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 void mlx5e_calibration_callout(void *arg);
743 static int mlx5e_calibration_duration = 20;
744 static int mlx5e_fast_calibration = 1;
745 static int mlx5e_normal_calibration = 30;
747 static SYSCTL_NODE(_hw_mlx5, OID_AUTO, calibr, CTLFLAG_RW, 0,
748 "MLX5 timestamp calibration parameteres");
750 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, duration, CTLFLAG_RWTUN,
751 &mlx5e_calibration_duration, 0,
752 "Duration of initial calibration");
753 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, fast, CTLFLAG_RWTUN,
754 &mlx5e_fast_calibration, 0,
755 "Recalibration interval during initial calibration");
756 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, normal, CTLFLAG_RWTUN,
757 &mlx5e_normal_calibration, 0,
758 "Recalibration interval during normal operations");
761 * Ignites the calibration process.
764 mlx5e_reset_calibration_callout(struct mlx5e_priv *priv)
767 if (priv->clbr_done == 0)
768 mlx5e_calibration_callout(priv);
770 callout_reset_curcpu(&priv->tstmp_clbr, (priv->clbr_done <
771 mlx5e_calibration_duration ? mlx5e_fast_calibration :
772 mlx5e_normal_calibration) * hz, mlx5e_calibration_callout,
777 mlx5e_timespec2usec(const struct timespec *ts)
780 return ((uint64_t)ts->tv_sec * 1000000000 + ts->tv_nsec);
784 mlx5e_hw_clock(struct mlx5e_priv *priv)
786 struct mlx5_init_seg *iseg;
787 uint32_t hw_h, hw_h1, hw_l;
789 iseg = priv->mdev->iseg;
791 hw_h = ioread32be(&iseg->internal_timer_h);
792 hw_l = ioread32be(&iseg->internal_timer_l);
793 hw_h1 = ioread32be(&iseg->internal_timer_h);
794 } while (hw_h1 != hw_h);
795 return (((uint64_t)hw_h << 32) | hw_l);
799 * The calibration callout, it runs either in the context of the
800 * thread which enables calibration, or in callout. It takes the
801 * snapshot of system and adapter clocks, then advances the pointers to
802 * the calibration point to allow rx path to read the consistent data
806 mlx5e_calibration_callout(void *arg)
808 struct mlx5e_priv *priv;
809 struct mlx5e_clbr_point *next, *curr;
814 curr = &priv->clbr_points[priv->clbr_curr];
815 clbr_curr_next = priv->clbr_curr + 1;
816 if (clbr_curr_next >= nitems(priv->clbr_points))
818 next = &priv->clbr_points[clbr_curr_next];
820 next->base_prev = curr->base_curr;
821 next->clbr_hw_prev = curr->clbr_hw_curr;
823 next->clbr_hw_curr = mlx5e_hw_clock(priv);
824 if (((next->clbr_hw_curr - curr->clbr_hw_prev) >> MLX5E_TSTMP_PREC) ==
826 if_printf(priv->ifp, "HW failed tstmp frozen %#jx %#jx,"
827 "disabling\n", next->clbr_hw_curr, curr->clbr_hw_prev);
833 next->base_curr = mlx5e_timespec2usec(&ts);
836 atomic_thread_fence_rel();
837 priv->clbr_curr = clbr_curr_next;
838 atomic_store_rel_int(&next->clbr_gen, ++(priv->clbr_gen));
840 if (priv->clbr_done < mlx5e_calibration_duration)
842 mlx5e_reset_calibration_callout(priv);
845 static const char *mlx5e_rq_stats_desc[] = {
846 MLX5E_RQ_STATS(MLX5E_STATS_DESC)
850 mlx5e_create_rq(struct mlx5e_channel *c,
851 struct mlx5e_rq_param *param,
854 struct mlx5e_priv *priv = c->priv;
855 struct mlx5_core_dev *mdev = priv->mdev;
857 void *rqc = param->rqc;
858 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
864 err = mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
868 /* Create DMA descriptor TAG */
869 if ((err = -bus_dma_tag_create(
870 bus_get_dma_tag(mdev->pdev->dev.bsddev),
871 1, /* any alignment */
873 BUS_SPACE_MAXADDR, /* lowaddr */
874 BUS_SPACE_MAXADDR, /* highaddr */
875 NULL, NULL, /* filter, filterarg */
876 nsegs * MLX5E_MAX_RX_BYTES, /* maxsize */
877 nsegs, /* nsegments */
878 nsegs * MLX5E_MAX_RX_BYTES, /* maxsegsize */
880 NULL, NULL, /* lockfunc, lockfuncarg */
884 err = mlx5_wq_ll_create(mdev, ¶m->wq, rqc_wq, &rq->wq,
887 goto err_free_dma_tag;
889 rq->wq.db = &rq->wq.db[MLX5_RCV_DBR];
891 err = mlx5e_get_wqe_sz(priv, &rq->wqe_sz, &rq->nsegs);
893 goto err_rq_wq_destroy;
895 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
897 err = -tcp_lro_init_args(&rq->lro, c->tag.m_snd_tag.ifp, TCP_LRO_ENTRIES, wq_sz);
899 goto err_rq_wq_destroy;
901 rq->mbuf = malloc(wq_sz * sizeof(rq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
902 for (i = 0; i != wq_sz; i++) {
903 struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
904 #if (MLX5E_MAX_RX_SEGS == 1)
905 uint32_t byte_count = rq->wqe_sz - MLX5E_NET_IP_ALIGN;
910 err = -bus_dmamap_create(rq->dma_tag, 0, &rq->mbuf[i].dma_map);
913 bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
914 goto err_rq_mbuf_free;
917 /* set value for constant fields */
918 #if (MLX5E_MAX_RX_SEGS == 1)
919 wqe->data[0].lkey = c->mkey_be;
920 wqe->data[0].byte_count = cpu_to_be32(byte_count | MLX5_HW_START_PADDING);
922 for (j = 0; j < rq->nsegs; j++)
923 wqe->data[j].lkey = c->mkey_be;
927 INIT_WORK(&rq->dim.work, mlx5e_dim_work);
928 if (priv->params.rx_cq_moderation_mode < 2) {
929 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
931 void *cqc = container_of(param,
932 struct mlx5e_channel_param, rq)->rx_cq.cqc;
934 switch (MLX5_GET(cqc, cqc, cq_period_mode)) {
935 case MLX5_CQ_PERIOD_MODE_START_FROM_EQE:
936 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
938 case MLX5_CQ_PERIOD_MODE_START_FROM_CQE:
939 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
942 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
947 rq->ifp = c->tag.m_snd_tag.ifp;
951 snprintf(buffer, sizeof(buffer), "rxstat%d", c->ix);
952 mlx5e_create_stats(&rq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
953 buffer, mlx5e_rq_stats_desc, MLX5E_RQ_STATS_NUM,
958 free(rq->mbuf, M_MLX5EN);
959 tcp_lro_free(&rq->lro);
961 mlx5_wq_destroy(&rq->wq_ctrl);
963 bus_dma_tag_destroy(rq->dma_tag);
969 mlx5e_destroy_rq(struct mlx5e_rq *rq)
974 /* destroy all sysctl nodes */
975 sysctl_ctx_free(&rq->stats.ctx);
977 /* free leftover LRO packets, if any */
978 tcp_lro_free(&rq->lro);
980 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
981 for (i = 0; i != wq_sz; i++) {
982 if (rq->mbuf[i].mbuf != NULL) {
983 bus_dmamap_unload(rq->dma_tag, rq->mbuf[i].dma_map);
984 m_freem(rq->mbuf[i].mbuf);
986 bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
988 free(rq->mbuf, M_MLX5EN);
989 mlx5_wq_destroy(&rq->wq_ctrl);
993 mlx5e_enable_rq(struct mlx5e_rq *rq, struct mlx5e_rq_param *param)
995 struct mlx5e_channel *c = rq->channel;
996 struct mlx5e_priv *priv = c->priv;
997 struct mlx5_core_dev *mdev = priv->mdev;
1005 inlen = MLX5_ST_SZ_BYTES(create_rq_in) +
1006 sizeof(u64) * rq->wq_ctrl.buf.npages;
1007 in = mlx5_vzalloc(inlen);
1011 rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
1012 wq = MLX5_ADDR_OF(rqc, rqc, wq);
1014 memcpy(rqc, param->rqc, sizeof(param->rqc));
1016 MLX5_SET(rqc, rqc, cqn, c->rq.cq.mcq.cqn);
1017 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST);
1018 MLX5_SET(rqc, rqc, flush_in_error_en, 1);
1019 if (priv->counter_set_id >= 0)
1020 MLX5_SET(rqc, rqc, counter_set_id, priv->counter_set_id);
1021 MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift -
1023 MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma);
1025 mlx5_fill_page_array(&rq->wq_ctrl.buf,
1026 (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
1028 err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn);
1036 mlx5e_modify_rq(struct mlx5e_rq *rq, int curr_state, int next_state)
1038 struct mlx5e_channel *c = rq->channel;
1039 struct mlx5e_priv *priv = c->priv;
1040 struct mlx5_core_dev *mdev = priv->mdev;
1047 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
1048 in = mlx5_vzalloc(inlen);
1052 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
1054 MLX5_SET(modify_rq_in, in, rqn, rq->rqn);
1055 MLX5_SET(modify_rq_in, in, rq_state, curr_state);
1056 MLX5_SET(rqc, rqc, state, next_state);
1058 err = mlx5_core_modify_rq(mdev, in, inlen);
1066 mlx5e_disable_rq(struct mlx5e_rq *rq)
1068 struct mlx5e_channel *c = rq->channel;
1069 struct mlx5e_priv *priv = c->priv;
1070 struct mlx5_core_dev *mdev = priv->mdev;
1072 mlx5_core_destroy_rq(mdev, rq->rqn);
1076 mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq)
1078 struct mlx5e_channel *c = rq->channel;
1079 struct mlx5e_priv *priv = c->priv;
1080 struct mlx5_wq_ll *wq = &rq->wq;
1083 for (i = 0; i < 1000; i++) {
1084 if (wq->cur_sz >= priv->params.min_rx_wqes)
1089 return (-ETIMEDOUT);
1093 mlx5e_open_rq(struct mlx5e_channel *c,
1094 struct mlx5e_rq_param *param,
1095 struct mlx5e_rq *rq)
1099 err = mlx5e_create_rq(c, param, rq);
1103 err = mlx5e_enable_rq(rq, param);
1105 goto err_destroy_rq;
1107 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
1109 goto err_disable_rq;
1116 mlx5e_disable_rq(rq);
1118 mlx5e_destroy_rq(rq);
1124 mlx5e_close_rq(struct mlx5e_rq *rq)
1128 callout_stop(&rq->watchdog);
1129 mtx_unlock(&rq->mtx);
1131 callout_drain(&rq->watchdog);
1133 mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
1137 mlx5e_close_rq_wait(struct mlx5e_rq *rq)
1139 struct mlx5_core_dev *mdev = rq->channel->priv->mdev;
1141 /* wait till RQ is empty */
1142 while (!mlx5_wq_ll_is_empty(&rq->wq) &&
1143 (mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR)) {
1145 rq->cq.mcq.comp(&rq->cq.mcq);
1148 cancel_work_sync(&rq->dim.work);
1149 mlx5e_disable_rq(rq);
1150 mlx5e_destroy_rq(rq);
1154 mlx5e_free_sq_db(struct mlx5e_sq *sq)
1156 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1159 for (x = 0; x != wq_sz; x++)
1160 bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1161 free(sq->mbuf, M_MLX5EN);
1165 mlx5e_alloc_sq_db(struct mlx5e_sq *sq)
1167 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1171 sq->mbuf = malloc(wq_sz * sizeof(sq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
1173 /* Create DMA descriptor MAPs */
1174 for (x = 0; x != wq_sz; x++) {
1175 err = -bus_dmamap_create(sq->dma_tag, 0, &sq->mbuf[x].dma_map);
1178 bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1179 free(sq->mbuf, M_MLX5EN);
1186 static const char *mlx5e_sq_stats_desc[] = {
1187 MLX5E_SQ_STATS(MLX5E_STATS_DESC)
1191 mlx5e_update_sq_inline(struct mlx5e_sq *sq)
1193 sq->max_inline = sq->priv->params.tx_max_inline;
1194 sq->min_inline_mode = sq->priv->params.tx_min_inline_mode;
1197 * Check if trust state is DSCP or if inline mode is NONE which
1198 * indicates CX-5 or newer hardware.
1200 if (sq->priv->params_ethtool.trust_state != MLX5_QPTS_TRUST_PCP ||
1201 sq->min_inline_mode == MLX5_INLINE_MODE_NONE) {
1202 if (MLX5_CAP_ETH(sq->priv->mdev, wqe_vlan_insert))
1203 sq->min_insert_caps = MLX5E_INSERT_VLAN | MLX5E_INSERT_NON_VLAN;
1205 sq->min_insert_caps = MLX5E_INSERT_NON_VLAN;
1207 sq->min_insert_caps = 0;
1212 mlx5e_refresh_sq_inline_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
1216 for (i = 0; i != c->num_tc; i++) {
1217 mtx_lock(&c->sq[i].lock);
1218 mlx5e_update_sq_inline(&c->sq[i]);
1219 mtx_unlock(&c->sq[i].lock);
1224 mlx5e_refresh_sq_inline(struct mlx5e_priv *priv)
1228 /* check if channels are closed */
1229 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
1232 for (i = 0; i < priv->params.num_channels; i++)
1233 mlx5e_refresh_sq_inline_sub(priv, &priv->channel[i]);
1237 mlx5e_create_sq(struct mlx5e_channel *c,
1239 struct mlx5e_sq_param *param,
1240 struct mlx5e_sq *sq)
1242 struct mlx5e_priv *priv = c->priv;
1243 struct mlx5_core_dev *mdev = priv->mdev;
1245 void *sqc = param->sqc;
1246 void *sqc_wq = MLX5_ADDR_OF(sqc, sqc, wq);
1249 /* Create DMA descriptor TAG */
1250 if ((err = -bus_dma_tag_create(
1251 bus_get_dma_tag(mdev->pdev->dev.bsddev),
1252 1, /* any alignment */
1253 0, /* no boundary */
1254 BUS_SPACE_MAXADDR, /* lowaddr */
1255 BUS_SPACE_MAXADDR, /* highaddr */
1256 NULL, NULL, /* filter, filterarg */
1257 MLX5E_MAX_TX_PAYLOAD_SIZE, /* maxsize */
1258 MLX5E_MAX_TX_MBUF_FRAGS, /* nsegments */
1259 MLX5E_MAX_TX_MBUF_SIZE, /* maxsegsize */
1261 NULL, NULL, /* lockfunc, lockfuncarg */
1265 err = mlx5_alloc_map_uar(mdev, &sq->uar);
1267 goto err_free_dma_tag;
1269 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, &sq->wq,
1272 goto err_unmap_free_uar;
1274 sq->wq.db = &sq->wq.db[MLX5_SND_DBR];
1275 sq->bf_buf_size = (1 << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2;
1277 err = mlx5e_alloc_sq_db(sq);
1279 goto err_sq_wq_destroy;
1281 sq->mkey_be = c->mkey_be;
1282 sq->ifp = priv->ifp;
1286 mlx5e_update_sq_inline(sq);
1288 snprintf(buffer, sizeof(buffer), "txstat%dtc%d", c->ix, tc);
1289 mlx5e_create_stats(&sq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
1290 buffer, mlx5e_sq_stats_desc, MLX5E_SQ_STATS_NUM,
1296 mlx5_wq_destroy(&sq->wq_ctrl);
1299 mlx5_unmap_free_uar(mdev, &sq->uar);
1302 bus_dma_tag_destroy(sq->dma_tag);
1308 mlx5e_destroy_sq(struct mlx5e_sq *sq)
1310 /* destroy all sysctl nodes */
1311 sysctl_ctx_free(&sq->stats.ctx);
1313 mlx5e_free_sq_db(sq);
1314 mlx5_wq_destroy(&sq->wq_ctrl);
1315 mlx5_unmap_free_uar(sq->priv->mdev, &sq->uar);
1319 mlx5e_enable_sq(struct mlx5e_sq *sq, struct mlx5e_sq_param *param,
1328 inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
1329 sizeof(u64) * sq->wq_ctrl.buf.npages;
1330 in = mlx5_vzalloc(inlen);
1334 sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
1335 wq = MLX5_ADDR_OF(sqc, sqc, wq);
1337 memcpy(sqc, param->sqc, sizeof(param->sqc));
1339 MLX5_SET(sqc, sqc, tis_num_0, tis_num);
1340 MLX5_SET(sqc, sqc, cqn, sq->cq.mcq.cqn);
1341 MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
1342 MLX5_SET(sqc, sqc, tis_lst_sz, 1);
1343 MLX5_SET(sqc, sqc, flush_in_error_en, 1);
1345 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
1346 MLX5_SET(wq, wq, uar_page, sq->uar.index);
1347 MLX5_SET(wq, wq, log_wq_pg_sz, sq->wq_ctrl.buf.page_shift -
1349 MLX5_SET64(wq, wq, dbr_addr, sq->wq_ctrl.db.dma);
1351 mlx5_fill_page_array(&sq->wq_ctrl.buf,
1352 (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
1354 err = mlx5_core_create_sq(sq->priv->mdev, in, inlen, &sq->sqn);
1362 mlx5e_modify_sq(struct mlx5e_sq *sq, int curr_state, int next_state)
1369 inlen = MLX5_ST_SZ_BYTES(modify_sq_in);
1370 in = mlx5_vzalloc(inlen);
1374 sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx);
1376 MLX5_SET(modify_sq_in, in, sqn, sq->sqn);
1377 MLX5_SET(modify_sq_in, in, sq_state, curr_state);
1378 MLX5_SET(sqc, sqc, state, next_state);
1380 err = mlx5_core_modify_sq(sq->priv->mdev, in, inlen);
1388 mlx5e_disable_sq(struct mlx5e_sq *sq)
1391 mlx5_core_destroy_sq(sq->priv->mdev, sq->sqn);
1395 mlx5e_open_sq(struct mlx5e_channel *c,
1397 struct mlx5e_sq_param *param,
1398 struct mlx5e_sq *sq)
1402 err = mlx5e_create_sq(c, tc, param, sq);
1406 err = mlx5e_enable_sq(sq, param, c->priv->tisn[tc]);
1408 goto err_destroy_sq;
1410 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY);
1412 goto err_disable_sq;
1414 WRITE_ONCE(sq->running, 1);
1419 mlx5e_disable_sq(sq);
1421 mlx5e_destroy_sq(sq);
1427 mlx5e_sq_send_nops_locked(struct mlx5e_sq *sq, int can_sleep)
1429 /* fill up remainder with NOPs */
1430 while (sq->cev_counter != 0) {
1431 while (!mlx5e_sq_has_room_for(sq, 1)) {
1432 if (can_sleep != 0) {
1433 mtx_unlock(&sq->lock);
1435 mtx_lock(&sq->lock);
1440 /* send a single NOP */
1441 mlx5e_send_nop(sq, 1);
1442 atomic_thread_fence_rel();
1445 /* Check if we need to write the doorbell */
1446 if (likely(sq->doorbell.d64 != 0)) {
1447 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
1448 sq->doorbell.d64 = 0;
1453 mlx5e_sq_cev_timeout(void *arg)
1455 struct mlx5e_sq *sq = arg;
1457 mtx_assert(&sq->lock, MA_OWNED);
1459 /* check next state */
1460 switch (sq->cev_next_state) {
1461 case MLX5E_CEV_STATE_SEND_NOPS:
1462 /* fill TX ring with NOPs, if any */
1463 mlx5e_sq_send_nops_locked(sq, 0);
1465 /* check if completed */
1466 if (sq->cev_counter == 0) {
1467 sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
1472 /* send NOPs on next timeout */
1473 sq->cev_next_state = MLX5E_CEV_STATE_SEND_NOPS;
1478 callout_reset_curcpu(&sq->cev_callout, hz, mlx5e_sq_cev_timeout, sq);
1482 mlx5e_drain_sq(struct mlx5e_sq *sq)
1485 struct mlx5_core_dev *mdev= sq->priv->mdev;
1488 * Check if already stopped.
1490 * NOTE: Serialization of this function is managed by the
1491 * caller ensuring the priv's state lock is locked or in case
1492 * of rate limit support, a single thread manages drain and
1493 * resume of SQs. The "running" variable can therefore safely
1494 * be read without any locks.
1496 if (READ_ONCE(sq->running) == 0)
1499 /* don't put more packets into the SQ */
1500 WRITE_ONCE(sq->running, 0);
1502 /* serialize access to DMA rings */
1503 mtx_lock(&sq->lock);
1505 /* teardown event factor timer, if any */
1506 sq->cev_next_state = MLX5E_CEV_STATE_HOLD_NOPS;
1507 callout_stop(&sq->cev_callout);
1509 /* send dummy NOPs in order to flush the transmit ring */
1510 mlx5e_sq_send_nops_locked(sq, 1);
1511 mtx_unlock(&sq->lock);
1513 /* make sure it is safe to free the callout */
1514 callout_drain(&sq->cev_callout);
1516 /* wait till SQ is empty or link is down */
1517 mtx_lock(&sq->lock);
1518 while (sq->cc != sq->pc &&
1519 (sq->priv->media_status_last & IFM_ACTIVE) != 0 &&
1520 mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1521 mtx_unlock(&sq->lock);
1523 sq->cq.mcq.comp(&sq->cq.mcq);
1524 mtx_lock(&sq->lock);
1526 mtx_unlock(&sq->lock);
1528 /* error out remaining requests */
1529 error = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RDY, MLX5_SQC_STATE_ERR);
1532 "mlx5e_modify_sq() from RDY to ERR failed: %d\n", error);
1535 /* wait till SQ is empty */
1536 mtx_lock(&sq->lock);
1537 while (sq->cc != sq->pc &&
1538 mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1539 mtx_unlock(&sq->lock);
1541 sq->cq.mcq.comp(&sq->cq.mcq);
1542 mtx_lock(&sq->lock);
1544 mtx_unlock(&sq->lock);
1548 mlx5e_close_sq_wait(struct mlx5e_sq *sq)
1552 mlx5e_disable_sq(sq);
1553 mlx5e_destroy_sq(sq);
1557 mlx5e_create_cq(struct mlx5e_priv *priv,
1558 struct mlx5e_cq_param *param,
1559 struct mlx5e_cq *cq,
1560 mlx5e_cq_comp_t *comp,
1563 struct mlx5_core_dev *mdev = priv->mdev;
1564 struct mlx5_core_cq *mcq = &cq->mcq;
1570 param->wq.buf_numa_node = 0;
1571 param->wq.db_numa_node = 0;
1573 err = mlx5_cqwq_create(mdev, ¶m->wq, param->cqc, &cq->wq,
1578 mlx5_vector2eqn(mdev, eq_ix, &eqn_not_used, &irqn);
1581 mcq->set_ci_db = cq->wq_ctrl.db.db;
1582 mcq->arm_db = cq->wq_ctrl.db.db + 1;
1583 *mcq->set_ci_db = 0;
1585 mcq->vector = eq_ix;
1587 mcq->event = mlx5e_cq_error_event;
1589 mcq->uar = &priv->cq_uar;
1591 for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) {
1592 struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i);
1603 mlx5e_destroy_cq(struct mlx5e_cq *cq)
1605 mlx5_wq_destroy(&cq->wq_ctrl);
1609 mlx5e_enable_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param, int eq_ix)
1611 struct mlx5_core_cq *mcq = &cq->mcq;
1619 inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
1620 sizeof(u64) * cq->wq_ctrl.buf.npages;
1621 in = mlx5_vzalloc(inlen);
1625 cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
1627 memcpy(cqc, param->cqc, sizeof(param->cqc));
1629 mlx5_fill_page_array(&cq->wq_ctrl.buf,
1630 (__be64 *) MLX5_ADDR_OF(create_cq_in, in, pas));
1632 mlx5_vector2eqn(cq->priv->mdev, eq_ix, &eqn, &irqn_not_used);
1634 MLX5_SET(cqc, cqc, c_eqn, eqn);
1635 MLX5_SET(cqc, cqc, uar_page, mcq->uar->index);
1636 MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.buf.page_shift -
1638 MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
1640 err = mlx5_core_create_cq(cq->priv->mdev, mcq, in, inlen);
1647 mlx5e_cq_arm(cq, MLX5_GET_DOORBELL_LOCK(&cq->priv->doorbell_lock));
1653 mlx5e_disable_cq(struct mlx5e_cq *cq)
1656 mlx5_core_destroy_cq(cq->priv->mdev, &cq->mcq);
1660 mlx5e_open_cq(struct mlx5e_priv *priv,
1661 struct mlx5e_cq_param *param,
1662 struct mlx5e_cq *cq,
1663 mlx5e_cq_comp_t *comp,
1668 err = mlx5e_create_cq(priv, param, cq, comp, eq_ix);
1672 err = mlx5e_enable_cq(cq, param, eq_ix);
1674 goto err_destroy_cq;
1679 mlx5e_destroy_cq(cq);
1685 mlx5e_close_cq(struct mlx5e_cq *cq)
1687 mlx5e_disable_cq(cq);
1688 mlx5e_destroy_cq(cq);
1692 mlx5e_open_tx_cqs(struct mlx5e_channel *c,
1693 struct mlx5e_channel_param *cparam)
1698 for (tc = 0; tc < c->num_tc; tc++) {
1699 /* open completion queue */
1700 err = mlx5e_open_cq(c->priv, &cparam->tx_cq, &c->sq[tc].cq,
1701 &mlx5e_tx_cq_comp, c->ix);
1703 goto err_close_tx_cqs;
1708 for (tc--; tc >= 0; tc--)
1709 mlx5e_close_cq(&c->sq[tc].cq);
1715 mlx5e_close_tx_cqs(struct mlx5e_channel *c)
1719 for (tc = 0; tc < c->num_tc; tc++)
1720 mlx5e_close_cq(&c->sq[tc].cq);
1724 mlx5e_open_sqs(struct mlx5e_channel *c,
1725 struct mlx5e_channel_param *cparam)
1730 for (tc = 0; tc < c->num_tc; tc++) {
1731 err = mlx5e_open_sq(c, tc, &cparam->sq, &c->sq[tc]);
1739 for (tc--; tc >= 0; tc--)
1740 mlx5e_close_sq_wait(&c->sq[tc]);
1746 mlx5e_close_sqs_wait(struct mlx5e_channel *c)
1750 for (tc = 0; tc < c->num_tc; tc++)
1751 mlx5e_close_sq_wait(&c->sq[tc]);
1755 mlx5e_chan_mtx_init(struct mlx5e_channel *c)
1759 mtx_init(&c->rq.mtx, "mlx5rx", MTX_NETWORK_LOCK, MTX_DEF);
1761 callout_init_mtx(&c->rq.watchdog, &c->rq.mtx, 0);
1763 for (tc = 0; tc < c->num_tc; tc++) {
1764 struct mlx5e_sq *sq = c->sq + tc;
1766 mtx_init(&sq->lock, "mlx5tx",
1767 MTX_NETWORK_LOCK " TX", MTX_DEF);
1768 mtx_init(&sq->comp_lock, "mlx5comp",
1769 MTX_NETWORK_LOCK " TX", MTX_DEF);
1771 callout_init_mtx(&sq->cev_callout, &sq->lock, 0);
1773 sq->cev_factor = c->priv->params_ethtool.tx_completion_fact;
1775 /* ensure the TX completion event factor is not zero */
1776 if (sq->cev_factor == 0)
1782 mlx5e_chan_mtx_destroy(struct mlx5e_channel *c)
1786 mtx_destroy(&c->rq.mtx);
1788 for (tc = 0; tc < c->num_tc; tc++) {
1789 mtx_destroy(&c->sq[tc].lock);
1790 mtx_destroy(&c->sq[tc].comp_lock);
1795 mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
1796 struct mlx5e_channel_param *cparam,
1797 struct mlx5e_channel *c)
1801 memset(c, 0, sizeof(*c));
1805 /* setup send tag */
1806 c->tag.m_snd_tag.ifp = priv->ifp;
1807 c->tag.type = IF_SND_TAG_TYPE_UNLIMITED;
1808 c->mkey_be = cpu_to_be32(priv->mr.key);
1809 c->num_tc = priv->num_tc;
1812 mlx5e_chan_mtx_init(c);
1814 /* open transmit completion queue */
1815 err = mlx5e_open_tx_cqs(c, cparam);
1819 /* open receive completion queue */
1820 err = mlx5e_open_cq(c->priv, &cparam->rx_cq, &c->rq.cq,
1821 &mlx5e_rx_cq_comp, c->ix);
1823 goto err_close_tx_cqs;
1825 err = mlx5e_open_sqs(c, cparam);
1827 goto err_close_rx_cq;
1829 err = mlx5e_open_rq(c, &cparam->rq, &c->rq);
1833 /* poll receive queue initially */
1834 c->rq.cq.mcq.comp(&c->rq.cq.mcq);
1839 mlx5e_close_sqs_wait(c);
1842 mlx5e_close_cq(&c->rq.cq);
1845 mlx5e_close_tx_cqs(c);
1848 /* destroy mutexes */
1849 mlx5e_chan_mtx_destroy(c);
1854 mlx5e_close_channel(struct mlx5e_channel *c)
1856 mlx5e_close_rq(&c->rq);
1860 mlx5e_close_channel_wait(struct mlx5e_channel *c)
1862 mlx5e_close_rq_wait(&c->rq);
1863 mlx5e_close_sqs_wait(c);
1864 mlx5e_close_cq(&c->rq.cq);
1865 mlx5e_close_tx_cqs(c);
1866 /* destroy mutexes */
1867 mlx5e_chan_mtx_destroy(c);
1871 mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs)
1875 r = priv->params.hw_lro_en ? priv->params.lro_wqe_sz :
1876 MLX5E_SW2MB_MTU(priv->ifp->if_mtu);
1877 if (r > MJUM16BYTES)
1882 else if (r > MJUMPAGESIZE)
1884 else if (r > MCLBYTES)
1890 * n + 1 must be a power of two, because stride size must be.
1891 * Stride size is 16 * (n + 1), as the first segment is
1894 for (n = howmany(r, MLX5E_MAX_RX_BYTES); !powerof2(n + 1); n++)
1903 mlx5e_build_rq_param(struct mlx5e_priv *priv,
1904 struct mlx5e_rq_param *param)
1906 void *rqc = param->rqc;
1907 void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
1910 mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
1911 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
1912 MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN);
1913 MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe) +
1914 nsegs * sizeof(struct mlx5_wqe_data_seg)));
1915 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_rq_size);
1916 MLX5_SET(wq, wq, pd, priv->pdn);
1918 param->wq.buf_numa_node = 0;
1919 param->wq.db_numa_node = 0;
1920 param->wq.linear = 1;
1924 mlx5e_build_sq_param(struct mlx5e_priv *priv,
1925 struct mlx5e_sq_param *param)
1927 void *sqc = param->sqc;
1928 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
1930 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_sq_size);
1931 MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
1932 MLX5_SET(wq, wq, pd, priv->pdn);
1934 param->wq.buf_numa_node = 0;
1935 param->wq.db_numa_node = 0;
1936 param->wq.linear = 1;
1940 mlx5e_build_common_cq_param(struct mlx5e_priv *priv,
1941 struct mlx5e_cq_param *param)
1943 void *cqc = param->cqc;
1945 MLX5_SET(cqc, cqc, uar_page, priv->cq_uar.index);
1949 mlx5e_get_default_profile(struct mlx5e_priv *priv, int mode, struct net_dim_cq_moder *ptr)
1952 *ptr = net_dim_get_profile(mode, MLX5E_DIM_DEFAULT_PROFILE);
1954 /* apply LRO restrictions */
1955 if (priv->params.hw_lro_en &&
1956 ptr->pkts > MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO) {
1957 ptr->pkts = MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO;
1962 mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
1963 struct mlx5e_cq_param *param)
1965 struct net_dim_cq_moder curr;
1966 void *cqc = param->cqc;
1970 * TODO The sysctl to control on/off is a bool value for now, which means
1971 * we only support CSUM, once HASH is implemnted we'll need to address that.
1973 if (priv->params.cqe_zipping_en) {
1974 MLX5_SET(cqc, cqc, mini_cqe_res_format, MLX5_CQE_FORMAT_CSUM);
1975 MLX5_SET(cqc, cqc, cqe_compression_en, 1);
1978 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_rq_size);
1980 switch (priv->params.rx_cq_moderation_mode) {
1982 MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
1983 MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
1984 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1987 MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
1988 MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
1989 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
1990 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
1992 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1995 mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE, &curr);
1996 MLX5_SET(cqc, cqc, cq_period, curr.usec);
1997 MLX5_SET(cqc, cqc, cq_max_count, curr.pkts);
1998 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2001 mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE, &curr);
2002 MLX5_SET(cqc, cqc, cq_period, curr.usec);
2003 MLX5_SET(cqc, cqc, cq_max_count, curr.pkts);
2004 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
2005 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
2007 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2013 mlx5e_dim_build_cq_param(priv, param);
2015 mlx5e_build_common_cq_param(priv, param);
2019 mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
2020 struct mlx5e_cq_param *param)
2022 void *cqc = param->cqc;
2024 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_sq_size);
2025 MLX5_SET(cqc, cqc, cq_period, priv->params.tx_cq_moderation_usec);
2026 MLX5_SET(cqc, cqc, cq_max_count, priv->params.tx_cq_moderation_pkts);
2028 switch (priv->params.tx_cq_moderation_mode) {
2030 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2033 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
2034 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
2036 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2040 mlx5e_build_common_cq_param(priv, param);
2044 mlx5e_build_channel_param(struct mlx5e_priv *priv,
2045 struct mlx5e_channel_param *cparam)
2047 memset(cparam, 0, sizeof(*cparam));
2049 mlx5e_build_rq_param(priv, &cparam->rq);
2050 mlx5e_build_sq_param(priv, &cparam->sq);
2051 mlx5e_build_rx_cq_param(priv, &cparam->rx_cq);
2052 mlx5e_build_tx_cq_param(priv, &cparam->tx_cq);
2056 mlx5e_open_channels(struct mlx5e_priv *priv)
2058 struct mlx5e_channel_param cparam;
2063 mlx5e_build_channel_param(priv, &cparam);
2064 for (i = 0; i < priv->params.num_channels; i++) {
2065 err = mlx5e_open_channel(priv, i, &cparam, &priv->channel[i]);
2067 goto err_close_channels;
2070 for (j = 0; j < priv->params.num_channels; j++) {
2071 err = mlx5e_wait_for_min_rx_wqes(&priv->channel[j].rq);
2073 goto err_close_channels;
2079 mlx5e_close_channel(&priv->channel[i]);
2080 mlx5e_close_channel_wait(&priv->channel[i]);
2086 mlx5e_close_channels(struct mlx5e_priv *priv)
2090 for (i = 0; i < priv->params.num_channels; i++)
2091 mlx5e_close_channel(&priv->channel[i]);
2092 for (i = 0; i < priv->params.num_channels; i++)
2093 mlx5e_close_channel_wait(&priv->channel[i]);
2097 mlx5e_refresh_sq_params(struct mlx5e_priv *priv, struct mlx5e_sq *sq)
2100 if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
2103 switch (priv->params.tx_cq_moderation_mode) {
2106 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2109 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2113 return (mlx5_core_modify_cq_moderation_mode(priv->mdev, &sq->cq.mcq,
2114 priv->params.tx_cq_moderation_usec,
2115 priv->params.tx_cq_moderation_pkts,
2119 return (mlx5_core_modify_cq_moderation(priv->mdev, &sq->cq.mcq,
2120 priv->params.tx_cq_moderation_usec,
2121 priv->params.tx_cq_moderation_pkts));
2125 mlx5e_refresh_rq_params(struct mlx5e_priv *priv, struct mlx5e_rq *rq)
2128 if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
2133 switch (priv->params.rx_cq_moderation_mode) {
2136 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2137 dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
2140 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2141 dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
2145 /* tear down dynamic interrupt moderation */
2147 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
2148 mtx_unlock(&rq->mtx);
2150 /* wait for dynamic interrupt moderation work task, if any */
2151 cancel_work_sync(&rq->dim.work);
2153 if (priv->params.rx_cq_moderation_mode >= 2) {
2154 struct net_dim_cq_moder curr;
2156 mlx5e_get_default_profile(priv, dim_mode, &curr);
2158 retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2159 curr.usec, curr.pkts, cq_mode);
2161 /* set dynamic interrupt moderation mode and zero defaults */
2163 rq->dim.mode = dim_mode;
2165 rq->dim.profile_ix = MLX5E_DIM_DEFAULT_PROFILE;
2166 mtx_unlock(&rq->mtx);
2168 retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2169 priv->params.rx_cq_moderation_usec,
2170 priv->params.rx_cq_moderation_pkts,
2176 return (mlx5_core_modify_cq_moderation(priv->mdev, &rq->cq.mcq,
2177 priv->params.rx_cq_moderation_usec,
2178 priv->params.rx_cq_moderation_pkts));
2182 mlx5e_refresh_channel_params_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
2187 err = mlx5e_refresh_rq_params(priv, &c->rq);
2191 for (i = 0; i != c->num_tc; i++) {
2192 err = mlx5e_refresh_sq_params(priv, &c->sq[i]);
2201 mlx5e_refresh_channel_params(struct mlx5e_priv *priv)
2205 /* check if channels are closed */
2206 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2209 for (i = 0; i < priv->params.num_channels; i++) {
2212 err = mlx5e_refresh_channel_params_sub(priv, &priv->channel[i]);
2220 mlx5e_open_tis(struct mlx5e_priv *priv, int tc)
2222 struct mlx5_core_dev *mdev = priv->mdev;
2223 u32 in[MLX5_ST_SZ_DW(create_tis_in)];
2224 void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
2226 memset(in, 0, sizeof(in));
2228 MLX5_SET(tisc, tisc, prio, tc);
2229 MLX5_SET(tisc, tisc, transport_domain, priv->tdn);
2231 return (mlx5_core_create_tis(mdev, in, sizeof(in), &priv->tisn[tc]));
2235 mlx5e_close_tis(struct mlx5e_priv *priv, int tc)
2237 mlx5_core_destroy_tis(priv->mdev, priv->tisn[tc]);
2241 mlx5e_open_tises(struct mlx5e_priv *priv)
2243 int num_tc = priv->num_tc;
2247 for (tc = 0; tc < num_tc; tc++) {
2248 err = mlx5e_open_tis(priv, tc);
2250 goto err_close_tises;
2256 for (tc--; tc >= 0; tc--)
2257 mlx5e_close_tis(priv, tc);
2263 mlx5e_close_tises(struct mlx5e_priv *priv)
2265 int num_tc = priv->num_tc;
2268 for (tc = 0; tc < num_tc; tc++)
2269 mlx5e_close_tis(priv, tc);
2273 mlx5e_open_rqt(struct mlx5e_priv *priv)
2275 struct mlx5_core_dev *mdev = priv->mdev;
2277 u32 out[MLX5_ST_SZ_DW(create_rqt_out)] = {0};
2284 sz = 1 << priv->params.rx_hash_log_tbl_sz;
2286 inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
2287 in = mlx5_vzalloc(inlen);
2290 rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
2292 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
2293 MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
2295 for (i = 0; i < sz; i++) {
2298 ix = rss_get_indirection_to_bucket(ix);
2300 /* ensure we don't overflow */
2301 ix %= priv->params.num_channels;
2303 /* apply receive side scaling stride, if any */
2304 ix -= ix % (int)priv->params.channels_rsss;
2306 MLX5_SET(rqtc, rqtc, rq_num[i], priv->channel[ix].rq.rqn);
2309 MLX5_SET(create_rqt_in, in, opcode, MLX5_CMD_OP_CREATE_RQT);
2311 err = mlx5_cmd_exec(mdev, in, inlen, out, sizeof(out));
2313 priv->rqtn = MLX5_GET(create_rqt_out, out, rqtn);
2321 mlx5e_close_rqt(struct mlx5e_priv *priv)
2323 u32 in[MLX5_ST_SZ_DW(destroy_rqt_in)] = {0};
2324 u32 out[MLX5_ST_SZ_DW(destroy_rqt_out)] = {0};
2326 MLX5_SET(destroy_rqt_in, in, opcode, MLX5_CMD_OP_DESTROY_RQT);
2327 MLX5_SET(destroy_rqt_in, in, rqtn, priv->rqtn);
2329 mlx5_cmd_exec(priv->mdev, in, sizeof(in), out, sizeof(out));
2333 mlx5e_build_tir_ctx(struct mlx5e_priv *priv, u32 * tirc, int tt)
2335 void *hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
2338 MLX5_SET(tirc, tirc, transport_domain, priv->tdn);
2340 #define ROUGH_MAX_L2_L3_HDR_SZ 256
2342 #define MLX5_HASH_IP (MLX5_HASH_FIELD_SEL_SRC_IP |\
2343 MLX5_HASH_FIELD_SEL_DST_IP)
2345 #define MLX5_HASH_ALL (MLX5_HASH_FIELD_SEL_SRC_IP |\
2346 MLX5_HASH_FIELD_SEL_DST_IP |\
2347 MLX5_HASH_FIELD_SEL_L4_SPORT |\
2348 MLX5_HASH_FIELD_SEL_L4_DPORT)
2350 #define MLX5_HASH_IP_IPSEC_SPI (MLX5_HASH_FIELD_SEL_SRC_IP |\
2351 MLX5_HASH_FIELD_SEL_DST_IP |\
2352 MLX5_HASH_FIELD_SEL_IPSEC_SPI)
2354 if (priv->params.hw_lro_en) {
2355 MLX5_SET(tirc, tirc, lro_enable_mask,
2356 MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO |
2357 MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO);
2358 MLX5_SET(tirc, tirc, lro_max_msg_sz,
2359 (priv->params.lro_wqe_sz -
2360 ROUGH_MAX_L2_L3_HDR_SZ) >> 8);
2361 /* TODO: add the option to choose timer value dynamically */
2362 MLX5_SET(tirc, tirc, lro_timeout_period_usecs,
2363 MLX5_CAP_ETH(priv->mdev,
2364 lro_timer_supported_periods[2]));
2367 /* setup parameters for hashing TIR type, if any */
2370 MLX5_SET(tirc, tirc, disp_type,
2371 MLX5_TIRC_DISP_TYPE_DIRECT);
2372 MLX5_SET(tirc, tirc, inline_rqn,
2373 priv->channel[0].rq.rqn);
2376 MLX5_SET(tirc, tirc, disp_type,
2377 MLX5_TIRC_DISP_TYPE_INDIRECT);
2378 MLX5_SET(tirc, tirc, indirect_table,
2380 MLX5_SET(tirc, tirc, rx_hash_fn,
2381 MLX5_TIRC_RX_HASH_FN_HASH_TOEPLITZ);
2382 hkey = (__be32 *) MLX5_ADDR_OF(tirc, tirc, rx_hash_toeplitz_key);
2385 * The FreeBSD RSS implementation does currently not
2386 * support symmetric Toeplitz hashes:
2388 MLX5_SET(tirc, tirc, rx_hash_symmetric, 0);
2389 rss_getkey((uint8_t *)hkey);
2391 MLX5_SET(tirc, tirc, rx_hash_symmetric, 1);
2392 hkey[0] = cpu_to_be32(0xD181C62C);
2393 hkey[1] = cpu_to_be32(0xF7F4DB5B);
2394 hkey[2] = cpu_to_be32(0x1983A2FC);
2395 hkey[3] = cpu_to_be32(0x943E1ADB);
2396 hkey[4] = cpu_to_be32(0xD9389E6B);
2397 hkey[5] = cpu_to_be32(0xD1039C2C);
2398 hkey[6] = cpu_to_be32(0xA74499AD);
2399 hkey[7] = cpu_to_be32(0x593D56D9);
2400 hkey[8] = cpu_to_be32(0xF3253C06);
2401 hkey[9] = cpu_to_be32(0x2ADC1FFC);
2407 case MLX5E_TT_IPV4_TCP:
2408 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2409 MLX5_L3_PROT_TYPE_IPV4);
2410 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2411 MLX5_L4_PROT_TYPE_TCP);
2413 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4)) {
2414 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2418 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2422 case MLX5E_TT_IPV6_TCP:
2423 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2424 MLX5_L3_PROT_TYPE_IPV6);
2425 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2426 MLX5_L4_PROT_TYPE_TCP);
2428 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) {
2429 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2433 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2437 case MLX5E_TT_IPV4_UDP:
2438 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2439 MLX5_L3_PROT_TYPE_IPV4);
2440 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2441 MLX5_L4_PROT_TYPE_UDP);
2443 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4)) {
2444 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2448 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2452 case MLX5E_TT_IPV6_UDP:
2453 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2454 MLX5_L3_PROT_TYPE_IPV6);
2455 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2456 MLX5_L4_PROT_TYPE_UDP);
2458 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) {
2459 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2463 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2467 case MLX5E_TT_IPV4_IPSEC_AH:
2468 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2469 MLX5_L3_PROT_TYPE_IPV4);
2470 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2471 MLX5_HASH_IP_IPSEC_SPI);
2474 case MLX5E_TT_IPV6_IPSEC_AH:
2475 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2476 MLX5_L3_PROT_TYPE_IPV6);
2477 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2478 MLX5_HASH_IP_IPSEC_SPI);
2481 case MLX5E_TT_IPV4_IPSEC_ESP:
2482 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2483 MLX5_L3_PROT_TYPE_IPV4);
2484 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2485 MLX5_HASH_IP_IPSEC_SPI);
2488 case MLX5E_TT_IPV6_IPSEC_ESP:
2489 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2490 MLX5_L3_PROT_TYPE_IPV6);
2491 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2492 MLX5_HASH_IP_IPSEC_SPI);
2496 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2497 MLX5_L3_PROT_TYPE_IPV4);
2498 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2503 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2504 MLX5_L3_PROT_TYPE_IPV6);
2505 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2515 mlx5e_open_tir(struct mlx5e_priv *priv, int tt)
2517 struct mlx5_core_dev *mdev = priv->mdev;
2523 inlen = MLX5_ST_SZ_BYTES(create_tir_in);
2524 in = mlx5_vzalloc(inlen);
2527 tirc = MLX5_ADDR_OF(create_tir_in, in, tir_context);
2529 mlx5e_build_tir_ctx(priv, tirc, tt);
2531 err = mlx5_core_create_tir(mdev, in, inlen, &priv->tirn[tt]);
2539 mlx5e_close_tir(struct mlx5e_priv *priv, int tt)
2541 mlx5_core_destroy_tir(priv->mdev, priv->tirn[tt]);
2545 mlx5e_open_tirs(struct mlx5e_priv *priv)
2550 for (i = 0; i < MLX5E_NUM_TT; i++) {
2551 err = mlx5e_open_tir(priv, i);
2553 goto err_close_tirs;
2559 for (i--; i >= 0; i--)
2560 mlx5e_close_tir(priv, i);
2566 mlx5e_close_tirs(struct mlx5e_priv *priv)
2570 for (i = 0; i < MLX5E_NUM_TT; i++)
2571 mlx5e_close_tir(priv, i);
2575 * SW MTU does not include headers,
2576 * HW MTU includes all headers and checksums.
2579 mlx5e_set_dev_port_mtu(struct ifnet *ifp, int sw_mtu)
2581 struct mlx5e_priv *priv = ifp->if_softc;
2582 struct mlx5_core_dev *mdev = priv->mdev;
2586 hw_mtu = MLX5E_SW2HW_MTU(sw_mtu);
2588 err = mlx5_set_port_mtu(mdev, hw_mtu);
2590 if_printf(ifp, "%s: mlx5_set_port_mtu failed setting %d, err=%d\n",
2591 __func__, sw_mtu, err);
2595 /* Update vport context MTU */
2596 err = mlx5_set_vport_mtu(mdev, hw_mtu);
2598 if_printf(ifp, "%s: Failed updating vport context with MTU size, err=%d\n",
2602 ifp->if_mtu = sw_mtu;
2604 err = mlx5_query_vport_mtu(mdev, &hw_mtu);
2605 if (err || !hw_mtu) {
2606 /* fallback to port oper mtu */
2607 err = mlx5_query_port_oper_mtu(mdev, &hw_mtu);
2610 if_printf(ifp, "Query port MTU, after setting new "
2611 "MTU value, failed\n");
2613 } else if (MLX5E_HW2SW_MTU(hw_mtu) < sw_mtu) {
2615 if_printf(ifp, "Port MTU %d is smaller than "
2616 "ifp mtu %d\n", hw_mtu, sw_mtu);
2617 } else if (MLX5E_HW2SW_MTU(hw_mtu) > sw_mtu) {
2619 if_printf(ifp, "Port MTU %d is bigger than "
2620 "ifp mtu %d\n", hw_mtu, sw_mtu);
2622 priv->params_ethtool.hw_mtu = hw_mtu;
2628 mlx5e_open_locked(struct ifnet *ifp)
2630 struct mlx5e_priv *priv = ifp->if_softc;
2634 /* check if already opened */
2635 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
2639 if (rss_getnumbuckets() > priv->params.num_channels) {
2640 if_printf(ifp, "NOTE: There are more RSS buckets(%u) than "
2641 "channels(%u) available\n", rss_getnumbuckets(),
2642 priv->params.num_channels);
2645 err = mlx5e_open_tises(priv);
2647 if_printf(ifp, "%s: mlx5e_open_tises failed, %d\n",
2651 err = mlx5_vport_alloc_q_counter(priv->mdev,
2652 MLX5_INTERFACE_PROTOCOL_ETH, &set_id);
2654 if_printf(priv->ifp,
2655 "%s: mlx5_vport_alloc_q_counter failed: %d\n",
2657 goto err_close_tises;
2659 /* store counter set ID */
2660 priv->counter_set_id = set_id;
2662 err = mlx5e_open_channels(priv);
2664 if_printf(ifp, "%s: mlx5e_open_channels failed, %d\n",
2666 goto err_dalloc_q_counter;
2668 err = mlx5e_open_rqt(priv);
2670 if_printf(ifp, "%s: mlx5e_open_rqt failed, %d\n",
2672 goto err_close_channels;
2674 err = mlx5e_open_tirs(priv);
2676 if_printf(ifp, "%s: mlx5e_open_tir failed, %d\n",
2678 goto err_close_rqls;
2680 err = mlx5e_open_flow_table(priv);
2682 if_printf(ifp, "%s: mlx5e_open_flow_table failed, %d\n",
2684 goto err_close_tirs;
2686 err = mlx5e_add_all_vlan_rules(priv);
2688 if_printf(ifp, "%s: mlx5e_add_all_vlan_rules failed, %d\n",
2690 goto err_close_flow_table;
2692 set_bit(MLX5E_STATE_OPENED, &priv->state);
2694 mlx5e_update_carrier(priv);
2695 mlx5e_set_rx_mode_core(priv);
2699 err_close_flow_table:
2700 mlx5e_close_flow_table(priv);
2703 mlx5e_close_tirs(priv);
2706 mlx5e_close_rqt(priv);
2709 mlx5e_close_channels(priv);
2711 err_dalloc_q_counter:
2712 mlx5_vport_dealloc_q_counter(priv->mdev,
2713 MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
2716 mlx5e_close_tises(priv);
2722 mlx5e_open(void *arg)
2724 struct mlx5e_priv *priv = arg;
2727 if (mlx5_set_port_status(priv->mdev, MLX5_PORT_UP))
2728 if_printf(priv->ifp,
2729 "%s: Setting port status to up failed\n",
2732 mlx5e_open_locked(priv->ifp);
2733 priv->ifp->if_drv_flags |= IFF_DRV_RUNNING;
2738 mlx5e_close_locked(struct ifnet *ifp)
2740 struct mlx5e_priv *priv = ifp->if_softc;
2742 /* check if already closed */
2743 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2746 clear_bit(MLX5E_STATE_OPENED, &priv->state);
2748 mlx5e_set_rx_mode_core(priv);
2749 mlx5e_del_all_vlan_rules(priv);
2750 if_link_state_change(priv->ifp, LINK_STATE_DOWN);
2751 mlx5e_close_flow_table(priv);
2752 mlx5e_close_tirs(priv);
2753 mlx5e_close_rqt(priv);
2754 mlx5e_close_channels(priv);
2755 mlx5_vport_dealloc_q_counter(priv->mdev,
2756 MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
2757 mlx5e_close_tises(priv);
2762 #if (__FreeBSD_version >= 1100000)
2764 mlx5e_get_counter(struct ifnet *ifp, ift_counter cnt)
2766 struct mlx5e_priv *priv = ifp->if_softc;
2769 /* PRIV_LOCK(priv); XXX not allowed */
2771 case IFCOUNTER_IPACKETS:
2772 retval = priv->stats.vport.rx_packets;
2774 case IFCOUNTER_IERRORS:
2775 retval = priv->stats.vport.rx_error_packets +
2776 priv->stats.pport.alignment_err +
2777 priv->stats.pport.check_seq_err +
2778 priv->stats.pport.crc_align_errors +
2779 priv->stats.pport.in_range_len_errors +
2780 priv->stats.pport.jabbers +
2781 priv->stats.pport.out_of_range_len +
2782 priv->stats.pport.oversize_pkts +
2783 priv->stats.pport.symbol_err +
2784 priv->stats.pport.too_long_errors +
2785 priv->stats.pport.undersize_pkts +
2786 priv->stats.pport.unsupported_op_rx;
2788 case IFCOUNTER_IQDROPS:
2789 retval = priv->stats.vport.rx_out_of_buffer +
2790 priv->stats.pport.drop_events;
2792 case IFCOUNTER_OPACKETS:
2793 retval = priv->stats.vport.tx_packets;
2795 case IFCOUNTER_OERRORS:
2796 retval = priv->stats.vport.tx_error_packets;
2798 case IFCOUNTER_IBYTES:
2799 retval = priv->stats.vport.rx_bytes;
2801 case IFCOUNTER_OBYTES:
2802 retval = priv->stats.vport.tx_bytes;
2804 case IFCOUNTER_IMCASTS:
2805 retval = priv->stats.vport.rx_multicast_packets;
2807 case IFCOUNTER_OMCASTS:
2808 retval = priv->stats.vport.tx_multicast_packets;
2810 case IFCOUNTER_OQDROPS:
2811 retval = priv->stats.vport.tx_queue_dropped;
2813 case IFCOUNTER_COLLISIONS:
2814 retval = priv->stats.pport.collisions;
2817 retval = if_get_counter_default(ifp, cnt);
2820 /* PRIV_UNLOCK(priv); XXX not allowed */
2826 mlx5e_set_rx_mode(struct ifnet *ifp)
2828 struct mlx5e_priv *priv = ifp->if_softc;
2830 queue_work(priv->wq, &priv->set_rx_mode_work);
2834 mlx5e_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2836 struct mlx5e_priv *priv;
2838 struct ifi2creq i2c;
2847 priv = ifp->if_softc;
2849 /* check if detaching */
2850 if (priv == NULL || priv->gone != 0)
2855 ifr = (struct ifreq *)data;
2858 mlx5_query_port_max_mtu(priv->mdev, &max_mtu);
2860 if (ifr->ifr_mtu >= MLX5E_MTU_MIN &&
2861 ifr->ifr_mtu <= MIN(MLX5E_MTU_MAX, max_mtu)) {
2864 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2866 mlx5e_close_locked(ifp);
2869 mlx5e_set_dev_port_mtu(ifp, ifr->ifr_mtu);
2872 mlx5e_open_locked(ifp);
2875 if_printf(ifp, "Invalid MTU value. Min val: %d, Max val: %d\n",
2876 MLX5E_MTU_MIN, MIN(MLX5E_MTU_MAX, max_mtu));
2881 if ((ifp->if_flags & IFF_UP) &&
2882 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
2883 mlx5e_set_rx_mode(ifp);
2887 if (ifp->if_flags & IFF_UP) {
2888 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2889 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2890 mlx5e_open_locked(ifp);
2891 ifp->if_drv_flags |= IFF_DRV_RUNNING;
2892 mlx5_set_port_status(priv->mdev, MLX5_PORT_UP);
2895 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
2896 mlx5_set_port_status(priv->mdev,
2898 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
2899 mlx5e_close_locked(ifp);
2900 mlx5e_update_carrier(priv);
2901 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2908 mlx5e_set_rx_mode(ifp);
2913 ifr = (struct ifreq *)data;
2914 error = ifmedia_ioctl(ifp, ifr, &priv->media, command);
2917 ifr = (struct ifreq *)data;
2919 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2921 if (mask & IFCAP_TXCSUM) {
2922 ifp->if_capenable ^= IFCAP_TXCSUM;
2923 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
2925 if (IFCAP_TSO4 & ifp->if_capenable &&
2926 !(IFCAP_TXCSUM & ifp->if_capenable)) {
2927 ifp->if_capenable &= ~IFCAP_TSO4;
2928 ifp->if_hwassist &= ~CSUM_IP_TSO;
2930 "tso4 disabled due to -txcsum.\n");
2933 if (mask & IFCAP_TXCSUM_IPV6) {
2934 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
2935 ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
2937 if (IFCAP_TSO6 & ifp->if_capenable &&
2938 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2939 ifp->if_capenable &= ~IFCAP_TSO6;
2940 ifp->if_hwassist &= ~CSUM_IP6_TSO;
2942 "tso6 disabled due to -txcsum6.\n");
2945 if (mask & IFCAP_RXCSUM)
2946 ifp->if_capenable ^= IFCAP_RXCSUM;
2947 if (mask & IFCAP_RXCSUM_IPV6)
2948 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
2949 if (mask & IFCAP_TSO4) {
2950 if (!(IFCAP_TSO4 & ifp->if_capenable) &&
2951 !(IFCAP_TXCSUM & ifp->if_capenable)) {
2952 if_printf(ifp, "enable txcsum first.\n");
2956 ifp->if_capenable ^= IFCAP_TSO4;
2957 ifp->if_hwassist ^= CSUM_IP_TSO;
2959 if (mask & IFCAP_TSO6) {
2960 if (!(IFCAP_TSO6 & ifp->if_capenable) &&
2961 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2962 if_printf(ifp, "enable txcsum6 first.\n");
2966 ifp->if_capenable ^= IFCAP_TSO6;
2967 ifp->if_hwassist ^= CSUM_IP6_TSO;
2969 if (mask & IFCAP_VLAN_HWFILTER) {
2970 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2971 mlx5e_disable_vlan_filter(priv);
2973 mlx5e_enable_vlan_filter(priv);
2975 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
2977 if (mask & IFCAP_VLAN_HWTAGGING)
2978 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2979 if (mask & IFCAP_WOL_MAGIC)
2980 ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2982 VLAN_CAPABILITIES(ifp);
2983 /* turn off LRO means also turn of HW LRO - if it's on */
2984 if (mask & IFCAP_LRO) {
2985 int was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2986 bool need_restart = false;
2988 ifp->if_capenable ^= IFCAP_LRO;
2990 /* figure out if updating HW LRO is needed */
2991 if (!(ifp->if_capenable & IFCAP_LRO)) {
2992 if (priv->params.hw_lro_en) {
2993 priv->params.hw_lro_en = false;
2994 need_restart = true;
2997 if (priv->params.hw_lro_en == false &&
2998 priv->params_ethtool.hw_lro != 0) {
2999 priv->params.hw_lro_en = true;
3000 need_restart = true;
3003 if (was_opened && need_restart) {
3004 mlx5e_close_locked(ifp);
3005 mlx5e_open_locked(ifp);
3008 if (mask & IFCAP_HWRXTSTMP) {
3009 ifp->if_capenable ^= IFCAP_HWRXTSTMP;
3010 if (ifp->if_capenable & IFCAP_HWRXTSTMP) {
3011 if (priv->clbr_done == 0)
3012 mlx5e_reset_calibration_callout(priv);
3014 callout_drain(&priv->tstmp_clbr);
3015 priv->clbr_done = 0;
3023 ifr = (struct ifreq *)data;
3026 * Copy from the user-space address ifr_data to the
3027 * kernel-space address i2c
3029 error = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
3033 if (i2c.len > sizeof(i2c.data)) {
3039 /* Get module_num which is required for the query_eeprom */
3040 error = mlx5_query_module_num(priv->mdev, &module_num);
3042 if_printf(ifp, "Query module num failed, eeprom "
3043 "reading is not supported\n");
3047 /* Check if module is present before doing an access */
3048 module_status = mlx5_query_module_status(priv->mdev, module_num);
3049 if (module_status != MLX5_MODULE_STATUS_PLUGGED_ENABLED &&
3050 module_status != MLX5_MODULE_STATUS_PLUGGED_DISABLED) {
3055 * Currently 0XA0 and 0xA2 are the only addresses permitted.
3056 * The internal conversion is as follows:
3058 if (i2c.dev_addr == 0xA0)
3059 read_addr = MLX5E_I2C_ADDR_LOW;
3060 else if (i2c.dev_addr == 0xA2)
3061 read_addr = MLX5E_I2C_ADDR_HIGH;
3063 if_printf(ifp, "Query eeprom failed, "
3064 "Invalid Address: %X\n", i2c.dev_addr);
3068 error = mlx5_query_eeprom(priv->mdev,
3069 read_addr, MLX5E_EEPROM_LOW_PAGE,
3070 (uint32_t)i2c.offset, (uint32_t)i2c.len, module_num,
3071 (uint32_t *)i2c.data, &size_read);
3073 if_printf(ifp, "Query eeprom failed, eeprom "
3074 "reading is not supported\n");
3079 if (i2c.len > MLX5_EEPROM_MAX_BYTES) {
3080 error = mlx5_query_eeprom(priv->mdev,
3081 read_addr, MLX5E_EEPROM_LOW_PAGE,
3082 (uint32_t)(i2c.offset + size_read),
3083 (uint32_t)(i2c.len - size_read), module_num,
3084 (uint32_t *)(i2c.data + size_read), &size_read);
3087 if_printf(ifp, "Query eeprom failed, eeprom "
3088 "reading is not supported\n");
3093 error = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
3099 error = ether_ioctl(ifp, command, data);
3106 mlx5e_check_required_hca_cap(struct mlx5_core_dev *mdev)
3109 * TODO: uncoment once FW really sets all these bits if
3110 * (!mdev->caps.eth.rss_ind_tbl_cap || !mdev->caps.eth.csum_cap ||
3111 * !mdev->caps.eth.max_lso_cap || !mdev->caps.eth.vlan_cap ||
3112 * !(mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_SCQE_BRK_MOD)) return
3116 /* TODO: add more must-to-have features */
3118 if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
3125 mlx5e_get_max_inline_cap(struct mlx5_core_dev *mdev)
3127 uint32_t bf_buf_size = (1U << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2U;
3129 bf_buf_size -= sizeof(struct mlx5e_tx_wqe) - 2;
3131 /* verify against driver hardware limit */
3132 if (bf_buf_size > MLX5E_MAX_TX_INLINE)
3133 bf_buf_size = MLX5E_MAX_TX_INLINE;
3135 return (bf_buf_size);
3139 mlx5e_build_ifp_priv(struct mlx5_core_dev *mdev,
3140 struct mlx5e_priv *priv,
3141 int num_comp_vectors)
3146 * TODO: Consider link speed for setting "log_sq_size",
3147 * "log_rq_size" and "cq_moderation_xxx":
3149 priv->params.log_sq_size =
3150 MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
3151 priv->params.log_rq_size =
3152 MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
3153 priv->params.rx_cq_moderation_usec =
3154 MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
3155 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE :
3156 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
3157 priv->params.rx_cq_moderation_mode =
3158 MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ? 1 : 0;
3159 priv->params.rx_cq_moderation_pkts =
3160 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS;
3161 priv->params.tx_cq_moderation_usec =
3162 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC;
3163 priv->params.tx_cq_moderation_pkts =
3164 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
3165 priv->params.min_rx_wqes =
3166 MLX5E_PARAMS_DEFAULT_MIN_RX_WQES;
3167 priv->params.rx_hash_log_tbl_sz =
3168 (order_base_2(num_comp_vectors) >
3169 MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ) ?
3170 order_base_2(num_comp_vectors) :
3171 MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ;
3172 priv->params.num_tc = 1;
3173 priv->params.default_vlan_prio = 0;
3174 priv->counter_set_id = -1;
3175 priv->params.tx_max_inline = mlx5e_get_max_inline_cap(mdev);
3177 err = mlx5_query_min_inline(mdev, &priv->params.tx_min_inline_mode);
3182 * hw lro is currently defaulted to off. when it won't anymore we
3183 * will consider the HW capability: "!!MLX5_CAP_ETH(mdev, lro_cap)"
3185 priv->params.hw_lro_en = false;
3186 priv->params.lro_wqe_sz = MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ;
3188 priv->params.cqe_zipping_en = !!MLX5_CAP_GEN(mdev, cqe_compression);
3191 priv->params.num_channels = num_comp_vectors;
3192 priv->params.channels_rsss = 1;
3193 priv->order_base_2_num_channels = order_base_2(num_comp_vectors);
3194 priv->queue_mapping_channel_mask =
3195 roundup_pow_of_two(num_comp_vectors) - 1;
3196 priv->num_tc = priv->params.num_tc;
3197 priv->default_vlan_prio = priv->params.default_vlan_prio;
3199 INIT_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
3200 INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
3201 INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
3207 mlx5e_create_mkey(struct mlx5e_priv *priv, u32 pdn,
3208 struct mlx5_core_mr *mkey)
3210 struct ifnet *ifp = priv->ifp;
3211 struct mlx5_core_dev *mdev = priv->mdev;
3212 int inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
3217 in = mlx5_vzalloc(inlen);
3219 if_printf(ifp, "%s: failed to allocate inbox\n", __func__);
3223 mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
3224 MLX5_SET(mkc, mkc, access_mode, MLX5_ACCESS_MODE_PA);
3225 MLX5_SET(mkc, mkc, lw, 1);
3226 MLX5_SET(mkc, mkc, lr, 1);
3228 MLX5_SET(mkc, mkc, pd, pdn);
3229 MLX5_SET(mkc, mkc, length64, 1);
3230 MLX5_SET(mkc, mkc, qpn, 0xffffff);
3232 err = mlx5_core_create_mkey(mdev, mkey, in, inlen);
3234 if_printf(ifp, "%s: mlx5_core_create_mkey failed, %d\n",
3241 static const char *mlx5e_vport_stats_desc[] = {
3242 MLX5E_VPORT_STATS(MLX5E_STATS_DESC)
3245 static const char *mlx5e_pport_stats_desc[] = {
3246 MLX5E_PPORT_STATS(MLX5E_STATS_DESC)
3250 mlx5e_priv_mtx_init(struct mlx5e_priv *priv)
3252 mtx_init(&priv->async_events_mtx, "mlx5async", MTX_NETWORK_LOCK, MTX_DEF);
3253 sx_init(&priv->state_lock, "mlx5state");
3254 callout_init_mtx(&priv->watchdog, &priv->async_events_mtx, 0);
3255 MLX5_INIT_DOORBELL_LOCK(&priv->doorbell_lock);
3259 mlx5e_priv_mtx_destroy(struct mlx5e_priv *priv)
3261 mtx_destroy(&priv->async_events_mtx);
3262 sx_destroy(&priv->state_lock);
3266 sysctl_firmware(SYSCTL_HANDLER_ARGS)
3269 * %d.%d%.d the string format.
3270 * fw_rev_{maj,min,sub} return u16, 2^16 = 65536.
3271 * We need at most 5 chars to store that.
3272 * It also has: two "." and NULL at the end, which means we need 18
3273 * (5*3 + 3) chars at most.
3276 struct mlx5e_priv *priv = arg1;
3279 snprintf(fw, sizeof(fw), "%d.%d.%d", fw_rev_maj(priv->mdev), fw_rev_min(priv->mdev),
3280 fw_rev_sub(priv->mdev));
3281 error = sysctl_handle_string(oidp, fw, sizeof(fw), req);
3286 mlx5e_disable_tx_dma(struct mlx5e_channel *ch)
3290 for (i = 0; i < ch->num_tc; i++)
3291 mlx5e_drain_sq(&ch->sq[i]);
3295 mlx5e_reset_sq_doorbell_record(struct mlx5e_sq *sq)
3298 sq->doorbell.d32[0] = cpu_to_be32(MLX5_OPCODE_NOP);
3299 sq->doorbell.d32[1] = cpu_to_be32(sq->sqn << 8);
3300 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
3301 sq->doorbell.d64 = 0;
3305 mlx5e_resume_sq(struct mlx5e_sq *sq)
3309 /* check if already enabled */
3310 if (READ_ONCE(sq->running) != 0)
3313 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_ERR,
3314 MLX5_SQC_STATE_RST);
3317 "mlx5e_modify_sq() from ERR to RST failed: %d\n", err);
3323 /* reset doorbell prior to moving from RST to RDY */
3324 mlx5e_reset_sq_doorbell_record(sq);
3326 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST,
3327 MLX5_SQC_STATE_RDY);
3330 "mlx5e_modify_sq() from RST to RDY failed: %d\n", err);
3333 sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
3334 WRITE_ONCE(sq->running, 1);
3338 mlx5e_enable_tx_dma(struct mlx5e_channel *ch)
3342 for (i = 0; i < ch->num_tc; i++)
3343 mlx5e_resume_sq(&ch->sq[i]);
3347 mlx5e_disable_rx_dma(struct mlx5e_channel *ch)
3349 struct mlx5e_rq *rq = &ch->rq;
3354 callout_stop(&rq->watchdog);
3355 mtx_unlock(&rq->mtx);
3357 callout_drain(&rq->watchdog);
3359 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
3362 "mlx5e_modify_rq() from RDY to RST failed: %d\n", err);
3365 while (!mlx5_wq_ll_is_empty(&rq->wq)) {
3367 rq->cq.mcq.comp(&rq->cq.mcq);
3371 * Transitioning into RST state will allow the FW to track less ERR state queues,
3372 * thus reducing the recv queue flushing time
3374 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_ERR, MLX5_RQC_STATE_RST);
3377 "mlx5e_modify_rq() from ERR to RST failed: %d\n", err);
3382 mlx5e_enable_rx_dma(struct mlx5e_channel *ch)
3384 struct mlx5e_rq *rq = &ch->rq;
3388 mlx5_wq_ll_update_db_record(&rq->wq);
3389 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
3392 "mlx5e_modify_rq() from RST to RDY failed: %d\n", err);
3397 rq->cq.mcq.comp(&rq->cq.mcq);
3401 mlx5e_modify_tx_dma(struct mlx5e_priv *priv, uint8_t value)
3405 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3408 for (i = 0; i < priv->params.num_channels; i++) {
3410 mlx5e_disable_tx_dma(&priv->channel[i]);
3412 mlx5e_enable_tx_dma(&priv->channel[i]);
3417 mlx5e_modify_rx_dma(struct mlx5e_priv *priv, uint8_t value)
3421 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3424 for (i = 0; i < priv->params.num_channels; i++) {
3426 mlx5e_disable_rx_dma(&priv->channel[i]);
3428 mlx5e_enable_rx_dma(&priv->channel[i]);
3433 mlx5e_add_hw_stats(struct mlx5e_priv *priv)
3435 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3436 OID_AUTO, "fw_version", CTLTYPE_STRING | CTLFLAG_RD, priv, 0,
3437 sysctl_firmware, "A", "HCA firmware version");
3439 SYSCTL_ADD_STRING(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3440 OID_AUTO, "board_id", CTLFLAG_RD, priv->mdev->board_id, 0,
3445 mlx5e_sysctl_tx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3447 struct mlx5e_priv *priv = arg1;
3448 uint8_t temp[MLX5E_MAX_PRIORITY];
3455 tx_pfc = priv->params.tx_priority_flow_control;
3457 for (i = 0; i != MLX5E_MAX_PRIORITY; i++)
3458 temp[i] = (tx_pfc >> i) & 1;
3460 err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY);
3461 if (err || !req->newptr)
3463 err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY);
3467 priv->params.tx_priority_flow_control = 0;
3469 /* range check input value */
3470 for (i = 0; i != MLX5E_MAX_PRIORITY; i++) {
3475 priv->params.tx_priority_flow_control |= (temp[i] << i);
3478 /* check if update is required */
3479 if (tx_pfc != priv->params.tx_priority_flow_control)
3480 err = -mlx5e_set_port_pfc(priv);
3483 priv->params.tx_priority_flow_control= tx_pfc;
3490 mlx5e_sysctl_rx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3492 struct mlx5e_priv *priv = arg1;
3493 uint8_t temp[MLX5E_MAX_PRIORITY];
3500 rx_pfc = priv->params.rx_priority_flow_control;
3502 for (i = 0; i != MLX5E_MAX_PRIORITY; i++)
3503 temp[i] = (rx_pfc >> i) & 1;
3505 err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY);
3506 if (err || !req->newptr)
3508 err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY);
3512 priv->params.rx_priority_flow_control = 0;
3514 /* range check input value */
3515 for (i = 0; i != MLX5E_MAX_PRIORITY; i++) {
3520 priv->params.rx_priority_flow_control |= (temp[i] << i);
3523 /* check if update is required */
3524 if (rx_pfc != priv->params.rx_priority_flow_control)
3525 err = -mlx5e_set_port_pfc(priv);
3528 priv->params.rx_priority_flow_control= rx_pfc;
3535 mlx5e_setup_pauseframes(struct mlx5e_priv *priv)
3537 #if (__FreeBSD_version < 1100000)
3542 /* enable pauseframes by default */
3543 priv->params.tx_pauseframe_control = 1;
3544 priv->params.rx_pauseframe_control = 1;
3546 /* disable ports flow control, PFC, by default */
3547 priv->params.tx_priority_flow_control = 0;
3548 priv->params.rx_priority_flow_control = 0;
3550 #if (__FreeBSD_version < 1100000)
3551 /* compute path for sysctl */
3552 snprintf(path, sizeof(path), "dev.mce.%d.tx_pauseframe_control",
3553 device_get_unit(priv->mdev->pdev->dev.bsddev));
3555 /* try to fetch tunable, if any */
3556 TUNABLE_INT_FETCH(path, &priv->params.tx_pauseframe_control);
3558 /* compute path for sysctl */
3559 snprintf(path, sizeof(path), "dev.mce.%d.rx_pauseframe_control",
3560 device_get_unit(priv->mdev->pdev->dev.bsddev));
3562 /* try to fetch tunable, if any */
3563 TUNABLE_INT_FETCH(path, &priv->params.rx_pauseframe_control);
3566 /* register pauseframe SYSCTLs */
3567 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3568 OID_AUTO, "tx_pauseframe_control", CTLFLAG_RDTUN,
3569 &priv->params.tx_pauseframe_control, 0,
3570 "Set to enable TX pause frames. Clear to disable.");
3572 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3573 OID_AUTO, "rx_pauseframe_control", CTLFLAG_RDTUN,
3574 &priv->params.rx_pauseframe_control, 0,
3575 "Set to enable RX pause frames. Clear to disable.");
3577 /* register priority flow control, PFC, SYSCTLs */
3578 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3579 OID_AUTO, "tx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN |
3580 CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_tx_priority_flow_control, "CU",
3581 "Set to enable TX ports flow control frames for priorities 0..7. Clear to disable.");
3583 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3584 OID_AUTO, "rx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN |
3585 CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_rx_priority_flow_control, "CU",
3586 "Set to enable RX ports flow control frames for priorities 0..7. Clear to disable.");
3591 priv->params.tx_pauseframe_control =
3592 priv->params.tx_pauseframe_control ? 1 : 0;
3593 priv->params.rx_pauseframe_control =
3594 priv->params.rx_pauseframe_control ? 1 : 0;
3596 /* update firmware */
3597 error = mlx5e_set_port_pause_and_pfc(priv);
3598 if (error == -EINVAL) {
3599 if_printf(priv->ifp,
3600 "Global pauseframes must be disabled before enabling PFC.\n");
3601 priv->params.rx_priority_flow_control = 0;
3602 priv->params.tx_priority_flow_control = 0;
3604 /* update firmware */
3605 (void) mlx5e_set_port_pause_and_pfc(priv);
3611 mlx5e_ul_snd_tag_alloc(struct ifnet *ifp,
3612 union if_snd_tag_alloc_params *params,
3613 struct m_snd_tag **ppmt)
3615 struct mlx5e_priv *priv;
3616 struct mlx5e_channel *pch;
3618 priv = ifp->if_softc;
3620 if (unlikely(priv->gone || params->hdr.flowtype == M_HASHTYPE_NONE)) {
3621 return (EOPNOTSUPP);
3623 /* keep this code synced with mlx5e_select_queue() */
3624 u32 ch = priv->params.num_channels;
3628 if (rss_hash2bucket(params->hdr.flowid,
3629 params->hdr.flowtype, &temp) == 0)
3633 ch = (params->hdr.flowid % 128) % ch;
3636 * NOTE: The channels array is only freed at detach
3637 * and it safe to return a pointer to the send tag
3638 * inside the channels structure as long as we
3639 * reference the priv.
3641 pch = priv->channel + ch;
3643 /* check if send queue is not running */
3644 if (unlikely(pch->sq[0].running == 0))
3646 mlx5e_ref_channel(priv);
3647 *ppmt = &pch->tag.m_snd_tag;
3653 mlx5e_ul_snd_tag_query(struct m_snd_tag *pmt, union if_snd_tag_query_params *params)
3655 struct mlx5e_channel *pch =
3656 container_of(pmt, struct mlx5e_channel, tag.m_snd_tag);
3658 params->unlimited.max_rate = -1ULL;
3659 params->unlimited.queue_level = mlx5e_sq_queue_level(&pch->sq[0]);
3664 mlx5e_ul_snd_tag_free(struct m_snd_tag *pmt)
3666 struct mlx5e_channel *pch =
3667 container_of(pmt, struct mlx5e_channel, tag.m_snd_tag);
3669 mlx5e_unref_channel(pch->priv);
3673 mlx5e_snd_tag_alloc(struct ifnet *ifp,
3674 union if_snd_tag_alloc_params *params,
3675 struct m_snd_tag **ppmt)
3678 switch (params->hdr.type) {
3680 case IF_SND_TAG_TYPE_RATE_LIMIT:
3681 return (mlx5e_rl_snd_tag_alloc(ifp, params, ppmt));
3683 case IF_SND_TAG_TYPE_UNLIMITED:
3684 return (mlx5e_ul_snd_tag_alloc(ifp, params, ppmt));
3686 return (EOPNOTSUPP);
3691 mlx5e_snd_tag_modify(struct m_snd_tag *pmt, union if_snd_tag_modify_params *params)
3693 struct mlx5e_snd_tag *tag =
3694 container_of(pmt, struct mlx5e_snd_tag, m_snd_tag);
3696 switch (tag->type) {
3698 case IF_SND_TAG_TYPE_RATE_LIMIT:
3699 return (mlx5e_rl_snd_tag_modify(pmt, params));
3701 case IF_SND_TAG_TYPE_UNLIMITED:
3703 return (EOPNOTSUPP);
3708 mlx5e_snd_tag_query(struct m_snd_tag *pmt, union if_snd_tag_query_params *params)
3710 struct mlx5e_snd_tag *tag =
3711 container_of(pmt, struct mlx5e_snd_tag, m_snd_tag);
3713 switch (tag->type) {
3715 case IF_SND_TAG_TYPE_RATE_LIMIT:
3716 return (mlx5e_rl_snd_tag_query(pmt, params));
3718 case IF_SND_TAG_TYPE_UNLIMITED:
3719 return (mlx5e_ul_snd_tag_query(pmt, params));
3721 return (EOPNOTSUPP);
3726 mlx5e_snd_tag_free(struct m_snd_tag *pmt)
3728 struct mlx5e_snd_tag *tag =
3729 container_of(pmt, struct mlx5e_snd_tag, m_snd_tag);
3731 switch (tag->type) {
3733 case IF_SND_TAG_TYPE_RATE_LIMIT:
3734 mlx5e_rl_snd_tag_free(pmt);
3737 case IF_SND_TAG_TYPE_UNLIMITED:
3738 mlx5e_ul_snd_tag_free(pmt);
3746 mlx5e_create_ifp(struct mlx5_core_dev *mdev)
3749 struct mlx5e_priv *priv;
3750 u8 dev_addr[ETHER_ADDR_LEN] __aligned(4);
3751 struct sysctl_oid_list *child;
3752 int ncv = mdev->priv.eq_table.num_comp_vectors;
3758 if (mlx5e_check_required_hca_cap(mdev)) {
3759 mlx5_core_dbg(mdev, "mlx5e_check_required_hca_cap() failed\n");
3763 * Try to allocate the priv and make room for worst-case
3764 * number of channel structures:
3766 priv = malloc(sizeof(*priv) +
3767 (sizeof(priv->channel[0]) * mdev->priv.eq_table.num_comp_vectors),
3768 M_MLX5EN, M_WAITOK | M_ZERO);
3769 mlx5e_priv_mtx_init(priv);
3771 ifp = priv->ifp = if_alloc(IFT_ETHER);
3773 mlx5_core_err(mdev, "if_alloc() failed\n");
3776 ifp->if_softc = priv;
3777 if_initname(ifp, "mce", device_get_unit(mdev->pdev->dev.bsddev));
3778 ifp->if_mtu = ETHERMTU;
3779 ifp->if_init = mlx5e_open;
3780 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
3781 ifp->if_ioctl = mlx5e_ioctl;
3782 ifp->if_transmit = mlx5e_xmit;
3783 ifp->if_qflush = if_qflush;
3784 #if (__FreeBSD_version >= 1100000)
3785 ifp->if_get_counter = mlx5e_get_counter;
3787 ifp->if_snd.ifq_maxlen = ifqmaxlen;
3789 * Set driver features
3791 ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6;
3792 ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
3793 ifp->if_capabilities |= IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWFILTER;
3794 ifp->if_capabilities |= IFCAP_LINKSTATE | IFCAP_JUMBO_MTU;
3795 ifp->if_capabilities |= IFCAP_LRO;
3796 ifp->if_capabilities |= IFCAP_TSO | IFCAP_VLAN_HWTSO;
3797 ifp->if_capabilities |= IFCAP_HWSTATS | IFCAP_HWRXTSTMP;
3798 ifp->if_capabilities |= IFCAP_TXRTLMT;
3799 ifp->if_snd_tag_alloc = mlx5e_snd_tag_alloc;
3800 ifp->if_snd_tag_free = mlx5e_snd_tag_free;
3801 ifp->if_snd_tag_modify = mlx5e_snd_tag_modify;
3802 ifp->if_snd_tag_query = mlx5e_snd_tag_query;
3804 /* set TSO limits so that we don't have to drop TX packets */
3805 ifp->if_hw_tsomax = MLX5E_MAX_TX_PAYLOAD_SIZE - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
3806 ifp->if_hw_tsomaxsegcount = MLX5E_MAX_TX_MBUF_FRAGS - 1 /* hdr */;
3807 ifp->if_hw_tsomaxsegsize = MLX5E_MAX_TX_MBUF_SIZE;
3809 ifp->if_capenable = ifp->if_capabilities;
3810 ifp->if_hwassist = 0;
3811 if (ifp->if_capenable & IFCAP_TSO)
3812 ifp->if_hwassist |= CSUM_TSO;
3813 if (ifp->if_capenable & IFCAP_TXCSUM)
3814 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP | CSUM_IP);
3815 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3816 ifp->if_hwassist |= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
3818 /* ifnet sysctl tree */
3819 sysctl_ctx_init(&priv->sysctl_ctx);
3820 priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_dev),
3821 OID_AUTO, ifp->if_dname, CTLFLAG_RD, 0, "MLX5 ethernet - interface name");
3822 if (priv->sysctl_ifnet == NULL) {
3823 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3824 goto err_free_sysctl;
3826 snprintf(unit, sizeof(unit), "%d", ifp->if_dunit);
3827 priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3828 OID_AUTO, unit, CTLFLAG_RD, 0, "MLX5 ethernet - interface unit");
3829 if (priv->sysctl_ifnet == NULL) {
3830 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3831 goto err_free_sysctl;
3834 /* HW sysctl tree */
3835 child = SYSCTL_CHILDREN(device_get_sysctl_tree(mdev->pdev->dev.bsddev));
3836 priv->sysctl_hw = SYSCTL_ADD_NODE(&priv->sysctl_ctx, child,
3837 OID_AUTO, "hw", CTLFLAG_RD, 0, "MLX5 ethernet dev hw");
3838 if (priv->sysctl_hw == NULL) {
3839 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3840 goto err_free_sysctl;
3843 err = mlx5e_build_ifp_priv(mdev, priv, ncv);
3845 mlx5_core_err(mdev, "mlx5e_build_ifp_priv() failed (%d)\n", err);
3846 goto err_free_sysctl;
3849 snprintf(unit, sizeof(unit), "mce%u_wq",
3850 device_get_unit(mdev->pdev->dev.bsddev));
3851 priv->wq = alloc_workqueue(unit, 0, 1);
3852 if (priv->wq == NULL) {
3853 if_printf(ifp, "%s: alloc_workqueue failed\n", __func__);
3854 goto err_free_sysctl;
3857 err = mlx5_alloc_map_uar(mdev, &priv->cq_uar);
3859 if_printf(ifp, "%s: mlx5_alloc_map_uar failed, %d\n",
3863 err = mlx5_core_alloc_pd(mdev, &priv->pdn);
3865 if_printf(ifp, "%s: mlx5_core_alloc_pd failed, %d\n",
3867 goto err_unmap_free_uar;
3869 err = mlx5_alloc_transport_domain(mdev, &priv->tdn);
3871 if_printf(ifp, "%s: mlx5_alloc_transport_domain failed, %d\n",
3873 goto err_dealloc_pd;
3875 err = mlx5e_create_mkey(priv, priv->pdn, &priv->mr);
3877 if_printf(ifp, "%s: mlx5e_create_mkey failed, %d\n",
3879 goto err_dealloc_transport_domain;
3881 mlx5_query_nic_vport_mac_address(priv->mdev, 0, dev_addr);
3883 /* check if we should generate a random MAC address */
3884 if (MLX5_CAP_GEN(priv->mdev, vport_group_manager) == 0 &&
3885 is_zero_ether_addr(dev_addr)) {
3886 random_ether_addr(dev_addr);
3887 if_printf(ifp, "Assigned random MAC address\n");
3890 err = mlx5e_rl_init(priv);
3892 if_printf(ifp, "%s: mlx5e_rl_init failed, %d\n",
3894 goto err_create_mkey;
3898 /* set default MTU */
3899 mlx5e_set_dev_port_mtu(ifp, ifp->if_mtu);
3901 /* Set default media status */
3902 priv->media_status_last = IFM_AVALID;
3903 priv->media_active_last = IFM_ETHER | IFM_AUTO |
3904 IFM_ETH_RXPAUSE | IFM_FDX;
3906 /* setup default pauseframes configuration */
3907 mlx5e_setup_pauseframes(priv);
3909 err = mlx5_query_port_proto_cap(mdev, ð_proto_cap, MLX5_PTYS_EN);
3912 if_printf(ifp, "%s: Query port media capability failed, %d\n",
3916 /* Setup supported medias */
3917 ifmedia_init(&priv->media, IFM_IMASK | IFM_ETH_FMASK,
3918 mlx5e_media_change, mlx5e_media_status);
3920 for (i = 0; i < MLX5E_LINK_MODES_NUMBER; ++i) {
3921 if (mlx5e_mode_table[i].baudrate == 0)
3923 if (MLX5E_PROT_MASK(i) & eth_proto_cap) {
3924 ifmedia_add(&priv->media,
3925 mlx5e_mode_table[i].subtype |
3926 IFM_ETHER, 0, NULL);
3927 ifmedia_add(&priv->media,
3928 mlx5e_mode_table[i].subtype |
3929 IFM_ETHER | IFM_FDX |
3930 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3934 /* Additional supported medias */
3935 ifmedia_add(&priv->media, IFM_10G_LR | IFM_ETHER, 0, NULL);
3936 ifmedia_add(&priv->media, IFM_10G_LR |
3937 IFM_ETHER | IFM_FDX |
3938 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3940 ifmedia_add(&priv->media, IFM_40G_ER4 | IFM_ETHER, 0, NULL);
3941 ifmedia_add(&priv->media, IFM_40G_ER4 |
3942 IFM_ETHER | IFM_FDX |
3943 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3945 ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO, 0, NULL);
3946 ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
3947 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3949 /* Set autoselect by default */
3950 ifmedia_set(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
3951 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE);
3952 ether_ifattach(ifp, dev_addr);
3954 /* Register for VLAN events */
3955 priv->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
3956 mlx5e_vlan_rx_add_vid, priv, EVENTHANDLER_PRI_FIRST);
3957 priv->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
3958 mlx5e_vlan_rx_kill_vid, priv, EVENTHANDLER_PRI_FIRST);
3960 /* Link is down by default */
3961 if_link_state_change(ifp, LINK_STATE_DOWN);
3963 mlx5e_enable_async_events(priv);
3965 mlx5e_add_hw_stats(priv);
3967 mlx5e_create_stats(&priv->stats.vport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3968 "vstats", mlx5e_vport_stats_desc, MLX5E_VPORT_STATS_NUM,
3969 priv->stats.vport.arg);
3971 mlx5e_create_stats(&priv->stats.pport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3972 "pstats", mlx5e_pport_stats_desc, MLX5E_PPORT_STATS_NUM,
3973 priv->stats.pport.arg);
3975 mlx5e_create_ethtool(priv);
3977 mtx_lock(&priv->async_events_mtx);
3978 mlx5e_update_stats(priv);
3979 mtx_unlock(&priv->async_events_mtx);
3981 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3982 OID_AUTO, "rx_clbr_done", CTLFLAG_RD,
3983 &priv->clbr_done, 0,
3984 "RX timestamps calibration state");
3985 callout_init(&priv->tstmp_clbr, CALLOUT_DIRECT);
3986 mlx5e_reset_calibration_callout(priv);
3992 mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
3994 err_dealloc_transport_domain:
3995 mlx5_dealloc_transport_domain(mdev, priv->tdn);
3998 mlx5_core_dealloc_pd(mdev, priv->pdn);
4001 mlx5_unmap_free_uar(mdev, &priv->cq_uar);
4004 destroy_workqueue(priv->wq);
4007 sysctl_ctx_free(&priv->sysctl_ctx);
4008 if (priv->sysctl_debug)
4009 sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
4013 mlx5e_priv_mtx_destroy(priv);
4014 free(priv, M_MLX5EN);
4019 mlx5e_destroy_ifp(struct mlx5_core_dev *mdev, void *vpriv)
4021 struct mlx5e_priv *priv = vpriv;
4022 struct ifnet *ifp = priv->ifp;
4024 /* don't allow more IOCTLs */
4027 /* XXX wait a bit to allow IOCTL handlers to complete */
4032 * The kernel can have reference(s) via the m_snd_tag's into
4033 * the ratelimit channels, and these must go away before
4036 while (READ_ONCE(priv->rl.stats.tx_active_connections) != 0) {
4037 if_printf(priv->ifp, "Waiting for all ratelimit connections "
4042 /* stop watchdog timer */
4043 callout_drain(&priv->watchdog);
4045 callout_drain(&priv->tstmp_clbr);
4047 if (priv->vlan_attach != NULL)
4048 EVENTHANDLER_DEREGISTER(vlan_config, priv->vlan_attach);
4049 if (priv->vlan_detach != NULL)
4050 EVENTHANDLER_DEREGISTER(vlan_unconfig, priv->vlan_detach);
4052 /* make sure device gets closed */
4054 mlx5e_close_locked(ifp);
4057 /* wait for all unlimited send tags to go away */
4058 while (priv->channel_refs != 0) {
4059 if_printf(priv->ifp, "Waiting for all unlimited connections "
4064 /* unregister device */
4065 ifmedia_removeall(&priv->media);
4066 ether_ifdetach(ifp);
4070 mlx5e_rl_cleanup(priv);
4072 /* destroy all remaining sysctl nodes */
4073 sysctl_ctx_free(&priv->stats.vport.ctx);
4074 sysctl_ctx_free(&priv->stats.pport.ctx);
4075 if (priv->sysctl_debug)
4076 sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
4077 sysctl_ctx_free(&priv->sysctl_ctx);
4079 mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
4080 mlx5_dealloc_transport_domain(priv->mdev, priv->tdn);
4081 mlx5_core_dealloc_pd(priv->mdev, priv->pdn);
4082 mlx5_unmap_free_uar(priv->mdev, &priv->cq_uar);
4083 mlx5e_disable_async_events(priv);
4084 destroy_workqueue(priv->wq);
4085 mlx5e_priv_mtx_destroy(priv);
4086 free(priv, M_MLX5EN);
4090 mlx5e_get_ifp(void *vpriv)
4092 struct mlx5e_priv *priv = vpriv;
4097 static struct mlx5_interface mlx5e_interface = {
4098 .add = mlx5e_create_ifp,
4099 .remove = mlx5e_destroy_ifp,
4100 .event = mlx5e_async_event,
4101 .protocol = MLX5_INTERFACE_PROTOCOL_ETH,
4102 .get_dev = mlx5e_get_ifp,
4108 mlx5_register_interface(&mlx5e_interface);
4114 mlx5_unregister_interface(&mlx5e_interface);
4118 mlx5e_show_version(void __unused *arg)
4121 printf("%s", mlx5e_version);
4123 SYSINIT(mlx5e_show_version, SI_SUB_DRIVERS, SI_ORDER_ANY, mlx5e_show_version, NULL);
4125 module_init_order(mlx5e_init, SI_ORDER_THIRD);
4126 module_exit_order(mlx5e_cleanup, SI_ORDER_THIRD);
4128 #if (__FreeBSD_version >= 1100000)
4129 MODULE_DEPEND(mlx5en, linuxkpi, 1, 1, 1);
4131 MODULE_DEPEND(mlx5en, mlx5, 1, 1, 1);
4132 MODULE_VERSION(mlx5en, 1);