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
30 #include <sys/sockio.h>
31 #include <machine/atomic.h>
33 #ifndef ETH_DRIVER_VERSION
34 #define ETH_DRIVER_VERSION "3.5.0"
36 #define DRIVER_RELDATE "November 2018"
38 static const char mlx5e_version[] = "mlx5en: Mellanox Ethernet driver "
39 ETH_DRIVER_VERSION " (" DRIVER_RELDATE ")\n";
41 static int mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs);
43 struct mlx5e_channel_param {
44 struct mlx5e_rq_param rq;
45 struct mlx5e_sq_param sq;
46 struct mlx5e_cq_param rx_cq;
47 struct mlx5e_cq_param tx_cq;
53 } mlx5e_mode_table[MLX5E_LINK_MODES_NUMBER] = {
55 [MLX5E_1000BASE_CX_SGMII] = {
56 .subtype = IFM_1000_CX_SGMII,
57 .baudrate = IF_Mbps(1000ULL),
59 [MLX5E_1000BASE_KX] = {
60 .subtype = IFM_1000_KX,
61 .baudrate = IF_Mbps(1000ULL),
63 [MLX5E_10GBASE_CX4] = {
64 .subtype = IFM_10G_CX4,
65 .baudrate = IF_Gbps(10ULL),
67 [MLX5E_10GBASE_KX4] = {
68 .subtype = IFM_10G_KX4,
69 .baudrate = IF_Gbps(10ULL),
71 [MLX5E_10GBASE_KR] = {
72 .subtype = IFM_10G_KR,
73 .baudrate = IF_Gbps(10ULL),
75 [MLX5E_20GBASE_KR2] = {
76 .subtype = IFM_20G_KR2,
77 .baudrate = IF_Gbps(20ULL),
79 [MLX5E_40GBASE_CR4] = {
80 .subtype = IFM_40G_CR4,
81 .baudrate = IF_Gbps(40ULL),
83 [MLX5E_40GBASE_KR4] = {
84 .subtype = IFM_40G_KR4,
85 .baudrate = IF_Gbps(40ULL),
87 [MLX5E_56GBASE_R4] = {
88 .subtype = IFM_56G_R4,
89 .baudrate = IF_Gbps(56ULL),
91 [MLX5E_10GBASE_CR] = {
92 .subtype = IFM_10G_CR1,
93 .baudrate = IF_Gbps(10ULL),
95 [MLX5E_10GBASE_SR] = {
96 .subtype = IFM_10G_SR,
97 .baudrate = IF_Gbps(10ULL),
99 [MLX5E_10GBASE_ER] = {
100 .subtype = IFM_10G_ER,
101 .baudrate = IF_Gbps(10ULL),
103 [MLX5E_40GBASE_SR4] = {
104 .subtype = IFM_40G_SR4,
105 .baudrate = IF_Gbps(40ULL),
107 [MLX5E_40GBASE_LR4] = {
108 .subtype = IFM_40G_LR4,
109 .baudrate = IF_Gbps(40ULL),
111 [MLX5E_100GBASE_CR4] = {
112 .subtype = IFM_100G_CR4,
113 .baudrate = IF_Gbps(100ULL),
115 [MLX5E_100GBASE_SR4] = {
116 .subtype = IFM_100G_SR4,
117 .baudrate = IF_Gbps(100ULL),
119 [MLX5E_100GBASE_KR4] = {
120 .subtype = IFM_100G_KR4,
121 .baudrate = IF_Gbps(100ULL),
123 [MLX5E_100GBASE_LR4] = {
124 .subtype = IFM_100G_LR4,
125 .baudrate = IF_Gbps(100ULL),
127 [MLX5E_100BASE_TX] = {
128 .subtype = IFM_100_TX,
129 .baudrate = IF_Mbps(100ULL),
131 [MLX5E_1000BASE_T] = {
132 .subtype = IFM_1000_T,
133 .baudrate = IF_Mbps(1000ULL),
135 [MLX5E_10GBASE_T] = {
136 .subtype = IFM_10G_T,
137 .baudrate = IF_Gbps(10ULL),
139 [MLX5E_25GBASE_CR] = {
140 .subtype = IFM_25G_CR,
141 .baudrate = IF_Gbps(25ULL),
143 [MLX5E_25GBASE_KR] = {
144 .subtype = IFM_25G_KR,
145 .baudrate = IF_Gbps(25ULL),
147 [MLX5E_25GBASE_SR] = {
148 .subtype = IFM_25G_SR,
149 .baudrate = IF_Gbps(25ULL),
151 [MLX5E_50GBASE_CR2] = {
152 .subtype = IFM_50G_CR2,
153 .baudrate = IF_Gbps(50ULL),
155 [MLX5E_50GBASE_KR2] = {
156 .subtype = IFM_50G_KR2,
157 .baudrate = IF_Gbps(50ULL),
161 MALLOC_DEFINE(M_MLX5EN, "MLX5EN", "MLX5 Ethernet");
164 mlx5e_update_carrier(struct mlx5e_priv *priv)
166 struct mlx5_core_dev *mdev = priv->mdev;
167 u32 out[MLX5_ST_SZ_DW(ptys_reg)];
174 port_state = mlx5_query_vport_state(mdev,
175 MLX5_QUERY_VPORT_STATE_IN_OP_MOD_VNIC_VPORT, 0);
177 if (port_state == VPORT_STATE_UP) {
178 priv->media_status_last |= IFM_ACTIVE;
180 priv->media_status_last &= ~IFM_ACTIVE;
181 priv->media_active_last = IFM_ETHER;
182 if_link_state_change(priv->ifp, LINK_STATE_DOWN);
186 error = mlx5_query_port_ptys(mdev, out, sizeof(out), MLX5_PTYS_EN, 1);
188 priv->media_active_last = IFM_ETHER;
189 priv->ifp->if_baudrate = 1;
190 if_printf(priv->ifp, "%s: query port ptys failed: 0x%x\n",
194 eth_proto_oper = MLX5_GET(ptys_reg, out, eth_proto_oper);
196 for (i = 0; i != MLX5E_LINK_MODES_NUMBER; i++) {
197 if (mlx5e_mode_table[i].baudrate == 0)
199 if (MLX5E_PROT_MASK(i) & eth_proto_oper) {
200 u32 subtype = mlx5e_mode_table[i].subtype;
202 priv->ifp->if_baudrate =
203 mlx5e_mode_table[i].baudrate;
207 error = mlx5_query_pddr_range_info(mdev, 1, &is_er_type);
209 if_printf(priv->ifp, "%s: query port pddr failed: %d\n",
212 if (error != 0 || is_er_type == 0)
213 subtype = IFM_10G_LR;
216 error = mlx5_query_pddr_range_info(mdev, 1, &is_er_type);
218 if_printf(priv->ifp, "%s: query port pddr failed: %d\n",
221 if (error == 0 && is_er_type != 0)
222 subtype = IFM_40G_ER4;
225 priv->media_active_last = subtype | IFM_ETHER | IFM_FDX;
229 if_link_state_change(priv->ifp, LINK_STATE_UP);
233 mlx5e_media_status(struct ifnet *dev, struct ifmediareq *ifmr)
235 struct mlx5e_priv *priv = dev->if_softc;
237 ifmr->ifm_status = priv->media_status_last;
238 ifmr->ifm_active = priv->media_active_last |
239 (priv->params.rx_pauseframe_control ? IFM_ETH_RXPAUSE : 0) |
240 (priv->params.tx_pauseframe_control ? IFM_ETH_TXPAUSE : 0);
245 mlx5e_find_link_mode(u32 subtype)
252 subtype = IFM_10G_ER;
255 subtype = IFM_40G_LR4;
259 for (i = 0; i < MLX5E_LINK_MODES_NUMBER; ++i) {
260 if (mlx5e_mode_table[i].baudrate == 0)
262 if (mlx5e_mode_table[i].subtype == subtype)
263 link_mode |= MLX5E_PROT_MASK(i);
270 mlx5e_set_port_pause_and_pfc(struct mlx5e_priv *priv)
272 return (mlx5_set_port_pause_and_pfc(priv->mdev, 1,
273 priv->params.rx_pauseframe_control,
274 priv->params.tx_pauseframe_control,
275 priv->params.rx_priority_flow_control,
276 priv->params.tx_priority_flow_control));
280 mlx5e_set_port_pfc(struct mlx5e_priv *priv)
284 if (priv->gone != 0) {
286 } else if (priv->params.rx_pauseframe_control ||
287 priv->params.tx_pauseframe_control) {
289 "Global pauseframes must be disabled before enabling PFC.\n");
292 error = mlx5e_set_port_pause_and_pfc(priv);
298 mlx5e_media_change(struct ifnet *dev)
300 struct mlx5e_priv *priv = dev->if_softc;
301 struct mlx5_core_dev *mdev = priv->mdev;
308 locked = PRIV_LOCKED(priv);
312 if (IFM_TYPE(priv->media.ifm_media) != IFM_ETHER) {
316 link_mode = mlx5e_find_link_mode(IFM_SUBTYPE(priv->media.ifm_media));
318 /* query supported capabilities */
319 error = mlx5_query_port_proto_cap(mdev, ð_proto_cap, MLX5_PTYS_EN);
321 if_printf(dev, "Query port media capability failed\n");
324 /* check for autoselect */
325 if (IFM_SUBTYPE(priv->media.ifm_media) == IFM_AUTO) {
326 link_mode = eth_proto_cap;
327 if (link_mode == 0) {
328 if_printf(dev, "Port media capability is zero\n");
333 link_mode = link_mode & eth_proto_cap;
334 if (link_mode == 0) {
335 if_printf(dev, "Not supported link mode requested\n");
340 if (priv->media.ifm_media & (IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE)) {
341 /* check if PFC is enabled */
342 if (priv->params.rx_priority_flow_control ||
343 priv->params.tx_priority_flow_control) {
344 if_printf(dev, "PFC must be disabled before enabling global pauseframes.\n");
349 /* update pauseframe control bits */
350 priv->params.rx_pauseframe_control =
351 (priv->media.ifm_media & IFM_ETH_RXPAUSE) ? 1 : 0;
352 priv->params.tx_pauseframe_control =
353 (priv->media.ifm_media & IFM_ETH_TXPAUSE) ? 1 : 0;
355 /* check if device is opened */
356 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
358 /* reconfigure the hardware */
359 mlx5_set_port_status(mdev, MLX5_PORT_DOWN);
360 mlx5_set_port_proto(mdev, link_mode, MLX5_PTYS_EN);
361 error = -mlx5e_set_port_pause_and_pfc(priv);
363 mlx5_set_port_status(mdev, MLX5_PORT_UP);
372 mlx5e_update_carrier_work(struct work_struct *work)
374 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
375 update_carrier_work);
378 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
379 mlx5e_update_carrier(priv);
384 * This function reads the physical port counters from the firmware
385 * using a pre-defined layout defined by various MLX5E_PPORT_XXX()
386 * macros. The output is converted from big-endian 64-bit values into
387 * host endian ones and stored in the "priv->stats.pport" structure.
390 mlx5e_update_pport_counters(struct mlx5e_priv *priv)
392 struct mlx5_core_dev *mdev = priv->mdev;
393 struct mlx5e_pport_stats *s = &priv->stats.pport;
394 struct mlx5e_port_stats_debug *s_debug = &priv->stats.port_stats_debug;
398 unsigned sz = MLX5_ST_SZ_BYTES(ppcnt_reg);
403 /* allocate firmware request structures */
404 in = mlx5_vzalloc(sz);
405 out = mlx5_vzalloc(sz);
406 if (in == NULL || out == NULL)
410 * Get pointer to the 64-bit counter set which is located at a
411 * fixed offset in the output firmware request structure:
413 ptr = (const uint64_t *)MLX5_ADDR_OF(ppcnt_reg, out, counter_set);
415 MLX5_SET(ppcnt_reg, in, local_port, 1);
417 /* read IEEE802_3 counter group using predefined counter layout */
418 MLX5_SET(ppcnt_reg, in, grp, MLX5_IEEE_802_3_COUNTERS_GROUP);
419 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
420 for (x = 0, y = MLX5E_PPORT_PER_PRIO_STATS_NUM;
421 x != MLX5E_PPORT_IEEE802_3_STATS_NUM; x++, y++)
422 s->arg[y] = be64toh(ptr[x]);
424 /* read RFC2819 counter group using predefined counter layout */
425 MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2819_COUNTERS_GROUP);
426 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
427 for (x = 0; x != MLX5E_PPORT_RFC2819_STATS_NUM; x++, y++)
428 s->arg[y] = be64toh(ptr[x]);
429 for (y = 0; x != MLX5E_PPORT_RFC2819_STATS_NUM +
430 MLX5E_PPORT_RFC2819_STATS_DEBUG_NUM; x++, y++)
431 s_debug->arg[y] = be64toh(ptr[x]);
433 /* read RFC2863 counter group using predefined counter layout */
434 MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2863_COUNTERS_GROUP);
435 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
436 for (x = 0; x != MLX5E_PPORT_RFC2863_STATS_DEBUG_NUM; x++, y++)
437 s_debug->arg[y] = be64toh(ptr[x]);
439 /* read physical layer stats counter group using predefined counter layout */
440 MLX5_SET(ppcnt_reg, in, grp, MLX5_PHYSICAL_LAYER_COUNTERS_GROUP);
441 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
442 for (x = 0; x != MLX5E_PPORT_PHYSICAL_LAYER_STATS_DEBUG_NUM; x++, y++)
443 s_debug->arg[y] = be64toh(ptr[x]);
445 /* read Extended Ethernet counter group using predefined counter layout */
446 MLX5_SET(ppcnt_reg, in, grp, MLX5_ETHERNET_EXTENDED_COUNTERS_GROUP);
447 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
448 for (x = 0; x != MLX5E_PPORT_ETHERNET_EXTENDED_STATS_DEBUG_NUM; x++, y++)
449 s_debug->arg[y] = be64toh(ptr[x]);
451 /* read per-priority counters */
452 MLX5_SET(ppcnt_reg, in, grp, MLX5_PER_PRIORITY_COUNTERS_GROUP);
454 /* iterate all the priorities */
455 for (y = z = 0; z != MLX5E_PPORT_PER_PRIO_STATS_NUM_PRIO; z++) {
456 MLX5_SET(ppcnt_reg, in, prio_tc, z);
457 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
459 /* read per priority stats counter group using predefined counter layout */
460 for (x = 0; x != (MLX5E_PPORT_PER_PRIO_STATS_NUM /
461 MLX5E_PPORT_PER_PRIO_STATS_NUM_PRIO); x++, y++)
462 s->arg[y] = be64toh(ptr[x]);
466 /* free firmware request structures */
472 * This function is called regularly to collect all statistics
473 * counters from the firmware. The values can be viewed through the
474 * sysctl interface. Execution is serialized using the priv's global
475 * configuration lock.
478 mlx5e_update_stats_locked(struct mlx5e_priv *priv)
480 struct mlx5_core_dev *mdev = priv->mdev;
481 struct mlx5e_vport_stats *s = &priv->stats.vport;
482 struct mlx5e_sq_stats *sq_stats;
483 struct buf_ring *sq_br;
484 #if (__FreeBSD_version < 1100000)
485 struct ifnet *ifp = priv->ifp;
488 u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)];
490 int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
493 u64 tx_queue_dropped = 0;
494 u64 tx_defragged = 0;
495 u64 tx_offload_none = 0;
498 u64 sw_lro_queued = 0;
499 u64 sw_lro_flushed = 0;
500 u64 rx_csum_none = 0;
504 u32 rx_out_of_buffer = 0;
508 out = mlx5_vzalloc(outlen);
512 /* Collect firts the SW counters and then HW for consistency */
513 for (i = 0; i < priv->params.num_channels; i++) {
514 struct mlx5e_channel *pch = priv->channel + i;
515 struct mlx5e_rq *rq = &pch->rq;
516 struct mlx5e_rq_stats *rq_stats = &pch->rq.stats;
518 /* collect stats from LRO */
519 rq_stats->sw_lro_queued = rq->lro.lro_queued;
520 rq_stats->sw_lro_flushed = rq->lro.lro_flushed;
521 sw_lro_queued += rq_stats->sw_lro_queued;
522 sw_lro_flushed += rq_stats->sw_lro_flushed;
523 lro_packets += rq_stats->lro_packets;
524 lro_bytes += rq_stats->lro_bytes;
525 rx_csum_none += rq_stats->csum_none;
526 rx_wqe_err += rq_stats->wqe_err;
527 rx_packets += rq_stats->packets;
528 rx_bytes += rq_stats->bytes;
530 for (j = 0; j < priv->num_tc; j++) {
531 sq_stats = &pch->sq[j].stats;
532 sq_br = pch->sq[j].br;
534 tso_packets += sq_stats->tso_packets;
535 tso_bytes += sq_stats->tso_bytes;
536 tx_queue_dropped += sq_stats->dropped;
538 tx_queue_dropped += sq_br->br_drops;
539 tx_defragged += sq_stats->defragged;
540 tx_offload_none += sq_stats->csum_offload_none;
544 /* update counters */
545 s->tso_packets = tso_packets;
546 s->tso_bytes = tso_bytes;
547 s->tx_queue_dropped = tx_queue_dropped;
548 s->tx_defragged = tx_defragged;
549 s->lro_packets = lro_packets;
550 s->lro_bytes = lro_bytes;
551 s->sw_lro_queued = sw_lro_queued;
552 s->sw_lro_flushed = sw_lro_flushed;
553 s->rx_csum_none = rx_csum_none;
554 s->rx_wqe_err = rx_wqe_err;
555 s->rx_packets = rx_packets;
556 s->rx_bytes = rx_bytes;
559 memset(in, 0, sizeof(in));
561 MLX5_SET(query_vport_counter_in, in, opcode,
562 MLX5_CMD_OP_QUERY_VPORT_COUNTER);
563 MLX5_SET(query_vport_counter_in, in, op_mod, 0);
564 MLX5_SET(query_vport_counter_in, in, other_vport, 0);
566 memset(out, 0, outlen);
568 /* get number of out-of-buffer drops first */
569 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0 &&
570 mlx5_vport_query_out_of_rx_buffer(mdev, priv->counter_set_id,
571 &rx_out_of_buffer) == 0) {
572 /* accumulate difference into a 64-bit counter */
573 s->rx_out_of_buffer += (u64)(u32)(rx_out_of_buffer -
574 s->rx_out_of_buffer_prev);
575 s->rx_out_of_buffer_prev = rx_out_of_buffer;
578 /* get port statistics */
579 if (mlx5_cmd_exec(mdev, in, sizeof(in), out, outlen) == 0) {
580 #define MLX5_GET_CTR(out, x) \
581 MLX5_GET64(query_vport_counter_out, out, x)
583 s->rx_error_packets =
584 MLX5_GET_CTR(out, received_errors.packets);
586 MLX5_GET_CTR(out, received_errors.octets);
587 s->tx_error_packets =
588 MLX5_GET_CTR(out, transmit_errors.packets);
590 MLX5_GET_CTR(out, transmit_errors.octets);
592 s->rx_unicast_packets =
593 MLX5_GET_CTR(out, received_eth_unicast.packets);
594 s->rx_unicast_bytes =
595 MLX5_GET_CTR(out, received_eth_unicast.octets);
596 s->tx_unicast_packets =
597 MLX5_GET_CTR(out, transmitted_eth_unicast.packets);
598 s->tx_unicast_bytes =
599 MLX5_GET_CTR(out, transmitted_eth_unicast.octets);
601 s->rx_multicast_packets =
602 MLX5_GET_CTR(out, received_eth_multicast.packets);
603 s->rx_multicast_bytes =
604 MLX5_GET_CTR(out, received_eth_multicast.octets);
605 s->tx_multicast_packets =
606 MLX5_GET_CTR(out, transmitted_eth_multicast.packets);
607 s->tx_multicast_bytes =
608 MLX5_GET_CTR(out, transmitted_eth_multicast.octets);
610 s->rx_broadcast_packets =
611 MLX5_GET_CTR(out, received_eth_broadcast.packets);
612 s->rx_broadcast_bytes =
613 MLX5_GET_CTR(out, received_eth_broadcast.octets);
614 s->tx_broadcast_packets =
615 MLX5_GET_CTR(out, transmitted_eth_broadcast.packets);
616 s->tx_broadcast_bytes =
617 MLX5_GET_CTR(out, transmitted_eth_broadcast.octets);
619 s->tx_packets = s->tx_unicast_packets +
620 s->tx_multicast_packets + s->tx_broadcast_packets;
621 s->tx_bytes = s->tx_unicast_bytes + s->tx_multicast_bytes +
622 s->tx_broadcast_bytes;
624 /* Update calculated offload counters */
625 s->tx_csum_offload = s->tx_packets - tx_offload_none;
626 s->rx_csum_good = s->rx_packets - s->rx_csum_none;
629 /* Get physical port counters */
630 mlx5e_update_pport_counters(priv);
632 s->tx_jumbo_packets =
633 priv->stats.port_stats_debug.tx_stat_p1519to2047octets +
634 priv->stats.port_stats_debug.tx_stat_p2048to4095octets +
635 priv->stats.port_stats_debug.tx_stat_p4096to8191octets +
636 priv->stats.port_stats_debug.tx_stat_p8192to10239octets;
638 #if (__FreeBSD_version < 1100000)
639 /* no get_counters interface in fbsd 10 */
640 ifp->if_ipackets = s->rx_packets;
641 ifp->if_ierrors = priv->stats.pport.in_range_len_errors +
642 priv->stats.pport.out_of_range_len +
643 priv->stats.pport.too_long_errors +
644 priv->stats.pport.check_seq_err +
645 priv->stats.pport.alignment_err;
646 ifp->if_iqdrops = s->rx_out_of_buffer;
647 ifp->if_opackets = s->tx_packets;
648 ifp->if_oerrors = priv->stats.port_stats_debug.out_discards;
649 ifp->if_snd.ifq_drops = s->tx_queue_dropped;
650 ifp->if_ibytes = s->rx_bytes;
651 ifp->if_obytes = s->tx_bytes;
653 priv->stats.pport.collisions;
659 /* Update diagnostics, if any */
660 if (priv->params_ethtool.diag_pci_enable ||
661 priv->params_ethtool.diag_general_enable) {
662 int error = mlx5_core_get_diagnostics_full(mdev,
663 priv->params_ethtool.diag_pci_enable ? &priv->params_pci : NULL,
664 priv->params_ethtool.diag_general_enable ? &priv->params_general : NULL);
666 if_printf(priv->ifp, "Failed reading diagnostics: %d\n", error);
671 mlx5e_update_stats_work(struct work_struct *work)
673 struct mlx5e_priv *priv;
675 priv = container_of(work, struct mlx5e_priv, update_stats_work);
677 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
678 mlx5e_update_stats_locked(priv);
683 mlx5e_update_stats(void *arg)
685 struct mlx5e_priv *priv = arg;
687 queue_work(priv->wq, &priv->update_stats_work);
689 callout_reset(&priv->watchdog, hz, &mlx5e_update_stats, priv);
693 mlx5e_async_event_sub(struct mlx5e_priv *priv,
694 enum mlx5_dev_event event)
697 case MLX5_DEV_EVENT_PORT_UP:
698 case MLX5_DEV_EVENT_PORT_DOWN:
699 queue_work(priv->wq, &priv->update_carrier_work);
708 mlx5e_async_event(struct mlx5_core_dev *mdev, void *vpriv,
709 enum mlx5_dev_event event, unsigned long param)
711 struct mlx5e_priv *priv = vpriv;
713 mtx_lock(&priv->async_events_mtx);
714 if (test_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state))
715 mlx5e_async_event_sub(priv, event);
716 mtx_unlock(&priv->async_events_mtx);
720 mlx5e_enable_async_events(struct mlx5e_priv *priv)
722 set_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
726 mlx5e_disable_async_events(struct mlx5e_priv *priv)
728 mtx_lock(&priv->async_events_mtx);
729 clear_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
730 mtx_unlock(&priv->async_events_mtx);
733 static const char *mlx5e_rq_stats_desc[] = {
734 MLX5E_RQ_STATS(MLX5E_STATS_DESC)
738 mlx5e_create_rq(struct mlx5e_channel *c,
739 struct mlx5e_rq_param *param,
742 struct mlx5e_priv *priv = c->priv;
743 struct mlx5_core_dev *mdev = priv->mdev;
745 void *rqc = param->rqc;
746 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
752 err = mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
756 /* Create DMA descriptor TAG */
757 if ((err = -bus_dma_tag_create(
758 bus_get_dma_tag(mdev->pdev->dev.bsddev),
759 1, /* any alignment */
761 BUS_SPACE_MAXADDR, /* lowaddr */
762 BUS_SPACE_MAXADDR, /* highaddr */
763 NULL, NULL, /* filter, filterarg */
764 nsegs * MLX5E_MAX_RX_BYTES, /* maxsize */
765 nsegs, /* nsegments */
766 nsegs * MLX5E_MAX_RX_BYTES, /* maxsegsize */
768 NULL, NULL, /* lockfunc, lockfuncarg */
772 err = mlx5_wq_ll_create(mdev, ¶m->wq, rqc_wq, &rq->wq,
775 goto err_free_dma_tag;
777 rq->wq.db = &rq->wq.db[MLX5_RCV_DBR];
779 err = mlx5e_get_wqe_sz(priv, &rq->wqe_sz, &rq->nsegs);
781 goto err_rq_wq_destroy;
783 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
785 err = -tcp_lro_init_args(&rq->lro, c->ifp, TCP_LRO_ENTRIES, wq_sz);
787 goto err_rq_wq_destroy;
789 rq->mbuf = malloc(wq_sz * sizeof(rq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
790 for (i = 0; i != wq_sz; i++) {
791 struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
792 #if (MLX5E_MAX_RX_SEGS == 1)
793 uint32_t byte_count = rq->wqe_sz - MLX5E_NET_IP_ALIGN;
798 err = -bus_dmamap_create(rq->dma_tag, 0, &rq->mbuf[i].dma_map);
801 bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
802 goto err_rq_mbuf_free;
805 /* set value for constant fields */
806 #if (MLX5E_MAX_RX_SEGS == 1)
807 wqe->data[0].lkey = c->mkey_be;
808 wqe->data[0].byte_count = cpu_to_be32(byte_count | MLX5_HW_START_PADDING);
810 for (j = 0; j < rq->nsegs; j++)
811 wqe->data[j].lkey = c->mkey_be;
815 INIT_WORK(&rq->dim.work, mlx5e_dim_work);
816 if (priv->params.rx_cq_moderation_mode < 2) {
817 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
819 void *cqc = container_of(param,
820 struct mlx5e_channel_param, rq)->rx_cq.cqc;
822 switch (MLX5_GET(cqc, cqc, cq_period_mode)) {
823 case MLX5_CQ_PERIOD_MODE_START_FROM_EQE:
824 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
826 case MLX5_CQ_PERIOD_MODE_START_FROM_CQE:
827 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
830 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
839 snprintf(buffer, sizeof(buffer), "rxstat%d", c->ix);
840 mlx5e_create_stats(&rq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
841 buffer, mlx5e_rq_stats_desc, MLX5E_RQ_STATS_NUM,
846 free(rq->mbuf, M_MLX5EN);
847 tcp_lro_free(&rq->lro);
849 mlx5_wq_destroy(&rq->wq_ctrl);
851 bus_dma_tag_destroy(rq->dma_tag);
857 mlx5e_destroy_rq(struct mlx5e_rq *rq)
862 /* destroy all sysctl nodes */
863 sysctl_ctx_free(&rq->stats.ctx);
865 /* free leftover LRO packets, if any */
866 tcp_lro_free(&rq->lro);
868 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
869 for (i = 0; i != wq_sz; i++) {
870 if (rq->mbuf[i].mbuf != NULL) {
871 bus_dmamap_unload(rq->dma_tag, rq->mbuf[i].dma_map);
872 m_freem(rq->mbuf[i].mbuf);
874 bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
876 free(rq->mbuf, M_MLX5EN);
877 mlx5_wq_destroy(&rq->wq_ctrl);
881 mlx5e_enable_rq(struct mlx5e_rq *rq, struct mlx5e_rq_param *param)
883 struct mlx5e_channel *c = rq->channel;
884 struct mlx5e_priv *priv = c->priv;
885 struct mlx5_core_dev *mdev = priv->mdev;
893 inlen = MLX5_ST_SZ_BYTES(create_rq_in) +
894 sizeof(u64) * rq->wq_ctrl.buf.npages;
895 in = mlx5_vzalloc(inlen);
899 rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
900 wq = MLX5_ADDR_OF(rqc, rqc, wq);
902 memcpy(rqc, param->rqc, sizeof(param->rqc));
904 MLX5_SET(rqc, rqc, cqn, c->rq.cq.mcq.cqn);
905 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST);
906 MLX5_SET(rqc, rqc, flush_in_error_en, 1);
907 if (priv->counter_set_id >= 0)
908 MLX5_SET(rqc, rqc, counter_set_id, priv->counter_set_id);
909 MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift -
911 MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma);
913 mlx5_fill_page_array(&rq->wq_ctrl.buf,
914 (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
916 err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn);
924 mlx5e_modify_rq(struct mlx5e_rq *rq, int curr_state, int next_state)
926 struct mlx5e_channel *c = rq->channel;
927 struct mlx5e_priv *priv = c->priv;
928 struct mlx5_core_dev *mdev = priv->mdev;
935 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
936 in = mlx5_vzalloc(inlen);
940 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
942 MLX5_SET(modify_rq_in, in, rqn, rq->rqn);
943 MLX5_SET(modify_rq_in, in, rq_state, curr_state);
944 MLX5_SET(rqc, rqc, state, next_state);
946 err = mlx5_core_modify_rq(mdev, in, inlen);
954 mlx5e_disable_rq(struct mlx5e_rq *rq)
956 struct mlx5e_channel *c = rq->channel;
957 struct mlx5e_priv *priv = c->priv;
958 struct mlx5_core_dev *mdev = priv->mdev;
960 mlx5_core_destroy_rq(mdev, rq->rqn);
964 mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq)
966 struct mlx5e_channel *c = rq->channel;
967 struct mlx5e_priv *priv = c->priv;
968 struct mlx5_wq_ll *wq = &rq->wq;
971 for (i = 0; i < 1000; i++) {
972 if (wq->cur_sz >= priv->params.min_rx_wqes)
981 mlx5e_open_rq(struct mlx5e_channel *c,
982 struct mlx5e_rq_param *param,
987 err = mlx5e_create_rq(c, param, rq);
991 err = mlx5e_enable_rq(rq, param);
995 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
1004 mlx5e_disable_rq(rq);
1006 mlx5e_destroy_rq(rq);
1012 mlx5e_close_rq(struct mlx5e_rq *rq)
1016 callout_stop(&rq->watchdog);
1017 mtx_unlock(&rq->mtx);
1019 callout_drain(&rq->watchdog);
1021 mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
1025 mlx5e_close_rq_wait(struct mlx5e_rq *rq)
1028 mlx5e_disable_rq(rq);
1029 mlx5e_close_cq(&rq->cq);
1030 cancel_work_sync(&rq->dim.work);
1031 mlx5e_destroy_rq(rq);
1035 mlx5e_free_sq_db(struct mlx5e_sq *sq)
1037 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1040 for (x = 0; x != wq_sz; x++) {
1041 if (sq->mbuf[x].mbuf != NULL) {
1042 bus_dmamap_unload(sq->dma_tag, sq->mbuf[x].dma_map);
1043 m_freem(sq->mbuf[x].mbuf);
1045 bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1047 free(sq->mbuf, M_MLX5EN);
1051 mlx5e_alloc_sq_db(struct mlx5e_sq *sq)
1053 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1057 sq->mbuf = malloc(wq_sz * sizeof(sq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
1059 /* Create DMA descriptor MAPs */
1060 for (x = 0; x != wq_sz; x++) {
1061 err = -bus_dmamap_create(sq->dma_tag, 0, &sq->mbuf[x].dma_map);
1064 bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1065 free(sq->mbuf, M_MLX5EN);
1072 static const char *mlx5e_sq_stats_desc[] = {
1073 MLX5E_SQ_STATS(MLX5E_STATS_DESC)
1077 mlx5e_update_sq_inline(struct mlx5e_sq *sq)
1079 sq->max_inline = sq->priv->params.tx_max_inline;
1080 sq->min_inline_mode = sq->priv->params.tx_min_inline_mode;
1083 * Check if trust state is DSCP or if inline mode is NONE which
1084 * indicates CX-5 or newer hardware.
1086 if (sq->priv->params_ethtool.trust_state != MLX5_QPTS_TRUST_PCP ||
1087 sq->min_inline_mode == MLX5_INLINE_MODE_NONE) {
1088 if (MLX5_CAP_ETH(sq->priv->mdev, wqe_vlan_insert))
1089 sq->min_insert_caps = MLX5E_INSERT_VLAN | MLX5E_INSERT_NON_VLAN;
1091 sq->min_insert_caps = MLX5E_INSERT_NON_VLAN;
1093 sq->min_insert_caps = 0;
1098 mlx5e_refresh_sq_inline_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
1102 for (i = 0; i != c->num_tc; i++) {
1103 mtx_lock(&c->sq[i].lock);
1104 mlx5e_update_sq_inline(&c->sq[i]);
1105 mtx_unlock(&c->sq[i].lock);
1110 mlx5e_refresh_sq_inline(struct mlx5e_priv *priv)
1114 /* check if channels are closed */
1115 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
1118 for (i = 0; i < priv->params.num_channels; i++)
1119 mlx5e_refresh_sq_inline_sub(priv, &priv->channel[i]);
1123 mlx5e_create_sq(struct mlx5e_channel *c,
1125 struct mlx5e_sq_param *param,
1126 struct mlx5e_sq *sq)
1128 struct mlx5e_priv *priv = c->priv;
1129 struct mlx5_core_dev *mdev = priv->mdev;
1131 void *sqc = param->sqc;
1132 void *sqc_wq = MLX5_ADDR_OF(sqc, sqc, wq);
1135 /* Create DMA descriptor TAG */
1136 if ((err = -bus_dma_tag_create(
1137 bus_get_dma_tag(mdev->pdev->dev.bsddev),
1138 1, /* any alignment */
1139 0, /* no boundary */
1140 BUS_SPACE_MAXADDR, /* lowaddr */
1141 BUS_SPACE_MAXADDR, /* highaddr */
1142 NULL, NULL, /* filter, filterarg */
1143 MLX5E_MAX_TX_PAYLOAD_SIZE, /* maxsize */
1144 MLX5E_MAX_TX_MBUF_FRAGS, /* nsegments */
1145 MLX5E_MAX_TX_MBUF_SIZE, /* maxsegsize */
1147 NULL, NULL, /* lockfunc, lockfuncarg */
1151 err = mlx5_alloc_map_uar(mdev, &sq->uar);
1153 goto err_free_dma_tag;
1155 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, &sq->wq,
1158 goto err_unmap_free_uar;
1160 sq->wq.db = &sq->wq.db[MLX5_SND_DBR];
1161 sq->bf_buf_size = (1 << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2;
1163 err = mlx5e_alloc_sq_db(sq);
1165 goto err_sq_wq_destroy;
1167 sq->mkey_be = c->mkey_be;
1168 sq->ifp = priv->ifp;
1172 mlx5e_update_sq_inline(sq);
1174 snprintf(buffer, sizeof(buffer), "txstat%dtc%d", c->ix, tc);
1175 mlx5e_create_stats(&sq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
1176 buffer, mlx5e_sq_stats_desc, MLX5E_SQ_STATS_NUM,
1182 mlx5_wq_destroy(&sq->wq_ctrl);
1185 mlx5_unmap_free_uar(mdev, &sq->uar);
1188 bus_dma_tag_destroy(sq->dma_tag);
1194 mlx5e_destroy_sq(struct mlx5e_sq *sq)
1196 /* destroy all sysctl nodes */
1197 sysctl_ctx_free(&sq->stats.ctx);
1199 mlx5e_free_sq_db(sq);
1200 mlx5_wq_destroy(&sq->wq_ctrl);
1201 mlx5_unmap_free_uar(sq->priv->mdev, &sq->uar);
1205 mlx5e_enable_sq(struct mlx5e_sq *sq, struct mlx5e_sq_param *param,
1214 inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
1215 sizeof(u64) * sq->wq_ctrl.buf.npages;
1216 in = mlx5_vzalloc(inlen);
1220 sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
1221 wq = MLX5_ADDR_OF(sqc, sqc, wq);
1223 memcpy(sqc, param->sqc, sizeof(param->sqc));
1225 MLX5_SET(sqc, sqc, tis_num_0, tis_num);
1226 MLX5_SET(sqc, sqc, cqn, sq->cq.mcq.cqn);
1227 MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
1228 MLX5_SET(sqc, sqc, tis_lst_sz, 1);
1229 MLX5_SET(sqc, sqc, flush_in_error_en, 1);
1231 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
1232 MLX5_SET(wq, wq, uar_page, sq->uar.index);
1233 MLX5_SET(wq, wq, log_wq_pg_sz, sq->wq_ctrl.buf.page_shift -
1235 MLX5_SET64(wq, wq, dbr_addr, sq->wq_ctrl.db.dma);
1237 mlx5_fill_page_array(&sq->wq_ctrl.buf,
1238 (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
1240 err = mlx5_core_create_sq(sq->priv->mdev, in, inlen, &sq->sqn);
1248 mlx5e_modify_sq(struct mlx5e_sq *sq, int curr_state, int next_state)
1255 inlen = MLX5_ST_SZ_BYTES(modify_sq_in);
1256 in = mlx5_vzalloc(inlen);
1260 sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx);
1262 MLX5_SET(modify_sq_in, in, sqn, sq->sqn);
1263 MLX5_SET(modify_sq_in, in, sq_state, curr_state);
1264 MLX5_SET(sqc, sqc, state, next_state);
1266 err = mlx5_core_modify_sq(sq->priv->mdev, in, inlen);
1274 mlx5e_disable_sq(struct mlx5e_sq *sq)
1277 mlx5_core_destroy_sq(sq->priv->mdev, sq->sqn);
1281 mlx5e_open_sq(struct mlx5e_channel *c,
1283 struct mlx5e_sq_param *param,
1284 struct mlx5e_sq *sq)
1288 err = mlx5e_create_sq(c, tc, param, sq);
1292 err = mlx5e_enable_sq(sq, param, c->priv->tisn[tc]);
1294 goto err_destroy_sq;
1296 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY);
1298 goto err_disable_sq;
1300 WRITE_ONCE(sq->running, 1);
1305 mlx5e_disable_sq(sq);
1307 mlx5e_destroy_sq(sq);
1313 mlx5e_sq_send_nops_locked(struct mlx5e_sq *sq, int can_sleep)
1315 /* fill up remainder with NOPs */
1316 while (sq->cev_counter != 0) {
1317 while (!mlx5e_sq_has_room_for(sq, 1)) {
1318 if (can_sleep != 0) {
1319 mtx_unlock(&sq->lock);
1321 mtx_lock(&sq->lock);
1326 /* send a single NOP */
1327 mlx5e_send_nop(sq, 1);
1328 atomic_thread_fence_rel();
1331 /* Check if we need to write the doorbell */
1332 if (likely(sq->doorbell.d64 != 0)) {
1333 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
1334 sq->doorbell.d64 = 0;
1339 mlx5e_sq_cev_timeout(void *arg)
1341 struct mlx5e_sq *sq = arg;
1343 mtx_assert(&sq->lock, MA_OWNED);
1345 /* check next state */
1346 switch (sq->cev_next_state) {
1347 case MLX5E_CEV_STATE_SEND_NOPS:
1348 /* fill TX ring with NOPs, if any */
1349 mlx5e_sq_send_nops_locked(sq, 0);
1351 /* check if completed */
1352 if (sq->cev_counter == 0) {
1353 sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
1358 /* send NOPs on next timeout */
1359 sq->cev_next_state = MLX5E_CEV_STATE_SEND_NOPS;
1364 callout_reset_curcpu(&sq->cev_callout, hz, mlx5e_sq_cev_timeout, sq);
1368 mlx5e_drain_sq(struct mlx5e_sq *sq)
1371 struct mlx5_core_dev *mdev= sq->priv->mdev;
1374 * Check if already stopped.
1376 * NOTE: Serialization of this function is managed by the
1377 * caller ensuring the priv's state lock is locked or in case
1378 * of rate limit support, a single thread manages drain and
1379 * resume of SQs. The "running" variable can therefore safely
1380 * be read without any locks.
1382 if (READ_ONCE(sq->running) == 0)
1385 /* don't put more packets into the SQ */
1386 WRITE_ONCE(sq->running, 0);
1388 /* serialize access to DMA rings */
1389 mtx_lock(&sq->lock);
1391 /* teardown event factor timer, if any */
1392 sq->cev_next_state = MLX5E_CEV_STATE_HOLD_NOPS;
1393 callout_stop(&sq->cev_callout);
1395 /* send dummy NOPs in order to flush the transmit ring */
1396 mlx5e_sq_send_nops_locked(sq, 1);
1397 mtx_unlock(&sq->lock);
1399 /* make sure it is safe to free the callout */
1400 callout_drain(&sq->cev_callout);
1402 /* wait till SQ is empty or link is down */
1403 mtx_lock(&sq->lock);
1404 while (sq->cc != sq->pc &&
1405 (sq->priv->media_status_last & IFM_ACTIVE) != 0 &&
1406 mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1407 mtx_unlock(&sq->lock);
1409 sq->cq.mcq.comp(&sq->cq.mcq);
1410 mtx_lock(&sq->lock);
1412 mtx_unlock(&sq->lock);
1414 /* error out remaining requests */
1415 error = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RDY, MLX5_SQC_STATE_ERR);
1418 "mlx5e_modify_sq() from RDY to ERR failed: %d\n", error);
1421 /* wait till SQ is empty */
1422 mtx_lock(&sq->lock);
1423 while (sq->cc != sq->pc &&
1424 mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1425 mtx_unlock(&sq->lock);
1427 sq->cq.mcq.comp(&sq->cq.mcq);
1428 mtx_lock(&sq->lock);
1430 mtx_unlock(&sq->lock);
1434 mlx5e_close_sq_wait(struct mlx5e_sq *sq)
1438 mlx5e_disable_sq(sq);
1439 mlx5e_destroy_sq(sq);
1443 mlx5e_create_cq(struct mlx5e_priv *priv,
1444 struct mlx5e_cq_param *param,
1445 struct mlx5e_cq *cq,
1446 mlx5e_cq_comp_t *comp,
1449 struct mlx5_core_dev *mdev = priv->mdev;
1450 struct mlx5_core_cq *mcq = &cq->mcq;
1456 param->wq.buf_numa_node = 0;
1457 param->wq.db_numa_node = 0;
1459 err = mlx5_cqwq_create(mdev, ¶m->wq, param->cqc, &cq->wq,
1464 mlx5_vector2eqn(mdev, eq_ix, &eqn_not_used, &irqn);
1467 mcq->set_ci_db = cq->wq_ctrl.db.db;
1468 mcq->arm_db = cq->wq_ctrl.db.db + 1;
1469 *mcq->set_ci_db = 0;
1471 mcq->vector = eq_ix;
1473 mcq->event = mlx5e_cq_error_event;
1475 mcq->uar = &priv->cq_uar;
1477 for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) {
1478 struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i);
1489 mlx5e_destroy_cq(struct mlx5e_cq *cq)
1491 mlx5_wq_destroy(&cq->wq_ctrl);
1495 mlx5e_enable_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param, int eq_ix)
1497 struct mlx5_core_cq *mcq = &cq->mcq;
1505 inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
1506 sizeof(u64) * cq->wq_ctrl.buf.npages;
1507 in = mlx5_vzalloc(inlen);
1511 cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
1513 memcpy(cqc, param->cqc, sizeof(param->cqc));
1515 mlx5_fill_page_array(&cq->wq_ctrl.buf,
1516 (__be64 *) MLX5_ADDR_OF(create_cq_in, in, pas));
1518 mlx5_vector2eqn(cq->priv->mdev, eq_ix, &eqn, &irqn_not_used);
1520 MLX5_SET(cqc, cqc, c_eqn, eqn);
1521 MLX5_SET(cqc, cqc, uar_page, mcq->uar->index);
1522 MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.buf.page_shift -
1524 MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
1526 err = mlx5_core_create_cq(cq->priv->mdev, mcq, in, inlen);
1533 mlx5e_cq_arm(cq, MLX5_GET_DOORBELL_LOCK(&cq->priv->doorbell_lock));
1539 mlx5e_disable_cq(struct mlx5e_cq *cq)
1542 mlx5_core_destroy_cq(cq->priv->mdev, &cq->mcq);
1546 mlx5e_open_cq(struct mlx5e_priv *priv,
1547 struct mlx5e_cq_param *param,
1548 struct mlx5e_cq *cq,
1549 mlx5e_cq_comp_t *comp,
1554 err = mlx5e_create_cq(priv, param, cq, comp, eq_ix);
1558 err = mlx5e_enable_cq(cq, param, eq_ix);
1560 goto err_destroy_cq;
1565 mlx5e_destroy_cq(cq);
1571 mlx5e_close_cq(struct mlx5e_cq *cq)
1573 mlx5e_disable_cq(cq);
1574 mlx5e_destroy_cq(cq);
1578 mlx5e_open_tx_cqs(struct mlx5e_channel *c,
1579 struct mlx5e_channel_param *cparam)
1584 for (tc = 0; tc < c->num_tc; tc++) {
1585 /* open completion queue */
1586 err = mlx5e_open_cq(c->priv, &cparam->tx_cq, &c->sq[tc].cq,
1587 &mlx5e_tx_cq_comp, c->ix);
1589 goto err_close_tx_cqs;
1594 for (tc--; tc >= 0; tc--)
1595 mlx5e_close_cq(&c->sq[tc].cq);
1601 mlx5e_close_tx_cqs(struct mlx5e_channel *c)
1605 for (tc = 0; tc < c->num_tc; tc++)
1606 mlx5e_close_cq(&c->sq[tc].cq);
1610 mlx5e_open_sqs(struct mlx5e_channel *c,
1611 struct mlx5e_channel_param *cparam)
1616 for (tc = 0; tc < c->num_tc; tc++) {
1617 err = mlx5e_open_sq(c, tc, &cparam->sq, &c->sq[tc]);
1625 for (tc--; tc >= 0; tc--)
1626 mlx5e_close_sq_wait(&c->sq[tc]);
1632 mlx5e_close_sqs_wait(struct mlx5e_channel *c)
1636 for (tc = 0; tc < c->num_tc; tc++)
1637 mlx5e_close_sq_wait(&c->sq[tc]);
1641 mlx5e_chan_mtx_init(struct mlx5e_channel *c)
1645 mtx_init(&c->rq.mtx, "mlx5rx", MTX_NETWORK_LOCK, MTX_DEF);
1647 callout_init_mtx(&c->rq.watchdog, &c->rq.mtx, 0);
1649 for (tc = 0; tc < c->num_tc; tc++) {
1650 struct mlx5e_sq *sq = c->sq + tc;
1652 mtx_init(&sq->lock, "mlx5tx",
1653 MTX_NETWORK_LOCK " TX", MTX_DEF);
1654 mtx_init(&sq->comp_lock, "mlx5comp",
1655 MTX_NETWORK_LOCK " TX", MTX_DEF);
1657 callout_init_mtx(&sq->cev_callout, &sq->lock, 0);
1659 sq->cev_factor = c->priv->params_ethtool.tx_completion_fact;
1661 /* ensure the TX completion event factor is not zero */
1662 if (sq->cev_factor == 0)
1668 mlx5e_chan_mtx_destroy(struct mlx5e_channel *c)
1672 mtx_destroy(&c->rq.mtx);
1674 for (tc = 0; tc < c->num_tc; tc++) {
1675 mtx_destroy(&c->sq[tc].lock);
1676 mtx_destroy(&c->sq[tc].comp_lock);
1681 mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
1682 struct mlx5e_channel_param *cparam,
1683 struct mlx5e_channel *c)
1687 memset(c, 0, sizeof(*c));
1692 c->mkey_be = cpu_to_be32(priv->mr.key);
1693 c->num_tc = priv->num_tc;
1696 mlx5e_chan_mtx_init(c);
1698 /* open transmit completion queue */
1699 err = mlx5e_open_tx_cqs(c, cparam);
1703 /* open receive completion queue */
1704 err = mlx5e_open_cq(c->priv, &cparam->rx_cq, &c->rq.cq,
1705 &mlx5e_rx_cq_comp, c->ix);
1707 goto err_close_tx_cqs;
1709 err = mlx5e_open_sqs(c, cparam);
1711 goto err_close_rx_cq;
1713 err = mlx5e_open_rq(c, &cparam->rq, &c->rq);
1717 /* poll receive queue initially */
1718 c->rq.cq.mcq.comp(&c->rq.cq.mcq);
1723 mlx5e_close_sqs_wait(c);
1726 mlx5e_close_cq(&c->rq.cq);
1729 mlx5e_close_tx_cqs(c);
1732 /* destroy mutexes */
1733 mlx5e_chan_mtx_destroy(c);
1738 mlx5e_close_channel(struct mlx5e_channel *c)
1740 mlx5e_close_rq(&c->rq);
1744 mlx5e_close_channel_wait(struct mlx5e_channel *c)
1746 mlx5e_close_rq_wait(&c->rq);
1747 mlx5e_close_sqs_wait(c);
1748 mlx5e_close_tx_cqs(c);
1749 /* destroy mutexes */
1750 mlx5e_chan_mtx_destroy(c);
1754 mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs)
1758 r = priv->params.hw_lro_en ? priv->params.lro_wqe_sz :
1759 MLX5E_SW2MB_MTU(priv->ifp->if_mtu);
1760 if (r > MJUM16BYTES)
1765 else if (r > MJUMPAGESIZE)
1767 else if (r > MCLBYTES)
1773 * n + 1 must be a power of two, because stride size must be.
1774 * Stride size is 16 * (n + 1), as the first segment is
1777 for (n = howmany(r, MLX5E_MAX_RX_BYTES); !powerof2(n + 1); n++)
1786 mlx5e_build_rq_param(struct mlx5e_priv *priv,
1787 struct mlx5e_rq_param *param)
1789 void *rqc = param->rqc;
1790 void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
1793 mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
1794 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
1795 MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN);
1796 MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe) +
1797 nsegs * sizeof(struct mlx5_wqe_data_seg)));
1798 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_rq_size);
1799 MLX5_SET(wq, wq, pd, priv->pdn);
1801 param->wq.buf_numa_node = 0;
1802 param->wq.db_numa_node = 0;
1803 param->wq.linear = 1;
1807 mlx5e_build_sq_param(struct mlx5e_priv *priv,
1808 struct mlx5e_sq_param *param)
1810 void *sqc = param->sqc;
1811 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
1813 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_sq_size);
1814 MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
1815 MLX5_SET(wq, wq, pd, priv->pdn);
1817 param->wq.buf_numa_node = 0;
1818 param->wq.db_numa_node = 0;
1819 param->wq.linear = 1;
1823 mlx5e_build_common_cq_param(struct mlx5e_priv *priv,
1824 struct mlx5e_cq_param *param)
1826 void *cqc = param->cqc;
1828 MLX5_SET(cqc, cqc, uar_page, priv->cq_uar.index);
1832 mlx5e_get_default_profile(struct mlx5e_priv *priv, int mode, struct net_dim_cq_moder *ptr)
1835 *ptr = net_dim_get_profile(mode, MLX5E_DIM_DEFAULT_PROFILE);
1837 /* apply LRO restrictions */
1838 if (priv->params.hw_lro_en &&
1839 ptr->pkts > MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO) {
1840 ptr->pkts = MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO;
1845 mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
1846 struct mlx5e_cq_param *param)
1848 struct net_dim_cq_moder curr;
1849 void *cqc = param->cqc;
1852 * We use MLX5_CQE_FORMAT_HASH because the RX hash mini CQE
1853 * format is more beneficial for FreeBSD use case.
1855 * Adding support for MLX5_CQE_FORMAT_CSUM will require changes
1856 * in mlx5e_decompress_cqe.
1858 if (priv->params.cqe_zipping_en) {
1859 MLX5_SET(cqc, cqc, mini_cqe_res_format, MLX5_CQE_FORMAT_HASH);
1860 MLX5_SET(cqc, cqc, cqe_compression_en, 1);
1863 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_rq_size);
1865 switch (priv->params.rx_cq_moderation_mode) {
1867 MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
1868 MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
1869 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1872 MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
1873 MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
1874 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
1875 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
1877 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1880 mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE, &curr);
1881 MLX5_SET(cqc, cqc, cq_period, curr.usec);
1882 MLX5_SET(cqc, cqc, cq_max_count, curr.pkts);
1883 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1886 mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE, &curr);
1887 MLX5_SET(cqc, cqc, cq_period, curr.usec);
1888 MLX5_SET(cqc, cqc, cq_max_count, curr.pkts);
1889 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
1890 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
1892 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1898 mlx5e_dim_build_cq_param(priv, param);
1900 mlx5e_build_common_cq_param(priv, param);
1904 mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
1905 struct mlx5e_cq_param *param)
1907 void *cqc = param->cqc;
1909 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_sq_size);
1910 MLX5_SET(cqc, cqc, cq_period, priv->params.tx_cq_moderation_usec);
1911 MLX5_SET(cqc, cqc, cq_max_count, priv->params.tx_cq_moderation_pkts);
1913 switch (priv->params.tx_cq_moderation_mode) {
1915 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1918 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
1919 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
1921 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1925 mlx5e_build_common_cq_param(priv, param);
1929 mlx5e_build_channel_param(struct mlx5e_priv *priv,
1930 struct mlx5e_channel_param *cparam)
1932 memset(cparam, 0, sizeof(*cparam));
1934 mlx5e_build_rq_param(priv, &cparam->rq);
1935 mlx5e_build_sq_param(priv, &cparam->sq);
1936 mlx5e_build_rx_cq_param(priv, &cparam->rx_cq);
1937 mlx5e_build_tx_cq_param(priv, &cparam->tx_cq);
1941 mlx5e_open_channels(struct mlx5e_priv *priv)
1943 struct mlx5e_channel_param cparam;
1948 mlx5e_build_channel_param(priv, &cparam);
1949 for (i = 0; i < priv->params.num_channels; i++) {
1950 err = mlx5e_open_channel(priv, i, &cparam, &priv->channel[i]);
1952 goto err_close_channels;
1955 for (j = 0; j < priv->params.num_channels; j++) {
1956 err = mlx5e_wait_for_min_rx_wqes(&priv->channel[j].rq);
1958 goto err_close_channels;
1965 mlx5e_close_channel(&priv->channel[i]);
1966 mlx5e_close_channel_wait(&priv->channel[i]);
1972 mlx5e_close_channels(struct mlx5e_priv *priv)
1976 for (i = 0; i < priv->params.num_channels; i++)
1977 mlx5e_close_channel(&priv->channel[i]);
1978 for (i = 0; i < priv->params.num_channels; i++)
1979 mlx5e_close_channel_wait(&priv->channel[i]);
1983 mlx5e_refresh_sq_params(struct mlx5e_priv *priv, struct mlx5e_sq *sq)
1986 if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
1989 switch (priv->params.tx_cq_moderation_mode) {
1992 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
1995 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
1999 return (mlx5_core_modify_cq_moderation_mode(priv->mdev, &sq->cq.mcq,
2000 priv->params.tx_cq_moderation_usec,
2001 priv->params.tx_cq_moderation_pkts,
2005 return (mlx5_core_modify_cq_moderation(priv->mdev, &sq->cq.mcq,
2006 priv->params.tx_cq_moderation_usec,
2007 priv->params.tx_cq_moderation_pkts));
2011 mlx5e_refresh_rq_params(struct mlx5e_priv *priv, struct mlx5e_rq *rq)
2014 if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
2019 switch (priv->params.rx_cq_moderation_mode) {
2022 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2023 dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
2026 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2027 dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
2031 /* tear down dynamic interrupt moderation */
2033 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
2034 mtx_unlock(&rq->mtx);
2036 /* wait for dynamic interrupt moderation work task, if any */
2037 cancel_work_sync(&rq->dim.work);
2039 if (priv->params.rx_cq_moderation_mode >= 2) {
2040 struct net_dim_cq_moder curr;
2042 mlx5e_get_default_profile(priv, dim_mode, &curr);
2044 retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2045 curr.usec, curr.pkts, cq_mode);
2047 /* set dynamic interrupt moderation mode and zero defaults */
2049 rq->dim.mode = dim_mode;
2051 rq->dim.profile_ix = MLX5E_DIM_DEFAULT_PROFILE;
2052 mtx_unlock(&rq->mtx);
2054 retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2055 priv->params.rx_cq_moderation_usec,
2056 priv->params.rx_cq_moderation_pkts,
2062 return (mlx5_core_modify_cq_moderation(priv->mdev, &rq->cq.mcq,
2063 priv->params.rx_cq_moderation_usec,
2064 priv->params.rx_cq_moderation_pkts));
2068 mlx5e_refresh_channel_params_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
2073 err = mlx5e_refresh_rq_params(priv, &c->rq);
2077 for (i = 0; i != c->num_tc; i++) {
2078 err = mlx5e_refresh_sq_params(priv, &c->sq[i]);
2087 mlx5e_refresh_channel_params(struct mlx5e_priv *priv)
2091 /* check if channels are closed */
2092 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2095 for (i = 0; i < priv->params.num_channels; i++) {
2098 err = mlx5e_refresh_channel_params_sub(priv, &priv->channel[i]);
2106 mlx5e_open_tis(struct mlx5e_priv *priv, int tc)
2108 struct mlx5_core_dev *mdev = priv->mdev;
2109 u32 in[MLX5_ST_SZ_DW(create_tis_in)];
2110 void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
2112 memset(in, 0, sizeof(in));
2114 MLX5_SET(tisc, tisc, prio, tc);
2115 MLX5_SET(tisc, tisc, transport_domain, priv->tdn);
2117 return (mlx5_core_create_tis(mdev, in, sizeof(in), &priv->tisn[tc]));
2121 mlx5e_close_tis(struct mlx5e_priv *priv, int tc)
2123 mlx5_core_destroy_tis(priv->mdev, priv->tisn[tc]);
2127 mlx5e_open_tises(struct mlx5e_priv *priv)
2129 int num_tc = priv->num_tc;
2133 for (tc = 0; tc < num_tc; tc++) {
2134 err = mlx5e_open_tis(priv, tc);
2136 goto err_close_tises;
2142 for (tc--; tc >= 0; tc--)
2143 mlx5e_close_tis(priv, tc);
2149 mlx5e_close_tises(struct mlx5e_priv *priv)
2151 int num_tc = priv->num_tc;
2154 for (tc = 0; tc < num_tc; tc++)
2155 mlx5e_close_tis(priv, tc);
2159 mlx5e_open_rqt(struct mlx5e_priv *priv)
2161 struct mlx5_core_dev *mdev = priv->mdev;
2163 u32 out[MLX5_ST_SZ_DW(create_rqt_out)] = {0};
2170 sz = 1 << priv->params.rx_hash_log_tbl_sz;
2172 inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
2173 in = mlx5_vzalloc(inlen);
2176 rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
2178 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
2179 MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
2181 for (i = 0; i < sz; i++) {
2184 ix = rss_get_indirection_to_bucket(ix);
2186 /* ensure we don't overflow */
2187 ix %= priv->params.num_channels;
2189 /* apply receive side scaling stride, if any */
2190 ix -= ix % (int)priv->params.channels_rsss;
2192 MLX5_SET(rqtc, rqtc, rq_num[i], priv->channel[ix].rq.rqn);
2195 MLX5_SET(create_rqt_in, in, opcode, MLX5_CMD_OP_CREATE_RQT);
2197 err = mlx5_cmd_exec(mdev, in, inlen, out, sizeof(out));
2199 priv->rqtn = MLX5_GET(create_rqt_out, out, rqtn);
2207 mlx5e_close_rqt(struct mlx5e_priv *priv)
2209 u32 in[MLX5_ST_SZ_DW(destroy_rqt_in)] = {0};
2210 u32 out[MLX5_ST_SZ_DW(destroy_rqt_out)] = {0};
2212 MLX5_SET(destroy_rqt_in, in, opcode, MLX5_CMD_OP_DESTROY_RQT);
2213 MLX5_SET(destroy_rqt_in, in, rqtn, priv->rqtn);
2215 mlx5_cmd_exec(priv->mdev, in, sizeof(in), out, sizeof(out));
2219 mlx5e_build_tir_ctx(struct mlx5e_priv *priv, u32 * tirc, int tt)
2221 void *hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
2224 MLX5_SET(tirc, tirc, transport_domain, priv->tdn);
2226 #define ROUGH_MAX_L2_L3_HDR_SZ 256
2228 #define MLX5_HASH_IP (MLX5_HASH_FIELD_SEL_SRC_IP |\
2229 MLX5_HASH_FIELD_SEL_DST_IP)
2231 #define MLX5_HASH_ALL (MLX5_HASH_FIELD_SEL_SRC_IP |\
2232 MLX5_HASH_FIELD_SEL_DST_IP |\
2233 MLX5_HASH_FIELD_SEL_L4_SPORT |\
2234 MLX5_HASH_FIELD_SEL_L4_DPORT)
2236 #define MLX5_HASH_IP_IPSEC_SPI (MLX5_HASH_FIELD_SEL_SRC_IP |\
2237 MLX5_HASH_FIELD_SEL_DST_IP |\
2238 MLX5_HASH_FIELD_SEL_IPSEC_SPI)
2240 if (priv->params.hw_lro_en) {
2241 MLX5_SET(tirc, tirc, lro_enable_mask,
2242 MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO |
2243 MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO);
2244 MLX5_SET(tirc, tirc, lro_max_msg_sz,
2245 (priv->params.lro_wqe_sz -
2246 ROUGH_MAX_L2_L3_HDR_SZ) >> 8);
2247 /* TODO: add the option to choose timer value dynamically */
2248 MLX5_SET(tirc, tirc, lro_timeout_period_usecs,
2249 MLX5_CAP_ETH(priv->mdev,
2250 lro_timer_supported_periods[2]));
2253 /* setup parameters for hashing TIR type, if any */
2256 MLX5_SET(tirc, tirc, disp_type,
2257 MLX5_TIRC_DISP_TYPE_DIRECT);
2258 MLX5_SET(tirc, tirc, inline_rqn,
2259 priv->channel[0].rq.rqn);
2262 MLX5_SET(tirc, tirc, disp_type,
2263 MLX5_TIRC_DISP_TYPE_INDIRECT);
2264 MLX5_SET(tirc, tirc, indirect_table,
2266 MLX5_SET(tirc, tirc, rx_hash_fn,
2267 MLX5_TIRC_RX_HASH_FN_HASH_TOEPLITZ);
2268 hkey = (__be32 *) MLX5_ADDR_OF(tirc, tirc, rx_hash_toeplitz_key);
2271 * The FreeBSD RSS implementation does currently not
2272 * support symmetric Toeplitz hashes:
2274 MLX5_SET(tirc, tirc, rx_hash_symmetric, 0);
2275 rss_getkey((uint8_t *)hkey);
2277 MLX5_SET(tirc, tirc, rx_hash_symmetric, 1);
2278 hkey[0] = cpu_to_be32(0xD181C62C);
2279 hkey[1] = cpu_to_be32(0xF7F4DB5B);
2280 hkey[2] = cpu_to_be32(0x1983A2FC);
2281 hkey[3] = cpu_to_be32(0x943E1ADB);
2282 hkey[4] = cpu_to_be32(0xD9389E6B);
2283 hkey[5] = cpu_to_be32(0xD1039C2C);
2284 hkey[6] = cpu_to_be32(0xA74499AD);
2285 hkey[7] = cpu_to_be32(0x593D56D9);
2286 hkey[8] = cpu_to_be32(0xF3253C06);
2287 hkey[9] = cpu_to_be32(0x2ADC1FFC);
2293 case MLX5E_TT_IPV4_TCP:
2294 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2295 MLX5_L3_PROT_TYPE_IPV4);
2296 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2297 MLX5_L4_PROT_TYPE_TCP);
2299 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4)) {
2300 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2304 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2308 case MLX5E_TT_IPV6_TCP:
2309 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2310 MLX5_L3_PROT_TYPE_IPV6);
2311 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2312 MLX5_L4_PROT_TYPE_TCP);
2314 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) {
2315 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2319 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2323 case MLX5E_TT_IPV4_UDP:
2324 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2325 MLX5_L3_PROT_TYPE_IPV4);
2326 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2327 MLX5_L4_PROT_TYPE_UDP);
2329 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4)) {
2330 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2334 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2338 case MLX5E_TT_IPV6_UDP:
2339 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2340 MLX5_L3_PROT_TYPE_IPV6);
2341 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2342 MLX5_L4_PROT_TYPE_UDP);
2344 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) {
2345 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2349 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2353 case MLX5E_TT_IPV4_IPSEC_AH:
2354 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2355 MLX5_L3_PROT_TYPE_IPV4);
2356 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2357 MLX5_HASH_IP_IPSEC_SPI);
2360 case MLX5E_TT_IPV6_IPSEC_AH:
2361 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2362 MLX5_L3_PROT_TYPE_IPV6);
2363 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2364 MLX5_HASH_IP_IPSEC_SPI);
2367 case MLX5E_TT_IPV4_IPSEC_ESP:
2368 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2369 MLX5_L3_PROT_TYPE_IPV4);
2370 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2371 MLX5_HASH_IP_IPSEC_SPI);
2374 case MLX5E_TT_IPV6_IPSEC_ESP:
2375 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2376 MLX5_L3_PROT_TYPE_IPV6);
2377 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2378 MLX5_HASH_IP_IPSEC_SPI);
2382 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2383 MLX5_L3_PROT_TYPE_IPV4);
2384 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2389 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2390 MLX5_L3_PROT_TYPE_IPV6);
2391 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2401 mlx5e_open_tir(struct mlx5e_priv *priv, int tt)
2403 struct mlx5_core_dev *mdev = priv->mdev;
2409 inlen = MLX5_ST_SZ_BYTES(create_tir_in);
2410 in = mlx5_vzalloc(inlen);
2413 tirc = MLX5_ADDR_OF(create_tir_in, in, tir_context);
2415 mlx5e_build_tir_ctx(priv, tirc, tt);
2417 err = mlx5_core_create_tir(mdev, in, inlen, &priv->tirn[tt]);
2425 mlx5e_close_tir(struct mlx5e_priv *priv, int tt)
2427 mlx5_core_destroy_tir(priv->mdev, priv->tirn[tt]);
2431 mlx5e_open_tirs(struct mlx5e_priv *priv)
2436 for (i = 0; i < MLX5E_NUM_TT; i++) {
2437 err = mlx5e_open_tir(priv, i);
2439 goto err_close_tirs;
2445 for (i--; i >= 0; i--)
2446 mlx5e_close_tir(priv, i);
2452 mlx5e_close_tirs(struct mlx5e_priv *priv)
2456 for (i = 0; i < MLX5E_NUM_TT; i++)
2457 mlx5e_close_tir(priv, i);
2461 * SW MTU does not include headers,
2462 * HW MTU includes all headers and checksums.
2465 mlx5e_set_dev_port_mtu(struct ifnet *ifp, int sw_mtu)
2467 struct mlx5e_priv *priv = ifp->if_softc;
2468 struct mlx5_core_dev *mdev = priv->mdev;
2472 hw_mtu = MLX5E_SW2HW_MTU(sw_mtu);
2474 err = mlx5_set_port_mtu(mdev, hw_mtu);
2476 if_printf(ifp, "%s: mlx5_set_port_mtu failed setting %d, err=%d\n",
2477 __func__, sw_mtu, err);
2481 /* Update vport context MTU */
2482 err = mlx5_set_vport_mtu(mdev, hw_mtu);
2484 if_printf(ifp, "%s: Failed updating vport context with MTU size, err=%d\n",
2488 ifp->if_mtu = sw_mtu;
2490 err = mlx5_query_vport_mtu(mdev, &hw_mtu);
2491 if (err || !hw_mtu) {
2492 /* fallback to port oper mtu */
2493 err = mlx5_query_port_oper_mtu(mdev, &hw_mtu);
2496 if_printf(ifp, "Query port MTU, after setting new "
2497 "MTU value, failed\n");
2499 } else if (MLX5E_HW2SW_MTU(hw_mtu) < sw_mtu) {
2501 if_printf(ifp, "Port MTU %d is smaller than "
2502 "ifp mtu %d\n", hw_mtu, sw_mtu);
2503 } else if (MLX5E_HW2SW_MTU(hw_mtu) > sw_mtu) {
2505 if_printf(ifp, "Port MTU %d is bigger than "
2506 "ifp mtu %d\n", hw_mtu, sw_mtu);
2508 priv->params_ethtool.hw_mtu = hw_mtu;
2514 mlx5e_open_locked(struct ifnet *ifp)
2516 struct mlx5e_priv *priv = ifp->if_softc;
2520 /* check if already opened */
2521 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
2525 if (rss_getnumbuckets() > priv->params.num_channels) {
2526 if_printf(ifp, "NOTE: There are more RSS buckets(%u) than "
2527 "channels(%u) available\n", rss_getnumbuckets(),
2528 priv->params.num_channels);
2531 err = mlx5e_open_tises(priv);
2533 if_printf(ifp, "%s: mlx5e_open_tises failed, %d\n",
2537 err = mlx5_vport_alloc_q_counter(priv->mdev,
2538 MLX5_INTERFACE_PROTOCOL_ETH, &set_id);
2540 if_printf(priv->ifp,
2541 "%s: mlx5_vport_alloc_q_counter failed: %d\n",
2543 goto err_close_tises;
2545 /* store counter set ID */
2546 priv->counter_set_id = set_id;
2548 err = mlx5e_open_channels(priv);
2550 if_printf(ifp, "%s: mlx5e_open_channels failed, %d\n",
2552 goto err_dalloc_q_counter;
2554 err = mlx5e_open_rqt(priv);
2556 if_printf(ifp, "%s: mlx5e_open_rqt failed, %d\n",
2558 goto err_close_channels;
2560 err = mlx5e_open_tirs(priv);
2562 if_printf(ifp, "%s: mlx5e_open_tir failed, %d\n",
2564 goto err_close_rqls;
2566 err = mlx5e_open_flow_table(priv);
2568 if_printf(ifp, "%s: mlx5e_open_flow_table failed, %d\n",
2570 goto err_close_tirs;
2572 err = mlx5e_add_all_vlan_rules(priv);
2574 if_printf(ifp, "%s: mlx5e_add_all_vlan_rules failed, %d\n",
2576 goto err_close_flow_table;
2578 set_bit(MLX5E_STATE_OPENED, &priv->state);
2580 mlx5e_update_carrier(priv);
2581 mlx5e_set_rx_mode_core(priv);
2585 err_close_flow_table:
2586 mlx5e_close_flow_table(priv);
2589 mlx5e_close_tirs(priv);
2592 mlx5e_close_rqt(priv);
2595 mlx5e_close_channels(priv);
2597 err_dalloc_q_counter:
2598 mlx5_vport_dealloc_q_counter(priv->mdev,
2599 MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
2602 mlx5e_close_tises(priv);
2608 mlx5e_open(void *arg)
2610 struct mlx5e_priv *priv = arg;
2613 if (mlx5_set_port_status(priv->mdev, MLX5_PORT_UP))
2614 if_printf(priv->ifp,
2615 "%s: Setting port status to up failed\n",
2618 mlx5e_open_locked(priv->ifp);
2619 priv->ifp->if_drv_flags |= IFF_DRV_RUNNING;
2624 mlx5e_close_locked(struct ifnet *ifp)
2626 struct mlx5e_priv *priv = ifp->if_softc;
2628 /* check if already closed */
2629 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2632 clear_bit(MLX5E_STATE_OPENED, &priv->state);
2634 mlx5e_set_rx_mode_core(priv);
2635 mlx5e_del_all_vlan_rules(priv);
2636 if_link_state_change(priv->ifp, LINK_STATE_DOWN);
2637 mlx5e_close_flow_table(priv);
2638 mlx5e_close_tirs(priv);
2639 mlx5e_close_rqt(priv);
2640 mlx5e_close_channels(priv);
2641 mlx5_vport_dealloc_q_counter(priv->mdev,
2642 MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
2643 mlx5e_close_tises(priv);
2648 #if (__FreeBSD_version >= 1100000)
2650 mlx5e_get_counter(struct ifnet *ifp, ift_counter cnt)
2652 struct mlx5e_priv *priv = ifp->if_softc;
2655 /* PRIV_LOCK(priv); XXX not allowed */
2657 case IFCOUNTER_IPACKETS:
2658 retval = priv->stats.vport.rx_packets;
2660 case IFCOUNTER_IERRORS:
2661 retval = priv->stats.pport.in_range_len_errors +
2662 priv->stats.pport.out_of_range_len +
2663 priv->stats.pport.too_long_errors +
2664 priv->stats.pport.check_seq_err +
2665 priv->stats.pport.alignment_err;
2667 case IFCOUNTER_IQDROPS:
2668 retval = priv->stats.vport.rx_out_of_buffer;
2670 case IFCOUNTER_OPACKETS:
2671 retval = priv->stats.vport.tx_packets;
2673 case IFCOUNTER_OERRORS:
2674 retval = priv->stats.port_stats_debug.out_discards;
2676 case IFCOUNTER_IBYTES:
2677 retval = priv->stats.vport.rx_bytes;
2679 case IFCOUNTER_OBYTES:
2680 retval = priv->stats.vport.tx_bytes;
2682 case IFCOUNTER_IMCASTS:
2683 retval = priv->stats.vport.rx_multicast_packets;
2685 case IFCOUNTER_OMCASTS:
2686 retval = priv->stats.vport.tx_multicast_packets;
2688 case IFCOUNTER_OQDROPS:
2689 retval = priv->stats.vport.tx_queue_dropped;
2691 case IFCOUNTER_COLLISIONS:
2692 retval = priv->stats.pport.collisions;
2695 retval = if_get_counter_default(ifp, cnt);
2698 /* PRIV_UNLOCK(priv); XXX not allowed */
2704 mlx5e_set_rx_mode(struct ifnet *ifp)
2706 struct mlx5e_priv *priv = ifp->if_softc;
2708 queue_work(priv->wq, &priv->set_rx_mode_work);
2712 mlx5e_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2714 struct mlx5e_priv *priv;
2716 struct ifi2creq i2c;
2725 priv = ifp->if_softc;
2727 /* check if detaching */
2728 if (priv == NULL || priv->gone != 0)
2733 ifr = (struct ifreq *)data;
2736 mlx5_query_port_max_mtu(priv->mdev, &max_mtu);
2738 if (ifr->ifr_mtu >= MLX5E_MTU_MIN &&
2739 ifr->ifr_mtu <= MIN(MLX5E_MTU_MAX, max_mtu)) {
2742 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2744 mlx5e_close_locked(ifp);
2747 mlx5e_set_dev_port_mtu(ifp, ifr->ifr_mtu);
2750 mlx5e_open_locked(ifp);
2753 if_printf(ifp, "Invalid MTU value. Min val: %d, Max val: %d\n",
2754 MLX5E_MTU_MIN, MIN(MLX5E_MTU_MAX, max_mtu));
2759 if ((ifp->if_flags & IFF_UP) &&
2760 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
2761 mlx5e_set_rx_mode(ifp);
2765 if (ifp->if_flags & IFF_UP) {
2766 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2767 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2768 mlx5e_open_locked(ifp);
2769 ifp->if_drv_flags |= IFF_DRV_RUNNING;
2770 mlx5_set_port_status(priv->mdev, MLX5_PORT_UP);
2773 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
2774 mlx5_set_port_status(priv->mdev,
2776 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
2777 mlx5e_close_locked(ifp);
2778 mlx5e_update_carrier(priv);
2779 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2786 mlx5e_set_rx_mode(ifp);
2791 ifr = (struct ifreq *)data;
2792 error = ifmedia_ioctl(ifp, ifr, &priv->media, command);
2795 ifr = (struct ifreq *)data;
2797 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2799 if (mask & IFCAP_TXCSUM) {
2800 ifp->if_capenable ^= IFCAP_TXCSUM;
2801 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
2803 if (IFCAP_TSO4 & ifp->if_capenable &&
2804 !(IFCAP_TXCSUM & ifp->if_capenable)) {
2805 ifp->if_capenable &= ~IFCAP_TSO4;
2806 ifp->if_hwassist &= ~CSUM_IP_TSO;
2808 "tso4 disabled due to -txcsum.\n");
2811 if (mask & IFCAP_TXCSUM_IPV6) {
2812 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
2813 ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
2815 if (IFCAP_TSO6 & ifp->if_capenable &&
2816 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2817 ifp->if_capenable &= ~IFCAP_TSO6;
2818 ifp->if_hwassist &= ~CSUM_IP6_TSO;
2820 "tso6 disabled due to -txcsum6.\n");
2823 if (mask & IFCAP_RXCSUM)
2824 ifp->if_capenable ^= IFCAP_RXCSUM;
2825 if (mask & IFCAP_RXCSUM_IPV6)
2826 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
2827 if (mask & IFCAP_TSO4) {
2828 if (!(IFCAP_TSO4 & ifp->if_capenable) &&
2829 !(IFCAP_TXCSUM & ifp->if_capenable)) {
2830 if_printf(ifp, "enable txcsum first.\n");
2834 ifp->if_capenable ^= IFCAP_TSO4;
2835 ifp->if_hwassist ^= CSUM_IP_TSO;
2837 if (mask & IFCAP_TSO6) {
2838 if (!(IFCAP_TSO6 & ifp->if_capenable) &&
2839 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2840 if_printf(ifp, "enable txcsum6 first.\n");
2844 ifp->if_capenable ^= IFCAP_TSO6;
2845 ifp->if_hwassist ^= CSUM_IP6_TSO;
2847 if (mask & IFCAP_VLAN_HWFILTER) {
2848 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2849 mlx5e_disable_vlan_filter(priv);
2851 mlx5e_enable_vlan_filter(priv);
2853 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
2855 if (mask & IFCAP_VLAN_HWTAGGING)
2856 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2857 if (mask & IFCAP_WOL_MAGIC)
2858 ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2860 VLAN_CAPABILITIES(ifp);
2861 /* turn off LRO means also turn of HW LRO - if it's on */
2862 if (mask & IFCAP_LRO) {
2863 int was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2864 bool need_restart = false;
2866 ifp->if_capenable ^= IFCAP_LRO;
2868 /* figure out if updating HW LRO is needed */
2869 if (!(ifp->if_capenable & IFCAP_LRO)) {
2870 if (priv->params.hw_lro_en) {
2871 priv->params.hw_lro_en = false;
2872 need_restart = true;
2875 if (priv->params.hw_lro_en == false &&
2876 priv->params_ethtool.hw_lro != 0) {
2877 priv->params.hw_lro_en = true;
2878 need_restart = true;
2881 if (was_opened && need_restart) {
2882 mlx5e_close_locked(ifp);
2883 mlx5e_open_locked(ifp);
2891 ifr = (struct ifreq *)data;
2894 * Copy from the user-space address ifr_data to the
2895 * kernel-space address i2c
2897 error = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
2901 if (i2c.len > sizeof(i2c.data)) {
2907 /* Get module_num which is required for the query_eeprom */
2908 error = mlx5_query_module_num(priv->mdev, &module_num);
2910 if_printf(ifp, "Query module num failed, eeprom "
2911 "reading is not supported\n");
2915 /* Check if module is present before doing an access */
2916 module_status = mlx5_query_module_status(priv->mdev, module_num);
2917 if (module_status != MLX5_MODULE_STATUS_PLUGGED_ENABLED &&
2918 module_status != MLX5_MODULE_STATUS_PLUGGED_DISABLED) {
2923 * Currently 0XA0 and 0xA2 are the only addresses permitted.
2924 * The internal conversion is as follows:
2926 if (i2c.dev_addr == 0xA0)
2927 read_addr = MLX5E_I2C_ADDR_LOW;
2928 else if (i2c.dev_addr == 0xA2)
2929 read_addr = MLX5E_I2C_ADDR_HIGH;
2931 if_printf(ifp, "Query eeprom failed, "
2932 "Invalid Address: %X\n", i2c.dev_addr);
2936 error = mlx5_query_eeprom(priv->mdev,
2937 read_addr, MLX5E_EEPROM_LOW_PAGE,
2938 (uint32_t)i2c.offset, (uint32_t)i2c.len, module_num,
2939 (uint32_t *)i2c.data, &size_read);
2941 if_printf(ifp, "Query eeprom failed, eeprom "
2942 "reading is not supported\n");
2947 if (i2c.len > MLX5_EEPROM_MAX_BYTES) {
2948 error = mlx5_query_eeprom(priv->mdev,
2949 read_addr, MLX5E_EEPROM_LOW_PAGE,
2950 (uint32_t)(i2c.offset + size_read),
2951 (uint32_t)(i2c.len - size_read), module_num,
2952 (uint32_t *)(i2c.data + size_read), &size_read);
2955 if_printf(ifp, "Query eeprom failed, eeprom "
2956 "reading is not supported\n");
2961 error = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
2967 error = ether_ioctl(ifp, command, data);
2974 mlx5e_check_required_hca_cap(struct mlx5_core_dev *mdev)
2977 * TODO: uncoment once FW really sets all these bits if
2978 * (!mdev->caps.eth.rss_ind_tbl_cap || !mdev->caps.eth.csum_cap ||
2979 * !mdev->caps.eth.max_lso_cap || !mdev->caps.eth.vlan_cap ||
2980 * !(mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_SCQE_BRK_MOD)) return
2984 /* TODO: add more must-to-have features */
2986 if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
2993 mlx5e_get_max_inline_cap(struct mlx5_core_dev *mdev)
2995 uint32_t bf_buf_size = (1U << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2U;
2997 bf_buf_size -= sizeof(struct mlx5e_tx_wqe) - 2;
2999 /* verify against driver hardware limit */
3000 if (bf_buf_size > MLX5E_MAX_TX_INLINE)
3001 bf_buf_size = MLX5E_MAX_TX_INLINE;
3003 return (bf_buf_size);
3007 mlx5e_build_ifp_priv(struct mlx5_core_dev *mdev,
3008 struct mlx5e_priv *priv,
3009 int num_comp_vectors)
3014 * TODO: Consider link speed for setting "log_sq_size",
3015 * "log_rq_size" and "cq_moderation_xxx":
3017 priv->params.log_sq_size =
3018 MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
3019 priv->params.log_rq_size =
3020 MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
3021 priv->params.rx_cq_moderation_usec =
3022 MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
3023 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE :
3024 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
3025 priv->params.rx_cq_moderation_mode =
3026 MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ? 1 : 0;
3027 priv->params.rx_cq_moderation_pkts =
3028 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS;
3029 priv->params.tx_cq_moderation_usec =
3030 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC;
3031 priv->params.tx_cq_moderation_pkts =
3032 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
3033 priv->params.min_rx_wqes =
3034 MLX5E_PARAMS_DEFAULT_MIN_RX_WQES;
3035 priv->params.rx_hash_log_tbl_sz =
3036 (order_base_2(num_comp_vectors) >
3037 MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ) ?
3038 order_base_2(num_comp_vectors) :
3039 MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ;
3040 priv->params.num_tc = 1;
3041 priv->params.default_vlan_prio = 0;
3042 priv->counter_set_id = -1;
3043 priv->params.tx_max_inline = mlx5e_get_max_inline_cap(mdev);
3045 err = mlx5_query_min_inline(mdev, &priv->params.tx_min_inline_mode);
3050 * hw lro is currently defaulted to off. when it won't anymore we
3051 * will consider the HW capability: "!!MLX5_CAP_ETH(mdev, lro_cap)"
3053 priv->params.hw_lro_en = false;
3054 priv->params.lro_wqe_sz = MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ;
3057 * CQE zipping is currently defaulted to off. when it won't
3058 * anymore we will consider the HW capability:
3059 * "!!MLX5_CAP_GEN(mdev, cqe_compression)"
3061 priv->params.cqe_zipping_en = false;
3064 priv->params.num_channels = num_comp_vectors;
3065 priv->params.channels_rsss = 1;
3066 priv->order_base_2_num_channels = order_base_2(num_comp_vectors);
3067 priv->queue_mapping_channel_mask =
3068 roundup_pow_of_two(num_comp_vectors) - 1;
3069 priv->num_tc = priv->params.num_tc;
3070 priv->default_vlan_prio = priv->params.default_vlan_prio;
3072 INIT_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
3073 INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
3074 INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
3080 mlx5e_create_mkey(struct mlx5e_priv *priv, u32 pdn,
3081 struct mlx5_core_mr *mkey)
3083 struct ifnet *ifp = priv->ifp;
3084 struct mlx5_core_dev *mdev = priv->mdev;
3085 int inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
3090 in = mlx5_vzalloc(inlen);
3092 if_printf(ifp, "%s: failed to allocate inbox\n", __func__);
3096 mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
3097 MLX5_SET(mkc, mkc, access_mode, MLX5_ACCESS_MODE_PA);
3098 MLX5_SET(mkc, mkc, lw, 1);
3099 MLX5_SET(mkc, mkc, lr, 1);
3101 MLX5_SET(mkc, mkc, pd, pdn);
3102 MLX5_SET(mkc, mkc, length64, 1);
3103 MLX5_SET(mkc, mkc, qpn, 0xffffff);
3105 err = mlx5_core_create_mkey(mdev, mkey, in, inlen);
3107 if_printf(ifp, "%s: mlx5_core_create_mkey failed, %d\n",
3114 static const char *mlx5e_vport_stats_desc[] = {
3115 MLX5E_VPORT_STATS(MLX5E_STATS_DESC)
3118 static const char *mlx5e_pport_stats_desc[] = {
3119 MLX5E_PPORT_STATS(MLX5E_STATS_DESC)
3123 mlx5e_priv_mtx_init(struct mlx5e_priv *priv)
3125 mtx_init(&priv->async_events_mtx, "mlx5async", MTX_NETWORK_LOCK, MTX_DEF);
3126 sx_init(&priv->state_lock, "mlx5state");
3127 callout_init_mtx(&priv->watchdog, &priv->async_events_mtx, 0);
3128 MLX5_INIT_DOORBELL_LOCK(&priv->doorbell_lock);
3132 mlx5e_priv_mtx_destroy(struct mlx5e_priv *priv)
3134 mtx_destroy(&priv->async_events_mtx);
3135 sx_destroy(&priv->state_lock);
3139 sysctl_firmware(SYSCTL_HANDLER_ARGS)
3142 * %d.%d%.d the string format.
3143 * fw_rev_{maj,min,sub} return u16, 2^16 = 65536.
3144 * We need at most 5 chars to store that.
3145 * It also has: two "." and NULL at the end, which means we need 18
3146 * (5*3 + 3) chars at most.
3149 struct mlx5e_priv *priv = arg1;
3152 snprintf(fw, sizeof(fw), "%d.%d.%d", fw_rev_maj(priv->mdev), fw_rev_min(priv->mdev),
3153 fw_rev_sub(priv->mdev));
3154 error = sysctl_handle_string(oidp, fw, sizeof(fw), req);
3159 mlx5e_disable_tx_dma(struct mlx5e_channel *ch)
3163 for (i = 0; i < ch->num_tc; i++)
3164 mlx5e_drain_sq(&ch->sq[i]);
3168 mlx5e_reset_sq_doorbell_record(struct mlx5e_sq *sq)
3171 sq->doorbell.d32[0] = cpu_to_be32(MLX5_OPCODE_NOP);
3172 sq->doorbell.d32[1] = cpu_to_be32(sq->sqn << 8);
3173 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
3174 sq->doorbell.d64 = 0;
3178 mlx5e_resume_sq(struct mlx5e_sq *sq)
3182 /* check if already enabled */
3183 if (READ_ONCE(sq->running) != 0)
3186 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_ERR,
3187 MLX5_SQC_STATE_RST);
3190 "mlx5e_modify_sq() from ERR to RST failed: %d\n", err);
3196 /* reset doorbell prior to moving from RST to RDY */
3197 mlx5e_reset_sq_doorbell_record(sq);
3199 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST,
3200 MLX5_SQC_STATE_RDY);
3203 "mlx5e_modify_sq() from RST to RDY failed: %d\n", err);
3206 sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
3207 WRITE_ONCE(sq->running, 1);
3211 mlx5e_enable_tx_dma(struct mlx5e_channel *ch)
3215 for (i = 0; i < ch->num_tc; i++)
3216 mlx5e_resume_sq(&ch->sq[i]);
3220 mlx5e_disable_rx_dma(struct mlx5e_channel *ch)
3222 struct mlx5e_rq *rq = &ch->rq;
3227 callout_stop(&rq->watchdog);
3228 mtx_unlock(&rq->mtx);
3230 callout_drain(&rq->watchdog);
3232 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
3235 "mlx5e_modify_rq() from RDY to RST failed: %d\n", err);
3238 while (!mlx5_wq_ll_is_empty(&rq->wq)) {
3240 rq->cq.mcq.comp(&rq->cq.mcq);
3244 * Transitioning into RST state will allow the FW to track less ERR state queues,
3245 * thus reducing the recv queue flushing time
3247 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_ERR, MLX5_RQC_STATE_RST);
3250 "mlx5e_modify_rq() from ERR to RST failed: %d\n", err);
3255 mlx5e_enable_rx_dma(struct mlx5e_channel *ch)
3257 struct mlx5e_rq *rq = &ch->rq;
3261 mlx5_wq_ll_update_db_record(&rq->wq);
3262 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
3265 "mlx5e_modify_rq() from RST to RDY failed: %d\n", err);
3270 rq->cq.mcq.comp(&rq->cq.mcq);
3274 mlx5e_modify_tx_dma(struct mlx5e_priv *priv, uint8_t value)
3278 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3281 for (i = 0; i < priv->params.num_channels; i++) {
3283 mlx5e_disable_tx_dma(&priv->channel[i]);
3285 mlx5e_enable_tx_dma(&priv->channel[i]);
3290 mlx5e_modify_rx_dma(struct mlx5e_priv *priv, uint8_t value)
3294 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3297 for (i = 0; i < priv->params.num_channels; i++) {
3299 mlx5e_disable_rx_dma(&priv->channel[i]);
3301 mlx5e_enable_rx_dma(&priv->channel[i]);
3306 mlx5e_add_hw_stats(struct mlx5e_priv *priv)
3308 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3309 OID_AUTO, "fw_version", CTLTYPE_STRING | CTLFLAG_RD, priv, 0,
3310 sysctl_firmware, "A", "HCA firmware version");
3312 SYSCTL_ADD_STRING(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3313 OID_AUTO, "board_id", CTLFLAG_RD, priv->mdev->board_id, 0,
3318 mlx5e_sysctl_tx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3320 struct mlx5e_priv *priv = arg1;
3321 uint8_t temp[MLX5E_MAX_PRIORITY];
3328 tx_pfc = priv->params.tx_priority_flow_control;
3330 for (i = 0; i != MLX5E_MAX_PRIORITY; i++)
3331 temp[i] = (tx_pfc >> i) & 1;
3333 err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY);
3334 if (err || !req->newptr)
3336 err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY);
3340 priv->params.tx_priority_flow_control = 0;
3342 /* range check input value */
3343 for (i = 0; i != MLX5E_MAX_PRIORITY; i++) {
3348 priv->params.tx_priority_flow_control |= (temp[i] << i);
3351 /* check if update is required */
3352 if (tx_pfc != priv->params.tx_priority_flow_control)
3353 err = -mlx5e_set_port_pfc(priv);
3356 priv->params.tx_priority_flow_control= tx_pfc;
3363 mlx5e_sysctl_rx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3365 struct mlx5e_priv *priv = arg1;
3366 uint8_t temp[MLX5E_MAX_PRIORITY];
3373 rx_pfc = priv->params.rx_priority_flow_control;
3375 for (i = 0; i != MLX5E_MAX_PRIORITY; i++)
3376 temp[i] = (rx_pfc >> i) & 1;
3378 err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY);
3379 if (err || !req->newptr)
3381 err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY);
3385 priv->params.rx_priority_flow_control = 0;
3387 /* range check input value */
3388 for (i = 0; i != MLX5E_MAX_PRIORITY; i++) {
3393 priv->params.rx_priority_flow_control |= (temp[i] << i);
3396 /* check if update is required */
3397 if (rx_pfc != priv->params.rx_priority_flow_control)
3398 err = -mlx5e_set_port_pfc(priv);
3401 priv->params.rx_priority_flow_control= rx_pfc;
3408 mlx5e_setup_pauseframes(struct mlx5e_priv *priv)
3410 #if (__FreeBSD_version < 1100000)
3415 /* enable pauseframes by default */
3416 priv->params.tx_pauseframe_control = 1;
3417 priv->params.rx_pauseframe_control = 1;
3419 /* disable ports flow control, PFC, by default */
3420 priv->params.tx_priority_flow_control = 0;
3421 priv->params.rx_priority_flow_control = 0;
3423 #if (__FreeBSD_version < 1100000)
3424 /* compute path for sysctl */
3425 snprintf(path, sizeof(path), "dev.mce.%d.tx_pauseframe_control",
3426 device_get_unit(priv->mdev->pdev->dev.bsddev));
3428 /* try to fetch tunable, if any */
3429 TUNABLE_INT_FETCH(path, &priv->params.tx_pauseframe_control);
3431 /* compute path for sysctl */
3432 snprintf(path, sizeof(path), "dev.mce.%d.rx_pauseframe_control",
3433 device_get_unit(priv->mdev->pdev->dev.bsddev));
3435 /* try to fetch tunable, if any */
3436 TUNABLE_INT_FETCH(path, &priv->params.rx_pauseframe_control);
3439 /* register pauseframe SYSCTLs */
3440 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3441 OID_AUTO, "tx_pauseframe_control", CTLFLAG_RDTUN,
3442 &priv->params.tx_pauseframe_control, 0,
3443 "Set to enable TX pause frames. Clear to disable.");
3445 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3446 OID_AUTO, "rx_pauseframe_control", CTLFLAG_RDTUN,
3447 &priv->params.rx_pauseframe_control, 0,
3448 "Set to enable RX pause frames. Clear to disable.");
3450 /* register priority flow control, PFC, SYSCTLs */
3451 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3452 OID_AUTO, "tx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN |
3453 CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_tx_priority_flow_control, "CU",
3454 "Set to enable TX ports flow control frames for priorities 0..7. Clear to disable.");
3456 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3457 OID_AUTO, "rx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN |
3458 CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_rx_priority_flow_control, "CU",
3459 "Set to enable RX ports flow control frames for priorities 0..7. Clear to disable.");
3464 priv->params.tx_pauseframe_control =
3465 priv->params.tx_pauseframe_control ? 1 : 0;
3466 priv->params.rx_pauseframe_control =
3467 priv->params.rx_pauseframe_control ? 1 : 0;
3469 /* update firmware */
3470 error = mlx5e_set_port_pause_and_pfc(priv);
3471 if (error == -EINVAL) {
3472 if_printf(priv->ifp,
3473 "Global pauseframes must be disabled before enabling PFC.\n");
3474 priv->params.rx_priority_flow_control = 0;
3475 priv->params.tx_priority_flow_control = 0;
3477 /* update firmware */
3478 (void) mlx5e_set_port_pause_and_pfc(priv);
3484 mlx5e_create_ifp(struct mlx5_core_dev *mdev)
3487 struct mlx5e_priv *priv;
3488 u8 dev_addr[ETHER_ADDR_LEN] __aligned(4);
3489 struct sysctl_oid_list *child;
3490 int ncv = mdev->priv.eq_table.num_comp_vectors;
3496 if (mlx5e_check_required_hca_cap(mdev)) {
3497 mlx5_core_dbg(mdev, "mlx5e_check_required_hca_cap() failed\n");
3501 * Try to allocate the priv and make room for worst-case
3502 * number of channel structures:
3504 priv = malloc(sizeof(*priv) +
3505 (sizeof(priv->channel[0]) * mdev->priv.eq_table.num_comp_vectors),
3506 M_MLX5EN, M_WAITOK | M_ZERO);
3507 mlx5e_priv_mtx_init(priv);
3509 ifp = priv->ifp = if_alloc(IFT_ETHER);
3511 mlx5_core_err(mdev, "if_alloc() failed\n");
3514 ifp->if_softc = priv;
3515 if_initname(ifp, "mce", device_get_unit(mdev->pdev->dev.bsddev));
3516 ifp->if_mtu = ETHERMTU;
3517 ifp->if_init = mlx5e_open;
3518 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
3519 ifp->if_ioctl = mlx5e_ioctl;
3520 ifp->if_transmit = mlx5e_xmit;
3521 ifp->if_qflush = if_qflush;
3522 #if (__FreeBSD_version >= 1100000)
3523 ifp->if_get_counter = mlx5e_get_counter;
3525 ifp->if_snd.ifq_maxlen = ifqmaxlen;
3527 * Set driver features
3529 ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6;
3530 ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
3531 ifp->if_capabilities |= IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWFILTER;
3532 ifp->if_capabilities |= IFCAP_LINKSTATE | IFCAP_JUMBO_MTU;
3533 ifp->if_capabilities |= IFCAP_LRO;
3534 ifp->if_capabilities |= IFCAP_TSO | IFCAP_VLAN_HWTSO;
3535 ifp->if_capabilities |= IFCAP_HWSTATS;
3537 /* set TSO limits so that we don't have to drop TX packets */
3538 ifp->if_hw_tsomax = MLX5E_MAX_TX_PAYLOAD_SIZE - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
3539 ifp->if_hw_tsomaxsegcount = MLX5E_MAX_TX_MBUF_FRAGS - 1 /* hdr */;
3540 ifp->if_hw_tsomaxsegsize = MLX5E_MAX_TX_MBUF_SIZE;
3542 ifp->if_capenable = ifp->if_capabilities;
3543 ifp->if_hwassist = 0;
3544 if (ifp->if_capenable & IFCAP_TSO)
3545 ifp->if_hwassist |= CSUM_TSO;
3546 if (ifp->if_capenable & IFCAP_TXCSUM)
3547 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP | CSUM_IP);
3548 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3549 ifp->if_hwassist |= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
3551 /* ifnet sysctl tree */
3552 sysctl_ctx_init(&priv->sysctl_ctx);
3553 priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_dev),
3554 OID_AUTO, ifp->if_dname, CTLFLAG_RD, 0, "MLX5 ethernet - interface name");
3555 if (priv->sysctl_ifnet == NULL) {
3556 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3557 goto err_free_sysctl;
3559 snprintf(unit, sizeof(unit), "%d", ifp->if_dunit);
3560 priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3561 OID_AUTO, unit, CTLFLAG_RD, 0, "MLX5 ethernet - interface unit");
3562 if (priv->sysctl_ifnet == NULL) {
3563 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3564 goto err_free_sysctl;
3567 /* HW sysctl tree */
3568 child = SYSCTL_CHILDREN(device_get_sysctl_tree(mdev->pdev->dev.bsddev));
3569 priv->sysctl_hw = SYSCTL_ADD_NODE(&priv->sysctl_ctx, child,
3570 OID_AUTO, "hw", CTLFLAG_RD, 0, "MLX5 ethernet dev hw");
3571 if (priv->sysctl_hw == NULL) {
3572 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3573 goto err_free_sysctl;
3576 err = mlx5e_build_ifp_priv(mdev, priv, ncv);
3578 mlx5_core_err(mdev, "mlx5e_build_ifp_priv() failed (%d)\n", err);
3579 goto err_free_sysctl;
3582 snprintf(unit, sizeof(unit), "mce%u_wq",
3583 device_get_unit(mdev->pdev->dev.bsddev));
3584 priv->wq = alloc_workqueue(unit, 0, 1);
3585 if (priv->wq == NULL) {
3586 if_printf(ifp, "%s: alloc_workqueue failed\n", __func__);
3587 goto err_free_sysctl;
3590 err = mlx5_alloc_map_uar(mdev, &priv->cq_uar);
3592 if_printf(ifp, "%s: mlx5_alloc_map_uar failed, %d\n",
3596 err = mlx5_core_alloc_pd(mdev, &priv->pdn);
3598 if_printf(ifp, "%s: mlx5_core_alloc_pd failed, %d\n",
3600 goto err_unmap_free_uar;
3602 err = mlx5_alloc_transport_domain(mdev, &priv->tdn);
3604 if_printf(ifp, "%s: mlx5_alloc_transport_domain failed, %d\n",
3606 goto err_dealloc_pd;
3608 err = mlx5e_create_mkey(priv, priv->pdn, &priv->mr);
3610 if_printf(ifp, "%s: mlx5e_create_mkey failed, %d\n",
3612 goto err_dealloc_transport_domain;
3614 mlx5_query_nic_vport_mac_address(priv->mdev, 0, dev_addr);
3616 /* check if we should generate a random MAC address */
3617 if (MLX5_CAP_GEN(priv->mdev, vport_group_manager) == 0 &&
3618 is_zero_ether_addr(dev_addr)) {
3619 random_ether_addr(dev_addr);
3620 if_printf(ifp, "Assigned random MAC address\n");
3623 /* set default MTU */
3624 mlx5e_set_dev_port_mtu(ifp, ifp->if_mtu);
3626 /* Set default media status */
3627 priv->media_status_last = IFM_AVALID;
3628 priv->media_active_last = IFM_ETHER | IFM_AUTO |
3629 IFM_ETH_RXPAUSE | IFM_FDX;
3631 /* setup default pauseframes configuration */
3632 mlx5e_setup_pauseframes(priv);
3634 err = mlx5_query_port_proto_cap(mdev, ð_proto_cap, MLX5_PTYS_EN);
3637 if_printf(ifp, "%s: Query port media capability failed, %d\n",
3641 /* Setup supported medias */
3642 ifmedia_init(&priv->media, IFM_IMASK | IFM_ETH_FMASK,
3643 mlx5e_media_change, mlx5e_media_status);
3645 for (i = 0; i < MLX5E_LINK_MODES_NUMBER; ++i) {
3646 if (mlx5e_mode_table[i].baudrate == 0)
3648 if (MLX5E_PROT_MASK(i) & eth_proto_cap) {
3649 ifmedia_add(&priv->media,
3650 mlx5e_mode_table[i].subtype |
3651 IFM_ETHER, 0, NULL);
3652 ifmedia_add(&priv->media,
3653 mlx5e_mode_table[i].subtype |
3654 IFM_ETHER | IFM_FDX |
3655 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3659 /* Additional supported medias */
3660 ifmedia_add(&priv->media, IFM_10G_LR | IFM_ETHER, 0, NULL);
3661 ifmedia_add(&priv->media, IFM_10G_LR |
3662 IFM_ETHER | IFM_FDX |
3663 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3665 ifmedia_add(&priv->media, IFM_40G_ER4 | IFM_ETHER, 0, NULL);
3666 ifmedia_add(&priv->media, IFM_40G_ER4 |
3667 IFM_ETHER | IFM_FDX |
3668 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3670 ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO, 0, NULL);
3671 ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
3672 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3674 /* Set autoselect by default */
3675 ifmedia_set(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
3676 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE);
3677 ether_ifattach(ifp, dev_addr);
3679 /* Register for VLAN events */
3680 priv->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
3681 mlx5e_vlan_rx_add_vid, priv, EVENTHANDLER_PRI_FIRST);
3682 priv->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
3683 mlx5e_vlan_rx_kill_vid, priv, EVENTHANDLER_PRI_FIRST);
3685 /* Link is down by default */
3686 if_link_state_change(ifp, LINK_STATE_DOWN);
3688 mlx5e_enable_async_events(priv);
3690 mlx5e_add_hw_stats(priv);
3692 mlx5e_create_stats(&priv->stats.vport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3693 "vstats", mlx5e_vport_stats_desc, MLX5E_VPORT_STATS_NUM,
3694 priv->stats.vport.arg);
3696 mlx5e_create_stats(&priv->stats.pport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3697 "pstats", mlx5e_pport_stats_desc, MLX5E_PPORT_STATS_NUM,
3698 priv->stats.pport.arg);
3700 mlx5e_create_ethtool(priv);
3702 mtx_lock(&priv->async_events_mtx);
3703 mlx5e_update_stats(priv);
3704 mtx_unlock(&priv->async_events_mtx);
3708 err_dealloc_transport_domain:
3709 mlx5_dealloc_transport_domain(mdev, priv->tdn);
3712 mlx5_core_dealloc_pd(mdev, priv->pdn);
3715 mlx5_unmap_free_uar(mdev, &priv->cq_uar);
3718 destroy_workqueue(priv->wq);
3721 sysctl_ctx_free(&priv->sysctl_ctx);
3722 if (priv->sysctl_debug)
3723 sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
3727 mlx5e_priv_mtx_destroy(priv);
3728 free(priv, M_MLX5EN);
3733 mlx5e_destroy_ifp(struct mlx5_core_dev *mdev, void *vpriv)
3735 struct mlx5e_priv *priv = vpriv;
3736 struct ifnet *ifp = priv->ifp;
3738 /* don't allow more IOCTLs */
3741 /* XXX wait a bit to allow IOCTL handlers to complete */
3744 /* stop watchdog timer */
3745 callout_drain(&priv->watchdog);
3747 if (priv->vlan_attach != NULL)
3748 EVENTHANDLER_DEREGISTER(vlan_config, priv->vlan_attach);
3749 if (priv->vlan_detach != NULL)
3750 EVENTHANDLER_DEREGISTER(vlan_unconfig, priv->vlan_detach);
3752 /* make sure device gets closed */
3754 mlx5e_close_locked(ifp);
3757 /* unregister device */
3758 ifmedia_removeall(&priv->media);
3759 ether_ifdetach(ifp);
3762 /* destroy all remaining sysctl nodes */
3763 sysctl_ctx_free(&priv->stats.vport.ctx);
3764 sysctl_ctx_free(&priv->stats.pport.ctx);
3765 if (priv->sysctl_debug)
3766 sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
3767 sysctl_ctx_free(&priv->sysctl_ctx);
3769 mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
3770 mlx5_dealloc_transport_domain(priv->mdev, priv->tdn);
3771 mlx5_core_dealloc_pd(priv->mdev, priv->pdn);
3772 mlx5_unmap_free_uar(priv->mdev, &priv->cq_uar);
3773 mlx5e_disable_async_events(priv);
3774 destroy_workqueue(priv->wq);
3775 mlx5e_priv_mtx_destroy(priv);
3776 free(priv, M_MLX5EN);
3780 mlx5e_get_ifp(void *vpriv)
3782 struct mlx5e_priv *priv = vpriv;
3787 static struct mlx5_interface mlx5e_interface = {
3788 .add = mlx5e_create_ifp,
3789 .remove = mlx5e_destroy_ifp,
3790 .event = mlx5e_async_event,
3791 .protocol = MLX5_INTERFACE_PROTOCOL_ETH,
3792 .get_dev = mlx5e_get_ifp,
3798 mlx5_register_interface(&mlx5e_interface);
3804 mlx5_unregister_interface(&mlx5e_interface);
3808 mlx5e_show_version(void __unused *arg)
3811 printf("%s", mlx5e_version);
3813 SYSINIT(mlx5e_show_version, SI_SUB_DRIVERS, SI_ORDER_ANY, mlx5e_show_version, NULL);
3815 module_init_order(mlx5e_init, SI_ORDER_THIRD);
3816 module_exit_order(mlx5e_cleanup, SI_ORDER_THIRD);
3818 #if (__FreeBSD_version >= 1100000)
3819 MODULE_DEPEND(mlx5en, linuxkpi, 1, 1, 1);
3821 MODULE_DEPEND(mlx5en, mlx5, 1, 1, 1);
3822 MODULE_VERSION(mlx5en, 1);