xref: /illumos-gate/usr/src/uts/common/io/i40e/i40e_gld.c (revision 71531557)
1 /*
2  * This file and its contents are supplied under the terms of the
3  * Common Development and Distribution License ("CDDL"), version 1.0.
4  * You may only use this file in accordance with the terms of version
5  * 1.0 of the CDDL.
6  *
7  * A full copy of the text of the CDDL should have accompanied this
8  * source.  A copy of the CDDL is also available via the Internet at
9  * http://www.illumos.org/license/CDDL.
10  */
11 
12 /*
13  * Copyright 2015 OmniTI Computer Consulting, Inc. All rights reserved.
14  * Copyright 2016 Joyent, Inc.
15  * Copyright 2017 Tegile Systems, Inc.  All rights reserved.
16  */
17 
18 /*
19  * For more information, please see the big theory statement in i40e_main.c.
20  */
21 
22 #include "i40e_sw.h"
23 
24 #define	I40E_PROP_RX_DMA_THRESH	"_rx_dma_threshold"
25 #define	I40E_PROP_TX_DMA_THRESH	"_tx_dma_threshold"
26 #define	I40E_PROP_RX_ITR	"_rx_intr_throttle"
27 #define	I40E_PROP_TX_ITR	"_tx_intr_throttle"
28 #define	I40E_PROP_OTHER_ITR	"_other_intr_throttle"
29 
30 char *i40e_priv_props[] = {
31 	I40E_PROP_RX_DMA_THRESH,
32 	I40E_PROP_TX_DMA_THRESH,
33 	I40E_PROP_RX_ITR,
34 	I40E_PROP_TX_ITR,
35 	I40E_PROP_OTHER_ITR,
36 	NULL
37 };
38 
39 static int
40 i40e_group_remove_mac(void *arg, const uint8_t *mac_addr)
41 {
42 	i40e_t *i40e = arg;
43 	struct i40e_aqc_remove_macvlan_element_data filt;
44 	struct i40e_hw *hw = &i40e->i40e_hw_space;
45 	int ret, i, last;
46 	i40e_uaddr_t *iua;
47 
48 	if (I40E_IS_MULTICAST(mac_addr))
49 		return (EINVAL);
50 
51 	mutex_enter(&i40e->i40e_general_lock);
52 
53 	if (i40e->i40e_state & I40E_SUSPENDED) {
54 		ret = ECANCELED;
55 		goto done;
56 	}
57 
58 	for (i = 0; i < i40e->i40e_resources.ifr_nmacfilt_used; i++) {
59 		if (bcmp(mac_addr, i40e->i40e_uaddrs[i].iua_mac,
60 		    ETHERADDRL) == 0)
61 			break;
62 	}
63 
64 	if (i == i40e->i40e_resources.ifr_nmacfilt_used) {
65 		ret = ENOENT;
66 		goto done;
67 	}
68 
69 	iua = &i40e->i40e_uaddrs[i];
70 	ASSERT(i40e->i40e_resources.ifr_nmacfilt_used > 0);
71 
72 	bzero(&filt, sizeof (filt));
73 	bcopy(mac_addr, filt.mac_addr, ETHERADDRL);
74 	filt.flags = I40E_AQC_MACVLAN_DEL_PERFECT_MATCH |
75 	    I40E_AQC_MACVLAN_DEL_IGNORE_VLAN;
76 
77 	if (i40e_aq_remove_macvlan(hw, iua->iua_vsi, &filt, 1, NULL) !=
78 	    I40E_SUCCESS) {
79 		i40e_error(i40e, "failed to remove mac address "
80 		    "%2x:%2x:%2x:%2x:%2x:%2x from unicast filter: %d",
81 		    mac_addr[0], mac_addr[1], mac_addr[2], mac_addr[3],
82 		    mac_addr[4], mac_addr[5], filt.error_code);
83 		ret = EIO;
84 		goto done;
85 	}
86 
87 	last = i40e->i40e_resources.ifr_nmacfilt_used - 1;
88 	if (i != last) {
89 		i40e_uaddr_t *src = &i40e->i40e_uaddrs[last];
90 		bcopy(src, iua, sizeof (i40e_uaddr_t));
91 	}
92 
93 	/*
94 	 * Set the multicast bit in the last one to indicate to ourselves that
95 	 * it's invalid.
96 	 */
97 	bzero(&i40e->i40e_uaddrs[last], sizeof (i40e_uaddr_t));
98 	i40e->i40e_uaddrs[last].iua_mac[0] = 0x01;
99 	i40e->i40e_resources.ifr_nmacfilt_used--;
100 	ret = 0;
101 done:
102 	mutex_exit(&i40e->i40e_general_lock);
103 
104 	return (ret);
105 }
106 
107 static int
108 i40e_group_add_mac(void *arg, const uint8_t *mac_addr)
109 {
110 	i40e_t *i40e = arg;
111 	struct i40e_hw *hw = &i40e->i40e_hw_space;
112 	int i, ret;
113 	i40e_uaddr_t *iua;
114 	struct i40e_aqc_add_macvlan_element_data filt;
115 
116 	if (I40E_IS_MULTICAST(mac_addr))
117 		return (EINVAL);
118 
119 	mutex_enter(&i40e->i40e_general_lock);
120 	if (i40e->i40e_state & I40E_SUSPENDED) {
121 		ret = ECANCELED;
122 		goto done;
123 	}
124 
125 	if (i40e->i40e_resources.ifr_nmacfilt ==
126 	    i40e->i40e_resources.ifr_nmacfilt_used) {
127 		ret = ENOSPC;
128 		goto done;
129 	}
130 
131 	for (i = 0; i < i40e->i40e_resources.ifr_nmacfilt_used; i++) {
132 		if (bcmp(mac_addr, i40e->i40e_uaddrs[i].iua_mac,
133 		    ETHERADDRL) == 0) {
134 			ret = EEXIST;
135 			goto done;
136 		}
137 	}
138 
139 	/*
140 	 * Note, the general use of the i40e_vsi_id will have to be refactored
141 	 * when we have proper group support.
142 	 */
143 	bzero(&filt, sizeof (filt));
144 	bcopy(mac_addr, filt.mac_addr, ETHERADDRL);
145 	filt.flags = I40E_AQC_MACVLAN_ADD_PERFECT_MATCH	|
146 	    I40E_AQC_MACVLAN_ADD_IGNORE_VLAN;
147 
148 	if ((ret = i40e_aq_add_macvlan(hw, i40e->i40e_vsi_id, &filt, 1,
149 	    NULL)) != I40E_SUCCESS) {
150 		i40e_error(i40e, "failed to add mac address "
151 		    "%2x:%2x:%2x:%2x:%2x:%2x to unicast filter: %d",
152 		    mac_addr[0], mac_addr[1], mac_addr[2], mac_addr[3],
153 		    mac_addr[4], mac_addr[5], ret);
154 		ret = EIO;
155 		goto done;
156 	}
157 
158 	iua = &i40e->i40e_uaddrs[i40e->i40e_resources.ifr_nmacfilt_used];
159 	bcopy(mac_addr, iua->iua_mac, ETHERADDRL);
160 	iua->iua_vsi = i40e->i40e_vsi_id;
161 	i40e->i40e_resources.ifr_nmacfilt_used++;
162 	ASSERT(i40e->i40e_resources.ifr_nmacfilt_used <=
163 	    i40e->i40e_resources.ifr_nmacfilt);
164 	ret = 0;
165 done:
166 	mutex_exit(&i40e->i40e_general_lock);
167 	return (ret);
168 }
169 
170 static int
171 i40e_m_start(void *arg)
172 {
173 	i40e_t *i40e = arg;
174 	int rc = 0;
175 
176 	mutex_enter(&i40e->i40e_general_lock);
177 	if (i40e->i40e_state & I40E_SUSPENDED) {
178 		rc = ECANCELED;
179 		goto done;
180 	}
181 
182 	if (!i40e_start(i40e, B_TRUE)) {
183 		rc = EIO;
184 		goto done;
185 	}
186 
187 	atomic_or_32(&i40e->i40e_state, I40E_STARTED);
188 done:
189 	mutex_exit(&i40e->i40e_general_lock);
190 
191 	return (rc);
192 }
193 
194 static void
195 i40e_m_stop(void *arg)
196 {
197 	i40e_t *i40e = arg;
198 
199 	mutex_enter(&i40e->i40e_general_lock);
200 
201 	if (i40e->i40e_state & I40E_SUSPENDED)
202 		goto done;
203 
204 	atomic_and_32(&i40e->i40e_state, ~I40E_STARTED);
205 	i40e_stop(i40e, B_TRUE);
206 done:
207 	mutex_exit(&i40e->i40e_general_lock);
208 }
209 
210 /*
211  * Enable and disable promiscuous mode as requested. We have to toggle both
212  * unicast and multicast. Note that multicast may already be enabled due to the
213  * i40e_m_multicast may toggle it itself. See i40e_main.c for more information
214  * on this.
215  */
216 static int
217 i40e_m_promisc(void *arg, boolean_t on)
218 {
219 	i40e_t *i40e = arg;
220 	struct i40e_hw *hw = &i40e->i40e_hw_space;
221 	int ret = 0, err = 0;
222 
223 	mutex_enter(&i40e->i40e_general_lock);
224 	if (i40e->i40e_state & I40E_SUSPENDED) {
225 		ret = ECANCELED;
226 		goto done;
227 	}
228 
229 
230 	ret = i40e_aq_set_vsi_unicast_promiscuous(hw, i40e->i40e_vsi_id,
231 	    on, NULL, B_FALSE);
232 	if (ret != I40E_SUCCESS) {
233 		i40e_error(i40e, "failed to %s unicast promiscuity on "
234 		    "the default VSI: %d", on == B_TRUE ? "enable" : "disable",
235 		    ret);
236 		err = EIO;
237 		goto done;
238 	}
239 
240 	/*
241 	 * If we have a non-zero mcast_promisc_count, then it has already been
242 	 * enabled or we need to leave it that way and not touch it.
243 	 */
244 	if (i40e->i40e_mcast_promisc_count > 0) {
245 		i40e->i40e_promisc_on = on;
246 		goto done;
247 	}
248 
249 	ret = i40e_aq_set_vsi_multicast_promiscuous(hw, i40e->i40e_vsi_id,
250 	    on, NULL);
251 	if (ret != I40E_SUCCESS) {
252 		i40e_error(i40e, "failed to %s multicast promiscuity on "
253 		    "the default VSI: %d", on == B_TRUE ? "enable" : "disable",
254 		    ret);
255 
256 		/*
257 		 * Try our best to put us back into a state that MAC expects us
258 		 * to be in.
259 		 */
260 		ret = i40e_aq_set_vsi_unicast_promiscuous(hw, i40e->i40e_vsi_id,
261 		    !on, NULL, B_FALSE);
262 		if (ret != I40E_SUCCESS) {
263 			i40e_error(i40e, "failed to %s unicast promiscuity on "
264 			    "the default VSI after toggling multicast failed: "
265 			    "%d", on == B_TRUE ? "disable" : "enable", ret);
266 		}
267 
268 		err = EIO;
269 		goto done;
270 	} else {
271 		i40e->i40e_promisc_on = on;
272 	}
273 
274 done:
275 	mutex_exit(&i40e->i40e_general_lock);
276 	return (err);
277 }
278 
279 /*
280  * See the big theory statement in i40e_main.c for multicast address management.
281  */
282 static int
283 i40e_multicast_add(i40e_t *i40e, const uint8_t *multicast_address)
284 {
285 	struct i40e_hw *hw = &i40e->i40e_hw_space;
286 	struct i40e_aqc_add_macvlan_element_data filt;
287 	i40e_maddr_t *mc;
288 	int ret;
289 
290 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
291 
292 	if (i40e->i40e_resources.ifr_nmcastfilt_used ==
293 	    i40e->i40e_resources.ifr_nmcastfilt) {
294 		if (i40e->i40e_mcast_promisc_count == 0 &&
295 		    i40e->i40e_promisc_on == B_FALSE) {
296 			ret = i40e_aq_set_vsi_multicast_promiscuous(hw,
297 			    i40e->i40e_vsi_id, B_TRUE, NULL);
298 			if (ret != I40E_SUCCESS) {
299 				i40e_error(i40e, "failed to enable multicast "
300 				    "promiscuous mode on VSI %d: %d",
301 				    i40e->i40e_vsi_id, ret);
302 				return (EIO);
303 			}
304 		}
305 		i40e->i40e_mcast_promisc_count++;
306 		return (0);
307 	}
308 
309 	mc = &i40e->i40e_maddrs[i40e->i40e_resources.ifr_nmcastfilt_used];
310 	bzero(&filt, sizeof (filt));
311 	bcopy(multicast_address, filt.mac_addr, ETHERADDRL);
312 	filt.flags = I40E_AQC_MACVLAN_ADD_HASH_MATCH |
313 	    I40E_AQC_MACVLAN_ADD_IGNORE_VLAN;
314 
315 	if ((ret = i40e_aq_add_macvlan(hw, i40e->i40e_vsi_id, &filt, 1,
316 	    NULL)) != I40E_SUCCESS) {
317 		i40e_error(i40e, "failed to add mac address "
318 		    "%2x:%2x:%2x:%2x:%2x:%2x to multicast filter: %d",
319 		    multicast_address[0], multicast_address[1],
320 		    multicast_address[2], multicast_address[3],
321 		    multicast_address[4], multicast_address[5],
322 		    ret);
323 		return (EIO);
324 	}
325 
326 	bcopy(multicast_address, mc->ima_mac, ETHERADDRL);
327 	i40e->i40e_resources.ifr_nmcastfilt_used++;
328 	return (0);
329 }
330 
331 /*
332  * See the big theory statement in i40e_main.c for multicast address management.
333  */
334 static int
335 i40e_multicast_remove(i40e_t *i40e, const uint8_t *multicast_address)
336 {
337 	int i, ret;
338 	struct i40e_hw *hw = &i40e->i40e_hw_space;
339 
340 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
341 
342 	for (i = 0; i < i40e->i40e_resources.ifr_nmcastfilt_used; i++) {
343 		struct i40e_aqc_remove_macvlan_element_data filt;
344 		int last;
345 
346 		if (bcmp(multicast_address, i40e->i40e_maddrs[i].ima_mac,
347 		    ETHERADDRL) != 0) {
348 			continue;
349 		}
350 
351 		bzero(&filt, sizeof (filt));
352 		bcopy(multicast_address, filt.mac_addr, ETHERADDRL);
353 		filt.flags = I40E_AQC_MACVLAN_DEL_HASH_MATCH |
354 		    I40E_AQC_MACVLAN_DEL_IGNORE_VLAN;
355 
356 		if (i40e_aq_remove_macvlan(hw, i40e->i40e_vsi_id,
357 		    &filt, 1, NULL) != I40E_SUCCESS) {
358 			i40e_error(i40e, "failed to remove mac address "
359 			    "%2x:%2x:%2x:%2x:%2x:%2x from multicast "
360 			    "filter: %d",
361 			    multicast_address[0], multicast_address[1],
362 			    multicast_address[2], multicast_address[3],
363 			    multicast_address[4], multicast_address[5],
364 			    filt.error_code);
365 			return (EIO);
366 		}
367 
368 		last = i40e->i40e_resources.ifr_nmcastfilt_used - 1;
369 		if (i != last) {
370 			bcopy(&i40e->i40e_maddrs[last], &i40e->i40e_maddrs[i],
371 			    sizeof (i40e_maddr_t));
372 			bzero(&i40e->i40e_maddrs[last], sizeof (i40e_maddr_t));
373 		}
374 
375 		ASSERT(i40e->i40e_resources.ifr_nmcastfilt_used > 0);
376 		i40e->i40e_resources.ifr_nmcastfilt_used--;
377 		return (0);
378 	}
379 
380 	if (i40e->i40e_mcast_promisc_count > 0) {
381 		if (i40e->i40e_mcast_promisc_count == 1 &&
382 		    i40e->i40e_promisc_on == B_FALSE) {
383 			ret = i40e_aq_set_vsi_multicast_promiscuous(hw,
384 			    i40e->i40e_vsi_id, B_FALSE, NULL);
385 			if (ret != I40E_SUCCESS) {
386 				i40e_error(i40e, "failed to disable "
387 				    "multicast promiscuous mode on VSI %d: %d",
388 				    i40e->i40e_vsi_id, ret);
389 				return (EIO);
390 			}
391 		}
392 		i40e->i40e_mcast_promisc_count--;
393 
394 		return (0);
395 	}
396 
397 	return (ENOENT);
398 }
399 
400 static int
401 i40e_m_multicast(void *arg, boolean_t add, const uint8_t *multicast_address)
402 {
403 	i40e_t *i40e = arg;
404 	int rc;
405 
406 	mutex_enter(&i40e->i40e_general_lock);
407 
408 	if (i40e->i40e_state & I40E_SUSPENDED) {
409 		mutex_exit(&i40e->i40e_general_lock);
410 		return (ECANCELED);
411 	}
412 
413 	if (add == B_TRUE) {
414 		rc = i40e_multicast_add(i40e, multicast_address);
415 	} else {
416 		rc = i40e_multicast_remove(i40e, multicast_address);
417 	}
418 
419 	mutex_exit(&i40e->i40e_general_lock);
420 	return (rc);
421 }
422 
423 /* ARGSUSED */
424 static void
425 i40e_m_ioctl(void *arg, queue_t *q, mblk_t *mp)
426 {
427 	/*
428 	 * At this time, we don't support toggling i40e into loopback mode. It's
429 	 * questionable how much value this has when there's no clear way to
430 	 * toggle this behavior from a supported way in userland.
431 	 */
432 	miocnak(q, mp, 0, EINVAL);
433 }
434 
435 static int
436 i40e_ring_start(mac_ring_driver_t rh, uint64_t gen_num)
437 {
438 	i40e_trqpair_t *itrq = (i40e_trqpair_t *)rh;
439 
440 	/*
441 	 * GLDv3 requires we keep track of a generation number, as it uses
442 	 * that number to keep track of whether or not a ring is active.
443 	 */
444 	mutex_enter(&itrq->itrq_rx_lock);
445 	itrq->itrq_rxgen = gen_num;
446 	mutex_exit(&itrq->itrq_rx_lock);
447 	return (0);
448 }
449 
450 /* ARGSUSED */
451 static int
452 i40e_rx_ring_intr_enable(mac_intr_handle_t intrh)
453 {
454 	i40e_trqpair_t *itrq = (i40e_trqpair_t *)intrh;
455 	i40e_t *i40e = itrq->itrq_i40e;
456 
457 	mutex_enter(&itrq->itrq_rx_lock);
458 	ASSERT(itrq->itrq_intr_poll == B_TRUE);
459 	i40e_intr_rx_queue_enable(i40e, itrq->itrq_index);
460 	itrq->itrq_intr_poll = B_FALSE;
461 	mutex_exit(&itrq->itrq_rx_lock);
462 
463 	return (0);
464 }
465 
466 /* ARGSUSED */
467 static int
468 i40e_rx_ring_intr_disable(mac_intr_handle_t intrh)
469 {
470 	i40e_trqpair_t *itrq = (i40e_trqpair_t *)intrh;
471 	i40e_t *i40e = itrq->itrq_i40e;
472 
473 	mutex_enter(&itrq->itrq_rx_lock);
474 	i40e_intr_rx_queue_disable(i40e, itrq->itrq_index);
475 	itrq->itrq_intr_poll = B_TRUE;
476 	mutex_exit(&itrq->itrq_rx_lock);
477 
478 	return (0);
479 }
480 
481 /* ARGSUSED */
482 static void
483 i40e_fill_tx_ring(void *arg, mac_ring_type_t rtype, const int group_index,
484     const int ring_index, mac_ring_info_t *infop, mac_ring_handle_t rh)
485 {
486 	i40e_t *i40e = arg;
487 	mac_intr_t *mintr = &infop->mri_intr;
488 	i40e_trqpair_t *itrq = &(i40e->i40e_trqpairs[ring_index]);
489 
490 	/*
491 	 * Note the group index here is expected to be -1 due to the fact that
492 	 * we're not actually grouping things tx-wise at this time.
493 	 */
494 	ASSERT(group_index == -1);
495 	ASSERT(ring_index < i40e->i40e_num_trqpairs);
496 
497 	itrq->itrq_mactxring = rh;
498 	infop->mri_driver = (mac_ring_driver_t)itrq;
499 	infop->mri_start = NULL;
500 	infop->mri_stop = NULL;
501 	infop->mri_tx = i40e_ring_tx;
502 	infop->mri_stat = i40e_tx_ring_stat;
503 
504 	/*
505 	 * We only provide the handle in cases where we have MSI-X interrupts,
506 	 * to indicate that we'd actually support retargetting.
507 	 */
508 	if (i40e->i40e_intr_type & DDI_INTR_TYPE_MSIX) {
509 		mintr->mi_ddi_handle =
510 		    i40e->i40e_intr_handles[itrq->itrq_tx_intrvec];
511 	}
512 }
513 
514 /* ARGSUSED */
515 static void
516 i40e_fill_rx_ring(void *arg, mac_ring_type_t rtype, const int group_index,
517     const int ring_index, mac_ring_info_t *infop, mac_ring_handle_t rh)
518 {
519 	i40e_t *i40e = arg;
520 	mac_intr_t *mintr = &infop->mri_intr;
521 	i40e_trqpair_t *itrq = &i40e->i40e_trqpairs[ring_index];
522 
523 	/*
524 	 * We assert the group number and ring index to help sanity check
525 	 * ourselves and mark that we'll need to rework this when we have
526 	 * multiple groups.
527 	 */
528 	ASSERT3S(group_index, ==, 0);
529 	ASSERT3S(ring_index, <, i40e->i40e_num_trqpairs);
530 
531 	itrq->itrq_macrxring = rh;
532 	infop->mri_driver = (mac_ring_driver_t)itrq;
533 	infop->mri_start = i40e_ring_start;
534 	infop->mri_stop = NULL;
535 	infop->mri_poll = i40e_ring_rx_poll;
536 	infop->mri_stat = i40e_rx_ring_stat;
537 	mintr->mi_handle = (mac_intr_handle_t)itrq;
538 	mintr->mi_enable = i40e_rx_ring_intr_enable;
539 	mintr->mi_disable = i40e_rx_ring_intr_disable;
540 
541 	/*
542 	 * We only provide the handle in cases where we have MSI-X interrupts,
543 	 * to indicate that we'd actually support retargetting.
544 	 */
545 	if (i40e->i40e_intr_type & DDI_INTR_TYPE_MSIX) {
546 		mintr->mi_ddi_handle =
547 		    i40e->i40e_intr_handles[itrq->itrq_rx_intrvec];
548 	}
549 }
550 
551 /* ARGSUSED */
552 static void
553 i40e_fill_rx_group(void *arg, mac_ring_type_t rtype, const int index,
554     mac_group_info_t *infop, mac_group_handle_t gh)
555 {
556 	i40e_t *i40e = arg;
557 
558 	if (rtype != MAC_RING_TYPE_RX)
559 		return;
560 
561 	/*
562 	 * Note, this is a simplified view of a group, given that we only have a
563 	 * single group and a single ring at the moment. We'll want to expand
564 	 * upon this as we leverage more hardware functionality.
565 	 */
566 	i40e->i40e_rx_group_handle = gh;
567 	infop->mgi_driver = (mac_group_driver_t)i40e;
568 	infop->mgi_start = NULL;
569 	infop->mgi_stop = NULL;
570 	infop->mgi_addmac = i40e_group_add_mac;
571 	infop->mgi_remmac = i40e_group_remove_mac;
572 
573 	ASSERT(i40e->i40e_num_rx_groups == I40E_GROUP_MAX);
574 	infop->mgi_count = i40e->i40e_num_trqpairs;
575 }
576 
577 static boolean_t
578 i40e_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
579 {
580 	i40e_t *i40e = arg;
581 	mac_capab_rings_t *cap_rings;
582 
583 	switch (cap) {
584 	case MAC_CAPAB_HCKSUM: {
585 		uint32_t *txflags = cap_data;
586 
587 		*txflags = 0;
588 		if (i40e->i40e_tx_hcksum_enable == B_TRUE)
589 			*txflags = HCKSUM_INET_PARTIAL | HCKSUM_IPHDRCKSUM;
590 		break;
591 	}
592 
593 	case MAC_CAPAB_RINGS:
594 		cap_rings = cap_data;
595 		cap_rings->mr_group_type = MAC_GROUP_TYPE_STATIC;
596 		switch (cap_rings->mr_type) {
597 		case MAC_RING_TYPE_TX:
598 			/*
599 			 * Note, saying we have no rings, but some number of
600 			 * groups indicates to MAC that it should create
601 			 * psuedo-groups with one for each TX ring. This may not
602 			 * be the long term behavior we want, but it'll work for
603 			 * now.
604 			 */
605 			cap_rings->mr_gnum = 0;
606 			cap_rings->mr_rnum = i40e->i40e_num_trqpairs;
607 			cap_rings->mr_rget = i40e_fill_tx_ring;
608 			cap_rings->mr_gget = NULL;
609 			cap_rings->mr_gaddring = NULL;
610 			cap_rings->mr_gremring = NULL;
611 			break;
612 		case MAC_RING_TYPE_RX:
613 			cap_rings->mr_rnum = i40e->i40e_num_trqpairs;
614 			cap_rings->mr_rget = i40e_fill_rx_ring;
615 			cap_rings->mr_gnum = I40E_GROUP_MAX;
616 			cap_rings->mr_gget = i40e_fill_rx_group;
617 			cap_rings->mr_gaddring = NULL;
618 			cap_rings->mr_gremring = NULL;
619 			break;
620 		default:
621 			return (B_FALSE);
622 		}
623 		break;
624 	default:
625 		return (B_FALSE);
626 	}
627 
628 	return (B_TRUE);
629 }
630 
631 /* ARGSUSED */
632 static int
633 i40e_m_setprop_private(i40e_t *i40e, const char *pr_name, uint_t pr_valsize,
634     const void *pr_val)
635 {
636 	int ret;
637 	long val;
638 	char *eptr;
639 
640 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
641 
642 	if ((ret = ddi_strtol(pr_val, &eptr, 10, &val)) != 0 ||
643 	    *eptr != '\0') {
644 		return (ret);
645 	}
646 
647 	if (strcmp(pr_name, I40E_PROP_RX_DMA_THRESH) == 0) {
648 		if (val < I40E_MIN_RX_DMA_THRESH ||
649 		    val > I40E_MAX_RX_DMA_THRESH) {
650 			return (EINVAL);
651 		}
652 		i40e->i40e_rx_dma_min = (uint32_t)val;
653 		return (0);
654 	}
655 
656 	if (strcmp(pr_name, I40E_PROP_TX_DMA_THRESH) == 0) {
657 		if (val < I40E_MIN_TX_DMA_THRESH ||
658 		    val > I40E_MAX_TX_DMA_THRESH) {
659 			return (EINVAL);
660 		}
661 		i40e->i40e_tx_dma_min = (uint32_t)val;
662 		return (0);
663 	}
664 
665 	if (strcmp(pr_name, I40E_PROP_RX_ITR) == 0) {
666 		if (val < I40E_MIN_ITR ||
667 		    val > I40E_MAX_ITR) {
668 			return (EINVAL);
669 		}
670 		i40e->i40e_rx_itr = (uint32_t)val;
671 		i40e_intr_set_itr(i40e, I40E_ITR_INDEX_RX, i40e->i40e_rx_itr);
672 		return (0);
673 	}
674 
675 	if (strcmp(pr_name, I40E_PROP_TX_ITR) == 0) {
676 		if (val < I40E_MIN_ITR ||
677 		    val > I40E_MAX_ITR) {
678 			return (EINVAL);
679 		}
680 		i40e->i40e_tx_itr = (uint32_t)val;
681 		i40e_intr_set_itr(i40e, I40E_ITR_INDEX_TX, i40e->i40e_tx_itr);
682 		return (0);
683 	}
684 
685 	if (strcmp(pr_name, I40E_PROP_OTHER_ITR) == 0) {
686 		if (val < I40E_MIN_ITR ||
687 		    val > I40E_MAX_ITR) {
688 			return (EINVAL);
689 		}
690 		i40e->i40e_tx_itr = (uint32_t)val;
691 		i40e_intr_set_itr(i40e, I40E_ITR_INDEX_OTHER,
692 		    i40e->i40e_other_itr);
693 		return (0);
694 	}
695 
696 	return (ENOTSUP);
697 }
698 
699 static int
700 i40e_m_getprop_private(i40e_t *i40e, const char *pr_name, uint_t pr_valsize,
701     void *pr_val)
702 {
703 	uint32_t val;
704 
705 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
706 
707 	if (strcmp(pr_name, I40E_PROP_RX_DMA_THRESH) == 0) {
708 		val = i40e->i40e_rx_dma_min;
709 	} else if (strcmp(pr_name, I40E_PROP_TX_DMA_THRESH) == 0) {
710 		val = i40e->i40e_tx_dma_min;
711 	} else if (strcmp(pr_name, I40E_PROP_RX_ITR) == 0) {
712 		val = i40e->i40e_rx_itr;
713 	} else if (strcmp(pr_name, I40E_PROP_TX_ITR) == 0) {
714 		val = i40e->i40e_tx_itr;
715 	} else if (strcmp(pr_name, I40E_PROP_OTHER_ITR) == 0) {
716 		val = i40e->i40e_other_itr;
717 	} else {
718 		return (ENOTSUP);
719 	}
720 
721 	if (snprintf(pr_val, pr_valsize, "%d", val) >= pr_valsize)
722 		return (ERANGE);
723 	return (0);
724 }
725 
726 /*
727  * Annoyingly for private properties MAC seems to ignore default values that
728  * aren't strings. That means that we have to translate all of these into
729  * uint32_t's and instead we size the buffer to be large enough to hold a
730  * uint32_t.
731  */
732 /* ARGSUSED */
733 static void
734 i40e_m_propinfo_private(i40e_t *i40e, const char *pr_name,
735     mac_prop_info_handle_t prh)
736 {
737 	char buf[64];
738 	uint32_t def;
739 
740 	if (strcmp(pr_name, I40E_PROP_RX_DMA_THRESH) == 0) {
741 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_RW);
742 		def = I40E_DEF_RX_DMA_THRESH;
743 		mac_prop_info_set_range_uint32(prh,
744 		    I40E_MIN_RX_DMA_THRESH,
745 		    I40E_MAX_RX_DMA_THRESH);
746 	} else if (strcmp(pr_name, I40E_PROP_TX_DMA_THRESH) == 0) {
747 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_RW);
748 		def = I40E_DEF_TX_DMA_THRESH;
749 		mac_prop_info_set_range_uint32(prh,
750 		    I40E_MIN_TX_DMA_THRESH,
751 		    I40E_MAX_TX_DMA_THRESH);
752 	} else if (strcmp(pr_name, I40E_PROP_RX_ITR) == 0) {
753 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_RW);
754 		def = I40E_DEF_RX_ITR;
755 		mac_prop_info_set_range_uint32(prh, I40E_MIN_ITR, I40E_MAX_ITR);
756 	} else if (strcmp(pr_name, I40E_PROP_TX_ITR) == 0) {
757 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_RW);
758 		def = I40E_DEF_TX_ITR;
759 		mac_prop_info_set_range_uint32(prh, I40E_MIN_ITR, I40E_MAX_ITR);
760 	} else if (strcmp(pr_name, I40E_PROP_OTHER_ITR) == 0) {
761 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_RW);
762 		def = I40E_DEF_OTHER_ITR;
763 		mac_prop_info_set_range_uint32(prh, I40E_MIN_ITR, I40E_MAX_ITR);
764 	} else {
765 		return;
766 	}
767 
768 	(void) snprintf(buf, sizeof (buf), "%d", def);
769 	mac_prop_info_set_default_str(prh, buf);
770 }
771 
772 static int
773 i40e_m_setprop(void *arg, const char *pr_name, mac_prop_id_t pr_num,
774     uint_t pr_valsize, const void *pr_val)
775 {
776 	uint32_t new_mtu;
777 	i40e_t *i40e = arg;
778 	int ret = 0;
779 
780 	mutex_enter(&i40e->i40e_general_lock);
781 	if (i40e->i40e_state & I40E_SUSPENDED) {
782 		mutex_exit(&i40e->i40e_general_lock);
783 		return (ECANCELED);
784 	}
785 
786 	switch (pr_num) {
787 	/*
788 	 * These properties are always read-only across every device.
789 	 */
790 	case MAC_PROP_DUPLEX:
791 	case MAC_PROP_SPEED:
792 	case MAC_PROP_STATUS:
793 	case MAC_PROP_ADV_100FDX_CAP:
794 	case MAC_PROP_ADV_1000FDX_CAP:
795 	case MAC_PROP_ADV_10GFDX_CAP:
796 	case MAC_PROP_ADV_25GFDX_CAP:
797 	case MAC_PROP_ADV_40GFDX_CAP:
798 		ret = ENOTSUP;
799 		break;
800 	/*
801 	 * These are read-only at this time as we don't support configuring
802 	 * auto-negotiation. See the theory statement in i40e_main.c.
803 	 */
804 	case MAC_PROP_EN_100FDX_CAP:
805 	case MAC_PROP_EN_1000FDX_CAP:
806 	case MAC_PROP_EN_10GFDX_CAP:
807 	case MAC_PROP_EN_25GFDX_CAP:
808 	case MAC_PROP_EN_40GFDX_CAP:
809 	case MAC_PROP_AUTONEG:
810 	case MAC_PROP_FLOWCTRL:
811 		ret = ENOTSUP;
812 		break;
813 
814 	case MAC_PROP_MTU:
815 		bcopy(pr_val, &new_mtu, sizeof (new_mtu));
816 		if (new_mtu == i40e->i40e_sdu)
817 			break;
818 
819 		if (new_mtu < I40E_MIN_MTU ||
820 		    new_mtu > I40E_MAX_MTU) {
821 			ret = EINVAL;
822 			break;
823 		}
824 
825 		if (i40e->i40e_state & I40E_STARTED) {
826 			ret = EBUSY;
827 			break;
828 		}
829 
830 		ret = mac_maxsdu_update(i40e->i40e_mac_hdl, new_mtu);
831 		if (ret == 0) {
832 			i40e->i40e_sdu = new_mtu;
833 			i40e_update_mtu(i40e);
834 		}
835 		break;
836 
837 	case MAC_PROP_PRIVATE:
838 		ret = i40e_m_setprop_private(i40e, pr_name, pr_valsize, pr_val);
839 		break;
840 	default:
841 		ret = ENOTSUP;
842 		break;
843 	}
844 
845 	mutex_exit(&i40e->i40e_general_lock);
846 	return (ret);
847 }
848 
849 static int
850 i40e_m_getprop(void *arg, const char *pr_name, mac_prop_id_t pr_num,
851     uint_t pr_valsize, void *pr_val)
852 {
853 	i40e_t *i40e = arg;
854 	uint64_t speed;
855 	int ret = 0;
856 	uint8_t *u8;
857 	link_flowctrl_t fctl;
858 
859 	mutex_enter(&i40e->i40e_general_lock);
860 
861 	switch (pr_num) {
862 	case MAC_PROP_DUPLEX:
863 		if (pr_valsize < sizeof (link_duplex_t)) {
864 			ret = EOVERFLOW;
865 			break;
866 		}
867 		bcopy(&i40e->i40e_link_duplex, pr_val, sizeof (link_duplex_t));
868 		break;
869 	case MAC_PROP_SPEED:
870 		if (pr_valsize < sizeof (uint64_t)) {
871 			ret = EOVERFLOW;
872 			break;
873 		}
874 		speed = i40e->i40e_link_speed * 1000000ULL;
875 		bcopy(&speed, pr_val, sizeof (speed));
876 		break;
877 	case MAC_PROP_STATUS:
878 		if (pr_valsize < sizeof (link_state_t)) {
879 			ret = EOVERFLOW;
880 			break;
881 		}
882 		bcopy(&i40e->i40e_link_state, pr_val, sizeof (link_state_t));
883 		break;
884 	case MAC_PROP_AUTONEG:
885 		if (pr_valsize < sizeof (uint8_t)) {
886 			ret = EOVERFLOW;
887 			break;
888 		}
889 		u8 = pr_val;
890 		*u8 = 1;
891 		break;
892 	case MAC_PROP_FLOWCTRL:
893 		/*
894 		 * Because we don't currently support hardware flow control, we
895 		 * just hardcode this to be none.
896 		 */
897 		if (pr_valsize < sizeof (link_flowctrl_t)) {
898 			ret = EOVERFLOW;
899 			break;
900 		}
901 		fctl = LINK_FLOWCTRL_NONE;
902 		bcopy(&fctl, pr_val, sizeof (link_flowctrl_t));
903 		break;
904 	case MAC_PROP_MTU:
905 		if (pr_valsize < sizeof (uint32_t)) {
906 			ret = EOVERFLOW;
907 			break;
908 		}
909 		bcopy(&i40e->i40e_sdu, pr_val, sizeof (uint32_t));
910 		break;
911 
912 	/*
913 	 * Because we don't let users control the speeds we may auto-negotiate
914 	 * to, the values of the ADV_ and EN_ will always be the same.
915 	 */
916 	case MAC_PROP_ADV_100FDX_CAP:
917 	case MAC_PROP_EN_100FDX_CAP:
918 		if (pr_valsize < sizeof (uint8_t)) {
919 			ret = EOVERFLOW;
920 			break;
921 		}
922 		u8 = pr_val;
923 		*u8 = (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_100MB) != 0;
924 		break;
925 	case MAC_PROP_ADV_1000FDX_CAP:
926 	case MAC_PROP_EN_1000FDX_CAP:
927 		if (pr_valsize < sizeof (uint8_t)) {
928 			ret = EOVERFLOW;
929 			break;
930 		}
931 		u8 = pr_val;
932 		*u8 = (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_1GB) != 0;
933 		break;
934 	case MAC_PROP_ADV_10GFDX_CAP:
935 	case MAC_PROP_EN_10GFDX_CAP:
936 		if (pr_valsize < sizeof (uint8_t)) {
937 			ret = EOVERFLOW;
938 			break;
939 		}
940 		u8 = pr_val;
941 		*u8 = (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_10GB) != 0;
942 		break;
943 	case MAC_PROP_ADV_25GFDX_CAP:
944 	case MAC_PROP_EN_25GFDX_CAP:
945 		if (pr_valsize < sizeof (uint8_t)) {
946 			ret = EOVERFLOW;
947 			break;
948 		}
949 		u8 = pr_val;
950 		*u8 = (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_25GB) != 0;
951 		break;
952 	case MAC_PROP_ADV_40GFDX_CAP:
953 	case MAC_PROP_EN_40GFDX_CAP:
954 		if (pr_valsize < sizeof (uint8_t)) {
955 			ret = EOVERFLOW;
956 			break;
957 		}
958 		u8 = pr_val;
959 		*u8 = (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_40GB) != 0;
960 		break;
961 	case MAC_PROP_PRIVATE:
962 		ret = i40e_m_getprop_private(i40e, pr_name, pr_valsize, pr_val);
963 		break;
964 	default:
965 		ret = ENOTSUP;
966 		break;
967 	}
968 
969 	mutex_exit(&i40e->i40e_general_lock);
970 
971 	return (ret);
972 }
973 
974 static void
975 i40e_m_propinfo(void *arg, const char *pr_name, mac_prop_id_t pr_num,
976     mac_prop_info_handle_t prh)
977 {
978 	i40e_t *i40e = arg;
979 
980 	mutex_enter(&i40e->i40e_general_lock);
981 
982 	switch (pr_num) {
983 	case MAC_PROP_DUPLEX:
984 	case MAC_PROP_SPEED:
985 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
986 		break;
987 	case MAC_PROP_FLOWCTRL:
988 		/*
989 		 * At the moment, the driver doesn't support flow control, hence
990 		 * why this is set to read-only and none.
991 		 */
992 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
993 		mac_prop_info_set_default_link_flowctrl(prh,
994 		    LINK_FLOWCTRL_NONE);
995 		break;
996 	case MAC_PROP_MTU:
997 		mac_prop_info_set_range_uint32(prh, I40E_MIN_MTU, I40E_MAX_MTU);
998 		break;
999 
1000 	/*
1001 	 * We set the defaults for these based upon the phy's ability to
1002 	 * support the speeds. Note, auto-negotiation is required for fiber,
1003 	 * hence it is read-only and always enabled. When we have access to
1004 	 * copper phys we can revisit this.
1005 	 */
1006 	case MAC_PROP_AUTONEG:
1007 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1008 		mac_prop_info_set_default_uint8(prh, 1);
1009 		break;
1010 	case MAC_PROP_ADV_100FDX_CAP:
1011 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1012 		mac_prop_info_set_default_uint8(prh,
1013 		    (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_100MB) != 0);
1014 		break;
1015 	case MAC_PROP_EN_100FDX_CAP:
1016 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1017 		mac_prop_info_set_default_uint8(prh,
1018 		    (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_100MB) != 0);
1019 		break;
1020 	case MAC_PROP_ADV_1000FDX_CAP:
1021 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1022 		mac_prop_info_set_default_uint8(prh,
1023 		    (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_1GB) != 0);
1024 		break;
1025 	case MAC_PROP_EN_1000FDX_CAP:
1026 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1027 		mac_prop_info_set_default_uint8(prh,
1028 		    (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_1GB) != 0);
1029 		break;
1030 	case MAC_PROP_ADV_10GFDX_CAP:
1031 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1032 		mac_prop_info_set_default_uint8(prh,
1033 		    (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_10GB) != 0);
1034 		break;
1035 	case MAC_PROP_EN_10GFDX_CAP:
1036 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1037 		mac_prop_info_set_default_uint8(prh,
1038 		    (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_10GB) != 0);
1039 		break;
1040 	case MAC_PROP_ADV_25GFDX_CAP:
1041 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1042 		mac_prop_info_set_default_uint8(prh,
1043 		    (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_25GB) != 0);
1044 		break;
1045 	case MAC_PROP_EN_25GFDX_CAP:
1046 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1047 		mac_prop_info_set_default_uint8(prh,
1048 		    (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_25GB) != 0);
1049 		break;
1050 	case MAC_PROP_ADV_40GFDX_CAP:
1051 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1052 		mac_prop_info_set_default_uint8(prh,
1053 		    (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_40GB) != 0);
1054 		break;
1055 	case MAC_PROP_EN_40GFDX_CAP:
1056 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
1057 		mac_prop_info_set_default_uint8(prh,
1058 		    (i40e->i40e_phy.link_speed & I40E_LINK_SPEED_40GB) != 0);
1059 		break;
1060 	case MAC_PROP_PRIVATE:
1061 		i40e_m_propinfo_private(i40e, pr_name, prh);
1062 		break;
1063 	default:
1064 		break;
1065 	}
1066 
1067 	mutex_exit(&i40e->i40e_general_lock);
1068 }
1069 
1070 #define	I40E_M_CALLBACK_FLAGS \
1071 	(MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP | MC_PROPINFO)
1072 
1073 static mac_callbacks_t i40e_m_callbacks = {
1074 	I40E_M_CALLBACK_FLAGS,
1075 	i40e_m_stat,
1076 	i40e_m_start,
1077 	i40e_m_stop,
1078 	i40e_m_promisc,
1079 	i40e_m_multicast,
1080 	NULL,
1081 	NULL,
1082 	NULL,
1083 	i40e_m_ioctl,
1084 	i40e_m_getcapab,
1085 	NULL,
1086 	NULL,
1087 	i40e_m_setprop,
1088 	i40e_m_getprop,
1089 	i40e_m_propinfo
1090 };
1091 
1092 boolean_t
1093 i40e_register_mac(i40e_t *i40e)
1094 {
1095 	struct i40e_hw *hw = &i40e->i40e_hw_space;
1096 	int status;
1097 	mac_register_t *mac = mac_alloc(MAC_VERSION);
1098 
1099 	if (mac == NULL)
1100 		return (B_FALSE);
1101 
1102 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
1103 	mac->m_driver = i40e;
1104 	mac->m_dip = i40e->i40e_dip;
1105 	mac->m_src_addr = hw->mac.addr;
1106 	mac->m_callbacks = &i40e_m_callbacks;
1107 	mac->m_min_sdu = 0;
1108 	mac->m_max_sdu = i40e->i40e_sdu;
1109 	mac->m_margin = VLAN_TAGSZ;
1110 	mac->m_priv_props = i40e_priv_props;
1111 	mac->m_v12n = MAC_VIRT_LEVEL1;
1112 
1113 	status = mac_register(mac, &i40e->i40e_mac_hdl);
1114 	if (status != 0)
1115 		i40e_error(i40e, "mac_register() returned %d", status);
1116 	mac_free(mac);
1117 
1118 	return (status == 0);
1119 }
1120