2 * Copyright (c) 2015-2016, Mellanox Technologies. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <net/tc_act/tc_gact.h>
34 #include <net/pkt_cls.h>
35 #include <linux/mlx5/fs.h>
36 #include <net/vxlan.h>
37 #include <linux/bpf.h>
43 struct mlx5e_rq_param {
44 u32 rqc[MLX5_ST_SZ_DW(rqc)];
45 struct mlx5_wq_param wq;
49 struct mlx5e_sq_param {
50 u32 sqc[MLX5_ST_SZ_DW(sqc)];
51 struct mlx5_wq_param wq;
54 enum mlx5e_sq_type type;
57 struct mlx5e_cq_param {
58 u32 cqc[MLX5_ST_SZ_DW(cqc)];
59 struct mlx5_wq_param wq;
64 struct mlx5e_channel_param {
65 struct mlx5e_rq_param rq;
66 struct mlx5e_sq_param sq;
67 struct mlx5e_sq_param xdp_sq;
68 struct mlx5e_sq_param icosq;
69 struct mlx5e_cq_param rx_cq;
70 struct mlx5e_cq_param tx_cq;
71 struct mlx5e_cq_param icosq_cq;
74 static bool mlx5e_check_fragmented_striding_rq_cap(struct mlx5_core_dev *mdev)
76 return MLX5_CAP_GEN(mdev, striding_rq) &&
77 MLX5_CAP_GEN(mdev, umr_ptr_rlky) &&
78 MLX5_CAP_ETH(mdev, reg_umr_sq);
81 static void mlx5e_set_rq_type_params(struct mlx5e_priv *priv, u8 rq_type)
83 priv->params.rq_wq_type = rq_type;
84 switch (priv->params.rq_wq_type) {
85 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
86 priv->params.log_rq_size = MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE_MPW;
87 priv->params.mpwqe_log_stride_sz =
88 MLX5E_GET_PFLAG(priv, MLX5E_PFLAG_RX_CQE_COMPRESS) ?
89 MLX5_MPWRQ_LOG_STRIDE_SIZE_CQE_COMPRESS :
90 MLX5_MPWRQ_LOG_STRIDE_SIZE;
91 priv->params.mpwqe_log_num_strides = MLX5_MPWRQ_LOG_WQE_SZ -
92 priv->params.mpwqe_log_stride_sz;
94 default: /* MLX5_WQ_TYPE_LINKED_LIST */
95 priv->params.log_rq_size = MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
97 priv->params.min_rx_wqes = mlx5_min_rx_wqes(priv->params.rq_wq_type,
98 BIT(priv->params.log_rq_size));
100 mlx5_core_info(priv->mdev,
101 "MLX5E: StrdRq(%d) RqSz(%ld) StrdSz(%ld) RxCqeCmprss(%d)\n",
102 priv->params.rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ,
103 BIT(priv->params.log_rq_size),
104 BIT(priv->params.mpwqe_log_stride_sz),
105 MLX5E_GET_PFLAG(priv, MLX5E_PFLAG_RX_CQE_COMPRESS));
108 static void mlx5e_set_rq_priv_params(struct mlx5e_priv *priv)
110 u8 rq_type = mlx5e_check_fragmented_striding_rq_cap(priv->mdev) &&
112 MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ :
113 MLX5_WQ_TYPE_LINKED_LIST;
114 mlx5e_set_rq_type_params(priv, rq_type);
117 static void mlx5e_update_carrier(struct mlx5e_priv *priv)
119 struct mlx5_core_dev *mdev = priv->mdev;
122 port_state = mlx5_query_vport_state(mdev,
123 MLX5_QUERY_VPORT_STATE_IN_OP_MOD_VNIC_VPORT, 0);
125 if (port_state == VPORT_STATE_UP) {
126 netdev_info(priv->netdev, "Link up\n");
127 netif_carrier_on(priv->netdev);
129 netdev_info(priv->netdev, "Link down\n");
130 netif_carrier_off(priv->netdev);
134 static void mlx5e_update_carrier_work(struct work_struct *work)
136 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
137 update_carrier_work);
139 mutex_lock(&priv->state_lock);
140 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
141 mlx5e_update_carrier(priv);
142 mutex_unlock(&priv->state_lock);
145 static void mlx5e_tx_timeout_work(struct work_struct *work)
147 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
152 mutex_lock(&priv->state_lock);
153 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
155 mlx5e_close_locked(priv->netdev);
156 err = mlx5e_open_locked(priv->netdev);
158 netdev_err(priv->netdev, "mlx5e_open_locked failed recovering from a tx_timeout, err(%d).\n",
161 mutex_unlock(&priv->state_lock);
165 static void mlx5e_update_sw_counters(struct mlx5e_priv *priv)
167 struct mlx5e_sw_stats *s = &priv->stats.sw;
168 struct mlx5e_rq_stats *rq_stats;
169 struct mlx5e_sq_stats *sq_stats;
170 u64 tx_offload_none = 0;
173 memset(s, 0, sizeof(*s));
174 for (i = 0; i < priv->params.num_channels; i++) {
175 rq_stats = &priv->channel[i]->rq.stats;
177 s->rx_packets += rq_stats->packets;
178 s->rx_bytes += rq_stats->bytes;
179 s->rx_lro_packets += rq_stats->lro_packets;
180 s->rx_lro_bytes += rq_stats->lro_bytes;
181 s->rx_csum_none += rq_stats->csum_none;
182 s->rx_csum_complete += rq_stats->csum_complete;
183 s->rx_csum_unnecessary_inner += rq_stats->csum_unnecessary_inner;
184 s->rx_xdp_drop += rq_stats->xdp_drop;
185 s->rx_xdp_tx += rq_stats->xdp_tx;
186 s->rx_xdp_tx_full += rq_stats->xdp_tx_full;
187 s->rx_wqe_err += rq_stats->wqe_err;
188 s->rx_mpwqe_filler += rq_stats->mpwqe_filler;
189 s->rx_buff_alloc_err += rq_stats->buff_alloc_err;
190 s->rx_cqe_compress_blks += rq_stats->cqe_compress_blks;
191 s->rx_cqe_compress_pkts += rq_stats->cqe_compress_pkts;
192 s->rx_cache_reuse += rq_stats->cache_reuse;
193 s->rx_cache_full += rq_stats->cache_full;
194 s->rx_cache_empty += rq_stats->cache_empty;
195 s->rx_cache_busy += rq_stats->cache_busy;
197 for (j = 0; j < priv->params.num_tc; j++) {
198 sq_stats = &priv->channel[i]->sq[j].stats;
200 s->tx_packets += sq_stats->packets;
201 s->tx_bytes += sq_stats->bytes;
202 s->tx_tso_packets += sq_stats->tso_packets;
203 s->tx_tso_bytes += sq_stats->tso_bytes;
204 s->tx_tso_inner_packets += sq_stats->tso_inner_packets;
205 s->tx_tso_inner_bytes += sq_stats->tso_inner_bytes;
206 s->tx_queue_stopped += sq_stats->stopped;
207 s->tx_queue_wake += sq_stats->wake;
208 s->tx_queue_dropped += sq_stats->dropped;
209 s->tx_xmit_more += sq_stats->xmit_more;
210 s->tx_csum_partial_inner += sq_stats->csum_partial_inner;
211 tx_offload_none += sq_stats->csum_none;
215 /* Update calculated offload counters */
216 s->tx_csum_partial = s->tx_packets - tx_offload_none - s->tx_csum_partial_inner;
217 s->rx_csum_unnecessary = s->rx_packets - s->rx_csum_none - s->rx_csum_complete;
219 s->link_down_events_phy = MLX5_GET(ppcnt_reg,
220 priv->stats.pport.phy_counters,
221 counter_set.phys_layer_cntrs.link_down_events);
224 static void mlx5e_update_vport_counters(struct mlx5e_priv *priv)
226 int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
227 u32 *out = (u32 *)priv->stats.vport.query_vport_out;
228 u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)] = {0};
229 struct mlx5_core_dev *mdev = priv->mdev;
231 MLX5_SET(query_vport_counter_in, in, opcode,
232 MLX5_CMD_OP_QUERY_VPORT_COUNTER);
233 MLX5_SET(query_vport_counter_in, in, op_mod, 0);
234 MLX5_SET(query_vport_counter_in, in, other_vport, 0);
236 memset(out, 0, outlen);
237 mlx5_cmd_exec(mdev, in, sizeof(in), out, outlen);
240 static void mlx5e_update_pport_counters(struct mlx5e_priv *priv)
242 struct mlx5e_pport_stats *pstats = &priv->stats.pport;
243 struct mlx5_core_dev *mdev = priv->mdev;
244 int sz = MLX5_ST_SZ_BYTES(ppcnt_reg);
249 in = mlx5_vzalloc(sz);
253 MLX5_SET(ppcnt_reg, in, local_port, 1);
255 out = pstats->IEEE_802_3_counters;
256 MLX5_SET(ppcnt_reg, in, grp, MLX5_IEEE_802_3_COUNTERS_GROUP);
257 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
259 out = pstats->RFC_2863_counters;
260 MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2863_COUNTERS_GROUP);
261 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
263 out = pstats->RFC_2819_counters;
264 MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2819_COUNTERS_GROUP);
265 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
267 out = pstats->phy_counters;
268 MLX5_SET(ppcnt_reg, in, grp, MLX5_PHYSICAL_LAYER_COUNTERS_GROUP);
269 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
271 MLX5_SET(ppcnt_reg, in, grp, MLX5_PER_PRIORITY_COUNTERS_GROUP);
272 for (prio = 0; prio < NUM_PPORT_PRIO; prio++) {
273 out = pstats->per_prio_counters[prio];
274 MLX5_SET(ppcnt_reg, in, prio_tc, prio);
275 mlx5_core_access_reg(mdev, in, sz, out, sz,
276 MLX5_REG_PPCNT, 0, 0);
283 static void mlx5e_update_q_counter(struct mlx5e_priv *priv)
285 struct mlx5e_qcounter_stats *qcnt = &priv->stats.qcnt;
287 if (!priv->q_counter)
290 mlx5_core_query_out_of_buffer(priv->mdev, priv->q_counter,
291 &qcnt->rx_out_of_buffer);
294 static void mlx5e_update_pcie_counters(struct mlx5e_priv *priv)
296 struct mlx5e_pcie_stats *pcie_stats = &priv->stats.pcie;
297 struct mlx5_core_dev *mdev = priv->mdev;
298 int sz = MLX5_ST_SZ_BYTES(mpcnt_reg);
302 in = mlx5_vzalloc(sz);
306 out = pcie_stats->pcie_perf_counters;
307 MLX5_SET(mpcnt_reg, in, grp, MLX5_PCIE_PERFORMANCE_COUNTERS_GROUP);
308 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_MPCNT, 0, 0);
310 out = pcie_stats->pcie_tas_counters;
311 MLX5_SET(mpcnt_reg, in, grp, MLX5_PCIE_TIMERS_AND_STATES_COUNTERS_GROUP);
312 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_MPCNT, 0, 0);
317 void mlx5e_update_stats(struct mlx5e_priv *priv)
319 mlx5e_update_q_counter(priv);
320 mlx5e_update_vport_counters(priv);
321 mlx5e_update_pport_counters(priv);
322 mlx5e_update_sw_counters(priv);
323 mlx5e_update_pcie_counters(priv);
326 void mlx5e_update_stats_work(struct work_struct *work)
328 struct delayed_work *dwork = to_delayed_work(work);
329 struct mlx5e_priv *priv = container_of(dwork, struct mlx5e_priv,
331 mutex_lock(&priv->state_lock);
332 if (test_bit(MLX5E_STATE_OPENED, &priv->state)) {
333 priv->profile->update_stats(priv);
334 queue_delayed_work(priv->wq, dwork,
335 msecs_to_jiffies(MLX5E_UPDATE_STATS_INTERVAL));
337 mutex_unlock(&priv->state_lock);
340 static void mlx5e_async_event(struct mlx5_core_dev *mdev, void *vpriv,
341 enum mlx5_dev_event event, unsigned long param)
343 struct mlx5e_priv *priv = vpriv;
345 if (!test_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLED, &priv->state))
349 case MLX5_DEV_EVENT_PORT_UP:
350 case MLX5_DEV_EVENT_PORT_DOWN:
351 queue_work(priv->wq, &priv->update_carrier_work);
359 static void mlx5e_enable_async_events(struct mlx5e_priv *priv)
361 set_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLED, &priv->state);
364 static void mlx5e_disable_async_events(struct mlx5e_priv *priv)
366 clear_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLED, &priv->state);
367 synchronize_irq(mlx5_get_msix_vec(priv->mdev, MLX5_EQ_VEC_ASYNC));
370 #define MLX5E_HW2SW_MTU(hwmtu) (hwmtu - (ETH_HLEN + VLAN_HLEN + ETH_FCS_LEN))
371 #define MLX5E_SW2HW_MTU(swmtu) (swmtu + (ETH_HLEN + VLAN_HLEN + ETH_FCS_LEN))
373 static inline int mlx5e_get_wqe_mtt_sz(void)
375 /* UMR copies MTTs in units of MLX5_UMR_MTT_ALIGNMENT bytes.
376 * To avoid copying garbage after the mtt array, we allocate
379 return ALIGN(MLX5_MPWRQ_PAGES_PER_WQE * sizeof(__be64),
380 MLX5_UMR_MTT_ALIGNMENT);
383 static inline void mlx5e_build_umr_wqe(struct mlx5e_rq *rq, struct mlx5e_sq *sq,
384 struct mlx5e_umr_wqe *wqe, u16 ix)
386 struct mlx5_wqe_ctrl_seg *cseg = &wqe->ctrl;
387 struct mlx5_wqe_umr_ctrl_seg *ucseg = &wqe->uctrl;
388 struct mlx5_wqe_data_seg *dseg = &wqe->data;
389 struct mlx5e_mpw_info *wi = &rq->mpwqe.info[ix];
390 u8 ds_cnt = DIV_ROUND_UP(sizeof(*wqe), MLX5_SEND_WQE_DS);
391 u32 umr_wqe_mtt_offset = mlx5e_get_wqe_mtt_offset(rq, ix);
393 cseg->qpn_ds = cpu_to_be32((sq->sqn << MLX5_WQE_CTRL_QPN_SHIFT) |
395 cseg->fm_ce_se = MLX5_WQE_CTRL_CQ_UPDATE;
396 cseg->imm = rq->mkey_be;
398 ucseg->flags = MLX5_UMR_TRANSLATION_OFFSET_EN;
399 ucseg->klm_octowords =
400 cpu_to_be16(MLX5_MTT_OCTW(MLX5_MPWRQ_PAGES_PER_WQE));
401 ucseg->bsf_octowords =
402 cpu_to_be16(MLX5_MTT_OCTW(umr_wqe_mtt_offset));
403 ucseg->mkey_mask = cpu_to_be64(MLX5_MKEY_MASK_FREE);
405 dseg->lkey = sq->mkey_be;
406 dseg->addr = cpu_to_be64(wi->umr.mtt_addr);
409 static int mlx5e_rq_alloc_mpwqe_info(struct mlx5e_rq *rq,
410 struct mlx5e_channel *c)
412 int wq_sz = mlx5_wq_ll_get_size(&rq->wq);
413 int mtt_sz = mlx5e_get_wqe_mtt_sz();
414 int mtt_alloc = mtt_sz + MLX5_UMR_ALIGN - 1;
417 rq->mpwqe.info = kzalloc_node(wq_sz * sizeof(*rq->mpwqe.info),
418 GFP_KERNEL, cpu_to_node(c->cpu));
422 /* We allocate more than mtt_sz as we will align the pointer */
423 rq->mpwqe.mtt_no_align = kzalloc_node(mtt_alloc * wq_sz, GFP_KERNEL,
424 cpu_to_node(c->cpu));
425 if (unlikely(!rq->mpwqe.mtt_no_align))
426 goto err_free_wqe_info;
428 for (i = 0; i < wq_sz; i++) {
429 struct mlx5e_mpw_info *wi = &rq->mpwqe.info[i];
431 wi->umr.mtt = PTR_ALIGN(rq->mpwqe.mtt_no_align + i * mtt_alloc,
433 wi->umr.mtt_addr = dma_map_single(c->pdev, wi->umr.mtt, mtt_sz,
435 if (unlikely(dma_mapping_error(c->pdev, wi->umr.mtt_addr)))
438 mlx5e_build_umr_wqe(rq, &c->icosq, &wi->umr.wqe, i);
445 struct mlx5e_mpw_info *wi = &rq->mpwqe.info[i];
447 dma_unmap_single(c->pdev, wi->umr.mtt_addr, mtt_sz,
450 kfree(rq->mpwqe.mtt_no_align);
452 kfree(rq->mpwqe.info);
458 static void mlx5e_rq_free_mpwqe_info(struct mlx5e_rq *rq)
460 int wq_sz = mlx5_wq_ll_get_size(&rq->wq);
461 int mtt_sz = mlx5e_get_wqe_mtt_sz();
464 for (i = 0; i < wq_sz; i++) {
465 struct mlx5e_mpw_info *wi = &rq->mpwqe.info[i];
467 dma_unmap_single(rq->pdev, wi->umr.mtt_addr, mtt_sz,
470 kfree(rq->mpwqe.mtt_no_align);
471 kfree(rq->mpwqe.info);
474 static int mlx5e_create_rq(struct mlx5e_channel *c,
475 struct mlx5e_rq_param *param,
478 struct mlx5e_priv *priv = c->priv;
479 struct mlx5_core_dev *mdev = priv->mdev;
480 void *rqc = param->rqc;
481 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
489 param->wq.db_numa_node = cpu_to_node(c->cpu);
491 err = mlx5_wq_ll_create(mdev, ¶m->wq, rqc_wq, &rq->wq,
496 rq->wq.db = &rq->wq.db[MLX5_RCV_DBR];
498 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
500 rq->wq_type = priv->params.rq_wq_type;
502 rq->netdev = c->netdev;
503 rq->tstamp = &priv->tstamp;
508 rq->xdp_prog = priv->xdp_prog ? bpf_prog_inc(priv->xdp_prog) : NULL;
509 if (IS_ERR(rq->xdp_prog)) {
510 err = PTR_ERR(rq->xdp_prog);
512 goto err_rq_wq_destroy;
515 rq->buff.map_dir = DMA_FROM_DEVICE;
517 rq->buff.map_dir = DMA_BIDIRECTIONAL;
519 switch (priv->params.rq_wq_type) {
520 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
521 if (mlx5e_is_vf_vport_rep(priv)) {
523 goto err_rq_wq_destroy;
526 rq->handle_rx_cqe = mlx5e_handle_rx_cqe_mpwrq;
527 rq->alloc_wqe = mlx5e_alloc_rx_mpwqe;
528 rq->dealloc_wqe = mlx5e_dealloc_rx_mpwqe;
530 rq->mpwqe.mtt_offset = c->ix *
531 MLX5E_REQUIRED_MTTS(1, BIT(priv->params.log_rq_size));
533 rq->mpwqe_stride_sz = BIT(priv->params.mpwqe_log_stride_sz);
534 rq->mpwqe_num_strides = BIT(priv->params.mpwqe_log_num_strides);
536 rq->buff.wqe_sz = rq->mpwqe_stride_sz * rq->mpwqe_num_strides;
537 byte_count = rq->buff.wqe_sz;
538 rq->mkey_be = cpu_to_be32(c->priv->umr_mkey.key);
539 err = mlx5e_rq_alloc_mpwqe_info(rq, c);
541 goto err_rq_wq_destroy;
543 default: /* MLX5_WQ_TYPE_LINKED_LIST */
544 rq->dma_info = kzalloc_node(wq_sz * sizeof(*rq->dma_info),
545 GFP_KERNEL, cpu_to_node(c->cpu));
548 goto err_rq_wq_destroy;
551 if (mlx5e_is_vf_vport_rep(priv))
552 rq->handle_rx_cqe = mlx5e_handle_rx_cqe_rep;
554 rq->handle_rx_cqe = mlx5e_handle_rx_cqe;
556 rq->alloc_wqe = mlx5e_alloc_rx_wqe;
557 rq->dealloc_wqe = mlx5e_dealloc_rx_wqe;
559 rq->buff.wqe_sz = (priv->params.lro_en) ?
560 priv->params.lro_wqe_sz :
561 MLX5E_SW2HW_MTU(priv->netdev->mtu);
562 byte_count = rq->buff.wqe_sz;
564 /* calc the required page order */
565 frag_sz = MLX5_RX_HEADROOM +
566 byte_count /* packet data */ +
567 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
568 frag_sz = SKB_DATA_ALIGN(frag_sz);
570 npages = DIV_ROUND_UP(frag_sz, PAGE_SIZE);
571 rq->buff.page_order = order_base_2(npages);
573 byte_count |= MLX5_HW_START_PADDING;
574 rq->mkey_be = c->mkey_be;
577 for (i = 0; i < wq_sz; i++) {
578 struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
580 wqe->data.byte_count = cpu_to_be32(byte_count);
581 wqe->data.lkey = rq->mkey_be;
584 INIT_WORK(&rq->am.work, mlx5e_rx_am_work);
585 rq->am.mode = priv->params.rx_cq_period_mode;
587 rq->page_cache.head = 0;
588 rq->page_cache.tail = 0;
594 bpf_prog_put(rq->xdp_prog);
595 mlx5_wq_destroy(&rq->wq_ctrl);
600 static void mlx5e_destroy_rq(struct mlx5e_rq *rq)
605 bpf_prog_put(rq->xdp_prog);
607 switch (rq->wq_type) {
608 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
609 mlx5e_rq_free_mpwqe_info(rq);
611 default: /* MLX5_WQ_TYPE_LINKED_LIST */
615 for (i = rq->page_cache.head; i != rq->page_cache.tail;
616 i = (i + 1) & (MLX5E_CACHE_SIZE - 1)) {
617 struct mlx5e_dma_info *dma_info = &rq->page_cache.page_cache[i];
619 mlx5e_page_release(rq, dma_info, false);
621 mlx5_wq_destroy(&rq->wq_ctrl);
624 static int mlx5e_enable_rq(struct mlx5e_rq *rq, struct mlx5e_rq_param *param)
626 struct mlx5e_priv *priv = rq->priv;
627 struct mlx5_core_dev *mdev = priv->mdev;
635 inlen = MLX5_ST_SZ_BYTES(create_rq_in) +
636 sizeof(u64) * rq->wq_ctrl.buf.npages;
637 in = mlx5_vzalloc(inlen);
641 rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
642 wq = MLX5_ADDR_OF(rqc, rqc, wq);
644 memcpy(rqc, param->rqc, sizeof(param->rqc));
646 MLX5_SET(rqc, rqc, cqn, rq->cq.mcq.cqn);
647 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST);
648 MLX5_SET(rqc, rqc, vsd, priv->params.vlan_strip_disable);
649 MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift -
650 MLX5_ADAPTER_PAGE_SHIFT);
651 MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma);
653 mlx5_fill_page_array(&rq->wq_ctrl.buf,
654 (__be64 *)MLX5_ADDR_OF(wq, wq, pas));
656 err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn);
663 static int mlx5e_modify_rq_state(struct mlx5e_rq *rq, int curr_state,
666 struct mlx5e_channel *c = rq->channel;
667 struct mlx5e_priv *priv = c->priv;
668 struct mlx5_core_dev *mdev = priv->mdev;
675 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
676 in = mlx5_vzalloc(inlen);
680 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
682 MLX5_SET(modify_rq_in, in, rq_state, curr_state);
683 MLX5_SET(rqc, rqc, state, next_state);
685 err = mlx5_core_modify_rq(mdev, rq->rqn, in, inlen);
692 static int mlx5e_modify_rq_vsd(struct mlx5e_rq *rq, bool vsd)
694 struct mlx5e_channel *c = rq->channel;
695 struct mlx5e_priv *priv = c->priv;
696 struct mlx5_core_dev *mdev = priv->mdev;
703 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
704 in = mlx5_vzalloc(inlen);
708 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
710 MLX5_SET(modify_rq_in, in, rq_state, MLX5_RQC_STATE_RDY);
711 MLX5_SET64(modify_rq_in, in, modify_bitmask,
712 MLX5_MODIFY_RQ_IN_MODIFY_BITMASK_VSD);
713 MLX5_SET(rqc, rqc, vsd, vsd);
714 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RDY);
716 err = mlx5_core_modify_rq(mdev, rq->rqn, in, inlen);
723 static void mlx5e_disable_rq(struct mlx5e_rq *rq)
725 mlx5_core_destroy_rq(rq->priv->mdev, rq->rqn);
728 static int mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq)
730 unsigned long exp_time = jiffies + msecs_to_jiffies(20000);
731 struct mlx5e_channel *c = rq->channel;
732 struct mlx5e_priv *priv = c->priv;
733 struct mlx5_wq_ll *wq = &rq->wq;
735 while (time_before(jiffies, exp_time)) {
736 if (wq->cur_sz >= priv->params.min_rx_wqes)
745 static void mlx5e_free_rx_descs(struct mlx5e_rq *rq)
747 struct mlx5_wq_ll *wq = &rq->wq;
748 struct mlx5e_rx_wqe *wqe;
752 /* UMR WQE (if in progress) is always at wq->head */
753 if (test_bit(MLX5E_RQ_STATE_UMR_WQE_IN_PROGRESS, &rq->state))
754 mlx5e_free_rx_mpwqe(rq, &rq->mpwqe.info[wq->head]);
756 while (!mlx5_wq_ll_is_empty(wq)) {
757 wqe_ix_be = *wq->tail_next;
758 wqe_ix = be16_to_cpu(wqe_ix_be);
759 wqe = mlx5_wq_ll_get_wqe(&rq->wq, wqe_ix);
760 rq->dealloc_wqe(rq, wqe_ix);
761 mlx5_wq_ll_pop(&rq->wq, wqe_ix_be,
762 &wqe->next.next_wqe_index);
766 static int mlx5e_open_rq(struct mlx5e_channel *c,
767 struct mlx5e_rq_param *param,
770 struct mlx5e_sq *sq = &c->icosq;
771 u16 pi = sq->pc & sq->wq.sz_m1;
774 err = mlx5e_create_rq(c, param, rq);
778 err = mlx5e_enable_rq(rq, param);
782 err = mlx5e_modify_rq_state(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
786 if (param->am_enabled)
787 set_bit(MLX5E_RQ_STATE_AM, &c->rq.state);
789 sq->db.ico_wqe[pi].opcode = MLX5_OPCODE_NOP;
790 sq->db.ico_wqe[pi].num_wqebbs = 1;
791 mlx5e_send_nop(sq, true); /* trigger mlx5e_post_rx_wqes() */
796 mlx5e_disable_rq(rq);
798 mlx5e_destroy_rq(rq);
803 static void mlx5e_close_rq(struct mlx5e_rq *rq)
805 set_bit(MLX5E_RQ_STATE_FLUSH, &rq->state);
806 napi_synchronize(&rq->channel->napi); /* prevent mlx5e_post_rx_wqes */
807 cancel_work_sync(&rq->am.work);
809 mlx5e_disable_rq(rq);
810 mlx5e_free_rx_descs(rq);
811 mlx5e_destroy_rq(rq);
814 static void mlx5e_free_sq_xdp_db(struct mlx5e_sq *sq)
816 kfree(sq->db.xdp.di);
817 kfree(sq->db.xdp.wqe_info);
820 static int mlx5e_alloc_sq_xdp_db(struct mlx5e_sq *sq, int numa)
822 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
824 sq->db.xdp.di = kzalloc_node(sizeof(*sq->db.xdp.di) * wq_sz,
826 sq->db.xdp.wqe_info = kzalloc_node(sizeof(*sq->db.xdp.wqe_info) * wq_sz,
828 if (!sq->db.xdp.di || !sq->db.xdp.wqe_info) {
829 mlx5e_free_sq_xdp_db(sq);
836 static void mlx5e_free_sq_ico_db(struct mlx5e_sq *sq)
838 kfree(sq->db.ico_wqe);
841 static int mlx5e_alloc_sq_ico_db(struct mlx5e_sq *sq, int numa)
843 u8 wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
845 sq->db.ico_wqe = kzalloc_node(sizeof(*sq->db.ico_wqe) * wq_sz,
853 static void mlx5e_free_sq_txq_db(struct mlx5e_sq *sq)
855 kfree(sq->db.txq.wqe_info);
856 kfree(sq->db.txq.dma_fifo);
857 kfree(sq->db.txq.skb);
860 static int mlx5e_alloc_sq_txq_db(struct mlx5e_sq *sq, int numa)
862 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
863 int df_sz = wq_sz * MLX5_SEND_WQEBB_NUM_DS;
865 sq->db.txq.skb = kzalloc_node(wq_sz * sizeof(*sq->db.txq.skb),
867 sq->db.txq.dma_fifo = kzalloc_node(df_sz * sizeof(*sq->db.txq.dma_fifo),
869 sq->db.txq.wqe_info = kzalloc_node(wq_sz * sizeof(*sq->db.txq.wqe_info),
871 if (!sq->db.txq.skb || !sq->db.txq.dma_fifo || !sq->db.txq.wqe_info) {
872 mlx5e_free_sq_txq_db(sq);
876 sq->dma_fifo_mask = df_sz - 1;
881 static void mlx5e_free_sq_db(struct mlx5e_sq *sq)
885 mlx5e_free_sq_txq_db(sq);
888 mlx5e_free_sq_ico_db(sq);
891 mlx5e_free_sq_xdp_db(sq);
896 static int mlx5e_alloc_sq_db(struct mlx5e_sq *sq, int numa)
900 return mlx5e_alloc_sq_txq_db(sq, numa);
902 return mlx5e_alloc_sq_ico_db(sq, numa);
904 return mlx5e_alloc_sq_xdp_db(sq, numa);
910 static int mlx5e_sq_get_max_wqebbs(u8 sq_type)
914 return MLX5E_ICOSQ_MAX_WQEBBS;
916 return MLX5E_XDP_TX_WQEBBS;
918 return MLX5_SEND_WQE_MAX_WQEBBS;
921 static int mlx5e_create_sq(struct mlx5e_channel *c,
923 struct mlx5e_sq_param *param,
926 struct mlx5e_priv *priv = c->priv;
927 struct mlx5_core_dev *mdev = priv->mdev;
929 void *sqc = param->sqc;
930 void *sqc_wq = MLX5_ADDR_OF(sqc, sqc, wq);
933 sq->type = param->type;
935 sq->tstamp = &priv->tstamp;
936 sq->mkey_be = c->mkey_be;
940 err = mlx5_alloc_map_uar(mdev, &sq->uar, !!MLX5_CAP_GEN(mdev, bf));
944 param->wq.db_numa_node = cpu_to_node(c->cpu);
946 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, &sq->wq,
949 goto err_unmap_free_uar;
951 sq->wq.db = &sq->wq.db[MLX5_SND_DBR];
952 if (sq->uar.bf_map) {
953 set_bit(MLX5E_SQ_STATE_BF_ENABLE, &sq->state);
954 sq->uar_map = sq->uar.bf_map;
956 sq->uar_map = sq->uar.map;
958 sq->bf_buf_size = (1 << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2;
959 sq->max_inline = param->max_inline;
960 sq->min_inline_mode =
961 MLX5_CAP_ETH(mdev, wqe_inline_mode) == MLX5_CAP_INLINE_MODE_VPORT_CONTEXT ?
962 param->min_inline_mode : 0;
964 err = mlx5e_alloc_sq_db(sq, cpu_to_node(c->cpu));
966 goto err_sq_wq_destroy;
968 if (sq->type == MLX5E_SQ_TXQ) {
971 txq_ix = c->ix + tc * priv->params.num_channels;
972 sq->txq = netdev_get_tx_queue(priv->netdev, txq_ix);
973 priv->txq_to_sq_map[txq_ix] = sq;
976 sq->edge = (sq->wq.sz_m1 + 1) - mlx5e_sq_get_max_wqebbs(sq->type);
977 sq->bf_budget = MLX5E_SQ_BF_BUDGET;
982 mlx5_wq_destroy(&sq->wq_ctrl);
985 mlx5_unmap_free_uar(mdev, &sq->uar);
990 static void mlx5e_destroy_sq(struct mlx5e_sq *sq)
992 struct mlx5e_channel *c = sq->channel;
993 struct mlx5e_priv *priv = c->priv;
995 mlx5e_free_sq_db(sq);
996 mlx5_wq_destroy(&sq->wq_ctrl);
997 mlx5_unmap_free_uar(priv->mdev, &sq->uar);
1000 static int mlx5e_enable_sq(struct mlx5e_sq *sq, struct mlx5e_sq_param *param)
1002 struct mlx5e_channel *c = sq->channel;
1003 struct mlx5e_priv *priv = c->priv;
1004 struct mlx5_core_dev *mdev = priv->mdev;
1012 inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
1013 sizeof(u64) * sq->wq_ctrl.buf.npages;
1014 in = mlx5_vzalloc(inlen);
1018 sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
1019 wq = MLX5_ADDR_OF(sqc, sqc, wq);
1021 memcpy(sqc, param->sqc, sizeof(param->sqc));
1023 MLX5_SET(sqc, sqc, tis_num_0, param->type == MLX5E_SQ_ICO ?
1024 0 : priv->tisn[sq->tc]);
1025 MLX5_SET(sqc, sqc, cqn, sq->cq.mcq.cqn);
1026 MLX5_SET(sqc, sqc, min_wqe_inline_mode, sq->min_inline_mode);
1027 MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
1028 MLX5_SET(sqc, sqc, tis_lst_sz, param->type == MLX5E_SQ_ICO ? 0 : 1);
1029 MLX5_SET(sqc, sqc, flush_in_error_en, 1);
1031 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
1032 MLX5_SET(wq, wq, uar_page, sq->uar.index);
1033 MLX5_SET(wq, wq, log_wq_pg_sz, sq->wq_ctrl.buf.page_shift -
1034 MLX5_ADAPTER_PAGE_SHIFT);
1035 MLX5_SET64(wq, wq, dbr_addr, sq->wq_ctrl.db.dma);
1037 mlx5_fill_page_array(&sq->wq_ctrl.buf,
1038 (__be64 *)MLX5_ADDR_OF(wq, wq, pas));
1040 err = mlx5_core_create_sq(mdev, in, inlen, &sq->sqn);
1047 static int mlx5e_modify_sq(struct mlx5e_sq *sq, int curr_state,
1048 int next_state, bool update_rl, int rl_index)
1050 struct mlx5e_channel *c = sq->channel;
1051 struct mlx5e_priv *priv = c->priv;
1052 struct mlx5_core_dev *mdev = priv->mdev;
1059 inlen = MLX5_ST_SZ_BYTES(modify_sq_in);
1060 in = mlx5_vzalloc(inlen);
1064 sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx);
1066 MLX5_SET(modify_sq_in, in, sq_state, curr_state);
1067 MLX5_SET(sqc, sqc, state, next_state);
1068 if (update_rl && next_state == MLX5_SQC_STATE_RDY) {
1069 MLX5_SET64(modify_sq_in, in, modify_bitmask, 1);
1070 MLX5_SET(sqc, sqc, packet_pacing_rate_limit_index, rl_index);
1073 err = mlx5_core_modify_sq(mdev, sq->sqn, in, inlen);
1080 static void mlx5e_disable_sq(struct mlx5e_sq *sq)
1082 struct mlx5e_channel *c = sq->channel;
1083 struct mlx5e_priv *priv = c->priv;
1084 struct mlx5_core_dev *mdev = priv->mdev;
1086 mlx5_core_destroy_sq(mdev, sq->sqn);
1088 mlx5_rl_remove_rate(mdev, sq->rate_limit);
1091 static int mlx5e_open_sq(struct mlx5e_channel *c,
1093 struct mlx5e_sq_param *param,
1094 struct mlx5e_sq *sq)
1098 err = mlx5e_create_sq(c, tc, param, sq);
1102 err = mlx5e_enable_sq(sq, param);
1104 goto err_destroy_sq;
1106 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY,
1109 goto err_disable_sq;
1112 netdev_tx_reset_queue(sq->txq);
1113 netif_tx_start_queue(sq->txq);
1119 mlx5e_disable_sq(sq);
1121 mlx5e_destroy_sq(sq);
1126 static inline void netif_tx_disable_queue(struct netdev_queue *txq)
1128 __netif_tx_lock_bh(txq);
1129 netif_tx_stop_queue(txq);
1130 __netif_tx_unlock_bh(txq);
1133 static void mlx5e_close_sq(struct mlx5e_sq *sq)
1135 set_bit(MLX5E_SQ_STATE_FLUSH, &sq->state);
1136 /* prevent netif_tx_wake_queue */
1137 napi_synchronize(&sq->channel->napi);
1140 netif_tx_disable_queue(sq->txq);
1142 /* last doorbell out, godspeed .. */
1143 if (mlx5e_sq_has_room_for(sq, 1)) {
1144 sq->db.txq.skb[(sq->pc & sq->wq.sz_m1)] = NULL;
1145 mlx5e_send_nop(sq, true);
1149 mlx5e_disable_sq(sq);
1150 mlx5e_free_sq_descs(sq);
1151 mlx5e_destroy_sq(sq);
1154 static int mlx5e_create_cq(struct mlx5e_channel *c,
1155 struct mlx5e_cq_param *param,
1156 struct mlx5e_cq *cq)
1158 struct mlx5e_priv *priv = c->priv;
1159 struct mlx5_core_dev *mdev = priv->mdev;
1160 struct mlx5_core_cq *mcq = &cq->mcq;
1166 param->wq.buf_numa_node = cpu_to_node(c->cpu);
1167 param->wq.db_numa_node = cpu_to_node(c->cpu);
1168 param->eq_ix = c->ix;
1170 err = mlx5_cqwq_create(mdev, ¶m->wq, param->cqc, &cq->wq,
1175 mlx5_vector2eqn(mdev, param->eq_ix, &eqn_not_used, &irqn);
1177 cq->napi = &c->napi;
1180 mcq->set_ci_db = cq->wq_ctrl.db.db;
1181 mcq->arm_db = cq->wq_ctrl.db.db + 1;
1182 *mcq->set_ci_db = 0;
1184 mcq->vector = param->eq_ix;
1185 mcq->comp = mlx5e_completion_event;
1186 mcq->event = mlx5e_cq_error_event;
1188 mcq->uar = &mdev->mlx5e_res.cq_uar;
1190 for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) {
1191 struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i);
1202 static void mlx5e_destroy_cq(struct mlx5e_cq *cq)
1204 mlx5_cqwq_destroy(&cq->wq_ctrl);
1207 static int mlx5e_enable_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param)
1209 struct mlx5e_priv *priv = cq->priv;
1210 struct mlx5_core_dev *mdev = priv->mdev;
1211 struct mlx5_core_cq *mcq = &cq->mcq;
1216 unsigned int irqn_not_used;
1220 inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
1221 sizeof(u64) * cq->wq_ctrl.frag_buf.npages;
1222 in = mlx5_vzalloc(inlen);
1226 cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
1228 memcpy(cqc, param->cqc, sizeof(param->cqc));
1230 mlx5_fill_page_frag_array(&cq->wq_ctrl.frag_buf,
1231 (__be64 *)MLX5_ADDR_OF(create_cq_in, in, pas));
1233 mlx5_vector2eqn(mdev, param->eq_ix, &eqn, &irqn_not_used);
1235 MLX5_SET(cqc, cqc, cq_period_mode, param->cq_period_mode);
1236 MLX5_SET(cqc, cqc, c_eqn, eqn);
1237 MLX5_SET(cqc, cqc, uar_page, mcq->uar->index);
1238 MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.frag_buf.page_shift -
1239 MLX5_ADAPTER_PAGE_SHIFT);
1240 MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
1242 err = mlx5_core_create_cq(mdev, mcq, in, inlen);
1254 static void mlx5e_disable_cq(struct mlx5e_cq *cq)
1256 struct mlx5e_priv *priv = cq->priv;
1257 struct mlx5_core_dev *mdev = priv->mdev;
1259 mlx5_core_destroy_cq(mdev, &cq->mcq);
1262 static int mlx5e_open_cq(struct mlx5e_channel *c,
1263 struct mlx5e_cq_param *param,
1264 struct mlx5e_cq *cq,
1265 struct mlx5e_cq_moder moderation)
1268 struct mlx5e_priv *priv = c->priv;
1269 struct mlx5_core_dev *mdev = priv->mdev;
1271 err = mlx5e_create_cq(c, param, cq);
1275 err = mlx5e_enable_cq(cq, param);
1277 goto err_destroy_cq;
1279 if (MLX5_CAP_GEN(mdev, cq_moderation))
1280 mlx5_core_modify_cq_moderation(mdev, &cq->mcq,
1286 mlx5e_destroy_cq(cq);
1291 static void mlx5e_close_cq(struct mlx5e_cq *cq)
1293 mlx5e_disable_cq(cq);
1294 mlx5e_destroy_cq(cq);
1297 static int mlx5e_get_cpu(struct mlx5e_priv *priv, int ix)
1299 return cpumask_first(priv->mdev->priv.irq_info[ix].mask);
1302 static int mlx5e_open_tx_cqs(struct mlx5e_channel *c,
1303 struct mlx5e_channel_param *cparam)
1305 struct mlx5e_priv *priv = c->priv;
1309 for (tc = 0; tc < c->num_tc; tc++) {
1310 err = mlx5e_open_cq(c, &cparam->tx_cq, &c->sq[tc].cq,
1311 priv->params.tx_cq_moderation);
1313 goto err_close_tx_cqs;
1319 for (tc--; tc >= 0; tc--)
1320 mlx5e_close_cq(&c->sq[tc].cq);
1325 static void mlx5e_close_tx_cqs(struct mlx5e_channel *c)
1329 for (tc = 0; tc < c->num_tc; tc++)
1330 mlx5e_close_cq(&c->sq[tc].cq);
1333 static int mlx5e_open_sqs(struct mlx5e_channel *c,
1334 struct mlx5e_channel_param *cparam)
1339 for (tc = 0; tc < c->num_tc; tc++) {
1340 err = mlx5e_open_sq(c, tc, &cparam->sq, &c->sq[tc]);
1348 for (tc--; tc >= 0; tc--)
1349 mlx5e_close_sq(&c->sq[tc]);
1354 static void mlx5e_close_sqs(struct mlx5e_channel *c)
1358 for (tc = 0; tc < c->num_tc; tc++)
1359 mlx5e_close_sq(&c->sq[tc]);
1362 static void mlx5e_build_channeltc_to_txq_map(struct mlx5e_priv *priv, int ix)
1366 for (i = 0; i < priv->profile->max_tc; i++)
1367 priv->channeltc_to_txq_map[ix][i] =
1368 ix + i * priv->params.num_channels;
1371 static int mlx5e_set_sq_maxrate(struct net_device *dev,
1372 struct mlx5e_sq *sq, u32 rate)
1374 struct mlx5e_priv *priv = netdev_priv(dev);
1375 struct mlx5_core_dev *mdev = priv->mdev;
1379 if (rate == sq->rate_limit)
1384 /* remove current rl index to free space to next ones */
1385 mlx5_rl_remove_rate(mdev, sq->rate_limit);
1390 err = mlx5_rl_add_rate(mdev, rate, &rl_index);
1392 netdev_err(dev, "Failed configuring rate %u: %d\n",
1398 err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RDY,
1399 MLX5_SQC_STATE_RDY, true, rl_index);
1401 netdev_err(dev, "Failed configuring rate %u: %d\n",
1403 /* remove the rate from the table */
1405 mlx5_rl_remove_rate(mdev, rate);
1409 sq->rate_limit = rate;
1413 static int mlx5e_set_tx_maxrate(struct net_device *dev, int index, u32 rate)
1415 struct mlx5e_priv *priv = netdev_priv(dev);
1416 struct mlx5_core_dev *mdev = priv->mdev;
1417 struct mlx5e_sq *sq = priv->txq_to_sq_map[index];
1420 if (!mlx5_rl_is_supported(mdev)) {
1421 netdev_err(dev, "Rate limiting is not supported on this device\n");
1425 /* rate is given in Mb/sec, HW config is in Kb/sec */
1428 /* Check whether rate in valid range, 0 is always valid */
1429 if (rate && !mlx5_rl_is_in_range(mdev, rate)) {
1430 netdev_err(dev, "TX rate %u, is not in range\n", rate);
1434 mutex_lock(&priv->state_lock);
1435 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
1436 err = mlx5e_set_sq_maxrate(dev, sq, rate);
1438 priv->tx_rates[index] = rate;
1439 mutex_unlock(&priv->state_lock);
1444 static int mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
1445 struct mlx5e_channel_param *cparam,
1446 struct mlx5e_channel **cp)
1448 struct mlx5e_cq_moder icosq_cq_moder = {0, 0};
1449 struct net_device *netdev = priv->netdev;
1450 struct mlx5e_cq_moder rx_cq_profile;
1451 int cpu = mlx5e_get_cpu(priv, ix);
1452 struct mlx5e_channel *c;
1453 struct mlx5e_sq *sq;
1457 c = kzalloc_node(sizeof(*c), GFP_KERNEL, cpu_to_node(cpu));
1464 c->pdev = &priv->mdev->pdev->dev;
1465 c->netdev = priv->netdev;
1466 c->mkey_be = cpu_to_be32(priv->mdev->mlx5e_res.mkey.key);
1467 c->num_tc = priv->params.num_tc;
1468 c->xdp = !!priv->xdp_prog;
1470 if (priv->params.rx_am_enabled)
1471 rx_cq_profile = mlx5e_am_get_def_profile(priv->params.rx_cq_period_mode);
1473 rx_cq_profile = priv->params.rx_cq_moderation;
1475 mlx5e_build_channeltc_to_txq_map(priv, ix);
1477 netif_napi_add(netdev, &c->napi, mlx5e_napi_poll, 64);
1479 err = mlx5e_open_cq(c, &cparam->icosq_cq, &c->icosq.cq, icosq_cq_moder);
1483 err = mlx5e_open_tx_cqs(c, cparam);
1485 goto err_close_icosq_cq;
1487 err = mlx5e_open_cq(c, &cparam->rx_cq, &c->rq.cq,
1490 goto err_close_tx_cqs;
1492 /* XDP SQ CQ params are same as normal TXQ sq CQ params */
1493 err = c->xdp ? mlx5e_open_cq(c, &cparam->tx_cq, &c->xdp_sq.cq,
1494 priv->params.tx_cq_moderation) : 0;
1496 goto err_close_rx_cq;
1498 napi_enable(&c->napi);
1500 err = mlx5e_open_sq(c, 0, &cparam->icosq, &c->icosq);
1502 goto err_disable_napi;
1504 err = mlx5e_open_sqs(c, cparam);
1506 goto err_close_icosq;
1508 for (i = 0; i < priv->params.num_tc; i++) {
1509 u32 txq_ix = priv->channeltc_to_txq_map[ix][i];
1511 if (priv->tx_rates[txq_ix]) {
1512 sq = priv->txq_to_sq_map[txq_ix];
1513 mlx5e_set_sq_maxrate(priv->netdev, sq,
1514 priv->tx_rates[txq_ix]);
1518 err = c->xdp ? mlx5e_open_sq(c, 0, &cparam->xdp_sq, &c->xdp_sq) : 0;
1522 err = mlx5e_open_rq(c, &cparam->rq, &c->rq);
1524 goto err_close_xdp_sq;
1526 netif_set_xps_queue(netdev, get_cpu_mask(c->cpu), ix);
1532 mlx5e_close_sq(&c->xdp_sq);
1538 mlx5e_close_sq(&c->icosq);
1541 napi_disable(&c->napi);
1543 mlx5e_close_cq(&c->xdp_sq.cq);
1546 mlx5e_close_cq(&c->rq.cq);
1549 mlx5e_close_tx_cqs(c);
1552 mlx5e_close_cq(&c->icosq.cq);
1555 netif_napi_del(&c->napi);
1561 static void mlx5e_close_channel(struct mlx5e_channel *c)
1563 mlx5e_close_rq(&c->rq);
1565 mlx5e_close_sq(&c->xdp_sq);
1567 mlx5e_close_sq(&c->icosq);
1568 napi_disable(&c->napi);
1570 mlx5e_close_cq(&c->xdp_sq.cq);
1571 mlx5e_close_cq(&c->rq.cq);
1572 mlx5e_close_tx_cqs(c);
1573 mlx5e_close_cq(&c->icosq.cq);
1574 netif_napi_del(&c->napi);
1579 static void mlx5e_build_rq_param(struct mlx5e_priv *priv,
1580 struct mlx5e_rq_param *param)
1582 void *rqc = param->rqc;
1583 void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
1585 switch (priv->params.rq_wq_type) {
1586 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
1587 MLX5_SET(wq, wq, log_wqe_num_of_strides,
1588 priv->params.mpwqe_log_num_strides - 9);
1589 MLX5_SET(wq, wq, log_wqe_stride_size,
1590 priv->params.mpwqe_log_stride_sz - 6);
1591 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ);
1593 default: /* MLX5_WQ_TYPE_LINKED_LIST */
1594 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
1597 MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN);
1598 MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe)));
1599 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_rq_size);
1600 MLX5_SET(wq, wq, pd, priv->mdev->mlx5e_res.pdn);
1601 MLX5_SET(rqc, rqc, counter_set_id, priv->q_counter);
1603 param->wq.buf_numa_node = dev_to_node(&priv->mdev->pdev->dev);
1604 param->wq.linear = 1;
1606 param->am_enabled = priv->params.rx_am_enabled;
1609 static void mlx5e_build_drop_rq_param(struct mlx5e_rq_param *param)
1611 void *rqc = param->rqc;
1612 void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
1614 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
1615 MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe)));
1618 static void mlx5e_build_sq_param_common(struct mlx5e_priv *priv,
1619 struct mlx5e_sq_param *param)
1621 void *sqc = param->sqc;
1622 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
1624 MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
1625 MLX5_SET(wq, wq, pd, priv->mdev->mlx5e_res.pdn);
1627 param->wq.buf_numa_node = dev_to_node(&priv->mdev->pdev->dev);
1630 static void mlx5e_build_sq_param(struct mlx5e_priv *priv,
1631 struct mlx5e_sq_param *param)
1633 void *sqc = param->sqc;
1634 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
1636 mlx5e_build_sq_param_common(priv, param);
1637 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_sq_size);
1639 param->max_inline = priv->params.tx_max_inline;
1640 param->min_inline_mode = priv->params.tx_min_inline_mode;
1641 param->type = MLX5E_SQ_TXQ;
1644 static void mlx5e_build_common_cq_param(struct mlx5e_priv *priv,
1645 struct mlx5e_cq_param *param)
1647 void *cqc = param->cqc;
1649 MLX5_SET(cqc, cqc, uar_page, priv->mdev->mlx5e_res.cq_uar.index);
1652 static void mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
1653 struct mlx5e_cq_param *param)
1655 void *cqc = param->cqc;
1658 switch (priv->params.rq_wq_type) {
1659 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
1660 log_cq_size = priv->params.log_rq_size +
1661 priv->params.mpwqe_log_num_strides;
1663 default: /* MLX5_WQ_TYPE_LINKED_LIST */
1664 log_cq_size = priv->params.log_rq_size;
1667 MLX5_SET(cqc, cqc, log_cq_size, log_cq_size);
1668 if (MLX5E_GET_PFLAG(priv, MLX5E_PFLAG_RX_CQE_COMPRESS)) {
1669 MLX5_SET(cqc, cqc, mini_cqe_res_format, MLX5_CQE_FORMAT_CSUM);
1670 MLX5_SET(cqc, cqc, cqe_comp_en, 1);
1673 mlx5e_build_common_cq_param(priv, param);
1675 param->cq_period_mode = priv->params.rx_cq_period_mode;
1678 static void mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
1679 struct mlx5e_cq_param *param)
1681 void *cqc = param->cqc;
1683 MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_sq_size);
1685 mlx5e_build_common_cq_param(priv, param);
1687 param->cq_period_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
1690 static void mlx5e_build_ico_cq_param(struct mlx5e_priv *priv,
1691 struct mlx5e_cq_param *param,
1694 void *cqc = param->cqc;
1696 MLX5_SET(cqc, cqc, log_cq_size, log_wq_size);
1698 mlx5e_build_common_cq_param(priv, param);
1700 param->cq_period_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
1703 static void mlx5e_build_icosq_param(struct mlx5e_priv *priv,
1704 struct mlx5e_sq_param *param,
1707 void *sqc = param->sqc;
1708 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
1710 mlx5e_build_sq_param_common(priv, param);
1712 MLX5_SET(wq, wq, log_wq_sz, log_wq_size);
1713 MLX5_SET(sqc, sqc, reg_umr, MLX5_CAP_ETH(priv->mdev, reg_umr_sq));
1715 param->type = MLX5E_SQ_ICO;
1718 static void mlx5e_build_xdpsq_param(struct mlx5e_priv *priv,
1719 struct mlx5e_sq_param *param)
1721 void *sqc = param->sqc;
1722 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
1724 mlx5e_build_sq_param_common(priv, param);
1725 MLX5_SET(wq, wq, log_wq_sz, priv->params.log_sq_size);
1727 param->max_inline = priv->params.tx_max_inline;
1728 /* FOR XDP SQs will support only L2 inline mode */
1729 param->min_inline_mode = MLX5_INLINE_MODE_NONE;
1730 param->type = MLX5E_SQ_XDP;
1733 static void mlx5e_build_channel_param(struct mlx5e_priv *priv, struct mlx5e_channel_param *cparam)
1735 u8 icosq_log_wq_sz = MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE;
1737 mlx5e_build_rq_param(priv, &cparam->rq);
1738 mlx5e_build_sq_param(priv, &cparam->sq);
1739 mlx5e_build_xdpsq_param(priv, &cparam->xdp_sq);
1740 mlx5e_build_icosq_param(priv, &cparam->icosq, icosq_log_wq_sz);
1741 mlx5e_build_rx_cq_param(priv, &cparam->rx_cq);
1742 mlx5e_build_tx_cq_param(priv, &cparam->tx_cq);
1743 mlx5e_build_ico_cq_param(priv, &cparam->icosq_cq, icosq_log_wq_sz);
1746 static int mlx5e_open_channels(struct mlx5e_priv *priv)
1748 struct mlx5e_channel_param *cparam;
1749 int nch = priv->params.num_channels;
1754 priv->channel = kcalloc(nch, sizeof(struct mlx5e_channel *),
1757 priv->txq_to_sq_map = kcalloc(nch * priv->params.num_tc,
1758 sizeof(struct mlx5e_sq *), GFP_KERNEL);
1760 cparam = kzalloc(sizeof(struct mlx5e_channel_param), GFP_KERNEL);
1762 if (!priv->channel || !priv->txq_to_sq_map || !cparam)
1763 goto err_free_txq_to_sq_map;
1765 mlx5e_build_channel_param(priv, cparam);
1767 for (i = 0; i < nch; i++) {
1768 err = mlx5e_open_channel(priv, i, cparam, &priv->channel[i]);
1770 goto err_close_channels;
1773 for (j = 0; j < nch; j++) {
1774 err = mlx5e_wait_for_min_rx_wqes(&priv->channel[j]->rq);
1776 goto err_close_channels;
1779 /* FIXME: This is a W/A for tx timeout watch dog false alarm when
1780 * polling for inactive tx queues.
1782 netif_tx_start_all_queues(priv->netdev);
1788 for (i--; i >= 0; i--)
1789 mlx5e_close_channel(priv->channel[i]);
1791 err_free_txq_to_sq_map:
1792 kfree(priv->txq_to_sq_map);
1793 kfree(priv->channel);
1799 static void mlx5e_close_channels(struct mlx5e_priv *priv)
1803 /* FIXME: This is a W/A only for tx timeout watch dog false alarm when
1804 * polling for inactive tx queues.
1806 netif_tx_stop_all_queues(priv->netdev);
1807 netif_tx_disable(priv->netdev);
1809 for (i = 0; i < priv->params.num_channels; i++)
1810 mlx5e_close_channel(priv->channel[i]);
1812 kfree(priv->txq_to_sq_map);
1813 kfree(priv->channel);
1816 static int mlx5e_rx_hash_fn(int hfunc)
1818 return (hfunc == ETH_RSS_HASH_TOP) ?
1819 MLX5_RX_HASH_FN_TOEPLITZ :
1820 MLX5_RX_HASH_FN_INVERTED_XOR8;
1823 static int mlx5e_bits_invert(unsigned long a, int size)
1828 for (i = 0; i < size; i++)
1829 inv |= (test_bit(size - i - 1, &a) ? 1 : 0) << i;
1834 static void mlx5e_fill_indir_rqt_rqns(struct mlx5e_priv *priv, void *rqtc)
1838 for (i = 0; i < MLX5E_INDIR_RQT_SIZE; i++) {
1842 if (priv->params.rss_hfunc == ETH_RSS_HASH_XOR)
1843 ix = mlx5e_bits_invert(i, MLX5E_LOG_INDIR_RQT_SIZE);
1845 ix = priv->params.indirection_rqt[ix];
1846 rqn = test_bit(MLX5E_STATE_OPENED, &priv->state) ?
1847 priv->channel[ix]->rq.rqn :
1849 MLX5_SET(rqtc, rqtc, rq_num[i], rqn);
1853 static void mlx5e_fill_direct_rqt_rqn(struct mlx5e_priv *priv, void *rqtc,
1856 u32 rqn = test_bit(MLX5E_STATE_OPENED, &priv->state) ?
1857 priv->channel[ix]->rq.rqn :
1860 MLX5_SET(rqtc, rqtc, rq_num[0], rqn);
1863 static int mlx5e_create_rqt(struct mlx5e_priv *priv, int sz,
1864 int ix, struct mlx5e_rqt *rqt)
1866 struct mlx5_core_dev *mdev = priv->mdev;
1872 inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
1873 in = mlx5_vzalloc(inlen);
1877 rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
1879 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
1880 MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
1882 if (sz > 1) /* RSS */
1883 mlx5e_fill_indir_rqt_rqns(priv, rqtc);
1885 mlx5e_fill_direct_rqt_rqn(priv, rqtc, ix);
1887 err = mlx5_core_create_rqt(mdev, in, inlen, &rqt->rqtn);
1889 rqt->enabled = true;
1895 void mlx5e_destroy_rqt(struct mlx5e_priv *priv, struct mlx5e_rqt *rqt)
1897 rqt->enabled = false;
1898 mlx5_core_destroy_rqt(priv->mdev, rqt->rqtn);
1901 static int mlx5e_create_indirect_rqts(struct mlx5e_priv *priv)
1903 struct mlx5e_rqt *rqt = &priv->indir_rqt;
1905 return mlx5e_create_rqt(priv, MLX5E_INDIR_RQT_SIZE, 0, rqt);
1908 int mlx5e_create_direct_rqts(struct mlx5e_priv *priv)
1910 struct mlx5e_rqt *rqt;
1914 for (ix = 0; ix < priv->profile->max_nch(priv->mdev); ix++) {
1915 rqt = &priv->direct_tir[ix].rqt;
1916 err = mlx5e_create_rqt(priv, 1 /*size */, ix, rqt);
1918 goto err_destroy_rqts;
1924 for (ix--; ix >= 0; ix--)
1925 mlx5e_destroy_rqt(priv, &priv->direct_tir[ix].rqt);
1930 int mlx5e_redirect_rqt(struct mlx5e_priv *priv, u32 rqtn, int sz, int ix)
1932 struct mlx5_core_dev *mdev = priv->mdev;
1938 inlen = MLX5_ST_SZ_BYTES(modify_rqt_in) + sizeof(u32) * sz;
1939 in = mlx5_vzalloc(inlen);
1943 rqtc = MLX5_ADDR_OF(modify_rqt_in, in, ctx);
1945 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
1946 if (sz > 1) /* RSS */
1947 mlx5e_fill_indir_rqt_rqns(priv, rqtc);
1949 mlx5e_fill_direct_rqt_rqn(priv, rqtc, ix);
1951 MLX5_SET(modify_rqt_in, in, bitmask.rqn_list, 1);
1953 err = mlx5_core_modify_rqt(mdev, rqtn, in, inlen);
1960 static void mlx5e_redirect_rqts(struct mlx5e_priv *priv)
1965 if (priv->indir_rqt.enabled) {
1966 rqtn = priv->indir_rqt.rqtn;
1967 mlx5e_redirect_rqt(priv, rqtn, MLX5E_INDIR_RQT_SIZE, 0);
1970 for (ix = 0; ix < priv->params.num_channels; ix++) {
1971 if (!priv->direct_tir[ix].rqt.enabled)
1973 rqtn = priv->direct_tir[ix].rqt.rqtn;
1974 mlx5e_redirect_rqt(priv, rqtn, 1, ix);
1978 static void mlx5e_build_tir_ctx_lro(void *tirc, struct mlx5e_priv *priv)
1980 if (!priv->params.lro_en)
1983 #define ROUGH_MAX_L2_L3_HDR_SZ 256
1985 MLX5_SET(tirc, tirc, lro_enable_mask,
1986 MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO |
1987 MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO);
1988 MLX5_SET(tirc, tirc, lro_max_ip_payload_size,
1989 (priv->params.lro_wqe_sz -
1990 ROUGH_MAX_L2_L3_HDR_SZ) >> 8);
1991 MLX5_SET(tirc, tirc, lro_timeout_period_usecs, priv->params.lro_timeout);
1994 void mlx5e_build_tir_ctx_hash(void *tirc, struct mlx5e_priv *priv)
1996 MLX5_SET(tirc, tirc, rx_hash_fn,
1997 mlx5e_rx_hash_fn(priv->params.rss_hfunc));
1998 if (priv->params.rss_hfunc == ETH_RSS_HASH_TOP) {
1999 void *rss_key = MLX5_ADDR_OF(tirc, tirc,
2000 rx_hash_toeplitz_key);
2001 size_t len = MLX5_FLD_SZ_BYTES(tirc,
2002 rx_hash_toeplitz_key);
2004 MLX5_SET(tirc, tirc, rx_hash_symmetric, 1);
2005 memcpy(rss_key, priv->params.toeplitz_hash_key, len);
2009 static int mlx5e_modify_tirs_lro(struct mlx5e_priv *priv)
2011 struct mlx5_core_dev *mdev = priv->mdev;
2020 inlen = MLX5_ST_SZ_BYTES(modify_tir_in);
2021 in = mlx5_vzalloc(inlen);
2025 MLX5_SET(modify_tir_in, in, bitmask.lro, 1);
2026 tirc = MLX5_ADDR_OF(modify_tir_in, in, ctx);
2028 mlx5e_build_tir_ctx_lro(tirc, priv);
2030 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
2031 err = mlx5_core_modify_tir(mdev, priv->indir_tir[tt].tirn, in,
2037 for (ix = 0; ix < priv->profile->max_nch(priv->mdev); ix++) {
2038 err = mlx5_core_modify_tir(mdev, priv->direct_tir[ix].tirn,
2050 static int mlx5e_set_mtu(struct mlx5e_priv *priv, u16 mtu)
2052 struct mlx5_core_dev *mdev = priv->mdev;
2053 u16 hw_mtu = MLX5E_SW2HW_MTU(mtu);
2056 err = mlx5_set_port_mtu(mdev, hw_mtu, 1);
2060 /* Update vport context MTU */
2061 mlx5_modify_nic_vport_mtu(mdev, hw_mtu);
2065 static void mlx5e_query_mtu(struct mlx5e_priv *priv, u16 *mtu)
2067 struct mlx5_core_dev *mdev = priv->mdev;
2071 err = mlx5_query_nic_vport_mtu(mdev, &hw_mtu);
2072 if (err || !hw_mtu) /* fallback to port oper mtu */
2073 mlx5_query_port_oper_mtu(mdev, &hw_mtu, 1);
2075 *mtu = MLX5E_HW2SW_MTU(hw_mtu);
2078 static int mlx5e_set_dev_port_mtu(struct net_device *netdev)
2080 struct mlx5e_priv *priv = netdev_priv(netdev);
2084 err = mlx5e_set_mtu(priv, netdev->mtu);
2088 mlx5e_query_mtu(priv, &mtu);
2089 if (mtu != netdev->mtu)
2090 netdev_warn(netdev, "%s: VPort MTU %d is different than netdev mtu %d\n",
2091 __func__, mtu, netdev->mtu);
2097 static void mlx5e_netdev_set_tcs(struct net_device *netdev)
2099 struct mlx5e_priv *priv = netdev_priv(netdev);
2100 int nch = priv->params.num_channels;
2101 int ntc = priv->params.num_tc;
2104 netdev_reset_tc(netdev);
2109 netdev_set_num_tc(netdev, ntc);
2111 /* Map netdev TCs to offset 0
2112 * We have our own UP to TXQ mapping for QoS
2114 for (tc = 0; tc < ntc; tc++)
2115 netdev_set_tc_queue(netdev, tc, nch, 0);
2118 int mlx5e_open_locked(struct net_device *netdev)
2120 struct mlx5e_priv *priv = netdev_priv(netdev);
2121 struct mlx5_core_dev *mdev = priv->mdev;
2125 set_bit(MLX5E_STATE_OPENED, &priv->state);
2127 mlx5e_netdev_set_tcs(netdev);
2129 num_txqs = priv->params.num_channels * priv->params.num_tc;
2130 netif_set_real_num_tx_queues(netdev, num_txqs);
2131 netif_set_real_num_rx_queues(netdev, priv->params.num_channels);
2133 err = mlx5e_open_channels(priv);
2135 netdev_err(netdev, "%s: mlx5e_open_channels failed, %d\n",
2137 goto err_clear_state_opened_flag;
2140 err = mlx5e_refresh_tirs_self_loopback(priv->mdev, false);
2142 netdev_err(netdev, "%s: mlx5e_refresh_tirs_self_loopback_enable failed, %d\n",
2144 goto err_close_channels;
2147 mlx5e_redirect_rqts(priv);
2148 mlx5e_update_carrier(priv);
2149 mlx5e_timestamp_init(priv);
2150 #ifdef CONFIG_RFS_ACCEL
2151 priv->netdev->rx_cpu_rmap = priv->mdev->rmap;
2153 if (priv->profile->update_stats)
2154 queue_delayed_work(priv->wq, &priv->update_stats_work, 0);
2156 if (MLX5_CAP_GEN(mdev, vport_group_manager)) {
2157 err = mlx5e_add_sqs_fwd_rules(priv);
2159 goto err_close_channels;
2164 mlx5e_close_channels(priv);
2165 err_clear_state_opened_flag:
2166 clear_bit(MLX5E_STATE_OPENED, &priv->state);
2170 int mlx5e_open(struct net_device *netdev)
2172 struct mlx5e_priv *priv = netdev_priv(netdev);
2175 mutex_lock(&priv->state_lock);
2176 err = mlx5e_open_locked(netdev);
2177 mutex_unlock(&priv->state_lock);
2182 int mlx5e_close_locked(struct net_device *netdev)
2184 struct mlx5e_priv *priv = netdev_priv(netdev);
2185 struct mlx5_core_dev *mdev = priv->mdev;
2187 /* May already be CLOSED in case a previous configuration operation
2188 * (e.g RX/TX queue size change) that involves close&open failed.
2190 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
2193 clear_bit(MLX5E_STATE_OPENED, &priv->state);
2195 if (MLX5_CAP_GEN(mdev, vport_group_manager))
2196 mlx5e_remove_sqs_fwd_rules(priv);
2198 mlx5e_timestamp_cleanup(priv);
2199 netif_carrier_off(priv->netdev);
2200 mlx5e_redirect_rqts(priv);
2201 mlx5e_close_channels(priv);
2206 int mlx5e_close(struct net_device *netdev)
2208 struct mlx5e_priv *priv = netdev_priv(netdev);
2211 if (!netif_device_present(netdev))
2214 mutex_lock(&priv->state_lock);
2215 err = mlx5e_close_locked(netdev);
2216 mutex_unlock(&priv->state_lock);
2221 static int mlx5e_create_drop_rq(struct mlx5e_priv *priv,
2222 struct mlx5e_rq *rq,
2223 struct mlx5e_rq_param *param)
2225 struct mlx5_core_dev *mdev = priv->mdev;
2226 void *rqc = param->rqc;
2227 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
2230 param->wq.db_numa_node = param->wq.buf_numa_node;
2232 err = mlx5_wq_ll_create(mdev, ¶m->wq, rqc_wq, &rq->wq,
2242 static int mlx5e_create_drop_cq(struct mlx5e_priv *priv,
2243 struct mlx5e_cq *cq,
2244 struct mlx5e_cq_param *param)
2246 struct mlx5_core_dev *mdev = priv->mdev;
2247 struct mlx5_core_cq *mcq = &cq->mcq;
2252 err = mlx5_cqwq_create(mdev, ¶m->wq, param->cqc, &cq->wq,
2257 mlx5_vector2eqn(mdev, param->eq_ix, &eqn_not_used, &irqn);
2260 mcq->set_ci_db = cq->wq_ctrl.db.db;
2261 mcq->arm_db = cq->wq_ctrl.db.db + 1;
2262 *mcq->set_ci_db = 0;
2264 mcq->vector = param->eq_ix;
2265 mcq->comp = mlx5e_completion_event;
2266 mcq->event = mlx5e_cq_error_event;
2268 mcq->uar = &mdev->mlx5e_res.cq_uar;
2275 static int mlx5e_open_drop_rq(struct mlx5e_priv *priv)
2277 struct mlx5e_cq_param cq_param;
2278 struct mlx5e_rq_param rq_param;
2279 struct mlx5e_rq *rq = &priv->drop_rq;
2280 struct mlx5e_cq *cq = &priv->drop_rq.cq;
2283 memset(&cq_param, 0, sizeof(cq_param));
2284 memset(&rq_param, 0, sizeof(rq_param));
2285 mlx5e_build_drop_rq_param(&rq_param);
2287 err = mlx5e_create_drop_cq(priv, cq, &cq_param);
2291 err = mlx5e_enable_cq(cq, &cq_param);
2293 goto err_destroy_cq;
2295 err = mlx5e_create_drop_rq(priv, rq, &rq_param);
2297 goto err_disable_cq;
2299 err = mlx5e_enable_rq(rq, &rq_param);
2301 goto err_destroy_rq;
2306 mlx5e_destroy_rq(&priv->drop_rq);
2309 mlx5e_disable_cq(&priv->drop_rq.cq);
2312 mlx5e_destroy_cq(&priv->drop_rq.cq);
2317 static void mlx5e_close_drop_rq(struct mlx5e_priv *priv)
2319 mlx5e_disable_rq(&priv->drop_rq);
2320 mlx5e_destroy_rq(&priv->drop_rq);
2321 mlx5e_disable_cq(&priv->drop_rq.cq);
2322 mlx5e_destroy_cq(&priv->drop_rq.cq);
2325 static int mlx5e_create_tis(struct mlx5e_priv *priv, int tc)
2327 struct mlx5_core_dev *mdev = priv->mdev;
2328 u32 in[MLX5_ST_SZ_DW(create_tis_in)] = {0};
2329 void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
2331 MLX5_SET(tisc, tisc, prio, tc << 1);
2332 MLX5_SET(tisc, tisc, transport_domain, mdev->mlx5e_res.td.tdn);
2334 if (mlx5_lag_is_lacp_owner(mdev))
2335 MLX5_SET(tisc, tisc, strict_lag_tx_port_affinity, 1);
2337 return mlx5_core_create_tis(mdev, in, sizeof(in), &priv->tisn[tc]);
2340 static void mlx5e_destroy_tis(struct mlx5e_priv *priv, int tc)
2342 mlx5_core_destroy_tis(priv->mdev, priv->tisn[tc]);
2345 int mlx5e_create_tises(struct mlx5e_priv *priv)
2350 for (tc = 0; tc < priv->profile->max_tc; tc++) {
2351 err = mlx5e_create_tis(priv, tc);
2353 goto err_close_tises;
2359 for (tc--; tc >= 0; tc--)
2360 mlx5e_destroy_tis(priv, tc);
2365 void mlx5e_cleanup_nic_tx(struct mlx5e_priv *priv)
2369 for (tc = 0; tc < priv->profile->max_tc; tc++)
2370 mlx5e_destroy_tis(priv, tc);
2373 static void mlx5e_build_indir_tir_ctx(struct mlx5e_priv *priv, u32 *tirc,
2374 enum mlx5e_traffic_types tt)
2376 void *hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
2378 MLX5_SET(tirc, tirc, transport_domain, priv->mdev->mlx5e_res.td.tdn);
2380 #define MLX5_HASH_IP (MLX5_HASH_FIELD_SEL_SRC_IP |\
2381 MLX5_HASH_FIELD_SEL_DST_IP)
2383 #define MLX5_HASH_IP_L4PORTS (MLX5_HASH_FIELD_SEL_SRC_IP |\
2384 MLX5_HASH_FIELD_SEL_DST_IP |\
2385 MLX5_HASH_FIELD_SEL_L4_SPORT |\
2386 MLX5_HASH_FIELD_SEL_L4_DPORT)
2388 #define MLX5_HASH_IP_IPSEC_SPI (MLX5_HASH_FIELD_SEL_SRC_IP |\
2389 MLX5_HASH_FIELD_SEL_DST_IP |\
2390 MLX5_HASH_FIELD_SEL_IPSEC_SPI)
2392 mlx5e_build_tir_ctx_lro(tirc, priv);
2394 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT);
2395 MLX5_SET(tirc, tirc, indirect_table, priv->indir_rqt.rqtn);
2396 mlx5e_build_tir_ctx_hash(tirc, priv);
2399 case MLX5E_TT_IPV4_TCP:
2400 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2401 MLX5_L3_PROT_TYPE_IPV4);
2402 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2403 MLX5_L4_PROT_TYPE_TCP);
2404 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2405 MLX5_HASH_IP_L4PORTS);
2408 case MLX5E_TT_IPV6_TCP:
2409 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2410 MLX5_L3_PROT_TYPE_IPV6);
2411 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2412 MLX5_L4_PROT_TYPE_TCP);
2413 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2414 MLX5_HASH_IP_L4PORTS);
2417 case MLX5E_TT_IPV4_UDP:
2418 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2419 MLX5_L3_PROT_TYPE_IPV4);
2420 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2421 MLX5_L4_PROT_TYPE_UDP);
2422 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2423 MLX5_HASH_IP_L4PORTS);
2426 case MLX5E_TT_IPV6_UDP:
2427 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2428 MLX5_L3_PROT_TYPE_IPV6);
2429 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2430 MLX5_L4_PROT_TYPE_UDP);
2431 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2432 MLX5_HASH_IP_L4PORTS);
2435 case MLX5E_TT_IPV4_IPSEC_AH:
2436 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2437 MLX5_L3_PROT_TYPE_IPV4);
2438 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2439 MLX5_HASH_IP_IPSEC_SPI);
2442 case MLX5E_TT_IPV6_IPSEC_AH:
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, selected_fields,
2446 MLX5_HASH_IP_IPSEC_SPI);
2449 case MLX5E_TT_IPV4_IPSEC_ESP:
2450 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2451 MLX5_L3_PROT_TYPE_IPV4);
2452 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2453 MLX5_HASH_IP_IPSEC_SPI);
2456 case MLX5E_TT_IPV6_IPSEC_ESP:
2457 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2458 MLX5_L3_PROT_TYPE_IPV6);
2459 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2460 MLX5_HASH_IP_IPSEC_SPI);
2464 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2465 MLX5_L3_PROT_TYPE_IPV4);
2466 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2471 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2472 MLX5_L3_PROT_TYPE_IPV6);
2473 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2478 "mlx5e_build_indir_tir_ctx: bad traffic type!\n");
2482 static void mlx5e_build_direct_tir_ctx(struct mlx5e_priv *priv, u32 *tirc,
2485 MLX5_SET(tirc, tirc, transport_domain, priv->mdev->mlx5e_res.td.tdn);
2487 mlx5e_build_tir_ctx_lro(tirc, priv);
2489 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT);
2490 MLX5_SET(tirc, tirc, indirect_table, rqtn);
2491 MLX5_SET(tirc, tirc, rx_hash_fn, MLX5_RX_HASH_FN_INVERTED_XOR8);
2494 static int mlx5e_create_indirect_tirs(struct mlx5e_priv *priv)
2496 struct mlx5e_tir *tir;
2503 inlen = MLX5_ST_SZ_BYTES(create_tir_in);
2504 in = mlx5_vzalloc(inlen);
2508 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
2509 memset(in, 0, inlen);
2510 tir = &priv->indir_tir[tt];
2511 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
2512 mlx5e_build_indir_tir_ctx(priv, tirc, tt);
2513 err = mlx5e_create_tir(priv->mdev, tir, in, inlen);
2515 goto err_destroy_tirs;
2523 for (tt--; tt >= 0; tt--)
2524 mlx5e_destroy_tir(priv->mdev, &priv->indir_tir[tt]);
2531 int mlx5e_create_direct_tirs(struct mlx5e_priv *priv)
2533 int nch = priv->profile->max_nch(priv->mdev);
2534 struct mlx5e_tir *tir;
2541 inlen = MLX5_ST_SZ_BYTES(create_tir_in);
2542 in = mlx5_vzalloc(inlen);
2546 for (ix = 0; ix < nch; ix++) {
2547 memset(in, 0, inlen);
2548 tir = &priv->direct_tir[ix];
2549 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
2550 mlx5e_build_direct_tir_ctx(priv, tirc,
2551 priv->direct_tir[ix].rqt.rqtn);
2552 err = mlx5e_create_tir(priv->mdev, tir, in, inlen);
2554 goto err_destroy_ch_tirs;
2561 err_destroy_ch_tirs:
2562 for (ix--; ix >= 0; ix--)
2563 mlx5e_destroy_tir(priv->mdev, &priv->direct_tir[ix]);
2570 static void mlx5e_destroy_indirect_tirs(struct mlx5e_priv *priv)
2574 for (i = 0; i < MLX5E_NUM_INDIR_TIRS; i++)
2575 mlx5e_destroy_tir(priv->mdev, &priv->indir_tir[i]);
2578 void mlx5e_destroy_direct_tirs(struct mlx5e_priv *priv)
2580 int nch = priv->profile->max_nch(priv->mdev);
2583 for (i = 0; i < nch; i++)
2584 mlx5e_destroy_tir(priv->mdev, &priv->direct_tir[i]);
2587 int mlx5e_modify_rqs_vsd(struct mlx5e_priv *priv, bool vsd)
2592 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
2595 for (i = 0; i < priv->params.num_channels; i++) {
2596 err = mlx5e_modify_rq_vsd(&priv->channel[i]->rq, vsd);
2604 static int mlx5e_setup_tc(struct net_device *netdev, u8 tc)
2606 struct mlx5e_priv *priv = netdev_priv(netdev);
2610 if (tc && tc != MLX5E_MAX_NUM_TC)
2613 mutex_lock(&priv->state_lock);
2615 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2617 mlx5e_close_locked(priv->netdev);
2619 priv->params.num_tc = tc ? tc : 1;
2622 err = mlx5e_open_locked(priv->netdev);
2624 mutex_unlock(&priv->state_lock);
2629 static int mlx5e_ndo_setup_tc(struct net_device *dev, u32 handle,
2630 __be16 proto, struct tc_to_netdev *tc)
2632 struct mlx5e_priv *priv = netdev_priv(dev);
2634 if (TC_H_MAJ(handle) != TC_H_MAJ(TC_H_INGRESS))
2638 case TC_SETUP_CLSFLOWER:
2639 switch (tc->cls_flower->command) {
2640 case TC_CLSFLOWER_REPLACE:
2641 return mlx5e_configure_flower(priv, proto, tc->cls_flower);
2642 case TC_CLSFLOWER_DESTROY:
2643 return mlx5e_delete_flower(priv, tc->cls_flower);
2644 case TC_CLSFLOWER_STATS:
2645 return mlx5e_stats_flower(priv, tc->cls_flower);
2652 if (tc->type != TC_SETUP_MQPRIO)
2655 return mlx5e_setup_tc(dev, tc->tc);
2658 static struct rtnl_link_stats64 *
2659 mlx5e_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats)
2661 struct mlx5e_priv *priv = netdev_priv(dev);
2662 struct mlx5e_sw_stats *sstats = &priv->stats.sw;
2663 struct mlx5e_vport_stats *vstats = &priv->stats.vport;
2664 struct mlx5e_pport_stats *pstats = &priv->stats.pport;
2666 if (mlx5e_is_uplink_rep(priv)) {
2667 stats->rx_packets = PPORT_802_3_GET(pstats, a_frames_received_ok);
2668 stats->rx_bytes = PPORT_802_3_GET(pstats, a_octets_received_ok);
2669 stats->tx_packets = PPORT_802_3_GET(pstats, a_frames_transmitted_ok);
2670 stats->tx_bytes = PPORT_802_3_GET(pstats, a_octets_transmitted_ok);
2672 stats->rx_packets = sstats->rx_packets;
2673 stats->rx_bytes = sstats->rx_bytes;
2674 stats->tx_packets = sstats->tx_packets;
2675 stats->tx_bytes = sstats->tx_bytes;
2676 stats->tx_dropped = sstats->tx_queue_dropped;
2679 stats->rx_dropped = priv->stats.qcnt.rx_out_of_buffer;
2681 stats->rx_length_errors =
2682 PPORT_802_3_GET(pstats, a_in_range_length_errors) +
2683 PPORT_802_3_GET(pstats, a_out_of_range_length_field) +
2684 PPORT_802_3_GET(pstats, a_frame_too_long_errors);
2685 stats->rx_crc_errors =
2686 PPORT_802_3_GET(pstats, a_frame_check_sequence_errors);
2687 stats->rx_frame_errors = PPORT_802_3_GET(pstats, a_alignment_errors);
2688 stats->tx_aborted_errors = PPORT_2863_GET(pstats, if_out_discards);
2689 stats->tx_carrier_errors =
2690 PPORT_802_3_GET(pstats, a_symbol_error_during_carrier);
2691 stats->rx_errors = stats->rx_length_errors + stats->rx_crc_errors +
2692 stats->rx_frame_errors;
2693 stats->tx_errors = stats->tx_aborted_errors + stats->tx_carrier_errors;
2695 /* vport multicast also counts packets that are dropped due to steering
2696 * or rx out of buffer
2699 VPORT_COUNTER_GET(vstats, received_eth_multicast.packets);
2704 static void mlx5e_set_rx_mode(struct net_device *dev)
2706 struct mlx5e_priv *priv = netdev_priv(dev);
2708 queue_work(priv->wq, &priv->set_rx_mode_work);
2711 static int mlx5e_set_mac(struct net_device *netdev, void *addr)
2713 struct mlx5e_priv *priv = netdev_priv(netdev);
2714 struct sockaddr *saddr = addr;
2716 if (!is_valid_ether_addr(saddr->sa_data))
2717 return -EADDRNOTAVAIL;
2719 netif_addr_lock_bh(netdev);
2720 ether_addr_copy(netdev->dev_addr, saddr->sa_data);
2721 netif_addr_unlock_bh(netdev);
2723 queue_work(priv->wq, &priv->set_rx_mode_work);
2728 #define MLX5E_SET_FEATURE(netdev, feature, enable) \
2731 netdev->features |= feature; \
2733 netdev->features &= ~feature; \
2736 typedef int (*mlx5e_feature_handler)(struct net_device *netdev, bool enable);
2738 static int set_feature_lro(struct net_device *netdev, bool enable)
2740 struct mlx5e_priv *priv = netdev_priv(netdev);
2741 bool was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2744 mutex_lock(&priv->state_lock);
2746 if (was_opened && (priv->params.rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST))
2747 mlx5e_close_locked(priv->netdev);
2749 priv->params.lro_en = enable;
2750 err = mlx5e_modify_tirs_lro(priv);
2752 netdev_err(netdev, "lro modify failed, %d\n", err);
2753 priv->params.lro_en = !enable;
2756 if (was_opened && (priv->params.rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST))
2757 mlx5e_open_locked(priv->netdev);
2759 mutex_unlock(&priv->state_lock);
2764 static int set_feature_vlan_filter(struct net_device *netdev, bool enable)
2766 struct mlx5e_priv *priv = netdev_priv(netdev);
2769 mlx5e_enable_vlan_filter(priv);
2771 mlx5e_disable_vlan_filter(priv);
2776 static int set_feature_tc_num_filters(struct net_device *netdev, bool enable)
2778 struct mlx5e_priv *priv = netdev_priv(netdev);
2780 if (!enable && mlx5e_tc_num_filters(priv)) {
2782 "Active offloaded tc filters, can't turn hw_tc_offload off\n");
2789 static int set_feature_rx_all(struct net_device *netdev, bool enable)
2791 struct mlx5e_priv *priv = netdev_priv(netdev);
2792 struct mlx5_core_dev *mdev = priv->mdev;
2794 return mlx5_set_port_fcs(mdev, !enable);
2797 static int set_feature_rx_vlan(struct net_device *netdev, bool enable)
2799 struct mlx5e_priv *priv = netdev_priv(netdev);
2802 mutex_lock(&priv->state_lock);
2804 priv->params.vlan_strip_disable = !enable;
2805 err = mlx5e_modify_rqs_vsd(priv, !enable);
2807 priv->params.vlan_strip_disable = enable;
2809 mutex_unlock(&priv->state_lock);
2814 #ifdef CONFIG_RFS_ACCEL
2815 static int set_feature_arfs(struct net_device *netdev, bool enable)
2817 struct mlx5e_priv *priv = netdev_priv(netdev);
2821 err = mlx5e_arfs_enable(priv);
2823 err = mlx5e_arfs_disable(priv);
2829 static int mlx5e_handle_feature(struct net_device *netdev,
2830 netdev_features_t wanted_features,
2831 netdev_features_t feature,
2832 mlx5e_feature_handler feature_handler)
2834 netdev_features_t changes = wanted_features ^ netdev->features;
2835 bool enable = !!(wanted_features & feature);
2838 if (!(changes & feature))
2841 err = feature_handler(netdev, enable);
2843 netdev_err(netdev, "%s feature 0x%llx failed err %d\n",
2844 enable ? "Enable" : "Disable", feature, err);
2848 MLX5E_SET_FEATURE(netdev, feature, enable);
2852 static int mlx5e_set_features(struct net_device *netdev,
2853 netdev_features_t features)
2857 err = mlx5e_handle_feature(netdev, features, NETIF_F_LRO,
2859 err |= mlx5e_handle_feature(netdev, features,
2860 NETIF_F_HW_VLAN_CTAG_FILTER,
2861 set_feature_vlan_filter);
2862 err |= mlx5e_handle_feature(netdev, features, NETIF_F_HW_TC,
2863 set_feature_tc_num_filters);
2864 err |= mlx5e_handle_feature(netdev, features, NETIF_F_RXALL,
2865 set_feature_rx_all);
2866 err |= mlx5e_handle_feature(netdev, features, NETIF_F_HW_VLAN_CTAG_RX,
2867 set_feature_rx_vlan);
2868 #ifdef CONFIG_RFS_ACCEL
2869 err |= mlx5e_handle_feature(netdev, features, NETIF_F_NTUPLE,
2873 return err ? -EINVAL : 0;
2876 static int mlx5e_change_mtu(struct net_device *netdev, int new_mtu)
2878 struct mlx5e_priv *priv = netdev_priv(netdev);
2883 mutex_lock(&priv->state_lock);
2885 reset = !priv->params.lro_en &&
2886 (priv->params.rq_wq_type !=
2887 MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ);
2889 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
2890 if (was_opened && reset)
2891 mlx5e_close_locked(netdev);
2893 netdev->mtu = new_mtu;
2894 mlx5e_set_dev_port_mtu(netdev);
2896 if (was_opened && reset)
2897 err = mlx5e_open_locked(netdev);
2899 mutex_unlock(&priv->state_lock);
2904 static int mlx5e_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2908 return mlx5e_hwstamp_set(dev, ifr);
2910 return mlx5e_hwstamp_get(dev, ifr);
2916 static int mlx5e_set_vf_mac(struct net_device *dev, int vf, u8 *mac)
2918 struct mlx5e_priv *priv = netdev_priv(dev);
2919 struct mlx5_core_dev *mdev = priv->mdev;
2921 return mlx5_eswitch_set_vport_mac(mdev->priv.eswitch, vf + 1, mac);
2924 static int mlx5e_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
2927 struct mlx5e_priv *priv = netdev_priv(dev);
2928 struct mlx5_core_dev *mdev = priv->mdev;
2930 if (vlan_proto != htons(ETH_P_8021Q))
2931 return -EPROTONOSUPPORT;
2933 return mlx5_eswitch_set_vport_vlan(mdev->priv.eswitch, vf + 1,
2937 static int mlx5e_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2939 struct mlx5e_priv *priv = netdev_priv(dev);
2940 struct mlx5_core_dev *mdev = priv->mdev;
2942 return mlx5_eswitch_set_vport_spoofchk(mdev->priv.eswitch, vf + 1, setting);
2945 static int mlx5e_set_vf_trust(struct net_device *dev, int vf, bool setting)
2947 struct mlx5e_priv *priv = netdev_priv(dev);
2948 struct mlx5_core_dev *mdev = priv->mdev;
2950 return mlx5_eswitch_set_vport_trust(mdev->priv.eswitch, vf + 1, setting);
2953 static int mlx5e_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2956 struct mlx5e_priv *priv = netdev_priv(dev);
2957 struct mlx5_core_dev *mdev = priv->mdev;
2962 return mlx5_eswitch_set_vport_rate(mdev->priv.eswitch, vf + 1,
2966 static int mlx5_vport_link2ifla(u8 esw_link)
2969 case MLX5_ESW_VPORT_ADMIN_STATE_DOWN:
2970 return IFLA_VF_LINK_STATE_DISABLE;
2971 case MLX5_ESW_VPORT_ADMIN_STATE_UP:
2972 return IFLA_VF_LINK_STATE_ENABLE;
2974 return IFLA_VF_LINK_STATE_AUTO;
2977 static int mlx5_ifla_link2vport(u8 ifla_link)
2979 switch (ifla_link) {
2980 case IFLA_VF_LINK_STATE_DISABLE:
2981 return MLX5_ESW_VPORT_ADMIN_STATE_DOWN;
2982 case IFLA_VF_LINK_STATE_ENABLE:
2983 return MLX5_ESW_VPORT_ADMIN_STATE_UP;
2985 return MLX5_ESW_VPORT_ADMIN_STATE_AUTO;
2988 static int mlx5e_set_vf_link_state(struct net_device *dev, int vf,
2991 struct mlx5e_priv *priv = netdev_priv(dev);
2992 struct mlx5_core_dev *mdev = priv->mdev;
2994 return mlx5_eswitch_set_vport_state(mdev->priv.eswitch, vf + 1,
2995 mlx5_ifla_link2vport(link_state));
2998 static int mlx5e_get_vf_config(struct net_device *dev,
2999 int vf, struct ifla_vf_info *ivi)
3001 struct mlx5e_priv *priv = netdev_priv(dev);
3002 struct mlx5_core_dev *mdev = priv->mdev;
3005 err = mlx5_eswitch_get_vport_config(mdev->priv.eswitch, vf + 1, ivi);
3008 ivi->linkstate = mlx5_vport_link2ifla(ivi->linkstate);
3012 static int mlx5e_get_vf_stats(struct net_device *dev,
3013 int vf, struct ifla_vf_stats *vf_stats)
3015 struct mlx5e_priv *priv = netdev_priv(dev);
3016 struct mlx5_core_dev *mdev = priv->mdev;
3018 return mlx5_eswitch_get_vport_stats(mdev->priv.eswitch, vf + 1,
3022 void mlx5e_add_vxlan_port(struct net_device *netdev,
3023 struct udp_tunnel_info *ti)
3025 struct mlx5e_priv *priv = netdev_priv(netdev);
3027 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3030 if (!mlx5e_vxlan_allowed(priv->mdev))
3033 mlx5e_vxlan_queue_work(priv, ti->sa_family, be16_to_cpu(ti->port), 1);
3036 void mlx5e_del_vxlan_port(struct net_device *netdev,
3037 struct udp_tunnel_info *ti)
3039 struct mlx5e_priv *priv = netdev_priv(netdev);
3041 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3044 if (!mlx5e_vxlan_allowed(priv->mdev))
3047 mlx5e_vxlan_queue_work(priv, ti->sa_family, be16_to_cpu(ti->port), 0);
3050 static netdev_features_t mlx5e_vxlan_features_check(struct mlx5e_priv *priv,
3051 struct sk_buff *skb,
3052 netdev_features_t features)
3054 struct udphdr *udph;
3058 switch (vlan_get_protocol(skb)) {
3059 case htons(ETH_P_IP):
3060 proto = ip_hdr(skb)->protocol;
3062 case htons(ETH_P_IPV6):
3063 proto = ipv6_hdr(skb)->nexthdr;
3069 if (proto == IPPROTO_UDP) {
3070 udph = udp_hdr(skb);
3071 port = be16_to_cpu(udph->dest);
3074 /* Verify if UDP port is being offloaded by HW */
3075 if (port && mlx5e_vxlan_lookup_port(priv, port))
3079 /* Disable CSUM and GSO if the udp dport is not offloaded by HW */
3080 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3083 static netdev_features_t mlx5e_features_check(struct sk_buff *skb,
3084 struct net_device *netdev,
3085 netdev_features_t features)
3087 struct mlx5e_priv *priv = netdev_priv(netdev);
3089 features = vlan_features_check(skb, features);
3090 features = vxlan_features_check(skb, features);
3092 /* Validate if the tunneled packet is being offloaded by HW */
3093 if (skb->encapsulation &&
3094 (features & NETIF_F_CSUM_MASK || features & NETIF_F_GSO_MASK))
3095 return mlx5e_vxlan_features_check(priv, skb, features);
3100 static void mlx5e_tx_timeout(struct net_device *dev)
3102 struct mlx5e_priv *priv = netdev_priv(dev);
3103 bool sched_work = false;
3106 netdev_err(dev, "TX timeout detected\n");
3108 for (i = 0; i < priv->params.num_channels * priv->params.num_tc; i++) {
3109 struct mlx5e_sq *sq = priv->txq_to_sq_map[i];
3111 if (!netif_xmit_stopped(netdev_get_tx_queue(dev, i)))
3114 set_bit(MLX5E_SQ_STATE_FLUSH, &sq->state);
3115 netdev_err(dev, "TX timeout on queue: %d, SQ: 0x%x, CQ: 0x%x, SQ Cons: 0x%x SQ Prod: 0x%x\n",
3116 i, sq->sqn, sq->cq.mcq.cqn, sq->cc, sq->pc);
3119 if (sched_work && test_bit(MLX5E_STATE_OPENED, &priv->state))
3120 schedule_work(&priv->tx_timeout_work);
3123 static int mlx5e_xdp_set(struct net_device *netdev, struct bpf_prog *prog)
3125 struct mlx5e_priv *priv = netdev_priv(netdev);
3126 struct bpf_prog *old_prog;
3128 bool reset, was_opened;
3131 mutex_lock(&priv->state_lock);
3133 if ((netdev->features & NETIF_F_LRO) && prog) {
3134 netdev_warn(netdev, "can't set XDP while LRO is on, disable LRO first\n");
3139 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
3140 /* no need for full reset when exchanging programs */
3141 reset = (!priv->xdp_prog || !prog);
3143 if (was_opened && reset)
3144 mlx5e_close_locked(netdev);
3145 if (was_opened && !reset) {
3146 /* num_channels is invariant here, so we can take the
3147 * batched reference right upfront.
3149 prog = bpf_prog_add(prog, priv->params.num_channels);
3151 err = PTR_ERR(prog);
3156 /* exchange programs, extra prog reference we got from caller
3157 * as long as we don't fail from this point onwards.
3159 old_prog = xchg(&priv->xdp_prog, prog);
3161 bpf_prog_put(old_prog);
3163 if (reset) /* change RQ type according to priv->xdp_prog */
3164 mlx5e_set_rq_priv_params(priv);
3166 if (was_opened && reset)
3167 mlx5e_open_locked(netdev);
3169 if (!test_bit(MLX5E_STATE_OPENED, &priv->state) || reset)
3172 /* exchanging programs w/o reset, we update ref counts on behalf
3173 * of the channels RQs here.
3175 for (i = 0; i < priv->params.num_channels; i++) {
3176 struct mlx5e_channel *c = priv->channel[i];
3178 set_bit(MLX5E_RQ_STATE_FLUSH, &c->rq.state);
3179 napi_synchronize(&c->napi);
3180 /* prevent mlx5e_poll_rx_cq from accessing rq->xdp_prog */
3182 old_prog = xchg(&c->rq.xdp_prog, prog);
3184 clear_bit(MLX5E_RQ_STATE_FLUSH, &c->rq.state);
3185 /* napi_schedule in case we have missed anything */
3186 set_bit(MLX5E_CHANNEL_NAPI_SCHED, &c->flags);
3187 napi_schedule(&c->napi);
3190 bpf_prog_put(old_prog);
3194 mutex_unlock(&priv->state_lock);
3198 static bool mlx5e_xdp_attached(struct net_device *dev)
3200 struct mlx5e_priv *priv = netdev_priv(dev);
3202 return !!priv->xdp_prog;
3205 static int mlx5e_xdp(struct net_device *dev, struct netdev_xdp *xdp)
3207 switch (xdp->command) {
3208 case XDP_SETUP_PROG:
3209 return mlx5e_xdp_set(dev, xdp->prog);
3210 case XDP_QUERY_PROG:
3211 xdp->prog_attached = mlx5e_xdp_attached(dev);
3218 #ifdef CONFIG_NET_POLL_CONTROLLER
3219 /* Fake "interrupt" called by netpoll (eg netconsole) to send skbs without
3220 * reenabling interrupts.
3222 static void mlx5e_netpoll(struct net_device *dev)
3224 struct mlx5e_priv *priv = netdev_priv(dev);
3227 for (i = 0; i < priv->params.num_channels; i++)
3228 napi_schedule(&priv->channel[i]->napi);
3232 static const struct net_device_ops mlx5e_netdev_ops_basic = {
3233 .ndo_open = mlx5e_open,
3234 .ndo_stop = mlx5e_close,
3235 .ndo_start_xmit = mlx5e_xmit,
3236 .ndo_setup_tc = mlx5e_ndo_setup_tc,
3237 .ndo_select_queue = mlx5e_select_queue,
3238 .ndo_get_stats64 = mlx5e_get_stats,
3239 .ndo_set_rx_mode = mlx5e_set_rx_mode,
3240 .ndo_set_mac_address = mlx5e_set_mac,
3241 .ndo_vlan_rx_add_vid = mlx5e_vlan_rx_add_vid,
3242 .ndo_vlan_rx_kill_vid = mlx5e_vlan_rx_kill_vid,
3243 .ndo_set_features = mlx5e_set_features,
3244 .ndo_change_mtu = mlx5e_change_mtu,
3245 .ndo_do_ioctl = mlx5e_ioctl,
3246 .ndo_set_tx_maxrate = mlx5e_set_tx_maxrate,
3247 #ifdef CONFIG_RFS_ACCEL
3248 .ndo_rx_flow_steer = mlx5e_rx_flow_steer,
3250 .ndo_tx_timeout = mlx5e_tx_timeout,
3251 .ndo_xdp = mlx5e_xdp,
3252 #ifdef CONFIG_NET_POLL_CONTROLLER
3253 .ndo_poll_controller = mlx5e_netpoll,
3257 static const struct net_device_ops mlx5e_netdev_ops_sriov = {
3258 .ndo_open = mlx5e_open,
3259 .ndo_stop = mlx5e_close,
3260 .ndo_start_xmit = mlx5e_xmit,
3261 .ndo_setup_tc = mlx5e_ndo_setup_tc,
3262 .ndo_select_queue = mlx5e_select_queue,
3263 .ndo_get_stats64 = mlx5e_get_stats,
3264 .ndo_set_rx_mode = mlx5e_set_rx_mode,
3265 .ndo_set_mac_address = mlx5e_set_mac,
3266 .ndo_vlan_rx_add_vid = mlx5e_vlan_rx_add_vid,
3267 .ndo_vlan_rx_kill_vid = mlx5e_vlan_rx_kill_vid,
3268 .ndo_set_features = mlx5e_set_features,
3269 .ndo_change_mtu = mlx5e_change_mtu,
3270 .ndo_do_ioctl = mlx5e_ioctl,
3271 .ndo_udp_tunnel_add = mlx5e_add_vxlan_port,
3272 .ndo_udp_tunnel_del = mlx5e_del_vxlan_port,
3273 .ndo_set_tx_maxrate = mlx5e_set_tx_maxrate,
3274 .ndo_features_check = mlx5e_features_check,
3275 #ifdef CONFIG_RFS_ACCEL
3276 .ndo_rx_flow_steer = mlx5e_rx_flow_steer,
3278 .ndo_set_vf_mac = mlx5e_set_vf_mac,
3279 .ndo_set_vf_vlan = mlx5e_set_vf_vlan,
3280 .ndo_set_vf_spoofchk = mlx5e_set_vf_spoofchk,
3281 .ndo_set_vf_trust = mlx5e_set_vf_trust,
3282 .ndo_set_vf_rate = mlx5e_set_vf_rate,
3283 .ndo_get_vf_config = mlx5e_get_vf_config,
3284 .ndo_set_vf_link_state = mlx5e_set_vf_link_state,
3285 .ndo_get_vf_stats = mlx5e_get_vf_stats,
3286 .ndo_tx_timeout = mlx5e_tx_timeout,
3287 .ndo_xdp = mlx5e_xdp,
3288 #ifdef CONFIG_NET_POLL_CONTROLLER
3289 .ndo_poll_controller = mlx5e_netpoll,
3291 .ndo_has_offload_stats = mlx5e_has_offload_stats,
3292 .ndo_get_offload_stats = mlx5e_get_offload_stats,
3295 static int mlx5e_check_required_hca_cap(struct mlx5_core_dev *mdev)
3297 if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
3299 if (!MLX5_CAP_GEN(mdev, eth_net_offloads) ||
3300 !MLX5_CAP_GEN(mdev, nic_flow_table) ||
3301 !MLX5_CAP_ETH(mdev, csum_cap) ||
3302 !MLX5_CAP_ETH(mdev, max_lso_cap) ||
3303 !MLX5_CAP_ETH(mdev, vlan_cap) ||
3304 !MLX5_CAP_ETH(mdev, rss_ind_tbl_cap) ||
3305 MLX5_CAP_FLOWTABLE(mdev,
3306 flow_table_properties_nic_receive.max_ft_level)
3308 mlx5_core_warn(mdev,
3309 "Not creating net device, some required device capabilities are missing\n");
3312 if (!MLX5_CAP_ETH(mdev, self_lb_en_modifiable))
3313 mlx5_core_warn(mdev, "Self loop back prevention is not supported\n");
3314 if (!MLX5_CAP_GEN(mdev, cq_moderation))
3315 mlx5_core_warn(mdev, "CQ modiration is not supported\n");
3320 u16 mlx5e_get_max_inline_cap(struct mlx5_core_dev *mdev)
3322 int bf_buf_size = (1 << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2;
3324 return bf_buf_size -
3325 sizeof(struct mlx5e_tx_wqe) +
3326 2 /*sizeof(mlx5e_tx_wqe.inline_hdr_start)*/;
3329 void mlx5e_build_default_indir_rqt(struct mlx5_core_dev *mdev,
3330 u32 *indirection_rqt, int len,
3333 int node = mdev->priv.numa_node;
3334 int node_num_of_cores;
3338 node = first_online_node;
3340 node_num_of_cores = cpumask_weight(cpumask_of_node(node));
3342 if (node_num_of_cores)
3343 num_channels = min_t(int, num_channels, node_num_of_cores);
3345 for (i = 0; i < len; i++)
3346 indirection_rqt[i] = i % num_channels;
3349 static int mlx5e_get_pci_bw(struct mlx5_core_dev *mdev, u32 *pci_bw)
3351 enum pcie_link_width width;
3352 enum pci_bus_speed speed;
3355 err = pcie_get_minimum_link(mdev->pdev, &speed, &width);
3359 if (speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN)
3363 case PCIE_SPEED_2_5GT:
3364 *pci_bw = 2500 * width;
3366 case PCIE_SPEED_5_0GT:
3367 *pci_bw = 5000 * width;
3369 case PCIE_SPEED_8_0GT:
3370 *pci_bw = 8000 * width;
3379 static bool cqe_compress_heuristic(u32 link_speed, u32 pci_bw)
3381 return (link_speed && pci_bw &&
3382 (pci_bw < 40000) && (pci_bw < link_speed));
3385 void mlx5e_set_rx_cq_mode_params(struct mlx5e_params *params, u8 cq_period_mode)
3387 params->rx_cq_period_mode = cq_period_mode;
3389 params->rx_cq_moderation.pkts =
3390 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS;
3391 params->rx_cq_moderation.usec =
3392 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
3394 if (cq_period_mode == MLX5_CQ_PERIOD_MODE_START_FROM_CQE)
3395 params->rx_cq_moderation.usec =
3396 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE;
3399 static void mlx5e_query_min_inline(struct mlx5_core_dev *mdev,
3400 u8 *min_inline_mode)
3402 switch (MLX5_CAP_ETH(mdev, wqe_inline_mode)) {
3403 case MLX5_CAP_INLINE_MODE_L2:
3404 *min_inline_mode = MLX5_INLINE_MODE_L2;
3406 case MLX5_CAP_INLINE_MODE_VPORT_CONTEXT:
3407 mlx5_query_nic_vport_min_inline(mdev, 0, min_inline_mode);
3409 case MLX5_CAP_INLINE_MODE_NOT_REQUIRED:
3410 *min_inline_mode = MLX5_INLINE_MODE_NONE;
3415 u32 mlx5e_choose_lro_timeout(struct mlx5_core_dev *mdev, u32 wanted_timeout)
3419 /* The supported periods are organized in ascending order */
3420 for (i = 0; i < MLX5E_LRO_TIMEOUT_ARR_SIZE - 1; i++)
3421 if (MLX5_CAP_ETH(mdev, lro_timer_supported_periods[i]) >= wanted_timeout)
3424 return MLX5_CAP_ETH(mdev, lro_timer_supported_periods[i]);
3427 static void mlx5e_build_nic_netdev_priv(struct mlx5_core_dev *mdev,
3428 struct net_device *netdev,
3429 const struct mlx5e_profile *profile,
3432 struct mlx5e_priv *priv = netdev_priv(netdev);
3435 u8 cq_period_mode = MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
3436 MLX5_CQ_PERIOD_MODE_START_FROM_CQE :
3437 MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
3440 priv->netdev = netdev;
3441 priv->params.num_channels = profile->max_nch(mdev);
3442 priv->profile = profile;
3443 priv->ppriv = ppriv;
3445 priv->params.lro_timeout =
3446 mlx5e_choose_lro_timeout(mdev, MLX5E_DEFAULT_LRO_TIMEOUT);
3448 priv->params.log_sq_size = MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
3450 /* set CQE compression */
3451 priv->params.rx_cqe_compress_def = false;
3452 if (MLX5_CAP_GEN(mdev, cqe_compression) &&
3453 MLX5_CAP_GEN(mdev, vport_group_manager)) {
3454 mlx5e_get_max_linkspeed(mdev, &link_speed);
3455 mlx5e_get_pci_bw(mdev, &pci_bw);
3456 mlx5_core_dbg(mdev, "Max link speed = %d, PCI BW = %d\n",
3457 link_speed, pci_bw);
3458 priv->params.rx_cqe_compress_def =
3459 cqe_compress_heuristic(link_speed, pci_bw);
3462 mlx5e_set_rq_priv_params(priv);
3463 if (priv->params.rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ)
3464 priv->params.lro_en = true;
3466 priv->params.rx_am_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
3467 mlx5e_set_rx_cq_mode_params(&priv->params, cq_period_mode);
3469 priv->params.tx_cq_moderation.usec =
3470 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC;
3471 priv->params.tx_cq_moderation.pkts =
3472 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
3473 priv->params.tx_max_inline = mlx5e_get_max_inline_cap(mdev);
3474 mlx5e_query_min_inline(mdev, &priv->params.tx_min_inline_mode);
3475 priv->params.num_tc = 1;
3476 priv->params.rss_hfunc = ETH_RSS_HASH_XOR;
3478 netdev_rss_key_fill(priv->params.toeplitz_hash_key,
3479 sizeof(priv->params.toeplitz_hash_key));
3481 mlx5e_build_default_indir_rqt(mdev, priv->params.indirection_rqt,
3482 MLX5E_INDIR_RQT_SIZE, profile->max_nch(mdev));
3484 priv->params.lro_wqe_sz =
3485 MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ -
3486 /* Extra room needed for build_skb */
3488 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
3490 /* Initialize pflags */
3491 MLX5E_SET_PFLAG(priv, MLX5E_PFLAG_RX_CQE_BASED_MODER,
3492 priv->params.rx_cq_period_mode == MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
3493 MLX5E_SET_PFLAG(priv, MLX5E_PFLAG_RX_CQE_COMPRESS, priv->params.rx_cqe_compress_def);
3495 mutex_init(&priv->state_lock);
3497 INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
3498 INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
3499 INIT_WORK(&priv->tx_timeout_work, mlx5e_tx_timeout_work);
3500 INIT_DELAYED_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
3503 static void mlx5e_set_netdev_dev_addr(struct net_device *netdev)
3505 struct mlx5e_priv *priv = netdev_priv(netdev);
3507 mlx5_query_nic_vport_mac_address(priv->mdev, 0, netdev->dev_addr);
3508 if (is_zero_ether_addr(netdev->dev_addr) &&
3509 !MLX5_CAP_GEN(priv->mdev, vport_group_manager)) {
3510 eth_hw_addr_random(netdev);
3511 mlx5_core_info(priv->mdev, "Assigned random MAC address %pM\n", netdev->dev_addr);
3515 static const struct switchdev_ops mlx5e_switchdev_ops = {
3516 .switchdev_port_attr_get = mlx5e_attr_get,
3519 static void mlx5e_build_nic_netdev(struct net_device *netdev)
3521 struct mlx5e_priv *priv = netdev_priv(netdev);
3522 struct mlx5_core_dev *mdev = priv->mdev;
3526 SET_NETDEV_DEV(netdev, &mdev->pdev->dev);
3528 if (MLX5_CAP_GEN(mdev, vport_group_manager)) {
3529 netdev->netdev_ops = &mlx5e_netdev_ops_sriov;
3530 #ifdef CONFIG_MLX5_CORE_EN_DCB
3531 if (MLX5_CAP_GEN(mdev, qos))
3532 netdev->dcbnl_ops = &mlx5e_dcbnl_ops;
3535 netdev->netdev_ops = &mlx5e_netdev_ops_basic;
3538 netdev->watchdog_timeo = 15 * HZ;
3540 netdev->ethtool_ops = &mlx5e_ethtool_ops;
3542 netdev->vlan_features |= NETIF_F_SG;
3543 netdev->vlan_features |= NETIF_F_IP_CSUM;
3544 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
3545 netdev->vlan_features |= NETIF_F_GRO;
3546 netdev->vlan_features |= NETIF_F_TSO;
3547 netdev->vlan_features |= NETIF_F_TSO6;
3548 netdev->vlan_features |= NETIF_F_RXCSUM;
3549 netdev->vlan_features |= NETIF_F_RXHASH;
3551 if (!!MLX5_CAP_ETH(mdev, lro_cap))
3552 netdev->vlan_features |= NETIF_F_LRO;
3554 netdev->hw_features = netdev->vlan_features;
3555 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
3556 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
3557 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
3559 if (mlx5e_vxlan_allowed(mdev)) {
3560 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3561 NETIF_F_GSO_UDP_TUNNEL_CSUM |
3562 NETIF_F_GSO_PARTIAL;
3563 netdev->hw_enc_features |= NETIF_F_IP_CSUM;
3564 netdev->hw_enc_features |= NETIF_F_IPV6_CSUM;
3565 netdev->hw_enc_features |= NETIF_F_TSO;
3566 netdev->hw_enc_features |= NETIF_F_TSO6;
3567 netdev->hw_enc_features |= NETIF_F_GSO_UDP_TUNNEL;
3568 netdev->hw_enc_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM |
3569 NETIF_F_GSO_PARTIAL;
3570 netdev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
3573 mlx5_query_port_fcs(mdev, &fcs_supported, &fcs_enabled);
3576 netdev->hw_features |= NETIF_F_RXALL;
3578 netdev->features = netdev->hw_features;
3579 if (!priv->params.lro_en)
3580 netdev->features &= ~NETIF_F_LRO;
3583 netdev->features &= ~NETIF_F_RXALL;
3585 #define FT_CAP(f) MLX5_CAP_FLOWTABLE(mdev, flow_table_properties_nic_receive.f)
3586 if (FT_CAP(flow_modify_en) &&
3587 FT_CAP(modify_root) &&
3588 FT_CAP(identified_miss_table_mode) &&
3589 FT_CAP(flow_table_modify)) {
3590 netdev->hw_features |= NETIF_F_HW_TC;
3591 #ifdef CONFIG_RFS_ACCEL
3592 netdev->hw_features |= NETIF_F_NTUPLE;
3596 netdev->features |= NETIF_F_HIGHDMA;
3598 netdev->priv_flags |= IFF_UNICAST_FLT;
3600 mlx5e_set_netdev_dev_addr(netdev);
3602 #ifdef CONFIG_NET_SWITCHDEV
3603 if (MLX5_CAP_GEN(mdev, vport_group_manager))
3604 netdev->switchdev_ops = &mlx5e_switchdev_ops;
3608 static void mlx5e_create_q_counter(struct mlx5e_priv *priv)
3610 struct mlx5_core_dev *mdev = priv->mdev;
3613 err = mlx5_core_alloc_q_counter(mdev, &priv->q_counter);
3615 mlx5_core_warn(mdev, "alloc queue counter failed, %d\n", err);
3616 priv->q_counter = 0;
3620 static void mlx5e_destroy_q_counter(struct mlx5e_priv *priv)
3622 if (!priv->q_counter)
3625 mlx5_core_dealloc_q_counter(priv->mdev, priv->q_counter);
3628 static int mlx5e_create_umr_mkey(struct mlx5e_priv *priv)
3630 struct mlx5_core_dev *mdev = priv->mdev;
3631 u64 npages = MLX5E_REQUIRED_MTTS(priv->profile->max_nch(mdev),
3632 BIT(MLX5E_PARAMS_MAXIMUM_LOG_RQ_SIZE_MPW));
3633 int inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
3638 in = mlx5_vzalloc(inlen);
3642 mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
3644 npages = min_t(u32, ALIGN(U16_MAX, 4) * 2, npages);
3646 MLX5_SET(mkc, mkc, free, 1);
3647 MLX5_SET(mkc, mkc, umr_en, 1);
3648 MLX5_SET(mkc, mkc, lw, 1);
3649 MLX5_SET(mkc, mkc, lr, 1);
3650 MLX5_SET(mkc, mkc, access_mode, MLX5_MKC_ACCESS_MODE_MTT);
3652 MLX5_SET(mkc, mkc, qpn, 0xffffff);
3653 MLX5_SET(mkc, mkc, pd, mdev->mlx5e_res.pdn);
3654 MLX5_SET64(mkc, mkc, len, npages << PAGE_SHIFT);
3655 MLX5_SET(mkc, mkc, translations_octword_size,
3656 MLX5_MTT_OCTW(npages));
3657 MLX5_SET(mkc, mkc, log_page_size, PAGE_SHIFT);
3659 err = mlx5_core_create_mkey(mdev, &priv->umr_mkey, in, inlen);
3665 static void mlx5e_nic_init(struct mlx5_core_dev *mdev,
3666 struct net_device *netdev,
3667 const struct mlx5e_profile *profile,
3670 struct mlx5e_priv *priv = netdev_priv(netdev);
3672 mlx5e_build_nic_netdev_priv(mdev, netdev, profile, ppriv);
3673 mlx5e_build_nic_netdev(netdev);
3674 mlx5e_vxlan_init(priv);
3677 static void mlx5e_nic_cleanup(struct mlx5e_priv *priv)
3679 struct mlx5_core_dev *mdev = priv->mdev;
3680 struct mlx5_eswitch *esw = mdev->priv.eswitch;
3682 mlx5e_vxlan_cleanup(priv);
3684 if (MLX5_CAP_GEN(mdev, vport_group_manager))
3685 mlx5_eswitch_unregister_vport_rep(esw, 0);
3688 bpf_prog_put(priv->xdp_prog);
3691 static int mlx5e_init_nic_rx(struct mlx5e_priv *priv)
3693 struct mlx5_core_dev *mdev = priv->mdev;
3697 err = mlx5e_create_indirect_rqts(priv);
3699 mlx5_core_warn(mdev, "create indirect rqts failed, %d\n", err);
3703 err = mlx5e_create_direct_rqts(priv);
3705 mlx5_core_warn(mdev, "create direct rqts failed, %d\n", err);
3706 goto err_destroy_indirect_rqts;
3709 err = mlx5e_create_indirect_tirs(priv);
3711 mlx5_core_warn(mdev, "create indirect tirs failed, %d\n", err);
3712 goto err_destroy_direct_rqts;
3715 err = mlx5e_create_direct_tirs(priv);
3717 mlx5_core_warn(mdev, "create direct tirs failed, %d\n", err);
3718 goto err_destroy_indirect_tirs;
3721 err = mlx5e_create_flow_steering(priv);
3723 mlx5_core_warn(mdev, "create flow steering failed, %d\n", err);
3724 goto err_destroy_direct_tirs;
3727 err = mlx5e_tc_init(priv);
3729 goto err_destroy_flow_steering;
3733 err_destroy_flow_steering:
3734 mlx5e_destroy_flow_steering(priv);
3735 err_destroy_direct_tirs:
3736 mlx5e_destroy_direct_tirs(priv);
3737 err_destroy_indirect_tirs:
3738 mlx5e_destroy_indirect_tirs(priv);
3739 err_destroy_direct_rqts:
3740 for (i = 0; i < priv->profile->max_nch(mdev); i++)
3741 mlx5e_destroy_rqt(priv, &priv->direct_tir[i].rqt);
3742 err_destroy_indirect_rqts:
3743 mlx5e_destroy_rqt(priv, &priv->indir_rqt);
3747 static void mlx5e_cleanup_nic_rx(struct mlx5e_priv *priv)
3751 mlx5e_tc_cleanup(priv);
3752 mlx5e_destroy_flow_steering(priv);
3753 mlx5e_destroy_direct_tirs(priv);
3754 mlx5e_destroy_indirect_tirs(priv);
3755 for (i = 0; i < priv->profile->max_nch(priv->mdev); i++)
3756 mlx5e_destroy_rqt(priv, &priv->direct_tir[i].rqt);
3757 mlx5e_destroy_rqt(priv, &priv->indir_rqt);
3760 static int mlx5e_init_nic_tx(struct mlx5e_priv *priv)
3764 err = mlx5e_create_tises(priv);
3766 mlx5_core_warn(priv->mdev, "create tises failed, %d\n", err);
3770 #ifdef CONFIG_MLX5_CORE_EN_DCB
3771 mlx5e_dcbnl_initialize(priv);
3776 static void mlx5e_nic_enable(struct mlx5e_priv *priv)
3778 struct net_device *netdev = priv->netdev;
3779 struct mlx5_core_dev *mdev = priv->mdev;
3780 struct mlx5_eswitch *esw = mdev->priv.eswitch;
3781 struct mlx5_eswitch_rep rep;
3783 mlx5_lag_add(mdev, netdev);
3785 if (mlx5e_vxlan_allowed(mdev)) {
3787 udp_tunnel_get_rx_info(netdev);
3791 mlx5e_enable_async_events(priv);
3792 queue_work(priv->wq, &priv->set_rx_mode_work);
3794 if (MLX5_CAP_GEN(mdev, vport_group_manager)) {
3795 mlx5_query_nic_vport_mac_address(mdev, 0, rep.hw_id);
3796 rep.load = mlx5e_nic_rep_load;
3797 rep.unload = mlx5e_nic_rep_unload;
3798 rep.vport = FDB_UPLINK_VPORT;
3799 rep.priv_data = priv;
3800 mlx5_eswitch_register_vport_rep(esw, 0, &rep);
3804 static void mlx5e_nic_disable(struct mlx5e_priv *priv)
3806 queue_work(priv->wq, &priv->set_rx_mode_work);
3807 mlx5e_disable_async_events(priv);
3808 mlx5_lag_remove(priv->mdev);
3811 static const struct mlx5e_profile mlx5e_nic_profile = {
3812 .init = mlx5e_nic_init,
3813 .cleanup = mlx5e_nic_cleanup,
3814 .init_rx = mlx5e_init_nic_rx,
3815 .cleanup_rx = mlx5e_cleanup_nic_rx,
3816 .init_tx = mlx5e_init_nic_tx,
3817 .cleanup_tx = mlx5e_cleanup_nic_tx,
3818 .enable = mlx5e_nic_enable,
3819 .disable = mlx5e_nic_disable,
3820 .update_stats = mlx5e_update_stats,
3821 .max_nch = mlx5e_get_max_num_channels,
3822 .max_tc = MLX5E_MAX_NUM_TC,
3825 struct net_device *mlx5e_create_netdev(struct mlx5_core_dev *mdev,
3826 const struct mlx5e_profile *profile,
3829 int nch = profile->max_nch(mdev);
3830 struct net_device *netdev;
3831 struct mlx5e_priv *priv;
3833 netdev = alloc_etherdev_mqs(sizeof(struct mlx5e_priv),
3834 nch * profile->max_tc,
3837 mlx5_core_err(mdev, "alloc_etherdev_mqs() failed\n");
3841 profile->init(mdev, netdev, profile, ppriv);
3843 netif_carrier_off(netdev);
3845 priv = netdev_priv(netdev);
3847 priv->wq = create_singlethread_workqueue("mlx5e");
3849 goto err_cleanup_nic;
3854 profile->cleanup(priv);
3855 free_netdev(netdev);
3860 int mlx5e_attach_netdev(struct mlx5_core_dev *mdev, struct net_device *netdev)
3862 const struct mlx5e_profile *profile;
3863 struct mlx5e_priv *priv;
3867 priv = netdev_priv(netdev);
3868 profile = priv->profile;
3869 clear_bit(MLX5E_STATE_DESTROYING, &priv->state);
3871 err = mlx5e_create_umr_mkey(priv);
3873 mlx5_core_err(mdev, "create umr mkey failed, %d\n", err);
3877 err = profile->init_tx(priv);
3879 goto err_destroy_umr_mkey;
3881 err = mlx5e_open_drop_rq(priv);
3883 mlx5_core_err(mdev, "open drop rq failed, %d\n", err);
3884 goto err_cleanup_tx;
3887 err = profile->init_rx(priv);
3889 goto err_close_drop_rq;
3891 mlx5e_create_q_counter(priv);
3893 mlx5e_init_l2_addr(priv);
3895 /* MTU range: 68 - hw-specific max */
3896 netdev->min_mtu = ETH_MIN_MTU;
3897 mlx5_query_port_max_mtu(priv->mdev, &max_mtu, 1);
3898 netdev->max_mtu = MLX5E_HW2SW_MTU(max_mtu);
3900 mlx5e_set_dev_port_mtu(netdev);
3902 if (profile->enable)
3903 profile->enable(priv);
3906 if (netif_running(netdev))
3908 netif_device_attach(netdev);
3914 mlx5e_close_drop_rq(priv);
3917 profile->cleanup_tx(priv);
3919 err_destroy_umr_mkey:
3920 mlx5_core_destroy_mkey(mdev, &priv->umr_mkey);
3926 static void mlx5e_register_vport_rep(struct mlx5_core_dev *mdev)
3928 struct mlx5_eswitch *esw = mdev->priv.eswitch;
3929 int total_vfs = MLX5_TOTAL_VPORTS(mdev);
3933 if (!MLX5_CAP_GEN(mdev, vport_group_manager))
3936 mlx5_query_nic_vport_mac_address(mdev, 0, mac);
3938 for (vport = 1; vport < total_vfs; vport++) {
3939 struct mlx5_eswitch_rep rep;
3941 rep.load = mlx5e_vport_rep_load;
3942 rep.unload = mlx5e_vport_rep_unload;
3944 ether_addr_copy(rep.hw_id, mac);
3945 mlx5_eswitch_register_vport_rep(esw, vport, &rep);
3949 void mlx5e_detach_netdev(struct mlx5_core_dev *mdev, struct net_device *netdev)
3951 struct mlx5e_priv *priv = netdev_priv(netdev);
3952 const struct mlx5e_profile *profile = priv->profile;
3954 set_bit(MLX5E_STATE_DESTROYING, &priv->state);
3955 if (profile->disable)
3956 profile->disable(priv);
3958 flush_workqueue(priv->wq);
3961 if (netif_running(netdev))
3962 mlx5e_close(netdev);
3963 netif_device_detach(netdev);
3966 mlx5e_destroy_q_counter(priv);
3967 profile->cleanup_rx(priv);
3968 mlx5e_close_drop_rq(priv);
3969 profile->cleanup_tx(priv);
3970 mlx5_core_destroy_mkey(priv->mdev, &priv->umr_mkey);
3971 cancel_delayed_work_sync(&priv->update_stats_work);
3974 /* mlx5e_attach and mlx5e_detach scope should be only creating/destroying
3975 * hardware contexts and to connect it to the current netdev.
3977 static int mlx5e_attach(struct mlx5_core_dev *mdev, void *vpriv)
3979 struct mlx5e_priv *priv = vpriv;
3980 struct net_device *netdev = priv->netdev;
3983 if (netif_device_present(netdev))
3986 err = mlx5e_create_mdev_resources(mdev);
3990 err = mlx5e_attach_netdev(mdev, netdev);
3992 mlx5e_destroy_mdev_resources(mdev);
3999 static void mlx5e_detach(struct mlx5_core_dev *mdev, void *vpriv)
4001 struct mlx5e_priv *priv = vpriv;
4002 struct net_device *netdev = priv->netdev;
4004 if (!netif_device_present(netdev))
4007 mlx5e_detach_netdev(mdev, netdev);
4008 mlx5e_destroy_mdev_resources(mdev);
4011 static void *mlx5e_add(struct mlx5_core_dev *mdev)
4013 struct mlx5_eswitch *esw = mdev->priv.eswitch;
4014 int total_vfs = MLX5_TOTAL_VPORTS(mdev);
4019 struct net_device *netdev;
4021 err = mlx5e_check_required_hca_cap(mdev);
4025 mlx5e_register_vport_rep(mdev);
4027 if (MLX5_CAP_GEN(mdev, vport_group_manager))
4028 ppriv = &esw->offloads.vport_reps[0];
4030 netdev = mlx5e_create_netdev(mdev, &mlx5e_nic_profile, ppriv);
4032 mlx5_core_err(mdev, "mlx5e_create_netdev failed\n");
4033 goto err_unregister_reps;
4036 priv = netdev_priv(netdev);
4038 err = mlx5e_attach(mdev, priv);
4040 mlx5_core_err(mdev, "mlx5e_attach failed, %d\n", err);
4041 goto err_destroy_netdev;
4044 err = register_netdev(netdev);
4046 mlx5_core_err(mdev, "register_netdev failed, %d\n", err);
4053 mlx5e_detach(mdev, priv);
4056 mlx5e_destroy_netdev(mdev, priv);
4058 err_unregister_reps:
4059 for (vport = 1; vport < total_vfs; vport++)
4060 mlx5_eswitch_unregister_vport_rep(esw, vport);
4065 void mlx5e_destroy_netdev(struct mlx5_core_dev *mdev, struct mlx5e_priv *priv)
4067 const struct mlx5e_profile *profile = priv->profile;
4068 struct net_device *netdev = priv->netdev;
4070 destroy_workqueue(priv->wq);
4071 if (profile->cleanup)
4072 profile->cleanup(priv);
4073 free_netdev(netdev);
4076 static void mlx5e_remove(struct mlx5_core_dev *mdev, void *vpriv)
4078 struct mlx5_eswitch *esw = mdev->priv.eswitch;
4079 int total_vfs = MLX5_TOTAL_VPORTS(mdev);
4080 struct mlx5e_priv *priv = vpriv;
4083 for (vport = 1; vport < total_vfs; vport++)
4084 mlx5_eswitch_unregister_vport_rep(esw, vport);
4086 unregister_netdev(priv->netdev);
4087 mlx5e_detach(mdev, vpriv);
4088 mlx5e_destroy_netdev(mdev, priv);
4091 static void *mlx5e_get_netdev(void *vpriv)
4093 struct mlx5e_priv *priv = vpriv;
4095 return priv->netdev;
4098 static struct mlx5_interface mlx5e_interface = {
4100 .remove = mlx5e_remove,
4101 .attach = mlx5e_attach,
4102 .detach = mlx5e_detach,
4103 .event = mlx5e_async_event,
4104 .protocol = MLX5_INTERFACE_PROTOCOL_ETH,
4105 .get_dev = mlx5e_get_netdev,
4108 void mlx5e_init(void)
4110 mlx5e_build_ptys2ethtool_map();
4111 mlx5_register_interface(&mlx5e_interface);
4114 void mlx5e_cleanup(void)
4116 mlx5_unregister_interface(&mlx5e_interface);