xref: /freebsd/sys/dev/mlx5/mlx5_en/mlx5_en_main.c (revision eaa797943eeac5614edfdc8f6309f332343c3dd2)
1 /*-
2  * Copyright (c) 2015 Mellanox Technologies. All rights reserved.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  * 1. Redistributions of source code must retain the above copyright
8  *    notice, this list of conditions and the following disclaimer.
9  * 2. Redistributions in binary form must reproduce the above copyright
10  *    notice, this list of conditions and the following disclaimer in the
11  *    documentation and/or other materials provided with the distribution.
12  *
13  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND
14  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16  * ARE DISCLAIMED.  IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
17  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
23  * SUCH DAMAGE.
24  *
25  * $FreeBSD$
26  */
27 
28 #include "en.h"
29 
30 #include <sys/sockio.h>
31 #include <machine/atomic.h>
32 
33 #define	ETH_DRIVER_VERSION	"3.1.0-dev"
34 char mlx5e_version[] = "Mellanox Ethernet driver"
35     " (" ETH_DRIVER_VERSION ")";
36 
37 struct mlx5e_channel_param {
38 	struct mlx5e_rq_param rq;
39 	struct mlx5e_sq_param sq;
40 	struct mlx5e_cq_param rx_cq;
41 	struct mlx5e_cq_param tx_cq;
42 };
43 
44 static const struct {
45 	u32	subtype;
46 	u64	baudrate;
47 }	mlx5e_mode_table[MLX5E_LINK_MODES_NUMBER] = {
48 
49 	[MLX5E_1000BASE_CX_SGMII] = {
50 		.subtype = IFM_1000_CX_SGMII,
51 		.baudrate = IF_Mbps(1000ULL),
52 	},
53 	[MLX5E_1000BASE_KX] = {
54 		.subtype = IFM_1000_KX,
55 		.baudrate = IF_Mbps(1000ULL),
56 	},
57 	[MLX5E_10GBASE_CX4] = {
58 		.subtype = IFM_10G_CX4,
59 		.baudrate = IF_Gbps(10ULL),
60 	},
61 	[MLX5E_10GBASE_KX4] = {
62 		.subtype = IFM_10G_KX4,
63 		.baudrate = IF_Gbps(10ULL),
64 	},
65 	[MLX5E_10GBASE_KR] = {
66 		.subtype = IFM_10G_KR,
67 		.baudrate = IF_Gbps(10ULL),
68 	},
69 	[MLX5E_20GBASE_KR2] = {
70 		.subtype = IFM_20G_KR2,
71 		.baudrate = IF_Gbps(20ULL),
72 	},
73 	[MLX5E_40GBASE_CR4] = {
74 		.subtype = IFM_40G_CR4,
75 		.baudrate = IF_Gbps(40ULL),
76 	},
77 	[MLX5E_40GBASE_KR4] = {
78 		.subtype = IFM_40G_KR4,
79 		.baudrate = IF_Gbps(40ULL),
80 	},
81 	[MLX5E_56GBASE_R4] = {
82 		.subtype = IFM_56G_R4,
83 		.baudrate = IF_Gbps(56ULL),
84 	},
85 	[MLX5E_10GBASE_CR] = {
86 		.subtype = IFM_10G_CR1,
87 		.baudrate = IF_Gbps(10ULL),
88 	},
89 	[MLX5E_10GBASE_SR] = {
90 		.subtype = IFM_10G_SR,
91 		.baudrate = IF_Gbps(10ULL),
92 	},
93 	[MLX5E_10GBASE_LR] = {
94 		.subtype = IFM_10G_LR,
95 		.baudrate = IF_Gbps(10ULL),
96 	},
97 	[MLX5E_40GBASE_SR4] = {
98 		.subtype = IFM_40G_SR4,
99 		.baudrate = IF_Gbps(40ULL),
100 	},
101 	[MLX5E_40GBASE_LR4] = {
102 		.subtype = IFM_40G_LR4,
103 		.baudrate = IF_Gbps(40ULL),
104 	},
105 	[MLX5E_100GBASE_CR4] = {
106 		.subtype = IFM_100G_CR4,
107 		.baudrate = IF_Gbps(100ULL),
108 	},
109 	[MLX5E_100GBASE_SR4] = {
110 		.subtype = IFM_100G_SR4,
111 		.baudrate = IF_Gbps(100ULL),
112 	},
113 	[MLX5E_100GBASE_KR4] = {
114 		.subtype = IFM_100G_KR4,
115 		.baudrate = IF_Gbps(100ULL),
116 	},
117 	[MLX5E_100GBASE_LR4] = {
118 		.subtype = IFM_100G_LR4,
119 		.baudrate = IF_Gbps(100ULL),
120 	},
121 	[MLX5E_100BASE_TX] = {
122 		.subtype = IFM_100_TX,
123 		.baudrate = IF_Mbps(100ULL),
124 	},
125 	[MLX5E_100BASE_T] = {
126 		.subtype = IFM_100_T,
127 		.baudrate = IF_Mbps(100ULL),
128 	},
129 	[MLX5E_10GBASE_T] = {
130 		.subtype = IFM_10G_T,
131 		.baudrate = IF_Gbps(10ULL),
132 	},
133 	[MLX5E_25GBASE_CR] = {
134 		.subtype = IFM_25G_CR,
135 		.baudrate = IF_Gbps(25ULL),
136 	},
137 	[MLX5E_25GBASE_KR] = {
138 		.subtype = IFM_25G_KR,
139 		.baudrate = IF_Gbps(25ULL),
140 	},
141 	[MLX5E_25GBASE_SR] = {
142 		.subtype = IFM_25G_SR,
143 		.baudrate = IF_Gbps(25ULL),
144 	},
145 	[MLX5E_50GBASE_CR2] = {
146 		.subtype = IFM_50G_CR2,
147 		.baudrate = IF_Gbps(50ULL),
148 	},
149 	[MLX5E_50GBASE_KR2] = {
150 		.subtype = IFM_50G_KR2,
151 		.baudrate = IF_Gbps(50ULL),
152 	},
153 };
154 
155 MALLOC_DEFINE(M_MLX5EN, "MLX5EN", "MLX5 Ethernet");
156 
157 static void
158 mlx5e_update_carrier(struct mlx5e_priv *priv)
159 {
160 	struct mlx5_core_dev *mdev = priv->mdev;
161 	u32 out[MLX5_ST_SZ_DW(ptys_reg)];
162 	u32 eth_proto_oper;
163 	int error;
164 	u8 port_state;
165 	u8 i;
166 
167 	port_state = mlx5_query_vport_state(mdev,
168 	    MLX5_QUERY_VPORT_STATE_IN_OP_MOD_VNIC_VPORT, 0);
169 
170 	if (port_state == VPORT_STATE_UP) {
171 		priv->media_status_last |= IFM_ACTIVE;
172 	} else {
173 		priv->media_status_last &= ~IFM_ACTIVE;
174 		priv->media_active_last = IFM_ETHER;
175 		if_link_state_change(priv->ifp, LINK_STATE_DOWN);
176 		return;
177 	}
178 
179 	error = mlx5_query_port_ptys(mdev, out, sizeof(out), MLX5_PTYS_EN);
180 	if (error) {
181 		priv->media_active_last = IFM_ETHER;
182 		priv->ifp->if_baudrate = 1;
183 		if_printf(priv->ifp, "%s: query port ptys failed: 0x%x\n",
184 		    __func__, error);
185 		return;
186 	}
187 	eth_proto_oper = MLX5_GET(ptys_reg, out, eth_proto_oper);
188 
189 	for (i = 0; i != MLX5E_LINK_MODES_NUMBER; i++) {
190 		if (mlx5e_mode_table[i].baudrate == 0)
191 			continue;
192 		if (MLX5E_PROT_MASK(i) & eth_proto_oper) {
193 			priv->ifp->if_baudrate =
194 			    mlx5e_mode_table[i].baudrate;
195 			priv->media_active_last =
196 			    mlx5e_mode_table[i].subtype | IFM_ETHER | IFM_FDX;
197 		}
198 	}
199 	if_link_state_change(priv->ifp, LINK_STATE_UP);
200 }
201 
202 static void
203 mlx5e_media_status(struct ifnet *dev, struct ifmediareq *ifmr)
204 {
205 	struct mlx5e_priv *priv = dev->if_softc;
206 
207 	ifmr->ifm_status = priv->media_status_last;
208 	ifmr->ifm_active = priv->media_active_last |
209 	    (priv->params.rx_pauseframe_control ? IFM_ETH_RXPAUSE : 0) |
210 	    (priv->params.tx_pauseframe_control ? IFM_ETH_TXPAUSE : 0);
211 
212 }
213 
214 static u32
215 mlx5e_find_link_mode(u32 subtype)
216 {
217 	u32 i;
218 	u32 link_mode = 0;
219 
220 	for (i = 0; i < MLX5E_LINK_MODES_NUMBER; ++i) {
221 		if (mlx5e_mode_table[i].baudrate == 0)
222 			continue;
223 		if (mlx5e_mode_table[i].subtype == subtype)
224 			link_mode |= MLX5E_PROT_MASK(i);
225 	}
226 
227 	return (link_mode);
228 }
229 
230 static int
231 mlx5e_media_change(struct ifnet *dev)
232 {
233 	struct mlx5e_priv *priv = dev->if_softc;
234 	struct mlx5_core_dev *mdev = priv->mdev;
235 	u32 eth_proto_cap;
236 	u32 link_mode;
237 	int was_opened;
238 	int locked;
239 	int error;
240 
241 	locked = PRIV_LOCKED(priv);
242 	if (!locked)
243 		PRIV_LOCK(priv);
244 
245 	if (IFM_TYPE(priv->media.ifm_media) != IFM_ETHER) {
246 		error = EINVAL;
247 		goto done;
248 	}
249 	link_mode = mlx5e_find_link_mode(IFM_SUBTYPE(priv->media.ifm_media));
250 
251 	/* query supported capabilities */
252 	error = mlx5_query_port_proto_cap(mdev, &eth_proto_cap, MLX5_PTYS_EN);
253 	if (error != 0) {
254 		if_printf(dev, "Query port media capability failed\n");
255 		goto done;
256 	}
257 	/* check for autoselect */
258 	if (IFM_SUBTYPE(priv->media.ifm_media) == IFM_AUTO) {
259 		link_mode = eth_proto_cap;
260 		if (link_mode == 0) {
261 			if_printf(dev, "Port media capability is zero\n");
262 			error = EINVAL;
263 			goto done;
264 		}
265 	} else {
266 		link_mode = link_mode & eth_proto_cap;
267 		if (link_mode == 0) {
268 			if_printf(dev, "Not supported link mode requested\n");
269 			error = EINVAL;
270 			goto done;
271 		}
272 	}
273 	/* update pauseframe control bits */
274 	priv->params.rx_pauseframe_control =
275 	    (priv->media.ifm_media & IFM_ETH_RXPAUSE) ? 1 : 0;
276 	priv->params.tx_pauseframe_control =
277 	    (priv->media.ifm_media & IFM_ETH_TXPAUSE) ? 1 : 0;
278 
279 	/* check if device is opened */
280 	was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
281 
282 	/* reconfigure the hardware */
283 	mlx5_set_port_status(mdev, MLX5_PORT_DOWN);
284 	mlx5_set_port_proto(mdev, link_mode, MLX5_PTYS_EN);
285 	mlx5_set_port_pause(mdev, 1,
286 	    priv->params.rx_pauseframe_control,
287 	    priv->params.tx_pauseframe_control);
288 	if (was_opened)
289 		mlx5_set_port_status(mdev, MLX5_PORT_UP);
290 
291 done:
292 	if (!locked)
293 		PRIV_UNLOCK(priv);
294 	return (error);
295 }
296 
297 static void
298 mlx5e_update_carrier_work(struct work_struct *work)
299 {
300 	struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
301 	    update_carrier_work);
302 
303 	PRIV_LOCK(priv);
304 	if (test_bit(MLX5E_STATE_OPENED, &priv->state))
305 		mlx5e_update_carrier(priv);
306 	PRIV_UNLOCK(priv);
307 }
308 
309 static void
310 mlx5e_update_pport_counters(struct mlx5e_priv *priv)
311 {
312 	struct mlx5_core_dev *mdev = priv->mdev;
313 	struct mlx5e_pport_stats *s = &priv->stats.pport;
314 	struct mlx5e_port_stats_debug *s_debug = &priv->stats.port_stats_debug;
315 	u32 *in;
316 	u32 *out;
317 	u64 *ptr;
318 	unsigned sz = MLX5_ST_SZ_BYTES(ppcnt_reg);
319 	unsigned x;
320 	unsigned y;
321 
322 	in = mlx5_vzalloc(sz);
323 	out = mlx5_vzalloc(sz);
324 	if (in == NULL || out == NULL)
325 		goto free_out;
326 
327 	ptr = (uint64_t *)MLX5_ADDR_OF(ppcnt_reg, out, counter_set);
328 
329 	MLX5_SET(ppcnt_reg, in, local_port, 1);
330 
331 	MLX5_SET(ppcnt_reg, in, grp, MLX5_IEEE_802_3_COUNTERS_GROUP);
332 	mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
333 	for (x = y = 0; x != MLX5E_PPORT_IEEE802_3_STATS_NUM; x++, y++)
334 		s->arg[y] = be64toh(ptr[x]);
335 
336 	MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2819_COUNTERS_GROUP);
337 	mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
338 	for (x = 0; x != MLX5E_PPORT_RFC2819_STATS_NUM; x++, y++)
339 		s->arg[y] = be64toh(ptr[x]);
340 	for (y = 0; x != MLX5E_PPORT_RFC2819_STATS_NUM +
341 	    MLX5E_PPORT_RFC2819_STATS_DEBUG_NUM; x++, y++)
342 		s_debug->arg[y] = be64toh(ptr[x]);
343 
344 	MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2863_COUNTERS_GROUP);
345 	mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
346 	for (x = 0; x != MLX5E_PPORT_RFC2863_STATS_DEBUG_NUM; x++, y++)
347 		s_debug->arg[y] = be64toh(ptr[x]);
348 
349 	MLX5_SET(ppcnt_reg, in, grp, MLX5_PHYSICAL_LAYER_COUNTERS_GROUP);
350 	mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
351 	for (x = 0; x != MLX5E_PPORT_PHYSICAL_LAYER_STATS_DEBUG_NUM; x++, y++)
352 		s_debug->arg[y] = be64toh(ptr[x]);
353 free_out:
354 	kvfree(in);
355 	kvfree(out);
356 }
357 
358 static void
359 mlx5e_update_stats_work(struct work_struct *work)
360 {
361 	struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
362 	    update_stats_work);
363 	struct mlx5_core_dev *mdev = priv->mdev;
364 	struct mlx5e_vport_stats *s = &priv->stats.vport;
365 	struct mlx5e_rq_stats *rq_stats;
366 	struct mlx5e_sq_stats *sq_stats;
367 	struct buf_ring *sq_br;
368 #if (__FreeBSD_version < 1100000)
369 	struct ifnet *ifp = priv->ifp;
370 #endif
371 
372 	u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)];
373 	u32 *out;
374 	int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
375 	u64 tso_packets = 0;
376 	u64 tso_bytes = 0;
377 	u64 tx_queue_dropped = 0;
378 	u64 tx_defragged = 0;
379 	u64 tx_offload_none = 0;
380 	u64 lro_packets = 0;
381 	u64 lro_bytes = 0;
382 	u64 sw_lro_queued = 0;
383 	u64 sw_lro_flushed = 0;
384 	u64 rx_csum_none = 0;
385 	u64 rx_wqe_err = 0;
386 	u32 rx_out_of_buffer = 0;
387 	int i;
388 	int j;
389 
390 	PRIV_LOCK(priv);
391 	out = mlx5_vzalloc(outlen);
392 	if (out == NULL)
393 		goto free_out;
394 	if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
395 		goto free_out;
396 
397 	/* Collect firts the SW counters and then HW for consistency */
398 	for (i = 0; i < priv->params.num_channels; i++) {
399 		struct mlx5e_rq *rq = &priv->channel[i]->rq;
400 
401 		rq_stats = &priv->channel[i]->rq.stats;
402 
403 		/* collect stats from LRO */
404 		rq_stats->sw_lro_queued = rq->lro.lro_queued;
405 		rq_stats->sw_lro_flushed = rq->lro.lro_flushed;
406 		sw_lro_queued += rq_stats->sw_lro_queued;
407 		sw_lro_flushed += rq_stats->sw_lro_flushed;
408 		lro_packets += rq_stats->lro_packets;
409 		lro_bytes += rq_stats->lro_bytes;
410 		rx_csum_none += rq_stats->csum_none;
411 		rx_wqe_err += rq_stats->wqe_err;
412 
413 		for (j = 0; j < priv->num_tc; j++) {
414 			sq_stats = &priv->channel[i]->sq[j].stats;
415 			sq_br = priv->channel[i]->sq[j].br;
416 
417 			tso_packets += sq_stats->tso_packets;
418 			tso_bytes += sq_stats->tso_bytes;
419 			tx_queue_dropped += sq_stats->dropped;
420 			tx_queue_dropped += sq_br->br_drops;
421 			tx_defragged += sq_stats->defragged;
422 			tx_offload_none += sq_stats->csum_offload_none;
423 		}
424 	}
425 
426 	/* update counters */
427 	s->tso_packets = tso_packets;
428 	s->tso_bytes = tso_bytes;
429 	s->tx_queue_dropped = tx_queue_dropped;
430 	s->tx_defragged = tx_defragged;
431 	s->lro_packets = lro_packets;
432 	s->lro_bytes = lro_bytes;
433 	s->sw_lro_queued = sw_lro_queued;
434 	s->sw_lro_flushed = sw_lro_flushed;
435 	s->rx_csum_none = rx_csum_none;
436 	s->rx_wqe_err = rx_wqe_err;
437 
438 	/* HW counters */
439 	memset(in, 0, sizeof(in));
440 
441 	MLX5_SET(query_vport_counter_in, in, opcode,
442 	    MLX5_CMD_OP_QUERY_VPORT_COUNTER);
443 	MLX5_SET(query_vport_counter_in, in, op_mod, 0);
444 	MLX5_SET(query_vport_counter_in, in, other_vport, 0);
445 
446 	memset(out, 0, outlen);
447 
448 	/* get number of out-of-buffer drops first */
449 	if (mlx5_vport_query_out_of_rx_buffer(mdev, priv->counter_set_id,
450 	    &rx_out_of_buffer))
451 		goto free_out;
452 
453 	/* accumulate difference into a 64-bit counter */
454 	s->rx_out_of_buffer += (u64)(u32)(rx_out_of_buffer - s->rx_out_of_buffer_prev);
455 	s->rx_out_of_buffer_prev = rx_out_of_buffer;
456 
457 	/* get port statistics */
458 	if (mlx5_cmd_exec(mdev, in, sizeof(in), out, outlen))
459 		goto free_out;
460 
461 #define	MLX5_GET_CTR(out, x) \
462 	MLX5_GET64(query_vport_counter_out, out, x)
463 
464 	s->rx_error_packets =
465 	    MLX5_GET_CTR(out, received_errors.packets);
466 	s->rx_error_bytes =
467 	    MLX5_GET_CTR(out, received_errors.octets);
468 	s->tx_error_packets =
469 	    MLX5_GET_CTR(out, transmit_errors.packets);
470 	s->tx_error_bytes =
471 	    MLX5_GET_CTR(out, transmit_errors.octets);
472 
473 	s->rx_unicast_packets =
474 	    MLX5_GET_CTR(out, received_eth_unicast.packets);
475 	s->rx_unicast_bytes =
476 	    MLX5_GET_CTR(out, received_eth_unicast.octets);
477 	s->tx_unicast_packets =
478 	    MLX5_GET_CTR(out, transmitted_eth_unicast.packets);
479 	s->tx_unicast_bytes =
480 	    MLX5_GET_CTR(out, transmitted_eth_unicast.octets);
481 
482 	s->rx_multicast_packets =
483 	    MLX5_GET_CTR(out, received_eth_multicast.packets);
484 	s->rx_multicast_bytes =
485 	    MLX5_GET_CTR(out, received_eth_multicast.octets);
486 	s->tx_multicast_packets =
487 	    MLX5_GET_CTR(out, transmitted_eth_multicast.packets);
488 	s->tx_multicast_bytes =
489 	    MLX5_GET_CTR(out, transmitted_eth_multicast.octets);
490 
491 	s->rx_broadcast_packets =
492 	    MLX5_GET_CTR(out, received_eth_broadcast.packets);
493 	s->rx_broadcast_bytes =
494 	    MLX5_GET_CTR(out, received_eth_broadcast.octets);
495 	s->tx_broadcast_packets =
496 	    MLX5_GET_CTR(out, transmitted_eth_broadcast.packets);
497 	s->tx_broadcast_bytes =
498 	    MLX5_GET_CTR(out, transmitted_eth_broadcast.octets);
499 
500 	s->rx_packets =
501 	    s->rx_unicast_packets +
502 	    s->rx_multicast_packets +
503 	    s->rx_broadcast_packets -
504 	    s->rx_out_of_buffer;
505 	s->rx_bytes =
506 	    s->rx_unicast_bytes +
507 	    s->rx_multicast_bytes +
508 	    s->rx_broadcast_bytes;
509 	s->tx_packets =
510 	    s->tx_unicast_packets +
511 	    s->tx_multicast_packets +
512 	    s->tx_broadcast_packets;
513 	s->tx_bytes =
514 	    s->tx_unicast_bytes +
515 	    s->tx_multicast_bytes +
516 	    s->tx_broadcast_bytes;
517 
518 	/* Update calculated offload counters */
519 	s->tx_csum_offload = s->tx_packets - tx_offload_none;
520 	s->rx_csum_good = s->rx_packets - s->rx_csum_none;
521 
522 	/* Update per port counters */
523 	mlx5e_update_pport_counters(priv);
524 
525 #if (__FreeBSD_version < 1100000)
526 	/* no get_counters interface in fbsd 10 */
527 	ifp->if_ipackets = s->rx_packets;
528 	ifp->if_ierrors = s->rx_error_packets;
529 	ifp->if_iqdrops = s->rx_out_of_buffer;
530 	ifp->if_opackets = s->tx_packets;
531 	ifp->if_oerrors = s->tx_error_packets;
532 	ifp->if_snd.ifq_drops = s->tx_queue_dropped;
533 	ifp->if_ibytes = s->rx_bytes;
534 	ifp->if_obytes = s->tx_bytes;
535 #endif
536 
537 free_out:
538 	kvfree(out);
539 	PRIV_UNLOCK(priv);
540 }
541 
542 static void
543 mlx5e_update_stats(void *arg)
544 {
545 	struct mlx5e_priv *priv = arg;
546 
547 	schedule_work(&priv->update_stats_work);
548 
549 	callout_reset(&priv->watchdog, hz, &mlx5e_update_stats, priv);
550 }
551 
552 static void
553 mlx5e_async_event_sub(struct mlx5e_priv *priv,
554     enum mlx5_dev_event event)
555 {
556 	switch (event) {
557 	case MLX5_DEV_EVENT_PORT_UP:
558 	case MLX5_DEV_EVENT_PORT_DOWN:
559 		schedule_work(&priv->update_carrier_work);
560 		break;
561 
562 	default:
563 		break;
564 	}
565 }
566 
567 static void
568 mlx5e_async_event(struct mlx5_core_dev *mdev, void *vpriv,
569     enum mlx5_dev_event event, unsigned long param)
570 {
571 	struct mlx5e_priv *priv = vpriv;
572 
573 	mtx_lock(&priv->async_events_mtx);
574 	if (test_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state))
575 		mlx5e_async_event_sub(priv, event);
576 	mtx_unlock(&priv->async_events_mtx);
577 }
578 
579 static void
580 mlx5e_enable_async_events(struct mlx5e_priv *priv)
581 {
582 	set_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
583 }
584 
585 static void
586 mlx5e_disable_async_events(struct mlx5e_priv *priv)
587 {
588 	mtx_lock(&priv->async_events_mtx);
589 	clear_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
590 	mtx_unlock(&priv->async_events_mtx);
591 }
592 
593 static const char *mlx5e_rq_stats_desc[] = {
594 	MLX5E_RQ_STATS(MLX5E_STATS_DESC)
595 };
596 
597 static int
598 mlx5e_create_rq(struct mlx5e_channel *c,
599     struct mlx5e_rq_param *param,
600     struct mlx5e_rq *rq)
601 {
602 	struct mlx5e_priv *priv = c->priv;
603 	struct mlx5_core_dev *mdev = priv->mdev;
604 	char buffer[16];
605 	void *rqc = param->rqc;
606 	void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
607 	int wq_sz;
608 	int err;
609 	int i;
610 
611 	/* Create DMA descriptor TAG */
612 	if ((err = -bus_dma_tag_create(
613 	    bus_get_dma_tag(mdev->pdev->dev.bsddev),
614 	    1,				/* any alignment */
615 	    0,				/* no boundary */
616 	    BUS_SPACE_MAXADDR,		/* lowaddr */
617 	    BUS_SPACE_MAXADDR,		/* highaddr */
618 	    NULL, NULL,			/* filter, filterarg */
619 	    MJUM16BYTES,		/* maxsize */
620 	    1,				/* nsegments */
621 	    MJUM16BYTES,		/* maxsegsize */
622 	    0,				/* flags */
623 	    NULL, NULL,			/* lockfunc, lockfuncarg */
624 	    &rq->dma_tag)))
625 		goto done;
626 
627 	err = mlx5_wq_ll_create(mdev, &param->wq, rqc_wq, &rq->wq,
628 	    &rq->wq_ctrl);
629 	if (err)
630 		goto err_free_dma_tag;
631 
632 	rq->wq.db = &rq->wq.db[MLX5_RCV_DBR];
633 
634 	if (priv->params.hw_lro_en) {
635 		rq->wqe_sz = priv->params.lro_wqe_sz;
636 	} else {
637 		rq->wqe_sz = MLX5E_SW2MB_MTU(priv->ifp->if_mtu);
638 	}
639 	if (rq->wqe_sz > MJUM16BYTES) {
640 		err = -ENOMEM;
641 		goto err_rq_wq_destroy;
642 	} else if (rq->wqe_sz > MJUM9BYTES) {
643 		rq->wqe_sz = MJUM16BYTES;
644 	} else if (rq->wqe_sz > MJUMPAGESIZE) {
645 		rq->wqe_sz = MJUM9BYTES;
646 	} else if (rq->wqe_sz > MCLBYTES) {
647 		rq->wqe_sz = MJUMPAGESIZE;
648 	} else {
649 		rq->wqe_sz = MCLBYTES;
650 	}
651 
652 	wq_sz = mlx5_wq_ll_get_size(&rq->wq);
653 
654 	err = -tcp_lro_init_args(&rq->lro, c->ifp, TCP_LRO_ENTRIES, wq_sz);
655 	if (err)
656 		goto err_rq_wq_destroy;
657 
658 	rq->mbuf = malloc(wq_sz * sizeof(rq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
659 	if (rq->mbuf == NULL) {
660 		err = -ENOMEM;
661 		goto err_lro_init;
662 	}
663 	for (i = 0; i != wq_sz; i++) {
664 		struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
665 		uint32_t byte_count = rq->wqe_sz - MLX5E_NET_IP_ALIGN;
666 
667 		err = -bus_dmamap_create(rq->dma_tag, 0, &rq->mbuf[i].dma_map);
668 		if (err != 0) {
669 			while (i--)
670 				bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
671 			goto err_rq_mbuf_free;
672 		}
673 		wqe->data.lkey = c->mkey_be;
674 		wqe->data.byte_count = cpu_to_be32(byte_count | MLX5_HW_START_PADDING);
675 	}
676 
677 	rq->ifp = c->ifp;
678 	rq->channel = c;
679 	rq->ix = c->ix;
680 
681 	snprintf(buffer, sizeof(buffer), "rxstat%d", c->ix);
682 	mlx5e_create_stats(&rq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
683 	    buffer, mlx5e_rq_stats_desc, MLX5E_RQ_STATS_NUM,
684 	    rq->stats.arg);
685 	return (0);
686 
687 err_rq_mbuf_free:
688 	free(rq->mbuf, M_MLX5EN);
689 err_lro_init:
690 	tcp_lro_free(&rq->lro);
691 err_rq_wq_destroy:
692 	mlx5_wq_destroy(&rq->wq_ctrl);
693 err_free_dma_tag:
694 	bus_dma_tag_destroy(rq->dma_tag);
695 done:
696 	return (err);
697 }
698 
699 static void
700 mlx5e_destroy_rq(struct mlx5e_rq *rq)
701 {
702 	int wq_sz;
703 	int i;
704 
705 	/* destroy all sysctl nodes */
706 	sysctl_ctx_free(&rq->stats.ctx);
707 
708 	/* free leftover LRO packets, if any */
709 	tcp_lro_free(&rq->lro);
710 
711 	wq_sz = mlx5_wq_ll_get_size(&rq->wq);
712 	for (i = 0; i != wq_sz; i++) {
713 		if (rq->mbuf[i].mbuf != NULL) {
714 			bus_dmamap_unload(rq->dma_tag,
715 			    rq->mbuf[i].dma_map);
716 			m_freem(rq->mbuf[i].mbuf);
717 		}
718 		bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
719 	}
720 	free(rq->mbuf, M_MLX5EN);
721 	mlx5_wq_destroy(&rq->wq_ctrl);
722 }
723 
724 static int
725 mlx5e_enable_rq(struct mlx5e_rq *rq, struct mlx5e_rq_param *param)
726 {
727 	struct mlx5e_channel *c = rq->channel;
728 	struct mlx5e_priv *priv = c->priv;
729 	struct mlx5_core_dev *mdev = priv->mdev;
730 
731 	void *in;
732 	void *rqc;
733 	void *wq;
734 	int inlen;
735 	int err;
736 
737 	inlen = MLX5_ST_SZ_BYTES(create_rq_in) +
738 	    sizeof(u64) * rq->wq_ctrl.buf.npages;
739 	in = mlx5_vzalloc(inlen);
740 	if (in == NULL)
741 		return (-ENOMEM);
742 
743 	rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
744 	wq = MLX5_ADDR_OF(rqc, rqc, wq);
745 
746 	memcpy(rqc, param->rqc, sizeof(param->rqc));
747 
748 	MLX5_SET(rqc, rqc, cqn, c->rq.cq.mcq.cqn);
749 	MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST);
750 	MLX5_SET(rqc, rqc, flush_in_error_en, 1);
751 	if (priv->counter_set_id >= 0)
752 		MLX5_SET(rqc, rqc, counter_set_id, priv->counter_set_id);
753 	MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift -
754 	    PAGE_SHIFT);
755 	MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma);
756 
757 	mlx5_fill_page_array(&rq->wq_ctrl.buf,
758 	    (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
759 
760 	err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn);
761 
762 	kvfree(in);
763 
764 	return (err);
765 }
766 
767 static int
768 mlx5e_modify_rq(struct mlx5e_rq *rq, int curr_state, int next_state)
769 {
770 	struct mlx5e_channel *c = rq->channel;
771 	struct mlx5e_priv *priv = c->priv;
772 	struct mlx5_core_dev *mdev = priv->mdev;
773 
774 	void *in;
775 	void *rqc;
776 	int inlen;
777 	int err;
778 
779 	inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
780 	in = mlx5_vzalloc(inlen);
781 	if (in == NULL)
782 		return (-ENOMEM);
783 
784 	rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
785 
786 	MLX5_SET(modify_rq_in, in, rqn, rq->rqn);
787 	MLX5_SET(modify_rq_in, in, rq_state, curr_state);
788 	MLX5_SET(rqc, rqc, state, next_state);
789 
790 	err = mlx5_core_modify_rq(mdev, in, inlen);
791 
792 	kvfree(in);
793 
794 	return (err);
795 }
796 
797 static void
798 mlx5e_disable_rq(struct mlx5e_rq *rq)
799 {
800 	struct mlx5e_channel *c = rq->channel;
801 	struct mlx5e_priv *priv = c->priv;
802 	struct mlx5_core_dev *mdev = priv->mdev;
803 
804 	mlx5_core_destroy_rq(mdev, rq->rqn);
805 }
806 
807 static int
808 mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq)
809 {
810 	struct mlx5e_channel *c = rq->channel;
811 	struct mlx5e_priv *priv = c->priv;
812 	struct mlx5_wq_ll *wq = &rq->wq;
813 	int i;
814 
815 	for (i = 0; i < 1000; i++) {
816 		if (wq->cur_sz >= priv->params.min_rx_wqes)
817 			return (0);
818 
819 		msleep(4);
820 	}
821 	return (-ETIMEDOUT);
822 }
823 
824 static int
825 mlx5e_open_rq(struct mlx5e_channel *c,
826     struct mlx5e_rq_param *param,
827     struct mlx5e_rq *rq)
828 {
829 	int err;
830 
831 	err = mlx5e_create_rq(c, param, rq);
832 	if (err)
833 		return (err);
834 
835 	err = mlx5e_enable_rq(rq, param);
836 	if (err)
837 		goto err_destroy_rq;
838 
839 	err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
840 	if (err)
841 		goto err_disable_rq;
842 
843 	c->rq.enabled = 1;
844 
845 	return (0);
846 
847 err_disable_rq:
848 	mlx5e_disable_rq(rq);
849 err_destroy_rq:
850 	mlx5e_destroy_rq(rq);
851 
852 	return (err);
853 }
854 
855 static void
856 mlx5e_close_rq(struct mlx5e_rq *rq)
857 {
858 	rq->enabled = 0;
859 	mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
860 }
861 
862 static void
863 mlx5e_close_rq_wait(struct mlx5e_rq *rq)
864 {
865 	/* wait till RQ is empty */
866 	while (!mlx5_wq_ll_is_empty(&rq->wq)) {
867 		msleep(4);
868 		rq->cq.mcq.comp(&rq->cq.mcq);
869 	}
870 
871 	mlx5e_disable_rq(rq);
872 	mlx5e_destroy_rq(rq);
873 }
874 
875 void
876 mlx5e_free_sq_db(struct mlx5e_sq *sq)
877 {
878 	int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
879 	int x;
880 
881 	for (x = 0; x != wq_sz; x++)
882 		bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
883 	free(sq->mbuf, M_MLX5EN);
884 }
885 
886 int
887 mlx5e_alloc_sq_db(struct mlx5e_sq *sq)
888 {
889 	int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
890 	int err;
891 	int x;
892 
893 	sq->mbuf = malloc(wq_sz * sizeof(sq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
894 	if (sq->mbuf == NULL)
895 		return (-ENOMEM);
896 
897 	/* Create DMA descriptor MAPs */
898 	for (x = 0; x != wq_sz; x++) {
899 		err = -bus_dmamap_create(sq->dma_tag, 0, &sq->mbuf[x].dma_map);
900 		if (err != 0) {
901 			while (x--)
902 				bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
903 			free(sq->mbuf, M_MLX5EN);
904 			return (err);
905 		}
906 	}
907 	return (0);
908 }
909 
910 static const char *mlx5e_sq_stats_desc[] = {
911 	MLX5E_SQ_STATS(MLX5E_STATS_DESC)
912 };
913 
914 static int
915 mlx5e_create_sq(struct mlx5e_channel *c,
916     int tc,
917     struct mlx5e_sq_param *param,
918     struct mlx5e_sq *sq)
919 {
920 	struct mlx5e_priv *priv = c->priv;
921 	struct mlx5_core_dev *mdev = priv->mdev;
922 	char buffer[16];
923 
924 	void *sqc = param->sqc;
925 	void *sqc_wq = MLX5_ADDR_OF(sqc, sqc, wq);
926 #ifdef RSS
927 	cpuset_t cpu_mask;
928 	int cpu_id;
929 #endif
930 	int err;
931 
932 	/* Create DMA descriptor TAG */
933 	if ((err = -bus_dma_tag_create(
934 	    bus_get_dma_tag(mdev->pdev->dev.bsddev),
935 	    1,				/* any alignment */
936 	    0,				/* no boundary */
937 	    BUS_SPACE_MAXADDR,		/* lowaddr */
938 	    BUS_SPACE_MAXADDR,		/* highaddr */
939 	    NULL, NULL,			/* filter, filterarg */
940 	    MLX5E_MAX_TX_PAYLOAD_SIZE,	/* maxsize */
941 	    MLX5E_MAX_TX_MBUF_FRAGS,	/* nsegments */
942 	    MLX5E_MAX_TX_MBUF_SIZE,	/* maxsegsize */
943 	    0,				/* flags */
944 	    NULL, NULL,			/* lockfunc, lockfuncarg */
945 	    &sq->dma_tag)))
946 		goto done;
947 
948 	err = mlx5_alloc_map_uar(mdev, &sq->uar);
949 	if (err)
950 		goto err_free_dma_tag;
951 
952 	err = mlx5_wq_cyc_create(mdev, &param->wq, sqc_wq, &sq->wq,
953 	    &sq->wq_ctrl);
954 	if (err)
955 		goto err_unmap_free_uar;
956 
957 	sq->wq.db = &sq->wq.db[MLX5_SND_DBR];
958 	sq->bf_buf_size = (1 << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2;
959 
960 	err = mlx5e_alloc_sq_db(sq);
961 	if (err)
962 		goto err_sq_wq_destroy;
963 
964 	sq->mkey_be = c->mkey_be;
965 	sq->ifp = priv->ifp;
966 	sq->priv = priv;
967 	sq->tc = tc;
968 
969 	sq->br = buf_ring_alloc(MLX5E_SQ_TX_QUEUE_SIZE, M_MLX5EN,
970 	    M_WAITOK, &sq->lock);
971 	if (sq->br == NULL) {
972 		if_printf(c->ifp, "%s: Failed allocating sq drbr buffer\n",
973 		    __func__);
974 		err = -ENOMEM;
975 		goto err_free_sq_db;
976 	}
977 
978 	sq->sq_tq = taskqueue_create_fast("mlx5e_que", M_WAITOK,
979 	    taskqueue_thread_enqueue, &sq->sq_tq);
980 	if (sq->sq_tq == NULL) {
981 		if_printf(c->ifp, "%s: Failed allocating taskqueue\n",
982 		    __func__);
983 		err = -ENOMEM;
984 		goto err_free_drbr;
985 	}
986 
987 	TASK_INIT(&sq->sq_task, 0, mlx5e_tx_que, sq);
988 #ifdef RSS
989 	cpu_id = rss_getcpu(c->ix % rss_getnumbuckets());
990 	CPU_SETOF(cpu_id, &cpu_mask);
991 	taskqueue_start_threads_cpuset(&sq->sq_tq, 1, PI_NET, &cpu_mask,
992 	    "%s TX SQ%d.%d CPU%d", c->ifp->if_xname, c->ix, tc, cpu_id);
993 #else
994 	taskqueue_start_threads(&sq->sq_tq, 1, PI_NET,
995 	    "%s TX SQ%d.%d", c->ifp->if_xname, c->ix, tc);
996 #endif
997 	snprintf(buffer, sizeof(buffer), "txstat%dtc%d", c->ix, tc);
998 	mlx5e_create_stats(&sq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
999 	    buffer, mlx5e_sq_stats_desc, MLX5E_SQ_STATS_NUM,
1000 	    sq->stats.arg);
1001 
1002 	return (0);
1003 
1004 err_free_drbr:
1005 	buf_ring_free(sq->br, M_MLX5EN);
1006 err_free_sq_db:
1007 	mlx5e_free_sq_db(sq);
1008 err_sq_wq_destroy:
1009 	mlx5_wq_destroy(&sq->wq_ctrl);
1010 
1011 err_unmap_free_uar:
1012 	mlx5_unmap_free_uar(mdev, &sq->uar);
1013 
1014 err_free_dma_tag:
1015 	bus_dma_tag_destroy(sq->dma_tag);
1016 done:
1017 	return (err);
1018 }
1019 
1020 static void
1021 mlx5e_destroy_sq(struct mlx5e_sq *sq)
1022 {
1023 	/* destroy all sysctl nodes */
1024 	sysctl_ctx_free(&sq->stats.ctx);
1025 
1026 	mlx5e_free_sq_db(sq);
1027 	mlx5_wq_destroy(&sq->wq_ctrl);
1028 	mlx5_unmap_free_uar(sq->priv->mdev, &sq->uar);
1029 	taskqueue_drain(sq->sq_tq, &sq->sq_task);
1030 	taskqueue_free(sq->sq_tq);
1031 	buf_ring_free(sq->br, M_MLX5EN);
1032 }
1033 
1034 int
1035 mlx5e_enable_sq(struct mlx5e_sq *sq, struct mlx5e_sq_param *param,
1036     int tis_num)
1037 {
1038 	void *in;
1039 	void *sqc;
1040 	void *wq;
1041 	int inlen;
1042 	int err;
1043 
1044 	inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
1045 	    sizeof(u64) * sq->wq_ctrl.buf.npages;
1046 	in = mlx5_vzalloc(inlen);
1047 	if (in == NULL)
1048 		return (-ENOMEM);
1049 
1050 	sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
1051 	wq = MLX5_ADDR_OF(sqc, sqc, wq);
1052 
1053 	memcpy(sqc, param->sqc, sizeof(param->sqc));
1054 
1055 	MLX5_SET(sqc, sqc, tis_num_0, tis_num);
1056 	MLX5_SET(sqc, sqc, cqn, sq->cq.mcq.cqn);
1057 	MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
1058 	MLX5_SET(sqc, sqc, tis_lst_sz, 1);
1059 	MLX5_SET(sqc, sqc, flush_in_error_en, 1);
1060 
1061 	MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
1062 	MLX5_SET(wq, wq, uar_page, sq->uar.index);
1063 	MLX5_SET(wq, wq, log_wq_pg_sz, sq->wq_ctrl.buf.page_shift -
1064 	    PAGE_SHIFT);
1065 	MLX5_SET64(wq, wq, dbr_addr, sq->wq_ctrl.db.dma);
1066 
1067 	mlx5_fill_page_array(&sq->wq_ctrl.buf,
1068 	    (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
1069 
1070 	err = mlx5_core_create_sq(sq->priv->mdev, in, inlen, &sq->sqn);
1071 
1072 	kvfree(in);
1073 
1074 	return (err);
1075 }
1076 
1077 int
1078 mlx5e_modify_sq(struct mlx5e_sq *sq, int curr_state, int next_state)
1079 {
1080 	void *in;
1081 	void *sqc;
1082 	int inlen;
1083 	int err;
1084 
1085 	inlen = MLX5_ST_SZ_BYTES(modify_sq_in);
1086 	in = mlx5_vzalloc(inlen);
1087 	if (in == NULL)
1088 		return (-ENOMEM);
1089 
1090 	sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx);
1091 
1092 	MLX5_SET(modify_sq_in, in, sqn, sq->sqn);
1093 	MLX5_SET(modify_sq_in, in, sq_state, curr_state);
1094 	MLX5_SET(sqc, sqc, state, next_state);
1095 
1096 	err = mlx5_core_modify_sq(sq->priv->mdev, in, inlen);
1097 
1098 	kvfree(in);
1099 
1100 	return (err);
1101 }
1102 
1103 void
1104 mlx5e_disable_sq(struct mlx5e_sq *sq)
1105 {
1106 
1107 	mlx5_core_destroy_sq(sq->priv->mdev, sq->sqn);
1108 }
1109 
1110 static int
1111 mlx5e_open_sq(struct mlx5e_channel *c,
1112     int tc,
1113     struct mlx5e_sq_param *param,
1114     struct mlx5e_sq *sq)
1115 {
1116 	int err;
1117 
1118 	err = mlx5e_create_sq(c, tc, param, sq);
1119 	if (err)
1120 		return (err);
1121 
1122 	err = mlx5e_enable_sq(sq, param, c->priv->tisn[tc]);
1123 	if (err)
1124 		goto err_destroy_sq;
1125 
1126 	err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY);
1127 	if (err)
1128 		goto err_disable_sq;
1129 
1130 	atomic_store_rel_int(&sq->queue_state, MLX5E_SQ_READY);
1131 
1132 	return (0);
1133 
1134 err_disable_sq:
1135 	mlx5e_disable_sq(sq);
1136 err_destroy_sq:
1137 	mlx5e_destroy_sq(sq);
1138 
1139 	return (err);
1140 }
1141 
1142 static void
1143 mlx5e_sq_send_nops_locked(struct mlx5e_sq *sq, int can_sleep)
1144 {
1145 	/* fill up remainder with NOPs */
1146 	while (sq->cev_counter != 0) {
1147 		while (!mlx5e_sq_has_room_for(sq, 1)) {
1148 			if (can_sleep != 0) {
1149 				mtx_unlock(&sq->lock);
1150 				msleep(4);
1151 				mtx_lock(&sq->lock);
1152 			} else {
1153 				goto done;
1154 			}
1155 		}
1156 		/* send a single NOP */
1157 		mlx5e_send_nop(sq, 1);
1158 		wmb();
1159 	}
1160 done:
1161 	/* Check if we need to write the doorbell */
1162 	if (likely(sq->doorbell.d64 != 0)) {
1163 		mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
1164 		sq->doorbell.d64 = 0;
1165 	}
1166 	return;
1167 }
1168 
1169 void
1170 mlx5e_sq_cev_timeout(void *arg)
1171 {
1172 	struct mlx5e_sq *sq = arg;
1173 
1174 	mtx_assert(&sq->lock, MA_OWNED);
1175 
1176 	/* check next state */
1177 	switch (sq->cev_next_state) {
1178 	case MLX5E_CEV_STATE_SEND_NOPS:
1179 		/* fill TX ring with NOPs, if any */
1180 		mlx5e_sq_send_nops_locked(sq, 0);
1181 
1182 		/* check if completed */
1183 		if (sq->cev_counter == 0) {
1184 			sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
1185 			return;
1186 		}
1187 		break;
1188 	default:
1189 		/* send NOPs on next timeout */
1190 		sq->cev_next_state = MLX5E_CEV_STATE_SEND_NOPS;
1191 		break;
1192 	}
1193 
1194 	/* restart timer */
1195 	callout_reset_curcpu(&sq->cev_callout, hz, mlx5e_sq_cev_timeout, sq);
1196 }
1197 
1198 void
1199 mlx5e_drain_sq(struct mlx5e_sq *sq)
1200 {
1201 
1202 	mtx_lock(&sq->lock);
1203 	/* teardown event factor timer, if any */
1204 	sq->cev_next_state = MLX5E_CEV_STATE_HOLD_NOPS;
1205 	callout_stop(&sq->cev_callout);
1206 
1207 	/* send dummy NOPs in order to flush the transmit ring */
1208 	mlx5e_sq_send_nops_locked(sq, 1);
1209 	mtx_unlock(&sq->lock);
1210 
1211 	/* make sure it is safe to free the callout */
1212 	callout_drain(&sq->cev_callout);
1213 
1214 	/* error out remaining requests */
1215 	mlx5e_modify_sq(sq, MLX5_SQC_STATE_RDY, MLX5_SQC_STATE_ERR);
1216 
1217 	/* wait till SQ is empty */
1218 	mtx_lock(&sq->lock);
1219 	while (sq->cc != sq->pc) {
1220 		mtx_unlock(&sq->lock);
1221 		msleep(4);
1222 		sq->cq.mcq.comp(&sq->cq.mcq);
1223 		mtx_lock(&sq->lock);
1224 	}
1225 	mtx_unlock(&sq->lock);
1226 }
1227 
1228 static void
1229 mlx5e_close_sq_wait(struct mlx5e_sq *sq)
1230 {
1231 
1232 	mlx5e_drain_sq(sq);
1233 	mlx5e_disable_sq(sq);
1234 	mlx5e_destroy_sq(sq);
1235 }
1236 
1237 static int
1238 mlx5e_create_cq(struct mlx5e_priv *priv,
1239     struct mlx5e_cq_param *param,
1240     struct mlx5e_cq *cq,
1241     mlx5e_cq_comp_t *comp,
1242     int eq_ix)
1243 {
1244 	struct mlx5_core_dev *mdev = priv->mdev;
1245 	struct mlx5_core_cq *mcq = &cq->mcq;
1246 	int eqn_not_used;
1247 	int irqn;
1248 	int err;
1249 	u32 i;
1250 
1251 	param->wq.buf_numa_node = 0;
1252 	param->wq.db_numa_node = 0;
1253 
1254 	err = mlx5_cqwq_create(mdev, &param->wq, param->cqc, &cq->wq,
1255 	    &cq->wq_ctrl);
1256 	if (err)
1257 		return (err);
1258 
1259 	mlx5_vector2eqn(mdev, eq_ix, &eqn_not_used, &irqn);
1260 
1261 	mcq->cqe_sz = 64;
1262 	mcq->set_ci_db = cq->wq_ctrl.db.db;
1263 	mcq->arm_db = cq->wq_ctrl.db.db + 1;
1264 	*mcq->set_ci_db = 0;
1265 	*mcq->arm_db = 0;
1266 	mcq->vector = eq_ix;
1267 	mcq->comp = comp;
1268 	mcq->event = mlx5e_cq_error_event;
1269 	mcq->irqn = irqn;
1270 	mcq->uar = &priv->cq_uar;
1271 
1272 	for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) {
1273 		struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i);
1274 
1275 		cqe->op_own = 0xf1;
1276 	}
1277 
1278 	cq->priv = priv;
1279 
1280 	return (0);
1281 }
1282 
1283 static void
1284 mlx5e_destroy_cq(struct mlx5e_cq *cq)
1285 {
1286 	mlx5_wq_destroy(&cq->wq_ctrl);
1287 }
1288 
1289 static int
1290 mlx5e_enable_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param, int eq_ix)
1291 {
1292 	struct mlx5_core_cq *mcq = &cq->mcq;
1293 	void *in;
1294 	void *cqc;
1295 	int inlen;
1296 	int irqn_not_used;
1297 	int eqn;
1298 	int err;
1299 
1300 	inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
1301 	    sizeof(u64) * cq->wq_ctrl.buf.npages;
1302 	in = mlx5_vzalloc(inlen);
1303 	if (in == NULL)
1304 		return (-ENOMEM);
1305 
1306 	cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
1307 
1308 	memcpy(cqc, param->cqc, sizeof(param->cqc));
1309 
1310 	mlx5_fill_page_array(&cq->wq_ctrl.buf,
1311 	    (__be64 *) MLX5_ADDR_OF(create_cq_in, in, pas));
1312 
1313 	mlx5_vector2eqn(cq->priv->mdev, eq_ix, &eqn, &irqn_not_used);
1314 
1315 	MLX5_SET(cqc, cqc, c_eqn, eqn);
1316 	MLX5_SET(cqc, cqc, uar_page, mcq->uar->index);
1317 	MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.buf.page_shift -
1318 	    PAGE_SHIFT);
1319 	MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
1320 
1321 	err = mlx5_core_create_cq(cq->priv->mdev, mcq, in, inlen);
1322 
1323 	kvfree(in);
1324 
1325 	if (err)
1326 		return (err);
1327 
1328 	mlx5e_cq_arm(cq);
1329 
1330 	return (0);
1331 }
1332 
1333 static void
1334 mlx5e_disable_cq(struct mlx5e_cq *cq)
1335 {
1336 
1337 	mlx5_core_destroy_cq(cq->priv->mdev, &cq->mcq);
1338 }
1339 
1340 int
1341 mlx5e_open_cq(struct mlx5e_priv *priv,
1342     struct mlx5e_cq_param *param,
1343     struct mlx5e_cq *cq,
1344     mlx5e_cq_comp_t *comp,
1345     int eq_ix)
1346 {
1347 	int err;
1348 
1349 	err = mlx5e_create_cq(priv, param, cq, comp, eq_ix);
1350 	if (err)
1351 		return (err);
1352 
1353 	err = mlx5e_enable_cq(cq, param, eq_ix);
1354 	if (err)
1355 		goto err_destroy_cq;
1356 
1357 	return (0);
1358 
1359 err_destroy_cq:
1360 	mlx5e_destroy_cq(cq);
1361 
1362 	return (err);
1363 }
1364 
1365 void
1366 mlx5e_close_cq(struct mlx5e_cq *cq)
1367 {
1368 	mlx5e_disable_cq(cq);
1369 	mlx5e_destroy_cq(cq);
1370 }
1371 
1372 static int
1373 mlx5e_open_tx_cqs(struct mlx5e_channel *c,
1374     struct mlx5e_channel_param *cparam)
1375 {
1376 	int err;
1377 	int tc;
1378 
1379 	for (tc = 0; tc < c->num_tc; tc++) {
1380 		/* open completion queue */
1381 		err = mlx5e_open_cq(c->priv, &cparam->tx_cq, &c->sq[tc].cq,
1382 		    &mlx5e_tx_cq_comp, c->ix);
1383 		if (err)
1384 			goto err_close_tx_cqs;
1385 	}
1386 	return (0);
1387 
1388 err_close_tx_cqs:
1389 	for (tc--; tc >= 0; tc--)
1390 		mlx5e_close_cq(&c->sq[tc].cq);
1391 
1392 	return (err);
1393 }
1394 
1395 static void
1396 mlx5e_close_tx_cqs(struct mlx5e_channel *c)
1397 {
1398 	int tc;
1399 
1400 	for (tc = 0; tc < c->num_tc; tc++)
1401 		mlx5e_close_cq(&c->sq[tc].cq);
1402 }
1403 
1404 static int
1405 mlx5e_open_sqs(struct mlx5e_channel *c,
1406     struct mlx5e_channel_param *cparam)
1407 {
1408 	int err;
1409 	int tc;
1410 
1411 	for (tc = 0; tc < c->num_tc; tc++) {
1412 		err = mlx5e_open_sq(c, tc, &cparam->sq, &c->sq[tc]);
1413 		if (err)
1414 			goto err_close_sqs;
1415 	}
1416 
1417 	return (0);
1418 
1419 err_close_sqs:
1420 	for (tc--; tc >= 0; tc--)
1421 		mlx5e_close_sq_wait(&c->sq[tc]);
1422 
1423 	return (err);
1424 }
1425 
1426 static void
1427 mlx5e_close_sqs_wait(struct mlx5e_channel *c)
1428 {
1429 	int tc;
1430 
1431 	for (tc = 0; tc < c->num_tc; tc++)
1432 		mlx5e_close_sq_wait(&c->sq[tc]);
1433 }
1434 
1435 static void
1436 mlx5e_chan_mtx_init(struct mlx5e_channel *c)
1437 {
1438 	int tc;
1439 
1440 	mtx_init(&c->rq.mtx, "mlx5rx", MTX_NETWORK_LOCK, MTX_DEF);
1441 
1442 	for (tc = 0; tc < c->num_tc; tc++) {
1443 		struct mlx5e_sq *sq = c->sq + tc;
1444 
1445 		mtx_init(&sq->lock, "mlx5tx", MTX_NETWORK_LOCK, MTX_DEF);
1446 		mtx_init(&sq->comp_lock, "mlx5comp", MTX_NETWORK_LOCK,
1447 		    MTX_DEF);
1448 
1449 		callout_init_mtx(&sq->cev_callout, &sq->lock, 0);
1450 
1451 		sq->cev_factor = c->priv->params_ethtool.tx_completion_fact;
1452 
1453 		/* ensure the TX completion event factor is not zero */
1454 		if (sq->cev_factor == 0)
1455 			sq->cev_factor = 1;
1456 	}
1457 }
1458 
1459 static void
1460 mlx5e_chan_mtx_destroy(struct mlx5e_channel *c)
1461 {
1462 	int tc;
1463 
1464 	mtx_destroy(&c->rq.mtx);
1465 
1466 	for (tc = 0; tc < c->num_tc; tc++) {
1467 		mtx_destroy(&c->sq[tc].lock);
1468 		mtx_destroy(&c->sq[tc].comp_lock);
1469 	}
1470 }
1471 
1472 static int
1473 mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
1474     struct mlx5e_channel_param *cparam,
1475     struct mlx5e_channel *volatile *cp)
1476 {
1477 	struct mlx5e_channel *c;
1478 	int err;
1479 
1480 	c = malloc(sizeof(*c), M_MLX5EN, M_WAITOK | M_ZERO);
1481 	if (c == NULL)
1482 		return (-ENOMEM);
1483 
1484 	c->priv = priv;
1485 	c->ix = ix;
1486 	c->cpu = 0;
1487 	c->ifp = priv->ifp;
1488 	c->mkey_be = cpu_to_be32(priv->mr.key);
1489 	c->num_tc = priv->num_tc;
1490 
1491 	/* init mutexes */
1492 	mlx5e_chan_mtx_init(c);
1493 
1494 	/* open transmit completion queue */
1495 	err = mlx5e_open_tx_cqs(c, cparam);
1496 	if (err)
1497 		goto err_free;
1498 
1499 	/* open receive completion queue */
1500 	err = mlx5e_open_cq(c->priv, &cparam->rx_cq, &c->rq.cq,
1501 	    &mlx5e_rx_cq_comp, c->ix);
1502 	if (err)
1503 		goto err_close_tx_cqs;
1504 
1505 	err = mlx5e_open_sqs(c, cparam);
1506 	if (err)
1507 		goto err_close_rx_cq;
1508 
1509 	err = mlx5e_open_rq(c, &cparam->rq, &c->rq);
1510 	if (err)
1511 		goto err_close_sqs;
1512 
1513 	/* store channel pointer */
1514 	*cp = c;
1515 
1516 	/* poll receive queue initially */
1517 	c->rq.cq.mcq.comp(&c->rq.cq.mcq);
1518 
1519 	return (0);
1520 
1521 err_close_sqs:
1522 	mlx5e_close_sqs_wait(c);
1523 
1524 err_close_rx_cq:
1525 	mlx5e_close_cq(&c->rq.cq);
1526 
1527 err_close_tx_cqs:
1528 	mlx5e_close_tx_cqs(c);
1529 
1530 err_free:
1531 	/* destroy mutexes */
1532 	mlx5e_chan_mtx_destroy(c);
1533 	free(c, M_MLX5EN);
1534 	return (err);
1535 }
1536 
1537 static void
1538 mlx5e_close_channel(struct mlx5e_channel *volatile *pp)
1539 {
1540 	struct mlx5e_channel *c = *pp;
1541 
1542 	/* check if channel is already closed */
1543 	if (c == NULL)
1544 		return;
1545 	mlx5e_close_rq(&c->rq);
1546 }
1547 
1548 static void
1549 mlx5e_close_channel_wait(struct mlx5e_channel *volatile *pp)
1550 {
1551 	struct mlx5e_channel *c = *pp;
1552 
1553 	/* check if channel is already closed */
1554 	if (c == NULL)
1555 		return;
1556 	/* ensure channel pointer is no longer used */
1557 	*pp = NULL;
1558 
1559 	mlx5e_close_rq_wait(&c->rq);
1560 	mlx5e_close_sqs_wait(c);
1561 	mlx5e_close_cq(&c->rq.cq);
1562 	mlx5e_close_tx_cqs(c);
1563 	/* destroy mutexes */
1564 	mlx5e_chan_mtx_destroy(c);
1565 	free(c, M_MLX5EN);
1566 }
1567 
1568 static void
1569 mlx5e_build_rq_param(struct mlx5e_priv *priv,
1570     struct mlx5e_rq_param *param)
1571 {
1572 	void *rqc = param->rqc;
1573 	void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
1574 
1575 	MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
1576 	MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN);
1577 	MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe)));
1578 	MLX5_SET(wq, wq, log_wq_sz, priv->params.log_rq_size);
1579 	MLX5_SET(wq, wq, pd, priv->pdn);
1580 
1581 	param->wq.buf_numa_node = 0;
1582 	param->wq.db_numa_node = 0;
1583 	param->wq.linear = 1;
1584 }
1585 
1586 static void
1587 mlx5e_build_sq_param(struct mlx5e_priv *priv,
1588     struct mlx5e_sq_param *param)
1589 {
1590 	void *sqc = param->sqc;
1591 	void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
1592 
1593 	MLX5_SET(wq, wq, log_wq_sz, priv->params.log_sq_size);
1594 	MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
1595 	MLX5_SET(wq, wq, pd, priv->pdn);
1596 
1597 	param->wq.buf_numa_node = 0;
1598 	param->wq.db_numa_node = 0;
1599 	param->wq.linear = 1;
1600 }
1601 
1602 static void
1603 mlx5e_build_common_cq_param(struct mlx5e_priv *priv,
1604     struct mlx5e_cq_param *param)
1605 {
1606 	void *cqc = param->cqc;
1607 
1608 	MLX5_SET(cqc, cqc, uar_page, priv->cq_uar.index);
1609 }
1610 
1611 static void
1612 mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
1613     struct mlx5e_cq_param *param)
1614 {
1615 	void *cqc = param->cqc;
1616 
1617 
1618 	/*
1619 	 * TODO The sysctl to control on/off is a bool value for now, which means
1620 	 * we only support CSUM, once HASH is implemnted we'll need to address that.
1621 	 */
1622 	if (priv->params.cqe_zipping_en) {
1623 		MLX5_SET(cqc, cqc, mini_cqe_res_format, MLX5_CQE_FORMAT_CSUM);
1624 		MLX5_SET(cqc, cqc, cqe_compression_en, 1);
1625 	}
1626 
1627 	MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_rq_size);
1628 	MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
1629 	MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
1630 
1631 	switch (priv->params.rx_cq_moderation_mode) {
1632 	case 0:
1633 		MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1634 		break;
1635 	default:
1636 		if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
1637 			MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
1638 		else
1639 			MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1640 		break;
1641 	}
1642 
1643 	mlx5e_build_common_cq_param(priv, param);
1644 }
1645 
1646 static void
1647 mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
1648     struct mlx5e_cq_param *param)
1649 {
1650 	void *cqc = param->cqc;
1651 
1652 	MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_sq_size);
1653 	MLX5_SET(cqc, cqc, cq_period, priv->params.tx_cq_moderation_usec);
1654 	MLX5_SET(cqc, cqc, cq_max_count, priv->params.tx_cq_moderation_pkts);
1655 
1656 	switch (priv->params.tx_cq_moderation_mode) {
1657 	case 0:
1658 		MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1659 		break;
1660 	default:
1661 		if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
1662 			MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
1663 		else
1664 			MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1665 		break;
1666 	}
1667 
1668 	mlx5e_build_common_cq_param(priv, param);
1669 }
1670 
1671 static void
1672 mlx5e_build_channel_param(struct mlx5e_priv *priv,
1673     struct mlx5e_channel_param *cparam)
1674 {
1675 	memset(cparam, 0, sizeof(*cparam));
1676 
1677 	mlx5e_build_rq_param(priv, &cparam->rq);
1678 	mlx5e_build_sq_param(priv, &cparam->sq);
1679 	mlx5e_build_rx_cq_param(priv, &cparam->rx_cq);
1680 	mlx5e_build_tx_cq_param(priv, &cparam->tx_cq);
1681 }
1682 
1683 static int
1684 mlx5e_open_channels(struct mlx5e_priv *priv)
1685 {
1686 	struct mlx5e_channel_param cparam;
1687 	void *ptr;
1688 	int err;
1689 	int i;
1690 	int j;
1691 
1692 	priv->channel = malloc(priv->params.num_channels *
1693 	    sizeof(struct mlx5e_channel *), M_MLX5EN, M_WAITOK | M_ZERO);
1694 	if (priv->channel == NULL)
1695 		return (-ENOMEM);
1696 
1697 	mlx5e_build_channel_param(priv, &cparam);
1698 	for (i = 0; i < priv->params.num_channels; i++) {
1699 		err = mlx5e_open_channel(priv, i, &cparam, &priv->channel[i]);
1700 		if (err)
1701 			goto err_close_channels;
1702 	}
1703 
1704 	for (j = 0; j < priv->params.num_channels; j++) {
1705 		err = mlx5e_wait_for_min_rx_wqes(&priv->channel[j]->rq);
1706 		if (err)
1707 			goto err_close_channels;
1708 	}
1709 
1710 	return (0);
1711 
1712 err_close_channels:
1713 	for (i--; i >= 0; i--) {
1714 		mlx5e_close_channel(&priv->channel[i]);
1715 		mlx5e_close_channel_wait(&priv->channel[i]);
1716 	}
1717 
1718 	/* remove "volatile" attribute from "channel" pointer */
1719 	ptr = __DECONST(void *, priv->channel);
1720 	priv->channel = NULL;
1721 
1722 	free(ptr, M_MLX5EN);
1723 
1724 	return (err);
1725 }
1726 
1727 static void
1728 mlx5e_close_channels(struct mlx5e_priv *priv)
1729 {
1730 	void *ptr;
1731 	int i;
1732 
1733 	if (priv->channel == NULL)
1734 		return;
1735 
1736 	for (i = 0; i < priv->params.num_channels; i++)
1737 		mlx5e_close_channel(&priv->channel[i]);
1738 	for (i = 0; i < priv->params.num_channels; i++)
1739 		mlx5e_close_channel_wait(&priv->channel[i]);
1740 
1741 	/* remove "volatile" attribute from "channel" pointer */
1742 	ptr = __DECONST(void *, priv->channel);
1743 	priv->channel = NULL;
1744 
1745 	free(ptr, M_MLX5EN);
1746 }
1747 
1748 static int
1749 mlx5e_refresh_sq_params(struct mlx5e_priv *priv, struct mlx5e_sq *sq)
1750 {
1751 	return (mlx5_core_modify_cq_moderation(priv->mdev, &sq->cq.mcq,
1752 	    priv->params.tx_cq_moderation_usec,
1753 	    priv->params.tx_cq_moderation_pkts));
1754 }
1755 
1756 static int
1757 mlx5e_refresh_rq_params(struct mlx5e_priv *priv, struct mlx5e_rq *rq)
1758 {
1759 	return (mlx5_core_modify_cq_moderation(priv->mdev, &rq->cq.mcq,
1760 	    priv->params.rx_cq_moderation_usec,
1761 	    priv->params.rx_cq_moderation_pkts));
1762 }
1763 
1764 static int
1765 mlx5e_refresh_channel_params_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
1766 {
1767 	int err;
1768 	int i;
1769 
1770 	if (c == NULL)
1771 		return (EINVAL);
1772 
1773 	err = mlx5e_refresh_rq_params(priv, &c->rq);
1774 	if (err)
1775 		goto done;
1776 
1777 	for (i = 0; i != c->num_tc; i++) {
1778 		err = mlx5e_refresh_sq_params(priv, &c->sq[i]);
1779 		if (err)
1780 			goto done;
1781 	}
1782 done:
1783 	return (err);
1784 }
1785 
1786 int
1787 mlx5e_refresh_channel_params(struct mlx5e_priv *priv)
1788 {
1789 	int i;
1790 
1791 	if (priv->channel == NULL)
1792 		return (EINVAL);
1793 
1794 	for (i = 0; i < priv->params.num_channels; i++) {
1795 		int err;
1796 
1797 		err = mlx5e_refresh_channel_params_sub(priv, priv->channel[i]);
1798 		if (err)
1799 			return (err);
1800 	}
1801 	return (0);
1802 }
1803 
1804 static int
1805 mlx5e_open_tis(struct mlx5e_priv *priv, int tc)
1806 {
1807 	struct mlx5_core_dev *mdev = priv->mdev;
1808 	u32 in[MLX5_ST_SZ_DW(create_tis_in)];
1809 	void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
1810 
1811 	memset(in, 0, sizeof(in));
1812 
1813 	MLX5_SET(tisc, tisc, prio, tc);
1814 	MLX5_SET(tisc, tisc, transport_domain, priv->tdn);
1815 
1816 	return (mlx5_core_create_tis(mdev, in, sizeof(in), &priv->tisn[tc]));
1817 }
1818 
1819 static void
1820 mlx5e_close_tis(struct mlx5e_priv *priv, int tc)
1821 {
1822 	mlx5_core_destroy_tis(priv->mdev, priv->tisn[tc]);
1823 }
1824 
1825 static int
1826 mlx5e_open_tises(struct mlx5e_priv *priv)
1827 {
1828 	int num_tc = priv->num_tc;
1829 	int err;
1830 	int tc;
1831 
1832 	for (tc = 0; tc < num_tc; tc++) {
1833 		err = mlx5e_open_tis(priv, tc);
1834 		if (err)
1835 			goto err_close_tises;
1836 	}
1837 
1838 	return (0);
1839 
1840 err_close_tises:
1841 	for (tc--; tc >= 0; tc--)
1842 		mlx5e_close_tis(priv, tc);
1843 
1844 	return (err);
1845 }
1846 
1847 static void
1848 mlx5e_close_tises(struct mlx5e_priv *priv)
1849 {
1850 	int num_tc = priv->num_tc;
1851 	int tc;
1852 
1853 	for (tc = 0; tc < num_tc; tc++)
1854 		mlx5e_close_tis(priv, tc);
1855 }
1856 
1857 static int
1858 mlx5e_open_rqt(struct mlx5e_priv *priv)
1859 {
1860 	struct mlx5_core_dev *mdev = priv->mdev;
1861 	u32 *in;
1862 	u32 out[MLX5_ST_SZ_DW(create_rqt_out)];
1863 	void *rqtc;
1864 	int inlen;
1865 	int err;
1866 	int sz;
1867 	int i;
1868 
1869 	sz = 1 << priv->params.rx_hash_log_tbl_sz;
1870 
1871 	inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
1872 	in = mlx5_vzalloc(inlen);
1873 	if (in == NULL)
1874 		return (-ENOMEM);
1875 	rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
1876 
1877 	MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
1878 	MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
1879 
1880 	for (i = 0; i < sz; i++) {
1881 		int ix;
1882 #ifdef RSS
1883 		ix = rss_get_indirection_to_bucket(i);
1884 #else
1885 		ix = i;
1886 #endif
1887 		/* ensure we don't overflow */
1888 		ix %= priv->params.num_channels;
1889 		MLX5_SET(rqtc, rqtc, rq_num[i], priv->channel[ix]->rq.rqn);
1890 	}
1891 
1892 	MLX5_SET(create_rqt_in, in, opcode, MLX5_CMD_OP_CREATE_RQT);
1893 
1894 	memset(out, 0, sizeof(out));
1895 	err = mlx5_cmd_exec_check_status(mdev, in, inlen, out, sizeof(out));
1896 	if (!err)
1897 		priv->rqtn = MLX5_GET(create_rqt_out, out, rqtn);
1898 
1899 	kvfree(in);
1900 
1901 	return (err);
1902 }
1903 
1904 static void
1905 mlx5e_close_rqt(struct mlx5e_priv *priv)
1906 {
1907 	u32 in[MLX5_ST_SZ_DW(destroy_rqt_in)];
1908 	u32 out[MLX5_ST_SZ_DW(destroy_rqt_out)];
1909 
1910 	memset(in, 0, sizeof(in));
1911 
1912 	MLX5_SET(destroy_rqt_in, in, opcode, MLX5_CMD_OP_DESTROY_RQT);
1913 	MLX5_SET(destroy_rqt_in, in, rqtn, priv->rqtn);
1914 
1915 	mlx5_cmd_exec_check_status(priv->mdev, in, sizeof(in), out,
1916 	    sizeof(out));
1917 }
1918 
1919 static void
1920 mlx5e_build_tir_ctx(struct mlx5e_priv *priv, u32 * tirc, int tt)
1921 {
1922 	void *hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
1923 	__be32 *hkey;
1924 
1925 	MLX5_SET(tirc, tirc, transport_domain, priv->tdn);
1926 
1927 #define	ROUGH_MAX_L2_L3_HDR_SZ 256
1928 
1929 #define	MLX5_HASH_IP     (MLX5_HASH_FIELD_SEL_SRC_IP   |\
1930 			  MLX5_HASH_FIELD_SEL_DST_IP)
1931 
1932 #define	MLX5_HASH_ALL    (MLX5_HASH_FIELD_SEL_SRC_IP   |\
1933 			  MLX5_HASH_FIELD_SEL_DST_IP   |\
1934 			  MLX5_HASH_FIELD_SEL_L4_SPORT |\
1935 			  MLX5_HASH_FIELD_SEL_L4_DPORT)
1936 
1937 #define	MLX5_HASH_IP_IPSEC_SPI	(MLX5_HASH_FIELD_SEL_SRC_IP   |\
1938 				 MLX5_HASH_FIELD_SEL_DST_IP   |\
1939 				 MLX5_HASH_FIELD_SEL_IPSEC_SPI)
1940 
1941 	if (priv->params.hw_lro_en) {
1942 		MLX5_SET(tirc, tirc, lro_enable_mask,
1943 		    MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO |
1944 		    MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO);
1945 		MLX5_SET(tirc, tirc, lro_max_msg_sz,
1946 		    (priv->params.lro_wqe_sz -
1947 		    ROUGH_MAX_L2_L3_HDR_SZ) >> 8);
1948 		/* TODO: add the option to choose timer value dynamically */
1949 		MLX5_SET(tirc, tirc, lro_timeout_period_usecs,
1950 		    MLX5_CAP_ETH(priv->mdev,
1951 		    lro_timer_supported_periods[2]));
1952 	}
1953 
1954 	/* setup parameters for hashing TIR type, if any */
1955 	switch (tt) {
1956 	case MLX5E_TT_ANY:
1957 		MLX5_SET(tirc, tirc, disp_type,
1958 		    MLX5_TIRC_DISP_TYPE_DIRECT);
1959 		MLX5_SET(tirc, tirc, inline_rqn,
1960 		    priv->channel[0]->rq.rqn);
1961 		break;
1962 	default:
1963 		MLX5_SET(tirc, tirc, disp_type,
1964 		    MLX5_TIRC_DISP_TYPE_INDIRECT);
1965 		MLX5_SET(tirc, tirc, indirect_table,
1966 		    priv->rqtn);
1967 		MLX5_SET(tirc, tirc, rx_hash_fn,
1968 		    MLX5_TIRC_RX_HASH_FN_HASH_TOEPLITZ);
1969 		hkey = (__be32 *) MLX5_ADDR_OF(tirc, tirc, rx_hash_toeplitz_key);
1970 #ifdef RSS
1971 		/*
1972 		 * The FreeBSD RSS implementation does currently not
1973 		 * support symmetric Toeplitz hashes:
1974 		 */
1975 		MLX5_SET(tirc, tirc, rx_hash_symmetric, 0);
1976 		rss_getkey((uint8_t *)hkey);
1977 #else
1978 		MLX5_SET(tirc, tirc, rx_hash_symmetric, 1);
1979 		hkey[0] = cpu_to_be32(0xD181C62C);
1980 		hkey[1] = cpu_to_be32(0xF7F4DB5B);
1981 		hkey[2] = cpu_to_be32(0x1983A2FC);
1982 		hkey[3] = cpu_to_be32(0x943E1ADB);
1983 		hkey[4] = cpu_to_be32(0xD9389E6B);
1984 		hkey[5] = cpu_to_be32(0xD1039C2C);
1985 		hkey[6] = cpu_to_be32(0xA74499AD);
1986 		hkey[7] = cpu_to_be32(0x593D56D9);
1987 		hkey[8] = cpu_to_be32(0xF3253C06);
1988 		hkey[9] = cpu_to_be32(0x2ADC1FFC);
1989 #endif
1990 		break;
1991 	}
1992 
1993 	switch (tt) {
1994 	case MLX5E_TT_IPV4_TCP:
1995 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
1996 		    MLX5_L3_PROT_TYPE_IPV4);
1997 		MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
1998 		    MLX5_L4_PROT_TYPE_TCP);
1999 #ifdef RSS
2000 		if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4)) {
2001 			MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2002 			    MLX5_HASH_IP);
2003 		} else
2004 #endif
2005 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2006 		    MLX5_HASH_ALL);
2007 		break;
2008 
2009 	case MLX5E_TT_IPV6_TCP:
2010 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2011 		    MLX5_L3_PROT_TYPE_IPV6);
2012 		MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2013 		    MLX5_L4_PROT_TYPE_TCP);
2014 #ifdef RSS
2015 		if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) {
2016 			MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2017 			    MLX5_HASH_IP);
2018 		} else
2019 #endif
2020 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2021 		    MLX5_HASH_ALL);
2022 		break;
2023 
2024 	case MLX5E_TT_IPV4_UDP:
2025 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2026 		    MLX5_L3_PROT_TYPE_IPV4);
2027 		MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2028 		    MLX5_L4_PROT_TYPE_UDP);
2029 #ifdef RSS
2030 		if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4)) {
2031 			MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2032 			    MLX5_HASH_IP);
2033 		} else
2034 #endif
2035 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2036 		    MLX5_HASH_ALL);
2037 		break;
2038 
2039 	case MLX5E_TT_IPV6_UDP:
2040 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2041 		    MLX5_L3_PROT_TYPE_IPV6);
2042 		MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2043 		    MLX5_L4_PROT_TYPE_UDP);
2044 #ifdef RSS
2045 		if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) {
2046 			MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2047 			    MLX5_HASH_IP);
2048 		} else
2049 #endif
2050 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2051 		    MLX5_HASH_ALL);
2052 		break;
2053 
2054 	case MLX5E_TT_IPV4_IPSEC_AH:
2055 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2056 		    MLX5_L3_PROT_TYPE_IPV4);
2057 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2058 		    MLX5_HASH_IP_IPSEC_SPI);
2059 		break;
2060 
2061 	case MLX5E_TT_IPV6_IPSEC_AH:
2062 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2063 		    MLX5_L3_PROT_TYPE_IPV6);
2064 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2065 		    MLX5_HASH_IP_IPSEC_SPI);
2066 		break;
2067 
2068 	case MLX5E_TT_IPV4_IPSEC_ESP:
2069 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2070 		    MLX5_L3_PROT_TYPE_IPV4);
2071 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2072 		    MLX5_HASH_IP_IPSEC_SPI);
2073 		break;
2074 
2075 	case MLX5E_TT_IPV6_IPSEC_ESP:
2076 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2077 		    MLX5_L3_PROT_TYPE_IPV6);
2078 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2079 		    MLX5_HASH_IP_IPSEC_SPI);
2080 		break;
2081 
2082 	case MLX5E_TT_IPV4:
2083 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2084 		    MLX5_L3_PROT_TYPE_IPV4);
2085 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2086 		    MLX5_HASH_IP);
2087 		break;
2088 
2089 	case MLX5E_TT_IPV6:
2090 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2091 		    MLX5_L3_PROT_TYPE_IPV6);
2092 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2093 		    MLX5_HASH_IP);
2094 		break;
2095 
2096 	default:
2097 		break;
2098 	}
2099 }
2100 
2101 static int
2102 mlx5e_open_tir(struct mlx5e_priv *priv, int tt)
2103 {
2104 	struct mlx5_core_dev *mdev = priv->mdev;
2105 	u32 *in;
2106 	void *tirc;
2107 	int inlen;
2108 	int err;
2109 
2110 	inlen = MLX5_ST_SZ_BYTES(create_tir_in);
2111 	in = mlx5_vzalloc(inlen);
2112 	if (in == NULL)
2113 		return (-ENOMEM);
2114 	tirc = MLX5_ADDR_OF(create_tir_in, in, tir_context);
2115 
2116 	mlx5e_build_tir_ctx(priv, tirc, tt);
2117 
2118 	err = mlx5_core_create_tir(mdev, in, inlen, &priv->tirn[tt]);
2119 
2120 	kvfree(in);
2121 
2122 	return (err);
2123 }
2124 
2125 static void
2126 mlx5e_close_tir(struct mlx5e_priv *priv, int tt)
2127 {
2128 	mlx5_core_destroy_tir(priv->mdev, priv->tirn[tt]);
2129 }
2130 
2131 static int
2132 mlx5e_open_tirs(struct mlx5e_priv *priv)
2133 {
2134 	int err;
2135 	int i;
2136 
2137 	for (i = 0; i < MLX5E_NUM_TT; i++) {
2138 		err = mlx5e_open_tir(priv, i);
2139 		if (err)
2140 			goto err_close_tirs;
2141 	}
2142 
2143 	return (0);
2144 
2145 err_close_tirs:
2146 	for (i--; i >= 0; i--)
2147 		mlx5e_close_tir(priv, i);
2148 
2149 	return (err);
2150 }
2151 
2152 static void
2153 mlx5e_close_tirs(struct mlx5e_priv *priv)
2154 {
2155 	int i;
2156 
2157 	for (i = 0; i < MLX5E_NUM_TT; i++)
2158 		mlx5e_close_tir(priv, i);
2159 }
2160 
2161 /*
2162  * SW MTU does not include headers,
2163  * HW MTU includes all headers and checksums.
2164  */
2165 static int
2166 mlx5e_set_dev_port_mtu(struct ifnet *ifp, int sw_mtu)
2167 {
2168 	struct mlx5e_priv *priv = ifp->if_softc;
2169 	struct mlx5_core_dev *mdev = priv->mdev;
2170 	int hw_mtu;
2171 	int err;
2172 
2173 	err = mlx5_set_port_mtu(mdev, MLX5E_SW2HW_MTU(sw_mtu));
2174 	if (err) {
2175 		if_printf(ifp, "%s: mlx5_set_port_mtu failed setting %d, err=%d\n",
2176 		    __func__, sw_mtu, err);
2177 		return (err);
2178 	}
2179 	err = mlx5_query_port_oper_mtu(mdev, &hw_mtu);
2180 	if (err) {
2181 		if_printf(ifp, "Query port MTU, after setting new "
2182 		    "MTU value, failed\n");
2183 	} else if (MLX5E_HW2SW_MTU(hw_mtu) < sw_mtu) {
2184 		err = -E2BIG,
2185 		if_printf(ifp, "Port MTU %d is smaller than "
2186                     "ifp mtu %d\n", hw_mtu, sw_mtu);
2187 	} else if (MLX5E_HW2SW_MTU(hw_mtu) > sw_mtu) {
2188 		err = -EINVAL;
2189                 if_printf(ifp, "Port MTU %d is bigger than "
2190                     "ifp mtu %d\n", hw_mtu, sw_mtu);
2191 	}
2192 	ifp->if_mtu = sw_mtu;
2193 	return (err);
2194 }
2195 
2196 int
2197 mlx5e_open_locked(struct ifnet *ifp)
2198 {
2199 	struct mlx5e_priv *priv = ifp->if_softc;
2200 	int err;
2201 	u16 set_id;
2202 
2203 	/* check if already opened */
2204 	if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
2205 		return (0);
2206 
2207 #ifdef RSS
2208 	if (rss_getnumbuckets() > priv->params.num_channels) {
2209 		if_printf(ifp, "NOTE: There are more RSS buckets(%u) than "
2210 		    "channels(%u) available\n", rss_getnumbuckets(),
2211 		    priv->params.num_channels);
2212 	}
2213 #endif
2214 	err = mlx5e_open_tises(priv);
2215 	if (err) {
2216 		if_printf(ifp, "%s: mlx5e_open_tises failed, %d\n",
2217 		    __func__, err);
2218 		return (err);
2219 	}
2220 	err = mlx5_vport_alloc_q_counter(priv->mdev,
2221 	    MLX5_INTERFACE_PROTOCOL_ETH, &set_id);
2222 	if (err) {
2223 		if_printf(priv->ifp,
2224 		    "%s: mlx5_vport_alloc_q_counter failed: %d\n",
2225 		    __func__, err);
2226 		goto err_close_tises;
2227 	}
2228 	/* store counter set ID */
2229 	priv->counter_set_id = set_id;
2230 
2231 	err = mlx5e_open_channels(priv);
2232 	if (err) {
2233 		if_printf(ifp, "%s: mlx5e_open_channels failed, %d\n",
2234 		    __func__, err);
2235 		goto err_dalloc_q_counter;
2236 	}
2237 	err = mlx5e_open_rqt(priv);
2238 	if (err) {
2239 		if_printf(ifp, "%s: mlx5e_open_rqt failed, %d\n",
2240 		    __func__, err);
2241 		goto err_close_channels;
2242 	}
2243 	err = mlx5e_open_tirs(priv);
2244 	if (err) {
2245 		if_printf(ifp, "%s: mlx5e_open_tir failed, %d\n",
2246 		    __func__, err);
2247 		goto err_close_rqls;
2248 	}
2249 	err = mlx5e_open_flow_table(priv);
2250 	if (err) {
2251 		if_printf(ifp, "%s: mlx5e_open_flow_table failed, %d\n",
2252 		    __func__, err);
2253 		goto err_close_tirs;
2254 	}
2255 	err = mlx5e_add_all_vlan_rules(priv);
2256 	if (err) {
2257 		if_printf(ifp, "%s: mlx5e_add_all_vlan_rules failed, %d\n",
2258 		    __func__, err);
2259 		goto err_close_flow_table;
2260 	}
2261 	set_bit(MLX5E_STATE_OPENED, &priv->state);
2262 
2263 	mlx5e_update_carrier(priv);
2264 	mlx5e_set_rx_mode_core(priv);
2265 
2266 	return (0);
2267 
2268 err_close_flow_table:
2269 	mlx5e_close_flow_table(priv);
2270 
2271 err_close_tirs:
2272 	mlx5e_close_tirs(priv);
2273 
2274 err_close_rqls:
2275 	mlx5e_close_rqt(priv);
2276 
2277 err_close_channels:
2278 	mlx5e_close_channels(priv);
2279 
2280 err_dalloc_q_counter:
2281 	mlx5_vport_dealloc_q_counter(priv->mdev,
2282 	    MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
2283 
2284 err_close_tises:
2285 	mlx5e_close_tises(priv);
2286 
2287 	return (err);
2288 }
2289 
2290 static void
2291 mlx5e_open(void *arg)
2292 {
2293 	struct mlx5e_priv *priv = arg;
2294 
2295 	PRIV_LOCK(priv);
2296 	if (mlx5_set_port_status(priv->mdev, MLX5_PORT_UP))
2297 		if_printf(priv->ifp,
2298 		    "%s: Setting port status to up failed\n",
2299 		    __func__);
2300 
2301 	mlx5e_open_locked(priv->ifp);
2302 	priv->ifp->if_drv_flags |= IFF_DRV_RUNNING;
2303 	PRIV_UNLOCK(priv);
2304 }
2305 
2306 int
2307 mlx5e_close_locked(struct ifnet *ifp)
2308 {
2309 	struct mlx5e_priv *priv = ifp->if_softc;
2310 
2311 	/* check if already closed */
2312 	if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2313 		return (0);
2314 
2315 	clear_bit(MLX5E_STATE_OPENED, &priv->state);
2316 
2317 	mlx5e_set_rx_mode_core(priv);
2318 	mlx5e_del_all_vlan_rules(priv);
2319 	if_link_state_change(priv->ifp, LINK_STATE_DOWN);
2320 	mlx5e_close_flow_table(priv);
2321 	mlx5e_close_tirs(priv);
2322 	mlx5e_close_rqt(priv);
2323 	mlx5e_close_channels(priv);
2324 	mlx5_vport_dealloc_q_counter(priv->mdev,
2325 	    MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
2326 	mlx5e_close_tises(priv);
2327 
2328 	return (0);
2329 }
2330 
2331 #if (__FreeBSD_version >= 1100000)
2332 static uint64_t
2333 mlx5e_get_counter(struct ifnet *ifp, ift_counter cnt)
2334 {
2335 	struct mlx5e_priv *priv = ifp->if_softc;
2336 	u64 retval;
2337 
2338 	/* PRIV_LOCK(priv); XXX not allowed */
2339 	switch (cnt) {
2340 	case IFCOUNTER_IPACKETS:
2341 		retval = priv->stats.vport.rx_packets;
2342 		break;
2343 	case IFCOUNTER_IERRORS:
2344 		retval = priv->stats.vport.rx_error_packets;
2345 		break;
2346 	case IFCOUNTER_IQDROPS:
2347 		retval = priv->stats.vport.rx_out_of_buffer;
2348 		break;
2349 	case IFCOUNTER_OPACKETS:
2350 		retval = priv->stats.vport.tx_packets;
2351 		break;
2352 	case IFCOUNTER_OERRORS:
2353 		retval = priv->stats.vport.tx_error_packets;
2354 		break;
2355 	case IFCOUNTER_IBYTES:
2356 		retval = priv->stats.vport.rx_bytes;
2357 		break;
2358 	case IFCOUNTER_OBYTES:
2359 		retval = priv->stats.vport.tx_bytes;
2360 		break;
2361 	case IFCOUNTER_IMCASTS:
2362 		retval = priv->stats.vport.rx_multicast_packets;
2363 		break;
2364 	case IFCOUNTER_OMCASTS:
2365 		retval = priv->stats.vport.tx_multicast_packets;
2366 		break;
2367 	case IFCOUNTER_OQDROPS:
2368 		retval = priv->stats.vport.tx_queue_dropped;
2369 		break;
2370 	default:
2371 		retval = if_get_counter_default(ifp, cnt);
2372 		break;
2373 	}
2374 	/* PRIV_UNLOCK(priv); XXX not allowed */
2375 	return (retval);
2376 }
2377 #endif
2378 
2379 static void
2380 mlx5e_set_rx_mode(struct ifnet *ifp)
2381 {
2382 	struct mlx5e_priv *priv = ifp->if_softc;
2383 
2384 	schedule_work(&priv->set_rx_mode_work);
2385 }
2386 
2387 static int
2388 mlx5e_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2389 {
2390 	struct mlx5e_priv *priv;
2391 	struct ifreq *ifr;
2392 	struct ifi2creq i2c;
2393 	int error = 0;
2394 	int mask = 0;
2395 	int size_read = 0;
2396 	int module_num;
2397 	int max_mtu;
2398 	uint8_t read_addr;
2399 
2400 	priv = ifp->if_softc;
2401 
2402 	/* check if detaching */
2403 	if (priv == NULL || priv->gone != 0)
2404 		return (ENXIO);
2405 
2406 	switch (command) {
2407 	case SIOCSIFMTU:
2408 		ifr = (struct ifreq *)data;
2409 
2410 		PRIV_LOCK(priv);
2411 		mlx5_query_port_max_mtu(priv->mdev, &max_mtu);
2412 
2413 		if (ifr->ifr_mtu >= MLX5E_MTU_MIN &&
2414 		    ifr->ifr_mtu <= MIN(MLX5E_MTU_MAX, max_mtu)) {
2415 			int was_opened;
2416 
2417 			was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2418 			if (was_opened)
2419 				mlx5e_close_locked(ifp);
2420 
2421 			/* set new MTU */
2422 			mlx5e_set_dev_port_mtu(ifp, ifr->ifr_mtu);
2423 
2424 			if (was_opened)
2425 				mlx5e_open_locked(ifp);
2426 		} else {
2427 			error = EINVAL;
2428 			if_printf(ifp, "Invalid MTU value. Min val: %d, Max val: %d\n",
2429 			    MLX5E_MTU_MIN, MIN(MLX5E_MTU_MAX, max_mtu));
2430 		}
2431 		PRIV_UNLOCK(priv);
2432 		break;
2433 	case SIOCSIFFLAGS:
2434 		if ((ifp->if_flags & IFF_UP) &&
2435 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
2436 			mlx5e_set_rx_mode(ifp);
2437 			break;
2438 		}
2439 		PRIV_LOCK(priv);
2440 		if (ifp->if_flags & IFF_UP) {
2441 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2442 				if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2443 					mlx5e_open_locked(ifp);
2444 				ifp->if_drv_flags |= IFF_DRV_RUNNING;
2445 				mlx5_set_port_status(priv->mdev, MLX5_PORT_UP);
2446 			}
2447 		} else {
2448 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
2449 				mlx5_set_port_status(priv->mdev,
2450 				    MLX5_PORT_DOWN);
2451 				if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
2452 					mlx5e_close_locked(ifp);
2453 				mlx5e_update_carrier(priv);
2454 				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2455 			}
2456 		}
2457 		PRIV_UNLOCK(priv);
2458 		break;
2459 	case SIOCADDMULTI:
2460 	case SIOCDELMULTI:
2461 		mlx5e_set_rx_mode(ifp);
2462 		break;
2463 	case SIOCSIFMEDIA:
2464 	case SIOCGIFMEDIA:
2465 	case SIOCGIFXMEDIA:
2466 		ifr = (struct ifreq *)data;
2467 		error = ifmedia_ioctl(ifp, ifr, &priv->media, command);
2468 		break;
2469 	case SIOCSIFCAP:
2470 		ifr = (struct ifreq *)data;
2471 		PRIV_LOCK(priv);
2472 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2473 
2474 		if (mask & IFCAP_TXCSUM) {
2475 			ifp->if_capenable ^= IFCAP_TXCSUM;
2476 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
2477 
2478 			if (IFCAP_TSO4 & ifp->if_capenable &&
2479 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
2480 				ifp->if_capenable &= ~IFCAP_TSO4;
2481 				ifp->if_hwassist &= ~CSUM_IP_TSO;
2482 				if_printf(ifp,
2483 				    "tso4 disabled due to -txcsum.\n");
2484 			}
2485 		}
2486 		if (mask & IFCAP_TXCSUM_IPV6) {
2487 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
2488 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
2489 
2490 			if (IFCAP_TSO6 & ifp->if_capenable &&
2491 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2492 				ifp->if_capenable &= ~IFCAP_TSO6;
2493 				ifp->if_hwassist &= ~CSUM_IP6_TSO;
2494 				if_printf(ifp,
2495 				    "tso6 disabled due to -txcsum6.\n");
2496 			}
2497 		}
2498 		if (mask & IFCAP_RXCSUM)
2499 			ifp->if_capenable ^= IFCAP_RXCSUM;
2500 		if (mask & IFCAP_RXCSUM_IPV6)
2501 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
2502 		if (mask & IFCAP_TSO4) {
2503 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
2504 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
2505 				if_printf(ifp, "enable txcsum first.\n");
2506 				error = EAGAIN;
2507 				goto out;
2508 			}
2509 			ifp->if_capenable ^= IFCAP_TSO4;
2510 			ifp->if_hwassist ^= CSUM_IP_TSO;
2511 		}
2512 		if (mask & IFCAP_TSO6) {
2513 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
2514 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2515 				if_printf(ifp, "enable txcsum6 first.\n");
2516 				error = EAGAIN;
2517 				goto out;
2518 			}
2519 			ifp->if_capenable ^= IFCAP_TSO6;
2520 			ifp->if_hwassist ^= CSUM_IP6_TSO;
2521 		}
2522 		if (mask & IFCAP_VLAN_HWFILTER) {
2523 			if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2524 				mlx5e_disable_vlan_filter(priv);
2525 			else
2526 				mlx5e_enable_vlan_filter(priv);
2527 
2528 			ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
2529 		}
2530 		if (mask & IFCAP_VLAN_HWTAGGING)
2531 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2532 		if (mask & IFCAP_WOL_MAGIC)
2533 			ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2534 
2535 		VLAN_CAPABILITIES(ifp);
2536 		/* turn off LRO means also turn of HW LRO - if it's on */
2537 		if (mask & IFCAP_LRO) {
2538 			int was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2539 			bool need_restart = false;
2540 
2541 			ifp->if_capenable ^= IFCAP_LRO;
2542 			if (!(ifp->if_capenable & IFCAP_LRO)) {
2543 				if (priv->params.hw_lro_en) {
2544 					priv->params.hw_lro_en = false;
2545 					need_restart = true;
2546 					/* Not sure this is the correct way */
2547 					priv->params_ethtool.hw_lro = priv->params.hw_lro_en;
2548 				}
2549 			}
2550 			if (was_opened && need_restart) {
2551 				mlx5e_close_locked(ifp);
2552 				mlx5e_open_locked(ifp);
2553 			}
2554 		}
2555 out:
2556 		PRIV_UNLOCK(priv);
2557 		break;
2558 
2559 	case SIOCGI2C:
2560 		ifr = (struct ifreq *)data;
2561 
2562 		/*
2563 		 * Copy from the user-space address ifr_data to the
2564 		 * kernel-space address i2c
2565 		 */
2566 		error = copyin(ifr->ifr_data, &i2c, sizeof(i2c));
2567 		if (error)
2568 			break;
2569 
2570 		if (i2c.len > sizeof(i2c.data)) {
2571 			error = EINVAL;
2572 			break;
2573 		}
2574 
2575 		PRIV_LOCK(priv);
2576 		/* Get module_num which is required for the query_eeprom */
2577 		error = mlx5_query_module_num(priv->mdev, &module_num);
2578 		if (error) {
2579 			if_printf(ifp, "Query module num failed, eeprom "
2580 			    "reading is not supported\n");
2581 			error = EINVAL;
2582 			goto err_i2c;
2583 		}
2584 		/* Check if module is present before doing an access */
2585 		if (mlx5_query_module_status(priv->mdev, module_num) !=
2586 		    MLX5_MODULE_STATUS_PLUGGED) {
2587 			error = EINVAL;
2588 			goto err_i2c;
2589 		}
2590 		/*
2591 		 * Currently 0XA0 and 0xA2 are the only addresses permitted.
2592 		 * The internal conversion is as follows:
2593 		 */
2594 		if (i2c.dev_addr == 0xA0)
2595 			read_addr = MLX5E_I2C_ADDR_LOW;
2596 		else if (i2c.dev_addr == 0xA2)
2597 			read_addr = MLX5E_I2C_ADDR_HIGH;
2598 		else {
2599 			if_printf(ifp, "Query eeprom failed, "
2600 			    "Invalid Address: %X\n", i2c.dev_addr);
2601 			error = EINVAL;
2602 			goto err_i2c;
2603 		}
2604 		error = mlx5_query_eeprom(priv->mdev,
2605 		    read_addr, MLX5E_EEPROM_LOW_PAGE,
2606 		    (uint32_t)i2c.offset, (uint32_t)i2c.len, module_num,
2607 		    (uint32_t *)i2c.data, &size_read);
2608 		if (error) {
2609 			if_printf(ifp, "Query eeprom failed, eeprom "
2610 			    "reading is not supported\n");
2611 			error = EINVAL;
2612 			goto err_i2c;
2613 		}
2614 
2615 		if (i2c.len > MLX5_EEPROM_MAX_BYTES) {
2616 			error = mlx5_query_eeprom(priv->mdev,
2617 			    read_addr, MLX5E_EEPROM_LOW_PAGE,
2618 			    (uint32_t)(i2c.offset + size_read),
2619 			    (uint32_t)(i2c.len - size_read), module_num,
2620 			    (uint32_t *)(i2c.data + size_read), &size_read);
2621 		}
2622 		if (error) {
2623 			if_printf(ifp, "Query eeprom failed, eeprom "
2624 			    "reading is not supported\n");
2625 			error = EINVAL;
2626 			goto err_i2c;
2627 		}
2628 
2629 		error = copyout(&i2c, ifr->ifr_data, sizeof(i2c));
2630 err_i2c:
2631 		PRIV_UNLOCK(priv);
2632 		break;
2633 
2634 	default:
2635 		error = ether_ioctl(ifp, command, data);
2636 		break;
2637 	}
2638 	return (error);
2639 }
2640 
2641 static int
2642 mlx5e_check_required_hca_cap(struct mlx5_core_dev *mdev)
2643 {
2644 	/*
2645 	 * TODO: uncoment once FW really sets all these bits if
2646 	 * (!mdev->caps.eth.rss_ind_tbl_cap || !mdev->caps.eth.csum_cap ||
2647 	 * !mdev->caps.eth.max_lso_cap || !mdev->caps.eth.vlan_cap ||
2648 	 * !(mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_SCQE_BRK_MOD)) return
2649 	 * -ENOTSUPP;
2650 	 */
2651 
2652 	/* TODO: add more must-to-have features */
2653 
2654 	if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
2655 		return (-ENODEV);
2656 
2657 	return (0);
2658 }
2659 
2660 static void
2661 mlx5e_build_ifp_priv(struct mlx5_core_dev *mdev,
2662     struct mlx5e_priv *priv,
2663     int num_comp_vectors)
2664 {
2665 	/*
2666 	 * TODO: Consider link speed for setting "log_sq_size",
2667 	 * "log_rq_size" and "cq_moderation_xxx":
2668 	 */
2669 	priv->params.log_sq_size =
2670 	    MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
2671 	priv->params.log_rq_size =
2672 	    MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
2673 	priv->params.rx_cq_moderation_usec =
2674 	    MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
2675 	    MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE :
2676 	    MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
2677 	priv->params.rx_cq_moderation_mode =
2678 	    MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ? 1 : 0;
2679 	priv->params.rx_cq_moderation_pkts =
2680 	    MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS;
2681 	priv->params.tx_cq_moderation_usec =
2682 	    MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC;
2683 	priv->params.tx_cq_moderation_pkts =
2684 	    MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
2685 	priv->params.min_rx_wqes =
2686 	    MLX5E_PARAMS_DEFAULT_MIN_RX_WQES;
2687 	priv->params.rx_hash_log_tbl_sz =
2688 	    (order_base_2(num_comp_vectors) >
2689 	    MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ) ?
2690 	    order_base_2(num_comp_vectors) :
2691 	    MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ;
2692 	priv->params.num_tc = 1;
2693 	priv->params.default_vlan_prio = 0;
2694 	priv->counter_set_id = -1;
2695 
2696 	/*
2697 	 * hw lro is currently defaulted to off. when it won't anymore we
2698 	 * will consider the HW capability: "!!MLX5_CAP_ETH(mdev, lro_cap)"
2699 	 */
2700 	priv->params.hw_lro_en = false;
2701 	priv->params.lro_wqe_sz = MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ;
2702 
2703 	priv->params.cqe_zipping_en = !!MLX5_CAP_GEN(mdev, cqe_compression);
2704 
2705 	priv->mdev = mdev;
2706 	priv->params.num_channels = num_comp_vectors;
2707 	priv->order_base_2_num_channels = order_base_2(num_comp_vectors);
2708 	priv->queue_mapping_channel_mask =
2709 	    roundup_pow_of_two(num_comp_vectors) - 1;
2710 	priv->num_tc = priv->params.num_tc;
2711 	priv->default_vlan_prio = priv->params.default_vlan_prio;
2712 
2713 	INIT_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
2714 	INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
2715 	INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
2716 }
2717 
2718 static int
2719 mlx5e_create_mkey(struct mlx5e_priv *priv, u32 pdn,
2720     struct mlx5_core_mr *mr)
2721 {
2722 	struct ifnet *ifp = priv->ifp;
2723 	struct mlx5_core_dev *mdev = priv->mdev;
2724 	struct mlx5_create_mkey_mbox_in *in;
2725 	int err;
2726 
2727 	in = mlx5_vzalloc(sizeof(*in));
2728 	if (in == NULL) {
2729 		if_printf(ifp, "%s: failed to allocate inbox\n", __func__);
2730 		return (-ENOMEM);
2731 	}
2732 	in->seg.flags = MLX5_PERM_LOCAL_WRITE |
2733 	    MLX5_PERM_LOCAL_READ |
2734 	    MLX5_ACCESS_MODE_PA;
2735 	in->seg.flags_pd = cpu_to_be32(pdn | MLX5_MKEY_LEN64);
2736 	in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
2737 
2738 	err = mlx5_core_create_mkey(mdev, mr, in, sizeof(*in), NULL, NULL,
2739 	    NULL);
2740 	if (err)
2741 		if_printf(ifp, "%s: mlx5_core_create_mkey failed, %d\n",
2742 		    __func__, err);
2743 
2744 	kvfree(in);
2745 
2746 	return (err);
2747 }
2748 
2749 static const char *mlx5e_vport_stats_desc[] = {
2750 	MLX5E_VPORT_STATS(MLX5E_STATS_DESC)
2751 };
2752 
2753 static const char *mlx5e_pport_stats_desc[] = {
2754 	MLX5E_PPORT_STATS(MLX5E_STATS_DESC)
2755 };
2756 
2757 static void
2758 mlx5e_priv_mtx_init(struct mlx5e_priv *priv)
2759 {
2760 	mtx_init(&priv->async_events_mtx, "mlx5async", MTX_NETWORK_LOCK, MTX_DEF);
2761 	sx_init(&priv->state_lock, "mlx5state");
2762 	callout_init_mtx(&priv->watchdog, &priv->async_events_mtx, 0);
2763 	MLX5_INIT_DOORBELL_LOCK(&priv->doorbell_lock);
2764 }
2765 
2766 static void
2767 mlx5e_priv_mtx_destroy(struct mlx5e_priv *priv)
2768 {
2769 	mtx_destroy(&priv->async_events_mtx);
2770 	sx_destroy(&priv->state_lock);
2771 }
2772 
2773 static int
2774 sysctl_firmware(SYSCTL_HANDLER_ARGS)
2775 {
2776 	/*
2777 	 * %d.%d%.d the string format.
2778 	 * fw_rev_{maj,min,sub} return u16, 2^16 = 65536.
2779 	 * We need at most 5 chars to store that.
2780 	 * It also has: two "." and NULL at the end, which means we need 18
2781 	 * (5*3 + 3) chars at most.
2782 	 */
2783 	char fw[18];
2784 	struct mlx5e_priv *priv = arg1;
2785 	int error;
2786 
2787 	snprintf(fw, sizeof(fw), "%d.%d.%d", fw_rev_maj(priv->mdev), fw_rev_min(priv->mdev),
2788 	    fw_rev_sub(priv->mdev));
2789 	error = sysctl_handle_string(oidp, fw, sizeof(fw), req);
2790 	return (error);
2791 }
2792 
2793 static void
2794 mlx5e_add_hw_stats(struct mlx5e_priv *priv)
2795 {
2796 	SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
2797 	    OID_AUTO, "fw_version", CTLTYPE_STRING | CTLFLAG_RD, priv, 0,
2798 	    sysctl_firmware, "A", "HCA firmware version");
2799 
2800 	SYSCTL_ADD_STRING(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
2801 	    OID_AUTO, "board_id", CTLFLAG_RD, priv->mdev->board_id, 0,
2802 	    "Board ID");
2803 }
2804 
2805 static void
2806 mlx5e_setup_pauseframes(struct mlx5e_priv *priv)
2807 {
2808 #if (__FreeBSD_version < 1100000)
2809 	char path[64];
2810 
2811 #endif
2812 	/* Only receiving pauseframes is enabled by default */
2813 	priv->params.tx_pauseframe_control = 0;
2814 	priv->params.rx_pauseframe_control = 1;
2815 
2816 #if (__FreeBSD_version < 1100000)
2817 	/* compute path for sysctl */
2818 	snprintf(path, sizeof(path), "dev.mce.%d.tx_pauseframe_control",
2819 	    device_get_unit(priv->mdev->pdev->dev.bsddev));
2820 
2821 	/* try to fetch tunable, if any */
2822 	TUNABLE_INT_FETCH(path, &priv->params.tx_pauseframe_control);
2823 
2824 	/* compute path for sysctl */
2825 	snprintf(path, sizeof(path), "dev.mce.%d.rx_pauseframe_control",
2826 	    device_get_unit(priv->mdev->pdev->dev.bsddev));
2827 
2828 	/* try to fetch tunable, if any */
2829 	TUNABLE_INT_FETCH(path, &priv->params.rx_pauseframe_control);
2830 #endif
2831 
2832 	/* register pausframe SYSCTLs */
2833 	SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
2834 	    OID_AUTO, "tx_pauseframe_control", CTLFLAG_RDTUN,
2835 	    &priv->params.tx_pauseframe_control, 0,
2836 	    "Set to enable TX pause frames. Clear to disable.");
2837 
2838 	SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
2839 	    OID_AUTO, "rx_pauseframe_control", CTLFLAG_RDTUN,
2840 	    &priv->params.rx_pauseframe_control, 0,
2841 	    "Set to enable RX pause frames. Clear to disable.");
2842 
2843 	/* range check */
2844 	priv->params.tx_pauseframe_control =
2845 	    priv->params.tx_pauseframe_control ? 1 : 0;
2846 	priv->params.rx_pauseframe_control =
2847 	    priv->params.rx_pauseframe_control ? 1 : 0;
2848 
2849 	/* update firmware */
2850 	mlx5_set_port_pause(priv->mdev, 1,
2851 	    priv->params.rx_pauseframe_control,
2852 	    priv->params.tx_pauseframe_control);
2853 }
2854 
2855 static void *
2856 mlx5e_create_ifp(struct mlx5_core_dev *mdev)
2857 {
2858 	static volatile int mlx5_en_unit;
2859 	struct ifnet *ifp;
2860 	struct mlx5e_priv *priv;
2861 	u8 dev_addr[ETHER_ADDR_LEN] __aligned(4);
2862 	struct sysctl_oid_list *child;
2863 	int ncv = mdev->priv.eq_table.num_comp_vectors;
2864 	char unit[16];
2865 	int err;
2866 	int i;
2867 	u32 eth_proto_cap;
2868 
2869 	if (mlx5e_check_required_hca_cap(mdev)) {
2870 		mlx5_core_dbg(mdev, "mlx5e_check_required_hca_cap() failed\n");
2871 		return (NULL);
2872 	}
2873 	priv = malloc(sizeof(*priv), M_MLX5EN, M_WAITOK | M_ZERO);
2874 	if (priv == NULL) {
2875 		mlx5_core_err(mdev, "malloc() failed\n");
2876 		return (NULL);
2877 	}
2878 	mlx5e_priv_mtx_init(priv);
2879 
2880 	ifp = priv->ifp = if_alloc(IFT_ETHER);
2881 	if (ifp == NULL) {
2882 		mlx5_core_err(mdev, "if_alloc() failed\n");
2883 		goto err_free_priv;
2884 	}
2885 	ifp->if_softc = priv;
2886 	if_initname(ifp, "mce", atomic_fetchadd_int(&mlx5_en_unit, 1));
2887 	ifp->if_mtu = ETHERMTU;
2888 	ifp->if_init = mlx5e_open;
2889 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2890 	ifp->if_ioctl = mlx5e_ioctl;
2891 	ifp->if_transmit = mlx5e_xmit;
2892 	ifp->if_qflush = if_qflush;
2893 #if (__FreeBSD_version >= 1100000)
2894 	ifp->if_get_counter = mlx5e_get_counter;
2895 #endif
2896 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
2897 	/*
2898          * Set driver features
2899          */
2900 	ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6;
2901 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
2902 	ifp->if_capabilities |= IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWFILTER;
2903 	ifp->if_capabilities |= IFCAP_LINKSTATE | IFCAP_JUMBO_MTU;
2904 	ifp->if_capabilities |= IFCAP_LRO;
2905 	ifp->if_capabilities |= IFCAP_TSO | IFCAP_VLAN_HWTSO;
2906 
2907 	/* set TSO limits so that we don't have to drop TX packets */
2908 	ifp->if_hw_tsomax = MLX5E_MAX_TX_PAYLOAD_SIZE - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2909 	ifp->if_hw_tsomaxsegcount = MLX5E_MAX_TX_MBUF_FRAGS - 1 /* hdr */;
2910 	ifp->if_hw_tsomaxsegsize = MLX5E_MAX_TX_MBUF_SIZE;
2911 
2912 	ifp->if_capenable = ifp->if_capabilities;
2913 	ifp->if_hwassist = 0;
2914 	if (ifp->if_capenable & IFCAP_TSO)
2915 		ifp->if_hwassist |= CSUM_TSO;
2916 	if (ifp->if_capenable & IFCAP_TXCSUM)
2917 		ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP | CSUM_IP);
2918 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
2919 		ifp->if_hwassist |= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
2920 
2921 	/* ifnet sysctl tree */
2922 	sysctl_ctx_init(&priv->sysctl_ctx);
2923 	priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_dev),
2924 	    OID_AUTO, ifp->if_dname, CTLFLAG_RD, 0, "MLX5 ethernet - interface name");
2925 	if (priv->sysctl_ifnet == NULL) {
2926 		mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
2927 		goto err_free_sysctl;
2928 	}
2929 	snprintf(unit, sizeof(unit), "%d", ifp->if_dunit);
2930 	priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
2931 	    OID_AUTO, unit, CTLFLAG_RD, 0, "MLX5 ethernet - interface unit");
2932 	if (priv->sysctl_ifnet == NULL) {
2933 		mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
2934 		goto err_free_sysctl;
2935 	}
2936 
2937 	/* HW sysctl tree */
2938 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(mdev->pdev->dev.bsddev));
2939 	priv->sysctl_hw = SYSCTL_ADD_NODE(&priv->sysctl_ctx, child,
2940 	    OID_AUTO, "hw", CTLFLAG_RD, 0, "MLX5 ethernet dev hw");
2941 	if (priv->sysctl_hw == NULL) {
2942 		mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
2943 		goto err_free_sysctl;
2944 	}
2945 	mlx5e_build_ifp_priv(mdev, priv, ncv);
2946 	err = mlx5_alloc_map_uar(mdev, &priv->cq_uar);
2947 	if (err) {
2948 		if_printf(ifp, "%s: mlx5_alloc_map_uar failed, %d\n",
2949 		    __func__, err);
2950 		goto err_free_sysctl;
2951 	}
2952 	err = mlx5_core_alloc_pd(mdev, &priv->pdn);
2953 	if (err) {
2954 		if_printf(ifp, "%s: mlx5_core_alloc_pd failed, %d\n",
2955 		    __func__, err);
2956 		goto err_unmap_free_uar;
2957 	}
2958 	err = mlx5_alloc_transport_domain(mdev, &priv->tdn);
2959 	if (err) {
2960 		if_printf(ifp, "%s: mlx5_alloc_transport_domain failed, %d\n",
2961 		    __func__, err);
2962 		goto err_dealloc_pd;
2963 	}
2964 	err = mlx5e_create_mkey(priv, priv->pdn, &priv->mr);
2965 	if (err) {
2966 		if_printf(ifp, "%s: mlx5e_create_mkey failed, %d\n",
2967 		    __func__, err);
2968 		goto err_dealloc_transport_domain;
2969 	}
2970 	mlx5_query_nic_vport_mac_address(priv->mdev, 0, dev_addr);
2971 
2972 	/* check if we should generate a random MAC address */
2973 	if (MLX5_CAP_GEN(priv->mdev, vport_group_manager) == 0 &&
2974 	    is_zero_ether_addr(dev_addr)) {
2975 		random_ether_addr(dev_addr);
2976 		if_printf(ifp, "Assigned random MAC address\n");
2977 	}
2978 
2979 	/* set default MTU */
2980 	mlx5e_set_dev_port_mtu(ifp, ifp->if_mtu);
2981 
2982 	/* Set desc */
2983 	device_set_desc(mdev->pdev->dev.bsddev, mlx5e_version);
2984 
2985 	/* Set default media status */
2986 	priv->media_status_last = IFM_AVALID;
2987 	priv->media_active_last = IFM_ETHER | IFM_AUTO |
2988 	    IFM_ETH_RXPAUSE | IFM_FDX;
2989 
2990 	/* setup default pauseframes configuration */
2991 	mlx5e_setup_pauseframes(priv);
2992 
2993 	err = mlx5_query_port_proto_cap(mdev, &eth_proto_cap, MLX5_PTYS_EN);
2994 	if (err) {
2995 		eth_proto_cap = 0;
2996 		if_printf(ifp, "%s: Query port media capability failed, %d\n",
2997 		    __func__, err);
2998 	}
2999 
3000 	/* Setup supported medias */
3001 	ifmedia_init(&priv->media, IFM_IMASK | IFM_ETH_FMASK,
3002 	    mlx5e_media_change, mlx5e_media_status);
3003 
3004 	for (i = 0; i < MLX5E_LINK_MODES_NUMBER; ++i) {
3005 		if (mlx5e_mode_table[i].baudrate == 0)
3006 			continue;
3007 		if (MLX5E_PROT_MASK(i) & eth_proto_cap) {
3008 			ifmedia_add(&priv->media,
3009 			    mlx5e_mode_table[i].subtype |
3010 			    IFM_ETHER, 0, NULL);
3011 			ifmedia_add(&priv->media,
3012 			    mlx5e_mode_table[i].subtype |
3013 			    IFM_ETHER | IFM_FDX |
3014 			    IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3015 		}
3016 	}
3017 
3018 	ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO, 0, NULL);
3019 	ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
3020 	    IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3021 
3022 	/* Set autoselect by default */
3023 	ifmedia_set(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
3024 	    IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE);
3025 	ether_ifattach(ifp, dev_addr);
3026 
3027 	/* Register for VLAN events */
3028 	priv->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
3029 	    mlx5e_vlan_rx_add_vid, priv, EVENTHANDLER_PRI_FIRST);
3030 	priv->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
3031 	    mlx5e_vlan_rx_kill_vid, priv, EVENTHANDLER_PRI_FIRST);
3032 
3033 	/* Link is down by default */
3034 	if_link_state_change(ifp, LINK_STATE_DOWN);
3035 
3036 	mlx5e_enable_async_events(priv);
3037 
3038 	mlx5e_add_hw_stats(priv);
3039 
3040 	mlx5e_create_stats(&priv->stats.vport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3041 	    "vstats", mlx5e_vport_stats_desc, MLX5E_VPORT_STATS_NUM,
3042 	    priv->stats.vport.arg);
3043 
3044 	mlx5e_create_stats(&priv->stats.pport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3045 	    "pstats", mlx5e_pport_stats_desc, MLX5E_PPORT_STATS_NUM,
3046 	    priv->stats.pport.arg);
3047 
3048 	mlx5e_create_ethtool(priv);
3049 
3050 	mtx_lock(&priv->async_events_mtx);
3051 	mlx5e_update_stats(priv);
3052 	mtx_unlock(&priv->async_events_mtx);
3053 
3054 	return (priv);
3055 
3056 err_dealloc_transport_domain:
3057 	mlx5_dealloc_transport_domain(mdev, priv->tdn);
3058 
3059 err_dealloc_pd:
3060 	mlx5_core_dealloc_pd(mdev, priv->pdn);
3061 
3062 err_unmap_free_uar:
3063 	mlx5_unmap_free_uar(mdev, &priv->cq_uar);
3064 
3065 err_free_sysctl:
3066 	sysctl_ctx_free(&priv->sysctl_ctx);
3067 
3068 	if_free(ifp);
3069 
3070 err_free_priv:
3071 	mlx5e_priv_mtx_destroy(priv);
3072 	free(priv, M_MLX5EN);
3073 	return (NULL);
3074 }
3075 
3076 static void
3077 mlx5e_destroy_ifp(struct mlx5_core_dev *mdev, void *vpriv)
3078 {
3079 	struct mlx5e_priv *priv = vpriv;
3080 	struct ifnet *ifp = priv->ifp;
3081 
3082 	/* don't allow more IOCTLs */
3083 	priv->gone = 1;
3084 
3085 	/*
3086 	 * Clear the device description to avoid use after free,
3087 	 * because the bsddev is not destroyed when this module is
3088 	 * unloaded:
3089 	 */
3090 	device_set_desc(mdev->pdev->dev.bsddev, NULL);
3091 
3092 	/* XXX wait a bit to allow IOCTL handlers to complete */
3093 	pause("W", hz);
3094 
3095 	/* stop watchdog timer */
3096 	callout_drain(&priv->watchdog);
3097 
3098 	if (priv->vlan_attach != NULL)
3099 		EVENTHANDLER_DEREGISTER(vlan_config, priv->vlan_attach);
3100 	if (priv->vlan_detach != NULL)
3101 		EVENTHANDLER_DEREGISTER(vlan_unconfig, priv->vlan_detach);
3102 
3103 	/* make sure device gets closed */
3104 	PRIV_LOCK(priv);
3105 	mlx5e_close_locked(ifp);
3106 	PRIV_UNLOCK(priv);
3107 
3108 	/* unregister device */
3109 	ifmedia_removeall(&priv->media);
3110 	ether_ifdetach(ifp);
3111 	if_free(ifp);
3112 
3113 	/* destroy all remaining sysctl nodes */
3114 	if (priv->sysctl_debug)
3115 		sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
3116 	sysctl_ctx_free(&priv->stats.vport.ctx);
3117 	sysctl_ctx_free(&priv->stats.pport.ctx);
3118 	sysctl_ctx_free(&priv->sysctl_ctx);
3119 
3120 	mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
3121 	mlx5_dealloc_transport_domain(priv->mdev, priv->tdn);
3122 	mlx5_core_dealloc_pd(priv->mdev, priv->pdn);
3123 	mlx5_unmap_free_uar(priv->mdev, &priv->cq_uar);
3124 	mlx5e_disable_async_events(priv);
3125 	flush_scheduled_work();
3126 	mlx5e_priv_mtx_destroy(priv);
3127 	free(priv, M_MLX5EN);
3128 }
3129 
3130 static void *
3131 mlx5e_get_ifp(void *vpriv)
3132 {
3133 	struct mlx5e_priv *priv = vpriv;
3134 
3135 	return (priv->ifp);
3136 }
3137 
3138 static struct mlx5_interface mlx5e_interface = {
3139 	.add = mlx5e_create_ifp,
3140 	.remove = mlx5e_destroy_ifp,
3141 	.event = mlx5e_async_event,
3142 	.protocol = MLX5_INTERFACE_PROTOCOL_ETH,
3143 	.get_dev = mlx5e_get_ifp,
3144 };
3145 
3146 void
3147 mlx5e_init(void)
3148 {
3149 	mlx5_register_interface(&mlx5e_interface);
3150 }
3151 
3152 void
3153 mlx5e_cleanup(void)
3154 {
3155 	mlx5_unregister_interface(&mlx5e_interface);
3156 }
3157 
3158 module_init_order(mlx5e_init, SI_ORDER_THIRD);
3159 module_exit_order(mlx5e_cleanup, SI_ORDER_THIRD);
3160 
3161 #if (__FreeBSD_version >= 1100000)
3162 MODULE_DEPEND(mlx5en, linuxkpi, 1, 1, 1);
3163 #endif
3164 MODULE_DEPEND(mlx5en, mlx5, 1, 1, 1);
3165 MODULE_VERSION(mlx5en, 1);
3166