xref: /illumos-gate/usr/src/uts/common/io/ixgbe/ixgbe_main.c (revision 2f7f7a62d7a3e8a2e75eb88b95bc65871b6b90cb)
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 (c) 2012, Joyent, Inc. All rights reserved.
29  * Copyright 2012 Nexenta Systems, Inc. All rights reserved.
30  * Copyright (c) 2013 Saso Kiselkov. All rights reserved.
31  * Copyright (c) 2013 OSN Online Service Nuernberg GmbH. All rights reserved.
32  * Copyright 2016 OmniTI Computer Consulting, Inc. All rights reserved.
33  */
34 
35 #include "ixgbe_sw.h"
36 
37 static char ixgbe_ident[] = "Intel 10Gb Ethernet";
38 
39 /*
40  * Local function protoypes
41  */
42 static int ixgbe_register_mac(ixgbe_t *);
43 static int ixgbe_identify_hardware(ixgbe_t *);
44 static int ixgbe_regs_map(ixgbe_t *);
45 static void ixgbe_init_properties(ixgbe_t *);
46 static int ixgbe_init_driver_settings(ixgbe_t *);
47 static void ixgbe_init_locks(ixgbe_t *);
48 static void ixgbe_destroy_locks(ixgbe_t *);
49 static int ixgbe_init(ixgbe_t *);
50 static int ixgbe_chip_start(ixgbe_t *);
51 static void ixgbe_chip_stop(ixgbe_t *);
52 static int ixgbe_reset(ixgbe_t *);
53 static void ixgbe_tx_clean(ixgbe_t *);
54 static boolean_t ixgbe_tx_drain(ixgbe_t *);
55 static boolean_t ixgbe_rx_drain(ixgbe_t *);
56 static int ixgbe_alloc_rings(ixgbe_t *);
57 static void ixgbe_free_rings(ixgbe_t *);
58 static int ixgbe_alloc_rx_data(ixgbe_t *);
59 static void ixgbe_free_rx_data(ixgbe_t *);
60 static void ixgbe_setup_rings(ixgbe_t *);
61 static void ixgbe_setup_rx(ixgbe_t *);
62 static void ixgbe_setup_tx(ixgbe_t *);
63 static void ixgbe_setup_rx_ring(ixgbe_rx_ring_t *);
64 static void ixgbe_setup_tx_ring(ixgbe_tx_ring_t *);
65 static void ixgbe_setup_rss(ixgbe_t *);
66 static void ixgbe_setup_vmdq(ixgbe_t *);
67 static void ixgbe_setup_vmdq_rss(ixgbe_t *);
68 static void ixgbe_setup_rss_table(ixgbe_t *);
69 static void ixgbe_init_unicst(ixgbe_t *);
70 static int ixgbe_unicst_find(ixgbe_t *, const uint8_t *);
71 static void ixgbe_setup_multicst(ixgbe_t *);
72 static void ixgbe_get_hw_state(ixgbe_t *);
73 static void ixgbe_setup_vmdq_rss_conf(ixgbe_t *ixgbe);
74 static void ixgbe_get_conf(ixgbe_t *);
75 static void ixgbe_init_params(ixgbe_t *);
76 static int ixgbe_get_prop(ixgbe_t *, char *, int, int, int);
77 static void ixgbe_driver_link_check(ixgbe_t *);
78 static void ixgbe_sfp_check(void *);
79 static void ixgbe_overtemp_check(void *);
80 static void ixgbe_phy_check(void *);
81 static void ixgbe_link_timer(void *);
82 static void ixgbe_local_timer(void *);
83 static void ixgbe_arm_watchdog_timer(ixgbe_t *);
84 static void ixgbe_restart_watchdog_timer(ixgbe_t *);
85 static void ixgbe_disable_adapter_interrupts(ixgbe_t *);
86 static void ixgbe_enable_adapter_interrupts(ixgbe_t *);
87 static boolean_t is_valid_mac_addr(uint8_t *);
88 static boolean_t ixgbe_stall_check(ixgbe_t *);
89 static boolean_t ixgbe_set_loopback_mode(ixgbe_t *, uint32_t);
90 static void ixgbe_set_internal_mac_loopback(ixgbe_t *);
91 static boolean_t ixgbe_find_mac_address(ixgbe_t *);
92 static int ixgbe_alloc_intrs(ixgbe_t *);
93 static int ixgbe_alloc_intr_handles(ixgbe_t *, int);
94 static int ixgbe_add_intr_handlers(ixgbe_t *);
95 static void ixgbe_map_rxring_to_vector(ixgbe_t *, int, int);
96 static void ixgbe_map_txring_to_vector(ixgbe_t *, int, int);
97 static void ixgbe_setup_ivar(ixgbe_t *, uint16_t, uint8_t, int8_t);
98 static void ixgbe_enable_ivar(ixgbe_t *, uint16_t, int8_t);
99 static void ixgbe_disable_ivar(ixgbe_t *, uint16_t, int8_t);
100 static uint32_t ixgbe_get_hw_rx_index(ixgbe_t *ixgbe, uint32_t sw_rx_index);
101 static int ixgbe_map_intrs_to_vectors(ixgbe_t *);
102 static void ixgbe_setup_adapter_vector(ixgbe_t *);
103 static void ixgbe_rem_intr_handlers(ixgbe_t *);
104 static void ixgbe_rem_intrs(ixgbe_t *);
105 static int ixgbe_enable_intrs(ixgbe_t *);
106 static int ixgbe_disable_intrs(ixgbe_t *);
107 static uint_t ixgbe_intr_legacy(void *, void *);
108 static uint_t ixgbe_intr_msi(void *, void *);
109 static uint_t ixgbe_intr_msix(void *, void *);
110 static void ixgbe_intr_rx_work(ixgbe_rx_ring_t *);
111 static void ixgbe_intr_tx_work(ixgbe_tx_ring_t *);
112 static void ixgbe_intr_other_work(ixgbe_t *, uint32_t);
113 static void ixgbe_get_driver_control(struct ixgbe_hw *);
114 static int ixgbe_addmac(void *, const uint8_t *);
115 static int ixgbe_remmac(void *, const uint8_t *);
116 static void ixgbe_release_driver_control(struct ixgbe_hw *);
117 
118 static int ixgbe_attach(dev_info_t *, ddi_attach_cmd_t);
119 static int ixgbe_detach(dev_info_t *, ddi_detach_cmd_t);
120 static int ixgbe_resume(dev_info_t *);
121 static int ixgbe_suspend(dev_info_t *);
122 static int ixgbe_quiesce(dev_info_t *);
123 static void ixgbe_unconfigure(dev_info_t *, ixgbe_t *);
124 static uint8_t *ixgbe_mc_table_itr(struct ixgbe_hw *, uint8_t **, uint32_t *);
125 static int ixgbe_cbfunc(dev_info_t *, ddi_cb_action_t, void *, void *, void *);
126 static int ixgbe_intr_cb_register(ixgbe_t *);
127 static int ixgbe_intr_adjust(ixgbe_t *, ddi_cb_action_t, int);
128 
129 static int ixgbe_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err,
130     const void *impl_data);
131 static void ixgbe_fm_init(ixgbe_t *);
132 static void ixgbe_fm_fini(ixgbe_t *);
133 
134 char *ixgbe_priv_props[] = {
135 	"_tx_copy_thresh",
136 	"_tx_recycle_thresh",
137 	"_tx_overload_thresh",
138 	"_tx_resched_thresh",
139 	"_rx_copy_thresh",
140 	"_rx_limit_per_intr",
141 	"_intr_throttling",
142 	"_adv_pause_cap",
143 	"_adv_asym_pause_cap",
144 	NULL
145 };
146 
147 #define	IXGBE_MAX_PRIV_PROPS \
148 	(sizeof (ixgbe_priv_props) / sizeof (mac_priv_prop_t))
149 
150 static struct cb_ops ixgbe_cb_ops = {
151 	nulldev,		/* cb_open */
152 	nulldev,		/* cb_close */
153 	nodev,			/* cb_strategy */
154 	nodev,			/* cb_print */
155 	nodev,			/* cb_dump */
156 	nodev,			/* cb_read */
157 	nodev,			/* cb_write */
158 	nodev,			/* cb_ioctl */
159 	nodev,			/* cb_devmap */
160 	nodev,			/* cb_mmap */
161 	nodev,			/* cb_segmap */
162 	nochpoll,		/* cb_chpoll */
163 	ddi_prop_op,		/* cb_prop_op */
164 	NULL,			/* cb_stream */
165 	D_MP | D_HOTPLUG,	/* cb_flag */
166 	CB_REV,			/* cb_rev */
167 	nodev,			/* cb_aread */
168 	nodev			/* cb_awrite */
169 };
170 
171 static struct dev_ops ixgbe_dev_ops = {
172 	DEVO_REV,		/* devo_rev */
173 	0,			/* devo_refcnt */
174 	NULL,			/* devo_getinfo */
175 	nulldev,		/* devo_identify */
176 	nulldev,		/* devo_probe */
177 	ixgbe_attach,		/* devo_attach */
178 	ixgbe_detach,		/* devo_detach */
179 	nodev,			/* devo_reset */
180 	&ixgbe_cb_ops,		/* devo_cb_ops */
181 	NULL,			/* devo_bus_ops */
182 	ddi_power,		/* devo_power */
183 	ixgbe_quiesce,		/* devo_quiesce */
184 };
185 
186 static struct modldrv ixgbe_modldrv = {
187 	&mod_driverops,		/* Type of module.  This one is a driver */
188 	ixgbe_ident,		/* Discription string */
189 	&ixgbe_dev_ops		/* driver ops */
190 };
191 
192 static struct modlinkage ixgbe_modlinkage = {
193 	MODREV_1, &ixgbe_modldrv, NULL
194 };
195 
196 /*
197  * Access attributes for register mapping
198  */
199 ddi_device_acc_attr_t ixgbe_regs_acc_attr = {
200 	DDI_DEVICE_ATTR_V1,
201 	DDI_STRUCTURE_LE_ACC,
202 	DDI_STRICTORDER_ACC,
203 	DDI_FLAGERR_ACC
204 };
205 
206 /*
207  * Loopback property
208  */
209 static lb_property_t lb_normal = {
210 	normal,	"normal", IXGBE_LB_NONE
211 };
212 
213 static lb_property_t lb_mac = {
214 	internal, "MAC", IXGBE_LB_INTERNAL_MAC
215 };
216 
217 static lb_property_t lb_external = {
218 	external, "External", IXGBE_LB_EXTERNAL
219 };
220 
221 #define	IXGBE_M_CALLBACK_FLAGS \
222 	(MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP | MC_PROPINFO)
223 
224 static mac_callbacks_t ixgbe_m_callbacks = {
225 	IXGBE_M_CALLBACK_FLAGS,
226 	ixgbe_m_stat,
227 	ixgbe_m_start,
228 	ixgbe_m_stop,
229 	ixgbe_m_promisc,
230 	ixgbe_m_multicst,
231 	NULL,
232 	NULL,
233 	NULL,
234 	ixgbe_m_ioctl,
235 	ixgbe_m_getcapab,
236 	NULL,
237 	NULL,
238 	ixgbe_m_setprop,
239 	ixgbe_m_getprop,
240 	ixgbe_m_propinfo
241 };
242 
243 /*
244  * Initialize capabilities of each supported adapter type
245  */
246 static adapter_info_t ixgbe_82598eb_cap = {
247 	64,		/* maximum number of rx queues */
248 	1,		/* minimum number of rx queues */
249 	64,		/* default number of rx queues */
250 	16,		/* maximum number of rx groups */
251 	1,		/* minimum number of rx groups */
252 	1,		/* default number of rx groups */
253 	32,		/* maximum number of tx queues */
254 	1,		/* minimum number of tx queues */
255 	8,		/* default number of tx queues */
256 	16366,		/* maximum MTU size */
257 	0xFFFF,		/* maximum interrupt throttle rate */
258 	0,		/* minimum interrupt throttle rate */
259 	200,		/* default interrupt throttle rate */
260 	18,		/* maximum total msix vectors */
261 	16,		/* maximum number of ring vectors */
262 	2,		/* maximum number of other vectors */
263 	IXGBE_EICR_LSC,	/* "other" interrupt types handled */
264 	0,		/* "other" interrupt types enable mask */
265 	(IXGBE_FLAG_DCA_CAPABLE	/* capability flags */
266 	| IXGBE_FLAG_RSS_CAPABLE
267 	| IXGBE_FLAG_VMDQ_CAPABLE)
268 };
269 
270 static adapter_info_t ixgbe_82599eb_cap = {
271 	128,		/* maximum number of rx queues */
272 	1,		/* minimum number of rx queues */
273 	128,		/* default number of rx queues */
274 	64,		/* maximum number of rx groups */
275 	1,		/* minimum number of rx groups */
276 	1,		/* default number of rx groups */
277 	128,		/* maximum number of tx queues */
278 	1,		/* minimum number of tx queues */
279 	8,		/* default number of tx queues */
280 	15500,		/* maximum MTU size */
281 	0xFF8,		/* maximum interrupt throttle rate */
282 	0,		/* minimum interrupt throttle rate */
283 	200,		/* default interrupt throttle rate */
284 	64,		/* maximum total msix vectors */
285 	16,		/* maximum number of ring vectors */
286 	2,		/* maximum number of other vectors */
287 	(IXGBE_EICR_LSC
288 	| IXGBE_EICR_GPI_SDP1
289 	| IXGBE_EICR_GPI_SDP2), /* "other" interrupt types handled */
290 
291 	(IXGBE_SDP1_GPIEN
292 	| IXGBE_SDP2_GPIEN), /* "other" interrupt types enable mask */
293 
294 	(IXGBE_FLAG_DCA_CAPABLE
295 	| IXGBE_FLAG_RSS_CAPABLE
296 	| IXGBE_FLAG_VMDQ_CAPABLE
297 	| IXGBE_FLAG_RSC_CAPABLE
298 	| IXGBE_FLAG_SFP_PLUG_CAPABLE) /* capability flags */
299 };
300 
301 static adapter_info_t ixgbe_X540_cap = {
302 	128,		/* maximum number of rx queues */
303 	1,		/* minimum number of rx queues */
304 	128,		/* default number of rx queues */
305 	64,		/* maximum number of rx groups */
306 	1,		/* minimum number of rx groups */
307 	1,		/* default number of rx groups */
308 	128,		/* maximum number of tx queues */
309 	1,		/* minimum number of tx queues */
310 	8,		/* default number of tx queues */
311 	15500,		/* maximum MTU size */
312 	0xFF8,		/* maximum interrupt throttle rate */
313 	0,		/* minimum interrupt throttle rate */
314 	200,		/* default interrupt throttle rate */
315 	64,		/* maximum total msix vectors */
316 	16,		/* maximum number of ring vectors */
317 	2,		/* maximum number of other vectors */
318 	(IXGBE_EICR_LSC
319 	| IXGBE_EICR_GPI_SDP1_X540
320 	| IXGBE_EICR_GPI_SDP2_X540), /* "other" interrupt types handled */
321 
322 	(IXGBE_SDP1_GPIEN_X540
323 	| IXGBE_SDP2_GPIEN_X540), /* "other" interrupt types enable mask */
324 
325 	(IXGBE_FLAG_DCA_CAPABLE
326 	| IXGBE_FLAG_RSS_CAPABLE
327 	| IXGBE_FLAG_VMDQ_CAPABLE
328 	| IXGBE_FLAG_RSC_CAPABLE) /* capability flags */
329 };
330 
331 static adapter_info_t ixgbe_X550_cap = {
332 	128,		/* maximum number of rx queues */
333 	1,		/* minimum number of rx queues */
334 	128,		/* default number of rx queues */
335 	64,		/* maximum number of rx groups */
336 	1,		/* minimum number of rx groups */
337 	1,		/* default number of rx groups */
338 	128,		/* maximum number of tx queues */
339 	1,		/* minimum number of tx queues */
340 	8,		/* default number of tx queues */
341 	15500,		/* maximum MTU size */
342 	0xFF8,		/* maximum interrupt throttle rate */
343 	0,		/* minimum interrupt throttle rate */
344 	0x200,		/* default interrupt throttle rate */
345 	64,		/* maximum total msix vectors */
346 	16,		/* maximum number of ring vectors */
347 	2,		/* maximum number of other vectors */
348 	IXGBE_EICR_LSC,	/* "other" interrupt types handled */
349 	0,		/* "other" interrupt types enable mask */
350 	(IXGBE_FLAG_RSS_CAPABLE
351 	| IXGBE_FLAG_VMDQ_CAPABLE
352 	| IXGBE_FLAG_RSC_CAPABLE) /* capability flags */
353 };
354 
355 /*
356  * Module Initialization Functions.
357  */
358 
359 int
360 _init(void)
361 {
362 	int status;
363 
364 	mac_init_ops(&ixgbe_dev_ops, MODULE_NAME);
365 
366 	status = mod_install(&ixgbe_modlinkage);
367 
368 	if (status != DDI_SUCCESS) {
369 		mac_fini_ops(&ixgbe_dev_ops);
370 	}
371 
372 	return (status);
373 }
374 
375 int
376 _fini(void)
377 {
378 	int status;
379 
380 	status = mod_remove(&ixgbe_modlinkage);
381 
382 	if (status == DDI_SUCCESS) {
383 		mac_fini_ops(&ixgbe_dev_ops);
384 	}
385 
386 	return (status);
387 }
388 
389 int
390 _info(struct modinfo *modinfop)
391 {
392 	int status;
393 
394 	status = mod_info(&ixgbe_modlinkage, modinfop);
395 
396 	return (status);
397 }
398 
399 /*
400  * ixgbe_attach - Driver attach.
401  *
402  * This function is the device specific initialization entry
403  * point. This entry point is required and must be written.
404  * The DDI_ATTACH command must be provided in the attach entry
405  * point. When attach() is called with cmd set to DDI_ATTACH,
406  * all normal kernel services (such as kmem_alloc(9F)) are
407  * available for use by the driver.
408  *
409  * The attach() function will be called once for each instance
410  * of  the  device  on  the  system with cmd set to DDI_ATTACH.
411  * Until attach() succeeds, the only driver entry points which
412  * may be called are open(9E) and getinfo(9E).
413  */
414 static int
415 ixgbe_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
416 {
417 	ixgbe_t *ixgbe;
418 	struct ixgbe_osdep *osdep;
419 	struct ixgbe_hw *hw;
420 	int instance;
421 	char taskqname[32];
422 
423 	/*
424 	 * Check the command and perform corresponding operations
425 	 */
426 	switch (cmd) {
427 	default:
428 		return (DDI_FAILURE);
429 
430 	case DDI_RESUME:
431 		return (ixgbe_resume(devinfo));
432 
433 	case DDI_ATTACH:
434 		break;
435 	}
436 
437 	/* Get the device instance */
438 	instance = ddi_get_instance(devinfo);
439 
440 	/* Allocate memory for the instance data structure */
441 	ixgbe = kmem_zalloc(sizeof (ixgbe_t), KM_SLEEP);
442 
443 	ixgbe->dip = devinfo;
444 	ixgbe->instance = instance;
445 
446 	hw = &ixgbe->hw;
447 	osdep = &ixgbe->osdep;
448 	hw->back = osdep;
449 	osdep->ixgbe = ixgbe;
450 
451 	/* Attach the instance pointer to the dev_info data structure */
452 	ddi_set_driver_private(devinfo, ixgbe);
453 
454 	/*
455 	 * Initialize for FMA support
456 	 */
457 	ixgbe->fm_capabilities = ixgbe_get_prop(ixgbe, PROP_FM_CAPABLE,
458 	    0, 0x0f, DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
459 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
460 	ixgbe_fm_init(ixgbe);
461 	ixgbe->attach_progress |= ATTACH_PROGRESS_FM_INIT;
462 
463 	/*
464 	 * Map PCI config space registers
465 	 */
466 	if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
467 		ixgbe_error(ixgbe, "Failed to map PCI configurations");
468 		goto attach_fail;
469 	}
470 	ixgbe->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
471 
472 	/*
473 	 * Identify the chipset family
474 	 */
475 	if (ixgbe_identify_hardware(ixgbe) != IXGBE_SUCCESS) {
476 		ixgbe_error(ixgbe, "Failed to identify hardware");
477 		goto attach_fail;
478 	}
479 
480 	/*
481 	 * Map device registers
482 	 */
483 	if (ixgbe_regs_map(ixgbe) != IXGBE_SUCCESS) {
484 		ixgbe_error(ixgbe, "Failed to map device registers");
485 		goto attach_fail;
486 	}
487 	ixgbe->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
488 
489 	/*
490 	 * Initialize driver parameters
491 	 */
492 	ixgbe_init_properties(ixgbe);
493 	ixgbe->attach_progress |= ATTACH_PROGRESS_PROPS;
494 
495 	/*
496 	 * Register interrupt callback
497 	 */
498 	if (ixgbe_intr_cb_register(ixgbe) != IXGBE_SUCCESS) {
499 		ixgbe_error(ixgbe, "Failed to register interrupt callback");
500 		goto attach_fail;
501 	}
502 
503 	/*
504 	 * Allocate interrupts
505 	 */
506 	if (ixgbe_alloc_intrs(ixgbe) != IXGBE_SUCCESS) {
507 		ixgbe_error(ixgbe, "Failed to allocate interrupts");
508 		goto attach_fail;
509 	}
510 	ixgbe->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
511 
512 	/*
513 	 * Allocate rx/tx rings based on the ring numbers.
514 	 * The actual numbers of rx/tx rings are decided by the number of
515 	 * allocated interrupt vectors, so we should allocate the rings after
516 	 * interrupts are allocated.
517 	 */
518 	if (ixgbe_alloc_rings(ixgbe) != IXGBE_SUCCESS) {
519 		ixgbe_error(ixgbe, "Failed to allocate rx and tx rings");
520 		goto attach_fail;
521 	}
522 	ixgbe->attach_progress |= ATTACH_PROGRESS_ALLOC_RINGS;
523 
524 	/*
525 	 * Map rings to interrupt vectors
526 	 */
527 	if (ixgbe_map_intrs_to_vectors(ixgbe) != IXGBE_SUCCESS) {
528 		ixgbe_error(ixgbe, "Failed to map interrupts to vectors");
529 		goto attach_fail;
530 	}
531 
532 	/*
533 	 * Add interrupt handlers
534 	 */
535 	if (ixgbe_add_intr_handlers(ixgbe) != IXGBE_SUCCESS) {
536 		ixgbe_error(ixgbe, "Failed to add interrupt handlers");
537 		goto attach_fail;
538 	}
539 	ixgbe->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
540 
541 	/*
542 	 * Create a taskq for sfp-change
543 	 */
544 	(void) sprintf(taskqname, "ixgbe%d_sfp_taskq", instance);
545 	if ((ixgbe->sfp_taskq = ddi_taskq_create(devinfo, taskqname,
546 	    1, TASKQ_DEFAULTPRI, 0)) == NULL) {
547 		ixgbe_error(ixgbe, "sfp_taskq create failed");
548 		goto attach_fail;
549 	}
550 	ixgbe->attach_progress |= ATTACH_PROGRESS_SFP_TASKQ;
551 
552 	/*
553 	 * Create a taskq for over-temp
554 	 */
555 	(void) sprintf(taskqname, "ixgbe%d_overtemp_taskq", instance);
556 	if ((ixgbe->overtemp_taskq = ddi_taskq_create(devinfo, taskqname,
557 	    1, TASKQ_DEFAULTPRI, 0)) == NULL) {
558 		ixgbe_error(ixgbe, "overtemp_taskq create failed");
559 		goto attach_fail;
560 	}
561 	ixgbe->attach_progress |= ATTACH_PROGRESS_OVERTEMP_TASKQ;
562 
563 	/*
564 	 * Create a taskq for processing external PHY interrupts
565 	 */
566 	(void) sprintf(taskqname, "ixgbe%d_phy_taskq", instance);
567 	if ((ixgbe->phy_taskq = ddi_taskq_create(devinfo, taskqname,
568 	    1, TASKQ_DEFAULTPRI, 0)) == NULL) {
569 		ixgbe_error(ixgbe, "phy_taskq create failed");
570 		goto attach_fail;
571 	}
572 	ixgbe->attach_progress |= ATTACH_PROGRESS_PHY_TASKQ;
573 
574 	/*
575 	 * Initialize driver parameters
576 	 */
577 	if (ixgbe_init_driver_settings(ixgbe) != IXGBE_SUCCESS) {
578 		ixgbe_error(ixgbe, "Failed to initialize driver settings");
579 		goto attach_fail;
580 	}
581 
582 	/*
583 	 * Initialize mutexes for this device.
584 	 * Do this before enabling the interrupt handler and
585 	 * register the softint to avoid the condition where
586 	 * interrupt handler can try using uninitialized mutex.
587 	 */
588 	ixgbe_init_locks(ixgbe);
589 	ixgbe->attach_progress |= ATTACH_PROGRESS_LOCKS;
590 
591 	/*
592 	 * Initialize chipset hardware
593 	 */
594 	if (ixgbe_init(ixgbe) != IXGBE_SUCCESS) {
595 		ixgbe_error(ixgbe, "Failed to initialize adapter");
596 		goto attach_fail;
597 	}
598 	ixgbe->link_check_complete = B_FALSE;
599 	ixgbe->link_check_hrtime = gethrtime() +
600 	    (IXGBE_LINK_UP_TIME * 100000000ULL);
601 	ixgbe->attach_progress |= ATTACH_PROGRESS_INIT;
602 
603 	if (ixgbe_check_acc_handle(ixgbe->osdep.cfg_handle) != DDI_FM_OK) {
604 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
605 		goto attach_fail;
606 	}
607 
608 	/*
609 	 * Initialize adapter capabilities
610 	 */
611 	ixgbe_init_params(ixgbe);
612 
613 	/*
614 	 * Initialize statistics
615 	 */
616 	if (ixgbe_init_stats(ixgbe) != IXGBE_SUCCESS) {
617 		ixgbe_error(ixgbe, "Failed to initialize statistics");
618 		goto attach_fail;
619 	}
620 	ixgbe->attach_progress |= ATTACH_PROGRESS_STATS;
621 
622 	/*
623 	 * Register the driver to the MAC
624 	 */
625 	if (ixgbe_register_mac(ixgbe) != IXGBE_SUCCESS) {
626 		ixgbe_error(ixgbe, "Failed to register MAC");
627 		goto attach_fail;
628 	}
629 	mac_link_update(ixgbe->mac_hdl, LINK_STATE_UNKNOWN);
630 	ixgbe->attach_progress |= ATTACH_PROGRESS_MAC;
631 
632 	ixgbe->periodic_id = ddi_periodic_add(ixgbe_link_timer, ixgbe,
633 	    IXGBE_CYCLIC_PERIOD, DDI_IPL_0);
634 	if (ixgbe->periodic_id == 0) {
635 		ixgbe_error(ixgbe, "Failed to add the link check timer");
636 		goto attach_fail;
637 	}
638 	ixgbe->attach_progress |= ATTACH_PROGRESS_LINK_TIMER;
639 
640 	/*
641 	 * Now that mutex locks are initialized, and the chip is also
642 	 * initialized, enable interrupts.
643 	 */
644 	if (ixgbe_enable_intrs(ixgbe) != IXGBE_SUCCESS) {
645 		ixgbe_error(ixgbe, "Failed to enable DDI interrupts");
646 		goto attach_fail;
647 	}
648 	ixgbe->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
649 
650 	ixgbe_log(ixgbe, "%s", ixgbe_ident);
651 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_INITIALIZED);
652 
653 	return (DDI_SUCCESS);
654 
655 attach_fail:
656 	ixgbe_unconfigure(devinfo, ixgbe);
657 	return (DDI_FAILURE);
658 }
659 
660 /*
661  * ixgbe_detach - Driver detach.
662  *
663  * The detach() function is the complement of the attach routine.
664  * If cmd is set to DDI_DETACH, detach() is used to remove  the
665  * state  associated  with  a  given  instance of a device node
666  * prior to the removal of that instance from the system.
667  *
668  * The detach() function will be called once for each  instance
669  * of the device for which there has been a successful attach()
670  * once there are no longer  any  opens  on  the  device.
671  *
672  * Interrupts routine are disabled, All memory allocated by this
673  * driver are freed.
674  */
675 static int
676 ixgbe_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
677 {
678 	ixgbe_t *ixgbe;
679 
680 	/*
681 	 * Check detach command
682 	 */
683 	switch (cmd) {
684 	default:
685 		return (DDI_FAILURE);
686 
687 	case DDI_SUSPEND:
688 		return (ixgbe_suspend(devinfo));
689 
690 	case DDI_DETACH:
691 		break;
692 	}
693 
694 	/*
695 	 * Get the pointer to the driver private data structure
696 	 */
697 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
698 	if (ixgbe == NULL)
699 		return (DDI_FAILURE);
700 
701 	/*
702 	 * If the device is still running, it needs to be stopped first.
703 	 * This check is necessary because under some specific circumstances,
704 	 * the detach routine can be called without stopping the interface
705 	 * first.
706 	 */
707 	if (ixgbe->ixgbe_state & IXGBE_STARTED) {
708 		atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_STARTED);
709 		mutex_enter(&ixgbe->gen_lock);
710 		ixgbe_stop(ixgbe, B_TRUE);
711 		mutex_exit(&ixgbe->gen_lock);
712 		/* Disable and stop the watchdog timer */
713 		ixgbe_disable_watchdog_timer(ixgbe);
714 	}
715 
716 	/*
717 	 * Check if there are still rx buffers held by the upper layer.
718 	 * If so, fail the detach.
719 	 */
720 	if (!ixgbe_rx_drain(ixgbe))
721 		return (DDI_FAILURE);
722 
723 	/*
724 	 * Do the remaining unconfigure routines
725 	 */
726 	ixgbe_unconfigure(devinfo, ixgbe);
727 
728 	return (DDI_SUCCESS);
729 }
730 
731 /*
732  * quiesce(9E) entry point.
733  *
734  * This function is called when the system is single-threaded at high
735  * PIL with preemption disabled. Therefore, this function must not be
736  * blocked.
737  *
738  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
739  * DDI_FAILURE indicates an error condition and should almost never happen.
740  */
741 static int
742 ixgbe_quiesce(dev_info_t *devinfo)
743 {
744 	ixgbe_t *ixgbe;
745 	struct ixgbe_hw *hw;
746 
747 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
748 
749 	if (ixgbe == NULL)
750 		return (DDI_FAILURE);
751 
752 	hw = &ixgbe->hw;
753 
754 	/*
755 	 * Disable the adapter interrupts
756 	 */
757 	ixgbe_disable_adapter_interrupts(ixgbe);
758 
759 	/*
760 	 * Tell firmware driver is no longer in control
761 	 */
762 	ixgbe_release_driver_control(hw);
763 
764 	/*
765 	 * Reset the chipset
766 	 */
767 	(void) ixgbe_reset_hw(hw);
768 
769 	/*
770 	 * Reset PHY
771 	 */
772 	(void) ixgbe_reset_phy(hw);
773 
774 	return (DDI_SUCCESS);
775 }
776 
777 static void
778 ixgbe_unconfigure(dev_info_t *devinfo, ixgbe_t *ixgbe)
779 {
780 	/*
781 	 * Disable interrupt
782 	 */
783 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
784 		(void) ixgbe_disable_intrs(ixgbe);
785 	}
786 
787 	/*
788 	 * remove the link check timer
789 	 */
790 	if (ixgbe->attach_progress & ATTACH_PROGRESS_LINK_TIMER) {
791 		if (ixgbe->periodic_id != NULL) {
792 			ddi_periodic_delete(ixgbe->periodic_id);
793 			ixgbe->periodic_id = NULL;
794 		}
795 	}
796 
797 	/*
798 	 * Unregister MAC
799 	 */
800 	if (ixgbe->attach_progress & ATTACH_PROGRESS_MAC) {
801 		(void) mac_unregister(ixgbe->mac_hdl);
802 	}
803 
804 	/*
805 	 * Free statistics
806 	 */
807 	if (ixgbe->attach_progress & ATTACH_PROGRESS_STATS) {
808 		kstat_delete((kstat_t *)ixgbe->ixgbe_ks);
809 	}
810 
811 	/*
812 	 * Remove interrupt handlers
813 	 */
814 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
815 		ixgbe_rem_intr_handlers(ixgbe);
816 	}
817 
818 	/*
819 	 * Remove taskq for sfp-status-change
820 	 */
821 	if (ixgbe->attach_progress & ATTACH_PROGRESS_SFP_TASKQ) {
822 		ddi_taskq_destroy(ixgbe->sfp_taskq);
823 	}
824 
825 	/*
826 	 * Remove taskq for over-temp
827 	 */
828 	if (ixgbe->attach_progress & ATTACH_PROGRESS_OVERTEMP_TASKQ) {
829 		ddi_taskq_destroy(ixgbe->overtemp_taskq);
830 	}
831 
832 	/*
833 	 * Remove taskq for external PHYs
834 	 */
835 	if (ixgbe->attach_progress & ATTACH_PROGRESS_PHY_TASKQ) {
836 		ddi_taskq_destroy(ixgbe->phy_taskq);
837 	}
838 
839 	/*
840 	 * Remove interrupts
841 	 */
842 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ALLOC_INTR) {
843 		ixgbe_rem_intrs(ixgbe);
844 	}
845 
846 	/*
847 	 * Unregister interrupt callback handler
848 	 */
849 	(void) ddi_cb_unregister(ixgbe->cb_hdl);
850 
851 	/*
852 	 * Remove driver properties
853 	 */
854 	if (ixgbe->attach_progress & ATTACH_PROGRESS_PROPS) {
855 		(void) ddi_prop_remove_all(devinfo);
856 	}
857 
858 	/*
859 	 * Stop the chipset
860 	 */
861 	if (ixgbe->attach_progress & ATTACH_PROGRESS_INIT) {
862 		mutex_enter(&ixgbe->gen_lock);
863 		ixgbe_chip_stop(ixgbe);
864 		mutex_exit(&ixgbe->gen_lock);
865 	}
866 
867 	/*
868 	 * Free register handle
869 	 */
870 	if (ixgbe->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
871 		if (ixgbe->osdep.reg_handle != NULL)
872 			ddi_regs_map_free(&ixgbe->osdep.reg_handle);
873 	}
874 
875 	/*
876 	 * Free PCI config handle
877 	 */
878 	if (ixgbe->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
879 		if (ixgbe->osdep.cfg_handle != NULL)
880 			pci_config_teardown(&ixgbe->osdep.cfg_handle);
881 	}
882 
883 	/*
884 	 * Free locks
885 	 */
886 	if (ixgbe->attach_progress & ATTACH_PROGRESS_LOCKS) {
887 		ixgbe_destroy_locks(ixgbe);
888 	}
889 
890 	/*
891 	 * Free the rx/tx rings
892 	 */
893 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
894 		ixgbe_free_rings(ixgbe);
895 	}
896 
897 	/*
898 	 * Unregister FMA capabilities
899 	 */
900 	if (ixgbe->attach_progress & ATTACH_PROGRESS_FM_INIT) {
901 		ixgbe_fm_fini(ixgbe);
902 	}
903 
904 	/*
905 	 * Free the driver data structure
906 	 */
907 	kmem_free(ixgbe, sizeof (ixgbe_t));
908 
909 	ddi_set_driver_private(devinfo, NULL);
910 }
911 
912 /*
913  * ixgbe_register_mac - Register the driver and its function pointers with
914  * the GLD interface.
915  */
916 static int
917 ixgbe_register_mac(ixgbe_t *ixgbe)
918 {
919 	struct ixgbe_hw *hw = &ixgbe->hw;
920 	mac_register_t *mac;
921 	int status;
922 
923 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
924 		return (IXGBE_FAILURE);
925 
926 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
927 	mac->m_driver = ixgbe;
928 	mac->m_dip = ixgbe->dip;
929 	mac->m_src_addr = hw->mac.addr;
930 	mac->m_callbacks = &ixgbe_m_callbacks;
931 	mac->m_min_sdu = 0;
932 	mac->m_max_sdu = ixgbe->default_mtu;
933 	mac->m_margin = VLAN_TAGSZ;
934 	mac->m_priv_props = ixgbe_priv_props;
935 	mac->m_v12n = MAC_VIRT_LEVEL1;
936 
937 	status = mac_register(mac, &ixgbe->mac_hdl);
938 
939 	mac_free(mac);
940 
941 	return ((status == 0) ? IXGBE_SUCCESS : IXGBE_FAILURE);
942 }
943 
944 /*
945  * ixgbe_identify_hardware - Identify the type of the chipset.
946  */
947 static int
948 ixgbe_identify_hardware(ixgbe_t *ixgbe)
949 {
950 	struct ixgbe_hw *hw = &ixgbe->hw;
951 	struct ixgbe_osdep *osdep = &ixgbe->osdep;
952 
953 	/*
954 	 * Get the device id
955 	 */
956 	hw->vendor_id =
957 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
958 	hw->device_id =
959 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
960 	hw->revision_id =
961 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
962 	hw->subsystem_device_id =
963 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
964 	hw->subsystem_vendor_id =
965 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
966 
967 	/*
968 	 * Set the mac type of the adapter based on the device id
969 	 */
970 	if (ixgbe_set_mac_type(hw) != IXGBE_SUCCESS) {
971 		return (IXGBE_FAILURE);
972 	}
973 
974 	/*
975 	 * Install adapter capabilities
976 	 */
977 	switch (hw->mac.type) {
978 	case ixgbe_mac_82598EB:
979 		IXGBE_DEBUGLOG_0(ixgbe, "identify 82598 adapter\n");
980 		ixgbe->capab = &ixgbe_82598eb_cap;
981 
982 		if (ixgbe_get_media_type(hw) == ixgbe_media_type_copper) {
983 			ixgbe->capab->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
984 			ixgbe->capab->other_intr |= IXGBE_EICR_GPI_SDP1;
985 			ixgbe->capab->other_gpie |= IXGBE_SDP1_GPIEN;
986 		}
987 		break;
988 
989 	case ixgbe_mac_82599EB:
990 		IXGBE_DEBUGLOG_0(ixgbe, "identify 82599 adapter\n");
991 		ixgbe->capab = &ixgbe_82599eb_cap;
992 
993 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM) {
994 			ixgbe->capab->flags |= IXGBE_FLAG_TEMP_SENSOR_CAPABLE;
995 			ixgbe->capab->other_intr |= IXGBE_EICR_GPI_SDP0;
996 			ixgbe->capab->other_gpie |= IXGBE_SDP0_GPIEN;
997 		}
998 		break;
999 
1000 	case ixgbe_mac_X540:
1001 		IXGBE_DEBUGLOG_0(ixgbe, "identify X540 adapter\n");
1002 		ixgbe->capab = &ixgbe_X540_cap;
1003 		/*
1004 		 * For now, X540 is all set in its capab structure.
1005 		 * As other X540 variants show up, things can change here.
1006 		 */
1007 		break;
1008 
1009 	case ixgbe_mac_X550:
1010 	case ixgbe_mac_X550EM_x:
1011 		IXGBE_DEBUGLOG_0(ixgbe, "identify X550 adapter\n");
1012 		ixgbe->capab = &ixgbe_X550_cap;
1013 
1014 		if (hw->device_id == IXGBE_DEV_ID_X550EM_X_SFP)
1015 			ixgbe->capab->flags |= IXGBE_FLAG_SFP_PLUG_CAPABLE;
1016 
1017 		/*
1018 		 * Link detection on X552 SFP+ and X552/X557-AT
1019 		 */
1020 		if (hw->device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
1021 		    hw->device_id == IXGBE_DEV_ID_X550EM_X_10G_T) {
1022 			ixgbe->capab->other_intr |=
1023 			    IXGBE_EIMS_GPI_SDP0_BY_MAC(hw);
1024 			ixgbe->capab->other_gpie |= IXGBE_SDP0_GPIEN_X540;
1025 		}
1026 		break;
1027 
1028 	default:
1029 		IXGBE_DEBUGLOG_1(ixgbe,
1030 		    "adapter not supported in ixgbe_identify_hardware(): %d\n",
1031 		    hw->mac.type);
1032 		return (IXGBE_FAILURE);
1033 	}
1034 
1035 	return (IXGBE_SUCCESS);
1036 }
1037 
1038 /*
1039  * ixgbe_regs_map - Map the device registers.
1040  *
1041  */
1042 static int
1043 ixgbe_regs_map(ixgbe_t *ixgbe)
1044 {
1045 	dev_info_t *devinfo = ixgbe->dip;
1046 	struct ixgbe_hw *hw = &ixgbe->hw;
1047 	struct ixgbe_osdep *osdep = &ixgbe->osdep;
1048 	off_t mem_size;
1049 
1050 	/*
1051 	 * First get the size of device registers to be mapped.
1052 	 */
1053 	if (ddi_dev_regsize(devinfo, IXGBE_ADAPTER_REGSET, &mem_size)
1054 	    != DDI_SUCCESS) {
1055 		return (IXGBE_FAILURE);
1056 	}
1057 
1058 	/*
1059 	 * Call ddi_regs_map_setup() to map registers
1060 	 */
1061 	if ((ddi_regs_map_setup(devinfo, IXGBE_ADAPTER_REGSET,
1062 	    (caddr_t *)&hw->hw_addr, 0,
1063 	    mem_size, &ixgbe_regs_acc_attr,
1064 	    &osdep->reg_handle)) != DDI_SUCCESS) {
1065 		return (IXGBE_FAILURE);
1066 	}
1067 
1068 	return (IXGBE_SUCCESS);
1069 }
1070 
1071 /*
1072  * ixgbe_init_properties - Initialize driver properties.
1073  */
1074 static void
1075 ixgbe_init_properties(ixgbe_t *ixgbe)
1076 {
1077 	/*
1078 	 * Get conf file properties, including link settings
1079 	 * jumbo frames, ring number, descriptor number, etc.
1080 	 */
1081 	ixgbe_get_conf(ixgbe);
1082 }
1083 
1084 /*
1085  * ixgbe_init_driver_settings - Initialize driver settings.
1086  *
1087  * The settings include hardware function pointers, bus information,
1088  * rx/tx rings settings, link state, and any other parameters that
1089  * need to be setup during driver initialization.
1090  */
1091 static int
1092 ixgbe_init_driver_settings(ixgbe_t *ixgbe)
1093 {
1094 	struct ixgbe_hw *hw = &ixgbe->hw;
1095 	dev_info_t *devinfo = ixgbe->dip;
1096 	ixgbe_rx_ring_t *rx_ring;
1097 	ixgbe_rx_group_t *rx_group;
1098 	ixgbe_tx_ring_t *tx_ring;
1099 	uint32_t rx_size;
1100 	uint32_t tx_size;
1101 	uint32_t ring_per_group;
1102 	int i;
1103 
1104 	/*
1105 	 * Initialize chipset specific hardware function pointers
1106 	 */
1107 	if (ixgbe_init_shared_code(hw) != IXGBE_SUCCESS) {
1108 		return (IXGBE_FAILURE);
1109 	}
1110 
1111 	/*
1112 	 * Get the system page size
1113 	 */
1114 	ixgbe->sys_page_size = ddi_ptob(devinfo, (ulong_t)1);
1115 
1116 	/*
1117 	 * Set rx buffer size
1118 	 *
1119 	 * The IP header alignment room is counted in the calculation.
1120 	 * The rx buffer size is in unit of 1K that is required by the
1121 	 * chipset hardware.
1122 	 */
1123 	rx_size = ixgbe->max_frame_size + IPHDR_ALIGN_ROOM;
1124 	ixgbe->rx_buf_size = ((rx_size >> 10) +
1125 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
1126 
1127 	/*
1128 	 * Set tx buffer size
1129 	 */
1130 	tx_size = ixgbe->max_frame_size;
1131 	ixgbe->tx_buf_size = ((tx_size >> 10) +
1132 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
1133 
1134 	/*
1135 	 * Initialize rx/tx rings/groups parameters
1136 	 */
1137 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
1138 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1139 		rx_ring = &ixgbe->rx_rings[i];
1140 		rx_ring->index = i;
1141 		rx_ring->ixgbe = ixgbe;
1142 		rx_ring->group_index = i / ring_per_group;
1143 		rx_ring->hw_index = ixgbe_get_hw_rx_index(ixgbe, i);
1144 	}
1145 
1146 	for (i = 0; i < ixgbe->num_rx_groups; i++) {
1147 		rx_group = &ixgbe->rx_groups[i];
1148 		rx_group->index = i;
1149 		rx_group->ixgbe = ixgbe;
1150 	}
1151 
1152 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1153 		tx_ring = &ixgbe->tx_rings[i];
1154 		tx_ring->index = i;
1155 		tx_ring->ixgbe = ixgbe;
1156 		if (ixgbe->tx_head_wb_enable)
1157 			tx_ring->tx_recycle = ixgbe_tx_recycle_head_wb;
1158 		else
1159 			tx_ring->tx_recycle = ixgbe_tx_recycle_legacy;
1160 
1161 		tx_ring->ring_size = ixgbe->tx_ring_size;
1162 		tx_ring->free_list_size = ixgbe->tx_ring_size +
1163 		    (ixgbe->tx_ring_size >> 1);
1164 	}
1165 
1166 	/*
1167 	 * Initialize values of interrupt throttling rate
1168 	 */
1169 	for (i = 1; i < MAX_INTR_VECTOR; i++)
1170 		ixgbe->intr_throttling[i] = ixgbe->intr_throttling[0];
1171 
1172 	/*
1173 	 * The initial link state should be "unknown"
1174 	 */
1175 	ixgbe->link_state = LINK_STATE_UNKNOWN;
1176 
1177 	return (IXGBE_SUCCESS);
1178 }
1179 
1180 /*
1181  * ixgbe_init_locks - Initialize locks.
1182  */
1183 static void
1184 ixgbe_init_locks(ixgbe_t *ixgbe)
1185 {
1186 	ixgbe_rx_ring_t *rx_ring;
1187 	ixgbe_tx_ring_t *tx_ring;
1188 	int i;
1189 
1190 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1191 		rx_ring = &ixgbe->rx_rings[i];
1192 		mutex_init(&rx_ring->rx_lock, NULL,
1193 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1194 	}
1195 
1196 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1197 		tx_ring = &ixgbe->tx_rings[i];
1198 		mutex_init(&tx_ring->tx_lock, NULL,
1199 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1200 		mutex_init(&tx_ring->recycle_lock, NULL,
1201 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1202 		mutex_init(&tx_ring->tcb_head_lock, NULL,
1203 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1204 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
1205 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1206 	}
1207 
1208 	mutex_init(&ixgbe->gen_lock, NULL,
1209 	    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1210 
1211 	mutex_init(&ixgbe->watchdog_lock, NULL,
1212 	    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1213 }
1214 
1215 /*
1216  * ixgbe_destroy_locks - Destroy locks.
1217  */
1218 static void
1219 ixgbe_destroy_locks(ixgbe_t *ixgbe)
1220 {
1221 	ixgbe_rx_ring_t *rx_ring;
1222 	ixgbe_tx_ring_t *tx_ring;
1223 	int i;
1224 
1225 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1226 		rx_ring = &ixgbe->rx_rings[i];
1227 		mutex_destroy(&rx_ring->rx_lock);
1228 	}
1229 
1230 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1231 		tx_ring = &ixgbe->tx_rings[i];
1232 		mutex_destroy(&tx_ring->tx_lock);
1233 		mutex_destroy(&tx_ring->recycle_lock);
1234 		mutex_destroy(&tx_ring->tcb_head_lock);
1235 		mutex_destroy(&tx_ring->tcb_tail_lock);
1236 	}
1237 
1238 	mutex_destroy(&ixgbe->gen_lock);
1239 	mutex_destroy(&ixgbe->watchdog_lock);
1240 }
1241 
1242 static int
1243 ixgbe_resume(dev_info_t *devinfo)
1244 {
1245 	ixgbe_t *ixgbe;
1246 	int i;
1247 
1248 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
1249 	if (ixgbe == NULL)
1250 		return (DDI_FAILURE);
1251 
1252 	mutex_enter(&ixgbe->gen_lock);
1253 
1254 	if (ixgbe->ixgbe_state & IXGBE_STARTED) {
1255 		if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
1256 			mutex_exit(&ixgbe->gen_lock);
1257 			return (DDI_FAILURE);
1258 		}
1259 
1260 		/*
1261 		 * Enable and start the watchdog timer
1262 		 */
1263 		ixgbe_enable_watchdog_timer(ixgbe);
1264 	}
1265 
1266 	atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_SUSPENDED);
1267 
1268 	if (ixgbe->ixgbe_state & IXGBE_STARTED) {
1269 		for (i = 0; i < ixgbe->num_tx_rings; i++) {
1270 			mac_tx_ring_update(ixgbe->mac_hdl,
1271 			    ixgbe->tx_rings[i].ring_handle);
1272 		}
1273 	}
1274 
1275 	mutex_exit(&ixgbe->gen_lock);
1276 
1277 	return (DDI_SUCCESS);
1278 }
1279 
1280 static int
1281 ixgbe_suspend(dev_info_t *devinfo)
1282 {
1283 	ixgbe_t *ixgbe;
1284 
1285 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
1286 	if (ixgbe == NULL)
1287 		return (DDI_FAILURE);
1288 
1289 	mutex_enter(&ixgbe->gen_lock);
1290 
1291 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_SUSPENDED);
1292 	if (!(ixgbe->ixgbe_state & IXGBE_STARTED)) {
1293 		mutex_exit(&ixgbe->gen_lock);
1294 		return (DDI_SUCCESS);
1295 	}
1296 	ixgbe_stop(ixgbe, B_FALSE);
1297 
1298 	mutex_exit(&ixgbe->gen_lock);
1299 
1300 	/*
1301 	 * Disable and stop the watchdog timer
1302 	 */
1303 	ixgbe_disable_watchdog_timer(ixgbe);
1304 
1305 	return (DDI_SUCCESS);
1306 }
1307 
1308 /*
1309  * ixgbe_init - Initialize the device.
1310  */
1311 static int
1312 ixgbe_init(ixgbe_t *ixgbe)
1313 {
1314 	struct ixgbe_hw *hw = &ixgbe->hw;
1315 	u8 pbanum[IXGBE_PBANUM_LENGTH];
1316 	int rv;
1317 
1318 	mutex_enter(&ixgbe->gen_lock);
1319 
1320 	/*
1321 	 * Configure/Initialize hardware
1322 	 */
1323 	rv = ixgbe_init_hw(hw);
1324 	if (rv != IXGBE_SUCCESS) {
1325 		switch (rv) {
1326 
1327 		/*
1328 		 * The first three errors are not prohibitive to us progressing
1329 		 * further, and are maily advisory in nature. In the case of a
1330 		 * SFP module not being present or not deemed supported by the
1331 		 * common code, we adivse the operator of this fact but carry on
1332 		 * instead of failing hard, as SFPs can be inserted or replaced
1333 		 * while the driver is running. In the case of a unknown error,
1334 		 * we fail-hard, logging the reason and emitting a FMA event.
1335 		 */
1336 		case IXGBE_ERR_EEPROM_VERSION:
1337 			ixgbe_error(ixgbe,
1338 			    "This Intel 10Gb Ethernet device is pre-release and"
1339 			    " contains outdated firmware. Please contact your"
1340 			    " hardware vendor for a replacement.");
1341 			break;
1342 		case IXGBE_ERR_SFP_NOT_PRESENT:
1343 			ixgbe_error(ixgbe,
1344 			    "No SFP+ module detected on this interface. Please "
1345 			    "install a supported SFP+ module for this "
1346 			    "interface to become operational.");
1347 			break;
1348 		case IXGBE_ERR_SFP_NOT_SUPPORTED:
1349 			ixgbe_error(ixgbe,
1350 			    "Unsupported SFP+ module detected. Please replace "
1351 			    "it with a supported SFP+ module per Intel "
1352 			    "documentation, or bypass this check with "
1353 			    "allow_unsupported_sfp=1 in ixgbe.conf.");
1354 			break;
1355 		default:
1356 			ixgbe_error(ixgbe,
1357 			    "Failed to initialize hardware. ixgbe_init_hw "
1358 			    "returned %d", rv);
1359 			ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1360 			goto init_fail;
1361 		}
1362 	}
1363 
1364 	/*
1365 	 * Need to init eeprom before validating the checksum.
1366 	 */
1367 	if (ixgbe_init_eeprom_params(hw) < 0) {
1368 		ixgbe_error(ixgbe,
1369 		    "Unable to intitialize the eeprom interface.");
1370 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1371 		goto init_fail;
1372 	}
1373 
1374 	/*
1375 	 * NVM validation
1376 	 */
1377 	if (ixgbe_validate_eeprom_checksum(hw, NULL) < 0) {
1378 		/*
1379 		 * Some PCI-E parts fail the first check due to
1380 		 * the link being in sleep state.  Call it again,
1381 		 * if it fails a second time it's a real issue.
1382 		 */
1383 		if (ixgbe_validate_eeprom_checksum(hw, NULL) < 0) {
1384 			ixgbe_error(ixgbe,
1385 			    "Invalid NVM checksum. Please contact "
1386 			    "the vendor to update the NVM.");
1387 			ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1388 			goto init_fail;
1389 		}
1390 	}
1391 
1392 	/*
1393 	 * Setup default flow control thresholds - enable/disable
1394 	 * & flow control type is controlled by ixgbe.conf
1395 	 */
1396 	hw->fc.high_water[0] = DEFAULT_FCRTH;
1397 	hw->fc.low_water[0] = DEFAULT_FCRTL;
1398 	hw->fc.pause_time = DEFAULT_FCPAUSE;
1399 	hw->fc.send_xon = B_TRUE;
1400 
1401 	/*
1402 	 * Initialize flow control
1403 	 */
1404 	(void) ixgbe_start_hw(hw);
1405 
1406 	/*
1407 	 * Initialize link settings
1408 	 */
1409 	(void) ixgbe_driver_setup_link(ixgbe, B_FALSE);
1410 
1411 	/*
1412 	 * Initialize the chipset hardware
1413 	 */
1414 	if (ixgbe_chip_start(ixgbe) != IXGBE_SUCCESS) {
1415 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1416 		goto init_fail;
1417 	}
1418 
1419 	/*
1420 	 * Read identifying information and place in devinfo.
1421 	 */
1422 	pbanum[0] = '\0';
1423 	(void) ixgbe_read_pba_string(hw, pbanum, sizeof (pbanum));
1424 	if (*pbanum != '\0') {
1425 		(void) ddi_prop_update_string(DDI_DEV_T_NONE, ixgbe->dip,
1426 		    "printed-board-assembly", (char *)pbanum);
1427 	}
1428 
1429 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1430 		goto init_fail;
1431 	}
1432 
1433 	mutex_exit(&ixgbe->gen_lock);
1434 	return (IXGBE_SUCCESS);
1435 
1436 init_fail:
1437 	/*
1438 	 * Reset PHY
1439 	 */
1440 	(void) ixgbe_reset_phy(hw);
1441 
1442 	mutex_exit(&ixgbe->gen_lock);
1443 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1444 	return (IXGBE_FAILURE);
1445 }
1446 
1447 /*
1448  * ixgbe_chip_start - Initialize and start the chipset hardware.
1449  */
1450 static int
1451 ixgbe_chip_start(ixgbe_t *ixgbe)
1452 {
1453 	struct ixgbe_hw *hw = &ixgbe->hw;
1454 	int i;
1455 
1456 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1457 
1458 	/*
1459 	 * Get the mac address
1460 	 * This function should handle SPARC case correctly.
1461 	 */
1462 	if (!ixgbe_find_mac_address(ixgbe)) {
1463 		ixgbe_error(ixgbe, "Failed to get the mac address");
1464 		return (IXGBE_FAILURE);
1465 	}
1466 
1467 	/*
1468 	 * Validate the mac address
1469 	 */
1470 	(void) ixgbe_init_rx_addrs(hw);
1471 	if (!is_valid_mac_addr(hw->mac.addr)) {
1472 		ixgbe_error(ixgbe, "Invalid mac address");
1473 		return (IXGBE_FAILURE);
1474 	}
1475 
1476 	/*
1477 	 * Re-enable relaxed ordering for performance.  It is disabled
1478 	 * by default in the hardware init.
1479 	 */
1480 	if (ixgbe->relax_order_enable == B_TRUE)
1481 		ixgbe_enable_relaxed_ordering(hw);
1482 
1483 	/*
1484 	 * Setup adapter interrupt vectors
1485 	 */
1486 	ixgbe_setup_adapter_vector(ixgbe);
1487 
1488 	/*
1489 	 * Initialize unicast addresses.
1490 	 */
1491 	ixgbe_init_unicst(ixgbe);
1492 
1493 	/*
1494 	 * Setup and initialize the mctable structures.
1495 	 */
1496 	ixgbe_setup_multicst(ixgbe);
1497 
1498 	/*
1499 	 * Set interrupt throttling rate
1500 	 */
1501 	for (i = 0; i < ixgbe->intr_cnt; i++) {
1502 		IXGBE_WRITE_REG(hw, IXGBE_EITR(i), ixgbe->intr_throttling[i]);
1503 	}
1504 
1505 	/*
1506 	 * Disable Wake-on-LAN
1507 	 */
1508 	IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
1509 
1510 	/*
1511 	 * Some adapters offer Energy Efficient Ethernet (EEE) support.
1512 	 * Due to issues with EEE in e1000g/igb, we disable this by default
1513 	 * as a precautionary measure.
1514 	 *
1515 	 * Currently, the only known adapter which supports EEE in the ixgbe
1516 	 * line is 8086,15AB (IXGBE_DEV_ID_X550EM_X_KR), and only after the
1517 	 * first revision of it, as well as any X550 with MAC type 6 (non-EM)
1518 	 */
1519 	(void) ixgbe_setup_eee(hw, B_FALSE);
1520 
1521 	/*
1522 	 * Turn on any present SFP Tx laser
1523 	 */
1524 	ixgbe_enable_tx_laser(hw);
1525 
1526 	/*
1527 	 * Power on the PHY
1528 	 */
1529 	(void) ixgbe_set_phy_power(hw, B_TRUE);
1530 
1531 	/*
1532 	 * Save the state of the PHY
1533 	 */
1534 	ixgbe_get_hw_state(ixgbe);
1535 
1536 	/*
1537 	 * Make sure driver has control
1538 	 */
1539 	ixgbe_get_driver_control(hw);
1540 
1541 	return (IXGBE_SUCCESS);
1542 }
1543 
1544 /*
1545  * ixgbe_chip_stop - Stop the chipset hardware
1546  */
1547 static void
1548 ixgbe_chip_stop(ixgbe_t *ixgbe)
1549 {
1550 	struct ixgbe_hw *hw = &ixgbe->hw;
1551 	int rv;
1552 
1553 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1554 
1555 	/*
1556 	 * Stop interupt generation and disable Tx unit
1557 	 */
1558 	hw->adapter_stopped = B_FALSE;
1559 	(void) ixgbe_stop_adapter(hw);
1560 
1561 	/*
1562 	 * Reset the chipset
1563 	 */
1564 	(void) ixgbe_reset_hw(hw);
1565 
1566 	/*
1567 	 * Reset PHY
1568 	 */
1569 	(void) ixgbe_reset_phy(hw);
1570 
1571 	/*
1572 	 * Enter LPLU (Low Power, Link Up) mode, if available. Avoid resetting
1573 	 * the PHY while doing so. Else, just power down the PHY.
1574 	 */
1575 	if (hw->phy.ops.enter_lplu != NULL) {
1576 		hw->phy.reset_disable = B_TRUE;
1577 		rv = hw->phy.ops.enter_lplu(hw);
1578 		if (rv != IXGBE_SUCCESS)
1579 			ixgbe_error(ixgbe, "Error while entering LPLU: %d", rv);
1580 		hw->phy.reset_disable = B_FALSE;
1581 	} else {
1582 		(void) ixgbe_set_phy_power(hw, B_FALSE);
1583 	}
1584 
1585 	/*
1586 	 * Turn off any present SFP Tx laser
1587 	 * Expected for health and safety reasons
1588 	 */
1589 	ixgbe_disable_tx_laser(hw);
1590 
1591 	/*
1592 	 * Tell firmware driver is no longer in control
1593 	 */
1594 	ixgbe_release_driver_control(hw);
1595 
1596 }
1597 
1598 /*
1599  * ixgbe_reset - Reset the chipset and re-start the driver.
1600  *
1601  * It involves stopping and re-starting the chipset,
1602  * and re-configuring the rx/tx rings.
1603  */
1604 static int
1605 ixgbe_reset(ixgbe_t *ixgbe)
1606 {
1607 	int i;
1608 
1609 	/*
1610 	 * Disable and stop the watchdog timer
1611 	 */
1612 	ixgbe_disable_watchdog_timer(ixgbe);
1613 
1614 	mutex_enter(&ixgbe->gen_lock);
1615 
1616 	ASSERT(ixgbe->ixgbe_state & IXGBE_STARTED);
1617 	atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_STARTED);
1618 
1619 	ixgbe_stop(ixgbe, B_FALSE);
1620 
1621 	if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
1622 		mutex_exit(&ixgbe->gen_lock);
1623 		return (IXGBE_FAILURE);
1624 	}
1625 
1626 	/*
1627 	 * After resetting, need to recheck the link status.
1628 	 */
1629 	ixgbe->link_check_complete = B_FALSE;
1630 	ixgbe->link_check_hrtime = gethrtime() +
1631 	    (IXGBE_LINK_UP_TIME * 100000000ULL);
1632 
1633 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_STARTED);
1634 
1635 	if (!(ixgbe->ixgbe_state & IXGBE_SUSPENDED)) {
1636 		for (i = 0; i < ixgbe->num_tx_rings; i++) {
1637 			mac_tx_ring_update(ixgbe->mac_hdl,
1638 			    ixgbe->tx_rings[i].ring_handle);
1639 		}
1640 	}
1641 
1642 	mutex_exit(&ixgbe->gen_lock);
1643 
1644 	/*
1645 	 * Enable and start the watchdog timer
1646 	 */
1647 	ixgbe_enable_watchdog_timer(ixgbe);
1648 
1649 	return (IXGBE_SUCCESS);
1650 }
1651 
1652 /*
1653  * ixgbe_tx_clean - Clean the pending transmit packets and DMA resources.
1654  */
1655 static void
1656 ixgbe_tx_clean(ixgbe_t *ixgbe)
1657 {
1658 	ixgbe_tx_ring_t *tx_ring;
1659 	tx_control_block_t *tcb;
1660 	link_list_t pending_list;
1661 	uint32_t desc_num;
1662 	int i, j;
1663 
1664 	LINK_LIST_INIT(&pending_list);
1665 
1666 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1667 		tx_ring = &ixgbe->tx_rings[i];
1668 
1669 		mutex_enter(&tx_ring->recycle_lock);
1670 
1671 		/*
1672 		 * Clean the pending tx data - the pending packets in the
1673 		 * work_list that have no chances to be transmitted again.
1674 		 *
1675 		 * We must ensure the chipset is stopped or the link is down
1676 		 * before cleaning the transmit packets.
1677 		 */
1678 		desc_num = 0;
1679 		for (j = 0; j < tx_ring->ring_size; j++) {
1680 			tcb = tx_ring->work_list[j];
1681 			if (tcb != NULL) {
1682 				desc_num += tcb->desc_num;
1683 
1684 				tx_ring->work_list[j] = NULL;
1685 
1686 				ixgbe_free_tcb(tcb);
1687 
1688 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1689 			}
1690 		}
1691 
1692 		if (desc_num > 0) {
1693 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1694 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1695 
1696 			/*
1697 			 * Reset the head and tail pointers of the tbd ring;
1698 			 * Reset the writeback head if it's enable.
1699 			 */
1700 			tx_ring->tbd_head = 0;
1701 			tx_ring->tbd_tail = 0;
1702 			if (ixgbe->tx_head_wb_enable)
1703 				*tx_ring->tbd_head_wb = 0;
1704 
1705 			IXGBE_WRITE_REG(&ixgbe->hw,
1706 			    IXGBE_TDH(tx_ring->index), 0);
1707 			IXGBE_WRITE_REG(&ixgbe->hw,
1708 			    IXGBE_TDT(tx_ring->index), 0);
1709 		}
1710 
1711 		mutex_exit(&tx_ring->recycle_lock);
1712 
1713 		/*
1714 		 * Add the tx control blocks in the pending list to
1715 		 * the free list.
1716 		 */
1717 		ixgbe_put_free_list(tx_ring, &pending_list);
1718 	}
1719 }
1720 
1721 /*
1722  * ixgbe_tx_drain - Drain the tx rings to allow pending packets to be
1723  * transmitted.
1724  */
1725 static boolean_t
1726 ixgbe_tx_drain(ixgbe_t *ixgbe)
1727 {
1728 	ixgbe_tx_ring_t *tx_ring;
1729 	boolean_t done;
1730 	int i, j;
1731 
1732 	/*
1733 	 * Wait for a specific time to allow pending tx packets
1734 	 * to be transmitted.
1735 	 *
1736 	 * Check the counter tbd_free to see if transmission is done.
1737 	 * No lock protection is needed here.
1738 	 *
1739 	 * Return B_TRUE if all pending packets have been transmitted;
1740 	 * Otherwise return B_FALSE;
1741 	 */
1742 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1743 
1744 		done = B_TRUE;
1745 		for (j = 0; j < ixgbe->num_tx_rings; j++) {
1746 			tx_ring = &ixgbe->tx_rings[j];
1747 			done = done &&
1748 			    (tx_ring->tbd_free == tx_ring->ring_size);
1749 		}
1750 
1751 		if (done)
1752 			break;
1753 
1754 		msec_delay(1);
1755 	}
1756 
1757 	return (done);
1758 }
1759 
1760 /*
1761  * ixgbe_rx_drain - Wait for all rx buffers to be released by upper layer.
1762  */
1763 static boolean_t
1764 ixgbe_rx_drain(ixgbe_t *ixgbe)
1765 {
1766 	boolean_t done = B_TRUE;
1767 	int i;
1768 
1769 	/*
1770 	 * Polling the rx free list to check if those rx buffers held by
1771 	 * the upper layer are released.
1772 	 *
1773 	 * Check the counter rcb_free to see if all pending buffers are
1774 	 * released. No lock protection is needed here.
1775 	 *
1776 	 * Return B_TRUE if all pending buffers have been released;
1777 	 * Otherwise return B_FALSE;
1778 	 */
1779 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1780 		done = (ixgbe->rcb_pending == 0);
1781 
1782 		if (done)
1783 			break;
1784 
1785 		msec_delay(1);
1786 	}
1787 
1788 	return (done);
1789 }
1790 
1791 /*
1792  * ixgbe_start - Start the driver/chipset.
1793  */
1794 int
1795 ixgbe_start(ixgbe_t *ixgbe, boolean_t alloc_buffer)
1796 {
1797 	struct ixgbe_hw *hw = &ixgbe->hw;
1798 	int i;
1799 
1800 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1801 
1802 	if (alloc_buffer) {
1803 		if (ixgbe_alloc_rx_data(ixgbe) != IXGBE_SUCCESS) {
1804 			ixgbe_error(ixgbe,
1805 			    "Failed to allocate software receive rings");
1806 			return (IXGBE_FAILURE);
1807 		}
1808 
1809 		/* Allocate buffers for all the rx/tx rings */
1810 		if (ixgbe_alloc_dma(ixgbe) != IXGBE_SUCCESS) {
1811 			ixgbe_error(ixgbe, "Failed to allocate DMA resource");
1812 			return (IXGBE_FAILURE);
1813 		}
1814 
1815 		ixgbe->tx_ring_init = B_TRUE;
1816 	} else {
1817 		ixgbe->tx_ring_init = B_FALSE;
1818 	}
1819 
1820 	for (i = 0; i < ixgbe->num_rx_rings; i++)
1821 		mutex_enter(&ixgbe->rx_rings[i].rx_lock);
1822 	for (i = 0; i < ixgbe->num_tx_rings; i++)
1823 		mutex_enter(&ixgbe->tx_rings[i].tx_lock);
1824 
1825 	/*
1826 	 * Start the chipset hardware
1827 	 */
1828 	if (ixgbe_chip_start(ixgbe) != IXGBE_SUCCESS) {
1829 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1830 		goto start_failure;
1831 	}
1832 
1833 	/*
1834 	 * Configure link now for X550
1835 	 *
1836 	 * X550 possesses a LPLU (Low-Power Link Up) mode which keeps the
1837 	 * resting state of the adapter at a 1Gb FDX speed. Prior to the X550,
1838 	 * the resting state of the link would be the maximum speed that
1839 	 * autonegotiation will allow (usually 10Gb, infrastructure allowing)
1840 	 * so we never bothered with explicitly setting the link to 10Gb as it
1841 	 * would already be at that state on driver attach. With X550, we must
1842 	 * trigger a re-negotiation of the link in order to switch from a LPLU
1843 	 * 1Gb link to 10Gb (cable and link partner permitting.)
1844 	 */
1845 	if (hw->mac.type == ixgbe_mac_X550 ||
1846 	    hw->mac.type == ixgbe_mac_X550EM_x) {
1847 		(void) ixgbe_driver_setup_link(ixgbe, B_TRUE);
1848 		ixgbe_get_hw_state(ixgbe);
1849 	}
1850 
1851 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1852 		goto start_failure;
1853 	}
1854 
1855 	/*
1856 	 * Setup the rx/tx rings
1857 	 */
1858 	ixgbe_setup_rings(ixgbe);
1859 
1860 	/*
1861 	 * ixgbe_start() will be called when resetting, however if reset
1862 	 * happens, we need to clear the ERROR, STALL and OVERTEMP flags
1863 	 * before enabling the interrupts.
1864 	 */
1865 	atomic_and_32(&ixgbe->ixgbe_state, ~(IXGBE_ERROR
1866 	    | IXGBE_STALL| IXGBE_OVERTEMP));
1867 
1868 	/*
1869 	 * Enable adapter interrupts
1870 	 * The interrupts must be enabled after the driver state is START
1871 	 */
1872 	ixgbe_enable_adapter_interrupts(ixgbe);
1873 
1874 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1875 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1876 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1877 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1878 
1879 	return (IXGBE_SUCCESS);
1880 
1881 start_failure:
1882 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1883 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1884 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1885 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1886 
1887 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1888 
1889 	return (IXGBE_FAILURE);
1890 }
1891 
1892 /*
1893  * ixgbe_stop - Stop the driver/chipset.
1894  */
1895 void
1896 ixgbe_stop(ixgbe_t *ixgbe, boolean_t free_buffer)
1897 {
1898 	int i;
1899 
1900 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1901 
1902 	/*
1903 	 * Disable the adapter interrupts
1904 	 */
1905 	ixgbe_disable_adapter_interrupts(ixgbe);
1906 
1907 	/*
1908 	 * Drain the pending tx packets
1909 	 */
1910 	(void) ixgbe_tx_drain(ixgbe);
1911 
1912 	for (i = 0; i < ixgbe->num_rx_rings; i++)
1913 		mutex_enter(&ixgbe->rx_rings[i].rx_lock);
1914 	for (i = 0; i < ixgbe->num_tx_rings; i++)
1915 		mutex_enter(&ixgbe->tx_rings[i].tx_lock);
1916 
1917 	/*
1918 	 * Stop the chipset hardware
1919 	 */
1920 	ixgbe_chip_stop(ixgbe);
1921 
1922 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1923 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1924 	}
1925 
1926 	/*
1927 	 * Clean the pending tx data/resources
1928 	 */
1929 	ixgbe_tx_clean(ixgbe);
1930 
1931 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1932 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1933 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1934 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1935 
1936 	if (ixgbe->link_state == LINK_STATE_UP) {
1937 		ixgbe->link_state = LINK_STATE_UNKNOWN;
1938 		mac_link_update(ixgbe->mac_hdl, ixgbe->link_state);
1939 	}
1940 
1941 	if (free_buffer) {
1942 		/*
1943 		 * Release the DMA/memory resources of rx/tx rings
1944 		 */
1945 		ixgbe_free_dma(ixgbe);
1946 		ixgbe_free_rx_data(ixgbe);
1947 	}
1948 }
1949 
1950 /*
1951  * ixgbe_cbfunc - Driver interface for generic DDI callbacks
1952  */
1953 /* ARGSUSED */
1954 static int
1955 ixgbe_cbfunc(dev_info_t *dip, ddi_cb_action_t cbaction, void *cbarg,
1956     void *arg1, void *arg2)
1957 {
1958 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
1959 
1960 	switch (cbaction) {
1961 	/* IRM callback */
1962 	int count;
1963 	case DDI_CB_INTR_ADD:
1964 	case DDI_CB_INTR_REMOVE:
1965 		count = (int)(uintptr_t)cbarg;
1966 		ASSERT(ixgbe->intr_type == DDI_INTR_TYPE_MSIX);
1967 		DTRACE_PROBE2(ixgbe__irm__callback, int, count,
1968 		    int, ixgbe->intr_cnt);
1969 		if (ixgbe_intr_adjust(ixgbe, cbaction, count) !=
1970 		    DDI_SUCCESS) {
1971 			ixgbe_error(ixgbe,
1972 			    "IRM CB: Failed to adjust interrupts");
1973 			goto cb_fail;
1974 		}
1975 		break;
1976 	default:
1977 		IXGBE_DEBUGLOG_1(ixgbe, "DDI CB: action 0x%x NOT supported",
1978 		    cbaction);
1979 		return (DDI_ENOTSUP);
1980 	}
1981 	return (DDI_SUCCESS);
1982 cb_fail:
1983 	return (DDI_FAILURE);
1984 }
1985 
1986 /*
1987  * ixgbe_intr_adjust - Adjust interrupt to respond to IRM request.
1988  */
1989 static int
1990 ixgbe_intr_adjust(ixgbe_t *ixgbe, ddi_cb_action_t cbaction, int count)
1991 {
1992 	int i, rc, actual;
1993 
1994 	if (count == 0)
1995 		return (DDI_SUCCESS);
1996 
1997 	if ((cbaction == DDI_CB_INTR_ADD &&
1998 	    ixgbe->intr_cnt + count > ixgbe->intr_cnt_max) ||
1999 	    (cbaction == DDI_CB_INTR_REMOVE &&
2000 	    ixgbe->intr_cnt - count < ixgbe->intr_cnt_min))
2001 		return (DDI_FAILURE);
2002 
2003 	if (!(ixgbe->ixgbe_state & IXGBE_STARTED)) {
2004 		return (DDI_FAILURE);
2005 	}
2006 
2007 	for (i = 0; i < ixgbe->num_rx_rings; i++)
2008 		mac_ring_intr_set(ixgbe->rx_rings[i].ring_handle, NULL);
2009 	for (i = 0; i < ixgbe->num_tx_rings; i++)
2010 		mac_ring_intr_set(ixgbe->tx_rings[i].ring_handle, NULL);
2011 
2012 	mutex_enter(&ixgbe->gen_lock);
2013 	ixgbe->ixgbe_state &= ~IXGBE_STARTED;
2014 	ixgbe->ixgbe_state |= IXGBE_INTR_ADJUST;
2015 	ixgbe->ixgbe_state |= IXGBE_SUSPENDED;
2016 	mac_link_update(ixgbe->mac_hdl, LINK_STATE_UNKNOWN);
2017 
2018 	ixgbe_stop(ixgbe, B_FALSE);
2019 	/*
2020 	 * Disable interrupts
2021 	 */
2022 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
2023 		rc = ixgbe_disable_intrs(ixgbe);
2024 		ASSERT(rc == IXGBE_SUCCESS);
2025 	}
2026 	ixgbe->attach_progress &= ~ATTACH_PROGRESS_ENABLE_INTR;
2027 
2028 	/*
2029 	 * Remove interrupt handlers
2030 	 */
2031 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
2032 		ixgbe_rem_intr_handlers(ixgbe);
2033 	}
2034 	ixgbe->attach_progress &= ~ATTACH_PROGRESS_ADD_INTR;
2035 
2036 	/*
2037 	 * Clear vect_map
2038 	 */
2039 	bzero(&ixgbe->vect_map, sizeof (ixgbe->vect_map));
2040 	switch (cbaction) {
2041 	case DDI_CB_INTR_ADD:
2042 		rc = ddi_intr_alloc(ixgbe->dip, ixgbe->htable,
2043 		    DDI_INTR_TYPE_MSIX, ixgbe->intr_cnt, count, &actual,
2044 		    DDI_INTR_ALLOC_NORMAL);
2045 		if (rc != DDI_SUCCESS || actual != count) {
2046 			ixgbe_log(ixgbe, "Adjust interrupts failed."
2047 			    "return: %d, irm cb size: %d, actual: %d",
2048 			    rc, count, actual);
2049 			goto intr_adjust_fail;
2050 		}
2051 		ixgbe->intr_cnt += count;
2052 		break;
2053 
2054 	case DDI_CB_INTR_REMOVE:
2055 		for (i = ixgbe->intr_cnt - count;
2056 		    i < ixgbe->intr_cnt; i ++) {
2057 			rc = ddi_intr_free(ixgbe->htable[i]);
2058 			ixgbe->htable[i] = NULL;
2059 			if (rc != DDI_SUCCESS) {
2060 				ixgbe_log(ixgbe, "Adjust interrupts failed."
2061 				    "return: %d, irm cb size: %d, actual: %d",
2062 				    rc, count, actual);
2063 				goto intr_adjust_fail;
2064 			}
2065 		}
2066 		ixgbe->intr_cnt -= count;
2067 		break;
2068 	}
2069 
2070 	/*
2071 	 * Get priority for first vector, assume remaining are all the same
2072 	 */
2073 	rc = ddi_intr_get_pri(ixgbe->htable[0], &ixgbe->intr_pri);
2074 	if (rc != DDI_SUCCESS) {
2075 		ixgbe_log(ixgbe,
2076 		    "Get interrupt priority failed: %d", rc);
2077 		goto intr_adjust_fail;
2078 	}
2079 	rc = ddi_intr_get_cap(ixgbe->htable[0], &ixgbe->intr_cap);
2080 	if (rc != DDI_SUCCESS) {
2081 		ixgbe_log(ixgbe, "Get interrupt cap failed: %d", rc);
2082 		goto intr_adjust_fail;
2083 	}
2084 	ixgbe->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
2085 
2086 	/*
2087 	 * Map rings to interrupt vectors
2088 	 */
2089 	if (ixgbe_map_intrs_to_vectors(ixgbe) != IXGBE_SUCCESS) {
2090 		ixgbe_error(ixgbe,
2091 		    "IRM CB: Failed to map interrupts to vectors");
2092 		goto intr_adjust_fail;
2093 	}
2094 
2095 	/*
2096 	 * Add interrupt handlers
2097 	 */
2098 	if (ixgbe_add_intr_handlers(ixgbe) != IXGBE_SUCCESS) {
2099 		ixgbe_error(ixgbe, "IRM CB: Failed to add interrupt handlers");
2100 		goto intr_adjust_fail;
2101 	}
2102 	ixgbe->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
2103 
2104 	/*
2105 	 * Now that mutex locks are initialized, and the chip is also
2106 	 * initialized, enable interrupts.
2107 	 */
2108 	if (ixgbe_enable_intrs(ixgbe) != IXGBE_SUCCESS) {
2109 		ixgbe_error(ixgbe, "IRM CB: Failed to enable DDI interrupts");
2110 		goto intr_adjust_fail;
2111 	}
2112 	ixgbe->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
2113 	if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
2114 		ixgbe_error(ixgbe, "IRM CB: Failed to start");
2115 		goto intr_adjust_fail;
2116 	}
2117 	ixgbe->ixgbe_state &= ~IXGBE_INTR_ADJUST;
2118 	ixgbe->ixgbe_state &= ~IXGBE_SUSPENDED;
2119 	ixgbe->ixgbe_state |= IXGBE_STARTED;
2120 	mutex_exit(&ixgbe->gen_lock);
2121 
2122 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2123 		mac_ring_intr_set(ixgbe->rx_rings[i].ring_handle,
2124 		    ixgbe->htable[ixgbe->rx_rings[i].intr_vector]);
2125 	}
2126 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2127 		mac_ring_intr_set(ixgbe->tx_rings[i].ring_handle,
2128 		    ixgbe->htable[ixgbe->tx_rings[i].intr_vector]);
2129 	}
2130 
2131 	/* Wakeup all Tx rings */
2132 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2133 		mac_tx_ring_update(ixgbe->mac_hdl,
2134 		    ixgbe->tx_rings[i].ring_handle);
2135 	}
2136 
2137 	IXGBE_DEBUGLOG_3(ixgbe,
2138 	    "IRM CB: interrupts new value: 0x%x(0x%x:0x%x).",
2139 	    ixgbe->intr_cnt, ixgbe->intr_cnt_min, ixgbe->intr_cnt_max);
2140 	return (DDI_SUCCESS);
2141 
2142 intr_adjust_fail:
2143 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
2144 	mutex_exit(&ixgbe->gen_lock);
2145 	return (DDI_FAILURE);
2146 }
2147 
2148 /*
2149  * ixgbe_intr_cb_register - Register interrupt callback function.
2150  */
2151 static int
2152 ixgbe_intr_cb_register(ixgbe_t *ixgbe)
2153 {
2154 	if (ddi_cb_register(ixgbe->dip, DDI_CB_FLAG_INTR, ixgbe_cbfunc,
2155 	    ixgbe, NULL, &ixgbe->cb_hdl) != DDI_SUCCESS) {
2156 		return (IXGBE_FAILURE);
2157 	}
2158 	IXGBE_DEBUGLOG_0(ixgbe, "Interrupt callback function registered.");
2159 	return (IXGBE_SUCCESS);
2160 }
2161 
2162 /*
2163  * ixgbe_alloc_rings - Allocate memory space for rx/tx rings.
2164  */
2165 static int
2166 ixgbe_alloc_rings(ixgbe_t *ixgbe)
2167 {
2168 	/*
2169 	 * Allocate memory space for rx rings
2170 	 */
2171 	ixgbe->rx_rings = kmem_zalloc(
2172 	    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings,
2173 	    KM_NOSLEEP);
2174 
2175 	if (ixgbe->rx_rings == NULL) {
2176 		return (IXGBE_FAILURE);
2177 	}
2178 
2179 	/*
2180 	 * Allocate memory space for tx rings
2181 	 */
2182 	ixgbe->tx_rings = kmem_zalloc(
2183 	    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings,
2184 	    KM_NOSLEEP);
2185 
2186 	if (ixgbe->tx_rings == NULL) {
2187 		kmem_free(ixgbe->rx_rings,
2188 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
2189 		ixgbe->rx_rings = NULL;
2190 		return (IXGBE_FAILURE);
2191 	}
2192 
2193 	/*
2194 	 * Allocate memory space for rx ring groups
2195 	 */
2196 	ixgbe->rx_groups = kmem_zalloc(
2197 	    sizeof (ixgbe_rx_group_t) * ixgbe->num_rx_groups,
2198 	    KM_NOSLEEP);
2199 
2200 	if (ixgbe->rx_groups == NULL) {
2201 		kmem_free(ixgbe->rx_rings,
2202 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
2203 		kmem_free(ixgbe->tx_rings,
2204 		    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings);
2205 		ixgbe->rx_rings = NULL;
2206 		ixgbe->tx_rings = NULL;
2207 		return (IXGBE_FAILURE);
2208 	}
2209 
2210 	return (IXGBE_SUCCESS);
2211 }
2212 
2213 /*
2214  * ixgbe_free_rings - Free the memory space of rx/tx rings.
2215  */
2216 static void
2217 ixgbe_free_rings(ixgbe_t *ixgbe)
2218 {
2219 	if (ixgbe->rx_rings != NULL) {
2220 		kmem_free(ixgbe->rx_rings,
2221 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
2222 		ixgbe->rx_rings = NULL;
2223 	}
2224 
2225 	if (ixgbe->tx_rings != NULL) {
2226 		kmem_free(ixgbe->tx_rings,
2227 		    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings);
2228 		ixgbe->tx_rings = NULL;
2229 	}
2230 
2231 	if (ixgbe->rx_groups != NULL) {
2232 		kmem_free(ixgbe->rx_groups,
2233 		    sizeof (ixgbe_rx_group_t) * ixgbe->num_rx_groups);
2234 		ixgbe->rx_groups = NULL;
2235 	}
2236 }
2237 
2238 static int
2239 ixgbe_alloc_rx_data(ixgbe_t *ixgbe)
2240 {
2241 	ixgbe_rx_ring_t *rx_ring;
2242 	int i;
2243 
2244 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2245 		rx_ring = &ixgbe->rx_rings[i];
2246 		if (ixgbe_alloc_rx_ring_data(rx_ring) != IXGBE_SUCCESS)
2247 			goto alloc_rx_rings_failure;
2248 	}
2249 	return (IXGBE_SUCCESS);
2250 
2251 alloc_rx_rings_failure:
2252 	ixgbe_free_rx_data(ixgbe);
2253 	return (IXGBE_FAILURE);
2254 }
2255 
2256 static void
2257 ixgbe_free_rx_data(ixgbe_t *ixgbe)
2258 {
2259 	ixgbe_rx_ring_t *rx_ring;
2260 	ixgbe_rx_data_t *rx_data;
2261 	int i;
2262 
2263 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2264 		rx_ring = &ixgbe->rx_rings[i];
2265 
2266 		mutex_enter(&ixgbe->rx_pending_lock);
2267 		rx_data = rx_ring->rx_data;
2268 
2269 		if (rx_data != NULL) {
2270 			rx_data->flag |= IXGBE_RX_STOPPED;
2271 
2272 			if (rx_data->rcb_pending == 0) {
2273 				ixgbe_free_rx_ring_data(rx_data);
2274 				rx_ring->rx_data = NULL;
2275 			}
2276 		}
2277 
2278 		mutex_exit(&ixgbe->rx_pending_lock);
2279 	}
2280 }
2281 
2282 /*
2283  * ixgbe_setup_rings - Setup rx/tx rings.
2284  */
2285 static void
2286 ixgbe_setup_rings(ixgbe_t *ixgbe)
2287 {
2288 	/*
2289 	 * Setup the rx/tx rings, including the following:
2290 	 *
2291 	 * 1. Setup the descriptor ring and the control block buffers;
2292 	 * 2. Initialize necessary registers for receive/transmit;
2293 	 * 3. Initialize software pointers/parameters for receive/transmit;
2294 	 */
2295 	ixgbe_setup_rx(ixgbe);
2296 
2297 	ixgbe_setup_tx(ixgbe);
2298 }
2299 
2300 static void
2301 ixgbe_setup_rx_ring(ixgbe_rx_ring_t *rx_ring)
2302 {
2303 	ixgbe_t *ixgbe = rx_ring->ixgbe;
2304 	ixgbe_rx_data_t *rx_data = rx_ring->rx_data;
2305 	struct ixgbe_hw *hw = &ixgbe->hw;
2306 	rx_control_block_t *rcb;
2307 	union ixgbe_adv_rx_desc	*rbd;
2308 	uint32_t size;
2309 	uint32_t buf_low;
2310 	uint32_t buf_high;
2311 	uint32_t reg_val;
2312 	int i;
2313 
2314 	ASSERT(mutex_owned(&rx_ring->rx_lock));
2315 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2316 
2317 	for (i = 0; i < ixgbe->rx_ring_size; i++) {
2318 		rcb = rx_data->work_list[i];
2319 		rbd = &rx_data->rbd_ring[i];
2320 
2321 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
2322 		rbd->read.hdr_addr = NULL;
2323 	}
2324 
2325 	/*
2326 	 * Initialize the length register
2327 	 */
2328 	size = rx_data->ring_size * sizeof (union ixgbe_adv_rx_desc);
2329 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(rx_ring->hw_index), size);
2330 
2331 	/*
2332 	 * Initialize the base address registers
2333 	 */
2334 	buf_low = (uint32_t)rx_data->rbd_area.dma_address;
2335 	buf_high = (uint32_t)(rx_data->rbd_area.dma_address >> 32);
2336 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(rx_ring->hw_index), buf_high);
2337 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(rx_ring->hw_index), buf_low);
2338 
2339 	/*
2340 	 * Setup head & tail pointers
2341 	 */
2342 	IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->hw_index),
2343 	    rx_data->ring_size - 1);
2344 	IXGBE_WRITE_REG(hw, IXGBE_RDH(rx_ring->hw_index), 0);
2345 
2346 	rx_data->rbd_next = 0;
2347 	rx_data->lro_first = 0;
2348 
2349 	/*
2350 	 * Setup the Receive Descriptor Control Register (RXDCTL)
2351 	 * PTHRESH=32 descriptors (half the internal cache)
2352 	 * HTHRESH=0 descriptors (to minimize latency on fetch)
2353 	 * WTHRESH defaults to 1 (writeback each descriptor)
2354 	 */
2355 	reg_val = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rx_ring->hw_index));
2356 	reg_val |= IXGBE_RXDCTL_ENABLE;	/* enable queue */
2357 
2358 	/* Not a valid value for 82599, X540 or X550 */
2359 	if (hw->mac.type == ixgbe_mac_82598EB) {
2360 		reg_val |= 0x0020;	/* pthresh */
2361 	}
2362 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rx_ring->hw_index), reg_val);
2363 
2364 	if (hw->mac.type == ixgbe_mac_82599EB ||
2365 	    hw->mac.type == ixgbe_mac_X540 ||
2366 	    hw->mac.type == ixgbe_mac_X550 ||
2367 	    hw->mac.type == ixgbe_mac_X550EM_x) {
2368 		reg_val = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2369 		reg_val |= (IXGBE_RDRXCTL_CRCSTRIP | IXGBE_RDRXCTL_AGGDIS);
2370 		IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, reg_val);
2371 	}
2372 
2373 	/*
2374 	 * Setup the Split and Replication Receive Control Register.
2375 	 * Set the rx buffer size and the advanced descriptor type.
2376 	 */
2377 	reg_val = (ixgbe->rx_buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) |
2378 	    IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2379 	reg_val |= IXGBE_SRRCTL_DROP_EN;
2380 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rx_ring->hw_index), reg_val);
2381 }
2382 
2383 static void
2384 ixgbe_setup_rx(ixgbe_t *ixgbe)
2385 {
2386 	ixgbe_rx_ring_t *rx_ring;
2387 	struct ixgbe_hw *hw = &ixgbe->hw;
2388 	uint32_t reg_val;
2389 	uint32_t ring_mapping;
2390 	uint32_t i, index;
2391 	uint32_t psrtype_rss_bit;
2392 
2393 	/*
2394 	 * Ensure that Rx is disabled while setting up
2395 	 * the Rx unit and Rx descriptor ring(s)
2396 	 */
2397 	ixgbe_disable_rx(hw);
2398 
2399 	/* PSRTYPE must be configured for 82599 */
2400 	if (ixgbe->classify_mode != IXGBE_CLASSIFY_VMDQ &&
2401 	    ixgbe->classify_mode != IXGBE_CLASSIFY_VMDQ_RSS) {
2402 		reg_val = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
2403 		    IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR;
2404 		reg_val |= IXGBE_PSRTYPE_L2HDR;
2405 		reg_val |= 0x80000000;
2406 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), reg_val);
2407 	} else {
2408 		if (ixgbe->num_rx_groups > 32) {
2409 			psrtype_rss_bit = 0x20000000;
2410 		} else {
2411 			psrtype_rss_bit = 0x40000000;
2412 		}
2413 		for (i = 0; i < ixgbe->capab->max_rx_grp_num; i++) {
2414 			reg_val = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
2415 			    IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR;
2416 			reg_val |= IXGBE_PSRTYPE_L2HDR;
2417 			reg_val |= psrtype_rss_bit;
2418 			IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(i), reg_val);
2419 		}
2420 	}
2421 
2422 	/*
2423 	 * Set filter control in FCTRL to determine types of packets are passed
2424 	 * up to the driver.
2425 	 * - Pass broadcast packets.
2426 	 * - Do not pass flow control pause frames (82598-specific)
2427 	 */
2428 	reg_val = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2429 	reg_val |= IXGBE_FCTRL_BAM; /* Broadcast Accept Mode */
2430 	if (hw->mac.type == ixgbe_mac_82598EB) {
2431 		reg_val |= IXGBE_FCTRL_DPF; /* Discard Pause Frames */
2432 	}
2433 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, reg_val);
2434 
2435 	/*
2436 	 * Hardware checksum settings
2437 	 */
2438 	if (ixgbe->rx_hcksum_enable) {
2439 		reg_val = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2440 		reg_val |= IXGBE_RXCSUM_IPPCSE;	/* IP checksum */
2441 		IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, reg_val);
2442 	}
2443 
2444 	/*
2445 	 * Setup VMDq and RSS for multiple receive queues
2446 	 */
2447 	switch (ixgbe->classify_mode) {
2448 	case IXGBE_CLASSIFY_RSS:
2449 		/*
2450 		 * One group, only RSS is needed when more than
2451 		 * one ring enabled.
2452 		 */
2453 		ixgbe_setup_rss(ixgbe);
2454 		break;
2455 
2456 	case IXGBE_CLASSIFY_VMDQ:
2457 		/*
2458 		 * Multiple groups, each group has one ring,
2459 		 * only VMDq is needed.
2460 		 */
2461 		ixgbe_setup_vmdq(ixgbe);
2462 		break;
2463 
2464 	case IXGBE_CLASSIFY_VMDQ_RSS:
2465 		/*
2466 		 * Multiple groups and multiple rings, both
2467 		 * VMDq and RSS are needed.
2468 		 */
2469 		ixgbe_setup_vmdq_rss(ixgbe);
2470 		break;
2471 
2472 	default:
2473 		break;
2474 	}
2475 
2476 	/*
2477 	 * Enable the receive unit.  This must be done after filter
2478 	 * control is set in FCTRL. On 82598, we disable the descriptor monitor.
2479 	 * 82598 is the only adapter which defines this RXCTRL option.
2480 	 */
2481 	reg_val = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2482 	if (hw->mac.type == ixgbe_mac_82598EB)
2483 		reg_val |= IXGBE_RXCTRL_DMBYPS; /* descriptor monitor bypass */
2484 	reg_val |= IXGBE_RXCTRL_RXEN;
2485 	(void) ixgbe_enable_rx_dma(hw, reg_val);
2486 
2487 	/*
2488 	 * ixgbe_setup_rx_ring must be called after configuring RXCTRL
2489 	 */
2490 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2491 		rx_ring = &ixgbe->rx_rings[i];
2492 		ixgbe_setup_rx_ring(rx_ring);
2493 	}
2494 
2495 	/*
2496 	 * Setup the per-ring statistics mapping.
2497 	 */
2498 	ring_mapping = 0;
2499 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2500 		index = ixgbe->rx_rings[i].hw_index;
2501 		ring_mapping = IXGBE_READ_REG(hw, IXGBE_RQSMR(index >> 2));
2502 		ring_mapping |= (i & 0xF) << (8 * (index & 0x3));
2503 		IXGBE_WRITE_REG(hw, IXGBE_RQSMR(index >> 2), ring_mapping);
2504 	}
2505 
2506 	/*
2507 	 * The Max Frame Size in MHADD/MAXFRS will be internally increased
2508 	 * by four bytes if the packet has a VLAN field, so includes MTU,
2509 	 * ethernet header and frame check sequence.
2510 	 * Register is MAXFRS in 82599.
2511 	 */
2512 	reg_val = IXGBE_READ_REG(hw, IXGBE_MHADD);
2513 	reg_val &= ~IXGBE_MHADD_MFS_MASK;
2514 	reg_val |= (ixgbe->default_mtu + sizeof (struct ether_header)
2515 	    + ETHERFCSL) << IXGBE_MHADD_MFS_SHIFT;
2516 	IXGBE_WRITE_REG(hw, IXGBE_MHADD, reg_val);
2517 
2518 	/*
2519 	 * Setup Jumbo Frame enable bit
2520 	 */
2521 	reg_val = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2522 	if (ixgbe->default_mtu > ETHERMTU)
2523 		reg_val |= IXGBE_HLREG0_JUMBOEN;
2524 	else
2525 		reg_val &= ~IXGBE_HLREG0_JUMBOEN;
2526 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, reg_val);
2527 
2528 	/*
2529 	 * Setup RSC for multiple receive queues.
2530 	 */
2531 	if (ixgbe->lro_enable) {
2532 		for (i = 0; i < ixgbe->num_rx_rings; i++) {
2533 			/*
2534 			 * Make sure rx_buf_size * MAXDESC not greater
2535 			 * than 65535.
2536 			 * Intel recommends 4 for MAXDESC field value.
2537 			 */
2538 			reg_val = IXGBE_READ_REG(hw, IXGBE_RSCCTL(i));
2539 			reg_val |= IXGBE_RSCCTL_RSCEN;
2540 			if (ixgbe->rx_buf_size == IXGBE_PKG_BUF_16k)
2541 				reg_val |= IXGBE_RSCCTL_MAXDESC_1;
2542 			else
2543 				reg_val |= IXGBE_RSCCTL_MAXDESC_4;
2544 			IXGBE_WRITE_REG(hw,  IXGBE_RSCCTL(i), reg_val);
2545 		}
2546 
2547 		reg_val = IXGBE_READ_REG(hw, IXGBE_RSCDBU);
2548 		reg_val |= IXGBE_RSCDBU_RSCACKDIS;
2549 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU, reg_val);
2550 
2551 		reg_val = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2552 		reg_val |= IXGBE_RDRXCTL_RSCACKC;
2553 		reg_val |= IXGBE_RDRXCTL_FCOE_WRFIX;
2554 		reg_val &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2555 
2556 		IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, reg_val);
2557 	}
2558 }
2559 
2560 static void
2561 ixgbe_setup_tx_ring(ixgbe_tx_ring_t *tx_ring)
2562 {
2563 	ixgbe_t *ixgbe = tx_ring->ixgbe;
2564 	struct ixgbe_hw *hw = &ixgbe->hw;
2565 	uint32_t size;
2566 	uint32_t buf_low;
2567 	uint32_t buf_high;
2568 	uint32_t reg_val;
2569 
2570 	ASSERT(mutex_owned(&tx_ring->tx_lock));
2571 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2572 
2573 	/*
2574 	 * Initialize the length register
2575 	 */
2576 	size = tx_ring->ring_size * sizeof (union ixgbe_adv_tx_desc);
2577 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(tx_ring->index), size);
2578 
2579 	/*
2580 	 * Initialize the base address registers
2581 	 */
2582 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
2583 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
2584 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(tx_ring->index), buf_low);
2585 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(tx_ring->index), buf_high);
2586 
2587 	/*
2588 	 * Setup head & tail pointers
2589 	 */
2590 	IXGBE_WRITE_REG(hw, IXGBE_TDH(tx_ring->index), 0);
2591 	IXGBE_WRITE_REG(hw, IXGBE_TDT(tx_ring->index), 0);
2592 
2593 	/*
2594 	 * Setup head write-back
2595 	 */
2596 	if (ixgbe->tx_head_wb_enable) {
2597 		/*
2598 		 * The memory of the head write-back is allocated using
2599 		 * the extra tbd beyond the tail of the tbd ring.
2600 		 */
2601 		tx_ring->tbd_head_wb = (uint32_t *)
2602 		    ((uintptr_t)tx_ring->tbd_area.address + size);
2603 		*tx_ring->tbd_head_wb = 0;
2604 
2605 		buf_low = (uint32_t)
2606 		    (tx_ring->tbd_area.dma_address + size);
2607 		buf_high = (uint32_t)
2608 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
2609 
2610 		/* Set the head write-back enable bit */
2611 		buf_low |= IXGBE_TDWBAL_HEAD_WB_ENABLE;
2612 
2613 		IXGBE_WRITE_REG(hw, IXGBE_TDWBAL(tx_ring->index), buf_low);
2614 		IXGBE_WRITE_REG(hw, IXGBE_TDWBAH(tx_ring->index), buf_high);
2615 
2616 		/*
2617 		 * Turn off relaxed ordering for head write back or it will
2618 		 * cause problems with the tx recycling
2619 		 */
2620 
2621 		reg_val = (hw->mac.type == ixgbe_mac_82598EB) ?
2622 		    IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(tx_ring->index)) :
2623 		    IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(tx_ring->index));
2624 		reg_val &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
2625 		if (hw->mac.type == ixgbe_mac_82598EB) {
2626 			IXGBE_WRITE_REG(hw,
2627 			    IXGBE_DCA_TXCTRL(tx_ring->index), reg_val);
2628 		} else {
2629 			IXGBE_WRITE_REG(hw,
2630 			    IXGBE_DCA_TXCTRL_82599(tx_ring->index), reg_val);
2631 		}
2632 	} else {
2633 		tx_ring->tbd_head_wb = NULL;
2634 	}
2635 
2636 	tx_ring->tbd_head = 0;
2637 	tx_ring->tbd_tail = 0;
2638 	tx_ring->tbd_free = tx_ring->ring_size;
2639 
2640 	if (ixgbe->tx_ring_init == B_TRUE) {
2641 		tx_ring->tcb_head = 0;
2642 		tx_ring->tcb_tail = 0;
2643 		tx_ring->tcb_free = tx_ring->free_list_size;
2644 	}
2645 
2646 	/*
2647 	 * Initialize the s/w context structure
2648 	 */
2649 	bzero(&tx_ring->tx_context, sizeof (ixgbe_tx_context_t));
2650 }
2651 
2652 static void
2653 ixgbe_setup_tx(ixgbe_t *ixgbe)
2654 {
2655 	struct ixgbe_hw *hw = &ixgbe->hw;
2656 	ixgbe_tx_ring_t *tx_ring;
2657 	uint32_t reg_val;
2658 	uint32_t ring_mapping;
2659 	int i;
2660 
2661 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2662 		tx_ring = &ixgbe->tx_rings[i];
2663 		ixgbe_setup_tx_ring(tx_ring);
2664 	}
2665 
2666 	/*
2667 	 * Setup the per-ring statistics mapping.
2668 	 */
2669 	ring_mapping = 0;
2670 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2671 		ring_mapping |= (i & 0xF) << (8 * (i & 0x3));
2672 		if ((i & 0x3) == 0x3) {
2673 			switch (hw->mac.type) {
2674 			case ixgbe_mac_82598EB:
2675 				IXGBE_WRITE_REG(hw, IXGBE_TQSMR(i >> 2),
2676 				    ring_mapping);
2677 				break;
2678 
2679 			case ixgbe_mac_82599EB:
2680 			case ixgbe_mac_X540:
2681 			case ixgbe_mac_X550:
2682 			case ixgbe_mac_X550EM_x:
2683 				IXGBE_WRITE_REG(hw, IXGBE_TQSM(i >> 2),
2684 				    ring_mapping);
2685 				break;
2686 
2687 			default:
2688 				break;
2689 			}
2690 
2691 			ring_mapping = 0;
2692 		}
2693 	}
2694 	if (i & 0x3) {
2695 		switch (hw->mac.type) {
2696 		case ixgbe_mac_82598EB:
2697 			IXGBE_WRITE_REG(hw, IXGBE_TQSMR(i >> 2), ring_mapping);
2698 			break;
2699 
2700 		case ixgbe_mac_82599EB:
2701 		case ixgbe_mac_X540:
2702 		case ixgbe_mac_X550:
2703 		case ixgbe_mac_X550EM_x:
2704 			IXGBE_WRITE_REG(hw, IXGBE_TQSM(i >> 2), ring_mapping);
2705 			break;
2706 
2707 		default:
2708 			break;
2709 		}
2710 	}
2711 
2712 	/*
2713 	 * Enable CRC appending and TX padding (for short tx frames)
2714 	 */
2715 	reg_val = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2716 	reg_val |= IXGBE_HLREG0_TXCRCEN | IXGBE_HLREG0_TXPADEN;
2717 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, reg_val);
2718 
2719 	/*
2720 	 * enable DMA for 82599, X540 and X550 parts
2721 	 */
2722 	if (hw->mac.type == ixgbe_mac_82599EB ||
2723 	    hw->mac.type == ixgbe_mac_X540 ||
2724 	    hw->mac.type == ixgbe_mac_X550 ||
2725 	    hw->mac.type == ixgbe_mac_X550EM_x) {
2726 		/* DMATXCTL.TE must be set after all Tx config is complete */
2727 		reg_val = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2728 		reg_val |= IXGBE_DMATXCTL_TE;
2729 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, reg_val);
2730 
2731 		/* Disable arbiter to set MTQC */
2732 		reg_val = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2733 		reg_val |= IXGBE_RTTDCS_ARBDIS;
2734 		IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg_val);
2735 		IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2736 		reg_val &= ~IXGBE_RTTDCS_ARBDIS;
2737 		IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg_val);
2738 	}
2739 
2740 	/*
2741 	 * Enabling tx queues ..
2742 	 * For 82599 must be done after DMATXCTL.TE is set
2743 	 */
2744 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2745 		tx_ring = &ixgbe->tx_rings[i];
2746 		reg_val = IXGBE_READ_REG(hw, IXGBE_TXDCTL(tx_ring->index));
2747 		reg_val |= IXGBE_TXDCTL_ENABLE;
2748 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(tx_ring->index), reg_val);
2749 	}
2750 }
2751 
2752 /*
2753  * ixgbe_setup_rss - Setup receive-side scaling feature.
2754  */
2755 static void
2756 ixgbe_setup_rss(ixgbe_t *ixgbe)
2757 {
2758 	struct ixgbe_hw *hw = &ixgbe->hw;
2759 	uint32_t mrqc;
2760 
2761 	/*
2762 	 * Initialize RETA/ERETA table
2763 	 */
2764 	ixgbe_setup_rss_table(ixgbe);
2765 
2766 	/*
2767 	 * Enable RSS & perform hash on these packet types
2768 	 */
2769 	mrqc = IXGBE_MRQC_RSSEN |
2770 	    IXGBE_MRQC_RSS_FIELD_IPV4 |
2771 	    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2772 	    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2773 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2774 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2775 	    IXGBE_MRQC_RSS_FIELD_IPV6 |
2776 	    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2777 	    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2778 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2779 	IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2780 }
2781 
2782 /*
2783  * ixgbe_setup_vmdq - Setup MAC classification feature
2784  */
2785 static void
2786 ixgbe_setup_vmdq(ixgbe_t *ixgbe)
2787 {
2788 	struct ixgbe_hw *hw = &ixgbe->hw;
2789 	uint32_t vmdctl, i, vtctl;
2790 
2791 	/*
2792 	 * Setup the VMDq Control register, enable VMDq based on
2793 	 * packet destination MAC address:
2794 	 */
2795 	switch (hw->mac.type) {
2796 	case ixgbe_mac_82598EB:
2797 		/*
2798 		 * VMDq Enable = 1;
2799 		 * VMDq Filter = 0; MAC filtering
2800 		 * Default VMDq output index = 0;
2801 		 */
2802 		vmdctl = IXGBE_VMD_CTL_VMDQ_EN;
2803 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
2804 		break;
2805 
2806 	case ixgbe_mac_82599EB:
2807 	case ixgbe_mac_X540:
2808 	case ixgbe_mac_X550:
2809 	case ixgbe_mac_X550EM_x:
2810 		/*
2811 		 * Enable VMDq-only.
2812 		 */
2813 		vmdctl = IXGBE_MRQC_VMDQEN;
2814 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, vmdctl);
2815 
2816 		for (i = 0; i < hw->mac.num_rar_entries; i++) {
2817 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(i), 0);
2818 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(i), 0);
2819 		}
2820 
2821 		/*
2822 		 * Enable Virtualization and Replication.
2823 		 */
2824 		vtctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2825 		IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vtctl);
2826 
2827 		/*
2828 		 * Enable receiving packets to all VFs
2829 		 */
2830 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), IXGBE_VFRE_ENABLE_ALL);
2831 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), IXGBE_VFRE_ENABLE_ALL);
2832 		break;
2833 
2834 	default:
2835 		break;
2836 	}
2837 }
2838 
2839 /*
2840  * ixgbe_setup_vmdq_rss - Setup both vmdq feature and rss feature.
2841  */
2842 static void
2843 ixgbe_setup_vmdq_rss(ixgbe_t *ixgbe)
2844 {
2845 	struct ixgbe_hw *hw = &ixgbe->hw;
2846 	uint32_t i, mrqc;
2847 	uint32_t vtctl, vmdctl;
2848 
2849 	/*
2850 	 * Initialize RETA/ERETA table
2851 	 */
2852 	ixgbe_setup_rss_table(ixgbe);
2853 
2854 	/*
2855 	 * Enable and setup RSS and VMDq
2856 	 */
2857 	switch (hw->mac.type) {
2858 	case ixgbe_mac_82598EB:
2859 		/*
2860 		 * Enable RSS & Setup RSS Hash functions
2861 		 */
2862 		mrqc = IXGBE_MRQC_RSSEN |
2863 		    IXGBE_MRQC_RSS_FIELD_IPV4 |
2864 		    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2865 		    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2866 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2867 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2868 		    IXGBE_MRQC_RSS_FIELD_IPV6 |
2869 		    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2870 		    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2871 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2872 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2873 
2874 		/*
2875 		 * Enable and Setup VMDq
2876 		 * VMDq Filter = 0; MAC filtering
2877 		 * Default VMDq output index = 0;
2878 		 */
2879 		vmdctl = IXGBE_VMD_CTL_VMDQ_EN;
2880 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
2881 		break;
2882 
2883 	case ixgbe_mac_82599EB:
2884 	case ixgbe_mac_X540:
2885 	case ixgbe_mac_X550:
2886 	case ixgbe_mac_X550EM_x:
2887 		/*
2888 		 * Enable RSS & Setup RSS Hash functions
2889 		 */
2890 		mrqc = IXGBE_MRQC_RSS_FIELD_IPV4 |
2891 		    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2892 		    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2893 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2894 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2895 		    IXGBE_MRQC_RSS_FIELD_IPV6 |
2896 		    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2897 		    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2898 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2899 
2900 		/*
2901 		 * Enable VMDq+RSS.
2902 		 */
2903 		if (ixgbe->num_rx_groups > 32)  {
2904 			mrqc = mrqc | IXGBE_MRQC_VMDQRSS64EN;
2905 		} else {
2906 			mrqc = mrqc | IXGBE_MRQC_VMDQRSS32EN;
2907 		}
2908 
2909 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2910 
2911 		for (i = 0; i < hw->mac.num_rar_entries; i++) {
2912 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(i), 0);
2913 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(i), 0);
2914 		}
2915 		break;
2916 
2917 	default:
2918 		break;
2919 
2920 	}
2921 
2922 	if (hw->mac.type == ixgbe_mac_82599EB ||
2923 	    hw->mac.type == ixgbe_mac_X540 ||
2924 	    hw->mac.type == ixgbe_mac_X550 ||
2925 	    hw->mac.type == ixgbe_mac_X550EM_x) {
2926 		/*
2927 		 * Enable Virtualization and Replication.
2928 		 */
2929 		vtctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2930 		IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vtctl);
2931 
2932 		/*
2933 		 * Enable receiving packets to all VFs
2934 		 */
2935 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), IXGBE_VFRE_ENABLE_ALL);
2936 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), IXGBE_VFRE_ENABLE_ALL);
2937 	}
2938 }
2939 
2940 /*
2941  * ixgbe_setup_rss_table - Setup RSS table
2942  */
2943 static void
2944 ixgbe_setup_rss_table(ixgbe_t *ixgbe)
2945 {
2946 	struct ixgbe_hw *hw = &ixgbe->hw;
2947 	uint32_t i, j;
2948 	uint32_t random;
2949 	uint32_t reta;
2950 	uint32_t ring_per_group;
2951 	uint32_t ring;
2952 	uint32_t table_size;
2953 	uint32_t index_mult;
2954 	uint32_t rxcsum;
2955 
2956 	/*
2957 	 * Set multiplier for RETA setup and table size based on MAC type.
2958 	 * RETA table sizes vary by model:
2959 	 *
2960 	 * 82598, 82599, X540: 128 table entries.
2961 	 * X550: 512 table entries.
2962 	 */
2963 	index_mult = 0x1;
2964 	table_size = 128;
2965 	switch (ixgbe->hw.mac.type) {
2966 	case ixgbe_mac_82598EB:
2967 		index_mult = 0x11;
2968 		break;
2969 	case ixgbe_mac_X550:
2970 	case ixgbe_mac_X550EM_x:
2971 		table_size = 512;
2972 		break;
2973 	default:
2974 		break;
2975 	}
2976 
2977 	/*
2978 	 * Fill out RSS redirection table. The configuation of the indices is
2979 	 * hardware-dependent.
2980 	 *
2981 	 *  82598: 8 bits wide containing two 4 bit RSS indices
2982 	 *  82599, X540: 8 bits wide containing one 4 bit RSS index
2983 	 *  X550: 8 bits wide containing one 6 bit RSS index
2984 	 */
2985 	reta = 0;
2986 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
2987 
2988 	for (i = 0, j = 0; i < table_size; i++, j++) {
2989 		if (j == ring_per_group) j = 0;
2990 
2991 		/*
2992 		 * The low 8 bits are for hash value (n+0);
2993 		 * The next 8 bits are for hash value (n+1), etc.
2994 		 */
2995 		ring = (j * index_mult);
2996 		reta = reta >> 8;
2997 		reta = reta | (((uint32_t)ring) << 24);
2998 
2999 		if ((i & 3) == 3)
3000 			/*
3001 			 * The first 128 table entries are programmed into the
3002 			 * RETA register, with any beyond that (eg; on X550)
3003 			 * into ERETA.
3004 			 */
3005 			if (i < 128)
3006 				IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3007 			else
3008 				IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3009 				    reta);
3010 			reta = 0;
3011 	}
3012 
3013 	/*
3014 	 * Fill out hash function seeds with a random constant
3015 	 */
3016 	for (i = 0; i < 10; i++) {
3017 		(void) random_get_pseudo_bytes((uint8_t *)&random,
3018 		    sizeof (uint32_t));
3019 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), random);
3020 	}
3021 
3022 	/*
3023 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
3024 	 * It is an adapter hardware limitation that Packet Checksum is
3025 	 * mutually exclusive with RSS.
3026 	 */
3027 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3028 	rxcsum |= IXGBE_RXCSUM_PCSD;
3029 	rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
3030 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3031 }
3032 
3033 /*
3034  * ixgbe_init_unicst - Initialize the unicast addresses.
3035  */
3036 static void
3037 ixgbe_init_unicst(ixgbe_t *ixgbe)
3038 {
3039 	struct ixgbe_hw *hw = &ixgbe->hw;
3040 	uint8_t *mac_addr;
3041 	int slot;
3042 	/*
3043 	 * Here we should consider two situations:
3044 	 *
3045 	 * 1. Chipset is initialized at the first time,
3046 	 *    Clear all the multiple unicast addresses.
3047 	 *
3048 	 * 2. Chipset is reset
3049 	 *    Recover the multiple unicast addresses from the
3050 	 *    software data structure to the RAR registers.
3051 	 */
3052 	if (!ixgbe->unicst_init) {
3053 		/*
3054 		 * Initialize the multiple unicast addresses
3055 		 */
3056 		ixgbe->unicst_total = hw->mac.num_rar_entries;
3057 		ixgbe->unicst_avail = ixgbe->unicst_total;
3058 		for (slot = 0; slot < ixgbe->unicst_total; slot++) {
3059 			mac_addr = ixgbe->unicst_addr[slot].mac.addr;
3060 			bzero(mac_addr, ETHERADDRL);
3061 			(void) ixgbe_set_rar(hw, slot, mac_addr, NULL, NULL);
3062 			ixgbe->unicst_addr[slot].mac.set = 0;
3063 		}
3064 		ixgbe->unicst_init = B_TRUE;
3065 	} else {
3066 		/* Re-configure the RAR registers */
3067 		for (slot = 0; slot < ixgbe->unicst_total; slot++) {
3068 			mac_addr = ixgbe->unicst_addr[slot].mac.addr;
3069 			if (ixgbe->unicst_addr[slot].mac.set == 1) {
3070 				(void) ixgbe_set_rar(hw, slot, mac_addr,
3071 				    ixgbe->unicst_addr[slot].mac.group_index,
3072 				    IXGBE_RAH_AV);
3073 			} else {
3074 				bzero(mac_addr, ETHERADDRL);
3075 				(void) ixgbe_set_rar(hw, slot, mac_addr,
3076 				    NULL, NULL);
3077 			}
3078 		}
3079 	}
3080 }
3081 
3082 /*
3083  * ixgbe_unicst_find - Find the slot for the specified unicast address
3084  */
3085 int
3086 ixgbe_unicst_find(ixgbe_t *ixgbe, const uint8_t *mac_addr)
3087 {
3088 	int slot;
3089 
3090 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3091 
3092 	for (slot = 0; slot < ixgbe->unicst_total; slot++) {
3093 		if (bcmp(ixgbe->unicst_addr[slot].mac.addr,
3094 		    mac_addr, ETHERADDRL) == 0)
3095 			return (slot);
3096 	}
3097 
3098 	return (-1);
3099 }
3100 
3101 /*
3102  * ixgbe_multicst_add - Add a multicst address.
3103  */
3104 int
3105 ixgbe_multicst_add(ixgbe_t *ixgbe, const uint8_t *multiaddr)
3106 {
3107 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3108 
3109 	if ((multiaddr[0] & 01) == 0) {
3110 		return (EINVAL);
3111 	}
3112 
3113 	if (ixgbe->mcast_count >= MAX_NUM_MULTICAST_ADDRESSES) {
3114 		return (ENOENT);
3115 	}
3116 
3117 	bcopy(multiaddr,
3118 	    &ixgbe->mcast_table[ixgbe->mcast_count], ETHERADDRL);
3119 	ixgbe->mcast_count++;
3120 
3121 	/*
3122 	 * Update the multicast table in the hardware
3123 	 */
3124 	ixgbe_setup_multicst(ixgbe);
3125 
3126 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
3127 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3128 		return (EIO);
3129 	}
3130 
3131 	return (0);
3132 }
3133 
3134 /*
3135  * ixgbe_multicst_remove - Remove a multicst address.
3136  */
3137 int
3138 ixgbe_multicst_remove(ixgbe_t *ixgbe, const uint8_t *multiaddr)
3139 {
3140 	int i;
3141 
3142 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3143 
3144 	for (i = 0; i < ixgbe->mcast_count; i++) {
3145 		if (bcmp(multiaddr, &ixgbe->mcast_table[i],
3146 		    ETHERADDRL) == 0) {
3147 			for (i++; i < ixgbe->mcast_count; i++) {
3148 				ixgbe->mcast_table[i - 1] =
3149 				    ixgbe->mcast_table[i];
3150 			}
3151 			ixgbe->mcast_count--;
3152 			break;
3153 		}
3154 	}
3155 
3156 	/*
3157 	 * Update the multicast table in the hardware
3158 	 */
3159 	ixgbe_setup_multicst(ixgbe);
3160 
3161 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
3162 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3163 		return (EIO);
3164 	}
3165 
3166 	return (0);
3167 }
3168 
3169 /*
3170  * ixgbe_setup_multicast - Setup multicast data structures.
3171  *
3172  * This routine initializes all of the multicast related structures
3173  * and save them in the hardware registers.
3174  */
3175 static void
3176 ixgbe_setup_multicst(ixgbe_t *ixgbe)
3177 {
3178 	uint8_t *mc_addr_list;
3179 	uint32_t mc_addr_count;
3180 	struct ixgbe_hw *hw = &ixgbe->hw;
3181 
3182 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3183 
3184 	ASSERT(ixgbe->mcast_count <= MAX_NUM_MULTICAST_ADDRESSES);
3185 
3186 	mc_addr_list = (uint8_t *)ixgbe->mcast_table;
3187 	mc_addr_count = ixgbe->mcast_count;
3188 
3189 	/*
3190 	 * Update the multicast addresses to the MTA registers
3191 	 */
3192 	(void) ixgbe_update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
3193 	    ixgbe_mc_table_itr, TRUE);
3194 }
3195 
3196 /*
3197  * ixgbe_setup_vmdq_rss_conf - Configure vmdq and rss (number and mode).
3198  *
3199  * Configure the rx classification mode (vmdq & rss) and vmdq & rss numbers.
3200  * Different chipsets may have different allowed configuration of vmdq and rss.
3201  */
3202 static void
3203 ixgbe_setup_vmdq_rss_conf(ixgbe_t *ixgbe)
3204 {
3205 	struct ixgbe_hw *hw = &ixgbe->hw;
3206 	uint32_t ring_per_group;
3207 
3208 	switch (hw->mac.type) {
3209 	case ixgbe_mac_82598EB:
3210 		/*
3211 		 * 82598 supports the following combination:
3212 		 * vmdq no. x rss no.
3213 		 * [5..16]  x 1
3214 		 * [1..4]   x [1..16]
3215 		 * However 8 rss queue per pool (vmdq) is sufficient for
3216 		 * most cases.
3217 		 */
3218 		ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
3219 		if (ixgbe->num_rx_groups > 4) {
3220 			ixgbe->num_rx_rings = ixgbe->num_rx_groups;
3221 		} else {
3222 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3223 			    min(8, ring_per_group);
3224 		}
3225 
3226 		break;
3227 
3228 	case ixgbe_mac_82599EB:
3229 	case ixgbe_mac_X540:
3230 	case ixgbe_mac_X550:
3231 	case ixgbe_mac_X550EM_x:
3232 		/*
3233 		 * 82599 supports the following combination:
3234 		 * vmdq no. x rss no.
3235 		 * [33..64] x [1..2]
3236 		 * [2..32]  x [1..4]
3237 		 * 1 x [1..16]
3238 		 * However 8 rss queue per pool (vmdq) is sufficient for
3239 		 * most cases.
3240 		 *
3241 		 * For now, treat X540 and X550 like the 82599.
3242 		 */
3243 		ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
3244 		if (ixgbe->num_rx_groups == 1) {
3245 			ixgbe->num_rx_rings = min(8, ring_per_group);
3246 		} else if (ixgbe->num_rx_groups <= 32) {
3247 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3248 			    min(4, ring_per_group);
3249 		} else if (ixgbe->num_rx_groups <= 64) {
3250 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3251 			    min(2, ring_per_group);
3252 		}
3253 		break;
3254 
3255 	default:
3256 		break;
3257 	}
3258 
3259 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
3260 
3261 	if (ixgbe->num_rx_groups == 1 && ring_per_group == 1) {
3262 		ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
3263 	} else if (ixgbe->num_rx_groups != 1 && ring_per_group == 1) {
3264 		ixgbe->classify_mode = IXGBE_CLASSIFY_VMDQ;
3265 	} else if (ixgbe->num_rx_groups != 1 && ring_per_group != 1) {
3266 		ixgbe->classify_mode = IXGBE_CLASSIFY_VMDQ_RSS;
3267 	} else {
3268 		ixgbe->classify_mode = IXGBE_CLASSIFY_RSS;
3269 	}
3270 
3271 	IXGBE_DEBUGLOG_2(ixgbe, "rx group number:%d, rx ring number:%d",
3272 	    ixgbe->num_rx_groups, ixgbe->num_rx_rings);
3273 }
3274 
3275 /*
3276  * ixgbe_get_conf - Get driver configurations set in driver.conf.
3277  *
3278  * This routine gets user-configured values out of the configuration
3279  * file ixgbe.conf.
3280  *
3281  * For each configurable value, there is a minimum, a maximum, and a
3282  * default.
3283  * If user does not configure a value, use the default.
3284  * If user configures below the minimum, use the minumum.
3285  * If user configures above the maximum, use the maxumum.
3286  */
3287 static void
3288 ixgbe_get_conf(ixgbe_t *ixgbe)
3289 {
3290 	struct ixgbe_hw *hw = &ixgbe->hw;
3291 	uint32_t flow_control;
3292 
3293 	/*
3294 	 * ixgbe driver supports the following user configurations:
3295 	 *
3296 	 * Jumbo frame configuration:
3297 	 *    default_mtu
3298 	 *
3299 	 * Ethernet flow control configuration:
3300 	 *    flow_control
3301 	 *
3302 	 * Multiple rings configurations:
3303 	 *    tx_queue_number
3304 	 *    tx_ring_size
3305 	 *    rx_queue_number
3306 	 *    rx_ring_size
3307 	 *
3308 	 * Call ixgbe_get_prop() to get the value for a specific
3309 	 * configuration parameter.
3310 	 */
3311 
3312 	/*
3313 	 * Jumbo frame configuration - max_frame_size controls host buffer
3314 	 * allocation, so includes MTU, ethernet header, vlan tag and
3315 	 * frame check sequence.
3316 	 */
3317 	ixgbe->default_mtu = ixgbe_get_prop(ixgbe, PROP_DEFAULT_MTU,
3318 	    MIN_MTU, ixgbe->capab->max_mtu, DEFAULT_MTU);
3319 
3320 	ixgbe->max_frame_size = ixgbe->default_mtu +
3321 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
3322 
3323 	/*
3324 	 * Ethernet flow control configuration
3325 	 */
3326 	flow_control = ixgbe_get_prop(ixgbe, PROP_FLOW_CONTROL,
3327 	    ixgbe_fc_none, 3, ixgbe_fc_none);
3328 	if (flow_control == 3)
3329 		flow_control = ixgbe_fc_default;
3330 
3331 	/*
3332 	 * fc.requested mode is what the user requests.  After autoneg,
3333 	 * fc.current_mode will be the flow_control mode that was negotiated.
3334 	 */
3335 	hw->fc.requested_mode = flow_control;
3336 
3337 	/*
3338 	 * Multiple rings configurations
3339 	 */
3340 	ixgbe->num_tx_rings = ixgbe_get_prop(ixgbe, PROP_TX_QUEUE_NUM,
3341 	    ixgbe->capab->min_tx_que_num,
3342 	    ixgbe->capab->max_tx_que_num,
3343 	    ixgbe->capab->def_tx_que_num);
3344 	ixgbe->tx_ring_size = ixgbe_get_prop(ixgbe, PROP_TX_RING_SIZE,
3345 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
3346 
3347 	ixgbe->num_rx_rings = ixgbe_get_prop(ixgbe, PROP_RX_QUEUE_NUM,
3348 	    ixgbe->capab->min_rx_que_num,
3349 	    ixgbe->capab->max_rx_que_num,
3350 	    ixgbe->capab->def_rx_que_num);
3351 	ixgbe->rx_ring_size = ixgbe_get_prop(ixgbe, PROP_RX_RING_SIZE,
3352 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
3353 
3354 	/*
3355 	 * Multiple groups configuration
3356 	 */
3357 	ixgbe->num_rx_groups = ixgbe_get_prop(ixgbe, PROP_RX_GROUP_NUM,
3358 	    ixgbe->capab->min_rx_grp_num, ixgbe->capab->max_rx_grp_num,
3359 	    ixgbe->capab->def_rx_grp_num);
3360 
3361 	ixgbe->mr_enable = ixgbe_get_prop(ixgbe, PROP_MR_ENABLE,
3362 	    0, 1, DEFAULT_MR_ENABLE);
3363 
3364 	if (ixgbe->mr_enable == B_FALSE) {
3365 		ixgbe->num_tx_rings = 1;
3366 		ixgbe->num_rx_rings = 1;
3367 		ixgbe->num_rx_groups = 1;
3368 		ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
3369 	} else {
3370 		ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3371 		    max(ixgbe->num_rx_rings / ixgbe->num_rx_groups, 1);
3372 		/*
3373 		 * The combination of num_rx_rings and num_rx_groups
3374 		 * may be not supported by h/w. We need to adjust
3375 		 * them to appropriate values.
3376 		 */
3377 		ixgbe_setup_vmdq_rss_conf(ixgbe);
3378 	}
3379 
3380 	/*
3381 	 * Tunable used to force an interrupt type. The only use is
3382 	 * for testing of the lesser interrupt types.
3383 	 * 0 = don't force interrupt type
3384 	 * 1 = force interrupt type MSI-X
3385 	 * 2 = force interrupt type MSI
3386 	 * 3 = force interrupt type Legacy
3387 	 */
3388 	ixgbe->intr_force = ixgbe_get_prop(ixgbe, PROP_INTR_FORCE,
3389 	    IXGBE_INTR_NONE, IXGBE_INTR_LEGACY, IXGBE_INTR_NONE);
3390 
3391 	ixgbe->tx_hcksum_enable = ixgbe_get_prop(ixgbe, PROP_TX_HCKSUM_ENABLE,
3392 	    0, 1, DEFAULT_TX_HCKSUM_ENABLE);
3393 	ixgbe->rx_hcksum_enable = ixgbe_get_prop(ixgbe, PROP_RX_HCKSUM_ENABLE,
3394 	    0, 1, DEFAULT_RX_HCKSUM_ENABLE);
3395 	ixgbe->lso_enable = ixgbe_get_prop(ixgbe, PROP_LSO_ENABLE,
3396 	    0, 1, DEFAULT_LSO_ENABLE);
3397 	ixgbe->lro_enable = ixgbe_get_prop(ixgbe, PROP_LRO_ENABLE,
3398 	    0, 1, DEFAULT_LRO_ENABLE);
3399 	ixgbe->tx_head_wb_enable = ixgbe_get_prop(ixgbe, PROP_TX_HEAD_WB_ENABLE,
3400 	    0, 1, DEFAULT_TX_HEAD_WB_ENABLE);
3401 	ixgbe->relax_order_enable = ixgbe_get_prop(ixgbe,
3402 	    PROP_RELAX_ORDER_ENABLE, 0, 1, DEFAULT_RELAX_ORDER_ENABLE);
3403 
3404 	/* Head Write Back not recommended for 82599, X540 and X550 */
3405 	if (hw->mac.type == ixgbe_mac_82599EB ||
3406 	    hw->mac.type == ixgbe_mac_X540 ||
3407 	    hw->mac.type == ixgbe_mac_X550 ||
3408 	    hw->mac.type == ixgbe_mac_X550EM_x) {
3409 		ixgbe->tx_head_wb_enable = B_FALSE;
3410 	}
3411 
3412 	/*
3413 	 * ixgbe LSO needs the tx h/w checksum support.
3414 	 * LSO will be disabled if tx h/w checksum is not
3415 	 * enabled.
3416 	 */
3417 	if (ixgbe->tx_hcksum_enable == B_FALSE) {
3418 		ixgbe->lso_enable = B_FALSE;
3419 	}
3420 
3421 	/*
3422 	 * ixgbe LRO needs the rx h/w checksum support.
3423 	 * LRO will be disabled if rx h/w checksum is not
3424 	 * enabled.
3425 	 */
3426 	if (ixgbe->rx_hcksum_enable == B_FALSE) {
3427 		ixgbe->lro_enable = B_FALSE;
3428 	}
3429 
3430 	/*
3431 	 * ixgbe LRO only supported by 82599, X540 and X550
3432 	 */
3433 	if (hw->mac.type == ixgbe_mac_82598EB) {
3434 		ixgbe->lro_enable = B_FALSE;
3435 	}
3436 	ixgbe->tx_copy_thresh = ixgbe_get_prop(ixgbe, PROP_TX_COPY_THRESHOLD,
3437 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
3438 	    DEFAULT_TX_COPY_THRESHOLD);
3439 	ixgbe->tx_recycle_thresh = ixgbe_get_prop(ixgbe,
3440 	    PROP_TX_RECYCLE_THRESHOLD, MIN_TX_RECYCLE_THRESHOLD,
3441 	    MAX_TX_RECYCLE_THRESHOLD, DEFAULT_TX_RECYCLE_THRESHOLD);
3442 	ixgbe->tx_overload_thresh = ixgbe_get_prop(ixgbe,
3443 	    PROP_TX_OVERLOAD_THRESHOLD, MIN_TX_OVERLOAD_THRESHOLD,
3444 	    MAX_TX_OVERLOAD_THRESHOLD, DEFAULT_TX_OVERLOAD_THRESHOLD);
3445 	ixgbe->tx_resched_thresh = ixgbe_get_prop(ixgbe,
3446 	    PROP_TX_RESCHED_THRESHOLD, MIN_TX_RESCHED_THRESHOLD,
3447 	    MAX_TX_RESCHED_THRESHOLD, DEFAULT_TX_RESCHED_THRESHOLD);
3448 
3449 	ixgbe->rx_copy_thresh = ixgbe_get_prop(ixgbe, PROP_RX_COPY_THRESHOLD,
3450 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
3451 	    DEFAULT_RX_COPY_THRESHOLD);
3452 	ixgbe->rx_limit_per_intr = ixgbe_get_prop(ixgbe, PROP_RX_LIMIT_PER_INTR,
3453 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
3454 	    DEFAULT_RX_LIMIT_PER_INTR);
3455 
3456 	ixgbe->intr_throttling[0] = ixgbe_get_prop(ixgbe, PROP_INTR_THROTTLING,
3457 	    ixgbe->capab->min_intr_throttle,
3458 	    ixgbe->capab->max_intr_throttle,
3459 	    ixgbe->capab->def_intr_throttle);
3460 	/*
3461 	 * 82599, X540 and X550 require the interrupt throttling rate is
3462 	 * a multiple of 8. This is enforced by the register definiton.
3463 	 */
3464 	if (hw->mac.type == ixgbe_mac_82599EB ||
3465 	    hw->mac.type == ixgbe_mac_X540 ||
3466 	    hw->mac.type == ixgbe_mac_X550 ||
3467 	    hw->mac.type == ixgbe_mac_X550EM_x)
3468 		ixgbe->intr_throttling[0] = ixgbe->intr_throttling[0] & 0xFF8;
3469 
3470 	hw->allow_unsupported_sfp = ixgbe_get_prop(ixgbe,
3471 	    PROP_ALLOW_UNSUPPORTED_SFP, 0, 1, DEFAULT_ALLOW_UNSUPPORTED_SFP);
3472 }
3473 
3474 static void
3475 ixgbe_init_params(ixgbe_t *ixgbe)
3476 {
3477 	struct ixgbe_hw *hw = &ixgbe->hw;
3478 	ixgbe_link_speed speeds_supported = 0;
3479 	boolean_t negotiate;
3480 
3481 	/*
3482 	 * Get a list of speeds the adapter supports. If the hw struct hasn't
3483 	 * been populated with this information yet, retrieve it from the
3484 	 * adapter and save it to our own variable.
3485 	 *
3486 	 * On certain adapters, such as ones which use SFPs, the contents of
3487 	 * hw->phy.speeds_supported (and hw->phy.autoneg_advertised) are not
3488 	 * updated, so we must rely on calling ixgbe_get_link_capabilities()
3489 	 * in order to ascertain the speeds which we are capable of supporting,
3490 	 * and in the case of SFP-equipped adapters, which speed we are
3491 	 * advertising. If ixgbe_get_link_capabilities() fails for some reason,
3492 	 * we'll go with a default list of speeds as a last resort.
3493 	 */
3494 	speeds_supported = hw->phy.speeds_supported;
3495 
3496 	if (speeds_supported == 0) {
3497 		if (ixgbe_get_link_capabilities(hw, &speeds_supported,
3498 		    &negotiate) != IXGBE_SUCCESS) {
3499 			if (hw->mac.type == ixgbe_mac_82598EB) {
3500 				speeds_supported =
3501 				    IXGBE_LINK_SPEED_82598_AUTONEG;
3502 			} else {
3503 				speeds_supported =
3504 				    IXGBE_LINK_SPEED_82599_AUTONEG;
3505 			}
3506 		}
3507 	}
3508 	ixgbe->speeds_supported = speeds_supported;
3509 
3510 	/*
3511 	 * By default, all supported speeds are enabled and advertised.
3512 	 */
3513 	if (speeds_supported & IXGBE_LINK_SPEED_10GB_FULL) {
3514 		ixgbe->param_en_10000fdx_cap = 1;
3515 		ixgbe->param_adv_10000fdx_cap = 1;
3516 	} else {
3517 		ixgbe->param_en_10000fdx_cap = 0;
3518 		ixgbe->param_adv_10000fdx_cap = 0;
3519 	}
3520 
3521 	if (speeds_supported & IXGBE_LINK_SPEED_5GB_FULL) {
3522 		ixgbe->param_en_5000fdx_cap = 1;
3523 		ixgbe->param_adv_5000fdx_cap = 1;
3524 	} else {
3525 		ixgbe->param_en_5000fdx_cap = 0;
3526 		ixgbe->param_adv_5000fdx_cap = 0;
3527 	}
3528 
3529 	if (speeds_supported & IXGBE_LINK_SPEED_2_5GB_FULL) {
3530 		ixgbe->param_en_2500fdx_cap = 1;
3531 		ixgbe->param_adv_2500fdx_cap = 1;
3532 	} else {
3533 		ixgbe->param_en_2500fdx_cap = 0;
3534 		ixgbe->param_adv_2500fdx_cap = 0;
3535 	}
3536 
3537 	if (speeds_supported & IXGBE_LINK_SPEED_1GB_FULL) {
3538 		ixgbe->param_en_1000fdx_cap = 1;
3539 		ixgbe->param_adv_1000fdx_cap = 1;
3540 	} else {
3541 		ixgbe->param_en_1000fdx_cap = 0;
3542 		ixgbe->param_adv_1000fdx_cap = 0;
3543 	}
3544 
3545 	if (speeds_supported & IXGBE_LINK_SPEED_100_FULL) {
3546 		ixgbe->param_en_100fdx_cap = 1;
3547 		ixgbe->param_adv_100fdx_cap = 1;
3548 	} else {
3549 		ixgbe->param_en_100fdx_cap = 0;
3550 		ixgbe->param_adv_100fdx_cap = 0;
3551 	}
3552 
3553 	ixgbe->param_pause_cap = 1;
3554 	ixgbe->param_asym_pause_cap = 1;
3555 	ixgbe->param_rem_fault = 0;
3556 
3557 	ixgbe->param_adv_autoneg_cap = 1;
3558 	ixgbe->param_adv_pause_cap = 1;
3559 	ixgbe->param_adv_asym_pause_cap = 1;
3560 	ixgbe->param_adv_rem_fault = 0;
3561 
3562 	ixgbe->param_lp_10000fdx_cap = 0;
3563 	ixgbe->param_lp_5000fdx_cap = 0;
3564 	ixgbe->param_lp_2500fdx_cap = 0;
3565 	ixgbe->param_lp_1000fdx_cap = 0;
3566 	ixgbe->param_lp_100fdx_cap = 0;
3567 	ixgbe->param_lp_autoneg_cap = 0;
3568 	ixgbe->param_lp_pause_cap = 0;
3569 	ixgbe->param_lp_asym_pause_cap = 0;
3570 	ixgbe->param_lp_rem_fault = 0;
3571 }
3572 
3573 /*
3574  * ixgbe_get_prop - Get a property value out of the configuration file
3575  * ixgbe.conf.
3576  *
3577  * Caller provides the name of the property, a default value, a minimum
3578  * value, and a maximum value.
3579  *
3580  * Return configured value of the property, with default, minimum and
3581  * maximum properly applied.
3582  */
3583 static int
3584 ixgbe_get_prop(ixgbe_t *ixgbe,
3585     char *propname,	/* name of the property */
3586     int minval,		/* minimum acceptable value */
3587     int maxval,		/* maximim acceptable value */
3588     int defval)		/* default value */
3589 {
3590 	int value;
3591 
3592 	/*
3593 	 * Call ddi_prop_get_int() to read the conf settings
3594 	 */
3595 	value = ddi_prop_get_int(DDI_DEV_T_ANY, ixgbe->dip,
3596 	    DDI_PROP_DONTPASS, propname, defval);
3597 	if (value > maxval)
3598 		value = maxval;
3599 
3600 	if (value < minval)
3601 		value = minval;
3602 
3603 	return (value);
3604 }
3605 
3606 /*
3607  * ixgbe_driver_setup_link - Using the link properties to setup the link.
3608  */
3609 int
3610 ixgbe_driver_setup_link(ixgbe_t *ixgbe, boolean_t setup_hw)
3611 {
3612 	struct ixgbe_hw *hw = &ixgbe->hw;
3613 	ixgbe_link_speed advertised = 0;
3614 
3615 	/*
3616 	 * Assemble a list of enabled speeds to auto-negotiate with.
3617 	 */
3618 	if (ixgbe->param_en_10000fdx_cap == 1)
3619 		advertised |= IXGBE_LINK_SPEED_10GB_FULL;
3620 
3621 	if (ixgbe->param_en_5000fdx_cap == 1)
3622 		advertised |= IXGBE_LINK_SPEED_5GB_FULL;
3623 
3624 	if (ixgbe->param_en_2500fdx_cap == 1)
3625 		advertised |= IXGBE_LINK_SPEED_2_5GB_FULL;
3626 
3627 	if (ixgbe->param_en_1000fdx_cap == 1)
3628 		advertised |= IXGBE_LINK_SPEED_1GB_FULL;
3629 
3630 	if (ixgbe->param_en_100fdx_cap == 1)
3631 		advertised |= IXGBE_LINK_SPEED_100_FULL;
3632 
3633 	/*
3634 	 * As a last resort, autoneg with a default list of speeds.
3635 	 */
3636 	if (ixgbe->param_adv_autoneg_cap == 1 && advertised == 0) {
3637 		ixgbe_notice(ixgbe, "Invalid link settings. Setting link "
3638 		    "to autonegotiate with full capabilities.");
3639 
3640 		if (hw->mac.type == ixgbe_mac_82598EB)
3641 			advertised = IXGBE_LINK_SPEED_82598_AUTONEG;
3642 		else
3643 			advertised = IXGBE_LINK_SPEED_82599_AUTONEG;
3644 	}
3645 
3646 	if (setup_hw) {
3647 		if (ixgbe_setup_link(&ixgbe->hw, advertised,
3648 		    ixgbe->param_adv_autoneg_cap) != IXGBE_SUCCESS) {
3649 			ixgbe_notice(ixgbe, "Setup link failed on this "
3650 			    "device.");
3651 			return (IXGBE_FAILURE);
3652 		}
3653 	}
3654 
3655 	return (IXGBE_SUCCESS);
3656 }
3657 
3658 /*
3659  * ixgbe_driver_link_check - Link status processing.
3660  *
3661  * This function can be called in both kernel context and interrupt context
3662  */
3663 static void
3664 ixgbe_driver_link_check(ixgbe_t *ixgbe)
3665 {
3666 	struct ixgbe_hw *hw = &ixgbe->hw;
3667 	ixgbe_link_speed speed = IXGBE_LINK_SPEED_UNKNOWN;
3668 	boolean_t link_up = B_FALSE;
3669 	boolean_t link_changed = B_FALSE;
3670 
3671 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3672 
3673 	(void) ixgbe_check_link(hw, &speed, &link_up, B_FALSE);
3674 	if (link_up) {
3675 		ixgbe->link_check_complete = B_TRUE;
3676 
3677 		/* Link is up, enable flow control settings */
3678 		(void) ixgbe_fc_enable(hw);
3679 
3680 		/*
3681 		 * The Link is up, check whether it was marked as down earlier
3682 		 */
3683 		if (ixgbe->link_state != LINK_STATE_UP) {
3684 			switch (speed) {
3685 			case IXGBE_LINK_SPEED_10GB_FULL:
3686 				ixgbe->link_speed = SPEED_10GB;
3687 				break;
3688 			case IXGBE_LINK_SPEED_5GB_FULL:
3689 				ixgbe->link_speed = SPEED_5GB;
3690 				break;
3691 			case IXGBE_LINK_SPEED_2_5GB_FULL:
3692 				ixgbe->link_speed = SPEED_2_5GB;
3693 				break;
3694 			case IXGBE_LINK_SPEED_1GB_FULL:
3695 				ixgbe->link_speed = SPEED_1GB;
3696 				break;
3697 			case IXGBE_LINK_SPEED_100_FULL:
3698 				ixgbe->link_speed = SPEED_100;
3699 			}
3700 			ixgbe->link_duplex = LINK_DUPLEX_FULL;
3701 			ixgbe->link_state = LINK_STATE_UP;
3702 			link_changed = B_TRUE;
3703 		}
3704 	} else {
3705 		if (ixgbe->link_check_complete == B_TRUE ||
3706 		    (ixgbe->link_check_complete == B_FALSE &&
3707 		    gethrtime() >= ixgbe->link_check_hrtime)) {
3708 			/*
3709 			 * The link is really down
3710 			 */
3711 			ixgbe->link_check_complete = B_TRUE;
3712 
3713 			if (ixgbe->link_state != LINK_STATE_DOWN) {
3714 				ixgbe->link_speed = 0;
3715 				ixgbe->link_duplex = LINK_DUPLEX_UNKNOWN;
3716 				ixgbe->link_state = LINK_STATE_DOWN;
3717 				link_changed = B_TRUE;
3718 			}
3719 		}
3720 	}
3721 
3722 	/*
3723 	 * If we are in an interrupt context, need to re-enable the
3724 	 * interrupt, which was automasked
3725 	 */
3726 	if (servicing_interrupt() != 0) {
3727 		ixgbe->eims |= IXGBE_EICR_LSC;
3728 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
3729 	}
3730 
3731 	if (link_changed) {
3732 		mac_link_update(ixgbe->mac_hdl, ixgbe->link_state);
3733 	}
3734 }
3735 
3736 /*
3737  * ixgbe_sfp_check - sfp module processing done in taskq only for 82599.
3738  */
3739 static void
3740 ixgbe_sfp_check(void *arg)
3741 {
3742 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3743 	uint32_t eicr = ixgbe->eicr;
3744 	struct ixgbe_hw *hw = &ixgbe->hw;
3745 
3746 	mutex_enter(&ixgbe->gen_lock);
3747 	if (eicr & IXGBE_EICR_GPI_SDP1_BY_MAC(hw)) {
3748 		/* clear the interrupt */
3749 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1_BY_MAC(hw));
3750 
3751 		/* if link up, do multispeed fiber setup */
3752 		(void) ixgbe_setup_link(hw, IXGBE_LINK_SPEED_82599_AUTONEG,
3753 		    B_TRUE);
3754 		ixgbe_driver_link_check(ixgbe);
3755 		ixgbe_get_hw_state(ixgbe);
3756 	} else if (eicr & IXGBE_EICR_GPI_SDP2_BY_MAC(hw)) {
3757 		/* clear the interrupt */
3758 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2_BY_MAC(hw));
3759 
3760 		/* if link up, do sfp module setup */
3761 		(void) hw->mac.ops.setup_sfp(hw);
3762 
3763 		/* do multispeed fiber setup */
3764 		(void) ixgbe_setup_link(hw, IXGBE_LINK_SPEED_82599_AUTONEG,
3765 		    B_TRUE);
3766 		ixgbe_driver_link_check(ixgbe);
3767 		ixgbe_get_hw_state(ixgbe);
3768 	}
3769 	mutex_exit(&ixgbe->gen_lock);
3770 
3771 	/*
3772 	 * We need to fully re-check the link later.
3773 	 */
3774 	ixgbe->link_check_complete = B_FALSE;
3775 	ixgbe->link_check_hrtime = gethrtime() +
3776 	    (IXGBE_LINK_UP_TIME * 100000000ULL);
3777 }
3778 
3779 /*
3780  * ixgbe_overtemp_check - overtemp module processing done in taskq
3781  *
3782  * This routine will only be called on adapters with temperature sensor.
3783  * The indication of over-temperature can be either SDP0 interrupt or the link
3784  * status change interrupt.
3785  */
3786 static void
3787 ixgbe_overtemp_check(void *arg)
3788 {
3789 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3790 	struct ixgbe_hw *hw = &ixgbe->hw;
3791 	uint32_t eicr = ixgbe->eicr;
3792 	ixgbe_link_speed speed;
3793 	boolean_t link_up;
3794 
3795 	mutex_enter(&ixgbe->gen_lock);
3796 
3797 	/* make sure we know current state of link */
3798 	(void) ixgbe_check_link(hw, &speed, &link_up, B_FALSE);
3799 
3800 	/* check over-temp condition */
3801 	if (((eicr & IXGBE_EICR_GPI_SDP0_BY_MAC(hw)) && (!link_up)) ||
3802 	    (eicr & IXGBE_EICR_LSC)) {
3803 		if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP) {
3804 			atomic_or_32(&ixgbe->ixgbe_state, IXGBE_OVERTEMP);
3805 
3806 			/*
3807 			 * Disable the adapter interrupts
3808 			 */
3809 			ixgbe_disable_adapter_interrupts(ixgbe);
3810 
3811 			/*
3812 			 * Disable Rx/Tx units
3813 			 */
3814 			(void) ixgbe_stop_adapter(hw);
3815 
3816 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
3817 			ixgbe_error(ixgbe,
3818 			    "Problem: Network adapter has been stopped "
3819 			    "because it has overheated");
3820 			ixgbe_error(ixgbe,
3821 			    "Action: Restart the computer. "
3822 			    "If the problem persists, power off the system "
3823 			    "and replace the adapter");
3824 		}
3825 	}
3826 
3827 	/* write to clear the interrupt */
3828 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3829 
3830 	mutex_exit(&ixgbe->gen_lock);
3831 }
3832 
3833 /*
3834  * ixgbe_phy_check - taskq to process interrupts from an external PHY
3835  *
3836  * This routine will only be called on adapters with external PHYs
3837  * (such as X550) that may be trying to raise our attention to some event.
3838  * Currently, this is limited to claiming PHY overtemperature and link status
3839  * change (LSC) events, however this may expand to include other things in
3840  * future adapters.
3841  */
3842 static void
3843 ixgbe_phy_check(void *arg)
3844 {
3845 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3846 	struct ixgbe_hw *hw = &ixgbe->hw;
3847 	int rv;
3848 
3849 	mutex_enter(&ixgbe->gen_lock);
3850 
3851 	/*
3852 	 * X550 baseT PHY overtemp and LSC events are handled here.
3853 	 *
3854 	 * If an overtemp event occurs, it will be reflected in the
3855 	 * return value of phy.ops.handle_lasi() and the common code will
3856 	 * automatically power off the baseT PHY. This is our cue to trigger
3857 	 * an FMA event.
3858 	 *
3859 	 * If a link status change event occurs, phy.ops.handle_lasi() will
3860 	 * automatically initiate a link setup between the integrated KR PHY
3861 	 * and the external X557 PHY to ensure that the link speed between
3862 	 * them matches the link speed of the baseT link.
3863 	 */
3864 	rv = ixgbe_handle_lasi(hw);
3865 
3866 	if (rv == IXGBE_ERR_OVERTEMP) {
3867 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_OVERTEMP);
3868 
3869 		/*
3870 		 * Disable the adapter interrupts
3871 		 */
3872 		ixgbe_disable_adapter_interrupts(ixgbe);
3873 
3874 		/*
3875 		 * Disable Rx/Tx units
3876 		 */
3877 		(void) ixgbe_stop_adapter(hw);
3878 
3879 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
3880 		ixgbe_error(ixgbe,
3881 		    "Problem: Network adapter has been stopped due to a "
3882 		    "overtemperature event being detected.");
3883 		ixgbe_error(ixgbe,
3884 		    "Action: Shut down or restart the computer. If the issue "
3885 		    "persists, please take action in accordance with the "
3886 		    "recommendations from your system vendor.");
3887 	}
3888 
3889 	mutex_exit(&ixgbe->gen_lock);
3890 }
3891 
3892 /*
3893  * ixgbe_link_timer - timer for link status detection
3894  */
3895 static void
3896 ixgbe_link_timer(void *arg)
3897 {
3898 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3899 
3900 	mutex_enter(&ixgbe->gen_lock);
3901 	ixgbe_driver_link_check(ixgbe);
3902 	mutex_exit(&ixgbe->gen_lock);
3903 }
3904 
3905 /*
3906  * ixgbe_local_timer - Driver watchdog function.
3907  *
3908  * This function will handle the transmit stall check and other routines.
3909  */
3910 static void
3911 ixgbe_local_timer(void *arg)
3912 {
3913 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3914 
3915 	if (ixgbe->ixgbe_state & IXGBE_OVERTEMP)
3916 		goto out;
3917 
3918 	if (ixgbe->ixgbe_state & IXGBE_ERROR) {
3919 		ixgbe->reset_count++;
3920 		if (ixgbe_reset(ixgbe) == IXGBE_SUCCESS)
3921 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_RESTORED);
3922 		goto out;
3923 	}
3924 
3925 	if (ixgbe_stall_check(ixgbe)) {
3926 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_STALL);
3927 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3928 
3929 		ixgbe->reset_count++;
3930 		if (ixgbe_reset(ixgbe) == IXGBE_SUCCESS)
3931 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_RESTORED);
3932 	}
3933 
3934 out:
3935 	ixgbe_restart_watchdog_timer(ixgbe);
3936 }
3937 
3938 /*
3939  * ixgbe_stall_check - Check for transmit stall.
3940  *
3941  * This function checks if the adapter is stalled (in transmit).
3942  *
3943  * It is called each time the watchdog timeout is invoked.
3944  * If the transmit descriptor reclaim continuously fails,
3945  * the watchdog value will increment by 1. If the watchdog
3946  * value exceeds the threshold, the ixgbe is assumed to
3947  * have stalled and need to be reset.
3948  */
3949 static boolean_t
3950 ixgbe_stall_check(ixgbe_t *ixgbe)
3951 {
3952 	ixgbe_tx_ring_t *tx_ring;
3953 	boolean_t result;
3954 	int i;
3955 
3956 	if (ixgbe->link_state != LINK_STATE_UP)
3957 		return (B_FALSE);
3958 
3959 	/*
3960 	 * If any tx ring is stalled, we'll reset the chipset
3961 	 */
3962 	result = B_FALSE;
3963 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
3964 		tx_ring = &ixgbe->tx_rings[i];
3965 		if (tx_ring->tbd_free <= ixgbe->tx_recycle_thresh) {
3966 			tx_ring->tx_recycle(tx_ring);
3967 		}
3968 
3969 		if (tx_ring->recycle_fail > 0)
3970 			tx_ring->stall_watchdog++;
3971 		else
3972 			tx_ring->stall_watchdog = 0;
3973 
3974 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
3975 			result = B_TRUE;
3976 			break;
3977 		}
3978 	}
3979 
3980 	if (result) {
3981 		tx_ring->stall_watchdog = 0;
3982 		tx_ring->recycle_fail = 0;
3983 	}
3984 
3985 	return (result);
3986 }
3987 
3988 
3989 /*
3990  * is_valid_mac_addr - Check if the mac address is valid.
3991  */
3992 static boolean_t
3993 is_valid_mac_addr(uint8_t *mac_addr)
3994 {
3995 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
3996 	const uint8_t addr_test2[6] =
3997 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
3998 
3999 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
4000 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
4001 		return (B_FALSE);
4002 
4003 	return (B_TRUE);
4004 }
4005 
4006 static boolean_t
4007 ixgbe_find_mac_address(ixgbe_t *ixgbe)
4008 {
4009 #ifdef __sparc
4010 	struct ixgbe_hw *hw = &ixgbe->hw;
4011 	uchar_t *bytes;
4012 	struct ether_addr sysaddr;
4013 	uint_t nelts;
4014 	int err;
4015 	boolean_t found = B_FALSE;
4016 
4017 	/*
4018 	 * The "vendor's factory-set address" may already have
4019 	 * been extracted from the chip, but if the property
4020 	 * "local-mac-address" is set we use that instead.
4021 	 *
4022 	 * We check whether it looks like an array of 6
4023 	 * bytes (which it should, if OBP set it).  If we can't
4024 	 * make sense of it this way, we'll ignore it.
4025 	 */
4026 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip,
4027 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
4028 	if (err == DDI_PROP_SUCCESS) {
4029 		if (nelts == ETHERADDRL) {
4030 			while (nelts--)
4031 				hw->mac.addr[nelts] = bytes[nelts];
4032 			found = B_TRUE;
4033 		}
4034 		ddi_prop_free(bytes);
4035 	}
4036 
4037 	/*
4038 	 * Look up the OBP property "local-mac-address?". If the user has set
4039 	 * 'local-mac-address? = false', use "the system address" instead.
4040 	 */
4041 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip, 0,
4042 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
4043 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
4044 			if (localetheraddr(NULL, &sysaddr) != 0) {
4045 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
4046 				found = B_TRUE;
4047 			}
4048 		}
4049 		ddi_prop_free(bytes);
4050 	}
4051 
4052 	/*
4053 	 * Finally(!), if there's a valid "mac-address" property (created
4054 	 * if we netbooted from this interface), we must use this instead
4055 	 * of any of the above to ensure that the NFS/install server doesn't
4056 	 * get confused by the address changing as illumos takes over!
4057 	 */
4058 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip,
4059 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
4060 	if (err == DDI_PROP_SUCCESS) {
4061 		if (nelts == ETHERADDRL) {
4062 			while (nelts--)
4063 				hw->mac.addr[nelts] = bytes[nelts];
4064 			found = B_TRUE;
4065 		}
4066 		ddi_prop_free(bytes);
4067 	}
4068 
4069 	if (found) {
4070 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
4071 		return (B_TRUE);
4072 	}
4073 #else
4074 	_NOTE(ARGUNUSED(ixgbe));
4075 #endif
4076 
4077 	return (B_TRUE);
4078 }
4079 
4080 #pragma inline(ixgbe_arm_watchdog_timer)
4081 static void
4082 ixgbe_arm_watchdog_timer(ixgbe_t *ixgbe)
4083 {
4084 	/*
4085 	 * Fire a watchdog timer
4086 	 */
4087 	ixgbe->watchdog_tid =
4088 	    timeout(ixgbe_local_timer,
4089 	    (void *)ixgbe, 1 * drv_usectohz(1000000));
4090 
4091 }
4092 
4093 /*
4094  * ixgbe_enable_watchdog_timer - Enable and start the driver watchdog timer.
4095  */
4096 void
4097 ixgbe_enable_watchdog_timer(ixgbe_t *ixgbe)
4098 {
4099 	mutex_enter(&ixgbe->watchdog_lock);
4100 
4101 	if (!ixgbe->watchdog_enable) {
4102 		ixgbe->watchdog_enable = B_TRUE;
4103 		ixgbe->watchdog_start = B_TRUE;
4104 		ixgbe_arm_watchdog_timer(ixgbe);
4105 	}
4106 
4107 	mutex_exit(&ixgbe->watchdog_lock);
4108 }
4109 
4110 /*
4111  * ixgbe_disable_watchdog_timer - Disable and stop the driver watchdog timer.
4112  */
4113 void
4114 ixgbe_disable_watchdog_timer(ixgbe_t *ixgbe)
4115 {
4116 	timeout_id_t tid;
4117 
4118 	mutex_enter(&ixgbe->watchdog_lock);
4119 
4120 	ixgbe->watchdog_enable = B_FALSE;
4121 	ixgbe->watchdog_start = B_FALSE;
4122 	tid = ixgbe->watchdog_tid;
4123 	ixgbe->watchdog_tid = 0;
4124 
4125 	mutex_exit(&ixgbe->watchdog_lock);
4126 
4127 	if (tid != 0)
4128 		(void) untimeout(tid);
4129 }
4130 
4131 /*
4132  * ixgbe_start_watchdog_timer - Start the driver watchdog timer.
4133  */
4134 void
4135 ixgbe_start_watchdog_timer(ixgbe_t *ixgbe)
4136 {
4137 	mutex_enter(&ixgbe->watchdog_lock);
4138 
4139 	if (ixgbe->watchdog_enable) {
4140 		if (!ixgbe->watchdog_start) {
4141 			ixgbe->watchdog_start = B_TRUE;
4142 			ixgbe_arm_watchdog_timer(ixgbe);
4143 		}
4144 	}
4145 
4146 	mutex_exit(&ixgbe->watchdog_lock);
4147 }
4148 
4149 /*
4150  * ixgbe_restart_watchdog_timer - Restart the driver watchdog timer.
4151  */
4152 static void
4153 ixgbe_restart_watchdog_timer(ixgbe_t *ixgbe)
4154 {
4155 	mutex_enter(&ixgbe->watchdog_lock);
4156 
4157 	if (ixgbe->watchdog_start)
4158 		ixgbe_arm_watchdog_timer(ixgbe);
4159 
4160 	mutex_exit(&ixgbe->watchdog_lock);
4161 }
4162 
4163 /*
4164  * ixgbe_stop_watchdog_timer - Stop the driver watchdog timer.
4165  */
4166 void
4167 ixgbe_stop_watchdog_timer(ixgbe_t *ixgbe)
4168 {
4169 	timeout_id_t tid;
4170 
4171 	mutex_enter(&ixgbe->watchdog_lock);
4172 
4173 	ixgbe->watchdog_start = B_FALSE;
4174 	tid = ixgbe->watchdog_tid;
4175 	ixgbe->watchdog_tid = 0;
4176 
4177 	mutex_exit(&ixgbe->watchdog_lock);
4178 
4179 	if (tid != 0)
4180 		(void) untimeout(tid);
4181 }
4182 
4183 /*
4184  * ixgbe_disable_adapter_interrupts - Disable all adapter interrupts.
4185  */
4186 static void
4187 ixgbe_disable_adapter_interrupts(ixgbe_t *ixgbe)
4188 {
4189 	struct ixgbe_hw *hw = &ixgbe->hw;
4190 
4191 	/*
4192 	 * mask all interrupts off
4193 	 */
4194 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, 0xffffffff);
4195 
4196 	/*
4197 	 * for MSI-X, also disable autoclear
4198 	 */
4199 	if (ixgbe->intr_type == DDI_INTR_TYPE_MSIX) {
4200 		IXGBE_WRITE_REG(hw, IXGBE_EIAC, 0x0);
4201 	}
4202 
4203 	IXGBE_WRITE_FLUSH(hw);
4204 }
4205 
4206 /*
4207  * ixgbe_enable_adapter_interrupts - Enable all hardware interrupts.
4208  */
4209 static void
4210 ixgbe_enable_adapter_interrupts(ixgbe_t *ixgbe)
4211 {
4212 	struct ixgbe_hw *hw = &ixgbe->hw;
4213 	uint32_t eiac, eiam;
4214 	uint32_t gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
4215 
4216 	/* interrupt types to enable */
4217 	ixgbe->eims = IXGBE_EIMS_ENABLE_MASK;	/* shared code default */
4218 	ixgbe->eims &= ~IXGBE_EIMS_TCP_TIMER;	/* minus tcp timer */
4219 	ixgbe->eims |= ixgbe->capab->other_intr; /* "other" interrupt types */
4220 
4221 	/* enable automask on "other" causes that this adapter can generate */
4222 	eiam = ixgbe->capab->other_intr;
4223 
4224 	/*
4225 	 * msi-x mode
4226 	 */
4227 	if (ixgbe->intr_type == DDI_INTR_TYPE_MSIX) {
4228 		/* enable autoclear but not on bits 29:20 */
4229 		eiac = (ixgbe->eims & ~IXGBE_OTHER_INTR);
4230 
4231 		/* general purpose interrupt enable */
4232 		gpie |= (IXGBE_GPIE_MSIX_MODE
4233 		    | IXGBE_GPIE_PBA_SUPPORT
4234 		    | IXGBE_GPIE_OCD
4235 		    | IXGBE_GPIE_EIAME);
4236 	/*
4237 	 * non-msi-x mode
4238 	 */
4239 	} else {
4240 
4241 		/* disable autoclear, leave gpie at default */
4242 		eiac = 0;
4243 
4244 		/*
4245 		 * General purpose interrupt enable.
4246 		 * For 82599, X540 and X550, extended interrupt
4247 		 * automask enable only in MSI or MSI-X mode
4248 		 */
4249 		if ((hw->mac.type == ixgbe_mac_82598EB) ||
4250 		    (ixgbe->intr_type == DDI_INTR_TYPE_MSI)) {
4251 			gpie |= IXGBE_GPIE_EIAME;
4252 		}
4253 	}
4254 
4255 	/* Enable specific "other" interrupt types */
4256 	switch (hw->mac.type) {
4257 	case ixgbe_mac_82598EB:
4258 		gpie |= ixgbe->capab->other_gpie;
4259 		break;
4260 
4261 	case ixgbe_mac_82599EB:
4262 	case ixgbe_mac_X540:
4263 	case ixgbe_mac_X550:
4264 	case ixgbe_mac_X550EM_x:
4265 		gpie |= ixgbe->capab->other_gpie;
4266 
4267 		/* Enable RSC Delay 8us when LRO enabled  */
4268 		if (ixgbe->lro_enable) {
4269 			gpie |= (1 << IXGBE_GPIE_RSC_DELAY_SHIFT);
4270 		}
4271 		break;
4272 
4273 	default:
4274 		break;
4275 	}
4276 
4277 	/* write to interrupt control registers */
4278 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4279 	IXGBE_WRITE_REG(hw, IXGBE_EIAC, eiac);
4280 	IXGBE_WRITE_REG(hw, IXGBE_EIAM, eiam);
4281 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4282 	IXGBE_WRITE_FLUSH(hw);
4283 }
4284 
4285 /*
4286  * ixgbe_loopback_ioctl - Loopback support.
4287  */
4288 enum ioc_reply
4289 ixgbe_loopback_ioctl(ixgbe_t *ixgbe, struct iocblk *iocp, mblk_t *mp)
4290 {
4291 	lb_info_sz_t *lbsp;
4292 	lb_property_t *lbpp;
4293 	uint32_t *lbmp;
4294 	uint32_t size;
4295 	uint32_t value;
4296 
4297 	if (mp->b_cont == NULL)
4298 		return (IOC_INVAL);
4299 
4300 	switch (iocp->ioc_cmd) {
4301 	default:
4302 		return (IOC_INVAL);
4303 
4304 	case LB_GET_INFO_SIZE:
4305 		size = sizeof (lb_info_sz_t);
4306 		if (iocp->ioc_count != size)
4307 			return (IOC_INVAL);
4308 
4309 		value = sizeof (lb_normal);
4310 		value += sizeof (lb_mac);
4311 		value += sizeof (lb_external);
4312 
4313 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
4314 		*lbsp = value;
4315 		break;
4316 
4317 	case LB_GET_INFO:
4318 		value = sizeof (lb_normal);
4319 		value += sizeof (lb_mac);
4320 		value += sizeof (lb_external);
4321 
4322 		size = value;
4323 		if (iocp->ioc_count != size)
4324 			return (IOC_INVAL);
4325 
4326 		value = 0;
4327 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
4328 
4329 		lbpp[value++] = lb_normal;
4330 		lbpp[value++] = lb_mac;
4331 		lbpp[value++] = lb_external;
4332 		break;
4333 
4334 	case LB_GET_MODE:
4335 		size = sizeof (uint32_t);
4336 		if (iocp->ioc_count != size)
4337 			return (IOC_INVAL);
4338 
4339 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
4340 		*lbmp = ixgbe->loopback_mode;
4341 		break;
4342 
4343 	case LB_SET_MODE:
4344 		size = 0;
4345 		if (iocp->ioc_count != sizeof (uint32_t))
4346 			return (IOC_INVAL);
4347 
4348 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
4349 		if (!ixgbe_set_loopback_mode(ixgbe, *lbmp))
4350 			return (IOC_INVAL);
4351 		break;
4352 	}
4353 
4354 	iocp->ioc_count = size;
4355 	iocp->ioc_error = 0;
4356 
4357 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
4358 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4359 		return (IOC_INVAL);
4360 	}
4361 
4362 	return (IOC_REPLY);
4363 }
4364 
4365 /*
4366  * ixgbe_set_loopback_mode - Setup loopback based on the loopback mode.
4367  */
4368 static boolean_t
4369 ixgbe_set_loopback_mode(ixgbe_t *ixgbe, uint32_t mode)
4370 {
4371 	if (mode == ixgbe->loopback_mode)
4372 		return (B_TRUE);
4373 
4374 	ixgbe->loopback_mode = mode;
4375 
4376 	if (mode == IXGBE_LB_NONE) {
4377 		/*
4378 		 * Reset the chip
4379 		 */
4380 		(void) ixgbe_reset(ixgbe);
4381 		return (B_TRUE);
4382 	}
4383 
4384 	mutex_enter(&ixgbe->gen_lock);
4385 
4386 	switch (mode) {
4387 	default:
4388 		mutex_exit(&ixgbe->gen_lock);
4389 		return (B_FALSE);
4390 
4391 	case IXGBE_LB_EXTERNAL:
4392 		break;
4393 
4394 	case IXGBE_LB_INTERNAL_MAC:
4395 		ixgbe_set_internal_mac_loopback(ixgbe);
4396 		break;
4397 	}
4398 
4399 	mutex_exit(&ixgbe->gen_lock);
4400 
4401 	return (B_TRUE);
4402 }
4403 
4404 /*
4405  * ixgbe_set_internal_mac_loopback - Set the internal MAC loopback mode.
4406  */
4407 static void
4408 ixgbe_set_internal_mac_loopback(ixgbe_t *ixgbe)
4409 {
4410 	struct ixgbe_hw *hw;
4411 	uint32_t reg;
4412 	uint8_t atlas;
4413 
4414 	hw = &ixgbe->hw;
4415 
4416 	/*
4417 	 * Setup MAC loopback
4418 	 */
4419 	reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_HLREG0);
4420 	reg |= IXGBE_HLREG0_LPBK;
4421 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_HLREG0, reg);
4422 
4423 	reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_AUTOC);
4424 	reg &= ~IXGBE_AUTOC_LMS_MASK;
4425 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_AUTOC, reg);
4426 
4427 	/*
4428 	 * Disable Atlas Tx lanes to keep packets in loopback and not on wire
4429 	 */
4430 	switch (hw->mac.type) {
4431 	case ixgbe_mac_82598EB:
4432 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_LPBK,
4433 		    &atlas);
4434 		atlas |= IXGBE_ATLAS_PDN_TX_REG_EN;
4435 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_LPBK,
4436 		    atlas);
4437 
4438 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_10G,
4439 		    &atlas);
4440 		atlas |= IXGBE_ATLAS_PDN_TX_10G_QL_ALL;
4441 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_10G,
4442 		    atlas);
4443 
4444 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_1G,
4445 		    &atlas);
4446 		atlas |= IXGBE_ATLAS_PDN_TX_1G_QL_ALL;
4447 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_1G,
4448 		    atlas);
4449 
4450 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_AN,
4451 		    &atlas);
4452 		atlas |= IXGBE_ATLAS_PDN_TX_AN_QL_ALL;
4453 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_AN,
4454 		    atlas);
4455 		break;
4456 
4457 	case ixgbe_mac_82599EB:
4458 	case ixgbe_mac_X540:
4459 	case ixgbe_mac_X550:
4460 	case ixgbe_mac_X550EM_x:
4461 		reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_AUTOC);
4462 		reg |= (IXGBE_AUTOC_FLU |
4463 		    IXGBE_AUTOC_10G_KX4);
4464 		IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_AUTOC, reg);
4465 
4466 		(void) ixgbe_setup_link(&ixgbe->hw, IXGBE_LINK_SPEED_10GB_FULL,
4467 		    B_FALSE);
4468 		break;
4469 
4470 	default:
4471 		break;
4472 	}
4473 }
4474 
4475 #pragma inline(ixgbe_intr_rx_work)
4476 /*
4477  * ixgbe_intr_rx_work - RX processing of ISR.
4478  */
4479 static void
4480 ixgbe_intr_rx_work(ixgbe_rx_ring_t *rx_ring)
4481 {
4482 	mblk_t *mp;
4483 
4484 	mutex_enter(&rx_ring->rx_lock);
4485 
4486 	mp = ixgbe_ring_rx(rx_ring, IXGBE_POLL_NULL);
4487 	mutex_exit(&rx_ring->rx_lock);
4488 
4489 	if (mp != NULL)
4490 		mac_rx_ring(rx_ring->ixgbe->mac_hdl, rx_ring->ring_handle, mp,
4491 		    rx_ring->ring_gen_num);
4492 }
4493 
4494 #pragma inline(ixgbe_intr_tx_work)
4495 /*
4496  * ixgbe_intr_tx_work - TX processing of ISR.
4497  */
4498 static void
4499 ixgbe_intr_tx_work(ixgbe_tx_ring_t *tx_ring)
4500 {
4501 	ixgbe_t *ixgbe = tx_ring->ixgbe;
4502 
4503 	/*
4504 	 * Recycle the tx descriptors
4505 	 */
4506 	tx_ring->tx_recycle(tx_ring);
4507 
4508 	/*
4509 	 * Schedule the re-transmit
4510 	 */
4511 	if (tx_ring->reschedule &&
4512 	    (tx_ring->tbd_free >= ixgbe->tx_resched_thresh)) {
4513 		tx_ring->reschedule = B_FALSE;
4514 		mac_tx_ring_update(tx_ring->ixgbe->mac_hdl,
4515 		    tx_ring->ring_handle);
4516 		IXGBE_DEBUG_STAT(tx_ring->stat_reschedule);
4517 	}
4518 }
4519 
4520 #pragma inline(ixgbe_intr_other_work)
4521 /*
4522  * ixgbe_intr_other_work - Process interrupt types other than tx/rx
4523  */
4524 static void
4525 ixgbe_intr_other_work(ixgbe_t *ixgbe, uint32_t eicr)
4526 {
4527 	struct ixgbe_hw *hw = &ixgbe->hw;
4528 
4529 	ASSERT(mutex_owned(&ixgbe->gen_lock));
4530 
4531 	/*
4532 	 * handle link status change
4533 	 */
4534 	if (eicr & IXGBE_EICR_LSC) {
4535 		ixgbe_driver_link_check(ixgbe);
4536 		ixgbe_get_hw_state(ixgbe);
4537 	}
4538 
4539 	/*
4540 	 * check for fan failure on adapters with fans
4541 	 */
4542 	if ((ixgbe->capab->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
4543 	    (eicr & IXGBE_EICR_GPI_SDP1)) {
4544 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_OVERTEMP);
4545 
4546 		/*
4547 		 * Disable the adapter interrupts
4548 		 */
4549 		ixgbe_disable_adapter_interrupts(ixgbe);
4550 
4551 		/*
4552 		 * Disable Rx/Tx units
4553 		 */
4554 		(void) ixgbe_stop_adapter(&ixgbe->hw);
4555 
4556 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
4557 		ixgbe_error(ixgbe,
4558 		    "Problem: Network adapter has been stopped "
4559 		    "because the fan has stopped.\n");
4560 		ixgbe_error(ixgbe,
4561 		    "Action: Replace the adapter.\n");
4562 
4563 		/* re-enable the interrupt, which was automasked */
4564 		ixgbe->eims |= IXGBE_EICR_GPI_SDP1;
4565 	}
4566 
4567 	/*
4568 	 * Do SFP check for adapters with hot-plug capability
4569 	 */
4570 	if ((ixgbe->capab->flags & IXGBE_FLAG_SFP_PLUG_CAPABLE) &&
4571 	    ((eicr & IXGBE_EICR_GPI_SDP1_BY_MAC(hw)) ||
4572 	    (eicr & IXGBE_EICR_GPI_SDP2_BY_MAC(hw)))) {
4573 		ixgbe->eicr = eicr;
4574 		if ((ddi_taskq_dispatch(ixgbe->sfp_taskq,
4575 		    ixgbe_sfp_check, (void *)ixgbe,
4576 		    DDI_NOSLEEP)) != DDI_SUCCESS) {
4577 			ixgbe_log(ixgbe, "No memory available to dispatch "
4578 			    "taskq for SFP check");
4579 		}
4580 	}
4581 
4582 	/*
4583 	 * Do over-temperature check for adapters with temp sensor
4584 	 */
4585 	if ((ixgbe->capab->flags & IXGBE_FLAG_TEMP_SENSOR_CAPABLE) &&
4586 	    ((eicr & IXGBE_EICR_GPI_SDP0_BY_MAC(hw)) ||
4587 	    (eicr & IXGBE_EICR_LSC))) {
4588 		ixgbe->eicr = eicr;
4589 		if ((ddi_taskq_dispatch(ixgbe->overtemp_taskq,
4590 		    ixgbe_overtemp_check, (void *)ixgbe,
4591 		    DDI_NOSLEEP)) != DDI_SUCCESS) {
4592 			ixgbe_log(ixgbe, "No memory available to dispatch "
4593 			    "taskq for overtemp check");
4594 		}
4595 	}
4596 
4597 	/*
4598 	 * Process an external PHY interrupt
4599 	 */
4600 	if (hw->device_id == IXGBE_DEV_ID_X550EM_X_10G_T &&
4601 	    (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
4602 		ixgbe->eicr = eicr;
4603 		if ((ddi_taskq_dispatch(ixgbe->phy_taskq,
4604 		    ixgbe_phy_check, (void *)ixgbe,
4605 		    DDI_NOSLEEP)) != DDI_SUCCESS) {
4606 			ixgbe_log(ixgbe, "No memory available to dispatch "
4607 			    "taskq for PHY check");
4608 		}
4609 	}
4610 }
4611 
4612 /*
4613  * ixgbe_intr_legacy - Interrupt handler for legacy interrupts.
4614  */
4615 static uint_t
4616 ixgbe_intr_legacy(void *arg1, void *arg2)
4617 {
4618 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
4619 	struct ixgbe_hw *hw = &ixgbe->hw;
4620 	ixgbe_tx_ring_t *tx_ring;
4621 	ixgbe_rx_ring_t *rx_ring;
4622 	uint32_t eicr;
4623 	mblk_t *mp;
4624 	boolean_t tx_reschedule;
4625 	uint_t result;
4626 
4627 	_NOTE(ARGUNUSED(arg2));
4628 
4629 	mutex_enter(&ixgbe->gen_lock);
4630 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
4631 		mutex_exit(&ixgbe->gen_lock);
4632 		return (DDI_INTR_UNCLAIMED);
4633 	}
4634 
4635 	mp = NULL;
4636 	tx_reschedule = B_FALSE;
4637 
4638 	/*
4639 	 * Any bit set in eicr: claim this interrupt
4640 	 */
4641 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4642 
4643 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
4644 		mutex_exit(&ixgbe->gen_lock);
4645 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4646 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4647 		return (DDI_INTR_CLAIMED);
4648 	}
4649 
4650 	if (eicr) {
4651 		/*
4652 		 * For legacy interrupt, we have only one interrupt,
4653 		 * so we have only one rx ring and one tx ring enabled.
4654 		 */
4655 		ASSERT(ixgbe->num_rx_rings == 1);
4656 		ASSERT(ixgbe->num_tx_rings == 1);
4657 
4658 		/*
4659 		 * For legacy interrupt, rx rings[0] will use RTxQ[0].
4660 		 */
4661 		if (eicr & 0x1) {
4662 			ixgbe->eimc |= IXGBE_EICR_RTX_QUEUE;
4663 			IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4664 			ixgbe->eims |= IXGBE_EICR_RTX_QUEUE;
4665 			/*
4666 			 * Clean the rx descriptors
4667 			 */
4668 			rx_ring = &ixgbe->rx_rings[0];
4669 			mp = ixgbe_ring_rx(rx_ring, IXGBE_POLL_NULL);
4670 		}
4671 
4672 		/*
4673 		 * For legacy interrupt, tx rings[0] will use RTxQ[1].
4674 		 */
4675 		if (eicr & 0x2) {
4676 			/*
4677 			 * Recycle the tx descriptors
4678 			 */
4679 			tx_ring = &ixgbe->tx_rings[0];
4680 			tx_ring->tx_recycle(tx_ring);
4681 
4682 			/*
4683 			 * Schedule the re-transmit
4684 			 */
4685 			tx_reschedule = (tx_ring->reschedule &&
4686 			    (tx_ring->tbd_free >= ixgbe->tx_resched_thresh));
4687 		}
4688 
4689 		/* any interrupt type other than tx/rx */
4690 		if (eicr & ixgbe->capab->other_intr) {
4691 			switch (hw->mac.type) {
4692 			case ixgbe_mac_82598EB:
4693 				ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4694 				break;
4695 
4696 			case ixgbe_mac_82599EB:
4697 			case ixgbe_mac_X540:
4698 			case ixgbe_mac_X550:
4699 			case ixgbe_mac_X550EM_x:
4700 				ixgbe->eimc = IXGBE_82599_OTHER_INTR;
4701 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4702 				break;
4703 
4704 			default:
4705 				break;
4706 			}
4707 			ixgbe_intr_other_work(ixgbe, eicr);
4708 			ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4709 		}
4710 
4711 		mutex_exit(&ixgbe->gen_lock);
4712 
4713 		result = DDI_INTR_CLAIMED;
4714 	} else {
4715 		mutex_exit(&ixgbe->gen_lock);
4716 
4717 		/*
4718 		 * No interrupt cause bits set: don't claim this interrupt.
4719 		 */
4720 		result = DDI_INTR_UNCLAIMED;
4721 	}
4722 
4723 	/* re-enable the interrupts which were automasked */
4724 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4725 
4726 	/*
4727 	 * Do the following work outside of the gen_lock
4728 	 */
4729 	if (mp != NULL) {
4730 		mac_rx_ring(rx_ring->ixgbe->mac_hdl, rx_ring->ring_handle, mp,
4731 		    rx_ring->ring_gen_num);
4732 	}
4733 
4734 	if (tx_reschedule)  {
4735 		tx_ring->reschedule = B_FALSE;
4736 		mac_tx_ring_update(ixgbe->mac_hdl, tx_ring->ring_handle);
4737 		IXGBE_DEBUG_STAT(tx_ring->stat_reschedule);
4738 	}
4739 
4740 	return (result);
4741 }
4742 
4743 /*
4744  * ixgbe_intr_msi - Interrupt handler for MSI.
4745  */
4746 static uint_t
4747 ixgbe_intr_msi(void *arg1, void *arg2)
4748 {
4749 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
4750 	struct ixgbe_hw *hw = &ixgbe->hw;
4751 	uint32_t eicr;
4752 
4753 	_NOTE(ARGUNUSED(arg2));
4754 
4755 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4756 
4757 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
4758 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4759 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4760 		return (DDI_INTR_CLAIMED);
4761 	}
4762 
4763 	/*
4764 	 * For MSI interrupt, we have only one vector,
4765 	 * so we have only one rx ring and one tx ring enabled.
4766 	 */
4767 	ASSERT(ixgbe->num_rx_rings == 1);
4768 	ASSERT(ixgbe->num_tx_rings == 1);
4769 
4770 	/*
4771 	 * For MSI interrupt, rx rings[0] will use RTxQ[0].
4772 	 */
4773 	if (eicr & 0x1) {
4774 		ixgbe_intr_rx_work(&ixgbe->rx_rings[0]);
4775 	}
4776 
4777 	/*
4778 	 * For MSI interrupt, tx rings[0] will use RTxQ[1].
4779 	 */
4780 	if (eicr & 0x2) {
4781 		ixgbe_intr_tx_work(&ixgbe->tx_rings[0]);
4782 	}
4783 
4784 	/* any interrupt type other than tx/rx */
4785 	if (eicr & ixgbe->capab->other_intr) {
4786 		mutex_enter(&ixgbe->gen_lock);
4787 		switch (hw->mac.type) {
4788 		case ixgbe_mac_82598EB:
4789 			ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4790 			break;
4791 
4792 		case ixgbe_mac_82599EB:
4793 		case ixgbe_mac_X540:
4794 		case ixgbe_mac_X550:
4795 		case ixgbe_mac_X550EM_x:
4796 			ixgbe->eimc = IXGBE_82599_OTHER_INTR;
4797 			IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4798 			break;
4799 
4800 		default:
4801 			break;
4802 		}
4803 		ixgbe_intr_other_work(ixgbe, eicr);
4804 		ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4805 		mutex_exit(&ixgbe->gen_lock);
4806 	}
4807 
4808 	/* re-enable the interrupts which were automasked */
4809 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4810 
4811 	return (DDI_INTR_CLAIMED);
4812 }
4813 
4814 /*
4815  * ixgbe_intr_msix - Interrupt handler for MSI-X.
4816  */
4817 static uint_t
4818 ixgbe_intr_msix(void *arg1, void *arg2)
4819 {
4820 	ixgbe_intr_vector_t *vect = (ixgbe_intr_vector_t *)arg1;
4821 	ixgbe_t *ixgbe = vect->ixgbe;
4822 	struct ixgbe_hw *hw = &ixgbe->hw;
4823 	uint32_t eicr;
4824 	int r_idx = 0;
4825 
4826 	_NOTE(ARGUNUSED(arg2));
4827 
4828 	/*
4829 	 * Clean each rx ring that has its bit set in the map
4830 	 */
4831 	r_idx = bt_getlowbit(vect->rx_map, 0, (ixgbe->num_rx_rings - 1));
4832 	while (r_idx >= 0) {
4833 		ixgbe_intr_rx_work(&ixgbe->rx_rings[r_idx]);
4834 		r_idx = bt_getlowbit(vect->rx_map, (r_idx + 1),
4835 		    (ixgbe->num_rx_rings - 1));
4836 	}
4837 
4838 	/*
4839 	 * Clean each tx ring that has its bit set in the map
4840 	 */
4841 	r_idx = bt_getlowbit(vect->tx_map, 0, (ixgbe->num_tx_rings - 1));
4842 	while (r_idx >= 0) {
4843 		ixgbe_intr_tx_work(&ixgbe->tx_rings[r_idx]);
4844 		r_idx = bt_getlowbit(vect->tx_map, (r_idx + 1),
4845 		    (ixgbe->num_tx_rings - 1));
4846 	}
4847 
4848 
4849 	/*
4850 	 * Clean other interrupt (link change) that has its bit set in the map
4851 	 */
4852 	if (BT_TEST(vect->other_map, 0) == 1) {
4853 		eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4854 
4855 		if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) !=
4856 		    DDI_FM_OK) {
4857 			ddi_fm_service_impact(ixgbe->dip,
4858 			    DDI_SERVICE_DEGRADED);
4859 			atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4860 			return (DDI_INTR_CLAIMED);
4861 		}
4862 
4863 		/*
4864 		 * Check "other" cause bits: any interrupt type other than tx/rx
4865 		 */
4866 		if (eicr & ixgbe->capab->other_intr) {
4867 			mutex_enter(&ixgbe->gen_lock);
4868 			switch (hw->mac.type) {
4869 			case ixgbe_mac_82598EB:
4870 				ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4871 				ixgbe_intr_other_work(ixgbe, eicr);
4872 				break;
4873 
4874 			case ixgbe_mac_82599EB:
4875 			case ixgbe_mac_X540:
4876 			case ixgbe_mac_X550:
4877 			case ixgbe_mac_X550EM_x:
4878 				ixgbe->eims |= IXGBE_EICR_RTX_QUEUE;
4879 				ixgbe_intr_other_work(ixgbe, eicr);
4880 				break;
4881 
4882 			default:
4883 				break;
4884 			}
4885 			mutex_exit(&ixgbe->gen_lock);
4886 		}
4887 
4888 		/* re-enable the interrupts which were automasked */
4889 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4890 	}
4891 
4892 	return (DDI_INTR_CLAIMED);
4893 }
4894 
4895 /*
4896  * ixgbe_alloc_intrs - Allocate interrupts for the driver.
4897  *
4898  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
4899  * if not successful, try Legacy.
4900  * ixgbe->intr_force can be used to force sequence to start with
4901  * any of the 3 types.
4902  * If MSI-X is not used, number of tx/rx rings is forced to 1.
4903  */
4904 static int
4905 ixgbe_alloc_intrs(ixgbe_t *ixgbe)
4906 {
4907 	dev_info_t *devinfo;
4908 	int intr_types;
4909 	int rc;
4910 
4911 	devinfo = ixgbe->dip;
4912 
4913 	/*
4914 	 * Get supported interrupt types
4915 	 */
4916 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
4917 
4918 	if (rc != DDI_SUCCESS) {
4919 		ixgbe_log(ixgbe,
4920 		    "Get supported interrupt types failed: %d", rc);
4921 		return (IXGBE_FAILURE);
4922 	}
4923 	IXGBE_DEBUGLOG_1(ixgbe, "Supported interrupt types: %x", intr_types);
4924 
4925 	ixgbe->intr_type = 0;
4926 
4927 	/*
4928 	 * Install MSI-X interrupts
4929 	 */
4930 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
4931 	    (ixgbe->intr_force <= IXGBE_INTR_MSIX)) {
4932 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_MSIX);
4933 		if (rc == IXGBE_SUCCESS)
4934 			return (IXGBE_SUCCESS);
4935 
4936 		ixgbe_log(ixgbe,
4937 		    "Allocate MSI-X failed, trying MSI interrupts...");
4938 	}
4939 
4940 	/*
4941 	 * MSI-X not used, force rings and groups to 1
4942 	 */
4943 	ixgbe->num_rx_rings = 1;
4944 	ixgbe->num_rx_groups = 1;
4945 	ixgbe->num_tx_rings = 1;
4946 	ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
4947 	ixgbe_log(ixgbe,
4948 	    "MSI-X not used, force rings and groups number to 1");
4949 
4950 	/*
4951 	 * Install MSI interrupts
4952 	 */
4953 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
4954 	    (ixgbe->intr_force <= IXGBE_INTR_MSI)) {
4955 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_MSI);
4956 		if (rc == IXGBE_SUCCESS)
4957 			return (IXGBE_SUCCESS);
4958 
4959 		ixgbe_log(ixgbe,
4960 		    "Allocate MSI failed, trying Legacy interrupts...");
4961 	}
4962 
4963 	/*
4964 	 * Install legacy interrupts
4965 	 */
4966 	if (intr_types & DDI_INTR_TYPE_FIXED) {
4967 		/*
4968 		 * Disallow legacy interrupts for X550. X550 has a silicon
4969 		 * bug which prevents Shared Legacy interrupts from working.
4970 		 * For details, please reference:
4971 		 *
4972 		 * Intel Ethernet Controller X550 Specification Update rev. 2.1
4973 		 * May 2016, erratum 22: PCIe Interrupt Status Bit
4974 		 */
4975 		if (ixgbe->hw.mac.type == ixgbe_mac_X550 ||
4976 		    ixgbe->hw.mac.type == ixgbe_mac_X550EM_x ||
4977 		    ixgbe->hw.mac.type == ixgbe_mac_X550_vf ||
4978 		    ixgbe->hw.mac.type == ixgbe_mac_X550EM_x_vf) {
4979 			ixgbe_log(ixgbe,
4980 			    "Legacy interrupts are not supported on this "
4981 			    "adapter. Please use MSI or MSI-X instead.");
4982 			return (IXGBE_FAILURE);
4983 		}
4984 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_FIXED);
4985 		if (rc == IXGBE_SUCCESS)
4986 			return (IXGBE_SUCCESS);
4987 
4988 		ixgbe_log(ixgbe,
4989 		    "Allocate Legacy interrupts failed");
4990 	}
4991 
4992 	/*
4993 	 * If none of the 3 types succeeded, return failure
4994 	 */
4995 	return (IXGBE_FAILURE);
4996 }
4997 
4998 /*
4999  * ixgbe_alloc_intr_handles - Allocate interrupt handles.
5000  *
5001  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
5002  * if fewer than 2 handles are available, return failure.
5003  * Upon success, this maps the vectors to rx and tx rings for
5004  * interrupts.
5005  */
5006 static int
5007 ixgbe_alloc_intr_handles(ixgbe_t *ixgbe, int intr_type)
5008 {
5009 	dev_info_t *devinfo;
5010 	int request, count, actual;
5011 	int minimum;
5012 	int rc;
5013 	uint32_t ring_per_group;
5014 
5015 	devinfo = ixgbe->dip;
5016 
5017 	switch (intr_type) {
5018 	case DDI_INTR_TYPE_FIXED:
5019 		request = 1;	/* Request 1 legacy interrupt handle */
5020 		minimum = 1;
5021 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: legacy");
5022 		break;
5023 
5024 	case DDI_INTR_TYPE_MSI:
5025 		request = 1;	/* Request 1 MSI interrupt handle */
5026 		minimum = 1;
5027 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: MSI");
5028 		break;
5029 
5030 	case DDI_INTR_TYPE_MSIX:
5031 		/*
5032 		 * Best number of vectors for the adapter is
5033 		 * (# rx rings + # tx rings), however we will
5034 		 * limit the request number.
5035 		 */
5036 		request = min(16, ixgbe->num_rx_rings + ixgbe->num_tx_rings);
5037 		if (request > ixgbe->capab->max_ring_vect)
5038 			request = ixgbe->capab->max_ring_vect;
5039 		minimum = 1;
5040 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: MSI-X");
5041 		break;
5042 
5043 	default:
5044 		ixgbe_log(ixgbe,
5045 		    "invalid call to ixgbe_alloc_intr_handles(): %d\n",
5046 		    intr_type);
5047 		return (IXGBE_FAILURE);
5048 	}
5049 	IXGBE_DEBUGLOG_2(ixgbe, "interrupt handles requested: %d  minimum: %d",
5050 	    request, minimum);
5051 
5052 	/*
5053 	 * Get number of supported interrupts
5054 	 */
5055 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
5056 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
5057 		ixgbe_log(ixgbe,
5058 		    "Get interrupt number failed. Return: %d, count: %d",
5059 		    rc, count);
5060 		return (IXGBE_FAILURE);
5061 	}
5062 	IXGBE_DEBUGLOG_1(ixgbe, "interrupts supported: %d", count);
5063 
5064 	actual = 0;
5065 	ixgbe->intr_cnt = 0;
5066 	ixgbe->intr_cnt_max = 0;
5067 	ixgbe->intr_cnt_min = 0;
5068 
5069 	/*
5070 	 * Allocate an array of interrupt handles
5071 	 */
5072 	ixgbe->intr_size = request * sizeof (ddi_intr_handle_t);
5073 	ixgbe->htable = kmem_alloc(ixgbe->intr_size, KM_SLEEP);
5074 
5075 	rc = ddi_intr_alloc(devinfo, ixgbe->htable, intr_type, 0,
5076 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
5077 	if (rc != DDI_SUCCESS) {
5078 		ixgbe_log(ixgbe, "Allocate interrupts failed. "
5079 		    "return: %d, request: %d, actual: %d",
5080 		    rc, request, actual);
5081 		goto alloc_handle_fail;
5082 	}
5083 	IXGBE_DEBUGLOG_1(ixgbe, "interrupts actually allocated: %d", actual);
5084 
5085 	/*
5086 	 * upper/lower limit of interrupts
5087 	 */
5088 	ixgbe->intr_cnt = actual;
5089 	ixgbe->intr_cnt_max = request;
5090 	ixgbe->intr_cnt_min = minimum;
5091 
5092 	/*
5093 	 * rss number per group should not exceed the rx interrupt number,
5094 	 * else need to adjust rx ring number.
5095 	 */
5096 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
5097 	ASSERT((ixgbe->num_rx_rings % ixgbe->num_rx_groups) == 0);
5098 	if (actual < ring_per_group) {
5099 		ixgbe->num_rx_rings = ixgbe->num_rx_groups * actual;
5100 		ixgbe_setup_vmdq_rss_conf(ixgbe);
5101 	}
5102 
5103 	/*
5104 	 * Now we know the actual number of vectors.  Here we map the vector
5105 	 * to other, rx rings and tx ring.
5106 	 */
5107 	if (actual < minimum) {
5108 		ixgbe_log(ixgbe, "Insufficient interrupt handles available: %d",
5109 		    actual);
5110 		goto alloc_handle_fail;
5111 	}
5112 
5113 	/*
5114 	 * Get priority for first vector, assume remaining are all the same
5115 	 */
5116 	rc = ddi_intr_get_pri(ixgbe->htable[0], &ixgbe->intr_pri);
5117 	if (rc != DDI_SUCCESS) {
5118 		ixgbe_log(ixgbe,
5119 		    "Get interrupt priority failed: %d", rc);
5120 		goto alloc_handle_fail;
5121 	}
5122 
5123 	rc = ddi_intr_get_cap(ixgbe->htable[0], &ixgbe->intr_cap);
5124 	if (rc != DDI_SUCCESS) {
5125 		ixgbe_log(ixgbe,
5126 		    "Get interrupt cap failed: %d", rc);
5127 		goto alloc_handle_fail;
5128 	}
5129 
5130 	ixgbe->intr_type = intr_type;
5131 
5132 	return (IXGBE_SUCCESS);
5133 
5134 alloc_handle_fail:
5135 	ixgbe_rem_intrs(ixgbe);
5136 
5137 	return (IXGBE_FAILURE);
5138 }
5139 
5140 /*
5141  * ixgbe_add_intr_handlers - Add interrupt handlers based on the interrupt type.
5142  *
5143  * Before adding the interrupt handlers, the interrupt vectors have
5144  * been allocated, and the rx/tx rings have also been allocated.
5145  */
5146 static int
5147 ixgbe_add_intr_handlers(ixgbe_t *ixgbe)
5148 {
5149 	int vector = 0;
5150 	int rc;
5151 
5152 	switch (ixgbe->intr_type) {
5153 	case DDI_INTR_TYPE_MSIX:
5154 		/*
5155 		 * Add interrupt handler for all vectors
5156 		 */
5157 		for (vector = 0; vector < ixgbe->intr_cnt; vector++) {
5158 			/*
5159 			 * install pointer to vect_map[vector]
5160 			 */
5161 			rc = ddi_intr_add_handler(ixgbe->htable[vector],
5162 			    (ddi_intr_handler_t *)ixgbe_intr_msix,
5163 			    (void *)&ixgbe->vect_map[vector], NULL);
5164 
5165 			if (rc != DDI_SUCCESS) {
5166 				ixgbe_log(ixgbe,
5167 				    "Add interrupt handler failed. "
5168 				    "return: %d, vector: %d", rc, vector);
5169 				for (vector--; vector >= 0; vector--) {
5170 					(void) ddi_intr_remove_handler(
5171 					    ixgbe->htable[vector]);
5172 				}
5173 				return (IXGBE_FAILURE);
5174 			}
5175 		}
5176 
5177 		break;
5178 
5179 	case DDI_INTR_TYPE_MSI:
5180 		/*
5181 		 * Add interrupt handlers for the only vector
5182 		 */
5183 		rc = ddi_intr_add_handler(ixgbe->htable[vector],
5184 		    (ddi_intr_handler_t *)ixgbe_intr_msi,
5185 		    (void *)ixgbe, NULL);
5186 
5187 		if (rc != DDI_SUCCESS) {
5188 			ixgbe_log(ixgbe,
5189 			    "Add MSI interrupt handler failed: %d", rc);
5190 			return (IXGBE_FAILURE);
5191 		}
5192 
5193 		break;
5194 
5195 	case DDI_INTR_TYPE_FIXED:
5196 		/*
5197 		 * Add interrupt handlers for the only vector
5198 		 */
5199 		rc = ddi_intr_add_handler(ixgbe->htable[vector],
5200 		    (ddi_intr_handler_t *)ixgbe_intr_legacy,
5201 		    (void *)ixgbe, NULL);
5202 
5203 		if (rc != DDI_SUCCESS) {
5204 			ixgbe_log(ixgbe,
5205 			    "Add legacy interrupt handler failed: %d", rc);
5206 			return (IXGBE_FAILURE);
5207 		}
5208 
5209 		break;
5210 
5211 	default:
5212 		return (IXGBE_FAILURE);
5213 	}
5214 
5215 	return (IXGBE_SUCCESS);
5216 }
5217 
5218 #pragma inline(ixgbe_map_rxring_to_vector)
5219 /*
5220  * ixgbe_map_rxring_to_vector - Map given rx ring to given interrupt vector.
5221  */
5222 static void
5223 ixgbe_map_rxring_to_vector(ixgbe_t *ixgbe, int r_idx, int v_idx)
5224 {
5225 	/*
5226 	 * Set bit in map
5227 	 */
5228 	BT_SET(ixgbe->vect_map[v_idx].rx_map, r_idx);
5229 
5230 	/*
5231 	 * Count bits set
5232 	 */
5233 	ixgbe->vect_map[v_idx].rxr_cnt++;
5234 
5235 	/*
5236 	 * Remember bit position
5237 	 */
5238 	ixgbe->rx_rings[r_idx].intr_vector = v_idx;
5239 	ixgbe->rx_rings[r_idx].vect_bit = 1 << v_idx;
5240 }
5241 
5242 #pragma inline(ixgbe_map_txring_to_vector)
5243 /*
5244  * ixgbe_map_txring_to_vector - Map given tx ring to given interrupt vector.
5245  */
5246 static void
5247 ixgbe_map_txring_to_vector(ixgbe_t *ixgbe, int t_idx, int v_idx)
5248 {
5249 	/*
5250 	 * Set bit in map
5251 	 */
5252 	BT_SET(ixgbe->vect_map[v_idx].tx_map, t_idx);
5253 
5254 	/*
5255 	 * Count bits set
5256 	 */
5257 	ixgbe->vect_map[v_idx].txr_cnt++;
5258 
5259 	/*
5260 	 * Remember bit position
5261 	 */
5262 	ixgbe->tx_rings[t_idx].intr_vector = v_idx;
5263 	ixgbe->tx_rings[t_idx].vect_bit = 1 << v_idx;
5264 }
5265 
5266 /*
5267  * ixgbe_setup_ivar - Set the given entry in the given interrupt vector
5268  * allocation register (IVAR).
5269  * cause:
5270  *   -1 : other cause
5271  *    0 : rx
5272  *    1 : tx
5273  */
5274 static void
5275 ixgbe_setup_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, uint8_t msix_vector,
5276     int8_t cause)
5277 {
5278 	struct ixgbe_hw *hw = &ixgbe->hw;
5279 	u32 ivar, index;
5280 
5281 	switch (hw->mac.type) {
5282 	case ixgbe_mac_82598EB:
5283 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
5284 		if (cause == -1) {
5285 			cause = 0;
5286 		}
5287 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
5288 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
5289 		ivar &= ~(0xFF << (8 * (intr_alloc_entry & 0x3)));
5290 		ivar |= (msix_vector << (8 * (intr_alloc_entry & 0x3)));
5291 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
5292 		break;
5293 
5294 	case ixgbe_mac_82599EB:
5295 	case ixgbe_mac_X540:
5296 	case ixgbe_mac_X550:
5297 	case ixgbe_mac_X550EM_x:
5298 		if (cause == -1) {
5299 			/* other causes */
5300 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
5301 			index = (intr_alloc_entry & 1) * 8;
5302 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
5303 			ivar &= ~(0xFF << index);
5304 			ivar |= (msix_vector << index);
5305 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
5306 		} else {
5307 			/* tx or rx causes */
5308 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
5309 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
5310 			ivar = IXGBE_READ_REG(hw,
5311 			    IXGBE_IVAR(intr_alloc_entry >> 1));
5312 			ivar &= ~(0xFF << index);
5313 			ivar |= (msix_vector << index);
5314 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
5315 			    ivar);
5316 		}
5317 		break;
5318 
5319 	default:
5320 		break;
5321 	}
5322 }
5323 
5324 /*
5325  * ixgbe_enable_ivar - Enable the given entry by setting the VAL bit of
5326  * given interrupt vector allocation register (IVAR).
5327  * cause:
5328  *   -1 : other cause
5329  *    0 : rx
5330  *    1 : tx
5331  */
5332 static void
5333 ixgbe_enable_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, int8_t cause)
5334 {
5335 	struct ixgbe_hw *hw = &ixgbe->hw;
5336 	u32 ivar, index;
5337 
5338 	switch (hw->mac.type) {
5339 	case ixgbe_mac_82598EB:
5340 		if (cause == -1) {
5341 			cause = 0;
5342 		}
5343 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
5344 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
5345 		ivar |= (IXGBE_IVAR_ALLOC_VAL << (8 *
5346 		    (intr_alloc_entry & 0x3)));
5347 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
5348 		break;
5349 
5350 	case ixgbe_mac_82599EB:
5351 	case ixgbe_mac_X540:
5352 	case ixgbe_mac_X550:
5353 	case ixgbe_mac_X550EM_x:
5354 		if (cause == -1) {
5355 			/* other causes */
5356 			index = (intr_alloc_entry & 1) * 8;
5357 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
5358 			ivar |= (IXGBE_IVAR_ALLOC_VAL << index);
5359 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
5360 		} else {
5361 			/* tx or rx causes */
5362 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
5363 			ivar = IXGBE_READ_REG(hw,
5364 			    IXGBE_IVAR(intr_alloc_entry >> 1));
5365 			ivar |= (IXGBE_IVAR_ALLOC_VAL << index);
5366 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
5367 			    ivar);
5368 		}
5369 		break;
5370 
5371 	default:
5372 		break;
5373 	}
5374 }
5375 
5376 /*
5377  * ixgbe_disable_ivar - Disble the given entry by clearing the VAL bit of
5378  * given interrupt vector allocation register (IVAR).
5379  * cause:
5380  *   -1 : other cause
5381  *    0 : rx
5382  *    1 : tx
5383  */
5384 static void
5385 ixgbe_disable_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, int8_t cause)
5386 {
5387 	struct ixgbe_hw *hw = &ixgbe->hw;
5388 	u32 ivar, index;
5389 
5390 	switch (hw->mac.type) {
5391 	case ixgbe_mac_82598EB:
5392 		if (cause == -1) {
5393 			cause = 0;
5394 		}
5395 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
5396 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
5397 		ivar &= ~(IXGBE_IVAR_ALLOC_VAL<< (8 *
5398 		    (intr_alloc_entry & 0x3)));
5399 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
5400 		break;
5401 
5402 	case ixgbe_mac_82599EB:
5403 	case ixgbe_mac_X540:
5404 	case ixgbe_mac_X550:
5405 	case ixgbe_mac_X550EM_x:
5406 		if (cause == -1) {
5407 			/* other causes */
5408 			index = (intr_alloc_entry & 1) * 8;
5409 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
5410 			ivar &= ~(IXGBE_IVAR_ALLOC_VAL << index);
5411 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
5412 		} else {
5413 			/* tx or rx causes */
5414 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
5415 			ivar = IXGBE_READ_REG(hw,
5416 			    IXGBE_IVAR(intr_alloc_entry >> 1));
5417 			ivar &= ~(IXGBE_IVAR_ALLOC_VAL << index);
5418 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
5419 			    ivar);
5420 		}
5421 		break;
5422 
5423 	default:
5424 		break;
5425 	}
5426 }
5427 
5428 /*
5429  * Convert the rx ring index driver maintained to the rx ring index
5430  * in h/w.
5431  */
5432 static uint32_t
5433 ixgbe_get_hw_rx_index(ixgbe_t *ixgbe, uint32_t sw_rx_index)
5434 {
5435 
5436 	struct ixgbe_hw *hw = &ixgbe->hw;
5437 	uint32_t rx_ring_per_group, hw_rx_index;
5438 
5439 	if (ixgbe->classify_mode == IXGBE_CLASSIFY_RSS ||
5440 	    ixgbe->classify_mode == IXGBE_CLASSIFY_NONE) {
5441 		return (sw_rx_index);
5442 	} else if (ixgbe->classify_mode == IXGBE_CLASSIFY_VMDQ) {
5443 		switch (hw->mac.type) {
5444 		case ixgbe_mac_82598EB:
5445 			return (sw_rx_index);
5446 
5447 		case ixgbe_mac_82599EB:
5448 		case ixgbe_mac_X540:
5449 		case ixgbe_mac_X550:
5450 		case ixgbe_mac_X550EM_x:
5451 			return (sw_rx_index * 2);
5452 
5453 		default:
5454 			break;
5455 		}
5456 	} else if (ixgbe->classify_mode == IXGBE_CLASSIFY_VMDQ_RSS) {
5457 		rx_ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
5458 
5459 		switch (hw->mac.type) {
5460 		case ixgbe_mac_82598EB:
5461 			hw_rx_index = (sw_rx_index / rx_ring_per_group) *
5462 			    16 + (sw_rx_index % rx_ring_per_group);
5463 			return (hw_rx_index);
5464 
5465 		case ixgbe_mac_82599EB:
5466 		case ixgbe_mac_X540:
5467 		case ixgbe_mac_X550:
5468 		case ixgbe_mac_X550EM_x:
5469 			if (ixgbe->num_rx_groups > 32) {
5470 				hw_rx_index = (sw_rx_index /
5471 				    rx_ring_per_group) * 2 +
5472 				    (sw_rx_index % rx_ring_per_group);
5473 			} else {
5474 				hw_rx_index = (sw_rx_index /
5475 				    rx_ring_per_group) * 4 +
5476 				    (sw_rx_index % rx_ring_per_group);
5477 			}
5478 			return (hw_rx_index);
5479 
5480 		default:
5481 			break;
5482 		}
5483 	}
5484 
5485 	/*
5486 	 * Should never reach. Just to make compiler happy.
5487 	 */
5488 	return (sw_rx_index);
5489 }
5490 
5491 /*
5492  * ixgbe_map_intrs_to_vectors - Map different interrupts to MSI-X vectors.
5493  *
5494  * For MSI-X, here will map rx interrupt, tx interrupt and other interrupt
5495  * to vector[0 - (intr_cnt -1)].
5496  */
5497 static int
5498 ixgbe_map_intrs_to_vectors(ixgbe_t *ixgbe)
5499 {
5500 	int i, vector = 0;
5501 
5502 	/* initialize vector map */
5503 	bzero(&ixgbe->vect_map, sizeof (ixgbe->vect_map));
5504 	for (i = 0; i < ixgbe->intr_cnt; i++) {
5505 		ixgbe->vect_map[i].ixgbe = ixgbe;
5506 	}
5507 
5508 	/*
5509 	 * non-MSI-X case is very simple: rx rings[0] on RTxQ[0],
5510 	 * tx rings[0] on RTxQ[1].
5511 	 */
5512 	if (ixgbe->intr_type != DDI_INTR_TYPE_MSIX) {
5513 		ixgbe_map_rxring_to_vector(ixgbe, 0, 0);
5514 		ixgbe_map_txring_to_vector(ixgbe, 0, 1);
5515 		return (IXGBE_SUCCESS);
5516 	}
5517 
5518 	/*
5519 	 * Interrupts/vectors mapping for MSI-X
5520 	 */
5521 
5522 	/*
5523 	 * Map other interrupt to vector 0,
5524 	 * Set bit in map and count the bits set.
5525 	 */
5526 	BT_SET(ixgbe->vect_map[vector].other_map, 0);
5527 	ixgbe->vect_map[vector].other_cnt++;
5528 
5529 	/*
5530 	 * Map rx ring interrupts to vectors
5531 	 */
5532 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
5533 		ixgbe_map_rxring_to_vector(ixgbe, i, vector);
5534 		vector = (vector +1) % ixgbe->intr_cnt;
5535 	}
5536 
5537 	/*
5538 	 * Map tx ring interrupts to vectors
5539 	 */
5540 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
5541 		ixgbe_map_txring_to_vector(ixgbe, i, vector);
5542 		vector = (vector +1) % ixgbe->intr_cnt;
5543 	}
5544 
5545 	return (IXGBE_SUCCESS);
5546 }
5547 
5548 /*
5549  * ixgbe_setup_adapter_vector - Setup the adapter interrupt vector(s).
5550  *
5551  * This relies on ring/vector mapping already set up in the
5552  * vect_map[] structures
5553  */
5554 static void
5555 ixgbe_setup_adapter_vector(ixgbe_t *ixgbe)
5556 {
5557 	struct ixgbe_hw *hw = &ixgbe->hw;
5558 	ixgbe_intr_vector_t *vect;	/* vector bitmap */
5559 	int r_idx;	/* ring index */
5560 	int v_idx;	/* vector index */
5561 	uint32_t hw_index;
5562 
5563 	/*
5564 	 * Clear any previous entries
5565 	 */
5566 	switch (hw->mac.type) {
5567 	case ixgbe_mac_82598EB:
5568 		for (v_idx = 0; v_idx < 25; v_idx++)
5569 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(v_idx), 0);
5570 		break;
5571 
5572 	case ixgbe_mac_82599EB:
5573 	case ixgbe_mac_X540:
5574 	case ixgbe_mac_X550:
5575 	case ixgbe_mac_X550EM_x:
5576 		for (v_idx = 0; v_idx < 64; v_idx++)
5577 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(v_idx), 0);
5578 		IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, 0);
5579 		break;
5580 
5581 	default:
5582 		break;
5583 	}
5584 
5585 	/*
5586 	 * For non MSI-X interrupt, rx rings[0] will use RTxQ[0], and
5587 	 * tx rings[0] will use RTxQ[1].
5588 	 */
5589 	if (ixgbe->intr_type != DDI_INTR_TYPE_MSIX) {
5590 		ixgbe_setup_ivar(ixgbe, 0, 0, 0);
5591 		ixgbe_setup_ivar(ixgbe, 0, 1, 1);
5592 		return;
5593 	}
5594 
5595 	/*
5596 	 * For MSI-X interrupt, "Other" is always on vector[0].
5597 	 */
5598 	ixgbe_setup_ivar(ixgbe, IXGBE_IVAR_OTHER_CAUSES_INDEX, 0, -1);
5599 
5600 	/*
5601 	 * For each interrupt vector, populate the IVAR table
5602 	 */
5603 	for (v_idx = 0; v_idx < ixgbe->intr_cnt; v_idx++) {
5604 		vect = &ixgbe->vect_map[v_idx];
5605 
5606 		/*
5607 		 * For each rx ring bit set
5608 		 */
5609 		r_idx = bt_getlowbit(vect->rx_map, 0,
5610 		    (ixgbe->num_rx_rings - 1));
5611 
5612 		while (r_idx >= 0) {
5613 			hw_index = ixgbe->rx_rings[r_idx].hw_index;
5614 			ixgbe_setup_ivar(ixgbe, hw_index, v_idx, 0);
5615 			r_idx = bt_getlowbit(vect->rx_map, (r_idx + 1),
5616 			    (ixgbe->num_rx_rings - 1));
5617 		}
5618 
5619 		/*
5620 		 * For each tx ring bit set
5621 		 */
5622 		r_idx = bt_getlowbit(vect->tx_map, 0,
5623 		    (ixgbe->num_tx_rings - 1));
5624 
5625 		while (r_idx >= 0) {
5626 			ixgbe_setup_ivar(ixgbe, r_idx, v_idx, 1);
5627 			r_idx = bt_getlowbit(vect->tx_map, (r_idx + 1),
5628 			    (ixgbe->num_tx_rings - 1));
5629 		}
5630 	}
5631 }
5632 
5633 /*
5634  * ixgbe_rem_intr_handlers - Remove the interrupt handlers.
5635  */
5636 static void
5637 ixgbe_rem_intr_handlers(ixgbe_t *ixgbe)
5638 {
5639 	int i;
5640 	int rc;
5641 
5642 	for (i = 0; i < ixgbe->intr_cnt; i++) {
5643 		rc = ddi_intr_remove_handler(ixgbe->htable[i]);
5644 		if (rc != DDI_SUCCESS) {
5645 			IXGBE_DEBUGLOG_1(ixgbe,
5646 			    "Remove intr handler failed: %d", rc);
5647 		}
5648 	}
5649 }
5650 
5651 /*
5652  * ixgbe_rem_intrs - Remove the allocated interrupts.
5653  */
5654 static void
5655 ixgbe_rem_intrs(ixgbe_t *ixgbe)
5656 {
5657 	int i;
5658 	int rc;
5659 
5660 	for (i = 0; i < ixgbe->intr_cnt; i++) {
5661 		rc = ddi_intr_free(ixgbe->htable[i]);
5662 		if (rc != DDI_SUCCESS) {
5663 			IXGBE_DEBUGLOG_1(ixgbe,
5664 			    "Free intr failed: %d", rc);
5665 		}
5666 	}
5667 
5668 	kmem_free(ixgbe->htable, ixgbe->intr_size);
5669 	ixgbe->htable = NULL;
5670 }
5671 
5672 /*
5673  * ixgbe_enable_intrs - Enable all the ddi interrupts.
5674  */
5675 static int
5676 ixgbe_enable_intrs(ixgbe_t *ixgbe)
5677 {
5678 	int i;
5679 	int rc;
5680 
5681 	/*
5682 	 * Enable interrupts
5683 	 */
5684 	if (ixgbe->intr_cap & DDI_INTR_FLAG_BLOCK) {
5685 		/*
5686 		 * Call ddi_intr_block_enable() for MSI
5687 		 */
5688 		rc = ddi_intr_block_enable(ixgbe->htable, ixgbe->intr_cnt);
5689 		if (rc != DDI_SUCCESS) {
5690 			ixgbe_log(ixgbe,
5691 			    "Enable block intr failed: %d", rc);
5692 			return (IXGBE_FAILURE);
5693 		}
5694 	} else {
5695 		/*
5696 		 * Call ddi_intr_enable() for Legacy/MSI non block enable
5697 		 */
5698 		for (i = 0; i < ixgbe->intr_cnt; i++) {
5699 			rc = ddi_intr_enable(ixgbe->htable[i]);
5700 			if (rc != DDI_SUCCESS) {
5701 				ixgbe_log(ixgbe,
5702 				    "Enable intr failed: %d", rc);
5703 				return (IXGBE_FAILURE);
5704 			}
5705 		}
5706 	}
5707 
5708 	return (IXGBE_SUCCESS);
5709 }
5710 
5711 /*
5712  * ixgbe_disable_intrs - Disable all the interrupts.
5713  */
5714 static int
5715 ixgbe_disable_intrs(ixgbe_t *ixgbe)
5716 {
5717 	int i;
5718 	int rc;
5719 
5720 	/*
5721 	 * Disable all interrupts
5722 	 */
5723 	if (ixgbe->intr_cap & DDI_INTR_FLAG_BLOCK) {
5724 		rc = ddi_intr_block_disable(ixgbe->htable, ixgbe->intr_cnt);
5725 		if (rc != DDI_SUCCESS) {
5726 			ixgbe_log(ixgbe,
5727 			    "Disable block intr failed: %d", rc);
5728 			return (IXGBE_FAILURE);
5729 		}
5730 	} else {
5731 		for (i = 0; i < ixgbe->intr_cnt; i++) {
5732 			rc = ddi_intr_disable(ixgbe->htable[i]);
5733 			if (rc != DDI_SUCCESS) {
5734 				ixgbe_log(ixgbe,
5735 				    "Disable intr failed: %d", rc);
5736 				return (IXGBE_FAILURE);
5737 			}
5738 		}
5739 	}
5740 
5741 	return (IXGBE_SUCCESS);
5742 }
5743 
5744 /*
5745  * ixgbe_get_hw_state - Get and save parameters related to adapter hardware.
5746  */
5747 static void
5748 ixgbe_get_hw_state(ixgbe_t *ixgbe)
5749 {
5750 	struct ixgbe_hw *hw = &ixgbe->hw;
5751 	ixgbe_link_speed speed = 0;
5752 	boolean_t link_up = B_FALSE;
5753 	uint32_t pcs1g_anlp = 0;
5754 
5755 	ASSERT(mutex_owned(&ixgbe->gen_lock));
5756 	ixgbe->param_lp_1000fdx_cap = 0;
5757 	ixgbe->param_lp_100fdx_cap  = 0;
5758 
5759 	/* check for link, don't wait */
5760 	(void) ixgbe_check_link(hw, &speed, &link_up, B_FALSE);
5761 
5762 	/*
5763 	 * Update the observed Link Partner's capabilities. Not all adapters
5764 	 * can provide full information on the LP's capable speeds, so we
5765 	 * provide what we can.
5766 	 */
5767 	if (link_up) {
5768 		pcs1g_anlp = IXGBE_READ_REG(hw, IXGBE_PCS1GANLP);
5769 
5770 		ixgbe->param_lp_1000fdx_cap =
5771 		    (pcs1g_anlp & IXGBE_PCS1GANLP_LPFD) ? 1 : 0;
5772 		ixgbe->param_lp_100fdx_cap =
5773 		    (pcs1g_anlp & IXGBE_PCS1GANLP_LPFD) ? 1 : 0;
5774 	}
5775 
5776 	/*
5777 	 * Update GLD's notion of the adapter's currently advertised speeds.
5778 	 * Since the common code doesn't always record the current autonegotiate
5779 	 * settings in the phy struct for all parts (specifically, adapters with
5780 	 * SFPs) we first test to see if it is 0, and if so, we fall back to
5781 	 * using the adapter's speed capabilities which we saved during instance
5782 	 * init in ixgbe_init_params().
5783 	 *
5784 	 * Adapters with SFPs will always be shown as advertising all of their
5785 	 * supported speeds, and adapters with baseT PHYs (where the phy struct
5786 	 * is maintained by the common code) will always have a factual view of
5787 	 * their currently-advertised speeds. In the case of SFPs, this is
5788 	 * acceptable as we default to advertising all speeds that the adapter
5789 	 * claims to support, and those properties are immutable; unlike on
5790 	 * baseT (copper) PHYs, where speeds can be enabled or disabled at will.
5791 	 */
5792 	speed = hw->phy.autoneg_advertised;
5793 	if (speed == 0)
5794 		speed = ixgbe->speeds_supported;
5795 
5796 	ixgbe->param_adv_10000fdx_cap =
5797 	    (speed & IXGBE_LINK_SPEED_10GB_FULL) ? 1 : 0;
5798 	ixgbe->param_adv_5000fdx_cap =
5799 	    (speed & IXGBE_LINK_SPEED_5GB_FULL) ? 1 : 0;
5800 	ixgbe->param_adv_2500fdx_cap =
5801 	    (speed & IXGBE_LINK_SPEED_2_5GB_FULL) ? 1 : 0;
5802 	ixgbe->param_adv_1000fdx_cap =
5803 	    (speed & IXGBE_LINK_SPEED_1GB_FULL) ? 1 : 0;
5804 	ixgbe->param_adv_100fdx_cap =
5805 	    (speed & IXGBE_LINK_SPEED_100_FULL) ? 1 : 0;
5806 }
5807 
5808 /*
5809  * ixgbe_get_driver_control - Notify that driver is in control of device.
5810  */
5811 static void
5812 ixgbe_get_driver_control(struct ixgbe_hw *hw)
5813 {
5814 	uint32_t ctrl_ext;
5815 
5816 	/*
5817 	 * Notify firmware that driver is in control of device
5818 	 */
5819 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5820 	ctrl_ext |= IXGBE_CTRL_EXT_DRV_LOAD;
5821 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5822 }
5823 
5824 /*
5825  * ixgbe_release_driver_control - Notify that driver is no longer in control
5826  * of device.
5827  */
5828 static void
5829 ixgbe_release_driver_control(struct ixgbe_hw *hw)
5830 {
5831 	uint32_t ctrl_ext;
5832 
5833 	/*
5834 	 * Notify firmware that driver is no longer in control of device
5835 	 */
5836 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5837 	ctrl_ext &= ~IXGBE_CTRL_EXT_DRV_LOAD;
5838 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5839 }
5840 
5841 /*
5842  * ixgbe_atomic_reserve - Atomic decrease operation.
5843  */
5844 int
5845 ixgbe_atomic_reserve(uint32_t *count_p, uint32_t n)
5846 {
5847 	uint32_t oldval;
5848 	uint32_t newval;
5849 
5850 	/*
5851 	 * ATOMICALLY
5852 	 */
5853 	do {
5854 		oldval = *count_p;
5855 		if (oldval < n)
5856 			return (-1);
5857 		newval = oldval - n;
5858 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
5859 
5860 	return (newval);
5861 }
5862 
5863 /*
5864  * ixgbe_mc_table_itr - Traverse the entries in the multicast table.
5865  */
5866 static uint8_t *
5867 ixgbe_mc_table_itr(struct ixgbe_hw *hw, uint8_t **upd_ptr, uint32_t *vmdq)
5868 {
5869 	uint8_t *addr = *upd_ptr;
5870 	uint8_t *new_ptr;
5871 
5872 	_NOTE(ARGUNUSED(hw));
5873 	_NOTE(ARGUNUSED(vmdq));
5874 
5875 	new_ptr = addr + IXGBE_ETH_LENGTH_OF_ADDRESS;
5876 	*upd_ptr = new_ptr;
5877 	return (addr);
5878 }
5879 
5880 /*
5881  * FMA support
5882  */
5883 int
5884 ixgbe_check_acc_handle(ddi_acc_handle_t handle)
5885 {
5886 	ddi_fm_error_t de;
5887 
5888 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
5889 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
5890 	return (de.fme_status);
5891 }
5892 
5893 int
5894 ixgbe_check_dma_handle(ddi_dma_handle_t handle)
5895 {
5896 	ddi_fm_error_t de;
5897 
5898 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
5899 	return (de.fme_status);
5900 }
5901 
5902 /*
5903  * ixgbe_fm_error_cb - The IO fault service error handling callback function.
5904  */
5905 static int
5906 ixgbe_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
5907 {
5908 	_NOTE(ARGUNUSED(impl_data));
5909 	/*
5910 	 * as the driver can always deal with an error in any dma or
5911 	 * access handle, we can just return the fme_status value.
5912 	 */
5913 	pci_ereport_post(dip, err, NULL);
5914 	return (err->fme_status);
5915 }
5916 
5917 static void
5918 ixgbe_fm_init(ixgbe_t *ixgbe)
5919 {
5920 	ddi_iblock_cookie_t iblk;
5921 	int fma_dma_flag;
5922 
5923 	/*
5924 	 * Only register with IO Fault Services if we have some capability
5925 	 */
5926 	if (ixgbe->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
5927 		ixgbe_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
5928 	} else {
5929 		ixgbe_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
5930 	}
5931 
5932 	if (ixgbe->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
5933 		fma_dma_flag = 1;
5934 	} else {
5935 		fma_dma_flag = 0;
5936 	}
5937 
5938 	ixgbe_set_fma_flags(fma_dma_flag);
5939 
5940 	if (ixgbe->fm_capabilities) {
5941 
5942 		/*
5943 		 * Register capabilities with IO Fault Services
5944 		 */
5945 		ddi_fm_init(ixgbe->dip, &ixgbe->fm_capabilities, &iblk);
5946 
5947 		/*
5948 		 * Initialize pci ereport capabilities if ereport capable
5949 		 */
5950 		if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities) ||
5951 		    DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5952 			pci_ereport_setup(ixgbe->dip);
5953 
5954 		/*
5955 		 * Register error callback if error callback capable
5956 		 */
5957 		if (DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5958 			ddi_fm_handler_register(ixgbe->dip,
5959 			    ixgbe_fm_error_cb, (void*) ixgbe);
5960 	}
5961 }
5962 
5963 static void
5964 ixgbe_fm_fini(ixgbe_t *ixgbe)
5965 {
5966 	/*
5967 	 * Only unregister FMA capabilities if they are registered
5968 	 */
5969 	if (ixgbe->fm_capabilities) {
5970 
5971 		/*
5972 		 * Release any resources allocated by pci_ereport_setup()
5973 		 */
5974 		if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities) ||
5975 		    DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5976 			pci_ereport_teardown(ixgbe->dip);
5977 
5978 		/*
5979 		 * Un-register error callback if error callback capable
5980 		 */
5981 		if (DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5982 			ddi_fm_handler_unregister(ixgbe->dip);
5983 
5984 		/*
5985 		 * Unregister from IO Fault Service
5986 		 */
5987 		ddi_fm_fini(ixgbe->dip);
5988 	}
5989 }
5990 
5991 void
5992 ixgbe_fm_ereport(ixgbe_t *ixgbe, char *detail)
5993 {
5994 	uint64_t ena;
5995 	char buf[FM_MAX_CLASS];
5996 
5997 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
5998 	ena = fm_ena_generate(0, FM_ENA_FMT1);
5999 	if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities)) {
6000 		ddi_fm_ereport_post(ixgbe->dip, buf, ena, DDI_NOSLEEP,
6001 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
6002 	}
6003 }
6004 
6005 static int
6006 ixgbe_ring_start(mac_ring_driver_t rh, uint64_t mr_gen_num)
6007 {
6008 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)rh;
6009 
6010 	mutex_enter(&rx_ring->rx_lock);
6011 	rx_ring->ring_gen_num = mr_gen_num;
6012 	mutex_exit(&rx_ring->rx_lock);
6013 	return (0);
6014 }
6015 
6016 /*
6017  * Get the global ring index by a ring index within a group.
6018  */
6019 static int
6020 ixgbe_get_rx_ring_index(ixgbe_t *ixgbe, int gindex, int rindex)
6021 {
6022 	ixgbe_rx_ring_t *rx_ring;
6023 	int i;
6024 
6025 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
6026 		rx_ring = &ixgbe->rx_rings[i];
6027 		if (rx_ring->group_index == gindex)
6028 			rindex--;
6029 		if (rindex < 0)
6030 			return (i);
6031 	}
6032 
6033 	return (-1);
6034 }
6035 
6036 /*
6037  * Callback funtion for MAC layer to register all rings.
6038  */
6039 /* ARGSUSED */
6040 void
6041 ixgbe_fill_ring(void *arg, mac_ring_type_t rtype, const int group_index,
6042     const int ring_index, mac_ring_info_t *infop, mac_ring_handle_t rh)
6043 {
6044 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
6045 	mac_intr_t *mintr = &infop->mri_intr;
6046 
6047 	switch (rtype) {
6048 	case MAC_RING_TYPE_RX: {
6049 		/*
6050 		 * 'index' is the ring index within the group.
6051 		 * Need to get the global ring index by searching in groups.
6052 		 */
6053 		int global_ring_index = ixgbe_get_rx_ring_index(
6054 		    ixgbe, group_index, ring_index);
6055 
6056 		ASSERT(global_ring_index >= 0);
6057 
6058 		ixgbe_rx_ring_t *rx_ring = &ixgbe->rx_rings[global_ring_index];
6059 		rx_ring->ring_handle = rh;
6060 
6061 		infop->mri_driver = (mac_ring_driver_t)rx_ring;
6062 		infop->mri_start = ixgbe_ring_start;
6063 		infop->mri_stop = NULL;
6064 		infop->mri_poll = ixgbe_ring_rx_poll;
6065 		infop->mri_stat = ixgbe_rx_ring_stat;
6066 
6067 		mintr->mi_handle = (mac_intr_handle_t)rx_ring;
6068 		mintr->mi_enable = ixgbe_rx_ring_intr_enable;
6069 		mintr->mi_disable = ixgbe_rx_ring_intr_disable;
6070 		if (ixgbe->intr_type &
6071 		    (DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI)) {
6072 			mintr->mi_ddi_handle =
6073 			    ixgbe->htable[rx_ring->intr_vector];
6074 		}
6075 
6076 		break;
6077 	}
6078 	case MAC_RING_TYPE_TX: {
6079 		ASSERT(group_index == -1);
6080 		ASSERT(ring_index < ixgbe->num_tx_rings);
6081 
6082 		ixgbe_tx_ring_t *tx_ring = &ixgbe->tx_rings[ring_index];
6083 		tx_ring->ring_handle = rh;
6084 
6085 		infop->mri_driver = (mac_ring_driver_t)tx_ring;
6086 		infop->mri_start = NULL;
6087 		infop->mri_stop = NULL;
6088 		infop->mri_tx = ixgbe_ring_tx;
6089 		infop->mri_stat = ixgbe_tx_ring_stat;
6090 		if (ixgbe->intr_type &
6091 		    (DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI)) {
6092 			mintr->mi_ddi_handle =
6093 			    ixgbe->htable[tx_ring->intr_vector];
6094 		}
6095 		break;
6096 	}
6097 	default:
6098 		break;
6099 	}
6100 }
6101 
6102 /*
6103  * Callback funtion for MAC layer to register all groups.
6104  */
6105 void
6106 ixgbe_fill_group(void *arg, mac_ring_type_t rtype, const int index,
6107     mac_group_info_t *infop, mac_group_handle_t gh)
6108 {
6109 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
6110 
6111 	switch (rtype) {
6112 	case MAC_RING_TYPE_RX: {
6113 		ixgbe_rx_group_t *rx_group;
6114 
6115 		rx_group = &ixgbe->rx_groups[index];
6116 		rx_group->group_handle = gh;
6117 
6118 		infop->mgi_driver = (mac_group_driver_t)rx_group;
6119 		infop->mgi_start = NULL;
6120 		infop->mgi_stop = NULL;
6121 		infop->mgi_addmac = ixgbe_addmac;
6122 		infop->mgi_remmac = ixgbe_remmac;
6123 		infop->mgi_count = (ixgbe->num_rx_rings / ixgbe->num_rx_groups);
6124 
6125 		break;
6126 	}
6127 	case MAC_RING_TYPE_TX:
6128 		break;
6129 	default:
6130 		break;
6131 	}
6132 }
6133 
6134 /*
6135  * Enable interrupt on the specificed rx ring.
6136  */
6137 int
6138 ixgbe_rx_ring_intr_enable(mac_intr_handle_t intrh)
6139 {
6140 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)intrh;
6141 	ixgbe_t *ixgbe = rx_ring->ixgbe;
6142 	int r_idx = rx_ring->index;
6143 	int hw_r_idx = rx_ring->hw_index;
6144 	int v_idx = rx_ring->intr_vector;
6145 
6146 	mutex_enter(&ixgbe->gen_lock);
6147 	if (ixgbe->ixgbe_state & IXGBE_INTR_ADJUST) {
6148 		mutex_exit(&ixgbe->gen_lock);
6149 		/*
6150 		 * Simply return 0.
6151 		 * Interrupts are being adjusted. ixgbe_intr_adjust()
6152 		 * will eventually re-enable the interrupt when it's
6153 		 * done with the adjustment.
6154 		 */
6155 		return (0);
6156 	}
6157 
6158 	/*
6159 	 * To enable interrupt by setting the VAL bit of given interrupt
6160 	 * vector allocation register (IVAR).
6161 	 */
6162 	ixgbe_enable_ivar(ixgbe, hw_r_idx, 0);
6163 
6164 	BT_SET(ixgbe->vect_map[v_idx].rx_map, r_idx);
6165 
6166 	/*
6167 	 * Trigger a Rx interrupt on this ring
6168 	 */
6169 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_EICS, (1 << v_idx));
6170 	IXGBE_WRITE_FLUSH(&ixgbe->hw);
6171 
6172 	mutex_exit(&ixgbe->gen_lock);
6173 
6174 	return (0);
6175 }
6176 
6177 /*
6178  * Disable interrupt on the specificed rx ring.
6179  */
6180 int
6181 ixgbe_rx_ring_intr_disable(mac_intr_handle_t intrh)
6182 {
6183 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)intrh;
6184 	ixgbe_t *ixgbe = rx_ring->ixgbe;
6185 	int r_idx = rx_ring->index;
6186 	int hw_r_idx = rx_ring->hw_index;
6187 	int v_idx = rx_ring->intr_vector;
6188 
6189 	mutex_enter(&ixgbe->gen_lock);
6190 	if (ixgbe->ixgbe_state & IXGBE_INTR_ADJUST) {
6191 		mutex_exit(&ixgbe->gen_lock);
6192 		/*
6193 		 * Simply return 0.
6194 		 * In the rare case where an interrupt is being
6195 		 * disabled while interrupts are being adjusted,
6196 		 * we don't fail the operation. No interrupts will
6197 		 * be generated while they are adjusted, and
6198 		 * ixgbe_intr_adjust() will cause the interrupts
6199 		 * to be re-enabled once it completes. Note that
6200 		 * in this case, packets may be delivered to the
6201 		 * stack via interrupts before xgbe_rx_ring_intr_enable()
6202 		 * is called again. This is acceptable since interrupt
6203 		 * adjustment is infrequent, and the stack will be
6204 		 * able to handle these packets.
6205 		 */
6206 		return (0);
6207 	}
6208 
6209 	/*
6210 	 * To disable interrupt by clearing the VAL bit of given interrupt
6211 	 * vector allocation register (IVAR).
6212 	 */
6213 	ixgbe_disable_ivar(ixgbe, hw_r_idx, 0);
6214 
6215 	BT_CLEAR(ixgbe->vect_map[v_idx].rx_map, r_idx);
6216 
6217 	mutex_exit(&ixgbe->gen_lock);
6218 
6219 	return (0);
6220 }
6221 
6222 /*
6223  * Add a mac address.
6224  */
6225 static int
6226 ixgbe_addmac(void *arg, const uint8_t *mac_addr)
6227 {
6228 	ixgbe_rx_group_t *rx_group = (ixgbe_rx_group_t *)arg;
6229 	ixgbe_t *ixgbe = rx_group->ixgbe;
6230 	struct ixgbe_hw *hw = &ixgbe->hw;
6231 	int slot, i;
6232 
6233 	mutex_enter(&ixgbe->gen_lock);
6234 
6235 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
6236 		mutex_exit(&ixgbe->gen_lock);
6237 		return (ECANCELED);
6238 	}
6239 
6240 	if (ixgbe->unicst_avail == 0) {
6241 		/* no slots available */
6242 		mutex_exit(&ixgbe->gen_lock);
6243 		return (ENOSPC);
6244 	}
6245 
6246 	/*
6247 	 * The first ixgbe->num_rx_groups slots are reserved for each respective
6248 	 * group. The rest slots are shared by all groups. While adding a
6249 	 * MAC address, reserved slots are firstly checked then the shared
6250 	 * slots are searched.
6251 	 */
6252 	slot = -1;
6253 	if (ixgbe->unicst_addr[rx_group->index].mac.set == 1) {
6254 		for (i = ixgbe->num_rx_groups; i < ixgbe->unicst_total; i++) {
6255 			if (ixgbe->unicst_addr[i].mac.set == 0) {
6256 				slot = i;
6257 				break;
6258 			}
6259 		}
6260 	} else {
6261 		slot = rx_group->index;
6262 	}
6263 
6264 	if (slot == -1) {
6265 		/* no slots available */
6266 		mutex_exit(&ixgbe->gen_lock);
6267 		return (ENOSPC);
6268 	}
6269 
6270 	bcopy(mac_addr, ixgbe->unicst_addr[slot].mac.addr, ETHERADDRL);
6271 	(void) ixgbe_set_rar(hw, slot, ixgbe->unicst_addr[slot].mac.addr,
6272 	    rx_group->index, IXGBE_RAH_AV);
6273 	ixgbe->unicst_addr[slot].mac.set = 1;
6274 	ixgbe->unicst_addr[slot].mac.group_index = rx_group->index;
6275 	ixgbe->unicst_avail--;
6276 
6277 	mutex_exit(&ixgbe->gen_lock);
6278 
6279 	return (0);
6280 }
6281 
6282 /*
6283  * Remove a mac address.
6284  */
6285 static int
6286 ixgbe_remmac(void *arg, const uint8_t *mac_addr)
6287 {
6288 	ixgbe_rx_group_t *rx_group = (ixgbe_rx_group_t *)arg;
6289 	ixgbe_t *ixgbe = rx_group->ixgbe;
6290 	struct ixgbe_hw *hw = &ixgbe->hw;
6291 	int slot;
6292 
6293 	mutex_enter(&ixgbe->gen_lock);
6294 
6295 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
6296 		mutex_exit(&ixgbe->gen_lock);
6297 		return (ECANCELED);
6298 	}
6299 
6300 	slot = ixgbe_unicst_find(ixgbe, mac_addr);
6301 	if (slot == -1) {
6302 		mutex_exit(&ixgbe->gen_lock);
6303 		return (EINVAL);
6304 	}
6305 
6306 	if (ixgbe->unicst_addr[slot].mac.set == 0) {
6307 		mutex_exit(&ixgbe->gen_lock);
6308 		return (EINVAL);
6309 	}
6310 
6311 	bzero(ixgbe->unicst_addr[slot].mac.addr, ETHERADDRL);
6312 	(void) ixgbe_clear_rar(hw, slot);
6313 	ixgbe->unicst_addr[slot].mac.set = 0;
6314 	ixgbe->unicst_avail++;
6315 
6316 	mutex_exit(&ixgbe->gen_lock);
6317 
6318 	return (0);
6319 }
6320