xref: /illumos-gate/usr/src/uts/common/io/ixgbe/ixgbe_main.c (revision 379728489ed47862c4927c75771e767b9476c9c4)
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 2020 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  * Copyright 2020 Oxide Computer Company
34  */
35 
36 #include "ixgbe_sw.h"
37 
38 static char ixgbe_ident[] = "Intel 10Gb Ethernet";
39 
40 /*
41  * Local function protoypes
42  */
43 static int ixgbe_register_mac(ixgbe_t *);
44 static int ixgbe_identify_hardware(ixgbe_t *);
45 static int ixgbe_regs_map(ixgbe_t *);
46 static void ixgbe_init_properties(ixgbe_t *);
47 static int ixgbe_init_driver_settings(ixgbe_t *);
48 static void ixgbe_init_locks(ixgbe_t *);
49 static void ixgbe_destroy_locks(ixgbe_t *);
50 static int ixgbe_init(ixgbe_t *);
51 static int ixgbe_chip_start(ixgbe_t *);
52 static void ixgbe_chip_stop(ixgbe_t *);
53 static int ixgbe_reset(ixgbe_t *);
54 static void ixgbe_tx_clean(ixgbe_t *);
55 static boolean_t ixgbe_tx_drain(ixgbe_t *);
56 static boolean_t ixgbe_rx_drain(ixgbe_t *);
57 static int ixgbe_alloc_rings(ixgbe_t *);
58 static void ixgbe_free_rings(ixgbe_t *);
59 static int ixgbe_alloc_rx_data(ixgbe_t *);
60 static void ixgbe_free_rx_data(ixgbe_t *);
61 static int ixgbe_setup_rings(ixgbe_t *);
62 static int ixgbe_setup_rx(ixgbe_t *);
63 static void ixgbe_setup_tx(ixgbe_t *);
64 static void ixgbe_setup_rx_ring(ixgbe_rx_ring_t *);
65 static void ixgbe_setup_tx_ring(ixgbe_tx_ring_t *);
66 static void ixgbe_setup_rss(ixgbe_t *);
67 static void ixgbe_setup_vmdq(ixgbe_t *);
68 static void ixgbe_setup_vmdq_rss(ixgbe_t *);
69 static void ixgbe_setup_rss_table(ixgbe_t *);
70 static void ixgbe_init_unicst(ixgbe_t *);
71 static int ixgbe_init_vlan(ixgbe_t *);
72 static int ixgbe_unicst_find(ixgbe_t *, const uint8_t *);
73 static void ixgbe_setup_multicst(ixgbe_t *);
74 static void ixgbe_get_hw_state(ixgbe_t *);
75 static void ixgbe_setup_vmdq_rss_conf(ixgbe_t *ixgbe);
76 static void ixgbe_get_conf(ixgbe_t *);
77 static void ixgbe_init_params(ixgbe_t *);
78 static int ixgbe_get_prop(ixgbe_t *, char *, int, int, int);
79 static void ixgbe_driver_link_check(ixgbe_t *);
80 static void ixgbe_sfp_check(void *);
81 static void ixgbe_overtemp_check(void *);
82 static void ixgbe_phy_check(void *);
83 static void ixgbe_link_timer(void *);
84 static void ixgbe_local_timer(void *);
85 static void ixgbe_arm_watchdog_timer(ixgbe_t *);
86 static void ixgbe_restart_watchdog_timer(ixgbe_t *);
87 static void ixgbe_disable_adapter_interrupts(ixgbe_t *);
88 static void ixgbe_enable_adapter_interrupts(ixgbe_t *);
89 static boolean_t is_valid_mac_addr(uint8_t *);
90 static boolean_t ixgbe_stall_check(ixgbe_t *);
91 static boolean_t ixgbe_set_loopback_mode(ixgbe_t *, uint32_t);
92 static void ixgbe_set_internal_mac_loopback(ixgbe_t *);
93 static boolean_t ixgbe_find_mac_address(ixgbe_t *);
94 static int ixgbe_alloc_intrs(ixgbe_t *);
95 static int ixgbe_alloc_intr_handles(ixgbe_t *, int);
96 static int ixgbe_add_intr_handlers(ixgbe_t *);
97 static void ixgbe_map_rxring_to_vector(ixgbe_t *, int, int);
98 static void ixgbe_map_txring_to_vector(ixgbe_t *, int, int);
99 static void ixgbe_setup_ivar(ixgbe_t *, uint16_t, uint8_t, int8_t);
100 static void ixgbe_enable_ivar(ixgbe_t *, uint16_t, int8_t);
101 static void ixgbe_disable_ivar(ixgbe_t *, uint16_t, int8_t);
102 static uint32_t ixgbe_get_hw_rx_index(ixgbe_t *ixgbe, uint32_t sw_rx_index);
103 static int ixgbe_map_intrs_to_vectors(ixgbe_t *);
104 static void ixgbe_setup_adapter_vector(ixgbe_t *);
105 static void ixgbe_rem_intr_handlers(ixgbe_t *);
106 static void ixgbe_rem_intrs(ixgbe_t *);
107 static int ixgbe_enable_intrs(ixgbe_t *);
108 static int ixgbe_disable_intrs(ixgbe_t *);
109 static uint_t ixgbe_intr_legacy(void *, void *);
110 static uint_t ixgbe_intr_msi(void *, void *);
111 static uint_t ixgbe_intr_msix(void *, void *);
112 static void ixgbe_intr_rx_work(ixgbe_rx_ring_t *);
113 static void ixgbe_intr_tx_work(ixgbe_tx_ring_t *);
114 static void ixgbe_intr_other_work(ixgbe_t *, uint32_t);
115 static void ixgbe_get_driver_control(struct ixgbe_hw *);
116 static int ixgbe_addmac(void *, const uint8_t *);
117 static int ixgbe_remmac(void *, const uint8_t *);
118 static int ixgbe_addvlan(mac_group_driver_t, uint16_t);
119 static int ixgbe_remvlan(mac_group_driver_t, uint16_t);
120 static void ixgbe_release_driver_control(struct ixgbe_hw *);
121 
122 static int ixgbe_attach(dev_info_t *, ddi_attach_cmd_t);
123 static int ixgbe_detach(dev_info_t *, ddi_detach_cmd_t);
124 static int ixgbe_resume(dev_info_t *);
125 static int ixgbe_suspend(dev_info_t *);
126 static int ixgbe_quiesce(dev_info_t *);
127 static void ixgbe_unconfigure(dev_info_t *, ixgbe_t *);
128 static uint8_t *ixgbe_mc_table_itr(struct ixgbe_hw *, uint8_t **, uint32_t *);
129 static int ixgbe_cbfunc(dev_info_t *, ddi_cb_action_t, void *, void *, void *);
130 static int ixgbe_intr_cb_register(ixgbe_t *);
131 static int ixgbe_intr_adjust(ixgbe_t *, ddi_cb_action_t, int);
132 
133 static int ixgbe_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err,
134     const void *impl_data);
135 static void ixgbe_fm_init(ixgbe_t *);
136 static void ixgbe_fm_fini(ixgbe_t *);
137 static int ixgbe_ufm_fill_image(ddi_ufm_handle_t *, void *arg, uint_t,
138     ddi_ufm_image_t *);
139 static int ixgbe_ufm_fill_slot(ddi_ufm_handle_t *, void *, uint_t, uint_t,
140     ddi_ufm_slot_t *);
141 static int ixgbe_ufm_getcaps(ddi_ufm_handle_t *, void *, ddi_ufm_cap_t *);
142 static int ixgbe_ufm_readimg(ddi_ufm_handle_t *, void *, uint_t, uint_t,
143     uint64_t, uint64_t, void *, uint64_t *);
144 
145 char *ixgbe_priv_props[] = {
146 	"_tx_copy_thresh",
147 	"_tx_recycle_thresh",
148 	"_tx_overload_thresh",
149 	"_tx_resched_thresh",
150 	"_rx_copy_thresh",
151 	"_rx_limit_per_intr",
152 	"_intr_throttling",
153 	"_adv_pause_cap",
154 	"_adv_asym_pause_cap",
155 	NULL
156 };
157 
158 #define	IXGBE_MAX_PRIV_PROPS \
159 	(sizeof (ixgbe_priv_props) / sizeof (mac_priv_prop_t))
160 
161 static struct cb_ops ixgbe_cb_ops = {
162 	nulldev,		/* cb_open */
163 	nulldev,		/* cb_close */
164 	nodev,			/* cb_strategy */
165 	nodev,			/* cb_print */
166 	nodev,			/* cb_dump */
167 	nodev,			/* cb_read */
168 	nodev,			/* cb_write */
169 	nodev,			/* cb_ioctl */
170 	nodev,			/* cb_devmap */
171 	nodev,			/* cb_mmap */
172 	nodev,			/* cb_segmap */
173 	nochpoll,		/* cb_chpoll */
174 	ddi_prop_op,		/* cb_prop_op */
175 	NULL,			/* cb_stream */
176 	D_MP | D_HOTPLUG,	/* cb_flag */
177 	CB_REV,			/* cb_rev */
178 	nodev,			/* cb_aread */
179 	nodev			/* cb_awrite */
180 };
181 
182 static struct dev_ops ixgbe_dev_ops = {
183 	DEVO_REV,		/* devo_rev */
184 	0,			/* devo_refcnt */
185 	NULL,			/* devo_getinfo */
186 	nulldev,		/* devo_identify */
187 	nulldev,		/* devo_probe */
188 	ixgbe_attach,		/* devo_attach */
189 	ixgbe_detach,		/* devo_detach */
190 	nodev,			/* devo_reset */
191 	&ixgbe_cb_ops,		/* devo_cb_ops */
192 	NULL,			/* devo_bus_ops */
193 	ddi_power,		/* devo_power */
194 	ixgbe_quiesce,		/* devo_quiesce */
195 };
196 
197 static struct modldrv ixgbe_modldrv = {
198 	&mod_driverops,		/* Type of module.  This one is a driver */
199 	ixgbe_ident,		/* Discription string */
200 	&ixgbe_dev_ops		/* driver ops */
201 };
202 
203 static struct modlinkage ixgbe_modlinkage = {
204 	MODREV_1, &ixgbe_modldrv, NULL
205 };
206 
207 /*
208  * Access attributes for register mapping
209  */
210 ddi_device_acc_attr_t ixgbe_regs_acc_attr = {
211 	DDI_DEVICE_ATTR_V1,
212 	DDI_STRUCTURE_LE_ACC,
213 	DDI_STRICTORDER_ACC,
214 	DDI_FLAGERR_ACC
215 };
216 
217 /*
218  * Loopback property
219  */
220 static lb_property_t lb_normal = {
221 	normal,	"normal", IXGBE_LB_NONE
222 };
223 
224 static lb_property_t lb_mac = {
225 	internal, "MAC", IXGBE_LB_INTERNAL_MAC
226 };
227 
228 static lb_property_t lb_external = {
229 	external, "External", IXGBE_LB_EXTERNAL
230 };
231 
232 #define	IXGBE_M_CALLBACK_FLAGS \
233 	(MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP | MC_PROPINFO)
234 
235 static mac_callbacks_t ixgbe_m_callbacks = {
236 	IXGBE_M_CALLBACK_FLAGS,
237 	ixgbe_m_stat,
238 	ixgbe_m_start,
239 	ixgbe_m_stop,
240 	ixgbe_m_promisc,
241 	ixgbe_m_multicst,
242 	NULL,
243 	NULL,
244 	NULL,
245 	ixgbe_m_ioctl,
246 	ixgbe_m_getcapab,
247 	NULL,
248 	NULL,
249 	ixgbe_m_setprop,
250 	ixgbe_m_getprop,
251 	ixgbe_m_propinfo
252 };
253 
254 /*
255  * Initialize capabilities of each supported adapter type
256  */
257 static adapter_info_t ixgbe_82598eb_cap = {
258 	64,		/* maximum number of rx queues */
259 	1,		/* minimum number of rx queues */
260 	64,		/* default number of rx queues */
261 	16,		/* maximum number of rx groups */
262 	1,		/* minimum number of rx groups */
263 	1,		/* default number of rx groups */
264 	32,		/* maximum number of tx queues */
265 	1,		/* minimum number of tx queues */
266 	8,		/* default number of tx queues */
267 	16366,		/* maximum MTU size */
268 	0xFFFF,		/* maximum interrupt throttle rate */
269 	0,		/* minimum interrupt throttle rate */
270 	200,		/* default interrupt throttle rate */
271 	18,		/* maximum total msix vectors */
272 	16,		/* maximum number of ring vectors */
273 	2,		/* maximum number of other vectors */
274 	IXGBE_EICR_LSC,	/* "other" interrupt types handled */
275 	0,		/* "other" interrupt types enable mask */
276 	(IXGBE_FLAG_DCA_CAPABLE	/* capability flags */
277 	| IXGBE_FLAG_RSS_CAPABLE
278 	| IXGBE_FLAG_VMDQ_CAPABLE)
279 };
280 
281 static adapter_info_t ixgbe_82599eb_cap = {
282 	128,		/* maximum number of rx queues */
283 	1,		/* minimum number of rx queues */
284 	128,		/* default number of rx queues */
285 	64,		/* maximum number of rx groups */
286 	1,		/* minimum number of rx groups */
287 	1,		/* default number of rx groups */
288 	128,		/* maximum number of tx queues */
289 	1,		/* minimum number of tx queues */
290 	8,		/* default number of tx queues */
291 	15500,		/* maximum MTU size */
292 	0xFF8,		/* maximum interrupt throttle rate */
293 	0,		/* minimum interrupt throttle rate */
294 	200,		/* default interrupt throttle rate */
295 	64,		/* maximum total msix vectors */
296 	16,		/* maximum number of ring vectors */
297 	2,		/* maximum number of other vectors */
298 	(IXGBE_EICR_LSC
299 	| IXGBE_EICR_GPI_SDP1
300 	| IXGBE_EICR_GPI_SDP2), /* "other" interrupt types handled */
301 
302 	(IXGBE_SDP1_GPIEN
303 	| IXGBE_SDP2_GPIEN), /* "other" interrupt types enable mask */
304 
305 	(IXGBE_FLAG_DCA_CAPABLE
306 	| IXGBE_FLAG_RSS_CAPABLE
307 	| IXGBE_FLAG_VMDQ_CAPABLE
308 	| IXGBE_FLAG_RSC_CAPABLE
309 	| IXGBE_FLAG_SFP_PLUG_CAPABLE) /* capability flags */
310 };
311 
312 static adapter_info_t ixgbe_X540_cap = {
313 	128,		/* maximum number of rx queues */
314 	1,		/* minimum number of rx queues */
315 	128,		/* default number of rx queues */
316 	64,		/* maximum number of rx groups */
317 	1,		/* minimum number of rx groups */
318 	1,		/* default number of rx groups */
319 	128,		/* maximum number of tx queues */
320 	1,		/* minimum number of tx queues */
321 	8,		/* default number of tx queues */
322 	15500,		/* maximum MTU size */
323 	0xFF8,		/* maximum interrupt throttle rate */
324 	0,		/* minimum interrupt throttle rate */
325 	200,		/* default interrupt throttle rate */
326 	64,		/* maximum total msix vectors */
327 	16,		/* maximum number of ring vectors */
328 	2,		/* maximum number of other vectors */
329 	(IXGBE_EICR_LSC
330 	| IXGBE_EICR_GPI_SDP1_X540
331 	| IXGBE_EICR_GPI_SDP2_X540), /* "other" interrupt types handled */
332 
333 	(IXGBE_SDP1_GPIEN_X540
334 	| IXGBE_SDP2_GPIEN_X540), /* "other" interrupt types enable mask */
335 
336 	(IXGBE_FLAG_DCA_CAPABLE
337 	| IXGBE_FLAG_RSS_CAPABLE
338 	| IXGBE_FLAG_VMDQ_CAPABLE
339 	| IXGBE_FLAG_RSC_CAPABLE) /* capability flags */
340 };
341 
342 static adapter_info_t ixgbe_X550_cap = {
343 	128,		/* maximum number of rx queues */
344 	1,		/* minimum number of rx queues */
345 	128,		/* default number of rx queues */
346 	64,		/* maximum number of rx groups */
347 	1,		/* minimum number of rx groups */
348 	1,		/* default number of rx groups */
349 	128,		/* maximum number of tx queues */
350 	1,		/* minimum number of tx queues */
351 	8,		/* default number of tx queues */
352 	15500,		/* maximum MTU size */
353 	0xFF8,		/* maximum interrupt throttle rate */
354 	0,		/* minimum interrupt throttle rate */
355 	0x200,		/* default interrupt throttle rate */
356 	64,		/* maximum total msix vectors */
357 	16,		/* maximum number of ring vectors */
358 	2,		/* maximum number of other vectors */
359 	IXGBE_EICR_LSC,	/* "other" interrupt types handled */
360 	0,		/* "other" interrupt types enable mask */
361 	(IXGBE_FLAG_RSS_CAPABLE
362 	| IXGBE_FLAG_VMDQ_CAPABLE
363 	| IXGBE_FLAG_RSC_CAPABLE) /* capability flags */
364 };
365 
366 static ddi_ufm_ops_t ixgbe_ufm_ops = {
367 	.ddi_ufm_op_fill_image = ixgbe_ufm_fill_image,
368 	.ddi_ufm_op_fill_slot = ixgbe_ufm_fill_slot,
369 	.ddi_ufm_op_getcaps = ixgbe_ufm_getcaps,
370 	.ddi_ufm_op_readimg = ixgbe_ufm_readimg
371 };
372 
373 
374 /*
375  * Module Initialization Functions.
376  */
377 
378 int
379 _init(void)
380 {
381 	int status;
382 
383 	mac_init_ops(&ixgbe_dev_ops, MODULE_NAME);
384 
385 	status = mod_install(&ixgbe_modlinkage);
386 
387 	if (status != DDI_SUCCESS) {
388 		mac_fini_ops(&ixgbe_dev_ops);
389 	}
390 
391 	return (status);
392 }
393 
394 int
395 _fini(void)
396 {
397 	int status;
398 
399 	status = mod_remove(&ixgbe_modlinkage);
400 
401 	if (status == DDI_SUCCESS) {
402 		mac_fini_ops(&ixgbe_dev_ops);
403 	}
404 
405 	return (status);
406 }
407 
408 int
409 _info(struct modinfo *modinfop)
410 {
411 	int status;
412 
413 	status = mod_info(&ixgbe_modlinkage, modinfop);
414 
415 	return (status);
416 }
417 
418 /*
419  * ixgbe_attach - Driver attach.
420  *
421  * This function is the device specific initialization entry
422  * point. This entry point is required and must be written.
423  * The DDI_ATTACH command must be provided in the attach entry
424  * point. When attach() is called with cmd set to DDI_ATTACH,
425  * all normal kernel services (such as kmem_alloc(9F)) are
426  * available for use by the driver.
427  *
428  * The attach() function will be called once for each instance
429  * of  the  device  on  the  system with cmd set to DDI_ATTACH.
430  * Until attach() succeeds, the only driver entry points which
431  * may be called are open(9E) and getinfo(9E).
432  */
433 static int
434 ixgbe_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
435 {
436 	ixgbe_t *ixgbe;
437 	struct ixgbe_osdep *osdep;
438 	struct ixgbe_hw *hw;
439 	int instance;
440 	char taskqname[32];
441 
442 	/*
443 	 * Check the command and perform corresponding operations
444 	 */
445 	switch (cmd) {
446 	default:
447 		return (DDI_FAILURE);
448 
449 	case DDI_RESUME:
450 		return (ixgbe_resume(devinfo));
451 
452 	case DDI_ATTACH:
453 		break;
454 	}
455 
456 	/* Get the device instance */
457 	instance = ddi_get_instance(devinfo);
458 
459 	/* Allocate memory for the instance data structure */
460 	ixgbe = kmem_zalloc(sizeof (ixgbe_t), KM_SLEEP);
461 
462 	ixgbe->dip = devinfo;
463 	ixgbe->instance = instance;
464 
465 	hw = &ixgbe->hw;
466 	osdep = &ixgbe->osdep;
467 	hw->back = osdep;
468 	osdep->ixgbe = ixgbe;
469 
470 	/* Attach the instance pointer to the dev_info data structure */
471 	ddi_set_driver_private(devinfo, ixgbe);
472 
473 	/*
474 	 * Initialize for FMA support
475 	 */
476 	ixgbe->fm_capabilities = ixgbe_get_prop(ixgbe, PROP_FM_CAPABLE,
477 	    0, 0x0f, DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
478 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
479 	ixgbe_fm_init(ixgbe);
480 	ixgbe->attach_progress |= ATTACH_PROGRESS_FM_INIT;
481 
482 	/*
483 	 * Map PCI config space registers
484 	 */
485 	if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
486 		ixgbe_error(ixgbe, "Failed to map PCI configurations");
487 		goto attach_fail;
488 	}
489 	ixgbe->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
490 
491 	/*
492 	 * Identify the chipset family
493 	 */
494 	if (ixgbe_identify_hardware(ixgbe) != IXGBE_SUCCESS) {
495 		ixgbe_error(ixgbe, "Failed to identify hardware");
496 		goto attach_fail;
497 	}
498 
499 	/*
500 	 * Map device registers
501 	 */
502 	if (ixgbe_regs_map(ixgbe) != IXGBE_SUCCESS) {
503 		ixgbe_error(ixgbe, "Failed to map device registers");
504 		goto attach_fail;
505 	}
506 	ixgbe->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
507 
508 	/*
509 	 * Initialize driver parameters
510 	 */
511 	ixgbe_init_properties(ixgbe);
512 	ixgbe->attach_progress |= ATTACH_PROGRESS_PROPS;
513 
514 	/*
515 	 * Register interrupt callback
516 	 */
517 	if (ixgbe_intr_cb_register(ixgbe) != IXGBE_SUCCESS) {
518 		ixgbe_error(ixgbe, "Failed to register interrupt callback");
519 		goto attach_fail;
520 	}
521 
522 	/*
523 	 * Allocate interrupts
524 	 */
525 	if (ixgbe_alloc_intrs(ixgbe) != IXGBE_SUCCESS) {
526 		ixgbe_error(ixgbe, "Failed to allocate interrupts");
527 		goto attach_fail;
528 	}
529 	ixgbe->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
530 
531 	/*
532 	 * Allocate rx/tx rings based on the ring numbers.
533 	 * The actual numbers of rx/tx rings are decided by the number of
534 	 * allocated interrupt vectors, so we should allocate the rings after
535 	 * interrupts are allocated.
536 	 */
537 	if (ixgbe_alloc_rings(ixgbe) != IXGBE_SUCCESS) {
538 		ixgbe_error(ixgbe, "Failed to allocate rx and tx rings");
539 		goto attach_fail;
540 	}
541 	ixgbe->attach_progress |= ATTACH_PROGRESS_ALLOC_RINGS;
542 
543 	/*
544 	 * Map rings to interrupt vectors
545 	 */
546 	if (ixgbe_map_intrs_to_vectors(ixgbe) != IXGBE_SUCCESS) {
547 		ixgbe_error(ixgbe, "Failed to map interrupts to vectors");
548 		goto attach_fail;
549 	}
550 
551 	/*
552 	 * Add interrupt handlers
553 	 */
554 	if (ixgbe_add_intr_handlers(ixgbe) != IXGBE_SUCCESS) {
555 		ixgbe_error(ixgbe, "Failed to add interrupt handlers");
556 		goto attach_fail;
557 	}
558 	ixgbe->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
559 
560 	/*
561 	 * Create a taskq for sfp-change
562 	 */
563 	(void) sprintf(taskqname, "ixgbe%d_sfp_taskq", instance);
564 	if ((ixgbe->sfp_taskq = ddi_taskq_create(devinfo, taskqname,
565 	    1, TASKQ_DEFAULTPRI, 0)) == NULL) {
566 		ixgbe_error(ixgbe, "sfp_taskq create failed");
567 		goto attach_fail;
568 	}
569 	ixgbe->attach_progress |= ATTACH_PROGRESS_SFP_TASKQ;
570 
571 	/*
572 	 * Create a taskq for over-temp
573 	 */
574 	(void) sprintf(taskqname, "ixgbe%d_overtemp_taskq", instance);
575 	if ((ixgbe->overtemp_taskq = ddi_taskq_create(devinfo, taskqname,
576 	    1, TASKQ_DEFAULTPRI, 0)) == NULL) {
577 		ixgbe_error(ixgbe, "overtemp_taskq create failed");
578 		goto attach_fail;
579 	}
580 	ixgbe->attach_progress |= ATTACH_PROGRESS_OVERTEMP_TASKQ;
581 
582 	/*
583 	 * Create a taskq for processing external PHY interrupts
584 	 */
585 	(void) sprintf(taskqname, "ixgbe%d_phy_taskq", instance);
586 	if ((ixgbe->phy_taskq = ddi_taskq_create(devinfo, taskqname,
587 	    1, TASKQ_DEFAULTPRI, 0)) == NULL) {
588 		ixgbe_error(ixgbe, "phy_taskq create failed");
589 		goto attach_fail;
590 	}
591 	ixgbe->attach_progress |= ATTACH_PROGRESS_PHY_TASKQ;
592 
593 	/*
594 	 * Initialize driver parameters
595 	 */
596 	if (ixgbe_init_driver_settings(ixgbe) != IXGBE_SUCCESS) {
597 		ixgbe_error(ixgbe, "Failed to initialize driver settings");
598 		goto attach_fail;
599 	}
600 
601 	/*
602 	 * Initialize mutexes for this device.
603 	 * Do this before enabling the interrupt handler and
604 	 * register the softint to avoid the condition where
605 	 * interrupt handler can try using uninitialized mutex.
606 	 */
607 	ixgbe_init_locks(ixgbe);
608 	ixgbe->attach_progress |= ATTACH_PROGRESS_LOCKS;
609 
610 	/*
611 	 * Initialize chipset hardware
612 	 */
613 	if (ixgbe_init(ixgbe) != IXGBE_SUCCESS) {
614 		ixgbe_error(ixgbe, "Failed to initialize adapter");
615 		goto attach_fail;
616 	}
617 	ixgbe->link_check_complete = B_FALSE;
618 	ixgbe->link_check_hrtime = gethrtime() +
619 	    (IXGBE_LINK_UP_TIME * 100000000ULL);
620 	ixgbe->attach_progress |= ATTACH_PROGRESS_INIT;
621 
622 	if (ixgbe_check_acc_handle(ixgbe->osdep.cfg_handle) != DDI_FM_OK) {
623 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
624 		goto attach_fail;
625 	}
626 
627 	/*
628 	 * Initialize adapter capabilities
629 	 */
630 	ixgbe_init_params(ixgbe);
631 
632 	/*
633 	 * Initialize statistics
634 	 */
635 	if (ixgbe_init_stats(ixgbe) != IXGBE_SUCCESS) {
636 		ixgbe_error(ixgbe, "Failed to initialize statistics");
637 		goto attach_fail;
638 	}
639 	ixgbe->attach_progress |= ATTACH_PROGRESS_STATS;
640 
641 	/*
642 	 * Register the driver to the MAC
643 	 */
644 	if (ixgbe_register_mac(ixgbe) != IXGBE_SUCCESS) {
645 		ixgbe_error(ixgbe, "Failed to register MAC");
646 		goto attach_fail;
647 	}
648 	mac_link_update(ixgbe->mac_hdl, LINK_STATE_UNKNOWN);
649 	ixgbe->attach_progress |= ATTACH_PROGRESS_MAC;
650 
651 	ixgbe->periodic_id = ddi_periodic_add(ixgbe_link_timer, ixgbe,
652 	    IXGBE_CYCLIC_PERIOD, DDI_IPL_0);
653 	if (ixgbe->periodic_id == 0) {
654 		ixgbe_error(ixgbe, "Failed to add the link check timer");
655 		goto attach_fail;
656 	}
657 	ixgbe->attach_progress |= ATTACH_PROGRESS_LINK_TIMER;
658 
659 	/*
660 	 * Now that mutex locks are initialized, and the chip is also
661 	 * initialized, enable interrupts.
662 	 */
663 	if (ixgbe_enable_intrs(ixgbe) != IXGBE_SUCCESS) {
664 		ixgbe_error(ixgbe, "Failed to enable DDI interrupts");
665 		goto attach_fail;
666 	}
667 	ixgbe->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
668 
669 	if (ixgbe->hw.bus.func == 0) {
670 		if (ddi_ufm_init(devinfo, DDI_UFM_CURRENT_VERSION,
671 		    &ixgbe_ufm_ops, &ixgbe->ixgbe_ufmh, ixgbe) != 0) {
672 			ixgbe_error(ixgbe, "Failed to enable DDI UFM support");
673 			goto attach_fail;
674 		}
675 		ixgbe->attach_progress |= ATTACH_PROGRESS_UFM;
676 		ddi_ufm_update(ixgbe->ixgbe_ufmh);
677 	}
678 
679 	ixgbe_log(ixgbe, "%s", ixgbe_ident);
680 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_INITIALIZED);
681 
682 	return (DDI_SUCCESS);
683 
684 attach_fail:
685 	ixgbe_unconfigure(devinfo, ixgbe);
686 	return (DDI_FAILURE);
687 }
688 
689 /*
690  * ixgbe_detach - Driver detach.
691  *
692  * The detach() function is the complement of the attach routine.
693  * If cmd is set to DDI_DETACH, detach() is used to remove  the
694  * state  associated  with  a  given  instance of a device node
695  * prior to the removal of that instance from the system.
696  *
697  * The detach() function will be called once for each  instance
698  * of the device for which there has been a successful attach()
699  * once there are no longer  any  opens  on  the  device.
700  *
701  * Interrupts routine are disabled, All memory allocated by this
702  * driver are freed.
703  */
704 static int
705 ixgbe_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
706 {
707 	ixgbe_t *ixgbe;
708 
709 	/*
710 	 * Check detach command
711 	 */
712 	switch (cmd) {
713 	default:
714 		return (DDI_FAILURE);
715 
716 	case DDI_SUSPEND:
717 		return (ixgbe_suspend(devinfo));
718 
719 	case DDI_DETACH:
720 		break;
721 	}
722 
723 	/*
724 	 * Get the pointer to the driver private data structure
725 	 */
726 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
727 	if (ixgbe == NULL)
728 		return (DDI_FAILURE);
729 
730 	/*
731 	 * If the device is still running, it needs to be stopped first.
732 	 * This check is necessary because under some specific circumstances,
733 	 * the detach routine can be called without stopping the interface
734 	 * first.
735 	 */
736 	if (ixgbe->ixgbe_state & IXGBE_STARTED) {
737 		atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_STARTED);
738 		mutex_enter(&ixgbe->gen_lock);
739 		ixgbe_stop(ixgbe, B_TRUE);
740 		mutex_exit(&ixgbe->gen_lock);
741 		/* Disable and stop the watchdog timer */
742 		ixgbe_disable_watchdog_timer(ixgbe);
743 	}
744 
745 	/*
746 	 * Check if there are still rx buffers held by the upper layer.
747 	 * If so, fail the detach.
748 	 */
749 	if (!ixgbe_rx_drain(ixgbe))
750 		return (DDI_FAILURE);
751 
752 	/*
753 	 * Do the remaining unconfigure routines
754 	 */
755 	ixgbe_unconfigure(devinfo, ixgbe);
756 
757 	return (DDI_SUCCESS);
758 }
759 
760 /*
761  * quiesce(9E) entry point.
762  *
763  * This function is called when the system is single-threaded at high
764  * PIL with preemption disabled. Therefore, this function must not be
765  * blocked.
766  *
767  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
768  * DDI_FAILURE indicates an error condition and should almost never happen.
769  */
770 static int
771 ixgbe_quiesce(dev_info_t *devinfo)
772 {
773 	ixgbe_t *ixgbe;
774 	struct ixgbe_hw *hw;
775 
776 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
777 
778 	if (ixgbe == NULL)
779 		return (DDI_FAILURE);
780 
781 	hw = &ixgbe->hw;
782 
783 	/*
784 	 * Disable the adapter interrupts
785 	 */
786 	ixgbe_disable_adapter_interrupts(ixgbe);
787 
788 	/*
789 	 * Tell firmware driver is no longer in control
790 	 */
791 	ixgbe_release_driver_control(hw);
792 
793 	/*
794 	 * Reset the chipset
795 	 */
796 	(void) ixgbe_reset_hw(hw);
797 
798 	/*
799 	 * Reset PHY
800 	 */
801 	(void) ixgbe_reset_phy(hw);
802 
803 	return (DDI_SUCCESS);
804 }
805 
806 static void
807 ixgbe_unconfigure(dev_info_t *devinfo, ixgbe_t *ixgbe)
808 {
809 	/*
810 	 * Disable interrupt
811 	 */
812 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
813 		(void) ixgbe_disable_intrs(ixgbe);
814 	}
815 
816 	/*
817 	 * remove the link check timer
818 	 */
819 	if (ixgbe->attach_progress & ATTACH_PROGRESS_LINK_TIMER) {
820 		if (ixgbe->periodic_id != NULL) {
821 			ddi_periodic_delete(ixgbe->periodic_id);
822 			ixgbe->periodic_id = NULL;
823 		}
824 	}
825 
826 	/*
827 	 * Clean up the UFM subsystem. Note this only is set on function 0.
828 	 */
829 	if (ixgbe->attach_progress & ATTACH_PROGRESS_UFM) {
830 		ddi_ufm_fini(ixgbe->ixgbe_ufmh);
831 	}
832 
833 	/*
834 	 * Unregister MAC
835 	 */
836 	if (ixgbe->attach_progress & ATTACH_PROGRESS_MAC) {
837 		(void) mac_unregister(ixgbe->mac_hdl);
838 	}
839 
840 	/*
841 	 * Free statistics
842 	 */
843 	if (ixgbe->attach_progress & ATTACH_PROGRESS_STATS) {
844 		kstat_delete((kstat_t *)ixgbe->ixgbe_ks);
845 	}
846 
847 	/*
848 	 * Remove interrupt handlers
849 	 */
850 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
851 		ixgbe_rem_intr_handlers(ixgbe);
852 	}
853 
854 	/*
855 	 * Remove taskq for sfp-status-change
856 	 */
857 	if (ixgbe->attach_progress & ATTACH_PROGRESS_SFP_TASKQ) {
858 		ddi_taskq_destroy(ixgbe->sfp_taskq);
859 	}
860 
861 	/*
862 	 * Remove taskq for over-temp
863 	 */
864 	if (ixgbe->attach_progress & ATTACH_PROGRESS_OVERTEMP_TASKQ) {
865 		ddi_taskq_destroy(ixgbe->overtemp_taskq);
866 	}
867 
868 	/*
869 	 * Remove taskq for external PHYs
870 	 */
871 	if (ixgbe->attach_progress & ATTACH_PROGRESS_PHY_TASKQ) {
872 		ddi_taskq_destroy(ixgbe->phy_taskq);
873 	}
874 
875 	/*
876 	 * Remove interrupts
877 	 */
878 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ALLOC_INTR) {
879 		ixgbe_rem_intrs(ixgbe);
880 	}
881 
882 	/*
883 	 * Unregister interrupt callback handler
884 	 */
885 	if (ixgbe->cb_hdl != NULL) {
886 		(void) ddi_cb_unregister(ixgbe->cb_hdl);
887 	}
888 
889 	/*
890 	 * Remove driver properties
891 	 */
892 	if (ixgbe->attach_progress & ATTACH_PROGRESS_PROPS) {
893 		(void) ddi_prop_remove_all(devinfo);
894 	}
895 
896 	/*
897 	 * Stop the chipset
898 	 */
899 	if (ixgbe->attach_progress & ATTACH_PROGRESS_INIT) {
900 		mutex_enter(&ixgbe->gen_lock);
901 		ixgbe_chip_stop(ixgbe);
902 		mutex_exit(&ixgbe->gen_lock);
903 	}
904 
905 	/*
906 	 * Free register handle
907 	 */
908 	if (ixgbe->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
909 		if (ixgbe->osdep.reg_handle != NULL)
910 			ddi_regs_map_free(&ixgbe->osdep.reg_handle);
911 	}
912 
913 	/*
914 	 * Free PCI config handle
915 	 */
916 	if (ixgbe->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
917 		if (ixgbe->osdep.cfg_handle != NULL)
918 			pci_config_teardown(&ixgbe->osdep.cfg_handle);
919 	}
920 
921 	/*
922 	 * Free locks
923 	 */
924 	if (ixgbe->attach_progress & ATTACH_PROGRESS_LOCKS) {
925 		ixgbe_destroy_locks(ixgbe);
926 	}
927 
928 	/*
929 	 * Free the rx/tx rings
930 	 */
931 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
932 		ixgbe_free_rings(ixgbe);
933 	}
934 
935 	/*
936 	 * Unregister FMA capabilities
937 	 */
938 	if (ixgbe->attach_progress & ATTACH_PROGRESS_FM_INIT) {
939 		ixgbe_fm_fini(ixgbe);
940 	}
941 
942 	/*
943 	 * Free the driver data structure
944 	 */
945 	kmem_free(ixgbe, sizeof (ixgbe_t));
946 
947 	ddi_set_driver_private(devinfo, NULL);
948 }
949 
950 /*
951  * ixgbe_register_mac - Register the driver and its function pointers with
952  * the GLD interface.
953  */
954 static int
955 ixgbe_register_mac(ixgbe_t *ixgbe)
956 {
957 	struct ixgbe_hw *hw = &ixgbe->hw;
958 	mac_register_t *mac;
959 	int status;
960 
961 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
962 		return (IXGBE_FAILURE);
963 
964 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
965 	mac->m_driver = ixgbe;
966 	mac->m_dip = ixgbe->dip;
967 	mac->m_src_addr = hw->mac.addr;
968 	mac->m_callbacks = &ixgbe_m_callbacks;
969 	mac->m_min_sdu = 0;
970 	mac->m_max_sdu = ixgbe->default_mtu;
971 	mac->m_margin = VLAN_TAGSZ;
972 	mac->m_priv_props = ixgbe_priv_props;
973 	mac->m_v12n = MAC_VIRT_LEVEL1;
974 
975 	status = mac_register(mac, &ixgbe->mac_hdl);
976 
977 	mac_free(mac);
978 
979 	return ((status == 0) ? IXGBE_SUCCESS : IXGBE_FAILURE);
980 }
981 
982 /*
983  * ixgbe_identify_hardware - Identify the type of the chipset.
984  */
985 static int
986 ixgbe_identify_hardware(ixgbe_t *ixgbe)
987 {
988 	struct ixgbe_hw *hw = &ixgbe->hw;
989 	struct ixgbe_osdep *osdep = &ixgbe->osdep;
990 
991 	/*
992 	 * Get the device id
993 	 */
994 	hw->vendor_id =
995 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
996 	hw->device_id =
997 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
998 	hw->revision_id =
999 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
1000 	hw->subsystem_device_id =
1001 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
1002 	hw->subsystem_vendor_id =
1003 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
1004 
1005 	/*
1006 	 * Set the mac type of the adapter based on the device id
1007 	 */
1008 	if (ixgbe_set_mac_type(hw) != IXGBE_SUCCESS) {
1009 		return (IXGBE_FAILURE);
1010 	}
1011 
1012 	/*
1013 	 * Install adapter capabilities
1014 	 */
1015 	switch (hw->mac.type) {
1016 	case ixgbe_mac_82598EB:
1017 		IXGBE_DEBUGLOG_0(ixgbe, "identify 82598 adapter\n");
1018 		ixgbe->capab = &ixgbe_82598eb_cap;
1019 
1020 		if (ixgbe_get_media_type(hw) == ixgbe_media_type_copper) {
1021 			ixgbe->capab->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
1022 			ixgbe->capab->other_intr |= IXGBE_EICR_GPI_SDP1;
1023 			ixgbe->capab->other_gpie |= IXGBE_SDP1_GPIEN;
1024 		}
1025 		break;
1026 
1027 	case ixgbe_mac_82599EB:
1028 		IXGBE_DEBUGLOG_0(ixgbe, "identify 82599 adapter\n");
1029 		ixgbe->capab = &ixgbe_82599eb_cap;
1030 
1031 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM) {
1032 			ixgbe->capab->flags |= IXGBE_FLAG_TEMP_SENSOR_CAPABLE;
1033 			ixgbe->capab->other_intr |= IXGBE_EICR_GPI_SDP0;
1034 			ixgbe->capab->other_gpie |= IXGBE_SDP0_GPIEN;
1035 		}
1036 		break;
1037 
1038 	case ixgbe_mac_X540:
1039 		IXGBE_DEBUGLOG_0(ixgbe, "identify X540 adapter\n");
1040 		ixgbe->capab = &ixgbe_X540_cap;
1041 		/*
1042 		 * For now, X540 is all set in its capab structure.
1043 		 * As other X540 variants show up, things can change here.
1044 		 */
1045 		break;
1046 
1047 	case ixgbe_mac_X550:
1048 	case ixgbe_mac_X550EM_x:
1049 	case ixgbe_mac_X550EM_a:
1050 		IXGBE_DEBUGLOG_0(ixgbe, "identify X550 adapter\n");
1051 		ixgbe->capab = &ixgbe_X550_cap;
1052 
1053 		if (hw->device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
1054 		    hw->device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
1055 		    hw->device_id == IXGBE_DEV_ID_X550EM_A_SFP_N ||
1056 		    hw->device_id == IXGBE_DEV_ID_X550EM_A_QSFP ||
1057 		    hw->device_id == IXGBE_DEV_ID_X550EM_A_QSFP_N) {
1058 			ixgbe->capab->flags |= IXGBE_FLAG_SFP_PLUG_CAPABLE;
1059 		}
1060 
1061 		/*
1062 		 * Link detection on X552 SFP+ and X552/X557-AT
1063 		 */
1064 		if (hw->device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
1065 		    hw->device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
1066 		    hw->device_id == IXGBE_DEV_ID_X550EM_A_SFP_N ||
1067 		    hw->device_id == IXGBE_DEV_ID_X550EM_X_10G_T) {
1068 			ixgbe->capab->other_intr |=
1069 			    IXGBE_EIMS_GPI_SDP0_BY_MAC(hw);
1070 		}
1071 		if (hw->phy.type == ixgbe_phy_x550em_ext_t) {
1072 			ixgbe->capab->other_gpie |= IXGBE_SDP0_GPIEN_X540;
1073 		}
1074 		break;
1075 
1076 	default:
1077 		IXGBE_DEBUGLOG_1(ixgbe,
1078 		    "adapter not supported in ixgbe_identify_hardware(): %d\n",
1079 		    hw->mac.type);
1080 		return (IXGBE_FAILURE);
1081 	}
1082 
1083 	return (IXGBE_SUCCESS);
1084 }
1085 
1086 /*
1087  * ixgbe_regs_map - Map the device registers.
1088  *
1089  */
1090 static int
1091 ixgbe_regs_map(ixgbe_t *ixgbe)
1092 {
1093 	dev_info_t *devinfo = ixgbe->dip;
1094 	struct ixgbe_hw *hw = &ixgbe->hw;
1095 	struct ixgbe_osdep *osdep = &ixgbe->osdep;
1096 	off_t mem_size;
1097 
1098 	/*
1099 	 * First get the size of device registers to be mapped.
1100 	 */
1101 	if (ddi_dev_regsize(devinfo, IXGBE_ADAPTER_REGSET, &mem_size)
1102 	    != DDI_SUCCESS) {
1103 		return (IXGBE_FAILURE);
1104 	}
1105 
1106 	/*
1107 	 * Call ddi_regs_map_setup() to map registers
1108 	 */
1109 	if ((ddi_regs_map_setup(devinfo, IXGBE_ADAPTER_REGSET,
1110 	    (caddr_t *)&hw->hw_addr, 0,
1111 	    mem_size, &ixgbe_regs_acc_attr,
1112 	    &osdep->reg_handle)) != DDI_SUCCESS) {
1113 		return (IXGBE_FAILURE);
1114 	}
1115 
1116 	return (IXGBE_SUCCESS);
1117 }
1118 
1119 /*
1120  * ixgbe_init_properties - Initialize driver properties.
1121  */
1122 static void
1123 ixgbe_init_properties(ixgbe_t *ixgbe)
1124 {
1125 	/*
1126 	 * Get conf file properties, including link settings
1127 	 * jumbo frames, ring number, descriptor number, etc.
1128 	 */
1129 	ixgbe_get_conf(ixgbe);
1130 }
1131 
1132 /*
1133  * ixgbe_init_driver_settings - Initialize driver settings.
1134  *
1135  * The settings include hardware function pointers, bus information,
1136  * rx/tx rings settings, link state, and any other parameters that
1137  * need to be setup during driver initialization.
1138  */
1139 static int
1140 ixgbe_init_driver_settings(ixgbe_t *ixgbe)
1141 {
1142 	struct ixgbe_hw *hw = &ixgbe->hw;
1143 	dev_info_t *devinfo = ixgbe->dip;
1144 	ixgbe_rx_ring_t *rx_ring;
1145 	ixgbe_rx_group_t *rx_group;
1146 	ixgbe_tx_ring_t *tx_ring;
1147 	uint32_t rx_size;
1148 	uint32_t tx_size;
1149 	uint32_t ring_per_group;
1150 	int i;
1151 
1152 	/*
1153 	 * Initialize chipset specific hardware function pointers
1154 	 */
1155 	if (ixgbe_init_shared_code(hw) != IXGBE_SUCCESS) {
1156 		return (IXGBE_FAILURE);
1157 	}
1158 
1159 	/*
1160 	 * Get the system page size
1161 	 */
1162 	ixgbe->sys_page_size = ddi_ptob(devinfo, (ulong_t)1);
1163 
1164 	/*
1165 	 * Set rx buffer size
1166 	 *
1167 	 * The IP header alignment room is counted in the calculation.
1168 	 * The rx buffer size is in unit of 1K that is required by the
1169 	 * chipset hardware.
1170 	 */
1171 	rx_size = ixgbe->max_frame_size + IPHDR_ALIGN_ROOM;
1172 	ixgbe->rx_buf_size = ((rx_size >> 10) +
1173 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
1174 
1175 	/*
1176 	 * Set tx buffer size
1177 	 */
1178 	tx_size = ixgbe->max_frame_size;
1179 	ixgbe->tx_buf_size = ((tx_size >> 10) +
1180 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
1181 
1182 	/*
1183 	 * Initialize rx/tx rings/groups parameters
1184 	 */
1185 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
1186 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1187 		rx_ring = &ixgbe->rx_rings[i];
1188 		rx_ring->index = i;
1189 		rx_ring->ixgbe = ixgbe;
1190 		rx_ring->group_index = i / ring_per_group;
1191 		rx_ring->hw_index = ixgbe_get_hw_rx_index(ixgbe, i);
1192 	}
1193 
1194 	for (i = 0; i < ixgbe->num_rx_groups; i++) {
1195 		rx_group = &ixgbe->rx_groups[i];
1196 		rx_group->index = i;
1197 		rx_group->ixgbe = ixgbe;
1198 		list_create(&rx_group->vlans, sizeof (ixgbe_vlan_t),
1199 		    offsetof(ixgbe_vlan_t, ixvl_link));
1200 	}
1201 
1202 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1203 		tx_ring = &ixgbe->tx_rings[i];
1204 		tx_ring->index = i;
1205 		tx_ring->ixgbe = ixgbe;
1206 		if (ixgbe->tx_head_wb_enable)
1207 			tx_ring->tx_recycle = ixgbe_tx_recycle_head_wb;
1208 		else
1209 			tx_ring->tx_recycle = ixgbe_tx_recycle_legacy;
1210 
1211 		tx_ring->ring_size = ixgbe->tx_ring_size;
1212 		tx_ring->free_list_size = ixgbe->tx_ring_size +
1213 		    (ixgbe->tx_ring_size >> 1);
1214 	}
1215 
1216 	/*
1217 	 * Initialize values of interrupt throttling rate
1218 	 */
1219 	for (i = 1; i < MAX_INTR_VECTOR; i++)
1220 		ixgbe->intr_throttling[i] = ixgbe->intr_throttling[0];
1221 
1222 	/*
1223 	 * The initial link state should be "unknown"
1224 	 */
1225 	ixgbe->link_state = LINK_STATE_UNKNOWN;
1226 
1227 	return (IXGBE_SUCCESS);
1228 }
1229 
1230 /*
1231  * ixgbe_init_locks - Initialize locks.
1232  */
1233 static void
1234 ixgbe_init_locks(ixgbe_t *ixgbe)
1235 {
1236 	ixgbe_rx_ring_t *rx_ring;
1237 	ixgbe_tx_ring_t *tx_ring;
1238 	int i;
1239 
1240 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1241 		rx_ring = &ixgbe->rx_rings[i];
1242 		mutex_init(&rx_ring->rx_lock, NULL,
1243 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1244 	}
1245 
1246 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1247 		tx_ring = &ixgbe->tx_rings[i];
1248 		mutex_init(&tx_ring->tx_lock, NULL,
1249 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1250 		mutex_init(&tx_ring->recycle_lock, NULL,
1251 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1252 		mutex_init(&tx_ring->tcb_head_lock, NULL,
1253 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1254 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
1255 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1256 	}
1257 
1258 	mutex_init(&ixgbe->gen_lock, NULL,
1259 	    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1260 
1261 	mutex_init(&ixgbe->watchdog_lock, NULL,
1262 	    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1263 }
1264 
1265 /*
1266  * ixgbe_destroy_locks - Destroy locks.
1267  */
1268 static void
1269 ixgbe_destroy_locks(ixgbe_t *ixgbe)
1270 {
1271 	ixgbe_rx_ring_t *rx_ring;
1272 	ixgbe_tx_ring_t *tx_ring;
1273 	int i;
1274 
1275 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1276 		rx_ring = &ixgbe->rx_rings[i];
1277 		mutex_destroy(&rx_ring->rx_lock);
1278 	}
1279 
1280 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1281 		tx_ring = &ixgbe->tx_rings[i];
1282 		mutex_destroy(&tx_ring->tx_lock);
1283 		mutex_destroy(&tx_ring->recycle_lock);
1284 		mutex_destroy(&tx_ring->tcb_head_lock);
1285 		mutex_destroy(&tx_ring->tcb_tail_lock);
1286 	}
1287 
1288 	mutex_destroy(&ixgbe->gen_lock);
1289 	mutex_destroy(&ixgbe->watchdog_lock);
1290 }
1291 
1292 /*
1293  * We need to try and determine which LED index in hardware corresponds to the
1294  * link/activity LED. This is the one that'll be overwritten when we perform
1295  * GLDv3 LED activity.
1296  */
1297 static void
1298 ixgbe_led_init(ixgbe_t *ixgbe)
1299 {
1300 	uint32_t reg, i;
1301 	struct ixgbe_hw *hw = &ixgbe->hw;
1302 
1303 	reg = IXGBE_READ_REG(hw, IXGBE_LEDCTL);
1304 	for (i = 0; i < 4; i++) {
1305 		if (((reg >> IXGBE_LED_MODE_SHIFT(i)) &
1306 		    IXGBE_LED_MODE_MASK_BASE) == IXGBE_LED_LINK_ACTIVE) {
1307 			ixgbe->ixgbe_led_index = i;
1308 			return;
1309 		}
1310 	}
1311 
1312 	/*
1313 	 * If we couldn't determine this, we use the default for various MACs
1314 	 * based on information Intel has inserted into other drivers over the
1315 	 * years.
1316 	 */
1317 	switch (hw->mac.type) {
1318 	case ixgbe_mac_X550EM_a:
1319 		ixgbe->ixgbe_led_index = 0;
1320 		break;
1321 	case ixgbe_mac_X550EM_x:
1322 		ixgbe->ixgbe_led_index = 1;
1323 		break;
1324 	default:
1325 		ixgbe->ixgbe_led_index = 2;
1326 		break;
1327 	}
1328 }
1329 
1330 static int
1331 ixgbe_resume(dev_info_t *devinfo)
1332 {
1333 	ixgbe_t *ixgbe;
1334 	int i;
1335 
1336 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
1337 	if (ixgbe == NULL)
1338 		return (DDI_FAILURE);
1339 
1340 	mutex_enter(&ixgbe->gen_lock);
1341 
1342 	if (ixgbe->ixgbe_state & IXGBE_STARTED) {
1343 		if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
1344 			mutex_exit(&ixgbe->gen_lock);
1345 			return (DDI_FAILURE);
1346 		}
1347 
1348 		/*
1349 		 * Enable and start the watchdog timer
1350 		 */
1351 		ixgbe_enable_watchdog_timer(ixgbe);
1352 	}
1353 
1354 	atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_SUSPENDED);
1355 
1356 	if (ixgbe->ixgbe_state & IXGBE_STARTED) {
1357 		for (i = 0; i < ixgbe->num_tx_rings; i++) {
1358 			mac_tx_ring_update(ixgbe->mac_hdl,
1359 			    ixgbe->tx_rings[i].ring_handle);
1360 		}
1361 	}
1362 
1363 	mutex_exit(&ixgbe->gen_lock);
1364 
1365 	return (DDI_SUCCESS);
1366 }
1367 
1368 static int
1369 ixgbe_suspend(dev_info_t *devinfo)
1370 {
1371 	ixgbe_t *ixgbe;
1372 
1373 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
1374 	if (ixgbe == NULL)
1375 		return (DDI_FAILURE);
1376 
1377 	mutex_enter(&ixgbe->gen_lock);
1378 
1379 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_SUSPENDED);
1380 	if (!(ixgbe->ixgbe_state & IXGBE_STARTED)) {
1381 		mutex_exit(&ixgbe->gen_lock);
1382 		return (DDI_SUCCESS);
1383 	}
1384 	ixgbe_stop(ixgbe, B_FALSE);
1385 
1386 	mutex_exit(&ixgbe->gen_lock);
1387 
1388 	/*
1389 	 * Disable and stop the watchdog timer
1390 	 */
1391 	ixgbe_disable_watchdog_timer(ixgbe);
1392 
1393 	return (DDI_SUCCESS);
1394 }
1395 
1396 /*
1397  * ixgbe_init - Initialize the device.
1398  */
1399 static int
1400 ixgbe_init(ixgbe_t *ixgbe)
1401 {
1402 	struct ixgbe_hw *hw = &ixgbe->hw;
1403 	u8 pbanum[IXGBE_PBANUM_LENGTH];
1404 	int rv;
1405 
1406 	mutex_enter(&ixgbe->gen_lock);
1407 
1408 	/*
1409 	 * Configure/Initialize hardware
1410 	 */
1411 	rv = ixgbe_init_hw(hw);
1412 	if (rv != IXGBE_SUCCESS) {
1413 		switch (rv) {
1414 
1415 		/*
1416 		 * The first three errors are not prohibitive to us progressing
1417 		 * further, and are maily advisory in nature. In the case of a
1418 		 * SFP module not being present or not deemed supported by the
1419 		 * common code, we adivse the operator of this fact but carry on
1420 		 * instead of failing hard, as SFPs can be inserted or replaced
1421 		 * while the driver is running. In the case of a unknown error,
1422 		 * we fail-hard, logging the reason and emitting a FMA event.
1423 		 */
1424 		case IXGBE_ERR_EEPROM_VERSION:
1425 			ixgbe_error(ixgbe,
1426 			    "This Intel 10Gb Ethernet device is pre-release and"
1427 			    " contains outdated firmware. Please contact your"
1428 			    " hardware vendor for a replacement.");
1429 			break;
1430 		case IXGBE_ERR_SFP_NOT_PRESENT:
1431 			ixgbe_error(ixgbe,
1432 			    "No SFP+ module detected on this interface. Please "
1433 			    "install a supported SFP+ module for this "
1434 			    "interface to become operational.");
1435 			break;
1436 		case IXGBE_ERR_SFP_NOT_SUPPORTED:
1437 			ixgbe_error(ixgbe,
1438 			    "Unsupported SFP+ module detected. Please replace "
1439 			    "it with a supported SFP+ module per Intel "
1440 			    "documentation, or bypass this check with "
1441 			    "allow_unsupported_sfp=1 in ixgbe.conf.");
1442 			break;
1443 		default:
1444 			ixgbe_error(ixgbe,
1445 			    "Failed to initialize hardware. ixgbe_init_hw "
1446 			    "returned %d", rv);
1447 			ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1448 			goto init_fail;
1449 		}
1450 	}
1451 
1452 	/*
1453 	 * Need to init eeprom before validating the checksum.
1454 	 */
1455 	if (ixgbe_init_eeprom_params(hw) < 0) {
1456 		ixgbe_error(ixgbe,
1457 		    "Unable to intitialize the eeprom interface.");
1458 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1459 		goto init_fail;
1460 	}
1461 
1462 	/*
1463 	 * NVM validation
1464 	 */
1465 	if (ixgbe_validate_eeprom_checksum(hw, NULL) < 0) {
1466 		/*
1467 		 * Some PCI-E parts fail the first check due to
1468 		 * the link being in sleep state.  Call it again,
1469 		 * if it fails a second time it's a real issue.
1470 		 */
1471 		if (ixgbe_validate_eeprom_checksum(hw, NULL) < 0) {
1472 			ixgbe_error(ixgbe,
1473 			    "Invalid NVM checksum. Please contact "
1474 			    "the vendor to update the NVM.");
1475 			ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1476 			goto init_fail;
1477 		}
1478 	}
1479 
1480 	/*
1481 	 * Setup default flow control thresholds - enable/disable
1482 	 * & flow control type is controlled by ixgbe.conf
1483 	 */
1484 	hw->fc.high_water[0] = DEFAULT_FCRTH;
1485 	hw->fc.low_water[0] = DEFAULT_FCRTL;
1486 	hw->fc.pause_time = DEFAULT_FCPAUSE;
1487 	hw->fc.send_xon = B_TRUE;
1488 
1489 	/*
1490 	 * Initialize flow control
1491 	 */
1492 	(void) ixgbe_start_hw(hw);
1493 
1494 	/*
1495 	 * Initialize link settings
1496 	 */
1497 	(void) ixgbe_driver_setup_link(ixgbe, B_FALSE);
1498 
1499 	/*
1500 	 * Initialize the chipset hardware
1501 	 */
1502 	if (ixgbe_chip_start(ixgbe) != IXGBE_SUCCESS) {
1503 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1504 		goto init_fail;
1505 	}
1506 
1507 	/*
1508 	 * Read identifying information and place in devinfo.
1509 	 */
1510 	pbanum[0] = '\0';
1511 	(void) ixgbe_read_pba_string(hw, pbanum, sizeof (pbanum));
1512 	if (*pbanum != '\0') {
1513 		(void) ddi_prop_update_string(DDI_DEV_T_NONE, ixgbe->dip,
1514 		    "printed-board-assembly", (char *)pbanum);
1515 	}
1516 
1517 	/*
1518 	 * Determine LED index.
1519 	 */
1520 	ixgbe_led_init(ixgbe);
1521 
1522 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1523 		goto init_fail;
1524 	}
1525 
1526 	mutex_exit(&ixgbe->gen_lock);
1527 	return (IXGBE_SUCCESS);
1528 
1529 init_fail:
1530 	/*
1531 	 * Reset PHY
1532 	 */
1533 	(void) ixgbe_reset_phy(hw);
1534 
1535 	mutex_exit(&ixgbe->gen_lock);
1536 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1537 	return (IXGBE_FAILURE);
1538 }
1539 
1540 /*
1541  * ixgbe_chip_start - Initialize and start the chipset hardware.
1542  */
1543 static int
1544 ixgbe_chip_start(ixgbe_t *ixgbe)
1545 {
1546 	struct ixgbe_hw *hw = &ixgbe->hw;
1547 	int i;
1548 
1549 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1550 
1551 	/*
1552 	 * Get the mac address
1553 	 * This function should handle SPARC case correctly.
1554 	 */
1555 	if (!ixgbe_find_mac_address(ixgbe)) {
1556 		ixgbe_error(ixgbe, "Failed to get the mac address");
1557 		return (IXGBE_FAILURE);
1558 	}
1559 
1560 	/*
1561 	 * Validate the mac address
1562 	 */
1563 	(void) ixgbe_init_rx_addrs(hw);
1564 	if (!is_valid_mac_addr(hw->mac.addr)) {
1565 		ixgbe_error(ixgbe, "Invalid mac address");
1566 		return (IXGBE_FAILURE);
1567 	}
1568 
1569 	/*
1570 	 * Re-enable relaxed ordering for performance.  It is disabled
1571 	 * by default in the hardware init.
1572 	 */
1573 	if (ixgbe->relax_order_enable == B_TRUE)
1574 		ixgbe_enable_relaxed_ordering(hw);
1575 
1576 	/*
1577 	 * Setup adapter interrupt vectors
1578 	 */
1579 	ixgbe_setup_adapter_vector(ixgbe);
1580 
1581 	/*
1582 	 * Initialize unicast addresses.
1583 	 */
1584 	ixgbe_init_unicst(ixgbe);
1585 
1586 	/*
1587 	 * Setup and initialize the mctable structures.
1588 	 */
1589 	ixgbe_setup_multicst(ixgbe);
1590 
1591 	/*
1592 	 * Set interrupt throttling rate
1593 	 */
1594 	for (i = 0; i < ixgbe->intr_cnt; i++) {
1595 		IXGBE_WRITE_REG(hw, IXGBE_EITR(i), ixgbe->intr_throttling[i]);
1596 	}
1597 
1598 	/*
1599 	 * Disable Wake-on-LAN
1600 	 */
1601 	IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
1602 
1603 	/*
1604 	 * Some adapters offer Energy Efficient Ethernet (EEE) support.
1605 	 * Due to issues with EEE in e1000g/igb, we disable this by default
1606 	 * as a precautionary measure.
1607 	 *
1608 	 * Currently, this is present on a number of the X550 family parts.
1609 	 */
1610 	(void) ixgbe_setup_eee(hw, B_FALSE);
1611 
1612 	/*
1613 	 * Turn on any present SFP Tx laser
1614 	 */
1615 	ixgbe_enable_tx_laser(hw);
1616 
1617 	/*
1618 	 * Power on the PHY
1619 	 */
1620 	(void) ixgbe_set_phy_power(hw, B_TRUE);
1621 
1622 	/*
1623 	 * Save the state of the PHY
1624 	 */
1625 	ixgbe_get_hw_state(ixgbe);
1626 
1627 	/*
1628 	 * Make sure driver has control
1629 	 */
1630 	ixgbe_get_driver_control(hw);
1631 
1632 	return (IXGBE_SUCCESS);
1633 }
1634 
1635 /*
1636  * ixgbe_chip_stop - Stop the chipset hardware
1637  */
1638 static void
1639 ixgbe_chip_stop(ixgbe_t *ixgbe)
1640 {
1641 	struct ixgbe_hw *hw = &ixgbe->hw;
1642 	int rv;
1643 
1644 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1645 
1646 	/*
1647 	 * Stop interupt generation and disable Tx unit
1648 	 */
1649 	hw->adapter_stopped = B_FALSE;
1650 	(void) ixgbe_stop_adapter(hw);
1651 
1652 	/*
1653 	 * Reset the chipset
1654 	 */
1655 	(void) ixgbe_reset_hw(hw);
1656 
1657 	/*
1658 	 * Reset PHY
1659 	 */
1660 	(void) ixgbe_reset_phy(hw);
1661 
1662 	/*
1663 	 * Enter LPLU (Low Power, Link Up) mode, if available. Avoid resetting
1664 	 * the PHY while doing so. Else, just power down the PHY.
1665 	 */
1666 	if (hw->phy.ops.enter_lplu != NULL) {
1667 		hw->phy.reset_disable = B_TRUE;
1668 		rv = hw->phy.ops.enter_lplu(hw);
1669 		if (rv != IXGBE_SUCCESS)
1670 			ixgbe_error(ixgbe, "Error while entering LPLU: %d", rv);
1671 		hw->phy.reset_disable = B_FALSE;
1672 	} else {
1673 		(void) ixgbe_set_phy_power(hw, B_FALSE);
1674 	}
1675 
1676 	/*
1677 	 * Turn off any present SFP Tx laser
1678 	 * Expected for health and safety reasons
1679 	 */
1680 	ixgbe_disable_tx_laser(hw);
1681 
1682 	/*
1683 	 * Tell firmware driver is no longer in control
1684 	 */
1685 	ixgbe_release_driver_control(hw);
1686 
1687 }
1688 
1689 /*
1690  * ixgbe_reset - Reset the chipset and re-start the driver.
1691  *
1692  * It involves stopping and re-starting the chipset,
1693  * and re-configuring the rx/tx rings.
1694  */
1695 static int
1696 ixgbe_reset(ixgbe_t *ixgbe)
1697 {
1698 	int i;
1699 
1700 	/*
1701 	 * Disable and stop the watchdog timer
1702 	 */
1703 	ixgbe_disable_watchdog_timer(ixgbe);
1704 
1705 	mutex_enter(&ixgbe->gen_lock);
1706 
1707 	ASSERT(ixgbe->ixgbe_state & IXGBE_STARTED);
1708 	atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_STARTED);
1709 
1710 	ixgbe_stop(ixgbe, B_FALSE);
1711 
1712 	if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
1713 		mutex_exit(&ixgbe->gen_lock);
1714 		return (IXGBE_FAILURE);
1715 	}
1716 
1717 	/*
1718 	 * After resetting, need to recheck the link status.
1719 	 */
1720 	ixgbe->link_check_complete = B_FALSE;
1721 	ixgbe->link_check_hrtime = gethrtime() +
1722 	    (IXGBE_LINK_UP_TIME * 100000000ULL);
1723 
1724 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_STARTED);
1725 
1726 	if (!(ixgbe->ixgbe_state & IXGBE_SUSPENDED)) {
1727 		for (i = 0; i < ixgbe->num_tx_rings; i++) {
1728 			mac_tx_ring_update(ixgbe->mac_hdl,
1729 			    ixgbe->tx_rings[i].ring_handle);
1730 		}
1731 	}
1732 
1733 	mutex_exit(&ixgbe->gen_lock);
1734 
1735 	/*
1736 	 * Enable and start the watchdog timer
1737 	 */
1738 	ixgbe_enable_watchdog_timer(ixgbe);
1739 
1740 	return (IXGBE_SUCCESS);
1741 }
1742 
1743 /*
1744  * ixgbe_tx_clean - Clean the pending transmit packets and DMA resources.
1745  */
1746 static void
1747 ixgbe_tx_clean(ixgbe_t *ixgbe)
1748 {
1749 	ixgbe_tx_ring_t *tx_ring;
1750 	tx_control_block_t *tcb;
1751 	link_list_t pending_list;
1752 	uint32_t desc_num;
1753 	int i, j;
1754 
1755 	LINK_LIST_INIT(&pending_list);
1756 
1757 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1758 		tx_ring = &ixgbe->tx_rings[i];
1759 
1760 		mutex_enter(&tx_ring->recycle_lock);
1761 
1762 		/*
1763 		 * Clean the pending tx data - the pending packets in the
1764 		 * work_list that have no chances to be transmitted again.
1765 		 *
1766 		 * We must ensure the chipset is stopped or the link is down
1767 		 * before cleaning the transmit packets.
1768 		 */
1769 		desc_num = 0;
1770 		for (j = 0; j < tx_ring->ring_size; j++) {
1771 			tcb = tx_ring->work_list[j];
1772 			if (tcb != NULL) {
1773 				desc_num += tcb->desc_num;
1774 
1775 				tx_ring->work_list[j] = NULL;
1776 
1777 				ixgbe_free_tcb(tcb);
1778 
1779 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1780 			}
1781 		}
1782 
1783 		if (desc_num > 0) {
1784 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1785 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1786 
1787 			/*
1788 			 * Reset the head and tail pointers of the tbd ring;
1789 			 * Reset the writeback head if it's enable.
1790 			 */
1791 			tx_ring->tbd_head = 0;
1792 			tx_ring->tbd_tail = 0;
1793 			if (ixgbe->tx_head_wb_enable)
1794 				*tx_ring->tbd_head_wb = 0;
1795 
1796 			IXGBE_WRITE_REG(&ixgbe->hw,
1797 			    IXGBE_TDH(tx_ring->index), 0);
1798 			IXGBE_WRITE_REG(&ixgbe->hw,
1799 			    IXGBE_TDT(tx_ring->index), 0);
1800 		}
1801 
1802 		mutex_exit(&tx_ring->recycle_lock);
1803 
1804 		/*
1805 		 * Add the tx control blocks in the pending list to
1806 		 * the free list.
1807 		 */
1808 		ixgbe_put_free_list(tx_ring, &pending_list);
1809 	}
1810 }
1811 
1812 /*
1813  * ixgbe_tx_drain - Drain the tx rings to allow pending packets to be
1814  * transmitted.
1815  */
1816 static boolean_t
1817 ixgbe_tx_drain(ixgbe_t *ixgbe)
1818 {
1819 	ixgbe_tx_ring_t *tx_ring;
1820 	boolean_t done;
1821 	int i, j;
1822 
1823 	/*
1824 	 * Wait for a specific time to allow pending tx packets
1825 	 * to be transmitted.
1826 	 *
1827 	 * Check the counter tbd_free to see if transmission is done.
1828 	 * No lock protection is needed here.
1829 	 *
1830 	 * Return B_TRUE if all pending packets have been transmitted;
1831 	 * Otherwise return B_FALSE;
1832 	 */
1833 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1834 
1835 		done = B_TRUE;
1836 		for (j = 0; j < ixgbe->num_tx_rings; j++) {
1837 			tx_ring = &ixgbe->tx_rings[j];
1838 			done = done &&
1839 			    (tx_ring->tbd_free == tx_ring->ring_size);
1840 		}
1841 
1842 		if (done)
1843 			break;
1844 
1845 		msec_delay(1);
1846 	}
1847 
1848 	return (done);
1849 }
1850 
1851 /*
1852  * ixgbe_rx_drain - Wait for all rx buffers to be released by upper layer.
1853  */
1854 static boolean_t
1855 ixgbe_rx_drain(ixgbe_t *ixgbe)
1856 {
1857 	boolean_t done = B_TRUE;
1858 	int i;
1859 
1860 	/*
1861 	 * Polling the rx free list to check if those rx buffers held by
1862 	 * the upper layer are released.
1863 	 *
1864 	 * Check the counter rcb_free to see if all pending buffers are
1865 	 * released. No lock protection is needed here.
1866 	 *
1867 	 * Return B_TRUE if all pending buffers have been released;
1868 	 * Otherwise return B_FALSE;
1869 	 */
1870 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1871 		done = (ixgbe->rcb_pending == 0);
1872 
1873 		if (done)
1874 			break;
1875 
1876 		msec_delay(1);
1877 	}
1878 
1879 	return (done);
1880 }
1881 
1882 /*
1883  * ixgbe_start - Start the driver/chipset.
1884  */
1885 int
1886 ixgbe_start(ixgbe_t *ixgbe, boolean_t alloc_buffer)
1887 {
1888 	struct ixgbe_hw *hw = &ixgbe->hw;
1889 	int i;
1890 
1891 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1892 
1893 	if (alloc_buffer) {
1894 		if (ixgbe_alloc_rx_data(ixgbe) != IXGBE_SUCCESS) {
1895 			ixgbe_error(ixgbe,
1896 			    "Failed to allocate software receive rings");
1897 			return (IXGBE_FAILURE);
1898 		}
1899 
1900 		/* Allocate buffers for all the rx/tx rings */
1901 		if (ixgbe_alloc_dma(ixgbe) != IXGBE_SUCCESS) {
1902 			ixgbe_error(ixgbe, "Failed to allocate DMA resource");
1903 			return (IXGBE_FAILURE);
1904 		}
1905 
1906 		ixgbe->tx_ring_init = B_TRUE;
1907 	} else {
1908 		ixgbe->tx_ring_init = B_FALSE;
1909 	}
1910 
1911 	for (i = 0; i < ixgbe->num_rx_rings; i++)
1912 		mutex_enter(&ixgbe->rx_rings[i].rx_lock);
1913 	for (i = 0; i < ixgbe->num_tx_rings; i++)
1914 		mutex_enter(&ixgbe->tx_rings[i].tx_lock);
1915 
1916 	/*
1917 	 * Start the chipset hardware
1918 	 */
1919 	if (ixgbe_chip_start(ixgbe) != IXGBE_SUCCESS) {
1920 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1921 		goto start_failure;
1922 	}
1923 
1924 	/*
1925 	 * Configure link now for X550
1926 	 *
1927 	 * X550 possesses a LPLU (Low-Power Link Up) mode which keeps the
1928 	 * resting state of the adapter at a 1Gb FDX speed. Prior to the X550,
1929 	 * the resting state of the link would be the maximum speed that
1930 	 * autonegotiation will allow (usually 10Gb, infrastructure allowing)
1931 	 * so we never bothered with explicitly setting the link to 10Gb as it
1932 	 * would already be at that state on driver attach. With X550, we must
1933 	 * trigger a re-negotiation of the link in order to switch from a LPLU
1934 	 * 1Gb link to 10Gb (cable and link partner permitting.)
1935 	 */
1936 	if (hw->mac.type == ixgbe_mac_X550 ||
1937 	    hw->mac.type == ixgbe_mac_X550EM_a ||
1938 	    hw->mac.type == ixgbe_mac_X550EM_x) {
1939 		(void) ixgbe_driver_setup_link(ixgbe, B_TRUE);
1940 		ixgbe_get_hw_state(ixgbe);
1941 	}
1942 
1943 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1944 		goto start_failure;
1945 	}
1946 
1947 	/*
1948 	 * Setup the rx/tx rings
1949 	 */
1950 	if (ixgbe_setup_rings(ixgbe) != IXGBE_SUCCESS)
1951 		goto start_failure;
1952 
1953 	/*
1954 	 * ixgbe_start() will be called when resetting, however if reset
1955 	 * happens, we need to clear the ERROR, STALL and OVERTEMP flags
1956 	 * before enabling the interrupts.
1957 	 */
1958 	atomic_and_32(&ixgbe->ixgbe_state, ~(IXGBE_ERROR
1959 	    | IXGBE_STALL| IXGBE_OVERTEMP));
1960 
1961 	/*
1962 	 * Enable adapter interrupts
1963 	 * The interrupts must be enabled after the driver state is START
1964 	 */
1965 	ixgbe_enable_adapter_interrupts(ixgbe);
1966 
1967 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1968 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1969 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1970 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1971 
1972 	return (IXGBE_SUCCESS);
1973 
1974 start_failure:
1975 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1976 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1977 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1978 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1979 
1980 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1981 
1982 	return (IXGBE_FAILURE);
1983 }
1984 
1985 /*
1986  * ixgbe_stop - Stop the driver/chipset.
1987  */
1988 void
1989 ixgbe_stop(ixgbe_t *ixgbe, boolean_t free_buffer)
1990 {
1991 	int i;
1992 
1993 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1994 
1995 	/*
1996 	 * Disable the adapter interrupts
1997 	 */
1998 	ixgbe_disable_adapter_interrupts(ixgbe);
1999 
2000 	/*
2001 	 * Drain the pending tx packets
2002 	 */
2003 	(void) ixgbe_tx_drain(ixgbe);
2004 
2005 	for (i = 0; i < ixgbe->num_rx_rings; i++)
2006 		mutex_enter(&ixgbe->rx_rings[i].rx_lock);
2007 	for (i = 0; i < ixgbe->num_tx_rings; i++)
2008 		mutex_enter(&ixgbe->tx_rings[i].tx_lock);
2009 
2010 	/*
2011 	 * Stop the chipset hardware
2012 	 */
2013 	ixgbe_chip_stop(ixgbe);
2014 
2015 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
2016 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
2017 	}
2018 
2019 	/*
2020 	 * Clean the pending tx data/resources
2021 	 */
2022 	ixgbe_tx_clean(ixgbe);
2023 
2024 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
2025 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
2026 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
2027 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
2028 
2029 	if (ixgbe->link_state == LINK_STATE_UP) {
2030 		ixgbe->link_state = LINK_STATE_UNKNOWN;
2031 		mac_link_update(ixgbe->mac_hdl, ixgbe->link_state);
2032 	}
2033 
2034 	if (free_buffer) {
2035 		/*
2036 		 * Release the DMA/memory resources of rx/tx rings
2037 		 */
2038 		ixgbe_free_dma(ixgbe);
2039 		ixgbe_free_rx_data(ixgbe);
2040 	}
2041 }
2042 
2043 /*
2044  * ixgbe_cbfunc - Driver interface for generic DDI callbacks
2045  */
2046 /* ARGSUSED */
2047 static int
2048 ixgbe_cbfunc(dev_info_t *dip, ddi_cb_action_t cbaction, void *cbarg,
2049     void *arg1, void *arg2)
2050 {
2051 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
2052 
2053 	switch (cbaction) {
2054 	/* IRM callback */
2055 	int count;
2056 	case DDI_CB_INTR_ADD:
2057 	case DDI_CB_INTR_REMOVE:
2058 		count = (int)(uintptr_t)cbarg;
2059 		ASSERT(ixgbe->intr_type == DDI_INTR_TYPE_MSIX);
2060 		DTRACE_PROBE2(ixgbe__irm__callback, int, count,
2061 		    int, ixgbe->intr_cnt);
2062 		if (ixgbe_intr_adjust(ixgbe, cbaction, count) !=
2063 		    DDI_SUCCESS) {
2064 			ixgbe_error(ixgbe,
2065 			    "IRM CB: Failed to adjust interrupts");
2066 			goto cb_fail;
2067 		}
2068 		break;
2069 	default:
2070 		IXGBE_DEBUGLOG_1(ixgbe, "DDI CB: action 0x%x NOT supported",
2071 		    cbaction);
2072 		return (DDI_ENOTSUP);
2073 	}
2074 	return (DDI_SUCCESS);
2075 cb_fail:
2076 	return (DDI_FAILURE);
2077 }
2078 
2079 /*
2080  * ixgbe_intr_adjust - Adjust interrupt to respond to IRM request.
2081  */
2082 static int
2083 ixgbe_intr_adjust(ixgbe_t *ixgbe, ddi_cb_action_t cbaction, int count)
2084 {
2085 	int i, rc, actual;
2086 
2087 	if (count == 0)
2088 		return (DDI_SUCCESS);
2089 
2090 	if ((cbaction == DDI_CB_INTR_ADD &&
2091 	    ixgbe->intr_cnt + count > ixgbe->intr_cnt_max) ||
2092 	    (cbaction == DDI_CB_INTR_REMOVE &&
2093 	    ixgbe->intr_cnt - count < ixgbe->intr_cnt_min))
2094 		return (DDI_FAILURE);
2095 
2096 	if (!(ixgbe->ixgbe_state & IXGBE_STARTED)) {
2097 		return (DDI_FAILURE);
2098 	}
2099 
2100 	for (i = 0; i < ixgbe->num_rx_rings; i++)
2101 		mac_ring_intr_set(ixgbe->rx_rings[i].ring_handle, NULL);
2102 	for (i = 0; i < ixgbe->num_tx_rings; i++)
2103 		mac_ring_intr_set(ixgbe->tx_rings[i].ring_handle, NULL);
2104 
2105 	mutex_enter(&ixgbe->gen_lock);
2106 	ixgbe->ixgbe_state &= ~IXGBE_STARTED;
2107 	ixgbe->ixgbe_state |= IXGBE_INTR_ADJUST;
2108 	ixgbe->ixgbe_state |= IXGBE_SUSPENDED;
2109 	mac_link_update(ixgbe->mac_hdl, LINK_STATE_UNKNOWN);
2110 
2111 	ixgbe_stop(ixgbe, B_FALSE);
2112 	/*
2113 	 * Disable interrupts
2114 	 */
2115 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
2116 		rc = ixgbe_disable_intrs(ixgbe);
2117 		ASSERT(rc == IXGBE_SUCCESS);
2118 	}
2119 	ixgbe->attach_progress &= ~ATTACH_PROGRESS_ENABLE_INTR;
2120 
2121 	/*
2122 	 * Remove interrupt handlers
2123 	 */
2124 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
2125 		ixgbe_rem_intr_handlers(ixgbe);
2126 	}
2127 	ixgbe->attach_progress &= ~ATTACH_PROGRESS_ADD_INTR;
2128 
2129 	/*
2130 	 * Clear vect_map
2131 	 */
2132 	bzero(&ixgbe->vect_map, sizeof (ixgbe->vect_map));
2133 	switch (cbaction) {
2134 	case DDI_CB_INTR_ADD:
2135 		rc = ddi_intr_alloc(ixgbe->dip, ixgbe->htable,
2136 		    DDI_INTR_TYPE_MSIX, ixgbe->intr_cnt, count, &actual,
2137 		    DDI_INTR_ALLOC_NORMAL);
2138 		if (rc != DDI_SUCCESS || actual != count) {
2139 			ixgbe_log(ixgbe, "Adjust interrupts failed."
2140 			    "return: %d, irm cb size: %d, actual: %d",
2141 			    rc, count, actual);
2142 			goto intr_adjust_fail;
2143 		}
2144 		ixgbe->intr_cnt += count;
2145 		break;
2146 
2147 	case DDI_CB_INTR_REMOVE:
2148 		for (i = ixgbe->intr_cnt - count;
2149 		    i < ixgbe->intr_cnt; i ++) {
2150 			rc = ddi_intr_free(ixgbe->htable[i]);
2151 			ixgbe->htable[i] = NULL;
2152 			if (rc != DDI_SUCCESS) {
2153 				ixgbe_log(ixgbe, "Adjust interrupts failed."
2154 				    "return: %d, irm cb size: %d, actual: %d",
2155 				    rc, count, actual);
2156 				goto intr_adjust_fail;
2157 			}
2158 		}
2159 		ixgbe->intr_cnt -= count;
2160 		break;
2161 	}
2162 
2163 	/*
2164 	 * Get priority for first vector, assume remaining are all the same
2165 	 */
2166 	rc = ddi_intr_get_pri(ixgbe->htable[0], &ixgbe->intr_pri);
2167 	if (rc != DDI_SUCCESS) {
2168 		ixgbe_log(ixgbe,
2169 		    "Get interrupt priority failed: %d", rc);
2170 		goto intr_adjust_fail;
2171 	}
2172 	rc = ddi_intr_get_cap(ixgbe->htable[0], &ixgbe->intr_cap);
2173 	if (rc != DDI_SUCCESS) {
2174 		ixgbe_log(ixgbe, "Get interrupt cap failed: %d", rc);
2175 		goto intr_adjust_fail;
2176 	}
2177 	ixgbe->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
2178 
2179 	/*
2180 	 * Map rings to interrupt vectors
2181 	 */
2182 	if (ixgbe_map_intrs_to_vectors(ixgbe) != IXGBE_SUCCESS) {
2183 		ixgbe_error(ixgbe,
2184 		    "IRM CB: Failed to map interrupts to vectors");
2185 		goto intr_adjust_fail;
2186 	}
2187 
2188 	/*
2189 	 * Add interrupt handlers
2190 	 */
2191 	if (ixgbe_add_intr_handlers(ixgbe) != IXGBE_SUCCESS) {
2192 		ixgbe_error(ixgbe, "IRM CB: Failed to add interrupt handlers");
2193 		goto intr_adjust_fail;
2194 	}
2195 	ixgbe->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
2196 
2197 	/*
2198 	 * Now that mutex locks are initialized, and the chip is also
2199 	 * initialized, enable interrupts.
2200 	 */
2201 	if (ixgbe_enable_intrs(ixgbe) != IXGBE_SUCCESS) {
2202 		ixgbe_error(ixgbe, "IRM CB: Failed to enable DDI interrupts");
2203 		goto intr_adjust_fail;
2204 	}
2205 	ixgbe->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
2206 	if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
2207 		ixgbe_error(ixgbe, "IRM CB: Failed to start");
2208 		goto intr_adjust_fail;
2209 	}
2210 	ixgbe->ixgbe_state &= ~IXGBE_INTR_ADJUST;
2211 	ixgbe->ixgbe_state &= ~IXGBE_SUSPENDED;
2212 	ixgbe->ixgbe_state |= IXGBE_STARTED;
2213 	mutex_exit(&ixgbe->gen_lock);
2214 
2215 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2216 		mac_ring_intr_set(ixgbe->rx_rings[i].ring_handle,
2217 		    ixgbe->htable[ixgbe->rx_rings[i].intr_vector]);
2218 	}
2219 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2220 		mac_ring_intr_set(ixgbe->tx_rings[i].ring_handle,
2221 		    ixgbe->htable[ixgbe->tx_rings[i].intr_vector]);
2222 	}
2223 
2224 	/* Wakeup all Tx rings */
2225 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2226 		mac_tx_ring_update(ixgbe->mac_hdl,
2227 		    ixgbe->tx_rings[i].ring_handle);
2228 	}
2229 
2230 	IXGBE_DEBUGLOG_3(ixgbe,
2231 	    "IRM CB: interrupts new value: 0x%x(0x%x:0x%x).",
2232 	    ixgbe->intr_cnt, ixgbe->intr_cnt_min, ixgbe->intr_cnt_max);
2233 	return (DDI_SUCCESS);
2234 
2235 intr_adjust_fail:
2236 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
2237 	mutex_exit(&ixgbe->gen_lock);
2238 	return (DDI_FAILURE);
2239 }
2240 
2241 /*
2242  * ixgbe_intr_cb_register - Register interrupt callback function.
2243  */
2244 static int
2245 ixgbe_intr_cb_register(ixgbe_t *ixgbe)
2246 {
2247 	if (ddi_cb_register(ixgbe->dip, DDI_CB_FLAG_INTR, ixgbe_cbfunc,
2248 	    ixgbe, NULL, &ixgbe->cb_hdl) != DDI_SUCCESS) {
2249 		return (IXGBE_FAILURE);
2250 	}
2251 	IXGBE_DEBUGLOG_0(ixgbe, "Interrupt callback function registered.");
2252 	return (IXGBE_SUCCESS);
2253 }
2254 
2255 /*
2256  * ixgbe_alloc_rings - Allocate memory space for rx/tx rings.
2257  */
2258 static int
2259 ixgbe_alloc_rings(ixgbe_t *ixgbe)
2260 {
2261 	/*
2262 	 * Allocate memory space for rx rings
2263 	 */
2264 	ixgbe->rx_rings = kmem_zalloc(
2265 	    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings,
2266 	    KM_NOSLEEP);
2267 
2268 	if (ixgbe->rx_rings == NULL) {
2269 		return (IXGBE_FAILURE);
2270 	}
2271 
2272 	/*
2273 	 * Allocate memory space for tx rings
2274 	 */
2275 	ixgbe->tx_rings = kmem_zalloc(
2276 	    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings,
2277 	    KM_NOSLEEP);
2278 
2279 	if (ixgbe->tx_rings == NULL) {
2280 		kmem_free(ixgbe->rx_rings,
2281 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
2282 		ixgbe->rx_rings = NULL;
2283 		return (IXGBE_FAILURE);
2284 	}
2285 
2286 	/*
2287 	 * Allocate memory space for rx ring groups
2288 	 */
2289 	ixgbe->rx_groups = kmem_zalloc(
2290 	    sizeof (ixgbe_rx_group_t) * ixgbe->num_rx_groups,
2291 	    KM_NOSLEEP);
2292 
2293 	if (ixgbe->rx_groups == NULL) {
2294 		kmem_free(ixgbe->rx_rings,
2295 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
2296 		kmem_free(ixgbe->tx_rings,
2297 		    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings);
2298 		ixgbe->rx_rings = NULL;
2299 		ixgbe->tx_rings = NULL;
2300 		return (IXGBE_FAILURE);
2301 	}
2302 
2303 	return (IXGBE_SUCCESS);
2304 }
2305 
2306 /*
2307  * ixgbe_free_rings - Free the memory space of rx/tx rings.
2308  */
2309 static void
2310 ixgbe_free_rings(ixgbe_t *ixgbe)
2311 {
2312 	if (ixgbe->rx_rings != NULL) {
2313 		kmem_free(ixgbe->rx_rings,
2314 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
2315 		ixgbe->rx_rings = NULL;
2316 	}
2317 
2318 	if (ixgbe->tx_rings != NULL) {
2319 		kmem_free(ixgbe->tx_rings,
2320 		    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings);
2321 		ixgbe->tx_rings = NULL;
2322 	}
2323 
2324 	for (uint_t i = 0; i < ixgbe->num_rx_groups; i++) {
2325 		ixgbe_vlan_t *vlp;
2326 		ixgbe_rx_group_t *rx_group = &ixgbe->rx_groups[i];
2327 
2328 		while ((vlp = list_remove_head(&rx_group->vlans)) != NULL)
2329 			kmem_free(vlp, sizeof (ixgbe_vlan_t));
2330 
2331 		list_destroy(&rx_group->vlans);
2332 	}
2333 
2334 	if (ixgbe->rx_groups != NULL) {
2335 		kmem_free(ixgbe->rx_groups,
2336 		    sizeof (ixgbe_rx_group_t) * ixgbe->num_rx_groups);
2337 		ixgbe->rx_groups = NULL;
2338 	}
2339 }
2340 
2341 static int
2342 ixgbe_alloc_rx_data(ixgbe_t *ixgbe)
2343 {
2344 	ixgbe_rx_ring_t *rx_ring;
2345 	int i;
2346 
2347 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2348 		rx_ring = &ixgbe->rx_rings[i];
2349 		if (ixgbe_alloc_rx_ring_data(rx_ring) != IXGBE_SUCCESS)
2350 			goto alloc_rx_rings_failure;
2351 	}
2352 	return (IXGBE_SUCCESS);
2353 
2354 alloc_rx_rings_failure:
2355 	ixgbe_free_rx_data(ixgbe);
2356 	return (IXGBE_FAILURE);
2357 }
2358 
2359 static void
2360 ixgbe_free_rx_data(ixgbe_t *ixgbe)
2361 {
2362 	ixgbe_rx_ring_t *rx_ring;
2363 	ixgbe_rx_data_t *rx_data;
2364 	int i;
2365 
2366 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2367 		rx_ring = &ixgbe->rx_rings[i];
2368 
2369 		mutex_enter(&ixgbe->rx_pending_lock);
2370 		rx_data = rx_ring->rx_data;
2371 
2372 		if (rx_data != NULL) {
2373 			rx_data->flag |= IXGBE_RX_STOPPED;
2374 
2375 			if (rx_data->rcb_pending == 0) {
2376 				ixgbe_free_rx_ring_data(rx_data);
2377 				rx_ring->rx_data = NULL;
2378 			}
2379 		}
2380 
2381 		mutex_exit(&ixgbe->rx_pending_lock);
2382 	}
2383 }
2384 
2385 /*
2386  * ixgbe_setup_rings - Setup rx/tx rings.
2387  */
2388 static int
2389 ixgbe_setup_rings(ixgbe_t *ixgbe)
2390 {
2391 	/*
2392 	 * Setup the rx/tx rings, including the following:
2393 	 *
2394 	 * 1. Setup the descriptor ring and the control block buffers;
2395 	 * 2. Initialize necessary registers for receive/transmit;
2396 	 * 3. Initialize software pointers/parameters for receive/transmit;
2397 	 */
2398 	if (ixgbe_setup_rx(ixgbe) != IXGBE_SUCCESS)
2399 		return (IXGBE_FAILURE);
2400 
2401 	ixgbe_setup_tx(ixgbe);
2402 
2403 	return (IXGBE_SUCCESS);
2404 }
2405 
2406 static void
2407 ixgbe_setup_rx_ring(ixgbe_rx_ring_t *rx_ring)
2408 {
2409 	ixgbe_t *ixgbe = rx_ring->ixgbe;
2410 	ixgbe_rx_data_t *rx_data = rx_ring->rx_data;
2411 	struct ixgbe_hw *hw = &ixgbe->hw;
2412 	rx_control_block_t *rcb;
2413 	union ixgbe_adv_rx_desc	*rbd;
2414 	uint32_t size;
2415 	uint32_t buf_low;
2416 	uint32_t buf_high;
2417 	uint32_t reg_val;
2418 	int i;
2419 
2420 	ASSERT(mutex_owned(&rx_ring->rx_lock));
2421 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2422 
2423 	for (i = 0; i < ixgbe->rx_ring_size; i++) {
2424 		rcb = rx_data->work_list[i];
2425 		rbd = &rx_data->rbd_ring[i];
2426 
2427 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
2428 		rbd->read.hdr_addr = 0;
2429 	}
2430 
2431 	/*
2432 	 * Initialize the length register
2433 	 */
2434 	size = rx_data->ring_size * sizeof (union ixgbe_adv_rx_desc);
2435 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(rx_ring->hw_index), size);
2436 
2437 	/*
2438 	 * Initialize the base address registers
2439 	 */
2440 	buf_low = (uint32_t)rx_data->rbd_area.dma_address;
2441 	buf_high = (uint32_t)(rx_data->rbd_area.dma_address >> 32);
2442 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(rx_ring->hw_index), buf_high);
2443 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(rx_ring->hw_index), buf_low);
2444 
2445 	/*
2446 	 * Setup head & tail pointers
2447 	 */
2448 	IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->hw_index),
2449 	    rx_data->ring_size - 1);
2450 	IXGBE_WRITE_REG(hw, IXGBE_RDH(rx_ring->hw_index), 0);
2451 
2452 	rx_data->rbd_next = 0;
2453 	rx_data->lro_first = 0;
2454 
2455 	/*
2456 	 * Setup the Receive Descriptor Control Register (RXDCTL)
2457 	 * PTHRESH=32 descriptors (half the internal cache)
2458 	 * HTHRESH=0 descriptors (to minimize latency on fetch)
2459 	 * WTHRESH defaults to 1 (writeback each descriptor)
2460 	 */
2461 	reg_val = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rx_ring->hw_index));
2462 	reg_val |= IXGBE_RXDCTL_ENABLE;	/* enable queue */
2463 
2464 	/* Not a valid value for 82599, X540 or X550 */
2465 	if (hw->mac.type == ixgbe_mac_82598EB) {
2466 		reg_val |= 0x0020;	/* pthresh */
2467 	}
2468 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rx_ring->hw_index), reg_val);
2469 
2470 	if (hw->mac.type == ixgbe_mac_82599EB ||
2471 	    hw->mac.type == ixgbe_mac_X540 ||
2472 	    hw->mac.type == ixgbe_mac_X550 ||
2473 	    hw->mac.type == ixgbe_mac_X550EM_x ||
2474 	    hw->mac.type == ixgbe_mac_X550EM_a) {
2475 		reg_val = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2476 		reg_val |= (IXGBE_RDRXCTL_CRCSTRIP | IXGBE_RDRXCTL_AGGDIS);
2477 		IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, reg_val);
2478 	}
2479 
2480 	/*
2481 	 * Setup the Split and Replication Receive Control Register.
2482 	 * Set the rx buffer size and the advanced descriptor type.
2483 	 */
2484 	reg_val = (ixgbe->rx_buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) |
2485 	    IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2486 	reg_val |= IXGBE_SRRCTL_DROP_EN;
2487 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rx_ring->hw_index), reg_val);
2488 }
2489 
2490 static int
2491 ixgbe_setup_rx(ixgbe_t *ixgbe)
2492 {
2493 	ixgbe_rx_ring_t *rx_ring;
2494 	struct ixgbe_hw *hw = &ixgbe->hw;
2495 	uint32_t reg_val;
2496 	uint32_t i;
2497 	uint32_t psrtype_rss_bit;
2498 
2499 	/*
2500 	 * Ensure that Rx is disabled while setting up
2501 	 * the Rx unit and Rx descriptor ring(s)
2502 	 */
2503 	ixgbe_disable_rx(hw);
2504 
2505 	/* PSRTYPE must be configured for 82599 */
2506 	if (ixgbe->classify_mode != IXGBE_CLASSIFY_VMDQ &&
2507 	    ixgbe->classify_mode != IXGBE_CLASSIFY_VMDQ_RSS) {
2508 		reg_val = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
2509 		    IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR;
2510 		reg_val |= IXGBE_PSRTYPE_L2HDR;
2511 		reg_val |= 0x80000000;
2512 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), reg_val);
2513 	} else {
2514 		if (ixgbe->num_rx_groups > 32) {
2515 			psrtype_rss_bit = 0x20000000;
2516 		} else {
2517 			psrtype_rss_bit = 0x40000000;
2518 		}
2519 		for (i = 0; i < ixgbe->capab->max_rx_grp_num; i++) {
2520 			reg_val = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
2521 			    IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR;
2522 			reg_val |= IXGBE_PSRTYPE_L2HDR;
2523 			reg_val |= psrtype_rss_bit;
2524 			IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(i), reg_val);
2525 		}
2526 	}
2527 
2528 	/*
2529 	 * Set filter control in FCTRL to determine types of packets are passed
2530 	 * up to the driver.
2531 	 * - Pass broadcast packets.
2532 	 * - Do not pass flow control pause frames (82598-specific)
2533 	 */
2534 	reg_val = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2535 	reg_val |= IXGBE_FCTRL_BAM; /* Broadcast Accept Mode */
2536 	if (hw->mac.type == ixgbe_mac_82598EB) {
2537 		reg_val |= IXGBE_FCTRL_DPF; /* Discard Pause Frames */
2538 	}
2539 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, reg_val);
2540 
2541 	/*
2542 	 * Hardware checksum settings
2543 	 */
2544 	if (ixgbe->rx_hcksum_enable) {
2545 		reg_val = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2546 		reg_val |= IXGBE_RXCSUM_IPPCSE;	/* IP checksum */
2547 		IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, reg_val);
2548 	}
2549 
2550 	/*
2551 	 * Setup VMDq and RSS for multiple receive queues
2552 	 */
2553 	switch (ixgbe->classify_mode) {
2554 	case IXGBE_CLASSIFY_RSS:
2555 		/*
2556 		 * One group, only RSS is needed when more than
2557 		 * one ring enabled.
2558 		 */
2559 		ixgbe_setup_rss(ixgbe);
2560 		break;
2561 
2562 	case IXGBE_CLASSIFY_VMDQ:
2563 		/*
2564 		 * Multiple groups, each group has one ring,
2565 		 * only VMDq is needed.
2566 		 */
2567 		ixgbe_setup_vmdq(ixgbe);
2568 		break;
2569 
2570 	case IXGBE_CLASSIFY_VMDQ_RSS:
2571 		/*
2572 		 * Multiple groups and multiple rings, both
2573 		 * VMDq and RSS are needed.
2574 		 */
2575 		ixgbe_setup_vmdq_rss(ixgbe);
2576 		break;
2577 
2578 	default:
2579 		break;
2580 	}
2581 
2582 	/*
2583 	 * Initialize VLAN SW and HW state if VLAN filtering is
2584 	 * enabled.
2585 	 */
2586 	if (ixgbe->vlft_enabled) {
2587 		if (ixgbe_init_vlan(ixgbe) != IXGBE_SUCCESS)
2588 			return (IXGBE_FAILURE);
2589 	}
2590 
2591 	/*
2592 	 * Enable the receive unit.  This must be done after filter
2593 	 * control is set in FCTRL. On 82598, we disable the descriptor monitor.
2594 	 * 82598 is the only adapter which defines this RXCTRL option.
2595 	 */
2596 	reg_val = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2597 	if (hw->mac.type == ixgbe_mac_82598EB)
2598 		reg_val |= IXGBE_RXCTRL_DMBYPS; /* descriptor monitor bypass */
2599 	reg_val |= IXGBE_RXCTRL_RXEN;
2600 	(void) ixgbe_enable_rx_dma(hw, reg_val);
2601 
2602 	/*
2603 	 * ixgbe_setup_rx_ring must be called after configuring RXCTRL
2604 	 */
2605 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2606 		rx_ring = &ixgbe->rx_rings[i];
2607 		ixgbe_setup_rx_ring(rx_ring);
2608 	}
2609 
2610 	/*
2611 	 * The 82598 controller gives us the RNBC (Receive No Buffer
2612 	 * Count) register to determine the number of frames dropped
2613 	 * due to no available descriptors on the destination queue.
2614 	 * However, this register was removed starting with 82599 and
2615 	 * it was replaced with the RQSMR/QPRDC registers. The nice
2616 	 * thing about the new registers is that they allow you to map
2617 	 * groups of queues to specific stat registers. The bad thing
2618 	 * is there are only 16 slots in the stat registers, so this
2619 	 * won't work when we have 32 Rx groups. Instead, we map all
2620 	 * queues to the zero slot of the stat registers, giving us a
2621 	 * global counter at QPRDC[0] (with the equivalent semantics
2622 	 * of RNBC). Perhaps future controllers will have more slots
2623 	 * and we can implement per-group counters.
2624 	 */
2625 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2626 		uint32_t index = ixgbe->rx_rings[i].hw_index;
2627 		IXGBE_WRITE_REG(hw, IXGBE_RQSMR(index >> 2), 0);
2628 	}
2629 
2630 	/*
2631 	 * The Max Frame Size in MHADD/MAXFRS will be internally increased
2632 	 * by four bytes if the packet has a VLAN field, so includes MTU,
2633 	 * ethernet header and frame check sequence.
2634 	 * Register is MAXFRS in 82599.
2635 	 */
2636 	reg_val = IXGBE_READ_REG(hw, IXGBE_MHADD);
2637 	reg_val &= ~IXGBE_MHADD_MFS_MASK;
2638 	reg_val |= (ixgbe->default_mtu + sizeof (struct ether_header)
2639 	    + ETHERFCSL) << IXGBE_MHADD_MFS_SHIFT;
2640 	IXGBE_WRITE_REG(hw, IXGBE_MHADD, reg_val);
2641 
2642 	/*
2643 	 * Setup Jumbo Frame enable bit
2644 	 */
2645 	reg_val = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2646 	if (ixgbe->default_mtu > ETHERMTU)
2647 		reg_val |= IXGBE_HLREG0_JUMBOEN;
2648 	else
2649 		reg_val &= ~IXGBE_HLREG0_JUMBOEN;
2650 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, reg_val);
2651 
2652 	/*
2653 	 * Setup RSC for multiple receive queues.
2654 	 */
2655 	if (ixgbe->lro_enable) {
2656 		for (i = 0; i < ixgbe->num_rx_rings; i++) {
2657 			/*
2658 			 * Make sure rx_buf_size * MAXDESC not greater
2659 			 * than 65535.
2660 			 * Intel recommends 4 for MAXDESC field value.
2661 			 */
2662 			reg_val = IXGBE_READ_REG(hw, IXGBE_RSCCTL(i));
2663 			reg_val |= IXGBE_RSCCTL_RSCEN;
2664 			if (ixgbe->rx_buf_size == IXGBE_PKG_BUF_16k)
2665 				reg_val |= IXGBE_RSCCTL_MAXDESC_1;
2666 			else
2667 				reg_val |= IXGBE_RSCCTL_MAXDESC_4;
2668 			IXGBE_WRITE_REG(hw,  IXGBE_RSCCTL(i), reg_val);
2669 		}
2670 
2671 		reg_val = IXGBE_READ_REG(hw, IXGBE_RSCDBU);
2672 		reg_val |= IXGBE_RSCDBU_RSCACKDIS;
2673 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU, reg_val);
2674 
2675 		reg_val = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2676 		reg_val |= IXGBE_RDRXCTL_RSCACKC;
2677 		reg_val |= IXGBE_RDRXCTL_FCOE_WRFIX;
2678 		reg_val &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2679 
2680 		IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, reg_val);
2681 	}
2682 
2683 	return (IXGBE_SUCCESS);
2684 }
2685 
2686 static void
2687 ixgbe_setup_tx_ring(ixgbe_tx_ring_t *tx_ring)
2688 {
2689 	ixgbe_t *ixgbe = tx_ring->ixgbe;
2690 	struct ixgbe_hw *hw = &ixgbe->hw;
2691 	uint32_t size;
2692 	uint32_t buf_low;
2693 	uint32_t buf_high;
2694 	uint32_t reg_val;
2695 
2696 	ASSERT(mutex_owned(&tx_ring->tx_lock));
2697 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2698 
2699 	/*
2700 	 * Initialize the length register
2701 	 */
2702 	size = tx_ring->ring_size * sizeof (union ixgbe_adv_tx_desc);
2703 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(tx_ring->index), size);
2704 
2705 	/*
2706 	 * Initialize the base address registers
2707 	 */
2708 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
2709 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
2710 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(tx_ring->index), buf_low);
2711 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(tx_ring->index), buf_high);
2712 
2713 	/*
2714 	 * Setup head & tail pointers
2715 	 */
2716 	IXGBE_WRITE_REG(hw, IXGBE_TDH(tx_ring->index), 0);
2717 	IXGBE_WRITE_REG(hw, IXGBE_TDT(tx_ring->index), 0);
2718 
2719 	/*
2720 	 * Setup head write-back
2721 	 */
2722 	if (ixgbe->tx_head_wb_enable) {
2723 		/*
2724 		 * The memory of the head write-back is allocated using
2725 		 * the extra tbd beyond the tail of the tbd ring.
2726 		 */
2727 		tx_ring->tbd_head_wb = (uint32_t *)
2728 		    ((uintptr_t)tx_ring->tbd_area.address + size);
2729 		*tx_ring->tbd_head_wb = 0;
2730 
2731 		buf_low = (uint32_t)
2732 		    (tx_ring->tbd_area.dma_address + size);
2733 		buf_high = (uint32_t)
2734 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
2735 
2736 		/* Set the head write-back enable bit */
2737 		buf_low |= IXGBE_TDWBAL_HEAD_WB_ENABLE;
2738 
2739 		IXGBE_WRITE_REG(hw, IXGBE_TDWBAL(tx_ring->index), buf_low);
2740 		IXGBE_WRITE_REG(hw, IXGBE_TDWBAH(tx_ring->index), buf_high);
2741 
2742 		/*
2743 		 * Turn off relaxed ordering for head write back or it will
2744 		 * cause problems with the tx recycling
2745 		 */
2746 
2747 		reg_val = (hw->mac.type == ixgbe_mac_82598EB) ?
2748 		    IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(tx_ring->index)) :
2749 		    IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(tx_ring->index));
2750 		reg_val &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
2751 		if (hw->mac.type == ixgbe_mac_82598EB) {
2752 			IXGBE_WRITE_REG(hw,
2753 			    IXGBE_DCA_TXCTRL(tx_ring->index), reg_val);
2754 		} else {
2755 			IXGBE_WRITE_REG(hw,
2756 			    IXGBE_DCA_TXCTRL_82599(tx_ring->index), reg_val);
2757 		}
2758 	} else {
2759 		tx_ring->tbd_head_wb = NULL;
2760 	}
2761 
2762 	tx_ring->tbd_head = 0;
2763 	tx_ring->tbd_tail = 0;
2764 	tx_ring->tbd_free = tx_ring->ring_size;
2765 
2766 	if (ixgbe->tx_ring_init == B_TRUE) {
2767 		tx_ring->tcb_head = 0;
2768 		tx_ring->tcb_tail = 0;
2769 		tx_ring->tcb_free = tx_ring->free_list_size;
2770 	}
2771 
2772 	/*
2773 	 * Initialize the s/w context structure
2774 	 */
2775 	bzero(&tx_ring->tx_context, sizeof (ixgbe_tx_context_t));
2776 }
2777 
2778 static void
2779 ixgbe_setup_tx(ixgbe_t *ixgbe)
2780 {
2781 	struct ixgbe_hw *hw = &ixgbe->hw;
2782 	ixgbe_tx_ring_t *tx_ring;
2783 	uint32_t reg_val;
2784 	int i;
2785 
2786 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2787 		tx_ring = &ixgbe->tx_rings[i];
2788 		ixgbe_setup_tx_ring(tx_ring);
2789 	}
2790 
2791 	/*
2792 	 * Setup the per-ring statistics mapping. We map all Tx queues
2793 	 * to slot 0 to stay consistent with Rx.
2794 	 */
2795 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2796 		switch (hw->mac.type) {
2797 		case ixgbe_mac_82598EB:
2798 			IXGBE_WRITE_REG(hw, IXGBE_TQSMR(i >> 2), 0);
2799 			break;
2800 
2801 		default:
2802 			IXGBE_WRITE_REG(hw, IXGBE_TQSM(i >> 2), 0);
2803 			break;
2804 		}
2805 	}
2806 
2807 	/*
2808 	 * Enable CRC appending and TX padding (for short tx frames)
2809 	 */
2810 	reg_val = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2811 	reg_val |= IXGBE_HLREG0_TXCRCEN | IXGBE_HLREG0_TXPADEN;
2812 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, reg_val);
2813 
2814 	/*
2815 	 * enable DMA for 82599, X540 and X550 parts
2816 	 */
2817 	if (hw->mac.type == ixgbe_mac_82599EB ||
2818 	    hw->mac.type == ixgbe_mac_X540 ||
2819 	    hw->mac.type == ixgbe_mac_X550 ||
2820 	    hw->mac.type == ixgbe_mac_X550EM_x ||
2821 	    hw->mac.type == ixgbe_mac_X550EM_a) {
2822 		/* DMATXCTL.TE must be set after all Tx config is complete */
2823 		reg_val = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2824 		reg_val |= IXGBE_DMATXCTL_TE;
2825 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, reg_val);
2826 
2827 		/* Disable arbiter to set MTQC */
2828 		reg_val = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2829 		reg_val |= IXGBE_RTTDCS_ARBDIS;
2830 		IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg_val);
2831 		IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2832 		reg_val &= ~IXGBE_RTTDCS_ARBDIS;
2833 		IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg_val);
2834 	}
2835 
2836 	/*
2837 	 * Enabling tx queues ..
2838 	 * For 82599 must be done after DMATXCTL.TE is set
2839 	 */
2840 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2841 		tx_ring = &ixgbe->tx_rings[i];
2842 		reg_val = IXGBE_READ_REG(hw, IXGBE_TXDCTL(tx_ring->index));
2843 		reg_val |= IXGBE_TXDCTL_ENABLE;
2844 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(tx_ring->index), reg_val);
2845 	}
2846 }
2847 
2848 /*
2849  * ixgbe_setup_rss - Setup receive-side scaling feature.
2850  */
2851 static void
2852 ixgbe_setup_rss(ixgbe_t *ixgbe)
2853 {
2854 	struct ixgbe_hw *hw = &ixgbe->hw;
2855 	uint32_t mrqc;
2856 
2857 	/*
2858 	 * Initialize RETA/ERETA table
2859 	 */
2860 	ixgbe_setup_rss_table(ixgbe);
2861 
2862 	/*
2863 	 * Enable RSS & perform hash on these packet types
2864 	 */
2865 	mrqc = IXGBE_MRQC_RSSEN |
2866 	    IXGBE_MRQC_RSS_FIELD_IPV4 |
2867 	    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2868 	    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2869 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2870 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2871 	    IXGBE_MRQC_RSS_FIELD_IPV6 |
2872 	    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2873 	    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2874 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2875 	IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2876 }
2877 
2878 /*
2879  * ixgbe_setup_vmdq - Setup MAC classification feature
2880  */
2881 static void
2882 ixgbe_setup_vmdq(ixgbe_t *ixgbe)
2883 {
2884 	struct ixgbe_hw *hw = &ixgbe->hw;
2885 	uint32_t vmdctl, i, vtctl, vlnctl;
2886 
2887 	/*
2888 	 * Setup the VMDq Control register, enable VMDq based on
2889 	 * packet destination MAC address:
2890 	 */
2891 	switch (hw->mac.type) {
2892 	case ixgbe_mac_82598EB:
2893 		/*
2894 		 * VMDq Enable = 1;
2895 		 * VMDq Filter = 0; MAC filtering
2896 		 * Default VMDq output index = 0;
2897 		 */
2898 		vmdctl = IXGBE_VMD_CTL_VMDQ_EN;
2899 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
2900 		break;
2901 
2902 	case ixgbe_mac_82599EB:
2903 	case ixgbe_mac_X540:
2904 	case ixgbe_mac_X550:
2905 	case ixgbe_mac_X550EM_x:
2906 	case ixgbe_mac_X550EM_a:
2907 		/*
2908 		 * Enable VMDq-only.
2909 		 */
2910 		vmdctl = IXGBE_MRQC_VMDQEN;
2911 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, vmdctl);
2912 
2913 		for (i = 0; i < hw->mac.num_rar_entries; i++) {
2914 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(i), 0);
2915 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(i), 0);
2916 		}
2917 
2918 		/*
2919 		 * Enable Virtualization and Replication.
2920 		 */
2921 		vtctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2922 		ixgbe->rx_def_group = vtctl & IXGBE_VT_CTL_POOL_MASK;
2923 		vtctl |= IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2924 		IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vtctl);
2925 
2926 		/*
2927 		 * Enable VLAN filtering and switching (VFTA and VLVF).
2928 		 */
2929 		vlnctl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2930 		vlnctl |= IXGBE_VLNCTRL_VFE;
2931 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctl);
2932 		ixgbe->vlft_enabled = B_TRUE;
2933 
2934 		/*
2935 		 * Enable receiving packets to all VFs
2936 		 */
2937 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), IXGBE_VFRE_ENABLE_ALL);
2938 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), IXGBE_VFRE_ENABLE_ALL);
2939 		break;
2940 
2941 	default:
2942 		break;
2943 	}
2944 }
2945 
2946 /*
2947  * ixgbe_setup_vmdq_rss - Setup both vmdq feature and rss feature.
2948  */
2949 static void
2950 ixgbe_setup_vmdq_rss(ixgbe_t *ixgbe)
2951 {
2952 	struct ixgbe_hw *hw = &ixgbe->hw;
2953 	uint32_t i, mrqc;
2954 	uint32_t vtctl, vmdctl, vlnctl;
2955 
2956 	/*
2957 	 * Initialize RETA/ERETA table
2958 	 */
2959 	ixgbe_setup_rss_table(ixgbe);
2960 
2961 	/*
2962 	 * Enable and setup RSS and VMDq
2963 	 */
2964 	switch (hw->mac.type) {
2965 	case ixgbe_mac_82598EB:
2966 		/*
2967 		 * Enable RSS & Setup RSS Hash functions
2968 		 */
2969 		mrqc = IXGBE_MRQC_RSSEN |
2970 		    IXGBE_MRQC_RSS_FIELD_IPV4 |
2971 		    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2972 		    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2973 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2974 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2975 		    IXGBE_MRQC_RSS_FIELD_IPV6 |
2976 		    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2977 		    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2978 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2979 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2980 
2981 		/*
2982 		 * Enable and Setup VMDq
2983 		 * VMDq Filter = 0; MAC filtering
2984 		 * Default VMDq output index = 0;
2985 		 */
2986 		vmdctl = IXGBE_VMD_CTL_VMDQ_EN;
2987 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
2988 		break;
2989 
2990 	case ixgbe_mac_82599EB:
2991 	case ixgbe_mac_X540:
2992 	case ixgbe_mac_X550:
2993 	case ixgbe_mac_X550EM_x:
2994 	case ixgbe_mac_X550EM_a:
2995 		/*
2996 		 * Enable RSS & Setup RSS Hash functions
2997 		 */
2998 		mrqc = IXGBE_MRQC_RSS_FIELD_IPV4 |
2999 		    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3000 		    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
3001 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
3002 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
3003 		    IXGBE_MRQC_RSS_FIELD_IPV6 |
3004 		    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
3005 		    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
3006 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
3007 
3008 		/*
3009 		 * Enable VMDq+RSS.
3010 		 */
3011 		if (ixgbe->num_rx_groups > 32)  {
3012 			mrqc = mrqc | IXGBE_MRQC_VMDQRSS64EN;
3013 		} else {
3014 			mrqc = mrqc | IXGBE_MRQC_VMDQRSS32EN;
3015 		}
3016 
3017 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3018 
3019 		for (i = 0; i < hw->mac.num_rar_entries; i++) {
3020 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(i), 0);
3021 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(i), 0);
3022 		}
3023 		break;
3024 
3025 	default:
3026 		break;
3027 
3028 	}
3029 
3030 	if (hw->mac.type == ixgbe_mac_82599EB ||
3031 	    hw->mac.type == ixgbe_mac_X540 ||
3032 	    hw->mac.type == ixgbe_mac_X550 ||
3033 	    hw->mac.type == ixgbe_mac_X550EM_x ||
3034 	    hw->mac.type == ixgbe_mac_X550EM_a) {
3035 		/*
3036 		 * Enable Virtualization and Replication.
3037 		 */
3038 		vtctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3039 		ixgbe->rx_def_group = vtctl & IXGBE_VT_CTL_POOL_MASK;
3040 		vtctl |= IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
3041 		vtctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
3042 		IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vtctl);
3043 
3044 		/*
3045 		 * Enable VLAN filtering and switching (VFTA and VLVF).
3046 		 */
3047 		vlnctl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3048 		vlnctl |= IXGBE_VLNCTRL_VFE;
3049 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctl);
3050 		ixgbe->vlft_enabled = B_TRUE;
3051 
3052 		/*
3053 		 * Enable receiving packets to all VFs
3054 		 */
3055 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), IXGBE_VFRE_ENABLE_ALL);
3056 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), IXGBE_VFRE_ENABLE_ALL);
3057 	}
3058 }
3059 
3060 /*
3061  * ixgbe_setup_rss_table - Setup RSS table
3062  */
3063 static void
3064 ixgbe_setup_rss_table(ixgbe_t *ixgbe)
3065 {
3066 	struct ixgbe_hw *hw = &ixgbe->hw;
3067 	uint32_t i, j;
3068 	uint32_t random;
3069 	uint32_t reta;
3070 	uint32_t ring_per_group;
3071 	uint32_t ring;
3072 	uint32_t table_size;
3073 	uint32_t index_mult;
3074 	uint32_t rxcsum;
3075 
3076 	/*
3077 	 * Set multiplier for RETA setup and table size based on MAC type.
3078 	 * RETA table sizes vary by model:
3079 	 *
3080 	 * 82598, 82599, X540: 128 table entries.
3081 	 * X550: 512 table entries.
3082 	 */
3083 	index_mult = 0x1;
3084 	table_size = 128;
3085 	switch (ixgbe->hw.mac.type) {
3086 	case ixgbe_mac_82598EB:
3087 		index_mult = 0x11;
3088 		break;
3089 	case ixgbe_mac_X550:
3090 	case ixgbe_mac_X550EM_x:
3091 	case ixgbe_mac_X550EM_a:
3092 		table_size = 512;
3093 		break;
3094 	default:
3095 		break;
3096 	}
3097 
3098 	/*
3099 	 * Fill out RSS redirection table. The configuation of the indices is
3100 	 * hardware-dependent.
3101 	 *
3102 	 *  82598: 8 bits wide containing two 4 bit RSS indices
3103 	 *  82599, X540: 8 bits wide containing one 4 bit RSS index
3104 	 *  X550: 8 bits wide containing one 6 bit RSS index
3105 	 */
3106 	reta = 0;
3107 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
3108 
3109 	for (i = 0, j = 0; i < table_size; i++, j++) {
3110 		if (j == ring_per_group) j = 0;
3111 
3112 		/*
3113 		 * The low 8 bits are for hash value (n+0);
3114 		 * The next 8 bits are for hash value (n+1), etc.
3115 		 */
3116 		ring = (j * index_mult);
3117 		reta = reta >> 8;
3118 		reta = reta | (((uint32_t)ring) << 24);
3119 
3120 		if ((i & 3) == 3) {
3121 			/*
3122 			 * The first 128 table entries are programmed into the
3123 			 * RETA register, with any beyond that (eg; on X550)
3124 			 * into ERETA.
3125 			 */
3126 			if (i < 128)
3127 				IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3128 			else
3129 				IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3130 				    reta);
3131 			reta = 0;
3132 		}
3133 	}
3134 
3135 	/*
3136 	 * Fill out hash function seeds with a random constant
3137 	 */
3138 	for (i = 0; i < 10; i++) {
3139 		(void) random_get_pseudo_bytes((uint8_t *)&random,
3140 		    sizeof (uint32_t));
3141 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), random);
3142 	}
3143 
3144 	/*
3145 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
3146 	 * It is an adapter hardware limitation that Packet Checksum is
3147 	 * mutually exclusive with RSS.
3148 	 */
3149 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3150 	rxcsum |= IXGBE_RXCSUM_PCSD;
3151 	rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
3152 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3153 }
3154 
3155 /*
3156  * ixgbe_init_unicst - Initialize the unicast addresses.
3157  */
3158 static void
3159 ixgbe_init_unicst(ixgbe_t *ixgbe)
3160 {
3161 	struct ixgbe_hw *hw = &ixgbe->hw;
3162 	uint8_t *mac_addr;
3163 	int slot;
3164 	/*
3165 	 * Here we should consider two situations:
3166 	 *
3167 	 * 1. Chipset is initialized at the first time,
3168 	 *    Clear all the multiple unicast addresses.
3169 	 *
3170 	 * 2. Chipset is reset
3171 	 *    Recover the multiple unicast addresses from the
3172 	 *    software data structure to the RAR registers.
3173 	 */
3174 	if (!ixgbe->unicst_init) {
3175 		/*
3176 		 * Initialize the multiple unicast addresses
3177 		 */
3178 		ixgbe->unicst_total = hw->mac.num_rar_entries;
3179 		ixgbe->unicst_avail = ixgbe->unicst_total;
3180 		for (slot = 0; slot < ixgbe->unicst_total; slot++) {
3181 			mac_addr = ixgbe->unicst_addr[slot].mac.addr;
3182 			bzero(mac_addr, ETHERADDRL);
3183 			(void) ixgbe_set_rar(hw, slot, mac_addr, 0, 0);
3184 			ixgbe->unicst_addr[slot].mac.set = 0;
3185 		}
3186 		ixgbe->unicst_init = B_TRUE;
3187 	} else {
3188 		/* Re-configure the RAR registers */
3189 		for (slot = 0; slot < ixgbe->unicst_total; slot++) {
3190 			mac_addr = ixgbe->unicst_addr[slot].mac.addr;
3191 			if (ixgbe->unicst_addr[slot].mac.set == 1) {
3192 				(void) ixgbe_set_rar(hw, slot, mac_addr,
3193 				    ixgbe->unicst_addr[slot].mac.group_index,
3194 				    IXGBE_RAH_AV);
3195 			} else {
3196 				bzero(mac_addr, ETHERADDRL);
3197 				(void) ixgbe_set_rar(hw, slot, mac_addr, 0, 0);
3198 			}
3199 		}
3200 	}
3201 }
3202 
3203 /*
3204  * ixgbe_unicst_find - Find the slot for the specified unicast address
3205  */
3206 int
3207 ixgbe_unicst_find(ixgbe_t *ixgbe, const uint8_t *mac_addr)
3208 {
3209 	int slot;
3210 
3211 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3212 
3213 	for (slot = 0; slot < ixgbe->unicst_total; slot++) {
3214 		if (bcmp(ixgbe->unicst_addr[slot].mac.addr,
3215 		    mac_addr, ETHERADDRL) == 0)
3216 			return (slot);
3217 	}
3218 
3219 	return (-1);
3220 }
3221 
3222 /*
3223  * Restore the HW state to match the SW state during restart.
3224  */
3225 static int
3226 ixgbe_init_vlan(ixgbe_t *ixgbe)
3227 {
3228 	/*
3229 	 * The device is starting for the first time; there is nothing
3230 	 * to do.
3231 	 */
3232 	if (!ixgbe->vlft_init) {
3233 		ixgbe->vlft_init = B_TRUE;
3234 		return (IXGBE_SUCCESS);
3235 	}
3236 
3237 	for (uint_t i = 0; i < ixgbe->num_rx_groups; i++) {
3238 		int			ret;
3239 		boolean_t		vlvf_bypass;
3240 		ixgbe_rx_group_t	*rxg = &ixgbe->rx_groups[i];
3241 		struct ixgbe_hw		*hw = &ixgbe->hw;
3242 
3243 		if (rxg->aupe) {
3244 			uint32_t vml2flt;
3245 
3246 			vml2flt = IXGBE_READ_REG(hw, IXGBE_VMOLR(rxg->index));
3247 			vml2flt |= IXGBE_VMOLR_AUPE;
3248 			IXGBE_WRITE_REG(hw, IXGBE_VMOLR(rxg->index), vml2flt);
3249 		}
3250 
3251 		vlvf_bypass = (rxg->index == ixgbe->rx_def_group);
3252 		for (ixgbe_vlan_t *vlp = list_head(&rxg->vlans); vlp != NULL;
3253 		    vlp = list_next(&rxg->vlans, vlp)) {
3254 			ret = ixgbe_set_vfta(hw, vlp->ixvl_vid, rxg->index,
3255 			    B_TRUE, vlvf_bypass);
3256 
3257 			if (ret != IXGBE_SUCCESS) {
3258 				ixgbe_error(ixgbe, "Failed to program VFTA"
3259 				    " for group %u, VID: %u, ret: %d.",
3260 				    rxg->index, vlp->ixvl_vid, ret);
3261 				return (IXGBE_FAILURE);
3262 			}
3263 		}
3264 	}
3265 
3266 	return (IXGBE_SUCCESS);
3267 }
3268 
3269 /*
3270  * ixgbe_multicst_add - Add a multicst address.
3271  */
3272 int
3273 ixgbe_multicst_add(ixgbe_t *ixgbe, const uint8_t *multiaddr)
3274 {
3275 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3276 
3277 	if ((multiaddr[0] & 01) == 0) {
3278 		return (EINVAL);
3279 	}
3280 
3281 	if (ixgbe->mcast_count >= MAX_NUM_MULTICAST_ADDRESSES) {
3282 		return (ENOENT);
3283 	}
3284 
3285 	bcopy(multiaddr,
3286 	    &ixgbe->mcast_table[ixgbe->mcast_count], ETHERADDRL);
3287 	ixgbe->mcast_count++;
3288 
3289 	/*
3290 	 * Update the multicast table in the hardware
3291 	 */
3292 	ixgbe_setup_multicst(ixgbe);
3293 
3294 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
3295 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3296 		return (EIO);
3297 	}
3298 
3299 	return (0);
3300 }
3301 
3302 /*
3303  * ixgbe_multicst_remove - Remove a multicst address.
3304  */
3305 int
3306 ixgbe_multicst_remove(ixgbe_t *ixgbe, const uint8_t *multiaddr)
3307 {
3308 	int i;
3309 
3310 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3311 
3312 	for (i = 0; i < ixgbe->mcast_count; i++) {
3313 		if (bcmp(multiaddr, &ixgbe->mcast_table[i],
3314 		    ETHERADDRL) == 0) {
3315 			for (i++; i < ixgbe->mcast_count; i++) {
3316 				ixgbe->mcast_table[i - 1] =
3317 				    ixgbe->mcast_table[i];
3318 			}
3319 			ixgbe->mcast_count--;
3320 			break;
3321 		}
3322 	}
3323 
3324 	/*
3325 	 * Update the multicast table in the hardware
3326 	 */
3327 	ixgbe_setup_multicst(ixgbe);
3328 
3329 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
3330 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3331 		return (EIO);
3332 	}
3333 
3334 	return (0);
3335 }
3336 
3337 /*
3338  * ixgbe_setup_multicast - Setup multicast data structures.
3339  *
3340  * This routine initializes all of the multicast related structures
3341  * and save them in the hardware registers.
3342  */
3343 static void
3344 ixgbe_setup_multicst(ixgbe_t *ixgbe)
3345 {
3346 	uint8_t *mc_addr_list;
3347 	uint32_t mc_addr_count;
3348 	struct ixgbe_hw *hw = &ixgbe->hw;
3349 
3350 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3351 
3352 	ASSERT(ixgbe->mcast_count <= MAX_NUM_MULTICAST_ADDRESSES);
3353 
3354 	mc_addr_list = (uint8_t *)ixgbe->mcast_table;
3355 	mc_addr_count = ixgbe->mcast_count;
3356 
3357 	/*
3358 	 * Update the multicast addresses to the MTA registers
3359 	 */
3360 	(void) ixgbe_update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
3361 	    ixgbe_mc_table_itr, TRUE);
3362 }
3363 
3364 /*
3365  * ixgbe_setup_vmdq_rss_conf - Configure vmdq and rss (number and mode).
3366  *
3367  * Configure the rx classification mode (vmdq & rss) and vmdq & rss numbers.
3368  * Different chipsets may have different allowed configuration of vmdq and rss.
3369  */
3370 static void
3371 ixgbe_setup_vmdq_rss_conf(ixgbe_t *ixgbe)
3372 {
3373 	struct ixgbe_hw *hw = &ixgbe->hw;
3374 	uint32_t ring_per_group;
3375 
3376 	switch (hw->mac.type) {
3377 	case ixgbe_mac_82598EB:
3378 		/*
3379 		 * 82598 supports the following combination:
3380 		 * vmdq no. x rss no.
3381 		 * [5..16]  x 1
3382 		 * [1..4]   x [1..16]
3383 		 * However 8 rss queue per pool (vmdq) is sufficient for
3384 		 * most cases.
3385 		 */
3386 		ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
3387 		if (ixgbe->num_rx_groups > 4) {
3388 			ixgbe->num_rx_rings = ixgbe->num_rx_groups;
3389 		} else {
3390 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3391 			    min(8, ring_per_group);
3392 		}
3393 
3394 		break;
3395 
3396 	case ixgbe_mac_82599EB:
3397 	case ixgbe_mac_X540:
3398 	case ixgbe_mac_X550:
3399 	case ixgbe_mac_X550EM_x:
3400 	case ixgbe_mac_X550EM_a:
3401 		/*
3402 		 * 82599 supports the following combination:
3403 		 * vmdq no. x rss no.
3404 		 * [33..64] x [1..2]
3405 		 * [2..32]  x [1..4]
3406 		 * 1 x [1..16]
3407 		 * However 8 rss queue per pool (vmdq) is sufficient for
3408 		 * most cases.
3409 		 *
3410 		 * For now, treat X540 and X550 like the 82599.
3411 		 */
3412 		ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
3413 		if (ixgbe->num_rx_groups == 1) {
3414 			ixgbe->num_rx_rings = min(8, ring_per_group);
3415 		} else if (ixgbe->num_rx_groups <= 32) {
3416 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3417 			    min(4, ring_per_group);
3418 		} else if (ixgbe->num_rx_groups <= 64) {
3419 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3420 			    min(2, ring_per_group);
3421 		}
3422 		break;
3423 
3424 	default:
3425 		break;
3426 	}
3427 
3428 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
3429 
3430 	if (ixgbe->num_rx_groups == 1 && ring_per_group == 1) {
3431 		ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
3432 	} else if (ixgbe->num_rx_groups != 1 && ring_per_group == 1) {
3433 		ixgbe->classify_mode = IXGBE_CLASSIFY_VMDQ;
3434 	} else if (ixgbe->num_rx_groups != 1 && ring_per_group != 1) {
3435 		ixgbe->classify_mode = IXGBE_CLASSIFY_VMDQ_RSS;
3436 	} else {
3437 		ixgbe->classify_mode = IXGBE_CLASSIFY_RSS;
3438 	}
3439 
3440 	IXGBE_DEBUGLOG_2(ixgbe, "rx group number:%d, rx ring number:%d",
3441 	    ixgbe->num_rx_groups, ixgbe->num_rx_rings);
3442 }
3443 
3444 /*
3445  * ixgbe_get_conf - Get driver configurations set in driver.conf.
3446  *
3447  * This routine gets user-configured values out of the configuration
3448  * file ixgbe.conf.
3449  *
3450  * For each configurable value, there is a minimum, a maximum, and a
3451  * default.
3452  * If user does not configure a value, use the default.
3453  * If user configures below the minimum, use the minumum.
3454  * If user configures above the maximum, use the maxumum.
3455  */
3456 static void
3457 ixgbe_get_conf(ixgbe_t *ixgbe)
3458 {
3459 	struct ixgbe_hw *hw = &ixgbe->hw;
3460 	uint32_t flow_control;
3461 
3462 	/*
3463 	 * ixgbe driver supports the following user configurations:
3464 	 *
3465 	 * Jumbo frame configuration:
3466 	 *    default_mtu
3467 	 *
3468 	 * Ethernet flow control configuration:
3469 	 *    flow_control
3470 	 *
3471 	 * Multiple rings configurations:
3472 	 *    tx_queue_number
3473 	 *    tx_ring_size
3474 	 *    rx_queue_number
3475 	 *    rx_ring_size
3476 	 *
3477 	 * Call ixgbe_get_prop() to get the value for a specific
3478 	 * configuration parameter.
3479 	 */
3480 
3481 	/*
3482 	 * Jumbo frame configuration - max_frame_size controls host buffer
3483 	 * allocation, so includes MTU, ethernet header, vlan tag and
3484 	 * frame check sequence.
3485 	 */
3486 	ixgbe->default_mtu = ixgbe_get_prop(ixgbe, PROP_DEFAULT_MTU,
3487 	    MIN_MTU, ixgbe->capab->max_mtu, DEFAULT_MTU);
3488 
3489 	ixgbe->max_frame_size = ixgbe->default_mtu +
3490 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
3491 
3492 	/*
3493 	 * Ethernet flow control configuration
3494 	 */
3495 	flow_control = ixgbe_get_prop(ixgbe, PROP_FLOW_CONTROL,
3496 	    ixgbe_fc_none, 3, ixgbe_fc_none);
3497 	if (flow_control == 3)
3498 		flow_control = ixgbe_fc_default;
3499 
3500 	/*
3501 	 * fc.requested mode is what the user requests.  After autoneg,
3502 	 * fc.current_mode will be the flow_control mode that was negotiated.
3503 	 */
3504 	hw->fc.requested_mode = flow_control;
3505 
3506 	/*
3507 	 * Multiple rings configurations
3508 	 */
3509 	ixgbe->num_tx_rings = ixgbe_get_prop(ixgbe, PROP_TX_QUEUE_NUM,
3510 	    ixgbe->capab->min_tx_que_num,
3511 	    ixgbe->capab->max_tx_que_num,
3512 	    ixgbe->capab->def_tx_que_num);
3513 	ixgbe->tx_ring_size = ixgbe_get_prop(ixgbe, PROP_TX_RING_SIZE,
3514 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
3515 
3516 	ixgbe->num_rx_rings = ixgbe_get_prop(ixgbe, PROP_RX_QUEUE_NUM,
3517 	    ixgbe->capab->min_rx_que_num,
3518 	    ixgbe->capab->max_rx_que_num,
3519 	    ixgbe->capab->def_rx_que_num);
3520 	ixgbe->rx_ring_size = ixgbe_get_prop(ixgbe, PROP_RX_RING_SIZE,
3521 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
3522 
3523 	/*
3524 	 * Multiple groups configuration
3525 	 */
3526 	ixgbe->num_rx_groups = ixgbe_get_prop(ixgbe, PROP_RX_GROUP_NUM,
3527 	    ixgbe->capab->min_rx_grp_num, ixgbe->capab->max_rx_grp_num,
3528 	    ixgbe->capab->def_rx_grp_num);
3529 
3530 	ixgbe->mr_enable = ixgbe_get_prop(ixgbe, PROP_MR_ENABLE,
3531 	    0, 1, DEFAULT_MR_ENABLE);
3532 
3533 	if (ixgbe->mr_enable == B_FALSE) {
3534 		ixgbe->num_tx_rings = 1;
3535 		ixgbe->num_rx_rings = 1;
3536 		ixgbe->num_rx_groups = 1;
3537 		ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
3538 	} else {
3539 		ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3540 		    max(ixgbe->num_rx_rings / ixgbe->num_rx_groups, 1);
3541 		/*
3542 		 * The combination of num_rx_rings and num_rx_groups
3543 		 * may be not supported by h/w. We need to adjust
3544 		 * them to appropriate values.
3545 		 */
3546 		ixgbe_setup_vmdq_rss_conf(ixgbe);
3547 	}
3548 
3549 	/*
3550 	 * Tunable used to force an interrupt type. The only use is
3551 	 * for testing of the lesser interrupt types.
3552 	 * 0 = don't force interrupt type
3553 	 * 1 = force interrupt type MSI-X
3554 	 * 2 = force interrupt type MSI
3555 	 * 3 = force interrupt type Legacy
3556 	 */
3557 	ixgbe->intr_force = ixgbe_get_prop(ixgbe, PROP_INTR_FORCE,
3558 	    IXGBE_INTR_NONE, IXGBE_INTR_LEGACY, IXGBE_INTR_NONE);
3559 
3560 	ixgbe->tx_hcksum_enable = ixgbe_get_prop(ixgbe, PROP_TX_HCKSUM_ENABLE,
3561 	    0, 1, DEFAULT_TX_HCKSUM_ENABLE);
3562 	ixgbe->rx_hcksum_enable = ixgbe_get_prop(ixgbe, PROP_RX_HCKSUM_ENABLE,
3563 	    0, 1, DEFAULT_RX_HCKSUM_ENABLE);
3564 	ixgbe->lso_enable = ixgbe_get_prop(ixgbe, PROP_LSO_ENABLE,
3565 	    0, 1, DEFAULT_LSO_ENABLE);
3566 	ixgbe->lro_enable = ixgbe_get_prop(ixgbe, PROP_LRO_ENABLE,
3567 	    0, 1, DEFAULT_LRO_ENABLE);
3568 	ixgbe->tx_head_wb_enable = ixgbe_get_prop(ixgbe, PROP_TX_HEAD_WB_ENABLE,
3569 	    0, 1, DEFAULT_TX_HEAD_WB_ENABLE);
3570 	ixgbe->relax_order_enable = ixgbe_get_prop(ixgbe,
3571 	    PROP_RELAX_ORDER_ENABLE, 0, 1, DEFAULT_RELAX_ORDER_ENABLE);
3572 
3573 	/* Head Write Back not recommended for 82599, X540 and X550 */
3574 	if (hw->mac.type == ixgbe_mac_82599EB ||
3575 	    hw->mac.type == ixgbe_mac_X540 ||
3576 	    hw->mac.type == ixgbe_mac_X550 ||
3577 	    hw->mac.type == ixgbe_mac_X550EM_x ||
3578 	    hw->mac.type == ixgbe_mac_X550EM_a) {
3579 		ixgbe->tx_head_wb_enable = B_FALSE;
3580 	}
3581 
3582 	/*
3583 	 * ixgbe LSO needs the tx h/w checksum support.
3584 	 * LSO will be disabled if tx h/w checksum is not
3585 	 * enabled.
3586 	 */
3587 	if (ixgbe->tx_hcksum_enable == B_FALSE) {
3588 		ixgbe->lso_enable = B_FALSE;
3589 	}
3590 
3591 	/*
3592 	 * ixgbe LRO needs the rx h/w checksum support.
3593 	 * LRO will be disabled if rx h/w checksum is not
3594 	 * enabled.
3595 	 */
3596 	if (ixgbe->rx_hcksum_enable == B_FALSE) {
3597 		ixgbe->lro_enable = B_FALSE;
3598 	}
3599 
3600 	/*
3601 	 * ixgbe LRO only supported by 82599, X540 and X550
3602 	 */
3603 	if (hw->mac.type == ixgbe_mac_82598EB) {
3604 		ixgbe->lro_enable = B_FALSE;
3605 	}
3606 	ixgbe->tx_copy_thresh = ixgbe_get_prop(ixgbe, PROP_TX_COPY_THRESHOLD,
3607 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
3608 	    DEFAULT_TX_COPY_THRESHOLD);
3609 	ixgbe->tx_recycle_thresh = ixgbe_get_prop(ixgbe,
3610 	    PROP_TX_RECYCLE_THRESHOLD, MIN_TX_RECYCLE_THRESHOLD,
3611 	    MAX_TX_RECYCLE_THRESHOLD, DEFAULT_TX_RECYCLE_THRESHOLD);
3612 	ixgbe->tx_overload_thresh = ixgbe_get_prop(ixgbe,
3613 	    PROP_TX_OVERLOAD_THRESHOLD, MIN_TX_OVERLOAD_THRESHOLD,
3614 	    MAX_TX_OVERLOAD_THRESHOLD, DEFAULT_TX_OVERLOAD_THRESHOLD);
3615 	ixgbe->tx_resched_thresh = ixgbe_get_prop(ixgbe,
3616 	    PROP_TX_RESCHED_THRESHOLD, MIN_TX_RESCHED_THRESHOLD,
3617 	    MAX_TX_RESCHED_THRESHOLD, DEFAULT_TX_RESCHED_THRESHOLD);
3618 
3619 	ixgbe->rx_copy_thresh = ixgbe_get_prop(ixgbe, PROP_RX_COPY_THRESHOLD,
3620 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
3621 	    DEFAULT_RX_COPY_THRESHOLD);
3622 	ixgbe->rx_limit_per_intr = ixgbe_get_prop(ixgbe, PROP_RX_LIMIT_PER_INTR,
3623 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
3624 	    DEFAULT_RX_LIMIT_PER_INTR);
3625 
3626 	ixgbe->intr_throttling[0] = ixgbe_get_prop(ixgbe, PROP_INTR_THROTTLING,
3627 	    ixgbe->capab->min_intr_throttle,
3628 	    ixgbe->capab->max_intr_throttle,
3629 	    ixgbe->capab->def_intr_throttle);
3630 	/*
3631 	 * 82599, X540 and X550 require the interrupt throttling rate is
3632 	 * a multiple of 8. This is enforced by the register definiton.
3633 	 */
3634 	if (hw->mac.type == ixgbe_mac_82599EB ||
3635 	    hw->mac.type == ixgbe_mac_X540 ||
3636 	    hw->mac.type == ixgbe_mac_X550 ||
3637 	    hw->mac.type == ixgbe_mac_X550EM_x ||
3638 	    hw->mac.type == ixgbe_mac_X550EM_a)
3639 		ixgbe->intr_throttling[0] = ixgbe->intr_throttling[0] & 0xFF8;
3640 
3641 	hw->allow_unsupported_sfp = ixgbe_get_prop(ixgbe,
3642 	    PROP_ALLOW_UNSUPPORTED_SFP, 0, 1, DEFAULT_ALLOW_UNSUPPORTED_SFP);
3643 }
3644 
3645 static void
3646 ixgbe_init_params(ixgbe_t *ixgbe)
3647 {
3648 	struct ixgbe_hw *hw = &ixgbe->hw;
3649 	ixgbe_link_speed speeds_supported = 0;
3650 	boolean_t negotiate;
3651 
3652 	/*
3653 	 * Get a list of speeds the adapter supports. If the hw struct hasn't
3654 	 * been populated with this information yet, retrieve it from the
3655 	 * adapter and save it to our own variable.
3656 	 *
3657 	 * On certain adapters, such as ones which use SFPs, the contents of
3658 	 * hw->phy.speeds_supported (and hw->phy.autoneg_advertised) are not
3659 	 * updated, so we must rely on calling ixgbe_get_link_capabilities()
3660 	 * in order to ascertain the speeds which we are capable of supporting,
3661 	 * and in the case of SFP-equipped adapters, which speed we are
3662 	 * advertising. If ixgbe_get_link_capabilities() fails for some reason,
3663 	 * we'll go with a default list of speeds as a last resort.
3664 	 */
3665 	speeds_supported = hw->phy.speeds_supported;
3666 
3667 	if (speeds_supported == 0) {
3668 		if (ixgbe_get_link_capabilities(hw, &speeds_supported,
3669 		    &negotiate) != IXGBE_SUCCESS) {
3670 			if (hw->mac.type == ixgbe_mac_82598EB) {
3671 				speeds_supported =
3672 				    IXGBE_LINK_SPEED_82598_AUTONEG;
3673 			} else {
3674 				speeds_supported =
3675 				    IXGBE_LINK_SPEED_82599_AUTONEG;
3676 			}
3677 		}
3678 	}
3679 	ixgbe->speeds_supported = speeds_supported;
3680 
3681 	/*
3682 	 * By default, all supported speeds are enabled and advertised.
3683 	 */
3684 	if (speeds_supported & IXGBE_LINK_SPEED_10GB_FULL) {
3685 		ixgbe->param_en_10000fdx_cap = 1;
3686 		ixgbe->param_adv_10000fdx_cap = 1;
3687 	} else {
3688 		ixgbe->param_en_10000fdx_cap = 0;
3689 		ixgbe->param_adv_10000fdx_cap = 0;
3690 	}
3691 
3692 	if (speeds_supported & IXGBE_LINK_SPEED_5GB_FULL) {
3693 		ixgbe->param_en_5000fdx_cap = 1;
3694 		ixgbe->param_adv_5000fdx_cap = 1;
3695 	} else {
3696 		ixgbe->param_en_5000fdx_cap = 0;
3697 		ixgbe->param_adv_5000fdx_cap = 0;
3698 	}
3699 
3700 	if (speeds_supported & IXGBE_LINK_SPEED_2_5GB_FULL) {
3701 		ixgbe->param_en_2500fdx_cap = 1;
3702 		ixgbe->param_adv_2500fdx_cap = 1;
3703 	} else {
3704 		ixgbe->param_en_2500fdx_cap = 0;
3705 		ixgbe->param_adv_2500fdx_cap = 0;
3706 	}
3707 
3708 	if (speeds_supported & IXGBE_LINK_SPEED_1GB_FULL) {
3709 		ixgbe->param_en_1000fdx_cap = 1;
3710 		ixgbe->param_adv_1000fdx_cap = 1;
3711 	} else {
3712 		ixgbe->param_en_1000fdx_cap = 0;
3713 		ixgbe->param_adv_1000fdx_cap = 0;
3714 	}
3715 
3716 	if (speeds_supported & IXGBE_LINK_SPEED_100_FULL) {
3717 		ixgbe->param_en_100fdx_cap = 1;
3718 		ixgbe->param_adv_100fdx_cap = 1;
3719 	} else {
3720 		ixgbe->param_en_100fdx_cap = 0;
3721 		ixgbe->param_adv_100fdx_cap = 0;
3722 	}
3723 
3724 	ixgbe->param_pause_cap = 1;
3725 	ixgbe->param_asym_pause_cap = 1;
3726 	ixgbe->param_rem_fault = 0;
3727 
3728 	ixgbe->param_adv_autoneg_cap = 1;
3729 	ixgbe->param_adv_pause_cap = 1;
3730 	ixgbe->param_adv_asym_pause_cap = 1;
3731 	ixgbe->param_adv_rem_fault = 0;
3732 
3733 	ixgbe->param_lp_10000fdx_cap = 0;
3734 	ixgbe->param_lp_5000fdx_cap = 0;
3735 	ixgbe->param_lp_2500fdx_cap = 0;
3736 	ixgbe->param_lp_1000fdx_cap = 0;
3737 	ixgbe->param_lp_100fdx_cap = 0;
3738 	ixgbe->param_lp_autoneg_cap = 0;
3739 	ixgbe->param_lp_pause_cap = 0;
3740 	ixgbe->param_lp_asym_pause_cap = 0;
3741 	ixgbe->param_lp_rem_fault = 0;
3742 }
3743 
3744 /*
3745  * ixgbe_get_prop - Get a property value out of the configuration file
3746  * ixgbe.conf.
3747  *
3748  * Caller provides the name of the property, a default value, a minimum
3749  * value, and a maximum value.
3750  *
3751  * Return configured value of the property, with default, minimum and
3752  * maximum properly applied.
3753  */
3754 static int
3755 ixgbe_get_prop(ixgbe_t *ixgbe,
3756     char *propname,	/* name of the property */
3757     int minval,		/* minimum acceptable value */
3758     int maxval,		/* maximim acceptable value */
3759     int defval)		/* default value */
3760 {
3761 	int value;
3762 
3763 	/*
3764 	 * Call ddi_prop_get_int() to read the conf settings
3765 	 */
3766 	value = ddi_prop_get_int(DDI_DEV_T_ANY, ixgbe->dip,
3767 	    DDI_PROP_DONTPASS, propname, defval);
3768 	if (value > maxval)
3769 		value = maxval;
3770 
3771 	if (value < minval)
3772 		value = minval;
3773 
3774 	return (value);
3775 }
3776 
3777 /*
3778  * ixgbe_driver_setup_link - Using the link properties to setup the link.
3779  */
3780 int
3781 ixgbe_driver_setup_link(ixgbe_t *ixgbe, boolean_t setup_hw)
3782 {
3783 	struct ixgbe_hw *hw = &ixgbe->hw;
3784 	ixgbe_link_speed advertised = 0;
3785 
3786 	/*
3787 	 * Assemble a list of enabled speeds to auto-negotiate with.
3788 	 */
3789 	if (ixgbe->param_en_10000fdx_cap == 1)
3790 		advertised |= IXGBE_LINK_SPEED_10GB_FULL;
3791 
3792 	if (ixgbe->param_en_5000fdx_cap == 1)
3793 		advertised |= IXGBE_LINK_SPEED_5GB_FULL;
3794 
3795 	if (ixgbe->param_en_2500fdx_cap == 1)
3796 		advertised |= IXGBE_LINK_SPEED_2_5GB_FULL;
3797 
3798 	if (ixgbe->param_en_1000fdx_cap == 1)
3799 		advertised |= IXGBE_LINK_SPEED_1GB_FULL;
3800 
3801 	if (ixgbe->param_en_100fdx_cap == 1)
3802 		advertised |= IXGBE_LINK_SPEED_100_FULL;
3803 
3804 	/*
3805 	 * As a last resort, autoneg with a default list of speeds.
3806 	 */
3807 	if (ixgbe->param_adv_autoneg_cap == 1 && advertised == 0) {
3808 		ixgbe_notice(ixgbe, "Invalid link settings. Setting link "
3809 		    "to autonegotiate with full capabilities.");
3810 
3811 		if (hw->mac.type == ixgbe_mac_82598EB)
3812 			advertised = IXGBE_LINK_SPEED_82598_AUTONEG;
3813 		else
3814 			advertised = IXGBE_LINK_SPEED_82599_AUTONEG;
3815 	}
3816 
3817 	if (setup_hw) {
3818 		if (ixgbe_setup_link(&ixgbe->hw, advertised,
3819 		    ixgbe->param_adv_autoneg_cap) != IXGBE_SUCCESS) {
3820 			ixgbe_notice(ixgbe, "Setup link failed on this "
3821 			    "device.");
3822 			return (IXGBE_FAILURE);
3823 		}
3824 	}
3825 
3826 	return (IXGBE_SUCCESS);
3827 }
3828 
3829 /*
3830  * ixgbe_driver_link_check - Link status processing.
3831  *
3832  * This function can be called in both kernel context and interrupt context
3833  */
3834 static void
3835 ixgbe_driver_link_check(ixgbe_t *ixgbe)
3836 {
3837 	struct ixgbe_hw *hw = &ixgbe->hw;
3838 	ixgbe_link_speed speed = IXGBE_LINK_SPEED_UNKNOWN;
3839 	boolean_t link_up = B_FALSE;
3840 	boolean_t link_changed = B_FALSE;
3841 
3842 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3843 
3844 	(void) ixgbe_check_link(hw, &speed, &link_up, B_FALSE);
3845 	if (link_up) {
3846 		ixgbe->link_check_complete = B_TRUE;
3847 
3848 		/* Link is up, enable flow control settings */
3849 		(void) ixgbe_fc_enable(hw);
3850 
3851 		/*
3852 		 * The Link is up, check whether it was marked as down earlier
3853 		 */
3854 		if (ixgbe->link_state != LINK_STATE_UP) {
3855 			switch (speed) {
3856 			case IXGBE_LINK_SPEED_10GB_FULL:
3857 				ixgbe->link_speed = SPEED_10GB;
3858 				break;
3859 			case IXGBE_LINK_SPEED_5GB_FULL:
3860 				ixgbe->link_speed = SPEED_5GB;
3861 				break;
3862 			case IXGBE_LINK_SPEED_2_5GB_FULL:
3863 				ixgbe->link_speed = SPEED_2_5GB;
3864 				break;
3865 			case IXGBE_LINK_SPEED_1GB_FULL:
3866 				ixgbe->link_speed = SPEED_1GB;
3867 				break;
3868 			case IXGBE_LINK_SPEED_100_FULL:
3869 				ixgbe->link_speed = SPEED_100;
3870 			}
3871 			ixgbe->link_duplex = LINK_DUPLEX_FULL;
3872 			ixgbe->link_state = LINK_STATE_UP;
3873 			link_changed = B_TRUE;
3874 		}
3875 	} else {
3876 		if (ixgbe->link_check_complete == B_TRUE ||
3877 		    (ixgbe->link_check_complete == B_FALSE &&
3878 		    gethrtime() >= ixgbe->link_check_hrtime)) {
3879 			/*
3880 			 * The link is really down
3881 			 */
3882 			ixgbe->link_check_complete = B_TRUE;
3883 
3884 			if (ixgbe->link_state != LINK_STATE_DOWN) {
3885 				ixgbe->link_speed = 0;
3886 				ixgbe->link_duplex = LINK_DUPLEX_UNKNOWN;
3887 				ixgbe->link_state = LINK_STATE_DOWN;
3888 				link_changed = B_TRUE;
3889 			}
3890 		}
3891 	}
3892 
3893 	/*
3894 	 * If we are in an interrupt context, need to re-enable the
3895 	 * interrupt, which was automasked
3896 	 */
3897 	if (servicing_interrupt() != 0) {
3898 		ixgbe->eims |= IXGBE_EICR_LSC;
3899 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
3900 	}
3901 
3902 	if (link_changed) {
3903 		mac_link_update(ixgbe->mac_hdl, ixgbe->link_state);
3904 	}
3905 }
3906 
3907 /*
3908  * ixgbe_sfp_check - sfp module processing done in taskq only for 82599.
3909  */
3910 static void
3911 ixgbe_sfp_check(void *arg)
3912 {
3913 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3914 	uint32_t eicr = ixgbe->eicr;
3915 	struct ixgbe_hw *hw = &ixgbe->hw;
3916 
3917 	mutex_enter(&ixgbe->gen_lock);
3918 	(void) hw->phy.ops.identify_sfp(hw);
3919 	if (eicr & IXGBE_EICR_GPI_SDP1_BY_MAC(hw)) {
3920 		/* clear the interrupt */
3921 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1_BY_MAC(hw));
3922 
3923 		/* if link up, do multispeed fiber setup */
3924 		(void) ixgbe_setup_link(hw, IXGBE_LINK_SPEED_82599_AUTONEG,
3925 		    B_TRUE);
3926 		ixgbe_driver_link_check(ixgbe);
3927 		ixgbe_get_hw_state(ixgbe);
3928 	} else if (eicr & IXGBE_EICR_GPI_SDP2_BY_MAC(hw)) {
3929 		/* clear the interrupt */
3930 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2_BY_MAC(hw));
3931 
3932 		/* if link up, do sfp module setup */
3933 		(void) hw->mac.ops.setup_sfp(hw);
3934 
3935 		/* do multispeed fiber setup */
3936 		(void) ixgbe_setup_link(hw, IXGBE_LINK_SPEED_82599_AUTONEG,
3937 		    B_TRUE);
3938 		ixgbe_driver_link_check(ixgbe);
3939 		ixgbe_get_hw_state(ixgbe);
3940 	}
3941 	mutex_exit(&ixgbe->gen_lock);
3942 
3943 	/*
3944 	 * We need to fully re-check the link later.
3945 	 */
3946 	ixgbe->link_check_complete = B_FALSE;
3947 	ixgbe->link_check_hrtime = gethrtime() +
3948 	    (IXGBE_LINK_UP_TIME * 100000000ULL);
3949 }
3950 
3951 /*
3952  * ixgbe_overtemp_check - overtemp module processing done in taskq
3953  *
3954  * This routine will only be called on adapters with temperature sensor.
3955  * The indication of over-temperature can be either SDP0 interrupt or the link
3956  * status change interrupt.
3957  */
3958 static void
3959 ixgbe_overtemp_check(void *arg)
3960 {
3961 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3962 	struct ixgbe_hw *hw = &ixgbe->hw;
3963 	uint32_t eicr = ixgbe->eicr;
3964 	ixgbe_link_speed speed;
3965 	boolean_t link_up;
3966 
3967 	mutex_enter(&ixgbe->gen_lock);
3968 
3969 	/* make sure we know current state of link */
3970 	(void) ixgbe_check_link(hw, &speed, &link_up, B_FALSE);
3971 
3972 	/* check over-temp condition */
3973 	if (((eicr & IXGBE_EICR_GPI_SDP0_BY_MAC(hw)) && (!link_up)) ||
3974 	    (eicr & IXGBE_EICR_LSC)) {
3975 		if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP) {
3976 			atomic_or_32(&ixgbe->ixgbe_state, IXGBE_OVERTEMP);
3977 
3978 			/*
3979 			 * Disable the adapter interrupts
3980 			 */
3981 			ixgbe_disable_adapter_interrupts(ixgbe);
3982 
3983 			/*
3984 			 * Disable Rx/Tx units
3985 			 */
3986 			(void) ixgbe_stop_adapter(hw);
3987 
3988 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
3989 			ixgbe_error(ixgbe,
3990 			    "Problem: Network adapter has been stopped "
3991 			    "because it has overheated");
3992 			ixgbe_error(ixgbe,
3993 			    "Action: Restart the computer. "
3994 			    "If the problem persists, power off the system "
3995 			    "and replace the adapter");
3996 		}
3997 	}
3998 
3999 	/* write to clear the interrupt */
4000 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
4001 
4002 	mutex_exit(&ixgbe->gen_lock);
4003 }
4004 
4005 /*
4006  * ixgbe_phy_check - taskq to process interrupts from an external PHY
4007  *
4008  * This routine will only be called on adapters with external PHYs
4009  * (such as X550) that may be trying to raise our attention to some event.
4010  * Currently, this is limited to claiming PHY overtemperature and link status
4011  * change (LSC) events, however this may expand to include other things in
4012  * future adapters.
4013  */
4014 static void
4015 ixgbe_phy_check(void *arg)
4016 {
4017 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
4018 	struct ixgbe_hw *hw = &ixgbe->hw;
4019 	int rv;
4020 
4021 	mutex_enter(&ixgbe->gen_lock);
4022 
4023 	/*
4024 	 * X550 baseT PHY overtemp and LSC events are handled here.
4025 	 *
4026 	 * If an overtemp event occurs, it will be reflected in the
4027 	 * return value of phy.ops.handle_lasi() and the common code will
4028 	 * automatically power off the baseT PHY. This is our cue to trigger
4029 	 * an FMA event.
4030 	 *
4031 	 * If a link status change event occurs, phy.ops.handle_lasi() will
4032 	 * automatically initiate a link setup between the integrated KR PHY
4033 	 * and the external X557 PHY to ensure that the link speed between
4034 	 * them matches the link speed of the baseT link.
4035 	 */
4036 	rv = ixgbe_handle_lasi(hw);
4037 
4038 	if (rv == IXGBE_ERR_OVERTEMP) {
4039 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_OVERTEMP);
4040 
4041 		/*
4042 		 * Disable the adapter interrupts
4043 		 */
4044 		ixgbe_disable_adapter_interrupts(ixgbe);
4045 
4046 		/*
4047 		 * Disable Rx/Tx units
4048 		 */
4049 		(void) ixgbe_stop_adapter(hw);
4050 
4051 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
4052 		ixgbe_error(ixgbe,
4053 		    "Problem: Network adapter has been stopped due to a "
4054 		    "overtemperature event being detected.");
4055 		ixgbe_error(ixgbe,
4056 		    "Action: Shut down or restart the computer. If the issue "
4057 		    "persists, please take action in accordance with the "
4058 		    "recommendations from your system vendor.");
4059 	}
4060 
4061 	mutex_exit(&ixgbe->gen_lock);
4062 }
4063 
4064 /*
4065  * ixgbe_link_timer - timer for link status detection
4066  */
4067 static void
4068 ixgbe_link_timer(void *arg)
4069 {
4070 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
4071 
4072 	mutex_enter(&ixgbe->gen_lock);
4073 	ixgbe_driver_link_check(ixgbe);
4074 	mutex_exit(&ixgbe->gen_lock);
4075 }
4076 
4077 /*
4078  * ixgbe_local_timer - Driver watchdog function.
4079  *
4080  * This function will handle the transmit stall check and other routines.
4081  */
4082 static void
4083 ixgbe_local_timer(void *arg)
4084 {
4085 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
4086 
4087 	if (ixgbe->ixgbe_state & IXGBE_OVERTEMP)
4088 		goto out;
4089 
4090 	if (ixgbe->ixgbe_state & IXGBE_ERROR) {
4091 		ixgbe->reset_count++;
4092 		if (ixgbe_reset(ixgbe) == IXGBE_SUCCESS)
4093 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_RESTORED);
4094 		goto out;
4095 	}
4096 
4097 	if (ixgbe_stall_check(ixgbe)) {
4098 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_STALL);
4099 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4100 
4101 		ixgbe->reset_count++;
4102 		if (ixgbe_reset(ixgbe) == IXGBE_SUCCESS)
4103 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_RESTORED);
4104 	}
4105 
4106 out:
4107 	ixgbe_restart_watchdog_timer(ixgbe);
4108 }
4109 
4110 /*
4111  * ixgbe_stall_check - Check for transmit stall.
4112  *
4113  * This function checks if the adapter is stalled (in transmit).
4114  *
4115  * It is called each time the watchdog timeout is invoked.
4116  * If the transmit descriptor reclaim continuously fails,
4117  * the watchdog value will increment by 1. If the watchdog
4118  * value exceeds the threshold, the ixgbe is assumed to
4119  * have stalled and need to be reset.
4120  */
4121 static boolean_t
4122 ixgbe_stall_check(ixgbe_t *ixgbe)
4123 {
4124 	ixgbe_tx_ring_t *tx_ring;
4125 	boolean_t result;
4126 	int i;
4127 
4128 	if (ixgbe->link_state != LINK_STATE_UP)
4129 		return (B_FALSE);
4130 
4131 	/*
4132 	 * If any tx ring is stalled, we'll reset the chipset
4133 	 */
4134 	result = B_FALSE;
4135 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
4136 		tx_ring = &ixgbe->tx_rings[i];
4137 		if (tx_ring->tbd_free <= ixgbe->tx_recycle_thresh) {
4138 			tx_ring->tx_recycle(tx_ring);
4139 		}
4140 
4141 		if (tx_ring->recycle_fail > 0)
4142 			tx_ring->stall_watchdog++;
4143 		else
4144 			tx_ring->stall_watchdog = 0;
4145 
4146 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
4147 			result = B_TRUE;
4148 			break;
4149 		}
4150 	}
4151 
4152 	if (result) {
4153 		tx_ring->stall_watchdog = 0;
4154 		tx_ring->recycle_fail = 0;
4155 	}
4156 
4157 	return (result);
4158 }
4159 
4160 
4161 /*
4162  * is_valid_mac_addr - Check if the mac address is valid.
4163  */
4164 static boolean_t
4165 is_valid_mac_addr(uint8_t *mac_addr)
4166 {
4167 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
4168 	const uint8_t addr_test2[6] =
4169 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
4170 
4171 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
4172 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
4173 		return (B_FALSE);
4174 
4175 	return (B_TRUE);
4176 }
4177 
4178 static boolean_t
4179 ixgbe_find_mac_address(ixgbe_t *ixgbe)
4180 {
4181 #ifdef __sparc
4182 	struct ixgbe_hw *hw = &ixgbe->hw;
4183 	uchar_t *bytes;
4184 	struct ether_addr sysaddr;
4185 	uint_t nelts;
4186 	int err;
4187 	boolean_t found = B_FALSE;
4188 
4189 	/*
4190 	 * The "vendor's factory-set address" may already have
4191 	 * been extracted from the chip, but if the property
4192 	 * "local-mac-address" is set we use that instead.
4193 	 *
4194 	 * We check whether it looks like an array of 6
4195 	 * bytes (which it should, if OBP set it).  If we can't
4196 	 * make sense of it this way, we'll ignore it.
4197 	 */
4198 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip,
4199 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
4200 	if (err == DDI_PROP_SUCCESS) {
4201 		if (nelts == ETHERADDRL) {
4202 			while (nelts--)
4203 				hw->mac.addr[nelts] = bytes[nelts];
4204 			found = B_TRUE;
4205 		}
4206 		ddi_prop_free(bytes);
4207 	}
4208 
4209 	/*
4210 	 * Look up the OBP property "local-mac-address?". If the user has set
4211 	 * 'local-mac-address? = false', use "the system address" instead.
4212 	 */
4213 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip, 0,
4214 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
4215 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
4216 			if (localetheraddr(NULL, &sysaddr) != 0) {
4217 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
4218 				found = B_TRUE;
4219 			}
4220 		}
4221 		ddi_prop_free(bytes);
4222 	}
4223 
4224 	/*
4225 	 * Finally(!), if there's a valid "mac-address" property (created
4226 	 * if we netbooted from this interface), we must use this instead
4227 	 * of any of the above to ensure that the NFS/install server doesn't
4228 	 * get confused by the address changing as illumos takes over!
4229 	 */
4230 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip,
4231 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
4232 	if (err == DDI_PROP_SUCCESS) {
4233 		if (nelts == ETHERADDRL) {
4234 			while (nelts--)
4235 				hw->mac.addr[nelts] = bytes[nelts];
4236 			found = B_TRUE;
4237 		}
4238 		ddi_prop_free(bytes);
4239 	}
4240 
4241 	if (found) {
4242 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
4243 		return (B_TRUE);
4244 	}
4245 #else
4246 	_NOTE(ARGUNUSED(ixgbe));
4247 #endif
4248 
4249 	return (B_TRUE);
4250 }
4251 
4252 #pragma inline(ixgbe_arm_watchdog_timer)
4253 static void
4254 ixgbe_arm_watchdog_timer(ixgbe_t *ixgbe)
4255 {
4256 	/*
4257 	 * Fire a watchdog timer
4258 	 */
4259 	ixgbe->watchdog_tid =
4260 	    timeout(ixgbe_local_timer,
4261 	    (void *)ixgbe, 1 * drv_usectohz(1000000));
4262 
4263 }
4264 
4265 /*
4266  * ixgbe_enable_watchdog_timer - Enable and start the driver watchdog timer.
4267  */
4268 void
4269 ixgbe_enable_watchdog_timer(ixgbe_t *ixgbe)
4270 {
4271 	mutex_enter(&ixgbe->watchdog_lock);
4272 
4273 	if (!ixgbe->watchdog_enable) {
4274 		ixgbe->watchdog_enable = B_TRUE;
4275 		ixgbe->watchdog_start = B_TRUE;
4276 		ixgbe_arm_watchdog_timer(ixgbe);
4277 	}
4278 
4279 	mutex_exit(&ixgbe->watchdog_lock);
4280 }
4281 
4282 /*
4283  * ixgbe_disable_watchdog_timer - Disable and stop the driver watchdog timer.
4284  */
4285 void
4286 ixgbe_disable_watchdog_timer(ixgbe_t *ixgbe)
4287 {
4288 	timeout_id_t tid;
4289 
4290 	mutex_enter(&ixgbe->watchdog_lock);
4291 
4292 	ixgbe->watchdog_enable = B_FALSE;
4293 	ixgbe->watchdog_start = B_FALSE;
4294 	tid = ixgbe->watchdog_tid;
4295 	ixgbe->watchdog_tid = 0;
4296 
4297 	mutex_exit(&ixgbe->watchdog_lock);
4298 
4299 	if (tid != 0)
4300 		(void) untimeout(tid);
4301 }
4302 
4303 /*
4304  * ixgbe_start_watchdog_timer - Start the driver watchdog timer.
4305  */
4306 void
4307 ixgbe_start_watchdog_timer(ixgbe_t *ixgbe)
4308 {
4309 	mutex_enter(&ixgbe->watchdog_lock);
4310 
4311 	if (ixgbe->watchdog_enable) {
4312 		if (!ixgbe->watchdog_start) {
4313 			ixgbe->watchdog_start = B_TRUE;
4314 			ixgbe_arm_watchdog_timer(ixgbe);
4315 		}
4316 	}
4317 
4318 	mutex_exit(&ixgbe->watchdog_lock);
4319 }
4320 
4321 /*
4322  * ixgbe_restart_watchdog_timer - Restart the driver watchdog timer.
4323  */
4324 static void
4325 ixgbe_restart_watchdog_timer(ixgbe_t *ixgbe)
4326 {
4327 	mutex_enter(&ixgbe->watchdog_lock);
4328 
4329 	if (ixgbe->watchdog_start)
4330 		ixgbe_arm_watchdog_timer(ixgbe);
4331 
4332 	mutex_exit(&ixgbe->watchdog_lock);
4333 }
4334 
4335 /*
4336  * ixgbe_stop_watchdog_timer - Stop the driver watchdog timer.
4337  */
4338 void
4339 ixgbe_stop_watchdog_timer(ixgbe_t *ixgbe)
4340 {
4341 	timeout_id_t tid;
4342 
4343 	mutex_enter(&ixgbe->watchdog_lock);
4344 
4345 	ixgbe->watchdog_start = B_FALSE;
4346 	tid = ixgbe->watchdog_tid;
4347 	ixgbe->watchdog_tid = 0;
4348 
4349 	mutex_exit(&ixgbe->watchdog_lock);
4350 
4351 	if (tid != 0)
4352 		(void) untimeout(tid);
4353 }
4354 
4355 /*
4356  * ixgbe_disable_adapter_interrupts - Disable all adapter interrupts.
4357  */
4358 static void
4359 ixgbe_disable_adapter_interrupts(ixgbe_t *ixgbe)
4360 {
4361 	struct ixgbe_hw *hw = &ixgbe->hw;
4362 
4363 	/*
4364 	 * mask all interrupts off
4365 	 */
4366 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, 0xffffffff);
4367 
4368 	/*
4369 	 * for MSI-X, also disable autoclear
4370 	 */
4371 	if (ixgbe->intr_type == DDI_INTR_TYPE_MSIX) {
4372 		IXGBE_WRITE_REG(hw, IXGBE_EIAC, 0x0);
4373 	}
4374 
4375 	IXGBE_WRITE_FLUSH(hw);
4376 }
4377 
4378 /*
4379  * ixgbe_enable_adapter_interrupts - Enable all hardware interrupts.
4380  */
4381 static void
4382 ixgbe_enable_adapter_interrupts(ixgbe_t *ixgbe)
4383 {
4384 	struct ixgbe_hw *hw = &ixgbe->hw;
4385 	uint32_t eiac, eiam;
4386 	uint32_t gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
4387 
4388 	/* interrupt types to enable */
4389 	ixgbe->eims = IXGBE_EIMS_ENABLE_MASK;	/* shared code default */
4390 	ixgbe->eims &= ~IXGBE_EIMS_TCP_TIMER;	/* minus tcp timer */
4391 	ixgbe->eims |= ixgbe->capab->other_intr; /* "other" interrupt types */
4392 
4393 	/* enable automask on "other" causes that this adapter can generate */
4394 	eiam = ixgbe->capab->other_intr;
4395 
4396 	/*
4397 	 * msi-x mode
4398 	 */
4399 	if (ixgbe->intr_type == DDI_INTR_TYPE_MSIX) {
4400 		/* enable autoclear but not on bits 29:20 */
4401 		eiac = (ixgbe->eims & ~IXGBE_OTHER_INTR);
4402 
4403 		/* general purpose interrupt enable */
4404 		gpie |= (IXGBE_GPIE_MSIX_MODE
4405 		    | IXGBE_GPIE_PBA_SUPPORT
4406 		    | IXGBE_GPIE_OCD
4407 		    | IXGBE_GPIE_EIAME);
4408 	/*
4409 	 * non-msi-x mode
4410 	 */
4411 	} else {
4412 
4413 		/* disable autoclear, leave gpie at default */
4414 		eiac = 0;
4415 
4416 		/*
4417 		 * General purpose interrupt enable.
4418 		 * For 82599, X540 and X550, extended interrupt
4419 		 * automask enable only in MSI or MSI-X mode
4420 		 */
4421 		if ((hw->mac.type == ixgbe_mac_82598EB) ||
4422 		    (ixgbe->intr_type == DDI_INTR_TYPE_MSI)) {
4423 			gpie |= IXGBE_GPIE_EIAME;
4424 		}
4425 	}
4426 
4427 	/* Enable specific "other" interrupt types */
4428 	switch (hw->mac.type) {
4429 	case ixgbe_mac_82598EB:
4430 		gpie |= ixgbe->capab->other_gpie;
4431 		break;
4432 
4433 	case ixgbe_mac_82599EB:
4434 	case ixgbe_mac_X540:
4435 	case ixgbe_mac_X550:
4436 	case ixgbe_mac_X550EM_x:
4437 	case ixgbe_mac_X550EM_a:
4438 		gpie |= ixgbe->capab->other_gpie;
4439 
4440 		/* Enable RSC Delay 8us when LRO enabled  */
4441 		if (ixgbe->lro_enable) {
4442 			gpie |= (1 << IXGBE_GPIE_RSC_DELAY_SHIFT);
4443 		}
4444 		break;
4445 
4446 	default:
4447 		break;
4448 	}
4449 
4450 	/* write to interrupt control registers */
4451 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4452 	IXGBE_WRITE_REG(hw, IXGBE_EIAC, eiac);
4453 	IXGBE_WRITE_REG(hw, IXGBE_EIAM, eiam);
4454 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4455 	IXGBE_WRITE_FLUSH(hw);
4456 }
4457 
4458 /*
4459  * ixgbe_loopback_ioctl - Loopback support.
4460  */
4461 enum ioc_reply
4462 ixgbe_loopback_ioctl(ixgbe_t *ixgbe, struct iocblk *iocp, mblk_t *mp)
4463 {
4464 	lb_info_sz_t *lbsp;
4465 	lb_property_t *lbpp;
4466 	uint32_t *lbmp;
4467 	uint32_t size;
4468 	uint32_t value;
4469 
4470 	if (mp->b_cont == NULL)
4471 		return (IOC_INVAL);
4472 
4473 	switch (iocp->ioc_cmd) {
4474 	default:
4475 		return (IOC_INVAL);
4476 
4477 	case LB_GET_INFO_SIZE:
4478 		size = sizeof (lb_info_sz_t);
4479 		if (iocp->ioc_count != size)
4480 			return (IOC_INVAL);
4481 
4482 		value = sizeof (lb_normal);
4483 		value += sizeof (lb_mac);
4484 		value += sizeof (lb_external);
4485 
4486 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
4487 		*lbsp = value;
4488 		break;
4489 
4490 	case LB_GET_INFO:
4491 		value = sizeof (lb_normal);
4492 		value += sizeof (lb_mac);
4493 		value += sizeof (lb_external);
4494 
4495 		size = value;
4496 		if (iocp->ioc_count != size)
4497 			return (IOC_INVAL);
4498 
4499 		value = 0;
4500 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
4501 
4502 		lbpp[value++] = lb_normal;
4503 		lbpp[value++] = lb_mac;
4504 		lbpp[value++] = lb_external;
4505 		break;
4506 
4507 	case LB_GET_MODE:
4508 		size = sizeof (uint32_t);
4509 		if (iocp->ioc_count != size)
4510 			return (IOC_INVAL);
4511 
4512 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
4513 		*lbmp = ixgbe->loopback_mode;
4514 		break;
4515 
4516 	case LB_SET_MODE:
4517 		size = 0;
4518 		if (iocp->ioc_count != sizeof (uint32_t))
4519 			return (IOC_INVAL);
4520 
4521 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
4522 		if (!ixgbe_set_loopback_mode(ixgbe, *lbmp))
4523 			return (IOC_INVAL);
4524 		break;
4525 	}
4526 
4527 	iocp->ioc_count = size;
4528 	iocp->ioc_error = 0;
4529 
4530 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
4531 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4532 		return (IOC_INVAL);
4533 	}
4534 
4535 	return (IOC_REPLY);
4536 }
4537 
4538 /*
4539  * ixgbe_set_loopback_mode - Setup loopback based on the loopback mode.
4540  */
4541 static boolean_t
4542 ixgbe_set_loopback_mode(ixgbe_t *ixgbe, uint32_t mode)
4543 {
4544 	if (mode == ixgbe->loopback_mode)
4545 		return (B_TRUE);
4546 
4547 	ixgbe->loopback_mode = mode;
4548 
4549 	if (mode == IXGBE_LB_NONE) {
4550 		/*
4551 		 * Reset the chip
4552 		 */
4553 		(void) ixgbe_reset(ixgbe);
4554 		return (B_TRUE);
4555 	}
4556 
4557 	mutex_enter(&ixgbe->gen_lock);
4558 
4559 	switch (mode) {
4560 	default:
4561 		mutex_exit(&ixgbe->gen_lock);
4562 		return (B_FALSE);
4563 
4564 	case IXGBE_LB_EXTERNAL:
4565 		break;
4566 
4567 	case IXGBE_LB_INTERNAL_MAC:
4568 		ixgbe_set_internal_mac_loopback(ixgbe);
4569 		break;
4570 	}
4571 
4572 	mutex_exit(&ixgbe->gen_lock);
4573 
4574 	return (B_TRUE);
4575 }
4576 
4577 /*
4578  * ixgbe_set_internal_mac_loopback - Set the internal MAC loopback mode.
4579  */
4580 static void
4581 ixgbe_set_internal_mac_loopback(ixgbe_t *ixgbe)
4582 {
4583 	struct ixgbe_hw *hw;
4584 	uint32_t reg;
4585 	uint8_t atlas;
4586 
4587 	hw = &ixgbe->hw;
4588 
4589 	/*
4590 	 * Setup MAC loopback
4591 	 */
4592 	reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_HLREG0);
4593 	reg |= IXGBE_HLREG0_LPBK;
4594 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_HLREG0, reg);
4595 
4596 	reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_AUTOC);
4597 	reg &= ~IXGBE_AUTOC_LMS_MASK;
4598 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_AUTOC, reg);
4599 
4600 	/*
4601 	 * Disable Atlas Tx lanes to keep packets in loopback and not on wire
4602 	 */
4603 	switch (hw->mac.type) {
4604 	case ixgbe_mac_82598EB:
4605 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_LPBK,
4606 		    &atlas);
4607 		atlas |= IXGBE_ATLAS_PDN_TX_REG_EN;
4608 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_LPBK,
4609 		    atlas);
4610 
4611 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_10G,
4612 		    &atlas);
4613 		atlas |= IXGBE_ATLAS_PDN_TX_10G_QL_ALL;
4614 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_10G,
4615 		    atlas);
4616 
4617 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_1G,
4618 		    &atlas);
4619 		atlas |= IXGBE_ATLAS_PDN_TX_1G_QL_ALL;
4620 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_1G,
4621 		    atlas);
4622 
4623 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_AN,
4624 		    &atlas);
4625 		atlas |= IXGBE_ATLAS_PDN_TX_AN_QL_ALL;
4626 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_AN,
4627 		    atlas);
4628 		break;
4629 
4630 	case ixgbe_mac_82599EB:
4631 	case ixgbe_mac_X540:
4632 	case ixgbe_mac_X550:
4633 	case ixgbe_mac_X550EM_x:
4634 	case ixgbe_mac_X550EM_a:
4635 		reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_AUTOC);
4636 		reg |= (IXGBE_AUTOC_FLU |
4637 		    IXGBE_AUTOC_10G_KX4);
4638 		IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_AUTOC, reg);
4639 
4640 		(void) ixgbe_setup_link(&ixgbe->hw, IXGBE_LINK_SPEED_10GB_FULL,
4641 		    B_FALSE);
4642 		break;
4643 
4644 	default:
4645 		break;
4646 	}
4647 }
4648 
4649 #pragma inline(ixgbe_intr_rx_work)
4650 /*
4651  * ixgbe_intr_rx_work - RX processing of ISR.
4652  */
4653 static void
4654 ixgbe_intr_rx_work(ixgbe_rx_ring_t *rx_ring)
4655 {
4656 	mblk_t *mp;
4657 
4658 	mutex_enter(&rx_ring->rx_lock);
4659 
4660 	mp = ixgbe_ring_rx(rx_ring, IXGBE_POLL_NULL);
4661 	mutex_exit(&rx_ring->rx_lock);
4662 
4663 	if (mp != NULL)
4664 		mac_rx_ring(rx_ring->ixgbe->mac_hdl, rx_ring->ring_handle, mp,
4665 		    rx_ring->ring_gen_num);
4666 }
4667 
4668 #pragma inline(ixgbe_intr_tx_work)
4669 /*
4670  * ixgbe_intr_tx_work - TX processing of ISR.
4671  */
4672 static void
4673 ixgbe_intr_tx_work(ixgbe_tx_ring_t *tx_ring)
4674 {
4675 	ixgbe_t *ixgbe = tx_ring->ixgbe;
4676 
4677 	/*
4678 	 * Recycle the tx descriptors
4679 	 */
4680 	tx_ring->tx_recycle(tx_ring);
4681 
4682 	/*
4683 	 * Schedule the re-transmit
4684 	 */
4685 	if (tx_ring->reschedule &&
4686 	    (tx_ring->tbd_free >= ixgbe->tx_resched_thresh)) {
4687 		tx_ring->reschedule = B_FALSE;
4688 		mac_tx_ring_update(tx_ring->ixgbe->mac_hdl,
4689 		    tx_ring->ring_handle);
4690 		tx_ring->stat_reschedule++;
4691 	}
4692 }
4693 
4694 #pragma inline(ixgbe_intr_other_work)
4695 /*
4696  * ixgbe_intr_other_work - Process interrupt types other than tx/rx
4697  */
4698 static void
4699 ixgbe_intr_other_work(ixgbe_t *ixgbe, uint32_t eicr)
4700 {
4701 	struct ixgbe_hw *hw = &ixgbe->hw;
4702 
4703 	ASSERT(mutex_owned(&ixgbe->gen_lock));
4704 
4705 	/*
4706 	 * handle link status change
4707 	 */
4708 	if (eicr & IXGBE_EICR_LSC) {
4709 		ixgbe_driver_link_check(ixgbe);
4710 		ixgbe_get_hw_state(ixgbe);
4711 	}
4712 
4713 	/*
4714 	 * check for fan failure on adapters with fans
4715 	 */
4716 	if ((ixgbe->capab->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
4717 	    (eicr & IXGBE_EICR_GPI_SDP1)) {
4718 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_OVERTEMP);
4719 
4720 		/*
4721 		 * Disable the adapter interrupts
4722 		 */
4723 		ixgbe_disable_adapter_interrupts(ixgbe);
4724 
4725 		/*
4726 		 * Disable Rx/Tx units
4727 		 */
4728 		(void) ixgbe_stop_adapter(&ixgbe->hw);
4729 
4730 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
4731 		ixgbe_error(ixgbe,
4732 		    "Problem: Network adapter has been stopped "
4733 		    "because the fan has stopped.\n");
4734 		ixgbe_error(ixgbe,
4735 		    "Action: Replace the adapter.\n");
4736 
4737 		/* re-enable the interrupt, which was automasked */
4738 		ixgbe->eims |= IXGBE_EICR_GPI_SDP1;
4739 	}
4740 
4741 	/*
4742 	 * Do SFP check for adapters with hot-plug capability
4743 	 */
4744 	if ((ixgbe->capab->flags & IXGBE_FLAG_SFP_PLUG_CAPABLE) &&
4745 	    ((eicr & IXGBE_EICR_GPI_SDP1_BY_MAC(hw)) ||
4746 	    (eicr & IXGBE_EICR_GPI_SDP2_BY_MAC(hw)))) {
4747 		ixgbe->eicr = eicr;
4748 		if ((ddi_taskq_dispatch(ixgbe->sfp_taskq,
4749 		    ixgbe_sfp_check, (void *)ixgbe,
4750 		    DDI_NOSLEEP)) != DDI_SUCCESS) {
4751 			ixgbe_log(ixgbe, "No memory available to dispatch "
4752 			    "taskq for SFP check");
4753 		}
4754 	}
4755 
4756 	/*
4757 	 * Do over-temperature check for adapters with temp sensor
4758 	 */
4759 	if ((ixgbe->capab->flags & IXGBE_FLAG_TEMP_SENSOR_CAPABLE) &&
4760 	    ((eicr & IXGBE_EICR_GPI_SDP0_BY_MAC(hw)) ||
4761 	    (eicr & IXGBE_EICR_LSC))) {
4762 		ixgbe->eicr = eicr;
4763 		if ((ddi_taskq_dispatch(ixgbe->overtemp_taskq,
4764 		    ixgbe_overtemp_check, (void *)ixgbe,
4765 		    DDI_NOSLEEP)) != DDI_SUCCESS) {
4766 			ixgbe_log(ixgbe, "No memory available to dispatch "
4767 			    "taskq for overtemp check");
4768 		}
4769 	}
4770 
4771 	/*
4772 	 * Process an external PHY interrupt
4773 	 */
4774 	if (hw->device_id == IXGBE_DEV_ID_X550EM_X_10G_T &&
4775 	    (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
4776 		ixgbe->eicr = eicr;
4777 		if ((ddi_taskq_dispatch(ixgbe->phy_taskq,
4778 		    ixgbe_phy_check, (void *)ixgbe,
4779 		    DDI_NOSLEEP)) != DDI_SUCCESS) {
4780 			ixgbe_log(ixgbe, "No memory available to dispatch "
4781 			    "taskq for PHY check");
4782 		}
4783 	}
4784 }
4785 
4786 /*
4787  * ixgbe_intr_legacy - Interrupt handler for legacy interrupts.
4788  */
4789 static uint_t
4790 ixgbe_intr_legacy(void *arg1, void *arg2)
4791 {
4792 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
4793 	struct ixgbe_hw *hw = &ixgbe->hw;
4794 	ixgbe_tx_ring_t *tx_ring;
4795 	ixgbe_rx_ring_t *rx_ring;
4796 	uint32_t eicr;
4797 	mblk_t *mp;
4798 	boolean_t tx_reschedule;
4799 	uint_t result;
4800 
4801 	_NOTE(ARGUNUSED(arg2));
4802 
4803 	mutex_enter(&ixgbe->gen_lock);
4804 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
4805 		mutex_exit(&ixgbe->gen_lock);
4806 		return (DDI_INTR_UNCLAIMED);
4807 	}
4808 
4809 	mp = NULL;
4810 	tx_reschedule = B_FALSE;
4811 
4812 	/*
4813 	 * Any bit set in eicr: claim this interrupt
4814 	 */
4815 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4816 
4817 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
4818 		mutex_exit(&ixgbe->gen_lock);
4819 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4820 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4821 		return (DDI_INTR_CLAIMED);
4822 	}
4823 
4824 	if (eicr) {
4825 		/*
4826 		 * For legacy interrupt, we have only one interrupt,
4827 		 * so we have only one rx ring and one tx ring enabled.
4828 		 */
4829 		ASSERT(ixgbe->num_rx_rings == 1);
4830 		ASSERT(ixgbe->num_tx_rings == 1);
4831 
4832 		/*
4833 		 * For legacy interrupt, rx rings[0] will use RTxQ[0].
4834 		 */
4835 		if (eicr & 0x1) {
4836 			ixgbe->eimc |= IXGBE_EICR_RTX_QUEUE;
4837 			IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4838 			ixgbe->eims |= IXGBE_EICR_RTX_QUEUE;
4839 			/*
4840 			 * Clean the rx descriptors
4841 			 */
4842 			rx_ring = &ixgbe->rx_rings[0];
4843 			mp = ixgbe_ring_rx(rx_ring, IXGBE_POLL_NULL);
4844 		}
4845 
4846 		/*
4847 		 * For legacy interrupt, tx rings[0] will use RTxQ[1].
4848 		 */
4849 		if (eicr & 0x2) {
4850 			/*
4851 			 * Recycle the tx descriptors
4852 			 */
4853 			tx_ring = &ixgbe->tx_rings[0];
4854 			tx_ring->tx_recycle(tx_ring);
4855 
4856 			/*
4857 			 * Schedule the re-transmit
4858 			 */
4859 			tx_reschedule = (tx_ring->reschedule &&
4860 			    (tx_ring->tbd_free >= ixgbe->tx_resched_thresh));
4861 		}
4862 
4863 		/* any interrupt type other than tx/rx */
4864 		if (eicr & ixgbe->capab->other_intr) {
4865 			switch (hw->mac.type) {
4866 			case ixgbe_mac_82598EB:
4867 				ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4868 				break;
4869 
4870 			case ixgbe_mac_82599EB:
4871 			case ixgbe_mac_X540:
4872 			case ixgbe_mac_X550:
4873 			case ixgbe_mac_X550EM_x:
4874 			case ixgbe_mac_X550EM_a:
4875 				ixgbe->eimc = IXGBE_82599_OTHER_INTR;
4876 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4877 				break;
4878 
4879 			default:
4880 				break;
4881 			}
4882 			ixgbe_intr_other_work(ixgbe, eicr);
4883 			ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4884 		}
4885 
4886 		mutex_exit(&ixgbe->gen_lock);
4887 
4888 		result = DDI_INTR_CLAIMED;
4889 	} else {
4890 		mutex_exit(&ixgbe->gen_lock);
4891 
4892 		/*
4893 		 * No interrupt cause bits set: don't claim this interrupt.
4894 		 */
4895 		result = DDI_INTR_UNCLAIMED;
4896 	}
4897 
4898 	/* re-enable the interrupts which were automasked */
4899 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4900 
4901 	/*
4902 	 * Do the following work outside of the gen_lock
4903 	 */
4904 	if (mp != NULL) {
4905 		mac_rx_ring(rx_ring->ixgbe->mac_hdl, rx_ring->ring_handle, mp,
4906 		    rx_ring->ring_gen_num);
4907 	}
4908 
4909 	if (tx_reschedule)  {
4910 		tx_ring->reschedule = B_FALSE;
4911 		mac_tx_ring_update(ixgbe->mac_hdl, tx_ring->ring_handle);
4912 		tx_ring->stat_reschedule++;
4913 	}
4914 
4915 	return (result);
4916 }
4917 
4918 /*
4919  * ixgbe_intr_msi - Interrupt handler for MSI.
4920  */
4921 static uint_t
4922 ixgbe_intr_msi(void *arg1, void *arg2)
4923 {
4924 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
4925 	struct ixgbe_hw *hw = &ixgbe->hw;
4926 	uint32_t eicr;
4927 
4928 	_NOTE(ARGUNUSED(arg2));
4929 
4930 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4931 
4932 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
4933 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4934 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4935 		return (DDI_INTR_CLAIMED);
4936 	}
4937 
4938 	/*
4939 	 * For MSI interrupt, we have only one vector,
4940 	 * so we have only one rx ring and one tx ring enabled.
4941 	 */
4942 	ASSERT(ixgbe->num_rx_rings == 1);
4943 	ASSERT(ixgbe->num_tx_rings == 1);
4944 
4945 	/*
4946 	 * For MSI interrupt, rx rings[0] will use RTxQ[0].
4947 	 */
4948 	if (eicr & 0x1) {
4949 		ixgbe_intr_rx_work(&ixgbe->rx_rings[0]);
4950 	}
4951 
4952 	/*
4953 	 * For MSI interrupt, tx rings[0] will use RTxQ[1].
4954 	 */
4955 	if (eicr & 0x2) {
4956 		ixgbe_intr_tx_work(&ixgbe->tx_rings[0]);
4957 	}
4958 
4959 	/* any interrupt type other than tx/rx */
4960 	if (eicr & ixgbe->capab->other_intr) {
4961 		mutex_enter(&ixgbe->gen_lock);
4962 		switch (hw->mac.type) {
4963 		case ixgbe_mac_82598EB:
4964 			ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4965 			break;
4966 
4967 		case ixgbe_mac_82599EB:
4968 		case ixgbe_mac_X540:
4969 		case ixgbe_mac_X550:
4970 		case ixgbe_mac_X550EM_x:
4971 		case ixgbe_mac_X550EM_a:
4972 			ixgbe->eimc = IXGBE_82599_OTHER_INTR;
4973 			IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4974 			break;
4975 
4976 		default:
4977 			break;
4978 		}
4979 		ixgbe_intr_other_work(ixgbe, eicr);
4980 		ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4981 		mutex_exit(&ixgbe->gen_lock);
4982 	}
4983 
4984 	/* re-enable the interrupts which were automasked */
4985 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4986 
4987 	return (DDI_INTR_CLAIMED);
4988 }
4989 
4990 /*
4991  * ixgbe_intr_msix - Interrupt handler for MSI-X.
4992  */
4993 static uint_t
4994 ixgbe_intr_msix(void *arg1, void *arg2)
4995 {
4996 	ixgbe_intr_vector_t *vect = (ixgbe_intr_vector_t *)arg1;
4997 	ixgbe_t *ixgbe = vect->ixgbe;
4998 	struct ixgbe_hw *hw = &ixgbe->hw;
4999 	uint32_t eicr;
5000 	int r_idx = 0;
5001 
5002 	_NOTE(ARGUNUSED(arg2));
5003 
5004 	/*
5005 	 * Clean each rx ring that has its bit set in the map
5006 	 */
5007 	r_idx = bt_getlowbit(vect->rx_map, 0, (ixgbe->num_rx_rings - 1));
5008 	while (r_idx >= 0) {
5009 		ixgbe_intr_rx_work(&ixgbe->rx_rings[r_idx]);
5010 		r_idx = bt_getlowbit(vect->rx_map, (r_idx + 1),
5011 		    (ixgbe->num_rx_rings - 1));
5012 	}
5013 
5014 	/*
5015 	 * Clean each tx ring that has its bit set in the map
5016 	 */
5017 	r_idx = bt_getlowbit(vect->tx_map, 0, (ixgbe->num_tx_rings - 1));
5018 	while (r_idx >= 0) {
5019 		ixgbe_intr_tx_work(&ixgbe->tx_rings[r_idx]);
5020 		r_idx = bt_getlowbit(vect->tx_map, (r_idx + 1),
5021 		    (ixgbe->num_tx_rings - 1));
5022 	}
5023 
5024 
5025 	/*
5026 	 * Clean other interrupt (link change) that has its bit set in the map
5027 	 */
5028 	if (BT_TEST(vect->other_map, 0) == 1) {
5029 		eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
5030 
5031 		if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) !=
5032 		    DDI_FM_OK) {
5033 			ddi_fm_service_impact(ixgbe->dip,
5034 			    DDI_SERVICE_DEGRADED);
5035 			atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
5036 			return (DDI_INTR_CLAIMED);
5037 		}
5038 
5039 		/*
5040 		 * Check "other" cause bits: any interrupt type other than tx/rx
5041 		 */
5042 		if (eicr & ixgbe->capab->other_intr) {
5043 			mutex_enter(&ixgbe->gen_lock);
5044 			switch (hw->mac.type) {
5045 			case ixgbe_mac_82598EB:
5046 				ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
5047 				ixgbe_intr_other_work(ixgbe, eicr);
5048 				break;
5049 
5050 			case ixgbe_mac_82599EB:
5051 			case ixgbe_mac_X540:
5052 			case ixgbe_mac_X550:
5053 			case ixgbe_mac_X550EM_x:
5054 			case ixgbe_mac_X550EM_a:
5055 				ixgbe->eims |= IXGBE_EICR_RTX_QUEUE;
5056 				ixgbe_intr_other_work(ixgbe, eicr);
5057 				break;
5058 
5059 			default:
5060 				break;
5061 			}
5062 			mutex_exit(&ixgbe->gen_lock);
5063 		}
5064 
5065 		/* re-enable the interrupts which were automasked */
5066 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
5067 	}
5068 
5069 	return (DDI_INTR_CLAIMED);
5070 }
5071 
5072 /*
5073  * ixgbe_alloc_intrs - Allocate interrupts for the driver.
5074  *
5075  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
5076  * if not successful, try Legacy.
5077  * ixgbe->intr_force can be used to force sequence to start with
5078  * any of the 3 types.
5079  * If MSI-X is not used, number of tx/rx rings is forced to 1.
5080  */
5081 static int
5082 ixgbe_alloc_intrs(ixgbe_t *ixgbe)
5083 {
5084 	dev_info_t *devinfo;
5085 	int intr_types;
5086 	int rc;
5087 
5088 	devinfo = ixgbe->dip;
5089 
5090 	/*
5091 	 * Get supported interrupt types
5092 	 */
5093 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
5094 
5095 	if (rc != DDI_SUCCESS) {
5096 		ixgbe_log(ixgbe,
5097 		    "Get supported interrupt types failed: %d", rc);
5098 		return (IXGBE_FAILURE);
5099 	}
5100 	IXGBE_DEBUGLOG_1(ixgbe, "Supported interrupt types: %x", intr_types);
5101 
5102 	ixgbe->intr_type = 0;
5103 
5104 	/*
5105 	 * Install MSI-X interrupts
5106 	 */
5107 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
5108 	    (ixgbe->intr_force <= IXGBE_INTR_MSIX)) {
5109 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_MSIX);
5110 		if (rc == IXGBE_SUCCESS)
5111 			return (IXGBE_SUCCESS);
5112 
5113 		ixgbe_log(ixgbe,
5114 		    "Allocate MSI-X failed, trying MSI interrupts...");
5115 	}
5116 
5117 	/*
5118 	 * MSI-X not used, force rings and groups to 1
5119 	 */
5120 	ixgbe->num_rx_rings = 1;
5121 	ixgbe->num_rx_groups = 1;
5122 	ixgbe->num_tx_rings = 1;
5123 	ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
5124 	ixgbe_log(ixgbe,
5125 	    "MSI-X not used, force rings and groups number to 1");
5126 
5127 	/*
5128 	 * Install MSI interrupts
5129 	 */
5130 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
5131 	    (ixgbe->intr_force <= IXGBE_INTR_MSI)) {
5132 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_MSI);
5133 		if (rc == IXGBE_SUCCESS)
5134 			return (IXGBE_SUCCESS);
5135 
5136 		ixgbe_log(ixgbe,
5137 		    "Allocate MSI failed, trying Legacy interrupts...");
5138 	}
5139 
5140 	/*
5141 	 * Install legacy interrupts
5142 	 */
5143 	if (intr_types & DDI_INTR_TYPE_FIXED) {
5144 		/*
5145 		 * Disallow legacy interrupts for X550. X550 has a silicon
5146 		 * bug which prevents Shared Legacy interrupts from working.
5147 		 * For details, please reference:
5148 		 *
5149 		 * Intel Ethernet Controller X550 Specification Update rev. 2.1
5150 		 * May 2016, erratum 22: PCIe Interrupt Status Bit
5151 		 */
5152 		if (ixgbe->hw.mac.type == ixgbe_mac_X550 ||
5153 		    ixgbe->hw.mac.type == ixgbe_mac_X550EM_x ||
5154 		    ixgbe->hw.mac.type == ixgbe_mac_X550EM_a ||
5155 		    ixgbe->hw.mac.type == ixgbe_mac_X550_vf ||
5156 		    ixgbe->hw.mac.type == ixgbe_mac_X550EM_x_vf ||
5157 		    ixgbe->hw.mac.type == ixgbe_mac_X550EM_a_vf) {
5158 			ixgbe_log(ixgbe,
5159 			    "Legacy interrupts are not supported on this "
5160 			    "adapter. Please use MSI or MSI-X instead.");
5161 			return (IXGBE_FAILURE);
5162 		}
5163 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_FIXED);
5164 		if (rc == IXGBE_SUCCESS)
5165 			return (IXGBE_SUCCESS);
5166 
5167 		ixgbe_log(ixgbe,
5168 		    "Allocate Legacy interrupts failed");
5169 	}
5170 
5171 	/*
5172 	 * If none of the 3 types succeeded, return failure
5173 	 */
5174 	return (IXGBE_FAILURE);
5175 }
5176 
5177 /*
5178  * ixgbe_alloc_intr_handles - Allocate interrupt handles.
5179  *
5180  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
5181  * if fewer than 2 handles are available, return failure.
5182  * Upon success, this maps the vectors to rx and tx rings for
5183  * interrupts.
5184  */
5185 static int
5186 ixgbe_alloc_intr_handles(ixgbe_t *ixgbe, int intr_type)
5187 {
5188 	dev_info_t *devinfo;
5189 	int request, count, actual;
5190 	int minimum;
5191 	int rc;
5192 	uint32_t ring_per_group;
5193 
5194 	devinfo = ixgbe->dip;
5195 
5196 	switch (intr_type) {
5197 	case DDI_INTR_TYPE_FIXED:
5198 		request = 1;	/* Request 1 legacy interrupt handle */
5199 		minimum = 1;
5200 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: legacy");
5201 		break;
5202 
5203 	case DDI_INTR_TYPE_MSI:
5204 		request = 1;	/* Request 1 MSI interrupt handle */
5205 		minimum = 1;
5206 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: MSI");
5207 		break;
5208 
5209 	case DDI_INTR_TYPE_MSIX:
5210 		/*
5211 		 * Best number of vectors for the adapter is
5212 		 * (# rx rings + # tx rings), however we will
5213 		 * limit the request number.
5214 		 */
5215 		request = min(16, ixgbe->num_rx_rings + ixgbe->num_tx_rings);
5216 		if (request > ixgbe->capab->max_ring_vect)
5217 			request = ixgbe->capab->max_ring_vect;
5218 		minimum = 1;
5219 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: MSI-X");
5220 		break;
5221 
5222 	default:
5223 		ixgbe_log(ixgbe,
5224 		    "invalid call to ixgbe_alloc_intr_handles(): %d\n",
5225 		    intr_type);
5226 		return (IXGBE_FAILURE);
5227 	}
5228 	IXGBE_DEBUGLOG_2(ixgbe, "interrupt handles requested: %d  minimum: %d",
5229 	    request, minimum);
5230 
5231 	/*
5232 	 * Get number of supported interrupts
5233 	 */
5234 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
5235 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
5236 		ixgbe_log(ixgbe,
5237 		    "Get interrupt number failed. Return: %d, count: %d",
5238 		    rc, count);
5239 		return (IXGBE_FAILURE);
5240 	}
5241 	IXGBE_DEBUGLOG_1(ixgbe, "interrupts supported: %d", count);
5242 
5243 	actual = 0;
5244 	ixgbe->intr_cnt = 0;
5245 	ixgbe->intr_cnt_max = 0;
5246 	ixgbe->intr_cnt_min = 0;
5247 
5248 	/*
5249 	 * Allocate an array of interrupt handles
5250 	 */
5251 	ixgbe->intr_size = request * sizeof (ddi_intr_handle_t);
5252 	ixgbe->htable = kmem_alloc(ixgbe->intr_size, KM_SLEEP);
5253 
5254 	rc = ddi_intr_alloc(devinfo, ixgbe->htable, intr_type, 0,
5255 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
5256 	if (rc != DDI_SUCCESS) {
5257 		ixgbe_log(ixgbe, "Allocate interrupts failed. "
5258 		    "return: %d, request: %d, actual: %d",
5259 		    rc, request, actual);
5260 		goto alloc_handle_fail;
5261 	}
5262 	IXGBE_DEBUGLOG_1(ixgbe, "interrupts actually allocated: %d", actual);
5263 
5264 	/*
5265 	 * upper/lower limit of interrupts
5266 	 */
5267 	ixgbe->intr_cnt = actual;
5268 	ixgbe->intr_cnt_max = request;
5269 	ixgbe->intr_cnt_min = minimum;
5270 
5271 	/*
5272 	 * rss number per group should not exceed the rx interrupt number,
5273 	 * else need to adjust rx ring number.
5274 	 */
5275 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
5276 	ASSERT((ixgbe->num_rx_rings % ixgbe->num_rx_groups) == 0);
5277 	if (actual < ring_per_group) {
5278 		ixgbe->num_rx_rings = ixgbe->num_rx_groups * actual;
5279 		ixgbe_setup_vmdq_rss_conf(ixgbe);
5280 	}
5281 
5282 	/*
5283 	 * Now we know the actual number of vectors.  Here we map the vector
5284 	 * to other, rx rings and tx ring.
5285 	 */
5286 	if (actual < minimum) {
5287 		ixgbe_log(ixgbe, "Insufficient interrupt handles available: %d",
5288 		    actual);
5289 		goto alloc_handle_fail;
5290 	}
5291 
5292 	/*
5293 	 * Get priority for first vector, assume remaining are all the same
5294 	 */
5295 	rc = ddi_intr_get_pri(ixgbe->htable[0], &ixgbe->intr_pri);
5296 	if (rc != DDI_SUCCESS) {
5297 		ixgbe_log(ixgbe,
5298 		    "Get interrupt priority failed: %d", rc);
5299 		goto alloc_handle_fail;
5300 	}
5301 
5302 	rc = ddi_intr_get_cap(ixgbe->htable[0], &ixgbe->intr_cap);
5303 	if (rc != DDI_SUCCESS) {
5304 		ixgbe_log(ixgbe,
5305 		    "Get interrupt cap failed: %d", rc);
5306 		goto alloc_handle_fail;
5307 	}
5308 
5309 	ixgbe->intr_type = intr_type;
5310 
5311 	return (IXGBE_SUCCESS);
5312 
5313 alloc_handle_fail:
5314 	ixgbe_rem_intrs(ixgbe);
5315 
5316 	return (IXGBE_FAILURE);
5317 }
5318 
5319 /*
5320  * ixgbe_add_intr_handlers - Add interrupt handlers based on the interrupt type.
5321  *
5322  * Before adding the interrupt handlers, the interrupt vectors have
5323  * been allocated, and the rx/tx rings have also been allocated.
5324  */
5325 static int
5326 ixgbe_add_intr_handlers(ixgbe_t *ixgbe)
5327 {
5328 	int vector = 0;
5329 	int rc;
5330 
5331 	switch (ixgbe->intr_type) {
5332 	case DDI_INTR_TYPE_MSIX:
5333 		/*
5334 		 * Add interrupt handler for all vectors
5335 		 */
5336 		for (vector = 0; vector < ixgbe->intr_cnt; vector++) {
5337 			/*
5338 			 * install pointer to vect_map[vector]
5339 			 */
5340 			rc = ddi_intr_add_handler(ixgbe->htable[vector],
5341 			    (ddi_intr_handler_t *)ixgbe_intr_msix,
5342 			    (void *)&ixgbe->vect_map[vector], NULL);
5343 
5344 			if (rc != DDI_SUCCESS) {
5345 				ixgbe_log(ixgbe,
5346 				    "Add interrupt handler failed. "
5347 				    "return: %d, vector: %d", rc, vector);
5348 				for (vector--; vector >= 0; vector--) {
5349 					(void) ddi_intr_remove_handler(
5350 					    ixgbe->htable[vector]);
5351 				}
5352 				return (IXGBE_FAILURE);
5353 			}
5354 		}
5355 
5356 		break;
5357 
5358 	case DDI_INTR_TYPE_MSI:
5359 		/*
5360 		 * Add interrupt handlers for the only vector
5361 		 */
5362 		rc = ddi_intr_add_handler(ixgbe->htable[vector],
5363 		    (ddi_intr_handler_t *)ixgbe_intr_msi,
5364 		    (void *)ixgbe, NULL);
5365 
5366 		if (rc != DDI_SUCCESS) {
5367 			ixgbe_log(ixgbe,
5368 			    "Add MSI interrupt handler failed: %d", rc);
5369 			return (IXGBE_FAILURE);
5370 		}
5371 
5372 		break;
5373 
5374 	case DDI_INTR_TYPE_FIXED:
5375 		/*
5376 		 * Add interrupt handlers for the only vector
5377 		 */
5378 		rc = ddi_intr_add_handler(ixgbe->htable[vector],
5379 		    (ddi_intr_handler_t *)ixgbe_intr_legacy,
5380 		    (void *)ixgbe, NULL);
5381 
5382 		if (rc != DDI_SUCCESS) {
5383 			ixgbe_log(ixgbe,
5384 			    "Add legacy interrupt handler failed: %d", rc);
5385 			return (IXGBE_FAILURE);
5386 		}
5387 
5388 		break;
5389 
5390 	default:
5391 		return (IXGBE_FAILURE);
5392 	}
5393 
5394 	return (IXGBE_SUCCESS);
5395 }
5396 
5397 #pragma inline(ixgbe_map_rxring_to_vector)
5398 /*
5399  * ixgbe_map_rxring_to_vector - Map given rx ring to given interrupt vector.
5400  */
5401 static void
5402 ixgbe_map_rxring_to_vector(ixgbe_t *ixgbe, int r_idx, int v_idx)
5403 {
5404 	/*
5405 	 * Set bit in map
5406 	 */
5407 	BT_SET(ixgbe->vect_map[v_idx].rx_map, r_idx);
5408 
5409 	/*
5410 	 * Count bits set
5411 	 */
5412 	ixgbe->vect_map[v_idx].rxr_cnt++;
5413 
5414 	/*
5415 	 * Remember bit position
5416 	 */
5417 	ixgbe->rx_rings[r_idx].intr_vector = v_idx;
5418 	ixgbe->rx_rings[r_idx].vect_bit = 1 << v_idx;
5419 }
5420 
5421 #pragma inline(ixgbe_map_txring_to_vector)
5422 /*
5423  * ixgbe_map_txring_to_vector - Map given tx ring to given interrupt vector.
5424  */
5425 static void
5426 ixgbe_map_txring_to_vector(ixgbe_t *ixgbe, int t_idx, int v_idx)
5427 {
5428 	/*
5429 	 * Set bit in map
5430 	 */
5431 	BT_SET(ixgbe->vect_map[v_idx].tx_map, t_idx);
5432 
5433 	/*
5434 	 * Count bits set
5435 	 */
5436 	ixgbe->vect_map[v_idx].txr_cnt++;
5437 
5438 	/*
5439 	 * Remember bit position
5440 	 */
5441 	ixgbe->tx_rings[t_idx].intr_vector = v_idx;
5442 	ixgbe->tx_rings[t_idx].vect_bit = 1 << v_idx;
5443 }
5444 
5445 /*
5446  * ixgbe_setup_ivar - Set the given entry in the given interrupt vector
5447  * allocation register (IVAR).
5448  * cause:
5449  *   -1 : other cause
5450  *    0 : rx
5451  *    1 : tx
5452  */
5453 static void
5454 ixgbe_setup_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, uint8_t msix_vector,
5455     int8_t cause)
5456 {
5457 	struct ixgbe_hw *hw = &ixgbe->hw;
5458 	u32 ivar, index;
5459 
5460 	switch (hw->mac.type) {
5461 	case ixgbe_mac_82598EB:
5462 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
5463 		if (cause == -1) {
5464 			cause = 0;
5465 		}
5466 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
5467 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
5468 		ivar &= ~(0xFF << (8 * (intr_alloc_entry & 0x3)));
5469 		ivar |= (msix_vector << (8 * (intr_alloc_entry & 0x3)));
5470 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
5471 		break;
5472 
5473 	case ixgbe_mac_82599EB:
5474 	case ixgbe_mac_X540:
5475 	case ixgbe_mac_X550:
5476 	case ixgbe_mac_X550EM_x:
5477 	case ixgbe_mac_X550EM_a:
5478 		if (cause == -1) {
5479 			/* other causes */
5480 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
5481 			index = (intr_alloc_entry & 1) * 8;
5482 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
5483 			ivar &= ~(0xFF << index);
5484 			ivar |= (msix_vector << index);
5485 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
5486 		} else {
5487 			/* tx or rx causes */
5488 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
5489 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
5490 			ivar = IXGBE_READ_REG(hw,
5491 			    IXGBE_IVAR(intr_alloc_entry >> 1));
5492 			ivar &= ~(0xFF << index);
5493 			ivar |= (msix_vector << index);
5494 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
5495 			    ivar);
5496 		}
5497 		break;
5498 
5499 	default:
5500 		break;
5501 	}
5502 }
5503 
5504 /*
5505  * ixgbe_enable_ivar - Enable the given entry by setting the VAL bit of
5506  * given interrupt vector allocation register (IVAR).
5507  * cause:
5508  *   -1 : other cause
5509  *    0 : rx
5510  *    1 : tx
5511  */
5512 static void
5513 ixgbe_enable_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, int8_t cause)
5514 {
5515 	struct ixgbe_hw *hw = &ixgbe->hw;
5516 	u32 ivar, index;
5517 
5518 	switch (hw->mac.type) {
5519 	case ixgbe_mac_82598EB:
5520 		if (cause == -1) {
5521 			cause = 0;
5522 		}
5523 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
5524 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
5525 		ivar |= (IXGBE_IVAR_ALLOC_VAL << (8 *
5526 		    (intr_alloc_entry & 0x3)));
5527 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
5528 		break;
5529 
5530 	case ixgbe_mac_82599EB:
5531 	case ixgbe_mac_X540:
5532 	case ixgbe_mac_X550:
5533 	case ixgbe_mac_X550EM_x:
5534 	case ixgbe_mac_X550EM_a:
5535 		if (cause == -1) {
5536 			/* other causes */
5537 			index = (intr_alloc_entry & 1) * 8;
5538 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
5539 			ivar |= (IXGBE_IVAR_ALLOC_VAL << index);
5540 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
5541 		} else {
5542 			/* tx or rx causes */
5543 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
5544 			ivar = IXGBE_READ_REG(hw,
5545 			    IXGBE_IVAR(intr_alloc_entry >> 1));
5546 			ivar |= (IXGBE_IVAR_ALLOC_VAL << index);
5547 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
5548 			    ivar);
5549 		}
5550 		break;
5551 
5552 	default:
5553 		break;
5554 	}
5555 }
5556 
5557 /*
5558  * ixgbe_disable_ivar - Disble the given entry by clearing the VAL bit of
5559  * given interrupt vector allocation register (IVAR).
5560  * cause:
5561  *   -1 : other cause
5562  *    0 : rx
5563  *    1 : tx
5564  */
5565 static void
5566 ixgbe_disable_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, int8_t cause)
5567 {
5568 	struct ixgbe_hw *hw = &ixgbe->hw;
5569 	u32 ivar, index;
5570 
5571 	switch (hw->mac.type) {
5572 	case ixgbe_mac_82598EB:
5573 		if (cause == -1) {
5574 			cause = 0;
5575 		}
5576 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
5577 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
5578 		ivar &= ~(IXGBE_IVAR_ALLOC_VAL<< (8 *
5579 		    (intr_alloc_entry & 0x3)));
5580 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
5581 		break;
5582 
5583 	case ixgbe_mac_82599EB:
5584 	case ixgbe_mac_X540:
5585 	case ixgbe_mac_X550:
5586 	case ixgbe_mac_X550EM_x:
5587 	case ixgbe_mac_X550EM_a:
5588 		if (cause == -1) {
5589 			/* other causes */
5590 			index = (intr_alloc_entry & 1) * 8;
5591 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
5592 			ivar &= ~(IXGBE_IVAR_ALLOC_VAL << index);
5593 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
5594 		} else {
5595 			/* tx or rx causes */
5596 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
5597 			ivar = IXGBE_READ_REG(hw,
5598 			    IXGBE_IVAR(intr_alloc_entry >> 1));
5599 			ivar &= ~(IXGBE_IVAR_ALLOC_VAL << index);
5600 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
5601 			    ivar);
5602 		}
5603 		break;
5604 
5605 	default:
5606 		break;
5607 	}
5608 }
5609 
5610 /*
5611  * Convert the rx ring index driver maintained to the rx ring index
5612  * in h/w.
5613  */
5614 static uint32_t
5615 ixgbe_get_hw_rx_index(ixgbe_t *ixgbe, uint32_t sw_rx_index)
5616 {
5617 
5618 	struct ixgbe_hw *hw = &ixgbe->hw;
5619 	uint32_t rx_ring_per_group, hw_rx_index;
5620 
5621 	if (ixgbe->classify_mode == IXGBE_CLASSIFY_RSS ||
5622 	    ixgbe->classify_mode == IXGBE_CLASSIFY_NONE) {
5623 		return (sw_rx_index);
5624 	} else if (ixgbe->classify_mode == IXGBE_CLASSIFY_VMDQ) {
5625 		switch (hw->mac.type) {
5626 		case ixgbe_mac_82598EB:
5627 			return (sw_rx_index);
5628 
5629 		case ixgbe_mac_82599EB:
5630 		case ixgbe_mac_X540:
5631 		case ixgbe_mac_X550:
5632 		case ixgbe_mac_X550EM_x:
5633 		case ixgbe_mac_X550EM_a:
5634 			return (sw_rx_index * 2);
5635 
5636 		default:
5637 			break;
5638 		}
5639 	} else if (ixgbe->classify_mode == IXGBE_CLASSIFY_VMDQ_RSS) {
5640 		rx_ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
5641 
5642 		switch (hw->mac.type) {
5643 		case ixgbe_mac_82598EB:
5644 			hw_rx_index = (sw_rx_index / rx_ring_per_group) *
5645 			    16 + (sw_rx_index % rx_ring_per_group);
5646 			return (hw_rx_index);
5647 
5648 		case ixgbe_mac_82599EB:
5649 		case ixgbe_mac_X540:
5650 		case ixgbe_mac_X550:
5651 		case ixgbe_mac_X550EM_x:
5652 		case ixgbe_mac_X550EM_a:
5653 			if (ixgbe->num_rx_groups > 32) {
5654 				hw_rx_index = (sw_rx_index /
5655 				    rx_ring_per_group) * 2 +
5656 				    (sw_rx_index % rx_ring_per_group);
5657 			} else {
5658 				hw_rx_index = (sw_rx_index /
5659 				    rx_ring_per_group) * 4 +
5660 				    (sw_rx_index % rx_ring_per_group);
5661 			}
5662 			return (hw_rx_index);
5663 
5664 		default:
5665 			break;
5666 		}
5667 	}
5668 
5669 	/*
5670 	 * Should never reach. Just to make compiler happy.
5671 	 */
5672 	return (sw_rx_index);
5673 }
5674 
5675 /*
5676  * ixgbe_map_intrs_to_vectors - Map different interrupts to MSI-X vectors.
5677  *
5678  * For MSI-X, here will map rx interrupt, tx interrupt and other interrupt
5679  * to vector[0 - (intr_cnt -1)].
5680  */
5681 static int
5682 ixgbe_map_intrs_to_vectors(ixgbe_t *ixgbe)
5683 {
5684 	int i, vector = 0;
5685 
5686 	/* initialize vector map */
5687 	bzero(&ixgbe->vect_map, sizeof (ixgbe->vect_map));
5688 	for (i = 0; i < ixgbe->intr_cnt; i++) {
5689 		ixgbe->vect_map[i].ixgbe = ixgbe;
5690 	}
5691 
5692 	/*
5693 	 * non-MSI-X case is very simple: rx rings[0] on RTxQ[0],
5694 	 * tx rings[0] on RTxQ[1].
5695 	 */
5696 	if (ixgbe->intr_type != DDI_INTR_TYPE_MSIX) {
5697 		ixgbe_map_rxring_to_vector(ixgbe, 0, 0);
5698 		ixgbe_map_txring_to_vector(ixgbe, 0, 1);
5699 		return (IXGBE_SUCCESS);
5700 	}
5701 
5702 	/*
5703 	 * Interrupts/vectors mapping for MSI-X
5704 	 */
5705 
5706 	/*
5707 	 * Map other interrupt to vector 0,
5708 	 * Set bit in map and count the bits set.
5709 	 */
5710 	BT_SET(ixgbe->vect_map[vector].other_map, 0);
5711 	ixgbe->vect_map[vector].other_cnt++;
5712 
5713 	/*
5714 	 * Map rx ring interrupts to vectors
5715 	 */
5716 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
5717 		ixgbe_map_rxring_to_vector(ixgbe, i, vector);
5718 		vector = (vector +1) % ixgbe->intr_cnt;
5719 	}
5720 
5721 	/*
5722 	 * Map tx ring interrupts to vectors
5723 	 */
5724 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
5725 		ixgbe_map_txring_to_vector(ixgbe, i, vector);
5726 		vector = (vector +1) % ixgbe->intr_cnt;
5727 	}
5728 
5729 	return (IXGBE_SUCCESS);
5730 }
5731 
5732 /*
5733  * ixgbe_setup_adapter_vector - Setup the adapter interrupt vector(s).
5734  *
5735  * This relies on ring/vector mapping already set up in the
5736  * vect_map[] structures
5737  */
5738 static void
5739 ixgbe_setup_adapter_vector(ixgbe_t *ixgbe)
5740 {
5741 	struct ixgbe_hw *hw = &ixgbe->hw;
5742 	ixgbe_intr_vector_t *vect;	/* vector bitmap */
5743 	int r_idx;	/* ring index */
5744 	int v_idx;	/* vector index */
5745 	uint32_t hw_index;
5746 
5747 	/*
5748 	 * Clear any previous entries
5749 	 */
5750 	switch (hw->mac.type) {
5751 	case ixgbe_mac_82598EB:
5752 		for (v_idx = 0; v_idx < 25; v_idx++)
5753 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(v_idx), 0);
5754 		break;
5755 
5756 	case ixgbe_mac_82599EB:
5757 	case ixgbe_mac_X540:
5758 	case ixgbe_mac_X550:
5759 	case ixgbe_mac_X550EM_x:
5760 	case ixgbe_mac_X550EM_a:
5761 		for (v_idx = 0; v_idx < 64; v_idx++)
5762 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(v_idx), 0);
5763 		IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, 0);
5764 		break;
5765 
5766 	default:
5767 		break;
5768 	}
5769 
5770 	/*
5771 	 * For non MSI-X interrupt, rx rings[0] will use RTxQ[0], and
5772 	 * tx rings[0] will use RTxQ[1].
5773 	 */
5774 	if (ixgbe->intr_type != DDI_INTR_TYPE_MSIX) {
5775 		ixgbe_setup_ivar(ixgbe, 0, 0, 0);
5776 		ixgbe_setup_ivar(ixgbe, 0, 1, 1);
5777 		return;
5778 	}
5779 
5780 	/*
5781 	 * For MSI-X interrupt, "Other" is always on vector[0].
5782 	 */
5783 	ixgbe_setup_ivar(ixgbe, IXGBE_IVAR_OTHER_CAUSES_INDEX, 0, -1);
5784 
5785 	/*
5786 	 * For each interrupt vector, populate the IVAR table
5787 	 */
5788 	for (v_idx = 0; v_idx < ixgbe->intr_cnt; v_idx++) {
5789 		vect = &ixgbe->vect_map[v_idx];
5790 
5791 		/*
5792 		 * For each rx ring bit set
5793 		 */
5794 		r_idx = bt_getlowbit(vect->rx_map, 0,
5795 		    (ixgbe->num_rx_rings - 1));
5796 
5797 		while (r_idx >= 0) {
5798 			hw_index = ixgbe->rx_rings[r_idx].hw_index;
5799 			ixgbe_setup_ivar(ixgbe, hw_index, v_idx, 0);
5800 			r_idx = bt_getlowbit(vect->rx_map, (r_idx + 1),
5801 			    (ixgbe->num_rx_rings - 1));
5802 		}
5803 
5804 		/*
5805 		 * For each tx ring bit set
5806 		 */
5807 		r_idx = bt_getlowbit(vect->tx_map, 0,
5808 		    (ixgbe->num_tx_rings - 1));
5809 
5810 		while (r_idx >= 0) {
5811 			ixgbe_setup_ivar(ixgbe, r_idx, v_idx, 1);
5812 			r_idx = bt_getlowbit(vect->tx_map, (r_idx + 1),
5813 			    (ixgbe->num_tx_rings - 1));
5814 		}
5815 	}
5816 }
5817 
5818 /*
5819  * ixgbe_rem_intr_handlers - Remove the interrupt handlers.
5820  */
5821 static void
5822 ixgbe_rem_intr_handlers(ixgbe_t *ixgbe)
5823 {
5824 	int i;
5825 	int rc;
5826 
5827 	for (i = 0; i < ixgbe->intr_cnt; i++) {
5828 		rc = ddi_intr_remove_handler(ixgbe->htable[i]);
5829 		if (rc != DDI_SUCCESS) {
5830 			IXGBE_DEBUGLOG_1(ixgbe,
5831 			    "Remove intr handler failed: %d", rc);
5832 		}
5833 	}
5834 }
5835 
5836 /*
5837  * ixgbe_rem_intrs - Remove the allocated interrupts.
5838  */
5839 static void
5840 ixgbe_rem_intrs(ixgbe_t *ixgbe)
5841 {
5842 	int i;
5843 	int rc;
5844 
5845 	for (i = 0; i < ixgbe->intr_cnt; i++) {
5846 		rc = ddi_intr_free(ixgbe->htable[i]);
5847 		if (rc != DDI_SUCCESS) {
5848 			IXGBE_DEBUGLOG_1(ixgbe,
5849 			    "Free intr failed: %d", rc);
5850 		}
5851 	}
5852 
5853 	kmem_free(ixgbe->htable, ixgbe->intr_size);
5854 	ixgbe->htable = NULL;
5855 }
5856 
5857 /*
5858  * ixgbe_enable_intrs - Enable all the ddi interrupts.
5859  */
5860 static int
5861 ixgbe_enable_intrs(ixgbe_t *ixgbe)
5862 {
5863 	int i;
5864 	int rc;
5865 
5866 	/*
5867 	 * Enable interrupts
5868 	 */
5869 	if (ixgbe->intr_cap & DDI_INTR_FLAG_BLOCK) {
5870 		/*
5871 		 * Call ddi_intr_block_enable() for MSI
5872 		 */
5873 		rc = ddi_intr_block_enable(ixgbe->htable, ixgbe->intr_cnt);
5874 		if (rc != DDI_SUCCESS) {
5875 			ixgbe_log(ixgbe,
5876 			    "Enable block intr failed: %d", rc);
5877 			return (IXGBE_FAILURE);
5878 		}
5879 	} else {
5880 		/*
5881 		 * Call ddi_intr_enable() for Legacy/MSI non block enable
5882 		 */
5883 		for (i = 0; i < ixgbe->intr_cnt; i++) {
5884 			rc = ddi_intr_enable(ixgbe->htable[i]);
5885 			if (rc != DDI_SUCCESS) {
5886 				ixgbe_log(ixgbe,
5887 				    "Enable intr failed: %d", rc);
5888 				return (IXGBE_FAILURE);
5889 			}
5890 		}
5891 	}
5892 
5893 	return (IXGBE_SUCCESS);
5894 }
5895 
5896 /*
5897  * ixgbe_disable_intrs - Disable all the interrupts.
5898  */
5899 static int
5900 ixgbe_disable_intrs(ixgbe_t *ixgbe)
5901 {
5902 	int i;
5903 	int rc;
5904 
5905 	/*
5906 	 * Disable all interrupts
5907 	 */
5908 	if (ixgbe->intr_cap & DDI_INTR_FLAG_BLOCK) {
5909 		rc = ddi_intr_block_disable(ixgbe->htable, ixgbe->intr_cnt);
5910 		if (rc != DDI_SUCCESS) {
5911 			ixgbe_log(ixgbe,
5912 			    "Disable block intr failed: %d", rc);
5913 			return (IXGBE_FAILURE);
5914 		}
5915 	} else {
5916 		for (i = 0; i < ixgbe->intr_cnt; i++) {
5917 			rc = ddi_intr_disable(ixgbe->htable[i]);
5918 			if (rc != DDI_SUCCESS) {
5919 				ixgbe_log(ixgbe,
5920 				    "Disable intr failed: %d", rc);
5921 				return (IXGBE_FAILURE);
5922 			}
5923 		}
5924 	}
5925 
5926 	return (IXGBE_SUCCESS);
5927 }
5928 
5929 /*
5930  * ixgbe_get_hw_state - Get and save parameters related to adapter hardware.
5931  */
5932 static void
5933 ixgbe_get_hw_state(ixgbe_t *ixgbe)
5934 {
5935 	struct ixgbe_hw *hw = &ixgbe->hw;
5936 	ixgbe_link_speed speed = 0;
5937 	boolean_t link_up = B_FALSE;
5938 	uint32_t pcs1g_anlp = 0;
5939 
5940 	ASSERT(mutex_owned(&ixgbe->gen_lock));
5941 	ixgbe->param_lp_1000fdx_cap = 0;
5942 	ixgbe->param_lp_100fdx_cap  = 0;
5943 
5944 	/* check for link, don't wait */
5945 	(void) ixgbe_check_link(hw, &speed, &link_up, B_FALSE);
5946 
5947 	/*
5948 	 * Update the observed Link Partner's capabilities. Not all adapters
5949 	 * can provide full information on the LP's capable speeds, so we
5950 	 * provide what we can.
5951 	 */
5952 	if (link_up) {
5953 		pcs1g_anlp = IXGBE_READ_REG(hw, IXGBE_PCS1GANLP);
5954 
5955 		ixgbe->param_lp_1000fdx_cap =
5956 		    (pcs1g_anlp & IXGBE_PCS1GANLP_LPFD) ? 1 : 0;
5957 		ixgbe->param_lp_100fdx_cap =
5958 		    (pcs1g_anlp & IXGBE_PCS1GANLP_LPFD) ? 1 : 0;
5959 	}
5960 
5961 	/*
5962 	 * Update GLD's notion of the adapter's currently advertised speeds.
5963 	 * Since the common code doesn't always record the current autonegotiate
5964 	 * settings in the phy struct for all parts (specifically, adapters with
5965 	 * SFPs) we first test to see if it is 0, and if so, we fall back to
5966 	 * using the adapter's speed capabilities which we saved during instance
5967 	 * init in ixgbe_init_params().
5968 	 *
5969 	 * Adapters with SFPs will always be shown as advertising all of their
5970 	 * supported speeds, and adapters with baseT PHYs (where the phy struct
5971 	 * is maintained by the common code) will always have a factual view of
5972 	 * their currently-advertised speeds. In the case of SFPs, this is
5973 	 * acceptable as we default to advertising all speeds that the adapter
5974 	 * claims to support, and those properties are immutable; unlike on
5975 	 * baseT (copper) PHYs, where speeds can be enabled or disabled at will.
5976 	 */
5977 	speed = hw->phy.autoneg_advertised;
5978 	if (speed == 0)
5979 		speed = ixgbe->speeds_supported;
5980 
5981 	ixgbe->param_adv_10000fdx_cap =
5982 	    (speed & IXGBE_LINK_SPEED_10GB_FULL) ? 1 : 0;
5983 	ixgbe->param_adv_5000fdx_cap =
5984 	    (speed & IXGBE_LINK_SPEED_5GB_FULL) ? 1 : 0;
5985 	ixgbe->param_adv_2500fdx_cap =
5986 	    (speed & IXGBE_LINK_SPEED_2_5GB_FULL) ? 1 : 0;
5987 	ixgbe->param_adv_1000fdx_cap =
5988 	    (speed & IXGBE_LINK_SPEED_1GB_FULL) ? 1 : 0;
5989 	ixgbe->param_adv_100fdx_cap =
5990 	    (speed & IXGBE_LINK_SPEED_100_FULL) ? 1 : 0;
5991 }
5992 
5993 /*
5994  * ixgbe_get_driver_control - Notify that driver is in control of device.
5995  */
5996 static void
5997 ixgbe_get_driver_control(struct ixgbe_hw *hw)
5998 {
5999 	uint32_t ctrl_ext;
6000 
6001 	/*
6002 	 * Notify firmware that driver is in control of device
6003 	 */
6004 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
6005 	ctrl_ext |= IXGBE_CTRL_EXT_DRV_LOAD;
6006 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
6007 }
6008 
6009 /*
6010  * ixgbe_release_driver_control - Notify that driver is no longer in control
6011  * of device.
6012  */
6013 static void
6014 ixgbe_release_driver_control(struct ixgbe_hw *hw)
6015 {
6016 	uint32_t ctrl_ext;
6017 
6018 	/*
6019 	 * Notify firmware that driver is no longer in control of device
6020 	 */
6021 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
6022 	ctrl_ext &= ~IXGBE_CTRL_EXT_DRV_LOAD;
6023 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
6024 }
6025 
6026 /*
6027  * ixgbe_atomic_reserve - Atomic decrease operation.
6028  */
6029 int
6030 ixgbe_atomic_reserve(uint32_t *count_p, uint32_t n)
6031 {
6032 	uint32_t oldval;
6033 	uint32_t newval;
6034 
6035 	/*
6036 	 * ATOMICALLY
6037 	 */
6038 	do {
6039 		oldval = *count_p;
6040 		if (oldval < n)
6041 			return (-1);
6042 		newval = oldval - n;
6043 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
6044 
6045 	return (newval);
6046 }
6047 
6048 /*
6049  * ixgbe_mc_table_itr - Traverse the entries in the multicast table.
6050  */
6051 static uint8_t *
6052 ixgbe_mc_table_itr(struct ixgbe_hw *hw, uint8_t **upd_ptr, uint32_t *vmdq)
6053 {
6054 	uint8_t *addr = *upd_ptr;
6055 	uint8_t *new_ptr;
6056 
6057 	_NOTE(ARGUNUSED(hw));
6058 	_NOTE(ARGUNUSED(vmdq));
6059 
6060 	new_ptr = addr + IXGBE_ETH_LENGTH_OF_ADDRESS;
6061 	*upd_ptr = new_ptr;
6062 	return (addr);
6063 }
6064 
6065 /*
6066  * FMA support
6067  */
6068 int
6069 ixgbe_check_acc_handle(ddi_acc_handle_t handle)
6070 {
6071 	ddi_fm_error_t de;
6072 
6073 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
6074 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
6075 	return (de.fme_status);
6076 }
6077 
6078 int
6079 ixgbe_check_dma_handle(ddi_dma_handle_t handle)
6080 {
6081 	ddi_fm_error_t de;
6082 
6083 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
6084 	return (de.fme_status);
6085 }
6086 
6087 /*
6088  * ixgbe_fm_error_cb - The IO fault service error handling callback function.
6089  */
6090 static int
6091 ixgbe_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
6092 {
6093 	_NOTE(ARGUNUSED(impl_data));
6094 	/*
6095 	 * as the driver can always deal with an error in any dma or
6096 	 * access handle, we can just return the fme_status value.
6097 	 */
6098 	pci_ereport_post(dip, err, NULL);
6099 	return (err->fme_status);
6100 }
6101 
6102 static void
6103 ixgbe_fm_init(ixgbe_t *ixgbe)
6104 {
6105 	ddi_iblock_cookie_t iblk;
6106 	int fma_dma_flag;
6107 
6108 	/*
6109 	 * Only register with IO Fault Services if we have some capability
6110 	 */
6111 	if (ixgbe->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
6112 		ixgbe_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
6113 	} else {
6114 		ixgbe_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
6115 	}
6116 
6117 	if (ixgbe->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
6118 		fma_dma_flag = 1;
6119 	} else {
6120 		fma_dma_flag = 0;
6121 	}
6122 
6123 	ixgbe_set_fma_flags(fma_dma_flag);
6124 
6125 	if (ixgbe->fm_capabilities) {
6126 
6127 		/*
6128 		 * Register capabilities with IO Fault Services
6129 		 */
6130 		ddi_fm_init(ixgbe->dip, &ixgbe->fm_capabilities, &iblk);
6131 
6132 		/*
6133 		 * Initialize pci ereport capabilities if ereport capable
6134 		 */
6135 		if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities) ||
6136 		    DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
6137 			pci_ereport_setup(ixgbe->dip);
6138 
6139 		/*
6140 		 * Register error callback if error callback capable
6141 		 */
6142 		if (DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
6143 			ddi_fm_handler_register(ixgbe->dip,
6144 			    ixgbe_fm_error_cb, (void*) ixgbe);
6145 	}
6146 }
6147 
6148 static void
6149 ixgbe_fm_fini(ixgbe_t *ixgbe)
6150 {
6151 	/*
6152 	 * Only unregister FMA capabilities if they are registered
6153 	 */
6154 	if (ixgbe->fm_capabilities) {
6155 
6156 		/*
6157 		 * Release any resources allocated by pci_ereport_setup()
6158 		 */
6159 		if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities) ||
6160 		    DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
6161 			pci_ereport_teardown(ixgbe->dip);
6162 
6163 		/*
6164 		 * Un-register error callback if error callback capable
6165 		 */
6166 		if (DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
6167 			ddi_fm_handler_unregister(ixgbe->dip);
6168 
6169 		/*
6170 		 * Unregister from IO Fault Service
6171 		 */
6172 		ddi_fm_fini(ixgbe->dip);
6173 	}
6174 }
6175 
6176 void
6177 ixgbe_fm_ereport(ixgbe_t *ixgbe, char *detail)
6178 {
6179 	uint64_t ena;
6180 	char buf[FM_MAX_CLASS];
6181 
6182 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
6183 	ena = fm_ena_generate(0, FM_ENA_FMT1);
6184 	if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities)) {
6185 		ddi_fm_ereport_post(ixgbe->dip, buf, ena, DDI_NOSLEEP,
6186 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
6187 	}
6188 }
6189 
6190 static int
6191 ixgbe_ring_start(mac_ring_driver_t rh, uint64_t mr_gen_num)
6192 {
6193 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)rh;
6194 
6195 	mutex_enter(&rx_ring->rx_lock);
6196 	rx_ring->ring_gen_num = mr_gen_num;
6197 	mutex_exit(&rx_ring->rx_lock);
6198 	return (0);
6199 }
6200 
6201 /*
6202  * Get the global ring index by a ring index within a group.
6203  */
6204 static int
6205 ixgbe_get_rx_ring_index(ixgbe_t *ixgbe, int gindex, int rindex)
6206 {
6207 	ixgbe_rx_ring_t *rx_ring;
6208 	int i;
6209 
6210 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
6211 		rx_ring = &ixgbe->rx_rings[i];
6212 		if (rx_ring->group_index == gindex)
6213 			rindex--;
6214 		if (rindex < 0)
6215 			return (i);
6216 	}
6217 
6218 	return (-1);
6219 }
6220 
6221 /*
6222  * Callback funtion for MAC layer to register all rings.
6223  */
6224 /* ARGSUSED */
6225 void
6226 ixgbe_fill_ring(void *arg, mac_ring_type_t rtype, const int group_index,
6227     const int ring_index, mac_ring_info_t *infop, mac_ring_handle_t rh)
6228 {
6229 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
6230 	mac_intr_t *mintr = &infop->mri_intr;
6231 
6232 	switch (rtype) {
6233 	case MAC_RING_TYPE_RX: {
6234 		/*
6235 		 * 'index' is the ring index within the group.
6236 		 * Need to get the global ring index by searching in groups.
6237 		 */
6238 		int global_ring_index = ixgbe_get_rx_ring_index(
6239 		    ixgbe, group_index, ring_index);
6240 
6241 		ASSERT(global_ring_index >= 0);
6242 
6243 		ixgbe_rx_ring_t *rx_ring = &ixgbe->rx_rings[global_ring_index];
6244 		rx_ring->ring_handle = rh;
6245 
6246 		infop->mri_driver = (mac_ring_driver_t)rx_ring;
6247 		infop->mri_start = ixgbe_ring_start;
6248 		infop->mri_stop = NULL;
6249 		infop->mri_poll = ixgbe_ring_rx_poll;
6250 		infop->mri_stat = ixgbe_rx_ring_stat;
6251 
6252 		mintr->mi_handle = (mac_intr_handle_t)rx_ring;
6253 		mintr->mi_enable = ixgbe_rx_ring_intr_enable;
6254 		mintr->mi_disable = ixgbe_rx_ring_intr_disable;
6255 		if (ixgbe->intr_type &
6256 		    (DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI)) {
6257 			mintr->mi_ddi_handle =
6258 			    ixgbe->htable[rx_ring->intr_vector];
6259 		}
6260 
6261 		break;
6262 	}
6263 	case MAC_RING_TYPE_TX: {
6264 		ASSERT(group_index == -1);
6265 		ASSERT(ring_index < ixgbe->num_tx_rings);
6266 
6267 		ixgbe_tx_ring_t *tx_ring = &ixgbe->tx_rings[ring_index];
6268 		tx_ring->ring_handle = rh;
6269 
6270 		infop->mri_driver = (mac_ring_driver_t)tx_ring;
6271 		infop->mri_start = NULL;
6272 		infop->mri_stop = NULL;
6273 		infop->mri_tx = ixgbe_ring_tx;
6274 		infop->mri_stat = ixgbe_tx_ring_stat;
6275 		if (ixgbe->intr_type &
6276 		    (DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI)) {
6277 			mintr->mi_ddi_handle =
6278 			    ixgbe->htable[tx_ring->intr_vector];
6279 		}
6280 		break;
6281 	}
6282 	default:
6283 		break;
6284 	}
6285 }
6286 
6287 /*
6288  * Callback funtion for MAC layer to register all groups.
6289  */
6290 void
6291 ixgbe_fill_group(void *arg, mac_ring_type_t rtype, const int index,
6292     mac_group_info_t *infop, mac_group_handle_t gh)
6293 {
6294 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
6295 	struct ixgbe_hw *hw = &ixgbe->hw;
6296 
6297 	switch (rtype) {
6298 	case MAC_RING_TYPE_RX: {
6299 		ixgbe_rx_group_t *rx_group;
6300 
6301 		rx_group = &ixgbe->rx_groups[index];
6302 		rx_group->group_handle = gh;
6303 
6304 		infop->mgi_driver = (mac_group_driver_t)rx_group;
6305 		infop->mgi_start = NULL;
6306 		infop->mgi_stop = NULL;
6307 		infop->mgi_addmac = ixgbe_addmac;
6308 		infop->mgi_remmac = ixgbe_remmac;
6309 
6310 		if ((ixgbe->classify_mode == IXGBE_CLASSIFY_VMDQ ||
6311 		    ixgbe->classify_mode == IXGBE_CLASSIFY_VMDQ_RSS) &&
6312 		    (hw->mac.type == ixgbe_mac_82599EB ||
6313 		    hw->mac.type == ixgbe_mac_X540 ||
6314 		    hw->mac.type == ixgbe_mac_X550 ||
6315 		    hw->mac.type == ixgbe_mac_X550EM_x)) {
6316 			infop->mgi_addvlan = ixgbe_addvlan;
6317 			infop->mgi_remvlan = ixgbe_remvlan;
6318 		} else {
6319 			infop->mgi_addvlan = NULL;
6320 			infop->mgi_remvlan = NULL;
6321 		}
6322 
6323 		infop->mgi_count = (ixgbe->num_rx_rings / ixgbe->num_rx_groups);
6324 
6325 		break;
6326 	}
6327 	case MAC_RING_TYPE_TX:
6328 		break;
6329 	default:
6330 		break;
6331 	}
6332 }
6333 
6334 /*
6335  * Enable interrupt on the specificed rx ring.
6336  */
6337 int
6338 ixgbe_rx_ring_intr_enable(mac_intr_handle_t intrh)
6339 {
6340 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)intrh;
6341 	ixgbe_t *ixgbe = rx_ring->ixgbe;
6342 	int r_idx = rx_ring->index;
6343 	int hw_r_idx = rx_ring->hw_index;
6344 	int v_idx = rx_ring->intr_vector;
6345 
6346 	mutex_enter(&ixgbe->gen_lock);
6347 	if (ixgbe->ixgbe_state & IXGBE_INTR_ADJUST) {
6348 		mutex_exit(&ixgbe->gen_lock);
6349 		/*
6350 		 * Simply return 0.
6351 		 * Interrupts are being adjusted. ixgbe_intr_adjust()
6352 		 * will eventually re-enable the interrupt when it's
6353 		 * done with the adjustment.
6354 		 */
6355 		return (0);
6356 	}
6357 
6358 	/*
6359 	 * To enable interrupt by setting the VAL bit of given interrupt
6360 	 * vector allocation register (IVAR).
6361 	 */
6362 	ixgbe_enable_ivar(ixgbe, hw_r_idx, 0);
6363 
6364 	BT_SET(ixgbe->vect_map[v_idx].rx_map, r_idx);
6365 
6366 	/*
6367 	 * Trigger a Rx interrupt on this ring
6368 	 */
6369 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_EICS, (1 << v_idx));
6370 	IXGBE_WRITE_FLUSH(&ixgbe->hw);
6371 
6372 	mutex_exit(&ixgbe->gen_lock);
6373 
6374 	return (0);
6375 }
6376 
6377 /*
6378  * Disable interrupt on the specificed rx ring.
6379  */
6380 int
6381 ixgbe_rx_ring_intr_disable(mac_intr_handle_t intrh)
6382 {
6383 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)intrh;
6384 	ixgbe_t *ixgbe = rx_ring->ixgbe;
6385 	int r_idx = rx_ring->index;
6386 	int hw_r_idx = rx_ring->hw_index;
6387 	int v_idx = rx_ring->intr_vector;
6388 
6389 	mutex_enter(&ixgbe->gen_lock);
6390 	if (ixgbe->ixgbe_state & IXGBE_INTR_ADJUST) {
6391 		mutex_exit(&ixgbe->gen_lock);
6392 		/*
6393 		 * Simply return 0.
6394 		 * In the rare case where an interrupt is being
6395 		 * disabled while interrupts are being adjusted,
6396 		 * we don't fail the operation. No interrupts will
6397 		 * be generated while they are adjusted, and
6398 		 * ixgbe_intr_adjust() will cause the interrupts
6399 		 * to be re-enabled once it completes. Note that
6400 		 * in this case, packets may be delivered to the
6401 		 * stack via interrupts before xgbe_rx_ring_intr_enable()
6402 		 * is called again. This is acceptable since interrupt
6403 		 * adjustment is infrequent, and the stack will be
6404 		 * able to handle these packets.
6405 		 */
6406 		return (0);
6407 	}
6408 
6409 	/*
6410 	 * To disable interrupt by clearing the VAL bit of given interrupt
6411 	 * vector allocation register (IVAR).
6412 	 */
6413 	ixgbe_disable_ivar(ixgbe, hw_r_idx, 0);
6414 
6415 	BT_CLEAR(ixgbe->vect_map[v_idx].rx_map, r_idx);
6416 
6417 	mutex_exit(&ixgbe->gen_lock);
6418 
6419 	return (0);
6420 }
6421 
6422 static ixgbe_vlan_t *
6423 ixgbe_find_vlan(ixgbe_rx_group_t *rx_group, uint16_t vid)
6424 {
6425 	for (ixgbe_vlan_t *vlp = list_head(&rx_group->vlans); vlp != NULL;
6426 	    vlp = list_next(&rx_group->vlans, vlp)) {
6427 		if (vlp->ixvl_vid == vid)
6428 			return (vlp);
6429 	}
6430 
6431 	return (NULL);
6432 }
6433 
6434 /*
6435  * Attempt to use a VLAN HW filter for this group. If the group is
6436  * interested in untagged packets then set AUPE only. If the group is
6437  * the default then only set the VFTA. Leave the VLVF slots open for
6438  * reserved groups to guarantee their use of HW filtering.
6439  */
6440 static int
6441 ixgbe_addvlan(mac_group_driver_t gdriver, uint16_t vid)
6442 {
6443 	ixgbe_rx_group_t	*rx_group = (ixgbe_rx_group_t *)gdriver;
6444 	ixgbe_t			*ixgbe = rx_group->ixgbe;
6445 	struct ixgbe_hw		*hw = &ixgbe->hw;
6446 	ixgbe_vlan_t		*vlp;
6447 	int			ret;
6448 	boolean_t		is_def_grp;
6449 
6450 	mutex_enter(&ixgbe->gen_lock);
6451 
6452 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
6453 		mutex_exit(&ixgbe->gen_lock);
6454 		return (ECANCELED);
6455 	}
6456 
6457 	/*
6458 	 * Let's be sure VLAN filtering is enabled.
6459 	 */
6460 	VERIFY3B(ixgbe->vlft_enabled, ==, B_TRUE);
6461 	is_def_grp = (rx_group->index == ixgbe->rx_def_group);
6462 
6463 	/*
6464 	 * VLAN filtering is enabled but we want to receive untagged
6465 	 * traffic on this group -- set the AUPE bit on the group and
6466 	 * leave the VLAN tables alone.
6467 	 */
6468 	if (vid == MAC_VLAN_UNTAGGED) {
6469 		/*
6470 		 * We never enable AUPE on the default group; it is
6471 		 * redundant. Untagged traffic which passes L2
6472 		 * filtering is delivered to the default group if no
6473 		 * other group is interested.
6474 		 */
6475 		if (!is_def_grp) {
6476 			uint32_t vml2flt;
6477 
6478 			vml2flt = IXGBE_READ_REG(hw,
6479 			    IXGBE_VMOLR(rx_group->index));
6480 			vml2flt |= IXGBE_VMOLR_AUPE;
6481 			IXGBE_WRITE_REG(hw, IXGBE_VMOLR(rx_group->index),
6482 			    vml2flt);
6483 			rx_group->aupe = B_TRUE;
6484 		}
6485 
6486 		mutex_exit(&ixgbe->gen_lock);
6487 		return (0);
6488 	}
6489 
6490 	vlp = ixgbe_find_vlan(rx_group, vid);
6491 	if (vlp != NULL) {
6492 		/* Only the default group supports multiple clients. */
6493 		VERIFY3B(is_def_grp, ==, B_TRUE);
6494 		vlp->ixvl_refs++;
6495 		mutex_exit(&ixgbe->gen_lock);
6496 		return (0);
6497 	}
6498 
6499 	/*
6500 	 * The default group doesn't require a VLVF entry, only a VFTA
6501 	 * entry. All traffic passing L2 filtering (MPSAR + VFTA) is
6502 	 * delivered to the default group if no other group is
6503 	 * interested. The fourth argument, vlvf_bypass, tells the
6504 	 * ixgbe common code to avoid using a VLVF slot if one isn't
6505 	 * already allocated to this VLAN.
6506 	 *
6507 	 * This logic is meant to reserve VLVF slots for use by
6508 	 * reserved groups: guaranteeing their use of HW filtering.
6509 	 */
6510 	ret = ixgbe_set_vfta(hw, vid, rx_group->index, B_TRUE, is_def_grp);
6511 
6512 	if (ret == IXGBE_SUCCESS) {
6513 		vlp = kmem_zalloc(sizeof (ixgbe_vlan_t), KM_SLEEP);
6514 		vlp->ixvl_vid = vid;
6515 		vlp->ixvl_refs = 1;
6516 		list_insert_tail(&rx_group->vlans, vlp);
6517 		mutex_exit(&ixgbe->gen_lock);
6518 		return (0);
6519 	}
6520 
6521 	/*
6522 	 * We should actually never return ENOSPC because we've set
6523 	 * things up so that every reserved group is guaranteed to
6524 	 * have a VLVF slot.
6525 	 */
6526 	if (ret == IXGBE_ERR_PARAM)
6527 		ret = EINVAL;
6528 	else if (ret == IXGBE_ERR_NO_SPACE)
6529 		ret = ENOSPC;
6530 	else
6531 		ret = EIO;
6532 
6533 	mutex_exit(&ixgbe->gen_lock);
6534 	return (ret);
6535 }
6536 
6537 /*
6538  * Attempt to remove the VLAN HW filter associated with this group. If
6539  * we are removing a HW filter for the default group then we know only
6540  * the VFTA was set (VLVF is reserved for non-default/reserved
6541  * groups). If the group wishes to stop receiving untagged traffic
6542  * then clear the AUPE but leave the VLAN filters alone.
6543  */
6544 static int
6545 ixgbe_remvlan(mac_group_driver_t gdriver, uint16_t vid)
6546 {
6547 	ixgbe_rx_group_t	*rx_group = (ixgbe_rx_group_t *)gdriver;
6548 	ixgbe_t			*ixgbe = rx_group->ixgbe;
6549 	struct ixgbe_hw		*hw = &ixgbe->hw;
6550 	int			ret;
6551 	ixgbe_vlan_t		*vlp;
6552 	boolean_t		is_def_grp;
6553 
6554 	mutex_enter(&ixgbe->gen_lock);
6555 
6556 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
6557 		mutex_exit(&ixgbe->gen_lock);
6558 		return (ECANCELED);
6559 	}
6560 
6561 	is_def_grp = (rx_group->index == ixgbe->rx_def_group);
6562 
6563 	/* See the AUPE comment in ixgbe_addvlan(). */
6564 	if (vid == MAC_VLAN_UNTAGGED) {
6565 		if (!is_def_grp) {
6566 			uint32_t vml2flt;
6567 
6568 			vml2flt = IXGBE_READ_REG(hw,
6569 			    IXGBE_VMOLR(rx_group->index));
6570 			vml2flt &= ~IXGBE_VMOLR_AUPE;
6571 			IXGBE_WRITE_REG(hw,
6572 			    IXGBE_VMOLR(rx_group->index), vml2flt);
6573 			rx_group->aupe = B_FALSE;
6574 		}
6575 		mutex_exit(&ixgbe->gen_lock);
6576 		return (0);
6577 	}
6578 
6579 	vlp = ixgbe_find_vlan(rx_group, vid);
6580 	if (vlp == NULL) {
6581 		mutex_exit(&ixgbe->gen_lock);
6582 		return (ENOENT);
6583 	}
6584 
6585 	/*
6586 	 * See the comment in ixgbe_addvlan() about is_def_grp and
6587 	 * vlvf_bypass.
6588 	 */
6589 	if (vlp->ixvl_refs == 1) {
6590 		ret = ixgbe_set_vfta(hw, vid, rx_group->index, B_FALSE,
6591 		    is_def_grp);
6592 	} else {
6593 		/*
6594 		 * Only the default group can have multiple clients.
6595 		 * If there is more than one client, leave the
6596 		 * VFTA[vid] bit alone.
6597 		 */
6598 		VERIFY3B(is_def_grp, ==, B_TRUE);
6599 		VERIFY3U(vlp->ixvl_refs, >, 1);
6600 		vlp->ixvl_refs--;
6601 		mutex_exit(&ixgbe->gen_lock);
6602 		return (0);
6603 	}
6604 
6605 	if (ret != IXGBE_SUCCESS) {
6606 		mutex_exit(&ixgbe->gen_lock);
6607 		/* IXGBE_ERR_PARAM should be the only possible error here. */
6608 		if (ret == IXGBE_ERR_PARAM)
6609 			return (EINVAL);
6610 		else
6611 			return (EIO);
6612 	}
6613 
6614 	VERIFY3U(vlp->ixvl_refs, ==, 1);
6615 	vlp->ixvl_refs = 0;
6616 	list_remove(&rx_group->vlans, vlp);
6617 	kmem_free(vlp, sizeof (ixgbe_vlan_t));
6618 
6619 	/*
6620 	 * Calling ixgbe_set_vfta() on a non-default group may have
6621 	 * cleared the VFTA[vid] bit even though the default group
6622 	 * still has clients using the vid. This happens because the
6623 	 * ixgbe common code doesn't ref count the use of VLANs. Check
6624 	 * for any use of vid on the default group and make sure the
6625 	 * VFTA[vid] bit is set. This operation is idempotent: setting
6626 	 * VFTA[vid] to true if already true won't hurt anything.
6627 	 */
6628 	if (!is_def_grp) {
6629 		ixgbe_rx_group_t *defgrp;
6630 
6631 		defgrp = &ixgbe->rx_groups[ixgbe->rx_def_group];
6632 		vlp = ixgbe_find_vlan(defgrp, vid);
6633 		if (vlp != NULL) {
6634 			/* This shouldn't fail, but if it does return EIO. */
6635 			ret = ixgbe_set_vfta(hw, vid, rx_group->index, B_TRUE,
6636 			    B_TRUE);
6637 			if (ret != IXGBE_SUCCESS) {
6638 				mutex_exit(&ixgbe->gen_lock);
6639 				return (EIO);
6640 			}
6641 		}
6642 	}
6643 
6644 	mutex_exit(&ixgbe->gen_lock);
6645 	return (0);
6646 }
6647 
6648 /*
6649  * Add a mac address.
6650  */
6651 static int
6652 ixgbe_addmac(void *arg, const uint8_t *mac_addr)
6653 {
6654 	ixgbe_rx_group_t *rx_group = (ixgbe_rx_group_t *)arg;
6655 	ixgbe_t *ixgbe = rx_group->ixgbe;
6656 	struct ixgbe_hw *hw = &ixgbe->hw;
6657 	int slot, i;
6658 
6659 	mutex_enter(&ixgbe->gen_lock);
6660 
6661 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
6662 		mutex_exit(&ixgbe->gen_lock);
6663 		return (ECANCELED);
6664 	}
6665 
6666 	if (ixgbe->unicst_avail == 0) {
6667 		/* no slots available */
6668 		mutex_exit(&ixgbe->gen_lock);
6669 		return (ENOSPC);
6670 	}
6671 
6672 	/*
6673 	 * The first ixgbe->num_rx_groups slots are reserved for each respective
6674 	 * group. The rest slots are shared by all groups. While adding a
6675 	 * MAC address, reserved slots are firstly checked then the shared
6676 	 * slots are searched.
6677 	 */
6678 	slot = -1;
6679 	if (ixgbe->unicst_addr[rx_group->index].mac.set == 1) {
6680 		for (i = ixgbe->num_rx_groups; i < ixgbe->unicst_total; i++) {
6681 			if (ixgbe->unicst_addr[i].mac.set == 0) {
6682 				slot = i;
6683 				break;
6684 			}
6685 		}
6686 	} else {
6687 		slot = rx_group->index;
6688 	}
6689 
6690 	if (slot == -1) {
6691 		/* no slots available */
6692 		mutex_exit(&ixgbe->gen_lock);
6693 		return (ENOSPC);
6694 	}
6695 
6696 	bcopy(mac_addr, ixgbe->unicst_addr[slot].mac.addr, ETHERADDRL);
6697 	(void) ixgbe_set_rar(hw, slot, ixgbe->unicst_addr[slot].mac.addr,
6698 	    rx_group->index, IXGBE_RAH_AV);
6699 	ixgbe->unicst_addr[slot].mac.set = 1;
6700 	ixgbe->unicst_addr[slot].mac.group_index = rx_group->index;
6701 	ixgbe->unicst_avail--;
6702 
6703 	mutex_exit(&ixgbe->gen_lock);
6704 
6705 	return (0);
6706 }
6707 
6708 /*
6709  * Remove a mac address.
6710  */
6711 static int
6712 ixgbe_remmac(void *arg, const uint8_t *mac_addr)
6713 {
6714 	ixgbe_rx_group_t *rx_group = (ixgbe_rx_group_t *)arg;
6715 	ixgbe_t *ixgbe = rx_group->ixgbe;
6716 	struct ixgbe_hw *hw = &ixgbe->hw;
6717 	int slot;
6718 
6719 	mutex_enter(&ixgbe->gen_lock);
6720 
6721 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
6722 		mutex_exit(&ixgbe->gen_lock);
6723 		return (ECANCELED);
6724 	}
6725 
6726 	slot = ixgbe_unicst_find(ixgbe, mac_addr);
6727 	if (slot == -1) {
6728 		mutex_exit(&ixgbe->gen_lock);
6729 		return (EINVAL);
6730 	}
6731 
6732 	if (ixgbe->unicst_addr[slot].mac.set == 0) {
6733 		mutex_exit(&ixgbe->gen_lock);
6734 		return (EINVAL);
6735 	}
6736 
6737 	bzero(ixgbe->unicst_addr[slot].mac.addr, ETHERADDRL);
6738 	(void) ixgbe_clear_rar(hw, slot);
6739 	ixgbe->unicst_addr[slot].mac.set = 0;
6740 	ixgbe->unicst_avail++;
6741 
6742 	mutex_exit(&ixgbe->gen_lock);
6743 
6744 	return (0);
6745 }
6746 
6747 static int
6748 ixgbe_ufm_fill_image(ddi_ufm_handle_t *ufmh, void *arg, uint_t imgno,
6749     ddi_ufm_image_t *imgp)
6750 {
6751 	ixgbe_t *ixgbe = arg;
6752 	const char *type;
6753 
6754 	if (imgno != 0) {
6755 		return (EINVAL);
6756 	}
6757 
6758 	ddi_ufm_image_set_desc(imgp, "NVM");
6759 	ddi_ufm_image_set_nslots(imgp, 1);
6760 	switch (ixgbe->hw.eeprom.type) {
6761 	case ixgbe_eeprom_spi:
6762 		type = "SPI EEPROM";
6763 		break;
6764 	case ixgbe_flash:
6765 		type = "Flash";
6766 		break;
6767 	default:
6768 		type = NULL;
6769 		break;
6770 	}
6771 
6772 	if (type != NULL) {
6773 		nvlist_t *nvl;
6774 
6775 		nvl = fnvlist_alloc();
6776 		fnvlist_add_string(nvl, "image-type", type);
6777 		/*
6778 		 * The DDI takes ownership of the nvlist_t at this point.
6779 		 */
6780 		ddi_ufm_image_set_misc(imgp, nvl);
6781 	}
6782 
6783 	return (0);
6784 }
6785 
6786 static int
6787 ixgbe_ufm_fill_slot(ddi_ufm_handle_t *ufmh, void *arg, uint_t imgno,
6788     uint_t slotno, ddi_ufm_slot_t *slotp)
6789 {
6790 	ixgbe_t *ixgbe = arg;
6791 
6792 	if (imgno != 0 || slotno != 0) {
6793 		return (EINVAL);
6794 	}
6795 
6796 	/*
6797 	 * Unfortunately there is no generic versioning in the ixgbe family
6798 	 * eeprom parts.
6799 	 */
6800 	ddi_ufm_slot_set_version(slotp, "unknown");
6801 	ddi_ufm_slot_set_attrs(slotp, DDI_UFM_ATTR_ACTIVE |
6802 	    DDI_UFM_ATTR_READABLE | DDI_UFM_ATTR_WRITEABLE);
6803 	ddi_ufm_slot_set_imgsize(slotp, ixgbe->hw.eeprom.word_size * 2);
6804 
6805 	return (0);
6806 }
6807 
6808 static int
6809 ixgbe_ufm_getcaps(ddi_ufm_handle_t *ufmh, void *arg, ddi_ufm_cap_t *caps)
6810 {
6811 	ixgbe_t *ixgbe = arg;
6812 
6813 	*caps = 0;
6814 	switch (ixgbe->hw.eeprom.type) {
6815 	case ixgbe_eeprom_spi:
6816 	case ixgbe_flash:
6817 		*caps |= DDI_UFM_CAP_REPORT;
6818 		if (ixgbe->hw.eeprom.ops.read_buffer != NULL) {
6819 			*caps |= DDI_UFM_CAP_READIMG;
6820 		}
6821 		break;
6822 	default:
6823 		break;
6824 	}
6825 
6826 	return (0);
6827 }
6828 
6829 static int
6830 ixgbe_ufm_readimg(ddi_ufm_handle_t *ufmh, void *arg, uint_t imgno,
6831     uint_t slotno, uint64_t len, uint64_t offset, void *buf, uint64_t *nread)
6832 {
6833 	int ret;
6834 	uint16_t wordoff, nwords, *buf16 = buf;
6835 	ixgbe_t *ixgbe = arg;
6836 	uint32_t imgsize = ixgbe->hw.eeprom.word_size * 2;
6837 
6838 	if (imgno != 0 || slotno != 0) {
6839 		return (EINVAL);
6840 	}
6841 
6842 	if (len > imgsize || offset > imgsize || len + offset > imgsize) {
6843 		return (EINVAL);
6844 	}
6845 
6846 	if (ixgbe->hw.eeprom.ops.read_buffer == NULL) {
6847 		return (ENOTSUP);
6848 	}
6849 
6850 	/*
6851 	 * Hardware provides us a means to read 16-bit words. For the time
6852 	 * being, restrict offset and length to be 2 byte aligned. We should
6853 	 * probably reduce this restriction. We could probably just use a bounce
6854 	 * buffer.
6855 	 */
6856 	if ((offset % 2) != 0 || (len % 2) != 0) {
6857 		return (EINVAL);
6858 	}
6859 
6860 	wordoff = offset >> 1;
6861 	nwords = len >> 1;
6862 	mutex_enter(&ixgbe->gen_lock);
6863 	ret = ixgbe_read_eeprom_buffer(&ixgbe->hw, wordoff, nwords, buf16);
6864 	mutex_exit(&ixgbe->gen_lock);
6865 
6866 	if (ret == 0) {
6867 		uint16_t i;
6868 		*nread = len;
6869 		for (i = 0; i < nwords; i++) {
6870 			buf16[i] = LE_16(buf16[i]);
6871 		}
6872 	} else {
6873 		ret = EIO;
6874 	}
6875 
6876 	return (ret);
6877 }
6878