2 * Copyright (c) 2015-2018 Mellanox Technologies. All rights reserved.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
13 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/sockio.h>
31 #include <machine/atomic.h>
33 #ifndef ETH_DRIVER_VERSION
34 #define ETH_DRIVER_VERSION "3.5.0"
36 #define DRIVER_RELDATE "November 2018"
38 static const char mlx5e_version[] = "mlx5en: Mellanox Ethernet driver "
39 ETH_DRIVER_VERSION " (" DRIVER_RELDATE ")\n";
41 static int mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs);
43 struct mlx5e_channel_param {
44 struct mlx5e_rq_param rq;
45 struct mlx5e_sq_param sq;
46 struct mlx5e_cq_param rx_cq;
47 struct mlx5e_cq_param tx_cq;
53 } mlx5e_mode_table[MLX5E_LINK_MODES_NUMBER] = {
55 [MLX5E_1000BASE_CX_SGMII] = {
56 .subtype = IFM_1000_CX_SGMII,
57 .baudrate = IF_Mbps(1000ULL),
59 [MLX5E_1000BASE_KX] = {
60 .subtype = IFM_1000_KX,
61 .baudrate = IF_Mbps(1000ULL),
63 [MLX5E_10GBASE_CX4] = {
64 .subtype = IFM_10G_CX4,
65 .baudrate = IF_Gbps(10ULL),
67 [MLX5E_10GBASE_KX4] = {
68 .subtype = IFM_10G_KX4,
69 .baudrate = IF_Gbps(10ULL),
71 [MLX5E_10GBASE_KR] = {
72 .subtype = IFM_10G_KR,
73 .baudrate = IF_Gbps(10ULL),
75 [MLX5E_20GBASE_KR2] = {
76 .subtype = IFM_20G_KR2,
77 .baudrate = IF_Gbps(20ULL),
79 [MLX5E_40GBASE_CR4] = {
80 .subtype = IFM_40G_CR4,
81 .baudrate = IF_Gbps(40ULL),
83 [MLX5E_40GBASE_KR4] = {
84 .subtype = IFM_40G_KR4,
85 .baudrate = IF_Gbps(40ULL),
87 [MLX5E_56GBASE_R4] = {
88 .subtype = IFM_56G_R4,
89 .baudrate = IF_Gbps(56ULL),
91 [MLX5E_10GBASE_CR] = {
92 .subtype = IFM_10G_CR1,
93 .baudrate = IF_Gbps(10ULL),
95 [MLX5E_10GBASE_SR] = {
96 .subtype = IFM_10G_SR,
97 .baudrate = IF_Gbps(10ULL),
99 [MLX5E_10GBASE_ER] = {
100 .subtype = IFM_10G_ER,
101 .baudrate = IF_Gbps(10ULL),
103 [MLX5E_40GBASE_SR4] = {
104 .subtype = IFM_40G_SR4,
105 .baudrate = IF_Gbps(40ULL),
107 [MLX5E_40GBASE_LR4] = {
108 .subtype = IFM_40G_LR4,
109 .baudrate = IF_Gbps(40ULL),
111 [MLX5E_100GBASE_CR4] = {
112 .subtype = IFM_100G_CR4,
113 .baudrate = IF_Gbps(100ULL),
115 [MLX5E_100GBASE_SR4] = {
116 .subtype = IFM_100G_SR4,
117 .baudrate = IF_Gbps(100ULL),
119 [MLX5E_100GBASE_KR4] = {
120 .subtype = IFM_100G_KR4,
121 .baudrate = IF_Gbps(100ULL),
123 [MLX5E_100GBASE_LR4] = {
124 .subtype = IFM_100G_LR4,
125 .baudrate = IF_Gbps(100ULL),
127 [MLX5E_100BASE_TX] = {
128 .subtype = IFM_100_TX,
129 .baudrate = IF_Mbps(100ULL),
131 [MLX5E_1000BASE_T] = {
132 .subtype = IFM_1000_T,
133 .baudrate = IF_Mbps(1000ULL),
135 [MLX5E_10GBASE_T] = {
136 .subtype = IFM_10G_T,
137 .baudrate = IF_Gbps(10ULL),
139 [MLX5E_25GBASE_CR] = {
140 .subtype = IFM_25G_CR,
141 .baudrate = IF_Gbps(25ULL),
143 [MLX5E_25GBASE_KR] = {
144 .subtype = IFM_25G_KR,
145 .baudrate = IF_Gbps(25ULL),
147 [MLX5E_25GBASE_SR] = {
148 .subtype = IFM_25G_SR,
149 .baudrate = IF_Gbps(25ULL),
151 [MLX5E_50GBASE_CR2] = {
152 .subtype = IFM_50G_CR2,
153 .baudrate = IF_Gbps(50ULL),
155 [MLX5E_50GBASE_KR2] = {
156 .subtype = IFM_50G_KR2,
157 .baudrate = IF_Gbps(50ULL),
161 MALLOC_DEFINE(M_MLX5EN, "MLX5EN", "MLX5 Ethernet");
164 mlx5e_update_carrier(struct mlx5e_priv *priv)
166 struct mlx5_core_dev *mdev = priv->mdev;
167 u32 out[MLX5_ST_SZ_DW(ptys_reg)];
174 port_state = mlx5_query_vport_state(mdev,
175 MLX5_QUERY_VPORT_STATE_IN_OP_MOD_VNIC_VPORT, 0);
177 if (port_state == VPORT_STATE_UP) {
178 priv->media_status_last |= IFM_ACTIVE;
180 priv->media_status_last &= ~IFM_ACTIVE;
181 priv->media_active_last = IFM_ETHER;
182 if_link_state_change(priv->ifp, LINK_STATE_DOWN);
186 error = mlx5_query_port_ptys(mdev, out, sizeof(out), MLX5_PTYS_EN, 1);
188 priv->media_active_last = IFM_ETHER;
189 priv->ifp->if_baudrate = 1;
190 if_printf(priv->ifp, "%s: query port ptys failed: 0x%x\n",
194 eth_proto_oper = MLX5_GET(ptys_reg, out, eth_proto_oper);
196 for (i = 0; i != MLX5E_LINK_MODES_NUMBER; i++) {
197 if (mlx5e_mode_table[i].baudrate == 0)
199 if (MLX5E_PROT_MASK(i) & eth_proto_oper) {
200 u32 subtype = mlx5e_mode_table[i].subtype;
202 priv->ifp->if_baudrate =
203 mlx5e_mode_table[i].baudrate;
207 error = mlx5_query_pddr_range_info(mdev, 1, &is_er_type);
209 if_printf(priv->ifp, "%s: query port pddr failed: %d\n",
212 if (error != 0 || is_er_type == 0)
213 subtype = IFM_10G_LR;
216 error = mlx5_query_pddr_range_info(mdev, 1, &is_er_type);
218 if_printf(priv->ifp, "%s: query port pddr failed: %d\n",
221 if (error == 0 && is_er_type != 0)
222 subtype = IFM_40G_ER4;
225 priv->media_active_last = subtype | IFM_ETHER | IFM_FDX;
229 if_link_state_change(priv->ifp, LINK_STATE_UP);
233 mlx5e_media_status(struct ifnet *dev, struct ifmediareq *ifmr)
235 struct mlx5e_priv *priv = dev->if_softc;
237 ifmr->ifm_status = priv->media_status_last;
238 ifmr->ifm_active = priv->media_active_last |
239 (priv->params.rx_pauseframe_control ? IFM_ETH_RXPAUSE : 0) |
240 (priv->params.tx_pauseframe_control ? IFM_ETH_TXPAUSE : 0);
245 mlx5e_find_link_mode(u32 subtype)
252 subtype = IFM_10G_ER;
255 subtype = IFM_40G_LR4;
259 for (i = 0; i < MLX5E_LINK_MODES_NUMBER; ++i) {
260 if (mlx5e_mode_table[i].baudrate == 0)
262 if (mlx5e_mode_table[i].subtype == subtype)
263 link_mode |= MLX5E_PROT_MASK(i);
270 mlx5e_set_port_pause_and_pfc(struct mlx5e_priv *priv)
272 return (mlx5_set_port_pause_and_pfc(priv->mdev, 1,
273 priv->params.rx_pauseframe_control,
274 priv->params.tx_pauseframe_control,
275 priv->params.rx_priority_flow_control,
276 priv->params.tx_priority_flow_control));
280 mlx5e_set_port_pfc(struct mlx5e_priv *priv)
284 if (priv->gone != 0) {
286 } else if (priv->params.rx_pauseframe_control ||
287 priv->params.tx_pauseframe_control) {
289 "Global pauseframes must be disabled before enabling PFC.\n");
292 error = mlx5e_set_port_pause_and_pfc(priv);
298 mlx5e_media_change(struct ifnet *dev)
300 struct mlx5e_priv *priv = dev->if_softc;
301 struct mlx5_core_dev *mdev = priv->mdev;
308 locked = PRIV_LOCKED(priv);
312 if (IFM_TYPE(priv->media.ifm_media) != IFM_ETHER) {
316 link_mode = mlx5e_find_link_mode(IFM_SUBTYPE(priv->media.ifm_media));
318 /* query supported capabilities */
319 error = mlx5_query_port_proto_cap(mdev, ð_proto_cap, MLX5_PTYS_EN);
321 if_printf(dev, "Query port media capability failed\n");
324 /* check for autoselect */
325 if (IFM_SUBTYPE(priv->media.ifm_media) == IFM_AUTO) {
326 link_mode = eth_proto_cap;
327 if (link_mode == 0) {
328 if_printf(dev, "Port media capability is zero\n");
333 link_mode = link_mode & eth_proto_cap;
334 if (link_mode == 0) {
335 if_printf(dev, "Not supported link mode requested\n");
340 if (priv->media.ifm_media & (IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE)) {
341 /* check if PFC is enabled */
342 if (priv->params.rx_priority_flow_control ||
343 priv->params.tx_priority_flow_control) {
344 if_printf(dev, "PFC must be disabled before enabling global pauseframes.\n");
349 /* update pauseframe control bits */
350 priv->params.rx_pauseframe_control =
351 (priv->media.ifm_media & IFM_ETH_RXPAUSE) ? 1 : 0;
352 priv->params.tx_pauseframe_control =
353 (priv->media.ifm_media & IFM_ETH_TXPAUSE) ? 1 : 0;
355 /* check if device is opened */
356 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
358 /* reconfigure the hardware */
359 mlx5_set_port_status(mdev, MLX5_PORT_DOWN);
360 mlx5_set_port_proto(mdev, link_mode, MLX5_PTYS_EN);
361 error = -mlx5e_set_port_pause_and_pfc(priv);
363 mlx5_set_port_status(mdev, MLX5_PORT_UP);
372 mlx5e_update_carrier_work(struct work_struct *work)
374 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
375 update_carrier_work);
378 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
379 mlx5e_update_carrier(priv);
384 * This function reads the physical port counters from the firmware
385 * using a pre-defined layout defined by various MLX5E_PPORT_XXX()
386 * macros. The output is converted from big-endian 64-bit values into
387 * host endian ones and stored in the "priv->stats.pport" structure.
390 mlx5e_update_pport_counters(struct mlx5e_priv *priv)
392 struct mlx5_core_dev *mdev = priv->mdev;
393 struct mlx5e_pport_stats *s = &priv->stats.pport;
394 struct mlx5e_port_stats_debug *s_debug = &priv->stats.port_stats_debug;
398 unsigned sz = MLX5_ST_SZ_BYTES(ppcnt_reg);
403 /* allocate firmware request structures */
404 in = mlx5_vzalloc(sz);
405 out = mlx5_vzalloc(sz);
406 if (in == NULL || out == NULL)
410 * Get pointer to the 64-bit counter set which is located at a
411 * fixed offset in the output firmware request structure:
413 ptr = (const uint64_t *)MLX5_ADDR_OF(ppcnt_reg, out, counter_set);
415 MLX5_SET(ppcnt_reg, in, local_port, 1);
417 /* read IEEE802_3 counter group using predefined counter layout */
418 MLX5_SET(ppcnt_reg, in, grp, MLX5_IEEE_802_3_COUNTERS_GROUP);
419 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
420 for (x = 0, y = MLX5E_PPORT_PER_PRIO_STATS_NUM;
421 x != MLX5E_PPORT_IEEE802_3_STATS_NUM; x++, y++)
422 s->arg[y] = be64toh(ptr[x]);
424 /* read RFC2819 counter group using predefined counter layout */
425 MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2819_COUNTERS_GROUP);
426 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
427 for (x = 0; x != MLX5E_PPORT_RFC2819_STATS_NUM; x++, y++)
428 s->arg[y] = be64toh(ptr[x]);
429 for (y = 0; x != MLX5E_PPORT_RFC2819_STATS_NUM +
430 MLX5E_PPORT_RFC2819_STATS_DEBUG_NUM; x++, y++)
431 s_debug->arg[y] = be64toh(ptr[x]);
433 /* read RFC2863 counter group using predefined counter layout */
434 MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2863_COUNTERS_GROUP);
435 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
436 for (x = 0; x != MLX5E_PPORT_RFC2863_STATS_DEBUG_NUM; x++, y++)
437 s_debug->arg[y] = be64toh(ptr[x]);
439 /* read physical layer stats counter group using predefined counter layout */
440 MLX5_SET(ppcnt_reg, in, grp, MLX5_PHYSICAL_LAYER_COUNTERS_GROUP);
441 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
442 for (x = 0; x != MLX5E_PPORT_PHYSICAL_LAYER_STATS_DEBUG_NUM; x++, y++)
443 s_debug->arg[y] = be64toh(ptr[x]);
445 /* read Extended Ethernet counter group using predefined counter layout */
446 MLX5_SET(ppcnt_reg, in, grp, MLX5_ETHERNET_EXTENDED_COUNTERS_GROUP);
447 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
448 for (x = 0; x != MLX5E_PPORT_ETHERNET_EXTENDED_STATS_DEBUG_NUM; x++, y++)
449 s_debug->arg[y] = be64toh(ptr[x]);
451 /* read per-priority counters */
452 MLX5_SET(ppcnt_reg, in, grp, MLX5_PER_PRIORITY_COUNTERS_GROUP);
454 /* iterate all the priorities */
455 for (y = z = 0; z != MLX5E_PPORT_PER_PRIO_STATS_NUM_PRIO; z++) {
456 MLX5_SET(ppcnt_reg, in, prio_tc, z);
457 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
459 /* read per priority stats counter group using predefined counter layout */
460 for (x = 0; x != (MLX5E_PPORT_PER_PRIO_STATS_NUM /
461 MLX5E_PPORT_PER_PRIO_STATS_NUM_PRIO); x++, y++)
462 s->arg[y] = be64toh(ptr[x]);
466 /* free firmware request structures */
472 * This function is called regularly to collect all statistics
473 * counters from the firmware. The values can be viewed through the
474 * sysctl interface. Execution is serialized using the priv's global
475 * configuration lock.
478 mlx5e_update_stats_locked(struct mlx5e_priv *priv)
480 struct mlx5_core_dev *mdev = priv->mdev;
481 struct mlx5e_vport_stats *s = &priv->stats.vport;
482 struct mlx5e_sq_stats *sq_stats;
483 struct buf_ring *sq_br;
484 #if (__FreeBSD_version < 1100000)
485 struct ifnet *ifp = priv->ifp;
488 u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)];
490 int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
493 u64 tx_queue_dropped = 0;
494 u64 tx_defragged = 0;
495 u64 tx_offload_none = 0;
498 u64 sw_lro_queued = 0;
499 u64 sw_lro_flushed = 0;
500 u64 rx_csum_none = 0;
502 u32 rx_out_of_buffer = 0;
506 out = mlx5_vzalloc(outlen);
510 /* Collect firts the SW counters and then HW for consistency */
511 for (i = 0; i < priv->params.num_channels; i++) {
512 struct mlx5e_channel *pch = priv->channel + i;
513 struct mlx5e_rq *rq = &pch->rq;
514 struct mlx5e_rq_stats *rq_stats = &pch->rq.stats;
516 /* collect stats from LRO */
517 rq_stats->sw_lro_queued = rq->lro.lro_queued;
518 rq_stats->sw_lro_flushed = rq->lro.lro_flushed;
519 sw_lro_queued += rq_stats->sw_lro_queued;
520 sw_lro_flushed += rq_stats->sw_lro_flushed;
521 lro_packets += rq_stats->lro_packets;
522 lro_bytes += rq_stats->lro_bytes;
523 rx_csum_none += rq_stats->csum_none;
524 rx_wqe_err += rq_stats->wqe_err;
526 for (j = 0; j < priv->num_tc; j++) {
527 sq_stats = &pch->sq[j].stats;
528 sq_br = pch->sq[j].br;
530 tso_packets += sq_stats->tso_packets;
531 tso_bytes += sq_stats->tso_bytes;
532 tx_queue_dropped += sq_stats->dropped;
534 tx_queue_dropped += sq_br->br_drops;
535 tx_defragged += sq_stats->defragged;
536 tx_offload_none += sq_stats->csum_offload_none;
540 /* update counters */
541 s->tso_packets = tso_packets;
542 s->tso_bytes = tso_bytes;
543 s->tx_queue_dropped = tx_queue_dropped;
544 s->tx_defragged = tx_defragged;
545 s->lro_packets = lro_packets;
546 s->lro_bytes = lro_bytes;
547 s->sw_lro_queued = sw_lro_queued;
548 s->sw_lro_flushed = sw_lro_flushed;
549 s->rx_csum_none = rx_csum_none;
550 s->rx_wqe_err = rx_wqe_err;
553 memset(in, 0, sizeof(in));
555 MLX5_SET(query_vport_counter_in, in, opcode,
556 MLX5_CMD_OP_QUERY_VPORT_COUNTER);
557 MLX5_SET(query_vport_counter_in, in, op_mod, 0);
558 MLX5_SET(query_vport_counter_in, in, other_vport, 0);
560 memset(out, 0, outlen);
562 /* get number of out-of-buffer drops first */
563 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0 &&
564 mlx5_vport_query_out_of_rx_buffer(mdev, priv->counter_set_id,
565 &rx_out_of_buffer) == 0) {
566 /* accumulate difference into a 64-bit counter */
567 s->rx_out_of_buffer += (u64)(u32)(rx_out_of_buffer -
568 s->rx_out_of_buffer_prev);
569 s->rx_out_of_buffer_prev = rx_out_of_buffer;
572 /* get port statistics */
573 if (mlx5_cmd_exec(mdev, in, sizeof(in), out, outlen) == 0) {
574 #define MLX5_GET_CTR(out, x) \
575 MLX5_GET64(query_vport_counter_out, out, x)
577 s->rx_error_packets =
578 MLX5_GET_CTR(out, received_errors.packets);
580 MLX5_GET_CTR(out, received_errors.octets);
581 s->tx_error_packets =
582 MLX5_GET_CTR(out, transmit_errors.packets);
584 MLX5_GET_CTR(out, transmit_errors.octets);
586 s->rx_unicast_packets =
587 MLX5_GET_CTR(out, received_eth_unicast.packets);
588 s->rx_unicast_bytes =
589 MLX5_GET_CTR(out, received_eth_unicast.octets);
590 s->tx_unicast_packets =
591 MLX5_GET_CTR(out, transmitted_eth_unicast.packets);
592 s->tx_unicast_bytes =
593 MLX5_GET_CTR(out, transmitted_eth_unicast.octets);
595 s->rx_multicast_packets =
596 MLX5_GET_CTR(out, received_eth_multicast.packets);
597 s->rx_multicast_bytes =
598 MLX5_GET_CTR(out, received_eth_multicast.octets);
599 s->tx_multicast_packets =
600 MLX5_GET_CTR(out, transmitted_eth_multicast.packets);
601 s->tx_multicast_bytes =
602 MLX5_GET_CTR(out, transmitted_eth_multicast.octets);
604 s->rx_broadcast_packets =
605 MLX5_GET_CTR(out, received_eth_broadcast.packets);
606 s->rx_broadcast_bytes =
607 MLX5_GET_CTR(out, received_eth_broadcast.octets);
608 s->tx_broadcast_packets =
609 MLX5_GET_CTR(out, transmitted_eth_broadcast.packets);
610 s->tx_broadcast_bytes =
611 MLX5_GET_CTR(out, transmitted_eth_broadcast.octets);
613 s->rx_packets = s->rx_unicast_packets +
614 s->rx_multicast_packets + s->rx_broadcast_packets -
616 s->rx_bytes = s->rx_unicast_bytes + s->rx_multicast_bytes +
617 s->rx_broadcast_bytes;
618 s->tx_packets = s->tx_unicast_packets +
619 s->tx_multicast_packets + s->tx_broadcast_packets;
620 s->tx_bytes = s->tx_unicast_bytes + s->tx_multicast_bytes +
621 s->tx_broadcast_bytes;
623 /* Update calculated offload counters */
624 s->tx_csum_offload = s->tx_packets - tx_offload_none;
625 s->rx_csum_good = s->rx_packets - s->rx_csum_none;
628 /* Get physical port counters */
629 mlx5e_update_pport_counters(priv);
631 s->tx_jumbo_packets =
632 priv->stats.port_stats_debug.tx_stat_p1519to2047octets +
633 priv->stats.port_stats_debug.tx_stat_p2048to4095octets +
634 priv->stats.port_stats_debug.tx_stat_p4096to8191octets +
635 priv->stats.port_stats_debug.tx_stat_p8192to10239octets;
637 #if (__FreeBSD_version < 1100000)
638 /* no get_counters interface in fbsd 10 */
639 ifp->if_ipackets = s->rx_packets;
640 ifp->if_ierrors = s->rx_error_packets +
641 priv->stats.pport.alignment_err +
642 priv->stats.pport.check_seq_err +
643 priv->stats.pport.crc_align_errors +
644 priv->stats.pport.in_range_len_errors +
645 priv->stats.pport.jabbers +
646 priv->stats.pport.out_of_range_len +
647 priv->stats.pport.oversize_pkts +
648 priv->stats.pport.symbol_err +
649 priv->stats.pport.too_long_errors +
650 priv->stats.pport.undersize_pkts +
651 priv->stats.pport.unsupported_op_rx;
652 ifp->if_iqdrops = s->rx_out_of_buffer +
653 priv->stats.pport.drop_events;
654 ifp->if_opackets = s->tx_packets;
655 ifp->if_oerrors = s->tx_error_packets;
656 ifp->if_snd.ifq_drops = s->tx_queue_dropped;
657 ifp->if_ibytes = s->rx_bytes;
658 ifp->if_obytes = s->tx_bytes;
660 priv->stats.pport.collisions;
666 /* Update diagnostics, if any */
667 if (priv->params_ethtool.diag_pci_enable ||
668 priv->params_ethtool.diag_general_enable) {
669 int error = mlx5_core_get_diagnostics_full(mdev,
670 priv->params_ethtool.diag_pci_enable ? &priv->params_pci : NULL,
671 priv->params_ethtool.diag_general_enable ? &priv->params_general : NULL);
673 if_printf(priv->ifp, "Failed reading diagnostics: %d\n", error);
678 mlx5e_update_stats_work(struct work_struct *work)
680 struct mlx5e_priv *priv;
682 priv = container_of(work, struct mlx5e_priv, update_stats_work);
684 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
685 mlx5e_update_stats_locked(priv);
690 mlx5e_update_stats(void *arg)
692 struct mlx5e_priv *priv = arg;
694 queue_work(priv->wq, &priv->update_stats_work);
696 callout_reset(&priv->watchdog, hz, &mlx5e_update_stats, priv);
700 mlx5e_async_event_sub(struct mlx5e_priv *priv,
701 enum mlx5_dev_event event)
704 case MLX5_DEV_EVENT_PORT_UP:
705 case MLX5_DEV_EVENT_PORT_DOWN:
706 queue_work(priv->wq, &priv->update_carrier_work);
715 mlx5e_async_event(struct mlx5_core_dev *mdev, void *vpriv,
716 enum mlx5_dev_event event, unsigned long param)
718 struct mlx5e_priv *priv = vpriv;
720 mtx_lock(&priv->async_events_mtx);
721 if (test_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state))
722 mlx5e_async_event_sub(priv, event);
723 mtx_unlock(&priv->async_events_mtx);
727 mlx5e_enable_async_events(struct mlx5e_priv *priv)
729 set_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
733 mlx5e_disable_async_events(struct mlx5e_priv *priv)
735 mtx_lock(&priv->async_events_mtx);
736 clear_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
737 mtx_unlock(&priv->async_events_mtx);
740 static void mlx5e_calibration_callout(void *arg);
741 static int mlx5e_calibration_duration = 20;
742 static int mlx5e_fast_calibration = 1;
743 static int mlx5e_normal_calibration = 30;
745 static SYSCTL_NODE(_hw_mlx5, OID_AUTO, calibr, CTLFLAG_RW, 0,
746 "MLX5 timestamp calibration parameteres");
748 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, duration, CTLFLAG_RWTUN,
749 &mlx5e_calibration_duration, 0,
750 "Duration of initial calibration");
751 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, fast, CTLFLAG_RWTUN,
752 &mlx5e_fast_calibration, 0,
753 "Recalibration interval during initial calibration");
754 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, normal, CTLFLAG_RWTUN,
755 &mlx5e_normal_calibration, 0,
756 "Recalibration interval during normal operations");
759 * Ignites the calibration process.
762 mlx5e_reset_calibration_callout(struct mlx5e_priv *priv)
765 if (priv->clbr_done == 0)
766 mlx5e_calibration_callout(priv);
768 callout_reset_curcpu(&priv->tstmp_clbr, (priv->clbr_done <
769 mlx5e_calibration_duration ? mlx5e_fast_calibration :
770 mlx5e_normal_calibration) * hz, mlx5e_calibration_callout,
775 mlx5e_timespec2usec(const struct timespec *ts)
778 return ((uint64_t)ts->tv_sec * 1000000000 + ts->tv_nsec);
782 mlx5e_hw_clock(struct mlx5e_priv *priv)
784 struct mlx5_init_seg *iseg;
785 uint32_t hw_h, hw_h1, hw_l;
787 iseg = priv->mdev->iseg;
789 hw_h = ioread32be(&iseg->internal_timer_h);
790 hw_l = ioread32be(&iseg->internal_timer_l);
791 hw_h1 = ioread32be(&iseg->internal_timer_h);
792 } while (hw_h1 != hw_h);
793 return (((uint64_t)hw_h << 32) | hw_l);
797 * The calibration callout, it runs either in the context of the
798 * thread which enables calibration, or in callout. It takes the
799 * snapshot of system and adapter clocks, then advances the pointers to
800 * the calibration point to allow rx path to read the consistent data
804 mlx5e_calibration_callout(void *arg)
806 struct mlx5e_priv *priv;
807 struct mlx5e_clbr_point *next, *curr;
812 curr = &priv->clbr_points[priv->clbr_curr];
813 clbr_curr_next = priv->clbr_curr + 1;
814 if (clbr_curr_next >= nitems(priv->clbr_points))
816 next = &priv->clbr_points[clbr_curr_next];
818 next->base_prev = curr->base_curr;
819 next->clbr_hw_prev = curr->clbr_hw_curr;
821 next->clbr_hw_curr = mlx5e_hw_clock(priv);
822 if (((next->clbr_hw_curr - curr->clbr_hw_prev) >> MLX5E_TSTMP_PREC) ==
824 if_printf(priv->ifp, "HW failed tstmp frozen %#jx %#jx,"
825 "disabling\n", next->clbr_hw_curr, curr->clbr_hw_prev);
831 next->base_curr = mlx5e_timespec2usec(&ts);
834 atomic_thread_fence_rel();
835 priv->clbr_curr = clbr_curr_next;
836 atomic_store_rel_int(&next->clbr_gen, ++(priv->clbr_gen));
838 if (priv->clbr_done < mlx5e_calibration_duration)
840 mlx5e_reset_calibration_callout(priv);
843 static const char *mlx5e_rq_stats_desc[] = {
844 MLX5E_RQ_STATS(MLX5E_STATS_DESC)
848 mlx5e_create_rq(struct mlx5e_channel *c,
849 struct mlx5e_rq_param *param,
852 struct mlx5e_priv *priv = c->priv;
853 struct mlx5_core_dev *mdev = priv->mdev;
855 void *rqc = param->rqc;
856 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
862 err = mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
866 /* Create DMA descriptor TAG */
867 if ((err = -bus_dma_tag_create(
868 bus_get_dma_tag(mdev->pdev->dev.bsddev),
869 1, /* any alignment */
871 BUS_SPACE_MAXADDR, /* lowaddr */
872 BUS_SPACE_MAXADDR, /* highaddr */
873 NULL, NULL, /* filter, filterarg */
874 nsegs * MLX5E_MAX_RX_BYTES, /* maxsize */
875 nsegs, /* nsegments */
876 nsegs * MLX5E_MAX_RX_BYTES, /* maxsegsize */
878 NULL, NULL, /* lockfunc, lockfuncarg */
882 err = mlx5_wq_ll_create(mdev, ¶m->wq, rqc_wq, &rq->wq,
885 goto err_free_dma_tag;
887 rq->wq.db = &rq->wq.db[MLX5_RCV_DBR];
889 err = mlx5e_get_wqe_sz(priv, &rq->wqe_sz, &rq->nsegs);
891 goto err_rq_wq_destroy;
893 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
895 err = -tcp_lro_init_args(&rq->lro, c->tag.m_snd_tag.ifp, TCP_LRO_ENTRIES, wq_sz);
897 goto err_rq_wq_destroy;
899 rq->mbuf = malloc(wq_sz * sizeof(rq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
900 for (i = 0; i != wq_sz; i++) {
901 struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
902 #if (MLX5E_MAX_RX_SEGS == 1)
903 uint32_t byte_count = rq->wqe_sz - MLX5E_NET_IP_ALIGN;
908 err = -bus_dmamap_create(rq->dma_tag, 0, &rq->mbuf[i].dma_map);
911 bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
912 goto err_rq_mbuf_free;
915 /* set value for constant fields */
916 #if (MLX5E_MAX_RX_SEGS == 1)
917 wqe->data[0].lkey = c->mkey_be;
918 wqe->data[0].byte_count = cpu_to_be32(byte_count | MLX5_HW_START_PADDING);
920 for (j = 0; j < rq->nsegs; j++)
921 wqe->data[j].lkey = c->mkey_be;
925 INIT_WORK(&rq->dim.work, mlx5e_dim_work);
926 if (priv->params.rx_cq_moderation_mode < 2) {
927 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
929 void *cqc = container_of(param,
930 struct mlx5e_channel_param, rq)->rx_cq.cqc;
932 switch (MLX5_GET(cqc, cqc, cq_period_mode)) {
933 case MLX5_CQ_PERIOD_MODE_START_FROM_EQE:
934 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
936 case MLX5_CQ_PERIOD_MODE_START_FROM_CQE:
937 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
940 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
945 rq->ifp = c->tag.m_snd_tag.ifp;
949 snprintf(buffer, sizeof(buffer), "rxstat%d", c->ix);
950 mlx5e_create_stats(&rq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
951 buffer, mlx5e_rq_stats_desc, MLX5E_RQ_STATS_NUM,
956 free(rq->mbuf, M_MLX5EN);
957 tcp_lro_free(&rq->lro);
959 mlx5_wq_destroy(&rq->wq_ctrl);
961 bus_dma_tag_destroy(rq->dma_tag);
967 mlx5e_destroy_rq(struct mlx5e_rq *rq)
972 /* destroy all sysctl nodes */
973 sysctl_ctx_free(&rq->stats.ctx);
975 /* free leftover LRO packets, if any */
976 tcp_lro_free(&rq->lro);
978 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
979 for (i = 0; i != wq_sz; i++) {
980 if (rq->mbuf[i].mbuf != NULL) {
981 bus_dmamap_unload(rq->dma_tag, rq->mbuf[i].dma_map);
982 m_freem(rq->mbuf[i].mbuf);
984 bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
986 free(rq->mbuf, M_MLX5EN);
987 mlx5_wq_destroy(&rq->wq_ctrl);
991 mlx5e_enable_rq(struct mlx5e_rq *rq, struct mlx5e_rq_param *param)
993 struct mlx5e_channel *c = rq->channel;
994 struct mlx5e_priv *priv = c->priv;
995 struct mlx5_core_dev *mdev = priv->mdev;
1003 inlen = MLX5_ST_SZ_BYTES(create_rq_in) +
1004 sizeof(u64) * rq->wq_ctrl.buf.npages;
1005 in = mlx5_vzalloc(inlen);
1009 rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
1010 wq = MLX5_ADDR_OF(rqc, rqc, wq);
1012 memcpy(rqc, param->rqc, sizeof(param->rqc));
1014 MLX5_SET(rqc, rqc, cqn, c->rq.cq.mcq.cqn);
1015 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST);
1016 MLX5_SET(rqc, rqc, flush_in_error_en, 1);
1017 if (priv->counter_set_id >= 0)
1018 MLX5_SET(rqc, rqc, counter_set_id, priv->counter_set_id);
1019 MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift -
1021 MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma);
1023 mlx5_fill_page_array(&rq->wq_ctrl.buf,
1024 (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
1026 err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn);
1034 mlx5e_modify_rq(struct mlx5e_rq *rq, int curr_state, int next_state)
1036 struct mlx5e_channel *c = rq->channel;
1037 struct mlx5e_priv *priv = c->priv;
1038 struct mlx5_core_dev *mdev = priv->mdev;
1045 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
1046 in = mlx5_vzalloc(inlen);
1050 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
1052 MLX5_SET(modify_rq_in, in, rqn, rq->rqn);
1053 MLX5_SET(modify_rq_in, in, rq_state, curr_state);
1054 MLX5_SET(rqc, rqc, state, next_state);
1056 err = mlx5_core_modify_rq(mdev, in, inlen);
1064 mlx5e_disable_rq(struct mlx5e_rq *rq)
1066 struct mlx5e_channel *c = rq->channel;
1067 struct mlx5e_priv *priv = c->priv;
1068 struct mlx5_core_dev *mdev = priv->mdev;
1070 mlx5_core_destroy_rq(mdev, rq->rqn);
1074 mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq)
1076 struct mlx5e_channel *c = rq->channel;
1077 struct mlx5e_priv *priv = c->priv;
1078 struct mlx5_wq_ll *wq = &rq->wq;
1081 for (i = 0; i < 1000; i++) {
1082 if (wq->cur_sz >= priv->params.min_rx_wqes)
1087 return (-ETIMEDOUT);
1091 mlx5e_open_rq(struct mlx5e_channel *c,
1092 struct mlx5e_rq_param *param,
1093 struct mlx5e_rq *rq)
1097 err = mlx5e_create_rq(c, param, rq);
1101 err = mlx5e_enable_rq(rq, param);
1103 goto err_destroy_rq;
1105 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
1107 goto err_disable_rq;
1114 mlx5e_disable_rq(rq);
1116 mlx5e_destroy_rq(rq);
1122 mlx5e_close_rq(struct mlx5e_rq *rq)
1126 callout_stop(&rq->watchdog);
1127 mtx_unlock(&rq->mtx);
1129 callout_drain(&rq->watchdog);
1131 mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
1135 mlx5e_close_rq_wait(struct mlx5e_rq *rq)
1138 mlx5e_disable_rq(rq);
1139 mlx5e_close_cq(&rq->cq);
1140 cancel_work_sync(&rq->dim.work);
1141 mlx5e_destroy_rq(rq);
1145 mlx5e_free_sq_db(struct mlx5e_sq *sq)
1147 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1150 for (x = 0; x != wq_sz; x++) {
1151 if (sq->mbuf[x].mbuf != NULL) {
1152 bus_dmamap_unload(sq->dma_tag, sq->mbuf[x].dma_map);
1153 m_freem(sq->mbuf[x].mbuf);
1155 bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1157 free(sq->mbuf, M_MLX5EN);
1161 mlx5e_alloc_sq_db(struct mlx5e_sq *sq)
1163 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1167 sq->mbuf = malloc(wq_sz * sizeof(sq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
1169 /* Create DMA descriptor MAPs */
1170 for (x = 0; x != wq_sz; x++) {
1171 err = -bus_dmamap_create(sq->dma_tag, 0, &sq->mbuf[x].dma_map);
1174 bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1175 free(sq->mbuf, M_MLX5EN);
1182 static const char *mlx5e_sq_stats_desc[] = {
1183 MLX5E_SQ_STATS(MLX5E_STATS_DESC)
1187 mlx5e_update_sq_inline(struct mlx5e_sq *sq)
1189 sq->max_inline = sq->priv->params.tx_max_inline;
1190 sq->min_inline_mode = sq->priv->params.tx_min_inline_mode;
1193 * Check if trust state is DSCP or if inline mode is NONE which
1194 * indicates CX-5 or newer hardware.
1196 if (sq->priv->params_ethtool.trust_state != MLX5_QPTS_TRUST_PCP ||
1197 sq->min_inline_mode == MLX5_INLINE_MODE_NONE) {
1198 if (MLX5_CAP_ETH(sq->priv->mdev, wqe_vlan_insert))
1199 sq->min_insert_caps = MLX5E_INSERT_VLAN | MLX5E_INSERT_NON_VLAN;
1201 sq->min_insert_caps = MLX5E_INSERT_NON_VLAN;
1203 sq->min_insert_caps = 0;
1208 mlx5e_refresh_sq_inline_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
1212 for (i = 0; i != c->num_tc; i++) {
1213 mtx_lock(&c->sq[i].lock);
1214 mlx5e_update_sq_inline(&c->sq[i]);
1215 mtx_unlock(&c->sq[i].lock);
1220 mlx5e_refresh_sq_inline(struct mlx5e_priv *priv)
1224 /* check if channels are closed */
1225 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
1228 for (i = 0; i < priv->params.num_channels; i++)
1229 mlx5e_refresh_sq_inline_sub(priv, &priv->channel[i]);
1233 mlx5e_create_sq(struct mlx5e_channel *c,
1235 struct mlx5e_sq_param *param,
1236 struct mlx5e_sq *sq)
1238 struct mlx5e_priv *priv = c->priv;
1239 struct mlx5_core_dev *mdev = priv->mdev;
1241 void *sqc = param->sqc;
1242 void *sqc_wq = MLX5_ADDR_OF(sqc, sqc, wq);
1245 /* Create DMA descriptor TAG */
1246 if ((err = -bus_dma_tag_create(
1247 bus_get_dma_tag(mdev->pdev->dev.bsddev),
1248 1, /* any alignment */
1249 0, /* no boundary */
1250 BUS_SPACE_MAXADDR, /* lowaddr */
1251 BUS_SPACE_MAXADDR, /* highaddr */
1252 NULL, NULL, /* filter, filterarg */
1253 MLX5E_MAX_TX_PAYLOAD_SIZE, /* maxsize */
1254 MLX5E_MAX_TX_MBUF_FRAGS, /* nsegments */
1255 MLX5E_MAX_TX_MBUF_SIZE, /* maxsegsize */
1257 NULL, NULL, /* lockfunc, lockfuncarg */
1261 err = mlx5_alloc_map_uar(mdev, &sq->uar);
1263 goto err_free_dma_tag;
1265 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, &sq->wq,
1268 goto err_unmap_free_uar;
1270 sq->wq.db = &sq->wq.db[MLX5_SND_DBR];
1271 sq->bf_buf_size = (1 << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2;
1273 err = mlx5e_alloc_sq_db(sq);
1275 goto err_sq_wq_destroy;
1277 sq->mkey_be = c->mkey_be;
1278 sq->ifp = priv->ifp;
1282 mlx5e_update_sq_inline(sq);
1284 snprintf(buffer, sizeof(buffer), "txstat%dtc%d", c->ix, tc);
1285 mlx5e_create_stats(&sq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
1286 buffer, mlx5e_sq_stats_desc, MLX5E_SQ_STATS_NUM,
1292 mlx5_wq_destroy(&sq->wq_ctrl);
1295 mlx5_unmap_free_uar(mdev, &sq->uar);
1298 bus_dma_tag_destroy(sq->dma_tag);
1304 mlx5e_destroy_sq(struct mlx5e_sq *sq)
1306 /* destroy all sysctl nodes */
1307 sysctl_ctx_free(&sq->stats.ctx);
1309 mlx5e_free_sq_db(sq);
1310 mlx5_wq_destroy(&sq->wq_ctrl);
1311 mlx5_unmap_free_uar(sq->priv->mdev, &sq->uar);
1315 mlx5e_enable_sq(struct mlx5e_sq *sq, struct mlx5e_sq_param *param,
1324 inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
1325 sizeof(u64) * sq->wq_ctrl.buf.npages;
1326 in = mlx5_vzalloc(inlen);
1330 sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
1331 wq = MLX5_ADDR_OF(sqc, sqc, wq);
1333 memcpy(sqc, param->sqc, sizeof(param->sqc));
1335 MLX5_SET(sqc, sqc, tis_num_0, tis_num);
1336 MLX5_SET(sqc, sqc, cqn, sq->cq.mcq.cqn);
1337 MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
1338 MLX5_SET(sqc, sqc, tis_lst_sz, 1);
1339 MLX5_SET(sqc, sqc, flush_in_error_en, 1);
1341 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
1342 MLX5_SET(wq, wq, uar_page, sq->uar.index);
1343 MLX5_SET(wq, wq, log_wq_pg_sz, sq->wq_ctrl.buf.page_shift -
1345 MLX5_SET64(wq, wq, dbr_addr, sq->wq_ctrl.db.dma);
1347 mlx5_fill_page_array(&sq->wq_ctrl.buf,
1348 (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
1350 err = mlx5_core_create_sq(sq->priv->mdev, in, inlen, &sq->sqn);
1358 mlx5e_modify_sq(struct mlx5e_sq *sq, int curr_state, int next_state)
1365 inlen = MLX5_ST_SZ_BYTES(modify_sq_in);
1366 in = mlx5_vzalloc(inlen);
1370 sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx);
1372 MLX5_SET(modify_sq_in, in, sqn, sq->sqn);
1373 MLX5_SET(modify_sq_in, in, sq_state, curr_state);
1374 MLX5_SET(sqc, sqc, state, next_state);
1376 err = mlx5_core_modify_sq(sq->priv->mdev, in, inlen);
1384 mlx5e_disable_sq(struct mlx5e_sq *sq)
1387 mlx5_core_destroy_sq(sq->priv->mdev, sq->sqn);
1391 mlx5e_open_sq(struct mlx5e_channel *c,
1393 struct mlx5e_sq_param *param,
1394 struct mlx5e_sq *sq)
1398 err = mlx5e_create_sq(c, tc, param, sq);
1402 err = mlx5e_enable_sq(sq, param, c->priv->tisn[tc]);
1404 goto err_destroy_sq;
1406 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY);
1408 goto err_disable_sq;
1410 WRITE_ONCE(sq->running, 1);
1415 mlx5e_disable_sq(sq);
1417 mlx5e_destroy_sq(sq);
1423 mlx5e_sq_send_nops_locked(struct mlx5e_sq *sq, int can_sleep)
1425 /* fill up remainder with NOPs */
1426 while (sq->cev_counter != 0) {
1427 while (!mlx5e_sq_has_room_for(sq, 1)) {
1428 if (can_sleep != 0) {
1429 mtx_unlock(&sq->lock);
1431 mtx_lock(&sq->lock);
1436 /* send a single NOP */
1437 mlx5e_send_nop(sq, 1);
1438 atomic_thread_fence_rel();
1441 /* Check if we need to write the doorbell */
1442 if (likely(sq->doorbell.d64 != 0)) {
1443 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
1444 sq->doorbell.d64 = 0;
1449 mlx5e_sq_cev_timeout(void *arg)
1451 struct mlx5e_sq *sq = arg;
1453 mtx_assert(&sq->lock, MA_OWNED);
1455 /* check next state */
1456 switch (sq->cev_next_state) {
1457 case MLX5E_CEV_STATE_SEND_NOPS:
1458 /* fill TX ring with NOPs, if any */
1459 mlx5e_sq_send_nops_locked(sq, 0);
1461 /* check if completed */
1462 if (sq->cev_counter == 0) {
1463 sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
1468 /* send NOPs on next timeout */
1469 sq->cev_next_state = MLX5E_CEV_STATE_SEND_NOPS;
1474 callout_reset_curcpu(&sq->cev_callout, hz, mlx5e_sq_cev_timeout, sq);
1478 mlx5e_drain_sq(struct mlx5e_sq *sq)
1481 struct mlx5_core_dev *mdev= sq->priv->mdev;
1484 * Check if already stopped.
1486 * NOTE: Serialization of this function is managed by the
1487 * caller ensuring the priv's state lock is locked or in case
1488 * of rate limit support, a single thread manages drain and
1489 * resume of SQs. The "running" variable can therefore safely
1490 * be read without any locks.
1492 if (READ_ONCE(sq->running) == 0)
1495 /* don't put more packets into the SQ */
1496 WRITE_ONCE(sq->running, 0);
1498 /* serialize access to DMA rings */
1499 mtx_lock(&sq->lock);
1501 /* teardown event factor timer, if any */
1502 sq->cev_next_state = MLX5E_CEV_STATE_HOLD_NOPS;
1503 callout_stop(&sq->cev_callout);
1505 /* send dummy NOPs in order to flush the transmit ring */
1506 mlx5e_sq_send_nops_locked(sq, 1);
1507 mtx_unlock(&sq->lock);
1509 /* make sure it is safe to free the callout */
1510 callout_drain(&sq->cev_callout);
1512 /* wait till SQ is empty or link is down */
1513 mtx_lock(&sq->lock);
1514 while (sq->cc != sq->pc &&
1515 (sq->priv->media_status_last & IFM_ACTIVE) != 0 &&
1516 mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1517 mtx_unlock(&sq->lock);
1519 sq->cq.mcq.comp(&sq->cq.mcq);
1520 mtx_lock(&sq->lock);
1522 mtx_unlock(&sq->lock);
1524 /* error out remaining requests */
1525 error = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RDY, MLX5_SQC_STATE_ERR);
1528 "mlx5e_modify_sq() from RDY to ERR failed: %d\n", error);
1531 /* wait till SQ is empty */
1532 mtx_lock(&sq->lock);
1533 while (sq->cc != sq->pc &&
1534 mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1535 mtx_unlock(&sq->lock);
1537 sq->cq.mcq.comp(&sq->cq.mcq);
1538 mtx_lock(&sq->lock);
1540 mtx_unlock(&sq->lock);
1544 mlx5e_close_sq_wait(struct mlx5e_sq *sq)
1548 mlx5e_disable_sq(sq);
1549 mlx5e_destroy_sq(sq);
1553 mlx5e_create_cq(struct mlx5e_priv *priv,
1554 struct mlx5e_cq_param *param,
1555 struct mlx5e_cq *cq,
1556 mlx5e_cq_comp_t *comp,
1559 struct mlx5_core_dev *mdev = priv->mdev;
1560 struct mlx5_core_cq *mcq = &cq->mcq;
1566 param->wq.buf_numa_node = 0;
1567 param->wq.db_numa_node = 0;
1569 err = mlx5_cqwq_create(mdev, ¶m->wq, param->cqc, &cq->wq,
1574 mlx5_vector2eqn(mdev, eq_ix, &eqn_not_used, &irqn);
1577 mcq->set_ci_db = cq->wq_ctrl.db.db;
1578 mcq->arm_db = cq->wq_ctrl.db.db + 1;
1579 *mcq->set_ci_db = 0;
1581 mcq->vector = eq_ix;
1583 mcq->event = mlx5e_cq_error_event;
1585 mcq->uar = &priv->cq_uar;
1587 for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) {
1588 struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i);
1599 mlx5e_destroy_cq(struct mlx5e_cq *cq)
1601 mlx5_wq_destroy(&cq->wq_ctrl);
1605 mlx5e_enable_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param, int eq_ix)
1607 struct mlx5_core_cq *mcq = &cq->mcq;
1615 inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
1616 sizeof(u64) * cq->wq_ctrl.buf.npages;
1617 in = mlx5_vzalloc(inlen);
1621 cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
1623 memcpy(cqc, param->cqc, sizeof(param->cqc));
1625 mlx5_fill_page_array(&cq->wq_ctrl.buf,
1626 (__be64 *) MLX5_ADDR_OF(create_cq_in, in, pas));
1628 mlx5_vector2eqn(cq->priv->mdev, eq_ix, &eqn, &irqn_not_used);
1630 MLX5_SET(cqc, cqc, c_eqn, eqn);
1631 MLX5_SET(cqc, cqc, uar_page, mcq->uar->index);
1632 MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.buf.page_shift -
1634 MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
1636 err = mlx5_core_create_cq(cq->priv->mdev, mcq, in, inlen);
1643 mlx5e_cq_arm(cq, MLX5_GET_DOORBELL_LOCK(&cq->priv->doorbell_lock));
1649 mlx5e_disable_cq(struct mlx5e_cq *cq)
1652 mlx5_core_destroy_cq(cq->priv->mdev, &cq->mcq);
1656 mlx5e_open_cq(struct mlx5e_priv *priv,
1657 struct mlx5e_cq_param *param,
1658 struct mlx5e_cq *cq,
1659 mlx5e_cq_comp_t *comp,
1664 err = mlx5e_create_cq(priv, param, cq, comp, eq_ix);
1668 err = mlx5e_enable_cq(cq, param, eq_ix);
1670 goto err_destroy_cq;
1675 mlx5e_destroy_cq(cq);
1681 mlx5e_close_cq(struct mlx5e_cq *cq)
1683 mlx5e_disable_cq(cq);
1684 mlx5e_destroy_cq(cq);
1688 mlx5e_open_tx_cqs(struct mlx5e_channel *c,
1689 struct mlx5e_channel_param *cparam)
1694 for (tc = 0; tc < c->num_tc; tc++) {
1695 /* open completion queue */
1696 err = mlx5e_open_cq(c->priv, &cparam->tx_cq, &c->sq[tc].cq,
1697 &mlx5e_tx_cq_comp, c->ix);
1699 goto err_close_tx_cqs;
1704 for (tc--; tc >= 0; tc--)
1705 mlx5e_close_cq(&c->sq[tc].cq);
1711 mlx5e_close_tx_cqs(struct mlx5e_channel *c)
1715 for (tc = 0; tc < c->num_tc; tc++)
1716 mlx5e_close_cq(&c->sq[tc].cq);
1720 mlx5e_open_sqs(struct mlx5e_channel *c,
1721 struct mlx5e_channel_param *cparam)
1726 for (tc = 0; tc < c->num_tc; tc++) {
1727 err = mlx5e_open_sq(c, tc, &cparam->sq, &c->sq[tc]);
1735 for (tc--; tc >= 0; tc--)
1736 mlx5e_close_sq_wait(&c->sq[tc]);
1742 mlx5e_close_sqs_wait(struct mlx5e_channel *c)
1746 for (tc = 0; tc < c->num_tc; tc++)
1747 mlx5e_close_sq_wait(&c->sq[tc]);
1751 mlx5e_chan_mtx_init(struct mlx5e_channel *c)
1755 mtx_init(&c->rq.mtx, "mlx5rx", MTX_NETWORK_LOCK, MTX_DEF);
1757 callout_init_mtx(&c->rq.watchdog, &c->rq.mtx, 0);
1759 for (tc = 0; tc < c->num_tc; tc++) {
1760 struct mlx5e_sq *sq = c->sq + tc;
1762 mtx_init(&sq->lock, "mlx5tx",
1763 MTX_NETWORK_LOCK " TX", MTX_DEF);
1764 mtx_init(&sq->comp_lock, "mlx5comp",
1765 MTX_NETWORK_LOCK " TX", MTX_DEF);
1767 callout_init_mtx(&sq->cev_callout, &sq->lock, 0);
1769 sq->cev_factor = c->priv->params_ethtool.tx_completion_fact;
1771 /* ensure the TX completion event factor is not zero */
1772 if (sq->cev_factor == 0)
1778 mlx5e_chan_mtx_destroy(struct mlx5e_channel *c)
1782 mtx_destroy(&c->rq.mtx);
1784 for (tc = 0; tc < c->num_tc; tc++) {
1785 mtx_destroy(&c->sq[tc].lock);
1786 mtx_destroy(&c->sq[tc].comp_lock);
1791 mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
1792 struct mlx5e_channel_param *cparam,
1793 struct mlx5e_channel *c)
1797 memset(c, 0, sizeof(*c));
1801 /* setup send tag */
1802 c->tag.m_snd_tag.ifp = priv->ifp;
1803 c->tag.type = IF_SND_TAG_TYPE_UNLIMITED;
1804 c->mkey_be = cpu_to_be32(priv->mr.key);
1805 c->num_tc = priv->num_tc;
1808 mlx5e_chan_mtx_init(c);
1810 /* open transmit completion queue */
1811 err = mlx5e_open_tx_cqs(c, cparam);
1815 /* open receive completion queue */
1816 err = mlx5e_open_cq(c->priv, &cparam->rx_cq, &c->rq.cq,
1817 &mlx5e_rx_cq_comp, c->ix);
1819 goto err_close_tx_cqs;
1821 err = mlx5e_open_sqs(c, cparam);
1823 goto err_close_rx_cq;
1825 err = mlx5e_open_rq(c, &cparam->rq, &c->rq);
1829 /* poll receive queue initially */
1830 c->rq.cq.mcq.comp(&c->rq.cq.mcq);
1835 mlx5e_close_sqs_wait(c);
1838 mlx5e_close_cq(&c->rq.cq);
1841 mlx5e_close_tx_cqs(c);
1844 /* destroy mutexes */
1845 mlx5e_chan_mtx_destroy(c);
1850 mlx5e_close_channel(struct mlx5e_channel *c)
1852 mlx5e_close_rq(&c->rq);
1856 mlx5e_close_channel_wait(struct mlx5e_channel *c)
1858 mlx5e_close_rq_wait(&c->rq);
1859 mlx5e_close_sqs_wait(c);
1860 mlx5e_close_tx_cqs(c);
1861 /* destroy mutexes */
1862 mlx5e_chan_mtx_destroy(c);
1866 mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs)
1870 r = priv->params.hw_lro_en ? priv->params.lro_wqe_sz :
1871 MLX5E_SW2MB_MTU(priv->ifp->if_mtu);
1872 if (r > MJUM16BYTES)
1877 else if (r > MJUMPAGESIZE)
1879 else if (r > MCLBYTES)
1885 * n + 1 must be a power of two, because stride size must be.
1886 * Stride size is 16 * (n + 1), as the first segment is
1889 for (n = howmany(r, MLX5E_MAX_RX_BYTES); !powerof2(n + 1); n++)
1898 mlx5e_build_rq_param(struct mlx5e_priv *priv,
1899 struct mlx5e_rq_param *param)
1901 void *rqc = param->rqc;
1902 void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
1905 mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
1906 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
1907 MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN);
1908 MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe) +
1909 nsegs * sizeof(struct mlx5_wqe_data_seg)));
1910 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_rq_size);
1911 MLX5_SET(wq, wq, pd, priv->pdn);
1913 param->wq.buf_numa_node = 0;
1914 param->wq.db_numa_node = 0;
1915 param->wq.linear = 1;
1919 mlx5e_build_sq_param(struct mlx5e_priv *priv,
1920 struct mlx5e_sq_param *param)
1922 void *sqc = param->sqc;
1923 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
1925 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_sq_size);
1926 MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
1927 MLX5_SET(wq, wq, pd, priv->pdn);
1929 param->wq.buf_numa_node = 0;
1930 param->wq.db_numa_node = 0;
1931 param->wq.linear = 1;
1935 mlx5e_build_common_cq_param(struct mlx5e_priv *priv,
1936 struct mlx5e_cq_param *param)
1938 void *cqc = param->cqc;
1940 MLX5_SET(cqc, cqc, uar_page, priv->cq_uar.index);
1944 mlx5e_get_default_profile(struct mlx5e_priv *priv, int mode, struct net_dim_cq_moder *ptr)
1947 *ptr = net_dim_get_profile(mode, MLX5E_DIM_DEFAULT_PROFILE);
1949 /* apply LRO restrictions */
1950 if (priv->params.hw_lro_en &&
1951 ptr->pkts > MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO) {
1952 ptr->pkts = MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO;
1957 mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
1958 struct mlx5e_cq_param *param)
1960 struct net_dim_cq_moder curr;
1961 void *cqc = param->cqc;
1964 * We use MLX5_CQE_FORMAT_HASH because the RX hash mini CQE
1965 * format is more beneficial for FreeBSD use case.
1967 * Adding support for MLX5_CQE_FORMAT_CSUM will require changes
1968 * in mlx5e_decompress_cqe.
1970 if (priv->params.cqe_zipping_en) {
1971 MLX5_SET(cqc, cqc, mini_cqe_res_format, MLX5_CQE_FORMAT_HASH);
1972 MLX5_SET(cqc, cqc, cqe_compression_en, 1);
1975 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_rq_size);
1977 switch (priv->params.rx_cq_moderation_mode) {
1979 MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
1980 MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
1981 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1984 MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
1985 MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
1986 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
1987 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
1989 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1992 mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE, &curr);
1993 MLX5_SET(cqc, cqc, cq_period, curr.usec);
1994 MLX5_SET(cqc, cqc, cq_max_count, curr.pkts);
1995 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1998 mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE, &curr);
1999 MLX5_SET(cqc, cqc, cq_period, curr.usec);
2000 MLX5_SET(cqc, cqc, cq_max_count, curr.pkts);
2001 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
2002 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
2004 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2010 mlx5e_dim_build_cq_param(priv, param);
2012 mlx5e_build_common_cq_param(priv, param);
2016 mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
2017 struct mlx5e_cq_param *param)
2019 void *cqc = param->cqc;
2021 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_sq_size);
2022 MLX5_SET(cqc, cqc, cq_period, priv->params.tx_cq_moderation_usec);
2023 MLX5_SET(cqc, cqc, cq_max_count, priv->params.tx_cq_moderation_pkts);
2025 switch (priv->params.tx_cq_moderation_mode) {
2027 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2030 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
2031 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
2033 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2037 mlx5e_build_common_cq_param(priv, param);
2041 mlx5e_build_channel_param(struct mlx5e_priv *priv,
2042 struct mlx5e_channel_param *cparam)
2044 memset(cparam, 0, sizeof(*cparam));
2046 mlx5e_build_rq_param(priv, &cparam->rq);
2047 mlx5e_build_sq_param(priv, &cparam->sq);
2048 mlx5e_build_rx_cq_param(priv, &cparam->rx_cq);
2049 mlx5e_build_tx_cq_param(priv, &cparam->tx_cq);
2053 mlx5e_open_channels(struct mlx5e_priv *priv)
2055 struct mlx5e_channel_param cparam;
2060 mlx5e_build_channel_param(priv, &cparam);
2061 for (i = 0; i < priv->params.num_channels; i++) {
2062 err = mlx5e_open_channel(priv, i, &cparam, &priv->channel[i]);
2064 goto err_close_channels;
2067 for (j = 0; j < priv->params.num_channels; j++) {
2068 err = mlx5e_wait_for_min_rx_wqes(&priv->channel[j].rq);
2070 goto err_close_channels;
2076 mlx5e_close_channel(&priv->channel[i]);
2077 mlx5e_close_channel_wait(&priv->channel[i]);
2083 mlx5e_close_channels(struct mlx5e_priv *priv)
2087 for (i = 0; i < priv->params.num_channels; i++)
2088 mlx5e_close_channel(&priv->channel[i]);
2089 for (i = 0; i < priv->params.num_channels; i++)
2090 mlx5e_close_channel_wait(&priv->channel[i]);
2094 mlx5e_refresh_sq_params(struct mlx5e_priv *priv, struct mlx5e_sq *sq)
2097 if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
2100 switch (priv->params.tx_cq_moderation_mode) {
2103 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2106 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2110 return (mlx5_core_modify_cq_moderation_mode(priv->mdev, &sq->cq.mcq,
2111 priv->params.tx_cq_moderation_usec,
2112 priv->params.tx_cq_moderation_pkts,
2116 return (mlx5_core_modify_cq_moderation(priv->mdev, &sq->cq.mcq,
2117 priv->params.tx_cq_moderation_usec,
2118 priv->params.tx_cq_moderation_pkts));
2122 mlx5e_refresh_rq_params(struct mlx5e_priv *priv, struct mlx5e_rq *rq)
2125 if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
2130 switch (priv->params.rx_cq_moderation_mode) {
2133 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2134 dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
2137 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2138 dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
2142 /* tear down dynamic interrupt moderation */
2144 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
2145 mtx_unlock(&rq->mtx);
2147 /* wait for dynamic interrupt moderation work task, if any */
2148 cancel_work_sync(&rq->dim.work);
2150 if (priv->params.rx_cq_moderation_mode >= 2) {
2151 struct net_dim_cq_moder curr;
2153 mlx5e_get_default_profile(priv, dim_mode, &curr);
2155 retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2156 curr.usec, curr.pkts, cq_mode);
2158 /* set dynamic interrupt moderation mode and zero defaults */
2160 rq->dim.mode = dim_mode;
2162 rq->dim.profile_ix = MLX5E_DIM_DEFAULT_PROFILE;
2163 mtx_unlock(&rq->mtx);
2165 retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2166 priv->params.rx_cq_moderation_usec,
2167 priv->params.rx_cq_moderation_pkts,
2173 return (mlx5_core_modify_cq_moderation(priv->mdev, &rq->cq.mcq,
2174 priv->params.rx_cq_moderation_usec,
2175 priv->params.rx_cq_moderation_pkts));
2179 mlx5e_refresh_channel_params_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
2184 err = mlx5e_refresh_rq_params(priv, &c->rq);
2188 for (i = 0; i != c->num_tc; i++) {
2189 err = mlx5e_refresh_sq_params(priv, &c->sq[i]);
2198 mlx5e_refresh_channel_params(struct mlx5e_priv *priv)
2202 /* check if channels are closed */
2203 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2206 for (i = 0; i < priv->params.num_channels; i++) {
2209 err = mlx5e_refresh_channel_params_sub(priv, &priv->channel[i]);
2217 mlx5e_open_tis(struct mlx5e_priv *priv, int tc)
2219 struct mlx5_core_dev *mdev = priv->mdev;
2220 u32 in[MLX5_ST_SZ_DW(create_tis_in)];
2221 void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
2223 memset(in, 0, sizeof(in));
2225 MLX5_SET(tisc, tisc, prio, tc);
2226 MLX5_SET(tisc, tisc, transport_domain, priv->tdn);
2228 return (mlx5_core_create_tis(mdev, in, sizeof(in), &priv->tisn[tc]));
2232 mlx5e_close_tis(struct mlx5e_priv *priv, int tc)
2234 mlx5_core_destroy_tis(priv->mdev, priv->tisn[tc]);
2238 mlx5e_open_tises(struct mlx5e_priv *priv)
2240 int num_tc = priv->num_tc;
2244 for (tc = 0; tc < num_tc; tc++) {
2245 err = mlx5e_open_tis(priv, tc);
2247 goto err_close_tises;
2253 for (tc--; tc >= 0; tc--)
2254 mlx5e_close_tis(priv, tc);
2260 mlx5e_close_tises(struct mlx5e_priv *priv)
2262 int num_tc = priv->num_tc;
2265 for (tc = 0; tc < num_tc; tc++)
2266 mlx5e_close_tis(priv, tc);
2270 mlx5e_open_rqt(struct mlx5e_priv *priv)
2272 struct mlx5_core_dev *mdev = priv->mdev;
2274 u32 out[MLX5_ST_SZ_DW(create_rqt_out)] = {0};
2281 sz = 1 << priv->params.rx_hash_log_tbl_sz;
2283 inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
2284 in = mlx5_vzalloc(inlen);
2287 rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
2289 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
2290 MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
2292 for (i = 0; i < sz; i++) {
2295 ix = rss_get_indirection_to_bucket(ix);
2297 /* ensure we don't overflow */
2298 ix %= priv->params.num_channels;
2300 /* apply receive side scaling stride, if any */
2301 ix -= ix % (int)priv->params.channels_rsss;
2303 MLX5_SET(rqtc, rqtc, rq_num[i], priv->channel[ix].rq.rqn);
2306 MLX5_SET(create_rqt_in, in, opcode, MLX5_CMD_OP_CREATE_RQT);
2308 err = mlx5_cmd_exec(mdev, in, inlen, out, sizeof(out));
2310 priv->rqtn = MLX5_GET(create_rqt_out, out, rqtn);
2318 mlx5e_close_rqt(struct mlx5e_priv *priv)
2320 u32 in[MLX5_ST_SZ_DW(destroy_rqt_in)] = {0};
2321 u32 out[MLX5_ST_SZ_DW(destroy_rqt_out)] = {0};
2323 MLX5_SET(destroy_rqt_in, in, opcode, MLX5_CMD_OP_DESTROY_RQT);
2324 MLX5_SET(destroy_rqt_in, in, rqtn, priv->rqtn);
2326 mlx5_cmd_exec(priv->mdev, in, sizeof(in), out, sizeof(out));
2330 mlx5e_build_tir_ctx(struct mlx5e_priv *priv, u32 * tirc, int tt)
2332 void *hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
2335 MLX5_SET(tirc, tirc, transport_domain, priv->tdn);
2337 #define ROUGH_MAX_L2_L3_HDR_SZ 256
2339 #define MLX5_HASH_IP (MLX5_HASH_FIELD_SEL_SRC_IP |\
2340 MLX5_HASH_FIELD_SEL_DST_IP)
2342 #define MLX5_HASH_ALL (MLX5_HASH_FIELD_SEL_SRC_IP |\
2343 MLX5_HASH_FIELD_SEL_DST_IP |\
2344 MLX5_HASH_FIELD_SEL_L4_SPORT |\
2345 MLX5_HASH_FIELD_SEL_L4_DPORT)
2347 #define MLX5_HASH_IP_IPSEC_SPI (MLX5_HASH_FIELD_SEL_SRC_IP |\
2348 MLX5_HASH_FIELD_SEL_DST_IP |\
2349 MLX5_HASH_FIELD_SEL_IPSEC_SPI)
2351 if (priv->params.hw_lro_en) {
2352 MLX5_SET(tirc, tirc, lro_enable_mask,
2353 MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO |
2354 MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO);
2355 MLX5_SET(tirc, tirc, lro_max_msg_sz,
2356 (priv->params.lro_wqe_sz -
2357 ROUGH_MAX_L2_L3_HDR_SZ) >> 8);
2358 /* TODO: add the option to choose timer value dynamically */
2359 MLX5_SET(tirc, tirc, lro_timeout_period_usecs,
2360 MLX5_CAP_ETH(priv->mdev,
2361 lro_timer_supported_periods[2]));
2364 /* setup parameters for hashing TIR type, if any */
2367 MLX5_SET(tirc, tirc, disp_type,
2368 MLX5_TIRC_DISP_TYPE_DIRECT);
2369 MLX5_SET(tirc, tirc, inline_rqn,
2370 priv->channel[0].rq.rqn);
2373 MLX5_SET(tirc, tirc, disp_type,
2374 MLX5_TIRC_DISP_TYPE_INDIRECT);
2375 MLX5_SET(tirc, tirc, indirect_table,
2377 MLX5_SET(tirc, tirc, rx_hash_fn,
2378 MLX5_TIRC_RX_HASH_FN_HASH_TOEPLITZ);
2379 hkey = (__be32 *) MLX5_ADDR_OF(tirc, tirc, rx_hash_toeplitz_key);
2382 * The FreeBSD RSS implementation does currently not
2383 * support symmetric Toeplitz hashes:
2385 MLX5_SET(tirc, tirc, rx_hash_symmetric, 0);
2386 rss_getkey((uint8_t *)hkey);
2388 MLX5_SET(tirc, tirc, rx_hash_symmetric, 1);
2389 hkey[0] = cpu_to_be32(0xD181C62C);
2390 hkey[1] = cpu_to_be32(0xF7F4DB5B);
2391 hkey[2] = cpu_to_be32(0x1983A2FC);
2392 hkey[3] = cpu_to_be32(0x943E1ADB);
2393 hkey[4] = cpu_to_be32(0xD9389E6B);
2394 hkey[5] = cpu_to_be32(0xD1039C2C);
2395 hkey[6] = cpu_to_be32(0xA74499AD);
2396 hkey[7] = cpu_to_be32(0x593D56D9);
2397 hkey[8] = cpu_to_be32(0xF3253C06);
2398 hkey[9] = cpu_to_be32(0x2ADC1FFC);
2404 case MLX5E_TT_IPV4_TCP:
2405 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2406 MLX5_L3_PROT_TYPE_IPV4);
2407 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2408 MLX5_L4_PROT_TYPE_TCP);
2410 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4)) {
2411 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2415 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2419 case MLX5E_TT_IPV6_TCP:
2420 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2421 MLX5_L3_PROT_TYPE_IPV6);
2422 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2423 MLX5_L4_PROT_TYPE_TCP);
2425 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) {
2426 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2430 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2434 case MLX5E_TT_IPV4_UDP:
2435 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2436 MLX5_L3_PROT_TYPE_IPV4);
2437 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2438 MLX5_L4_PROT_TYPE_UDP);
2440 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4)) {
2441 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2445 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2449 case MLX5E_TT_IPV6_UDP:
2450 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2451 MLX5_L3_PROT_TYPE_IPV6);
2452 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2453 MLX5_L4_PROT_TYPE_UDP);
2455 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) {
2456 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2460 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2464 case MLX5E_TT_IPV4_IPSEC_AH:
2465 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2466 MLX5_L3_PROT_TYPE_IPV4);
2467 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2468 MLX5_HASH_IP_IPSEC_SPI);
2471 case MLX5E_TT_IPV6_IPSEC_AH:
2472 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2473 MLX5_L3_PROT_TYPE_IPV6);
2474 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2475 MLX5_HASH_IP_IPSEC_SPI);
2478 case MLX5E_TT_IPV4_IPSEC_ESP:
2479 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2480 MLX5_L3_PROT_TYPE_IPV4);
2481 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2482 MLX5_HASH_IP_IPSEC_SPI);
2485 case MLX5E_TT_IPV6_IPSEC_ESP:
2486 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2487 MLX5_L3_PROT_TYPE_IPV6);
2488 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2489 MLX5_HASH_IP_IPSEC_SPI);
2493 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2494 MLX5_L3_PROT_TYPE_IPV4);
2495 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2500 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2501 MLX5_L3_PROT_TYPE_IPV6);
2502 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2512 mlx5e_open_tir(struct mlx5e_priv *priv, int tt)
2514 struct mlx5_core_dev *mdev = priv->mdev;
2520 inlen = MLX5_ST_SZ_BYTES(create_tir_in);
2521 in = mlx5_vzalloc(inlen);
2524 tirc = MLX5_ADDR_OF(create_tir_in, in, tir_context);
2526 mlx5e_build_tir_ctx(priv, tirc, tt);
2528 err = mlx5_core_create_tir(mdev, in, inlen, &priv->tirn[tt]);
2536 mlx5e_close_tir(struct mlx5e_priv *priv, int tt)
2538 mlx5_core_destroy_tir(priv->mdev, priv->tirn[tt]);
2542 mlx5e_open_tirs(struct mlx5e_priv *priv)
2547 for (i = 0; i < MLX5E_NUM_TT; i++) {
2548 err = mlx5e_open_tir(priv, i);
2550 goto err_close_tirs;
2556 for (i--; i >= 0; i--)
2557 mlx5e_close_tir(priv, i);
2563 mlx5e_close_tirs(struct mlx5e_priv *priv)
2567 for (i = 0; i < MLX5E_NUM_TT; i++)
2568 mlx5e_close_tir(priv, i);
2572 * SW MTU does not include headers,
2573 * HW MTU includes all headers and checksums.
2576 mlx5e_set_dev_port_mtu(struct ifnet *ifp, int sw_mtu)
2578 struct mlx5e_priv *priv = ifp->if_softc;
2579 struct mlx5_core_dev *mdev = priv->mdev;
2583 hw_mtu = MLX5E_SW2HW_MTU(sw_mtu);
2585 err = mlx5_set_port_mtu(mdev, hw_mtu);
2587 if_printf(ifp, "%s: mlx5_set_port_mtu failed setting %d, err=%d\n",
2588 __func__, sw_mtu, err);
2592 /* Update vport context MTU */
2593 err = mlx5_set_vport_mtu(mdev, hw_mtu);
2595 if_printf(ifp, "%s: Failed updating vport context with MTU size, err=%d\n",
2599 ifp->if_mtu = sw_mtu;
2601 err = mlx5_query_vport_mtu(mdev, &hw_mtu);
2602 if (err || !hw_mtu) {
2603 /* fallback to port oper mtu */
2604 err = mlx5_query_port_oper_mtu(mdev, &hw_mtu);
2607 if_printf(ifp, "Query port MTU, after setting new "
2608 "MTU value, failed\n");
2610 } else if (MLX5E_HW2SW_MTU(hw_mtu) < sw_mtu) {
2612 if_printf(ifp, "Port MTU %d is smaller than "
2613 "ifp mtu %d\n", hw_mtu, sw_mtu);
2614 } else if (MLX5E_HW2SW_MTU(hw_mtu) > sw_mtu) {
2616 if_printf(ifp, "Port MTU %d is bigger than "
2617 "ifp mtu %d\n", hw_mtu, sw_mtu);
2619 priv->params_ethtool.hw_mtu = hw_mtu;
2625 mlx5e_open_locked(struct ifnet *ifp)
2627 struct mlx5e_priv *priv = ifp->if_softc;
2631 /* check if already opened */
2632 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
2636 if (rss_getnumbuckets() > priv->params.num_channels) {
2637 if_printf(ifp, "NOTE: There are more RSS buckets(%u) than "
2638 "channels(%u) available\n", rss_getnumbuckets(),
2639 priv->params.num_channels);
2642 err = mlx5e_open_tises(priv);
2644 if_printf(ifp, "%s: mlx5e_open_tises failed, %d\n",
2648 err = mlx5_vport_alloc_q_counter(priv->mdev,
2649 MLX5_INTERFACE_PROTOCOL_ETH, &set_id);
2651 if_printf(priv->ifp,
2652 "%s: mlx5_vport_alloc_q_counter failed: %d\n",
2654 goto err_close_tises;
2656 /* store counter set ID */
2657 priv->counter_set_id = set_id;
2659 err = mlx5e_open_channels(priv);
2661 if_printf(ifp, "%s: mlx5e_open_channels failed, %d\n",
2663 goto err_dalloc_q_counter;
2665 err = mlx5e_open_rqt(priv);
2667 if_printf(ifp, "%s: mlx5e_open_rqt failed, %d\n",
2669 goto err_close_channels;
2671 err = mlx5e_open_tirs(priv);
2673 if_printf(ifp, "%s: mlx5e_open_tir failed, %d\n",
2675 goto err_close_rqls;
2677 err = mlx5e_open_flow_table(priv);
2679 if_printf(ifp, "%s: mlx5e_open_flow_table failed, %d\n",
2681 goto err_close_tirs;
2683 err = mlx5e_add_all_vlan_rules(priv);
2685 if_printf(ifp, "%s: mlx5e_add_all_vlan_rules failed, %d\n",
2687 goto err_close_flow_table;
2689 set_bit(MLX5E_STATE_OPENED, &priv->state);
2691 mlx5e_update_carrier(priv);
2692 mlx5e_set_rx_mode_core(priv);
2696 err_close_flow_table:
2697 mlx5e_close_flow_table(priv);
2700 mlx5e_close_tirs(priv);
2703 mlx5e_close_rqt(priv);
2706 mlx5e_close_channels(priv);
2708 err_dalloc_q_counter:
2709 mlx5_vport_dealloc_q_counter(priv->mdev,
2710 MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
2713 mlx5e_close_tises(priv);
2719 mlx5e_open(void *arg)
2721 struct mlx5e_priv *priv = arg;
2724 if (mlx5_set_port_status(priv->mdev, MLX5_PORT_UP))
2725 if_printf(priv->ifp,
2726 "%s: Setting port status to up failed\n",
2729 mlx5e_open_locked(priv->ifp);
2730 priv->ifp->if_drv_flags |= IFF_DRV_RUNNING;
2735 mlx5e_close_locked(struct ifnet *ifp)
2737 struct mlx5e_priv *priv = ifp->if_softc;
2739 /* check if already closed */
2740 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2743 clear_bit(MLX5E_STATE_OPENED, &priv->state);
2745 mlx5e_set_rx_mode_core(priv);
2746 mlx5e_del_all_vlan_rules(priv);
2747 if_link_state_change(priv->ifp, LINK_STATE_DOWN);
2748 mlx5e_close_flow_table(priv);
2749 mlx5e_close_tirs(priv);
2750 mlx5e_close_rqt(priv);
2751 mlx5e_close_channels(priv);
2752 mlx5_vport_dealloc_q_counter(priv->mdev,
2753 MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
2754 mlx5e_close_tises(priv);
2759 #if (__FreeBSD_version >= 1100000)
2761 mlx5e_get_counter(struct ifnet *ifp, ift_counter cnt)
2763 struct mlx5e_priv *priv = ifp->if_softc;
2766 /* PRIV_LOCK(priv); XXX not allowed */
2768 case IFCOUNTER_IPACKETS:
2769 retval = priv->stats.vport.rx_packets;
2771 case IFCOUNTER_IERRORS:
2772 retval = priv->stats.vport.rx_error_packets +
2773 priv->stats.pport.alignment_err +
2774 priv->stats.pport.check_seq_err +
2775 priv->stats.pport.crc_align_errors +
2776 priv->stats.pport.in_range_len_errors +
2777 priv->stats.pport.jabbers +
2778 priv->stats.pport.out_of_range_len +
2779 priv->stats.pport.oversize_pkts +
2780 priv->stats.pport.symbol_err +
2781 priv->stats.pport.too_long_errors +
2782 priv->stats.pport.undersize_pkts +
2783 priv->stats.pport.unsupported_op_rx;
2785 case IFCOUNTER_IQDROPS:
2786 retval = priv->stats.vport.rx_out_of_buffer +
2787 priv->stats.pport.drop_events;
2789 case IFCOUNTER_OPACKETS:
2790 retval = priv->stats.vport.tx_packets;
2792 case IFCOUNTER_OERRORS:
2793 retval = priv->stats.vport.tx_error_packets;
2795 case IFCOUNTER_IBYTES:
2796 retval = priv->stats.vport.rx_bytes;
2798 case IFCOUNTER_OBYTES:
2799 retval = priv->stats.vport.tx_bytes;
2801 case IFCOUNTER_IMCASTS:
2802 retval = priv->stats.vport.rx_multicast_packets;
2804 case IFCOUNTER_OMCASTS:
2805 retval = priv->stats.vport.tx_multicast_packets;
2807 case IFCOUNTER_OQDROPS:
2808 retval = priv->stats.vport.tx_queue_dropped;
2810 case IFCOUNTER_COLLISIONS:
2811 retval = priv->stats.pport.collisions;
2814 retval = if_get_counter_default(ifp, cnt);
2817 /* PRIV_UNLOCK(priv); XXX not allowed */
2823 mlx5e_set_rx_mode(struct ifnet *ifp)
2825 struct mlx5e_priv *priv = ifp->if_softc;
2827 queue_work(priv->wq, &priv->set_rx_mode_work);
2831 mlx5e_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2833 struct mlx5e_priv *priv;
2835 struct ifi2creq i2c;
2844 priv = ifp->if_softc;
2846 /* check if detaching */
2847 if (priv == NULL || priv->gone != 0)
2852 ifr = (struct ifreq *)data;
2855 mlx5_query_port_max_mtu(priv->mdev, &max_mtu);
2857 if (ifr->ifr_mtu >= MLX5E_MTU_MIN &&
2858 ifr->ifr_mtu <= MIN(MLX5E_MTU_MAX, max_mtu)) {
2861 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2863 mlx5e_close_locked(ifp);
2866 mlx5e_set_dev_port_mtu(ifp, ifr->ifr_mtu);
2869 mlx5e_open_locked(ifp);
2872 if_printf(ifp, "Invalid MTU value. Min val: %d, Max val: %d\n",
2873 MLX5E_MTU_MIN, MIN(MLX5E_MTU_MAX, max_mtu));
2878 if ((ifp->if_flags & IFF_UP) &&
2879 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
2880 mlx5e_set_rx_mode(ifp);
2884 if (ifp->if_flags & IFF_UP) {
2885 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2886 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2887 mlx5e_open_locked(ifp);
2888 ifp->if_drv_flags |= IFF_DRV_RUNNING;
2889 mlx5_set_port_status(priv->mdev, MLX5_PORT_UP);
2892 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
2893 mlx5_set_port_status(priv->mdev,
2895 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
2896 mlx5e_close_locked(ifp);
2897 mlx5e_update_carrier(priv);
2898 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2905 mlx5e_set_rx_mode(ifp);
2910 ifr = (struct ifreq *)data;
2911 error = ifmedia_ioctl(ifp, ifr, &priv->media, command);
2914 ifr = (struct ifreq *)data;
2916 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2918 if (mask & IFCAP_TXCSUM) {
2919 ifp->if_capenable ^= IFCAP_TXCSUM;
2920 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
2922 if (IFCAP_TSO4 & ifp->if_capenable &&
2923 !(IFCAP_TXCSUM & ifp->if_capenable)) {
2924 ifp->if_capenable &= ~IFCAP_TSO4;
2925 ifp->if_hwassist &= ~CSUM_IP_TSO;
2927 "tso4 disabled due to -txcsum.\n");
2930 if (mask & IFCAP_TXCSUM_IPV6) {
2931 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
2932 ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
2934 if (IFCAP_TSO6 & ifp->if_capenable &&
2935 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2936 ifp->if_capenable &= ~IFCAP_TSO6;
2937 ifp->if_hwassist &= ~CSUM_IP6_TSO;
2939 "tso6 disabled due to -txcsum6.\n");
2942 if (mask & IFCAP_RXCSUM)
2943 ifp->if_capenable ^= IFCAP_RXCSUM;
2944 if (mask & IFCAP_RXCSUM_IPV6)
2945 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
2946 if (mask & IFCAP_TSO4) {
2947 if (!(IFCAP_TSO4 & ifp->if_capenable) &&
2948 !(IFCAP_TXCSUM & ifp->if_capenable)) {
2949 if_printf(ifp, "enable txcsum first.\n");
2953 ifp->if_capenable ^= IFCAP_TSO4;
2954 ifp->if_hwassist ^= CSUM_IP_TSO;
2956 if (mask & IFCAP_TSO6) {
2957 if (!(IFCAP_TSO6 & ifp->if_capenable) &&
2958 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2959 if_printf(ifp, "enable txcsum6 first.\n");
2963 ifp->if_capenable ^= IFCAP_TSO6;
2964 ifp->if_hwassist ^= CSUM_IP6_TSO;
2966 if (mask & IFCAP_VLAN_HWFILTER) {
2967 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2968 mlx5e_disable_vlan_filter(priv);
2970 mlx5e_enable_vlan_filter(priv);
2972 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
2974 if (mask & IFCAP_VLAN_HWTAGGING)
2975 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2976 if (mask & IFCAP_WOL_MAGIC)
2977 ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2979 VLAN_CAPABILITIES(ifp);
2980 /* turn off LRO means also turn of HW LRO - if it's on */
2981 if (mask & IFCAP_LRO) {
2982 int was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2983 bool need_restart = false;
2985 ifp->if_capenable ^= IFCAP_LRO;
2987 /* figure out if updating HW LRO is needed */
2988 if (!(ifp->if_capenable & IFCAP_LRO)) {
2989 if (priv->params.hw_lro_en) {
2990 priv->params.hw_lro_en = false;
2991 need_restart = true;
2994 if (priv->params.hw_lro_en == false &&
2995 priv->params_ethtool.hw_lro != 0) {
2996 priv->params.hw_lro_en = true;
2997 need_restart = true;
3000 if (was_opened && need_restart) {
3001 mlx5e_close_locked(ifp);
3002 mlx5e_open_locked(ifp);
3005 if (mask & IFCAP_HWRXTSTMP) {
3006 ifp->if_capenable ^= IFCAP_HWRXTSTMP;
3007 if (ifp->if_capenable & IFCAP_HWRXTSTMP) {
3008 if (priv->clbr_done == 0)
3009 mlx5e_reset_calibration_callout(priv);
3011 callout_drain(&priv->tstmp_clbr);
3012 priv->clbr_done = 0;
3020 ifr = (struct ifreq *)data;
3023 * Copy from the user-space address ifr_data to the
3024 * kernel-space address i2c
3026 error = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
3030 if (i2c.len > sizeof(i2c.data)) {
3036 /* Get module_num which is required for the query_eeprom */
3037 error = mlx5_query_module_num(priv->mdev, &module_num);
3039 if_printf(ifp, "Query module num failed, eeprom "
3040 "reading is not supported\n");
3044 /* Check if module is present before doing an access */
3045 module_status = mlx5_query_module_status(priv->mdev, module_num);
3046 if (module_status != MLX5_MODULE_STATUS_PLUGGED_ENABLED &&
3047 module_status != MLX5_MODULE_STATUS_PLUGGED_DISABLED) {
3052 * Currently 0XA0 and 0xA2 are the only addresses permitted.
3053 * The internal conversion is as follows:
3055 if (i2c.dev_addr == 0xA0)
3056 read_addr = MLX5E_I2C_ADDR_LOW;
3057 else if (i2c.dev_addr == 0xA2)
3058 read_addr = MLX5E_I2C_ADDR_HIGH;
3060 if_printf(ifp, "Query eeprom failed, "
3061 "Invalid Address: %X\n", i2c.dev_addr);
3065 error = mlx5_query_eeprom(priv->mdev,
3066 read_addr, MLX5E_EEPROM_LOW_PAGE,
3067 (uint32_t)i2c.offset, (uint32_t)i2c.len, module_num,
3068 (uint32_t *)i2c.data, &size_read);
3070 if_printf(ifp, "Query eeprom failed, eeprom "
3071 "reading is not supported\n");
3076 if (i2c.len > MLX5_EEPROM_MAX_BYTES) {
3077 error = mlx5_query_eeprom(priv->mdev,
3078 read_addr, MLX5E_EEPROM_LOW_PAGE,
3079 (uint32_t)(i2c.offset + size_read),
3080 (uint32_t)(i2c.len - size_read), module_num,
3081 (uint32_t *)(i2c.data + size_read), &size_read);
3084 if_printf(ifp, "Query eeprom failed, eeprom "
3085 "reading is not supported\n");
3090 error = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
3096 error = ether_ioctl(ifp, command, data);
3103 mlx5e_check_required_hca_cap(struct mlx5_core_dev *mdev)
3106 * TODO: uncoment once FW really sets all these bits if
3107 * (!mdev->caps.eth.rss_ind_tbl_cap || !mdev->caps.eth.csum_cap ||
3108 * !mdev->caps.eth.max_lso_cap || !mdev->caps.eth.vlan_cap ||
3109 * !(mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_SCQE_BRK_MOD)) return
3113 /* TODO: add more must-to-have features */
3115 if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
3122 mlx5e_get_max_inline_cap(struct mlx5_core_dev *mdev)
3124 uint32_t bf_buf_size = (1U << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2U;
3126 bf_buf_size -= sizeof(struct mlx5e_tx_wqe) - 2;
3128 /* verify against driver hardware limit */
3129 if (bf_buf_size > MLX5E_MAX_TX_INLINE)
3130 bf_buf_size = MLX5E_MAX_TX_INLINE;
3132 return (bf_buf_size);
3136 mlx5e_build_ifp_priv(struct mlx5_core_dev *mdev,
3137 struct mlx5e_priv *priv,
3138 int num_comp_vectors)
3143 * TODO: Consider link speed for setting "log_sq_size",
3144 * "log_rq_size" and "cq_moderation_xxx":
3146 priv->params.log_sq_size =
3147 MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
3148 priv->params.log_rq_size =
3149 MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
3150 priv->params.rx_cq_moderation_usec =
3151 MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
3152 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE :
3153 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
3154 priv->params.rx_cq_moderation_mode =
3155 MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ? 1 : 0;
3156 priv->params.rx_cq_moderation_pkts =
3157 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS;
3158 priv->params.tx_cq_moderation_usec =
3159 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC;
3160 priv->params.tx_cq_moderation_pkts =
3161 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
3162 priv->params.min_rx_wqes =
3163 MLX5E_PARAMS_DEFAULT_MIN_RX_WQES;
3164 priv->params.rx_hash_log_tbl_sz =
3165 (order_base_2(num_comp_vectors) >
3166 MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ) ?
3167 order_base_2(num_comp_vectors) :
3168 MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ;
3169 priv->params.num_tc = 1;
3170 priv->params.default_vlan_prio = 0;
3171 priv->counter_set_id = -1;
3172 priv->params.tx_max_inline = mlx5e_get_max_inline_cap(mdev);
3174 err = mlx5_query_min_inline(mdev, &priv->params.tx_min_inline_mode);
3179 * hw lro is currently defaulted to off. when it won't anymore we
3180 * will consider the HW capability: "!!MLX5_CAP_ETH(mdev, lro_cap)"
3182 priv->params.hw_lro_en = false;
3183 priv->params.lro_wqe_sz = MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ;
3186 * CQE zipping is currently defaulted to off. when it won't
3187 * anymore we will consider the HW capability:
3188 * "!!MLX5_CAP_GEN(mdev, cqe_compression)"
3190 priv->params.cqe_zipping_en = false;
3193 priv->params.num_channels = num_comp_vectors;
3194 priv->params.channels_rsss = 1;
3195 priv->order_base_2_num_channels = order_base_2(num_comp_vectors);
3196 priv->queue_mapping_channel_mask =
3197 roundup_pow_of_two(num_comp_vectors) - 1;
3198 priv->num_tc = priv->params.num_tc;
3199 priv->default_vlan_prio = priv->params.default_vlan_prio;
3201 INIT_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
3202 INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
3203 INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
3209 mlx5e_create_mkey(struct mlx5e_priv *priv, u32 pdn,
3210 struct mlx5_core_mr *mkey)
3212 struct ifnet *ifp = priv->ifp;
3213 struct mlx5_core_dev *mdev = priv->mdev;
3214 int inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
3219 in = mlx5_vzalloc(inlen);
3221 if_printf(ifp, "%s: failed to allocate inbox\n", __func__);
3225 mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
3226 MLX5_SET(mkc, mkc, access_mode, MLX5_ACCESS_MODE_PA);
3227 MLX5_SET(mkc, mkc, lw, 1);
3228 MLX5_SET(mkc, mkc, lr, 1);
3230 MLX5_SET(mkc, mkc, pd, pdn);
3231 MLX5_SET(mkc, mkc, length64, 1);
3232 MLX5_SET(mkc, mkc, qpn, 0xffffff);
3234 err = mlx5_core_create_mkey(mdev, mkey, in, inlen);
3236 if_printf(ifp, "%s: mlx5_core_create_mkey failed, %d\n",
3243 static const char *mlx5e_vport_stats_desc[] = {
3244 MLX5E_VPORT_STATS(MLX5E_STATS_DESC)
3247 static const char *mlx5e_pport_stats_desc[] = {
3248 MLX5E_PPORT_STATS(MLX5E_STATS_DESC)
3252 mlx5e_priv_mtx_init(struct mlx5e_priv *priv)
3254 mtx_init(&priv->async_events_mtx, "mlx5async", MTX_NETWORK_LOCK, MTX_DEF);
3255 sx_init(&priv->state_lock, "mlx5state");
3256 callout_init_mtx(&priv->watchdog, &priv->async_events_mtx, 0);
3257 MLX5_INIT_DOORBELL_LOCK(&priv->doorbell_lock);
3261 mlx5e_priv_mtx_destroy(struct mlx5e_priv *priv)
3263 mtx_destroy(&priv->async_events_mtx);
3264 sx_destroy(&priv->state_lock);
3268 sysctl_firmware(SYSCTL_HANDLER_ARGS)
3271 * %d.%d%.d the string format.
3272 * fw_rev_{maj,min,sub} return u16, 2^16 = 65536.
3273 * We need at most 5 chars to store that.
3274 * It also has: two "." and NULL at the end, which means we need 18
3275 * (5*3 + 3) chars at most.
3278 struct mlx5e_priv *priv = arg1;
3281 snprintf(fw, sizeof(fw), "%d.%d.%d", fw_rev_maj(priv->mdev), fw_rev_min(priv->mdev),
3282 fw_rev_sub(priv->mdev));
3283 error = sysctl_handle_string(oidp, fw, sizeof(fw), req);
3288 mlx5e_disable_tx_dma(struct mlx5e_channel *ch)
3292 for (i = 0; i < ch->num_tc; i++)
3293 mlx5e_drain_sq(&ch->sq[i]);
3297 mlx5e_reset_sq_doorbell_record(struct mlx5e_sq *sq)
3300 sq->doorbell.d32[0] = cpu_to_be32(MLX5_OPCODE_NOP);
3301 sq->doorbell.d32[1] = cpu_to_be32(sq->sqn << 8);
3302 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
3303 sq->doorbell.d64 = 0;
3307 mlx5e_resume_sq(struct mlx5e_sq *sq)
3311 /* check if already enabled */
3312 if (READ_ONCE(sq->running) != 0)
3315 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_ERR,
3316 MLX5_SQC_STATE_RST);
3319 "mlx5e_modify_sq() from ERR to RST failed: %d\n", err);
3325 /* reset doorbell prior to moving from RST to RDY */
3326 mlx5e_reset_sq_doorbell_record(sq);
3328 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST,
3329 MLX5_SQC_STATE_RDY);
3332 "mlx5e_modify_sq() from RST to RDY failed: %d\n", err);
3335 sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
3336 WRITE_ONCE(sq->running, 1);
3340 mlx5e_enable_tx_dma(struct mlx5e_channel *ch)
3344 for (i = 0; i < ch->num_tc; i++)
3345 mlx5e_resume_sq(&ch->sq[i]);
3349 mlx5e_disable_rx_dma(struct mlx5e_channel *ch)
3351 struct mlx5e_rq *rq = &ch->rq;
3356 callout_stop(&rq->watchdog);
3357 mtx_unlock(&rq->mtx);
3359 callout_drain(&rq->watchdog);
3361 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
3364 "mlx5e_modify_rq() from RDY to RST failed: %d\n", err);
3367 while (!mlx5_wq_ll_is_empty(&rq->wq)) {
3369 rq->cq.mcq.comp(&rq->cq.mcq);
3373 * Transitioning into RST state will allow the FW to track less ERR state queues,
3374 * thus reducing the recv queue flushing time
3376 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_ERR, MLX5_RQC_STATE_RST);
3379 "mlx5e_modify_rq() from ERR to RST failed: %d\n", err);
3384 mlx5e_enable_rx_dma(struct mlx5e_channel *ch)
3386 struct mlx5e_rq *rq = &ch->rq;
3390 mlx5_wq_ll_update_db_record(&rq->wq);
3391 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
3394 "mlx5e_modify_rq() from RST to RDY failed: %d\n", err);
3399 rq->cq.mcq.comp(&rq->cq.mcq);
3403 mlx5e_modify_tx_dma(struct mlx5e_priv *priv, uint8_t value)
3407 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3410 for (i = 0; i < priv->params.num_channels; i++) {
3412 mlx5e_disable_tx_dma(&priv->channel[i]);
3414 mlx5e_enable_tx_dma(&priv->channel[i]);
3419 mlx5e_modify_rx_dma(struct mlx5e_priv *priv, uint8_t value)
3423 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3426 for (i = 0; i < priv->params.num_channels; i++) {
3428 mlx5e_disable_rx_dma(&priv->channel[i]);
3430 mlx5e_enable_rx_dma(&priv->channel[i]);
3435 mlx5e_add_hw_stats(struct mlx5e_priv *priv)
3437 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3438 OID_AUTO, "fw_version", CTLTYPE_STRING | CTLFLAG_RD, priv, 0,
3439 sysctl_firmware, "A", "HCA firmware version");
3441 SYSCTL_ADD_STRING(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3442 OID_AUTO, "board_id", CTLFLAG_RD, priv->mdev->board_id, 0,
3447 mlx5e_sysctl_tx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3449 struct mlx5e_priv *priv = arg1;
3450 uint8_t temp[MLX5E_MAX_PRIORITY];
3457 tx_pfc = priv->params.tx_priority_flow_control;
3459 for (i = 0; i != MLX5E_MAX_PRIORITY; i++)
3460 temp[i] = (tx_pfc >> i) & 1;
3462 err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY);
3463 if (err || !req->newptr)
3465 err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY);
3469 priv->params.tx_priority_flow_control = 0;
3471 /* range check input value */
3472 for (i = 0; i != MLX5E_MAX_PRIORITY; i++) {
3477 priv->params.tx_priority_flow_control |= (temp[i] << i);
3480 /* check if update is required */
3481 if (tx_pfc != priv->params.tx_priority_flow_control)
3482 err = -mlx5e_set_port_pfc(priv);
3485 priv->params.tx_priority_flow_control= tx_pfc;
3492 mlx5e_sysctl_rx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3494 struct mlx5e_priv *priv = arg1;
3495 uint8_t temp[MLX5E_MAX_PRIORITY];
3502 rx_pfc = priv->params.rx_priority_flow_control;
3504 for (i = 0; i != MLX5E_MAX_PRIORITY; i++)
3505 temp[i] = (rx_pfc >> i) & 1;
3507 err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY);
3508 if (err || !req->newptr)
3510 err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY);
3514 priv->params.rx_priority_flow_control = 0;
3516 /* range check input value */
3517 for (i = 0; i != MLX5E_MAX_PRIORITY; i++) {
3522 priv->params.rx_priority_flow_control |= (temp[i] << i);
3525 /* check if update is required */
3526 if (rx_pfc != priv->params.rx_priority_flow_control)
3527 err = -mlx5e_set_port_pfc(priv);
3530 priv->params.rx_priority_flow_control= rx_pfc;
3537 mlx5e_setup_pauseframes(struct mlx5e_priv *priv)
3539 #if (__FreeBSD_version < 1100000)
3544 /* enable pauseframes by default */
3545 priv->params.tx_pauseframe_control = 1;
3546 priv->params.rx_pauseframe_control = 1;
3548 /* disable ports flow control, PFC, by default */
3549 priv->params.tx_priority_flow_control = 0;
3550 priv->params.rx_priority_flow_control = 0;
3552 #if (__FreeBSD_version < 1100000)
3553 /* compute path for sysctl */
3554 snprintf(path, sizeof(path), "dev.mce.%d.tx_pauseframe_control",
3555 device_get_unit(priv->mdev->pdev->dev.bsddev));
3557 /* try to fetch tunable, if any */
3558 TUNABLE_INT_FETCH(path, &priv->params.tx_pauseframe_control);
3560 /* compute path for sysctl */
3561 snprintf(path, sizeof(path), "dev.mce.%d.rx_pauseframe_control",
3562 device_get_unit(priv->mdev->pdev->dev.bsddev));
3564 /* try to fetch tunable, if any */
3565 TUNABLE_INT_FETCH(path, &priv->params.rx_pauseframe_control);
3568 /* register pauseframe SYSCTLs */
3569 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3570 OID_AUTO, "tx_pauseframe_control", CTLFLAG_RDTUN,
3571 &priv->params.tx_pauseframe_control, 0,
3572 "Set to enable TX pause frames. Clear to disable.");
3574 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3575 OID_AUTO, "rx_pauseframe_control", CTLFLAG_RDTUN,
3576 &priv->params.rx_pauseframe_control, 0,
3577 "Set to enable RX pause frames. Clear to disable.");
3579 /* register priority flow control, PFC, SYSCTLs */
3580 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3581 OID_AUTO, "tx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN |
3582 CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_tx_priority_flow_control, "CU",
3583 "Set to enable TX ports flow control frames for priorities 0..7. Clear to disable.");
3585 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3586 OID_AUTO, "rx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN |
3587 CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_rx_priority_flow_control, "CU",
3588 "Set to enable RX ports flow control frames for priorities 0..7. Clear to disable.");
3593 priv->params.tx_pauseframe_control =
3594 priv->params.tx_pauseframe_control ? 1 : 0;
3595 priv->params.rx_pauseframe_control =
3596 priv->params.rx_pauseframe_control ? 1 : 0;
3598 /* update firmware */
3599 error = mlx5e_set_port_pause_and_pfc(priv);
3600 if (error == -EINVAL) {
3601 if_printf(priv->ifp,
3602 "Global pauseframes must be disabled before enabling PFC.\n");
3603 priv->params.rx_priority_flow_control = 0;
3604 priv->params.tx_priority_flow_control = 0;
3606 /* update firmware */
3607 (void) mlx5e_set_port_pause_and_pfc(priv);
3613 mlx5e_ul_snd_tag_alloc(struct ifnet *ifp,
3614 union if_snd_tag_alloc_params *params,
3615 struct m_snd_tag **ppmt)
3617 struct mlx5e_priv *priv;
3618 struct mlx5e_channel *pch;
3620 priv = ifp->if_softc;
3622 if (unlikely(priv->gone || params->hdr.flowtype == M_HASHTYPE_NONE)) {
3623 return (EOPNOTSUPP);
3625 /* keep this code synced with mlx5e_select_queue() */
3626 u32 ch = priv->params.num_channels;
3630 if (rss_hash2bucket(params->hdr.flowid,
3631 params->hdr.flowtype, &temp) == 0)
3635 ch = (params->hdr.flowid % 128) % ch;
3638 * NOTE: The channels array is only freed at detach
3639 * and it safe to return a pointer to the send tag
3640 * inside the channels structure as long as we
3641 * reference the priv.
3643 pch = priv->channel + ch;
3645 /* check if send queue is not running */
3646 if (unlikely(pch->sq[0].running == 0))
3648 mlx5e_ref_channel(priv);
3649 *ppmt = &pch->tag.m_snd_tag;
3655 mlx5e_ul_snd_tag_query(struct m_snd_tag *pmt, union if_snd_tag_query_params *params)
3657 struct mlx5e_channel *pch =
3658 container_of(pmt, struct mlx5e_channel, tag.m_snd_tag);
3660 params->unlimited.max_rate = -1ULL;
3661 params->unlimited.queue_level = mlx5e_sq_queue_level(&pch->sq[0]);
3666 mlx5e_ul_snd_tag_free(struct m_snd_tag *pmt)
3668 struct mlx5e_channel *pch =
3669 container_of(pmt, struct mlx5e_channel, tag.m_snd_tag);
3671 mlx5e_unref_channel(pch->priv);
3675 mlx5e_snd_tag_alloc(struct ifnet *ifp,
3676 union if_snd_tag_alloc_params *params,
3677 struct m_snd_tag **ppmt)
3680 switch (params->hdr.type) {
3682 case IF_SND_TAG_TYPE_RATE_LIMIT:
3683 return (mlx5e_rl_snd_tag_alloc(ifp, params, ppmt));
3685 case IF_SND_TAG_TYPE_UNLIMITED:
3686 return (mlx5e_ul_snd_tag_alloc(ifp, params, ppmt));
3688 return (EOPNOTSUPP);
3693 mlx5e_snd_tag_modify(struct m_snd_tag *pmt, union if_snd_tag_modify_params *params)
3695 struct mlx5e_snd_tag *tag =
3696 container_of(pmt, struct mlx5e_snd_tag, m_snd_tag);
3698 switch (tag->type) {
3700 case IF_SND_TAG_TYPE_RATE_LIMIT:
3701 return (mlx5e_rl_snd_tag_modify(pmt, params));
3703 case IF_SND_TAG_TYPE_UNLIMITED:
3705 return (EOPNOTSUPP);
3710 mlx5e_snd_tag_query(struct m_snd_tag *pmt, union if_snd_tag_query_params *params)
3712 struct mlx5e_snd_tag *tag =
3713 container_of(pmt, struct mlx5e_snd_tag, m_snd_tag);
3715 switch (tag->type) {
3717 case IF_SND_TAG_TYPE_RATE_LIMIT:
3718 return (mlx5e_rl_snd_tag_query(pmt, params));
3720 case IF_SND_TAG_TYPE_UNLIMITED:
3721 return (mlx5e_ul_snd_tag_query(pmt, params));
3723 return (EOPNOTSUPP);
3728 mlx5e_snd_tag_free(struct m_snd_tag *pmt)
3730 struct mlx5e_snd_tag *tag =
3731 container_of(pmt, struct mlx5e_snd_tag, m_snd_tag);
3733 switch (tag->type) {
3735 case IF_SND_TAG_TYPE_RATE_LIMIT:
3736 mlx5e_rl_snd_tag_free(pmt);
3739 case IF_SND_TAG_TYPE_UNLIMITED:
3740 mlx5e_ul_snd_tag_free(pmt);
3748 mlx5e_create_ifp(struct mlx5_core_dev *mdev)
3751 struct mlx5e_priv *priv;
3752 u8 dev_addr[ETHER_ADDR_LEN] __aligned(4);
3753 struct sysctl_oid_list *child;
3754 int ncv = mdev->priv.eq_table.num_comp_vectors;
3760 if (mlx5e_check_required_hca_cap(mdev)) {
3761 mlx5_core_dbg(mdev, "mlx5e_check_required_hca_cap() failed\n");
3765 * Try to allocate the priv and make room for worst-case
3766 * number of channel structures:
3768 priv = malloc(sizeof(*priv) +
3769 (sizeof(priv->channel[0]) * mdev->priv.eq_table.num_comp_vectors),
3770 M_MLX5EN, M_WAITOK | M_ZERO);
3771 mlx5e_priv_mtx_init(priv);
3773 ifp = priv->ifp = if_alloc(IFT_ETHER);
3775 mlx5_core_err(mdev, "if_alloc() failed\n");
3778 ifp->if_softc = priv;
3779 if_initname(ifp, "mce", device_get_unit(mdev->pdev->dev.bsddev));
3780 ifp->if_mtu = ETHERMTU;
3781 ifp->if_init = mlx5e_open;
3782 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
3783 ifp->if_ioctl = mlx5e_ioctl;
3784 ifp->if_transmit = mlx5e_xmit;
3785 ifp->if_qflush = if_qflush;
3786 #if (__FreeBSD_version >= 1100000)
3787 ifp->if_get_counter = mlx5e_get_counter;
3789 ifp->if_snd.ifq_maxlen = ifqmaxlen;
3791 * Set driver features
3793 ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6;
3794 ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
3795 ifp->if_capabilities |= IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWFILTER;
3796 ifp->if_capabilities |= IFCAP_LINKSTATE | IFCAP_JUMBO_MTU;
3797 ifp->if_capabilities |= IFCAP_LRO;
3798 ifp->if_capabilities |= IFCAP_TSO | IFCAP_VLAN_HWTSO;
3799 ifp->if_capabilities |= IFCAP_HWSTATS | IFCAP_HWRXTSTMP;
3800 ifp->if_capabilities |= IFCAP_TXRTLMT;
3801 ifp->if_snd_tag_alloc = mlx5e_snd_tag_alloc;
3802 ifp->if_snd_tag_free = mlx5e_snd_tag_free;
3803 ifp->if_snd_tag_modify = mlx5e_snd_tag_modify;
3804 ifp->if_snd_tag_query = mlx5e_snd_tag_query;
3806 /* set TSO limits so that we don't have to drop TX packets */
3807 ifp->if_hw_tsomax = MLX5E_MAX_TX_PAYLOAD_SIZE - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
3808 ifp->if_hw_tsomaxsegcount = MLX5E_MAX_TX_MBUF_FRAGS - 1 /* hdr */;
3809 ifp->if_hw_tsomaxsegsize = MLX5E_MAX_TX_MBUF_SIZE;
3811 ifp->if_capenable = ifp->if_capabilities;
3812 ifp->if_hwassist = 0;
3813 if (ifp->if_capenable & IFCAP_TSO)
3814 ifp->if_hwassist |= CSUM_TSO;
3815 if (ifp->if_capenable & IFCAP_TXCSUM)
3816 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP | CSUM_IP);
3817 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3818 ifp->if_hwassist |= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
3820 /* ifnet sysctl tree */
3821 sysctl_ctx_init(&priv->sysctl_ctx);
3822 priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_dev),
3823 OID_AUTO, ifp->if_dname, CTLFLAG_RD, 0, "MLX5 ethernet - interface name");
3824 if (priv->sysctl_ifnet == NULL) {
3825 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3826 goto err_free_sysctl;
3828 snprintf(unit, sizeof(unit), "%d", ifp->if_dunit);
3829 priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3830 OID_AUTO, unit, CTLFLAG_RD, 0, "MLX5 ethernet - interface unit");
3831 if (priv->sysctl_ifnet == NULL) {
3832 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3833 goto err_free_sysctl;
3836 /* HW sysctl tree */
3837 child = SYSCTL_CHILDREN(device_get_sysctl_tree(mdev->pdev->dev.bsddev));
3838 priv->sysctl_hw = SYSCTL_ADD_NODE(&priv->sysctl_ctx, child,
3839 OID_AUTO, "hw", CTLFLAG_RD, 0, "MLX5 ethernet dev hw");
3840 if (priv->sysctl_hw == NULL) {
3841 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3842 goto err_free_sysctl;
3845 err = mlx5e_build_ifp_priv(mdev, priv, ncv);
3847 mlx5_core_err(mdev, "mlx5e_build_ifp_priv() failed (%d)\n", err);
3848 goto err_free_sysctl;
3851 snprintf(unit, sizeof(unit), "mce%u_wq",
3852 device_get_unit(mdev->pdev->dev.bsddev));
3853 priv->wq = alloc_workqueue(unit, 0, 1);
3854 if (priv->wq == NULL) {
3855 if_printf(ifp, "%s: alloc_workqueue failed\n", __func__);
3856 goto err_free_sysctl;
3859 err = mlx5_alloc_map_uar(mdev, &priv->cq_uar);
3861 if_printf(ifp, "%s: mlx5_alloc_map_uar failed, %d\n",
3865 err = mlx5_core_alloc_pd(mdev, &priv->pdn);
3867 if_printf(ifp, "%s: mlx5_core_alloc_pd failed, %d\n",
3869 goto err_unmap_free_uar;
3871 err = mlx5_alloc_transport_domain(mdev, &priv->tdn);
3873 if_printf(ifp, "%s: mlx5_alloc_transport_domain failed, %d\n",
3875 goto err_dealloc_pd;
3877 err = mlx5e_create_mkey(priv, priv->pdn, &priv->mr);
3879 if_printf(ifp, "%s: mlx5e_create_mkey failed, %d\n",
3881 goto err_dealloc_transport_domain;
3883 mlx5_query_nic_vport_mac_address(priv->mdev, 0, dev_addr);
3885 /* check if we should generate a random MAC address */
3886 if (MLX5_CAP_GEN(priv->mdev, vport_group_manager) == 0 &&
3887 is_zero_ether_addr(dev_addr)) {
3888 random_ether_addr(dev_addr);
3889 if_printf(ifp, "Assigned random MAC address\n");
3892 err = mlx5e_rl_init(priv);
3894 if_printf(ifp, "%s: mlx5e_rl_init failed, %d\n",
3896 goto err_create_mkey;
3900 /* set default MTU */
3901 mlx5e_set_dev_port_mtu(ifp, ifp->if_mtu);
3903 /* Set default media status */
3904 priv->media_status_last = IFM_AVALID;
3905 priv->media_active_last = IFM_ETHER | IFM_AUTO |
3906 IFM_ETH_RXPAUSE | IFM_FDX;
3908 /* setup default pauseframes configuration */
3909 mlx5e_setup_pauseframes(priv);
3911 err = mlx5_query_port_proto_cap(mdev, ð_proto_cap, MLX5_PTYS_EN);
3914 if_printf(ifp, "%s: Query port media capability failed, %d\n",
3918 /* Setup supported medias */
3919 ifmedia_init(&priv->media, IFM_IMASK | IFM_ETH_FMASK,
3920 mlx5e_media_change, mlx5e_media_status);
3922 for (i = 0; i < MLX5E_LINK_MODES_NUMBER; ++i) {
3923 if (mlx5e_mode_table[i].baudrate == 0)
3925 if (MLX5E_PROT_MASK(i) & eth_proto_cap) {
3926 ifmedia_add(&priv->media,
3927 mlx5e_mode_table[i].subtype |
3928 IFM_ETHER, 0, NULL);
3929 ifmedia_add(&priv->media,
3930 mlx5e_mode_table[i].subtype |
3931 IFM_ETHER | IFM_FDX |
3932 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3936 /* Additional supported medias */
3937 ifmedia_add(&priv->media, IFM_10G_LR | IFM_ETHER, 0, NULL);
3938 ifmedia_add(&priv->media, IFM_10G_LR |
3939 IFM_ETHER | IFM_FDX |
3940 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3942 ifmedia_add(&priv->media, IFM_40G_ER4 | IFM_ETHER, 0, NULL);
3943 ifmedia_add(&priv->media, IFM_40G_ER4 |
3944 IFM_ETHER | IFM_FDX |
3945 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3947 ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO, 0, NULL);
3948 ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
3949 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3951 /* Set autoselect by default */
3952 ifmedia_set(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
3953 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE);
3954 ether_ifattach(ifp, dev_addr);
3956 /* Register for VLAN events */
3957 priv->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
3958 mlx5e_vlan_rx_add_vid, priv, EVENTHANDLER_PRI_FIRST);
3959 priv->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
3960 mlx5e_vlan_rx_kill_vid, priv, EVENTHANDLER_PRI_FIRST);
3962 /* Link is down by default */
3963 if_link_state_change(ifp, LINK_STATE_DOWN);
3965 mlx5e_enable_async_events(priv);
3967 mlx5e_add_hw_stats(priv);
3969 mlx5e_create_stats(&priv->stats.vport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3970 "vstats", mlx5e_vport_stats_desc, MLX5E_VPORT_STATS_NUM,
3971 priv->stats.vport.arg);
3973 mlx5e_create_stats(&priv->stats.pport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3974 "pstats", mlx5e_pport_stats_desc, MLX5E_PPORT_STATS_NUM,
3975 priv->stats.pport.arg);
3977 mlx5e_create_ethtool(priv);
3979 mtx_lock(&priv->async_events_mtx);
3980 mlx5e_update_stats(priv);
3981 mtx_unlock(&priv->async_events_mtx);
3983 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3984 OID_AUTO, "rx_clbr_done", CTLFLAG_RD,
3985 &priv->clbr_done, 0,
3986 "RX timestamps calibration state");
3987 callout_init(&priv->tstmp_clbr, CALLOUT_DIRECT);
3988 mlx5e_reset_calibration_callout(priv);
3994 mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
3996 err_dealloc_transport_domain:
3997 mlx5_dealloc_transport_domain(mdev, priv->tdn);
4000 mlx5_core_dealloc_pd(mdev, priv->pdn);
4003 mlx5_unmap_free_uar(mdev, &priv->cq_uar);
4006 destroy_workqueue(priv->wq);
4009 sysctl_ctx_free(&priv->sysctl_ctx);
4010 if (priv->sysctl_debug)
4011 sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
4015 mlx5e_priv_mtx_destroy(priv);
4016 free(priv, M_MLX5EN);
4021 mlx5e_destroy_ifp(struct mlx5_core_dev *mdev, void *vpriv)
4023 struct mlx5e_priv *priv = vpriv;
4024 struct ifnet *ifp = priv->ifp;
4026 /* don't allow more IOCTLs */
4029 /* XXX wait a bit to allow IOCTL handlers to complete */
4034 * The kernel can have reference(s) via the m_snd_tag's into
4035 * the ratelimit channels, and these must go away before
4038 while (READ_ONCE(priv->rl.stats.tx_active_connections) != 0) {
4039 if_printf(priv->ifp, "Waiting for all ratelimit connections "
4044 /* stop watchdog timer */
4045 callout_drain(&priv->watchdog);
4047 callout_drain(&priv->tstmp_clbr);
4049 if (priv->vlan_attach != NULL)
4050 EVENTHANDLER_DEREGISTER(vlan_config, priv->vlan_attach);
4051 if (priv->vlan_detach != NULL)
4052 EVENTHANDLER_DEREGISTER(vlan_unconfig, priv->vlan_detach);
4054 /* make sure device gets closed */
4056 mlx5e_close_locked(ifp);
4059 /* wait for all unlimited send tags to go away */
4060 while (priv->channel_refs != 0) {
4061 if_printf(priv->ifp, "Waiting for all unlimited connections "
4066 /* unregister device */
4067 ifmedia_removeall(&priv->media);
4068 ether_ifdetach(ifp);
4072 mlx5e_rl_cleanup(priv);
4074 /* destroy all remaining sysctl nodes */
4075 sysctl_ctx_free(&priv->stats.vport.ctx);
4076 sysctl_ctx_free(&priv->stats.pport.ctx);
4077 if (priv->sysctl_debug)
4078 sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
4079 sysctl_ctx_free(&priv->sysctl_ctx);
4081 mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
4082 mlx5_dealloc_transport_domain(priv->mdev, priv->tdn);
4083 mlx5_core_dealloc_pd(priv->mdev, priv->pdn);
4084 mlx5_unmap_free_uar(priv->mdev, &priv->cq_uar);
4085 mlx5e_disable_async_events(priv);
4086 destroy_workqueue(priv->wq);
4087 mlx5e_priv_mtx_destroy(priv);
4088 free(priv, M_MLX5EN);
4092 mlx5e_get_ifp(void *vpriv)
4094 struct mlx5e_priv *priv = vpriv;
4099 static struct mlx5_interface mlx5e_interface = {
4100 .add = mlx5e_create_ifp,
4101 .remove = mlx5e_destroy_ifp,
4102 .event = mlx5e_async_event,
4103 .protocol = MLX5_INTERFACE_PROTOCOL_ETH,
4104 .get_dev = mlx5e_get_ifp,
4110 mlx5_register_interface(&mlx5e_interface);
4116 mlx5_unregister_interface(&mlx5e_interface);
4120 mlx5e_show_version(void __unused *arg)
4123 printf("%s", mlx5e_version);
4125 SYSINIT(mlx5e_show_version, SI_SUB_DRIVERS, SI_ORDER_ANY, mlx5e_show_version, NULL);
4127 module_init_order(mlx5e_init, SI_ORDER_THIRD);
4128 module_exit_order(mlx5e_cleanup, SI_ORDER_THIRD);
4130 #if (__FreeBSD_version >= 1100000)
4131 MODULE_DEPEND(mlx5en, linuxkpi, 1, 1, 1);
4133 MODULE_DEPEND(mlx5en, mlx5, 1, 1, 1);
4134 MODULE_VERSION(mlx5en, 1);