(ixgbe);
1927 
1928 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1929 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1930 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1931 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1932 
1933 	return (IXGBE_SUCCESS);
1934 
1935 start_failure:
1936 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1937 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1938 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1939 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1940 
1941 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1942 
1943 	return (IXGBE_FAILURE);
1944 }
1945 
1946 /*
1947  * ixgbe_stop - Stop the driver/chipset.
1948  */
1949 void
1950 ixgbe_stop(ixgbe_t *ixgbe, boolean_t free_buffer)
1951 {
1952 	int i;
1953 
1954 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1955 
1956 	/*
1957 	 * Disable the adapter interrupts
1958 	 */
1959 	ixgbe_disable_adapter_interrupts(ixgbe);
1960 
1961 	/*
1962 	 * Drain the pending tx packets
1963 	 */
1964 	(void) ixgbe_tx_drain(ixgbe);
1965 
1966 	for (i = 0; i < ixgbe->num_rx_rings; i++)
1967 		mutex_enter(&ixgbe->rx_rings[i].rx_lock);
1968 	for (i = 0; i < ixgbe->num_tx_rings; i++)
1969 		mutex_enter(&ixgbe->tx_rings[i].tx_lock);
1970 
1971 	/*
1972 	 * Stop the chipset hardware
1973 	 */
1974 	ixgbe_chip_stop(ixgbe);
1975 
1976 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1977 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1978 	}
1979 
1980 	/*
1981 	 * Clean the pending tx data/resources
1982 	 */
1983 	ixgbe_tx_clean(ixgbe);
1984 
1985 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1986 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1987 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1988 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1989 
1990 	if (ixgbe->link_state == LINK_STATE_UP) {
1991 		ixgbe->link_state = LINK_STATE_UNKNOWN;
1992 		mac_link_update(ixgbe->mac_hdl, ixgbe->link_state);
1993 	}
1994 
1995 	if (free_buffer) {
1996 		/*
1997 		 * Release the DMA/memory resources of rx/tx rings
1998 		 */
1999 		ixgbe_free_dma(ixgbe);
2000 		ixgbe_free_rx_data(ixgbe);
2001 	}
2002 }
2003 
2004 /*
2005  * ixgbe_cbfunc - Driver interface for generic DDI callbacks
2006  */
2007 /* ARGSUSED */
2008 static int
2009 ixgbe_cbfunc(dev_info_t *dip, ddi_cb_action_t cbaction, void *cbarg,
2010     void *arg1, void *arg2)
2011 {
2012 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
2013 
2014 	switch (cbaction) {
2015 	/* IRM callback */
2016 	int count;
2017 	case DDI_CB_INTR_ADD:
2018 	case DDI_CB_INTR_REMOVE:
2019 		count = (int)(uintptr_t)cbarg;
2020 		ASSERT(ixgbe->intr_type == DDI_INTR_TYPE_MSIX);
2021 		DTRACE_PROBE2(ixgbe__irm__callback, int, count,
2022 		    int, ixgbe->intr_cnt);
2023 		if (ixgbe_intr_adjust(ixgbe, cbaction, count) !=
2024 		    DDI_SUCCESS) {
2025 			ixgbe_error(ixgbe,
2026 			    "IRM CB: Failed to adjust interrupts");
2027 			goto cb_fail;
2028 		}
2029 		break;
2030 	default:
2031 		IXGBE_DEBUGLOG_1(ixgbe, "DDI CB: action 0x%x NOT supported",
2032 		    cbaction);
2033 		return (DDI_ENOTSUP);
2034 	}
2035 	return (DDI_SUCCESS);
2036 cb_fail:
2037 	return (DDI_FAILURE);
2038 }
2039 
2040 /*
2041  * ixgbe_intr_adjust - Adjust interrupt to respond to IRM request.
2042  */
2043 static int
2044 ixgbe_intr_adjust(ixgbe_t *ixgbe, ddi_cb_action_t cbaction, int count)
2045 {
2046 	int i, rc, actual;
2047 
2048 	if (count == 0)
2049 		return (DDI_SUCCESS);
2050 
2051 	if ((cbaction == DDI_CB_INTR_ADD &&
2052 	    ixgbe->intr_cnt + count > ixgbe->intr_cnt_max) ||
2053 	    (cbaction == DDI_CB_INTR_REMOVE &&
2054 	    ixgbe->intr_cnt - count < ixgbe->intr_cnt_min))
2055 		return (DDI_FAILURE);
2056 
2057 	if (!(ixgbe->ixgbe_state & IXGBE_STARTED)) {
2058 		return (DDI_FAILURE);
2059 	}
2060 
2061 	for (i = 0; i < ixgbe->num_rx_rings; i++)
2062 		mac_ring_intr_set(ixgbe->rx_rings[i].ring_handle, NULL);
2063 	for (i = 0; i < ixgbe->num_tx_rings; i++)
2064 		mac_ring_intr_set(ixgbe->tx_rings[i].ring_handle, NULL);
2065 
2066 	mutex_enter(&ixgbe->gen_lock);
2067 	ixgbe->ixgbe_state &= ~IXGBE_STARTED;
2068 	ixgbe->ixgbe_state |= IXGBE_INTR_ADJUST;
2069 	ixgbe->ixgbe_state |= IXGBE_SUSPENDED;
2070 	mac_link_update(ixgbe->mac_hdl, LINK_STATE_UNKNOWN);
2071 
2072 	ixgbe_stop(ixgbe, B_FALSE);
2073 	/*
2074 	 * Disable interrupts
2075 	 */
2076 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
2077 		rc = ixgbe_disable_intrs(ixgbe);
2078 		ASSERT(rc == IXGBE_SUCCESS);
2079 	}
2080 	ixgbe->attach_progress &= ~ATTACH_PROGRESS_ENABLE_INTR;
2081 
2082 	/*
2083 	 * Remove interrupt handlers
2084 	 */
2085 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
2086 		ixgbe_rem_intr_handlers(ixgbe);
2087 	}
2088 	ixgbe->attach_progress &= ~ATTACH_PROGRESS_ADD_INTR;
2089 
2090 	/*
2091 	 * Clear vect_map
2092 	 */
2093 	bzero(&ixgbe->vect_map, sizeof (ixgbe->vect_map));
2094 	switch (cbaction) {
2095 	case DDI_CB_INTR_ADD:
2096 		rc = ddi_intr_alloc(ixgbe->dip, ixgbe->htable,
2097 		    DDI_INTR_TYPE_MSIX, ixgbe->intr_cnt, count, &actual,
2098 		    DDI_INTR_ALLOC_NORMAL);
2099 		if (rc != DDI_SUCCESS || actual != count) {
2100 			ixgbe_log(ixgbe, "Adjust interrupts failed."
2101 			    "return: %d, irm cb size: %d, actual: %d",
2102 			    rc, count, actual);
2103 			goto intr_adjust_fail;
2104 		}
2105 		ixgbe->intr_cnt += count;
2106 		break;
2107 
2108 	case DDI_CB_INTR_REMOVE:
2109 		for (i = ixgbe->intr_cnt - count;
2110 		    i < ixgbe->intr_cnt; i ++) {
2111 			rc = ddi_intr_free(ixgbe->htable[i]);
2112 			ixgbe->htable[i] = NULL;
2113 			if (rc != DDI_SUCCESS) {
2114 				ixgbe_log(ixgbe, "Adjust interrupts failed."
2115 				    "return: %d, irm cb size: %d, actual: %d",
2116 				    rc, count, actual);
2117 				goto intr_adjust_fail;
2118 			}
2119 		}
2120 		ixgbe->intr_cnt -= count;
2121 		break;
2122 	}
2123 
2124 	/*
2125 	 * Get priority for first vector, assume remaining are all the same
2126 	 */
2127 	rc = ddi_intr_get_pri(ixgbe->htable[0], &ixgbe->intr_pri);
2128 	if (rc != DDI_SUCCESS) {
2129 		ixgbe_log(ixgbe,
2130 		    "Get interrupt priority failed: %d", rc);
2131 		goto intr_adjust_fail;
2132 	}
2133 	rc = ddi_intr_get_cap(ixgbe->htable[0], &ixgbe->intr_cap);
2134 	if (rc != DDI_SUCCESS) {
2135 		ixgbe_log(ixgbe, "Get interrupt cap failed: %d", rc);
2136 		goto intr_adjust_fail;
2137 	}
2138 	ixgbe->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
2139 
2140 	/*
2141 	 * Map rings to interrupt vectors
2142 	 */
2143 	if (ixgbe_map_intrs_to_vectors(ixgbe) != IXGBE_SUCCESS) {
2144 		ixgbe_error(ixgbe,
2145 		    "IRM CB: Failed to map interrupts to vectors");
2146 		goto intr_adjust_fail;
2147 	}
2148 
2149 	/*
2150 	 * Add interrupt handlers
2151 	 */
2152 	if (ixgbe_add_intr_handlers(ixgbe) != IXGBE_SUCCESS) {
2153 		ixgbe_error(ixgbe, "IRM CB: Failed to add interrupt handlers");
2154 		goto intr_adjust_fail;
2155 	}
2156 	ixgbe->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
2157 
2158 	/*
2159 	 * Now that mutex locks are initialized, and the chip is also
2160 	 * initialized, enable interrupts.
2161 	 */
2162 	if (ixgbe_enable_intrs(ixgbe) != IXGBE_SUCCESS) {
2163 		ixgbe_error(ixgbe, "IRM CB: Failed to enable DDI interrupts");
2164 		goto intr_adjust_fail;
2165 	}
2166 	ixgbe->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
2167 	if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
2168 		ixgbe_error(ixgbe, "IRM CB: Failed to start");
2169 		goto intr_adjust_fail;
2170 	}
2171 	ixgbe->ixgbe_state &= ~IXGBE_INTR_ADJUST;
2172 	ixgbe->ixgbe_state &= ~IXGBE_SUSPENDED;
2173 	ixgbe->ixgbe_state |= IXGBE_STARTED;
2174 	mutex_exit(&ixgbe->gen_lock);
2175 
2176 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2177 		mac_ring_intr_set(ixgbe->rx_rings[i].ring_handle,
2178 		    ixgbe->htable[ixgbe->rx_rings[i].intr_vector]);
2179 	}
2180 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2181 		mac_ring_intr_set(ixgbe->tx_rings[i].ring_handle,
2182 		    ixgbe->htable[ixgbe->tx_rings[i].intr_vector]);
2183 	}
2184 
2185 	/* Wakeup all Tx rings */
2186 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2187 		mac_tx_ring_update(ixgbe->mac_hdl,
2188 		    ixgbe->tx_rings[i].ring_handle);
2189 	}
2190 
2191 	IXGBE_DEBUGLOG_3(ixgbe,
2192 	    "IRM CB: interrupts new value: 0x%x(0x%x:0x%x).",
2193 	    ixgbe->intr_cnt, ixgbe->intr_cnt_min, ixgbe->intr_cnt_max);
2194 	return (DDI_SUCCESS);
2195 
2196 intr_adjust_fail:
2197 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
2198 	mutex_exit(&ixgbe->gen_lock);
2199 	return (DDI_FAILURE);
2200 }
2201 
2202 /*
2203  * ixgbe_intr_cb_register - Register interrupt callback function.
2204  */
2205 static int
2206 ixgbe_intr_cb_register(ixgbe_t *ixgbe)
2207 {
2208 	if (ddi_cb_register(ixgbe->dip, DDI_CB_FLAG_INTR, ixgbe_cbfunc,
2209 	    ixgbe, NULL, &ixgbe->cb_hdl) != DDI_SUCCESS) {
2210 		return (IXGBE_FAILURE);
2211 	}
2212 	IXGBE_DEBUGLOG_0(ixgbe, "Interrupt callback function registered.");
2213 	return (IXGBE_SUCCESS);
2214 }
2215 
2216 /*
2217  * ixgbe_alloc_rings - Allocate memory space for rx/tx rings.
2218  */
2219 static int
2220 ixgbe_alloc_rings(ixgbe_t *ixgbe)
2221 {
2222 	/*
2223 	 * Allocate memory space for rx rings
2224 	 */
2225 	ixgbe->rx_rings = kmem_zalloc(
2226 	    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings,
2227 	    KM_NOSLEEP);
2228 
2229 	if (ixgbe->rx_rings == NULL) {
2230 		return (IXGBE_FAILURE);
2231 	}
2232 
2233 	/*
2234 	 * Allocate memory space for tx rings
2235 	 */
2236 	ixgbe->tx_rings = kmem_zalloc(
2237 	    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings,
2238 	    KM_NOSLEEP);
2239 
2240 	if (ixgbe->tx_rings == NULL) {
2241 		kmem_free(ixgbe->rx_rings,
2242 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
2243 		ixgbe->rx_rings = NULL;
2244 		return (IXGBE_FAILURE);
2245 	}
2246 
2247 	/*
2248 	 * Allocate memory space for rx ring groups
2249 	 */
2250 	ixgbe->rx_groups = kmem_zalloc(
2251 	    sizeof (ixgbe_rx_group_t) * ixgbe->num_rx_groups,
2252 	    KM_NOSLEEP);
2253 
2254 	if (ixgbe->rx_groups == NULL) {
2255 		kmem_free(ixgbe->rx_rings,
2256 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
2257 		kmem_free(ixgbe->tx_rings,
2258 		    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings);
2259 		ixgbe->rx_rings = NULL;
2260 		ixgbe->tx_rings = NULL;
2261 		return (IXGBE_FAILURE);
2262 	}
2263 
2264 	return (IXGBE_SUCCESS);
2265 }
2266 
2267 /*
2268  * ixgbe_free_rings - Free the memory space of rx/tx rings.
2269  */
2270 static void
2271 ixgbe_free_rings(ixgbe_t *ixgbe)
2272 {
2273 	if (ixgbe->rx_rings != NULL) {
2274 		kmem_free(ixgbe->rx_rings,
2275 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
2276 		ixgbe->rx_rings = NULL;
2277 	}
2278 
2279 	if (ixgbe->tx_rings != NULL) {
2280 		kmem_free(ixgbe->tx_rings,
2281 		    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings);
2282 		ixgbe->tx_rings = NULL;
2283 	}
2284 
2285 	if (ixgbe->rx_groups != NULL) {
2286 		kmem_free(ixgbe->rx_groups,
2287 		    sizeof (ixgbe_rx_group_t) * ixgbe->num_rx_groups);
2288 		ixgbe->rx_groups = NULL;
2289 	}
2290 }
2291 
2292 static int
2293 ixgbe_alloc_rx_data(ixgbe_t *ixgbe)
2294 {
2295 	ixgbe_rx_ring_t *rx_ring;
2296 	int i;
2297 
2298 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2299 		rx_ring = &ixgbe->rx_rings[i];
2300 		if (ixgbe_alloc_rx_ring_data(rx_ring) != IXGBE_SUCCESS)
2301 			goto alloc_rx_rings_failure;
2302 	}
2303 	return (IXGBE_SUCCESS);
2304 
2305 alloc_rx_rings_failure:
2306 	ixgbe_free_rx_data(ixgbe);
2307 	return (IXGBE_FAILURE);
2308 }
2309 
2310 static void
2311 ixgbe_free_rx_data(ixgbe_t *ixgbe)
2312 {
2313 	ixgbe_rx_ring_t *rx_ring;
2314 	ixgbe_rx_data_t *rx_data;
2315 	int i;
2316 
2317 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2318 		rx_ring = &ixgbe->rx_rings[i];
2319 
2320 		mutex_enter(&ixgbe->rx_pending_lock);
2321 		rx_data = rx_ring->rx_data;
2322 
2323 		if (rx_data != NULL) {
2324 			rx_data->flag |= IXGBE_RX_STOPPED;
2325 
2326 			if (rx_data->rcb_pending == 0) {
2327 				ixgbe_free_rx_ring_data(rx_data);
2328 				rx_ring->rx_data = NULL;
2329 			}
2330 		}
2331 
2332 		mutex_exit(&ixgbe->rx_pending_lock);
2333 	}
2334 }
2335 
2336 /*
2337  * ixgbe_setup_rings - Setup rx/tx rings.
2338  */
2339 static void
2340 ixgbe_setup_rings(ixgbe_t *ixgbe)
2341 {
2342 	/*
2343 	 * Setup the rx/tx rings, including the following:
2344 	 *
2345 	 * 1. Setup the descriptor ring and the control block buffers;
2346 	 * 2. Initialize necessary registers for receive/transmit;
2347 	 * 3. Initialize software pointers/parameters for receive/transmit;
2348 	 */
2349 	ixgbe_setup_rx(ixgbe);
2350 
2351 	ixgbe_setup_tx(ixgbe);
2352 }
2353 
2354 static void
2355 ixgbe_setup_rx_ring(ixgbe_rx_ring_t *rx_ring)
2356 {
2357 	ixgbe_t *ixgbe = rx_ring->ixgbe;
2358 	ixgbe_rx_data_t *rx_data = rx_ring->rx_data;
2359 	struct ixgbe_hw *hw = &ixgbe->hw;
2360 	rx_control_block_t *rcb;
2361 	union ixgbe_adv_rx_desc	*rbd;
2362 	uint32_t size;
2363 	uint32_t buf_low;
2364 	uint32_t buf_high;
2365 	uint32_t reg_val;
2366 	int i;
2367 
2368 	ASSERT(mutex_owned(&rx_ring->rx_lock));
2369 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2370 
2371 	for (i = 0; i < ixgbe->rx_ring_size; i++) {
2372 		rcb = rx_data->work_list[i];
2373 		rbd = &rx_data->rbd_ring[i];
2374 
2375 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
2376 		rbd->read.hdr_addr = 0;
2377 	}
2378 
2379 	/*
2380 	 * Initialize the length register
2381 	 */
2382 	size = rx_data->ring_size * sizeof (union ixgbe_adv_rx_desc);
2383 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(rx_ring->hw_index), size);
2384 
2385 	/*
2386 	 * Initialize the base address registers
2387 	 */
2388 	buf_low = (uint32_t)rx_data->rbd_area.dma_address;
2389 	buf_high = (uint32_t)(rx_data->rbd_area.dma_address >> 32);
2390 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(rx_ring->hw_index), buf_high);
2391 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(rx_ring->hw_index), buf_low);
2392 
2393 	/*
2394 	 * Setup head & tail pointers
2395 	 */
2396 	IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->hw_index),
2397 	    rx_data->ring_size - 1);
2398 	IXGBE_WRITE_REG(hw, IXGBE_RDH(rx_ring->hw_index), 0);
2399 
2400 	rx_data->rbd_next = 0;
2401 	rx_data->lro_first = 0;
2402 
2403 	/*
2404 	 * Setup the Receive Descriptor Control Register (RXDCTL)
2405 	 * PTHRESH=32 descriptors (half the internal cache)
2406 	 * HTHRESH=0 descriptors (to minimize latency on fetch)
2407 	 * WTHRESH defaults to 1 (writeback each descriptor)
2408 	 */
2409 	reg_val = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rx_ring->hw_index));
2410 	reg_val |= IXGBE_RXDCTL_ENABLE;	/* enable queue */
2411 
2412 	/* Not a valid value for 82599, X540 or X550 */
2413 	if (hw->mac.type == ixgbe_mac_82598EB) {
2414 		reg_val |= 0x0020;	/* pthresh */
2415 	}
2416 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rx_ring->hw_index), reg_val);
2417 
2418 	if (hw->mac.type == ixgbe_mac_82599EB ||
2419 	    hw->mac.type == ixgbe_mac_X540 ||
2420 	    hw->mac.type == ixgbe_mac_X550 ||
2421 	    hw->mac.type == ixgbe_mac_X550EM_x ||
2422 	    hw->mac.type == ixgbe_mac_X550EM_a) {
2423 		reg_val = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2424 		reg_val |= (IXGBE_RDRXCTL_CRCSTRIP | IXGBE_RDRXCTL_AGGDIS);
2425 		IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, reg_val);
2426 	}
2427 
2428 	/*
2429 	 * Setup the Split and Replication Receive Control Register.
2430 	 * Set the rx buffer size and the advanced descriptor type.
2431 	 */
2432 	reg_val = (ixgbe->rx_buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) |
2433 	    IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2434 	reg_val |= IXGBE_SRRCTL_DROP_EN;
2435 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rx_ring->hw_index), reg_val);
2436 }
2437 
2438 static void
2439 ixgbe_setup_rx(ixgbe_t *ixgbe)
2440 {
2441 	ixgbe_rx_ring_t *rx_ring;
2442 	struct ixgbe_hw *hw = &ixgbe->hw;
2443 	uint32_t reg_val;
2444 	uint32_t i;
2445 	uint32_t psrtype_rss_bit;
2446 
2447 	/*
2448 	 * Ensure that Rx is disabled while setting up
2449 	 * the Rx unit and Rx descriptor ring(s)
2450 	 */
2451 	ixgbe_disable_rx(hw);
2452 
2453 	/* PSRTYPE must be configured for 82599 */
2454 	if (ixgbe->classify_mode != IXGBE_CLASSIFY_VMDQ &&
2455 	    ixgbe->classify_mode != IXGBE_CLASSIFY_VMDQ_RSS) {
2456 		reg_val = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
2457 		    IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR;
2458 		reg_val |= IXGBE_PSRTYPE_L2HDR;
2459 		reg_val |= 0x80000000;
2460 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), reg_val);
2461 	} else {
2462 		if (ixgbe->num_rx_groups > 32) {
2463 			psrtype_rss_bit = 0x20000000;
2464 		} else {
2465 			psrtype_rss_bit = 0x40000000;
2466 		}
2467 		for (i = 0; i < ixgbe->capab->max_rx_grp_num; i++) {
2468 			reg_val = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
2469 			    IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR;
2470 			reg_val |= IXGBE_PSRTYPE_L2HDR;
2471 			reg_val |= psrtype_rss_bit;
2472 			IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(i), reg_val);
2473 		}
2474 	}
2475 
2476 	/*
2477 	 * Set filter control in FCTRL to determine types of packets are passed
2478 	 * up to the driver.
2479 	 * - Pass broadcast packets.
2480 	 * - Do not pass flow control pause frames (82598-specific)
2481 	 */
2482 	reg_val = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2483 	reg_val |= IXGBE_FCTRL_BAM; /* Broadcast Accept Mode */
2484 	if (hw->mac.type == ixgbe_mac_82598EB) {
2485 		reg_val |= IXGBE_FCTRL_DPF; /* Discard Pause Frames */
2486 	}
2487 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, reg_val);
2488 
2489 	/*
2490 	 * Hardware checksum settings
2491 	 */
2492 	if (ixgbe->rx_hcksum_enable) {
2493 		reg_val = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2494 		reg_val |= IXGBE_RXCSUM_IPPCSE;	/* IP checksum */
2495 		IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, reg_val);
2496 	}
2497 
2498 	/*
2499 	 * Setup VMDq and RSS for multiple receive queues
2500 	 */
2501 	switch (ixgbe->classify_mode) {
2502 	case IXGBE_CLASSIFY_RSS:
2503 		/*
2504 		 * One group, only RSS is needed when more than
2505 		 * one ring enabled.
2506 		 */
2507 		ixgbe_setup_rss(ixgbe);
2508 		break;
2509 
2510 	case IXGBE_CLASSIFY_VMDQ:
2511 		/*
2512 		 * Multiple groups, each group has one ring,
2513 		 * only VMDq is needed.
2514 		 */
2515 		ixgbe_setup_vmdq(ixgbe);
2516 		break;
2517 
2518 	case IXGBE_CLASSIFY_VMDQ_RSS:
2519 		/*
2520 		 * Multiple groups and multiple rings, both
2521 		 * VMDq and RSS are needed.
2522 		 */
2523 		ixgbe_setup_vmdq_rss(ixgbe);
2524 		break;
2525 
2526 	default:
2527 		break;
2528 	}
2529 
2530 	/*
2531 	 * Enable the receive unit.  This must be done after filter
2532 	 * control is set in FCTRL. On 82598, we disable the descriptor monitor.
2533 	 * 82598 is the only adapter which defines this RXCTRL option.
2534 	 */
2535 	reg_val = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2536 	if (hw->mac.type == ixgbe_mac_82598EB)
2537 		reg_val |= IXGBE_RXCTRL_DMBYPS; /* descriptor monitor bypass */
2538 	reg_val |= IXGBE_RXCTRL_RXEN;
2539 	(void) ixgbe_enable_rx_dma(hw, reg_val);
2540 
2541 	/*
2542 	 * ixgbe_setup_rx_ring must be called after configuring RXCTRL
2543 	 */
2544 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2545 		rx_ring = &ixgbe->rx_rings[i];
2546 		ixgbe_setup_rx_ring(rx_ring);
2547 	}
2548 
2549 	/*
2550 	 * The 82598 controller gives us the RNBC (Receive No Buffer
2551 	 * Count) register to determine the number of frames dropped
2552 	 * due to no available descriptors on the destination queue.
2553 	 * However, this register was removed starting with 82599 and
2554 	 * it was replaced with the RQSMR/QPRDC registers. The nice
2555 	 * thing about the new registers is that they allow you to map
2556 	 * groups of queues to specific stat registers. The bad thing
2557 	 * is there are only 16 slots in the stat registers, so this
2558 	 * won't work when we have 32 Rx groups. Instead, we map all
2559 	 * queues to the zero slot of the stat registers, giving us a
2560 	 * global counter at QPRDC[0] (with the equivalent semantics
2561 	 * of RNBC). Perhaps future controllers will have more slots
2562 	 * and we can implement per-group counters.
2563 	 */
2564 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2565 		uint32_t index = ixgbe->rx_rings[i].hw_index;
2566 		IXGBE_WRITE_REG(hw, IXGBE_RQSMR(index >> 2), 0);
2567 	}
2568 
2569 	/*
2570 	 * The Max Frame Size in MHADD/MAXFRS will be internally increased
2571 	 * by four bytes if the packet has a VLAN field, so includes MTU,
2572 	 * ethernet header and frame check sequence.
2573 	 * Register is MAXFRS in 82599.
2574 	 */
2575 	reg_val = IXGBE_READ_REG(hw, IXGBE_MHADD);
2576 	reg_val &= ~IXGBE_MHADD_MFS_MASK;
2577 	reg_val |= (ixgbe->default_mtu + sizeof (struct ether_header)
2578 	    + ETHERFCSL) << IXGBE_MHADD_MFS_SHIFT;
2579 	IXGBE_WRITE_REG(hw, IXGBE_MHADD, reg_val);
2580 
2581 	/*
2582 	 * Setup Jumbo Frame enable bit
2583 	 */
2584 	reg_val = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2585 	if (ixgbe->default_mtu > ETHERMTU)
2586 		reg_val |= IXGBE_HLREG0_JUMBOEN;
2587 	else
2588 		reg_val &= ~IXGBE_HLREG0_JUMBOEN;
2589 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, reg_val);
2590 
2591 	/*
2592 	 * Setup RSC for multiple receive queues.
2593 	 */
2594 	if (ixgbe->lro_enable) {
2595 		for (i = 0; i < ixgbe->num_rx_rings; i++) {
2596 			/*
2597 			 * Make sure rx_buf_size * MAXDESC not greater
2598 			 * than 65535.
2599 			 * Intel recommends 4 for MAXDESC field value.
2600 			 */
2601 			reg_val = IXGBE_READ_REG(hw, IXGBE_RSCCTL(i));
2602 			reg_val |= IXGBE_RSCCTL_RSCEN;
2603 			if (ixgbe->rx_buf_size == IXGBE_PKG_BUF_16k)
2604 				reg_val |= IXGBE_RSCCTL_MAXDESC_1;
2605 			else
2606 				reg_val |= IXGBE_RSCCTL_MAXDESC_4;
2607 			IXGBE_WRITE_REG(hw,  IXGBE_RSCCTL(i), reg_val);
2608 		}
2609 
2610 		reg_val = IXGBE_READ_REG(hw, IXGBE_RSCDBU);
2611 		reg_val |= IXGBE_RSCDBU_RSCACKDIS;
2612 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU, reg_val);
2613 
2614 		reg_val = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2615 		reg_val |= IXGBE_RDRXCTL_RSCACKC;
2616 		reg_val |= IXGBE_RDRXCTL_FCOE_WRFIX;
2617 		reg_val &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2618 
2619 		IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, reg_val);
2620 	}
2621 }
2622 
2623 static void
2624 ixgbe_setup_tx_ring(ixgbe_tx_ring_t *tx_ring)
2625 {
2626 	ixgbe_t *ixgbe = tx_ring->ixgbe;
2627 	struct ixgbe_hw *hw = &ixgbe->hw;
2628 	uint32_t size;
2629 	uint32_t buf_low;
2630 	uint32_t buf_high;
2631 	uint32_t reg_val;
2632 
2633 	ASSERT(mutex_owned(&tx_ring->tx_lock));
2634 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2635 
2636 	/*
2637 	 * Initialize the length register
2638 	 */
2639 	size = tx_ring->ring_size * sizeof (union ixgbe_adv_tx_desc);
2640 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(tx_ring->index), size);
2641 
2642 	/*
2643 	 * Initialize the base address registers
2644 	 */
2645 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
2646 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
2647 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(tx_ring->index), buf_low);
2648 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(tx_ring->index), buf_high);
2649 
2650 	/*
2651 	 * Setup head & tail pointers
2652 	 */
2653 	IXGBE_WRITE_REG(hw, IXGBE_TDH(tx_ring->index), 0);
2654 	IXGBE_WRITE_REG(hw, IXGBE_TDT(tx_ring->index), 0);
2655 
2656 	/*
2657 	 * Setup head write-back
2658 	 */
2659 	if (ixgbe->tx_head_wb_enable) {
2660 		/*
2661 		 * The memory of the head write-back is allocated using
2662 		 * the extra tbd beyond the tail of the tbd ring.
2663 		 */
2664 		tx_ring->tbd_head_wb = (uint32_t *)
2665 		    ((uintptr_t)tx_ring->tbd_area.address + size);
2666 		*tx_ring->tbd_head_wb = 0;
2667 
2668 		buf_low = (uint32_t)
2669 		    (tx_ring->tbd_area.dma_address + size);
2670 		buf_high = (uint32_t)
2671 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
2672 
2673 		/* Set the head write-back enable bit */
2674 		buf_low |= IXGBE_TDWBAL_HEAD_WB_ENABLE;
2675 
2676 		IXGBE_WRITE_REG(hw, IXGBE_TDWBAL(tx_ring->index), buf_low);
2677 		IXGBE_WRITE_REG(hw, IXGBE_TDWBAH(tx_ring->index), buf_high);
2678 
2679 		/*
2680 		 * Turn off relaxed ordering for head write back or it will
2681 		 * cause problems with the tx recycling
2682 		 */
2683 
2684 		reg_val = (hw->mac.type == ixgbe_mac_82598EB) ?
2685 		    IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(tx_ring->index)) :
2686 		    IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(tx_ring->index));
2687 		reg_val &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
2688 		if (hw->mac.type == ixgbe_mac_82598EB) {
2689 			IXGBE_WRITE_REG(hw,
2690 			    IXGBE_DCA_TXCTRL(tx_ring->index), reg_val);
2691 		} else {
2692 			IXGBE_WRITE_REG(hw,
2693 			    IXGBE_DCA_TXCTRL_82599(tx_ring->index), reg_val);
2694 		}
2695 	} else {
2696 		tx_ring->tbd_head_wb = NULL;
2697 	}
2698 
2699 	tx_ring->tbd_head = 0;
2700 	tx_ring->tbd_tail = 0;
2701 	tx_ring->tbd_free = tx_ring->ring_size;
2702 
2703 	if (ixgbe->tx_ring_init == B_TRUE) {
2704 		tx_ring->tcb_head = 0;
2705 		tx_ring->tcb_tail = 0;
2706 		tx_ring->tcb_free = tx_ring->free_list_size;
2707 	}
2708 
2709 	/*
2710 	 * Initialize the s/w context structure
2711 	 */
2712 	bzero(&tx_ring->tx_context, sizeof (ixgbe_tx_context_t));
2713 }
2714 
2715 static void
2716 ixgbe_setup_tx(ixgbe_t *ixgbe)
2717 {
2718 	struct ixgbe_hw *hw = &ixgbe->hw;
2719 	ixgbe_tx_ring_t *tx_ring;
2720 	uint32_t reg_val;
2721 	int i;
2722 
2723 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2724 		tx_ring = &ixgbe->tx_rings[i];
2725 		ixgbe_setup_tx_ring(tx_ring);
2726 	}
2727 
2728 	/*
2729 	 * Setup the per-ring statistics mapping. We map all Tx queues
2730 	 * to slot 0 to stay consistent with Rx.
2731 	 */
2732 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2733 		switch (hw->mac.type) {
2734 		case ixgbe_mac_82598EB:
2735 			IXGBE_WRITE_REG(hw, IXGBE_TQSMR(i >> 2), 0);
2736 			break;
2737 
2738 		default:
2739 			IXGBE_WRITE_REG(hw, IXGBE_TQSM(i >> 2), 0);
2740 			break;
2741 		}
2742 	}
2743 
2744 	/*
2745 	 * Enable CRC appending and TX padding (for short tx frames)
2746 	 */
2747 	reg_val = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2748 	reg_val |= IXGBE_HLREG0_TXCRCEN | IXGBE_HLREG0_TXPADEN;
2749 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, reg_val);
2750 
2751 	/*
2752 	 * enable DMA for 82599, X540 and X550 parts
2753 	 */
2754 	if (hw->mac.type == ixgbe_mac_82599EB ||
2755 	    hw->mac.type == ixgbe_mac_X540 ||
2756 	    hw->mac.type == ixgbe_mac_X550 ||
2757 	    hw->mac.type == ixgbe_mac_X550EM_x ||
2758 	    hw->mac.type == ixgbe_mac_X550EM_a) {
2759 		/* DMATXCTL.TE must be set after all Tx config is complete */
2760 		reg_val = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2761 		reg_val |= IXGBE_DMATXCTL_TE;
2762 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, reg_val);
2763 
2764 		/* Disable arbiter to set MTQC */
2765 		reg_val = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2766 		reg_val |= IXGBE_RTTDCS_ARBDIS;
2767 		IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg_val);
2768 		IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2769 		reg_val &= ~IXGBE_RTTDCS_ARBDIS;
2770 		IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg_val);
2771 	}
2772 
2773 	/*
2774 	 * Enabling tx queues ..
2775 	 * For 82599 must be done after DMATXCTL.TE is set
2776 	 */
2777 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2778 		tx_ring = &ixgbe->tx_rings[i];
2779 		reg_val = IXGBE_READ_REG(hw, IXGBE_TXDCTL(tx_ring->index));
2780 		reg_val |= IXGBE_TXDCTL_ENABLE;
2781 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(tx_ring->index), reg_val);
2782 	}
2783 }
2784 
2785 /*
2786  * ixgbe_setup_rss - Setup receive-side scaling feature.
2787  */
2788 static void
2789 ixgbe_setup_rss(ixgbe_t *ixgbe)
2790 {
2791 	struct ixgbe_hw *hw = &ixgbe->hw;
2792 	uint32_t mrqc;
2793 
2794 	/*
2795 	 * Initialize RETA/ERETA table
2796 	 */
2797 	ixgbe_setup_rss_table(ixgbe);
2798 
2799 	/*
2800 	 * Enable RSS & perform hash on these packet types
2801 	 */
2802 	mrqc = IXGBE_MRQC_RSSEN |
2803 	    IXGBE_MRQC_RSS_FIELD_IPV4 |
2804 	    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2805 	    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2806 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2807 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2808 	    IXGBE_MRQC_RSS_FIELD_IPV6 |
2809 	    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2810 	    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2811 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2812 	IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2813 }
2814 
2815 /*
2816  * ixgbe_setup_vmdq - Setup MAC classification feature
2817  */
2818 static void
2819 ixgbe_setup_vmdq(ixgbe_t *ixgbe)
2820 {
2821 	struct ixgbe_hw *hw = &ixgbe->hw;
2822 	uint32_t vmdctl, i, vtctl;
2823 
2824 	/*
2825 	 * Setup the VMDq Control register, enable VMDq based on
2826 	 * packet destination MAC address:
2827 	 */
2828 	switch (hw->mac.type) {
2829 	case ixgbe_mac_82598EB:
2830 		/*
2831 		 * VMDq Enable = 1;
2832 		 * VMDq Filter = 0; MAC filtering
2833 		 * Default VMDq output index = 0;
2834 		 */
2835 		vmdctl = IXGBE_VMD_CTL_VMDQ_EN;
2836 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
2837 		break;
2838 
2839 	case ixgbe_mac_82599EB:
2840 	case ixgbe_mac_X540:
2841 	case ixgbe_mac_X550:
2842 	case ixgbe_mac_X550EM_x:
2843 	case ixgbe_mac_X550EM_a:
2844 		/*
2845 		 * Enable VMDq-only.
2846 		 */
2847 		vmdctl = IXGBE_MRQC_VMDQEN;
2848 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, vmdctl);
2849 
2850 		for (i = 0; i < hw->mac.num_rar_entries; i++) {
2851 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(i), 0);
2852 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(i), 0);
2853 		}
2854 
2855 		/*
2856 		 * Enable Virtualization and Replication.
2857 		 */
2858 		vtctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2859 		IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vtctl);
2860 
2861 		/*
2862 		 * Enable receiving packets to all VFs
2863 		 */
2864 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), IXGBE_VFRE_ENABLE_ALL);
2865 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), IXGBE_VFRE_ENABLE_ALL);
2866 		break;
2867 
2868 	default:
2869 		break;
2870 	}
2871 }
2872 
2873 /*
2874  * ixgbe_setup_vmdq_rss - Setup both vmdq feature and rss feature.
2875  */
2876 static void
2877 ixgbe_setup_vmdq_rss(ixgbe_t *ixgbe)
2878 {
2879 	struct ixgbe_hw *hw = &ixgbe->hw;
2880 	uint32_t i, mrqc;
2881 	uint32_t vtctl, vmdctl;
2882 
2883 	/*
2884 	 * Initialize RETA/ERETA table
2885 	 */
2886 	ixgbe_setup_rss_table(ixgbe);
2887 
2888 	/*
2889 	 * Enable and setup RSS and VMDq
2890 	 */
2891 	switch (hw->mac.type) {
2892 	case ixgbe_mac_82598EB:
2893 		/*
2894 		 * Enable RSS & Setup RSS Hash functions
2895 		 */
2896 		mrqc = IXGBE_MRQC_RSSEN |
2897 		    IXGBE_MRQC_RSS_FIELD_IPV4 |
2898 		    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2899 		    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2900 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2901 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2902 		    IXGBE_MRQC_RSS_FIELD_IPV6 |
2903 		    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2904 		    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2905 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2906 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2907 
2908 		/*
2909 		 * Enable and Setup VMDq
2910 		 * VMDq Filter = 0; MAC filtering
2911 		 * Default VMDq output index = 0;
2912 		 */
2913 		vmdctl = IXGBE_VMD_CTL_VMDQ_EN;
2914 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
2915 		break;
2916 
2917 	case ixgbe_mac_82599EB:
2918 	case ixgbe_mac_X540:
2919 	case ixgbe_mac_X550:
2920 	case ixgbe_mac_X550EM_x:
2921 	case ixgbe_mac_X550EM_a:
2922 		/*
2923 		 * Enable RSS & Setup RSS Hash functions
2924 		 */
2925 		mrqc = IXGBE_MRQC_RSS_FIELD_IPV4 |
2926 		    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2927 		    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2928 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2929 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2930 		    IXGBE_MRQC_RSS_FIELD_IPV6 |
2931 		    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2932 		    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2933 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2934 
2935 		/*
2936 		 * Enable VMDq+RSS.
2937 		 */
2938 		if (ixgbe->num_rx_groups > 32)  {
2939 			mrqc = mrqc | IXGBE_MRQC_VMDQRSS64EN;
2940 		} else {
2941 			mrqc = mrqc | IXGBE_MRQC_VMDQRSS32EN;
2942 		}
2943 
2944 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2945 
2946 		for (i = 0; i < hw->mac.num_rar_entries; i++) {
2947 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(i), 0);
2948 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(i), 0);
2949 		}
2950 		break;
2951 
2952 	default:
2953 		break;
2954 
2955 	}
2956 
2957 	if (hw->mac.type == ixgbe_mac_82599EB ||
2958 	    hw->mac.type == ixgbe_mac_X540 ||
2959 	    hw->mac.type == ixgbe_mac_X550 ||
2960 	    hw->mac.type == ixgbe_mac_X550EM_x ||
2961 	    hw->mac.type == ixgbe_mac_X550EM_a) {
2962 		/*
2963 		 * Enable Virtualization and Replication.
2964 		 */
2965 		vtctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2966 		IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vtctl);
2967 
2968 		/*
2969 		 * Enable receiving packets to all VFs
2970 		 */
2971 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), IXGBE_VFRE_ENABLE_ALL);
2972 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), IXGBE_VFRE_ENABLE_ALL);
2973 	}
2974 }
2975 
2976 /*
2977  * ixgbe_setup_rss_table - Setup RSS table
2978  */
2979 static void
2980 ixgbe_setup_rss_table(ixgbe_t *ixgbe)
2981 {
2982 	struct ixgbe_hw *hw = &ixgbe->hw;
2983 	uint32_t i, j;
2984 	uint32_t random;
2985 	uint32_t reta;
2986 	uint32_t ring_per_group;
2987 	uint32_t ring;
2988 	uint32_t table_size;
2989 	uint32_t index_mult;
2990 	uint32_t rxcsum;
2991 
2992 	/*
2993 	 * Set multiplier for RETA setup and table size based on MAC type.
2994 	 * RETA table sizes vary by model:
2995 	 *
2996 	 * 82598, 82599, X540: 128 table entries.
2997 	 * X550: 512 table entries.
2998 	 */
2999 	index_mult = 0x1;
3000 	table_size = 128;
3001 	switch (ixgbe->hw.mac.type) {
3002 	case ixgbe_mac_82598EB:
3003 		index_mult = 0x11;
3004 		break;
3005 	case ixgbe_mac_X550:
3006 	case ixgbe_mac_X550EM_x:
3007 	case ixgbe_mac_X550EM_a:
3008 		table_size = 512;
3009 		break;
3010 	default:
3011 		break;
3012 	}
3013 
3014 	/*
3015 	 * Fill out RSS redirection table. The configuation of the indices is
3016 	 * hardware-dependent.
3017 	 *
3018 	 *  82598: 8 bits wide containing two 4 bit RSS indices
3019 	 *  82599, X540: 8 bits wide containing one 4 bit RSS index
3020 	 *  X550: 8 bits wide containing one 6 bit RSS index
3021 	 */
3022 	reta = 0;
3023 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
3024 
3025 	for (i = 0, j = 0; i < table_size; i++, j++) {
3026 		if (j == ring_per_group) j = 0;
3027 
3028 		/*
3029 		 * The low 8 bits are for hash value (n+0);
3030 		 * The next 8 bits are for hash value (n+1), etc.
3031 		 */
3032 		ring = (j * index_mult);
3033 		reta = reta >> 8;
3034 		reta = reta | (((uint32_t)ring) << 24);
3035 
3036 		if ((i & 3) == 3) {
3037 			/*
3038 			 * The first 128 table entries are programmed into the
3039 			 * RETA register, with any beyond that (eg; on X550)
3040 			 * into ERETA.
3041 			 */
3042 			if (i < 128)
3043 				IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3044 			else
3045 				IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3046 				    reta);
3047 			reta = 0;
3048 		}
3049 	}
3050 
3051 	/*
3052 	 * Fill out hash function seeds with a random constant
3053 	 */
3054 	for (i = 0; i < 10; i++) {
3055 		(void) random_get_pseudo_bytes((uint8_t *)&random,
3056 		    sizeof (uint32_t));
3057 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), random);
3058 	}
3059 
3060 	/*
3061 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
3062 	 * It is an adapter hardware limitation that Packet Checksum is
3063 	 * mutually exclusive with RSS.
3064 	 */
3065 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3066 	rxcsum |= IXGBE_RXCSUM_PCSD;
3067 	rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
3068 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3069 }
3070 
3071 /*
3072  * ixgbe_init_unicst - Initialize the unicast addresses.
3073  */
3074 static void
3075 ixgbe_init_unicst(ixgbe_t *ixgbe)
3076 {
3077 	struct ixgbe_hw *hw = &ixgbe->hw;
3078 	uint8_t *mac_addr;
3079 	int slot;
3080 	/*
3081 	 * Here we should consider two situations:
3082 	 *
3083 	 * 1. Chipset is initialized at the first time,
3084 	 *    Clear all the multiple unicast addresses.
3085 	 *
3086 	 * 2. Chipset is reset
3087 	 *    Recover the multiple unicast addresses from the
3088 	 *    software data structure to the RAR registers.
3089 	 */
3090 	if (!ixgbe->unicst_init) {
3091 		/*
3092 		 * Initialize the multiple unicast addresses
3093 		 */
3094 		ixgbe->unicst_total = hw->mac.num_rar_entries;
3095 		ixgbe->unicst_avail = ixgbe->unicst_total;
3096 		for (slot = 0; slot < ixgbe->unicst_total; slot++) {
3097 			mac_addr = ixgbe->unicst_addr[slot].mac.addr;
3098 			bzero(mac_addr, ETHERADDRL);
3099 			(void) ixgbe_set_rar(hw, slot, mac_addr, 0, 0);
3100 			ixgbe->unicst_addr[slot].mac.set = 0;
3101 		}
3102 		ixgbe->unicst_init = B_TRUE;
3103 	} else {
3104 		/* Re-configure the RAR registers */
3105 		for (slot = 0; slot < ixgbe->unicst_total; slot++) {
3106 			mac_addr = ixgbe->unicst_addr[slot].mac.addr;
3107 			if (ixgbe->unicst_addr[slot].mac.set == 1) {
3108 				(void) ixgbe_set_rar(hw, slot, mac_addr,
3109 				    ixgbe->unicst_addr[slot].mac.group_index,
3110 				    IXGBE_RAH_AV);
3111 			} else {
3112 				bzero(mac_addr, ETHERADDRL);
3113 				(void) ixgbe_set_rar(hw, slot, mac_addr, 0, 0);
3114 			}
3115 		}
3116 	}
3117 }
3118 
3119 /*
3120  * ixgbe_unicst_find - Find the slot for the specified unicast address
3121  */
3122 int
3123 ixgbe_unicst_find(ixgbe_t *ixgbe, const uint8_t *mac_addr)
3124 {
3125 	int slot;
3126 
3127 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3128 
3129 	for (slot = 0; slot < ixgbe->unicst_total; slot++) {
3130 		if (bcmp(ixgbe->unicst_addr[slot].mac.addr,
3131 		    mac_addr, ETHERADDRL) == 0)
3132 			return (slot);
3133 	}
3134 
3135 	return (-1);
3136 }
3137 
3138 /*
3139  * ixgbe_multicst_add - Add a multicst address.
3140  */
3141 int
3142 ixgbe_multicst_add(ixgbe_t *ixgbe, const uint8_t *multiaddr)
3143 {
3144 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3145 
3146 	if ((multiaddr[0] & 01) == 0) {
3147 		return (EINVAL);
3148 	}
3149 
3150 	if (ixgbe->mcast_count >= MAX_NUM_MULTICAST_ADDRESSES) {
3151 		return (ENOENT);
3152 	}
3153 
3154 	bcopy(multiaddr,
3155 	    &ixgbe->mcast_table[ixgbe->mcast_count], ETHERADDRL);
3156 	ixgbe->mcast_count++;
3157 
3158 	/*
3159 	 * Update the multicast table in the hardware
3160 	 */
3161 	ixgbe_setup_multicst(ixgbe);
3162 
3163 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
3164 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3165 		return (EIO);
3166 	}
3167 
3168 	return (0);
3169 }
3170 
3171 /*
3172  * ixgbe_multicst_remove - Remove a multicst address.
3173  */
3174 int
3175 ixgbe_multicst_remove(ixgbe_t *ixgbe, const uint8_t *multiaddr)
3176 {
3177 	int i;
3178 
3179 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3180 
3181 	for (i = 0; i < ixgbe->mcast_count; i++) {
3182 		if (bcmp(multiaddr, &ixgbe->mcast_table[i],
3183 		    ETHERADDRL) == 0) {
3184 			for (i++; i < ixgbe->mcast_count; i++) {
3185 				ixgbe->mcast_table[i - 1] =
3186 				    ixgbe->mcast_table[i];
3187 			}
3188 			ixgbe->mcast_count--;
3189 			break;
3190 		}
3191 	}
3192 
3193 	/*
3194 	 * Update the multicast table in the hardware
3195 	 */
3196 	ixgbe_setup_multicst(ixgbe);
3197 
3198 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
3199 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3200 		return (EIO);
3201 	}
3202 
3203 	return (0);
3204 }
3205 
3206 /*
3207  * ixgbe_setup_multicast - Setup multicast data structures.
3208  *
3209  * This routine initializes all of the multicast related structures
3210  * and save them in the hardware registers.
3211  */
3212 static void
3213 ixgbe_setup_multicst(ixgbe_t *ixgbe)
3214 {
3215 	uint8_t *mc_addr_list;
3216 	uint32_t mc_addr_count;
3217 	struct ixgbe_hw *hw = &ixgbe->hw;
3218 
3219 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3220 
3221 	ASSERT(ixgbe->mcast_count <= MAX_NUM_MULTICAST_ADDRESSES);
3222 
3223 	mc_addr_list = (uint8_t *)ixgbe->mcast_table;
3224 	mc_addr_count = ixgbe->mcast_count;
3225 
3226 	/*
3227 	 * Update the multicast addresses to the MTA registers
3228 	 */
3229 	(void) ixgbe_update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
3230 	    ixgbe_mc_table_itr, TRUE);
3231 }
3232 
3233 /*
3234  * ixgbe_setup_vmdq_rss_conf - Configure vmdq and rss (number and mode).
3235  *
3236  * Configure the rx classification mode (vmdq & rss) and vmdq & rss numbers.
3237  * Different chipsets may have different allowed configuration of vmdq and rss.
3238  */
3239 static void
3240 ixgbe_setup_vmdq_rss_conf(ixgbe_t *ixgbe)
3241 {
3242 	struct ixgbe_hw *hw = &ixgbe->hw;
3243 	uint32_t ring_per_group;
3244 
3245 	switch (hw->mac.type) {
3246 	case ixgbe_mac_82598EB:
3247 		/*
3248 		 * 82598 supports the following combination:
3249 		 * vmdq no. x rss no.
3250 		 * [5..16]  x 1
3251 		 * [1..4]   x [1..16]
3252 		 * However 8 rss queue per pool (vmdq) is sufficient for
3253 		 * most cases.
3254 		 */
3255 		ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
3256 		if (ixgbe->num_rx_groups > 4) {
3257 			ixgbe->num_rx_rings = ixgbe->num_rx_groups;
3258 		} else {
3259 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3260 			    min(8, ring_per_group);
3261 		}
3262 
3263 		break;
3264 
3265 	case ixgbe_mac_82599EB:
3266 	case ixgbe_mac_X540:
3267 	case ixgbe_mac_X550:
3268 	case ixgbe_mac_X550EM_x:
3269 	case ixgbe_mac_X550EM_a:
3270 		/*
3271 		 * 82599 supports the following combination:
3272 		 * vmdq no. x rss no.
3273 		 * [33..64] x [1..2]
3274 		 * [2..32]  x [1..4]
3275 		 * 1 x [1..16]
3276 		 * However 8 rss queue per pool (vmdq) is sufficient for
3277 		 * most cases.
3278 		 *
3279 		 * For now, treat X540 and X550 like the 82599.
3280 		 */
3281 		ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
3282 		if (ixgbe->num_rx_groups == 1) {
3283 			ixgbe->num_rx_rings = min(8, ring_per_group);
3284 		} else if (ixgbe->num_rx_groups <= 32) {
3285 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3286 			    min(4, ring_per_group);
3287 		} else if (ixgbe->num_rx_groups <= 64) {
3288 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3289 			    min(2, ring_per_group);
3290 		}
3291 		break;
3292 
3293 	default:
3294 		break;
3295 	}
3296 
3297 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
3298 
3299 	if (ixgbe->num_rx_groups == 1 && ring_per_group == 1) {
3300 		ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
3301 	} else if (ixgbe->num_rx_groups != 1 && ring_per_group == 1) {
3302 		ixgbe->classify_mode = IXGBE_CLASSIFY_VMDQ;
3303 	} else if (ixgbe->num_rx_groups != 1 && ring_per_group != 1) {
3304 		ixgbe->classify_mode = IXGBE_CLASSIFY_VMDQ_RSS;
3305 	} else {
3306 		ixgbe->classify_mode = IXGBE_CLASSIFY_RSS;
3307 	}
3308 
3309 	IXGBE_DEBUGLOG_2(ixgbe, "rx group number:%d, rx ring number:%d",
3310 	    ixgbe->num_rx_groups, ixgbe->num_rx_rings);
3311 }
3312 
3313 /*
3314  * ixgbe_get_conf - Get driver configurations set in driver.conf.
3315  *
3316  * This routine gets user-configured values out of the configuration
3317  * file ixgbe.conf.
3318  *
3319  * For each configurable value, there is a minimum, a maximum, and a
3320  * default.
3321  * If user does not configure a value, use the default.
3322  * If user configures below the minimum, use the minumum.
3323  * If user configures above the maximum, use the maxumum.
3324  */
3325 static void
3326 ixgbe_get_conf(ixgbe_t *ixgbe)
3327 {
3328 	struct ixgbe_hw *hw = &ixgbe->hw;
3329 	uint32_t flow_control;
3330 
3331 	/*
3332 	 * ixgbe driver supports the following user configurations:
3333 	 *
3334 	 * Jumbo frame configuration:
3335 	 *    default_mtu
3336 	 *
3337 	 * Ethernet flow control configuration:
3338 	 *    flow_control
3339 	 *
3340 	 * Multiple rings configurations:
3341 	 *    tx_queue_number
3342 	 *    tx_ring_size
3343 	 *    rx_queue_number
3344 	 *    rx_ring_size
3345 	 *
3346 	 * Call ixgbe_get_prop() to get the value for a specific
3347 	 * configuration parameter.
3348 	 */
3349 
3350 	/*
3351 	 * Jumbo frame configuration - max_frame_size controls host buffer
3352 	 * allocation, so includes MTU, ethernet header, vlan tag and
3353 	 * frame check sequence.
3354 	 */
3355 	ixgbe->default_mtu = ixgbe_get_prop(ixgbe, PROP_DEFAULT_MTU,
3356 	    MIN_MTU, ixgbe->capab->max_mtu, DEFAULT_MTU);
3357 
3358 	ixgbe->max_frame_size = ixgbe->default_mtu +
3359 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
3360 
3361 	/*
3362 	 * Ethernet flow control configuration
3363 	 */
3364 	flow_control = ixgbe_get_prop(ixgbe, PROP_FLOW_CONTROL,
3365 	    ixgbe_fc_none, 3, ixgbe_fc_none);
3366 	if (flow_control == 3)
3367 		flow_control = ixgbe_fc_default;
3368 
3369 	/*
3370 	 * fc.requested mode is what the user requests.  After autoneg,
3371 	 * fc.current_mode will be the flow_control mode that was negotiated.
3372 	 */
3373 	hw->fc.requested_mode = flow_control;
3374 
3375 	/*
3376 	 * Multiple rings configurations
3377 	 */
3378 	ixgbe->num_tx_rings = ixgbe_get_prop(ixgbe, PROP_TX_QUEUE_NUM,
3379 	    ixgbe->capab->min_tx_que_num,
3380 	    ixgbe->capab->max_tx_que_num,
3381 	    ixgbe->capab->def_tx_que_num);
3382 	ixgbe->tx_ring_size = ixgbe_get_prop(ixgbe, PROP_TX_RING_SIZE,
3383 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
3384 
3385 	ixgbe->num_rx_rings = ixgbe_get_prop(ixgbe, PROP_RX_QUEUE_NUM,
3386 	    ixgbe->capab->min_rx_que_num,
3387 	    ixgbe->capab->max_rx_que_num,
3388 	    ixgbe->capab->def_rx_que_num);
3389 	ixgbe->rx_ring_size = ixgbe_get_prop(ixgbe, PROP_RX_RING_SIZE,
3390 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
3391 
3392 	/*
3393 	 * Multiple groups configuration
3394 	 */
3395 	ixgbe->num_rx_groups = ixgbe_get_prop(ixgbe, PROP_RX_GROUP_NUM,
3396 	    ixgbe->capab->min_rx_grp_num, ixgbe->capab->max_rx_grp_num,
3397 	    ixgbe->capab->def_rx_grp_num);
3398 
3399 	ixgbe->mr_enable = ixgbe_get_prop(ixgbe, PROP_MR_ENABLE,
3400 	    0, 1, DEFAULT_MR_ENABLE);
3401 
3402 	if (ixgbe->mr_enable == B_FALSE) {
3403 		ixgbe->num_tx_rings = 1;
3404 		ixgbe->num_rx_rings = 1;
3405 		ixgbe->num_rx_groups = 1;
3406 		ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
3407 	} else {
3408 		ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3409 		    max(ixgbe->num_rx_rings / ixgbe->num_rx_groups, 1);
3410 		/*
3411 		 * The combination of num_rx_rings and num_rx_groups
3412 		 * may be not supported by h/w. We need to adjust
3413 		 * them to appropriate values.
3414 		 */
3415 		ixgbe_setup_vmdq_rss_conf(ixgbe);
3416 	}
3417 
3418 	/*
3419 	 * Tunable used to force an interrupt type. The only use is
3420 	 * for testing of the lesser interrupt types.
3421 	 * 0 = don't force interrupt type
3422 	 * 1 = force interrupt type MSI-X
3423 	 * 2 = force interrupt type MSI
3424 	 * 3 = force interrupt type Legacy
3425 	 */
3426 	ixgbe->intr_force = ixgbe_get_prop(ixgbe, PROP_INTR_FORCE,
3427 	    IXGBE_INTR_NONE, IXGBE_INTR_LEGACY, IXGBE_INTR_NONE);
3428 
3429 	ixgbe->tx_hcksum_enable = ixgbe_get_prop(ixgbe, PROP_TX_HCKSUM_ENABLE,
3430 	    0, 1, DEFAULT_TX_HCKSUM_ENABLE);
3431 	ixgbe->rx_hcksum_enable = ixgbe_get_prop(ixgbe, PROP_RX_HCKSUM_ENABLE,
3432 	    0, 1, DEFAULT_RX_HCKSUM_ENABLE);
3433 	ixgbe->lso_enable = ixgbe_get_prop(ixgbe, PROP_LSO_ENABLE,
3434 	    0, 1, DEFAULT_LSO_ENABLE);
3435 	ixgbe->lro_enable = ixgbe_get_prop(ixgbe, PROP_LRO_ENABLE,
3436 	    0, 1, DEFAULT_LRO_ENABLE);
3437 	ixgbe->tx_head_wb_enable = ixgbe_get_prop(ixgbe, PROP_TX_HEAD_WB_ENABLE,
3438 	    0, 1, DEFAULT_TX_HEAD_WB_ENABLE);
3439 	ixgbe->relax_order_enable = ixgbe_get_prop(ixgbe,
3440 	    PROP_RELAX_ORDER_ENABLE, 0, 1, DEFAULT_RELAX_ORDER_ENABLE);
3441 
3442 	/* Head Write Back not recommended for 82599, X540 and X550 */
3443 	if (hw->mac.type == ixgbe_mac_82599EB ||
3444 	    hw->mac.type == ixgbe_mac_X540 ||
3445 	    hw->mac.type == ixgbe_mac_X550 ||
3446 	    hw->mac.type == ixgbe_mac_X550EM_x ||
3447 	    hw->mac.type == ixgbe_mac_X550EM_a) {
3448 		ixgbe->tx_head_wb_enable = B_FALSE;
3449 	}
3450 
3451 	/*
3452 	 * ixgbe LSO needs the tx h/w checksum support.
3453 	 * LSO will be disabled if tx h/w checksum is not
3454 	 * enabled.
3455 	 */
3456 	if (ixgbe->tx_hcksum_enable == B_FALSE) {
3457 		ixgbe->lso_enable = B_FALSE;
3458 	}
3459 
3460 	/*
3461 	 * ixgbe LRO needs the rx h/w checksum support.
3462 	 * LRO will be disabled if rx h/w checksum is not
3463 	 * enabled.
3464 	 */
3465 	if (ixgbe->rx_hcksum_enable == B_FALSE) {
3466 		ixgbe->lro_enable = B_FALSE;
3467 	}
3468 
3469 	/*
3470 	 * ixgbe LRO only supported by 82599, X540 and X550
3471 	 */
3472 	if (hw->mac.type == ixgbe_mac_82598EB) {
3473 		ixgbe->lro_enable = B_FALSE;
3474 	}
3475 	ixgbe->tx_copy_thresh = ixgbe_get_prop(ixgbe, PROP_TX_COPY_THRESHOLD,
3476 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
3477 	    DEFAULT_TX_COPY_THRESHOLD);
3478 	ixgbe->tx_recycle_thresh = ixgbe_get_prop(ixgbe,
3479 	    PROP_TX_RECYCLE_THRESHOLD, MIN_TX_RECYCLE_THRESHOLD,
3480 	    MAX_TX_RECYCLE_THRESHOLD, DEFAULT_TX_RECYCLE_THRESHOLD);
3481 	ixgbe->tx_overload_thresh = ixgbe_get_prop(ixgbe,
3482 	    PROP_TX_OVERLOAD_THRESHOLD, MIN_TX_OVERLOAD_THRESHOLD,
3483 	    MAX_TX_OVERLOAD_THRESHOLD, DEFAULT_TX_OVERLOAD_THRESHOLD);
3484 	ixgbe->tx_resched_thresh = ixgbe_get_prop(ixgbe,
3485 	    PROP_TX_RESCHED_THRESHOLD, MIN_TX_RESCHED_THRESHOLD,
3486 	    MAX_TX_RESCHED_THRESHOLD, DEFAULT_TX_RESCHED_THRESHOLD);
3487 
3488 	ixgbe->rx_copy_thresh = ixgbe_get_prop(ixgbe, PROP_RX_COPY_THRESHOLD,
3489 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
3490 	    DEFAULT_RX_COPY_THRESHOLD);
3491 	ixgbe->rx_limit_per_intr = ixgbe_get_prop(ixgbe, PROP_RX_LIMIT_PER_INTR,
3492 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
3493 	    DEFAULT_RX_LIMIT_PER_INTR);
3494 
3495 	ixgbe->intr_throttling[0] = ixgbe_get_prop(ixgbe, PROP_INTR_THROTTLING,
3496 	    ixgbe->capab->min_intr_throttle,
3497 	    ixgbe->capab->max_intr_throttle,
3498 	    ixgbe->capab->def_intr_throttle);
3499 	/*
3500 	 * 82599, X540 and X550 require the interrupt throttling rate is
3501 	 * a multiple of 8. This is enforced by the register definiton.
3502 	 */
3503 	if (hw->mac.type == ixgbe_mac_82599EB ||
3504 	    hw->mac.type == ixgbe_mac_X540 ||
3505 	    hw->mac.type == ixgbe_mac_X550 ||
3506 	    hw->mac.type == ixgbe_mac_X550EM_x ||
3507 	    hw->mac.type == ixgbe_mac_X550EM_a)
3508 		ixgbe->intr_throttling[0] = ixgbe->intr_throttling[0] & 0xFF8;
3509 
3510 	hw->allow_unsupported_sfp = ixgbe_get_prop(ixgbe,
3511 	    PROP_ALLOW_UNSUPPORTED_SFP, 0, 1, DEFAULT_ALLOW_UNSUPPORTED_SFP);
3512 }
3513 
3514 static void
3515 ixgbe_init_params(ixgbe_t *ixgbe)
3516 {
3517 	struct ixgbe_hw *hw = &ixgbe->hw;
3518 	ixgbe_link_speed speeds_supported = 0;
3519 	boolean_t negotiate;
3520 
3521 	/*
3522 	 * Get a list of speeds the adapter supports. If the hw struct hasn't
3523 	 * been populated with this information yet, retrieve it from the
3524 	 * adapter and save it to our own variable.
3525 	 *
3526 	 * On certain adapters, such as ones which use SFPs, the contents of
3527 	 * hw->phy.speeds_supported (and hw->phy.autoneg_advertised) are not
3528 	 * updated, so we must rely on calling ixgbe_get_link_capabilities()
3529 	 * in order to ascertain the speeds which we are capable of supporting,
3530 	 * and in the case of SFP-equipped adapters, which speed we are
3531 	 * advertising. If ixgbe_get_link_capabilities() fails for some reason,
3532 	 * we'll go with a default list of speeds as a last resort.
3533 	 */
3534 	speeds_supported = hw->phy.speeds_supported;
3535 
3536 	if (speeds_supported == 0) {
3537 		if (ixgbe_get_link_capabilities(hw, &speeds_supported,
3538 		    &negotiate) != IXGBE_SUCCESS) {
3539 			if (hw->mac.type == ixgbe_mac_82598EB) {
3540 				speeds_supported =
3541 				    IXGBE_LINK_SPEED_82598_AUTONEG;
3542 			} else {
3543 				speeds_supported =
3544 				    IXGBE_LINK_SPEED_82599_AUTONEG;
3545 			}
3546 		}
3547 	}
3548 	ixgbe->speeds_supported = speeds_supported;
3549 
3550 	/*
3551 	 * By default, all supported speeds are enabled and advertised.
3552 	 */
3553 	if (speeds_supported & IXGBE_LINK_SPEED_10GB_FULL) {
3554 		ixgbe->param_en_10000fdx_cap = 1;
3555 		ixgbe->param_adv_10000fdx_cap = 1;
3556 	} else {
3557 		ixgbe->param_en_10000fdx_cap = 0;
3558 		ixgbe->param_adv_10000fdx_cap = 0;
3559 	}
3560 
3561 	if (speeds_supported & IXGBE_LINK_SPEED_5GB_FULL) {
3562 		ixgbe->param_en_5000fdx_cap = 1;
3563 		ixgbe->param_adv_5000fdx_cap = 1;
3564 	} else {
3565 		ixgbe->param_en_5000fdx_cap = 0;
3566 		ixgbe->param_adv_5000fdx_cap = 0;
3567 	}
3568 
3569 	if (speeds_supported & IXGBE_LINK_SPEED_2_5GB_FULL) {
3570 		ixgbe->param_en_2500fdx_cap = 1;
3571 		ixgbe->param_adv_2500fdx_cap = 1;
3572 	} else {
3573 		ixgbe->param_en_2500fdx_cap = 0;
3574 		ixgbe->param_adv_2500fdx_cap = 0;
3575 	}
3576 
3577 	if (speeds_supported & IXGBE_LINK_SPEED_1GB_FULL) {
3578 		ixgbe->param_en_1000fdx_cap = 1;
3579 		ixgbe->param_adv_1000fdx_cap = 1;
3580 	} else {
3581 		ixgbe->param_en_1000fdx_cap = 0;
3582 		ixgbe->param_adv_1000fdx_cap = 0;
3583 	}
3584 
3585 	if (speeds_supported & IXGBE_LINK_SPEED_100_FULL) {
3586 		ixgbe->param_en_100fdx_cap = 1;
3587 		ixgbe->param_adv_100fdx_cap = 1;
3588 	} else {
3589 		ixgbe->param_en_100fdx_cap = 0;
3590 		ixgbe->param_adv_100fdx_cap = 0;
3591 	}
3592 
3593 	ixgbe->param_pause_cap = 1;
3594 	ixgbe->param_asym_pause_cap = 1;
3595 	ixgbe->param_rem_fault = 0;
3596 
3597 	ixgbe->param_adv_autoneg_cap = 1;
3598 	ixgbe->param_adv_pause_cap = 1;
3599 	ixgbe->param_adv_asym_pause_cap = 1;
3600 	ixgbe->param_adv_rem_fault = 0;
3601 
3602 	ixgbe->param_lp_10000fdx_cap = 0;
3603 	ixgbe->param_lp_5000fdx_cap = 0;
3604 	ixgbe->param_lp_2500fdx_cap = 0;
3605 	ixgbe->param_lp_1000fdx_cap = 0;
3606 	ixgbe->param_lp_100fdx_cap = 0;
3607 	ixgbe->param_lp_autoneg_cap = 0;
3608 	ixgbe->param_lp_pause_cap = 0;
3609 	ixgbe->param_lp_asym_pause_cap = 0;
3610 	ixgbe->param_lp_rem_fault = 0;
3611 }
3612 
3613 /*
3614  * ixgbe_get_prop - Get a property value out of the configuration file
3615  * ixgbe.conf.
3616  *
3617  * Caller provides the name of the property, a default value, a minimum
3618  * value, and a maximum value.
3619  *
3620  * Return configured value of the property, with default, minimum and
3621  * maximum properly applied.
3622  */
3623 static int
3624 ixgbe_get_prop(ixgbe_t *ixgbe,
3625     char *propname,	/* name of the property */
3626     int minval,		/* minimum acceptable value */
3627     int maxval,		/* maximim acceptable value */
3628     int defval)		/* default value */
3629 {
3630 	int value;
3631 
3632 	/*
3633 	 * Call ddi_prop_get_int() to read the conf settings
3634 	 */
3635 	value = ddi_prop_get_int(DDI_DEV_T_ANY, ixgbe->dip,
3636 	    DDI_PROP_DONTPASS, propname, defval);
3637 	if (value > maxval)
3638 		value = maxval;
3639 
3640 	if (value < minval)
3641 		value = minval;
3642 
3643 	return (value);
3644 }
3645 
3646 /*
3647  * ixgbe_driver_setup_link - Using the link properties to setup the link.
3648  */
3649 int
3650 ixgbe_driver_setup_link(ixgbe_t *ixgbe, boolean_t setup_hw)
3651 {
3652 	struct ixgbe_hw *hw = &ixgbe->hw;
3653 	ixgbe_link_speed advertised = 0;
3654 
3655 	/*
3656 	 * Assemble a list of enabled speeds to auto-negotiate with.
3657 	 */
3658 	if (ixgbe->param_en_10000fdx_cap == 1)
3659 		advertised |= IXGBE_LINK_SPEED_10GB_FULL;
3660 
3661 	if (ixgbe->param_en_5000fdx_cap == 1)
3662 		advertised |= IXGBE_LINK_SPEED_5GB_FULL;
3663 
3664 	if (ixgbe->param_en_2500fdx_cap == 1)
3665 		advertised |= IXGBE_LINK_SPEED_2_5GB_FULL;
3666 
3667 	if (ixgbe->param_en_1000fdx_cap == 1)
3668 		advertised |= IXGBE_LINK_SPEED_1GB_FULL;
3669 
3670 	if (ixgbe->param_en_100fdx_cap == 1)
3671 		advertised |= IXGBE_LINK_SPEED_100_FULL;
3672 
3673 	/*
3674 	 * As a last resort, autoneg with a default list of speeds.
3675 	 */
3676 	if (ixgbe->param_adv_autoneg_cap == 1 && advertised == 0) {
3677 		ixgbe_notice(ixgbe, "Invalid link settings. Setting link "
3678 		    "to autonegotiate with full capabilities.");
3679 
3680 		if (hw->mac.type == ixgbe_mac_82598EB)
3681 			advertised = IXGBE_LINK_SPEED_82598_AUTONEG;
3682 		else
3683 			advertised = IXGBE_LINK_SPEED_82599_AUTONEG;
3684 	}
3685 
3686 	if (setup_hw) {
3687 		if (ixgbe_setup_link(&ixgbe->hw, advertised,
3688 		    ixgbe->param_adv_autoneg_cap) != IXGBE_SUCCESS) {
3689 			ixgbe_notice(ixgbe, "Setup link failed on this "
3690 			    "device.");
3691 			return (IXGBE_FAILURE);
3692 		}
3693 	}
3694 
3695 	return (IXGBE_SUCCESS);
3696 }
3697 
3698 /*
3699  * ixgbe_driver_link_check - Link status processing.
3700  *
3701  * This function can be called in both kernel context and interrupt context
3702  */
3703 static void
3704 ixgbe_driver_link_check(ixgbe_t *ixgbe)
3705 {
3706 	struct ixgbe_hw *hw = &ixgbe->hw;
3707 	ixgbe_link_speed speed = IXGBE_LINK_SPEED_UNKNOWN;
3708 	boolean_t link_up = B_FALSE;
3709 	boolean_t link_changed = B_FALSE;
3710 
3711 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3712 
3713 	(void) ixgbe_check_link(hw, &speed, &link_up, B_FALSE);
3714 	if (link_up) {
3715 		ixgbe->link_check_complete = B_TRUE;
3716 
3717 		/* Link is up, enable flow control settings */
3718 		(void) ixgbe_fc_enable(hw);
3719 
3720 		/*
3721 		 * The Link is up, check whether it was marked as down earlier
3722 		 */
3723 		if (ixgbe->link_state != LINK_STATE_UP) {
3724 			switch (speed) {
3725 			case IXGBE_LINK_SPEED_10GB_FULL:
3726 				ixgbe->link_speed = SPEED_10GB;
3727 				break;
3728 			case IXGBE_LINK_SPEED_5GB_FULL:
3729 				ixgbe->link_speed = SPEED_5GB;
3730 				break;
3731 			case IXGBE_LINK_SPEED_2_5GB_FULL:
3732 				ixgbe->link_speed = SPEED_2_5GB;
3733 				break;
3734 			case IXGBE_LINK_SPEED_1GB_FULL:
3735 				ixgbe->link_speed = SPEED_1GB;
3736 				break;
3737 			case IXGBE_LINK_SPEED_100_FULL:
3738 				ixgbe->link_speed = SPEED_100;
3739 			}
3740 			ixgbe->link_duplex = LINK_DUPLEX_FULL;
3741 			ixgbe->link_state = LINK_STATE_UP;
3742 			link_changed = B_TRUE;
3743 		}
3744 	} else {
3745 		if (ixgbe->link_check_complete == B_TRUE ||
3746 		    (ixgbe->link_check_complete == B_FALSE &&
3747 		    gethrtime() >= ixgbe->link_check_hrtime)) {
3748 			/*
3749 			 * The link is really down
3750 			 */
3751 			ixgbe->link_check_complete = B_TRUE;
3752 
3753 			if (ixgbe->link_state != LINK_STATE_DOWN) {
3754 				ixgbe->link_speed = 0;
3755 				ixgbe->link_duplex = LINK_DUPLEX_UNKNOWN;
3756 				ixgbe->link_state = LINK_STATE_DOWN;
3757 				link_changed = B_TRUE;
3758 			}
3759 		}
3760 	}
3761 
3762 	/*
3763 	 * If we are in an interrupt context, need to re-enable the
3764 	 * interrupt, which was automasked
3765 	 */
3766 	if (servicing_interrupt() != 0) {
3767 		ixgbe->eims |= IXGBE_EICR_LSC;
3768 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
3769 	}
3770 
3771 	if (link_changed) {
3772 		mac_link_update(ixgbe->mac_hdl, ixgbe->link_state);
3773 	}
3774 }
3775 
3776 /*
3777  * ixgbe_sfp_check - sfp module processing done in taskq only for 82599.
3778  */
3779 static void
3780 ixgbe_sfp_check(void *arg)
3781 {
3782 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3783 	uint32_t eicr = ixgbe->eicr;
3784 	struct ixgbe_hw *hw = &ixgbe->hw;
3785 
3786 	mutex_enter(&ixgbe->gen_lock);
3787 	(void) hw->phy.ops.identify_sfp(hw);
3788 	if (eicr & IXGBE_EICR_GPI_SDP1_BY_MAC(hw)) {
3789 		/* clear the interrupt */
3790 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1_BY_MAC(hw));
3791 
3792 		/* if link up, do multispeed fiber setup */
3793 		(void) ixgbe_setup_link(hw, IXGBE_LINK_SPEED_82599_AUTONEG,
3794 		    B_TRUE);
3795 		ixgbe_driver_link_check(ixgbe);
3796 		ixgbe_get_hw_state(ixgbe);
3797 	} else if (eicr & IXGBE_EICR_GPI_SDP2_BY_MAC(hw)) {
3798 		/* clear the interrupt */
3799 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2_BY_MAC(hw));
3800 
3801 		/* if link up, do sfp module setup */
3802 		(void) hw->mac.ops.setup_sfp(hw);
3803 
3804 		/* do multispeed fiber setup */
3805 		(void) ixgbe_setup_link(hw, IXGBE_LINK_SPEED_82599_AUTONEG,
3806 		    B_TRUE);
3807 		ixgbe_driver_link_check(ixgbe);
3808 		ixgbe_get_hw_state(ixgbe);
3809 	}
3810 	mutex_exit(&ixgbe->gen_lock);
3811 
3812 	/*
3813 	 * We need to fully re-check the link later.
3814 	 */
3815 	ixgbe->link_check_complete = B_FALSE;
3816 	ixgbe->link_check_hrtime = gethrtime() +
3817 	    (IXGBE_LINK_UP_TIME * 100000000ULL);
3818 }
3819 
3820 /*
3821  * ixgbe_overtemp_check - overtemp module processing done in taskq
3822  *
3823  * This routine will only be called on adapters with temperature sensor.
3824  * The indication of over-temperature can be either SDP0 interrupt or the link
3825  * status change interrupt.
3826  */
3827 static void
3828 ixgbe_overtemp_check(void *arg)
3829 {
3830 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3831 	struct ixgbe_hw *hw = &ixgbe->hw;
3832 	uint32_t eicr = ixgbe->eicr;
3833 	ixgbe_link_speed speed;
3834 	boolean_t link_up;
3835 
3836 	mutex_enter(&ixgbe->gen_lock);
3837 
3838 	/* make sure we know current state of link */
3839 	(void) ixgbe_check_link(hw, &speed, &link_up, B_FALSE);
3840 
3841 	/* check over-temp condition */
3842 	if (((eicr & IXGBE_EICR_GPI_SDP0_BY_MAC(hw)) && (!link_up)) ||
3843 	    (eicr & IXGBE_EICR_LSC)) {
3844 		if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP) {
3845 			atomic_or_32(&ixgbe->ixgbe_state, IXGBE_OVERTEMP);
3846 
3847 			/*
3848 			 * Disable the adapter interrupts
3849 			 */
3850 			ixgbe_disable_adapter_interrupts(ixgbe);
3851 
3852 			/*
3853 			 * Disable Rx/Tx units
3854 			 */
3855 			(void) ixgbe_stop_adapter(hw);
3856 
3857 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
3858 			ixgbe_error(ixgbe,
3859 			    "Problem: Network adapter has been stopped "
3860 			    "because it has overheated");
3861 			ixgbe_error(ixgbe,
3862 			    "Action: Restart the computer. "
3863 			    "If the problem persists, power off the system "
3864 			    "and replace the adapter");
3865 		}
3866 	}
3867 
3868 	/* write to clear the interrupt */
3869 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3870 
3871 	mutex_exit(&ixgbe->gen_lock);
3872 }
3873 
3874 /*
3875  * ixgbe_phy_check - taskq to process interrupts from an external PHY
3876  *
3877  * This routine will only be called on adapters with external PHYs
3878  * (such as X550) that may be trying to raise our attention to some event.
3879  * Currently, this is limited to claiming PHY overtemperature and link status
3880  * change (LSC) events, however this may expand to include other things in
3881  * future adapters.
3882  */
3883 static void
3884 ixgbe_phy_check(void *arg)
3885 {
3886 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3887 	struct ixgbe_hw *hw = &ixgbe->hw;
3888 	int rv;
3889 
3890 	mutex_enter(&ixgbe->gen_lock);
3891 
3892 	/*
3893 	 * X550 baseT PHY overtemp and LSC events are handled here.
3894 	 *
3895 	 * If an overtemp event occurs, it will be reflected in the
3896 	 * return value of phy.ops.handle_lasi() and the common code will
3897 	 * automatically power off the baseT PHY. This is our cue to trigger
3898 	 * an FMA event.
3899 	 *
3900 	 * If a link status change event occurs, phy.ops.handle_lasi() will
3901 	 * automatically initiate a link setup between the integrated KR PHY
3902 	 * and the external X557 PHY to ensure that the link speed between
3903 	 * them matches the link speed of the baseT link.
3904 	 */
3905 	rv = ixgbe_handle_lasi(hw);
3906 
3907 	if (rv == IXGBE_ERR_OVERTEMP) {
3908 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_OVERTEMP);
3909 
3910 		/*
3911 		 * Disable the adapter interrupts
3912 		 */
3913 		ixgbe_disable_adapter_interrupts(ixgbe);
3914 
3915 		/*
3916 		 * Disable Rx/Tx units
3917 		 */
3918 		(void) ixgbe_stop_adapter(hw);
3919 
3920 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
3921 		ixgbe_error(ixgbe,
3922 		    "Problem: Network adapter has been stopped due to a "
3923 		    "overtemperature event being detected.");
3924 		ixgbe_error(ixgbe,
3925 		    "Action: Shut down or restart the computer. If the issue "
3926 		    "persists, please take action in accordance with the "
3927 		    "recommendations from your system vendor.");
3928 	}
3929 
3930 	mutex_exit(&ixgbe->gen_lock);
3931 }
3932 
3933 /*
3934  * ixgbe_link_timer - timer for link status detection
3935  */
3936 static void
3937 ixgbe_link_timer(void *arg)
3938 {
3939 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3940 
3941 	mutex_enter(&ixgbe->gen_lock);
3942 	ixgbe_driver_link_check(ixgbe);
3943 	mutex_exit(&ixgbe->gen_lock);
3944 }
3945 
3946 /*
3947  * ixgbe_local_timer - Driver watchdog function.
3948  *
3949  * This function will handle the transmit stall check and other routines.
3950  */
3951 static void
3952 ixgbe_local_timer(void *arg)
3953 {
3954 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3955 
3956 	if (ixgbe->ixgbe_state & IXGBE_OVERTEMP)
3957 		goto out;
3958 
3959 	if (ixgbe->ixgbe_state & IXGBE_ERROR) {
3960 		ixgbe->reset_count++;
3961 		if (ixgbe_reset(ixgbe) == IXGBE_SUCCESS)
3962 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_RESTORED);
3963 		goto out;
3964 	}
3965 
3966 	if (ixgbe_stall_check(ixgbe)) {
3967 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_STALL);
3968 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3969 
3970 		ixgbe->reset_count++;
3971 		if (ixgbe_reset(ixgbe) == IXGBE_SUCCESS)
3972 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_RESTORED);
3973 	}
3974 
3975 out:
3976 	ixgbe_restart_watchdog_timer(ixgbe);
3977 }
3978 
3979 /*
3980  * ixgbe_stall_check - Check for transmit stall.
3981  *
3982  * This function checks if the adapter is stalled (in transmit).
3983  *
3984  * It is called each time the watchdog timeout is invoked.
3985  * If the transmit descriptor reclaim continuously fails,
3986  * the watchdog value will increment by 1. If the watchdog
3987  * value exceeds the threshold, the ixgbe is assumed to
3988  * have stalled and need to be reset.
3989  */
3990 static boolean_t
3991 ixgbe_stall_check(ixgbe_t *ixgbe)
3992 {
3993 	ixgbe_tx_ring_t *tx_ring;
3994 	boolean_t result;
3995 	int i;
3996 
3997 	if (ixgbe->link_state != LINK_STATE_UP)
3998 		return (B_FALSE);
3999 
4000 	/*
4001 	 * If any tx ring is stalled, we'll reset the chipset
4002 	 */
4003 	result = B_FALSE;
4004 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
4005 		tx_ring = &ixgbe->tx_rings[i];
4006 		if (tx_ring->tbd_free <= ixgbe->tx_recycle_thresh) {
4007 			tx_ring->tx_recycle(tx_ring);
4008 		}
4009 
4010 		if (tx_ring->recycle_fail > 0)
4011 			tx_ring->stall_watchdog++;
4012 		else
4013 			tx_ring->stall_watchdog = 0;
4014 
4015 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
4016 			result = B_TRUE;
4017 			break;
4018 		}
4019 	}
4020 
4021 	if (result) {
4022 		tx_ring->stall_watchdog = 0;
4023 		tx_ring->recycle_fail = 0;
4024 	}
4025 
4026 	return (result);
4027 }
4028 
4029 
4030 /*
4031  * is_valid_mac_addr - Check if the mac address is valid.
4032  */
4033 static boolean_t
4034 is_valid_mac_addr(uint8_t *mac_addr)
4035 {
4036 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
4037 	const uint8_t addr_test2[6] =
4038 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
4039 
4040 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
4041 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
4042 		return (B_FALSE);
4043 
4044 	return (B_TRUE);
4045 }
4046 
4047 static boolean_t
4048 ixgbe_find_mac_address(ixgbe_t *ixgbe)
4049 {
4050 #ifdef __sparc
4051 	struct ixgbe_hw *hw = &ixgbe->hw;
4052 	uchar_t *bytes;
4053 	struct ether_addr sysaddr;
4054 	uint_t nelts;
4055 	int err;
4056 	boolean_t found = B_FALSE;
4057 
4058 	/*
4059 	 * The "vendor's factory-set address" may already have
4060 	 * been extracted from the chip, but if the property
4061 	 * "local-mac-address" is set we use that instead.
4062 	 *
4063 	 * We check whether it looks like an array of 6
4064 	 * bytes (which it should, if OBP set it).  If we can't
4065 	 * make sense of it this way, we'll ignore it.
4066 	 */
4067 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip,
4068 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
4069 	if (err == DDI_PROP_SUCCESS) {
4070 		if (nelts == ETHERADDRL) {
4071 			while (nelts--)
4072 				hw->mac.addr[nelts] = bytes[nelts];
4073 			found = B_TRUE;
4074 		}
4075 		ddi_prop_free(bytes);
4076 	}
4077 
4078 	/*
4079 	 * Look up the OBP property "local-mac-address?". If the user has set
4080 	 * 'local-mac-address? = false', use "the system address" instead.
4081 	 */
4082 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip, 0,
4083 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
4084 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
4085 			if (localetheraddr(NULL, &sysaddr) != 0) {
4086 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
4087 				found = B_TRUE;
4088 			}
4089 		}
4090 		ddi_prop_free(bytes);
4091 	}
4092 
4093 	/*
4094 	 * Finally(!), if there's a valid "mac-address" property (created
4095 	 * if we netbooted from this interface), we must use this instead
4096 	 * of any of the above to ensure that the NFS/install server doesn't
4097 	 * get confused by the address changing as illumos takes over!
4098 	 */
4099 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip,
4100 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
4101 	if (err == DDI_PROP_SUCCESS) {
4102 		if (nelts == ETHERADDRL) {
4103 			while (nelts--)
4104 				hw->mac.addr[nelts] = bytes[nelts];
4105 			found = B_TRUE;
4106 		}
4107 		ddi_prop_free(bytes);
4108 	}
4109 
4110 	if (found) {
4111 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
4112 		return (B_TRUE);
4113 	}
4114 #else
4115 	_NOTE(ARGUNUSED(ixgbe));
4116 #endif
4117 
4118 	return (B_TRUE);
4119 }
4120 
4121 #pragma inline(ixgbe_arm_watchdog_timer)
4122 static void
4123 ixgbe_arm_watchdog_timer(ixgbe_t *ixgbe)
4124 {
4125 	/*
4126 	 * Fire a watchdog timer
4127 	 */
4128 	ixgbe->watchdog_tid =
4129 	    timeout(ixgbe_local_timer,
4130 	    (void *)ixgbe, 1 * drv_usectohz(1000000));
4131 
4132 }
4133 
4134 /*
4135  * ixgbe_enable_watchdog_timer - Enable and start the driver watchdog timer.
4136  */
4137 void
4138 ixgbe_enable_watchdog_timer(ixgbe_t *ixgbe)
4139 {
4140 	mutex_enter(&ixgbe->watchdog_lock);
4141 
4142 	if (!ixgbe->watchdog_enable) {
4143 		ixgbe->watchdog_enable = B_TRUE;
4144 		ixgbe->watchdog_start = B_TRUE;
4145 		ixgbe_arm_watchdog_timer(ixgbe);
4146 	}
4147 
4148 	mutex_exit(&ixgbe->watchdog_lock);
4149 }
4150 
4151 /*
4152  * ixgbe_disable_watchdog_timer - Disable and stop the driver watchdog timer.
4153  */
4154 void
4155 ixgbe_disable_watchdog_timer(ixgbe_t *ixgbe)
4156 {
4157 	timeout_id_t tid;
4158 
4159 	mutex_enter(&ixgbe->watchdog_lock);
4160 
4161 	ixgbe->watchdog_enable = B_FALSE;
4162 	ixgbe->watchdog_start = B_FALSE;
4163 	tid = ixgbe->watchdog_tid;
4164 	ixgbe->watchdog_tid = 0;
4165 
4166 	mutex_exit(&ixgbe->watchdog_lock);
4167 
4168 	if (tid != 0)
4169 		(void) untimeout(tid);
4170 }
4171 
4172 /*
4173  * ixgbe_start_watchdog_timer - Start the driver watchdog timer.
4174  */
4175 void
4176 ixgbe_start_watchdog_timer(ixgbe_t *ixgbe)
4177 {
4178 	mutex_enter(&ixgbe->watchdog_lock);
4179 
4180 	if (ixgbe->watchdog_enable) {
4181 		if (!ixgbe->watchdog_start) {
4182 			ixgbe->watchdog_start = B_TRUE;
4183 			ixgbe_arm_watchdog_timer(ixgbe);
4184 		}
4185 	}
4186 
4187 	mutex_exit(&ixgbe->watchdog_lock);
4188 }
4189 
4190 /*
4191  * ixgbe_restart_watchdog_timer - Restart the driver watchdog timer.
4192  */
4193 static void
4194 ixgbe_restart_watchdog_timer(ixgbe_t *ixgbe)
4195 {
4196 	mutex_enter(&ixgbe->watchdog_lock);
4197 
4198 	if (ixgbe->watchdog_start)
4199 		ixgbe_arm_watchdog_timer(ixgbe);
4200 
4201 	mutex_exit(&ixgbe->watchdog_lock);
4202 }
4203 
4204 /*
4205  * ixgbe_stop_watchdog_timer - Stop the driver watchdog timer.
4206  */
4207 void
4208 ixgbe_stop_watchdog_timer(ixgbe_t *ixgbe)
4209 {
4210 	timeout_id_t tid;
4211 
4212 	mutex_enter(&ixgbe->watchdog_lock);
4213 
4214 	ixgbe->watchdog_start = B_FALSE;
4215 	tid = ixgbe->watchdog_tid;
4216 	ixgbe->watchdog_tid = 0;
4217 
4218 	mutex_exit(&ixgbe->watchdog_lock);
4219 
4220 	if (tid != 0)
4221 		(void) untimeout(tid);
4222 }
4223 
4224 /*
4225  * ixgbe_disable_adapter_interrupts - Disable all adapter interrupts.
4226  */
4227 static void
4228 ixgbe_disable_adapter_interrupts(ixgbe_t *ixgbe)
4229 {
4230 	struct ixgbe_hw *hw = &ixgbe->hw;
4231 
4232 	/*
4233 	 * mask all interrupts off
4234 	 */
4235 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, 0xffffffff);
4236 
4237 	/*
4238 	 * for MSI-X, also disable autoclear
4239 	 */
4240 	if (ixgbe->intr_type == DDI_INTR_TYPE_MSIX) {
4241 		IXGBE_WRITE_REG(hw, IXGBE_EIAC, 0x0);
4242 	}
4243 
4244 	IXGBE_WRITE_FLUSH(hw);
4245 }
4246 
4247 /*
4248  * ixgbe_enable_adapter_interrupts - Enable all hardware interrupts.
4249  */
4250 static void
4251 ixgbe_enable_adapter_interrupts(ixgbe_t *ixgbe)
4252 {
4253 	struct ixgbe_hw *hw = &ixgbe->hw;
4254 	uint32_t eiac, eiam;
4255 	uint32_t gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
4256 
4257 	/* interrupt types to enable */
4258 	ixgbe->eims = IXGBE_EIMS_ENABLE_MASK;	/* shared code default */
4259 	ixgbe->eims &= ~IXGBE_EIMS_TCP_TIMER;	/* minus tcp timer */
4260 	ixgbe->eims |= ixgbe->capab->other_intr; /* "other" interrupt types */
4261 
4262 	/* enable automask on "other" causes that this adapter can generate */
4263 	eiam = ixgbe->capab->other_intr;
4264 
4265 	/*
4266 	 * msi-x mode
4267 	 */
4268 	if (ixgbe->intr_type == DDI_INTR_TYPE_MSIX) {
4269 		/* enable autoclear but not on bits 29:20 */
4270 		eiac = (ixgbe->eims & ~IXGBE_OTHER_INTR);
4271 
4272 		/* general purpose interrupt enable */
4273 		gpie |= (IXGBE_GPIE_MSIX_MODE
4274 		    | IXGBE_GPIE_PBA_SUPPORT
4275 		    | IXGBE_GPIE_OCD
4276 		    | IXGBE_GPIE_EIAME);
4277 	/*
4278 	 * non-msi-x mode
4279 	 */
4280 	} else {
4281 
4282 		/* disable autoclear, leave gpie at default */
4283 		eiac = 0;
4284 
4285 		/*
4286 		 * General purpose interrupt enable.
4287 		 * For 82599, X540 and X550, extended interrupt
4288 		 * automask enable only in MSI or MSI-X mode
4289 		 */
4290 		if ((hw->mac.type == ixgbe_mac_82598EB) ||
4291 		    (ixgbe->intr_type == DDI_INTR_TYPE_MSI)) {
4292 			gpie |= IXGBE_GPIE_EIAME;
4293 		}
4294 	}
4295 
4296 	/* Enable specific "other" interrupt types */
4297 	switch (hw->mac.type) {
4298 	case ixgbe_mac_82598EB:
4299 		gpie |= ixgbe->capab->other_gpie;
4300 		break;
4301 
4302 	case ixgbe_mac_82599EB:
4303 	case ixgbe_mac_X540:
4304 	case ixgbe_mac_X550:
4305 	case ixgbe_mac_X550EM_x:
4306 	case ixgbe_mac_X550EM_a:
4307 		gpie |= ixgbe->capab->other_gpie;
4308 
4309 		/* Enable RSC Delay 8us when LRO enabled  */
4310 		if (ixgbe->lro_enable) {
4311 			gpie |= (1 << IXGBE_GPIE_RSC_DELAY_SHIFT);
4312 		}
4313 		break;
4314 
4315 	default:
4316 		break;
4317 	}
4318 
4319 	/* write to interrupt control registers */
4320 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4321 	IXGBE_WRITE_REG(hw, IXGBE_EIAC, eiac);
4322 	IXGBE_WRITE_REG(hw, IXGBE_EIAM, eiam);
4323 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4324 	IXGBE_WRITE_FLUSH(hw);
4325 }
4326 
4327 /*
4328  * ixgbe_loopback_ioctl - Loopback support.
4329  */
4330 enum ioc_reply
4331 ixgbe_loopback_ioctl(ixgbe_t *ixgbe, struct iocblk *iocp, mblk_t *mp)
4332 {
4333 	lb_info_sz_t *lbsp;
4334 	lb_property_t *lbpp;
4335 	uint32_t *lbmp;
4336 	uint32_t size;
4337 	uint32_t value;
4338 
4339 	if (mp->b_cont == NULL)
4340 		return (IOC_INVAL);
4341 
4342 	switch (iocp->ioc_cmd) {
4343 	default:
4344 		return (IOC_INVAL);
4345 
4346 	case LB_GET_INFO_SIZE:
4347 		size = sizeof (lb_info_sz_t);
4348 		if (iocp->ioc_count != size)
4349 			return (IOC_INVAL);
4350 
4351 		value = sizeof (lb_normal);
4352 		value += sizeof (lb_mac);
4353 		value += sizeof (lb_external);
4354 
4355 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
4356 		*lbsp = value;
4357 		break;
4358 
4359 	case LB_GET_INFO:
4360 		value = sizeof (lb_normal);
4361 		value += sizeof (lb_mac);
4362 		value += sizeof (lb_external);
4363 
4364 		size = value;
4365 		if (iocp->ioc_count != size)
4366 			return (IOC_INVAL);
4367 
4368 		value = 0;
4369 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
4370 
4371 		lbpp[value++] = lb_normal;
4372 		lbpp[value++] = lb_mac;
4373 		lbpp[value++] = lb_external;
4374 		break;
4375 
4376 	case LB_GET_MODE:
4377 		size = sizeof (uint32_t);
4378 		if (iocp->ioc_count != size)
4379 			return (IOC_INVAL);
4380 
4381 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
4382 		*lbmp = ixgbe->loopback_mode;
4383 		break;
4384 
4385 	case LB_SET_MODE:
4386 		size = 0;
4387 		if (iocp->ioc_count != sizeof (uint32_t))
4388 			return (IOC_INVAL);
4389 
4390 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
4391 		if (!ixgbe_set_loopback_mode(ixgbe, *lbmp))
4392 			return (IOC_INVAL);
4393 		break;
4394 	}
4395 
4396 	iocp->ioc_count = size;
4397 	iocp->ioc_error = 0;
4398 
4399 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
4400 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4401 		return (IOC_INVAL);
4402 	}
4403 
4404 	return (IOC_REPLY);
4405 }
4406 
4407 /*
4408  * ixgbe_set_loopback_mode - Setup loopback based on the loopback mode.
4409  */
4410 static boolean_t
4411 ixgbe_set_loopback_mode(ixgbe_t *ixgbe, uint32_t mode)
4412 {
4413 	if (mode == ixgbe->loopback_mode)
4414 		return (B_TRUE);
4415 
4416 	ixgbe->loopback_mode = mode;
4417 
4418 	if (mode == IXGBE_LB_NONE) {
4419 		/*
4420 		 * Reset the chip
4421 		 */
4422 		(void) ixgbe_reset(ixgbe);
4423 		return (B_TRUE);
4424 	}
4425 
4426 	mutex_enter(&ixgbe->gen_lock);
4427 
4428 	switch (mode) {
4429 	default:
4430 		mutex_exit(&ixgbe->gen_lock);
4431 		return (B_FALSE);
4432 
4433 	case IXGBE_LB_EXTERNAL:
4434 		break;
4435 
4436 	case IXGBE_LB_INTERNAL_MAC:
4437 		ixgbe_set_internal_mac_loopback(ixgbe);
4438 		break;
4439 	}
4440 
4441 	mutex_exit(&ixgbe->gen_lock);
4442 
4443 	return (B_TRUE);
4444 }
4445 
4446 /*
4447  * ixgbe_set_internal_mac_loopback - Set the internal MAC loopback mode.
4448  */
4449 static void
4450 ixgbe_set_internal_mac_loopback(ixgbe_t *ixgbe)
4451 {
4452 	struct ixgbe_hw *hw;
4453 	uint32_t reg;
4454 	uint8_t atlas;
4455 
4456 	hw = &ixgbe->hw;
4457 
4458 	/*
4459 	 * Setup MAC loopback
4460 	 */
4461 	reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_HLREG0);
4462 	reg |= IXGBE_HLREG0_LPBK;
4463 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_HLREG0, reg);
4464 
4465 	reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_AUTOC);
4466 	reg &= ~IXGBE_AUTOC_LMS_MASK;
4467 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_AUTOC, reg);
4468 
4469 	/*
4470 	 * Disable Atlas Tx lanes to keep packets in loopback and not on wire
4471 	 */
4472 	switch (hw->mac.type) {
4473 	case ixgbe_mac_82598EB:
4474 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_LPBK,
4475 		    &atlas);
4476 		atlas |= IXGBE_ATLAS_PDN_TX_REG_EN;
4477 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_LPBK,
4478 		    atlas);
4479 
4480 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_10G,
4481 		    &atlas);
4482 		atlas |= IXGBE_ATLAS_PDN_TX_10G_QL_ALL;
4483 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_10G,
4484 		    atlas);
4485 
4486 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_1G,
4487 		    &atlas);
4488 		atlas |= IXGBE_ATLAS_PDN_TX_1G_QL_ALL;
4489 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_1G,
4490 		    atlas);
4491 
4492 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_AN,
4493 		    &atlas);
4494 		atlas |= IXGBE_ATLAS_PDN_TX_AN_QL_ALL;
4495 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_AN,
4496 		    atlas);
4497 		break;
4498 
4499 	case ixgbe_mac_82599EB:
4500 	case ixgbe_mac_X540:
4501 	case ixgbe_mac_X550:
4502 	case ixgbe_mac_X550EM_x:
4503 	case ixgbe_mac_X550EM_a:
4504 		reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_AUTOC);
4505 		reg |= (IXGBE_AUTOC_FLU |
4506 		    IXGBE_AUTOC_10G_KX4);
4507 		IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_AUTOC, reg);
4508 
4509 		(void) ixgbe_setup_link(&ixgbe->hw, IXGBE_LINK_SPEED_10GB_FULL,
4510 		    B_FALSE);
4511 		break;
4512 
4513 	default:
4514 		break;
4515 	}
4516 }
4517 
4518 #pragma inline(ixgbe_intr_rx_work)
4519 /*
4520  * ixgbe_intr_rx_work - RX processing of ISR.
4521  */
4522 static void
4523 ixgbe_intr_rx_work(ixgbe_rx_ring_t *rx_ring)
4524 {
4525 	mblk_t *mp;
4526 
4527 	mutex_enter(&rx_ring->rx_lock);
4528 
4529 	mp = ixgbe_ring_rx(rx_ring, IXGBE_POLL_NULL);
4530 	mutex_exit(&rx_ring->rx_lock);
4531 
4532 	if (mp != NULL)
4533 		mac_rx_ring(rx_ring->ixgbe->mac_hdl, rx_ring->ring_handle, mp,
4534 		    rx_ring->ring_gen_num);
4535 }
4536 
4537 #pragma inline(ixgbe_intr_tx_work)
4538 /*
4539  * ixgbe_intr_tx_work - TX processing of ISR.
4540  */
4541 static void
4542 ixgbe_intr_tx_work(ixgbe_tx_ring_t *tx_ring)
4543 {
4544 	ixgbe_t *ixgbe = tx_ring->ixgbe;
4545 
4546 	/*
4547 	 * Recycle the tx descriptors
4548 	 */
4549 	tx_ring->tx_recycle(tx_ring);
4550 
4551 	/*
4552 	 * Schedule the re-transmit
4553 	 */
4554 	if (tx_ring->reschedule &&
4555 	    (tx_ring->tbd_free >= ixgbe->tx_resched_thresh)) {
4556 		tx_ring->reschedule = B_FALSE;
4557 		mac_tx_ring_update(tx_ring->ixgbe->mac_hdl,
4558 		    tx_ring->ring_handle);
4559 		tx_ring->stat_reschedule++;
4560 	}
4561 }
4562 
4563 #pragma inline(ixgbe_intr_other_work)
4564 /*
4565  * ixgbe_intr_other_work - Process interrupt types other than tx/rx
4566  */
4567 static void
4568 ixgbe_intr_other_work(ixgbe_t *ixgbe, uint32_t eicr)
4569 {
4570 	struct ixgbe_hw *hw = &ixgbe->hw;
4571 
4572 	ASSERT(mutex_owned(&ixgbe->gen_lock));
4573 
4574 	/*
4575 	 * handle link status change
4576 	 */
4577 	if (eicr & IXGBE_EICR_LSC) {
4578 		ixgbe_driver_link_check(ixgbe);
4579 		ixgbe_get_hw_state(ixgbe);
4580 	}
4581 
4582 	/*
4583 	 * check for fan failure on adapters with fans
4584 	 */
4585 	if ((ixgbe->capab->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
4586 	    (eicr & IXGBE_EICR_GPI_SDP1)) {
4587 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_OVERTEMP);
4588 
4589 		/*
4590 		 * Disable the adapter interrupts
4591 		 */
4592 		ixgbe_disable_adapter_interrupts(ixgbe);
4593 
4594 		/*
4595 		 * Disable Rx/Tx units
4596 		 */
4597 		(void) ixgbe_stop_adapter(&ixgbe->hw);
4598 
4599 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
4600 		ixgbe_error(ixgbe,
4601 		    "Problem: Network adapter has been stopped "
4602 		    "because the fan has stopped.\n");
4603 		ixgbe_error(ixgbe,
4604 		    "Action: Replace the adapter.\n");
4605 
4606 		/* re-enable the interrupt, which was automasked */
4607 		ixgbe->eims |= IXGBE_EICR_GPI_SDP1;
4608 	}
4609 
4610 	/*
4611 	 * Do SFP check for adapters with hot-plug capability
4612 	 */
4613 	if ((ixgbe->capab->flags & IXGBE_FLAG_SFP_PLUG_CAPABLE) &&
4614 	    ((eicr & IXGBE_EICR_GPI_SDP1_BY_MAC(hw)) ||
4615 	    (eicr & IXGBE_EICR_GPI_SDP2_BY_MAC(hw)))) {
4616 		ixgbe->eicr = eicr;
4617 		if ((ddi_taskq_dispatch(ixgbe->sfp_taskq,
4618 		    ixgbe_sfp_check, (void *)ixgbe,
4619 		    DDI_NOSLEEP)) != DDI_SUCCESS) {
4620 			ixgbe_log(ixgbe, "No memory available to dispatch "
4621 			    "taskq for SFP check");
4622 		}
4623 	}
4624 
4625 	/*
4626 	 * Do over-temperature check for adapters with temp sensor
4627 	 */
4628 	if ((ixgbe->capab->flags & IXGBE_FLAG_TEMP_SENSOR_CAPABLE) &&
4629 	    ((eicr & IXGBE_EICR_GPI_SDP0_BY_MAC(hw)) ||
4630 	    (eicr & IXGBE_EICR_LSC))) {
4631 		ixgbe->eicr = eicr;
4632 		if ((ddi_taskq_dispatch(ixgbe->overtemp_taskq,
4633 		    ixgbe_overtemp_check, (void *)ixgbe,
4634 		    DDI_NOSLEEP)) != DDI_SUCCESS) {
4635 			ixgbe_log(ixgbe, "No memory available to dispatch "
4636 			    "taskq for overtemp check");
4637 		}
4638 	}
4639 
4640 	/*
4641 	 * Process an external PHY interrupt
4642 	 */
4643 	if (hw->device_id == IXGBE_DEV_ID_X550EM_X_10G_T &&
4644 	    (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
4645 		ixgbe->eicr = eicr;
4646 		if ((ddi_taskq_dispatch(ixgbe->phy_taskq,
4647 		    ixgbe_phy_check, (void *)ixgbe,
4648 		    DDI_NOSLEEP)) != DDI_SUCCESS) {
4649 			ixgbe_log(ixgbe, "No memory available to dispatch "
4650 			    "taskq for PHY check");
4651 		}
4652 	}
4653 }
4654 
4655 /*
4656  * ixgbe_intr_legacy - Interrupt handler for legacy interrupts.
4657  */
4658 static uint_t
4659 ixgbe_intr_legacy(void *arg1, void *arg2)
4660 {
4661 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
4662 	struct ixgbe_hw *hw = &ixgbe->hw;
4663 	ixgbe_tx_ring_t *tx_ring;
4664 	ixgbe_rx_ring_t *rx_ring;
4665 	uint32_t eicr;
4666 	mblk_t *mp;
4667 	boolean_t tx_reschedule;
4668 	uint_t result;
4669 
4670 	_NOTE(ARGUNUSED(arg2));
4671 
4672 	mutex_enter(&ixgbe->gen_lock);
4673 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
4674 		mutex_exit(&ixgbe->gen_lock);
4675 		return (DDI_INTR_UNCLAIMED);
4676 	}
4677 
4678 	mp = NULL;
4679 	tx_reschedule = B_FALSE;
4680 
4681 	/*
4682 	 * Any bit set in eicr: claim this interrupt
4683 	 */
4684 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4685 
4686 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
4687 		mutex_exit(&ixgbe->gen_lock);
4688 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4689 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4690 		return (DDI_INTR_CLAIMED);
4691 	}
4692 
4693 	if (eicr) {
4694 		/*
4695 		 * For legacy interrupt, we have only one interrupt,
4696 		 * so we have only one rx ring and one tx ring enabled.
4697 		 */
4698 		ASSERT(ixgbe->num_rx_rings == 1);
4699 		ASSERT(ixgbe->num_tx_rings == 1);
4700 
4701 		/*
4702 		 * For legacy interrupt, rx rings[0] will use RTxQ[0].
4703 		 */
4704 		if (eicr & 0x1) {
4705 			ixgbe->eimc |= IXGBE_EICR_RTX_QUEUE;
4706 			IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4707 			ixgbe->eims |= IXGBE_EICR_RTX_QUEUE;
4708 			/*
4709 			 * Clean the rx descriptors
4710 			 */
4711 			rx_ring = &ixgbe->rx_rings[0];
4712 			mp = ixgbe_ring_rx(rx_ring, IXGBE_POLL_NULL);
4713 		}
4714 
4715 		/*
4716 		 * For legacy interrupt, tx rings[0] will use RTxQ[1].
4717 		 */
4718 		if (eicr & 0x2) {
4719 			/*
4720 			 * Recycle the tx descriptors
4721 			 */
4722 			tx_ring = &ixgbe->tx_rings[0];
4723 			tx_ring->tx_recycle(tx_ring);
4724 
4725 			/*
4726 			 * Schedule the re-transmit
4727 			 */
4728 			tx_reschedule = (tx_ring->reschedule &&
4729 			    (tx_ring->tbd_free >= ixgbe->tx_resched_thresh));
4730 		}
4731 
4732 		/* any interrupt type other than tx/rx */
4733 		if (eicr & ixgbe->capab->other_intr) {
4734 			switch (hw->mac.type) {
4735 			case ixgbe_mac_82598EB:
4736 				ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4737 				break;
4738 
4739 			case ixgbe_mac_82599EB:
4740 			case ixgbe_mac_X540:
4741 			case ixgbe_mac_X550:
4742 			case ixgbe_mac_X550EM_x:
4743 			case ixgbe_mac_X550EM_a:
4744 				ixgbe->eimc = IXGBE_82599_OTHER_INTR;
4745 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4746 				break;
4747 
4748 			default:
4749 				break;
4750 			}
4751 			ixgbe_intr_other_work(ixgbe, eicr);
4752 			ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4753 		}
4754 
4755 		mutex_exit(&ixgbe->gen_lock);
4756 
4757 		result = DDI_INTR_CLAIMED;
4758 	} else {
4759 		mutex_exit(&ixgbe->gen_lock);
4760 
4761 		/*
4762 		 * No interrupt cause bits set: don't claim this interrupt.
4763 		 */
4764 		result = DDI_INTR_UNCLAIMED;
4765 	}
4766 
4767 	/* re-enable the interrupts which were automasked */
4768 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4769 
4770 	/*
4771 	 * Do the following work outside of the gen_lock
4772 	 */
4773 	if (mp != NULL) {
4774 		mac_rx_ring(rx_ring->ixgbe->mac_hdl, rx_ring->ring_handle, mp,
4775 		    rx_ring->ring_gen_num);
4776 	}
4777 
4778 	if (tx_reschedule)  {
4779 		tx_ring->reschedule = B_FALSE;
4780 		mac_tx_ring_update(ixgbe->mac_hdl, tx_ring->ring_handle);
4781 		tx_ring->stat_reschedule++;
4782 	}
4783 
4784 	return (result);
4785 }
4786 
4787 /*
4788  * ixgbe_intr_msi - Interrupt handler for MSI.
4789  */
4790 static uint_t
4791 ixgbe_intr_msi(void *arg1, void *arg2)
4792 {
4793 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
4794 	struct ixgbe_hw *hw = &ixgbe->hw;
4795 	uint32_t eicr;
4796 
4797 	_NOTE(ARGUNUSED(arg2));
4798 
4799 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4800 
4801 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
4802 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4803 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4804 		return (DDI_INTR_CLAIMED);
4805 	}
4806 
4807 	/*
4808 	 * For MSI interrupt, we have only one vector,
4809 	 * so we have only one rx ring and one tx ring enabled.
4810 	 */
4811 	ASSERT(ixgbe->num_rx_rings == 1);
4812 	ASSERT(ixgbe->num_tx_rings == 1);
4813 
4814 	/*
4815 	 * For MSI interrupt, rx rings[0] will use RTxQ[0].
4816 	 */
4817 	if (eicr & 0x1) {
4818 		ixgbe_intr_rx_work(&ixgbe->rx_rings[0]);
4819 	}
4820 
4821 	/*
4822 	 * For MSI interrupt, tx rings[0] will use RTxQ[1].
4823 	 */
4824 	if (eicr & 0x2) {
4825 		ixgbe_intr_tx_work(&ixgbe->tx_rings[0]);
4826 	}
4827 
4828 	/* any interrupt type other than tx/rx */
4829 	if (eicr & ixgbe->capab->other_intr) {
4830 		mutex_enter(&ixgbe->gen_lock);
4831 		switch (hw->mac.type) {
4832 		case ixgbe_mac_82598EB:
4833 			ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4834 			break;
4835 
4836 		case ixgbe_mac_82599EB:
4837 		case ixgbe_mac_X540:
4838 		case ixgbe_mac_X550:
4839 		case ixgbe_mac_X550EM_x:
4840 		case ixgbe_mac_X550EM_a:
4841 			ixgbe->eimc = IXGBE_82599_OTHER_INTR;
4842 			IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4843 			break;
4844 
4845 		default:
4846 			break;
4847 		}
4848 		ixgbe_intr_other_work(ixgbe, eicr);
4849 		ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4850 		mutex_exit(&ixgbe->gen_lock);
4851 	}
4852 
4853 	/* re-enable the interrupts which were automasked */
4854 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4855 
4856 	return (DDI_INTR_CLAIMED);
4857 }
4858 
4859 /*
4860  * ixgbe_intr_msix - Interrupt handler for MSI-X.
4861  */
4862 static uint_t
4863 ixgbe_intr_msix(void *arg1, void *arg2)
4864 {
4865 	ixgbe_intr_vector_t *vect = (ixgbe_intr_vector_t *)arg1;
4866 	ixgbe_t *ixgbe = vect->ixgbe;
4867 	struct ixgbe_hw *hw = &ixgbe->hw;
4868 	uint32_t eicr;
4869 	int r_idx = 0;
4870 
4871 	_NOTE(ARGUNUSED(arg2));
4872 
4873 	/*
4874 	 * Clean each rx ring that has its bit set in the map
4875 	 */
4876 	r_idx = bt_getlowbit(vect->rx_map, 0, (ixgbe->num_rx_rings - 1));
4877 	while (r_idx >= 0) {
4878 		ixgbe_intr_rx_work(&ixgbe->rx_rings[r_idx]);
4879 		r_idx = bt_getlowbit(vect->rx_map, (r_idx + 1),
4880 		    (ixgbe->num_rx_rings - 1));
4881 	}
4882 
4883 	/*
4884 	 * Clean each tx ring that has its bit set in the map
4885 	 */
4886 	r_idx = bt_getlowbit(vect->tx_map, 0, (ixgbe->num_tx_rings - 1));
4887 	while (r_idx >= 0) {
4888 		ixgbe_intr_tx_work(&ixgbe->tx_rings[r_idx]);
4889 		r_idx = bt_getlowbit(vect->tx_map, (r_idx + 1),
4890 		    (ixgbe->num_tx_rings - 1));
4891 	}
4892 
4893 
4894 	/*
4895 	 * Clean other interrupt (link change) that has its bit set in the map
4896 	 */
4897 	if (BT_TEST(vect->other_map, 0) == 1) {
4898 		eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4899 
4900 		if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) !=
4901 		    DDI_FM_OK) {
4902 			ddi_fm_service_impact(ixgbe->dip,
4903 			    DDI_SERVICE_DEGRADED);
4904 			atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4905 			return (DDI_INTR_CLAIMED);
4906 		}
4907 
4908 		/*
4909 		 * Check "other" cause bits: any interrupt type other than tx/rx
4910 		 */
4911 		if (eicr & ixgbe->capab->other_intr) {
4912 			mutex_enter(&ixgbe->gen_lock);
4913 			switch (hw->mac.type) {
4914 			case ixgbe_mac_82598EB:
4915 				ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4916 				ixgbe_intr_other_work(ixgbe, eicr);
4917 				break;
4918 
4919 			case ixgbe_mac_82599EB:
4920 			case ixgbe_mac_X540:
4921 			case ixgbe_mac_X550:
4922 			case ixgbe_mac_X550EM_x:
4923 			case ixgbe_mac_X550EM_a:
4924 				ixgbe->eims |= IXGBE_EICR_RTX_QUEUE;
4925 				ixgbe_intr_other_work(ixgbe, eicr);
4926 				break;
4927 
4928 			default:
4929 				break;
4930 			}
4931 			mutex_exit(&ixgbe->gen_lock);
4932 		}
4933 
4934 		/* re-enable the interrupts which were automasked */
4935 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4936 	}
4937 
4938 	return (DDI_INTR_CLAIMED);
4939 }
4940 
4941 /*
4942  * ixgbe_alloc_intrs - Allocate interrupts for the driver.
4943  *
4944  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
4945  * if not successful, try Legacy.
4946  * ixgbe->intr_force can be used to force sequence to start with
4947  * any of the 3 types.
4948  * If MSI-X is not used, number of tx/rx rings is forced to 1.
4949  */
4950 static int
4951 ixgbe_alloc_intrs(ixgbe_t *ixgbe)
4952 {
4953 	dev_info_t *devinfo;
4954 	int intr_types;
4955 	int rc;
4956 
4957 	devinfo = ixgbe->dip;
4958 
4959 	/*
4960 	 * Get supported interrupt types
4961 	 */
4962 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
4963 
4964 	if (rc != DDI_SUCCESS) {
4965 		ixgbe_log(ixgbe,
4966 		    "Get supported interrupt types failed: %d", rc);
4967 		return (IXGBE_FAILURE);
4968 	}
4969 	IXGBE_DEBUGLOG_1(ixgbe, "Supported interrupt types: %x", intr_types);
4970 
4971 	ixgbe->intr_type = 0;
4972 
4973 	/*
4974 	 * Install MSI-X interrupts
4975 	 */
4976 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
4977 	    (ixgbe->intr_force <= IXGBE_INTR_MSIX)) {
4978 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_MSIX);
4979 		if (rc == IXGBE_SUCCESS)
4980 			return (IXGBE_SUCCESS);
4981 
4982 		ixgbe_log(ixgbe,
4983 		    "Allocate MSI-X failed, trying MSI interrupts...");
4984 	}
4985 
4986 	/*
4987 	 * MSI-X not used, force rings and groups to 1
4988 	 */
4989 	ixgbe->num_rx_rings = 1;
4990 	ixgbe->num_rx_groups = 1;
4991 	ixgbe->num_tx_rings = 1;
4992 	ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
4993 	ixgbe_log(ixgbe,
4994 	    "MSI-X not used, force rings and groups number to 1");
4995 
4996 	/*
4997 	 * Install MSI interrupts
4998 	 */
4999 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
5000 	    (ixgbe->intr_force <= IXGBE_INTR_MSI)) {
5001 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_MSI);
5002 		if (rc == IXGBE_SUCCESS)
5003 			return (IXGBE_SUCCESS);
5004 
5005 		ixgbe_log(ixgbe,
5006 		    "Allocate MSI failed, trying Legacy interrupts...");
5007 	}
5008 
5009 	/*
5010 	 * Install legacy interrupts
5011 	 */
5012 	if (intr_types & DDI_INTR_TYPE_FIXED) {
5013 		/*
5014 		 * Disallow legacy interrupts for X550. X550 has a silicon
5015 		 * bug which prevents Shared Legacy interrupts from working.
5016 		 * For details, please reference:
5017 		 *
5018 		 * Intel Ethernet Controller X550 Specification Update rev. 2.1
5019 		 * May 2016, erratum 22: PCIe Interrupt Status Bit
5020 		 */
5021 		if (ixgbe->hw.mac.type == ixgbe_mac_X550 ||
5022 		    ixgbe->hw.mac.type == ixgbe_mac_X550EM_x ||
5023 		    ixgbe->hw.mac.type == ixgbe_mac_X550EM_a ||
5024 		    ixgbe->hw.mac.type == ixgbe_mac_X550_vf ||
5025 		    ixgbe->hw.mac.type == ixgbe_mac_X550EM_x_vf ||
5026 		    ixgbe->hw.mac.type == ixgbe_mac_X550EM_a_vf) {
5027 			ixgbe_log(ixgbe,
5028 			    "Legacy interrupts are not supported on this "
5029 			    "adapter. Please use MSI or MSI-X instead.");
5030 			return (IXGBE_FAILURE);
5031 		}
5032 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_FIXED);
5033 		if (rc == IXGBE_SUCCESS)
5034 			return (IXGBE_SUCCESS);
5035 
5036 		ixgbe_log(ixgbe,
5037 		    "Allocate Legacy interrupts failed");
5038 	}
5039 
5040 	/*
5041 	 * If none of the 3 types succeeded, return failure
5042 	 */
5043 	return (IXGBE_FAILURE);
5044 }
5045 
5046 /*
5047  * ixgbe_alloc_intr_handles - Allocate interrupt handles.
5048  *
5049  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
5050  * if fewer than 2 handles are available, return failure.
5051  * Upon success, this maps the vectors to rx and tx rings for
5052  * interrupts.
5053  */
5054 static int
5055 ixgbe_alloc_intr_handles(ixgbe_t *ixgbe, int intr_type)
5056 {
5057 	dev_info_t *devinfo;
5058 	int request, count, actual;
5059 	int minimum;
5060 	int rc;
5061 	uint32_t ring_per_group;
5062 
5063 	devinfo = ixgbe->dip;
5064 
5065 	switch (intr_type) {
5066 	case DDI_INTR_TYPE_FIXED:
5067 		request = 1;	/* Request 1 legacy interrupt handle */
5068 		minimum = 1;
5069 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: legacy");
5070 		break;
5071 
5072 	case DDI_INTR_TYPE_MSI:
5073 		request = 1;	/* Request 1 MSI interrupt handle */
5074 		minimum = 1;
5075 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: MSI");
5076 		break;
5077 
5078 	case DDI_INTR_TYPE_MSIX:
5079 		/*
5080 		 * Best number of vectors for the adapter is
5081 		 * (# rx rings + # tx rings), however we will
5082 		 * limit the request number.
5083 		 */
5084 		request = min(16, ixgbe->num_rx_rings + ixgbe->num_tx_rings);
5085 		if (request > ixgbe->capab->max_ring_vect)
5086 			request = ixgbe->capab->max_ring_vect;
5087 		minimum = 1;
5088 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: MSI-X");
5089 		break;
5090 
5091 	default:
5092 		ixgbe_log(ixgbe,
5093 		    "invalid call to ixgbe_alloc_intr_handles(): %d\n",
5094 		    intr_type);
5095 		return (IXGBE_FAILURE);
5096 	}
5097 	IXGBE_DEBUGLOG_2(ixgbe, "interrupt handles requested: %d  minimum: %d",
5098 	    request, minimum);
5099 
5100 	/*
5101 	 * Get number of supported interrupts
5102 	 */
5103 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
5104 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
5105 		ixgbe_log(ixgbe,
5106 		    "Get interrupt number failed. Return: %d, count: %d",
5107 		    rc, count);
5108 		return (IXGBE_FAILURE);
5109 	}
5110 	IXGBE_DEBUGLOG_1(ixgbe, "interrupts supported: %d", count);
5111 
5112 	actual = 0;
5113 	ixgbe->intr_cnt = 0;
5114 	ixgbe->intr_cnt_max = 0;
5115 	ixgbe->intr_cnt_min = 0;
5116 
5117 	/*
5118 	 * Allocate an array of interrupt handles
5119 	 */
5120 	ixgbe->intr_size = request * sizeof (ddi_intr_handle_t);
5121 	ixgbe->htable = kmem_alloc(ixgbe->intr_size, KM_SLEEP);
5122 
5123 	rc = ddi_intr_alloc(devinfo, ixgbe->htable, intr_type, 0,
5124 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
5125 	if (rc != DDI_SUCCESS) {
5126 		ixgbe_log(ixgbe, "Allocate interrupts failed. "
5127 		    "return: %d, request: %d, actual: %d",
5128 		    rc, request, actual);
5129 		goto alloc_handle_fail;
5130 	}
5131 	IXGBE_DEBUGLOG_1(ixgbe, "interrupts actually allocated: %d", actual);
5132 
5133 	/*
5134 	 * upper/lower limit of interrupts
5135 	 */
5136 	ixgbe->intr_cnt = actual;
5137 	ixgbe->intr_cnt_max = request;
5138 	ixgbe->intr_cnt_min = minimum;
5139 
5140 	/*
5141 	 * rss number per group should not exceed the rx interrupt number,
5142 	 * else need to adjust rx ring number.
5143 	 */
5144 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
5145 	ASSERT((ixgbe->num_rx_rings % ixgbe->num_rx_groups) == 0);
5146 	if (actual < ring_per_group) {
5147 		ixgbe->num_rx_rings = ixgbe->num_rx_groups * actual;
5148 		ixgbe_setup_vmdq_rss_conf(ixgbe);
5149 	}
5150 
5151 	/*
5152 	 * Now we know the actual number of vectors.  Here we map the vector
5153 	 * to other, rx rings and tx ring.
5154 	 */
5155 	if (actual < minimum) {
5156 		ixgbe_log(ixgbe, "Insufficient interrupt handles available: %d",
5157 		    actual);
5158 		goto alloc_handle_fail;
5159 	}
5160 
5161 	/*
5162 	 * Get priority for first vector, assume remaining are all the same
5163 	 */
5164 	rc = ddi_intr_get_pri(ixgbe->htable[0], &ixgbe->intr_pri);
5165 	if (rc != DDI_SUCCESS) {
5166 		ixgbe_log(ixgbe,
5167 		    "Get interrupt priority failed: %d", rc);
5168 		goto alloc_handle_fail;
5169 	}
5170 
5171 	rc = ddi_intr_get_cap(ixgbe->htable[0], &ixgbe->intr_cap);
5172 	if (rc != DDI_SUCCESS) {
5173 		ixgbe_log(ixgbe,
5174 		    "Get interrupt cap failed: %d", rc);
5175 		goto alloc_handle_fail;
5176 	}
5177 
5178 	ixgbe->intr_type = intr_type;
5179 
5180 	return (IXGBE_SUCCESS);
5181 
5182 alloc_handle_fail:
5183 	ixgbe_rem_intrs(ixgbe);
5184 
5185 	return (IXGBE_FAILURE);
5186 }
5187 
5188 /*
5189  * ixgbe_add_intr_handlers - Add interrupt handlers based on the interrupt type.
5190  *
5191  * Before adding the interrupt handlers, the interrupt vectors have
5192  * been allocated, and the rx/tx rings have also been allocated.
5193  */
5194 static int
5195 ixgbe_add_intr_handlers(ixgbe_t *ixgbe)
5196 {
5197 	int vector = 0;
5198 	int rc;
5199 
5200 	switch (ixgbe->intr_type) {
5201 	case DDI_INTR_TYPE_MSIX:
5202 		/*
5203 		 * Add interrupt handler for all vectors
5204 		 */
5205 		for (vector = 0; vector < ixgbe->intr_cnt; vector++) {
5206 			/*
5207 			 * install pointer to vect_map[vector]
5208 			 */
5209 			rc = ddi_intr_add_handler(ixgbe->htable[vector],
5210 			    (ddi_intr_handler_t *)ixgbe_intr_msix,
5211 			    (void *)&ixgbe->vect_map[vector], NULL);
5212 
5213 			if (rc != DDI_SUCCESS) {
5214 				ixgbe_log(ixgbe,
5215 				    "Add interrupt handler failed. "
5216 				    "return: %d, vector: %d", rc, vector);
5217 				for (vector--; vector >= 0; vector--) {
5218 					(void) ddi_intr_remove_handler(
5219 					    ixgbe->htable[vector]);
5220 				}
5221 				return (IXGBE_FAILURE);
5222 			}
5223 		}
5224 
5225 		break;
5226 
5227 	case DDI_INTR_TYPE_MSI:
5228 		/*
5229 		 * Add interrupt handlers for the only vector
5230 		 */
5231 		rc = ddi_intr_add_handler(ixgbe->htable[vector],
5232 		    (ddi_intr_handler_t *)ixgbe_intr_msi,
5233 		    (void *)ixgbe, NULL);
5234 
5235 		if (rc != DDI_SUCCESS) {
5236 			ixgbe_log(ixgbe,
5237 			    "Add MSI interrupt handler failed: %d", rc);
5238 			return (IXGBE_FAILURE);
5239 		}
5240 
5241 		break;
5242 
5243 	case DDI_INTR_TYPE_FIXED:
5244 		/*
5245 		 * Add interrupt handlers for the only vector
5246 		 */
5247 		rc = ddi_intr_add_handler(ixgbe->htable[vector],
5248 		    (ddi_intr_handler_t *)ixgbe_intr_legacy,
5249 		    (void *)ixgbe, NULL);
5250 
5251 		if (rc != DDI_SUCCESS) {
5252 			ixgbe_log(ixgbe,
5253 			    "Add legacy interrupt handler failed: %d", rc);
5254 			return (IXGBE_FAILURE);
5255 		}
5256 
5257 		break;
5258 
5259 	default:
5260 		return (IXGBE_FAILURE);
5261 	}
5262 
5263 	return (IXGBE_SUCCESS);
5264 }
5265 
5266 #pragma inline(ixgbe_map_rxring_to_vector)
5267 /*
5268  * ixgbe_map_rxring_to_vector - Map given rx ring to given interrupt vector.
5269  */
5270 static void
5271 ixgbe_map_rxring_to_vector(ixgbe_t *ixgbe, int r_idx, int v_idx)
5272 {
5273 	/*
5274 	 * Set bit in map
5275 	 */
5276 	BT_SET(ixgbe->vect_map[v_idx].rx_map, r_idx);
5277 
5278 	/*
5279 	 * Count bits set
5280 	 */
5281 	ixgbe->vect_map[v_idx].rxr_cnt++;
5282 
5283 	/*
5284 	 * Remember bit position
5285 	 */
5286 	ixgbe->rx_rings[r_idx].intr_vector = v_idx;
5287 	ixgbe->rx_rings[r_idx].vect_bit = 1 << v_idx;
5288 }
5289 
5290 #pragma inline(ixgbe_map_txring_to_vector)
5291 /*
5292  * ixgbe_map_txring_to_vector - Map given tx ring to given interrupt vector.
5293  */
5294 static void
5295 ixgbe_map_txring_to_vector(ixgbe_t *ixgbe, int t_idx, int v_idx)
5296 {
5297 	/*
5298 	 * Set bit in map
5299 	 */
5300 	BT_SET(ixgbe->vect_map[v_idx].tx_map, t_idx);
5301 
5302 	/*
5303 	 * Count bits set
5304 	 */
5305 	ixgbe->vect_map[v_idx].txr_cnt++;
5306 
5307 	/*
5308 	 * Remember bit position
5309 	 */
5310 	ixgbe->tx_rings[t_idx].intr_vector = v_idx;
5311 	ixgbe->tx_rings[t_idx].vect_bit = 1 << v_idx;
5312 }
5313 
5314 /*
5315  * ixgbe_setup_ivar - Set the given entry in the given interrupt vector
5316  * allocation register (IVAR).
5317  * cause:
5318  *   -1 : other cause
5319  *    0 : rx
5320  *    1 : tx
5321  */
5322 static void
5323 ixgbe_setup_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, uint8_t msix_vector,
5324     int8_t cause)
5325 {
5326 	struct ixgbe_hw *hw = &ixgbe->hw;
5327 	u32 ivar, index;
5328 
5329 	switch (hw->mac.type) {
5330 	case ixgbe_mac_82598EB:
5331 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
5332 		if (cause == -1) {
5333 			cause = 0;
5334 		}
5335 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
5336 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
5337 		ivar &= ~(0xFF << (8 * (intr_alloc_entry & 0x3)));
5338 		ivar |= (msix_vector << (8 * (intr_alloc_entry & 0x3)));
5339 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
5340 		break;
5341 
5342 	case ixgbe_mac_82599EB:
5343 	case ixgbe_mac_X540:
5344 	case ixgbe_mac_X550:
5345 	case ixgbe_mac_X550EM_x:
5346 	case ixgbe_mac_X550EM_a:
5347 		if (cause == -1) {
5348 			/* other causes */
5349 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
5350 			index = (intr_alloc_entry & 1) * 8;
5351 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
5352 			ivar &= ~(0xFF << index);
5353 			ivar |= (msix_vector << index);
5354 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
5355 		} else {
5356 			/* tx or rx causes */
5357 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
5358 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
5359 			ivar = IXGBE_READ_REG(hw,
5360 			    IXGBE_IVAR(intr_alloc_entry >> 1));
5361 			ivar &= ~(0xFF << index);
5362 			ivar |= (msix_vector << index);
5363 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
5364 			    ivar);
5365 		}
5366 		break;
5367 
5368 	default:
5369 		break;
5370 	}
5371 }
5372 
5373 /*
5374  * ixgbe_enable_ivar - Enable the given entry by setting the VAL bit of
5375  * given interrupt vector allocation register (IVAR).
5376  * cause:
5377  *   -1 : other cause
5378  *    0 : rx
5379  *    1 : tx
5380  */
5381 static void
5382 ixgbe_enable_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, int8_t cause)
5383 {
5384 	struct ixgbe_hw *hw = &ixgbe->hw;
5385 	u32 ivar, index;
5386 
5387 	switch (hw->mac.type) {
5388 	case ixgbe_mac_82598EB:
5389 		if (cause == -1) {
5390 			cause = 0;
5391 		}
5392 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
5393 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
5394 		ivar |= (IXGBE_IVAR_ALLOC_VAL << (8 *
5395 		    (intr_alloc_entry & 0x3)));
5396 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
5397 		break;
5398 
5399 	case ixgbe_mac_82599EB:
5400 	case ixgbe_mac_X540:
5401 	case ixgbe_mac_X550:
5402 	case ixgbe_mac_X550EM_x:
5403 	case ixgbe_mac_X550EM_a:
5404 		if (cause == -1) {
5405 			/* other causes */
5406 			index = (intr_alloc_entry & 1) * 8;
5407 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
5408 			ivar |= (IXGBE_IVAR_ALLOC_VAL << index);
5409 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
5410 		} else {
5411 			/* tx or rx causes */
5412 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
5413 			ivar = IXGBE_READ_REG(hw,
5414 			    IXGBE_IVAR(intr_alloc_entry >> 1));
5415 			ivar |= (IXGBE_IVAR_ALLOC_VAL << index);
5416 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
5417 			    ivar);
5418 		}
5419 		break;
5420 
5421 	default:
5422 		break;
5423 	}
5424 }
5425 
5426 /*
5427  * ixgbe_disable_ivar - Disble the given entry by clearing the VAL bit of
5428  * given interrupt vector allocation register (IVAR).
5429  * cause:
5430  *   -1 : other cause
5431  *    0 : rx
5432  *    1 : tx
5433  */
5434 static void
5435 ixgbe_disable_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, int8_t cause)
5436 {
5437 	struct ixgbe_hw *hw = &ixgbe->hw;
5438 	u32 ivar, index;
5439 
5440 	switch (hw->mac.type) {
5441 	case ixgbe_mac_82598EB:
5442 		if (cause == -1) {
5443 			cause = 0;
5444 		}
5445 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
5446 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
5447 		ivar &= ~(IXGBE_IVAR_ALLOC_VAL<< (8 *
5448 		    (intr_alloc_entry & 0x3)));
5449 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
5450 		break;
5451 
5452 	case ixgbe_mac_82599EB:
5453 	case ixgbe_mac_X540:
5454 	case ixgbe_mac_X550:
5455 	case ixgbe_mac_X550EM_x:
5456 	case ixgbe_mac_X550EM_a:
5457 		if (cause == -1) {
5458 			/* other causes */
5459 			index = (intr_alloc_entry & 1) * 8;
5460 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
5461 			ivar &= ~(IXGBE_IVAR_ALLOC_VAL << index);
5462 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
5463 		} else {
5464 			/* tx or rx causes */
5465 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
5466 			ivar = IXGBE_READ_REG(hw,
5467 			    IXGBE_IVAR(intr_alloc_entry >> 1));
5468 			ivar &= ~(IXGBE_IVAR_ALLOC_VAL << index);
5469 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
5470 			    ivar);
5471 		}
5472 		break;
5473 
5474 	default:
5475 		break;
5476 	}
5477 }
5478 
5479 /*
5480  * Convert the rx ring index driver maintained to the rx ring index
5481  * in h/w.
5482  */
5483 static uint32_t
5484 ixgbe_get_hw_rx_index(ixgbe_t *ixgbe, uint32_t sw_rx_index)
5485 {
5486 
5487 	struct ixgbe_hw *hw = &ixgbe->hw;
5488 	uint32_t rx_ring_per_group, hw_rx_index;
5489 
5490 	if (ixgbe->classify_mode == IXGBE_CLASSIFY_RSS ||
5491 	    ixgbe->classify_mode == IXGBE_CLASSIFY_NONE) {
5492 		return (sw_rx_index);
5493 	} else if (ixgbe->classify_mode == IXGBE_CLASSIFY_VMDQ) {
5494 		switch (hw->mac.type) {
5495 		case ixgbe_mac_82598EB:
5496 			return (sw_rx_index);
5497 
5498 		case ixgbe_mac_82599EB:
5499 		case ixgbe_mac_X540:
5500 		case ixgbe_mac_X550:
5501 		case ixgbe_mac_X550EM_x:
5502 		case ixgbe_mac_X550EM_a:
5503 			return (sw_rx_index * 2);
5504 
5505 		default:
5506 			break;
5507 		}
5508 	} else if (ixgbe->classify_mode == IXGBE_CLASSIFY_VMDQ_RSS) {
5509 		rx_ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
5510 
5511 		switch (hw->mac.type) {
5512 		case ixgbe_mac_82598EB:
5513 			hw_rx_index = (sw_rx_index / rx_ring_per_group) *
5514 			    16 + (sw_rx_index % rx_ring_per_group);
5515 			return (hw_rx_index);
5516 
5517 		case ixgbe_mac_82599EB:
5518 		case ixgbe_mac_X540:
5519 		case ixgbe_mac_X550:
5520 		case ixgbe_mac_X550EM_x:
5521 		case ixgbe_mac_X550EM_a:
5522 			if (ixgbe->num_rx_groups > 32) {
5523 				hw_rx_index = (sw_rx_index /
5524 				    rx_ring_per_group) * 2 +
5525 				    (sw_rx_index % rx_ring_per_group);
5526 			} else {
5527 				hw_rx_index = (sw_rx_index /
5528 				    rx_ring_per_group) * 4 +
5529 				    (sw_rx_index % rx_ring_per_group);
5530 			}
5531 			return (hw_rx_index);
5532 
5533 		default:
5534 			break;
5535 		}
5536 	}
5537 
5538 	/*
5539 	 * Should never reach. Just to make compiler happy.
5540 	 */
5541 	return (sw_rx_index);
5542 }
5543 
5544 /*
5545  * ixgbe_map_intrs_to_vectors - Map different interrupts to MSI-X vectors.
5546  *
5547  * For MSI-X, here will map rx interrupt, tx interrupt and other interrupt
5548  * to vector[0 - (intr_cnt -1)].
5549  */
5550 static int
5551 ixgbe_map_intrs_to_vectors(ixgbe_t *ixgbe)
5552 {
5553 	int i, vector = 0;
5554 
5555 	/* initialize vector map */
5556 	bzero(&ixgbe->vect_map, sizeof (ixgbe->vect_map));
5557 	for (i = 0; i < ixgbe->intr_cnt; i++) {
5558 		ixgbe->vect_map[i].ixgbe = ixgbe;
5559 	}
5560 
5561 	/*
5562 	 * non-MSI-X case is very simple: rx rings[0] on RTxQ[0],
5563 	 * tx rings[0] on RTxQ[1].
5564 	 */
5565 	if (ixgbe->intr_type != DDI_INTR_TYPE_MSIX) {
5566 		ixgbe_map_rxring_to_vector(ixgbe, 0, 0);
5567 		ixgbe_map_txring_to_vector(ixgbe, 0, 1);
5568 		return (IXGBE_SUCCESS);
5569 	}
5570 
5571 	/*
5572 	 * Interrupts/vectors mapping for MSI-X
5573 	 */
5574 
5575 	/*
5576 	 * Map other interrupt to vector 0,
5577 	 * Set bit in map and count the bits set.
5578 	 */
5579 	BT_SET(ixgbe->vect_map[vector].other_map, 0);
5580 	ixgbe->vect_map[vector].other_cnt++;
5581 
5582 	/*
5583 	 * Map rx ring interrupts to vectors
5584 	 */
5585 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
5586 		ixgbe_map_rxring_to_vector(ixgbe, i, vector);
5587 		vector = (vector +1) % ixgbe->intr_cnt;
5588 	}
5589 
5590 	/*
5591 	 * Map tx ring interrupts to vectors
5592 	 */
5593 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
5594 		ixgbe_map_txring_to_vector(ixgbe, i, vector);
5595 		vector = (vector +1) % ixgbe->intr_cnt;
5596 	}
5597 
5598 	return (IXGBE_SUCCESS);
5599 }
5600 
5601 /*
5602  * ixgbe_setup_adapter_vector - Setup the adapter interrupt vector(s).
5603  *
5604  * This relies on ring/vector mapping already set up in the
5605  * vect_map[] structures
5606  */
5607 static void
5608 ixgbe_setup_adapter_vector(ixgbe_t *ixgbe)
5609 {
5610 	struct ixgbe_hw *hw = &ixgbe->hw;
5611 	ixgbe_intr_vector_t *vect;	/* vector bitmap */
5612 	int r_idx;	/* ring index */
5613 	int v_idx;	/* vector index */
5614 	uint32_t hw_index;
5615 
5616 	/*
5617 	 * Clear any previous entries
5618 	 */
5619 	switch (hw->mac.type) {
5620 	case ixgbe_mac_82598EB:
5621 		for (v_idx = 0; v_idx < 25; v_idx++)
5622 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(v_idx), 0);
5623 		break;
5624 
5625 	case ixgbe_mac_82599EB:
5626 	case ixgbe_mac_X540:
5627 	case ixgbe_mac_X550:
5628 	case ixgbe_mac_X550EM_x:
5629 	case ixgbe_mac_X550EM_a:
5630 		for (v_idx = 0; v_idx < 64; v_idx++)
5631 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(v_idx), 0);
5632 		IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, 0);
5633 		break;
5634 
5635 	default:
5636 		break;
5637 	}
5638 
5639 	/*
5640 	 * For non MSI-X interrupt, rx rings[0] will use RTxQ[0], and
5641 	 * tx rings[0] will use RTxQ[1].
5642 	 */
5643 	if (ixgbe->intr_type != DDI_INTR_TYPE_MSIX) {
5644 		ixgbe_setup_ivar(ixgbe, 0, 0, 0);
5645 		ixgbe_setup_ivar(ixgbe, 0, 1, 1);
5646 		return;
5647 	}
5648 
5649 	/*
5650 	 * For MSI-X interrupt, "Other" is always on vector[0].
5651 	 */
5652 	ixgbe_setup_ivar(ixgbe, IXGBE_IVAR_OTHER_CAUSES_INDEX, 0, -1);
5653 
5654 	/*
5655 	 * For each interrupt vector, populate the IVAR table
5656 	 */
5657 	for (v_idx = 0; v_idx < ixgbe->intr_cnt; v_idx++) {
5658 		vect = &ixgbe->vect_map[v_idx];
5659 
5660 		/*
5661 		 * For each rx ring bit set
5662 		 */
5663 		r_idx = bt_getlowbit(vect->rx_map, 0,
5664 		    (ixgbe->num_rx_rings - 1));
5665 
5666 		while (r_idx >= 0) {
5667 			hw_index = ixgbe->rx_rings[r_idx].hw_index;
5668 			ixgbe_setup_ivar(ixgbe, hw_index, v_idx, 0);
5669 			r_idx = bt_getlowbit(vect->rx_map, (r_idx + 1),
5670 			    (ixgbe->num_rx_rings - 1));
5671 		}
5672 
5673 		/*
5674 		 * For each tx ring bit set
5675 		 */
5676 		r_idx = bt_getlowbit(vect->tx_map, 0,
5677 		    (ixgbe->num_tx_rings - 1));
5678 
5679 		while (r_idx >= 0) {
5680 			ixgbe_setup_ivar(ixgbe, r_idx, v_idx, 1);
5681 			r_idx = bt_getlowbit(vect->tx_map, (r_idx + 1),
5682 			    (ixgbe->num_tx_rings - 1));
5683 		}
5684 	}
5685 }
5686 
5687 /*
5688  * ixgbe_rem_intr_handlers - Remove the interrupt handlers.
5689  */
5690 static void
5691 ixgbe_rem_intr_handlers(ixgbe_t *ixgbe)
5692 {
5693 	int i;
5694 	int rc;
5695 
5696 	for (i = 0; i < ixgbe->intr_cnt; i++) {
5697 		rc = ddi_intr_remove_handler(ixgbe->htable[i]);
5698 		if (rc != DDI_SUCCESS) {
5699 			IXGBE_DEBUGLOG_1(ixgbe,
5700 			    "Remove intr handler failed: %d", rc);
5701 		}
5702 	}
5703 }
5704 
5705 /*
5706  * ixgbe_rem_intrs - Remove the allocated interrupts.
5707  */
5708 static void
5709 ixgbe_rem_intrs(ixgbe_t *ixgbe)
5710 {
5711 	int i;
5712 	int rc;
5713 
5714 	for (i = 0; i < ixgbe->intr_cnt; i++) {
5715 		rc = ddi_intr_free(ixgbe->htable[i]);
5716 		if (rc != DDI_SUCCESS) {
5717 			IXGBE_DEBUGLOG_1(ixgbe,
5718 			    "Free intr failed: %d", rc);
5719 		}
5720 	}
5721 
5722 	kmem_free(ixgbe->htable, ixgbe->intr_size);
5723 	ixgbe->htable = NULL;
5724 }
5725 
5726 /*
5727  * ixgbe_enable_intrs - Enable all the ddi interrupts.
5728  */
5729 static int
5730 ixgbe_enable_intrs(ixgbe_t *ixgbe)
5731 {
5732 	int i;
5733 	int rc;
5734 
5735 	/*
5736 	 * Enable interrupts
5737 	 */
5738 	if (ixgbe->intr_cap & DDI_INTR_FLAG_BLOCK) {
5739 		/*
5740 		 * Call ddi_intr_block_enable() for MSI
5741 		 */
5742 		rc = ddi_intr_block_enable(ixgbe->htable, ixgbe->intr_cnt);
5743 		if (rc != DDI_SUCCESS) {
5744 			ixgbe_log(ixgbe,
5745 			    "Enable block intr failed: %d", rc);
5746 			return (IXGBE_FAILURE);
5747 		}
5748 	} else {
5749 		/*
5750 		 * Call ddi_intr_enable() for Legacy/MSI non block enable
5751 		 */
5752 		for (i = 0; i < ixgbe->intr_cnt; i++) {
5753 			rc = ddi_intr_enable(ixgbe->htable[i]);
5754 			if (rc != DDI_SUCCESS) {
5755 				ixgbe_log(ixgbe,
5756 				    "Enable intr failed: %d", rc);
5757 				return (IXGBE_FAILURE);
5758 			}
5759 		}
5760 	}
5761 
5762 	return (IXGBE_SUCCESS);
5763 }
5764 
5765 /*
5766  * ixgbe_disable_intrs - Disable all the interrupts.
5767  */
5768 static int
5769 ixgbe_disable_intrs(ixgbe_t *ixgbe)
5770 {
5771 	int i;
5772 	int rc;
5773 
5774 	/*
5775 	 * Disable all interrupts
5776 	 */
5777 	if (ixgbe->intr_cap & DDI_INTR_FLAG_BLOCK) {
5778 		rc = ddi_intr_block_disable(ixgbe->htable, ixgbe->intr_cnt);
5779 		if (rc != DDI_SUCCESS) {
5780 			ixgbe_log(ixgbe,
5781 			    "Disable block intr failed: %d", rc);
5782 			return (IXGBE_FAILURE);
5783 		}
5784 	} else {
5785 		for (i = 0; i < ixgbe->intr_cnt; i++) {
5786 			rc = ddi_intr_disable(ixgbe->htable[i]);
5787 			if (rc != DDI_SUCCESS) {
5788 				ixgbe_log(ixgbe,
5789 				    "Disable intr failed: %d", rc);
5790 				return (IXGBE_FAILURE);
5791 			}
5792 		}
5793 	}
5794 
5795 	return (IXGBE_SUCCESS);
5796 }
5797 
5798 /*
5799  * ixgbe_get_hw_state - Get and save parameters related to adapter hardware.
5800  */
5801 static void
5802 ixgbe_get_hw_state(ixgbe_t *ixgbe)
5803 {
5804 	struct ixgbe_hw *hw = &ixgbe->hw;
5805 	ixgbe_link_speed speed = 0;
5806 	boolean_t link_up = B_FALSE;
5807 	uint32_t pcs1g_anlp = 0;
5808 
5809 	ASSERT(mutex_owned(&ixgbe->gen_lock));
5810 	ixgbe->param_lp_1000fdx_cap = 0;
5811 	ixgbe->param_lp_100fdx_cap  = 0;
5812 
5813 	/* check for link, don't wait */
5814 	(void) ixgbe_check_link(hw, &speed, &link_up, B_FALSE);
5815 
5816 	/*
5817 	 * Update the observed Link Partner's capabilities. Not all adapters
5818 	 * can provide full information on the LP's capable speeds, so we
5819 	 * provide what we can.
5820 	 */
5821 	if (link_up) {
5822 		pcs1g_anlp = IXGBE_READ_REG(hw, IXGBE_PCS1GANLP);
5823 
5824 		ixgbe->param_lp_1000fdx_cap =
5825 		    (pcs1g_anlp & IXGBE_PCS1GANLP_LPFD) ? 1 : 0;
5826 		ixgbe->param_lp_100fdx_cap =
5827 		    (pcs1g_anlp & IXGBE_PCS1GANLP_LPFD) ? 1 : 0;
5828 	}
5829 
5830 	/*
5831 	 * Update GLD's notion of the adapter's currently advertised speeds.
5832 	 * Since the common code doesn't always record the current autonegotiate
5833 	 * settings in the phy struct for all parts (specifically, adapters with
5834 	 * SFPs) we first test to see if it is 0, and if so, we fall back to
5835 	 * using the adapter's speed capabilities which we saved during instance
5836 	 * init in ixgbe_init_params().
5837 	 *
5838 	 * Adapters with SFPs will always be shown as advertising all of their
5839 	 * supported speeds, and adapters with baseT PHYs (where the phy struct
5840 	 * is maintained by the common code) will always have a factual view of
5841 	 * their currently-advertised speeds. In the case of SFPs, this is
5842 	 * acceptable as we default to advertising all speeds that the adapter
5843 	 * claims to support, and those properties are immutable; unlike on
5844 	 * baseT (copper) PHYs, where speeds can be enabled or disabled at will.
5845 	 */
5846 	speed = hw->phy.autoneg_advertised;
5847 	if (speed == 0)
5848 		speed = ixgbe->speeds_supported;
5849 
5850 	ixgbe->param_adv_10000fdx_cap =
5851 	    (speed & IXGBE_LINK_SPEED_10GB_FULL) ? 1 : 0;
5852 	ixgbe->param_adv_5000fdx_cap =
5853 	    (speed & IXGBE_LINK_SPEED_5GB_FULL) ? 1 : 0;
5854 	ixgbe->param_adv_2500fdx_cap =
5855 	    (speed & IXGBE_LINK_SPEED_2_5GB_FULL) ? 1 : 0;
5856 	ixgbe->param_adv_1000fdx_cap =
5857 	    (speed & IXGBE_LINK_SPEED_1GB_FULL) ? 1 : 0;
5858 	ixgbe->param_adv_100fdx_cap =
5859 	    (speed & IXGBE_LINK_SPEED_100_FULL) ? 1 : 0;
5860 }
5861 
5862 /*
5863  * ixgbe_get_driver_control - Notify that driver is in control of device.
5864  */
5865 static void
5866 ixgbe_get_driver_control(struct ixgbe_hw *hw)
5867 {
5868 	uint32_t ctrl_ext;
5869 
5870 	/*
5871 	 * Notify firmware that driver is in control of device
5872 	 */
5873 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5874 	ctrl_ext |= IXGBE_CTRL_EXT_DRV_LOAD;
5875 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5876 }
5877 
5878 /*
5879  * ixgbe_release_driver_control - Notify that driver is no longer in control
5880  * of device.
5881  */
5882 static void
5883 ixgbe_release_driver_control(struct ixgbe_hw *hw)
5884 {
5885 	uint32_t ctrl_ext;
5886 
5887 	/*
5888 	 * Notify firmware that driver is no longer in control of device
5889 	 */
5890 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5891 	ctrl_ext &= ~IXGBE_CTRL_EXT_DRV_LOAD;
5892 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5893 }
5894 
5895 /*
5896  * ixgbe_atomic_reserve - Atomic decrease operation.
5897  */
5898 int
5899 ixgbe_atomic_reserve(uint32_t *count_p, uint32_t n)
5900 {
5901 	uint32_t oldval;
5902 	uint32_t newval;
5903 
5904 	/*
5905 	 * ATOMICALLY
5906 	 */
5907 	do {
5908 		oldval = *count_p;
5909 		if (oldval < n)
5910 			return (-1);
5911 		newval = oldval - n;
5912 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
5913 
5914 	return (newval);
5915 }
5916 
5917 /*
5918  * ixgbe_mc_table_itr - Traverse the entries in the multicast table.
5919  */
5920 static uint8_t *
5921 ixgbe_mc_table_itr(struct ixgbe_hw *hw, uint8_t **upd_ptr, uint32_t *vmdq)
5922 {
5923 	uint8_t *addr = *upd_ptr;
5924 	uint8_t *new_ptr;
5925 
5926 	_NOTE(ARGUNUSED(hw));
5927 	_NOTE(ARGUNUSED(vmdq));
5928 
5929 	new_ptr = addr + IXGBE_ETH_LENGTH_OF_ADDRESS;
5930 	*upd_ptr = new_ptr;
5931 	return (addr);
5932 }
5933 
5934 /*
5935  * FMA support
5936  */
5937 int
5938 ixgbe_check_acc_handle(ddi_acc_handle_t handle)
5939 {
5940 	ddi_fm_error_t de;
5941 
5942 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
5943 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
5944 	return (de.fme_status);
5945 }
5946 
5947 int
5948 ixgbe_check_dma_handle(ddi_dma_handle_t handle)
5949 {
5950 	ddi_fm_error_t de;
5951 
5952 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
5953 	return (de.fme_status);
5954 }
5955 
5956 /*
5957  * ixgbe_fm_error_cb - The IO fault service error handling callback function.
5958  */
5959 static int
5960 ixgbe_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
5961 {
5962 	_NOTE(ARGUNUSED(impl_data));
5963 	/*
5964 	 * as the driver can always deal with an error in any dma or
5965 	 * access handle, we can just return the fme_status value.
5966 	 */
5967 	pci_ereport_post(dip, err, NULL);
5968 	return (err->fme_status);
5969 }
5970 
5971 static void
5972 ixgbe_fm_init(ixgbe_t *ixgbe)
5973 {
5974 	ddi_iblock_cookie_t iblk;
5975 	int fma_dma_flag;
5976 
5977 	/*
5978 	 * Only register with IO Fault Services if we have some capability
5979 	 */
5980 	if (ixgbe->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
5981 		ixgbe_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
5982 	} else {
5983 		ixgbe_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
5984 	}
5985 
5986 	if (ixgbe->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
5987 		fma_dma_flag = 1;
5988 	} else {
5989 		fma_dma_flag = 0;
5990 	}
5991 
5992 	ixgbe_set_fma_flags(fma_dma_flag);
5993 
5994 	if (ixgbe->fm_capabilities) {
5995 
5996 		/*
5997 		 * Register capabilities with IO Fault Services
5998 		 */
5999 		ddi_fm_init(ixgbe->dip, &ixgbe->fm_capabilities, &iblk);
6000 
6001 		/*
6002 		 * Initialize pci ereport capabilities if ereport capable
6003 		 */
6004 		if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities) ||
6005 		    DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
6006 			pci_ereport_setup(ixgbe->dip);
6007 
6008 		/*
6009 		 * Register error callback if error callback capable
6010 		 */
6011 		if (DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
6012 			ddi_fm_handler_register(ixgbe->dip,
6013 			    ixgbe_fm_error_cb, (void*) ixgbe);
6014 	}
6015 }
6016 
6017 static void
6018 ixgbe_fm_fini(ixgbe_t *ixgbe)
6019 {
6020 	/*
6021 	 * Only unregister FMA capabilities if they are registered
6022 	 */
6023 	if (ixgbe->fm_capabilities) {
6024 
6025 		/*
6026 		 * Release any resources allocated by pci_ereport_setup()
6027 		 */
6028 		if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities) ||
6029 		    DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
6030 			pci_ereport_teardown(ixgbe->dip);
6031 
6032 		/*
6033 		 * Un-register error callback if error callback capable
6034 		 */
6035 		if (DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
6036 			ddi_fm_handler_unregister(ixgbe->dip);
6037 
6038 		/*
6039 		 * Unregister from IO Fault Service
6040 		 */
6041 		ddi_fm_fini(ixgbe->dip);
6042 	}
6043 }
6044 
6045 void
6046 ixgbe_fm_ereport(ixgbe_t *ixgbe, char *detail)
6047 {
6048 	uint64_t ena;
6049 	char buf[FM_MAX_CLASS];
6050 
6051 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
6052 	ena = fm_ena_generate(0, FM_ENA_FMT1);
6053 	if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities)) {
6054 		ddi_fm_ereport_post(ixgbe->dip, buf, ena, DDI_NOSLEEP,
6055 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
6056 	}
6057 }
6058 
6059 static int
6060 ixgbe_ring_start(mac_ring_driver_t rh, uint64_t mr_gen_num)
6061 {
6062 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)rh;
6063 
6064 	mutex_enter(&rx_ring->rx_lock);
6065 	rx_ring->ring_gen_num = mr_gen_num;
6066 	mutex_exit(&rx_ring->rx_lock);
6067 	return (0);
6068 }
6069 
6070 /*
6071  * Get the global ring index by a ring index within a group.
6072  */
6073 static int
6074 ixgbe_get_rx_ring_index(ixgbe_t *ixgbe, int gindex, int rindex)
6075 {
6076 	ixgbe_rx_ring_t *rx_ring;
6077 	int i;
6078 
6079 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
6080 		rx_ring = &ixgbe->rx_rings[i];
6081 		if (rx_ring->group_index == gindex)
6082 			rindex--;
6083 		if (rindex < 0)
6084 			return (i);
6085 	}
6086 
6087 	return (-1);
6088 }
6089 
6090 /*
6091  * Callback funtion for MAC layer to register all rings.
6092  */
6093 /* ARGSUSED */
6094 void
6095 ixgbe_fill_ring(void *arg, mac_ring_type_t rtype, const int group_index,
6096     const int ring_index, mac_ring_info_t *infop, mac_ring_handle_t rh)
6097 {
6098 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
6099 	mac_intr_t *mintr = &infop->mri_intr;
6100 
6101 	switch (rtype) {
6102 	case MAC_RING_TYPE_RX: {
6103 		/*
6104 		 * 'index' is the ring index within the group.
6105 		 * Need to get the global ring index by searching in groups.
6106 		 */
6107 		int global_ring_index = ixgbe_get_rx_ring_index(
6108 		    ixgbe, group_index, ring_index);
6109 
6110 		ASSERT(global_ring_index >= 0);
6111 
6112 		ixgbe_rx_ring_t *rx_ring = &ixgbe->rx_rings[global_ring_index];
6113 		rx_ring->ring_handle = rh;
6114 
6115 		infop->mri_driver = (mac_ring_driver_t)rx_ring;
6116 		infop->mri_start = ixgbe_ring_start;
6117 		infop->mri_stop = NULL;
6118 		infop->mri_poll = ixgbe_ring_rx_poll;
6119 		infop->mri_stat = ixgbe_rx_ring_stat;
6120 
6121 		mintr->mi_handle = (mac_intr_handle_t)rx_ring;
6122 		mintr->mi_enable = ixgbe_rx_ring_intr_enable;
6123 		mintr->mi_disable = ixgbe_rx_ring_intr_disable;
6124 		if (ixgbe->intr_type &
6125 		    (DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI)) {
6126 			mintr->mi_ddi_handle =
6127 			    ixgbe->htable[rx_ring->intr_vector];
6128 		}
6129 
6130 		break;
6131 	}
6132 	case MAC_RING_TYPE_TX: {
6133 		ASSERT(group_index == -1);
6134 		ASSERT(ring_index < ixgbe->num_tx_rings);
6135 
6136 		ixgbe_tx_ring_t *tx_ring = &ixgbe->tx_rings[ring_index];
6137 		tx_ring->ring_handle = rh;
6138 
6139 		infop->mri_driver = (mac_ring_driver_t)tx_ring;
6140 		infop->mri_start = NULL;
6141 		infop->mri_stop = NULL;
6142 		infop->mri_tx = ixgbe_ring_tx;
6143 		infop->mri_stat = ixgbe_tx_ring_stat;
6144 		if (ixgbe->intr_type &
6145 		    (DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI)) {
6146 			mintr->mi_ddi_handle =
6147 			    ixgbe->htable[tx_ring->intr_vector];
6148 		}
6149 		break;
6150 	}
6151 	default:
6152 		break;
6153 	}
6154 }
6155 
6156 /*
6157  * Callback funtion for MAC layer to register all groups.
6158  */
6159 void
6160 ixgbe_fill_group(void *arg, mac_ring_type_t rtype, const int index,
6161     mac_group_info_t *infop, mac_group_handle_t gh)
6162 {
6163 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
6164 
6165 	switch (rtype) {
6166 	case MAC_RING_TYPE_RX: {
6167 		ixgbe_rx_group_t *rx_group;
6168 
6169 		rx_group = &ixgbe->rx_groups[index];
6170 		rx_group->group_handle = gh;
6171 
6172 		infop->mgi_driver = (mac_group_driver_t)rx_group;
6173 		infop->mgi_start = NULL;
6174 		infop->mgi_stop = NULL;
6175 		infop->mgi_addmac = ixgbe_addmac;
6176 		infop->mgi_remmac = ixgbe_remmac;
6177 		infop->mgi_count = (ixgbe->num_rx_rings / ixgbe->num_rx_groups);
6178 
6179 		break;
6180 	}
6181 	case MAC_RING_TYPE_TX:
6182 		break;
6183 	default:
6184 		break;
6185 	}
6186 }
6187 
6188 /*
6189  * Enable interrupt on the specificed rx ring.
6190  */
6191 int
6192 ixgbe_rx_ring_intr_enable(mac_intr_handle_t intrh)
6193 {
6194 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)intrh;
6195 	ixgbe_t *ixgbe = rx_ring->ixgbe;
6196 	int r_idx = rx_ring->index;
6197 	int hw_r_idx = rx_ring->hw_index;
6198 	int v_idx = rx_ring->intr_vector;
6199 
6200 	mutex_enter(&ixgbe->gen_lock);
6201 	if (ixgbe->ixgbe_state & IXGBE_INTR_ADJUST) {
6202 		mutex_exit(&ixgbe->gen_lock);
6203 		/*
6204 		 * Simply return 0.
6205 		 * Interrupts are being adjusted. ixgbe_intr_adjust()
6206 		 * will eventually re-enable the interrupt when it's
6207 		 * done with the adjustment.
6208 		 */
6209 		return (0);
6210 	}
6211 
6212 	/*
6213 	 * To enable interrupt by setting the VAL bit of given interrupt
6214 	 * vector allocation register (IVAR).
6215 	 */
6216 	ixgbe_enable_ivar(ixgbe, hw_r_idx, 0);
6217 
6218 	BT_SET(ixgbe->vect_map[v_idx].rx_map, r_idx);
6219 
6220 	/*
6221 	 * Trigger a Rx interrupt on this ring
6222 	 */
6223 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_EICS, (1 << v_idx));
6224 	IXGBE_WRITE_FLUSH(&ixgbe->hw);
6225 
6226 	mutex_exit(&ixgbe->gen_lock);
6227 
6228 	return (0);
6229 }
6230 
6231 /*
6232  * Disable interrupt on the specificed rx ring.
6233  */
6234 int
6235 ixgbe_rx_ring_intr_disable(mac_intr_handle_t intrh)
6236 {
6237 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)intrh;
6238 	ixgbe_t *ixgbe = rx_ring->ixgbe;
6239 	int r_idx = rx_ring->index;
6240 	int hw_r_idx = rx_ring->hw_index;
6241 	int v_idx = rx_ring->intr_vector;
6242 
6243 	mutex_enter(&ixgbe->gen_lock);
6244 	if (ixgbe->ixgbe_state & IXGBE_INTR_ADJUST) {
6245 		mutex_exit(&ixgbe->gen_lock);
6246 		/*
6247 		 * Simply return 0.
6248 		 * In the rare case where an interrupt is being
6249 		 * disabled while interrupts are being adjusted,
6250 		 * we don't fail the operation. No interrupts will
6251 		 * be generated while they are adjusted, and
6252 		 * ixgbe_intr_adjust() will cause the interrupts
6253 		 * to be re-enabled once it completes. Note that
6254 		 * in this case, packets may be delivered to the
6255 		 * stack via interrupts before xgbe_rx_ring_intr_enable()
6256 		 * is called again. This is acceptable since interrupt
6257 		 * adjustment is infrequent, and the stack will be
6258 		 * able to handle these packets.
6259 		 */
6260 		return (0);
6261 	}
6262 
6263 	/*
6264 	 * To disable interrupt by clearing the VAL bit of given interrupt
6265 	 * vector allocation register (IVAR).
6266 	 */
6267 	ixgbe_disable_ivar(ixgbe, hw_r_idx, 0);
6268 
6269 	BT_CLEAR(ixgbe->vect_map[v_idx].rx_map, r_idx);
6270 
6271 	mutex_exit(&ixgbe->gen_lock);
6272 
6273 	return (0);
6274 }
6275 
6276 /*
6277  * Add a mac address.
6278  */
6279 static int
6280 ixgbe_addmac(void *arg, const uint8_t *mac_addr)
6281 {
6282 	ixgbe_rx_group_t *rx_group = (ixgbe_rx_group_t *)arg;
6283 	ixgbe_t *ixgbe = rx_group->ixgbe;
6284 	struct ixgbe_hw *hw = &ixgbe->hw;
6285 	int slot, i;
6286 
6287 	mutex_enter(&ixgbe->gen_lock);
6288 
6289 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
6290 		mutex_exit(&ixgbe->gen_lock);
6291 		return (ECANCELED);
6292 	}
6293 
6294 	if (ixgbe->unicst_avail == 0) {
6295 		/* no slots available */
6296 		mutex_exit(&ixgbe->gen_lock);
6297 		return (ENOSPC);
6298 	}
6299 
6300 	/*
6301 	 * The first ixgbe->num_rx_groups slots are reserved for each respective
6302 	 * group. The rest slots are shared by all groups. While adding a
6303 	 * MAC address, reserved slots are firstly checked then the shared
6304 	 * slots are searched.
6305 	 */
6306 	slot = -1;
6307 	if (ixgbe->unicst_addr[rx_group->index].mac.set == 1) {
6308 		for (i = ixgbe->num_rx_groups; i < ixgbe->unicst_total; i++) {
6309 			if (ixgbe->unicst_addr[i].mac.set == 0) {
6310 				slot = i;
6311 				break;
6312 			}
6313 		}
6314 	} else {
6315 		slot = rx_group->index;
6316 	}
6317 
6318 	if (slot == -1) {
6319 		/* no slots available */
6320 		mutex_exit(&ixgbe->gen_lock);
6321 		return (ENOSPC);
6322 	}
6323 
6324 	bcopy(mac_addr, ixgbe->unicst_addr[slot].mac.addr, ETHERADDRL);
6325 	(void) ixgbe_set_rar(hw, slot, ixgbe->unicst_addr[slot].mac.addr,
6326 	    rx_group->index, IXGBE_RAH_AV);
6327 	ixgbe->unicst_addr[slot].mac.set = 1;
6328 	ixgbe->unicst_addr[slot].mac.group_index = rx_group->index;
6329 	ixgbe->unicst_avail--;
6330 
6331 	mutex_exit(&ixgbe->gen_lock);
6332 
6333 	return (0);
6334 }
6335 
6336 /*
6337  * Remove a mac address.
6338  */
6339 static int
6340 ixgbe_remmac(void *arg, const uint8_t *mac_addr)
6341 {
6342 	ixgbe_rx_group_t *rx_group = (ixgbe_rx_group_t *)arg;
6343 	ixgbe_t *ixgbe = rx_group->ixgbe;
6344 	struct ixgbe_hw *hw = &ixgbe->hw;
6345 	int slot;
6346 
6347 	mutex_enter(&ixgbe->gen_lock);
6348 
6349 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
6350 		mutex_exit(&ixgbe->gen_lock);
6351 		return (ECANCELED);
6352 	}
6353 
6354 	slot = ixgbe_unicst_find(ixgbe, mac_addr);
6355 	if (slot == -1) {
6356 		mutex_exit(&ixgbe->gen_lock);
6357 		return (EINVAL);
6358 	}
6359 
6360 	if (ixgbe->unicst_addr[slot].mac.set == 0) {
6361 		mutex_exit(&ixgbe->gen_lock);
6362 		return (EINVAL);
6363 	}
6364 
6365 	bzero(ixgbe->unicst_addr[slot].mac.addr, ETHERADDRL);
6366 	(void) ixgbe_clear_rar(hw, slot);
6367 	ixgbe->unicst_addr[slot].mac.set = 0;
6368 	ixgbe->unicst_avail++;
6369 
6370 	mutex_exit(&ixgbe->gen_lock);
6371 
6372 	return (0);
6373 }
6374