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>
42 #include "en_accel/ipsec.h"
43 #include "en_accel/ipsec_rxtx.h"
44 #include "accel/ipsec.h"
47 struct mlx5e_rq_param {
48 u32 rqc[MLX5_ST_SZ_DW(rqc)];
49 struct mlx5_wq_param wq;
52 struct mlx5e_sq_param {
53 u32 sqc[MLX5_ST_SZ_DW(sqc)];
54 struct mlx5_wq_param wq;
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 void mlx5e_init_rq_type_params(struct mlx5_core_dev *mdev,
82 struct mlx5e_params *params, u8 rq_type)
84 params->rq_wq_type = rq_type;
85 params->lro_wqe_sz = MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ;
86 switch (params->rq_wq_type) {
87 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
88 params->log_rq_size = is_kdump_kernel() ?
89 MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE_MPW :
90 MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE_MPW;
91 params->mpwqe_log_stride_sz = MLX5E_MPWQE_STRIDE_SZ(mdev,
92 MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS));
93 params->mpwqe_log_num_strides = MLX5_MPWRQ_LOG_WQE_SZ -
94 params->mpwqe_log_stride_sz;
96 default: /* MLX5_WQ_TYPE_LINKED_LIST */
97 params->log_rq_size = is_kdump_kernel() ?
98 MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE :
99 MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
100 params->rq_headroom = params->xdp_prog ?
101 XDP_PACKET_HEADROOM : MLX5_RX_HEADROOM;
102 params->rq_headroom += NET_IP_ALIGN;
104 /* Extra room needed for build_skb */
105 params->lro_wqe_sz -= params->rq_headroom +
106 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
109 mlx5_core_info(mdev, "MLX5E: StrdRq(%d) RqSz(%ld) StrdSz(%ld) RxCqeCmprss(%d)\n",
110 params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ,
111 BIT(params->log_rq_size),
112 BIT(params->mpwqe_log_stride_sz),
113 MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS));
116 static void mlx5e_set_rq_params(struct mlx5_core_dev *mdev,
117 struct mlx5e_params *params)
119 u8 rq_type = mlx5e_check_fragmented_striding_rq_cap(mdev) &&
120 !params->xdp_prog && !MLX5_IPSEC_DEV(mdev) ?
121 MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ :
122 MLX5_WQ_TYPE_LINKED_LIST;
123 mlx5e_init_rq_type_params(mdev, params, rq_type);
126 static void mlx5e_update_carrier(struct mlx5e_priv *priv)
128 struct mlx5_core_dev *mdev = priv->mdev;
131 port_state = mlx5_query_vport_state(mdev,
132 MLX5_QUERY_VPORT_STATE_IN_OP_MOD_VNIC_VPORT,
135 if (port_state == VPORT_STATE_UP) {
136 netdev_info(priv->netdev, "Link up\n");
137 netif_carrier_on(priv->netdev);
139 netdev_info(priv->netdev, "Link down\n");
140 netif_carrier_off(priv->netdev);
144 static void mlx5e_update_carrier_work(struct work_struct *work)
146 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
147 update_carrier_work);
149 mutex_lock(&priv->state_lock);
150 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
151 if (priv->profile->update_carrier)
152 priv->profile->update_carrier(priv);
153 mutex_unlock(&priv->state_lock);
156 static void mlx5e_tx_timeout_work(struct work_struct *work)
158 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
163 mutex_lock(&priv->state_lock);
164 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
166 mlx5e_close_locked(priv->netdev);
167 err = mlx5e_open_locked(priv->netdev);
169 netdev_err(priv->netdev, "mlx5e_open_locked failed recovering from a tx_timeout, err(%d).\n",
172 mutex_unlock(&priv->state_lock);
176 static void mlx5e_update_sw_counters(struct mlx5e_priv *priv)
178 struct mlx5e_sw_stats temp, *s = &temp;
179 struct mlx5e_rq_stats *rq_stats;
180 struct mlx5e_sq_stats *sq_stats;
183 memset(s, 0, sizeof(*s));
184 for (i = 0; i < priv->channels.num; i++) {
185 struct mlx5e_channel *c = priv->channels.c[i];
187 rq_stats = &c->rq.stats;
189 s->rx_packets += rq_stats->packets;
190 s->rx_bytes += rq_stats->bytes;
191 s->rx_lro_packets += rq_stats->lro_packets;
192 s->rx_lro_bytes += rq_stats->lro_bytes;
193 s->rx_removed_vlan_packets += rq_stats->removed_vlan_packets;
194 s->rx_csum_none += rq_stats->csum_none;
195 s->rx_csum_complete += rq_stats->csum_complete;
196 s->rx_csum_unnecessary += rq_stats->csum_unnecessary;
197 s->rx_csum_unnecessary_inner += rq_stats->csum_unnecessary_inner;
198 s->rx_xdp_drop += rq_stats->xdp_drop;
199 s->rx_xdp_tx += rq_stats->xdp_tx;
200 s->rx_xdp_tx_full += rq_stats->xdp_tx_full;
201 s->rx_wqe_err += rq_stats->wqe_err;
202 s->rx_mpwqe_filler += rq_stats->mpwqe_filler;
203 s->rx_buff_alloc_err += rq_stats->buff_alloc_err;
204 s->rx_cqe_compress_blks += rq_stats->cqe_compress_blks;
205 s->rx_cqe_compress_pkts += rq_stats->cqe_compress_pkts;
206 s->rx_page_reuse += rq_stats->page_reuse;
207 s->rx_cache_reuse += rq_stats->cache_reuse;
208 s->rx_cache_full += rq_stats->cache_full;
209 s->rx_cache_empty += rq_stats->cache_empty;
210 s->rx_cache_busy += rq_stats->cache_busy;
211 s->rx_cache_waive += rq_stats->cache_waive;
213 for (j = 0; j < priv->channels.params.num_tc; j++) {
214 sq_stats = &c->sq[j].stats;
216 s->tx_packets += sq_stats->packets;
217 s->tx_bytes += sq_stats->bytes;
218 s->tx_tso_packets += sq_stats->tso_packets;
219 s->tx_tso_bytes += sq_stats->tso_bytes;
220 s->tx_tso_inner_packets += sq_stats->tso_inner_packets;
221 s->tx_tso_inner_bytes += sq_stats->tso_inner_bytes;
222 s->tx_added_vlan_packets += sq_stats->added_vlan_packets;
223 s->tx_queue_stopped += sq_stats->stopped;
224 s->tx_queue_wake += sq_stats->wake;
225 s->tx_queue_dropped += sq_stats->dropped;
226 s->tx_xmit_more += sq_stats->xmit_more;
227 s->tx_csum_partial_inner += sq_stats->csum_partial_inner;
228 s->tx_csum_none += sq_stats->csum_none;
229 s->tx_csum_partial += sq_stats->csum_partial;
233 s->link_down_events_phy = MLX5_GET(ppcnt_reg,
234 priv->stats.pport.phy_counters,
235 counter_set.phys_layer_cntrs.link_down_events);
236 memcpy(&priv->stats.sw, s, sizeof(*s));
239 static void mlx5e_update_vport_counters(struct mlx5e_priv *priv)
241 int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
242 u32 *out = (u32 *)priv->stats.vport.query_vport_out;
243 u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)] = {0};
244 struct mlx5_core_dev *mdev = priv->mdev;
246 MLX5_SET(query_vport_counter_in, in, opcode,
247 MLX5_CMD_OP_QUERY_VPORT_COUNTER);
248 MLX5_SET(query_vport_counter_in, in, op_mod, 0);
249 MLX5_SET(query_vport_counter_in, in, other_vport, 0);
251 mlx5_cmd_exec(mdev, in, sizeof(in), out, outlen);
254 static void mlx5e_update_pport_counters(struct mlx5e_priv *priv, bool full)
256 struct mlx5e_pport_stats *pstats = &priv->stats.pport;
257 struct mlx5_core_dev *mdev = priv->mdev;
258 u32 in[MLX5_ST_SZ_DW(ppcnt_reg)] = {0};
259 int sz = MLX5_ST_SZ_BYTES(ppcnt_reg);
263 MLX5_SET(ppcnt_reg, in, local_port, 1);
265 out = pstats->IEEE_802_3_counters;
266 MLX5_SET(ppcnt_reg, in, grp, MLX5_IEEE_802_3_COUNTERS_GROUP);
267 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
272 out = pstats->RFC_2863_counters;
273 MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2863_COUNTERS_GROUP);
274 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
276 out = pstats->RFC_2819_counters;
277 MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2819_COUNTERS_GROUP);
278 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
280 out = pstats->phy_counters;
281 MLX5_SET(ppcnt_reg, in, grp, MLX5_PHYSICAL_LAYER_COUNTERS_GROUP);
282 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
284 if (MLX5_CAP_PCAM_FEATURE(mdev, ppcnt_statistical_group)) {
285 out = pstats->phy_statistical_counters;
286 MLX5_SET(ppcnt_reg, in, grp, MLX5_PHYSICAL_LAYER_STATISTICAL_GROUP);
287 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
290 if (MLX5_CAP_PCAM_FEATURE(mdev, rx_buffer_fullness_counters)) {
291 out = pstats->eth_ext_counters;
292 MLX5_SET(ppcnt_reg, in, grp, MLX5_ETHERNET_EXTENDED_COUNTERS_GROUP);
293 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0);
296 MLX5_SET(ppcnt_reg, in, grp, MLX5_PER_PRIORITY_COUNTERS_GROUP);
297 for (prio = 0; prio < NUM_PPORT_PRIO; prio++) {
298 out = pstats->per_prio_counters[prio];
299 MLX5_SET(ppcnt_reg, in, prio_tc, prio);
300 mlx5_core_access_reg(mdev, in, sz, out, sz,
301 MLX5_REG_PPCNT, 0, 0);
305 static void mlx5e_update_q_counter(struct mlx5e_priv *priv)
307 struct mlx5e_qcounter_stats *qcnt = &priv->stats.qcnt;
308 u32 out[MLX5_ST_SZ_DW(query_q_counter_out)];
311 if (!priv->q_counter)
314 err = mlx5_core_query_q_counter(priv->mdev, priv->q_counter, 0, out, sizeof(out));
318 qcnt->rx_out_of_buffer = MLX5_GET(query_q_counter_out, out, out_of_buffer);
321 static void mlx5e_update_pcie_counters(struct mlx5e_priv *priv)
323 struct mlx5e_pcie_stats *pcie_stats = &priv->stats.pcie;
324 struct mlx5_core_dev *mdev = priv->mdev;
325 u32 in[MLX5_ST_SZ_DW(mpcnt_reg)] = {0};
326 int sz = MLX5_ST_SZ_BYTES(mpcnt_reg);
329 if (!MLX5_CAP_MCAM_FEATURE(mdev, pcie_performance_group))
332 out = pcie_stats->pcie_perf_counters;
333 MLX5_SET(mpcnt_reg, in, grp, MLX5_PCIE_PERFORMANCE_COUNTERS_GROUP);
334 mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_MPCNT, 0, 0);
337 void mlx5e_update_stats(struct mlx5e_priv *priv, bool full)
340 mlx5e_update_pcie_counters(priv);
341 mlx5e_ipsec_update_stats(priv);
343 mlx5e_update_pport_counters(priv, full);
344 mlx5e_update_vport_counters(priv);
345 mlx5e_update_q_counter(priv);
346 mlx5e_update_sw_counters(priv);
349 static void mlx5e_update_ndo_stats(struct mlx5e_priv *priv)
351 mlx5e_update_stats(priv, false);
354 void mlx5e_update_stats_work(struct work_struct *work)
356 struct delayed_work *dwork = to_delayed_work(work);
357 struct mlx5e_priv *priv = container_of(dwork, struct mlx5e_priv,
359 mutex_lock(&priv->state_lock);
360 if (test_bit(MLX5E_STATE_OPENED, &priv->state)) {
361 priv->profile->update_stats(priv);
362 queue_delayed_work(priv->wq, dwork,
363 msecs_to_jiffies(MLX5E_UPDATE_STATS_INTERVAL));
365 mutex_unlock(&priv->state_lock);
368 static void mlx5e_async_event(struct mlx5_core_dev *mdev, void *vpriv,
369 enum mlx5_dev_event event, unsigned long param)
371 struct mlx5e_priv *priv = vpriv;
373 if (!test_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLED, &priv->state))
377 case MLX5_DEV_EVENT_PORT_UP:
378 case MLX5_DEV_EVENT_PORT_DOWN:
379 queue_work(priv->wq, &priv->update_carrier_work);
386 static void mlx5e_enable_async_events(struct mlx5e_priv *priv)
388 set_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLED, &priv->state);
391 static void mlx5e_disable_async_events(struct mlx5e_priv *priv)
393 clear_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLED, &priv->state);
394 synchronize_irq(pci_irq_vector(priv->mdev->pdev, MLX5_EQ_VEC_ASYNC));
397 static inline int mlx5e_get_wqe_mtt_sz(void)
399 /* UMR copies MTTs in units of MLX5_UMR_MTT_ALIGNMENT bytes.
400 * To avoid copying garbage after the mtt array, we allocate
403 return ALIGN(MLX5_MPWRQ_PAGES_PER_WQE * sizeof(__be64),
404 MLX5_UMR_MTT_ALIGNMENT);
407 static inline void mlx5e_build_umr_wqe(struct mlx5e_rq *rq,
408 struct mlx5e_icosq *sq,
409 struct mlx5e_umr_wqe *wqe,
412 struct mlx5_wqe_ctrl_seg *cseg = &wqe->ctrl;
413 struct mlx5_wqe_umr_ctrl_seg *ucseg = &wqe->uctrl;
414 struct mlx5_wqe_data_seg *dseg = &wqe->data;
415 struct mlx5e_mpw_info *wi = &rq->mpwqe.info[ix];
416 u8 ds_cnt = DIV_ROUND_UP(sizeof(*wqe), MLX5_SEND_WQE_DS);
417 u32 umr_wqe_mtt_offset = mlx5e_get_wqe_mtt_offset(rq, ix);
419 cseg->qpn_ds = cpu_to_be32((sq->sqn << MLX5_WQE_CTRL_QPN_SHIFT) |
421 cseg->fm_ce_se = MLX5_WQE_CTRL_CQ_UPDATE;
422 cseg->imm = rq->mkey_be;
424 ucseg->flags = MLX5_UMR_TRANSLATION_OFFSET_EN;
425 ucseg->xlt_octowords =
426 cpu_to_be16(MLX5_MTT_OCTW(MLX5_MPWRQ_PAGES_PER_WQE));
427 ucseg->bsf_octowords =
428 cpu_to_be16(MLX5_MTT_OCTW(umr_wqe_mtt_offset));
429 ucseg->mkey_mask = cpu_to_be64(MLX5_MKEY_MASK_FREE);
431 dseg->lkey = sq->mkey_be;
432 dseg->addr = cpu_to_be64(wi->umr.mtt_addr);
435 static int mlx5e_rq_alloc_mpwqe_info(struct mlx5e_rq *rq,
436 struct mlx5e_channel *c)
438 int wq_sz = mlx5_wq_ll_get_size(&rq->wq);
439 int mtt_sz = mlx5e_get_wqe_mtt_sz();
440 int mtt_alloc = mtt_sz + MLX5_UMR_ALIGN - 1;
443 rq->mpwqe.info = kzalloc_node(wq_sz * sizeof(*rq->mpwqe.info),
444 GFP_KERNEL, cpu_to_node(c->cpu));
448 /* We allocate more than mtt_sz as we will align the pointer */
449 rq->mpwqe.mtt_no_align = kzalloc_node(mtt_alloc * wq_sz, GFP_KERNEL,
450 cpu_to_node(c->cpu));
451 if (unlikely(!rq->mpwqe.mtt_no_align))
452 goto err_free_wqe_info;
454 for (i = 0; i < wq_sz; i++) {
455 struct mlx5e_mpw_info *wi = &rq->mpwqe.info[i];
457 wi->umr.mtt = PTR_ALIGN(rq->mpwqe.mtt_no_align + i * mtt_alloc,
459 wi->umr.mtt_addr = dma_map_single(c->pdev, wi->umr.mtt, mtt_sz,
461 if (unlikely(dma_mapping_error(c->pdev, wi->umr.mtt_addr)))
464 mlx5e_build_umr_wqe(rq, &c->icosq, &wi->umr.wqe, i);
471 struct mlx5e_mpw_info *wi = &rq->mpwqe.info[i];
473 dma_unmap_single(c->pdev, wi->umr.mtt_addr, mtt_sz,
476 kfree(rq->mpwqe.mtt_no_align);
478 kfree(rq->mpwqe.info);
484 static void mlx5e_rq_free_mpwqe_info(struct mlx5e_rq *rq)
486 int wq_sz = mlx5_wq_ll_get_size(&rq->wq);
487 int mtt_sz = mlx5e_get_wqe_mtt_sz();
490 for (i = 0; i < wq_sz; i++) {
491 struct mlx5e_mpw_info *wi = &rq->mpwqe.info[i];
493 dma_unmap_single(rq->pdev, wi->umr.mtt_addr, mtt_sz,
496 kfree(rq->mpwqe.mtt_no_align);
497 kfree(rq->mpwqe.info);
500 static int mlx5e_create_umr_mkey(struct mlx5_core_dev *mdev,
501 u64 npages, u8 page_shift,
502 struct mlx5_core_mkey *umr_mkey)
504 int inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
509 if (!MLX5E_VALID_NUM_MTTS(npages))
512 in = kvzalloc(inlen, GFP_KERNEL);
516 mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
518 MLX5_SET(mkc, mkc, free, 1);
519 MLX5_SET(mkc, mkc, umr_en, 1);
520 MLX5_SET(mkc, mkc, lw, 1);
521 MLX5_SET(mkc, mkc, lr, 1);
522 MLX5_SET(mkc, mkc, access_mode, MLX5_MKC_ACCESS_MODE_MTT);
524 MLX5_SET(mkc, mkc, qpn, 0xffffff);
525 MLX5_SET(mkc, mkc, pd, mdev->mlx5e_res.pdn);
526 MLX5_SET64(mkc, mkc, len, npages << page_shift);
527 MLX5_SET(mkc, mkc, translations_octword_size,
528 MLX5_MTT_OCTW(npages));
529 MLX5_SET(mkc, mkc, log_page_size, page_shift);
531 err = mlx5_core_create_mkey(mdev, umr_mkey, in, inlen);
537 static int mlx5e_create_rq_umr_mkey(struct mlx5_core_dev *mdev, struct mlx5e_rq *rq)
539 u64 num_mtts = MLX5E_REQUIRED_MTTS(mlx5_wq_ll_get_size(&rq->wq));
541 return mlx5e_create_umr_mkey(mdev, num_mtts, PAGE_SHIFT, &rq->umr_mkey);
544 static int mlx5e_alloc_rq(struct mlx5e_channel *c,
545 struct mlx5e_params *params,
546 struct mlx5e_rq_param *rqp,
549 struct mlx5_core_dev *mdev = c->mdev;
550 void *rqc = rqp->rqc;
551 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
558 rqp->wq.db_numa_node = cpu_to_node(c->cpu);
560 err = mlx5_wq_ll_create(mdev, &rqp->wq, rqc_wq, &rq->wq,
565 rq->wq.db = &rq->wq.db[MLX5_RCV_DBR];
567 wq_sz = mlx5_wq_ll_get_size(&rq->wq);
569 rq->wq_type = params->rq_wq_type;
571 rq->netdev = c->netdev;
572 rq->tstamp = c->tstamp;
573 rq->clock = &mdev->clock;
578 rq->xdp_prog = params->xdp_prog ? bpf_prog_inc(params->xdp_prog) : NULL;
579 if (IS_ERR(rq->xdp_prog)) {
580 err = PTR_ERR(rq->xdp_prog);
582 goto err_rq_wq_destroy;
585 rq->buff.map_dir = rq->xdp_prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
586 rq->buff.headroom = params->rq_headroom;
588 switch (rq->wq_type) {
589 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
591 rq->post_wqes = mlx5e_post_rx_mpwqes;
592 rq->dealloc_wqe = mlx5e_dealloc_rx_mpwqe;
594 rq->handle_rx_cqe = c->priv->profile->rx_handlers.handle_rx_cqe_mpwqe;
595 #ifdef CONFIG_MLX5_EN_IPSEC
596 if (MLX5_IPSEC_DEV(mdev)) {
598 netdev_err(c->netdev, "MPWQE RQ with IPSec offload not supported\n");
599 goto err_rq_wq_destroy;
602 if (!rq->handle_rx_cqe) {
604 netdev_err(c->netdev, "RX handler of MPWQE RQ is not set, err %d\n", err);
605 goto err_rq_wq_destroy;
608 rq->mpwqe.log_stride_sz = params->mpwqe_log_stride_sz;
609 rq->mpwqe.num_strides = BIT(params->mpwqe_log_num_strides);
611 byte_count = rq->mpwqe.num_strides << rq->mpwqe.log_stride_sz;
613 err = mlx5e_create_rq_umr_mkey(mdev, rq);
615 goto err_rq_wq_destroy;
616 rq->mkey_be = cpu_to_be32(rq->umr_mkey.key);
618 err = mlx5e_rq_alloc_mpwqe_info(rq, c);
620 goto err_destroy_umr_mkey;
622 default: /* MLX5_WQ_TYPE_LINKED_LIST */
624 kzalloc_node(wq_sz * sizeof(*rq->wqe.frag_info),
625 GFP_KERNEL, cpu_to_node(c->cpu));
626 if (!rq->wqe.frag_info) {
628 goto err_rq_wq_destroy;
630 rq->post_wqes = mlx5e_post_rx_wqes;
631 rq->dealloc_wqe = mlx5e_dealloc_rx_wqe;
633 #ifdef CONFIG_MLX5_EN_IPSEC
635 rq->handle_rx_cqe = mlx5e_ipsec_handle_rx_cqe;
638 rq->handle_rx_cqe = c->priv->profile->rx_handlers.handle_rx_cqe;
639 if (!rq->handle_rx_cqe) {
640 kfree(rq->wqe.frag_info);
642 netdev_err(c->netdev, "RX handler of RQ is not set, err %d\n", err);
643 goto err_rq_wq_destroy;
646 byte_count = params->lro_en ?
648 MLX5E_SW2HW_MTU(c->priv, c->netdev->mtu);
649 #ifdef CONFIG_MLX5_EN_IPSEC
650 if (MLX5_IPSEC_DEV(mdev))
651 byte_count += MLX5E_METADATA_ETHER_LEN;
653 rq->wqe.page_reuse = !params->xdp_prog && !params->lro_en;
655 /* calc the required page order */
656 rq->wqe.frag_sz = MLX5_SKB_FRAG_SZ(rq->buff.headroom + byte_count);
657 npages = DIV_ROUND_UP(rq->wqe.frag_sz, PAGE_SIZE);
658 rq->buff.page_order = order_base_2(npages);
660 byte_count |= MLX5_HW_START_PADDING;
661 rq->mkey_be = c->mkey_be;
664 for (i = 0; i < wq_sz; i++) {
665 struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
667 if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
668 u64 dma_offset = (u64)mlx5e_get_wqe_mtt_offset(rq, i) << PAGE_SHIFT;
670 wqe->data.addr = cpu_to_be64(dma_offset);
673 wqe->data.byte_count = cpu_to_be32(byte_count);
674 wqe->data.lkey = rq->mkey_be;
677 INIT_WORK(&rq->am.work, mlx5e_rx_am_work);
678 rq->am.mode = params->rx_cq_moderation.cq_period_mode;
679 rq->page_cache.head = 0;
680 rq->page_cache.tail = 0;
684 err_destroy_umr_mkey:
685 mlx5_core_destroy_mkey(mdev, &rq->umr_mkey);
689 bpf_prog_put(rq->xdp_prog);
690 mlx5_wq_destroy(&rq->wq_ctrl);
695 static void mlx5e_free_rq(struct mlx5e_rq *rq)
700 bpf_prog_put(rq->xdp_prog);
702 switch (rq->wq_type) {
703 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
704 mlx5e_rq_free_mpwqe_info(rq);
705 mlx5_core_destroy_mkey(rq->mdev, &rq->umr_mkey);
707 default: /* MLX5_WQ_TYPE_LINKED_LIST */
708 kfree(rq->wqe.frag_info);
711 for (i = rq->page_cache.head; i != rq->page_cache.tail;
712 i = (i + 1) & (MLX5E_CACHE_SIZE - 1)) {
713 struct mlx5e_dma_info *dma_info = &rq->page_cache.page_cache[i];
715 mlx5e_page_release(rq, dma_info, false);
717 mlx5_wq_destroy(&rq->wq_ctrl);
720 static int mlx5e_create_rq(struct mlx5e_rq *rq,
721 struct mlx5e_rq_param *param)
723 struct mlx5_core_dev *mdev = rq->mdev;
731 inlen = MLX5_ST_SZ_BYTES(create_rq_in) +
732 sizeof(u64) * rq->wq_ctrl.buf.npages;
733 in = kvzalloc(inlen, GFP_KERNEL);
737 rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
738 wq = MLX5_ADDR_OF(rqc, rqc, wq);
740 memcpy(rqc, param->rqc, sizeof(param->rqc));
742 MLX5_SET(rqc, rqc, cqn, rq->cq.mcq.cqn);
743 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST);
744 MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift -
745 MLX5_ADAPTER_PAGE_SHIFT);
746 MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma);
748 mlx5_fill_page_array(&rq->wq_ctrl.buf,
749 (__be64 *)MLX5_ADDR_OF(wq, wq, pas));
751 err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn);
758 static int mlx5e_modify_rq_state(struct mlx5e_rq *rq, int curr_state,
761 struct mlx5e_channel *c = rq->channel;
762 struct mlx5_core_dev *mdev = c->mdev;
769 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
770 in = kvzalloc(inlen, GFP_KERNEL);
774 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
776 MLX5_SET(modify_rq_in, in, rq_state, curr_state);
777 MLX5_SET(rqc, rqc, state, next_state);
779 err = mlx5_core_modify_rq(mdev, rq->rqn, in, inlen);
786 static int mlx5e_modify_rq_scatter_fcs(struct mlx5e_rq *rq, bool enable)
788 struct mlx5e_channel *c = rq->channel;
789 struct mlx5e_priv *priv = c->priv;
790 struct mlx5_core_dev *mdev = priv->mdev;
797 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
798 in = kvzalloc(inlen, GFP_KERNEL);
802 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
804 MLX5_SET(modify_rq_in, in, rq_state, MLX5_RQC_STATE_RDY);
805 MLX5_SET64(modify_rq_in, in, modify_bitmask,
806 MLX5_MODIFY_RQ_IN_MODIFY_BITMASK_SCATTER_FCS);
807 MLX5_SET(rqc, rqc, scatter_fcs, enable);
808 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RDY);
810 err = mlx5_core_modify_rq(mdev, rq->rqn, in, inlen);
817 static int mlx5e_modify_rq_vsd(struct mlx5e_rq *rq, bool vsd)
819 struct mlx5e_channel *c = rq->channel;
820 struct mlx5_core_dev *mdev = c->mdev;
826 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
827 in = kvzalloc(inlen, GFP_KERNEL);
831 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
833 MLX5_SET(modify_rq_in, in, rq_state, MLX5_RQC_STATE_RDY);
834 MLX5_SET64(modify_rq_in, in, modify_bitmask,
835 MLX5_MODIFY_RQ_IN_MODIFY_BITMASK_VSD);
836 MLX5_SET(rqc, rqc, vsd, vsd);
837 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RDY);
839 err = mlx5_core_modify_rq(mdev, rq->rqn, in, inlen);
846 static void mlx5e_destroy_rq(struct mlx5e_rq *rq)
848 mlx5_core_destroy_rq(rq->mdev, rq->rqn);
851 static int mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq)
853 unsigned long exp_time = jiffies + msecs_to_jiffies(20000);
854 struct mlx5e_channel *c = rq->channel;
856 struct mlx5_wq_ll *wq = &rq->wq;
857 u16 min_wqes = mlx5_min_rx_wqes(rq->wq_type, mlx5_wq_ll_get_size(wq));
859 while (time_before(jiffies, exp_time)) {
860 if (wq->cur_sz >= min_wqes)
866 netdev_warn(c->netdev, "Failed to get min RX wqes on RQN[0x%x] wq cur_sz(%d) min_rx_wqes(%d)\n",
867 rq->rqn, wq->cur_sz, min_wqes);
871 static void mlx5e_free_rx_descs(struct mlx5e_rq *rq)
873 struct mlx5_wq_ll *wq = &rq->wq;
874 struct mlx5e_rx_wqe *wqe;
878 /* UMR WQE (if in progress) is always at wq->head */
879 if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ &&
880 rq->mpwqe.umr_in_progress)
881 mlx5e_free_rx_mpwqe(rq, &rq->mpwqe.info[wq->head]);
883 while (!mlx5_wq_ll_is_empty(wq)) {
884 wqe_ix_be = *wq->tail_next;
885 wqe_ix = be16_to_cpu(wqe_ix_be);
886 wqe = mlx5_wq_ll_get_wqe(&rq->wq, wqe_ix);
887 rq->dealloc_wqe(rq, wqe_ix);
888 mlx5_wq_ll_pop(&rq->wq, wqe_ix_be,
889 &wqe->next.next_wqe_index);
892 if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST && rq->wqe.page_reuse) {
893 /* Clean outstanding pages on handled WQEs that decided to do page-reuse,
894 * but yet to be re-posted.
896 int wq_sz = mlx5_wq_ll_get_size(&rq->wq);
898 for (wqe_ix = 0; wqe_ix < wq_sz; wqe_ix++)
899 rq->dealloc_wqe(rq, wqe_ix);
903 static int mlx5e_open_rq(struct mlx5e_channel *c,
904 struct mlx5e_params *params,
905 struct mlx5e_rq_param *param,
910 err = mlx5e_alloc_rq(c, params, param, rq);
914 err = mlx5e_create_rq(rq, param);
918 err = mlx5e_modify_rq_state(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
922 if (params->rx_am_enabled)
923 c->rq.state |= BIT(MLX5E_RQ_STATE_AM);
928 mlx5e_destroy_rq(rq);
935 static void mlx5e_activate_rq(struct mlx5e_rq *rq)
937 struct mlx5e_icosq *sq = &rq->channel->icosq;
938 u16 pi = sq->pc & sq->wq.sz_m1;
939 struct mlx5e_tx_wqe *nopwqe;
941 set_bit(MLX5E_RQ_STATE_ENABLED, &rq->state);
942 sq->db.ico_wqe[pi].opcode = MLX5_OPCODE_NOP;
943 nopwqe = mlx5e_post_nop(&sq->wq, sq->sqn, &sq->pc);
944 mlx5e_notify_hw(&sq->wq, sq->pc, sq->uar_map, &nopwqe->ctrl);
947 static void mlx5e_deactivate_rq(struct mlx5e_rq *rq)
949 clear_bit(MLX5E_RQ_STATE_ENABLED, &rq->state);
950 napi_synchronize(&rq->channel->napi); /* prevent mlx5e_post_rx_wqes */
953 static void mlx5e_close_rq(struct mlx5e_rq *rq)
955 cancel_work_sync(&rq->am.work);
956 mlx5e_destroy_rq(rq);
957 mlx5e_free_rx_descs(rq);
961 static void mlx5e_free_xdpsq_db(struct mlx5e_xdpsq *sq)
966 static int mlx5e_alloc_xdpsq_db(struct mlx5e_xdpsq *sq, int numa)
968 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
970 sq->db.di = kzalloc_node(sizeof(*sq->db.di) * wq_sz,
973 mlx5e_free_xdpsq_db(sq);
980 static int mlx5e_alloc_xdpsq(struct mlx5e_channel *c,
981 struct mlx5e_params *params,
982 struct mlx5e_sq_param *param,
983 struct mlx5e_xdpsq *sq)
985 void *sqc_wq = MLX5_ADDR_OF(sqc, param->sqc, wq);
986 struct mlx5_core_dev *mdev = c->mdev;
990 sq->mkey_be = c->mkey_be;
992 sq->uar_map = mdev->mlx5e_res.bfreg.map;
993 sq->min_inline_mode = params->tx_min_inline_mode;
995 param->wq.db_numa_node = cpu_to_node(c->cpu);
996 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, &sq->wq, &sq->wq_ctrl);
999 sq->wq.db = &sq->wq.db[MLX5_SND_DBR];
1001 err = mlx5e_alloc_xdpsq_db(sq, cpu_to_node(c->cpu));
1003 goto err_sq_wq_destroy;
1008 mlx5_wq_destroy(&sq->wq_ctrl);
1013 static void mlx5e_free_xdpsq(struct mlx5e_xdpsq *sq)
1015 mlx5e_free_xdpsq_db(sq);
1016 mlx5_wq_destroy(&sq->wq_ctrl);
1019 static void mlx5e_free_icosq_db(struct mlx5e_icosq *sq)
1021 kfree(sq->db.ico_wqe);
1024 static int mlx5e_alloc_icosq_db(struct mlx5e_icosq *sq, int numa)
1026 u8 wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1028 sq->db.ico_wqe = kzalloc_node(sizeof(*sq->db.ico_wqe) * wq_sz,
1030 if (!sq->db.ico_wqe)
1036 static int mlx5e_alloc_icosq(struct mlx5e_channel *c,
1037 struct mlx5e_sq_param *param,
1038 struct mlx5e_icosq *sq)
1040 void *sqc_wq = MLX5_ADDR_OF(sqc, param->sqc, wq);
1041 struct mlx5_core_dev *mdev = c->mdev;
1044 sq->mkey_be = c->mkey_be;
1046 sq->uar_map = mdev->mlx5e_res.bfreg.map;
1048 param->wq.db_numa_node = cpu_to_node(c->cpu);
1049 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, &sq->wq, &sq->wq_ctrl);
1052 sq->wq.db = &sq->wq.db[MLX5_SND_DBR];
1054 err = mlx5e_alloc_icosq_db(sq, cpu_to_node(c->cpu));
1056 goto err_sq_wq_destroy;
1058 sq->edge = (sq->wq.sz_m1 + 1) - MLX5E_ICOSQ_MAX_WQEBBS;
1063 mlx5_wq_destroy(&sq->wq_ctrl);
1068 static void mlx5e_free_icosq(struct mlx5e_icosq *sq)
1070 mlx5e_free_icosq_db(sq);
1071 mlx5_wq_destroy(&sq->wq_ctrl);
1074 static void mlx5e_free_txqsq_db(struct mlx5e_txqsq *sq)
1076 kfree(sq->db.wqe_info);
1077 kfree(sq->db.dma_fifo);
1080 static int mlx5e_alloc_txqsq_db(struct mlx5e_txqsq *sq, int numa)
1082 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1083 int df_sz = wq_sz * MLX5_SEND_WQEBB_NUM_DS;
1085 sq->db.dma_fifo = kzalloc_node(df_sz * sizeof(*sq->db.dma_fifo),
1087 sq->db.wqe_info = kzalloc_node(wq_sz * sizeof(*sq->db.wqe_info),
1089 if (!sq->db.dma_fifo || !sq->db.wqe_info) {
1090 mlx5e_free_txqsq_db(sq);
1094 sq->dma_fifo_mask = df_sz - 1;
1099 static int mlx5e_alloc_txqsq(struct mlx5e_channel *c,
1101 struct mlx5e_params *params,
1102 struct mlx5e_sq_param *param,
1103 struct mlx5e_txqsq *sq)
1105 void *sqc_wq = MLX5_ADDR_OF(sqc, param->sqc, wq);
1106 struct mlx5_core_dev *mdev = c->mdev;
1110 sq->tstamp = c->tstamp;
1111 sq->clock = &mdev->clock;
1112 sq->mkey_be = c->mkey_be;
1114 sq->txq_ix = txq_ix;
1115 sq->uar_map = mdev->mlx5e_res.bfreg.map;
1116 sq->max_inline = params->tx_max_inline;
1117 sq->min_inline_mode = params->tx_min_inline_mode;
1118 if (MLX5_IPSEC_DEV(c->priv->mdev))
1119 set_bit(MLX5E_SQ_STATE_IPSEC, &sq->state);
1121 param->wq.db_numa_node = cpu_to_node(c->cpu);
1122 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, &sq->wq, &sq->wq_ctrl);
1125 sq->wq.db = &sq->wq.db[MLX5_SND_DBR];
1127 err = mlx5e_alloc_txqsq_db(sq, cpu_to_node(c->cpu));
1129 goto err_sq_wq_destroy;
1131 sq->edge = (sq->wq.sz_m1 + 1) - MLX5_SEND_WQE_MAX_WQEBBS;
1136 mlx5_wq_destroy(&sq->wq_ctrl);
1141 static void mlx5e_free_txqsq(struct mlx5e_txqsq *sq)
1143 mlx5e_free_txqsq_db(sq);
1144 mlx5_wq_destroy(&sq->wq_ctrl);
1147 struct mlx5e_create_sq_param {
1148 struct mlx5_wq_ctrl *wq_ctrl;
1155 static int mlx5e_create_sq(struct mlx5_core_dev *mdev,
1156 struct mlx5e_sq_param *param,
1157 struct mlx5e_create_sq_param *csp,
1166 inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
1167 sizeof(u64) * csp->wq_ctrl->buf.npages;
1168 in = kvzalloc(inlen, GFP_KERNEL);
1172 sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
1173 wq = MLX5_ADDR_OF(sqc, sqc, wq);
1175 memcpy(sqc, param->sqc, sizeof(param->sqc));
1176 MLX5_SET(sqc, sqc, tis_lst_sz, csp->tis_lst_sz);
1177 MLX5_SET(sqc, sqc, tis_num_0, csp->tisn);
1178 MLX5_SET(sqc, sqc, cqn, csp->cqn);
1180 if (MLX5_CAP_ETH(mdev, wqe_inline_mode) == MLX5_CAP_INLINE_MODE_VPORT_CONTEXT)
1181 MLX5_SET(sqc, sqc, min_wqe_inline_mode, csp->min_inline_mode);
1183 MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
1185 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
1186 MLX5_SET(wq, wq, uar_page, mdev->mlx5e_res.bfreg.index);
1187 MLX5_SET(wq, wq, log_wq_pg_sz, csp->wq_ctrl->buf.page_shift -
1188 MLX5_ADAPTER_PAGE_SHIFT);
1189 MLX5_SET64(wq, wq, dbr_addr, csp->wq_ctrl->db.dma);
1191 mlx5_fill_page_array(&csp->wq_ctrl->buf, (__be64 *)MLX5_ADDR_OF(wq, wq, pas));
1193 err = mlx5_core_create_sq(mdev, in, inlen, sqn);
1200 struct mlx5e_modify_sq_param {
1207 static int mlx5e_modify_sq(struct mlx5_core_dev *mdev, u32 sqn,
1208 struct mlx5e_modify_sq_param *p)
1215 inlen = MLX5_ST_SZ_BYTES(modify_sq_in);
1216 in = kvzalloc(inlen, GFP_KERNEL);
1220 sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx);
1222 MLX5_SET(modify_sq_in, in, sq_state, p->curr_state);
1223 MLX5_SET(sqc, sqc, state, p->next_state);
1224 if (p->rl_update && p->next_state == MLX5_SQC_STATE_RDY) {
1225 MLX5_SET64(modify_sq_in, in, modify_bitmask, 1);
1226 MLX5_SET(sqc, sqc, packet_pacing_rate_limit_index, p->rl_index);
1229 err = mlx5_core_modify_sq(mdev, sqn, in, inlen);
1236 static void mlx5e_destroy_sq(struct mlx5_core_dev *mdev, u32 sqn)
1238 mlx5_core_destroy_sq(mdev, sqn);
1241 static int mlx5e_create_sq_rdy(struct mlx5_core_dev *mdev,
1242 struct mlx5e_sq_param *param,
1243 struct mlx5e_create_sq_param *csp,
1246 struct mlx5e_modify_sq_param msp = {0};
1249 err = mlx5e_create_sq(mdev, param, csp, sqn);
1253 msp.curr_state = MLX5_SQC_STATE_RST;
1254 msp.next_state = MLX5_SQC_STATE_RDY;
1255 err = mlx5e_modify_sq(mdev, *sqn, &msp);
1257 mlx5e_destroy_sq(mdev, *sqn);
1262 static int mlx5e_set_sq_maxrate(struct net_device *dev,
1263 struct mlx5e_txqsq *sq, u32 rate);
1265 static int mlx5e_open_txqsq(struct mlx5e_channel *c,
1268 struct mlx5e_params *params,
1269 struct mlx5e_sq_param *param,
1270 struct mlx5e_txqsq *sq)
1272 struct mlx5e_create_sq_param csp = {};
1276 err = mlx5e_alloc_txqsq(c, txq_ix, params, param, sq);
1282 csp.cqn = sq->cq.mcq.cqn;
1283 csp.wq_ctrl = &sq->wq_ctrl;
1284 csp.min_inline_mode = sq->min_inline_mode;
1285 err = mlx5e_create_sq_rdy(c->mdev, param, &csp, &sq->sqn);
1287 goto err_free_txqsq;
1289 tx_rate = c->priv->tx_rates[sq->txq_ix];
1291 mlx5e_set_sq_maxrate(c->netdev, sq, tx_rate);
1296 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1297 mlx5e_free_txqsq(sq);
1302 static void mlx5e_activate_txqsq(struct mlx5e_txqsq *sq)
1304 sq->txq = netdev_get_tx_queue(sq->channel->netdev, sq->txq_ix);
1305 set_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1306 netdev_tx_reset_queue(sq->txq);
1307 netif_tx_start_queue(sq->txq);
1310 static inline void netif_tx_disable_queue(struct netdev_queue *txq)
1312 __netif_tx_lock_bh(txq);
1313 netif_tx_stop_queue(txq);
1314 __netif_tx_unlock_bh(txq);
1317 static void mlx5e_deactivate_txqsq(struct mlx5e_txqsq *sq)
1319 struct mlx5e_channel *c = sq->channel;
1321 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1322 /* prevent netif_tx_wake_queue */
1323 napi_synchronize(&c->napi);
1325 netif_tx_disable_queue(sq->txq);
1327 /* last doorbell out, godspeed .. */
1328 if (mlx5e_wqc_has_room_for(&sq->wq, sq->cc, sq->pc, 1)) {
1329 struct mlx5e_tx_wqe *nop;
1331 sq->db.wqe_info[(sq->pc & sq->wq.sz_m1)].skb = NULL;
1332 nop = mlx5e_post_nop(&sq->wq, sq->sqn, &sq->pc);
1333 mlx5e_notify_hw(&sq->wq, sq->pc, sq->uar_map, &nop->ctrl);
1337 static void mlx5e_close_txqsq(struct mlx5e_txqsq *sq)
1339 struct mlx5e_channel *c = sq->channel;
1340 struct mlx5_core_dev *mdev = c->mdev;
1342 mlx5e_destroy_sq(mdev, sq->sqn);
1344 mlx5_rl_remove_rate(mdev, sq->rate_limit);
1345 mlx5e_free_txqsq_descs(sq);
1346 mlx5e_free_txqsq(sq);
1349 static int mlx5e_open_icosq(struct mlx5e_channel *c,
1350 struct mlx5e_params *params,
1351 struct mlx5e_sq_param *param,
1352 struct mlx5e_icosq *sq)
1354 struct mlx5e_create_sq_param csp = {};
1357 err = mlx5e_alloc_icosq(c, param, sq);
1361 csp.cqn = sq->cq.mcq.cqn;
1362 csp.wq_ctrl = &sq->wq_ctrl;
1363 csp.min_inline_mode = params->tx_min_inline_mode;
1364 set_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1365 err = mlx5e_create_sq_rdy(c->mdev, param, &csp, &sq->sqn);
1367 goto err_free_icosq;
1372 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1373 mlx5e_free_icosq(sq);
1378 static void mlx5e_close_icosq(struct mlx5e_icosq *sq)
1380 struct mlx5e_channel *c = sq->channel;
1382 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1383 napi_synchronize(&c->napi);
1385 mlx5e_destroy_sq(c->mdev, sq->sqn);
1386 mlx5e_free_icosq(sq);
1389 static int mlx5e_open_xdpsq(struct mlx5e_channel *c,
1390 struct mlx5e_params *params,
1391 struct mlx5e_sq_param *param,
1392 struct mlx5e_xdpsq *sq)
1394 unsigned int ds_cnt = MLX5E_XDP_TX_DS_COUNT;
1395 struct mlx5e_create_sq_param csp = {};
1396 unsigned int inline_hdr_sz = 0;
1400 err = mlx5e_alloc_xdpsq(c, params, param, sq);
1405 csp.tisn = c->priv->tisn[0]; /* tc = 0 */
1406 csp.cqn = sq->cq.mcq.cqn;
1407 csp.wq_ctrl = &sq->wq_ctrl;
1408 csp.min_inline_mode = sq->min_inline_mode;
1409 set_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1410 err = mlx5e_create_sq_rdy(c->mdev, param, &csp, &sq->sqn);
1412 goto err_free_xdpsq;
1414 if (sq->min_inline_mode != MLX5_INLINE_MODE_NONE) {
1415 inline_hdr_sz = MLX5E_XDP_MIN_INLINE;
1419 /* Pre initialize fixed WQE fields */
1420 for (i = 0; i < mlx5_wq_cyc_get_size(&sq->wq); i++) {
1421 struct mlx5e_tx_wqe *wqe = mlx5_wq_cyc_get_wqe(&sq->wq, i);
1422 struct mlx5_wqe_ctrl_seg *cseg = &wqe->ctrl;
1423 struct mlx5_wqe_eth_seg *eseg = &wqe->eth;
1424 struct mlx5_wqe_data_seg *dseg;
1426 cseg->qpn_ds = cpu_to_be32((sq->sqn << 8) | ds_cnt);
1427 eseg->inline_hdr.sz = cpu_to_be16(inline_hdr_sz);
1429 dseg = (struct mlx5_wqe_data_seg *)cseg + (ds_cnt - 1);
1430 dseg->lkey = sq->mkey_be;
1436 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1437 mlx5e_free_xdpsq(sq);
1442 static void mlx5e_close_xdpsq(struct mlx5e_xdpsq *sq)
1444 struct mlx5e_channel *c = sq->channel;
1446 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1447 napi_synchronize(&c->napi);
1449 mlx5e_destroy_sq(c->mdev, sq->sqn);
1450 mlx5e_free_xdpsq_descs(sq);
1451 mlx5e_free_xdpsq(sq);
1454 static int mlx5e_alloc_cq_common(struct mlx5_core_dev *mdev,
1455 struct mlx5e_cq_param *param,
1456 struct mlx5e_cq *cq)
1458 struct mlx5_core_cq *mcq = &cq->mcq;
1464 err = mlx5_cqwq_create(mdev, ¶m->wq, param->cqc, &cq->wq,
1469 mlx5_vector2eqn(mdev, param->eq_ix, &eqn_not_used, &irqn);
1472 mcq->set_ci_db = cq->wq_ctrl.db.db;
1473 mcq->arm_db = cq->wq_ctrl.db.db + 1;
1474 *mcq->set_ci_db = 0;
1476 mcq->vector = param->eq_ix;
1477 mcq->comp = mlx5e_completion_event;
1478 mcq->event = mlx5e_cq_error_event;
1481 for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) {
1482 struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i);
1492 static int mlx5e_alloc_cq(struct mlx5e_channel *c,
1493 struct mlx5e_cq_param *param,
1494 struct mlx5e_cq *cq)
1496 struct mlx5_core_dev *mdev = c->priv->mdev;
1499 param->wq.buf_numa_node = cpu_to_node(c->cpu);
1500 param->wq.db_numa_node = cpu_to_node(c->cpu);
1501 param->eq_ix = c->ix;
1503 err = mlx5e_alloc_cq_common(mdev, param, cq);
1505 cq->napi = &c->napi;
1511 static void mlx5e_free_cq(struct mlx5e_cq *cq)
1513 mlx5_cqwq_destroy(&cq->wq_ctrl);
1516 static int mlx5e_create_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param)
1518 struct mlx5_core_dev *mdev = cq->mdev;
1519 struct mlx5_core_cq *mcq = &cq->mcq;
1524 unsigned int irqn_not_used;
1528 inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
1529 sizeof(u64) * cq->wq_ctrl.frag_buf.npages;
1530 in = kvzalloc(inlen, GFP_KERNEL);
1534 cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
1536 memcpy(cqc, param->cqc, sizeof(param->cqc));
1538 mlx5_fill_page_frag_array(&cq->wq_ctrl.frag_buf,
1539 (__be64 *)MLX5_ADDR_OF(create_cq_in, in, pas));
1541 mlx5_vector2eqn(mdev, param->eq_ix, &eqn, &irqn_not_used);
1543 MLX5_SET(cqc, cqc, cq_period_mode, param->cq_period_mode);
1544 MLX5_SET(cqc, cqc, c_eqn, eqn);
1545 MLX5_SET(cqc, cqc, uar_page, mdev->priv.uar->index);
1546 MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.frag_buf.page_shift -
1547 MLX5_ADAPTER_PAGE_SHIFT);
1548 MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
1550 err = mlx5_core_create_cq(mdev, mcq, in, inlen);
1562 static void mlx5e_destroy_cq(struct mlx5e_cq *cq)
1564 mlx5_core_destroy_cq(cq->mdev, &cq->mcq);
1567 static int mlx5e_open_cq(struct mlx5e_channel *c,
1568 struct mlx5e_cq_moder moder,
1569 struct mlx5e_cq_param *param,
1570 struct mlx5e_cq *cq)
1572 struct mlx5_core_dev *mdev = c->mdev;
1575 err = mlx5e_alloc_cq(c, param, cq);
1579 err = mlx5e_create_cq(cq, param);
1583 if (MLX5_CAP_GEN(mdev, cq_moderation))
1584 mlx5_core_modify_cq_moderation(mdev, &cq->mcq, moder.usec, moder.pkts);
1593 static void mlx5e_close_cq(struct mlx5e_cq *cq)
1595 mlx5e_destroy_cq(cq);
1599 static int mlx5e_get_cpu(struct mlx5e_priv *priv, int ix)
1601 return cpumask_first(priv->mdev->priv.irq_info[ix].mask);
1604 static int mlx5e_open_tx_cqs(struct mlx5e_channel *c,
1605 struct mlx5e_params *params,
1606 struct mlx5e_channel_param *cparam)
1611 for (tc = 0; tc < c->num_tc; tc++) {
1612 err = mlx5e_open_cq(c, params->tx_cq_moderation,
1613 &cparam->tx_cq, &c->sq[tc].cq);
1615 goto err_close_tx_cqs;
1621 for (tc--; tc >= 0; tc--)
1622 mlx5e_close_cq(&c->sq[tc].cq);
1627 static void mlx5e_close_tx_cqs(struct mlx5e_channel *c)
1631 for (tc = 0; tc < c->num_tc; tc++)
1632 mlx5e_close_cq(&c->sq[tc].cq);
1635 static int mlx5e_open_sqs(struct mlx5e_channel *c,
1636 struct mlx5e_params *params,
1637 struct mlx5e_channel_param *cparam)
1642 for (tc = 0; tc < params->num_tc; tc++) {
1643 int txq_ix = c->ix + tc * params->num_channels;
1645 err = mlx5e_open_txqsq(c, c->priv->tisn[tc], txq_ix,
1646 params, &cparam->sq, &c->sq[tc]);
1654 for (tc--; tc >= 0; tc--)
1655 mlx5e_close_txqsq(&c->sq[tc]);
1660 static void mlx5e_close_sqs(struct mlx5e_channel *c)
1664 for (tc = 0; tc < c->num_tc; tc++)
1665 mlx5e_close_txqsq(&c->sq[tc]);
1668 static int mlx5e_set_sq_maxrate(struct net_device *dev,
1669 struct mlx5e_txqsq *sq, u32 rate)
1671 struct mlx5e_priv *priv = netdev_priv(dev);
1672 struct mlx5_core_dev *mdev = priv->mdev;
1673 struct mlx5e_modify_sq_param msp = {0};
1677 if (rate == sq->rate_limit)
1682 /* remove current rl index to free space to next ones */
1683 mlx5_rl_remove_rate(mdev, sq->rate_limit);
1688 err = mlx5_rl_add_rate(mdev, rate, &rl_index);
1690 netdev_err(dev, "Failed configuring rate %u: %d\n",
1696 msp.curr_state = MLX5_SQC_STATE_RDY;
1697 msp.next_state = MLX5_SQC_STATE_RDY;
1698 msp.rl_index = rl_index;
1699 msp.rl_update = true;
1700 err = mlx5e_modify_sq(mdev, sq->sqn, &msp);
1702 netdev_err(dev, "Failed configuring rate %u: %d\n",
1704 /* remove the rate from the table */
1706 mlx5_rl_remove_rate(mdev, rate);
1710 sq->rate_limit = rate;
1714 static int mlx5e_set_tx_maxrate(struct net_device *dev, int index, u32 rate)
1716 struct mlx5e_priv *priv = netdev_priv(dev);
1717 struct mlx5_core_dev *mdev = priv->mdev;
1718 struct mlx5e_txqsq *sq = priv->txq2sq[index];
1721 if (!mlx5_rl_is_supported(mdev)) {
1722 netdev_err(dev, "Rate limiting is not supported on this device\n");
1726 /* rate is given in Mb/sec, HW config is in Kb/sec */
1729 /* Check whether rate in valid range, 0 is always valid */
1730 if (rate && !mlx5_rl_is_in_range(mdev, rate)) {
1731 netdev_err(dev, "TX rate %u, is not in range\n", rate);
1735 mutex_lock(&priv->state_lock);
1736 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
1737 err = mlx5e_set_sq_maxrate(dev, sq, rate);
1739 priv->tx_rates[index] = rate;
1740 mutex_unlock(&priv->state_lock);
1745 static int mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
1746 struct mlx5e_params *params,
1747 struct mlx5e_channel_param *cparam,
1748 struct mlx5e_channel **cp)
1750 struct mlx5e_cq_moder icocq_moder = {0, 0};
1751 struct net_device *netdev = priv->netdev;
1752 int cpu = mlx5e_get_cpu(priv, ix);
1753 struct mlx5e_channel *c;
1758 c = kzalloc_node(sizeof(*c), GFP_KERNEL, cpu_to_node(cpu));
1763 c->mdev = priv->mdev;
1764 c->tstamp = &priv->tstamp;
1767 c->pdev = &priv->mdev->pdev->dev;
1768 c->netdev = priv->netdev;
1769 c->mkey_be = cpu_to_be32(priv->mdev->mlx5e_res.mkey.key);
1770 c->num_tc = params->num_tc;
1771 c->xdp = !!params->xdp_prog;
1773 mlx5_vector2eqn(priv->mdev, ix, &eqn, &irq);
1774 c->irq_desc = irq_to_desc(irq);
1776 netif_napi_add(netdev, &c->napi, mlx5e_napi_poll, 64);
1778 err = mlx5e_open_cq(c, icocq_moder, &cparam->icosq_cq, &c->icosq.cq);
1782 err = mlx5e_open_tx_cqs(c, params, cparam);
1784 goto err_close_icosq_cq;
1786 err = mlx5e_open_cq(c, params->rx_cq_moderation, &cparam->rx_cq, &c->rq.cq);
1788 goto err_close_tx_cqs;
1790 /* XDP SQ CQ params are same as normal TXQ sq CQ params */
1791 err = c->xdp ? mlx5e_open_cq(c, params->tx_cq_moderation,
1792 &cparam->tx_cq, &c->rq.xdpsq.cq) : 0;
1794 goto err_close_rx_cq;
1796 napi_enable(&c->napi);
1798 err = mlx5e_open_icosq(c, params, &cparam->icosq, &c->icosq);
1800 goto err_disable_napi;
1802 err = mlx5e_open_sqs(c, params, cparam);
1804 goto err_close_icosq;
1806 err = c->xdp ? mlx5e_open_xdpsq(c, params, &cparam->xdp_sq, &c->rq.xdpsq) : 0;
1810 err = mlx5e_open_rq(c, params, &cparam->rq, &c->rq);
1812 goto err_close_xdp_sq;
1819 mlx5e_close_xdpsq(&c->rq.xdpsq);
1825 mlx5e_close_icosq(&c->icosq);
1828 napi_disable(&c->napi);
1830 mlx5e_close_cq(&c->rq.xdpsq.cq);
1833 mlx5e_close_cq(&c->rq.cq);
1836 mlx5e_close_tx_cqs(c);
1839 mlx5e_close_cq(&c->icosq.cq);
1842 netif_napi_del(&c->napi);
1848 static void mlx5e_activate_channel(struct mlx5e_channel *c)
1852 for (tc = 0; tc < c->num_tc; tc++)
1853 mlx5e_activate_txqsq(&c->sq[tc]);
1854 mlx5e_activate_rq(&c->rq);
1855 netif_set_xps_queue(c->netdev, get_cpu_mask(c->cpu), c->ix);
1858 static void mlx5e_deactivate_channel(struct mlx5e_channel *c)
1862 mlx5e_deactivate_rq(&c->rq);
1863 for (tc = 0; tc < c->num_tc; tc++)
1864 mlx5e_deactivate_txqsq(&c->sq[tc]);
1867 static void mlx5e_close_channel(struct mlx5e_channel *c)
1869 mlx5e_close_rq(&c->rq);
1871 mlx5e_close_xdpsq(&c->rq.xdpsq);
1873 mlx5e_close_icosq(&c->icosq);
1874 napi_disable(&c->napi);
1876 mlx5e_close_cq(&c->rq.xdpsq.cq);
1877 mlx5e_close_cq(&c->rq.cq);
1878 mlx5e_close_tx_cqs(c);
1879 mlx5e_close_cq(&c->icosq.cq);
1880 netif_napi_del(&c->napi);
1885 static void mlx5e_build_rq_param(struct mlx5e_priv *priv,
1886 struct mlx5e_params *params,
1887 struct mlx5e_rq_param *param)
1889 void *rqc = param->rqc;
1890 void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
1892 switch (params->rq_wq_type) {
1893 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
1894 MLX5_SET(wq, wq, log_wqe_num_of_strides, params->mpwqe_log_num_strides - 9);
1895 MLX5_SET(wq, wq, log_wqe_stride_size, params->mpwqe_log_stride_sz - 6);
1896 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ);
1898 default: /* MLX5_WQ_TYPE_LINKED_LIST */
1899 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
1902 MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN);
1903 MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe)));
1904 MLX5_SET(wq, wq, log_wq_sz, params->log_rq_size);
1905 MLX5_SET(wq, wq, pd, priv->mdev->mlx5e_res.pdn);
1906 MLX5_SET(rqc, rqc, counter_set_id, priv->q_counter);
1907 MLX5_SET(rqc, rqc, vsd, params->vlan_strip_disable);
1908 MLX5_SET(rqc, rqc, scatter_fcs, params->scatter_fcs_en);
1910 param->wq.buf_numa_node = dev_to_node(&priv->mdev->pdev->dev);
1911 param->wq.linear = 1;
1914 static void mlx5e_build_drop_rq_param(struct mlx5e_rq_param *param)
1916 void *rqc = param->rqc;
1917 void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
1919 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
1920 MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe)));
1923 static void mlx5e_build_sq_param_common(struct mlx5e_priv *priv,
1924 struct mlx5e_sq_param *param)
1926 void *sqc = param->sqc;
1927 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
1929 MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
1930 MLX5_SET(wq, wq, pd, priv->mdev->mlx5e_res.pdn);
1932 param->wq.buf_numa_node = dev_to_node(&priv->mdev->pdev->dev);
1935 static void mlx5e_build_sq_param(struct mlx5e_priv *priv,
1936 struct mlx5e_params *params,
1937 struct mlx5e_sq_param *param)
1939 void *sqc = param->sqc;
1940 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
1942 mlx5e_build_sq_param_common(priv, param);
1943 MLX5_SET(wq, wq, log_wq_sz, params->log_sq_size);
1944 MLX5_SET(sqc, sqc, allow_swp, !!MLX5_IPSEC_DEV(priv->mdev));
1947 static void mlx5e_build_common_cq_param(struct mlx5e_priv *priv,
1948 struct mlx5e_cq_param *param)
1950 void *cqc = param->cqc;
1952 MLX5_SET(cqc, cqc, uar_page, priv->mdev->priv.uar->index);
1955 static void mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
1956 struct mlx5e_params *params,
1957 struct mlx5e_cq_param *param)
1959 void *cqc = param->cqc;
1962 switch (params->rq_wq_type) {
1963 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
1964 log_cq_size = params->log_rq_size + params->mpwqe_log_num_strides;
1966 default: /* MLX5_WQ_TYPE_LINKED_LIST */
1967 log_cq_size = params->log_rq_size;
1970 MLX5_SET(cqc, cqc, log_cq_size, log_cq_size);
1971 if (MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS)) {
1972 MLX5_SET(cqc, cqc, mini_cqe_res_format, MLX5_CQE_FORMAT_CSUM);
1973 MLX5_SET(cqc, cqc, cqe_comp_en, 1);
1976 mlx5e_build_common_cq_param(priv, param);
1977 param->cq_period_mode = params->rx_cq_moderation.cq_period_mode;
1980 static void mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
1981 struct mlx5e_params *params,
1982 struct mlx5e_cq_param *param)
1984 void *cqc = param->cqc;
1986 MLX5_SET(cqc, cqc, log_cq_size, params->log_sq_size);
1988 mlx5e_build_common_cq_param(priv, param);
1989 param->cq_period_mode = params->tx_cq_moderation.cq_period_mode;
1992 static void mlx5e_build_ico_cq_param(struct mlx5e_priv *priv,
1994 struct mlx5e_cq_param *param)
1996 void *cqc = param->cqc;
1998 MLX5_SET(cqc, cqc, log_cq_size, log_wq_size);
2000 mlx5e_build_common_cq_param(priv, param);
2002 param->cq_period_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
2005 static void mlx5e_build_icosq_param(struct mlx5e_priv *priv,
2007 struct mlx5e_sq_param *param)
2009 void *sqc = param->sqc;
2010 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
2012 mlx5e_build_sq_param_common(priv, param);
2014 MLX5_SET(wq, wq, log_wq_sz, log_wq_size);
2015 MLX5_SET(sqc, sqc, reg_umr, MLX5_CAP_ETH(priv->mdev, reg_umr_sq));
2018 static void mlx5e_build_xdpsq_param(struct mlx5e_priv *priv,
2019 struct mlx5e_params *params,
2020 struct mlx5e_sq_param *param)
2022 void *sqc = param->sqc;
2023 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
2025 mlx5e_build_sq_param_common(priv, param);
2026 MLX5_SET(wq, wq, log_wq_sz, params->log_sq_size);
2029 static void mlx5e_build_channel_param(struct mlx5e_priv *priv,
2030 struct mlx5e_params *params,
2031 struct mlx5e_channel_param *cparam)
2033 u8 icosq_log_wq_sz = MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE;
2035 mlx5e_build_rq_param(priv, params, &cparam->rq);
2036 mlx5e_build_sq_param(priv, params, &cparam->sq);
2037 mlx5e_build_xdpsq_param(priv, params, &cparam->xdp_sq);
2038 mlx5e_build_icosq_param(priv, icosq_log_wq_sz, &cparam->icosq);
2039 mlx5e_build_rx_cq_param(priv, params, &cparam->rx_cq);
2040 mlx5e_build_tx_cq_param(priv, params, &cparam->tx_cq);
2041 mlx5e_build_ico_cq_param(priv, icosq_log_wq_sz, &cparam->icosq_cq);
2044 int mlx5e_open_channels(struct mlx5e_priv *priv,
2045 struct mlx5e_channels *chs)
2047 struct mlx5e_channel_param *cparam;
2051 chs->num = chs->params.num_channels;
2053 chs->c = kcalloc(chs->num, sizeof(struct mlx5e_channel *), GFP_KERNEL);
2054 cparam = kzalloc(sizeof(struct mlx5e_channel_param), GFP_KERNEL);
2055 if (!chs->c || !cparam)
2058 mlx5e_build_channel_param(priv, &chs->params, cparam);
2059 for (i = 0; i < chs->num; i++) {
2060 err = mlx5e_open_channel(priv, i, &chs->params, cparam, &chs->c[i]);
2062 goto err_close_channels;
2069 for (i--; i >= 0; i--)
2070 mlx5e_close_channel(chs->c[i]);
2079 static void mlx5e_activate_channels(struct mlx5e_channels *chs)
2083 for (i = 0; i < chs->num; i++)
2084 mlx5e_activate_channel(chs->c[i]);
2087 static int mlx5e_wait_channels_min_rx_wqes(struct mlx5e_channels *chs)
2092 for (i = 0; i < chs->num; i++) {
2093 err = mlx5e_wait_for_min_rx_wqes(&chs->c[i]->rq);
2101 static void mlx5e_deactivate_channels(struct mlx5e_channels *chs)
2105 for (i = 0; i < chs->num; i++)
2106 mlx5e_deactivate_channel(chs->c[i]);
2109 void mlx5e_close_channels(struct mlx5e_channels *chs)
2113 for (i = 0; i < chs->num; i++)
2114 mlx5e_close_channel(chs->c[i]);
2121 mlx5e_create_rqt(struct mlx5e_priv *priv, int sz, struct mlx5e_rqt *rqt)
2123 struct mlx5_core_dev *mdev = priv->mdev;
2130 inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
2131 in = kvzalloc(inlen, GFP_KERNEL);
2135 rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
2137 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
2138 MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
2140 for (i = 0; i < sz; i++)
2141 MLX5_SET(rqtc, rqtc, rq_num[i], priv->drop_rq.rqn);
2143 err = mlx5_core_create_rqt(mdev, in, inlen, &rqt->rqtn);
2145 rqt->enabled = true;
2151 void mlx5e_destroy_rqt(struct mlx5e_priv *priv, struct mlx5e_rqt *rqt)
2153 rqt->enabled = false;
2154 mlx5_core_destroy_rqt(priv->mdev, rqt->rqtn);
2157 int mlx5e_create_indirect_rqt(struct mlx5e_priv *priv)
2159 struct mlx5e_rqt *rqt = &priv->indir_rqt;
2162 err = mlx5e_create_rqt(priv, MLX5E_INDIR_RQT_SIZE, rqt);
2164 mlx5_core_warn(priv->mdev, "create indirect rqts failed, %d\n", err);
2168 int mlx5e_create_direct_rqts(struct mlx5e_priv *priv)
2170 struct mlx5e_rqt *rqt;
2174 for (ix = 0; ix < priv->profile->max_nch(priv->mdev); ix++) {
2175 rqt = &priv->direct_tir[ix].rqt;
2176 err = mlx5e_create_rqt(priv, 1 /*size */, rqt);
2178 goto err_destroy_rqts;
2184 mlx5_core_warn(priv->mdev, "create direct rqts failed, %d\n", err);
2185 for (ix--; ix >= 0; ix--)
2186 mlx5e_destroy_rqt(priv, &priv->direct_tir[ix].rqt);
2191 void mlx5e_destroy_direct_rqts(struct mlx5e_priv *priv)
2195 for (i = 0; i < priv->profile->max_nch(priv->mdev); i++)
2196 mlx5e_destroy_rqt(priv, &priv->direct_tir[i].rqt);
2199 static int mlx5e_rx_hash_fn(int hfunc)
2201 return (hfunc == ETH_RSS_HASH_TOP) ?
2202 MLX5_RX_HASH_FN_TOEPLITZ :
2203 MLX5_RX_HASH_FN_INVERTED_XOR8;
2206 static int mlx5e_bits_invert(unsigned long a, int size)
2211 for (i = 0; i < size; i++)
2212 inv |= (test_bit(size - i - 1, &a) ? 1 : 0) << i;
2217 static void mlx5e_fill_rqt_rqns(struct mlx5e_priv *priv, int sz,
2218 struct mlx5e_redirect_rqt_param rrp, void *rqtc)
2222 for (i = 0; i < sz; i++) {
2228 if (rrp.rss.hfunc == ETH_RSS_HASH_XOR)
2229 ix = mlx5e_bits_invert(i, ilog2(sz));
2231 ix = priv->channels.params.indirection_rqt[ix];
2232 rqn = rrp.rss.channels->c[ix]->rq.rqn;
2236 MLX5_SET(rqtc, rqtc, rq_num[i], rqn);
2240 int mlx5e_redirect_rqt(struct mlx5e_priv *priv, u32 rqtn, int sz,
2241 struct mlx5e_redirect_rqt_param rrp)
2243 struct mlx5_core_dev *mdev = priv->mdev;
2249 inlen = MLX5_ST_SZ_BYTES(modify_rqt_in) + sizeof(u32) * sz;
2250 in = kvzalloc(inlen, GFP_KERNEL);
2254 rqtc = MLX5_ADDR_OF(modify_rqt_in, in, ctx);
2256 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
2257 MLX5_SET(modify_rqt_in, in, bitmask.rqn_list, 1);
2258 mlx5e_fill_rqt_rqns(priv, sz, rrp, rqtc);
2259 err = mlx5_core_modify_rqt(mdev, rqtn, in, inlen);
2265 static u32 mlx5e_get_direct_rqn(struct mlx5e_priv *priv, int ix,
2266 struct mlx5e_redirect_rqt_param rrp)
2271 if (ix >= rrp.rss.channels->num)
2272 return priv->drop_rq.rqn;
2274 return rrp.rss.channels->c[ix]->rq.rqn;
2277 static void mlx5e_redirect_rqts(struct mlx5e_priv *priv,
2278 struct mlx5e_redirect_rqt_param rrp)
2283 if (priv->indir_rqt.enabled) {
2285 rqtn = priv->indir_rqt.rqtn;
2286 mlx5e_redirect_rqt(priv, rqtn, MLX5E_INDIR_RQT_SIZE, rrp);
2289 for (ix = 0; ix < priv->profile->max_nch(priv->mdev); ix++) {
2290 struct mlx5e_redirect_rqt_param direct_rrp = {
2293 .rqn = mlx5e_get_direct_rqn(priv, ix, rrp)
2297 /* Direct RQ Tables */
2298 if (!priv->direct_tir[ix].rqt.enabled)
2301 rqtn = priv->direct_tir[ix].rqt.rqtn;
2302 mlx5e_redirect_rqt(priv, rqtn, 1, direct_rrp);
2306 static void mlx5e_redirect_rqts_to_channels(struct mlx5e_priv *priv,
2307 struct mlx5e_channels *chs)
2309 struct mlx5e_redirect_rqt_param rrp = {
2314 .hfunc = chs->params.rss_hfunc,
2319 mlx5e_redirect_rqts(priv, rrp);
2322 static void mlx5e_redirect_rqts_to_drop(struct mlx5e_priv *priv)
2324 struct mlx5e_redirect_rqt_param drop_rrp = {
2327 .rqn = priv->drop_rq.rqn,
2331 mlx5e_redirect_rqts(priv, drop_rrp);
2334 static void mlx5e_build_tir_ctx_lro(struct mlx5e_params *params, void *tirc)
2336 if (!params->lro_en)
2339 #define ROUGH_MAX_L2_L3_HDR_SZ 256
2341 MLX5_SET(tirc, tirc, lro_enable_mask,
2342 MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO |
2343 MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO);
2344 MLX5_SET(tirc, tirc, lro_max_ip_payload_size,
2345 (params->lro_wqe_sz - ROUGH_MAX_L2_L3_HDR_SZ) >> 8);
2346 MLX5_SET(tirc, tirc, lro_timeout_period_usecs, params->lro_timeout);
2349 void mlx5e_build_indir_tir_ctx_hash(struct mlx5e_params *params,
2350 enum mlx5e_traffic_types tt,
2351 void *tirc, bool inner)
2353 void *hfso = inner ? MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_inner) :
2354 MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
2356 #define MLX5_HASH_IP (MLX5_HASH_FIELD_SEL_SRC_IP |\
2357 MLX5_HASH_FIELD_SEL_DST_IP)
2359 #define MLX5_HASH_IP_L4PORTS (MLX5_HASH_FIELD_SEL_SRC_IP |\
2360 MLX5_HASH_FIELD_SEL_DST_IP |\
2361 MLX5_HASH_FIELD_SEL_L4_SPORT |\
2362 MLX5_HASH_FIELD_SEL_L4_DPORT)
2364 #define MLX5_HASH_IP_IPSEC_SPI (MLX5_HASH_FIELD_SEL_SRC_IP |\
2365 MLX5_HASH_FIELD_SEL_DST_IP |\
2366 MLX5_HASH_FIELD_SEL_IPSEC_SPI)
2368 MLX5_SET(tirc, tirc, rx_hash_fn, mlx5e_rx_hash_fn(params->rss_hfunc));
2369 if (params->rss_hfunc == ETH_RSS_HASH_TOP) {
2370 void *rss_key = MLX5_ADDR_OF(tirc, tirc,
2371 rx_hash_toeplitz_key);
2372 size_t len = MLX5_FLD_SZ_BYTES(tirc,
2373 rx_hash_toeplitz_key);
2375 MLX5_SET(tirc, tirc, rx_hash_symmetric, 1);
2376 memcpy(rss_key, params->toeplitz_hash_key, len);
2380 case MLX5E_TT_IPV4_TCP:
2381 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2382 MLX5_L3_PROT_TYPE_IPV4);
2383 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2384 MLX5_L4_PROT_TYPE_TCP);
2385 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2386 MLX5_HASH_IP_L4PORTS);
2389 case MLX5E_TT_IPV6_TCP:
2390 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2391 MLX5_L3_PROT_TYPE_IPV6);
2392 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2393 MLX5_L4_PROT_TYPE_TCP);
2394 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2395 MLX5_HASH_IP_L4PORTS);
2398 case MLX5E_TT_IPV4_UDP:
2399 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2400 MLX5_L3_PROT_TYPE_IPV4);
2401 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2402 MLX5_L4_PROT_TYPE_UDP);
2403 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2404 MLX5_HASH_IP_L4PORTS);
2407 case MLX5E_TT_IPV6_UDP:
2408 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2409 MLX5_L3_PROT_TYPE_IPV6);
2410 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2411 MLX5_L4_PROT_TYPE_UDP);
2412 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2413 MLX5_HASH_IP_L4PORTS);
2416 case MLX5E_TT_IPV4_IPSEC_AH:
2417 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2418 MLX5_L3_PROT_TYPE_IPV4);
2419 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2420 MLX5_HASH_IP_IPSEC_SPI);
2423 case MLX5E_TT_IPV6_IPSEC_AH:
2424 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2425 MLX5_L3_PROT_TYPE_IPV6);
2426 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2427 MLX5_HASH_IP_IPSEC_SPI);
2430 case MLX5E_TT_IPV4_IPSEC_ESP:
2431 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2432 MLX5_L3_PROT_TYPE_IPV4);
2433 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2434 MLX5_HASH_IP_IPSEC_SPI);
2437 case MLX5E_TT_IPV6_IPSEC_ESP:
2438 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2439 MLX5_L3_PROT_TYPE_IPV6);
2440 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2441 MLX5_HASH_IP_IPSEC_SPI);
2445 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2446 MLX5_L3_PROT_TYPE_IPV4);
2447 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2452 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2453 MLX5_L3_PROT_TYPE_IPV6);
2454 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2458 WARN_ONCE(true, "%s: bad traffic type!\n", __func__);
2462 static int mlx5e_modify_tirs_lro(struct mlx5e_priv *priv)
2464 struct mlx5_core_dev *mdev = priv->mdev;
2473 inlen = MLX5_ST_SZ_BYTES(modify_tir_in);
2474 in = kvzalloc(inlen, GFP_KERNEL);
2478 MLX5_SET(modify_tir_in, in, bitmask.lro, 1);
2479 tirc = MLX5_ADDR_OF(modify_tir_in, in, ctx);
2481 mlx5e_build_tir_ctx_lro(&priv->channels.params, tirc);
2483 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
2484 err = mlx5_core_modify_tir(mdev, priv->indir_tir[tt].tirn, in,
2490 for (ix = 0; ix < priv->profile->max_nch(priv->mdev); ix++) {
2491 err = mlx5_core_modify_tir(mdev, priv->direct_tir[ix].tirn,
2503 static void mlx5e_build_inner_indir_tir_ctx(struct mlx5e_priv *priv,
2504 enum mlx5e_traffic_types tt,
2507 MLX5_SET(tirc, tirc, transport_domain, priv->mdev->mlx5e_res.td.tdn);
2509 mlx5e_build_tir_ctx_lro(&priv->channels.params, tirc);
2511 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT);
2512 MLX5_SET(tirc, tirc, indirect_table, priv->indir_rqt.rqtn);
2513 MLX5_SET(tirc, tirc, tunneled_offload_en, 0x1);
2515 mlx5e_build_indir_tir_ctx_hash(&priv->channels.params, tt, tirc, true);
2518 static int mlx5e_set_mtu(struct mlx5e_priv *priv, u16 mtu)
2520 struct mlx5_core_dev *mdev = priv->mdev;
2521 u16 hw_mtu = MLX5E_SW2HW_MTU(priv, mtu);
2524 err = mlx5_set_port_mtu(mdev, hw_mtu, 1);
2528 /* Update vport context MTU */
2529 mlx5_modify_nic_vport_mtu(mdev, hw_mtu);
2533 static void mlx5e_query_mtu(struct mlx5e_priv *priv, u16 *mtu)
2535 struct mlx5_core_dev *mdev = priv->mdev;
2539 err = mlx5_query_nic_vport_mtu(mdev, &hw_mtu);
2540 if (err || !hw_mtu) /* fallback to port oper mtu */
2541 mlx5_query_port_oper_mtu(mdev, &hw_mtu, 1);
2543 *mtu = MLX5E_HW2SW_MTU(priv, hw_mtu);
2546 static int mlx5e_set_dev_port_mtu(struct mlx5e_priv *priv)
2548 struct net_device *netdev = priv->netdev;
2552 err = mlx5e_set_mtu(priv, netdev->mtu);
2556 mlx5e_query_mtu(priv, &mtu);
2557 if (mtu != netdev->mtu)
2558 netdev_warn(netdev, "%s: VPort MTU %d is different than netdev mtu %d\n",
2559 __func__, mtu, netdev->mtu);
2565 static void mlx5e_netdev_set_tcs(struct net_device *netdev)
2567 struct mlx5e_priv *priv = netdev_priv(netdev);
2568 int nch = priv->channels.params.num_channels;
2569 int ntc = priv->channels.params.num_tc;
2572 netdev_reset_tc(netdev);
2577 netdev_set_num_tc(netdev, ntc);
2579 /* Map netdev TCs to offset 0
2580 * We have our own UP to TXQ mapping for QoS
2582 for (tc = 0; tc < ntc; tc++)
2583 netdev_set_tc_queue(netdev, tc, nch, 0);
2586 static void mlx5e_build_channels_tx_maps(struct mlx5e_priv *priv)
2588 struct mlx5e_channel *c;
2589 struct mlx5e_txqsq *sq;
2592 for (i = 0; i < priv->channels.num; i++)
2593 for (tc = 0; tc < priv->profile->max_tc; tc++)
2594 priv->channel_tc2txq[i][tc] = i + tc * priv->channels.num;
2596 for (i = 0; i < priv->channels.num; i++) {
2597 c = priv->channels.c[i];
2598 for (tc = 0; tc < c->num_tc; tc++) {
2600 priv->txq2sq[sq->txq_ix] = sq;
2605 void mlx5e_activate_priv_channels(struct mlx5e_priv *priv)
2607 int num_txqs = priv->channels.num * priv->channels.params.num_tc;
2608 struct net_device *netdev = priv->netdev;
2610 mlx5e_netdev_set_tcs(netdev);
2611 netif_set_real_num_tx_queues(netdev, num_txqs);
2612 netif_set_real_num_rx_queues(netdev, priv->channels.num);
2614 mlx5e_build_channels_tx_maps(priv);
2615 mlx5e_activate_channels(&priv->channels);
2616 netif_tx_start_all_queues(priv->netdev);
2618 if (MLX5_VPORT_MANAGER(priv->mdev))
2619 mlx5e_add_sqs_fwd_rules(priv);
2621 mlx5e_wait_channels_min_rx_wqes(&priv->channels);
2622 mlx5e_redirect_rqts_to_channels(priv, &priv->channels);
2625 void mlx5e_deactivate_priv_channels(struct mlx5e_priv *priv)
2627 mlx5e_redirect_rqts_to_drop(priv);
2629 if (MLX5_VPORT_MANAGER(priv->mdev))
2630 mlx5e_remove_sqs_fwd_rules(priv);
2632 /* FIXME: This is a W/A only for tx timeout watch dog false alarm when
2633 * polling for inactive tx queues.
2635 netif_tx_stop_all_queues(priv->netdev);
2636 netif_tx_disable(priv->netdev);
2637 mlx5e_deactivate_channels(&priv->channels);
2640 void mlx5e_switch_priv_channels(struct mlx5e_priv *priv,
2641 struct mlx5e_channels *new_chs,
2642 mlx5e_fp_hw_modify hw_modify)
2644 struct net_device *netdev = priv->netdev;
2647 new_num_txqs = new_chs->num * new_chs->params.num_tc;
2649 carrier_ok = netif_carrier_ok(netdev);
2650 netif_carrier_off(netdev);
2652 if (new_num_txqs < netdev->real_num_tx_queues)
2653 netif_set_real_num_tx_queues(netdev, new_num_txqs);
2655 mlx5e_deactivate_priv_channels(priv);
2656 mlx5e_close_channels(&priv->channels);
2658 priv->channels = *new_chs;
2660 /* New channels are ready to roll, modify HW settings if needed */
2664 mlx5e_refresh_tirs(priv, false);
2665 mlx5e_activate_priv_channels(priv);
2667 /* return carrier back if needed */
2669 netif_carrier_on(netdev);
2672 void mlx5e_timestamp_set(struct mlx5e_priv *priv)
2674 priv->tstamp.tx_type = HWTSTAMP_TX_OFF;
2675 priv->tstamp.rx_filter = HWTSTAMP_FILTER_NONE;
2678 int mlx5e_open_locked(struct net_device *netdev)
2680 struct mlx5e_priv *priv = netdev_priv(netdev);
2683 set_bit(MLX5E_STATE_OPENED, &priv->state);
2685 err = mlx5e_open_channels(priv, &priv->channels);
2687 goto err_clear_state_opened_flag;
2689 mlx5e_refresh_tirs(priv, false);
2690 mlx5e_activate_priv_channels(priv);
2691 if (priv->profile->update_carrier)
2692 priv->profile->update_carrier(priv);
2693 mlx5e_timestamp_set(priv);
2695 if (priv->profile->update_stats)
2696 queue_delayed_work(priv->wq, &priv->update_stats_work, 0);
2700 err_clear_state_opened_flag:
2701 clear_bit(MLX5E_STATE_OPENED, &priv->state);
2705 int mlx5e_open(struct net_device *netdev)
2707 struct mlx5e_priv *priv = netdev_priv(netdev);
2710 mutex_lock(&priv->state_lock);
2711 err = mlx5e_open_locked(netdev);
2713 mlx5_set_port_admin_status(priv->mdev, MLX5_PORT_UP);
2714 mutex_unlock(&priv->state_lock);
2719 int mlx5e_close_locked(struct net_device *netdev)
2721 struct mlx5e_priv *priv = netdev_priv(netdev);
2723 /* May already be CLOSED in case a previous configuration operation
2724 * (e.g RX/TX queue size change) that involves close&open failed.
2726 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
2729 clear_bit(MLX5E_STATE_OPENED, &priv->state);
2731 netif_carrier_off(priv->netdev);
2732 mlx5e_deactivate_priv_channels(priv);
2733 mlx5e_close_channels(&priv->channels);
2738 int mlx5e_close(struct net_device *netdev)
2740 struct mlx5e_priv *priv = netdev_priv(netdev);
2743 if (!netif_device_present(netdev))
2746 mutex_lock(&priv->state_lock);
2747 mlx5_set_port_admin_status(priv->mdev, MLX5_PORT_DOWN);
2748 err = mlx5e_close_locked(netdev);
2749 mutex_unlock(&priv->state_lock);
2754 static int mlx5e_alloc_drop_rq(struct mlx5_core_dev *mdev,
2755 struct mlx5e_rq *rq,
2756 struct mlx5e_rq_param *param)
2758 void *rqc = param->rqc;
2759 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
2762 param->wq.db_numa_node = param->wq.buf_numa_node;
2764 err = mlx5_wq_ll_create(mdev, ¶m->wq, rqc_wq, &rq->wq,
2774 static int mlx5e_alloc_drop_cq(struct mlx5_core_dev *mdev,
2775 struct mlx5e_cq *cq,
2776 struct mlx5e_cq_param *param)
2778 return mlx5e_alloc_cq_common(mdev, param, cq);
2781 static int mlx5e_open_drop_rq(struct mlx5_core_dev *mdev,
2782 struct mlx5e_rq *drop_rq)
2784 struct mlx5e_cq_param cq_param = {};
2785 struct mlx5e_rq_param rq_param = {};
2786 struct mlx5e_cq *cq = &drop_rq->cq;
2789 mlx5e_build_drop_rq_param(&rq_param);
2791 err = mlx5e_alloc_drop_cq(mdev, cq, &cq_param);
2795 err = mlx5e_create_cq(cq, &cq_param);
2799 err = mlx5e_alloc_drop_rq(mdev, drop_rq, &rq_param);
2801 goto err_destroy_cq;
2803 err = mlx5e_create_rq(drop_rq, &rq_param);
2810 mlx5e_free_rq(drop_rq);
2813 mlx5e_destroy_cq(cq);
2821 static void mlx5e_close_drop_rq(struct mlx5e_rq *drop_rq)
2823 mlx5e_destroy_rq(drop_rq);
2824 mlx5e_free_rq(drop_rq);
2825 mlx5e_destroy_cq(&drop_rq->cq);
2826 mlx5e_free_cq(&drop_rq->cq);
2829 int mlx5e_create_tis(struct mlx5_core_dev *mdev, int tc,
2830 u32 underlay_qpn, u32 *tisn)
2832 u32 in[MLX5_ST_SZ_DW(create_tis_in)] = {0};
2833 void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
2835 MLX5_SET(tisc, tisc, prio, tc << 1);
2836 MLX5_SET(tisc, tisc, underlay_qpn, underlay_qpn);
2837 MLX5_SET(tisc, tisc, transport_domain, mdev->mlx5e_res.td.tdn);
2839 if (mlx5_lag_is_lacp_owner(mdev))
2840 MLX5_SET(tisc, tisc, strict_lag_tx_port_affinity, 1);
2842 return mlx5_core_create_tis(mdev, in, sizeof(in), tisn);
2845 void mlx5e_destroy_tis(struct mlx5_core_dev *mdev, u32 tisn)
2847 mlx5_core_destroy_tis(mdev, tisn);
2850 int mlx5e_create_tises(struct mlx5e_priv *priv)
2855 for (tc = 0; tc < priv->profile->max_tc; tc++) {
2856 err = mlx5e_create_tis(priv->mdev, tc, 0, &priv->tisn[tc]);
2858 goto err_close_tises;
2864 for (tc--; tc >= 0; tc--)
2865 mlx5e_destroy_tis(priv->mdev, priv->tisn[tc]);
2870 void mlx5e_cleanup_nic_tx(struct mlx5e_priv *priv)
2874 for (tc = 0; tc < priv->profile->max_tc; tc++)
2875 mlx5e_destroy_tis(priv->mdev, priv->tisn[tc]);
2878 static void mlx5e_build_indir_tir_ctx(struct mlx5e_priv *priv,
2879 enum mlx5e_traffic_types tt,
2882 MLX5_SET(tirc, tirc, transport_domain, priv->mdev->mlx5e_res.td.tdn);
2884 mlx5e_build_tir_ctx_lro(&priv->channels.params, tirc);
2886 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT);
2887 MLX5_SET(tirc, tirc, indirect_table, priv->indir_rqt.rqtn);
2888 mlx5e_build_indir_tir_ctx_hash(&priv->channels.params, tt, tirc, false);
2891 static void mlx5e_build_direct_tir_ctx(struct mlx5e_priv *priv, u32 rqtn, u32 *tirc)
2893 MLX5_SET(tirc, tirc, transport_domain, priv->mdev->mlx5e_res.td.tdn);
2895 mlx5e_build_tir_ctx_lro(&priv->channels.params, tirc);
2897 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT);
2898 MLX5_SET(tirc, tirc, indirect_table, rqtn);
2899 MLX5_SET(tirc, tirc, rx_hash_fn, MLX5_RX_HASH_FN_INVERTED_XOR8);
2902 int mlx5e_create_indirect_tirs(struct mlx5e_priv *priv)
2904 struct mlx5e_tir *tir;
2912 inlen = MLX5_ST_SZ_BYTES(create_tir_in);
2913 in = kvzalloc(inlen, GFP_KERNEL);
2917 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
2918 memset(in, 0, inlen);
2919 tir = &priv->indir_tir[tt];
2920 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
2921 mlx5e_build_indir_tir_ctx(priv, tt, tirc);
2922 err = mlx5e_create_tir(priv->mdev, tir, in, inlen);
2924 mlx5_core_warn(priv->mdev, "create indirect tirs failed, %d\n", err);
2925 goto err_destroy_inner_tirs;
2929 if (!mlx5e_tunnel_inner_ft_supported(priv->mdev))
2932 for (i = 0; i < MLX5E_NUM_INDIR_TIRS; i++) {
2933 memset(in, 0, inlen);
2934 tir = &priv->inner_indir_tir[i];
2935 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
2936 mlx5e_build_inner_indir_tir_ctx(priv, i, tirc);
2937 err = mlx5e_create_tir(priv->mdev, tir, in, inlen);
2939 mlx5_core_warn(priv->mdev, "create inner indirect tirs failed, %d\n", err);
2940 goto err_destroy_inner_tirs;
2949 err_destroy_inner_tirs:
2950 for (i--; i >= 0; i--)
2951 mlx5e_destroy_tir(priv->mdev, &priv->inner_indir_tir[i]);
2953 for (tt--; tt >= 0; tt--)
2954 mlx5e_destroy_tir(priv->mdev, &priv->indir_tir[tt]);
2961 int mlx5e_create_direct_tirs(struct mlx5e_priv *priv)
2963 int nch = priv->profile->max_nch(priv->mdev);
2964 struct mlx5e_tir *tir;
2971 inlen = MLX5_ST_SZ_BYTES(create_tir_in);
2972 in = kvzalloc(inlen, GFP_KERNEL);
2976 for (ix = 0; ix < nch; ix++) {
2977 memset(in, 0, inlen);
2978 tir = &priv->direct_tir[ix];
2979 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
2980 mlx5e_build_direct_tir_ctx(priv, priv->direct_tir[ix].rqt.rqtn, tirc);
2981 err = mlx5e_create_tir(priv->mdev, tir, in, inlen);
2983 goto err_destroy_ch_tirs;
2990 err_destroy_ch_tirs:
2991 mlx5_core_warn(priv->mdev, "create direct tirs failed, %d\n", err);
2992 for (ix--; ix >= 0; ix--)
2993 mlx5e_destroy_tir(priv->mdev, &priv->direct_tir[ix]);
3000 void mlx5e_destroy_indirect_tirs(struct mlx5e_priv *priv)
3004 for (i = 0; i < MLX5E_NUM_INDIR_TIRS; i++)
3005 mlx5e_destroy_tir(priv->mdev, &priv->indir_tir[i]);
3007 if (!mlx5e_tunnel_inner_ft_supported(priv->mdev))
3010 for (i = 0; i < MLX5E_NUM_INDIR_TIRS; i++)
3011 mlx5e_destroy_tir(priv->mdev, &priv->inner_indir_tir[i]);
3014 void mlx5e_destroy_direct_tirs(struct mlx5e_priv *priv)
3016 int nch = priv->profile->max_nch(priv->mdev);
3019 for (i = 0; i < nch; i++)
3020 mlx5e_destroy_tir(priv->mdev, &priv->direct_tir[i]);
3023 static int mlx5e_modify_channels_scatter_fcs(struct mlx5e_channels *chs, bool enable)
3028 for (i = 0; i < chs->num; i++) {
3029 err = mlx5e_modify_rq_scatter_fcs(&chs->c[i]->rq, enable);
3037 static int mlx5e_modify_channels_vsd(struct mlx5e_channels *chs, bool vsd)
3042 for (i = 0; i < chs->num; i++) {
3043 err = mlx5e_modify_rq_vsd(&chs->c[i]->rq, vsd);
3051 static int mlx5e_setup_tc_mqprio(struct net_device *netdev,
3052 struct tc_mqprio_qopt *mqprio)
3054 struct mlx5e_priv *priv = netdev_priv(netdev);
3055 struct mlx5e_channels new_channels = {};
3056 u8 tc = mqprio->num_tc;
3059 mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
3061 if (tc && tc != MLX5E_MAX_NUM_TC)
3064 mutex_lock(&priv->state_lock);
3066 new_channels.params = priv->channels.params;
3067 new_channels.params.num_tc = tc ? tc : 1;
3069 if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
3070 priv->channels.params = new_channels.params;
3074 err = mlx5e_open_channels(priv, &new_channels);
3078 mlx5e_switch_priv_channels(priv, &new_channels, NULL);
3080 mutex_unlock(&priv->state_lock);
3084 #ifdef CONFIG_MLX5_ESWITCH
3085 static int mlx5e_setup_tc_cls_flower(struct mlx5e_priv *priv,
3086 struct tc_cls_flower_offload *cls_flower)
3088 if (cls_flower->common.chain_index)
3091 switch (cls_flower->command) {
3092 case TC_CLSFLOWER_REPLACE:
3093 return mlx5e_configure_flower(priv, cls_flower);
3094 case TC_CLSFLOWER_DESTROY:
3095 return mlx5e_delete_flower(priv, cls_flower);
3096 case TC_CLSFLOWER_STATS:
3097 return mlx5e_stats_flower(priv, cls_flower);
3103 int mlx5e_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
3106 struct mlx5e_priv *priv = cb_priv;
3108 if (!tc_can_offload(priv->netdev))
3112 case TC_SETUP_CLSFLOWER:
3113 return mlx5e_setup_tc_cls_flower(priv, type_data);
3119 static int mlx5e_setup_tc_block(struct net_device *dev,
3120 struct tc_block_offload *f)
3122 struct mlx5e_priv *priv = netdev_priv(dev);
3124 if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
3127 switch (f->command) {
3129 return tcf_block_cb_register(f->block, mlx5e_setup_tc_block_cb,
3131 case TC_BLOCK_UNBIND:
3132 tcf_block_cb_unregister(f->block, mlx5e_setup_tc_block_cb,
3141 int mlx5e_setup_tc(struct net_device *dev, enum tc_setup_type type,
3145 #ifdef CONFIG_MLX5_ESWITCH
3146 case TC_SETUP_BLOCK:
3147 return mlx5e_setup_tc_block(dev, type_data);
3149 case TC_SETUP_QDISC_MQPRIO:
3150 return mlx5e_setup_tc_mqprio(dev, type_data);
3157 mlx5e_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats)
3159 struct mlx5e_priv *priv = netdev_priv(dev);
3160 struct mlx5e_sw_stats *sstats = &priv->stats.sw;
3161 struct mlx5e_vport_stats *vstats = &priv->stats.vport;
3162 struct mlx5e_pport_stats *pstats = &priv->stats.pport;
3164 if (mlx5e_is_uplink_rep(priv)) {
3165 stats->rx_packets = PPORT_802_3_GET(pstats, a_frames_received_ok);
3166 stats->rx_bytes = PPORT_802_3_GET(pstats, a_octets_received_ok);
3167 stats->tx_packets = PPORT_802_3_GET(pstats, a_frames_transmitted_ok);
3168 stats->tx_bytes = PPORT_802_3_GET(pstats, a_octets_transmitted_ok);
3170 stats->rx_packets = sstats->rx_packets;
3171 stats->rx_bytes = sstats->rx_bytes;
3172 stats->tx_packets = sstats->tx_packets;
3173 stats->tx_bytes = sstats->tx_bytes;
3174 stats->tx_dropped = sstats->tx_queue_dropped;
3177 stats->rx_dropped = priv->stats.qcnt.rx_out_of_buffer;
3179 stats->rx_length_errors =
3180 PPORT_802_3_GET(pstats, a_in_range_length_errors) +
3181 PPORT_802_3_GET(pstats, a_out_of_range_length_field) +
3182 PPORT_802_3_GET(pstats, a_frame_too_long_errors);
3183 stats->rx_crc_errors =
3184 PPORT_802_3_GET(pstats, a_frame_check_sequence_errors);
3185 stats->rx_frame_errors = PPORT_802_3_GET(pstats, a_alignment_errors);
3186 stats->tx_aborted_errors = PPORT_2863_GET(pstats, if_out_discards);
3187 stats->rx_errors = stats->rx_length_errors + stats->rx_crc_errors +
3188 stats->rx_frame_errors;
3189 stats->tx_errors = stats->tx_aborted_errors + stats->tx_carrier_errors;
3191 /* vport multicast also counts packets that are dropped due to steering
3192 * or rx out of buffer
3195 VPORT_COUNTER_GET(vstats, received_eth_multicast.packets);
3198 static void mlx5e_set_rx_mode(struct net_device *dev)
3200 struct mlx5e_priv *priv = netdev_priv(dev);
3202 queue_work(priv->wq, &priv->set_rx_mode_work);
3205 static int mlx5e_set_mac(struct net_device *netdev, void *addr)
3207 struct mlx5e_priv *priv = netdev_priv(netdev);
3208 struct sockaddr *saddr = addr;
3210 if (!is_valid_ether_addr(saddr->sa_data))
3211 return -EADDRNOTAVAIL;
3213 netif_addr_lock_bh(netdev);
3214 ether_addr_copy(netdev->dev_addr, saddr->sa_data);
3215 netif_addr_unlock_bh(netdev);
3217 queue_work(priv->wq, &priv->set_rx_mode_work);
3222 #define MLX5E_SET_FEATURE(netdev, feature, enable) \
3225 netdev->features |= feature; \
3227 netdev->features &= ~feature; \
3230 typedef int (*mlx5e_feature_handler)(struct net_device *netdev, bool enable);
3232 static int set_feature_lro(struct net_device *netdev, bool enable)
3234 struct mlx5e_priv *priv = netdev_priv(netdev);
3235 struct mlx5e_channels new_channels = {};
3239 mutex_lock(&priv->state_lock);
3241 reset = (priv->channels.params.rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST);
3242 reset = reset && test_bit(MLX5E_STATE_OPENED, &priv->state);
3244 new_channels.params = priv->channels.params;
3245 new_channels.params.lro_en = enable;
3248 priv->channels.params = new_channels.params;
3249 err = mlx5e_modify_tirs_lro(priv);
3253 err = mlx5e_open_channels(priv, &new_channels);
3257 mlx5e_switch_priv_channels(priv, &new_channels, mlx5e_modify_tirs_lro);
3259 mutex_unlock(&priv->state_lock);
3263 static int set_feature_cvlan_filter(struct net_device *netdev, bool enable)
3265 struct mlx5e_priv *priv = netdev_priv(netdev);
3268 mlx5e_enable_cvlan_filter(priv);
3270 mlx5e_disable_cvlan_filter(priv);
3275 static int set_feature_tc_num_filters(struct net_device *netdev, bool enable)
3277 struct mlx5e_priv *priv = netdev_priv(netdev);
3279 if (!enable && mlx5e_tc_num_filters(priv)) {
3281 "Active offloaded tc filters, can't turn hw_tc_offload off\n");
3288 static int set_feature_rx_all(struct net_device *netdev, bool enable)
3290 struct mlx5e_priv *priv = netdev_priv(netdev);
3291 struct mlx5_core_dev *mdev = priv->mdev;
3293 return mlx5_set_port_fcs(mdev, !enable);
3296 static int set_feature_rx_fcs(struct net_device *netdev, bool enable)
3298 struct mlx5e_priv *priv = netdev_priv(netdev);
3301 mutex_lock(&priv->state_lock);
3303 priv->channels.params.scatter_fcs_en = enable;
3304 err = mlx5e_modify_channels_scatter_fcs(&priv->channels, enable);
3306 priv->channels.params.scatter_fcs_en = !enable;
3308 mutex_unlock(&priv->state_lock);
3313 static int set_feature_rx_vlan(struct net_device *netdev, bool enable)
3315 struct mlx5e_priv *priv = netdev_priv(netdev);
3318 mutex_lock(&priv->state_lock);
3320 priv->channels.params.vlan_strip_disable = !enable;
3321 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
3324 err = mlx5e_modify_channels_vsd(&priv->channels, !enable);
3326 priv->channels.params.vlan_strip_disable = enable;
3329 mutex_unlock(&priv->state_lock);
3334 #ifdef CONFIG_RFS_ACCEL
3335 static int set_feature_arfs(struct net_device *netdev, bool enable)
3337 struct mlx5e_priv *priv = netdev_priv(netdev);
3341 err = mlx5e_arfs_enable(priv);
3343 err = mlx5e_arfs_disable(priv);
3349 static int mlx5e_handle_feature(struct net_device *netdev,
3350 netdev_features_t wanted_features,
3351 netdev_features_t feature,
3352 mlx5e_feature_handler feature_handler)
3354 netdev_features_t changes = wanted_features ^ netdev->features;
3355 bool enable = !!(wanted_features & feature);
3358 if (!(changes & feature))
3361 err = feature_handler(netdev, enable);
3363 netdev_err(netdev, "%s feature %pNF failed, err %d\n",
3364 enable ? "Enable" : "Disable", &feature, err);
3368 MLX5E_SET_FEATURE(netdev, feature, enable);
3372 static int mlx5e_set_features(struct net_device *netdev,
3373 netdev_features_t features)
3377 err = mlx5e_handle_feature(netdev, features, NETIF_F_LRO,
3379 err |= mlx5e_handle_feature(netdev, features,
3380 NETIF_F_HW_VLAN_CTAG_FILTER,
3381 set_feature_cvlan_filter);
3382 err |= mlx5e_handle_feature(netdev, features, NETIF_F_HW_TC,
3383 set_feature_tc_num_filters);
3384 err |= mlx5e_handle_feature(netdev, features, NETIF_F_RXALL,
3385 set_feature_rx_all);
3386 err |= mlx5e_handle_feature(netdev, features, NETIF_F_RXFCS,
3387 set_feature_rx_fcs);
3388 err |= mlx5e_handle_feature(netdev, features, NETIF_F_HW_VLAN_CTAG_RX,
3389 set_feature_rx_vlan);
3390 #ifdef CONFIG_RFS_ACCEL
3391 err |= mlx5e_handle_feature(netdev, features, NETIF_F_NTUPLE,
3395 return err ? -EINVAL : 0;
3398 static netdev_features_t mlx5e_fix_features(struct net_device *netdev,
3399 netdev_features_t features)
3401 struct mlx5e_priv *priv = netdev_priv(netdev);
3403 mutex_lock(&priv->state_lock);
3404 if (!bitmap_empty(priv->fs.vlan.active_svlans, VLAN_N_VID)) {
3405 /* HW strips the outer C-tag header, this is a problem
3406 * for S-tag traffic.
3408 features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3409 if (!priv->channels.params.vlan_strip_disable)
3410 netdev_warn(netdev, "Dropping C-tag vlan stripping offload due to S-tag vlan\n");
3412 mutex_unlock(&priv->state_lock);
3417 static int mlx5e_change_mtu(struct net_device *netdev, int new_mtu)
3419 struct mlx5e_priv *priv = netdev_priv(netdev);
3420 struct mlx5e_channels new_channels = {};
3425 mutex_lock(&priv->state_lock);
3427 reset = !priv->channels.params.lro_en &&
3428 (priv->channels.params.rq_wq_type !=
3429 MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ);
3431 reset = reset && test_bit(MLX5E_STATE_OPENED, &priv->state);
3433 curr_mtu = netdev->mtu;
3434 netdev->mtu = new_mtu;
3437 mlx5e_set_dev_port_mtu(priv);
3441 new_channels.params = priv->channels.params;
3442 err = mlx5e_open_channels(priv, &new_channels);
3444 netdev->mtu = curr_mtu;
3448 mlx5e_switch_priv_channels(priv, &new_channels, mlx5e_set_dev_port_mtu);
3451 mutex_unlock(&priv->state_lock);
3455 int mlx5e_hwstamp_set(struct mlx5e_priv *priv, struct ifreq *ifr)
3457 struct hwtstamp_config config;
3460 if (!MLX5_CAP_GEN(priv->mdev, device_frequency_khz))
3463 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
3466 /* TX HW timestamp */
3467 switch (config.tx_type) {
3468 case HWTSTAMP_TX_OFF:
3469 case HWTSTAMP_TX_ON:
3475 mutex_lock(&priv->state_lock);
3476 /* RX HW timestamp */
3477 switch (config.rx_filter) {
3478 case HWTSTAMP_FILTER_NONE:
3479 /* Reset CQE compression to Admin default */
3480 mlx5e_modify_rx_cqe_compression_locked(priv, priv->channels.params.rx_cqe_compress_def);
3482 case HWTSTAMP_FILTER_ALL:
3483 case HWTSTAMP_FILTER_SOME:
3484 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3485 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3486 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3487 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3488 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3489 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3490 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3491 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3492 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3493 case HWTSTAMP_FILTER_PTP_V2_EVENT:
3494 case HWTSTAMP_FILTER_PTP_V2_SYNC:
3495 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3496 case HWTSTAMP_FILTER_NTP_ALL:
3497 /* Disable CQE compression */
3498 netdev_warn(priv->netdev, "Disabling cqe compression");
3499 err = mlx5e_modify_rx_cqe_compression_locked(priv, false);
3501 netdev_err(priv->netdev, "Failed disabling cqe compression err=%d\n", err);
3502 mutex_unlock(&priv->state_lock);
3505 config.rx_filter = HWTSTAMP_FILTER_ALL;
3508 mutex_unlock(&priv->state_lock);
3512 memcpy(&priv->tstamp, &config, sizeof(config));
3513 mutex_unlock(&priv->state_lock);
3515 return copy_to_user(ifr->ifr_data, &config,
3516 sizeof(config)) ? -EFAULT : 0;
3519 int mlx5e_hwstamp_get(struct mlx5e_priv *priv, struct ifreq *ifr)
3521 struct hwtstamp_config *cfg = &priv->tstamp;
3523 if (!MLX5_CAP_GEN(priv->mdev, device_frequency_khz))
3526 return copy_to_user(ifr->ifr_data, cfg, sizeof(*cfg)) ? -EFAULT : 0;
3529 static int mlx5e_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
3531 struct mlx5e_priv *priv = netdev_priv(dev);
3535 return mlx5e_hwstamp_set(priv, ifr);
3537 return mlx5e_hwstamp_get(priv, ifr);
3543 #ifdef CONFIG_MLX5_ESWITCH
3544 static int mlx5e_set_vf_mac(struct net_device *dev, int vf, u8 *mac)
3546 struct mlx5e_priv *priv = netdev_priv(dev);
3547 struct mlx5_core_dev *mdev = priv->mdev;
3549 return mlx5_eswitch_set_vport_mac(mdev->priv.eswitch, vf + 1, mac);
3552 static int mlx5e_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
3555 struct mlx5e_priv *priv = netdev_priv(dev);
3556 struct mlx5_core_dev *mdev = priv->mdev;
3558 if (vlan_proto != htons(ETH_P_8021Q))
3559 return -EPROTONOSUPPORT;
3561 return mlx5_eswitch_set_vport_vlan(mdev->priv.eswitch, vf + 1,
3565 static int mlx5e_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
3567 struct mlx5e_priv *priv = netdev_priv(dev);
3568 struct mlx5_core_dev *mdev = priv->mdev;
3570 return mlx5_eswitch_set_vport_spoofchk(mdev->priv.eswitch, vf + 1, setting);
3573 static int mlx5e_set_vf_trust(struct net_device *dev, int vf, bool setting)
3575 struct mlx5e_priv *priv = netdev_priv(dev);
3576 struct mlx5_core_dev *mdev = priv->mdev;
3578 return mlx5_eswitch_set_vport_trust(mdev->priv.eswitch, vf + 1, setting);
3581 static int mlx5e_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
3584 struct mlx5e_priv *priv = netdev_priv(dev);
3585 struct mlx5_core_dev *mdev = priv->mdev;
3587 return mlx5_eswitch_set_vport_rate(mdev->priv.eswitch, vf + 1,
3588 max_tx_rate, min_tx_rate);
3591 static int mlx5_vport_link2ifla(u8 esw_link)
3594 case MLX5_ESW_VPORT_ADMIN_STATE_DOWN:
3595 return IFLA_VF_LINK_STATE_DISABLE;
3596 case MLX5_ESW_VPORT_ADMIN_STATE_UP:
3597 return IFLA_VF_LINK_STATE_ENABLE;
3599 return IFLA_VF_LINK_STATE_AUTO;
3602 static int mlx5_ifla_link2vport(u8 ifla_link)
3604 switch (ifla_link) {
3605 case IFLA_VF_LINK_STATE_DISABLE:
3606 return MLX5_ESW_VPORT_ADMIN_STATE_DOWN;
3607 case IFLA_VF_LINK_STATE_ENABLE:
3608 return MLX5_ESW_VPORT_ADMIN_STATE_UP;
3610 return MLX5_ESW_VPORT_ADMIN_STATE_AUTO;
3613 static int mlx5e_set_vf_link_state(struct net_device *dev, int vf,
3616 struct mlx5e_priv *priv = netdev_priv(dev);
3617 struct mlx5_core_dev *mdev = priv->mdev;
3619 return mlx5_eswitch_set_vport_state(mdev->priv.eswitch, vf + 1,
3620 mlx5_ifla_link2vport(link_state));
3623 static int mlx5e_get_vf_config(struct net_device *dev,
3624 int vf, struct ifla_vf_info *ivi)
3626 struct mlx5e_priv *priv = netdev_priv(dev);
3627 struct mlx5_core_dev *mdev = priv->mdev;
3630 err = mlx5_eswitch_get_vport_config(mdev->priv.eswitch, vf + 1, ivi);
3633 ivi->linkstate = mlx5_vport_link2ifla(ivi->linkstate);
3637 static int mlx5e_get_vf_stats(struct net_device *dev,
3638 int vf, struct ifla_vf_stats *vf_stats)
3640 struct mlx5e_priv *priv = netdev_priv(dev);
3641 struct mlx5_core_dev *mdev = priv->mdev;
3643 return mlx5_eswitch_get_vport_stats(mdev->priv.eswitch, vf + 1,
3648 static void mlx5e_add_vxlan_port(struct net_device *netdev,
3649 struct udp_tunnel_info *ti)
3651 struct mlx5e_priv *priv = netdev_priv(netdev);
3653 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3656 if (!mlx5e_vxlan_allowed(priv->mdev))
3659 mlx5e_vxlan_queue_work(priv, ti->sa_family, be16_to_cpu(ti->port), 1);
3662 static void mlx5e_del_vxlan_port(struct net_device *netdev,
3663 struct udp_tunnel_info *ti)
3665 struct mlx5e_priv *priv = netdev_priv(netdev);
3667 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3670 if (!mlx5e_vxlan_allowed(priv->mdev))
3673 mlx5e_vxlan_queue_work(priv, ti->sa_family, be16_to_cpu(ti->port), 0);
3676 static netdev_features_t mlx5e_tunnel_features_check(struct mlx5e_priv *priv,
3677 struct sk_buff *skb,
3678 netdev_features_t features)
3680 unsigned int offset = 0;
3681 struct udphdr *udph;
3685 switch (vlan_get_protocol(skb)) {
3686 case htons(ETH_P_IP):
3687 proto = ip_hdr(skb)->protocol;
3689 case htons(ETH_P_IPV6):
3690 proto = ipv6_find_hdr(skb, &offset, -1, NULL, NULL);
3700 udph = udp_hdr(skb);
3701 port = be16_to_cpu(udph->dest);
3703 /* Verify if UDP port is being offloaded by HW */
3704 if (mlx5e_vxlan_lookup_port(priv, port))
3709 /* Disable CSUM and GSO if the udp dport is not offloaded by HW */
3710 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3713 static netdev_features_t mlx5e_features_check(struct sk_buff *skb,
3714 struct net_device *netdev,
3715 netdev_features_t features)
3717 struct mlx5e_priv *priv = netdev_priv(netdev);
3719 features = vlan_features_check(skb, features);
3720 features = vxlan_features_check(skb, features);
3722 #ifdef CONFIG_MLX5_EN_IPSEC
3723 if (mlx5e_ipsec_feature_check(skb, netdev, features))
3727 /* Validate if the tunneled packet is being offloaded by HW */
3728 if (skb->encapsulation &&
3729 (features & NETIF_F_CSUM_MASK || features & NETIF_F_GSO_MASK))
3730 return mlx5e_tunnel_features_check(priv, skb, features);
3735 static void mlx5e_tx_timeout(struct net_device *dev)
3737 struct mlx5e_priv *priv = netdev_priv(dev);
3738 bool sched_work = false;
3741 netdev_err(dev, "TX timeout detected\n");
3743 for (i = 0; i < priv->channels.num * priv->channels.params.num_tc; i++) {
3744 struct mlx5e_txqsq *sq = priv->txq2sq[i];
3746 if (!netif_xmit_stopped(netdev_get_tx_queue(dev, i)))
3749 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
3750 netdev_err(dev, "TX timeout on queue: %d, SQ: 0x%x, CQ: 0x%x, SQ Cons: 0x%x SQ Prod: 0x%x\n",
3751 i, sq->sqn, sq->cq.mcq.cqn, sq->cc, sq->pc);
3754 if (sched_work && test_bit(MLX5E_STATE_OPENED, &priv->state))
3755 schedule_work(&priv->tx_timeout_work);
3758 static int mlx5e_xdp_set(struct net_device *netdev, struct bpf_prog *prog)
3760 struct mlx5e_priv *priv = netdev_priv(netdev);
3761 struct bpf_prog *old_prog;
3763 bool reset, was_opened;
3766 mutex_lock(&priv->state_lock);
3768 if ((netdev->features & NETIF_F_LRO) && prog) {
3769 netdev_warn(netdev, "can't set XDP while LRO is on, disable LRO first\n");
3774 if ((netdev->features & NETIF_F_HW_ESP) && prog) {
3775 netdev_warn(netdev, "can't set XDP with IPSec offload\n");
3780 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
3781 /* no need for full reset when exchanging programs */
3782 reset = (!priv->channels.params.xdp_prog || !prog);
3784 if (was_opened && reset)
3785 mlx5e_close_locked(netdev);
3786 if (was_opened && !reset) {
3787 /* num_channels is invariant here, so we can take the
3788 * batched reference right upfront.
3790 prog = bpf_prog_add(prog, priv->channels.num);
3792 err = PTR_ERR(prog);
3797 /* exchange programs, extra prog reference we got from caller
3798 * as long as we don't fail from this point onwards.
3800 old_prog = xchg(&priv->channels.params.xdp_prog, prog);
3802 bpf_prog_put(old_prog);
3804 if (reset) /* change RQ type according to priv->xdp_prog */
3805 mlx5e_set_rq_params(priv->mdev, &priv->channels.params);
3807 if (was_opened && reset)
3808 mlx5e_open_locked(netdev);
3810 if (!test_bit(MLX5E_STATE_OPENED, &priv->state) || reset)
3813 /* exchanging programs w/o reset, we update ref counts on behalf
3814 * of the channels RQs here.
3816 for (i = 0; i < priv->channels.num; i++) {
3817 struct mlx5e_channel *c = priv->channels.c[i];
3819 clear_bit(MLX5E_RQ_STATE_ENABLED, &c->rq.state);
3820 napi_synchronize(&c->napi);
3821 /* prevent mlx5e_poll_rx_cq from accessing rq->xdp_prog */
3823 old_prog = xchg(&c->rq.xdp_prog, prog);
3825 set_bit(MLX5E_RQ_STATE_ENABLED, &c->rq.state);
3826 /* napi_schedule in case we have missed anything */
3827 napi_schedule(&c->napi);
3830 bpf_prog_put(old_prog);
3834 mutex_unlock(&priv->state_lock);
3838 static u32 mlx5e_xdp_query(struct net_device *dev)
3840 struct mlx5e_priv *priv = netdev_priv(dev);
3841 const struct bpf_prog *xdp_prog;
3844 mutex_lock(&priv->state_lock);
3845 xdp_prog = priv->channels.params.xdp_prog;
3847 prog_id = xdp_prog->aux->id;
3848 mutex_unlock(&priv->state_lock);
3853 static int mlx5e_xdp(struct net_device *dev, struct netdev_bpf *xdp)
3855 switch (xdp->command) {
3856 case XDP_SETUP_PROG:
3857 return mlx5e_xdp_set(dev, xdp->prog);
3858 case XDP_QUERY_PROG:
3859 xdp->prog_id = mlx5e_xdp_query(dev);
3860 xdp->prog_attached = !!xdp->prog_id;
3867 #ifdef CONFIG_NET_POLL_CONTROLLER
3868 /* Fake "interrupt" called by netpoll (eg netconsole) to send skbs without
3869 * reenabling interrupts.
3871 static void mlx5e_netpoll(struct net_device *dev)
3873 struct mlx5e_priv *priv = netdev_priv(dev);
3874 struct mlx5e_channels *chs = &priv->channels;
3878 for (i = 0; i < chs->num; i++)
3879 napi_schedule(&chs->c[i]->napi);
3883 static const struct net_device_ops mlx5e_netdev_ops = {
3884 .ndo_open = mlx5e_open,
3885 .ndo_stop = mlx5e_close,
3886 .ndo_start_xmit = mlx5e_xmit,
3887 .ndo_setup_tc = mlx5e_setup_tc,
3888 .ndo_select_queue = mlx5e_select_queue,
3889 .ndo_get_stats64 = mlx5e_get_stats,
3890 .ndo_set_rx_mode = mlx5e_set_rx_mode,
3891 .ndo_set_mac_address = mlx5e_set_mac,
3892 .ndo_vlan_rx_add_vid = mlx5e_vlan_rx_add_vid,
3893 .ndo_vlan_rx_kill_vid = mlx5e_vlan_rx_kill_vid,
3894 .ndo_set_features = mlx5e_set_features,
3895 .ndo_fix_features = mlx5e_fix_features,
3896 .ndo_change_mtu = mlx5e_change_mtu,
3897 .ndo_do_ioctl = mlx5e_ioctl,
3898 .ndo_set_tx_maxrate = mlx5e_set_tx_maxrate,
3899 .ndo_udp_tunnel_add = mlx5e_add_vxlan_port,
3900 .ndo_udp_tunnel_del = mlx5e_del_vxlan_port,
3901 .ndo_features_check = mlx5e_features_check,
3902 #ifdef CONFIG_RFS_ACCEL
3903 .ndo_rx_flow_steer = mlx5e_rx_flow_steer,
3905 .ndo_tx_timeout = mlx5e_tx_timeout,
3906 .ndo_bpf = mlx5e_xdp,
3907 #ifdef CONFIG_NET_POLL_CONTROLLER
3908 .ndo_poll_controller = mlx5e_netpoll,
3910 #ifdef CONFIG_MLX5_ESWITCH
3911 /* SRIOV E-Switch NDOs */
3912 .ndo_set_vf_mac = mlx5e_set_vf_mac,
3913 .ndo_set_vf_vlan = mlx5e_set_vf_vlan,
3914 .ndo_set_vf_spoofchk = mlx5e_set_vf_spoofchk,
3915 .ndo_set_vf_trust = mlx5e_set_vf_trust,
3916 .ndo_set_vf_rate = mlx5e_set_vf_rate,
3917 .ndo_get_vf_config = mlx5e_get_vf_config,
3918 .ndo_set_vf_link_state = mlx5e_set_vf_link_state,
3919 .ndo_get_vf_stats = mlx5e_get_vf_stats,
3920 .ndo_has_offload_stats = mlx5e_has_offload_stats,
3921 .ndo_get_offload_stats = mlx5e_get_offload_stats,
3925 static int mlx5e_check_required_hca_cap(struct mlx5_core_dev *mdev)
3927 if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
3929 if (!MLX5_CAP_GEN(mdev, eth_net_offloads) ||
3930 !MLX5_CAP_GEN(mdev, nic_flow_table) ||
3931 !MLX5_CAP_ETH(mdev, csum_cap) ||
3932 !MLX5_CAP_ETH(mdev, max_lso_cap) ||
3933 !MLX5_CAP_ETH(mdev, vlan_cap) ||
3934 !MLX5_CAP_ETH(mdev, rss_ind_tbl_cap) ||
3935 MLX5_CAP_FLOWTABLE(mdev,
3936 flow_table_properties_nic_receive.max_ft_level)
3938 mlx5_core_warn(mdev,
3939 "Not creating net device, some required device capabilities are missing\n");
3942 if (!MLX5_CAP_ETH(mdev, self_lb_en_modifiable))
3943 mlx5_core_warn(mdev, "Self loop back prevention is not supported\n");
3944 if (!MLX5_CAP_GEN(mdev, cq_moderation))
3945 mlx5_core_warn(mdev, "CQ moderation is not supported\n");
3950 u16 mlx5e_get_max_inline_cap(struct mlx5_core_dev *mdev)
3952 int bf_buf_size = (1 << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2;
3954 return bf_buf_size -
3955 sizeof(struct mlx5e_tx_wqe) +
3956 2 /*sizeof(mlx5e_tx_wqe.inline_hdr_start)*/;
3959 void mlx5e_build_default_indir_rqt(u32 *indirection_rqt, int len,
3964 for (i = 0; i < len; i++)
3965 indirection_rqt[i] = i % num_channels;
3968 static int mlx5e_get_pci_bw(struct mlx5_core_dev *mdev, u32 *pci_bw)
3970 enum pcie_link_width width;
3971 enum pci_bus_speed speed;
3974 err = pcie_get_minimum_link(mdev->pdev, &speed, &width);
3978 if (speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN)
3982 case PCIE_SPEED_2_5GT:
3983 *pci_bw = 2500 * width;
3985 case PCIE_SPEED_5_0GT:
3986 *pci_bw = 5000 * width;
3988 case PCIE_SPEED_8_0GT:
3989 *pci_bw = 8000 * width;
3998 static bool cqe_compress_heuristic(u32 link_speed, u32 pci_bw)
4000 return (link_speed && pci_bw &&
4001 (pci_bw < 40000) && (pci_bw < link_speed));
4004 static bool hw_lro_heuristic(u32 link_speed, u32 pci_bw)
4006 return !(link_speed && pci_bw &&
4007 (pci_bw <= 16000) && (pci_bw < link_speed));
4010 void mlx5e_set_tx_cq_mode_params(struct mlx5e_params *params, u8 cq_period_mode)
4012 params->tx_cq_moderation.cq_period_mode = cq_period_mode;
4014 params->tx_cq_moderation.pkts =
4015 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
4016 params->tx_cq_moderation.usec =
4017 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC;
4019 if (cq_period_mode == MLX5_CQ_PERIOD_MODE_START_FROM_CQE)
4020 params->tx_cq_moderation.usec =
4021 MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC_FROM_CQE;
4023 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_TX_CQE_BASED_MODER,
4024 params->tx_cq_moderation.cq_period_mode ==
4025 MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
4028 void mlx5e_set_rx_cq_mode_params(struct mlx5e_params *params, u8 cq_period_mode)
4030 params->rx_cq_moderation.cq_period_mode = cq_period_mode;
4032 params->rx_cq_moderation.pkts =
4033 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS;
4034 params->rx_cq_moderation.usec =
4035 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
4037 if (cq_period_mode == MLX5_CQ_PERIOD_MODE_START_FROM_CQE)
4038 params->rx_cq_moderation.usec =
4039 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE;
4041 if (params->rx_am_enabled)
4042 params->rx_cq_moderation =
4043 mlx5e_am_get_def_profile(cq_period_mode);
4045 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_CQE_BASED_MODER,
4046 params->rx_cq_moderation.cq_period_mode ==
4047 MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
4050 u32 mlx5e_choose_lro_timeout(struct mlx5_core_dev *mdev, u32 wanted_timeout)
4054 /* The supported periods are organized in ascending order */
4055 for (i = 0; i < MLX5E_LRO_TIMEOUT_ARR_SIZE - 1; i++)
4056 if (MLX5_CAP_ETH(mdev, lro_timer_supported_periods[i]) >= wanted_timeout)
4059 return MLX5_CAP_ETH(mdev, lro_timer_supported_periods[i]);
4062 void mlx5e_build_nic_params(struct mlx5_core_dev *mdev,
4063 struct mlx5e_params *params,
4066 u8 cq_period_mode = 0;
4070 params->num_channels = max_channels;
4073 mlx5e_get_max_linkspeed(mdev, &link_speed);
4074 mlx5e_get_pci_bw(mdev, &pci_bw);
4075 mlx5_core_dbg(mdev, "Max link speed = %d, PCI BW = %d\n",
4076 link_speed, pci_bw);
4079 params->log_sq_size = is_kdump_kernel() ?
4080 MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE :
4081 MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
4083 /* set CQE compression */
4084 params->rx_cqe_compress_def = false;
4085 if (MLX5_CAP_GEN(mdev, cqe_compression) &&
4086 MLX5_CAP_GEN(mdev, vport_group_manager))
4087 params->rx_cqe_compress_def = cqe_compress_heuristic(link_speed, pci_bw);
4089 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS, params->rx_cqe_compress_def);
4092 mlx5e_set_rq_params(mdev, params);
4096 /* TODO: && MLX5_CAP_ETH(mdev, lro_cap) */
4097 if (params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ)
4098 params->lro_en = hw_lro_heuristic(link_speed, pci_bw);
4099 params->lro_timeout = mlx5e_choose_lro_timeout(mdev, MLX5E_DEFAULT_LRO_TIMEOUT);
4101 /* CQ moderation params */
4102 cq_period_mode = MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
4103 MLX5_CQ_PERIOD_MODE_START_FROM_CQE :
4104 MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
4105 params->rx_am_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
4106 mlx5e_set_rx_cq_mode_params(params, cq_period_mode);
4107 mlx5e_set_tx_cq_mode_params(params, cq_period_mode);
4110 params->tx_max_inline = mlx5e_get_max_inline_cap(mdev);
4111 params->tx_min_inline_mode = mlx5e_params_calculate_tx_min_inline(mdev);
4114 params->rss_hfunc = ETH_RSS_HASH_XOR;
4115 netdev_rss_key_fill(params->toeplitz_hash_key, sizeof(params->toeplitz_hash_key));
4116 mlx5e_build_default_indir_rqt(params->indirection_rqt,
4117 MLX5E_INDIR_RQT_SIZE, max_channels);
4120 static void mlx5e_build_nic_netdev_priv(struct mlx5_core_dev *mdev,
4121 struct net_device *netdev,
4122 const struct mlx5e_profile *profile,
4125 struct mlx5e_priv *priv = netdev_priv(netdev);
4128 priv->netdev = netdev;
4129 priv->profile = profile;
4130 priv->ppriv = ppriv;
4131 priv->msglevel = MLX5E_MSG_LEVEL;
4132 priv->hard_mtu = MLX5E_ETH_HARD_MTU;
4134 mlx5e_build_nic_params(mdev, &priv->channels.params, profile->max_nch(mdev));
4136 mutex_init(&priv->state_lock);
4138 INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
4139 INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
4140 INIT_WORK(&priv->tx_timeout_work, mlx5e_tx_timeout_work);
4141 INIT_DELAYED_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
4144 static void mlx5e_set_netdev_dev_addr(struct net_device *netdev)
4146 struct mlx5e_priv *priv = netdev_priv(netdev);
4148 mlx5_query_nic_vport_mac_address(priv->mdev, 0, netdev->dev_addr);
4149 if (is_zero_ether_addr(netdev->dev_addr) &&
4150 !MLX5_CAP_GEN(priv->mdev, vport_group_manager)) {
4151 eth_hw_addr_random(netdev);
4152 mlx5_core_info(priv->mdev, "Assigned random MAC address %pM\n", netdev->dev_addr);
4156 #if IS_ENABLED(CONFIG_NET_SWITCHDEV) && IS_ENABLED(CONFIG_MLX5_ESWITCH)
4157 static const struct switchdev_ops mlx5e_switchdev_ops = {
4158 .switchdev_port_attr_get = mlx5e_attr_get,
4162 static void mlx5e_build_nic_netdev(struct net_device *netdev)
4164 struct mlx5e_priv *priv = netdev_priv(netdev);
4165 struct mlx5_core_dev *mdev = priv->mdev;
4169 SET_NETDEV_DEV(netdev, &mdev->pdev->dev);
4171 netdev->netdev_ops = &mlx5e_netdev_ops;
4173 #ifdef CONFIG_MLX5_CORE_EN_DCB
4174 if (MLX5_CAP_GEN(mdev, vport_group_manager) && MLX5_CAP_GEN(mdev, qos))
4175 netdev->dcbnl_ops = &mlx5e_dcbnl_ops;
4178 netdev->watchdog_timeo = 15 * HZ;
4180 netdev->ethtool_ops = &mlx5e_ethtool_ops;
4182 netdev->vlan_features |= NETIF_F_SG;
4183 netdev->vlan_features |= NETIF_F_IP_CSUM;
4184 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
4185 netdev->vlan_features |= NETIF_F_GRO;
4186 netdev->vlan_features |= NETIF_F_TSO;
4187 netdev->vlan_features |= NETIF_F_TSO6;
4188 netdev->vlan_features |= NETIF_F_RXCSUM;
4189 netdev->vlan_features |= NETIF_F_RXHASH;
4191 if (!!MLX5_CAP_ETH(mdev, lro_cap))
4192 netdev->vlan_features |= NETIF_F_LRO;
4194 netdev->hw_features = netdev->vlan_features;
4195 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
4196 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
4197 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
4198 netdev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
4200 if (mlx5e_vxlan_allowed(mdev) || MLX5_CAP_ETH(mdev, tunnel_stateless_gre)) {
4201 netdev->hw_features |= NETIF_F_GSO_PARTIAL;
4202 netdev->hw_enc_features |= NETIF_F_IP_CSUM;
4203 netdev->hw_enc_features |= NETIF_F_IPV6_CSUM;
4204 netdev->hw_enc_features |= NETIF_F_TSO;
4205 netdev->hw_enc_features |= NETIF_F_TSO6;
4206 netdev->hw_enc_features |= NETIF_F_GSO_PARTIAL;
4209 if (mlx5e_vxlan_allowed(mdev)) {
4210 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
4211 NETIF_F_GSO_UDP_TUNNEL_CSUM;
4212 netdev->hw_enc_features |= NETIF_F_GSO_UDP_TUNNEL |
4213 NETIF_F_GSO_UDP_TUNNEL_CSUM;
4214 netdev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
4217 if (MLX5_CAP_ETH(mdev, tunnel_stateless_gre)) {
4218 netdev->hw_features |= NETIF_F_GSO_GRE |
4219 NETIF_F_GSO_GRE_CSUM;
4220 netdev->hw_enc_features |= NETIF_F_GSO_GRE |
4221 NETIF_F_GSO_GRE_CSUM;
4222 netdev->gso_partial_features |= NETIF_F_GSO_GRE |
4223 NETIF_F_GSO_GRE_CSUM;
4226 mlx5_query_port_fcs(mdev, &fcs_supported, &fcs_enabled);
4229 netdev->hw_features |= NETIF_F_RXALL;
4231 if (MLX5_CAP_ETH(mdev, scatter_fcs))
4232 netdev->hw_features |= NETIF_F_RXFCS;
4234 netdev->features = netdev->hw_features;
4235 if (!priv->channels.params.lro_en)
4236 netdev->features &= ~NETIF_F_LRO;
4239 netdev->features &= ~NETIF_F_RXALL;
4241 if (!priv->channels.params.scatter_fcs_en)
4242 netdev->features &= ~NETIF_F_RXFCS;
4244 #define FT_CAP(f) MLX5_CAP_FLOWTABLE(mdev, flow_table_properties_nic_receive.f)
4245 if (FT_CAP(flow_modify_en) &&
4246 FT_CAP(modify_root) &&
4247 FT_CAP(identified_miss_table_mode) &&
4248 FT_CAP(flow_table_modify)) {
4249 netdev->hw_features |= NETIF_F_HW_TC;
4250 #ifdef CONFIG_RFS_ACCEL
4251 netdev->hw_features |= NETIF_F_NTUPLE;
4255 netdev->features |= NETIF_F_HIGHDMA;
4256 netdev->features |= NETIF_F_HW_VLAN_STAG_FILTER;
4258 netdev->priv_flags |= IFF_UNICAST_FLT;
4260 mlx5e_set_netdev_dev_addr(netdev);
4262 #if IS_ENABLED(CONFIG_NET_SWITCHDEV) && IS_ENABLED(CONFIG_MLX5_ESWITCH)
4263 if (MLX5_VPORT_MANAGER(mdev))
4264 netdev->switchdev_ops = &mlx5e_switchdev_ops;
4267 mlx5e_ipsec_build_netdev(priv);
4270 static void mlx5e_create_q_counter(struct mlx5e_priv *priv)
4272 struct mlx5_core_dev *mdev = priv->mdev;
4275 err = mlx5_core_alloc_q_counter(mdev, &priv->q_counter);
4277 mlx5_core_warn(mdev, "alloc queue counter failed, %d\n", err);
4278 priv->q_counter = 0;
4282 static void mlx5e_destroy_q_counter(struct mlx5e_priv *priv)
4284 if (!priv->q_counter)
4287 mlx5_core_dealloc_q_counter(priv->mdev, priv->q_counter);
4290 static void mlx5e_nic_init(struct mlx5_core_dev *mdev,
4291 struct net_device *netdev,
4292 const struct mlx5e_profile *profile,
4295 struct mlx5e_priv *priv = netdev_priv(netdev);
4298 mlx5e_build_nic_netdev_priv(mdev, netdev, profile, ppriv);
4299 err = mlx5e_ipsec_init(priv);
4301 mlx5_core_err(mdev, "IPSec initialization failed, %d\n", err);
4302 mlx5e_build_nic_netdev(netdev);
4303 mlx5e_vxlan_init(priv);
4306 static void mlx5e_nic_cleanup(struct mlx5e_priv *priv)
4308 mlx5e_ipsec_cleanup(priv);
4309 mlx5e_vxlan_cleanup(priv);
4311 if (priv->channels.params.xdp_prog)
4312 bpf_prog_put(priv->channels.params.xdp_prog);
4315 static int mlx5e_init_nic_rx(struct mlx5e_priv *priv)
4317 struct mlx5_core_dev *mdev = priv->mdev;
4320 err = mlx5e_create_indirect_rqt(priv);
4324 err = mlx5e_create_direct_rqts(priv);
4326 goto err_destroy_indirect_rqts;
4328 err = mlx5e_create_indirect_tirs(priv);
4330 goto err_destroy_direct_rqts;
4332 err = mlx5e_create_direct_tirs(priv);
4334 goto err_destroy_indirect_tirs;
4336 err = mlx5e_create_flow_steering(priv);
4338 mlx5_core_warn(mdev, "create flow steering failed, %d\n", err);
4339 goto err_destroy_direct_tirs;
4342 err = mlx5e_tc_init(priv);
4344 goto err_destroy_flow_steering;
4348 err_destroy_flow_steering:
4349 mlx5e_destroy_flow_steering(priv);
4350 err_destroy_direct_tirs:
4351 mlx5e_destroy_direct_tirs(priv);
4352 err_destroy_indirect_tirs:
4353 mlx5e_destroy_indirect_tirs(priv);
4354 err_destroy_direct_rqts:
4355 mlx5e_destroy_direct_rqts(priv);
4356 err_destroy_indirect_rqts:
4357 mlx5e_destroy_rqt(priv, &priv->indir_rqt);
4361 static void mlx5e_cleanup_nic_rx(struct mlx5e_priv *priv)
4363 mlx5e_tc_cleanup(priv);
4364 mlx5e_destroy_flow_steering(priv);
4365 mlx5e_destroy_direct_tirs(priv);
4366 mlx5e_destroy_indirect_tirs(priv);
4367 mlx5e_destroy_direct_rqts(priv);
4368 mlx5e_destroy_rqt(priv, &priv->indir_rqt);
4371 static int mlx5e_init_nic_tx(struct mlx5e_priv *priv)
4375 err = mlx5e_create_tises(priv);
4377 mlx5_core_warn(priv->mdev, "create tises failed, %d\n", err);
4381 #ifdef CONFIG_MLX5_CORE_EN_DCB
4382 mlx5e_dcbnl_initialize(priv);
4387 static void mlx5e_nic_enable(struct mlx5e_priv *priv)
4389 struct net_device *netdev = priv->netdev;
4390 struct mlx5_core_dev *mdev = priv->mdev;
4393 mlx5e_init_l2_addr(priv);
4395 /* Marking the link as currently not needed by the Driver */
4396 if (!netif_running(netdev))
4397 mlx5_set_port_admin_status(mdev, MLX5_PORT_DOWN);
4399 /* MTU range: 68 - hw-specific max */
4400 netdev->min_mtu = ETH_MIN_MTU;
4401 mlx5_query_port_max_mtu(priv->mdev, &max_mtu, 1);
4402 netdev->max_mtu = MLX5E_HW2SW_MTU(priv, max_mtu);
4403 mlx5e_set_dev_port_mtu(priv);
4405 mlx5_lag_add(mdev, netdev);
4407 mlx5e_enable_async_events(priv);
4409 if (MLX5_VPORT_MANAGER(priv->mdev))
4410 mlx5e_register_vport_reps(priv);
4412 if (netdev->reg_state != NETREG_REGISTERED)
4414 #ifdef CONFIG_MLX5_CORE_EN_DCB
4415 mlx5e_dcbnl_init_app(priv);
4417 /* Device already registered: sync netdev system state */
4418 if (mlx5e_vxlan_allowed(mdev)) {
4420 udp_tunnel_get_rx_info(netdev);
4424 queue_work(priv->wq, &priv->set_rx_mode_work);
4427 if (netif_running(netdev))
4429 netif_device_attach(netdev);
4433 static void mlx5e_nic_disable(struct mlx5e_priv *priv)
4435 struct mlx5_core_dev *mdev = priv->mdev;
4437 #ifdef CONFIG_MLX5_CORE_EN_DCB
4438 if (priv->netdev->reg_state == NETREG_REGISTERED)
4439 mlx5e_dcbnl_delete_app(priv);
4443 if (netif_running(priv->netdev))
4444 mlx5e_close(priv->netdev);
4445 netif_device_detach(priv->netdev);
4448 queue_work(priv->wq, &priv->set_rx_mode_work);
4450 if (MLX5_VPORT_MANAGER(priv->mdev))
4451 mlx5e_unregister_vport_reps(priv);
4453 mlx5e_disable_async_events(priv);
4454 mlx5_lag_remove(mdev);
4457 static const struct mlx5e_profile mlx5e_nic_profile = {
4458 .init = mlx5e_nic_init,
4459 .cleanup = mlx5e_nic_cleanup,
4460 .init_rx = mlx5e_init_nic_rx,
4461 .cleanup_rx = mlx5e_cleanup_nic_rx,
4462 .init_tx = mlx5e_init_nic_tx,
4463 .cleanup_tx = mlx5e_cleanup_nic_tx,
4464 .enable = mlx5e_nic_enable,
4465 .disable = mlx5e_nic_disable,
4466 .update_stats = mlx5e_update_ndo_stats,
4467 .max_nch = mlx5e_get_max_num_channels,
4468 .update_carrier = mlx5e_update_carrier,
4469 .rx_handlers.handle_rx_cqe = mlx5e_handle_rx_cqe,
4470 .rx_handlers.handle_rx_cqe_mpwqe = mlx5e_handle_rx_cqe_mpwrq,
4471 .max_tc = MLX5E_MAX_NUM_TC,
4474 /* mlx5e generic netdev management API (move to en_common.c) */
4476 struct net_device *mlx5e_create_netdev(struct mlx5_core_dev *mdev,
4477 const struct mlx5e_profile *profile,
4480 int nch = profile->max_nch(mdev);
4481 struct net_device *netdev;
4482 struct mlx5e_priv *priv;
4484 netdev = alloc_etherdev_mqs(sizeof(struct mlx5e_priv),
4485 nch * profile->max_tc,
4488 mlx5_core_err(mdev, "alloc_etherdev_mqs() failed\n");
4492 #ifdef CONFIG_RFS_ACCEL
4493 netdev->rx_cpu_rmap = mdev->rmap;
4496 profile->init(mdev, netdev, profile, ppriv);
4498 netif_carrier_off(netdev);
4500 priv = netdev_priv(netdev);
4502 priv->wq = create_singlethread_workqueue("mlx5e");
4504 goto err_cleanup_nic;
4509 if (profile->cleanup)
4510 profile->cleanup(priv);
4511 free_netdev(netdev);
4516 int mlx5e_attach_netdev(struct mlx5e_priv *priv)
4518 struct mlx5_core_dev *mdev = priv->mdev;
4519 const struct mlx5e_profile *profile;
4522 profile = priv->profile;
4523 clear_bit(MLX5E_STATE_DESTROYING, &priv->state);
4525 err = profile->init_tx(priv);
4529 err = mlx5e_open_drop_rq(mdev, &priv->drop_rq);
4531 mlx5_core_err(mdev, "open drop rq failed, %d\n", err);
4532 goto err_cleanup_tx;
4535 err = profile->init_rx(priv);
4537 goto err_close_drop_rq;
4539 mlx5e_create_q_counter(priv);
4541 if (profile->enable)
4542 profile->enable(priv);
4547 mlx5e_close_drop_rq(&priv->drop_rq);
4550 profile->cleanup_tx(priv);
4556 void mlx5e_detach_netdev(struct mlx5e_priv *priv)
4558 const struct mlx5e_profile *profile = priv->profile;
4560 set_bit(MLX5E_STATE_DESTROYING, &priv->state);
4562 if (profile->disable)
4563 profile->disable(priv);
4564 flush_workqueue(priv->wq);
4566 mlx5e_destroy_q_counter(priv);
4567 profile->cleanup_rx(priv);
4568 mlx5e_close_drop_rq(&priv->drop_rq);
4569 profile->cleanup_tx(priv);
4570 cancel_delayed_work_sync(&priv->update_stats_work);
4573 void mlx5e_destroy_netdev(struct mlx5e_priv *priv)
4575 const struct mlx5e_profile *profile = priv->profile;
4576 struct net_device *netdev = priv->netdev;
4578 destroy_workqueue(priv->wq);
4579 if (profile->cleanup)
4580 profile->cleanup(priv);
4581 free_netdev(netdev);
4584 /* mlx5e_attach and mlx5e_detach scope should be only creating/destroying
4585 * hardware contexts and to connect it to the current netdev.
4587 static int mlx5e_attach(struct mlx5_core_dev *mdev, void *vpriv)
4589 struct mlx5e_priv *priv = vpriv;
4590 struct net_device *netdev = priv->netdev;
4593 if (netif_device_present(netdev))
4596 err = mlx5e_create_mdev_resources(mdev);
4600 err = mlx5e_attach_netdev(priv);
4602 mlx5e_destroy_mdev_resources(mdev);
4609 static void mlx5e_detach(struct mlx5_core_dev *mdev, void *vpriv)
4611 struct mlx5e_priv *priv = vpriv;
4612 struct net_device *netdev = priv->netdev;
4614 if (!netif_device_present(netdev))
4617 mlx5e_detach_netdev(priv);
4618 mlx5e_destroy_mdev_resources(mdev);
4621 static void *mlx5e_add(struct mlx5_core_dev *mdev)
4623 struct net_device *netdev;
4628 err = mlx5e_check_required_hca_cap(mdev);
4632 #ifdef CONFIG_MLX5_ESWITCH
4633 if (MLX5_VPORT_MANAGER(mdev)) {
4634 rpriv = mlx5e_alloc_nic_rep_priv(mdev);
4636 mlx5_core_warn(mdev, "Failed to alloc NIC rep priv data\n");
4642 netdev = mlx5e_create_netdev(mdev, &mlx5e_nic_profile, rpriv);
4644 mlx5_core_err(mdev, "mlx5e_create_netdev failed\n");
4645 goto err_free_rpriv;
4648 priv = netdev_priv(netdev);
4650 err = mlx5e_attach(mdev, priv);
4652 mlx5_core_err(mdev, "mlx5e_attach failed, %d\n", err);
4653 goto err_destroy_netdev;
4656 err = register_netdev(netdev);
4658 mlx5_core_err(mdev, "register_netdev failed, %d\n", err);
4662 #ifdef CONFIG_MLX5_CORE_EN_DCB
4663 mlx5e_dcbnl_init_app(priv);
4668 mlx5e_detach(mdev, priv);
4670 mlx5e_destroy_netdev(priv);
4676 static void mlx5e_remove(struct mlx5_core_dev *mdev, void *vpriv)
4678 struct mlx5e_priv *priv = vpriv;
4679 void *ppriv = priv->ppriv;
4681 #ifdef CONFIG_MLX5_CORE_EN_DCB
4682 mlx5e_dcbnl_delete_app(priv);
4684 unregister_netdev(priv->netdev);
4685 mlx5e_detach(mdev, vpriv);
4686 mlx5e_destroy_netdev(priv);
4690 static void *mlx5e_get_netdev(void *vpriv)
4692 struct mlx5e_priv *priv = vpriv;
4694 return priv->netdev;
4697 static struct mlx5_interface mlx5e_interface = {
4699 .remove = mlx5e_remove,
4700 .attach = mlx5e_attach,
4701 .detach = mlx5e_detach,
4702 .event = mlx5e_async_event,
4703 .protocol = MLX5_INTERFACE_PROTOCOL_ETH,
4704 .get_dev = mlx5e_get_netdev,
4707 void mlx5e_init(void)
4709 mlx5e_ipsec_build_inverse_table();
4710 mlx5e_build_ptys2ethtool_map();
4711 mlx5_register_interface(&mlx5e_interface);
4714 void mlx5e_cleanup(void)
4716 mlx5_unregister_interface(&mlx5e_interface);