2 * Copyright (c) 2015-2018 Mellanox Technologies. All rights reserved.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
13 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/sockio.h>
31 #include <machine/atomic.h>
33 #ifndef ETH_DRIVER_VERSION
34 #define ETH_DRIVER_VERSION "3.5.0"
36 #define DRIVER_RELDATE "November 2018"
38 static const char mlx5e_version[] = "mlx5en: Mellanox Ethernet driver "
39 ETH_DRIVER_VERSION " (" DRIVER_RELDATE ")\n";
41 static int mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs);
43 struct mlx5e_channel_param {
44 struct mlx5e_rq_param rq;
45 struct mlx5e_sq_param sq;
46 struct mlx5e_cq_param rx_cq;
47 struct mlx5e_cq_param tx_cq;
53 } mlx5e_mode_table[MLX5E_LINK_MODES_NUMBER] = {
55 [MLX5E_1000BASE_CX_SGMII] = {
56 .subtype = IFM_1000_CX_SGMII,
57 .baudrate = IF_Mbps(1000ULL),
59 [MLX5E_1000BASE_KX] = {
60 .subtype = IFM_1000_KX,
61 .baudrate = IF_Mbps(1000ULL),
63 [MLX5E_10GBASE_CX4] = {
64 .subtype = IFM_10G_CX4,
65 .baudrate = IF_Gbps(10ULL),
67 [MLX5E_10GBASE_KX4] = {
68 .subtype = IFM_10G_KX4,
69 .baudrate = IF_Gbps(10ULL),
71 [MLX5E_10GBASE_KR] = {
72 .subtype = IFM_10G_KR,
73 .baudrate = IF_Gbps(10ULL),
75 [MLX5E_20GBASE_KR2] = {
76 .subtype = IFM_20G_KR2,
77 .baudrate = IF_Gbps(20ULL),
79 [MLX5E_40GBASE_CR4] = {
80 .subtype = IFM_40G_CR4,
81 .baudrate = IF_Gbps(40ULL),
83 [MLX5E_40GBASE_KR4] = {
84 .subtype = IFM_40G_KR4,
85 .baudrate = IF_Gbps(40ULL),
87 [MLX5E_56GBASE_R4] = {
88 .subtype = IFM_56G_R4,
89 .baudrate = IF_Gbps(56ULL),
91 [MLX5E_10GBASE_CR] = {
92 .subtype = IFM_10G_CR1,
93 .baudrate = IF_Gbps(10ULL),
95 [MLX5E_10GBASE_SR] = {
96 .subtype = IFM_10G_SR,
97 .baudrate = IF_Gbps(10ULL),
99 [MLX5E_10GBASE_ER] = {
100 .subtype = IFM_10G_ER,
101 .baudrate = IF_Gbps(10ULL),
103 [MLX5E_40GBASE_SR4] = {
104 .subtype = IFM_40G_SR4,
105 .baudrate = IF_Gbps(40ULL),
107 [MLX5E_40GBASE_LR4] = {
108 .subtype = IFM_40G_LR4,
109 .baudrate = IF_Gbps(40ULL),
111 [MLX5E_100GBASE_CR4] = {
112 .subtype = IFM_100G_CR4,
113 .baudrate = IF_Gbps(100ULL),
115 [MLX5E_100GBASE_SR4] = {
116 .subtype = IFM_100G_SR4,
117 .baudrate = IF_Gbps(100ULL),
119 [MLX5E_100GBASE_KR4] = {
120 .subtype = IFM_100G_KR4,
121 .baudrate = IF_Gbps(100ULL),
123 [MLX5E_100GBASE_LR4] = {
124 .subtype = IFM_100G_LR4,
125 .baudrate = IF_Gbps(100ULL),
127 [MLX5E_100BASE_TX] = {
128 .subtype = IFM_100_TX,
129 .baudrate = IF_Mbps(100ULL),
131 [MLX5E_1000BASE_T] = {
132 .subtype = IFM_1000_T,
133 .baudrate = IF_Mbps(1000ULL),
135 [MLX5E_10GBASE_T] = {
136 .subtype = IFM_10G_T,
137 .baudrate = IF_Gbps(10ULL),
139 [MLX5E_25GBASE_CR] = {
140 .subtype = IFM_25G_CR,
141 .baudrate = IF_Gbps(25ULL),
143 [MLX5E_25GBASE_KR] = {
144 .subtype = IFM_25G_KR,
145 .baudrate = IF_Gbps(25ULL),
147 [MLX5E_25GBASE_SR] = {
148 .subtype = IFM_25G_SR,
149 .baudrate = IF_Gbps(25ULL),
151 [MLX5E_50GBASE_CR2] = {
152 .subtype = IFM_50G_CR2,
153 .baudrate = IF_Gbps(50ULL),
155 [MLX5E_50GBASE_KR2] = {
156 .subtype = IFM_50G_KR2,
157 .baudrate = IF_Gbps(50ULL),
161 MALLOC_DEFINE(M_MLX5EN, "MLX5EN", "MLX5 Ethernet");
164 mlx5e_update_carrier(struct mlx5e_priv *priv)
166 struct mlx5_core_dev *mdev = priv->mdev;
167 u32 out[MLX5_ST_SZ_DW(ptys_reg)];
174 port_state = mlx5_query_vport_state(mdev,
175 MLX5_QUERY_VPORT_STATE_IN_OP_MOD_VNIC_VPORT, 0);
177 if (port_state == VPORT_STATE_UP) {
178 priv->media_status_last |= IFM_ACTIVE;
180 priv->media_status_last &= ~IFM_ACTIVE;
181 priv->media_active_last = IFM_ETHER;
182 if_link_state_change(priv->ifp, LINK_STATE_DOWN);
186 error = mlx5_query_port_ptys(mdev, out, sizeof(out), MLX5_PTYS_EN, 1);
188 priv->media_active_last = IFM_ETHER;
189 priv->ifp->if_baudrate = 1;
190 if_printf(priv->ifp, "%s: query port ptys failed: 0x%x\n",
194 eth_proto_oper = MLX5_GET(ptys_reg, out, eth_proto_oper);
196 for (i = 0; i != MLX5E_LINK_MODES_NUMBER; i++) {
197 if (mlx5e_mode_table[i].baudrate == 0)
199 if (MLX5E_PROT_MASK(i) & eth_proto_oper) {
200 u32 subtype = mlx5e_mode_table[i].subtype;
202 priv->ifp->if_baudrate =
203 mlx5e_mode_table[i].baudrate;
207 error = mlx5_query_pddr_range_info(mdev, 1, &is_er_type);
209 if_printf(priv->ifp, "%s: query port pddr failed: %d\n",
212 if (error != 0 || is_er_type == 0)
213 subtype = IFM_10G_LR;
216 error = mlx5_query_pddr_range_info(mdev, 1, &is_er_type);
218 if_printf(priv->ifp, "%s: query port pddr failed: %d\n",
221 if (error == 0 && is_er_type != 0)
222 subtype = IFM_40G_ER4;
225 priv->media_active_last = subtype | IFM_ETHER | IFM_FDX;
229 if_link_state_change(priv->ifp, LINK_STATE_UP);
233 mlx5e_media_status(struct ifnet *dev, struct ifmediareq *ifmr)
235 struct mlx5e_priv *priv = dev->if_softc;
237 ifmr->ifm_status = priv->media_status_last;
238 ifmr->ifm_active = priv->media_active_last |
239 (priv->params.rx_pauseframe_control ? IFM_ETH_RXPAUSE : 0) |
240 (priv->params.tx_pauseframe_control ? IFM_ETH_TXPAUSE : 0);
245 mlx5e_find_link_mode(u32 subtype)
252 subtype = IFM_10G_ER;
255 subtype = IFM_40G_LR4;
259 for (i = 0; i < MLX5E_LINK_MODES_NUMBER; ++i) {
260 if (mlx5e_mode_table[i].baudrate == 0)
262 if (mlx5e_mode_table[i].subtype == subtype)
263 link_mode |= MLX5E_PROT_MASK(i);
270 mlx5e_set_port_pause_and_pfc(struct mlx5e_priv *priv)
272 return (mlx5_set_port_pause_and_pfc(priv->mdev, 1,
273 priv->params.rx_pauseframe_control,
274 priv->params.tx_pauseframe_control,
275 priv->params.rx_priority_flow_control,
276 priv->params.tx_priority_flow_control));
280 mlx5e_set_port_pfc(struct mlx5e_priv *priv)
284 if (priv->gone != 0) {
286 } else if (priv->params.rx_pauseframe_control ||
287 priv->params.tx_pauseframe_control) {
289 "Global pauseframes must be disabled before enabling PFC.\n");
292 error = mlx5e_set_port_pause_and_pfc(priv);
298 mlx5e_media_change(struct ifnet *dev)
300 struct mlx5e_priv *priv = dev->if_softc;
301 struct mlx5_core_dev *mdev = priv->mdev;
308 locked = PRIV_LOCKED(priv);
312 if (IFM_TYPE(priv->media.ifm_media) != IFM_ETHER) {
316 link_mode = mlx5e_find_link_mode(IFM_SUBTYPE(priv->media.ifm_media));
318 /* query supported capabilities */
319 error = mlx5_query_port_proto_cap(mdev, ð_proto_cap, MLX5_PTYS_EN);
321 if_printf(dev, "Query port media capability failed\n");
324 /* check for autoselect */
325 if (IFM_SUBTYPE(priv->media.ifm_media) == IFM_AUTO) {
326 link_mode = eth_proto_cap;
327 if (link_mode == 0) {
328 if_printf(dev, "Port media capability is zero\n");
333 link_mode = link_mode & eth_proto_cap;
334 if (link_mode == 0) {
335 if_printf(dev, "Not supported link mode requested\n");
340 if (priv->media.ifm_media & (IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE)) {
341 /* check if PFC is enabled */
342 if (priv->params.rx_priority_flow_control ||
343 priv->params.tx_priority_flow_control) {
344 if_printf(dev, "PFC must be disabled before enabling global pauseframes.\n");
349 /* update pauseframe control bits */
350 priv->params.rx_pauseframe_control =
351 (priv->media.ifm_media & IFM_ETH_RXPAUSE) ? 1 : 0;
352 priv->params.tx_pauseframe_control =
353 (priv->media.ifm_media & IFM_ETH_TXPAUSE) ? 1 : 0;
355 /* check if device is opened */
356 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
358 /* reconfigure the hardware */
359 mlx5_set_port_status(mdev, MLX5_PORT_DOWN);
360 mlx5_set_port_proto(mdev, link_mode, MLX5_PTYS_EN);
361 error = -mlx5e_set_port_pause_and_pfc(priv);
363 mlx5_set_port_status(mdev, MLX5_PORT_UP);
372 mlx5e_update_carrier_work(struct work_struct *work)
374 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
375 update_carrier_work);
378 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
379 mlx5e_update_carrier(priv);
384 * This function reads the physical port counters from the firmware
385 * using a pre-defined layout defined by various MLX5E_PPORT_XXX()
386 * macros. The output is converted from big-endian 64-bit values into
387 * host endian ones and stored in the "priv->stats.pport" structure.
390 mlx5e_update_pport_counters(struct mlx5e_priv *priv)
392 struct mlx5_core_dev *mdev = priv->mdev;
393 struct mlx5e_pport_stats *s = &priv->stats.pport;
394 struct mlx5e_port_stats_debug *s_debug = &priv->stats.port_stats_debug;
398 unsigned sz = MLX5_ST_SZ_BYTES(ppcnt_reg);
403 /* allocate firmware request structures */
404 in = mlx5_vzalloc(sz);
405 out = mlx5_vzalloc(sz);
406 if (in == NULL || out == NULL)
410 * Get pointer to the 64-bit counter set which is located at a
411 * fixed offset in the output firmware request structure:
413 ptr = (const uint64_t *)MLX5_ADDR_OF(ppcnt_reg, out, counter_set);
415 MLX5_SET(ppcnt_reg, in, local_port, 1);
417 /* read IEEE802_3 counter group using predefined counter layout */
418 MLX5_SET(ppcnt_reg, in, grp, MLX5_IEEE_802_3_COUNTERS_GROUP);
419 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
420 for (x = 0, y = MLX5E_PPORT_PER_PRIO_STATS_NUM;
421 x != MLX5E_PPORT_IEEE802_3_STATS_NUM; x++, y++)
422 s->arg[y] = be64toh(ptr[x]);
424 /* read RFC2819 counter group using predefined counter layout */
425 MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2819_COUNTERS_GROUP);
426 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
427 for (x = 0; x != MLX5E_PPORT_RFC2819_STATS_NUM; x++, y++)
428 s->arg[y] = be64toh(ptr[x]);
429 for (y = 0; x != MLX5E_PPORT_RFC2819_STATS_NUM +
430 MLX5E_PPORT_RFC2819_STATS_DEBUG_NUM; x++, y++)
431 s_debug->arg[y] = be64toh(ptr[x]);
433 /* read RFC2863 counter group using predefined counter layout */
434 MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2863_COUNTERS_GROUP);
435 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
436 for (x = 0; x != MLX5E_PPORT_RFC2863_STATS_DEBUG_NUM; x++, y++)
437 s_debug->arg[y] = be64toh(ptr[x]);
439 /* read physical layer stats counter group using predefined counter layout */
440 MLX5_SET(ppcnt_reg, in, grp, MLX5_PHYSICAL_LAYER_COUNTERS_GROUP);
441 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
442 for (x = 0; x != MLX5E_PPORT_PHYSICAL_LAYER_STATS_DEBUG_NUM; x++, y++)
443 s_debug->arg[y] = be64toh(ptr[x]);
445 /* read Extended Ethernet counter group using predefined counter layout */
446 MLX5_SET(ppcnt_reg, in, grp, MLX5_ETHERNET_EXTENDED_COUNTERS_GROUP);
447 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
448 for (x = 0; x != MLX5E_PPORT_ETHERNET_EXTENDED_STATS_DEBUG_NUM; x++, y++)
449 s_debug->arg[y] = be64toh(ptr[x]);
451 /* read per-priority counters */
452 MLX5_SET(ppcnt_reg, in, grp, MLX5_PER_PRIORITY_COUNTERS_GROUP);
454 /* iterate all the priorities */
455 for (y = z = 0; z != MLX5E_PPORT_PER_PRIO_STATS_NUM_PRIO; z++) {
456 MLX5_SET(ppcnt_reg, in, prio_tc, z);
457 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
459 /* read per priority stats counter group using predefined counter layout */
460 for (x = 0; x != (MLX5E_PPORT_PER_PRIO_STATS_NUM /
461 MLX5E_PPORT_PER_PRIO_STATS_NUM_PRIO); x++, y++)
462 s->arg[y] = be64toh(ptr[x]);
466 /* free firmware request structures */
472 * This function is called regularly to collect all statistics
473 * counters from the firmware. The values can be viewed through the
474 * sysctl interface. Execution is serialized using the priv's global
475 * configuration lock.
478 mlx5e_update_stats_locked(struct mlx5e_priv *priv)
480 struct mlx5_core_dev *mdev = priv->mdev;
481 struct mlx5e_vport_stats *s = &priv->stats.vport;
482 struct mlx5e_sq_stats *sq_stats;
483 struct buf_ring *sq_br;
484 #if (__FreeBSD_version < 1100000)
485 struct ifnet *ifp = priv->ifp;
488 u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)];
490 int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
493 u64 tx_queue_dropped = 0;
494 u64 tx_defragged = 0;
495 u64 tx_offload_none = 0;
498 u64 sw_lro_queued = 0;
499 u64 sw_lro_flushed = 0;
500 u64 rx_csum_none = 0;
504 u32 rx_out_of_buffer = 0;
508 out = mlx5_vzalloc(outlen);
512 /* Collect firts the SW counters and then HW for consistency */
513 for (i = 0; i < priv->params.num_channels; i++) {
514 struct mlx5e_channel *pch = priv->channel + i;
515 struct mlx5e_rq *rq = &pch->rq;
516 struct mlx5e_rq_stats *rq_stats = &pch->rq.stats;
518 /* collect stats from LRO */
519 rq_stats->sw_lro_queued = rq->lro.lro_queued;
520 rq_stats->sw_lro_flushed = rq->lro.lro_flushed;
521 sw_lro_queued += rq_stats->sw_lro_queued;
522 sw_lro_flushed += rq_stats->sw_lro_flushed;
523 lro_packets += rq_stats->lro_packets;
524 lro_bytes += rq_stats->lro_bytes;
525 rx_csum_none += rq_stats->csum_none;
526 rx_wqe_err += rq_stats->wqe_err;
527 rx_packets += rq_stats->packets;
528 rx_bytes += rq_stats->bytes;
530 for (j = 0; j < priv->num_tc; j++) {
531 sq_stats = &pch->sq[j].stats;
532 sq_br = pch->sq[j].br;
534 tso_packets += sq_stats->tso_packets;
535 tso_bytes += sq_stats->tso_bytes;
536 tx_queue_dropped += sq_stats->dropped;
538 tx_queue_dropped += sq_br->br_drops;
539 tx_defragged += sq_stats->defragged;
540 tx_offload_none += sq_stats->csum_offload_none;
544 /* update counters */
545 s->tso_packets = tso_packets;
546 s->tso_bytes = tso_bytes;
547 s->tx_queue_dropped = tx_queue_dropped;
548 s->tx_defragged = tx_defragged;
549 s->lro_packets = lro_packets;
550 s->lro_bytes = lro_bytes;
551 s->sw_lro_queued = sw_lro_queued;
552 s->sw_lro_flushed = sw_lro_flushed;
553 s->rx_csum_none = rx_csum_none;
554 s->rx_wqe_err = rx_wqe_err;
555 s->rx_packets = rx_packets;
556 s->rx_bytes = rx_bytes;
559 memset(in, 0, sizeof(in));
561 MLX5_SET(query_vport_counter_in, in, opcode,
562 MLX5_CMD_OP_QUERY_VPORT_COUNTER);
563 MLX5_SET(query_vport_counter_in, in, op_mod, 0);
564 MLX5_SET(query_vport_counter_in, in, other_vport, 0);
566 memset(out, 0, outlen);
568 /* get number of out-of-buffer drops first */
569 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0 &&
570 mlx5_vport_query_out_of_rx_buffer(mdev, priv->counter_set_id,
571 &rx_out_of_buffer) == 0) {
572 /* accumulate difference into a 64-bit counter */
573 s->rx_out_of_buffer += (u64)(u32)(rx_out_of_buffer -
574 s->rx_out_of_buffer_prev);
575 s->rx_out_of_buffer_prev = rx_out_of_buffer;
578 /* get port statistics */
579 if (mlx5_cmd_exec(mdev, in, sizeof(in), out, outlen) == 0) {
580 #define MLX5_GET_CTR(out, x) \
581 MLX5_GET64(query_vport_counter_out, out, x)
583 s->rx_error_packets =
584 MLX5_GET_CTR(out, received_errors.packets);
586 MLX5_GET_CTR(out, received_errors.octets);
587 s->tx_error_packets =
588 MLX5_GET_CTR(out, transmit_errors.packets);
590 MLX5_GET_CTR(out, transmit_errors.octets);
592 s->rx_unicast_packets =
593 MLX5_GET_CTR(out, received_eth_unicast.packets);
594 s->rx_unicast_bytes =
595 MLX5_GET_CTR(out, received_eth_unicast.octets);
596 s->tx_unicast_packets =
597 MLX5_GET_CTR(out, transmitted_eth_unicast.packets);
598 s->tx_unicast_bytes =
599 MLX5_GET_CTR(out, transmitted_eth_unicast.octets);
601 s->rx_multicast_packets =
602 MLX5_GET_CTR(out, received_eth_multicast.packets);
603 s->rx_multicast_bytes =
604 MLX5_GET_CTR(out, received_eth_multicast.octets);
605 s->tx_multicast_packets =
606 MLX5_GET_CTR(out, transmitted_eth_multicast.packets);
607 s->tx_multicast_bytes =
608 MLX5_GET_CTR(out, transmitted_eth_multicast.octets);
610 s->rx_broadcast_packets =
611 MLX5_GET_CTR(out, received_eth_broadcast.packets);
612 s->rx_broadcast_bytes =
613 MLX5_GET_CTR(out, received_eth_broadcast.octets);
614 s->tx_broadcast_packets =
615 MLX5_GET_CTR(out, transmitted_eth_broadcast.packets);
616 s->tx_broadcast_bytes =
617 MLX5_GET_CTR(out, transmitted_eth_broadcast.octets);
619 s->tx_packets = s->tx_unicast_packets +
620 s->tx_multicast_packets + s->tx_broadcast_packets;
621 s->tx_bytes = s->tx_unicast_bytes + s->tx_multicast_bytes +
622 s->tx_broadcast_bytes;
624 /* Update calculated offload counters */
625 s->tx_csum_offload = s->tx_packets - tx_offload_none;
626 s->rx_csum_good = s->rx_packets - s->rx_csum_none;
629 /* Get physical port counters */
630 mlx5e_update_pport_counters(priv);
632 s->tx_jumbo_packets =
633 priv->stats.port_stats_debug.tx_stat_p1519to2047octets +
634 priv->stats.port_stats_debug.tx_stat_p2048to4095octets +
635 priv->stats.port_stats_debug.tx_stat_p4096to8191octets +
636 priv->stats.port_stats_debug.tx_stat_p8192to10239octets;
638 #if (__FreeBSD_version < 1100000)
639 /* no get_counters interface in fbsd 10 */
640 ifp->if_ipackets = s->rx_packets;
641 ifp->if_ierrors = priv->stats.pport.in_range_len_errors +
642 priv->stats.pport.out_of_range_len +
643 priv->stats.pport.too_long_errors +
644 priv->stats.pport.check_seq_err +
645 priv->stats.pport.alignment_err;
646 ifp->if_iqdrops = s->rx_out_of_buffer;
647 ifp->if_opackets = s->tx_packets;
648 ifp->if_oerrors = priv->stats.port_stats_debug.out_discards;
649 ifp->if_snd.ifq_drops = s->tx_queue_dropped;
650 ifp->if_ibytes = s->rx_bytes;
651 ifp->if_obytes = s->tx_bytes;
653 priv->stats.pport.collisions;
659 /* Update diagnostics, if any */
660 if (priv->params_ethtool.diag_pci_enable ||
661 priv->params_ethtool.diag_general_enable) {
662 int error = mlx5_core_get_diagnostics_full(mdev,
663 priv->params_ethtool.diag_pci_enable ? &priv->params_pci : NULL,
664 priv->params_ethtool.diag_general_enable ? &priv->params_general : NULL);
666 if_printf(priv->ifp, "Failed reading diagnostics: %d\n", error);
671 mlx5e_update_stats_work(struct work_struct *work)
673 struct mlx5e_priv *priv;
675 priv = container_of(work, struct mlx5e_priv, update_stats_work);
677 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
678 mlx5e_update_stats_locked(priv);
683 mlx5e_update_stats(void *arg)
685 struct mlx5e_priv *priv = arg;
687 queue_work(priv->wq, &priv->update_stats_work);
689 callout_reset(&priv->watchdog, hz, &mlx5e_update_stats, priv);
693 mlx5e_async_event_sub(struct mlx5e_priv *priv,
694 enum mlx5_dev_event event)
697 case MLX5_DEV_EVENT_PORT_UP:
698 case MLX5_DEV_EVENT_PORT_DOWN:
699 queue_work(priv->wq, &priv->update_carrier_work);
708 mlx5e_async_event(struct mlx5_core_dev *mdev, void *vpriv,
709 enum mlx5_dev_event event, unsigned long param)
711 struct mlx5e_priv *priv = vpriv;
713 mtx_lock(&priv->async_events_mtx);
714 if (test_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state))
715 mlx5e_async_event_sub(priv, event);
716 mtx_unlock(&priv->async_events_mtx);
720 mlx5e_enable_async_events(struct mlx5e_priv *priv)
722 set_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
726 mlx5e_disable_async_events(struct mlx5e_priv *priv)
728 mtx_lock(&priv->async_events_mtx);
729 clear_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
730 mtx_unlock(&priv->async_events_mtx);
733 static void mlx5e_calibration_callout(void *arg);
734 static int mlx5e_calibration_duration = 20;
735 static int mlx5e_fast_calibration = 1;
736 static int mlx5e_normal_calibration = 30;
738 static SYSCTL_NODE(_hw_mlx5, OID_AUTO, calibr, CTLFLAG_RW, 0,
739 "MLX5 timestamp calibration parameteres");
741 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, duration, CTLFLAG_RWTUN,
742 &mlx5e_calibration_duration, 0,
743 "Duration of initial calibration");
744 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, fast, CTLFLAG_RWTUN,
745 &mlx5e_fast_calibration, 0,
746 "Recalibration interval during initial calibration");
747 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, normal, CTLFLAG_RWTUN,
748 &mlx5e_normal_calibration, 0,
749 "Recalibration interval during normal operations");
752 * Ignites the calibration process.
755 mlx5e_reset_calibration_callout(struct mlx5e_priv *priv)
758 if (priv->clbr_done == 0)
759 mlx5e_calibration_callout(priv);
761 callout_reset_curcpu(&priv->tstmp_clbr, (priv->clbr_done <
762 mlx5e_calibration_duration ? mlx5e_fast_calibration :
763 mlx5e_normal_calibration) * hz, mlx5e_calibration_callout,
768 mlx5e_timespec2usec(const struct timespec *ts)
771 return ((uint64_t)ts->tv_sec * 1000000000 + ts->tv_nsec);
775 mlx5e_hw_clock(struct mlx5e_priv *priv)
777 struct mlx5_init_seg *iseg;
778 uint32_t hw_h, hw_h1, hw_l;
780 iseg = priv->mdev->iseg;
782 hw_h = ioread32be(&iseg->internal_timer_h);
783 hw_l = ioread32be(&iseg->internal_timer_l);
784 hw_h1 = ioread32be(&iseg->internal_timer_h);
785 } while (hw_h1 != hw_h);
786 return (((uint64_t)hw_h << 32) | hw_l);
790 * The calibration callout, it runs either in the context of the
791 * thread which enables calibration, or in callout. It takes the
792 * snapshot of system and adapter clocks, then advances the pointers to
793 * the calibration point to allow rx path to read the consistent data
797 mlx5e_calibration_callout(void *arg)
799 struct mlx5e_priv *priv;
800 struct mlx5e_clbr_point *next, *curr;
805 curr = &priv->clbr_points[priv->clbr_curr];
806 clbr_curr_next = priv->clbr_curr + 1;
807 if (clbr_curr_next >= nitems(priv->clbr_points))
809 next = &priv->clbr_points[clbr_curr_next];
811 next->base_prev = curr->base_curr;
812 next->clbr_hw_prev = curr->clbr_hw_curr;
814 next->clbr_hw_curr = mlx5e_hw_clock(priv);
815 if (((next->clbr_hw_curr - curr->clbr_hw_prev) >> MLX5E_TSTMP_PREC) ==
817 if_printf(priv->ifp, "HW failed tstmp frozen %#jx %#jx,"
818 "disabling\n", next->clbr_hw_curr, curr->clbr_hw_prev);
824 next->base_curr = mlx5e_timespec2usec(&ts);
827 atomic_thread_fence_rel();
828 priv->clbr_curr = clbr_curr_next;
829 atomic_store_rel_int(&next->clbr_gen, ++(priv->clbr_gen));
831 if (priv->clbr_done < mlx5e_calibration_duration)
833 mlx5e_reset_calibration_callout(priv);
836 static const char *mlx5e_rq_stats_desc[] = {
837 MLX5E_RQ_STATS(MLX5E_STATS_DESC)
841 mlx5e_create_rq(struct mlx5e_channel *c,
842 struct mlx5e_rq_param *param,
845 struct mlx5e_priv *priv = c->priv;
846 struct mlx5_core_dev *mdev = priv->mdev;
848 void *rqc = param->rqc;
849 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
855 err = mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
859 /* Create DMA descriptor TAG */
860 if ((err = -bus_dma_tag_create(
861 bus_get_dma_tag(mdev->pdev->dev.bsddev),
862 1, /* any alignment */
864 BUS_SPACE_MAXADDR, /* lowaddr */
865 BUS_SPACE_MAXADDR, /* highaddr */
866 NULL, NULL, /* filter, filterarg */
867 nsegs * MLX5E_MAX_RX_BYTES, /* maxsize */
868 nsegs, /* nsegments */
869 nsegs * MLX5E_MAX_RX_BYTES, /* maxsegsize */
871 NULL, NULL, /* lockfunc, lockfuncarg */
875 err = mlx5_wq_ll_create(mdev, ¶m->wq, rqc_wq, &rq->wq,
878 goto err_free_dma_tag;
880 rq->wq.db = &rq->wq.db[MLX5_RCV_DBR];
882 err = mlx5e_get_wqe_sz(priv, &rq->wqe_sz, &rq->nsegs);
884 goto err_rq_wq_destroy;
886 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
888 err = -tcp_lro_init_args(&rq->lro, c->tag.m_snd_tag.ifp, TCP_LRO_ENTRIES, wq_sz);
890 goto err_rq_wq_destroy;
892 rq->mbuf = malloc(wq_sz * sizeof(rq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
893 for (i = 0; i != wq_sz; i++) {
894 struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
895 #if (MLX5E_MAX_RX_SEGS == 1)
896 uint32_t byte_count = rq->wqe_sz - MLX5E_NET_IP_ALIGN;
901 err = -bus_dmamap_create(rq->dma_tag, 0, &rq->mbuf[i].dma_map);
904 bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
905 goto err_rq_mbuf_free;
908 /* set value for constant fields */
909 #if (MLX5E_MAX_RX_SEGS == 1)
910 wqe->data[0].lkey = c->mkey_be;
911 wqe->data[0].byte_count = cpu_to_be32(byte_count | MLX5_HW_START_PADDING);
913 for (j = 0; j < rq->nsegs; j++)
914 wqe->data[j].lkey = c->mkey_be;
918 INIT_WORK(&rq->dim.work, mlx5e_dim_work);
919 if (priv->params.rx_cq_moderation_mode < 2) {
920 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
922 void *cqc = container_of(param,
923 struct mlx5e_channel_param, rq)->rx_cq.cqc;
925 switch (MLX5_GET(cqc, cqc, cq_period_mode)) {
926 case MLX5_CQ_PERIOD_MODE_START_FROM_EQE:
927 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
929 case MLX5_CQ_PERIOD_MODE_START_FROM_CQE:
930 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
933 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
938 rq->ifp = c->tag.m_snd_tag.ifp;
942 snprintf(buffer, sizeof(buffer), "rxstat%d", c->ix);
943 mlx5e_create_stats(&rq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
944 buffer, mlx5e_rq_stats_desc, MLX5E_RQ_STATS_NUM,
949 free(rq->mbuf, M_MLX5EN);
950 tcp_lro_free(&rq->lro);
952 mlx5_wq_destroy(&rq->wq_ctrl);
954 bus_dma_tag_destroy(rq->dma_tag);
960 mlx5e_destroy_rq(struct mlx5e_rq *rq)
965 /* destroy all sysctl nodes */
966 sysctl_ctx_free(&rq->stats.ctx);
968 /* free leftover LRO packets, if any */
969 tcp_lro_free(&rq->lro);
971 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
972 for (i = 0; i != wq_sz; i++) {
973 if (rq->mbuf[i].mbuf != NULL) {
974 bus_dmamap_unload(rq->dma_tag, rq->mbuf[i].dma_map);
975 m_freem(rq->mbuf[i].mbuf);
977 bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
979 free(rq->mbuf, M_MLX5EN);
980 mlx5_wq_destroy(&rq->wq_ctrl);
984 mlx5e_enable_rq(struct mlx5e_rq *rq, struct mlx5e_rq_param *param)
986 struct mlx5e_channel *c = rq->channel;
987 struct mlx5e_priv *priv = c->priv;
988 struct mlx5_core_dev *mdev = priv->mdev;
996 inlen = MLX5_ST_SZ_BYTES(create_rq_in) +
997 sizeof(u64) * rq->wq_ctrl.buf.npages;
998 in = mlx5_vzalloc(inlen);
1002 rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
1003 wq = MLX5_ADDR_OF(rqc, rqc, wq);
1005 memcpy(rqc, param->rqc, sizeof(param->rqc));
1007 MLX5_SET(rqc, rqc, cqn, c->rq.cq.mcq.cqn);
1008 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST);
1009 MLX5_SET(rqc, rqc, flush_in_error_en, 1);
1010 if (priv->counter_set_id >= 0)
1011 MLX5_SET(rqc, rqc, counter_set_id, priv->counter_set_id);
1012 MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift -
1014 MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma);
1016 mlx5_fill_page_array(&rq->wq_ctrl.buf,
1017 (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
1019 err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn);
1027 mlx5e_modify_rq(struct mlx5e_rq *rq, int curr_state, int next_state)
1029 struct mlx5e_channel *c = rq->channel;
1030 struct mlx5e_priv *priv = c->priv;
1031 struct mlx5_core_dev *mdev = priv->mdev;
1038 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
1039 in = mlx5_vzalloc(inlen);
1043 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
1045 MLX5_SET(modify_rq_in, in, rqn, rq->rqn);
1046 MLX5_SET(modify_rq_in, in, rq_state, curr_state);
1047 MLX5_SET(rqc, rqc, state, next_state);
1049 err = mlx5_core_modify_rq(mdev, in, inlen);
1057 mlx5e_disable_rq(struct mlx5e_rq *rq)
1059 struct mlx5e_channel *c = rq->channel;
1060 struct mlx5e_priv *priv = c->priv;
1061 struct mlx5_core_dev *mdev = priv->mdev;
1063 mlx5_core_destroy_rq(mdev, rq->rqn);
1067 mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq)
1069 struct mlx5e_channel *c = rq->channel;
1070 struct mlx5e_priv *priv = c->priv;
1071 struct mlx5_wq_ll *wq = &rq->wq;
1074 for (i = 0; i < 1000; i++) {
1075 if (wq->cur_sz >= priv->params.min_rx_wqes)
1080 return (-ETIMEDOUT);
1084 mlx5e_open_rq(struct mlx5e_channel *c,
1085 struct mlx5e_rq_param *param,
1086 struct mlx5e_rq *rq)
1090 err = mlx5e_create_rq(c, param, rq);
1094 err = mlx5e_enable_rq(rq, param);
1096 goto err_destroy_rq;
1098 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
1100 goto err_disable_rq;
1107 mlx5e_disable_rq(rq);
1109 mlx5e_destroy_rq(rq);
1115 mlx5e_close_rq(struct mlx5e_rq *rq)
1119 callout_stop(&rq->watchdog);
1120 mtx_unlock(&rq->mtx);
1122 callout_drain(&rq->watchdog);
1124 mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
1128 mlx5e_close_rq_wait(struct mlx5e_rq *rq)
1131 mlx5e_disable_rq(rq);
1132 mlx5e_close_cq(&rq->cq);
1133 cancel_work_sync(&rq->dim.work);
1134 mlx5e_destroy_rq(rq);
1138 mlx5e_free_sq_db(struct mlx5e_sq *sq)
1140 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1143 for (x = 0; x != wq_sz; x++) {
1144 if (sq->mbuf[x].mbuf != NULL) {
1145 bus_dmamap_unload(sq->dma_tag, sq->mbuf[x].dma_map);
1146 m_freem(sq->mbuf[x].mbuf);
1148 bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1150 free(sq->mbuf, M_MLX5EN);
1154 mlx5e_alloc_sq_db(struct mlx5e_sq *sq)
1156 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1160 sq->mbuf = malloc(wq_sz * sizeof(sq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
1162 /* Create DMA descriptor MAPs */
1163 for (x = 0; x != wq_sz; x++) {
1164 err = -bus_dmamap_create(sq->dma_tag, 0, &sq->mbuf[x].dma_map);
1167 bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1168 free(sq->mbuf, M_MLX5EN);
1175 static const char *mlx5e_sq_stats_desc[] = {
1176 MLX5E_SQ_STATS(MLX5E_STATS_DESC)
1180 mlx5e_update_sq_inline(struct mlx5e_sq *sq)
1182 sq->max_inline = sq->priv->params.tx_max_inline;
1183 sq->min_inline_mode = sq->priv->params.tx_min_inline_mode;
1186 * Check if trust state is DSCP or if inline mode is NONE which
1187 * indicates CX-5 or newer hardware.
1189 if (sq->priv->params_ethtool.trust_state != MLX5_QPTS_TRUST_PCP ||
1190 sq->min_inline_mode == MLX5_INLINE_MODE_NONE) {
1191 if (MLX5_CAP_ETH(sq->priv->mdev, wqe_vlan_insert))
1192 sq->min_insert_caps = MLX5E_INSERT_VLAN | MLX5E_INSERT_NON_VLAN;
1194 sq->min_insert_caps = MLX5E_INSERT_NON_VLAN;
1196 sq->min_insert_caps = 0;
1201 mlx5e_refresh_sq_inline_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
1205 for (i = 0; i != c->num_tc; i++) {
1206 mtx_lock(&c->sq[i].lock);
1207 mlx5e_update_sq_inline(&c->sq[i]);
1208 mtx_unlock(&c->sq[i].lock);
1213 mlx5e_refresh_sq_inline(struct mlx5e_priv *priv)
1217 /* check if channels are closed */
1218 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
1221 for (i = 0; i < priv->params.num_channels; i++)
1222 mlx5e_refresh_sq_inline_sub(priv, &priv->channel[i]);
1226 mlx5e_create_sq(struct mlx5e_channel *c,
1228 struct mlx5e_sq_param *param,
1229 struct mlx5e_sq *sq)
1231 struct mlx5e_priv *priv = c->priv;
1232 struct mlx5_core_dev *mdev = priv->mdev;
1234 void *sqc = param->sqc;
1235 void *sqc_wq = MLX5_ADDR_OF(sqc, sqc, wq);
1238 /* Create DMA descriptor TAG */
1239 if ((err = -bus_dma_tag_create(
1240 bus_get_dma_tag(mdev->pdev->dev.bsddev),
1241 1, /* any alignment */
1242 0, /* no boundary */
1243 BUS_SPACE_MAXADDR, /* lowaddr */
1244 BUS_SPACE_MAXADDR, /* highaddr */
1245 NULL, NULL, /* filter, filterarg */
1246 MLX5E_MAX_TX_PAYLOAD_SIZE, /* maxsize */
1247 MLX5E_MAX_TX_MBUF_FRAGS, /* nsegments */
1248 MLX5E_MAX_TX_MBUF_SIZE, /* maxsegsize */
1250 NULL, NULL, /* lockfunc, lockfuncarg */
1254 err = mlx5_alloc_map_uar(mdev, &sq->uar);
1256 goto err_free_dma_tag;
1258 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, &sq->wq,
1261 goto err_unmap_free_uar;
1263 sq->wq.db = &sq->wq.db[MLX5_SND_DBR];
1264 sq->bf_buf_size = (1 << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2;
1266 err = mlx5e_alloc_sq_db(sq);
1268 goto err_sq_wq_destroy;
1270 sq->mkey_be = c->mkey_be;
1271 sq->ifp = priv->ifp;
1275 mlx5e_update_sq_inline(sq);
1277 snprintf(buffer, sizeof(buffer), "txstat%dtc%d", c->ix, tc);
1278 mlx5e_create_stats(&sq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
1279 buffer, mlx5e_sq_stats_desc, MLX5E_SQ_STATS_NUM,
1285 mlx5_wq_destroy(&sq->wq_ctrl);
1288 mlx5_unmap_free_uar(mdev, &sq->uar);
1291 bus_dma_tag_destroy(sq->dma_tag);
1297 mlx5e_destroy_sq(struct mlx5e_sq *sq)
1299 /* destroy all sysctl nodes */
1300 sysctl_ctx_free(&sq->stats.ctx);
1302 mlx5e_free_sq_db(sq);
1303 mlx5_wq_destroy(&sq->wq_ctrl);
1304 mlx5_unmap_free_uar(sq->priv->mdev, &sq->uar);
1308 mlx5e_enable_sq(struct mlx5e_sq *sq, struct mlx5e_sq_param *param,
1317 inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
1318 sizeof(u64) * sq->wq_ctrl.buf.npages;
1319 in = mlx5_vzalloc(inlen);
1323 sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
1324 wq = MLX5_ADDR_OF(sqc, sqc, wq);
1326 memcpy(sqc, param->sqc, sizeof(param->sqc));
1328 MLX5_SET(sqc, sqc, tis_num_0, tis_num);
1329 MLX5_SET(sqc, sqc, cqn, sq->cq.mcq.cqn);
1330 MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
1331 MLX5_SET(sqc, sqc, tis_lst_sz, 1);
1332 MLX5_SET(sqc, sqc, flush_in_error_en, 1);
1334 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
1335 MLX5_SET(wq, wq, uar_page, sq->uar.index);
1336 MLX5_SET(wq, wq, log_wq_pg_sz, sq->wq_ctrl.buf.page_shift -
1338 MLX5_SET64(wq, wq, dbr_addr, sq->wq_ctrl.db.dma);
1340 mlx5_fill_page_array(&sq->wq_ctrl.buf,
1341 (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
1343 err = mlx5_core_create_sq(sq->priv->mdev, in, inlen, &sq->sqn);
1351 mlx5e_modify_sq(struct mlx5e_sq *sq, int curr_state, int next_state)
1358 inlen = MLX5_ST_SZ_BYTES(modify_sq_in);
1359 in = mlx5_vzalloc(inlen);
1363 sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx);
1365 MLX5_SET(modify_sq_in, in, sqn, sq->sqn);
1366 MLX5_SET(modify_sq_in, in, sq_state, curr_state);
1367 MLX5_SET(sqc, sqc, state, next_state);
1369 err = mlx5_core_modify_sq(sq->priv->mdev, in, inlen);
1377 mlx5e_disable_sq(struct mlx5e_sq *sq)
1380 mlx5_core_destroy_sq(sq->priv->mdev, sq->sqn);
1384 mlx5e_open_sq(struct mlx5e_channel *c,
1386 struct mlx5e_sq_param *param,
1387 struct mlx5e_sq *sq)
1391 err = mlx5e_create_sq(c, tc, param, sq);
1395 err = mlx5e_enable_sq(sq, param, c->priv->tisn[tc]);
1397 goto err_destroy_sq;
1399 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY);
1401 goto err_disable_sq;
1403 WRITE_ONCE(sq->running, 1);
1408 mlx5e_disable_sq(sq);
1410 mlx5e_destroy_sq(sq);
1416 mlx5e_sq_send_nops_locked(struct mlx5e_sq *sq, int can_sleep)
1418 /* fill up remainder with NOPs */
1419 while (sq->cev_counter != 0) {
1420 while (!mlx5e_sq_has_room_for(sq, 1)) {
1421 if (can_sleep != 0) {
1422 mtx_unlock(&sq->lock);
1424 mtx_lock(&sq->lock);
1429 /* send a single NOP */
1430 mlx5e_send_nop(sq, 1);
1431 atomic_thread_fence_rel();
1434 /* Check if we need to write the doorbell */
1435 if (likely(sq->doorbell.d64 != 0)) {
1436 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
1437 sq->doorbell.d64 = 0;
1442 mlx5e_sq_cev_timeout(void *arg)
1444 struct mlx5e_sq *sq = arg;
1446 mtx_assert(&sq->lock, MA_OWNED);
1448 /* check next state */
1449 switch (sq->cev_next_state) {
1450 case MLX5E_CEV_STATE_SEND_NOPS:
1451 /* fill TX ring with NOPs, if any */
1452 mlx5e_sq_send_nops_locked(sq, 0);
1454 /* check if completed */
1455 if (sq->cev_counter == 0) {
1456 sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
1461 /* send NOPs on next timeout */
1462 sq->cev_next_state = MLX5E_CEV_STATE_SEND_NOPS;
1467 callout_reset_curcpu(&sq->cev_callout, hz, mlx5e_sq_cev_timeout, sq);
1471 mlx5e_drain_sq(struct mlx5e_sq *sq)
1474 struct mlx5_core_dev *mdev= sq->priv->mdev;
1477 * Check if already stopped.
1479 * NOTE: Serialization of this function is managed by the
1480 * caller ensuring the priv's state lock is locked or in case
1481 * of rate limit support, a single thread manages drain and
1482 * resume of SQs. The "running" variable can therefore safely
1483 * be read without any locks.
1485 if (READ_ONCE(sq->running) == 0)
1488 /* don't put more packets into the SQ */
1489 WRITE_ONCE(sq->running, 0);
1491 /* serialize access to DMA rings */
1492 mtx_lock(&sq->lock);
1494 /* teardown event factor timer, if any */
1495 sq->cev_next_state = MLX5E_CEV_STATE_HOLD_NOPS;
1496 callout_stop(&sq->cev_callout);
1498 /* send dummy NOPs in order to flush the transmit ring */
1499 mlx5e_sq_send_nops_locked(sq, 1);
1500 mtx_unlock(&sq->lock);
1502 /* make sure it is safe to free the callout */
1503 callout_drain(&sq->cev_callout);
1505 /* wait till SQ is empty or link is down */
1506 mtx_lock(&sq->lock);
1507 while (sq->cc != sq->pc &&
1508 (sq->priv->media_status_last & IFM_ACTIVE) != 0 &&
1509 mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1510 mtx_unlock(&sq->lock);
1512 sq->cq.mcq.comp(&sq->cq.mcq);
1513 mtx_lock(&sq->lock);
1515 mtx_unlock(&sq->lock);
1517 /* error out remaining requests */
1518 error = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RDY, MLX5_SQC_STATE_ERR);
1521 "mlx5e_modify_sq() from RDY to ERR failed: %d\n", error);
1524 /* wait till SQ is empty */
1525 mtx_lock(&sq->lock);
1526 while (sq->cc != sq->pc &&
1527 mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1528 mtx_unlock(&sq->lock);
1530 sq->cq.mcq.comp(&sq->cq.mcq);
1531 mtx_lock(&sq->lock);
1533 mtx_unlock(&sq->lock);
1537 mlx5e_close_sq_wait(struct mlx5e_sq *sq)
1541 mlx5e_disable_sq(sq);
1542 mlx5e_destroy_sq(sq);
1546 mlx5e_create_cq(struct mlx5e_priv *priv,
1547 struct mlx5e_cq_param *param,
1548 struct mlx5e_cq *cq,
1549 mlx5e_cq_comp_t *comp,
1552 struct mlx5_core_dev *mdev = priv->mdev;
1553 struct mlx5_core_cq *mcq = &cq->mcq;
1559 param->wq.buf_numa_node = 0;
1560 param->wq.db_numa_node = 0;
1562 err = mlx5_cqwq_create(mdev, ¶m->wq, param->cqc, &cq->wq,
1567 mlx5_vector2eqn(mdev, eq_ix, &eqn_not_used, &irqn);
1570 mcq->set_ci_db = cq->wq_ctrl.db.db;
1571 mcq->arm_db = cq->wq_ctrl.db.db + 1;
1572 *mcq->set_ci_db = 0;
1574 mcq->vector = eq_ix;
1576 mcq->event = mlx5e_cq_error_event;
1578 mcq->uar = &priv->cq_uar;
1580 for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) {
1581 struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i);
1592 mlx5e_destroy_cq(struct mlx5e_cq *cq)
1594 mlx5_wq_destroy(&cq->wq_ctrl);
1598 mlx5e_enable_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param, int eq_ix)
1600 struct mlx5_core_cq *mcq = &cq->mcq;
1608 inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
1609 sizeof(u64) * cq->wq_ctrl.buf.npages;
1610 in = mlx5_vzalloc(inlen);
1614 cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
1616 memcpy(cqc, param->cqc, sizeof(param->cqc));
1618 mlx5_fill_page_array(&cq->wq_ctrl.buf,
1619 (__be64 *) MLX5_ADDR_OF(create_cq_in, in, pas));
1621 mlx5_vector2eqn(cq->priv->mdev, eq_ix, &eqn, &irqn_not_used);
1623 MLX5_SET(cqc, cqc, c_eqn, eqn);
1624 MLX5_SET(cqc, cqc, uar_page, mcq->uar->index);
1625 MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.buf.page_shift -
1627 MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
1629 err = mlx5_core_create_cq(cq->priv->mdev, mcq, in, inlen);
1636 mlx5e_cq_arm(cq, MLX5_GET_DOORBELL_LOCK(&cq->priv->doorbell_lock));
1642 mlx5e_disable_cq(struct mlx5e_cq *cq)
1645 mlx5_core_destroy_cq(cq->priv->mdev, &cq->mcq);
1649 mlx5e_open_cq(struct mlx5e_priv *priv,
1650 struct mlx5e_cq_param *param,
1651 struct mlx5e_cq *cq,
1652 mlx5e_cq_comp_t *comp,
1657 err = mlx5e_create_cq(priv, param, cq, comp, eq_ix);
1661 err = mlx5e_enable_cq(cq, param, eq_ix);
1663 goto err_destroy_cq;
1668 mlx5e_destroy_cq(cq);
1674 mlx5e_close_cq(struct mlx5e_cq *cq)
1676 mlx5e_disable_cq(cq);
1677 mlx5e_destroy_cq(cq);
1681 mlx5e_open_tx_cqs(struct mlx5e_channel *c,
1682 struct mlx5e_channel_param *cparam)
1687 for (tc = 0; tc < c->num_tc; tc++) {
1688 /* open completion queue */
1689 err = mlx5e_open_cq(c->priv, &cparam->tx_cq, &c->sq[tc].cq,
1690 &mlx5e_tx_cq_comp, c->ix);
1692 goto err_close_tx_cqs;
1697 for (tc--; tc >= 0; tc--)
1698 mlx5e_close_cq(&c->sq[tc].cq);
1704 mlx5e_close_tx_cqs(struct mlx5e_channel *c)
1708 for (tc = 0; tc < c->num_tc; tc++)
1709 mlx5e_close_cq(&c->sq[tc].cq);
1713 mlx5e_open_sqs(struct mlx5e_channel *c,
1714 struct mlx5e_channel_param *cparam)
1719 for (tc = 0; tc < c->num_tc; tc++) {
1720 err = mlx5e_open_sq(c, tc, &cparam->sq, &c->sq[tc]);
1728 for (tc--; tc >= 0; tc--)
1729 mlx5e_close_sq_wait(&c->sq[tc]);
1735 mlx5e_close_sqs_wait(struct mlx5e_channel *c)
1739 for (tc = 0; tc < c->num_tc; tc++)
1740 mlx5e_close_sq_wait(&c->sq[tc]);
1744 mlx5e_chan_mtx_init(struct mlx5e_channel *c)
1748 mtx_init(&c->rq.mtx, "mlx5rx", MTX_NETWORK_LOCK, MTX_DEF);
1750 callout_init_mtx(&c->rq.watchdog, &c->rq.mtx, 0);
1752 for (tc = 0; tc < c->num_tc; tc++) {
1753 struct mlx5e_sq *sq = c->sq + tc;
1755 mtx_init(&sq->lock, "mlx5tx",
1756 MTX_NETWORK_LOCK " TX", MTX_DEF);
1757 mtx_init(&sq->comp_lock, "mlx5comp",
1758 MTX_NETWORK_LOCK " TX", MTX_DEF);
1760 callout_init_mtx(&sq->cev_callout, &sq->lock, 0);
1762 sq->cev_factor = c->priv->params_ethtool.tx_completion_fact;
1764 /* ensure the TX completion event factor is not zero */
1765 if (sq->cev_factor == 0)
1771 mlx5e_chan_mtx_destroy(struct mlx5e_channel *c)
1775 mtx_destroy(&c->rq.mtx);
1777 for (tc = 0; tc < c->num_tc; tc++) {
1778 mtx_destroy(&c->sq[tc].lock);
1779 mtx_destroy(&c->sq[tc].comp_lock);
1784 mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
1785 struct mlx5e_channel_param *cparam,
1786 struct mlx5e_channel *c)
1790 memset(c, 0, sizeof(*c));
1794 /* setup send tag */
1795 c->tag.m_snd_tag.ifp = priv->ifp;
1796 c->tag.type = IF_SND_TAG_TYPE_UNLIMITED;
1797 c->mkey_be = cpu_to_be32(priv->mr.key);
1798 c->num_tc = priv->num_tc;
1801 mlx5e_chan_mtx_init(c);
1803 /* open transmit completion queue */
1804 err = mlx5e_open_tx_cqs(c, cparam);
1808 /* open receive completion queue */
1809 err = mlx5e_open_cq(c->priv, &cparam->rx_cq, &c->rq.cq,
1810 &mlx5e_rx_cq_comp, c->ix);
1812 goto err_close_tx_cqs;
1814 err = mlx5e_open_sqs(c, cparam);
1816 goto err_close_rx_cq;
1818 err = mlx5e_open_rq(c, &cparam->rq, &c->rq);
1822 /* poll receive queue initially */
1823 c->rq.cq.mcq.comp(&c->rq.cq.mcq);
1828 mlx5e_close_sqs_wait(c);
1831 mlx5e_close_cq(&c->rq.cq);
1834 mlx5e_close_tx_cqs(c);
1837 /* destroy mutexes */
1838 mlx5e_chan_mtx_destroy(c);
1843 mlx5e_close_channel(struct mlx5e_channel *c)
1845 mlx5e_close_rq(&c->rq);
1849 mlx5e_close_channel_wait(struct mlx5e_channel *c)
1851 mlx5e_close_rq_wait(&c->rq);
1852 mlx5e_close_sqs_wait(c);
1853 mlx5e_close_tx_cqs(c);
1854 /* destroy mutexes */
1855 mlx5e_chan_mtx_destroy(c);
1859 mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs)
1863 r = priv->params.hw_lro_en ? priv->params.lro_wqe_sz :
1864 MLX5E_SW2MB_MTU(priv->ifp->if_mtu);
1865 if (r > MJUM16BYTES)
1870 else if (r > MJUMPAGESIZE)
1872 else if (r > MCLBYTES)
1878 * n + 1 must be a power of two, because stride size must be.
1879 * Stride size is 16 * (n + 1), as the first segment is
1882 for (n = howmany(r, MLX5E_MAX_RX_BYTES); !powerof2(n + 1); n++)
1891 mlx5e_build_rq_param(struct mlx5e_priv *priv,
1892 struct mlx5e_rq_param *param)
1894 void *rqc = param->rqc;
1895 void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
1898 mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
1899 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
1900 MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN);
1901 MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe) +
1902 nsegs * sizeof(struct mlx5_wqe_data_seg)));
1903 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_rq_size);
1904 MLX5_SET(wq, wq, pd, priv->pdn);
1906 param->wq.buf_numa_node = 0;
1907 param->wq.db_numa_node = 0;
1908 param->wq.linear = 1;
1912 mlx5e_build_sq_param(struct mlx5e_priv *priv,
1913 struct mlx5e_sq_param *param)
1915 void *sqc = param->sqc;
1916 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
1918 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_sq_size);
1919 MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
1920 MLX5_SET(wq, wq, pd, priv->pdn);
1922 param->wq.buf_numa_node = 0;
1923 param->wq.db_numa_node = 0;
1924 param->wq.linear = 1;
1928 mlx5e_build_common_cq_param(struct mlx5e_priv *priv,
1929 struct mlx5e_cq_param *param)
1931 void *cqc = param->cqc;
1933 MLX5_SET(cqc, cqc, uar_page, priv->cq_uar.index);
1937 mlx5e_get_default_profile(struct mlx5e_priv *priv, int mode, struct net_dim_cq_moder *ptr)
1940 *ptr = net_dim_get_profile(mode, MLX5E_DIM_DEFAULT_PROFILE);
1942 /* apply LRO restrictions */
1943 if (priv->params.hw_lro_en &&
1944 ptr->pkts > MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO) {
1945 ptr->pkts = MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO;
1950 mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
1951 struct mlx5e_cq_param *param)
1953 struct net_dim_cq_moder curr;
1954 void *cqc = param->cqc;
1957 * We use MLX5_CQE_FORMAT_HASH because the RX hash mini CQE
1958 * format is more beneficial for FreeBSD use case.
1960 * Adding support for MLX5_CQE_FORMAT_CSUM will require changes
1961 * in mlx5e_decompress_cqe.
1963 if (priv->params.cqe_zipping_en) {
1964 MLX5_SET(cqc, cqc, mini_cqe_res_format, MLX5_CQE_FORMAT_HASH);
1965 MLX5_SET(cqc, cqc, cqe_compression_en, 1);
1968 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_rq_size);
1970 switch (priv->params.rx_cq_moderation_mode) {
1972 MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
1973 MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
1974 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1977 MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
1978 MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
1979 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
1980 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
1982 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1985 mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE, &curr);
1986 MLX5_SET(cqc, cqc, cq_period, curr.usec);
1987 MLX5_SET(cqc, cqc, cq_max_count, curr.pkts);
1988 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1991 mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE, &curr);
1992 MLX5_SET(cqc, cqc, cq_period, curr.usec);
1993 MLX5_SET(cqc, cqc, cq_max_count, curr.pkts);
1994 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
1995 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
1997 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2003 mlx5e_dim_build_cq_param(priv, param);
2005 mlx5e_build_common_cq_param(priv, param);
2009 mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
2010 struct mlx5e_cq_param *param)
2012 void *cqc = param->cqc;
2014 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_sq_size);
2015 MLX5_SET(cqc, cqc, cq_period, priv->params.tx_cq_moderation_usec);
2016 MLX5_SET(cqc, cqc, cq_max_count, priv->params.tx_cq_moderation_pkts);
2018 switch (priv->params.tx_cq_moderation_mode) {
2020 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2023 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
2024 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
2026 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
2030 mlx5e_build_common_cq_param(priv, param);
2034 mlx5e_build_channel_param(struct mlx5e_priv *priv,
2035 struct mlx5e_channel_param *cparam)
2037 memset(cparam, 0, sizeof(*cparam));
2039 mlx5e_build_rq_param(priv, &cparam->rq);
2040 mlx5e_build_sq_param(priv, &cparam->sq);
2041 mlx5e_build_rx_cq_param(priv, &cparam->rx_cq);
2042 mlx5e_build_tx_cq_param(priv, &cparam->tx_cq);
2046 mlx5e_open_channels(struct mlx5e_priv *priv)
2048 struct mlx5e_channel_param cparam;
2053 mlx5e_build_channel_param(priv, &cparam);
2054 for (i = 0; i < priv->params.num_channels; i++) {
2055 err = mlx5e_open_channel(priv, i, &cparam, &priv->channel[i]);
2057 goto err_close_channels;
2060 for (j = 0; j < priv->params.num_channels; j++) {
2061 err = mlx5e_wait_for_min_rx_wqes(&priv->channel[j].rq);
2063 goto err_close_channels;
2069 mlx5e_close_channel(&priv->channel[i]);
2070 mlx5e_close_channel_wait(&priv->channel[i]);
2076 mlx5e_close_channels(struct mlx5e_priv *priv)
2080 for (i = 0; i < priv->params.num_channels; i++)
2081 mlx5e_close_channel(&priv->channel[i]);
2082 for (i = 0; i < priv->params.num_channels; i++)
2083 mlx5e_close_channel_wait(&priv->channel[i]);
2087 mlx5e_refresh_sq_params(struct mlx5e_priv *priv, struct mlx5e_sq *sq)
2090 if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
2093 switch (priv->params.tx_cq_moderation_mode) {
2096 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2099 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2103 return (mlx5_core_modify_cq_moderation_mode(priv->mdev, &sq->cq.mcq,
2104 priv->params.tx_cq_moderation_usec,
2105 priv->params.tx_cq_moderation_pkts,
2109 return (mlx5_core_modify_cq_moderation(priv->mdev, &sq->cq.mcq,
2110 priv->params.tx_cq_moderation_usec,
2111 priv->params.tx_cq_moderation_pkts));
2115 mlx5e_refresh_rq_params(struct mlx5e_priv *priv, struct mlx5e_rq *rq)
2118 if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
2123 switch (priv->params.rx_cq_moderation_mode) {
2126 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2127 dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE;
2130 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2131 dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE;
2135 /* tear down dynamic interrupt moderation */
2137 rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED;
2138 mtx_unlock(&rq->mtx);
2140 /* wait for dynamic interrupt moderation work task, if any */
2141 cancel_work_sync(&rq->dim.work);
2143 if (priv->params.rx_cq_moderation_mode >= 2) {
2144 struct net_dim_cq_moder curr;
2146 mlx5e_get_default_profile(priv, dim_mode, &curr);
2148 retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2149 curr.usec, curr.pkts, cq_mode);
2151 /* set dynamic interrupt moderation mode and zero defaults */
2153 rq->dim.mode = dim_mode;
2155 rq->dim.profile_ix = MLX5E_DIM_DEFAULT_PROFILE;
2156 mtx_unlock(&rq->mtx);
2158 retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2159 priv->params.rx_cq_moderation_usec,
2160 priv->params.rx_cq_moderation_pkts,
2166 return (mlx5_core_modify_cq_moderation(priv->mdev, &rq->cq.mcq,
2167 priv->params.rx_cq_moderation_usec,
2168 priv->params.rx_cq_moderation_pkts));
2172 mlx5e_refresh_channel_params_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
2177 err = mlx5e_refresh_rq_params(priv, &c->rq);
2181 for (i = 0; i != c->num_tc; i++) {
2182 err = mlx5e_refresh_sq_params(priv, &c->sq[i]);
2191 mlx5e_refresh_channel_params(struct mlx5e_priv *priv)
2195 /* check if channels are closed */
2196 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2199 for (i = 0; i < priv->params.num_channels; i++) {
2202 err = mlx5e_refresh_channel_params_sub(priv, &priv->channel[i]);
2210 mlx5e_open_tis(struct mlx5e_priv *priv, int tc)
2212 struct mlx5_core_dev *mdev = priv->mdev;
2213 u32 in[MLX5_ST_SZ_DW(create_tis_in)];
2214 void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
2216 memset(in, 0, sizeof(in));
2218 MLX5_SET(tisc, tisc, prio, tc);
2219 MLX5_SET(tisc, tisc, transport_domain, priv->tdn);
2221 return (mlx5_core_create_tis(mdev, in, sizeof(in), &priv->tisn[tc]));
2225 mlx5e_close_tis(struct mlx5e_priv *priv, int tc)
2227 mlx5_core_destroy_tis(priv->mdev, priv->tisn[tc]);
2231 mlx5e_open_tises(struct mlx5e_priv *priv)
2233 int num_tc = priv->num_tc;
2237 for (tc = 0; tc < num_tc; tc++) {
2238 err = mlx5e_open_tis(priv, tc);
2240 goto err_close_tises;
2246 for (tc--; tc >= 0; tc--)
2247 mlx5e_close_tis(priv, tc);
2253 mlx5e_close_tises(struct mlx5e_priv *priv)
2255 int num_tc = priv->num_tc;
2258 for (tc = 0; tc < num_tc; tc++)
2259 mlx5e_close_tis(priv, tc);
2263 mlx5e_open_rqt(struct mlx5e_priv *priv)
2265 struct mlx5_core_dev *mdev = priv->mdev;
2267 u32 out[MLX5_ST_SZ_DW(create_rqt_out)] = {0};
2274 sz = 1 << priv->params.rx_hash_log_tbl_sz;
2276 inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
2277 in = mlx5_vzalloc(inlen);
2280 rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
2282 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
2283 MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
2285 for (i = 0; i < sz; i++) {
2288 ix = rss_get_indirection_to_bucket(ix);
2290 /* ensure we don't overflow */
2291 ix %= priv->params.num_channels;
2293 /* apply receive side scaling stride, if any */
2294 ix -= ix % (int)priv->params.channels_rsss;
2296 MLX5_SET(rqtc, rqtc, rq_num[i], priv->channel[ix].rq.rqn);
2299 MLX5_SET(create_rqt_in, in, opcode, MLX5_CMD_OP_CREATE_RQT);
2301 err = mlx5_cmd_exec(mdev, in, inlen, out, sizeof(out));
2303 priv->rqtn = MLX5_GET(create_rqt_out, out, rqtn);
2311 mlx5e_close_rqt(struct mlx5e_priv *priv)
2313 u32 in[MLX5_ST_SZ_DW(destroy_rqt_in)] = {0};
2314 u32 out[MLX5_ST_SZ_DW(destroy_rqt_out)] = {0};
2316 MLX5_SET(destroy_rqt_in, in, opcode, MLX5_CMD_OP_DESTROY_RQT);
2317 MLX5_SET(destroy_rqt_in, in, rqtn, priv->rqtn);
2319 mlx5_cmd_exec(priv->mdev, in, sizeof(in), out, sizeof(out));
2323 mlx5e_build_tir_ctx(struct mlx5e_priv *priv, u32 * tirc, int tt)
2325 void *hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
2328 MLX5_SET(tirc, tirc, transport_domain, priv->tdn);
2330 #define ROUGH_MAX_L2_L3_HDR_SZ 256
2332 #define MLX5_HASH_IP (MLX5_HASH_FIELD_SEL_SRC_IP |\
2333 MLX5_HASH_FIELD_SEL_DST_IP)
2335 #define MLX5_HASH_ALL (MLX5_HASH_FIELD_SEL_SRC_IP |\
2336 MLX5_HASH_FIELD_SEL_DST_IP |\
2337 MLX5_HASH_FIELD_SEL_L4_SPORT |\
2338 MLX5_HASH_FIELD_SEL_L4_DPORT)
2340 #define MLX5_HASH_IP_IPSEC_SPI (MLX5_HASH_FIELD_SEL_SRC_IP |\
2341 MLX5_HASH_FIELD_SEL_DST_IP |\
2342 MLX5_HASH_FIELD_SEL_IPSEC_SPI)
2344 if (priv->params.hw_lro_en) {
2345 MLX5_SET(tirc, tirc, lro_enable_mask,
2346 MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO |
2347 MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO);
2348 MLX5_SET(tirc, tirc, lro_max_msg_sz,
2349 (priv->params.lro_wqe_sz -
2350 ROUGH_MAX_L2_L3_HDR_SZ) >> 8);
2351 /* TODO: add the option to choose timer value dynamically */
2352 MLX5_SET(tirc, tirc, lro_timeout_period_usecs,
2353 MLX5_CAP_ETH(priv->mdev,
2354 lro_timer_supported_periods[2]));
2357 /* setup parameters for hashing TIR type, if any */
2360 MLX5_SET(tirc, tirc, disp_type,
2361 MLX5_TIRC_DISP_TYPE_DIRECT);
2362 MLX5_SET(tirc, tirc, inline_rqn,
2363 priv->channel[0].rq.rqn);
2366 MLX5_SET(tirc, tirc, disp_type,
2367 MLX5_TIRC_DISP_TYPE_INDIRECT);
2368 MLX5_SET(tirc, tirc, indirect_table,
2370 MLX5_SET(tirc, tirc, rx_hash_fn,
2371 MLX5_TIRC_RX_HASH_FN_HASH_TOEPLITZ);
2372 hkey = (__be32 *) MLX5_ADDR_OF(tirc, tirc, rx_hash_toeplitz_key);
2375 * The FreeBSD RSS implementation does currently not
2376 * support symmetric Toeplitz hashes:
2378 MLX5_SET(tirc, tirc, rx_hash_symmetric, 0);
2379 rss_getkey((uint8_t *)hkey);
2381 MLX5_SET(tirc, tirc, rx_hash_symmetric, 1);
2382 hkey[0] = cpu_to_be32(0xD181C62C);
2383 hkey[1] = cpu_to_be32(0xF7F4DB5B);
2384 hkey[2] = cpu_to_be32(0x1983A2FC);
2385 hkey[3] = cpu_to_be32(0x943E1ADB);
2386 hkey[4] = cpu_to_be32(0xD9389E6B);
2387 hkey[5] = cpu_to_be32(0xD1039C2C);
2388 hkey[6] = cpu_to_be32(0xA74499AD);
2389 hkey[7] = cpu_to_be32(0x593D56D9);
2390 hkey[8] = cpu_to_be32(0xF3253C06);
2391 hkey[9] = cpu_to_be32(0x2ADC1FFC);
2397 case MLX5E_TT_IPV4_TCP:
2398 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2399 MLX5_L3_PROT_TYPE_IPV4);
2400 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2401 MLX5_L4_PROT_TYPE_TCP);
2403 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4)) {
2404 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2408 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2412 case MLX5E_TT_IPV6_TCP:
2413 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2414 MLX5_L3_PROT_TYPE_IPV6);
2415 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2416 MLX5_L4_PROT_TYPE_TCP);
2418 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) {
2419 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2423 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2427 case MLX5E_TT_IPV4_UDP:
2428 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2429 MLX5_L3_PROT_TYPE_IPV4);
2430 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2431 MLX5_L4_PROT_TYPE_UDP);
2433 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4)) {
2434 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2438 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2442 case MLX5E_TT_IPV6_UDP:
2443 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2444 MLX5_L3_PROT_TYPE_IPV6);
2445 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2446 MLX5_L4_PROT_TYPE_UDP);
2448 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) {
2449 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2453 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2457 case MLX5E_TT_IPV4_IPSEC_AH:
2458 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2459 MLX5_L3_PROT_TYPE_IPV4);
2460 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2461 MLX5_HASH_IP_IPSEC_SPI);
2464 case MLX5E_TT_IPV6_IPSEC_AH:
2465 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2466 MLX5_L3_PROT_TYPE_IPV6);
2467 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2468 MLX5_HASH_IP_IPSEC_SPI);
2471 case MLX5E_TT_IPV4_IPSEC_ESP:
2472 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2473 MLX5_L3_PROT_TYPE_IPV4);
2474 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2475 MLX5_HASH_IP_IPSEC_SPI);
2478 case MLX5E_TT_IPV6_IPSEC_ESP:
2479 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2480 MLX5_L3_PROT_TYPE_IPV6);
2481 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2482 MLX5_HASH_IP_IPSEC_SPI);
2486 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2487 MLX5_L3_PROT_TYPE_IPV4);
2488 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2493 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2494 MLX5_L3_PROT_TYPE_IPV6);
2495 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2505 mlx5e_open_tir(struct mlx5e_priv *priv, int tt)
2507 struct mlx5_core_dev *mdev = priv->mdev;
2513 inlen = MLX5_ST_SZ_BYTES(create_tir_in);
2514 in = mlx5_vzalloc(inlen);
2517 tirc = MLX5_ADDR_OF(create_tir_in, in, tir_context);
2519 mlx5e_build_tir_ctx(priv, tirc, tt);
2521 err = mlx5_core_create_tir(mdev, in, inlen, &priv->tirn[tt]);
2529 mlx5e_close_tir(struct mlx5e_priv *priv, int tt)
2531 mlx5_core_destroy_tir(priv->mdev, priv->tirn[tt]);
2535 mlx5e_open_tirs(struct mlx5e_priv *priv)
2540 for (i = 0; i < MLX5E_NUM_TT; i++) {
2541 err = mlx5e_open_tir(priv, i);
2543 goto err_close_tirs;
2549 for (i--; i >= 0; i--)
2550 mlx5e_close_tir(priv, i);
2556 mlx5e_close_tirs(struct mlx5e_priv *priv)
2560 for (i = 0; i < MLX5E_NUM_TT; i++)
2561 mlx5e_close_tir(priv, i);
2565 * SW MTU does not include headers,
2566 * HW MTU includes all headers and checksums.
2569 mlx5e_set_dev_port_mtu(struct ifnet *ifp, int sw_mtu)
2571 struct mlx5e_priv *priv = ifp->if_softc;
2572 struct mlx5_core_dev *mdev = priv->mdev;
2576 hw_mtu = MLX5E_SW2HW_MTU(sw_mtu);
2578 err = mlx5_set_port_mtu(mdev, hw_mtu);
2580 if_printf(ifp, "%s: mlx5_set_port_mtu failed setting %d, err=%d\n",
2581 __func__, sw_mtu, err);
2585 /* Update vport context MTU */
2586 err = mlx5_set_vport_mtu(mdev, hw_mtu);
2588 if_printf(ifp, "%s: Failed updating vport context with MTU size, err=%d\n",
2592 ifp->if_mtu = sw_mtu;
2594 err = mlx5_query_vport_mtu(mdev, &hw_mtu);
2595 if (err || !hw_mtu) {
2596 /* fallback to port oper mtu */
2597 err = mlx5_query_port_oper_mtu(mdev, &hw_mtu);
2600 if_printf(ifp, "Query port MTU, after setting new "
2601 "MTU value, failed\n");
2603 } else if (MLX5E_HW2SW_MTU(hw_mtu) < sw_mtu) {
2605 if_printf(ifp, "Port MTU %d is smaller than "
2606 "ifp mtu %d\n", hw_mtu, sw_mtu);
2607 } else if (MLX5E_HW2SW_MTU(hw_mtu) > sw_mtu) {
2609 if_printf(ifp, "Port MTU %d is bigger than "
2610 "ifp mtu %d\n", hw_mtu, sw_mtu);
2612 priv->params_ethtool.hw_mtu = hw_mtu;
2618 mlx5e_open_locked(struct ifnet *ifp)
2620 struct mlx5e_priv *priv = ifp->if_softc;
2624 /* check if already opened */
2625 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
2629 if (rss_getnumbuckets() > priv->params.num_channels) {
2630 if_printf(ifp, "NOTE: There are more RSS buckets(%u) than "
2631 "channels(%u) available\n", rss_getnumbuckets(),
2632 priv->params.num_channels);
2635 err = mlx5e_open_tises(priv);
2637 if_printf(ifp, "%s: mlx5e_open_tises failed, %d\n",
2641 err = mlx5_vport_alloc_q_counter(priv->mdev,
2642 MLX5_INTERFACE_PROTOCOL_ETH, &set_id);
2644 if_printf(priv->ifp,
2645 "%s: mlx5_vport_alloc_q_counter failed: %d\n",
2647 goto err_close_tises;
2649 /* store counter set ID */
2650 priv->counter_set_id = set_id;
2652 err = mlx5e_open_channels(priv);
2654 if_printf(ifp, "%s: mlx5e_open_channels failed, %d\n",
2656 goto err_dalloc_q_counter;
2658 err = mlx5e_open_rqt(priv);
2660 if_printf(ifp, "%s: mlx5e_open_rqt failed, %d\n",
2662 goto err_close_channels;
2664 err = mlx5e_open_tirs(priv);
2666 if_printf(ifp, "%s: mlx5e_open_tir failed, %d\n",
2668 goto err_close_rqls;
2670 err = mlx5e_open_flow_table(priv);
2672 if_printf(ifp, "%s: mlx5e_open_flow_table failed, %d\n",
2674 goto err_close_tirs;
2676 err = mlx5e_add_all_vlan_rules(priv);
2678 if_printf(ifp, "%s: mlx5e_add_all_vlan_rules failed, %d\n",
2680 goto err_close_flow_table;
2682 set_bit(MLX5E_STATE_OPENED, &priv->state);
2684 mlx5e_update_carrier(priv);
2685 mlx5e_set_rx_mode_core(priv);
2689 err_close_flow_table:
2690 mlx5e_close_flow_table(priv);
2693 mlx5e_close_tirs(priv);
2696 mlx5e_close_rqt(priv);
2699 mlx5e_close_channels(priv);
2701 err_dalloc_q_counter:
2702 mlx5_vport_dealloc_q_counter(priv->mdev,
2703 MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
2706 mlx5e_close_tises(priv);
2712 mlx5e_open(void *arg)
2714 struct mlx5e_priv *priv = arg;
2717 if (mlx5_set_port_status(priv->mdev, MLX5_PORT_UP))
2718 if_printf(priv->ifp,
2719 "%s: Setting port status to up failed\n",
2722 mlx5e_open_locked(priv->ifp);
2723 priv->ifp->if_drv_flags |= IFF_DRV_RUNNING;
2728 mlx5e_close_locked(struct ifnet *ifp)
2730 struct mlx5e_priv *priv = ifp->if_softc;
2732 /* check if already closed */
2733 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2736 clear_bit(MLX5E_STATE_OPENED, &priv->state);
2738 mlx5e_set_rx_mode_core(priv);
2739 mlx5e_del_all_vlan_rules(priv);
2740 if_link_state_change(priv->ifp, LINK_STATE_DOWN);
2741 mlx5e_close_flow_table(priv);
2742 mlx5e_close_tirs(priv);
2743 mlx5e_close_rqt(priv);
2744 mlx5e_close_channels(priv);
2745 mlx5_vport_dealloc_q_counter(priv->mdev,
2746 MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
2747 mlx5e_close_tises(priv);
2752 #if (__FreeBSD_version >= 1100000)
2754 mlx5e_get_counter(struct ifnet *ifp, ift_counter cnt)
2756 struct mlx5e_priv *priv = ifp->if_softc;
2759 /* PRIV_LOCK(priv); XXX not allowed */
2761 case IFCOUNTER_IPACKETS:
2762 retval = priv->stats.vport.rx_packets;
2764 case IFCOUNTER_IERRORS:
2765 retval = priv->stats.pport.in_range_len_errors +
2766 priv->stats.pport.out_of_range_len +
2767 priv->stats.pport.too_long_errors +
2768 priv->stats.pport.check_seq_err +
2769 priv->stats.pport.alignment_err;
2771 case IFCOUNTER_IQDROPS:
2772 retval = priv->stats.vport.rx_out_of_buffer;
2774 case IFCOUNTER_OPACKETS:
2775 retval = priv->stats.vport.tx_packets;
2777 case IFCOUNTER_OERRORS:
2778 retval = priv->stats.port_stats_debug.out_discards;
2780 case IFCOUNTER_IBYTES:
2781 retval = priv->stats.vport.rx_bytes;
2783 case IFCOUNTER_OBYTES:
2784 retval = priv->stats.vport.tx_bytes;
2786 case IFCOUNTER_IMCASTS:
2787 retval = priv->stats.vport.rx_multicast_packets;
2789 case IFCOUNTER_OMCASTS:
2790 retval = priv->stats.vport.tx_multicast_packets;
2792 case IFCOUNTER_OQDROPS:
2793 retval = priv->stats.vport.tx_queue_dropped;
2795 case IFCOUNTER_COLLISIONS:
2796 retval = priv->stats.pport.collisions;
2799 retval = if_get_counter_default(ifp, cnt);
2802 /* PRIV_UNLOCK(priv); XXX not allowed */
2808 mlx5e_set_rx_mode(struct ifnet *ifp)
2810 struct mlx5e_priv *priv = ifp->if_softc;
2812 queue_work(priv->wq, &priv->set_rx_mode_work);
2816 mlx5e_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2818 struct mlx5e_priv *priv;
2820 struct ifi2creq i2c;
2829 priv = ifp->if_softc;
2831 /* check if detaching */
2832 if (priv == NULL || priv->gone != 0)
2837 ifr = (struct ifreq *)data;
2840 mlx5_query_port_max_mtu(priv->mdev, &max_mtu);
2842 if (ifr->ifr_mtu >= MLX5E_MTU_MIN &&
2843 ifr->ifr_mtu <= MIN(MLX5E_MTU_MAX, max_mtu)) {
2846 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2848 mlx5e_close_locked(ifp);
2851 mlx5e_set_dev_port_mtu(ifp, ifr->ifr_mtu);
2854 mlx5e_open_locked(ifp);
2857 if_printf(ifp, "Invalid MTU value. Min val: %d, Max val: %d\n",
2858 MLX5E_MTU_MIN, MIN(MLX5E_MTU_MAX, max_mtu));
2863 if ((ifp->if_flags & IFF_UP) &&
2864 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
2865 mlx5e_set_rx_mode(ifp);
2869 if (ifp->if_flags & IFF_UP) {
2870 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2871 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2872 mlx5e_open_locked(ifp);
2873 ifp->if_drv_flags |= IFF_DRV_RUNNING;
2874 mlx5_set_port_status(priv->mdev, MLX5_PORT_UP);
2877 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
2878 mlx5_set_port_status(priv->mdev,
2880 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
2881 mlx5e_close_locked(ifp);
2882 mlx5e_update_carrier(priv);
2883 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2890 mlx5e_set_rx_mode(ifp);
2895 ifr = (struct ifreq *)data;
2896 error = ifmedia_ioctl(ifp, ifr, &priv->media, command);
2899 ifr = (struct ifreq *)data;
2901 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2903 if (mask & IFCAP_TXCSUM) {
2904 ifp->if_capenable ^= IFCAP_TXCSUM;
2905 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
2907 if (IFCAP_TSO4 & ifp->if_capenable &&
2908 !(IFCAP_TXCSUM & ifp->if_capenable)) {
2909 ifp->if_capenable &= ~IFCAP_TSO4;
2910 ifp->if_hwassist &= ~CSUM_IP_TSO;
2912 "tso4 disabled due to -txcsum.\n");
2915 if (mask & IFCAP_TXCSUM_IPV6) {
2916 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
2917 ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
2919 if (IFCAP_TSO6 & ifp->if_capenable &&
2920 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2921 ifp->if_capenable &= ~IFCAP_TSO6;
2922 ifp->if_hwassist &= ~CSUM_IP6_TSO;
2924 "tso6 disabled due to -txcsum6.\n");
2927 if (mask & IFCAP_RXCSUM)
2928 ifp->if_capenable ^= IFCAP_RXCSUM;
2929 if (mask & IFCAP_RXCSUM_IPV6)
2930 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
2931 if (mask & IFCAP_TSO4) {
2932 if (!(IFCAP_TSO4 & ifp->if_capenable) &&
2933 !(IFCAP_TXCSUM & ifp->if_capenable)) {
2934 if_printf(ifp, "enable txcsum first.\n");
2938 ifp->if_capenable ^= IFCAP_TSO4;
2939 ifp->if_hwassist ^= CSUM_IP_TSO;
2941 if (mask & IFCAP_TSO6) {
2942 if (!(IFCAP_TSO6 & ifp->if_capenable) &&
2943 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2944 if_printf(ifp, "enable txcsum6 first.\n");
2948 ifp->if_capenable ^= IFCAP_TSO6;
2949 ifp->if_hwassist ^= CSUM_IP6_TSO;
2951 if (mask & IFCAP_VLAN_HWFILTER) {
2952 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2953 mlx5e_disable_vlan_filter(priv);
2955 mlx5e_enable_vlan_filter(priv);
2957 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
2959 if (mask & IFCAP_VLAN_HWTAGGING)
2960 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2961 if (mask & IFCAP_WOL_MAGIC)
2962 ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2964 VLAN_CAPABILITIES(ifp);
2965 /* turn off LRO means also turn of HW LRO - if it's on */
2966 if (mask & IFCAP_LRO) {
2967 int was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2968 bool need_restart = false;
2970 ifp->if_capenable ^= IFCAP_LRO;
2972 /* figure out if updating HW LRO is needed */
2973 if (!(ifp->if_capenable & IFCAP_LRO)) {
2974 if (priv->params.hw_lro_en) {
2975 priv->params.hw_lro_en = false;
2976 need_restart = true;
2979 if (priv->params.hw_lro_en == false &&
2980 priv->params_ethtool.hw_lro != 0) {
2981 priv->params.hw_lro_en = true;
2982 need_restart = true;
2985 if (was_opened && need_restart) {
2986 mlx5e_close_locked(ifp);
2987 mlx5e_open_locked(ifp);
2990 if (mask & IFCAP_HWRXTSTMP) {
2991 ifp->if_capenable ^= IFCAP_HWRXTSTMP;
2992 if (ifp->if_capenable & IFCAP_HWRXTSTMP) {
2993 if (priv->clbr_done == 0)
2994 mlx5e_reset_calibration_callout(priv);
2996 callout_drain(&priv->tstmp_clbr);
2997 priv->clbr_done = 0;
3005 ifr = (struct ifreq *)data;
3008 * Copy from the user-space address ifr_data to the
3009 * kernel-space address i2c
3011 error = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
3015 if (i2c.len > sizeof(i2c.data)) {
3021 /* Get module_num which is required for the query_eeprom */
3022 error = mlx5_query_module_num(priv->mdev, &module_num);
3024 if_printf(ifp, "Query module num failed, eeprom "
3025 "reading is not supported\n");
3029 /* Check if module is present before doing an access */
3030 module_status = mlx5_query_module_status(priv->mdev, module_num);
3031 if (module_status != MLX5_MODULE_STATUS_PLUGGED_ENABLED &&
3032 module_status != MLX5_MODULE_STATUS_PLUGGED_DISABLED) {
3037 * Currently 0XA0 and 0xA2 are the only addresses permitted.
3038 * The internal conversion is as follows:
3040 if (i2c.dev_addr == 0xA0)
3041 read_addr = MLX5E_I2C_ADDR_LOW;
3042 else if (i2c.dev_addr == 0xA2)
3043 read_addr = MLX5E_I2C_ADDR_HIGH;
3045 if_printf(ifp, "Query eeprom failed, "
3046 "Invalid Address: %X\n", i2c.dev_addr);
3050 error = mlx5_query_eeprom(priv->mdev,
3051 read_addr, MLX5E_EEPROM_LOW_PAGE,
3052 (uint32_t)i2c.offset, (uint32_t)i2c.len, module_num,
3053 (uint32_t *)i2c.data, &size_read);
3055 if_printf(ifp, "Query eeprom failed, eeprom "
3056 "reading is not supported\n");
3061 if (i2c.len > MLX5_EEPROM_MAX_BYTES) {
3062 error = mlx5_query_eeprom(priv->mdev,
3063 read_addr, MLX5E_EEPROM_LOW_PAGE,
3064 (uint32_t)(i2c.offset + size_read),
3065 (uint32_t)(i2c.len - size_read), module_num,
3066 (uint32_t *)(i2c.data + size_read), &size_read);
3069 if_printf(ifp, "Query eeprom failed, eeprom "
3070 "reading is not supported\n");
3075 error = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
3081 error = ether_ioctl(ifp, command, data);
3088 mlx5e_check_required_hca_cap(struct mlx5_core_dev *mdev)
3091 * TODO: uncoment once FW really sets all these bits if
3092 * (!mdev->caps.eth.rss_ind_tbl_cap || !mdev->caps.eth.csum_cap ||
3093 * !mdev->caps.eth.max_lso_cap || !mdev->caps.eth.vlan_cap ||
3094 * !(mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_SCQE_BRK_MOD)) return
3098 /* TODO: add more must-to-have features */
3100 if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
3107 mlx5e_get_max_inline_cap(struct mlx5_core_dev *mdev)
3109 uint32_t bf_buf_size = (1U << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2U;
3111 bf_buf_size -= sizeof(struct mlx5e_tx_wqe) - 2;
3113 /* verify against driver hardware limit */
3114 if (bf_buf_size > MLX5E_MAX_TX_INLINE)
3115 bf_buf_size = MLX5E_MAX_TX_INLINE;
3117 return (bf_buf_size);
3121 mlx5e_build_ifp_priv(struct mlx5_core_dev *mdev,
3122 struct mlx5e_priv *priv,
3123 int num_comp_vectors)
3128 * TODO: Consider link speed for setting "log_sq_size",
3129 * "log_rq_size" and "cq_moderation_xxx":
3131 priv->params.log_sq_size =
3132 MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
3133 priv->params.log_rq_size =
3134 MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
3135 priv->params.rx_cq_moderation_usec =
3136 MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
3137 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE :
3138 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
3139 priv->params.rx_cq_moderation_mode =
3140 MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ? 1 : 0;
3141 priv->params.rx_cq_moderation_pkts =
3142 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS;
3143 priv->params.tx_cq_moderation_usec =
3144 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC;
3145 priv->params.tx_cq_moderation_pkts =
3146 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
3147 priv->params.min_rx_wqes =
3148 MLX5E_PARAMS_DEFAULT_MIN_RX_WQES;
3149 priv->params.rx_hash_log_tbl_sz =
3150 (order_base_2(num_comp_vectors) >
3151 MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ) ?
3152 order_base_2(num_comp_vectors) :
3153 MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ;
3154 priv->params.num_tc = 1;
3155 priv->params.default_vlan_prio = 0;
3156 priv->counter_set_id = -1;
3157 priv->params.tx_max_inline = mlx5e_get_max_inline_cap(mdev);
3159 err = mlx5_query_min_inline(mdev, &priv->params.tx_min_inline_mode);
3164 * hw lro is currently defaulted to off. when it won't anymore we
3165 * will consider the HW capability: "!!MLX5_CAP_ETH(mdev, lro_cap)"
3167 priv->params.hw_lro_en = false;
3168 priv->params.lro_wqe_sz = MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ;
3171 * CQE zipping is currently defaulted to off. when it won't
3172 * anymore we will consider the HW capability:
3173 * "!!MLX5_CAP_GEN(mdev, cqe_compression)"
3175 priv->params.cqe_zipping_en = false;
3178 priv->params.num_channels = num_comp_vectors;
3179 priv->params.channels_rsss = 1;
3180 priv->order_base_2_num_channels = order_base_2(num_comp_vectors);
3181 priv->queue_mapping_channel_mask =
3182 roundup_pow_of_two(num_comp_vectors) - 1;
3183 priv->num_tc = priv->params.num_tc;
3184 priv->default_vlan_prio = priv->params.default_vlan_prio;
3186 INIT_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
3187 INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
3188 INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
3194 mlx5e_create_mkey(struct mlx5e_priv *priv, u32 pdn,
3195 struct mlx5_core_mr *mkey)
3197 struct ifnet *ifp = priv->ifp;
3198 struct mlx5_core_dev *mdev = priv->mdev;
3199 int inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
3204 in = mlx5_vzalloc(inlen);
3206 if_printf(ifp, "%s: failed to allocate inbox\n", __func__);
3210 mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
3211 MLX5_SET(mkc, mkc, access_mode, MLX5_ACCESS_MODE_PA);
3212 MLX5_SET(mkc, mkc, lw, 1);
3213 MLX5_SET(mkc, mkc, lr, 1);
3215 MLX5_SET(mkc, mkc, pd, pdn);
3216 MLX5_SET(mkc, mkc, length64, 1);
3217 MLX5_SET(mkc, mkc, qpn, 0xffffff);
3219 err = mlx5_core_create_mkey(mdev, mkey, in, inlen);
3221 if_printf(ifp, "%s: mlx5_core_create_mkey failed, %d\n",
3228 static const char *mlx5e_vport_stats_desc[] = {
3229 MLX5E_VPORT_STATS(MLX5E_STATS_DESC)
3232 static const char *mlx5e_pport_stats_desc[] = {
3233 MLX5E_PPORT_STATS(MLX5E_STATS_DESC)
3237 mlx5e_priv_mtx_init(struct mlx5e_priv *priv)
3239 mtx_init(&priv->async_events_mtx, "mlx5async", MTX_NETWORK_LOCK, MTX_DEF);
3240 sx_init(&priv->state_lock, "mlx5state");
3241 callout_init_mtx(&priv->watchdog, &priv->async_events_mtx, 0);
3242 MLX5_INIT_DOORBELL_LOCK(&priv->doorbell_lock);
3246 mlx5e_priv_mtx_destroy(struct mlx5e_priv *priv)
3248 mtx_destroy(&priv->async_events_mtx);
3249 sx_destroy(&priv->state_lock);
3253 sysctl_firmware(SYSCTL_HANDLER_ARGS)
3256 * %d.%d%.d the string format.
3257 * fw_rev_{maj,min,sub} return u16, 2^16 = 65536.
3258 * We need at most 5 chars to store that.
3259 * It also has: two "." and NULL at the end, which means we need 18
3260 * (5*3 + 3) chars at most.
3263 struct mlx5e_priv *priv = arg1;
3266 snprintf(fw, sizeof(fw), "%d.%d.%d", fw_rev_maj(priv->mdev), fw_rev_min(priv->mdev),
3267 fw_rev_sub(priv->mdev));
3268 error = sysctl_handle_string(oidp, fw, sizeof(fw), req);
3273 mlx5e_disable_tx_dma(struct mlx5e_channel *ch)
3277 for (i = 0; i < ch->num_tc; i++)
3278 mlx5e_drain_sq(&ch->sq[i]);
3282 mlx5e_reset_sq_doorbell_record(struct mlx5e_sq *sq)
3285 sq->doorbell.d32[0] = cpu_to_be32(MLX5_OPCODE_NOP);
3286 sq->doorbell.d32[1] = cpu_to_be32(sq->sqn << 8);
3287 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
3288 sq->doorbell.d64 = 0;
3292 mlx5e_resume_sq(struct mlx5e_sq *sq)
3296 /* check if already enabled */
3297 if (READ_ONCE(sq->running) != 0)
3300 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_ERR,
3301 MLX5_SQC_STATE_RST);
3304 "mlx5e_modify_sq() from ERR to RST failed: %d\n", err);
3310 /* reset doorbell prior to moving from RST to RDY */
3311 mlx5e_reset_sq_doorbell_record(sq);
3313 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST,
3314 MLX5_SQC_STATE_RDY);
3317 "mlx5e_modify_sq() from RST to RDY failed: %d\n", err);
3320 sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
3321 WRITE_ONCE(sq->running, 1);
3325 mlx5e_enable_tx_dma(struct mlx5e_channel *ch)
3329 for (i = 0; i < ch->num_tc; i++)
3330 mlx5e_resume_sq(&ch->sq[i]);
3334 mlx5e_disable_rx_dma(struct mlx5e_channel *ch)
3336 struct mlx5e_rq *rq = &ch->rq;
3341 callout_stop(&rq->watchdog);
3342 mtx_unlock(&rq->mtx);
3344 callout_drain(&rq->watchdog);
3346 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
3349 "mlx5e_modify_rq() from RDY to RST failed: %d\n", err);
3352 while (!mlx5_wq_ll_is_empty(&rq->wq)) {
3354 rq->cq.mcq.comp(&rq->cq.mcq);
3358 * Transitioning into RST state will allow the FW to track less ERR state queues,
3359 * thus reducing the recv queue flushing time
3361 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_ERR, MLX5_RQC_STATE_RST);
3364 "mlx5e_modify_rq() from ERR to RST failed: %d\n", err);
3369 mlx5e_enable_rx_dma(struct mlx5e_channel *ch)
3371 struct mlx5e_rq *rq = &ch->rq;
3375 mlx5_wq_ll_update_db_record(&rq->wq);
3376 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
3379 "mlx5e_modify_rq() from RST to RDY failed: %d\n", err);
3384 rq->cq.mcq.comp(&rq->cq.mcq);
3388 mlx5e_modify_tx_dma(struct mlx5e_priv *priv, uint8_t value)
3392 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3395 for (i = 0; i < priv->params.num_channels; i++) {
3397 mlx5e_disable_tx_dma(&priv->channel[i]);
3399 mlx5e_enable_tx_dma(&priv->channel[i]);
3404 mlx5e_modify_rx_dma(struct mlx5e_priv *priv, uint8_t value)
3408 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
3411 for (i = 0; i < priv->params.num_channels; i++) {
3413 mlx5e_disable_rx_dma(&priv->channel[i]);
3415 mlx5e_enable_rx_dma(&priv->channel[i]);
3420 mlx5e_add_hw_stats(struct mlx5e_priv *priv)
3422 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3423 OID_AUTO, "fw_version", CTLTYPE_STRING | CTLFLAG_RD, priv, 0,
3424 sysctl_firmware, "A", "HCA firmware version");
3426 SYSCTL_ADD_STRING(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3427 OID_AUTO, "board_id", CTLFLAG_RD, priv->mdev->board_id, 0,
3432 mlx5e_sysctl_tx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3434 struct mlx5e_priv *priv = arg1;
3435 uint8_t temp[MLX5E_MAX_PRIORITY];
3442 tx_pfc = priv->params.tx_priority_flow_control;
3444 for (i = 0; i != MLX5E_MAX_PRIORITY; i++)
3445 temp[i] = (tx_pfc >> i) & 1;
3447 err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY);
3448 if (err || !req->newptr)
3450 err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY);
3454 priv->params.tx_priority_flow_control = 0;
3456 /* range check input value */
3457 for (i = 0; i != MLX5E_MAX_PRIORITY; i++) {
3462 priv->params.tx_priority_flow_control |= (temp[i] << i);
3465 /* check if update is required */
3466 if (tx_pfc != priv->params.tx_priority_flow_control)
3467 err = -mlx5e_set_port_pfc(priv);
3470 priv->params.tx_priority_flow_control= tx_pfc;
3477 mlx5e_sysctl_rx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3479 struct mlx5e_priv *priv = arg1;
3480 uint8_t temp[MLX5E_MAX_PRIORITY];
3487 rx_pfc = priv->params.rx_priority_flow_control;
3489 for (i = 0; i != MLX5E_MAX_PRIORITY; i++)
3490 temp[i] = (rx_pfc >> i) & 1;
3492 err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY);
3493 if (err || !req->newptr)
3495 err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY);
3499 priv->params.rx_priority_flow_control = 0;
3501 /* range check input value */
3502 for (i = 0; i != MLX5E_MAX_PRIORITY; i++) {
3507 priv->params.rx_priority_flow_control |= (temp[i] << i);
3510 /* check if update is required */
3511 if (rx_pfc != priv->params.rx_priority_flow_control)
3512 err = -mlx5e_set_port_pfc(priv);
3515 priv->params.rx_priority_flow_control= rx_pfc;
3522 mlx5e_setup_pauseframes(struct mlx5e_priv *priv)
3524 #if (__FreeBSD_version < 1100000)
3529 /* enable pauseframes by default */
3530 priv->params.tx_pauseframe_control = 1;
3531 priv->params.rx_pauseframe_control = 1;
3533 /* disable ports flow control, PFC, by default */
3534 priv->params.tx_priority_flow_control = 0;
3535 priv->params.rx_priority_flow_control = 0;
3537 #if (__FreeBSD_version < 1100000)
3538 /* compute path for sysctl */
3539 snprintf(path, sizeof(path), "dev.mce.%d.tx_pauseframe_control",
3540 device_get_unit(priv->mdev->pdev->dev.bsddev));
3542 /* try to fetch tunable, if any */
3543 TUNABLE_INT_FETCH(path, &priv->params.tx_pauseframe_control);
3545 /* compute path for sysctl */
3546 snprintf(path, sizeof(path), "dev.mce.%d.rx_pauseframe_control",
3547 device_get_unit(priv->mdev->pdev->dev.bsddev));
3549 /* try to fetch tunable, if any */
3550 TUNABLE_INT_FETCH(path, &priv->params.rx_pauseframe_control);
3553 /* register pauseframe SYSCTLs */
3554 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3555 OID_AUTO, "tx_pauseframe_control", CTLFLAG_RDTUN,
3556 &priv->params.tx_pauseframe_control, 0,
3557 "Set to enable TX pause frames. Clear to disable.");
3559 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3560 OID_AUTO, "rx_pauseframe_control", CTLFLAG_RDTUN,
3561 &priv->params.rx_pauseframe_control, 0,
3562 "Set to enable RX pause frames. Clear to disable.");
3564 /* register priority flow control, PFC, SYSCTLs */
3565 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3566 OID_AUTO, "tx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN |
3567 CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_tx_priority_flow_control, "CU",
3568 "Set to enable TX ports flow control frames for priorities 0..7. Clear to disable.");
3570 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3571 OID_AUTO, "rx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN |
3572 CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_rx_priority_flow_control, "CU",
3573 "Set to enable RX ports flow control frames for priorities 0..7. Clear to disable.");
3578 priv->params.tx_pauseframe_control =
3579 priv->params.tx_pauseframe_control ? 1 : 0;
3580 priv->params.rx_pauseframe_control =
3581 priv->params.rx_pauseframe_control ? 1 : 0;
3583 /* update firmware */
3584 error = mlx5e_set_port_pause_and_pfc(priv);
3585 if (error == -EINVAL) {
3586 if_printf(priv->ifp,
3587 "Global pauseframes must be disabled before enabling PFC.\n");
3588 priv->params.rx_priority_flow_control = 0;
3589 priv->params.tx_priority_flow_control = 0;
3591 /* update firmware */
3592 (void) mlx5e_set_port_pause_and_pfc(priv);
3598 mlx5e_ul_snd_tag_alloc(struct ifnet *ifp,
3599 union if_snd_tag_alloc_params *params,
3600 struct m_snd_tag **ppmt)
3602 struct mlx5e_priv *priv;
3603 struct mlx5e_channel *pch;
3605 priv = ifp->if_softc;
3607 if (unlikely(priv->gone || params->hdr.flowtype == M_HASHTYPE_NONE)) {
3608 return (EOPNOTSUPP);
3610 /* keep this code synced with mlx5e_select_queue() */
3611 u32 ch = priv->params.num_channels;
3615 if (rss_hash2bucket(params->hdr.flowid,
3616 params->hdr.flowtype, &temp) == 0)
3620 ch = (params->hdr.flowid % 128) % ch;
3623 * NOTE: The channels array is only freed at detach
3624 * and it safe to return a pointer to the send tag
3625 * inside the channels structure as long as we
3626 * reference the priv.
3628 pch = priv->channel + ch;
3630 /* check if send queue is not running */
3631 if (unlikely(pch->sq[0].running == 0))
3633 mlx5e_ref_channel(priv);
3634 *ppmt = &pch->tag.m_snd_tag;
3640 mlx5e_ul_snd_tag_query(struct m_snd_tag *pmt, union if_snd_tag_query_params *params)
3642 struct mlx5e_channel *pch =
3643 container_of(pmt, struct mlx5e_channel, tag.m_snd_tag);
3645 params->unlimited.max_rate = -1ULL;
3646 params->unlimited.queue_level = mlx5e_sq_queue_level(&pch->sq[0]);
3651 mlx5e_ul_snd_tag_free(struct m_snd_tag *pmt)
3653 struct mlx5e_channel *pch =
3654 container_of(pmt, struct mlx5e_channel, tag.m_snd_tag);
3656 mlx5e_unref_channel(pch->priv);
3660 mlx5e_snd_tag_alloc(struct ifnet *ifp,
3661 union if_snd_tag_alloc_params *params,
3662 struct m_snd_tag **ppmt)
3665 switch (params->hdr.type) {
3667 case IF_SND_TAG_TYPE_RATE_LIMIT:
3668 return (mlx5e_rl_snd_tag_alloc(ifp, params, ppmt));
3670 case IF_SND_TAG_TYPE_UNLIMITED:
3671 return (mlx5e_ul_snd_tag_alloc(ifp, params, ppmt));
3673 return (EOPNOTSUPP);
3678 mlx5e_snd_tag_modify(struct m_snd_tag *pmt, union if_snd_tag_modify_params *params)
3680 struct mlx5e_snd_tag *tag =
3681 container_of(pmt, struct mlx5e_snd_tag, m_snd_tag);
3683 switch (tag->type) {
3685 case IF_SND_TAG_TYPE_RATE_LIMIT:
3686 return (mlx5e_rl_snd_tag_modify(pmt, params));
3688 case IF_SND_TAG_TYPE_UNLIMITED:
3690 return (EOPNOTSUPP);
3695 mlx5e_snd_tag_query(struct m_snd_tag *pmt, union if_snd_tag_query_params *params)
3697 struct mlx5e_snd_tag *tag =
3698 container_of(pmt, struct mlx5e_snd_tag, m_snd_tag);
3700 switch (tag->type) {
3702 case IF_SND_TAG_TYPE_RATE_LIMIT:
3703 return (mlx5e_rl_snd_tag_query(pmt, params));
3705 case IF_SND_TAG_TYPE_UNLIMITED:
3706 return (mlx5e_ul_snd_tag_query(pmt, params));
3708 return (EOPNOTSUPP);
3713 mlx5e_snd_tag_free(struct m_snd_tag *pmt)
3715 struct mlx5e_snd_tag *tag =
3716 container_of(pmt, struct mlx5e_snd_tag, m_snd_tag);
3718 switch (tag->type) {
3720 case IF_SND_TAG_TYPE_RATE_LIMIT:
3721 mlx5e_rl_snd_tag_free(pmt);
3724 case IF_SND_TAG_TYPE_UNLIMITED:
3725 mlx5e_ul_snd_tag_free(pmt);
3733 mlx5e_create_ifp(struct mlx5_core_dev *mdev)
3736 struct mlx5e_priv *priv;
3737 u8 dev_addr[ETHER_ADDR_LEN] __aligned(4);
3738 struct sysctl_oid_list *child;
3739 int ncv = mdev->priv.eq_table.num_comp_vectors;
3745 if (mlx5e_check_required_hca_cap(mdev)) {
3746 mlx5_core_dbg(mdev, "mlx5e_check_required_hca_cap() failed\n");
3750 * Try to allocate the priv and make room for worst-case
3751 * number of channel structures:
3753 priv = malloc(sizeof(*priv) +
3754 (sizeof(priv->channel[0]) * mdev->priv.eq_table.num_comp_vectors),
3755 M_MLX5EN, M_WAITOK | M_ZERO);
3756 mlx5e_priv_mtx_init(priv);
3758 ifp = priv->ifp = if_alloc(IFT_ETHER);
3760 mlx5_core_err(mdev, "if_alloc() failed\n");
3763 ifp->if_softc = priv;
3764 if_initname(ifp, "mce", device_get_unit(mdev->pdev->dev.bsddev));
3765 ifp->if_mtu = ETHERMTU;
3766 ifp->if_init = mlx5e_open;
3767 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
3768 ifp->if_ioctl = mlx5e_ioctl;
3769 ifp->if_transmit = mlx5e_xmit;
3770 ifp->if_qflush = if_qflush;
3771 #if (__FreeBSD_version >= 1100000)
3772 ifp->if_get_counter = mlx5e_get_counter;
3774 ifp->if_snd.ifq_maxlen = ifqmaxlen;
3776 * Set driver features
3778 ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6;
3779 ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
3780 ifp->if_capabilities |= IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWFILTER;
3781 ifp->if_capabilities |= IFCAP_LINKSTATE | IFCAP_JUMBO_MTU;
3782 ifp->if_capabilities |= IFCAP_LRO;
3783 ifp->if_capabilities |= IFCAP_TSO | IFCAP_VLAN_HWTSO;
3784 ifp->if_capabilities |= IFCAP_HWSTATS | IFCAP_HWRXTSTMP;
3785 ifp->if_capabilities |= IFCAP_TXRTLMT;
3786 ifp->if_snd_tag_alloc = mlx5e_snd_tag_alloc;
3787 ifp->if_snd_tag_free = mlx5e_snd_tag_free;
3788 ifp->if_snd_tag_modify = mlx5e_snd_tag_modify;
3789 ifp->if_snd_tag_query = mlx5e_snd_tag_query;
3791 /* set TSO limits so that we don't have to drop TX packets */
3792 ifp->if_hw_tsomax = MLX5E_MAX_TX_PAYLOAD_SIZE - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
3793 ifp->if_hw_tsomaxsegcount = MLX5E_MAX_TX_MBUF_FRAGS - 1 /* hdr */;
3794 ifp->if_hw_tsomaxsegsize = MLX5E_MAX_TX_MBUF_SIZE;
3796 ifp->if_capenable = ifp->if_capabilities;
3797 ifp->if_hwassist = 0;
3798 if (ifp->if_capenable & IFCAP_TSO)
3799 ifp->if_hwassist |= CSUM_TSO;
3800 if (ifp->if_capenable & IFCAP_TXCSUM)
3801 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP | CSUM_IP);
3802 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3803 ifp->if_hwassist |= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
3805 /* ifnet sysctl tree */
3806 sysctl_ctx_init(&priv->sysctl_ctx);
3807 priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_dev),
3808 OID_AUTO, ifp->if_dname, CTLFLAG_RD, 0, "MLX5 ethernet - interface name");
3809 if (priv->sysctl_ifnet == NULL) {
3810 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3811 goto err_free_sysctl;
3813 snprintf(unit, sizeof(unit), "%d", ifp->if_dunit);
3814 priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3815 OID_AUTO, unit, CTLFLAG_RD, 0, "MLX5 ethernet - interface unit");
3816 if (priv->sysctl_ifnet == NULL) {
3817 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3818 goto err_free_sysctl;
3821 /* HW sysctl tree */
3822 child = SYSCTL_CHILDREN(device_get_sysctl_tree(mdev->pdev->dev.bsddev));
3823 priv->sysctl_hw = SYSCTL_ADD_NODE(&priv->sysctl_ctx, child,
3824 OID_AUTO, "hw", CTLFLAG_RD, 0, "MLX5 ethernet dev hw");
3825 if (priv->sysctl_hw == NULL) {
3826 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3827 goto err_free_sysctl;
3830 err = mlx5e_build_ifp_priv(mdev, priv, ncv);
3832 mlx5_core_err(mdev, "mlx5e_build_ifp_priv() failed (%d)\n", err);
3833 goto err_free_sysctl;
3836 snprintf(unit, sizeof(unit), "mce%u_wq",
3837 device_get_unit(mdev->pdev->dev.bsddev));
3838 priv->wq = alloc_workqueue(unit, 0, 1);
3839 if (priv->wq == NULL) {
3840 if_printf(ifp, "%s: alloc_workqueue failed\n", __func__);
3841 goto err_free_sysctl;
3844 err = mlx5_alloc_map_uar(mdev, &priv->cq_uar);
3846 if_printf(ifp, "%s: mlx5_alloc_map_uar failed, %d\n",
3850 err = mlx5_core_alloc_pd(mdev, &priv->pdn);
3852 if_printf(ifp, "%s: mlx5_core_alloc_pd failed, %d\n",
3854 goto err_unmap_free_uar;
3856 err = mlx5_alloc_transport_domain(mdev, &priv->tdn);
3858 if_printf(ifp, "%s: mlx5_alloc_transport_domain failed, %d\n",
3860 goto err_dealloc_pd;
3862 err = mlx5e_create_mkey(priv, priv->pdn, &priv->mr);
3864 if_printf(ifp, "%s: mlx5e_create_mkey failed, %d\n",
3866 goto err_dealloc_transport_domain;
3868 mlx5_query_nic_vport_mac_address(priv->mdev, 0, dev_addr);
3870 /* check if we should generate a random MAC address */
3871 if (MLX5_CAP_GEN(priv->mdev, vport_group_manager) == 0 &&
3872 is_zero_ether_addr(dev_addr)) {
3873 random_ether_addr(dev_addr);
3874 if_printf(ifp, "Assigned random MAC address\n");
3877 err = mlx5e_rl_init(priv);
3879 if_printf(ifp, "%s: mlx5e_rl_init failed, %d\n",
3881 goto err_create_mkey;
3885 /* set default MTU */
3886 mlx5e_set_dev_port_mtu(ifp, ifp->if_mtu);
3888 /* Set default media status */
3889 priv->media_status_last = IFM_AVALID;
3890 priv->media_active_last = IFM_ETHER | IFM_AUTO |
3891 IFM_ETH_RXPAUSE | IFM_FDX;
3893 /* setup default pauseframes configuration */
3894 mlx5e_setup_pauseframes(priv);
3896 err = mlx5_query_port_proto_cap(mdev, ð_proto_cap, MLX5_PTYS_EN);
3899 if_printf(ifp, "%s: Query port media capability failed, %d\n",
3903 /* Setup supported medias */
3904 ifmedia_init(&priv->media, IFM_IMASK | IFM_ETH_FMASK,
3905 mlx5e_media_change, mlx5e_media_status);
3907 for (i = 0; i < MLX5E_LINK_MODES_NUMBER; ++i) {
3908 if (mlx5e_mode_table[i].baudrate == 0)
3910 if (MLX5E_PROT_MASK(i) & eth_proto_cap) {
3911 ifmedia_add(&priv->media,
3912 mlx5e_mode_table[i].subtype |
3913 IFM_ETHER, 0, NULL);
3914 ifmedia_add(&priv->media,
3915 mlx5e_mode_table[i].subtype |
3916 IFM_ETHER | IFM_FDX |
3917 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3921 /* Additional supported medias */
3922 ifmedia_add(&priv->media, IFM_10G_LR | IFM_ETHER, 0, NULL);
3923 ifmedia_add(&priv->media, IFM_10G_LR |
3924 IFM_ETHER | IFM_FDX |
3925 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3927 ifmedia_add(&priv->media, IFM_40G_ER4 | IFM_ETHER, 0, NULL);
3928 ifmedia_add(&priv->media, IFM_40G_ER4 |
3929 IFM_ETHER | IFM_FDX |
3930 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3932 ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO, 0, NULL);
3933 ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
3934 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3936 /* Set autoselect by default */
3937 ifmedia_set(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
3938 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE);
3939 ether_ifattach(ifp, dev_addr);
3941 /* Register for VLAN events */
3942 priv->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
3943 mlx5e_vlan_rx_add_vid, priv, EVENTHANDLER_PRI_FIRST);
3944 priv->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
3945 mlx5e_vlan_rx_kill_vid, priv, EVENTHANDLER_PRI_FIRST);
3947 /* Link is down by default */
3948 if_link_state_change(ifp, LINK_STATE_DOWN);
3950 mlx5e_enable_async_events(priv);
3952 mlx5e_add_hw_stats(priv);
3954 mlx5e_create_stats(&priv->stats.vport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3955 "vstats", mlx5e_vport_stats_desc, MLX5E_VPORT_STATS_NUM,
3956 priv->stats.vport.arg);
3958 mlx5e_create_stats(&priv->stats.pport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3959 "pstats", mlx5e_pport_stats_desc, MLX5E_PPORT_STATS_NUM,
3960 priv->stats.pport.arg);
3962 mlx5e_create_ethtool(priv);
3964 mtx_lock(&priv->async_events_mtx);
3965 mlx5e_update_stats(priv);
3966 mtx_unlock(&priv->async_events_mtx);
3968 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3969 OID_AUTO, "rx_clbr_done", CTLFLAG_RD,
3970 &priv->clbr_done, 0,
3971 "RX timestamps calibration state");
3972 callout_init(&priv->tstmp_clbr, CALLOUT_DIRECT);
3973 mlx5e_reset_calibration_callout(priv);
3979 mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
3981 err_dealloc_transport_domain:
3982 mlx5_dealloc_transport_domain(mdev, priv->tdn);
3985 mlx5_core_dealloc_pd(mdev, priv->pdn);
3988 mlx5_unmap_free_uar(mdev, &priv->cq_uar);
3991 destroy_workqueue(priv->wq);
3994 sysctl_ctx_free(&priv->sysctl_ctx);
3995 if (priv->sysctl_debug)
3996 sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
4000 mlx5e_priv_mtx_destroy(priv);
4001 free(priv, M_MLX5EN);
4006 mlx5e_destroy_ifp(struct mlx5_core_dev *mdev, void *vpriv)
4008 struct mlx5e_priv *priv = vpriv;
4009 struct ifnet *ifp = priv->ifp;
4011 /* don't allow more IOCTLs */
4014 /* XXX wait a bit to allow IOCTL handlers to complete */
4019 * The kernel can have reference(s) via the m_snd_tag's into
4020 * the ratelimit channels, and these must go away before
4023 while (READ_ONCE(priv->rl.stats.tx_active_connections) != 0) {
4024 if_printf(priv->ifp, "Waiting for all ratelimit connections "
4029 /* stop watchdog timer */
4030 callout_drain(&priv->watchdog);
4032 callout_drain(&priv->tstmp_clbr);
4034 if (priv->vlan_attach != NULL)
4035 EVENTHANDLER_DEREGISTER(vlan_config, priv->vlan_attach);
4036 if (priv->vlan_detach != NULL)
4037 EVENTHANDLER_DEREGISTER(vlan_unconfig, priv->vlan_detach);
4039 /* make sure device gets closed */
4041 mlx5e_close_locked(ifp);
4044 /* wait for all unlimited send tags to go away */
4045 while (priv->channel_refs != 0) {
4046 if_printf(priv->ifp, "Waiting for all unlimited connections "
4051 /* unregister device */
4052 ifmedia_removeall(&priv->media);
4053 ether_ifdetach(ifp);
4057 mlx5e_rl_cleanup(priv);
4059 /* destroy all remaining sysctl nodes */
4060 sysctl_ctx_free(&priv->stats.vport.ctx);
4061 sysctl_ctx_free(&priv->stats.pport.ctx);
4062 if (priv->sysctl_debug)
4063 sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
4064 sysctl_ctx_free(&priv->sysctl_ctx);
4066 mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
4067 mlx5_dealloc_transport_domain(priv->mdev, priv->tdn);
4068 mlx5_core_dealloc_pd(priv->mdev, priv->pdn);
4069 mlx5_unmap_free_uar(priv->mdev, &priv->cq_uar);
4070 mlx5e_disable_async_events(priv);
4071 destroy_workqueue(priv->wq);
4072 mlx5e_priv_mtx_destroy(priv);
4073 free(priv, M_MLX5EN);
4077 mlx5e_get_ifp(void *vpriv)
4079 struct mlx5e_priv *priv = vpriv;
4084 static struct mlx5_interface mlx5e_interface = {
4085 .add = mlx5e_create_ifp,
4086 .remove = mlx5e_destroy_ifp,
4087 .event = mlx5e_async_event,
4088 .protocol = MLX5_INTERFACE_PROTOCOL_ETH,
4089 .get_dev = mlx5e_get_ifp,
4095 mlx5_register_interface(&mlx5e_interface);
4101 mlx5_unregister_interface(&mlx5e_interface);
4105 mlx5e_show_version(void __unused *arg)
4108 printf("%s", mlx5e_version);
4110 SYSINIT(mlx5e_show_version, SI_SUB_DRIVERS, SI_ORDER_ANY, mlx5e_show_version, NULL);
4112 module_init_order(mlx5e_init, SI_ORDER_THIRD);
4113 module_exit_order(mlx5e_cleanup, SI_ORDER_THIRD);
4115 #if (__FreeBSD_version >= 1100000)
4116 MODULE_DEPEND(mlx5en, linuxkpi, 1, 1, 1);
4118 MODULE_DEPEND(mlx5en, mlx5, 1, 1, 1);
4119 MODULE_VERSION(mlx5en, 1);