2 * Copyright (c) 2015-2018 Mellanox Technologies. All rights reserved.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
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.
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
28 #include "opt_kern_tls.h"
32 #include <sys/eventhandler.h>
33 #include <sys/sockio.h>
34 #include <machine/atomic.h>
36 #include <net/debugnet.h>
38 #ifndef ETH_DRIVER_VERSION
39 #define ETH_DRIVER_VERSION "3.5.2"
41 #define DRIVER_RELDATE "September 2019"
43 static const char mlx5e_version[] = "mlx5en: Mellanox Ethernet driver "
44 ETH_DRIVER_VERSION " (" DRIVER_RELDATE ")\n";
46 static int mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs);
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;
60 static const struct media mlx5e_mode_table[MLX5E_LINK_SPEEDS_NUMBER][MLX5E_LINK_MODES_NUMBER] = {
62 [MLX5E_1000BASE_CX_SGMII][MLX5E_SGMII] = {
63 .subtype = IFM_1000_CX_SGMII,
64 .baudrate = IF_Mbps(1000ULL),
66 [MLX5E_1000BASE_KX][MLX5E_KX] = {
67 .subtype = IFM_1000_KX,
68 .baudrate = IF_Mbps(1000ULL),
70 [MLX5E_10GBASE_CX4][MLX5E_CX4] = {
71 .subtype = IFM_10G_CX4,
72 .baudrate = IF_Gbps(10ULL),
74 [MLX5E_10GBASE_KX4][MLX5E_KX4] = {
75 .subtype = IFM_10G_KX4,
76 .baudrate = IF_Gbps(10ULL),
78 [MLX5E_10GBASE_KR][MLX5E_KR] = {
79 .subtype = IFM_10G_KR,
80 .baudrate = IF_Gbps(10ULL),
82 [MLX5E_20GBASE_KR2][MLX5E_KR2] = {
83 .subtype = IFM_20G_KR2,
84 .baudrate = IF_Gbps(20ULL),
86 [MLX5E_40GBASE_CR4][MLX5E_CR4] = {
87 .subtype = IFM_40G_CR4,
88 .baudrate = IF_Gbps(40ULL),
90 [MLX5E_40GBASE_KR4][MLX5E_KR4] = {
91 .subtype = IFM_40G_KR4,
92 .baudrate = IF_Gbps(40ULL),
94 [MLX5E_56GBASE_R4][MLX5E_R] = {
95 .subtype = IFM_56G_R4,
96 .baudrate = IF_Gbps(56ULL),
98 [MLX5E_10GBASE_CR][MLX5E_CR1] = {
99 .subtype = IFM_10G_CR1,
100 .baudrate = IF_Gbps(10ULL),
102 [MLX5E_10GBASE_SR][MLX5E_SR] = {
103 .subtype = IFM_10G_SR,
104 .baudrate = IF_Gbps(10ULL),
106 [MLX5E_10GBASE_ER_LR][MLX5E_ER] = {
107 .subtype = IFM_10G_ER,
108 .baudrate = IF_Gbps(10ULL),
110 [MLX5E_10GBASE_ER_LR][MLX5E_LR] = {
111 .subtype = IFM_10G_LR,
112 .baudrate = IF_Gbps(10ULL),
114 [MLX5E_40GBASE_SR4][MLX5E_SR4] = {
115 .subtype = IFM_40G_SR4,
116 .baudrate = IF_Gbps(40ULL),
118 [MLX5E_40GBASE_LR4_ER4][MLX5E_LR4] = {
119 .subtype = IFM_40G_LR4,
120 .baudrate = IF_Gbps(40ULL),
122 [MLX5E_40GBASE_LR4_ER4][MLX5E_ER4] = {
123 .subtype = IFM_40G_ER4,
124 .baudrate = IF_Gbps(40ULL),
126 [MLX5E_100GBASE_CR4][MLX5E_CR4] = {
127 .subtype = IFM_100G_CR4,
128 .baudrate = IF_Gbps(100ULL),
130 [MLX5E_100GBASE_SR4][MLX5E_SR4] = {
131 .subtype = IFM_100G_SR4,
132 .baudrate = IF_Gbps(100ULL),
134 [MLX5E_100GBASE_KR4][MLX5E_KR4] = {
135 .subtype = IFM_100G_KR4,
136 .baudrate = IF_Gbps(100ULL),
138 [MLX5E_100GBASE_LR4][MLX5E_LR4] = {
139 .subtype = IFM_100G_LR4,
140 .baudrate = IF_Gbps(100ULL),
142 [MLX5E_100BASE_TX][MLX5E_TX] = {
143 .subtype = IFM_100_TX,
144 .baudrate = IF_Mbps(100ULL),
146 [MLX5E_1000BASE_T][MLX5E_T] = {
147 .subtype = IFM_1000_T,
148 .baudrate = IF_Mbps(1000ULL),
150 [MLX5E_10GBASE_T][MLX5E_T] = {
151 .subtype = IFM_10G_T,
152 .baudrate = IF_Gbps(10ULL),
154 [MLX5E_25GBASE_CR][MLX5E_CR] = {
155 .subtype = IFM_25G_CR,
156 .baudrate = IF_Gbps(25ULL),
158 [MLX5E_25GBASE_KR][MLX5E_KR] = {
159 .subtype = IFM_25G_KR,
160 .baudrate = IF_Gbps(25ULL),
162 [MLX5E_25GBASE_SR][MLX5E_SR] = {
163 .subtype = IFM_25G_SR,
164 .baudrate = IF_Gbps(25ULL),
166 [MLX5E_50GBASE_CR2][MLX5E_CR2] = {
167 .subtype = IFM_50G_CR2,
168 .baudrate = IF_Gbps(50ULL),
170 [MLX5E_50GBASE_KR2][MLX5E_KR2] = {
171 .subtype = IFM_50G_KR2,
172 .baudrate = IF_Gbps(50ULL),
174 [MLX5E_50GBASE_KR4][MLX5E_KR4] = {
175 .subtype = IFM_50G_KR4,
176 .baudrate = IF_Gbps(50ULL),
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),
185 [MLX5E_1000BASE_X_SGMII][MLX5E_KX] = {
186 .subtype = IFM_1000_KX,
187 .baudrate = IF_Mbps(1000),
189 [MLX5E_1000BASE_X_SGMII][MLX5E_CX_SGMII] = {
190 .subtype = IFM_1000_CX_SGMII,
191 .baudrate = IF_Mbps(1000),
193 [MLX5E_1000BASE_X_SGMII][MLX5E_CX] = {
194 .subtype = IFM_1000_CX,
195 .baudrate = IF_Mbps(1000),
197 [MLX5E_1000BASE_X_SGMII][MLX5E_LX] = {
198 .subtype = IFM_1000_LX,
199 .baudrate = IF_Mbps(1000),
201 [MLX5E_1000BASE_X_SGMII][MLX5E_SX] = {
202 .subtype = IFM_1000_SX,
203 .baudrate = IF_Mbps(1000),
205 [MLX5E_1000BASE_X_SGMII][MLX5E_T] = {
206 .subtype = IFM_1000_T,
207 .baudrate = IF_Mbps(1000),
209 [MLX5E_5GBASE_R][MLX5E_T] = {
210 .subtype = IFM_5000_T,
211 .baudrate = IF_Mbps(5000),
213 [MLX5E_5GBASE_R][MLX5E_KR] = {
214 .subtype = IFM_5000_KR,
215 .baudrate = IF_Mbps(5000),
217 [MLX5E_5GBASE_R][MLX5E_KR1] = {
218 .subtype = IFM_5000_KR1,
219 .baudrate = IF_Mbps(5000),
221 [MLX5E_5GBASE_R][MLX5E_KR_S] = {
222 .subtype = IFM_5000_KR_S,
223 .baudrate = IF_Mbps(5000),
225 [MLX5E_10GBASE_XFI_XAUI_1][MLX5E_ER] = {
226 .subtype = IFM_10G_ER,
227 .baudrate = IF_Gbps(10ULL),
229 [MLX5E_10GBASE_XFI_XAUI_1][MLX5E_KR] = {
230 .subtype = IFM_10G_KR,
231 .baudrate = IF_Gbps(10ULL),
233 [MLX5E_10GBASE_XFI_XAUI_1][MLX5E_LR] = {
234 .subtype = IFM_10G_LR,
235 .baudrate = IF_Gbps(10ULL),
237 [MLX5E_10GBASE_XFI_XAUI_1][MLX5E_SR] = {
238 .subtype = IFM_10G_SR,
239 .baudrate = IF_Gbps(10ULL),
241 [MLX5E_10GBASE_XFI_XAUI_1][MLX5E_T] = {
242 .subtype = IFM_10G_T,
243 .baudrate = IF_Gbps(10ULL),
245 [MLX5E_10GBASE_XFI_XAUI_1][MLX5E_AOC] = {
246 .subtype = IFM_10G_AOC,
247 .baudrate = IF_Gbps(10ULL),
249 [MLX5E_10GBASE_XFI_XAUI_1][MLX5E_CR1] = {
250 .subtype = IFM_10G_CR1,
251 .baudrate = IF_Gbps(10ULL),
253 [MLX5E_40GBASE_XLAUI_4_XLPPI_4][MLX5E_CR4] = {
254 .subtype = IFM_40G_CR4,
255 .baudrate = IF_Gbps(40ULL),
257 [MLX5E_40GBASE_XLAUI_4_XLPPI_4][MLX5E_KR4] = {
258 .subtype = IFM_40G_KR4,
259 .baudrate = IF_Gbps(40ULL),
261 [MLX5E_40GBASE_XLAUI_4_XLPPI_4][MLX5E_LR4] = {
262 .subtype = IFM_40G_LR4,
263 .baudrate = IF_Gbps(40ULL),
265 [MLX5E_40GBASE_XLAUI_4_XLPPI_4][MLX5E_SR4] = {
266 .subtype = IFM_40G_SR4,
267 .baudrate = IF_Gbps(40ULL),
269 [MLX5E_40GBASE_XLAUI_4_XLPPI_4][MLX5E_ER4] = {
270 .subtype = IFM_40G_ER4,
271 .baudrate = IF_Gbps(40ULL),
274 [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_CR] = {
275 .subtype = IFM_25G_CR,
276 .baudrate = IF_Gbps(25ULL),
278 [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_KR] = {
279 .subtype = IFM_25G_KR,
280 .baudrate = IF_Gbps(25ULL),
282 [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_SR] = {
283 .subtype = IFM_25G_SR,
284 .baudrate = IF_Gbps(25ULL),
286 [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_ACC] = {
287 .subtype = IFM_25G_ACC,
288 .baudrate = IF_Gbps(25ULL),
290 [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_AOC] = {
291 .subtype = IFM_25G_AOC,
292 .baudrate = IF_Gbps(25ULL),
294 [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_CR1] = {
295 .subtype = IFM_25G_CR1,
296 .baudrate = IF_Gbps(25ULL),
298 [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_CR_S] = {
299 .subtype = IFM_25G_CR_S,
300 .baudrate = IF_Gbps(25ULL),
302 [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_KR1] = {
303 .subtype = IFM_5000_KR1,
304 .baudrate = IF_Gbps(25ULL),
306 [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_KR_S] = {
307 .subtype = IFM_25G_KR_S,
308 .baudrate = IF_Gbps(25ULL),
310 [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_LR] = {
311 .subtype = IFM_25G_LR,
312 .baudrate = IF_Gbps(25ULL),
314 [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_T] = {
315 .subtype = IFM_25G_T,
316 .baudrate = IF_Gbps(25ULL),
318 [MLX5E_50GAUI_2_LAUI_2_50GBASE_CR2_KR2][MLX5E_CR2] = {
319 .subtype = IFM_50G_CR2,
320 .baudrate = IF_Gbps(50ULL),
322 [MLX5E_50GAUI_2_LAUI_2_50GBASE_CR2_KR2][MLX5E_KR2] = {
323 .subtype = IFM_50G_KR2,
324 .baudrate = IF_Gbps(50ULL),
326 [MLX5E_50GAUI_2_LAUI_2_50GBASE_CR2_KR2][MLX5E_KR4] = {
327 .subtype = IFM_50G_KR4,
328 .baudrate = IF_Gbps(50ULL),
330 [MLX5E_50GAUI_2_LAUI_2_50GBASE_CR2_KR2][MLX5E_SR2] = {
331 .subtype = IFM_50G_SR2,
332 .baudrate = IF_Gbps(50ULL),
334 [MLX5E_50GAUI_2_LAUI_2_50GBASE_CR2_KR2][MLX5E_LR2] = {
335 .subtype = IFM_50G_LR2,
336 .baudrate = IF_Gbps(50ULL),
338 [MLX5E_50GAUI_1_LAUI_1_50GBASE_CR_KR][MLX5E_LR] = {
339 .subtype = IFM_50G_LR,
340 .baudrate = IF_Gbps(50ULL),
342 [MLX5E_50GAUI_1_LAUI_1_50GBASE_CR_KR][MLX5E_SR] = {
343 .subtype = IFM_50G_SR,
344 .baudrate = IF_Gbps(50ULL),
346 [MLX5E_50GAUI_1_LAUI_1_50GBASE_CR_KR][MLX5E_CP] = {
347 .subtype = IFM_50G_CP,
348 .baudrate = IF_Gbps(50ULL),
350 [MLX5E_50GAUI_1_LAUI_1_50GBASE_CR_KR][MLX5E_FR] = {
351 .subtype = IFM_50G_FR,
352 .baudrate = IF_Gbps(50ULL),
354 [MLX5E_50GAUI_1_LAUI_1_50GBASE_CR_KR][MLX5E_KR_PAM4] = {
355 .subtype = IFM_50G_KR_PAM4,
356 .baudrate = IF_Gbps(50ULL),
358 [MLX5E_CAUI_4_100GBASE_CR4_KR4][MLX5E_CR4] = {
359 .subtype = IFM_100G_CR4,
360 .baudrate = IF_Gbps(100ULL),
362 [MLX5E_CAUI_4_100GBASE_CR4_KR4][MLX5E_KR4] = {
363 .subtype = IFM_100G_KR4,
364 .baudrate = IF_Gbps(100ULL),
366 [MLX5E_CAUI_4_100GBASE_CR4_KR4][MLX5E_LR4] = {
367 .subtype = IFM_100G_LR4,
368 .baudrate = IF_Gbps(100ULL),
370 [MLX5E_CAUI_4_100GBASE_CR4_KR4][MLX5E_SR4] = {
371 .subtype = IFM_100G_SR4,
372 .baudrate = IF_Gbps(100ULL),
374 [MLX5E_100GAUI_2_100GBASE_CR2_KR2][MLX5E_SR2] = {
375 .subtype = IFM_100G_SR2,
376 .baudrate = IF_Gbps(100ULL),
378 [MLX5E_100GAUI_2_100GBASE_CR2_KR2][MLX5E_CP2] = {
379 .subtype = IFM_100G_CP2,
380 .baudrate = IF_Gbps(100ULL),
382 [MLX5E_100GAUI_2_100GBASE_CR2_KR2][MLX5E_KR2_PAM4] = {
383 .subtype = IFM_100G_KR2_PAM4,
384 .baudrate = IF_Gbps(100ULL),
386 [MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_DR4] = {
387 .subtype = IFM_200G_DR4,
388 .baudrate = IF_Gbps(200ULL),
390 [MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_LR4] = {
391 .subtype = IFM_200G_LR4,
392 .baudrate = IF_Gbps(200ULL),
394 [MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_SR4] = {
395 .subtype = IFM_200G_SR4,
396 .baudrate = IF_Gbps(200ULL),
398 [MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_FR4] = {
399 .subtype = IFM_200G_FR4,
400 .baudrate = IF_Gbps(200ULL),
402 [MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_CR4_PAM4] = {
403 .subtype = IFM_200G_CR4_PAM4,
404 .baudrate = IF_Gbps(200ULL),
406 [MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_KR4_PAM4] = {
407 .subtype = IFM_200G_KR4_PAM4,
408 .baudrate = IF_Gbps(200ULL),
412 DEBUGNET_DEFINE(mlx5_en);
414 MALLOC_DEFINE(M_MLX5EN, "MLX5EN", "MLX5 Ethernet");
417 mlx5e_update_carrier(struct mlx5e_priv *priv)
419 struct mlx5_core_dev *mdev = priv->mdev;
420 u32 out[MLX5_ST_SZ_DW(ptys_reg)];
427 struct media media_entry = {};
429 port_state = mlx5_query_vport_state(mdev,
430 MLX5_QUERY_VPORT_STATE_IN_OP_MOD_VNIC_VPORT, 0);
432 if (port_state == VPORT_STATE_UP) {
433 priv->media_status_last |= IFM_ACTIVE;
435 priv->media_status_last &= ~IFM_ACTIVE;
436 priv->media_active_last = IFM_ETHER;
437 if_link_state_change(priv->ifp, LINK_STATE_DOWN);
441 error = mlx5_query_port_ptys(mdev, out, sizeof(out),
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",
451 ext = MLX5_CAP_PCAM_FEATURE(mdev, ptys_extended_ethernet);
452 eth_proto_oper = MLX5_GET_ETH_PROTO(ptys_reg, out, ext,
455 i = ilog2(eth_proto_oper);
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)
464 if (media_entry.subtype == 0) {
465 mlx5_en_err(priv->ifp,
466 "Could not find operational media subtype\n");
470 switch (media_entry.subtype) {
472 error = mlx5_query_pddr_range_info(mdev, 1, &is_er_type);
474 mlx5_en_err(priv->ifp,
475 "query port pddr failed: %d\n", error);
477 if (error != 0 || is_er_type == 0)
478 media_entry.subtype = IFM_10G_LR;
481 error = mlx5_query_pddr_range_info(mdev, 1, &is_er_type);
483 mlx5_en_err(priv->ifp,
484 "query port pddr failed: %d\n", error);
486 if (error == 0 && is_er_type != 0)
487 media_entry.subtype = IFM_40G_ER4;
490 priv->media_active_last = media_entry.subtype | IFM_ETHER | IFM_FDX;
491 priv->ifp->if_baudrate = media_entry.baudrate;
493 if_link_state_change(priv->ifp, LINK_STATE_UP);
497 mlx5e_media_status(struct ifnet *dev, struct ifmediareq *ifmr)
499 struct mlx5e_priv *priv = dev->if_softc;
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);
509 mlx5e_find_link_mode(u32 subtype, bool ext)
515 struct media media_entry = {};
519 subtype = IFM_10G_ER;
522 subtype = IFM_40G_LR4;
526 speeds_num = ext ? MLX5E_EXT_LINK_SPEEDS_NUMBER :
527 MLX5E_LINK_SPEEDS_NUMBER;
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)
535 if (media_entry.subtype == subtype) {
536 link_mode |= MLX5E_PROT_MASK(i);
545 mlx5e_set_port_pause_and_pfc(struct mlx5e_priv *priv)
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));
555 mlx5e_set_port_pfc(struct mlx5e_priv *priv)
559 if (priv->gone != 0) {
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");
567 error = mlx5e_set_port_pause_and_pfc(priv);
573 mlx5e_media_change(struct ifnet *dev)
575 struct mlx5e_priv *priv = dev->if_softc;
576 struct mlx5_core_dev *mdev = priv->mdev;
579 u32 out[MLX5_ST_SZ_DW(ptys_reg)];
585 locked = PRIV_LOCKED(priv);
589 if (IFM_TYPE(priv->media.ifm_media) != IFM_ETHER) {
594 error = mlx5_query_port_ptys(mdev, out, sizeof(out),
597 mlx5_en_err(dev, "Query port media capability failed\n");
601 ext = MLX5_CAP_PCAM_FEATURE(mdev, ptys_extended_ethernet);
602 link_mode = mlx5e_find_link_mode(IFM_SUBTYPE(priv->media.ifm_media), ext);
604 /* query supported capabilities */
605 eth_proto_cap = MLX5_GET_ETH_PROTO(ptys_reg, out, ext,
606 eth_proto_capability);
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");
617 link_mode = link_mode & eth_proto_cap;
618 if (link_mode == 0) {
619 mlx5_en_err(dev, "Not supported link mode requested\n");
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");
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;
639 /* check if device is opened */
640 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
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);
647 mlx5_set_port_status(mdev, MLX5_PORT_UP);
656 mlx5e_update_carrier_work(struct work_struct *work)
658 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
659 update_carrier_work);
662 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
663 mlx5e_update_carrier(priv);
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);
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);
674 mlx5e_update_pcie_counters(struct mlx5e_priv *priv)
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);
683 /* allocate firmware request structures */
684 in = mlx5_vzalloc(sz);
685 out = mlx5_vzalloc(sz);
686 if (in == NULL || out == NULL)
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);
694 MLX5E_PCIE_PERFORMANCE_COUNTERS_64(MLX5E_PCIE_PERF_GET_64)
695 MLX5E_PCIE_PERFORMANCE_COUNTERS_32(MLX5E_PCIE_PERF_GET_32)
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);
702 MLX5E_PCIE_TIMERS_AND_STATES_COUNTERS_32(MLX5E_PCIE_PERF_GET_32)
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);
709 MLX5E_PCIE_LANE_COUNTERS_32(MLX5E_PCIE_PERF_GET_32)
712 /* free firmware request structures */
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.
724 mlx5e_update_pport_counters(struct mlx5e_priv *priv)
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;
732 unsigned sz = MLX5_ST_SZ_BYTES(ppcnt_reg);
737 /* allocate firmware request structures */
738 in = mlx5_vzalloc(sz);
739 out = mlx5_vzalloc(sz);
740 if (in == NULL || out == NULL)
744 * Get pointer to the 64-bit counter set which is located at a
745 * fixed offset in the output firmware request structure:
747 ptr = (const uint64_t *)MLX5_ADDR_OF(ppcnt_reg, out, counter_set);
749 MLX5_SET(ppcnt_reg, in, local_port, 1);
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]);
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]);
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]);
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]);
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]);
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]);
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);
794 for (x = 0; x != MLX5E_PPORT_STATISTICAL_DEBUG_NUM; x++, y++)
795 s_debug->arg[y] = be64toh(ptr[x]);
798 /* read PCIE counters */
799 mlx5e_update_pcie_counters(priv);
801 /* read per-priority counters */
802 MLX5_SET(ppcnt_reg, in, grp, MLX5_PER_PRIORITY_COUNTERS_GROUP);
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);
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]);
816 /* free firmware request structures */
822 mlx5e_grp_vnic_env_update_stats(struct mlx5e_priv *priv)
824 u32 out[MLX5_ST_SZ_DW(query_vnic_env_out)] = {};
825 u32 in[MLX5_ST_SZ_DW(query_vnic_env_in)] = {};
827 if (!MLX5_CAP_GEN(priv->mdev, nic_receive_steering_discard))
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);
835 if (mlx5_cmd_exec(priv->mdev, in, sizeof(in), out, sizeof(out)) != 0)
838 priv->stats.vport.rx_steer_missed_packets =
839 MLX5_GET64(query_vnic_env_out, out,
840 vport_env.nic_receive_steering_discard);
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.
850 mlx5e_update_stats_locked(struct mlx5e_priv *priv)
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;
859 u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)];
861 int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
864 u64 tx_queue_dropped = 0;
865 u64 tx_defragged = 0;
866 u64 tx_offload_none = 0;
869 u64 sw_lro_queued = 0;
870 u64 sw_lro_flushed = 0;
871 u64 rx_csum_none = 0;
875 u32 rx_out_of_buffer = 0;
880 out = mlx5_vzalloc(outlen);
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;
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;
902 for (j = 0; j < priv->num_tc; j++) {
903 sq_stats = &pch->sq[j].stats;
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;
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;
928 mlx5e_grp_vnic_env_update_stats(priv);
931 memset(in, 0, sizeof(in));
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);
938 memset(out, 0, outlen);
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;
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)
952 s->rx_error_packets =
953 MLX5_GET_CTR(out, received_errors.packets);
955 MLX5_GET_CTR(out, received_errors.octets);
956 s->tx_error_packets =
957 MLX5_GET_CTR(out, transmit_errors.packets);
959 MLX5_GET_CTR(out, transmit_errors.octets);
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);
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);
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);
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;
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;
998 /* Get physical port counters */
999 mlx5e_update_pport_counters(priv);
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;
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;
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);
1035 mlx5_en_err(priv->ifp,
1036 "Failed reading diagnostics: %d\n", error);
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);
1048 mlx5e_update_stats_work(struct work_struct *work)
1050 struct mlx5e_priv *priv;
1052 priv = container_of(work, struct mlx5e_priv, update_stats_work);
1054 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0 &&
1055 !test_bit(MLX5_INTERFACE_STATE_TEARDOWN, &priv->mdev->intf_state))
1056 mlx5e_update_stats_locked(priv);
1061 mlx5e_update_stats(void *arg)
1063 struct mlx5e_priv *priv = arg;
1065 queue_work(priv->wq, &priv->update_stats_work);
1067 callout_reset(&priv->watchdog, hz, &mlx5e_update_stats, priv);
1071 mlx5e_async_event_sub(struct mlx5e_priv *priv,
1072 enum mlx5_dev_event event)
1075 case MLX5_DEV_EVENT_PORT_UP:
1076 case MLX5_DEV_EVENT_PORT_DOWN:
1077 queue_work(priv->wq, &priv->update_carrier_work);
1086 mlx5e_async_event(struct mlx5_core_dev *mdev, void *vpriv,
1087 enum mlx5_dev_event event, unsigned long param)
1089 struct mlx5e_priv *priv = vpriv;
1091 mtx_lock(&priv->async_events_mtx);
1092 if (test_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state))
1093 mlx5e_async_event_sub(priv, event);
1094 mtx_unlock(&priv->async_events_mtx);
1098 mlx5e_enable_async_events(struct mlx5e_priv *priv)
1100 set_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
1104 mlx5e_disable_async_events(struct mlx5e_priv *priv)
1106 mtx_lock(&priv->async_events_mtx);
1107 clear_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
1108 mtx_unlock(&priv->async_events_mtx);
1111 static void mlx5e_calibration_callout(void *arg);
1112 static int mlx5e_calibration_duration = 20;
1113 static int mlx5e_fast_calibration = 1;
1114 static int mlx5e_normal_calibration = 30;
1116 static SYSCTL_NODE(_hw_mlx5, OID_AUTO, calibr, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
1117 "MLX5 timestamp calibration parameteres");
1119 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, duration, CTLFLAG_RWTUN,
1120 &mlx5e_calibration_duration, 0,
1121 "Duration of initial calibration");
1122 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, fast, CTLFLAG_RWTUN,
1123 &mlx5e_fast_calibration, 0,
1124 "Recalibration interval during initial calibration");
1125 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, normal, CTLFLAG_RWTUN,
1126 &mlx5e_normal_calibration, 0,
1127 "Recalibration interval during normal operations");
1130 * Ignites the calibration process.
1133 mlx5e_reset_calibration_callout(struct mlx5e_priv *priv)
1136 if (priv->clbr_done == 0)
1137 mlx5e_calibration_callout(priv);
1139 callout_reset_curcpu(&priv->tstmp_clbr, (priv->clbr_done <
1140 mlx5e_calibration_duration ? mlx5e_fast_calibration :
1141 mlx5e_normal_calibration) * hz, mlx5e_calibration_callout,
1146 mlx5e_timespec2usec(const struct timespec *ts)
1149 return ((uint64_t)ts->tv_sec * 1000000000 + ts->tv_nsec);
1153 mlx5e_hw_clock(struct mlx5e_priv *priv)
1155 struct mlx5_init_seg *iseg;
1156 uint32_t hw_h, hw_h1, hw_l;
1158 iseg = priv->mdev->iseg;
1160 hw_h = ioread32be(&iseg->internal_timer_h);
1161 hw_l = ioread32be(&iseg->internal_timer_l);
1162 hw_h1 = ioread32be(&iseg->internal_timer_h);
1163 } while (hw_h1 != hw_h);
1164 return (((uint64_t)hw_h << 32) | hw_l);
1168 * The calibration callout, it runs either in the context of the
1169 * thread which enables calibration, or in callout. It takes the
1170 * snapshot of system and adapter clocks, then advances the pointers to
1171 * the calibration point to allow rx path to read the consistent data
1175 mlx5e_calibration_callout(void *arg)
1177 struct mlx5e_priv *priv;
1178 struct mlx5e_clbr_point *next, *curr;
1183 curr = &priv->clbr_points[priv->clbr_curr];
1184 clbr_curr_next = priv->clbr_curr + 1;
1185 if (clbr_curr_next >= nitems(priv->clbr_points))
1187 next = &priv->clbr_points[clbr_curr_next];
1189 next->base_prev = curr->base_curr;
1190 next->clbr_hw_prev = curr->clbr_hw_curr;
1192 next->clbr_hw_curr = mlx5e_hw_clock(priv);
1193 if (((next->clbr_hw_curr - curr->clbr_hw_curr) >> MLX5E_TSTMP_PREC) ==
1195 if (priv->clbr_done != 0) {
1196 mlx5_en_err(priv->ifp,
1197 "HW failed tstmp frozen %#jx %#jx, disabling\n",
1198 next->clbr_hw_curr, curr->clbr_hw_prev);
1199 priv->clbr_done = 0;
1201 atomic_store_rel_int(&curr->clbr_gen, 0);
1206 next->base_curr = mlx5e_timespec2usec(&ts);
1209 atomic_thread_fence_rel();
1210 priv->clbr_curr = clbr_curr_next;
1211 atomic_store_rel_int(&next->clbr_gen, ++(priv->clbr_gen));
1213 if (priv->clbr_done < mlx5e_calibration_duration)
1215 mlx5e_reset_calibration_callout(priv);
1218 static const char *mlx5e_rq_stats_desc[] = {
1219 MLX5E_RQ_STATS(MLX5E_STATS_DESC)
1223 mlx5e_create_rq(struct mlx5e_channel *c,
1224 struct mlx5e_rq_param *param,
1225 struct mlx5e_rq *rq)
1227 struct mlx5e_priv *priv = c->priv;
1228 struct mlx5_core_dev *mdev = priv->mdev;
1230 void *rqc = param->rqc;
1231 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
1237 err = mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
1241 /* Create DMA descriptor TAG */
1242 if ((err = -bus_dma_tag_create(
1243 bus_get_dma_tag(mdev->pdev->dev.bsddev),
1244 1, /* any alignment */
1245 0, /* no boundary */
1246 BUS_SPACE_MAXADDR, /* lowaddr */
1247 BUS_SPACE_MAXADDR, /* highaddr */
1248 NULL, NULL, /* filter, filterarg */
1249 nsegs * MLX5E_MAX_RX_BYTES, /* maxsize */
1250 nsegs, /* nsegments */
1251 nsegs * MLX5E_MAX_RX_BYTES, /* maxsegsize */
1253 NULL, NULL, /* lockfunc, lockfuncarg */
1257 err = mlx5_wq_ll_create(mdev, ¶m->wq, rqc_wq, &rq->wq,
1260 goto err_free_dma_tag;
1262 rq->wq.db = &rq->wq.db[MLX5_RCV_DBR];
1264 err = mlx5e_get_wqe_sz(priv, &rq->wqe_sz, &rq->nsegs);
1266 goto err_rq_wq_destroy;
1268 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
1270 err = -tcp_lro_init_args(&rq->lro, priv->ifp, TCP_LRO_ENTRIES, wq_sz);
1272 goto err_rq_wq_destroy;
1274 rq->mbuf = malloc(wq_sz * sizeof(rq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
1275 for (i = 0; i != wq_sz; i++) {
1276 struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
1279 err = -bus_dmamap_create(rq->dma_tag, 0, &rq->mbuf[i].dma_map);
1282 bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
1283 goto err_rq_mbuf_free;
1286 /* set value for constant fields */
1287 for (j = 0; j < rq->nsegs; j++)
1288 wqe->data[j].lkey = cpu_to_be32(priv->mr.key);
1291 INIT_WORK(&rq->dim.work, mlx5e_dim_work);
1292 if (priv->params.rx_cq_moderation_mode < 2) {
1293 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
1295 void *cqc = container_of(param,
1296 struct mlx5e_channel_param, rq)->rx_cq.cqc;
1298 switch (MLX5_GET(cqc, cqc, cq_period_mode)) {
1299 case MLX5_CQ_PERIOD_MODE_START_FROM_EQE:
1300 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
1302 case MLX5_CQ_PERIOD_MODE_START_FROM_CQE:
1303 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
1306 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
1311 rq->ifp = priv->ifp;
1315 snprintf(buffer, sizeof(buffer), "rxstat%d", c->ix);
1316 mlx5e_create_stats(&rq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
1317 buffer, mlx5e_rq_stats_desc, MLX5E_RQ_STATS_NUM,
1322 free(rq->mbuf, M_MLX5EN);
1323 tcp_lro_free(&rq->lro);
1325 mlx5_wq_destroy(&rq->wq_ctrl);
1327 bus_dma_tag_destroy(rq->dma_tag);
1333 mlx5e_destroy_rq(struct mlx5e_rq *rq)
1338 /* destroy all sysctl nodes */
1339 sysctl_ctx_free(&rq->stats.ctx);
1341 /* free leftover LRO packets, if any */
1342 tcp_lro_free(&rq->lro);
1344 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
1345 for (i = 0; i != wq_sz; i++) {
1346 if (rq->mbuf[i].mbuf != NULL) {
1347 bus_dmamap_unload(rq->dma_tag, rq->mbuf[i].dma_map);
1348 m_freem(rq->mbuf[i].mbuf);
1350 bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
1352 free(rq->mbuf, M_MLX5EN);
1353 mlx5_wq_destroy(&rq->wq_ctrl);
1354 bus_dma_tag_destroy(rq->dma_tag);
1358 mlx5e_enable_rq(struct mlx5e_rq *rq, struct mlx5e_rq_param *param)
1360 struct mlx5e_channel *c = rq->channel;
1361 struct mlx5e_priv *priv = c->priv;
1362 struct mlx5_core_dev *mdev = priv->mdev;
1370 inlen = MLX5_ST_SZ_BYTES(create_rq_in) +
1371 sizeof(u64) * rq->wq_ctrl.buf.npages;
1372 in = mlx5_vzalloc(inlen);
1376 rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
1377 wq = MLX5_ADDR_OF(rqc, rqc, wq);
1379 memcpy(rqc, param->rqc, sizeof(param->rqc));
1381 MLX5_SET(rqc, rqc, cqn, c->rq.cq.mcq.cqn);
1382 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST);
1383 MLX5_SET(rqc, rqc, flush_in_error_en, 1);
1384 if (priv->counter_set_id >= 0)
1385 MLX5_SET(rqc, rqc, counter_set_id, priv->counter_set_id);
1386 MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift -
1388 MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma);
1390 mlx5_fill_page_array(&rq->wq_ctrl.buf,
1391 (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
1393 err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn);
1401 mlx5e_modify_rq(struct mlx5e_rq *rq, int curr_state, int next_state)
1403 struct mlx5e_channel *c = rq->channel;
1404 struct mlx5e_priv *priv = c->priv;
1405 struct mlx5_core_dev *mdev = priv->mdev;
1412 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
1413 in = mlx5_vzalloc(inlen);
1417 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
1419 MLX5_SET(modify_rq_in, in, rqn, rq->rqn);
1420 MLX5_SET(modify_rq_in, in, rq_state, curr_state);
1421 MLX5_SET(rqc, rqc, state, next_state);
1423 err = mlx5_core_modify_rq(mdev, in, inlen);
1431 mlx5e_disable_rq(struct mlx5e_rq *rq)
1433 struct mlx5e_channel *c = rq->channel;
1434 struct mlx5e_priv *priv = c->priv;
1435 struct mlx5_core_dev *mdev = priv->mdev;
1437 mlx5_core_destroy_rq(mdev, rq->rqn);
1441 mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq)
1443 struct mlx5e_channel *c = rq->channel;
1444 struct mlx5e_priv *priv = c->priv;
1445 struct mlx5_wq_ll *wq = &rq->wq;
1448 for (i = 0; i < 1000; i++) {
1449 if (wq->cur_sz >= priv->params.min_rx_wqes)
1454 return (-ETIMEDOUT);
1458 mlx5e_open_rq(struct mlx5e_channel *c,
1459 struct mlx5e_rq_param *param,
1460 struct mlx5e_rq *rq)
1464 err = mlx5e_create_rq(c, param, rq);
1468 err = mlx5e_enable_rq(rq, param);
1470 goto err_destroy_rq;
1472 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
1474 goto err_disable_rq;
1481 mlx5e_disable_rq(rq);
1483 mlx5e_destroy_rq(rq);
1489 mlx5e_close_rq(struct mlx5e_rq *rq)
1493 callout_stop(&rq->watchdog);
1494 mtx_unlock(&rq->mtx);
1496 mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
1500 mlx5e_close_rq_wait(struct mlx5e_rq *rq)
1503 mlx5e_disable_rq(rq);
1504 mlx5e_close_cq(&rq->cq);
1505 cancel_work_sync(&rq->dim.work);
1506 mlx5e_destroy_rq(rq);
1510 mlx5e_free_sq_db(struct mlx5e_sq *sq)
1512 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1515 for (x = 0; x != wq_sz; x++) {
1516 if (unlikely(sq->mbuf[x].p_refcount != NULL)) {
1517 atomic_add_int(sq->mbuf[x].p_refcount, -1);
1518 sq->mbuf[x].p_refcount = NULL;
1520 if (sq->mbuf[x].mbuf != NULL) {
1521 bus_dmamap_unload(sq->dma_tag, sq->mbuf[x].dma_map);
1522 m_freem(sq->mbuf[x].mbuf);
1524 bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1526 free(sq->mbuf, M_MLX5EN);
1530 mlx5e_alloc_sq_db(struct mlx5e_sq *sq)
1532 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1536 sq->mbuf = malloc(wq_sz * sizeof(sq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
1538 /* Create DMA descriptor MAPs */
1539 for (x = 0; x != wq_sz; x++) {
1540 err = -bus_dmamap_create(sq->dma_tag, 0, &sq->mbuf[x].dma_map);
1543 bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1544 free(sq->mbuf, M_MLX5EN);
1551 static const char *mlx5e_sq_stats_desc[] = {
1552 MLX5E_SQ_STATS(MLX5E_STATS_DESC)
1556 mlx5e_update_sq_inline(struct mlx5e_sq *sq)
1558 sq->max_inline = sq->priv->params.tx_max_inline;
1559 sq->min_inline_mode = sq->priv->params.tx_min_inline_mode;
1562 * Check if trust state is DSCP or if inline mode is NONE which
1563 * indicates CX-5 or newer hardware.
1565 if (sq->priv->params_ethtool.trust_state != MLX5_QPTS_TRUST_PCP ||
1566 sq->min_inline_mode == MLX5_INLINE_MODE_NONE) {
1567 if (MLX5_CAP_ETH(sq->priv->mdev, wqe_vlan_insert))
1568 sq->min_insert_caps = MLX5E_INSERT_VLAN | MLX5E_INSERT_NON_VLAN;
1570 sq->min_insert_caps = MLX5E_INSERT_NON_VLAN;
1572 sq->min_insert_caps = 0;
1577 mlx5e_refresh_sq_inline_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
1581 for (i = 0; i != priv->num_tc; i++) {
1582 mtx_lock(&c->sq[i].lock);
1583 mlx5e_update_sq_inline(&c->sq[i]);
1584 mtx_unlock(&c->sq[i].lock);
1589 mlx5e_refresh_sq_inline(struct mlx5e_priv *priv)
1593 /* check if channels are closed */
1594 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
1597 for (i = 0; i < priv->params.num_channels; i++)
1598 mlx5e_refresh_sq_inline_sub(priv, &priv->channel[i]);
1602 mlx5e_create_sq(struct mlx5e_channel *c,
1604 struct mlx5e_sq_param *param,
1605 struct mlx5e_sq *sq)
1607 struct mlx5e_priv *priv = c->priv;
1608 struct mlx5_core_dev *mdev = priv->mdev;
1610 void *sqc = param->sqc;
1611 void *sqc_wq = MLX5_ADDR_OF(sqc, sqc, wq);
1614 /* Create DMA descriptor TAG */
1615 if ((err = -bus_dma_tag_create(
1616 bus_get_dma_tag(mdev->pdev->dev.bsddev),
1617 1, /* any alignment */
1618 0, /* no boundary */
1619 BUS_SPACE_MAXADDR, /* lowaddr */
1620 BUS_SPACE_MAXADDR, /* highaddr */
1621 NULL, NULL, /* filter, filterarg */
1622 MLX5E_MAX_TX_PAYLOAD_SIZE, /* maxsize */
1623 MLX5E_MAX_TX_MBUF_FRAGS, /* nsegments */
1624 MLX5E_MAX_TX_MBUF_SIZE, /* maxsegsize */
1626 NULL, NULL, /* lockfunc, lockfuncarg */
1630 err = mlx5_alloc_map_uar(mdev, &sq->uar);
1632 goto err_free_dma_tag;
1634 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, &sq->wq,
1637 goto err_unmap_free_uar;
1639 sq->wq.db = &sq->wq.db[MLX5_SND_DBR];
1640 sq->bf_buf_size = (1 << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2;
1642 err = mlx5e_alloc_sq_db(sq);
1644 goto err_sq_wq_destroy;
1646 sq->mkey_be = cpu_to_be32(priv->mr.key);
1647 sq->ifp = priv->ifp;
1651 mlx5e_update_sq_inline(sq);
1653 snprintf(buffer, sizeof(buffer), "txstat%dtc%d", c->ix, tc);
1654 mlx5e_create_stats(&sq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
1655 buffer, mlx5e_sq_stats_desc, MLX5E_SQ_STATS_NUM,
1661 mlx5_wq_destroy(&sq->wq_ctrl);
1664 mlx5_unmap_free_uar(mdev, &sq->uar);
1667 bus_dma_tag_destroy(sq->dma_tag);
1673 mlx5e_destroy_sq(struct mlx5e_sq *sq)
1675 /* destroy all sysctl nodes */
1676 sysctl_ctx_free(&sq->stats.ctx);
1678 mlx5e_free_sq_db(sq);
1679 mlx5_wq_destroy(&sq->wq_ctrl);
1680 mlx5_unmap_free_uar(sq->priv->mdev, &sq->uar);
1681 bus_dma_tag_destroy(sq->dma_tag);
1685 mlx5e_enable_sq(struct mlx5e_sq *sq, struct mlx5e_sq_param *param,
1694 inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
1695 sizeof(u64) * sq->wq_ctrl.buf.npages;
1696 in = mlx5_vzalloc(inlen);
1700 sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
1701 wq = MLX5_ADDR_OF(sqc, sqc, wq);
1703 memcpy(sqc, param->sqc, sizeof(param->sqc));
1705 MLX5_SET(sqc, sqc, tis_num_0, tis_num);
1706 MLX5_SET(sqc, sqc, cqn, sq->cq.mcq.cqn);
1707 MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
1708 MLX5_SET(sqc, sqc, tis_lst_sz, 1);
1709 MLX5_SET(sqc, sqc, flush_in_error_en, 1);
1711 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
1712 MLX5_SET(wq, wq, uar_page, sq->uar.index);
1713 MLX5_SET(wq, wq, log_wq_pg_sz, sq->wq_ctrl.buf.page_shift -
1715 MLX5_SET64(wq, wq, dbr_addr, sq->wq_ctrl.db.dma);
1717 mlx5_fill_page_array(&sq->wq_ctrl.buf,
1718 (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
1720 err = mlx5_core_create_sq(sq->priv->mdev, in, inlen, &sq->sqn);
1728 mlx5e_modify_sq(struct mlx5e_sq *sq, int curr_state, int next_state)
1735 inlen = MLX5_ST_SZ_BYTES(modify_sq_in);
1736 in = mlx5_vzalloc(inlen);
1740 sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx);
1742 MLX5_SET(modify_sq_in, in, sqn, sq->sqn);
1743 MLX5_SET(modify_sq_in, in, sq_state, curr_state);
1744 MLX5_SET(sqc, sqc, state, next_state);
1746 err = mlx5_core_modify_sq(sq->priv->mdev, in, inlen);
1754 mlx5e_disable_sq(struct mlx5e_sq *sq)
1757 mlx5_core_destroy_sq(sq->priv->mdev, sq->sqn);
1761 mlx5e_open_sq(struct mlx5e_channel *c,
1763 struct mlx5e_sq_param *param,
1764 struct mlx5e_sq *sq)
1768 sq->cev_factor = c->priv->params_ethtool.tx_completion_fact;
1770 /* ensure the TX completion event factor is not zero */
1771 if (sq->cev_factor == 0)
1774 err = mlx5e_create_sq(c, tc, param, sq);
1778 err = mlx5e_enable_sq(sq, param, c->priv->tisn[tc]);
1780 goto err_destroy_sq;
1782 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY);
1784 goto err_disable_sq;
1786 WRITE_ONCE(sq->running, 1);
1791 mlx5e_disable_sq(sq);
1793 mlx5e_destroy_sq(sq);
1799 mlx5e_sq_send_nops_locked(struct mlx5e_sq *sq, int can_sleep)
1801 /* fill up remainder with NOPs */
1802 while (sq->cev_counter != 0) {
1803 while (!mlx5e_sq_has_room_for(sq, 1)) {
1804 if (can_sleep != 0) {
1805 mtx_unlock(&sq->lock);
1807 mtx_lock(&sq->lock);
1812 /* send a single NOP */
1813 mlx5e_send_nop(sq, 1);
1814 atomic_thread_fence_rel();
1817 /* Check if we need to write the doorbell */
1818 if (likely(sq->doorbell.d64 != 0)) {
1819 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
1820 sq->doorbell.d64 = 0;
1825 mlx5e_sq_cev_timeout(void *arg)
1827 struct mlx5e_sq *sq = arg;
1829 mtx_assert(&sq->lock, MA_OWNED);
1831 /* check next state */
1832 switch (sq->cev_next_state) {
1833 case MLX5E_CEV_STATE_SEND_NOPS:
1834 /* fill TX ring with NOPs, if any */
1835 mlx5e_sq_send_nops_locked(sq, 0);
1837 /* check if completed */
1838 if (sq->cev_counter == 0) {
1839 sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
1844 /* send NOPs on next timeout */
1845 sq->cev_next_state = MLX5E_CEV_STATE_SEND_NOPS;
1850 callout_reset_curcpu(&sq->cev_callout, hz, mlx5e_sq_cev_timeout, sq);
1854 mlx5e_drain_sq(struct mlx5e_sq *sq)
1857 struct mlx5_core_dev *mdev= sq->priv->mdev;
1860 * Check if already stopped.
1862 * NOTE: Serialization of this function is managed by the
1863 * caller ensuring the priv's state lock is locked or in case
1864 * of rate limit support, a single thread manages drain and
1865 * resume of SQs. The "running" variable can therefore safely
1866 * be read without any locks.
1868 if (READ_ONCE(sq->running) == 0)
1871 /* don't put more packets into the SQ */
1872 WRITE_ONCE(sq->running, 0);
1874 /* serialize access to DMA rings */
1875 mtx_lock(&sq->lock);
1877 /* teardown event factor timer, if any */
1878 sq->cev_next_state = MLX5E_CEV_STATE_HOLD_NOPS;
1879 callout_stop(&sq->cev_callout);
1881 /* send dummy NOPs in order to flush the transmit ring */
1882 mlx5e_sq_send_nops_locked(sq, 1);
1883 mtx_unlock(&sq->lock);
1885 /* wait till SQ is empty or link is down */
1886 mtx_lock(&sq->lock);
1887 while (sq->cc != sq->pc &&
1888 (sq->priv->media_status_last & IFM_ACTIVE) != 0 &&
1889 mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1890 mtx_unlock(&sq->lock);
1892 sq->cq.mcq.comp(&sq->cq.mcq);
1893 mtx_lock(&sq->lock);
1895 mtx_unlock(&sq->lock);
1897 /* error out remaining requests */
1898 error = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RDY, MLX5_SQC_STATE_ERR);
1900 mlx5_en_err(sq->ifp,
1901 "mlx5e_modify_sq() from RDY to ERR failed: %d\n", error);
1904 /* wait till SQ is empty */
1905 mtx_lock(&sq->lock);
1906 while (sq->cc != sq->pc &&
1907 mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1908 mtx_unlock(&sq->lock);
1910 sq->cq.mcq.comp(&sq->cq.mcq);
1911 mtx_lock(&sq->lock);
1913 mtx_unlock(&sq->lock);
1917 mlx5e_close_sq_wait(struct mlx5e_sq *sq)
1921 mlx5e_disable_sq(sq);
1922 mlx5e_destroy_sq(sq);
1926 mlx5e_create_cq(struct mlx5e_priv *priv,
1927 struct mlx5e_cq_param *param,
1928 struct mlx5e_cq *cq,
1929 mlx5e_cq_comp_t *comp,
1932 struct mlx5_core_dev *mdev = priv->mdev;
1933 struct mlx5_core_cq *mcq = &cq->mcq;
1939 param->wq.buf_numa_node = 0;
1940 param->wq.db_numa_node = 0;
1942 err = mlx5_vector2eqn(mdev, eq_ix, &eqn_not_used, &irqn);
1946 err = mlx5_cqwq_create(mdev, ¶m->wq, param->cqc, &cq->wq,
1952 mcq->set_ci_db = cq->wq_ctrl.db.db;
1953 mcq->arm_db = cq->wq_ctrl.db.db + 1;
1954 *mcq->set_ci_db = 0;
1956 mcq->vector = eq_ix;
1958 mcq->event = mlx5e_cq_error_event;
1960 mcq->uar = &priv->cq_uar;
1962 for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) {
1963 struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i);
1974 mlx5e_destroy_cq(struct mlx5e_cq *cq)
1976 mlx5_wq_destroy(&cq->wq_ctrl);
1980 mlx5e_enable_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param, int eq_ix)
1982 struct mlx5_core_cq *mcq = &cq->mcq;
1990 inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
1991 sizeof(u64) * cq->wq_ctrl.buf.npages;
1992 in = mlx5_vzalloc(inlen);
1996 cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
1998 memcpy(cqc, param->cqc, sizeof(param->cqc));
2000 mlx5_fill_page_array(&cq->wq_ctrl.buf,
2001 (__be64 *) MLX5_ADDR_OF(create_cq_in, in, pas));
2003 mlx5_vector2eqn(cq->priv->mdev, eq_ix, &eqn, &irqn_not_used);
2005 MLX5_SET(cqc, cqc, c_eqn, eqn);
2006 MLX5_SET(cqc, cqc, uar_page, mcq->uar->index);
2007 MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.buf.page_shift -
2009 MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
2011 err = mlx5_core_create_cq(cq->priv->mdev, mcq, in, inlen);
2018 mlx5e_cq_arm(cq, MLX5_GET_DOORBELL_LOCK(&cq->priv->doorbell_lock));
2024 mlx5e_disable_cq(struct mlx5e_cq *cq)
2027 mlx5_core_destroy_cq(cq->priv->mdev, &cq->mcq);
2031 mlx5e_open_cq(struct mlx5e_priv *priv,
2032 struct mlx5e_cq_param *param,
2033 struct mlx5e_cq *cq,
2034 mlx5e_cq_comp_t *comp,
2039 err = mlx5e_create_cq(priv, param, cq, comp, eq_ix);
2043 err = mlx5e_enable_cq(cq, param, eq_ix);
2045 goto err_destroy_cq;
2050 mlx5e_destroy_cq(cq);
2056 mlx5e_close_cq(struct mlx5e_cq *cq)
2058 mlx5e_disable_cq(cq);
2059 mlx5e_destroy_cq(cq);
2063 mlx5e_open_tx_cqs(struct mlx5e_channel *c,
2064 struct mlx5e_channel_param *cparam)
2069 for (tc = 0; tc < c->priv->num_tc; tc++) {
2070 /* open completion queue */
2071 err = mlx5e_open_cq(c->priv, &cparam->tx_cq, &c->sq[tc].cq,
2072 &mlx5e_tx_cq_comp, c->ix);
2074 goto err_close_tx_cqs;
2079 for (tc--; tc >= 0; tc--)
2080 mlx5e_close_cq(&c->sq[tc].cq);
2086 mlx5e_close_tx_cqs(struct mlx5e_channel *c)
2090 for (tc = 0; tc < c->priv->num_tc; tc++)
2091 mlx5e_close_cq(&c->sq[tc].cq);
2095 mlx5e_open_sqs(struct mlx5e_channel *c,
2096 struct mlx5e_channel_param *cparam)
2101 for (tc = 0; tc < c->priv->num_tc; tc++) {
2102 err = mlx5e_open_sq(c, tc, &cparam->sq, &c->sq[tc]);
2110 for (tc--; tc >= 0; tc--)
2111 mlx5e_close_sq_wait(&c->sq[tc]);
2117 mlx5e_close_sqs_wait(struct mlx5e_channel *c)
2121 for (tc = 0; tc < c->priv->num_tc; tc++)
2122 mlx5e_close_sq_wait(&c->sq[tc]);
2126 mlx5e_chan_static_init(struct mlx5e_priv *priv, struct mlx5e_channel *c, int ix)
2130 /* setup priv and channel number */
2134 /* setup send tag */
2135 c->tag.type = IF_SND_TAG_TYPE_UNLIMITED;
2136 m_snd_tag_init(&c->tag.m_snd_tag, c->priv->ifp);
2138 init_completion(&c->completion);
2140 mtx_init(&c->rq.mtx, "mlx5rx", MTX_NETWORK_LOCK, MTX_DEF);
2142 callout_init_mtx(&c->rq.watchdog, &c->rq.mtx, 0);
2144 for (tc = 0; tc != MLX5E_MAX_TX_NUM_TC; tc++) {
2145 struct mlx5e_sq *sq = c->sq + tc;
2147 mtx_init(&sq->lock, "mlx5tx",
2148 MTX_NETWORK_LOCK " TX", MTX_DEF);
2149 mtx_init(&sq->comp_lock, "mlx5comp",
2150 MTX_NETWORK_LOCK " TX", MTX_DEF);
2152 callout_init_mtx(&sq->cev_callout, &sq->lock, 0);
2157 mlx5e_chan_wait_for_completion(struct mlx5e_channel *c)
2160 m_snd_tag_rele(&c->tag.m_snd_tag);
2161 wait_for_completion(&c->completion);
2165 mlx5e_priv_wait_for_completion(struct mlx5e_priv *priv, const uint32_t channels)
2169 for (x = 0; x != channels; x++)
2170 mlx5e_chan_wait_for_completion(&priv->channel[x]);
2174 mlx5e_chan_static_destroy(struct mlx5e_channel *c)
2178 callout_drain(&c->rq.watchdog);
2180 mtx_destroy(&c->rq.mtx);
2182 for (tc = 0; tc != MLX5E_MAX_TX_NUM_TC; tc++) {
2183 callout_drain(&c->sq[tc].cev_callout);
2184 mtx_destroy(&c->sq[tc].lock);
2185 mtx_destroy(&c->sq[tc].comp_lock);
2190 mlx5e_open_channel(struct mlx5e_priv *priv,
2191 struct mlx5e_channel_param *cparam,
2192 struct mlx5e_channel *c)
2194 struct epoch_tracker et;
2197 /* zero non-persistant data */
2198 MLX5E_ZERO(&c->rq, mlx5e_rq_zero_start);
2199 for (i = 0; i != priv->num_tc; i++)
2200 MLX5E_ZERO(&c->sq[i], mlx5e_sq_zero_start);
2202 /* open transmit completion queue */
2203 err = mlx5e_open_tx_cqs(c, cparam);
2207 /* open receive completion queue */
2208 err = mlx5e_open_cq(c->priv, &cparam->rx_cq, &c->rq.cq,
2209 &mlx5e_rx_cq_comp, c->ix);
2211 goto err_close_tx_cqs;
2213 err = mlx5e_open_sqs(c, cparam);
2215 goto err_close_rx_cq;
2217 err = mlx5e_open_rq(c, &cparam->rq, &c->rq);
2221 /* poll receive queue initially */
2222 NET_EPOCH_ENTER(et);
2223 c->rq.cq.mcq.comp(&c->rq.cq.mcq);
2229 mlx5e_close_sqs_wait(c);
2232 mlx5e_close_cq(&c->rq.cq);
2235 mlx5e_close_tx_cqs(c);
2242 mlx5e_close_channel(struct mlx5e_channel *c)
2244 mlx5e_close_rq(&c->rq);
2248 mlx5e_close_channel_wait(struct mlx5e_channel *c)
2250 mlx5e_close_rq_wait(&c->rq);
2251 mlx5e_close_sqs_wait(c);
2252 mlx5e_close_tx_cqs(c);
2256 mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs)
2260 r = priv->params.hw_lro_en ? priv->params.lro_wqe_sz :
2261 MLX5E_SW2MB_MTU(priv->ifp->if_mtu);
2262 if (r > MJUM16BYTES)
2267 else if (r > MJUMPAGESIZE)
2269 else if (r > MCLBYTES)
2275 * n + 1 must be a power of two, because stride size must be.
2276 * Stride size is 16 * (n + 1), as the first segment is
2279 for (n = howmany(r, MLX5E_MAX_RX_BYTES); !powerof2(n + 1); n++)
2282 if (n > MLX5E_MAX_BUSDMA_RX_SEGS)
2291 mlx5e_build_rq_param(struct mlx5e_priv *priv,
2292 struct mlx5e_rq_param *param)
2294 void *rqc = param->rqc;
2295 void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
2298 mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
2299 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
2300 MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN);
2301 MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe) +
2302 nsegs * sizeof(struct mlx5_wqe_data_seg)));
2303 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_rq_size);
2304 MLX5_SET(wq, wq, pd, priv->pdn);
2306 param->wq.buf_numa_node = 0;
2307 param->wq.db_numa_node = 0;
2308 param->wq.linear = 1;
2312 mlx5e_build_sq_param(struct mlx5e_priv *priv,
2313 struct mlx5e_sq_param *param)
2315 void *sqc = param->sqc;
2316 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
2318 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_sq_size);
2319 MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
2320 MLX5_SET(wq, wq, pd, priv->pdn);
2322 param->wq.buf_numa_node = 0;
2323 param->wq.db_numa_node = 0;
2324 param->wq.linear = 1;
2328 mlx5e_build_common_cq_param(struct mlx5e_priv *priv,
2329 struct mlx5e_cq_param *param)
2331 void *cqc = param->cqc;
2333 MLX5_SET(cqc, cqc, uar_page, priv->cq_uar.index);
2337 mlx5e_get_default_profile(struct mlx5e_priv *priv, int mode, struct net_dim_cq_moder *ptr)
2340 *ptr = net_dim_get_profile(mode, MLX5E_DIM_DEFAULT_PROFILE);
2342 /* apply LRO restrictions */
2343 if (priv->params.hw_lro_en &&
2344 ptr->pkts > MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO) {
2345 ptr->pkts = MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO;
2350 mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
2351 struct mlx5e_cq_param *param)
2353 struct net_dim_cq_moder curr;
2354 void *cqc = param->cqc;
2357 * We use MLX5_CQE_FORMAT_HASH because the RX hash mini CQE
2358 * format is more beneficial for FreeBSD use case.
2360 * Adding support for MLX5_CQE_FORMAT_CSUM will require changes
2361 * in mlx5e_decompress_cqe.
2363 if (priv->params.cqe_zipping_en) {
2364 MLX5_SET(cqc, cqc, mini_cqe_res_format, MLX5_CQE_FORMAT_HASH);
2365 MLX5_SET(cqc, cqc, cqe_compression_en, 1);
2368 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_rq_size);
2370 switch (priv->params.rx_cq_moderation_mode) {
2372 MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
2373 MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
2374 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2377 MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
2378 MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
2379 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
2380 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
2382 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2385 mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE, &curr);
2386 MLX5_SET(cqc, cqc, cq_period, curr.usec);
2387 MLX5_SET(cqc, cqc, cq_max_count, curr.pkts);
2388 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2391 mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE, &curr);
2392 MLX5_SET(cqc, cqc, cq_period, curr.usec);
2393 MLX5_SET(cqc, cqc, cq_max_count, curr.pkts);
2394 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
2395 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
2397 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2403 mlx5e_dim_build_cq_param(priv, param);
2405 mlx5e_build_common_cq_param(priv, param);
2409 mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
2410 struct mlx5e_cq_param *param)
2412 void *cqc = param->cqc;
2414 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_sq_size);
2415 MLX5_SET(cqc, cqc, cq_period, priv->params.tx_cq_moderation_usec);
2416 MLX5_SET(cqc, cqc, cq_max_count, priv->params.tx_cq_moderation_pkts);
2418 switch (priv->params.tx_cq_moderation_mode) {
2420 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2423 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
2424 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
2426 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2430 mlx5e_build_common_cq_param(priv, param);
2434 mlx5e_build_channel_param(struct mlx5e_priv *priv,
2435 struct mlx5e_channel_param *cparam)
2437 memset(cparam, 0, sizeof(*cparam));
2439 mlx5e_build_rq_param(priv, &cparam->rq);
2440 mlx5e_build_sq_param(priv, &cparam->sq);
2441 mlx5e_build_rx_cq_param(priv, &cparam->rx_cq);
2442 mlx5e_build_tx_cq_param(priv, &cparam->tx_cq);
2446 mlx5e_open_channels(struct mlx5e_priv *priv)
2448 struct mlx5e_channel_param *cparam;
2453 cparam = malloc(sizeof(*cparam), M_MLX5EN, M_WAITOK);
2455 mlx5e_build_channel_param(priv, cparam);
2456 for (i = 0; i < priv->params.num_channels; i++) {
2457 err = mlx5e_open_channel(priv, cparam, &priv->channel[i]);
2459 goto err_close_channels;
2462 for (j = 0; j < priv->params.num_channels; j++) {
2463 err = mlx5e_wait_for_min_rx_wqes(&priv->channel[j].rq);
2465 goto err_close_channels;
2467 free(cparam, M_MLX5EN);
2472 mlx5e_close_channel(&priv->channel[i]);
2473 mlx5e_close_channel_wait(&priv->channel[i]);
2475 free(cparam, M_MLX5EN);
2480 mlx5e_close_channels(struct mlx5e_priv *priv)
2484 for (i = 0; i < priv->params.num_channels; i++)
2485 mlx5e_close_channel(&priv->channel[i]);
2486 for (i = 0; i < priv->params.num_channels; i++)
2487 mlx5e_close_channel_wait(&priv->channel[i]);
2491 mlx5e_refresh_sq_params(struct mlx5e_priv *priv, struct mlx5e_sq *sq)
2494 if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
2497 switch (priv->params.tx_cq_moderation_mode) {
2500 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2503 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2507 return (mlx5_core_modify_cq_moderation_mode(priv->mdev, &sq->cq.mcq,
2508 priv->params.tx_cq_moderation_usec,
2509 priv->params.tx_cq_moderation_pkts,
2513 return (mlx5_core_modify_cq_moderation(priv->mdev, &sq->cq.mcq,
2514 priv->params.tx_cq_moderation_usec,
2515 priv->params.tx_cq_moderation_pkts));
2519 mlx5e_refresh_rq_params(struct mlx5e_priv *priv, struct mlx5e_rq *rq)
2522 if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
2527 switch (priv->params.rx_cq_moderation_mode) {
2530 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2531 dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
2534 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2535 dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
2539 /* tear down dynamic interrupt moderation */
2541 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
2542 mtx_unlock(&rq->mtx);
2544 /* wait for dynamic interrupt moderation work task, if any */
2545 cancel_work_sync(&rq->dim.work);
2547 if (priv->params.rx_cq_moderation_mode >= 2) {
2548 struct net_dim_cq_moder curr;
2550 mlx5e_get_default_profile(priv, dim_mode, &curr);
2552 retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2553 curr.usec, curr.pkts, cq_mode);
2555 /* set dynamic interrupt moderation mode and zero defaults */
2557 rq->dim.mode = dim_mode;
2559 rq->dim.profile_ix = MLX5E_DIM_DEFAULT_PROFILE;
2560 mtx_unlock(&rq->mtx);
2562 retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2563 priv->params.rx_cq_moderation_usec,
2564 priv->params.rx_cq_moderation_pkts,
2570 return (mlx5_core_modify_cq_moderation(priv->mdev, &rq->cq.mcq,
2571 priv->params.rx_cq_moderation_usec,
2572 priv->params.rx_cq_moderation_pkts));
2576 mlx5e_refresh_channel_params_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
2581 err = mlx5e_refresh_rq_params(priv, &c->rq);
2585 for (i = 0; i != priv->num_tc; i++) {
2586 err = mlx5e_refresh_sq_params(priv, &c->sq[i]);
2595 mlx5e_refresh_channel_params(struct mlx5e_priv *priv)
2599 /* check if channels are closed */
2600 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2603 for (i = 0; i < priv->params.num_channels; i++) {
2606 err = mlx5e_refresh_channel_params_sub(priv, &priv->channel[i]);
2614 mlx5e_open_tis(struct mlx5e_priv *priv, int tc)
2616 struct mlx5_core_dev *mdev = priv->mdev;
2617 u32 in[MLX5_ST_SZ_DW(create_tis_in)];
2618 void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
2620 memset(in, 0, sizeof(in));
2622 MLX5_SET(tisc, tisc, prio, tc);
2623 MLX5_SET(tisc, tisc, transport_domain, priv->tdn);
2625 return (mlx5_core_create_tis(mdev, in, sizeof(in), &priv->tisn[tc]));
2629 mlx5e_close_tis(struct mlx5e_priv *priv, int tc)
2631 mlx5_core_destroy_tis(priv->mdev, priv->tisn[tc]);
2635 mlx5e_open_tises(struct mlx5e_priv *priv)
2637 int num_tc = priv->num_tc;
2641 for (tc = 0; tc < num_tc; tc++) {
2642 err = mlx5e_open_tis(priv, tc);
2644 goto err_close_tises;
2650 for (tc--; tc >= 0; tc--)
2651 mlx5e_close_tis(priv, tc);
2657 mlx5e_close_tises(struct mlx5e_priv *priv)
2659 int num_tc = priv->num_tc;
2662 for (tc = 0; tc < num_tc; tc++)
2663 mlx5e_close_tis(priv, tc);
2667 mlx5e_open_rqt(struct mlx5e_priv *priv)
2669 struct mlx5_core_dev *mdev = priv->mdev;
2671 u32 out[MLX5_ST_SZ_DW(create_rqt_out)] = {0};
2678 sz = 1 << priv->params.rx_hash_log_tbl_sz;
2680 inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
2681 in = mlx5_vzalloc(inlen);
2684 rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
2686 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
2687 MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
2689 for (i = 0; i < sz; i++) {
2692 ix = rss_get_indirection_to_bucket(ix);
2694 /* ensure we don't overflow */
2695 ix %= priv->params.num_channels;
2697 /* apply receive side scaling stride, if any */
2698 ix -= ix % (int)priv->params.channels_rsss;
2700 MLX5_SET(rqtc, rqtc, rq_num[i], priv->channel[ix].rq.rqn);
2703 MLX5_SET(create_rqt_in, in, opcode, MLX5_CMD_OP_CREATE_RQT);
2705 err = mlx5_cmd_exec(mdev, in, inlen, out, sizeof(out));
2707 priv->rqtn = MLX5_GET(create_rqt_out, out, rqtn);
2715 mlx5e_close_rqt(struct mlx5e_priv *priv)
2717 u32 in[MLX5_ST_SZ_DW(destroy_rqt_in)] = {0};
2718 u32 out[MLX5_ST_SZ_DW(destroy_rqt_out)] = {0};
2720 MLX5_SET(destroy_rqt_in, in, opcode, MLX5_CMD_OP_DESTROY_RQT);
2721 MLX5_SET(destroy_rqt_in, in, rqtn, priv->rqtn);
2723 mlx5_cmd_exec(priv->mdev, in, sizeof(in), out, sizeof(out));
2727 mlx5e_build_tir_ctx(struct mlx5e_priv *priv, u32 * tirc, int tt)
2729 void *hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
2732 MLX5_SET(tirc, tirc, transport_domain, priv->tdn);
2734 #define ROUGH_MAX_L2_L3_HDR_SZ 256
2736 #define MLX5_HASH_IP (MLX5_HASH_FIELD_SEL_SRC_IP |\
2737 MLX5_HASH_FIELD_SEL_DST_IP)
2739 #define MLX5_HASH_ALL (MLX5_HASH_FIELD_SEL_SRC_IP |\
2740 MLX5_HASH_FIELD_SEL_DST_IP |\
2741 MLX5_HASH_FIELD_SEL_L4_SPORT |\
2742 MLX5_HASH_FIELD_SEL_L4_DPORT)
2744 #define MLX5_HASH_IP_IPSEC_SPI (MLX5_HASH_FIELD_SEL_SRC_IP |\
2745 MLX5_HASH_FIELD_SEL_DST_IP |\
2746 MLX5_HASH_FIELD_SEL_IPSEC_SPI)
2748 if (priv->params.hw_lro_en) {
2749 MLX5_SET(tirc, tirc, lro_enable_mask,
2750 MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO |
2751 MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO);
2752 MLX5_SET(tirc, tirc, lro_max_msg_sz,
2753 (priv->params.lro_wqe_sz -
2754 ROUGH_MAX_L2_L3_HDR_SZ) >> 8);
2755 /* TODO: add the option to choose timer value dynamically */
2756 MLX5_SET(tirc, tirc, lro_timeout_period_usecs,
2757 MLX5_CAP_ETH(priv->mdev,
2758 lro_timer_supported_periods[2]));
2761 /* setup parameters for hashing TIR type, if any */
2764 MLX5_SET(tirc, tirc, disp_type,
2765 MLX5_TIRC_DISP_TYPE_DIRECT);
2766 MLX5_SET(tirc, tirc, inline_rqn,
2767 priv->channel[0].rq.rqn);
2770 MLX5_SET(tirc, tirc, disp_type,
2771 MLX5_TIRC_DISP_TYPE_INDIRECT);
2772 MLX5_SET(tirc, tirc, indirect_table,
2774 MLX5_SET(tirc, tirc, rx_hash_fn,
2775 MLX5_TIRC_RX_HASH_FN_HASH_TOEPLITZ);
2776 hkey = (__be32 *) MLX5_ADDR_OF(tirc, tirc, rx_hash_toeplitz_key);
2779 * The FreeBSD RSS implementation does currently not
2780 * support symmetric Toeplitz hashes:
2782 MLX5_SET(tirc, tirc, rx_hash_symmetric, 0);
2783 rss_getkey((uint8_t *)hkey);
2785 MLX5_SET(tirc, tirc, rx_hash_symmetric, 1);
2786 hkey[0] = cpu_to_be32(0xD181C62C);
2787 hkey[1] = cpu_to_be32(0xF7F4DB5B);
2788 hkey[2] = cpu_to_be32(0x1983A2FC);
2789 hkey[3] = cpu_to_be32(0x943E1ADB);
2790 hkey[4] = cpu_to_be32(0xD9389E6B);
2791 hkey[5] = cpu_to_be32(0xD1039C2C);
2792 hkey[6] = cpu_to_be32(0xA74499AD);
2793 hkey[7] = cpu_to_be32(0x593D56D9);
2794 hkey[8] = cpu_to_be32(0xF3253C06);
2795 hkey[9] = cpu_to_be32(0x2ADC1FFC);
2801 case MLX5E_TT_IPV4_TCP:
2802 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2803 MLX5_L3_PROT_TYPE_IPV4);
2804 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2805 MLX5_L4_PROT_TYPE_TCP);
2807 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4)) {
2808 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2812 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2816 case MLX5E_TT_IPV6_TCP:
2817 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2818 MLX5_L3_PROT_TYPE_IPV6);
2819 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2820 MLX5_L4_PROT_TYPE_TCP);
2822 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) {
2823 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2827 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2831 case MLX5E_TT_IPV4_UDP:
2832 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2833 MLX5_L3_PROT_TYPE_IPV4);
2834 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2835 MLX5_L4_PROT_TYPE_UDP);
2837 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4)) {
2838 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2842 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2846 case MLX5E_TT_IPV6_UDP:
2847 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2848 MLX5_L3_PROT_TYPE_IPV6);
2849 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2850 MLX5_L4_PROT_TYPE_UDP);
2852 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) {
2853 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2857 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2861 case MLX5E_TT_IPV4_IPSEC_AH:
2862 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2863 MLX5_L3_PROT_TYPE_IPV4);
2864 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2865 MLX5_HASH_IP_IPSEC_SPI);
2868 case MLX5E_TT_IPV6_IPSEC_AH:
2869 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2870 MLX5_L3_PROT_TYPE_IPV6);
2871 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2872 MLX5_HASH_IP_IPSEC_SPI);
2875 case MLX5E_TT_IPV4_IPSEC_ESP:
2876 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2877 MLX5_L3_PROT_TYPE_IPV4);
2878 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2879 MLX5_HASH_IP_IPSEC_SPI);
2882 case MLX5E_TT_IPV6_IPSEC_ESP:
2883 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2884 MLX5_L3_PROT_TYPE_IPV6);
2885 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2886 MLX5_HASH_IP_IPSEC_SPI);
2890 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2891 MLX5_L3_PROT_TYPE_IPV4);
2892 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2897 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2898 MLX5_L3_PROT_TYPE_IPV6);
2899 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2909 mlx5e_open_tir(struct mlx5e_priv *priv, int tt)
2911 struct mlx5_core_dev *mdev = priv->mdev;
2917 inlen = MLX5_ST_SZ_BYTES(create_tir_in);
2918 in = mlx5_vzalloc(inlen);
2921 tirc = MLX5_ADDR_OF(create_tir_in, in, tir_context);
2923 mlx5e_build_tir_ctx(priv, tirc, tt);
2925 err = mlx5_core_create_tir(mdev, in, inlen, &priv->tirn[tt]);
2933 mlx5e_close_tir(struct mlx5e_priv *priv, int tt)
2935 mlx5_core_destroy_tir(priv->mdev, priv->tirn[tt]);
2939 mlx5e_open_tirs(struct mlx5e_priv *priv)
2944 for (i = 0; i < MLX5E_NUM_TT; i++) {
2945 err = mlx5e_open_tir(priv, i);
2947 goto err_close_tirs;
2953 for (i--; i >= 0; i--)
2954 mlx5e_close_tir(priv, i);
2960 mlx5e_close_tirs(struct mlx5e_priv *priv)
2964 for (i = 0; i < MLX5E_NUM_TT; i++)
2965 mlx5e_close_tir(priv, i);
2969 * SW MTU does not include headers,
2970 * HW MTU includes all headers and checksums.
2973 mlx5e_set_dev_port_mtu(struct ifnet *ifp, int sw_mtu)
2975 struct mlx5e_priv *priv = ifp->if_softc;
2976 struct mlx5_core_dev *mdev = priv->mdev;
2980 hw_mtu = MLX5E_SW2HW_MTU(sw_mtu);
2982 err = mlx5_set_port_mtu(mdev, hw_mtu);
2984 mlx5_en_err(ifp, "mlx5_set_port_mtu failed setting %d, err=%d\n",
2989 /* Update vport context MTU */
2990 err = mlx5_set_vport_mtu(mdev, hw_mtu);
2993 "Failed updating vport context with MTU size, err=%d\n",
2997 ifp->if_mtu = sw_mtu;
2999 err = mlx5_query_vport_mtu(mdev, &hw_mtu);
3000 if (err || !hw_mtu) {
3001 /* fallback to port oper mtu */
3002 err = mlx5_query_port_oper_mtu(mdev, &hw_mtu);
3006 "Query port MTU, after setting new MTU value, failed\n");
3008 } else if (MLX5E_HW2SW_MTU(hw_mtu) < sw_mtu) {
3011 "Port MTU %d is smaller than ifp mtu %d\n",
3013 } else if (MLX5E_HW2SW_MTU(hw_mtu) > sw_mtu) {
3016 "Port MTU %d is bigger than ifp mtu %d\n",
3019 priv->params_ethtool.hw_mtu = hw_mtu;
3022 while (hw_mtu & (hw_mtu - 1))
3023 hw_mtu &= (hw_mtu - 1);
3024 priv->params_ethtool.hw_mtu_msb = hw_mtu;
3030 mlx5e_open_locked(struct ifnet *ifp)
3032 struct mlx5e_priv *priv = ifp->if_softc;
3036 /* check if already opened */
3037 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
3041 if (rss_getnumbuckets() > priv->params.num_channels) {
3043 "NOTE: There are more RSS buckets(%u) than channels(%u) available\n",
3044 rss_getnumbuckets(), priv->params.num_channels);
3047 err = mlx5e_open_tises(priv);
3049 mlx5_en_err(ifp, "mlx5e_open_tises failed, %d\n", err);
3052 err = mlx5_vport_alloc_q_counter(priv->mdev,
3053 MLX5_INTERFACE_PROTOCOL_ETH, &set_id);
3055 mlx5_en_err(priv->ifp,
3056 "mlx5_vport_alloc_q_counter failed: %d\n", err);
3057 goto err_close_tises;
3059 /* store counter set ID */
3060 priv->counter_set_id = set_id;
3062 err = mlx5e_open_channels(priv);
3065 "mlx5e_open_channels failed, %d\n", err);
3066 goto err_dalloc_q_counter;
3068 err = mlx5e_open_rqt(priv);
3070 mlx5_en_err(ifp, "mlx5e_open_rqt failed, %d\n", err);
3071 goto err_close_channels;
3073 err = mlx5e_open_tirs(priv);
3075 mlx5_en_err(ifp, "mlx5e_open_tir failed, %d\n", err);
3076 goto err_close_rqls;
3078 err = mlx5e_open_flow_table(priv);
3081 "mlx5e_open_flow_table failed, %d\n", err);
3082 goto err_close_tirs;
3084 err = mlx5e_add_all_vlan_rules(priv);
3087 "mlx5e_add_all_vlan_rules failed, %d\n", err);
3088 goto err_close_flow_table;
3090 set_bit(MLX5E_STATE_OPENED, &priv->state);
3092 mlx5e_update_carrier(priv);
3093 mlx5e_set_rx_mode_core(priv);
3097 err_close_flow_table:
3098 mlx5e_close_flow_table(priv);
3101 mlx5e_close_tirs(priv);
3104 mlx5e_close_rqt(priv);
3107 mlx5e_close_channels(priv);
3109 err_dalloc_q_counter:
3110 mlx5_vport_dealloc_q_counter(priv->mdev,
3111 MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
3114 mlx5e_close_tises(priv);
3120 mlx5e_open(void *arg)
3122 struct mlx5e_priv *priv = arg;
3125 if (mlx5_set_port_status(priv->mdev, MLX5_PORT_UP))
3126 mlx5_en_err(priv->ifp,
3127 "Setting port status to up failed\n");
3129 mlx5e_open_locked(priv->ifp);
3130 priv->ifp->if_drv_flags |= IFF_DRV_RUNNING;
3135 mlx5e_close_locked(struct ifnet *ifp)
3137 struct mlx5e_priv *priv = ifp->if_softc;
3139 /* check if already closed */
3140 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3143 clear_bit(MLX5E_STATE_OPENED, &priv->state);
3145 mlx5e_set_rx_mode_core(priv);
3146 mlx5e_del_all_vlan_rules(priv);
3147 if_link_state_change(priv->ifp, LINK_STATE_DOWN);
3148 mlx5e_close_flow_table(priv);
3149 mlx5e_close_tirs(priv);
3150 mlx5e_close_rqt(priv);
3151 mlx5e_close_channels(priv);
3152 mlx5_vport_dealloc_q_counter(priv->mdev,
3153 MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
3154 mlx5e_close_tises(priv);
3159 #if (__FreeBSD_version >= 1100000)
3161 mlx5e_get_counter(struct ifnet *ifp, ift_counter cnt)
3163 struct mlx5e_priv *priv = ifp->if_softc;
3166 /* PRIV_LOCK(priv); XXX not allowed */
3168 case IFCOUNTER_IPACKETS:
3169 retval = priv->stats.vport.rx_packets;
3171 case IFCOUNTER_IERRORS:
3172 retval = priv->stats.pport.in_range_len_errors +
3173 priv->stats.pport.out_of_range_len +
3174 priv->stats.pport.too_long_errors +
3175 priv->stats.pport.check_seq_err +
3176 priv->stats.pport.alignment_err;
3178 case IFCOUNTER_IQDROPS:
3179 retval = priv->stats.vport.rx_out_of_buffer;
3181 case IFCOUNTER_OPACKETS:
3182 retval = priv->stats.vport.tx_packets;
3184 case IFCOUNTER_OERRORS:
3185 retval = priv->stats.port_stats_debug.out_discards;
3187 case IFCOUNTER_IBYTES:
3188 retval = priv->stats.vport.rx_bytes;
3190 case IFCOUNTER_OBYTES:
3191 retval = priv->stats.vport.tx_bytes;
3193 case IFCOUNTER_IMCASTS:
3194 retval = priv->stats.vport.rx_multicast_packets;
3196 case IFCOUNTER_OMCASTS:
3197 retval = priv->stats.vport.tx_multicast_packets;
3199 case IFCOUNTER_OQDROPS:
3200 retval = priv->stats.vport.tx_queue_dropped;
3202 case IFCOUNTER_COLLISIONS:
3203 retval = priv->stats.pport.collisions;
3206 retval = if_get_counter_default(ifp, cnt);
3209 /* PRIV_UNLOCK(priv); XXX not allowed */
3215 mlx5e_set_rx_mode(struct ifnet *ifp)
3217 struct mlx5e_priv *priv = ifp->if_softc;
3219 queue_work(priv->wq, &priv->set_rx_mode_work);
3223 mlx5e_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3225 struct mlx5e_priv *priv;
3227 struct ifi2creq i2c;
3236 priv = ifp->if_softc;
3238 /* check if detaching */
3239 if (priv == NULL || priv->gone != 0)
3244 ifr = (struct ifreq *)data;
3247 mlx5_query_port_max_mtu(priv->mdev, &max_mtu);
3249 if (ifr->ifr_mtu >= MLX5E_MTU_MIN &&
3250 ifr->ifr_mtu <= MIN(MLX5E_MTU_MAX, max_mtu)) {
3253 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
3255 mlx5e_close_locked(ifp);
3258 mlx5e_set_dev_port_mtu(ifp, ifr->ifr_mtu);
3261 mlx5e_open_locked(ifp);
3265 "Invalid MTU value. Min val: %d, Max val: %d\n",
3266 MLX5E_MTU_MIN, MIN(MLX5E_MTU_MAX, max_mtu));
3271 if ((ifp->if_flags & IFF_UP) &&
3272 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
3273 mlx5e_set_rx_mode(ifp);
3277 if (ifp->if_flags & IFF_UP) {
3278 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3279 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3280 mlx5e_open_locked(ifp);
3281 ifp->if_drv_flags |= IFF_DRV_RUNNING;
3282 mlx5_set_port_status(priv->mdev, MLX5_PORT_UP);
3285 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
3286 mlx5_set_port_status(priv->mdev,
3288 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
3289 mlx5e_close_locked(ifp);
3290 mlx5e_update_carrier(priv);
3291 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3298 mlx5e_set_rx_mode(ifp);
3303 ifr = (struct ifreq *)data;
3304 error = ifmedia_ioctl(ifp, ifr, &priv->media, command);
3307 ifr = (struct ifreq *)data;
3309 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3311 if (mask & IFCAP_TXCSUM) {
3312 ifp->if_capenable ^= IFCAP_TXCSUM;
3313 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
3315 if (IFCAP_TSO4 & ifp->if_capenable &&
3316 !(IFCAP_TXCSUM & ifp->if_capenable)) {
3317 ifp->if_capenable &= ~IFCAP_TSO4;
3318 ifp->if_hwassist &= ~CSUM_IP_TSO;
3320 "tso4 disabled due to -txcsum.\n");
3323 if (mask & IFCAP_TXCSUM_IPV6) {
3324 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
3325 ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
3327 if (IFCAP_TSO6 & ifp->if_capenable &&
3328 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
3329 ifp->if_capenable &= ~IFCAP_TSO6;
3330 ifp->if_hwassist &= ~CSUM_IP6_TSO;
3332 "tso6 disabled due to -txcsum6.\n");
3335 if (mask & IFCAP_NOMAP)
3336 ifp->if_capenable ^= IFCAP_NOMAP;
3337 if (mask & IFCAP_TXTLS4)
3338 ifp->if_capenable ^= IFCAP_TXTLS4;
3339 if (mask & IFCAP_TXTLS6)
3340 ifp->if_capenable ^= IFCAP_TXTLS6;
3341 if (mask & IFCAP_RXCSUM)
3342 ifp->if_capenable ^= IFCAP_RXCSUM;
3343 if (mask & IFCAP_RXCSUM_IPV6)
3344 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
3345 if (mask & IFCAP_TSO4) {
3346 if (!(IFCAP_TSO4 & ifp->if_capenable) &&
3347 !(IFCAP_TXCSUM & ifp->if_capenable)) {
3348 mlx5_en_err(ifp, "enable txcsum first.\n");
3352 ifp->if_capenable ^= IFCAP_TSO4;
3353 ifp->if_hwassist ^= CSUM_IP_TSO;
3355 if (mask & IFCAP_TSO6) {
3356 if (!(IFCAP_TSO6 & ifp->if_capenable) &&
3357 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
3358 mlx5_en_err(ifp, "enable txcsum6 first.\n");
3362 ifp->if_capenable ^= IFCAP_TSO6;
3363 ifp->if_hwassist ^= CSUM_IP6_TSO;
3365 if (mask & IFCAP_VLAN_HWFILTER) {
3366 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
3367 mlx5e_disable_vlan_filter(priv);
3369 mlx5e_enable_vlan_filter(priv);
3371 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
3373 if (mask & IFCAP_VLAN_HWTAGGING)
3374 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3375 if (mask & IFCAP_WOL_MAGIC)
3376 ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3378 VLAN_CAPABILITIES(ifp);
3379 /* turn off LRO means also turn of HW LRO - if it's on */
3380 if (mask & IFCAP_LRO) {
3381 int was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
3382 bool need_restart = false;
3384 ifp->if_capenable ^= IFCAP_LRO;
3386 /* figure out if updating HW LRO is needed */
3387 if (!(ifp->if_capenable & IFCAP_LRO)) {
3388 if (priv->params.hw_lro_en) {
3389 priv->params.hw_lro_en = false;
3390 need_restart = true;
3393 if (priv->params.hw_lro_en == false &&
3394 priv->params_ethtool.hw_lro != 0) {
3395 priv->params.hw_lro_en = true;
3396 need_restart = true;
3399 if (was_opened && need_restart) {
3400 mlx5e_close_locked(ifp);
3401 mlx5e_open_locked(ifp);
3404 if (mask & IFCAP_HWRXTSTMP) {
3405 ifp->if_capenable ^= IFCAP_HWRXTSTMP;
3406 if (ifp->if_capenable & IFCAP_HWRXTSTMP) {
3407 if (priv->clbr_done == 0)
3408 mlx5e_reset_calibration_callout(priv);
3410 callout_drain(&priv->tstmp_clbr);
3411 priv->clbr_done = 0;
3419 ifr = (struct ifreq *)data;
3422 * Copy from the user-space address ifr_data to the
3423 * kernel-space address i2c
3425 error = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
3429 if (i2c.len > sizeof(i2c.data)) {
3435 /* Get module_num which is required for the query_eeprom */
3436 error = mlx5_query_module_num(priv->mdev, &module_num);
3439 "Query module num failed, eeprom reading is not supported\n");
3443 /* Check if module is present before doing an access */
3444 module_status = mlx5_query_module_status(priv->mdev, module_num);
3445 if (module_status != MLX5_MODULE_STATUS_PLUGGED_ENABLED) {
3450 * Currently 0XA0 and 0xA2 are the only addresses permitted.
3451 * The internal conversion is as follows:
3453 if (i2c.dev_addr == 0xA0)
3454 read_addr = MLX5_I2C_ADDR_LOW;
3455 else if (i2c.dev_addr == 0xA2)
3456 read_addr = MLX5_I2C_ADDR_HIGH;
3459 "Query eeprom failed, Invalid Address: %X\n",
3464 error = mlx5_query_eeprom(priv->mdev,
3465 read_addr, MLX5_EEPROM_LOW_PAGE,
3466 (uint32_t)i2c.offset, (uint32_t)i2c.len, module_num,
3467 (uint32_t *)i2c.data, &size_read);
3470 "Query eeprom failed, eeprom reading is not supported\n");
3475 if (i2c.len > MLX5_EEPROM_MAX_BYTES) {
3476 error = mlx5_query_eeprom(priv->mdev,
3477 read_addr, MLX5_EEPROM_LOW_PAGE,
3478 (uint32_t)(i2c.offset + size_read),
3479 (uint32_t)(i2c.len - size_read), module_num,
3480 (uint32_t *)(i2c.data + size_read), &size_read);
3484 "Query eeprom failed, eeprom reading is not supported\n");
3489 error = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
3495 error = ether_ioctl(ifp, command, data);
3502 mlx5e_check_required_hca_cap(struct mlx5_core_dev *mdev)
3505 * TODO: uncoment once FW really sets all these bits if
3506 * (!mdev->caps.eth.rss_ind_tbl_cap || !mdev->caps.eth.csum_cap ||
3507 * !mdev->caps.eth.max_lso_cap || !mdev->caps.eth.vlan_cap ||
3508 * !(mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_SCQE_BRK_MOD)) return
3512 /* TODO: add more must-to-have features */
3514 if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
3521 mlx5e_get_max_inline_cap(struct mlx5_core_dev *mdev)
3523 const int min_size = ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN;
3524 const int max_size = MLX5E_MAX_TX_INLINE;
3525 const int bf_buf_size =
3526 ((1U << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2U) -
3527 (sizeof(struct mlx5e_tx_wqe) - 2);
3529 /* verify against driver limits */
3530 if (bf_buf_size > max_size)
3532 else if (bf_buf_size < min_size)
3535 return (bf_buf_size);
3539 mlx5e_build_ifp_priv(struct mlx5_core_dev *mdev,
3540 struct mlx5e_priv *priv,
3541 int num_comp_vectors)
3546 * TODO: Consider link speed for setting "log_sq_size",
3547 * "log_rq_size" and "cq_moderation_xxx":
3549 priv->params.log_sq_size =
3550 MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
3551 priv->params.log_rq_size =
3552 MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
3553 priv->params.rx_cq_moderation_usec =
3554 MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
3555 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE :
3556 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
3557 priv->params.rx_cq_moderation_mode =
3558 MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ? 1 : 0;
3559 priv->params.rx_cq_moderation_pkts =
3560 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS;
3561 priv->params.tx_cq_moderation_usec =
3562 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC;
3563 priv->params.tx_cq_moderation_pkts =
3564 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
3565 priv->params.min_rx_wqes =
3566 MLX5E_PARAMS_DEFAULT_MIN_RX_WQES;
3567 priv->params.rx_hash_log_tbl_sz =
3568 (order_base_2(num_comp_vectors) >
3569 MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ) ?
3570 order_base_2(num_comp_vectors) :
3571 MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ;
3572 priv->params.num_tc = 1;
3573 priv->params.default_vlan_prio = 0;
3574 priv->counter_set_id = -1;
3575 priv->params.tx_max_inline = mlx5e_get_max_inline_cap(mdev);
3577 err = mlx5_query_min_inline(mdev, &priv->params.tx_min_inline_mode);
3582 * hw lro is currently defaulted to off. when it won't anymore we
3583 * will consider the HW capability: "!!MLX5_CAP_ETH(mdev, lro_cap)"
3585 priv->params.hw_lro_en = false;
3586 priv->params.lro_wqe_sz = MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ;
3589 * CQE zipping is currently defaulted to off. when it won't
3590 * anymore we will consider the HW capability:
3591 * "!!MLX5_CAP_GEN(mdev, cqe_compression)"
3593 priv->params.cqe_zipping_en = false;
3596 priv->params.num_channels = num_comp_vectors;
3597 priv->params.channels_rsss = 1;
3598 priv->order_base_2_num_channels = order_base_2(num_comp_vectors);
3599 priv->queue_mapping_channel_mask =
3600 roundup_pow_of_two(num_comp_vectors) - 1;
3601 priv->num_tc = priv->params.num_tc;
3602 priv->default_vlan_prio = priv->params.default_vlan_prio;
3604 INIT_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
3605 INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
3606 INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
3612 mlx5e_create_mkey(struct mlx5e_priv *priv, u32 pdn,
3613 struct mlx5_core_mr *mkey)
3615 struct ifnet *ifp = priv->ifp;
3616 struct mlx5_core_dev *mdev = priv->mdev;
3617 int inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
3622 in = mlx5_vzalloc(inlen);
3624 mlx5_en_err(ifp, "failed to allocate inbox\n");
3628 mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
3629 MLX5_SET(mkc, mkc, access_mode, MLX5_ACCESS_MODE_PA);
3630 MLX5_SET(mkc, mkc, umr_en, 1); /* used by HW TLS */
3631 MLX5_SET(mkc, mkc, lw, 1);
3632 MLX5_SET(mkc, mkc, lr, 1);
3634 MLX5_SET(mkc, mkc, pd, pdn);
3635 MLX5_SET(mkc, mkc, length64, 1);
3636 MLX5_SET(mkc, mkc, qpn, 0xffffff);
3638 err = mlx5_core_create_mkey(mdev, mkey, in, inlen);
3640 mlx5_en_err(ifp, "mlx5_core_create_mkey failed, %d\n",
3647 static const char *mlx5e_vport_stats_desc[] = {
3648 MLX5E_VPORT_STATS(MLX5E_STATS_DESC)
3651 static const char *mlx5e_pport_stats_desc[] = {
3652 MLX5E_PPORT_STATS(MLX5E_STATS_DESC)
3656 mlx5e_priv_static_init(struct mlx5e_priv *priv, const uint32_t channels)
3660 mtx_init(&priv->async_events_mtx, "mlx5async", MTX_NETWORK_LOCK, MTX_DEF);
3661 sx_init(&priv->state_lock, "mlx5state");
3662 callout_init_mtx(&priv->watchdog, &priv->async_events_mtx, 0);
3663 MLX5_INIT_DOORBELL_LOCK(&priv->doorbell_lock);
3664 for (x = 0; x != channels; x++)
3665 mlx5e_chan_static_init(priv, &priv->channel[x], x);
3669 mlx5e_priv_static_destroy(struct mlx5e_priv *priv, const uint32_t channels)
3673 for (x = 0; x != channels; x++)
3674 mlx5e_chan_static_destroy(&priv->channel[x]);
3675 callout_drain(&priv->watchdog);
3676 mtx_destroy(&priv->async_events_mtx);
3677 sx_destroy(&priv->state_lock);
3681 sysctl_firmware(SYSCTL_HANDLER_ARGS)
3684 * %d.%d%.d the string format.
3685 * fw_rev_{maj,min,sub} return u16, 2^16 = 65536.
3686 * We need at most 5 chars to store that.
3687 * It also has: two "." and NULL at the end, which means we need 18
3688 * (5*3 + 3) chars at most.
3691 struct mlx5e_priv *priv = arg1;
3694 snprintf(fw, sizeof(fw), "%d.%d.%d", fw_rev_maj(priv->mdev), fw_rev_min(priv->mdev),
3695 fw_rev_sub(priv->mdev));
3696 error = sysctl_handle_string(oidp, fw, sizeof(fw), req);
3701 mlx5e_disable_tx_dma(struct mlx5e_channel *ch)
3705 for (i = 0; i < ch->priv->num_tc; i++)
3706 mlx5e_drain_sq(&ch->sq[i]);
3710 mlx5e_reset_sq_doorbell_record(struct mlx5e_sq *sq)
3713 sq->doorbell.d32[0] = cpu_to_be32(MLX5_OPCODE_NOP);
3714 sq->doorbell.d32[1] = cpu_to_be32(sq->sqn << 8);
3715 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
3716 sq->doorbell.d64 = 0;
3720 mlx5e_resume_sq(struct mlx5e_sq *sq)
3724 /* check if already enabled */
3725 if (READ_ONCE(sq->running) != 0)
3728 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_ERR,
3729 MLX5_SQC_STATE_RST);
3731 mlx5_en_err(sq->ifp,
3732 "mlx5e_modify_sq() from ERR to RST failed: %d\n", err);
3738 /* reset doorbell prior to moving from RST to RDY */
3739 mlx5e_reset_sq_doorbell_record(sq);
3741 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST,
3742 MLX5_SQC_STATE_RDY);
3744 mlx5_en_err(sq->ifp,
3745 "mlx5e_modify_sq() from RST to RDY failed: %d\n", err);
3748 sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
3749 WRITE_ONCE(sq->running, 1);
3753 mlx5e_enable_tx_dma(struct mlx5e_channel *ch)
3757 for (i = 0; i < ch->priv->num_tc; i++)
3758 mlx5e_resume_sq(&ch->sq[i]);
3762 mlx5e_disable_rx_dma(struct mlx5e_channel *ch)
3764 struct mlx5e_rq *rq = &ch->rq;
3765 struct epoch_tracker et;
3770 callout_stop(&rq->watchdog);
3771 mtx_unlock(&rq->mtx);
3773 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
3775 mlx5_en_err(rq->ifp,
3776 "mlx5e_modify_rq() from RDY to RST failed: %d\n", err);
3779 while (!mlx5_wq_ll_is_empty(&rq->wq)) {
3781 NET_EPOCH_ENTER(et);
3782 rq->cq.mcq.comp(&rq->cq.mcq);
3787 * Transitioning into RST state will allow the FW to track less ERR state queues,
3788 * thus reducing the recv queue flushing time
3790 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_ERR, MLX5_RQC_STATE_RST);
3792 mlx5_en_err(rq->ifp,
3793 "mlx5e_modify_rq() from ERR to RST failed: %d\n", err);
3798 mlx5e_enable_rx_dma(struct mlx5e_channel *ch)
3800 struct mlx5e_rq *rq = &ch->rq;
3801 struct epoch_tracker et;
3805 mlx5_wq_ll_update_db_record(&rq->wq);
3806 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
3808 mlx5_en_err(rq->ifp,
3809 "mlx5e_modify_rq() from RST to RDY failed: %d\n", err);
3814 NET_EPOCH_ENTER(et);
3815 rq->cq.mcq.comp(&rq->cq.mcq);
3820 mlx5e_modify_tx_dma(struct mlx5e_priv *priv, uint8_t value)
3824 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3827 for (i = 0; i < priv->params.num_channels; i++) {
3829 mlx5e_disable_tx_dma(&priv->channel[i]);
3831 mlx5e_enable_tx_dma(&priv->channel[i]);
3836 mlx5e_modify_rx_dma(struct mlx5e_priv *priv, uint8_t value)
3840 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3843 for (i = 0; i < priv->params.num_channels; i++) {
3845 mlx5e_disable_rx_dma(&priv->channel[i]);
3847 mlx5e_enable_rx_dma(&priv->channel[i]);
3852 mlx5e_add_hw_stats(struct mlx5e_priv *priv)
3854 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3855 OID_AUTO, "fw_version", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
3856 priv, 0, sysctl_firmware, "A", "HCA firmware version");
3858 SYSCTL_ADD_STRING(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3859 OID_AUTO, "board_id", CTLFLAG_RD, priv->mdev->board_id, 0,
3864 mlx5e_sysctl_tx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3866 struct mlx5e_priv *priv = arg1;
3867 uint8_t temp[MLX5E_MAX_PRIORITY];
3874 tx_pfc = priv->params.tx_priority_flow_control;
3876 for (i = 0; i != MLX5E_MAX_PRIORITY; i++)
3877 temp[i] = (tx_pfc >> i) & 1;
3879 err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY);
3880 if (err || !req->newptr)
3882 err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY);
3886 priv->params.tx_priority_flow_control = 0;
3888 /* range check input value */
3889 for (i = 0; i != MLX5E_MAX_PRIORITY; i++) {
3894 priv->params.tx_priority_flow_control |= (temp[i] << i);
3897 /* check if update is required */
3898 if (tx_pfc != priv->params.tx_priority_flow_control)
3899 err = -mlx5e_set_port_pfc(priv);
3902 priv->params.tx_priority_flow_control= tx_pfc;
3909 mlx5e_sysctl_rx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3911 struct mlx5e_priv *priv = arg1;
3912 uint8_t temp[MLX5E_MAX_PRIORITY];
3919 rx_pfc = priv->params.rx_priority_flow_control;
3921 for (i = 0; i != MLX5E_MAX_PRIORITY; i++)
3922 temp[i] = (rx_pfc >> i) & 1;
3924 err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY);
3925 if (err || !req->newptr)
3927 err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY);
3931 priv->params.rx_priority_flow_control = 0;
3933 /* range check input value */
3934 for (i = 0; i != MLX5E_MAX_PRIORITY; i++) {
3939 priv->params.rx_priority_flow_control |= (temp[i] << i);
3942 /* check if update is required */
3943 if (rx_pfc != priv->params.rx_priority_flow_control) {
3944 err = -mlx5e_set_port_pfc(priv);
3945 if (err == 0 && priv->sw_is_port_buf_owner)
3946 err = mlx5e_update_buf_lossy(priv);
3950 priv->params.rx_priority_flow_control= rx_pfc;
3957 mlx5e_setup_pauseframes(struct mlx5e_priv *priv)
3959 #if (__FreeBSD_version < 1100000)
3964 /* enable pauseframes by default */
3965 priv->params.tx_pauseframe_control = 1;
3966 priv->params.rx_pauseframe_control = 1;
3968 /* disable ports flow control, PFC, by default */
3969 priv->params.tx_priority_flow_control = 0;
3970 priv->params.rx_priority_flow_control = 0;
3972 #if (__FreeBSD_version < 1100000)
3973 /* compute path for sysctl */
3974 snprintf(path, sizeof(path), "dev.mce.%d.tx_pauseframe_control",
3975 device_get_unit(priv->mdev->pdev->dev.bsddev));
3977 /* try to fetch tunable, if any */
3978 TUNABLE_INT_FETCH(path, &priv->params.tx_pauseframe_control);
3980 /* compute path for sysctl */
3981 snprintf(path, sizeof(path), "dev.mce.%d.rx_pauseframe_control",
3982 device_get_unit(priv->mdev->pdev->dev.bsddev));
3984 /* try to fetch tunable, if any */
3985 TUNABLE_INT_FETCH(path, &priv->params.rx_pauseframe_control);
3988 /* register pauseframe SYSCTLs */
3989 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3990 OID_AUTO, "tx_pauseframe_control", CTLFLAG_RDTUN,
3991 &priv->params.tx_pauseframe_control, 0,
3992 "Set to enable TX pause frames. Clear to disable.");
3994 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3995 OID_AUTO, "rx_pauseframe_control", CTLFLAG_RDTUN,
3996 &priv->params.rx_pauseframe_control, 0,
3997 "Set to enable RX pause frames. Clear to disable.");
3999 /* register priority flow control, PFC, SYSCTLs */
4000 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
4001 OID_AUTO, "tx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN |
4002 CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_tx_priority_flow_control, "CU",
4003 "Set to enable TX ports flow control frames for priorities 0..7. Clear to disable.");
4005 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
4006 OID_AUTO, "rx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN |
4007 CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_rx_priority_flow_control, "CU",
4008 "Set to enable RX ports flow control frames for priorities 0..7. Clear to disable.");
4013 priv->params.tx_pauseframe_control =
4014 priv->params.tx_pauseframe_control ? 1 : 0;
4015 priv->params.rx_pauseframe_control =
4016 priv->params.rx_pauseframe_control ? 1 : 0;
4018 /* update firmware */
4019 error = mlx5e_set_port_pause_and_pfc(priv);
4020 if (error == -EINVAL) {
4021 mlx5_en_err(priv->ifp,
4022 "Global pauseframes must be disabled before enabling PFC.\n");
4023 priv->params.rx_priority_flow_control = 0;
4024 priv->params.tx_priority_flow_control = 0;
4026 /* update firmware */
4027 (void) mlx5e_set_port_pause_and_pfc(priv);
4033 mlx5e_ul_snd_tag_alloc(struct ifnet *ifp,
4034 union if_snd_tag_alloc_params *params,
4035 struct m_snd_tag **ppmt)
4037 struct mlx5e_priv *priv;
4038 struct mlx5e_channel *pch;
4040 priv = ifp->if_softc;
4042 if (unlikely(priv->gone || params->hdr.flowtype == M_HASHTYPE_NONE)) {
4043 return (EOPNOTSUPP);
4045 /* keep this code synced with mlx5e_select_queue() */
4046 u32 ch = priv->params.num_channels;
4050 if (rss_hash2bucket(params->hdr.flowid,
4051 params->hdr.flowtype, &temp) == 0)
4055 ch = (params->hdr.flowid % 128) % ch;
4058 * NOTE: The channels array is only freed at detach
4059 * and it safe to return a pointer to the send tag
4060 * inside the channels structure as long as we
4061 * reference the priv.
4063 pch = priv->channel + ch;
4065 /* check if send queue is not running */
4066 if (unlikely(pch->sq[0].running == 0))
4068 m_snd_tag_ref(&pch->tag.m_snd_tag);
4069 *ppmt = &pch->tag.m_snd_tag;
4075 mlx5e_ul_snd_tag_query(struct m_snd_tag *pmt, union if_snd_tag_query_params *params)
4077 struct mlx5e_channel *pch =
4078 container_of(pmt, struct mlx5e_channel, tag.m_snd_tag);
4080 params->unlimited.max_rate = -1ULL;
4081 params->unlimited.queue_level = mlx5e_sq_queue_level(&pch->sq[0]);
4086 mlx5e_ul_snd_tag_free(struct m_snd_tag *pmt)
4088 struct mlx5e_channel *pch =
4089 container_of(pmt, struct mlx5e_channel, tag.m_snd_tag);
4091 complete(&pch->completion);
4095 mlx5e_snd_tag_alloc(struct ifnet *ifp,
4096 union if_snd_tag_alloc_params *params,
4097 struct m_snd_tag **ppmt)
4100 switch (params->hdr.type) {
4102 case IF_SND_TAG_TYPE_RATE_LIMIT:
4103 return (mlx5e_rl_snd_tag_alloc(ifp, params, ppmt));
4104 #if defined(KERN_TLS) && defined(IF_SND_TAG_TYPE_TLS_RATE_LIMIT)
4105 case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
4106 return (mlx5e_tls_snd_tag_alloc(ifp, params, ppmt));
4109 case IF_SND_TAG_TYPE_UNLIMITED:
4110 return (mlx5e_ul_snd_tag_alloc(ifp, params, ppmt));
4112 case IF_SND_TAG_TYPE_TLS:
4113 return (mlx5e_tls_snd_tag_alloc(ifp, params, ppmt));
4116 return (EOPNOTSUPP);
4121 mlx5e_snd_tag_modify(struct m_snd_tag *pmt, union if_snd_tag_modify_params *params)
4123 struct mlx5e_snd_tag *tag =
4124 container_of(pmt, struct mlx5e_snd_tag, m_snd_tag);
4126 switch (tag->type) {
4128 case IF_SND_TAG_TYPE_RATE_LIMIT:
4129 return (mlx5e_rl_snd_tag_modify(pmt, params));
4130 #if defined(KERN_TLS) && defined(IF_SND_TAG_TYPE_TLS_RATE_LIMIT)
4131 case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
4132 return (mlx5e_tls_snd_tag_modify(pmt, params));
4135 case IF_SND_TAG_TYPE_UNLIMITED:
4137 case IF_SND_TAG_TYPE_TLS:
4140 return (EOPNOTSUPP);
4145 mlx5e_snd_tag_query(struct m_snd_tag *pmt, union if_snd_tag_query_params *params)
4147 struct mlx5e_snd_tag *tag =
4148 container_of(pmt, struct mlx5e_snd_tag, m_snd_tag);
4150 switch (tag->type) {
4152 case IF_SND_TAG_TYPE_RATE_LIMIT:
4153 return (mlx5e_rl_snd_tag_query(pmt, params));
4154 #if defined(KERN_TLS) && defined(IF_SND_TAG_TYPE_TLS_RATE_LIMIT)
4155 case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
4156 return (mlx5e_tls_snd_tag_query(pmt, params));
4159 case IF_SND_TAG_TYPE_UNLIMITED:
4160 return (mlx5e_ul_snd_tag_query(pmt, params));
4162 case IF_SND_TAG_TYPE_TLS:
4163 return (mlx5e_tls_snd_tag_query(pmt, params));
4166 return (EOPNOTSUPP);
4171 #define NUM_HDWR_RATES_MLX 13
4172 static const uint64_t adapter_rates_mlx[NUM_HDWR_RATES_MLX] = {
4173 135375, /* 1,083,000 */
4174 180500, /* 1,444,000 */
4175 270750, /* 2,166,000 */
4176 361000, /* 2,888,000 */
4177 541500, /* 4,332,000 */
4178 721875, /* 5,775,000 */
4179 1082875, /* 8,663,000 */
4180 1443875, /* 11,551,000 */
4181 2165750, /* 17,326,000 */
4182 2887750, /* 23,102,000 */
4183 4331625, /* 34,653,000 */
4184 5775500, /* 46,204,000 */
4185 8663125 /* 69,305,000 */
4189 mlx5e_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q)
4192 * This function needs updating by the driver maintainer!
4193 * For the MLX card there are currently (ConectX-4?) 13
4194 * pre-set rates and others i.e. ConnectX-5, 6, 7??
4196 * This will change based on later adapters
4197 * and this code should be updated to look at ifp
4198 * and figure out the specific adapter type
4199 * settings i.e. how many rates as well
4200 * as if they are fixed (as is shown here) or
4201 * if they are dynamic (example chelsio t4). Also if there
4202 * is a maximum number of flows that the adapter
4203 * can handle that too needs to be updated in
4204 * the max_flows field.
4206 q->rate_table = adapter_rates_mlx;
4207 q->flags = RT_IS_FIXED_TABLE;
4208 q->max_flows = 0; /* mlx has no limit */
4209 q->number_of_rates = NUM_HDWR_RATES_MLX;
4210 q->min_segment_burst = 1;
4215 mlx5e_snd_tag_free(struct m_snd_tag *pmt)
4217 struct mlx5e_snd_tag *tag =
4218 container_of(pmt, struct mlx5e_snd_tag, m_snd_tag);
4220 switch (tag->type) {
4222 case IF_SND_TAG_TYPE_RATE_LIMIT:
4223 mlx5e_rl_snd_tag_free(pmt);
4225 #if defined(KERN_TLS) && defined(IF_SND_TAG_TYPE_TLS_RATE_LIMIT)
4226 case IF_SND_TAG_TYPE_TLS_RATE_LIMIT:
4227 mlx5e_tls_snd_tag_free(pmt);
4231 case IF_SND_TAG_TYPE_UNLIMITED:
4232 mlx5e_ul_snd_tag_free(pmt);
4235 case IF_SND_TAG_TYPE_TLS:
4236 mlx5e_tls_snd_tag_free(pmt);
4245 mlx5e_create_ifp(struct mlx5_core_dev *mdev)
4248 struct mlx5e_priv *priv;
4249 u8 dev_addr[ETHER_ADDR_LEN] __aligned(4);
4251 struct sysctl_oid_list *child;
4252 int ncv = mdev->priv.eq_table.num_comp_vectors;
4254 struct pfil_head_args pa;
4258 u32 out[MLX5_ST_SZ_DW(ptys_reg)];
4261 struct media media_entry = {};
4263 if (mlx5e_check_required_hca_cap(mdev)) {
4264 mlx5_core_dbg(mdev, "mlx5e_check_required_hca_cap() failed\n");
4268 * Try to allocate the priv and make room for worst-case
4269 * number of channel structures:
4271 priv = malloc(sizeof(*priv) +
4272 (sizeof(priv->channel[0]) * mdev->priv.eq_table.num_comp_vectors),
4273 M_MLX5EN, M_WAITOK | M_ZERO);
4275 ifp = priv->ifp = if_alloc_dev(IFT_ETHER, mdev->pdev->dev.bsddev);
4277 mlx5_core_err(mdev, "if_alloc() failed\n");
4280 /* setup all static fields */
4281 mlx5e_priv_static_init(priv, mdev->priv.eq_table.num_comp_vectors);
4283 ifp->if_softc = priv;
4284 if_initname(ifp, "mce", device_get_unit(mdev->pdev->dev.bsddev));
4285 ifp->if_mtu = ETHERMTU;
4286 ifp->if_init = mlx5e_open;
4287 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST |
4289 ifp->if_ioctl = mlx5e_ioctl;
4290 ifp->if_transmit = mlx5e_xmit;
4291 ifp->if_qflush = if_qflush;
4292 #if (__FreeBSD_version >= 1100000)
4293 ifp->if_get_counter = mlx5e_get_counter;
4295 ifp->if_snd.ifq_maxlen = ifqmaxlen;
4297 * Set driver features
4299 ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6;
4300 ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
4301 ifp->if_capabilities |= IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWFILTER;
4302 ifp->if_capabilities |= IFCAP_LINKSTATE | IFCAP_JUMBO_MTU;
4303 ifp->if_capabilities |= IFCAP_LRO;
4304 ifp->if_capabilities |= IFCAP_TSO | IFCAP_VLAN_HWTSO;
4305 ifp->if_capabilities |= IFCAP_HWSTATS | IFCAP_HWRXTSTMP;
4306 ifp->if_capabilities |= IFCAP_NOMAP;
4307 ifp->if_capabilities |= IFCAP_TXTLS4 | IFCAP_TXTLS6;
4308 ifp->if_capabilities |= IFCAP_TXRTLMT;
4309 ifp->if_snd_tag_alloc = mlx5e_snd_tag_alloc;
4310 ifp->if_snd_tag_free = mlx5e_snd_tag_free;
4311 ifp->if_snd_tag_modify = mlx5e_snd_tag_modify;
4312 ifp->if_snd_tag_query = mlx5e_snd_tag_query;
4314 ifp->if_ratelimit_query = mlx5e_ratelimit_query;
4316 /* set TSO limits so that we don't have to drop TX packets */
4317 ifp->if_hw_tsomax = MLX5E_MAX_TX_PAYLOAD_SIZE - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
4318 ifp->if_hw_tsomaxsegcount = MLX5E_MAX_TX_MBUF_FRAGS - 1 /* hdr */;
4319 ifp->if_hw_tsomaxsegsize = MLX5E_MAX_TX_MBUF_SIZE;
4321 ifp->if_capenable = ifp->if_capabilities;
4322 ifp->if_hwassist = 0;
4323 if (ifp->if_capenable & IFCAP_TSO)
4324 ifp->if_hwassist |= CSUM_TSO;
4325 if (ifp->if_capenable & IFCAP_TXCSUM)
4326 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP | CSUM_IP);
4327 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
4328 ifp->if_hwassist |= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
4330 /* ifnet sysctl tree */
4331 sysctl_ctx_init(&priv->sysctl_ctx);
4332 priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_dev),
4333 OID_AUTO, ifp->if_dname, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
4334 "MLX5 ethernet - interface name");
4335 if (priv->sysctl_ifnet == NULL) {
4336 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
4337 goto err_free_sysctl;
4339 snprintf(unit, sizeof(unit), "%d", ifp->if_dunit);
4340 priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
4341 OID_AUTO, unit, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
4342 "MLX5 ethernet - interface unit");
4343 if (priv->sysctl_ifnet == NULL) {
4344 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
4345 goto err_free_sysctl;
4348 /* HW sysctl tree */
4349 child = SYSCTL_CHILDREN(device_get_sysctl_tree(mdev->pdev->dev.bsddev));
4350 priv->sysctl_hw = SYSCTL_ADD_NODE(&priv->sysctl_ctx, child,
4351 OID_AUTO, "hw", CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
4352 "MLX5 ethernet dev hw");
4353 if (priv->sysctl_hw == NULL) {
4354 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
4355 goto err_free_sysctl;
4358 err = mlx5e_build_ifp_priv(mdev, priv, ncv);
4360 mlx5_core_err(mdev, "mlx5e_build_ifp_priv() failed (%d)\n", err);
4361 goto err_free_sysctl;
4364 /* reuse mlx5core's watchdog workqueue */
4365 priv->wq = mdev->priv.health.wq_watchdog;
4367 err = mlx5_alloc_map_uar(mdev, &priv->cq_uar);
4369 mlx5_en_err(ifp, "mlx5_alloc_map_uar failed, %d\n", err);
4372 err = mlx5_core_alloc_pd(mdev, &priv->pdn);
4374 mlx5_en_err(ifp, "mlx5_core_alloc_pd failed, %d\n", err);
4375 goto err_unmap_free_uar;
4377 err = mlx5_alloc_transport_domain(mdev, &priv->tdn);
4380 "mlx5_alloc_transport_domain failed, %d\n", err);
4381 goto err_dealloc_pd;
4383 err = mlx5e_create_mkey(priv, priv->pdn, &priv->mr);
4385 mlx5_en_err(ifp, "mlx5e_create_mkey failed, %d\n", err);
4386 goto err_dealloc_transport_domain;
4388 mlx5_query_nic_vport_mac_address(priv->mdev, 0, dev_addr);
4390 /* check if we should generate a random MAC address */
4391 if (MLX5_CAP_GEN(priv->mdev, vport_group_manager) == 0 &&
4392 is_zero_ether_addr(dev_addr)) {
4393 random_ether_addr(dev_addr);
4394 mlx5_en_err(ifp, "Assigned random MAC address\n");
4397 err = mlx5e_rl_init(priv);
4399 mlx5_en_err(ifp, "mlx5e_rl_init failed, %d\n", err);
4400 goto err_create_mkey;
4403 err = mlx5e_tls_init(priv);
4405 if_printf(ifp, "%s: mlx5e_tls_init failed\n", __func__);
4409 /* set default MTU */
4410 mlx5e_set_dev_port_mtu(ifp, ifp->if_mtu);
4412 /* Set default media status */
4413 priv->media_status_last = IFM_AVALID;
4414 priv->media_active_last = IFM_ETHER | IFM_AUTO |
4415 IFM_ETH_RXPAUSE | IFM_FDX;
4417 /* setup default pauseframes configuration */
4418 mlx5e_setup_pauseframes(priv);
4420 /* Setup supported medias */
4421 //TODO: If we failed to query ptys is it ok to proceed??
4422 if (!mlx5_query_port_ptys(mdev, out, sizeof(out), MLX5_PTYS_EN, 1)) {
4423 ext = MLX5_CAP_PCAM_FEATURE(mdev,
4424 ptys_extended_ethernet);
4425 eth_proto_cap = MLX5_GET_ETH_PROTO(ptys_reg, out, ext,
4426 eth_proto_capability);
4427 if (MLX5_CAP_PCAM_FEATURE(mdev, ptys_connector_type))
4428 connector_type = MLX5_GET(ptys_reg, out,
4432 mlx5_en_err(ifp, "Query port media capability failed, %d\n", err);
4435 ifmedia_init(&priv->media, IFM_IMASK | IFM_ETH_FMASK,
4436 mlx5e_media_change, mlx5e_media_status);
4438 speeds_num = ext ? MLX5E_EXT_LINK_SPEEDS_NUMBER : MLX5E_LINK_SPEEDS_NUMBER;
4439 for (i = 0; i != speeds_num; i++) {
4440 for (j = 0; j < MLX5E_LINK_MODES_NUMBER ; ++j) {
4441 media_entry = ext ? mlx5e_ext_mode_table[i][j] :
4442 mlx5e_mode_table[i][j];
4443 if (media_entry.baudrate == 0)
4445 if (MLX5E_PROT_MASK(i) & eth_proto_cap) {
4446 ifmedia_add(&priv->media,
4447 media_entry.subtype |
4448 IFM_ETHER, 0, NULL);
4449 ifmedia_add(&priv->media,
4450 media_entry.subtype |
4451 IFM_ETHER | IFM_FDX |
4452 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
4457 ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO, 0, NULL);
4458 ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
4459 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
4461 /* Set autoselect by default */
4462 ifmedia_set(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
4463 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE);
4465 DEBUGNET_SET(ifp, mlx5_en);
4467 ether_ifattach(ifp, dev_addr);
4469 /* Register for VLAN events */
4470 priv->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
4471 mlx5e_vlan_rx_add_vid, priv, EVENTHANDLER_PRI_FIRST);
4472 priv->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
4473 mlx5e_vlan_rx_kill_vid, priv, EVENTHANDLER_PRI_FIRST);
4475 /* Link is down by default */
4476 if_link_state_change(ifp, LINK_STATE_DOWN);
4478 mlx5e_enable_async_events(priv);
4480 mlx5e_add_hw_stats(priv);
4482 mlx5e_create_stats(&priv->stats.vport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
4483 "vstats", mlx5e_vport_stats_desc, MLX5E_VPORT_STATS_NUM,
4484 priv->stats.vport.arg);
4486 mlx5e_create_stats(&priv->stats.pport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
4487 "pstats", mlx5e_pport_stats_desc, MLX5E_PPORT_STATS_NUM,
4488 priv->stats.pport.arg);
4490 mlx5e_create_ethtool(priv);
4492 mtx_lock(&priv->async_events_mtx);
4493 mlx5e_update_stats(priv);
4494 mtx_unlock(&priv->async_events_mtx);
4496 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
4497 OID_AUTO, "rx_clbr_done", CTLFLAG_RD,
4498 &priv->clbr_done, 0,
4499 "RX timestamps calibration state");
4500 callout_init(&priv->tstmp_clbr, CALLOUT_DIRECT);
4501 mlx5e_reset_calibration_callout(priv);
4503 pa.pa_version = PFIL_VERSION;
4504 pa.pa_flags = PFIL_IN;
4505 pa.pa_type = PFIL_TYPE_ETHERNET;
4506 pa.pa_headname = ifp->if_xname;
4507 priv->pfil = pfil_head_register(&pa);
4512 mlx5e_rl_cleanup(priv);
4515 mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
4517 err_dealloc_transport_domain:
4518 mlx5_dealloc_transport_domain(mdev, priv->tdn);
4521 mlx5_core_dealloc_pd(mdev, priv->pdn);
4524 mlx5_unmap_free_uar(mdev, &priv->cq_uar);
4527 flush_workqueue(priv->wq);
4530 sysctl_ctx_free(&priv->sysctl_ctx);
4531 if (priv->sysctl_debug)
4532 sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
4533 mlx5e_priv_static_destroy(priv, mdev->priv.eq_table.num_comp_vectors);
4537 free(priv, M_MLX5EN);
4542 mlx5e_destroy_ifp(struct mlx5_core_dev *mdev, void *vpriv)
4544 struct mlx5e_priv *priv = vpriv;
4545 struct ifnet *ifp = priv->ifp;
4547 /* don't allow more IOCTLs */
4550 /* XXX wait a bit to allow IOCTL handlers to complete */
4555 * The kernel can have reference(s) via the m_snd_tag's into
4556 * the ratelimit channels, and these must go away before
4559 while (READ_ONCE(priv->rl.stats.tx_active_connections) != 0) {
4560 mlx5_en_err(priv->ifp,
4561 "Waiting for all ratelimit connections to terminate\n");
4565 /* wait for all unlimited send tags to complete */
4566 mlx5e_priv_wait_for_completion(priv, mdev->priv.eq_table.num_comp_vectors);
4568 /* stop watchdog timer */
4569 callout_drain(&priv->watchdog);
4571 callout_drain(&priv->tstmp_clbr);
4573 if (priv->vlan_attach != NULL)
4574 EVENTHANDLER_DEREGISTER(vlan_config, priv->vlan_attach);
4575 if (priv->vlan_detach != NULL)
4576 EVENTHANDLER_DEREGISTER(vlan_unconfig, priv->vlan_detach);
4578 /* make sure device gets closed */
4580 mlx5e_close_locked(ifp);
4583 /* deregister pfil */
4584 if (priv->pfil != NULL) {
4585 pfil_head_unregister(priv->pfil);
4589 /* unregister device */
4590 ifmedia_removeall(&priv->media);
4591 ether_ifdetach(ifp);
4593 mlx5e_tls_cleanup(priv);
4594 mlx5e_rl_cleanup(priv);
4596 /* destroy all remaining sysctl nodes */
4597 sysctl_ctx_free(&priv->stats.vport.ctx);
4598 sysctl_ctx_free(&priv->stats.pport.ctx);
4599 if (priv->sysctl_debug)
4600 sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
4601 sysctl_ctx_free(&priv->sysctl_ctx);
4603 mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
4604 mlx5_dealloc_transport_domain(priv->mdev, priv->tdn);
4605 mlx5_core_dealloc_pd(priv->mdev, priv->pdn);
4606 mlx5_unmap_free_uar(priv->mdev, &priv->cq_uar);
4607 mlx5e_disable_async_events(priv);
4608 flush_workqueue(priv->wq);
4609 mlx5e_priv_static_destroy(priv, mdev->priv.eq_table.num_comp_vectors);
4611 free(priv, M_MLX5EN);
4616 mlx5_en_debugnet_init(struct ifnet *dev, int *nrxr, int *ncl, int *clsize)
4618 struct mlx5e_priv *priv = if_getsoftc(dev);
4621 *nrxr = priv->params.num_channels;
4622 *ncl = DEBUGNET_MAX_IN_FLIGHT;
4623 *clsize = MLX5E_MAX_RX_BYTES;
4628 mlx5_en_debugnet_event(struct ifnet *dev, enum debugnet_ev event)
4633 mlx5_en_debugnet_transmit(struct ifnet *dev, struct mbuf *m)
4635 struct mlx5e_priv *priv = if_getsoftc(dev);
4636 struct mlx5e_sq *sq;
4639 if ((if_getdrvflags(dev) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4640 IFF_DRV_RUNNING || (priv->media_status_last & IFM_ACTIVE) == 0)
4643 sq = &priv->channel[0].sq[0];
4645 if (sq->running == 0) {
4650 if (mlx5e_sq_xmit(sq, &m) != 0) {
4657 if (likely(sq->doorbell.d64 != 0)) {
4658 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
4659 sq->doorbell.d64 = 0;
4665 mlx5_en_debugnet_poll(struct ifnet *dev, int count)
4667 struct mlx5e_priv *priv = if_getsoftc(dev);
4669 if ((if_getdrvflags(dev) & IFF_DRV_RUNNING) == 0 ||
4670 (priv->media_status_last & IFM_ACTIVE) == 0)
4673 mlx5_poll_interrupts(priv->mdev);
4677 #endif /* DEBUGNET */
4680 mlx5e_get_ifp(void *vpriv)
4682 struct mlx5e_priv *priv = vpriv;
4687 static struct mlx5_interface mlx5e_interface = {
4688 .add = mlx5e_create_ifp,
4689 .remove = mlx5e_destroy_ifp,
4690 .event = mlx5e_async_event,
4691 .protocol = MLX5_INTERFACE_PROTOCOL_ETH,
4692 .get_dev = mlx5e_get_ifp,
4698 mlx5_register_interface(&mlx5e_interface);
4704 mlx5_unregister_interface(&mlx5e_interface);
4708 mlx5e_show_version(void __unused *arg)
4711 printf("%s", mlx5e_version);
4713 SYSINIT(mlx5e_show_version, SI_SUB_DRIVERS, SI_ORDER_ANY, mlx5e_show_version, NULL);
4715 module_init_order(mlx5e_init, SI_ORDER_THIRD);
4716 module_exit_order(mlx5e_cleanup, SI_ORDER_THIRD);
4718 #if (__FreeBSD_version >= 1100000)
4719 MODULE_DEPEND(mlx5en, linuxkpi, 1, 1, 1);
4721 MODULE_DEPEND(mlx5en, mlx5, 1, 1, 1);
4722 MODULE_VERSION(mlx5en, 1);