]> git.ipfire.org Git - thirdparty/linux.git/blob - drivers/net/ethernet/mellanox/mlx5/core/en_main.c
projects / thirdparty / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
riscv: select ARCH_HAS_STRICT_KERNEL_RWX only if MMU
[thirdparty/linux.git] / drivers / net / ethernet / mellanox / mlx5 / core / en_main.c
1 /*
2 * Copyright (c) 2015-2016, Mellanox Technologies. All rights reserved.
3 *
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:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
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.
22 *
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
30 * SOFTWARE.
31 */
32
33 #include <net/tc_act/tc_gact.h>
34 #include <net/pkt_cls.h>
35 #include <linux/mlx5/fs.h>
36 #include <net/vxlan.h>
37 #include <net/geneve.h>
38 #include <linux/bpf.h>
39 #include <linux/if_bridge.h>
40 #include <net/page_pool.h>
41 #include <net/xdp_sock.h>
42 #include "eswitch.h"
43 #include "en.h"
44 #include "en/txrx.h"
45 #include "en_tc.h"
46 #include "en_rep.h"
47 #include "en_accel/ipsec.h"
48 #include "en_accel/ipsec_rxtx.h"
49 #include "en_accel/en_accel.h"
50 #include "en_accel/tls.h"
51 #include "accel/ipsec.h"
52 #include "accel/tls.h"
53 #include "lib/vxlan.h"
54 #include "lib/clock.h"
55 #include "en/port.h"
56 #include "en/xdp.h"
57 #include "lib/eq.h"
58 #include "en/monitor_stats.h"
59 #include "en/health.h"
60 #include "en/params.h"
61 #include "en/xsk/umem.h"
62 #include "en/xsk/setup.h"
63 #include "en/xsk/rx.h"
64 #include "en/xsk/tx.h"
65 #include "en/hv_vhca_stats.h"
66 #include "en/devlink.h"
67 #include "lib/mlx5.h"
68
69
70 bool mlx5e_check_fragmented_striding_rq_cap(struct mlx5_core_dev *mdev)
71 {
72 bool striding_rq_umr = MLX5_CAP_GEN(mdev, striding_rq) &&
73 MLX5_CAP_GEN(mdev, umr_ptr_rlky) &&
74 MLX5_CAP_ETH(mdev, reg_umr_sq);
75 u16 max_wqe_sz_cap = MLX5_CAP_GEN(mdev, max_wqe_sz_sq);
76 bool inline_umr = MLX5E_UMR_WQE_INLINE_SZ <= max_wqe_sz_cap;
77
78 if (!striding_rq_umr)
79 return false;
80 if (!inline_umr) {
81 mlx5_core_warn(mdev, "Cannot support Striding RQ: UMR WQE size (%d) exceeds maximum supported (%d).\n",
82 (int)MLX5E_UMR_WQE_INLINE_SZ, max_wqe_sz_cap);
83 return false;
84 }
85 return true;
86 }
87
88 void mlx5e_init_rq_type_params(struct mlx5_core_dev *mdev,
89 struct mlx5e_params *params)
90 {
91 params->log_rq_mtu_frames = is_kdump_kernel() ?
92 MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE :
93 MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
94
95 mlx5_core_info(mdev, "MLX5E: StrdRq(%d) RqSz(%ld) StrdSz(%ld) RxCqeCmprss(%d)\n",
96 params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ,
97 params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ ?
98 BIT(mlx5e_mpwqe_get_log_rq_size(params, NULL)) :
99 BIT(params->log_rq_mtu_frames),
100 BIT(mlx5e_mpwqe_get_log_stride_size(mdev, params, NULL)),
101 MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS));
102 }
103
104 bool mlx5e_striding_rq_possible(struct mlx5_core_dev *mdev,
105 struct mlx5e_params *params)
106 {
107 if (!mlx5e_check_fragmented_striding_rq_cap(mdev))
108 return false;
109
110 if (MLX5_IPSEC_DEV(mdev))
111 return false;
112
113 if (params->xdp_prog) {
114 /* XSK params are not considered here. If striding RQ is in use,
115 * and an XSK is being opened, mlx5e_rx_mpwqe_is_linear_skb will
116 * be called with the known XSK params.
117 */
118 if (!mlx5e_rx_mpwqe_is_linear_skb(mdev, params, NULL))
119 return false;
120 }
121
122 return true;
123 }
124
125 void mlx5e_set_rq_type(struct mlx5_core_dev *mdev, struct mlx5e_params *params)
126 {
127 params->rq_wq_type = mlx5e_striding_rq_possible(mdev, params) &&
128 MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_STRIDING_RQ) ?
129 MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ :
130 MLX5_WQ_TYPE_CYCLIC;
131 }
132
133 void mlx5e_update_carrier(struct mlx5e_priv *priv)
134 {
135 struct mlx5_core_dev *mdev = priv->mdev;
136 u8 port_state;
137
138 port_state = mlx5_query_vport_state(mdev,
139 MLX5_VPORT_STATE_OP_MOD_VNIC_VPORT,
140 0);
141
142 if (port_state == VPORT_STATE_UP) {
143 netdev_info(priv->netdev, "Link up\n");
144 netif_carrier_on(priv->netdev);
145 } else {
146 netdev_info(priv->netdev, "Link down\n");
147 netif_carrier_off(priv->netdev);
148 }
149 }
150
151 static void mlx5e_update_carrier_work(struct work_struct *work)
152 {
153 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
154 update_carrier_work);
155
156 mutex_lock(&priv->state_lock);
157 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
158 if (priv->profile->update_carrier)
159 priv->profile->update_carrier(priv);
160 mutex_unlock(&priv->state_lock);
161 }
162
163 void mlx5e_update_ndo_stats(struct mlx5e_priv *priv)
164 {
165 int i;
166
167 for (i = mlx5e_nic_stats_grps_num(priv) - 1; i >= 0; i--)
168 if (mlx5e_nic_stats_grps[i]->update_stats_mask &
169 MLX5E_NDO_UPDATE_STATS)
170 mlx5e_nic_stats_grps[i]->update_stats(priv);
171 }
172
173 static void mlx5e_update_stats_work(struct work_struct *work)
174 {
175 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
176 update_stats_work);
177
178 mutex_lock(&priv->state_lock);
179 priv->profile->update_stats(priv);
180 mutex_unlock(&priv->state_lock);
181 }
182
183 void mlx5e_queue_update_stats(struct mlx5e_priv *priv)
184 {
185 if (!priv->profile->update_stats)
186 return;
187
188 if (unlikely(test_bit(MLX5E_STATE_DESTROYING, &priv->state)))
189 return;
190
191 queue_work(priv->wq, &priv->update_stats_work);
192 }
193
194 static int async_event(struct notifier_block *nb, unsigned long event, void *data)
195 {
196 struct mlx5e_priv *priv = container_of(nb, struct mlx5e_priv, events_nb);
197 struct mlx5_eqe *eqe = data;
198
199 if (event != MLX5_EVENT_TYPE_PORT_CHANGE)
200 return NOTIFY_DONE;
201
202 switch (eqe->sub_type) {
203 case MLX5_PORT_CHANGE_SUBTYPE_DOWN:
204 case MLX5_PORT_CHANGE_SUBTYPE_ACTIVE:
205 queue_work(priv->wq, &priv->update_carrier_work);
206 break;
207 default:
208 return NOTIFY_DONE;
209 }
210
211 return NOTIFY_OK;
212 }
213
214 static void mlx5e_enable_async_events(struct mlx5e_priv *priv)
215 {
216 priv->events_nb.notifier_call = async_event;
217 mlx5_notifier_register(priv->mdev, &priv->events_nb);
218 }
219
220 static void mlx5e_disable_async_events(struct mlx5e_priv *priv)
221 {
222 mlx5_notifier_unregister(priv->mdev, &priv->events_nb);
223 }
224
225 static inline void mlx5e_build_umr_wqe(struct mlx5e_rq *rq,
226 struct mlx5e_icosq *sq,
227 struct mlx5e_umr_wqe *wqe)
228 {
229 struct mlx5_wqe_ctrl_seg *cseg = &wqe->ctrl;
230 struct mlx5_wqe_umr_ctrl_seg *ucseg = &wqe->uctrl;
231 u8 ds_cnt = DIV_ROUND_UP(MLX5E_UMR_WQE_INLINE_SZ, MLX5_SEND_WQE_DS);
232
233 cseg->qpn_ds = cpu_to_be32((sq->sqn << MLX5_WQE_CTRL_QPN_SHIFT) |
234 ds_cnt);
235 cseg->fm_ce_se = MLX5_WQE_CTRL_CQ_UPDATE;
236 cseg->imm = rq->mkey_be;
237
238 ucseg->flags = MLX5_UMR_TRANSLATION_OFFSET_EN | MLX5_UMR_INLINE;
239 ucseg->xlt_octowords =
240 cpu_to_be16(MLX5_MTT_OCTW(MLX5_MPWRQ_PAGES_PER_WQE));
241 ucseg->mkey_mask = cpu_to_be64(MLX5_MKEY_MASK_FREE);
242 }
243
244 static int mlx5e_rq_alloc_mpwqe_info(struct mlx5e_rq *rq,
245 struct mlx5e_channel *c)
246 {
247 int wq_sz = mlx5_wq_ll_get_size(&rq->mpwqe.wq);
248
249 rq->mpwqe.info = kvzalloc_node(array_size(wq_sz,
250 sizeof(*rq->mpwqe.info)),
251 GFP_KERNEL, cpu_to_node(c->cpu));
252 if (!rq->mpwqe.info)
253 return -ENOMEM;
254
255 mlx5e_build_umr_wqe(rq, &c->icosq, &rq->mpwqe.umr_wqe);
256
257 return 0;
258 }
259
260 static int mlx5e_create_umr_mkey(struct mlx5_core_dev *mdev,
261 u64 npages, u8 page_shift,
262 struct mlx5_core_mkey *umr_mkey)
263 {
264 int inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
265 void *mkc;
266 u32 *in;
267 int err;
268
269 in = kvzalloc(inlen, GFP_KERNEL);
270 if (!in)
271 return -ENOMEM;
272
273 mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
274
275 MLX5_SET(mkc, mkc, free, 1);
276 MLX5_SET(mkc, mkc, umr_en, 1);
277 MLX5_SET(mkc, mkc, lw, 1);
278 MLX5_SET(mkc, mkc, lr, 1);
279 MLX5_SET(mkc, mkc, access_mode_1_0, MLX5_MKC_ACCESS_MODE_MTT);
280
281 MLX5_SET(mkc, mkc, qpn, 0xffffff);
282 MLX5_SET(mkc, mkc, pd, mdev->mlx5e_res.pdn);
283 MLX5_SET64(mkc, mkc, len, npages << page_shift);
284 MLX5_SET(mkc, mkc, translations_octword_size,
285 MLX5_MTT_OCTW(npages));
286 MLX5_SET(mkc, mkc, log_page_size, page_shift);
287
288 err = mlx5_core_create_mkey(mdev, umr_mkey, in, inlen);
289
290 kvfree(in);
291 return err;
292 }
293
294 static int mlx5e_create_rq_umr_mkey(struct mlx5_core_dev *mdev, struct mlx5e_rq *rq)
295 {
296 u64 num_mtts = MLX5E_REQUIRED_MTTS(mlx5_wq_ll_get_size(&rq->mpwqe.wq));
297
298 return mlx5e_create_umr_mkey(mdev, num_mtts, PAGE_SHIFT, &rq->umr_mkey);
299 }
300
301 static inline u64 mlx5e_get_mpwqe_offset(struct mlx5e_rq *rq, u16 wqe_ix)
302 {
303 return (wqe_ix << MLX5E_LOG_ALIGNED_MPWQE_PPW) << PAGE_SHIFT;
304 }
305
306 static void mlx5e_init_frags_partition(struct mlx5e_rq *rq)
307 {
308 struct mlx5e_wqe_frag_info next_frag = {};
309 struct mlx5e_wqe_frag_info *prev = NULL;
310 int i;
311
312 next_frag.di = &rq->wqe.di[0];
313
314 for (i = 0; i < mlx5_wq_cyc_get_size(&rq->wqe.wq); i++) {
315 struct mlx5e_rq_frag_info *frag_info = &rq->wqe.info.arr[0];
316 struct mlx5e_wqe_frag_info *frag =
317 &rq->wqe.frags[i << rq->wqe.info.log_num_frags];
318 int f;
319
320 for (f = 0; f < rq->wqe.info.num_frags; f++, frag++) {
321 if (next_frag.offset + frag_info[f].frag_stride > PAGE_SIZE) {
322 next_frag.di++;
323 next_frag.offset = 0;
324 if (prev)
325 prev->last_in_page = true;
326 }
327 *frag = next_frag;
328
329 /* prepare next */
330 next_frag.offset += frag_info[f].frag_stride;
331 prev = frag;
332 }
333 }
334
335 if (prev)
336 prev->last_in_page = true;
337 }
338
339 static int mlx5e_init_di_list(struct mlx5e_rq *rq,
340 int wq_sz, int cpu)
341 {
342 int len = wq_sz << rq->wqe.info.log_num_frags;
343
344 rq->wqe.di = kvzalloc_node(array_size(len, sizeof(*rq->wqe.di)),
345 GFP_KERNEL, cpu_to_node(cpu));
346 if (!rq->wqe.di)
347 return -ENOMEM;
348
349 mlx5e_init_frags_partition(rq);
350
351 return 0;
352 }
353
354 static void mlx5e_free_di_list(struct mlx5e_rq *rq)
355 {
356 kvfree(rq->wqe.di);
357 }
358
359 static void mlx5e_rq_err_cqe_work(struct work_struct *recover_work)
360 {
361 struct mlx5e_rq *rq = container_of(recover_work, struct mlx5e_rq, recover_work);
362
363 mlx5e_reporter_rq_cqe_err(rq);
364 }
365
366 static int mlx5e_alloc_rq(struct mlx5e_channel *c,
367 struct mlx5e_params *params,
368 struct mlx5e_xsk_param *xsk,
369 struct xdp_umem *umem,
370 struct mlx5e_rq_param *rqp,
371 struct mlx5e_rq *rq)
372 {
373 struct page_pool_params pp_params = { 0 };
374 struct mlx5_core_dev *mdev = c->mdev;
375 void *rqc = rqp->rqc;
376 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
377 u32 num_xsk_frames = 0;
378 u32 rq_xdp_ix;
379 u32 pool_size;
380 int wq_sz;
381 int err;
382 int i;
383
384 rqp->wq.db_numa_node = cpu_to_node(c->cpu);
385
386 rq->wq_type = params->rq_wq_type;
387 rq->pdev = c->pdev;
388 rq->netdev = c->netdev;
389 rq->tstamp = c->tstamp;
390 rq->clock = &mdev->clock;
391 rq->channel = c;
392 rq->ix = c->ix;
393 rq->mdev = mdev;
394 rq->hw_mtu = MLX5E_SW2HW_MTU(params, params->sw_mtu);
395 rq->xdpsq = &c->rq_xdpsq;
396 rq->umem = umem;
397
398 if (rq->umem)
399 rq->stats = &c->priv->channel_stats[c->ix].xskrq;
400 else
401 rq->stats = &c->priv->channel_stats[c->ix].rq;
402 INIT_WORK(&rq->recover_work, mlx5e_rq_err_cqe_work);
403
404 if (params->xdp_prog)
405 bpf_prog_inc(params->xdp_prog);
406 rq->xdp_prog = params->xdp_prog;
407
408 rq_xdp_ix = rq->ix;
409 if (xsk)
410 rq_xdp_ix += params->num_channels * MLX5E_RQ_GROUP_XSK;
411 err = xdp_rxq_info_reg(&rq->xdp_rxq, rq->netdev, rq_xdp_ix);
412 if (err < 0)
413 goto err_rq_wq_destroy;
414
415 rq->buff.map_dir = rq->xdp_prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
416 rq->buff.headroom = mlx5e_get_rq_headroom(mdev, params, xsk);
417 rq->buff.umem_headroom = xsk ? xsk->headroom : 0;
418 pool_size = 1 << params->log_rq_mtu_frames;
419
420 switch (rq->wq_type) {
421 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
422 err = mlx5_wq_ll_create(mdev, &rqp->wq, rqc_wq, &rq->mpwqe.wq,
423 &rq->wq_ctrl);
424 if (err)
425 return err;
426
427 rq->mpwqe.wq.db = &rq->mpwqe.wq.db[MLX5_RCV_DBR];
428
429 wq_sz = mlx5_wq_ll_get_size(&rq->mpwqe.wq);
430
431 if (xsk)
432 num_xsk_frames = wq_sz <<
433 mlx5e_mpwqe_get_log_num_strides(mdev, params, xsk);
434
435 pool_size = MLX5_MPWRQ_PAGES_PER_WQE <<
436 mlx5e_mpwqe_get_log_rq_size(params, xsk);
437
438 rq->post_wqes = mlx5e_post_rx_mpwqes;
439 rq->dealloc_wqe = mlx5e_dealloc_rx_mpwqe;
440
441 rq->handle_rx_cqe = c->priv->profile->rx_handlers.handle_rx_cqe_mpwqe;
442 #ifdef CONFIG_MLX5_EN_IPSEC
443 if (MLX5_IPSEC_DEV(mdev)) {
444 err = -EINVAL;
445 netdev_err(c->netdev, "MPWQE RQ with IPSec offload not supported\n");
446 goto err_rq_wq_destroy;
447 }
448 #endif
449 if (!rq->handle_rx_cqe) {
450 err = -EINVAL;
451 netdev_err(c->netdev, "RX handler of MPWQE RQ is not set, err %d\n", err);
452 goto err_rq_wq_destroy;
453 }
454
455 rq->mpwqe.skb_from_cqe_mpwrq = xsk ?
456 mlx5e_xsk_skb_from_cqe_mpwrq_linear :
457 mlx5e_rx_mpwqe_is_linear_skb(mdev, params, NULL) ?
458 mlx5e_skb_from_cqe_mpwrq_linear :
459 mlx5e_skb_from_cqe_mpwrq_nonlinear;
460
461 rq->mpwqe.log_stride_sz = mlx5e_mpwqe_get_log_stride_size(mdev, params, xsk);
462 rq->mpwqe.num_strides =
463 BIT(mlx5e_mpwqe_get_log_num_strides(mdev, params, xsk));
464
465 err = mlx5e_create_rq_umr_mkey(mdev, rq);
466 if (err)
467 goto err_rq_wq_destroy;
468 rq->mkey_be = cpu_to_be32(rq->umr_mkey.key);
469
470 err = mlx5e_rq_alloc_mpwqe_info(rq, c);
471 if (err)
472 goto err_free;
473 break;
474 default: /* MLX5_WQ_TYPE_CYCLIC */
475 err = mlx5_wq_cyc_create(mdev, &rqp->wq, rqc_wq, &rq->wqe.wq,
476 &rq->wq_ctrl);
477 if (err)
478 return err;
479
480 rq->wqe.wq.db = &rq->wqe.wq.db[MLX5_RCV_DBR];
481
482 wq_sz = mlx5_wq_cyc_get_size(&rq->wqe.wq);
483
484 if (xsk)
485 num_xsk_frames = wq_sz << rq->wqe.info.log_num_frags;
486
487 rq->wqe.info = rqp->frags_info;
488 rq->wqe.frags =
489 kvzalloc_node(array_size(sizeof(*rq->wqe.frags),
490 (wq_sz << rq->wqe.info.log_num_frags)),
491 GFP_KERNEL, cpu_to_node(c->cpu));
492 if (!rq->wqe.frags) {
493 err = -ENOMEM;
494 goto err_free;
495 }
496
497 err = mlx5e_init_di_list(rq, wq_sz, c->cpu);
498 if (err)
499 goto err_free;
500
501 rq->post_wqes = mlx5e_post_rx_wqes;
502 rq->dealloc_wqe = mlx5e_dealloc_rx_wqe;
503
504 #ifdef CONFIG_MLX5_EN_IPSEC
505 if (c->priv->ipsec)
506 rq->handle_rx_cqe = mlx5e_ipsec_handle_rx_cqe;
507 else
508 #endif
509 rq->handle_rx_cqe = c->priv->profile->rx_handlers.handle_rx_cqe;
510 if (!rq->handle_rx_cqe) {
511 err = -EINVAL;
512 netdev_err(c->netdev, "RX handler of RQ is not set, err %d\n", err);
513 goto err_free;
514 }
515
516 rq->wqe.skb_from_cqe = xsk ?
517 mlx5e_xsk_skb_from_cqe_linear :
518 mlx5e_rx_is_linear_skb(params, NULL) ?
519 mlx5e_skb_from_cqe_linear :
520 mlx5e_skb_from_cqe_nonlinear;
521 rq->mkey_be = c->mkey_be;
522 }
523
524 if (xsk) {
525 err = mlx5e_xsk_resize_reuseq(umem, num_xsk_frames);
526 if (unlikely(err)) {
527 mlx5_core_err(mdev, "Unable to allocate the Reuse Ring for %u frames\n",
528 num_xsk_frames);
529 goto err_free;
530 }
531
532 rq->zca.free = mlx5e_xsk_zca_free;
533 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
534 MEM_TYPE_ZERO_COPY,
535 &rq->zca);
536 } else {
537 /* Create a page_pool and register it with rxq */
538 pp_params.order = 0;
539 pp_params.flags = 0; /* No-internal DMA mapping in page_pool */
540 pp_params.pool_size = pool_size;
541 pp_params.nid = cpu_to_node(c->cpu);
542 pp_params.dev = c->pdev;
543 pp_params.dma_dir = rq->buff.map_dir;
544
545 /* page_pool can be used even when there is no rq->xdp_prog,
546 * given page_pool does not handle DMA mapping there is no
547 * required state to clear. And page_pool gracefully handle
548 * elevated refcnt.
549 */
550 rq->page_pool = page_pool_create(&pp_params);
551 if (IS_ERR(rq->page_pool)) {
552 err = PTR_ERR(rq->page_pool);
553 rq->page_pool = NULL;
554 goto err_free;
555 }
556 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
557 MEM_TYPE_PAGE_POOL, rq->page_pool);
558 }
559 if (err)
560 goto err_free;
561
562 for (i = 0; i < wq_sz; i++) {
563 if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
564 struct mlx5e_rx_wqe_ll *wqe =
565 mlx5_wq_ll_get_wqe(&rq->mpwqe.wq, i);
566 u32 byte_count =
567 rq->mpwqe.num_strides << rq->mpwqe.log_stride_sz;
568 u64 dma_offset = mlx5e_get_mpwqe_offset(rq, i);
569
570 wqe->data[0].addr = cpu_to_be64(dma_offset + rq->buff.headroom);
571 wqe->data[0].byte_count = cpu_to_be32(byte_count);
572 wqe->data[0].lkey = rq->mkey_be;
573 } else {
574 struct mlx5e_rx_wqe_cyc *wqe =
575 mlx5_wq_cyc_get_wqe(&rq->wqe.wq, i);
576 int f;
577
578 for (f = 0; f < rq->wqe.info.num_frags; f++) {
579 u32 frag_size = rq->wqe.info.arr[f].frag_size |
580 MLX5_HW_START_PADDING;
581
582 wqe->data[f].byte_count = cpu_to_be32(frag_size);
583 wqe->data[f].lkey = rq->mkey_be;
584 }
585 /* check if num_frags is not a pow of two */
586 if (rq->wqe.info.num_frags < (1 << rq->wqe.info.log_num_frags)) {
587 wqe->data[f].byte_count = 0;
588 wqe->data[f].lkey = cpu_to_be32(MLX5_INVALID_LKEY);
589 wqe->data[f].addr = 0;
590 }
591 }
592 }
593
594 INIT_WORK(&rq->dim.work, mlx5e_rx_dim_work);
595
596 switch (params->rx_cq_moderation.cq_period_mode) {
597 case MLX5_CQ_PERIOD_MODE_START_FROM_CQE:
598 rq->dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_CQE;
599 break;
600 case MLX5_CQ_PERIOD_MODE_START_FROM_EQE:
601 default:
602 rq->dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
603 }
604
605 rq->page_cache.head = 0;
606 rq->page_cache.tail = 0;
607
608 return 0;
609
610 err_free:
611 switch (rq->wq_type) {
612 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
613 kvfree(rq->mpwqe.info);
614 mlx5_core_destroy_mkey(mdev, &rq->umr_mkey);
615 break;
616 default: /* MLX5_WQ_TYPE_CYCLIC */
617 kvfree(rq->wqe.frags);
618 mlx5e_free_di_list(rq);
619 }
620
621 err_rq_wq_destroy:
622 if (rq->xdp_prog)
623 bpf_prog_put(rq->xdp_prog);
624 xdp_rxq_info_unreg(&rq->xdp_rxq);
625 page_pool_destroy(rq->page_pool);
626 mlx5_wq_destroy(&rq->wq_ctrl);
627
628 return err;
629 }
630
631 static void mlx5e_free_rq(struct mlx5e_rq *rq)
632 {
633 int i;
634
635 if (rq->xdp_prog)
636 bpf_prog_put(rq->xdp_prog);
637
638 switch (rq->wq_type) {
639 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
640 kvfree(rq->mpwqe.info);
641 mlx5_core_destroy_mkey(rq->mdev, &rq->umr_mkey);
642 break;
643 default: /* MLX5_WQ_TYPE_CYCLIC */
644 kvfree(rq->wqe.frags);
645 mlx5e_free_di_list(rq);
646 }
647
648 for (i = rq->page_cache.head; i != rq->page_cache.tail;
649 i = (i + 1) & (MLX5E_CACHE_SIZE - 1)) {
650 struct mlx5e_dma_info *dma_info = &rq->page_cache.page_cache[i];
651
652 /* With AF_XDP, page_cache is not used, so this loop is not
653 * entered, and it's safe to call mlx5e_page_release_dynamic
654 * directly.
655 */
656 mlx5e_page_release_dynamic(rq, dma_info, false);
657 }
658
659 xdp_rxq_info_unreg(&rq->xdp_rxq);
660 page_pool_destroy(rq->page_pool);
661 mlx5_wq_destroy(&rq->wq_ctrl);
662 }
663
664 static int mlx5e_create_rq(struct mlx5e_rq *rq,
665 struct mlx5e_rq_param *param)
666 {
667 struct mlx5_core_dev *mdev = rq->mdev;
668
669 void *in;
670 void *rqc;
671 void *wq;
672 int inlen;
673 int err;
674
675 inlen = MLX5_ST_SZ_BYTES(create_rq_in) +
676 sizeof(u64) * rq->wq_ctrl.buf.npages;
677 in = kvzalloc(inlen, GFP_KERNEL);
678 if (!in)
679 return -ENOMEM;
680
681 rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
682 wq = MLX5_ADDR_OF(rqc, rqc, wq);
683
684 memcpy(rqc, param->rqc, sizeof(param->rqc));
685
686 MLX5_SET(rqc, rqc, cqn, rq->cq.mcq.cqn);
687 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST);
688 MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift -
689 MLX5_ADAPTER_PAGE_SHIFT);
690 MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma);
691
692 mlx5_fill_page_frag_array(&rq->wq_ctrl.buf,
693 (__be64 *)MLX5_ADDR_OF(wq, wq, pas));
694
695 err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn);
696
697 kvfree(in);
698
699 return err;
700 }
701
702 int mlx5e_modify_rq_state(struct mlx5e_rq *rq, int curr_state, int next_state)
703 {
704 struct mlx5_core_dev *mdev = rq->mdev;
705
706 void *in;
707 void *rqc;
708 int inlen;
709 int err;
710
711 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
712 in = kvzalloc(inlen, GFP_KERNEL);
713 if (!in)
714 return -ENOMEM;
715
716 if (curr_state == MLX5_RQC_STATE_RST && next_state == MLX5_RQC_STATE_RDY)
717 mlx5e_rqwq_reset(rq);
718
719 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
720
721 MLX5_SET(modify_rq_in, in, rq_state, curr_state);
722 MLX5_SET(rqc, rqc, state, next_state);
723
724 err = mlx5_core_modify_rq(mdev, rq->rqn, in, inlen);
725
726 kvfree(in);
727
728 return err;
729 }
730
731 static int mlx5e_modify_rq_scatter_fcs(struct mlx5e_rq *rq, bool enable)
732 {
733 struct mlx5e_channel *c = rq->channel;
734 struct mlx5e_priv *priv = c->priv;
735 struct mlx5_core_dev *mdev = priv->mdev;
736
737 void *in;
738 void *rqc;
739 int inlen;
740 int err;
741
742 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
743 in = kvzalloc(inlen, GFP_KERNEL);
744 if (!in)
745 return -ENOMEM;
746
747 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
748
749 MLX5_SET(modify_rq_in, in, rq_state, MLX5_RQC_STATE_RDY);
750 MLX5_SET64(modify_rq_in, in, modify_bitmask,
751 MLX5_MODIFY_RQ_IN_MODIFY_BITMASK_SCATTER_FCS);
752 MLX5_SET(rqc, rqc, scatter_fcs, enable);
753 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RDY);
754
755 err = mlx5_core_modify_rq(mdev, rq->rqn, in, inlen);
756
757 kvfree(in);
758
759 return err;
760 }
761
762 static int mlx5e_modify_rq_vsd(struct mlx5e_rq *rq, bool vsd)
763 {
764 struct mlx5e_channel *c = rq->channel;
765 struct mlx5_core_dev *mdev = c->mdev;
766 void *in;
767 void *rqc;
768 int inlen;
769 int err;
770
771 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
772 in = kvzalloc(inlen, GFP_KERNEL);
773 if (!in)
774 return -ENOMEM;
775
776 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
777
778 MLX5_SET(modify_rq_in, in, rq_state, MLX5_RQC_STATE_RDY);
779 MLX5_SET64(modify_rq_in, in, modify_bitmask,
780 MLX5_MODIFY_RQ_IN_MODIFY_BITMASK_VSD);
781 MLX5_SET(rqc, rqc, vsd, vsd);
782 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RDY);
783
784 err = mlx5_core_modify_rq(mdev, rq->rqn, in, inlen);
785
786 kvfree(in);
787
788 return err;
789 }
790
791 static void mlx5e_destroy_rq(struct mlx5e_rq *rq)
792 {
793 mlx5_core_destroy_rq(rq->mdev, rq->rqn);
794 }
795
796 int mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq, int wait_time)
797 {
798 unsigned long exp_time = jiffies + msecs_to_jiffies(wait_time);
799 struct mlx5e_channel *c = rq->channel;
800
801 u16 min_wqes = mlx5_min_rx_wqes(rq->wq_type, mlx5e_rqwq_get_size(rq));
802
803 do {
804 if (mlx5e_rqwq_get_cur_sz(rq) >= min_wqes)
805 return 0;
806
807 msleep(20);
808 } while (time_before(jiffies, exp_time));
809
810 netdev_warn(c->netdev, "Failed to get min RX wqes on Channel[%d] RQN[0x%x] wq cur_sz(%d) min_rx_wqes(%d)\n",
811 c->ix, rq->rqn, mlx5e_rqwq_get_cur_sz(rq), min_wqes);
812
813 mlx5e_reporter_rx_timeout(rq);
814 return -ETIMEDOUT;
815 }
816
817 void mlx5e_free_rx_in_progress_descs(struct mlx5e_rq *rq)
818 {
819 struct mlx5_wq_ll *wq;
820 u16 head;
821 int i;
822
823 if (rq->wq_type != MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ)
824 return;
825
826 wq = &rq->mpwqe.wq;
827 head = wq->head;
828
829 /* Outstanding UMR WQEs (in progress) start at wq->head */
830 for (i = 0; i < rq->mpwqe.umr_in_progress; i++) {
831 rq->dealloc_wqe(rq, head);
832 head = mlx5_wq_ll_get_wqe_next_ix(wq, head);
833 }
834
835 rq->mpwqe.actual_wq_head = wq->head;
836 rq->mpwqe.umr_in_progress = 0;
837 rq->mpwqe.umr_completed = 0;
838 }
839
840 void mlx5e_free_rx_descs(struct mlx5e_rq *rq)
841 {
842 __be16 wqe_ix_be;
843 u16 wqe_ix;
844
845 if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
846 struct mlx5_wq_ll *wq = &rq->mpwqe.wq;
847
848 mlx5e_free_rx_in_progress_descs(rq);
849
850 while (!mlx5_wq_ll_is_empty(wq)) {
851 struct mlx5e_rx_wqe_ll *wqe;
852
853 wqe_ix_be = *wq->tail_next;
854 wqe_ix = be16_to_cpu(wqe_ix_be);
855 wqe = mlx5_wq_ll_get_wqe(wq, wqe_ix);
856 rq->dealloc_wqe(rq, wqe_ix);
857 mlx5_wq_ll_pop(wq, wqe_ix_be,
858 &wqe->next.next_wqe_index);
859 }
860 } else {
861 struct mlx5_wq_cyc *wq = &rq->wqe.wq;
862
863 while (!mlx5_wq_cyc_is_empty(wq)) {
864 wqe_ix = mlx5_wq_cyc_get_tail(wq);
865 rq->dealloc_wqe(rq, wqe_ix);
866 mlx5_wq_cyc_pop(wq);
867 }
868 }
869
870 }
871
872 int mlx5e_open_rq(struct mlx5e_channel *c, struct mlx5e_params *params,
873 struct mlx5e_rq_param *param, struct mlx5e_xsk_param *xsk,
874 struct xdp_umem *umem, struct mlx5e_rq *rq)
875 {
876 int err;
877
878 err = mlx5e_alloc_rq(c, params, xsk, umem, param, rq);
879 if (err)
880 return err;
881
882 err = mlx5e_create_rq(rq, param);
883 if (err)
884 goto err_free_rq;
885
886 err = mlx5e_modify_rq_state(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
887 if (err)
888 goto err_destroy_rq;
889
890 if (MLX5_CAP_ETH(c->mdev, cqe_checksum_full))
891 __set_bit(MLX5E_RQ_STATE_CSUM_FULL, &c->rq.state);
892
893 if (params->rx_dim_enabled)
894 __set_bit(MLX5E_RQ_STATE_AM, &c->rq.state);
895
896 /* We disable csum_complete when XDP is enabled since
897 * XDP programs might manipulate packets which will render
898 * skb->checksum incorrect.
899 */
900 if (MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_NO_CSUM_COMPLETE) || c->xdp)
901 __set_bit(MLX5E_RQ_STATE_NO_CSUM_COMPLETE, &c->rq.state);
902
903 return 0;
904
905 err_destroy_rq:
906 mlx5e_destroy_rq(rq);
907 err_free_rq:
908 mlx5e_free_rq(rq);
909
910 return err;
911 }
912
913 void mlx5e_activate_rq(struct mlx5e_rq *rq)
914 {
915 set_bit(MLX5E_RQ_STATE_ENABLED, &rq->state);
916 mlx5e_trigger_irq(&rq->channel->icosq);
917 }
918
919 void mlx5e_deactivate_rq(struct mlx5e_rq *rq)
920 {
921 clear_bit(MLX5E_RQ_STATE_ENABLED, &rq->state);
922 napi_synchronize(&rq->channel->napi); /* prevent mlx5e_post_rx_wqes */
923 }
924
925 void mlx5e_close_rq(struct mlx5e_rq *rq)
926 {
927 cancel_work_sync(&rq->dim.work);
928 cancel_work_sync(&rq->channel->icosq.recover_work);
929 cancel_work_sync(&rq->recover_work);
930 mlx5e_destroy_rq(rq);
931 mlx5e_free_rx_descs(rq);
932 mlx5e_free_rq(rq);
933 }
934
935 static void mlx5e_free_xdpsq_db(struct mlx5e_xdpsq *sq)
936 {
937 kvfree(sq->db.xdpi_fifo.xi);
938 kvfree(sq->db.wqe_info);
939 }
940
941 static int mlx5e_alloc_xdpsq_fifo(struct mlx5e_xdpsq *sq, int numa)
942 {
943 struct mlx5e_xdp_info_fifo *xdpi_fifo = &sq->db.xdpi_fifo;
944 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
945 int dsegs_per_wq = wq_sz * MLX5_SEND_WQEBB_NUM_DS;
946
947 xdpi_fifo->xi = kvzalloc_node(sizeof(*xdpi_fifo->xi) * dsegs_per_wq,
948 GFP_KERNEL, numa);
949 if (!xdpi_fifo->xi)
950 return -ENOMEM;
951
952 xdpi_fifo->pc = &sq->xdpi_fifo_pc;
953 xdpi_fifo->cc = &sq->xdpi_fifo_cc;
954 xdpi_fifo->mask = dsegs_per_wq - 1;
955
956 return 0;
957 }
958
959 static int mlx5e_alloc_xdpsq_db(struct mlx5e_xdpsq *sq, int numa)
960 {
961 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
962 int err;
963
964 sq->db.wqe_info = kvzalloc_node(sizeof(*sq->db.wqe_info) * wq_sz,
965 GFP_KERNEL, numa);
966 if (!sq->db.wqe_info)
967 return -ENOMEM;
968
969 err = mlx5e_alloc_xdpsq_fifo(sq, numa);
970 if (err) {
971 mlx5e_free_xdpsq_db(sq);
972 return err;
973 }
974
975 return 0;
976 }
977
978 static int mlx5e_alloc_xdpsq(struct mlx5e_channel *c,
979 struct mlx5e_params *params,
980 struct xdp_umem *umem,
981 struct mlx5e_sq_param *param,
982 struct mlx5e_xdpsq *sq,
983 bool is_redirect)
984 {
985 void *sqc_wq = MLX5_ADDR_OF(sqc, param->sqc, wq);
986 struct mlx5_core_dev *mdev = c->mdev;
987 struct mlx5_wq_cyc *wq = &sq->wq;
988 int err;
989
990 sq->pdev = c->pdev;
991 sq->mkey_be = c->mkey_be;
992 sq->channel = c;
993 sq->uar_map = mdev->mlx5e_res.bfreg.map;
994 sq->min_inline_mode = params->tx_min_inline_mode;
995 sq->hw_mtu = MLX5E_SW2HW_MTU(params, params->sw_mtu);
996 sq->umem = umem;
997
998 sq->stats = sq->umem ?
999 &c->priv->channel_stats[c->ix].xsksq :
1000 is_redirect ?
1001 &c->priv->channel_stats[c->ix].xdpsq :
1002 &c->priv->channel_stats[c->ix].rq_xdpsq;
1003
1004 param->wq.db_numa_node = cpu_to_node(c->cpu);
1005 err = mlx5_wq_cyc_create(mdev, &param->wq, sqc_wq, wq, &sq->wq_ctrl);
1006 if (err)
1007 return err;
1008 wq->db = &wq->db[MLX5_SND_DBR];
1009
1010 err = mlx5e_alloc_xdpsq_db(sq, cpu_to_node(c->cpu));
1011 if (err)
1012 goto err_sq_wq_destroy;
1013
1014 return 0;
1015
1016 err_sq_wq_destroy:
1017 mlx5_wq_destroy(&sq->wq_ctrl);
1018
1019 return err;
1020 }
1021
1022 static void mlx5e_free_xdpsq(struct mlx5e_xdpsq *sq)
1023 {
1024 mlx5e_free_xdpsq_db(sq);
1025 mlx5_wq_destroy(&sq->wq_ctrl);
1026 }
1027
1028 static void mlx5e_free_icosq_db(struct mlx5e_icosq *sq)
1029 {
1030 kvfree(sq->db.ico_wqe);
1031 }
1032
1033 static int mlx5e_alloc_icosq_db(struct mlx5e_icosq *sq, int numa)
1034 {
1035 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1036
1037 sq->db.ico_wqe = kvzalloc_node(array_size(wq_sz,
1038 sizeof(*sq->db.ico_wqe)),
1039 GFP_KERNEL, numa);
1040 if (!sq->db.ico_wqe)
1041 return -ENOMEM;
1042
1043 return 0;
1044 }
1045
1046 static void mlx5e_icosq_err_cqe_work(struct work_struct *recover_work)
1047 {
1048 struct mlx5e_icosq *sq = container_of(recover_work, struct mlx5e_icosq,
1049 recover_work);
1050
1051 mlx5e_reporter_icosq_cqe_err(sq);
1052 }
1053
1054 static int mlx5e_alloc_icosq(struct mlx5e_channel *c,
1055 struct mlx5e_sq_param *param,
1056 struct mlx5e_icosq *sq)
1057 {
1058 void *sqc_wq = MLX5_ADDR_OF(sqc, param->sqc, wq);
1059 struct mlx5_core_dev *mdev = c->mdev;
1060 struct mlx5_wq_cyc *wq = &sq->wq;
1061 int err;
1062
1063 sq->channel = c;
1064 sq->uar_map = mdev->mlx5e_res.bfreg.map;
1065
1066 param->wq.db_numa_node = cpu_to_node(c->cpu);
1067 err = mlx5_wq_cyc_create(mdev, &param->wq, sqc_wq, wq, &sq->wq_ctrl);
1068 if (err)
1069 return err;
1070 wq->db = &wq->db[MLX5_SND_DBR];
1071
1072 err = mlx5e_alloc_icosq_db(sq, cpu_to_node(c->cpu));
1073 if (err)
1074 goto err_sq_wq_destroy;
1075
1076 INIT_WORK(&sq->recover_work, mlx5e_icosq_err_cqe_work);
1077
1078 return 0;
1079
1080 err_sq_wq_destroy:
1081 mlx5_wq_destroy(&sq->wq_ctrl);
1082
1083 return err;
1084 }
1085
1086 static void mlx5e_free_icosq(struct mlx5e_icosq *sq)
1087 {
1088 mlx5e_free_icosq_db(sq);
1089 mlx5_wq_destroy(&sq->wq_ctrl);
1090 }
1091
1092 static void mlx5e_free_txqsq_db(struct mlx5e_txqsq *sq)
1093 {
1094 kvfree(sq->db.wqe_info);
1095 kvfree(sq->db.dma_fifo);
1096 }
1097
1098 static int mlx5e_alloc_txqsq_db(struct mlx5e_txqsq *sq, int numa)
1099 {
1100 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1101 int df_sz = wq_sz * MLX5_SEND_WQEBB_NUM_DS;
1102
1103 sq->db.dma_fifo = kvzalloc_node(array_size(df_sz,
1104 sizeof(*sq->db.dma_fifo)),
1105 GFP_KERNEL, numa);
1106 sq->db.wqe_info = kvzalloc_node(array_size(wq_sz,
1107 sizeof(*sq->db.wqe_info)),
1108 GFP_KERNEL, numa);
1109 if (!sq->db.dma_fifo || !sq->db.wqe_info) {
1110 mlx5e_free_txqsq_db(sq);
1111 return -ENOMEM;
1112 }
1113
1114 sq->dma_fifo_mask = df_sz - 1;
1115
1116 return 0;
1117 }
1118
1119 static void mlx5e_tx_err_cqe_work(struct work_struct *recover_work);
1120 static int mlx5e_alloc_txqsq(struct mlx5e_channel *c,
1121 int txq_ix,
1122 struct mlx5e_params *params,
1123 struct mlx5e_sq_param *param,
1124 struct mlx5e_txqsq *sq,
1125 int tc)
1126 {
1127 void *sqc_wq = MLX5_ADDR_OF(sqc, param->sqc, wq);
1128 struct mlx5_core_dev *mdev = c->mdev;
1129 struct mlx5_wq_cyc *wq = &sq->wq;
1130 int err;
1131
1132 sq->pdev = c->pdev;
1133 sq->tstamp = c->tstamp;
1134 sq->clock = &mdev->clock;
1135 sq->mkey_be = c->mkey_be;
1136 sq->channel = c;
1137 sq->ch_ix = c->ix;
1138 sq->txq_ix = txq_ix;
1139 sq->uar_map = mdev->mlx5e_res.bfreg.map;
1140 sq->min_inline_mode = params->tx_min_inline_mode;
1141 sq->hw_mtu = MLX5E_SW2HW_MTU(params, params->sw_mtu);
1142 sq->stats = &c->priv->channel_stats[c->ix].sq[tc];
1143 sq->stop_room = MLX5E_SQ_STOP_ROOM;
1144 INIT_WORK(&sq->recover_work, mlx5e_tx_err_cqe_work);
1145 if (!MLX5_CAP_ETH(mdev, wqe_vlan_insert))
1146 set_bit(MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE, &sq->state);
1147 if (MLX5_IPSEC_DEV(c->priv->mdev))
1148 set_bit(MLX5E_SQ_STATE_IPSEC, &sq->state);
1149 #ifdef CONFIG_MLX5_EN_TLS
1150 if (mlx5_accel_is_tls_device(c->priv->mdev)) {
1151 set_bit(MLX5E_SQ_STATE_TLS, &sq->state);
1152 sq->stop_room += MLX5E_SQ_TLS_ROOM +
1153 mlx5e_ktls_dumps_num_wqebbs(sq, MAX_SKB_FRAGS,
1154 TLS_MAX_PAYLOAD_SIZE);
1155 }
1156 #endif
1157
1158 param->wq.db_numa_node = cpu_to_node(c->cpu);
1159 err = mlx5_wq_cyc_create(mdev, &param->wq, sqc_wq, wq, &sq->wq_ctrl);
1160 if (err)
1161 return err;
1162 wq->db = &wq->db[MLX5_SND_DBR];
1163
1164 err = mlx5e_alloc_txqsq_db(sq, cpu_to_node(c->cpu));
1165 if (err)
1166 goto err_sq_wq_destroy;
1167
1168 INIT_WORK(&sq->dim.work, mlx5e_tx_dim_work);
1169 sq->dim.mode = params->tx_cq_moderation.cq_period_mode;
1170
1171 return 0;
1172
1173 err_sq_wq_destroy:
1174 mlx5_wq_destroy(&sq->wq_ctrl);
1175
1176 return err;
1177 }
1178
1179 static void mlx5e_free_txqsq(struct mlx5e_txqsq *sq)
1180 {
1181 mlx5e_free_txqsq_db(sq);
1182 mlx5_wq_destroy(&sq->wq_ctrl);
1183 }
1184
1185 struct mlx5e_create_sq_param {
1186 struct mlx5_wq_ctrl *wq_ctrl;
1187 u32 cqn;
1188 u32 tisn;
1189 u8 tis_lst_sz;
1190 u8 min_inline_mode;
1191 };
1192
1193 static int mlx5e_create_sq(struct mlx5_core_dev *mdev,
1194 struct mlx5e_sq_param *param,
1195 struct mlx5e_create_sq_param *csp,
1196 u32 *sqn)
1197 {
1198 void *in;
1199 void *sqc;
1200 void *wq;
1201 int inlen;
1202 int err;
1203
1204 inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
1205 sizeof(u64) * csp->wq_ctrl->buf.npages;
1206 in = kvzalloc(inlen, GFP_KERNEL);
1207 if (!in)
1208 return -ENOMEM;
1209
1210 sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
1211 wq = MLX5_ADDR_OF(sqc, sqc, wq);
1212
1213 memcpy(sqc, param->sqc, sizeof(param->sqc));
1214 MLX5_SET(sqc, sqc, tis_lst_sz, csp->tis_lst_sz);
1215 MLX5_SET(sqc, sqc, tis_num_0, csp->tisn);
1216 MLX5_SET(sqc, sqc, cqn, csp->cqn);
1217
1218 if (MLX5_CAP_ETH(mdev, wqe_inline_mode) == MLX5_CAP_INLINE_MODE_VPORT_CONTEXT)
1219 MLX5_SET(sqc, sqc, min_wqe_inline_mode, csp->min_inline_mode);
1220
1221 MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
1222 MLX5_SET(sqc, sqc, flush_in_error_en, 1);
1223
1224 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
1225 MLX5_SET(wq, wq, uar_page, mdev->mlx5e_res.bfreg.index);
1226 MLX5_SET(wq, wq, log_wq_pg_sz, csp->wq_ctrl->buf.page_shift -
1227 MLX5_ADAPTER_PAGE_SHIFT);
1228 MLX5_SET64(wq, wq, dbr_addr, csp->wq_ctrl->db.dma);
1229
1230 mlx5_fill_page_frag_array(&csp->wq_ctrl->buf,
1231 (__be64 *)MLX5_ADDR_OF(wq, wq, pas));
1232
1233 err = mlx5_core_create_sq(mdev, in, inlen, sqn);
1234
1235 kvfree(in);
1236
1237 return err;
1238 }
1239
1240 int mlx5e_modify_sq(struct mlx5_core_dev *mdev, u32 sqn,
1241 struct mlx5e_modify_sq_param *p)
1242 {
1243 void *in;
1244 void *sqc;
1245 int inlen;
1246 int err;
1247
1248 inlen = MLX5_ST_SZ_BYTES(modify_sq_in);
1249 in = kvzalloc(inlen, GFP_KERNEL);
1250 if (!in)
1251 return -ENOMEM;
1252
1253 sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx);
1254
1255 MLX5_SET(modify_sq_in, in, sq_state, p->curr_state);
1256 MLX5_SET(sqc, sqc, state, p->next_state);
1257 if (p->rl_update && p->next_state == MLX5_SQC_STATE_RDY) {
1258 MLX5_SET64(modify_sq_in, in, modify_bitmask, 1);
1259 MLX5_SET(sqc, sqc, packet_pacing_rate_limit_index, p->rl_index);
1260 }
1261
1262 err = mlx5_core_modify_sq(mdev, sqn, in, inlen);
1263
1264 kvfree(in);
1265
1266 return err;
1267 }
1268
1269 static void mlx5e_destroy_sq(struct mlx5_core_dev *mdev, u32 sqn)
1270 {
1271 mlx5_core_destroy_sq(mdev, sqn);
1272 }
1273
1274 static int mlx5e_create_sq_rdy(struct mlx5_core_dev *mdev,
1275 struct mlx5e_sq_param *param,
1276 struct mlx5e_create_sq_param *csp,
1277 u32 *sqn)
1278 {
1279 struct mlx5e_modify_sq_param msp = {0};
1280 int err;
1281
1282 err = mlx5e_create_sq(mdev, param, csp, sqn);
1283 if (err)
1284 return err;
1285
1286 msp.curr_state = MLX5_SQC_STATE_RST;
1287 msp.next_state = MLX5_SQC_STATE_RDY;
1288 err = mlx5e_modify_sq(mdev, *sqn, &msp);
1289 if (err)
1290 mlx5e_destroy_sq(mdev, *sqn);
1291
1292 return err;
1293 }
1294
1295 static int mlx5e_set_sq_maxrate(struct net_device *dev,
1296 struct mlx5e_txqsq *sq, u32 rate);
1297
1298 static int mlx5e_open_txqsq(struct mlx5e_channel *c,
1299 u32 tisn,
1300 int txq_ix,
1301 struct mlx5e_params *params,
1302 struct mlx5e_sq_param *param,
1303 struct mlx5e_txqsq *sq,
1304 int tc)
1305 {
1306 struct mlx5e_create_sq_param csp = {};
1307 u32 tx_rate;
1308 int err;
1309
1310 err = mlx5e_alloc_txqsq(c, txq_ix, params, param, sq, tc);
1311 if (err)
1312 return err;
1313
1314 csp.tisn = tisn;
1315 csp.tis_lst_sz = 1;
1316 csp.cqn = sq->cq.mcq.cqn;
1317 csp.wq_ctrl = &sq->wq_ctrl;
1318 csp.min_inline_mode = sq->min_inline_mode;
1319 err = mlx5e_create_sq_rdy(c->mdev, param, &csp, &sq->sqn);
1320 if (err)
1321 goto err_free_txqsq;
1322
1323 tx_rate = c->priv->tx_rates[sq->txq_ix];
1324 if (tx_rate)
1325 mlx5e_set_sq_maxrate(c->netdev, sq, tx_rate);
1326
1327 if (params->tx_dim_enabled)
1328 sq->state |= BIT(MLX5E_SQ_STATE_AM);
1329
1330 return 0;
1331
1332 err_free_txqsq:
1333 mlx5e_free_txqsq(sq);
1334
1335 return err;
1336 }
1337
1338 void mlx5e_activate_txqsq(struct mlx5e_txqsq *sq)
1339 {
1340 sq->txq = netdev_get_tx_queue(sq->channel->netdev, sq->txq_ix);
1341 set_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1342 netdev_tx_reset_queue(sq->txq);
1343 netif_tx_start_queue(sq->txq);
1344 }
1345
1346 void mlx5e_tx_disable_queue(struct netdev_queue *txq)
1347 {
1348 __netif_tx_lock_bh(txq);
1349 netif_tx_stop_queue(txq);
1350 __netif_tx_unlock_bh(txq);
1351 }
1352
1353 static void mlx5e_deactivate_txqsq(struct mlx5e_txqsq *sq)
1354 {
1355 struct mlx5e_channel *c = sq->channel;
1356 struct mlx5_wq_cyc *wq = &sq->wq;
1357
1358 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1359 /* prevent netif_tx_wake_queue */
1360 napi_synchronize(&c->napi);
1361
1362 mlx5e_tx_disable_queue(sq->txq);
1363
1364 /* last doorbell out, godspeed .. */
1365 if (mlx5e_wqc_has_room_for(wq, sq->cc, sq->pc, 1)) {
1366 u16 pi = mlx5_wq_cyc_ctr2ix(wq, sq->pc);
1367 struct mlx5e_tx_wqe_info *wi;
1368 struct mlx5e_tx_wqe *nop;
1369
1370 wi = &sq->db.wqe_info[pi];
1371
1372 memset(wi, 0, sizeof(*wi));
1373 wi->num_wqebbs = 1;
1374 nop = mlx5e_post_nop(wq, sq->sqn, &sq->pc);
1375 mlx5e_notify_hw(wq, sq->pc, sq->uar_map, &nop->ctrl);
1376 }
1377 }
1378
1379 static void mlx5e_close_txqsq(struct mlx5e_txqsq *sq)
1380 {
1381 struct mlx5e_channel *c = sq->channel;
1382 struct mlx5_core_dev *mdev = c->mdev;
1383 struct mlx5_rate_limit rl = {0};
1384
1385 cancel_work_sync(&sq->dim.work);
1386 cancel_work_sync(&sq->recover_work);
1387 mlx5e_destroy_sq(mdev, sq->sqn);
1388 if (sq->rate_limit) {
1389 rl.rate = sq->rate_limit;
1390 mlx5_rl_remove_rate(mdev, &rl);
1391 }
1392 mlx5e_free_txqsq_descs(sq);
1393 mlx5e_free_txqsq(sq);
1394 }
1395
1396 static void mlx5e_tx_err_cqe_work(struct work_struct *recover_work)
1397 {
1398 struct mlx5e_txqsq *sq = container_of(recover_work, struct mlx5e_txqsq,
1399 recover_work);
1400
1401 mlx5e_reporter_tx_err_cqe(sq);
1402 }
1403
1404 int mlx5e_open_icosq(struct mlx5e_channel *c, struct mlx5e_params *params,
1405 struct mlx5e_sq_param *param, struct mlx5e_icosq *sq)
1406 {
1407 struct mlx5e_create_sq_param csp = {};
1408 int err;
1409
1410 err = mlx5e_alloc_icosq(c, param, sq);
1411 if (err)
1412 return err;
1413
1414 csp.cqn = sq->cq.mcq.cqn;
1415 csp.wq_ctrl = &sq->wq_ctrl;
1416 csp.min_inline_mode = params->tx_min_inline_mode;
1417 err = mlx5e_create_sq_rdy(c->mdev, param, &csp, &sq->sqn);
1418 if (err)
1419 goto err_free_icosq;
1420
1421 return 0;
1422
1423 err_free_icosq:
1424 mlx5e_free_icosq(sq);
1425
1426 return err;
1427 }
1428
1429 void mlx5e_activate_icosq(struct mlx5e_icosq *icosq)
1430 {
1431 set_bit(MLX5E_SQ_STATE_ENABLED, &icosq->state);
1432 }
1433
1434 void mlx5e_deactivate_icosq(struct mlx5e_icosq *icosq)
1435 {
1436 struct mlx5e_channel *c = icosq->channel;
1437
1438 clear_bit(MLX5E_SQ_STATE_ENABLED, &icosq->state);
1439 napi_synchronize(&c->napi);
1440 }
1441
1442 void mlx5e_close_icosq(struct mlx5e_icosq *sq)
1443 {
1444 struct mlx5e_channel *c = sq->channel;
1445
1446 mlx5e_destroy_sq(c->mdev, sq->sqn);
1447 mlx5e_free_icosq(sq);
1448 }
1449
1450 int mlx5e_open_xdpsq(struct mlx5e_channel *c, struct mlx5e_params *params,
1451 struct mlx5e_sq_param *param, struct xdp_umem *umem,
1452 struct mlx5e_xdpsq *sq, bool is_redirect)
1453 {
1454 struct mlx5e_create_sq_param csp = {};
1455 int err;
1456
1457 err = mlx5e_alloc_xdpsq(c, params, umem, param, sq, is_redirect);
1458 if (err)
1459 return err;
1460
1461 csp.tis_lst_sz = 1;
1462 csp.tisn = c->priv->tisn[c->lag_port][0]; /* tc = 0 */
1463 csp.cqn = sq->cq.mcq.cqn;
1464 csp.wq_ctrl = &sq->wq_ctrl;
1465 csp.min_inline_mode = sq->min_inline_mode;
1466 set_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1467 err = mlx5e_create_sq_rdy(c->mdev, param, &csp, &sq->sqn);
1468 if (err)
1469 goto err_free_xdpsq;
1470
1471 mlx5e_set_xmit_fp(sq, param->is_mpw);
1472
1473 if (!param->is_mpw) {
1474 unsigned int ds_cnt = MLX5E_XDP_TX_DS_COUNT;
1475 unsigned int inline_hdr_sz = 0;
1476 int i;
1477
1478 if (sq->min_inline_mode != MLX5_INLINE_MODE_NONE) {
1479 inline_hdr_sz = MLX5E_XDP_MIN_INLINE;
1480 ds_cnt++;
1481 }
1482
1483 /* Pre initialize fixed WQE fields */
1484 for (i = 0; i < mlx5_wq_cyc_get_size(&sq->wq); i++) {
1485 struct mlx5e_xdp_wqe_info *wi = &sq->db.wqe_info[i];
1486 struct mlx5e_tx_wqe *wqe = mlx5_wq_cyc_get_wqe(&sq->wq, i);
1487 struct mlx5_wqe_ctrl_seg *cseg = &wqe->ctrl;
1488 struct mlx5_wqe_eth_seg *eseg = &wqe->eth;
1489 struct mlx5_wqe_data_seg *dseg;
1490
1491 cseg->qpn_ds = cpu_to_be32((sq->sqn << 8) | ds_cnt);
1492 eseg->inline_hdr.sz = cpu_to_be16(inline_hdr_sz);
1493
1494 dseg = (struct mlx5_wqe_data_seg *)cseg + (ds_cnt - 1);
1495 dseg->lkey = sq->mkey_be;
1496
1497 wi->num_wqebbs = 1;
1498 wi->num_pkts = 1;
1499 }
1500 }
1501
1502 return 0;
1503
1504 err_free_xdpsq:
1505 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1506 mlx5e_free_xdpsq(sq);
1507
1508 return err;
1509 }
1510
1511 void mlx5e_close_xdpsq(struct mlx5e_xdpsq *sq)
1512 {
1513 struct mlx5e_channel *c = sq->channel;
1514
1515 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1516 napi_synchronize(&c->napi);
1517
1518 mlx5e_destroy_sq(c->mdev, sq->sqn);
1519 mlx5e_free_xdpsq_descs(sq);
1520 mlx5e_free_xdpsq(sq);
1521 }
1522
1523 static int mlx5e_alloc_cq_common(struct mlx5_core_dev *mdev,
1524 struct mlx5e_cq_param *param,
1525 struct mlx5e_cq *cq)
1526 {
1527 struct mlx5_core_cq *mcq = &cq->mcq;
1528 int eqn_not_used;
1529 unsigned int irqn;
1530 int err;
1531 u32 i;
1532
1533 err = mlx5_vector2eqn(mdev, param->eq_ix, &eqn_not_used, &irqn);
1534 if (err)
1535 return err;
1536
1537 err = mlx5_cqwq_create(mdev, &param->wq, param->cqc, &cq->wq,
1538 &cq->wq_ctrl);
1539 if (err)
1540 return err;
1541
1542 mcq->cqe_sz = 64;
1543 mcq->set_ci_db = cq->wq_ctrl.db.db;
1544 mcq->arm_db = cq->wq_ctrl.db.db + 1;
1545 *mcq->set_ci_db = 0;
1546 *mcq->arm_db = 0;
1547 mcq->vector = param->eq_ix;
1548 mcq->comp = mlx5e_completion_event;
1549 mcq->event = mlx5e_cq_error_event;
1550 mcq->irqn = irqn;
1551
1552 for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) {
1553 struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i);
1554
1555 cqe->op_own = 0xf1;
1556 }
1557
1558 cq->mdev = mdev;
1559
1560 return 0;
1561 }
1562
1563 static int mlx5e_alloc_cq(struct mlx5e_channel *c,
1564 struct mlx5e_cq_param *param,
1565 struct mlx5e_cq *cq)
1566 {
1567 struct mlx5_core_dev *mdev = c->priv->mdev;
1568 int err;
1569
1570 param->wq.buf_numa_node = cpu_to_node(c->cpu);
1571 param->wq.db_numa_node = cpu_to_node(c->cpu);
1572 param->eq_ix = c->ix;
1573
1574 err = mlx5e_alloc_cq_common(mdev, param, cq);
1575
1576 cq->napi = &c->napi;
1577 cq->channel = c;
1578
1579 return err;
1580 }
1581
1582 static void mlx5e_free_cq(struct mlx5e_cq *cq)
1583 {
1584 mlx5_wq_destroy(&cq->wq_ctrl);
1585 }
1586
1587 static int mlx5e_create_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param)
1588 {
1589 u32 out[MLX5_ST_SZ_DW(create_cq_out)];
1590 struct mlx5_core_dev *mdev = cq->mdev;
1591 struct mlx5_core_cq *mcq = &cq->mcq;
1592
1593 void *in;
1594 void *cqc;
1595 int inlen;
1596 unsigned int irqn_not_used;
1597 int eqn;
1598 int err;
1599
1600 err = mlx5_vector2eqn(mdev, param->eq_ix, &eqn, &irqn_not_used);
1601 if (err)
1602 return err;
1603
1604 inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
1605 sizeof(u64) * cq->wq_ctrl.buf.npages;
1606 in = kvzalloc(inlen, GFP_KERNEL);
1607 if (!in)
1608 return -ENOMEM;
1609
1610 cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
1611
1612 memcpy(cqc, param->cqc, sizeof(param->cqc));
1613
1614 mlx5_fill_page_frag_array(&cq->wq_ctrl.buf,
1615 (__be64 *)MLX5_ADDR_OF(create_cq_in, in, pas));
1616
1617 MLX5_SET(cqc, cqc, cq_period_mode, param->cq_period_mode);
1618 MLX5_SET(cqc, cqc, c_eqn, eqn);
1619 MLX5_SET(cqc, cqc, uar_page, mdev->priv.uar->index);
1620 MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.buf.page_shift -
1621 MLX5_ADAPTER_PAGE_SHIFT);
1622 MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
1623
1624 err = mlx5_core_create_cq(mdev, mcq, in, inlen, out, sizeof(out));
1625
1626 kvfree(in);
1627
1628 if (err)
1629 return err;
1630
1631 mlx5e_cq_arm(cq);
1632
1633 return 0;
1634 }
1635
1636 static void mlx5e_destroy_cq(struct mlx5e_cq *cq)
1637 {
1638 mlx5_core_destroy_cq(cq->mdev, &cq->mcq);
1639 }
1640
1641 int mlx5e_open_cq(struct mlx5e_channel *c, struct dim_cq_moder moder,
1642 struct mlx5e_cq_param *param, struct mlx5e_cq *cq)
1643 {
1644 struct mlx5_core_dev *mdev = c->mdev;
1645 int err;
1646
1647 err = mlx5e_alloc_cq(c, param, cq);
1648 if (err)
1649 return err;
1650
1651 err = mlx5e_create_cq(cq, param);
1652 if (err)
1653 goto err_free_cq;
1654
1655 if (MLX5_CAP_GEN(mdev, cq_moderation))
1656 mlx5_core_modify_cq_moderation(mdev, &cq->mcq, moder.usec, moder.pkts);
1657 return 0;
1658
1659 err_free_cq:
1660 mlx5e_free_cq(cq);
1661
1662 return err;
1663 }
1664
1665 void mlx5e_close_cq(struct mlx5e_cq *cq)
1666 {
1667 mlx5e_destroy_cq(cq);
1668 mlx5e_free_cq(cq);
1669 }
1670
1671 static int mlx5e_open_tx_cqs(struct mlx5e_channel *c,
1672 struct mlx5e_params *params,
1673 struct mlx5e_channel_param *cparam)
1674 {
1675 int err;
1676 int tc;
1677
1678 for (tc = 0; tc < c->num_tc; tc++) {
1679 err = mlx5e_open_cq(c, params->tx_cq_moderation,
1680 &cparam->tx_cq, &c->sq[tc].cq);
1681 if (err)
1682 goto err_close_tx_cqs;
1683 }
1684
1685 return 0;
1686
1687 err_close_tx_cqs:
1688 for (tc--; tc >= 0; tc--)
1689 mlx5e_close_cq(&c->sq[tc].cq);
1690
1691 return err;
1692 }
1693
1694 static void mlx5e_close_tx_cqs(struct mlx5e_channel *c)
1695 {
1696 int tc;
1697
1698 for (tc = 0; tc < c->num_tc; tc++)
1699 mlx5e_close_cq(&c->sq[tc].cq);
1700 }
1701
1702 static int mlx5e_open_sqs(struct mlx5e_channel *c,
1703 struct mlx5e_params *params,
1704 struct mlx5e_channel_param *cparam)
1705 {
1706 int err, tc;
1707
1708 for (tc = 0; tc < params->num_tc; tc++) {
1709 int txq_ix = c->ix + tc * params->num_channels;
1710
1711 err = mlx5e_open_txqsq(c, c->priv->tisn[c->lag_port][tc], txq_ix,
1712 params, &cparam->sq, &c->sq[tc], tc);
1713 if (err)
1714 goto err_close_sqs;
1715 }
1716
1717 return 0;
1718
1719 err_close_sqs:
1720 for (tc--; tc >= 0; tc--)
1721 mlx5e_close_txqsq(&c->sq[tc]);
1722
1723 return err;
1724 }
1725
1726 static void mlx5e_close_sqs(struct mlx5e_channel *c)
1727 {
1728 int tc;
1729
1730 for (tc = 0; tc < c->num_tc; tc++)
1731 mlx5e_close_txqsq(&c->sq[tc]);
1732 }
1733
1734 static int mlx5e_set_sq_maxrate(struct net_device *dev,
1735 struct mlx5e_txqsq *sq, u32 rate)
1736 {
1737 struct mlx5e_priv *priv = netdev_priv(dev);
1738 struct mlx5_core_dev *mdev = priv->mdev;
1739 struct mlx5e_modify_sq_param msp = {0};
1740 struct mlx5_rate_limit rl = {0};
1741 u16 rl_index = 0;
1742 int err;
1743
1744 if (rate == sq->rate_limit)
1745 /* nothing to do */
1746 return 0;
1747
1748 if (sq->rate_limit) {
1749 rl.rate = sq->rate_limit;
1750 /* remove current rl index to free space to next ones */
1751 mlx5_rl_remove_rate(mdev, &rl);
1752 }
1753
1754 sq->rate_limit = 0;
1755
1756 if (rate) {
1757 rl.rate = rate;
1758 err = mlx5_rl_add_rate(mdev, &rl_index, &rl);
1759 if (err) {
1760 netdev_err(dev, "Failed configuring rate %u: %d\n",
1761 rate, err);
1762 return err;
1763 }
1764 }
1765
1766 msp.curr_state = MLX5_SQC_STATE_RDY;
1767 msp.next_state = MLX5_SQC_STATE_RDY;
1768 msp.rl_index = rl_index;
1769 msp.rl_update = true;
1770 err = mlx5e_modify_sq(mdev, sq->sqn, &msp);
1771 if (err) {
1772 netdev_err(dev, "Failed configuring rate %u: %d\n",
1773 rate, err);
1774 /* remove the rate from the table */
1775 if (rate)
1776 mlx5_rl_remove_rate(mdev, &rl);
1777 return err;
1778 }
1779
1780 sq->rate_limit = rate;
1781 return 0;
1782 }
1783
1784 static int mlx5e_set_tx_maxrate(struct net_device *dev, int index, u32 rate)
1785 {
1786 struct mlx5e_priv *priv = netdev_priv(dev);
1787 struct mlx5_core_dev *mdev = priv->mdev;
1788 struct mlx5e_txqsq *sq = priv->txq2sq[index];
1789 int err = 0;
1790
1791 if (!mlx5_rl_is_supported(mdev)) {
1792 netdev_err(dev, "Rate limiting is not supported on this device\n");
1793 return -EINVAL;
1794 }
1795
1796 /* rate is given in Mb/sec, HW config is in Kb/sec */
1797 rate = rate << 10;
1798
1799 /* Check whether rate in valid range, 0 is always valid */
1800 if (rate && !mlx5_rl_is_in_range(mdev, rate)) {
1801 netdev_err(dev, "TX rate %u, is not in range\n", rate);
1802 return -ERANGE;
1803 }
1804
1805 mutex_lock(&priv->state_lock);
1806 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
1807 err = mlx5e_set_sq_maxrate(dev, sq, rate);
1808 if (!err)
1809 priv->tx_rates[index] = rate;
1810 mutex_unlock(&priv->state_lock);
1811
1812 return err;
1813 }
1814
1815 static int mlx5e_open_queues(struct mlx5e_channel *c,
1816 struct mlx5e_params *params,
1817 struct mlx5e_channel_param *cparam)
1818 {
1819 struct dim_cq_moder icocq_moder = {0, 0};
1820 int err;
1821
1822 err = mlx5e_open_cq(c, icocq_moder, &cparam->icosq_cq, &c->icosq.cq);
1823 if (err)
1824 return err;
1825
1826 err = mlx5e_open_tx_cqs(c, params, cparam);
1827 if (err)
1828 goto err_close_icosq_cq;
1829
1830 err = mlx5e_open_cq(c, params->tx_cq_moderation, &cparam->tx_cq, &c->xdpsq.cq);
1831 if (err)
1832 goto err_close_tx_cqs;
1833
1834 err = mlx5e_open_cq(c, params->rx_cq_moderation, &cparam->rx_cq, &c->rq.cq);
1835 if (err)
1836 goto err_close_xdp_tx_cqs;
1837
1838 /* XDP SQ CQ params are same as normal TXQ sq CQ params */
1839 err = c->xdp ? mlx5e_open_cq(c, params->tx_cq_moderation,
1840 &cparam->tx_cq, &c->rq_xdpsq.cq) : 0;
1841 if (err)
1842 goto err_close_rx_cq;
1843
1844 napi_enable(&c->napi);
1845
1846 err = mlx5e_open_icosq(c, params, &cparam->icosq, &c->icosq);
1847 if (err)
1848 goto err_disable_napi;
1849
1850 err = mlx5e_open_sqs(c, params, cparam);
1851 if (err)
1852 goto err_close_icosq;
1853
1854 if (c->xdp) {
1855 err = mlx5e_open_xdpsq(c, params, &cparam->xdp_sq, NULL,
1856 &c->rq_xdpsq, false);
1857 if (err)
1858 goto err_close_sqs;
1859 }
1860
1861 err = mlx5e_open_rq(c, params, &cparam->rq, NULL, NULL, &c->rq);
1862 if (err)
1863 goto err_close_xdp_sq;
1864
1865 err = mlx5e_open_xdpsq(c, params, &cparam->xdp_sq, NULL, &c->xdpsq, true);
1866 if (err)
1867 goto err_close_rq;
1868
1869 return 0;
1870
1871 err_close_rq:
1872 mlx5e_close_rq(&c->rq);
1873
1874 err_close_xdp_sq:
1875 if (c->xdp)
1876 mlx5e_close_xdpsq(&c->rq_xdpsq);
1877
1878 err_close_sqs:
1879 mlx5e_close_sqs(c);
1880
1881 err_close_icosq:
1882 mlx5e_close_icosq(&c->icosq);
1883
1884 err_disable_napi:
1885 napi_disable(&c->napi);
1886
1887 if (c->xdp)
1888 mlx5e_close_cq(&c->rq_xdpsq.cq);
1889
1890 err_close_rx_cq:
1891 mlx5e_close_cq(&c->rq.cq);
1892
1893 err_close_xdp_tx_cqs:
1894 mlx5e_close_cq(&c->xdpsq.cq);
1895
1896 err_close_tx_cqs:
1897 mlx5e_close_tx_cqs(c);
1898
1899 err_close_icosq_cq:
1900 mlx5e_close_cq(&c->icosq.cq);
1901
1902 return err;
1903 }
1904
1905 static void mlx5e_close_queues(struct mlx5e_channel *c)
1906 {
1907 mlx5e_close_xdpsq(&c->xdpsq);
1908 mlx5e_close_rq(&c->rq);
1909 if (c->xdp)
1910 mlx5e_close_xdpsq(&c->rq_xdpsq);
1911 mlx5e_close_sqs(c);
1912 mlx5e_close_icosq(&c->icosq);
1913 napi_disable(&c->napi);
1914 if (c->xdp)
1915 mlx5e_close_cq(&c->rq_xdpsq.cq);
1916 mlx5e_close_cq(&c->rq.cq);
1917 mlx5e_close_cq(&c->xdpsq.cq);
1918 mlx5e_close_tx_cqs(c);
1919 mlx5e_close_cq(&c->icosq.cq);
1920 }
1921
1922 static u8 mlx5e_enumerate_lag_port(struct mlx5_core_dev *mdev, int ix)
1923 {
1924 u16 port_aff_bias = mlx5_core_is_pf(mdev) ? 0 : MLX5_CAP_GEN(mdev, vhca_id);
1925
1926 return (ix + port_aff_bias) % mlx5e_get_num_lag_ports(mdev);
1927 }
1928
1929 static int mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
1930 struct mlx5e_params *params,
1931 struct mlx5e_channel_param *cparam,
1932 struct xdp_umem *umem,
1933 struct mlx5e_channel **cp)
1934 {
1935 int cpu = cpumask_first(mlx5_comp_irq_get_affinity_mask(priv->mdev, ix));
1936 struct net_device *netdev = priv->netdev;
1937 struct mlx5e_xsk_param xsk;
1938 struct mlx5e_channel *c;
1939 unsigned int irq;
1940 int err;
1941 int eqn;
1942
1943 err = mlx5_vector2eqn(priv->mdev, ix, &eqn, &irq);
1944 if (err)
1945 return err;
1946
1947 c = kvzalloc_node(sizeof(*c), GFP_KERNEL, cpu_to_node(cpu));
1948 if (!c)
1949 return -ENOMEM;
1950
1951 c->priv = priv;
1952 c->mdev = priv->mdev;
1953 c->tstamp = &priv->tstamp;
1954 c->ix = ix;
1955 c->cpu = cpu;
1956 c->pdev = priv->mdev->device;
1957 c->netdev = priv->netdev;
1958 c->mkey_be = cpu_to_be32(priv->mdev->mlx5e_res.mkey.key);
1959 c->num_tc = params->num_tc;
1960 c->xdp = !!params->xdp_prog;
1961 c->stats = &priv->channel_stats[ix].ch;
1962 c->irq_desc = irq_to_desc(irq);
1963 c->lag_port = mlx5e_enumerate_lag_port(priv->mdev, ix);
1964
1965 netif_napi_add(netdev, &c->napi, mlx5e_napi_poll, 64);
1966
1967 err = mlx5e_open_queues(c, params, cparam);
1968 if (unlikely(err))
1969 goto err_napi_del;
1970
1971 if (umem) {
1972 mlx5e_build_xsk_param(umem, &xsk);
1973 err = mlx5e_open_xsk(priv, params, &xsk, umem, c);
1974 if (unlikely(err))
1975 goto err_close_queues;
1976 }
1977
1978 *cp = c;
1979
1980 return 0;
1981
1982 err_close_queues:
1983 mlx5e_close_queues(c);
1984
1985 err_napi_del:
1986 netif_napi_del(&c->napi);
1987
1988 kvfree(c);
1989
1990 return err;
1991 }
1992
1993 static void mlx5e_activate_channel(struct mlx5e_channel *c)
1994 {
1995 int tc;
1996
1997 for (tc = 0; tc < c->num_tc; tc++)
1998 mlx5e_activate_txqsq(&c->sq[tc]);
1999 mlx5e_activate_icosq(&c->icosq);
2000 mlx5e_activate_rq(&c->rq);
2001
2002 if (test_bit(MLX5E_CHANNEL_STATE_XSK, c->state))
2003 mlx5e_activate_xsk(c);
2004 }
2005
2006 static void mlx5e_deactivate_channel(struct mlx5e_channel *c)
2007 {
2008 int tc;
2009
2010 if (test_bit(MLX5E_CHANNEL_STATE_XSK, c->state))
2011 mlx5e_deactivate_xsk(c);
2012
2013 mlx5e_deactivate_rq(&c->rq);
2014 mlx5e_deactivate_icosq(&c->icosq);
2015 for (tc = 0; tc < c->num_tc; tc++)
2016 mlx5e_deactivate_txqsq(&c->sq[tc]);
2017 }
2018
2019 static void mlx5e_close_channel(struct mlx5e_channel *c)
2020 {
2021 if (test_bit(MLX5E_CHANNEL_STATE_XSK, c->state))
2022 mlx5e_close_xsk(c);
2023 mlx5e_close_queues(c);
2024 netif_napi_del(&c->napi);
2025
2026 kvfree(c);
2027 }
2028
2029 #define DEFAULT_FRAG_SIZE (2048)
2030
2031 static void mlx5e_build_rq_frags_info(struct mlx5_core_dev *mdev,
2032 struct mlx5e_params *params,
2033 struct mlx5e_xsk_param *xsk,
2034 struct mlx5e_rq_frags_info *info)
2035 {
2036 u32 byte_count = MLX5E_SW2HW_MTU(params, params->sw_mtu);
2037 int frag_size_max = DEFAULT_FRAG_SIZE;
2038 u32 buf_size = 0;
2039 int i;
2040
2041 #ifdef CONFIG_MLX5_EN_IPSEC
2042 if (MLX5_IPSEC_DEV(mdev))
2043 byte_count += MLX5E_METADATA_ETHER_LEN;
2044 #endif
2045
2046 if (mlx5e_rx_is_linear_skb(params, xsk)) {
2047 int frag_stride;
2048
2049 frag_stride = mlx5e_rx_get_linear_frag_sz(params, xsk);
2050 frag_stride = roundup_pow_of_two(frag_stride);
2051
2052 info->arr[0].frag_size = byte_count;
2053 info->arr[0].frag_stride = frag_stride;
2054 info->num_frags = 1;
2055 info->wqe_bulk = PAGE_SIZE / frag_stride;
2056 goto out;
2057 }
2058
2059 if (byte_count > PAGE_SIZE +
2060 (MLX5E_MAX_RX_FRAGS - 1) * frag_size_max)
2061 frag_size_max = PAGE_SIZE;
2062
2063 i = 0;
2064 while (buf_size < byte_count) {
2065 int frag_size = byte_count - buf_size;
2066
2067 if (i < MLX5E_MAX_RX_FRAGS - 1)
2068 frag_size = min(frag_size, frag_size_max);
2069
2070 info->arr[i].frag_size = frag_size;
2071 info->arr[i].frag_stride = roundup_pow_of_two(frag_size);
2072
2073 buf_size += frag_size;
2074 i++;
2075 }
2076 info->num_frags = i;
2077 /* number of different wqes sharing a page */
2078 info->wqe_bulk = 1 + (info->num_frags % 2);
2079
2080 out:
2081 info->wqe_bulk = max_t(u8, info->wqe_bulk, 8);
2082 info->log_num_frags = order_base_2(info->num_frags);
2083 }
2084
2085 static inline u8 mlx5e_get_rqwq_log_stride(u8 wq_type, int ndsegs)
2086 {
2087 int sz = sizeof(struct mlx5_wqe_data_seg) * ndsegs;
2088
2089 switch (wq_type) {
2090 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
2091 sz += sizeof(struct mlx5e_rx_wqe_ll);
2092 break;
2093 default: /* MLX5_WQ_TYPE_CYCLIC */
2094 sz += sizeof(struct mlx5e_rx_wqe_cyc);
2095 }
2096
2097 return order_base_2(sz);
2098 }
2099
2100 static u8 mlx5e_get_rq_log_wq_sz(void *rqc)
2101 {
2102 void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
2103
2104 return MLX5_GET(wq, wq, log_wq_sz);
2105 }
2106
2107 void mlx5e_build_rq_param(struct mlx5e_priv *priv,
2108 struct mlx5e_params *params,
2109 struct mlx5e_xsk_param *xsk,
2110 struct mlx5e_rq_param *param)
2111 {
2112 struct mlx5_core_dev *mdev = priv->mdev;
2113 void *rqc = param->rqc;
2114 void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
2115 int ndsegs = 1;
2116
2117 switch (params->rq_wq_type) {
2118 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
2119 MLX5_SET(wq, wq, log_wqe_num_of_strides,
2120 mlx5e_mpwqe_get_log_num_strides(mdev, params, xsk) -
2121 MLX5_MPWQE_LOG_NUM_STRIDES_BASE);
2122 MLX5_SET(wq, wq, log_wqe_stride_size,
2123 mlx5e_mpwqe_get_log_stride_size(mdev, params, xsk) -
2124 MLX5_MPWQE_LOG_STRIDE_SZ_BASE);
2125 MLX5_SET(wq, wq, log_wq_sz, mlx5e_mpwqe_get_log_rq_size(params, xsk));
2126 break;
2127 default: /* MLX5_WQ_TYPE_CYCLIC */
2128 MLX5_SET(wq, wq, log_wq_sz, params->log_rq_mtu_frames);
2129 mlx5e_build_rq_frags_info(mdev, params, xsk, &param->frags_info);
2130 ndsegs = param->frags_info.num_frags;
2131 }
2132
2133 MLX5_SET(wq, wq, wq_type, params->rq_wq_type);
2134 MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN);
2135 MLX5_SET(wq, wq, log_wq_stride,
2136 mlx5e_get_rqwq_log_stride(params->rq_wq_type, ndsegs));
2137 MLX5_SET(wq, wq, pd, mdev->mlx5e_res.pdn);
2138 MLX5_SET(rqc, rqc, counter_set_id, priv->q_counter);
2139 MLX5_SET(rqc, rqc, vsd, params->vlan_strip_disable);
2140 MLX5_SET(rqc, rqc, scatter_fcs, params->scatter_fcs_en);
2141
2142 param->wq.buf_numa_node = dev_to_node(mdev->device);
2143 }
2144
2145 static void mlx5e_build_drop_rq_param(struct mlx5e_priv *priv,
2146 struct mlx5e_rq_param *param)
2147 {
2148 struct mlx5_core_dev *mdev = priv->mdev;
2149 void *rqc = param->rqc;
2150 void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
2151
2152 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
2153 MLX5_SET(wq, wq, log_wq_stride,
2154 mlx5e_get_rqwq_log_stride(MLX5_WQ_TYPE_CYCLIC, 1));
2155 MLX5_SET(rqc, rqc, counter_set_id, priv->drop_rq_q_counter);
2156
2157 param->wq.buf_numa_node = dev_to_node(mdev->device);
2158 }
2159
2160 void mlx5e_build_sq_param_common(struct mlx5e_priv *priv,
2161 struct mlx5e_sq_param *param)
2162 {
2163 void *sqc = param->sqc;
2164 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
2165
2166 MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
2167 MLX5_SET(wq, wq, pd, priv->mdev->mlx5e_res.pdn);
2168
2169 param->wq.buf_numa_node = dev_to_node(priv->mdev->device);
2170 }
2171
2172 static void mlx5e_build_sq_param(struct mlx5e_priv *priv,
2173 struct mlx5e_params *params,
2174 struct mlx5e_sq_param *param)
2175 {
2176 void *sqc = param->sqc;
2177 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
2178 bool allow_swp;
2179
2180 allow_swp = mlx5_geneve_tx_allowed(priv->mdev) ||
2181 !!MLX5_IPSEC_DEV(priv->mdev);
2182 mlx5e_build_sq_param_common(priv, param);
2183 MLX5_SET(wq, wq, log_wq_sz, params->log_sq_size);
2184 MLX5_SET(sqc, sqc, allow_swp, allow_swp);
2185 }
2186
2187 static void mlx5e_build_common_cq_param(struct mlx5e_priv *priv,
2188 struct mlx5e_cq_param *param)
2189 {
2190 void *cqc = param->cqc;
2191
2192 MLX5_SET(cqc, cqc, uar_page, priv->mdev->priv.uar->index);
2193 if (MLX5_CAP_GEN(priv->mdev, cqe_128_always) && cache_line_size() >= 128)
2194 MLX5_SET(cqc, cqc, cqe_sz, CQE_STRIDE_128_PAD);
2195 }
2196
2197 void mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
2198 struct mlx5e_params *params,
2199 struct mlx5e_xsk_param *xsk,
2200 struct mlx5e_cq_param *param)
2201 {
2202 struct mlx5_core_dev *mdev = priv->mdev;
2203 void *cqc = param->cqc;
2204 u8 log_cq_size;
2205
2206 switch (params->rq_wq_type) {
2207 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
2208 log_cq_size = mlx5e_mpwqe_get_log_rq_size(params, xsk) +
2209 mlx5e_mpwqe_get_log_num_strides(mdev, params, xsk);
2210 break;
2211 default: /* MLX5_WQ_TYPE_CYCLIC */
2212 log_cq_size = params->log_rq_mtu_frames;
2213 }
2214
2215 MLX5_SET(cqc, cqc, log_cq_size, log_cq_size);
2216 if (MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS)) {
2217 MLX5_SET(cqc, cqc, mini_cqe_res_format, MLX5_CQE_FORMAT_CSUM);
2218 MLX5_SET(cqc, cqc, cqe_comp_en, 1);
2219 }
2220
2221 mlx5e_build_common_cq_param(priv, param);
2222 param->cq_period_mode = params->rx_cq_moderation.cq_period_mode;
2223 }
2224
2225 void mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
2226 struct mlx5e_params *params,
2227 struct mlx5e_cq_param *param)
2228 {
2229 void *cqc = param->cqc;
2230
2231 MLX5_SET(cqc, cqc, log_cq_size, params->log_sq_size);
2232
2233 mlx5e_build_common_cq_param(priv, param);
2234 param->cq_period_mode = params->tx_cq_moderation.cq_period_mode;
2235 }
2236
2237 void mlx5e_build_ico_cq_param(struct mlx5e_priv *priv,
2238 u8 log_wq_size,
2239 struct mlx5e_cq_param *param)
2240 {
2241 void *cqc = param->cqc;
2242
2243 MLX5_SET(cqc, cqc, log_cq_size, log_wq_size);
2244
2245 mlx5e_build_common_cq_param(priv, param);
2246
2247 param->cq_period_mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
2248 }
2249
2250 void mlx5e_build_icosq_param(struct mlx5e_priv *priv,
2251 u8 log_wq_size,
2252 struct mlx5e_sq_param *param)
2253 {
2254 void *sqc = param->sqc;
2255 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
2256
2257 mlx5e_build_sq_param_common(priv, param);
2258
2259 MLX5_SET(wq, wq, log_wq_sz, log_wq_size);
2260 MLX5_SET(sqc, sqc, reg_umr, MLX5_CAP_ETH(priv->mdev, reg_umr_sq));
2261 }
2262
2263 void mlx5e_build_xdpsq_param(struct mlx5e_priv *priv,
2264 struct mlx5e_params *params,
2265 struct mlx5e_sq_param *param)
2266 {
2267 void *sqc = param->sqc;
2268 void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
2269
2270 mlx5e_build_sq_param_common(priv, param);
2271 MLX5_SET(wq, wq, log_wq_sz, params->log_sq_size);
2272 param->is_mpw = MLX5E_GET_PFLAG(params, MLX5E_PFLAG_XDP_TX_MPWQE);
2273 }
2274
2275 static u8 mlx5e_build_icosq_log_wq_sz(struct mlx5e_params *params,
2276 struct mlx5e_rq_param *rqp)
2277 {
2278 switch (params->rq_wq_type) {
2279 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
2280 return order_base_2(MLX5E_UMR_WQEBBS) +
2281 mlx5e_get_rq_log_wq_sz(rqp->rqc);
2282 default: /* MLX5_WQ_TYPE_CYCLIC */
2283 return MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE;
2284 }
2285 }
2286
2287 static void mlx5e_build_channel_param(struct mlx5e_priv *priv,
2288 struct mlx5e_params *params,
2289 struct mlx5e_channel_param *cparam)
2290 {
2291 u8 icosq_log_wq_sz;
2292
2293 mlx5e_build_rq_param(priv, params, NULL, &cparam->rq);
2294
2295 icosq_log_wq_sz = mlx5e_build_icosq_log_wq_sz(params, &cparam->rq);
2296
2297 mlx5e_build_sq_param(priv, params, &cparam->sq);
2298 mlx5e_build_xdpsq_param(priv, params, &cparam->xdp_sq);
2299 mlx5e_build_icosq_param(priv, icosq_log_wq_sz, &cparam->icosq);
2300 mlx5e_build_rx_cq_param(priv, params, NULL, &cparam->rx_cq);
2301 mlx5e_build_tx_cq_param(priv, params, &cparam->tx_cq);
2302 mlx5e_build_ico_cq_param(priv, icosq_log_wq_sz, &cparam->icosq_cq);
2303 }
2304
2305 int mlx5e_open_channels(struct mlx5e_priv *priv,
2306 struct mlx5e_channels *chs)
2307 {
2308 struct mlx5e_channel_param *cparam;
2309 int err = -ENOMEM;
2310 int i;
2311
2312 chs->num = chs->params.num_channels;
2313
2314 chs->c = kcalloc(chs->num, sizeof(struct mlx5e_channel *), GFP_KERNEL);
2315 cparam = kvzalloc(sizeof(struct mlx5e_channel_param), GFP_KERNEL);
2316 if (!chs->c || !cparam)
2317 goto err_free;
2318
2319 mlx5e_build_channel_param(priv, &chs->params, cparam);
2320 for (i = 0; i < chs->num; i++) {
2321 struct xdp_umem *umem = NULL;
2322
2323 if (chs->params.xdp_prog)
2324 umem = mlx5e_xsk_get_umem(&chs->params, chs->params.xsk, i);
2325
2326 err = mlx5e_open_channel(priv, i, &chs->params, cparam, umem, &chs->c[i]);
2327 if (err)
2328 goto err_close_channels;
2329 }
2330
2331 mlx5e_health_channels_update(priv);
2332 kvfree(cparam);
2333 return 0;
2334
2335 err_close_channels:
2336 for (i--; i >= 0; i--)
2337 mlx5e_close_channel(chs->c[i]);
2338
2339 err_free:
2340 kfree(chs->c);
2341 kvfree(cparam);
2342 chs->num = 0;
2343 return err;
2344 }
2345
2346 static void mlx5e_activate_channels(struct mlx5e_channels *chs)
2347 {
2348 int i;
2349
2350 for (i = 0; i < chs->num; i++)
2351 mlx5e_activate_channel(chs->c[i]);
2352 }
2353
2354 #define MLX5E_RQ_WQES_TIMEOUT 20000 /* msecs */
2355
2356 static int mlx5e_wait_channels_min_rx_wqes(struct mlx5e_channels *chs)
2357 {
2358 int err = 0;
2359 int i;
2360
2361 for (i = 0; i < chs->num; i++) {
2362 int timeout = err ? 0 : MLX5E_RQ_WQES_TIMEOUT;
2363
2364 err |= mlx5e_wait_for_min_rx_wqes(&chs->c[i]->rq, timeout);
2365
2366 /* Don't wait on the XSK RQ, because the newer xdpsock sample
2367 * doesn't provide any Fill Ring entries at the setup stage.
2368 */
2369 }
2370
2371 return err ? -ETIMEDOUT : 0;
2372 }
2373
2374 static void mlx5e_deactivate_channels(struct mlx5e_channels *chs)
2375 {
2376 int i;
2377
2378 for (i = 0; i < chs->num; i++)
2379 mlx5e_deactivate_channel(chs->c[i]);
2380 }
2381
2382 void mlx5e_close_channels(struct mlx5e_channels *chs)
2383 {
2384 int i;
2385
2386 for (i = 0; i < chs->num; i++)
2387 mlx5e_close_channel(chs->c[i]);
2388
2389 kfree(chs->c);
2390 chs->num = 0;
2391 }
2392
2393 static int
2394 mlx5e_create_rqt(struct mlx5e_priv *priv, int sz, struct mlx5e_rqt *rqt)
2395 {
2396 struct mlx5_core_dev *mdev = priv->mdev;
2397 void *rqtc;
2398 int inlen;
2399 int err;
2400 u32 *in;
2401 int i;
2402
2403 inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz;
2404 in = kvzalloc(inlen, GFP_KERNEL);
2405 if (!in)
2406 return -ENOMEM;
2407
2408 rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context);
2409
2410 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
2411 MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
2412
2413 for (i = 0; i < sz; i++)
2414 MLX5_SET(rqtc, rqtc, rq_num[i], priv->drop_rq.rqn);
2415
2416 err = mlx5_core_create_rqt(mdev, in, inlen, &rqt->rqtn);
2417 if (!err)
2418 rqt->enabled = true;
2419
2420 kvfree(in);
2421 return err;
2422 }
2423
2424 void mlx5e_destroy_rqt(struct mlx5e_priv *priv, struct mlx5e_rqt *rqt)
2425 {
2426 rqt->enabled = false;
2427 mlx5_core_destroy_rqt(priv->mdev, rqt->rqtn);
2428 }
2429
2430 int mlx5e_create_indirect_rqt(struct mlx5e_priv *priv)
2431 {
2432 struct mlx5e_rqt *rqt = &priv->indir_rqt;
2433 int err;
2434
2435 err = mlx5e_create_rqt(priv, MLX5E_INDIR_RQT_SIZE, rqt);
2436 if (err)
2437 mlx5_core_warn(priv->mdev, "create indirect rqts failed, %d\n", err);
2438 return err;
2439 }
2440
2441 int mlx5e_create_direct_rqts(struct mlx5e_priv *priv, struct mlx5e_tir *tirs)
2442 {
2443 int err;
2444 int ix;
2445
2446 for (ix = 0; ix < priv->max_nch; ix++) {
2447 err = mlx5e_create_rqt(priv, 1 /*size */, &tirs[ix].rqt);
2448 if (unlikely(err))
2449 goto err_destroy_rqts;
2450 }
2451
2452 return 0;
2453
2454 err_destroy_rqts:
2455 mlx5_core_warn(priv->mdev, "create rqts failed, %d\n", err);
2456 for (ix--; ix >= 0; ix--)
2457 mlx5e_destroy_rqt(priv, &tirs[ix].rqt);
2458
2459 return err;
2460 }
2461
2462 void mlx5e_destroy_direct_rqts(struct mlx5e_priv *priv, struct mlx5e_tir *tirs)
2463 {
2464 int i;
2465
2466 for (i = 0; i < priv->max_nch; i++)
2467 mlx5e_destroy_rqt(priv, &tirs[i].rqt);
2468 }
2469
2470 static int mlx5e_rx_hash_fn(int hfunc)
2471 {
2472 return (hfunc == ETH_RSS_HASH_TOP) ?
2473 MLX5_RX_HASH_FN_TOEPLITZ :
2474 MLX5_RX_HASH_FN_INVERTED_XOR8;
2475 }
2476
2477 int mlx5e_bits_invert(unsigned long a, int size)
2478 {
2479 int inv = 0;
2480 int i;
2481
2482 for (i = 0; i < size; i++)
2483 inv |= (test_bit(size - i - 1, &a) ? 1 : 0) << i;
2484
2485 return inv;
2486 }
2487
2488 static void mlx5e_fill_rqt_rqns(struct mlx5e_priv *priv, int sz,
2489 struct mlx5e_redirect_rqt_param rrp, void *rqtc)
2490 {
2491 int i;
2492
2493 for (i = 0; i < sz; i++) {
2494 u32 rqn;
2495
2496 if (rrp.is_rss) {
2497 int ix = i;
2498
2499 if (rrp.rss.hfunc == ETH_RSS_HASH_XOR)
2500 ix = mlx5e_bits_invert(i, ilog2(sz));
2501
2502 ix = priv->rss_params.indirection_rqt[ix];
2503 rqn = rrp.rss.channels->c[ix]->rq.rqn;
2504 } else {
2505 rqn = rrp.rqn;
2506 }
2507 MLX5_SET(rqtc, rqtc, rq_num[i], rqn);
2508 }
2509 }
2510
2511 int mlx5e_redirect_rqt(struct mlx5e_priv *priv, u32 rqtn, int sz,
2512 struct mlx5e_redirect_rqt_param rrp)
2513 {
2514 struct mlx5_core_dev *mdev = priv->mdev;
2515 void *rqtc;
2516 int inlen;
2517 u32 *in;
2518 int err;
2519
2520 inlen = MLX5_ST_SZ_BYTES(modify_rqt_in) + sizeof(u32) * sz;
2521 in = kvzalloc(inlen, GFP_KERNEL);
2522 if (!in)
2523 return -ENOMEM;
2524
2525 rqtc = MLX5_ADDR_OF(modify_rqt_in, in, ctx);
2526
2527 MLX5_SET(rqtc, rqtc, rqt_actual_size, sz);
2528 MLX5_SET(modify_rqt_in, in, bitmask.rqn_list, 1);
2529 mlx5e_fill_rqt_rqns(priv, sz, rrp, rqtc);
2530 err = mlx5_core_modify_rqt(mdev, rqtn, in, inlen);
2531
2532 kvfree(in);
2533 return err;
2534 }
2535
2536 static u32 mlx5e_get_direct_rqn(struct mlx5e_priv *priv, int ix,
2537 struct mlx5e_redirect_rqt_param rrp)
2538 {
2539 if (!rrp.is_rss)
2540 return rrp.rqn;
2541
2542 if (ix >= rrp.rss.channels->num)
2543 return priv->drop_rq.rqn;
2544
2545 return rrp.rss.channels->c[ix]->rq.rqn;
2546 }
2547
2548 static void mlx5e_redirect_rqts(struct mlx5e_priv *priv,
2549 struct mlx5e_redirect_rqt_param rrp)
2550 {
2551 u32 rqtn;
2552 int ix;
2553
2554 if (priv->indir_rqt.enabled) {
2555 /* RSS RQ table */
2556 rqtn = priv->indir_rqt.rqtn;
2557 mlx5e_redirect_rqt(priv, rqtn, MLX5E_INDIR_RQT_SIZE, rrp);
2558 }
2559
2560 for (ix = 0; ix < priv->max_nch; ix++) {
2561 struct mlx5e_redirect_rqt_param direct_rrp = {
2562 .is_rss = false,
2563 {
2564 .rqn = mlx5e_get_direct_rqn(priv, ix, rrp)
2565 },
2566 };
2567
2568 /* Direct RQ Tables */
2569 if (!priv->direct_tir[ix].rqt.enabled)
2570 continue;
2571
2572 rqtn = priv->direct_tir[ix].rqt.rqtn;
2573 mlx5e_redirect_rqt(priv, rqtn, 1, direct_rrp);
2574 }
2575 }
2576
2577 static void mlx5e_redirect_rqts_to_channels(struct mlx5e_priv *priv,
2578 struct mlx5e_channels *chs)
2579 {
2580 struct mlx5e_redirect_rqt_param rrp = {
2581 .is_rss = true,
2582 {
2583 .rss = {
2584 .channels = chs,
2585 .hfunc = priv->rss_params.hfunc,
2586 }
2587 },
2588 };
2589
2590 mlx5e_redirect_rqts(priv, rrp);
2591 }
2592
2593 static void mlx5e_redirect_rqts_to_drop(struct mlx5e_priv *priv)
2594 {
2595 struct mlx5e_redirect_rqt_param drop_rrp = {
2596 .is_rss = false,
2597 {
2598 .rqn = priv->drop_rq.rqn,
2599 },
2600 };
2601
2602 mlx5e_redirect_rqts(priv, drop_rrp);
2603 }
2604
2605 static const struct mlx5e_tirc_config tirc_default_config[MLX5E_NUM_INDIR_TIRS] = {
2606 [MLX5E_TT_IPV4_TCP] = { .l3_prot_type = MLX5_L3_PROT_TYPE_IPV4,
2607 .l4_prot_type = MLX5_L4_PROT_TYPE_TCP,
2608 .rx_hash_fields = MLX5_HASH_IP_L4PORTS,
2609 },
2610 [MLX5E_TT_IPV6_TCP] = { .l3_prot_type = MLX5_L3_PROT_TYPE_IPV6,
2611 .l4_prot_type = MLX5_L4_PROT_TYPE_TCP,
2612 .rx_hash_fields = MLX5_HASH_IP_L4PORTS,
2613 },
2614 [MLX5E_TT_IPV4_UDP] = { .l3_prot_type = MLX5_L3_PROT_TYPE_IPV4,
2615 .l4_prot_type = MLX5_L4_PROT_TYPE_UDP,
2616 .rx_hash_fields = MLX5_HASH_IP_L4PORTS,
2617 },
2618 [MLX5E_TT_IPV6_UDP] = { .l3_prot_type = MLX5_L3_PROT_TYPE_IPV6,
2619 .l4_prot_type = MLX5_L4_PROT_TYPE_UDP,
2620 .rx_hash_fields = MLX5_HASH_IP_L4PORTS,
2621 },
2622 [MLX5E_TT_IPV4_IPSEC_AH] = { .l3_prot_type = MLX5_L3_PROT_TYPE_IPV4,
2623 .l4_prot_type = 0,
2624 .rx_hash_fields = MLX5_HASH_IP_IPSEC_SPI,
2625 },
2626 [MLX5E_TT_IPV6_IPSEC_AH] = { .l3_prot_type = MLX5_L3_PROT_TYPE_IPV6,
2627 .l4_prot_type = 0,
2628 .rx_hash_fields = MLX5_HASH_IP_IPSEC_SPI,
2629 },
2630 [MLX5E_TT_IPV4_IPSEC_ESP] = { .l3_prot_type = MLX5_L3_PROT_TYPE_IPV4,
2631 .l4_prot_type = 0,
2632 .rx_hash_fields = MLX5_HASH_IP_IPSEC_SPI,
2633 },
2634 [MLX5E_TT_IPV6_IPSEC_ESP] = { .l3_prot_type = MLX5_L3_PROT_TYPE_IPV6,
2635 .l4_prot_type = 0,
2636 .rx_hash_fields = MLX5_HASH_IP_IPSEC_SPI,
2637 },
2638 [MLX5E_TT_IPV4] = { .l3_prot_type = MLX5_L3_PROT_TYPE_IPV4,
2639 .l4_prot_type = 0,
2640 .rx_hash_fields = MLX5_HASH_IP,
2641 },
2642 [MLX5E_TT_IPV6] = { .l3_prot_type = MLX5_L3_PROT_TYPE_IPV6,
2643 .l4_prot_type = 0,
2644 .rx_hash_fields = MLX5_HASH_IP,
2645 },
2646 };
2647
2648 struct mlx5e_tirc_config mlx5e_tirc_get_default_config(enum mlx5e_traffic_types tt)
2649 {
2650 return tirc_default_config[tt];
2651 }
2652
2653 static void mlx5e_build_tir_ctx_lro(struct mlx5e_params *params, void *tirc)
2654 {
2655 if (!params->lro_en)
2656 return;
2657
2658 #define ROUGH_MAX_L2_L3_HDR_SZ 256
2659
2660 MLX5_SET(tirc, tirc, lro_enable_mask,
2661 MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO |
2662 MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO);
2663 MLX5_SET(tirc, tirc, lro_max_ip_payload_size,
2664 (MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ - ROUGH_MAX_L2_L3_HDR_SZ) >> 8);
2665 MLX5_SET(tirc, tirc, lro_timeout_period_usecs, params->lro_timeout);
2666 }
2667
2668 void mlx5e_build_indir_tir_ctx_hash(struct mlx5e_rss_params *rss_params,
2669 const struct mlx5e_tirc_config *ttconfig,
2670 void *tirc, bool inner)
2671 {
2672 void *hfso = inner ? MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_inner) :
2673 MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
2674
2675 MLX5_SET(tirc, tirc, rx_hash_fn, mlx5e_rx_hash_fn(rss_params->hfunc));
2676 if (rss_params->hfunc == ETH_RSS_HASH_TOP) {
2677 void *rss_key = MLX5_ADDR_OF(tirc, tirc,
2678 rx_hash_toeplitz_key);
2679 size_t len = MLX5_FLD_SZ_BYTES(tirc,
2680 rx_hash_toeplitz_key);
2681
2682 MLX5_SET(tirc, tirc, rx_hash_symmetric, 1);
2683 memcpy(rss_key, rss_params->toeplitz_hash_key, len);
2684 }
2685 MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
2686 ttconfig->l3_prot_type);
2687 MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
2688 ttconfig->l4_prot_type);
2689 MLX5_SET(rx_hash_field_select, hfso, selected_fields,
2690 ttconfig->rx_hash_fields);
2691 }
2692
2693 static void mlx5e_update_rx_hash_fields(struct mlx5e_tirc_config *ttconfig,
2694 enum mlx5e_traffic_types tt,
2695 u32 rx_hash_fields)
2696 {
2697 *ttconfig = tirc_default_config[tt];
2698 ttconfig->rx_hash_fields = rx_hash_fields;
2699 }
2700
2701 void mlx5e_modify_tirs_hash(struct mlx5e_priv *priv, void *in, int inlen)
2702 {
2703 void *tirc = MLX5_ADDR_OF(modify_tir_in, in, ctx);
2704 struct mlx5e_rss_params *rss = &priv->rss_params;
2705 struct mlx5_core_dev *mdev = priv->mdev;
2706 int ctxlen = MLX5_ST_SZ_BYTES(tirc);
2707 struct mlx5e_tirc_config ttconfig;
2708 int tt;
2709
2710 MLX5_SET(modify_tir_in, in, bitmask.hash, 1);
2711
2712 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
2713 memset(tirc, 0, ctxlen);
2714 mlx5e_update_rx_hash_fields(&ttconfig, tt,
2715 rss->rx_hash_fields[tt]);
2716 mlx5e_build_indir_tir_ctx_hash(rss, &ttconfig, tirc, false);
2717 mlx5_core_modify_tir(mdev, priv->indir_tir[tt].tirn, in, inlen);
2718 }
2719
2720 if (!mlx5e_tunnel_inner_ft_supported(priv->mdev))
2721 return;
2722
2723 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
2724 memset(tirc, 0, ctxlen);
2725 mlx5e_update_rx_hash_fields(&ttconfig, tt,
2726 rss->rx_hash_fields[tt]);
2727 mlx5e_build_indir_tir_ctx_hash(rss, &ttconfig, tirc, true);
2728 mlx5_core_modify_tir(mdev, priv->inner_indir_tir[tt].tirn, in,
2729 inlen);
2730 }
2731 }
2732
2733 static int mlx5e_modify_tirs_lro(struct mlx5e_priv *priv)
2734 {
2735 struct mlx5_core_dev *mdev = priv->mdev;
2736
2737 void *in;
2738 void *tirc;
2739 int inlen;
2740 int err;
2741 int tt;
2742 int ix;
2743
2744 inlen = MLX5_ST_SZ_BYTES(modify_tir_in);
2745 in = kvzalloc(inlen, GFP_KERNEL);
2746 if (!in)
2747 return -ENOMEM;
2748
2749 MLX5_SET(modify_tir_in, in, bitmask.lro, 1);
2750 tirc = MLX5_ADDR_OF(modify_tir_in, in, ctx);
2751
2752 mlx5e_build_tir_ctx_lro(&priv->channels.params, tirc);
2753
2754 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
2755 err = mlx5_core_modify_tir(mdev, priv->indir_tir[tt].tirn, in,
2756 inlen);
2757 if (err)
2758 goto free_in;
2759 }
2760
2761 for (ix = 0; ix < priv->max_nch; ix++) {
2762 err = mlx5_core_modify_tir(mdev, priv->direct_tir[ix].tirn,
2763 in, inlen);
2764 if (err)
2765 goto free_in;
2766 }
2767
2768 free_in:
2769 kvfree(in);
2770
2771 return err;
2772 }
2773
2774 static MLX5E_DEFINE_PREACTIVATE_WRAPPER_CTX(mlx5e_modify_tirs_lro);
2775
2776 static int mlx5e_set_mtu(struct mlx5_core_dev *mdev,
2777 struct mlx5e_params *params, u16 mtu)
2778 {
2779 u16 hw_mtu = MLX5E_SW2HW_MTU(params, mtu);
2780 int err;
2781
2782 err = mlx5_set_port_mtu(mdev, hw_mtu, 1);
2783 if (err)
2784 return err;
2785
2786 /* Update vport context MTU */
2787 mlx5_modify_nic_vport_mtu(mdev, hw_mtu);
2788 return 0;
2789 }
2790
2791 static void mlx5e_query_mtu(struct mlx5_core_dev *mdev,
2792 struct mlx5e_params *params, u16 *mtu)
2793 {
2794 u16 hw_mtu = 0;
2795 int err;
2796
2797 err = mlx5_query_nic_vport_mtu(mdev, &hw_mtu);
2798 if (err || !hw_mtu) /* fallback to port oper mtu */
2799 mlx5_query_port_oper_mtu(mdev, &hw_mtu, 1);
2800
2801 *mtu = MLX5E_HW2SW_MTU(params, hw_mtu);
2802 }
2803
2804 int mlx5e_set_dev_port_mtu(struct mlx5e_priv *priv)
2805 {
2806 struct mlx5e_params *params = &priv->channels.params;
2807 struct net_device *netdev = priv->netdev;
2808 struct mlx5_core_dev *mdev = priv->mdev;
2809 u16 mtu;
2810 int err;
2811
2812 err = mlx5e_set_mtu(mdev, params, params->sw_mtu);
2813 if (err)
2814 return err;
2815
2816 mlx5e_query_mtu(mdev, params, &mtu);
2817 if (mtu != params->sw_mtu)
2818 netdev_warn(netdev, "%s: VPort MTU %d is different than netdev mtu %d\n",
2819 __func__, mtu, params->sw_mtu);
2820
2821 params->sw_mtu = mtu;
2822 return 0;
2823 }
2824
2825 MLX5E_DEFINE_PREACTIVATE_WRAPPER_CTX(mlx5e_set_dev_port_mtu);
2826
2827 void mlx5e_set_netdev_mtu_boundaries(struct mlx5e_priv *priv)
2828 {
2829 struct mlx5e_params *params = &priv->channels.params;
2830 struct net_device *netdev = priv->netdev;
2831 struct mlx5_core_dev *mdev = priv->mdev;
2832 u16 max_mtu;
2833
2834 /* MTU range: 68 - hw-specific max */
2835 netdev->min_mtu = ETH_MIN_MTU;
2836
2837 mlx5_query_port_max_mtu(mdev, &max_mtu, 1);
2838 netdev->max_mtu = min_t(unsigned int, MLX5E_HW2SW_MTU(params, max_mtu),
2839 ETH_MAX_MTU);
2840 }
2841
2842 static void mlx5e_netdev_set_tcs(struct net_device *netdev)
2843 {
2844 struct mlx5e_priv *priv = netdev_priv(netdev);
2845 int nch = priv->channels.params.num_channels;
2846 int ntc = priv->channels.params.num_tc;
2847 int tc;
2848
2849 netdev_reset_tc(netdev);
2850
2851 if (ntc == 1)
2852 return;
2853
2854 netdev_set_num_tc(netdev, ntc);
2855
2856 /* Map netdev TCs to offset 0
2857 * We have our own UP to TXQ mapping for QoS
2858 */
2859 for (tc = 0; tc < ntc; tc++)
2860 netdev_set_tc_queue(netdev, tc, nch, 0);
2861 }
2862
2863 static void mlx5e_update_netdev_queues(struct mlx5e_priv *priv, u16 count)
2864 {
2865 int num_txqs = count * priv->channels.params.num_tc;
2866 int num_rxqs = count * priv->profile->rq_groups;
2867 struct net_device *netdev = priv->netdev;
2868
2869 mlx5e_netdev_set_tcs(netdev);
2870 netif_set_real_num_tx_queues(netdev, num_txqs);
2871 netif_set_real_num_rx_queues(netdev, num_rxqs);
2872 }
2873
2874 static void mlx5e_set_default_xps_cpumasks(struct mlx5e_priv *priv,
2875 struct mlx5e_params *params)
2876 {
2877 struct mlx5_core_dev *mdev = priv->mdev;
2878 int num_comp_vectors, ix, irq;
2879
2880 num_comp_vectors = mlx5_comp_vectors_count(mdev);
2881
2882 for (ix = 0; ix < params->num_channels; ix++) {
2883 cpumask_clear(priv->scratchpad.cpumask);
2884
2885 for (irq = ix; irq < num_comp_vectors; irq += params->num_channels) {
2886 int cpu = cpumask_first(mlx5_comp_irq_get_affinity_mask(mdev, irq));
2887
2888 cpumask_set_cpu(cpu, priv->scratchpad.cpumask);
2889 }
2890
2891 netif_set_xps_queue(priv->netdev, priv->scratchpad.cpumask, ix);
2892 }
2893 }
2894
2895 int mlx5e_num_channels_changed(struct mlx5e_priv *priv)
2896 {
2897 u16 count = priv->channels.params.num_channels;
2898
2899 mlx5e_update_netdev_queues(priv, count);
2900 mlx5e_set_default_xps_cpumasks(priv, &priv->channels.params);
2901
2902 if (!netif_is_rxfh_configured(priv->netdev))
2903 mlx5e_build_default_indir_rqt(priv->rss_params.indirection_rqt,
2904 MLX5E_INDIR_RQT_SIZE, count);
2905
2906 return 0;
2907 }
2908
2909 MLX5E_DEFINE_PREACTIVATE_WRAPPER_CTX(mlx5e_num_channels_changed);
2910
2911 static void mlx5e_build_txq_maps(struct mlx5e_priv *priv)
2912 {
2913 int i, ch;
2914
2915 ch = priv->channels.num;
2916
2917 for (i = 0; i < ch; i++) {
2918 int tc;
2919
2920 for (tc = 0; tc < priv->channels.params.num_tc; tc++) {
2921 struct mlx5e_channel *c = priv->channels.c[i];
2922 struct mlx5e_txqsq *sq = &c->sq[tc];
2923
2924 priv->txq2sq[sq->txq_ix] = sq;
2925 priv->channel_tc2realtxq[i][tc] = i + tc * ch;
2926 }
2927 }
2928 }
2929
2930 void mlx5e_activate_priv_channels(struct mlx5e_priv *priv)
2931 {
2932 mlx5e_build_txq_maps(priv);
2933 mlx5e_activate_channels(&priv->channels);
2934 mlx5e_xdp_tx_enable(priv);
2935 netif_tx_start_all_queues(priv->netdev);
2936
2937 if (mlx5e_is_vport_rep(priv))
2938 mlx5e_add_sqs_fwd_rules(priv);
2939
2940 mlx5e_wait_channels_min_rx_wqes(&priv->channels);
2941 mlx5e_redirect_rqts_to_channels(priv, &priv->channels);
2942
2943 mlx5e_xsk_redirect_rqts_to_channels(priv, &priv->channels);
2944 }
2945
2946 void mlx5e_deactivate_priv_channels(struct mlx5e_priv *priv)
2947 {
2948 mlx5e_xsk_redirect_rqts_to_drop(priv, &priv->channels);
2949
2950 mlx5e_redirect_rqts_to_drop(priv);
2951
2952 if (mlx5e_is_vport_rep(priv))
2953 mlx5e_remove_sqs_fwd_rules(priv);
2954
2955 /* FIXME: This is a W/A only for tx timeout watch dog false alarm when
2956 * polling for inactive tx queues.
2957 */
2958 netif_tx_stop_all_queues(priv->netdev);
2959 netif_tx_disable(priv->netdev);
2960 mlx5e_xdp_tx_disable(priv);
2961 mlx5e_deactivate_channels(&priv->channels);
2962 }
2963
2964 static int mlx5e_switch_priv_channels(struct mlx5e_priv *priv,
2965 struct mlx5e_channels *new_chs,
2966 mlx5e_fp_preactivate preactivate,
2967 void *context)
2968 {
2969 struct net_device *netdev = priv->netdev;
2970 struct mlx5e_channels old_chs;
2971 int carrier_ok;
2972 int err = 0;
2973
2974 carrier_ok = netif_carrier_ok(netdev);
2975 netif_carrier_off(netdev);
2976
2977 mlx5e_deactivate_priv_channels(priv);
2978
2979 old_chs = priv->channels;
2980 priv->channels = *new_chs;
2981
2982 /* New channels are ready to roll, call the preactivate hook if needed
2983 * to modify HW settings or update kernel parameters.
2984 */
2985 if (preactivate) {
2986 err = preactivate(priv, context);
2987 if (err) {
2988 priv->channels = old_chs;
2989 goto out;
2990 }
2991 }
2992
2993 mlx5e_close_channels(&old_chs);
2994 priv->profile->update_rx(priv);
2995
2996 out:
2997 mlx5e_activate_priv_channels(priv);
2998
2999 /* return carrier back if needed */
3000 if (carrier_ok)
3001 netif_carrier_on(netdev);
3002
3003 return err;
3004 }
3005
3006 int mlx5e_safe_switch_channels(struct mlx5e_priv *priv,
3007 struct mlx5e_channels *new_chs,
3008 mlx5e_fp_preactivate preactivate,
3009 void *context)
3010 {
3011 int err;
3012
3013 err = mlx5e_open_channels(priv, new_chs);
3014 if (err)
3015 return err;
3016
3017 err = mlx5e_switch_priv_channels(priv, new_chs, preactivate, context);
3018 if (err)
3019 goto err_close;
3020
3021 return 0;
3022
3023 err_close:
3024 mlx5e_close_channels(new_chs);
3025
3026 return err;
3027 }
3028
3029 int mlx5e_safe_reopen_channels(struct mlx5e_priv *priv)
3030 {
3031 struct mlx5e_channels new_channels = {};
3032
3033 new_channels.params = priv->channels.params;
3034 return mlx5e_safe_switch_channels(priv, &new_channels, NULL, NULL);
3035 }
3036
3037 void mlx5e_timestamp_init(struct mlx5e_priv *priv)
3038 {
3039 priv->tstamp.tx_type = HWTSTAMP_TX_OFF;
3040 priv->tstamp.rx_filter = HWTSTAMP_FILTER_NONE;
3041 }
3042
3043 int mlx5e_open_locked(struct net_device *netdev)
3044 {
3045 struct mlx5e_priv *priv = netdev_priv(netdev);
3046 int err;
3047
3048 set_bit(MLX5E_STATE_OPENED, &priv->state);
3049
3050 err = mlx5e_open_channels(priv, &priv->channels);
3051 if (err)
3052 goto err_clear_state_opened_flag;
3053
3054 priv->profile->update_rx(priv);
3055 mlx5e_activate_priv_channels(priv);
3056 if (priv->profile->update_carrier)
3057 priv->profile->update_carrier(priv);
3058
3059 mlx5e_queue_update_stats(priv);
3060 return 0;
3061
3062 err_clear_state_opened_flag:
3063 clear_bit(MLX5E_STATE_OPENED, &priv->state);
3064 return err;
3065 }
3066
3067 int mlx5e_open(struct net_device *netdev)
3068 {
3069 struct mlx5e_priv *priv = netdev_priv(netdev);
3070 int err;
3071
3072 mutex_lock(&priv->state_lock);
3073 err = mlx5e_open_locked(netdev);
3074 if (!err)
3075 mlx5_set_port_admin_status(priv->mdev, MLX5_PORT_UP);
3076 mutex_unlock(&priv->state_lock);
3077
3078 if (mlx5_vxlan_allowed(priv->mdev->vxlan))
3079 udp_tunnel_get_rx_info(netdev);
3080
3081 return err;
3082 }
3083
3084 int mlx5e_close_locked(struct net_device *netdev)
3085 {
3086 struct mlx5e_priv *priv = netdev_priv(netdev);
3087
3088 /* May already be CLOSED in case a previous configuration operation
3089 * (e.g RX/TX queue size change) that involves close&open failed.
3090 */
3091 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
3092 return 0;
3093
3094 clear_bit(MLX5E_STATE_OPENED, &priv->state);
3095
3096 netif_carrier_off(priv->netdev);
3097 mlx5e_deactivate_priv_channels(priv);
3098 mlx5e_close_channels(&priv->channels);
3099
3100 return 0;
3101 }
3102
3103 int mlx5e_close(struct net_device *netdev)
3104 {
3105 struct mlx5e_priv *priv = netdev_priv(netdev);
3106 int err;
3107
3108 if (!netif_device_present(netdev))
3109 return -ENODEV;
3110
3111 mutex_lock(&priv->state_lock);
3112 mlx5_set_port_admin_status(priv->mdev, MLX5_PORT_DOWN);
3113 err = mlx5e_close_locked(netdev);
3114 mutex_unlock(&priv->state_lock);
3115
3116 return err;
3117 }
3118
3119 static int mlx5e_alloc_drop_rq(struct mlx5_core_dev *mdev,
3120 struct mlx5e_rq *rq,
3121 struct mlx5e_rq_param *param)
3122 {
3123 void *rqc = param->rqc;
3124 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
3125 int err;
3126
3127 param->wq.db_numa_node = param->wq.buf_numa_node;
3128
3129 err = mlx5_wq_cyc_create(mdev, &param->wq, rqc_wq, &rq->wqe.wq,
3130 &rq->wq_ctrl);
3131 if (err)
3132 return err;
3133
3134 /* Mark as unused given "Drop-RQ" packets never reach XDP */
3135 xdp_rxq_info_unused(&rq->xdp_rxq);
3136
3137 rq->mdev = mdev;
3138
3139 return 0;
3140 }
3141
3142 static int mlx5e_alloc_drop_cq(struct mlx5_core_dev *mdev,
3143 struct mlx5e_cq *cq,
3144 struct mlx5e_cq_param *param)
3145 {
3146 param->wq.buf_numa_node = dev_to_node(mdev->device);
3147 param->wq.db_numa_node = dev_to_node(mdev->device);
3148
3149 return mlx5e_alloc_cq_common(mdev, param, cq);
3150 }
3151
3152 int mlx5e_open_drop_rq(struct mlx5e_priv *priv,
3153 struct mlx5e_rq *drop_rq)
3154 {
3155 struct mlx5_core_dev *mdev = priv->mdev;
3156 struct mlx5e_cq_param cq_param = {};
3157 struct mlx5e_rq_param rq_param = {};
3158 struct mlx5e_cq *cq = &drop_rq->cq;
3159 int err;
3160
3161 mlx5e_build_drop_rq_param(priv, &rq_param);
3162
3163 err = mlx5e_alloc_drop_cq(mdev, cq, &cq_param);
3164 if (err)
3165 return err;
3166
3167 err = mlx5e_create_cq(cq, &cq_param);
3168 if (err)
3169 goto err_free_cq;
3170
3171 err = mlx5e_alloc_drop_rq(mdev, drop_rq, &rq_param);
3172 if (err)
3173 goto err_destroy_cq;
3174
3175 err = mlx5e_create_rq(drop_rq, &rq_param);
3176 if (err)
3177 goto err_free_rq;
3178
3179 err = mlx5e_modify_rq_state(drop_rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
3180 if (err)
3181 mlx5_core_warn(priv->mdev, "modify_rq_state failed, rx_if_down_packets won't be counted %d\n", err);
3182
3183 return 0;
3184
3185 err_free_rq:
3186 mlx5e_free_rq(drop_rq);
3187
3188 err_destroy_cq:
3189 mlx5e_destroy_cq(cq);
3190
3191 err_free_cq:
3192 mlx5e_free_cq(cq);
3193
3194 return err;
3195 }
3196
3197 void mlx5e_close_drop_rq(struct mlx5e_rq *drop_rq)
3198 {
3199 mlx5e_destroy_rq(drop_rq);
3200 mlx5e_free_rq(drop_rq);
3201 mlx5e_destroy_cq(&drop_rq->cq);
3202 mlx5e_free_cq(&drop_rq->cq);
3203 }
3204
3205 int mlx5e_create_tis(struct mlx5_core_dev *mdev, void *in, u32 *tisn)
3206 {
3207 void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
3208
3209 MLX5_SET(tisc, tisc, transport_domain, mdev->mlx5e_res.td.tdn);
3210
3211 if (MLX5_GET(tisc, tisc, tls_en))
3212 MLX5_SET(tisc, tisc, pd, mdev->mlx5e_res.pdn);
3213
3214 if (mlx5_lag_is_lacp_owner(mdev))
3215 MLX5_SET(tisc, tisc, strict_lag_tx_port_affinity, 1);
3216
3217 return mlx5_core_create_tis(mdev, in, MLX5_ST_SZ_BYTES(create_tis_in), tisn);
3218 }
3219
3220 void mlx5e_destroy_tis(struct mlx5_core_dev *mdev, u32 tisn)
3221 {
3222 mlx5_core_destroy_tis(mdev, tisn);
3223 }
3224
3225 void mlx5e_destroy_tises(struct mlx5e_priv *priv)
3226 {
3227 int tc, i;
3228
3229 for (i = 0; i < mlx5e_get_num_lag_ports(priv->mdev); i++)
3230 for (tc = 0; tc < priv->profile->max_tc; tc++)
3231 mlx5e_destroy_tis(priv->mdev, priv->tisn[i][tc]);
3232 }
3233
3234 static bool mlx5e_lag_should_assign_affinity(struct mlx5_core_dev *mdev)
3235 {
3236 return MLX5_CAP_GEN(mdev, lag_tx_port_affinity) && mlx5e_get_num_lag_ports(mdev) > 1;
3237 }
3238
3239 int mlx5e_create_tises(struct mlx5e_priv *priv)
3240 {
3241 int tc, i;
3242 int err;
3243
3244 for (i = 0; i < mlx5e_get_num_lag_ports(priv->mdev); i++) {
3245 for (tc = 0; tc < priv->profile->max_tc; tc++) {
3246 u32 in[MLX5_ST_SZ_DW(create_tis_in)] = {};
3247 void *tisc;
3248
3249 tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
3250
3251 MLX5_SET(tisc, tisc, prio, tc << 1);
3252
3253 if (mlx5e_lag_should_assign_affinity(priv->mdev))
3254 MLX5_SET(tisc, tisc, lag_tx_port_affinity, i + 1);
3255
3256 err = mlx5e_create_tis(priv->mdev, in, &priv->tisn[i][tc]);
3257 if (err)
3258 goto err_close_tises;
3259 }
3260 }
3261
3262 return 0;
3263
3264 err_close_tises:
3265 for (; i >= 0; i--) {
3266 for (tc--; tc >= 0; tc--)
3267 mlx5e_destroy_tis(priv->mdev, priv->tisn[i][tc]);
3268 tc = priv->profile->max_tc;
3269 }
3270
3271 return err;
3272 }
3273
3274 static void mlx5e_cleanup_nic_tx(struct mlx5e_priv *priv)
3275 {
3276 mlx5e_destroy_tises(priv);
3277 }
3278
3279 static void mlx5e_build_indir_tir_ctx_common(struct mlx5e_priv *priv,
3280 u32 rqtn, u32 *tirc)
3281 {
3282 MLX5_SET(tirc, tirc, transport_domain, priv->mdev->mlx5e_res.td.tdn);
3283 MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT);
3284 MLX5_SET(tirc, tirc, indirect_table, rqtn);
3285 MLX5_SET(tirc, tirc, tunneled_offload_en,
3286 priv->channels.params.tunneled_offload_en);
3287
3288 mlx5e_build_tir_ctx_lro(&priv->channels.params, tirc);
3289 }
3290
3291 static void mlx5e_build_indir_tir_ctx(struct mlx5e_priv *priv,
3292 enum mlx5e_traffic_types tt,
3293 u32 *tirc)
3294 {
3295 mlx5e_build_indir_tir_ctx_common(priv, priv->indir_rqt.rqtn, tirc);
3296 mlx5e_build_indir_tir_ctx_hash(&priv->rss_params,
3297 &tirc_default_config[tt], tirc, false);
3298 }
3299
3300 static void mlx5e_build_direct_tir_ctx(struct mlx5e_priv *priv, u32 rqtn, u32 *tirc)
3301 {
3302 mlx5e_build_indir_tir_ctx_common(priv, rqtn, tirc);
3303 MLX5_SET(tirc, tirc, rx_hash_fn, MLX5_RX_HASH_FN_INVERTED_XOR8);
3304 }
3305
3306 static void mlx5e_build_inner_indir_tir_ctx(struct mlx5e_priv *priv,
3307 enum mlx5e_traffic_types tt,
3308 u32 *tirc)
3309 {
3310 mlx5e_build_indir_tir_ctx_common(priv, priv->indir_rqt.rqtn, tirc);
3311 mlx5e_build_indir_tir_ctx_hash(&priv->rss_params,
3312 &tirc_default_config[tt], tirc, true);
3313 }
3314
3315 int mlx5e_create_indirect_tirs(struct mlx5e_priv *priv, bool inner_ttc)
3316 {
3317 struct mlx5e_tir *tir;
3318 void *tirc;
3319 int inlen;
3320 int i = 0;
3321 int err;
3322 u32 *in;
3323 int tt;
3324
3325 inlen = MLX5_ST_SZ_BYTES(create_tir_in);
3326 in = kvzalloc(inlen, GFP_KERNEL);
3327 if (!in)
3328 return -ENOMEM;
3329
3330 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
3331 memset(in, 0, inlen);
3332 tir = &priv->indir_tir[tt];
3333 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
3334 mlx5e_build_indir_tir_ctx(priv, tt, tirc);
3335 err = mlx5e_create_tir(priv->mdev, tir, in, inlen);
3336 if (err) {
3337 mlx5_core_warn(priv->mdev, "create indirect tirs failed, %d\n", err);
3338 goto err_destroy_inner_tirs;
3339 }
3340 }
3341
3342 if (!inner_ttc || !mlx5e_tunnel_inner_ft_supported(priv->mdev))
3343 goto out;
3344
3345 for (i = 0; i < MLX5E_NUM_INDIR_TIRS; i++) {
3346 memset(in, 0, inlen);
3347 tir = &priv->inner_indir_tir[i];
3348 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
3349 mlx5e_build_inner_indir_tir_ctx(priv, i, tirc);
3350 err = mlx5e_create_tir(priv->mdev, tir, in, inlen);
3351 if (err) {
3352 mlx5_core_warn(priv->mdev, "create inner indirect tirs failed, %d\n", err);
3353 goto err_destroy_inner_tirs;
3354 }
3355 }
3356
3357 out:
3358 kvfree(in);
3359
3360 return 0;
3361
3362 err_destroy_inner_tirs:
3363 for (i--; i >= 0; i--)
3364 mlx5e_destroy_tir(priv->mdev, &priv->inner_indir_tir[i]);
3365
3366 for (tt--; tt >= 0; tt--)
3367 mlx5e_destroy_tir(priv->mdev, &priv->indir_tir[tt]);
3368
3369 kvfree(in);
3370
3371 return err;
3372 }
3373
3374 int mlx5e_create_direct_tirs(struct mlx5e_priv *priv, struct mlx5e_tir *tirs)
3375 {
3376 struct mlx5e_tir *tir;
3377 void *tirc;
3378 int inlen;
3379 int err = 0;
3380 u32 *in;
3381 int ix;
3382
3383 inlen = MLX5_ST_SZ_BYTES(create_tir_in);
3384 in = kvzalloc(inlen, GFP_KERNEL);
3385 if (!in)
3386 return -ENOMEM;
3387
3388 for (ix = 0; ix < priv->max_nch; ix++) {
3389 memset(in, 0, inlen);
3390 tir = &tirs[ix];
3391 tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
3392 mlx5e_build_direct_tir_ctx(priv, tir->rqt.rqtn, tirc);
3393 err = mlx5e_create_tir(priv->mdev, tir, in, inlen);
3394 if (unlikely(err))
3395 goto err_destroy_ch_tirs;
3396 }
3397
3398 goto out;
3399
3400 err_destroy_ch_tirs:
3401 mlx5_core_warn(priv->mdev, "create tirs failed, %d\n", err);
3402 for (ix--; ix >= 0; ix--)
3403 mlx5e_destroy_tir(priv->mdev, &tirs[ix]);
3404
3405 out:
3406 kvfree(in);
3407
3408 return err;
3409 }
3410
3411 void mlx5e_destroy_indirect_tirs(struct mlx5e_priv *priv, bool inner_ttc)
3412 {
3413 int i;
3414
3415 for (i = 0; i < MLX5E_NUM_INDIR_TIRS; i++)
3416 mlx5e_destroy_tir(priv->mdev, &priv->indir_tir[i]);
3417
3418 if (!inner_ttc || !mlx5e_tunnel_inner_ft_supported(priv->mdev))
3419 return;
3420
3421 for (i = 0; i < MLX5E_NUM_INDIR_TIRS; i++)
3422 mlx5e_destroy_tir(priv->mdev, &priv->inner_indir_tir[i]);
3423 }
3424
3425 void mlx5e_destroy_direct_tirs(struct mlx5e_priv *priv, struct mlx5e_tir *tirs)
3426 {
3427 int i;
3428
3429 for (i = 0; i < priv->max_nch; i++)
3430 mlx5e_destroy_tir(priv->mdev, &tirs[i]);
3431 }
3432
3433 static int mlx5e_modify_channels_scatter_fcs(struct mlx5e_channels *chs, bool enable)
3434 {
3435 int err = 0;
3436 int i;
3437
3438 for (i = 0; i < chs->num; i++) {
3439 err = mlx5e_modify_rq_scatter_fcs(&chs->c[i]->rq, enable);
3440 if (err)
3441 return err;
3442 }
3443
3444 return 0;
3445 }
3446
3447 static int mlx5e_modify_channels_vsd(struct mlx5e_channels *chs, bool vsd)
3448 {
3449 int err = 0;
3450 int i;
3451
3452 for (i = 0; i < chs->num; i++) {
3453 err = mlx5e_modify_rq_vsd(&chs->c[i]->rq, vsd);
3454 if (err)
3455 return err;
3456 }
3457
3458 return 0;
3459 }
3460
3461 static int mlx5e_setup_tc_mqprio(struct mlx5e_priv *priv,
3462 struct tc_mqprio_qopt *mqprio)
3463 {
3464 struct mlx5e_channels new_channels = {};
3465 u8 tc = mqprio->num_tc;
3466 int err = 0;
3467
3468 mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
3469
3470 if (tc && tc != MLX5E_MAX_NUM_TC)
3471 return -EINVAL;
3472
3473 mutex_lock(&priv->state_lock);
3474
3475 new_channels.params = priv->channels.params;
3476 new_channels.params.num_tc = tc ? tc : 1;
3477
3478 if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
3479 priv->channels.params = new_channels.params;
3480 goto out;
3481 }
3482
3483 err = mlx5e_safe_switch_channels(priv, &new_channels,
3484 mlx5e_num_channels_changed_ctx, NULL);
3485 if (err)
3486 goto out;
3487
3488 priv->max_opened_tc = max_t(u8, priv->max_opened_tc,
3489 new_channels.params.num_tc);
3490 out:
3491 mutex_unlock(&priv->state_lock);
3492 return err;
3493 }
3494
3495 #ifdef CONFIG_MLX5_ESWITCH
3496 static int mlx5e_setup_tc_cls_flower(struct mlx5e_priv *priv,
3497 struct flow_cls_offload *cls_flower,
3498 unsigned long flags)
3499 {
3500 switch (cls_flower->command) {
3501 case FLOW_CLS_REPLACE:
3502 return mlx5e_configure_flower(priv->netdev, priv, cls_flower,
3503 flags);
3504 case FLOW_CLS_DESTROY:
3505 return mlx5e_delete_flower(priv->netdev, priv, cls_flower,
3506 flags);
3507 case FLOW_CLS_STATS:
3508 return mlx5e_stats_flower(priv->netdev, priv, cls_flower,
3509 flags);
3510 default:
3511 return -EOPNOTSUPP;
3512 }
3513 }
3514
3515 static int mlx5e_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
3516 void *cb_priv)
3517 {
3518 unsigned long flags = MLX5_TC_FLAG(INGRESS) | MLX5_TC_FLAG(NIC_OFFLOAD);
3519 struct mlx5e_priv *priv = cb_priv;
3520
3521 switch (type) {
3522 case TC_SETUP_CLSFLOWER:
3523 return mlx5e_setup_tc_cls_flower(priv, type_data, flags);
3524 default:
3525 return -EOPNOTSUPP;
3526 }
3527 }
3528 #endif
3529
3530 static LIST_HEAD(mlx5e_block_cb_list);
3531
3532 static int mlx5e_setup_tc(struct net_device *dev, enum tc_setup_type type,
3533 void *type_data)
3534 {
3535 struct mlx5e_priv *priv = netdev_priv(dev);
3536
3537 switch (type) {
3538 #ifdef CONFIG_MLX5_ESWITCH
3539 case TC_SETUP_BLOCK: {
3540 struct flow_block_offload *f = type_data;
3541
3542 f->unlocked_driver_cb = true;
3543 return flow_block_cb_setup_simple(type_data,
3544 &mlx5e_block_cb_list,
3545 mlx5e_setup_tc_block_cb,
3546 priv, priv, true);
3547 }
3548 #endif
3549 case TC_SETUP_QDISC_MQPRIO:
3550 return mlx5e_setup_tc_mqprio(priv, type_data);
3551 default:
3552 return -EOPNOTSUPP;
3553 }
3554 }
3555
3556 void mlx5e_fold_sw_stats64(struct mlx5e_priv *priv, struct rtnl_link_stats64 *s)
3557 {
3558 int i;
3559
3560 for (i = 0; i < priv->max_nch; i++) {
3561 struct mlx5e_channel_stats *channel_stats = &priv->channel_stats[i];
3562 struct mlx5e_rq_stats *xskrq_stats = &channel_stats->xskrq;
3563 struct mlx5e_rq_stats *rq_stats = &channel_stats->rq;
3564 int j;
3565
3566 s->rx_packets += rq_stats->packets + xskrq_stats->packets;
3567 s->rx_bytes += rq_stats->bytes + xskrq_stats->bytes;
3568
3569 for (j = 0; j < priv->max_opened_tc; j++) {
3570 struct mlx5e_sq_stats *sq_stats = &channel_stats->sq[j];
3571
3572 s->tx_packets += sq_stats->packets;
3573 s->tx_bytes += sq_stats->bytes;
3574 s->tx_dropped += sq_stats->dropped;
3575 }
3576 }
3577 }
3578
3579 void
3580 mlx5e_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats)
3581 {
3582 struct mlx5e_priv *priv = netdev_priv(dev);
3583 struct mlx5e_vport_stats *vstats = &priv->stats.vport;
3584 struct mlx5e_pport_stats *pstats = &priv->stats.pport;
3585
3586 if (!mlx5e_monitor_counter_supported(priv)) {
3587 /* update HW stats in background for next time */
3588 mlx5e_queue_update_stats(priv);
3589 }
3590
3591 if (mlx5e_is_uplink_rep(priv)) {
3592 stats->rx_packets = PPORT_802_3_GET(pstats, a_frames_received_ok);
3593 stats->rx_bytes = PPORT_802_3_GET(pstats, a_octets_received_ok);
3594 stats->tx_packets = PPORT_802_3_GET(pstats, a_frames_transmitted_ok);
3595 stats->tx_bytes = PPORT_802_3_GET(pstats, a_octets_transmitted_ok);
3596 } else {
3597 mlx5e_fold_sw_stats64(priv, stats);
3598 }
3599
3600 stats->rx_dropped = priv->stats.qcnt.rx_out_of_buffer;
3601
3602 stats->rx_length_errors =
3603 PPORT_802_3_GET(pstats, a_in_range_length_errors) +
3604 PPORT_802_3_GET(pstats, a_out_of_range_length_field) +
3605 PPORT_802_3_GET(pstats, a_frame_too_long_errors);
3606 stats->rx_crc_errors =
3607 PPORT_802_3_GET(pstats, a_frame_check_sequence_errors);
3608 stats->rx_frame_errors = PPORT_802_3_GET(pstats, a_alignment_errors);
3609 stats->tx_aborted_errors = PPORT_2863_GET(pstats, if_out_discards);
3610 stats->rx_errors = stats->rx_length_errors + stats->rx_crc_errors +
3611 stats->rx_frame_errors;
3612 stats->tx_errors = stats->tx_aborted_errors + stats->tx_carrier_errors;
3613
3614 /* vport multicast also counts packets that are dropped due to steering
3615 * or rx out of buffer
3616 */
3617 stats->multicast =
3618 VPORT_COUNTER_GET(vstats, received_eth_multicast.packets);
3619 }
3620
3621 static void mlx5e_set_rx_mode(struct net_device *dev)
3622 {
3623 struct mlx5e_priv *priv = netdev_priv(dev);
3624
3625 queue_work(priv->wq, &priv->set_rx_mode_work);
3626 }
3627
3628 static int mlx5e_set_mac(struct net_device *netdev, void *addr)
3629 {
3630 struct mlx5e_priv *priv = netdev_priv(netdev);
3631 struct sockaddr *saddr = addr;
3632
3633 if (!is_valid_ether_addr(saddr->sa_data))
3634 return -EADDRNOTAVAIL;
3635
3636 netif_addr_lock_bh(netdev);
3637 ether_addr_copy(netdev->dev_addr, saddr->sa_data);
3638 netif_addr_unlock_bh(netdev);
3639
3640 queue_work(priv->wq, &priv->set_rx_mode_work);
3641
3642 return 0;
3643 }
3644
3645 #define MLX5E_SET_FEATURE(features, feature, enable) \
3646 do { \
3647 if (enable) \
3648 *features |= feature; \
3649 else \
3650 *features &= ~feature; \
3651 } while (0)
3652
3653 typedef int (*mlx5e_feature_handler)(struct net_device *netdev, bool enable);
3654
3655 static int set_feature_lro(struct net_device *netdev, bool enable)
3656 {
3657 struct mlx5e_priv *priv = netdev_priv(netdev);
3658 struct mlx5_core_dev *mdev = priv->mdev;
3659 struct mlx5e_channels new_channels = {};
3660 struct mlx5e_params *old_params;
3661 int err = 0;
3662 bool reset;
3663
3664 mutex_lock(&priv->state_lock);
3665
3666 if (enable && priv->xsk.refcnt) {
3667 netdev_warn(netdev, "LRO is incompatible with AF_XDP (%hu XSKs are active)\n",
3668 priv->xsk.refcnt);
3669 err = -EINVAL;
3670 goto out;
3671 }
3672
3673 old_params = &priv->channels.params;
3674 if (enable && !MLX5E_GET_PFLAG(old_params, MLX5E_PFLAG_RX_STRIDING_RQ)) {
3675 netdev_warn(netdev, "can't set LRO with legacy RQ\n");
3676 err = -EINVAL;
3677 goto out;
3678 }
3679
3680 reset = test_bit(MLX5E_STATE_OPENED, &priv->state);
3681
3682 new_channels.params = *old_params;
3683 new_channels.params.lro_en = enable;
3684
3685 if (old_params->rq_wq_type != MLX5_WQ_TYPE_CYCLIC) {
3686 if (mlx5e_rx_mpwqe_is_linear_skb(mdev, old_params, NULL) ==
3687 mlx5e_rx_mpwqe_is_linear_skb(mdev, &new_channels.params, NULL))
3688 reset = false;
3689 }
3690
3691 if (!reset) {
3692 *old_params = new_channels.params;
3693 err = mlx5e_modify_tirs_lro(priv);
3694 goto out;
3695 }
3696
3697 err = mlx5e_safe_switch_channels(priv, &new_channels,
3698 mlx5e_modify_tirs_lro_ctx, NULL);
3699 out:
3700 mutex_unlock(&priv->state_lock);
3701 return err;
3702 }
3703
3704 static int set_feature_cvlan_filter(struct net_device *netdev, bool enable)
3705 {
3706 struct mlx5e_priv *priv = netdev_priv(netdev);
3707
3708 if (enable)
3709 mlx5e_enable_cvlan_filter(priv);
3710 else
3711 mlx5e_disable_cvlan_filter(priv);
3712
3713 return 0;
3714 }
3715
3716 #ifdef CONFIG_MLX5_ESWITCH
3717 static int set_feature_tc_num_filters(struct net_device *netdev, bool enable)
3718 {
3719 struct mlx5e_priv *priv = netdev_priv(netdev);
3720
3721 if (!enable && mlx5e_tc_num_filters(priv, MLX5_TC_FLAG(NIC_OFFLOAD))) {
3722 netdev_err(netdev,
3723 "Active offloaded tc filters, can't turn hw_tc_offload off\n");
3724 return -EINVAL;
3725 }
3726
3727 return 0;
3728 }
3729 #endif
3730
3731 static int set_feature_rx_all(struct net_device *netdev, bool enable)
3732 {
3733 struct mlx5e_priv *priv = netdev_priv(netdev);
3734 struct mlx5_core_dev *mdev = priv->mdev;
3735
3736 return mlx5_set_port_fcs(mdev, !enable);
3737 }
3738
3739 static int set_feature_rx_fcs(struct net_device *netdev, bool enable)
3740 {
3741 struct mlx5e_priv *priv = netdev_priv(netdev);
3742 int err;
3743
3744 mutex_lock(&priv->state_lock);
3745
3746 priv->channels.params.scatter_fcs_en = enable;
3747 err = mlx5e_modify_channels_scatter_fcs(&priv->channels, enable);
3748 if (err)
3749 priv->channels.params.scatter_fcs_en = !enable;
3750
3751 mutex_unlock(&priv->state_lock);
3752
3753 return err;
3754 }
3755
3756 static int set_feature_rx_vlan(struct net_device *netdev, bool enable)
3757 {
3758 struct mlx5e_priv *priv = netdev_priv(netdev);
3759 int err = 0;
3760
3761 mutex_lock(&priv->state_lock);
3762
3763 priv->channels.params.vlan_strip_disable = !enable;
3764 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
3765 goto unlock;
3766
3767 err = mlx5e_modify_channels_vsd(&priv->channels, !enable);
3768 if (err)
3769 priv->channels.params.vlan_strip_disable = enable;
3770
3771 unlock:
3772 mutex_unlock(&priv->state_lock);
3773
3774 return err;
3775 }
3776
3777 #ifdef CONFIG_MLX5_EN_ARFS
3778 static int set_feature_arfs(struct net_device *netdev, bool enable)
3779 {
3780 struct mlx5e_priv *priv = netdev_priv(netdev);
3781 int err;
3782
3783 if (enable)
3784 err = mlx5e_arfs_enable(priv);
3785 else
3786 err = mlx5e_arfs_disable(priv);
3787
3788 return err;
3789 }
3790 #endif
3791
3792 static int mlx5e_handle_feature(struct net_device *netdev,
3793 netdev_features_t *features,
3794 netdev_features_t wanted_features,
3795 netdev_features_t feature,
3796 mlx5e_feature_handler feature_handler)
3797 {
3798 netdev_features_t changes = wanted_features ^ netdev->features;
3799 bool enable = !!(wanted_features & feature);
3800 int err;
3801
3802 if (!(changes & feature))
3803 return 0;
3804
3805 err = feature_handler(netdev, enable);
3806 if (err) {
3807 netdev_err(netdev, "%s feature %pNF failed, err %d\n",
3808 enable ? "Enable" : "Disable", &feature, err);
3809 return err;
3810 }
3811
3812 MLX5E_SET_FEATURE(features, feature, enable);
3813 return 0;
3814 }
3815
3816 int mlx5e_set_features(struct net_device *netdev, netdev_features_t features)
3817 {
3818 netdev_features_t oper_features = netdev->features;
3819 int err = 0;
3820
3821 #define MLX5E_HANDLE_FEATURE(feature, handler) \
3822 mlx5e_handle_feature(netdev, &oper_features, features, feature, handler)
3823
3824 err |= MLX5E_HANDLE_FEATURE(NETIF_F_LRO, set_feature_lro);
3825 err |= MLX5E_HANDLE_FEATURE(NETIF_F_HW_VLAN_CTAG_FILTER,
3826 set_feature_cvlan_filter);
3827 #ifdef CONFIG_MLX5_ESWITCH
3828 err |= MLX5E_HANDLE_FEATURE(NETIF_F_HW_TC, set_feature_tc_num_filters);
3829 #endif
3830 err |= MLX5E_HANDLE_FEATURE(NETIF_F_RXALL, set_feature_rx_all);
3831 err |= MLX5E_HANDLE_FEATURE(NETIF_F_RXFCS, set_feature_rx_fcs);
3832 err |= MLX5E_HANDLE_FEATURE(NETIF_F_HW_VLAN_CTAG_RX, set_feature_rx_vlan);
3833 #ifdef CONFIG_MLX5_EN_ARFS
3834 err |= MLX5E_HANDLE_FEATURE(NETIF_F_NTUPLE, set_feature_arfs);
3835 #endif
3836
3837 if (err) {
3838 netdev->features = oper_features;
3839 return -EINVAL;
3840 }
3841
3842 return 0;
3843 }
3844
3845 static netdev_features_t mlx5e_fix_features(struct net_device *netdev,
3846 netdev_features_t features)
3847 {
3848 struct mlx5e_priv *priv = netdev_priv(netdev);
3849 struct mlx5e_params *params;
3850
3851 mutex_lock(&priv->state_lock);
3852 params = &priv->channels.params;
3853 if (!bitmap_empty(priv->fs.vlan.active_svlans, VLAN_N_VID)) {
3854 /* HW strips the outer C-tag header, this is a problem
3855 * for S-tag traffic.
3856 */
3857 features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3858 if (!params->vlan_strip_disable)
3859 netdev_warn(netdev, "Dropping C-tag vlan stripping offload due to S-tag vlan\n");
3860 }
3861 if (!MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_STRIDING_RQ)) {
3862 if (features & NETIF_F_LRO) {
3863 netdev_warn(netdev, "Disabling LRO, not supported in legacy RQ\n");
3864 features &= ~NETIF_F_LRO;
3865 }
3866 }
3867
3868 if (MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS)) {
3869 features &= ~NETIF_F_RXHASH;
3870 if (netdev->features & NETIF_F_RXHASH)
3871 netdev_warn(netdev, "Disabling rxhash, not supported when CQE compress is active\n");
3872 }
3873
3874 mutex_unlock(&priv->state_lock);
3875
3876 return features;
3877 }
3878
3879 static bool mlx5e_xsk_validate_mtu(struct net_device *netdev,
3880 struct mlx5e_channels *chs,
3881 struct mlx5e_params *new_params,
3882 struct mlx5_core_dev *mdev)
3883 {
3884 u16 ix;
3885
3886 for (ix = 0; ix < chs->params.num_channels; ix++) {
3887 struct xdp_umem *umem = mlx5e_xsk_get_umem(&chs->params, chs->params.xsk, ix);
3888 struct mlx5e_xsk_param xsk;
3889
3890 if (!umem)
3891 continue;
3892
3893 mlx5e_build_xsk_param(umem, &xsk);
3894
3895 if (!mlx5e_validate_xsk_param(new_params, &xsk, mdev)) {
3896 u32 hr = mlx5e_get_linear_rq_headroom(new_params, &xsk);
3897 int max_mtu_frame, max_mtu_page, max_mtu;
3898
3899 /* Two criteria must be met:
3900 * 1. HW MTU + all headrooms <= XSK frame size.
3901 * 2. Size of SKBs allocated on XDP_PASS <= PAGE_SIZE.
3902 */
3903 max_mtu_frame = MLX5E_HW2SW_MTU(new_params, xsk.chunk_size - hr);
3904 max_mtu_page = mlx5e_xdp_max_mtu(new_params, &xsk);
3905 max_mtu = min(max_mtu_frame, max_mtu_page);
3906
3907 netdev_err(netdev, "MTU %d is too big for an XSK running on channel %hu. Try MTU <= %d\n",
3908 new_params->sw_mtu, ix, max_mtu);
3909 return false;
3910 }
3911 }
3912
3913 return true;
3914 }
3915
3916 int mlx5e_change_mtu(struct net_device *netdev, int new_mtu,
3917 mlx5e_fp_preactivate preactivate)
3918 {
3919 struct mlx5e_priv *priv = netdev_priv(netdev);
3920 struct mlx5e_channels new_channels = {};
3921 struct mlx5e_params *params;
3922 int err = 0;
3923 bool reset;
3924
3925 mutex_lock(&priv->state_lock);
3926
3927 params = &priv->channels.params;
3928
3929 reset = !params->lro_en;
3930 reset = reset && test_bit(MLX5E_STATE_OPENED, &priv->state);
3931
3932 new_channels.params = *params;
3933 new_channels.params.sw_mtu = new_mtu;
3934
3935 if (params->xdp_prog &&
3936 !mlx5e_rx_is_linear_skb(&new_channels.params, NULL)) {
3937 netdev_err(netdev, "MTU(%d) > %d is not allowed while XDP enabled\n",
3938 new_mtu, mlx5e_xdp_max_mtu(params, NULL));
3939 err = -EINVAL;
3940 goto out;
3941 }
3942
3943 if (priv->xsk.refcnt &&
3944 !mlx5e_xsk_validate_mtu(netdev, &priv->channels,
3945 &new_channels.params, priv->mdev)) {
3946 err = -EINVAL;
3947 goto out;
3948 }
3949
3950 if (params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
3951 bool is_linear = mlx5e_rx_mpwqe_is_linear_skb(priv->mdev,
3952 &new_channels.params,
3953 NULL);
3954 u8 ppw_old = mlx5e_mpwqe_log_pkts_per_wqe(params, NULL);
3955 u8 ppw_new = mlx5e_mpwqe_log_pkts_per_wqe(&new_channels.params, NULL);
3956
3957 /* If XSK is active, XSK RQs are linear. */
3958 is_linear |= priv->xsk.refcnt;
3959
3960 /* Always reset in linear mode - hw_mtu is used in data path. */
3961 reset = reset && (is_linear || (ppw_old != ppw_new));
3962 }
3963
3964 if (!reset) {
3965 params->sw_mtu = new_mtu;
3966 if (preactivate)
3967 preactivate(priv, NULL);
3968 netdev->mtu = params->sw_mtu;
3969 goto out;
3970 }
3971
3972 err = mlx5e_safe_switch_channels(priv, &new_channels, preactivate, NULL);
3973 if (err)
3974 goto out;
3975
3976 netdev->mtu = new_channels.params.sw_mtu;
3977
3978 out:
3979 mutex_unlock(&priv->state_lock);
3980 return err;
3981 }
3982
3983 static int mlx5e_change_nic_mtu(struct net_device *netdev, int new_mtu)
3984 {
3985 return mlx5e_change_mtu(netdev, new_mtu, mlx5e_set_dev_port_mtu_ctx);
3986 }
3987
3988 int mlx5e_hwstamp_set(struct mlx5e_priv *priv, struct ifreq *ifr)
3989 {
3990 struct hwtstamp_config config;
3991 int err;
3992
3993 if (!MLX5_CAP_GEN(priv->mdev, device_frequency_khz) ||
3994 (mlx5_clock_get_ptp_index(priv->mdev) == -1))
3995 return -EOPNOTSUPP;
3996
3997 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
3998 return -EFAULT;
3999
4000 /* TX HW timestamp */
4001 switch (config.tx_type) {
4002 case HWTSTAMP_TX_OFF:
4003 case HWTSTAMP_TX_ON:
4004 break;
4005 default:
4006 return -ERANGE;
4007 }
4008
4009 mutex_lock(&priv->state_lock);
4010 /* RX HW timestamp */
4011 switch (config.rx_filter) {
4012 case HWTSTAMP_FILTER_NONE:
4013 /* Reset CQE compression to Admin default */
4014 mlx5e_modify_rx_cqe_compression_locked(priv, priv->channels.params.rx_cqe_compress_def);
4015 break;
4016 case HWTSTAMP_FILTER_ALL:
4017 case HWTSTAMP_FILTER_SOME:
4018 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4019 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4020 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4021 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4022 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4023 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4024 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4025 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4026 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4027 case HWTSTAMP_FILTER_PTP_V2_EVENT:
4028 case HWTSTAMP_FILTER_PTP_V2_SYNC:
4029 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4030 case HWTSTAMP_FILTER_NTP_ALL:
4031 /* Disable CQE compression */
4032 if (MLX5E_GET_PFLAG(&priv->channels.params, MLX5E_PFLAG_RX_CQE_COMPRESS))
4033 netdev_warn(priv->netdev, "Disabling RX cqe compression\n");
4034 err = mlx5e_modify_rx_cqe_compression_locked(priv, false);
4035 if (err) {
4036 netdev_err(priv->netdev, "Failed disabling cqe compression err=%d\n", err);
4037 mutex_unlock(&priv->state_lock);
4038 return err;
4039 }
4040 config.rx_filter = HWTSTAMP_FILTER_ALL;
4041 break;
4042 default:
4043 mutex_unlock(&priv->state_lock);
4044 return -ERANGE;
4045 }
4046
4047 memcpy(&priv->tstamp, &config, sizeof(config));
4048 mutex_unlock(&priv->state_lock);
4049
4050 /* might need to fix some features */
4051 netdev_update_features(priv->netdev);
4052
4053 return copy_to_user(ifr->ifr_data, &config,
4054 sizeof(config)) ? -EFAULT : 0;
4055 }
4056
4057 int mlx5e_hwstamp_get(struct mlx5e_priv *priv, struct ifreq *ifr)
4058 {
4059 struct hwtstamp_config *cfg = &priv->tstamp;
4060
4061 if (!MLX5_CAP_GEN(priv->mdev, device_frequency_khz))
4062 return -EOPNOTSUPP;
4063
4064 return copy_to_user(ifr->ifr_data, cfg, sizeof(*cfg)) ? -EFAULT : 0;
4065 }
4066
4067 static int mlx5e_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
4068 {
4069 struct mlx5e_priv *priv = netdev_priv(dev);
4070
4071 switch (cmd) {
4072 case SIOCSHWTSTAMP:
4073 return mlx5e_hwstamp_set(priv, ifr);
4074 case SIOCGHWTSTAMP:
4075 return mlx5e_hwstamp_get(priv, ifr);
4076 default:
4077 return -EOPNOTSUPP;
4078 }
4079 }
4080
4081 #ifdef CONFIG_MLX5_ESWITCH
4082 int mlx5e_set_vf_mac(struct net_device *dev, int vf, u8 *mac)
4083 {
4084 struct mlx5e_priv *priv = netdev_priv(dev);
4085 struct mlx5_core_dev *mdev = priv->mdev;
4086
4087 return mlx5_eswitch_set_vport_mac(mdev->priv.eswitch, vf + 1, mac);
4088 }
4089
4090 static int mlx5e_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
4091 __be16 vlan_proto)
4092 {
4093 struct mlx5e_priv *priv = netdev_priv(dev);
4094 struct mlx5_core_dev *mdev = priv->mdev;
4095
4096 if (vlan_proto != htons(ETH_P_8021Q))
4097 return -EPROTONOSUPPORT;
4098
4099 return mlx5_eswitch_set_vport_vlan(mdev->priv.eswitch, vf + 1,
4100 vlan, qos);
4101 }
4102
4103 static int mlx5e_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
4104 {
4105 struct mlx5e_priv *priv = netdev_priv(dev);
4106 struct mlx5_core_dev *mdev = priv->mdev;
4107
4108 return mlx5_eswitch_set_vport_spoofchk(mdev->priv.eswitch, vf + 1, setting);
4109 }
4110
4111 static int mlx5e_set_vf_trust(struct net_device *dev, int vf, bool setting)
4112 {
4113 struct mlx5e_priv *priv = netdev_priv(dev);
4114 struct mlx5_core_dev *mdev = priv->mdev;
4115
4116 return mlx5_eswitch_set_vport_trust(mdev->priv.eswitch, vf + 1, setting);
4117 }
4118
4119 int mlx5e_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
4120 int max_tx_rate)
4121 {
4122 struct mlx5e_priv *priv = netdev_priv(dev);
4123 struct mlx5_core_dev *mdev = priv->mdev;
4124
4125 return mlx5_eswitch_set_vport_rate(mdev->priv.eswitch, vf + 1,
4126 max_tx_rate, min_tx_rate);
4127 }
4128
4129 static int mlx5_vport_link2ifla(u8 esw_link)
4130 {
4131 switch (esw_link) {
4132 case MLX5_VPORT_ADMIN_STATE_DOWN:
4133 return IFLA_VF_LINK_STATE_DISABLE;
4134 case MLX5_VPORT_ADMIN_STATE_UP:
4135 return IFLA_VF_LINK_STATE_ENABLE;
4136 }
4137 return IFLA_VF_LINK_STATE_AUTO;
4138 }
4139
4140 static int mlx5_ifla_link2vport(u8 ifla_link)
4141 {
4142 switch (ifla_link) {
4143 case IFLA_VF_LINK_STATE_DISABLE:
4144 return MLX5_VPORT_ADMIN_STATE_DOWN;
4145 case IFLA_VF_LINK_STATE_ENABLE:
4146 return MLX5_VPORT_ADMIN_STATE_UP;
4147 }
4148 return MLX5_VPORT_ADMIN_STATE_AUTO;
4149 }
4150
4151 static int mlx5e_set_vf_link_state(struct net_device *dev, int vf,
4152 int link_state)
4153 {
4154 struct mlx5e_priv *priv = netdev_priv(dev);
4155 struct mlx5_core_dev *mdev = priv->mdev;
4156
4157 return mlx5_eswitch_set_vport_state(mdev->priv.eswitch, vf + 1,
4158 mlx5_ifla_link2vport(link_state));
4159 }
4160
4161 int mlx5e_get_vf_config(struct net_device *dev,
4162 int vf, struct ifla_vf_info *ivi)
4163 {
4164 struct mlx5e_priv *priv = netdev_priv(dev);
4165 struct mlx5_core_dev *mdev = priv->mdev;
4166 int err;
4167
4168 err = mlx5_eswitch_get_vport_config(mdev->priv.eswitch, vf + 1, ivi);
4169 if (err)
4170 return err;
4171 ivi->linkstate = mlx5_vport_link2ifla(ivi->linkstate);
4172 return 0;
4173 }
4174
4175 int mlx5e_get_vf_stats(struct net_device *dev,
4176 int vf, struct ifla_vf_stats *vf_stats)
4177 {
4178 struct mlx5e_priv *priv = netdev_priv(dev);
4179 struct mlx5_core_dev *mdev = priv->mdev;
4180
4181 return mlx5_eswitch_get_vport_stats(mdev->priv.eswitch, vf + 1,
4182 vf_stats);
4183 }
4184 #endif
4185
4186 struct mlx5e_vxlan_work {
4187 struct work_struct work;
4188 struct mlx5e_priv *priv;
4189 u16 port;
4190 };
4191
4192 static void mlx5e_vxlan_add_work(struct work_struct *work)
4193 {
4194 struct mlx5e_vxlan_work *vxlan_work =
4195 container_of(work, struct mlx5e_vxlan_work, work);
4196 struct mlx5e_priv *priv = vxlan_work->priv;
4197 u16 port = vxlan_work->port;
4198
4199 mutex_lock(&priv->state_lock);
4200 mlx5_vxlan_add_port(priv->mdev->vxlan, port);
4201 mutex_unlock(&priv->state_lock);
4202
4203 kfree(vxlan_work);
4204 }
4205
4206 static void mlx5e_vxlan_del_work(struct work_struct *work)
4207 {
4208 struct mlx5e_vxlan_work *vxlan_work =
4209 container_of(work, struct mlx5e_vxlan_work, work);
4210 struct mlx5e_priv *priv = vxlan_work->priv;
4211 u16 port = vxlan_work->port;
4212
4213 mutex_lock(&priv->state_lock);
4214 mlx5_vxlan_del_port(priv->mdev->vxlan, port);
4215 mutex_unlock(&priv->state_lock);
4216 kfree(vxlan_work);
4217 }
4218
4219 static void mlx5e_vxlan_queue_work(struct mlx5e_priv *priv, u16 port, int add)
4220 {
4221 struct mlx5e_vxlan_work *vxlan_work;
4222
4223 vxlan_work = kmalloc(sizeof(*vxlan_work), GFP_ATOMIC);
4224 if (!vxlan_work)
4225 return;
4226
4227 if (add)
4228 INIT_WORK(&vxlan_work->work, mlx5e_vxlan_add_work);
4229 else
4230 INIT_WORK(&vxlan_work->work, mlx5e_vxlan_del_work);
4231
4232 vxlan_work->priv = priv;
4233 vxlan_work->port = port;
4234 queue_work(priv->wq, &vxlan_work->work);
4235 }
4236
4237 void mlx5e_add_vxlan_port(struct net_device *netdev, struct udp_tunnel_info *ti)
4238 {
4239 struct mlx5e_priv *priv = netdev_priv(netdev);
4240
4241 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
4242 return;
4243
4244 if (!mlx5_vxlan_allowed(priv->mdev->vxlan))
4245 return;
4246
4247 mlx5e_vxlan_queue_work(priv, be16_to_cpu(ti->port), 1);
4248 }
4249
4250 void mlx5e_del_vxlan_port(struct net_device *netdev, struct udp_tunnel_info *ti)
4251 {
4252 struct mlx5e_priv *priv = netdev_priv(netdev);
4253
4254 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
4255 return;
4256
4257 if (!mlx5_vxlan_allowed(priv->mdev->vxlan))
4258 return;
4259
4260 mlx5e_vxlan_queue_work(priv, be16_to_cpu(ti->port), 0);
4261 }
4262
4263 static netdev_features_t mlx5e_tunnel_features_check(struct mlx5e_priv *priv,
4264 struct sk_buff *skb,
4265 netdev_features_t features)
4266 {
4267 unsigned int offset = 0;
4268 struct udphdr *udph;
4269 u8 proto;
4270 u16 port;
4271
4272 switch (vlan_get_protocol(skb)) {
4273 case htons(ETH_P_IP):
4274 proto = ip_hdr(skb)->protocol;
4275 break;
4276 case htons(ETH_P_IPV6):
4277 proto = ipv6_find_hdr(skb, &offset, -1, NULL, NULL);
4278 break;
4279 default:
4280 goto out;
4281 }
4282
4283 switch (proto) {
4284 case IPPROTO_GRE:
4285 return features;
4286 case IPPROTO_IPIP:
4287 case IPPROTO_IPV6:
4288 if (mlx5e_tunnel_proto_supported(priv->mdev, IPPROTO_IPIP))
4289 return features;
4290 break;
4291 case IPPROTO_UDP:
4292 udph = udp_hdr(skb);
4293 port = be16_to_cpu(udph->dest);
4294
4295 /* Verify if UDP port is being offloaded by HW */
4296 if (mlx5_vxlan_lookup_port(priv->mdev->vxlan, port))
4297 return features;
4298
4299 #if IS_ENABLED(CONFIG_GENEVE)
4300 /* Support Geneve offload for default UDP port */
4301 if (port == GENEVE_UDP_PORT && mlx5_geneve_tx_allowed(priv->mdev))
4302 return features;
4303 #endif
4304 }
4305
4306 out:
4307 /* Disable CSUM and GSO if the udp dport is not offloaded by HW */
4308 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
4309 }
4310
4311 netdev_features_t mlx5e_features_check(struct sk_buff *skb,
4312 struct net_device *netdev,
4313 netdev_features_t features)
4314 {
4315 struct mlx5e_priv *priv = netdev_priv(netdev);
4316
4317 features = vlan_features_check(skb, features);
4318 features = vxlan_features_check(skb, features);
4319
4320 #ifdef CONFIG_MLX5_EN_IPSEC
4321 if (mlx5e_ipsec_feature_check(skb, netdev, features))
4322 return features;
4323 #endif
4324
4325 /* Validate if the tunneled packet is being offloaded by HW */
4326 if (skb->encapsulation &&
4327 (features & NETIF_F_CSUM_MASK || features & NETIF_F_GSO_MASK))
4328 return mlx5e_tunnel_features_check(priv, skb, features);
4329
4330 return features;
4331 }
4332
4333 static void mlx5e_tx_timeout_work(struct work_struct *work)
4334 {
4335 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
4336 tx_timeout_work);
4337 bool report_failed = false;
4338 int err;
4339 int i;
4340
4341 rtnl_lock();
4342 mutex_lock(&priv->state_lock);
4343
4344 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
4345 goto unlock;
4346
4347 for (i = 0; i < priv->channels.num * priv->channels.params.num_tc; i++) {
4348 struct netdev_queue *dev_queue =
4349 netdev_get_tx_queue(priv->netdev, i);
4350 struct mlx5e_txqsq *sq = priv->txq2sq[i];
4351
4352 if (!netif_xmit_stopped(dev_queue))
4353 continue;
4354
4355 if (mlx5e_reporter_tx_timeout(sq))
4356 report_failed = true;
4357 }
4358
4359 if (!report_failed)
4360 goto unlock;
4361
4362 err = mlx5e_safe_reopen_channels(priv);
4363 if (err)
4364 netdev_err(priv->netdev,
4365 "mlx5e_safe_reopen_channels failed recovering from a tx_timeout, err(%d).\n",
4366 err);
4367
4368 unlock:
4369 mutex_unlock(&priv->state_lock);
4370 rtnl_unlock();
4371 }
4372
4373 static void mlx5e_tx_timeout(struct net_device *dev, unsigned int txqueue)
4374 {
4375 struct mlx5e_priv *priv = netdev_priv(dev);
4376
4377 netdev_err(dev, "TX timeout detected\n");
4378 queue_work(priv->wq, &priv->tx_timeout_work);
4379 }
4380
4381 static int mlx5e_xdp_allowed(struct mlx5e_priv *priv, struct bpf_prog *prog)
4382 {
4383 struct net_device *netdev = priv->netdev;
4384 struct mlx5e_channels new_channels = {};
4385
4386 if (priv->channels.params.lro_en) {
4387 netdev_warn(netdev, "can't set XDP while LRO is on, disable LRO first\n");
4388 return -EINVAL;
4389 }
4390
4391 if (MLX5_IPSEC_DEV(priv->mdev)) {
4392 netdev_warn(netdev, "can't set XDP with IPSec offload\n");
4393 return -EINVAL;
4394 }
4395
4396 new_channels.params = priv->channels.params;
4397 new_channels.params.xdp_prog = prog;
4398
4399 /* No XSK params: AF_XDP can't be enabled yet at the point of setting
4400 * the XDP program.
4401 */
4402 if (!mlx5e_rx_is_linear_skb(&new_channels.params, NULL)) {
4403 netdev_warn(netdev, "XDP is not allowed with MTU(%d) > %d\n",
4404 new_channels.params.sw_mtu,
4405 mlx5e_xdp_max_mtu(&new_channels.params, NULL));
4406 return -EINVAL;
4407 }
4408
4409 return 0;
4410 }
4411
4412 static int mlx5e_xdp_set(struct net_device *netdev, struct bpf_prog *prog)
4413 {
4414 struct mlx5e_priv *priv = netdev_priv(netdev);
4415 struct bpf_prog *old_prog;
4416 bool reset, was_opened;
4417 int err = 0;
4418 int i;
4419
4420 mutex_lock(&priv->state_lock);
4421
4422 if (prog) {
4423 err = mlx5e_xdp_allowed(priv, prog);
4424 if (err)
4425 goto unlock;
4426 }
4427
4428 was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
4429 /* no need for full reset when exchanging programs */
4430 reset = (!priv->channels.params.xdp_prog || !prog);
4431
4432 if (was_opened && !reset)
4433 /* num_channels is invariant here, so we can take the
4434 * batched reference right upfront.
4435 */
4436 bpf_prog_add(prog, priv->channels.num);
4437
4438 if (was_opened && reset) {
4439 struct mlx5e_channels new_channels = {};
4440
4441 new_channels.params = priv->channels.params;
4442 new_channels.params.xdp_prog = prog;
4443 mlx5e_set_rq_type(priv->mdev, &new_channels.params);
4444 old_prog = priv->channels.params.xdp_prog;
4445
4446 err = mlx5e_safe_switch_channels(priv, &new_channels, NULL, NULL);
4447 if (err)
4448 goto unlock;
4449 } else {
4450 /* exchange programs, extra prog reference we got from caller
4451 * as long as we don't fail from this point onwards.
4452 */
4453 old_prog = xchg(&priv->channels.params.xdp_prog, prog);
4454 }
4455
4456 if (old_prog)
4457 bpf_prog_put(old_prog);
4458
4459 if (!was_opened && reset) /* change RQ type according to priv->xdp_prog */
4460 mlx5e_set_rq_type(priv->mdev, &priv->channels.params);
4461
4462 if (!was_opened || reset)
4463 goto unlock;
4464
4465 /* exchanging programs w/o reset, we update ref counts on behalf
4466 * of the channels RQs here.
4467 */
4468 for (i = 0; i < priv->channels.num; i++) {
4469 struct mlx5e_channel *c = priv->channels.c[i];
4470 bool xsk_open = test_bit(MLX5E_CHANNEL_STATE_XSK, c->state);
4471
4472 clear_bit(MLX5E_RQ_STATE_ENABLED, &c->rq.state);
4473 if (xsk_open)
4474 clear_bit(MLX5E_RQ_STATE_ENABLED, &c->xskrq.state);
4475 napi_synchronize(&c->napi);
4476 /* prevent mlx5e_poll_rx_cq from accessing rq->xdp_prog */
4477
4478 old_prog = xchg(&c->rq.xdp_prog, prog);
4479 if (old_prog)
4480 bpf_prog_put(old_prog);
4481
4482 if (xsk_open) {
4483 old_prog = xchg(&c->xskrq.xdp_prog, prog);
4484 if (old_prog)
4485 bpf_prog_put(old_prog);
4486 }
4487
4488 set_bit(MLX5E_RQ_STATE_ENABLED, &c->rq.state);
4489 if (xsk_open)
4490 set_bit(MLX5E_RQ_STATE_ENABLED, &c->xskrq.state);
4491 /* napi_schedule in case we have missed anything */
4492 napi_schedule(&c->napi);
4493 }
4494
4495 unlock:
4496 mutex_unlock(&priv->state_lock);
4497 return err;
4498 }
4499
4500 static u32 mlx5e_xdp_query(struct net_device *dev)
4501 {
4502 struct mlx5e_priv *priv = netdev_priv(dev);
4503 const struct bpf_prog *xdp_prog;
4504 u32 prog_id = 0;
4505
4506 mutex_lock(&priv->state_lock);
4507 xdp_prog = priv->channels.params.xdp_prog;
4508 if (xdp_prog)
4509 prog_id = xdp_prog->aux->id;
4510 mutex_unlock(&priv->state_lock);
4511
4512 return prog_id;
4513 }
4514
4515 static int mlx5e_xdp(struct net_device *dev, struct netdev_bpf *xdp)
4516 {
4517 switch (xdp->command) {
4518 case XDP_SETUP_PROG:
4519 return mlx5e_xdp_set(dev, xdp->prog);
4520 case XDP_QUERY_PROG:
4521 xdp->prog_id = mlx5e_xdp_query(dev);
4522 return 0;
4523 case XDP_SETUP_XSK_UMEM:
4524 return mlx5e_xsk_setup_umem(dev, xdp->xsk.umem,
4525 xdp->xsk.queue_id);
4526 default:
4527 return -EINVAL;
4528 }
4529 }
4530
4531 #ifdef CONFIG_MLX5_ESWITCH
4532 static int mlx5e_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4533 struct net_device *dev, u32 filter_mask,
4534 int nlflags)
4535 {
4536 struct mlx5e_priv *priv = netdev_priv(dev);
4537 struct mlx5_core_dev *mdev = priv->mdev;
4538 u8 mode, setting;
4539 int err;
4540
4541 err = mlx5_eswitch_get_vepa(mdev->priv.eswitch, &setting);
4542 if (err)
4543 return err;
4544 mode = setting ? BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB;
4545 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4546 mode,
4547 0, 0, nlflags, filter_mask, NULL);
4548 }
4549
4550 static int mlx5e_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4551 u16 flags, struct netlink_ext_ack *extack)
4552 {
4553 struct mlx5e_priv *priv = netdev_priv(dev);
4554 struct mlx5_core_dev *mdev = priv->mdev;
4555 struct nlattr *attr, *br_spec;
4556 u16 mode = BRIDGE_MODE_UNDEF;
4557 u8 setting;
4558 int rem;
4559
4560 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4561 if (!br_spec)
4562 return -EINVAL;
4563
4564 nla_for_each_nested(attr, br_spec, rem) {
4565 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4566 continue;
4567
4568 if (nla_len(attr) < sizeof(mode))
4569 return -EINVAL;
4570
4571 mode = nla_get_u16(attr);
4572 if (mode > BRIDGE_MODE_VEPA)
4573 return -EINVAL;
4574
4575 break;
4576 }
4577
4578 if (mode == BRIDGE_MODE_UNDEF)
4579 return -EINVAL;
4580
4581 setting = (mode == BRIDGE_MODE_VEPA) ? 1 : 0;
4582 return mlx5_eswitch_set_vepa(mdev->priv.eswitch, setting);
4583 }
4584 #endif
4585
4586 const struct net_device_ops mlx5e_netdev_ops = {
4587 .ndo_open = mlx5e_open,
4588 .ndo_stop = mlx5e_close,
4589 .ndo_start_xmit = mlx5e_xmit,
4590 .ndo_setup_tc = mlx5e_setup_tc,
4591 .ndo_select_queue = mlx5e_select_queue,
4592 .ndo_get_stats64 = mlx5e_get_stats,
4593 .ndo_set_rx_mode = mlx5e_set_rx_mode,
4594 .ndo_set_mac_address = mlx5e_set_mac,
4595 .ndo_vlan_rx_add_vid = mlx5e_vlan_rx_add_vid,
4596 .ndo_vlan_rx_kill_vid = mlx5e_vlan_rx_kill_vid,
4597 .ndo_set_features = mlx5e_set_features,
4598 .ndo_fix_features = mlx5e_fix_features,
4599 .ndo_change_mtu = mlx5e_change_nic_mtu,
4600 .ndo_do_ioctl = mlx5e_ioctl,
4601 .ndo_set_tx_maxrate = mlx5e_set_tx_maxrate,
4602 .ndo_udp_tunnel_add = mlx5e_add_vxlan_port,
4603 .ndo_udp_tunnel_del = mlx5e_del_vxlan_port,
4604 .ndo_features_check = mlx5e_features_check,
4605 .ndo_tx_timeout = mlx5e_tx_timeout,
4606 .ndo_bpf = mlx5e_xdp,
4607 .ndo_xdp_xmit = mlx5e_xdp_xmit,
4608 .ndo_xsk_wakeup = mlx5e_xsk_wakeup,
4609 #ifdef CONFIG_MLX5_EN_ARFS
4610 .ndo_rx_flow_steer = mlx5e_rx_flow_steer,
4611 #endif
4612 #ifdef CONFIG_MLX5_ESWITCH
4613 .ndo_bridge_setlink = mlx5e_bridge_setlink,
4614 .ndo_bridge_getlink = mlx5e_bridge_getlink,
4615
4616 /* SRIOV E-Switch NDOs */
4617 .ndo_set_vf_mac = mlx5e_set_vf_mac,
4618 .ndo_set_vf_vlan = mlx5e_set_vf_vlan,
4619 .ndo_set_vf_spoofchk = mlx5e_set_vf_spoofchk,
4620 .ndo_set_vf_trust = mlx5e_set_vf_trust,
4621 .ndo_set_vf_rate = mlx5e_set_vf_rate,
4622 .ndo_get_vf_config = mlx5e_get_vf_config,
4623 .ndo_set_vf_link_state = mlx5e_set_vf_link_state,
4624 .ndo_get_vf_stats = mlx5e_get_vf_stats,
4625 #endif
4626 .ndo_get_devlink_port = mlx5e_get_devlink_port,
4627 };
4628
4629 static int mlx5e_check_required_hca_cap(struct mlx5_core_dev *mdev)
4630 {
4631 if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
4632 return -EOPNOTSUPP;
4633 if (!MLX5_CAP_GEN(mdev, eth_net_offloads) ||
4634 !MLX5_CAP_GEN(mdev, nic_flow_table) ||
4635 !MLX5_CAP_ETH(mdev, csum_cap) ||
4636 !MLX5_CAP_ETH(mdev, max_lso_cap) ||
4637 !MLX5_CAP_ETH(mdev, vlan_cap) ||
4638 !MLX5_CAP_ETH(mdev, rss_ind_tbl_cap) ||
4639 MLX5_CAP_FLOWTABLE(mdev,
4640 flow_table_properties_nic_receive.max_ft_level)
4641 < 3) {
4642 mlx5_core_warn(mdev,
4643 "Not creating net device, some required device capabilities are missing\n");
4644 return -EOPNOTSUPP;
4645 }
4646 if (!MLX5_CAP_ETH(mdev, self_lb_en_modifiable))
4647 mlx5_core_warn(mdev, "Self loop back prevention is not supported\n");
4648 if (!MLX5_CAP_GEN(mdev, cq_moderation))
4649 mlx5_core_warn(mdev, "CQ moderation is not supported\n");
4650
4651 return 0;
4652 }
4653
4654 void mlx5e_build_default_indir_rqt(u32 *indirection_rqt, int len,
4655 int num_channels)
4656 {
4657 int i;
4658
4659 for (i = 0; i < len; i++)
4660 indirection_rqt[i] = i % num_channels;
4661 }
4662
4663 static bool slow_pci_heuristic(struct mlx5_core_dev *mdev)
4664 {
4665 u32 link_speed = 0;
4666 u32 pci_bw = 0;
4667
4668 mlx5e_port_max_linkspeed(mdev, &link_speed);
4669 pci_bw = pcie_bandwidth_available(mdev->pdev, NULL, NULL, NULL);
4670 mlx5_core_dbg_once(mdev, "Max link speed = %d, PCI BW = %d\n",
4671 link_speed, pci_bw);
4672
4673 #define MLX5E_SLOW_PCI_RATIO (2)
4674
4675 return link_speed && pci_bw &&
4676 link_speed > MLX5E_SLOW_PCI_RATIO * pci_bw;
4677 }
4678
4679 static struct dim_cq_moder mlx5e_get_def_tx_moderation(u8 cq_period_mode)
4680 {
4681 struct dim_cq_moder moder;
4682
4683 moder.cq_period_mode = cq_period_mode;
4684 moder.pkts = MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
4685 moder.usec = MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC;
4686 if (cq_period_mode == MLX5_CQ_PERIOD_MODE_START_FROM_CQE)
4687 moder.usec = MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC_FROM_CQE;
4688
4689 return moder;
4690 }
4691
4692 static struct dim_cq_moder mlx5e_get_def_rx_moderation(u8 cq_period_mode)
4693 {
4694 struct dim_cq_moder moder;
4695
4696 moder.cq_period_mode = cq_period_mode;
4697 moder.pkts = MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS;
4698 moder.usec = MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
4699 if (cq_period_mode == MLX5_CQ_PERIOD_MODE_START_FROM_CQE)
4700 moder.usec = MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE;
4701
4702 return moder;
4703 }
4704
4705 static u8 mlx5_to_net_dim_cq_period_mode(u8 cq_period_mode)
4706 {
4707 return cq_period_mode == MLX5_CQ_PERIOD_MODE_START_FROM_CQE ?
4708 DIM_CQ_PERIOD_MODE_START_FROM_CQE :
4709 DIM_CQ_PERIOD_MODE_START_FROM_EQE;
4710 }
4711
4712 void mlx5e_set_tx_cq_mode_params(struct mlx5e_params *params, u8 cq_period_mode)
4713 {
4714 if (params->tx_dim_enabled) {
4715 u8 dim_period_mode = mlx5_to_net_dim_cq_period_mode(cq_period_mode);
4716
4717 params->tx_cq_moderation = net_dim_get_def_tx_moderation(dim_period_mode);
4718 } else {
4719 params->tx_cq_moderation = mlx5e_get_def_tx_moderation(cq_period_mode);
4720 }
4721
4722 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_TX_CQE_BASED_MODER,
4723 params->tx_cq_moderation.cq_period_mode ==
4724 MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
4725 }
4726
4727 void mlx5e_set_rx_cq_mode_params(struct mlx5e_params *params, u8 cq_period_mode)
4728 {
4729 if (params->rx_dim_enabled) {
4730 u8 dim_period_mode = mlx5_to_net_dim_cq_period_mode(cq_period_mode);
4731
4732 params->rx_cq_moderation = net_dim_get_def_rx_moderation(dim_period_mode);
4733 } else {
4734 params->rx_cq_moderation = mlx5e_get_def_rx_moderation(cq_period_mode);
4735 }
4736
4737 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_CQE_BASED_MODER,
4738 params->rx_cq_moderation.cq_period_mode ==
4739 MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
4740 }
4741
4742 static u32 mlx5e_choose_lro_timeout(struct mlx5_core_dev *mdev, u32 wanted_timeout)
4743 {
4744 int i;
4745
4746 /* The supported periods are organized in ascending order */
4747 for (i = 0; i < MLX5E_LRO_TIMEOUT_ARR_SIZE - 1; i++)
4748 if (MLX5_CAP_ETH(mdev, lro_timer_supported_periods[i]) >= wanted_timeout)
4749 break;
4750
4751 return MLX5_CAP_ETH(mdev, lro_timer_supported_periods[i]);
4752 }
4753
4754 void mlx5e_build_rq_params(struct mlx5_core_dev *mdev,
4755 struct mlx5e_params *params)
4756 {
4757 /* Prefer Striding RQ, unless any of the following holds:
4758 * - Striding RQ configuration is not possible/supported.
4759 * - Slow PCI heuristic.
4760 * - Legacy RQ would use linear SKB while Striding RQ would use non-linear.
4761 *
4762 * No XSK params: checking the availability of striding RQ in general.
4763 */
4764 if (!slow_pci_heuristic(mdev) &&
4765 mlx5e_striding_rq_possible(mdev, params) &&
4766 (mlx5e_rx_mpwqe_is_linear_skb(mdev, params, NULL) ||
4767 !mlx5e_rx_is_linear_skb(params, NULL)))
4768 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_STRIDING_RQ, true);
4769 mlx5e_set_rq_type(mdev, params);
4770 mlx5e_init_rq_type_params(mdev, params);
4771 }
4772
4773 void mlx5e_build_rss_params(struct mlx5e_rss_params *rss_params,
4774 u16 num_channels)
4775 {
4776 enum mlx5e_traffic_types tt;
4777
4778 rss_params->hfunc = ETH_RSS_HASH_TOP;
4779 netdev_rss_key_fill(rss_params->toeplitz_hash_key,
4780 sizeof(rss_params->toeplitz_hash_key));
4781 mlx5e_build_default_indir_rqt(rss_params->indirection_rqt,
4782 MLX5E_INDIR_RQT_SIZE, num_channels);
4783 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
4784 rss_params->rx_hash_fields[tt] =
4785 tirc_default_config[tt].rx_hash_fields;
4786 }
4787
4788 void mlx5e_build_nic_params(struct mlx5e_priv *priv,
4789 struct mlx5e_xsk *xsk,
4790 struct mlx5e_rss_params *rss_params,
4791 struct mlx5e_params *params,
4792 u16 mtu)
4793 {
4794 struct mlx5_core_dev *mdev = priv->mdev;
4795 u8 rx_cq_period_mode;
4796
4797 params->sw_mtu = mtu;
4798 params->hard_mtu = MLX5E_ETH_HARD_MTU;
4799 params->num_channels = min_t(unsigned int, MLX5E_MAX_NUM_CHANNELS / 2,
4800 priv->max_nch);
4801 params->num_tc = 1;
4802
4803 /* SQ */
4804 params->log_sq_size = is_kdump_kernel() ?
4805 MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE :
4806 MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
4807
4808 /* XDP SQ */
4809 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_XDP_TX_MPWQE,
4810 MLX5_CAP_ETH(mdev, enhanced_multi_pkt_send_wqe));
4811
4812 /* set CQE compression */
4813 params->rx_cqe_compress_def = false;
4814 if (MLX5_CAP_GEN(mdev, cqe_compression) &&
4815 MLX5_CAP_GEN(mdev, vport_group_manager))
4816 params->rx_cqe_compress_def = slow_pci_heuristic(mdev);
4817
4818 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS, params->rx_cqe_compress_def);
4819 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_NO_CSUM_COMPLETE, false);
4820
4821 /* RQ */
4822 mlx5e_build_rq_params(mdev, params);
4823
4824 /* HW LRO */
4825 if (MLX5_CAP_ETH(mdev, lro_cap) &&
4826 params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
4827 /* No XSK params: checking the availability of striding RQ in general. */
4828 if (!mlx5e_rx_mpwqe_is_linear_skb(mdev, params, NULL))
4829 params->lro_en = !slow_pci_heuristic(mdev);
4830 }
4831 params->lro_timeout = mlx5e_choose_lro_timeout(mdev, MLX5E_DEFAULT_LRO_TIMEOUT);
4832
4833 /* CQ moderation params */
4834 rx_cq_period_mode = MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
4835 MLX5_CQ_PERIOD_MODE_START_FROM_CQE :
4836 MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
4837 params->rx_dim_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
4838 params->tx_dim_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
4839 mlx5e_set_rx_cq_mode_params(params, rx_cq_period_mode);
4840 mlx5e_set_tx_cq_mode_params(params, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
4841
4842 /* TX inline */
4843 mlx5_query_min_inline(mdev, &params->tx_min_inline_mode);
4844
4845 /* RSS */
4846 mlx5e_build_rss_params(rss_params, params->num_channels);
4847 params->tunneled_offload_en =
4848 mlx5e_tunnel_inner_ft_supported(mdev);
4849
4850 /* AF_XDP */
4851 params->xsk = xsk;
4852 }
4853
4854 static void mlx5e_set_netdev_dev_addr(struct net_device *netdev)
4855 {
4856 struct mlx5e_priv *priv = netdev_priv(netdev);
4857
4858 mlx5_query_mac_address(priv->mdev, netdev->dev_addr);
4859 if (is_zero_ether_addr(netdev->dev_addr) &&
4860 !MLX5_CAP_GEN(priv->mdev, vport_group_manager)) {
4861 eth_hw_addr_random(netdev);
4862 mlx5_core_info(priv->mdev, "Assigned random MAC address %pM\n", netdev->dev_addr);
4863 }
4864 }
4865
4866 static void mlx5e_build_nic_netdev(struct net_device *netdev)
4867 {
4868 struct mlx5e_priv *priv = netdev_priv(netdev);
4869 struct mlx5_core_dev *mdev = priv->mdev;
4870 bool fcs_supported;
4871 bool fcs_enabled;
4872
4873 SET_NETDEV_DEV(netdev, mdev->device);
4874
4875 netdev->netdev_ops = &mlx5e_netdev_ops;
4876
4877 #ifdef CONFIG_MLX5_CORE_EN_DCB
4878 if (MLX5_CAP_GEN(mdev, vport_group_manager) && MLX5_CAP_GEN(mdev, qos))
4879 netdev->dcbnl_ops = &mlx5e_dcbnl_ops;
4880 #endif
4881
4882 netdev->watchdog_timeo = 15 * HZ;
4883
4884 netdev->ethtool_ops = &mlx5e_ethtool_ops;
4885
4886 netdev->vlan_features |= NETIF_F_SG;
4887 netdev->vlan_features |= NETIF_F_HW_CSUM;
4888 netdev->vlan_features |= NETIF_F_GRO;
4889 netdev->vlan_features |= NETIF_F_TSO;
4890 netdev->vlan_features |= NETIF_F_TSO6;
4891 netdev->vlan_features |= NETIF_F_RXCSUM;
4892 netdev->vlan_features |= NETIF_F_RXHASH;
4893
4894 netdev->mpls_features |= NETIF_F_SG;
4895 netdev->mpls_features |= NETIF_F_HW_CSUM;
4896 netdev->mpls_features |= NETIF_F_TSO;
4897 netdev->mpls_features |= NETIF_F_TSO6;
4898
4899 netdev->hw_enc_features |= NETIF_F_HW_VLAN_CTAG_TX;
4900 netdev->hw_enc_features |= NETIF_F_HW_VLAN_CTAG_RX;
4901
4902 if (!!MLX5_CAP_ETH(mdev, lro_cap) &&
4903 mlx5e_check_fragmented_striding_rq_cap(mdev))
4904 netdev->vlan_features |= NETIF_F_LRO;
4905
4906 netdev->hw_features = netdev->vlan_features;
4907 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
4908 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
4909 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
4910 netdev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
4911
4912 if (mlx5_vxlan_allowed(mdev->vxlan) || mlx5_geneve_tx_allowed(mdev) ||
4913 mlx5e_any_tunnel_proto_supported(mdev)) {
4914 netdev->hw_enc_features |= NETIF_F_HW_CSUM;
4915 netdev->hw_enc_features |= NETIF_F_TSO;
4916 netdev->hw_enc_features |= NETIF_F_TSO6;
4917 netdev->hw_enc_features |= NETIF_F_GSO_PARTIAL;
4918 }
4919
4920 if (mlx5_vxlan_allowed(mdev->vxlan) || mlx5_geneve_tx_allowed(mdev)) {
4921 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
4922 NETIF_F_GSO_UDP_TUNNEL_CSUM;
4923 netdev->hw_enc_features |= NETIF_F_GSO_UDP_TUNNEL |
4924 NETIF_F_GSO_UDP_TUNNEL_CSUM;
4925 netdev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
4926 netdev->vlan_features |= NETIF_F_GSO_UDP_TUNNEL |
4927 NETIF_F_GSO_UDP_TUNNEL_CSUM;
4928 }
4929
4930 if (mlx5e_tunnel_proto_supported(mdev, IPPROTO_GRE)) {
4931 netdev->hw_features |= NETIF_F_GSO_GRE |
4932 NETIF_F_GSO_GRE_CSUM;
4933 netdev->hw_enc_features |= NETIF_F_GSO_GRE |
4934 NETIF_F_GSO_GRE_CSUM;
4935 netdev->gso_partial_features |= NETIF_F_GSO_GRE |
4936 NETIF_F_GSO_GRE_CSUM;
4937 }
4938
4939 if (mlx5e_tunnel_proto_supported(mdev, IPPROTO_IPIP)) {
4940 netdev->hw_features |= NETIF_F_GSO_IPXIP4 |
4941 NETIF_F_GSO_IPXIP6;
4942 netdev->hw_enc_features |= NETIF_F_GSO_IPXIP4 |
4943 NETIF_F_GSO_IPXIP6;
4944 netdev->gso_partial_features |= NETIF_F_GSO_IPXIP4 |
4945 NETIF_F_GSO_IPXIP6;
4946 }
4947
4948 netdev->hw_features |= NETIF_F_GSO_PARTIAL;
4949 netdev->gso_partial_features |= NETIF_F_GSO_UDP_L4;
4950 netdev->hw_features |= NETIF_F_GSO_UDP_L4;
4951 netdev->features |= NETIF_F_GSO_UDP_L4;
4952
4953 mlx5_query_port_fcs(mdev, &fcs_supported, &fcs_enabled);
4954
4955 if (fcs_supported)
4956 netdev->hw_features |= NETIF_F_RXALL;
4957
4958 if (MLX5_CAP_ETH(mdev, scatter_fcs))
4959 netdev->hw_features |= NETIF_F_RXFCS;
4960
4961 netdev->features = netdev->hw_features;
4962 if (!priv->channels.params.lro_en)
4963 netdev->features &= ~NETIF_F_LRO;
4964
4965 if (fcs_enabled)
4966 netdev->features &= ~NETIF_F_RXALL;
4967
4968 if (!priv->channels.params.scatter_fcs_en)
4969 netdev->features &= ~NETIF_F_RXFCS;
4970
4971 /* prefere CQE compression over rxhash */
4972 if (MLX5E_GET_PFLAG(&priv->channels.params, MLX5E_PFLAG_RX_CQE_COMPRESS))
4973 netdev->features &= ~NETIF_F_RXHASH;
4974
4975 #define FT_CAP(f) MLX5_CAP_FLOWTABLE(mdev, flow_table_properties_nic_receive.f)
4976 if (FT_CAP(flow_modify_en) &&
4977 FT_CAP(modify_root) &&
4978 FT_CAP(identified_miss_table_mode) &&
4979 FT_CAP(flow_table_modify)) {
4980 #ifdef CONFIG_MLX5_ESWITCH
4981 netdev->hw_features |= NETIF_F_HW_TC;
4982 #endif
4983 #ifdef CONFIG_MLX5_EN_ARFS
4984 netdev->hw_features |= NETIF_F_NTUPLE;
4985 #endif
4986 }
4987
4988 netdev->features |= NETIF_F_HIGHDMA;
4989 netdev->features |= NETIF_F_HW_VLAN_STAG_FILTER;
4990
4991 netdev->priv_flags |= IFF_UNICAST_FLT;
4992
4993 mlx5e_set_netdev_dev_addr(netdev);
4994 mlx5e_ipsec_build_netdev(priv);
4995 mlx5e_tls_build_netdev(priv);
4996 }
4997
4998 void mlx5e_create_q_counters(struct mlx5e_priv *priv)
4999 {
5000 struct mlx5_core_dev *mdev = priv->mdev;
5001 int err;
5002
5003 err = mlx5_core_alloc_q_counter(mdev, &priv->q_counter);
5004 if (err) {
5005 mlx5_core_warn(mdev, "alloc queue counter failed, %d\n", err);
5006 priv->q_counter = 0;
5007 }
5008
5009 err = mlx5_core_alloc_q_counter(mdev, &priv->drop_rq_q_counter);
5010 if (err) {
5011 mlx5_core_warn(mdev, "alloc drop RQ counter failed, %d\n", err);
5012 priv->drop_rq_q_counter = 0;
5013 }
5014 }
5015
5016 void mlx5e_destroy_q_counters(struct mlx5e_priv *priv)
5017 {
5018 if (priv->q_counter)
5019 mlx5_core_dealloc_q_counter(priv->mdev, priv->q_counter);
5020
5021 if (priv->drop_rq_q_counter)
5022 mlx5_core_dealloc_q_counter(priv->mdev, priv->drop_rq_q_counter);
5023 }
5024
5025 static int mlx5e_nic_init(struct mlx5_core_dev *mdev,
5026 struct net_device *netdev,
5027 const struct mlx5e_profile *profile,
5028 void *ppriv)
5029 {
5030 struct mlx5e_priv *priv = netdev_priv(netdev);
5031 struct mlx5e_rss_params *rss = &priv->rss_params;
5032 int err;
5033
5034 err = mlx5e_netdev_init(netdev, priv, mdev, profile, ppriv);
5035 if (err)
5036 return err;
5037
5038 mlx5e_build_nic_params(priv, &priv->xsk, rss, &priv->channels.params,
5039 netdev->mtu);
5040
5041 mlx5e_timestamp_init(priv);
5042
5043 err = mlx5e_ipsec_init(priv);
5044 if (err)
5045 mlx5_core_err(mdev, "IPSec initialization failed, %d\n", err);
5046 err = mlx5e_tls_init(priv);
5047 if (err)
5048 mlx5_core_err(mdev, "TLS initialization failed, %d\n", err);
5049 mlx5e_build_nic_netdev(netdev);
5050 mlx5e_health_create_reporters(priv);
5051
5052 return 0;
5053 }
5054
5055 static void mlx5e_nic_cleanup(struct mlx5e_priv *priv)
5056 {
5057 mlx5e_health_destroy_reporters(priv);
5058 mlx5e_tls_cleanup(priv);
5059 mlx5e_ipsec_cleanup(priv);
5060 mlx5e_netdev_cleanup(priv->netdev, priv);
5061 }
5062
5063 static int mlx5e_init_nic_rx(struct mlx5e_priv *priv)
5064 {
5065 struct mlx5_core_dev *mdev = priv->mdev;
5066 int err;
5067
5068 mlx5e_create_q_counters(priv);
5069
5070 err = mlx5e_open_drop_rq(priv, &priv->drop_rq);
5071 if (err) {
5072 mlx5_core_err(mdev, "open drop rq failed, %d\n", err);
5073 goto err_destroy_q_counters;
5074 }
5075
5076 err = mlx5e_create_indirect_rqt(priv);
5077 if (err)
5078 goto err_close_drop_rq;
5079
5080 err = mlx5e_create_direct_rqts(priv, priv->direct_tir);
5081 if (err)
5082 goto err_destroy_indirect_rqts;
5083
5084 err = mlx5e_create_indirect_tirs(priv, true);
5085 if (err)
5086 goto err_destroy_direct_rqts;
5087
5088 err = mlx5e_create_direct_tirs(priv, priv->direct_tir);
5089 if (err)
5090 goto err_destroy_indirect_tirs;
5091
5092 err = mlx5e_create_direct_rqts(priv, priv->xsk_tir);
5093 if (unlikely(err))
5094 goto err_destroy_direct_tirs;
5095
5096 err = mlx5e_create_direct_tirs(priv, priv->xsk_tir);
5097 if (unlikely(err))
5098 goto err_destroy_xsk_rqts;
5099
5100 err = mlx5e_create_flow_steering(priv);
5101 if (err) {
5102 mlx5_core_warn(mdev, "create flow steering failed, %d\n", err);
5103 goto err_destroy_xsk_tirs;
5104 }
5105
5106 err = mlx5e_tc_nic_init(priv);
5107 if (err)
5108 goto err_destroy_flow_steering;
5109
5110 return 0;
5111
5112 err_destroy_flow_steering:
5113 mlx5e_destroy_flow_steering(priv);
5114 err_destroy_xsk_tirs:
5115 mlx5e_destroy_direct_tirs(priv, priv->xsk_tir);
5116 err_destroy_xsk_rqts:
5117 mlx5e_destroy_direct_rqts(priv, priv->xsk_tir);
5118 err_destroy_direct_tirs:
5119 mlx5e_destroy_direct_tirs(priv, priv->direct_tir);
5120 err_destroy_indirect_tirs:
5121 mlx5e_destroy_indirect_tirs(priv, true);
5122 err_destroy_direct_rqts:
5123 mlx5e_destroy_direct_rqts(priv, priv->direct_tir);
5124 err_destroy_indirect_rqts:
5125 mlx5e_destroy_rqt(priv, &priv->indir_rqt);
5126 err_close_drop_rq:
5127 mlx5e_close_drop_rq(&priv->drop_rq);
5128 err_destroy_q_counters:
5129 mlx5e_destroy_q_counters(priv);
5130 return err;
5131 }
5132
5133 static void mlx5e_cleanup_nic_rx(struct mlx5e_priv *priv)
5134 {
5135 mlx5e_tc_nic_cleanup(priv);
5136 mlx5e_destroy_flow_steering(priv);
5137 mlx5e_destroy_direct_tirs(priv, priv->xsk_tir);
5138 mlx5e_destroy_direct_rqts(priv, priv->xsk_tir);
5139 mlx5e_destroy_direct_tirs(priv, priv->direct_tir);
5140 mlx5e_destroy_indirect_tirs(priv, true);
5141 mlx5e_destroy_direct_rqts(priv, priv->direct_tir);
5142 mlx5e_destroy_rqt(priv, &priv->indir_rqt);
5143 mlx5e_close_drop_rq(&priv->drop_rq);
5144 mlx5e_destroy_q_counters(priv);
5145 }
5146
5147 static int mlx5e_init_nic_tx(struct mlx5e_priv *priv)
5148 {
5149 int err;
5150
5151 err = mlx5e_create_tises(priv);
5152 if (err) {
5153 mlx5_core_warn(priv->mdev, "create tises failed, %d\n", err);
5154 return err;
5155 }
5156
5157 #ifdef CONFIG_MLX5_CORE_EN_DCB
5158 mlx5e_dcbnl_initialize(priv);
5159 #endif
5160 return 0;
5161 }
5162
5163 static void mlx5e_nic_enable(struct mlx5e_priv *priv)
5164 {
5165 struct net_device *netdev = priv->netdev;
5166 struct mlx5_core_dev *mdev = priv->mdev;
5167
5168 mlx5e_init_l2_addr(priv);
5169
5170 /* Marking the link as currently not needed by the Driver */
5171 if (!netif_running(netdev))
5172 mlx5_set_port_admin_status(mdev, MLX5_PORT_DOWN);
5173
5174 mlx5e_set_netdev_mtu_boundaries(priv);
5175 mlx5e_set_dev_port_mtu(priv);
5176
5177 mlx5_lag_add(mdev, netdev);
5178
5179 mlx5e_enable_async_events(priv);
5180 if (mlx5e_monitor_counter_supported(priv))
5181 mlx5e_monitor_counter_init(priv);
5182
5183 mlx5e_hv_vhca_stats_create(priv);
5184 if (netdev->reg_state != NETREG_REGISTERED)
5185 return;
5186 #ifdef CONFIG_MLX5_CORE_EN_DCB
5187 mlx5e_dcbnl_init_app(priv);
5188 #endif
5189
5190 queue_work(priv->wq, &priv->set_rx_mode_work);
5191
5192 rtnl_lock();
5193 if (netif_running(netdev))
5194 mlx5e_open(netdev);
5195 netif_device_attach(netdev);
5196 rtnl_unlock();
5197 }
5198
5199 static void mlx5e_nic_disable(struct mlx5e_priv *priv)
5200 {
5201 struct mlx5_core_dev *mdev = priv->mdev;
5202
5203 #ifdef CONFIG_MLX5_CORE_EN_DCB
5204 if (priv->netdev->reg_state == NETREG_REGISTERED)
5205 mlx5e_dcbnl_delete_app(priv);
5206 #endif
5207
5208 rtnl_lock();
5209 if (netif_running(priv->netdev))
5210 mlx5e_close(priv->netdev);
5211 netif_device_detach(priv->netdev);
5212 rtnl_unlock();
5213
5214 queue_work(priv->wq, &priv->set_rx_mode_work);
5215
5216 mlx5e_hv_vhca_stats_destroy(priv);
5217 if (mlx5e_monitor_counter_supported(priv))
5218 mlx5e_monitor_counter_cleanup(priv);
5219
5220 mlx5e_disable_async_events(priv);
5221 mlx5_lag_remove(mdev);
5222 }
5223
5224 int mlx5e_update_nic_rx(struct mlx5e_priv *priv)
5225 {
5226 return mlx5e_refresh_tirs(priv, false);
5227 }
5228
5229 static const struct mlx5e_profile mlx5e_nic_profile = {
5230 .init = mlx5e_nic_init,
5231 .cleanup = mlx5e_nic_cleanup,
5232 .init_rx = mlx5e_init_nic_rx,
5233 .cleanup_rx = mlx5e_cleanup_nic_rx,
5234 .init_tx = mlx5e_init_nic_tx,
5235 .cleanup_tx = mlx5e_cleanup_nic_tx,
5236 .enable = mlx5e_nic_enable,
5237 .disable = mlx5e_nic_disable,
5238 .update_rx = mlx5e_update_nic_rx,
5239 .update_stats = mlx5e_update_ndo_stats,
5240 .update_carrier = mlx5e_update_carrier,
5241 .rx_handlers.handle_rx_cqe = mlx5e_handle_rx_cqe,
5242 .rx_handlers.handle_rx_cqe_mpwqe = mlx5e_handle_rx_cqe_mpwrq,
5243 .max_tc = MLX5E_MAX_NUM_TC,
5244 .rq_groups = MLX5E_NUM_RQ_GROUPS(XSK),
5245 .stats_grps = mlx5e_nic_stats_grps,
5246 .stats_grps_num = mlx5e_nic_stats_grps_num,
5247 };
5248
5249 /* mlx5e generic netdev management API (move to en_common.c) */
5250
5251 /* mlx5e_netdev_init/cleanup must be called from profile->init/cleanup callbacks */
5252 int mlx5e_netdev_init(struct net_device *netdev,
5253 struct mlx5e_priv *priv,
5254 struct mlx5_core_dev *mdev,
5255 const struct mlx5e_profile *profile,
5256 void *ppriv)
5257 {
5258 /* priv init */
5259 priv->mdev = mdev;
5260 priv->netdev = netdev;
5261 priv->profile = profile;
5262 priv->ppriv = ppriv;
5263 priv->msglevel = MLX5E_MSG_LEVEL;
5264 priv->max_nch = netdev->num_rx_queues / max_t(u8, profile->rq_groups, 1);
5265 priv->max_opened_tc = 1;
5266
5267 if (!alloc_cpumask_var(&priv->scratchpad.cpumask, GFP_KERNEL))
5268 return -ENOMEM;
5269
5270 mutex_init(&priv->state_lock);
5271 INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
5272 INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
5273 INIT_WORK(&priv->tx_timeout_work, mlx5e_tx_timeout_work);
5274 INIT_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
5275
5276 priv->wq = create_singlethread_workqueue("mlx5e");
5277 if (!priv->wq)
5278 goto err_free_cpumask;
5279
5280 /* netdev init */
5281 netif_carrier_off(netdev);
5282
5283 #ifdef CONFIG_MLX5_EN_ARFS
5284 netdev->rx_cpu_rmap = mlx5_eq_table_get_rmap(mdev);
5285 #endif
5286
5287 return 0;
5288
5289 err_free_cpumask:
5290 free_cpumask_var(priv->scratchpad.cpumask);
5291
5292 return -ENOMEM;
5293 }
5294
5295 void mlx5e_netdev_cleanup(struct net_device *netdev, struct mlx5e_priv *priv)
5296 {
5297 destroy_workqueue(priv->wq);
5298 free_cpumask_var(priv->scratchpad.cpumask);
5299 }
5300
5301 struct net_device *mlx5e_create_netdev(struct mlx5_core_dev *mdev,
5302 const struct mlx5e_profile *profile,
5303 int nch,
5304 void *ppriv)
5305 {
5306 struct net_device *netdev;
5307 int err;
5308
5309 netdev = alloc_etherdev_mqs(sizeof(struct mlx5e_priv),
5310 nch * profile->max_tc,
5311 nch * profile->rq_groups);
5312 if (!netdev) {
5313 mlx5_core_err(mdev, "alloc_etherdev_mqs() failed\n");
5314 return NULL;
5315 }
5316
5317 err = profile->init(mdev, netdev, profile, ppriv);
5318 if (err) {
5319 mlx5_core_err(mdev, "failed to init mlx5e profile %d\n", err);
5320 goto err_free_netdev;
5321 }
5322
5323 return netdev;
5324
5325 err_free_netdev:
5326 free_netdev(netdev);
5327
5328 return NULL;
5329 }
5330
5331 int mlx5e_attach_netdev(struct mlx5e_priv *priv)
5332 {
5333 const bool take_rtnl = priv->netdev->reg_state == NETREG_REGISTERED;
5334 const struct mlx5e_profile *profile;
5335 int max_nch;
5336 int err;
5337
5338 profile = priv->profile;
5339 clear_bit(MLX5E_STATE_DESTROYING, &priv->state);
5340
5341 /* max number of channels may have changed */
5342 max_nch = mlx5e_get_max_num_channels(priv->mdev);
5343 if (priv->channels.params.num_channels > max_nch) {
5344 mlx5_core_warn(priv->mdev, "MLX5E: Reducing number of channels to %d\n", max_nch);
5345 /* Reducing the number of channels - RXFH has to be reset, and
5346 * mlx5e_num_channels_changed below will build the RQT.
5347 */
5348 priv->netdev->priv_flags &= ~IFF_RXFH_CONFIGURED;
5349 priv->channels.params.num_channels = max_nch;
5350 }
5351 /* 1. Set the real number of queues in the kernel the first time.
5352 * 2. Set our default XPS cpumask.
5353 * 3. Build the RQT.
5354 *
5355 * rtnl_lock is required by netif_set_real_num_*_queues in case the
5356 * netdev has been registered by this point (if this function was called
5357 * in the reload or resume flow).
5358 */
5359 if (take_rtnl)
5360 rtnl_lock();
5361 mlx5e_num_channels_changed(priv);
5362 if (take_rtnl)
5363 rtnl_unlock();
5364
5365 err = profile->init_tx(priv);
5366 if (err)
5367 goto out;
5368
5369 err = profile->init_rx(priv);
5370 if (err)
5371 goto err_cleanup_tx;
5372
5373 if (profile->enable)
5374 profile->enable(priv);
5375
5376 return 0;
5377
5378 err_cleanup_tx:
5379 profile->cleanup_tx(priv);
5380
5381 out:
5382 return err;
5383 }
5384
5385 void mlx5e_detach_netdev(struct mlx5e_priv *priv)
5386 {
5387 const struct mlx5e_profile *profile = priv->profile;
5388
5389 set_bit(MLX5E_STATE_DESTROYING, &priv->state);
5390
5391 if (profile->disable)
5392 profile->disable(priv);
5393 flush_workqueue(priv->wq);
5394
5395 profile->cleanup_rx(priv);
5396 profile->cleanup_tx(priv);
5397 cancel_work_sync(&priv->update_stats_work);
5398 }
5399
5400 void mlx5e_destroy_netdev(struct mlx5e_priv *priv)
5401 {
5402 const struct mlx5e_profile *profile = priv->profile;
5403 struct net_device *netdev = priv->netdev;
5404
5405 if (profile->cleanup)
5406 profile->cleanup(priv);
5407 free_netdev(netdev);
5408 }
5409
5410 /* mlx5e_attach and mlx5e_detach scope should be only creating/destroying
5411 * hardware contexts and to connect it to the current netdev.
5412 */
5413 static int mlx5e_attach(struct mlx5_core_dev *mdev, void *vpriv)
5414 {
5415 struct mlx5e_priv *priv = vpriv;
5416 struct net_device *netdev = priv->netdev;
5417 int err;
5418
5419 if (netif_device_present(netdev))
5420 return 0;
5421
5422 err = mlx5e_create_mdev_resources(mdev);
5423 if (err)
5424 return err;
5425
5426 err = mlx5e_attach_netdev(priv);
5427 if (err) {
5428 mlx5e_destroy_mdev_resources(mdev);
5429 return err;
5430 }
5431
5432 return 0;
5433 }
5434
5435 static void mlx5e_detach(struct mlx5_core_dev *mdev, void *vpriv)
5436 {
5437 struct mlx5e_priv *priv = vpriv;
5438 struct net_device *netdev = priv->netdev;
5439
5440 #ifdef CONFIG_MLX5_ESWITCH
5441 if (MLX5_ESWITCH_MANAGER(mdev) && vpriv == mdev)
5442 return;
5443 #endif
5444
5445 if (!netif_device_present(netdev))
5446 return;
5447
5448 mlx5e_detach_netdev(priv);
5449 mlx5e_destroy_mdev_resources(mdev);
5450 }
5451
5452 static void *mlx5e_add(struct mlx5_core_dev *mdev)
5453 {
5454 struct net_device *netdev;
5455 void *priv;
5456 int err;
5457 int nch;
5458
5459 err = mlx5e_check_required_hca_cap(mdev);
5460 if (err)
5461 return NULL;
5462
5463 #ifdef CONFIG_MLX5_ESWITCH
5464 if (MLX5_ESWITCH_MANAGER(mdev) &&
5465 mlx5_eswitch_mode(mdev->priv.eswitch) == MLX5_ESWITCH_OFFLOADS) {
5466 mlx5e_rep_register_vport_reps(mdev);
5467 return mdev;
5468 }
5469 #endif
5470
5471 nch = mlx5e_get_max_num_channels(mdev);
5472 netdev = mlx5e_create_netdev(mdev, &mlx5e_nic_profile, nch, NULL);
5473 if (!netdev) {
5474 mlx5_core_err(mdev, "mlx5e_create_netdev failed\n");
5475 return NULL;
5476 }
5477
5478 dev_net_set(netdev, mlx5_core_net(mdev));
5479 priv = netdev_priv(netdev);
5480
5481 err = mlx5e_attach(mdev, priv);
5482 if (err) {
5483 mlx5_core_err(mdev, "mlx5e_attach failed, %d\n", err);
5484 goto err_destroy_netdev;
5485 }
5486
5487 err = mlx5e_devlink_port_register(priv);
5488 if (err) {
5489 mlx5_core_err(mdev, "mlx5e_devlink_port_register failed, %d\n", err);
5490 goto err_detach;
5491 }
5492
5493 err = register_netdev(netdev);
5494 if (err) {
5495 mlx5_core_err(mdev, "register_netdev failed, %d\n", err);
5496 goto err_devlink_port_unregister;
5497 }
5498
5499 mlx5e_devlink_port_type_eth_set(priv);
5500
5501 #ifdef CONFIG_MLX5_CORE_EN_DCB
5502 mlx5e_dcbnl_init_app(priv);
5503 #endif
5504 return priv;
5505
5506 err_devlink_port_unregister:
5507 mlx5e_devlink_port_unregister(priv);
5508 err_detach:
5509 mlx5e_detach(mdev, priv);
5510 err_destroy_netdev:
5511 mlx5e_destroy_netdev(priv);
5512 return NULL;
5513 }
5514
5515 static void mlx5e_remove(struct mlx5_core_dev *mdev, void *vpriv)
5516 {
5517 struct mlx5e_priv *priv;
5518
5519 #ifdef CONFIG_MLX5_ESWITCH
5520 if (MLX5_ESWITCH_MANAGER(mdev) && vpriv == mdev) {
5521 mlx5e_rep_unregister_vport_reps(mdev);
5522 return;
5523 }
5524 #endif
5525 priv = vpriv;
5526 #ifdef CONFIG_MLX5_CORE_EN_DCB
5527 mlx5e_dcbnl_delete_app(priv);
5528 #endif
5529 unregister_netdev(priv->netdev);
5530 mlx5e_devlink_port_unregister(priv);
5531 mlx5e_detach(mdev, vpriv);
5532 mlx5e_destroy_netdev(priv);
5533 }
5534
5535 static struct mlx5_interface mlx5e_interface = {
5536 .add = mlx5e_add,
5537 .remove = mlx5e_remove,
5538 .attach = mlx5e_attach,
5539 .detach = mlx5e_detach,
5540 .protocol = MLX5_INTERFACE_PROTOCOL_ETH,
5541 };
5542
5543 void mlx5e_init(void)
5544 {
5545 mlx5e_ipsec_build_inverse_table();
5546 mlx5e_build_ptys2ethtool_map();
5547 mlx5_register_interface(&mlx5e_interface);
5548 }
5549
5550 void mlx5e_cleanup(void)
5551 {
5552 mlx5_unregister_interface(&mlx5e_interface);
5553 }
A mirror of Linus' kernel repository