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_work(struct work_struct *work)
480 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
482 struct mlx5_core_dev *mdev = priv->mdev;
483 struct mlx5e_vport_stats *s = &priv->stats.vport;
484 struct mlx5e_sq_stats *sq_stats;
485 struct buf_ring *sq_br;
486 #if (__FreeBSD_version < 1100000)
487 struct ifnet *ifp = priv->ifp;
490 u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)];
492 int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
495 u64 tx_queue_dropped = 0;
496 u64 tx_defragged = 0;
497 u64 tx_offload_none = 0;
500 u64 sw_lro_queued = 0;
501 u64 sw_lro_flushed = 0;
502 u64 rx_csum_none = 0;
504 u32 rx_out_of_buffer = 0;
509 out = mlx5_vzalloc(outlen);
512 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
515 /* Collect firts the SW counters and then HW for consistency */
516 for (i = 0; i < priv->params.num_channels; i++) {
517 struct mlx5e_channel *pch = priv->channel + i;
518 struct mlx5e_rq *rq = &pch->rq;
519 struct mlx5e_rq_stats *rq_stats = &pch->rq.stats;
521 /* collect stats from LRO */
522 rq_stats->sw_lro_queued = rq->lro.lro_queued;
523 rq_stats->sw_lro_flushed = rq->lro.lro_flushed;
524 sw_lro_queued += rq_stats->sw_lro_queued;
525 sw_lro_flushed += rq_stats->sw_lro_flushed;
526 lro_packets += rq_stats->lro_packets;
527 lro_bytes += rq_stats->lro_bytes;
528 rx_csum_none += rq_stats->csum_none;
529 rx_wqe_err += rq_stats->wqe_err;
531 for (j = 0; j < priv->num_tc; j++) {
532 sq_stats = &pch->sq[j].stats;
533 sq_br = pch->sq[j].br;
535 tso_packets += sq_stats->tso_packets;
536 tso_bytes += sq_stats->tso_bytes;
537 tx_queue_dropped += sq_stats->dropped;
539 tx_queue_dropped += sq_br->br_drops;
540 tx_defragged += sq_stats->defragged;
541 tx_offload_none += sq_stats->csum_offload_none;
545 /* update counters */
546 s->tso_packets = tso_packets;
547 s->tso_bytes = tso_bytes;
548 s->tx_queue_dropped = tx_queue_dropped;
549 s->tx_defragged = tx_defragged;
550 s->lro_packets = lro_packets;
551 s->lro_bytes = lro_bytes;
552 s->sw_lro_queued = sw_lro_queued;
553 s->sw_lro_flushed = sw_lro_flushed;
554 s->rx_csum_none = rx_csum_none;
555 s->rx_wqe_err = rx_wqe_err;
558 memset(in, 0, sizeof(in));
560 MLX5_SET(query_vport_counter_in, in, opcode,
561 MLX5_CMD_OP_QUERY_VPORT_COUNTER);
562 MLX5_SET(query_vport_counter_in, in, op_mod, 0);
563 MLX5_SET(query_vport_counter_in, in, other_vport, 0);
565 memset(out, 0, outlen);
567 /* get number of out-of-buffer drops first */
568 if (mlx5_vport_query_out_of_rx_buffer(mdev, priv->counter_set_id,
572 /* accumulate difference into a 64-bit counter */
573 s->rx_out_of_buffer += (u64)(u32)(rx_out_of_buffer - s->rx_out_of_buffer_prev);
574 s->rx_out_of_buffer_prev = rx_out_of_buffer;
576 /* get port statistics */
577 if (mlx5_cmd_exec(mdev, in, sizeof(in), out, outlen))
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);
620 s->rx_unicast_packets +
621 s->rx_multicast_packets +
622 s->rx_broadcast_packets -
625 s->rx_unicast_bytes +
626 s->rx_multicast_bytes +
627 s->rx_broadcast_bytes;
629 s->tx_unicast_packets +
630 s->tx_multicast_packets +
631 s->tx_broadcast_packets;
633 s->tx_unicast_bytes +
634 s->tx_multicast_bytes +
635 s->tx_broadcast_bytes;
637 /* Update calculated offload counters */
638 s->tx_csum_offload = s->tx_packets - tx_offload_none;
639 s->rx_csum_good = s->rx_packets - s->rx_csum_none;
641 /* Get physical port counters */
642 mlx5e_update_pport_counters(priv);
644 s->tx_jumbo_packets =
645 priv->stats.port_stats_debug.tx_stat_p1519to2047octets +
646 priv->stats.port_stats_debug.tx_stat_p2048to4095octets +
647 priv->stats.port_stats_debug.tx_stat_p4096to8191octets +
648 priv->stats.port_stats_debug.tx_stat_p8192to10239octets;
650 #if (__FreeBSD_version < 1100000)
651 /* no get_counters interface in fbsd 10 */
652 ifp->if_ipackets = s->rx_packets;
653 ifp->if_ierrors = s->rx_error_packets +
654 priv->stats.pport.alignment_err +
655 priv->stats.pport.check_seq_err +
656 priv->stats.pport.crc_align_errors +
657 priv->stats.pport.in_range_len_errors +
658 priv->stats.pport.jabbers +
659 priv->stats.pport.out_of_range_len +
660 priv->stats.pport.oversize_pkts +
661 priv->stats.pport.symbol_err +
662 priv->stats.pport.too_long_errors +
663 priv->stats.pport.undersize_pkts +
664 priv->stats.pport.unsupported_op_rx;
665 ifp->if_iqdrops = s->rx_out_of_buffer +
666 priv->stats.pport.drop_events;
667 ifp->if_opackets = s->tx_packets;
668 ifp->if_oerrors = s->tx_error_packets;
669 ifp->if_snd.ifq_drops = s->tx_queue_dropped;
670 ifp->if_ibytes = s->rx_bytes;
671 ifp->if_obytes = s->tx_bytes;
673 priv->stats.pport.collisions;
679 /* Update diagnostics, if any */
680 if (priv->params_ethtool.diag_pci_enable ||
681 priv->params_ethtool.diag_general_enable) {
682 int error = mlx5_core_get_diagnostics_full(mdev,
683 priv->params_ethtool.diag_pci_enable ? &priv->params_pci : NULL,
684 priv->params_ethtool.diag_general_enable ? &priv->params_general : NULL);
686 if_printf(priv->ifp, "Failed reading diagnostics: %d\n", error);
692 mlx5e_update_stats(void *arg)
694 struct mlx5e_priv *priv = arg;
696 queue_work(priv->wq, &priv->update_stats_work);
698 callout_reset(&priv->watchdog, hz, &mlx5e_update_stats, priv);
702 mlx5e_async_event_sub(struct mlx5e_priv *priv,
703 enum mlx5_dev_event event)
706 case MLX5_DEV_EVENT_PORT_UP:
707 case MLX5_DEV_EVENT_PORT_DOWN:
708 queue_work(priv->wq, &priv->update_carrier_work);
717 mlx5e_async_event(struct mlx5_core_dev *mdev, void *vpriv,
718 enum mlx5_dev_event event, unsigned long param)
720 struct mlx5e_priv *priv = vpriv;
722 mtx_lock(&priv->async_events_mtx);
723 if (test_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state))
724 mlx5e_async_event_sub(priv, event);
725 mtx_unlock(&priv->async_events_mtx);
729 mlx5e_enable_async_events(struct mlx5e_priv *priv)
731 set_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
735 mlx5e_disable_async_events(struct mlx5e_priv *priv)
737 mtx_lock(&priv->async_events_mtx);
738 clear_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
739 mtx_unlock(&priv->async_events_mtx);
742 static const char *mlx5e_rq_stats_desc[] = {
743 MLX5E_RQ_STATS(MLX5E_STATS_DESC)
747 mlx5e_create_rq(struct mlx5e_channel *c,
748 struct mlx5e_rq_param *param,
751 struct mlx5e_priv *priv = c->priv;
752 struct mlx5_core_dev *mdev = priv->mdev;
754 void *rqc = param->rqc;
755 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
761 err = mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
765 /* Create DMA descriptor TAG */
766 if ((err = -bus_dma_tag_create(
767 bus_get_dma_tag(mdev->pdev->dev.bsddev),
768 1, /* any alignment */
770 BUS_SPACE_MAXADDR, /* lowaddr */
771 BUS_SPACE_MAXADDR, /* highaddr */
772 NULL, NULL, /* filter, filterarg */
773 nsegs * MLX5E_MAX_RX_BYTES, /* maxsize */
774 nsegs, /* nsegments */
775 nsegs * MLX5E_MAX_RX_BYTES, /* maxsegsize */
777 NULL, NULL, /* lockfunc, lockfuncarg */
781 err = mlx5_wq_ll_create(mdev, ¶m->wq, rqc_wq, &rq->wq,
784 goto err_free_dma_tag;
786 rq->wq.db = &rq->wq.db[MLX5_RCV_DBR];
788 err = mlx5e_get_wqe_sz(priv, &rq->wqe_sz, &rq->nsegs);
790 goto err_rq_wq_destroy;
792 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
794 err = -tcp_lro_init_args(&rq->lro, c->ifp, TCP_LRO_ENTRIES, wq_sz);
796 goto err_rq_wq_destroy;
798 rq->mbuf = malloc(wq_sz * sizeof(rq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
799 for (i = 0; i != wq_sz; i++) {
800 struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
801 #if (MLX5E_MAX_RX_SEGS == 1)
802 uint32_t byte_count = rq->wqe_sz - MLX5E_NET_IP_ALIGN;
807 err = -bus_dmamap_create(rq->dma_tag, 0, &rq->mbuf[i].dma_map);
810 bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
811 goto err_rq_mbuf_free;
814 /* set value for constant fields */
815 #if (MLX5E_MAX_RX_SEGS == 1)
816 wqe->data[0].lkey = c->mkey_be;
817 wqe->data[0].byte_count = cpu_to_be32(byte_count | MLX5_HW_START_PADDING);
819 for (j = 0; j < rq->nsegs; j++)
820 wqe->data[j].lkey = c->mkey_be;
824 INIT_WORK(&rq->dim.work, mlx5e_dim_work);
825 if (priv->params.rx_cq_moderation_mode < 2) {
826 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
828 void *cqc = container_of(param,
829 struct mlx5e_channel_param, rq)->rx_cq.cqc;
831 switch (MLX5_GET(cqc, cqc, cq_period_mode)) {
832 case MLX5_CQ_PERIOD_MODE_START_FROM_EQE:
833 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
835 case MLX5_CQ_PERIOD_MODE_START_FROM_CQE:
836 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
839 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
848 snprintf(buffer, sizeof(buffer), "rxstat%d", c->ix);
849 mlx5e_create_stats(&rq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
850 buffer, mlx5e_rq_stats_desc, MLX5E_RQ_STATS_NUM,
855 free(rq->mbuf, M_MLX5EN);
856 tcp_lro_free(&rq->lro);
858 mlx5_wq_destroy(&rq->wq_ctrl);
860 bus_dma_tag_destroy(rq->dma_tag);
866 mlx5e_destroy_rq(struct mlx5e_rq *rq)
871 /* destroy all sysctl nodes */
872 sysctl_ctx_free(&rq->stats.ctx);
874 /* free leftover LRO packets, if any */
875 tcp_lro_free(&rq->lro);
877 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
878 for (i = 0; i != wq_sz; i++) {
879 if (rq->mbuf[i].mbuf != NULL) {
880 bus_dmamap_unload(rq->dma_tag, rq->mbuf[i].dma_map);
881 m_freem(rq->mbuf[i].mbuf);
883 bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
885 free(rq->mbuf, M_MLX5EN);
886 mlx5_wq_destroy(&rq->wq_ctrl);
890 mlx5e_enable_rq(struct mlx5e_rq *rq, struct mlx5e_rq_param *param)
892 struct mlx5e_channel *c = rq->channel;
893 struct mlx5e_priv *priv = c->priv;
894 struct mlx5_core_dev *mdev = priv->mdev;
902 inlen = MLX5_ST_SZ_BYTES(create_rq_in) +
903 sizeof(u64) * rq->wq_ctrl.buf.npages;
904 in = mlx5_vzalloc(inlen);
908 rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
909 wq = MLX5_ADDR_OF(rqc, rqc, wq);
911 memcpy(rqc, param->rqc, sizeof(param->rqc));
913 MLX5_SET(rqc, rqc, cqn, c->rq.cq.mcq.cqn);
914 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST);
915 MLX5_SET(rqc, rqc, flush_in_error_en, 1);
916 if (priv->counter_set_id >= 0)
917 MLX5_SET(rqc, rqc, counter_set_id, priv->counter_set_id);
918 MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift -
920 MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma);
922 mlx5_fill_page_array(&rq->wq_ctrl.buf,
923 (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
925 err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn);
933 mlx5e_modify_rq(struct mlx5e_rq *rq, int curr_state, int next_state)
935 struct mlx5e_channel *c = rq->channel;
936 struct mlx5e_priv *priv = c->priv;
937 struct mlx5_core_dev *mdev = priv->mdev;
944 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
945 in = mlx5_vzalloc(inlen);
949 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
951 MLX5_SET(modify_rq_in, in, rqn, rq->rqn);
952 MLX5_SET(modify_rq_in, in, rq_state, curr_state);
953 MLX5_SET(rqc, rqc, state, next_state);
955 err = mlx5_core_modify_rq(mdev, in, inlen);
963 mlx5e_disable_rq(struct mlx5e_rq *rq)
965 struct mlx5e_channel *c = rq->channel;
966 struct mlx5e_priv *priv = c->priv;
967 struct mlx5_core_dev *mdev = priv->mdev;
969 mlx5_core_destroy_rq(mdev, rq->rqn);
973 mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq)
975 struct mlx5e_channel *c = rq->channel;
976 struct mlx5e_priv *priv = c->priv;
977 struct mlx5_wq_ll *wq = &rq->wq;
980 for (i = 0; i < 1000; i++) {
981 if (wq->cur_sz >= priv->params.min_rx_wqes)
990 mlx5e_open_rq(struct mlx5e_channel *c,
991 struct mlx5e_rq_param *param,
996 err = mlx5e_create_rq(c, param, rq);
1000 err = mlx5e_enable_rq(rq, param);
1002 goto err_destroy_rq;
1004 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
1006 goto err_disable_rq;
1013 mlx5e_disable_rq(rq);
1015 mlx5e_destroy_rq(rq);
1021 mlx5e_close_rq(struct mlx5e_rq *rq)
1025 callout_stop(&rq->watchdog);
1026 mtx_unlock(&rq->mtx);
1028 callout_drain(&rq->watchdog);
1030 mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
1034 mlx5e_close_rq_wait(struct mlx5e_rq *rq)
1037 mlx5e_disable_rq(rq);
1038 mlx5e_close_cq(&rq->cq);
1039 cancel_work_sync(&rq->dim.work);
1040 mlx5e_destroy_rq(rq);
1044 mlx5e_free_sq_db(struct mlx5e_sq *sq)
1046 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1049 for (x = 0; x != wq_sz; x++)
1050 bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1051 free(sq->mbuf, M_MLX5EN);
1055 mlx5e_alloc_sq_db(struct mlx5e_sq *sq)
1057 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1061 sq->mbuf = malloc(wq_sz * sizeof(sq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
1063 /* Create DMA descriptor MAPs */
1064 for (x = 0; x != wq_sz; x++) {
1065 err = -bus_dmamap_create(sq->dma_tag, 0, &sq->mbuf[x].dma_map);
1068 bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1069 free(sq->mbuf, M_MLX5EN);
1076 static const char *mlx5e_sq_stats_desc[] = {
1077 MLX5E_SQ_STATS(MLX5E_STATS_DESC)
1081 mlx5e_update_sq_inline(struct mlx5e_sq *sq)
1083 sq->max_inline = sq->priv->params.tx_max_inline;
1084 sq->min_inline_mode = sq->priv->params.tx_min_inline_mode;
1087 * Check if trust state is DSCP or if inline mode is NONE which
1088 * indicates CX-5 or newer hardware.
1090 if (sq->priv->params_ethtool.trust_state != MLX5_QPTS_TRUST_PCP ||
1091 sq->min_inline_mode == MLX5_INLINE_MODE_NONE) {
1092 if (MLX5_CAP_ETH(sq->priv->mdev, wqe_vlan_insert))
1093 sq->min_insert_caps = MLX5E_INSERT_VLAN | MLX5E_INSERT_NON_VLAN;
1095 sq->min_insert_caps = MLX5E_INSERT_NON_VLAN;
1097 sq->min_insert_caps = 0;
1102 mlx5e_refresh_sq_inline_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
1106 for (i = 0; i != c->num_tc; i++) {
1107 mtx_lock(&c->sq[i].lock);
1108 mlx5e_update_sq_inline(&c->sq[i]);
1109 mtx_unlock(&c->sq[i].lock);
1114 mlx5e_refresh_sq_inline(struct mlx5e_priv *priv)
1118 /* check if channels are closed */
1119 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
1122 for (i = 0; i < priv->params.num_channels; i++)
1123 mlx5e_refresh_sq_inline_sub(priv, &priv->channel[i]);
1127 mlx5e_create_sq(struct mlx5e_channel *c,
1129 struct mlx5e_sq_param *param,
1130 struct mlx5e_sq *sq)
1132 struct mlx5e_priv *priv = c->priv;
1133 struct mlx5_core_dev *mdev = priv->mdev;
1135 void *sqc = param->sqc;
1136 void *sqc_wq = MLX5_ADDR_OF(sqc, sqc, wq);
1139 /* Create DMA descriptor TAG */
1140 if ((err = -bus_dma_tag_create(
1141 bus_get_dma_tag(mdev->pdev->dev.bsddev),
1142 1, /* any alignment */
1143 0, /* no boundary */
1144 BUS_SPACE_MAXADDR, /* lowaddr */
1145 BUS_SPACE_MAXADDR, /* highaddr */
1146 NULL, NULL, /* filter, filterarg */
1147 MLX5E_MAX_TX_PAYLOAD_SIZE, /* maxsize */
1148 MLX5E_MAX_TX_MBUF_FRAGS, /* nsegments */
1149 MLX5E_MAX_TX_MBUF_SIZE, /* maxsegsize */
1151 NULL, NULL, /* lockfunc, lockfuncarg */
1155 err = mlx5_alloc_map_uar(mdev, &sq->uar);
1157 goto err_free_dma_tag;
1159 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, &sq->wq,
1162 goto err_unmap_free_uar;
1164 sq->wq.db = &sq->wq.db[MLX5_SND_DBR];
1165 sq->bf_buf_size = (1 << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2;
1167 err = mlx5e_alloc_sq_db(sq);
1169 goto err_sq_wq_destroy;
1171 sq->mkey_be = c->mkey_be;
1172 sq->ifp = priv->ifp;
1176 mlx5e_update_sq_inline(sq);
1178 snprintf(buffer, sizeof(buffer), "txstat%dtc%d", c->ix, tc);
1179 mlx5e_create_stats(&sq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
1180 buffer, mlx5e_sq_stats_desc, MLX5E_SQ_STATS_NUM,
1186 mlx5_wq_destroy(&sq->wq_ctrl);
1189 mlx5_unmap_free_uar(mdev, &sq->uar);
1192 bus_dma_tag_destroy(sq->dma_tag);
1198 mlx5e_destroy_sq(struct mlx5e_sq *sq)
1200 /* destroy all sysctl nodes */
1201 sysctl_ctx_free(&sq->stats.ctx);
1203 mlx5e_free_sq_db(sq);
1204 mlx5_wq_destroy(&sq->wq_ctrl);
1205 mlx5_unmap_free_uar(sq->priv->mdev, &sq->uar);
1209 mlx5e_enable_sq(struct mlx5e_sq *sq, struct mlx5e_sq_param *param,
1218 inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
1219 sizeof(u64) * sq->wq_ctrl.buf.npages;
1220 in = mlx5_vzalloc(inlen);
1224 sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
1225 wq = MLX5_ADDR_OF(sqc, sqc, wq);
1227 memcpy(sqc, param->sqc, sizeof(param->sqc));
1229 MLX5_SET(sqc, sqc, tis_num_0, tis_num);
1230 MLX5_SET(sqc, sqc, cqn, sq->cq.mcq.cqn);
1231 MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
1232 MLX5_SET(sqc, sqc, tis_lst_sz, 1);
1233 MLX5_SET(sqc, sqc, flush_in_error_en, 1);
1235 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
1236 MLX5_SET(wq, wq, uar_page, sq->uar.index);
1237 MLX5_SET(wq, wq, log_wq_pg_sz, sq->wq_ctrl.buf.page_shift -
1239 MLX5_SET64(wq, wq, dbr_addr, sq->wq_ctrl.db.dma);
1241 mlx5_fill_page_array(&sq->wq_ctrl.buf,
1242 (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
1244 err = mlx5_core_create_sq(sq->priv->mdev, in, inlen, &sq->sqn);
1252 mlx5e_modify_sq(struct mlx5e_sq *sq, int curr_state, int next_state)
1259 inlen = MLX5_ST_SZ_BYTES(modify_sq_in);
1260 in = mlx5_vzalloc(inlen);
1264 sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx);
1266 MLX5_SET(modify_sq_in, in, sqn, sq->sqn);
1267 MLX5_SET(modify_sq_in, in, sq_state, curr_state);
1268 MLX5_SET(sqc, sqc, state, next_state);
1270 err = mlx5_core_modify_sq(sq->priv->mdev, in, inlen);
1278 mlx5e_disable_sq(struct mlx5e_sq *sq)
1281 mlx5_core_destroy_sq(sq->priv->mdev, sq->sqn);
1285 mlx5e_open_sq(struct mlx5e_channel *c,
1287 struct mlx5e_sq_param *param,
1288 struct mlx5e_sq *sq)
1292 err = mlx5e_create_sq(c, tc, param, sq);
1296 err = mlx5e_enable_sq(sq, param, c->priv->tisn[tc]);
1298 goto err_destroy_sq;
1300 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY);
1302 goto err_disable_sq;
1304 WRITE_ONCE(sq->running, 1);
1309 mlx5e_disable_sq(sq);
1311 mlx5e_destroy_sq(sq);
1317 mlx5e_sq_send_nops_locked(struct mlx5e_sq *sq, int can_sleep)
1319 /* fill up remainder with NOPs */
1320 while (sq->cev_counter != 0) {
1321 while (!mlx5e_sq_has_room_for(sq, 1)) {
1322 if (can_sleep != 0) {
1323 mtx_unlock(&sq->lock);
1325 mtx_lock(&sq->lock);
1330 /* send a single NOP */
1331 mlx5e_send_nop(sq, 1);
1332 atomic_thread_fence_rel();
1335 /* Check if we need to write the doorbell */
1336 if (likely(sq->doorbell.d64 != 0)) {
1337 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
1338 sq->doorbell.d64 = 0;
1343 mlx5e_sq_cev_timeout(void *arg)
1345 struct mlx5e_sq *sq = arg;
1347 mtx_assert(&sq->lock, MA_OWNED);
1349 /* check next state */
1350 switch (sq->cev_next_state) {
1351 case MLX5E_CEV_STATE_SEND_NOPS:
1352 /* fill TX ring with NOPs, if any */
1353 mlx5e_sq_send_nops_locked(sq, 0);
1355 /* check if completed */
1356 if (sq->cev_counter == 0) {
1357 sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
1362 /* send NOPs on next timeout */
1363 sq->cev_next_state = MLX5E_CEV_STATE_SEND_NOPS;
1368 callout_reset_curcpu(&sq->cev_callout, hz, mlx5e_sq_cev_timeout, sq);
1372 mlx5e_drain_sq(struct mlx5e_sq *sq)
1375 struct mlx5_core_dev *mdev= sq->priv->mdev;
1378 * Check if already stopped.
1380 * NOTE: Serialization of this function is managed by the
1381 * caller ensuring the priv's state lock is locked or in case
1382 * of rate limit support, a single thread manages drain and
1383 * resume of SQs. The "running" variable can therefore safely
1384 * be read without any locks.
1386 if (READ_ONCE(sq->running) == 0)
1389 /* don't put more packets into the SQ */
1390 WRITE_ONCE(sq->running, 0);
1392 /* serialize access to DMA rings */
1393 mtx_lock(&sq->lock);
1395 /* teardown event factor timer, if any */
1396 sq->cev_next_state = MLX5E_CEV_STATE_HOLD_NOPS;
1397 callout_stop(&sq->cev_callout);
1399 /* send dummy NOPs in order to flush the transmit ring */
1400 mlx5e_sq_send_nops_locked(sq, 1);
1401 mtx_unlock(&sq->lock);
1403 /* make sure it is safe to free the callout */
1404 callout_drain(&sq->cev_callout);
1406 /* wait till SQ is empty or link is down */
1407 mtx_lock(&sq->lock);
1408 while (sq->cc != sq->pc &&
1409 (sq->priv->media_status_last & IFM_ACTIVE) != 0 &&
1410 mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1411 mtx_unlock(&sq->lock);
1413 sq->cq.mcq.comp(&sq->cq.mcq);
1414 mtx_lock(&sq->lock);
1416 mtx_unlock(&sq->lock);
1418 /* error out remaining requests */
1419 error = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RDY, MLX5_SQC_STATE_ERR);
1422 "mlx5e_modify_sq() from RDY to ERR failed: %d\n", error);
1425 /* wait till SQ is empty */
1426 mtx_lock(&sq->lock);
1427 while (sq->cc != sq->pc &&
1428 mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1429 mtx_unlock(&sq->lock);
1431 sq->cq.mcq.comp(&sq->cq.mcq);
1432 mtx_lock(&sq->lock);
1434 mtx_unlock(&sq->lock);
1438 mlx5e_close_sq_wait(struct mlx5e_sq *sq)
1442 mlx5e_disable_sq(sq);
1443 mlx5e_destroy_sq(sq);
1447 mlx5e_create_cq(struct mlx5e_priv *priv,
1448 struct mlx5e_cq_param *param,
1449 struct mlx5e_cq *cq,
1450 mlx5e_cq_comp_t *comp,
1453 struct mlx5_core_dev *mdev = priv->mdev;
1454 struct mlx5_core_cq *mcq = &cq->mcq;
1460 param->wq.buf_numa_node = 0;
1461 param->wq.db_numa_node = 0;
1463 err = mlx5_cqwq_create(mdev, ¶m->wq, param->cqc, &cq->wq,
1468 mlx5_vector2eqn(mdev, eq_ix, &eqn_not_used, &irqn);
1471 mcq->set_ci_db = cq->wq_ctrl.db.db;
1472 mcq->arm_db = cq->wq_ctrl.db.db + 1;
1473 *mcq->set_ci_db = 0;
1475 mcq->vector = eq_ix;
1477 mcq->event = mlx5e_cq_error_event;
1479 mcq->uar = &priv->cq_uar;
1481 for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) {
1482 struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i);
1493 mlx5e_destroy_cq(struct mlx5e_cq *cq)
1495 mlx5_wq_destroy(&cq->wq_ctrl);
1499 mlx5e_enable_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param, int eq_ix)
1501 struct mlx5_core_cq *mcq = &cq->mcq;
1509 inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
1510 sizeof(u64) * cq->wq_ctrl.buf.npages;
1511 in = mlx5_vzalloc(inlen);
1515 cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
1517 memcpy(cqc, param->cqc, sizeof(param->cqc));
1519 mlx5_fill_page_array(&cq->wq_ctrl.buf,
1520 (__be64 *) MLX5_ADDR_OF(create_cq_in, in, pas));
1522 mlx5_vector2eqn(cq->priv->mdev, eq_ix, &eqn, &irqn_not_used);
1524 MLX5_SET(cqc, cqc, c_eqn, eqn);
1525 MLX5_SET(cqc, cqc, uar_page, mcq->uar->index);
1526 MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.buf.page_shift -
1528 MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
1530 err = mlx5_core_create_cq(cq->priv->mdev, mcq, in, inlen);
1537 mlx5e_cq_arm(cq, MLX5_GET_DOORBELL_LOCK(&cq->priv->doorbell_lock));
1543 mlx5e_disable_cq(struct mlx5e_cq *cq)
1546 mlx5_core_destroy_cq(cq->priv->mdev, &cq->mcq);
1550 mlx5e_open_cq(struct mlx5e_priv *priv,
1551 struct mlx5e_cq_param *param,
1552 struct mlx5e_cq *cq,
1553 mlx5e_cq_comp_t *comp,
1558 err = mlx5e_create_cq(priv, param, cq, comp, eq_ix);
1562 err = mlx5e_enable_cq(cq, param, eq_ix);
1564 goto err_destroy_cq;
1569 mlx5e_destroy_cq(cq);
1575 mlx5e_close_cq(struct mlx5e_cq *cq)
1577 mlx5e_disable_cq(cq);
1578 mlx5e_destroy_cq(cq);
1582 mlx5e_open_tx_cqs(struct mlx5e_channel *c,
1583 struct mlx5e_channel_param *cparam)
1588 for (tc = 0; tc < c->num_tc; tc++) {
1589 /* open completion queue */
1590 err = mlx5e_open_cq(c->priv, &cparam->tx_cq, &c->sq[tc].cq,
1591 &mlx5e_tx_cq_comp, c->ix);
1593 goto err_close_tx_cqs;
1598 for (tc--; tc >= 0; tc--)
1599 mlx5e_close_cq(&c->sq[tc].cq);
1605 mlx5e_close_tx_cqs(struct mlx5e_channel *c)
1609 for (tc = 0; tc < c->num_tc; tc++)
1610 mlx5e_close_cq(&c->sq[tc].cq);
1614 mlx5e_open_sqs(struct mlx5e_channel *c,
1615 struct mlx5e_channel_param *cparam)
1620 for (tc = 0; tc < c->num_tc; tc++) {
1621 err = mlx5e_open_sq(c, tc, &cparam->sq, &c->sq[tc]);
1629 for (tc--; tc >= 0; tc--)
1630 mlx5e_close_sq_wait(&c->sq[tc]);
1636 mlx5e_close_sqs_wait(struct mlx5e_channel *c)
1640 for (tc = 0; tc < c->num_tc; tc++)
1641 mlx5e_close_sq_wait(&c->sq[tc]);
1645 mlx5e_chan_mtx_init(struct mlx5e_channel *c)
1649 mtx_init(&c->rq.mtx, "mlx5rx", MTX_NETWORK_LOCK, MTX_DEF);
1651 callout_init_mtx(&c->rq.watchdog, &c->rq.mtx, 0);
1653 for (tc = 0; tc < c->num_tc; tc++) {
1654 struct mlx5e_sq *sq = c->sq + tc;
1656 mtx_init(&sq->lock, "mlx5tx",
1657 MTX_NETWORK_LOCK " TX", MTX_DEF);
1658 mtx_init(&sq->comp_lock, "mlx5comp",
1659 MTX_NETWORK_LOCK " TX", MTX_DEF);
1661 callout_init_mtx(&sq->cev_callout, &sq->lock, 0);
1663 sq->cev_factor = c->priv->params_ethtool.tx_completion_fact;
1665 /* ensure the TX completion event factor is not zero */
1666 if (sq->cev_factor == 0)
1672 mlx5e_chan_mtx_destroy(struct mlx5e_channel *c)
1676 mtx_destroy(&c->rq.mtx);
1678 for (tc = 0; tc < c->num_tc; tc++) {
1679 mtx_destroy(&c->sq[tc].lock);
1680 mtx_destroy(&c->sq[tc].comp_lock);
1685 mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
1686 struct mlx5e_channel_param *cparam,
1687 struct mlx5e_channel *c)
1691 memset(c, 0, sizeof(*c));
1696 c->mkey_be = cpu_to_be32(priv->mr.key);
1697 c->num_tc = priv->num_tc;
1700 mlx5e_chan_mtx_init(c);
1702 /* open transmit completion queue */
1703 err = mlx5e_open_tx_cqs(c, cparam);
1707 /* open receive completion queue */
1708 err = mlx5e_open_cq(c->priv, &cparam->rx_cq, &c->rq.cq,
1709 &mlx5e_rx_cq_comp, c->ix);
1711 goto err_close_tx_cqs;
1713 err = mlx5e_open_sqs(c, cparam);
1715 goto err_close_rx_cq;
1717 err = mlx5e_open_rq(c, &cparam->rq, &c->rq);
1721 /* poll receive queue initially */
1722 c->rq.cq.mcq.comp(&c->rq.cq.mcq);
1727 mlx5e_close_sqs_wait(c);
1730 mlx5e_close_cq(&c->rq.cq);
1733 mlx5e_close_tx_cqs(c);
1736 /* destroy mutexes */
1737 mlx5e_chan_mtx_destroy(c);
1742 mlx5e_close_channel(struct mlx5e_channel *c)
1744 mlx5e_close_rq(&c->rq);
1748 mlx5e_close_channel_wait(struct mlx5e_channel *c)
1750 mlx5e_close_rq_wait(&c->rq);
1751 mlx5e_close_sqs_wait(c);
1752 mlx5e_close_tx_cqs(c);
1753 /* destroy mutexes */
1754 mlx5e_chan_mtx_destroy(c);
1758 mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs)
1762 r = priv->params.hw_lro_en ? priv->params.lro_wqe_sz :
1763 MLX5E_SW2MB_MTU(priv->ifp->if_mtu);
1764 if (r > MJUM16BYTES)
1769 else if (r > MJUMPAGESIZE)
1771 else if (r > MCLBYTES)
1777 * n + 1 must be a power of two, because stride size must be.
1778 * Stride size is 16 * (n + 1), as the first segment is
1781 for (n = howmany(r, MLX5E_MAX_RX_BYTES); !powerof2(n + 1); n++)
1790 mlx5e_build_rq_param(struct mlx5e_priv *priv,
1791 struct mlx5e_rq_param *param)
1793 void *rqc = param->rqc;
1794 void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
1797 mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
1798 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
1799 MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN);
1800 MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe) +
1801 nsegs * sizeof(struct mlx5_wqe_data_seg)));
1802 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_rq_size);
1803 MLX5_SET(wq, wq, pd, priv->pdn);
1805 param->wq.buf_numa_node = 0;
1806 param->wq.db_numa_node = 0;
1807 param->wq.linear = 1;
1811 mlx5e_build_sq_param(struct mlx5e_priv *priv,
1812 struct mlx5e_sq_param *param)
1814 void *sqc = param->sqc;
1815 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
1817 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_sq_size);
1818 MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
1819 MLX5_SET(wq, wq, pd, priv->pdn);
1821 param->wq.buf_numa_node = 0;
1822 param->wq.db_numa_node = 0;
1823 param->wq.linear = 1;
1827 mlx5e_build_common_cq_param(struct mlx5e_priv *priv,
1828 struct mlx5e_cq_param *param)
1830 void *cqc = param->cqc;
1832 MLX5_SET(cqc, cqc, uar_page, priv->cq_uar.index);
1836 mlx5e_get_default_profile(struct mlx5e_priv *priv, int mode, struct net_dim_cq_moder *ptr)
1839 *ptr = net_dim_get_profile(mode, MLX5E_DIM_DEFAULT_PROFILE);
1841 /* apply LRO restrictions */
1842 if (priv->params.hw_lro_en &&
1843 ptr->pkts > MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO) {
1844 ptr->pkts = MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO;
1849 mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
1850 struct mlx5e_cq_param *param)
1852 struct net_dim_cq_moder curr;
1853 void *cqc = param->cqc;
1857 * TODO The sysctl to control on/off is a bool value for now, which means
1858 * we only support CSUM, once HASH is implemnted we'll need to address that.
1860 if (priv->params.cqe_zipping_en) {
1861 MLX5_SET(cqc, cqc, mini_cqe_res_format, MLX5_CQE_FORMAT_CSUM);
1862 MLX5_SET(cqc, cqc, cqe_compression_en, 1);
1865 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_rq_size);
1867 switch (priv->params.rx_cq_moderation_mode) {
1869 MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
1870 MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
1871 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1874 MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
1875 MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
1876 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
1877 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
1879 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1882 mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE, &curr);
1883 MLX5_SET(cqc, cqc, cq_period, curr.usec);
1884 MLX5_SET(cqc, cqc, cq_max_count, curr.pkts);
1885 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1888 mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE, &curr);
1889 MLX5_SET(cqc, cqc, cq_period, curr.usec);
1890 MLX5_SET(cqc, cqc, cq_max_count, curr.pkts);
1891 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
1892 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
1894 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1900 mlx5e_dim_build_cq_param(priv, param);
1902 mlx5e_build_common_cq_param(priv, param);
1906 mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
1907 struct mlx5e_cq_param *param)
1909 void *cqc = param->cqc;
1911 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_sq_size);
1912 MLX5_SET(cqc, cqc, cq_period, priv->params.tx_cq_moderation_usec);
1913 MLX5_SET(cqc, cqc, cq_max_count, priv->params.tx_cq_moderation_pkts);
1915 switch (priv->params.tx_cq_moderation_mode) {
1917 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1920 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
1921 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
1923 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1927 mlx5e_build_common_cq_param(priv, param);
1931 mlx5e_build_channel_param(struct mlx5e_priv *priv,
1932 struct mlx5e_channel_param *cparam)
1934 memset(cparam, 0, sizeof(*cparam));
1936 mlx5e_build_rq_param(priv, &cparam->rq);
1937 mlx5e_build_sq_param(priv, &cparam->sq);
1938 mlx5e_build_rx_cq_param(priv, &cparam->rx_cq);
1939 mlx5e_build_tx_cq_param(priv, &cparam->tx_cq);
1943 mlx5e_open_channels(struct mlx5e_priv *priv)
1945 struct mlx5e_channel_param cparam;
1950 mlx5e_build_channel_param(priv, &cparam);
1951 for (i = 0; i < priv->params.num_channels; i++) {
1952 err = mlx5e_open_channel(priv, i, &cparam, &priv->channel[i]);
1954 goto err_close_channels;
1957 for (j = 0; j < priv->params.num_channels; j++) {
1958 err = mlx5e_wait_for_min_rx_wqes(&priv->channel[j].rq);
1960 goto err_close_channels;
1967 mlx5e_close_channel(&priv->channel[i]);
1968 mlx5e_close_channel_wait(&priv->channel[i]);
1974 mlx5e_close_channels(struct mlx5e_priv *priv)
1978 for (i = 0; i < priv->params.num_channels; i++)
1979 mlx5e_close_channel(&priv->channel[i]);
1980 for (i = 0; i < priv->params.num_channels; i++)
1981 mlx5e_close_channel_wait(&priv->channel[i]);
1985 mlx5e_refresh_sq_params(struct mlx5e_priv *priv, struct mlx5e_sq *sq)
1988 if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
1991 switch (priv->params.tx_cq_moderation_mode) {
1994 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
1997 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2001 return (mlx5_core_modify_cq_moderation_mode(priv->mdev, &sq->cq.mcq,
2002 priv->params.tx_cq_moderation_usec,
2003 priv->params.tx_cq_moderation_pkts,
2007 return (mlx5_core_modify_cq_moderation(priv->mdev, &sq->cq.mcq,
2008 priv->params.tx_cq_moderation_usec,
2009 priv->params.tx_cq_moderation_pkts));
2013 mlx5e_refresh_rq_params(struct mlx5e_priv *priv, struct mlx5e_rq *rq)
2016 if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
2021 switch (priv->params.rx_cq_moderation_mode) {
2024 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2025 dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
2028 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2029 dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
2033 /* tear down dynamic interrupt moderation */
2035 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
2036 mtx_unlock(&rq->mtx);
2038 /* wait for dynamic interrupt moderation work task, if any */
2039 cancel_work_sync(&rq->dim.work);
2041 if (priv->params.rx_cq_moderation_mode >= 2) {
2042 struct net_dim_cq_moder curr;
2044 mlx5e_get_default_profile(priv, dim_mode, &curr);
2046 retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2047 curr.usec, curr.pkts, cq_mode);
2049 /* set dynamic interrupt moderation mode and zero defaults */
2051 rq->dim.mode = dim_mode;
2053 rq->dim.profile_ix = MLX5E_DIM_DEFAULT_PROFILE;
2054 mtx_unlock(&rq->mtx);
2056 retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2057 priv->params.rx_cq_moderation_usec,
2058 priv->params.rx_cq_moderation_pkts,
2064 return (mlx5_core_modify_cq_moderation(priv->mdev, &rq->cq.mcq,
2065 priv->params.rx_cq_moderation_usec,
2066 priv->params.rx_cq_moderation_pkts));
2070 mlx5e_refresh_channel_params_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
2075 err = mlx5e_refresh_rq_params(priv, &c->rq);
2079 for (i = 0; i != c->num_tc; i++) {
2080 err = mlx5e_refresh_sq_params(priv, &c->sq[i]);
2089 mlx5e_refresh_channel_params(struct mlx5e_priv *priv)
2093 /* check if channels are closed */
2094 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2097 for (i = 0; i < priv->params.num_channels; i++) {
2100 err = mlx5e_refresh_channel_params_sub(priv, &priv->channel[i]);
2108 mlx5e_open_tis(struct mlx5e_priv *priv, int tc)
2110 struct mlx5_core_dev *mdev = priv->mdev;
2111 u32 in[MLX5_ST_SZ_DW(create_tis_in)];
2112 void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
2114 memset(in, 0, sizeof(in));
2116 MLX5_SET(tisc, tisc, prio, tc);
2117 MLX5_SET(tisc, tisc, transport_domain, priv->tdn);
2119 return (mlx5_core_create_tis(mdev, in, sizeof(in), &priv->tisn[tc]));
2123 mlx5e_close_tis(struct mlx5e_priv *priv, int tc)
2125 mlx5_core_destroy_tis(priv->mdev, priv->tisn[tc]);
2129 mlx5e_open_tises(struct mlx5e_priv *priv)
2131 int num_tc = priv->num_tc;
2135 for (tc = 0; tc < num_tc; tc++) {
2136 err = mlx5e_open_tis(priv, tc);
2138 goto err_close_tises;
2144 for (tc--; tc >= 0; tc--)
2145 mlx5e_close_tis(priv, tc);
2151 mlx5e_close_tises(struct mlx5e_priv *priv)
2153 int num_tc = priv->num_tc;
2156 for (tc = 0; tc < num_tc; tc++)
2157 mlx5e_close_tis(priv, tc);
2161 mlx5e_open_rqt(struct mlx5e_priv *priv)
2163 struct mlx5_core_dev *mdev = priv->mdev;
2165 u32 out[MLX5_ST_SZ_DW(create_rqt_out)] = {0};
2172 sz = 1 << priv->params.rx_hash_log_tbl_sz;
2174 inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
2175 in = mlx5_vzalloc(inlen);
2178 rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
2180 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
2181 MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
2183 for (i = 0; i < sz; i++) {
2186 ix = rss_get_indirection_to_bucket(ix);
2188 /* ensure we don't overflow */
2189 ix %= priv->params.num_channels;
2191 /* apply receive side scaling stride, if any */
2192 ix -= ix % (int)priv->params.channels_rsss;
2194 MLX5_SET(rqtc, rqtc, rq_num[i], priv->channel[ix].rq.rqn);
2197 MLX5_SET(create_rqt_in, in, opcode, MLX5_CMD_OP_CREATE_RQT);
2199 err = mlx5_cmd_exec(mdev, in, inlen, out, sizeof(out));
2201 priv->rqtn = MLX5_GET(create_rqt_out, out, rqtn);
2209 mlx5e_close_rqt(struct mlx5e_priv *priv)
2211 u32 in[MLX5_ST_SZ_DW(destroy_rqt_in)] = {0};
2212 u32 out[MLX5_ST_SZ_DW(destroy_rqt_out)] = {0};
2214 MLX5_SET(destroy_rqt_in, in, opcode, MLX5_CMD_OP_DESTROY_RQT);
2215 MLX5_SET(destroy_rqt_in, in, rqtn, priv->rqtn);
2217 mlx5_cmd_exec(priv->mdev, in, sizeof(in), out, sizeof(out));
2221 mlx5e_build_tir_ctx(struct mlx5e_priv *priv, u32 * tirc, int tt)
2223 void *hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
2226 MLX5_SET(tirc, tirc, transport_domain, priv->tdn);
2228 #define ROUGH_MAX_L2_L3_HDR_SZ 256
2230 #define MLX5_HASH_IP (MLX5_HASH_FIELD_SEL_SRC_IP |\
2231 MLX5_HASH_FIELD_SEL_DST_IP)
2233 #define MLX5_HASH_ALL (MLX5_HASH_FIELD_SEL_SRC_IP |\
2234 MLX5_HASH_FIELD_SEL_DST_IP |\
2235 MLX5_HASH_FIELD_SEL_L4_SPORT |\
2236 MLX5_HASH_FIELD_SEL_L4_DPORT)
2238 #define MLX5_HASH_IP_IPSEC_SPI (MLX5_HASH_FIELD_SEL_SRC_IP |\
2239 MLX5_HASH_FIELD_SEL_DST_IP |\
2240 MLX5_HASH_FIELD_SEL_IPSEC_SPI)
2242 if (priv->params.hw_lro_en) {
2243 MLX5_SET(tirc, tirc, lro_enable_mask,
2244 MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO |
2245 MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO);
2246 MLX5_SET(tirc, tirc, lro_max_msg_sz,
2247 (priv->params.lro_wqe_sz -
2248 ROUGH_MAX_L2_L3_HDR_SZ) >> 8);
2249 /* TODO: add the option to choose timer value dynamically */
2250 MLX5_SET(tirc, tirc, lro_timeout_period_usecs,
2251 MLX5_CAP_ETH(priv->mdev,
2252 lro_timer_supported_periods[2]));
2255 /* setup parameters for hashing TIR type, if any */
2258 MLX5_SET(tirc, tirc, disp_type,
2259 MLX5_TIRC_DISP_TYPE_DIRECT);
2260 MLX5_SET(tirc, tirc, inline_rqn,
2261 priv->channel[0].rq.rqn);
2264 MLX5_SET(tirc, tirc, disp_type,
2265 MLX5_TIRC_DISP_TYPE_INDIRECT);
2266 MLX5_SET(tirc, tirc, indirect_table,
2268 MLX5_SET(tirc, tirc, rx_hash_fn,
2269 MLX5_TIRC_RX_HASH_FN_HASH_TOEPLITZ);
2270 hkey = (__be32 *) MLX5_ADDR_OF(tirc, tirc, rx_hash_toeplitz_key);
2273 * The FreeBSD RSS implementation does currently not
2274 * support symmetric Toeplitz hashes:
2276 MLX5_SET(tirc, tirc, rx_hash_symmetric, 0);
2277 rss_getkey((uint8_t *)hkey);
2279 MLX5_SET(tirc, tirc, rx_hash_symmetric, 1);
2280 hkey[0] = cpu_to_be32(0xD181C62C);
2281 hkey[1] = cpu_to_be32(0xF7F4DB5B);
2282 hkey[2] = cpu_to_be32(0x1983A2FC);
2283 hkey[3] = cpu_to_be32(0x943E1ADB);
2284 hkey[4] = cpu_to_be32(0xD9389E6B);
2285 hkey[5] = cpu_to_be32(0xD1039C2C);
2286 hkey[6] = cpu_to_be32(0xA74499AD);
2287 hkey[7] = cpu_to_be32(0x593D56D9);
2288 hkey[8] = cpu_to_be32(0xF3253C06);
2289 hkey[9] = cpu_to_be32(0x2ADC1FFC);
2295 case MLX5E_TT_IPV4_TCP:
2296 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2297 MLX5_L3_PROT_TYPE_IPV4);
2298 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2299 MLX5_L4_PROT_TYPE_TCP);
2301 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4)) {
2302 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2306 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2310 case MLX5E_TT_IPV6_TCP:
2311 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2312 MLX5_L3_PROT_TYPE_IPV6);
2313 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2314 MLX5_L4_PROT_TYPE_TCP);
2316 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) {
2317 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2321 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2325 case MLX5E_TT_IPV4_UDP:
2326 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2327 MLX5_L3_PROT_TYPE_IPV4);
2328 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2329 MLX5_L4_PROT_TYPE_UDP);
2331 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4)) {
2332 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2336 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2340 case MLX5E_TT_IPV6_UDP:
2341 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2342 MLX5_L3_PROT_TYPE_IPV6);
2343 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2344 MLX5_L4_PROT_TYPE_UDP);
2346 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) {
2347 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2351 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2355 case MLX5E_TT_IPV4_IPSEC_AH:
2356 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2357 MLX5_L3_PROT_TYPE_IPV4);
2358 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2359 MLX5_HASH_IP_IPSEC_SPI);
2362 case MLX5E_TT_IPV6_IPSEC_AH:
2363 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2364 MLX5_L3_PROT_TYPE_IPV6);
2365 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2366 MLX5_HASH_IP_IPSEC_SPI);
2369 case MLX5E_TT_IPV4_IPSEC_ESP:
2370 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2371 MLX5_L3_PROT_TYPE_IPV4);
2372 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2373 MLX5_HASH_IP_IPSEC_SPI);
2376 case MLX5E_TT_IPV6_IPSEC_ESP:
2377 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2378 MLX5_L3_PROT_TYPE_IPV6);
2379 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2380 MLX5_HASH_IP_IPSEC_SPI);
2384 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2385 MLX5_L3_PROT_TYPE_IPV4);
2386 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2391 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2392 MLX5_L3_PROT_TYPE_IPV6);
2393 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2403 mlx5e_open_tir(struct mlx5e_priv *priv, int tt)
2405 struct mlx5_core_dev *mdev = priv->mdev;
2411 inlen = MLX5_ST_SZ_BYTES(create_tir_in);
2412 in = mlx5_vzalloc(inlen);
2415 tirc = MLX5_ADDR_OF(create_tir_in, in, tir_context);
2417 mlx5e_build_tir_ctx(priv, tirc, tt);
2419 err = mlx5_core_create_tir(mdev, in, inlen, &priv->tirn[tt]);
2427 mlx5e_close_tir(struct mlx5e_priv *priv, int tt)
2429 mlx5_core_destroy_tir(priv->mdev, priv->tirn[tt]);
2433 mlx5e_open_tirs(struct mlx5e_priv *priv)
2438 for (i = 0; i < MLX5E_NUM_TT; i++) {
2439 err = mlx5e_open_tir(priv, i);
2441 goto err_close_tirs;
2447 for (i--; i >= 0; i--)
2448 mlx5e_close_tir(priv, i);
2454 mlx5e_close_tirs(struct mlx5e_priv *priv)
2458 for (i = 0; i < MLX5E_NUM_TT; i++)
2459 mlx5e_close_tir(priv, i);
2463 * SW MTU does not include headers,
2464 * HW MTU includes all headers and checksums.
2467 mlx5e_set_dev_port_mtu(struct ifnet *ifp, int sw_mtu)
2469 struct mlx5e_priv *priv = ifp->if_softc;
2470 struct mlx5_core_dev *mdev = priv->mdev;
2474 hw_mtu = MLX5E_SW2HW_MTU(sw_mtu);
2476 err = mlx5_set_port_mtu(mdev, hw_mtu);
2478 if_printf(ifp, "%s: mlx5_set_port_mtu failed setting %d, err=%d\n",
2479 __func__, sw_mtu, err);
2483 /* Update vport context MTU */
2484 err = mlx5_set_vport_mtu(mdev, hw_mtu);
2486 if_printf(ifp, "%s: Failed updating vport context with MTU size, err=%d\n",
2490 ifp->if_mtu = sw_mtu;
2492 err = mlx5_query_vport_mtu(mdev, &hw_mtu);
2493 if (err || !hw_mtu) {
2494 /* fallback to port oper mtu */
2495 err = mlx5_query_port_oper_mtu(mdev, &hw_mtu);
2498 if_printf(ifp, "Query port MTU, after setting new "
2499 "MTU value, failed\n");
2501 } else if (MLX5E_HW2SW_MTU(hw_mtu) < sw_mtu) {
2503 if_printf(ifp, "Port MTU %d is smaller than "
2504 "ifp mtu %d\n", hw_mtu, sw_mtu);
2505 } else if (MLX5E_HW2SW_MTU(hw_mtu) > sw_mtu) {
2507 if_printf(ifp, "Port MTU %d is bigger than "
2508 "ifp mtu %d\n", hw_mtu, sw_mtu);
2510 priv->params_ethtool.hw_mtu = hw_mtu;
2516 mlx5e_open_locked(struct ifnet *ifp)
2518 struct mlx5e_priv *priv = ifp->if_softc;
2522 /* check if already opened */
2523 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
2527 if (rss_getnumbuckets() > priv->params.num_channels) {
2528 if_printf(ifp, "NOTE: There are more RSS buckets(%u) than "
2529 "channels(%u) available\n", rss_getnumbuckets(),
2530 priv->params.num_channels);
2533 err = mlx5e_open_tises(priv);
2535 if_printf(ifp, "%s: mlx5e_open_tises failed, %d\n",
2539 err = mlx5_vport_alloc_q_counter(priv->mdev,
2540 MLX5_INTERFACE_PROTOCOL_ETH, &set_id);
2542 if_printf(priv->ifp,
2543 "%s: mlx5_vport_alloc_q_counter failed: %d\n",
2545 goto err_close_tises;
2547 /* store counter set ID */
2548 priv->counter_set_id = set_id;
2550 err = mlx5e_open_channels(priv);
2552 if_printf(ifp, "%s: mlx5e_open_channels failed, %d\n",
2554 goto err_dalloc_q_counter;
2556 err = mlx5e_open_rqt(priv);
2558 if_printf(ifp, "%s: mlx5e_open_rqt failed, %d\n",
2560 goto err_close_channels;
2562 err = mlx5e_open_tirs(priv);
2564 if_printf(ifp, "%s: mlx5e_open_tir failed, %d\n",
2566 goto err_close_rqls;
2568 err = mlx5e_open_flow_table(priv);
2570 if_printf(ifp, "%s: mlx5e_open_flow_table failed, %d\n",
2572 goto err_close_tirs;
2574 err = mlx5e_add_all_vlan_rules(priv);
2576 if_printf(ifp, "%s: mlx5e_add_all_vlan_rules failed, %d\n",
2578 goto err_close_flow_table;
2580 set_bit(MLX5E_STATE_OPENED, &priv->state);
2582 mlx5e_update_carrier(priv);
2583 mlx5e_set_rx_mode_core(priv);
2587 err_close_flow_table:
2588 mlx5e_close_flow_table(priv);
2591 mlx5e_close_tirs(priv);
2594 mlx5e_close_rqt(priv);
2597 mlx5e_close_channels(priv);
2599 err_dalloc_q_counter:
2600 mlx5_vport_dealloc_q_counter(priv->mdev,
2601 MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
2604 mlx5e_close_tises(priv);
2610 mlx5e_open(void *arg)
2612 struct mlx5e_priv *priv = arg;
2615 if (mlx5_set_port_status(priv->mdev, MLX5_PORT_UP))
2616 if_printf(priv->ifp,
2617 "%s: Setting port status to up failed\n",
2620 mlx5e_open_locked(priv->ifp);
2621 priv->ifp->if_drv_flags |= IFF_DRV_RUNNING;
2626 mlx5e_close_locked(struct ifnet *ifp)
2628 struct mlx5e_priv *priv = ifp->if_softc;
2630 /* check if already closed */
2631 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2634 clear_bit(MLX5E_STATE_OPENED, &priv->state);
2636 mlx5e_set_rx_mode_core(priv);
2637 mlx5e_del_all_vlan_rules(priv);
2638 if_link_state_change(priv->ifp, LINK_STATE_DOWN);
2639 mlx5e_close_flow_table(priv);
2640 mlx5e_close_tirs(priv);
2641 mlx5e_close_rqt(priv);
2642 mlx5e_close_channels(priv);
2643 mlx5_vport_dealloc_q_counter(priv->mdev,
2644 MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
2645 mlx5e_close_tises(priv);
2650 #if (__FreeBSD_version >= 1100000)
2652 mlx5e_get_counter(struct ifnet *ifp, ift_counter cnt)
2654 struct mlx5e_priv *priv = ifp->if_softc;
2657 /* PRIV_LOCK(priv); XXX not allowed */
2659 case IFCOUNTER_IPACKETS:
2660 retval = priv->stats.vport.rx_packets;
2662 case IFCOUNTER_IERRORS:
2663 retval = priv->stats.vport.rx_error_packets +
2664 priv->stats.pport.alignment_err +
2665 priv->stats.pport.check_seq_err +
2666 priv->stats.pport.crc_align_errors +
2667 priv->stats.pport.in_range_len_errors +
2668 priv->stats.pport.jabbers +
2669 priv->stats.pport.out_of_range_len +
2670 priv->stats.pport.oversize_pkts +
2671 priv->stats.pport.symbol_err +
2672 priv->stats.pport.too_long_errors +
2673 priv->stats.pport.undersize_pkts +
2674 priv->stats.pport.unsupported_op_rx;
2676 case IFCOUNTER_IQDROPS:
2677 retval = priv->stats.vport.rx_out_of_buffer +
2678 priv->stats.pport.drop_events;
2680 case IFCOUNTER_OPACKETS:
2681 retval = priv->stats.vport.tx_packets;
2683 case IFCOUNTER_OERRORS:
2684 retval = priv->stats.vport.tx_error_packets;
2686 case IFCOUNTER_IBYTES:
2687 retval = priv->stats.vport.rx_bytes;
2689 case IFCOUNTER_OBYTES:
2690 retval = priv->stats.vport.tx_bytes;
2692 case IFCOUNTER_IMCASTS:
2693 retval = priv->stats.vport.rx_multicast_packets;
2695 case IFCOUNTER_OMCASTS:
2696 retval = priv->stats.vport.tx_multicast_packets;
2698 case IFCOUNTER_OQDROPS:
2699 retval = priv->stats.vport.tx_queue_dropped;
2701 case IFCOUNTER_COLLISIONS:
2702 retval = priv->stats.pport.collisions;
2705 retval = if_get_counter_default(ifp, cnt);
2708 /* PRIV_UNLOCK(priv); XXX not allowed */
2714 mlx5e_set_rx_mode(struct ifnet *ifp)
2716 struct mlx5e_priv *priv = ifp->if_softc;
2718 queue_work(priv->wq, &priv->set_rx_mode_work);
2722 mlx5e_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2724 struct mlx5e_priv *priv;
2726 struct ifi2creq i2c;
2735 priv = ifp->if_softc;
2737 /* check if detaching */
2738 if (priv == NULL || priv->gone != 0)
2743 ifr = (struct ifreq *)data;
2746 mlx5_query_port_max_mtu(priv->mdev, &max_mtu);
2748 if (ifr->ifr_mtu >= MLX5E_MTU_MIN &&
2749 ifr->ifr_mtu <= MIN(MLX5E_MTU_MAX, max_mtu)) {
2752 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2754 mlx5e_close_locked(ifp);
2757 mlx5e_set_dev_port_mtu(ifp, ifr->ifr_mtu);
2760 mlx5e_open_locked(ifp);
2763 if_printf(ifp, "Invalid MTU value. Min val: %d, Max val: %d\n",
2764 MLX5E_MTU_MIN, MIN(MLX5E_MTU_MAX, max_mtu));
2769 if ((ifp->if_flags & IFF_UP) &&
2770 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
2771 mlx5e_set_rx_mode(ifp);
2775 if (ifp->if_flags & IFF_UP) {
2776 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2777 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2778 mlx5e_open_locked(ifp);
2779 ifp->if_drv_flags |= IFF_DRV_RUNNING;
2780 mlx5_set_port_status(priv->mdev, MLX5_PORT_UP);
2783 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
2784 mlx5_set_port_status(priv->mdev,
2786 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
2787 mlx5e_close_locked(ifp);
2788 mlx5e_update_carrier(priv);
2789 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2796 mlx5e_set_rx_mode(ifp);
2801 ifr = (struct ifreq *)data;
2802 error = ifmedia_ioctl(ifp, ifr, &priv->media, command);
2805 ifr = (struct ifreq *)data;
2807 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2809 if (mask & IFCAP_TXCSUM) {
2810 ifp->if_capenable ^= IFCAP_TXCSUM;
2811 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
2813 if (IFCAP_TSO4 & ifp->if_capenable &&
2814 !(IFCAP_TXCSUM & ifp->if_capenable)) {
2815 ifp->if_capenable &= ~IFCAP_TSO4;
2816 ifp->if_hwassist &= ~CSUM_IP_TSO;
2818 "tso4 disabled due to -txcsum.\n");
2821 if (mask & IFCAP_TXCSUM_IPV6) {
2822 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
2823 ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
2825 if (IFCAP_TSO6 & ifp->if_capenable &&
2826 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2827 ifp->if_capenable &= ~IFCAP_TSO6;
2828 ifp->if_hwassist &= ~CSUM_IP6_TSO;
2830 "tso6 disabled due to -txcsum6.\n");
2833 if (mask & IFCAP_RXCSUM)
2834 ifp->if_capenable ^= IFCAP_RXCSUM;
2835 if (mask & IFCAP_RXCSUM_IPV6)
2836 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
2837 if (mask & IFCAP_TSO4) {
2838 if (!(IFCAP_TSO4 & ifp->if_capenable) &&
2839 !(IFCAP_TXCSUM & ifp->if_capenable)) {
2840 if_printf(ifp, "enable txcsum first.\n");
2844 ifp->if_capenable ^= IFCAP_TSO4;
2845 ifp->if_hwassist ^= CSUM_IP_TSO;
2847 if (mask & IFCAP_TSO6) {
2848 if (!(IFCAP_TSO6 & ifp->if_capenable) &&
2849 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2850 if_printf(ifp, "enable txcsum6 first.\n");
2854 ifp->if_capenable ^= IFCAP_TSO6;
2855 ifp->if_hwassist ^= CSUM_IP6_TSO;
2857 if (mask & IFCAP_VLAN_HWFILTER) {
2858 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2859 mlx5e_disable_vlan_filter(priv);
2861 mlx5e_enable_vlan_filter(priv);
2863 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
2865 if (mask & IFCAP_VLAN_HWTAGGING)
2866 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2867 if (mask & IFCAP_WOL_MAGIC)
2868 ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2870 VLAN_CAPABILITIES(ifp);
2871 /* turn off LRO means also turn of HW LRO - if it's on */
2872 if (mask & IFCAP_LRO) {
2873 int was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2874 bool need_restart = false;
2876 ifp->if_capenable ^= IFCAP_LRO;
2878 /* figure out if updating HW LRO is needed */
2879 if (!(ifp->if_capenable & IFCAP_LRO)) {
2880 if (priv->params.hw_lro_en) {
2881 priv->params.hw_lro_en = false;
2882 need_restart = true;
2885 if (priv->params.hw_lro_en == false &&
2886 priv->params_ethtool.hw_lro != 0) {
2887 priv->params.hw_lro_en = true;
2888 need_restart = true;
2891 if (was_opened && need_restart) {
2892 mlx5e_close_locked(ifp);
2893 mlx5e_open_locked(ifp);
2901 ifr = (struct ifreq *)data;
2904 * Copy from the user-space address ifr_data to the
2905 * kernel-space address i2c
2907 error = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
2911 if (i2c.len > sizeof(i2c.data)) {
2917 /* Get module_num which is required for the query_eeprom */
2918 error = mlx5_query_module_num(priv->mdev, &module_num);
2920 if_printf(ifp, "Query module num failed, eeprom "
2921 "reading is not supported\n");
2925 /* Check if module is present before doing an access */
2926 module_status = mlx5_query_module_status(priv->mdev, module_num);
2927 if (module_status != MLX5_MODULE_STATUS_PLUGGED_ENABLED &&
2928 module_status != MLX5_MODULE_STATUS_PLUGGED_DISABLED) {
2933 * Currently 0XA0 and 0xA2 are the only addresses permitted.
2934 * The internal conversion is as follows:
2936 if (i2c.dev_addr == 0xA0)
2937 read_addr = MLX5E_I2C_ADDR_LOW;
2938 else if (i2c.dev_addr == 0xA2)
2939 read_addr = MLX5E_I2C_ADDR_HIGH;
2941 if_printf(ifp, "Query eeprom failed, "
2942 "Invalid Address: %X\n", i2c.dev_addr);
2946 error = mlx5_query_eeprom(priv->mdev,
2947 read_addr, MLX5E_EEPROM_LOW_PAGE,
2948 (uint32_t)i2c.offset, (uint32_t)i2c.len, module_num,
2949 (uint32_t *)i2c.data, &size_read);
2951 if_printf(ifp, "Query eeprom failed, eeprom "
2952 "reading is not supported\n");
2957 if (i2c.len > MLX5_EEPROM_MAX_BYTES) {
2958 error = mlx5_query_eeprom(priv->mdev,
2959 read_addr, MLX5E_EEPROM_LOW_PAGE,
2960 (uint32_t)(i2c.offset + size_read),
2961 (uint32_t)(i2c.len - size_read), module_num,
2962 (uint32_t *)(i2c.data + size_read), &size_read);
2965 if_printf(ifp, "Query eeprom failed, eeprom "
2966 "reading is not supported\n");
2971 error = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
2977 error = ether_ioctl(ifp, command, data);
2984 mlx5e_check_required_hca_cap(struct mlx5_core_dev *mdev)
2987 * TODO: uncoment once FW really sets all these bits if
2988 * (!mdev->caps.eth.rss_ind_tbl_cap || !mdev->caps.eth.csum_cap ||
2989 * !mdev->caps.eth.max_lso_cap || !mdev->caps.eth.vlan_cap ||
2990 * !(mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_SCQE_BRK_MOD)) return
2994 /* TODO: add more must-to-have features */
2996 if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
3003 mlx5e_get_max_inline_cap(struct mlx5_core_dev *mdev)
3005 uint32_t bf_buf_size = (1U << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2U;
3007 bf_buf_size -= sizeof(struct mlx5e_tx_wqe) - 2;
3009 /* verify against driver hardware limit */
3010 if (bf_buf_size > MLX5E_MAX_TX_INLINE)
3011 bf_buf_size = MLX5E_MAX_TX_INLINE;
3013 return (bf_buf_size);
3017 mlx5e_build_ifp_priv(struct mlx5_core_dev *mdev,
3018 struct mlx5e_priv *priv,
3019 int num_comp_vectors)
3024 * TODO: Consider link speed for setting "log_sq_size",
3025 * "log_rq_size" and "cq_moderation_xxx":
3027 priv->params.log_sq_size =
3028 MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
3029 priv->params.log_rq_size =
3030 MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
3031 priv->params.rx_cq_moderation_usec =
3032 MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
3033 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE :
3034 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
3035 priv->params.rx_cq_moderation_mode =
3036 MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ? 1 : 0;
3037 priv->params.rx_cq_moderation_pkts =
3038 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS;
3039 priv->params.tx_cq_moderation_usec =
3040 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC;
3041 priv->params.tx_cq_moderation_pkts =
3042 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
3043 priv->params.min_rx_wqes =
3044 MLX5E_PARAMS_DEFAULT_MIN_RX_WQES;
3045 priv->params.rx_hash_log_tbl_sz =
3046 (order_base_2(num_comp_vectors) >
3047 MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ) ?
3048 order_base_2(num_comp_vectors) :
3049 MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ;
3050 priv->params.num_tc = 1;
3051 priv->params.default_vlan_prio = 0;
3052 priv->counter_set_id = -1;
3053 priv->params.tx_max_inline = mlx5e_get_max_inline_cap(mdev);
3055 err = mlx5_query_min_inline(mdev, &priv->params.tx_min_inline_mode);
3060 * hw lro is currently defaulted to off. when it won't anymore we
3061 * will consider the HW capability: "!!MLX5_CAP_ETH(mdev, lro_cap)"
3063 priv->params.hw_lro_en = false;
3064 priv->params.lro_wqe_sz = MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ;
3066 priv->params.cqe_zipping_en = !!MLX5_CAP_GEN(mdev, cqe_compression);
3069 priv->params.num_channels = num_comp_vectors;
3070 priv->params.channels_rsss = 1;
3071 priv->order_base_2_num_channels = order_base_2(num_comp_vectors);
3072 priv->queue_mapping_channel_mask =
3073 roundup_pow_of_two(num_comp_vectors) - 1;
3074 priv->num_tc = priv->params.num_tc;
3075 priv->default_vlan_prio = priv->params.default_vlan_prio;
3077 INIT_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
3078 INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
3079 INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
3085 mlx5e_create_mkey(struct mlx5e_priv *priv, u32 pdn,
3086 struct mlx5_core_mr *mkey)
3088 struct ifnet *ifp = priv->ifp;
3089 struct mlx5_core_dev *mdev = priv->mdev;
3090 int inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
3095 in = mlx5_vzalloc(inlen);
3097 if_printf(ifp, "%s: failed to allocate inbox\n", __func__);
3101 mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
3102 MLX5_SET(mkc, mkc, access_mode, MLX5_ACCESS_MODE_PA);
3103 MLX5_SET(mkc, mkc, lw, 1);
3104 MLX5_SET(mkc, mkc, lr, 1);
3106 MLX5_SET(mkc, mkc, pd, pdn);
3107 MLX5_SET(mkc, mkc, length64, 1);
3108 MLX5_SET(mkc, mkc, qpn, 0xffffff);
3110 err = mlx5_core_create_mkey(mdev, mkey, in, inlen);
3112 if_printf(ifp, "%s: mlx5_core_create_mkey failed, %d\n",
3119 static const char *mlx5e_vport_stats_desc[] = {
3120 MLX5E_VPORT_STATS(MLX5E_STATS_DESC)
3123 static const char *mlx5e_pport_stats_desc[] = {
3124 MLX5E_PPORT_STATS(MLX5E_STATS_DESC)
3128 mlx5e_priv_mtx_init(struct mlx5e_priv *priv)
3130 mtx_init(&priv->async_events_mtx, "mlx5async", MTX_NETWORK_LOCK, MTX_DEF);
3131 sx_init(&priv->state_lock, "mlx5state");
3132 callout_init_mtx(&priv->watchdog, &priv->async_events_mtx, 0);
3133 MLX5_INIT_DOORBELL_LOCK(&priv->doorbell_lock);
3137 mlx5e_priv_mtx_destroy(struct mlx5e_priv *priv)
3139 mtx_destroy(&priv->async_events_mtx);
3140 sx_destroy(&priv->state_lock);
3144 sysctl_firmware(SYSCTL_HANDLER_ARGS)
3147 * %d.%d%.d the string format.
3148 * fw_rev_{maj,min,sub} return u16, 2^16 = 65536.
3149 * We need at most 5 chars to store that.
3150 * It also has: two "." and NULL at the end, which means we need 18
3151 * (5*3 + 3) chars at most.
3154 struct mlx5e_priv *priv = arg1;
3157 snprintf(fw, sizeof(fw), "%d.%d.%d", fw_rev_maj(priv->mdev), fw_rev_min(priv->mdev),
3158 fw_rev_sub(priv->mdev));
3159 error = sysctl_handle_string(oidp, fw, sizeof(fw), req);
3164 mlx5e_disable_tx_dma(struct mlx5e_channel *ch)
3168 for (i = 0; i < ch->num_tc; i++)
3169 mlx5e_drain_sq(&ch->sq[i]);
3173 mlx5e_reset_sq_doorbell_record(struct mlx5e_sq *sq)
3176 sq->doorbell.d32[0] = cpu_to_be32(MLX5_OPCODE_NOP);
3177 sq->doorbell.d32[1] = cpu_to_be32(sq->sqn << 8);
3178 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
3179 sq->doorbell.d64 = 0;
3183 mlx5e_resume_sq(struct mlx5e_sq *sq)
3187 /* check if already enabled */
3188 if (READ_ONCE(sq->running) != 0)
3191 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_ERR,
3192 MLX5_SQC_STATE_RST);
3195 "mlx5e_modify_sq() from ERR to RST failed: %d\n", err);
3201 /* reset doorbell prior to moving from RST to RDY */
3202 mlx5e_reset_sq_doorbell_record(sq);
3204 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST,
3205 MLX5_SQC_STATE_RDY);
3208 "mlx5e_modify_sq() from RST to RDY failed: %d\n", err);
3211 sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
3212 WRITE_ONCE(sq->running, 1);
3216 mlx5e_enable_tx_dma(struct mlx5e_channel *ch)
3220 for (i = 0; i < ch->num_tc; i++)
3221 mlx5e_resume_sq(&ch->sq[i]);
3225 mlx5e_disable_rx_dma(struct mlx5e_channel *ch)
3227 struct mlx5e_rq *rq = &ch->rq;
3232 callout_stop(&rq->watchdog);
3233 mtx_unlock(&rq->mtx);
3235 callout_drain(&rq->watchdog);
3237 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
3240 "mlx5e_modify_rq() from RDY to RST failed: %d\n", err);
3243 while (!mlx5_wq_ll_is_empty(&rq->wq)) {
3245 rq->cq.mcq.comp(&rq->cq.mcq);
3249 * Transitioning into RST state will allow the FW to track less ERR state queues,
3250 * thus reducing the recv queue flushing time
3252 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_ERR, MLX5_RQC_STATE_RST);
3255 "mlx5e_modify_rq() from ERR to RST failed: %d\n", err);
3260 mlx5e_enable_rx_dma(struct mlx5e_channel *ch)
3262 struct mlx5e_rq *rq = &ch->rq;
3266 mlx5_wq_ll_update_db_record(&rq->wq);
3267 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
3270 "mlx5e_modify_rq() from RST to RDY failed: %d\n", err);
3275 rq->cq.mcq.comp(&rq->cq.mcq);
3279 mlx5e_modify_tx_dma(struct mlx5e_priv *priv, uint8_t value)
3283 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3286 for (i = 0; i < priv->params.num_channels; i++) {
3288 mlx5e_disable_tx_dma(&priv->channel[i]);
3290 mlx5e_enable_tx_dma(&priv->channel[i]);
3295 mlx5e_modify_rx_dma(struct mlx5e_priv *priv, uint8_t value)
3299 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3302 for (i = 0; i < priv->params.num_channels; i++) {
3304 mlx5e_disable_rx_dma(&priv->channel[i]);
3306 mlx5e_enable_rx_dma(&priv->channel[i]);
3311 mlx5e_add_hw_stats(struct mlx5e_priv *priv)
3313 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3314 OID_AUTO, "fw_version", CTLTYPE_STRING | CTLFLAG_RD, priv, 0,
3315 sysctl_firmware, "A", "HCA firmware version");
3317 SYSCTL_ADD_STRING(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3318 OID_AUTO, "board_id", CTLFLAG_RD, priv->mdev->board_id, 0,
3323 mlx5e_sysctl_tx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3325 struct mlx5e_priv *priv = arg1;
3326 uint8_t temp[MLX5E_MAX_PRIORITY];
3333 tx_pfc = priv->params.tx_priority_flow_control;
3335 for (i = 0; i != MLX5E_MAX_PRIORITY; i++)
3336 temp[i] = (tx_pfc >> i) & 1;
3338 err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY);
3339 if (err || !req->newptr)
3341 err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY);
3345 priv->params.tx_priority_flow_control = 0;
3347 /* range check input value */
3348 for (i = 0; i != MLX5E_MAX_PRIORITY; i++) {
3353 priv->params.tx_priority_flow_control |= (temp[i] << i);
3356 /* check if update is required */
3357 if (tx_pfc != priv->params.tx_priority_flow_control)
3358 err = -mlx5e_set_port_pfc(priv);
3361 priv->params.tx_priority_flow_control= tx_pfc;
3368 mlx5e_sysctl_rx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3370 struct mlx5e_priv *priv = arg1;
3371 uint8_t temp[MLX5E_MAX_PRIORITY];
3378 rx_pfc = priv->params.rx_priority_flow_control;
3380 for (i = 0; i != MLX5E_MAX_PRIORITY; i++)
3381 temp[i] = (rx_pfc >> i) & 1;
3383 err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY);
3384 if (err || !req->newptr)
3386 err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY);
3390 priv->params.rx_priority_flow_control = 0;
3392 /* range check input value */
3393 for (i = 0; i != MLX5E_MAX_PRIORITY; i++) {
3398 priv->params.rx_priority_flow_control |= (temp[i] << i);
3401 /* check if update is required */
3402 if (rx_pfc != priv->params.rx_priority_flow_control)
3403 err = -mlx5e_set_port_pfc(priv);
3406 priv->params.rx_priority_flow_control= rx_pfc;
3413 mlx5e_setup_pauseframes(struct mlx5e_priv *priv)
3415 #if (__FreeBSD_version < 1100000)
3420 /* enable pauseframes by default */
3421 priv->params.tx_pauseframe_control = 1;
3422 priv->params.rx_pauseframe_control = 1;
3424 /* disable ports flow control, PFC, by default */
3425 priv->params.tx_priority_flow_control = 0;
3426 priv->params.rx_priority_flow_control = 0;
3428 #if (__FreeBSD_version < 1100000)
3429 /* compute path for sysctl */
3430 snprintf(path, sizeof(path), "dev.mce.%d.tx_pauseframe_control",
3431 device_get_unit(priv->mdev->pdev->dev.bsddev));
3433 /* try to fetch tunable, if any */
3434 TUNABLE_INT_FETCH(path, &priv->params.tx_pauseframe_control);
3436 /* compute path for sysctl */
3437 snprintf(path, sizeof(path), "dev.mce.%d.rx_pauseframe_control",
3438 device_get_unit(priv->mdev->pdev->dev.bsddev));
3440 /* try to fetch tunable, if any */
3441 TUNABLE_INT_FETCH(path, &priv->params.rx_pauseframe_control);
3444 /* register pauseframe SYSCTLs */
3445 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3446 OID_AUTO, "tx_pauseframe_control", CTLFLAG_RDTUN,
3447 &priv->params.tx_pauseframe_control, 0,
3448 "Set to enable TX pause frames. Clear to disable.");
3450 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3451 OID_AUTO, "rx_pauseframe_control", CTLFLAG_RDTUN,
3452 &priv->params.rx_pauseframe_control, 0,
3453 "Set to enable RX pause frames. Clear to disable.");
3455 /* register priority flow control, PFC, SYSCTLs */
3456 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3457 OID_AUTO, "tx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN |
3458 CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_tx_priority_flow_control, "CU",
3459 "Set to enable TX ports flow control frames for priorities 0..7. Clear to disable.");
3461 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3462 OID_AUTO, "rx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN |
3463 CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_rx_priority_flow_control, "CU",
3464 "Set to enable RX ports flow control frames for priorities 0..7. Clear to disable.");
3469 priv->params.tx_pauseframe_control =
3470 priv->params.tx_pauseframe_control ? 1 : 0;
3471 priv->params.rx_pauseframe_control =
3472 priv->params.rx_pauseframe_control ? 1 : 0;
3474 /* update firmware */
3475 error = mlx5e_set_port_pause_and_pfc(priv);
3476 if (error == -EINVAL) {
3477 if_printf(priv->ifp,
3478 "Global pauseframes must be disabled before enabling PFC.\n");
3479 priv->params.rx_priority_flow_control = 0;
3480 priv->params.tx_priority_flow_control = 0;
3482 /* update firmware */
3483 (void) mlx5e_set_port_pause_and_pfc(priv);
3489 mlx5e_create_ifp(struct mlx5_core_dev *mdev)
3492 struct mlx5e_priv *priv;
3493 u8 dev_addr[ETHER_ADDR_LEN] __aligned(4);
3494 struct sysctl_oid_list *child;
3495 int ncv = mdev->priv.eq_table.num_comp_vectors;
3501 if (mlx5e_check_required_hca_cap(mdev)) {
3502 mlx5_core_dbg(mdev, "mlx5e_check_required_hca_cap() failed\n");
3506 * Try to allocate the priv and make room for worst-case
3507 * number of channel structures:
3509 priv = malloc(sizeof(*priv) +
3510 (sizeof(priv->channel[0]) * mdev->priv.eq_table.num_comp_vectors),
3511 M_MLX5EN, M_WAITOK | M_ZERO);
3512 mlx5e_priv_mtx_init(priv);
3514 ifp = priv->ifp = if_alloc(IFT_ETHER);
3516 mlx5_core_err(mdev, "if_alloc() failed\n");
3519 ifp->if_softc = priv;
3520 if_initname(ifp, "mce", device_get_unit(mdev->pdev->dev.bsddev));
3521 ifp->if_mtu = ETHERMTU;
3522 ifp->if_init = mlx5e_open;
3523 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
3524 ifp->if_ioctl = mlx5e_ioctl;
3525 ifp->if_transmit = mlx5e_xmit;
3526 ifp->if_qflush = if_qflush;
3527 #if (__FreeBSD_version >= 1100000)
3528 ifp->if_get_counter = mlx5e_get_counter;
3530 ifp->if_snd.ifq_maxlen = ifqmaxlen;
3532 * Set driver features
3534 ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6;
3535 ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
3536 ifp->if_capabilities |= IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWFILTER;
3537 ifp->if_capabilities |= IFCAP_LINKSTATE | IFCAP_JUMBO_MTU;
3538 ifp->if_capabilities |= IFCAP_LRO;
3539 ifp->if_capabilities |= IFCAP_TSO | IFCAP_VLAN_HWTSO;
3540 ifp->if_capabilities |= IFCAP_HWSTATS;
3542 /* set TSO limits so that we don't have to drop TX packets */
3543 ifp->if_hw_tsomax = MLX5E_MAX_TX_PAYLOAD_SIZE - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
3544 ifp->if_hw_tsomaxsegcount = MLX5E_MAX_TX_MBUF_FRAGS - 1 /* hdr */;
3545 ifp->if_hw_tsomaxsegsize = MLX5E_MAX_TX_MBUF_SIZE;
3547 ifp->if_capenable = ifp->if_capabilities;
3548 ifp->if_hwassist = 0;
3549 if (ifp->if_capenable & IFCAP_TSO)
3550 ifp->if_hwassist |= CSUM_TSO;
3551 if (ifp->if_capenable & IFCAP_TXCSUM)
3552 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP | CSUM_IP);
3553 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3554 ifp->if_hwassist |= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
3556 /* ifnet sysctl tree */
3557 sysctl_ctx_init(&priv->sysctl_ctx);
3558 priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_dev),
3559 OID_AUTO, ifp->if_dname, CTLFLAG_RD, 0, "MLX5 ethernet - interface name");
3560 if (priv->sysctl_ifnet == NULL) {
3561 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3562 goto err_free_sysctl;
3564 snprintf(unit, sizeof(unit), "%d", ifp->if_dunit);
3565 priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3566 OID_AUTO, unit, CTLFLAG_RD, 0, "MLX5 ethernet - interface unit");
3567 if (priv->sysctl_ifnet == NULL) {
3568 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3569 goto err_free_sysctl;
3572 /* HW sysctl tree */
3573 child = SYSCTL_CHILDREN(device_get_sysctl_tree(mdev->pdev->dev.bsddev));
3574 priv->sysctl_hw = SYSCTL_ADD_NODE(&priv->sysctl_ctx, child,
3575 OID_AUTO, "hw", CTLFLAG_RD, 0, "MLX5 ethernet dev hw");
3576 if (priv->sysctl_hw == NULL) {
3577 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3578 goto err_free_sysctl;
3581 err = mlx5e_build_ifp_priv(mdev, priv, ncv);
3583 mlx5_core_err(mdev, "mlx5e_build_ifp_priv() failed (%d)\n", err);
3584 goto err_free_sysctl;
3587 snprintf(unit, sizeof(unit), "mce%u_wq",
3588 device_get_unit(mdev->pdev->dev.bsddev));
3589 priv->wq = alloc_workqueue(unit, 0, 1);
3590 if (priv->wq == NULL) {
3591 if_printf(ifp, "%s: alloc_workqueue failed\n", __func__);
3592 goto err_free_sysctl;
3595 err = mlx5_alloc_map_uar(mdev, &priv->cq_uar);
3597 if_printf(ifp, "%s: mlx5_alloc_map_uar failed, %d\n",
3601 err = mlx5_core_alloc_pd(mdev, &priv->pdn);
3603 if_printf(ifp, "%s: mlx5_core_alloc_pd failed, %d\n",
3605 goto err_unmap_free_uar;
3607 err = mlx5_alloc_transport_domain(mdev, &priv->tdn);
3609 if_printf(ifp, "%s: mlx5_alloc_transport_domain failed, %d\n",
3611 goto err_dealloc_pd;
3613 err = mlx5e_create_mkey(priv, priv->pdn, &priv->mr);
3615 if_printf(ifp, "%s: mlx5e_create_mkey failed, %d\n",
3617 goto err_dealloc_transport_domain;
3619 mlx5_query_nic_vport_mac_address(priv->mdev, 0, dev_addr);
3621 /* check if we should generate a random MAC address */
3622 if (MLX5_CAP_GEN(priv->mdev, vport_group_manager) == 0 &&
3623 is_zero_ether_addr(dev_addr)) {
3624 random_ether_addr(dev_addr);
3625 if_printf(ifp, "Assigned random MAC address\n");
3628 /* set default MTU */
3629 mlx5e_set_dev_port_mtu(ifp, ifp->if_mtu);
3631 /* Set default media status */
3632 priv->media_status_last = IFM_AVALID;
3633 priv->media_active_last = IFM_ETHER | IFM_AUTO |
3634 IFM_ETH_RXPAUSE | IFM_FDX;
3636 /* setup default pauseframes configuration */
3637 mlx5e_setup_pauseframes(priv);
3639 err = mlx5_query_port_proto_cap(mdev, ð_proto_cap, MLX5_PTYS_EN);
3642 if_printf(ifp, "%s: Query port media capability failed, %d\n",
3646 /* Setup supported medias */
3647 ifmedia_init(&priv->media, IFM_IMASK | IFM_ETH_FMASK,
3648 mlx5e_media_change, mlx5e_media_status);
3650 for (i = 0; i < MLX5E_LINK_MODES_NUMBER; ++i) {
3651 if (mlx5e_mode_table[i].baudrate == 0)
3653 if (MLX5E_PROT_MASK(i) & eth_proto_cap) {
3654 ifmedia_add(&priv->media,
3655 mlx5e_mode_table[i].subtype |
3656 IFM_ETHER, 0, NULL);
3657 ifmedia_add(&priv->media,
3658 mlx5e_mode_table[i].subtype |
3659 IFM_ETHER | IFM_FDX |
3660 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3664 /* Additional supported medias */
3665 ifmedia_add(&priv->media, IFM_10G_LR | IFM_ETHER, 0, NULL);
3666 ifmedia_add(&priv->media, IFM_10G_LR |
3667 IFM_ETHER | IFM_FDX |
3668 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3670 ifmedia_add(&priv->media, IFM_40G_ER4 | IFM_ETHER, 0, NULL);
3671 ifmedia_add(&priv->media, IFM_40G_ER4 |
3672 IFM_ETHER | IFM_FDX |
3673 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3675 ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO, 0, NULL);
3676 ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
3677 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3679 /* Set autoselect by default */
3680 ifmedia_set(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
3681 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE);
3682 ether_ifattach(ifp, dev_addr);
3684 /* Register for VLAN events */
3685 priv->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
3686 mlx5e_vlan_rx_add_vid, priv, EVENTHANDLER_PRI_FIRST);
3687 priv->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
3688 mlx5e_vlan_rx_kill_vid, priv, EVENTHANDLER_PRI_FIRST);
3690 /* Link is down by default */
3691 if_link_state_change(ifp, LINK_STATE_DOWN);
3693 mlx5e_enable_async_events(priv);
3695 mlx5e_add_hw_stats(priv);
3697 mlx5e_create_stats(&priv->stats.vport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3698 "vstats", mlx5e_vport_stats_desc, MLX5E_VPORT_STATS_NUM,
3699 priv->stats.vport.arg);
3701 mlx5e_create_stats(&priv->stats.pport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3702 "pstats", mlx5e_pport_stats_desc, MLX5E_PPORT_STATS_NUM,
3703 priv->stats.pport.arg);
3705 mlx5e_create_ethtool(priv);
3707 mtx_lock(&priv->async_events_mtx);
3708 mlx5e_update_stats(priv);
3709 mtx_unlock(&priv->async_events_mtx);
3713 err_dealloc_transport_domain:
3714 mlx5_dealloc_transport_domain(mdev, priv->tdn);
3717 mlx5_core_dealloc_pd(mdev, priv->pdn);
3720 mlx5_unmap_free_uar(mdev, &priv->cq_uar);
3723 destroy_workqueue(priv->wq);
3726 sysctl_ctx_free(&priv->sysctl_ctx);
3727 if (priv->sysctl_debug)
3728 sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
3732 mlx5e_priv_mtx_destroy(priv);
3733 free(priv, M_MLX5EN);
3738 mlx5e_destroy_ifp(struct mlx5_core_dev *mdev, void *vpriv)
3740 struct mlx5e_priv *priv = vpriv;
3741 struct ifnet *ifp = priv->ifp;
3743 /* don't allow more IOCTLs */
3746 /* XXX wait a bit to allow IOCTL handlers to complete */
3749 /* stop watchdog timer */
3750 callout_drain(&priv->watchdog);
3752 if (priv->vlan_attach != NULL)
3753 EVENTHANDLER_DEREGISTER(vlan_config, priv->vlan_attach);
3754 if (priv->vlan_detach != NULL)
3755 EVENTHANDLER_DEREGISTER(vlan_unconfig, priv->vlan_detach);
3757 /* make sure device gets closed */
3759 mlx5e_close_locked(ifp);
3762 /* unregister device */
3763 ifmedia_removeall(&priv->media);
3764 ether_ifdetach(ifp);
3767 /* destroy all remaining sysctl nodes */
3768 sysctl_ctx_free(&priv->stats.vport.ctx);
3769 sysctl_ctx_free(&priv->stats.pport.ctx);
3770 if (priv->sysctl_debug)
3771 sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
3772 sysctl_ctx_free(&priv->sysctl_ctx);
3774 mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
3775 mlx5_dealloc_transport_domain(priv->mdev, priv->tdn);
3776 mlx5_core_dealloc_pd(priv->mdev, priv->pdn);
3777 mlx5_unmap_free_uar(priv->mdev, &priv->cq_uar);
3778 mlx5e_disable_async_events(priv);
3779 destroy_workqueue(priv->wq);
3780 mlx5e_priv_mtx_destroy(priv);
3781 free(priv, M_MLX5EN);
3785 mlx5e_get_ifp(void *vpriv)
3787 struct mlx5e_priv *priv = vpriv;
3792 static struct mlx5_interface mlx5e_interface = {
3793 .add = mlx5e_create_ifp,
3794 .remove = mlx5e_destroy_ifp,
3795 .event = mlx5e_async_event,
3796 .protocol = MLX5_INTERFACE_PROTOCOL_ETH,
3797 .get_dev = mlx5e_get_ifp,
3803 mlx5_register_interface(&mlx5e_interface);
3809 mlx5_unregister_interface(&mlx5e_interface);
3813 mlx5e_show_version(void __unused *arg)
3816 printf("%s", mlx5e_version);
3818 SYSINIT(mlx5e_show_version, SI_SUB_DRIVERS, SI_ORDER_ANY, mlx5e_show_version, NULL);
3820 module_init_order(mlx5e_init, SI_ORDER_THIRD);
3821 module_exit_order(mlx5e_cleanup, SI_ORDER_THIRD);
3823 #if (__FreeBSD_version >= 1100000)
3824 MODULE_DEPEND(mlx5en, linuxkpi, 1, 1, 1);
3826 MODULE_DEPEND(mlx5en, mlx5, 1, 1, 1);
3827 MODULE_VERSION(mlx5en, 1);