xref: /illumos-gate/usr/src/uts/common/io/ixgbe/ixgbe_main.c (revision fb2a9bae0030340ad72b9c26ba1ffee2ee3cafec)
1 /*
2  * CDDL HEADER START
3  *
4  * Copyright(c) 2007-2010 Intel Corporation. All rights reserved.
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 
23 /*
24  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
25  * Use is subject to license terms.
26  */
27 
28 #include "ixgbe_sw.h"
29 
30 static char ixgbe_ident[] = "Intel 10Gb Ethernet";
31 static char ixgbe_version[] = "ixgbe 1.1.5";
32 
33 /*
34  * Local function protoypes
35  */
36 static int ixgbe_register_mac(ixgbe_t *);
37 static int ixgbe_identify_hardware(ixgbe_t *);
38 static int ixgbe_regs_map(ixgbe_t *);
39 static void ixgbe_init_properties(ixgbe_t *);
40 static int ixgbe_init_driver_settings(ixgbe_t *);
41 static void ixgbe_init_locks(ixgbe_t *);
42 static void ixgbe_destroy_locks(ixgbe_t *);
43 static int ixgbe_init(ixgbe_t *);
44 static int ixgbe_chip_start(ixgbe_t *);
45 static void ixgbe_chip_stop(ixgbe_t *);
46 static int ixgbe_reset(ixgbe_t *);
47 static void ixgbe_tx_clean(ixgbe_t *);
48 static boolean_t ixgbe_tx_drain(ixgbe_t *);
49 static boolean_t ixgbe_rx_drain(ixgbe_t *);
50 static int ixgbe_alloc_rings(ixgbe_t *);
51 static void ixgbe_free_rings(ixgbe_t *);
52 static int ixgbe_alloc_rx_data(ixgbe_t *);
53 static void ixgbe_free_rx_data(ixgbe_t *);
54 static void ixgbe_setup_rings(ixgbe_t *);
55 static void ixgbe_setup_rx(ixgbe_t *);
56 static void ixgbe_setup_tx(ixgbe_t *);
57 static void ixgbe_setup_rx_ring(ixgbe_rx_ring_t *);
58 static void ixgbe_setup_tx_ring(ixgbe_tx_ring_t *);
59 static void ixgbe_setup_rss(ixgbe_t *);
60 static void ixgbe_setup_vmdq(ixgbe_t *);
61 static void ixgbe_setup_vmdq_rss(ixgbe_t *);
62 static void ixgbe_init_unicst(ixgbe_t *);
63 static int ixgbe_unicst_find(ixgbe_t *, const uint8_t *);
64 static void ixgbe_setup_multicst(ixgbe_t *);
65 static void ixgbe_get_hw_state(ixgbe_t *);
66 static void ixgbe_setup_vmdq_rss_conf(ixgbe_t *ixgbe);
67 static void ixgbe_get_conf(ixgbe_t *);
68 static void ixgbe_init_params(ixgbe_t *);
69 static int ixgbe_get_prop(ixgbe_t *, char *, int, int, int);
70 static void ixgbe_driver_link_check(ixgbe_t *);
71 static void ixgbe_sfp_check(void *);
72 static void ixgbe_link_timer(void *);
73 static void ixgbe_local_timer(void *);
74 static void ixgbe_arm_watchdog_timer(ixgbe_t *);
75 static void ixgbe_restart_watchdog_timer(ixgbe_t *);
76 static void ixgbe_disable_adapter_interrupts(ixgbe_t *);
77 static void ixgbe_enable_adapter_interrupts(ixgbe_t *);
78 static boolean_t is_valid_mac_addr(uint8_t *);
79 static boolean_t ixgbe_stall_check(ixgbe_t *);
80 static boolean_t ixgbe_set_loopback_mode(ixgbe_t *, uint32_t);
81 static void ixgbe_set_internal_mac_loopback(ixgbe_t *);
82 static boolean_t ixgbe_find_mac_address(ixgbe_t *);
83 static int ixgbe_alloc_intrs(ixgbe_t *);
84 static int ixgbe_alloc_intr_handles(ixgbe_t *, int);
85 static int ixgbe_add_intr_handlers(ixgbe_t *);
86 static void ixgbe_map_rxring_to_vector(ixgbe_t *, int, int);
87 static void ixgbe_map_txring_to_vector(ixgbe_t *, int, int);
88 static void ixgbe_setup_ivar(ixgbe_t *, uint16_t, uint8_t, int8_t);
89 static void ixgbe_enable_ivar(ixgbe_t *, uint16_t, int8_t);
90 static void ixgbe_disable_ivar(ixgbe_t *, uint16_t, int8_t);
91 static uint32_t ixgbe_get_hw_rx_index(ixgbe_t *ixgbe, uint32_t sw_rx_index);
92 static int ixgbe_map_intrs_to_vectors(ixgbe_t *);
93 static void ixgbe_setup_adapter_vector(ixgbe_t *);
94 static void ixgbe_rem_intr_handlers(ixgbe_t *);
95 static void ixgbe_rem_intrs(ixgbe_t *);
96 static int ixgbe_enable_intrs(ixgbe_t *);
97 static int ixgbe_disable_intrs(ixgbe_t *);
98 static uint_t ixgbe_intr_legacy(void *, void *);
99 static uint_t ixgbe_intr_msi(void *, void *);
100 static uint_t ixgbe_intr_msix(void *, void *);
101 static void ixgbe_intr_rx_work(ixgbe_rx_ring_t *);
102 static void ixgbe_intr_tx_work(ixgbe_tx_ring_t *);
103 static void ixgbe_intr_other_work(ixgbe_t *, uint32_t);
104 static void ixgbe_get_driver_control(struct ixgbe_hw *);
105 static int ixgbe_addmac(void *, const uint8_t *);
106 static int ixgbe_remmac(void *, const uint8_t *);
107 static void ixgbe_release_driver_control(struct ixgbe_hw *);
108 
109 static int ixgbe_attach(dev_info_t *, ddi_attach_cmd_t);
110 static int ixgbe_detach(dev_info_t *, ddi_detach_cmd_t);
111 static int ixgbe_resume(dev_info_t *);
112 static int ixgbe_suspend(dev_info_t *);
113 static void ixgbe_unconfigure(dev_info_t *, ixgbe_t *);
114 static uint8_t *ixgbe_mc_table_itr(struct ixgbe_hw *, uint8_t **, uint32_t *);
115 static int ixgbe_cbfunc(dev_info_t *, ddi_cb_action_t, void *, void *, void *);
116 static int ixgbe_intr_cb_register(ixgbe_t *);
117 static int ixgbe_intr_adjust(ixgbe_t *, ddi_cb_action_t, int);
118 
119 static int ixgbe_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err,
120     const void *impl_data);
121 static void ixgbe_fm_init(ixgbe_t *);
122 static void ixgbe_fm_fini(ixgbe_t *);
123 
124 char *ixgbe_priv_props[] = {
125 	"_tx_copy_thresh",
126 	"_tx_recycle_thresh",
127 	"_tx_overload_thresh",
128 	"_tx_resched_thresh",
129 	"_rx_copy_thresh",
130 	"_rx_limit_per_intr",
131 	"_intr_throttling",
132 	"_adv_pause_cap",
133 	"_adv_asym_pause_cap",
134 	NULL
135 };
136 
137 #define	IXGBE_MAX_PRIV_PROPS \
138 	(sizeof (ixgbe_priv_props) / sizeof (mac_priv_prop_t))
139 
140 static struct cb_ops ixgbe_cb_ops = {
141 	nulldev,		/* cb_open */
142 	nulldev,		/* cb_close */
143 	nodev,			/* cb_strategy */
144 	nodev,			/* cb_print */
145 	nodev,			/* cb_dump */
146 	nodev,			/* cb_read */
147 	nodev,			/* cb_write */
148 	nodev,			/* cb_ioctl */
149 	nodev,			/* cb_devmap */
150 	nodev,			/* cb_mmap */
151 	nodev,			/* cb_segmap */
152 	nochpoll,		/* cb_chpoll */
153 	ddi_prop_op,		/* cb_prop_op */
154 	NULL,			/* cb_stream */
155 	D_MP | D_HOTPLUG,	/* cb_flag */
156 	CB_REV,			/* cb_rev */
157 	nodev,			/* cb_aread */
158 	nodev			/* cb_awrite */
159 };
160 
161 static struct dev_ops ixgbe_dev_ops = {
162 	DEVO_REV,		/* devo_rev */
163 	0,			/* devo_refcnt */
164 	NULL,			/* devo_getinfo */
165 	nulldev,		/* devo_identify */
166 	nulldev,		/* devo_probe */
167 	ixgbe_attach,		/* devo_attach */
168 	ixgbe_detach,		/* devo_detach */
169 	nodev,			/* devo_reset */
170 	&ixgbe_cb_ops,		/* devo_cb_ops */
171 	NULL,			/* devo_bus_ops */
172 	ddi_power,		/* devo_power */
173 	ddi_quiesce_not_supported,	/* devo_quiesce */
174 };
175 
176 static struct modldrv ixgbe_modldrv = {
177 	&mod_driverops,		/* Type of module.  This one is a driver */
178 	ixgbe_ident,		/* Discription string */
179 	&ixgbe_dev_ops		/* driver ops */
180 };
181 
182 static struct modlinkage ixgbe_modlinkage = {
183 	MODREV_1, &ixgbe_modldrv, NULL
184 };
185 
186 /*
187  * Access attributes for register mapping
188  */
189 ddi_device_acc_attr_t ixgbe_regs_acc_attr = {
190 	DDI_DEVICE_ATTR_V1,
191 	DDI_STRUCTURE_LE_ACC,
192 	DDI_STRICTORDER_ACC,
193 	DDI_FLAGERR_ACC
194 };
195 
196 /*
197  * Loopback property
198  */
199 static lb_property_t lb_normal = {
200 	normal,	"normal", IXGBE_LB_NONE
201 };
202 
203 static lb_property_t lb_mac = {
204 	internal, "MAC", IXGBE_LB_INTERNAL_MAC
205 };
206 
207 static lb_property_t lb_external = {
208 	external, "External", IXGBE_LB_EXTERNAL
209 };
210 
211 #define	IXGBE_M_CALLBACK_FLAGS \
212 	(MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP | MC_PROPINFO)
213 
214 static mac_callbacks_t ixgbe_m_callbacks = {
215 	IXGBE_M_CALLBACK_FLAGS,
216 	ixgbe_m_stat,
217 	ixgbe_m_start,
218 	ixgbe_m_stop,
219 	ixgbe_m_promisc,
220 	ixgbe_m_multicst,
221 	NULL,
222 	NULL,
223 	NULL,
224 	ixgbe_m_ioctl,
225 	ixgbe_m_getcapab,
226 	NULL,
227 	NULL,
228 	ixgbe_m_setprop,
229 	ixgbe_m_getprop,
230 	ixgbe_m_propinfo
231 };
232 
233 /*
234  * Initialize capabilities of each supported adapter type
235  */
236 static adapter_info_t ixgbe_82598eb_cap = {
237 	64,		/* maximum number of rx queues */
238 	1,		/* minimum number of rx queues */
239 	64,		/* default number of rx queues */
240 	16,		/* maximum number of rx groups */
241 	1,		/* minimum number of rx groups */
242 	1,		/* default number of rx groups */
243 	32,		/* maximum number of tx queues */
244 	1,		/* minimum number of tx queues */
245 	8,		/* default number of tx queues */
246 	16366,		/* maximum MTU size */
247 	0xFFFF,		/* maximum interrupt throttle rate */
248 	0,		/* minimum interrupt throttle rate */
249 	200,		/* default interrupt throttle rate */
250 	18,		/* maximum total msix vectors */
251 	16,		/* maximum number of ring vectors */
252 	2,		/* maximum number of other vectors */
253 	IXGBE_EICR_LSC,	/* "other" interrupt types handled */
254 	(IXGBE_FLAG_DCA_CAPABLE	/* capability flags */
255 	| IXGBE_FLAG_RSS_CAPABLE
256 	| IXGBE_FLAG_VMDQ_CAPABLE)
257 };
258 
259 static adapter_info_t ixgbe_82599eb_cap = {
260 	128,		/* maximum number of rx queues */
261 	1,		/* minimum number of rx queues */
262 	128,		/* default number of rx queues */
263 	64,		/* maximum number of rx groups */
264 	1,		/* minimum number of rx groups */
265 	1,		/* default number of rx groups */
266 	128,		/* maximum number of tx queues */
267 	1,		/* minimum number of tx queues */
268 	8,		/* default number of tx queues */
269 	15500,		/* maximum MTU size */
270 	0xFF8,		/* maximum interrupt throttle rate */
271 	0,		/* minimum interrupt throttle rate */
272 	200,		/* default interrupt throttle rate */
273 	64,		/* maximum total msix vectors */
274 	16,		/* maximum number of ring vectors */
275 	2,		/* maximum number of other vectors */
276 	IXGBE_EICR_LSC,	/* "other" interrupt types handled */
277 	(IXGBE_FLAG_DCA_CAPABLE	/* capability flags */
278 	| IXGBE_FLAG_RSS_CAPABLE
279 	| IXGBE_FLAG_VMDQ_CAPABLE
280 	| IXGBE_FLAG_RSC_CAPABLE)
281 };
282 
283 /*
284  * Module Initialization Functions.
285  */
286 
287 int
288 _init(void)
289 {
290 	int status;
291 
292 	mac_init_ops(&ixgbe_dev_ops, MODULE_NAME);
293 
294 	status = mod_install(&ixgbe_modlinkage);
295 
296 	if (status != DDI_SUCCESS) {
297 		mac_fini_ops(&ixgbe_dev_ops);
298 	}
299 
300 	return (status);
301 }
302 
303 int
304 _fini(void)
305 {
306 	int status;
307 
308 	status = mod_remove(&ixgbe_modlinkage);
309 
310 	if (status == DDI_SUCCESS) {
311 		mac_fini_ops(&ixgbe_dev_ops);
312 	}
313 
314 	return (status);
315 }
316 
317 int
318 _info(struct modinfo *modinfop)
319 {
320 	int status;
321 
322 	status = mod_info(&ixgbe_modlinkage, modinfop);
323 
324 	return (status);
325 }
326 
327 /*
328  * ixgbe_attach - Driver attach.
329  *
330  * This function is the device specific initialization entry
331  * point. This entry point is required and must be written.
332  * The DDI_ATTACH command must be provided in the attach entry
333  * point. When attach() is called with cmd set to DDI_ATTACH,
334  * all normal kernel services (such as kmem_alloc(9F)) are
335  * available for use by the driver.
336  *
337  * The attach() function will be called once for each instance
338  * of  the  device  on  the  system with cmd set to DDI_ATTACH.
339  * Until attach() succeeds, the only driver entry points which
340  * may be called are open(9E) and getinfo(9E).
341  */
342 static int
343 ixgbe_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
344 {
345 	ixgbe_t *ixgbe;
346 	struct ixgbe_osdep *osdep;
347 	struct ixgbe_hw *hw;
348 	int instance;
349 	char taskqname[32];
350 
351 	/*
352 	 * Check the command and perform corresponding operations
353 	 */
354 	switch (cmd) {
355 	default:
356 		return (DDI_FAILURE);
357 
358 	case DDI_RESUME:
359 		return (ixgbe_resume(devinfo));
360 
361 	case DDI_ATTACH:
362 		break;
363 	}
364 
365 	/* Get the device instance */
366 	instance = ddi_get_instance(devinfo);
367 
368 	/* Allocate memory for the instance data structure */
369 	ixgbe = kmem_zalloc(sizeof (ixgbe_t), KM_SLEEP);
370 
371 	ixgbe->dip = devinfo;
372 	ixgbe->instance = instance;
373 
374 	hw = &ixgbe->hw;
375 	osdep = &ixgbe->osdep;
376 	hw->back = osdep;
377 	osdep->ixgbe = ixgbe;
378 
379 	/* Attach the instance pointer to the dev_info data structure */
380 	ddi_set_driver_private(devinfo, ixgbe);
381 
382 	/*
383 	 * Initialize for fma support
384 	 */
385 	ixgbe->fm_capabilities = ixgbe_get_prop(ixgbe, PROP_FM_CAPABLE,
386 	    0, 0x0f, DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
387 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
388 	ixgbe_fm_init(ixgbe);
389 	ixgbe->attach_progress |= ATTACH_PROGRESS_FM_INIT;
390 
391 	/*
392 	 * Map PCI config space registers
393 	 */
394 	if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
395 		ixgbe_error(ixgbe, "Failed to map PCI configurations");
396 		goto attach_fail;
397 	}
398 	ixgbe->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
399 
400 	/*
401 	 * Identify the chipset family
402 	 */
403 	if (ixgbe_identify_hardware(ixgbe) != IXGBE_SUCCESS) {
404 		ixgbe_error(ixgbe, "Failed to identify hardware");
405 		goto attach_fail;
406 	}
407 
408 	/*
409 	 * Map device registers
410 	 */
411 	if (ixgbe_regs_map(ixgbe) != IXGBE_SUCCESS) {
412 		ixgbe_error(ixgbe, "Failed to map device registers");
413 		goto attach_fail;
414 	}
415 	ixgbe->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
416 
417 	/*
418 	 * Initialize driver parameters
419 	 */
420 	ixgbe_init_properties(ixgbe);
421 	ixgbe->attach_progress |= ATTACH_PROGRESS_PROPS;
422 
423 	/*
424 	 * Register interrupt callback
425 	 */
426 	if (ixgbe_intr_cb_register(ixgbe) != IXGBE_SUCCESS) {
427 		ixgbe_error(ixgbe, "Failed to register interrupt callback");
428 		goto attach_fail;
429 	}
430 
431 	/*
432 	 * Allocate interrupts
433 	 */
434 	if (ixgbe_alloc_intrs(ixgbe) != IXGBE_SUCCESS) {
435 		ixgbe_error(ixgbe, "Failed to allocate interrupts");
436 		goto attach_fail;
437 	}
438 	ixgbe->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
439 
440 	/*
441 	 * Allocate rx/tx rings based on the ring numbers.
442 	 * The actual numbers of rx/tx rings are decided by the number of
443 	 * allocated interrupt vectors, so we should allocate the rings after
444 	 * interrupts are allocated.
445 	 */
446 	if (ixgbe_alloc_rings(ixgbe) != IXGBE_SUCCESS) {
447 		ixgbe_error(ixgbe, "Failed to allocate rx and tx rings");
448 		goto attach_fail;
449 	}
450 	ixgbe->attach_progress |= ATTACH_PROGRESS_ALLOC_RINGS;
451 
452 	/*
453 	 * Map rings to interrupt vectors
454 	 */
455 	if (ixgbe_map_intrs_to_vectors(ixgbe) != IXGBE_SUCCESS) {
456 		ixgbe_error(ixgbe, "Failed to map interrupts to vectors");
457 		goto attach_fail;
458 	}
459 
460 	/*
461 	 * Add interrupt handlers
462 	 */
463 	if (ixgbe_add_intr_handlers(ixgbe) != IXGBE_SUCCESS) {
464 		ixgbe_error(ixgbe, "Failed to add interrupt handlers");
465 		goto attach_fail;
466 	}
467 	ixgbe->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
468 
469 	/*
470 	 * Create a taskq for sfp-change
471 	 */
472 	(void) sprintf(taskqname, "ixgbe%d_taskq", instance);
473 	if ((ixgbe->sfp_taskq = ddi_taskq_create(devinfo, taskqname,
474 	    1, TASKQ_DEFAULTPRI, 0)) == NULL) {
475 		ixgbe_error(ixgbe, "taskq_create failed");
476 		goto attach_fail;
477 	}
478 	ixgbe->attach_progress |= ATTACH_PROGRESS_SFP_TASKQ;
479 
480 	/*
481 	 * Initialize driver parameters
482 	 */
483 	if (ixgbe_init_driver_settings(ixgbe) != IXGBE_SUCCESS) {
484 		ixgbe_error(ixgbe, "Failed to initialize driver settings");
485 		goto attach_fail;
486 	}
487 
488 	/*
489 	 * Initialize mutexes for this device.
490 	 * Do this before enabling the interrupt handler and
491 	 * register the softint to avoid the condition where
492 	 * interrupt handler can try using uninitialized mutex.
493 	 */
494 	ixgbe_init_locks(ixgbe);
495 	ixgbe->attach_progress |= ATTACH_PROGRESS_LOCKS;
496 
497 	/*
498 	 * Initialize chipset hardware
499 	 */
500 	if (ixgbe_init(ixgbe) != IXGBE_SUCCESS) {
501 		ixgbe_error(ixgbe, "Failed to initialize adapter");
502 		goto attach_fail;
503 	}
504 	ixgbe->link_check_complete = B_FALSE;
505 	ixgbe->link_check_hrtime = gethrtime() +
506 	    (IXGBE_LINK_UP_TIME * 100000000ULL);
507 	ixgbe->attach_progress |= ATTACH_PROGRESS_INIT;
508 
509 	if (ixgbe_check_acc_handle(ixgbe->osdep.cfg_handle) != DDI_FM_OK) {
510 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
511 		goto attach_fail;
512 	}
513 
514 	/*
515 	 * Initialize statistics
516 	 */
517 	if (ixgbe_init_stats(ixgbe) != IXGBE_SUCCESS) {
518 		ixgbe_error(ixgbe, "Failed to initialize statistics");
519 		goto attach_fail;
520 	}
521 	ixgbe->attach_progress |= ATTACH_PROGRESS_STATS;
522 
523 	/*
524 	 * Register the driver to the MAC
525 	 */
526 	if (ixgbe_register_mac(ixgbe) != IXGBE_SUCCESS) {
527 		ixgbe_error(ixgbe, "Failed to register MAC");
528 		goto attach_fail;
529 	}
530 	mac_link_update(ixgbe->mac_hdl, LINK_STATE_UNKNOWN);
531 	ixgbe->attach_progress |= ATTACH_PROGRESS_MAC;
532 
533 	ixgbe->periodic_id = ddi_periodic_add(ixgbe_link_timer, ixgbe,
534 	    IXGBE_CYCLIC_PERIOD, DDI_IPL_0);
535 	if (ixgbe->periodic_id == 0) {
536 		ixgbe_error(ixgbe, "Failed to add the link check timer");
537 		goto attach_fail;
538 	}
539 	ixgbe->attach_progress |= ATTACH_PROGRESS_LINK_TIMER;
540 
541 	/*
542 	 * Now that mutex locks are initialized, and the chip is also
543 	 * initialized, enable interrupts.
544 	 */
545 	if (ixgbe_enable_intrs(ixgbe) != IXGBE_SUCCESS) {
546 		ixgbe_error(ixgbe, "Failed to enable DDI interrupts");
547 		goto attach_fail;
548 	}
549 	ixgbe->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
550 
551 	ixgbe_log(ixgbe, "%s, %s", ixgbe_ident, ixgbe_version);
552 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_INITIALIZED);
553 
554 	return (DDI_SUCCESS);
555 
556 attach_fail:
557 	ixgbe_unconfigure(devinfo, ixgbe);
558 	return (DDI_FAILURE);
559 }
560 
561 /*
562  * ixgbe_detach - Driver detach.
563  *
564  * The detach() function is the complement of the attach routine.
565  * If cmd is set to DDI_DETACH, detach() is used to remove  the
566  * state  associated  with  a  given  instance of a device node
567  * prior to the removal of that instance from the system.
568  *
569  * The detach() function will be called once for each  instance
570  * of the device for which there has been a successful attach()
571  * once there are no longer  any  opens  on  the  device.
572  *
573  * Interrupts routine are disabled, All memory allocated by this
574  * driver are freed.
575  */
576 static int
577 ixgbe_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
578 {
579 	ixgbe_t *ixgbe;
580 
581 	/*
582 	 * Check detach command
583 	 */
584 	switch (cmd) {
585 	default:
586 		return (DDI_FAILURE);
587 
588 	case DDI_SUSPEND:
589 		return (ixgbe_suspend(devinfo));
590 
591 	case DDI_DETACH:
592 		break;
593 	}
594 
595 	/*
596 	 * Get the pointer to the driver private data structure
597 	 */
598 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
599 	if (ixgbe == NULL)
600 		return (DDI_FAILURE);
601 
602 	/*
603 	 * If the device is still running, it needs to be stopped first.
604 	 * This check is necessary because under some specific circumstances,
605 	 * the detach routine can be called without stopping the interface
606 	 * first.
607 	 */
608 	if (ixgbe->ixgbe_state & IXGBE_STARTED) {
609 		atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_STARTED);
610 		mutex_enter(&ixgbe->gen_lock);
611 		ixgbe_stop(ixgbe, B_TRUE);
612 		mutex_exit(&ixgbe->gen_lock);
613 		/* Disable and stop the watchdog timer */
614 		ixgbe_disable_watchdog_timer(ixgbe);
615 	}
616 
617 	/*
618 	 * Check if there are still rx buffers held by the upper layer.
619 	 * If so, fail the detach.
620 	 */
621 	if (!ixgbe_rx_drain(ixgbe))
622 		return (DDI_FAILURE);
623 
624 	/*
625 	 * Do the remaining unconfigure routines
626 	 */
627 	ixgbe_unconfigure(devinfo, ixgbe);
628 
629 	return (DDI_SUCCESS);
630 }
631 
632 static void
633 ixgbe_unconfigure(dev_info_t *devinfo, ixgbe_t *ixgbe)
634 {
635 	/*
636 	 * Disable interrupt
637 	 */
638 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
639 		(void) ixgbe_disable_intrs(ixgbe);
640 	}
641 
642 	/*
643 	 * remove the link check timer
644 	 */
645 	if (ixgbe->attach_progress & ATTACH_PROGRESS_LINK_TIMER) {
646 		if (ixgbe->periodic_id != NULL) {
647 			ddi_periodic_delete(ixgbe->periodic_id);
648 			ixgbe->periodic_id = NULL;
649 		}
650 	}
651 
652 	/*
653 	 * Unregister MAC
654 	 */
655 	if (ixgbe->attach_progress & ATTACH_PROGRESS_MAC) {
656 		(void) mac_unregister(ixgbe->mac_hdl);
657 	}
658 
659 	/*
660 	 * Free statistics
661 	 */
662 	if (ixgbe->attach_progress & ATTACH_PROGRESS_STATS) {
663 		kstat_delete((kstat_t *)ixgbe->ixgbe_ks);
664 	}
665 
666 	/*
667 	 * Remove interrupt handlers
668 	 */
669 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
670 		ixgbe_rem_intr_handlers(ixgbe);
671 	}
672 
673 	/*
674 	 * Remove taskq for sfp-status-change
675 	 */
676 	if (ixgbe->attach_progress & ATTACH_PROGRESS_SFP_TASKQ) {
677 		ddi_taskq_destroy(ixgbe->sfp_taskq);
678 	}
679 
680 	/*
681 	 * Remove interrupts
682 	 */
683 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ALLOC_INTR) {
684 		ixgbe_rem_intrs(ixgbe);
685 	}
686 
687 	/*
688 	 * Unregister interrupt callback handler
689 	 */
690 	(void) ddi_cb_unregister(ixgbe->cb_hdl);
691 
692 	/*
693 	 * Remove driver properties
694 	 */
695 	if (ixgbe->attach_progress & ATTACH_PROGRESS_PROPS) {
696 		(void) ddi_prop_remove_all(devinfo);
697 	}
698 
699 	/*
700 	 * Stop the chipset
701 	 */
702 	if (ixgbe->attach_progress & ATTACH_PROGRESS_INIT) {
703 		mutex_enter(&ixgbe->gen_lock);
704 		ixgbe_chip_stop(ixgbe);
705 		mutex_exit(&ixgbe->gen_lock);
706 	}
707 
708 	/*
709 	 * Free register handle
710 	 */
711 	if (ixgbe->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
712 		if (ixgbe->osdep.reg_handle != NULL)
713 			ddi_regs_map_free(&ixgbe->osdep.reg_handle);
714 	}
715 
716 	/*
717 	 * Free PCI config handle
718 	 */
719 	if (ixgbe->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
720 		if (ixgbe->osdep.cfg_handle != NULL)
721 			pci_config_teardown(&ixgbe->osdep.cfg_handle);
722 	}
723 
724 	/*
725 	 * Free locks
726 	 */
727 	if (ixgbe->attach_progress & ATTACH_PROGRESS_LOCKS) {
728 		ixgbe_destroy_locks(ixgbe);
729 	}
730 
731 	/*
732 	 * Free the rx/tx rings
733 	 */
734 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
735 		ixgbe_free_rings(ixgbe);
736 	}
737 
738 	/*
739 	 * Unregister FMA capabilities
740 	 */
741 	if (ixgbe->attach_progress & ATTACH_PROGRESS_FM_INIT) {
742 		ixgbe_fm_fini(ixgbe);
743 	}
744 
745 	/*
746 	 * Free the driver data structure
747 	 */
748 	kmem_free(ixgbe, sizeof (ixgbe_t));
749 
750 	ddi_set_driver_private(devinfo, NULL);
751 }
752 
753 /*
754  * ixgbe_register_mac - Register the driver and its function pointers with
755  * the GLD interface.
756  */
757 static int
758 ixgbe_register_mac(ixgbe_t *ixgbe)
759 {
760 	struct ixgbe_hw *hw = &ixgbe->hw;
761 	mac_register_t *mac;
762 	int status;
763 
764 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
765 		return (IXGBE_FAILURE);
766 
767 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
768 	mac->m_driver = ixgbe;
769 	mac->m_dip = ixgbe->dip;
770 	mac->m_src_addr = hw->mac.addr;
771 	mac->m_callbacks = &ixgbe_m_callbacks;
772 	mac->m_min_sdu = 0;
773 	mac->m_max_sdu = ixgbe->default_mtu;
774 	mac->m_margin = VLAN_TAGSZ;
775 	mac->m_priv_props = ixgbe_priv_props;
776 	mac->m_v12n = MAC_VIRT_LEVEL1;
777 
778 	status = mac_register(mac, &ixgbe->mac_hdl);
779 
780 	mac_free(mac);
781 
782 	return ((status == 0) ? IXGBE_SUCCESS : IXGBE_FAILURE);
783 }
784 
785 /*
786  * ixgbe_identify_hardware - Identify the type of the chipset.
787  */
788 static int
789 ixgbe_identify_hardware(ixgbe_t *ixgbe)
790 {
791 	struct ixgbe_hw *hw = &ixgbe->hw;
792 	struct ixgbe_osdep *osdep = &ixgbe->osdep;
793 
794 	/*
795 	 * Get the device id
796 	 */
797 	hw->vendor_id =
798 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
799 	hw->device_id =
800 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
801 	hw->revision_id =
802 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
803 	hw->subsystem_device_id =
804 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
805 	hw->subsystem_vendor_id =
806 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
807 
808 	/*
809 	 * Set the mac type of the adapter based on the device id
810 	 */
811 	if (ixgbe_set_mac_type(hw) != IXGBE_SUCCESS) {
812 		return (IXGBE_FAILURE);
813 	}
814 
815 	/*
816 	 * Install adapter capabilities
817 	 */
818 	switch (hw->mac.type) {
819 	case ixgbe_mac_82598EB:
820 		IXGBE_DEBUGLOG_0(ixgbe, "identify 82598 adapter\n");
821 		ixgbe->capab = &ixgbe_82598eb_cap;
822 
823 		if (ixgbe_get_media_type(hw) == ixgbe_media_type_copper) {
824 			ixgbe->capab->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
825 			ixgbe->capab->other_intr |= IXGBE_EICR_GPI_SDP1;
826 		}
827 		ixgbe->capab->other_intr |= IXGBE_EICR_LSC;
828 
829 		break;
830 	case ixgbe_mac_82599EB:
831 		IXGBE_DEBUGLOG_0(ixgbe, "identify 82599 adapter\n");
832 		ixgbe->capab = &ixgbe_82599eb_cap;
833 
834 		ixgbe->capab->other_intr = (IXGBE_EICR_GPI_SDP1 |
835 		    IXGBE_EICR_GPI_SDP2 | IXGBE_EICR_LSC);
836 
837 		break;
838 	default:
839 		IXGBE_DEBUGLOG_1(ixgbe,
840 		    "adapter not supported in ixgbe_identify_hardware(): %d\n",
841 		    hw->mac.type);
842 		return (IXGBE_FAILURE);
843 	}
844 
845 	return (IXGBE_SUCCESS);
846 }
847 
848 /*
849  * ixgbe_regs_map - Map the device registers.
850  *
851  */
852 static int
853 ixgbe_regs_map(ixgbe_t *ixgbe)
854 {
855 	dev_info_t *devinfo = ixgbe->dip;
856 	struct ixgbe_hw *hw = &ixgbe->hw;
857 	struct ixgbe_osdep *osdep = &ixgbe->osdep;
858 	off_t mem_size;
859 
860 	/*
861 	 * First get the size of device registers to be mapped.
862 	 */
863 	if (ddi_dev_regsize(devinfo, IXGBE_ADAPTER_REGSET, &mem_size)
864 	    != DDI_SUCCESS) {
865 		return (IXGBE_FAILURE);
866 	}
867 
868 	/*
869 	 * Call ddi_regs_map_setup() to map registers
870 	 */
871 	if ((ddi_regs_map_setup(devinfo, IXGBE_ADAPTER_REGSET,
872 	    (caddr_t *)&hw->hw_addr, 0,
873 	    mem_size, &ixgbe_regs_acc_attr,
874 	    &osdep->reg_handle)) != DDI_SUCCESS) {
875 		return (IXGBE_FAILURE);
876 	}
877 
878 	return (IXGBE_SUCCESS);
879 }
880 
881 /*
882  * ixgbe_init_properties - Initialize driver properties.
883  */
884 static void
885 ixgbe_init_properties(ixgbe_t *ixgbe)
886 {
887 	/*
888 	 * Get conf file properties, including link settings
889 	 * jumbo frames, ring number, descriptor number, etc.
890 	 */
891 	ixgbe_get_conf(ixgbe);
892 
893 	ixgbe_init_params(ixgbe);
894 }
895 
896 /*
897  * ixgbe_init_driver_settings - Initialize driver settings.
898  *
899  * The settings include hardware function pointers, bus information,
900  * rx/tx rings settings, link state, and any other parameters that
901  * need to be setup during driver initialization.
902  */
903 static int
904 ixgbe_init_driver_settings(ixgbe_t *ixgbe)
905 {
906 	struct ixgbe_hw *hw = &ixgbe->hw;
907 	dev_info_t *devinfo = ixgbe->dip;
908 	ixgbe_rx_ring_t *rx_ring;
909 	ixgbe_rx_group_t *rx_group;
910 	ixgbe_tx_ring_t *tx_ring;
911 	uint32_t rx_size;
912 	uint32_t tx_size;
913 	uint32_t ring_per_group;
914 	int i;
915 
916 	/*
917 	 * Initialize chipset specific hardware function pointers
918 	 */
919 	if (ixgbe_init_shared_code(hw) != IXGBE_SUCCESS) {
920 		return (IXGBE_FAILURE);
921 	}
922 
923 	/*
924 	 * Get the system page size
925 	 */
926 	ixgbe->sys_page_size = ddi_ptob(devinfo, (ulong_t)1);
927 
928 	/*
929 	 * Set rx buffer size
930 	 *
931 	 * The IP header alignment room is counted in the calculation.
932 	 * The rx buffer size is in unit of 1K that is required by the
933 	 * chipset hardware.
934 	 */
935 	rx_size = ixgbe->max_frame_size + IPHDR_ALIGN_ROOM;
936 	ixgbe->rx_buf_size = ((rx_size >> 10) +
937 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
938 
939 	/*
940 	 * Set tx buffer size
941 	 */
942 	tx_size = ixgbe->max_frame_size;
943 	ixgbe->tx_buf_size = ((tx_size >> 10) +
944 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
945 
946 	/*
947 	 * Initialize rx/tx rings/groups parameters
948 	 */
949 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
950 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
951 		rx_ring = &ixgbe->rx_rings[i];
952 		rx_ring->index = i;
953 		rx_ring->ixgbe = ixgbe;
954 		rx_ring->group_index = i / ring_per_group;
955 		rx_ring->hw_index = ixgbe_get_hw_rx_index(ixgbe, i);
956 	}
957 
958 	for (i = 0; i < ixgbe->num_rx_groups; i++) {
959 		rx_group = &ixgbe->rx_groups[i];
960 		rx_group->index = i;
961 		rx_group->ixgbe = ixgbe;
962 	}
963 
964 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
965 		tx_ring = &ixgbe->tx_rings[i];
966 		tx_ring->index = i;
967 		tx_ring->ixgbe = ixgbe;
968 		if (ixgbe->tx_head_wb_enable)
969 			tx_ring->tx_recycle = ixgbe_tx_recycle_head_wb;
970 		else
971 			tx_ring->tx_recycle = ixgbe_tx_recycle_legacy;
972 
973 		tx_ring->ring_size = ixgbe->tx_ring_size;
974 		tx_ring->free_list_size = ixgbe->tx_ring_size +
975 		    (ixgbe->tx_ring_size >> 1);
976 	}
977 
978 	/*
979 	 * Initialize values of interrupt throttling rate
980 	 */
981 	for (i = 1; i < MAX_INTR_VECTOR; i++)
982 		ixgbe->intr_throttling[i] = ixgbe->intr_throttling[0];
983 
984 	/*
985 	 * The initial link state should be "unknown"
986 	 */
987 	ixgbe->link_state = LINK_STATE_UNKNOWN;
988 
989 	return (IXGBE_SUCCESS);
990 }
991 
992 /*
993  * ixgbe_init_locks - Initialize locks.
994  */
995 static void
996 ixgbe_init_locks(ixgbe_t *ixgbe)
997 {
998 	ixgbe_rx_ring_t *rx_ring;
999 	ixgbe_tx_ring_t *tx_ring;
1000 	int i;
1001 
1002 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1003 		rx_ring = &ixgbe->rx_rings[i];
1004 		mutex_init(&rx_ring->rx_lock, NULL,
1005 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1006 	}
1007 
1008 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1009 		tx_ring = &ixgbe->tx_rings[i];
1010 		mutex_init(&tx_ring->tx_lock, NULL,
1011 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1012 		mutex_init(&tx_ring->recycle_lock, NULL,
1013 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1014 		mutex_init(&tx_ring->tcb_head_lock, NULL,
1015 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1016 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
1017 		    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1018 	}
1019 
1020 	mutex_init(&ixgbe->gen_lock, NULL,
1021 	    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1022 
1023 	mutex_init(&ixgbe->watchdog_lock, NULL,
1024 	    MUTEX_DRIVER, DDI_INTR_PRI(ixgbe->intr_pri));
1025 }
1026 
1027 /*
1028  * ixgbe_destroy_locks - Destroy locks.
1029  */
1030 static void
1031 ixgbe_destroy_locks(ixgbe_t *ixgbe)
1032 {
1033 	ixgbe_rx_ring_t *rx_ring;
1034 	ixgbe_tx_ring_t *tx_ring;
1035 	int i;
1036 
1037 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1038 		rx_ring = &ixgbe->rx_rings[i];
1039 		mutex_destroy(&rx_ring->rx_lock);
1040 	}
1041 
1042 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1043 		tx_ring = &ixgbe->tx_rings[i];
1044 		mutex_destroy(&tx_ring->tx_lock);
1045 		mutex_destroy(&tx_ring->recycle_lock);
1046 		mutex_destroy(&tx_ring->tcb_head_lock);
1047 		mutex_destroy(&tx_ring->tcb_tail_lock);
1048 	}
1049 
1050 	mutex_destroy(&ixgbe->gen_lock);
1051 	mutex_destroy(&ixgbe->watchdog_lock);
1052 }
1053 
1054 static int
1055 ixgbe_resume(dev_info_t *devinfo)
1056 {
1057 	ixgbe_t *ixgbe;
1058 	int i;
1059 
1060 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
1061 	if (ixgbe == NULL)
1062 		return (DDI_FAILURE);
1063 
1064 	mutex_enter(&ixgbe->gen_lock);
1065 
1066 	if (ixgbe->ixgbe_state & IXGBE_STARTED) {
1067 		if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
1068 			mutex_exit(&ixgbe->gen_lock);
1069 			return (DDI_FAILURE);
1070 		}
1071 
1072 		/*
1073 		 * Enable and start the watchdog timer
1074 		 */
1075 		ixgbe_enable_watchdog_timer(ixgbe);
1076 	}
1077 
1078 	atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_SUSPENDED);
1079 
1080 	if (ixgbe->ixgbe_state & IXGBE_STARTED) {
1081 		for (i = 0; i < ixgbe->num_tx_rings; i++) {
1082 			mac_tx_ring_update(ixgbe->mac_hdl,
1083 			    ixgbe->tx_rings[i].ring_handle);
1084 		}
1085 	}
1086 
1087 	mutex_exit(&ixgbe->gen_lock);
1088 
1089 	return (DDI_SUCCESS);
1090 }
1091 
1092 static int
1093 ixgbe_suspend(dev_info_t *devinfo)
1094 {
1095 	ixgbe_t *ixgbe;
1096 
1097 	ixgbe = (ixgbe_t *)ddi_get_driver_private(devinfo);
1098 	if (ixgbe == NULL)
1099 		return (DDI_FAILURE);
1100 
1101 	mutex_enter(&ixgbe->gen_lock);
1102 
1103 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_SUSPENDED);
1104 	if (!(ixgbe->ixgbe_state & IXGBE_STARTED)) {
1105 		mutex_exit(&ixgbe->gen_lock);
1106 		return (DDI_SUCCESS);
1107 	}
1108 	ixgbe_stop(ixgbe, B_FALSE);
1109 
1110 	mutex_exit(&ixgbe->gen_lock);
1111 
1112 	/*
1113 	 * Disable and stop the watchdog timer
1114 	 */
1115 	ixgbe_disable_watchdog_timer(ixgbe);
1116 
1117 	return (DDI_SUCCESS);
1118 }
1119 
1120 /*
1121  * ixgbe_init - Initialize the device.
1122  */
1123 static int
1124 ixgbe_init(ixgbe_t *ixgbe)
1125 {
1126 	struct ixgbe_hw *hw = &ixgbe->hw;
1127 
1128 	mutex_enter(&ixgbe->gen_lock);
1129 
1130 	/*
1131 	 * Reset chipset to put the hardware in a known state
1132 	 * before we try to do anything with the eeprom.
1133 	 */
1134 	if (ixgbe_reset_hw(hw) != IXGBE_SUCCESS) {
1135 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1136 		goto init_fail;
1137 	}
1138 
1139 	/*
1140 	 * Need to init eeprom before validating the checksum.
1141 	 */
1142 	if (ixgbe_init_eeprom_params(hw) < 0) {
1143 		ixgbe_error(ixgbe,
1144 		    "Unable to intitialize the eeprom interface.");
1145 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1146 		goto init_fail;
1147 	}
1148 
1149 	/*
1150 	 * NVM validation
1151 	 */
1152 	if (ixgbe_validate_eeprom_checksum(hw, NULL) < 0) {
1153 		/*
1154 		 * Some PCI-E parts fail the first check due to
1155 		 * the link being in sleep state.  Call it again,
1156 		 * if it fails a second time it's a real issue.
1157 		 */
1158 		if (ixgbe_validate_eeprom_checksum(hw, NULL) < 0) {
1159 			ixgbe_error(ixgbe,
1160 			    "Invalid NVM checksum. Please contact "
1161 			    "the vendor to update the NVM.");
1162 			ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1163 			goto init_fail;
1164 		}
1165 	}
1166 
1167 	/*
1168 	 * Setup default flow control thresholds - enable/disable
1169 	 * & flow control type is controlled by ixgbe.conf
1170 	 */
1171 	hw->fc.high_water = DEFAULT_FCRTH;
1172 	hw->fc.low_water = DEFAULT_FCRTL;
1173 	hw->fc.pause_time = DEFAULT_FCPAUSE;
1174 	hw->fc.send_xon = B_TRUE;
1175 
1176 	/*
1177 	 * Initialize link settings
1178 	 */
1179 	(void) ixgbe_driver_setup_link(ixgbe, B_FALSE);
1180 
1181 	/*
1182 	 * Initialize the chipset hardware
1183 	 */
1184 	if (ixgbe_chip_start(ixgbe) != IXGBE_SUCCESS) {
1185 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1186 		goto init_fail;
1187 	}
1188 
1189 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1190 		goto init_fail;
1191 	}
1192 
1193 	mutex_exit(&ixgbe->gen_lock);
1194 	return (IXGBE_SUCCESS);
1195 
1196 init_fail:
1197 	/*
1198 	 * Reset PHY
1199 	 */
1200 	(void) ixgbe_reset_phy(hw);
1201 
1202 	mutex_exit(&ixgbe->gen_lock);
1203 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1204 	return (IXGBE_FAILURE);
1205 }
1206 
1207 /*
1208  * ixgbe_chip_start - Initialize and start the chipset hardware.
1209  */
1210 static int
1211 ixgbe_chip_start(ixgbe_t *ixgbe)
1212 {
1213 	struct ixgbe_hw *hw = &ixgbe->hw;
1214 	int ret_val, i;
1215 
1216 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1217 
1218 	/*
1219 	 * Get the mac address
1220 	 * This function should handle SPARC case correctly.
1221 	 */
1222 	if (!ixgbe_find_mac_address(ixgbe)) {
1223 		ixgbe_error(ixgbe, "Failed to get the mac address");
1224 		return (IXGBE_FAILURE);
1225 	}
1226 
1227 	/*
1228 	 * Validate the mac address
1229 	 */
1230 	(void) ixgbe_init_rx_addrs(hw);
1231 	if (!is_valid_mac_addr(hw->mac.addr)) {
1232 		ixgbe_error(ixgbe, "Invalid mac address");
1233 		return (IXGBE_FAILURE);
1234 	}
1235 
1236 	/*
1237 	 * Configure/Initialize hardware
1238 	 */
1239 	ret_val = ixgbe_init_hw(hw);
1240 	if (ret_val != IXGBE_SUCCESS) {
1241 		if (ret_val == IXGBE_ERR_EEPROM_VERSION) {
1242 			ixgbe_error(ixgbe,
1243 			    "This 82599 device is pre-release and contains"
1244 			    " outdated firmware, please contact your hardware"
1245 			    " vendor for a replacement.");
1246 		} else {
1247 			ixgbe_error(ixgbe, "Failed to initialize hardware");
1248 			return (IXGBE_FAILURE);
1249 		}
1250 	}
1251 
1252 	/*
1253 	 * Re-enable relaxed ordering for performance.  It is disabled
1254 	 * by default in the hardware init.
1255 	 */
1256 	ixgbe_enable_relaxed_ordering(hw);
1257 
1258 	/*
1259 	 * Setup adapter interrupt vectors
1260 	 */
1261 	ixgbe_setup_adapter_vector(ixgbe);
1262 
1263 	/*
1264 	 * Initialize unicast addresses.
1265 	 */
1266 	ixgbe_init_unicst(ixgbe);
1267 
1268 	/*
1269 	 * Setup and initialize the mctable structures.
1270 	 */
1271 	ixgbe_setup_multicst(ixgbe);
1272 
1273 	/*
1274 	 * Set interrupt throttling rate
1275 	 */
1276 	for (i = 0; i < ixgbe->intr_cnt; i++) {
1277 		IXGBE_WRITE_REG(hw, IXGBE_EITR(i), ixgbe->intr_throttling[i]);
1278 	}
1279 
1280 	/*
1281 	 * Save the state of the phy
1282 	 */
1283 	ixgbe_get_hw_state(ixgbe);
1284 
1285 	/*
1286 	 * Make sure driver has control
1287 	 */
1288 	ixgbe_get_driver_control(hw);
1289 
1290 	return (IXGBE_SUCCESS);
1291 }
1292 
1293 /*
1294  * ixgbe_chip_stop - Stop the chipset hardware
1295  */
1296 static void
1297 ixgbe_chip_stop(ixgbe_t *ixgbe)
1298 {
1299 	struct ixgbe_hw *hw = &ixgbe->hw;
1300 
1301 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1302 
1303 	/*
1304 	 * Tell firmware driver is no longer in control
1305 	 */
1306 	ixgbe_release_driver_control(hw);
1307 
1308 	/*
1309 	 * Reset the chipset
1310 	 */
1311 	(void) ixgbe_reset_hw(hw);
1312 
1313 	/*
1314 	 * Reset PHY
1315 	 */
1316 	(void) ixgbe_reset_phy(hw);
1317 }
1318 
1319 /*
1320  * ixgbe_reset - Reset the chipset and re-start the driver.
1321  *
1322  * It involves stopping and re-starting the chipset,
1323  * and re-configuring the rx/tx rings.
1324  */
1325 static int
1326 ixgbe_reset(ixgbe_t *ixgbe)
1327 {
1328 	int i;
1329 
1330 	/*
1331 	 * Disable and stop the watchdog timer
1332 	 */
1333 	ixgbe_disable_watchdog_timer(ixgbe);
1334 
1335 	mutex_enter(&ixgbe->gen_lock);
1336 
1337 	ASSERT(ixgbe->ixgbe_state & IXGBE_STARTED);
1338 	atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_STARTED);
1339 
1340 	ixgbe_stop(ixgbe, B_FALSE);
1341 
1342 	if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
1343 		mutex_exit(&ixgbe->gen_lock);
1344 		return (IXGBE_FAILURE);
1345 	}
1346 
1347 	/*
1348 	 * After resetting, need to recheck the link status.
1349 	 */
1350 	ixgbe->link_check_complete = B_FALSE;
1351 	ixgbe->link_check_hrtime = gethrtime() +
1352 	    (IXGBE_LINK_UP_TIME * 100000000ULL);
1353 
1354 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_STARTED);
1355 
1356 	if (!(ixgbe->ixgbe_state & IXGBE_SUSPENDED)) {
1357 		for (i = 0; i < ixgbe->num_tx_rings; i++) {
1358 			mac_tx_ring_update(ixgbe->mac_hdl,
1359 			    ixgbe->tx_rings[i].ring_handle);
1360 		}
1361 	}
1362 
1363 	mutex_exit(&ixgbe->gen_lock);
1364 
1365 	/*
1366 	 * Enable and start the watchdog timer
1367 	 */
1368 	ixgbe_enable_watchdog_timer(ixgbe);
1369 
1370 	return (IXGBE_SUCCESS);
1371 }
1372 
1373 /*
1374  * ixgbe_tx_clean - Clean the pending transmit packets and DMA resources.
1375  */
1376 static void
1377 ixgbe_tx_clean(ixgbe_t *ixgbe)
1378 {
1379 	ixgbe_tx_ring_t *tx_ring;
1380 	tx_control_block_t *tcb;
1381 	link_list_t pending_list;
1382 	uint32_t desc_num;
1383 	int i, j;
1384 
1385 	LINK_LIST_INIT(&pending_list);
1386 
1387 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1388 		tx_ring = &ixgbe->tx_rings[i];
1389 
1390 		mutex_enter(&tx_ring->recycle_lock);
1391 
1392 		/*
1393 		 * Clean the pending tx data - the pending packets in the
1394 		 * work_list that have no chances to be transmitted again.
1395 		 *
1396 		 * We must ensure the chipset is stopped or the link is down
1397 		 * before cleaning the transmit packets.
1398 		 */
1399 		desc_num = 0;
1400 		for (j = 0; j < tx_ring->ring_size; j++) {
1401 			tcb = tx_ring->work_list[j];
1402 			if (tcb != NULL) {
1403 				desc_num += tcb->desc_num;
1404 
1405 				tx_ring->work_list[j] = NULL;
1406 
1407 				ixgbe_free_tcb(tcb);
1408 
1409 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1410 			}
1411 		}
1412 
1413 		if (desc_num > 0) {
1414 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1415 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1416 
1417 			/*
1418 			 * Reset the head and tail pointers of the tbd ring;
1419 			 * Reset the writeback head if it's enable.
1420 			 */
1421 			tx_ring->tbd_head = 0;
1422 			tx_ring->tbd_tail = 0;
1423 			if (ixgbe->tx_head_wb_enable)
1424 				*tx_ring->tbd_head_wb = 0;
1425 
1426 			IXGBE_WRITE_REG(&ixgbe->hw,
1427 			    IXGBE_TDH(tx_ring->index), 0);
1428 			IXGBE_WRITE_REG(&ixgbe->hw,
1429 			    IXGBE_TDT(tx_ring->index), 0);
1430 		}
1431 
1432 		mutex_exit(&tx_ring->recycle_lock);
1433 
1434 		/*
1435 		 * Add the tx control blocks in the pending list to
1436 		 * the free list.
1437 		 */
1438 		ixgbe_put_free_list(tx_ring, &pending_list);
1439 	}
1440 }
1441 
1442 /*
1443  * ixgbe_tx_drain - Drain the tx rings to allow pending packets to be
1444  * transmitted.
1445  */
1446 static boolean_t
1447 ixgbe_tx_drain(ixgbe_t *ixgbe)
1448 {
1449 	ixgbe_tx_ring_t *tx_ring;
1450 	boolean_t done;
1451 	int i, j;
1452 
1453 	/*
1454 	 * Wait for a specific time to allow pending tx packets
1455 	 * to be transmitted.
1456 	 *
1457 	 * Check the counter tbd_free to see if transmission is done.
1458 	 * No lock protection is needed here.
1459 	 *
1460 	 * Return B_TRUE if all pending packets have been transmitted;
1461 	 * Otherwise return B_FALSE;
1462 	 */
1463 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1464 
1465 		done = B_TRUE;
1466 		for (j = 0; j < ixgbe->num_tx_rings; j++) {
1467 			tx_ring = &ixgbe->tx_rings[j];
1468 			done = done &&
1469 			    (tx_ring->tbd_free == tx_ring->ring_size);
1470 		}
1471 
1472 		if (done)
1473 			break;
1474 
1475 		msec_delay(1);
1476 	}
1477 
1478 	return (done);
1479 }
1480 
1481 /*
1482  * ixgbe_rx_drain - Wait for all rx buffers to be released by upper layer.
1483  */
1484 static boolean_t
1485 ixgbe_rx_drain(ixgbe_t *ixgbe)
1486 {
1487 	boolean_t done = B_TRUE;
1488 	int i;
1489 
1490 	/*
1491 	 * Polling the rx free list to check if those rx buffers held by
1492 	 * the upper layer are released.
1493 	 *
1494 	 * Check the counter rcb_free to see if all pending buffers are
1495 	 * released. No lock protection is needed here.
1496 	 *
1497 	 * Return B_TRUE if all pending buffers have been released;
1498 	 * Otherwise return B_FALSE;
1499 	 */
1500 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1501 		done = (ixgbe->rcb_pending == 0);
1502 
1503 		if (done)
1504 			break;
1505 
1506 		msec_delay(1);
1507 	}
1508 
1509 	return (done);
1510 }
1511 
1512 /*
1513  * ixgbe_start - Start the driver/chipset.
1514  */
1515 int
1516 ixgbe_start(ixgbe_t *ixgbe, boolean_t alloc_buffer)
1517 {
1518 	int i;
1519 
1520 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1521 
1522 	if (alloc_buffer) {
1523 		if (ixgbe_alloc_rx_data(ixgbe) != IXGBE_SUCCESS) {
1524 			ixgbe_error(ixgbe,
1525 			    "Failed to allocate software receive rings");
1526 			return (IXGBE_FAILURE);
1527 		}
1528 
1529 		/* Allocate buffers for all the rx/tx rings */
1530 		if (ixgbe_alloc_dma(ixgbe) != IXGBE_SUCCESS) {
1531 			ixgbe_error(ixgbe, "Failed to allocate DMA resource");
1532 			return (IXGBE_FAILURE);
1533 		}
1534 
1535 		ixgbe->tx_ring_init = B_TRUE;
1536 	} else {
1537 		ixgbe->tx_ring_init = B_FALSE;
1538 	}
1539 
1540 	for (i = 0; i < ixgbe->num_rx_rings; i++)
1541 		mutex_enter(&ixgbe->rx_rings[i].rx_lock);
1542 	for (i = 0; i < ixgbe->num_tx_rings; i++)
1543 		mutex_enter(&ixgbe->tx_rings[i].tx_lock);
1544 
1545 	/*
1546 	 * Start the chipset hardware
1547 	 */
1548 	if (ixgbe_chip_start(ixgbe) != IXGBE_SUCCESS) {
1549 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1550 		goto start_failure;
1551 	}
1552 
1553 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1554 		goto start_failure;
1555 	}
1556 
1557 	/*
1558 	 * Setup the rx/tx rings
1559 	 */
1560 	ixgbe_setup_rings(ixgbe);
1561 
1562 	/*
1563 	 * ixgbe_start() will be called when resetting, however if reset
1564 	 * happens, we need to clear the ERROR and STALL flags before
1565 	 * enabling the interrupts.
1566 	 */
1567 	atomic_and_32(&ixgbe->ixgbe_state, ~(IXGBE_ERROR | IXGBE_STALL));
1568 
1569 	/*
1570 	 * Enable adapter interrupts
1571 	 * The interrupts must be enabled after the driver state is START
1572 	 */
1573 	ixgbe_enable_adapter_interrupts(ixgbe);
1574 
1575 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1576 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1577 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1578 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1579 
1580 	return (IXGBE_SUCCESS);
1581 
1582 start_failure:
1583 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1584 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1585 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1586 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1587 
1588 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1589 
1590 	return (IXGBE_FAILURE);
1591 }
1592 
1593 /*
1594  * ixgbe_stop - Stop the driver/chipset.
1595  */
1596 void
1597 ixgbe_stop(ixgbe_t *ixgbe, boolean_t free_buffer)
1598 {
1599 	int i;
1600 
1601 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1602 
1603 	/*
1604 	 * Disable the adapter interrupts
1605 	 */
1606 	ixgbe_disable_adapter_interrupts(ixgbe);
1607 
1608 	/*
1609 	 * Drain the pending tx packets
1610 	 */
1611 	(void) ixgbe_tx_drain(ixgbe);
1612 
1613 	for (i = 0; i < ixgbe->num_rx_rings; i++)
1614 		mutex_enter(&ixgbe->rx_rings[i].rx_lock);
1615 	for (i = 0; i < ixgbe->num_tx_rings; i++)
1616 		mutex_enter(&ixgbe->tx_rings[i].tx_lock);
1617 
1618 	/*
1619 	 * Stop the chipset hardware
1620 	 */
1621 	ixgbe_chip_stop(ixgbe);
1622 
1623 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1624 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1625 	}
1626 
1627 	/*
1628 	 * Clean the pending tx data/resources
1629 	 */
1630 	ixgbe_tx_clean(ixgbe);
1631 
1632 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1633 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1634 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1635 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1636 
1637 	if (ixgbe->link_state == LINK_STATE_UP) {
1638 		ixgbe->link_state = LINK_STATE_UNKNOWN;
1639 		mac_link_update(ixgbe->mac_hdl, ixgbe->link_state);
1640 	}
1641 
1642 	if (free_buffer) {
1643 		/*
1644 		 * Release the DMA/memory resources of rx/tx rings
1645 		 */
1646 		ixgbe_free_dma(ixgbe);
1647 		ixgbe_free_rx_data(ixgbe);
1648 	}
1649 }
1650 
1651 /*
1652  * ixgbe_cbfunc - Driver interface for generic DDI callbacks
1653  */
1654 /* ARGSUSED */
1655 static int
1656 ixgbe_cbfunc(dev_info_t *dip, ddi_cb_action_t cbaction, void *cbarg,
1657     void *arg1, void *arg2)
1658 {
1659 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
1660 
1661 	switch (cbaction) {
1662 	/* IRM callback */
1663 	int count;
1664 	case DDI_CB_INTR_ADD:
1665 	case DDI_CB_INTR_REMOVE:
1666 		count = (int)(uintptr_t)cbarg;
1667 		ASSERT(ixgbe->intr_type == DDI_INTR_TYPE_MSIX);
1668 		DTRACE_PROBE2(ixgbe__irm__callback, int, count,
1669 		    int, ixgbe->intr_cnt);
1670 		if (ixgbe_intr_adjust(ixgbe, cbaction, count) !=
1671 		    DDI_SUCCESS) {
1672 			ixgbe_error(ixgbe,
1673 			    "IRM CB: Failed to adjust interrupts");
1674 			goto cb_fail;
1675 		}
1676 		break;
1677 	default:
1678 		IXGBE_DEBUGLOG_1(ixgbe, "DDI CB: action 0x%x NOT supported",
1679 		    cbaction);
1680 		return (DDI_ENOTSUP);
1681 	}
1682 	return (DDI_SUCCESS);
1683 cb_fail:
1684 	return (DDI_FAILURE);
1685 }
1686 
1687 /*
1688  * ixgbe_intr_adjust - Adjust interrupt to respond to IRM request.
1689  */
1690 static int
1691 ixgbe_intr_adjust(ixgbe_t *ixgbe, ddi_cb_action_t cbaction, int count)
1692 {
1693 	int i, rc, actual;
1694 
1695 	if (count == 0)
1696 		return (DDI_SUCCESS);
1697 
1698 	if ((cbaction == DDI_CB_INTR_ADD &&
1699 	    ixgbe->intr_cnt + count > ixgbe->intr_cnt_max) ||
1700 	    (cbaction == DDI_CB_INTR_REMOVE &&
1701 	    ixgbe->intr_cnt - count < ixgbe->intr_cnt_min))
1702 		return (DDI_FAILURE);
1703 
1704 	if (!(ixgbe->ixgbe_state & IXGBE_STARTED)) {
1705 		return (DDI_FAILURE);
1706 	}
1707 
1708 	for (i = 0; i < ixgbe->num_rx_rings; i++)
1709 		mac_ring_intr_set(ixgbe->rx_rings[i].ring_handle, NULL);
1710 	for (i = 0; i < ixgbe->num_tx_rings; i++)
1711 		mac_ring_intr_set(ixgbe->tx_rings[i].ring_handle, NULL);
1712 
1713 	mutex_enter(&ixgbe->gen_lock);
1714 	ixgbe->ixgbe_state &= ~IXGBE_STARTED;
1715 	ixgbe->ixgbe_state |= IXGBE_INTR_ADJUST;
1716 	ixgbe->ixgbe_state |= IXGBE_SUSPENDED;
1717 	mac_link_update(ixgbe->mac_hdl, LINK_STATE_UNKNOWN);
1718 
1719 	ixgbe_stop(ixgbe, B_FALSE);
1720 	/*
1721 	 * Disable interrupts
1722 	 */
1723 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
1724 		rc = ixgbe_disable_intrs(ixgbe);
1725 		ASSERT(rc == IXGBE_SUCCESS);
1726 	}
1727 	ixgbe->attach_progress &= ~ATTACH_PROGRESS_ENABLE_INTR;
1728 
1729 	/*
1730 	 * Remove interrupt handlers
1731 	 */
1732 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
1733 		ixgbe_rem_intr_handlers(ixgbe);
1734 	}
1735 	ixgbe->attach_progress &= ~ATTACH_PROGRESS_ADD_INTR;
1736 
1737 	/*
1738 	 * Clear vect_map
1739 	 */
1740 	bzero(&ixgbe->vect_map, sizeof (ixgbe->vect_map));
1741 	switch (cbaction) {
1742 	case DDI_CB_INTR_ADD:
1743 		rc = ddi_intr_alloc(ixgbe->dip, ixgbe->htable,
1744 		    DDI_INTR_TYPE_MSIX, ixgbe->intr_cnt, count, &actual,
1745 		    DDI_INTR_ALLOC_NORMAL);
1746 		if (rc != DDI_SUCCESS || actual != count) {
1747 			ixgbe_log(ixgbe, "Adjust interrupts failed."
1748 			    "return: %d, irm cb size: %d, actual: %d",
1749 			    rc, count, actual);
1750 			goto intr_adjust_fail;
1751 		}
1752 		ixgbe->intr_cnt += count;
1753 		break;
1754 
1755 	case DDI_CB_INTR_REMOVE:
1756 		for (i = ixgbe->intr_cnt - count;
1757 		    i < ixgbe->intr_cnt; i ++) {
1758 			rc = ddi_intr_free(ixgbe->htable[i]);
1759 			ixgbe->htable[i] = NULL;
1760 			if (rc != DDI_SUCCESS) {
1761 				ixgbe_log(ixgbe, "Adjust interrupts failed."
1762 				    "return: %d, irm cb size: %d, actual: %d",
1763 				    rc, count, actual);
1764 				goto intr_adjust_fail;
1765 			}
1766 		}
1767 		ixgbe->intr_cnt -= count;
1768 		break;
1769 	}
1770 
1771 	/*
1772 	 * Get priority for first vector, assume remaining are all the same
1773 	 */
1774 	rc = ddi_intr_get_pri(ixgbe->htable[0], &ixgbe->intr_pri);
1775 	if (rc != DDI_SUCCESS) {
1776 		ixgbe_log(ixgbe,
1777 		    "Get interrupt priority failed: %d", rc);
1778 		goto intr_adjust_fail;
1779 	}
1780 	rc = ddi_intr_get_cap(ixgbe->htable[0], &ixgbe->intr_cap);
1781 	if (rc != DDI_SUCCESS) {
1782 		ixgbe_log(ixgbe, "Get interrupt cap failed: %d", rc);
1783 		goto intr_adjust_fail;
1784 	}
1785 	ixgbe->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
1786 
1787 	/*
1788 	 * Map rings to interrupt vectors
1789 	 */
1790 	if (ixgbe_map_intrs_to_vectors(ixgbe) != IXGBE_SUCCESS) {
1791 		ixgbe_error(ixgbe,
1792 		    "IRM CB: Failed to map interrupts to vectors");
1793 		goto intr_adjust_fail;
1794 	}
1795 
1796 	/*
1797 	 * Add interrupt handlers
1798 	 */
1799 	if (ixgbe_add_intr_handlers(ixgbe) != IXGBE_SUCCESS) {
1800 		ixgbe_error(ixgbe, "IRM CB: Failed to add interrupt handlers");
1801 		goto intr_adjust_fail;
1802 	}
1803 	ixgbe->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
1804 
1805 	/*
1806 	 * Now that mutex locks are initialized, and the chip is also
1807 	 * initialized, enable interrupts.
1808 	 */
1809 	if (ixgbe_enable_intrs(ixgbe) != IXGBE_SUCCESS) {
1810 		ixgbe_error(ixgbe, "IRM CB: Failed to enable DDI interrupts");
1811 		goto intr_adjust_fail;
1812 	}
1813 	ixgbe->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
1814 	if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
1815 		ixgbe_error(ixgbe, "IRM CB: Failed to start");
1816 		goto intr_adjust_fail;
1817 	}
1818 	ixgbe->ixgbe_state &= ~IXGBE_INTR_ADJUST;
1819 	ixgbe->ixgbe_state &= ~IXGBE_SUSPENDED;
1820 	ixgbe->ixgbe_state |= IXGBE_STARTED;
1821 	mutex_exit(&ixgbe->gen_lock);
1822 
1823 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1824 		mac_ring_intr_set(ixgbe->rx_rings[i].ring_handle,
1825 		    ixgbe->htable[ixgbe->rx_rings[i].intr_vector]);
1826 	}
1827 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1828 		mac_ring_intr_set(ixgbe->tx_rings[i].ring_handle,
1829 		    ixgbe->htable[ixgbe->tx_rings[i].intr_vector]);
1830 	}
1831 
1832 	/* Wakeup all Tx rings */
1833 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1834 		mac_tx_ring_update(ixgbe->mac_hdl,
1835 		    ixgbe->tx_rings[i].ring_handle);
1836 	}
1837 
1838 	IXGBE_DEBUGLOG_3(ixgbe,
1839 	    "IRM CB: interrupts new value: 0x%x(0x%x:0x%x).",
1840 	    ixgbe->intr_cnt, ixgbe->intr_cnt_min, ixgbe->intr_cnt_max);
1841 	return (DDI_SUCCESS);
1842 
1843 intr_adjust_fail:
1844 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1845 	mutex_exit(&ixgbe->gen_lock);
1846 	return (DDI_FAILURE);
1847 }
1848 
1849 /*
1850  * ixgbe_intr_cb_register - Register interrupt callback function.
1851  */
1852 static int
1853 ixgbe_intr_cb_register(ixgbe_t *ixgbe)
1854 {
1855 	if (ddi_cb_register(ixgbe->dip, DDI_CB_FLAG_INTR, ixgbe_cbfunc,
1856 	    ixgbe, NULL, &ixgbe->cb_hdl) != DDI_SUCCESS) {
1857 		return (IXGBE_FAILURE);
1858 	}
1859 	IXGBE_DEBUGLOG_0(ixgbe, "Interrupt callback function registered.");
1860 	return (IXGBE_SUCCESS);
1861 }
1862 
1863 /*
1864  * ixgbe_alloc_rings - Allocate memory space for rx/tx rings.
1865  */
1866 static int
1867 ixgbe_alloc_rings(ixgbe_t *ixgbe)
1868 {
1869 	/*
1870 	 * Allocate memory space for rx rings
1871 	 */
1872 	ixgbe->rx_rings = kmem_zalloc(
1873 	    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings,
1874 	    KM_NOSLEEP);
1875 
1876 	if (ixgbe->rx_rings == NULL) {
1877 		return (IXGBE_FAILURE);
1878 	}
1879 
1880 	/*
1881 	 * Allocate memory space for tx rings
1882 	 */
1883 	ixgbe->tx_rings = kmem_zalloc(
1884 	    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings,
1885 	    KM_NOSLEEP);
1886 
1887 	if (ixgbe->tx_rings == NULL) {
1888 		kmem_free(ixgbe->rx_rings,
1889 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
1890 		ixgbe->rx_rings = NULL;
1891 		return (IXGBE_FAILURE);
1892 	}
1893 
1894 	/*
1895 	 * Allocate memory space for rx ring groups
1896 	 */
1897 	ixgbe->rx_groups = kmem_zalloc(
1898 	    sizeof (ixgbe_rx_group_t) * ixgbe->num_rx_groups,
1899 	    KM_NOSLEEP);
1900 
1901 	if (ixgbe->rx_groups == NULL) {
1902 		kmem_free(ixgbe->rx_rings,
1903 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
1904 		kmem_free(ixgbe->tx_rings,
1905 		    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings);
1906 		ixgbe->rx_rings = NULL;
1907 		ixgbe->tx_rings = NULL;
1908 		return (IXGBE_FAILURE);
1909 	}
1910 
1911 	return (IXGBE_SUCCESS);
1912 }
1913 
1914 /*
1915  * ixgbe_free_rings - Free the memory space of rx/tx rings.
1916  */
1917 static void
1918 ixgbe_free_rings(ixgbe_t *ixgbe)
1919 {
1920 	if (ixgbe->rx_rings != NULL) {
1921 		kmem_free(ixgbe->rx_rings,
1922 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
1923 		ixgbe->rx_rings = NULL;
1924 	}
1925 
1926 	if (ixgbe->tx_rings != NULL) {
1927 		kmem_free(ixgbe->tx_rings,
1928 		    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings);
1929 		ixgbe->tx_rings = NULL;
1930 	}
1931 
1932 	if (ixgbe->rx_groups != NULL) {
1933 		kmem_free(ixgbe->rx_groups,
1934 		    sizeof (ixgbe_rx_group_t) * ixgbe->num_rx_groups);
1935 		ixgbe->rx_groups = NULL;
1936 	}
1937 }
1938 
1939 static int
1940 ixgbe_alloc_rx_data(ixgbe_t *ixgbe)
1941 {
1942 	ixgbe_rx_ring_t *rx_ring;
1943 	int i;
1944 
1945 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1946 		rx_ring = &ixgbe->rx_rings[i];
1947 		if (ixgbe_alloc_rx_ring_data(rx_ring) != IXGBE_SUCCESS)
1948 			goto alloc_rx_rings_failure;
1949 	}
1950 	return (IXGBE_SUCCESS);
1951 
1952 alloc_rx_rings_failure:
1953 	ixgbe_free_rx_data(ixgbe);
1954 	return (IXGBE_FAILURE);
1955 }
1956 
1957 static void
1958 ixgbe_free_rx_data(ixgbe_t *ixgbe)
1959 {
1960 	ixgbe_rx_ring_t *rx_ring;
1961 	ixgbe_rx_data_t *rx_data;
1962 	int i;
1963 
1964 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1965 		rx_ring = &ixgbe->rx_rings[i];
1966 
1967 		mutex_enter(&ixgbe->rx_pending_lock);
1968 		rx_data = rx_ring->rx_data;
1969 
1970 		if (rx_data != NULL) {
1971 			rx_data->flag |= IXGBE_RX_STOPPED;
1972 
1973 			if (rx_data->rcb_pending == 0) {
1974 				ixgbe_free_rx_ring_data(rx_data);
1975 				rx_ring->rx_data = NULL;
1976 			}
1977 		}
1978 
1979 		mutex_exit(&ixgbe->rx_pending_lock);
1980 	}
1981 }
1982 
1983 /*
1984  * ixgbe_setup_rings - Setup rx/tx rings.
1985  */
1986 static void
1987 ixgbe_setup_rings(ixgbe_t *ixgbe)
1988 {
1989 	/*
1990 	 * Setup the rx/tx rings, including the following:
1991 	 *
1992 	 * 1. Setup the descriptor ring and the control block buffers;
1993 	 * 2. Initialize necessary registers for receive/transmit;
1994 	 * 3. Initialize software pointers/parameters for receive/transmit;
1995 	 */
1996 	ixgbe_setup_rx(ixgbe);
1997 
1998 	ixgbe_setup_tx(ixgbe);
1999 }
2000 
2001 static void
2002 ixgbe_setup_rx_ring(ixgbe_rx_ring_t *rx_ring)
2003 {
2004 	ixgbe_t *ixgbe = rx_ring->ixgbe;
2005 	ixgbe_rx_data_t *rx_data = rx_ring->rx_data;
2006 	struct ixgbe_hw *hw = &ixgbe->hw;
2007 	rx_control_block_t *rcb;
2008 	union ixgbe_adv_rx_desc	*rbd;
2009 	uint32_t size;
2010 	uint32_t buf_low;
2011 	uint32_t buf_high;
2012 	uint32_t reg_val;
2013 	int i;
2014 
2015 	ASSERT(mutex_owned(&rx_ring->rx_lock));
2016 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2017 
2018 	for (i = 0; i < ixgbe->rx_ring_size; i++) {
2019 		rcb = rx_data->work_list[i];
2020 		rbd = &rx_data->rbd_ring[i];
2021 
2022 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
2023 		rbd->read.hdr_addr = NULL;
2024 	}
2025 
2026 	/*
2027 	 * Initialize the length register
2028 	 */
2029 	size = rx_data->ring_size * sizeof (union ixgbe_adv_rx_desc);
2030 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(rx_ring->hw_index), size);
2031 
2032 	/*
2033 	 * Initialize the base address registers
2034 	 */
2035 	buf_low = (uint32_t)rx_data->rbd_area.dma_address;
2036 	buf_high = (uint32_t)(rx_data->rbd_area.dma_address >> 32);
2037 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(rx_ring->hw_index), buf_high);
2038 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(rx_ring->hw_index), buf_low);
2039 
2040 	/*
2041 	 * Setup head & tail pointers
2042 	 */
2043 	IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->hw_index),
2044 	    rx_data->ring_size - 1);
2045 	IXGBE_WRITE_REG(hw, IXGBE_RDH(rx_ring->hw_index), 0);
2046 
2047 	rx_data->rbd_next = 0;
2048 	rx_data->lro_first = 0;
2049 
2050 	/*
2051 	 * Setup the Receive Descriptor Control Register (RXDCTL)
2052 	 * PTHRESH=32 descriptors (half the internal cache)
2053 	 * HTHRESH=0 descriptors (to minimize latency on fetch)
2054 	 * WTHRESH defaults to 1 (writeback each descriptor)
2055 	 */
2056 	reg_val = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rx_ring->hw_index));
2057 	reg_val |= IXGBE_RXDCTL_ENABLE;	/* enable queue */
2058 
2059 	/* Not a valid value for 82599 */
2060 	if (hw->mac.type < ixgbe_mac_82599EB) {
2061 		reg_val |= 0x0020;	/* pthresh */
2062 	}
2063 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rx_ring->hw_index), reg_val);
2064 
2065 	if (hw->mac.type == ixgbe_mac_82599EB) {
2066 		reg_val = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2067 		reg_val |= (IXGBE_RDRXCTL_CRCSTRIP | IXGBE_RDRXCTL_AGGDIS);
2068 		IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, reg_val);
2069 	}
2070 
2071 	/*
2072 	 * Setup the Split and Replication Receive Control Register.
2073 	 * Set the rx buffer size and the advanced descriptor type.
2074 	 */
2075 	reg_val = (ixgbe->rx_buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) |
2076 	    IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2077 	reg_val |= IXGBE_SRRCTL_DROP_EN;
2078 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rx_ring->hw_index), reg_val);
2079 }
2080 
2081 static void
2082 ixgbe_setup_rx(ixgbe_t *ixgbe)
2083 {
2084 	ixgbe_rx_ring_t *rx_ring;
2085 	struct ixgbe_hw *hw = &ixgbe->hw;
2086 	uint32_t reg_val;
2087 	uint32_t ring_mapping;
2088 	uint32_t i, index;
2089 	uint32_t psrtype_rss_bit;
2090 
2091 	/* PSRTYPE must be configured for 82599 */
2092 	if (ixgbe->classify_mode != IXGBE_CLASSIFY_VMDQ &&
2093 	    ixgbe->classify_mode != IXGBE_CLASSIFY_VMDQ_RSS) {
2094 		reg_val = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
2095 		    IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR;
2096 		reg_val |= IXGBE_PSRTYPE_L2HDR;
2097 		reg_val |= 0x80000000;
2098 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), reg_val);
2099 	} else {
2100 		if (ixgbe->num_rx_groups > 32) {
2101 			psrtype_rss_bit = 0x20000000;
2102 		} else {
2103 			psrtype_rss_bit = 0x40000000;
2104 		}
2105 		for (i = 0; i < ixgbe->capab->max_rx_grp_num; i++) {
2106 			reg_val = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
2107 			    IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR;
2108 			reg_val |= IXGBE_PSRTYPE_L2HDR;
2109 			reg_val |= psrtype_rss_bit;
2110 			IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(i), reg_val);
2111 		}
2112 	}
2113 
2114 	/*
2115 	 * Set filter control in FCTRL to accept broadcast packets and do
2116 	 * not pass pause frames to host.  Flow control settings are already
2117 	 * in this register, so preserve them.
2118 	 */
2119 	reg_val = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2120 	reg_val |= IXGBE_FCTRL_BAM;	/* broadcast accept mode */
2121 	reg_val |= IXGBE_FCTRL_DPF;	/* discard pause frames */
2122 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, reg_val);
2123 
2124 	/*
2125 	 * Hardware checksum settings
2126 	 */
2127 	if (ixgbe->rx_hcksum_enable) {
2128 		reg_val = IXGBE_RXCSUM_IPPCSE;	/* IP checksum */
2129 		IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, reg_val);
2130 	}
2131 
2132 	/*
2133 	 * Setup VMDq and RSS for multiple receive queues
2134 	 */
2135 	switch (ixgbe->classify_mode) {
2136 	case IXGBE_CLASSIFY_RSS:
2137 		/*
2138 		 * One group, only RSS is needed when more than
2139 		 * one ring enabled.
2140 		 */
2141 		ixgbe_setup_rss(ixgbe);
2142 		break;
2143 
2144 	case IXGBE_CLASSIFY_VMDQ:
2145 		/*
2146 		 * Multiple groups, each group has one ring,
2147 		 * only VMDq is needed.
2148 		 */
2149 		ixgbe_setup_vmdq(ixgbe);
2150 		break;
2151 
2152 	case IXGBE_CLASSIFY_VMDQ_RSS:
2153 		/*
2154 		 * Multiple groups and multiple rings, both
2155 		 * VMDq and RSS are needed.
2156 		 */
2157 		ixgbe_setup_vmdq_rss(ixgbe);
2158 		break;
2159 
2160 	default:
2161 		break;
2162 	}
2163 
2164 	/*
2165 	 * Enable the receive unit.  This must be done after filter
2166 	 * control is set in FCTRL.
2167 	 */
2168 	reg_val = (IXGBE_RXCTRL_RXEN	/* Enable Receive Unit */
2169 	    | IXGBE_RXCTRL_DMBYPS);	/* descriptor monitor bypass */
2170 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, reg_val);
2171 
2172 	/*
2173 	 * ixgbe_setup_rx_ring must be called after configuring RXCTRL
2174 	 */
2175 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2176 		rx_ring = &ixgbe->rx_rings[i];
2177 		ixgbe_setup_rx_ring(rx_ring);
2178 	}
2179 
2180 	/*
2181 	 * Setup the per-ring statistics mapping.
2182 	 */
2183 	ring_mapping = 0;
2184 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2185 		index = ixgbe->rx_rings[i].hw_index;
2186 		ring_mapping = IXGBE_READ_REG(hw, IXGBE_RQSMR(index >> 2));
2187 		ring_mapping |= (i & 0xF) << (8 * (index & 0x3));
2188 		IXGBE_WRITE_REG(hw, IXGBE_RQSMR(index >> 2), ring_mapping);
2189 	}
2190 
2191 	/*
2192 	 * The Max Frame Size in MHADD/MAXFRS will be internally increased
2193 	 * by four bytes if the packet has a VLAN field, so includes MTU,
2194 	 * ethernet header and frame check sequence.
2195 	 * Register is MAXFRS in 82599.
2196 	 */
2197 	reg_val = (ixgbe->default_mtu + sizeof (struct ether_header)
2198 	    + ETHERFCSL) << IXGBE_MHADD_MFS_SHIFT;
2199 	IXGBE_WRITE_REG(hw, IXGBE_MHADD, reg_val);
2200 
2201 	/*
2202 	 * Setup Jumbo Frame enable bit
2203 	 */
2204 	if (ixgbe->default_mtu > ETHERMTU) {
2205 		reg_val = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2206 		reg_val |= IXGBE_HLREG0_JUMBOEN;
2207 		IXGBE_WRITE_REG(hw, IXGBE_HLREG0, reg_val);
2208 	}
2209 
2210 	/*
2211 	 * Setup RSC for multiple receive queues.
2212 	 */
2213 	if (ixgbe->lro_enable) {
2214 		for (i = 0; i < ixgbe->num_rx_rings; i++) {
2215 			/*
2216 			 * Make sure rx_buf_size * MAXDESC not greater
2217 			 * than 65535.
2218 			 * Intel recommends 4 for MAXDESC field value.
2219 			 */
2220 			reg_val = IXGBE_READ_REG(hw, IXGBE_RSCCTL(i));
2221 			reg_val |= IXGBE_RSCCTL_RSCEN;
2222 			if (ixgbe->rx_buf_size == IXGBE_PKG_BUF_16k)
2223 				reg_val |= IXGBE_RSCCTL_MAXDESC_1;
2224 			else
2225 				reg_val |= IXGBE_RSCCTL_MAXDESC_4;
2226 			IXGBE_WRITE_REG(hw,  IXGBE_RSCCTL(i), reg_val);
2227 		}
2228 
2229 		reg_val = IXGBE_READ_REG(hw, IXGBE_RSCDBU);
2230 		reg_val |= IXGBE_RSCDBU_RSCACKDIS;
2231 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU, reg_val);
2232 
2233 		reg_val = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2234 		reg_val |= IXGBE_RDRXCTL_RSCACKC;
2235 		reg_val |= IXGBE_RDRXCTL_FCOE_WRFIX;
2236 		reg_val &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2237 
2238 		IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, reg_val);
2239 	}
2240 }
2241 
2242 static void
2243 ixgbe_setup_tx_ring(ixgbe_tx_ring_t *tx_ring)
2244 {
2245 	ixgbe_t *ixgbe = tx_ring->ixgbe;
2246 	struct ixgbe_hw *hw = &ixgbe->hw;
2247 	uint32_t size;
2248 	uint32_t buf_low;
2249 	uint32_t buf_high;
2250 	uint32_t reg_val;
2251 
2252 	ASSERT(mutex_owned(&tx_ring->tx_lock));
2253 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2254 
2255 	/*
2256 	 * Initialize the length register
2257 	 */
2258 	size = tx_ring->ring_size * sizeof (union ixgbe_adv_tx_desc);
2259 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(tx_ring->index), size);
2260 
2261 	/*
2262 	 * Initialize the base address registers
2263 	 */
2264 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
2265 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
2266 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(tx_ring->index), buf_low);
2267 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(tx_ring->index), buf_high);
2268 
2269 	/*
2270 	 * Setup head & tail pointers
2271 	 */
2272 	IXGBE_WRITE_REG(hw, IXGBE_TDH(tx_ring->index), 0);
2273 	IXGBE_WRITE_REG(hw, IXGBE_TDT(tx_ring->index), 0);
2274 
2275 	/*
2276 	 * Setup head write-back
2277 	 */
2278 	if (ixgbe->tx_head_wb_enable) {
2279 		/*
2280 		 * The memory of the head write-back is allocated using
2281 		 * the extra tbd beyond the tail of the tbd ring.
2282 		 */
2283 		tx_ring->tbd_head_wb = (uint32_t *)
2284 		    ((uintptr_t)tx_ring->tbd_area.address + size);
2285 		*tx_ring->tbd_head_wb = 0;
2286 
2287 		buf_low = (uint32_t)
2288 		    (tx_ring->tbd_area.dma_address + size);
2289 		buf_high = (uint32_t)
2290 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
2291 
2292 		/* Set the head write-back enable bit */
2293 		buf_low |= IXGBE_TDWBAL_HEAD_WB_ENABLE;
2294 
2295 		IXGBE_WRITE_REG(hw, IXGBE_TDWBAL(tx_ring->index), buf_low);
2296 		IXGBE_WRITE_REG(hw, IXGBE_TDWBAH(tx_ring->index), buf_high);
2297 
2298 		/*
2299 		 * Turn off relaxed ordering for head write back or it will
2300 		 * cause problems with the tx recycling
2301 		 */
2302 		reg_val = IXGBE_READ_REG(hw,
2303 		    IXGBE_DCA_TXCTRL(tx_ring->index));
2304 		reg_val &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
2305 		IXGBE_WRITE_REG(hw,
2306 		    IXGBE_DCA_TXCTRL(tx_ring->index), reg_val);
2307 	} else {
2308 		tx_ring->tbd_head_wb = NULL;
2309 	}
2310 
2311 	tx_ring->tbd_head = 0;
2312 	tx_ring->tbd_tail = 0;
2313 	tx_ring->tbd_free = tx_ring->ring_size;
2314 
2315 	if (ixgbe->tx_ring_init == B_TRUE) {
2316 		tx_ring->tcb_head = 0;
2317 		tx_ring->tcb_tail = 0;
2318 		tx_ring->tcb_free = tx_ring->free_list_size;
2319 	}
2320 
2321 	/*
2322 	 * Initialize the s/w context structure
2323 	 */
2324 	bzero(&tx_ring->tx_context, sizeof (ixgbe_tx_context_t));
2325 }
2326 
2327 static void
2328 ixgbe_setup_tx(ixgbe_t *ixgbe)
2329 {
2330 	struct ixgbe_hw *hw = &ixgbe->hw;
2331 	ixgbe_tx_ring_t *tx_ring;
2332 	uint32_t reg_val;
2333 	uint32_t ring_mapping;
2334 	int i;
2335 
2336 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2337 		tx_ring = &ixgbe->tx_rings[i];
2338 		ixgbe_setup_tx_ring(tx_ring);
2339 	}
2340 
2341 	/*
2342 	 * Setup the per-ring statistics mapping.
2343 	 */
2344 	ring_mapping = 0;
2345 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2346 		ring_mapping |= (i & 0xF) << (8 * (i & 0x3));
2347 		if ((i & 0x3) == 0x3) {
2348 			if (hw->mac.type >= ixgbe_mac_82599EB) {
2349 				IXGBE_WRITE_REG(hw, IXGBE_TQSM(i >> 2),
2350 				    ring_mapping);
2351 			} else {
2352 				IXGBE_WRITE_REG(hw, IXGBE_TQSMR(i >> 2),
2353 				    ring_mapping);
2354 			}
2355 			ring_mapping = 0;
2356 		}
2357 	}
2358 	if ((i & 0x3) != 0x3)
2359 		if (hw->mac.type >= ixgbe_mac_82599EB) {
2360 			IXGBE_WRITE_REG(hw, IXGBE_TQSM(i >> 2), ring_mapping);
2361 		} else {
2362 			IXGBE_WRITE_REG(hw, IXGBE_TQSMR(i >> 2), ring_mapping);
2363 		}
2364 
2365 	/*
2366 	 * Enable CRC appending and TX padding (for short tx frames)
2367 	 */
2368 	reg_val = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2369 	reg_val |= IXGBE_HLREG0_TXCRCEN | IXGBE_HLREG0_TXPADEN;
2370 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, reg_val);
2371 
2372 	/*
2373 	 * enable DMA for 82599 parts
2374 	 */
2375 	if (hw->mac.type == ixgbe_mac_82599EB) {
2376 	/* DMATXCTL.TE must be set after all Tx config is complete */
2377 		reg_val = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2378 		reg_val |= IXGBE_DMATXCTL_TE;
2379 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, reg_val);
2380 	}
2381 
2382 	/*
2383 	 * Enabling tx queues ..
2384 	 * For 82599 must be done after DMATXCTL.TE is set
2385 	 */
2386 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2387 		tx_ring = &ixgbe->tx_rings[i];
2388 		reg_val = IXGBE_READ_REG(hw, IXGBE_TXDCTL(tx_ring->index));
2389 		reg_val |= IXGBE_TXDCTL_ENABLE;
2390 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(tx_ring->index), reg_val);
2391 	}
2392 }
2393 
2394 /*
2395  * ixgbe_setup_rss - Setup receive-side scaling feature.
2396  */
2397 static void
2398 ixgbe_setup_rss(ixgbe_t *ixgbe)
2399 {
2400 	struct ixgbe_hw *hw = &ixgbe->hw;
2401 	uint32_t i, mrqc, rxcsum;
2402 	uint32_t random;
2403 	uint32_t reta;
2404 	uint32_t ring_per_group;
2405 
2406 	/*
2407 	 * Fill out redirection table
2408 	 */
2409 	reta = 0;
2410 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
2411 
2412 	for (i = 0; i < 128; i++) {
2413 		reta = (reta << 8) | (i % ring_per_group) |
2414 		    ((i % ring_per_group) << 4);
2415 		if ((i & 3) == 3)
2416 			IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2417 	}
2418 
2419 	/*
2420 	 * Fill out hash function seeds with a random constant
2421 	 */
2422 	for (i = 0; i < 10; i++) {
2423 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2424 		    sizeof (uint32_t));
2425 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), random);
2426 	}
2427 
2428 	/*
2429 	 * Enable RSS & perform hash on these packet types
2430 	 */
2431 	mrqc = IXGBE_MRQC_RSSEN |
2432 	    IXGBE_MRQC_RSS_FIELD_IPV4 |
2433 	    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2434 	    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2435 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2436 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2437 	    IXGBE_MRQC_RSS_FIELD_IPV6 |
2438 	    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2439 	    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2440 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2441 	IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2442 
2443 	/*
2444 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2445 	 * It is an adapter hardware limitation that Packet Checksum is
2446 	 * mutually exclusive with RSS.
2447 	 */
2448 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2449 	rxcsum |= IXGBE_RXCSUM_PCSD;
2450 	rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
2451 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2452 }
2453 
2454 /*
2455  * ixgbe_setup_vmdq - Setup MAC classification feature
2456  */
2457 static void
2458 ixgbe_setup_vmdq(ixgbe_t *ixgbe)
2459 {
2460 	struct ixgbe_hw *hw = &ixgbe->hw;
2461 	uint32_t vmdctl, i, vtctl;
2462 
2463 	/*
2464 	 * Setup the VMDq Control register, enable VMDq based on
2465 	 * packet destination MAC address:
2466 	 */
2467 	switch (hw->mac.type) {
2468 	case ixgbe_mac_82598EB:
2469 		/*
2470 		 * VMDq Enable = 1;
2471 		 * VMDq Filter = 0; MAC filtering
2472 		 * Default VMDq output index = 0;
2473 		 */
2474 		vmdctl = IXGBE_VMD_CTL_VMDQ_EN;
2475 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
2476 		break;
2477 
2478 	case ixgbe_mac_82599EB:
2479 		/*
2480 		 * Enable VMDq-only.
2481 		 */
2482 		vmdctl = IXGBE_MRQC_VMDQEN;
2483 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, vmdctl);
2484 
2485 		for (i = 0; i < hw->mac.num_rar_entries; i++) {
2486 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(i), 0);
2487 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(i), 0);
2488 		}
2489 
2490 		/*
2491 		 * Enable Virtualization and Replication.
2492 		 */
2493 		vtctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2494 		IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vtctl);
2495 
2496 		/*
2497 		 * Enable receiving packets to all VFs
2498 		 */
2499 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), IXGBE_VFRE_ENABLE_ALL);
2500 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), IXGBE_VFRE_ENABLE_ALL);
2501 
2502 		break;
2503 
2504 	default:
2505 		break;
2506 	}
2507 }
2508 
2509 /*
2510  * ixgbe_setup_vmdq_rss - Setup both vmdq feature and rss feature.
2511  */
2512 static void
2513 ixgbe_setup_vmdq_rss(ixgbe_t *ixgbe)
2514 {
2515 	struct ixgbe_hw *hw = &ixgbe->hw;
2516 	uint32_t i, mrqc, rxcsum;
2517 	uint32_t random;
2518 	uint32_t reta;
2519 	uint32_t ring_per_group;
2520 	uint32_t vmdctl, vtctl;
2521 
2522 	/*
2523 	 * Fill out redirection table
2524 	 */
2525 	reta = 0;
2526 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
2527 	for (i = 0; i < 128; i++) {
2528 		reta = (reta << 8) | (i % ring_per_group) |
2529 		    ((i % ring_per_group) << 4);
2530 		if ((i & 3) == 3)
2531 			IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2532 	}
2533 
2534 	/*
2535 	 * Fill out hash function seeds with a random constant
2536 	 */
2537 	for (i = 0; i < 10; i++) {
2538 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2539 		    sizeof (uint32_t));
2540 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), random);
2541 	}
2542 
2543 	/*
2544 	 * Enable and setup RSS and VMDq
2545 	 */
2546 	switch (hw->mac.type) {
2547 	case ixgbe_mac_82598EB:
2548 		/*
2549 		 * Enable RSS & Setup RSS Hash functions
2550 		 */
2551 		mrqc = IXGBE_MRQC_RSSEN |
2552 		    IXGBE_MRQC_RSS_FIELD_IPV4 |
2553 		    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2554 		    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2555 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2556 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2557 		    IXGBE_MRQC_RSS_FIELD_IPV6 |
2558 		    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2559 		    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2560 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2561 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2562 
2563 		/*
2564 		 * Enable and Setup VMDq
2565 		 * VMDq Filter = 0; MAC filtering
2566 		 * Default VMDq output index = 0;
2567 		 */
2568 		vmdctl = IXGBE_VMD_CTL_VMDQ_EN;
2569 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
2570 		break;
2571 
2572 	case ixgbe_mac_82599EB:
2573 		/*
2574 		 * Enable RSS & Setup RSS Hash functions
2575 		 */
2576 		mrqc = IXGBE_MRQC_RSS_FIELD_IPV4 |
2577 		    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2578 		    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2579 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2580 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2581 		    IXGBE_MRQC_RSS_FIELD_IPV6 |
2582 		    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2583 		    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2584 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2585 
2586 		/*
2587 		 * Enable VMDq+RSS.
2588 		 */
2589 		if (ixgbe->num_rx_groups > 32)  {
2590 			mrqc = mrqc | IXGBE_MRQC_VMDQRSS64EN;
2591 		} else {
2592 			mrqc = mrqc | IXGBE_MRQC_VMDQRSS32EN;
2593 		}
2594 
2595 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2596 
2597 		for (i = 0; i < hw->mac.num_rar_entries; i++) {
2598 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(i), 0);
2599 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(i), 0);
2600 		}
2601 		break;
2602 
2603 	default:
2604 		break;
2605 
2606 	}
2607 
2608 	/*
2609 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2610 	 * It is an adapter hardware limitation that Packet Checksum is
2611 	 * mutually exclusive with RSS.
2612 	 */
2613 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2614 	rxcsum |= IXGBE_RXCSUM_PCSD;
2615 	rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
2616 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2617 
2618 	if (hw->mac.type == ixgbe_mac_82599EB) {
2619 		/*
2620 		 * Enable Virtualization and Replication.
2621 		 */
2622 		vtctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2623 		IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vtctl);
2624 
2625 		/*
2626 		 * Enable receiving packets to all VFs
2627 		 */
2628 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), IXGBE_VFRE_ENABLE_ALL);
2629 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), IXGBE_VFRE_ENABLE_ALL);
2630 	}
2631 }
2632 
2633 /*
2634  * ixgbe_init_unicst - Initialize the unicast addresses.
2635  */
2636 static void
2637 ixgbe_init_unicst(ixgbe_t *ixgbe)
2638 {
2639 	struct ixgbe_hw *hw = &ixgbe->hw;
2640 	uint8_t *mac_addr;
2641 	int slot;
2642 	/*
2643 	 * Here we should consider two situations:
2644 	 *
2645 	 * 1. Chipset is initialized at the first time,
2646 	 *    Clear all the multiple unicast addresses.
2647 	 *
2648 	 * 2. Chipset is reset
2649 	 *    Recover the multiple unicast addresses from the
2650 	 *    software data structure to the RAR registers.
2651 	 */
2652 	if (!ixgbe->unicst_init) {
2653 		/*
2654 		 * Initialize the multiple unicast addresses
2655 		 */
2656 		ixgbe->unicst_total = hw->mac.num_rar_entries;
2657 		ixgbe->unicst_avail = ixgbe->unicst_total;
2658 		for (slot = 0; slot < ixgbe->unicst_total; slot++) {
2659 			mac_addr = ixgbe->unicst_addr[slot].mac.addr;
2660 			bzero(mac_addr, ETHERADDRL);
2661 			(void) ixgbe_set_rar(hw, slot, mac_addr, NULL, NULL);
2662 			ixgbe->unicst_addr[slot].mac.set = 0;
2663 		}
2664 		ixgbe->unicst_init = B_TRUE;
2665 	} else {
2666 		/* Re-configure the RAR registers */
2667 		for (slot = 0; slot < ixgbe->unicst_total; slot++) {
2668 			mac_addr = ixgbe->unicst_addr[slot].mac.addr;
2669 			if (ixgbe->unicst_addr[slot].mac.set == 1) {
2670 				(void) ixgbe_set_rar(hw, slot, mac_addr,
2671 				    ixgbe->unicst_addr[slot].mac.group_index,
2672 				    IXGBE_RAH_AV);
2673 			} else {
2674 				bzero(mac_addr, ETHERADDRL);
2675 				(void) ixgbe_set_rar(hw, slot, mac_addr,
2676 				    NULL, NULL);
2677 			}
2678 		}
2679 	}
2680 }
2681 
2682 /*
2683  * ixgbe_unicst_find - Find the slot for the specified unicast address
2684  */
2685 int
2686 ixgbe_unicst_find(ixgbe_t *ixgbe, const uint8_t *mac_addr)
2687 {
2688 	int slot;
2689 
2690 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2691 
2692 	for (slot = 0; slot < ixgbe->unicst_total; slot++) {
2693 		if (bcmp(ixgbe->unicst_addr[slot].mac.addr,
2694 		    mac_addr, ETHERADDRL) == 0)
2695 			return (slot);
2696 	}
2697 
2698 	return (-1);
2699 }
2700 
2701 /*
2702  * ixgbe_multicst_add - Add a multicst address.
2703  */
2704 int
2705 ixgbe_multicst_add(ixgbe_t *ixgbe, const uint8_t *multiaddr)
2706 {
2707 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2708 
2709 	if ((multiaddr[0] & 01) == 0) {
2710 		return (EINVAL);
2711 	}
2712 
2713 	if (ixgbe->mcast_count >= MAX_NUM_MULTICAST_ADDRESSES) {
2714 		return (ENOENT);
2715 	}
2716 
2717 	bcopy(multiaddr,
2718 	    &ixgbe->mcast_table[ixgbe->mcast_count], ETHERADDRL);
2719 	ixgbe->mcast_count++;
2720 
2721 	/*
2722 	 * Update the multicast table in the hardware
2723 	 */
2724 	ixgbe_setup_multicst(ixgbe);
2725 
2726 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
2727 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
2728 		return (EIO);
2729 	}
2730 
2731 	return (0);
2732 }
2733 
2734 /*
2735  * ixgbe_multicst_remove - Remove a multicst address.
2736  */
2737 int
2738 ixgbe_multicst_remove(ixgbe_t *ixgbe, const uint8_t *multiaddr)
2739 {
2740 	int i;
2741 
2742 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2743 
2744 	for (i = 0; i < ixgbe->mcast_count; i++) {
2745 		if (bcmp(multiaddr, &ixgbe->mcast_table[i],
2746 		    ETHERADDRL) == 0) {
2747 			for (i++; i < ixgbe->mcast_count; i++) {
2748 				ixgbe->mcast_table[i - 1] =
2749 				    ixgbe->mcast_table[i];
2750 			}
2751 			ixgbe->mcast_count--;
2752 			break;
2753 		}
2754 	}
2755 
2756 	/*
2757 	 * Update the multicast table in the hardware
2758 	 */
2759 	ixgbe_setup_multicst(ixgbe);
2760 
2761 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
2762 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
2763 		return (EIO);
2764 	}
2765 
2766 	return (0);
2767 }
2768 
2769 /*
2770  * ixgbe_setup_multicast - Setup multicast data structures.
2771  *
2772  * This routine initializes all of the multicast related structures
2773  * and save them in the hardware registers.
2774  */
2775 static void
2776 ixgbe_setup_multicst(ixgbe_t *ixgbe)
2777 {
2778 	uint8_t *mc_addr_list;
2779 	uint32_t mc_addr_count;
2780 	struct ixgbe_hw *hw = &ixgbe->hw;
2781 
2782 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2783 
2784 	ASSERT(ixgbe->mcast_count <= MAX_NUM_MULTICAST_ADDRESSES);
2785 
2786 	mc_addr_list = (uint8_t *)ixgbe->mcast_table;
2787 	mc_addr_count = ixgbe->mcast_count;
2788 
2789 	/*
2790 	 * Update the multicast addresses to the MTA registers
2791 	 */
2792 	(void) ixgbe_update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2793 	    ixgbe_mc_table_itr);
2794 }
2795 
2796 /*
2797  * ixgbe_setup_vmdq_rss_conf - Configure vmdq and rss (number and mode).
2798  *
2799  * Configure the rx classification mode (vmdq & rss) and vmdq & rss numbers.
2800  * Different chipsets may have different allowed configuration of vmdq and rss.
2801  */
2802 static void
2803 ixgbe_setup_vmdq_rss_conf(ixgbe_t *ixgbe)
2804 {
2805 	struct ixgbe_hw *hw = &ixgbe->hw;
2806 	uint32_t ring_per_group;
2807 
2808 	switch (hw->mac.type) {
2809 	case ixgbe_mac_82598EB:
2810 		/*
2811 		 * 82598 supports the following combination:
2812 		 * vmdq no. x rss no.
2813 		 * [5..16]  x 1
2814 		 * [1..4]   x [1..16]
2815 		 * However 8 rss queue per pool (vmdq) is sufficient for
2816 		 * most cases.
2817 		 */
2818 		ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
2819 		if (ixgbe->num_rx_groups > 4) {
2820 			ixgbe->num_rx_rings = ixgbe->num_rx_groups;
2821 		} else {
2822 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
2823 			    min(8, ring_per_group);
2824 		}
2825 
2826 		break;
2827 
2828 	case ixgbe_mac_82599EB:
2829 		/*
2830 		 * 82599 supports the following combination:
2831 		 * vmdq no. x rss no.
2832 		 * [33..64] x [1..2]
2833 		 * [2..32]  x [1..4]
2834 		 * 1 x [1..16]
2835 		 * However 8 rss queue per pool (vmdq) is sufficient for
2836 		 * most cases.
2837 		 */
2838 		ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
2839 		if (ixgbe->num_rx_groups == 1) {
2840 			ixgbe->num_rx_rings = min(8, ring_per_group);
2841 		} else if (ixgbe->num_rx_groups <= 32) {
2842 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
2843 			    min(4, ring_per_group);
2844 		} else if (ixgbe->num_rx_groups <= 64) {
2845 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
2846 			    min(2, ring_per_group);
2847 		}
2848 
2849 		break;
2850 
2851 	default:
2852 		break;
2853 	}
2854 
2855 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
2856 
2857 	if (ixgbe->num_rx_groups == 1 && ring_per_group == 1) {
2858 		ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
2859 	} else if (ixgbe->num_rx_groups != 1 && ring_per_group == 1) {
2860 		ixgbe->classify_mode = IXGBE_CLASSIFY_VMDQ;
2861 	} else if (ixgbe->num_rx_groups != 1 && ring_per_group != 1) {
2862 		ixgbe->classify_mode = IXGBE_CLASSIFY_VMDQ_RSS;
2863 	} else {
2864 		ixgbe->classify_mode = IXGBE_CLASSIFY_RSS;
2865 	}
2866 
2867 	IXGBE_DEBUGLOG_2(ixgbe, "rx group number:%d, rx ring number:%d",
2868 	    ixgbe->num_rx_groups, ixgbe->num_rx_rings);
2869 }
2870 
2871 /*
2872  * ixgbe_get_conf - Get driver configurations set in driver.conf.
2873  *
2874  * This routine gets user-configured values out of the configuration
2875  * file ixgbe.conf.
2876  *
2877  * For each configurable value, there is a minimum, a maximum, and a
2878  * default.
2879  * If user does not configure a value, use the default.
2880  * If user configures below the minimum, use the minumum.
2881  * If user configures above the maximum, use the maxumum.
2882  */
2883 static void
2884 ixgbe_get_conf(ixgbe_t *ixgbe)
2885 {
2886 	struct ixgbe_hw *hw = &ixgbe->hw;
2887 	uint32_t flow_control;
2888 
2889 	/*
2890 	 * ixgbe driver supports the following user configurations:
2891 	 *
2892 	 * Jumbo frame configuration:
2893 	 *    default_mtu
2894 	 *
2895 	 * Ethernet flow control configuration:
2896 	 *    flow_control
2897 	 *
2898 	 * Multiple rings configurations:
2899 	 *    tx_queue_number
2900 	 *    tx_ring_size
2901 	 *    rx_queue_number
2902 	 *    rx_ring_size
2903 	 *
2904 	 * Call ixgbe_get_prop() to get the value for a specific
2905 	 * configuration parameter.
2906 	 */
2907 
2908 	/*
2909 	 * Jumbo frame configuration - max_frame_size controls host buffer
2910 	 * allocation, so includes MTU, ethernet header, vlan tag and
2911 	 * frame check sequence.
2912 	 */
2913 	ixgbe->default_mtu = ixgbe_get_prop(ixgbe, PROP_DEFAULT_MTU,
2914 	    MIN_MTU, ixgbe->capab->max_mtu, DEFAULT_MTU);
2915 
2916 	ixgbe->max_frame_size = ixgbe->default_mtu +
2917 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
2918 
2919 	/*
2920 	 * Ethernet flow control configuration
2921 	 */
2922 	flow_control = ixgbe_get_prop(ixgbe, PROP_FLOW_CONTROL,
2923 	    ixgbe_fc_none, 3, ixgbe_fc_none);
2924 	if (flow_control == 3)
2925 		flow_control = ixgbe_fc_default;
2926 
2927 	/*
2928 	 * fc.requested mode is what the user requests.  After autoneg,
2929 	 * fc.current_mode will be the flow_control mode that was negotiated.
2930 	 */
2931 	hw->fc.requested_mode = flow_control;
2932 
2933 	/*
2934 	 * Multiple rings configurations
2935 	 */
2936 	ixgbe->num_tx_rings = ixgbe_get_prop(ixgbe, PROP_TX_QUEUE_NUM,
2937 	    ixgbe->capab->min_tx_que_num,
2938 	    ixgbe->capab->max_tx_que_num,
2939 	    ixgbe->capab->def_tx_que_num);
2940 	ixgbe->tx_ring_size = ixgbe_get_prop(ixgbe, PROP_TX_RING_SIZE,
2941 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
2942 
2943 	ixgbe->num_rx_rings = ixgbe_get_prop(ixgbe, PROP_RX_QUEUE_NUM,
2944 	    ixgbe->capab->min_rx_que_num,
2945 	    ixgbe->capab->max_rx_que_num,
2946 	    ixgbe->capab->def_rx_que_num);
2947 	ixgbe->rx_ring_size = ixgbe_get_prop(ixgbe, PROP_RX_RING_SIZE,
2948 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
2949 
2950 	/*
2951 	 * Multiple groups configuration
2952 	 */
2953 	ixgbe->num_rx_groups = ixgbe_get_prop(ixgbe, PROP_RX_GROUP_NUM,
2954 	    ixgbe->capab->min_rx_grp_num, ixgbe->capab->max_rx_grp_num,
2955 	    ixgbe->capab->def_rx_grp_num);
2956 
2957 	ixgbe->mr_enable = ixgbe_get_prop(ixgbe, PROP_MR_ENABLE,
2958 	    0, 1, DEFAULT_MR_ENABLE);
2959 
2960 	if (ixgbe->mr_enable == B_FALSE) {
2961 		ixgbe->num_tx_rings = 1;
2962 		ixgbe->num_rx_rings = 1;
2963 		ixgbe->num_rx_groups = 1;
2964 		ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
2965 	} else {
2966 		ixgbe->num_rx_rings = ixgbe->num_rx_groups *
2967 		    max(ixgbe->num_rx_rings / ixgbe->num_rx_groups, 1);
2968 		/*
2969 		 * The combination of num_rx_rings and num_rx_groups
2970 		 * may be not supported by h/w. We need to adjust
2971 		 * them to appropriate values.
2972 		 */
2973 		ixgbe_setup_vmdq_rss_conf(ixgbe);
2974 	}
2975 
2976 	/*
2977 	 * Tunable used to force an interrupt type. The only use is
2978 	 * for testing of the lesser interrupt types.
2979 	 * 0 = don't force interrupt type
2980 	 * 1 = force interrupt type MSI-X
2981 	 * 2 = force interrupt type MSI
2982 	 * 3 = force interrupt type Legacy
2983 	 */
2984 	ixgbe->intr_force = ixgbe_get_prop(ixgbe, PROP_INTR_FORCE,
2985 	    IXGBE_INTR_NONE, IXGBE_INTR_LEGACY, IXGBE_INTR_NONE);
2986 
2987 	ixgbe->tx_hcksum_enable = ixgbe_get_prop(ixgbe, PROP_TX_HCKSUM_ENABLE,
2988 	    0, 1, DEFAULT_TX_HCKSUM_ENABLE);
2989 	ixgbe->rx_hcksum_enable = ixgbe_get_prop(ixgbe, PROP_RX_HCKSUM_ENABLE,
2990 	    0, 1, DEFAULT_RX_HCKSUM_ENABLE);
2991 	ixgbe->lso_enable = ixgbe_get_prop(ixgbe, PROP_LSO_ENABLE,
2992 	    0, 1, DEFAULT_LSO_ENABLE);
2993 	ixgbe->lro_enable = ixgbe_get_prop(ixgbe, PROP_LRO_ENABLE,
2994 	    0, 1, DEFAULT_LRO_ENABLE);
2995 	ixgbe->tx_head_wb_enable = ixgbe_get_prop(ixgbe, PROP_TX_HEAD_WB_ENABLE,
2996 	    0, 1, DEFAULT_TX_HEAD_WB_ENABLE);
2997 
2998 	/* Head Write Back not recommended for 82599 */
2999 	if (hw->mac.type >= ixgbe_mac_82599EB) {
3000 		ixgbe->tx_head_wb_enable = B_FALSE;
3001 	}
3002 
3003 	/*
3004 	 * ixgbe LSO needs the tx h/w checksum support.
3005 	 * LSO will be disabled if tx h/w checksum is not
3006 	 * enabled.
3007 	 */
3008 	if (ixgbe->tx_hcksum_enable == B_FALSE) {
3009 		ixgbe->lso_enable = B_FALSE;
3010 	}
3011 
3012 	/*
3013 	 * ixgbe LRO needs the rx h/w checksum support.
3014 	 * LRO will be disabled if rx h/w checksum is not
3015 	 * enabled.
3016 	 */
3017 	if (ixgbe->rx_hcksum_enable == B_FALSE) {
3018 		ixgbe->lro_enable = B_FALSE;
3019 	}
3020 
3021 	/*
3022 	 * ixgbe LRO only been supported by 82599 now
3023 	 */
3024 	if (hw->mac.type != ixgbe_mac_82599EB) {
3025 		ixgbe->lro_enable = B_FALSE;
3026 	}
3027 	ixgbe->tx_copy_thresh = ixgbe_get_prop(ixgbe, PROP_TX_COPY_THRESHOLD,
3028 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
3029 	    DEFAULT_TX_COPY_THRESHOLD);
3030 	ixgbe->tx_recycle_thresh = ixgbe_get_prop(ixgbe,
3031 	    PROP_TX_RECYCLE_THRESHOLD, MIN_TX_RECYCLE_THRESHOLD,
3032 	    MAX_TX_RECYCLE_THRESHOLD, DEFAULT_TX_RECYCLE_THRESHOLD);
3033 	ixgbe->tx_overload_thresh = ixgbe_get_prop(ixgbe,
3034 	    PROP_TX_OVERLOAD_THRESHOLD, MIN_TX_OVERLOAD_THRESHOLD,
3035 	    MAX_TX_OVERLOAD_THRESHOLD, DEFAULT_TX_OVERLOAD_THRESHOLD);
3036 	ixgbe->tx_resched_thresh = ixgbe_get_prop(ixgbe,
3037 	    PROP_TX_RESCHED_THRESHOLD, MIN_TX_RESCHED_THRESHOLD,
3038 	    MAX_TX_RESCHED_THRESHOLD, DEFAULT_TX_RESCHED_THRESHOLD);
3039 
3040 	ixgbe->rx_copy_thresh = ixgbe_get_prop(ixgbe, PROP_RX_COPY_THRESHOLD,
3041 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
3042 	    DEFAULT_RX_COPY_THRESHOLD);
3043 	ixgbe->rx_limit_per_intr = ixgbe_get_prop(ixgbe, PROP_RX_LIMIT_PER_INTR,
3044 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
3045 	    DEFAULT_RX_LIMIT_PER_INTR);
3046 
3047 	ixgbe->intr_throttling[0] = ixgbe_get_prop(ixgbe, PROP_INTR_THROTTLING,
3048 	    ixgbe->capab->min_intr_throttle,
3049 	    ixgbe->capab->max_intr_throttle,
3050 	    ixgbe->capab->def_intr_throttle);
3051 	/*
3052 	 * 82599 requires the interupt throttling rate is
3053 	 * a multiple of 8. This is enforced by the register
3054 	 * definiton.
3055 	 */
3056 	if (hw->mac.type == ixgbe_mac_82599EB)
3057 		ixgbe->intr_throttling[0] = ixgbe->intr_throttling[0] & 0xFF8;
3058 }
3059 
3060 static void
3061 ixgbe_init_params(ixgbe_t *ixgbe)
3062 {
3063 	ixgbe->param_en_10000fdx_cap = 1;
3064 	ixgbe->param_en_1000fdx_cap = 1;
3065 	ixgbe->param_en_100fdx_cap = 1;
3066 	ixgbe->param_adv_10000fdx_cap = 1;
3067 	ixgbe->param_adv_1000fdx_cap = 1;
3068 	ixgbe->param_adv_100fdx_cap = 1;
3069 
3070 	ixgbe->param_pause_cap = 1;
3071 	ixgbe->param_asym_pause_cap = 1;
3072 	ixgbe->param_rem_fault = 0;
3073 
3074 	ixgbe->param_adv_autoneg_cap = 1;
3075 	ixgbe->param_adv_pause_cap = 1;
3076 	ixgbe->param_adv_asym_pause_cap = 1;
3077 	ixgbe->param_adv_rem_fault = 0;
3078 
3079 	ixgbe->param_lp_10000fdx_cap = 0;
3080 	ixgbe->param_lp_1000fdx_cap = 0;
3081 	ixgbe->param_lp_100fdx_cap = 0;
3082 	ixgbe->param_lp_autoneg_cap = 0;
3083 	ixgbe->param_lp_pause_cap = 0;
3084 	ixgbe->param_lp_asym_pause_cap = 0;
3085 	ixgbe->param_lp_rem_fault = 0;
3086 }
3087 
3088 /*
3089  * ixgbe_get_prop - Get a property value out of the configuration file
3090  * ixgbe.conf.
3091  *
3092  * Caller provides the name of the property, a default value, a minimum
3093  * value, and a maximum value.
3094  *
3095  * Return configured value of the property, with default, minimum and
3096  * maximum properly applied.
3097  */
3098 static int
3099 ixgbe_get_prop(ixgbe_t *ixgbe,
3100     char *propname,	/* name of the property */
3101     int minval,		/* minimum acceptable value */
3102     int maxval,		/* maximim acceptable value */
3103     int defval)		/* default value */
3104 {
3105 	int value;
3106 
3107 	/*
3108 	 * Call ddi_prop_get_int() to read the conf settings
3109 	 */
3110 	value = ddi_prop_get_int(DDI_DEV_T_ANY, ixgbe->dip,
3111 	    DDI_PROP_DONTPASS, propname, defval);
3112 	if (value > maxval)
3113 		value = maxval;
3114 
3115 	if (value < minval)
3116 		value = minval;
3117 
3118 	return (value);
3119 }
3120 
3121 /*
3122  * ixgbe_driver_setup_link - Using the link properties to setup the link.
3123  */
3124 int
3125 ixgbe_driver_setup_link(ixgbe_t *ixgbe, boolean_t setup_hw)
3126 {
3127 	u32 autoneg_advertised = 0;
3128 
3129 	/*
3130 	 * No half duplex support with 10Gb parts
3131 	 */
3132 	if (ixgbe->param_adv_10000fdx_cap == 1)
3133 		autoneg_advertised |= IXGBE_LINK_SPEED_10GB_FULL;
3134 
3135 	if (ixgbe->param_adv_1000fdx_cap == 1)
3136 		autoneg_advertised |= IXGBE_LINK_SPEED_1GB_FULL;
3137 
3138 	if (ixgbe->param_adv_100fdx_cap == 1)
3139 		autoneg_advertised |= IXGBE_LINK_SPEED_100_FULL;
3140 
3141 	if (ixgbe->param_adv_autoneg_cap == 1 && autoneg_advertised == 0) {
3142 		ixgbe_notice(ixgbe, "Invalid link settings. Setup link "
3143 		    "to autonegotiation with full link capabilities.");
3144 
3145 		autoneg_advertised = IXGBE_LINK_SPEED_10GB_FULL |
3146 		    IXGBE_LINK_SPEED_1GB_FULL |
3147 		    IXGBE_LINK_SPEED_100_FULL;
3148 	}
3149 
3150 	if (setup_hw) {
3151 		if (ixgbe_setup_link(&ixgbe->hw, autoneg_advertised,
3152 		    ixgbe->param_adv_autoneg_cap, B_TRUE) != IXGBE_SUCCESS) {
3153 			ixgbe_notice(ixgbe, "Setup link failed on this "
3154 			    "device.");
3155 			return (IXGBE_FAILURE);
3156 		}
3157 	}
3158 
3159 	return (IXGBE_SUCCESS);
3160 }
3161 
3162 /*
3163  * ixgbe_driver_link_check - Link status processing.
3164  *
3165  * This function can be called in both kernel context and interrupt context
3166  */
3167 static void
3168 ixgbe_driver_link_check(ixgbe_t *ixgbe)
3169 {
3170 	struct ixgbe_hw *hw = &ixgbe->hw;
3171 	ixgbe_link_speed speed = IXGBE_LINK_SPEED_UNKNOWN;
3172 	boolean_t link_up = B_FALSE;
3173 	boolean_t link_changed = B_FALSE;
3174 
3175 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3176 
3177 	(void) ixgbe_check_link(hw, &speed, &link_up, false);
3178 	if (link_up) {
3179 		ixgbe->link_check_complete = B_TRUE;
3180 
3181 		/* Link is up, enable flow control settings */
3182 		(void) ixgbe_fc_enable(hw, 0);
3183 
3184 		/*
3185 		 * The Link is up, check whether it was marked as down earlier
3186 		 */
3187 		if (ixgbe->link_state != LINK_STATE_UP) {
3188 			switch (speed) {
3189 			case IXGBE_LINK_SPEED_10GB_FULL:
3190 				ixgbe->link_speed = SPEED_10GB;
3191 				break;
3192 			case IXGBE_LINK_SPEED_1GB_FULL:
3193 				ixgbe->link_speed = SPEED_1GB;
3194 				break;
3195 			case IXGBE_LINK_SPEED_100_FULL:
3196 				ixgbe->link_speed = SPEED_100;
3197 			}
3198 			ixgbe->link_duplex = LINK_DUPLEX_FULL;
3199 			ixgbe->link_state = LINK_STATE_UP;
3200 			link_changed = B_TRUE;
3201 		}
3202 	} else {
3203 		if (ixgbe->link_check_complete == B_TRUE ||
3204 		    (ixgbe->link_check_complete == B_FALSE &&
3205 		    gethrtime() >= ixgbe->link_check_hrtime)) {
3206 			/*
3207 			 * The link is really down
3208 			 */
3209 			ixgbe->link_check_complete = B_TRUE;
3210 
3211 			if (ixgbe->link_state != LINK_STATE_DOWN) {
3212 				ixgbe->link_speed = 0;
3213 				ixgbe->link_duplex = LINK_DUPLEX_UNKNOWN;
3214 				ixgbe->link_state = LINK_STATE_DOWN;
3215 				link_changed = B_TRUE;
3216 			}
3217 		}
3218 	}
3219 
3220 	/*
3221 	 * this is only reached after a link-status-change interrupt
3222 	 * so always get new phy state
3223 	 */
3224 	ixgbe_get_hw_state(ixgbe);
3225 
3226 	/*
3227 	 * If we are in an interrupt context, need to re-enable the
3228 	 * interrupt, which was automasked
3229 	 */
3230 	if (servicing_interrupt() != 0) {
3231 		ixgbe->eims |= IXGBE_EICR_LSC;
3232 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
3233 	}
3234 
3235 	if (link_changed) {
3236 		mac_link_update(ixgbe->mac_hdl, ixgbe->link_state);
3237 	}
3238 }
3239 
3240 /*
3241  * ixgbe_sfp_check - sfp module processing done in taskq only for 82599.
3242  */
3243 static void
3244 ixgbe_sfp_check(void *arg)
3245 {
3246 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3247 	uint32_t eicr = ixgbe->eicr;
3248 	struct ixgbe_hw *hw = &ixgbe->hw;
3249 
3250 	mutex_enter(&ixgbe->gen_lock);
3251 	if (eicr & IXGBE_EICR_GPI_SDP1) {
3252 		/* clear the interrupt */
3253 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
3254 
3255 		/* if link up, do multispeed fiber setup */
3256 		(void) ixgbe_setup_link(hw, IXGBE_LINK_SPEED_82599_AUTONEG,
3257 		    B_TRUE, B_TRUE);
3258 		ixgbe_driver_link_check(ixgbe);
3259 	} else if (eicr & IXGBE_EICR_GPI_SDP2) {
3260 		/* clear the interrupt */
3261 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
3262 
3263 		/* if link up, do sfp module setup */
3264 		(void) hw->mac.ops.setup_sfp(hw);
3265 
3266 		/* do multispeed fiber setup */
3267 		(void) ixgbe_setup_link(hw, IXGBE_LINK_SPEED_82599_AUTONEG,
3268 		    B_TRUE, B_TRUE);
3269 		ixgbe_driver_link_check(ixgbe);
3270 	}
3271 	mutex_exit(&ixgbe->gen_lock);
3272 }
3273 
3274 /*
3275  * ixgbe_link_timer - timer for link status detection
3276  */
3277 static void
3278 ixgbe_link_timer(void *arg)
3279 {
3280 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3281 
3282 	mutex_enter(&ixgbe->gen_lock);
3283 	ixgbe_driver_link_check(ixgbe);
3284 	mutex_exit(&ixgbe->gen_lock);
3285 }
3286 
3287 /*
3288  * ixgbe_local_timer - Driver watchdog function.
3289  *
3290  * This function will handle the transmit stall check and other routines.
3291  */
3292 static void
3293 ixgbe_local_timer(void *arg)
3294 {
3295 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3296 
3297 	if (ixgbe->ixgbe_state & IXGBE_ERROR) {
3298 		ixgbe->reset_count++;
3299 		if (ixgbe_reset(ixgbe) == IXGBE_SUCCESS)
3300 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_RESTORED);
3301 		ixgbe_restart_watchdog_timer(ixgbe);
3302 		return;
3303 	}
3304 
3305 	if (ixgbe_stall_check(ixgbe)) {
3306 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_STALL);
3307 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3308 
3309 		ixgbe->reset_count++;
3310 		if (ixgbe_reset(ixgbe) == IXGBE_SUCCESS)
3311 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_RESTORED);
3312 	}
3313 
3314 	ixgbe_restart_watchdog_timer(ixgbe);
3315 }
3316 
3317 /*
3318  * ixgbe_stall_check - Check for transmit stall.
3319  *
3320  * This function checks if the adapter is stalled (in transmit).
3321  *
3322  * It is called each time the watchdog timeout is invoked.
3323  * If the transmit descriptor reclaim continuously fails,
3324  * the watchdog value will increment by 1. If the watchdog
3325  * value exceeds the threshold, the ixgbe is assumed to
3326  * have stalled and need to be reset.
3327  */
3328 static boolean_t
3329 ixgbe_stall_check(ixgbe_t *ixgbe)
3330 {
3331 	ixgbe_tx_ring_t *tx_ring;
3332 	boolean_t result;
3333 	int i;
3334 
3335 	if (ixgbe->link_state != LINK_STATE_UP)
3336 		return (B_FALSE);
3337 
3338 	/*
3339 	 * If any tx ring is stalled, we'll reset the chipset
3340 	 */
3341 	result = B_FALSE;
3342 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
3343 		tx_ring = &ixgbe->tx_rings[i];
3344 		if (tx_ring->tbd_free <= ixgbe->tx_recycle_thresh) {
3345 			tx_ring->tx_recycle(tx_ring);
3346 		}
3347 
3348 		if (tx_ring->recycle_fail > 0)
3349 			tx_ring->stall_watchdog++;
3350 		else
3351 			tx_ring->stall_watchdog = 0;
3352 
3353 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
3354 			result = B_TRUE;
3355 			break;
3356 		}
3357 	}
3358 
3359 	if (result) {
3360 		tx_ring->stall_watchdog = 0;
3361 		tx_ring->recycle_fail = 0;
3362 	}
3363 
3364 	return (result);
3365 }
3366 
3367 
3368 /*
3369  * is_valid_mac_addr - Check if the mac address is valid.
3370  */
3371 static boolean_t
3372 is_valid_mac_addr(uint8_t *mac_addr)
3373 {
3374 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
3375 	const uint8_t addr_test2[6] =
3376 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
3377 
3378 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
3379 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
3380 		return (B_FALSE);
3381 
3382 	return (B_TRUE);
3383 }
3384 
3385 static boolean_t
3386 ixgbe_find_mac_address(ixgbe_t *ixgbe)
3387 {
3388 #ifdef __sparc
3389 	struct ixgbe_hw *hw = &ixgbe->hw;
3390 	uchar_t *bytes;
3391 	struct ether_addr sysaddr;
3392 	uint_t nelts;
3393 	int err;
3394 	boolean_t found = B_FALSE;
3395 
3396 	/*
3397 	 * The "vendor's factory-set address" may already have
3398 	 * been extracted from the chip, but if the property
3399 	 * "local-mac-address" is set we use that instead.
3400 	 *
3401 	 * We check whether it looks like an array of 6
3402 	 * bytes (which it should, if OBP set it).  If we can't
3403 	 * make sense of it this way, we'll ignore it.
3404 	 */
3405 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip,
3406 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
3407 	if (err == DDI_PROP_SUCCESS) {
3408 		if (nelts == ETHERADDRL) {
3409 			while (nelts--)
3410 				hw->mac.addr[nelts] = bytes[nelts];
3411 			found = B_TRUE;
3412 		}
3413 		ddi_prop_free(bytes);
3414 	}
3415 
3416 	/*
3417 	 * Look up the OBP property "local-mac-address?". If the user has set
3418 	 * 'local-mac-address? = false', use "the system address" instead.
3419 	 */
3420 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip, 0,
3421 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
3422 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
3423 			if (localetheraddr(NULL, &sysaddr) != 0) {
3424 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
3425 				found = B_TRUE;
3426 			}
3427 		}
3428 		ddi_prop_free(bytes);
3429 	}
3430 
3431 	/*
3432 	 * Finally(!), if there's a valid "mac-address" property (created
3433 	 * if we netbooted from this interface), we must use this instead
3434 	 * of any of the above to ensure that the NFS/install server doesn't
3435 	 * get confused by the address changing as Solaris takes over!
3436 	 */
3437 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip,
3438 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
3439 	if (err == DDI_PROP_SUCCESS) {
3440 		if (nelts == ETHERADDRL) {
3441 			while (nelts--)
3442 				hw->mac.addr[nelts] = bytes[nelts];
3443 			found = B_TRUE;
3444 		}
3445 		ddi_prop_free(bytes);
3446 	}
3447 
3448 	if (found) {
3449 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
3450 		return (B_TRUE);
3451 	}
3452 #else
3453 	_NOTE(ARGUNUSED(ixgbe));
3454 #endif
3455 
3456 	return (B_TRUE);
3457 }
3458 
3459 #pragma inline(ixgbe_arm_watchdog_timer)
3460 static void
3461 ixgbe_arm_watchdog_timer(ixgbe_t *ixgbe)
3462 {
3463 	/*
3464 	 * Fire a watchdog timer
3465 	 */
3466 	ixgbe->watchdog_tid =
3467 	    timeout(ixgbe_local_timer,
3468 	    (void *)ixgbe, 1 * drv_usectohz(1000000));
3469 
3470 }
3471 
3472 /*
3473  * ixgbe_enable_watchdog_timer - Enable and start the driver watchdog timer.
3474  */
3475 void
3476 ixgbe_enable_watchdog_timer(ixgbe_t *ixgbe)
3477 {
3478 	mutex_enter(&ixgbe->watchdog_lock);
3479 
3480 	if (!ixgbe->watchdog_enable) {
3481 		ixgbe->watchdog_enable = B_TRUE;
3482 		ixgbe->watchdog_start = B_TRUE;
3483 		ixgbe_arm_watchdog_timer(ixgbe);
3484 	}
3485 
3486 	mutex_exit(&ixgbe->watchdog_lock);
3487 }
3488 
3489 /*
3490  * ixgbe_disable_watchdog_timer - Disable and stop the driver watchdog timer.
3491  */
3492 void
3493 ixgbe_disable_watchdog_timer(ixgbe_t *ixgbe)
3494 {
3495 	timeout_id_t tid;
3496 
3497 	mutex_enter(&ixgbe->watchdog_lock);
3498 
3499 	ixgbe->watchdog_enable = B_FALSE;
3500 	ixgbe->watchdog_start = B_FALSE;
3501 	tid = ixgbe->watchdog_tid;
3502 	ixgbe->watchdog_tid = 0;
3503 
3504 	mutex_exit(&ixgbe->watchdog_lock);
3505 
3506 	if (tid != 0)
3507 		(void) untimeout(tid);
3508 }
3509 
3510 /*
3511  * ixgbe_start_watchdog_timer - Start the driver watchdog timer.
3512  */
3513 void
3514 ixgbe_start_watchdog_timer(ixgbe_t *ixgbe)
3515 {
3516 	mutex_enter(&ixgbe->watchdog_lock);
3517 
3518 	if (ixgbe->watchdog_enable) {
3519 		if (!ixgbe->watchdog_start) {
3520 			ixgbe->watchdog_start = B_TRUE;
3521 			ixgbe_arm_watchdog_timer(ixgbe);
3522 		}
3523 	}
3524 
3525 	mutex_exit(&ixgbe->watchdog_lock);
3526 }
3527 
3528 /*
3529  * ixgbe_restart_watchdog_timer - Restart the driver watchdog timer.
3530  */
3531 static void
3532 ixgbe_restart_watchdog_timer(ixgbe_t *ixgbe)
3533 {
3534 	mutex_enter(&ixgbe->watchdog_lock);
3535 
3536 	if (ixgbe->watchdog_start)
3537 		ixgbe_arm_watchdog_timer(ixgbe);
3538 
3539 	mutex_exit(&ixgbe->watchdog_lock);
3540 }
3541 
3542 /*
3543  * ixgbe_stop_watchdog_timer - Stop the driver watchdog timer.
3544  */
3545 void
3546 ixgbe_stop_watchdog_timer(ixgbe_t *ixgbe)
3547 {
3548 	timeout_id_t tid;
3549 
3550 	mutex_enter(&ixgbe->watchdog_lock);
3551 
3552 	ixgbe->watchdog_start = B_FALSE;
3553 	tid = ixgbe->watchdog_tid;
3554 	ixgbe->watchdog_tid = 0;
3555 
3556 	mutex_exit(&ixgbe->watchdog_lock);
3557 
3558 	if (tid != 0)
3559 		(void) untimeout(tid);
3560 }
3561 
3562 /*
3563  * ixgbe_disable_adapter_interrupts - Disable all adapter interrupts.
3564  */
3565 static void
3566 ixgbe_disable_adapter_interrupts(ixgbe_t *ixgbe)
3567 {
3568 	struct ixgbe_hw *hw = &ixgbe->hw;
3569 
3570 	/*
3571 	 * mask all interrupts off
3572 	 */
3573 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, 0xffffffff);
3574 
3575 	/*
3576 	 * for MSI-X, also disable autoclear
3577 	 */
3578 	if (ixgbe->intr_type == DDI_INTR_TYPE_MSIX) {
3579 		IXGBE_WRITE_REG(hw, IXGBE_EIAC, 0x0);
3580 	}
3581 
3582 	IXGBE_WRITE_FLUSH(hw);
3583 }
3584 
3585 /*
3586  * ixgbe_enable_adapter_interrupts - Enable all hardware interrupts.
3587  */
3588 static void
3589 ixgbe_enable_adapter_interrupts(ixgbe_t *ixgbe)
3590 {
3591 	struct ixgbe_hw *hw = &ixgbe->hw;
3592 	uint32_t eiac, eiam;
3593 	uint32_t gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
3594 
3595 	/* interrupt types to enable */
3596 	ixgbe->eims = IXGBE_EIMS_ENABLE_MASK;	/* shared code default */
3597 	ixgbe->eims &= ~IXGBE_EIMS_TCP_TIMER;	/* minus tcp timer */
3598 	ixgbe->eims |= ixgbe->capab->other_intr; /* "other" interrupt types */
3599 
3600 	/* enable automask on "other" causes that this adapter can generate */
3601 	eiam = ixgbe->capab->other_intr;
3602 
3603 	/*
3604 	 * msi-x mode
3605 	 */
3606 	if (ixgbe->intr_type == DDI_INTR_TYPE_MSIX) {
3607 		/* enable autoclear but not on bits 29:20 */
3608 		eiac = (ixgbe->eims & ~IXGBE_OTHER_INTR);
3609 
3610 		/* general purpose interrupt enable */
3611 		gpie |= (IXGBE_GPIE_MSIX_MODE
3612 		    | IXGBE_GPIE_PBA_SUPPORT
3613 		    | IXGBE_GPIE_OCD
3614 		    | IXGBE_GPIE_EIAME);
3615 	/*
3616 	 * non-msi-x mode
3617 	 */
3618 	} else {
3619 
3620 		/* disable autoclear, leave gpie at default */
3621 		eiac = 0;
3622 
3623 		/*
3624 		 * General purpose interrupt enable.
3625 		 * For 82599, extended interrupt automask enable
3626 		 * only in MSI or MSI-X mode
3627 		 */
3628 		if ((hw->mac.type < ixgbe_mac_82599EB) ||
3629 		    (ixgbe->intr_type == DDI_INTR_TYPE_MSI)) {
3630 			gpie |= IXGBE_GPIE_EIAME;
3631 		}
3632 	}
3633 	/* Enable specific interrupts for 82599  */
3634 	if (hw->mac.type == ixgbe_mac_82599EB) {
3635 		gpie |= IXGBE_SDP2_GPIEN; /* pluggable optics intr */
3636 		gpie |= IXGBE_SDP1_GPIEN; /* LSC interrupt */
3637 	}
3638 	/* Enable RSC Dealy 8us for 82599  */
3639 	if (ixgbe->lro_enable) {
3640 		gpie |= (1 << IXGBE_GPIE_RSC_DELAY_SHIFT);
3641 	}
3642 	/* write to interrupt control registers */
3643 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
3644 	IXGBE_WRITE_REG(hw, IXGBE_EIAC, eiac);
3645 	IXGBE_WRITE_REG(hw, IXGBE_EIAM, eiam);
3646 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3647 	IXGBE_WRITE_FLUSH(hw);
3648 }
3649 
3650 /*
3651  * ixgbe_loopback_ioctl - Loopback support.
3652  */
3653 enum ioc_reply
3654 ixgbe_loopback_ioctl(ixgbe_t *ixgbe, struct iocblk *iocp, mblk_t *mp)
3655 {
3656 	lb_info_sz_t *lbsp;
3657 	lb_property_t *lbpp;
3658 	uint32_t *lbmp;
3659 	uint32_t size;
3660 	uint32_t value;
3661 
3662 	if (mp->b_cont == NULL)
3663 		return (IOC_INVAL);
3664 
3665 	switch (iocp->ioc_cmd) {
3666 	default:
3667 		return (IOC_INVAL);
3668 
3669 	case LB_GET_INFO_SIZE:
3670 		size = sizeof (lb_info_sz_t);
3671 		if (iocp->ioc_count != size)
3672 			return (IOC_INVAL);
3673 
3674 		value = sizeof (lb_normal);
3675 		value += sizeof (lb_mac);
3676 		value += sizeof (lb_external);
3677 
3678 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
3679 		*lbsp = value;
3680 		break;
3681 
3682 	case LB_GET_INFO:
3683 		value = sizeof (lb_normal);
3684 		value += sizeof (lb_mac);
3685 		value += sizeof (lb_external);
3686 
3687 		size = value;
3688 		if (iocp->ioc_count != size)
3689 			return (IOC_INVAL);
3690 
3691 		value = 0;
3692 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
3693 
3694 		lbpp[value++] = lb_normal;
3695 		lbpp[value++] = lb_mac;
3696 		lbpp[value++] = lb_external;
3697 		break;
3698 
3699 	case LB_GET_MODE:
3700 		size = sizeof (uint32_t);
3701 		if (iocp->ioc_count != size)
3702 			return (IOC_INVAL);
3703 
3704 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3705 		*lbmp = ixgbe->loopback_mode;
3706 		break;
3707 
3708 	case LB_SET_MODE:
3709 		size = 0;
3710 		if (iocp->ioc_count != sizeof (uint32_t))
3711 			return (IOC_INVAL);
3712 
3713 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3714 		if (!ixgbe_set_loopback_mode(ixgbe, *lbmp))
3715 			return (IOC_INVAL);
3716 		break;
3717 	}
3718 
3719 	iocp->ioc_count = size;
3720 	iocp->ioc_error = 0;
3721 
3722 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
3723 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3724 		return (IOC_INVAL);
3725 	}
3726 
3727 	return (IOC_REPLY);
3728 }
3729 
3730 /*
3731  * ixgbe_set_loopback_mode - Setup loopback based on the loopback mode.
3732  */
3733 static boolean_t
3734 ixgbe_set_loopback_mode(ixgbe_t *ixgbe, uint32_t mode)
3735 {
3736 	if (mode == ixgbe->loopback_mode)
3737 		return (B_TRUE);
3738 
3739 	ixgbe->loopback_mode = mode;
3740 
3741 	if (mode == IXGBE_LB_NONE) {
3742 		/*
3743 		 * Reset the chip
3744 		 */
3745 		(void) ixgbe_reset(ixgbe);
3746 		return (B_TRUE);
3747 	}
3748 
3749 	mutex_enter(&ixgbe->gen_lock);
3750 
3751 	switch (mode) {
3752 	default:
3753 		mutex_exit(&ixgbe->gen_lock);
3754 		return (B_FALSE);
3755 
3756 	case IXGBE_LB_EXTERNAL:
3757 		break;
3758 
3759 	case IXGBE_LB_INTERNAL_MAC:
3760 		ixgbe_set_internal_mac_loopback(ixgbe);
3761 		break;
3762 	}
3763 
3764 	mutex_exit(&ixgbe->gen_lock);
3765 
3766 	return (B_TRUE);
3767 }
3768 
3769 /*
3770  * ixgbe_set_internal_mac_loopback - Set the internal MAC loopback mode.
3771  */
3772 static void
3773 ixgbe_set_internal_mac_loopback(ixgbe_t *ixgbe)
3774 {
3775 	struct ixgbe_hw *hw;
3776 	uint32_t reg;
3777 	uint8_t atlas;
3778 
3779 	hw = &ixgbe->hw;
3780 
3781 	/*
3782 	 * Setup MAC loopback
3783 	 */
3784 	reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_HLREG0);
3785 	reg |= IXGBE_HLREG0_LPBK;
3786 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_HLREG0, reg);
3787 
3788 	reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_AUTOC);
3789 	reg &= ~IXGBE_AUTOC_LMS_MASK;
3790 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_AUTOC, reg);
3791 
3792 	/*
3793 	 * Disable Atlas Tx lanes to keep packets in loopback and not on wire
3794 	 */
3795 	if (hw->mac.type == ixgbe_mac_82598EB) {
3796 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_LPBK,
3797 		    &atlas);
3798 		atlas |= IXGBE_ATLAS_PDN_TX_REG_EN;
3799 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_LPBK,
3800 		    atlas);
3801 
3802 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_10G,
3803 		    &atlas);
3804 		atlas |= IXGBE_ATLAS_PDN_TX_10G_QL_ALL;
3805 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_10G,
3806 		    atlas);
3807 
3808 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_1G,
3809 		    &atlas);
3810 		atlas |= IXGBE_ATLAS_PDN_TX_1G_QL_ALL;
3811 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_1G,
3812 		    atlas);
3813 
3814 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_AN,
3815 		    &atlas);
3816 		atlas |= IXGBE_ATLAS_PDN_TX_AN_QL_ALL;
3817 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_AN,
3818 		    atlas);
3819 	}
3820 }
3821 
3822 #pragma inline(ixgbe_intr_rx_work)
3823 /*
3824  * ixgbe_intr_rx_work - RX processing of ISR.
3825  */
3826 static void
3827 ixgbe_intr_rx_work(ixgbe_rx_ring_t *rx_ring)
3828 {
3829 	mblk_t *mp;
3830 
3831 	mutex_enter(&rx_ring->rx_lock);
3832 
3833 	mp = ixgbe_ring_rx(rx_ring, IXGBE_POLL_NULL);
3834 	mutex_exit(&rx_ring->rx_lock);
3835 
3836 	if (mp != NULL)
3837 		mac_rx_ring(rx_ring->ixgbe->mac_hdl, rx_ring->ring_handle, mp,
3838 		    rx_ring->ring_gen_num);
3839 }
3840 
3841 #pragma inline(ixgbe_intr_tx_work)
3842 /*
3843  * ixgbe_intr_tx_work - TX processing of ISR.
3844  */
3845 static void
3846 ixgbe_intr_tx_work(ixgbe_tx_ring_t *tx_ring)
3847 {
3848 	ixgbe_t *ixgbe = tx_ring->ixgbe;
3849 
3850 	/*
3851 	 * Recycle the tx descriptors
3852 	 */
3853 	tx_ring->tx_recycle(tx_ring);
3854 
3855 	/*
3856 	 * Schedule the re-transmit
3857 	 */
3858 	if (tx_ring->reschedule &&
3859 	    (tx_ring->tbd_free >= ixgbe->tx_resched_thresh)) {
3860 		tx_ring->reschedule = B_FALSE;
3861 		mac_tx_ring_update(tx_ring->ixgbe->mac_hdl,
3862 		    tx_ring->ring_handle);
3863 		IXGBE_DEBUG_STAT(tx_ring->stat_reschedule);
3864 	}
3865 }
3866 
3867 #pragma inline(ixgbe_intr_other_work)
3868 /*
3869  * ixgbe_intr_other_work - Process interrupt types other than tx/rx
3870  */
3871 static void
3872 ixgbe_intr_other_work(ixgbe_t *ixgbe, uint32_t eicr)
3873 {
3874 	struct ixgbe_hw *hw = &ixgbe->hw;
3875 
3876 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3877 
3878 	/*
3879 	 * handle link status change
3880 	 */
3881 	if (eicr & IXGBE_EICR_LSC) {
3882 		ixgbe_driver_link_check(ixgbe);
3883 	}
3884 
3885 	/*
3886 	 * check for fan failure on adapters with fans
3887 	 */
3888 	if ((ixgbe->capab->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
3889 	    (eicr & IXGBE_EICR_GPI_SDP1)) {
3890 		if (hw->mac.type < ixgbe_mac_82599EB) {
3891 			ixgbe_log(ixgbe,
3892 			    "Fan has stopped, replace the adapter\n");
3893 
3894 			/* re-enable the interrupt, which was automasked */
3895 			ixgbe->eims |= IXGBE_EICR_GPI_SDP1;
3896 		}
3897 	}
3898 
3899 	/*
3900 	 * Do SFP check for 82599
3901 	 */
3902 	if (hw->mac.type == ixgbe_mac_82599EB) {
3903 		if ((ddi_taskq_dispatch(ixgbe->sfp_taskq,
3904 		    ixgbe_sfp_check, (void *)ixgbe,
3905 		    DDI_NOSLEEP)) != DDI_SUCCESS) {
3906 			ixgbe_log(ixgbe, "No memory available to dispatch "
3907 			    "taskq for SFP check");
3908 		}
3909 
3910 		/*
3911 		 * We need to fully re-check the link later.
3912 		 */
3913 		ixgbe->link_check_complete = B_FALSE;
3914 		ixgbe->link_check_hrtime = gethrtime() +
3915 		    (IXGBE_LINK_UP_TIME * 100000000ULL);
3916 	}
3917 }
3918 
3919 /*
3920  * ixgbe_intr_legacy - Interrupt handler for legacy interrupts.
3921  */
3922 static uint_t
3923 ixgbe_intr_legacy(void *arg1, void *arg2)
3924 {
3925 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
3926 	struct ixgbe_hw *hw = &ixgbe->hw;
3927 	ixgbe_tx_ring_t *tx_ring;
3928 	ixgbe_rx_ring_t *rx_ring;
3929 	uint32_t eicr;
3930 	mblk_t *mp;
3931 	boolean_t tx_reschedule;
3932 	uint_t result;
3933 
3934 	_NOTE(ARGUNUSED(arg2));
3935 
3936 	mutex_enter(&ixgbe->gen_lock);
3937 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
3938 		mutex_exit(&ixgbe->gen_lock);
3939 		return (DDI_INTR_UNCLAIMED);
3940 	}
3941 
3942 	mp = NULL;
3943 	tx_reschedule = B_FALSE;
3944 
3945 	/*
3946 	 * Any bit set in eicr: claim this interrupt
3947 	 */
3948 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3949 
3950 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
3951 		mutex_exit(&ixgbe->gen_lock);
3952 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3953 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
3954 		return (DDI_INTR_CLAIMED);
3955 	}
3956 
3957 	if (eicr) {
3958 		/*
3959 		 * For legacy interrupt, we have only one interrupt,
3960 		 * so we have only one rx ring and one tx ring enabled.
3961 		 */
3962 		ASSERT(ixgbe->num_rx_rings == 1);
3963 		ASSERT(ixgbe->num_tx_rings == 1);
3964 
3965 		/*
3966 		 * For legacy interrupt, rx rings[0] will use RTxQ[0].
3967 		 */
3968 		if (eicr & 0x1) {
3969 			ixgbe->eimc |= IXGBE_EICR_RTX_QUEUE;
3970 			IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
3971 			ixgbe->eims |= IXGBE_EICR_RTX_QUEUE;
3972 			/*
3973 			 * Clean the rx descriptors
3974 			 */
3975 			rx_ring = &ixgbe->rx_rings[0];
3976 			mp = ixgbe_ring_rx(rx_ring, IXGBE_POLL_NULL);
3977 		}
3978 
3979 		/*
3980 		 * For legacy interrupt, tx rings[0] will use RTxQ[1].
3981 		 */
3982 		if (eicr & 0x2) {
3983 			/*
3984 			 * Recycle the tx descriptors
3985 			 */
3986 			tx_ring = &ixgbe->tx_rings[0];
3987 			tx_ring->tx_recycle(tx_ring);
3988 
3989 			/*
3990 			 * Schedule the re-transmit
3991 			 */
3992 			tx_reschedule = (tx_ring->reschedule &&
3993 			    (tx_ring->tbd_free >= ixgbe->tx_resched_thresh));
3994 		}
3995 
3996 		/* any interrupt type other than tx/rx */
3997 		if (eicr & ixgbe->capab->other_intr) {
3998 			if (hw->mac.type < ixgbe_mac_82599EB) {
3999 				ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4000 			}
4001 			if (hw->mac.type == ixgbe_mac_82599EB) {
4002 				ixgbe->eimc = IXGBE_82599_OTHER_INTR;
4003 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4004 			}
4005 			ixgbe_intr_other_work(ixgbe, eicr);
4006 			ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4007 		}
4008 
4009 		mutex_exit(&ixgbe->gen_lock);
4010 
4011 		result = DDI_INTR_CLAIMED;
4012 	} else {
4013 		mutex_exit(&ixgbe->gen_lock);
4014 
4015 		/*
4016 		 * No interrupt cause bits set: don't claim this interrupt.
4017 		 */
4018 		result = DDI_INTR_UNCLAIMED;
4019 	}
4020 
4021 	/* re-enable the interrupts which were automasked */
4022 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4023 
4024 	/*
4025 	 * Do the following work outside of the gen_lock
4026 	 */
4027 	if (mp != NULL) {
4028 		mac_rx_ring(rx_ring->ixgbe->mac_hdl, rx_ring->ring_handle, mp,
4029 		    rx_ring->ring_gen_num);
4030 	}
4031 
4032 	if (tx_reschedule)  {
4033 		tx_ring->reschedule = B_FALSE;
4034 		mac_tx_ring_update(ixgbe->mac_hdl, tx_ring->ring_handle);
4035 		IXGBE_DEBUG_STAT(tx_ring->stat_reschedule);
4036 	}
4037 
4038 	return (result);
4039 }
4040 
4041 /*
4042  * ixgbe_intr_msi - Interrupt handler for MSI.
4043  */
4044 static uint_t
4045 ixgbe_intr_msi(void *arg1, void *arg2)
4046 {
4047 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
4048 	struct ixgbe_hw *hw = &ixgbe->hw;
4049 	uint32_t eicr;
4050 
4051 	_NOTE(ARGUNUSED(arg2));
4052 
4053 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4054 
4055 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
4056 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4057 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4058 		return (DDI_INTR_CLAIMED);
4059 	}
4060 
4061 	/*
4062 	 * For MSI interrupt, we have only one vector,
4063 	 * so we have only one rx ring and one tx ring enabled.
4064 	 */
4065 	ASSERT(ixgbe->num_rx_rings == 1);
4066 	ASSERT(ixgbe->num_tx_rings == 1);
4067 
4068 	/*
4069 	 * For MSI interrupt, rx rings[0] will use RTxQ[0].
4070 	 */
4071 	if (eicr & 0x1) {
4072 		ixgbe_intr_rx_work(&ixgbe->rx_rings[0]);
4073 	}
4074 
4075 	/*
4076 	 * For MSI interrupt, tx rings[0] will use RTxQ[1].
4077 	 */
4078 	if (eicr & 0x2) {
4079 		ixgbe_intr_tx_work(&ixgbe->tx_rings[0]);
4080 	}
4081 
4082 	/* any interrupt type other than tx/rx */
4083 	if (eicr & ixgbe->capab->other_intr) {
4084 		mutex_enter(&ixgbe->gen_lock);
4085 		if (hw->mac.type < ixgbe_mac_82599EB) {
4086 			ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4087 		}
4088 		if (hw->mac.type == ixgbe_mac_82599EB) {
4089 			ixgbe->eimc = IXGBE_82599_OTHER_INTR;
4090 			IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4091 		}
4092 		ixgbe_intr_other_work(ixgbe, eicr);
4093 		ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4094 		mutex_exit(&ixgbe->gen_lock);
4095 	}
4096 
4097 	/* re-enable the interrupts which were automasked */
4098 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4099 
4100 	return (DDI_INTR_CLAIMED);
4101 }
4102 
4103 /*
4104  * ixgbe_intr_msix - Interrupt handler for MSI-X.
4105  */
4106 static uint_t
4107 ixgbe_intr_msix(void *arg1, void *arg2)
4108 {
4109 	ixgbe_intr_vector_t *vect = (ixgbe_intr_vector_t *)arg1;
4110 	ixgbe_t *ixgbe = vect->ixgbe;
4111 	struct ixgbe_hw *hw = &ixgbe->hw;
4112 	uint32_t eicr;
4113 	int r_idx = 0;
4114 
4115 	_NOTE(ARGUNUSED(arg2));
4116 
4117 	/*
4118 	 * Clean each rx ring that has its bit set in the map
4119 	 */
4120 	r_idx = bt_getlowbit(vect->rx_map, 0, (ixgbe->num_rx_rings - 1));
4121 	while (r_idx >= 0) {
4122 		ixgbe_intr_rx_work(&ixgbe->rx_rings[r_idx]);
4123 		r_idx = bt_getlowbit(vect->rx_map, (r_idx + 1),
4124 		    (ixgbe->num_rx_rings - 1));
4125 	}
4126 
4127 	/*
4128 	 * Clean each tx ring that has its bit set in the map
4129 	 */
4130 	r_idx = bt_getlowbit(vect->tx_map, 0, (ixgbe->num_tx_rings - 1));
4131 	while (r_idx >= 0) {
4132 		ixgbe_intr_tx_work(&ixgbe->tx_rings[r_idx]);
4133 		r_idx = bt_getlowbit(vect->tx_map, (r_idx + 1),
4134 		    (ixgbe->num_tx_rings - 1));
4135 	}
4136 
4137 
4138 	/*
4139 	 * Clean other interrupt (link change) that has its bit set in the map
4140 	 */
4141 	if (BT_TEST(vect->other_map, 0) == 1) {
4142 		eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4143 
4144 		if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) !=
4145 		    DDI_FM_OK) {
4146 			ddi_fm_service_impact(ixgbe->dip,
4147 			    DDI_SERVICE_DEGRADED);
4148 			atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4149 			return (DDI_INTR_CLAIMED);
4150 		}
4151 
4152 		/*
4153 		 * Need check cause bits and only other causes will
4154 		 * be processed
4155 		 */
4156 		/* any interrupt type other than tx/rx */
4157 		if (eicr & ixgbe->capab->other_intr) {
4158 			if (hw->mac.type < ixgbe_mac_82599EB) {
4159 				mutex_enter(&ixgbe->gen_lock);
4160 				ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4161 				ixgbe_intr_other_work(ixgbe, eicr);
4162 				mutex_exit(&ixgbe->gen_lock);
4163 			} else {
4164 				if (hw->mac.type == ixgbe_mac_82599EB) {
4165 					mutex_enter(&ixgbe->gen_lock);
4166 					ixgbe->eims |= IXGBE_EICR_RTX_QUEUE;
4167 					ixgbe_intr_other_work(ixgbe, eicr);
4168 					mutex_exit(&ixgbe->gen_lock);
4169 				}
4170 			}
4171 		}
4172 
4173 		/* re-enable the interrupts which were automasked */
4174 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4175 	}
4176 
4177 	return (DDI_INTR_CLAIMED);
4178 }
4179 
4180 /*
4181  * ixgbe_alloc_intrs - Allocate interrupts for the driver.
4182  *
4183  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
4184  * if not successful, try Legacy.
4185  * ixgbe->intr_force can be used to force sequence to start with
4186  * any of the 3 types.
4187  * If MSI-X is not used, number of tx/rx rings is forced to 1.
4188  */
4189 static int
4190 ixgbe_alloc_intrs(ixgbe_t *ixgbe)
4191 {
4192 	dev_info_t *devinfo;
4193 	int intr_types;
4194 	int rc;
4195 
4196 	devinfo = ixgbe->dip;
4197 
4198 	/*
4199 	 * Get supported interrupt types
4200 	 */
4201 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
4202 
4203 	if (rc != DDI_SUCCESS) {
4204 		ixgbe_log(ixgbe,
4205 		    "Get supported interrupt types failed: %d", rc);
4206 		return (IXGBE_FAILURE);
4207 	}
4208 	IXGBE_DEBUGLOG_1(ixgbe, "Supported interrupt types: %x", intr_types);
4209 
4210 	ixgbe->intr_type = 0;
4211 
4212 	/*
4213 	 * Install MSI-X interrupts
4214 	 */
4215 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
4216 	    (ixgbe->intr_force <= IXGBE_INTR_MSIX)) {
4217 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_MSIX);
4218 		if (rc == IXGBE_SUCCESS)
4219 			return (IXGBE_SUCCESS);
4220 
4221 		ixgbe_log(ixgbe,
4222 		    "Allocate MSI-X failed, trying MSI interrupts...");
4223 	}
4224 
4225 	/*
4226 	 * MSI-X not used, force rings and groups to 1
4227 	 */
4228 	ixgbe->num_rx_rings = 1;
4229 	ixgbe->num_rx_groups = 1;
4230 	ixgbe->num_tx_rings = 1;
4231 	ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
4232 	ixgbe_log(ixgbe,
4233 	    "MSI-X not used, force rings and groups number to 1");
4234 
4235 	/*
4236 	 * Install MSI interrupts
4237 	 */
4238 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
4239 	    (ixgbe->intr_force <= IXGBE_INTR_MSI)) {
4240 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_MSI);
4241 		if (rc == IXGBE_SUCCESS)
4242 			return (IXGBE_SUCCESS);
4243 
4244 		ixgbe_log(ixgbe,
4245 		    "Allocate MSI failed, trying Legacy interrupts...");
4246 	}
4247 
4248 	/*
4249 	 * Install legacy interrupts
4250 	 */
4251 	if (intr_types & DDI_INTR_TYPE_FIXED) {
4252 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_FIXED);
4253 		if (rc == IXGBE_SUCCESS)
4254 			return (IXGBE_SUCCESS);
4255 
4256 		ixgbe_log(ixgbe,
4257 		    "Allocate Legacy interrupts failed");
4258 	}
4259 
4260 	/*
4261 	 * If none of the 3 types succeeded, return failure
4262 	 */
4263 	return (IXGBE_FAILURE);
4264 }
4265 
4266 /*
4267  * ixgbe_alloc_intr_handles - Allocate interrupt handles.
4268  *
4269  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
4270  * if fewer than 2 handles are available, return failure.
4271  * Upon success, this maps the vectors to rx and tx rings for
4272  * interrupts.
4273  */
4274 static int
4275 ixgbe_alloc_intr_handles(ixgbe_t *ixgbe, int intr_type)
4276 {
4277 	dev_info_t *devinfo;
4278 	int request, count, actual;
4279 	int minimum;
4280 	int rc;
4281 	uint32_t ring_per_group;
4282 
4283 	devinfo = ixgbe->dip;
4284 
4285 	switch (intr_type) {
4286 	case DDI_INTR_TYPE_FIXED:
4287 		request = 1;	/* Request 1 legacy interrupt handle */
4288 		minimum = 1;
4289 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: legacy");
4290 		break;
4291 
4292 	case DDI_INTR_TYPE_MSI:
4293 		request = 1;	/* Request 1 MSI interrupt handle */
4294 		minimum = 1;
4295 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: MSI");
4296 		break;
4297 
4298 	case DDI_INTR_TYPE_MSIX:
4299 		/*
4300 		 * Best number of vectors for the adapter is
4301 		 * (# rx rings + # tx rings), however we will
4302 		 * limit the request number.
4303 		 */
4304 		request = min(16, ixgbe->num_rx_rings + ixgbe->num_tx_rings);
4305 		if (request > ixgbe->capab->max_ring_vect)
4306 			request = ixgbe->capab->max_ring_vect;
4307 		minimum = 1;
4308 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: MSI-X");
4309 		break;
4310 
4311 	default:
4312 		ixgbe_log(ixgbe,
4313 		    "invalid call to ixgbe_alloc_intr_handles(): %d\n",
4314 		    intr_type);
4315 		return (IXGBE_FAILURE);
4316 	}
4317 	IXGBE_DEBUGLOG_2(ixgbe, "interrupt handles requested: %d  minimum: %d",
4318 	    request, minimum);
4319 
4320 	/*
4321 	 * Get number of supported interrupts
4322 	 */
4323 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
4324 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
4325 		ixgbe_log(ixgbe,
4326 		    "Get interrupt number failed. Return: %d, count: %d",
4327 		    rc, count);
4328 		return (IXGBE_FAILURE);
4329 	}
4330 	IXGBE_DEBUGLOG_1(ixgbe, "interrupts supported: %d", count);
4331 
4332 	actual = 0;
4333 	ixgbe->intr_cnt = 0;
4334 	ixgbe->intr_cnt_max = 0;
4335 	ixgbe->intr_cnt_min = 0;
4336 
4337 	/*
4338 	 * Allocate an array of interrupt handles
4339 	 */
4340 	ixgbe->intr_size = request * sizeof (ddi_intr_handle_t);
4341 	ixgbe->htable = kmem_alloc(ixgbe->intr_size, KM_SLEEP);
4342 
4343 	rc = ddi_intr_alloc(devinfo, ixgbe->htable, intr_type, 0,
4344 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
4345 	if (rc != DDI_SUCCESS) {
4346 		ixgbe_log(ixgbe, "Allocate interrupts failed. "
4347 		    "return: %d, request: %d, actual: %d",
4348 		    rc, request, actual);
4349 		goto alloc_handle_fail;
4350 	}
4351 	IXGBE_DEBUGLOG_1(ixgbe, "interrupts actually allocated: %d", actual);
4352 
4353 	/*
4354 	 * upper/lower limit of interrupts
4355 	 */
4356 	ixgbe->intr_cnt = actual;
4357 	ixgbe->intr_cnt_max = request;
4358 	ixgbe->intr_cnt_min = minimum;
4359 
4360 	/*
4361 	 * rss number per group should not exceed the rx interrupt number,
4362 	 * else need to adjust rx ring number.
4363 	 */
4364 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
4365 	ASSERT((ixgbe->num_rx_rings % ixgbe->num_rx_groups) == 0);
4366 	if (min(actual, ixgbe->num_rx_rings) < ring_per_group) {
4367 		ixgbe->num_rx_rings = ixgbe->num_rx_groups *
4368 		    min(actual, ixgbe->num_rx_rings);
4369 		ixgbe_setup_vmdq_rss_conf(ixgbe);
4370 	}
4371 
4372 	/*
4373 	 * Now we know the actual number of vectors.  Here we map the vector
4374 	 * to other, rx rings and tx ring.
4375 	 */
4376 	if (actual < minimum) {
4377 		ixgbe_log(ixgbe, "Insufficient interrupt handles available: %d",
4378 		    actual);
4379 		goto alloc_handle_fail;
4380 	}
4381 
4382 	/*
4383 	 * Get priority for first vector, assume remaining are all the same
4384 	 */
4385 	rc = ddi_intr_get_pri(ixgbe->htable[0], &ixgbe->intr_pri);
4386 	if (rc != DDI_SUCCESS) {
4387 		ixgbe_log(ixgbe,
4388 		    "Get interrupt priority failed: %d", rc);
4389 		goto alloc_handle_fail;
4390 	}
4391 
4392 	rc = ddi_intr_get_cap(ixgbe->htable[0], &ixgbe->intr_cap);
4393 	if (rc != DDI_SUCCESS) {
4394 		ixgbe_log(ixgbe,
4395 		    "Get interrupt cap failed: %d", rc);
4396 		goto alloc_handle_fail;
4397 	}
4398 
4399 	ixgbe->intr_type = intr_type;
4400 
4401 	return (IXGBE_SUCCESS);
4402 
4403 alloc_handle_fail:
4404 	ixgbe_rem_intrs(ixgbe);
4405 
4406 	return (IXGBE_FAILURE);
4407 }
4408 
4409 /*
4410  * ixgbe_add_intr_handlers - Add interrupt handlers based on the interrupt type.
4411  *
4412  * Before adding the interrupt handlers, the interrupt vectors have
4413  * been allocated, and the rx/tx rings have also been allocated.
4414  */
4415 static int
4416 ixgbe_add_intr_handlers(ixgbe_t *ixgbe)
4417 {
4418 	int vector = 0;
4419 	int rc;
4420 
4421 	switch (ixgbe->intr_type) {
4422 	case DDI_INTR_TYPE_MSIX:
4423 		/*
4424 		 * Add interrupt handler for all vectors
4425 		 */
4426 		for (vector = 0; vector < ixgbe->intr_cnt; vector++) {
4427 			/*
4428 			 * install pointer to vect_map[vector]
4429 			 */
4430 			rc = ddi_intr_add_handler(ixgbe->htable[vector],
4431 			    (ddi_intr_handler_t *)ixgbe_intr_msix,
4432 			    (void *)&ixgbe->vect_map[vector], NULL);
4433 
4434 			if (rc != DDI_SUCCESS) {
4435 				ixgbe_log(ixgbe,
4436 				    "Add interrupt handler failed. "
4437 				    "return: %d, vector: %d", rc, vector);
4438 				for (vector--; vector >= 0; vector--) {
4439 					(void) ddi_intr_remove_handler(
4440 					    ixgbe->htable[vector]);
4441 				}
4442 				return (IXGBE_FAILURE);
4443 			}
4444 		}
4445 
4446 		break;
4447 
4448 	case DDI_INTR_TYPE_MSI:
4449 		/*
4450 		 * Add interrupt handlers for the only vector
4451 		 */
4452 		rc = ddi_intr_add_handler(ixgbe->htable[vector],
4453 		    (ddi_intr_handler_t *)ixgbe_intr_msi,
4454 		    (void *)ixgbe, NULL);
4455 
4456 		if (rc != DDI_SUCCESS) {
4457 			ixgbe_log(ixgbe,
4458 			    "Add MSI interrupt handler failed: %d", rc);
4459 			return (IXGBE_FAILURE);
4460 		}
4461 
4462 		break;
4463 
4464 	case DDI_INTR_TYPE_FIXED:
4465 		/*
4466 		 * Add interrupt handlers for the only vector
4467 		 */
4468 		rc = ddi_intr_add_handler(ixgbe->htable[vector],
4469 		    (ddi_intr_handler_t *)ixgbe_intr_legacy,
4470 		    (void *)ixgbe, NULL);
4471 
4472 		if (rc != DDI_SUCCESS) {
4473 			ixgbe_log(ixgbe,
4474 			    "Add legacy interrupt handler failed: %d", rc);
4475 			return (IXGBE_FAILURE);
4476 		}
4477 
4478 		break;
4479 
4480 	default:
4481 		return (IXGBE_FAILURE);
4482 	}
4483 
4484 	return (IXGBE_SUCCESS);
4485 }
4486 
4487 #pragma inline(ixgbe_map_rxring_to_vector)
4488 /*
4489  * ixgbe_map_rxring_to_vector - Map given rx ring to given interrupt vector.
4490  */
4491 static void
4492 ixgbe_map_rxring_to_vector(ixgbe_t *ixgbe, int r_idx, int v_idx)
4493 {
4494 	/*
4495 	 * Set bit in map
4496 	 */
4497 	BT_SET(ixgbe->vect_map[v_idx].rx_map, r_idx);
4498 
4499 	/*
4500 	 * Count bits set
4501 	 */
4502 	ixgbe->vect_map[v_idx].rxr_cnt++;
4503 
4504 	/*
4505 	 * Remember bit position
4506 	 */
4507 	ixgbe->rx_rings[r_idx].intr_vector = v_idx;
4508 	ixgbe->rx_rings[r_idx].vect_bit = 1 << v_idx;
4509 }
4510 
4511 #pragma inline(ixgbe_map_txring_to_vector)
4512 /*
4513  * ixgbe_map_txring_to_vector - Map given tx ring to given interrupt vector.
4514  */
4515 static void
4516 ixgbe_map_txring_to_vector(ixgbe_t *ixgbe, int t_idx, int v_idx)
4517 {
4518 	/*
4519 	 * Set bit in map
4520 	 */
4521 	BT_SET(ixgbe->vect_map[v_idx].tx_map, t_idx);
4522 
4523 	/*
4524 	 * Count bits set
4525 	 */
4526 	ixgbe->vect_map[v_idx].txr_cnt++;
4527 
4528 	/*
4529 	 * Remember bit position
4530 	 */
4531 	ixgbe->tx_rings[t_idx].intr_vector = v_idx;
4532 	ixgbe->tx_rings[t_idx].vect_bit = 1 << v_idx;
4533 }
4534 
4535 /*
4536  * ixgbe_setup_ivar - Set the given entry in the given interrupt vector
4537  * allocation register (IVAR).
4538  * cause:
4539  *   -1 : other cause
4540  *    0 : rx
4541  *    1 : tx
4542  */
4543 static void
4544 ixgbe_setup_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, uint8_t msix_vector,
4545     int8_t cause)
4546 {
4547 	struct ixgbe_hw *hw = &ixgbe->hw;
4548 	u32 ivar, index;
4549 
4550 	switch (hw->mac.type) {
4551 	case ixgbe_mac_82598EB:
4552 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
4553 		if (cause == -1) {
4554 			cause = 0;
4555 		}
4556 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
4557 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
4558 		ivar &= ~(0xFF << (8 * (intr_alloc_entry & 0x3)));
4559 		ivar |= (msix_vector << (8 * (intr_alloc_entry & 0x3)));
4560 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
4561 		break;
4562 	case ixgbe_mac_82599EB:
4563 		if (cause == -1) {
4564 			/* other causes */
4565 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
4566 			index = (intr_alloc_entry & 1) * 8;
4567 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
4568 			ivar &= ~(0xFF << index);
4569 			ivar |= (msix_vector << index);
4570 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
4571 		} else {
4572 			/* tx or rx causes */
4573 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
4574 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
4575 			ivar = IXGBE_READ_REG(hw,
4576 			    IXGBE_IVAR(intr_alloc_entry >> 1));
4577 			ivar &= ~(0xFF << index);
4578 			ivar |= (msix_vector << index);
4579 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
4580 			    ivar);
4581 		}
4582 		break;
4583 	default:
4584 		break;
4585 	}
4586 }
4587 
4588 /*
4589  * ixgbe_enable_ivar - Enable the given entry by setting the VAL bit of
4590  * given interrupt vector allocation register (IVAR).
4591  * cause:
4592  *   -1 : other cause
4593  *    0 : rx
4594  *    1 : tx
4595  */
4596 static void
4597 ixgbe_enable_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, int8_t cause)
4598 {
4599 	struct ixgbe_hw *hw = &ixgbe->hw;
4600 	u32 ivar, index;
4601 
4602 	switch (hw->mac.type) {
4603 	case ixgbe_mac_82598EB:
4604 		if (cause == -1) {
4605 			cause = 0;
4606 		}
4607 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
4608 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
4609 		ivar |= (IXGBE_IVAR_ALLOC_VAL << (8 *
4610 		    (intr_alloc_entry & 0x3)));
4611 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
4612 		break;
4613 	case ixgbe_mac_82599EB:
4614 		if (cause == -1) {
4615 			/* other causes */
4616 			index = (intr_alloc_entry & 1) * 8;
4617 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
4618 			ivar |= (IXGBE_IVAR_ALLOC_VAL << index);
4619 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
4620 		} else {
4621 			/* tx or rx causes */
4622 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
4623 			ivar = IXGBE_READ_REG(hw,
4624 			    IXGBE_IVAR(intr_alloc_entry >> 1));
4625 			ivar |= (IXGBE_IVAR_ALLOC_VAL << index);
4626 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
4627 			    ivar);
4628 		}
4629 		break;
4630 	default:
4631 		break;
4632 	}
4633 }
4634 
4635 /*
4636  * ixgbe_disable_ivar - Disble the given entry by clearing the VAL bit of
4637  * given interrupt vector allocation register (IVAR).
4638  * cause:
4639  *   -1 : other cause
4640  *    0 : rx
4641  *    1 : tx
4642  */
4643 static void
4644 ixgbe_disable_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, int8_t cause)
4645 {
4646 	struct ixgbe_hw *hw = &ixgbe->hw;
4647 	u32 ivar, index;
4648 
4649 	switch (hw->mac.type) {
4650 	case ixgbe_mac_82598EB:
4651 		if (cause == -1) {
4652 			cause = 0;
4653 		}
4654 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
4655 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
4656 		ivar &= ~(IXGBE_IVAR_ALLOC_VAL<< (8 *
4657 		    (intr_alloc_entry & 0x3)));
4658 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
4659 		break;
4660 	case ixgbe_mac_82599EB:
4661 		if (cause == -1) {
4662 			/* other causes */
4663 			index = (intr_alloc_entry & 1) * 8;
4664 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
4665 			ivar &= ~(IXGBE_IVAR_ALLOC_VAL << index);
4666 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
4667 		} else {
4668 			/* tx or rx causes */
4669 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
4670 			ivar = IXGBE_READ_REG(hw,
4671 			    IXGBE_IVAR(intr_alloc_entry >> 1));
4672 			ivar &= ~(IXGBE_IVAR_ALLOC_VAL << index);
4673 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
4674 			    ivar);
4675 		}
4676 		break;
4677 	default:
4678 		break;
4679 	}
4680 }
4681 
4682 /*
4683  * Convert the rx ring index driver maintained to the rx ring index
4684  * in h/w.
4685  */
4686 static uint32_t
4687 ixgbe_get_hw_rx_index(ixgbe_t *ixgbe, uint32_t sw_rx_index)
4688 {
4689 
4690 	struct ixgbe_hw *hw = &ixgbe->hw;
4691 	uint32_t rx_ring_per_group, hw_rx_index;
4692 
4693 	if (ixgbe->classify_mode == IXGBE_CLASSIFY_RSS ||
4694 	    ixgbe->classify_mode == IXGBE_CLASSIFY_NONE) {
4695 		return (sw_rx_index);
4696 	} else if (ixgbe->classify_mode == IXGBE_CLASSIFY_VMDQ) {
4697 		if (hw->mac.type == ixgbe_mac_82598EB) {
4698 			return (sw_rx_index);
4699 		} else if (hw->mac.type == ixgbe_mac_82599EB) {
4700 			return (sw_rx_index * 2);
4701 		}
4702 	} else if (ixgbe->classify_mode == IXGBE_CLASSIFY_VMDQ_RSS) {
4703 		rx_ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
4704 
4705 		if (hw->mac.type == ixgbe_mac_82598EB) {
4706 			hw_rx_index = (sw_rx_index / rx_ring_per_group) *
4707 			    16 + (sw_rx_index % rx_ring_per_group);
4708 			return (hw_rx_index);
4709 		} else if (hw->mac.type == ixgbe_mac_82599EB) {
4710 			if (ixgbe->num_rx_groups > 32) {
4711 				hw_rx_index = (sw_rx_index /
4712 				    rx_ring_per_group) * 2 +
4713 				    (sw_rx_index % rx_ring_per_group);
4714 			} else {
4715 				hw_rx_index = (sw_rx_index /
4716 				    rx_ring_per_group) * 4 +
4717 				    (sw_rx_index % rx_ring_per_group);
4718 			}
4719 			return (hw_rx_index);
4720 		}
4721 	}
4722 
4723 	/*
4724 	 * Should never reach. Just to make compiler happy.
4725 	 */
4726 	return (sw_rx_index);
4727 }
4728 
4729 /*
4730  * ixgbe_map_intrs_to_vectors - Map different interrupts to MSI-X vectors.
4731  *
4732  * For MSI-X, here will map rx interrupt, tx interrupt and other interrupt
4733  * to vector[0 - (intr_cnt -1)].
4734  */
4735 static int
4736 ixgbe_map_intrs_to_vectors(ixgbe_t *ixgbe)
4737 {
4738 	int i, vector = 0;
4739 
4740 	/* initialize vector map */
4741 	bzero(&ixgbe->vect_map, sizeof (ixgbe->vect_map));
4742 	for (i = 0; i < ixgbe->intr_cnt; i++) {
4743 		ixgbe->vect_map[i].ixgbe = ixgbe;
4744 	}
4745 
4746 	/*
4747 	 * non-MSI-X case is very simple: rx rings[0] on RTxQ[0],
4748 	 * tx rings[0] on RTxQ[1].
4749 	 */
4750 	if (ixgbe->intr_type != DDI_INTR_TYPE_MSIX) {
4751 		ixgbe_map_rxring_to_vector(ixgbe, 0, 0);
4752 		ixgbe_map_txring_to_vector(ixgbe, 0, 1);
4753 		return (IXGBE_SUCCESS);
4754 	}
4755 
4756 	/*
4757 	 * Interrupts/vectors mapping for MSI-X
4758 	 */
4759 
4760 	/*
4761 	 * Map other interrupt to vector 0,
4762 	 * Set bit in map and count the bits set.
4763 	 */
4764 	BT_SET(ixgbe->vect_map[vector].other_map, 0);
4765 	ixgbe->vect_map[vector].other_cnt++;
4766 
4767 	/*
4768 	 * Map rx ring interrupts to vectors
4769 	 */
4770 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
4771 		ixgbe_map_rxring_to_vector(ixgbe, i, vector);
4772 		vector = (vector +1) % ixgbe->intr_cnt;
4773 	}
4774 
4775 	/*
4776 	 * Map tx ring interrupts to vectors
4777 	 */
4778 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
4779 		ixgbe_map_txring_to_vector(ixgbe, i, vector);
4780 		vector = (vector +1) % ixgbe->intr_cnt;
4781 	}
4782 
4783 	return (IXGBE_SUCCESS);
4784 }
4785 
4786 /*
4787  * ixgbe_setup_adapter_vector - Setup the adapter interrupt vector(s).
4788  *
4789  * This relies on ring/vector mapping already set up in the
4790  * vect_map[] structures
4791  */
4792 static void
4793 ixgbe_setup_adapter_vector(ixgbe_t *ixgbe)
4794 {
4795 	struct ixgbe_hw *hw = &ixgbe->hw;
4796 	ixgbe_intr_vector_t *vect;	/* vector bitmap */
4797 	int r_idx;	/* ring index */
4798 	int v_idx;	/* vector index */
4799 	uint32_t hw_index;
4800 
4801 	/*
4802 	 * Clear any previous entries
4803 	 */
4804 	switch (hw->mac.type) {
4805 	case ixgbe_mac_82598EB:
4806 		for (v_idx = 0; v_idx < 25; v_idx++)
4807 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(v_idx), 0);
4808 
4809 		break;
4810 	case ixgbe_mac_82599EB:
4811 		for (v_idx = 0; v_idx < 64; v_idx++)
4812 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(v_idx), 0);
4813 		IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, 0);
4814 
4815 		break;
4816 	default:
4817 		break;
4818 	}
4819 
4820 	/*
4821 	 * For non MSI-X interrupt, rx rings[0] will use RTxQ[0], and
4822 	 * tx rings[0] will use RTxQ[1].
4823 	 */
4824 	if (ixgbe->intr_type != DDI_INTR_TYPE_MSIX) {
4825 		ixgbe_setup_ivar(ixgbe, 0, 0, 0);
4826 		ixgbe_setup_ivar(ixgbe, 0, 1, 1);
4827 		return;
4828 	}
4829 
4830 	/*
4831 	 * For MSI-X interrupt, "Other" is always on vector[0].
4832 	 */
4833 	ixgbe_setup_ivar(ixgbe, IXGBE_IVAR_OTHER_CAUSES_INDEX, 0, -1);
4834 
4835 	/*
4836 	 * For each interrupt vector, populate the IVAR table
4837 	 */
4838 	for (v_idx = 0; v_idx < ixgbe->intr_cnt; v_idx++) {
4839 		vect = &ixgbe->vect_map[v_idx];
4840 
4841 		/*
4842 		 * For each rx ring bit set
4843 		 */
4844 		r_idx = bt_getlowbit(vect->rx_map, 0,
4845 		    (ixgbe->num_rx_rings - 1));
4846 
4847 		while (r_idx >= 0) {
4848 			hw_index = ixgbe->rx_rings[r_idx].hw_index;
4849 			ixgbe_setup_ivar(ixgbe, hw_index, v_idx, 0);
4850 			r_idx = bt_getlowbit(vect->rx_map, (r_idx + 1),
4851 			    (ixgbe->num_rx_rings - 1));
4852 		}
4853 
4854 		/*
4855 		 * For each tx ring bit set
4856 		 */
4857 		r_idx = bt_getlowbit(vect->tx_map, 0,
4858 		    (ixgbe->num_tx_rings - 1));
4859 
4860 		while (r_idx >= 0) {
4861 			ixgbe_setup_ivar(ixgbe, r_idx, v_idx, 1);
4862 			r_idx = bt_getlowbit(vect->tx_map, (r_idx + 1),
4863 			    (ixgbe->num_tx_rings - 1));
4864 		}
4865 	}
4866 }
4867 
4868 /*
4869  * ixgbe_rem_intr_handlers - Remove the interrupt handlers.
4870  */
4871 static void
4872 ixgbe_rem_intr_handlers(ixgbe_t *ixgbe)
4873 {
4874 	int i;
4875 	int rc;
4876 
4877 	for (i = 0; i < ixgbe->intr_cnt; i++) {
4878 		rc = ddi_intr_remove_handler(ixgbe->htable[i]);
4879 		if (rc != DDI_SUCCESS) {
4880 			IXGBE_DEBUGLOG_1(ixgbe,
4881 			    "Remove intr handler failed: %d", rc);
4882 		}
4883 	}
4884 }
4885 
4886 /*
4887  * ixgbe_rem_intrs - Remove the allocated interrupts.
4888  */
4889 static void
4890 ixgbe_rem_intrs(ixgbe_t *ixgbe)
4891 {
4892 	int i;
4893 	int rc;
4894 
4895 	for (i = 0; i < ixgbe->intr_cnt; i++) {
4896 		rc = ddi_intr_free(ixgbe->htable[i]);
4897 		if (rc != DDI_SUCCESS) {
4898 			IXGBE_DEBUGLOG_1(ixgbe,
4899 			    "Free intr failed: %d", rc);
4900 		}
4901 	}
4902 
4903 	kmem_free(ixgbe->htable, ixgbe->intr_size);
4904 	ixgbe->htable = NULL;
4905 }
4906 
4907 /*
4908  * ixgbe_enable_intrs - Enable all the ddi interrupts.
4909  */
4910 static int
4911 ixgbe_enable_intrs(ixgbe_t *ixgbe)
4912 {
4913 	int i;
4914 	int rc;
4915 
4916 	/*
4917 	 * Enable interrupts
4918 	 */
4919 	if (ixgbe->intr_cap & DDI_INTR_FLAG_BLOCK) {
4920 		/*
4921 		 * Call ddi_intr_block_enable() for MSI
4922 		 */
4923 		rc = ddi_intr_block_enable(ixgbe->htable, ixgbe->intr_cnt);
4924 		if (rc != DDI_SUCCESS) {
4925 			ixgbe_log(ixgbe,
4926 			    "Enable block intr failed: %d", rc);
4927 			return (IXGBE_FAILURE);
4928 		}
4929 	} else {
4930 		/*
4931 		 * Call ddi_intr_enable() for Legacy/MSI non block enable
4932 		 */
4933 		for (i = 0; i < ixgbe->intr_cnt; i++) {
4934 			rc = ddi_intr_enable(ixgbe->htable[i]);
4935 			if (rc != DDI_SUCCESS) {
4936 				ixgbe_log(ixgbe,
4937 				    "Enable intr failed: %d", rc);
4938 				return (IXGBE_FAILURE);
4939 			}
4940 		}
4941 	}
4942 
4943 	return (IXGBE_SUCCESS);
4944 }
4945 
4946 /*
4947  * ixgbe_disable_intrs - Disable all the interrupts.
4948  */
4949 static int
4950 ixgbe_disable_intrs(ixgbe_t *ixgbe)
4951 {
4952 	int i;
4953 	int rc;
4954 
4955 	/*
4956 	 * Disable all interrupts
4957 	 */
4958 	if (ixgbe->intr_cap & DDI_INTR_FLAG_BLOCK) {
4959 		rc = ddi_intr_block_disable(ixgbe->htable, ixgbe->intr_cnt);
4960 		if (rc != DDI_SUCCESS) {
4961 			ixgbe_log(ixgbe,
4962 			    "Disable block intr failed: %d", rc);
4963 			return (IXGBE_FAILURE);
4964 		}
4965 	} else {
4966 		for (i = 0; i < ixgbe->intr_cnt; i++) {
4967 			rc = ddi_intr_disable(ixgbe->htable[i]);
4968 			if (rc != DDI_SUCCESS) {
4969 				ixgbe_log(ixgbe,
4970 				    "Disable intr failed: %d", rc);
4971 				return (IXGBE_FAILURE);
4972 			}
4973 		}
4974 	}
4975 
4976 	return (IXGBE_SUCCESS);
4977 }
4978 
4979 /*
4980  * ixgbe_get_hw_state - Get and save parameters related to adapter hardware.
4981  */
4982 static void
4983 ixgbe_get_hw_state(ixgbe_t *ixgbe)
4984 {
4985 	struct ixgbe_hw *hw = &ixgbe->hw;
4986 	ixgbe_link_speed speed = IXGBE_LINK_SPEED_UNKNOWN;
4987 	boolean_t link_up = B_FALSE;
4988 	uint32_t pcs1g_anlp = 0;
4989 	uint32_t pcs1g_ana = 0;
4990 
4991 	ASSERT(mutex_owned(&ixgbe->gen_lock));
4992 	ixgbe->param_lp_1000fdx_cap = 0;
4993 	ixgbe->param_lp_100fdx_cap  = 0;
4994 
4995 	/* check for link, don't wait */
4996 	(void) ixgbe_check_link(hw, &speed, &link_up, false);
4997 	if (link_up) {
4998 		pcs1g_anlp = IXGBE_READ_REG(hw, IXGBE_PCS1GANLP);
4999 		pcs1g_ana = IXGBE_READ_REG(hw, IXGBE_PCS1GANA);
5000 
5001 		ixgbe->param_lp_1000fdx_cap =
5002 		    (pcs1g_anlp & IXGBE_PCS1GANLP_LPFD) ? 1 : 0;
5003 		ixgbe->param_lp_100fdx_cap =
5004 		    (pcs1g_anlp & IXGBE_PCS1GANLP_LPFD) ? 1 : 0;
5005 	}
5006 
5007 	ixgbe->param_adv_1000fdx_cap =
5008 	    (pcs1g_ana & IXGBE_PCS1GANA_FDC)  ? 1 : 0;
5009 	ixgbe->param_adv_100fdx_cap = (pcs1g_ana & IXGBE_PCS1GANA_FDC)  ? 1 : 0;
5010 }
5011 
5012 /*
5013  * ixgbe_get_driver_control - Notify that driver is in control of device.
5014  */
5015 static void
5016 ixgbe_get_driver_control(struct ixgbe_hw *hw)
5017 {
5018 	uint32_t ctrl_ext;
5019 
5020 	/*
5021 	 * Notify firmware that driver is in control of device
5022 	 */
5023 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5024 	ctrl_ext |= IXGBE_CTRL_EXT_DRV_LOAD;
5025 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5026 }
5027 
5028 /*
5029  * ixgbe_release_driver_control - Notify that driver is no longer in control
5030  * of device.
5031  */
5032 static void
5033 ixgbe_release_driver_control(struct ixgbe_hw *hw)
5034 {
5035 	uint32_t ctrl_ext;
5036 
5037 	/*
5038 	 * Notify firmware that driver is no longer in control of device
5039 	 */
5040 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5041 	ctrl_ext &= ~IXGBE_CTRL_EXT_DRV_LOAD;
5042 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5043 }
5044 
5045 /*
5046  * ixgbe_atomic_reserve - Atomic decrease operation.
5047  */
5048 int
5049 ixgbe_atomic_reserve(uint32_t *count_p, uint32_t n)
5050 {
5051 	uint32_t oldval;
5052 	uint32_t newval;
5053 
5054 	/*
5055 	 * ATOMICALLY
5056 	 */
5057 	do {
5058 		oldval = *count_p;
5059 		if (oldval < n)
5060 			return (-1);
5061 		newval = oldval - n;
5062 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
5063 
5064 	return (newval);
5065 }
5066 
5067 /*
5068  * ixgbe_mc_table_itr - Traverse the entries in the multicast table.
5069  */
5070 static uint8_t *
5071 ixgbe_mc_table_itr(struct ixgbe_hw *hw, uint8_t **upd_ptr, uint32_t *vmdq)
5072 {
5073 	uint8_t *addr = *upd_ptr;
5074 	uint8_t *new_ptr;
5075 
5076 	_NOTE(ARGUNUSED(hw));
5077 	_NOTE(ARGUNUSED(vmdq));
5078 
5079 	new_ptr = addr + IXGBE_ETH_LENGTH_OF_ADDRESS;
5080 	*upd_ptr = new_ptr;
5081 	return (addr);
5082 }
5083 
5084 /*
5085  * FMA support
5086  */
5087 int
5088 ixgbe_check_acc_handle(ddi_acc_handle_t handle)
5089 {
5090 	ddi_fm_error_t de;
5091 
5092 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
5093 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
5094 	return (de.fme_status);
5095 }
5096 
5097 int
5098 ixgbe_check_dma_handle(ddi_dma_handle_t handle)
5099 {
5100 	ddi_fm_error_t de;
5101 
5102 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
5103 	return (de.fme_status);
5104 }
5105 
5106 /*
5107  * ixgbe_fm_error_cb - The IO fault service error handling callback function.
5108  */
5109 static int
5110 ixgbe_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
5111 {
5112 	_NOTE(ARGUNUSED(impl_data));
5113 	/*
5114 	 * as the driver can always deal with an error in any dma or
5115 	 * access handle, we can just return the fme_status value.
5116 	 */
5117 	pci_ereport_post(dip, err, NULL);
5118 	return (err->fme_status);
5119 }
5120 
5121 static void
5122 ixgbe_fm_init(ixgbe_t *ixgbe)
5123 {
5124 	ddi_iblock_cookie_t iblk;
5125 	int fma_dma_flag;
5126 
5127 	/*
5128 	 * Only register with IO Fault Services if we have some capability
5129 	 */
5130 	if (ixgbe->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
5131 		ixgbe_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
5132 	} else {
5133 		ixgbe_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
5134 	}
5135 
5136 	if (ixgbe->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
5137 		fma_dma_flag = 1;
5138 	} else {
5139 		fma_dma_flag = 0;
5140 	}
5141 
5142 	ixgbe_set_fma_flags(fma_dma_flag);
5143 
5144 	if (ixgbe->fm_capabilities) {
5145 
5146 		/*
5147 		 * Register capabilities with IO Fault Services
5148 		 */
5149 		ddi_fm_init(ixgbe->dip, &ixgbe->fm_capabilities, &iblk);
5150 
5151 		/*
5152 		 * Initialize pci ereport capabilities if ereport capable
5153 		 */
5154 		if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities) ||
5155 		    DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5156 			pci_ereport_setup(ixgbe->dip);
5157 
5158 		/*
5159 		 * Register error callback if error callback capable
5160 		 */
5161 		if (DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5162 			ddi_fm_handler_register(ixgbe->dip,
5163 			    ixgbe_fm_error_cb, (void*) ixgbe);
5164 	}
5165 }
5166 
5167 static void
5168 ixgbe_fm_fini(ixgbe_t *ixgbe)
5169 {
5170 	/*
5171 	 * Only unregister FMA capabilities if they are registered
5172 	 */
5173 	if (ixgbe->fm_capabilities) {
5174 
5175 		/*
5176 		 * Release any resources allocated by pci_ereport_setup()
5177 		 */
5178 		if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities) ||
5179 		    DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5180 			pci_ereport_teardown(ixgbe->dip);
5181 
5182 		/*
5183 		 * Un-register error callback if error callback capable
5184 		 */
5185 		if (DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5186 			ddi_fm_handler_unregister(ixgbe->dip);
5187 
5188 		/*
5189 		 * Unregister from IO Fault Service
5190 		 */
5191 		ddi_fm_fini(ixgbe->dip);
5192 	}
5193 }
5194 
5195 void
5196 ixgbe_fm_ereport(ixgbe_t *ixgbe, char *detail)
5197 {
5198 	uint64_t ena;
5199 	char buf[FM_MAX_CLASS];
5200 
5201 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
5202 	ena = fm_ena_generate(0, FM_ENA_FMT1);
5203 	if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities)) {
5204 		ddi_fm_ereport_post(ixgbe->dip, buf, ena, DDI_NOSLEEP,
5205 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
5206 	}
5207 }
5208 
5209 static int
5210 ixgbe_ring_start(mac_ring_driver_t rh, uint64_t mr_gen_num)
5211 {
5212 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)rh;
5213 
5214 	mutex_enter(&rx_ring->rx_lock);
5215 	rx_ring->ring_gen_num = mr_gen_num;
5216 	mutex_exit(&rx_ring->rx_lock);
5217 	return (0);
5218 }
5219 
5220 /*
5221  * Get the global ring index by a ring index within a group.
5222  */
5223 static int
5224 ixgbe_get_rx_ring_index(ixgbe_t *ixgbe, int gindex, int rindex)
5225 {
5226 	ixgbe_rx_ring_t *rx_ring;
5227 	int i;
5228 
5229 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
5230 		rx_ring = &ixgbe->rx_rings[i];
5231 		if (rx_ring->group_index == gindex)
5232 			rindex--;
5233 		if (rindex < 0)
5234 			return (i);
5235 	}
5236 
5237 	return (-1);
5238 }
5239 
5240 /*
5241  * Callback funtion for MAC layer to register all rings.
5242  */
5243 /* ARGSUSED */
5244 void
5245 ixgbe_fill_ring(void *arg, mac_ring_type_t rtype, const int group_index,
5246     const int ring_index, mac_ring_info_t *infop, mac_ring_handle_t rh)
5247 {
5248 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
5249 	mac_intr_t *mintr = &infop->mri_intr;
5250 
5251 	switch (rtype) {
5252 	case MAC_RING_TYPE_RX: {
5253 		/*
5254 		 * 'index' is the ring index within the group.
5255 		 * Need to get the global ring index by searching in groups.
5256 		 */
5257 		int global_ring_index = ixgbe_get_rx_ring_index(
5258 		    ixgbe, group_index, ring_index);
5259 
5260 		ASSERT(global_ring_index >= 0);
5261 
5262 		ixgbe_rx_ring_t *rx_ring = &ixgbe->rx_rings[global_ring_index];
5263 		rx_ring->ring_handle = rh;
5264 
5265 		infop->mri_driver = (mac_ring_driver_t)rx_ring;
5266 		infop->mri_start = ixgbe_ring_start;
5267 		infop->mri_stop = NULL;
5268 		infop->mri_poll = ixgbe_ring_rx_poll;
5269 		infop->mri_stat = ixgbe_rx_ring_stat;
5270 
5271 		mintr->mi_handle = (mac_intr_handle_t)rx_ring;
5272 		mintr->mi_enable = ixgbe_rx_ring_intr_enable;
5273 		mintr->mi_disable = ixgbe_rx_ring_intr_disable;
5274 		if (ixgbe->intr_type &
5275 		    (DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI)) {
5276 			mintr->mi_ddi_handle =
5277 			    ixgbe->htable[rx_ring->intr_vector];
5278 		}
5279 
5280 		break;
5281 	}
5282 	case MAC_RING_TYPE_TX: {
5283 		ASSERT(group_index == -1);
5284 		ASSERT(ring_index < ixgbe->num_tx_rings);
5285 
5286 		ixgbe_tx_ring_t *tx_ring = &ixgbe->tx_rings[ring_index];
5287 		tx_ring->ring_handle = rh;
5288 
5289 		infop->mri_driver = (mac_ring_driver_t)tx_ring;
5290 		infop->mri_start = NULL;
5291 		infop->mri_stop = NULL;
5292 		infop->mri_tx = ixgbe_ring_tx;
5293 		infop->mri_stat = ixgbe_tx_ring_stat;
5294 		if (ixgbe->intr_type &
5295 		    (DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI)) {
5296 			mintr->mi_ddi_handle =
5297 			    ixgbe->htable[tx_ring->intr_vector];
5298 		}
5299 		break;
5300 	}
5301 	default:
5302 		break;
5303 	}
5304 }
5305 
5306 /*
5307  * Callback funtion for MAC layer to register all groups.
5308  */
5309 void
5310 ixgbe_fill_group(void *arg, mac_ring_type_t rtype, const int index,
5311     mac_group_info_t *infop, mac_group_handle_t gh)
5312 {
5313 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
5314 
5315 	switch (rtype) {
5316 	case MAC_RING_TYPE_RX: {
5317 		ixgbe_rx_group_t *rx_group;
5318 
5319 		rx_group = &ixgbe->rx_groups[index];
5320 		rx_group->group_handle = gh;
5321 
5322 		infop->mgi_driver = (mac_group_driver_t)rx_group;
5323 		infop->mgi_start = NULL;
5324 		infop->mgi_stop = NULL;
5325 		infop->mgi_addmac = ixgbe_addmac;
5326 		infop->mgi_remmac = ixgbe_remmac;
5327 		infop->mgi_count = (ixgbe->num_rx_rings / ixgbe->num_rx_groups);
5328 
5329 		break;
5330 	}
5331 	case MAC_RING_TYPE_TX:
5332 		break;
5333 	default:
5334 		break;
5335 	}
5336 }
5337 
5338 /*
5339  * Enable interrupt on the specificed rx ring.
5340  */
5341 int
5342 ixgbe_rx_ring_intr_enable(mac_intr_handle_t intrh)
5343 {
5344 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)intrh;
5345 	ixgbe_t *ixgbe = rx_ring->ixgbe;
5346 	int r_idx = rx_ring->index;
5347 	int hw_r_idx = rx_ring->hw_index;
5348 	int v_idx = rx_ring->intr_vector;
5349 
5350 	mutex_enter(&ixgbe->gen_lock);
5351 	if (ixgbe->ixgbe_state & IXGBE_INTR_ADJUST) {
5352 		mutex_exit(&ixgbe->gen_lock);
5353 		/*
5354 		 * Simply return 0.
5355 		 * Interrupts are being adjusted. ixgbe_intr_adjust()
5356 		 * will eventually re-enable the interrupt when it's
5357 		 * done with the adjustment.
5358 		 */
5359 		return (0);
5360 	}
5361 
5362 	/*
5363 	 * To enable interrupt by setting the VAL bit of given interrupt
5364 	 * vector allocation register (IVAR).
5365 	 */
5366 	ixgbe_enable_ivar(ixgbe, hw_r_idx, 0);
5367 
5368 	BT_SET(ixgbe->vect_map[v_idx].rx_map, r_idx);
5369 
5370 	/*
5371 	 * Trigger a Rx interrupt on this ring
5372 	 */
5373 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_EICS, (1 << v_idx));
5374 	IXGBE_WRITE_FLUSH(&ixgbe->hw);
5375 
5376 	mutex_exit(&ixgbe->gen_lock);
5377 
5378 	return (0);
5379 }
5380 
5381 /*
5382  * Disable interrupt on the specificed rx ring.
5383  */
5384 int
5385 ixgbe_rx_ring_intr_disable(mac_intr_handle_t intrh)
5386 {
5387 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)intrh;
5388 	ixgbe_t *ixgbe = rx_ring->ixgbe;
5389 	int r_idx = rx_ring->index;
5390 	int hw_r_idx = rx_ring->hw_index;
5391 	int v_idx = rx_ring->intr_vector;
5392 
5393 	mutex_enter(&ixgbe->gen_lock);
5394 	if (ixgbe->ixgbe_state & IXGBE_INTR_ADJUST) {
5395 		mutex_exit(&ixgbe->gen_lock);
5396 		/*
5397 		 * Simply return 0.
5398 		 * In the rare case where an interrupt is being
5399 		 * disabled while interrupts are being adjusted,
5400 		 * we don't fail the operation. No interrupts will
5401 		 * be generated while they are adjusted, and
5402 		 * ixgbe_intr_adjust() will cause the interrupts
5403 		 * to be re-enabled once it completes. Note that
5404 		 * in this case, packets may be delivered to the
5405 		 * stack via interrupts before xgbe_rx_ring_intr_enable()
5406 		 * is called again. This is acceptable since interrupt
5407 		 * adjustment is infrequent, and the stack will be
5408 		 * able to handle these packets.
5409 		 */
5410 		return (0);
5411 	}
5412 
5413 	/*
5414 	 * To disable interrupt by clearing the VAL bit of given interrupt
5415 	 * vector allocation register (IVAR).
5416 	 */
5417 	ixgbe_disable_ivar(ixgbe, hw_r_idx, 0);
5418 
5419 	BT_CLEAR(ixgbe->vect_map[v_idx].rx_map, r_idx);
5420 
5421 	mutex_exit(&ixgbe->gen_lock);
5422 
5423 	return (0);
5424 }
5425 
5426 /*
5427  * Add a mac address.
5428  */
5429 static int
5430 ixgbe_addmac(void *arg, const uint8_t *mac_addr)
5431 {
5432 	ixgbe_rx_group_t *rx_group = (ixgbe_rx_group_t *)arg;
5433 	ixgbe_t *ixgbe = rx_group->ixgbe;
5434 	struct ixgbe_hw *hw = &ixgbe->hw;
5435 	int slot, i;
5436 
5437 	mutex_enter(&ixgbe->gen_lock);
5438 
5439 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
5440 		mutex_exit(&ixgbe->gen_lock);
5441 		return (ECANCELED);
5442 	}
5443 
5444 	if (ixgbe->unicst_avail == 0) {
5445 		/* no slots available */
5446 		mutex_exit(&ixgbe->gen_lock);
5447 		return (ENOSPC);
5448 	}
5449 
5450 	/*
5451 	 * The first ixgbe->num_rx_groups slots are reserved for each respective
5452 	 * group. The rest slots are shared by all groups. While adding a
5453 	 * MAC address, reserved slots are firstly checked then the shared
5454 	 * slots are searched.
5455 	 */
5456 	slot = -1;
5457 	if (ixgbe->unicst_addr[rx_group->index].mac.set == 1) {
5458 		for (i = ixgbe->num_rx_groups; i < ixgbe->unicst_total; i++) {
5459 			if (ixgbe->unicst_addr[i].mac.set == 0) {
5460 				slot = i;
5461 				break;
5462 			}
5463 		}
5464 	} else {
5465 		slot = rx_group->index;
5466 	}
5467 
5468 	if (slot == -1) {
5469 		/* no slots available */
5470 		mutex_exit(&ixgbe->gen_lock);
5471 		return (ENOSPC);
5472 	}
5473 
5474 	bcopy(mac_addr, ixgbe->unicst_addr[slot].mac.addr, ETHERADDRL);
5475 	(void) ixgbe_set_rar(hw, slot, ixgbe->unicst_addr[slot].mac.addr,
5476 	    rx_group->index, IXGBE_RAH_AV);
5477 	ixgbe->unicst_addr[slot].mac.set = 1;
5478 	ixgbe->unicst_addr[slot].mac.group_index = rx_group->index;
5479 	ixgbe->unicst_avail--;
5480 
5481 	mutex_exit(&ixgbe->gen_lock);
5482 
5483 	return (0);
5484 }
5485 
5486 /*
5487  * Remove a mac address.
5488  */
5489 static int
5490 ixgbe_remmac(void *arg, const uint8_t *mac_addr)
5491 {
5492 	ixgbe_rx_group_t *rx_group = (ixgbe_rx_group_t *)arg;
5493 	ixgbe_t *ixgbe = rx_group->ixgbe;
5494 	struct ixgbe_hw *hw = &ixgbe->hw;
5495 	int slot;
5496 
5497 	mutex_enter(&ixgbe->gen_lock);
5498 
5499 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
5500 		mutex_exit(&ixgbe->gen_lock);
5501 		return (ECANCELED);
5502 	}
5503 
5504 	slot = ixgbe_unicst_find(ixgbe, mac_addr);
5505 	if (slot == -1) {
5506 		mutex_exit(&ixgbe->gen_lock);
5507 		return (EINVAL);
5508 	}
5509 
5510 	if (ixgbe->unicst_addr[slot].mac.set == 0) {
5511 		mutex_exit(&ixgbe->gen_lock);
5512 		return (EINVAL);
5513 	}
5514 
5515 	bzero(ixgbe->unicst_addr[slot].mac.addr, ETHERADDRL);
5516 	(void) ixgbe_clear_rar(hw, slot);
5517 	ixgbe->unicst_addr[slot].mac.set = 0;
5518 	ixgbe->unicst_avail++;
5519 
5520 	mutex_exit(&ixgbe->gen_lock);
5521 
5522 	return (0);
5523 }
5524