2 * Copyright (c) 2015 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 "3.4.2"
37 char mlx5e_version[] = "Mellanox Ethernet driver"
38 " (" ETH_DRIVER_VERSION ")";
40 static int mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs);
42 struct mlx5e_channel_param {
43 struct mlx5e_rq_param rq;
44 struct mlx5e_sq_param sq;
45 struct mlx5e_cq_param rx_cq;
46 struct mlx5e_cq_param tx_cq;
52 } mlx5e_mode_table[MLX5E_LINK_MODES_NUMBER] = {
54 [MLX5E_1000BASE_CX_SGMII] = {
55 .subtype = IFM_1000_CX_SGMII,
56 .baudrate = IF_Mbps(1000ULL),
58 [MLX5E_1000BASE_KX] = {
59 .subtype = IFM_1000_KX,
60 .baudrate = IF_Mbps(1000ULL),
62 [MLX5E_10GBASE_CX4] = {
63 .subtype = IFM_10G_CX4,
64 .baudrate = IF_Gbps(10ULL),
66 [MLX5E_10GBASE_KX4] = {
67 .subtype = IFM_10G_KX4,
68 .baudrate = IF_Gbps(10ULL),
70 [MLX5E_10GBASE_KR] = {
71 .subtype = IFM_10G_KR,
72 .baudrate = IF_Gbps(10ULL),
74 [MLX5E_20GBASE_KR2] = {
75 .subtype = IFM_20G_KR2,
76 .baudrate = IF_Gbps(20ULL),
78 [MLX5E_40GBASE_CR4] = {
79 .subtype = IFM_40G_CR4,
80 .baudrate = IF_Gbps(40ULL),
82 [MLX5E_40GBASE_KR4] = {
83 .subtype = IFM_40G_KR4,
84 .baudrate = IF_Gbps(40ULL),
86 [MLX5E_56GBASE_R4] = {
87 .subtype = IFM_56G_R4,
88 .baudrate = IF_Gbps(56ULL),
90 [MLX5E_10GBASE_CR] = {
91 .subtype = IFM_10G_CR1,
92 .baudrate = IF_Gbps(10ULL),
94 [MLX5E_10GBASE_SR] = {
95 .subtype = IFM_10G_SR,
96 .baudrate = IF_Gbps(10ULL),
98 [MLX5E_10GBASE_ER] = {
99 .subtype = IFM_10G_ER,
100 .baudrate = IF_Gbps(10ULL),
102 [MLX5E_40GBASE_SR4] = {
103 .subtype = IFM_40G_SR4,
104 .baudrate = IF_Gbps(40ULL),
106 [MLX5E_40GBASE_LR4] = {
107 .subtype = IFM_40G_LR4,
108 .baudrate = IF_Gbps(40ULL),
110 [MLX5E_100GBASE_CR4] = {
111 .subtype = IFM_100G_CR4,
112 .baudrate = IF_Gbps(100ULL),
114 [MLX5E_100GBASE_SR4] = {
115 .subtype = IFM_100G_SR4,
116 .baudrate = IF_Gbps(100ULL),
118 [MLX5E_100GBASE_KR4] = {
119 .subtype = IFM_100G_KR4,
120 .baudrate = IF_Gbps(100ULL),
122 [MLX5E_100GBASE_LR4] = {
123 .subtype = IFM_100G_LR4,
124 .baudrate = IF_Gbps(100ULL),
126 [MLX5E_100BASE_TX] = {
127 .subtype = IFM_100_TX,
128 .baudrate = IF_Mbps(100ULL),
130 [MLX5E_1000BASE_T] = {
131 .subtype = IFM_1000_T,
132 .baudrate = IF_Mbps(1000ULL),
134 [MLX5E_10GBASE_T] = {
135 .subtype = IFM_10G_T,
136 .baudrate = IF_Gbps(10ULL),
138 [MLX5E_25GBASE_CR] = {
139 .subtype = IFM_25G_CR,
140 .baudrate = IF_Gbps(25ULL),
142 [MLX5E_25GBASE_KR] = {
143 .subtype = IFM_25G_KR,
144 .baudrate = IF_Gbps(25ULL),
146 [MLX5E_25GBASE_SR] = {
147 .subtype = IFM_25G_SR,
148 .baudrate = IF_Gbps(25ULL),
150 [MLX5E_50GBASE_CR2] = {
151 .subtype = IFM_50G_CR2,
152 .baudrate = IF_Gbps(50ULL),
154 [MLX5E_50GBASE_KR2] = {
155 .subtype = IFM_50G_KR2,
156 .baudrate = IF_Gbps(50ULL),
160 MALLOC_DEFINE(M_MLX5EN, "MLX5EN", "MLX5 Ethernet");
162 static SYSCTL_NODE(_hw, OID_AUTO, mlx5, CTLFLAG_RW, 0, "MLX5 driver parameters");
165 mlx5e_update_carrier(struct mlx5e_priv *priv)
167 struct mlx5_core_dev *mdev = priv->mdev;
168 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 priv->ifp->if_baudrate =
201 mlx5e_mode_table[i].baudrate;
202 priv->media_active_last =
203 mlx5e_mode_table[i].subtype | IFM_ETHER | IFM_FDX;
206 if_link_state_change(priv->ifp, LINK_STATE_UP);
210 mlx5e_media_status(struct ifnet *dev, struct ifmediareq *ifmr)
212 struct mlx5e_priv *priv = dev->if_softc;
214 ifmr->ifm_status = priv->media_status_last;
215 ifmr->ifm_active = priv->media_active_last |
216 (priv->params.rx_pauseframe_control ? IFM_ETH_RXPAUSE : 0) |
217 (priv->params.tx_pauseframe_control ? IFM_ETH_TXPAUSE : 0);
222 mlx5e_find_link_mode(u32 subtype)
227 for (i = 0; i < MLX5E_LINK_MODES_NUMBER; ++i) {
228 if (mlx5e_mode_table[i].baudrate == 0)
230 if (mlx5e_mode_table[i].subtype == subtype)
231 link_mode |= MLX5E_PROT_MASK(i);
238 mlx5e_set_port_pause_and_pfc(struct mlx5e_priv *priv)
240 return (mlx5_set_port_pause_and_pfc(priv->mdev, 1,
241 priv->params.rx_pauseframe_control,
242 priv->params.tx_pauseframe_control,
243 priv->params.rx_priority_flow_control,
244 priv->params.tx_priority_flow_control));
248 mlx5e_set_port_pfc(struct mlx5e_priv *priv)
252 if (priv->params.rx_pauseframe_control ||
253 priv->params.tx_pauseframe_control) {
255 "Global pauseframes must be disabled before enabling PFC.\n");
258 error = mlx5e_set_port_pause_and_pfc(priv);
264 mlx5e_media_change(struct ifnet *dev)
266 struct mlx5e_priv *priv = dev->if_softc;
267 struct mlx5_core_dev *mdev = priv->mdev;
274 locked = PRIV_LOCKED(priv);
278 if (IFM_TYPE(priv->media.ifm_media) != IFM_ETHER) {
282 link_mode = mlx5e_find_link_mode(IFM_SUBTYPE(priv->media.ifm_media));
284 /* query supported capabilities */
285 error = mlx5_query_port_proto_cap(mdev, ð_proto_cap, MLX5_PTYS_EN);
287 if_printf(dev, "Query port media capability failed\n");
290 /* check for autoselect */
291 if (IFM_SUBTYPE(priv->media.ifm_media) == IFM_AUTO) {
292 link_mode = eth_proto_cap;
293 if (link_mode == 0) {
294 if_printf(dev, "Port media capability is zero\n");
299 link_mode = link_mode & eth_proto_cap;
300 if (link_mode == 0) {
301 if_printf(dev, "Not supported link mode requested\n");
306 if (priv->media.ifm_media & (IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE)) {
307 /* check if PFC is enabled */
308 if (priv->params.rx_priority_flow_control ||
309 priv->params.tx_priority_flow_control) {
310 if_printf(dev, "PFC must be disabled before enabling global pauseframes.\n");
315 /* update pauseframe control bits */
316 priv->params.rx_pauseframe_control =
317 (priv->media.ifm_media & IFM_ETH_RXPAUSE) ? 1 : 0;
318 priv->params.tx_pauseframe_control =
319 (priv->media.ifm_media & IFM_ETH_TXPAUSE) ? 1 : 0;
321 /* check if device is opened */
322 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
324 /* reconfigure the hardware */
325 mlx5_set_port_status(mdev, MLX5_PORT_DOWN);
326 mlx5_set_port_proto(mdev, link_mode, MLX5_PTYS_EN);
327 error = -mlx5e_set_port_pause_and_pfc(priv);
329 mlx5_set_port_status(mdev, MLX5_PORT_UP);
338 mlx5e_update_carrier_work(struct work_struct *work)
340 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
341 update_carrier_work);
344 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
345 mlx5e_update_carrier(priv);
350 * This function reads the physical port counters from the firmware
351 * using a pre-defined layout defined by various MLX5E_PPORT_XXX()
352 * macros. The output is converted from big-endian 64-bit values into
353 * host endian ones and stored in the "priv->stats.pport" structure.
356 mlx5e_update_pport_counters(struct mlx5e_priv *priv)
358 struct mlx5_core_dev *mdev = priv->mdev;
359 struct mlx5e_pport_stats *s = &priv->stats.pport;
360 struct mlx5e_port_stats_debug *s_debug = &priv->stats.port_stats_debug;
364 unsigned sz = MLX5_ST_SZ_BYTES(ppcnt_reg);
369 /* allocate firmware request structures */
370 in = mlx5_vzalloc(sz);
371 out = mlx5_vzalloc(sz);
372 if (in == NULL || out == NULL)
376 * Get pointer to the 64-bit counter set which is located at a
377 * fixed offset in the output firmware request structure:
379 ptr = (const uint64_t *)MLX5_ADDR_OF(ppcnt_reg, out, counter_set);
381 MLX5_SET(ppcnt_reg, in, local_port, 1);
383 /* read IEEE802_3 counter group using predefined counter layout */
384 MLX5_SET(ppcnt_reg, in, grp, MLX5_IEEE_802_3_COUNTERS_GROUP);
385 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
386 for (x = 0, y = MLX5E_PPORT_PER_PRIO_STATS_NUM;
387 x != MLX5E_PPORT_IEEE802_3_STATS_NUM; x++, y++)
388 s->arg[y] = be64toh(ptr[x]);
390 /* read RFC2819 counter group using predefined counter layout */
391 MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2819_COUNTERS_GROUP);
392 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
393 for (x = 0; x != MLX5E_PPORT_RFC2819_STATS_NUM; x++, y++)
394 s->arg[y] = be64toh(ptr[x]);
395 for (y = 0; x != MLX5E_PPORT_RFC2819_STATS_NUM +
396 MLX5E_PPORT_RFC2819_STATS_DEBUG_NUM; x++, y++)
397 s_debug->arg[y] = be64toh(ptr[x]);
399 /* read RFC2863 counter group using predefined counter layout */
400 MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2863_COUNTERS_GROUP);
401 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
402 for (x = 0; x != MLX5E_PPORT_RFC2863_STATS_DEBUG_NUM; x++, y++)
403 s_debug->arg[y] = be64toh(ptr[x]);
405 /* read physical layer stats counter group using predefined counter layout */
406 MLX5_SET(ppcnt_reg, in, grp, MLX5_PHYSICAL_LAYER_COUNTERS_GROUP);
407 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
408 for (x = 0; x != MLX5E_PPORT_PHYSICAL_LAYER_STATS_DEBUG_NUM; x++, y++)
409 s_debug->arg[y] = be64toh(ptr[x]);
411 /* read per-priority counters */
412 MLX5_SET(ppcnt_reg, in, grp, MLX5_PER_PRIORITY_COUNTERS_GROUP);
414 /* iterate all the priorities */
415 for (y = z = 0; z != MLX5E_PPORT_PER_PRIO_STATS_NUM_PRIO; z++) {
416 MLX5_SET(ppcnt_reg, in, prio_tc, z);
417 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
419 /* read per priority stats counter group using predefined counter layout */
420 for (x = 0; x != (MLX5E_PPORT_PER_PRIO_STATS_NUM /
421 MLX5E_PPORT_PER_PRIO_STATS_NUM_PRIO); x++, y++)
422 s->arg[y] = be64toh(ptr[x]);
425 /* free firmware request structures */
431 * This function is called regularly to collect all statistics
432 * counters from the firmware. The values can be viewed through the
433 * sysctl interface. Execution is serialized using the priv's global
434 * configuration lock.
437 mlx5e_update_stats_work(struct work_struct *work)
439 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
441 struct mlx5_core_dev *mdev = priv->mdev;
442 struct mlx5e_vport_stats *s = &priv->stats.vport;
443 struct mlx5e_rq_stats *rq_stats;
444 struct mlx5e_sq_stats *sq_stats;
445 struct buf_ring *sq_br;
446 #if (__FreeBSD_version < 1100000)
447 struct ifnet *ifp = priv->ifp;
450 u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)];
452 int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
455 u64 tx_queue_dropped = 0;
456 u64 tx_defragged = 0;
457 u64 tx_offload_none = 0;
460 u64 sw_lro_queued = 0;
461 u64 sw_lro_flushed = 0;
462 u64 rx_csum_none = 0;
464 u32 rx_out_of_buffer = 0;
469 out = mlx5_vzalloc(outlen);
472 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
475 /* Collect firts the SW counters and then HW for consistency */
476 for (i = 0; i < priv->params.num_channels; i++) {
477 struct mlx5e_rq *rq = &priv->channel[i]->rq;
479 rq_stats = &priv->channel[i]->rq.stats;
481 /* collect stats from LRO */
482 rq_stats->sw_lro_queued = rq->lro.lro_queued;
483 rq_stats->sw_lro_flushed = rq->lro.lro_flushed;
484 sw_lro_queued += rq_stats->sw_lro_queued;
485 sw_lro_flushed += rq_stats->sw_lro_flushed;
486 lro_packets += rq_stats->lro_packets;
487 lro_bytes += rq_stats->lro_bytes;
488 rx_csum_none += rq_stats->csum_none;
489 rx_wqe_err += rq_stats->wqe_err;
491 for (j = 0; j < priv->num_tc; j++) {
492 sq_stats = &priv->channel[i]->sq[j].stats;
493 sq_br = priv->channel[i]->sq[j].br;
495 tso_packets += sq_stats->tso_packets;
496 tso_bytes += sq_stats->tso_bytes;
497 tx_queue_dropped += sq_stats->dropped;
499 tx_queue_dropped += sq_br->br_drops;
500 tx_defragged += sq_stats->defragged;
501 tx_offload_none += sq_stats->csum_offload_none;
505 /* update counters */
506 s->tso_packets = tso_packets;
507 s->tso_bytes = tso_bytes;
508 s->tx_queue_dropped = tx_queue_dropped;
509 s->tx_defragged = tx_defragged;
510 s->lro_packets = lro_packets;
511 s->lro_bytes = lro_bytes;
512 s->sw_lro_queued = sw_lro_queued;
513 s->sw_lro_flushed = sw_lro_flushed;
514 s->rx_csum_none = rx_csum_none;
515 s->rx_wqe_err = rx_wqe_err;
518 memset(in, 0, sizeof(in));
520 MLX5_SET(query_vport_counter_in, in, opcode,
521 MLX5_CMD_OP_QUERY_VPORT_COUNTER);
522 MLX5_SET(query_vport_counter_in, in, op_mod, 0);
523 MLX5_SET(query_vport_counter_in, in, other_vport, 0);
525 memset(out, 0, outlen);
527 /* get number of out-of-buffer drops first */
528 if (mlx5_vport_query_out_of_rx_buffer(mdev, priv->counter_set_id,
532 /* accumulate difference into a 64-bit counter */
533 s->rx_out_of_buffer += (u64)(u32)(rx_out_of_buffer - s->rx_out_of_buffer_prev);
534 s->rx_out_of_buffer_prev = rx_out_of_buffer;
536 /* get port statistics */
537 if (mlx5_cmd_exec(mdev, in, sizeof(in), out, outlen))
540 #define MLX5_GET_CTR(out, x) \
541 MLX5_GET64(query_vport_counter_out, out, x)
543 s->rx_error_packets =
544 MLX5_GET_CTR(out, received_errors.packets);
546 MLX5_GET_CTR(out, received_errors.octets);
547 s->tx_error_packets =
548 MLX5_GET_CTR(out, transmit_errors.packets);
550 MLX5_GET_CTR(out, transmit_errors.octets);
552 s->rx_unicast_packets =
553 MLX5_GET_CTR(out, received_eth_unicast.packets);
554 s->rx_unicast_bytes =
555 MLX5_GET_CTR(out, received_eth_unicast.octets);
556 s->tx_unicast_packets =
557 MLX5_GET_CTR(out, transmitted_eth_unicast.packets);
558 s->tx_unicast_bytes =
559 MLX5_GET_CTR(out, transmitted_eth_unicast.octets);
561 s->rx_multicast_packets =
562 MLX5_GET_CTR(out, received_eth_multicast.packets);
563 s->rx_multicast_bytes =
564 MLX5_GET_CTR(out, received_eth_multicast.octets);
565 s->tx_multicast_packets =
566 MLX5_GET_CTR(out, transmitted_eth_multicast.packets);
567 s->tx_multicast_bytes =
568 MLX5_GET_CTR(out, transmitted_eth_multicast.octets);
570 s->rx_broadcast_packets =
571 MLX5_GET_CTR(out, received_eth_broadcast.packets);
572 s->rx_broadcast_bytes =
573 MLX5_GET_CTR(out, received_eth_broadcast.octets);
574 s->tx_broadcast_packets =
575 MLX5_GET_CTR(out, transmitted_eth_broadcast.packets);
576 s->tx_broadcast_bytes =
577 MLX5_GET_CTR(out, transmitted_eth_broadcast.octets);
580 s->rx_unicast_packets +
581 s->rx_multicast_packets +
582 s->rx_broadcast_packets -
585 s->rx_unicast_bytes +
586 s->rx_multicast_bytes +
587 s->rx_broadcast_bytes;
589 s->tx_unicast_packets +
590 s->tx_multicast_packets +
591 s->tx_broadcast_packets;
593 s->tx_unicast_bytes +
594 s->tx_multicast_bytes +
595 s->tx_broadcast_bytes;
597 /* Update calculated offload counters */
598 s->tx_csum_offload = s->tx_packets - tx_offload_none;
599 s->rx_csum_good = s->rx_packets - s->rx_csum_none;
601 /* Get physical port counters */
602 mlx5e_update_pport_counters(priv);
604 #if (__FreeBSD_version < 1100000)
605 /* no get_counters interface in fbsd 10 */
606 ifp->if_ipackets = s->rx_packets;
607 ifp->if_ierrors = s->rx_error_packets +
608 priv->stats.pport.alignment_err +
609 priv->stats.pport.check_seq_err +
610 priv->stats.pport.crc_align_errors +
611 priv->stats.pport.in_range_len_errors +
612 priv->stats.pport.jabbers +
613 priv->stats.pport.out_of_range_len +
614 priv->stats.pport.oversize_pkts +
615 priv->stats.pport.symbol_err +
616 priv->stats.pport.too_long_errors +
617 priv->stats.pport.undersize_pkts +
618 priv->stats.pport.unsupported_op_rx;
619 ifp->if_iqdrops = s->rx_out_of_buffer +
620 priv->stats.pport.drop_events;
621 ifp->if_opackets = s->tx_packets;
622 ifp->if_oerrors = s->tx_error_packets;
623 ifp->if_snd.ifq_drops = s->tx_queue_dropped;
624 ifp->if_ibytes = s->rx_bytes;
625 ifp->if_obytes = s->tx_bytes;
627 priv->stats.pport.collisions;
633 /* Update diagnostics, if any */
634 if (priv->params_ethtool.diag_pci_enable ||
635 priv->params_ethtool.diag_general_enable) {
636 int error = mlx5_core_get_diagnostics_full(mdev,
637 priv->params_ethtool.diag_pci_enable ? &priv->params_pci : NULL,
638 priv->params_ethtool.diag_general_enable ? &priv->params_general : NULL);
640 if_printf(priv->ifp, "Failed reading diagnostics: %d\n", error);
646 mlx5e_update_stats(void *arg)
648 struct mlx5e_priv *priv = arg;
650 queue_work(priv->wq, &priv->update_stats_work);
652 callout_reset(&priv->watchdog, hz, &mlx5e_update_stats, priv);
656 mlx5e_async_event_sub(struct mlx5e_priv *priv,
657 enum mlx5_dev_event event)
660 case MLX5_DEV_EVENT_PORT_UP:
661 case MLX5_DEV_EVENT_PORT_DOWN:
662 queue_work(priv->wq, &priv->update_carrier_work);
671 mlx5e_async_event(struct mlx5_core_dev *mdev, void *vpriv,
672 enum mlx5_dev_event event, unsigned long param)
674 struct mlx5e_priv *priv = vpriv;
676 mtx_lock(&priv->async_events_mtx);
677 if (test_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state))
678 mlx5e_async_event_sub(priv, event);
679 mtx_unlock(&priv->async_events_mtx);
683 mlx5e_enable_async_events(struct mlx5e_priv *priv)
685 set_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
689 mlx5e_disable_async_events(struct mlx5e_priv *priv)
691 mtx_lock(&priv->async_events_mtx);
692 clear_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state);
693 mtx_unlock(&priv->async_events_mtx);
696 static void mlx5e_calibration_callout(void *arg);
697 static int mlx5e_calibration_duration = 20;
698 static int mlx5e_fast_calibration = 1;
699 static int mlx5e_normal_calibration = 30;
701 static SYSCTL_NODE(_hw_mlx5, OID_AUTO, calibr, CTLFLAG_RW, 0,
702 "MLX5 timestamp calibration parameteres");
704 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, duration, CTLFLAG_RWTUN,
705 &mlx5e_calibration_duration, 0,
706 "Duration of initial calibration");
707 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, fast, CTLFLAG_RWTUN,
708 &mlx5e_fast_calibration, 0,
709 "Recalibration interval during initial calibration");
710 SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, normal, CTLFLAG_RWTUN,
711 &mlx5e_normal_calibration, 0,
712 "Recalibration interval during normal operations");
715 * Ignites the calibration process.
718 mlx5e_reset_calibration_callout(struct mlx5e_priv *priv)
721 if (priv->clbr_done == 0)
722 mlx5e_calibration_callout(priv);
724 callout_reset_curcpu(&priv->tstmp_clbr, (priv->clbr_done <
725 mlx5e_calibration_duration ? mlx5e_fast_calibration :
726 mlx5e_normal_calibration) * hz, mlx5e_calibration_callout,
731 mlx5e_timespec2usec(const struct timespec *ts)
734 return ((uint64_t)ts->tv_sec * 1000000000 + ts->tv_nsec);
738 mlx5e_hw_clock(struct mlx5e_priv *priv)
740 struct mlx5_init_seg *iseg;
741 uint32_t hw_h, hw_h1, hw_l;
743 iseg = priv->mdev->iseg;
745 hw_h = ioread32be(&iseg->internal_timer_h);
746 hw_l = ioread32be(&iseg->internal_timer_l);
747 hw_h1 = ioread32be(&iseg->internal_timer_h);
748 } while (hw_h1 != hw_h);
749 return (((uint64_t)hw_h << 32) | hw_l);
753 * The calibration callout, it runs either in the context of the
754 * thread which enables calibration, or in callout. It takes the
755 * snapshot of system and adapter clocks, then advances the pointers to
756 * the calibration point to allow rx path to read the consistent data
760 mlx5e_calibration_callout(void *arg)
762 struct mlx5e_priv *priv;
763 struct mlx5e_clbr_point *next, *curr;
768 curr = &priv->clbr_points[priv->clbr_curr];
769 clbr_curr_next = priv->clbr_curr + 1;
770 if (clbr_curr_next >= nitems(priv->clbr_points))
772 next = &priv->clbr_points[clbr_curr_next];
774 next->base_prev = curr->base_curr;
775 next->clbr_hw_prev = curr->clbr_hw_curr;
777 next->clbr_hw_curr = mlx5e_hw_clock(priv);
778 if (((next->clbr_hw_curr - curr->clbr_hw_prev) >> MLX5E_TSTMP_PREC) ==
780 if_printf(priv->ifp, "HW failed tstmp frozen %#jx %#jx,"
781 "disabling\n", next->clbr_hw_curr, curr->clbr_hw_prev);
787 next->base_curr = mlx5e_timespec2usec(&ts);
790 atomic_thread_fence_rel();
791 priv->clbr_curr = clbr_curr_next;
792 atomic_store_rel_int(&next->clbr_gen, ++(priv->clbr_gen));
794 if (priv->clbr_done < mlx5e_calibration_duration)
796 mlx5e_reset_calibration_callout(priv);
799 static const char *mlx5e_rq_stats_desc[] = {
800 MLX5E_RQ_STATS(MLX5E_STATS_DESC)
804 mlx5e_create_rq(struct mlx5e_channel *c,
805 struct mlx5e_rq_param *param,
808 struct mlx5e_priv *priv = c->priv;
809 struct mlx5_core_dev *mdev = priv->mdev;
811 void *rqc = param->rqc;
812 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
818 err = mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
822 /* Create DMA descriptor TAG */
823 if ((err = -bus_dma_tag_create(
824 bus_get_dma_tag(mdev->pdev->dev.bsddev),
825 1, /* any alignment */
827 BUS_SPACE_MAXADDR, /* lowaddr */
828 BUS_SPACE_MAXADDR, /* highaddr */
829 NULL, NULL, /* filter, filterarg */
830 nsegs * MLX5E_MAX_RX_BYTES, /* maxsize */
831 nsegs, /* nsegments */
832 nsegs * MLX5E_MAX_RX_BYTES, /* maxsegsize */
834 NULL, NULL, /* lockfunc, lockfuncarg */
838 err = mlx5_wq_ll_create(mdev, ¶m->wq, rqc_wq, &rq->wq,
841 goto err_free_dma_tag;
843 rq->wq.db = &rq->wq.db[MLX5_RCV_DBR];
845 err = mlx5e_get_wqe_sz(priv, &rq->wqe_sz, &rq->nsegs);
847 goto err_rq_wq_destroy;
849 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
851 err = -tcp_lro_init_args(&rq->lro, c->ifp, TCP_LRO_ENTRIES, wq_sz);
853 goto err_rq_wq_destroy;
855 rq->mbuf = malloc(wq_sz * sizeof(rq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
856 for (i = 0; i != wq_sz; i++) {
857 struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
858 #if (MLX5E_MAX_RX_SEGS == 1)
859 uint32_t byte_count = rq->wqe_sz - MLX5E_NET_IP_ALIGN;
864 err = -bus_dmamap_create(rq->dma_tag, 0, &rq->mbuf[i].dma_map);
867 bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
868 goto err_rq_mbuf_free;
871 /* set value for constant fields */
872 #if (MLX5E_MAX_RX_SEGS == 1)
873 wqe->data[0].lkey = c->mkey_be;
874 wqe->data[0].byte_count = cpu_to_be32(byte_count | MLX5_HW_START_PADDING);
876 for (j = 0; j < rq->nsegs; j++)
877 wqe->data[j].lkey = c->mkey_be;
885 snprintf(buffer, sizeof(buffer), "rxstat%d", c->ix);
886 mlx5e_create_stats(&rq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
887 buffer, mlx5e_rq_stats_desc, MLX5E_RQ_STATS_NUM,
892 free(rq->mbuf, M_MLX5EN);
893 tcp_lro_free(&rq->lro);
895 mlx5_wq_destroy(&rq->wq_ctrl);
897 bus_dma_tag_destroy(rq->dma_tag);
903 mlx5e_destroy_rq(struct mlx5e_rq *rq)
908 /* destroy all sysctl nodes */
909 sysctl_ctx_free(&rq->stats.ctx);
911 /* free leftover LRO packets, if any */
912 tcp_lro_free(&rq->lro);
914 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
915 for (i = 0; i != wq_sz; i++) {
916 if (rq->mbuf[i].mbuf != NULL) {
917 bus_dmamap_unload(rq->dma_tag, rq->mbuf[i].dma_map);
918 m_freem(rq->mbuf[i].mbuf);
920 bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map);
922 free(rq->mbuf, M_MLX5EN);
923 mlx5_wq_destroy(&rq->wq_ctrl);
927 mlx5e_enable_rq(struct mlx5e_rq *rq, struct mlx5e_rq_param *param)
929 struct mlx5e_channel *c = rq->channel;
930 struct mlx5e_priv *priv = c->priv;
931 struct mlx5_core_dev *mdev = priv->mdev;
939 inlen = MLX5_ST_SZ_BYTES(create_rq_in) +
940 sizeof(u64) * rq->wq_ctrl.buf.npages;
941 in = mlx5_vzalloc(inlen);
945 rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
946 wq = MLX5_ADDR_OF(rqc, rqc, wq);
948 memcpy(rqc, param->rqc, sizeof(param->rqc));
950 MLX5_SET(rqc, rqc, cqn, c->rq.cq.mcq.cqn);
951 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST);
952 MLX5_SET(rqc, rqc, flush_in_error_en, 1);
953 if (priv->counter_set_id >= 0)
954 MLX5_SET(rqc, rqc, counter_set_id, priv->counter_set_id);
955 MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift -
957 MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma);
959 mlx5_fill_page_array(&rq->wq_ctrl.buf,
960 (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
962 err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn);
970 mlx5e_modify_rq(struct mlx5e_rq *rq, int curr_state, int next_state)
972 struct mlx5e_channel *c = rq->channel;
973 struct mlx5e_priv *priv = c->priv;
974 struct mlx5_core_dev *mdev = priv->mdev;
981 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
982 in = mlx5_vzalloc(inlen);
986 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
988 MLX5_SET(modify_rq_in, in, rqn, rq->rqn);
989 MLX5_SET(modify_rq_in, in, rq_state, curr_state);
990 MLX5_SET(rqc, rqc, state, next_state);
992 err = mlx5_core_modify_rq(mdev, in, inlen);
1000 mlx5e_disable_rq(struct mlx5e_rq *rq)
1002 struct mlx5e_channel *c = rq->channel;
1003 struct mlx5e_priv *priv = c->priv;
1004 struct mlx5_core_dev *mdev = priv->mdev;
1006 mlx5_core_destroy_rq(mdev, rq->rqn);
1010 mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq)
1012 struct mlx5e_channel *c = rq->channel;
1013 struct mlx5e_priv *priv = c->priv;
1014 struct mlx5_wq_ll *wq = &rq->wq;
1017 for (i = 0; i < 1000; i++) {
1018 if (wq->cur_sz >= priv->params.min_rx_wqes)
1023 return (-ETIMEDOUT);
1027 mlx5e_open_rq(struct mlx5e_channel *c,
1028 struct mlx5e_rq_param *param,
1029 struct mlx5e_rq *rq)
1033 err = mlx5e_create_rq(c, param, rq);
1037 err = mlx5e_enable_rq(rq, param);
1039 goto err_destroy_rq;
1041 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
1043 goto err_disable_rq;
1050 mlx5e_disable_rq(rq);
1052 mlx5e_destroy_rq(rq);
1058 mlx5e_close_rq(struct mlx5e_rq *rq)
1062 callout_stop(&rq->watchdog);
1063 mtx_unlock(&rq->mtx);
1065 callout_drain(&rq->watchdog);
1067 mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
1071 mlx5e_close_rq_wait(struct mlx5e_rq *rq)
1073 struct mlx5_core_dev *mdev = rq->channel->priv->mdev;
1075 /* wait till RQ is empty */
1076 while (!mlx5_wq_ll_is_empty(&rq->wq) &&
1077 (mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR)) {
1079 rq->cq.mcq.comp(&rq->cq.mcq);
1082 mlx5e_disable_rq(rq);
1083 mlx5e_destroy_rq(rq);
1087 mlx5e_free_sq_db(struct mlx5e_sq *sq)
1089 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1092 for (x = 0; x != wq_sz; x++)
1093 bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1094 free(sq->mbuf, M_MLX5EN);
1098 mlx5e_alloc_sq_db(struct mlx5e_sq *sq)
1100 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1104 sq->mbuf = malloc(wq_sz * sizeof(sq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO);
1106 /* Create DMA descriptor MAPs */
1107 for (x = 0; x != wq_sz; x++) {
1108 err = -bus_dmamap_create(sq->dma_tag, 0, &sq->mbuf[x].dma_map);
1111 bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map);
1112 free(sq->mbuf, M_MLX5EN);
1119 static const char *mlx5e_sq_stats_desc[] = {
1120 MLX5E_SQ_STATS(MLX5E_STATS_DESC)
1124 mlx5e_create_sq(struct mlx5e_channel *c,
1126 struct mlx5e_sq_param *param,
1127 struct mlx5e_sq *sq)
1129 struct mlx5e_priv *priv = c->priv;
1130 struct mlx5_core_dev *mdev = priv->mdev;
1133 void *sqc = param->sqc;
1134 void *sqc_wq = MLX5_ADDR_OF(sqc, sqc, wq);
1141 /* Create DMA descriptor TAG */
1142 if ((err = -bus_dma_tag_create(
1143 bus_get_dma_tag(mdev->pdev->dev.bsddev),
1144 1, /* any alignment */
1145 0, /* no boundary */
1146 BUS_SPACE_MAXADDR, /* lowaddr */
1147 BUS_SPACE_MAXADDR, /* highaddr */
1148 NULL, NULL, /* filter, filterarg */
1149 MLX5E_MAX_TX_PAYLOAD_SIZE, /* maxsize */
1150 MLX5E_MAX_TX_MBUF_FRAGS, /* nsegments */
1151 MLX5E_MAX_TX_MBUF_SIZE, /* maxsegsize */
1153 NULL, NULL, /* lockfunc, lockfuncarg */
1157 err = mlx5_alloc_map_uar(mdev, &sq->uar);
1159 goto err_free_dma_tag;
1161 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, &sq->wq,
1164 goto err_unmap_free_uar;
1166 sq->wq.db = &sq->wq.db[MLX5_SND_DBR];
1167 sq->bf_buf_size = (1 << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2;
1169 err = mlx5e_alloc_sq_db(sq);
1171 goto err_sq_wq_destroy;
1173 sq->mkey_be = c->mkey_be;
1174 sq->ifp = priv->ifp;
1177 sq->max_inline = priv->params.tx_max_inline;
1178 sq->min_inline_mode = priv->params.tx_min_inline_mode;
1179 sq->vlan_inline_cap = MLX5_CAP_ETH(mdev, wqe_vlan_insert);
1181 /* check if we should allocate a second packet buffer */
1182 if (priv->params_ethtool.tx_bufring_disable == 0) {
1183 sq->br = buf_ring_alloc(MLX5E_SQ_TX_QUEUE_SIZE, M_MLX5EN,
1184 M_WAITOK, &sq->lock);
1185 if (sq->br == NULL) {
1186 if_printf(c->ifp, "%s: Failed allocating sq drbr buffer\n",
1189 goto err_free_sq_db;
1192 sq->sq_tq = taskqueue_create_fast("mlx5e_que", M_WAITOK,
1193 taskqueue_thread_enqueue, &sq->sq_tq);
1194 if (sq->sq_tq == NULL) {
1195 if_printf(c->ifp, "%s: Failed allocating taskqueue\n",
1201 TASK_INIT(&sq->sq_task, 0, mlx5e_tx_que, sq);
1203 cpu_id = rss_getcpu(c->ix % rss_getnumbuckets());
1204 CPU_SETOF(cpu_id, &cpu_mask);
1205 taskqueue_start_threads_cpuset(&sq->sq_tq, 1, PI_NET, &cpu_mask,
1206 "%s TX SQ%d.%d CPU%d", c->ifp->if_xname, c->ix, tc, cpu_id);
1208 taskqueue_start_threads(&sq->sq_tq, 1, PI_NET,
1209 "%s TX SQ%d.%d", c->ifp->if_xname, c->ix, tc);
1212 snprintf(buffer, sizeof(buffer), "txstat%dtc%d", c->ix, tc);
1213 mlx5e_create_stats(&sq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
1214 buffer, mlx5e_sq_stats_desc, MLX5E_SQ_STATS_NUM,
1220 buf_ring_free(sq->br, M_MLX5EN);
1222 mlx5e_free_sq_db(sq);
1224 mlx5_wq_destroy(&sq->wq_ctrl);
1227 mlx5_unmap_free_uar(mdev, &sq->uar);
1230 bus_dma_tag_destroy(sq->dma_tag);
1236 mlx5e_destroy_sq(struct mlx5e_sq *sq)
1238 /* destroy all sysctl nodes */
1239 sysctl_ctx_free(&sq->stats.ctx);
1241 mlx5e_free_sq_db(sq);
1242 mlx5_wq_destroy(&sq->wq_ctrl);
1243 mlx5_unmap_free_uar(sq->priv->mdev, &sq->uar);
1244 if (sq->sq_tq != NULL) {
1245 taskqueue_drain(sq->sq_tq, &sq->sq_task);
1246 taskqueue_free(sq->sq_tq);
1249 buf_ring_free(sq->br, M_MLX5EN);
1253 mlx5e_enable_sq(struct mlx5e_sq *sq, struct mlx5e_sq_param *param,
1262 inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
1263 sizeof(u64) * sq->wq_ctrl.buf.npages;
1264 in = mlx5_vzalloc(inlen);
1268 sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
1269 wq = MLX5_ADDR_OF(sqc, sqc, wq);
1271 memcpy(sqc, param->sqc, sizeof(param->sqc));
1273 MLX5_SET(sqc, sqc, tis_num_0, tis_num);
1274 MLX5_SET(sqc, sqc, cqn, sq->cq.mcq.cqn);
1275 MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
1276 MLX5_SET(sqc, sqc, tis_lst_sz, 1);
1277 MLX5_SET(sqc, sqc, flush_in_error_en, 1);
1279 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
1280 MLX5_SET(wq, wq, uar_page, sq->uar.index);
1281 MLX5_SET(wq, wq, log_wq_pg_sz, sq->wq_ctrl.buf.page_shift -
1283 MLX5_SET64(wq, wq, dbr_addr, sq->wq_ctrl.db.dma);
1285 mlx5_fill_page_array(&sq->wq_ctrl.buf,
1286 (__be64 *) MLX5_ADDR_OF(wq, wq, pas));
1288 err = mlx5_core_create_sq(sq->priv->mdev, in, inlen, &sq->sqn);
1296 mlx5e_modify_sq(struct mlx5e_sq *sq, int curr_state, int next_state)
1303 inlen = MLX5_ST_SZ_BYTES(modify_sq_in);
1304 in = mlx5_vzalloc(inlen);
1308 sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx);
1310 MLX5_SET(modify_sq_in, in, sqn, sq->sqn);
1311 MLX5_SET(modify_sq_in, in, sq_state, curr_state);
1312 MLX5_SET(sqc, sqc, state, next_state);
1314 err = mlx5_core_modify_sq(sq->priv->mdev, in, inlen);
1322 mlx5e_disable_sq(struct mlx5e_sq *sq)
1325 mlx5_core_destroy_sq(sq->priv->mdev, sq->sqn);
1329 mlx5e_open_sq(struct mlx5e_channel *c,
1331 struct mlx5e_sq_param *param,
1332 struct mlx5e_sq *sq)
1336 err = mlx5e_create_sq(c, tc, param, sq);
1340 err = mlx5e_enable_sq(sq, param, c->priv->tisn[tc]);
1342 goto err_destroy_sq;
1344 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY);
1346 goto err_disable_sq;
1348 WRITE_ONCE(sq->queue_state, MLX5E_SQ_READY);
1353 mlx5e_disable_sq(sq);
1355 mlx5e_destroy_sq(sq);
1361 mlx5e_sq_send_nops_locked(struct mlx5e_sq *sq, int can_sleep)
1363 /* fill up remainder with NOPs */
1364 while (sq->cev_counter != 0) {
1365 while (!mlx5e_sq_has_room_for(sq, 1)) {
1366 if (can_sleep != 0) {
1367 mtx_unlock(&sq->lock);
1369 mtx_lock(&sq->lock);
1374 /* send a single NOP */
1375 mlx5e_send_nop(sq, 1);
1376 atomic_thread_fence_rel();
1379 /* Check if we need to write the doorbell */
1380 if (likely(sq->doorbell.d64 != 0)) {
1381 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
1382 sq->doorbell.d64 = 0;
1387 mlx5e_sq_cev_timeout(void *arg)
1389 struct mlx5e_sq *sq = arg;
1391 mtx_assert(&sq->lock, MA_OWNED);
1393 /* check next state */
1394 switch (sq->cev_next_state) {
1395 case MLX5E_CEV_STATE_SEND_NOPS:
1396 /* fill TX ring with NOPs, if any */
1397 mlx5e_sq_send_nops_locked(sq, 0);
1399 /* check if completed */
1400 if (sq->cev_counter == 0) {
1401 sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
1406 /* send NOPs on next timeout */
1407 sq->cev_next_state = MLX5E_CEV_STATE_SEND_NOPS;
1412 callout_reset_curcpu(&sq->cev_callout, hz, mlx5e_sq_cev_timeout, sq);
1416 mlx5e_drain_sq(struct mlx5e_sq *sq)
1419 struct mlx5_core_dev *mdev= sq->priv->mdev;
1422 * Check if already stopped.
1424 * NOTE: The "stopped" variable is only written when both the
1425 * priv's configuration lock and the SQ's lock is locked. It
1426 * can therefore safely be read when only one of the two locks
1427 * is locked. This function is always called when the priv's
1428 * configuration lock is locked.
1430 if (sq->stopped != 0)
1433 mtx_lock(&sq->lock);
1435 /* don't put more packets into the SQ */
1438 /* teardown event factor timer, if any */
1439 sq->cev_next_state = MLX5E_CEV_STATE_HOLD_NOPS;
1440 callout_stop(&sq->cev_callout);
1442 /* send dummy NOPs in order to flush the transmit ring */
1443 mlx5e_sq_send_nops_locked(sq, 1);
1444 mtx_unlock(&sq->lock);
1446 /* make sure it is safe to free the callout */
1447 callout_drain(&sq->cev_callout);
1449 /* wait till SQ is empty or link is down */
1450 mtx_lock(&sq->lock);
1451 while (sq->cc != sq->pc &&
1452 (sq->priv->media_status_last & IFM_ACTIVE) != 0 &&
1453 mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1454 mtx_unlock(&sq->lock);
1456 sq->cq.mcq.comp(&sq->cq.mcq);
1457 mtx_lock(&sq->lock);
1459 mtx_unlock(&sq->lock);
1461 /* error out remaining requests */
1462 error = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RDY, MLX5_SQC_STATE_ERR);
1465 "mlx5e_modify_sq() from RDY to ERR failed: %d\n", error);
1468 /* wait till SQ is empty */
1469 mtx_lock(&sq->lock);
1470 while (sq->cc != sq->pc &&
1471 mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1472 mtx_unlock(&sq->lock);
1474 sq->cq.mcq.comp(&sq->cq.mcq);
1475 mtx_lock(&sq->lock);
1477 mtx_unlock(&sq->lock);
1481 mlx5e_close_sq_wait(struct mlx5e_sq *sq)
1485 mlx5e_disable_sq(sq);
1486 mlx5e_destroy_sq(sq);
1490 mlx5e_create_cq(struct mlx5e_priv *priv,
1491 struct mlx5e_cq_param *param,
1492 struct mlx5e_cq *cq,
1493 mlx5e_cq_comp_t *comp,
1496 struct mlx5_core_dev *mdev = priv->mdev;
1497 struct mlx5_core_cq *mcq = &cq->mcq;
1503 param->wq.buf_numa_node = 0;
1504 param->wq.db_numa_node = 0;
1506 err = mlx5_cqwq_create(mdev, ¶m->wq, param->cqc, &cq->wq,
1511 mlx5_vector2eqn(mdev, eq_ix, &eqn_not_used, &irqn);
1514 mcq->set_ci_db = cq->wq_ctrl.db.db;
1515 mcq->arm_db = cq->wq_ctrl.db.db + 1;
1516 *mcq->set_ci_db = 0;
1518 mcq->vector = eq_ix;
1520 mcq->event = mlx5e_cq_error_event;
1522 mcq->uar = &priv->cq_uar;
1524 for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) {
1525 struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i);
1536 mlx5e_destroy_cq(struct mlx5e_cq *cq)
1538 mlx5_wq_destroy(&cq->wq_ctrl);
1542 mlx5e_enable_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param, int eq_ix)
1544 struct mlx5_core_cq *mcq = &cq->mcq;
1552 inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
1553 sizeof(u64) * cq->wq_ctrl.buf.npages;
1554 in = mlx5_vzalloc(inlen);
1558 cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
1560 memcpy(cqc, param->cqc, sizeof(param->cqc));
1562 mlx5_fill_page_array(&cq->wq_ctrl.buf,
1563 (__be64 *) MLX5_ADDR_OF(create_cq_in, in, pas));
1565 mlx5_vector2eqn(cq->priv->mdev, eq_ix, &eqn, &irqn_not_used);
1567 MLX5_SET(cqc, cqc, c_eqn, eqn);
1568 MLX5_SET(cqc, cqc, uar_page, mcq->uar->index);
1569 MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.buf.page_shift -
1571 MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
1573 err = mlx5_core_create_cq(cq->priv->mdev, mcq, in, inlen);
1580 mlx5e_cq_arm(cq, MLX5_GET_DOORBELL_LOCK(&cq->priv->doorbell_lock));
1586 mlx5e_disable_cq(struct mlx5e_cq *cq)
1589 mlx5_core_destroy_cq(cq->priv->mdev, &cq->mcq);
1593 mlx5e_open_cq(struct mlx5e_priv *priv,
1594 struct mlx5e_cq_param *param,
1595 struct mlx5e_cq *cq,
1596 mlx5e_cq_comp_t *comp,
1601 err = mlx5e_create_cq(priv, param, cq, comp, eq_ix);
1605 err = mlx5e_enable_cq(cq, param, eq_ix);
1607 goto err_destroy_cq;
1612 mlx5e_destroy_cq(cq);
1618 mlx5e_close_cq(struct mlx5e_cq *cq)
1620 mlx5e_disable_cq(cq);
1621 mlx5e_destroy_cq(cq);
1625 mlx5e_open_tx_cqs(struct mlx5e_channel *c,
1626 struct mlx5e_channel_param *cparam)
1631 for (tc = 0; tc < c->num_tc; tc++) {
1632 /* open completion queue */
1633 err = mlx5e_open_cq(c->priv, &cparam->tx_cq, &c->sq[tc].cq,
1634 &mlx5e_tx_cq_comp, c->ix);
1636 goto err_close_tx_cqs;
1641 for (tc--; tc >= 0; tc--)
1642 mlx5e_close_cq(&c->sq[tc].cq);
1648 mlx5e_close_tx_cqs(struct mlx5e_channel *c)
1652 for (tc = 0; tc < c->num_tc; tc++)
1653 mlx5e_close_cq(&c->sq[tc].cq);
1657 mlx5e_open_sqs(struct mlx5e_channel *c,
1658 struct mlx5e_channel_param *cparam)
1663 for (tc = 0; tc < c->num_tc; tc++) {
1664 err = mlx5e_open_sq(c, tc, &cparam->sq, &c->sq[tc]);
1672 for (tc--; tc >= 0; tc--)
1673 mlx5e_close_sq_wait(&c->sq[tc]);
1679 mlx5e_close_sqs_wait(struct mlx5e_channel *c)
1683 for (tc = 0; tc < c->num_tc; tc++)
1684 mlx5e_close_sq_wait(&c->sq[tc]);
1688 mlx5e_chan_mtx_init(struct mlx5e_channel *c)
1692 mtx_init(&c->rq.mtx, "mlx5rx", MTX_NETWORK_LOCK, MTX_DEF);
1694 callout_init_mtx(&c->rq.watchdog, &c->rq.mtx, 0);
1696 for (tc = 0; tc < c->num_tc; tc++) {
1697 struct mlx5e_sq *sq = c->sq + tc;
1699 mtx_init(&sq->lock, "mlx5tx",
1700 MTX_NETWORK_LOCK " TX", MTX_DEF);
1701 mtx_init(&sq->comp_lock, "mlx5comp",
1702 MTX_NETWORK_LOCK " TX", MTX_DEF);
1704 callout_init_mtx(&sq->cev_callout, &sq->lock, 0);
1706 sq->cev_factor = c->priv->params_ethtool.tx_completion_fact;
1708 /* ensure the TX completion event factor is not zero */
1709 if (sq->cev_factor == 0)
1715 mlx5e_chan_mtx_destroy(struct mlx5e_channel *c)
1719 mtx_destroy(&c->rq.mtx);
1721 for (tc = 0; tc < c->num_tc; tc++) {
1722 mtx_destroy(&c->sq[tc].lock);
1723 mtx_destroy(&c->sq[tc].comp_lock);
1728 mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
1729 struct mlx5e_channel_param *cparam,
1730 struct mlx5e_channel *volatile *cp)
1732 struct mlx5e_channel *c;
1735 c = malloc(sizeof(*c), M_MLX5EN, M_WAITOK | M_ZERO);
1740 c->mkey_be = cpu_to_be32(priv->mr.key);
1741 c->num_tc = priv->num_tc;
1744 mlx5e_chan_mtx_init(c);
1746 /* open transmit completion queue */
1747 err = mlx5e_open_tx_cqs(c, cparam);
1751 /* open receive completion queue */
1752 err = mlx5e_open_cq(c->priv, &cparam->rx_cq, &c->rq.cq,
1753 &mlx5e_rx_cq_comp, c->ix);
1755 goto err_close_tx_cqs;
1757 err = mlx5e_open_sqs(c, cparam);
1759 goto err_close_rx_cq;
1761 err = mlx5e_open_rq(c, &cparam->rq, &c->rq);
1765 /* store channel pointer */
1768 /* poll receive queue initially */
1769 c->rq.cq.mcq.comp(&c->rq.cq.mcq);
1774 mlx5e_close_sqs_wait(c);
1777 mlx5e_close_cq(&c->rq.cq);
1780 mlx5e_close_tx_cqs(c);
1783 /* destroy mutexes */
1784 mlx5e_chan_mtx_destroy(c);
1790 mlx5e_close_channel(struct mlx5e_channel *volatile *pp)
1792 struct mlx5e_channel *c = *pp;
1794 /* check if channel is already closed */
1797 mlx5e_close_rq(&c->rq);
1801 mlx5e_close_channel_wait(struct mlx5e_channel *volatile *pp)
1803 struct mlx5e_channel *c = *pp;
1805 /* check if channel is already closed */
1808 /* ensure channel pointer is no longer used */
1811 mlx5e_close_rq_wait(&c->rq);
1812 mlx5e_close_sqs_wait(c);
1813 mlx5e_close_cq(&c->rq.cq);
1814 mlx5e_close_tx_cqs(c);
1815 /* destroy mutexes */
1816 mlx5e_chan_mtx_destroy(c);
1821 mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs)
1825 r = priv->params.hw_lro_en ? priv->params.lro_wqe_sz :
1826 MLX5E_SW2MB_MTU(priv->ifp->if_mtu);
1827 if (r > MJUM16BYTES)
1832 else if (r > MJUMPAGESIZE)
1834 else if (r > MCLBYTES)
1840 * n + 1 must be a power of two, because stride size must be.
1841 * Stride size is 16 * (n + 1), as the first segment is
1844 for (n = howmany(r, MLX5E_MAX_RX_BYTES); !powerof2(n + 1); n++)
1853 mlx5e_build_rq_param(struct mlx5e_priv *priv,
1854 struct mlx5e_rq_param *param)
1856 void *rqc = param->rqc;
1857 void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
1860 mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs);
1861 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
1862 MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN);
1863 MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe) +
1864 nsegs * sizeof(struct mlx5_wqe_data_seg)));
1865 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_rq_size);
1866 MLX5_SET(wq, wq, pd, priv->pdn);
1868 param->wq.buf_numa_node = 0;
1869 param->wq.db_numa_node = 0;
1870 param->wq.linear = 1;
1874 mlx5e_build_sq_param(struct mlx5e_priv *priv,
1875 struct mlx5e_sq_param *param)
1877 void *sqc = param->sqc;
1878 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
1880 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_sq_size);
1881 MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
1882 MLX5_SET(wq, wq, pd, priv->pdn);
1884 param->wq.buf_numa_node = 0;
1885 param->wq.db_numa_node = 0;
1886 param->wq.linear = 1;
1890 mlx5e_build_common_cq_param(struct mlx5e_priv *priv,
1891 struct mlx5e_cq_param *param)
1893 void *cqc = param->cqc;
1895 MLX5_SET(cqc, cqc, uar_page, priv->cq_uar.index);
1899 mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
1900 struct mlx5e_cq_param *param)
1902 void *cqc = param->cqc;
1906 * TODO The sysctl to control on/off is a bool value for now, which means
1907 * we only support CSUM, once HASH is implemnted we'll need to address that.
1909 if (priv->params.cqe_zipping_en) {
1910 MLX5_SET(cqc, cqc, mini_cqe_res_format, MLX5_CQE_FORMAT_CSUM);
1911 MLX5_SET(cqc, cqc, cqe_compression_en, 1);
1914 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_rq_size);
1915 MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec);
1916 MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts);
1918 switch (priv->params.rx_cq_moderation_mode) {
1920 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1923 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
1924 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
1926 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1930 mlx5e_build_common_cq_param(priv, param);
1934 mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
1935 struct mlx5e_cq_param *param)
1937 void *cqc = param->cqc;
1939 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_sq_size);
1940 MLX5_SET(cqc, cqc, cq_period, priv->params.tx_cq_moderation_usec);
1941 MLX5_SET(cqc, cqc, cq_max_count, priv->params.tx_cq_moderation_pkts);
1943 switch (priv->params.tx_cq_moderation_mode) {
1945 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1948 if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe))
1949 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
1951 MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
1955 mlx5e_build_common_cq_param(priv, param);
1959 mlx5e_build_channel_param(struct mlx5e_priv *priv,
1960 struct mlx5e_channel_param *cparam)
1962 memset(cparam, 0, sizeof(*cparam));
1964 mlx5e_build_rq_param(priv, &cparam->rq);
1965 mlx5e_build_sq_param(priv, &cparam->sq);
1966 mlx5e_build_rx_cq_param(priv, &cparam->rx_cq);
1967 mlx5e_build_tx_cq_param(priv, &cparam->tx_cq);
1971 mlx5e_open_channels(struct mlx5e_priv *priv)
1973 struct mlx5e_channel_param cparam;
1979 priv->channel = malloc(priv->params.num_channels *
1980 sizeof(struct mlx5e_channel *), M_MLX5EN, M_WAITOK | M_ZERO);
1982 mlx5e_build_channel_param(priv, &cparam);
1983 for (i = 0; i < priv->params.num_channels; i++) {
1984 err = mlx5e_open_channel(priv, i, &cparam, &priv->channel[i]);
1986 goto err_close_channels;
1989 for (j = 0; j < priv->params.num_channels; j++) {
1990 err = mlx5e_wait_for_min_rx_wqes(&priv->channel[j]->rq);
1992 goto err_close_channels;
1998 for (i--; i >= 0; i--) {
1999 mlx5e_close_channel(&priv->channel[i]);
2000 mlx5e_close_channel_wait(&priv->channel[i]);
2003 /* remove "volatile" attribute from "channel" pointer */
2004 ptr = __DECONST(void *, priv->channel);
2005 priv->channel = NULL;
2007 free(ptr, M_MLX5EN);
2013 mlx5e_close_channels(struct mlx5e_priv *priv)
2018 if (priv->channel == NULL)
2021 for (i = 0; i < priv->params.num_channels; i++)
2022 mlx5e_close_channel(&priv->channel[i]);
2023 for (i = 0; i < priv->params.num_channels; i++)
2024 mlx5e_close_channel_wait(&priv->channel[i]);
2026 /* remove "volatile" attribute from "channel" pointer */
2027 ptr = __DECONST(void *, priv->channel);
2028 priv->channel = NULL;
2030 free(ptr, M_MLX5EN);
2034 mlx5e_refresh_sq_params(struct mlx5e_priv *priv, struct mlx5e_sq *sq)
2037 if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
2040 switch (priv->params.tx_cq_moderation_mode) {
2042 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2045 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2049 return (mlx5_core_modify_cq_moderation_mode(priv->mdev, &sq->cq.mcq,
2050 priv->params.tx_cq_moderation_usec,
2051 priv->params.tx_cq_moderation_pkts,
2055 return (mlx5_core_modify_cq_moderation(priv->mdev, &sq->cq.mcq,
2056 priv->params.tx_cq_moderation_usec,
2057 priv->params.tx_cq_moderation_pkts));
2061 mlx5e_refresh_rq_params(struct mlx5e_priv *priv, struct mlx5e_rq *rq)
2064 if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) {
2068 switch (priv->params.rx_cq_moderation_mode) {
2070 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2073 cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE;
2077 retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq,
2078 priv->params.rx_cq_moderation_usec,
2079 priv->params.rx_cq_moderation_pkts,
2085 return (mlx5_core_modify_cq_moderation(priv->mdev, &rq->cq.mcq,
2086 priv->params.rx_cq_moderation_usec,
2087 priv->params.rx_cq_moderation_pkts));
2091 mlx5e_refresh_channel_params_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c)
2099 err = mlx5e_refresh_rq_params(priv, &c->rq);
2103 for (i = 0; i != c->num_tc; i++) {
2104 err = mlx5e_refresh_sq_params(priv, &c->sq[i]);
2113 mlx5e_refresh_channel_params(struct mlx5e_priv *priv)
2117 if (priv->channel == NULL)
2120 for (i = 0; i < priv->params.num_channels; i++) {
2123 err = mlx5e_refresh_channel_params_sub(priv, priv->channel[i]);
2131 mlx5e_open_tis(struct mlx5e_priv *priv, int tc)
2133 struct mlx5_core_dev *mdev = priv->mdev;
2134 u32 in[MLX5_ST_SZ_DW(create_tis_in)];
2135 void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
2137 memset(in, 0, sizeof(in));
2139 MLX5_SET(tisc, tisc, prio, tc);
2140 MLX5_SET(tisc, tisc, transport_domain, priv->tdn);
2142 return (mlx5_core_create_tis(mdev, in, sizeof(in), &priv->tisn[tc]));
2146 mlx5e_close_tis(struct mlx5e_priv *priv, int tc)
2148 mlx5_core_destroy_tis(priv->mdev, priv->tisn[tc]);
2152 mlx5e_open_tises(struct mlx5e_priv *priv)
2154 int num_tc = priv->num_tc;
2158 for (tc = 0; tc < num_tc; tc++) {
2159 err = mlx5e_open_tis(priv, tc);
2161 goto err_close_tises;
2167 for (tc--; tc >= 0; tc--)
2168 mlx5e_close_tis(priv, tc);
2174 mlx5e_close_tises(struct mlx5e_priv *priv)
2176 int num_tc = priv->num_tc;
2179 for (tc = 0; tc < num_tc; tc++)
2180 mlx5e_close_tis(priv, tc);
2184 mlx5e_open_rqt(struct mlx5e_priv *priv)
2186 struct mlx5_core_dev *mdev = priv->mdev;
2188 u32 out[MLX5_ST_SZ_DW(create_rqt_out)] = {0};
2195 sz = 1 << priv->params.rx_hash_log_tbl_sz;
2197 inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
2198 in = mlx5_vzalloc(inlen);
2201 rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
2203 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
2204 MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
2206 for (i = 0; i < sz; i++) {
2209 ix = rss_get_indirection_to_bucket(ix);
2211 /* ensure we don't overflow */
2212 ix %= priv->params.num_channels;
2214 /* apply receive side scaling stride, if any */
2215 ix -= ix % (int)priv->params.channels_rsss;
2217 MLX5_SET(rqtc, rqtc, rq_num[i], priv->channel[ix]->rq.rqn);
2220 MLX5_SET(create_rqt_in, in, opcode, MLX5_CMD_OP_CREATE_RQT);
2222 err = mlx5_cmd_exec(mdev, in, inlen, out, sizeof(out));
2224 priv->rqtn = MLX5_GET(create_rqt_out, out, rqtn);
2232 mlx5e_close_rqt(struct mlx5e_priv *priv)
2234 u32 in[MLX5_ST_SZ_DW(destroy_rqt_in)] = {0};
2235 u32 out[MLX5_ST_SZ_DW(destroy_rqt_out)] = {0};
2237 MLX5_SET(destroy_rqt_in, in, opcode, MLX5_CMD_OP_DESTROY_RQT);
2238 MLX5_SET(destroy_rqt_in, in, rqtn, priv->rqtn);
2240 mlx5_cmd_exec(priv->mdev, in, sizeof(in), out, sizeof(out));
2244 mlx5e_build_tir_ctx(struct mlx5e_priv *priv, u32 * tirc, int tt)
2246 void *hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
2249 MLX5_SET(tirc, tirc, transport_domain, priv->tdn);
2251 #define ROUGH_MAX_L2_L3_HDR_SZ 256
2253 #define MLX5_HASH_IP (MLX5_HASH_FIELD_SEL_SRC_IP |\
2254 MLX5_HASH_FIELD_SEL_DST_IP)
2256 #define MLX5_HASH_ALL (MLX5_HASH_FIELD_SEL_SRC_IP |\
2257 MLX5_HASH_FIELD_SEL_DST_IP |\
2258 MLX5_HASH_FIELD_SEL_L4_SPORT |\
2259 MLX5_HASH_FIELD_SEL_L4_DPORT)
2261 #define MLX5_HASH_IP_IPSEC_SPI (MLX5_HASH_FIELD_SEL_SRC_IP |\
2262 MLX5_HASH_FIELD_SEL_DST_IP |\
2263 MLX5_HASH_FIELD_SEL_IPSEC_SPI)
2265 if (priv->params.hw_lro_en) {
2266 MLX5_SET(tirc, tirc, lro_enable_mask,
2267 MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO |
2268 MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO);
2269 MLX5_SET(tirc, tirc, lro_max_msg_sz,
2270 (priv->params.lro_wqe_sz -
2271 ROUGH_MAX_L2_L3_HDR_SZ) >> 8);
2272 /* TODO: add the option to choose timer value dynamically */
2273 MLX5_SET(tirc, tirc, lro_timeout_period_usecs,
2274 MLX5_CAP_ETH(priv->mdev,
2275 lro_timer_supported_periods[2]));
2278 /* setup parameters for hashing TIR type, if any */
2281 MLX5_SET(tirc, tirc, disp_type,
2282 MLX5_TIRC_DISP_TYPE_DIRECT);
2283 MLX5_SET(tirc, tirc, inline_rqn,
2284 priv->channel[0]->rq.rqn);
2287 MLX5_SET(tirc, tirc, disp_type,
2288 MLX5_TIRC_DISP_TYPE_INDIRECT);
2289 MLX5_SET(tirc, tirc, indirect_table,
2291 MLX5_SET(tirc, tirc, rx_hash_fn,
2292 MLX5_TIRC_RX_HASH_FN_HASH_TOEPLITZ);
2293 hkey = (__be32 *) MLX5_ADDR_OF(tirc, tirc, rx_hash_toeplitz_key);
2296 * The FreeBSD RSS implementation does currently not
2297 * support symmetric Toeplitz hashes:
2299 MLX5_SET(tirc, tirc, rx_hash_symmetric, 0);
2300 rss_getkey((uint8_t *)hkey);
2302 MLX5_SET(tirc, tirc, rx_hash_symmetric, 1);
2303 hkey[0] = cpu_to_be32(0xD181C62C);
2304 hkey[1] = cpu_to_be32(0xF7F4DB5B);
2305 hkey[2] = cpu_to_be32(0x1983A2FC);
2306 hkey[3] = cpu_to_be32(0x943E1ADB);
2307 hkey[4] = cpu_to_be32(0xD9389E6B);
2308 hkey[5] = cpu_to_be32(0xD1039C2C);
2309 hkey[6] = cpu_to_be32(0xA74499AD);
2310 hkey[7] = cpu_to_be32(0x593D56D9);
2311 hkey[8] = cpu_to_be32(0xF3253C06);
2312 hkey[9] = cpu_to_be32(0x2ADC1FFC);
2318 case MLX5E_TT_IPV4_TCP:
2319 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2320 MLX5_L3_PROT_TYPE_IPV4);
2321 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2322 MLX5_L4_PROT_TYPE_TCP);
2324 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4)) {
2325 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2329 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2333 case MLX5E_TT_IPV6_TCP:
2334 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2335 MLX5_L3_PROT_TYPE_IPV6);
2336 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2337 MLX5_L4_PROT_TYPE_TCP);
2339 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) {
2340 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2344 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2348 case MLX5E_TT_IPV4_UDP:
2349 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2350 MLX5_L3_PROT_TYPE_IPV4);
2351 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2352 MLX5_L4_PROT_TYPE_UDP);
2354 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4)) {
2355 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2359 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2363 case MLX5E_TT_IPV6_UDP:
2364 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2365 MLX5_L3_PROT_TYPE_IPV6);
2366 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2367 MLX5_L4_PROT_TYPE_UDP);
2369 if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) {
2370 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2374 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2378 case MLX5E_TT_IPV4_IPSEC_AH:
2379 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2380 MLX5_L3_PROT_TYPE_IPV4);
2381 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2382 MLX5_HASH_IP_IPSEC_SPI);
2385 case MLX5E_TT_IPV6_IPSEC_AH:
2386 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2387 MLX5_L3_PROT_TYPE_IPV6);
2388 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2389 MLX5_HASH_IP_IPSEC_SPI);
2392 case MLX5E_TT_IPV4_IPSEC_ESP:
2393 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2394 MLX5_L3_PROT_TYPE_IPV4);
2395 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2396 MLX5_HASH_IP_IPSEC_SPI);
2399 case MLX5E_TT_IPV6_IPSEC_ESP:
2400 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2401 MLX5_L3_PROT_TYPE_IPV6);
2402 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2403 MLX5_HASH_IP_IPSEC_SPI);
2407 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2408 MLX5_L3_PROT_TYPE_IPV4);
2409 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2414 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2415 MLX5_L3_PROT_TYPE_IPV6);
2416 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2426 mlx5e_open_tir(struct mlx5e_priv *priv, int tt)
2428 struct mlx5_core_dev *mdev = priv->mdev;
2434 inlen = MLX5_ST_SZ_BYTES(create_tir_in);
2435 in = mlx5_vzalloc(inlen);
2438 tirc = MLX5_ADDR_OF(create_tir_in, in, tir_context);
2440 mlx5e_build_tir_ctx(priv, tirc, tt);
2442 err = mlx5_core_create_tir(mdev, in, inlen, &priv->tirn[tt]);
2450 mlx5e_close_tir(struct mlx5e_priv *priv, int tt)
2452 mlx5_core_destroy_tir(priv->mdev, priv->tirn[tt]);
2456 mlx5e_open_tirs(struct mlx5e_priv *priv)
2461 for (i = 0; i < MLX5E_NUM_TT; i++) {
2462 err = mlx5e_open_tir(priv, i);
2464 goto err_close_tirs;
2470 for (i--; i >= 0; i--)
2471 mlx5e_close_tir(priv, i);
2477 mlx5e_close_tirs(struct mlx5e_priv *priv)
2481 for (i = 0; i < MLX5E_NUM_TT; i++)
2482 mlx5e_close_tir(priv, i);
2486 * SW MTU does not include headers,
2487 * HW MTU includes all headers and checksums.
2490 mlx5e_set_dev_port_mtu(struct ifnet *ifp, int sw_mtu)
2492 struct mlx5e_priv *priv = ifp->if_softc;
2493 struct mlx5_core_dev *mdev = priv->mdev;
2497 hw_mtu = MLX5E_SW2HW_MTU(sw_mtu);
2499 err = mlx5_set_port_mtu(mdev, hw_mtu);
2501 if_printf(ifp, "%s: mlx5_set_port_mtu failed setting %d, err=%d\n",
2502 __func__, sw_mtu, err);
2506 /* Update vport context MTU */
2507 err = mlx5_set_vport_mtu(mdev, hw_mtu);
2509 if_printf(ifp, "%s: Failed updating vport context with MTU size, err=%d\n",
2513 ifp->if_mtu = sw_mtu;
2515 err = mlx5_query_vport_mtu(mdev, &hw_mtu);
2516 if (err || !hw_mtu) {
2517 /* fallback to port oper mtu */
2518 err = mlx5_query_port_oper_mtu(mdev, &hw_mtu);
2521 if_printf(ifp, "Query port MTU, after setting new "
2522 "MTU value, failed\n");
2524 } else if (MLX5E_HW2SW_MTU(hw_mtu) < sw_mtu) {
2526 if_printf(ifp, "Port MTU %d is smaller than "
2527 "ifp mtu %d\n", hw_mtu, sw_mtu);
2528 } else if (MLX5E_HW2SW_MTU(hw_mtu) > sw_mtu) {
2530 if_printf(ifp, "Port MTU %d is bigger than "
2531 "ifp mtu %d\n", hw_mtu, sw_mtu);
2533 priv->params_ethtool.hw_mtu = hw_mtu;
2539 mlx5e_open_locked(struct ifnet *ifp)
2541 struct mlx5e_priv *priv = ifp->if_softc;
2545 /* check if already opened */
2546 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
2550 if (rss_getnumbuckets() > priv->params.num_channels) {
2551 if_printf(ifp, "NOTE: There are more RSS buckets(%u) than "
2552 "channels(%u) available\n", rss_getnumbuckets(),
2553 priv->params.num_channels);
2556 err = mlx5e_open_tises(priv);
2558 if_printf(ifp, "%s: mlx5e_open_tises failed, %d\n",
2562 err = mlx5_vport_alloc_q_counter(priv->mdev,
2563 MLX5_INTERFACE_PROTOCOL_ETH, &set_id);
2565 if_printf(priv->ifp,
2566 "%s: mlx5_vport_alloc_q_counter failed: %d\n",
2568 goto err_close_tises;
2570 /* store counter set ID */
2571 priv->counter_set_id = set_id;
2573 err = mlx5e_open_channels(priv);
2575 if_printf(ifp, "%s: mlx5e_open_channels failed, %d\n",
2577 goto err_dalloc_q_counter;
2579 err = mlx5e_open_rqt(priv);
2581 if_printf(ifp, "%s: mlx5e_open_rqt failed, %d\n",
2583 goto err_close_channels;
2585 err = mlx5e_open_tirs(priv);
2587 if_printf(ifp, "%s: mlx5e_open_tir failed, %d\n",
2589 goto err_close_rqls;
2591 err = mlx5e_open_flow_table(priv);
2593 if_printf(ifp, "%s: mlx5e_open_flow_table failed, %d\n",
2595 goto err_close_tirs;
2597 err = mlx5e_add_all_vlan_rules(priv);
2599 if_printf(ifp, "%s: mlx5e_add_all_vlan_rules failed, %d\n",
2601 goto err_close_flow_table;
2603 set_bit(MLX5E_STATE_OPENED, &priv->state);
2605 mlx5e_update_carrier(priv);
2606 mlx5e_set_rx_mode_core(priv);
2610 err_close_flow_table:
2611 mlx5e_close_flow_table(priv);
2614 mlx5e_close_tirs(priv);
2617 mlx5e_close_rqt(priv);
2620 mlx5e_close_channels(priv);
2622 err_dalloc_q_counter:
2623 mlx5_vport_dealloc_q_counter(priv->mdev,
2624 MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
2627 mlx5e_close_tises(priv);
2633 mlx5e_open(void *arg)
2635 struct mlx5e_priv *priv = arg;
2638 if (mlx5_set_port_status(priv->mdev, MLX5_PORT_UP))
2639 if_printf(priv->ifp,
2640 "%s: Setting port status to up failed\n",
2643 mlx5e_open_locked(priv->ifp);
2644 priv->ifp->if_drv_flags |= IFF_DRV_RUNNING;
2649 mlx5e_close_locked(struct ifnet *ifp)
2651 struct mlx5e_priv *priv = ifp->if_softc;
2653 /* check if already closed */
2654 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2657 clear_bit(MLX5E_STATE_OPENED, &priv->state);
2659 mlx5e_set_rx_mode_core(priv);
2660 mlx5e_del_all_vlan_rules(priv);
2661 if_link_state_change(priv->ifp, LINK_STATE_DOWN);
2662 mlx5e_close_flow_table(priv);
2663 mlx5e_close_tirs(priv);
2664 mlx5e_close_rqt(priv);
2665 mlx5e_close_channels(priv);
2666 mlx5_vport_dealloc_q_counter(priv->mdev,
2667 MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id);
2668 mlx5e_close_tises(priv);
2673 #if (__FreeBSD_version >= 1100000)
2675 mlx5e_get_counter(struct ifnet *ifp, ift_counter cnt)
2677 struct mlx5e_priv *priv = ifp->if_softc;
2680 /* PRIV_LOCK(priv); XXX not allowed */
2682 case IFCOUNTER_IPACKETS:
2683 retval = priv->stats.vport.rx_packets;
2685 case IFCOUNTER_IERRORS:
2686 retval = priv->stats.vport.rx_error_packets +
2687 priv->stats.pport.alignment_err +
2688 priv->stats.pport.check_seq_err +
2689 priv->stats.pport.crc_align_errors +
2690 priv->stats.pport.in_range_len_errors +
2691 priv->stats.pport.jabbers +
2692 priv->stats.pport.out_of_range_len +
2693 priv->stats.pport.oversize_pkts +
2694 priv->stats.pport.symbol_err +
2695 priv->stats.pport.too_long_errors +
2696 priv->stats.pport.undersize_pkts +
2697 priv->stats.pport.unsupported_op_rx;
2699 case IFCOUNTER_IQDROPS:
2700 retval = priv->stats.vport.rx_out_of_buffer +
2701 priv->stats.pport.drop_events;
2703 case IFCOUNTER_OPACKETS:
2704 retval = priv->stats.vport.tx_packets;
2706 case IFCOUNTER_OERRORS:
2707 retval = priv->stats.vport.tx_error_packets;
2709 case IFCOUNTER_IBYTES:
2710 retval = priv->stats.vport.rx_bytes;
2712 case IFCOUNTER_OBYTES:
2713 retval = priv->stats.vport.tx_bytes;
2715 case IFCOUNTER_IMCASTS:
2716 retval = priv->stats.vport.rx_multicast_packets;
2718 case IFCOUNTER_OMCASTS:
2719 retval = priv->stats.vport.tx_multicast_packets;
2721 case IFCOUNTER_OQDROPS:
2722 retval = priv->stats.vport.tx_queue_dropped;
2724 case IFCOUNTER_COLLISIONS:
2725 retval = priv->stats.pport.collisions;
2728 retval = if_get_counter_default(ifp, cnt);
2731 /* PRIV_UNLOCK(priv); XXX not allowed */
2737 mlx5e_set_rx_mode(struct ifnet *ifp)
2739 struct mlx5e_priv *priv = ifp->if_softc;
2741 queue_work(priv->wq, &priv->set_rx_mode_work);
2745 mlx5e_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2747 struct mlx5e_priv *priv;
2749 struct ifi2creq i2c;
2758 priv = ifp->if_softc;
2760 /* check if detaching */
2761 if (priv == NULL || priv->gone != 0)
2766 ifr = (struct ifreq *)data;
2769 mlx5_query_port_max_mtu(priv->mdev, &max_mtu);
2771 if (ifr->ifr_mtu >= MLX5E_MTU_MIN &&
2772 ifr->ifr_mtu <= MIN(MLX5E_MTU_MAX, max_mtu)) {
2775 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2777 mlx5e_close_locked(ifp);
2780 mlx5e_set_dev_port_mtu(ifp, ifr->ifr_mtu);
2783 mlx5e_open_locked(ifp);
2786 if_printf(ifp, "Invalid MTU value. Min val: %d, Max val: %d\n",
2787 MLX5E_MTU_MIN, MIN(MLX5E_MTU_MAX, max_mtu));
2792 if ((ifp->if_flags & IFF_UP) &&
2793 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
2794 mlx5e_set_rx_mode(ifp);
2798 if (ifp->if_flags & IFF_UP) {
2799 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2800 if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0)
2801 mlx5e_open_locked(ifp);
2802 ifp->if_drv_flags |= IFF_DRV_RUNNING;
2803 mlx5_set_port_status(priv->mdev, MLX5_PORT_UP);
2806 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
2807 mlx5_set_port_status(priv->mdev,
2809 if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0)
2810 mlx5e_close_locked(ifp);
2811 mlx5e_update_carrier(priv);
2812 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2819 mlx5e_set_rx_mode(ifp);
2824 ifr = (struct ifreq *)data;
2825 error = ifmedia_ioctl(ifp, ifr, &priv->media, command);
2828 ifr = (struct ifreq *)data;
2830 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2832 if (mask & IFCAP_TXCSUM) {
2833 ifp->if_capenable ^= IFCAP_TXCSUM;
2834 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
2836 if (IFCAP_TSO4 & ifp->if_capenable &&
2837 !(IFCAP_TXCSUM & ifp->if_capenable)) {
2838 ifp->if_capenable &= ~IFCAP_TSO4;
2839 ifp->if_hwassist &= ~CSUM_IP_TSO;
2841 "tso4 disabled due to -txcsum.\n");
2844 if (mask & IFCAP_TXCSUM_IPV6) {
2845 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
2846 ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
2848 if (IFCAP_TSO6 & ifp->if_capenable &&
2849 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2850 ifp->if_capenable &= ~IFCAP_TSO6;
2851 ifp->if_hwassist &= ~CSUM_IP6_TSO;
2853 "tso6 disabled due to -txcsum6.\n");
2856 if (mask & IFCAP_RXCSUM)
2857 ifp->if_capenable ^= IFCAP_RXCSUM;
2858 if (mask & IFCAP_RXCSUM_IPV6)
2859 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
2860 if (mask & IFCAP_TSO4) {
2861 if (!(IFCAP_TSO4 & ifp->if_capenable) &&
2862 !(IFCAP_TXCSUM & ifp->if_capenable)) {
2863 if_printf(ifp, "enable txcsum first.\n");
2867 ifp->if_capenable ^= IFCAP_TSO4;
2868 ifp->if_hwassist ^= CSUM_IP_TSO;
2870 if (mask & IFCAP_TSO6) {
2871 if (!(IFCAP_TSO6 & ifp->if_capenable) &&
2872 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2873 if_printf(ifp, "enable txcsum6 first.\n");
2877 ifp->if_capenable ^= IFCAP_TSO6;
2878 ifp->if_hwassist ^= CSUM_IP6_TSO;
2880 if (mask & IFCAP_VLAN_HWFILTER) {
2881 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2882 mlx5e_disable_vlan_filter(priv);
2884 mlx5e_enable_vlan_filter(priv);
2886 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
2888 if (mask & IFCAP_VLAN_HWTAGGING)
2889 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2890 if (mask & IFCAP_WOL_MAGIC)
2891 ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2893 VLAN_CAPABILITIES(ifp);
2894 /* turn off LRO means also turn of HW LRO - if it's on */
2895 if (mask & IFCAP_LRO) {
2896 int was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2897 bool need_restart = false;
2899 ifp->if_capenable ^= IFCAP_LRO;
2900 if (!(ifp->if_capenable & IFCAP_LRO)) {
2901 if (priv->params.hw_lro_en) {
2902 priv->params.hw_lro_en = false;
2903 need_restart = true;
2904 /* Not sure this is the correct way */
2905 priv->params_ethtool.hw_lro = priv->params.hw_lro_en;
2908 if (was_opened && need_restart) {
2909 mlx5e_close_locked(ifp);
2910 mlx5e_open_locked(ifp);
2913 if (mask & IFCAP_HWRXTSTMP) {
2914 ifp->if_capenable ^= IFCAP_HWRXTSTMP;
2915 if (ifp->if_capenable & IFCAP_HWRXTSTMP) {
2916 if (priv->clbr_done == 0)
2917 mlx5e_reset_calibration_callout(priv);
2919 callout_drain(&priv->tstmp_clbr);
2920 priv->clbr_done = 0;
2928 ifr = (struct ifreq *)data;
2931 * Copy from the user-space address ifr_data to the
2932 * kernel-space address i2c
2934 error = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
2938 if (i2c.len > sizeof(i2c.data)) {
2944 /* Get module_num which is required for the query_eeprom */
2945 error = mlx5_query_module_num(priv->mdev, &module_num);
2947 if_printf(ifp, "Query module num failed, eeprom "
2948 "reading is not supported\n");
2952 /* Check if module is present before doing an access */
2953 module_status = mlx5_query_module_status(priv->mdev, module_num);
2954 if (module_status != MLX5_MODULE_STATUS_PLUGGED_ENABLED &&
2955 module_status != MLX5_MODULE_STATUS_PLUGGED_DISABLED) {
2960 * Currently 0XA0 and 0xA2 are the only addresses permitted.
2961 * The internal conversion is as follows:
2963 if (i2c.dev_addr == 0xA0)
2964 read_addr = MLX5E_I2C_ADDR_LOW;
2965 else if (i2c.dev_addr == 0xA2)
2966 read_addr = MLX5E_I2C_ADDR_HIGH;
2968 if_printf(ifp, "Query eeprom failed, "
2969 "Invalid Address: %X\n", i2c.dev_addr);
2973 error = mlx5_query_eeprom(priv->mdev,
2974 read_addr, MLX5E_EEPROM_LOW_PAGE,
2975 (uint32_t)i2c.offset, (uint32_t)i2c.len, module_num,
2976 (uint32_t *)i2c.data, &size_read);
2978 if_printf(ifp, "Query eeprom failed, eeprom "
2979 "reading is not supported\n");
2984 if (i2c.len > MLX5_EEPROM_MAX_BYTES) {
2985 error = mlx5_query_eeprom(priv->mdev,
2986 read_addr, MLX5E_EEPROM_LOW_PAGE,
2987 (uint32_t)(i2c.offset + size_read),
2988 (uint32_t)(i2c.len - size_read), module_num,
2989 (uint32_t *)(i2c.data + size_read), &size_read);
2992 if_printf(ifp, "Query eeprom failed, eeprom "
2993 "reading is not supported\n");
2998 error = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
3004 error = ether_ioctl(ifp, command, data);
3011 mlx5e_check_required_hca_cap(struct mlx5_core_dev *mdev)
3014 * TODO: uncoment once FW really sets all these bits if
3015 * (!mdev->caps.eth.rss_ind_tbl_cap || !mdev->caps.eth.csum_cap ||
3016 * !mdev->caps.eth.max_lso_cap || !mdev->caps.eth.vlan_cap ||
3017 * !(mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_SCQE_BRK_MOD)) return
3021 /* TODO: add more must-to-have features */
3023 if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
3030 mlx5e_get_max_inline_cap(struct mlx5_core_dev *mdev)
3032 int bf_buf_size = (1 << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2;
3034 return bf_buf_size -
3035 sizeof(struct mlx5e_tx_wqe) +
3036 2 /*sizeof(mlx5e_tx_wqe.inline_hdr_start)*/;
3040 mlx5e_build_ifp_priv(struct mlx5_core_dev *mdev,
3041 struct mlx5e_priv *priv,
3042 int num_comp_vectors)
3045 * TODO: Consider link speed for setting "log_sq_size",
3046 * "log_rq_size" and "cq_moderation_xxx":
3048 priv->params.log_sq_size =
3049 MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
3050 priv->params.log_rq_size =
3051 MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
3052 priv->params.rx_cq_moderation_usec =
3053 MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
3054 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE :
3055 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
3056 priv->params.rx_cq_moderation_mode =
3057 MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ? 1 : 0;
3058 priv->params.rx_cq_moderation_pkts =
3059 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS;
3060 priv->params.tx_cq_moderation_usec =
3061 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC;
3062 priv->params.tx_cq_moderation_pkts =
3063 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
3064 priv->params.min_rx_wqes =
3065 MLX5E_PARAMS_DEFAULT_MIN_RX_WQES;
3066 priv->params.rx_hash_log_tbl_sz =
3067 (order_base_2(num_comp_vectors) >
3068 MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ) ?
3069 order_base_2(num_comp_vectors) :
3070 MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ;
3071 priv->params.num_tc = 1;
3072 priv->params.default_vlan_prio = 0;
3073 priv->counter_set_id = -1;
3074 priv->params.tx_max_inline = mlx5e_get_max_inline_cap(mdev);
3075 mlx5_query_min_inline(mdev, &priv->params.tx_min_inline_mode);
3078 * hw lro is currently defaulted to off. when it won't anymore we
3079 * will consider the HW capability: "!!MLX5_CAP_ETH(mdev, lro_cap)"
3081 priv->params.hw_lro_en = false;
3082 priv->params.lro_wqe_sz = MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ;
3084 priv->params.cqe_zipping_en = !!MLX5_CAP_GEN(mdev, cqe_compression);
3087 priv->params.num_channels = num_comp_vectors;
3088 priv->params.channels_rsss = 1;
3089 priv->order_base_2_num_channels = order_base_2(num_comp_vectors);
3090 priv->queue_mapping_channel_mask =
3091 roundup_pow_of_two(num_comp_vectors) - 1;
3092 priv->num_tc = priv->params.num_tc;
3093 priv->default_vlan_prio = priv->params.default_vlan_prio;
3095 INIT_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
3096 INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
3097 INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
3101 mlx5e_create_mkey(struct mlx5e_priv *priv, u32 pdn,
3102 struct mlx5_core_mr *mkey)
3104 struct ifnet *ifp = priv->ifp;
3105 struct mlx5_core_dev *mdev = priv->mdev;
3106 int inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
3111 in = mlx5_vzalloc(inlen);
3113 if_printf(ifp, "%s: failed to allocate inbox\n", __func__);
3117 mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
3118 MLX5_SET(mkc, mkc, access_mode, MLX5_ACCESS_MODE_PA);
3119 MLX5_SET(mkc, mkc, lw, 1);
3120 MLX5_SET(mkc, mkc, lr, 1);
3122 MLX5_SET(mkc, mkc, pd, pdn);
3123 MLX5_SET(mkc, mkc, length64, 1);
3124 MLX5_SET(mkc, mkc, qpn, 0xffffff);
3126 err = mlx5_core_create_mkey(mdev, mkey, in, inlen);
3128 if_printf(ifp, "%s: mlx5_core_create_mkey failed, %d\n",
3135 static const char *mlx5e_vport_stats_desc[] = {
3136 MLX5E_VPORT_STATS(MLX5E_STATS_DESC)
3139 static const char *mlx5e_pport_stats_desc[] = {
3140 MLX5E_PPORT_STATS(MLX5E_STATS_DESC)
3144 mlx5e_priv_mtx_init(struct mlx5e_priv *priv)
3146 mtx_init(&priv->async_events_mtx, "mlx5async", MTX_NETWORK_LOCK, MTX_DEF);
3147 sx_init(&priv->state_lock, "mlx5state");
3148 callout_init_mtx(&priv->watchdog, &priv->async_events_mtx, 0);
3149 MLX5_INIT_DOORBELL_LOCK(&priv->doorbell_lock);
3153 mlx5e_priv_mtx_destroy(struct mlx5e_priv *priv)
3155 mtx_destroy(&priv->async_events_mtx);
3156 sx_destroy(&priv->state_lock);
3160 sysctl_firmware(SYSCTL_HANDLER_ARGS)
3163 * %d.%d%.d the string format.
3164 * fw_rev_{maj,min,sub} return u16, 2^16 = 65536.
3165 * We need at most 5 chars to store that.
3166 * It also has: two "." and NULL at the end, which means we need 18
3167 * (5*3 + 3) chars at most.
3170 struct mlx5e_priv *priv = arg1;
3173 snprintf(fw, sizeof(fw), "%d.%d.%d", fw_rev_maj(priv->mdev), fw_rev_min(priv->mdev),
3174 fw_rev_sub(priv->mdev));
3175 error = sysctl_handle_string(oidp, fw, sizeof(fw), req);
3180 mlx5e_disable_tx_dma(struct mlx5e_channel *ch)
3184 for (i = 0; i < ch->num_tc; i++)
3185 mlx5e_drain_sq(&ch->sq[i]);
3189 mlx5e_reset_sq_doorbell_record(struct mlx5e_sq *sq)
3192 sq->doorbell.d32[0] = cpu_to_be32(MLX5_OPCODE_NOP);
3193 sq->doorbell.d32[1] = cpu_to_be32(sq->sqn << 8);
3194 mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
3195 sq->doorbell.d64 = 0;
3199 mlx5e_resume_sq(struct mlx5e_sq *sq)
3203 /* check if already enabled */
3204 if (sq->stopped == 0)
3207 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_ERR,
3208 MLX5_SQC_STATE_RST);
3211 "mlx5e_modify_sq() from ERR to RST failed: %d\n", err);
3217 /* reset doorbell prior to moving from RST to RDY */
3218 mlx5e_reset_sq_doorbell_record(sq);
3220 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST,
3221 MLX5_SQC_STATE_RDY);
3224 "mlx5e_modify_sq() from RST to RDY failed: %d\n", err);
3227 mtx_lock(&sq->lock);
3228 sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
3230 mtx_unlock(&sq->lock);
3235 mlx5e_enable_tx_dma(struct mlx5e_channel *ch)
3239 for (i = 0; i < ch->num_tc; i++)
3240 mlx5e_resume_sq(&ch->sq[i]);
3244 mlx5e_disable_rx_dma(struct mlx5e_channel *ch)
3246 struct mlx5e_rq *rq = &ch->rq;
3251 callout_stop(&rq->watchdog);
3252 mtx_unlock(&rq->mtx);
3254 callout_drain(&rq->watchdog);
3256 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
3259 "mlx5e_modify_rq() from RDY to RST failed: %d\n", err);
3262 while (!mlx5_wq_ll_is_empty(&rq->wq)) {
3264 rq->cq.mcq.comp(&rq->cq.mcq);
3268 * Transitioning into RST state will allow the FW to track less ERR state queues,
3269 * thus reducing the recv queue flushing time
3271 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_ERR, MLX5_RQC_STATE_RST);
3274 "mlx5e_modify_rq() from ERR to RST failed: %d\n", err);
3279 mlx5e_enable_rx_dma(struct mlx5e_channel *ch)
3281 struct mlx5e_rq *rq = &ch->rq;
3285 mlx5_wq_ll_update_db_record(&rq->wq);
3286 err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
3289 "mlx5e_modify_rq() from RST to RDY failed: %d\n", err);
3294 rq->cq.mcq.comp(&rq->cq.mcq);
3298 mlx5e_modify_tx_dma(struct mlx5e_priv *priv, uint8_t value)
3302 if (priv->channel == NULL)
3305 for (i = 0; i < priv->params.num_channels; i++) {
3307 if (!priv->channel[i])
3311 mlx5e_disable_tx_dma(priv->channel[i]);
3313 mlx5e_enable_tx_dma(priv->channel[i]);
3318 mlx5e_modify_rx_dma(struct mlx5e_priv *priv, uint8_t value)
3322 if (priv->channel == NULL)
3325 for (i = 0; i < priv->params.num_channels; i++) {
3327 if (!priv->channel[i])
3331 mlx5e_disable_rx_dma(priv->channel[i]);
3333 mlx5e_enable_rx_dma(priv->channel[i]);
3338 mlx5e_params_calculate_tx_min_inline(struct mlx5_core_dev *mdev)
3342 min_inline_mode = MLX5_INLINE_MODE_L2;
3343 mlx5_query_min_inline(mdev, &min_inline_mode);
3344 if (min_inline_mode == MLX5_INLINE_MODE_NONE &&
3345 !MLX5_CAP_ETH(mdev, wqe_vlan_insert))
3346 min_inline_mode = MLX5_INLINE_MODE_L2;
3348 return (min_inline_mode);
3352 mlx5e_add_hw_stats(struct mlx5e_priv *priv)
3354 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3355 OID_AUTO, "fw_version", CTLTYPE_STRING | CTLFLAG_RD, priv, 0,
3356 sysctl_firmware, "A", "HCA firmware version");
3358 SYSCTL_ADD_STRING(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw),
3359 OID_AUTO, "board_id", CTLFLAG_RD, priv->mdev->board_id, 0,
3364 mlx5e_sysctl_tx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3366 struct mlx5e_priv *priv = arg1;
3373 tx_pfc = priv->params.tx_priority_flow_control;
3375 /* get current value */
3376 value = (tx_pfc >> arg2) & 1;
3378 error = sysctl_handle_32(oidp, &value, 0, req);
3380 /* range check value */
3382 priv->params.tx_priority_flow_control |= (1 << arg2);
3384 priv->params.tx_priority_flow_control &= ~(1 << arg2);
3386 /* check if update is required */
3387 if (error == 0 && priv->gone == 0 &&
3388 tx_pfc != priv->params.tx_priority_flow_control) {
3389 error = -mlx5e_set_port_pfc(priv);
3390 /* restore previous value */
3392 priv->params.tx_priority_flow_control= tx_pfc;
3400 mlx5e_sysctl_rx_priority_flow_control(SYSCTL_HANDLER_ARGS)
3402 struct mlx5e_priv *priv = arg1;
3409 rx_pfc = priv->params.rx_priority_flow_control;
3411 /* get current value */
3412 value = (rx_pfc >> arg2) & 1;
3414 error = sysctl_handle_32(oidp, &value, 0, req);
3416 /* range check value */
3418 priv->params.rx_priority_flow_control |= (1 << arg2);
3420 priv->params.rx_priority_flow_control &= ~(1 << arg2);
3422 /* check if update is required */
3423 if (error == 0 && priv->gone == 0 &&
3424 rx_pfc != priv->params.rx_priority_flow_control) {
3425 error = -mlx5e_set_port_pfc(priv);
3426 /* restore previous value */
3428 priv->params.rx_priority_flow_control= rx_pfc;
3436 mlx5e_setup_pauseframes(struct mlx5e_priv *priv)
3442 /* enable pauseframes by default */
3443 priv->params.tx_pauseframe_control = 1;
3444 priv->params.rx_pauseframe_control = 1;
3446 /* disable ports flow control, PFC, by default */
3447 priv->params.tx_priority_flow_control = 0;
3448 priv->params.rx_priority_flow_control = 0;
3450 #if (__FreeBSD_version < 1100000)
3451 /* compute path for sysctl */
3452 snprintf(path, sizeof(path), "dev.mce.%d.tx_pauseframe_control",
3453 device_get_unit(priv->mdev->pdev->dev.bsddev));
3455 /* try to fetch tunable, if any */
3456 TUNABLE_INT_FETCH(path, &priv->params.tx_pauseframe_control);
3458 /* compute path for sysctl */
3459 snprintf(path, sizeof(path), "dev.mce.%d.rx_pauseframe_control",
3460 device_get_unit(priv->mdev->pdev->dev.bsddev));
3462 /* try to fetch tunable, if any */
3463 TUNABLE_INT_FETCH(path, &priv->params.rx_pauseframe_control);
3465 for (x = 0; x != 8; x++) {
3467 /* compute path for sysctl */
3468 snprintf(path, sizeof(path), "dev.mce.%d.tx_priority_flow_control_%u",
3469 device_get_unit(priv->mdev->pdev->dev.bsddev), x);
3471 /* try to fetch tunable, if any */
3472 if (TUNABLE_INT_FETCH(path, &value) == 0 && value != 0)
3473 priv->params.tx_priority_flow_control |= 1 << x;
3475 /* compute path for sysctl */
3476 snprintf(path, sizeof(path), "dev.mce.%d.rx_priority_flow_control_%u",
3477 device_get_unit(priv->mdev->pdev->dev.bsddev), x);
3479 /* try to fetch tunable, if any */
3480 if (TUNABLE_INT_FETCH(path, &value) == 0 && value != 0)
3481 priv->params.rx_priority_flow_control |= 1 << x;
3485 /* register pauseframe SYSCTLs */
3486 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3487 OID_AUTO, "tx_pauseframe_control", CTLFLAG_RDTUN,
3488 &priv->params.tx_pauseframe_control, 0,
3489 "Set to enable TX pause frames. Clear to disable.");
3491 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3492 OID_AUTO, "rx_pauseframe_control", CTLFLAG_RDTUN,
3493 &priv->params.rx_pauseframe_control, 0,
3494 "Set to enable RX pause frames. Clear to disable.");
3496 /* register priority_flow control, PFC, SYSCTLs */
3497 for (x = 0; x != 8; x++) {
3498 snprintf(path, sizeof(path), "tx_priority_flow_control_%u", x);
3500 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3501 OID_AUTO, path, CTLTYPE_UINT | CTLFLAG_RWTUN |
3502 CTLFLAG_MPSAFE, priv, x, &mlx5e_sysctl_tx_priority_flow_control, "IU",
3503 "Set to enable TX ports flow control frames for given priority. Clear to disable.");
3505 snprintf(path, sizeof(path), "rx_priority_flow_control_%u", x);
3507 SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3508 OID_AUTO, path, CTLTYPE_UINT | CTLFLAG_RWTUN |
3509 CTLFLAG_MPSAFE, priv, x, &mlx5e_sysctl_rx_priority_flow_control, "IU",
3510 "Set to enable RX ports flow control frames for given priority. Clear to disable.");
3516 priv->params.tx_pauseframe_control =
3517 priv->params.tx_pauseframe_control ? 1 : 0;
3518 priv->params.rx_pauseframe_control =
3519 priv->params.rx_pauseframe_control ? 1 : 0;
3521 /* update firmware */
3522 error = mlx5e_set_port_pause_and_pfc(priv);
3523 if (error == -EINVAL) {
3524 if_printf(priv->ifp,
3525 "Global pauseframes must be disabled before enabling PFC.\n");
3526 priv->params.rx_priority_flow_control = 0;
3527 priv->params.tx_priority_flow_control = 0;
3529 /* update firmware */
3530 (void) mlx5e_set_port_pause_and_pfc(priv);
3536 mlx5e_create_ifp(struct mlx5_core_dev *mdev)
3539 struct mlx5e_priv *priv;
3540 u8 dev_addr[ETHER_ADDR_LEN] __aligned(4);
3541 struct sysctl_oid_list *child;
3542 int ncv = mdev->priv.eq_table.num_comp_vectors;
3548 if (mlx5e_check_required_hca_cap(mdev)) {
3549 mlx5_core_dbg(mdev, "mlx5e_check_required_hca_cap() failed\n");
3552 priv = malloc(sizeof(*priv), M_MLX5EN, M_WAITOK | M_ZERO);
3553 mlx5e_priv_mtx_init(priv);
3555 ifp = priv->ifp = if_alloc(IFT_ETHER);
3557 mlx5_core_err(mdev, "if_alloc() failed\n");
3560 ifp->if_softc = priv;
3561 if_initname(ifp, "mce", device_get_unit(mdev->pdev->dev.bsddev));
3562 ifp->if_mtu = ETHERMTU;
3563 ifp->if_init = mlx5e_open;
3564 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
3565 ifp->if_ioctl = mlx5e_ioctl;
3566 ifp->if_transmit = mlx5e_xmit;
3567 ifp->if_qflush = if_qflush;
3568 #if (__FreeBSD_version >= 1100000)
3569 ifp->if_get_counter = mlx5e_get_counter;
3571 ifp->if_snd.ifq_maxlen = ifqmaxlen;
3573 * Set driver features
3575 ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6;
3576 ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
3577 ifp->if_capabilities |= IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWFILTER;
3578 ifp->if_capabilities |= IFCAP_LINKSTATE | IFCAP_JUMBO_MTU;
3579 ifp->if_capabilities |= IFCAP_LRO;
3580 ifp->if_capabilities |= IFCAP_TSO | IFCAP_VLAN_HWTSO;
3581 ifp->if_capabilities |= IFCAP_HWSTATS | IFCAP_HWRXTSTMP;
3583 ifp->if_capabilities |= IFCAP_TXRTLMT;
3584 ifp->if_snd_tag_alloc = mlx5e_rl_snd_tag_alloc;
3585 ifp->if_snd_tag_free = mlx5e_rl_snd_tag_free;
3586 ifp->if_snd_tag_modify = mlx5e_rl_snd_tag_modify;
3587 ifp->if_snd_tag_query = mlx5e_rl_snd_tag_query;
3590 /* set TSO limits so that we don't have to drop TX packets */
3591 ifp->if_hw_tsomax = MLX5E_MAX_TX_PAYLOAD_SIZE - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
3592 ifp->if_hw_tsomaxsegcount = MLX5E_MAX_TX_MBUF_FRAGS - 1 /* hdr */;
3593 ifp->if_hw_tsomaxsegsize = MLX5E_MAX_TX_MBUF_SIZE;
3595 ifp->if_capenable = ifp->if_capabilities;
3596 ifp->if_hwassist = 0;
3597 if (ifp->if_capenable & IFCAP_TSO)
3598 ifp->if_hwassist |= CSUM_TSO;
3599 if (ifp->if_capenable & IFCAP_TXCSUM)
3600 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP | CSUM_IP);
3601 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
3602 ifp->if_hwassist |= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
3604 sysctl_ctx_init(&priv->sysctl_ctx_channel_debug);
3606 /* ifnet sysctl tree */
3607 sysctl_ctx_init(&priv->sysctl_ctx);
3608 priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_dev),
3609 OID_AUTO, ifp->if_dname, CTLFLAG_RD, 0, "MLX5 ethernet - interface name");
3610 if (priv->sysctl_ifnet == NULL) {
3611 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3612 goto err_free_sysctl;
3614 snprintf(unit, sizeof(unit), "%d", ifp->if_dunit);
3615 priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3616 OID_AUTO, unit, CTLFLAG_RD, 0, "MLX5 ethernet - interface unit");
3617 if (priv->sysctl_ifnet == NULL) {
3618 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3619 goto err_free_sysctl;
3622 /* HW sysctl tree */
3623 child = SYSCTL_CHILDREN(device_get_sysctl_tree(mdev->pdev->dev.bsddev));
3624 priv->sysctl_hw = SYSCTL_ADD_NODE(&priv->sysctl_ctx, child,
3625 OID_AUTO, "hw", CTLFLAG_RD, 0, "MLX5 ethernet dev hw");
3626 if (priv->sysctl_hw == NULL) {
3627 mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n");
3628 goto err_free_sysctl;
3630 mlx5e_build_ifp_priv(mdev, priv, ncv);
3632 snprintf(unit, sizeof(unit), "mce%u_wq",
3633 device_get_unit(mdev->pdev->dev.bsddev));
3634 priv->wq = alloc_workqueue(unit, 0, 1);
3635 if (priv->wq == NULL) {
3636 if_printf(ifp, "%s: alloc_workqueue failed\n", __func__);
3637 goto err_free_sysctl;
3640 err = mlx5_alloc_map_uar(mdev, &priv->cq_uar);
3642 if_printf(ifp, "%s: mlx5_alloc_map_uar failed, %d\n",
3646 err = mlx5_core_alloc_pd(mdev, &priv->pdn);
3648 if_printf(ifp, "%s: mlx5_core_alloc_pd failed, %d\n",
3650 goto err_unmap_free_uar;
3652 err = mlx5_alloc_transport_domain(mdev, &priv->tdn);
3654 if_printf(ifp, "%s: mlx5_alloc_transport_domain failed, %d\n",
3656 goto err_dealloc_pd;
3658 err = mlx5e_create_mkey(priv, priv->pdn, &priv->mr);
3660 if_printf(ifp, "%s: mlx5e_create_mkey failed, %d\n",
3662 goto err_dealloc_transport_domain;
3664 mlx5_query_nic_vport_mac_address(priv->mdev, 0, dev_addr);
3666 /* check if we should generate a random MAC address */
3667 if (MLX5_CAP_GEN(priv->mdev, vport_group_manager) == 0 &&
3668 is_zero_ether_addr(dev_addr)) {
3669 random_ether_addr(dev_addr);
3670 if_printf(ifp, "Assigned random MAC address\n");
3673 err = mlx5e_rl_init(priv);
3675 if_printf(ifp, "%s: mlx5e_rl_init failed, %d\n",
3677 goto err_create_mkey;
3681 /* set default MTU */
3682 mlx5e_set_dev_port_mtu(ifp, ifp->if_mtu);
3685 device_set_desc(mdev->pdev->dev.bsddev, mlx5e_version);
3687 /* Set default media status */
3688 priv->media_status_last = IFM_AVALID;
3689 priv->media_active_last = IFM_ETHER | IFM_AUTO |
3690 IFM_ETH_RXPAUSE | IFM_FDX;
3692 /* setup default pauseframes configuration */
3693 mlx5e_setup_pauseframes(priv);
3695 err = mlx5_query_port_proto_cap(mdev, ð_proto_cap, MLX5_PTYS_EN);
3698 if_printf(ifp, "%s: Query port media capability failed, %d\n",
3702 /* Setup supported medias */
3703 ifmedia_init(&priv->media, IFM_IMASK | IFM_ETH_FMASK,
3704 mlx5e_media_change, mlx5e_media_status);
3706 for (i = 0; i < MLX5E_LINK_MODES_NUMBER; ++i) {
3707 if (mlx5e_mode_table[i].baudrate == 0)
3709 if (MLX5E_PROT_MASK(i) & eth_proto_cap) {
3710 ifmedia_add(&priv->media,
3711 mlx5e_mode_table[i].subtype |
3712 IFM_ETHER, 0, NULL);
3713 ifmedia_add(&priv->media,
3714 mlx5e_mode_table[i].subtype |
3715 IFM_ETHER | IFM_FDX |
3716 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3720 ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO, 0, NULL);
3721 ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
3722 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL);
3724 /* Set autoselect by default */
3725 ifmedia_set(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX |
3726 IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE);
3727 ether_ifattach(ifp, dev_addr);
3729 /* Register for VLAN events */
3730 priv->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
3731 mlx5e_vlan_rx_add_vid, priv, EVENTHANDLER_PRI_FIRST);
3732 priv->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
3733 mlx5e_vlan_rx_kill_vid, priv, EVENTHANDLER_PRI_FIRST);
3735 /* Link is down by default */
3736 if_link_state_change(ifp, LINK_STATE_DOWN);
3738 mlx5e_enable_async_events(priv);
3740 mlx5e_add_hw_stats(priv);
3742 mlx5e_create_stats(&priv->stats.vport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3743 "vstats", mlx5e_vport_stats_desc, MLX5E_VPORT_STATS_NUM,
3744 priv->stats.vport.arg);
3746 mlx5e_create_stats(&priv->stats.pport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3747 "pstats", mlx5e_pport_stats_desc, MLX5E_PPORT_STATS_NUM,
3748 priv->stats.pport.arg);
3750 mlx5e_create_ethtool(priv);
3752 mtx_lock(&priv->async_events_mtx);
3753 mlx5e_update_stats(priv);
3754 mtx_unlock(&priv->async_events_mtx);
3756 SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet),
3757 OID_AUTO, "rx_clbr_done", CTLFLAG_RD,
3758 &priv->clbr_done, 0,
3759 "RX timestamps calibration state");
3760 callout_init(&priv->tstmp_clbr, CALLOUT_DIRECT);
3761 mlx5e_reset_calibration_callout(priv);
3767 mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
3769 err_dealloc_transport_domain:
3770 mlx5_dealloc_transport_domain(mdev, priv->tdn);
3773 mlx5_core_dealloc_pd(mdev, priv->pdn);
3776 mlx5_unmap_free_uar(mdev, &priv->cq_uar);
3779 destroy_workqueue(priv->wq);
3782 sysctl_ctx_free(&priv->sysctl_ctx);
3783 sysctl_ctx_free(&priv->sysctl_ctx_channel_debug);
3788 mlx5e_priv_mtx_destroy(priv);
3789 free(priv, M_MLX5EN);
3794 mlx5e_destroy_ifp(struct mlx5_core_dev *mdev, void *vpriv)
3796 struct mlx5e_priv *priv = vpriv;
3797 struct ifnet *ifp = priv->ifp;
3799 /* don't allow more IOCTLs */
3803 * Clear the device description to avoid use after free,
3804 * because the bsddev is not destroyed when this module is
3807 device_set_desc(mdev->pdev->dev.bsddev, NULL);
3809 /* XXX wait a bit to allow IOCTL handlers to complete */
3814 * The kernel can have reference(s) via the m_snd_tag's into
3815 * the ratelimit channels, and these must go away before
3818 while (READ_ONCE(priv->rl.stats.tx_active_connections) != 0) {
3819 if_printf(priv->ifp, "Waiting for all ratelimit connections "
3824 /* stop watchdog timer */
3825 callout_drain(&priv->watchdog);
3827 callout_drain(&priv->tstmp_clbr);
3829 if (priv->vlan_attach != NULL)
3830 EVENTHANDLER_DEREGISTER(vlan_config, priv->vlan_attach);
3831 if (priv->vlan_detach != NULL)
3832 EVENTHANDLER_DEREGISTER(vlan_unconfig, priv->vlan_detach);
3834 /* make sure device gets closed */
3836 mlx5e_close_locked(ifp);
3839 /* unregister device */
3840 ifmedia_removeall(&priv->media);
3841 ether_ifdetach(ifp);
3845 mlx5e_rl_cleanup(priv);
3847 /* destroy all remaining sysctl nodes */
3848 if (priv->sysctl_debug) {
3849 sysctl_ctx_free(&priv->sysctl_ctx_channel_debug);
3850 sysctl_ctx_free(&priv->stats.port_stats_debug.ctx);
3852 sysctl_ctx_free(&priv->stats.vport.ctx);
3853 sysctl_ctx_free(&priv->stats.pport.ctx);
3854 sysctl_ctx_free(&priv->sysctl_ctx);
3856 mlx5_core_destroy_mkey(priv->mdev, &priv->mr);
3857 mlx5_dealloc_transport_domain(priv->mdev, priv->tdn);
3858 mlx5_core_dealloc_pd(priv->mdev, priv->pdn);
3859 mlx5_unmap_free_uar(priv->mdev, &priv->cq_uar);
3860 mlx5e_disable_async_events(priv);
3861 destroy_workqueue(priv->wq);
3862 mlx5e_priv_mtx_destroy(priv);
3863 free(priv, M_MLX5EN);
3867 mlx5e_get_ifp(void *vpriv)
3869 struct mlx5e_priv *priv = vpriv;
3874 static struct mlx5_interface mlx5e_interface = {
3875 .add = mlx5e_create_ifp,
3876 .remove = mlx5e_destroy_ifp,
3877 .event = mlx5e_async_event,
3878 .protocol = MLX5_INTERFACE_PROTOCOL_ETH,
3879 .get_dev = mlx5e_get_ifp,
3885 mlx5_register_interface(&mlx5e_interface);
3891 mlx5_unregister_interface(&mlx5e_interface);
3894 module_init_order(mlx5e_init, SI_ORDER_THIRD);
3895 module_exit_order(mlx5e_cleanup, SI_ORDER_THIRD);
3897 #if (__FreeBSD_version >= 1100000)
3898 MODULE_DEPEND(mlx5en, linuxkpi, 1, 1, 1);
3900 MODULE_DEPEND(mlx5en, mlx5, 1, 1, 1);
3901 MODULE_VERSION(mlx5en, 1);