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 	if (ixgbe->cb_hdl != NULL) {
850 		(void) ddi_cb_unregister(ixgbe->cb_hdl);
851 	}
852 
853 	/*
854 	 * Remove driver properties
855 	 */
856 	if (ixgbe->attach_progress & ATTACH_PROGRESS_PROPS) {
857 		(void) ddi_prop_remove_all(devinfo);
858 	}
859 
860 	/*
861 	 * Stop the chipset
862 	 */
863 	if (ixgbe->attach_progress & ATTACH_PROGRESS_INIT) {
864 		mutex_enter(&ixgbe->gen_lock);
865 		ixgbe_chip_stop(ixgbe);
866 		mutex_exit(&ixgbe->gen_lock);
867 	}
868 
869 	/*
870 	 * Free register handle
871 	 */
872 	if (ixgbe->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
873 		if (ixgbe->osdep.reg_handle != NULL)
874 			ddi_regs_map_free(&ixgbe->osdep.reg_handle);
875 	}
876 
877 	/*
878 	 * Free PCI config handle
879 	 */
880 	if (ixgbe->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
881 		if (ixgbe->osdep.cfg_handle != NULL)
882 			pci_config_teardown(&ixgbe->osdep.cfg_handle);
883 	}
884 
885 	/*
886 	 * Free locks
887 	 */
888 	if (ixgbe->attach_progress & ATTACH_PROGRESS_LOCKS) {
889 		ixgbe_destroy_locks(ixgbe);
890 	}
891 
892 	/*
893 	 * Free the rx/tx rings
894 	 */
895 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
896 		ixgbe_free_rings(ixgbe);
897 	}
898 
899 	/*
900 	 * Unregister FMA capabilities
901 	 */
902 	if (ixgbe->attach_progress & ATTACH_PROGRESS_FM_INIT) {
903 		ixgbe_fm_fini(ixgbe);
904 	}
905 
906 	/*
907 	 * Free the driver data structure
908 	 */
909 	kmem_free(ixgbe, sizeof (ixgbe_t));
910 
911 	ddi_set_driver_private(devinfo, NULL);
912 }
913 
914 /*
915  * ixgbe_register_mac - Register the driver and its function pointers with
916  * the GLD interface.
917  */
918 static int
919 ixgbe_register_mac(ixgbe_t *ixgbe)
920 {
921 	struct ixgbe_hw *hw = &ixgbe->hw;
922 	mac_register_t *mac;
923 	int status;
924 
925 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
926 		return (IXGBE_FAILURE);
927 
928 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
929 	mac->m_driver = ixgbe;
930 	mac->m_dip = ixgbe->dip;
931 	mac->m_src_addr = hw->mac.addr;
932 	mac->m_callbacks = &ixgbe_m_callbacks;
933 	mac->m_min_sdu = 0;
934 	mac->m_max_sdu = ixgbe->default_mtu;
935 	mac->m_margin = VLAN_TAGSZ;
936 	mac->m_priv_props = ixgbe_priv_props;
937 	mac->m_v12n = MAC_VIRT_LEVEL1;
938 
939 	status = mac_register(mac, &ixgbe->mac_hdl);
940 
941 	mac_free(mac);
942 
943 	return ((status == 0) ? IXGBE_SUCCESS : IXGBE_FAILURE);
944 }
945 
946 /*
947  * ixgbe_identify_hardware - Identify the type of the chipset.
948  */
949 static int
950 ixgbe_identify_hardware(ixgbe_t *ixgbe)
951 {
952 	struct ixgbe_hw *hw = &ixgbe->hw;
953 	struct ixgbe_osdep *osdep = &ixgbe->osdep;
954 
955 	/*
956 	 * Get the device id
957 	 */
958 	hw->vendor_id =
959 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
960 	hw->device_id =
961 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
962 	hw->revision_id =
963 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
964 	hw->subsystem_device_id =
965 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
966 	hw->subsystem_vendor_id =
967 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
968 
969 	/*
970 	 * Set the mac type of the adapter based on the device id
971 	 */
972 	if (ixgbe_set_mac_type(hw) != IXGBE_SUCCESS) {
973 		return (IXGBE_FAILURE);
974 	}
975 
976 	/*
977 	 * Install adapter capabilities
978 	 */
979 	switch (hw->mac.type) {
980 	case ixgbe_mac_82598EB:
981 		IXGBE_DEBUGLOG_0(ixgbe, "identify 82598 adapter\n");
982 		ixgbe->capab = &ixgbe_82598eb_cap;
983 
984 		if (ixgbe_get_media_type(hw) == ixgbe_media_type_copper) {
985 			ixgbe->capab->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
986 			ixgbe->capab->other_intr |= IXGBE_EICR_GPI_SDP1;
987 			ixgbe->capab->other_gpie |= IXGBE_SDP1_GPIEN;
988 		}
989 		break;
990 
991 	case ixgbe_mac_82599EB:
992 		IXGBE_DEBUGLOG_0(ixgbe, "identify 82599 adapter\n");
993 		ixgbe->capab = &ixgbe_82599eb_cap;
994 
995 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM) {
996 			ixgbe->capab->flags |= IXGBE_FLAG_TEMP_SENSOR_CAPABLE;
997 			ixgbe->capab->other_intr |= IXGBE_EICR_GPI_SDP0;
998 			ixgbe->capab->other_gpie |= IXGBE_SDP0_GPIEN;
999 		}
1000 		break;
1001 
1002 	case ixgbe_mac_X540:
1003 		IXGBE_DEBUGLOG_0(ixgbe, "identify X540 adapter\n");
1004 		ixgbe->capab = &ixgbe_X540_cap;
1005 		/*
1006 		 * For now, X540 is all set in its capab structure.
1007 		 * As other X540 variants show up, things can change here.
1008 		 */
1009 		break;
1010 
1011 	case ixgbe_mac_X550:
1012 	case ixgbe_mac_X550EM_x:
1013 		IXGBE_DEBUGLOG_0(ixgbe, "identify X550 adapter\n");
1014 		ixgbe->capab = &ixgbe_X550_cap;
1015 
1016 		if (hw->device_id == IXGBE_DEV_ID_X550EM_X_SFP)
1017 			ixgbe->capab->flags |= IXGBE_FLAG_SFP_PLUG_CAPABLE;
1018 
1019 		/*
1020 		 * Link detection on X552 SFP+ and X552/X557-AT
1021 		 */
1022 		if (hw->device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
1023 		    hw->device_id == IXGBE_DEV_ID_X550EM_X_10G_T) {
1024 			ixgbe->capab->other_intr |=
1025 			    IXGBE_EIMS_GPI_SDP0_BY_MAC(hw);
1026 			ixgbe->capab->other_gpie |= IXGBE_SDP0_GPIEN_X540;
1027 		}
1028 		break;
1029 
1030 	default:
1031 		IXGBE_DEBUGLOG_1(ixgbe,
1032 		    "adapter not supported in ixgbe_identify_hardware(): %d\n",
1033 		    hw->mac.type);
1034 		return (IXGBE_FAILURE);
1035 	}
1036 
1037 	return (IXGBE_SUCCESS);
1038 }
1039 
1040 /*
1041  * ixgbe_regs_map - Map the device registers.
1042  *
1043  */
1044 static int
1045 ixgbe_regs_map(ixgbe_t *ixgbe)
1046 {
1047 	dev_info_t *devinfo = ixgbe->dip;
1048 	struct ixgbe_hw *hw = &ixgbe->hw;
1049 	struct ixgbe_osdep *osdep = &ixgbe->osdep;
1050 	off_t mem_size;
1051 
1052 	/*
1053 	 * First get the size of device registers to be mapped.
1054 	 */
1055 	if (ddi_dev_regsize(devinfo, IXGBE_ADAPTER_REGSET, &mem_size)
1056 	    != DDI_SUCCESS) {
1057 		return (IXGBE_FAILURE);
1058 	}
1059 
1060 	/*
1061 	 * Call ddi_regs_map_setup() to map registers
1062 	 */
1063 	if ((ddi_regs_map_setup(devinfo, IXGBE_ADAPTER_REGSET,
1064 	    (caddr_t *)&hw->hw_addr, 0,
1065 	    mem_size, &ixgbe_regs_acc_attr,
1066 	    &osdep->reg_handle)) != DDI_SUCCESS) {
1067 		return (IXGBE_FAILURE);
1068 	}
1069 
1070 	return (IXGBE_SUCCESS);
1071 }
1072 
1073 /*
1074  * ixgbe_init_properties - Initialize driver properties.
1075  */
1076 static void
1077 ixgbe_init_properties(ixgbe_t *ixgbe)
1078 {
1079 	/*
1080 	 * Get conf file properties, including link settings
1081 	 * jumbo frames, ring number, descriptor number, etc.
1082 	 */
1083 	ixgbe_get_conf(ixgbe);
1084 }
1085 
1086 /*
1087  * ixgbe_init_driver_settings - Initialize driver settings.
1088  *
1089  * The settings include hardware function pointers, bus information,
1090  * rx/tx rings settings, link state, and any other parameters that
1091  * need to be setup during driver initialization.
1092  */
1093 static int
1094 ixgbe_init_driver_settings(ixgbe_t *ixgbe)
1095 {
1096 	struct ixgbe_hw *hw = &ixgbe->hw;
1097 	dev_info_t *devinfo = ixgbe->dip;
1098 	ixgbe_rx_ring_t *rx_ring;
1099 	ixgbe_rx_group_t *rx_group;
1100 	ixgbe_tx_ring_t *tx_ring;
1101 	uint32_t rx_size;
1102 	uint32_t tx_size;
1103 	uint32_t ring_per_group;
1104 	int i;
1105 
1106 	/*
1107 	 * Initialize chipset specific hardware function pointers
1108 	 */
1109 	if (ixgbe_init_shared_code(hw) != IXGBE_SUCCESS) {
1110 		return (IXGBE_FAILURE);
1111 	}
1112 
1113 	/*
1114 	 * Get the system page size
1115 	 */
1116 	ixgbe->sys_page_size = ddi_ptob(devinfo, (ulong_t)1);
1117 
1118 	/*
1119 	 * Set rx buffer size
1120 	 *
1121 	 * The IP header alignment room is counted in the calculation.
1122 	 * The rx buffer size is in unit of 1K that is required by the
1123 	 * chipset hardware.
1124 	 */
1125 	rx_size = ixgbe->max_frame_size + IPHDR_ALIGN_ROOM;
1126 	ixgbe->rx_buf_size = ((rx_size >> 10) +
1127 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
1128 
1129 	/*
1130 	 * Set tx buffer size
1131 	 */
1132 	tx_size = ixgbe->max_frame_size;
1133 	ixgbe->tx_buf_size = ((tx_size >> 10) +
1134 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
1135 
1136 	/*
1137 	 * Initialize rx/tx rings/groups parameters
1138 	 */
1139 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
1140 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1141 		rx_ring = &ixgbe->rx_rings[i];
1142 		rx_ring->index = i;
1143 		rx_ring->ixgbe = ixgbe;
1144 		rx_ring->group_index = i / ring_per_group;
1145 		rx_ring->hw_index = ixgbe_get_hw_rx_index(ixgbe, i);
1146 	}
1147 
1148 	for (i = 0; i < ixgbe->num_rx_groups; i++) {
1149 		rx_group = &ixgbe->rx_groups[i];
1150 		rx_group->index = i;
1151 		rx_group->ixgbe = ixgbe;
1152 	}
1153 
1154 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1155 		tx_ring = &ixgbe->tx_rings[i];
1156 		tx_ring->index = i;
1157 		tx_ring->ixgbe = ixgbe;
1158 		if (ixgbe->tx_head_wb_enable)
1159 			tx_ring->tx_recycle = ixgbe_tx_recycle_head_wb;
1160 		else
1161 			tx_ring->tx_recycle = ixgbe_tx_recycle_legacy;
1162 
1163 		tx_ring->ring_size = ixgbe->tx_ring_size;
1164 		tx_ring->free_list_size = ixgbe->tx_ring_size +
1165 		    (ixgbe->tx_ring_size >> 1);
1166 	}
1167 
1168 	/*
1169 	 * Initialize values of interrupt throttling rate
1170 	 */
1171 	for (i = 1; i < MAX_INTR_VECTOR; i++)
1172 		ixgbe->intr_throttling[i] = ixgbe->intr_throttling[0];
1173 
1174 	/*
1175 	 * The initial link state should be "unknown"
1176 	 */
1177 	ixgbe->link_state = LINK_STATE_UNKNOWN;
1178 
1179 	return (IXGBE_SUCCESS);
1180 }
1181 
1182 /*
1183  * ixgbe_init_locks - Initialize locks.
1184  */
1185 static void
1186 ixgbe_init_locks(ixgbe_t *ixgbe)
1187 {
1188 	ixgbe_rx_ring_t *rx_ring;
1189 	ixgbe_tx_ring_t *tx_ring;
1190 	int i;
1191 
1192 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1193 		rx_ring = &ixgbe->rx_rings[i];
1194 		mutex_init(&rx_ring->rx_lock, NULL,
1195 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1196 	}
1197 
1198 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1199 		tx_ring = &ixgbe->tx_rings[i];
1200 		mutex_init(&tx_ring->tx_lock, NULL,
1201 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1202 		mutex_init(&tx_ring->recycle_lock, NULL,
1203 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1204 		mutex_init(&tx_ring->tcb_head_lock, NULL,
1205 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1206 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
1207 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1208 	}
1209 
1210 	mutex_init(&ixgbe->gen_lock, NULL,
1211 	    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1212 
1213 	mutex_init(&ixgbe->watchdog_lock, NULL,
1214 	    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1215 }
1216 
1217 /*
1218  * ixgbe_destroy_locks - Destroy locks.
1219  */
1220 static void
1221 ixgbe_destroy_locks(ixgbe_t *ixgbe)
1222 {
1223 	ixgbe_rx_ring_t *rx_ring;
1224 	ixgbe_tx_ring_t *tx_ring;
1225 	int i;
1226 
1227 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1228 		rx_ring = &ixgbe->rx_rings[i];
1229 		mutex_destroy(&rx_ring->rx_lock);
1230 	}
1231 
1232 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1233 		tx_ring = &ixgbe->tx_rings[i];
1234 		mutex_destroy(&tx_ring->tx_lock);
1235 		mutex_destroy(&tx_ring->recycle_lock);
1236 		mutex_destroy(&tx_ring->tcb_head_lock);
1237 		mutex_destroy(&tx_ring->tcb_tail_lock);
1238 	}
1239 
1240 	mutex_destroy(&ixgbe->gen_lock);
1241 	mutex_destroy(&ixgbe->watchdog_lock);
1242 }
1243 
1244 static int
1245 ixgbe_resume(dev_info_t *devinfo)
1246 {
1247 	ixgbe_t *ixgbe;
1248 	int i;
1249 
1250 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
1251 	if (ixgbe == NULL)
1252 		return (DDI_FAILURE);
1253 
1254 	mutex_enter(&ixgbe->gen_lock);
1255 
1256 	if (ixgbe->ixgbe_state & IXGBE_STARTED) {
1257 		if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
1258 			mutex_exit(&ixgbe->gen_lock);
1259 			return (DDI_FAILURE);
1260 		}
1261 
1262 		/*
1263 		 * Enable and start the watchdog timer
1264 		 */
1265 		ixgbe_enable_watchdog_timer(ixgbe);
1266 	}
1267 
1268 	atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_SUSPENDED);
1269 
1270 	if (ixgbe->ixgbe_state & IXGBE_STARTED) {
1271 		for (i = 0; i < ixgbe->num_tx_rings; i++) {
1272 			mac_tx_ring_update(ixgbe->mac_hdl,
1273 			    ixgbe->tx_rings[i].ring_handle);
1274 		}
1275 	}
1276 
1277 	mutex_exit(&ixgbe->gen_lock);
1278 
1279 	return (DDI_SUCCESS);
1280 }
1281 
1282 static int
1283 ixgbe_suspend(dev_info_t *devinfo)
1284 {
1285 	ixgbe_t *ixgbe;
1286 
1287 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
1288 	if (ixgbe == NULL)
1289 		return (DDI_FAILURE);
1290 
1291 	mutex_enter(&ixgbe->gen_lock);
1292 
1293 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_SUSPENDED);
1294 	if (!(ixgbe->ixgbe_state & IXGBE_STARTED)) {
1295 		mutex_exit(&ixgbe->gen_lock);
1296 		return (DDI_SUCCESS);
1297 	}
1298 	ixgbe_stop(ixgbe, B_FALSE);
1299 
1300 	mutex_exit(&ixgbe->gen_lock);
1301 
1302 	/*
1303 	 * Disable and stop the watchdog timer
1304 	 */
1305 	ixgbe_disable_watchdog_timer(ixgbe);
1306 
1307 	return (DDI_SUCCESS);
1308 }
1309 
1310 /*
1311  * ixgbe_init - Initialize the device.
1312  */
1313 static int
1314 ixgbe_init(ixgbe_t *ixgbe)
1315 {
1316 	struct ixgbe_hw *hw = &ixgbe->hw;
1317 	u8 pbanum[IXGBE_PBANUM_LENGTH];
1318 	int rv;
1319 
1320 	mutex_enter(&ixgbe->gen_lock);
1321 
1322 	/*
1323 	 * Configure/Initialize hardware
1324 	 */
1325 	rv = ixgbe_init_hw(hw);
1326 	if (rv != IXGBE_SUCCESS) {
1327 		switch (rv) {
1328 
1329 		/*
1330 		 * The first three errors are not prohibitive to us progressing
1331 		 * further, and are maily advisory in nature. In the case of a
1332 		 * SFP module not being present or not deemed supported by the
1333 		 * common code, we adivse the operator of this fact but carry on
1334 		 * instead of failing hard, as SFPs can be inserted or replaced
1335 		 * while the driver is running. In the case of a unknown error,
1336 		 * we fail-hard, logging the reason and emitting a FMA event.
1337 		 */
1338 		case IXGBE_ERR_EEPROM_VERSION:
1339 			ixgbe_error(ixgbe,
1340 			    "This Intel 10Gb Ethernet device is pre-release and"
1341 			    " contains outdated firmware. Please contact your"
1342 			    " hardware vendor for a replacement.");
1343 			break;
1344 		case IXGBE_ERR_SFP_NOT_PRESENT:
1345 			ixgbe_error(ixgbe,
1346 			    "No SFP+ module detected on this interface. Please "
1347 			    "install a supported SFP+ module for this "
1348 			    "interface to become operational.");
1349 			break;
1350 		case IXGBE_ERR_SFP_NOT_SUPPORTED:
1351 			ixgbe_error(ixgbe,
1352 			    "Unsupported SFP+ module detected. Please replace "
1353 			    "it with a supported SFP+ module per Intel "
1354 			    "documentation, or bypass this check with "
1355 			    "allow_unsupported_sfp=1 in ixgbe.conf.");
1356 			break;
1357 		default:
1358 			ixgbe_error(ixgbe,
1359 			    "Failed to initialize hardware. ixgbe_init_hw "
1360 			    "returned %d", rv);
1361 			ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1362 			goto init_fail;
1363 		}
1364 	}
1365 
1366 	/*
1367 	 * Need to init eeprom before validating the checksum.
1368 	 */
1369 	if (ixgbe_init_eeprom_params(hw) < 0) {
1370 		ixgbe_error(ixgbe,
1371 		    "Unable to intitialize the eeprom interface.");
1372 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1373 		goto init_fail;
1374 	}
1375 
1376 	/*
1377 	 * NVM validation
1378 	 */
1379 	if (ixgbe_validate_eeprom_checksum(hw, NULL) < 0) {
1380 		/*
1381 		 * Some PCI-E parts fail the first check due to
1382 		 * the link being in sleep state.  Call it again,
1383 		 * if it fails a second time it's a real issue.
1384 		 */
1385 		if (ixgbe_validate_eeprom_checksum(hw, NULL) < 0) {
1386 			ixgbe_error(ixgbe,
1387 			    "Invalid NVM checksum. Please contact "
1388 			    "the vendor to update the NVM.");
1389 			ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1390 			goto init_fail;
1391 		}
1392 	}
1393 
1394 	/*
1395 	 * Setup default flow control thresholds - enable/disable
1396 	 * & flow control type is controlled by ixgbe.conf
1397 	 */
1398 	hw->fc.high_water[0] = DEFAULT_FCRTH;
1399 	hw->fc.low_water[0] = DEFAULT_FCRTL;
1400 	hw->fc.pause_time = DEFAULT_FCPAUSE;
1401 	hw->fc.send_xon = B_TRUE;
1402 
1403 	/*
1404 	 * Initialize flow control
1405 	 */
1406 	(void) ixgbe_start_hw(hw);
1407 
1408 	/*
1409 	 * Initialize link settings
1410 	 */
1411 	(void) ixgbe_driver_setup_link(ixgbe, B_FALSE);
1412 
1413 	/*
1414 	 * Initialize the chipset hardware
1415 	 */
1416 	if (ixgbe_chip_start(ixgbe) != IXGBE_SUCCESS) {
1417 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1418 		goto init_fail;
1419 	}
1420 
1421 	/*
1422 	 * Read identifying information and place in devinfo.
1423 	 */
1424 	pbanum[0] = '\0';
1425 	(void) ixgbe_read_pba_string(hw, pbanum, sizeof (pbanum));
1426 	if (*pbanum != '\0') {
1427 		(void) ddi_prop_update_string(DDI_DEV_T_NONE, ixgbe->dip,
1428 		    "printed-board-assembly", (char *)pbanum);
1429 	}
1430 
1431 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1432 		goto init_fail;
1433 	}
1434 
1435 	mutex_exit(&ixgbe->gen_lock);
1436 	return (IXGBE_SUCCESS);
1437 
1438 init_fail:
1439 	/*
1440 	 * Reset PHY
1441 	 */
1442 	(void) ixgbe_reset_phy(hw);
1443 
1444 	mutex_exit(&ixgbe->gen_lock);
1445 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1446 	return (IXGBE_FAILURE);
1447 }
1448 
1449 /*
1450  * ixgbe_chip_start - Initialize and start the chipset hardware.
1451  */
1452 static int
1453 ixgbe_chip_start(ixgbe_t *ixgbe)
1454 {
1455 	struct ixgbe_hw *hw = &ixgbe->hw;
1456 	int i;
1457 
1458 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1459 
1460 	/*
1461 	 * Get the mac address
1462 	 * This function should handle SPARC case correctly.
1463 	 */
1464 	if (!ixgbe_find_mac_address(ixgbe)) {
1465 		ixgbe_error(ixgbe, "Failed to get the mac address");
1466 		return (IXGBE_FAILURE);
1467 	}
1468 
1469 	/*
1470 	 * Validate the mac address
1471 	 */
1472 	(void) ixgbe_init_rx_addrs(hw);
1473 	if (!is_valid_mac_addr(hw->mac.addr)) {
1474 		ixgbe_error(ixgbe, "Invalid mac address");
1475 		return (IXGBE_FAILURE);
1476 	}
1477 
1478 	/*
1479 	 * Re-enable relaxed ordering for performance.  It is disabled
1480 	 * by default in the hardware init.
1481 	 */
1482 	if (ixgbe->relax_order_enable == B_TRUE)
1483 		ixgbe_enable_relaxed_ordering(hw);
1484 
1485 	/*
1486 	 * Setup adapter interrupt vectors
1487 	 */
1488 	ixgbe_setup_adapter_vector(ixgbe);
1489 
1490 	/*
1491 	 * Initialize unicast addresses.
1492 	 */
1493 	ixgbe_init_unicst(ixgbe);
1494 
1495 	/*
1496 	 * Setup and initialize the mctable structures.
1497 	 */
1498 	ixgbe_setup_multicst(ixgbe);
1499 
1500 	/*
1501 	 * Set interrupt throttling rate
1502 	 */
1503 	for (i = 0; i < ixgbe->intr_cnt; i++) {
1504 		IXGBE_WRITE_REG(hw, IXGBE_EITR(i), ixgbe->intr_throttling[i]);
1505 	}
1506 
1507 	/*
1508 	 * Disable Wake-on-LAN
1509 	 */
1510 	IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
1511 
1512 	/*
1513 	 * Some adapters offer Energy Efficient Ethernet (EEE) support.
1514 	 * Due to issues with EEE in e1000g/igb, we disable this by default
1515 	 * as a precautionary measure.
1516 	 *
1517 	 * Currently, the only known adapter which supports EEE in the ixgbe
1518 	 * line is 8086,15AB (IXGBE_DEV_ID_X550EM_X_KR), and only after the
1519 	 * first revision of it, as well as any X550 with MAC type 6 (non-EM)
1520 	 */
1521 	(void) ixgbe_setup_eee(hw, B_FALSE);
1522 
1523 	/*
1524 	 * Turn on any present SFP Tx laser
1525 	 */
1526 	ixgbe_enable_tx_laser(hw);
1527 
1528 	/*
1529 	 * Power on the PHY
1530 	 */
1531 	(void) ixgbe_set_phy_power(hw, B_TRUE);
1532 
1533 	/*
1534 	 * Save the state of the PHY
1535 	 */
1536 	ixgbe_get_hw_state(ixgbe);
1537 
1538 	/*
1539 	 * Make sure driver has control
1540 	 */
1541 	ixgbe_get_driver_control(hw);
1542 
1543 	return (IXGBE_SUCCESS);
1544 }
1545 
1546 /*
1547  * ixgbe_chip_stop - Stop the chipset hardware
1548  */
1549 static void
1550 ixgbe_chip_stop(ixgbe_t *ixgbe)
1551 {
1552 	struct ixgbe_hw *hw = &ixgbe->hw;
1553 	int rv;
1554 
1555 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1556 
1557 	/*
1558 	 * Stop interupt generation and disable Tx unit
1559 	 */
1560 	hw->adapter_stopped = B_FALSE;
1561 	(void) ixgbe_stop_adapter(hw);
1562 
1563 	/*
1564 	 * Reset the chipset
1565 	 */
1566 	(void) ixgbe_reset_hw(hw);
1567 
1568 	/*
1569 	 * Reset PHY
1570 	 */
1571 	(void) ixgbe_reset_phy(hw);
1572 
1573 	/*
1574 	 * Enter LPLU (Low Power, Link Up) mode, if available. Avoid resetting
1575 	 * the PHY while doing so. Else, just power down the PHY.
1576 	 */
1577 	if (hw->phy.ops.enter_lplu != NULL) {
1578 		hw->phy.reset_disable = B_TRUE;
1579 		rv = hw->phy.ops.enter_lplu(hw);
1580 		if (rv != IXGBE_SUCCESS)
1581 			ixgbe_error(ixgbe, "Error while entering LPLU: %d", rv);
1582 		hw->phy.reset_disable = B_FALSE;
1583 	} else {
1584 		(void) ixgbe_set_phy_power(hw, B_FALSE);
1585 	}
1586 
1587 	/*
1588 	 * Turn off any present SFP Tx laser
1589 	 * Expected for health and safety reasons
1590 	 */
1591 	ixgbe_disable_tx_laser(hw);
1592 
1593 	/*
1594 	 * Tell firmware driver is no longer in control
1595 	 */
1596 	ixgbe_release_driver_control(hw);
1597 
1598 }
1599 
1600 /*
1601  * ixgbe_reset - Reset the chipset and re-start the driver.
1602  *
1603  * It involves stopping and re-starting the chipset,
1604  * and re-configuring the rx/tx rings.
1605  */
1606 static int
1607 ixgbe_reset(ixgbe_t *ixgbe)
1608 {
1609 	int i;
1610 
1611 	/*
1612 	 * Disable and stop the watchdog timer
1613 	 */
1614 	ixgbe_disable_watchdog_timer(ixgbe);
1615 
1616 	mutex_enter(&ixgbe->gen_lock);
1617 
1618 	ASSERT(ixgbe->ixgbe_state & IXGBE_STARTED);
1619 	atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_STARTED);
1620 
1621 	ixgbe_stop(ixgbe, B_FALSE);
1622 
1623 	if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
1624 		mutex_exit(&ixgbe->gen_lock);
1625 		return (IXGBE_FAILURE);
1626 	}
1627 
1628 	/*
1629 	 * After resetting, need to recheck the link status.
1630 	 */
1631 	ixgbe->link_check_complete = B_FALSE;
1632 	ixgbe->link_check_hrtime = gethrtime() +
1633 	    (IXGBE_LINK_UP_TIME * 100000000ULL);
1634 
1635 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_STARTED);
1636 
1637 	if (!(ixgbe->ixgbe_state & IXGBE_SUSPENDED)) {
1638 		for (i = 0; i < ixgbe->num_tx_rings; i++) {
1639 			mac_tx_ring_update(ixgbe->mac_hdl,
1640 			    ixgbe->tx_rings[i].ring_handle);
1641 		}
1642 	}
1643 
1644 	mutex_exit(&ixgbe->gen_lock);
1645 
1646 	/*
1647 	 * Enable and start the watchdog timer
1648 	 */
1649 	ixgbe_enable_watchdog_timer(ixgbe);
1650 
1651 	return (IXGBE_SUCCESS);
1652 }
1653 
1654 /*
1655  * ixgbe_tx_clean - Clean the pending transmit packets and DMA resources.
1656  */
1657 static void
1658 ixgbe_tx_clean(ixgbe_t *ixgbe)
1659 {
1660 	ixgbe_tx_ring_t *tx_ring;
1661 	tx_control_block_t *tcb;
1662 	link_list_t pending_list;
1663 	uint32_t desc_num;
1664 	int i, j;
1665 
1666 	LINK_LIST_INIT(&pending_list);
1667 
1668 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1669 		tx_ring = &ixgbe->tx_rings[i];
1670 
1671 		mutex_enter(&tx_ring->recycle_lock);
1672 
1673 		/*
1674 		 * Clean the pending tx data - the pending packets in the
1675 		 * work_list that have no chances to be transmitted again.
1676 		 *
1677 		 * We must ensure the chipset is stopped or the link is down
1678 		 * before cleaning the transmit packets.
1679 		 */
1680 		desc_num = 0;
1681 		for (j = 0; j < tx_ring->ring_size; j++) {
1682 			tcb = tx_ring->work_list[j];
1683 			if (tcb != NULL) {
1684 				desc_num += tcb->desc_num;
1685 
1686 				tx_ring->work_list[j] = NULL;
1687 
1688 				ixgbe_free_tcb(tcb);
1689 
1690 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1691 			}
1692 		}
1693 
1694 		if (desc_num > 0) {
1695 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1696 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1697 
1698 			/*
1699 			 * Reset the head and tail pointers of the tbd ring;
1700 			 * Reset the writeback head if it's enable.
1701 			 */
1702 			tx_ring->tbd_head = 0;
1703 			tx_ring->tbd_tail = 0;
1704 			if (ixgbe->tx_head_wb_enable)
1705 				*tx_ring->tbd_head_wb = 0;
1706 
1707 			IXGBE_WRITE_REG(&ixgbe->hw,
1708 			    IXGBE_TDH(tx_ring->index), 0);
1709 			IXGBE_WRITE_REG(&ixgbe->hw,
1710 			    IXGBE_TDT(tx_ring->index), 0);
1711 		}
1712 
1713 		mutex_exit(&tx_ring->recycle_lock);
1714 
1715 		/*
1716 		 * Add the tx control blocks in the pending list to
1717 		 * the free list.
1718 		 */
1719 		ixgbe_put_free_list(tx_ring, &pending_list);
1720 	}
1721 }
1722 
1723 /*
1724  * ixgbe_tx_drain - Drain the tx rings to allow pending packets to be
1725  * transmitted.
1726  */
1727 static boolean_t
1728 ixgbe_tx_drain(ixgbe_t *ixgbe)
1729 {
1730 	ixgbe_tx_ring_t *tx_ring;
1731 	boolean_t done;
1732 	int i, j;
1733 
1734 	/*
1735 	 * Wait for a specific time to allow pending tx packets
1736 	 * to be transmitted.
1737 	 *
1738 	 * Check the counter tbd_free to see if transmission is done.
1739 	 * No lock protection is needed here.
1740 	 *
1741 	 * Return B_TRUE if all pending packets have been transmitted;
1742 	 * Otherwise return B_FALSE;
1743 	 */
1744 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1745 
1746 		done = B_TRUE;
1747 		for (j = 0; j < ixgbe->num_tx_rings; j++) {
1748 			tx_ring = &ixgbe->tx_rings[j];
1749 			done = done &&
1750 			    (tx_ring->tbd_free == tx_ring->ring_size);
1751 		}
1752 
1753 		if (done)
1754 			break;
1755 
1756 		msec_delay(1);
1757 	}
1758 
1759 	return (done);
1760 }
1761 
1762 /*
1763  * ixgbe_rx_drain - Wait for all rx buffers to be released by upper layer.
1764  */
1765 static boolean_t
1766 ixgbe_rx_drain(ixgbe_t *ixgbe)
1767 {
1768 	boolean_t done = B_TRUE;
1769 	int i;
1770 
1771 	/*
1772 	 * Polling the rx free list to check if those rx buffers held by
1773 	 * the upper layer are released.
1774 	 *
1775 	 * Check the counter rcb_free to see if all pending buffers are
1776 	 * released. No lock protection is needed here.
1777 	 *
1778 	 * Return B_TRUE if all pending buffers have been released;
1779 	 * Otherwise return B_FALSE;
1780 	 */
1781 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1782 		done = (ixgbe->rcb_pending == 0);
1783 
1784 		if (done)
1785 			break;
1786 
1787 		msec_delay(1);
1788 	}
1789 
1790 	return (done);
1791 }
1792 
1793 /*
1794  * ixgbe_start - Start the driver/chipset.
1795  */
1796 int
1797 ixgbe_start(ixgbe_t *ixgbe, boolean_t alloc_buffer)
1798 {
1799 	struct ixgbe_hw *hw = &ixgbe->hw;
1800 	int i;
1801 
1802 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1803 
1804 	if (alloc_buffer) {
1805 		if (ixgbe_alloc_rx_data(ixgbe) != IXGBE_SUCCESS) {
1806 			ixgbe_error(ixgbe,
1807 			    "Failed to allocate software receive rings");
1808 			return (IXGBE_FAILURE);
1809 		}
1810 
1811 		/* Allocate buffers for all the rx/tx rings */
1812 		if (ixgbe_alloc_dma(ixgbe) != IXGBE_SUCCESS) {
1813 			ixgbe_error(ixgbe, "Failed to allocate DMA resource");
1814 			return (IXGBE_FAILURE);
1815 		}
1816 
1817 		ixgbe->tx_ring_init = B_TRUE;
1818 	} else {
1819 		ixgbe->tx_ring_init = B_FALSE;
1820 	}
1821 
1822 	for (i = 0; i < ixgbe->num_rx_rings; i++)
1823 		mutex_enter(&ixgbe->rx_rings[i].rx_lock);
1824 	for (i = 0; i < ixgbe->num_tx_rings; i++)
1825 		mutex_enter(&ixgbe->tx_rings[i].tx_lock);
1826 
1827 	/*
1828 	 * Start the chipset hardware
1829 	 */
1830 	if (ixgbe_chip_start(ixgbe) != IXGBE_SUCCESS) {
1831 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1832 		goto start_failure;
1833 	}
1834 
1835 	/*
1836 	 * Configure link now for X550
1837 	 *
1838 	 * X550 possesses a LPLU (Low-Power Link Up) mode which keeps the
1839 	 * resting state of the adapter at a 1Gb FDX speed. Prior to the X550,
1840 	 * the resting state of the link would be the maximum speed that
1841 	 * autonegotiation will allow (usually 10Gb, infrastructure allowing)
1842 	 * so we never bothered with explicitly setting the link to 10Gb as it
1843 	 * would already be at that state on driver attach. With X550, we must
1844 	 * trigger a re-negotiation of the link in order to switch from a LPLU
1845 	 * 1Gb link to 10Gb (cable and link partner permitting.)
1846 	 */
1847 	if (hw->mac.type == ixgbe_mac_X550 ||
1848 	    hw->mac.type == ixgbe_mac_X550EM_x) {
1849 		(void) ixgbe_driver_setup_link(ixgbe, B_TRUE);
1850 		ixgbe_get_hw_state(ixgbe);
1851 	}
1852 
1853 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1854 		goto start_failure;
1855 	}
1856 
1857 	/*
1858 	 * Setup the rx/tx rings
1859 	 */
1860 	ixgbe_setup_rings(ixgbe);
1861 
1862 	/*
1863 	 * ixgbe_start() will be called when resetting, however if reset
1864 	 * happens, we need to clear the ERROR, STALL and OVERTEMP flags
1865 	 * before enabling the interrupts.
1866 	 */
1867 	atomic_and_32(&ixgbe->ixgbe_state, ~(IXGBE_ERROR
1868 	    | IXGBE_STALL| IXGBE_OVERTEMP));
1869 
1870 	/*
1871 	 * Enable adapter interrupts
1872 	 * The interrupts must be enabled after the driver state is START
1873 	 */
1874 	ixgbe_enable_adapter_interrupts(ixgbe);
1875 
1876 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1877 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1878 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1879 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1880 
1881 	return (IXGBE_SUCCESS);
1882 
1883 start_failure:
1884 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1885 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1886 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1887 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1888 
1889 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1890 
1891 	return (IXGBE_FAILURE);
1892 }
1893 
1894 /*
1895  * ixgbe_stop - Stop the driver/chipset.
1896  */
1897 void
1898 ixgbe_stop(ixgbe_t *ixgbe, boolean_t free_buffer)
1899 {
1900 	int i;
1901 
1902 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1903 
1904 	/*
1905 	 * Disable the adapter interrupts
1906 	 */
1907 	ixgbe_disable_adapter_interrupts(ixgbe);
1908 
1909 	/*
1910 	 * Drain the pending tx packets
1911 	 */
1912 	(void) ixgbe_tx_drain(ixgbe);
1913 
1914 	for (i = 0; i < ixgbe->num_rx_rings; i++)
1915 		mutex_enter(&ixgbe->rx_rings[i].rx_lock);
1916 	for (i = 0; i < ixgbe->num_tx_rings; i++)
1917 		mutex_enter(&ixgbe->tx_rings[i].tx_lock);
1918 
1919 	/*
1920 	 * Stop the chipset hardware
1921 	 */
1922 	ixgbe_chip_stop(ixgbe);
1923 
1924 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1925 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1926 	}
1927 
1928 	/*
1929 	 * Clean the pending tx data/resources
1930 	 */
1931 	ixgbe_tx_clean(ixgbe);
1932 
1933 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1934 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1935 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1936 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1937 
1938 	if (ixgbe->link_state == LINK_STATE_UP) {
1939 		ixgbe->link_state = LINK_STATE_UNKNOWN;
1940 		mac_link_update(ixgbe->mac_hdl, ixgbe->link_state);
1941 	}
1942 
1943 	if (free_buffer) {
1944 		/*
1945 		 * Release the DMA/memory resources of rx/tx rings
1946 		 */
1947 		ixgbe_free_dma(ixgbe);
1948 		ixgbe_free_rx_data(ixgbe);
1949 	}
1950 }
1951 
1952 /*
1953  * ixgbe_cbfunc - Driver interface for generic DDI callbacks
1954  */
1955 /* ARGSUSED */
1956 static int
1957 ixgbe_cbfunc(dev_info_t *dip, ddi_cb_action_t cbaction, void *cbarg,
1958     void *arg1, void *arg2)
1959 {
1960 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
1961 
1962 	switch (cbaction) {
1963 	/* IRM callback */
1964 	int count;
1965 	case DDI_CB_INTR_ADD:
1966 	case DDI_CB_INTR_REMOVE:
1967 		count = (int)(uintptr_t)cbarg;
1968 		ASSERT(ixgbe->intr_type == DDI_INTR_TYPE_MSIX);
1969 		DTRACE_PROBE2(ixgbe__irm__callback, int, count,
1970 		    int, ixgbe->intr_cnt);
1971 		if (ixgbe_intr_adjust(ixgbe, cbaction, count) !=
1972 		    DDI_SUCCESS) {
1973 			ixgbe_error(ixgbe,
1974 			    "IRM CB: Failed to adjust interrupts");
1975 			goto cb_fail;
1976 		}
1977 		break;
1978 	default:
1979 		IXGBE_DEBUGLOG_1(ixgbe, "DDI CB: action 0x%x NOT supported",
1980 		    cbaction);
1981 		return (DDI_ENOTSUP);
1982 	}
1983 	return (DDI_SUCCESS);
1984 cb_fail:
1985 	return (DDI_FAILURE);
1986 }
1987 
1988 /*
1989  * ixgbe_intr_adjust - Adjust interrupt to respond to IRM request.
1990  */
1991 static int
1992 ixgbe_intr_adjust(ixgbe_t *ixgbe, ddi_cb_action_t cbaction, int count)
1993 {
1994 	int i, rc, actual;
1995 
1996 	if (count == 0)
1997 		return (DDI_SUCCESS);
1998 
1999 	if ((cbaction == DDI_CB_INTR_ADD &&
2000 	    ixgbe->intr_cnt + count > ixgbe->intr_cnt_max) ||
2001 	    (cbaction == DDI_CB_INTR_REMOVE &&
2002 	    ixgbe->intr_cnt - count < ixgbe->intr_cnt_min))
2003 		return (DDI_FAILURE);
2004 
2005 	if (!(ixgbe->ixgbe_state & IXGBE_STARTED)) {
2006 		return (DDI_FAILURE);
2007 	}
2008 
2009 	for (i = 0; i < ixgbe->num_rx_rings; i++)
2010 		mac_ring_intr_set(ixgbe->rx_rings[i].ring_handle, NULL);
2011 	for (i = 0; i < ixgbe->num_tx_rings; i++)
2012 		mac_ring_intr_set(ixgbe->tx_rings[i].ring_handle, NULL);
2013 
2014 	mutex_enter(&ixgbe->gen_lock);
2015 	ixgbe->ixgbe_state &= ~IXGBE_STARTED;
2016 	ixgbe->ixgbe_state |= IXGBE_INTR_ADJUST;
2017 	ixgbe->ixgbe_state |= IXGBE_SUSPENDED;
2018 	mac_link_update(ixgbe->mac_hdl, LINK_STATE_UNKNOWN);
2019 
2020 	ixgbe_stop(ixgbe, B_FALSE);
2021 	/*
2022 	 * Disable interrupts
2023 	 */
2024 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
2025 		rc = ixgbe_disable_intrs(ixgbe);
2026 		ASSERT(rc == IXGBE_SUCCESS);
2027 	}
2028 	ixgbe->attach_progress &= ~ATTACH_PROGRESS_ENABLE_INTR;
2029 
2030 	/*
2031 	 * Remove interrupt handlers
2032 	 */
2033 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
2034 		ixgbe_rem_intr_handlers(ixgbe);
2035 	}
2036 	ixgbe->attach_progress &= ~ATTACH_PROGRESS_ADD_INTR;
2037 
2038 	/*
2039 	 * Clear vect_map
2040 	 */
2041 	bzero(&ixgbe->vect_map, sizeof (ixgbe->vect_map));
2042 	switch (cbaction) {
2043 	case DDI_CB_INTR_ADD:
2044 		rc = ddi_intr_alloc(ixgbe->dip, ixgbe->htable,
2045 		    DDI_INTR_TYPE_MSIX, ixgbe->intr_cnt, count, &actual,
2046 		    DDI_INTR_ALLOC_NORMAL);
2047 		if (rc != DDI_SUCCESS || actual != count) {
2048 			ixgbe_log(ixgbe, "Adjust interrupts failed."
2049 			    "return: %d, irm cb size: %d, actual: %d",
2050 			    rc, count, actual);
2051 			goto intr_adjust_fail;
2052 		}
2053 		ixgbe->intr_cnt += count;
2054 		break;
2055 
2056 	case DDI_CB_INTR_REMOVE:
2057 		for (i = ixgbe->intr_cnt - count;
2058 		    i < ixgbe->intr_cnt; i ++) {
2059 			rc = ddi_intr_free(ixgbe->htable[i]);
2060 			ixgbe->htable[i] = NULL;
2061 			if (rc != DDI_SUCCESS) {
2062 				ixgbe_log(ixgbe, "Adjust interrupts failed."
2063 				    "return: %d, irm cb size: %d, actual: %d",
2064 				    rc, count, actual);
2065 				goto intr_adjust_fail;
2066 			}
2067 		}
2068 		ixgbe->intr_cnt -= count;
2069 		break;
2070 	}
2071 
2072 	/*
2073 	 * Get priority for first vector, assume remaining are all the same
2074 	 */
2075 	rc = ddi_intr_get_pri(ixgbe->htable[0], &ixgbe->intr_pri);
2076 	if (rc != DDI_SUCCESS) {
2077 		ixgbe_log(ixgbe,
2078 		    "Get interrupt priority failed: %d", rc);
2079 		goto intr_adjust_fail;
2080 	}
2081 	rc = ddi_intr_get_cap(ixgbe->htable[0], &ixgbe->intr_cap);
2082 	if (rc != DDI_SUCCESS) {
2083 		ixgbe_log(ixgbe, "Get interrupt cap failed: %d", rc);
2084 		goto intr_adjust_fail;
2085 	}
2086 	ixgbe->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
2087 
2088 	/*
2089 	 * Map rings to interrupt vectors
2090 	 */
2091 	if (ixgbe_map_intrs_to_vectors(ixgbe) != IXGBE_SUCCESS) {
2092 		ixgbe_error(ixgbe,
2093 		    "IRM CB: Failed to map interrupts to vectors");
2094 		goto intr_adjust_fail;
2095 	}
2096 
2097 	/*
2098 	 * Add interrupt handlers
2099 	 */
2100 	if (ixgbe_add_intr_handlers(ixgbe) != IXGBE_SUCCESS) {
2101 		ixgbe_error(ixgbe, "IRM CB: Failed to add interrupt handlers");
2102 		goto intr_adjust_fail;
2103 	}
2104 	ixgbe->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
2105 
2106 	/*
2107 	 * Now that mutex locks are initialized, and the chip is also
2108 	 * initialized, enable interrupts.
2109 	 */
2110 	if (ixgbe_enable_intrs(ixgbe) != IXGBE_SUCCESS) {
2111 		ixgbe_error(ixgbe, "IRM CB: Failed to enable DDI interrupts");
2112 		goto intr_adjust_fail;
2113 	}
2114 	ixgbe->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
2115 	if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
2116 		ixgbe_error(ixgbe, "IRM CB: Failed to start");
2117 		goto intr_adjust_fail;
2118 	}
2119 	ixgbe->ixgbe_state &= ~IXGBE_INTR_ADJUST;
2120 	ixgbe->ixgbe_state &= ~IXGBE_SUSPENDED;
2121 	ixgbe->ixgbe_state |= IXGBE_STARTED;
2122 	mutex_exit(&ixgbe->gen_lock);
2123 
2124 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2125 		mac_ring_intr_set(ixgbe->rx_rings[i].ring_handle,
2126 		    ixgbe->htable[ixgbe->rx_rings[i].intr_vector]);
2127 	}
2128 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2129 		mac_ring_intr_set(ixgbe->tx_rings[i].ring_handle,
2130 		    ixgbe->htable[ixgbe->tx_rings[i].intr_vector]);
2131 	}
2132 
2133 	/* Wakeup all Tx rings */
2134 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2135 		mac_tx_ring_update(ixgbe->mac_hdl,
2136 		    ixgbe->tx_rings[i].ring_handle);
2137 	}
2138 
2139 	IXGBE_DEBUGLOG_3(ixgbe,
2140 	    "IRM CB: interrupts new value: 0x%x(0x%x:0x%x).",
2141 	    ixgbe->intr_cnt, ixgbe->intr_cnt_min, ixgbe->intr_cnt_max);
2142 	return (DDI_SUCCESS);
2143 
2144 intr_adjust_fail:
2145 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
2146 	mutex_exit(&ixgbe->gen_lock);
2147 	return (DDI_FAILURE);
2148 }
2149 
2150 /*
2151  * ixgbe_intr_cb_register - Register interrupt callback function.
2152  */
2153 static int
2154 ixgbe_intr_cb_register(ixgbe_t *ixgbe)
2155 {
2156 	if (ddi_cb_register(ixgbe->dip, DDI_CB_FLAG_INTR, ixgbe_cbfunc,
2157 	    ixgbe, NULL, &ixgbe->cb_hdl) != DDI_SUCCESS) {
2158 		return (IXGBE_FAILURE);
2159 	}
2160 	IXGBE_DEBUGLOG_0(ixgbe, "Interrupt callback function registered.");
2161 	return (IXGBE_SUCCESS);
2162 }
2163 
2164 /*
2165  * ixgbe_alloc_rings - Allocate memory space for rx/tx rings.
2166  */
2167 static int
2168 ixgbe_alloc_rings(ixgbe_t *ixgbe)
2169 {
2170 	/*
2171 	 * Allocate memory space for rx rings
2172 	 */
2173 	ixgbe->rx_rings = kmem_zalloc(
2174 	    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings,
2175 	    KM_NOSLEEP);
2176 
2177 	if (ixgbe->rx_rings == NULL) {
2178 		return (IXGBE_FAILURE);
2179 	}
2180 
2181 	/*
2182 	 * Allocate memory space for tx rings
2183 	 */
2184 	ixgbe->tx_rings = kmem_zalloc(
2185 	    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings,
2186 	    KM_NOSLEEP);
2187 
2188 	if (ixgbe->tx_rings == NULL) {
2189 		kmem_free(ixgbe->rx_rings,
2190 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
2191 		ixgbe->rx_rings = NULL;
2192 		return (IXGBE_FAILURE);
2193 	}
2194 
2195 	/*
2196 	 * Allocate memory space for rx ring groups
2197 	 */
2198 	ixgbe->rx_groups = kmem_zalloc(
2199 	    sizeof (ixgbe_rx_group_t) * ixgbe->num_rx_groups,
2200 	    KM_NOSLEEP);
2201 
2202 	if (ixgbe->rx_groups == NULL) {
2203 		kmem_free(ixgbe->rx_rings,
2204 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
2205 		kmem_free(ixgbe->tx_rings,
2206 		    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings);
2207 		ixgbe->rx_rings = NULL;
2208 		ixgbe->tx_rings = NULL;
2209 		return (IXGBE_FAILURE);
2210 	}
2211 
2212 	return (IXGBE_SUCCESS);
2213 }
2214 
2215 /*
2216  * ixgbe_free_rings - Free the memory space of rx/tx rings.
2217  */
2218 static void
2219 ixgbe_free_rings(ixgbe_t *ixgbe)
2220 {
2221 	if (ixgbe->rx_rings != NULL) {
2222 		kmem_free(ixgbe->rx_rings,
2223 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
2224 		ixgbe->rx_rings = NULL;
2225 	}
2226 
2227 	if (ixgbe->tx_rings != NULL) {
2228 		kmem_free(ixgbe->tx_rings,
2229 		    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings);
2230 		ixgbe->tx_rings = NULL;
2231 	}
2232 
2233 	if (ixgbe->rx_groups != NULL) {
2234 		kmem_free(ixgbe->rx_groups,
2235 		    sizeof (ixgbe_rx_group_t) * ixgbe->num_rx_groups);
2236 		ixgbe->rx_groups = NULL;
2237 	}
2238 }
2239 
2240 static int
2241 ixgbe_alloc_rx_data(ixgbe_t *ixgbe)
2242 {
2243 	ixgbe_rx_ring_t *rx_ring;
2244 	int i;
2245 
2246 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2247 		rx_ring = &ixgbe->rx_rings[i];
2248 		if (ixgbe_alloc_rx_ring_data(rx_ring) != IXGBE_SUCCESS)
2249 			goto alloc_rx_rings_failure;
2250 	}
2251 	return (IXGBE_SUCCESS);
2252 
2253 alloc_rx_rings_failure:
2254 	ixgbe_free_rx_data(ixgbe);
2255 	return (IXGBE_FAILURE);
2256 }
2257 
2258 static void
2259 ixgbe_free_rx_data(ixgbe_t *ixgbe)
2260 {
2261 	ixgbe_rx_ring_t *rx_ring;
2262 	ixgbe_rx_data_t *rx_data;
2263 	int i;
2264 
2265 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2266 		rx_ring = &ixgbe->rx_rings[i];
2267 
2268 		mutex_enter(&ixgbe->rx_pending_lock);
2269 		rx_data = rx_ring->rx_data;
2270 
2271 		if (rx_data != NULL) {
2272 			rx_data->flag |= IXGBE_RX_STOPPED;
2273 
2274 			if (rx_data->rcb_pending == 0) {
2275 				ixgbe_free_rx_ring_data(rx_data);
2276 				rx_ring->rx_data = NULL;
2277 			}
2278 		}
2279 
2280 		mutex_exit(&ixgbe->rx_pending_lock);
2281 	}
2282 }
2283 
2284 /*
2285  * ixgbe_setup_rings - Setup rx/tx rings.
2286  */
2287 static void
2288 ixgbe_setup_rings(ixgbe_t *ixgbe)
2289 {
2290 	/*
2291 	 * Setup the rx/tx rings, including the following:
2292 	 *
2293 	 * 1. Setup the descriptor ring and the control block buffers;
2294 	 * 2. Initialize necessary registers for receive/transmit;
2295 	 * 3. Initialize software pointers/parameters for receive/transmit;
2296 	 */
2297 	ixgbe_setup_rx(ixgbe);
2298 
2299 	ixgbe_setup_tx(ixgbe);
2300 }
2301 
2302 static void
2303 ixgbe_setup_rx_ring(ixgbe_rx_ring_t *rx_ring)
2304 {
2305 	ixgbe_t *ixgbe = rx_ring->ixgbe;
2306 	ixgbe_rx_data_t *rx_data = rx_ring->rx_data;
2307 	struct ixgbe_hw *hw = &ixgbe->hw;
2308 	rx_control_block_t *rcb;
2309 	union ixgbe_adv_rx_desc	*rbd;
2310 	uint32_t size;
2311 	uint32_t buf_low;
2312 	uint32_t buf_high;
2313 	uint32_t reg_val;
2314 	int i;
2315 
2316 	ASSERT(mutex_owned(&rx_ring->rx_lock));
2317 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2318 
2319 	for (i = 0; i < ixgbe->rx_ring_size; i++) {
2320 		rcb = rx_data->work_list[i];
2321 		rbd = &rx_data->rbd_ring[i];
2322 
2323 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
2324 		rbd->read.hdr_addr = NULL;
2325 	}
2326 
2327 	/*
2328 	 * Initialize the length register
2329 	 */
2330 	size = rx_data->ring_size * sizeof (union ixgbe_adv_rx_desc);
2331 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(rx_ring->hw_index), size);
2332 
2333 	/*
2334 	 * Initialize the base address registers
2335 	 */
2336 	buf_low = (uint32_t)rx_data->rbd_area.dma_address;
2337 	buf_high = (uint32_t)(rx_data->rbd_area.dma_address >> 32);
2338 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(rx_ring->hw_index), buf_high);
2339 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(rx_ring->hw_index), buf_low);
2340 
2341 	/*
2342 	 * Setup head & tail pointers
2343 	 */
2344 	IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->hw_index),
2345 	    rx_data->ring_size - 1);
2346 	IXGBE_WRITE_REG(hw, IXGBE_RDH(rx_ring->hw_index), 0);
2347 
2348 	rx_data->rbd_next = 0;
2349 	rx_data->lro_first = 0;
2350 
2351 	/*
2352 	 * Setup the Receive Descriptor Control Register (RXDCTL)
2353 	 * PTHRESH=32 descriptors (half the internal cache)
2354 	 * HTHRESH=0 descriptors (to minimize latency on fetch)
2355 	 * WTHRESH defaults to 1 (writeback each descriptor)
2356 	 */
2357 	reg_val = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rx_ring->hw_index));
2358 	reg_val |= IXGBE_RXDCTL_ENABLE;	/* enable queue */
2359 
2360 	/* Not a valid value for 82599, X540 or X550 */
2361 	if (hw->mac.type == ixgbe_mac_82598EB) {
2362 		reg_val |= 0x0020;	/* pthresh */
2363 	}
2364 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rx_ring->hw_index), reg_val);
2365 
2366 	if (hw->mac.type == ixgbe_mac_82599EB ||
2367 	    hw->mac.type == ixgbe_mac_X540 ||
2368 	    hw->mac.type == ixgbe_mac_X550 ||
2369 	    hw->mac.type == ixgbe_mac_X550EM_x) {
2370 		reg_val = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2371 		reg_val |= (IXGBE_RDRXCTL_CRCSTRIP | IXGBE_RDRXCTL_AGGDIS);
2372 		IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, reg_val);
2373 	}
2374 
2375 	/*
2376 	 * Setup the Split and Replication Receive Control Register.
2377 	 * Set the rx buffer size and the advanced descriptor type.
2378 	 */
2379 	reg_val = (ixgbe->rx_buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) |
2380 	    IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2381 	reg_val |= IXGBE_SRRCTL_DROP_EN;
2382 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rx_ring->hw_index), reg_val);
2383 }
2384 
2385 static void
2386 ixgbe_setup_rx(ixgbe_t *ixgbe)
2387 {
2388 	ixgbe_rx_ring_t *rx_ring;
2389 	struct ixgbe_hw *hw = &ixgbe->hw;
2390 	uint32_t reg_val;
2391 	uint32_t ring_mapping;
2392 	uint32_t i, index;
2393 	uint32_t psrtype_rss_bit;
2394 
2395 	/*
2396 	 * Ensure that Rx is disabled while setting up
2397 	 * the Rx unit and Rx descriptor ring(s)
2398 	 */
2399 	ixgbe_disable_rx(hw);
2400 
2401 	/* PSRTYPE must be configured for 82599 */
2402 	if (ixgbe->classify_mode != IXGBE_CLASSIFY_VMDQ &&
2403 	    ixgbe->classify_mode != IXGBE_CLASSIFY_VMDQ_RSS) {
2404 		reg_val = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
2405 		    IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR;
2406 		reg_val |= IXGBE_PSRTYPE_L2HDR;
2407 		reg_val |= 0x80000000;
2408 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), reg_val);
2409 	} else {
2410 		if (ixgbe->num_rx_groups > 32) {
2411 			psrtype_rss_bit = 0x20000000;
2412 		} else {
2413 			psrtype_rss_bit = 0x40000000;
2414 		}
2415 		for (i = 0; i < ixgbe->capab->max_rx_grp_num; i++) {
2416 			reg_val = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
2417 			    IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR;
2418 			reg_val |= IXGBE_PSRTYPE_L2HDR;
2419 			reg_val |= psrtype_rss_bit;
2420 			IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(i), reg_val);
2421 		}
2422 	}
2423 
2424 	/*
2425 	 * Set filter control in FCTRL to determine types of packets are passed
2426 	 * up to the driver.
2427 	 * - Pass broadcast packets.
2428 	 * - Do not pass flow control pause frames (82598-specific)
2429 	 */
2430 	reg_val = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2431 	reg_val |= IXGBE_FCTRL_BAM; /* Broadcast Accept Mode */
2432 	if (hw->mac.type == ixgbe_mac_82598EB) {
2433 		reg_val |= IXGBE_FCTRL_DPF; /* Discard Pause Frames */
2434 	}
2435 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, reg_val);
2436 
2437 	/*
2438 	 * Hardware checksum settings
2439 	 */
2440 	if (ixgbe->rx_hcksum_enable) {
2441 		reg_val = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2442 		reg_val |= IXGBE_RXCSUM_IPPCSE;	/* IP checksum */
2443 		IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, reg_val);
2444 	}
2445 
2446 	/*
2447 	 * Setup VMDq and RSS for multiple receive queues
2448 	 */
2449 	switch (ixgbe->classify_mode) {
2450 	case IXGBE_CLASSIFY_RSS:
2451 		/*
2452 		 * One group, only RSS is needed when more than
2453 		 * one ring enabled.
2454 		 */
2455 		ixgbe_setup_rss(ixgbe);
2456 		break;
2457 
2458 	case IXGBE_CLASSIFY_VMDQ:
2459 		/*
2460 		 * Multiple groups, each group has one ring,
2461 		 * only VMDq is needed.
2462 		 */
2463 		ixgbe_setup_vmdq(ixgbe);
2464 		break;
2465 
2466 	case IXGBE_CLASSIFY_VMDQ_RSS:
2467 		/*
2468 		 * Multiple groups and multiple rings, both
2469 		 * VMDq and RSS are needed.
2470 		 */
2471 		ixgbe_setup_vmdq_rss(ixgbe);
2472 		break;
2473 
2474 	default:
2475 		break;
2476 	}
2477 
2478 	/*
2479 	 * Enable the receive unit.  This must be done after filter
2480 	 * control is set in FCTRL. On 82598, we disable the descriptor monitor.
2481 	 * 82598 is the only adapter which defines this RXCTRL option.
2482 	 */
2483 	reg_val = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2484 	if (hw->mac.type == ixgbe_mac_82598EB)
2485 		reg_val |= IXGBE_RXCTRL_DMBYPS; /* descriptor monitor bypass */
2486 	reg_val |= IXGBE_RXCTRL_RXEN;
2487 	(void) ixgbe_enable_rx_dma(hw, reg_val);
2488 
2489 	/*
2490 	 * ixgbe_setup_rx_ring must be called after configuring RXCTRL
2491 	 */
2492 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2493 		rx_ring = &ixgbe->rx_rings[i];
2494 		ixgbe_setup_rx_ring(rx_ring);
2495 	}
2496 
2497 	/*
2498 	 * Setup the per-ring statistics mapping.
2499 	 */
2500 	ring_mapping = 0;
2501 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2502 		index = ixgbe->rx_rings[i].hw_index;
2503 		ring_mapping = IXGBE_READ_REG(hw, IXGBE_RQSMR(index >> 2));
2504 		ring_mapping |= (i & 0xF) << (8 * (index & 0x3));
2505 		IXGBE_WRITE_REG(hw, IXGBE_RQSMR(index >> 2), ring_mapping);
2506 	}
2507 
2508 	/*
2509 	 * The Max Frame Size in MHADD/MAXFRS will be internally increased
2510 	 * by four bytes if the packet has a VLAN field, so includes MTU,
2511 	 * ethernet header and frame check sequence.
2512 	 * Register is MAXFRS in 82599.
2513 	 */
2514 	reg_val = IXGBE_READ_REG(hw, IXGBE_MHADD);
2515 	reg_val &= ~IXGBE_MHADD_MFS_MASK;
2516 	reg_val |= (ixgbe->default_mtu + sizeof (struct ether_header)
2517 	    + ETHERFCSL) << IXGBE_MHADD_MFS_SHIFT;
2518 	IXGBE_WRITE_REG(hw, IXGBE_MHADD, reg_val);
2519 
2520 	/*
2521 	 * Setup Jumbo Frame enable bit
2522 	 */
2523 	reg_val = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2524 	if (ixgbe->default_mtu > ETHERMTU)
2525 		reg_val |= IXGBE_HLREG0_JUMBOEN;
2526 	else
2527 		reg_val &= ~IXGBE_HLREG0_JUMBOEN;
2528 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, reg_val);
2529 
2530 	/*
2531 	 * Setup RSC for multiple receive queues.
2532 	 */
2533 	if (ixgbe->lro_enable) {
2534 		for (i = 0; i < ixgbe->num_rx_rings; i++) {
2535 			/*
2536 			 * Make sure rx_buf_size * MAXDESC not greater
2537 			 * than 65535.
2538 			 * Intel recommends 4 for MAXDESC field value.
2539 			 */
2540 			reg_val = IXGBE_READ_REG(hw, IXGBE_RSCCTL(i));
2541 			reg_val |= IXGBE_RSCCTL_RSCEN;
2542 			if (ixgbe->rx_buf_size == IXGBE_PKG_BUF_16k)
2543 				reg_val |= IXGBE_RSCCTL_MAXDESC_1;
2544 			else
2545 				reg_val |= IXGBE_RSCCTL_MAXDESC_4;
2546 			IXGBE_WRITE_REG(hw,  IXGBE_RSCCTL(i), reg_val);
2547 		}
2548 
2549 		reg_val = IXGBE_READ_REG(hw, IXGBE_RSCDBU);
2550 		reg_val |= IXGBE_RSCDBU_RSCACKDIS;
2551 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU, reg_val);
2552 
2553 		reg_val = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2554 		reg_val |= IXGBE_RDRXCTL_RSCACKC;
2555 		reg_val |= IXGBE_RDRXCTL_FCOE_WRFIX;
2556 		reg_val &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2557 
2558 		IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, reg_val);
2559 	}
2560 }
2561 
2562 static void
2563 ixgbe_setup_tx_ring(ixgbe_tx_ring_t *tx_ring)
2564 {
2565 	ixgbe_t *ixgbe = tx_ring->ixgbe;
2566 	struct ixgbe_hw *hw = &ixgbe->hw;
2567 	uint32_t size;
2568 	uint32_t buf_low;
2569 	uint32_t buf_high;
2570 	uint32_t reg_val;
2571 
2572 	ASSERT(mutex_owned(&tx_ring->tx_lock));
2573 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2574 
2575 	/*
2576 	 * Initialize the length register
2577 	 */
2578 	size = tx_ring->ring_size * sizeof (union ixgbe_adv_tx_desc);
2579 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(tx_ring->index), size);
2580 
2581 	/*
2582 	 * Initialize the base address registers
2583 	 */
2584 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
2585 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
2586 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(tx_ring->index), buf_low);
2587 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(tx_ring->index), buf_high);
2588 
2589 	/*
2590 	 * Setup head & tail pointers
2591 	 */
2592 	IXGBE_WRITE_REG(hw, IXGBE_TDH(tx_ring->index), 0);
2593 	IXGBE_WRITE_REG(hw, IXGBE_TDT(tx_ring->index), 0);
2594 
2595 	/*
2596 	 * Setup head write-back
2597 	 */
2598 	if (ixgbe->tx_head_wb_enable) {
2599 		/*
2600 		 * The memory of the head write-back is allocated using
2601 		 * the extra tbd beyond the tail of the tbd ring.
2602 		 */
2603 		tx_ring->tbd_head_wb = (uint32_t *)
2604 		    ((uintptr_t)tx_ring->tbd_area.address + size);
2605 		*tx_ring->tbd_head_wb = 0;
2606 
2607 		buf_low = (uint32_t)
2608 		    (tx_ring->tbd_area.dma_address + size);
2609 		buf_high = (uint32_t)
2610 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
2611 
2612 		/* Set the head write-back enable bit */
2613 		buf_low |= IXGBE_TDWBAL_HEAD_WB_ENABLE;
2614 
2615 		IXGBE_WRITE_REG(hw, IXGBE_TDWBAL(tx_ring->index), buf_low);
2616 		IXGBE_WRITE_REG(hw, IXGBE_TDWBAH(tx_ring->index), buf_high);
2617 
2618 		/*
2619 		 * Turn off relaxed ordering for head write back or it will
2620 		 * cause problems with the tx recycling
2621 		 */
2622 
2623 		reg_val = (hw->mac.type == ixgbe_mac_82598EB) ?
2624 		    IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(tx_ring->index)) :
2625 		    IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(tx_ring->index));
2626 		reg_val &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
2627 		if (hw->mac.type == ixgbe_mac_82598EB) {
2628 			IXGBE_WRITE_REG(hw,
2629 			    IXGBE_DCA_TXCTRL(tx_ring->index), reg_val);
2630 		} else {
2631 			IXGBE_WRITE_REG(hw,
2632 			    IXGBE_DCA_TXCTRL_82599(tx_ring->index), reg_val);
2633 		}
2634 	} else {
2635 		tx_ring->tbd_head_wb = NULL;
2636 	}
2637 
2638 	tx_ring->tbd_head = 0;
2639 	tx_ring->tbd_tail = 0;
2640 	tx_ring->tbd_free = tx_ring->ring_size;
2641 
2642 	if (ixgbe->tx_ring_init == B_TRUE) {
2643 		tx_ring->tcb_head = 0;
2644 		tx_ring->tcb_tail = 0;
2645 		tx_ring->tcb_free = tx_ring->free_list_size;
2646 	}
2647 
2648 	/*
2649 	 * Initialize the s/w context structure
2650 	 */
2651 	bzero(&tx_ring->tx_context, sizeof (ixgbe_tx_context_t));
2652 }
2653 
2654 static void
2655 ixgbe_setup_tx(ixgbe_t *ixgbe)
2656 {
2657 	struct ixgbe_hw *hw = &ixgbe->hw;
2658 	ixgbe_tx_ring_t *tx_ring;
2659 	uint32_t reg_val;
2660 	uint32_t ring_mapping;
2661 	int i;
2662 
2663 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2664 		tx_ring = &ixgbe->tx_rings[i];
2665 		ixgbe_setup_tx_ring(tx_ring);
2666 	}
2667 
2668 	/*
2669 	 * Setup the per-ring statistics mapping.
2670 	 */
2671 	ring_mapping = 0;
2672 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2673 		ring_mapping |= (i & 0xF) << (8 * (i & 0x3));
2674 		if ((i & 0x3) == 0x3) {
2675 			switch (hw->mac.type) {
2676 			case ixgbe_mac_82598EB:
2677 				IXGBE_WRITE_REG(hw, IXGBE_TQSMR(i >> 2),
2678 				    ring_mapping);
2679 				break;
2680 
2681 			case ixgbe_mac_82599EB:
2682 			case ixgbe_mac_X540:
2683 			case ixgbe_mac_X550:
2684 			case ixgbe_mac_X550EM_x:
2685 				IXGBE_WRITE_REG(hw, IXGBE_TQSM(i >> 2),
2686 				    ring_mapping);
2687 				break;
2688 
2689 			default:
2690 				break;
2691 			}
2692 
2693 			ring_mapping = 0;
2694 		}
2695 	}
2696 	if (i & 0x3) {
2697 		switch (hw->mac.type) {
2698 		case ixgbe_mac_82598EB:
2699 			IXGBE_WRITE_REG(hw, IXGBE_TQSMR(i >> 2), ring_mapping);
2700 			break;
2701 
2702 		case ixgbe_mac_82599EB:
2703 		case ixgbe_mac_X540:
2704 		case ixgbe_mac_X550:
2705 		case ixgbe_mac_X550EM_x:
2706 			IXGBE_WRITE_REG(hw, IXGBE_TQSM(i >> 2), ring_mapping);
2707 			break;
2708 
2709 		default:
2710 			break;
2711 		}
2712 	}
2713 
2714 	/*
2715 	 * Enable CRC appending and TX padding (for short tx frames)
2716 	 */
2717 	reg_val = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2718 	reg_val |= IXGBE_HLREG0_TXCRCEN | IXGBE_HLREG0_TXPADEN;
2719 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, reg_val);
2720 
2721 	/*
2722 	 * enable DMA for 82599, X540 and X550 parts
2723 	 */
2724 	if (hw->mac.type == ixgbe_mac_82599EB ||
2725 	    hw->mac.type == ixgbe_mac_X540 ||
2726 	    hw->mac.type == ixgbe_mac_X550 ||
2727 	    hw->mac.type == ixgbe_mac_X550EM_x) {
2728 		/* DMATXCTL.TE must be set after all Tx config is complete */
2729 		reg_val = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2730 		reg_val |= IXGBE_DMATXCTL_TE;
2731 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, reg_val);
2732 
2733 		/* Disable arbiter to set MTQC */
2734 		reg_val = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2735 		reg_val |= IXGBE_RTTDCS_ARBDIS;
2736 		IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg_val);
2737 		IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2738 		reg_val &= ~IXGBE_RTTDCS_ARBDIS;
2739 		IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg_val);
2740 	}
2741 
2742 	/*
2743 	 * Enabling tx queues ..
2744 	 * For 82599 must be done after DMATXCTL.TE is set
2745 	 */
2746 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2747 		tx_ring = &ixgbe->tx_rings[i];
2748 		reg_val = IXGBE_READ_REG(hw, IXGBE_TXDCTL(tx_ring->index));
2749 		reg_val |= IXGBE_TXDCTL_ENABLE;
2750 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(tx_ring->index), reg_val);
2751 	}
2752 }
2753 
2754 /*
2755  * ixgbe_setup_rss - Setup receive-side scaling feature.
2756  */
2757 static void
2758 ixgbe_setup_rss(ixgbe_t *ixgbe)
2759 {
2760 	struct ixgbe_hw *hw = &ixgbe->hw;
2761 	uint32_t mrqc;
2762 
2763 	/*
2764 	 * Initialize RETA/ERETA table
2765 	 */
2766 	ixgbe_setup_rss_table(ixgbe);
2767 
2768 	/*
2769 	 * Enable RSS & perform hash on these packet types
2770 	 */
2771 	mrqc = IXGBE_MRQC_RSSEN |
2772 	    IXGBE_MRQC_RSS_FIELD_IPV4 |
2773 	    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2774 	    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2775 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2776 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2777 	    IXGBE_MRQC_RSS_FIELD_IPV6 |
2778 	    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2779 	    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2780 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2781 	IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2782 }
2783 
2784 /*
2785  * ixgbe_setup_vmdq - Setup MAC classification feature
2786  */
2787 static void
2788 ixgbe_setup_vmdq(ixgbe_t *ixgbe)
2789 {
2790 	struct ixgbe_hw *hw = &ixgbe->hw;
2791 	uint32_t vmdctl, i, vtctl;
2792 
2793 	/*
2794 	 * Setup the VMDq Control register, enable VMDq based on
2795 	 * packet destination MAC address:
2796 	 */
2797 	switch (hw->mac.type) {
2798 	case ixgbe_mac_82598EB:
2799 		/*
2800 		 * VMDq Enable = 1;
2801 		 * VMDq Filter = 0; MAC filtering
2802 		 * Default VMDq output index = 0;
2803 		 */
2804 		vmdctl = IXGBE_VMD_CTL_VMDQ_EN;
2805 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
2806 		break;
2807 
2808 	case ixgbe_mac_82599EB:
2809 	case ixgbe_mac_X540:
2810 	case ixgbe_mac_X550:
2811 	case ixgbe_mac_X550EM_x:
2812 		/*
2813 		 * Enable VMDq-only.
2814 		 */
2815 		vmdctl = IXGBE_MRQC_VMDQEN;
2816 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, vmdctl);
2817 
2818 		for (i = 0; i < hw->mac.num_rar_entries; i++) {
2819 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(i), 0);
2820 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(i), 0);
2821 		}
2822 
2823 		/*
2824 		 * Enable Virtualization and Replication.
2825 		 */
2826 		vtctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2827 		IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vtctl);
2828 
2829 		/*
2830 		 * Enable receiving packets to all VFs
2831 		 */
2832 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), IXGBE_VFRE_ENABLE_ALL);
2833 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), IXGBE_VFRE_ENABLE_ALL);
2834 		break;
2835 
2836 	default:
2837 		break;
2838 	}
2839 }
2840 
2841 /*
2842  * ixgbe_setup_vmdq_rss - Setup both vmdq feature and rss feature.
2843  */
2844 static void
2845 ixgbe_setup_vmdq_rss(ixgbe_t *ixgbe)
2846 {
2847 	struct ixgbe_hw *hw = &ixgbe->hw;
2848 	uint32_t i, mrqc;
2849 	uint32_t vtctl, vmdctl;
2850 
2851 	/*
2852 	 * Initialize RETA/ERETA table
2853 	 */
2854 	ixgbe_setup_rss_table(ixgbe);
2855 
2856 	/*
2857 	 * Enable and setup RSS and VMDq
2858 	 */
2859 	switch (hw->mac.type) {
2860 	case ixgbe_mac_82598EB:
2861 		/*
2862 		 * Enable RSS & Setup RSS Hash functions
2863 		 */
2864 		mrqc = IXGBE_MRQC_RSSEN |
2865 		    IXGBE_MRQC_RSS_FIELD_IPV4 |
2866 		    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2867 		    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2868 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2869 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2870 		    IXGBE_MRQC_RSS_FIELD_IPV6 |
2871 		    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2872 		    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2873 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2874 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2875 
2876 		/*
2877 		 * Enable and Setup VMDq
2878 		 * VMDq Filter = 0; MAC filtering
2879 		 * Default VMDq output index = 0;
2880 		 */
2881 		vmdctl = IXGBE_VMD_CTL_VMDQ_EN;
2882 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
2883 		break;
2884 
2885 	case ixgbe_mac_82599EB:
2886 	case ixgbe_mac_X540:
2887 	case ixgbe_mac_X550:
2888 	case ixgbe_mac_X550EM_x:
2889 		/*
2890 		 * Enable RSS & Setup RSS Hash functions
2891 		 */
2892 		mrqc = IXGBE_MRQC_RSS_FIELD_IPV4 |
2893 		    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2894 		    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2895 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2896 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2897 		    IXGBE_MRQC_RSS_FIELD_IPV6 |
2898 		    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2899 		    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2900 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2901 
2902 		/*
2903 		 * Enable VMDq+RSS.
2904 		 */
2905 		if (ixgbe->num_rx_groups > 32)  {
2906 			mrqc = mrqc | IXGBE_MRQC_VMDQRSS64EN;
2907 		} else {
2908 			mrqc = mrqc | IXGBE_MRQC_VMDQRSS32EN;
2909 		}
2910 
2911 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2912 
2913 		for (i = 0; i < hw->mac.num_rar_entries; i++) {
2914 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(i), 0);
2915 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(i), 0);
2916 		}
2917 		break;
2918 
2919 	default:
2920 		break;
2921 
2922 	}
2923 
2924 	if (hw->mac.type == ixgbe_mac_82599EB ||
2925 	    hw->mac.type == ixgbe_mac_X540 ||
2926 	    hw->mac.type == ixgbe_mac_X550 ||
2927 	    hw->mac.type == ixgbe_mac_X550EM_x) {
2928 		/*
2929 		 * Enable Virtualization and Replication.
2930 		 */
2931 		vtctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2932 		IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vtctl);
2933 
2934 		/*
2935 		 * Enable receiving packets to all VFs
2936 		 */
2937 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), IXGBE_VFRE_ENABLE_ALL);
2938 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), IXGBE_VFRE_ENABLE_ALL);
2939 	}
2940 }
2941 
2942 /*
2943  * ixgbe_setup_rss_table - Setup RSS table
2944  */
2945 static void
2946 ixgbe_setup_rss_table(ixgbe_t *ixgbe)
2947 {
2948 	struct ixgbe_hw *hw = &ixgbe->hw;
2949 	uint32_t i, j;
2950 	uint32_t random;
2951 	uint32_t reta;
2952 	uint32_t ring_per_group;
2953 	uint32_t ring;
2954 	uint32_t table_size;
2955 	uint32_t index_mult;
2956 	uint32_t rxcsum;
2957 
2958 	/*
2959 	 * Set multiplier for RETA setup and table size based on MAC type.
2960 	 * RETA table sizes vary by model:
2961 	 *
2962 	 * 82598, 82599, X540: 128 table entries.
2963 	 * X550: 512 table entries.
2964 	 */
2965 	index_mult = 0x1;
2966 	table_size = 128;
2967 	switch (ixgbe->hw.mac.type) {
2968 	case ixgbe_mac_82598EB:
2969 		index_mult = 0x11;
2970 		break;
2971 	case ixgbe_mac_X550:
2972 	case ixgbe_mac_X550EM_x:
2973 		table_size = 512;
2974 		break;
2975 	default:
2976 		break;
2977 	}
2978 
2979 	/*
2980 	 * Fill out RSS redirection table. The configuation of the indices is
2981 	 * hardware-dependent.
2982 	 *
2983 	 *  82598: 8 bits wide containing two 4 bit RSS indices
2984 	 *  82599, X540: 8 bits wide containing one 4 bit RSS index
2985 	 *  X550: 8 bits wide containing one 6 bit RSS index
2986 	 */
2987 	reta = 0;
2988 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
2989 
2990 	for (i = 0, j = 0; i < table_size; i++, j++) {
2991 		if (j == ring_per_group) j = 0;
2992 
2993 		/*
2994 		 * The low 8 bits are for hash value (n+0);
2995 		 * The next 8 bits are for hash value (n+1), etc.
2996 		 */
2997 		ring = (j * index_mult);
2998 		reta = reta >> 8;
2999 		reta = reta | (((uint32_t)ring) << 24);
3000 
3001 		if ((i & 3) == 3) {
3002 			/*
3003 			 * The first 128 table entries are programmed into the
3004 			 * RETA register, with any beyond that (eg; on X550)
3005 			 * into ERETA.
3006 			 */
3007 			if (i < 128)
3008 				IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3009 			else
3010 				IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3011 				    reta);
3012 			reta = 0;
3013 		}
3014 	}
3015 
3016 	/*
3017 	 * Fill out hash function seeds with a random constant
3018 	 */
3019 	for (i = 0; i < 10; i++) {
3020 		(void) random_get_pseudo_bytes((uint8_t *)&random,
3021 		    sizeof (uint32_t));
3022 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), random);
3023 	}
3024 
3025 	/*
3026 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
3027 	 * It is an adapter hardware limitation that Packet Checksum is
3028 	 * mutually exclusive with RSS.
3029 	 */
3030 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3031 	rxcsum |= IXGBE_RXCSUM_PCSD;
3032 	rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
3033 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3034 }
3035 
3036 /*
3037  * ixgbe_init_unicst - Initialize the unicast addresses.
3038  */
3039 static void
3040 ixgbe_init_unicst(ixgbe_t *ixgbe)
3041 {
3042 	struct ixgbe_hw *hw = &ixgbe->hw;
3043 	uint8_t *mac_addr;
3044 	int slot;
3045 	/*
3046 	 * Here we should consider two situations:
3047 	 *
3048 	 * 1. Chipset is initialized at the first time,
3049 	 *    Clear all the multiple unicast addresses.
3050 	 *
3051 	 * 2. Chipset is reset
3052 	 *    Recover the multiple unicast addresses from the
3053 	 *    software data structure to the RAR registers.
3054 	 */
3055 	if (!ixgbe->unicst_init) {
3056 		/*
3057 		 * Initialize the multiple unicast addresses
3058 		 */
3059 		ixgbe->unicst_total = hw->mac.num_rar_entries;
3060 		ixgbe->unicst_avail = ixgbe->unicst_total;
3061 		for (slot = 0; slot < ixgbe->unicst_total; slot++) {
3062 			mac_addr = ixgbe->unicst_addr[slot].mac.addr;
3063 			bzero(mac_addr, ETHERADDRL);
3064 			(void) ixgbe_set_rar(hw, slot, mac_addr, NULL, NULL);
3065 			ixgbe->unicst_addr[slot].mac.set = 0;
3066 		}
3067 		ixgbe->unicst_init = B_TRUE;
3068 	} else {
3069 		/* Re-configure the RAR registers */
3070 		for (slot = 0; slot < ixgbe->unicst_total; slot++) {
3071 			mac_addr = ixgbe->unicst_addr[slot].mac.addr;
3072 			if (ixgbe->unicst_addr[slot].mac.set == 1) {
3073 				(void) ixgbe_set_rar(hw, slot, mac_addr,
3074 				    ixgbe->unicst_addr[slot].mac.group_index,
3075 				    IXGBE_RAH_AV);
3076 			} else {
3077 				bzero(mac_addr, ETHERADDRL);
3078 				(void) ixgbe_set_rar(hw, slot, mac_addr,
3079 				    NULL, NULL);
3080 			}
3081 		}
3082 	}
3083 }
3084 
3085 /*
3086  * ixgbe_unicst_find - Find the slot for the specified unicast address
3087  */
3088 int
3089 ixgbe_unicst_find(ixgbe_t *ixgbe, const uint8_t *mac_addr)
3090 {
3091 	int slot;
3092 
3093 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3094 
3095 	for (slot = 0; slot < ixgbe->unicst_total; slot++) {
3096 		if (bcmp(ixgbe->unicst_addr[slot].mac.addr,
3097 		    mac_addr, ETHERADDRL) == 0)
3098 			return (slot);
3099 	}
3100 
3101 	return (-1);
3102 }
3103 
3104 /*
3105  * ixgbe_multicst_add - Add a multicst address.
3106  */
3107 int
3108 ixgbe_multicst_add(ixgbe_t *ixgbe, const uint8_t *multiaddr)
3109 {
3110 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3111 
3112 	if ((multiaddr[0] & 01) == 0) {
3113 		return (EINVAL);
3114 	}
3115 
3116 	if (ixgbe->mcast_count >= MAX_NUM_MULTICAST_ADDRESSES) {
3117 		return (ENOENT);
3118 	}
3119 
3120 	bcopy(multiaddr,
3121 	    &ixgbe->mcast_table[ixgbe->mcast_count], ETHERADDRL);
3122 	ixgbe->mcast_count++;
3123 
3124 	/*
3125 	 * Update the multicast table in the hardware
3126 	 */
3127 	ixgbe_setup_multicst(ixgbe);
3128 
3129 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
3130 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3131 		return (EIO);
3132 	}
3133 
3134 	return (0);
3135 }
3136 
3137 /*
3138  * ixgbe_multicst_remove - Remove a multicst address.
3139  */
3140 int
3141 ixgbe_multicst_remove(ixgbe_t *ixgbe, const uint8_t *multiaddr)
3142 {
3143 	int i;
3144 
3145 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3146 
3147 	for (i = 0; i < ixgbe->mcast_count; i++) {
3148 		if (bcmp(multiaddr, &ixgbe->mcast_table[i],
3149 		    ETHERADDRL) == 0) {
3150 			for (i++; i < ixgbe->mcast_count; i++) {
3151 				ixgbe->mcast_table[i - 1] =
3152 				    ixgbe->mcast_table[i];
3153 			}
3154 			ixgbe->mcast_count--;
3155 			break;
3156 		}
3157 	}
3158 
3159 	/*
3160 	 * Update the multicast table in the hardware
3161 	 */
3162 	ixgbe_setup_multicst(ixgbe);
3163 
3164 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
3165 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3166 		return (EIO);
3167 	}
3168 
3169 	return (0);
3170 }
3171 
3172 /*
3173  * ixgbe_setup_multicast - Setup multicast data structures.
3174  *
3175  * This routine initializes all of the multicast related structures
3176  * and save them in the hardware registers.
3177  */
3178 static void
3179 ixgbe_setup_multicst(ixgbe_t *ixgbe)
3180 {
3181 	uint8_t *mc_addr_list;
3182 	uint32_t mc_addr_count;
3183 	struct ixgbe_hw *hw = &ixgbe->hw;
3184 
3185 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3186 
3187 	ASSERT(ixgbe->mcast_count <= MAX_NUM_MULTICAST_ADDRESSES);
3188 
3189 	mc_addr_list = (uint8_t *)ixgbe->mcast_table;
3190 	mc_addr_count = ixgbe->mcast_count;
3191 
3192 	/*
3193 	 * Update the multicast addresses to the MTA registers
3194 	 */
3195 	(void) ixgbe_update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
3196 	    ixgbe_mc_table_itr, TRUE);
3197 }
3198 
3199 /*
3200  * ixgbe_setup_vmdq_rss_conf - Configure vmdq and rss (number and mode).
3201  *
3202  * Configure the rx classification mode (vmdq & rss) and vmdq & rss numbers.
3203  * Different chipsets may have different allowed configuration of vmdq and rss.
3204  */
3205 static void
3206 ixgbe_setup_vmdq_rss_conf(ixgbe_t *ixgbe)
3207 {
3208 	struct ixgbe_hw *hw = &ixgbe->hw;
3209 	uint32_t ring_per_group;
3210 
3211 	switch (hw->mac.type) {
3212 	case ixgbe_mac_82598EB:
3213 		/*
3214 		 * 82598 supports the following combination:
3215 		 * vmdq no. x rss no.
3216 		 * [5..16]  x 1
3217 		 * [1..4]   x [1..16]
3218 		 * However 8 rss queue per pool (vmdq) is sufficient for
3219 		 * most cases.
3220 		 */
3221 		ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
3222 		if (ixgbe->num_rx_groups > 4) {
3223 			ixgbe->num_rx_rings = ixgbe->num_rx_groups;
3224 		} else {
3225 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3226 			    min(8, ring_per_group);
3227 		}
3228 
3229 		break;
3230 
3231 	case ixgbe_mac_82599EB:
3232 	case ixgbe_mac_X540:
3233 	case ixgbe_mac_X550:
3234 	case ixgbe_mac_X550EM_x:
3235 		/*
3236 		 * 82599 supports the following combination:
3237 		 * vmdq no. x rss no.
3238 		 * [33..64] x [1..2]
3239 		 * [2..32]  x [1..4]
3240 		 * 1 x [1..16]
3241 		 * However 8 rss queue per pool (vmdq) is sufficient for
3242 		 * most cases.
3243 		 *
3244 		 * For now, treat X540 and X550 like the 82599.
3245 		 */
3246 		ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
3247 		if (ixgbe->num_rx_groups == 1) {
3248 			ixgbe->num_rx_rings = min(8, ring_per_group);
3249 		} else if (ixgbe->num_rx_groups <= 32) {
3250 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3251 			    min(4, ring_per_group);
3252 		} else if (ixgbe->num_rx_groups <= 64) {
3253 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3254 			    min(2, ring_per_group);
3255 		}
3256 		break;
3257 
3258 	default:
3259 		break;
3260 	}
3261 
3262 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
3263 
3264 	if (ixgbe->num_rx_groups == 1 && ring_per_group == 1) {
3265 		ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
3266 	} else if (ixgbe->num_rx_groups != 1 && ring_per_group == 1) {
3267 		ixgbe->classify_mode = IXGBE_CLASSIFY_VMDQ;
3268 	} else if (ixgbe->num_rx_groups != 1 && ring_per_group != 1) {
3269 		ixgbe->classify_mode = IXGBE_CLASSIFY_VMDQ_RSS;
3270 	} else {
3271 		ixgbe->classify_mode = IXGBE_CLASSIFY_RSS;
3272 	}
3273 
3274 	IXGBE_DEBUGLOG_2(ixgbe, "rx group number:%d, rx ring number:%d",
3275 	    ixgbe->num_rx_groups, ixgbe->num_rx_rings);
3276 }
3277 
3278 /*
3279  * ixgbe_get_conf - Get driver configurations set in driver.conf.
3280  *
3281  * This routine gets user-configured values out of the configuration
3282  * file ixgbe.conf.
3283  *
3284  * For each configurable value, there is a minimum, a maximum, and a
3285  * default.
3286  * If user does not configure a value, use the default.
3287  * If user configures below the minimum, use the minumum.
3288  * If user configures above the maximum, use the maxumum.
3289  */
3290 static void
3291 ixgbe_get_conf(ixgbe_t *ixgbe)
3292 {
3293 	struct ixgbe_hw *hw = &ixgbe->hw;
3294 	uint32_t flow_control;
3295 
3296 	/*
3297 	 * ixgbe driver supports the following user configurations:
3298 	 *
3299 	 * Jumbo frame configuration:
3300 	 *    default_mtu
3301 	 *
3302 	 * Ethernet flow control configuration:
3303 	 *    flow_control
3304 	 *
3305 	 * Multiple rings configurations:
3306 	 *    tx_queue_number
3307 	 *    tx_ring_size
3308 	 *    rx_queue_number
3309 	 *    rx_ring_size
3310 	 *
3311 	 * Call ixgbe_get_prop() to get the value for a specific
3312 	 * configuration parameter.
3313 	 */
3314 
3315 	/*
3316 	 * Jumbo frame configuration - max_frame_size controls host buffer
3317 	 * allocation, so includes MTU, ethernet header, vlan tag and
3318 	 * frame check sequence.
3319 	 */
3320 	ixgbe->default_mtu = ixgbe_get_prop(ixgbe, PROP_DEFAULT_MTU,
3321 	    MIN_MTU, ixgbe->capab->max_mtu, DEFAULT_MTU);
3322 
3323 	ixgbe->max_frame_size = ixgbe->default_mtu +
3324 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
3325 
3326 	/*
3327 	 * Ethernet flow control configuration
3328 	 */
3329 	flow_control = ixgbe_get_prop(ixgbe, PROP_FLOW_CONTROL,
3330 	    ixgbe_fc_none, 3, ixgbe_fc_none);
3331 	if (flow_control == 3)
3332 		flow_control = ixgbe_fc_default;
3333 
3334 	/*
3335 	 * fc.requested mode is what the user requests.  After autoneg,
3336 	 * fc.current_mode will be the flow_control mode that was negotiated.
3337 	 */
3338 	hw->fc.requested_mode = flow_control;
3339 
3340 	/*
3341 	 * Multiple rings configurations
3342 	 */
3343 	ixgbe->num_tx_rings = ixgbe_get_prop(ixgbe, PROP_TX_QUEUE_NUM,
3344 	    ixgbe->capab->min_tx_que_num,
3345 	    ixgbe->capab->max_tx_que_num,
3346 	    ixgbe->capab->def_tx_que_num);
3347 	ixgbe->tx_ring_size = ixgbe_get_prop(ixgbe, PROP_TX_RING_SIZE,
3348 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
3349 
3350 	ixgbe->num_rx_rings = ixgbe_get_prop(ixgbe, PROP_RX_QUEUE_NUM,
3351 	    ixgbe->capab->min_rx_que_num,
3352 	    ixgbe->capab->max_rx_que_num,
3353 	    ixgbe->capab->def_rx_que_num);
3354 	ixgbe->rx_ring_size = ixgbe_get_prop(ixgbe, PROP_RX_RING_SIZE,
3355 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
3356 
3357 	/*
3358 	 * Multiple groups configuration
3359 	 */
3360 	ixgbe->num_rx_groups = ixgbe_get_prop(ixgbe, PROP_RX_GROUP_NUM,
3361 	    ixgbe->capab->min_rx_grp_num, ixgbe->capab->max_rx_grp_num,
3362 	    ixgbe->capab->def_rx_grp_num);
3363 
3364 	ixgbe->mr_enable = ixgbe_get_prop(ixgbe, PROP_MR_ENABLE,
3365 	    0, 1, DEFAULT_MR_ENABLE);
3366 
3367 	if (ixgbe->mr_enable == B_FALSE) {
3368 		ixgbe->num_tx_rings = 1;
3369 		ixgbe->num_rx_rings = 1;
3370 		ixgbe->num_rx_groups = 1;
3371 		ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
3372 	} else {
3373 		ixgbe->num_rx_rings = ixgbe->num_rx_groups *
3374 		    max(ixgbe->num_rx_rings / ixgbe->num_rx_groups, 1);
3375 		/*
3376 		 * The combination of num_rx_rings and num_rx_groups
3377 		 * may be not supported by h/w. We need to adjust
3378 		 * them to appropriate values.
3379 		 */
3380 		ixgbe_setup_vmdq_rss_conf(ixgbe);
3381 	}
3382 
3383 	/*
3384 	 * Tunable used to force an interrupt type. The only use is
3385 	 * for testing of the lesser interrupt types.
3386 	 * 0 = don't force interrupt type
3387 	 * 1 = force interrupt type MSI-X
3388 	 * 2 = force interrupt type MSI
3389 	 * 3 = force interrupt type Legacy
3390 	 */
3391 	ixgbe->intr_force = ixgbe_get_prop(ixgbe, PROP_INTR_FORCE,
3392 	    IXGBE_INTR_NONE, IXGBE_INTR_LEGACY, IXGBE_INTR_NONE);
3393 
3394 	ixgbe->tx_hcksum_enable = ixgbe_get_prop(ixgbe, PROP_TX_HCKSUM_ENABLE,
3395 	    0, 1, DEFAULT_TX_HCKSUM_ENABLE);
3396 	ixgbe->rx_hcksum_enable = ixgbe_get_prop(ixgbe, PROP_RX_HCKSUM_ENABLE,
3397 	    0, 1, DEFAULT_RX_HCKSUM_ENABLE);
3398 	ixgbe->lso_enable = ixgbe_get_prop(ixgbe, PROP_LSO_ENABLE,
3399 	    0, 1, DEFAULT_LSO_ENABLE);
3400 	ixgbe->lro_enable = ixgbe_get_prop(ixgbe, PROP_LRO_ENABLE,
3401 	    0, 1, DEFAULT_LRO_ENABLE);
3402 	ixgbe->tx_head_wb_enable = ixgbe_get_prop(ixgbe, PROP_TX_HEAD_WB_ENABLE,
3403 	    0, 1, DEFAULT_TX_HEAD_WB_ENABLE);
3404 	ixgbe->relax_order_enable = ixgbe_get_prop(ixgbe,
3405 	    PROP_RELAX_ORDER_ENABLE, 0, 1, DEFAULT_RELAX_ORDER_ENABLE);
3406 
3407 	/* Head Write Back not recommended for 82599, X540 and X550 */
3408 	if (hw->mac.type == ixgbe_mac_82599EB ||
3409 	    hw->mac.type == ixgbe_mac_X540 ||
3410 	    hw->mac.type == ixgbe_mac_X550 ||
3411 	    hw->mac.type == ixgbe_mac_X550EM_x) {
3412 		ixgbe->tx_head_wb_enable = B_FALSE;
3413 	}
3414 
3415 	/*
3416 	 * ixgbe LSO needs the tx h/w checksum support.
3417 	 * LSO will be disabled if tx h/w checksum is not
3418 	 * enabled.
3419 	 */
3420 	if (ixgbe->tx_hcksum_enable == B_FALSE) {
3421 		ixgbe->lso_enable = B_FALSE;
3422 	}
3423 
3424 	/*
3425 	 * ixgbe LRO needs the rx h/w checksum support.
3426 	 * LRO will be disabled if rx h/w checksum is not
3427 	 * enabled.
3428 	 */
3429 	if (ixgbe->rx_hcksum_enable == B_FALSE) {
3430 		ixgbe->lro_enable = B_FALSE;
3431 	}
3432 
3433 	/*
3434 	 * ixgbe LRO only supported by 82599, X540 and X550
3435 	 */
3436 	if (hw->mac.type == ixgbe_mac_82598EB) {
3437 		ixgbe->lro_enable = B_FALSE;
3438 	}
3439 	ixgbe->tx_copy_thresh = ixgbe_get_prop(ixgbe, PROP_TX_COPY_THRESHOLD,
3440 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
3441 	    DEFAULT_TX_COPY_THRESHOLD);
3442 	ixgbe->tx_recycle_thresh = ixgbe_get_prop(ixgbe,
3443 	    PROP_TX_RECYCLE_THRESHOLD, MIN_TX_RECYCLE_THRESHOLD,
3444 	    MAX_TX_RECYCLE_THRESHOLD, DEFAULT_TX_RECYCLE_THRESHOLD);
3445 	ixgbe->tx_overload_thresh = ixgbe_get_prop(ixgbe,
3446 	    PROP_TX_OVERLOAD_THRESHOLD, MIN_TX_OVERLOAD_THRESHOLD,
3447 	    MAX_TX_OVERLOAD_THRESHOLD, DEFAULT_TX_OVERLOAD_THRESHOLD);
3448 	ixgbe->tx_resched_thresh = ixgbe_get_prop(ixgbe,
3449 	    PROP_TX_RESCHED_THRESHOLD, MIN_TX_RESCHED_THRESHOLD,
3450 	    MAX_TX_RESCHED_THRESHOLD, DEFAULT_TX_RESCHED_THRESHOLD);
3451 
3452 	ixgbe->rx_copy_thresh = ixgbe_get_prop(ixgbe, PROP_RX_COPY_THRESHOLD,
3453 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
3454 	    DEFAULT_RX_COPY_THRESHOLD);
3455 	ixgbe->rx_limit_per_intr = ixgbe_get_prop(ixgbe, PROP_RX_LIMIT_PER_INTR,
3456 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
3457 	    DEFAULT_RX_LIMIT_PER_INTR);
3458 
3459 	ixgbe->intr_throttling[0] = ixgbe_get_prop(ixgbe, PROP_INTR_THROTTLING,
3460 	    ixgbe->capab->min_intr_throttle,
3461 	    ixgbe->capab->max_intr_throttle,
3462 	    ixgbe->capab->def_intr_throttle);
3463 	/*
3464 	 * 82599, X540 and X550 require the interrupt throttling rate is
3465 	 * a multiple of 8. This is enforced by the register definiton.
3466 	 */
3467 	if (hw->mac.type == ixgbe_mac_82599EB ||
3468 	    hw->mac.type == ixgbe_mac_X540 ||
3469 	    hw->mac.type == ixgbe_mac_X550 ||
3470 	    hw->mac.type == ixgbe_mac_X550EM_x)
3471 		ixgbe->intr_throttling[0] = ixgbe->intr_throttling[0] & 0xFF8;
3472 
3473 	hw->allow_unsupported_sfp = ixgbe_get_prop(ixgbe,
3474 	    PROP_ALLOW_UNSUPPORTED_SFP, 0, 1, DEFAULT_ALLOW_UNSUPPORTED_SFP);
3475 }
3476 
3477 static void
3478 ixgbe_init_params(ixgbe_t *ixgbe)
3479 {
3480 	struct ixgbe_hw *hw = &ixgbe->hw;
3481 	ixgbe_link_speed speeds_supported = 0;
3482 	boolean_t negotiate;
3483 
3484 	/*
3485 	 * Get a list of speeds the adapter supports. If the hw struct hasn't
3486 	 * been populated with this information yet, retrieve it from the
3487 	 * adapter and save it to our own variable.
3488 	 *
3489 	 * On certain adapters, such as ones which use SFPs, the contents of
3490 	 * hw->phy.speeds_supported (and hw->phy.autoneg_advertised) are not
3491 	 * updated, so we must rely on calling ixgbe_get_link_capabilities()
3492 	 * in order to ascertain the speeds which we are capable of supporting,
3493 	 * and in the case of SFP-equipped adapters, which speed we are
3494 	 * advertising. If ixgbe_get_link_capabilities() fails for some reason,
3495 	 * we'll go with a default list of speeds as a last resort.
3496 	 */
3497 	speeds_supported = hw->phy.speeds_supported;
3498 
3499 	if (speeds_supported == 0) {
3500 		if (ixgbe_get_link_capabilities(hw, &speeds_supported,
3501 		    &negotiate) != IXGBE_SUCCESS) {
3502 			if (hw->mac.type == ixgbe_mac_82598EB) {
3503 				speeds_supported =
3504 				    IXGBE_LINK_SPEED_82598_AUTONEG;
3505 			} else {
3506 				speeds_supported =
3507 				    IXGBE_LINK_SPEED_82599_AUTONEG;
3508 			}
3509 		}
3510 	}
3511 	ixgbe->speeds_supported = speeds_supported;
3512 
3513 	/*
3514 	 * By default, all supported speeds are enabled and advertised.
3515 	 */
3516 	if (speeds_supported & IXGBE_LINK_SPEED_10GB_FULL) {
3517 		ixgbe->param_en_10000fdx_cap = 1;
3518 		ixgbe->param_adv_10000fdx_cap = 1;
3519 	} else {
3520 		ixgbe->param_en_10000fdx_cap = 0;
3521 		ixgbe->param_adv_10000fdx_cap = 0;
3522 	}
3523 
3524 	if (speeds_supported & IXGBE_LINK_SPEED_5GB_FULL) {
3525 		ixgbe->param_en_5000fdx_cap = 1;
3526 		ixgbe->param_adv_5000fdx_cap = 1;
3527 	} else {
3528 		ixgbe->param_en_5000fdx_cap = 0;
3529 		ixgbe->param_adv_5000fdx_cap = 0;
3530 	}
3531 
3532 	if (speeds_supported & IXGBE_LINK_SPEED_2_5GB_FULL) {
3533 		ixgbe->param_en_2500fdx_cap = 1;
3534 		ixgbe->param_adv_2500fdx_cap = 1;
3535 	} else {
3536 		ixgbe->param_en_2500fdx_cap = 0;
3537 		ixgbe->param_adv_2500fdx_cap = 0;
3538 	}
3539 
3540 	if (speeds_supported & IXGBE_LINK_SPEED_1GB_FULL) {
3541 		ixgbe->param_en_1000fdx_cap = 1;
3542 		ixgbe->param_adv_1000fdx_cap = 1;
3543 	} else {
3544 		ixgbe->param_en_1000fdx_cap = 0;
3545 		ixgbe->param_adv_1000fdx_cap = 0;
3546 	}
3547 
3548 	if (speeds_supported & IXGBE_LINK_SPEED_100_FULL) {
3549 		ixgbe->param_en_100fdx_cap = 1;
3550 		ixgbe->param_adv_100fdx_cap = 1;
3551 	} else {
3552 		ixgbe->param_en_100fdx_cap = 0;
3553 		ixgbe->param_adv_100fdx_cap = 0;
3554 	}
3555 
3556 	ixgbe->param_pause_cap = 1;
3557 	ixgbe->param_asym_pause_cap = 1;
3558 	ixgbe->param_rem_fault = 0;
3559 
3560 	ixgbe->param_adv_autoneg_cap = 1;
3561 	ixgbe->param_adv_pause_cap = 1;
3562 	ixgbe->param_adv_asym_pause_cap = 1;
3563 	ixgbe->param_adv_rem_fault = 0;
3564 
3565 	ixgbe->param_lp_10000fdx_cap = 0;
3566 	ixgbe->param_lp_5000fdx_cap = 0;
3567 	ixgbe->param_lp_2500fdx_cap = 0;
3568 	ixgbe->param_lp_1000fdx_cap = 0;
3569 	ixgbe->param_lp_100fdx_cap = 0;
3570 	ixgbe->param_lp_autoneg_cap = 0;
3571 	ixgbe->param_lp_pause_cap = 0;
3572 	ixgbe->param_lp_asym_pause_cap = 0;
3573 	ixgbe->param_lp_rem_fault = 0;
3574 }
3575 
3576 /*
3577  * ixgbe_get_prop - Get a property value out of the configuration file
3578  * ixgbe.conf.
3579  *
3580  * Caller provides the name of the property, a default value, a minimum
3581  * value, and a maximum value.
3582  *
3583  * Return configured value of the property, with default, minimum and
3584  * maximum properly applied.
3585  */
3586 static int
3587 ixgbe_get_prop(ixgbe_t *ixgbe,
3588     char *propname,	/* name of the property */
3589     int minval,		/* minimum acceptable value */
3590     int maxval,		/* maximim acceptable value */
3591     int defval)		/* default value */
3592 {
3593 	int value;
3594 
3595 	/*
3596 	 * Call ddi_prop_get_int() to read the conf settings
3597 	 */
3598 	value = ddi_prop_get_int(DDI_DEV_T_ANY, ixgbe->dip,
3599 	    DDI_PROP_DONTPASS, propname, defval);
3600 	if (value > maxval)
3601 		value = maxval;
3602 
3603 	if (value < minval)
3604 		value = minval;
3605 
3606 	return (value);
3607 }
3608 
3609 /*
3610  * ixgbe_driver_setup_link - Using the link properties to setup the link.
3611  */
3612 int
3613 ixgbe_driver_setup_link(ixgbe_t *ixgbe, boolean_t setup_hw)
3614 {
3615 	struct ixgbe_hw *hw = &ixgbe->hw;
3616 	ixgbe_link_speed advertised = 0;
3617 
3618 	/*
3619 	 * Assemble a list of enabled speeds to auto-negotiate with.
3620 	 */
3621 	if (ixgbe->param_en_10000fdx_cap == 1)
3622 		advertised |= IXGBE_LINK_SPEED_10GB_FULL;
3623 
3624 	if (ixgbe->param_en_5000fdx_cap == 1)
3625 		advertised |= IXGBE_LINK_SPEED_5GB_FULL;
3626 
3627 	if (ixgbe->param_en_2500fdx_cap == 1)
3628 		advertised |= IXGBE_LINK_SPEED_2_5GB_FULL;
3629 
3630 	if (ixgbe->param_en_1000fdx_cap == 1)
3631 		advertised |= IXGBE_LINK_SPEED_1GB_FULL;
3632 
3633 	if (ixgbe->param_en_100fdx_cap == 1)
3634 		advertised |= IXGBE_LINK_SPEED_100_FULL;
3635 
3636 	/*
3637 	 * As a last resort, autoneg with a default list of speeds.
3638 	 */
3639 	if (ixgbe->param_adv_autoneg_cap == 1 && advertised == 0) {
3640 		ixgbe_notice(ixgbe, "Invalid link settings. Setting link "
3641 		    "to autonegotiate with full capabilities.");
3642 
3643 		if (hw->mac.type == ixgbe_mac_82598EB)
3644 			advertised = IXGBE_LINK_SPEED_82598_AUTONEG;
3645 		else
3646 			advertised = IXGBE_LINK_SPEED_82599_AUTONEG;
3647 	}
3648 
3649 	if (setup_hw) {
3650 		if (ixgbe_setup_link(&ixgbe->hw, advertised,
3651 		    ixgbe->param_adv_autoneg_cap) != IXGBE_SUCCESS) {
3652 			ixgbe_notice(ixgbe, "Setup link failed on this "
3653 			    "device.");
3654 			return (IXGBE_FAILURE);
3655 		}
3656 	}
3657 
3658 	return (IXGBE_SUCCESS);
3659 }
3660 
3661 /*
3662  * ixgbe_driver_link_check - Link status processing.
3663  *
3664  * This function can be called in both kernel context and interrupt context
3665  */
3666 static void
3667 ixgbe_driver_link_check(ixgbe_t *ixgbe)
3668 {
3669 	struct ixgbe_hw *hw = &ixgbe->hw;
3670 	ixgbe_link_speed speed = IXGBE_LINK_SPEED_UNKNOWN;
3671 	boolean_t link_up = B_FALSE;
3672 	boolean_t link_changed = B_FALSE;
3673 
3674 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3675 
3676 	(void) ixgbe_check_link(hw, &speed, &link_up, B_FALSE);
3677 	if (link_up) {
3678 		ixgbe->link_check_complete = B_TRUE;
3679 
3680 		/* Link is up, enable flow control settings */
3681 		(void) ixgbe_fc_enable(hw);
3682 
3683 		/*
3684 		 * The Link is up, check whether it was marked as down earlier
3685 		 */
3686 		if (ixgbe->link_state != LINK_STATE_UP) {
3687 			switch (speed) {
3688 			case IXGBE_LINK_SPEED_10GB_FULL:
3689 				ixgbe->link_speed = SPEED_10GB;
3690 				break;
3691 			case IXGBE_LINK_SPEED_5GB_FULL:
3692 				ixgbe->link_speed = SPEED_5GB;
3693 				break;
3694 			case IXGBE_LINK_SPEED_2_5GB_FULL:
3695 				ixgbe->link_speed = SPEED_2_5GB;
3696 				break;
3697 			case IXGBE_LINK_SPEED_1GB_FULL:
3698 				ixgbe->link_speed = SPEED_1GB;
3699 				break;
3700 			case IXGBE_LINK_SPEED_100_FULL:
3701 				ixgbe->link_speed = SPEED_100;
3702 			}
3703 			ixgbe->link_duplex = LINK_DUPLEX_FULL;
3704 			ixgbe->link_state = LINK_STATE_UP;
3705 			link_changed = B_TRUE;
3706 		}
3707 	} else {
3708 		if (ixgbe->link_check_complete == B_TRUE ||
3709 		    (ixgbe->link_check_complete == B_FALSE &&
3710 		    gethrtime() >= ixgbe->link_check_hrtime)) {
3711 			/*
3712 			 * The link is really down
3713 			 */
3714 			ixgbe->link_check_complete = B_TRUE;
3715 
3716 			if (ixgbe->link_state != LINK_STATE_DOWN) {
3717 				ixgbe->link_speed = 0;
3718 				ixgbe->link_duplex = LINK_DUPLEX_UNKNOWN;
3719 				ixgbe->link_state = LINK_STATE_DOWN;
3720 				link_changed = B_TRUE;
3721 			}
3722 		}
3723 	}
3724 
3725 	/*
3726 	 * If we are in an interrupt context, need to re-enable the
3727 	 * interrupt, which was automasked
3728 	 */
3729 	if (servicing_interrupt() != 0) {
3730 		ixgbe->eims |= IXGBE_EICR_LSC;
3731 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
3732 	}
3733 
3734 	if (link_changed) {
3735 		mac_link_update(ixgbe->mac_hdl, ixgbe->link_state);
3736 	}
3737 }
3738 
3739 /*
3740  * ixgbe_sfp_check - sfp module processing done in taskq only for 82599.
3741  */
3742 static void
3743 ixgbe_sfp_check(void *arg)
3744 {
3745 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3746 	uint32_t eicr = ixgbe->eicr;
3747 	struct ixgbe_hw *hw = &ixgbe->hw;
3748 
3749 	mutex_enter(&ixgbe->gen_lock);
3750 	if (eicr & IXGBE_EICR_GPI_SDP1_BY_MAC(hw)) {
3751 		/* clear the interrupt */
3752 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1_BY_MAC(hw));
3753 
3754 		/* if link up, do multispeed fiber setup */
3755 		(void) ixgbe_setup_link(hw, IXGBE_LINK_SPEED_82599_AUTONEG,
3756 		    B_TRUE);
3757 		ixgbe_driver_link_check(ixgbe);
3758 		ixgbe_get_hw_state(ixgbe);
3759 	} else if (eicr & IXGBE_EICR_GPI_SDP2_BY_MAC(hw)) {
3760 		/* clear the interrupt */
3761 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2_BY_MAC(hw));
3762 
3763 		/* if link up, do sfp module setup */
3764 		(void) hw->mac.ops.setup_sfp(hw);
3765 
3766 		/* do multispeed fiber setup */
3767 		(void) ixgbe_setup_link(hw, IXGBE_LINK_SPEED_82599_AUTONEG,
3768 		    B_TRUE);
3769 		ixgbe_driver_link_check(ixgbe);
3770 		ixgbe_get_hw_state(ixgbe);
3771 	}
3772 	mutex_exit(&ixgbe->gen_lock);
3773 
3774 	/*
3775 	 * We need to fully re-check the link later.
3776 	 */
3777 	ixgbe->link_check_complete = B_FALSE;
3778 	ixgbe->link_check_hrtime = gethrtime() +
3779 	    (IXGBE_LINK_UP_TIME * 100000000ULL);
3780 }
3781 
3782 /*
3783  * ixgbe_overtemp_check - overtemp module processing done in taskq
3784  *
3785  * This routine will only be called on adapters with temperature sensor.
3786  * The indication of over-temperature can be either SDP0 interrupt or the link
3787  * status change interrupt.
3788  */
3789 static void
3790 ixgbe_overtemp_check(void *arg)
3791 {
3792 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3793 	struct ixgbe_hw *hw = &ixgbe->hw;
3794 	uint32_t eicr = ixgbe->eicr;
3795 	ixgbe_link_speed speed;
3796 	boolean_t link_up;
3797 
3798 	mutex_enter(&ixgbe->gen_lock);
3799 
3800 	/* make sure we know current state of link */
3801 	(void) ixgbe_check_link(hw, &speed, &link_up, B_FALSE);
3802 
3803 	/* check over-temp condition */
3804 	if (((eicr & IXGBE_EICR_GPI_SDP0_BY_MAC(hw)) && (!link_up)) ||
3805 	    (eicr & IXGBE_EICR_LSC)) {
3806 		if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP) {
3807 			atomic_or_32(&ixgbe->ixgbe_state, IXGBE_OVERTEMP);
3808 
3809 			/*
3810 			 * Disable the adapter interrupts
3811 			 */
3812 			ixgbe_disable_adapter_interrupts(ixgbe);
3813 
3814 			/*
3815 			 * Disable Rx/Tx units
3816 			 */
3817 			(void) ixgbe_stop_adapter(hw);
3818 
3819 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
3820 			ixgbe_error(ixgbe,
3821 			    "Problem: Network adapter has been stopped "
3822 			    "because it has overheated");
3823 			ixgbe_error(ixgbe,
3824 			    "Action: Restart the computer. "
3825 			    "If the problem persists, power off the system "
3826 			    "and replace the adapter");
3827 		}
3828 	}
3829 
3830 	/* write to clear the interrupt */
3831 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3832 
3833 	mutex_exit(&ixgbe->gen_lock);
3834 }
3835 
3836 /*
3837  * ixgbe_phy_check - taskq to process interrupts from an external PHY
3838  *
3839  * This routine will only be called on adapters with external PHYs
3840  * (such as X550) that may be trying to raise our attention to some event.
3841  * Currently, this is limited to claiming PHY overtemperature and link status
3842  * change (LSC) events, however this may expand to include other things in
3843  * future adapters.
3844  */
3845 static void
3846 ixgbe_phy_check(void *arg)
3847 {
3848 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3849 	struct ixgbe_hw *hw = &ixgbe->hw;
3850 	int rv;
3851 
3852 	mutex_enter(&ixgbe->gen_lock);
3853 
3854 	/*
3855 	 * X550 baseT PHY overtemp and LSC events are handled here.
3856 	 *
3857 	 * If an overtemp event occurs, it will be reflected in the
3858 	 * return value of phy.ops.handle_lasi() and the common code will
3859 	 * automatically power off the baseT PHY. This is our cue to trigger
3860 	 * an FMA event.
3861 	 *
3862 	 * If a link status change event occurs, phy.ops.handle_lasi() will
3863 	 * automatically initiate a link setup between the integrated KR PHY
3864 	 * and the external X557 PHY to ensure that the link speed between
3865 	 * them matches the link speed of the baseT link.
3866 	 */
3867 	rv = ixgbe_handle_lasi(hw);
3868 
3869 	if (rv == IXGBE_ERR_OVERTEMP) {
3870 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_OVERTEMP);
3871 
3872 		/*
3873 		 * Disable the adapter interrupts
3874 		 */
3875 		ixgbe_disable_adapter_interrupts(ixgbe);
3876 
3877 		/*
3878 		 * Disable Rx/Tx units
3879 		 */
3880 		(void) ixgbe_stop_adapter(hw);
3881 
3882 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
3883 		ixgbe_error(ixgbe,
3884 		    "Problem: Network adapter has been stopped due to a "
3885 		    "overtemperature event being detected.");
3886 		ixgbe_error(ixgbe,
3887 		    "Action: Shut down or restart the computer. If the issue "
3888 		    "persists, please take action in accordance with the "
3889 		    "recommendations from your system vendor.");
3890 	}
3891 
3892 	mutex_exit(&ixgbe->gen_lock);
3893 }
3894 
3895 /*
3896  * ixgbe_link_timer - timer for link status detection
3897  */
3898 static void
3899 ixgbe_link_timer(void *arg)
3900 {
3901 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3902 
3903 	mutex_enter(&ixgbe->gen_lock);
3904 	ixgbe_driver_link_check(ixgbe);
3905 	mutex_exit(&ixgbe->gen_lock);
3906 }
3907 
3908 /*
3909  * ixgbe_local_timer - Driver watchdog function.
3910  *
3911  * This function will handle the transmit stall check and other routines.
3912  */
3913 static void
3914 ixgbe_local_timer(void *arg)
3915 {
3916 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3917 
3918 	if (ixgbe->ixgbe_state & IXGBE_OVERTEMP)
3919 		goto out;
3920 
3921 	if (ixgbe->ixgbe_state & IXGBE_ERROR) {
3922 		ixgbe->reset_count++;
3923 		if (ixgbe_reset(ixgbe) == IXGBE_SUCCESS)
3924 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_RESTORED);
3925 		goto out;
3926 	}
3927 
3928 	if (ixgbe_stall_check(ixgbe)) {
3929 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_STALL);
3930 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3931 
3932 		ixgbe->reset_count++;
3933 		if (ixgbe_reset(ixgbe) == IXGBE_SUCCESS)
3934 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_RESTORED);
3935 	}
3936 
3937 out:
3938 	ixgbe_restart_watchdog_timer(ixgbe);
3939 }
3940 
3941 /*
3942  * ixgbe_stall_check - Check for transmit stall.
3943  *
3944  * This function checks if the adapter is stalled (in transmit).
3945  *
3946  * It is called each time the watchdog timeout is invoked.
3947  * If the transmit descriptor reclaim continuously fails,
3948  * the watchdog value will increment by 1. If the watchdog
3949  * value exceeds the threshold, the ixgbe is assumed to
3950  * have stalled and need to be reset.
3951  */
3952 static boolean_t
3953 ixgbe_stall_check(ixgbe_t *ixgbe)
3954 {
3955 	ixgbe_tx_ring_t *tx_ring;
3956 	boolean_t result;
3957 	int i;
3958 
3959 	if (ixgbe->link_state != LINK_STATE_UP)
3960 		return (B_FALSE);
3961 
3962 	/*
3963 	 * If any tx ring is stalled, we'll reset the chipset
3964 	 */
3965 	result = B_FALSE;
3966 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
3967 		tx_ring = &ixgbe->tx_rings[i];
3968 		if (tx_ring->tbd_free <= ixgbe->tx_recycle_thresh) {
3969 			tx_ring->tx_recycle(tx_ring);
3970 		}
3971 
3972 		if (tx_ring->recycle_fail > 0)
3973 			tx_ring->stall_watchdog++;
3974 		else
3975 			tx_ring->stall_watchdog = 0;
3976 
3977 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
3978 			result = B_TRUE;
3979 			break;
3980 		}
3981 	}
3982 
3983 	if (result) {
3984 		tx_ring->stall_watchdog = 0;
3985 		tx_ring->recycle_fail = 0;
3986 	}
3987 
3988 	return (result);
3989 }
3990 
3991 
3992 /*
3993  * is_valid_mac_addr - Check if the mac address is valid.
3994  */
3995 static boolean_t
3996 is_valid_mac_addr(uint8_t *mac_addr)
3997 {
3998 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
3999 	const uint8_t addr_test2[6] =
4000 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
4001 
4002 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
4003 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
4004 		return (B_FALSE);
4005 
4006 	return (B_TRUE);
4007 }
4008 
4009 static boolean_t
4010 ixgbe_find_mac_address(ixgbe_t *ixgbe)
4011 {
4012 #ifdef __sparc
4013 	struct ixgbe_hw *hw = &ixgbe->hw;
4014 	uchar_t *bytes;
4015 	struct ether_addr sysaddr;
4016 	uint_t nelts;
4017 	int err;
4018 	boolean_t found = B_FALSE;
4019 
4020 	/*
4021 	 * The "vendor's factory-set address" may already have
4022 	 * been extracted from the chip, but if the property
4023 	 * "local-mac-address" is set we use that instead.
4024 	 *
4025 	 * We check whether it looks like an array of 6
4026 	 * bytes (which it should, if OBP set it).  If we can't
4027 	 * make sense of it this way, we'll ignore it.
4028 	 */
4029 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip,
4030 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
4031 	if (err == DDI_PROP_SUCCESS) {
4032 		if (nelts == ETHERADDRL) {
4033 			while (nelts--)
4034 				hw->mac.addr[nelts] = bytes[nelts];
4035 			found = B_TRUE;
4036 		}
4037 		ddi_prop_free(bytes);
4038 	}
4039 
4040 	/*
4041 	 * Look up the OBP property "local-mac-address?". If the user has set
4042 	 * 'local-mac-address? = false', use "the system address" instead.
4043 	 */
4044 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip, 0,
4045 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
4046 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
4047 			if (localetheraddr(NULL, &sysaddr) != 0) {
4048 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
4049 				found = B_TRUE;
4050 			}
4051 		}
4052 		ddi_prop_free(bytes);
4053 	}
4054 
4055 	/*
4056 	 * Finally(!), if there's a valid "mac-address" property (created
4057 	 * if we netbooted from this interface), we must use this instead
4058 	 * of any of the above to ensure that the NFS/install server doesn't
4059 	 * get confused by the address changing as illumos takes over!
4060 	 */
4061 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip,
4062 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
4063 	if (err == DDI_PROP_SUCCESS) {
4064 		if (nelts == ETHERADDRL) {
4065 			while (nelts--)
4066 				hw->mac.addr[nelts] = bytes[nelts];
4067 			found = B_TRUE;
4068 		}
4069 		ddi_prop_free(bytes);
4070 	}
4071 
4072 	if (found) {
4073 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
4074 		return (B_TRUE);
4075 	}
4076 #else
4077 	_NOTE(ARGUNUSED(ixgbe));
4078 #endif
4079 
4080 	return (B_TRUE);
4081 }
4082 
4083 #pragma inline(ixgbe_arm_watchdog_timer)
4084 static void
4085 ixgbe_arm_watchdog_timer(ixgbe_t *ixgbe)
4086 {
4087 	/*
4088 	 * Fire a watchdog timer
4089 	 */
4090 	ixgbe->watchdog_tid =
4091 	    timeout(ixgbe_local_timer,
4092 	    (void *)ixgbe, 1 * drv_usectohz(1000000));
4093 
4094 }
4095 
4096 /*
4097  * ixgbe_enable_watchdog_timer - Enable and start the driver watchdog timer.
4098  */
4099 void
4100 ixgbe_enable_watchdog_timer(ixgbe_t *ixgbe)
4101 {
4102 	mutex_enter(&ixgbe->watchdog_lock);
4103 
4104 	if (!ixgbe->watchdog_enable) {
4105 		ixgbe->watchdog_enable = B_TRUE;
4106 		ixgbe->watchdog_start = B_TRUE;
4107 		ixgbe_arm_watchdog_timer(ixgbe);
4108 	}
4109 
4110 	mutex_exit(&ixgbe->watchdog_lock);
4111 }
4112 
4113 /*
4114  * ixgbe_disable_watchdog_timer - Disable and stop the driver watchdog timer.
4115  */
4116 void
4117 ixgbe_disable_watchdog_timer(ixgbe_t *ixgbe)
4118 {
4119 	timeout_id_t tid;
4120 
4121 	mutex_enter(&ixgbe->watchdog_lock);
4122 
4123 	ixgbe->watchdog_enable = B_FALSE;
4124 	ixgbe->watchdog_start = B_FALSE;
4125 	tid = ixgbe->watchdog_tid;
4126 	ixgbe->watchdog_tid = 0;
4127 
4128 	mutex_exit(&ixgbe->watchdog_lock);
4129 
4130 	if (tid != 0)
4131 		(void) untimeout(tid);
4132 }
4133 
4134 /*
4135  * ixgbe_start_watchdog_timer - Start the driver watchdog timer.
4136  */
4137 void
4138 ixgbe_start_watchdog_timer(ixgbe_t *ixgbe)
4139 {
4140 	mutex_enter(&ixgbe->watchdog_lock);
4141 
4142 	if (ixgbe->watchdog_enable) {
4143 		if (!ixgbe->watchdog_start) {
4144 			ixgbe->watchdog_start = B_TRUE;
4145 			ixgbe_arm_watchdog_timer(ixgbe);
4146 		}
4147 	}
4148 
4149 	mutex_exit(&ixgbe->watchdog_lock);
4150 }
4151 
4152 /*
4153  * ixgbe_restart_watchdog_timer - Restart the driver watchdog timer.
4154  */
4155 static void
4156 ixgbe_restart_watchdog_timer(ixgbe_t *ixgbe)
4157 {
4158 	mutex_enter(&ixgbe->watchdog_lock);
4159 
4160 	if (ixgbe->watchdog_start)
4161 		ixgbe_arm_watchdog_timer(ixgbe);
4162 
4163 	mutex_exit(&ixgbe->watchdog_lock);
4164 }
4165 
4166 /*
4167  * ixgbe_stop_watchdog_timer - Stop the driver watchdog timer.
4168  */
4169 void
4170 ixgbe_stop_watchdog_timer(ixgbe_t *ixgbe)
4171 {
4172 	timeout_id_t tid;
4173 
4174 	mutex_enter(&ixgbe->watchdog_lock);
4175 
4176 	ixgbe->watchdog_start = B_FALSE;
4177 	tid = ixgbe->watchdog_tid;
4178 	ixgbe->watchdog_tid = 0;
4179 
4180 	mutex_exit(&ixgbe->watchdog_lock);
4181 
4182 	if (tid != 0)
4183 		(void) untimeout(tid);
4184 }
4185 
4186 /*
4187  * ixgbe_disable_adapter_interrupts - Disable all adapter interrupts.
4188  */
4189 static void
4190 ixgbe_disable_adapter_interrupts(ixgbe_t *ixgbe)
4191 {
4192 	struct ixgbe_hw *hw = &ixgbe->hw;
4193 
4194 	/*
4195 	 * mask all interrupts off
4196 	 */
4197 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, 0xffffffff);
4198 
4199 	/*
4200 	 * for MSI-X, also disable autoclear
4201 	 */
4202 	if (ixgbe->intr_type == DDI_INTR_TYPE_MSIX) {
4203 		IXGBE_WRITE_REG(hw, IXGBE_EIAC, 0x0);
4204 	}
4205 
4206 	IXGBE_WRITE_FLUSH(hw);
4207 }
4208 
4209 /*
4210  * ixgbe_enable_adapter_interrupts - Enable all hardware interrupts.
4211  */
4212 static void
4213 ixgbe_enable_adapter_interrupts(ixgbe_t *ixgbe)
4214 {
4215 	struct ixgbe_hw *hw = &ixgbe->hw;
4216 	uint32_t eiac, eiam;
4217 	uint32_t gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
4218 
4219 	/* interrupt types to enable */
4220 	ixgbe->eims = IXGBE_EIMS_ENABLE_MASK;	/* shared code default */
4221 	ixgbe->eims &= ~IXGBE_EIMS_TCP_TIMER;	/* minus tcp timer */
4222 	ixgbe->eims |= ixgbe->capab->other_intr; /* "other" interrupt types */
4223 
4224 	/* enable automask on "other" causes that this adapter can generate */
4225 	eiam = ixgbe->capab->other_intr;
4226 
4227 	/*
4228 	 * msi-x mode
4229 	 */
4230 	if (ixgbe->intr_type == DDI_INTR_TYPE_MSIX) {
4231 		/* enable autoclear but not on bits 29:20 */
4232 		eiac = (ixgbe->eims & ~IXGBE_OTHER_INTR);
4233 
4234 		/* general purpose interrupt enable */
4235 		gpie |= (IXGBE_GPIE_MSIX_MODE
4236 		    | IXGBE_GPIE_PBA_SUPPORT
4237 		    | IXGBE_GPIE_OCD
4238 		    | IXGBE_GPIE_EIAME);
4239 	/*
4240 	 * non-msi-x mode
4241 	 */
4242 	} else {
4243 
4244 		/* disable autoclear, leave gpie at default */
4245 		eiac = 0;
4246 
4247 		/*
4248 		 * General purpose interrupt enable.
4249 		 * For 82599, X540 and X550, extended interrupt
4250 		 * automask enable only in MSI or MSI-X mode
4251 		 */
4252 		if ((hw->mac.type == ixgbe_mac_82598EB) ||
4253 		    (ixgbe->intr_type == DDI_INTR_TYPE_MSI)) {
4254 			gpie |= IXGBE_GPIE_EIAME;
4255 		}
4256 	}
4257 
4258 	/* Enable specific "other" interrupt types */
4259 	switch (hw->mac.type) {
4260 	case ixgbe_mac_82598EB:
4261 		gpie |= ixgbe->capab->other_gpie;
4262 		break;
4263 
4264 	case ixgbe_mac_82599EB:
4265 	case ixgbe_mac_X540:
4266 	case ixgbe_mac_X550:
4267 	case ixgbe_mac_X550EM_x:
4268 		gpie |= ixgbe->capab->other_gpie;
4269 
4270 		/* Enable RSC Delay 8us when LRO enabled  */
4271 		if (ixgbe->lro_enable) {
4272 			gpie |= (1 << IXGBE_GPIE_RSC_DELAY_SHIFT);
4273 		}
4274 		break;
4275 
4276 	default:
4277 		break;
4278 	}
4279 
4280 	/* write to interrupt control registers */
4281 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4282 	IXGBE_WRITE_REG(hw, IXGBE_EIAC, eiac);
4283 	IXGBE_WRITE_REG(hw, IXGBE_EIAM, eiam);
4284 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4285 	IXGBE_WRITE_FLUSH(hw);
4286 }
4287 
4288 /*
4289  * ixgbe_loopback_ioctl - Loopback support.
4290  */
4291 enum ioc_reply
4292 ixgbe_loopback_ioctl(ixgbe_t *ixgbe, struct iocblk *iocp, mblk_t *mp)
4293 {
4294 	lb_info_sz_t *lbsp;
4295 	lb_property_t *lbpp;
4296 	uint32_t *lbmp;
4297 	uint32_t size;
4298 	uint32_t value;
4299 
4300 	if (mp->b_cont == NULL)
4301 		return (IOC_INVAL);
4302 
4303 	switch (iocp->ioc_cmd) {
4304 	default:
4305 		return (IOC_INVAL);
4306 
4307 	case LB_GET_INFO_SIZE:
4308 		size = sizeof (lb_info_sz_t);
4309 		if (iocp->ioc_count != size)
4310 			return (IOC_INVAL);
4311 
4312 		value = sizeof (lb_normal);
4313 		value += sizeof (lb_mac);
4314 		value += sizeof (lb_external);
4315 
4316 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
4317 		*lbsp = value;
4318 		break;
4319 
4320 	case LB_GET_INFO:
4321 		value = sizeof (lb_normal);
4322 		value += sizeof (lb_mac);
4323 		value += sizeof (lb_external);
4324 
4325 		size = value;
4326 		if (iocp->ioc_count != size)
4327 			return (IOC_INVAL);
4328 
4329 		value = 0;
4330 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
4331 
4332 		lbpp[value++] = lb_normal;
4333 		lbpp[value++] = lb_mac;
4334 		lbpp[value++] = lb_external;
4335 		break;
4336 
4337 	case LB_GET_MODE:
4338 		size = sizeof (uint32_t);
4339 		if (iocp->ioc_count != size)
4340 			return (IOC_INVAL);
4341 
4342 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
4343 		*lbmp = ixgbe->loopback_mode;
4344 		break;
4345 
4346 	case LB_SET_MODE:
4347 		size = 0;
4348 		if (iocp->ioc_count != sizeof (uint32_t))
4349 			return (IOC_INVAL);
4350 
4351 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
4352 		if (!ixgbe_set_loopback_mode(ixgbe, *lbmp))
4353 			return (IOC_INVAL);
4354 		break;
4355 	}
4356 
4357 	iocp->ioc_count = size;
4358 	iocp->ioc_error = 0;
4359 
4360 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
4361 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4362 		return (IOC_INVAL);
4363 	}
4364 
4365 	return (IOC_REPLY);
4366 }
4367 
4368 /*
4369  * ixgbe_set_loopback_mode - Setup loopback based on the loopback mode.
4370  */
4371 static boolean_t
4372 ixgbe_set_loopback_mode(ixgbe_t *ixgbe, uint32_t mode)
4373 {
4374 	if (mode == ixgbe->loopback_mode)
4375 		return (B_TRUE);
4376 
4377 	ixgbe->loopback_mode = mode;
4378 
4379 	if (mode == IXGBE_LB_NONE) {
4380 		/*
4381 		 * Reset the chip
4382 		 */
4383 		(void) ixgbe_reset(ixgbe);
4384 		return (B_TRUE);
4385 	}
4386 
4387 	mutex_enter(&ixgbe->gen_lock);
4388 
4389 	switch (mode) {
4390 	default:
4391 		mutex_exit(&ixgbe->gen_lock);
4392 		return (B_FALSE);
4393 
4394 	case IXGBE_LB_EXTERNAL:
4395 		break;
4396 
4397 	case IXGBE_LB_INTERNAL_MAC:
4398 		ixgbe_set_internal_mac_loopback(ixgbe);
4399 		break;
4400 	}
4401 
4402 	mutex_exit(&ixgbe->gen_lock);
4403 
4404 	return (B_TRUE);
4405 }
4406 
4407 /*
4408  * ixgbe_set_internal_mac_loopback - Set the internal MAC loopback mode.
4409  */
4410 static void
4411 ixgbe_set_internal_mac_loopback(ixgbe_t *ixgbe)
4412 {
4413 	struct ixgbe_hw *hw;
4414 	uint32_t reg;
4415 	uint8_t atlas;
4416 
4417 	hw = &ixgbe->hw;
4418 
4419 	/*
4420 	 * Setup MAC loopback
4421 	 */
4422 	reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_HLREG0);
4423 	reg |= IXGBE_HLREG0_LPBK;
4424 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_HLREG0, reg);
4425 
4426 	reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_AUTOC);
4427 	reg &= ~IXGBE_AUTOC_LMS_MASK;
4428 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_AUTOC, reg);
4429 
4430 	/*
4431 	 * Disable Atlas Tx lanes to keep packets in loopback and not on wire
4432 	 */
4433 	switch (hw->mac.type) {
4434 	case ixgbe_mac_82598EB:
4435 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_LPBK,
4436 		    &atlas);
4437 		atlas |= IXGBE_ATLAS_PDN_TX_REG_EN;
4438 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_LPBK,
4439 		    atlas);
4440 
4441 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_10G,
4442 		    &atlas);
4443 		atlas |= IXGBE_ATLAS_PDN_TX_10G_QL_ALL;
4444 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_10G,
4445 		    atlas);
4446 
4447 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_1G,
4448 		    &atlas);
4449 		atlas |= IXGBE_ATLAS_PDN_TX_1G_QL_ALL;
4450 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_1G,
4451 		    atlas);
4452 
4453 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_AN,
4454 		    &atlas);
4455 		atlas |= IXGBE_ATLAS_PDN_TX_AN_QL_ALL;
4456 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_AN,
4457 		    atlas);
4458 		break;
4459 
4460 	case ixgbe_mac_82599EB:
4461 	case ixgbe_mac_X540:
4462 	case ixgbe_mac_X550:
4463 	case ixgbe_mac_X550EM_x:
4464 		reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_AUTOC);
4465 		reg |= (IXGBE_AUTOC_FLU |
4466 		    IXGBE_AUTOC_10G_KX4);
4467 		IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_AUTOC, reg);
4468 
4469 		(void) ixgbe_setup_link(&ixgbe->hw, IXGBE_LINK_SPEED_10GB_FULL,
4470 		    B_FALSE);
4471 		break;
4472 
4473 	default:
4474 		break;
4475 	}
4476 }
4477 
4478 #pragma inline(ixgbe_intr_rx_work)
4479 /*
4480  * ixgbe_intr_rx_work - RX processing of ISR.
4481  */
4482 static void
4483 ixgbe_intr_rx_work(ixgbe_rx_ring_t *rx_ring)
4484 {
4485 	mblk_t *mp;
4486 
4487 	mutex_enter(&rx_ring->rx_lock);
4488 
4489 	mp = ixgbe_ring_rx(rx_ring, IXGBE_POLL_NULL);
4490 	mutex_exit(&rx_ring->rx_lock);
4491 
4492 	if (mp != NULL)
4493 		mac_rx_ring(rx_ring->ixgbe->mac_hdl, rx_ring->ring_handle, mp,
4494 		    rx_ring->ring_gen_num);
4495 }
4496 
4497 #pragma inline(ixgbe_intr_tx_work)
4498 /*
4499  * ixgbe_intr_tx_work - TX processing of ISR.
4500  */
4501 static void
4502 ixgbe_intr_tx_work(ixgbe_tx_ring_t *tx_ring)
4503 {
4504 	ixgbe_t *ixgbe = tx_ring->ixgbe;
4505 
4506 	/*
4507 	 * Recycle the tx descriptors
4508 	 */
4509 	tx_ring->tx_recycle(tx_ring);
4510 
4511 	/*
4512 	 * Schedule the re-transmit
4513 	 */
4514 	if (tx_ring->reschedule &&
4515 	    (tx_ring->tbd_free >= ixgbe->tx_resched_thresh)) {
4516 		tx_ring->reschedule = B_FALSE;
4517 		mac_tx_ring_update(tx_ring->ixgbe->mac_hdl,
4518 		    tx_ring->ring_handle);
4519 		IXGBE_DEBUG_STAT(tx_ring->stat_reschedule);
4520 	}
4521 }
4522 
4523 #pragma inline(ixgbe_intr_other_work)
4524 /*
4525  * ixgbe_intr_other_work - Process interrupt types other than tx/rx
4526  */
4527 static void
4528 ixgbe_intr_other_work(ixgbe_t *ixgbe, uint32_t eicr)
4529 {
4530 	struct ixgbe_hw *hw = &ixgbe->hw;
4531 
4532 	ASSERT(mutex_owned(&ixgbe->gen_lock));
4533 
4534 	/*
4535 	 * handle link status change
4536 	 */
4537 	if (eicr & IXGBE_EICR_LSC) {
4538 		ixgbe_driver_link_check(ixgbe);
4539 		ixgbe_get_hw_state(ixgbe);
4540 	}
4541 
4542 	/*
4543 	 * check for fan failure on adapters with fans
4544 	 */
4545 	if ((ixgbe->capab->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
4546 	    (eicr & IXGBE_EICR_GPI_SDP1)) {
4547 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_OVERTEMP);
4548 
4549 		/*
4550 		 * Disable the adapter interrupts
4551 		 */
4552 		ixgbe_disable_adapter_interrupts(ixgbe);
4553 
4554 		/*
4555 		 * Disable Rx/Tx units
4556 		 */
4557 		(void) ixgbe_stop_adapter(&ixgbe->hw);
4558 
4559 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
4560 		ixgbe_error(ixgbe,
4561 		    "Problem: Network adapter has been stopped "
4562 		    "because the fan has stopped.\n");
4563 		ixgbe_error(ixgbe,
4564 		    "Action: Replace the adapter.\n");
4565 
4566 		/* re-enable the interrupt, which was automasked */
4567 		ixgbe->eims |= IXGBE_EICR_GPI_SDP1;
4568 	}
4569 
4570 	/*
4571 	 * Do SFP check for adapters with hot-plug capability
4572 	 */
4573 	if ((ixgbe->capab->flags & IXGBE_FLAG_SFP_PLUG_CAPABLE) &&
4574 	    ((eicr & IXGBE_EICR_GPI_SDP1_BY_MAC(hw)) ||
4575 	    (eicr & IXGBE_EICR_GPI_SDP2_BY_MAC(hw)))) {
4576 		ixgbe->eicr = eicr;
4577 		if ((ddi_taskq_dispatch(ixgbe->sfp_taskq,
4578 		    ixgbe_sfp_check, (void *)ixgbe,
4579 		    DDI_NOSLEEP)) != DDI_SUCCESS) {
4580 			ixgbe_log(ixgbe, "No memory available to dispatch "
4581 			    "taskq for SFP check");
4582 		}
4583 	}
4584 
4585 	/*
4586 	 * Do over-temperature check for adapters with temp sensor
4587 	 */
4588 	if ((ixgbe->capab->flags & IXGBE_FLAG_TEMP_SENSOR_CAPABLE) &&
4589 	    ((eicr & IXGBE_EICR_GPI_SDP0_BY_MAC(hw)) ||
4590 	    (eicr & IXGBE_EICR_LSC))) {
4591 		ixgbe->eicr = eicr;
4592 		if ((ddi_taskq_dispatch(ixgbe->overtemp_taskq,
4593 		    ixgbe_overtemp_check, (void *)ixgbe,
4594 		    DDI_NOSLEEP)) != DDI_SUCCESS) {
4595 			ixgbe_log(ixgbe, "No memory available to dispatch "
4596 			    "taskq for overtemp check");
4597 		}
4598 	}
4599 
4600 	/*
4601 	 * Process an external PHY interrupt
4602 	 */
4603 	if (hw->device_id == IXGBE_DEV_ID_X550EM_X_10G_T &&
4604 	    (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
4605 		ixgbe->eicr = eicr;
4606 		if ((ddi_taskq_dispatch(ixgbe->phy_taskq,
4607 		    ixgbe_phy_check, (void *)ixgbe,
4608 		    DDI_NOSLEEP)) != DDI_SUCCESS) {
4609 			ixgbe_log(ixgbe, "No memory available to dispatch "
4610 			    "taskq for PHY check");
4611 		}
4612 	}
4613 }
4614 
4615 /*
4616  * ixgbe_intr_legacy - Interrupt handler for legacy interrupts.
4617  */
4618 static uint_t
4619 ixgbe_intr_legacy(void *arg1, void *arg2)
4620 {
4621 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
4622 	struct ixgbe_hw *hw = &ixgbe->hw;
4623 	ixgbe_tx_ring_t *tx_ring;
4624 	ixgbe_rx_ring_t *rx_ring;
4625 	uint32_t eicr;
4626 	mblk_t *mp;
4627 	boolean_t tx_reschedule;
4628 	uint_t result;
4629 
4630 	_NOTE(ARGUNUSED(arg2));
4631 
4632 	mutex_enter(&ixgbe->gen_lock);
4633 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
4634 		mutex_exit(&ixgbe->gen_lock);
4635 		return (DDI_INTR_UNCLAIMED);
4636 	}
4637 
4638 	mp = NULL;
4639 	tx_reschedule = B_FALSE;
4640 
4641 	/*
4642 	 * Any bit set in eicr: claim this interrupt
4643 	 */
4644 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4645 
4646 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
4647 		mutex_exit(&ixgbe->gen_lock);
4648 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4649 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4650 		return (DDI_INTR_CLAIMED);
4651 	}
4652 
4653 	if (eicr) {
4654 		/*
4655 		 * For legacy interrupt, we have only one interrupt,
4656 		 * so we have only one rx ring and one tx ring enabled.
4657 		 */
4658 		ASSERT(ixgbe->num_rx_rings == 1);
4659 		ASSERT(ixgbe->num_tx_rings == 1);
4660 
4661 		/*
4662 		 * For legacy interrupt, rx rings[0] will use RTxQ[0].
4663 		 */
4664 		if (eicr & 0x1) {
4665 			ixgbe->eimc |= IXGBE_EICR_RTX_QUEUE;
4666 			IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4667 			ixgbe->eims |= IXGBE_EICR_RTX_QUEUE;
4668 			/*
4669 			 * Clean the rx descriptors
4670 			 */
4671 			rx_ring = &ixgbe->rx_rings[0];
4672 			mp = ixgbe_ring_rx(rx_ring, IXGBE_POLL_NULL);
4673 		}
4674 
4675 		/*
4676 		 * For legacy interrupt, tx rings[0] will use RTxQ[1].
4677 		 */
4678 		if (eicr & 0x2) {
4679 			/*
4680 			 * Recycle the tx descriptors
4681 			 */
4682 			tx_ring = &ixgbe->tx_rings[0];
4683 			tx_ring->tx_recycle(tx_ring);
4684 
4685 			/*
4686 			 * Schedule the re-transmit
4687 			 */
4688 			tx_reschedule = (tx_ring->reschedule &&
4689 			    (tx_ring->tbd_free >= ixgbe->tx_resched_thresh));
4690 		}
4691 
4692 		/* any interrupt type other than tx/rx */
4693 		if (eicr & ixgbe->capab->other_intr) {
4694 			switch (hw->mac.type) {
4695 			case ixgbe_mac_82598EB:
4696 				ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4697 				break;
4698 
4699 			case ixgbe_mac_82599EB:
4700 			case ixgbe_mac_X540:
4701 			case ixgbe_mac_X550:
4702 			case ixgbe_mac_X550EM_x:
4703 				ixgbe->eimc = IXGBE_82599_OTHER_INTR;
4704 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4705 				break;
4706 
4707 			default:
4708 				break;
4709 			}
4710 			ixgbe_intr_other_work(ixgbe, eicr);
4711 			ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4712 		}
4713 
4714 		mutex_exit(&ixgbe->gen_lock);
4715 
4716 		result = DDI_INTR_CLAIMED;
4717 	} else {
4718 		mutex_exit(&ixgbe->gen_lock);
4719 
4720 		/*
4721 		 * No interrupt cause bits set: don't claim this interrupt.
4722 		 */
4723 		result = DDI_INTR_UNCLAIMED;
4724 	}
4725 
4726 	/* re-enable the interrupts which were automasked */
4727 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4728 
4729 	/*
4730 	 * Do the following work outside of the gen_lock
4731 	 */
4732 	if (mp != NULL) {
4733 		mac_rx_ring(rx_ring->ixgbe->mac_hdl, rx_ring->ring_handle, mp,
4734 		    rx_ring->ring_gen_num);
4735 	}
4736 
4737 	if (tx_reschedule)  {
4738 		tx_ring->reschedule = B_FALSE;
4739 		mac_tx_ring_update(ixgbe->mac_hdl, tx_ring->ring_handle);
4740 		IXGBE_DEBUG_STAT(tx_ring->stat_reschedule);
4741 	}
4742 
4743 	return (result);
4744 }
4745 
4746 /*
4747  * ixgbe_intr_msi - Interrupt handler for MSI.
4748  */
4749 static uint_t
4750 ixgbe_intr_msi(void *arg1, void *arg2)
4751 {
4752 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
4753 	struct ixgbe_hw *hw = &ixgbe->hw;
4754 	uint32_t eicr;
4755 
4756 	_NOTE(ARGUNUSED(arg2));
4757 
4758 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4759 
4760 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
4761 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4762 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4763 		return (DDI_INTR_CLAIMED);
4764 	}
4765 
4766 	/*
4767 	 * For MSI interrupt, we have only one vector,
4768 	 * so we have only one rx ring and one tx ring enabled.
4769 	 */
4770 	ASSERT(ixgbe->num_rx_rings == 1);
4771 	ASSERT(ixgbe->num_tx_rings == 1);
4772 
4773 	/*
4774 	 * For MSI interrupt, rx rings[0] will use RTxQ[0].
4775 	 */
4776 	if (eicr & 0x1) {
4777 		ixgbe_intr_rx_work(&ixgbe->rx_rings[0]);
4778 	}
4779 
4780 	/*
4781 	 * For MSI interrupt, tx rings[0] will use RTxQ[1].
4782 	 */
4783 	if (eicr & 0x2) {
4784 		ixgbe_intr_tx_work(&ixgbe->tx_rings[0]);
4785 	}
4786 
4787 	/* any interrupt type other than tx/rx */
4788 	if (eicr & ixgbe->capab->other_intr) {
4789 		mutex_enter(&ixgbe->gen_lock);
4790 		switch (hw->mac.type) {
4791 		case ixgbe_mac_82598EB:
4792 			ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4793 			break;
4794 
4795 		case ixgbe_mac_82599EB:
4796 		case ixgbe_mac_X540:
4797 		case ixgbe_mac_X550:
4798 		case ixgbe_mac_X550EM_x:
4799 			ixgbe->eimc = IXGBE_82599_OTHER_INTR;
4800 			IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4801 			break;
4802 
4803 		default:
4804 			break;
4805 		}
4806 		ixgbe_intr_other_work(ixgbe, eicr);
4807 		ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4808 		mutex_exit(&ixgbe->gen_lock);
4809 	}
4810 
4811 	/* re-enable the interrupts which were automasked */
4812 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4813 
4814 	return (DDI_INTR_CLAIMED);
4815 }
4816 
4817 /*
4818  * ixgbe_intr_msix - Interrupt handler for MSI-X.
4819  */
4820 static uint_t
4821 ixgbe_intr_msix(void *arg1, void *arg2)
4822 {
4823 	ixgbe_intr_vector_t *vect = (ixgbe_intr_vector_t *)arg1;
4824 	ixgbe_t *ixgbe = vect->ixgbe;
4825 	struct ixgbe_hw *hw = &ixgbe->hw;
4826 	uint32_t eicr;
4827 	int r_idx = 0;
4828 
4829 	_NOTE(ARGUNUSED(arg2));
4830 
4831 	/*
4832 	 * Clean each rx ring that has its bit set in the map
4833 	 */
4834 	r_idx = bt_getlowbit(vect->rx_map, 0, (ixgbe->num_rx_rings - 1));
4835 	while (r_idx >= 0) {
4836 		ixgbe_intr_rx_work(&ixgbe->rx_rings[r_idx]);
4837 		r_idx = bt_getlowbit(vect->rx_map, (r_idx + 1),
4838 		    (ixgbe->num_rx_rings - 1));
4839 	}
4840 
4841 	/*
4842 	 * Clean each tx ring that has its bit set in the map
4843 	 */
4844 	r_idx = bt_getlowbit(vect->tx_map, 0, (ixgbe->num_tx_rings - 1));
4845 	while (r_idx >= 0) {
4846 		ixgbe_intr_tx_work(&ixgbe->tx_rings[r_idx]);
4847 		r_idx = bt_getlowbit(vect->tx_map, (r_idx + 1),
4848 		    (ixgbe->num_tx_rings - 1));
4849 	}
4850 
4851 
4852 	/*
4853 	 * Clean other interrupt (link change) that has its bit set in the map
4854 	 */
4855 	if (BT_TEST(vect->other_map, 0) == 1) {
4856 		eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4857 
4858 		if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) !=
4859 		    DDI_FM_OK) {
4860 			ddi_fm_service_impact(ixgbe->dip,
4861 			    DDI_SERVICE_DEGRADED);
4862 			atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4863 			return (DDI_INTR_CLAIMED);
4864 		}
4865 
4866 		/*
4867 		 * Check "other" cause bits: any interrupt type other than tx/rx
4868 		 */
4869 		if (eicr & ixgbe->capab->other_intr) {
4870 			mutex_enter(&ixgbe->gen_lock);
4871 			switch (hw->mac.type) {
4872 			case ixgbe_mac_82598EB:
4873 				ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4874 				ixgbe_intr_other_work(ixgbe, eicr);
4875 				break;
4876 
4877 			case ixgbe_mac_82599EB:
4878 			case ixgbe_mac_X540:
4879 			case ixgbe_mac_X550:
4880 			case ixgbe_mac_X550EM_x:
4881 				ixgbe->eims |= IXGBE_EICR_RTX_QUEUE;
4882 				ixgbe_intr_other_work(ixgbe, eicr);
4883 				break;
4884 
4885 			default:
4886 				break;
4887 			}
4888 			mutex_exit(&ixgbe->gen_lock);
4889 		}
4890 
4891 		/* re-enable the interrupts which were automasked */
4892 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4893 	}
4894 
4895 	return (DDI_INTR_CLAIMED);
4896 }
4897 
4898 /*
4899  * ixgbe_alloc_intrs - Allocate interrupts for the driver.
4900  *
4901  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
4902  * if not successful, try Legacy.
4903  * ixgbe->intr_force can be used to force sequence to start with
4904  * any of the 3 types.
4905  * If MSI-X is not used, number of tx/rx rings is forced to 1.
4906  */
4907 static int
4908 ixgbe_alloc_intrs(ixgbe_t *ixgbe)
4909 {
4910 	dev_info_t *devinfo;
4911 	int intr_types;
4912 	int rc;
4913 
4914 	devinfo = ixgbe->dip;
4915 
4916 	/*
4917 	 * Get supported interrupt types
4918 	 */
4919 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
4920 
4921 	if (rc != DDI_SUCCESS) {
4922 		ixgbe_log(ixgbe,
4923 		    "Get supported interrupt types failed: %d", rc);
4924 		return (IXGBE_FAILURE);
4925 	}
4926 	IXGBE_DEBUGLOG_1(ixgbe, "Supported interrupt types: %x", intr_types);
4927 
4928 	ixgbe->intr_type = 0;
4929 
4930 	/*
4931 	 * Install MSI-X interrupts
4932 	 */
4933 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
4934 	    (ixgbe->intr_force <= IXGBE_INTR_MSIX)) {
4935 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_MSIX);
4936 		if (rc == IXGBE_SUCCESS)
4937 			return (IXGBE_SUCCESS);
4938 
4939 		ixgbe_log(ixgbe,
4940 		    "Allocate MSI-X failed, trying MSI interrupts...");
4941 	}
4942 
4943 	/*
4944 	 * MSI-X not used, force rings and groups to 1
4945 	 */
4946 	ixgbe->num_rx_rings = 1;
4947 	ixgbe->num_rx_groups = 1;
4948 	ixgbe->num_tx_rings = 1;
4949 	ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
4950 	ixgbe_log(ixgbe,
4951 	    "MSI-X not used, force rings and groups number to 1");
4952 
4953 	/*
4954 	 * Install MSI interrupts
4955 	 */
4956 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
4957 	    (ixgbe->intr_force <= IXGBE_INTR_MSI)) {
4958 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_MSI);
4959 		if (rc == IXGBE_SUCCESS)
4960 			return (IXGBE_SUCCESS);
4961 
4962 		ixgbe_log(ixgbe,
4963 		    "Allocate MSI failed, trying Legacy interrupts...");
4964 	}
4965 
4966 	/*
4967 	 * Install legacy interrupts
4968 	 */
4969 	if (intr_types & DDI_INTR_TYPE_FIXED) {
4970 		/*
4971 		 * Disallow legacy interrupts for X550. X550 has a silicon
4972 		 * bug which prevents Shared Legacy interrupts from working.
4973 		 * For details, please reference:
4974 		 *
4975 		 * Intel Ethernet Controller X550 Specification Update rev. 2.1
4976 		 * May 2016, erratum 22: PCIe Interrupt Status Bit
4977 		 */
4978 		if (ixgbe->hw.mac.type == ixgbe_mac_X550 ||
4979 		    ixgbe->hw.mac.type == ixgbe_mac_X550EM_x ||
4980 		    ixgbe->hw.mac.type == ixgbe_mac_X550_vf ||
4981 		    ixgbe->hw.mac.type == ixgbe_mac_X550EM_x_vf) {
4982 			ixgbe_log(ixgbe,
4983 			    "Legacy interrupts are not supported on this "
4984 			    "adapter. Please use MSI or MSI-X instead.");
4985 			return (IXGBE_FAILURE);
4986 		}
4987 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_FIXED);
4988 		if (rc == IXGBE_SUCCESS)
4989 			return (IXGBE_SUCCESS);
4990 
4991 		ixgbe_log(ixgbe,
4992 		    "Allocate Legacy interrupts failed");
4993 	}
4994 
4995 	/*
4996 	 * If none of the 3 types succeeded, return failure
4997 	 */
4998 	return (IXGBE_FAILURE);
4999 }
5000 
5001 /*
5002  * ixgbe_alloc_intr_handles - Allocate interrupt handles.
5003  *
5004  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
5005  * if fewer than 2 handles are available, return failure.
5006  * Upon success, this maps the vectors to rx and tx rings for
5007  * interrupts.
5008  */
5009 static int
5010 ixgbe_alloc_intr_handles(ixgbe_t *ixgbe, int intr_type)
5011 {
5012 	dev_info_t *devinfo;
5013 	int request, count, actual;
5014 	int minimum;
5015 	int rc;
5016 	uint32_t ring_per_group;
5017 
5018 	devinfo = ixgbe->dip;
5019 
5020 	switch (intr_type) {
5021 	case DDI_INTR_TYPE_FIXED:
5022 		request = 1;	/* Request 1 legacy interrupt handle */
5023 		minimum = 1;
5024 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: legacy");
5025 		break;
5026 
5027 	case DDI_INTR_TYPE_MSI:
5028 		request = 1;	/* Request 1 MSI interrupt handle */
5029 		minimum = 1;
5030 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: MSI");
5031 		break;
5032 
5033 	case DDI_INTR_TYPE_MSIX:
5034 		/*
5035 		 * Best number of vectors for the adapter is
5036 		 * (# rx rings + # tx rings), however we will
5037 		 * limit the request number.
5038 		 */
5039 		request = min(16, ixgbe->num_rx_rings + ixgbe->num_tx_rings);
5040 		if (request > ixgbe->capab->max_ring_vect)
5041 			request = ixgbe->capab->max_ring_vect;
5042 		minimum = 1;
5043 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: MSI-X");
5044 		break;
5045 
5046 	default:
5047 		ixgbe_log(ixgbe,
5048 		    "invalid call to ixgbe_alloc_intr_handles(): %d\n",
5049 		    intr_type);
5050 		return (IXGBE_FAILURE);
5051 	}
5052 	IXGBE_DEBUGLOG_2(ixgbe, "interrupt handles requested: %d  minimum: %d",
5053 	    request, minimum);
5054 
5055 	/*
5056 	 * Get number of supported interrupts
5057 	 */
5058 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
5059 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
5060 		ixgbe_log(ixgbe,
5061 		    "Get interrupt number failed. Return: %d, count: %d",
5062 		    rc, count);
5063 		return (IXGBE_FAILURE);
5064 	}
5065 	IXGBE_DEBUGLOG_1(ixgbe, "interrupts supported: %d", count);
5066 
5067 	actual = 0;
5068 	ixgbe->intr_cnt = 0;
5069 	ixgbe->intr_cnt_max = 0;
5070 	ixgbe->intr_cnt_min = 0;
5071 
5072 	/*
5073 	 * Allocate an array of interrupt handles
5074 	 */
5075 	ixgbe->intr_size = request * sizeof (ddi_intr_handle_t);
5076 	ixgbe->htable = kmem_alloc(ixgbe->intr_size, KM_SLEEP);
5077 
5078 	rc = ddi_intr_alloc(devinfo, ixgbe->htable, intr_type, 0,
5079 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
5080 	if (rc != DDI_SUCCESS) {
5081 		ixgbe_log(ixgbe, "Allocate interrupts failed. "
5082 		    "return: %d, request: %d, actual: %d",
5083 		    rc, request, actual);
5084 		goto alloc_handle_fail;
5085 	}
5086 	IXGBE_DEBUGLOG_1(ixgbe, "interrupts actually allocated: %d", actual);
5087 
5088 	/*
5089 	 * upper/lower limit of interrupts
5090 	 */
5091 	ixgbe->intr_cnt = actual;
5092 	ixgbe->intr_cnt_max = request;
5093 	ixgbe->intr_cnt_min = minimum;
5094 
5095 	/*
5096 	 * rss number per group should not exceed the rx interrupt number,
5097 	 * else need to adjust rx ring number.
5098 	 */
5099 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
5100 	ASSERT((ixgbe->num_rx_rings % ixgbe->num_rx_groups) == 0);
5101 	if (actual < ring_per_group) {
5102 		ixgbe->num_rx_rings = ixgbe->num_rx_groups * actual;
5103 		ixgbe_setup_vmdq_rss_conf(ixgbe);
5104 	}
5105 
5106 	/*
5107 	 * Now we know the actual number of vectors.  Here we map the vector
5108 	 * to other, rx rings and tx ring.
5109 	 */
5110 	if (actual < minimum) {
5111 		ixgbe_log(ixgbe, "Insufficient interrupt handles available: %d",
5112 		    actual);
5113 		goto alloc_handle_fail;
5114 	}
5115 
5116 	/*
5117 	 * Get priority for first vector, assume remaining are all the same
5118 	 */
5119 	rc = ddi_intr_get_pri(ixgbe->htable[0], &ixgbe->intr_pri);
5120 	if (rc != DDI_SUCCESS) {
5121 		ixgbe_log(ixgbe,
5122 		    "Get interrupt priority failed: %d", rc);
5123 		goto alloc_handle_fail;
5124 	}
5125 
5126 	rc = ddi_intr_get_cap(ixgbe->htable[0], &ixgbe->intr_cap);
5127 	if (rc != DDI_SUCCESS) {
5128 		ixgbe_log(ixgbe,
5129 		    "Get interrupt cap failed: %d", rc);
5130 		goto alloc_handle_fail;
5131 	}
5132 
5133 	ixgbe->intr_type = intr_type;
5134 
5135 	return (IXGBE_SUCCESS);
5136 
5137 alloc_handle_fail:
5138 	ixgbe_rem_intrs(ixgbe);
5139 
5140 	return (IXGBE_FAILURE);
5141 }
5142 
5143 /*
5144  * ixgbe_add_intr_handlers - Add interrupt handlers based on the interrupt type.
5145  *
5146  * Before adding the interrupt handlers, the interrupt vectors have
5147  * been allocated, and the rx/tx rings have also been allocated.
5148  */
5149 static int
5150 ixgbe_add_intr_handlers(ixgbe_t *ixgbe)
5151 {
5152 	int vector = 0;
5153 	int rc;
5154 
5155 	switch (ixgbe->intr_type) {
5156 	case DDI_INTR_TYPE_MSIX:
5157 		/*
5158 		 * Add interrupt handler for all vectors
5159 		 */
5160 		for (vector = 0; vector < ixgbe->intr_cnt; vector++) {
5161 			/*
5162 			 * install pointer to vect_map[vector]
5163 			 */
5164 			rc = ddi_intr_add_handler(ixgbe->htable[vector],
5165 			    (ddi_intr_handler_t *)ixgbe_intr_msix,
5166 			    (void *)&ixgbe->vect_map[vector], NULL);
5167 
5168 			if (rc != DDI_SUCCESS) {
5169 				ixgbe_log(ixgbe,
5170 				    "Add interrupt handler failed. "
5171 				    "return: %d, vector: %d", rc, vector);
5172 				for (vector--; vector >= 0; vector--) {
5173 					(void) ddi_intr_remove_handler(
5174 					    ixgbe->htable[vector]);
5175 				}
5176 				return (IXGBE_FAILURE);
5177 			}
5178 		}
5179 
5180 		break;
5181 
5182 	case DDI_INTR_TYPE_MSI:
5183 		/*
5184 		 * Add interrupt handlers for the only vector
5185 		 */
5186 		rc = ddi_intr_add_handler(ixgbe->htable[vector],
5187 		    (ddi_intr_handler_t *)ixgbe_intr_msi,
5188 		    (void *)ixgbe, NULL);
5189 
5190 		if (rc != DDI_SUCCESS) {
5191 			ixgbe_log(ixgbe,
5192 			    "Add MSI interrupt handler failed: %d", rc);
5193 			return (IXGBE_FAILURE);
5194 		}
5195 
5196 		break;
5197 
5198 	case DDI_INTR_TYPE_FIXED:
5199 		/*
5200 		 * Add interrupt handlers for the only vector
5201 		 */
5202 		rc = ddi_intr_add_handler(ixgbe->htable[vector],
5203 		    (ddi_intr_handler_t *)ixgbe_intr_legacy,
5204 		    (void *)ixgbe, NULL);
5205 
5206 		if (rc != DDI_SUCCESS) {
5207 			ixgbe_log(ixgbe,
5208 			    "Add legacy interrupt handler failed: %d", rc);
5209 			return (IXGBE_FAILURE);
5210 		}
5211 
5212 		break;
5213 
5214 	default:
5215 		return (IXGBE_FAILURE);
5216 	}
5217 
5218 	return (IXGBE_SUCCESS);
5219 }
5220 
5221 #pragma inline(ixgbe_map_rxring_to_vector)
5222 /*
5223  * ixgbe_map_rxring_to_vector - Map given rx ring to given interrupt vector.
5224  */
5225 static void
5226 ixgbe_map_rxring_to_vector(ixgbe_t *ixgbe, int r_idx, int v_idx)
5227 {
5228 	/*
5229 	 * Set bit in map
5230 	 */
5231 	BT_SET(ixgbe->vect_map[v_idx].rx_map, r_idx);
5232 
5233 	/*
5234 	 * Count bits set
5235 	 */
5236 	ixgbe->vect_map[v_idx].rxr_cnt++;
5237 
5238 	/*
5239 	 * Remember bit position
5240 	 */
5241 	ixgbe->rx_rings[r_idx].intr_vector = v_idx;
5242 	ixgbe->rx_rings[r_idx].vect_bit = 1 << v_idx;
5243 }
5244 
5245 #pragma inline(ixgbe_map_txring_to_vector)
5246 /*
5247  * ixgbe_map_txring_to_vector - Map given tx ring to given interrupt vector.
5248  */
5249 static void
5250 ixgbe_map_txring_to_vector(ixgbe_t *ixgbe, int t_idx, int v_idx)
5251 {
5252 	/*
5253 	 * Set bit in map
5254 	 */
5255 	BT_SET(ixgbe->vect_map[v_idx].tx_map, t_idx);
5256 
5257 	/*
5258 	 * Count bits set
5259 	 */
5260 	ixgbe->vect_map[v_idx].txr_cnt++;
5261 
5262 	/*
5263 	 * Remember bit position
5264 	 */
5265 	ixgbe->tx_rings[t_idx].intr_vector = v_idx;
5266 	ixgbe->tx_rings[t_idx].vect_bit = 1 << v_idx;
5267 }
5268 
5269 /*
5270  * ixgbe_setup_ivar - Set the given entry in the given interrupt vector
5271  * allocation register (IVAR).
5272  * cause:
5273  *   -1 : other cause
5274  *    0 : rx
5275  *    1 : tx
5276  */
5277 static void
5278 ixgbe_setup_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, uint8_t msix_vector,
5279     int8_t cause)
5280 {
5281 	struct ixgbe_hw *hw = &ixgbe->hw;
5282 	u32 ivar, index;
5283 
5284 	switch (hw->mac.type) {
5285 	case ixgbe_mac_82598EB:
5286 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
5287 		if (cause == -1) {
5288 			cause = 0;
5289 		}
5290 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
5291 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
5292 		ivar &= ~(0xFF << (8 * (intr_alloc_entry & 0x3)));
5293 		ivar |= (msix_vector << (8 * (intr_alloc_entry & 0x3)));
5294 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
5295 		break;
5296 
5297 	case ixgbe_mac_82599EB:
5298 	case ixgbe_mac_X540:
5299 	case ixgbe_mac_X550:
5300 	case ixgbe_mac_X550EM_x:
5301 		if (cause == -1) {
5302 			/* other causes */
5303 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
5304 			index = (intr_alloc_entry & 1) * 8;
5305 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
5306 			ivar &= ~(0xFF << index);
5307 			ivar |= (msix_vector << index);
5308 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
5309 		} else {
5310 			/* tx or rx causes */
5311 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
5312 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
5313 			ivar = IXGBE_READ_REG(hw,
5314 			    IXGBE_IVAR(intr_alloc_entry >> 1));
5315 			ivar &= ~(0xFF << index);
5316 			ivar |= (msix_vector << index);
5317 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
5318 			    ivar);
5319 		}
5320 		break;
5321 
5322 	default:
5323 		break;
5324 	}
5325 }
5326 
5327 /*
5328  * ixgbe_enable_ivar - Enable the given entry by setting the VAL bit of
5329  * given interrupt vector allocation register (IVAR).
5330  * cause:
5331  *   -1 : other cause
5332  *    0 : rx
5333  *    1 : tx
5334  */
5335 static void
5336 ixgbe_enable_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, int8_t cause)
5337 {
5338 	struct ixgbe_hw *hw = &ixgbe->hw;
5339 	u32 ivar, index;
5340 
5341 	switch (hw->mac.type) {
5342 	case ixgbe_mac_82598EB:
5343 		if (cause == -1) {
5344 			cause = 0;
5345 		}
5346 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
5347 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
5348 		ivar |= (IXGBE_IVAR_ALLOC_VAL << (8 *
5349 		    (intr_alloc_entry & 0x3)));
5350 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
5351 		break;
5352 
5353 	case ixgbe_mac_82599EB:
5354 	case ixgbe_mac_X540:
5355 	case ixgbe_mac_X550:
5356 	case ixgbe_mac_X550EM_x:
5357 		if (cause == -1) {
5358 			/* other causes */
5359 			index = (intr_alloc_entry & 1) * 8;
5360 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
5361 			ivar |= (IXGBE_IVAR_ALLOC_VAL << index);
5362 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
5363 		} else {
5364 			/* tx or rx causes */
5365 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
5366 			ivar = IXGBE_READ_REG(hw,
5367 			    IXGBE_IVAR(intr_alloc_entry >> 1));
5368 			ivar |= (IXGBE_IVAR_ALLOC_VAL << index);
5369 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
5370 			    ivar);
5371 		}
5372 		break;
5373 
5374 	default:
5375 		break;
5376 	}
5377 }
5378 
5379 /*
5380  * ixgbe_disable_ivar - Disble the given entry by clearing the VAL bit of
5381  * given interrupt vector allocation register (IVAR).
5382  * cause:
5383  *   -1 : other cause
5384  *    0 : rx
5385  *    1 : tx
5386  */
5387 static void
5388 ixgbe_disable_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, int8_t cause)
5389 {
5390 	struct ixgbe_hw *hw = &ixgbe->hw;
5391 	u32 ivar, index;
5392 
5393 	switch (hw->mac.type) {
5394 	case ixgbe_mac_82598EB:
5395 		if (cause == -1) {
5396 			cause = 0;
5397 		}
5398 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
5399 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
5400 		ivar &= ~(IXGBE_IVAR_ALLOC_VAL<< (8 *
5401 		    (intr_alloc_entry & 0x3)));
5402 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
5403 		break;
5404 
5405 	case ixgbe_mac_82599EB:
5406 	case ixgbe_mac_X540:
5407 	case ixgbe_mac_X550:
5408 	case ixgbe_mac_X550EM_x:
5409 		if (cause == -1) {
5410 			/* other causes */
5411 			index = (intr_alloc_entry & 1) * 8;
5412 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
5413 			ivar &= ~(IXGBE_IVAR_ALLOC_VAL << index);
5414 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
5415 		} else {
5416 			/* tx or rx causes */
5417 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
5418 			ivar = IXGBE_READ_REG(hw,
5419 			    IXGBE_IVAR(intr_alloc_entry >> 1));
5420 			ivar &= ~(IXGBE_IVAR_ALLOC_VAL << index);
5421 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
5422 			    ivar);
5423 		}
5424 		break;
5425 
5426 	default:
5427 		break;
5428 	}
5429 }
5430 
5431 /*
5432  * Convert the rx ring index driver maintained to the rx ring index
5433  * in h/w.
5434  */
5435 static uint32_t
5436 ixgbe_get_hw_rx_index(ixgbe_t *ixgbe, uint32_t sw_rx_index)
5437 {
5438 
5439 	struct ixgbe_hw *hw = &ixgbe->hw;
5440 	uint32_t rx_ring_per_group, hw_rx_index;
5441 
5442 	if (ixgbe->classify_mode == IXGBE_CLASSIFY_RSS ||
5443 	    ixgbe->classify_mode == IXGBE_CLASSIFY_NONE) {
5444 		return (sw_rx_index);
5445 	} else if (ixgbe->classify_mode == IXGBE_CLASSIFY_VMDQ) {
5446 		switch (hw->mac.type) {
5447 		case ixgbe_mac_82598EB:
5448 			return (sw_rx_index);
5449 
5450 		case ixgbe_mac_82599EB:
5451 		case ixgbe_mac_X540:
5452 		case ixgbe_mac_X550:
5453 		case ixgbe_mac_X550EM_x:
5454 			return (sw_rx_index * 2);
5455 
5456 		default:
5457 			break;
5458 		}
5459 	} else if (ixgbe->classify_mode == IXGBE_CLASSIFY_VMDQ_RSS) {
5460 		rx_ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
5461 
5462 		switch (hw->mac.type) {
5463 		case ixgbe_mac_82598EB:
5464 			hw_rx_index = (sw_rx_index / rx_ring_per_group) *
5465 			    16 + (sw_rx_index % rx_ring_per_group);
5466 			return (hw_rx_index);
5467 
5468 		case ixgbe_mac_82599EB:
5469 		case ixgbe_mac_X540:
5470 		case ixgbe_mac_X550:
5471 		case ixgbe_mac_X550EM_x:
5472 			if (ixgbe->num_rx_groups > 32) {
5473 				hw_rx_index = (sw_rx_index /
5474 				    rx_ring_per_group) * 2 +
5475 				    (sw_rx_index % rx_ring_per_group);
5476 			} else {
5477 				hw_rx_index = (sw_rx_index /
5478 				    rx_ring_per_group) * 4 +
5479 				    (sw_rx_index % rx_ring_per_group);
5480 			}
5481 			return (hw_rx_index);
5482 
5483 		default:
5484 			break;
5485 		}
5486 	}
5487 
5488 	/*
5489 	 * Should never reach. Just to make compiler happy.
5490 	 */
5491 	return (sw_rx_index);
5492 }
5493 
5494 /*
5495  * ixgbe_map_intrs_to_vectors - Map different interrupts to MSI-X vectors.
5496  *
5497  * For MSI-X, here will map rx interrupt, tx interrupt and other interrupt
5498  * to vector[0 - (intr_cnt -1)].
5499  */
5500 static int
5501 ixgbe_map_intrs_to_vectors(ixgbe_t *ixgbe)
5502 {
5503 	int i, vector = 0;
5504 
5505 	/* initialize vector map */
5506 	bzero(&ixgbe->vect_map, sizeof (ixgbe->vect_map));
5507 	for (i = 0; i < ixgbe->intr_cnt; i++) {
5508 		ixgbe->vect_map[i].ixgbe = ixgbe;
5509 	}
5510 
5511 	/*
5512 	 * non-MSI-X case is very simple: rx rings[0] on RTxQ[0],
5513 	 * tx rings[0] on RTxQ[1].
5514 	 */
5515 	if (ixgbe->intr_type != DDI_INTR_TYPE_MSIX) {
5516 		ixgbe_map_rxring_to_vector(ixgbe, 0, 0);
5517 		ixgbe_map_txring_to_vector(ixgbe, 0, 1);
5518 		return (IXGBE_SUCCESS);
5519 	}
5520 
5521 	/*
5522 	 * Interrupts/vectors mapping for MSI-X
5523 	 */
5524 
5525 	/*
5526 	 * Map other interrupt to vector 0,
5527 	 * Set bit in map and count the bits set.
5528 	 */
5529 	BT_SET(ixgbe->vect_map[vector].other_map, 0);
5530 	ixgbe->vect_map[vector].other_cnt++;
5531 
5532 	/*
5533 	 * Map rx ring interrupts to vectors
5534 	 */
5535 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
5536 		ixgbe_map_rxring_to_vector(ixgbe, i, vector);
5537 		vector = (vector +1) % ixgbe->intr_cnt;
5538 	}
5539 
5540 	/*
5541 	 * Map tx ring interrupts to vectors
5542 	 */
5543 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
5544 		ixgbe_map_txring_to_vector(ixgbe, i, vector);
5545 		vector = (vector +1) % ixgbe->intr_cnt;
5546 	}
5547 
5548 	return (IXGBE_SUCCESS);
5549 }
5550 
5551 /*
5552  * ixgbe_setup_adapter_vector - Setup the adapter interrupt vector(s).
5553  *
5554  * This relies on ring/vector mapping already set up in the
5555  * vect_map[] structures
5556  */
5557 static void
5558 ixgbe_setup_adapter_vector(ixgbe_t *ixgbe)
5559 {
5560 	struct ixgbe_hw *hw = &ixgbe->hw;
5561 	ixgbe_intr_vector_t *vect;	/* vector bitmap */
5562 	int r_idx;	/* ring index */
5563 	int v_idx;	/* vector index */
5564 	uint32_t hw_index;
5565 
5566 	/*
5567 	 * Clear any previous entries
5568 	 */
5569 	switch (hw->mac.type) {
5570 	case ixgbe_mac_82598EB:
5571 		for (v_idx = 0; v_idx < 25; v_idx++)
5572 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(v_idx), 0);
5573 		break;
5574 
5575 	case ixgbe_mac_82599EB:
5576 	case ixgbe_mac_X540:
5577 	case ixgbe_mac_X550:
5578 	case ixgbe_mac_X550EM_x:
5579 		for (v_idx = 0; v_idx < 64; v_idx++)
5580 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(v_idx), 0);
5581 		IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, 0);
5582 		break;
5583 
5584 	default:
5585 		break;
5586 	}
5587 
5588 	/*
5589 	 * For non MSI-X interrupt, rx rings[0] will use RTxQ[0], and
5590 	 * tx rings[0] will use RTxQ[1].
5591 	 */
5592 	if (ixgbe->intr_type != DDI_INTR_TYPE_MSIX) {
5593 		ixgbe_setup_ivar(ixgbe, 0, 0, 0);
5594 		ixgbe_setup_ivar(ixgbe, 0, 1, 1);
5595 		return;
5596 	}
5597 
5598 	/*
5599 	 * For MSI-X interrupt, "Other" is always on vector[0].
5600 	 */
5601 	ixgbe_setup_ivar(ixgbe, IXGBE_IVAR_OTHER_CAUSES_INDEX, 0, -1);
5602 
5603 	/*
5604 	 * For each interrupt vector, populate the IVAR table
5605 	 */
5606 	for (v_idx = 0; v_idx < ixgbe->intr_cnt; v_idx++) {
5607 		vect = &ixgbe->vect_map[v_idx];
5608 
5609 		/*
5610 		 * For each rx ring bit set
5611 		 */
5612 		r_idx = bt_getlowbit(vect->rx_map, 0,
5613 		    (ixgbe->num_rx_rings - 1));
5614 
5615 		while (r_idx >= 0) {
5616 			hw_index = ixgbe->rx_rings[r_idx].hw_index;
5617 			ixgbe_setup_ivar(ixgbe, hw_index, v_idx, 0);
5618 			r_idx = bt_getlowbit(vect->rx_map, (r_idx + 1),
5619 			    (ixgbe->num_rx_rings - 1));
5620 		}
5621 
5622 		/*
5623 		 * For each tx ring bit set
5624 		 */
5625 		r_idx = bt_getlowbit(vect->tx_map, 0,
5626 		    (ixgbe->num_tx_rings - 1));
5627 
5628 		while (r_idx >= 0) {
5629 			ixgbe_setup_ivar(ixgbe, r_idx, v_idx, 1);
5630 			r_idx = bt_getlowbit(vect->tx_map, (r_idx + 1),
5631 			    (ixgbe->num_tx_rings - 1));
5632 		}
5633 	}
5634 }
5635 
5636 /*
5637  * ixgbe_rem_intr_handlers - Remove the interrupt handlers.
5638  */
5639 static void
5640 ixgbe_rem_intr_handlers(ixgbe_t *ixgbe)
5641 {
5642 	int i;
5643 	int rc;
5644 
5645 	for (i = 0; i < ixgbe->intr_cnt; i++) {
5646 		rc = ddi_intr_remove_handler(ixgbe->htable[i]);
5647 		if (rc != DDI_SUCCESS) {
5648 			IXGBE_DEBUGLOG_1(ixgbe,
5649 			    "Remove intr handler failed: %d", rc);
5650 		}
5651 	}
5652 }
5653 
5654 /*
5655  * ixgbe_rem_intrs - Remove the allocated interrupts.
5656  */
5657 static void
5658 ixgbe_rem_intrs(ixgbe_t *ixgbe)
5659 {
5660 	int i;
5661 	int rc;
5662 
5663 	for (i = 0; i < ixgbe->intr_cnt; i++) {
5664 		rc = ddi_intr_free(ixgbe->htable[i]);
5665 		if (rc != DDI_SUCCESS) {
5666 			IXGBE_DEBUGLOG_1(ixgbe,
5667 			    "Free intr failed: %d", rc);
5668 		}
5669 	}
5670 
5671 	kmem_free(ixgbe->htable, ixgbe->intr_size);
5672 	ixgbe->htable = NULL;
5673 }
5674 
5675 /*
5676  * ixgbe_enable_intrs - Enable all the ddi interrupts.
5677  */
5678 static int
5679 ixgbe_enable_intrs(ixgbe_t *ixgbe)
5680 {
5681 	int i;
5682 	int rc;
5683 
5684 	/*
5685 	 * Enable interrupts
5686 	 */
5687 	if (ixgbe->intr_cap & DDI_INTR_FLAG_BLOCK) {
5688 		/*
5689 		 * Call ddi_intr_block_enable() for MSI
5690 		 */
5691 		rc = ddi_intr_block_enable(ixgbe->htable, ixgbe->intr_cnt);
5692 		if (rc != DDI_SUCCESS) {
5693 			ixgbe_log(ixgbe,
5694 			    "Enable block intr failed: %d", rc);
5695 			return (IXGBE_FAILURE);
5696 		}
5697 	} else {
5698 		/*
5699 		 * Call ddi_intr_enable() for Legacy/MSI non block enable
5700 		 */
5701 		for (i = 0; i < ixgbe->intr_cnt; i++) {
5702 			rc = ddi_intr_enable(ixgbe->htable[i]);
5703 			if (rc != DDI_SUCCESS) {
5704 				ixgbe_log(ixgbe,
5705 				    "Enable intr failed: %d", rc);
5706 				return (IXGBE_FAILURE);
5707 			}
5708 		}
5709 	}
5710 
5711 	return (IXGBE_SUCCESS);
5712 }
5713 
5714 /*
5715  * ixgbe_disable_intrs - Disable all the interrupts.
5716  */
5717 static int
5718 ixgbe_disable_intrs(ixgbe_t *ixgbe)
5719 {
5720 	int i;
5721 	int rc;
5722 
5723 	/*
5724 	 * Disable all interrupts
5725 	 */
5726 	if (ixgbe->intr_cap & DDI_INTR_FLAG_BLOCK) {
5727 		rc = ddi_intr_block_disable(ixgbe->htable, ixgbe->intr_cnt);
5728 		if (rc != DDI_SUCCESS) {
5729 			ixgbe_log(ixgbe,
5730 			    "Disable block intr failed: %d", rc);
5731 			return (IXGBE_FAILURE);
5732 		}
5733 	} else {
5734 		for (i = 0; i < ixgbe->intr_cnt; i++) {
5735 			rc = ddi_intr_disable(ixgbe->htable[i]);
5736 			if (rc != DDI_SUCCESS) {
5737 				ixgbe_log(ixgbe,
5738 				    "Disable intr failed: %d", rc);
5739 				return (IXGBE_FAILURE);
5740 			}
5741 		}
5742 	}
5743 
5744 	return (IXGBE_SUCCESS);
5745 }
5746 
5747 /*
5748  * ixgbe_get_hw_state - Get and save parameters related to adapter hardware.
5749  */
5750 static void
5751 ixgbe_get_hw_state(ixgbe_t *ixgbe)
5752 {
5753 	struct ixgbe_hw *hw = &ixgbe->hw;
5754 	ixgbe_link_speed speed = 0;
5755 	boolean_t link_up = B_FALSE;
5756 	uint32_t pcs1g_anlp = 0;
5757 
5758 	ASSERT(mutex_owned(&ixgbe->gen_lock));
5759 	ixgbe->param_lp_1000fdx_cap = 0;
5760 	ixgbe->param_lp_100fdx_cap  = 0;
5761 
5762 	/* check for link, don't wait */
5763 	(void) ixgbe_check_link(hw, &speed, &link_up, B_FALSE);
5764 
5765 	/*
5766 	 * Update the observed Link Partner's capabilities. Not all adapters
5767 	 * can provide full information on the LP's capable speeds, so we
5768 	 * provide what we can.
5769 	 */
5770 	if (link_up) {
5771 		pcs1g_anlp = IXGBE_READ_REG(hw, IXGBE_PCS1GANLP);
5772 
5773 		ixgbe->param_lp_1000fdx_cap =
5774 		    (pcs1g_anlp & IXGBE_PCS1GANLP_LPFD) ? 1 : 0;
5775 		ixgbe->param_lp_100fdx_cap =
5776 		    (pcs1g_anlp & IXGBE_PCS1GANLP_LPFD) ? 1 : 0;
5777 	}
5778 
5779 	/*
5780 	 * Update GLD's notion of the adapter's currently advertised speeds.
5781 	 * Since the common code doesn't always record the current autonegotiate
5782 	 * settings in the phy struct for all parts (specifically, adapters with
5783 	 * SFPs) we first test to see if it is 0, and if so, we fall back to
5784 	 * using the adapter's speed capabilities which we saved during instance
5785 	 * init in ixgbe_init_params().
5786 	 *
5787 	 * Adapters with SFPs will always be shown as advertising all of their
5788 	 * supported speeds, and adapters with baseT PHYs (where the phy struct
5789 	 * is maintained by the common code) will always have a factual view of
5790 	 * their currently-advertised speeds. In the case of SFPs, this is
5791 	 * acceptable as we default to advertising all speeds that the adapter
5792 	 * claims to support, and those properties are immutable; unlike on
5793 	 * baseT (copper) PHYs, where speeds can be enabled or disabled at will.
5794 	 */
5795 	speed = hw->phy.autoneg_advertised;
5796 	if (speed == 0)
5797 		speed = ixgbe->speeds_supported;
5798 
5799 	ixgbe->param_adv_10000fdx_cap =
5800 	    (speed & IXGBE_LINK_SPEED_10GB_FULL) ? 1 : 0;
5801 	ixgbe->param_adv_5000fdx_cap =
5802 	    (speed & IXGBE_LINK_SPEED_5GB_FULL) ? 1 : 0;
5803 	ixgbe->param_adv_2500fdx_cap =
5804 	    (speed & IXGBE_LINK_SPEED_2_5GB_FULL) ? 1 : 0;
5805 	ixgbe->param_adv_1000fdx_cap =
5806 	    (speed & IXGBE_LINK_SPEED_1GB_FULL) ? 1 : 0;
5807 	ixgbe->param_adv_100fdx_cap =
5808 	    (speed & IXGBE_LINK_SPEED_100_FULL) ? 1 : 0;
5809 }
5810 
5811 /*
5812  * ixgbe_get_driver_control - Notify that driver is in control of device.
5813  */
5814 static void
5815 ixgbe_get_driver_control(struct ixgbe_hw *hw)
5816 {
5817 	uint32_t ctrl_ext;
5818 
5819 	/*
5820 	 * Notify firmware that driver is in control of device
5821 	 */
5822 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5823 	ctrl_ext |= IXGBE_CTRL_EXT_DRV_LOAD;
5824 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5825 }
5826 
5827 /*
5828  * ixgbe_release_driver_control - Notify that driver is no longer in control
5829  * of device.
5830  */
5831 static void
5832 ixgbe_release_driver_control(struct ixgbe_hw *hw)
5833 {
5834 	uint32_t ctrl_ext;
5835 
5836 	/*
5837 	 * Notify firmware that driver is no longer in control of device
5838 	 */
5839 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5840 	ctrl_ext &= ~IXGBE_CTRL_EXT_DRV_LOAD;
5841 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5842 }
5843 
5844 /*
5845  * ixgbe_atomic_reserve - Atomic decrease operation.
5846  */
5847 int
5848 ixgbe_atomic_reserve(uint32_t *count_p, uint32_t n)
5849 {
5850 	uint32_t oldval;
5851 	uint32_t newval;
5852 
5853 	/*
5854 	 * ATOMICALLY
5855 	 */
5856 	do {
5857 		oldval = *count_p;
5858 		if (oldval < n)
5859 			return (-1);
5860 		newval = oldval - n;
5861 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
5862 
5863 	return (newval);
5864 }
5865 
5866 /*
5867  * ixgbe_mc_table_itr - Traverse the entries in the multicast table.
5868  */
5869 static uint8_t *
5870 ixgbe_mc_table_itr(struct ixgbe_hw *hw, uint8_t **upd_ptr, uint32_t *vmdq)
5871 {
5872 	uint8_t *addr = *upd_ptr;
5873 	uint8_t *new_ptr;
5874 
5875 	_NOTE(ARGUNUSED(hw));
5876 	_NOTE(ARGUNUSED(vmdq));
5877 
5878 	new_ptr = addr + IXGBE_ETH_LENGTH_OF_ADDRESS;
5879 	*upd_ptr = new_ptr;
5880 	return (addr);
5881 }
5882 
5883 /*
5884  * FMA support
5885  */
5886 int
5887 ixgbe_check_acc_handle(ddi_acc_handle_t handle)
5888 {
5889 	ddi_fm_error_t de;
5890 
5891 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
5892 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
5893 	return (de.fme_status);
5894 }
5895 
5896 int
5897 ixgbe_check_dma_handle(ddi_dma_handle_t handle)
5898 {
5899 	ddi_fm_error_t de;
5900 
5901 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
5902 	return (de.fme_status);
5903 }
5904 
5905 /*
5906  * ixgbe_fm_error_cb - The IO fault service error handling callback function.
5907  */
5908 static int
5909 ixgbe_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
5910 {
5911 	_NOTE(ARGUNUSED(impl_data));
5912 	/*
5913 	 * as the driver can always deal with an error in any dma or
5914 	 * access handle, we can just return the fme_status value.
5915 	 */
5916 	pci_ereport_post(dip, err, NULL);
5917 	return (err->fme_status);
5918 }
5919 
5920 static void
5921 ixgbe_fm_init(ixgbe_t *ixgbe)
5922 {
5923 	ddi_iblock_cookie_t iblk;
5924 	int fma_dma_flag;
5925 
5926 	/*
5927 	 * Only register with IO Fault Services if we have some capability
5928 	 */
5929 	if (ixgbe->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
5930 		ixgbe_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
5931 	} else {
5932 		ixgbe_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
5933 	}
5934 
5935 	if (ixgbe->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
5936 		fma_dma_flag = 1;
5937 	} else {
5938 		fma_dma_flag = 0;
5939 	}
5940 
5941 	ixgbe_set_fma_flags(fma_dma_flag);
5942 
5943 	if (ixgbe->fm_capabilities) {
5944 
5945 		/*
5946 		 * Register capabilities with IO Fault Services
5947 		 */
5948 		ddi_fm_init(ixgbe->dip, &ixgbe->fm_capabilities, &iblk);
5949 
5950 		/*
5951 		 * Initialize pci ereport capabilities if ereport capable
5952 		 */
5953 		if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities) ||
5954 		    DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5955 			pci_ereport_setup(ixgbe->dip);
5956 
5957 		/*
5958 		 * Register error callback if error callback capable
5959 		 */
5960 		if (DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5961 			ddi_fm_handler_register(ixgbe->dip,
5962 			    ixgbe_fm_error_cb, (void*) ixgbe);
5963 	}
5964 }
5965 
5966 static void
5967 ixgbe_fm_fini(ixgbe_t *ixgbe)
5968 {
5969 	/*
5970 	 * Only unregister FMA capabilities if they are registered
5971 	 */
5972 	if (ixgbe->fm_capabilities) {
5973 
5974 		/*
5975 		 * Release any resources allocated by pci_ereport_setup()
5976 		 */
5977 		if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities) ||
5978 		    DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5979 			pci_ereport_teardown(ixgbe->dip);
5980 
5981 		/*
5982 		 * Un-register error callback if error callback capable
5983 		 */
5984 		if (DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5985 			ddi_fm_handler_unregister(ixgbe->dip);
5986 
5987 		/*
5988 		 * Unregister from IO Fault Service
5989 		 */
5990 		ddi_fm_fini(ixgbe->dip);
5991 	}
5992 }
5993 
5994 void
5995 ixgbe_fm_ereport(ixgbe_t *ixgbe, char *detail)
5996 {
5997 	uint64_t ena;
5998 	char buf[FM_MAX_CLASS];
5999 
6000 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
6001 	ena = fm_ena_generate(0, FM_ENA_FMT1);
6002 	if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities)) {
6003 		ddi_fm_ereport_post(ixgbe->dip, buf, ena, DDI_NOSLEEP,
6004 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
6005 	}
6006 }
6007 
6008 static int
6009 ixgbe_ring_start(mac_ring_driver_t rh, uint64_t mr_gen_num)
6010 {
6011 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)rh;
6012 
6013 	mutex_enter(&rx_ring->rx_lock);
6014 	rx_ring->ring_gen_num = mr_gen_num;
6015 	mutex_exit(&rx_ring->rx_lock);
6016 	return (0);
6017 }
6018 
6019 /*
6020  * Get the global ring index by a ring index within a group.
6021  */
6022 static int
6023 ixgbe_get_rx_ring_index(ixgbe_t *ixgbe, int gindex, int rindex)
6024 {
6025 	ixgbe_rx_ring_t *rx_ring;
6026 	int i;
6027 
6028 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
6029 		rx_ring = &ixgbe->rx_rings[i];
6030 		if (rx_ring->group_index == gindex)
6031 			rindex--;
6032 		if (rindex < 0)
6033 			return (i);
6034 	}
6035 
6036 	return (-1);
6037 }
6038 
6039 /*
6040  * Callback funtion for MAC layer to register all rings.
6041  */
6042 /* ARGSUSED */
6043 void
6044 ixgbe_fill_ring(void *arg, mac_ring_type_t rtype, const int group_index,
6045     const int ring_index, mac_ring_info_t *infop, mac_ring_handle_t rh)
6046 {
6047 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
6048 	mac_intr_t *mintr = &infop->mri_intr;
6049 
6050 	switch (rtype) {
6051 	case MAC_RING_TYPE_RX: {
6052 		/*
6053 		 * 'index' is the ring index within the group.
6054 		 * Need to get the global ring index by searching in groups.
6055 		 */
6056 		int global_ring_index = ixgbe_get_rx_ring_index(
6057 		    ixgbe, group_index, ring_index);
6058 
6059 		ASSERT(global_ring_index >= 0);
6060 
6061 		ixgbe_rx_ring_t *rx_ring = &ixgbe->rx_rings[global_ring_index];
6062 		rx_ring->ring_handle = rh;
6063 
6064 		infop->mri_driver = (mac_ring_driver_t)rx_ring;
6065 		infop->mri_start = ixgbe_ring_start;
6066 		infop->mri_stop = NULL;
6067 		infop->mri_poll = ixgbe_ring_rx_poll;
6068 		infop->mri_stat = ixgbe_rx_ring_stat;
6069 
6070 		mintr->mi_handle = (mac_intr_handle_t)rx_ring;
6071 		mintr->mi_enable = ixgbe_rx_ring_intr_enable;
6072 		mintr->mi_disable = ixgbe_rx_ring_intr_disable;
6073 		if (ixgbe->intr_type &
6074 		    (DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI)) {
6075 			mintr->mi_ddi_handle =
6076 			    ixgbe->htable[rx_ring->intr_vector];
6077 		}
6078 
6079 		break;
6080 	}
6081 	case MAC_RING_TYPE_TX: {
6082 		ASSERT(group_index == -1);
6083 		ASSERT(ring_index < ixgbe->num_tx_rings);
6084 
6085 		ixgbe_tx_ring_t *tx_ring = &ixgbe->tx_rings[ring_index];
6086 		tx_ring->ring_handle = rh;
6087 
6088 		infop->mri_driver = (mac_ring_driver_t)tx_ring;
6089 		infop->mri_start = NULL;
6090 		infop->mri_stop = NULL;
6091 		infop->mri_tx = ixgbe_ring_tx;
6092 		infop->mri_stat = ixgbe_tx_ring_stat;
6093 		if (ixgbe->intr_type &
6094 		    (DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI)) {
6095 			mintr->mi_ddi_handle =
6096 			    ixgbe->htable[tx_ring->intr_vector];
6097 		}
6098 		break;
6099 	}
6100 	default:
6101 		break;
6102 	}
6103 }
6104 
6105 /*
6106  * Callback funtion for MAC layer to register all groups.
6107  */
6108 void
6109 ixgbe_fill_group(void *arg, mac_ring_type_t rtype, const int index,
6110     mac_group_info_t *infop, mac_group_handle_t gh)
6111 {
6112 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
6113 
6114 	switch (rtype) {
6115 	case MAC_RING_TYPE_RX: {
6116 		ixgbe_rx_group_t *rx_group;
6117 
6118 		rx_group = &ixgbe->rx_groups[index];
6119 		rx_group->group_handle = gh;
6120 
6121 		infop->mgi_driver = (mac_group_driver_t)rx_group;
6122 		infop->mgi_start = NULL;
6123 		infop->mgi_stop = NULL;
6124 		infop->mgi_addmac = ixgbe_addmac;
6125 		infop->mgi_remmac = ixgbe_remmac;
6126 		infop->mgi_count = (ixgbe->num_rx_rings / ixgbe->num_rx_groups);
6127 
6128 		break;
6129 	}
6130 	case MAC_RING_TYPE_TX:
6131 		break;
6132 	default:
6133 		break;
6134 	}
6135 }
6136 
6137 /*
6138  * Enable interrupt on the specificed rx ring.
6139  */
6140 int
6141 ixgbe_rx_ring_intr_enable(mac_intr_handle_t intrh)
6142 {
6143 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)intrh;
6144 	ixgbe_t *ixgbe = rx_ring->ixgbe;
6145 	int r_idx = rx_ring->index;
6146 	int hw_r_idx = rx_ring->hw_index;
6147 	int v_idx = rx_ring->intr_vector;
6148 
6149 	mutex_enter(&ixgbe->gen_lock);
6150 	if (ixgbe->ixgbe_state & IXGBE_INTR_ADJUST) {
6151 		mutex_exit(&ixgbe->gen_lock);
6152 		/*
6153 		 * Simply return 0.
6154 		 * Interrupts are being adjusted. ixgbe_intr_adjust()
6155 		 * will eventually re-enable the interrupt when it's
6156 		 * done with the adjustment.
6157 		 */
6158 		return (0);
6159 	}
6160 
6161 	/*
6162 	 * To enable interrupt by setting the VAL bit of given interrupt
6163 	 * vector allocation register (IVAR).
6164 	 */
6165 	ixgbe_enable_ivar(ixgbe, hw_r_idx, 0);
6166 
6167 	BT_SET(ixgbe->vect_map[v_idx].rx_map, r_idx);
6168 
6169 	/*
6170 	 * Trigger a Rx interrupt on this ring
6171 	 */
6172 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_EICS, (1 << v_idx));
6173 	IXGBE_WRITE_FLUSH(&ixgbe->hw);
6174 
6175 	mutex_exit(&ixgbe->gen_lock);
6176 
6177 	return (0);
6178 }
6179 
6180 /*
6181  * Disable interrupt on the specificed rx ring.
6182  */
6183 int
6184 ixgbe_rx_ring_intr_disable(mac_intr_handle_t intrh)
6185 {
6186 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)intrh;
6187 	ixgbe_t *ixgbe = rx_ring->ixgbe;
6188 	int r_idx = rx_ring->index;
6189 	int hw_r_idx = rx_ring->hw_index;
6190 	int v_idx = rx_ring->intr_vector;
6191 
6192 	mutex_enter(&ixgbe->gen_lock);
6193 	if (ixgbe->ixgbe_state & IXGBE_INTR_ADJUST) {
6194 		mutex_exit(&ixgbe->gen_lock);
6195 		/*
6196 		 * Simply return 0.
6197 		 * In the rare case where an interrupt is being
6198 		 * disabled while interrupts are being adjusted,
6199 		 * we don't fail the operation. No interrupts will
6200 		 * be generated while they are adjusted, and
6201 		 * ixgbe_intr_adjust() will cause the interrupts
6202 		 * to be re-enabled once it completes. Note that
6203 		 * in this case, packets may be delivered to the
6204 		 * stack via interrupts before xgbe_rx_ring_intr_enable()
6205 		 * is called again. This is acceptable since interrupt
6206 		 * adjustment is infrequent, and the stack will be
6207 		 * able to handle these packets.
6208 		 */
6209 		return (0);
6210 	}
6211 
6212 	/*
6213 	 * To disable interrupt by clearing the VAL bit of given interrupt
6214 	 * vector allocation register (IVAR).
6215 	 */
6216 	ixgbe_disable_ivar(ixgbe, hw_r_idx, 0);
6217 
6218 	BT_CLEAR(ixgbe->vect_map[v_idx].rx_map, r_idx);
6219 
6220 	mutex_exit(&ixgbe->gen_lock);
6221 
6222 	return (0);
6223 }
6224 
6225 /*
6226  * Add a mac address.
6227  */
6228 static int
6229 ixgbe_addmac(void *arg, const uint8_t *mac_addr)
6230 {
6231 	ixgbe_rx_group_t *rx_group = (ixgbe_rx_group_t *)arg;
6232 	ixgbe_t *ixgbe = rx_group->ixgbe;
6233 	struct ixgbe_hw *hw = &ixgbe->hw;
6234 	int slot, i;
6235 
6236 	mutex_enter(&ixgbe->gen_lock);
6237 
6238 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
6239 		mutex_exit(&ixgbe->gen_lock);
6240 		return (ECANCELED);
6241 	}
6242 
6243 	if (ixgbe->unicst_avail == 0) {
6244 		/* no slots available */
6245 		mutex_exit(&ixgbe->gen_lock);
6246 		return (ENOSPC);
6247 	}
6248 
6249 	/*
6250 	 * The first ixgbe->num_rx_groups slots are reserved for each respective
6251 	 * group. The rest slots are shared by all groups. While adding a
6252 	 * MAC address, reserved slots are firstly checked then the shared
6253 	 * slots are searched.
6254 	 */
6255 	slot = -1;
6256 	if (ixgbe->unicst_addr[rx_group->index].mac.set == 1) {
6257 		for (i = ixgbe->num_rx_groups; i < ixgbe->unicst_total; i++) {
6258 			if (ixgbe->unicst_addr[i].mac.set == 0) {
6259 				slot = i;
6260 				break;
6261 			}
6262 		}
6263 	} else {
6264 		slot = rx_group->index;
6265 	}
6266 
6267 	if (slot == -1) {
6268 		/* no slots available */
6269 		mutex_exit(&ixgbe->gen_lock);
6270 		return (ENOSPC);
6271 	}
6272 
6273 	bcopy(mac_addr, ixgbe->unicst_addr[slot].mac.addr, ETHERADDRL);
6274 	(void) ixgbe_set_rar(hw, slot, ixgbe->unicst_addr[slot].mac.addr,
6275 	    rx_group->index, IXGBE_RAH_AV);
6276 	ixgbe->unicst_addr[slot].mac.set = 1;
6277 	ixgbe->unicst_addr[slot].mac.group_index = rx_group->index;
6278 	ixgbe->unicst_avail--;
6279 
6280 	mutex_exit(&ixgbe->gen_lock);
6281 
6282 	return (0);
6283 }
6284 
6285 /*
6286  * Remove a mac address.
6287  */
6288 static int
6289 ixgbe_remmac(void *arg, const uint8_t *mac_addr)
6290 {
6291 	ixgbe_rx_group_t *rx_group = (ixgbe_rx_group_t *)arg;
6292 	ixgbe_t *ixgbe = rx_group->ixgbe;
6293 	struct ixgbe_hw *hw = &ixgbe->hw;
6294 	int slot;
6295 
6296 	mutex_enter(&ixgbe->gen_lock);
6297 
6298 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
6299 		mutex_exit(&ixgbe->gen_lock);
6300 		return (ECANCELED);
6301 	}
6302 
6303 	slot = ixgbe_unicst_find(ixgbe, mac_addr);
6304 	if (slot == -1) {
6305 		mutex_exit(&ixgbe->gen_lock);
6306 		return (EINVAL);
6307 	}
6308 
6309 	if (ixgbe->unicst_addr[slot].mac.set == 0) {
6310 		mutex_exit(&ixgbe->gen_lock);
6311 		return (EINVAL);
6312 	}
6313 
6314 	bzero(ixgbe->unicst_addr[slot].mac.addr, ETHERADDRL);
6315 	(void) ixgbe_clear_rar(hw, slot);
6316 	ixgbe->unicst_addr[slot].mac.set = 0;
6317 	ixgbe->unicst_avail++;
6318 
6319 	mutex_exit(&ixgbe->gen_lock);
6320 
6321 	return (0);
6322 }
6323