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 <linux/mlx5/fs.h>
35 #include <net/vxlan.h>
36 #include <net/geneve.h>
37 #include <linux/bpf.h>
38 #include <linux/if_bridge.h>
39 #include <linux/filter.h>
40 #include <net/page_pool.h>
41 #include <net/xdp_sock_drv.h>
47 #include "en_accel/ipsec.h"
48 #include "en_accel/en_accel.h"
49 #include "en_accel/ktls.h"
50 #include "lib/vxlan.h"
51 #include "lib/clock.h"
55 #include "en/monitor_stats.h"
56 #include "en/health.h"
57 #include "en/params.h"
58 #include "en/xsk/pool.h"
59 #include "en/xsk/setup.h"
60 #include "en/xsk/rx.h"
61 #include "en/xsk/tx.h"
62 #include "en/hv_vhca_stats.h"
63 #include "en/devlink.h"
70 bool mlx5e_check_fragmented_striding_rq_cap(struct mlx5_core_dev *mdev)
72 bool striding_rq_umr, inline_umr;
75 striding_rq_umr = MLX5_CAP_GEN(mdev, striding_rq) && MLX5_CAP_GEN(mdev, umr_ptr_rlky) &&
76 MLX5_CAP_ETH(mdev, reg_umr_sq);
77 max_wqe_sz_cap = mlx5e_get_max_sq_wqebbs(mdev) * MLX5_SEND_WQE_BB;
78 inline_umr = max_wqe_sz_cap >= MLX5E_UMR_WQE_INLINE_SZ;
82 mlx5_core_warn(mdev, "Cannot support Striding RQ: UMR WQE size (%d) exceeds maximum supported (%d).\n",
83 (int)MLX5E_UMR_WQE_INLINE_SZ, max_wqe_sz_cap);
89 void mlx5e_update_carrier(struct mlx5e_priv *priv)
91 struct mlx5_core_dev *mdev = priv->mdev;
95 port_state = mlx5_query_vport_state(mdev,
96 MLX5_VPORT_STATE_OP_MOD_VNIC_VPORT,
99 up = port_state == VPORT_STATE_UP;
100 if (up == netif_carrier_ok(priv->netdev))
101 netif_carrier_event(priv->netdev);
103 netdev_info(priv->netdev, "Link up\n");
104 netif_carrier_on(priv->netdev);
106 netdev_info(priv->netdev, "Link down\n");
107 netif_carrier_off(priv->netdev);
111 static void mlx5e_update_carrier_work(struct work_struct *work)
113 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
114 update_carrier_work);
116 mutex_lock(&priv->state_lock);
117 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
118 if (priv->profile->update_carrier)
119 priv->profile->update_carrier(priv);
120 mutex_unlock(&priv->state_lock);
123 static void mlx5e_update_stats_work(struct work_struct *work)
125 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
128 mutex_lock(&priv->state_lock);
129 priv->profile->update_stats(priv);
130 mutex_unlock(&priv->state_lock);
133 void mlx5e_queue_update_stats(struct mlx5e_priv *priv)
135 if (!priv->profile->update_stats)
138 if (unlikely(test_bit(MLX5E_STATE_DESTROYING, &priv->state)))
141 queue_work(priv->wq, &priv->update_stats_work);
144 static int async_event(struct notifier_block *nb, unsigned long event, void *data)
146 struct mlx5e_priv *priv = container_of(nb, struct mlx5e_priv, events_nb);
147 struct mlx5_eqe *eqe = data;
149 if (event != MLX5_EVENT_TYPE_PORT_CHANGE)
152 switch (eqe->sub_type) {
153 case MLX5_PORT_CHANGE_SUBTYPE_DOWN:
154 case MLX5_PORT_CHANGE_SUBTYPE_ACTIVE:
155 queue_work(priv->wq, &priv->update_carrier_work);
164 static void mlx5e_enable_async_events(struct mlx5e_priv *priv)
166 priv->events_nb.notifier_call = async_event;
167 mlx5_notifier_register(priv->mdev, &priv->events_nb);
170 static void mlx5e_disable_async_events(struct mlx5e_priv *priv)
172 mlx5_notifier_unregister(priv->mdev, &priv->events_nb);
175 static int blocking_event(struct notifier_block *nb, unsigned long event, void *data)
177 struct mlx5e_priv *priv = container_of(nb, struct mlx5e_priv, blocking_events_nb);
181 case MLX5_DRIVER_EVENT_TYPE_TRAP:
182 err = mlx5e_handle_trap_event(priv, data);
185 netdev_warn(priv->netdev, "Sync event: Unknown event %ld\n", event);
191 static void mlx5e_enable_blocking_events(struct mlx5e_priv *priv)
193 priv->blocking_events_nb.notifier_call = blocking_event;
194 mlx5_blocking_notifier_register(priv->mdev, &priv->blocking_events_nb);
197 static void mlx5e_disable_blocking_events(struct mlx5e_priv *priv)
199 mlx5_blocking_notifier_unregister(priv->mdev, &priv->blocking_events_nb);
202 static inline void mlx5e_build_umr_wqe(struct mlx5e_rq *rq,
203 struct mlx5e_icosq *sq,
204 struct mlx5e_umr_wqe *wqe)
206 struct mlx5_wqe_ctrl_seg *cseg = &wqe->ctrl;
207 struct mlx5_wqe_umr_ctrl_seg *ucseg = &wqe->uctrl;
208 u8 ds_cnt = DIV_ROUND_UP(MLX5E_UMR_WQE_INLINE_SZ, MLX5_SEND_WQE_DS);
210 cseg->qpn_ds = cpu_to_be32((sq->sqn << MLX5_WQE_CTRL_QPN_SHIFT) |
212 cseg->umr_mkey = rq->mkey_be;
214 ucseg->flags = MLX5_UMR_TRANSLATION_OFFSET_EN | MLX5_UMR_INLINE;
215 ucseg->xlt_octowords =
216 cpu_to_be16(MLX5_MTT_OCTW(MLX5_MPWRQ_PAGES_PER_WQE));
217 ucseg->mkey_mask = cpu_to_be64(MLX5_MKEY_MASK_FREE);
220 static int mlx5e_rq_shampo_hd_alloc(struct mlx5e_rq *rq, int node)
222 rq->mpwqe.shampo = kvzalloc_node(sizeof(*rq->mpwqe.shampo),
224 if (!rq->mpwqe.shampo)
229 static void mlx5e_rq_shampo_hd_free(struct mlx5e_rq *rq)
231 kvfree(rq->mpwqe.shampo);
234 static int mlx5e_rq_shampo_hd_info_alloc(struct mlx5e_rq *rq, int node)
236 struct mlx5e_shampo_hd *shampo = rq->mpwqe.shampo;
238 shampo->bitmap = bitmap_zalloc_node(shampo->hd_per_wq, GFP_KERNEL,
243 shampo->info = kvzalloc_node(array_size(shampo->hd_per_wq,
244 sizeof(*shampo->info)),
247 kvfree(shampo->bitmap);
253 static void mlx5e_rq_shampo_hd_info_free(struct mlx5e_rq *rq)
255 kvfree(rq->mpwqe.shampo->bitmap);
256 kvfree(rq->mpwqe.shampo->info);
259 static int mlx5e_rq_alloc_mpwqe_info(struct mlx5e_rq *rq, int node)
261 int wq_sz = mlx5_wq_ll_get_size(&rq->mpwqe.wq);
263 rq->mpwqe.info = kvzalloc_node(array_size(wq_sz,
264 sizeof(*rq->mpwqe.info)),
269 mlx5e_build_umr_wqe(rq, rq->icosq, &rq->mpwqe.umr_wqe);
274 static int mlx5e_create_umr_mtt_mkey(struct mlx5_core_dev *mdev,
275 u64 npages, u8 page_shift, u32 *umr_mkey,
276 dma_addr_t filler_addr)
278 struct mlx5_mtt *mtt;
285 inlen = MLX5_ST_SZ_BYTES(create_mkey_in) + sizeof(*mtt) * npages;
287 in = kvzalloc(inlen, GFP_KERNEL);
291 mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
293 MLX5_SET(mkc, mkc, free, 1);
294 MLX5_SET(mkc, mkc, umr_en, 1);
295 MLX5_SET(mkc, mkc, lw, 1);
296 MLX5_SET(mkc, mkc, lr, 1);
297 MLX5_SET(mkc, mkc, access_mode_1_0, MLX5_MKC_ACCESS_MODE_MTT);
298 mlx5e_mkey_set_relaxed_ordering(mdev, mkc);
299 MLX5_SET(mkc, mkc, qpn, 0xffffff);
300 MLX5_SET(mkc, mkc, pd, mdev->mlx5e_res.hw_objs.pdn);
301 MLX5_SET64(mkc, mkc, len, npages << page_shift);
302 MLX5_SET(mkc, mkc, translations_octword_size,
303 MLX5_MTT_OCTW(npages));
304 MLX5_SET(mkc, mkc, log_page_size, page_shift);
305 MLX5_SET(create_mkey_in, in, translations_octword_actual_size,
306 MLX5_MTT_OCTW(npages));
308 /* Initialize the mkey with all MTTs pointing to a default
309 * page (filler_addr). When the channels are activated, UMR
310 * WQEs will redirect the RX WQEs to the actual memory from
311 * the RQ's pool, while the gaps (wqe_overflow) remain mapped
312 * to the default page.
314 mtt = MLX5_ADDR_OF(create_mkey_in, in, klm_pas_mtt);
315 for (i = 0 ; i < npages ; i++)
316 mtt[i].ptag = cpu_to_be64(filler_addr);
318 err = mlx5_core_create_mkey(mdev, umr_mkey, in, inlen);
324 static int mlx5e_create_umr_klm_mkey(struct mlx5_core_dev *mdev,
333 inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
335 in = kvzalloc(inlen, GFP_KERNEL);
339 mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
341 MLX5_SET(mkc, mkc, free, 1);
342 MLX5_SET(mkc, mkc, umr_en, 1);
343 MLX5_SET(mkc, mkc, lw, 1);
344 MLX5_SET(mkc, mkc, lr, 1);
345 MLX5_SET(mkc, mkc, access_mode_1_0, MLX5_MKC_ACCESS_MODE_KLMS);
346 mlx5e_mkey_set_relaxed_ordering(mdev, mkc);
347 MLX5_SET(mkc, mkc, qpn, 0xffffff);
348 MLX5_SET(mkc, mkc, pd, mdev->mlx5e_res.hw_objs.pdn);
349 MLX5_SET(mkc, mkc, translations_octword_size, nentries);
350 MLX5_SET(mkc, mkc, length64, 1);
351 err = mlx5_core_create_mkey(mdev, umr_mkey, in, inlen);
357 static int mlx5e_create_rq_umr_mkey(struct mlx5_core_dev *mdev, struct mlx5e_rq *rq)
359 u64 num_mtts = MLX5E_REQUIRED_MTTS(mlx5_wq_ll_get_size(&rq->mpwqe.wq));
361 return mlx5e_create_umr_mtt_mkey(mdev, num_mtts, PAGE_SHIFT,
362 &rq->umr_mkey, rq->wqe_overflow.addr);
365 static int mlx5e_create_rq_hd_umr_mkey(struct mlx5_core_dev *mdev,
368 u32 max_klm_size = BIT(MLX5_CAP_GEN(mdev, log_max_klm_list_size));
370 if (max_klm_size < rq->mpwqe.shampo->hd_per_wq) {
371 mlx5_core_err(mdev, "max klm list size 0x%x is smaller than shampo header buffer list size 0x%x\n",
372 max_klm_size, rq->mpwqe.shampo->hd_per_wq);
375 return mlx5e_create_umr_klm_mkey(mdev, rq->mpwqe.shampo->hd_per_wq,
376 &rq->mpwqe.shampo->mkey);
379 static u64 mlx5e_get_mpwqe_offset(u16 wqe_ix)
381 return MLX5E_REQUIRED_MTTS(wqe_ix) << PAGE_SHIFT;
384 static void mlx5e_init_frags_partition(struct mlx5e_rq *rq)
386 struct mlx5e_wqe_frag_info next_frag = {};
387 struct mlx5e_wqe_frag_info *prev = NULL;
390 next_frag.di = &rq->wqe.di[0];
392 for (i = 0; i < mlx5_wq_cyc_get_size(&rq->wqe.wq); i++) {
393 struct mlx5e_rq_frag_info *frag_info = &rq->wqe.info.arr[0];
394 struct mlx5e_wqe_frag_info *frag =
395 &rq->wqe.frags[i << rq->wqe.info.log_num_frags];
398 for (f = 0; f < rq->wqe.info.num_frags; f++, frag++) {
399 if (next_frag.offset + frag_info[f].frag_stride > PAGE_SIZE) {
401 next_frag.offset = 0;
403 prev->last_in_page = true;
408 next_frag.offset += frag_info[f].frag_stride;
414 prev->last_in_page = true;
417 int mlx5e_init_di_list(struct mlx5e_rq *rq, int wq_sz, int node)
419 int len = wq_sz << rq->wqe.info.log_num_frags;
421 rq->wqe.di = kvzalloc_node(array_size(len, sizeof(*rq->wqe.di)), GFP_KERNEL, node);
425 mlx5e_init_frags_partition(rq);
430 void mlx5e_free_di_list(struct mlx5e_rq *rq)
435 static void mlx5e_rq_err_cqe_work(struct work_struct *recover_work)
437 struct mlx5e_rq *rq = container_of(recover_work, struct mlx5e_rq, recover_work);
439 mlx5e_reporter_rq_cqe_err(rq);
442 static int mlx5e_alloc_mpwqe_rq_drop_page(struct mlx5e_rq *rq)
444 rq->wqe_overflow.page = alloc_page(GFP_KERNEL);
445 if (!rq->wqe_overflow.page)
448 rq->wqe_overflow.addr = dma_map_page(rq->pdev, rq->wqe_overflow.page, 0,
449 PAGE_SIZE, rq->buff.map_dir);
450 if (dma_mapping_error(rq->pdev, rq->wqe_overflow.addr)) {
451 __free_page(rq->wqe_overflow.page);
457 static void mlx5e_free_mpwqe_rq_drop_page(struct mlx5e_rq *rq)
459 dma_unmap_page(rq->pdev, rq->wqe_overflow.addr, PAGE_SIZE,
461 __free_page(rq->wqe_overflow.page);
464 static int mlx5e_init_rxq_rq(struct mlx5e_channel *c, struct mlx5e_params *params,
467 struct mlx5_core_dev *mdev = c->mdev;
470 rq->wq_type = params->rq_wq_type;
472 rq->netdev = c->netdev;
474 rq->tstamp = c->tstamp;
475 rq->clock = &mdev->clock;
476 rq->icosq = &c->icosq;
480 rq->hw_mtu = MLX5E_SW2HW_MTU(params, params->sw_mtu);
481 rq->xdpsq = &c->rq_xdpsq;
482 rq->stats = &c->priv->channel_stats[c->ix]->rq;
483 rq->ptp_cyc2time = mlx5_rq_ts_translator(mdev);
484 err = mlx5e_rq_set_handlers(rq, params, NULL);
488 return xdp_rxq_info_reg(&rq->xdp_rxq, rq->netdev, rq->ix, 0);
491 static int mlx5_rq_shampo_alloc(struct mlx5_core_dev *mdev,
492 struct mlx5e_params *params,
493 struct mlx5e_rq_param *rqp,
498 void *wqc = MLX5_ADDR_OF(rqc, rqp->rqc, wq);
502 if (!test_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state))
504 err = mlx5e_rq_shampo_hd_alloc(rq, node);
507 rq->mpwqe.shampo->hd_per_wq =
508 mlx5e_shampo_hd_per_wq(mdev, params, rqp);
509 err = mlx5e_create_rq_hd_umr_mkey(mdev, rq);
512 err = mlx5e_rq_shampo_hd_info_alloc(rq, node);
514 goto err_shampo_info;
515 rq->hw_gro_data = kvzalloc_node(sizeof(*rq->hw_gro_data), GFP_KERNEL, node);
516 if (!rq->hw_gro_data) {
518 goto err_hw_gro_data;
520 rq->mpwqe.shampo->key =
521 cpu_to_be32(rq->mpwqe.shampo->mkey);
522 rq->mpwqe.shampo->hd_per_wqe =
523 mlx5e_shampo_hd_per_wqe(mdev, params, rqp);
524 wq_size = BIT(MLX5_GET(wq, wqc, log_wq_sz));
525 *pool_size += (rq->mpwqe.shampo->hd_per_wqe * wq_size) /
526 MLX5E_SHAMPO_WQ_HEADER_PER_PAGE;
530 mlx5e_rq_shampo_hd_info_free(rq);
532 mlx5_core_destroy_mkey(mdev, rq->mpwqe.shampo->mkey);
534 mlx5e_rq_shampo_hd_free(rq);
539 static void mlx5e_rq_free_shampo(struct mlx5e_rq *rq)
541 if (!test_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state))
544 kvfree(rq->hw_gro_data);
545 mlx5e_rq_shampo_hd_info_free(rq);
546 mlx5_core_destroy_mkey(rq->mdev, rq->mpwqe.shampo->mkey);
547 mlx5e_rq_shampo_hd_free(rq);
550 static int mlx5e_alloc_rq(struct mlx5e_params *params,
551 struct mlx5e_xsk_param *xsk,
552 struct mlx5e_rq_param *rqp,
553 int node, struct mlx5e_rq *rq)
555 struct page_pool_params pp_params = { 0 };
556 struct mlx5_core_dev *mdev = rq->mdev;
557 void *rqc = rqp->rqc;
558 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
564 rqp->wq.db_numa_node = node;
565 INIT_WORK(&rq->recover_work, mlx5e_rq_err_cqe_work);
567 if (params->xdp_prog)
568 bpf_prog_inc(params->xdp_prog);
569 RCU_INIT_POINTER(rq->xdp_prog, params->xdp_prog);
571 rq->buff.map_dir = params->xdp_prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
572 rq->buff.headroom = mlx5e_get_rq_headroom(mdev, params, xsk);
573 pool_size = 1 << params->log_rq_mtu_frames;
575 switch (rq->wq_type) {
576 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
577 err = mlx5_wq_ll_create(mdev, &rqp->wq, rqc_wq, &rq->mpwqe.wq,
580 goto err_rq_xdp_prog;
582 err = mlx5e_alloc_mpwqe_rq_drop_page(rq);
584 goto err_rq_wq_destroy;
586 rq->mpwqe.wq.db = &rq->mpwqe.wq.db[MLX5_RCV_DBR];
588 wq_sz = mlx5_wq_ll_get_size(&rq->mpwqe.wq);
590 pool_size = MLX5_MPWRQ_PAGES_PER_WQE <<
591 mlx5e_mpwqe_get_log_rq_size(params, xsk);
593 rq->mpwqe.log_stride_sz = mlx5e_mpwqe_get_log_stride_size(mdev, params, xsk);
594 rq->mpwqe.num_strides =
595 BIT(mlx5e_mpwqe_get_log_num_strides(mdev, params, xsk));
596 rq->mpwqe.min_wqe_bulk = mlx5e_mpwqe_get_min_wqe_bulk(wq_sz);
598 rq->buff.frame0_sz = (1 << rq->mpwqe.log_stride_sz);
600 err = mlx5e_create_rq_umr_mkey(mdev, rq);
602 goto err_rq_drop_page;
603 rq->mkey_be = cpu_to_be32(rq->umr_mkey);
605 err = mlx5e_rq_alloc_mpwqe_info(rq, node);
609 err = mlx5_rq_shampo_alloc(mdev, params, rqp, rq, &pool_size, node);
611 goto err_free_by_rq_type;
614 default: /* MLX5_WQ_TYPE_CYCLIC */
615 err = mlx5_wq_cyc_create(mdev, &rqp->wq, rqc_wq, &rq->wqe.wq,
618 goto err_rq_xdp_prog;
620 rq->wqe.wq.db = &rq->wqe.wq.db[MLX5_RCV_DBR];
622 wq_sz = mlx5_wq_cyc_get_size(&rq->wqe.wq);
624 rq->wqe.info = rqp->frags_info;
625 rq->buff.frame0_sz = rq->wqe.info.arr[0].frag_stride;
628 kvzalloc_node(array_size(sizeof(*rq->wqe.frags),
629 (wq_sz << rq->wqe.info.log_num_frags)),
631 if (!rq->wqe.frags) {
633 goto err_rq_wq_destroy;
636 err = mlx5e_init_di_list(rq, wq_sz, node);
640 rq->mkey_be = cpu_to_be32(mdev->mlx5e_res.hw_objs.mkey);
644 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
645 MEM_TYPE_XSK_BUFF_POOL, NULL);
646 xsk_pool_set_rxq_info(rq->xsk_pool, &rq->xdp_rxq);
648 /* Create a page_pool and register it with rxq */
650 pp_params.flags = 0; /* No-internal DMA mapping in page_pool */
651 pp_params.pool_size = pool_size;
652 pp_params.nid = node;
653 pp_params.dev = rq->pdev;
654 pp_params.dma_dir = rq->buff.map_dir;
656 /* page_pool can be used even when there is no rq->xdp_prog,
657 * given page_pool does not handle DMA mapping there is no
658 * required state to clear. And page_pool gracefully handle
661 rq->page_pool = page_pool_create(&pp_params);
662 if (IS_ERR(rq->page_pool)) {
663 err = PTR_ERR(rq->page_pool);
664 rq->page_pool = NULL;
665 goto err_free_shampo;
667 if (xdp_rxq_info_is_reg(&rq->xdp_rxq))
668 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
669 MEM_TYPE_PAGE_POOL, rq->page_pool);
672 goto err_free_shampo;
674 for (i = 0; i < wq_sz; i++) {
675 if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
676 struct mlx5e_rx_wqe_ll *wqe =
677 mlx5_wq_ll_get_wqe(&rq->mpwqe.wq, i);
679 rq->mpwqe.num_strides << rq->mpwqe.log_stride_sz;
680 u64 dma_offset = mlx5e_get_mpwqe_offset(i);
681 u16 headroom = test_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state) ?
682 0 : rq->buff.headroom;
684 wqe->data[0].addr = cpu_to_be64(dma_offset + headroom);
685 wqe->data[0].byte_count = cpu_to_be32(byte_count);
686 wqe->data[0].lkey = rq->mkey_be;
688 struct mlx5e_rx_wqe_cyc *wqe =
689 mlx5_wq_cyc_get_wqe(&rq->wqe.wq, i);
692 for (f = 0; f < rq->wqe.info.num_frags; f++) {
693 u32 frag_size = rq->wqe.info.arr[f].frag_size |
694 MLX5_HW_START_PADDING;
696 wqe->data[f].byte_count = cpu_to_be32(frag_size);
697 wqe->data[f].lkey = rq->mkey_be;
699 /* check if num_frags is not a pow of two */
700 if (rq->wqe.info.num_frags < (1 << rq->wqe.info.log_num_frags)) {
701 wqe->data[f].byte_count = 0;
702 wqe->data[f].lkey = cpu_to_be32(MLX5_INVALID_LKEY);
703 wqe->data[f].addr = 0;
708 INIT_WORK(&rq->dim.work, mlx5e_rx_dim_work);
710 switch (params->rx_cq_moderation.cq_period_mode) {
711 case MLX5_CQ_PERIOD_MODE_START_FROM_CQE:
712 rq->dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_CQE;
714 case MLX5_CQ_PERIOD_MODE_START_FROM_EQE:
716 rq->dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
719 rq->page_cache.head = 0;
720 rq->page_cache.tail = 0;
725 mlx5e_rq_free_shampo(rq);
727 switch (rq->wq_type) {
728 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
729 kvfree(rq->mpwqe.info);
731 mlx5_core_destroy_mkey(mdev, rq->umr_mkey);
733 mlx5e_free_mpwqe_rq_drop_page(rq);
735 default: /* MLX5_WQ_TYPE_CYCLIC */
736 mlx5e_free_di_list(rq);
738 kvfree(rq->wqe.frags);
741 mlx5_wq_destroy(&rq->wq_ctrl);
743 if (params->xdp_prog)
744 bpf_prog_put(params->xdp_prog);
749 static void mlx5e_free_rq(struct mlx5e_rq *rq)
751 struct bpf_prog *old_prog;
754 if (xdp_rxq_info_is_reg(&rq->xdp_rxq)) {
755 old_prog = rcu_dereference_protected(rq->xdp_prog,
756 lockdep_is_held(&rq->priv->state_lock));
758 bpf_prog_put(old_prog);
761 switch (rq->wq_type) {
762 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
763 kvfree(rq->mpwqe.info);
764 mlx5_core_destroy_mkey(rq->mdev, rq->umr_mkey);
765 mlx5e_free_mpwqe_rq_drop_page(rq);
766 mlx5e_rq_free_shampo(rq);
768 default: /* MLX5_WQ_TYPE_CYCLIC */
769 kvfree(rq->wqe.frags);
770 mlx5e_free_di_list(rq);
773 for (i = rq->page_cache.head; i != rq->page_cache.tail;
774 i = (i + 1) & (MLX5E_CACHE_SIZE - 1)) {
775 struct mlx5e_dma_info *dma_info = &rq->page_cache.page_cache[i];
777 /* With AF_XDP, page_cache is not used, so this loop is not
778 * entered, and it's safe to call mlx5e_page_release_dynamic
781 mlx5e_page_release_dynamic(rq, dma_info->page, false);
784 xdp_rxq_info_unreg(&rq->xdp_rxq);
785 page_pool_destroy(rq->page_pool);
786 mlx5_wq_destroy(&rq->wq_ctrl);
789 int mlx5e_create_rq(struct mlx5e_rq *rq, struct mlx5e_rq_param *param)
791 struct mlx5_core_dev *mdev = rq->mdev;
799 inlen = MLX5_ST_SZ_BYTES(create_rq_in) +
800 sizeof(u64) * rq->wq_ctrl.buf.npages;
801 in = kvzalloc(inlen, GFP_KERNEL);
805 ts_format = mlx5_is_real_time_rq(mdev) ?
806 MLX5_TIMESTAMP_FORMAT_REAL_TIME :
807 MLX5_TIMESTAMP_FORMAT_FREE_RUNNING;
808 rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
809 wq = MLX5_ADDR_OF(rqc, rqc, wq);
811 memcpy(rqc, param->rqc, sizeof(param->rqc));
813 MLX5_SET(rqc, rqc, cqn, rq->cq.mcq.cqn);
814 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST);
815 MLX5_SET(rqc, rqc, ts_format, ts_format);
816 MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift -
817 MLX5_ADAPTER_PAGE_SHIFT);
818 MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma);
820 if (test_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state)) {
821 MLX5_SET(wq, wq, log_headers_buffer_entry_num,
822 order_base_2(rq->mpwqe.shampo->hd_per_wq));
823 MLX5_SET(wq, wq, headers_mkey, rq->mpwqe.shampo->mkey);
826 mlx5_fill_page_frag_array(&rq->wq_ctrl.buf,
827 (__be64 *)MLX5_ADDR_OF(wq, wq, pas));
829 err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn);
836 int mlx5e_modify_rq_state(struct mlx5e_rq *rq, int curr_state, int next_state)
838 struct mlx5_core_dev *mdev = rq->mdev;
845 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
846 in = kvzalloc(inlen, GFP_KERNEL);
850 if (curr_state == MLX5_RQC_STATE_RST && next_state == MLX5_RQC_STATE_RDY)
851 mlx5e_rqwq_reset(rq);
853 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
855 MLX5_SET(modify_rq_in, in, rq_state, curr_state);
856 MLX5_SET(rqc, rqc, state, next_state);
858 err = mlx5_core_modify_rq(mdev, rq->rqn, in);
865 static int mlx5e_modify_rq_scatter_fcs(struct mlx5e_rq *rq, bool enable)
867 struct mlx5_core_dev *mdev = rq->mdev;
874 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
875 in = kvzalloc(inlen, GFP_KERNEL);
879 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
881 MLX5_SET(modify_rq_in, in, rq_state, MLX5_RQC_STATE_RDY);
882 MLX5_SET64(modify_rq_in, in, modify_bitmask,
883 MLX5_MODIFY_RQ_IN_MODIFY_BITMASK_SCATTER_FCS);
884 MLX5_SET(rqc, rqc, scatter_fcs, enable);
885 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RDY);
887 err = mlx5_core_modify_rq(mdev, rq->rqn, in);
894 static int mlx5e_modify_rq_vsd(struct mlx5e_rq *rq, bool vsd)
896 struct mlx5_core_dev *mdev = rq->mdev;
902 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
903 in = kvzalloc(inlen, GFP_KERNEL);
907 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
909 MLX5_SET(modify_rq_in, in, rq_state, MLX5_RQC_STATE_RDY);
910 MLX5_SET64(modify_rq_in, in, modify_bitmask,
911 MLX5_MODIFY_RQ_IN_MODIFY_BITMASK_VSD);
912 MLX5_SET(rqc, rqc, vsd, vsd);
913 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RDY);
915 err = mlx5_core_modify_rq(mdev, rq->rqn, in);
922 void mlx5e_destroy_rq(struct mlx5e_rq *rq)
924 mlx5_core_destroy_rq(rq->mdev, rq->rqn);
927 int mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq, int wait_time)
929 unsigned long exp_time = jiffies + msecs_to_jiffies(wait_time);
931 u16 min_wqes = mlx5_min_rx_wqes(rq->wq_type, mlx5e_rqwq_get_size(rq));
934 if (mlx5e_rqwq_get_cur_sz(rq) >= min_wqes)
938 } while (time_before(jiffies, exp_time));
940 netdev_warn(rq->netdev, "Failed to get min RX wqes on Channel[%d] RQN[0x%x] wq cur_sz(%d) min_rx_wqes(%d)\n",
941 rq->ix, rq->rqn, mlx5e_rqwq_get_cur_sz(rq), min_wqes);
943 mlx5e_reporter_rx_timeout(rq);
947 void mlx5e_free_rx_in_progress_descs(struct mlx5e_rq *rq)
949 struct mlx5_wq_ll *wq;
953 if (rq->wq_type != MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ)
959 /* Outstanding UMR WQEs (in progress) start at wq->head */
960 for (i = 0; i < rq->mpwqe.umr_in_progress; i++) {
961 rq->dealloc_wqe(rq, head);
962 head = mlx5_wq_ll_get_wqe_next_ix(wq, head);
965 if (test_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state)) {
968 len = (rq->mpwqe.shampo->pi - rq->mpwqe.shampo->ci) &
969 (rq->mpwqe.shampo->hd_per_wq - 1);
970 mlx5e_shampo_dealloc_hd(rq, len, rq->mpwqe.shampo->ci, false);
971 rq->mpwqe.shampo->pi = rq->mpwqe.shampo->ci;
974 rq->mpwqe.actual_wq_head = wq->head;
975 rq->mpwqe.umr_in_progress = 0;
976 rq->mpwqe.umr_completed = 0;
979 void mlx5e_free_rx_descs(struct mlx5e_rq *rq)
984 if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
985 struct mlx5_wq_ll *wq = &rq->mpwqe.wq;
987 mlx5e_free_rx_in_progress_descs(rq);
989 while (!mlx5_wq_ll_is_empty(wq)) {
990 struct mlx5e_rx_wqe_ll *wqe;
992 wqe_ix_be = *wq->tail_next;
993 wqe_ix = be16_to_cpu(wqe_ix_be);
994 wqe = mlx5_wq_ll_get_wqe(wq, wqe_ix);
995 rq->dealloc_wqe(rq, wqe_ix);
996 mlx5_wq_ll_pop(wq, wqe_ix_be,
997 &wqe->next.next_wqe_index);
1000 if (test_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state))
1001 mlx5e_shampo_dealloc_hd(rq, rq->mpwqe.shampo->hd_per_wq,
1004 struct mlx5_wq_cyc *wq = &rq->wqe.wq;
1006 while (!mlx5_wq_cyc_is_empty(wq)) {
1007 wqe_ix = mlx5_wq_cyc_get_tail(wq);
1008 rq->dealloc_wqe(rq, wqe_ix);
1009 mlx5_wq_cyc_pop(wq);
1015 int mlx5e_open_rq(struct mlx5e_params *params, struct mlx5e_rq_param *param,
1016 struct mlx5e_xsk_param *xsk, int node,
1017 struct mlx5e_rq *rq)
1019 struct mlx5_core_dev *mdev = rq->mdev;
1022 if (params->packet_merge.type == MLX5E_PACKET_MERGE_SHAMPO)
1023 __set_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state);
1025 err = mlx5e_alloc_rq(params, xsk, param, node, rq);
1029 err = mlx5e_create_rq(rq, param);
1033 err = mlx5e_modify_rq_state(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
1035 goto err_destroy_rq;
1037 if (MLX5_CAP_ETH(mdev, cqe_checksum_full))
1038 __set_bit(MLX5E_RQ_STATE_CSUM_FULL, &rq->state);
1040 if (params->rx_dim_enabled)
1041 __set_bit(MLX5E_RQ_STATE_AM, &rq->state);
1043 /* We disable csum_complete when XDP is enabled since
1044 * XDP programs might manipulate packets which will render
1045 * skb->checksum incorrect.
1047 if (MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_NO_CSUM_COMPLETE) || params->xdp_prog)
1048 __set_bit(MLX5E_RQ_STATE_NO_CSUM_COMPLETE, &rq->state);
1050 /* For CQE compression on striding RQ, use stride index provided by
1051 * HW if capability is supported.
1053 if (MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_STRIDING_RQ) &&
1054 MLX5_CAP_GEN(mdev, mini_cqe_resp_stride_index))
1055 __set_bit(MLX5E_RQ_STATE_MINI_CQE_HW_STRIDX, &rq->state);
1060 mlx5e_destroy_rq(rq);
1067 void mlx5e_activate_rq(struct mlx5e_rq *rq)
1069 set_bit(MLX5E_RQ_STATE_ENABLED, &rq->state);
1072 void mlx5e_deactivate_rq(struct mlx5e_rq *rq)
1074 clear_bit(MLX5E_RQ_STATE_ENABLED, &rq->state);
1075 synchronize_net(); /* Sync with NAPI to prevent mlx5e_post_rx_wqes. */
1078 void mlx5e_close_rq(struct mlx5e_rq *rq)
1080 cancel_work_sync(&rq->dim.work);
1081 cancel_work_sync(&rq->recover_work);
1082 mlx5e_destroy_rq(rq);
1083 mlx5e_free_rx_descs(rq);
1087 static void mlx5e_free_xdpsq_db(struct mlx5e_xdpsq *sq)
1089 kvfree(sq->db.xdpi_fifo.xi);
1090 kvfree(sq->db.wqe_info);
1093 static int mlx5e_alloc_xdpsq_fifo(struct mlx5e_xdpsq *sq, int numa)
1095 struct mlx5e_xdp_info_fifo *xdpi_fifo = &sq->db.xdpi_fifo;
1096 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1097 int dsegs_per_wq = wq_sz * MLX5_SEND_WQEBB_NUM_DS;
1100 size = array_size(sizeof(*xdpi_fifo->xi), dsegs_per_wq);
1101 xdpi_fifo->xi = kvzalloc_node(size, GFP_KERNEL, numa);
1105 xdpi_fifo->pc = &sq->xdpi_fifo_pc;
1106 xdpi_fifo->cc = &sq->xdpi_fifo_cc;
1107 xdpi_fifo->mask = dsegs_per_wq - 1;
1112 static int mlx5e_alloc_xdpsq_db(struct mlx5e_xdpsq *sq, int numa)
1114 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1118 size = array_size(sizeof(*sq->db.wqe_info), wq_sz);
1119 sq->db.wqe_info = kvzalloc_node(size, GFP_KERNEL, numa);
1120 if (!sq->db.wqe_info)
1123 err = mlx5e_alloc_xdpsq_fifo(sq, numa);
1125 mlx5e_free_xdpsq_db(sq);
1132 static int mlx5e_alloc_xdpsq(struct mlx5e_channel *c,
1133 struct mlx5e_params *params,
1134 struct xsk_buff_pool *xsk_pool,
1135 struct mlx5e_sq_param *param,
1136 struct mlx5e_xdpsq *sq,
1139 void *sqc_wq = MLX5_ADDR_OF(sqc, param->sqc, wq);
1140 struct mlx5_core_dev *mdev = c->mdev;
1141 struct mlx5_wq_cyc *wq = &sq->wq;
1145 sq->mkey_be = c->mkey_be;
1147 sq->uar_map = mdev->mlx5e_res.hw_objs.bfreg.map;
1148 sq->min_inline_mode = params->tx_min_inline_mode;
1149 sq->hw_mtu = MLX5E_SW2HW_MTU(params, params->sw_mtu);
1150 sq->xsk_pool = xsk_pool;
1152 sq->stats = sq->xsk_pool ?
1153 &c->priv->channel_stats[c->ix]->xsksq :
1155 &c->priv->channel_stats[c->ix]->xdpsq :
1156 &c->priv->channel_stats[c->ix]->rq_xdpsq;
1157 sq->max_sq_wqebbs = mlx5e_get_max_sq_wqebbs(mdev);
1158 sq->stop_room = MLX5E_STOP_ROOM(sq->max_sq_wqebbs);
1159 sq->max_sq_mpw_wqebbs = mlx5e_get_sw_max_sq_mpw_wqebbs(sq->max_sq_wqebbs);
1161 param->wq.db_numa_node = cpu_to_node(c->cpu);
1162 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, wq, &sq->wq_ctrl);
1165 wq->db = &wq->db[MLX5_SND_DBR];
1167 err = mlx5e_alloc_xdpsq_db(sq, cpu_to_node(c->cpu));
1169 goto err_sq_wq_destroy;
1174 mlx5_wq_destroy(&sq->wq_ctrl);
1179 static void mlx5e_free_xdpsq(struct mlx5e_xdpsq *sq)
1181 mlx5e_free_xdpsq_db(sq);
1182 mlx5_wq_destroy(&sq->wq_ctrl);
1185 static void mlx5e_free_icosq_db(struct mlx5e_icosq *sq)
1187 kvfree(sq->db.wqe_info);
1190 static int mlx5e_alloc_icosq_db(struct mlx5e_icosq *sq, int numa)
1192 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1195 size = array_size(wq_sz, sizeof(*sq->db.wqe_info));
1196 sq->db.wqe_info = kvzalloc_node(size, GFP_KERNEL, numa);
1197 if (!sq->db.wqe_info)
1203 static void mlx5e_icosq_err_cqe_work(struct work_struct *recover_work)
1205 struct mlx5e_icosq *sq = container_of(recover_work, struct mlx5e_icosq,
1208 mlx5e_reporter_icosq_cqe_err(sq);
1211 static void mlx5e_async_icosq_err_cqe_work(struct work_struct *recover_work)
1213 struct mlx5e_icosq *sq = container_of(recover_work, struct mlx5e_icosq,
1216 /* Not implemented yet. */
1218 netdev_warn(sq->channel->netdev, "async_icosq recovery is not implemented\n");
1221 static int mlx5e_alloc_icosq(struct mlx5e_channel *c,
1222 struct mlx5e_sq_param *param,
1223 struct mlx5e_icosq *sq,
1224 work_func_t recover_work_func)
1226 void *sqc_wq = MLX5_ADDR_OF(sqc, param->sqc, wq);
1227 struct mlx5_core_dev *mdev = c->mdev;
1228 struct mlx5_wq_cyc *wq = &sq->wq;
1232 sq->uar_map = mdev->mlx5e_res.hw_objs.bfreg.map;
1233 sq->reserved_room = param->stop_room;
1234 sq->max_sq_wqebbs = mlx5e_get_max_sq_wqebbs(mdev);
1236 param->wq.db_numa_node = cpu_to_node(c->cpu);
1237 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, wq, &sq->wq_ctrl);
1240 wq->db = &wq->db[MLX5_SND_DBR];
1242 err = mlx5e_alloc_icosq_db(sq, cpu_to_node(c->cpu));
1244 goto err_sq_wq_destroy;
1246 INIT_WORK(&sq->recover_work, recover_work_func);
1251 mlx5_wq_destroy(&sq->wq_ctrl);
1256 static void mlx5e_free_icosq(struct mlx5e_icosq *sq)
1258 mlx5e_free_icosq_db(sq);
1259 mlx5_wq_destroy(&sq->wq_ctrl);
1262 void mlx5e_free_txqsq_db(struct mlx5e_txqsq *sq)
1264 kvfree(sq->db.wqe_info);
1265 kvfree(sq->db.skb_fifo.fifo);
1266 kvfree(sq->db.dma_fifo);
1269 int mlx5e_alloc_txqsq_db(struct mlx5e_txqsq *sq, int numa)
1271 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1272 int df_sz = wq_sz * MLX5_SEND_WQEBB_NUM_DS;
1274 sq->db.dma_fifo = kvzalloc_node(array_size(df_sz,
1275 sizeof(*sq->db.dma_fifo)),
1277 sq->db.skb_fifo.fifo = kvzalloc_node(array_size(df_sz,
1278 sizeof(*sq->db.skb_fifo.fifo)),
1280 sq->db.wqe_info = kvzalloc_node(array_size(wq_sz,
1281 sizeof(*sq->db.wqe_info)),
1283 if (!sq->db.dma_fifo || !sq->db.skb_fifo.fifo || !sq->db.wqe_info) {
1284 mlx5e_free_txqsq_db(sq);
1288 sq->dma_fifo_mask = df_sz - 1;
1290 sq->db.skb_fifo.pc = &sq->skb_fifo_pc;
1291 sq->db.skb_fifo.cc = &sq->skb_fifo_cc;
1292 sq->db.skb_fifo.mask = df_sz - 1;
1297 static int mlx5e_alloc_txqsq(struct mlx5e_channel *c,
1299 struct mlx5e_params *params,
1300 struct mlx5e_sq_param *param,
1301 struct mlx5e_txqsq *sq,
1304 void *sqc_wq = MLX5_ADDR_OF(sqc, param->sqc, wq);
1305 struct mlx5_core_dev *mdev = c->mdev;
1306 struct mlx5_wq_cyc *wq = &sq->wq;
1310 sq->clock = &mdev->clock;
1311 sq->mkey_be = c->mkey_be;
1312 sq->netdev = c->netdev;
1316 sq->txq_ix = txq_ix;
1317 sq->uar_map = mdev->mlx5e_res.hw_objs.bfreg.map;
1318 sq->min_inline_mode = params->tx_min_inline_mode;
1319 sq->hw_mtu = MLX5E_SW2HW_MTU(params, params->sw_mtu);
1320 sq->max_sq_wqebbs = mlx5e_get_max_sq_wqebbs(mdev);
1321 sq->max_sq_mpw_wqebbs = mlx5e_get_sw_max_sq_mpw_wqebbs(sq->max_sq_wqebbs);
1322 INIT_WORK(&sq->recover_work, mlx5e_tx_err_cqe_work);
1323 if (!MLX5_CAP_ETH(mdev, wqe_vlan_insert))
1324 set_bit(MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE, &sq->state);
1325 if (mlx5_ipsec_device_caps(c->priv->mdev))
1326 set_bit(MLX5E_SQ_STATE_IPSEC, &sq->state);
1328 set_bit(MLX5E_SQ_STATE_MPWQE, &sq->state);
1329 sq->stop_room = param->stop_room;
1330 sq->ptp_cyc2time = mlx5_sq_ts_translator(mdev);
1332 param->wq.db_numa_node = cpu_to_node(c->cpu);
1333 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, wq, &sq->wq_ctrl);
1336 wq->db = &wq->db[MLX5_SND_DBR];
1338 err = mlx5e_alloc_txqsq_db(sq, cpu_to_node(c->cpu));
1340 goto err_sq_wq_destroy;
1342 INIT_WORK(&sq->dim.work, mlx5e_tx_dim_work);
1343 sq->dim.mode = params->tx_cq_moderation.cq_period_mode;
1348 mlx5_wq_destroy(&sq->wq_ctrl);
1353 void mlx5e_free_txqsq(struct mlx5e_txqsq *sq)
1355 mlx5e_free_txqsq_db(sq);
1356 mlx5_wq_destroy(&sq->wq_ctrl);
1359 static int mlx5e_create_sq(struct mlx5_core_dev *mdev,
1360 struct mlx5e_sq_param *param,
1361 struct mlx5e_create_sq_param *csp,
1371 inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
1372 sizeof(u64) * csp->wq_ctrl->buf.npages;
1373 in = kvzalloc(inlen, GFP_KERNEL);
1377 ts_format = mlx5_is_real_time_sq(mdev) ?
1378 MLX5_TIMESTAMP_FORMAT_REAL_TIME :
1379 MLX5_TIMESTAMP_FORMAT_FREE_RUNNING;
1380 sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
1381 wq = MLX5_ADDR_OF(sqc, sqc, wq);
1383 memcpy(sqc, param->sqc, sizeof(param->sqc));
1384 MLX5_SET(sqc, sqc, tis_lst_sz, csp->tis_lst_sz);
1385 MLX5_SET(sqc, sqc, tis_num_0, csp->tisn);
1386 MLX5_SET(sqc, sqc, cqn, csp->cqn);
1387 MLX5_SET(sqc, sqc, ts_cqe_to_dest_cqn, csp->ts_cqe_to_dest_cqn);
1388 MLX5_SET(sqc, sqc, ts_format, ts_format);
1391 if (MLX5_CAP_ETH(mdev, wqe_inline_mode) == MLX5_CAP_INLINE_MODE_VPORT_CONTEXT)
1392 MLX5_SET(sqc, sqc, min_wqe_inline_mode, csp->min_inline_mode);
1394 MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
1395 MLX5_SET(sqc, sqc, flush_in_error_en, 1);
1397 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
1398 MLX5_SET(wq, wq, uar_page, mdev->mlx5e_res.hw_objs.bfreg.index);
1399 MLX5_SET(wq, wq, log_wq_pg_sz, csp->wq_ctrl->buf.page_shift -
1400 MLX5_ADAPTER_PAGE_SHIFT);
1401 MLX5_SET64(wq, wq, dbr_addr, csp->wq_ctrl->db.dma);
1403 mlx5_fill_page_frag_array(&csp->wq_ctrl->buf,
1404 (__be64 *)MLX5_ADDR_OF(wq, wq, pas));
1406 err = mlx5_core_create_sq(mdev, in, inlen, sqn);
1413 int mlx5e_modify_sq(struct mlx5_core_dev *mdev, u32 sqn,
1414 struct mlx5e_modify_sq_param *p)
1422 inlen = MLX5_ST_SZ_BYTES(modify_sq_in);
1423 in = kvzalloc(inlen, GFP_KERNEL);
1427 sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx);
1429 MLX5_SET(modify_sq_in, in, sq_state, p->curr_state);
1430 MLX5_SET(sqc, sqc, state, p->next_state);
1431 if (p->rl_update && p->next_state == MLX5_SQC_STATE_RDY) {
1433 MLX5_SET(sqc, sqc, packet_pacing_rate_limit_index, p->rl_index);
1435 if (p->qos_update && p->next_state == MLX5_SQC_STATE_RDY) {
1437 MLX5_SET(sqc, sqc, qos_queue_group_id, p->qos_queue_group_id);
1439 MLX5_SET64(modify_sq_in, in, modify_bitmask, bitmask);
1441 err = mlx5_core_modify_sq(mdev, sqn, in);
1448 static void mlx5e_destroy_sq(struct mlx5_core_dev *mdev, u32 sqn)
1450 mlx5_core_destroy_sq(mdev, sqn);
1453 int mlx5e_create_sq_rdy(struct mlx5_core_dev *mdev,
1454 struct mlx5e_sq_param *param,
1455 struct mlx5e_create_sq_param *csp,
1456 u16 qos_queue_group_id,
1459 struct mlx5e_modify_sq_param msp = {0};
1462 err = mlx5e_create_sq(mdev, param, csp, sqn);
1466 msp.curr_state = MLX5_SQC_STATE_RST;
1467 msp.next_state = MLX5_SQC_STATE_RDY;
1468 if (qos_queue_group_id) {
1469 msp.qos_update = true;
1470 msp.qos_queue_group_id = qos_queue_group_id;
1472 err = mlx5e_modify_sq(mdev, *sqn, &msp);
1474 mlx5e_destroy_sq(mdev, *sqn);
1479 static int mlx5e_set_sq_maxrate(struct net_device *dev,
1480 struct mlx5e_txqsq *sq, u32 rate);
1482 int mlx5e_open_txqsq(struct mlx5e_channel *c, u32 tisn, int txq_ix,
1483 struct mlx5e_params *params, struct mlx5e_sq_param *param,
1484 struct mlx5e_txqsq *sq, int tc, u16 qos_queue_group_id,
1485 struct mlx5e_sq_stats *sq_stats)
1487 struct mlx5e_create_sq_param csp = {};
1491 err = mlx5e_alloc_txqsq(c, txq_ix, params, param, sq, tc);
1495 sq->stats = sq_stats;
1499 csp.cqn = sq->cq.mcq.cqn;
1500 csp.wq_ctrl = &sq->wq_ctrl;
1501 csp.min_inline_mode = sq->min_inline_mode;
1502 err = mlx5e_create_sq_rdy(c->mdev, param, &csp, qos_queue_group_id, &sq->sqn);
1504 goto err_free_txqsq;
1506 tx_rate = c->priv->tx_rates[sq->txq_ix];
1508 mlx5e_set_sq_maxrate(c->netdev, sq, tx_rate);
1510 if (params->tx_dim_enabled)
1511 sq->state |= BIT(MLX5E_SQ_STATE_AM);
1516 mlx5e_free_txqsq(sq);
1521 void mlx5e_activate_txqsq(struct mlx5e_txqsq *sq)
1523 sq->txq = netdev_get_tx_queue(sq->netdev, sq->txq_ix);
1524 set_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1525 netdev_tx_reset_queue(sq->txq);
1526 netif_tx_start_queue(sq->txq);
1529 void mlx5e_tx_disable_queue(struct netdev_queue *txq)
1531 __netif_tx_lock_bh(txq);
1532 netif_tx_stop_queue(txq);
1533 __netif_tx_unlock_bh(txq);
1536 void mlx5e_deactivate_txqsq(struct mlx5e_txqsq *sq)
1538 struct mlx5_wq_cyc *wq = &sq->wq;
1540 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1541 synchronize_net(); /* Sync with NAPI to prevent netif_tx_wake_queue. */
1543 mlx5e_tx_disable_queue(sq->txq);
1545 /* last doorbell out, godspeed .. */
1546 if (mlx5e_wqc_has_room_for(wq, sq->cc, sq->pc, 1)) {
1547 u16 pi = mlx5_wq_cyc_ctr2ix(wq, sq->pc);
1548 struct mlx5e_tx_wqe *nop;
1550 sq->db.wqe_info[pi] = (struct mlx5e_tx_wqe_info) {
1554 nop = mlx5e_post_nop(wq, sq->sqn, &sq->pc);
1555 mlx5e_notify_hw(wq, sq->pc, sq->uar_map, &nop->ctrl);
1559 void mlx5e_close_txqsq(struct mlx5e_txqsq *sq)
1561 struct mlx5_core_dev *mdev = sq->mdev;
1562 struct mlx5_rate_limit rl = {0};
1564 cancel_work_sync(&sq->dim.work);
1565 cancel_work_sync(&sq->recover_work);
1566 mlx5e_destroy_sq(mdev, sq->sqn);
1567 if (sq->rate_limit) {
1568 rl.rate = sq->rate_limit;
1569 mlx5_rl_remove_rate(mdev, &rl);
1571 mlx5e_free_txqsq_descs(sq);
1572 mlx5e_free_txqsq(sq);
1575 void mlx5e_tx_err_cqe_work(struct work_struct *recover_work)
1577 struct mlx5e_txqsq *sq = container_of(recover_work, struct mlx5e_txqsq,
1580 mlx5e_reporter_tx_err_cqe(sq);
1583 static int mlx5e_open_icosq(struct mlx5e_channel *c, struct mlx5e_params *params,
1584 struct mlx5e_sq_param *param, struct mlx5e_icosq *sq,
1585 work_func_t recover_work_func)
1587 struct mlx5e_create_sq_param csp = {};
1590 err = mlx5e_alloc_icosq(c, param, sq, recover_work_func);
1594 csp.cqn = sq->cq.mcq.cqn;
1595 csp.wq_ctrl = &sq->wq_ctrl;
1596 csp.min_inline_mode = params->tx_min_inline_mode;
1597 err = mlx5e_create_sq_rdy(c->mdev, param, &csp, 0, &sq->sqn);
1599 goto err_free_icosq;
1601 if (param->is_tls) {
1602 sq->ktls_resync = mlx5e_ktls_rx_resync_create_resp_list();
1603 if (IS_ERR(sq->ktls_resync)) {
1604 err = PTR_ERR(sq->ktls_resync);
1605 goto err_destroy_icosq;
1611 mlx5e_destroy_sq(c->mdev, sq->sqn);
1613 mlx5e_free_icosq(sq);
1618 void mlx5e_activate_icosq(struct mlx5e_icosq *icosq)
1620 set_bit(MLX5E_SQ_STATE_ENABLED, &icosq->state);
1623 void mlx5e_deactivate_icosq(struct mlx5e_icosq *icosq)
1625 clear_bit(MLX5E_SQ_STATE_ENABLED, &icosq->state);
1626 synchronize_net(); /* Sync with NAPI. */
1629 static void mlx5e_close_icosq(struct mlx5e_icosq *sq)
1631 struct mlx5e_channel *c = sq->channel;
1633 if (sq->ktls_resync)
1634 mlx5e_ktls_rx_resync_destroy_resp_list(sq->ktls_resync);
1635 mlx5e_destroy_sq(c->mdev, sq->sqn);
1636 mlx5e_free_icosq_descs(sq);
1637 mlx5e_free_icosq(sq);
1640 int mlx5e_open_xdpsq(struct mlx5e_channel *c, struct mlx5e_params *params,
1641 struct mlx5e_sq_param *param, struct xsk_buff_pool *xsk_pool,
1642 struct mlx5e_xdpsq *sq, bool is_redirect)
1644 struct mlx5e_create_sq_param csp = {};
1647 err = mlx5e_alloc_xdpsq(c, params, xsk_pool, param, sq, is_redirect);
1652 csp.tisn = c->priv->tisn[c->lag_port][0]; /* tc = 0 */
1653 csp.cqn = sq->cq.mcq.cqn;
1654 csp.wq_ctrl = &sq->wq_ctrl;
1655 csp.min_inline_mode = sq->min_inline_mode;
1656 set_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1658 /* Don't enable multi buffer on XDP_REDIRECT SQ, as it's not yet
1659 * supported by upstream, and there is no defined trigger to allow
1660 * transmitting redirected multi-buffer frames.
1662 if (param->is_xdp_mb && !is_redirect)
1663 set_bit(MLX5E_SQ_STATE_XDP_MULTIBUF, &sq->state);
1665 err = mlx5e_create_sq_rdy(c->mdev, param, &csp, 0, &sq->sqn);
1667 goto err_free_xdpsq;
1669 mlx5e_set_xmit_fp(sq, param->is_mpw);
1671 if (!param->is_mpw && !test_bit(MLX5E_SQ_STATE_XDP_MULTIBUF, &sq->state)) {
1672 unsigned int ds_cnt = MLX5E_TX_WQE_EMPTY_DS_COUNT + 1;
1673 unsigned int inline_hdr_sz = 0;
1676 if (sq->min_inline_mode != MLX5_INLINE_MODE_NONE) {
1677 inline_hdr_sz = MLX5E_XDP_MIN_INLINE;
1681 /* Pre initialize fixed WQE fields */
1682 for (i = 0; i < mlx5_wq_cyc_get_size(&sq->wq); i++) {
1683 struct mlx5e_tx_wqe *wqe = mlx5_wq_cyc_get_wqe(&sq->wq, i);
1684 struct mlx5_wqe_ctrl_seg *cseg = &wqe->ctrl;
1685 struct mlx5_wqe_eth_seg *eseg = &wqe->eth;
1686 struct mlx5_wqe_data_seg *dseg;
1688 sq->db.wqe_info[i] = (struct mlx5e_xdp_wqe_info) {
1693 cseg->qpn_ds = cpu_to_be32((sq->sqn << 8) | ds_cnt);
1694 eseg->inline_hdr.sz = cpu_to_be16(inline_hdr_sz);
1696 dseg = (struct mlx5_wqe_data_seg *)cseg + (ds_cnt - 1);
1697 dseg->lkey = sq->mkey_be;
1704 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1705 mlx5e_free_xdpsq(sq);
1710 void mlx5e_close_xdpsq(struct mlx5e_xdpsq *sq)
1712 struct mlx5e_channel *c = sq->channel;
1714 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1715 synchronize_net(); /* Sync with NAPI. */
1717 mlx5e_destroy_sq(c->mdev, sq->sqn);
1718 mlx5e_free_xdpsq_descs(sq);
1719 mlx5e_free_xdpsq(sq);
1722 static int mlx5e_alloc_cq_common(struct mlx5e_priv *priv,
1723 struct mlx5e_cq_param *param,
1724 struct mlx5e_cq *cq)
1726 struct mlx5_core_dev *mdev = priv->mdev;
1727 struct mlx5_core_cq *mcq = &cq->mcq;
1731 err = mlx5_cqwq_create(mdev, ¶m->wq, param->cqc, &cq->wq,
1737 mcq->set_ci_db = cq->wq_ctrl.db.db;
1738 mcq->arm_db = cq->wq_ctrl.db.db + 1;
1739 *mcq->set_ci_db = 0;
1741 mcq->vector = param->eq_ix;
1742 mcq->comp = mlx5e_completion_event;
1743 mcq->event = mlx5e_cq_error_event;
1745 for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) {
1746 struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i);
1752 cq->netdev = priv->netdev;
1758 static int mlx5e_alloc_cq(struct mlx5e_priv *priv,
1759 struct mlx5e_cq_param *param,
1760 struct mlx5e_create_cq_param *ccp,
1761 struct mlx5e_cq *cq)
1765 param->wq.buf_numa_node = ccp->node;
1766 param->wq.db_numa_node = ccp->node;
1767 param->eq_ix = ccp->ix;
1769 err = mlx5e_alloc_cq_common(priv, param, cq);
1771 cq->napi = ccp->napi;
1772 cq->ch_stats = ccp->ch_stats;
1777 static void mlx5e_free_cq(struct mlx5e_cq *cq)
1779 mlx5_wq_destroy(&cq->wq_ctrl);
1782 static int mlx5e_create_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param)
1784 u32 out[MLX5_ST_SZ_DW(create_cq_out)];
1785 struct mlx5_core_dev *mdev = cq->mdev;
1786 struct mlx5_core_cq *mcq = &cq->mcq;
1794 err = mlx5_vector2eqn(mdev, param->eq_ix, &eqn);
1798 inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
1799 sizeof(u64) * cq->wq_ctrl.buf.npages;
1800 in = kvzalloc(inlen, GFP_KERNEL);
1804 cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
1806 memcpy(cqc, param->cqc, sizeof(param->cqc));
1808 mlx5_fill_page_frag_array(&cq->wq_ctrl.buf,
1809 (__be64 *)MLX5_ADDR_OF(create_cq_in, in, pas));
1811 MLX5_SET(cqc, cqc, cq_period_mode, param->cq_period_mode);
1812 MLX5_SET(cqc, cqc, c_eqn_or_apu_element, eqn);
1813 MLX5_SET(cqc, cqc, uar_page, mdev->priv.uar->index);
1814 MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.buf.page_shift -
1815 MLX5_ADAPTER_PAGE_SHIFT);
1816 MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
1818 err = mlx5_core_create_cq(mdev, mcq, in, inlen, out, sizeof(out));
1830 static void mlx5e_destroy_cq(struct mlx5e_cq *cq)
1832 mlx5_core_destroy_cq(cq->mdev, &cq->mcq);
1835 int mlx5e_open_cq(struct mlx5e_priv *priv, struct dim_cq_moder moder,
1836 struct mlx5e_cq_param *param, struct mlx5e_create_cq_param *ccp,
1837 struct mlx5e_cq *cq)
1839 struct mlx5_core_dev *mdev = priv->mdev;
1842 err = mlx5e_alloc_cq(priv, param, ccp, cq);
1846 err = mlx5e_create_cq(cq, param);
1850 if (MLX5_CAP_GEN(mdev, cq_moderation))
1851 mlx5_core_modify_cq_moderation(mdev, &cq->mcq, moder.usec, moder.pkts);
1860 void mlx5e_close_cq(struct mlx5e_cq *cq)
1862 mlx5e_destroy_cq(cq);
1866 static int mlx5e_open_tx_cqs(struct mlx5e_channel *c,
1867 struct mlx5e_params *params,
1868 struct mlx5e_create_cq_param *ccp,
1869 struct mlx5e_channel_param *cparam)
1874 for (tc = 0; tc < c->num_tc; tc++) {
1875 err = mlx5e_open_cq(c->priv, params->tx_cq_moderation, &cparam->txq_sq.cqp,
1876 ccp, &c->sq[tc].cq);
1878 goto err_close_tx_cqs;
1884 for (tc--; tc >= 0; tc--)
1885 mlx5e_close_cq(&c->sq[tc].cq);
1890 static void mlx5e_close_tx_cqs(struct mlx5e_channel *c)
1894 for (tc = 0; tc < c->num_tc; tc++)
1895 mlx5e_close_cq(&c->sq[tc].cq);
1898 static int mlx5e_mqprio_txq_to_tc(struct netdev_tc_txq *tc_to_txq, unsigned int txq)
1902 for (tc = 0; tc < TC_MAX_QUEUE; tc++)
1903 if (txq - tc_to_txq[tc].offset < tc_to_txq[tc].count)
1906 WARN(1, "Unexpected TCs configuration. No match found for txq %u", txq);
1910 static int mlx5e_txq_get_qos_node_hw_id(struct mlx5e_params *params, int txq_ix,
1915 if (params->mqprio.mode != TC_MQPRIO_MODE_CHANNEL) {
1920 tc = mlx5e_mqprio_txq_to_tc(params->mqprio.tc_to_txq, txq_ix);
1924 if (tc >= params->mqprio.num_tc) {
1925 WARN(1, "Unexpected TCs configuration. tc %d is out of range of %u",
1926 tc, params->mqprio.num_tc);
1930 *hw_id = params->mqprio.channel.hw_id[tc];
1934 static int mlx5e_open_sqs(struct mlx5e_channel *c,
1935 struct mlx5e_params *params,
1936 struct mlx5e_channel_param *cparam)
1940 for (tc = 0; tc < mlx5e_get_dcb_num_tc(params); tc++) {
1941 int txq_ix = c->ix + tc * params->num_channels;
1942 u32 qos_queue_group_id;
1944 err = mlx5e_txq_get_qos_node_hw_id(params, txq_ix, &qos_queue_group_id);
1948 err = mlx5e_open_txqsq(c, c->priv->tisn[c->lag_port][tc], txq_ix,
1949 params, &cparam->txq_sq, &c->sq[tc], tc,
1951 &c->priv->channel_stats[c->ix]->sq[tc]);
1959 for (tc--; tc >= 0; tc--)
1960 mlx5e_close_txqsq(&c->sq[tc]);
1965 static void mlx5e_close_sqs(struct mlx5e_channel *c)
1969 for (tc = 0; tc < c->num_tc; tc++)
1970 mlx5e_close_txqsq(&c->sq[tc]);
1973 static int mlx5e_set_sq_maxrate(struct net_device *dev,
1974 struct mlx5e_txqsq *sq, u32 rate)
1976 struct mlx5e_priv *priv = netdev_priv(dev);
1977 struct mlx5_core_dev *mdev = priv->mdev;
1978 struct mlx5e_modify_sq_param msp = {0};
1979 struct mlx5_rate_limit rl = {0};
1983 if (rate == sq->rate_limit)
1987 if (sq->rate_limit) {
1988 rl.rate = sq->rate_limit;
1989 /* remove current rl index to free space to next ones */
1990 mlx5_rl_remove_rate(mdev, &rl);
1997 err = mlx5_rl_add_rate(mdev, &rl_index, &rl);
1999 netdev_err(dev, "Failed configuring rate %u: %d\n",
2005 msp.curr_state = MLX5_SQC_STATE_RDY;
2006 msp.next_state = MLX5_SQC_STATE_RDY;
2007 msp.rl_index = rl_index;
2008 msp.rl_update = true;
2009 err = mlx5e_modify_sq(mdev, sq->sqn, &msp);
2011 netdev_err(dev, "Failed configuring rate %u: %d\n",
2013 /* remove the rate from the table */
2015 mlx5_rl_remove_rate(mdev, &rl);
2019 sq->rate_limit = rate;
2023 static int mlx5e_set_tx_maxrate(struct net_device *dev, int index, u32 rate)
2025 struct mlx5e_priv *priv = netdev_priv(dev);
2026 struct mlx5_core_dev *mdev = priv->mdev;
2027 struct mlx5e_txqsq *sq = priv->txq2sq[index];
2030 if (!mlx5_rl_is_supported(mdev)) {
2031 netdev_err(dev, "Rate limiting is not supported on this device\n");
2035 /* rate is given in Mb/sec, HW config is in Kb/sec */
2038 /* Check whether rate in valid range, 0 is always valid */
2039 if (rate && !mlx5_rl_is_in_range(mdev, rate)) {
2040 netdev_err(dev, "TX rate %u, is not in range\n", rate);
2044 mutex_lock(&priv->state_lock);
2045 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
2046 err = mlx5e_set_sq_maxrate(dev, sq, rate);
2048 priv->tx_rates[index] = rate;
2049 mutex_unlock(&priv->state_lock);
2054 static int mlx5e_open_rxq_rq(struct mlx5e_channel *c, struct mlx5e_params *params,
2055 struct mlx5e_rq_param *rq_params)
2059 err = mlx5e_init_rxq_rq(c, params, &c->rq);
2063 return mlx5e_open_rq(params, rq_params, NULL, cpu_to_node(c->cpu), &c->rq);
2066 static int mlx5e_open_queues(struct mlx5e_channel *c,
2067 struct mlx5e_params *params,
2068 struct mlx5e_channel_param *cparam)
2070 struct dim_cq_moder icocq_moder = {0, 0};
2071 struct mlx5e_create_cq_param ccp;
2074 mlx5e_build_create_cq_param(&ccp, c);
2076 err = mlx5e_open_cq(c->priv, icocq_moder, &cparam->async_icosq.cqp, &ccp,
2077 &c->async_icosq.cq);
2081 err = mlx5e_open_cq(c->priv, icocq_moder, &cparam->icosq.cqp, &ccp,
2084 goto err_close_async_icosq_cq;
2086 err = mlx5e_open_tx_cqs(c, params, &ccp, cparam);
2088 goto err_close_icosq_cq;
2090 err = mlx5e_open_cq(c->priv, params->tx_cq_moderation, &cparam->xdp_sq.cqp, &ccp,
2093 goto err_close_tx_cqs;
2095 err = mlx5e_open_cq(c->priv, params->rx_cq_moderation, &cparam->rq.cqp, &ccp,
2098 goto err_close_xdp_tx_cqs;
2100 err = c->xdp ? mlx5e_open_cq(c->priv, params->tx_cq_moderation, &cparam->xdp_sq.cqp,
2101 &ccp, &c->rq_xdpsq.cq) : 0;
2103 goto err_close_rx_cq;
2105 spin_lock_init(&c->async_icosq_lock);
2107 err = mlx5e_open_icosq(c, params, &cparam->async_icosq, &c->async_icosq,
2108 mlx5e_async_icosq_err_cqe_work);
2110 goto err_close_xdpsq_cq;
2112 mutex_init(&c->icosq_recovery_lock);
2114 err = mlx5e_open_icosq(c, params, &cparam->icosq, &c->icosq,
2115 mlx5e_icosq_err_cqe_work);
2117 goto err_close_async_icosq;
2119 err = mlx5e_open_sqs(c, params, cparam);
2121 goto err_close_icosq;
2123 err = mlx5e_open_rxq_rq(c, params, &cparam->rq);
2128 err = mlx5e_open_xdpsq(c, params, &cparam->xdp_sq, NULL,
2129 &c->rq_xdpsq, false);
2134 err = mlx5e_open_xdpsq(c, params, &cparam->xdp_sq, NULL, &c->xdpsq, true);
2136 goto err_close_xdp_sq;
2142 mlx5e_close_xdpsq(&c->rq_xdpsq);
2145 mlx5e_close_rq(&c->rq);
2151 mlx5e_close_icosq(&c->icosq);
2153 err_close_async_icosq:
2154 mlx5e_close_icosq(&c->async_icosq);
2158 mlx5e_close_cq(&c->rq_xdpsq.cq);
2161 mlx5e_close_cq(&c->rq.cq);
2163 err_close_xdp_tx_cqs:
2164 mlx5e_close_cq(&c->xdpsq.cq);
2167 mlx5e_close_tx_cqs(c);
2170 mlx5e_close_cq(&c->icosq.cq);
2172 err_close_async_icosq_cq:
2173 mlx5e_close_cq(&c->async_icosq.cq);
2178 static void mlx5e_close_queues(struct mlx5e_channel *c)
2180 mlx5e_close_xdpsq(&c->xdpsq);
2182 mlx5e_close_xdpsq(&c->rq_xdpsq);
2183 /* The same ICOSQ is used for UMRs for both RQ and XSKRQ. */
2184 cancel_work_sync(&c->icosq.recover_work);
2185 mlx5e_close_rq(&c->rq);
2187 mlx5e_close_icosq(&c->icosq);
2188 mutex_destroy(&c->icosq_recovery_lock);
2189 mlx5e_close_icosq(&c->async_icosq);
2191 mlx5e_close_cq(&c->rq_xdpsq.cq);
2192 mlx5e_close_cq(&c->rq.cq);
2193 mlx5e_close_cq(&c->xdpsq.cq);
2194 mlx5e_close_tx_cqs(c);
2195 mlx5e_close_cq(&c->icosq.cq);
2196 mlx5e_close_cq(&c->async_icosq.cq);
2199 static u8 mlx5e_enumerate_lag_port(struct mlx5_core_dev *mdev, int ix)
2201 u16 port_aff_bias = mlx5_core_is_pf(mdev) ? 0 : MLX5_CAP_GEN(mdev, vhca_id);
2203 return (ix + port_aff_bias) % mlx5e_get_num_lag_ports(mdev);
2206 static int mlx5e_channel_stats_alloc(struct mlx5e_priv *priv, int ix, int cpu)
2208 if (ix > priv->stats_nch) {
2209 netdev_warn(priv->netdev, "Unexpected channel stats index %d > %d\n", ix,
2214 if (priv->channel_stats[ix])
2217 /* Asymmetric dynamic memory allocation.
2218 * Freed in mlx5e_priv_arrays_free, not on channel closure.
2220 mlx5e_dbg(DRV, priv, "Creating channel stats %d\n", ix);
2221 priv->channel_stats[ix] = kvzalloc_node(sizeof(**priv->channel_stats),
2222 GFP_KERNEL, cpu_to_node(cpu));
2223 if (!priv->channel_stats[ix])
2230 void mlx5e_trigger_napi_icosq(struct mlx5e_channel *c)
2232 spin_lock_bh(&c->async_icosq_lock);
2233 mlx5e_trigger_irq(&c->async_icosq);
2234 spin_unlock_bh(&c->async_icosq_lock);
2237 void mlx5e_trigger_napi_sched(struct napi_struct *napi)
2240 napi_schedule(napi);
2244 static int mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
2245 struct mlx5e_params *params,
2246 struct mlx5e_channel_param *cparam,
2247 struct xsk_buff_pool *xsk_pool,
2248 struct mlx5e_channel **cp)
2250 int cpu = cpumask_first(mlx5_comp_irq_get_affinity_mask(priv->mdev, ix));
2251 struct net_device *netdev = priv->netdev;
2252 struct mlx5e_xsk_param xsk;
2253 struct mlx5e_channel *c;
2257 err = mlx5_vector2irqn(priv->mdev, ix, &irq);
2261 err = mlx5e_channel_stats_alloc(priv, ix, cpu);
2265 c = kvzalloc_node(sizeof(*c), GFP_KERNEL, cpu_to_node(cpu));
2270 c->mdev = priv->mdev;
2271 c->tstamp = &priv->tstamp;
2274 c->pdev = mlx5_core_dma_dev(priv->mdev);
2275 c->netdev = priv->netdev;
2276 c->mkey_be = cpu_to_be32(priv->mdev->mlx5e_res.hw_objs.mkey);
2277 c->num_tc = mlx5e_get_dcb_num_tc(params);
2278 c->xdp = !!params->xdp_prog;
2279 c->stats = &priv->channel_stats[ix]->ch;
2280 c->aff_mask = irq_get_effective_affinity_mask(irq);
2281 c->lag_port = mlx5e_enumerate_lag_port(priv->mdev, ix);
2283 netif_napi_add(netdev, &c->napi, mlx5e_napi_poll, 64);
2285 err = mlx5e_open_queues(c, params, cparam);
2290 mlx5e_build_xsk_param(xsk_pool, &xsk);
2291 err = mlx5e_open_xsk(priv, params, &xsk, xsk_pool, c);
2293 goto err_close_queues;
2301 mlx5e_close_queues(c);
2304 netif_napi_del(&c->napi);
2311 static void mlx5e_activate_channel(struct mlx5e_channel *c)
2315 napi_enable(&c->napi);
2317 for (tc = 0; tc < c->num_tc; tc++)
2318 mlx5e_activate_txqsq(&c->sq[tc]);
2319 mlx5e_activate_icosq(&c->icosq);
2320 mlx5e_activate_icosq(&c->async_icosq);
2321 mlx5e_activate_rq(&c->rq);
2323 if (test_bit(MLX5E_CHANNEL_STATE_XSK, c->state))
2324 mlx5e_activate_xsk(c);
2326 mlx5e_trigger_napi_icosq(c);
2329 static void mlx5e_deactivate_channel(struct mlx5e_channel *c)
2333 if (test_bit(MLX5E_CHANNEL_STATE_XSK, c->state))
2334 mlx5e_deactivate_xsk(c);
2336 mlx5e_deactivate_rq(&c->rq);
2337 mlx5e_deactivate_icosq(&c->async_icosq);
2338 mlx5e_deactivate_icosq(&c->icosq);
2339 for (tc = 0; tc < c->num_tc; tc++)
2340 mlx5e_deactivate_txqsq(&c->sq[tc]);
2341 mlx5e_qos_deactivate_queues(c);
2343 napi_disable(&c->napi);
2346 static void mlx5e_close_channel(struct mlx5e_channel *c)
2348 if (test_bit(MLX5E_CHANNEL_STATE_XSK, c->state))
2350 mlx5e_close_queues(c);
2351 mlx5e_qos_close_queues(c);
2352 netif_napi_del(&c->napi);
2357 int mlx5e_open_channels(struct mlx5e_priv *priv,
2358 struct mlx5e_channels *chs)
2360 struct mlx5e_channel_param *cparam;
2364 chs->num = chs->params.num_channels;
2366 chs->c = kcalloc(chs->num, sizeof(struct mlx5e_channel *), GFP_KERNEL);
2367 cparam = kvzalloc(sizeof(struct mlx5e_channel_param), GFP_KERNEL);
2368 if (!chs->c || !cparam)
2371 err = mlx5e_build_channel_param(priv->mdev, &chs->params, priv->q_counter, cparam);
2375 for (i = 0; i < chs->num; i++) {
2376 struct xsk_buff_pool *xsk_pool = NULL;
2378 if (chs->params.xdp_prog)
2379 xsk_pool = mlx5e_xsk_get_pool(&chs->params, chs->params.xsk, i);
2381 err = mlx5e_open_channel(priv, i, &chs->params, cparam, xsk_pool, &chs->c[i]);
2383 goto err_close_channels;
2386 if (MLX5E_GET_PFLAG(&chs->params, MLX5E_PFLAG_TX_PORT_TS) || chs->params.ptp_rx) {
2387 err = mlx5e_ptp_open(priv, &chs->params, chs->c[0]->lag_port, &chs->ptp);
2389 goto err_close_channels;
2393 err = mlx5e_qos_open_queues(priv, chs);
2398 mlx5e_health_channels_update(priv);
2404 mlx5e_ptp_close(chs->ptp);
2407 for (i--; i >= 0; i--)
2408 mlx5e_close_channel(chs->c[i]);
2417 static void mlx5e_activate_channels(struct mlx5e_channels *chs)
2421 for (i = 0; i < chs->num; i++)
2422 mlx5e_activate_channel(chs->c[i]);
2425 mlx5e_ptp_activate_channel(chs->ptp);
2428 #define MLX5E_RQ_WQES_TIMEOUT 20000 /* msecs */
2430 static int mlx5e_wait_channels_min_rx_wqes(struct mlx5e_channels *chs)
2435 for (i = 0; i < chs->num; i++) {
2436 int timeout = err ? 0 : MLX5E_RQ_WQES_TIMEOUT;
2438 err |= mlx5e_wait_for_min_rx_wqes(&chs->c[i]->rq, timeout);
2440 /* Don't wait on the XSK RQ, because the newer xdpsock sample
2441 * doesn't provide any Fill Ring entries at the setup stage.
2445 return err ? -ETIMEDOUT : 0;
2448 static void mlx5e_deactivate_channels(struct mlx5e_channels *chs)
2453 mlx5e_ptp_deactivate_channel(chs->ptp);
2455 for (i = 0; i < chs->num; i++)
2456 mlx5e_deactivate_channel(chs->c[i]);
2459 void mlx5e_close_channels(struct mlx5e_channels *chs)
2464 mlx5e_ptp_close(chs->ptp);
2467 for (i = 0; i < chs->num; i++)
2468 mlx5e_close_channel(chs->c[i]);
2474 static int mlx5e_modify_tirs_packet_merge(struct mlx5e_priv *priv)
2476 struct mlx5e_rx_res *res = priv->rx_res;
2478 return mlx5e_rx_res_packet_merge_set_param(res, &priv->channels.params.packet_merge);
2481 static MLX5E_DEFINE_PREACTIVATE_WRAPPER_CTX(mlx5e_modify_tirs_packet_merge);
2483 static int mlx5e_set_mtu(struct mlx5_core_dev *mdev,
2484 struct mlx5e_params *params, u16 mtu)
2486 u16 hw_mtu = MLX5E_SW2HW_MTU(params, mtu);
2489 err = mlx5_set_port_mtu(mdev, hw_mtu, 1);
2493 /* Update vport context MTU */
2494 mlx5_modify_nic_vport_mtu(mdev, hw_mtu);
2498 static void mlx5e_query_mtu(struct mlx5_core_dev *mdev,
2499 struct mlx5e_params *params, u16 *mtu)
2504 err = mlx5_query_nic_vport_mtu(mdev, &hw_mtu);
2505 if (err || !hw_mtu) /* fallback to port oper mtu */
2506 mlx5_query_port_oper_mtu(mdev, &hw_mtu, 1);
2508 *mtu = MLX5E_HW2SW_MTU(params, hw_mtu);
2511 int mlx5e_set_dev_port_mtu(struct mlx5e_priv *priv)
2513 struct mlx5e_params *params = &priv->channels.params;
2514 struct net_device *netdev = priv->netdev;
2515 struct mlx5_core_dev *mdev = priv->mdev;
2519 err = mlx5e_set_mtu(mdev, params, params->sw_mtu);
2523 mlx5e_query_mtu(mdev, params, &mtu);
2524 if (mtu != params->sw_mtu)
2525 netdev_warn(netdev, "%s: VPort MTU %d is different than netdev mtu %d\n",
2526 __func__, mtu, params->sw_mtu);
2528 params->sw_mtu = mtu;
2532 MLX5E_DEFINE_PREACTIVATE_WRAPPER_CTX(mlx5e_set_dev_port_mtu);
2534 void mlx5e_set_netdev_mtu_boundaries(struct mlx5e_priv *priv)
2536 struct mlx5e_params *params = &priv->channels.params;
2537 struct net_device *netdev = priv->netdev;
2538 struct mlx5_core_dev *mdev = priv->mdev;
2541 /* MTU range: 68 - hw-specific max */
2542 netdev->min_mtu = ETH_MIN_MTU;
2544 mlx5_query_port_max_mtu(mdev, &max_mtu, 1);
2545 netdev->max_mtu = min_t(unsigned int, MLX5E_HW2SW_MTU(params, max_mtu),
2549 static int mlx5e_netdev_set_tcs(struct net_device *netdev, u16 nch, u8 ntc,
2550 struct netdev_tc_txq *tc_to_txq)
2554 netdev_reset_tc(netdev);
2559 err = netdev_set_num_tc(netdev, ntc);
2561 netdev_WARN(netdev, "netdev_set_num_tc failed (%d), ntc = %d\n", err, ntc);
2565 for (tc = 0; tc < ntc; tc++) {
2568 count = tc_to_txq[tc].count;
2569 offset = tc_to_txq[tc].offset;
2570 netdev_set_tc_queue(netdev, tc, count, offset);
2576 int mlx5e_update_tx_netdev_queues(struct mlx5e_priv *priv)
2578 int nch, ntc, num_txqs, err;
2582 qos_queues = mlx5e_htb_cur_leaf_nodes(priv->htb);
2584 nch = priv->channels.params.num_channels;
2585 ntc = mlx5e_get_dcb_num_tc(&priv->channels.params);
2586 num_txqs = nch * ntc + qos_queues;
2587 if (MLX5E_GET_PFLAG(&priv->channels.params, MLX5E_PFLAG_TX_PORT_TS))
2590 mlx5e_dbg(DRV, priv, "Setting num_txqs %d\n", num_txqs);
2591 err = netif_set_real_num_tx_queues(priv->netdev, num_txqs);
2593 netdev_warn(priv->netdev, "netif_set_real_num_tx_queues failed, %d\n", err);
2598 static int mlx5e_update_netdev_queues(struct mlx5e_priv *priv)
2600 struct netdev_tc_txq old_tc_to_txq[TC_MAX_QUEUE], *tc_to_txq;
2601 struct net_device *netdev = priv->netdev;
2602 int old_num_txqs, old_ntc;
2603 int num_rxqs, nch, ntc;
2607 old_num_txqs = netdev->real_num_tx_queues;
2608 old_ntc = netdev->num_tc ? : 1;
2609 for (i = 0; i < ARRAY_SIZE(old_tc_to_txq); i++)
2610 old_tc_to_txq[i] = netdev->tc_to_txq[i];
2612 nch = priv->channels.params.num_channels;
2613 ntc = priv->channels.params.mqprio.num_tc;
2614 num_rxqs = nch * priv->profile->rq_groups;
2615 tc_to_txq = priv->channels.params.mqprio.tc_to_txq;
2617 err = mlx5e_netdev_set_tcs(netdev, nch, ntc, tc_to_txq);
2620 err = mlx5e_update_tx_netdev_queues(priv);
2623 err = netif_set_real_num_rx_queues(netdev, num_rxqs);
2625 netdev_warn(netdev, "netif_set_real_num_rx_queues failed, %d\n", err);
2632 /* netif_set_real_num_rx_queues could fail only when nch increased. Only
2633 * one of nch and ntc is changed in this function. That means, the call
2634 * to netif_set_real_num_tx_queues below should not fail, because it
2635 * decreases the number of TX queues.
2637 WARN_ON_ONCE(netif_set_real_num_tx_queues(netdev, old_num_txqs));
2640 WARN_ON_ONCE(mlx5e_netdev_set_tcs(netdev, old_num_txqs / old_ntc, old_ntc,
2646 static MLX5E_DEFINE_PREACTIVATE_WRAPPER_CTX(mlx5e_update_netdev_queues);
2648 static void mlx5e_set_default_xps_cpumasks(struct mlx5e_priv *priv,
2649 struct mlx5e_params *params)
2651 struct mlx5_core_dev *mdev = priv->mdev;
2652 int num_comp_vectors, ix, irq;
2654 num_comp_vectors = mlx5_comp_vectors_count(mdev);
2656 for (ix = 0; ix < params->num_channels; ix++) {
2657 cpumask_clear(priv->scratchpad.cpumask);
2659 for (irq = ix; irq < num_comp_vectors; irq += params->num_channels) {
2660 int cpu = cpumask_first(mlx5_comp_irq_get_affinity_mask(mdev, irq));
2662 cpumask_set_cpu(cpu, priv->scratchpad.cpumask);
2665 netif_set_xps_queue(priv->netdev, priv->scratchpad.cpumask, ix);
2669 static int mlx5e_num_channels_changed(struct mlx5e_priv *priv)
2671 u16 count = priv->channels.params.num_channels;
2674 err = mlx5e_update_netdev_queues(priv);
2678 mlx5e_set_default_xps_cpumasks(priv, &priv->channels.params);
2680 /* This function may be called on attach, before priv->rx_res is created. */
2681 if (!netif_is_rxfh_configured(priv->netdev) && priv->rx_res)
2682 mlx5e_rx_res_rss_set_indir_uniform(priv->rx_res, count);
2687 MLX5E_DEFINE_PREACTIVATE_WRAPPER_CTX(mlx5e_num_channels_changed);
2689 static void mlx5e_build_txq_maps(struct mlx5e_priv *priv)
2691 int i, ch, tc, num_tc;
2693 ch = priv->channels.num;
2694 num_tc = mlx5e_get_dcb_num_tc(&priv->channels.params);
2696 for (i = 0; i < ch; i++) {
2697 for (tc = 0; tc < num_tc; tc++) {
2698 struct mlx5e_channel *c = priv->channels.c[i];
2699 struct mlx5e_txqsq *sq = &c->sq[tc];
2701 priv->txq2sq[sq->txq_ix] = sq;
2705 if (!priv->channels.ptp)
2708 if (!test_bit(MLX5E_PTP_STATE_TX, priv->channels.ptp->state))
2711 for (tc = 0; tc < num_tc; tc++) {
2712 struct mlx5e_ptp *c = priv->channels.ptp;
2713 struct mlx5e_txqsq *sq = &c->ptpsq[tc].txqsq;
2715 priv->txq2sq[sq->txq_ix] = sq;
2719 /* Make the change to txq2sq visible before the queue is started.
2720 * As mlx5e_xmit runs under a spinlock, there is an implicit ACQUIRE,
2721 * which pairs with this barrier.
2726 void mlx5e_activate_priv_channels(struct mlx5e_priv *priv)
2728 mlx5e_build_txq_maps(priv);
2729 mlx5e_activate_channels(&priv->channels);
2731 mlx5e_qos_activate_queues(priv);
2732 mlx5e_xdp_tx_enable(priv);
2734 /* dev_watchdog() wants all TX queues to be started when the carrier is
2735 * OK, including the ones in range real_num_tx_queues..num_tx_queues-1.
2736 * Make it happy to avoid TX timeout false alarms.
2738 netif_tx_start_all_queues(priv->netdev);
2740 if (mlx5e_is_vport_rep(priv))
2741 mlx5e_add_sqs_fwd_rules(priv);
2743 mlx5e_wait_channels_min_rx_wqes(&priv->channels);
2746 mlx5e_rx_res_channels_activate(priv->rx_res, &priv->channels);
2749 void mlx5e_deactivate_priv_channels(struct mlx5e_priv *priv)
2752 mlx5e_rx_res_channels_deactivate(priv->rx_res);
2754 if (mlx5e_is_vport_rep(priv))
2755 mlx5e_remove_sqs_fwd_rules(priv);
2757 /* The results of ndo_select_queue are unreliable, while netdev config
2758 * is being changed (real_num_tx_queues, num_tc). Stop all queues to
2759 * prevent ndo_start_xmit from being called, so that it can assume that
2760 * the selected queue is always valid.
2762 netif_tx_disable(priv->netdev);
2764 mlx5e_xdp_tx_disable(priv);
2765 mlx5e_deactivate_channels(&priv->channels);
2768 static int mlx5e_switch_priv_params(struct mlx5e_priv *priv,
2769 struct mlx5e_params *new_params,
2770 mlx5e_fp_preactivate preactivate,
2773 struct mlx5e_params old_params;
2775 old_params = priv->channels.params;
2776 priv->channels.params = *new_params;
2781 err = preactivate(priv, context);
2783 priv->channels.params = old_params;
2791 static int mlx5e_switch_priv_channels(struct mlx5e_priv *priv,
2792 struct mlx5e_channels *new_chs,
2793 mlx5e_fp_preactivate preactivate,
2796 struct net_device *netdev = priv->netdev;
2797 struct mlx5e_channels old_chs;
2801 carrier_ok = netif_carrier_ok(netdev);
2802 netif_carrier_off(netdev);
2804 mlx5e_deactivate_priv_channels(priv);
2806 old_chs = priv->channels;
2807 priv->channels = *new_chs;
2809 /* New channels are ready to roll, call the preactivate hook if needed
2810 * to modify HW settings or update kernel parameters.
2813 err = preactivate(priv, context);
2815 priv->channels = old_chs;
2820 mlx5e_close_channels(&old_chs);
2821 priv->profile->update_rx(priv);
2823 mlx5e_selq_apply(&priv->selq);
2825 mlx5e_activate_priv_channels(priv);
2827 /* return carrier back if needed */
2829 netif_carrier_on(netdev);
2834 int mlx5e_safe_switch_params(struct mlx5e_priv *priv,
2835 struct mlx5e_params *params,
2836 mlx5e_fp_preactivate preactivate,
2837 void *context, bool reset)
2839 struct mlx5e_channels new_chs = {};
2842 reset &= test_bit(MLX5E_STATE_OPENED, &priv->state);
2844 return mlx5e_switch_priv_params(priv, params, preactivate, context);
2846 new_chs.params = *params;
2848 mlx5e_selq_prepare_params(&priv->selq, &new_chs.params);
2850 err = mlx5e_open_channels(priv, &new_chs);
2852 goto err_cancel_selq;
2854 err = mlx5e_switch_priv_channels(priv, &new_chs, preactivate, context);
2861 mlx5e_close_channels(&new_chs);
2864 mlx5e_selq_cancel(&priv->selq);
2868 int mlx5e_safe_reopen_channels(struct mlx5e_priv *priv)
2870 return mlx5e_safe_switch_params(priv, &priv->channels.params, NULL, NULL, true);
2873 void mlx5e_timestamp_init(struct mlx5e_priv *priv)
2875 priv->tstamp.tx_type = HWTSTAMP_TX_OFF;
2876 priv->tstamp.rx_filter = HWTSTAMP_FILTER_NONE;
2879 static void mlx5e_modify_admin_state(struct mlx5_core_dev *mdev,
2880 enum mlx5_port_status state)
2882 struct mlx5_eswitch *esw = mdev->priv.eswitch;
2883 int vport_admin_state;
2885 mlx5_set_port_admin_status(mdev, state);
2887 if (mlx5_eswitch_mode(mdev) == MLX5_ESWITCH_OFFLOADS ||
2888 !MLX5_CAP_GEN(mdev, uplink_follow))
2891 if (state == MLX5_PORT_UP)
2892 vport_admin_state = MLX5_VPORT_ADMIN_STATE_AUTO;
2894 vport_admin_state = MLX5_VPORT_ADMIN_STATE_DOWN;
2896 mlx5_eswitch_set_vport_state(esw, MLX5_VPORT_UPLINK, vport_admin_state);
2899 int mlx5e_open_locked(struct net_device *netdev)
2901 struct mlx5e_priv *priv = netdev_priv(netdev);
2904 mlx5e_selq_prepare_params(&priv->selq, &priv->channels.params);
2906 set_bit(MLX5E_STATE_OPENED, &priv->state);
2908 err = mlx5e_open_channels(priv, &priv->channels);
2910 goto err_clear_state_opened_flag;
2912 priv->profile->update_rx(priv);
2913 mlx5e_selq_apply(&priv->selq);
2914 mlx5e_activate_priv_channels(priv);
2915 mlx5e_apply_traps(priv, true);
2916 if (priv->profile->update_carrier)
2917 priv->profile->update_carrier(priv);
2919 mlx5e_queue_update_stats(priv);
2922 err_clear_state_opened_flag:
2923 clear_bit(MLX5E_STATE_OPENED, &priv->state);
2924 mlx5e_selq_cancel(&priv->selq);
2928 int mlx5e_open(struct net_device *netdev)
2930 struct mlx5e_priv *priv = netdev_priv(netdev);
2933 mutex_lock(&priv->state_lock);
2934 err = mlx5e_open_locked(netdev);
2936 mlx5e_modify_admin_state(priv->mdev, MLX5_PORT_UP);
2937 mutex_unlock(&priv->state_lock);
2942 int mlx5e_close_locked(struct net_device *netdev)
2944 struct mlx5e_priv *priv = netdev_priv(netdev);
2946 /* May already be CLOSED in case a previous configuration operation
2947 * (e.g RX/TX queue size change) that involves close&open failed.
2949 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
2952 mlx5e_apply_traps(priv, false);
2953 clear_bit(MLX5E_STATE_OPENED, &priv->state);
2955 netif_carrier_off(priv->netdev);
2956 mlx5e_deactivate_priv_channels(priv);
2957 mlx5e_close_channels(&priv->channels);
2962 int mlx5e_close(struct net_device *netdev)
2964 struct mlx5e_priv *priv = netdev_priv(netdev);
2967 if (!netif_device_present(netdev))
2970 mutex_lock(&priv->state_lock);
2971 mlx5e_modify_admin_state(priv->mdev, MLX5_PORT_DOWN);
2972 err = mlx5e_close_locked(netdev);
2973 mutex_unlock(&priv->state_lock);
2978 static void mlx5e_free_drop_rq(struct mlx5e_rq *rq)
2980 mlx5_wq_destroy(&rq->wq_ctrl);
2983 static int mlx5e_alloc_drop_rq(struct mlx5_core_dev *mdev,
2984 struct mlx5e_rq *rq,
2985 struct mlx5e_rq_param *param)
2987 void *rqc = param->rqc;
2988 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
2991 param->wq.db_numa_node = param->wq.buf_numa_node;
2993 err = mlx5_wq_cyc_create(mdev, ¶m->wq, rqc_wq, &rq->wqe.wq,
2998 /* Mark as unused given "Drop-RQ" packets never reach XDP */
2999 xdp_rxq_info_unused(&rq->xdp_rxq);
3006 static int mlx5e_alloc_drop_cq(struct mlx5e_priv *priv,
3007 struct mlx5e_cq *cq,
3008 struct mlx5e_cq_param *param)
3010 struct mlx5_core_dev *mdev = priv->mdev;
3012 param->wq.buf_numa_node = dev_to_node(mlx5_core_dma_dev(mdev));
3013 param->wq.db_numa_node = dev_to_node(mlx5_core_dma_dev(mdev));
3015 return mlx5e_alloc_cq_common(priv, param, cq);
3018 int mlx5e_open_drop_rq(struct mlx5e_priv *priv,
3019 struct mlx5e_rq *drop_rq)
3021 struct mlx5_core_dev *mdev = priv->mdev;
3022 struct mlx5e_cq_param cq_param = {};
3023 struct mlx5e_rq_param rq_param = {};
3024 struct mlx5e_cq *cq = &drop_rq->cq;
3027 mlx5e_build_drop_rq_param(mdev, priv->drop_rq_q_counter, &rq_param);
3029 err = mlx5e_alloc_drop_cq(priv, cq, &cq_param);
3033 err = mlx5e_create_cq(cq, &cq_param);
3037 err = mlx5e_alloc_drop_rq(mdev, drop_rq, &rq_param);
3039 goto err_destroy_cq;
3041 err = mlx5e_create_rq(drop_rq, &rq_param);
3045 err = mlx5e_modify_rq_state(drop_rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
3047 mlx5_core_warn(priv->mdev, "modify_rq_state failed, rx_if_down_packets won't be counted %d\n", err);
3052 mlx5e_free_drop_rq(drop_rq);
3055 mlx5e_destroy_cq(cq);
3063 void mlx5e_close_drop_rq(struct mlx5e_rq *drop_rq)
3065 mlx5e_destroy_rq(drop_rq);
3066 mlx5e_free_drop_rq(drop_rq);
3067 mlx5e_destroy_cq(&drop_rq->cq);
3068 mlx5e_free_cq(&drop_rq->cq);
3071 int mlx5e_create_tis(struct mlx5_core_dev *mdev, void *in, u32 *tisn)
3073 void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
3075 MLX5_SET(tisc, tisc, transport_domain, mdev->mlx5e_res.hw_objs.td.tdn);
3077 if (MLX5_GET(tisc, tisc, tls_en))
3078 MLX5_SET(tisc, tisc, pd, mdev->mlx5e_res.hw_objs.pdn);
3080 if (mlx5_lag_is_lacp_owner(mdev))
3081 MLX5_SET(tisc, tisc, strict_lag_tx_port_affinity, 1);
3083 return mlx5_core_create_tis(mdev, in, tisn);
3086 void mlx5e_destroy_tis(struct mlx5_core_dev *mdev, u32 tisn)
3088 mlx5_core_destroy_tis(mdev, tisn);
3091 void mlx5e_destroy_tises(struct mlx5e_priv *priv)
3095 for (i = 0; i < mlx5e_get_num_lag_ports(priv->mdev); i++)
3096 for (tc = 0; tc < priv->profile->max_tc; tc++)
3097 mlx5e_destroy_tis(priv->mdev, priv->tisn[i][tc]);
3100 static bool mlx5e_lag_should_assign_affinity(struct mlx5_core_dev *mdev)
3102 return MLX5_CAP_GEN(mdev, lag_tx_port_affinity) && mlx5e_get_num_lag_ports(mdev) > 1;
3105 int mlx5e_create_tises(struct mlx5e_priv *priv)
3110 for (i = 0; i < mlx5e_get_num_lag_ports(priv->mdev); i++) {
3111 for (tc = 0; tc < priv->profile->max_tc; tc++) {
3112 u32 in[MLX5_ST_SZ_DW(create_tis_in)] = {};
3115 tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
3117 MLX5_SET(tisc, tisc, prio, tc << 1);
3119 if (mlx5e_lag_should_assign_affinity(priv->mdev))
3120 MLX5_SET(tisc, tisc, lag_tx_port_affinity, i + 1);
3122 err = mlx5e_create_tis(priv->mdev, in, &priv->tisn[i][tc]);
3124 goto err_close_tises;
3131 for (; i >= 0; i--) {
3132 for (tc--; tc >= 0; tc--)
3133 mlx5e_destroy_tis(priv->mdev, priv->tisn[i][tc]);
3134 tc = priv->profile->max_tc;
3140 static void mlx5e_cleanup_nic_tx(struct mlx5e_priv *priv)
3142 if (priv->mqprio_rl) {
3143 mlx5e_mqprio_rl_cleanup(priv->mqprio_rl);
3144 mlx5e_mqprio_rl_free(priv->mqprio_rl);
3145 priv->mqprio_rl = NULL;
3147 mlx5e_accel_cleanup_tx(priv);
3148 mlx5e_destroy_tises(priv);
3151 static int mlx5e_modify_channels_scatter_fcs(struct mlx5e_channels *chs, bool enable)
3156 for (i = 0; i < chs->num; i++) {
3157 err = mlx5e_modify_rq_scatter_fcs(&chs->c[i]->rq, enable);
3165 static int mlx5e_modify_channels_vsd(struct mlx5e_channels *chs, bool vsd)
3170 for (i = 0; i < chs->num; i++) {
3171 err = mlx5e_modify_rq_vsd(&chs->c[i]->rq, vsd);
3175 if (chs->ptp && test_bit(MLX5E_PTP_STATE_RX, chs->ptp->state))
3176 return mlx5e_modify_rq_vsd(&chs->ptp->rq, vsd);
3181 static void mlx5e_mqprio_build_default_tc_to_txq(struct netdev_tc_txq *tc_to_txq,
3186 memset(tc_to_txq, 0, sizeof(*tc_to_txq) * TC_MAX_QUEUE);
3188 /* Map netdev TCs to offset 0.
3189 * We have our own UP to TXQ mapping for DCB mode of QoS
3191 for (tc = 0; tc < ntc; tc++) {
3192 tc_to_txq[tc] = (struct netdev_tc_txq) {
3199 static void mlx5e_mqprio_build_tc_to_txq(struct netdev_tc_txq *tc_to_txq,
3200 struct tc_mqprio_qopt *qopt)
3204 for (tc = 0; tc < TC_MAX_QUEUE; tc++) {
3205 tc_to_txq[tc] = (struct netdev_tc_txq) {
3206 .count = qopt->count[tc],
3207 .offset = qopt->offset[tc],
3212 static void mlx5e_params_mqprio_dcb_set(struct mlx5e_params *params, u8 num_tc)
3214 params->mqprio.mode = TC_MQPRIO_MODE_DCB;
3215 params->mqprio.num_tc = num_tc;
3216 mlx5e_mqprio_build_default_tc_to_txq(params->mqprio.tc_to_txq, num_tc,
3217 params->num_channels);
3220 static void mlx5e_mqprio_rl_update_params(struct mlx5e_params *params,
3221 struct mlx5e_mqprio_rl *rl)
3225 for (tc = 0; tc < TC_MAX_QUEUE; tc++) {
3229 mlx5e_mqprio_rl_get_node_hw_id(rl, tc, &hw_id);
3230 params->mqprio.channel.hw_id[tc] = hw_id;
3234 static void mlx5e_params_mqprio_channel_set(struct mlx5e_params *params,
3235 struct tc_mqprio_qopt_offload *mqprio,
3236 struct mlx5e_mqprio_rl *rl)
3240 params->mqprio.mode = TC_MQPRIO_MODE_CHANNEL;
3241 params->mqprio.num_tc = mqprio->qopt.num_tc;
3243 for (tc = 0; tc < TC_MAX_QUEUE; tc++)
3244 params->mqprio.channel.max_rate[tc] = mqprio->max_rate[tc];
3246 mlx5e_mqprio_rl_update_params(params, rl);
3247 mlx5e_mqprio_build_tc_to_txq(params->mqprio.tc_to_txq, &mqprio->qopt);
3250 static void mlx5e_params_mqprio_reset(struct mlx5e_params *params)
3252 mlx5e_params_mqprio_dcb_set(params, 1);
3255 static int mlx5e_setup_tc_mqprio_dcb(struct mlx5e_priv *priv,
3256 struct tc_mqprio_qopt *mqprio)
3258 struct mlx5e_params new_params;
3259 u8 tc = mqprio->num_tc;
3262 mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
3264 if (tc && tc != MLX5E_MAX_NUM_TC)
3267 new_params = priv->channels.params;
3268 mlx5e_params_mqprio_dcb_set(&new_params, tc ? tc : 1);
3270 err = mlx5e_safe_switch_params(priv, &new_params,
3271 mlx5e_num_channels_changed_ctx, NULL, true);
3273 if (!err && priv->mqprio_rl) {
3274 mlx5e_mqprio_rl_cleanup(priv->mqprio_rl);
3275 mlx5e_mqprio_rl_free(priv->mqprio_rl);
3276 priv->mqprio_rl = NULL;
3279 priv->max_opened_tc = max_t(u8, priv->max_opened_tc,
3280 mlx5e_get_dcb_num_tc(&priv->channels.params));
3284 static int mlx5e_mqprio_channel_validate(struct mlx5e_priv *priv,
3285 struct tc_mqprio_qopt_offload *mqprio)
3287 struct net_device *netdev = priv->netdev;
3288 struct mlx5e_ptp *ptp_channel;
3292 ptp_channel = priv->channels.ptp;
3293 if (ptp_channel && test_bit(MLX5E_PTP_STATE_TX, ptp_channel->state)) {
3295 "Cannot activate MQPRIO mode channel since it conflicts with TX port TS\n");
3299 if (mqprio->qopt.offset[0] != 0 || mqprio->qopt.num_tc < 1 ||
3300 mqprio->qopt.num_tc > MLX5E_MAX_NUM_MQPRIO_CH_TC)
3303 for (i = 0; i < mqprio->qopt.num_tc; i++) {
3304 if (!mqprio->qopt.count[i]) {
3305 netdev_err(netdev, "Zero size for queue-group (%d) is not supported\n", i);
3308 if (mqprio->min_rate[i]) {
3309 netdev_err(netdev, "Min tx rate is not supported\n");
3313 if (mqprio->max_rate[i]) {
3316 err = mlx5e_qos_bytes_rate_check(priv->mdev, mqprio->max_rate[i]);
3321 if (mqprio->qopt.offset[i] != agg_count) {
3322 netdev_err(netdev, "Discontinuous queues config is not supported\n");
3325 agg_count += mqprio->qopt.count[i];
3328 if (priv->channels.params.num_channels != agg_count) {
3329 netdev_err(netdev, "Num of queues (%d) does not match available (%d)\n",
3330 agg_count, priv->channels.params.num_channels);
3337 static bool mlx5e_mqprio_rate_limit(u8 num_tc, u64 max_rate[])
3341 for (tc = 0; tc < num_tc; tc++)
3347 static struct mlx5e_mqprio_rl *mlx5e_mqprio_rl_create(struct mlx5_core_dev *mdev,
3348 u8 num_tc, u64 max_rate[])
3350 struct mlx5e_mqprio_rl *rl;
3353 if (!mlx5e_mqprio_rate_limit(num_tc, max_rate))
3356 rl = mlx5e_mqprio_rl_alloc();
3358 return ERR_PTR(-ENOMEM);
3360 err = mlx5e_mqprio_rl_init(rl, mdev, num_tc, max_rate);
3362 mlx5e_mqprio_rl_free(rl);
3363 return ERR_PTR(err);
3369 static int mlx5e_setup_tc_mqprio_channel(struct mlx5e_priv *priv,
3370 struct tc_mqprio_qopt_offload *mqprio)
3372 mlx5e_fp_preactivate preactivate;
3373 struct mlx5e_params new_params;
3374 struct mlx5e_mqprio_rl *rl;
3378 err = mlx5e_mqprio_channel_validate(priv, mqprio);
3382 rl = mlx5e_mqprio_rl_create(priv->mdev, mqprio->qopt.num_tc, mqprio->max_rate);
3386 new_params = priv->channels.params;
3387 mlx5e_params_mqprio_channel_set(&new_params, mqprio, rl);
3389 nch_changed = mlx5e_get_dcb_num_tc(&priv->channels.params) > 1;
3390 preactivate = nch_changed ? mlx5e_num_channels_changed_ctx :
3391 mlx5e_update_netdev_queues_ctx;
3392 err = mlx5e_safe_switch_params(priv, &new_params, preactivate, NULL, true);
3395 mlx5e_mqprio_rl_cleanup(rl);
3396 mlx5e_mqprio_rl_free(rl);
3401 if (priv->mqprio_rl) {
3402 mlx5e_mqprio_rl_cleanup(priv->mqprio_rl);
3403 mlx5e_mqprio_rl_free(priv->mqprio_rl);
3405 priv->mqprio_rl = rl;
3410 static int mlx5e_setup_tc_mqprio(struct mlx5e_priv *priv,
3411 struct tc_mqprio_qopt_offload *mqprio)
3413 /* MQPRIO is another toplevel qdisc that can't be attached
3414 * simultaneously with the offloaded HTB.
3416 if (WARN_ON(mlx5e_selq_is_htb_enabled(&priv->selq)))
3419 switch (mqprio->mode) {
3420 case TC_MQPRIO_MODE_DCB:
3421 return mlx5e_setup_tc_mqprio_dcb(priv, &mqprio->qopt);
3422 case TC_MQPRIO_MODE_CHANNEL:
3423 return mlx5e_setup_tc_mqprio_channel(priv, mqprio);
3429 static LIST_HEAD(mlx5e_block_cb_list);
3431 static int mlx5e_setup_tc(struct net_device *dev, enum tc_setup_type type,
3434 struct mlx5e_priv *priv = netdev_priv(dev);
3435 bool tc_unbind = false;
3438 if (type == TC_SETUP_BLOCK &&
3439 ((struct flow_block_offload *)type_data)->command == FLOW_BLOCK_UNBIND)
3442 if (!netif_device_present(dev) && !tc_unbind)
3446 case TC_SETUP_BLOCK: {
3447 struct flow_block_offload *f = type_data;
3449 f->unlocked_driver_cb = true;
3450 return flow_block_cb_setup_simple(type_data,
3451 &mlx5e_block_cb_list,
3452 mlx5e_setup_tc_block_cb,
3455 case TC_SETUP_QDISC_MQPRIO:
3456 mutex_lock(&priv->state_lock);
3457 err = mlx5e_setup_tc_mqprio(priv, type_data);
3458 mutex_unlock(&priv->state_lock);
3460 case TC_SETUP_QDISC_HTB:
3461 mutex_lock(&priv->state_lock);
3462 err = mlx5e_htb_setup_tc(priv, type_data);
3463 mutex_unlock(&priv->state_lock);
3470 void mlx5e_fold_sw_stats64(struct mlx5e_priv *priv, struct rtnl_link_stats64 *s)
3474 for (i = 0; i < priv->stats_nch; i++) {
3475 struct mlx5e_channel_stats *channel_stats = priv->channel_stats[i];
3476 struct mlx5e_rq_stats *xskrq_stats = &channel_stats->xskrq;
3477 struct mlx5e_rq_stats *rq_stats = &channel_stats->rq;
3480 s->rx_packets += rq_stats->packets + xskrq_stats->packets;
3481 s->rx_bytes += rq_stats->bytes + xskrq_stats->bytes;
3482 s->multicast += rq_stats->mcast_packets + xskrq_stats->mcast_packets;
3484 for (j = 0; j < priv->max_opened_tc; j++) {
3485 struct mlx5e_sq_stats *sq_stats = &channel_stats->sq[j];
3487 s->tx_packets += sq_stats->packets;
3488 s->tx_bytes += sq_stats->bytes;
3489 s->tx_dropped += sq_stats->dropped;
3492 if (priv->tx_ptp_opened) {
3493 for (i = 0; i < priv->max_opened_tc; i++) {
3494 struct mlx5e_sq_stats *sq_stats = &priv->ptp_stats.sq[i];
3496 s->tx_packets += sq_stats->packets;
3497 s->tx_bytes += sq_stats->bytes;
3498 s->tx_dropped += sq_stats->dropped;
3501 if (priv->rx_ptp_opened) {
3502 struct mlx5e_rq_stats *rq_stats = &priv->ptp_stats.rq;
3504 s->rx_packets += rq_stats->packets;
3505 s->rx_bytes += rq_stats->bytes;
3506 s->multicast += rq_stats->mcast_packets;
3511 mlx5e_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats)
3513 struct mlx5e_priv *priv = netdev_priv(dev);
3514 struct mlx5e_pport_stats *pstats = &priv->stats.pport;
3516 if (!netif_device_present(dev))
3519 /* In switchdev mode, monitor counters doesn't monitor
3520 * rx/tx stats of 802_3. The update stats mechanism
3521 * should keep the 802_3 layout counters updated
3523 if (!mlx5e_monitor_counter_supported(priv) ||
3524 mlx5e_is_uplink_rep(priv)) {
3525 /* update HW stats in background for next time */
3526 mlx5e_queue_update_stats(priv);
3529 if (mlx5e_is_uplink_rep(priv)) {
3530 struct mlx5e_vport_stats *vstats = &priv->stats.vport;
3532 stats->rx_packets = PPORT_802_3_GET(pstats, a_frames_received_ok);
3533 stats->rx_bytes = PPORT_802_3_GET(pstats, a_octets_received_ok);
3534 stats->tx_packets = PPORT_802_3_GET(pstats, a_frames_transmitted_ok);
3535 stats->tx_bytes = PPORT_802_3_GET(pstats, a_octets_transmitted_ok);
3537 /* vport multicast also counts packets that are dropped due to steering
3538 * or rx out of buffer
3540 stats->multicast = VPORT_COUNTER_GET(vstats, received_eth_multicast.packets);
3542 mlx5e_fold_sw_stats64(priv, stats);
3545 stats->rx_dropped = priv->stats.qcnt.rx_out_of_buffer;
3547 stats->rx_length_errors =
3548 PPORT_802_3_GET(pstats, a_in_range_length_errors) +
3549 PPORT_802_3_GET(pstats, a_out_of_range_length_field) +
3550 PPORT_802_3_GET(pstats, a_frame_too_long_errors);
3551 stats->rx_crc_errors =
3552 PPORT_802_3_GET(pstats, a_frame_check_sequence_errors);
3553 stats->rx_frame_errors = PPORT_802_3_GET(pstats, a_alignment_errors);
3554 stats->tx_aborted_errors = PPORT_2863_GET(pstats, if_out_discards);
3555 stats->rx_errors = stats->rx_length_errors + stats->rx_crc_errors +
3556 stats->rx_frame_errors;
3557 stats->tx_errors = stats->tx_aborted_errors + stats->tx_carrier_errors;
3560 static void mlx5e_nic_set_rx_mode(struct mlx5e_priv *priv)
3562 if (mlx5e_is_uplink_rep(priv))
3563 return; /* no rx mode for uplink rep */
3565 queue_work(priv->wq, &priv->set_rx_mode_work);
3568 static void mlx5e_set_rx_mode(struct net_device *dev)
3570 struct mlx5e_priv *priv = netdev_priv(dev);
3572 mlx5e_nic_set_rx_mode(priv);
3575 static int mlx5e_set_mac(struct net_device *netdev, void *addr)
3577 struct mlx5e_priv *priv = netdev_priv(netdev);
3578 struct sockaddr *saddr = addr;
3580 if (!is_valid_ether_addr(saddr->sa_data))
3581 return -EADDRNOTAVAIL;
3583 netif_addr_lock_bh(netdev);
3584 eth_hw_addr_set(netdev, saddr->sa_data);
3585 netif_addr_unlock_bh(netdev);
3587 mlx5e_nic_set_rx_mode(priv);
3592 #define MLX5E_SET_FEATURE(features, feature, enable) \
3595 *features |= feature; \
3597 *features &= ~feature; \
3600 typedef int (*mlx5e_feature_handler)(struct net_device *netdev, bool enable);
3602 static int set_feature_lro(struct net_device *netdev, bool enable)
3604 struct mlx5e_priv *priv = netdev_priv(netdev);
3605 struct mlx5_core_dev *mdev = priv->mdev;
3606 struct mlx5e_params *cur_params;
3607 struct mlx5e_params new_params;
3611 mutex_lock(&priv->state_lock);
3613 cur_params = &priv->channels.params;
3614 new_params = *cur_params;
3617 new_params.packet_merge.type = MLX5E_PACKET_MERGE_LRO;
3618 else if (new_params.packet_merge.type == MLX5E_PACKET_MERGE_LRO)
3619 new_params.packet_merge.type = MLX5E_PACKET_MERGE_NONE;
3623 if (!(cur_params->packet_merge.type == MLX5E_PACKET_MERGE_SHAMPO &&
3624 new_params.packet_merge.type == MLX5E_PACKET_MERGE_LRO)) {
3625 if (cur_params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
3626 if (mlx5e_rx_mpwqe_is_linear_skb(mdev, cur_params, NULL) ==
3627 mlx5e_rx_mpwqe_is_linear_skb(mdev, &new_params, NULL))
3632 err = mlx5e_safe_switch_params(priv, &new_params,
3633 mlx5e_modify_tirs_packet_merge_ctx, NULL, reset);
3635 mutex_unlock(&priv->state_lock);
3639 static int set_feature_hw_gro(struct net_device *netdev, bool enable)
3641 struct mlx5e_priv *priv = netdev_priv(netdev);
3642 struct mlx5e_params new_params;
3646 mutex_lock(&priv->state_lock);
3647 new_params = priv->channels.params;
3650 new_params.packet_merge.type = MLX5E_PACKET_MERGE_SHAMPO;
3651 new_params.packet_merge.shampo.match_criteria_type =
3652 MLX5_RQC_SHAMPO_MATCH_CRITERIA_TYPE_EXTENDED;
3653 new_params.packet_merge.shampo.alignment_granularity =
3654 MLX5_RQC_SHAMPO_NO_MATCH_ALIGNMENT_GRANULARITY_STRIDE;
3655 } else if (new_params.packet_merge.type == MLX5E_PACKET_MERGE_SHAMPO) {
3656 new_params.packet_merge.type = MLX5E_PACKET_MERGE_NONE;
3661 err = mlx5e_safe_switch_params(priv, &new_params, NULL, NULL, reset);
3663 mutex_unlock(&priv->state_lock);
3667 static int set_feature_cvlan_filter(struct net_device *netdev, bool enable)
3669 struct mlx5e_priv *priv = netdev_priv(netdev);
3672 mlx5e_enable_cvlan_filter(priv);
3674 mlx5e_disable_cvlan_filter(priv);
3679 static int set_feature_hw_tc(struct net_device *netdev, bool enable)
3681 struct mlx5e_priv *priv = netdev_priv(netdev);
3684 #if IS_ENABLED(CONFIG_MLX5_CLS_ACT)
3685 if (!enable && mlx5e_tc_num_filters(priv, MLX5_TC_FLAG(NIC_OFFLOAD))) {
3687 "Active offloaded tc filters, can't turn hw_tc_offload off\n");
3692 mutex_lock(&priv->state_lock);
3693 if (!enable && mlx5e_selq_is_htb_enabled(&priv->selq)) {
3694 netdev_err(netdev, "Active HTB offload, can't turn hw_tc_offload off\n");
3697 mutex_unlock(&priv->state_lock);
3702 static int set_feature_rx_all(struct net_device *netdev, bool enable)
3704 struct mlx5e_priv *priv = netdev_priv(netdev);
3705 struct mlx5_core_dev *mdev = priv->mdev;
3707 return mlx5_set_port_fcs(mdev, !enable);
3710 static int mlx5e_set_rx_port_ts(struct mlx5_core_dev *mdev, bool enable)
3712 u32 in[MLX5_ST_SZ_DW(pcmr_reg)] = {};
3713 bool supported, curr_state;
3716 if (!MLX5_CAP_GEN(mdev, ports_check))
3719 err = mlx5_query_ports_check(mdev, in, sizeof(in));
3723 supported = MLX5_GET(pcmr_reg, in, rx_ts_over_crc_cap);
3724 curr_state = MLX5_GET(pcmr_reg, in, rx_ts_over_crc);
3726 if (!supported || enable == curr_state)
3729 MLX5_SET(pcmr_reg, in, local_port, 1);
3730 MLX5_SET(pcmr_reg, in, rx_ts_over_crc, enable);
3732 return mlx5_set_ports_check(mdev, in, sizeof(in));
3735 static int set_feature_rx_fcs(struct net_device *netdev, bool enable)
3737 struct mlx5e_priv *priv = netdev_priv(netdev);
3738 struct mlx5e_channels *chs = &priv->channels;
3739 struct mlx5_core_dev *mdev = priv->mdev;
3742 mutex_lock(&priv->state_lock);
3745 err = mlx5e_set_rx_port_ts(mdev, false);
3749 chs->params.scatter_fcs_en = true;
3750 err = mlx5e_modify_channels_scatter_fcs(chs, true);
3752 chs->params.scatter_fcs_en = false;
3753 mlx5e_set_rx_port_ts(mdev, true);
3756 chs->params.scatter_fcs_en = false;
3757 err = mlx5e_modify_channels_scatter_fcs(chs, false);
3759 chs->params.scatter_fcs_en = true;
3762 err = mlx5e_set_rx_port_ts(mdev, true);
3764 mlx5_core_warn(mdev, "Failed to set RX port timestamp %d\n", err);
3770 mutex_unlock(&priv->state_lock);
3774 static int set_feature_rx_vlan(struct net_device *netdev, bool enable)
3776 struct mlx5e_priv *priv = netdev_priv(netdev);
3779 mutex_lock(&priv->state_lock);
3781 mlx5e_fs_set_vlan_strip_disable(priv->fs, !enable);
3782 priv->channels.params.vlan_strip_disable = !enable;
3784 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
3787 err = mlx5e_modify_channels_vsd(&priv->channels, !enable);
3789 mlx5e_fs_set_vlan_strip_disable(priv->fs, enable);
3790 priv->channels.params.vlan_strip_disable = enable;
3793 mutex_unlock(&priv->state_lock);
3798 int mlx5e_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
3800 struct mlx5e_priv *priv = netdev_priv(dev);
3801 struct mlx5e_flow_steering *fs = priv->fs;
3803 if (mlx5e_is_uplink_rep(priv))
3804 return 0; /* no vlan table for uplink rep */
3806 return mlx5e_fs_vlan_rx_add_vid(fs, dev, proto, vid);
3809 int mlx5e_vlan_rx_kill_vid(struct net_device *dev, __be16 proto, u16 vid)
3811 struct mlx5e_priv *priv = netdev_priv(dev);
3812 struct mlx5e_flow_steering *fs = priv->fs;
3814 if (mlx5e_is_uplink_rep(priv))
3815 return 0; /* no vlan table for uplink rep */
3817 return mlx5e_fs_vlan_rx_kill_vid(fs, dev, proto, vid);
3820 #ifdef CONFIG_MLX5_EN_ARFS
3821 static int set_feature_arfs(struct net_device *netdev, bool enable)
3823 struct mlx5e_priv *priv = netdev_priv(netdev);
3827 err = mlx5e_arfs_enable(priv);
3829 err = mlx5e_arfs_disable(priv);
3835 static int mlx5e_handle_feature(struct net_device *netdev,
3836 netdev_features_t *features,
3837 netdev_features_t feature,
3838 mlx5e_feature_handler feature_handler)
3840 netdev_features_t changes = *features ^ netdev->features;
3841 bool enable = !!(*features & feature);
3844 if (!(changes & feature))
3847 err = feature_handler(netdev, enable);
3849 MLX5E_SET_FEATURE(features, feature, !enable);
3850 netdev_err(netdev, "%s feature %pNF failed, err %d\n",
3851 enable ? "Enable" : "Disable", &feature, err);
3858 int mlx5e_set_features(struct net_device *netdev, netdev_features_t features)
3860 netdev_features_t oper_features = features;
3863 #define MLX5E_HANDLE_FEATURE(feature, handler) \
3864 mlx5e_handle_feature(netdev, &oper_features, feature, handler)
3866 err |= MLX5E_HANDLE_FEATURE(NETIF_F_LRO, set_feature_lro);
3867 err |= MLX5E_HANDLE_FEATURE(NETIF_F_GRO_HW, set_feature_hw_gro);
3868 err |= MLX5E_HANDLE_FEATURE(NETIF_F_HW_VLAN_CTAG_FILTER,
3869 set_feature_cvlan_filter);
3870 err |= MLX5E_HANDLE_FEATURE(NETIF_F_HW_TC, set_feature_hw_tc);
3871 err |= MLX5E_HANDLE_FEATURE(NETIF_F_RXALL, set_feature_rx_all);
3872 err |= MLX5E_HANDLE_FEATURE(NETIF_F_RXFCS, set_feature_rx_fcs);
3873 err |= MLX5E_HANDLE_FEATURE(NETIF_F_HW_VLAN_CTAG_RX, set_feature_rx_vlan);
3874 #ifdef CONFIG_MLX5_EN_ARFS
3875 err |= MLX5E_HANDLE_FEATURE(NETIF_F_NTUPLE, set_feature_arfs);
3877 err |= MLX5E_HANDLE_FEATURE(NETIF_F_HW_TLS_RX, mlx5e_ktls_set_feature_rx);
3880 netdev->features = oper_features;
3887 static netdev_features_t mlx5e_fix_uplink_rep_features(struct net_device *netdev,
3888 netdev_features_t features)
3890 features &= ~NETIF_F_HW_TLS_RX;
3891 if (netdev->features & NETIF_F_HW_TLS_RX)
3892 netdev_warn(netdev, "Disabling hw_tls_rx, not supported in switchdev mode\n");
3894 features &= ~NETIF_F_HW_TLS_TX;
3895 if (netdev->features & NETIF_F_HW_TLS_TX)
3896 netdev_warn(netdev, "Disabling hw_tls_tx, not supported in switchdev mode\n");
3898 features &= ~NETIF_F_NTUPLE;
3899 if (netdev->features & NETIF_F_NTUPLE)
3900 netdev_warn(netdev, "Disabling ntuple, not supported in switchdev mode\n");
3902 features &= ~NETIF_F_GRO_HW;
3903 if (netdev->features & NETIF_F_GRO_HW)
3904 netdev_warn(netdev, "Disabling HW_GRO, not supported in switchdev mode\n");
3909 static netdev_features_t mlx5e_fix_features(struct net_device *netdev,
3910 netdev_features_t features)
3912 struct mlx5e_priv *priv = netdev_priv(netdev);
3913 struct mlx5e_vlan_table *vlan;
3914 struct mlx5e_params *params;
3916 vlan = mlx5e_fs_get_vlan(priv->fs);
3917 mutex_lock(&priv->state_lock);
3918 params = &priv->channels.params;
3920 !bitmap_empty(mlx5e_vlan_get_active_svlans(vlan), VLAN_N_VID)) {
3921 /* HW strips the outer C-tag header, this is a problem
3922 * for S-tag traffic.
3924 features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3925 if (!params->vlan_strip_disable)
3926 netdev_warn(netdev, "Dropping C-tag vlan stripping offload due to S-tag vlan\n");
3929 if (!MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_STRIDING_RQ)) {
3930 if (features & NETIF_F_LRO) {
3931 netdev_warn(netdev, "Disabling LRO, not supported in legacy RQ\n");
3932 features &= ~NETIF_F_LRO;
3934 if (features & NETIF_F_GRO_HW) {
3935 netdev_warn(netdev, "Disabling HW-GRO, not supported in legacy RQ\n");
3936 features &= ~NETIF_F_GRO_HW;
3940 if (params->xdp_prog) {
3941 if (features & NETIF_F_LRO) {
3942 netdev_warn(netdev, "LRO is incompatible with XDP\n");
3943 features &= ~NETIF_F_LRO;
3945 if (features & NETIF_F_GRO_HW) {
3946 netdev_warn(netdev, "HW GRO is incompatible with XDP\n");
3947 features &= ~NETIF_F_GRO_HW;
3951 if (priv->xsk.refcnt) {
3952 if (features & NETIF_F_LRO) {
3953 netdev_warn(netdev, "LRO is incompatible with AF_XDP (%u XSKs are active)\n",
3955 features &= ~NETIF_F_LRO;
3957 if (features & NETIF_F_GRO_HW) {
3958 netdev_warn(netdev, "HW GRO is incompatible with AF_XDP (%u XSKs are active)\n",
3960 features &= ~NETIF_F_GRO_HW;
3964 if (MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS)) {
3965 features &= ~NETIF_F_RXHASH;
3966 if (netdev->features & NETIF_F_RXHASH)
3967 netdev_warn(netdev, "Disabling rxhash, not supported when CQE compress is active\n");
3969 if (features & NETIF_F_GRO_HW) {
3970 netdev_warn(netdev, "Disabling HW-GRO, not supported when CQE compress is active\n");
3971 features &= ~NETIF_F_GRO_HW;
3975 if (mlx5e_is_uplink_rep(priv))
3976 features = mlx5e_fix_uplink_rep_features(netdev, features);
3978 mutex_unlock(&priv->state_lock);
3983 static bool mlx5e_xsk_validate_mtu(struct net_device *netdev,
3984 struct mlx5e_channels *chs,
3985 struct mlx5e_params *new_params,
3986 struct mlx5_core_dev *mdev)
3990 for (ix = 0; ix < chs->params.num_channels; ix++) {
3991 struct xsk_buff_pool *xsk_pool =
3992 mlx5e_xsk_get_pool(&chs->params, chs->params.xsk, ix);
3993 struct mlx5e_xsk_param xsk;
3998 mlx5e_build_xsk_param(xsk_pool, &xsk);
4000 if (!mlx5e_validate_xsk_param(new_params, &xsk, mdev)) {
4001 u32 hr = mlx5e_get_linear_rq_headroom(new_params, &xsk);
4002 int max_mtu_frame, max_mtu_page, max_mtu;
4004 /* Two criteria must be met:
4005 * 1. HW MTU + all headrooms <= XSK frame size.
4006 * 2. Size of SKBs allocated on XDP_PASS <= PAGE_SIZE.
4008 max_mtu_frame = MLX5E_HW2SW_MTU(new_params, xsk.chunk_size - hr);
4009 max_mtu_page = mlx5e_xdp_max_mtu(new_params, &xsk);
4010 max_mtu = min(max_mtu_frame, max_mtu_page);
4012 netdev_err(netdev, "MTU %d is too big for an XSK running on channel %u. Try MTU <= %d\n",
4013 new_params->sw_mtu, ix, max_mtu);
4021 static bool mlx5e_params_validate_xdp(struct net_device *netdev, struct mlx5e_params *params)
4025 /* No XSK params: AF_XDP can't be enabled yet at the point of setting
4028 is_linear = mlx5e_rx_is_linear_skb(params, NULL);
4030 if (!is_linear && params->rq_wq_type != MLX5_WQ_TYPE_CYCLIC) {
4031 netdev_warn(netdev, "XDP is not allowed with striding RQ and MTU(%d) > %d\n",
4033 mlx5e_xdp_max_mtu(params, NULL));
4036 if (!is_linear && !params->xdp_prog->aux->xdp_has_frags) {
4037 netdev_warn(netdev, "MTU(%d) > %d, too big for an XDP program not aware of multi buffer\n",
4039 mlx5e_xdp_max_mtu(params, NULL));
4046 int mlx5e_change_mtu(struct net_device *netdev, int new_mtu,
4047 mlx5e_fp_preactivate preactivate)
4049 struct mlx5e_priv *priv = netdev_priv(netdev);
4050 struct mlx5e_params new_params;
4051 struct mlx5e_params *params;
4055 mutex_lock(&priv->state_lock);
4057 params = &priv->channels.params;
4059 new_params = *params;
4060 new_params.sw_mtu = new_mtu;
4061 err = mlx5e_validate_params(priv->mdev, &new_params);
4065 if (new_params.xdp_prog && !mlx5e_params_validate_xdp(netdev, &new_params)) {
4070 if (priv->xsk.refcnt &&
4071 !mlx5e_xsk_validate_mtu(netdev, &priv->channels,
4072 &new_params, priv->mdev)) {
4077 if (params->packet_merge.type == MLX5E_PACKET_MERGE_LRO)
4080 if (params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
4081 bool is_linear_old = mlx5e_rx_mpwqe_is_linear_skb(priv->mdev, params, NULL);
4082 bool is_linear_new = mlx5e_rx_mpwqe_is_linear_skb(priv->mdev,
4084 u8 ppw_old = mlx5e_mpwqe_log_pkts_per_wqe(params, NULL);
4085 u8 ppw_new = mlx5e_mpwqe_log_pkts_per_wqe(&new_params, NULL);
4087 /* Always reset in linear mode - hw_mtu is used in data path.
4088 * Check that the mode was non-linear and didn't change.
4089 * If XSK is active, XSK RQs are linear.
4091 if (!is_linear_old && !is_linear_new && !priv->xsk.refcnt &&
4096 err = mlx5e_safe_switch_params(priv, &new_params, preactivate, NULL, reset);
4099 netdev->mtu = params->sw_mtu;
4100 mutex_unlock(&priv->state_lock);
4104 static int mlx5e_change_nic_mtu(struct net_device *netdev, int new_mtu)
4106 return mlx5e_change_mtu(netdev, new_mtu, mlx5e_set_dev_port_mtu_ctx);
4109 int mlx5e_ptp_rx_manage_fs_ctx(struct mlx5e_priv *priv, void *ctx)
4111 bool set = *(bool *)ctx;
4113 return mlx5e_ptp_rx_manage_fs(priv, set);
4116 static int mlx5e_hwstamp_config_no_ptp_rx(struct mlx5e_priv *priv, bool rx_filter)
4118 bool rx_cqe_compress_def = priv->channels.params.rx_cqe_compress_def;
4122 /* Reset CQE compression to Admin default */
4123 return mlx5e_modify_rx_cqe_compression_locked(priv, rx_cqe_compress_def, false);
4125 if (!MLX5E_GET_PFLAG(&priv->channels.params, MLX5E_PFLAG_RX_CQE_COMPRESS))
4128 /* Disable CQE compression */
4129 netdev_warn(priv->netdev, "Disabling RX cqe compression\n");
4130 err = mlx5e_modify_rx_cqe_compression_locked(priv, false, true);
4132 netdev_err(priv->netdev, "Failed disabling cqe compression err=%d\n", err);
4137 static int mlx5e_hwstamp_config_ptp_rx(struct mlx5e_priv *priv, bool ptp_rx)
4139 struct mlx5e_params new_params;
4141 if (ptp_rx == priv->channels.params.ptp_rx)
4144 new_params = priv->channels.params;
4145 new_params.ptp_rx = ptp_rx;
4146 return mlx5e_safe_switch_params(priv, &new_params, mlx5e_ptp_rx_manage_fs_ctx,
4147 &new_params.ptp_rx, true);
4150 int mlx5e_hwstamp_set(struct mlx5e_priv *priv, struct ifreq *ifr)
4152 struct hwtstamp_config config;
4153 bool rx_cqe_compress_def;
4157 if (!MLX5_CAP_GEN(priv->mdev, device_frequency_khz) ||
4158 (mlx5_clock_get_ptp_index(priv->mdev) == -1))
4161 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
4164 /* TX HW timestamp */
4165 switch (config.tx_type) {
4166 case HWTSTAMP_TX_OFF:
4167 case HWTSTAMP_TX_ON:
4173 mutex_lock(&priv->state_lock);
4174 rx_cqe_compress_def = priv->channels.params.rx_cqe_compress_def;
4176 /* RX HW timestamp */
4177 switch (config.rx_filter) {
4178 case HWTSTAMP_FILTER_NONE:
4181 case HWTSTAMP_FILTER_ALL:
4182 case HWTSTAMP_FILTER_SOME:
4183 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4184 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4185 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4186 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4187 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4188 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4189 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4190 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4191 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4192 case HWTSTAMP_FILTER_PTP_V2_EVENT:
4193 case HWTSTAMP_FILTER_PTP_V2_SYNC:
4194 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4195 case HWTSTAMP_FILTER_NTP_ALL:
4196 config.rx_filter = HWTSTAMP_FILTER_ALL;
4197 /* ptp_rx is set if both HW TS is set and CQE
4198 * compression is set
4200 ptp_rx = rx_cqe_compress_def;
4207 if (!mlx5e_profile_feature_cap(priv->profile, PTP_RX))
4208 err = mlx5e_hwstamp_config_no_ptp_rx(priv,
4209 config.rx_filter != HWTSTAMP_FILTER_NONE);
4211 err = mlx5e_hwstamp_config_ptp_rx(priv, ptp_rx);
4215 memcpy(&priv->tstamp, &config, sizeof(config));
4216 mutex_unlock(&priv->state_lock);
4218 /* might need to fix some features */
4219 netdev_update_features(priv->netdev);
4221 return copy_to_user(ifr->ifr_data, &config,
4222 sizeof(config)) ? -EFAULT : 0;
4224 mutex_unlock(&priv->state_lock);
4228 int mlx5e_hwstamp_get(struct mlx5e_priv *priv, struct ifreq *ifr)
4230 struct hwtstamp_config *cfg = &priv->tstamp;
4232 if (!MLX5_CAP_GEN(priv->mdev, device_frequency_khz))
4235 return copy_to_user(ifr->ifr_data, cfg, sizeof(*cfg)) ? -EFAULT : 0;
4238 static int mlx5e_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
4240 struct mlx5e_priv *priv = netdev_priv(dev);
4244 return mlx5e_hwstamp_set(priv, ifr);
4246 return mlx5e_hwstamp_get(priv, ifr);
4252 #ifdef CONFIG_MLX5_ESWITCH
4253 int mlx5e_set_vf_mac(struct net_device *dev, int vf, u8 *mac)
4255 struct mlx5e_priv *priv = netdev_priv(dev);
4256 struct mlx5_core_dev *mdev = priv->mdev;
4258 return mlx5_eswitch_set_vport_mac(mdev->priv.eswitch, vf + 1, mac);
4261 static int mlx5e_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
4264 struct mlx5e_priv *priv = netdev_priv(dev);
4265 struct mlx5_core_dev *mdev = priv->mdev;
4267 if (vlan_proto != htons(ETH_P_8021Q))
4268 return -EPROTONOSUPPORT;
4270 return mlx5_eswitch_set_vport_vlan(mdev->priv.eswitch, vf + 1,
4274 static int mlx5e_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
4276 struct mlx5e_priv *priv = netdev_priv(dev);
4277 struct mlx5_core_dev *mdev = priv->mdev;
4279 return mlx5_eswitch_set_vport_spoofchk(mdev->priv.eswitch, vf + 1, setting);
4282 static int mlx5e_set_vf_trust(struct net_device *dev, int vf, bool setting)
4284 struct mlx5e_priv *priv = netdev_priv(dev);
4285 struct mlx5_core_dev *mdev = priv->mdev;
4287 return mlx5_eswitch_set_vport_trust(mdev->priv.eswitch, vf + 1, setting);
4290 int mlx5e_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
4293 struct mlx5e_priv *priv = netdev_priv(dev);
4294 struct mlx5_core_dev *mdev = priv->mdev;
4296 return mlx5_eswitch_set_vport_rate(mdev->priv.eswitch, vf + 1,
4297 max_tx_rate, min_tx_rate);
4300 static int mlx5_vport_link2ifla(u8 esw_link)
4303 case MLX5_VPORT_ADMIN_STATE_DOWN:
4304 return IFLA_VF_LINK_STATE_DISABLE;
4305 case MLX5_VPORT_ADMIN_STATE_UP:
4306 return IFLA_VF_LINK_STATE_ENABLE;
4308 return IFLA_VF_LINK_STATE_AUTO;
4311 static int mlx5_ifla_link2vport(u8 ifla_link)
4313 switch (ifla_link) {
4314 case IFLA_VF_LINK_STATE_DISABLE:
4315 return MLX5_VPORT_ADMIN_STATE_DOWN;
4316 case IFLA_VF_LINK_STATE_ENABLE:
4317 return MLX5_VPORT_ADMIN_STATE_UP;
4319 return MLX5_VPORT_ADMIN_STATE_AUTO;
4322 static int mlx5e_set_vf_link_state(struct net_device *dev, int vf,
4325 struct mlx5e_priv *priv = netdev_priv(dev);
4326 struct mlx5_core_dev *mdev = priv->mdev;
4328 if (mlx5e_is_uplink_rep(priv))
4331 return mlx5_eswitch_set_vport_state(mdev->priv.eswitch, vf + 1,
4332 mlx5_ifla_link2vport(link_state));
4335 int mlx5e_get_vf_config(struct net_device *dev,
4336 int vf, struct ifla_vf_info *ivi)
4338 struct mlx5e_priv *priv = netdev_priv(dev);
4339 struct mlx5_core_dev *mdev = priv->mdev;
4342 if (!netif_device_present(dev))
4345 err = mlx5_eswitch_get_vport_config(mdev->priv.eswitch, vf + 1, ivi);
4348 ivi->linkstate = mlx5_vport_link2ifla(ivi->linkstate);
4352 int mlx5e_get_vf_stats(struct net_device *dev,
4353 int vf, struct ifla_vf_stats *vf_stats)
4355 struct mlx5e_priv *priv = netdev_priv(dev);
4356 struct mlx5_core_dev *mdev = priv->mdev;
4358 return mlx5_eswitch_get_vport_stats(mdev->priv.eswitch, vf + 1,
4363 mlx5e_has_offload_stats(const struct net_device *dev, int attr_id)
4365 struct mlx5e_priv *priv = netdev_priv(dev);
4367 if (!netif_device_present(dev))
4370 if (!mlx5e_is_uplink_rep(priv))
4373 return mlx5e_rep_has_offload_stats(dev, attr_id);
4377 mlx5e_get_offload_stats(int attr_id, const struct net_device *dev,
4380 struct mlx5e_priv *priv = netdev_priv(dev);
4382 if (!mlx5e_is_uplink_rep(priv))
4385 return mlx5e_rep_get_offload_stats(attr_id, dev, sp);
4389 static bool mlx5e_tunnel_proto_supported_tx(struct mlx5_core_dev *mdev, u8 proto_type)
4391 switch (proto_type) {
4393 return MLX5_CAP_ETH(mdev, tunnel_stateless_gre);
4396 return (MLX5_CAP_ETH(mdev, tunnel_stateless_ip_over_ip) ||
4397 MLX5_CAP_ETH(mdev, tunnel_stateless_ip_over_ip_tx));
4403 static bool mlx5e_gre_tunnel_inner_proto_offload_supported(struct mlx5_core_dev *mdev,
4404 struct sk_buff *skb)
4406 switch (skb->inner_protocol) {
4407 case htons(ETH_P_IP):
4408 case htons(ETH_P_IPV6):
4409 case htons(ETH_P_TEB):
4411 case htons(ETH_P_MPLS_UC):
4412 case htons(ETH_P_MPLS_MC):
4413 return MLX5_CAP_ETH(mdev, tunnel_stateless_mpls_over_gre);
4418 static netdev_features_t mlx5e_tunnel_features_check(struct mlx5e_priv *priv,
4419 struct sk_buff *skb,
4420 netdev_features_t features)
4422 unsigned int offset = 0;
4423 struct udphdr *udph;
4427 switch (vlan_get_protocol(skb)) {
4428 case htons(ETH_P_IP):
4429 proto = ip_hdr(skb)->protocol;
4431 case htons(ETH_P_IPV6):
4432 proto = ipv6_find_hdr(skb, &offset, -1, NULL, NULL);
4440 if (mlx5e_gre_tunnel_inner_proto_offload_supported(priv->mdev, skb))
4445 if (mlx5e_tunnel_proto_supported_tx(priv->mdev, IPPROTO_IPIP))
4449 udph = udp_hdr(skb);
4450 port = be16_to_cpu(udph->dest);
4452 /* Verify if UDP port is being offloaded by HW */
4453 if (mlx5_vxlan_lookup_port(priv->mdev->vxlan, port))
4456 #if IS_ENABLED(CONFIG_GENEVE)
4457 /* Support Geneve offload for default UDP port */
4458 if (port == GENEVE_UDP_PORT && mlx5_geneve_tx_allowed(priv->mdev))
4462 #ifdef CONFIG_MLX5_EN_IPSEC
4464 return mlx5e_ipsec_feature_check(skb, features);
4469 /* Disable CSUM and GSO if the udp dport is not offloaded by HW */
4470 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
4473 netdev_features_t mlx5e_features_check(struct sk_buff *skb,
4474 struct net_device *netdev,
4475 netdev_features_t features)
4477 struct mlx5e_priv *priv = netdev_priv(netdev);
4479 features = vlan_features_check(skb, features);
4480 features = vxlan_features_check(skb, features);
4482 /* Validate if the tunneled packet is being offloaded by HW */
4483 if (skb->encapsulation &&
4484 (features & NETIF_F_CSUM_MASK || features & NETIF_F_GSO_MASK))
4485 return mlx5e_tunnel_features_check(priv, skb, features);
4490 static void mlx5e_tx_timeout_work(struct work_struct *work)
4492 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
4494 struct net_device *netdev = priv->netdev;
4498 mutex_lock(&priv->state_lock);
4500 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
4503 for (i = 0; i < netdev->real_num_tx_queues; i++) {
4504 struct netdev_queue *dev_queue =
4505 netdev_get_tx_queue(netdev, i);
4506 struct mlx5e_txqsq *sq = priv->txq2sq[i];
4508 if (!netif_xmit_stopped(dev_queue))
4511 if (mlx5e_reporter_tx_timeout(sq))
4512 /* break if tried to reopened channels */
4517 mutex_unlock(&priv->state_lock);
4521 static void mlx5e_tx_timeout(struct net_device *dev, unsigned int txqueue)
4523 struct mlx5e_priv *priv = netdev_priv(dev);
4525 netdev_err(dev, "TX timeout detected\n");
4526 queue_work(priv->wq, &priv->tx_timeout_work);
4529 static int mlx5e_xdp_allowed(struct mlx5e_priv *priv, struct bpf_prog *prog)
4531 struct net_device *netdev = priv->netdev;
4532 struct mlx5e_params new_params;
4534 if (priv->channels.params.packet_merge.type != MLX5E_PACKET_MERGE_NONE) {
4535 netdev_warn(netdev, "can't set XDP while HW-GRO/LRO is on, disable them first\n");
4539 new_params = priv->channels.params;
4540 new_params.xdp_prog = prog;
4542 if (!mlx5e_params_validate_xdp(netdev, &new_params))
4548 static void mlx5e_rq_replace_xdp_prog(struct mlx5e_rq *rq, struct bpf_prog *prog)
4550 struct bpf_prog *old_prog;
4552 old_prog = rcu_replace_pointer(rq->xdp_prog, prog,
4553 lockdep_is_held(&rq->priv->state_lock));
4555 bpf_prog_put(old_prog);
4558 static int mlx5e_xdp_set(struct net_device *netdev, struct bpf_prog *prog)
4560 struct mlx5e_priv *priv = netdev_priv(netdev);
4561 struct mlx5e_params new_params;
4562 struct bpf_prog *old_prog;
4567 mutex_lock(&priv->state_lock);
4570 err = mlx5e_xdp_allowed(priv, prog);
4575 /* no need for full reset when exchanging programs */
4576 reset = (!priv->channels.params.xdp_prog || !prog);
4578 new_params = priv->channels.params;
4579 new_params.xdp_prog = prog;
4581 mlx5e_set_rq_type(priv->mdev, &new_params);
4582 old_prog = priv->channels.params.xdp_prog;
4584 err = mlx5e_safe_switch_params(priv, &new_params, NULL, NULL, reset);
4589 bpf_prog_put(old_prog);
4591 if (!test_bit(MLX5E_STATE_OPENED, &priv->state) || reset)
4594 /* exchanging programs w/o reset, we update ref counts on behalf
4595 * of the channels RQs here.
4597 bpf_prog_add(prog, priv->channels.num);
4598 for (i = 0; i < priv->channels.num; i++) {
4599 struct mlx5e_channel *c = priv->channels.c[i];
4601 mlx5e_rq_replace_xdp_prog(&c->rq, prog);
4602 if (test_bit(MLX5E_CHANNEL_STATE_XSK, c->state)) {
4604 mlx5e_rq_replace_xdp_prog(&c->xskrq, prog);
4609 mutex_unlock(&priv->state_lock);
4611 /* Need to fix some features. */
4613 netdev_update_features(netdev);
4618 static int mlx5e_xdp(struct net_device *dev, struct netdev_bpf *xdp)
4620 switch (xdp->command) {
4621 case XDP_SETUP_PROG:
4622 return mlx5e_xdp_set(dev, xdp->prog);
4623 case XDP_SETUP_XSK_POOL:
4624 return mlx5e_xsk_setup_pool(dev, xdp->xsk.pool,
4631 #ifdef CONFIG_MLX5_ESWITCH
4632 static int mlx5e_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4633 struct net_device *dev, u32 filter_mask,
4636 struct mlx5e_priv *priv = netdev_priv(dev);
4637 struct mlx5_core_dev *mdev = priv->mdev;
4641 err = mlx5_eswitch_get_vepa(mdev->priv.eswitch, &setting);
4644 mode = setting ? BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB;
4645 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4647 0, 0, nlflags, filter_mask, NULL);
4650 static int mlx5e_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4651 u16 flags, struct netlink_ext_ack *extack)
4653 struct mlx5e_priv *priv = netdev_priv(dev);
4654 struct mlx5_core_dev *mdev = priv->mdev;
4655 struct nlattr *attr, *br_spec;
4656 u16 mode = BRIDGE_MODE_UNDEF;
4660 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4664 nla_for_each_nested(attr, br_spec, rem) {
4665 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4668 if (nla_len(attr) < sizeof(mode))
4671 mode = nla_get_u16(attr);
4672 if (mode > BRIDGE_MODE_VEPA)
4678 if (mode == BRIDGE_MODE_UNDEF)
4681 setting = (mode == BRIDGE_MODE_VEPA) ? 1 : 0;
4682 return mlx5_eswitch_set_vepa(mdev->priv.eswitch, setting);
4686 const struct net_device_ops mlx5e_netdev_ops = {
4687 .ndo_open = mlx5e_open,
4688 .ndo_stop = mlx5e_close,
4689 .ndo_start_xmit = mlx5e_xmit,
4690 .ndo_setup_tc = mlx5e_setup_tc,
4691 .ndo_select_queue = mlx5e_select_queue,
4692 .ndo_get_stats64 = mlx5e_get_stats,
4693 .ndo_set_rx_mode = mlx5e_set_rx_mode,
4694 .ndo_set_mac_address = mlx5e_set_mac,
4695 .ndo_vlan_rx_add_vid = mlx5e_vlan_rx_add_vid,
4696 .ndo_vlan_rx_kill_vid = mlx5e_vlan_rx_kill_vid,
4697 .ndo_set_features = mlx5e_set_features,
4698 .ndo_fix_features = mlx5e_fix_features,
4699 .ndo_change_mtu = mlx5e_change_nic_mtu,
4700 .ndo_eth_ioctl = mlx5e_ioctl,
4701 .ndo_set_tx_maxrate = mlx5e_set_tx_maxrate,
4702 .ndo_features_check = mlx5e_features_check,
4703 .ndo_tx_timeout = mlx5e_tx_timeout,
4704 .ndo_bpf = mlx5e_xdp,
4705 .ndo_xdp_xmit = mlx5e_xdp_xmit,
4706 .ndo_xsk_wakeup = mlx5e_xsk_wakeup,
4707 #ifdef CONFIG_MLX5_EN_ARFS
4708 .ndo_rx_flow_steer = mlx5e_rx_flow_steer,
4710 #ifdef CONFIG_MLX5_ESWITCH
4711 .ndo_bridge_setlink = mlx5e_bridge_setlink,
4712 .ndo_bridge_getlink = mlx5e_bridge_getlink,
4714 /* SRIOV E-Switch NDOs */
4715 .ndo_set_vf_mac = mlx5e_set_vf_mac,
4716 .ndo_set_vf_vlan = mlx5e_set_vf_vlan,
4717 .ndo_set_vf_spoofchk = mlx5e_set_vf_spoofchk,
4718 .ndo_set_vf_trust = mlx5e_set_vf_trust,
4719 .ndo_set_vf_rate = mlx5e_set_vf_rate,
4720 .ndo_get_vf_config = mlx5e_get_vf_config,
4721 .ndo_set_vf_link_state = mlx5e_set_vf_link_state,
4722 .ndo_get_vf_stats = mlx5e_get_vf_stats,
4723 .ndo_has_offload_stats = mlx5e_has_offload_stats,
4724 .ndo_get_offload_stats = mlx5e_get_offload_stats,
4726 .ndo_get_devlink_port = mlx5e_get_devlink_port,
4729 static u32 mlx5e_choose_lro_timeout(struct mlx5_core_dev *mdev, u32 wanted_timeout)
4733 /* The supported periods are organized in ascending order */
4734 for (i = 0; i < MLX5E_LRO_TIMEOUT_ARR_SIZE - 1; i++)
4735 if (MLX5_CAP_ETH(mdev, lro_timer_supported_periods[i]) >= wanted_timeout)
4738 return MLX5_CAP_ETH(mdev, lro_timer_supported_periods[i]);
4741 void mlx5e_build_nic_params(struct mlx5e_priv *priv, struct mlx5e_xsk *xsk, u16 mtu)
4743 struct mlx5e_params *params = &priv->channels.params;
4744 struct mlx5_core_dev *mdev = priv->mdev;
4745 u8 rx_cq_period_mode;
4747 params->sw_mtu = mtu;
4748 params->hard_mtu = MLX5E_ETH_HARD_MTU;
4749 params->num_channels = min_t(unsigned int, MLX5E_MAX_NUM_CHANNELS / 2,
4751 mlx5e_params_mqprio_reset(params);
4754 params->log_sq_size = is_kdump_kernel() ?
4755 MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE :
4756 MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
4757 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_SKB_TX_MPWQE, mlx5e_tx_mpwqe_supported(mdev));
4760 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_XDP_TX_MPWQE, mlx5e_tx_mpwqe_supported(mdev));
4762 /* set CQE compression */
4763 params->rx_cqe_compress_def = false;
4764 if (MLX5_CAP_GEN(mdev, cqe_compression) &&
4765 MLX5_CAP_GEN(mdev, vport_group_manager))
4766 params->rx_cqe_compress_def = slow_pci_heuristic(mdev);
4768 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS, params->rx_cqe_compress_def);
4769 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_NO_CSUM_COMPLETE, false);
4772 mlx5e_build_rq_params(mdev, params);
4775 if (MLX5_CAP_ETH(mdev, lro_cap) &&
4776 params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
4777 /* No XSK params: checking the availability of striding RQ in general. */
4778 if (!mlx5e_rx_mpwqe_is_linear_skb(mdev, params, NULL))
4779 params->packet_merge.type = slow_pci_heuristic(mdev) ?
4780 MLX5E_PACKET_MERGE_NONE : MLX5E_PACKET_MERGE_LRO;
4782 params->packet_merge.timeout = mlx5e_choose_lro_timeout(mdev, MLX5E_DEFAULT_LRO_TIMEOUT);
4784 /* CQ moderation params */
4785 rx_cq_period_mode = MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
4786 MLX5_CQ_PERIOD_MODE_START_FROM_CQE :
4787 MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
4788 params->rx_dim_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
4789 params->tx_dim_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
4790 mlx5e_set_rx_cq_mode_params(params, rx_cq_period_mode);
4791 mlx5e_set_tx_cq_mode_params(params, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
4794 mlx5_query_min_inline(mdev, ¶ms->tx_min_inline_mode);
4796 params->tunneled_offload_en = mlx5_tunnel_inner_ft_supported(mdev);
4801 /* Do not update netdev->features directly in here
4802 * on mlx5e_attach_netdev() we will call mlx5e_update_features()
4803 * To update netdev->features please modify mlx5e_fix_features()
4807 static void mlx5e_set_netdev_dev_addr(struct net_device *netdev)
4809 struct mlx5e_priv *priv = netdev_priv(netdev);
4812 mlx5_query_mac_address(priv->mdev, addr);
4813 if (is_zero_ether_addr(addr) &&
4814 !MLX5_CAP_GEN(priv->mdev, vport_group_manager)) {
4815 eth_hw_addr_random(netdev);
4816 mlx5_core_info(priv->mdev, "Assigned random MAC address %pM\n", netdev->dev_addr);
4820 eth_hw_addr_set(netdev, addr);
4823 static int mlx5e_vxlan_set_port(struct net_device *netdev, unsigned int table,
4824 unsigned int entry, struct udp_tunnel_info *ti)
4826 struct mlx5e_priv *priv = netdev_priv(netdev);
4828 return mlx5_vxlan_add_port(priv->mdev->vxlan, ntohs(ti->port));
4831 static int mlx5e_vxlan_unset_port(struct net_device *netdev, unsigned int table,
4832 unsigned int entry, struct udp_tunnel_info *ti)
4834 struct mlx5e_priv *priv = netdev_priv(netdev);
4836 return mlx5_vxlan_del_port(priv->mdev->vxlan, ntohs(ti->port));
4839 void mlx5e_vxlan_set_netdev_info(struct mlx5e_priv *priv)
4841 if (!mlx5_vxlan_allowed(priv->mdev->vxlan))
4844 priv->nic_info.set_port = mlx5e_vxlan_set_port;
4845 priv->nic_info.unset_port = mlx5e_vxlan_unset_port;
4846 priv->nic_info.flags = UDP_TUNNEL_NIC_INFO_MAY_SLEEP |
4847 UDP_TUNNEL_NIC_INFO_STATIC_IANA_VXLAN;
4848 priv->nic_info.tables[0].tunnel_types = UDP_TUNNEL_TYPE_VXLAN;
4849 /* Don't count the space hard-coded to the IANA port */
4850 priv->nic_info.tables[0].n_entries =
4851 mlx5_vxlan_max_udp_ports(priv->mdev) - 1;
4853 priv->netdev->udp_tunnel_nic_info = &priv->nic_info;
4856 static bool mlx5e_tunnel_any_tx_proto_supported(struct mlx5_core_dev *mdev)
4860 for (tt = 0; tt < MLX5_NUM_TUNNEL_TT; tt++) {
4861 if (mlx5e_tunnel_proto_supported_tx(mdev, mlx5_get_proto_by_tunnel_type(tt)))
4864 return (mlx5_vxlan_allowed(mdev->vxlan) || mlx5_geneve_tx_allowed(mdev));
4867 static void mlx5e_build_nic_netdev(struct net_device *netdev)
4869 struct mlx5e_priv *priv = netdev_priv(netdev);
4870 struct mlx5_core_dev *mdev = priv->mdev;
4874 SET_NETDEV_DEV(netdev, mdev->device);
4876 netdev->netdev_ops = &mlx5e_netdev_ops;
4878 mlx5e_dcbnl_build_netdev(netdev);
4880 netdev->watchdog_timeo = 15 * HZ;
4882 netdev->ethtool_ops = &mlx5e_ethtool_ops;
4884 netdev->vlan_features |= NETIF_F_SG;
4885 netdev->vlan_features |= NETIF_F_HW_CSUM;
4886 netdev->vlan_features |= NETIF_F_GRO;
4887 netdev->vlan_features |= NETIF_F_TSO;
4888 netdev->vlan_features |= NETIF_F_TSO6;
4889 netdev->vlan_features |= NETIF_F_RXCSUM;
4890 netdev->vlan_features |= NETIF_F_RXHASH;
4891 netdev->vlan_features |= NETIF_F_GSO_PARTIAL;
4893 netdev->mpls_features |= NETIF_F_SG;
4894 netdev->mpls_features |= NETIF_F_HW_CSUM;
4895 netdev->mpls_features |= NETIF_F_TSO;
4896 netdev->mpls_features |= NETIF_F_TSO6;
4898 netdev->hw_enc_features |= NETIF_F_HW_VLAN_CTAG_TX;
4899 netdev->hw_enc_features |= NETIF_F_HW_VLAN_CTAG_RX;
4901 /* Tunneled LRO is not supported in the driver, and the same RQs are
4902 * shared between inner and outer TIRs, so the driver can't disable LRO
4903 * for inner TIRs while having it enabled for outer TIRs. Due to this,
4904 * block LRO altogether if the firmware declares tunneled LRO support.
4906 if (!!MLX5_CAP_ETH(mdev, lro_cap) &&
4907 !MLX5_CAP_ETH(mdev, tunnel_lro_vxlan) &&
4908 !MLX5_CAP_ETH(mdev, tunnel_lro_gre) &&
4909 mlx5e_check_fragmented_striding_rq_cap(mdev))
4910 netdev->vlan_features |= NETIF_F_LRO;
4912 netdev->hw_features = netdev->vlan_features;
4913 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
4914 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
4915 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
4916 netdev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
4918 if (mlx5e_tunnel_any_tx_proto_supported(mdev)) {
4919 netdev->hw_enc_features |= NETIF_F_HW_CSUM;
4920 netdev->hw_enc_features |= NETIF_F_TSO;
4921 netdev->hw_enc_features |= NETIF_F_TSO6;
4922 netdev->hw_enc_features |= NETIF_F_GSO_PARTIAL;
4925 if (mlx5_vxlan_allowed(mdev->vxlan) || mlx5_geneve_tx_allowed(mdev)) {
4926 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
4927 NETIF_F_GSO_UDP_TUNNEL_CSUM;
4928 netdev->hw_enc_features |= NETIF_F_GSO_UDP_TUNNEL |
4929 NETIF_F_GSO_UDP_TUNNEL_CSUM;
4930 netdev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
4931 netdev->vlan_features |= NETIF_F_GSO_UDP_TUNNEL |
4932 NETIF_F_GSO_UDP_TUNNEL_CSUM;
4935 if (mlx5e_tunnel_proto_supported_tx(mdev, IPPROTO_GRE)) {
4936 netdev->hw_features |= NETIF_F_GSO_GRE |
4937 NETIF_F_GSO_GRE_CSUM;
4938 netdev->hw_enc_features |= NETIF_F_GSO_GRE |
4939 NETIF_F_GSO_GRE_CSUM;
4940 netdev->gso_partial_features |= NETIF_F_GSO_GRE |
4941 NETIF_F_GSO_GRE_CSUM;
4944 if (mlx5e_tunnel_proto_supported_tx(mdev, IPPROTO_IPIP)) {
4945 netdev->hw_features |= NETIF_F_GSO_IPXIP4 |
4947 netdev->hw_enc_features |= NETIF_F_GSO_IPXIP4 |
4949 netdev->gso_partial_features |= NETIF_F_GSO_IPXIP4 |
4953 netdev->gso_partial_features |= NETIF_F_GSO_UDP_L4;
4954 netdev->hw_features |= NETIF_F_GSO_UDP_L4;
4955 netdev->features |= NETIF_F_GSO_UDP_L4;
4957 mlx5_query_port_fcs(mdev, &fcs_supported, &fcs_enabled);
4960 netdev->hw_features |= NETIF_F_RXALL;
4962 if (MLX5_CAP_ETH(mdev, scatter_fcs))
4963 netdev->hw_features |= NETIF_F_RXFCS;
4965 if (mlx5_qos_is_supported(mdev))
4966 netdev->hw_features |= NETIF_F_HW_TC;
4968 netdev->features = netdev->hw_features;
4972 netdev->features &= ~NETIF_F_RXALL;
4973 netdev->features &= ~NETIF_F_LRO;
4974 netdev->features &= ~NETIF_F_GRO_HW;
4975 netdev->features &= ~NETIF_F_RXFCS;
4977 #define FT_CAP(f) MLX5_CAP_FLOWTABLE(mdev, flow_table_properties_nic_receive.f)
4978 if (FT_CAP(flow_modify_en) &&
4979 FT_CAP(modify_root) &&
4980 FT_CAP(identified_miss_table_mode) &&
4981 FT_CAP(flow_table_modify)) {
4982 #if IS_ENABLED(CONFIG_MLX5_CLS_ACT)
4983 netdev->hw_features |= NETIF_F_HW_TC;
4985 #ifdef CONFIG_MLX5_EN_ARFS
4986 netdev->hw_features |= NETIF_F_NTUPLE;
4990 netdev->features |= NETIF_F_HIGHDMA;
4991 netdev->features |= NETIF_F_HW_VLAN_STAG_FILTER;
4993 netdev->priv_flags |= IFF_UNICAST_FLT;
4995 netif_set_tso_max_size(netdev, GSO_MAX_SIZE);
4996 mlx5e_set_netdev_dev_addr(netdev);
4997 mlx5e_ipsec_build_netdev(priv);
4998 mlx5e_ktls_build_netdev(priv);
5001 void mlx5e_create_q_counters(struct mlx5e_priv *priv)
5003 u32 out[MLX5_ST_SZ_DW(alloc_q_counter_out)] = {};
5004 u32 in[MLX5_ST_SZ_DW(alloc_q_counter_in)] = {};
5005 struct mlx5_core_dev *mdev = priv->mdev;
5008 MLX5_SET(alloc_q_counter_in, in, opcode, MLX5_CMD_OP_ALLOC_Q_COUNTER);
5009 err = mlx5_cmd_exec_inout(mdev, alloc_q_counter, in, out);
5012 MLX5_GET(alloc_q_counter_out, out, counter_set_id);
5014 err = mlx5_cmd_exec_inout(mdev, alloc_q_counter, in, out);
5016 priv->drop_rq_q_counter =
5017 MLX5_GET(alloc_q_counter_out, out, counter_set_id);
5020 void mlx5e_destroy_q_counters(struct mlx5e_priv *priv)
5022 u32 in[MLX5_ST_SZ_DW(dealloc_q_counter_in)] = {};
5024 MLX5_SET(dealloc_q_counter_in, in, opcode,
5025 MLX5_CMD_OP_DEALLOC_Q_COUNTER);
5026 if (priv->q_counter) {
5027 MLX5_SET(dealloc_q_counter_in, in, counter_set_id,
5029 mlx5_cmd_exec_in(priv->mdev, dealloc_q_counter, in);
5032 if (priv->drop_rq_q_counter) {
5033 MLX5_SET(dealloc_q_counter_in, in, counter_set_id,
5034 priv->drop_rq_q_counter);
5035 mlx5_cmd_exec_in(priv->mdev, dealloc_q_counter, in);
5039 static int mlx5e_nic_init(struct mlx5_core_dev *mdev,
5040 struct net_device *netdev)
5042 struct mlx5e_priv *priv = netdev_priv(netdev);
5043 struct mlx5e_flow_steering *fs;
5046 mlx5e_build_nic_params(priv, &priv->xsk, netdev->mtu);
5047 mlx5e_vxlan_set_netdev_info(priv);
5049 mlx5e_timestamp_init(priv);
5051 fs = mlx5e_fs_init(priv->profile, mdev,
5052 !test_bit(MLX5E_STATE_DESTROYING, &priv->state));
5055 mlx5_core_err(mdev, "FS initialization failed, %d\n", err);
5060 err = mlx5e_ipsec_init(priv);
5062 mlx5_core_err(mdev, "IPSec initialization failed, %d\n", err);
5064 err = mlx5e_ktls_init(priv);
5066 mlx5_core_err(mdev, "TLS initialization failed, %d\n", err);
5068 mlx5e_health_create_reporters(priv);
5072 static void mlx5e_nic_cleanup(struct mlx5e_priv *priv)
5074 mlx5e_health_destroy_reporters(priv);
5075 mlx5e_ktls_cleanup(priv);
5076 mlx5e_ipsec_cleanup(priv);
5077 mlx5e_fs_cleanup(priv->fs);
5080 static int mlx5e_init_nic_rx(struct mlx5e_priv *priv)
5082 struct mlx5_core_dev *mdev = priv->mdev;
5083 enum mlx5e_rx_res_features features;
5086 priv->rx_res = mlx5e_rx_res_alloc();
5090 mlx5e_create_q_counters(priv);
5092 err = mlx5e_open_drop_rq(priv, &priv->drop_rq);
5094 mlx5_core_err(mdev, "open drop rq failed, %d\n", err);
5095 goto err_destroy_q_counters;
5098 features = MLX5E_RX_RES_FEATURE_XSK | MLX5E_RX_RES_FEATURE_PTP;
5099 if (priv->channels.params.tunneled_offload_en)
5100 features |= MLX5E_RX_RES_FEATURE_INNER_FT;
5101 err = mlx5e_rx_res_init(priv->rx_res, priv->mdev, features,
5102 priv->max_nch, priv->drop_rq.rqn,
5103 &priv->channels.params.packet_merge,
5104 priv->channels.params.num_channels);
5106 goto err_close_drop_rq;
5108 err = mlx5e_create_flow_steering(priv);
5110 mlx5_core_warn(mdev, "create flow steering failed, %d\n", err);
5111 goto err_destroy_rx_res;
5114 err = mlx5e_tc_nic_init(priv);
5116 goto err_destroy_flow_steering;
5118 err = mlx5e_accel_init_rx(priv);
5120 goto err_tc_nic_cleanup;
5122 #ifdef CONFIG_MLX5_EN_ARFS
5123 priv->netdev->rx_cpu_rmap = mlx5_eq_table_get_rmap(priv->mdev);
5129 mlx5e_tc_nic_cleanup(priv);
5130 err_destroy_flow_steering:
5131 mlx5e_destroy_flow_steering(priv);
5133 mlx5e_rx_res_destroy(priv->rx_res);
5135 mlx5e_close_drop_rq(&priv->drop_rq);
5136 err_destroy_q_counters:
5137 mlx5e_destroy_q_counters(priv);
5138 mlx5e_rx_res_free(priv->rx_res);
5139 priv->rx_res = NULL;
5143 static void mlx5e_cleanup_nic_rx(struct mlx5e_priv *priv)
5145 mlx5e_accel_cleanup_rx(priv);
5146 mlx5e_tc_nic_cleanup(priv);
5147 mlx5e_destroy_flow_steering(priv);
5148 mlx5e_rx_res_destroy(priv->rx_res);
5149 mlx5e_close_drop_rq(&priv->drop_rq);
5150 mlx5e_destroy_q_counters(priv);
5151 mlx5e_rx_res_free(priv->rx_res);
5152 priv->rx_res = NULL;
5155 static void mlx5e_set_mqprio_rl(struct mlx5e_priv *priv)
5157 struct mlx5e_params *params;
5158 struct mlx5e_mqprio_rl *rl;
5160 params = &priv->channels.params;
5161 if (params->mqprio.mode != TC_MQPRIO_MODE_CHANNEL)
5164 rl = mlx5e_mqprio_rl_create(priv->mdev, params->mqprio.num_tc,
5165 params->mqprio.channel.max_rate);
5168 priv->mqprio_rl = rl;
5169 mlx5e_mqprio_rl_update_params(params, rl);
5172 static int mlx5e_init_nic_tx(struct mlx5e_priv *priv)
5176 err = mlx5e_create_tises(priv);
5178 mlx5_core_warn(priv->mdev, "create tises failed, %d\n", err);
5182 err = mlx5e_accel_init_tx(priv);
5184 goto err_destroy_tises;
5186 mlx5e_set_mqprio_rl(priv);
5187 mlx5e_dcbnl_initialize(priv);
5191 mlx5e_destroy_tises(priv);
5195 static void mlx5e_nic_enable(struct mlx5e_priv *priv)
5197 struct net_device *netdev = priv->netdev;
5198 struct mlx5_core_dev *mdev = priv->mdev;
5200 mlx5e_fs_init_l2_addr(priv->fs, netdev);
5202 /* Marking the link as currently not needed by the Driver */
5203 if (!netif_running(netdev))
5204 mlx5e_modify_admin_state(mdev, MLX5_PORT_DOWN);
5206 mlx5e_set_netdev_mtu_boundaries(priv);
5207 mlx5e_set_dev_port_mtu(priv);
5209 mlx5_lag_add_netdev(mdev, netdev);
5211 mlx5e_enable_async_events(priv);
5212 mlx5e_enable_blocking_events(priv);
5213 if (mlx5e_monitor_counter_supported(priv))
5214 mlx5e_monitor_counter_init(priv);
5216 mlx5e_hv_vhca_stats_create(priv);
5217 if (netdev->reg_state != NETREG_REGISTERED)
5219 mlx5e_dcbnl_init_app(priv);
5221 mlx5e_nic_set_rx_mode(priv);
5224 if (netif_running(netdev))
5226 udp_tunnel_nic_reset_ntf(priv->netdev);
5227 netif_device_attach(netdev);
5231 static void mlx5e_nic_disable(struct mlx5e_priv *priv)
5233 struct mlx5_core_dev *mdev = priv->mdev;
5235 if (priv->netdev->reg_state == NETREG_REGISTERED)
5236 mlx5e_dcbnl_delete_app(priv);
5239 if (netif_running(priv->netdev))
5240 mlx5e_close(priv->netdev);
5241 netif_device_detach(priv->netdev);
5244 mlx5e_nic_set_rx_mode(priv);
5246 mlx5e_hv_vhca_stats_destroy(priv);
5247 if (mlx5e_monitor_counter_supported(priv))
5248 mlx5e_monitor_counter_cleanup(priv);
5250 mlx5e_disable_blocking_events(priv);
5251 if (priv->en_trap) {
5252 mlx5e_deactivate_trap(priv);
5253 mlx5e_close_trap(priv->en_trap);
5254 priv->en_trap = NULL;
5256 mlx5e_disable_async_events(priv);
5257 mlx5_lag_remove_netdev(mdev, priv->netdev);
5258 mlx5_vxlan_reset_to_default(mdev->vxlan);
5261 int mlx5e_update_nic_rx(struct mlx5e_priv *priv)
5263 return mlx5e_refresh_tirs(priv, false, false);
5266 static const struct mlx5e_profile mlx5e_nic_profile = {
5267 .init = mlx5e_nic_init,
5268 .cleanup = mlx5e_nic_cleanup,
5269 .init_rx = mlx5e_init_nic_rx,
5270 .cleanup_rx = mlx5e_cleanup_nic_rx,
5271 .init_tx = mlx5e_init_nic_tx,
5272 .cleanup_tx = mlx5e_cleanup_nic_tx,
5273 .enable = mlx5e_nic_enable,
5274 .disable = mlx5e_nic_disable,
5275 .update_rx = mlx5e_update_nic_rx,
5276 .update_stats = mlx5e_stats_update_ndo_stats,
5277 .update_carrier = mlx5e_update_carrier,
5278 .rx_handlers = &mlx5e_rx_handlers_nic,
5279 .max_tc = MLX5E_MAX_NUM_TC,
5280 .rq_groups = MLX5E_NUM_RQ_GROUPS(XSK),
5281 .stats_grps = mlx5e_nic_stats_grps,
5282 .stats_grps_num = mlx5e_nic_stats_grps_num,
5283 .features = BIT(MLX5E_PROFILE_FEATURE_PTP_RX) |
5284 BIT(MLX5E_PROFILE_FEATURE_PTP_TX) |
5285 BIT(MLX5E_PROFILE_FEATURE_QOS_HTB) |
5286 BIT(MLX5E_PROFILE_FEATURE_FS_VLAN) |
5287 BIT(MLX5E_PROFILE_FEATURE_FS_TC),
5290 static int mlx5e_profile_max_num_channels(struct mlx5_core_dev *mdev,
5291 const struct mlx5e_profile *profile)
5295 nch = mlx5e_get_max_num_channels(mdev);
5297 if (profile->max_nch_limit)
5298 nch = min_t(int, nch, profile->max_nch_limit(mdev));
5303 mlx5e_calc_max_nch(struct mlx5_core_dev *mdev, struct net_device *netdev,
5304 const struct mlx5e_profile *profile)
5307 unsigned int max_nch, tmp;
5309 /* core resources */
5310 max_nch = mlx5e_profile_max_num_channels(mdev, profile);
5312 /* netdev rx queues */
5313 tmp = netdev->num_rx_queues / max_t(u8, profile->rq_groups, 1);
5314 max_nch = min_t(unsigned int, max_nch, tmp);
5316 /* netdev tx queues */
5317 tmp = netdev->num_tx_queues;
5318 if (mlx5_qos_is_supported(mdev))
5319 tmp -= mlx5e_qos_max_leaf_nodes(mdev);
5320 if (MLX5_CAP_GEN(mdev, ts_cqe_to_dest_cqn))
5321 tmp -= profile->max_tc;
5322 tmp = tmp / profile->max_tc;
5323 max_nch = min_t(unsigned int, max_nch, tmp);
5328 int mlx5e_get_pf_num_tirs(struct mlx5_core_dev *mdev)
5330 /* Indirect TIRS: 2 sets of TTCs (inner + outer steering)
5331 * and 1 set of direct TIRS
5333 return 2 * MLX5E_NUM_INDIR_TIRS
5334 + mlx5e_profile_max_num_channels(mdev, &mlx5e_nic_profile);
5337 void mlx5e_set_rx_mode_work(struct work_struct *work)
5339 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
5342 return mlx5e_fs_set_rx_mode_work(priv->fs, priv->netdev);
5345 /* mlx5e generic netdev management API (move to en_common.c) */
5346 int mlx5e_priv_init(struct mlx5e_priv *priv,
5347 const struct mlx5e_profile *profile,
5348 struct net_device *netdev,
5349 struct mlx5_core_dev *mdev)
5351 int nch, num_txqs, node;
5354 num_txqs = netdev->num_tx_queues;
5355 nch = mlx5e_calc_max_nch(mdev, netdev, profile);
5356 node = dev_to_node(mlx5_core_dma_dev(mdev));
5360 priv->netdev = netdev;
5361 priv->msglevel = MLX5E_MSG_LEVEL;
5362 priv->max_nch = nch;
5363 priv->max_opened_tc = 1;
5365 if (!alloc_cpumask_var(&priv->scratchpad.cpumask, GFP_KERNEL))
5368 mutex_init(&priv->state_lock);
5370 err = mlx5e_selq_init(&priv->selq, &priv->state_lock);
5372 goto err_free_cpumask;
5374 INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
5375 INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
5376 INIT_WORK(&priv->tx_timeout_work, mlx5e_tx_timeout_work);
5377 INIT_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
5379 priv->wq = create_singlethread_workqueue("mlx5e");
5383 priv->txq2sq = kcalloc_node(num_txqs, sizeof(*priv->txq2sq), GFP_KERNEL, node);
5385 goto err_destroy_workqueue;
5387 priv->tx_rates = kcalloc_node(num_txqs, sizeof(*priv->tx_rates), GFP_KERNEL, node);
5388 if (!priv->tx_rates)
5389 goto err_free_txq2sq;
5391 priv->channel_stats =
5392 kcalloc_node(nch, sizeof(*priv->channel_stats), GFP_KERNEL, node);
5393 if (!priv->channel_stats)
5394 goto err_free_tx_rates;
5399 kfree(priv->tx_rates);
5401 kfree(priv->txq2sq);
5402 err_destroy_workqueue:
5403 destroy_workqueue(priv->wq);
5405 mlx5e_selq_cleanup(&priv->selq);
5407 free_cpumask_var(priv->scratchpad.cpumask);
5411 void mlx5e_priv_cleanup(struct mlx5e_priv *priv)
5415 /* bail if change profile failed and also rollback failed */
5419 for (i = 0; i < priv->stats_nch; i++)
5420 kvfree(priv->channel_stats[i]);
5421 kfree(priv->channel_stats);
5422 kfree(priv->tx_rates);
5423 kfree(priv->txq2sq);
5424 destroy_workqueue(priv->wq);
5425 mutex_lock(&priv->state_lock);
5426 mlx5e_selq_cleanup(&priv->selq);
5427 mutex_unlock(&priv->state_lock);
5428 free_cpumask_var(priv->scratchpad.cpumask);
5430 for (i = 0; i < priv->htb_max_qos_sqs; i++)
5431 kfree(priv->htb_qos_sq_stats[i]);
5432 kvfree(priv->htb_qos_sq_stats);
5434 memset(priv, 0, sizeof(*priv));
5437 static unsigned int mlx5e_get_max_num_txqs(struct mlx5_core_dev *mdev,
5438 const struct mlx5e_profile *profile)
5440 unsigned int nch, ptp_txqs, qos_txqs;
5442 nch = mlx5e_profile_max_num_channels(mdev, profile);
5444 ptp_txqs = MLX5_CAP_GEN(mdev, ts_cqe_to_dest_cqn) &&
5445 mlx5e_profile_feature_cap(profile, PTP_TX) ?
5446 profile->max_tc : 0;
5448 qos_txqs = mlx5_qos_is_supported(mdev) &&
5449 mlx5e_profile_feature_cap(profile, QOS_HTB) ?
5450 mlx5e_qos_max_leaf_nodes(mdev) : 0;
5452 return nch * profile->max_tc + ptp_txqs + qos_txqs;
5455 static unsigned int mlx5e_get_max_num_rxqs(struct mlx5_core_dev *mdev,
5456 const struct mlx5e_profile *profile)
5460 nch = mlx5e_profile_max_num_channels(mdev, profile);
5462 return nch * profile->rq_groups;
5466 mlx5e_create_netdev(struct mlx5_core_dev *mdev, const struct mlx5e_profile *profile)
5468 struct net_device *netdev;
5469 unsigned int txqs, rxqs;
5472 txqs = mlx5e_get_max_num_txqs(mdev, profile);
5473 rxqs = mlx5e_get_max_num_rxqs(mdev, profile);
5475 netdev = alloc_etherdev_mqs(sizeof(struct mlx5e_priv), txqs, rxqs);
5477 mlx5_core_err(mdev, "alloc_etherdev_mqs() failed\n");
5481 err = mlx5e_priv_init(netdev_priv(netdev), profile, netdev, mdev);
5483 mlx5_core_err(mdev, "mlx5e_priv_init failed, err=%d\n", err);
5484 goto err_free_netdev;
5487 netif_carrier_off(netdev);
5488 netif_tx_disable(netdev);
5489 dev_net_set(netdev, mlx5_core_net(mdev));
5494 free_netdev(netdev);
5499 static void mlx5e_update_features(struct net_device *netdev)
5501 if (netdev->reg_state != NETREG_REGISTERED)
5502 return; /* features will be updated on netdev registration */
5505 netdev_update_features(netdev);
5509 static void mlx5e_reset_channels(struct net_device *netdev)
5511 netdev_reset_tc(netdev);
5514 int mlx5e_attach_netdev(struct mlx5e_priv *priv)
5516 const bool take_rtnl = priv->netdev->reg_state == NETREG_REGISTERED;
5517 const struct mlx5e_profile *profile = priv->profile;
5521 clear_bit(MLX5E_STATE_DESTROYING, &priv->state);
5523 mlx5e_fs_set_state_destroy(priv->fs,
5524 !test_bit(MLX5E_STATE_DESTROYING, &priv->state));
5526 /* max number of channels may have changed */
5527 max_nch = mlx5e_calc_max_nch(priv->mdev, priv->netdev, profile);
5528 if (priv->channels.params.num_channels > max_nch) {
5529 mlx5_core_warn(priv->mdev, "MLX5E: Reducing number of channels to %d\n", max_nch);
5530 /* Reducing the number of channels - RXFH has to be reset, and
5531 * mlx5e_num_channels_changed below will build the RQT.
5533 priv->netdev->priv_flags &= ~IFF_RXFH_CONFIGURED;
5534 priv->channels.params.num_channels = max_nch;
5535 if (priv->channels.params.mqprio.mode == TC_MQPRIO_MODE_CHANNEL) {
5536 mlx5_core_warn(priv->mdev, "MLX5E: Disabling MQPRIO channel mode\n");
5537 mlx5e_params_mqprio_reset(&priv->channels.params);
5540 if (max_nch != priv->max_nch) {
5541 mlx5_core_warn(priv->mdev,
5542 "MLX5E: Updating max number of channels from %u to %u\n",
5543 priv->max_nch, max_nch);
5544 priv->max_nch = max_nch;
5547 /* 1. Set the real number of queues in the kernel the first time.
5548 * 2. Set our default XPS cpumask.
5551 * rtnl_lock is required by netif_set_real_num_*_queues in case the
5552 * netdev has been registered by this point (if this function was called
5553 * in the reload or resume flow).
5557 err = mlx5e_num_channels_changed(priv);
5563 err = profile->init_tx(priv);
5567 err = profile->init_rx(priv);
5569 goto err_cleanup_tx;
5571 if (profile->enable)
5572 profile->enable(priv);
5574 mlx5e_update_features(priv->netdev);
5579 profile->cleanup_tx(priv);
5582 mlx5e_reset_channels(priv->netdev);
5583 set_bit(MLX5E_STATE_DESTROYING, &priv->state);
5585 mlx5e_fs_set_state_destroy(priv->fs,
5586 !test_bit(MLX5E_STATE_DESTROYING, &priv->state));
5587 cancel_work_sync(&priv->update_stats_work);
5591 void mlx5e_detach_netdev(struct mlx5e_priv *priv)
5593 const struct mlx5e_profile *profile = priv->profile;
5595 set_bit(MLX5E_STATE_DESTROYING, &priv->state);
5597 mlx5e_fs_set_state_destroy(priv->fs,
5598 !test_bit(MLX5E_STATE_DESTROYING, &priv->state));
5600 if (profile->disable)
5601 profile->disable(priv);
5602 flush_workqueue(priv->wq);
5604 profile->cleanup_rx(priv);
5605 profile->cleanup_tx(priv);
5606 mlx5e_reset_channels(priv->netdev);
5607 cancel_work_sync(&priv->update_stats_work);
5611 mlx5e_netdev_attach_profile(struct net_device *netdev, struct mlx5_core_dev *mdev,
5612 const struct mlx5e_profile *new_profile, void *new_ppriv)
5614 struct mlx5e_priv *priv = netdev_priv(netdev);
5617 err = mlx5e_priv_init(priv, new_profile, netdev, mdev);
5619 mlx5_core_err(mdev, "mlx5e_priv_init failed, err=%d\n", err);
5622 netif_carrier_off(netdev);
5623 priv->profile = new_profile;
5624 priv->ppriv = new_ppriv;
5625 err = new_profile->init(priv->mdev, priv->netdev);
5628 err = mlx5e_attach_netdev(priv);
5630 goto profile_cleanup;
5634 new_profile->cleanup(priv);
5636 mlx5e_priv_cleanup(priv);
5640 int mlx5e_netdev_change_profile(struct mlx5e_priv *priv,
5641 const struct mlx5e_profile *new_profile, void *new_ppriv)
5643 const struct mlx5e_profile *orig_profile = priv->profile;
5644 struct net_device *netdev = priv->netdev;
5645 struct mlx5_core_dev *mdev = priv->mdev;
5646 void *orig_ppriv = priv->ppriv;
5647 int err, rollback_err;
5649 /* cleanup old profile */
5650 mlx5e_detach_netdev(priv);
5651 priv->profile->cleanup(priv);
5652 mlx5e_priv_cleanup(priv);
5654 err = mlx5e_netdev_attach_profile(netdev, mdev, new_profile, new_ppriv);
5655 if (err) { /* roll back to original profile */
5656 netdev_warn(netdev, "%s: new profile init failed, %d\n", __func__, err);
5663 rollback_err = mlx5e_netdev_attach_profile(netdev, mdev, orig_profile, orig_ppriv);
5665 netdev_err(netdev, "%s: failed to rollback to orig profile, %d\n",
5666 __func__, rollback_err);
5670 void mlx5e_netdev_attach_nic_profile(struct mlx5e_priv *priv)
5672 mlx5e_netdev_change_profile(priv, &mlx5e_nic_profile, NULL);
5675 void mlx5e_destroy_netdev(struct mlx5e_priv *priv)
5677 struct net_device *netdev = priv->netdev;
5679 mlx5e_priv_cleanup(priv);
5680 free_netdev(netdev);
5683 static int mlx5e_resume(struct auxiliary_device *adev)
5685 struct mlx5_adev *edev = container_of(adev, struct mlx5_adev, adev);
5686 struct mlx5e_priv *priv = auxiliary_get_drvdata(adev);
5687 struct net_device *netdev = priv->netdev;
5688 struct mlx5_core_dev *mdev = edev->mdev;
5691 if (netif_device_present(netdev))
5694 err = mlx5e_create_mdev_resources(mdev);
5698 err = mlx5e_attach_netdev(priv);
5700 mlx5e_destroy_mdev_resources(mdev);
5707 static int mlx5e_suspend(struct auxiliary_device *adev, pm_message_t state)
5709 struct mlx5e_priv *priv = auxiliary_get_drvdata(adev);
5710 struct net_device *netdev = priv->netdev;
5711 struct mlx5_core_dev *mdev = priv->mdev;
5713 if (!netif_device_present(netdev))
5716 mlx5e_detach_netdev(priv);
5717 mlx5e_destroy_mdev_resources(mdev);
5721 static int mlx5e_probe(struct auxiliary_device *adev,
5722 const struct auxiliary_device_id *id)
5724 struct mlx5_adev *edev = container_of(adev, struct mlx5_adev, adev);
5725 const struct mlx5e_profile *profile = &mlx5e_nic_profile;
5726 struct mlx5_core_dev *mdev = edev->mdev;
5727 struct net_device *netdev;
5728 pm_message_t state = {};
5729 struct mlx5e_priv *priv;
5732 netdev = mlx5e_create_netdev(mdev, profile);
5734 mlx5_core_err(mdev, "mlx5e_create_netdev failed\n");
5738 mlx5e_build_nic_netdev(netdev);
5740 priv = netdev_priv(netdev);
5741 auxiliary_set_drvdata(adev, priv);
5743 priv->profile = profile;
5746 err = mlx5e_devlink_port_register(priv);
5748 mlx5_core_err(mdev, "mlx5e_devlink_port_register failed, %d\n", err);
5749 goto err_destroy_netdev;
5752 err = profile->init(mdev, netdev);
5754 mlx5_core_err(mdev, "mlx5e_nic_profile init failed, %d\n", err);
5755 goto err_devlink_cleanup;
5758 err = mlx5e_resume(adev);
5760 mlx5_core_err(mdev, "mlx5e_resume failed, %d\n", err);
5761 goto err_profile_cleanup;
5764 err = register_netdev(netdev);
5766 mlx5_core_err(mdev, "register_netdev failed, %d\n", err);
5770 mlx5e_devlink_port_type_eth_set(priv);
5772 mlx5e_dcbnl_init_app(priv);
5773 mlx5_uplink_netdev_set(mdev, netdev);
5777 mlx5e_suspend(adev, state);
5778 err_profile_cleanup:
5779 profile->cleanup(priv);
5780 err_devlink_cleanup:
5781 mlx5e_devlink_port_unregister(priv);
5783 mlx5e_destroy_netdev(priv);
5787 static void mlx5e_remove(struct auxiliary_device *adev)
5789 struct mlx5e_priv *priv = auxiliary_get_drvdata(adev);
5790 pm_message_t state = {};
5792 mlx5e_dcbnl_delete_app(priv);
5793 unregister_netdev(priv->netdev);
5794 mlx5e_suspend(adev, state);
5795 priv->profile->cleanup(priv);
5796 mlx5e_devlink_port_unregister(priv);
5797 mlx5e_destroy_netdev(priv);
5800 static const struct auxiliary_device_id mlx5e_id_table[] = {
5801 { .name = MLX5_ADEV_NAME ".eth", },
5805 MODULE_DEVICE_TABLE(auxiliary, mlx5e_id_table);
5807 static struct auxiliary_driver mlx5e_driver = {
5809 .probe = mlx5e_probe,
5810 .remove = mlx5e_remove,
5811 .suspend = mlx5e_suspend,
5812 .resume = mlx5e_resume,
5813 .id_table = mlx5e_id_table,
5816 int mlx5e_init(void)
5820 mlx5e_build_ptys2ethtool_map();
5821 ret = auxiliary_driver_register(&mlx5e_driver);
5825 ret = mlx5e_rep_init();
5827 auxiliary_driver_unregister(&mlx5e_driver);
5831 void mlx5e_cleanup(void)
5833 mlx5e_rep_cleanup();
5834 auxiliary_driver_unregister(&mlx5e_driver);