xref: /freebsd/sys/dev/mlx5/mlx5_en/mlx5_en_main.c (revision ae41709ab46305df80f7f35bb478a3c8ebf22ebb)
1 /*-
2  * Copyright (c) 2015-2018 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 "opt_kern_tls.h"
29 
30 #include "en.h"
31 
32 #include <sys/eventhandler.h>
33 #include <sys/sockio.h>
34 #include <machine/atomic.h>
35 
36 #include <net/debugnet.h>
37 
38 #ifndef ETH_DRIVER_VERSION
39 #define	ETH_DRIVER_VERSION	"3.5.2"
40 #endif
41 #define DRIVER_RELDATE	"September 2019"
42 
43 static const char mlx5e_version[] = "mlx5en: Mellanox Ethernet driver "
44 	ETH_DRIVER_VERSION " (" DRIVER_RELDATE ")\n";
45 
46 static int mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs);
47 
48 struct mlx5e_channel_param {
49 	struct mlx5e_rq_param rq;
50 	struct mlx5e_sq_param sq;
51 	struct mlx5e_cq_param rx_cq;
52 	struct mlx5e_cq_param tx_cq;
53 };
54 
55 struct media {
56 	u32	subtype;
57 	u64	baudrate;
58 };
59 
60 static const struct media mlx5e_mode_table[MLX5E_LINK_SPEEDS_NUMBER][MLX5E_LINK_MODES_NUMBER] = {
61 
62 	[MLX5E_1000BASE_CX_SGMII][MLX5E_SGMII] = {
63 		.subtype = IFM_1000_CX_SGMII,
64 		.baudrate = IF_Mbps(1000ULL),
65 	},
66 	[MLX5E_1000BASE_KX][MLX5E_KX] = {
67 		.subtype = IFM_1000_KX,
68 		.baudrate = IF_Mbps(1000ULL),
69 	},
70 	[MLX5E_10GBASE_CX4][MLX5E_CX4] = {
71 		.subtype = IFM_10G_CX4,
72 		.baudrate = IF_Gbps(10ULL),
73 	},
74 	[MLX5E_10GBASE_KX4][MLX5E_KX4] = {
75 		.subtype = IFM_10G_KX4,
76 		.baudrate = IF_Gbps(10ULL),
77 	},
78 	[MLX5E_10GBASE_KR][MLX5E_KR] = {
79 		.subtype = IFM_10G_KR,
80 		.baudrate = IF_Gbps(10ULL),
81 	},
82 	[MLX5E_20GBASE_KR2][MLX5E_KR2] = {
83 		.subtype = IFM_20G_KR2,
84 		.baudrate = IF_Gbps(20ULL),
85 	},
86 	[MLX5E_40GBASE_CR4][MLX5E_CR4] = {
87 		.subtype = IFM_40G_CR4,
88 		.baudrate = IF_Gbps(40ULL),
89 	},
90 	[MLX5E_40GBASE_KR4][MLX5E_KR4] = {
91 		.subtype = IFM_40G_KR4,
92 		.baudrate = IF_Gbps(40ULL),
93 	},
94 	[MLX5E_56GBASE_R4][MLX5E_R] = {
95 		.subtype = IFM_56G_R4,
96 		.baudrate = IF_Gbps(56ULL),
97 	},
98 	[MLX5E_10GBASE_CR][MLX5E_CR1] = {
99 		.subtype = IFM_10G_CR1,
100 		.baudrate = IF_Gbps(10ULL),
101 	},
102 	[MLX5E_10GBASE_SR][MLX5E_SR] = {
103 		.subtype = IFM_10G_SR,
104 		.baudrate = IF_Gbps(10ULL),
105 	},
106 	[MLX5E_10GBASE_ER_LR][MLX5E_ER] = {
107 		.subtype = IFM_10G_ER,
108 		.baudrate = IF_Gbps(10ULL),
109 	},
110 	[MLX5E_10GBASE_ER_LR][MLX5E_LR] = {
111 		.subtype = IFM_10G_LR,
112 		.baudrate = IF_Gbps(10ULL),
113 	},
114 	[MLX5E_40GBASE_SR4][MLX5E_SR4] = {
115 		.subtype = IFM_40G_SR4,
116 		.baudrate = IF_Gbps(40ULL),
117 	},
118 	[MLX5E_40GBASE_LR4_ER4][MLX5E_LR4] = {
119 		.subtype = IFM_40G_LR4,
120 		.baudrate = IF_Gbps(40ULL),
121 	},
122 	[MLX5E_40GBASE_LR4_ER4][MLX5E_ER4] = {
123 		.subtype = IFM_40G_ER4,
124 		.baudrate = IF_Gbps(40ULL),
125 	},
126 	[MLX5E_100GBASE_CR4][MLX5E_CR4] = {
127 		.subtype = IFM_100G_CR4,
128 		.baudrate = IF_Gbps(100ULL),
129 	},
130 	[MLX5E_100GBASE_SR4][MLX5E_SR4] = {
131 		.subtype = IFM_100G_SR4,
132 		.baudrate = IF_Gbps(100ULL),
133 	},
134 	[MLX5E_100GBASE_KR4][MLX5E_KR4] = {
135 		.subtype = IFM_100G_KR4,
136 		.baudrate = IF_Gbps(100ULL),
137 	},
138 	[MLX5E_100GBASE_LR4][MLX5E_LR4] = {
139 		.subtype = IFM_100G_LR4,
140 		.baudrate = IF_Gbps(100ULL),
141 	},
142 	[MLX5E_100BASE_TX][MLX5E_TX] = {
143 		.subtype = IFM_100_TX,
144 		.baudrate = IF_Mbps(100ULL),
145 	},
146 	[MLX5E_1000BASE_T][MLX5E_T] = {
147 		.subtype = IFM_1000_T,
148 		.baudrate = IF_Mbps(1000ULL),
149 	},
150 	[MLX5E_10GBASE_T][MLX5E_T] = {
151 		.subtype = IFM_10G_T,
152 		.baudrate = IF_Gbps(10ULL),
153 	},
154 	[MLX5E_25GBASE_CR][MLX5E_CR] = {
155 		.subtype = IFM_25G_CR,
156 		.baudrate = IF_Gbps(25ULL),
157 	},
158 	[MLX5E_25GBASE_KR][MLX5E_KR] = {
159 		.subtype = IFM_25G_KR,
160 		.baudrate = IF_Gbps(25ULL),
161 	},
162 	[MLX5E_25GBASE_SR][MLX5E_SR] = {
163 		.subtype = IFM_25G_SR,
164 		.baudrate = IF_Gbps(25ULL),
165 	},
166 	[MLX5E_50GBASE_CR2][MLX5E_CR2] = {
167 		.subtype = IFM_50G_CR2,
168 		.baudrate = IF_Gbps(50ULL),
169 	},
170 	[MLX5E_50GBASE_KR2][MLX5E_KR2] = {
171 		.subtype = IFM_50G_KR2,
172 		.baudrate = IF_Gbps(50ULL),
173 	},
174 	[MLX5E_50GBASE_KR4][MLX5E_KR4] = {
175 		.subtype = IFM_50G_KR4,
176 		.baudrate = IF_Gbps(50ULL),
177 	},
178 };
179 
180 static const struct media mlx5e_ext_mode_table[MLX5E_EXT_LINK_SPEEDS_NUMBER][MLX5E_LINK_MODES_NUMBER] = {
181 	[MLX5E_SGMII_100M][MLX5E_SGMII] = {
182 		.subtype = IFM_100_SGMII,
183 		.baudrate = IF_Mbps(100),
184 	},
185 	[MLX5E_1000BASE_X_SGMII][MLX5E_KX] = {
186 		.subtype = IFM_1000_KX,
187 		.baudrate = IF_Mbps(1000),
188 	},
189 	[MLX5E_1000BASE_X_SGMII][MLX5E_CX_SGMII] = {
190 		.subtype = IFM_1000_CX_SGMII,
191 		.baudrate = IF_Mbps(1000),
192 	},
193 	[MLX5E_1000BASE_X_SGMII][MLX5E_CX] = {
194 		.subtype = IFM_1000_CX,
195 		.baudrate = IF_Mbps(1000),
196 	},
197 	[MLX5E_1000BASE_X_SGMII][MLX5E_LX] = {
198 		.subtype = IFM_1000_LX,
199 		.baudrate = IF_Mbps(1000),
200 	},
201 	[MLX5E_1000BASE_X_SGMII][MLX5E_SX] = {
202 		.subtype = IFM_1000_SX,
203 		.baudrate = IF_Mbps(1000),
204 	},
205 	[MLX5E_1000BASE_X_SGMII][MLX5E_T] = {
206 		.subtype = IFM_1000_T,
207 		.baudrate = IF_Mbps(1000),
208 	},
209 	[MLX5E_5GBASE_R][MLX5E_T] = {
210 		.subtype = IFM_5000_T,
211 		.baudrate = IF_Mbps(5000),
212 	},
213 	[MLX5E_5GBASE_R][MLX5E_KR] = {
214 		.subtype = IFM_5000_KR,
215 		.baudrate = IF_Mbps(5000),
216 	},
217 	[MLX5E_5GBASE_R][MLX5E_KR1] = {
218 		.subtype = IFM_5000_KR1,
219 		.baudrate = IF_Mbps(5000),
220 	},
221 	[MLX5E_5GBASE_R][MLX5E_KR_S] = {
222 		.subtype = IFM_5000_KR_S,
223 		.baudrate = IF_Mbps(5000),
224 	},
225 	[MLX5E_10GBASE_XFI_XAUI_1][MLX5E_ER] = {
226 		.subtype = IFM_10G_ER,
227 		.baudrate = IF_Gbps(10ULL),
228 	},
229 	[MLX5E_10GBASE_XFI_XAUI_1][MLX5E_KR] = {
230 		.subtype = IFM_10G_KR,
231 		.baudrate = IF_Gbps(10ULL),
232 	},
233 	[MLX5E_10GBASE_XFI_XAUI_1][MLX5E_LR] = {
234 		.subtype = IFM_10G_LR,
235 		.baudrate = IF_Gbps(10ULL),
236 	},
237 	[MLX5E_10GBASE_XFI_XAUI_1][MLX5E_SR] = {
238 		.subtype = IFM_10G_SR,
239 		.baudrate = IF_Gbps(10ULL),
240 	},
241 	[MLX5E_10GBASE_XFI_XAUI_1][MLX5E_T] = {
242 		.subtype = IFM_10G_T,
243 		.baudrate = IF_Gbps(10ULL),
244 	},
245 	[MLX5E_10GBASE_XFI_XAUI_1][MLX5E_AOC] = {
246 		.subtype = IFM_10G_AOC,
247 		.baudrate = IF_Gbps(10ULL),
248 	},
249 	[MLX5E_10GBASE_XFI_XAUI_1][MLX5E_CR1] = {
250 		.subtype = IFM_10G_CR1,
251 		.baudrate = IF_Gbps(10ULL),
252 	},
253 	[MLX5E_40GBASE_XLAUI_4_XLPPI_4][MLX5E_CR4] = {
254 		.subtype = IFM_40G_CR4,
255 		.baudrate = IF_Gbps(40ULL),
256 	},
257 	[MLX5E_40GBASE_XLAUI_4_XLPPI_4][MLX5E_KR4] = {
258 		.subtype = IFM_40G_KR4,
259 		.baudrate = IF_Gbps(40ULL),
260 	},
261 	[MLX5E_40GBASE_XLAUI_4_XLPPI_4][MLX5E_LR4] = {
262 		.subtype = IFM_40G_LR4,
263 		.baudrate = IF_Gbps(40ULL),
264 	},
265 	[MLX5E_40GBASE_XLAUI_4_XLPPI_4][MLX5E_SR4] = {
266 		.subtype = IFM_40G_SR4,
267 		.baudrate = IF_Gbps(40ULL),
268 	},
269 	[MLX5E_40GBASE_XLAUI_4_XLPPI_4][MLX5E_ER4] = {
270 		.subtype = IFM_40G_ER4,
271 		.baudrate = IF_Gbps(40ULL),
272 	},
273 
274 	[MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_CR] = {
275 		.subtype = IFM_25G_CR,
276 		.baudrate = IF_Gbps(25ULL),
277 	},
278 	[MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_KR] = {
279 		.subtype = IFM_25G_KR,
280 		.baudrate = IF_Gbps(25ULL),
281 	},
282 	[MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_SR] = {
283 		.subtype = IFM_25G_SR,
284 		.baudrate = IF_Gbps(25ULL),
285 	},
286 	[MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_ACC] = {
287 		.subtype = IFM_25G_ACC,
288 		.baudrate = IF_Gbps(25ULL),
289 	},
290 	[MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_AOC] = {
291 		.subtype = IFM_25G_AOC,
292 		.baudrate = IF_Gbps(25ULL),
293 	},
294 	[MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_CR1] = {
295 		.subtype = IFM_25G_CR1,
296 		.baudrate = IF_Gbps(25ULL),
297 	},
298 	[MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_CR_S] = {
299 		.subtype = IFM_25G_CR_S,
300 		.baudrate = IF_Gbps(25ULL),
301 	},
302 	[MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_KR1] = {
303 		.subtype = IFM_5000_KR1,
304 		.baudrate = IF_Gbps(25ULL),
305 	},
306 	[MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_KR_S] = {
307 		.subtype = IFM_25G_KR_S,
308 		.baudrate = IF_Gbps(25ULL),
309 	},
310 	[MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_LR] = {
311 		.subtype = IFM_25G_LR,
312 		.baudrate = IF_Gbps(25ULL),
313 	},
314 	[MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_T] = {
315 		.subtype = IFM_25G_T,
316 		.baudrate = IF_Gbps(25ULL),
317 	},
318 	[MLX5E_50GAUI_2_LAUI_2_50GBASE_CR2_KR2][MLX5E_CR2] = {
319 		.subtype = IFM_50G_CR2,
320 		.baudrate = IF_Gbps(50ULL),
321 	},
322 	[MLX5E_50GAUI_2_LAUI_2_50GBASE_CR2_KR2][MLX5E_KR2] = {
323 		.subtype = IFM_50G_KR2,
324 		.baudrate = IF_Gbps(50ULL),
325 	},
326 	[MLX5E_50GAUI_2_LAUI_2_50GBASE_CR2_KR2][MLX5E_KR4] = {
327 		.subtype = IFM_50G_KR4,
328 		.baudrate = IF_Gbps(50ULL),
329 	},
330 	[MLX5E_50GAUI_2_LAUI_2_50GBASE_CR2_KR2][MLX5E_SR2] = {
331 		.subtype = IFM_50G_SR2,
332 		.baudrate = IF_Gbps(50ULL),
333 	},
334 	[MLX5E_50GAUI_2_LAUI_2_50GBASE_CR2_KR2][MLX5E_LR2] = {
335 		.subtype = IFM_50G_LR2,
336 		.baudrate = IF_Gbps(50ULL),
337 	},
338 	[MLX5E_50GAUI_1_LAUI_1_50GBASE_CR_KR][MLX5E_LR] = {
339 		.subtype = IFM_50G_LR,
340 		.baudrate = IF_Gbps(50ULL),
341 	},
342 	[MLX5E_50GAUI_1_LAUI_1_50GBASE_CR_KR][MLX5E_SR] = {
343 		.subtype = IFM_50G_SR,
344 		.baudrate = IF_Gbps(50ULL),
345 	},
346 	[MLX5E_50GAUI_1_LAUI_1_50GBASE_CR_KR][MLX5E_CP] = {
347 		.subtype = IFM_50G_CP,
348 		.baudrate = IF_Gbps(50ULL),
349 	},
350 	[MLX5E_50GAUI_1_LAUI_1_50GBASE_CR_KR][MLX5E_FR] = {
351 		.subtype = IFM_50G_FR,
352 		.baudrate = IF_Gbps(50ULL),
353 	},
354 	[MLX5E_50GAUI_1_LAUI_1_50GBASE_CR_KR][MLX5E_KR_PAM4] = {
355 		.subtype = IFM_50G_KR_PAM4,
356 		.baudrate = IF_Gbps(50ULL),
357 	},
358 	[MLX5E_CAUI_4_100GBASE_CR4_KR4][MLX5E_CR4] = {
359 		.subtype = IFM_100G_CR4,
360 		.baudrate = IF_Gbps(100ULL),
361 	},
362 	[MLX5E_CAUI_4_100GBASE_CR4_KR4][MLX5E_KR4] = {
363 		.subtype = IFM_100G_KR4,
364 		.baudrate = IF_Gbps(100ULL),
365 	},
366 	[MLX5E_CAUI_4_100GBASE_CR4_KR4][MLX5E_LR4] = {
367 		.subtype = IFM_100G_LR4,
368 		.baudrate = IF_Gbps(100ULL),
369 	},
370 	[MLX5E_CAUI_4_100GBASE_CR4_KR4][MLX5E_SR4] = {
371 		.subtype = IFM_100G_SR4,
372 		.baudrate = IF_Gbps(100ULL),
373 	},
374 	[MLX5E_100GAUI_2_100GBASE_CR2_KR2][MLX5E_SR2] = {
375 		.subtype = IFM_100G_SR2,
376 		.baudrate = IF_Gbps(100ULL),
377 	},
378 	[MLX5E_100GAUI_2_100GBASE_CR2_KR2][MLX5E_CP2] = {
379 		.subtype = IFM_100G_CP2,
380 		.baudrate = IF_Gbps(100ULL),
381 	},
382 	[MLX5E_100GAUI_2_100GBASE_CR2_KR2][MLX5E_KR2_PAM4] = {
383 		.subtype = IFM_100G_KR2_PAM4,
384 		.baudrate = IF_Gbps(100ULL),
385 	},
386 	[MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_DR4] = {
387 		.subtype = IFM_200G_DR4,
388 		.baudrate = IF_Gbps(200ULL),
389 	},
390 	[MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_LR4] = {
391 		.subtype = IFM_200G_LR4,
392 		.baudrate = IF_Gbps(200ULL),
393 	},
394 	[MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_SR4] = {
395 		.subtype = IFM_200G_SR4,
396 		.baudrate = IF_Gbps(200ULL),
397 	},
398 	[MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_FR4] = {
399 		.subtype = IFM_200G_FR4,
400 		.baudrate = IF_Gbps(200ULL),
401 	},
402 	[MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_CR4_PAM4] = {
403 		.subtype = IFM_200G_CR4_PAM4,
404 		.baudrate = IF_Gbps(200ULL),
405 	},
406 	[MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_KR4_PAM4] = {
407 		.subtype = IFM_200G_KR4_PAM4,
408 		.baudrate = IF_Gbps(200ULL),
409 	},
410 };
411 
412 DEBUGNET_DEFINE(mlx5_en);
413 
414 MALLOC_DEFINE(M_MLX5EN, "MLX5EN", "MLX5 Ethernet");
415 
416 static void
417 mlx5e_update_carrier(struct mlx5e_priv *priv)
418 {
419 	struct mlx5_core_dev *mdev = priv->mdev;
420 	u32 out[MLX5_ST_SZ_DW(ptys_reg)];
421 	u32 eth_proto_oper;
422 	int error;
423 	u8 port_state;
424 	u8 is_er_type;
425 	u8 i, j;
426 	bool ext;
427 	struct media media_entry = {};
428 
429 	port_state = mlx5_query_vport_state(mdev,
430 	    MLX5_QUERY_VPORT_STATE_IN_OP_MOD_VNIC_VPORT, 0);
431 
432 	if (port_state == VPORT_STATE_UP) {
433 		priv->media_status_last |= IFM_ACTIVE;
434 	} else {
435 		priv->media_status_last &= ~IFM_ACTIVE;
436 		priv->media_active_last = IFM_ETHER;
437 		if_link_state_change(priv->ifp, LINK_STATE_DOWN);
438 		return;
439 	}
440 
441 	error = mlx5_query_port_ptys(mdev, out, sizeof(out),
442 	    MLX5_PTYS_EN, 1);
443 	if (error) {
444 		priv->media_active_last = IFM_ETHER;
445 		priv->ifp->if_baudrate = 1;
446 		mlx5_en_err(priv->ifp, "query port ptys failed: 0x%x\n",
447 		    error);
448 		return;
449 	}
450 
451 	ext = MLX5_CAP_PCAM_FEATURE(mdev, ptys_extended_ethernet);
452 	eth_proto_oper = MLX5_GET_ETH_PROTO(ptys_reg, out, ext,
453 	    eth_proto_oper);
454 
455 	i = ilog2(eth_proto_oper);
456 
457 	for (j = 0; j != MLX5E_LINK_MODES_NUMBER; j++) {
458 		media_entry = ext ? mlx5e_ext_mode_table[i][j] :
459 		    mlx5e_mode_table[i][j];
460 		if (media_entry.baudrate != 0)
461 			break;
462 	}
463 
464 	if (media_entry.subtype == 0) {
465 		mlx5_en_err(priv->ifp,
466 		    "Could not find operational media subtype\n");
467 		return;
468 	}
469 
470 	switch (media_entry.subtype) {
471 	case IFM_10G_ER:
472 		error = mlx5_query_pddr_range_info(mdev, 1, &is_er_type);
473 		if (error != 0) {
474 			mlx5_en_err(priv->ifp,
475 			    "query port pddr failed: %d\n", error);
476 		}
477 		if (error != 0 || is_er_type == 0)
478 			media_entry.subtype = IFM_10G_LR;
479 		break;
480 	case IFM_40G_LR4:
481 		error = mlx5_query_pddr_range_info(mdev, 1, &is_er_type);
482 		if (error != 0) {
483 			mlx5_en_err(priv->ifp,
484 			    "query port pddr failed: %d\n", error);
485 		}
486 		if (error == 0 && is_er_type != 0)
487 			media_entry.subtype = IFM_40G_ER4;
488 		break;
489 	}
490 	priv->media_active_last = media_entry.subtype | IFM_ETHER | IFM_FDX;
491 	priv->ifp->if_baudrate = media_entry.baudrate;
492 
493 	if_link_state_change(priv->ifp, LINK_STATE_UP);
494 }
495 
496 static void
497 mlx5e_media_status(struct ifnet *dev, struct ifmediareq *ifmr)
498 {
499 	struct mlx5e_priv *priv = dev->if_softc;
500 
501 	ifmr->ifm_status = priv->media_status_last;
502 	ifmr->ifm_active = priv->media_active_last |
503 	    (priv->params.rx_pauseframe_control ? IFM_ETH_RXPAUSE : 0) |
504 	    (priv->params.tx_pauseframe_control ? IFM_ETH_TXPAUSE : 0);
505 
506 }
507 
508 static u32
509 mlx5e_find_link_mode(u32 subtype, bool ext)
510 {
511 	u32 i;
512 	u32 j;
513 	u32 link_mode = 0;
514 	u32 speeds_num = 0;
515 	struct media media_entry = {};
516 
517 	switch (subtype) {
518 	case IFM_10G_LR:
519 		subtype = IFM_10G_ER;
520 		break;
521 	case IFM_40G_ER4:
522 		subtype = IFM_40G_LR4;
523 		break;
524 	}
525 
526 	speeds_num = ext ? MLX5E_EXT_LINK_SPEEDS_NUMBER :
527 	    MLX5E_LINK_SPEEDS_NUMBER;
528 
529 	for (i = 0; i != speeds_num; i++) {
530 		for (j = 0; j < MLX5E_LINK_MODES_NUMBER ; ++j) {
531 			media_entry = ext ? mlx5e_ext_mode_table[i][j] :
532 			    mlx5e_mode_table[i][j];
533 			if (media_entry.baudrate == 0)
534 				continue;
535 			if (media_entry.subtype == subtype) {
536 				link_mode |= MLX5E_PROT_MASK(i);
537 			}
538 		}
539 	}
540 
541 	return (link_mode);
542 }
543 
544 static int
545 mlx5e_set_port_pause_and_pfc(struct mlx5e_priv *priv)
546 {
547 	return (mlx5_set_port_pause_and_pfc(priv->mdev, 1,
548 	    priv->params.rx_pauseframe_control,
549 	    priv->params.tx_pauseframe_control,
550 	    priv->params.rx_priority_flow_control,
551 	    priv->params.tx_priority_flow_control));
552 }
553 
554 static int
555 mlx5e_set_port_pfc(struct mlx5e_priv *priv)
556 {
557 	int error;
558 
559 	if (priv->gone != 0) {
560 		error = -ENXIO;
561 	} else if (priv->params.rx_pauseframe_control ||
562 	    priv->params.tx_pauseframe_control) {
563 		mlx5_en_err(priv->ifp,
564 		    "Global pauseframes must be disabled before enabling PFC.\n");
565 		error = -EINVAL;
566 	} else {
567 		error = mlx5e_set_port_pause_and_pfc(priv);
568 	}
569 	return (error);
570 }
571 
572 static int
573 mlx5e_media_change(struct ifnet *dev)
574 {
575 	struct mlx5e_priv *priv = dev->if_softc;
576 	struct mlx5_core_dev *mdev = priv->mdev;
577 	u32 eth_proto_cap;
578 	u32 link_mode;
579 	u32 out[MLX5_ST_SZ_DW(ptys_reg)];
580 	int was_opened;
581 	int locked;
582 	int error;
583 	bool ext;
584 
585 	locked = PRIV_LOCKED(priv);
586 	if (!locked)
587 		PRIV_LOCK(priv);
588 
589 	if (IFM_TYPE(priv->media.ifm_media) != IFM_ETHER) {
590 		error = EINVAL;
591 		goto done;
592 	}
593 
594 	error = mlx5_query_port_ptys(mdev, out, sizeof(out),
595 	    MLX5_PTYS_EN, 1);
596 	if (error != 0) {
597 		mlx5_en_err(dev, "Query port media capability failed\n");
598 		goto done;
599 	}
600 
601 	ext = MLX5_CAP_PCAM_FEATURE(mdev, ptys_extended_ethernet);
602 	link_mode = mlx5e_find_link_mode(IFM_SUBTYPE(priv->media.ifm_media), ext);
603 
604 	/* query supported capabilities */
605 	eth_proto_cap = MLX5_GET_ETH_PROTO(ptys_reg, out, ext,
606 	    eth_proto_capability);
607 
608 	/* check for autoselect */
609 	if (IFM_SUBTYPE(priv->media.ifm_media) == IFM_AUTO) {
610 		link_mode = eth_proto_cap;
611 		if (link_mode == 0) {
612 			mlx5_en_err(dev, "Port media capability is zero\n");
613 			error = EINVAL;
614 			goto done;
615 		}
616 	} else {
617 		link_mode = link_mode & eth_proto_cap;
618 		if (link_mode == 0) {
619 			mlx5_en_err(dev, "Not supported link mode requested\n");
620 			error = EINVAL;
621 			goto done;
622 		}
623 	}
624 	if (priv->media.ifm_media & (IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE)) {
625 		/* check if PFC is enabled */
626 		if (priv->params.rx_priority_flow_control ||
627 		    priv->params.tx_priority_flow_control) {
628 			mlx5_en_err(dev, "PFC must be disabled before enabling global pauseframes.\n");
629 			error = EINVAL;
630 			goto done;
631 		}
632 	}
633 	/* update pauseframe control bits */
634 	priv->params.rx_pauseframe_control =
635 	    (priv->media.ifm_media & IFM_ETH_RXPAUSE) ? 1 : 0;
636 	priv->params.tx_pauseframe_control =
637 	    (priv->media.ifm_media & IFM_ETH_TXPAUSE) ? 1 : 0;
638 
639 	/* check if device is opened */
640 	was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
641 
642 	/* reconfigure the hardware */
643 	mlx5_set_port_status(mdev, MLX5_PORT_DOWN);
644 	mlx5_set_port_proto(mdev, link_mode, MLX5_PTYS_EN, ext);
645 	error = -mlx5e_set_port_pause_and_pfc(priv);
646 	if (was_opened)
647 		mlx5_set_port_status(mdev, MLX5_PORT_UP);
648 
649 done:
650 	if (!locked)
651 		PRIV_UNLOCK(priv);
652 	return (error);
653 }
654 
655 static void
656 mlx5e_update_carrier_work(struct work_struct *work)
657 {
658 	struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
659 	    update_carrier_work);
660 
661 	PRIV_LOCK(priv);
662 	if (test_bit(MLX5E_STATE_OPENED, &priv->state))
663 		mlx5e_update_carrier(priv);
664 	PRIV_UNLOCK(priv);
665 }
666 
667 #define	MLX5E_PCIE_PERF_GET_64(a,b,c,d,e,f)    \
668 	s_debug->c = MLX5_GET64(mpcnt_reg, out, counter_set.f.c);
669 
670 #define	MLX5E_PCIE_PERF_GET_32(a,b,c,d,e,f)    \
671 	s_debug->c = MLX5_GET(mpcnt_reg, out, counter_set.f.c);
672 
673 static void
674 mlx5e_update_pcie_counters(struct mlx5e_priv *priv)
675 {
676 	struct mlx5_core_dev *mdev = priv->mdev;
677 	struct mlx5e_port_stats_debug *s_debug = &priv->stats.port_stats_debug;
678 	const unsigned sz = MLX5_ST_SZ_BYTES(mpcnt_reg);
679 	void *out;
680 	void *in;
681 	int err;
682 
683 	/* allocate firmware request structures */
684 	in = mlx5_vzalloc(sz);
685 	out = mlx5_vzalloc(sz);
686 	if (in == NULL || out == NULL)
687 		goto free_out;
688 
689 	MLX5_SET(mpcnt_reg, in, grp, MLX5_PCIE_PERFORMANCE_COUNTERS_GROUP);
690 	err = mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_MPCNT, 0, 0);
691 	if (err != 0)
692 		goto free_out;
693 
694 	MLX5E_PCIE_PERFORMANCE_COUNTERS_64(MLX5E_PCIE_PERF_GET_64)
695 	MLX5E_PCIE_PERFORMANCE_COUNTERS_32(MLX5E_PCIE_PERF_GET_32)
696 
697 	MLX5_SET(mpcnt_reg, in, grp, MLX5_PCIE_TIMERS_AND_STATES_COUNTERS_GROUP);
698 	err = mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_MPCNT, 0, 0);
699 	if (err != 0)
700 		goto free_out;
701 
702 	MLX5E_PCIE_TIMERS_AND_STATES_COUNTERS_32(MLX5E_PCIE_PERF_GET_32)
703 
704 	MLX5_SET(mpcnt_reg, in, grp, MLX5_PCIE_LANE_COUNTERS_GROUP);
705 	err = mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_MPCNT, 0, 0);
706 	if (err != 0)
707 		goto free_out;
708 
709 	MLX5E_PCIE_LANE_COUNTERS_32(MLX5E_PCIE_PERF_GET_32)
710 
711 free_out:
712 	/* free firmware request structures */
713 	kvfree(in);
714 	kvfree(out);
715 }
716 
717 /*
718  * This function reads the physical port counters from the firmware
719  * using a pre-defined layout defined by various MLX5E_PPORT_XXX()
720  * macros. The output is converted from big-endian 64-bit values into
721  * host endian ones and stored in the "priv->stats.pport" structure.
722  */
723 static void
724 mlx5e_update_pport_counters(struct mlx5e_priv *priv)
725 {
726 	struct mlx5_core_dev *mdev = priv->mdev;
727 	struct mlx5e_pport_stats *s = &priv->stats.pport;
728 	struct mlx5e_port_stats_debug *s_debug = &priv->stats.port_stats_debug;
729 	u32 *in;
730 	u32 *out;
731 	const u64 *ptr;
732 	unsigned sz = MLX5_ST_SZ_BYTES(ppcnt_reg);
733 	unsigned x;
734 	unsigned y;
735 	unsigned z;
736 
737 	/* allocate firmware request structures */
738 	in = mlx5_vzalloc(sz);
739 	out = mlx5_vzalloc(sz);
740 	if (in == NULL || out == NULL)
741 		goto free_out;
742 
743 	/*
744 	 * Get pointer to the 64-bit counter set which is located at a
745 	 * fixed offset in the output firmware request structure:
746 	 */
747 	ptr = (const uint64_t *)MLX5_ADDR_OF(ppcnt_reg, out, counter_set);
748 
749 	MLX5_SET(ppcnt_reg, in, local_port, 1);
750 
751 	/* read IEEE802_3 counter group using predefined counter layout */
752 	MLX5_SET(ppcnt_reg, in, grp, MLX5_IEEE_802_3_COUNTERS_GROUP);
753 	mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
754 	for (x = 0, y = MLX5E_PPORT_PER_PRIO_STATS_NUM;
755 	     x != MLX5E_PPORT_IEEE802_3_STATS_NUM; x++, y++)
756 		s->arg[y] = be64toh(ptr[x]);
757 
758 	/* read RFC2819 counter group using predefined counter layout */
759 	MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2819_COUNTERS_GROUP);
760 	mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
761 	for (x = 0; x != MLX5E_PPORT_RFC2819_STATS_NUM; x++, y++)
762 		s->arg[y] = be64toh(ptr[x]);
763 
764 	for (y = 0; x != MLX5E_PPORT_RFC2819_STATS_NUM +
765 	    MLX5E_PPORT_RFC2819_STATS_DEBUG_NUM; x++, y++)
766 		s_debug->arg[y] = be64toh(ptr[x]);
767 
768 	/* read RFC2863 counter group using predefined counter layout */
769 	MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2863_COUNTERS_GROUP);
770 	mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
771 	for (x = 0; x != MLX5E_PPORT_RFC2863_STATS_DEBUG_NUM; x++, y++)
772 		s_debug->arg[y] = be64toh(ptr[x]);
773 
774 	/* read physical layer stats counter group using predefined counter layout */
775 	MLX5_SET(ppcnt_reg, in, grp, MLX5_PHYSICAL_LAYER_COUNTERS_GROUP);
776 	mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
777 	for (x = 0; x != MLX5E_PPORT_PHYSICAL_LAYER_STATS_DEBUG_NUM; x++, y++)
778 		s_debug->arg[y] = be64toh(ptr[x]);
779 
780 	/* read Extended Ethernet counter group using predefined counter layout */
781 	MLX5_SET(ppcnt_reg, in, grp, MLX5_ETHERNET_EXTENDED_COUNTERS_GROUP);
782 	mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
783 	for (x = 0; x != MLX5E_PPORT_ETHERNET_EXTENDED_STATS_DEBUG_NUM; x++, y++)
784 		s_debug->arg[y] = be64toh(ptr[x]);
785 
786 	/* read Extended Statistical Group */
787 	if (MLX5_CAP_GEN(mdev, pcam_reg) &&
788 	    MLX5_CAP_PCAM_FEATURE(mdev, ppcnt_statistical_group) &&
789 	    MLX5_CAP_PCAM_FEATURE(mdev, per_lane_error_counters)) {
790 		/* read Extended Statistical counter group using predefined counter layout */
791 		MLX5_SET(ppcnt_reg, in, grp, MLX5_PHYSICAL_LAYER_STATISTICAL_GROUP);
792 		mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
793 
794 		for (x = 0; x != MLX5E_PPORT_STATISTICAL_DEBUG_NUM; x++, y++)
795 			s_debug->arg[y] = be64toh(ptr[x]);
796 	}
797 
798 	/* read PCIE counters */
799 	mlx5e_update_pcie_counters(priv);
800 
801 	/* read per-priority counters */
802 	MLX5_SET(ppcnt_reg, in, grp, MLX5_PER_PRIORITY_COUNTERS_GROUP);
803 
804 	/* iterate all the priorities */
805 	for (y = z = 0; z != MLX5E_PPORT_PER_PRIO_STATS_NUM_PRIO; z++) {
806 		MLX5_SET(ppcnt_reg, in, prio_tc, z);
807 		mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
808 
809 		/* read per priority stats counter group using predefined counter layout */
810 		for (x = 0; x != (MLX5E_PPORT_PER_PRIO_STATS_NUM /
811 		    MLX5E_PPORT_PER_PRIO_STATS_NUM_PRIO); x++, y++)
812 			s->arg[y] = be64toh(ptr[x]);
813 	}
814 
815 free_out:
816 	/* free firmware request structures */
817 	kvfree(in);
818 	kvfree(out);
819 }
820 
821 static void
822 mlx5e_grp_vnic_env_update_stats(struct mlx5e_priv *priv)
823 {
824 	u32 out[MLX5_ST_SZ_DW(query_vnic_env_out)] = {};
825 	u32 in[MLX5_ST_SZ_DW(query_vnic_env_in)] = {};
826 
827 	if (!MLX5_CAP_GEN(priv->mdev, nic_receive_steering_discard))
828 		return;
829 
830 	MLX5_SET(query_vnic_env_in, in, opcode,
831 	    MLX5_CMD_OP_QUERY_VNIC_ENV);
832 	MLX5_SET(query_vnic_env_in, in, op_mod, 0);
833 	MLX5_SET(query_vnic_env_in, in, other_vport, 0);
834 
835 	if (mlx5_cmd_exec(priv->mdev, in, sizeof(in), out, sizeof(out)) != 0)
836 		return;
837 
838 	priv->stats.vport.rx_steer_missed_packets =
839 	    MLX5_GET64(query_vnic_env_out, out,
840 	    vport_env.nic_receive_steering_discard);
841 }
842 
843 /*
844  * This function is called regularly to collect all statistics
845  * counters from the firmware. The values can be viewed through the
846  * sysctl interface. Execution is serialized using the priv's global
847  * configuration lock.
848  */
849 static void
850 mlx5e_update_stats_locked(struct mlx5e_priv *priv)
851 {
852 	struct mlx5_core_dev *mdev = priv->mdev;
853 	struct mlx5e_vport_stats *s = &priv->stats.vport;
854 	struct mlx5e_sq_stats *sq_stats;
855 #if (__FreeBSD_version < 1100000)
856 	struct ifnet *ifp = priv->ifp;
857 #endif
858 
859 	u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)];
860 	u32 *out;
861 	int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
862 	u64 tso_packets = 0;
863 	u64 tso_bytes = 0;
864 	u64 tx_queue_dropped = 0;
865 	u64 tx_defragged = 0;
866 	u64 tx_offload_none = 0;
867 	u64 lro_packets = 0;
868 	u64 lro_bytes = 0;
869 	u64 sw_lro_queued = 0;
870 	u64 sw_lro_flushed = 0;
871 	u64 rx_csum_none = 0;
872 	u64 rx_wqe_err = 0;
873 	u64 rx_packets = 0;
874 	u64 rx_bytes = 0;
875 	u32 rx_out_of_buffer = 0;
876 	int error;
877 	int i;
878 	int j;
879 
880 	out = mlx5_vzalloc(outlen);
881 	if (out == NULL)
882 		goto free_out;
883 
884 	/* Collect firts the SW counters and then HW for consistency */
885 	for (i = 0; i < priv->params.num_channels; i++) {
886 		struct mlx5e_channel *pch = priv->channel + i;
887 		struct mlx5e_rq *rq = &pch->rq;
888 		struct mlx5e_rq_stats *rq_stats = &pch->rq.stats;
889 
890 		/* collect stats from LRO */
891 		rq_stats->sw_lro_queued = rq->lro.lro_queued;
892 		rq_stats->sw_lro_flushed = rq->lro.lro_flushed;
893 		sw_lro_queued += rq_stats->sw_lro_queued;
894 		sw_lro_flushed += rq_stats->sw_lro_flushed;
895 		lro_packets += rq_stats->lro_packets;
896 		lro_bytes += rq_stats->lro_bytes;
897 		rx_csum_none += rq_stats->csum_none;
898 		rx_wqe_err += rq_stats->wqe_err;
899 		rx_packets += rq_stats->packets;
900 		rx_bytes += rq_stats->bytes;
901 
902 		for (j = 0; j < priv->num_tc; j++) {
903 			sq_stats = &pch->sq[j].stats;
904 
905 			tso_packets += sq_stats->tso_packets;
906 			tso_bytes += sq_stats->tso_bytes;
907 			tx_queue_dropped += sq_stats->dropped;
908 			tx_queue_dropped += sq_stats->enobuf;
909 			tx_defragged += sq_stats->defragged;
910 			tx_offload_none += sq_stats->csum_offload_none;
911 		}
912 	}
913 
914 	/* update counters */
915 	s->tso_packets = tso_packets;
916 	s->tso_bytes = tso_bytes;
917 	s->tx_queue_dropped = tx_queue_dropped;
918 	s->tx_defragged = tx_defragged;
919 	s->lro_packets = lro_packets;
920 	s->lro_bytes = lro_bytes;
921 	s->sw_lro_queued = sw_lro_queued;
922 	s->sw_lro_flushed = sw_lro_flushed;
923 	s->rx_csum_none = rx_csum_none;
924 	s->rx_wqe_err = rx_wqe_err;
925 	s->rx_packets = rx_packets;
926 	s->rx_bytes = rx_bytes;
927 
928 	mlx5e_grp_vnic_env_update_stats(priv);
929 
930 	/* HW counters */
931 	memset(in, 0, sizeof(in));
932 
933 	MLX5_SET(query_vport_counter_in, in, opcode,
934 	    MLX5_CMD_OP_QUERY_VPORT_COUNTER);
935 	MLX5_SET(query_vport_counter_in, in, op_mod, 0);
936 	MLX5_SET(query_vport_counter_in, in, other_vport, 0);
937 
938 	memset(out, 0, outlen);
939 
940 	/* get number of out-of-buffer drops first */
941 	if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0 &&
942 	    mlx5_vport_query_out_of_rx_buffer(mdev, priv->counter_set_id,
943 	    &rx_out_of_buffer) == 0) {
944 		s->rx_out_of_buffer = rx_out_of_buffer;
945 	}
946 
947 	/* get port statistics */
948 	if (mlx5_cmd_exec(mdev, in, sizeof(in), out, outlen) == 0) {
949 #define	MLX5_GET_CTR(out, x) \
950 	MLX5_GET64(query_vport_counter_out, out, x)
951 
952 		s->rx_error_packets =
953 		    MLX5_GET_CTR(out, received_errors.packets);
954 		s->rx_error_bytes =
955 		    MLX5_GET_CTR(out, received_errors.octets);
956 		s->tx_error_packets =
957 		    MLX5_GET_CTR(out, transmit_errors.packets);
958 		s->tx_error_bytes =
959 		    MLX5_GET_CTR(out, transmit_errors.octets);
960 
961 		s->rx_unicast_packets =
962 		    MLX5_GET_CTR(out, received_eth_unicast.packets);
963 		s->rx_unicast_bytes =
964 		    MLX5_GET_CTR(out, received_eth_unicast.octets);
965 		s->tx_unicast_packets =
966 		    MLX5_GET_CTR(out, transmitted_eth_unicast.packets);
967 		s->tx_unicast_bytes =
968 		    MLX5_GET_CTR(out, transmitted_eth_unicast.octets);
969 
970 		s->rx_multicast_packets =
971 		    MLX5_GET_CTR(out, received_eth_multicast.packets);
972 		s->rx_multicast_bytes =
973 		    MLX5_GET_CTR(out, received_eth_multicast.octets);
974 		s->tx_multicast_packets =
975 		    MLX5_GET_CTR(out, transmitted_eth_multicast.packets);
976 		s->tx_multicast_bytes =
977 		    MLX5_GET_CTR(out, transmitted_eth_multicast.octets);
978 
979 		s->rx_broadcast_packets =
980 		    MLX5_GET_CTR(out, received_eth_broadcast.packets);
981 		s->rx_broadcast_bytes =
982 		    MLX5_GET_CTR(out, received_eth_broadcast.octets);
983 		s->tx_broadcast_packets =
984 		    MLX5_GET_CTR(out, transmitted_eth_broadcast.packets);
985 		s->tx_broadcast_bytes =
986 		    MLX5_GET_CTR(out, transmitted_eth_broadcast.octets);
987 
988 		s->tx_packets = s->tx_unicast_packets +
989 		    s->tx_multicast_packets + s->tx_broadcast_packets;
990 		s->tx_bytes = s->tx_unicast_bytes + s->tx_multicast_bytes +
991 		    s->tx_broadcast_bytes;
992 
993 		/* Update calculated offload counters */
994 		s->tx_csum_offload = s->tx_packets - tx_offload_none;
995 		s->rx_csum_good = s->rx_packets - s->rx_csum_none;
996 	}
997 
998 	/* Get physical port counters */
999 	mlx5e_update_pport_counters(priv);
1000 
1001 	s->tx_jumbo_packets =
1002 	    priv->stats.port_stats_debug.tx_stat_p1519to2047octets +
1003 	    priv->stats.port_stats_debug.tx_stat_p2048to4095octets +
1004 	    priv->stats.port_stats_debug.tx_stat_p4096to8191octets +
1005 	    priv->stats.port_stats_debug.tx_stat_p8192to10239octets;
1006 
1007 #if (__FreeBSD_version < 1100000)
1008 	/* no get_counters interface in fbsd 10 */
1009 	ifp->if_ipackets = s->rx_packets;
1010 	ifp->if_ierrors = priv->stats.pport.in_range_len_errors +
1011 	    priv->stats.pport.out_of_range_len +
1012 	    priv->stats.pport.too_long_errors +
1013 	    priv->stats.pport.check_seq_err +
1014 	    priv->stats.pport.alignment_err;
1015 	ifp->if_iqdrops = s->rx_out_of_buffer;
1016 	ifp->if_opackets = s->tx_packets;
1017 	ifp->if_oerrors = priv->stats.port_stats_debug.out_discards;
1018 	ifp->if_snd.ifq_drops = s->tx_queue_dropped;
1019 	ifp->if_ibytes = s->rx_bytes;
1020 	ifp->if_obytes = s->tx_bytes;
1021 	ifp->if_collisions =
1022 	    priv->stats.pport.collisions;
1023 #endif
1024 
1025 free_out:
1026 	kvfree(out);
1027 
1028 	/* Update diagnostics, if any */
1029 	if (priv->params_ethtool.diag_pci_enable ||
1030 	    priv->params_ethtool.diag_general_enable) {
1031 		error = mlx5_core_get_diagnostics_full(mdev,
1032 		    priv->params_ethtool.diag_pci_enable ? &priv->params_pci : NULL,
1033 		    priv->params_ethtool.diag_general_enable ? &priv->params_general : NULL);
1034 		if (error != 0)
1035 			mlx5_en_err(priv->ifp,
1036 			    "Failed reading diagnostics: %d\n", error);
1037 	}
1038 
1039 	/* Update FEC, if any */
1040 	error = mlx5e_fec_update(priv);
1041 	if (error != 0 && error != EOPNOTSUPP) {
1042 		mlx5_en_err(priv->ifp,
1043 		    "Updating FEC failed: %d\n", error);
1044 	}
1045 
1046 	/* Update temperature, if any */
1047 	if (priv->params_ethtool.hw_num_temp != 0) {
1048 		error = mlx5e_hw_temperature_update(priv);
1049 		if (error != 0 && error != EOPNOTSUPP) {
1050 			mlx5_en_err(priv->ifp,
1051 			    "Updating temperature failed: %d\n", error);
1052 		}
1053 	}
1054 }
1055 
1056 static void
1057 mlx5e_update_stats_work(struct work_struct *work)
1058 {
1059 	struct mlx5e_priv *priv;
1060 
1061 	priv = container_of(work, struct mlx5e_priv, update_stats_work);
1062 	PRIV_LOCK(priv);
1063 	if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0 &&
1064 	    !test_bit(MLX5_INTERFACE_STATE_TEARDOWN, &priv->mdev->intf_state))
1065 		mlx5e_update_stats_locked(priv);
1066 	PRIV_UNLOCK(priv);
1067 }
1068 
1069 static void
1070 mlx5e_update_stats(void *arg)
1071 {
1072 	struct mlx5e_priv *priv = arg;
1073 
1074 	queue_work(priv->wq, &priv->update_stats_work);
1075 
1076 	callout_reset(&priv->watchdog, hz / 4, &mlx5e_update_stats, priv);
1077 }
1078 
1079 static void
1080 mlx5e_async_event_sub(struct mlx5e_priv *priv,
1081     enum mlx5_dev_event event)
1082 {
1083 	switch (event) {
1084 	case MLX5_DEV_EVENT_PORT_UP:
1085 	case MLX5_DEV_EVENT_PORT_DOWN:
1086 		queue_work(priv->wq, &priv->update_carrier_work);
1087 		break;
1088 
1089 	default:
1090 		break;
1091 	}
1092 }
1093 
1094 static void
1095 mlx5e_async_event(struct mlx5_core_dev *mdev, void *vpriv,
1096     enum mlx5_dev_event event, unsigned long param)
1097 {
1098 	struct mlx5e_priv *priv = vpriv;
1099 
1100 	mtx_lock(&priv->async_events_mtx);
1101 	if (test_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state))
1102 		mlx5e_async_event_sub(priv, event);
1103 	mtx_unlock(&priv->async_events_mtx);
1104 }
1105 
1106 static void
1107 mlx5e_enable_async_events(struct mlx5e_priv *priv)
1108 {
1109 	set_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
1110 }
1111 
1112 static void
1113 mlx5e_disable_async_events(struct mlx5e_priv *priv)
1114 {
1115 	mtx_lock(&priv->async_events_mtx);
1116 	clear_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
1117 	mtx_unlock(&priv->async_events_mtx);
1118 }
1119 
1120 static void mlx5e_calibration_callout(void *arg);
1121 static int mlx5e_calibration_duration = 20;
1122 static int mlx5e_fast_calibration = 1;
1123 static int mlx5e_normal_calibration = 30;
1124 
1125 static SYSCTL_NODE(_hw_mlx5, OID_AUTO, calibr, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
1126     "MLX5 timestamp calibration parameteres");
1127 
1128 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, duration, CTLFLAG_RWTUN,
1129     &mlx5e_calibration_duration, 0,
1130     "Duration of initial calibration");
1131 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, fast, CTLFLAG_RWTUN,
1132     &mlx5e_fast_calibration, 0,
1133     "Recalibration interval during initial calibration");
1134 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, normal, CTLFLAG_RWTUN,
1135     &mlx5e_normal_calibration, 0,
1136     "Recalibration interval during normal operations");
1137 
1138 /*
1139  * Ignites the calibration process.
1140  */
1141 static void
1142 mlx5e_reset_calibration_callout(struct mlx5e_priv *priv)
1143 {
1144 
1145 	if (priv->clbr_done == 0)
1146 		mlx5e_calibration_callout(priv);
1147 	else
1148 		callout_reset_curcpu(&priv->tstmp_clbr, (priv->clbr_done <
1149 		    mlx5e_calibration_duration ? mlx5e_fast_calibration :
1150 		    mlx5e_normal_calibration) * hz, mlx5e_calibration_callout,
1151 		    priv);
1152 }
1153 
1154 static uint64_t
1155 mlx5e_timespec2usec(const struct timespec *ts)
1156 {
1157 
1158 	return ((uint64_t)ts->tv_sec * 1000000000 + ts->tv_nsec);
1159 }
1160 
1161 static uint64_t
1162 mlx5e_hw_clock(struct mlx5e_priv *priv)
1163 {
1164 	struct mlx5_init_seg *iseg;
1165 	uint32_t hw_h, hw_h1, hw_l;
1166 
1167 	iseg = priv->mdev->iseg;
1168 	do {
1169 		hw_h = ioread32be(&iseg->internal_timer_h);
1170 		hw_l = ioread32be(&iseg->internal_timer_l);
1171 		hw_h1 = ioread32be(&iseg->internal_timer_h);
1172 	} while (hw_h1 != hw_h);
1173 	return (((uint64_t)hw_h << 32) | hw_l);
1174 }
1175 
1176 /*
1177  * The calibration callout, it runs either in the context of the
1178  * thread which enables calibration, or in callout.  It takes the
1179  * snapshot of system and adapter clocks, then advances the pointers to
1180  * the calibration point to allow rx path to read the consistent data
1181  * lockless.
1182  */
1183 static void
1184 mlx5e_calibration_callout(void *arg)
1185 {
1186 	struct mlx5e_priv *priv;
1187 	struct mlx5e_clbr_point *next, *curr;
1188 	struct timespec ts;
1189 	int clbr_curr_next;
1190 
1191 	priv = arg;
1192 	curr = &priv->clbr_points[priv->clbr_curr];
1193 	clbr_curr_next = priv->clbr_curr + 1;
1194 	if (clbr_curr_next >= nitems(priv->clbr_points))
1195 		clbr_curr_next = 0;
1196 	next = &priv->clbr_points[clbr_curr_next];
1197 
1198 	next->base_prev = curr->base_curr;
1199 	next->clbr_hw_prev = curr->clbr_hw_curr;
1200 
1201 	next->clbr_hw_curr = mlx5e_hw_clock(priv);
1202 	if (((next->clbr_hw_curr - curr->clbr_hw_curr) >> MLX5E_TSTMP_PREC) ==
1203 	    0) {
1204 		if (priv->clbr_done != 0) {
1205 			mlx5_en_err(priv->ifp,
1206 			    "HW failed tstmp frozen %#jx %#jx, disabling\n",
1207 			     next->clbr_hw_curr, curr->clbr_hw_prev);
1208 			priv->clbr_done = 0;
1209 		}
1210 		atomic_store_rel_int(&curr->clbr_gen, 0);
1211 		return;
1212 	}
1213 
1214 	nanouptime(&ts);
1215 	next->base_curr = mlx5e_timespec2usec(&ts);
1216 
1217 	curr->clbr_gen = 0;
1218 	atomic_thread_fence_rel();
1219 	priv->clbr_curr = clbr_curr_next;
1220 	atomic_store_rel_int(&next->clbr_gen, ++(priv->clbr_gen));
1221 
1222 	if (priv->clbr_done < mlx5e_calibration_duration)
1223 		priv->clbr_done++;
1224 	mlx5e_reset_calibration_callout(priv);
1225 }
1226 
1227 static const char *mlx5e_rq_stats_desc[] = {
1228 	MLX5E_RQ_STATS(MLX5E_STATS_DESC)
1229 };
1230 
1231 static int
1232 mlx5e_create_rq(struct mlx5e_channel *c,
1233     struct mlx5e_rq_param *param,
1234     struct mlx5e_rq *rq)
1235 {
1236 	struct mlx5e_priv *priv = c->priv;
1237 	struct mlx5_core_dev *mdev = priv->mdev;
1238 	char buffer[16];
1239 	void *rqc = param->rqc;
1240 	void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
1241 	int wq_sz;
1242 	int err;
1243 	int i;
1244 	u32 nsegs, wqe_sz;
1245 
1246 	err = mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
1247 	if (err != 0)
1248 		goto done;
1249 
1250 	/* Create DMA descriptor TAG */
1251 	if ((err = -bus_dma_tag_create(
1252 	    bus_get_dma_tag(mdev->pdev->dev.bsddev),
1253 	    1,				/* any alignment */
1254 	    0,				/* no boundary */
1255 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1256 	    BUS_SPACE_MAXADDR,		/* highaddr */
1257 	    NULL, NULL,			/* filter, filterarg */
1258 	    nsegs * MLX5E_MAX_RX_BYTES,	/* maxsize */
1259 	    nsegs,			/* nsegments */
1260 	    nsegs * MLX5E_MAX_RX_BYTES,	/* maxsegsize */
1261 	    0,				/* flags */
1262 	    NULL, NULL,			/* lockfunc, lockfuncarg */
1263 	    &rq->dma_tag)))
1264 		goto done;
1265 
1266 	err = mlx5_wq_ll_create(mdev, &param->wq, rqc_wq, &rq->wq,
1267 	    &rq->wq_ctrl);
1268 	if (err)
1269 		goto err_free_dma_tag;
1270 
1271 	rq->wq.db = &rq->wq.db[MLX5_RCV_DBR];
1272 
1273 	err = mlx5e_get_wqe_sz(priv, &rq->wqe_sz, &rq->nsegs);
1274 	if (err != 0)
1275 		goto err_rq_wq_destroy;
1276 
1277 	wq_sz = mlx5_wq_ll_get_size(&rq->wq);
1278 
1279 	err = -tcp_lro_init_args(&rq->lro, priv->ifp, TCP_LRO_ENTRIES, wq_sz);
1280 	if (err)
1281 		goto err_rq_wq_destroy;
1282 
1283 	rq->mbuf = malloc(wq_sz * sizeof(rq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
1284 	for (i = 0; i != wq_sz; i++) {
1285 		struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
1286 		int j;
1287 
1288 		err = -bus_dmamap_create(rq->dma_tag, 0, &rq->mbuf[i].dma_map);
1289 		if (err != 0) {
1290 			while (i--)
1291 				bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
1292 			goto err_rq_mbuf_free;
1293 		}
1294 
1295 		/* set value for constant fields */
1296 		for (j = 0; j < rq->nsegs; j++)
1297 			wqe->data[j].lkey = cpu_to_be32(priv->mr.key);
1298 	}
1299 
1300 	INIT_WORK(&rq->dim.work, mlx5e_dim_work);
1301 	if (priv->params.rx_cq_moderation_mode < 2) {
1302 		rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
1303 	} else {
1304 		void *cqc = container_of(param,
1305 		    struct mlx5e_channel_param, rq)->rx_cq.cqc;
1306 
1307 		switch (MLX5_GET(cqc, cqc, cq_period_mode)) {
1308 		case MLX5_CQ_PERIOD_MODE_START_FROM_EQE:
1309 			rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
1310 			break;
1311 		case MLX5_CQ_PERIOD_MODE_START_FROM_CQE:
1312 			rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
1313 			break;
1314 		default:
1315 			rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
1316 			break;
1317 		}
1318 	}
1319 
1320 	rq->ifp = priv->ifp;
1321 	rq->channel = c;
1322 	rq->ix = c->ix;
1323 
1324 	snprintf(buffer, sizeof(buffer), "rxstat%d", c->ix);
1325 	mlx5e_create_stats(&rq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
1326 	    buffer, mlx5e_rq_stats_desc, MLX5E_RQ_STATS_NUM,
1327 	    rq->stats.arg);
1328 	return (0);
1329 
1330 err_rq_mbuf_free:
1331 	free(rq->mbuf, M_MLX5EN);
1332 	tcp_lro_free(&rq->lro);
1333 err_rq_wq_destroy:
1334 	mlx5_wq_destroy(&rq->wq_ctrl);
1335 err_free_dma_tag:
1336 	bus_dma_tag_destroy(rq->dma_tag);
1337 done:
1338 	return (err);
1339 }
1340 
1341 static void
1342 mlx5e_destroy_rq(struct mlx5e_rq *rq)
1343 {
1344 	int wq_sz;
1345 	int i;
1346 
1347 	/* destroy all sysctl nodes */
1348 	sysctl_ctx_free(&rq->stats.ctx);
1349 
1350 	/* free leftover LRO packets, if any */
1351 	tcp_lro_free(&rq->lro);
1352 
1353 	wq_sz = mlx5_wq_ll_get_size(&rq->wq);
1354 	for (i = 0; i != wq_sz; i++) {
1355 		if (rq->mbuf[i].mbuf != NULL) {
1356 			bus_dmamap_unload(rq->dma_tag, rq->mbuf[i].dma_map);
1357 			m_freem(rq->mbuf[i].mbuf);
1358 		}
1359 		bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
1360 	}
1361 	free(rq->mbuf, M_MLX5EN);
1362 	mlx5_wq_destroy(&rq->wq_ctrl);
1363 	bus_dma_tag_destroy(rq->dma_tag);
1364 }
1365 
1366 static int
1367 mlx5e_enable_rq(struct mlx5e_rq *rq, struct mlx5e_rq_param *param)
1368 {
1369 	struct mlx5e_channel *c = rq->channel;
1370 	struct mlx5e_priv *priv = c->priv;
1371 	struct mlx5_core_dev *mdev = priv->mdev;
1372 
1373 	void *in;
1374 	void *rqc;
1375 	void *wq;
1376 	int inlen;
1377 	int err;
1378 
1379 	inlen = MLX5_ST_SZ_BYTES(create_rq_in) +
1380 	    sizeof(u64) * rq->wq_ctrl.buf.npages;
1381 	in = mlx5_vzalloc(inlen);
1382 	if (in == NULL)
1383 		return (-ENOMEM);
1384 
1385 	rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
1386 	wq = MLX5_ADDR_OF(rqc, rqc, wq);
1387 
1388 	memcpy(rqc, param->rqc, sizeof(param->rqc));
1389 
1390 	MLX5_SET(rqc, rqc, cqn, c->rq.cq.mcq.cqn);
1391 	MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST);
1392 	MLX5_SET(rqc, rqc, flush_in_error_en, 1);
1393 	if (priv->counter_set_id >= 0)
1394 		MLX5_SET(rqc, rqc, counter_set_id, priv->counter_set_id);
1395 	MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift -
1396 	    PAGE_SHIFT);
1397 	MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma);
1398 
1399 	mlx5_fill_page_array(&rq->wq_ctrl.buf,
1400 	    (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
1401 
1402 	err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn);
1403 
1404 	kvfree(in);
1405 
1406 	return (err);
1407 }
1408 
1409 static int
1410 mlx5e_modify_rq(struct mlx5e_rq *rq, int curr_state, int next_state)
1411 {
1412 	struct mlx5e_channel *c = rq->channel;
1413 	struct mlx5e_priv *priv = c->priv;
1414 	struct mlx5_core_dev *mdev = priv->mdev;
1415 
1416 	void *in;
1417 	void *rqc;
1418 	int inlen;
1419 	int err;
1420 
1421 	inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
1422 	in = mlx5_vzalloc(inlen);
1423 	if (in == NULL)
1424 		return (-ENOMEM);
1425 
1426 	rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
1427 
1428 	MLX5_SET(modify_rq_in, in, rqn, rq->rqn);
1429 	MLX5_SET(modify_rq_in, in, rq_state, curr_state);
1430 	MLX5_SET(rqc, rqc, state, next_state);
1431 
1432 	err = mlx5_core_modify_rq(mdev, in, inlen);
1433 
1434 	kvfree(in);
1435 
1436 	return (err);
1437 }
1438 
1439 static void
1440 mlx5e_disable_rq(struct mlx5e_rq *rq)
1441 {
1442 	struct mlx5e_channel *c = rq->channel;
1443 	struct mlx5e_priv *priv = c->priv;
1444 	struct mlx5_core_dev *mdev = priv->mdev;
1445 
1446 	mlx5_core_destroy_rq(mdev, rq->rqn);
1447 }
1448 
1449 static int
1450 mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq)
1451 {
1452 	struct mlx5e_channel *c = rq->channel;
1453 	struct mlx5e_priv *priv = c->priv;
1454 	struct mlx5_wq_ll *wq = &rq->wq;
1455 	int i;
1456 
1457 	for (i = 0; i < 1000; i++) {
1458 		if (wq->cur_sz >= priv->params.min_rx_wqes)
1459 			return (0);
1460 
1461 		msleep(4);
1462 	}
1463 	return (-ETIMEDOUT);
1464 }
1465 
1466 static int
1467 mlx5e_open_rq(struct mlx5e_channel *c,
1468     struct mlx5e_rq_param *param,
1469     struct mlx5e_rq *rq)
1470 {
1471 	int err;
1472 
1473 	err = mlx5e_create_rq(c, param, rq);
1474 	if (err)
1475 		return (err);
1476 
1477 	err = mlx5e_enable_rq(rq, param);
1478 	if (err)
1479 		goto err_destroy_rq;
1480 
1481 	err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
1482 	if (err)
1483 		goto err_disable_rq;
1484 
1485 	c->rq.enabled = 1;
1486 
1487 	return (0);
1488 
1489 err_disable_rq:
1490 	mlx5e_disable_rq(rq);
1491 err_destroy_rq:
1492 	mlx5e_destroy_rq(rq);
1493 
1494 	return (err);
1495 }
1496 
1497 static void
1498 mlx5e_close_rq(struct mlx5e_rq *rq)
1499 {
1500 	mtx_lock(&rq->mtx);
1501 	rq->enabled = 0;
1502 	callout_stop(&rq->watchdog);
1503 	mtx_unlock(&rq->mtx);
1504 
1505 	mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
1506 }
1507 
1508 static void
1509 mlx5e_close_rq_wait(struct mlx5e_rq *rq)
1510 {
1511 
1512 	mlx5e_disable_rq(rq);
1513 	mlx5e_close_cq(&rq->cq);
1514 	cancel_work_sync(&rq->dim.work);
1515 	mlx5e_destroy_rq(rq);
1516 }
1517 
1518 void
1519 mlx5e_free_sq_db(struct mlx5e_sq *sq)
1520 {
1521 	int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1522 	int x;
1523 
1524 	for (x = 0; x != wq_sz; x++) {
1525 		if (unlikely(sq->mbuf[x].p_refcount != NULL)) {
1526 			atomic_add_int(sq->mbuf[x].p_refcount, -1);
1527 			sq->mbuf[x].p_refcount = NULL;
1528 		}
1529 		if (sq->mbuf[x].mbuf != NULL) {
1530 			bus_dmamap_unload(sq->dma_tag, sq->mbuf[x].dma_map);
1531 			m_freem(sq->mbuf[x].mbuf);
1532 		}
1533 		bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1534 	}
1535 	free(sq->mbuf, M_MLX5EN);
1536 }
1537 
1538 int
1539 mlx5e_alloc_sq_db(struct mlx5e_sq *sq)
1540 {
1541 	int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1542 	int err;
1543 	int x;
1544 
1545 	sq->mbuf = malloc(wq_sz * sizeof(sq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
1546 
1547 	/* Create DMA descriptor MAPs */
1548 	for (x = 0; x != wq_sz; x++) {
1549 		err = -bus_dmamap_create(sq->dma_tag, 0, &sq->mbuf[x].dma_map);
1550 		if (err != 0) {
1551 			while (x--)
1552 				bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1553 			free(sq->mbuf, M_MLX5EN);
1554 			return (err);
1555 		}
1556 	}
1557 	return (0);
1558 }
1559 
1560 static const char *mlx5e_sq_stats_desc[] = {
1561 	MLX5E_SQ_STATS(MLX5E_STATS_DESC)
1562 };
1563 
1564 void
1565 mlx5e_update_sq_inline(struct mlx5e_sq *sq)
1566 {
1567 	sq->max_inline = sq->priv->params.tx_max_inline;
1568 	sq->min_inline_mode = sq->priv->params.tx_min_inline_mode;
1569 
1570 	/*
1571 	 * Check if trust state is DSCP or if inline mode is NONE which
1572 	 * indicates CX-5 or newer hardware.
1573 	 */
1574 	if (sq->priv->params_ethtool.trust_state != MLX5_QPTS_TRUST_PCP ||
1575 	    sq->min_inline_mode == MLX5_INLINE_MODE_NONE) {
1576 		if (MLX5_CAP_ETH(sq->priv->mdev, wqe_vlan_insert))
1577 			sq->min_insert_caps = MLX5E_INSERT_VLAN | MLX5E_INSERT_NON_VLAN;
1578 		else
1579 			sq->min_insert_caps = MLX5E_INSERT_NON_VLAN;
1580 	} else {
1581 		sq->min_insert_caps = 0;
1582 	}
1583 }
1584 
1585 static void
1586 mlx5e_refresh_sq_inline_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
1587 {
1588 	int i;
1589 
1590 	for (i = 0; i != priv->num_tc; i++) {
1591 		mtx_lock(&c->sq[i].lock);
1592 		mlx5e_update_sq_inline(&c->sq[i]);
1593 		mtx_unlock(&c->sq[i].lock);
1594 	}
1595 }
1596 
1597 void
1598 mlx5e_refresh_sq_inline(struct mlx5e_priv *priv)
1599 {
1600 	int i;
1601 
1602 	/* check if channels are closed */
1603 	if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
1604 		return;
1605 
1606 	for (i = 0; i < priv->params.num_channels; i++)
1607 		mlx5e_refresh_sq_inline_sub(priv, &priv->channel[i]);
1608 }
1609 
1610 static int
1611 mlx5e_create_sq(struct mlx5e_channel *c,
1612     int tc,
1613     struct mlx5e_sq_param *param,
1614     struct mlx5e_sq *sq)
1615 {
1616 	struct mlx5e_priv *priv = c->priv;
1617 	struct mlx5_core_dev *mdev = priv->mdev;
1618 	char buffer[16];
1619 	void *sqc = param->sqc;
1620 	void *sqc_wq = MLX5_ADDR_OF(sqc, sqc, wq);
1621 	int err;
1622 
1623 	/* Create DMA descriptor TAG */
1624 	if ((err = -bus_dma_tag_create(
1625 	    bus_get_dma_tag(mdev->pdev->dev.bsddev),
1626 	    1,				/* any alignment */
1627 	    0,				/* no boundary */
1628 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1629 	    BUS_SPACE_MAXADDR,		/* highaddr */
1630 	    NULL, NULL,			/* filter, filterarg */
1631 	    MLX5E_MAX_TX_PAYLOAD_SIZE,	/* maxsize */
1632 	    MLX5E_MAX_TX_MBUF_FRAGS,	/* nsegments */
1633 	    MLX5E_MAX_TX_MBUF_SIZE,	/* maxsegsize */
1634 	    0,				/* flags */
1635 	    NULL, NULL,			/* lockfunc, lockfuncarg */
1636 	    &sq->dma_tag)))
1637 		goto done;
1638 
1639 	err = mlx5_alloc_map_uar(mdev, &sq->uar);
1640 	if (err)
1641 		goto err_free_dma_tag;
1642 
1643 	err = mlx5_wq_cyc_create(mdev, &param->wq, sqc_wq, &sq->wq,
1644 	    &sq->wq_ctrl);
1645 	if (err)
1646 		goto err_unmap_free_uar;
1647 
1648 	sq->wq.db = &sq->wq.db[MLX5_SND_DBR];
1649 	sq->bf_buf_size = (1 << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2;
1650 
1651 	err = mlx5e_alloc_sq_db(sq);
1652 	if (err)
1653 		goto err_sq_wq_destroy;
1654 
1655 	sq->mkey_be = cpu_to_be32(priv->mr.key);
1656 	sq->ifp = priv->ifp;
1657 	sq->priv = priv;
1658 	sq->tc = tc;
1659 
1660 	mlx5e_update_sq_inline(sq);
1661 
1662 	snprintf(buffer, sizeof(buffer), "txstat%dtc%d", c->ix, tc);
1663 	mlx5e_create_stats(&sq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
1664 	    buffer, mlx5e_sq_stats_desc, MLX5E_SQ_STATS_NUM,
1665 	    sq->stats.arg);
1666 
1667 	return (0);
1668 
1669 err_sq_wq_destroy:
1670 	mlx5_wq_destroy(&sq->wq_ctrl);
1671 
1672 err_unmap_free_uar:
1673 	mlx5_unmap_free_uar(mdev, &sq->uar);
1674 
1675 err_free_dma_tag:
1676 	bus_dma_tag_destroy(sq->dma_tag);
1677 done:
1678 	return (err);
1679 }
1680 
1681 static void
1682 mlx5e_destroy_sq(struct mlx5e_sq *sq)
1683 {
1684 	/* destroy all sysctl nodes */
1685 	sysctl_ctx_free(&sq->stats.ctx);
1686 
1687 	mlx5e_free_sq_db(sq);
1688 	mlx5_wq_destroy(&sq->wq_ctrl);
1689 	mlx5_unmap_free_uar(sq->priv->mdev, &sq->uar);
1690 	bus_dma_tag_destroy(sq->dma_tag);
1691 }
1692 
1693 int
1694 mlx5e_enable_sq(struct mlx5e_sq *sq, struct mlx5e_sq_param *param,
1695     int tis_num)
1696 {
1697 	void *in;
1698 	void *sqc;
1699 	void *wq;
1700 	int inlen;
1701 	int err;
1702 
1703 	inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
1704 	    sizeof(u64) * sq->wq_ctrl.buf.npages;
1705 	in = mlx5_vzalloc(inlen);
1706 	if (in == NULL)
1707 		return (-ENOMEM);
1708 
1709 	sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
1710 	wq = MLX5_ADDR_OF(sqc, sqc, wq);
1711 
1712 	memcpy(sqc, param->sqc, sizeof(param->sqc));
1713 
1714 	MLX5_SET(sqc, sqc, tis_num_0, tis_num);
1715 	MLX5_SET(sqc, sqc, cqn, sq->cq.mcq.cqn);
1716 	MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
1717 	MLX5_SET(sqc, sqc, tis_lst_sz, 1);
1718 	MLX5_SET(sqc, sqc, flush_in_error_en, 1);
1719 
1720 	MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
1721 	MLX5_SET(wq, wq, uar_page, sq->uar.index);
1722 	MLX5_SET(wq, wq, log_wq_pg_sz, sq->wq_ctrl.buf.page_shift -
1723 	    PAGE_SHIFT);
1724 	MLX5_SET64(wq, wq, dbr_addr, sq->wq_ctrl.db.dma);
1725 
1726 	mlx5_fill_page_array(&sq->wq_ctrl.buf,
1727 	    (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
1728 
1729 	err = mlx5_core_create_sq(sq->priv->mdev, in, inlen, &sq->sqn);
1730 
1731 	kvfree(in);
1732 
1733 	return (err);
1734 }
1735 
1736 int
1737 mlx5e_modify_sq(struct mlx5e_sq *sq, int curr_state, int next_state)
1738 {
1739 	void *in;
1740 	void *sqc;
1741 	int inlen;
1742 	int err;
1743 
1744 	inlen = MLX5_ST_SZ_BYTES(modify_sq_in);
1745 	in = mlx5_vzalloc(inlen);
1746 	if (in == NULL)
1747 		return (-ENOMEM);
1748 
1749 	sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx);
1750 
1751 	MLX5_SET(modify_sq_in, in, sqn, sq->sqn);
1752 	MLX5_SET(modify_sq_in, in, sq_state, curr_state);
1753 	MLX5_SET(sqc, sqc, state, next_state);
1754 
1755 	err = mlx5_core_modify_sq(sq->priv->mdev, in, inlen);
1756 
1757 	kvfree(in);
1758 
1759 	return (err);
1760 }
1761 
1762 void
1763 mlx5e_disable_sq(struct mlx5e_sq *sq)
1764 {
1765 
1766 	mlx5_core_destroy_sq(sq->priv->mdev, sq->sqn);
1767 }
1768 
1769 static int
1770 mlx5e_open_sq(struct mlx5e_channel *c,
1771     int tc,
1772     struct mlx5e_sq_param *param,
1773     struct mlx5e_sq *sq)
1774 {
1775 	int err;
1776 
1777 	sq->cev_factor = c->priv->params_ethtool.tx_completion_fact;
1778 
1779 	/* ensure the TX completion event factor is not zero */
1780 	if (sq->cev_factor == 0)
1781 		sq->cev_factor = 1;
1782 
1783 	err = mlx5e_create_sq(c, tc, param, sq);
1784 	if (err)
1785 		return (err);
1786 
1787 	err = mlx5e_enable_sq(sq, param, c->priv->tisn[tc]);
1788 	if (err)
1789 		goto err_destroy_sq;
1790 
1791 	err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY);
1792 	if (err)
1793 		goto err_disable_sq;
1794 
1795 	WRITE_ONCE(sq->running, 1);
1796 
1797 	return (0);
1798 
1799 err_disable_sq:
1800 	mlx5e_disable_sq(sq);
1801 err_destroy_sq:
1802 	mlx5e_destroy_sq(sq);
1803 
1804 	return (err);
1805 }
1806 
1807 static void
1808 mlx5e_sq_send_nops_locked(struct mlx5e_sq *sq, int can_sleep)
1809 {
1810 	/* fill up remainder with NOPs */
1811 	while (sq->cev_counter != 0) {
1812 		while (!mlx5e_sq_has_room_for(sq, 1)) {
1813 			if (can_sleep != 0) {
1814 				mtx_unlock(&sq->lock);
1815 				msleep(4);
1816 				mtx_lock(&sq->lock);
1817 			} else {
1818 				goto done;
1819 			}
1820 		}
1821 		/* send a single NOP */
1822 		mlx5e_send_nop(sq, 1);
1823 		atomic_thread_fence_rel();
1824 	}
1825 done:
1826 	/* Check if we need to write the doorbell */
1827 	if (likely(sq->doorbell.d64 != 0)) {
1828 		mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
1829 		sq->doorbell.d64 = 0;
1830 	}
1831 }
1832 
1833 void
1834 mlx5e_sq_cev_timeout(void *arg)
1835 {
1836 	struct mlx5e_sq *sq = arg;
1837 
1838 	mtx_assert(&sq->lock, MA_OWNED);
1839 
1840 	/* check next state */
1841 	switch (sq->cev_next_state) {
1842 	case MLX5E_CEV_STATE_SEND_NOPS:
1843 		/* fill TX ring with NOPs, if any */
1844 		mlx5e_sq_send_nops_locked(sq, 0);
1845 
1846 		/* check if completed */
1847 		if (sq->cev_counter == 0) {
1848 			sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
1849 			return;
1850 		}
1851 		break;
1852 	default:
1853 		/* send NOPs on next timeout */
1854 		sq->cev_next_state = MLX5E_CEV_STATE_SEND_NOPS;
1855 		break;
1856 	}
1857 
1858 	/* restart timer */
1859 	callout_reset_curcpu(&sq->cev_callout, hz, mlx5e_sq_cev_timeout, sq);
1860 }
1861 
1862 void
1863 mlx5e_drain_sq(struct mlx5e_sq *sq)
1864 {
1865 	int error;
1866 	struct mlx5_core_dev *mdev= sq->priv->mdev;
1867 
1868 	/*
1869 	 * Check if already stopped.
1870 	 *
1871 	 * NOTE: Serialization of this function is managed by the
1872 	 * caller ensuring the priv's state lock is locked or in case
1873 	 * of rate limit support, a single thread manages drain and
1874 	 * resume of SQs. The "running" variable can therefore safely
1875 	 * be read without any locks.
1876 	 */
1877 	if (READ_ONCE(sq->running) == 0)
1878 		return;
1879 
1880 	/* don't put more packets into the SQ */
1881 	WRITE_ONCE(sq->running, 0);
1882 
1883 	/* serialize access to DMA rings */
1884 	mtx_lock(&sq->lock);
1885 
1886 	/* teardown event factor timer, if any */
1887 	sq->cev_next_state = MLX5E_CEV_STATE_HOLD_NOPS;
1888 	callout_stop(&sq->cev_callout);
1889 
1890 	/* send dummy NOPs in order to flush the transmit ring */
1891 	mlx5e_sq_send_nops_locked(sq, 1);
1892 	mtx_unlock(&sq->lock);
1893 
1894 	/* wait till SQ is empty or link is down */
1895 	mtx_lock(&sq->lock);
1896 	while (sq->cc != sq->pc &&
1897 	    (sq->priv->media_status_last & IFM_ACTIVE) != 0 &&
1898 	    mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1899 		mtx_unlock(&sq->lock);
1900 		msleep(1);
1901 		sq->cq.mcq.comp(&sq->cq.mcq);
1902 		mtx_lock(&sq->lock);
1903 	}
1904 	mtx_unlock(&sq->lock);
1905 
1906 	/* error out remaining requests */
1907 	error = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RDY, MLX5_SQC_STATE_ERR);
1908 	if (error != 0) {
1909 		mlx5_en_err(sq->ifp,
1910 		    "mlx5e_modify_sq() from RDY to ERR failed: %d\n", error);
1911 	}
1912 
1913 	/* wait till SQ is empty */
1914 	mtx_lock(&sq->lock);
1915 	while (sq->cc != sq->pc &&
1916 	       mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1917 		mtx_unlock(&sq->lock);
1918 		msleep(1);
1919 		sq->cq.mcq.comp(&sq->cq.mcq);
1920 		mtx_lock(&sq->lock);
1921 	}
1922 	mtx_unlock(&sq->lock);
1923 }
1924 
1925 static void
1926 mlx5e_close_sq_wait(struct mlx5e_sq *sq)
1927 {
1928 
1929 	mlx5e_drain_sq(sq);
1930 	mlx5e_disable_sq(sq);
1931 	mlx5e_destroy_sq(sq);
1932 }
1933 
1934 static int
1935 mlx5e_create_cq(struct mlx5e_priv *priv,
1936     struct mlx5e_cq_param *param,
1937     struct mlx5e_cq *cq,
1938     mlx5e_cq_comp_t *comp,
1939     int eq_ix)
1940 {
1941 	struct mlx5_core_dev *mdev = priv->mdev;
1942 	struct mlx5_core_cq *mcq = &cq->mcq;
1943 	int eqn_not_used;
1944 	int irqn;
1945 	int err;
1946 	u32 i;
1947 
1948 	param->wq.buf_numa_node = 0;
1949 	param->wq.db_numa_node = 0;
1950 
1951 	err = mlx5_vector2eqn(mdev, eq_ix, &eqn_not_used, &irqn);
1952 	if (err)
1953 		return (err);
1954 
1955 	err = mlx5_cqwq_create(mdev, &param->wq, param->cqc, &cq->wq,
1956 	    &cq->wq_ctrl);
1957 	if (err)
1958 		return (err);
1959 
1960 	mcq->cqe_sz = 64;
1961 	mcq->set_ci_db = cq->wq_ctrl.db.db;
1962 	mcq->arm_db = cq->wq_ctrl.db.db + 1;
1963 	*mcq->set_ci_db = 0;
1964 	*mcq->arm_db = 0;
1965 	mcq->vector = eq_ix;
1966 	mcq->comp = comp;
1967 	mcq->event = mlx5e_cq_error_event;
1968 	mcq->irqn = irqn;
1969 	mcq->uar = &priv->cq_uar;
1970 
1971 	for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) {
1972 		struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i);
1973 
1974 		cqe->op_own = 0xf1;
1975 	}
1976 
1977 	cq->priv = priv;
1978 
1979 	return (0);
1980 }
1981 
1982 static void
1983 mlx5e_destroy_cq(struct mlx5e_cq *cq)
1984 {
1985 	mlx5_wq_destroy(&cq->wq_ctrl);
1986 }
1987 
1988 static int
1989 mlx5e_enable_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param, int eq_ix)
1990 {
1991 	struct mlx5_core_cq *mcq = &cq->mcq;
1992 	void *in;
1993 	void *cqc;
1994 	int inlen;
1995 	int irqn_not_used;
1996 	int eqn;
1997 	int err;
1998 
1999 	inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
2000 	    sizeof(u64) * cq->wq_ctrl.buf.npages;
2001 	in = mlx5_vzalloc(inlen);
2002 	if (in == NULL)
2003 		return (-ENOMEM);
2004 
2005 	cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
2006 
2007 	memcpy(cqc, param->cqc, sizeof(param->cqc));
2008 
2009 	mlx5_fill_page_array(&cq->wq_ctrl.buf,
2010 	    (__be64 *) MLX5_ADDR_OF(create_cq_in, in, pas));
2011 
2012 	mlx5_vector2eqn(cq->priv->mdev, eq_ix, &eqn, &irqn_not_used);
2013 
2014 	MLX5_SET(cqc, cqc, c_eqn, eqn);
2015 	MLX5_SET(cqc, cqc, uar_page, mcq->uar->index);
2016 	MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.buf.page_shift -
2017 	    PAGE_SHIFT);
2018 	MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
2019 
2020 	err = mlx5_core_create_cq(cq->priv->mdev, mcq, in, inlen);
2021 
2022 	kvfree(in);
2023 
2024 	if (err)
2025 		return (err);
2026 
2027 	mlx5e_cq_arm(cq, MLX5_GET_DOORBELL_LOCK(&cq->priv->doorbell_lock));
2028 
2029 	return (0);
2030 }
2031 
2032 static void
2033 mlx5e_disable_cq(struct mlx5e_cq *cq)
2034 {
2035 
2036 	mlx5_core_destroy_cq(cq->priv->mdev, &cq->mcq);
2037 }
2038 
2039 int
2040 mlx5e_open_cq(struct mlx5e_priv *priv,
2041     struct mlx5e_cq_param *param,
2042     struct mlx5e_cq *cq,
2043     mlx5e_cq_comp_t *comp,
2044     int eq_ix)
2045 {
2046 	int err;
2047 
2048 	err = mlx5e_create_cq(priv, param, cq, comp, eq_ix);
2049 	if (err)
2050 		return (err);
2051 
2052 	err = mlx5e_enable_cq(cq, param, eq_ix);
2053 	if (err)
2054 		goto err_destroy_cq;
2055 
2056 	return (0);
2057 
2058 err_destroy_cq:
2059 	mlx5e_destroy_cq(cq);
2060 
2061 	return (err);
2062 }
2063 
2064 void
2065 mlx5e_close_cq(struct mlx5e_cq *cq)
2066 {
2067 	mlx5e_disable_cq(cq);
2068 	mlx5e_destroy_cq(cq);
2069 }
2070 
2071 static int
2072 mlx5e_open_tx_cqs(struct mlx5e_channel *c,
2073     struct mlx5e_channel_param *cparam)
2074 {
2075 	int err;
2076 	int tc;
2077 
2078 	for (tc = 0; tc < c->priv->num_tc; tc++) {
2079 		/* open completion queue */
2080 		err = mlx5e_open_cq(c->priv, &cparam->tx_cq, &c->sq[tc].cq,
2081 		    &mlx5e_tx_cq_comp, c->ix);
2082 		if (err)
2083 			goto err_close_tx_cqs;
2084 	}
2085 	return (0);
2086 
2087 err_close_tx_cqs:
2088 	for (tc--; tc >= 0; tc--)
2089 		mlx5e_close_cq(&c->sq[tc].cq);
2090 
2091 	return (err);
2092 }
2093 
2094 static void
2095 mlx5e_close_tx_cqs(struct mlx5e_channel *c)
2096 {
2097 	int tc;
2098 
2099 	for (tc = 0; tc < c->priv->num_tc; tc++)
2100 		mlx5e_close_cq(&c->sq[tc].cq);
2101 }
2102 
2103 static int
2104 mlx5e_open_sqs(struct mlx5e_channel *c,
2105     struct mlx5e_channel_param *cparam)
2106 {
2107 	int err;
2108 	int tc;
2109 
2110 	for (tc = 0; tc < c->priv->num_tc; tc++) {
2111 		err = mlx5e_open_sq(c, tc, &cparam->sq, &c->sq[tc]);
2112 		if (err)
2113 			goto err_close_sqs;
2114 	}
2115 
2116 	return (0);
2117 
2118 err_close_sqs:
2119 	for (tc--; tc >= 0; tc--)
2120 		mlx5e_close_sq_wait(&c->sq[tc]);
2121 
2122 	return (err);
2123 }
2124 
2125 static void
2126 mlx5e_close_sqs_wait(struct mlx5e_channel *c)
2127 {
2128 	int tc;
2129 
2130 	for (tc = 0; tc < c->priv->num_tc; tc++)
2131 		mlx5e_close_sq_wait(&c->sq[tc]);
2132 }
2133 
2134 static void
2135 mlx5e_chan_static_init(struct mlx5e_priv *priv, struct mlx5e_channel *c, int ix)
2136 {
2137 	int tc;
2138 
2139 	/* setup priv and channel number */
2140 	c->priv = priv;
2141 	c->ix = ix;
2142 
2143 	/* setup send tag */
2144 	m_snd_tag_init(&c->tag, c->priv->ifp, IF_SND_TAG_TYPE_UNLIMITED);
2145 
2146 	init_completion(&c->completion);
2147 
2148 	mtx_init(&c->rq.mtx, "mlx5rx", MTX_NETWORK_LOCK, MTX_DEF);
2149 
2150 	callout_init_mtx(&c->rq.watchdog, &c->rq.mtx, 0);
2151 
2152 	for (tc = 0; tc != MLX5E_MAX_TX_NUM_TC; tc++) {
2153 		struct mlx5e_sq *sq = c->sq + tc;
2154 
2155 		mtx_init(&sq->lock, "mlx5tx",
2156 		    MTX_NETWORK_LOCK " TX", MTX_DEF);
2157 		mtx_init(&sq->comp_lock, "mlx5comp",
2158 		    MTX_NETWORK_LOCK " TX", MTX_DEF);
2159 
2160 		callout_init_mtx(&sq->cev_callout, &sq->lock, 0);
2161 	}
2162 }
2163 
2164 static void
2165 mlx5e_chan_wait_for_completion(struct mlx5e_channel *c)
2166 {
2167 
2168 	m_snd_tag_rele(&c->tag);
2169 	wait_for_completion(&c->completion);
2170 }
2171 
2172 static void
2173 mlx5e_priv_wait_for_completion(struct mlx5e_priv *priv, const uint32_t channels)
2174 {
2175 	uint32_t x;
2176 
2177 	for (x = 0; x != channels; x++)
2178 		mlx5e_chan_wait_for_completion(&priv->channel[x]);
2179 }
2180 
2181 static void
2182 mlx5e_chan_static_destroy(struct mlx5e_channel *c)
2183 {
2184 	int tc;
2185 
2186 	callout_drain(&c->rq.watchdog);
2187 
2188 	mtx_destroy(&c->rq.mtx);
2189 
2190 	for (tc = 0; tc != MLX5E_MAX_TX_NUM_TC; tc++) {
2191 		callout_drain(&c->sq[tc].cev_callout);
2192 		mtx_destroy(&c->sq[tc].lock);
2193 		mtx_destroy(&c->sq[tc].comp_lock);
2194 	}
2195 }
2196 
2197 static int
2198 mlx5e_open_channel(struct mlx5e_priv *priv,
2199     struct mlx5e_channel_param *cparam,
2200     struct mlx5e_channel *c)
2201 {
2202 	struct epoch_tracker et;
2203 	int i, err;
2204 
2205 	/* zero non-persistant data */
2206 	MLX5E_ZERO(&c->rq, mlx5e_rq_zero_start);
2207 	for (i = 0; i != priv->num_tc; i++)
2208 		MLX5E_ZERO(&c->sq[i], mlx5e_sq_zero_start);
2209 
2210 	/* open transmit completion queue */
2211 	err = mlx5e_open_tx_cqs(c, cparam);
2212 	if (err)
2213 		goto err_free;
2214 
2215 	/* open receive completion queue */
2216 	err = mlx5e_open_cq(c->priv, &cparam->rx_cq, &c->rq.cq,
2217 	    &mlx5e_rx_cq_comp, c->ix);
2218 	if (err)
2219 		goto err_close_tx_cqs;
2220 
2221 	err = mlx5e_open_sqs(c, cparam);
2222 	if (err)
2223 		goto err_close_rx_cq;
2224 
2225 	err = mlx5e_open_rq(c, &cparam->rq, &c->rq);
2226 	if (err)
2227 		goto err_close_sqs;
2228 
2229 	/* poll receive queue initially */
2230 	NET_EPOCH_ENTER(et);
2231 	c->rq.cq.mcq.comp(&c->rq.cq.mcq);
2232 	NET_EPOCH_EXIT(et);
2233 
2234 	return (0);
2235 
2236 err_close_sqs:
2237 	mlx5e_close_sqs_wait(c);
2238 
2239 err_close_rx_cq:
2240 	mlx5e_close_cq(&c->rq.cq);
2241 
2242 err_close_tx_cqs:
2243 	mlx5e_close_tx_cqs(c);
2244 
2245 err_free:
2246 	return (err);
2247 }
2248 
2249 static void
2250 mlx5e_close_channel(struct mlx5e_channel *c)
2251 {
2252 	mlx5e_close_rq(&c->rq);
2253 }
2254 
2255 static void
2256 mlx5e_close_channel_wait(struct mlx5e_channel *c)
2257 {
2258 	mlx5e_close_rq_wait(&c->rq);
2259 	mlx5e_close_sqs_wait(c);
2260 	mlx5e_close_tx_cqs(c);
2261 }
2262 
2263 static int
2264 mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs)
2265 {
2266 	u32 r, n;
2267 
2268 	r = priv->params.hw_lro_en ? priv->params.lro_wqe_sz :
2269 	    MLX5E_SW2MB_MTU(priv->ifp->if_mtu);
2270 	if (r > MJUM16BYTES)
2271 		return (-ENOMEM);
2272 
2273 	if (r > MJUM9BYTES)
2274 		r = MJUM16BYTES;
2275 	else if (r > MJUMPAGESIZE)
2276 		r = MJUM9BYTES;
2277 	else if (r > MCLBYTES)
2278 		r = MJUMPAGESIZE;
2279 	else
2280 		r = MCLBYTES;
2281 
2282 	/*
2283 	 * n + 1 must be a power of two, because stride size must be.
2284 	 * Stride size is 16 * (n + 1), as the first segment is
2285 	 * control.
2286 	 */
2287 	for (n = howmany(r, MLX5E_MAX_RX_BYTES); !powerof2(n + 1); n++)
2288 		;
2289 
2290 	if (n > MLX5E_MAX_BUSDMA_RX_SEGS)
2291 		return (-ENOMEM);
2292 
2293 	*wqe_sz = r;
2294 	*nsegs = n;
2295 	return (0);
2296 }
2297 
2298 static void
2299 mlx5e_build_rq_param(struct mlx5e_priv *priv,
2300     struct mlx5e_rq_param *param)
2301 {
2302 	void *rqc = param->rqc;
2303 	void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
2304 	u32 wqe_sz, nsegs;
2305 
2306 	mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
2307 	MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
2308 	MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN);
2309 	MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe) +
2310 	    nsegs * sizeof(struct mlx5_wqe_data_seg)));
2311 	MLX5_SET(wq, wq, log_wq_sz, priv->params.log_rq_size);
2312 	MLX5_SET(wq, wq, pd, priv->pdn);
2313 
2314 	param->wq.buf_numa_node = 0;
2315 	param->wq.db_numa_node = 0;
2316 	param->wq.linear = 1;
2317 }
2318 
2319 static void
2320 mlx5e_build_sq_param(struct mlx5e_priv *priv,
2321     struct mlx5e_sq_param *param)
2322 {
2323 	void *sqc = param->sqc;
2324 	void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
2325 
2326 	MLX5_SET(wq, wq, log_wq_sz, priv->params.log_sq_size);
2327 	MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
2328 	MLX5_SET(wq, wq, pd, priv->pdn);
2329 
2330 	param->wq.buf_numa_node = 0;
2331 	param->wq.db_numa_node = 0;
2332 	param->wq.linear = 1;
2333 }
2334 
2335 static void
2336 mlx5e_build_common_cq_param(struct mlx5e_priv *priv,
2337     struct mlx5e_cq_param *param)
2338 {
2339 	void *cqc = param->cqc;
2340 
2341 	MLX5_SET(cqc, cqc, uar_page, priv->cq_uar.index);
2342 }
2343 
2344 static void
2345 mlx5e_get_default_profile(struct mlx5e_priv *priv, int mode, struct net_dim_cq_moder *ptr)
2346 {
2347 
2348 	*ptr = net_dim_get_profile(mode, MLX5E_DIM_DEFAULT_PROFILE);
2349 
2350 	/* apply LRO restrictions */
2351 	if (priv->params.hw_lro_en &&
2352 	    ptr->pkts > MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO) {
2353 		ptr->pkts = MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO;
2354 	}
2355 }
2356 
2357 static void
2358 mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
2359     struct mlx5e_cq_param *param)
2360 {
2361 	struct net_dim_cq_moder curr;
2362 	void *cqc = param->cqc;
2363 
2364 	/*
2365 	 * We use MLX5_CQE_FORMAT_HASH because the RX hash mini CQE
2366 	 * format is more beneficial for FreeBSD use case.
2367 	 *
2368 	 * Adding support for MLX5_CQE_FORMAT_CSUM will require changes
2369 	 * in mlx5e_decompress_cqe.
2370 	 */
2371 	if (priv->params.cqe_zipping_en) {
2372 		MLX5_SET(cqc, cqc, mini_cqe_res_format, MLX5_CQE_FORMAT_HASH);
2373 		MLX5_SET(cqc, cqc, cqe_compression_en, 1);
2374 	}
2375 
2376 	MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_rq_size);
2377 
2378 	switch (priv->params.rx_cq_moderation_mode) {
2379 	case 0:
2380 		MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
2381 		MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
2382 		MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2383 		break;
2384 	case 1:
2385 		MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
2386 		MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
2387 		if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
2388 			MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
2389 		else
2390 			MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2391 		break;
2392 	case 2:
2393 		mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE, &curr);
2394 		MLX5_SET(cqc, cqc, cq_period, curr.usec);
2395 		MLX5_SET(cqc, cqc, cq_max_count, curr.pkts);
2396 		MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2397 		break;
2398 	case 3:
2399 		mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE, &curr);
2400 		MLX5_SET(cqc, cqc, cq_period, curr.usec);
2401 		MLX5_SET(cqc, cqc, cq_max_count, curr.pkts);
2402 		if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
2403 			MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
2404 		else
2405 			MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2406 		break;
2407 	default:
2408 		break;
2409 	}
2410 
2411 	mlx5e_dim_build_cq_param(priv, param);
2412 
2413 	mlx5e_build_common_cq_param(priv, param);
2414 }
2415 
2416 static void
2417 mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
2418     struct mlx5e_cq_param *param)
2419 {
2420 	void *cqc = param->cqc;
2421 
2422 	MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_sq_size);
2423 	MLX5_SET(cqc, cqc, cq_period, priv->params.tx_cq_moderation_usec);
2424 	MLX5_SET(cqc, cqc, cq_max_count, priv->params.tx_cq_moderation_pkts);
2425 
2426 	switch (priv->params.tx_cq_moderation_mode) {
2427 	case 0:
2428 		MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2429 		break;
2430 	default:
2431 		if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
2432 			MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
2433 		else
2434 			MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2435 		break;
2436 	}
2437 
2438 	mlx5e_build_common_cq_param(priv, param);
2439 }
2440 
2441 static void
2442 mlx5e_build_channel_param(struct mlx5e_priv *priv,
2443     struct mlx5e_channel_param *cparam)
2444 {
2445 	memset(cparam, 0, sizeof(*cparam));
2446 
2447 	mlx5e_build_rq_param(priv, &cparam->rq);
2448 	mlx5e_build_sq_param(priv, &cparam->sq);
2449 	mlx5e_build_rx_cq_param(priv, &cparam->rx_cq);
2450 	mlx5e_build_tx_cq_param(priv, &cparam->tx_cq);
2451 }
2452 
2453 static int
2454 mlx5e_open_channels(struct mlx5e_priv *priv)
2455 {
2456 	struct mlx5e_channel_param *cparam;
2457 	int err;
2458 	int i;
2459 	int j;
2460 
2461 	cparam = malloc(sizeof(*cparam), M_MLX5EN, M_WAITOK);
2462 
2463 	mlx5e_build_channel_param(priv, cparam);
2464 	for (i = 0; i < priv->params.num_channels; i++) {
2465 		err = mlx5e_open_channel(priv, cparam, &priv->channel[i]);
2466 		if (err)
2467 			goto err_close_channels;
2468 	}
2469 
2470 	for (j = 0; j < priv->params.num_channels; j++) {
2471 		err = mlx5e_wait_for_min_rx_wqes(&priv->channel[j].rq);
2472 		if (err)
2473 			goto err_close_channels;
2474 	}
2475 	free(cparam, M_MLX5EN);
2476 	return (0);
2477 
2478 err_close_channels:
2479 	while (i--) {
2480 		mlx5e_close_channel(&priv->channel[i]);
2481 		mlx5e_close_channel_wait(&priv->channel[i]);
2482 	}
2483 	free(cparam, M_MLX5EN);
2484 	return (err);
2485 }
2486 
2487 static void
2488 mlx5e_close_channels(struct mlx5e_priv *priv)
2489 {
2490 	int i;
2491 
2492 	for (i = 0; i < priv->params.num_channels; i++)
2493 		mlx5e_close_channel(&priv->channel[i]);
2494 	for (i = 0; i < priv->params.num_channels; i++)
2495 		mlx5e_close_channel_wait(&priv->channel[i]);
2496 }
2497 
2498 static int
2499 mlx5e_refresh_sq_params(struct mlx5e_priv *priv, struct mlx5e_sq *sq)
2500 {
2501 
2502 	if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
2503 		uint8_t cq_mode;
2504 
2505 		switch (priv->params.tx_cq_moderation_mode) {
2506 		case 0:
2507 		case 2:
2508 			cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2509 			break;
2510 		default:
2511 			cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2512 			break;
2513 		}
2514 
2515 		return (mlx5_core_modify_cq_moderation_mode(priv->mdev, &sq->cq.mcq,
2516 		    priv->params.tx_cq_moderation_usec,
2517 		    priv->params.tx_cq_moderation_pkts,
2518 		    cq_mode));
2519 	}
2520 
2521 	return (mlx5_core_modify_cq_moderation(priv->mdev, &sq->cq.mcq,
2522 	    priv->params.tx_cq_moderation_usec,
2523 	    priv->params.tx_cq_moderation_pkts));
2524 }
2525 
2526 static int
2527 mlx5e_refresh_rq_params(struct mlx5e_priv *priv, struct mlx5e_rq *rq)
2528 {
2529 
2530 	if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
2531 		uint8_t cq_mode;
2532 		uint8_t dim_mode;
2533 		int retval;
2534 
2535 		switch (priv->params.rx_cq_moderation_mode) {
2536 		case 0:
2537 		case 2:
2538 			cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2539 			dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
2540 			break;
2541 		default:
2542 			cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2543 			dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
2544 			break;
2545 		}
2546 
2547 		/* tear down dynamic interrupt moderation */
2548 		mtx_lock(&rq->mtx);
2549 		rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
2550 		mtx_unlock(&rq->mtx);
2551 
2552 		/* wait for dynamic interrupt moderation work task, if any */
2553 		cancel_work_sync(&rq->dim.work);
2554 
2555 		if (priv->params.rx_cq_moderation_mode >= 2) {
2556 			struct net_dim_cq_moder curr;
2557 
2558 			mlx5e_get_default_profile(priv, dim_mode, &curr);
2559 
2560 			retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2561 			    curr.usec, curr.pkts, cq_mode);
2562 
2563 			/* set dynamic interrupt moderation mode and zero defaults */
2564 			mtx_lock(&rq->mtx);
2565 			rq->dim.mode = dim_mode;
2566 			rq->dim.state = 0;
2567 			rq->dim.profile_ix = MLX5E_DIM_DEFAULT_PROFILE;
2568 			mtx_unlock(&rq->mtx);
2569 		} else {
2570 			retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2571 			    priv->params.rx_cq_moderation_usec,
2572 			    priv->params.rx_cq_moderation_pkts,
2573 			    cq_mode);
2574 		}
2575 		return (retval);
2576 	}
2577 
2578 	return (mlx5_core_modify_cq_moderation(priv->mdev, &rq->cq.mcq,
2579 	    priv->params.rx_cq_moderation_usec,
2580 	    priv->params.rx_cq_moderation_pkts));
2581 }
2582 
2583 static int
2584 mlx5e_refresh_channel_params_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
2585 {
2586 	int err;
2587 	int i;
2588 
2589 	err = mlx5e_refresh_rq_params(priv, &c->rq);
2590 	if (err)
2591 		goto done;
2592 
2593 	for (i = 0; i != priv->num_tc; i++) {
2594 		err = mlx5e_refresh_sq_params(priv, &c->sq[i]);
2595 		if (err)
2596 			goto done;
2597 	}
2598 done:
2599 	return (err);
2600 }
2601 
2602 int
2603 mlx5e_refresh_channel_params(struct mlx5e_priv *priv)
2604 {
2605 	int i;
2606 
2607 	/* check if channels are closed */
2608 	if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2609 		return (EINVAL);
2610 
2611 	for (i = 0; i < priv->params.num_channels; i++) {
2612 		int err;
2613 
2614 		err = mlx5e_refresh_channel_params_sub(priv, &priv->channel[i]);
2615 		if (err)
2616 			return (err);
2617 	}
2618 	return (0);
2619 }
2620 
2621 static int
2622 mlx5e_open_tis(struct mlx5e_priv *priv, int tc)
2623 {
2624 	struct mlx5_core_dev *mdev = priv->mdev;
2625 	u32 in[MLX5_ST_SZ_DW(create_tis_in)];
2626 	void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
2627 
2628 	memset(in, 0, sizeof(in));
2629 
2630 	MLX5_SET(tisc, tisc, prio, tc);
2631 	MLX5_SET(tisc, tisc, transport_domain, priv->tdn);
2632 
2633 	return (mlx5_core_create_tis(mdev, in, sizeof(in), &priv->tisn[tc]));
2634 }
2635 
2636 static void
2637 mlx5e_close_tis(struct mlx5e_priv *priv, int tc)
2638 {
2639 	mlx5_core_destroy_tis(priv->mdev, priv->tisn[tc]);
2640 }
2641 
2642 static int
2643 mlx5e_open_tises(struct mlx5e_priv *priv)
2644 {
2645 	int num_tc = priv->num_tc;
2646 	int err;
2647 	int tc;
2648 
2649 	for (tc = 0; tc < num_tc; tc++) {
2650 		err = mlx5e_open_tis(priv, tc);
2651 		if (err)
2652 			goto err_close_tises;
2653 	}
2654 
2655 	return (0);
2656 
2657 err_close_tises:
2658 	for (tc--; tc >= 0; tc--)
2659 		mlx5e_close_tis(priv, tc);
2660 
2661 	return (err);
2662 }
2663 
2664 static void
2665 mlx5e_close_tises(struct mlx5e_priv *priv)
2666 {
2667 	int num_tc = priv->num_tc;
2668 	int tc;
2669 
2670 	for (tc = 0; tc < num_tc; tc++)
2671 		mlx5e_close_tis(priv, tc);
2672 }
2673 
2674 static int
2675 mlx5e_open_rqt(struct mlx5e_priv *priv)
2676 {
2677 	struct mlx5_core_dev *mdev = priv->mdev;
2678 	u32 *in;
2679 	u32 out[MLX5_ST_SZ_DW(create_rqt_out)] = {0};
2680 	void *rqtc;
2681 	int inlen;
2682 	int err;
2683 	int sz;
2684 	int i;
2685 
2686 	sz = 1 << priv->params.rx_hash_log_tbl_sz;
2687 
2688 	inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
2689 	in = mlx5_vzalloc(inlen);
2690 	if (in == NULL)
2691 		return (-ENOMEM);
2692 	rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
2693 
2694 	MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
2695 	MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
2696 
2697 	for (i = 0; i < sz; i++) {
2698 		int ix = i;
2699 #ifdef RSS
2700 		ix = rss_get_indirection_to_bucket(ix);
2701 #endif
2702 		/* ensure we don't overflow */
2703 		ix %= priv->params.num_channels;
2704 
2705 		/* apply receive side scaling stride, if any */
2706 		ix -= ix % (int)priv->params.channels_rsss;
2707 
2708 		MLX5_SET(rqtc, rqtc, rq_num[i], priv->channel[ix].rq.rqn);
2709 	}
2710 
2711 	MLX5_SET(create_rqt_in, in, opcode, MLX5_CMD_OP_CREATE_RQT);
2712 
2713 	err = mlx5_cmd_exec(mdev, in, inlen, out, sizeof(out));
2714 	if (!err)
2715 		priv->rqtn = MLX5_GET(create_rqt_out, out, rqtn);
2716 
2717 	kvfree(in);
2718 
2719 	return (err);
2720 }
2721 
2722 static void
2723 mlx5e_close_rqt(struct mlx5e_priv *priv)
2724 {
2725 	u32 in[MLX5_ST_SZ_DW(destroy_rqt_in)] = {0};
2726 	u32 out[MLX5_ST_SZ_DW(destroy_rqt_out)] = {0};
2727 
2728 	MLX5_SET(destroy_rqt_in, in, opcode, MLX5_CMD_OP_DESTROY_RQT);
2729 	MLX5_SET(destroy_rqt_in, in, rqtn, priv->rqtn);
2730 
2731 	mlx5_cmd_exec(priv->mdev, in, sizeof(in), out, sizeof(out));
2732 }
2733 
2734 static void
2735 mlx5e_build_tir_ctx(struct mlx5e_priv *priv, u32 * tirc, int tt)
2736 {
2737 	void *hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
2738 	__be32 *hkey;
2739 
2740 	MLX5_SET(tirc, tirc, transport_domain, priv->tdn);
2741 
2742 #define	ROUGH_MAX_L2_L3_HDR_SZ 256
2743 
2744 #define	MLX5_HASH_IP     (MLX5_HASH_FIELD_SEL_SRC_IP   |\
2745 			  MLX5_HASH_FIELD_SEL_DST_IP)
2746 
2747 #define	MLX5_HASH_ALL    (MLX5_HASH_FIELD_SEL_SRC_IP   |\
2748 			  MLX5_HASH_FIELD_SEL_DST_IP   |\
2749 			  MLX5_HASH_FIELD_SEL_L4_SPORT |\
2750 			  MLX5_HASH_FIELD_SEL_L4_DPORT)
2751 
2752 #define	MLX5_HASH_IP_IPSEC_SPI	(MLX5_HASH_FIELD_SEL_SRC_IP   |\
2753 				 MLX5_HASH_FIELD_SEL_DST_IP   |\
2754 				 MLX5_HASH_FIELD_SEL_IPSEC_SPI)
2755 
2756 	if (priv->params.hw_lro_en) {
2757 		MLX5_SET(tirc, tirc, lro_enable_mask,
2758 		    MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO |
2759 		    MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO);
2760 		MLX5_SET(tirc, tirc, lro_max_msg_sz,
2761 		    (priv->params.lro_wqe_sz -
2762 		    ROUGH_MAX_L2_L3_HDR_SZ) >> 8);
2763 		/* TODO: add the option to choose timer value dynamically */
2764 		MLX5_SET(tirc, tirc, lro_timeout_period_usecs,
2765 		    MLX5_CAP_ETH(priv->mdev,
2766 		    lro_timer_supported_periods[2]));
2767 	}
2768 
2769 	/* setup parameters for hashing TIR type, if any */
2770 	switch (tt) {
2771 	case MLX5E_TT_ANY:
2772 		MLX5_SET(tirc, tirc, disp_type,
2773 		    MLX5_TIRC_DISP_TYPE_DIRECT);
2774 		MLX5_SET(tirc, tirc, inline_rqn,
2775 		    priv->channel[0].rq.rqn);
2776 		break;
2777 	default:
2778 		MLX5_SET(tirc, tirc, disp_type,
2779 		    MLX5_TIRC_DISP_TYPE_INDIRECT);
2780 		MLX5_SET(tirc, tirc, indirect_table,
2781 		    priv->rqtn);
2782 		MLX5_SET(tirc, tirc, rx_hash_fn,
2783 		    MLX5_TIRC_RX_HASH_FN_HASH_TOEPLITZ);
2784 		hkey = (__be32 *) MLX5_ADDR_OF(tirc, tirc, rx_hash_toeplitz_key);
2785 #ifdef RSS
2786 		/*
2787 		 * The FreeBSD RSS implementation does currently not
2788 		 * support symmetric Toeplitz hashes:
2789 		 */
2790 		MLX5_SET(tirc, tirc, rx_hash_symmetric, 0);
2791 		rss_getkey((uint8_t *)hkey);
2792 #else
2793 		MLX5_SET(tirc, tirc, rx_hash_symmetric, 1);
2794 		hkey[0] = cpu_to_be32(0xD181C62C);
2795 		hkey[1] = cpu_to_be32(0xF7F4DB5B);
2796 		hkey[2] = cpu_to_be32(0x1983A2FC);
2797 		hkey[3] = cpu_to_be32(0x943E1ADB);
2798 		hkey[4] = cpu_to_be32(0xD9389E6B);
2799 		hkey[5] = cpu_to_be32(0xD1039C2C);
2800 		hkey[6] = cpu_to_be32(0xA74499AD);
2801 		hkey[7] = cpu_to_be32(0x593D56D9);
2802 		hkey[8] = cpu_to_be32(0xF3253C06);
2803 		hkey[9] = cpu_to_be32(0x2ADC1FFC);
2804 #endif
2805 		break;
2806 	}
2807 
2808 	switch (tt) {
2809 	case MLX5E_TT_IPV4_TCP:
2810 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2811 		    MLX5_L3_PROT_TYPE_IPV4);
2812 		MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2813 		    MLX5_L4_PROT_TYPE_TCP);
2814 #ifdef RSS
2815 		if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4)) {
2816 			MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2817 			    MLX5_HASH_IP);
2818 		} else
2819 #endif
2820 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2821 		    MLX5_HASH_ALL);
2822 		break;
2823 
2824 	case MLX5E_TT_IPV6_TCP:
2825 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2826 		    MLX5_L3_PROT_TYPE_IPV6);
2827 		MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2828 		    MLX5_L4_PROT_TYPE_TCP);
2829 #ifdef RSS
2830 		if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) {
2831 			MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2832 			    MLX5_HASH_IP);
2833 		} else
2834 #endif
2835 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2836 		    MLX5_HASH_ALL);
2837 		break;
2838 
2839 	case MLX5E_TT_IPV4_UDP:
2840 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2841 		    MLX5_L3_PROT_TYPE_IPV4);
2842 		MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2843 		    MLX5_L4_PROT_TYPE_UDP);
2844 #ifdef RSS
2845 		if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4)) {
2846 			MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2847 			    MLX5_HASH_IP);
2848 		} else
2849 #endif
2850 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2851 		    MLX5_HASH_ALL);
2852 		break;
2853 
2854 	case MLX5E_TT_IPV6_UDP:
2855 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2856 		    MLX5_L3_PROT_TYPE_IPV6);
2857 		MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2858 		    MLX5_L4_PROT_TYPE_UDP);
2859 #ifdef RSS
2860 		if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) {
2861 			MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2862 			    MLX5_HASH_IP);
2863 		} else
2864 #endif
2865 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2866 		    MLX5_HASH_ALL);
2867 		break;
2868 
2869 	case MLX5E_TT_IPV4_IPSEC_AH:
2870 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2871 		    MLX5_L3_PROT_TYPE_IPV4);
2872 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2873 		    MLX5_HASH_IP_IPSEC_SPI);
2874 		break;
2875 
2876 	case MLX5E_TT_IPV6_IPSEC_AH:
2877 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2878 		    MLX5_L3_PROT_TYPE_IPV6);
2879 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2880 		    MLX5_HASH_IP_IPSEC_SPI);
2881 		break;
2882 
2883 	case MLX5E_TT_IPV4_IPSEC_ESP:
2884 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2885 		    MLX5_L3_PROT_TYPE_IPV4);
2886 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2887 		    MLX5_HASH_IP_IPSEC_SPI);
2888 		break;
2889 
2890 	case MLX5E_TT_IPV6_IPSEC_ESP:
2891 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2892 		    MLX5_L3_PROT_TYPE_IPV6);
2893 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2894 		    MLX5_HASH_IP_IPSEC_SPI);
2895 		break;
2896 
2897 	case MLX5E_TT_IPV4:
2898 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2899 		    MLX5_L3_PROT_TYPE_IPV4);
2900 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2901 		    MLX5_HASH_IP);
2902 		break;
2903 
2904 	case MLX5E_TT_IPV6:
2905 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2906 		    MLX5_L3_PROT_TYPE_IPV6);
2907 		MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2908 		    MLX5_HASH_IP);
2909 		break;
2910 
2911 	default:
2912 		break;
2913 	}
2914 }
2915 
2916 static int
2917 mlx5e_open_tir(struct mlx5e_priv *priv, int tt)
2918 {
2919 	struct mlx5_core_dev *mdev = priv->mdev;
2920 	u32 *in;
2921 	void *tirc;
2922 	int inlen;
2923 	int err;
2924 
2925 	inlen = MLX5_ST_SZ_BYTES(create_tir_in);
2926 	in = mlx5_vzalloc(inlen);
2927 	if (in == NULL)
2928 		return (-ENOMEM);
2929 	tirc = MLX5_ADDR_OF(create_tir_in, in, tir_context);
2930 
2931 	mlx5e_build_tir_ctx(priv, tirc, tt);
2932 
2933 	err = mlx5_core_create_tir(mdev, in, inlen, &priv->tirn[tt]);
2934 
2935 	kvfree(in);
2936 
2937 	return (err);
2938 }
2939 
2940 static void
2941 mlx5e_close_tir(struct mlx5e_priv *priv, int tt)
2942 {
2943 	mlx5_core_destroy_tir(priv->mdev, priv->tirn[tt]);
2944 }
2945 
2946 static int
2947 mlx5e_open_tirs(struct mlx5e_priv *priv)
2948 {
2949 	int err;
2950 	int i;
2951 
2952 	for (i = 0; i < MLX5E_NUM_TT; i++) {
2953 		err = mlx5e_open_tir(priv, i);
2954 		if (err)
2955 			goto err_close_tirs;
2956 	}
2957 
2958 	return (0);
2959 
2960 err_close_tirs:
2961 	for (i--; i >= 0; i--)
2962 		mlx5e_close_tir(priv, i);
2963 
2964 	return (err);
2965 }
2966 
2967 static void
2968 mlx5e_close_tirs(struct mlx5e_priv *priv)
2969 {
2970 	int i;
2971 
2972 	for (i = 0; i < MLX5E_NUM_TT; i++)
2973 		mlx5e_close_tir(priv, i);
2974 }
2975 
2976 /*
2977  * SW MTU does not include headers,
2978  * HW MTU includes all headers and checksums.
2979  */
2980 static int
2981 mlx5e_set_dev_port_mtu(struct ifnet *ifp, int sw_mtu)
2982 {
2983 	struct mlx5e_priv *priv = ifp->if_softc;
2984 	struct mlx5_core_dev *mdev = priv->mdev;
2985 	int hw_mtu;
2986 	int err;
2987 
2988 	hw_mtu = MLX5E_SW2HW_MTU(sw_mtu);
2989 
2990 	err = mlx5_set_port_mtu(mdev, hw_mtu);
2991 	if (err) {
2992 		mlx5_en_err(ifp, "mlx5_set_port_mtu failed setting %d, err=%d\n",
2993 		    sw_mtu, err);
2994 		return (err);
2995 	}
2996 
2997 	/* Update vport context MTU */
2998 	err = mlx5_set_vport_mtu(mdev, hw_mtu);
2999 	if (err) {
3000 		mlx5_en_err(ifp,
3001 		    "Failed updating vport context with MTU size, err=%d\n",
3002 		    err);
3003 	}
3004 
3005 	ifp->if_mtu = sw_mtu;
3006 
3007 	err = mlx5_query_vport_mtu(mdev, &hw_mtu);
3008 	if (err || !hw_mtu) {
3009 		/* fallback to port oper mtu */
3010 		err = mlx5_query_port_oper_mtu(mdev, &hw_mtu);
3011 	}
3012 	if (err) {
3013 		mlx5_en_err(ifp,
3014 		    "Query port MTU, after setting new MTU value, failed\n");
3015 		return (err);
3016 	} else if (MLX5E_HW2SW_MTU(hw_mtu) < sw_mtu) {
3017 		err = -E2BIG,
3018 		mlx5_en_err(ifp,
3019 		    "Port MTU %d is smaller than ifp mtu %d\n",
3020 		    hw_mtu, sw_mtu);
3021 	} else if (MLX5E_HW2SW_MTU(hw_mtu) > sw_mtu) {
3022 		err = -EINVAL;
3023                 mlx5_en_err(ifp,
3024 		    "Port MTU %d is bigger than ifp mtu %d\n",
3025 		    hw_mtu, sw_mtu);
3026 	}
3027 	priv->params_ethtool.hw_mtu = hw_mtu;
3028 
3029 	/* compute MSB */
3030 	while (hw_mtu & (hw_mtu - 1))
3031 		hw_mtu &= (hw_mtu - 1);
3032 	priv->params_ethtool.hw_mtu_msb = hw_mtu;
3033 
3034 	return (err);
3035 }
3036 
3037 int
3038 mlx5e_open_locked(struct ifnet *ifp)
3039 {
3040 	struct mlx5e_priv *priv = ifp->if_softc;
3041 	int err;
3042 	u16 set_id;
3043 
3044 	/* check if already opened */
3045 	if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
3046 		return (0);
3047 
3048 #ifdef RSS
3049 	if (rss_getnumbuckets() > priv->params.num_channels) {
3050 		mlx5_en_info(ifp,
3051 		    "NOTE: There are more RSS buckets(%u) than channels(%u) available\n",
3052 		    rss_getnumbuckets(), priv->params.num_channels);
3053 	}
3054 #endif
3055 	err = mlx5e_open_tises(priv);
3056 	if (err) {
3057 		mlx5_en_err(ifp, "mlx5e_open_tises failed, %d\n", err);
3058 		return (err);
3059 	}
3060 	err = mlx5_vport_alloc_q_counter(priv->mdev,
3061 	    MLX5_INTERFACE_PROTOCOL_ETH, &set_id);
3062 	if (err) {
3063 		mlx5_en_err(priv->ifp,
3064 		    "mlx5_vport_alloc_q_counter failed: %d\n", err);
3065 		goto err_close_tises;
3066 	}
3067 	/* store counter set ID */
3068 	priv->counter_set_id = set_id;
3069 
3070 	err = mlx5e_open_channels(priv);
3071 	if (err) {
3072 		mlx5_en_err(ifp,
3073 		    "mlx5e_open_channels failed, %d\n", err);
3074 		goto err_dalloc_q_counter;
3075 	}
3076 	err = mlx5e_open_rqt(priv);
3077 	if (err) {
3078 		mlx5_en_err(ifp, "mlx5e_open_rqt failed, %d\n", err);
3079 		goto err_close_channels;
3080 	}
3081 	err = mlx5e_open_tirs(priv);
3082 	if (err) {
3083 		mlx5_en_err(ifp, "mlx5e_open_tir failed, %d\n", err);
3084 		goto err_close_rqls;
3085 	}
3086 	err = mlx5e_open_flow_table(priv);
3087 	if (err) {
3088 		mlx5_en_err(ifp,
3089 		    "mlx5e_open_flow_table failed, %d\n", err);
3090 		goto err_close_tirs;
3091 	}
3092 	err = mlx5e_add_all_vlan_rules(priv);
3093 	if (err) {
3094 		mlx5_en_err(ifp,
3095 		    "mlx5e_add_all_vlan_rules failed, %d\n", err);
3096 		goto err_close_flow_table;
3097 	}
3098 	set_bit(MLX5E_STATE_OPENED, &priv->state);
3099 
3100 	mlx5e_update_carrier(priv);
3101 	mlx5e_set_rx_mode_core(priv);
3102 
3103 	return (0);
3104 
3105 err_close_flow_table:
3106 	mlx5e_close_flow_table(priv);
3107 
3108 err_close_tirs:
3109 	mlx5e_close_tirs(priv);
3110 
3111 err_close_rqls:
3112 	mlx5e_close_rqt(priv);
3113 
3114 err_close_channels:
3115 	mlx5e_close_channels(priv);
3116 
3117 err_dalloc_q_counter:
3118 	mlx5_vport_dealloc_q_counter(priv->mdev,
3119 	    MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
3120 
3121 err_close_tises:
3122 	mlx5e_close_tises(priv);
3123 
3124 	return (err);
3125 }
3126 
3127 static void
3128 mlx5e_open(void *arg)
3129 {
3130 	struct mlx5e_priv *priv = arg;
3131 
3132 	PRIV_LOCK(priv);
3133 	if (mlx5_set_port_status(priv->mdev, MLX5_PORT_UP))
3134 		mlx5_en_err(priv->ifp,
3135 		    "Setting port status to up failed\n");
3136 
3137 	mlx5e_open_locked(priv->ifp);
3138 	priv->ifp->if_drv_flags |= IFF_DRV_RUNNING;
3139 	PRIV_UNLOCK(priv);
3140 }
3141 
3142 int
3143 mlx5e_close_locked(struct ifnet *ifp)
3144 {
3145 	struct mlx5e_priv *priv = ifp->if_softc;
3146 
3147 	/* check if already closed */
3148 	if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3149 		return (0);
3150 
3151 	clear_bit(MLX5E_STATE_OPENED, &priv->state);
3152 
3153 	mlx5e_set_rx_mode_core(priv);
3154 	mlx5e_del_all_vlan_rules(priv);
3155 	if_link_state_change(priv->ifp, LINK_STATE_DOWN);
3156 	mlx5e_close_flow_table(priv);
3157 	mlx5e_close_tirs(priv);
3158 	mlx5e_close_rqt(priv);
3159 	mlx5e_close_channels(priv);
3160 	mlx5_vport_dealloc_q_counter(priv->mdev,
3161 	    MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
3162 	mlx5e_close_tises(priv);
3163 
3164 	return (0);
3165 }
3166 
3167 #if (__FreeBSD_version >= 1100000)
3168 static uint64_t
3169 mlx5e_get_counter(struct ifnet *ifp, ift_counter cnt)
3170 {
3171 	struct mlx5e_priv *priv = ifp->if_softc;
3172 	u64 retval;
3173 
3174 	/* PRIV_LOCK(priv); XXX not allowed */
3175 	switch (cnt) {
3176 	case IFCOUNTER_IPACKETS:
3177 		retval = priv->stats.vport.rx_packets;
3178 		break;
3179 	case IFCOUNTER_IERRORS:
3180 		retval = priv->stats.pport.in_range_len_errors +
3181 		    priv->stats.pport.out_of_range_len +
3182 		    priv->stats.pport.too_long_errors +
3183 		    priv->stats.pport.check_seq_err +
3184 		    priv->stats.pport.alignment_err;
3185 		break;
3186 	case IFCOUNTER_IQDROPS:
3187 		retval = priv->stats.vport.rx_out_of_buffer;
3188 		break;
3189 	case IFCOUNTER_OPACKETS:
3190 		retval = priv->stats.vport.tx_packets;
3191 		break;
3192 	case IFCOUNTER_OERRORS:
3193 		retval = priv->stats.port_stats_debug.out_discards;
3194 		break;
3195 	case IFCOUNTER_IBYTES:
3196 		retval = priv->stats.vport.rx_bytes;
3197 		break;
3198 	case IFCOUNTER_OBYTES:
3199 		retval = priv->stats.vport.tx_bytes;
3200 		break;
3201 	case IFCOUNTER_IMCASTS:
3202 		retval = priv->stats.vport.rx_multicast_packets;
3203 		break;
3204 	case IFCOUNTER_OMCASTS:
3205 		retval = priv->stats.vport.tx_multicast_packets;
3206 		break;
3207 	case IFCOUNTER_OQDROPS:
3208 		retval = priv->stats.vport.tx_queue_dropped;
3209 		break;
3210 	case IFCOUNTER_COLLISIONS:
3211 		retval = priv->stats.pport.collisions;
3212 		break;
3213 	default:
3214 		retval = if_get_counter_default(ifp, cnt);
3215 		break;
3216 	}
3217 	/* PRIV_UNLOCK(priv); XXX not allowed */
3218 	return (retval);
3219 }
3220 #endif
3221 
3222 static void
3223 mlx5e_set_rx_mode(struct ifnet *ifp)
3224 {
3225 	struct mlx5e_priv *priv = ifp->if_softc;
3226 
3227 	queue_work(priv->wq, &priv->set_rx_mode_work);
3228 }
3229 
3230 static int
3231 mlx5e_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3232 {
3233 	struct mlx5e_priv *priv;
3234 	struct ifreq *ifr;
3235 	struct ifdownreason *ifdr;
3236 	struct ifi2creq i2c;
3237 	int error = 0;
3238 	int mask = 0;
3239 	int size_read = 0;
3240 	int module_status;
3241 	int module_num;
3242 	int max_mtu;
3243 	uint8_t read_addr;
3244 
3245 	priv = ifp->if_softc;
3246 
3247 	/* check if detaching */
3248 	if (priv == NULL || priv->gone != 0)
3249 		return (ENXIO);
3250 
3251 	switch (command) {
3252 	case SIOCSIFMTU:
3253 		ifr = (struct ifreq *)data;
3254 
3255 		PRIV_LOCK(priv);
3256 		mlx5_query_port_max_mtu(priv->mdev, &max_mtu);
3257 
3258 		if (ifr->ifr_mtu >= MLX5E_MTU_MIN &&
3259 		    ifr->ifr_mtu <= MIN(MLX5E_MTU_MAX, max_mtu)) {
3260 			int was_opened;
3261 
3262 			was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
3263 			if (was_opened)
3264 				mlx5e_close_locked(ifp);
3265 
3266 			/* set new MTU */
3267 			mlx5e_set_dev_port_mtu(ifp, ifr->ifr_mtu);
3268 
3269 			if (was_opened)
3270 				mlx5e_open_locked(ifp);
3271 		} else {
3272 			error = EINVAL;
3273 			mlx5_en_err(ifp,
3274 			    "Invalid MTU value. Min val: %d, Max val: %d\n",
3275 			    MLX5E_MTU_MIN, MIN(MLX5E_MTU_MAX, max_mtu));
3276 		}
3277 		PRIV_UNLOCK(priv);
3278 		break;
3279 	case SIOCSIFFLAGS:
3280 		if ((ifp->if_flags & IFF_UP) &&
3281 		    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
3282 			mlx5e_set_rx_mode(ifp);
3283 			break;
3284 		}
3285 		PRIV_LOCK(priv);
3286 		if (ifp->if_flags & IFF_UP) {
3287 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3288 				if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3289 					mlx5e_open_locked(ifp);
3290 				ifp->if_drv_flags |= IFF_DRV_RUNNING;
3291 				mlx5_set_port_status(priv->mdev, MLX5_PORT_UP);
3292 			}
3293 		} else {
3294 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
3295 				mlx5_set_port_status(priv->mdev,
3296 				    MLX5_PORT_DOWN);
3297 				if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
3298 					mlx5e_close_locked(ifp);
3299 				mlx5e_update_carrier(priv);
3300 				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3301 			}
3302 		}
3303 		PRIV_UNLOCK(priv);
3304 		break;
3305 	case SIOCADDMULTI:
3306 	case SIOCDELMULTI:
3307 		mlx5e_set_rx_mode(ifp);
3308 		break;
3309 	case SIOCSIFMEDIA:
3310 	case SIOCGIFMEDIA:
3311 	case SIOCGIFXMEDIA:
3312 		ifr = (struct ifreq *)data;
3313 		error = ifmedia_ioctl(ifp, ifr, &priv->media, command);
3314 		break;
3315 	case SIOCSIFCAP:
3316 		ifr = (struct ifreq *)data;
3317 		PRIV_LOCK(priv);
3318 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3319 
3320 		if (mask & IFCAP_TXCSUM) {
3321 			ifp->if_capenable ^= IFCAP_TXCSUM;
3322 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
3323 
3324 			if (IFCAP_TSO4 & ifp->if_capenable &&
3325 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
3326 				mask &= ~IFCAP_TSO4;
3327 				ifp->if_capenable &= ~IFCAP_TSO4;
3328 				ifp->if_hwassist &= ~CSUM_IP_TSO;
3329 				mlx5_en_err(ifp,
3330 				    "tso4 disabled due to -txcsum.\n");
3331 			}
3332 		}
3333 		if (mask & IFCAP_TXCSUM_IPV6) {
3334 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
3335 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
3336 
3337 			if (IFCAP_TSO6 & ifp->if_capenable &&
3338 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
3339 				mask &= ~IFCAP_TSO6;
3340 				ifp->if_capenable &= ~IFCAP_TSO6;
3341 				ifp->if_hwassist &= ~CSUM_IP6_TSO;
3342 				mlx5_en_err(ifp,
3343 				    "tso6 disabled due to -txcsum6.\n");
3344 			}
3345 		}
3346 		if (mask & IFCAP_NOMAP)
3347 			ifp->if_capenable ^= IFCAP_NOMAP;
3348 		if (mask & IFCAP_TXTLS4)
3349 			ifp->if_capenable ^= IFCAP_TXTLS4;
3350 		if (mask & IFCAP_TXTLS6)
3351 			ifp->if_capenable ^= IFCAP_TXTLS6;
3352 #ifdef RATELIMIT
3353 		if (mask & IFCAP_TXTLS_RTLMT)
3354 			ifp->if_capenable ^= IFCAP_TXTLS_RTLMT;
3355 #endif
3356 		if (mask & IFCAP_RXCSUM)
3357 			ifp->if_capenable ^= IFCAP_RXCSUM;
3358 		if (mask & IFCAP_RXCSUM_IPV6)
3359 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
3360 		if (mask & IFCAP_TSO4) {
3361 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
3362 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
3363 				mlx5_en_err(ifp, "enable txcsum first.\n");
3364 				error = EAGAIN;
3365 				goto out;
3366 			}
3367 			ifp->if_capenable ^= IFCAP_TSO4;
3368 			ifp->if_hwassist ^= CSUM_IP_TSO;
3369 		}
3370 		if (mask & IFCAP_TSO6) {
3371 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
3372 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
3373 				mlx5_en_err(ifp, "enable txcsum6 first.\n");
3374 				error = EAGAIN;
3375 				goto out;
3376 			}
3377 			ifp->if_capenable ^= IFCAP_TSO6;
3378 			ifp->if_hwassist ^= CSUM_IP6_TSO;
3379 		}
3380 		if (mask & IFCAP_VLAN_HWFILTER) {
3381 			if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
3382 				mlx5e_disable_vlan_filter(priv);
3383 			else
3384 				mlx5e_enable_vlan_filter(priv);
3385 
3386 			ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
3387 		}
3388 		if (mask & IFCAP_VLAN_HWTAGGING)
3389 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3390 		if (mask & IFCAP_WOL_MAGIC)
3391 			ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3392 
3393 		VLAN_CAPABILITIES(ifp);
3394 		/* turn off LRO means also turn of HW LRO - if it's on */
3395 		if (mask & IFCAP_LRO) {
3396 			int was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
3397 			bool need_restart = false;
3398 
3399 			ifp->if_capenable ^= IFCAP_LRO;
3400 
3401 			/* figure out if updating HW LRO is needed */
3402 			if (!(ifp->if_capenable & IFCAP_LRO)) {
3403 				if (priv->params.hw_lro_en) {
3404 					priv->params.hw_lro_en = false;
3405 					need_restart = true;
3406 				}
3407 			} else {
3408 				if (priv->params.hw_lro_en == false &&
3409 				    priv->params_ethtool.hw_lro != 0) {
3410 					priv->params.hw_lro_en = true;
3411 					need_restart = true;
3412 				}
3413 			}
3414 			if (was_opened && need_restart) {
3415 				mlx5e_close_locked(ifp);
3416 				mlx5e_open_locked(ifp);
3417 			}
3418 		}
3419 		if (mask & IFCAP_HWRXTSTMP) {
3420 			ifp->if_capenable ^= IFCAP_HWRXTSTMP;
3421 			if (ifp->if_capenable & IFCAP_HWRXTSTMP) {
3422 				if (priv->clbr_done == 0)
3423 					mlx5e_reset_calibration_callout(priv);
3424 			} else {
3425 				callout_drain(&priv->tstmp_clbr);
3426 				priv->clbr_done = 0;
3427 			}
3428 		}
3429 out:
3430 		PRIV_UNLOCK(priv);
3431 		break;
3432 
3433 	case SIOCGI2C:
3434 		ifr = (struct ifreq *)data;
3435 
3436 		/*
3437 		 * Copy from the user-space address ifr_data to the
3438 		 * kernel-space address i2c
3439 		 */
3440 		error = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
3441 		if (error)
3442 			break;
3443 
3444 		if (i2c.len > sizeof(i2c.data)) {
3445 			error = EINVAL;
3446 			break;
3447 		}
3448 
3449 		PRIV_LOCK(priv);
3450 		/* Get module_num which is required for the query_eeprom */
3451 		error = mlx5_query_module_num(priv->mdev, &module_num);
3452 		if (error) {
3453 			mlx5_en_err(ifp,
3454 			    "Query module num failed, eeprom reading is not supported\n");
3455 			error = EINVAL;
3456 			goto err_i2c;
3457 		}
3458 		/* Check if module is present before doing an access */
3459 		module_status = mlx5_query_module_status(priv->mdev, module_num);
3460 		if (module_status != MLX5_MODULE_STATUS_PLUGGED_ENABLED) {
3461 			error = EINVAL;
3462 			goto err_i2c;
3463 		}
3464 		/*
3465 		 * Currently 0XA0 and 0xA2 are the only addresses permitted.
3466 		 * The internal conversion is as follows:
3467 		 */
3468 		if (i2c.dev_addr == 0xA0)
3469 			read_addr = MLX5_I2C_ADDR_LOW;
3470 		else if (i2c.dev_addr == 0xA2)
3471 			read_addr = MLX5_I2C_ADDR_HIGH;
3472 		else {
3473 			mlx5_en_err(ifp,
3474 			    "Query eeprom failed, Invalid Address: %X\n",
3475 			    i2c.dev_addr);
3476 			error = EINVAL;
3477 			goto err_i2c;
3478 		}
3479 		error = mlx5_query_eeprom(priv->mdev,
3480 		    read_addr, MLX5_EEPROM_LOW_PAGE,
3481 		    (uint32_t)i2c.offset, (uint32_t)i2c.len, module_num,
3482 		    (uint32_t *)i2c.data, &size_read);
3483 		if (error) {
3484 			mlx5_en_err(ifp,
3485 			    "Query eeprom failed, eeprom reading is not supported\n");
3486 			error = EINVAL;
3487 			goto err_i2c;
3488 		}
3489 
3490 		if (i2c.len > MLX5_EEPROM_MAX_BYTES) {
3491 			error = mlx5_query_eeprom(priv->mdev,
3492 			    read_addr, MLX5_EEPROM_LOW_PAGE,
3493 			    (uint32_t)(i2c.offset + size_read),
3494 			    (uint32_t)(i2c.len - size_read), module_num,
3495 			    (uint32_t *)(i2c.data + size_read), &size_read);
3496 		}
3497 		if (error) {
3498 			mlx5_en_err(ifp,
3499 			    "Query eeprom failed, eeprom reading is not supported\n");
3500 			error = EINVAL;
3501 			goto err_i2c;
3502 		}
3503 
3504 		error = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
3505 err_i2c:
3506 		PRIV_UNLOCK(priv);
3507 		break;
3508 	case SIOCGIFDOWNREASON:
3509 		ifdr = (struct ifdownreason *)data;
3510 		bzero(ifdr->ifdr_msg, sizeof(ifdr->ifdr_msg));
3511 		PRIV_LOCK(priv);
3512 		error = -mlx5_query_pddr_troubleshooting_info(priv->mdev, NULL,
3513 		    ifdr->ifdr_msg, sizeof(ifdr->ifdr_msg));
3514 		PRIV_UNLOCK(priv);
3515 		if (error == 0)
3516 			ifdr->ifdr_reason = IFDR_REASON_MSG;
3517 		break;
3518 
3519 	default:
3520 		error = ether_ioctl(ifp, command, data);
3521 		break;
3522 	}
3523 	return (error);
3524 }
3525 
3526 static int
3527 mlx5e_check_required_hca_cap(struct mlx5_core_dev *mdev)
3528 {
3529 	/*
3530 	 * TODO: uncoment once FW really sets all these bits if
3531 	 * (!mdev->caps.eth.rss_ind_tbl_cap || !mdev->caps.eth.csum_cap ||
3532 	 * !mdev->caps.eth.max_lso_cap || !mdev->caps.eth.vlan_cap ||
3533 	 * !(mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_SCQE_BRK_MOD)) return
3534 	 * -ENOTSUPP;
3535 	 */
3536 
3537 	/* TODO: add more must-to-have features */
3538 
3539 	if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
3540 		return (-ENODEV);
3541 
3542 	return (0);
3543 }
3544 
3545 static u16
3546 mlx5e_get_max_inline_cap(struct mlx5_core_dev *mdev)
3547 {
3548 	const int min_size = ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN;
3549 	const int max_size = MLX5E_MAX_TX_INLINE;
3550 	const int bf_buf_size =
3551 	    ((1U << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2U) -
3552 	    (sizeof(struct mlx5e_tx_wqe) - 2);
3553 
3554 	/* verify against driver limits */
3555 	if (bf_buf_size > max_size)
3556 		return (max_size);
3557 	else if (bf_buf_size < min_size)
3558 		return (min_size);
3559 	else
3560 		return (bf_buf_size);
3561 }
3562 
3563 static int
3564 mlx5e_build_ifp_priv(struct mlx5_core_dev *mdev,
3565     struct mlx5e_priv *priv,
3566     int num_comp_vectors)
3567 {
3568 	int err;
3569 
3570 	/*
3571 	 * TODO: Consider link speed for setting "log_sq_size",
3572 	 * "log_rq_size" and "cq_moderation_xxx":
3573 	 */
3574 	priv->params.log_sq_size =
3575 	    MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
3576 	priv->params.log_rq_size =
3577 	    MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
3578 	priv->params.rx_cq_moderation_usec =
3579 	    MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
3580 	    MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE :
3581 	    MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
3582 	priv->params.rx_cq_moderation_mode =
3583 	    MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ? 1 : 0;
3584 	priv->params.rx_cq_moderation_pkts =
3585 	    MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS;
3586 	priv->params.tx_cq_moderation_usec =
3587 	    MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC;
3588 	priv->params.tx_cq_moderation_pkts =
3589 	    MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
3590 	priv->params.min_rx_wqes =
3591 	    MLX5E_PARAMS_DEFAULT_MIN_RX_WQES;
3592 	priv->params.rx_hash_log_tbl_sz =
3593 	    (order_base_2(num_comp_vectors) >
3594 	    MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ) ?
3595 	    order_base_2(num_comp_vectors) :
3596 	    MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ;
3597 	priv->params.num_tc = 1;
3598 	priv->params.default_vlan_prio = 0;
3599 	priv->counter_set_id = -1;
3600 	priv->params.tx_max_inline = mlx5e_get_max_inline_cap(mdev);
3601 
3602 	err = mlx5_query_min_inline(mdev, &priv->params.tx_min_inline_mode);
3603 	if (err)
3604 		return (err);
3605 
3606 	/*
3607 	 * hw lro is currently defaulted to off. when it won't anymore we
3608 	 * will consider the HW capability: "!!MLX5_CAP_ETH(mdev, lro_cap)"
3609 	 */
3610 	priv->params.hw_lro_en = false;
3611 	priv->params.lro_wqe_sz = MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ;
3612 
3613 	/*
3614 	 * CQE zipping is currently defaulted to off. when it won't
3615 	 * anymore we will consider the HW capability:
3616 	 * "!!MLX5_CAP_GEN(mdev, cqe_compression)"
3617 	 */
3618 	priv->params.cqe_zipping_en = false;
3619 
3620 	priv->mdev = mdev;
3621 	priv->params.num_channels = num_comp_vectors;
3622 	priv->params.channels_rsss = 1;
3623 	priv->order_base_2_num_channels = order_base_2(num_comp_vectors);
3624 	priv->queue_mapping_channel_mask =
3625 	    roundup_pow_of_two(num_comp_vectors) - 1;
3626 	priv->num_tc = priv->params.num_tc;
3627 	priv->default_vlan_prio = priv->params.default_vlan_prio;
3628 
3629 	INIT_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
3630 	INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
3631 	INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
3632 
3633 	return (0);
3634 }
3635 
3636 static int
3637 mlx5e_create_mkey(struct mlx5e_priv *priv, u32 pdn,
3638 		  struct mlx5_core_mr *mkey)
3639 {
3640 	struct ifnet *ifp = priv->ifp;
3641 	struct mlx5_core_dev *mdev = priv->mdev;
3642 	int inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
3643 	void *mkc;
3644 	u32 *in;
3645 	int err;
3646 
3647 	in = mlx5_vzalloc(inlen);
3648 	if (in == NULL) {
3649 		mlx5_en_err(ifp, "failed to allocate inbox\n");
3650 		return (-ENOMEM);
3651 	}
3652 
3653 	mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
3654 	MLX5_SET(mkc, mkc, access_mode, MLX5_ACCESS_MODE_PA);
3655 	MLX5_SET(mkc, mkc, umr_en, 1);	/* used by HW TLS */
3656 	MLX5_SET(mkc, mkc, lw, 1);
3657 	MLX5_SET(mkc, mkc, lr, 1);
3658 
3659 	MLX5_SET(mkc, mkc, pd, pdn);
3660 	MLX5_SET(mkc, mkc, length64, 1);
3661 	MLX5_SET(mkc, mkc, qpn, 0xffffff);
3662 
3663 	err = mlx5_core_create_mkey(mdev, mkey, in, inlen);
3664 	if (err)
3665 		mlx5_en_err(ifp, "mlx5_core_create_mkey failed, %d\n",
3666 		    err);
3667 
3668 	kvfree(in);
3669 	return (err);
3670 }
3671 
3672 static const char *mlx5e_vport_stats_desc[] = {
3673 	MLX5E_VPORT_STATS(MLX5E_STATS_DESC)
3674 };
3675 
3676 static const char *mlx5e_pport_stats_desc[] = {
3677 	MLX5E_PPORT_STATS(MLX5E_STATS_DESC)
3678 };
3679 
3680 static void
3681 mlx5e_priv_static_init(struct mlx5e_priv *priv, const uint32_t channels)
3682 {
3683 	uint32_t x;
3684 
3685 	mtx_init(&priv->async_events_mtx, "mlx5async", MTX_NETWORK_LOCK, MTX_DEF);
3686 	sx_init(&priv->state_lock, "mlx5state");
3687 	callout_init_mtx(&priv->watchdog, &priv->async_events_mtx, 0);
3688 	MLX5_INIT_DOORBELL_LOCK(&priv->doorbell_lock);
3689 	for (x = 0; x != channels; x++)
3690 		mlx5e_chan_static_init(priv, &priv->channel[x], x);
3691 }
3692 
3693 static void
3694 mlx5e_priv_static_destroy(struct mlx5e_priv *priv, const uint32_t channels)
3695 {
3696 	uint32_t x;
3697 
3698 	for (x = 0; x != channels; x++)
3699 		mlx5e_chan_static_destroy(&priv->channel[x]);
3700 	callout_drain(&priv->watchdog);
3701 	mtx_destroy(&priv->async_events_mtx);
3702 	sx_destroy(&priv->state_lock);
3703 }
3704 
3705 static int
3706 sysctl_firmware(SYSCTL_HANDLER_ARGS)
3707 {
3708 	/*
3709 	 * %d.%d%.d the string format.
3710 	 * fw_rev_{maj,min,sub} return u16, 2^16 = 65536.
3711 	 * We need at most 5 chars to store that.
3712 	 * It also has: two "." and NULL at the end, which means we need 18
3713 	 * (5*3 + 3) chars at most.
3714 	 */
3715 	char fw[18];
3716 	struct mlx5e_priv *priv = arg1;
3717 	int error;
3718 
3719 	snprintf(fw, sizeof(fw), "%d.%d.%d", fw_rev_maj(priv->mdev), fw_rev_min(priv->mdev),
3720 	    fw_rev_sub(priv->mdev));
3721 	error = sysctl_handle_string(oidp, fw, sizeof(fw), req);
3722 	return (error);
3723 }
3724 
3725 static void
3726 mlx5e_disable_tx_dma(struct mlx5e_channel *ch)
3727 {
3728 	int i;
3729 
3730 	for (i = 0; i < ch->priv->num_tc; i++)
3731 		mlx5e_drain_sq(&ch->sq[i]);
3732 }
3733 
3734 static void
3735 mlx5e_reset_sq_doorbell_record(struct mlx5e_sq *sq)
3736 {
3737 
3738 	sq->doorbell.d32[0] = cpu_to_be32(MLX5_OPCODE_NOP);
3739 	sq->doorbell.d32[1] = cpu_to_be32(sq->sqn << 8);
3740 	mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
3741 	sq->doorbell.d64 = 0;
3742 }
3743 
3744 void
3745 mlx5e_resume_sq(struct mlx5e_sq *sq)
3746 {
3747 	int err;
3748 
3749 	/* check if already enabled */
3750 	if (READ_ONCE(sq->running) != 0)
3751 		return;
3752 
3753 	err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_ERR,
3754 	    MLX5_SQC_STATE_RST);
3755 	if (err != 0) {
3756 		mlx5_en_err(sq->ifp,
3757 		    "mlx5e_modify_sq() from ERR to RST failed: %d\n", err);
3758 	}
3759 
3760 	sq->cc = 0;
3761 	sq->pc = 0;
3762 
3763 	/* reset doorbell prior to moving from RST to RDY */
3764 	mlx5e_reset_sq_doorbell_record(sq);
3765 
3766 	err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST,
3767 	    MLX5_SQC_STATE_RDY);
3768 	if (err != 0) {
3769 		mlx5_en_err(sq->ifp,
3770 		    "mlx5e_modify_sq() from RST to RDY failed: %d\n", err);
3771 	}
3772 
3773 	sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
3774 	WRITE_ONCE(sq->running, 1);
3775 }
3776 
3777 static void
3778 mlx5e_enable_tx_dma(struct mlx5e_channel *ch)
3779 {
3780         int i;
3781 
3782 	for (i = 0; i < ch->priv->num_tc; i++)
3783 		mlx5e_resume_sq(&ch->sq[i]);
3784 }
3785 
3786 static void
3787 mlx5e_disable_rx_dma(struct mlx5e_channel *ch)
3788 {
3789 	struct mlx5e_rq *rq = &ch->rq;
3790 	struct epoch_tracker et;
3791 	int err;
3792 
3793 	mtx_lock(&rq->mtx);
3794 	rq->enabled = 0;
3795 	callout_stop(&rq->watchdog);
3796 	mtx_unlock(&rq->mtx);
3797 
3798 	err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
3799 	if (err != 0) {
3800 		mlx5_en_err(rq->ifp,
3801 		    "mlx5e_modify_rq() from RDY to RST failed: %d\n", err);
3802 	}
3803 
3804 	while (!mlx5_wq_ll_is_empty(&rq->wq)) {
3805 		msleep(1);
3806 		NET_EPOCH_ENTER(et);
3807 		rq->cq.mcq.comp(&rq->cq.mcq);
3808 		NET_EPOCH_EXIT(et);
3809 	}
3810 
3811 	/*
3812 	 * Transitioning into RST state will allow the FW to track less ERR state queues,
3813 	 * thus reducing the recv queue flushing time
3814 	 */
3815 	err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_ERR, MLX5_RQC_STATE_RST);
3816 	if (err != 0) {
3817 		mlx5_en_err(rq->ifp,
3818 		    "mlx5e_modify_rq() from ERR to RST failed: %d\n", err);
3819 	}
3820 }
3821 
3822 static void
3823 mlx5e_enable_rx_dma(struct mlx5e_channel *ch)
3824 {
3825 	struct mlx5e_rq *rq = &ch->rq;
3826 	struct epoch_tracker et;
3827 	int err;
3828 
3829 	rq->wq.wqe_ctr = 0;
3830 	mlx5_wq_ll_update_db_record(&rq->wq);
3831 	err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
3832 	if (err != 0) {
3833 		mlx5_en_err(rq->ifp,
3834 		    "mlx5e_modify_rq() from RST to RDY failed: %d\n", err);
3835         }
3836 
3837 	rq->enabled = 1;
3838 
3839 	NET_EPOCH_ENTER(et);
3840 	rq->cq.mcq.comp(&rq->cq.mcq);
3841 	NET_EPOCH_EXIT(et);
3842 }
3843 
3844 void
3845 mlx5e_modify_tx_dma(struct mlx5e_priv *priv, uint8_t value)
3846 {
3847 	int i;
3848 
3849 	if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3850 		return;
3851 
3852 	for (i = 0; i < priv->params.num_channels; i++) {
3853 		if (value)
3854 			mlx5e_disable_tx_dma(&priv->channel[i]);
3855 		else
3856 			mlx5e_enable_tx_dma(&priv->channel[i]);
3857 	}
3858 }
3859 
3860 void
3861 mlx5e_modify_rx_dma(struct mlx5e_priv *priv, uint8_t value)
3862 {
3863 	int i;
3864 
3865 	if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3866 		return;
3867 
3868 	for (i = 0; i < priv->params.num_channels; i++) {
3869 		if (value)
3870 			mlx5e_disable_rx_dma(&priv->channel[i]);
3871 		else
3872 			mlx5e_enable_rx_dma(&priv->channel[i]);
3873 	}
3874 }
3875 
3876 static void
3877 mlx5e_add_hw_stats(struct mlx5e_priv *priv)
3878 {
3879 	SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3880 	    OID_AUTO, "fw_version", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
3881 	    priv, 0, sysctl_firmware, "A", "HCA firmware version");
3882 
3883 	SYSCTL_ADD_STRING(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3884 	    OID_AUTO, "board_id", CTLFLAG_RD, priv->mdev->board_id, 0,
3885 	    "Board ID");
3886 }
3887 
3888 static int
3889 mlx5e_sysctl_tx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3890 {
3891 	struct mlx5e_priv *priv = arg1;
3892 	uint8_t temp[MLX5E_MAX_PRIORITY];
3893 	uint32_t tx_pfc;
3894 	int err;
3895 	int i;
3896 
3897 	PRIV_LOCK(priv);
3898 
3899 	tx_pfc = priv->params.tx_priority_flow_control;
3900 
3901 	for (i = 0; i != MLX5E_MAX_PRIORITY; i++)
3902 		temp[i] = (tx_pfc >> i) & 1;
3903 
3904 	err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY);
3905 	if (err || !req->newptr)
3906 		goto done;
3907 	err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY);
3908 	if (err)
3909 		goto done;
3910 
3911 	priv->params.tx_priority_flow_control = 0;
3912 
3913 	/* range check input value */
3914 	for (i = 0; i != MLX5E_MAX_PRIORITY; i++) {
3915 		if (temp[i] > 1) {
3916 			err = ERANGE;
3917 			goto done;
3918 		}
3919 		priv->params.tx_priority_flow_control |= (temp[i] << i);
3920 	}
3921 
3922 	/* check if update is required */
3923 	if (tx_pfc != priv->params.tx_priority_flow_control)
3924 		err = -mlx5e_set_port_pfc(priv);
3925 done:
3926 	if (err != 0)
3927 		priv->params.tx_priority_flow_control= tx_pfc;
3928 	PRIV_UNLOCK(priv);
3929 
3930 	return (err);
3931 }
3932 
3933 static int
3934 mlx5e_sysctl_rx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3935 {
3936 	struct mlx5e_priv *priv = arg1;
3937 	uint8_t temp[MLX5E_MAX_PRIORITY];
3938 	uint32_t rx_pfc;
3939 	int err;
3940 	int i;
3941 
3942 	PRIV_LOCK(priv);
3943 
3944 	rx_pfc = priv->params.rx_priority_flow_control;
3945 
3946 	for (i = 0; i != MLX5E_MAX_PRIORITY; i++)
3947 		temp[i] = (rx_pfc >> i) & 1;
3948 
3949 	err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY);
3950 	if (err || !req->newptr)
3951 		goto done;
3952 	err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY);
3953 	if (err)
3954 		goto done;
3955 
3956 	priv->params.rx_priority_flow_control = 0;
3957 
3958 	/* range check input value */
3959 	for (i = 0; i != MLX5E_MAX_PRIORITY; i++) {
3960 		if (temp[i] > 1) {
3961 			err = ERANGE;
3962 			goto done;
3963 		}
3964 		priv->params.rx_priority_flow_control |= (temp[i] << i);
3965 	}
3966 
3967 	/* check if update is required */
3968 	if (rx_pfc != priv->params.rx_priority_flow_control) {
3969 		err = -mlx5e_set_port_pfc(priv);
3970 		if (err == 0 && priv->sw_is_port_buf_owner)
3971 			err = mlx5e_update_buf_lossy(priv);
3972 	}
3973 done:
3974 	if (err != 0)
3975 		priv->params.rx_priority_flow_control= rx_pfc;
3976 	PRIV_UNLOCK(priv);
3977 
3978 	return (err);
3979 }
3980 
3981 static void
3982 mlx5e_setup_pauseframes(struct mlx5e_priv *priv)
3983 {
3984 #if (__FreeBSD_version < 1100000)
3985 	char path[96];
3986 #endif
3987 	int error;
3988 
3989 	/* enable pauseframes by default */
3990 	priv->params.tx_pauseframe_control = 1;
3991 	priv->params.rx_pauseframe_control = 1;
3992 
3993 	/* disable ports flow control, PFC, by default */
3994 	priv->params.tx_priority_flow_control = 0;
3995 	priv->params.rx_priority_flow_control = 0;
3996 
3997 #if (__FreeBSD_version < 1100000)
3998 	/* compute path for sysctl */
3999 	snprintf(path, sizeof(path), "dev.mce.%d.tx_pauseframe_control",
4000 	    device_get_unit(priv->mdev->pdev->dev.bsddev));
4001 
4002 	/* try to fetch tunable, if any */
4003 	TUNABLE_INT_FETCH(path, &priv->params.tx_pauseframe_control);
4004 
4005 	/* compute path for sysctl */
4006 	snprintf(path, sizeof(path), "dev.mce.%d.rx_pauseframe_control",
4007 	    device_get_unit(priv->mdev->pdev->dev.bsddev));
4008 
4009 	/* try to fetch tunable, if any */
4010 	TUNABLE_INT_FETCH(path, &priv->params.rx_pauseframe_control);
4011 #endif
4012 
4013 	/* register pauseframe SYSCTLs */
4014 	SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
4015 	    OID_AUTO, "tx_pauseframe_control", CTLFLAG_RDTUN,
4016 	    &priv->params.tx_pauseframe_control, 0,
4017 	    "Set to enable TX pause frames. Clear to disable.");
4018 
4019 	SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
4020 	    OID_AUTO, "rx_pauseframe_control", CTLFLAG_RDTUN,
4021 	    &priv->params.rx_pauseframe_control, 0,
4022 	    "Set to enable RX pause frames. Clear to disable.");
4023 
4024 	/* register priority flow control, PFC, SYSCTLs */
4025 	SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
4026 	    OID_AUTO, "tx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN |
4027 	    CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_tx_priority_flow_control, "CU",
4028 	    "Set to enable TX ports flow control frames for priorities 0..7. Clear to disable.");
4029 
4030 	SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
4031 	    OID_AUTO, "rx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN |
4032 	    CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_rx_priority_flow_control, "CU",
4033 	    "Set to enable RX ports flow control frames for priorities 0..7. Clear to disable.");
4034 
4035 	PRIV_LOCK(priv);
4036 
4037 	/* range check */
4038 	priv->params.tx_pauseframe_control =
4039 	    priv->params.tx_pauseframe_control ? 1 : 0;
4040 	priv->params.rx_pauseframe_control =
4041 	    priv->params.rx_pauseframe_control ? 1 : 0;
4042 
4043 	/* update firmware */
4044 	error = mlx5e_set_port_pause_and_pfc(priv);
4045 	if (error == -EINVAL) {
4046 		mlx5_en_err(priv->ifp,
4047 		    "Global pauseframes must be disabled before enabling PFC.\n");
4048 		priv->params.rx_priority_flow_control = 0;
4049 		priv->params.tx_priority_flow_control = 0;
4050 
4051 		/* update firmware */
4052 		(void) mlx5e_set_port_pause_and_pfc(priv);
4053 	}
4054 	PRIV_UNLOCK(priv);
4055 }
4056 
4057 int
4058 mlx5e_ul_snd_tag_alloc(struct ifnet *ifp,
4059     union if_snd_tag_alloc_params *params,
4060     struct m_snd_tag **ppmt)
4061 {
4062 	struct mlx5e_priv *priv;
4063 	struct mlx5e_channel *pch;
4064 
4065 	priv = ifp->if_softc;
4066 
4067 	if (unlikely(priv->gone || params->hdr.flowtype == M_HASHTYPE_NONE)) {
4068 		return (EOPNOTSUPP);
4069 	} else {
4070 		/* keep this code synced with mlx5e_select_queue() */
4071 		u32 ch = priv->params.num_channels;
4072 #ifdef RSS
4073 		u32 temp;
4074 
4075 		if (rss_hash2bucket(params->hdr.flowid,
4076 		    params->hdr.flowtype, &temp) == 0)
4077 			ch = temp % ch;
4078 		else
4079 #endif
4080 			ch = (params->hdr.flowid % 128) % ch;
4081 
4082 		/*
4083 		 * NOTE: The channels array is only freed at detach
4084 		 * and it safe to return a pointer to the send tag
4085 		 * inside the channels structure as long as we
4086 		 * reference the priv.
4087 		 */
4088 		pch = priv->channel + ch;
4089 
4090 		/* check if send queue is not running */
4091 		if (unlikely(pch->sq[0].running == 0))
4092 			return (ENXIO);
4093 		m_snd_tag_ref(&pch->tag);
4094 		*ppmt = &pch->tag;
4095 		return (0);
4096 	}
4097 }
4098 
4099 int
4100 mlx5e_ul_snd_tag_query(struct m_snd_tag *pmt, union if_snd_tag_query_params *params)
4101 {
4102 	struct mlx5e_channel *pch =
4103 	    container_of(pmt, struct mlx5e_channel, tag);
4104 
4105 	params->unlimited.max_rate = -1ULL;
4106 	params->unlimited.queue_level = mlx5e_sq_queue_level(&pch->sq[0]);
4107 	return (0);
4108 }
4109 
4110 void
4111 mlx5e_ul_snd_tag_free(struct m_snd_tag *pmt)
4112 {
4113 	struct mlx5e_channel *pch =
4114 	    container_of(pmt, struct mlx5e_channel, tag);
4115 
4116 	complete(&pch->completion);
4117 }
4118 
4119 static int
4120 mlx5e_snd_tag_alloc(struct ifnet *ifp,
4121     union if_snd_tag_alloc_params *params,
4122     struct m_snd_tag **ppmt)
4123 {
4124 
4125 	switch (params->hdr.type) {
4126 #ifdef RATELIMIT
4127 	case IF_SND_TAG_TYPE_RATE_LIMIT:
4128 		return (mlx5e_rl_snd_tag_alloc(ifp, params, ppmt));
4129 #ifdef KERN_TLS
4130 	case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
4131 		return (mlx5e_tls_snd_tag_alloc(ifp, params, ppmt));
4132 #endif
4133 #endif
4134 	case IF_SND_TAG_TYPE_UNLIMITED:
4135 		return (mlx5e_ul_snd_tag_alloc(ifp, params, ppmt));
4136 #ifdef KERN_TLS
4137 	case IF_SND_TAG_TYPE_TLS:
4138 		return (mlx5e_tls_snd_tag_alloc(ifp, params, ppmt));
4139 #endif
4140 	default:
4141 		return (EOPNOTSUPP);
4142 	}
4143 }
4144 
4145 static int
4146 mlx5e_snd_tag_modify(struct m_snd_tag *pmt, union if_snd_tag_modify_params *params)
4147 {
4148 
4149 	switch (pmt->type) {
4150 #ifdef RATELIMIT
4151 	case IF_SND_TAG_TYPE_RATE_LIMIT:
4152 		return (mlx5e_rl_snd_tag_modify(pmt, params));
4153 #ifdef KERN_TLS
4154 	case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
4155 		return (mlx5e_tls_snd_tag_modify(pmt, params));
4156 #endif
4157 #endif
4158 	case IF_SND_TAG_TYPE_UNLIMITED:
4159 #ifdef KERN_TLS
4160 	case IF_SND_TAG_TYPE_TLS:
4161 #endif
4162 	default:
4163 		return (EOPNOTSUPP);
4164 	}
4165 }
4166 
4167 static int
4168 mlx5e_snd_tag_query(struct m_snd_tag *pmt, union if_snd_tag_query_params *params)
4169 {
4170 
4171 	switch (pmt->type) {
4172 #ifdef RATELIMIT
4173 	case IF_SND_TAG_TYPE_RATE_LIMIT:
4174 		return (mlx5e_rl_snd_tag_query(pmt, params));
4175 #ifdef KERN_TLS
4176 	case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
4177 		return (mlx5e_tls_snd_tag_query(pmt, params));
4178 #endif
4179 #endif
4180 	case IF_SND_TAG_TYPE_UNLIMITED:
4181 		return (mlx5e_ul_snd_tag_query(pmt, params));
4182 #ifdef KERN_TLS
4183 	case IF_SND_TAG_TYPE_TLS:
4184 		return (mlx5e_tls_snd_tag_query(pmt, params));
4185 #endif
4186 	default:
4187 		return (EOPNOTSUPP);
4188 	}
4189 }
4190 
4191 #ifdef RATELIMIT
4192 #define NUM_HDWR_RATES_MLX 13
4193 static const uint64_t adapter_rates_mlx[NUM_HDWR_RATES_MLX] = {
4194 	135375,			/* 1,083,000 */
4195 	180500,			/* 1,444,000 */
4196 	270750,			/* 2,166,000 */
4197 	361000,			/* 2,888,000 */
4198 	541500,			/* 4,332,000 */
4199 	721875,			/* 5,775,000 */
4200 	1082875,		/* 8,663,000 */
4201 	1443875,		/* 11,551,000 */
4202 	2165750,		/* 17,326,000 */
4203 	2887750,		/* 23,102,000 */
4204 	4331625,		/* 34,653,000 */
4205 	5775500,		/* 46,204,000 */
4206 	8663125			/* 69,305,000 */
4207 };
4208 
4209 static void
4210 mlx5e_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
4211 {
4212 	/*
4213 	 * This function needs updating by the driver maintainer!
4214 	 * For the MLX card there are currently (ConectX-4?) 13
4215 	 * pre-set rates and others i.e. ConnectX-5, 6, 7??
4216 	 *
4217 	 * This will change based on later adapters
4218 	 * and this code should be updated to look at ifp
4219 	 * and figure out the specific adapter type
4220 	 * settings i.e. how many rates as well
4221 	 * as if they are fixed (as is shown here) or
4222 	 * if they are dynamic (example chelsio t4). Also if there
4223 	 * is a maximum number of flows that the adapter
4224 	 * can handle that too needs to be updated in
4225 	 * the max_flows field.
4226 	 */
4227 	q->rate_table = adapter_rates_mlx;
4228 	q->flags = RT_IS_FIXED_TABLE;
4229 	q->max_flows = 0;	/* mlx has no limit */
4230 	q->number_of_rates = NUM_HDWR_RATES_MLX;
4231 	q->min_segment_burst = 1;
4232 }
4233 #endif
4234 
4235 static void
4236 mlx5e_snd_tag_free(struct m_snd_tag *pmt)
4237 {
4238 
4239 	switch (pmt->type) {
4240 #ifdef RATELIMIT
4241 	case IF_SND_TAG_TYPE_RATE_LIMIT:
4242 		mlx5e_rl_snd_tag_free(pmt);
4243 		break;
4244 #ifdef KERN_TLS
4245 	case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
4246 		mlx5e_tls_snd_tag_free(pmt);
4247 		break;
4248 #endif
4249 #endif
4250 	case IF_SND_TAG_TYPE_UNLIMITED:
4251 		mlx5e_ul_snd_tag_free(pmt);
4252 		break;
4253 #ifdef KERN_TLS
4254 	case IF_SND_TAG_TYPE_TLS:
4255 		mlx5e_tls_snd_tag_free(pmt);
4256 		break;
4257 #endif
4258 	default:
4259 		break;
4260 	}
4261 }
4262 
4263 static void *
4264 mlx5e_create_ifp(struct mlx5_core_dev *mdev)
4265 {
4266 	struct ifnet *ifp;
4267 	struct mlx5e_priv *priv;
4268 	u8 dev_addr[ETHER_ADDR_LEN] __aligned(4);
4269 	u8 connector_type;
4270 	struct sysctl_oid_list *child;
4271 	int ncv = mdev->priv.eq_table.num_comp_vectors;
4272 	char unit[16];
4273 	struct pfil_head_args pa;
4274 	int err;
4275 	int i,j;
4276 	u32 eth_proto_cap;
4277 	u32 out[MLX5_ST_SZ_DW(ptys_reg)];
4278 	bool ext = 0;
4279 	u32 speeds_num;
4280 	struct media media_entry = {};
4281 
4282 	if (mlx5e_check_required_hca_cap(mdev)) {
4283 		mlx5_core_dbg(mdev, "mlx5e_check_required_hca_cap() failed\n");
4284 		return (NULL);
4285 	}
4286 	/*
4287 	 * Try to allocate the priv and make room for worst-case
4288 	 * number of channel structures:
4289 	 */
4290 	priv = malloc(sizeof(*priv) +
4291 	    (sizeof(priv->channel[0]) * mdev->priv.eq_table.num_comp_vectors),
4292 	    M_MLX5EN, M_WAITOK | M_ZERO);
4293 
4294 	ifp = priv->ifp = if_alloc_dev(IFT_ETHER, mdev->pdev->dev.bsddev);
4295 	if (ifp == NULL) {
4296 		mlx5_core_err(mdev, "if_alloc() failed\n");
4297 		goto err_free_priv;
4298 	}
4299 	/* setup all static fields */
4300 	mlx5e_priv_static_init(priv, mdev->priv.eq_table.num_comp_vectors);
4301 
4302 	ifp->if_softc = priv;
4303 	if_initname(ifp, "mce", device_get_unit(mdev->pdev->dev.bsddev));
4304 	ifp->if_mtu = ETHERMTU;
4305 	ifp->if_init = mlx5e_open;
4306 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST |
4307 	    IFF_KNOWSEPOCH;
4308 	ifp->if_ioctl = mlx5e_ioctl;
4309 	ifp->if_transmit = mlx5e_xmit;
4310 	ifp->if_qflush = if_qflush;
4311 #if (__FreeBSD_version >= 1100000)
4312 	ifp->if_get_counter = mlx5e_get_counter;
4313 #endif
4314 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
4315 	/*
4316          * Set driver features
4317          */
4318 	ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6;
4319 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
4320 	ifp->if_capabilities |= IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWFILTER;
4321 	ifp->if_capabilities |= IFCAP_LINKSTATE | IFCAP_JUMBO_MTU;
4322 	ifp->if_capabilities |= IFCAP_LRO;
4323 	ifp->if_capabilities |= IFCAP_TSO | IFCAP_VLAN_HWTSO;
4324 	ifp->if_capabilities |= IFCAP_HWSTATS | IFCAP_HWRXTSTMP;
4325 	ifp->if_capabilities |= IFCAP_NOMAP;
4326 	ifp->if_capabilities |= IFCAP_TXTLS4 | IFCAP_TXTLS6;
4327 #ifdef RATELIMIT
4328 	ifp->if_capabilities |= IFCAP_TXRTLMT | IFCAP_TXTLS_RTLMT;
4329 #endif
4330 	ifp->if_snd_tag_alloc = mlx5e_snd_tag_alloc;
4331 	ifp->if_snd_tag_free = mlx5e_snd_tag_free;
4332 	ifp->if_snd_tag_modify = mlx5e_snd_tag_modify;
4333 	ifp->if_snd_tag_query = mlx5e_snd_tag_query;
4334 #ifdef RATELIMIT
4335 	ifp->if_ratelimit_query = mlx5e_ratelimit_query;
4336 #endif
4337 	/* set TSO limits so that we don't have to drop TX packets */
4338 	ifp->if_hw_tsomax = MLX5E_MAX_TX_PAYLOAD_SIZE - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
4339 	ifp->if_hw_tsomaxsegcount = MLX5E_MAX_TX_MBUF_FRAGS - 1 /* hdr */;
4340 	ifp->if_hw_tsomaxsegsize = MLX5E_MAX_TX_MBUF_SIZE;
4341 
4342 	ifp->if_capenable = ifp->if_capabilities;
4343 	ifp->if_hwassist = 0;
4344 	if (ifp->if_capenable & IFCAP_TSO)
4345 		ifp->if_hwassist |= CSUM_TSO;
4346 	if (ifp->if_capenable & IFCAP_TXCSUM)
4347 		ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP | CSUM_IP);
4348 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
4349 		ifp->if_hwassist |= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
4350 
4351 	/* ifnet sysctl tree */
4352 	sysctl_ctx_init(&priv->sysctl_ctx);
4353 	priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_dev),
4354 	    OID_AUTO, ifp->if_dname, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
4355 	    "MLX5 ethernet - interface name");
4356 	if (priv->sysctl_ifnet == NULL) {
4357 		mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
4358 		goto err_free_sysctl;
4359 	}
4360 	snprintf(unit, sizeof(unit), "%d", ifp->if_dunit);
4361 	priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
4362 	    OID_AUTO, unit, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
4363 	    "MLX5 ethernet - interface unit");
4364 	if (priv->sysctl_ifnet == NULL) {
4365 		mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
4366 		goto err_free_sysctl;
4367 	}
4368 
4369 	/* HW sysctl tree */
4370 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(mdev->pdev->dev.bsddev));
4371 	priv->sysctl_hw = SYSCTL_ADD_NODE(&priv->sysctl_ctx, child,
4372 	    OID_AUTO, "hw", CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
4373 	    "MLX5 ethernet dev hw");
4374 	if (priv->sysctl_hw == NULL) {
4375 		mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
4376 		goto err_free_sysctl;
4377 	}
4378 
4379 	err = mlx5e_build_ifp_priv(mdev, priv, ncv);
4380 	if (err) {
4381 		mlx5_core_err(mdev, "mlx5e_build_ifp_priv() failed (%d)\n", err);
4382 		goto err_free_sysctl;
4383 	}
4384 
4385 	/* reuse mlx5core's watchdog workqueue */
4386 	priv->wq = mdev->priv.health.wq_watchdog;
4387 
4388 	err = mlx5_alloc_map_uar(mdev, &priv->cq_uar);
4389 	if (err) {
4390 		mlx5_en_err(ifp, "mlx5_alloc_map_uar failed, %d\n", err);
4391 		goto err_free_wq;
4392 	}
4393 	err = mlx5_core_alloc_pd(mdev, &priv->pdn);
4394 	if (err) {
4395 		mlx5_en_err(ifp, "mlx5_core_alloc_pd failed, %d\n", err);
4396 		goto err_unmap_free_uar;
4397 	}
4398 	err = mlx5_alloc_transport_domain(mdev, &priv->tdn);
4399 	if (err) {
4400 		mlx5_en_err(ifp,
4401 		    "mlx5_alloc_transport_domain failed, %d\n", err);
4402 		goto err_dealloc_pd;
4403 	}
4404 	err = mlx5e_create_mkey(priv, priv->pdn, &priv->mr);
4405 	if (err) {
4406 		mlx5_en_err(ifp, "mlx5e_create_mkey failed, %d\n", err);
4407 		goto err_dealloc_transport_domain;
4408 	}
4409 	mlx5_query_nic_vport_mac_address(priv->mdev, 0, dev_addr);
4410 
4411 	/* check if we should generate a random MAC address */
4412 	if (MLX5_CAP_GEN(priv->mdev, vport_group_manager) == 0 &&
4413 	    is_zero_ether_addr(dev_addr)) {
4414 		random_ether_addr(dev_addr);
4415 		mlx5_en_err(ifp, "Assigned random MAC address\n");
4416 	}
4417 
4418 	err = mlx5e_rl_init(priv);
4419 	if (err) {
4420 		mlx5_en_err(ifp, "mlx5e_rl_init failed, %d\n", err);
4421 		goto err_create_mkey;
4422 	}
4423 
4424 	err = mlx5e_tls_init(priv);
4425 	if (err) {
4426 		if_printf(ifp, "%s: mlx5e_tls_init failed\n", __func__);
4427 		goto err_rl_init;
4428 	}
4429 
4430 	/* set default MTU */
4431 	mlx5e_set_dev_port_mtu(ifp, ifp->if_mtu);
4432 
4433 	/* Set default media status */
4434 	priv->media_status_last = IFM_AVALID;
4435 	priv->media_active_last = IFM_ETHER | IFM_AUTO |
4436 	    IFM_ETH_RXPAUSE | IFM_FDX;
4437 
4438 	/* setup default pauseframes configuration */
4439 	mlx5e_setup_pauseframes(priv);
4440 
4441 	/* Setup supported medias */
4442 	//TODO: If we failed to query ptys is it ok to proceed??
4443 	if (!mlx5_query_port_ptys(mdev, out, sizeof(out), MLX5_PTYS_EN, 1)) {
4444 		ext = MLX5_CAP_PCAM_FEATURE(mdev,
4445 		    ptys_extended_ethernet);
4446 		eth_proto_cap = MLX5_GET_ETH_PROTO(ptys_reg, out, ext,
4447 		    eth_proto_capability);
4448 		if (MLX5_CAP_PCAM_FEATURE(mdev, ptys_connector_type))
4449 			connector_type = MLX5_GET(ptys_reg, out,
4450 			    connector_type);
4451 	} else {
4452 		eth_proto_cap = 0;
4453 		mlx5_en_err(ifp, "Query port media capability failed, %d\n", err);
4454 	}
4455 
4456 	ifmedia_init(&priv->media, IFM_IMASK | IFM_ETH_FMASK,
4457 	    mlx5e_media_change, mlx5e_media_status);
4458 
4459 	speeds_num = ext ? MLX5E_EXT_LINK_SPEEDS_NUMBER : MLX5E_LINK_SPEEDS_NUMBER;
4460 	for (i = 0; i != speeds_num; i++) {
4461 		for (j = 0; j < MLX5E_LINK_MODES_NUMBER ; ++j) {
4462 			media_entry = ext ? mlx5e_ext_mode_table[i][j] :
4463 			    mlx5e_mode_table[i][j];
4464 			if (media_entry.baudrate == 0)
4465 				continue;
4466 			if (MLX5E_PROT_MASK(i) & eth_proto_cap) {
4467 				ifmedia_add(&priv->media,
4468 				    media_entry.subtype |
4469 				    IFM_ETHER, 0, NULL);
4470 				ifmedia_add(&priv->media,
4471 				    media_entry.subtype |
4472 				    IFM_ETHER | IFM_FDX |
4473 				    IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
4474 			}
4475 		}
4476 	}
4477 
4478 	ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO, 0, NULL);
4479 	ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
4480 	    IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
4481 
4482 	/* Set autoselect by default */
4483 	ifmedia_set(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
4484 	    IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE);
4485 
4486 	DEBUGNET_SET(ifp, mlx5_en);
4487 
4488 	ether_ifattach(ifp, dev_addr);
4489 
4490 	/* Register for VLAN events */
4491 	priv->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
4492 	    mlx5e_vlan_rx_add_vid, priv, EVENTHANDLER_PRI_FIRST);
4493 	priv->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
4494 	    mlx5e_vlan_rx_kill_vid, priv, EVENTHANDLER_PRI_FIRST);
4495 
4496 	/* Link is down by default */
4497 	if_link_state_change(ifp, LINK_STATE_DOWN);
4498 
4499 	mlx5e_enable_async_events(priv);
4500 
4501 	mlx5e_add_hw_stats(priv);
4502 
4503 	mlx5e_create_stats(&priv->stats.vport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
4504 	    "vstats", mlx5e_vport_stats_desc, MLX5E_VPORT_STATS_NUM,
4505 	    priv->stats.vport.arg);
4506 
4507 	mlx5e_create_stats(&priv->stats.pport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
4508 	    "pstats", mlx5e_pport_stats_desc, MLX5E_PPORT_STATS_NUM,
4509 	    priv->stats.pport.arg);
4510 
4511 	mlx5e_create_ethtool(priv);
4512 
4513 	mtx_lock(&priv->async_events_mtx);
4514 	mlx5e_update_stats(priv);
4515 	mtx_unlock(&priv->async_events_mtx);
4516 
4517 	SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
4518 	    OID_AUTO, "rx_clbr_done", CTLFLAG_RD,
4519 	    &priv->clbr_done, 0,
4520 	    "RX timestamps calibration state");
4521 	callout_init(&priv->tstmp_clbr, CALLOUT_DIRECT);
4522 	mlx5e_reset_calibration_callout(priv);
4523 
4524 	pa.pa_version = PFIL_VERSION;
4525 	pa.pa_flags = PFIL_IN;
4526 	pa.pa_type = PFIL_TYPE_ETHERNET;
4527 	pa.pa_headname = ifp->if_xname;
4528 	priv->pfil = pfil_head_register(&pa);
4529 
4530 	return (priv);
4531 
4532 err_rl_init:
4533 	mlx5e_rl_cleanup(priv);
4534 
4535 err_create_mkey:
4536 	mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
4537 
4538 err_dealloc_transport_domain:
4539 	mlx5_dealloc_transport_domain(mdev, priv->tdn);
4540 
4541 err_dealloc_pd:
4542 	mlx5_core_dealloc_pd(mdev, priv->pdn);
4543 
4544 err_unmap_free_uar:
4545 	mlx5_unmap_free_uar(mdev, &priv->cq_uar);
4546 
4547 err_free_wq:
4548 	flush_workqueue(priv->wq);
4549 
4550 err_free_sysctl:
4551 	sysctl_ctx_free(&priv->sysctl_ctx);
4552 	if (priv->sysctl_debug)
4553 		sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
4554 	mlx5e_priv_static_destroy(priv, mdev->priv.eq_table.num_comp_vectors);
4555 	if_free(ifp);
4556 
4557 err_free_priv:
4558 	free(priv, M_MLX5EN);
4559 	return (NULL);
4560 }
4561 
4562 static void
4563 mlx5e_destroy_ifp(struct mlx5_core_dev *mdev, void *vpriv)
4564 {
4565 	struct mlx5e_priv *priv = vpriv;
4566 	struct ifnet *ifp = priv->ifp;
4567 
4568 	/* don't allow more IOCTLs */
4569 	priv->gone = 1;
4570 
4571 	/* XXX wait a bit to allow IOCTL handlers to complete */
4572 	pause("W", hz);
4573 
4574 #ifdef RATELIMIT
4575 	/*
4576 	 * The kernel can have reference(s) via the m_snd_tag's into
4577 	 * the ratelimit channels, and these must go away before
4578 	 * detaching:
4579 	 */
4580 	while (READ_ONCE(priv->rl.stats.tx_active_connections) != 0) {
4581 		mlx5_en_err(priv->ifp,
4582 		    "Waiting for all ratelimit connections to terminate\n");
4583 		pause("W", hz);
4584 	}
4585 #endif
4586 	/* wait for all unlimited send tags to complete */
4587 	mlx5e_priv_wait_for_completion(priv, mdev->priv.eq_table.num_comp_vectors);
4588 
4589 	/* stop watchdog timer */
4590 	callout_drain(&priv->watchdog);
4591 
4592 	callout_drain(&priv->tstmp_clbr);
4593 
4594 	if (priv->vlan_attach != NULL)
4595 		EVENTHANDLER_DEREGISTER(vlan_config, priv->vlan_attach);
4596 	if (priv->vlan_detach != NULL)
4597 		EVENTHANDLER_DEREGISTER(vlan_unconfig, priv->vlan_detach);
4598 
4599 	/* make sure device gets closed */
4600 	PRIV_LOCK(priv);
4601 	mlx5e_close_locked(ifp);
4602 	PRIV_UNLOCK(priv);
4603 
4604 	/* deregister pfil */
4605 	if (priv->pfil != NULL) {
4606 		pfil_head_unregister(priv->pfil);
4607 		priv->pfil = NULL;
4608 	}
4609 
4610 	/* unregister device */
4611 	ifmedia_removeall(&priv->media);
4612 	ether_ifdetach(ifp);
4613 
4614 	mlx5e_tls_cleanup(priv);
4615 	mlx5e_rl_cleanup(priv);
4616 
4617 	/* destroy all remaining sysctl nodes */
4618 	sysctl_ctx_free(&priv->stats.vport.ctx);
4619 	sysctl_ctx_free(&priv->stats.pport.ctx);
4620 	if (priv->sysctl_debug)
4621 		sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
4622 	sysctl_ctx_free(&priv->sysctl_ctx);
4623 
4624 	mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
4625 	mlx5_dealloc_transport_domain(priv->mdev, priv->tdn);
4626 	mlx5_core_dealloc_pd(priv->mdev, priv->pdn);
4627 	mlx5_unmap_free_uar(priv->mdev, &priv->cq_uar);
4628 	mlx5e_disable_async_events(priv);
4629 	flush_workqueue(priv->wq);
4630 	mlx5e_priv_static_destroy(priv, mdev->priv.eq_table.num_comp_vectors);
4631 	if_free(ifp);
4632 	free(priv, M_MLX5EN);
4633 }
4634 
4635 #ifdef DEBUGNET
4636 static void
4637 mlx5_en_debugnet_init(struct ifnet *dev, int *nrxr, int *ncl, int *clsize)
4638 {
4639 	struct mlx5e_priv *priv = if_getsoftc(dev);
4640 
4641 	PRIV_LOCK(priv);
4642 	*nrxr = priv->params.num_channels;
4643 	*ncl = DEBUGNET_MAX_IN_FLIGHT;
4644 	*clsize = MLX5E_MAX_RX_BYTES;
4645 	PRIV_UNLOCK(priv);
4646 }
4647 
4648 static void
4649 mlx5_en_debugnet_event(struct ifnet *dev, enum debugnet_ev event)
4650 {
4651 }
4652 
4653 static int
4654 mlx5_en_debugnet_transmit(struct ifnet *dev, struct mbuf *m)
4655 {
4656 	struct mlx5e_priv *priv = if_getsoftc(dev);
4657 	struct mlx5e_sq *sq;
4658 	int err;
4659 
4660 	if ((if_getdrvflags(dev) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4661 	    IFF_DRV_RUNNING || (priv->media_status_last & IFM_ACTIVE) == 0)
4662 		return (ENOENT);
4663 
4664 	sq = &priv->channel[0].sq[0];
4665 
4666 	if (sq->running == 0) {
4667 		m_freem(m);
4668 		return (ENOENT);
4669 	}
4670 
4671 	if (mlx5e_sq_xmit(sq, &m) != 0) {
4672 		m_freem(m);
4673 		err = ENOBUFS;
4674 	} else {
4675 		err = 0;
4676 	}
4677 
4678 	if (likely(sq->doorbell.d64 != 0)) {
4679 		mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
4680 		sq->doorbell.d64 = 0;
4681 	}
4682 	return (err);
4683 }
4684 
4685 static int
4686 mlx5_en_debugnet_poll(struct ifnet *dev, int count)
4687 {
4688 	struct mlx5e_priv *priv = if_getsoftc(dev);
4689 
4690 	if ((if_getdrvflags(dev) & IFF_DRV_RUNNING) == 0 ||
4691 	    (priv->media_status_last & IFM_ACTIVE) == 0)
4692 		return (ENOENT);
4693 
4694 	mlx5_poll_interrupts(priv->mdev);
4695 
4696 	return (0);
4697 }
4698 #endif /* DEBUGNET */
4699 
4700 static void *
4701 mlx5e_get_ifp(void *vpriv)
4702 {
4703 	struct mlx5e_priv *priv = vpriv;
4704 
4705 	return (priv->ifp);
4706 }
4707 
4708 static struct mlx5_interface mlx5e_interface = {
4709 	.add = mlx5e_create_ifp,
4710 	.remove = mlx5e_destroy_ifp,
4711 	.event = mlx5e_async_event,
4712 	.protocol = MLX5_INTERFACE_PROTOCOL_ETH,
4713 	.get_dev = mlx5e_get_ifp,
4714 };
4715 
4716 void
4717 mlx5e_init(void)
4718 {
4719 	mlx5_register_interface(&mlx5e_interface);
4720 }
4721 
4722 void
4723 mlx5e_cleanup(void)
4724 {
4725 	mlx5_unregister_interface(&mlx5e_interface);
4726 }
4727 
4728 static void
4729 mlx5e_show_version(void __unused *arg)
4730 {
4731 
4732 	printf("%s", mlx5e_version);
4733 }
4734 SYSINIT(mlx5e_show_version, SI_SUB_DRIVERS, SI_ORDER_ANY, mlx5e_show_version, NULL);
4735 
4736 module_init_order(mlx5e_init, SI_ORDER_SIXTH);
4737 module_exit_order(mlx5e_cleanup, SI_ORDER_SIXTH);
4738 
4739 #if (__FreeBSD_version >= 1100000)
4740 MODULE_DEPEND(mlx5en, linuxkpi, 1, 1, 1);
4741 #endif
4742 MODULE_DEPEND(mlx5en, mlx5, 1, 1, 1);
4743 MODULE_VERSION(mlx5en, 1);
4744