xref: /titanic_52/usr/src/uts/common/io/ixgbe/ixgbe_main.c (revision 3b133bec939f5230f040960ee1503dadd3dff343)
1 /*
2  * CDDL HEADER START
3  *
4  * Copyright(c) 2007-2009 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 1.1.4";
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_log(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_log(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_log(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 	 * Setup adapter interrupt vectors
1254 	 */
1255 	ixgbe_setup_adapter_vector(ixgbe);
1256 
1257 	/*
1258 	 * Initialize unicast addresses.
1259 	 */
1260 	ixgbe_init_unicst(ixgbe);
1261 
1262 	/*
1263 	 * Setup and initialize the mctable structures.
1264 	 */
1265 	ixgbe_setup_multicst(ixgbe);
1266 
1267 	/*
1268 	 * Set interrupt throttling rate
1269 	 */
1270 	for (i = 0; i < ixgbe->intr_cnt; i++) {
1271 		IXGBE_WRITE_REG(hw, IXGBE_EITR(i), ixgbe->intr_throttling[i]);
1272 	}
1273 
1274 	/*
1275 	 * Save the state of the phy
1276 	 */
1277 	ixgbe_get_hw_state(ixgbe);
1278 
1279 	/*
1280 	 * Make sure driver has control
1281 	 */
1282 	ixgbe_get_driver_control(hw);
1283 
1284 	return (IXGBE_SUCCESS);
1285 }
1286 
1287 /*
1288  * ixgbe_chip_stop - Stop the chipset hardware
1289  */
1290 static void
1291 ixgbe_chip_stop(ixgbe_t *ixgbe)
1292 {
1293 	struct ixgbe_hw *hw = &ixgbe->hw;
1294 
1295 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1296 
1297 	/*
1298 	 * Tell firmware driver is no longer in control
1299 	 */
1300 	ixgbe_release_driver_control(hw);
1301 
1302 	/*
1303 	 * Reset the chipset
1304 	 */
1305 	(void) ixgbe_reset_hw(hw);
1306 
1307 	/*
1308 	 * Reset PHY
1309 	 */
1310 	(void) ixgbe_reset_phy(hw);
1311 }
1312 
1313 /*
1314  * ixgbe_reset - Reset the chipset and re-start the driver.
1315  *
1316  * It involves stopping and re-starting the chipset,
1317  * and re-configuring the rx/tx rings.
1318  */
1319 static int
1320 ixgbe_reset(ixgbe_t *ixgbe)
1321 {
1322 	int i;
1323 
1324 	/*
1325 	 * Disable and stop the watchdog timer
1326 	 */
1327 	ixgbe_disable_watchdog_timer(ixgbe);
1328 
1329 	mutex_enter(&ixgbe->gen_lock);
1330 
1331 	ASSERT(ixgbe->ixgbe_state & IXGBE_STARTED);
1332 	atomic_and_32(&ixgbe->ixgbe_state, ~IXGBE_STARTED);
1333 
1334 	ixgbe_stop(ixgbe, B_FALSE);
1335 
1336 	if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
1337 		mutex_exit(&ixgbe->gen_lock);
1338 		return (IXGBE_FAILURE);
1339 	}
1340 
1341 	/*
1342 	 * After resetting, need to recheck the link status.
1343 	 */
1344 	ixgbe->link_check_complete = B_FALSE;
1345 	ixgbe->link_check_hrtime = gethrtime() +
1346 	    (IXGBE_LINK_UP_TIME * 100000000ULL);
1347 
1348 	atomic_or_32(&ixgbe->ixgbe_state, IXGBE_STARTED);
1349 
1350 	if (!(ixgbe->ixgbe_state & IXGBE_SUSPENDED)) {
1351 		for (i = 0; i < ixgbe->num_tx_rings; i++) {
1352 			mac_tx_ring_update(ixgbe->mac_hdl,
1353 			    ixgbe->tx_rings[i].ring_handle);
1354 		}
1355 	}
1356 
1357 	mutex_exit(&ixgbe->gen_lock);
1358 
1359 	/*
1360 	 * Enable and start the watchdog timer
1361 	 */
1362 	ixgbe_enable_watchdog_timer(ixgbe);
1363 
1364 	return (IXGBE_SUCCESS);
1365 }
1366 
1367 /*
1368  * ixgbe_tx_clean - Clean the pending transmit packets and DMA resources.
1369  */
1370 static void
1371 ixgbe_tx_clean(ixgbe_t *ixgbe)
1372 {
1373 	ixgbe_tx_ring_t *tx_ring;
1374 	tx_control_block_t *tcb;
1375 	link_list_t pending_list;
1376 	uint32_t desc_num;
1377 	int i, j;
1378 
1379 	LINK_LIST_INIT(&pending_list);
1380 
1381 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1382 		tx_ring = &ixgbe->tx_rings[i];
1383 
1384 		mutex_enter(&tx_ring->recycle_lock);
1385 
1386 		/*
1387 		 * Clean the pending tx data - the pending packets in the
1388 		 * work_list that have no chances to be transmitted again.
1389 		 *
1390 		 * We must ensure the chipset is stopped or the link is down
1391 		 * before cleaning the transmit packets.
1392 		 */
1393 		desc_num = 0;
1394 		for (j = 0; j < tx_ring->ring_size; j++) {
1395 			tcb = tx_ring->work_list[j];
1396 			if (tcb != NULL) {
1397 				desc_num += tcb->desc_num;
1398 
1399 				tx_ring->work_list[j] = NULL;
1400 
1401 				ixgbe_free_tcb(tcb);
1402 
1403 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1404 			}
1405 		}
1406 
1407 		if (desc_num > 0) {
1408 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1409 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1410 
1411 			/*
1412 			 * Reset the head and tail pointers of the tbd ring;
1413 			 * Reset the writeback head if it's enable.
1414 			 */
1415 			tx_ring->tbd_head = 0;
1416 			tx_ring->tbd_tail = 0;
1417 			if (ixgbe->tx_head_wb_enable)
1418 				*tx_ring->tbd_head_wb = 0;
1419 
1420 			IXGBE_WRITE_REG(&ixgbe->hw,
1421 			    IXGBE_TDH(tx_ring->index), 0);
1422 			IXGBE_WRITE_REG(&ixgbe->hw,
1423 			    IXGBE_TDT(tx_ring->index), 0);
1424 		}
1425 
1426 		mutex_exit(&tx_ring->recycle_lock);
1427 
1428 		/*
1429 		 * Add the tx control blocks in the pending list to
1430 		 * the free list.
1431 		 */
1432 		ixgbe_put_free_list(tx_ring, &pending_list);
1433 	}
1434 }
1435 
1436 /*
1437  * ixgbe_tx_drain - Drain the tx rings to allow pending packets to be
1438  * transmitted.
1439  */
1440 static boolean_t
1441 ixgbe_tx_drain(ixgbe_t *ixgbe)
1442 {
1443 	ixgbe_tx_ring_t *tx_ring;
1444 	boolean_t done;
1445 	int i, j;
1446 
1447 	/*
1448 	 * Wait for a specific time to allow pending tx packets
1449 	 * to be transmitted.
1450 	 *
1451 	 * Check the counter tbd_free to see if transmission is done.
1452 	 * No lock protection is needed here.
1453 	 *
1454 	 * Return B_TRUE if all pending packets have been transmitted;
1455 	 * Otherwise return B_FALSE;
1456 	 */
1457 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1458 
1459 		done = B_TRUE;
1460 		for (j = 0; j < ixgbe->num_tx_rings; j++) {
1461 			tx_ring = &ixgbe->tx_rings[j];
1462 			done = done &&
1463 			    (tx_ring->tbd_free == tx_ring->ring_size);
1464 		}
1465 
1466 		if (done)
1467 			break;
1468 
1469 		msec_delay(1);
1470 	}
1471 
1472 	return (done);
1473 }
1474 
1475 /*
1476  * ixgbe_rx_drain - Wait for all rx buffers to be released by upper layer.
1477  */
1478 static boolean_t
1479 ixgbe_rx_drain(ixgbe_t *ixgbe)
1480 {
1481 	boolean_t done = B_TRUE;
1482 	int i;
1483 
1484 	/*
1485 	 * Polling the rx free list to check if those rx buffers held by
1486 	 * the upper layer are released.
1487 	 *
1488 	 * Check the counter rcb_free to see if all pending buffers are
1489 	 * released. No lock protection is needed here.
1490 	 *
1491 	 * Return B_TRUE if all pending buffers have been released;
1492 	 * Otherwise return B_FALSE;
1493 	 */
1494 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1495 		done = (ixgbe->rcb_pending == 0);
1496 
1497 		if (done)
1498 			break;
1499 
1500 		msec_delay(1);
1501 	}
1502 
1503 	return (done);
1504 }
1505 
1506 /*
1507  * ixgbe_start - Start the driver/chipset.
1508  */
1509 int
1510 ixgbe_start(ixgbe_t *ixgbe, boolean_t alloc_buffer)
1511 {
1512 	int i;
1513 
1514 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1515 
1516 	if (alloc_buffer) {
1517 		if (ixgbe_alloc_rx_data(ixgbe) != IXGBE_SUCCESS) {
1518 			ixgbe_error(ixgbe,
1519 			    "Failed to allocate software receive rings");
1520 			return (IXGBE_FAILURE);
1521 		}
1522 
1523 		/* Allocate buffers for all the rx/tx rings */
1524 		if (ixgbe_alloc_dma(ixgbe) != IXGBE_SUCCESS) {
1525 			ixgbe_error(ixgbe, "Failed to allocate DMA resource");
1526 			return (IXGBE_FAILURE);
1527 		}
1528 
1529 		ixgbe->tx_ring_init = B_TRUE;
1530 	} else {
1531 		ixgbe->tx_ring_init = B_FALSE;
1532 	}
1533 
1534 	for (i = 0; i < ixgbe->num_rx_rings; i++)
1535 		mutex_enter(&ixgbe->rx_rings[i].rx_lock);
1536 	for (i = 0; i < ixgbe->num_tx_rings; i++)
1537 		mutex_enter(&ixgbe->tx_rings[i].tx_lock);
1538 
1539 	/*
1540 	 * Start the chipset hardware
1541 	 */
1542 	if (ixgbe_chip_start(ixgbe) != IXGBE_SUCCESS) {
1543 		ixgbe_fm_ereport(ixgbe, DDI_FM_DEVICE_INVAL_STATE);
1544 		goto start_failure;
1545 	}
1546 
1547 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1548 		goto start_failure;
1549 	}
1550 
1551 	/*
1552 	 * Setup the rx/tx rings
1553 	 */
1554 	ixgbe_setup_rings(ixgbe);
1555 
1556 	/*
1557 	 * ixgbe_start() will be called when resetting, however if reset
1558 	 * happens, we need to clear the ERROR and STALL flags before
1559 	 * enabling the interrupts.
1560 	 */
1561 	atomic_and_32(&ixgbe->ixgbe_state, ~(IXGBE_ERROR | IXGBE_STALL));
1562 
1563 	/*
1564 	 * Enable adapter interrupts
1565 	 * The interrupts must be enabled after the driver state is START
1566 	 */
1567 	ixgbe_enable_adapter_interrupts(ixgbe);
1568 
1569 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1570 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1571 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1572 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1573 
1574 	return (IXGBE_SUCCESS);
1575 
1576 start_failure:
1577 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1578 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1579 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1580 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1581 
1582 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1583 
1584 	return (IXGBE_FAILURE);
1585 }
1586 
1587 /*
1588  * ixgbe_stop - Stop the driver/chipset.
1589  */
1590 void
1591 ixgbe_stop(ixgbe_t *ixgbe, boolean_t free_buffer)
1592 {
1593 	int i;
1594 
1595 	ASSERT(mutex_owned(&ixgbe->gen_lock));
1596 
1597 	/*
1598 	 * Disable the adapter interrupts
1599 	 */
1600 	ixgbe_disable_adapter_interrupts(ixgbe);
1601 
1602 	/*
1603 	 * Drain the pending tx packets
1604 	 */
1605 	(void) ixgbe_tx_drain(ixgbe);
1606 
1607 	for (i = 0; i < ixgbe->num_rx_rings; i++)
1608 		mutex_enter(&ixgbe->rx_rings[i].rx_lock);
1609 	for (i = 0; i < ixgbe->num_tx_rings; i++)
1610 		mutex_enter(&ixgbe->tx_rings[i].tx_lock);
1611 
1612 	/*
1613 	 * Stop the chipset hardware
1614 	 */
1615 	ixgbe_chip_stop(ixgbe);
1616 
1617 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
1618 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1619 	}
1620 
1621 	/*
1622 	 * Clean the pending tx data/resources
1623 	 */
1624 	ixgbe_tx_clean(ixgbe);
1625 
1626 	for (i = ixgbe->num_tx_rings - 1; i >= 0; i--)
1627 		mutex_exit(&ixgbe->tx_rings[i].tx_lock);
1628 	for (i = ixgbe->num_rx_rings - 1; i >= 0; i--)
1629 		mutex_exit(&ixgbe->rx_rings[i].rx_lock);
1630 
1631 	if (ixgbe->link_state == LINK_STATE_UP) {
1632 		ixgbe->link_state = LINK_STATE_UNKNOWN;
1633 		mac_link_update(ixgbe->mac_hdl, ixgbe->link_state);
1634 	}
1635 
1636 	if (free_buffer) {
1637 		/*
1638 		 * Release the DMA/memory resources of rx/tx rings
1639 		 */
1640 		ixgbe_free_dma(ixgbe);
1641 		ixgbe_free_rx_data(ixgbe);
1642 	}
1643 }
1644 
1645 /*
1646  * ixgbe_cbfunc - Driver interface for generic DDI callbacks
1647  */
1648 /* ARGSUSED */
1649 static int
1650 ixgbe_cbfunc(dev_info_t *dip, ddi_cb_action_t cbaction, void *cbarg,
1651     void *arg1, void *arg2)
1652 {
1653 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
1654 
1655 	switch (cbaction) {
1656 	/* IRM callback */
1657 	int count;
1658 	case DDI_CB_INTR_ADD:
1659 	case DDI_CB_INTR_REMOVE:
1660 		count = (int)(uintptr_t)cbarg;
1661 		ASSERT(ixgbe->intr_type == DDI_INTR_TYPE_MSIX);
1662 		DTRACE_PROBE2(ixgbe__irm__callback, int, count,
1663 		    int, ixgbe->intr_cnt);
1664 		if (ixgbe_intr_adjust(ixgbe, cbaction, count) !=
1665 		    DDI_SUCCESS) {
1666 			ixgbe_error(ixgbe,
1667 			    "IRM CB: Failed to adjust interrupts");
1668 			goto cb_fail;
1669 		}
1670 		break;
1671 	default:
1672 		IXGBE_DEBUGLOG_1(ixgbe, "DDI CB: action 0x%x NOT supported",
1673 		    cbaction);
1674 		return (DDI_ENOTSUP);
1675 	}
1676 	return (DDI_SUCCESS);
1677 cb_fail:
1678 	return (DDI_FAILURE);
1679 }
1680 
1681 /*
1682  * ixgbe_intr_adjust - Adjust interrupt to respond to IRM request.
1683  */
1684 static int
1685 ixgbe_intr_adjust(ixgbe_t *ixgbe, ddi_cb_action_t cbaction, int count)
1686 {
1687 	int i, rc, actual;
1688 
1689 	if (count == 0)
1690 		return (DDI_SUCCESS);
1691 
1692 	if ((cbaction == DDI_CB_INTR_ADD &&
1693 	    ixgbe->intr_cnt + count > ixgbe->intr_cnt_max) ||
1694 	    (cbaction == DDI_CB_INTR_REMOVE &&
1695 	    ixgbe->intr_cnt - count < ixgbe->intr_cnt_min))
1696 		return (DDI_FAILURE);
1697 
1698 	if (!(ixgbe->ixgbe_state & IXGBE_STARTED)) {
1699 		return (DDI_FAILURE);
1700 	}
1701 
1702 	for (i = 0; i < ixgbe->num_rx_rings; i++)
1703 		mac_ring_intr_set(ixgbe->rx_rings[i].ring_handle, NULL);
1704 	for (i = 0; i < ixgbe->num_tx_rings; i++)
1705 		mac_ring_intr_set(ixgbe->tx_rings[i].ring_handle, NULL);
1706 
1707 	mutex_enter(&ixgbe->gen_lock);
1708 	ixgbe->ixgbe_state &= ~IXGBE_STARTED;
1709 	ixgbe->ixgbe_state |= IXGBE_INTR_ADJUST;
1710 	ixgbe->ixgbe_state |= IXGBE_SUSPENDED;
1711 	mac_link_update(ixgbe->mac_hdl, LINK_STATE_UNKNOWN);
1712 
1713 	ixgbe_stop(ixgbe, B_FALSE);
1714 	/*
1715 	 * Disable interrupts
1716 	 */
1717 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
1718 		rc = ixgbe_disable_intrs(ixgbe);
1719 		ASSERT(rc == IXGBE_SUCCESS);
1720 	}
1721 	ixgbe->attach_progress &= ~ATTACH_PROGRESS_ENABLE_INTR;
1722 
1723 	/*
1724 	 * Remove interrupt handlers
1725 	 */
1726 	if (ixgbe->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
1727 		ixgbe_rem_intr_handlers(ixgbe);
1728 	}
1729 	ixgbe->attach_progress &= ~ATTACH_PROGRESS_ADD_INTR;
1730 
1731 	/*
1732 	 * Clear vect_map
1733 	 */
1734 	bzero(&ixgbe->vect_map, sizeof (ixgbe->vect_map));
1735 	switch (cbaction) {
1736 	case DDI_CB_INTR_ADD:
1737 		rc = ddi_intr_alloc(ixgbe->dip, ixgbe->htable,
1738 		    DDI_INTR_TYPE_MSIX, ixgbe->intr_cnt, count, &actual,
1739 		    DDI_INTR_ALLOC_NORMAL);
1740 		if (rc != DDI_SUCCESS || actual != count) {
1741 			ixgbe_log(ixgbe, "Adjust interrupts failed."
1742 			    "return: %d, irm cb size: %d, actual: %d",
1743 			    rc, count, actual);
1744 			goto intr_adjust_fail;
1745 		}
1746 		ixgbe->intr_cnt += count;
1747 		break;
1748 
1749 	case DDI_CB_INTR_REMOVE:
1750 		for (i = ixgbe->intr_cnt - count;
1751 		    i < ixgbe->intr_cnt; i ++) {
1752 			rc = ddi_intr_free(ixgbe->htable[i]);
1753 			ixgbe->htable[i] = NULL;
1754 			if (rc != DDI_SUCCESS) {
1755 				ixgbe_log(ixgbe, "Adjust interrupts failed."
1756 				    "return: %d, irm cb size: %d, actual: %d",
1757 				    rc, count, actual);
1758 				goto intr_adjust_fail;
1759 			}
1760 		}
1761 		ixgbe->intr_cnt -= count;
1762 		break;
1763 	}
1764 
1765 	/*
1766 	 * Get priority for first vector, assume remaining are all the same
1767 	 */
1768 	rc = ddi_intr_get_pri(ixgbe->htable[0], &ixgbe->intr_pri);
1769 	if (rc != DDI_SUCCESS) {
1770 		ixgbe_log(ixgbe,
1771 		    "Get interrupt priority failed: %d", rc);
1772 		goto intr_adjust_fail;
1773 	}
1774 	rc = ddi_intr_get_cap(ixgbe->htable[0], &ixgbe->intr_cap);
1775 	if (rc != DDI_SUCCESS) {
1776 		ixgbe_log(ixgbe, "Get interrupt cap failed: %d", rc);
1777 		goto intr_adjust_fail;
1778 	}
1779 	ixgbe->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
1780 
1781 	/*
1782 	 * Map rings to interrupt vectors
1783 	 */
1784 	if (ixgbe_map_intrs_to_vectors(ixgbe) != IXGBE_SUCCESS) {
1785 		ixgbe_error(ixgbe,
1786 		    "IRM CB: Failed to map interrupts to vectors");
1787 		goto intr_adjust_fail;
1788 	}
1789 
1790 	/*
1791 	 * Add interrupt handlers
1792 	 */
1793 	if (ixgbe_add_intr_handlers(ixgbe) != IXGBE_SUCCESS) {
1794 		ixgbe_error(ixgbe, "IRM CB: Failed to add interrupt handlers");
1795 		goto intr_adjust_fail;
1796 	}
1797 	ixgbe->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
1798 
1799 	/*
1800 	 * Now that mutex locks are initialized, and the chip is also
1801 	 * initialized, enable interrupts.
1802 	 */
1803 	if (ixgbe_enable_intrs(ixgbe) != IXGBE_SUCCESS) {
1804 		ixgbe_error(ixgbe, "IRM CB: Failed to enable DDI interrupts");
1805 		goto intr_adjust_fail;
1806 	}
1807 	ixgbe->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
1808 	if (ixgbe_start(ixgbe, B_FALSE) != IXGBE_SUCCESS) {
1809 		ixgbe_error(ixgbe, "IRM CB: Failed to start");
1810 		goto intr_adjust_fail;
1811 	}
1812 	ixgbe->ixgbe_state &= ~IXGBE_INTR_ADJUST;
1813 	ixgbe->ixgbe_state &= ~IXGBE_SUSPENDED;
1814 	ixgbe->ixgbe_state |= IXGBE_STARTED;
1815 	mutex_exit(&ixgbe->gen_lock);
1816 
1817 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1818 		mac_ring_intr_set(ixgbe->rx_rings[i].ring_handle,
1819 		    ixgbe->htable[ixgbe->rx_rings[i].intr_vector]);
1820 	}
1821 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1822 		mac_ring_intr_set(ixgbe->tx_rings[i].ring_handle,
1823 		    ixgbe->htable[ixgbe->tx_rings[i].intr_vector]);
1824 	}
1825 
1826 	/* Wakeup all Tx rings */
1827 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
1828 		mac_tx_ring_update(ixgbe->mac_hdl,
1829 		    ixgbe->tx_rings[i].ring_handle);
1830 	}
1831 
1832 	IXGBE_DEBUGLOG_3(ixgbe,
1833 	    "IRM CB: interrupts new value: 0x%x(0x%x:0x%x).",
1834 	    ixgbe->intr_cnt, ixgbe->intr_cnt_min, ixgbe->intr_cnt_max);
1835 	return (DDI_SUCCESS);
1836 
1837 intr_adjust_fail:
1838 	ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_LOST);
1839 	mutex_exit(&ixgbe->gen_lock);
1840 	return (DDI_FAILURE);
1841 }
1842 
1843 /*
1844  * ixgbe_intr_cb_register - Register interrupt callback function.
1845  */
1846 static int
1847 ixgbe_intr_cb_register(ixgbe_t *ixgbe)
1848 {
1849 	if (ddi_cb_register(ixgbe->dip, DDI_CB_FLAG_INTR, ixgbe_cbfunc,
1850 	    ixgbe, NULL, &ixgbe->cb_hdl) != DDI_SUCCESS) {
1851 		return (IXGBE_FAILURE);
1852 	}
1853 	IXGBE_DEBUGLOG_0(ixgbe, "Interrupt callback function registered.");
1854 	return (IXGBE_SUCCESS);
1855 }
1856 
1857 /*
1858  * ixgbe_alloc_rings - Allocate memory space for rx/tx rings.
1859  */
1860 static int
1861 ixgbe_alloc_rings(ixgbe_t *ixgbe)
1862 {
1863 	/*
1864 	 * Allocate memory space for rx rings
1865 	 */
1866 	ixgbe->rx_rings = kmem_zalloc(
1867 	    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings,
1868 	    KM_NOSLEEP);
1869 
1870 	if (ixgbe->rx_rings == NULL) {
1871 		return (IXGBE_FAILURE);
1872 	}
1873 
1874 	/*
1875 	 * Allocate memory space for tx rings
1876 	 */
1877 	ixgbe->tx_rings = kmem_zalloc(
1878 	    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings,
1879 	    KM_NOSLEEP);
1880 
1881 	if (ixgbe->tx_rings == NULL) {
1882 		kmem_free(ixgbe->rx_rings,
1883 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
1884 		ixgbe->rx_rings = NULL;
1885 		return (IXGBE_FAILURE);
1886 	}
1887 
1888 	/*
1889 	 * Allocate memory space for rx ring groups
1890 	 */
1891 	ixgbe->rx_groups = kmem_zalloc(
1892 	    sizeof (ixgbe_rx_group_t) * ixgbe->num_rx_groups,
1893 	    KM_NOSLEEP);
1894 
1895 	if (ixgbe->rx_groups == NULL) {
1896 		kmem_free(ixgbe->rx_rings,
1897 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
1898 		kmem_free(ixgbe->tx_rings,
1899 		    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings);
1900 		ixgbe->rx_rings = NULL;
1901 		ixgbe->tx_rings = NULL;
1902 		return (IXGBE_FAILURE);
1903 	}
1904 
1905 	return (IXGBE_SUCCESS);
1906 }
1907 
1908 /*
1909  * ixgbe_free_rings - Free the memory space of rx/tx rings.
1910  */
1911 static void
1912 ixgbe_free_rings(ixgbe_t *ixgbe)
1913 {
1914 	if (ixgbe->rx_rings != NULL) {
1915 		kmem_free(ixgbe->rx_rings,
1916 		    sizeof (ixgbe_rx_ring_t) * ixgbe->num_rx_rings);
1917 		ixgbe->rx_rings = NULL;
1918 	}
1919 
1920 	if (ixgbe->tx_rings != NULL) {
1921 		kmem_free(ixgbe->tx_rings,
1922 		    sizeof (ixgbe_tx_ring_t) * ixgbe->num_tx_rings);
1923 		ixgbe->tx_rings = NULL;
1924 	}
1925 
1926 	if (ixgbe->rx_groups != NULL) {
1927 		kmem_free(ixgbe->rx_groups,
1928 		    sizeof (ixgbe_rx_group_t) * ixgbe->num_rx_groups);
1929 		ixgbe->rx_groups = NULL;
1930 	}
1931 }
1932 
1933 static int
1934 ixgbe_alloc_rx_data(ixgbe_t *ixgbe)
1935 {
1936 	ixgbe_rx_ring_t *rx_ring;
1937 	int i;
1938 
1939 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1940 		rx_ring = &ixgbe->rx_rings[i];
1941 		if (ixgbe_alloc_rx_ring_data(rx_ring) != IXGBE_SUCCESS)
1942 			goto alloc_rx_rings_failure;
1943 	}
1944 	return (IXGBE_SUCCESS);
1945 
1946 alloc_rx_rings_failure:
1947 	ixgbe_free_rx_data(ixgbe);
1948 	return (IXGBE_FAILURE);
1949 }
1950 
1951 static void
1952 ixgbe_free_rx_data(ixgbe_t *ixgbe)
1953 {
1954 	ixgbe_rx_ring_t *rx_ring;
1955 	ixgbe_rx_data_t *rx_data;
1956 	int i;
1957 
1958 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
1959 		rx_ring = &ixgbe->rx_rings[i];
1960 
1961 		mutex_enter(&ixgbe->rx_pending_lock);
1962 		rx_data = rx_ring->rx_data;
1963 
1964 		if (rx_data != NULL) {
1965 			rx_data->flag |= IXGBE_RX_STOPPED;
1966 
1967 			if (rx_data->rcb_pending == 0) {
1968 				ixgbe_free_rx_ring_data(rx_data);
1969 				rx_ring->rx_data = NULL;
1970 			}
1971 		}
1972 
1973 		mutex_exit(&ixgbe->rx_pending_lock);
1974 	}
1975 }
1976 
1977 /*
1978  * ixgbe_setup_rings - Setup rx/tx rings.
1979  */
1980 static void
1981 ixgbe_setup_rings(ixgbe_t *ixgbe)
1982 {
1983 	/*
1984 	 * Setup the rx/tx rings, including the following:
1985 	 *
1986 	 * 1. Setup the descriptor ring and the control block buffers;
1987 	 * 2. Initialize necessary registers for receive/transmit;
1988 	 * 3. Initialize software pointers/parameters for receive/transmit;
1989 	 */
1990 	ixgbe_setup_rx(ixgbe);
1991 
1992 	ixgbe_setup_tx(ixgbe);
1993 }
1994 
1995 static void
1996 ixgbe_setup_rx_ring(ixgbe_rx_ring_t *rx_ring)
1997 {
1998 	ixgbe_t *ixgbe = rx_ring->ixgbe;
1999 	ixgbe_rx_data_t *rx_data = rx_ring->rx_data;
2000 	struct ixgbe_hw *hw = &ixgbe->hw;
2001 	rx_control_block_t *rcb;
2002 	union ixgbe_adv_rx_desc	*rbd;
2003 	uint32_t size;
2004 	uint32_t buf_low;
2005 	uint32_t buf_high;
2006 	uint32_t reg_val;
2007 	int i;
2008 
2009 	ASSERT(mutex_owned(&rx_ring->rx_lock));
2010 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2011 
2012 	for (i = 0; i < ixgbe->rx_ring_size; i++) {
2013 		rcb = rx_data->work_list[i];
2014 		rbd = &rx_data->rbd_ring[i];
2015 
2016 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
2017 		rbd->read.hdr_addr = NULL;
2018 	}
2019 
2020 	/*
2021 	 * Initialize the length register
2022 	 */
2023 	size = rx_data->ring_size * sizeof (union ixgbe_adv_rx_desc);
2024 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(rx_ring->hw_index), size);
2025 
2026 	/*
2027 	 * Initialize the base address registers
2028 	 */
2029 	buf_low = (uint32_t)rx_data->rbd_area.dma_address;
2030 	buf_high = (uint32_t)(rx_data->rbd_area.dma_address >> 32);
2031 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(rx_ring->hw_index), buf_high);
2032 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(rx_ring->hw_index), buf_low);
2033 
2034 	/*
2035 	 * Setup head & tail pointers
2036 	 */
2037 	IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->hw_index),
2038 	    rx_data->ring_size - 1);
2039 	IXGBE_WRITE_REG(hw, IXGBE_RDH(rx_ring->hw_index), 0);
2040 
2041 	rx_data->rbd_next = 0;
2042 	rx_data->lro_first = 0;
2043 
2044 	/*
2045 	 * Setup the Receive Descriptor Control Register (RXDCTL)
2046 	 * PTHRESH=32 descriptors (half the internal cache)
2047 	 * HTHRESH=0 descriptors (to minimize latency on fetch)
2048 	 * WTHRESH defaults to 1 (writeback each descriptor)
2049 	 */
2050 	reg_val = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rx_ring->hw_index));
2051 	reg_val |= IXGBE_RXDCTL_ENABLE;	/* enable queue */
2052 
2053 	/* Not a valid value for 82599 */
2054 	if (hw->mac.type < ixgbe_mac_82599EB) {
2055 		reg_val |= 0x0020;	/* pthresh */
2056 	}
2057 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rx_ring->hw_index), reg_val);
2058 
2059 	if (hw->mac.type == ixgbe_mac_82599EB) {
2060 		reg_val = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2061 		reg_val |= (IXGBE_RDRXCTL_CRCSTRIP | IXGBE_RDRXCTL_AGGDIS);
2062 		IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, reg_val);
2063 	}
2064 
2065 	/*
2066 	 * Setup the Split and Replication Receive Control Register.
2067 	 * Set the rx buffer size and the advanced descriptor type.
2068 	 */
2069 	reg_val = (ixgbe->rx_buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) |
2070 	    IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2071 	reg_val |= IXGBE_SRRCTL_DROP_EN;
2072 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rx_ring->hw_index), reg_val);
2073 }
2074 
2075 static void
2076 ixgbe_setup_rx(ixgbe_t *ixgbe)
2077 {
2078 	ixgbe_rx_ring_t *rx_ring;
2079 	struct ixgbe_hw *hw = &ixgbe->hw;
2080 	uint32_t reg_val;
2081 	uint32_t ring_mapping;
2082 	uint32_t i, index;
2083 	uint32_t psrtype_rss_bit;
2084 
2085 	/* PSRTYPE must be configured for 82599 */
2086 	if (ixgbe->classify_mode != IXGBE_CLASSIFY_VMDQ &&
2087 	    ixgbe->classify_mode != IXGBE_CLASSIFY_VMDQ_RSS) {
2088 		reg_val = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
2089 		    IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR;
2090 		reg_val |= IXGBE_PSRTYPE_L2HDR;
2091 		reg_val |= 0x80000000;
2092 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), reg_val);
2093 	} else {
2094 		if (ixgbe->num_rx_groups > 32) {
2095 			psrtype_rss_bit = 0x20000000;
2096 		} else {
2097 			psrtype_rss_bit = 0x40000000;
2098 		}
2099 		for (i = 0; i < ixgbe->capab->max_rx_grp_num; i++) {
2100 			reg_val = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
2101 			    IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR;
2102 			reg_val |= IXGBE_PSRTYPE_L2HDR;
2103 			reg_val |= psrtype_rss_bit;
2104 			IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(i), reg_val);
2105 		}
2106 	}
2107 
2108 	/*
2109 	 * Set filter control in FCTRL to accept broadcast packets and do
2110 	 * not pass pause frames to host.  Flow control settings are already
2111 	 * in this register, so preserve them.
2112 	 */
2113 	reg_val = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2114 	reg_val |= IXGBE_FCTRL_BAM;	/* broadcast accept mode */
2115 	reg_val |= IXGBE_FCTRL_DPF;	/* discard pause frames */
2116 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, reg_val);
2117 
2118 	/*
2119 	 * Hardware checksum settings
2120 	 */
2121 	if (ixgbe->rx_hcksum_enable) {
2122 		reg_val = IXGBE_RXCSUM_IPPCSE;	/* IP checksum */
2123 		IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, reg_val);
2124 	}
2125 
2126 	/*
2127 	 * Setup VMDq and RSS for multiple receive queues
2128 	 */
2129 	switch (ixgbe->classify_mode) {
2130 	case IXGBE_CLASSIFY_RSS:
2131 		/*
2132 		 * One group, only RSS is needed when more than
2133 		 * one ring enabled.
2134 		 */
2135 		ixgbe_setup_rss(ixgbe);
2136 		break;
2137 
2138 	case IXGBE_CLASSIFY_VMDQ:
2139 		/*
2140 		 * Multiple groups, each group has one ring,
2141 		 * only VMDq is needed.
2142 		 */
2143 		ixgbe_setup_vmdq(ixgbe);
2144 		break;
2145 
2146 	case IXGBE_CLASSIFY_VMDQ_RSS:
2147 		/*
2148 		 * Multiple groups and multiple rings, both
2149 		 * VMDq and RSS are needed.
2150 		 */
2151 		ixgbe_setup_vmdq_rss(ixgbe);
2152 		break;
2153 
2154 	default:
2155 		break;
2156 	}
2157 
2158 	/*
2159 	 * Enable the receive unit.  This must be done after filter
2160 	 * control is set in FCTRL.
2161 	 */
2162 	reg_val = (IXGBE_RXCTRL_RXEN	/* Enable Receive Unit */
2163 	    | IXGBE_RXCTRL_DMBYPS);	/* descriptor monitor bypass */
2164 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, reg_val);
2165 
2166 	/*
2167 	 * ixgbe_setup_rx_ring must be called after configuring RXCTRL
2168 	 */
2169 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2170 		rx_ring = &ixgbe->rx_rings[i];
2171 		ixgbe_setup_rx_ring(rx_ring);
2172 	}
2173 
2174 	/*
2175 	 * Setup the per-ring statistics mapping.
2176 	 */
2177 	ring_mapping = 0;
2178 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
2179 		index = ixgbe->rx_rings[i].hw_index;
2180 		ring_mapping = IXGBE_READ_REG(hw, IXGBE_RQSMR(index >> 2));
2181 		ring_mapping |= (i & 0xF) << (8 * (index & 0x3));
2182 		IXGBE_WRITE_REG(hw, IXGBE_RQSMR(index >> 2), ring_mapping);
2183 	}
2184 
2185 	/*
2186 	 * The Max Frame Size in MHADD/MAXFRS will be internally increased
2187 	 * by four bytes if the packet has a VLAN field, so includes MTU,
2188 	 * ethernet header and frame check sequence.
2189 	 * Register is MAXFRS in 82599.
2190 	 */
2191 	reg_val = (ixgbe->default_mtu + sizeof (struct ether_header)
2192 	    + ETHERFCSL) << IXGBE_MHADD_MFS_SHIFT;
2193 	IXGBE_WRITE_REG(hw, IXGBE_MHADD, reg_val);
2194 
2195 	/*
2196 	 * Setup Jumbo Frame enable bit
2197 	 */
2198 	if (ixgbe->default_mtu > ETHERMTU) {
2199 		reg_val = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2200 		reg_val |= IXGBE_HLREG0_JUMBOEN;
2201 		IXGBE_WRITE_REG(hw, IXGBE_HLREG0, reg_val);
2202 	}
2203 
2204 	/*
2205 	 * Setup RSC for multiple receive queues.
2206 	 */
2207 	if (ixgbe->lro_enable) {
2208 		for (i = 0; i < ixgbe->num_rx_rings; i++) {
2209 			/*
2210 			 * Make sure rx_buf_size * MAXDESC not greater
2211 			 * than 65535.
2212 			 * Intel recommends 4 for MAXDESC field value.
2213 			 */
2214 			reg_val = IXGBE_READ_REG(hw, IXGBE_RSCCTL(i));
2215 			reg_val |= IXGBE_RSCCTL_RSCEN;
2216 			if (ixgbe->rx_buf_size == IXGBE_PKG_BUF_16k)
2217 				reg_val |= IXGBE_RSCCTL_MAXDESC_1;
2218 			else
2219 				reg_val |= IXGBE_RSCCTL_MAXDESC_4;
2220 			IXGBE_WRITE_REG(hw,  IXGBE_RSCCTL(i), reg_val);
2221 		}
2222 
2223 		reg_val = IXGBE_READ_REG(hw, IXGBE_RSCDBU);
2224 		reg_val |= IXGBE_RSCDBU_RSCACKDIS;
2225 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU, reg_val);
2226 
2227 		reg_val = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2228 		reg_val |= IXGBE_RDRXCTL_RSCACKC;
2229 		reg_val &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2230 
2231 		IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, reg_val);
2232 	}
2233 }
2234 
2235 static void
2236 ixgbe_setup_tx_ring(ixgbe_tx_ring_t *tx_ring)
2237 {
2238 	ixgbe_t *ixgbe = tx_ring->ixgbe;
2239 	struct ixgbe_hw *hw = &ixgbe->hw;
2240 	uint32_t size;
2241 	uint32_t buf_low;
2242 	uint32_t buf_high;
2243 	uint32_t reg_val;
2244 
2245 	ASSERT(mutex_owned(&tx_ring->tx_lock));
2246 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2247 
2248 	/*
2249 	 * Initialize the length register
2250 	 */
2251 	size = tx_ring->ring_size * sizeof (union ixgbe_adv_tx_desc);
2252 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(tx_ring->index), size);
2253 
2254 	/*
2255 	 * Initialize the base address registers
2256 	 */
2257 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
2258 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
2259 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(tx_ring->index), buf_low);
2260 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(tx_ring->index), buf_high);
2261 
2262 	/*
2263 	 * Setup head & tail pointers
2264 	 */
2265 	IXGBE_WRITE_REG(hw, IXGBE_TDH(tx_ring->index), 0);
2266 	IXGBE_WRITE_REG(hw, IXGBE_TDT(tx_ring->index), 0);
2267 
2268 	/*
2269 	 * Setup head write-back
2270 	 */
2271 	if (ixgbe->tx_head_wb_enable) {
2272 		/*
2273 		 * The memory of the head write-back is allocated using
2274 		 * the extra tbd beyond the tail of the tbd ring.
2275 		 */
2276 		tx_ring->tbd_head_wb = (uint32_t *)
2277 		    ((uintptr_t)tx_ring->tbd_area.address + size);
2278 		*tx_ring->tbd_head_wb = 0;
2279 
2280 		buf_low = (uint32_t)
2281 		    (tx_ring->tbd_area.dma_address + size);
2282 		buf_high = (uint32_t)
2283 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
2284 
2285 		/* Set the head write-back enable bit */
2286 		buf_low |= IXGBE_TDWBAL_HEAD_WB_ENABLE;
2287 
2288 		IXGBE_WRITE_REG(hw, IXGBE_TDWBAL(tx_ring->index), buf_low);
2289 		IXGBE_WRITE_REG(hw, IXGBE_TDWBAH(tx_ring->index), buf_high);
2290 
2291 		/*
2292 		 * Turn off relaxed ordering for head write back or it will
2293 		 * cause problems with the tx recycling
2294 		 */
2295 		reg_val = IXGBE_READ_REG(hw,
2296 		    IXGBE_DCA_TXCTRL(tx_ring->index));
2297 		reg_val &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
2298 		IXGBE_WRITE_REG(hw,
2299 		    IXGBE_DCA_TXCTRL(tx_ring->index), reg_val);
2300 	} else {
2301 		tx_ring->tbd_head_wb = NULL;
2302 	}
2303 
2304 	tx_ring->tbd_head = 0;
2305 	tx_ring->tbd_tail = 0;
2306 	tx_ring->tbd_free = tx_ring->ring_size;
2307 
2308 	if (ixgbe->tx_ring_init == B_TRUE) {
2309 		tx_ring->tcb_head = 0;
2310 		tx_ring->tcb_tail = 0;
2311 		tx_ring->tcb_free = tx_ring->free_list_size;
2312 	}
2313 
2314 	/*
2315 	 * Initialize the s/w context structure
2316 	 */
2317 	bzero(&tx_ring->tx_context, sizeof (ixgbe_tx_context_t));
2318 }
2319 
2320 static void
2321 ixgbe_setup_tx(ixgbe_t *ixgbe)
2322 {
2323 	struct ixgbe_hw *hw = &ixgbe->hw;
2324 	ixgbe_tx_ring_t *tx_ring;
2325 	uint32_t reg_val;
2326 	uint32_t ring_mapping;
2327 	int i;
2328 
2329 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2330 		tx_ring = &ixgbe->tx_rings[i];
2331 		ixgbe_setup_tx_ring(tx_ring);
2332 	}
2333 
2334 	/*
2335 	 * Setup the per-ring statistics mapping.
2336 	 */
2337 	ring_mapping = 0;
2338 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2339 		ring_mapping |= (i & 0xF) << (8 * (i & 0x3));
2340 		if ((i & 0x3) == 0x3) {
2341 			if (hw->mac.type >= ixgbe_mac_82599EB) {
2342 				IXGBE_WRITE_REG(hw, IXGBE_TQSM(i >> 2),
2343 				    ring_mapping);
2344 			} else {
2345 				IXGBE_WRITE_REG(hw, IXGBE_TQSMR(i >> 2),
2346 				    ring_mapping);
2347 			}
2348 			ring_mapping = 0;
2349 		}
2350 	}
2351 	if ((i & 0x3) != 0x3)
2352 		if (hw->mac.type >= ixgbe_mac_82599EB) {
2353 			IXGBE_WRITE_REG(hw, IXGBE_TQSM(i >> 2), ring_mapping);
2354 		} else {
2355 			IXGBE_WRITE_REG(hw, IXGBE_TQSMR(i >> 2), ring_mapping);
2356 		}
2357 
2358 	/*
2359 	 * Enable CRC appending and TX padding (for short tx frames)
2360 	 */
2361 	reg_val = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2362 	reg_val |= IXGBE_HLREG0_TXCRCEN | IXGBE_HLREG0_TXPADEN;
2363 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, reg_val);
2364 
2365 	/*
2366 	 * enable DMA for 82599 parts
2367 	 */
2368 	if (hw->mac.type == ixgbe_mac_82599EB) {
2369 	/* DMATXCTL.TE must be set after all Tx config is complete */
2370 		reg_val = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2371 		reg_val |= IXGBE_DMATXCTL_TE;
2372 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, reg_val);
2373 	}
2374 
2375 	/*
2376 	 * Enabling tx queues ..
2377 	 * For 82599 must be done after DMATXCTL.TE is set
2378 	 */
2379 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
2380 		tx_ring = &ixgbe->tx_rings[i];
2381 		reg_val = IXGBE_READ_REG(hw, IXGBE_TXDCTL(tx_ring->index));
2382 		reg_val |= IXGBE_TXDCTL_ENABLE;
2383 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(tx_ring->index), reg_val);
2384 	}
2385 }
2386 
2387 /*
2388  * ixgbe_setup_rss - Setup receive-side scaling feature.
2389  */
2390 static void
2391 ixgbe_setup_rss(ixgbe_t *ixgbe)
2392 {
2393 	struct ixgbe_hw *hw = &ixgbe->hw;
2394 	uint32_t i, mrqc, rxcsum;
2395 	uint32_t random;
2396 	uint32_t reta;
2397 	uint32_t ring_per_group;
2398 
2399 	/*
2400 	 * Fill out redirection table
2401 	 */
2402 	reta = 0;
2403 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
2404 
2405 	for (i = 0; i < 128; i++) {
2406 		reta = (reta << 8) | (i % ring_per_group) |
2407 		    ((i % ring_per_group) << 4);
2408 		if ((i & 3) == 3)
2409 			IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2410 	}
2411 
2412 	/*
2413 	 * Fill out hash function seeds with a random constant
2414 	 */
2415 	for (i = 0; i < 10; i++) {
2416 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2417 		    sizeof (uint32_t));
2418 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), random);
2419 	}
2420 
2421 	/*
2422 	 * Enable RSS & perform hash on these packet types
2423 	 */
2424 	mrqc = IXGBE_MRQC_RSSEN |
2425 	    IXGBE_MRQC_RSS_FIELD_IPV4 |
2426 	    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2427 	    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2428 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2429 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2430 	    IXGBE_MRQC_RSS_FIELD_IPV6 |
2431 	    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2432 	    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2433 	    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2434 	IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2435 
2436 	/*
2437 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2438 	 * It is an adapter hardware limitation that Packet Checksum is
2439 	 * mutually exclusive with RSS.
2440 	 */
2441 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2442 	rxcsum |= IXGBE_RXCSUM_PCSD;
2443 	rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
2444 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2445 }
2446 
2447 /*
2448  * ixgbe_setup_vmdq - Setup MAC classification feature
2449  */
2450 static void
2451 ixgbe_setup_vmdq(ixgbe_t *ixgbe)
2452 {
2453 	struct ixgbe_hw *hw = &ixgbe->hw;
2454 	uint32_t vmdctl, i, vtctl;
2455 
2456 	/*
2457 	 * Setup the VMDq Control register, enable VMDq based on
2458 	 * packet destination MAC address:
2459 	 */
2460 	switch (hw->mac.type) {
2461 	case ixgbe_mac_82598EB:
2462 		/*
2463 		 * VMDq Enable = 1;
2464 		 * VMDq Filter = 0; MAC filtering
2465 		 * Default VMDq output index = 0;
2466 		 */
2467 		vmdctl = IXGBE_VMD_CTL_VMDQ_EN;
2468 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
2469 		break;
2470 
2471 	case ixgbe_mac_82599EB:
2472 		/*
2473 		 * Enable VMDq-only.
2474 		 */
2475 		vmdctl = IXGBE_MRQC_VMDQEN;
2476 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, vmdctl);
2477 
2478 		for (i = 0; i < hw->mac.num_rar_entries; i++) {
2479 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(i), 0);
2480 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(i), 0);
2481 		}
2482 
2483 		/*
2484 		 * Enable Virtualization and Replication.
2485 		 */
2486 		vtctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2487 		IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vtctl);
2488 
2489 		/*
2490 		 * Enable receiving packets to all VFs
2491 		 */
2492 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), IXGBE_VFRE_ENABLE_ALL);
2493 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), IXGBE_VFRE_ENABLE_ALL);
2494 
2495 		break;
2496 
2497 	default:
2498 		break;
2499 	}
2500 }
2501 
2502 /*
2503  * ixgbe_setup_vmdq_rss - Setup both vmdq feature and rss feature.
2504  */
2505 static void
2506 ixgbe_setup_vmdq_rss(ixgbe_t *ixgbe)
2507 {
2508 	struct ixgbe_hw *hw = &ixgbe->hw;
2509 	uint32_t i, mrqc, rxcsum;
2510 	uint32_t random;
2511 	uint32_t reta;
2512 	uint32_t ring_per_group;
2513 	uint32_t vmdctl, vtctl;
2514 
2515 	/*
2516 	 * Fill out redirection table
2517 	 */
2518 	reta = 0;
2519 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
2520 	for (i = 0; i < 128; i++) {
2521 		reta = (reta << 8) | (i % ring_per_group) |
2522 		    ((i % ring_per_group) << 4);
2523 		if ((i & 3) == 3)
2524 			IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2525 	}
2526 
2527 	/*
2528 	 * Fill out hash function seeds with a random constant
2529 	 */
2530 	for (i = 0; i < 10; i++) {
2531 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2532 		    sizeof (uint32_t));
2533 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), random);
2534 	}
2535 
2536 	/*
2537 	 * Enable and setup RSS and VMDq
2538 	 */
2539 	switch (hw->mac.type) {
2540 	case ixgbe_mac_82598EB:
2541 		/*
2542 		 * Enable RSS & Setup RSS Hash functions
2543 		 */
2544 		mrqc = IXGBE_MRQC_RSSEN |
2545 		    IXGBE_MRQC_RSS_FIELD_IPV4 |
2546 		    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2547 		    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2548 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2549 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2550 		    IXGBE_MRQC_RSS_FIELD_IPV6 |
2551 		    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2552 		    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2553 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2554 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2555 
2556 		/*
2557 		 * Enable and Setup VMDq
2558 		 * VMDq Filter = 0; MAC filtering
2559 		 * Default VMDq output index = 0;
2560 		 */
2561 		vmdctl = IXGBE_VMD_CTL_VMDQ_EN;
2562 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
2563 		break;
2564 
2565 	case ixgbe_mac_82599EB:
2566 		/*
2567 		 * Enable RSS & Setup RSS Hash functions
2568 		 */
2569 		mrqc = IXGBE_MRQC_RSS_FIELD_IPV4 |
2570 		    IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
2571 		    IXGBE_MRQC_RSS_FIELD_IPV4_UDP |
2572 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP |
2573 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX |
2574 		    IXGBE_MRQC_RSS_FIELD_IPV6 |
2575 		    IXGBE_MRQC_RSS_FIELD_IPV6_TCP |
2576 		    IXGBE_MRQC_RSS_FIELD_IPV6_UDP |
2577 		    IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2578 
2579 		/*
2580 		 * Enable VMDq+RSS.
2581 		 */
2582 		if (ixgbe->num_rx_groups > 32)  {
2583 			mrqc = mrqc | IXGBE_MRQC_VMDQRSS64EN;
2584 		} else {
2585 			mrqc = mrqc | IXGBE_MRQC_VMDQRSS32EN;
2586 		}
2587 
2588 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2589 
2590 		for (i = 0; i < hw->mac.num_rar_entries; i++) {
2591 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(i), 0);
2592 			IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(i), 0);
2593 		}
2594 		break;
2595 
2596 	default:
2597 		break;
2598 
2599 	}
2600 
2601 	/*
2602 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2603 	 * It is an adapter hardware limitation that Packet Checksum is
2604 	 * mutually exclusive with RSS.
2605 	 */
2606 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2607 	rxcsum |= IXGBE_RXCSUM_PCSD;
2608 	rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
2609 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2610 
2611 	if (hw->mac.type == ixgbe_mac_82599EB) {
2612 		/*
2613 		 * Enable Virtualization and Replication.
2614 		 */
2615 		vtctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2616 		IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vtctl);
2617 
2618 		/*
2619 		 * Enable receiving packets to all VFs
2620 		 */
2621 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), IXGBE_VFRE_ENABLE_ALL);
2622 		IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), IXGBE_VFRE_ENABLE_ALL);
2623 	}
2624 }
2625 
2626 /*
2627  * ixgbe_init_unicst - Initialize the unicast addresses.
2628  */
2629 static void
2630 ixgbe_init_unicst(ixgbe_t *ixgbe)
2631 {
2632 	struct ixgbe_hw *hw = &ixgbe->hw;
2633 	uint8_t *mac_addr;
2634 	int slot;
2635 	/*
2636 	 * Here we should consider two situations:
2637 	 *
2638 	 * 1. Chipset is initialized at the first time,
2639 	 *    Clear all the multiple unicast addresses.
2640 	 *
2641 	 * 2. Chipset is reset
2642 	 *    Recover the multiple unicast addresses from the
2643 	 *    software data structure to the RAR registers.
2644 	 */
2645 	if (!ixgbe->unicst_init) {
2646 		/*
2647 		 * Initialize the multiple unicast addresses
2648 		 */
2649 		ixgbe->unicst_total = hw->mac.num_rar_entries;
2650 		ixgbe->unicst_avail = ixgbe->unicst_total;
2651 		for (slot = 0; slot < ixgbe->unicst_total; slot++) {
2652 			mac_addr = ixgbe->unicst_addr[slot].mac.addr;
2653 			bzero(mac_addr, ETHERADDRL);
2654 			(void) ixgbe_set_rar(hw, slot, mac_addr, NULL, NULL);
2655 			ixgbe->unicst_addr[slot].mac.set = 0;
2656 		}
2657 		ixgbe->unicst_init = B_TRUE;
2658 	} else {
2659 		/* Re-configure the RAR registers */
2660 		for (slot = 0; slot < ixgbe->unicst_total; slot++) {
2661 			mac_addr = ixgbe->unicst_addr[slot].mac.addr;
2662 			if (ixgbe->unicst_addr[slot].mac.set == 1) {
2663 				(void) ixgbe_set_rar(hw, slot, mac_addr,
2664 				    ixgbe->unicst_addr[slot].mac.group_index,
2665 				    IXGBE_RAH_AV);
2666 			} else {
2667 				bzero(mac_addr, ETHERADDRL);
2668 				(void) ixgbe_set_rar(hw, slot, mac_addr,
2669 				    NULL, NULL);
2670 			}
2671 		}
2672 	}
2673 }
2674 
2675 /*
2676  * ixgbe_unicst_find - Find the slot for the specified unicast address
2677  */
2678 int
2679 ixgbe_unicst_find(ixgbe_t *ixgbe, const uint8_t *mac_addr)
2680 {
2681 	int slot;
2682 
2683 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2684 
2685 	for (slot = 0; slot < ixgbe->unicst_total; slot++) {
2686 		if (bcmp(ixgbe->unicst_addr[slot].mac.addr,
2687 		    mac_addr, ETHERADDRL) == 0)
2688 			return (slot);
2689 	}
2690 
2691 	return (-1);
2692 }
2693 
2694 /*
2695  * ixgbe_multicst_add - Add a multicst address.
2696  */
2697 int
2698 ixgbe_multicst_add(ixgbe_t *ixgbe, const uint8_t *multiaddr)
2699 {
2700 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2701 
2702 	if ((multiaddr[0] & 01) == 0) {
2703 		return (EINVAL);
2704 	}
2705 
2706 	if (ixgbe->mcast_count >= MAX_NUM_MULTICAST_ADDRESSES) {
2707 		return (ENOENT);
2708 	}
2709 
2710 	bcopy(multiaddr,
2711 	    &ixgbe->mcast_table[ixgbe->mcast_count], ETHERADDRL);
2712 	ixgbe->mcast_count++;
2713 
2714 	/*
2715 	 * Update the multicast table in the hardware
2716 	 */
2717 	ixgbe_setup_multicst(ixgbe);
2718 
2719 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
2720 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
2721 		return (EIO);
2722 	}
2723 
2724 	return (0);
2725 }
2726 
2727 /*
2728  * ixgbe_multicst_remove - Remove a multicst address.
2729  */
2730 int
2731 ixgbe_multicst_remove(ixgbe_t *ixgbe, const uint8_t *multiaddr)
2732 {
2733 	int i;
2734 
2735 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2736 
2737 	for (i = 0; i < ixgbe->mcast_count; i++) {
2738 		if (bcmp(multiaddr, &ixgbe->mcast_table[i],
2739 		    ETHERADDRL) == 0) {
2740 			for (i++; i < ixgbe->mcast_count; i++) {
2741 				ixgbe->mcast_table[i - 1] =
2742 				    ixgbe->mcast_table[i];
2743 			}
2744 			ixgbe->mcast_count--;
2745 			break;
2746 		}
2747 	}
2748 
2749 	/*
2750 	 * Update the multicast table in the hardware
2751 	 */
2752 	ixgbe_setup_multicst(ixgbe);
2753 
2754 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
2755 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
2756 		return (EIO);
2757 	}
2758 
2759 	return (0);
2760 }
2761 
2762 /*
2763  * ixgbe_setup_multicast - Setup multicast data structures.
2764  *
2765  * This routine initializes all of the multicast related structures
2766  * and save them in the hardware registers.
2767  */
2768 static void
2769 ixgbe_setup_multicst(ixgbe_t *ixgbe)
2770 {
2771 	uint8_t *mc_addr_list;
2772 	uint32_t mc_addr_count;
2773 	struct ixgbe_hw *hw = &ixgbe->hw;
2774 
2775 	ASSERT(mutex_owned(&ixgbe->gen_lock));
2776 
2777 	ASSERT(ixgbe->mcast_count <= MAX_NUM_MULTICAST_ADDRESSES);
2778 
2779 	mc_addr_list = (uint8_t *)ixgbe->mcast_table;
2780 	mc_addr_count = ixgbe->mcast_count;
2781 
2782 	/*
2783 	 * Update the multicast addresses to the MTA registers
2784 	 */
2785 	(void) ixgbe_update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2786 	    ixgbe_mc_table_itr);
2787 }
2788 
2789 /*
2790  * ixgbe_setup_vmdq_rss_conf - Configure vmdq and rss (number and mode).
2791  *
2792  * Configure the rx classification mode (vmdq & rss) and vmdq & rss numbers.
2793  * Different chipsets may have different allowed configuration of vmdq and rss.
2794  */
2795 static void
2796 ixgbe_setup_vmdq_rss_conf(ixgbe_t *ixgbe)
2797 {
2798 	struct ixgbe_hw *hw = &ixgbe->hw;
2799 	uint32_t ring_per_group;
2800 
2801 	switch (hw->mac.type) {
2802 	case ixgbe_mac_82598EB:
2803 		/*
2804 		 * 82598 supports the following combination:
2805 		 * vmdq no. x rss no.
2806 		 * [5..16]  x 1
2807 		 * [1..4]   x [1..16]
2808 		 * However 8 rss queue per pool (vmdq) is sufficient for
2809 		 * most cases.
2810 		 */
2811 		ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
2812 		if (ixgbe->num_rx_groups > 4) {
2813 			ixgbe->num_rx_rings = ixgbe->num_rx_groups;
2814 		} else {
2815 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
2816 			    min(8, ring_per_group);
2817 		}
2818 
2819 		break;
2820 
2821 	case ixgbe_mac_82599EB:
2822 		/*
2823 		 * 82599 supports the following combination:
2824 		 * vmdq no. x rss no.
2825 		 * [33..64] x [1..2]
2826 		 * [2..32]  x [1..4]
2827 		 * 1 x [1..16]
2828 		 * However 8 rss queue per pool (vmdq) is sufficient for
2829 		 * most cases.
2830 		 */
2831 		ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
2832 		if (ixgbe->num_rx_groups == 1) {
2833 			ixgbe->num_rx_rings = min(8, ring_per_group);
2834 		} else if (ixgbe->num_rx_groups <= 32) {
2835 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
2836 			    min(4, ring_per_group);
2837 		} else if (ixgbe->num_rx_groups <= 64) {
2838 			ixgbe->num_rx_rings = ixgbe->num_rx_groups *
2839 			    min(2, ring_per_group);
2840 		}
2841 
2842 		break;
2843 
2844 	default:
2845 		break;
2846 	}
2847 
2848 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
2849 
2850 	if (ixgbe->num_rx_groups == 1 && ring_per_group == 1) {
2851 		ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
2852 	} else if (ixgbe->num_rx_groups != 1 && ring_per_group == 1) {
2853 		ixgbe->classify_mode = IXGBE_CLASSIFY_VMDQ;
2854 	} else if (ixgbe->num_rx_groups != 1 && ring_per_group != 1) {
2855 		ixgbe->classify_mode = IXGBE_CLASSIFY_VMDQ_RSS;
2856 	} else {
2857 		ixgbe->classify_mode = IXGBE_CLASSIFY_RSS;
2858 	}
2859 
2860 	ixgbe_log(ixgbe, "rx group number:%d, rx ring number:%d",
2861 	    ixgbe->num_rx_groups, ixgbe->num_rx_rings);
2862 }
2863 
2864 /*
2865  * ixgbe_get_conf - Get driver configurations set in driver.conf.
2866  *
2867  * This routine gets user-configured values out of the configuration
2868  * file ixgbe.conf.
2869  *
2870  * For each configurable value, there is a minimum, a maximum, and a
2871  * default.
2872  * If user does not configure a value, use the default.
2873  * If user configures below the minimum, use the minumum.
2874  * If user configures above the maximum, use the maxumum.
2875  */
2876 static void
2877 ixgbe_get_conf(ixgbe_t *ixgbe)
2878 {
2879 	struct ixgbe_hw *hw = &ixgbe->hw;
2880 	uint32_t flow_control;
2881 
2882 	/*
2883 	 * ixgbe driver supports the following user configurations:
2884 	 *
2885 	 * Jumbo frame configuration:
2886 	 *    default_mtu
2887 	 *
2888 	 * Ethernet flow control configuration:
2889 	 *    flow_control
2890 	 *
2891 	 * Multiple rings configurations:
2892 	 *    tx_queue_number
2893 	 *    tx_ring_size
2894 	 *    rx_queue_number
2895 	 *    rx_ring_size
2896 	 *
2897 	 * Call ixgbe_get_prop() to get the value for a specific
2898 	 * configuration parameter.
2899 	 */
2900 
2901 	/*
2902 	 * Jumbo frame configuration - max_frame_size controls host buffer
2903 	 * allocation, so includes MTU, ethernet header, vlan tag and
2904 	 * frame check sequence.
2905 	 */
2906 	ixgbe->default_mtu = ixgbe_get_prop(ixgbe, PROP_DEFAULT_MTU,
2907 	    MIN_MTU, ixgbe->capab->max_mtu, DEFAULT_MTU);
2908 
2909 	ixgbe->max_frame_size = ixgbe->default_mtu +
2910 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
2911 
2912 	/*
2913 	 * Ethernet flow control configuration
2914 	 */
2915 	flow_control = ixgbe_get_prop(ixgbe, PROP_FLOW_CONTROL,
2916 	    ixgbe_fc_none, 3, ixgbe_fc_none);
2917 	if (flow_control == 3)
2918 		flow_control = ixgbe_fc_default;
2919 
2920 	/*
2921 	 * fc.requested mode is what the user requests.  After autoneg,
2922 	 * fc.current_mode will be the flow_control mode that was negotiated.
2923 	 */
2924 	hw->fc.requested_mode = flow_control;
2925 
2926 	/*
2927 	 * Multiple rings configurations
2928 	 */
2929 	ixgbe->num_tx_rings = ixgbe_get_prop(ixgbe, PROP_TX_QUEUE_NUM,
2930 	    ixgbe->capab->min_tx_que_num,
2931 	    ixgbe->capab->max_tx_que_num,
2932 	    ixgbe->capab->def_tx_que_num);
2933 	ixgbe->tx_ring_size = ixgbe_get_prop(ixgbe, PROP_TX_RING_SIZE,
2934 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
2935 
2936 	ixgbe->num_rx_rings = ixgbe_get_prop(ixgbe, PROP_RX_QUEUE_NUM,
2937 	    ixgbe->capab->min_rx_que_num,
2938 	    ixgbe->capab->max_rx_que_num,
2939 	    ixgbe->capab->def_rx_que_num);
2940 	ixgbe->rx_ring_size = ixgbe_get_prop(ixgbe, PROP_RX_RING_SIZE,
2941 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
2942 
2943 	/*
2944 	 * Multiple groups configuration
2945 	 */
2946 	ixgbe->num_rx_groups = ixgbe_get_prop(ixgbe, PROP_RX_GROUP_NUM,
2947 	    ixgbe->capab->min_rx_grp_num, ixgbe->capab->max_rx_grp_num,
2948 	    ixgbe->capab->def_rx_grp_num);
2949 
2950 	ixgbe->mr_enable = ixgbe_get_prop(ixgbe, PROP_MR_ENABLE,
2951 	    0, 1, DEFAULT_MR_ENABLE);
2952 
2953 	if (ixgbe->mr_enable == B_FALSE) {
2954 		ixgbe->num_tx_rings = 1;
2955 		ixgbe->num_rx_rings = 1;
2956 		ixgbe->num_rx_groups = 1;
2957 		ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
2958 	} else {
2959 		ixgbe->num_rx_rings = ixgbe->num_rx_groups *
2960 		    max(ixgbe->num_rx_rings / ixgbe->num_rx_groups, 1);
2961 		/*
2962 		 * The combination of num_rx_rings and num_rx_groups
2963 		 * may be not supported by h/w. We need to adjust
2964 		 * them to appropriate values.
2965 		 */
2966 		ixgbe_setup_vmdq_rss_conf(ixgbe);
2967 	}
2968 
2969 	/*
2970 	 * Tunable used to force an interrupt type. The only use is
2971 	 * for testing of the lesser interrupt types.
2972 	 * 0 = don't force interrupt type
2973 	 * 1 = force interrupt type MSI-X
2974 	 * 2 = force interrupt type MSI
2975 	 * 3 = force interrupt type Legacy
2976 	 */
2977 	ixgbe->intr_force = ixgbe_get_prop(ixgbe, PROP_INTR_FORCE,
2978 	    IXGBE_INTR_NONE, IXGBE_INTR_LEGACY, IXGBE_INTR_NONE);
2979 
2980 	ixgbe->tx_hcksum_enable = ixgbe_get_prop(ixgbe, PROP_TX_HCKSUM_ENABLE,
2981 	    0, 1, DEFAULT_TX_HCKSUM_ENABLE);
2982 	ixgbe->rx_hcksum_enable = ixgbe_get_prop(ixgbe, PROP_RX_HCKSUM_ENABLE,
2983 	    0, 1, DEFAULT_RX_HCKSUM_ENABLE);
2984 	ixgbe->lso_enable = ixgbe_get_prop(ixgbe, PROP_LSO_ENABLE,
2985 	    0, 1, DEFAULT_LSO_ENABLE);
2986 	ixgbe->lro_enable = ixgbe_get_prop(ixgbe, PROP_LRO_ENABLE,
2987 	    0, 1, DEFAULT_LRO_ENABLE);
2988 	ixgbe->tx_head_wb_enable = ixgbe_get_prop(ixgbe, PROP_TX_HEAD_WB_ENABLE,
2989 	    0, 1, DEFAULT_TX_HEAD_WB_ENABLE);
2990 
2991 	/* Head Write Back not recommended for 82599 */
2992 	if (hw->mac.type >= ixgbe_mac_82599EB) {
2993 		ixgbe->tx_head_wb_enable = B_FALSE;
2994 	}
2995 
2996 	/*
2997 	 * ixgbe LSO needs the tx h/w checksum support.
2998 	 * LSO will be disabled if tx h/w checksum is not
2999 	 * enabled.
3000 	 */
3001 	if (ixgbe->tx_hcksum_enable == B_FALSE) {
3002 		ixgbe->lso_enable = B_FALSE;
3003 	}
3004 
3005 	/*
3006 	 * ixgbe LRO needs the rx h/w checksum support.
3007 	 * LRO will be disabled if rx h/w checksum is not
3008 	 * enabled.
3009 	 */
3010 	if (ixgbe->rx_hcksum_enable == B_FALSE) {
3011 		ixgbe->lro_enable = B_FALSE;
3012 	}
3013 
3014 	/*
3015 	 * ixgbe LRO only been supported by 82599 now
3016 	 */
3017 	if (hw->mac.type != ixgbe_mac_82599EB) {
3018 		ixgbe->lro_enable = B_FALSE;
3019 	}
3020 	ixgbe->tx_copy_thresh = ixgbe_get_prop(ixgbe, PROP_TX_COPY_THRESHOLD,
3021 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
3022 	    DEFAULT_TX_COPY_THRESHOLD);
3023 	ixgbe->tx_recycle_thresh = ixgbe_get_prop(ixgbe,
3024 	    PROP_TX_RECYCLE_THRESHOLD, MIN_TX_RECYCLE_THRESHOLD,
3025 	    MAX_TX_RECYCLE_THRESHOLD, DEFAULT_TX_RECYCLE_THRESHOLD);
3026 	ixgbe->tx_overload_thresh = ixgbe_get_prop(ixgbe,
3027 	    PROP_TX_OVERLOAD_THRESHOLD, MIN_TX_OVERLOAD_THRESHOLD,
3028 	    MAX_TX_OVERLOAD_THRESHOLD, DEFAULT_TX_OVERLOAD_THRESHOLD);
3029 	ixgbe->tx_resched_thresh = ixgbe_get_prop(ixgbe,
3030 	    PROP_TX_RESCHED_THRESHOLD, MIN_TX_RESCHED_THRESHOLD,
3031 	    MAX_TX_RESCHED_THRESHOLD, DEFAULT_TX_RESCHED_THRESHOLD);
3032 
3033 	ixgbe->rx_copy_thresh = ixgbe_get_prop(ixgbe, PROP_RX_COPY_THRESHOLD,
3034 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
3035 	    DEFAULT_RX_COPY_THRESHOLD);
3036 	ixgbe->rx_limit_per_intr = ixgbe_get_prop(ixgbe, PROP_RX_LIMIT_PER_INTR,
3037 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
3038 	    DEFAULT_RX_LIMIT_PER_INTR);
3039 
3040 	ixgbe->intr_throttling[0] = ixgbe_get_prop(ixgbe, PROP_INTR_THROTTLING,
3041 	    ixgbe->capab->min_intr_throttle,
3042 	    ixgbe->capab->max_intr_throttle,
3043 	    ixgbe->capab->def_intr_throttle);
3044 	/*
3045 	 * 82599 requires the interupt throttling rate is
3046 	 * a multiple of 8. This is enforced by the register
3047 	 * definiton.
3048 	 */
3049 	if (hw->mac.type == ixgbe_mac_82599EB)
3050 		ixgbe->intr_throttling[0] = ixgbe->intr_throttling[0] & 0xFF8;
3051 }
3052 
3053 static void
3054 ixgbe_init_params(ixgbe_t *ixgbe)
3055 {
3056 	ixgbe->param_en_10000fdx_cap = 1;
3057 	ixgbe->param_en_1000fdx_cap = 1;
3058 	ixgbe->param_en_100fdx_cap = 1;
3059 	ixgbe->param_adv_10000fdx_cap = 1;
3060 	ixgbe->param_adv_1000fdx_cap = 1;
3061 	ixgbe->param_adv_100fdx_cap = 1;
3062 
3063 	ixgbe->param_pause_cap = 1;
3064 	ixgbe->param_asym_pause_cap = 1;
3065 	ixgbe->param_rem_fault = 0;
3066 
3067 	ixgbe->param_adv_autoneg_cap = 1;
3068 	ixgbe->param_adv_pause_cap = 1;
3069 	ixgbe->param_adv_asym_pause_cap = 1;
3070 	ixgbe->param_adv_rem_fault = 0;
3071 
3072 	ixgbe->param_lp_10000fdx_cap = 0;
3073 	ixgbe->param_lp_1000fdx_cap = 0;
3074 	ixgbe->param_lp_100fdx_cap = 0;
3075 	ixgbe->param_lp_autoneg_cap = 0;
3076 	ixgbe->param_lp_pause_cap = 0;
3077 	ixgbe->param_lp_asym_pause_cap = 0;
3078 	ixgbe->param_lp_rem_fault = 0;
3079 }
3080 
3081 /*
3082  * ixgbe_get_prop - Get a property value out of the configuration file
3083  * ixgbe.conf.
3084  *
3085  * Caller provides the name of the property, a default value, a minimum
3086  * value, and a maximum value.
3087  *
3088  * Return configured value of the property, with default, minimum and
3089  * maximum properly applied.
3090  */
3091 static int
3092 ixgbe_get_prop(ixgbe_t *ixgbe,
3093     char *propname,	/* name of the property */
3094     int minval,		/* minimum acceptable value */
3095     int maxval,		/* maximim acceptable value */
3096     int defval)		/* default value */
3097 {
3098 	int value;
3099 
3100 	/*
3101 	 * Call ddi_prop_get_int() to read the conf settings
3102 	 */
3103 	value = ddi_prop_get_int(DDI_DEV_T_ANY, ixgbe->dip,
3104 	    DDI_PROP_DONTPASS, propname, defval);
3105 	if (value > maxval)
3106 		value = maxval;
3107 
3108 	if (value < minval)
3109 		value = minval;
3110 
3111 	return (value);
3112 }
3113 
3114 /*
3115  * ixgbe_driver_setup_link - Using the link properties to setup the link.
3116  */
3117 int
3118 ixgbe_driver_setup_link(ixgbe_t *ixgbe, boolean_t setup_hw)
3119 {
3120 	u32 autoneg_advertised = 0;
3121 
3122 	/*
3123 	 * No half duplex support with 10Gb parts
3124 	 */
3125 	if (ixgbe->param_adv_10000fdx_cap == 1)
3126 		autoneg_advertised |= IXGBE_LINK_SPEED_10GB_FULL;
3127 
3128 	if (ixgbe->param_adv_1000fdx_cap == 1)
3129 		autoneg_advertised |= IXGBE_LINK_SPEED_1GB_FULL;
3130 
3131 	if (ixgbe->param_adv_100fdx_cap == 1)
3132 		autoneg_advertised |= IXGBE_LINK_SPEED_100_FULL;
3133 
3134 	if (ixgbe->param_adv_autoneg_cap == 1 && autoneg_advertised == 0) {
3135 		ixgbe_notice(ixgbe, "Invalid link settings. Setup link "
3136 		    "to autonegotiation with full link capabilities.");
3137 
3138 		autoneg_advertised = IXGBE_LINK_SPEED_10GB_FULL |
3139 		    IXGBE_LINK_SPEED_1GB_FULL |
3140 		    IXGBE_LINK_SPEED_100_FULL;
3141 	}
3142 
3143 	if (setup_hw) {
3144 		if (ixgbe_setup_link(&ixgbe->hw, autoneg_advertised,
3145 		    ixgbe->param_adv_autoneg_cap, B_TRUE) != IXGBE_SUCCESS) {
3146 			ixgbe_notice(ixgbe, "Setup link failed on this "
3147 			    "device.");
3148 			return (IXGBE_FAILURE);
3149 		}
3150 	}
3151 
3152 	return (IXGBE_SUCCESS);
3153 }
3154 
3155 /*
3156  * ixgbe_driver_link_check - Link status processing.
3157  *
3158  * This function can be called in both kernel context and interrupt context
3159  */
3160 static void
3161 ixgbe_driver_link_check(ixgbe_t *ixgbe)
3162 {
3163 	struct ixgbe_hw *hw = &ixgbe->hw;
3164 	ixgbe_link_speed speed = IXGBE_LINK_SPEED_UNKNOWN;
3165 	boolean_t link_up = B_FALSE;
3166 	boolean_t link_changed = B_FALSE;
3167 
3168 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3169 
3170 	(void) ixgbe_check_link(hw, &speed, &link_up, false);
3171 	if (link_up) {
3172 		ixgbe->link_check_complete = B_TRUE;
3173 
3174 		/* Link is up, enable flow control settings */
3175 		(void) ixgbe_fc_enable(hw, 0);
3176 
3177 		/*
3178 		 * The Link is up, check whether it was marked as down earlier
3179 		 */
3180 		if (ixgbe->link_state != LINK_STATE_UP) {
3181 			switch (speed) {
3182 			case IXGBE_LINK_SPEED_10GB_FULL:
3183 				ixgbe->link_speed = SPEED_10GB;
3184 				break;
3185 			case IXGBE_LINK_SPEED_1GB_FULL:
3186 				ixgbe->link_speed = SPEED_1GB;
3187 				break;
3188 			case IXGBE_LINK_SPEED_100_FULL:
3189 				ixgbe->link_speed = SPEED_100;
3190 			}
3191 			ixgbe->link_duplex = LINK_DUPLEX_FULL;
3192 			ixgbe->link_state = LINK_STATE_UP;
3193 			link_changed = B_TRUE;
3194 		}
3195 	} else {
3196 		if (ixgbe->link_check_complete == B_TRUE ||
3197 		    (ixgbe->link_check_complete == B_FALSE &&
3198 		    gethrtime() >= ixgbe->link_check_hrtime)) {
3199 			/*
3200 			 * The link is really down
3201 			 */
3202 			ixgbe->link_check_complete = B_TRUE;
3203 
3204 			if (ixgbe->link_state != LINK_STATE_DOWN) {
3205 				ixgbe->link_speed = 0;
3206 				ixgbe->link_duplex = LINK_DUPLEX_UNKNOWN;
3207 				ixgbe->link_state = LINK_STATE_DOWN;
3208 				link_changed = B_TRUE;
3209 			}
3210 		}
3211 	}
3212 
3213 	/*
3214 	 * this is only reached after a link-status-change interrupt
3215 	 * so always get new phy state
3216 	 */
3217 	ixgbe_get_hw_state(ixgbe);
3218 
3219 	/*
3220 	 * If we are in an interrupt context, need to re-enable the
3221 	 * interrupt, which was automasked
3222 	 */
3223 	if (servicing_interrupt() != 0) {
3224 		ixgbe->eims |= IXGBE_EICR_LSC;
3225 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
3226 	}
3227 
3228 	if (link_changed) {
3229 		mac_link_update(ixgbe->mac_hdl, ixgbe->link_state);
3230 	}
3231 }
3232 
3233 /*
3234  * ixgbe_sfp_check - sfp module processing done in taskq only for 82599.
3235  */
3236 static void
3237 ixgbe_sfp_check(void *arg)
3238 {
3239 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3240 	uint32_t eicr = ixgbe->eicr;
3241 	struct ixgbe_hw *hw = &ixgbe->hw;
3242 
3243 	mutex_enter(&ixgbe->gen_lock);
3244 	if (eicr & IXGBE_EICR_GPI_SDP1) {
3245 		/* clear the interrupt */
3246 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
3247 
3248 		/* if link up, do multispeed fiber setup */
3249 		(void) ixgbe_setup_link(hw, IXGBE_LINK_SPEED_82599_AUTONEG,
3250 		    B_TRUE, B_TRUE);
3251 		ixgbe_driver_link_check(ixgbe);
3252 	} else if (eicr & IXGBE_EICR_GPI_SDP2) {
3253 		/* clear the interrupt */
3254 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
3255 
3256 		/* if link up, do sfp module setup */
3257 		(void) hw->mac.ops.setup_sfp(hw);
3258 
3259 		/* do multispeed fiber setup */
3260 		(void) ixgbe_setup_link(hw, IXGBE_LINK_SPEED_82599_AUTONEG,
3261 		    B_TRUE, B_TRUE);
3262 		ixgbe_driver_link_check(ixgbe);
3263 	}
3264 	mutex_exit(&ixgbe->gen_lock);
3265 }
3266 
3267 /*
3268  * ixgbe_link_timer - timer for link status detection
3269  */
3270 static void
3271 ixgbe_link_timer(void *arg)
3272 {
3273 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3274 
3275 	mutex_enter(&ixgbe->gen_lock);
3276 	ixgbe_driver_link_check(ixgbe);
3277 	mutex_exit(&ixgbe->gen_lock);
3278 }
3279 
3280 /*
3281  * ixgbe_local_timer - Driver watchdog function.
3282  *
3283  * This function will handle the transmit stall check and other routines.
3284  */
3285 static void
3286 ixgbe_local_timer(void *arg)
3287 {
3288 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
3289 
3290 	if (ixgbe->ixgbe_state & IXGBE_ERROR) {
3291 		ixgbe->reset_count++;
3292 		if (ixgbe_reset(ixgbe) == IXGBE_SUCCESS)
3293 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_RESTORED);
3294 		ixgbe_restart_watchdog_timer(ixgbe);
3295 		return;
3296 	}
3297 
3298 	if (ixgbe_stall_check(ixgbe)) {
3299 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_STALL);
3300 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3301 
3302 		ixgbe->reset_count++;
3303 		if (ixgbe_reset(ixgbe) == IXGBE_SUCCESS)
3304 			ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_RESTORED);
3305 	}
3306 
3307 	ixgbe_restart_watchdog_timer(ixgbe);
3308 }
3309 
3310 /*
3311  * ixgbe_stall_check - Check for transmit stall.
3312  *
3313  * This function checks if the adapter is stalled (in transmit).
3314  *
3315  * It is called each time the watchdog timeout is invoked.
3316  * If the transmit descriptor reclaim continuously fails,
3317  * the watchdog value will increment by 1. If the watchdog
3318  * value exceeds the threshold, the ixgbe is assumed to
3319  * have stalled and need to be reset.
3320  */
3321 static boolean_t
3322 ixgbe_stall_check(ixgbe_t *ixgbe)
3323 {
3324 	ixgbe_tx_ring_t *tx_ring;
3325 	boolean_t result;
3326 	int i;
3327 
3328 	if (ixgbe->link_state != LINK_STATE_UP)
3329 		return (B_FALSE);
3330 
3331 	/*
3332 	 * If any tx ring is stalled, we'll reset the chipset
3333 	 */
3334 	result = B_FALSE;
3335 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
3336 		tx_ring = &ixgbe->tx_rings[i];
3337 		if (tx_ring->tbd_free <= ixgbe->tx_recycle_thresh) {
3338 			tx_ring->tx_recycle(tx_ring);
3339 		}
3340 
3341 		if (tx_ring->recycle_fail > 0)
3342 			tx_ring->stall_watchdog++;
3343 		else
3344 			tx_ring->stall_watchdog = 0;
3345 
3346 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
3347 			result = B_TRUE;
3348 			break;
3349 		}
3350 	}
3351 
3352 	if (result) {
3353 		tx_ring->stall_watchdog = 0;
3354 		tx_ring->recycle_fail = 0;
3355 	}
3356 
3357 	return (result);
3358 }
3359 
3360 
3361 /*
3362  * is_valid_mac_addr - Check if the mac address is valid.
3363  */
3364 static boolean_t
3365 is_valid_mac_addr(uint8_t *mac_addr)
3366 {
3367 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
3368 	const uint8_t addr_test2[6] =
3369 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
3370 
3371 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
3372 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
3373 		return (B_FALSE);
3374 
3375 	return (B_TRUE);
3376 }
3377 
3378 static boolean_t
3379 ixgbe_find_mac_address(ixgbe_t *ixgbe)
3380 {
3381 #ifdef __sparc
3382 	struct ixgbe_hw *hw = &ixgbe->hw;
3383 	uchar_t *bytes;
3384 	struct ether_addr sysaddr;
3385 	uint_t nelts;
3386 	int err;
3387 	boolean_t found = B_FALSE;
3388 
3389 	/*
3390 	 * The "vendor's factory-set address" may already have
3391 	 * been extracted from the chip, but if the property
3392 	 * "local-mac-address" is set we use that instead.
3393 	 *
3394 	 * We check whether it looks like an array of 6
3395 	 * bytes (which it should, if OBP set it).  If we can't
3396 	 * make sense of it this way, we'll ignore it.
3397 	 */
3398 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip,
3399 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
3400 	if (err == DDI_PROP_SUCCESS) {
3401 		if (nelts == ETHERADDRL) {
3402 			while (nelts--)
3403 				hw->mac.addr[nelts] = bytes[nelts];
3404 			found = B_TRUE;
3405 		}
3406 		ddi_prop_free(bytes);
3407 	}
3408 
3409 	/*
3410 	 * Look up the OBP property "local-mac-address?". If the user has set
3411 	 * 'local-mac-address? = false', use "the system address" instead.
3412 	 */
3413 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip, 0,
3414 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
3415 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
3416 			if (localetheraddr(NULL, &sysaddr) != 0) {
3417 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
3418 				found = B_TRUE;
3419 			}
3420 		}
3421 		ddi_prop_free(bytes);
3422 	}
3423 
3424 	/*
3425 	 * Finally(!), if there's a valid "mac-address" property (created
3426 	 * if we netbooted from this interface), we must use this instead
3427 	 * of any of the above to ensure that the NFS/install server doesn't
3428 	 * get confused by the address changing as Solaris takes over!
3429 	 */
3430 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, ixgbe->dip,
3431 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
3432 	if (err == DDI_PROP_SUCCESS) {
3433 		if (nelts == ETHERADDRL) {
3434 			while (nelts--)
3435 				hw->mac.addr[nelts] = bytes[nelts];
3436 			found = B_TRUE;
3437 		}
3438 		ddi_prop_free(bytes);
3439 	}
3440 
3441 	if (found) {
3442 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
3443 		return (B_TRUE);
3444 	}
3445 #else
3446 	_NOTE(ARGUNUSED(ixgbe));
3447 #endif
3448 
3449 	return (B_TRUE);
3450 }
3451 
3452 #pragma inline(ixgbe_arm_watchdog_timer)
3453 static void
3454 ixgbe_arm_watchdog_timer(ixgbe_t *ixgbe)
3455 {
3456 	/*
3457 	 * Fire a watchdog timer
3458 	 */
3459 	ixgbe->watchdog_tid =
3460 	    timeout(ixgbe_local_timer,
3461 	    (void *)ixgbe, 1 * drv_usectohz(1000000));
3462 
3463 }
3464 
3465 /*
3466  * ixgbe_enable_watchdog_timer - Enable and start the driver watchdog timer.
3467  */
3468 void
3469 ixgbe_enable_watchdog_timer(ixgbe_t *ixgbe)
3470 {
3471 	mutex_enter(&ixgbe->watchdog_lock);
3472 
3473 	if (!ixgbe->watchdog_enable) {
3474 		ixgbe->watchdog_enable = B_TRUE;
3475 		ixgbe->watchdog_start = B_TRUE;
3476 		ixgbe_arm_watchdog_timer(ixgbe);
3477 	}
3478 
3479 	mutex_exit(&ixgbe->watchdog_lock);
3480 }
3481 
3482 /*
3483  * ixgbe_disable_watchdog_timer - Disable and stop the driver watchdog timer.
3484  */
3485 void
3486 ixgbe_disable_watchdog_timer(ixgbe_t *ixgbe)
3487 {
3488 	timeout_id_t tid;
3489 
3490 	mutex_enter(&ixgbe->watchdog_lock);
3491 
3492 	ixgbe->watchdog_enable = B_FALSE;
3493 	ixgbe->watchdog_start = B_FALSE;
3494 	tid = ixgbe->watchdog_tid;
3495 	ixgbe->watchdog_tid = 0;
3496 
3497 	mutex_exit(&ixgbe->watchdog_lock);
3498 
3499 	if (tid != 0)
3500 		(void) untimeout(tid);
3501 }
3502 
3503 /*
3504  * ixgbe_start_watchdog_timer - Start the driver watchdog timer.
3505  */
3506 void
3507 ixgbe_start_watchdog_timer(ixgbe_t *ixgbe)
3508 {
3509 	mutex_enter(&ixgbe->watchdog_lock);
3510 
3511 	if (ixgbe->watchdog_enable) {
3512 		if (!ixgbe->watchdog_start) {
3513 			ixgbe->watchdog_start = B_TRUE;
3514 			ixgbe_arm_watchdog_timer(ixgbe);
3515 		}
3516 	}
3517 
3518 	mutex_exit(&ixgbe->watchdog_lock);
3519 }
3520 
3521 /*
3522  * ixgbe_restart_watchdog_timer - Restart the driver watchdog timer.
3523  */
3524 static void
3525 ixgbe_restart_watchdog_timer(ixgbe_t *ixgbe)
3526 {
3527 	mutex_enter(&ixgbe->watchdog_lock);
3528 
3529 	if (ixgbe->watchdog_start)
3530 		ixgbe_arm_watchdog_timer(ixgbe);
3531 
3532 	mutex_exit(&ixgbe->watchdog_lock);
3533 }
3534 
3535 /*
3536  * ixgbe_stop_watchdog_timer - Stop the driver watchdog timer.
3537  */
3538 void
3539 ixgbe_stop_watchdog_timer(ixgbe_t *ixgbe)
3540 {
3541 	timeout_id_t tid;
3542 
3543 	mutex_enter(&ixgbe->watchdog_lock);
3544 
3545 	ixgbe->watchdog_start = B_FALSE;
3546 	tid = ixgbe->watchdog_tid;
3547 	ixgbe->watchdog_tid = 0;
3548 
3549 	mutex_exit(&ixgbe->watchdog_lock);
3550 
3551 	if (tid != 0)
3552 		(void) untimeout(tid);
3553 }
3554 
3555 /*
3556  * ixgbe_disable_adapter_interrupts - Disable all adapter interrupts.
3557  */
3558 static void
3559 ixgbe_disable_adapter_interrupts(ixgbe_t *ixgbe)
3560 {
3561 	struct ixgbe_hw *hw = &ixgbe->hw;
3562 
3563 	/*
3564 	 * mask all interrupts off
3565 	 */
3566 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, 0xffffffff);
3567 
3568 	/*
3569 	 * for MSI-X, also disable autoclear
3570 	 */
3571 	if (ixgbe->intr_type == DDI_INTR_TYPE_MSIX) {
3572 		IXGBE_WRITE_REG(hw, IXGBE_EIAC, 0x0);
3573 	}
3574 
3575 	IXGBE_WRITE_FLUSH(hw);
3576 }
3577 
3578 /*
3579  * ixgbe_enable_adapter_interrupts - Enable all hardware interrupts.
3580  */
3581 static void
3582 ixgbe_enable_adapter_interrupts(ixgbe_t *ixgbe)
3583 {
3584 	struct ixgbe_hw *hw = &ixgbe->hw;
3585 	uint32_t eiac, eiam;
3586 	uint32_t gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
3587 
3588 	/* interrupt types to enable */
3589 	ixgbe->eims = IXGBE_EIMS_ENABLE_MASK;	/* shared code default */
3590 	ixgbe->eims &= ~IXGBE_EIMS_TCP_TIMER;	/* minus tcp timer */
3591 	ixgbe->eims |= ixgbe->capab->other_intr; /* "other" interrupt types */
3592 
3593 	/* enable automask on "other" causes that this adapter can generate */
3594 	eiam = ixgbe->capab->other_intr;
3595 
3596 	/*
3597 	 * msi-x mode
3598 	 */
3599 	if (ixgbe->intr_type == DDI_INTR_TYPE_MSIX) {
3600 		/* enable autoclear but not on bits 29:20 */
3601 		eiac = (ixgbe->eims & ~IXGBE_OTHER_INTR);
3602 
3603 		/* general purpose interrupt enable */
3604 		gpie |= (IXGBE_GPIE_MSIX_MODE
3605 		    | IXGBE_GPIE_PBA_SUPPORT
3606 		    | IXGBE_GPIE_OCD
3607 		    | IXGBE_GPIE_EIAME);
3608 	/*
3609 	 * non-msi-x mode
3610 	 */
3611 	} else {
3612 
3613 		/* disable autoclear, leave gpie at default */
3614 		eiac = 0;
3615 
3616 		/*
3617 		 * General purpose interrupt enable.
3618 		 * For 82599, extended interrupt automask enable
3619 		 * only in MSI or MSI-X mode
3620 		 */
3621 		if ((hw->mac.type < ixgbe_mac_82599EB) ||
3622 		    (ixgbe->intr_type == DDI_INTR_TYPE_MSI)) {
3623 			gpie |= IXGBE_GPIE_EIAME;
3624 		}
3625 	}
3626 	/* Enable specific interrupts for 82599  */
3627 	if (hw->mac.type == ixgbe_mac_82599EB) {
3628 		gpie |= IXGBE_SDP2_GPIEN; /* pluggable optics intr */
3629 		gpie |= IXGBE_SDP1_GPIEN; /* LSC interrupt */
3630 	}
3631 	/* Enable RSC Dealy 8us for 82599  */
3632 	if (ixgbe->lro_enable) {
3633 		gpie |= (1 << IXGBE_GPIE_RSC_DELAY_SHIFT);
3634 	}
3635 	/* write to interrupt control registers */
3636 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
3637 	IXGBE_WRITE_REG(hw, IXGBE_EIAC, eiac);
3638 	IXGBE_WRITE_REG(hw, IXGBE_EIAM, eiam);
3639 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3640 	IXGBE_WRITE_FLUSH(hw);
3641 }
3642 
3643 /*
3644  * ixgbe_loopback_ioctl - Loopback support.
3645  */
3646 enum ioc_reply
3647 ixgbe_loopback_ioctl(ixgbe_t *ixgbe, struct iocblk *iocp, mblk_t *mp)
3648 {
3649 	lb_info_sz_t *lbsp;
3650 	lb_property_t *lbpp;
3651 	uint32_t *lbmp;
3652 	uint32_t size;
3653 	uint32_t value;
3654 
3655 	if (mp->b_cont == NULL)
3656 		return (IOC_INVAL);
3657 
3658 	switch (iocp->ioc_cmd) {
3659 	default:
3660 		return (IOC_INVAL);
3661 
3662 	case LB_GET_INFO_SIZE:
3663 		size = sizeof (lb_info_sz_t);
3664 		if (iocp->ioc_count != size)
3665 			return (IOC_INVAL);
3666 
3667 		value = sizeof (lb_normal);
3668 		value += sizeof (lb_mac);
3669 		value += sizeof (lb_external);
3670 
3671 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
3672 		*lbsp = value;
3673 		break;
3674 
3675 	case LB_GET_INFO:
3676 		value = sizeof (lb_normal);
3677 		value += sizeof (lb_mac);
3678 		value += sizeof (lb_external);
3679 
3680 		size = value;
3681 		if (iocp->ioc_count != size)
3682 			return (IOC_INVAL);
3683 
3684 		value = 0;
3685 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
3686 
3687 		lbpp[value++] = lb_normal;
3688 		lbpp[value++] = lb_mac;
3689 		lbpp[value++] = lb_external;
3690 		break;
3691 
3692 	case LB_GET_MODE:
3693 		size = sizeof (uint32_t);
3694 		if (iocp->ioc_count != size)
3695 			return (IOC_INVAL);
3696 
3697 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3698 		*lbmp = ixgbe->loopback_mode;
3699 		break;
3700 
3701 	case LB_SET_MODE:
3702 		size = 0;
3703 		if (iocp->ioc_count != sizeof (uint32_t))
3704 			return (IOC_INVAL);
3705 
3706 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3707 		if (!ixgbe_set_loopback_mode(ixgbe, *lbmp))
3708 			return (IOC_INVAL);
3709 		break;
3710 	}
3711 
3712 	iocp->ioc_count = size;
3713 	iocp->ioc_error = 0;
3714 
3715 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
3716 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3717 		return (IOC_INVAL);
3718 	}
3719 
3720 	return (IOC_REPLY);
3721 }
3722 
3723 /*
3724  * ixgbe_set_loopback_mode - Setup loopback based on the loopback mode.
3725  */
3726 static boolean_t
3727 ixgbe_set_loopback_mode(ixgbe_t *ixgbe, uint32_t mode)
3728 {
3729 	if (mode == ixgbe->loopback_mode)
3730 		return (B_TRUE);
3731 
3732 	ixgbe->loopback_mode = mode;
3733 
3734 	if (mode == IXGBE_LB_NONE) {
3735 		/*
3736 		 * Reset the chip
3737 		 */
3738 		(void) ixgbe_reset(ixgbe);
3739 		return (B_TRUE);
3740 	}
3741 
3742 	mutex_enter(&ixgbe->gen_lock);
3743 
3744 	switch (mode) {
3745 	default:
3746 		mutex_exit(&ixgbe->gen_lock);
3747 		return (B_FALSE);
3748 
3749 	case IXGBE_LB_EXTERNAL:
3750 		break;
3751 
3752 	case IXGBE_LB_INTERNAL_MAC:
3753 		ixgbe_set_internal_mac_loopback(ixgbe);
3754 		break;
3755 	}
3756 
3757 	mutex_exit(&ixgbe->gen_lock);
3758 
3759 	return (B_TRUE);
3760 }
3761 
3762 /*
3763  * ixgbe_set_internal_mac_loopback - Set the internal MAC loopback mode.
3764  */
3765 static void
3766 ixgbe_set_internal_mac_loopback(ixgbe_t *ixgbe)
3767 {
3768 	struct ixgbe_hw *hw;
3769 	uint32_t reg;
3770 	uint8_t atlas;
3771 
3772 	hw = &ixgbe->hw;
3773 
3774 	/*
3775 	 * Setup MAC loopback
3776 	 */
3777 	reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_HLREG0);
3778 	reg |= IXGBE_HLREG0_LPBK;
3779 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_HLREG0, reg);
3780 
3781 	reg = IXGBE_READ_REG(&ixgbe->hw, IXGBE_AUTOC);
3782 	reg &= ~IXGBE_AUTOC_LMS_MASK;
3783 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_AUTOC, reg);
3784 
3785 	/*
3786 	 * Disable Atlas Tx lanes to keep packets in loopback and not on wire
3787 	 */
3788 	if (hw->mac.type == ixgbe_mac_82598EB) {
3789 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_LPBK,
3790 		    &atlas);
3791 		atlas |= IXGBE_ATLAS_PDN_TX_REG_EN;
3792 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_LPBK,
3793 		    atlas);
3794 
3795 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_10G,
3796 		    &atlas);
3797 		atlas |= IXGBE_ATLAS_PDN_TX_10G_QL_ALL;
3798 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_10G,
3799 		    atlas);
3800 
3801 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_1G,
3802 		    &atlas);
3803 		atlas |= IXGBE_ATLAS_PDN_TX_1G_QL_ALL;
3804 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_1G,
3805 		    atlas);
3806 
3807 		(void) ixgbe_read_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_AN,
3808 		    &atlas);
3809 		atlas |= IXGBE_ATLAS_PDN_TX_AN_QL_ALL;
3810 		(void) ixgbe_write_analog_reg8(&ixgbe->hw, IXGBE_ATLAS_PDN_AN,
3811 		    atlas);
3812 	}
3813 }
3814 
3815 #pragma inline(ixgbe_intr_rx_work)
3816 /*
3817  * ixgbe_intr_rx_work - RX processing of ISR.
3818  */
3819 static void
3820 ixgbe_intr_rx_work(ixgbe_rx_ring_t *rx_ring)
3821 {
3822 	mblk_t *mp;
3823 
3824 	mutex_enter(&rx_ring->rx_lock);
3825 
3826 	mp = ixgbe_ring_rx(rx_ring, IXGBE_POLL_NULL);
3827 	mutex_exit(&rx_ring->rx_lock);
3828 
3829 	if (mp != NULL)
3830 		mac_rx_ring(rx_ring->ixgbe->mac_hdl, rx_ring->ring_handle, mp,
3831 		    rx_ring->ring_gen_num);
3832 }
3833 
3834 #pragma inline(ixgbe_intr_tx_work)
3835 /*
3836  * ixgbe_intr_tx_work - TX processing of ISR.
3837  */
3838 static void
3839 ixgbe_intr_tx_work(ixgbe_tx_ring_t *tx_ring)
3840 {
3841 	ixgbe_t *ixgbe = tx_ring->ixgbe;
3842 
3843 	/*
3844 	 * Recycle the tx descriptors
3845 	 */
3846 	tx_ring->tx_recycle(tx_ring);
3847 
3848 	/*
3849 	 * Schedule the re-transmit
3850 	 */
3851 	if (tx_ring->reschedule &&
3852 	    (tx_ring->tbd_free >= ixgbe->tx_resched_thresh)) {
3853 		tx_ring->reschedule = B_FALSE;
3854 		mac_tx_ring_update(tx_ring->ixgbe->mac_hdl,
3855 		    tx_ring->ring_handle);
3856 		IXGBE_DEBUG_STAT(tx_ring->stat_reschedule);
3857 	}
3858 }
3859 
3860 #pragma inline(ixgbe_intr_other_work)
3861 /*
3862  * ixgbe_intr_other_work - Process interrupt types other than tx/rx
3863  */
3864 static void
3865 ixgbe_intr_other_work(ixgbe_t *ixgbe, uint32_t eicr)
3866 {
3867 	struct ixgbe_hw *hw = &ixgbe->hw;
3868 
3869 	ASSERT(mutex_owned(&ixgbe->gen_lock));
3870 
3871 	/*
3872 	 * handle link status change
3873 	 */
3874 	if (eicr & IXGBE_EICR_LSC) {
3875 		ixgbe_driver_link_check(ixgbe);
3876 	}
3877 
3878 	/*
3879 	 * check for fan failure on adapters with fans
3880 	 */
3881 	if ((ixgbe->capab->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
3882 	    (eicr & IXGBE_EICR_GPI_SDP1)) {
3883 		if (hw->mac.type < ixgbe_mac_82599EB) {
3884 			ixgbe_log(ixgbe,
3885 			    "Fan has stopped, replace the adapter\n");
3886 
3887 			/* re-enable the interrupt, which was automasked */
3888 			ixgbe->eims |= IXGBE_EICR_GPI_SDP1;
3889 		}
3890 	}
3891 
3892 	/*
3893 	 * Do SFP check for 82599
3894 	 */
3895 	if (hw->mac.type == ixgbe_mac_82599EB) {
3896 		if ((ddi_taskq_dispatch(ixgbe->sfp_taskq,
3897 		    ixgbe_sfp_check, (void *)ixgbe,
3898 		    DDI_NOSLEEP)) != DDI_SUCCESS) {
3899 			ixgbe_log(ixgbe, "No memory available to dispatch "
3900 			    "taskq for SFP check");
3901 		}
3902 
3903 		/*
3904 		 * We need to fully re-check the link later.
3905 		 */
3906 		ixgbe->link_check_complete = B_FALSE;
3907 		ixgbe->link_check_hrtime = gethrtime() +
3908 		    (IXGBE_LINK_UP_TIME * 100000000ULL);
3909 	}
3910 }
3911 
3912 /*
3913  * ixgbe_intr_legacy - Interrupt handler for legacy interrupts.
3914  */
3915 static uint_t
3916 ixgbe_intr_legacy(void *arg1, void *arg2)
3917 {
3918 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
3919 	struct ixgbe_hw *hw = &ixgbe->hw;
3920 	ixgbe_tx_ring_t *tx_ring;
3921 	ixgbe_rx_ring_t *rx_ring;
3922 	uint32_t eicr;
3923 	mblk_t *mp;
3924 	boolean_t tx_reschedule;
3925 	uint_t result;
3926 
3927 	_NOTE(ARGUNUSED(arg2));
3928 
3929 	mutex_enter(&ixgbe->gen_lock);
3930 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
3931 		mutex_exit(&ixgbe->gen_lock);
3932 		return (DDI_INTR_UNCLAIMED);
3933 	}
3934 
3935 	mp = NULL;
3936 	tx_reschedule = B_FALSE;
3937 
3938 	/*
3939 	 * Any bit set in eicr: claim this interrupt
3940 	 */
3941 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3942 
3943 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
3944 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
3945 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
3946 		return (DDI_INTR_CLAIMED);
3947 	}
3948 
3949 	if (eicr) {
3950 		/*
3951 		 * For legacy interrupt, we have only one interrupt,
3952 		 * so we have only one rx ring and one tx ring enabled.
3953 		 */
3954 		ASSERT(ixgbe->num_rx_rings == 1);
3955 		ASSERT(ixgbe->num_tx_rings == 1);
3956 
3957 		/*
3958 		 * For legacy interrupt, rx rings[0] will use RTxQ[0].
3959 		 */
3960 		if (eicr & 0x1) {
3961 			ixgbe->eimc |= IXGBE_EICR_RTX_QUEUE;
3962 			IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
3963 			ixgbe->eims |= IXGBE_EICR_RTX_QUEUE;
3964 			/*
3965 			 * Clean the rx descriptors
3966 			 */
3967 			rx_ring = &ixgbe->rx_rings[0];
3968 			mp = ixgbe_ring_rx(rx_ring, IXGBE_POLL_NULL);
3969 		}
3970 
3971 		/*
3972 		 * For legacy interrupt, tx rings[0] will use RTxQ[1].
3973 		 */
3974 		if (eicr & 0x2) {
3975 			/*
3976 			 * Recycle the tx descriptors
3977 			 */
3978 			tx_ring = &ixgbe->tx_rings[0];
3979 			tx_ring->tx_recycle(tx_ring);
3980 
3981 			/*
3982 			 * Schedule the re-transmit
3983 			 */
3984 			tx_reschedule = (tx_ring->reschedule &&
3985 			    (tx_ring->tbd_free >= ixgbe->tx_resched_thresh));
3986 		}
3987 
3988 		/* any interrupt type other than tx/rx */
3989 		if (eicr & ixgbe->capab->other_intr) {
3990 			if (hw->mac.type < ixgbe_mac_82599EB) {
3991 				ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
3992 			}
3993 			if (hw->mac.type == ixgbe_mac_82599EB) {
3994 				ixgbe->eimc = IXGBE_82599_OTHER_INTR;
3995 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
3996 			}
3997 			ixgbe_intr_other_work(ixgbe, eicr);
3998 			ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
3999 		}
4000 
4001 		mutex_exit(&ixgbe->gen_lock);
4002 
4003 		result = DDI_INTR_CLAIMED;
4004 	} else {
4005 		mutex_exit(&ixgbe->gen_lock);
4006 
4007 		/*
4008 		 * No interrupt cause bits set: don't claim this interrupt.
4009 		 */
4010 		result = DDI_INTR_UNCLAIMED;
4011 	}
4012 
4013 	/* re-enable the interrupts which were automasked */
4014 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4015 
4016 	/*
4017 	 * Do the following work outside of the gen_lock
4018 	 */
4019 	if (mp != NULL) {
4020 		mac_rx_ring(rx_ring->ixgbe->mac_hdl, rx_ring->ring_handle, mp,
4021 		    rx_ring->ring_gen_num);
4022 	}
4023 
4024 	if (tx_reschedule)  {
4025 		tx_ring->reschedule = B_FALSE;
4026 		mac_tx_ring_update(ixgbe->mac_hdl, tx_ring->ring_handle);
4027 		IXGBE_DEBUG_STAT(tx_ring->stat_reschedule);
4028 	}
4029 
4030 	return (result);
4031 }
4032 
4033 /*
4034  * ixgbe_intr_msi - Interrupt handler for MSI.
4035  */
4036 static uint_t
4037 ixgbe_intr_msi(void *arg1, void *arg2)
4038 {
4039 	ixgbe_t *ixgbe = (ixgbe_t *)arg1;
4040 	struct ixgbe_hw *hw = &ixgbe->hw;
4041 	uint32_t eicr;
4042 
4043 	_NOTE(ARGUNUSED(arg2));
4044 
4045 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4046 
4047 	if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) != DDI_FM_OK) {
4048 		ddi_fm_service_impact(ixgbe->dip, DDI_SERVICE_DEGRADED);
4049 		atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4050 		return (DDI_INTR_CLAIMED);
4051 	}
4052 
4053 	/*
4054 	 * For MSI interrupt, we have only one vector,
4055 	 * so we have only one rx ring and one tx ring enabled.
4056 	 */
4057 	ASSERT(ixgbe->num_rx_rings == 1);
4058 	ASSERT(ixgbe->num_tx_rings == 1);
4059 
4060 	/*
4061 	 * For MSI interrupt, rx rings[0] will use RTxQ[0].
4062 	 */
4063 	if (eicr & 0x1) {
4064 		ixgbe_intr_rx_work(&ixgbe->rx_rings[0]);
4065 	}
4066 
4067 	/*
4068 	 * For MSI interrupt, tx rings[0] will use RTxQ[1].
4069 	 */
4070 	if (eicr & 0x2) {
4071 		ixgbe_intr_tx_work(&ixgbe->tx_rings[0]);
4072 	}
4073 
4074 	/* any interrupt type other than tx/rx */
4075 	if (eicr & ixgbe->capab->other_intr) {
4076 		mutex_enter(&ixgbe->gen_lock);
4077 		if (hw->mac.type < ixgbe_mac_82599EB) {
4078 			ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4079 		}
4080 		if (hw->mac.type == ixgbe_mac_82599EB) {
4081 			ixgbe->eimc = IXGBE_82599_OTHER_INTR;
4082 			IXGBE_WRITE_REG(hw, IXGBE_EIMC, ixgbe->eimc);
4083 		}
4084 		ixgbe_intr_other_work(ixgbe, eicr);
4085 		ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4086 		mutex_exit(&ixgbe->gen_lock);
4087 	}
4088 
4089 	/* re-enable the interrupts which were automasked */
4090 	IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4091 
4092 	return (DDI_INTR_CLAIMED);
4093 }
4094 
4095 /*
4096  * ixgbe_intr_msix - Interrupt handler for MSI-X.
4097  */
4098 static uint_t
4099 ixgbe_intr_msix(void *arg1, void *arg2)
4100 {
4101 	ixgbe_intr_vector_t *vect = (ixgbe_intr_vector_t *)arg1;
4102 	ixgbe_t *ixgbe = vect->ixgbe;
4103 	struct ixgbe_hw *hw = &ixgbe->hw;
4104 	uint32_t eicr;
4105 	int r_idx = 0;
4106 
4107 	_NOTE(ARGUNUSED(arg2));
4108 
4109 	/*
4110 	 * Clean each rx ring that has its bit set in the map
4111 	 */
4112 	r_idx = bt_getlowbit(vect->rx_map, 0, (ixgbe->num_rx_rings - 1));
4113 	while (r_idx >= 0) {
4114 		ixgbe_intr_rx_work(&ixgbe->rx_rings[r_idx]);
4115 		r_idx = bt_getlowbit(vect->rx_map, (r_idx + 1),
4116 		    (ixgbe->num_rx_rings - 1));
4117 	}
4118 
4119 	/*
4120 	 * Clean each tx ring that has its bit set in the map
4121 	 */
4122 	r_idx = bt_getlowbit(vect->tx_map, 0, (ixgbe->num_tx_rings - 1));
4123 	while (r_idx >= 0) {
4124 		ixgbe_intr_tx_work(&ixgbe->tx_rings[r_idx]);
4125 		r_idx = bt_getlowbit(vect->tx_map, (r_idx + 1),
4126 		    (ixgbe->num_tx_rings - 1));
4127 	}
4128 
4129 
4130 	/*
4131 	 * Clean other interrupt (link change) that has its bit set in the map
4132 	 */
4133 	if (BT_TEST(vect->other_map, 0) == 1) {
4134 		eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
4135 
4136 		if (ixgbe_check_acc_handle(ixgbe->osdep.reg_handle) !=
4137 		    DDI_FM_OK) {
4138 			ddi_fm_service_impact(ixgbe->dip,
4139 			    DDI_SERVICE_DEGRADED);
4140 			atomic_or_32(&ixgbe->ixgbe_state, IXGBE_ERROR);
4141 			return (DDI_INTR_CLAIMED);
4142 		}
4143 
4144 		/*
4145 		 * Need check cause bits and only other causes will
4146 		 * be processed
4147 		 */
4148 		/* any interrupt type other than tx/rx */
4149 		if (eicr & ixgbe->capab->other_intr) {
4150 			if (hw->mac.type < ixgbe_mac_82599EB) {
4151 				mutex_enter(&ixgbe->gen_lock);
4152 				ixgbe->eims &= ~(eicr & IXGBE_OTHER_INTR);
4153 				ixgbe_intr_other_work(ixgbe, eicr);
4154 				mutex_exit(&ixgbe->gen_lock);
4155 			} else {
4156 				if (hw->mac.type == ixgbe_mac_82599EB) {
4157 					mutex_enter(&ixgbe->gen_lock);
4158 					ixgbe->eims |= IXGBE_EICR_RTX_QUEUE;
4159 					ixgbe_intr_other_work(ixgbe, eicr);
4160 					mutex_exit(&ixgbe->gen_lock);
4161 				}
4162 			}
4163 		}
4164 
4165 		/* re-enable the interrupts which were automasked */
4166 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, ixgbe->eims);
4167 	}
4168 
4169 	return (DDI_INTR_CLAIMED);
4170 }
4171 
4172 /*
4173  * ixgbe_alloc_intrs - Allocate interrupts for the driver.
4174  *
4175  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
4176  * if not successful, try Legacy.
4177  * ixgbe->intr_force can be used to force sequence to start with
4178  * any of the 3 types.
4179  * If MSI-X is not used, number of tx/rx rings is forced to 1.
4180  */
4181 static int
4182 ixgbe_alloc_intrs(ixgbe_t *ixgbe)
4183 {
4184 	dev_info_t *devinfo;
4185 	int intr_types;
4186 	int rc;
4187 
4188 	devinfo = ixgbe->dip;
4189 
4190 	/*
4191 	 * Get supported interrupt types
4192 	 */
4193 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
4194 
4195 	if (rc != DDI_SUCCESS) {
4196 		ixgbe_log(ixgbe,
4197 		    "Get supported interrupt types failed: %d", rc);
4198 		return (IXGBE_FAILURE);
4199 	}
4200 	IXGBE_DEBUGLOG_1(ixgbe, "Supported interrupt types: %x", intr_types);
4201 
4202 	ixgbe->intr_type = 0;
4203 
4204 	/*
4205 	 * Install MSI-X interrupts
4206 	 */
4207 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
4208 	    (ixgbe->intr_force <= IXGBE_INTR_MSIX)) {
4209 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_MSIX);
4210 		if (rc == IXGBE_SUCCESS)
4211 			return (IXGBE_SUCCESS);
4212 
4213 		ixgbe_log(ixgbe,
4214 		    "Allocate MSI-X failed, trying MSI interrupts...");
4215 	}
4216 
4217 	/*
4218 	 * MSI-X not used, force rings and groups to 1
4219 	 */
4220 	ixgbe->num_rx_rings = 1;
4221 	ixgbe->num_rx_groups = 1;
4222 	ixgbe->num_tx_rings = 1;
4223 	ixgbe->classify_mode = IXGBE_CLASSIFY_NONE;
4224 	ixgbe_log(ixgbe,
4225 	    "MSI-X not used, force rings and groups number to 1");
4226 
4227 	/*
4228 	 * Install MSI interrupts
4229 	 */
4230 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
4231 	    (ixgbe->intr_force <= IXGBE_INTR_MSI)) {
4232 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_MSI);
4233 		if (rc == IXGBE_SUCCESS)
4234 			return (IXGBE_SUCCESS);
4235 
4236 		ixgbe_log(ixgbe,
4237 		    "Allocate MSI failed, trying Legacy interrupts...");
4238 	}
4239 
4240 	/*
4241 	 * Install legacy interrupts
4242 	 */
4243 	if (intr_types & DDI_INTR_TYPE_FIXED) {
4244 		rc = ixgbe_alloc_intr_handles(ixgbe, DDI_INTR_TYPE_FIXED);
4245 		if (rc == IXGBE_SUCCESS)
4246 			return (IXGBE_SUCCESS);
4247 
4248 		ixgbe_log(ixgbe,
4249 		    "Allocate Legacy interrupts failed");
4250 	}
4251 
4252 	/*
4253 	 * If none of the 3 types succeeded, return failure
4254 	 */
4255 	return (IXGBE_FAILURE);
4256 }
4257 
4258 /*
4259  * ixgbe_alloc_intr_handles - Allocate interrupt handles.
4260  *
4261  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
4262  * if fewer than 2 handles are available, return failure.
4263  * Upon success, this maps the vectors to rx and tx rings for
4264  * interrupts.
4265  */
4266 static int
4267 ixgbe_alloc_intr_handles(ixgbe_t *ixgbe, int intr_type)
4268 {
4269 	dev_info_t *devinfo;
4270 	int request, count, actual;
4271 	int minimum;
4272 	int rc;
4273 	uint32_t ring_per_group;
4274 
4275 	devinfo = ixgbe->dip;
4276 
4277 	switch (intr_type) {
4278 	case DDI_INTR_TYPE_FIXED:
4279 		request = 1;	/* Request 1 legacy interrupt handle */
4280 		minimum = 1;
4281 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: legacy");
4282 		break;
4283 
4284 	case DDI_INTR_TYPE_MSI:
4285 		request = 1;	/* Request 1 MSI interrupt handle */
4286 		minimum = 1;
4287 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: MSI");
4288 		break;
4289 
4290 	case DDI_INTR_TYPE_MSIX:
4291 		/*
4292 		 * Best number of vectors for the adapter is
4293 		 * (# rx rings + # tx rings), however we will
4294 		 * limit the request number.
4295 		 */
4296 		request = min(16, ixgbe->num_rx_rings + ixgbe->num_tx_rings);
4297 		if (request > ixgbe->capab->max_ring_vect)
4298 			request = ixgbe->capab->max_ring_vect;
4299 		minimum = 1;
4300 		IXGBE_DEBUGLOG_0(ixgbe, "interrupt type: MSI-X");
4301 		break;
4302 
4303 	default:
4304 		ixgbe_log(ixgbe,
4305 		    "invalid call to ixgbe_alloc_intr_handles(): %d\n",
4306 		    intr_type);
4307 		return (IXGBE_FAILURE);
4308 	}
4309 	IXGBE_DEBUGLOG_2(ixgbe, "interrupt handles requested: %d  minimum: %d",
4310 	    request, minimum);
4311 
4312 	/*
4313 	 * Get number of supported interrupts
4314 	 */
4315 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
4316 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
4317 		ixgbe_log(ixgbe,
4318 		    "Get interrupt number failed. Return: %d, count: %d",
4319 		    rc, count);
4320 		return (IXGBE_FAILURE);
4321 	}
4322 	IXGBE_DEBUGLOG_1(ixgbe, "interrupts supported: %d", count);
4323 
4324 	actual = 0;
4325 	ixgbe->intr_cnt = 0;
4326 	ixgbe->intr_cnt_max = 0;
4327 	ixgbe->intr_cnt_min = 0;
4328 
4329 	/*
4330 	 * Allocate an array of interrupt handles
4331 	 */
4332 	ixgbe->intr_size = request * sizeof (ddi_intr_handle_t);
4333 	ixgbe->htable = kmem_alloc(ixgbe->intr_size, KM_SLEEP);
4334 
4335 	rc = ddi_intr_alloc(devinfo, ixgbe->htable, intr_type, 0,
4336 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
4337 	if (rc != DDI_SUCCESS) {
4338 		ixgbe_log(ixgbe, "Allocate interrupts failed. "
4339 		    "return: %d, request: %d, actual: %d",
4340 		    rc, request, actual);
4341 		goto alloc_handle_fail;
4342 	}
4343 	IXGBE_DEBUGLOG_1(ixgbe, "interrupts actually allocated: %d", actual);
4344 
4345 	/*
4346 	 * upper/lower limit of interrupts
4347 	 */
4348 	ixgbe->intr_cnt = actual;
4349 	ixgbe->intr_cnt_max = request;
4350 	ixgbe->intr_cnt_min = minimum;
4351 
4352 	/*
4353 	 * rss number per group should not exceed the rx interrupt number,
4354 	 * else need to adjust rx ring number.
4355 	 */
4356 	ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
4357 	ASSERT((ixgbe->num_rx_rings % ixgbe->num_rx_groups) == 0);
4358 	if (min(actual, ixgbe->num_rx_rings) < ring_per_group) {
4359 		ixgbe->num_rx_rings = ixgbe->num_rx_groups *
4360 		    min(actual, ixgbe->num_rx_rings);
4361 		ixgbe_setup_vmdq_rss_conf(ixgbe);
4362 	}
4363 
4364 	/*
4365 	 * Now we know the actual number of vectors.  Here we map the vector
4366 	 * to other, rx rings and tx ring.
4367 	 */
4368 	if (actual < minimum) {
4369 		ixgbe_log(ixgbe, "Insufficient interrupt handles available: %d",
4370 		    actual);
4371 		goto alloc_handle_fail;
4372 	}
4373 
4374 	/*
4375 	 * Get priority for first vector, assume remaining are all the same
4376 	 */
4377 	rc = ddi_intr_get_pri(ixgbe->htable[0], &ixgbe->intr_pri);
4378 	if (rc != DDI_SUCCESS) {
4379 		ixgbe_log(ixgbe,
4380 		    "Get interrupt priority failed: %d", rc);
4381 		goto alloc_handle_fail;
4382 	}
4383 
4384 	rc = ddi_intr_get_cap(ixgbe->htable[0], &ixgbe->intr_cap);
4385 	if (rc != DDI_SUCCESS) {
4386 		ixgbe_log(ixgbe,
4387 		    "Get interrupt cap failed: %d", rc);
4388 		goto alloc_handle_fail;
4389 	}
4390 
4391 	ixgbe->intr_type = intr_type;
4392 
4393 	return (IXGBE_SUCCESS);
4394 
4395 alloc_handle_fail:
4396 	ixgbe_rem_intrs(ixgbe);
4397 
4398 	return (IXGBE_FAILURE);
4399 }
4400 
4401 /*
4402  * ixgbe_add_intr_handlers - Add interrupt handlers based on the interrupt type.
4403  *
4404  * Before adding the interrupt handlers, the interrupt vectors have
4405  * been allocated, and the rx/tx rings have also been allocated.
4406  */
4407 static int
4408 ixgbe_add_intr_handlers(ixgbe_t *ixgbe)
4409 {
4410 	int vector = 0;
4411 	int rc;
4412 
4413 	switch (ixgbe->intr_type) {
4414 	case DDI_INTR_TYPE_MSIX:
4415 		/*
4416 		 * Add interrupt handler for all vectors
4417 		 */
4418 		for (vector = 0; vector < ixgbe->intr_cnt; vector++) {
4419 			/*
4420 			 * install pointer to vect_map[vector]
4421 			 */
4422 			rc = ddi_intr_add_handler(ixgbe->htable[vector],
4423 			    (ddi_intr_handler_t *)ixgbe_intr_msix,
4424 			    (void *)&ixgbe->vect_map[vector], NULL);
4425 
4426 			if (rc != DDI_SUCCESS) {
4427 				ixgbe_log(ixgbe,
4428 				    "Add rx interrupt handler failed. "
4429 				    "return: %d, vector: %d", rc, vector);
4430 				for (vector--; vector >= 0; vector--) {
4431 					(void) ddi_intr_remove_handler(
4432 					    ixgbe->htable[vector]);
4433 				}
4434 				return (IXGBE_FAILURE);
4435 			}
4436 		}
4437 
4438 		break;
4439 
4440 	case DDI_INTR_TYPE_MSI:
4441 		/*
4442 		 * Add interrupt handlers for the only vector
4443 		 */
4444 		rc = ddi_intr_add_handler(ixgbe->htable[vector],
4445 		    (ddi_intr_handler_t *)ixgbe_intr_msi,
4446 		    (void *)ixgbe, NULL);
4447 
4448 		if (rc != DDI_SUCCESS) {
4449 			ixgbe_log(ixgbe,
4450 			    "Add MSI interrupt handler failed: %d", rc);
4451 			return (IXGBE_FAILURE);
4452 		}
4453 
4454 		break;
4455 
4456 	case DDI_INTR_TYPE_FIXED:
4457 		/*
4458 		 * Add interrupt handlers for the only vector
4459 		 */
4460 		rc = ddi_intr_add_handler(ixgbe->htable[vector],
4461 		    (ddi_intr_handler_t *)ixgbe_intr_legacy,
4462 		    (void *)ixgbe, NULL);
4463 
4464 		if (rc != DDI_SUCCESS) {
4465 			ixgbe_log(ixgbe,
4466 			    "Add legacy interrupt handler failed: %d", rc);
4467 			return (IXGBE_FAILURE);
4468 		}
4469 
4470 		break;
4471 
4472 	default:
4473 		return (IXGBE_FAILURE);
4474 	}
4475 
4476 	return (IXGBE_SUCCESS);
4477 }
4478 
4479 #pragma inline(ixgbe_map_rxring_to_vector)
4480 /*
4481  * ixgbe_map_rxring_to_vector - Map given rx ring to given interrupt vector.
4482  */
4483 static void
4484 ixgbe_map_rxring_to_vector(ixgbe_t *ixgbe, int r_idx, int v_idx)
4485 {
4486 	/*
4487 	 * Set bit in map
4488 	 */
4489 	BT_SET(ixgbe->vect_map[v_idx].rx_map, r_idx);
4490 
4491 	/*
4492 	 * Count bits set
4493 	 */
4494 	ixgbe->vect_map[v_idx].rxr_cnt++;
4495 
4496 	/*
4497 	 * Remember bit position
4498 	 */
4499 	ixgbe->rx_rings[r_idx].intr_vector = v_idx;
4500 	ixgbe->rx_rings[r_idx].vect_bit = 1 << v_idx;
4501 }
4502 
4503 #pragma inline(ixgbe_map_txring_to_vector)
4504 /*
4505  * ixgbe_map_txring_to_vector - Map given tx ring to given interrupt vector.
4506  */
4507 static void
4508 ixgbe_map_txring_to_vector(ixgbe_t *ixgbe, int t_idx, int v_idx)
4509 {
4510 	/*
4511 	 * Set bit in map
4512 	 */
4513 	BT_SET(ixgbe->vect_map[v_idx].tx_map, t_idx);
4514 
4515 	/*
4516 	 * Count bits set
4517 	 */
4518 	ixgbe->vect_map[v_idx].txr_cnt++;
4519 
4520 	/*
4521 	 * Remember bit position
4522 	 */
4523 	ixgbe->tx_rings[t_idx].intr_vector = v_idx;
4524 	ixgbe->tx_rings[t_idx].vect_bit = 1 << v_idx;
4525 }
4526 
4527 /*
4528  * ixgbe_setup_ivar - Set the given entry in the given interrupt vector
4529  * allocation register (IVAR).
4530  * cause:
4531  *   -1 : other cause
4532  *    0 : rx
4533  *    1 : tx
4534  */
4535 static void
4536 ixgbe_setup_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, uint8_t msix_vector,
4537     int8_t cause)
4538 {
4539 	struct ixgbe_hw *hw = &ixgbe->hw;
4540 	u32 ivar, index;
4541 
4542 	switch (hw->mac.type) {
4543 	case ixgbe_mac_82598EB:
4544 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
4545 		if (cause == -1) {
4546 			cause = 0;
4547 		}
4548 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
4549 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
4550 		ivar &= ~(0xFF << (8 * (intr_alloc_entry & 0x3)));
4551 		ivar |= (msix_vector << (8 * (intr_alloc_entry & 0x3)));
4552 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
4553 		break;
4554 	case ixgbe_mac_82599EB:
4555 		if (cause == -1) {
4556 			/* other causes */
4557 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
4558 			index = (intr_alloc_entry & 1) * 8;
4559 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
4560 			ivar &= ~(0xFF << index);
4561 			ivar |= (msix_vector << index);
4562 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
4563 		} else {
4564 			/* tx or rx causes */
4565 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
4566 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
4567 			ivar = IXGBE_READ_REG(hw,
4568 			    IXGBE_IVAR(intr_alloc_entry >> 1));
4569 			ivar &= ~(0xFF << index);
4570 			ivar |= (msix_vector << index);
4571 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
4572 			    ivar);
4573 		}
4574 		break;
4575 	default:
4576 		break;
4577 	}
4578 }
4579 
4580 /*
4581  * ixgbe_enable_ivar - Enable the given entry by setting the VAL bit of
4582  * given interrupt vector allocation register (IVAR).
4583  * cause:
4584  *   -1 : other cause
4585  *    0 : rx
4586  *    1 : tx
4587  */
4588 static void
4589 ixgbe_enable_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, int8_t cause)
4590 {
4591 	struct ixgbe_hw *hw = &ixgbe->hw;
4592 	u32 ivar, index;
4593 
4594 	switch (hw->mac.type) {
4595 	case ixgbe_mac_82598EB:
4596 		if (cause == -1) {
4597 			cause = 0;
4598 		}
4599 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
4600 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
4601 		ivar |= (IXGBE_IVAR_ALLOC_VAL << (8 *
4602 		    (intr_alloc_entry & 0x3)));
4603 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
4604 		break;
4605 	case ixgbe_mac_82599EB:
4606 		if (cause == -1) {
4607 			/* other causes */
4608 			index = (intr_alloc_entry & 1) * 8;
4609 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
4610 			ivar |= (IXGBE_IVAR_ALLOC_VAL << index);
4611 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
4612 		} else {
4613 			/* tx or rx causes */
4614 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
4615 			ivar = IXGBE_READ_REG(hw,
4616 			    IXGBE_IVAR(intr_alloc_entry >> 1));
4617 			ivar |= (IXGBE_IVAR_ALLOC_VAL << index);
4618 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
4619 			    ivar);
4620 		}
4621 		break;
4622 	default:
4623 		break;
4624 	}
4625 }
4626 
4627 /*
4628  * ixgbe_disable_ivar - Disble the given entry by clearing the VAL bit of
4629  * given interrupt vector allocation register (IVAR).
4630  * cause:
4631  *   -1 : other cause
4632  *    0 : rx
4633  *    1 : tx
4634  */
4635 static void
4636 ixgbe_disable_ivar(ixgbe_t *ixgbe, uint16_t intr_alloc_entry, int8_t cause)
4637 {
4638 	struct ixgbe_hw *hw = &ixgbe->hw;
4639 	u32 ivar, index;
4640 
4641 	switch (hw->mac.type) {
4642 	case ixgbe_mac_82598EB:
4643 		if (cause == -1) {
4644 			cause = 0;
4645 		}
4646 		index = (((cause * 64) + intr_alloc_entry) >> 2) & 0x1F;
4647 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
4648 		ivar &= ~(IXGBE_IVAR_ALLOC_VAL<< (8 *
4649 		    (intr_alloc_entry & 0x3)));
4650 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
4651 		break;
4652 	case ixgbe_mac_82599EB:
4653 		if (cause == -1) {
4654 			/* other causes */
4655 			index = (intr_alloc_entry & 1) * 8;
4656 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
4657 			ivar &= ~(IXGBE_IVAR_ALLOC_VAL << index);
4658 			IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
4659 		} else {
4660 			/* tx or rx causes */
4661 			index = ((16 * (intr_alloc_entry & 1)) + (8 * cause));
4662 			ivar = IXGBE_READ_REG(hw,
4663 			    IXGBE_IVAR(intr_alloc_entry >> 1));
4664 			ivar &= ~(IXGBE_IVAR_ALLOC_VAL << index);
4665 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(intr_alloc_entry >> 1),
4666 			    ivar);
4667 		}
4668 		break;
4669 	default:
4670 		break;
4671 	}
4672 }
4673 
4674 /*
4675  * Convert the rx ring index driver maintained to the rx ring index
4676  * in h/w.
4677  */
4678 static uint32_t
4679 ixgbe_get_hw_rx_index(ixgbe_t *ixgbe, uint32_t sw_rx_index)
4680 {
4681 
4682 	struct ixgbe_hw *hw = &ixgbe->hw;
4683 	uint32_t rx_ring_per_group, hw_rx_index;
4684 
4685 	if (ixgbe->classify_mode == IXGBE_CLASSIFY_RSS ||
4686 	    ixgbe->classify_mode == IXGBE_CLASSIFY_NONE) {
4687 		return (sw_rx_index);
4688 	} else if (ixgbe->classify_mode == IXGBE_CLASSIFY_VMDQ) {
4689 		if (hw->mac.type == ixgbe_mac_82598EB) {
4690 			return (sw_rx_index);
4691 		} else if (hw->mac.type == ixgbe_mac_82599EB) {
4692 			return (sw_rx_index * 2);
4693 		}
4694 	} else if (ixgbe->classify_mode == IXGBE_CLASSIFY_VMDQ_RSS) {
4695 		rx_ring_per_group = ixgbe->num_rx_rings / ixgbe->num_rx_groups;
4696 
4697 		if (hw->mac.type == ixgbe_mac_82598EB) {
4698 			hw_rx_index = (sw_rx_index / rx_ring_per_group) *
4699 			    16 + (sw_rx_index % rx_ring_per_group);
4700 			return (hw_rx_index);
4701 		} else if (hw->mac.type == ixgbe_mac_82599EB) {
4702 			if (ixgbe->num_rx_groups > 32) {
4703 				hw_rx_index = (sw_rx_index /
4704 				    rx_ring_per_group) * 2 +
4705 				    (sw_rx_index % rx_ring_per_group);
4706 			} else {
4707 				hw_rx_index = (sw_rx_index /
4708 				    rx_ring_per_group) * 4 +
4709 				    (sw_rx_index % rx_ring_per_group);
4710 			}
4711 			return (hw_rx_index);
4712 		}
4713 	}
4714 
4715 	/*
4716 	 * Should never reach. Just to make compiler happy.
4717 	 */
4718 	return (sw_rx_index);
4719 }
4720 
4721 /*
4722  * ixgbe_map_intrs_to_vectors - Map different interrupts to MSI-X vectors.
4723  *
4724  * For MSI-X, here will map rx interrupt, tx interrupt and other interrupt
4725  * to vector[0 - (intr_cnt -1)].
4726  */
4727 static int
4728 ixgbe_map_intrs_to_vectors(ixgbe_t *ixgbe)
4729 {
4730 	int i, vector = 0;
4731 
4732 	/* initialize vector map */
4733 	bzero(&ixgbe->vect_map, sizeof (ixgbe->vect_map));
4734 	for (i = 0; i < ixgbe->intr_cnt; i++) {
4735 		ixgbe->vect_map[i].ixgbe = ixgbe;
4736 	}
4737 
4738 	/*
4739 	 * non-MSI-X case is very simple: rx rings[0] on RTxQ[0],
4740 	 * tx rings[0] on RTxQ[1].
4741 	 */
4742 	if (ixgbe->intr_type != DDI_INTR_TYPE_MSIX) {
4743 		ixgbe_map_rxring_to_vector(ixgbe, 0, 0);
4744 		ixgbe_map_txring_to_vector(ixgbe, 0, 1);
4745 		return (IXGBE_SUCCESS);
4746 	}
4747 
4748 	/*
4749 	 * Interrupts/vectors mapping for MSI-X
4750 	 */
4751 
4752 	/*
4753 	 * Map other interrupt to vector 0,
4754 	 * Set bit in map and count the bits set.
4755 	 */
4756 	BT_SET(ixgbe->vect_map[vector].other_map, 0);
4757 	ixgbe->vect_map[vector].other_cnt++;
4758 
4759 	/*
4760 	 * Map rx ring interrupts to vectors
4761 	 */
4762 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
4763 		ixgbe_map_rxring_to_vector(ixgbe, i, vector);
4764 		vector = (vector +1) % ixgbe->intr_cnt;
4765 	}
4766 
4767 	/*
4768 	 * Map tx ring interrupts to vectors
4769 	 */
4770 	for (i = 0; i < ixgbe->num_tx_rings; i++) {
4771 		ixgbe_map_txring_to_vector(ixgbe, i, vector);
4772 		vector = (vector +1) % ixgbe->intr_cnt;
4773 	}
4774 
4775 	return (IXGBE_SUCCESS);
4776 }
4777 
4778 /*
4779  * ixgbe_setup_adapter_vector - Setup the adapter interrupt vector(s).
4780  *
4781  * This relies on ring/vector mapping already set up in the
4782  * vect_map[] structures
4783  */
4784 static void
4785 ixgbe_setup_adapter_vector(ixgbe_t *ixgbe)
4786 {
4787 	struct ixgbe_hw *hw = &ixgbe->hw;
4788 	ixgbe_intr_vector_t *vect;	/* vector bitmap */
4789 	int r_idx;	/* ring index */
4790 	int v_idx;	/* vector index */
4791 	uint32_t hw_index;
4792 
4793 	/*
4794 	 * Clear any previous entries
4795 	 */
4796 	switch (hw->mac.type) {
4797 	case ixgbe_mac_82598EB:
4798 		for (v_idx = 0; v_idx < 25; v_idx++)
4799 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(v_idx), 0);
4800 
4801 		break;
4802 	case ixgbe_mac_82599EB:
4803 		for (v_idx = 0; v_idx < 64; v_idx++)
4804 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(v_idx), 0);
4805 		IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, 0);
4806 
4807 		break;
4808 	default:
4809 		break;
4810 	}
4811 
4812 	/*
4813 	 * For non MSI-X interrupt, rx rings[0] will use RTxQ[0], and
4814 	 * tx rings[0] will use RTxQ[1].
4815 	 */
4816 	if (ixgbe->intr_type != DDI_INTR_TYPE_MSIX) {
4817 		ixgbe_setup_ivar(ixgbe, 0, 0, 0);
4818 		ixgbe_setup_ivar(ixgbe, 0, 1, 1);
4819 		return;
4820 	}
4821 
4822 	/*
4823 	 * For MSI-X interrupt, "Other" is always on vector[0].
4824 	 */
4825 	ixgbe_setup_ivar(ixgbe, IXGBE_IVAR_OTHER_CAUSES_INDEX, 0, -1);
4826 
4827 	/*
4828 	 * For each interrupt vector, populate the IVAR table
4829 	 */
4830 	for (v_idx = 0; v_idx < ixgbe->intr_cnt; v_idx++) {
4831 		vect = &ixgbe->vect_map[v_idx];
4832 
4833 		/*
4834 		 * For each rx ring bit set
4835 		 */
4836 		r_idx = bt_getlowbit(vect->rx_map, 0,
4837 		    (ixgbe->num_rx_rings - 1));
4838 
4839 		while (r_idx >= 0) {
4840 			hw_index = ixgbe->rx_rings[r_idx].hw_index;
4841 			ixgbe_setup_ivar(ixgbe, hw_index, v_idx, 0);
4842 			r_idx = bt_getlowbit(vect->rx_map, (r_idx + 1),
4843 			    (ixgbe->num_rx_rings - 1));
4844 		}
4845 
4846 		/*
4847 		 * For each tx ring bit set
4848 		 */
4849 		r_idx = bt_getlowbit(vect->tx_map, 0,
4850 		    (ixgbe->num_tx_rings - 1));
4851 
4852 		while (r_idx >= 0) {
4853 			ixgbe_setup_ivar(ixgbe, r_idx, v_idx, 1);
4854 			r_idx = bt_getlowbit(vect->tx_map, (r_idx + 1),
4855 			    (ixgbe->num_tx_rings - 1));
4856 		}
4857 	}
4858 }
4859 
4860 /*
4861  * ixgbe_rem_intr_handlers - Remove the interrupt handlers.
4862  */
4863 static void
4864 ixgbe_rem_intr_handlers(ixgbe_t *ixgbe)
4865 {
4866 	int i;
4867 	int rc;
4868 
4869 	for (i = 0; i < ixgbe->intr_cnt; i++) {
4870 		rc = ddi_intr_remove_handler(ixgbe->htable[i]);
4871 		if (rc != DDI_SUCCESS) {
4872 			IXGBE_DEBUGLOG_1(ixgbe,
4873 			    "Remove intr handler failed: %d", rc);
4874 		}
4875 	}
4876 }
4877 
4878 /*
4879  * ixgbe_rem_intrs - Remove the allocated interrupts.
4880  */
4881 static void
4882 ixgbe_rem_intrs(ixgbe_t *ixgbe)
4883 {
4884 	int i;
4885 	int rc;
4886 
4887 	for (i = 0; i < ixgbe->intr_cnt; i++) {
4888 		rc = ddi_intr_free(ixgbe->htable[i]);
4889 		if (rc != DDI_SUCCESS) {
4890 			IXGBE_DEBUGLOG_1(ixgbe,
4891 			    "Free intr failed: %d", rc);
4892 		}
4893 	}
4894 
4895 	kmem_free(ixgbe->htable, ixgbe->intr_size);
4896 	ixgbe->htable = NULL;
4897 }
4898 
4899 /*
4900  * ixgbe_enable_intrs - Enable all the ddi interrupts.
4901  */
4902 static int
4903 ixgbe_enable_intrs(ixgbe_t *ixgbe)
4904 {
4905 	int i;
4906 	int rc;
4907 
4908 	/*
4909 	 * Enable interrupts
4910 	 */
4911 	if (ixgbe->intr_cap & DDI_INTR_FLAG_BLOCK) {
4912 		/*
4913 		 * Call ddi_intr_block_enable() for MSI
4914 		 */
4915 		rc = ddi_intr_block_enable(ixgbe->htable, ixgbe->intr_cnt);
4916 		if (rc != DDI_SUCCESS) {
4917 			ixgbe_log(ixgbe,
4918 			    "Enable block intr failed: %d", rc);
4919 			return (IXGBE_FAILURE);
4920 		}
4921 	} else {
4922 		/*
4923 		 * Call ddi_intr_enable() for Legacy/MSI non block enable
4924 		 */
4925 		for (i = 0; i < ixgbe->intr_cnt; i++) {
4926 			rc = ddi_intr_enable(ixgbe->htable[i]);
4927 			if (rc != DDI_SUCCESS) {
4928 				ixgbe_log(ixgbe,
4929 				    "Enable intr failed: %d", rc);
4930 				return (IXGBE_FAILURE);
4931 			}
4932 		}
4933 	}
4934 
4935 	return (IXGBE_SUCCESS);
4936 }
4937 
4938 /*
4939  * ixgbe_disable_intrs - Disable all the interrupts.
4940  */
4941 static int
4942 ixgbe_disable_intrs(ixgbe_t *ixgbe)
4943 {
4944 	int i;
4945 	int rc;
4946 
4947 	/*
4948 	 * Disable all interrupts
4949 	 */
4950 	if (ixgbe->intr_cap & DDI_INTR_FLAG_BLOCK) {
4951 		rc = ddi_intr_block_disable(ixgbe->htable, ixgbe->intr_cnt);
4952 		if (rc != DDI_SUCCESS) {
4953 			ixgbe_log(ixgbe,
4954 			    "Disable block intr failed: %d", rc);
4955 			return (IXGBE_FAILURE);
4956 		}
4957 	} else {
4958 		for (i = 0; i < ixgbe->intr_cnt; i++) {
4959 			rc = ddi_intr_disable(ixgbe->htable[i]);
4960 			if (rc != DDI_SUCCESS) {
4961 				ixgbe_log(ixgbe,
4962 				    "Disable intr failed: %d", rc);
4963 				return (IXGBE_FAILURE);
4964 			}
4965 		}
4966 	}
4967 
4968 	return (IXGBE_SUCCESS);
4969 }
4970 
4971 /*
4972  * ixgbe_get_hw_state - Get and save parameters related to adapter hardware.
4973  */
4974 static void
4975 ixgbe_get_hw_state(ixgbe_t *ixgbe)
4976 {
4977 	struct ixgbe_hw *hw = &ixgbe->hw;
4978 	ixgbe_link_speed speed = IXGBE_LINK_SPEED_UNKNOWN;
4979 	boolean_t link_up = B_FALSE;
4980 	uint32_t pcs1g_anlp = 0;
4981 	uint32_t pcs1g_ana = 0;
4982 
4983 	ASSERT(mutex_owned(&ixgbe->gen_lock));
4984 	ixgbe->param_lp_1000fdx_cap = 0;
4985 	ixgbe->param_lp_100fdx_cap  = 0;
4986 
4987 	/* check for link, don't wait */
4988 	(void) ixgbe_check_link(hw, &speed, &link_up, false);
4989 	if (link_up) {
4990 		pcs1g_anlp = IXGBE_READ_REG(hw, IXGBE_PCS1GANLP);
4991 		pcs1g_ana = IXGBE_READ_REG(hw, IXGBE_PCS1GANA);
4992 
4993 		ixgbe->param_lp_1000fdx_cap =
4994 		    (pcs1g_anlp & IXGBE_PCS1GANLP_LPFD) ? 1 : 0;
4995 		ixgbe->param_lp_100fdx_cap =
4996 		    (pcs1g_anlp & IXGBE_PCS1GANLP_LPFD) ? 1 : 0;
4997 	}
4998 
4999 	ixgbe->param_adv_1000fdx_cap =
5000 	    (pcs1g_ana & IXGBE_PCS1GANA_FDC)  ? 1 : 0;
5001 	ixgbe->param_adv_100fdx_cap = (pcs1g_ana & IXGBE_PCS1GANA_FDC)  ? 1 : 0;
5002 }
5003 
5004 /*
5005  * ixgbe_get_driver_control - Notify that driver is in control of device.
5006  */
5007 static void
5008 ixgbe_get_driver_control(struct ixgbe_hw *hw)
5009 {
5010 	uint32_t ctrl_ext;
5011 
5012 	/*
5013 	 * Notify firmware that driver is in control of device
5014 	 */
5015 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5016 	ctrl_ext |= IXGBE_CTRL_EXT_DRV_LOAD;
5017 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5018 }
5019 
5020 /*
5021  * ixgbe_release_driver_control - Notify that driver is no longer in control
5022  * of device.
5023  */
5024 static void
5025 ixgbe_release_driver_control(struct ixgbe_hw *hw)
5026 {
5027 	uint32_t ctrl_ext;
5028 
5029 	/*
5030 	 * Notify firmware that driver is no longer in control of device
5031 	 */
5032 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5033 	ctrl_ext &= ~IXGBE_CTRL_EXT_DRV_LOAD;
5034 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5035 }
5036 
5037 /*
5038  * ixgbe_atomic_reserve - Atomic decrease operation.
5039  */
5040 int
5041 ixgbe_atomic_reserve(uint32_t *count_p, uint32_t n)
5042 {
5043 	uint32_t oldval;
5044 	uint32_t newval;
5045 
5046 	/*
5047 	 * ATOMICALLY
5048 	 */
5049 	do {
5050 		oldval = *count_p;
5051 		if (oldval < n)
5052 			return (-1);
5053 		newval = oldval - n;
5054 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
5055 
5056 	return (newval);
5057 }
5058 
5059 /*
5060  * ixgbe_mc_table_itr - Traverse the entries in the multicast table.
5061  */
5062 static uint8_t *
5063 ixgbe_mc_table_itr(struct ixgbe_hw *hw, uint8_t **upd_ptr, uint32_t *vmdq)
5064 {
5065 	uint8_t *addr = *upd_ptr;
5066 	uint8_t *new_ptr;
5067 
5068 	_NOTE(ARGUNUSED(hw));
5069 	_NOTE(ARGUNUSED(vmdq));
5070 
5071 	new_ptr = addr + IXGBE_ETH_LENGTH_OF_ADDRESS;
5072 	*upd_ptr = new_ptr;
5073 	return (addr);
5074 }
5075 
5076 /*
5077  * FMA support
5078  */
5079 int
5080 ixgbe_check_acc_handle(ddi_acc_handle_t handle)
5081 {
5082 	ddi_fm_error_t de;
5083 
5084 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
5085 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
5086 	return (de.fme_status);
5087 }
5088 
5089 int
5090 ixgbe_check_dma_handle(ddi_dma_handle_t handle)
5091 {
5092 	ddi_fm_error_t de;
5093 
5094 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
5095 	return (de.fme_status);
5096 }
5097 
5098 /*
5099  * ixgbe_fm_error_cb - The IO fault service error handling callback function.
5100  */
5101 static int
5102 ixgbe_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
5103 {
5104 	_NOTE(ARGUNUSED(impl_data));
5105 	/*
5106 	 * as the driver can always deal with an error in any dma or
5107 	 * access handle, we can just return the fme_status value.
5108 	 */
5109 	pci_ereport_post(dip, err, NULL);
5110 	return (err->fme_status);
5111 }
5112 
5113 static void
5114 ixgbe_fm_init(ixgbe_t *ixgbe)
5115 {
5116 	ddi_iblock_cookie_t iblk;
5117 	int fma_dma_flag;
5118 
5119 	/*
5120 	 * Only register with IO Fault Services if we have some capability
5121 	 */
5122 	if (ixgbe->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
5123 		ixgbe_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
5124 	} else {
5125 		ixgbe_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
5126 	}
5127 
5128 	if (ixgbe->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
5129 		fma_dma_flag = 1;
5130 	} else {
5131 		fma_dma_flag = 0;
5132 	}
5133 
5134 	ixgbe_set_fma_flags(fma_dma_flag);
5135 
5136 	if (ixgbe->fm_capabilities) {
5137 
5138 		/*
5139 		 * Register capabilities with IO Fault Services
5140 		 */
5141 		ddi_fm_init(ixgbe->dip, &ixgbe->fm_capabilities, &iblk);
5142 
5143 		/*
5144 		 * Initialize pci ereport capabilities if ereport capable
5145 		 */
5146 		if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities) ||
5147 		    DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5148 			pci_ereport_setup(ixgbe->dip);
5149 
5150 		/*
5151 		 * Register error callback if error callback capable
5152 		 */
5153 		if (DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5154 			ddi_fm_handler_register(ixgbe->dip,
5155 			    ixgbe_fm_error_cb, (void*) ixgbe);
5156 	}
5157 }
5158 
5159 static void
5160 ixgbe_fm_fini(ixgbe_t *ixgbe)
5161 {
5162 	/*
5163 	 * Only unregister FMA capabilities if they are registered
5164 	 */
5165 	if (ixgbe->fm_capabilities) {
5166 
5167 		/*
5168 		 * Release any resources allocated by pci_ereport_setup()
5169 		 */
5170 		if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities) ||
5171 		    DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5172 			pci_ereport_teardown(ixgbe->dip);
5173 
5174 		/*
5175 		 * Un-register error callback if error callback capable
5176 		 */
5177 		if (DDI_FM_ERRCB_CAP(ixgbe->fm_capabilities))
5178 			ddi_fm_handler_unregister(ixgbe->dip);
5179 
5180 		/*
5181 		 * Unregister from IO Fault Service
5182 		 */
5183 		ddi_fm_fini(ixgbe->dip);
5184 	}
5185 }
5186 
5187 void
5188 ixgbe_fm_ereport(ixgbe_t *ixgbe, char *detail)
5189 {
5190 	uint64_t ena;
5191 	char buf[FM_MAX_CLASS];
5192 
5193 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
5194 	ena = fm_ena_generate(0, FM_ENA_FMT1);
5195 	if (DDI_FM_EREPORT_CAP(ixgbe->fm_capabilities)) {
5196 		ddi_fm_ereport_post(ixgbe->dip, buf, ena, DDI_NOSLEEP,
5197 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
5198 	}
5199 }
5200 
5201 static int
5202 ixgbe_ring_start(mac_ring_driver_t rh, uint64_t mr_gen_num)
5203 {
5204 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)rh;
5205 
5206 	mutex_enter(&rx_ring->rx_lock);
5207 	rx_ring->ring_gen_num = mr_gen_num;
5208 	mutex_exit(&rx_ring->rx_lock);
5209 	return (0);
5210 }
5211 
5212 /*
5213  * Get the global ring index by a ring index within a group.
5214  */
5215 static int
5216 ixgbe_get_rx_ring_index(ixgbe_t *ixgbe, int gindex, int rindex)
5217 {
5218 	ixgbe_rx_ring_t *rx_ring;
5219 	int i;
5220 
5221 	for (i = 0; i < ixgbe->num_rx_rings; i++) {
5222 		rx_ring = &ixgbe->rx_rings[i];
5223 		if (rx_ring->group_index == gindex)
5224 			rindex--;
5225 		if (rindex < 0)
5226 			return (i);
5227 	}
5228 
5229 	return (-1);
5230 }
5231 
5232 /*
5233  * Callback funtion for MAC layer to register all rings.
5234  */
5235 /* ARGSUSED */
5236 void
5237 ixgbe_fill_ring(void *arg, mac_ring_type_t rtype, const int group_index,
5238     const int ring_index, mac_ring_info_t *infop, mac_ring_handle_t rh)
5239 {
5240 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
5241 	mac_intr_t *mintr = &infop->mri_intr;
5242 
5243 	switch (rtype) {
5244 	case MAC_RING_TYPE_RX: {
5245 		/*
5246 		 * 'index' is the ring index within the group.
5247 		 * Need to get the global ring index by searching in groups.
5248 		 */
5249 		int global_ring_index = ixgbe_get_rx_ring_index(
5250 		    ixgbe, group_index, ring_index);
5251 
5252 		ASSERT(global_ring_index >= 0);
5253 
5254 		ixgbe_rx_ring_t *rx_ring = &ixgbe->rx_rings[global_ring_index];
5255 		rx_ring->ring_handle = rh;
5256 
5257 		infop->mri_driver = (mac_ring_driver_t)rx_ring;
5258 		infop->mri_start = ixgbe_ring_start;
5259 		infop->mri_stop = NULL;
5260 		infop->mri_poll = ixgbe_ring_rx_poll;
5261 		infop->mri_stat = ixgbe_rx_ring_stat;
5262 
5263 		mintr->mi_handle = (mac_intr_handle_t)rx_ring;
5264 		mintr->mi_enable = ixgbe_rx_ring_intr_enable;
5265 		mintr->mi_disable = ixgbe_rx_ring_intr_disable;
5266 		if (ixgbe->intr_type &
5267 		    (DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI)) {
5268 			mintr->mi_ddi_handle =
5269 			    ixgbe->htable[rx_ring->intr_vector];
5270 		}
5271 
5272 		break;
5273 	}
5274 	case MAC_RING_TYPE_TX: {
5275 		ASSERT(group_index == -1);
5276 		ASSERT(ring_index < ixgbe->num_tx_rings);
5277 
5278 		ixgbe_tx_ring_t *tx_ring = &ixgbe->tx_rings[ring_index];
5279 		tx_ring->ring_handle = rh;
5280 
5281 		infop->mri_driver = (mac_ring_driver_t)tx_ring;
5282 		infop->mri_start = NULL;
5283 		infop->mri_stop = NULL;
5284 		infop->mri_tx = ixgbe_ring_tx;
5285 		infop->mri_stat = ixgbe_tx_ring_stat;
5286 		if (ixgbe->intr_type &
5287 		    (DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI)) {
5288 			mintr->mi_ddi_handle =
5289 			    ixgbe->htable[tx_ring->intr_vector];
5290 		}
5291 		break;
5292 	}
5293 	default:
5294 		break;
5295 	}
5296 }
5297 
5298 /*
5299  * Callback funtion for MAC layer to register all groups.
5300  */
5301 void
5302 ixgbe_fill_group(void *arg, mac_ring_type_t rtype, const int index,
5303     mac_group_info_t *infop, mac_group_handle_t gh)
5304 {
5305 	ixgbe_t *ixgbe = (ixgbe_t *)arg;
5306 
5307 	switch (rtype) {
5308 	case MAC_RING_TYPE_RX: {
5309 		ixgbe_rx_group_t *rx_group;
5310 
5311 		rx_group = &ixgbe->rx_groups[index];
5312 		rx_group->group_handle = gh;
5313 
5314 		infop->mgi_driver = (mac_group_driver_t)rx_group;
5315 		infop->mgi_start = NULL;
5316 		infop->mgi_stop = NULL;
5317 		infop->mgi_addmac = ixgbe_addmac;
5318 		infop->mgi_remmac = ixgbe_remmac;
5319 		infop->mgi_count = (ixgbe->num_rx_rings / ixgbe->num_rx_groups);
5320 
5321 		break;
5322 	}
5323 	case MAC_RING_TYPE_TX:
5324 		break;
5325 	default:
5326 		break;
5327 	}
5328 }
5329 
5330 /*
5331  * Enable interrupt on the specificed rx ring.
5332  */
5333 int
5334 ixgbe_rx_ring_intr_enable(mac_intr_handle_t intrh)
5335 {
5336 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)intrh;
5337 	ixgbe_t *ixgbe = rx_ring->ixgbe;
5338 	int r_idx = rx_ring->index;
5339 	int hw_r_idx = rx_ring->hw_index;
5340 	int v_idx = rx_ring->intr_vector;
5341 
5342 	mutex_enter(&ixgbe->gen_lock);
5343 	if (ixgbe->ixgbe_state & IXGBE_INTR_ADJUST) {
5344 		mutex_exit(&ixgbe->gen_lock);
5345 		/*
5346 		 * Simply return 0.
5347 		 * Interrupts are being adjusted. ixgbe_intr_adjust()
5348 		 * will eventually re-enable the interrupt when it's
5349 		 * done with the adjustment.
5350 		 */
5351 		return (0);
5352 	}
5353 
5354 	/*
5355 	 * To enable interrupt by setting the VAL bit of given interrupt
5356 	 * vector allocation register (IVAR).
5357 	 */
5358 	ixgbe_enable_ivar(ixgbe, hw_r_idx, 0);
5359 
5360 	BT_SET(ixgbe->vect_map[v_idx].rx_map, r_idx);
5361 
5362 	/*
5363 	 * To trigger a Rx interrupt to on this ring
5364 	 */
5365 	IXGBE_WRITE_REG(&ixgbe->hw, IXGBE_EICS, (1 << v_idx));
5366 	IXGBE_WRITE_FLUSH(&ixgbe->hw);
5367 
5368 	mutex_exit(&ixgbe->gen_lock);
5369 
5370 	return (0);
5371 }
5372 
5373 /*
5374  * Disable interrupt on the specificed rx ring.
5375  */
5376 int
5377 ixgbe_rx_ring_intr_disable(mac_intr_handle_t intrh)
5378 {
5379 	ixgbe_rx_ring_t *rx_ring = (ixgbe_rx_ring_t *)intrh;
5380 	ixgbe_t *ixgbe = rx_ring->ixgbe;
5381 	int r_idx = rx_ring->index;
5382 	int hw_r_idx = rx_ring->hw_index;
5383 	int v_idx = rx_ring->intr_vector;
5384 
5385 	mutex_enter(&ixgbe->gen_lock);
5386 	if (ixgbe->ixgbe_state & IXGBE_INTR_ADJUST) {
5387 		mutex_exit(&ixgbe->gen_lock);
5388 		/*
5389 		 * Simply return 0.
5390 		 * In the rare case where an interrupt is being
5391 		 * disabled while interrupts are being adjusted,
5392 		 * we don't fail the operation. No interrupts will
5393 		 * be generated while they are adjusted, and
5394 		 * ixgbe_intr_adjust() will cause the interrupts
5395 		 * to be re-enabled once it completes. Note that
5396 		 * in this case, packets may be delivered to the
5397 		 * stack via interrupts before xgbe_rx_ring_intr_enable()
5398 		 * is called again. This is acceptable since interrupt
5399 		 * adjustment is infrequent, and the stack will be
5400 		 * able to handle these packets.
5401 		 */
5402 		return (0);
5403 	}
5404 
5405 	/*
5406 	 * To disable interrupt by clearing the VAL bit of given interrupt
5407 	 * vector allocation register (IVAR).
5408 	 */
5409 	ixgbe_disable_ivar(ixgbe, hw_r_idx, 0);
5410 
5411 	BT_CLEAR(ixgbe->vect_map[v_idx].rx_map, r_idx);
5412 
5413 	mutex_exit(&ixgbe->gen_lock);
5414 
5415 	return (0);
5416 }
5417 
5418 /*
5419  * Add a mac address.
5420  */
5421 static int
5422 ixgbe_addmac(void *arg, const uint8_t *mac_addr)
5423 {
5424 	ixgbe_rx_group_t *rx_group = (ixgbe_rx_group_t *)arg;
5425 	ixgbe_t *ixgbe = rx_group->ixgbe;
5426 	struct ixgbe_hw *hw = &ixgbe->hw;
5427 	int slot, i;
5428 
5429 	mutex_enter(&ixgbe->gen_lock);
5430 
5431 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
5432 		mutex_exit(&ixgbe->gen_lock);
5433 		return (ECANCELED);
5434 	}
5435 
5436 	if (ixgbe->unicst_avail == 0) {
5437 		/* no slots available */
5438 		mutex_exit(&ixgbe->gen_lock);
5439 		return (ENOSPC);
5440 	}
5441 
5442 	/*
5443 	 * The first ixgbe->num_rx_groups slots are reserved for each respective
5444 	 * group. The rest slots are shared by all groups. While adding a
5445 	 * MAC address, reserved slots are firstly checked then the shared
5446 	 * slots are searched.
5447 	 */
5448 	slot = -1;
5449 	if (ixgbe->unicst_addr[rx_group->index].mac.set == 1) {
5450 		for (i = ixgbe->num_rx_groups; i < ixgbe->unicst_total; i++) {
5451 			if (ixgbe->unicst_addr[i].mac.set == 0) {
5452 				slot = i;
5453 				break;
5454 			}
5455 		}
5456 	} else {
5457 		slot = rx_group->index;
5458 	}
5459 
5460 	if (slot == -1) {
5461 		/* no slots available */
5462 		mutex_exit(&ixgbe->gen_lock);
5463 		return (ENOSPC);
5464 	}
5465 
5466 	bcopy(mac_addr, ixgbe->unicst_addr[slot].mac.addr, ETHERADDRL);
5467 	(void) ixgbe_set_rar(hw, slot, ixgbe->unicst_addr[slot].mac.addr,
5468 	    rx_group->index, IXGBE_RAH_AV);
5469 	ixgbe->unicst_addr[slot].mac.set = 1;
5470 	ixgbe->unicst_addr[slot].mac.group_index = rx_group->index;
5471 	ixgbe->unicst_avail--;
5472 
5473 	mutex_exit(&ixgbe->gen_lock);
5474 
5475 	return (0);
5476 }
5477 
5478 /*
5479  * Remove a mac address.
5480  */
5481 static int
5482 ixgbe_remmac(void *arg, const uint8_t *mac_addr)
5483 {
5484 	ixgbe_rx_group_t *rx_group = (ixgbe_rx_group_t *)arg;
5485 	ixgbe_t *ixgbe = rx_group->ixgbe;
5486 	struct ixgbe_hw *hw = &ixgbe->hw;
5487 	int slot;
5488 
5489 	mutex_enter(&ixgbe->gen_lock);
5490 
5491 	if (ixgbe->ixgbe_state & IXGBE_SUSPENDED) {
5492 		mutex_exit(&ixgbe->gen_lock);
5493 		return (ECANCELED);
5494 	}
5495 
5496 	slot = ixgbe_unicst_find(ixgbe, mac_addr);
5497 	if (slot == -1) {
5498 		mutex_exit(&ixgbe->gen_lock);
5499 		return (EINVAL);
5500 	}
5501 
5502 	if (ixgbe->unicst_addr[slot].mac.set == 0) {
5503 		mutex_exit(&ixgbe->gen_lock);
5504 		return (EINVAL);
5505 	}
5506 
5507 	bzero(ixgbe->unicst_addr[slot].mac.addr, ETHERADDRL);
5508 	(void) ixgbe_clear_rar(hw, slot);
5509 	ixgbe->unicst_addr[slot].mac.set = 0;
5510 	ixgbe->unicst_avail++;
5511 
5512 	mutex_exit(&ixgbe->gen_lock);
5513 
5514 	return (0);
5515 }
5516