2 * Copyright (c) 2015-2016, Mellanox Technologies. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <net/tc_act/tc_gact.h>
34 #include <linux/mlx5/fs.h>
35 #include <net/vxlan.h>
36 #include <net/geneve.h>
37 #include <linux/bpf.h>
38 #include <linux/debugfs.h>
39 #include <linux/if_bridge.h>
40 #include <linux/filter.h>
41 #include <net/page_pool/types.h>
42 #include <net/pkt_sched.h>
43 #include <net/xdp_sock_drv.h>
49 #include "en_accel/ipsec.h"
50 #include "en_accel/macsec.h"
51 #include "en_accel/en_accel.h"
52 #include "en_accel/ktls.h"
53 #include "lib/vxlan.h"
54 #include "lib/clock.h"
58 #include "en/monitor_stats.h"
59 #include "en/health.h"
60 #include "en/params.h"
61 #include "en/xsk/pool.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"
73 bool mlx5e_check_fragmented_striding_rq_cap(struct mlx5_core_dev *mdev, u8 page_shift,
74 enum mlx5e_mpwrq_umr_mode umr_mode)
76 u16 umr_wqebbs, max_wqebbs;
79 striding_rq_umr = MLX5_CAP_GEN(mdev, striding_rq) && MLX5_CAP_GEN(mdev, umr_ptr_rlky) &&
80 MLX5_CAP_ETH(mdev, reg_umr_sq);
84 umr_wqebbs = mlx5e_mpwrq_umr_wqebbs(mdev, page_shift, umr_mode);
85 max_wqebbs = mlx5e_get_max_sq_aligned_wqebbs(mdev);
86 /* Sanity check; should never happen, because mlx5e_mpwrq_umr_wqebbs is
87 * calculated from mlx5e_get_max_sq_aligned_wqebbs.
89 if (WARN_ON(umr_wqebbs > max_wqebbs))
95 void mlx5e_update_carrier(struct mlx5e_priv *priv)
97 struct mlx5_core_dev *mdev = priv->mdev;
101 port_state = mlx5_query_vport_state(mdev,
102 MLX5_VPORT_STATE_OP_MOD_VNIC_VPORT,
105 up = port_state == VPORT_STATE_UP;
106 if (up == netif_carrier_ok(priv->netdev))
107 netif_carrier_event(priv->netdev);
109 netdev_info(priv->netdev, "Link up\n");
110 netif_carrier_on(priv->netdev);
112 netdev_info(priv->netdev, "Link down\n");
113 netif_carrier_off(priv->netdev);
117 static void mlx5e_update_carrier_work(struct work_struct *work)
119 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
120 update_carrier_work);
122 mutex_lock(&priv->state_lock);
123 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
124 if (priv->profile->update_carrier)
125 priv->profile->update_carrier(priv);
126 mutex_unlock(&priv->state_lock);
129 static void mlx5e_update_stats_work(struct work_struct *work)
131 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
134 mutex_lock(&priv->state_lock);
135 priv->profile->update_stats(priv);
136 mutex_unlock(&priv->state_lock);
139 void mlx5e_queue_update_stats(struct mlx5e_priv *priv)
141 if (!priv->profile->update_stats)
144 if (unlikely(test_bit(MLX5E_STATE_DESTROYING, &priv->state)))
147 queue_work(priv->wq, &priv->update_stats_work);
150 static int async_event(struct notifier_block *nb, unsigned long event, void *data)
152 struct mlx5e_priv *priv = container_of(nb, struct mlx5e_priv, events_nb);
153 struct mlx5_eqe *eqe = data;
155 if (event != MLX5_EVENT_TYPE_PORT_CHANGE)
158 switch (eqe->sub_type) {
159 case MLX5_PORT_CHANGE_SUBTYPE_DOWN:
160 case MLX5_PORT_CHANGE_SUBTYPE_ACTIVE:
161 queue_work(priv->wq, &priv->update_carrier_work);
170 static void mlx5e_enable_async_events(struct mlx5e_priv *priv)
172 priv->events_nb.notifier_call = async_event;
173 mlx5_notifier_register(priv->mdev, &priv->events_nb);
176 static void mlx5e_disable_async_events(struct mlx5e_priv *priv)
178 mlx5_notifier_unregister(priv->mdev, &priv->events_nb);
181 static int blocking_event(struct notifier_block *nb, unsigned long event, void *data)
183 struct mlx5e_priv *priv = container_of(nb, struct mlx5e_priv, blocking_events_nb);
184 struct mlx5_devlink_trap_event_ctx *trap_event_ctx = data;
188 case MLX5_DRIVER_EVENT_TYPE_TRAP:
189 err = mlx5e_handle_trap_event(priv, trap_event_ctx->trap);
191 trap_event_ctx->err = err;
201 static void mlx5e_enable_blocking_events(struct mlx5e_priv *priv)
203 priv->blocking_events_nb.notifier_call = blocking_event;
204 mlx5_blocking_notifier_register(priv->mdev, &priv->blocking_events_nb);
207 static void mlx5e_disable_blocking_events(struct mlx5e_priv *priv)
209 mlx5_blocking_notifier_unregister(priv->mdev, &priv->blocking_events_nb);
212 static u16 mlx5e_mpwrq_umr_octowords(u32 entries, enum mlx5e_mpwrq_umr_mode umr_mode)
214 u8 umr_entry_size = mlx5e_mpwrq_umr_entry_size(umr_mode);
217 sz = ALIGN(entries * umr_entry_size, MLX5_UMR_FLEX_ALIGNMENT);
219 return sz / MLX5_OCTWORD;
222 static inline void mlx5e_build_umr_wqe(struct mlx5e_rq *rq,
223 struct mlx5e_icosq *sq,
224 struct mlx5e_umr_wqe *wqe)
226 struct mlx5_wqe_ctrl_seg *cseg = &wqe->ctrl;
227 struct mlx5_wqe_umr_ctrl_seg *ucseg = &wqe->uctrl;
231 ds_cnt = DIV_ROUND_UP(mlx5e_mpwrq_umr_wqe_sz(rq->mdev, rq->mpwqe.page_shift,
235 cseg->qpn_ds = cpu_to_be32((sq->sqn << MLX5_WQE_CTRL_QPN_SHIFT) |
237 cseg->umr_mkey = rq->mpwqe.umr_mkey_be;
239 ucseg->flags = MLX5_UMR_TRANSLATION_OFFSET_EN | MLX5_UMR_INLINE;
240 octowords = mlx5e_mpwrq_umr_octowords(rq->mpwqe.pages_per_wqe, rq->mpwqe.umr_mode);
241 ucseg->xlt_octowords = cpu_to_be16(octowords);
242 ucseg->mkey_mask = cpu_to_be64(MLX5_MKEY_MASK_FREE);
245 static int mlx5e_rq_shampo_hd_alloc(struct mlx5e_rq *rq, int node)
247 rq->mpwqe.shampo = kvzalloc_node(sizeof(*rq->mpwqe.shampo),
249 if (!rq->mpwqe.shampo)
254 static void mlx5e_rq_shampo_hd_free(struct mlx5e_rq *rq)
256 kvfree(rq->mpwqe.shampo);
259 static int mlx5e_rq_shampo_hd_info_alloc(struct mlx5e_rq *rq, int node)
261 struct mlx5e_shampo_hd *shampo = rq->mpwqe.shampo;
263 shampo->bitmap = bitmap_zalloc_node(shampo->hd_per_wq, GFP_KERNEL,
265 shampo->info = kvzalloc_node(array_size(shampo->hd_per_wq,
266 sizeof(*shampo->info)),
268 shampo->pages = kvzalloc_node(array_size(shampo->hd_per_wq,
269 sizeof(*shampo->pages)),
271 if (!shampo->bitmap || !shampo->info || !shampo->pages)
277 kvfree(shampo->info);
278 kvfree(shampo->bitmap);
279 kvfree(shampo->pages);
284 static void mlx5e_rq_shampo_hd_info_free(struct mlx5e_rq *rq)
286 kvfree(rq->mpwqe.shampo->bitmap);
287 kvfree(rq->mpwqe.shampo->info);
288 kvfree(rq->mpwqe.shampo->pages);
291 static int mlx5e_rq_alloc_mpwqe_info(struct mlx5e_rq *rq, int node)
293 int wq_sz = mlx5_wq_ll_get_size(&rq->mpwqe.wq);
296 alloc_size = array_size(wq_sz, struct_size(rq->mpwqe.info,
297 alloc_units.frag_pages,
298 rq->mpwqe.pages_per_wqe));
300 rq->mpwqe.info = kvzalloc_node(alloc_size, GFP_KERNEL, node);
304 /* For deferred page release (release right before alloc), make sure
305 * that on first round release is not called.
307 for (int i = 0; i < wq_sz; i++) {
308 struct mlx5e_mpw_info *wi = mlx5e_get_mpw_info(rq, i);
310 bitmap_fill(wi->skip_release_bitmap, rq->mpwqe.pages_per_wqe);
313 mlx5e_build_umr_wqe(rq, rq->icosq, &rq->mpwqe.umr_wqe);
319 static u8 mlx5e_mpwrq_access_mode(enum mlx5e_mpwrq_umr_mode umr_mode)
322 case MLX5E_MPWRQ_UMR_MODE_ALIGNED:
323 return MLX5_MKC_ACCESS_MODE_MTT;
324 case MLX5E_MPWRQ_UMR_MODE_UNALIGNED:
325 return MLX5_MKC_ACCESS_MODE_KSM;
326 case MLX5E_MPWRQ_UMR_MODE_OVERSIZED:
327 return MLX5_MKC_ACCESS_MODE_KLMS;
328 case MLX5E_MPWRQ_UMR_MODE_TRIPLE:
329 return MLX5_MKC_ACCESS_MODE_KSM;
331 WARN_ONCE(1, "MPWRQ UMR mode %d is not known\n", umr_mode);
335 static int mlx5e_create_umr_mkey(struct mlx5_core_dev *mdev,
336 u32 npages, u8 page_shift, u32 *umr_mkey,
337 dma_addr_t filler_addr,
338 enum mlx5e_mpwrq_umr_mode umr_mode,
341 struct mlx5_mtt *mtt;
342 struct mlx5_ksm *ksm;
343 struct mlx5_klm *klm;
351 if ((umr_mode == MLX5E_MPWRQ_UMR_MODE_UNALIGNED ||
352 umr_mode == MLX5E_MPWRQ_UMR_MODE_TRIPLE) &&
353 !MLX5_CAP_GEN(mdev, fixed_buffer_size)) {
354 mlx5_core_warn(mdev, "Unaligned AF_XDP requires fixed_buffer_size capability\n");
358 octwords = mlx5e_mpwrq_umr_octowords(npages, umr_mode);
360 inlen = MLX5_FLEXIBLE_INLEN(mdev, MLX5_ST_SZ_BYTES(create_mkey_in),
361 MLX5_OCTWORD, octwords);
365 in = kvzalloc(inlen, GFP_KERNEL);
369 mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
371 MLX5_SET(mkc, mkc, free, 1);
372 MLX5_SET(mkc, mkc, umr_en, 1);
373 MLX5_SET(mkc, mkc, lw, 1);
374 MLX5_SET(mkc, mkc, lr, 1);
375 MLX5_SET(mkc, mkc, access_mode_1_0, mlx5e_mpwrq_access_mode(umr_mode));
376 mlx5e_mkey_set_relaxed_ordering(mdev, mkc);
377 MLX5_SET(mkc, mkc, qpn, 0xffffff);
378 MLX5_SET(mkc, mkc, pd, mdev->mlx5e_res.hw_objs.pdn);
379 MLX5_SET64(mkc, mkc, len, npages << page_shift);
380 MLX5_SET(mkc, mkc, translations_octword_size, octwords);
381 if (umr_mode == MLX5E_MPWRQ_UMR_MODE_TRIPLE)
382 MLX5_SET(mkc, mkc, log_page_size, page_shift - 2);
383 else if (umr_mode != MLX5E_MPWRQ_UMR_MODE_OVERSIZED)
384 MLX5_SET(mkc, mkc, log_page_size, page_shift);
385 MLX5_SET(create_mkey_in, in, translations_octword_actual_size, octwords);
387 /* Initialize the mkey with all MTTs pointing to a default
388 * page (filler_addr). When the channels are activated, UMR
389 * WQEs will redirect the RX WQEs to the actual memory from
390 * the RQ's pool, while the gaps (wqe_overflow) remain mapped
391 * to the default page.
394 case MLX5E_MPWRQ_UMR_MODE_OVERSIZED:
395 klm = MLX5_ADDR_OF(create_mkey_in, in, klm_pas_mtt);
396 for (i = 0; i < npages; i++) {
397 klm[i << 1] = (struct mlx5_klm) {
398 .va = cpu_to_be64(filler_addr),
399 .bcount = cpu_to_be32(xsk_chunk_size),
400 .key = cpu_to_be32(mdev->mlx5e_res.hw_objs.mkey),
402 klm[(i << 1) + 1] = (struct mlx5_klm) {
403 .va = cpu_to_be64(filler_addr),
404 .bcount = cpu_to_be32((1 << page_shift) - xsk_chunk_size),
405 .key = cpu_to_be32(mdev->mlx5e_res.hw_objs.mkey),
409 case MLX5E_MPWRQ_UMR_MODE_UNALIGNED:
410 ksm = MLX5_ADDR_OF(create_mkey_in, in, klm_pas_mtt);
411 for (i = 0; i < npages; i++)
412 ksm[i] = (struct mlx5_ksm) {
413 .key = cpu_to_be32(mdev->mlx5e_res.hw_objs.mkey),
414 .va = cpu_to_be64(filler_addr),
417 case MLX5E_MPWRQ_UMR_MODE_ALIGNED:
418 mtt = MLX5_ADDR_OF(create_mkey_in, in, klm_pas_mtt);
419 for (i = 0; i < npages; i++)
420 mtt[i] = (struct mlx5_mtt) {
421 .ptag = cpu_to_be64(filler_addr),
424 case MLX5E_MPWRQ_UMR_MODE_TRIPLE:
425 ksm = MLX5_ADDR_OF(create_mkey_in, in, klm_pas_mtt);
426 for (i = 0; i < npages * 4; i++) {
427 ksm[i] = (struct mlx5_ksm) {
428 .key = cpu_to_be32(mdev->mlx5e_res.hw_objs.mkey),
429 .va = cpu_to_be64(filler_addr),
435 err = mlx5_core_create_mkey(mdev, umr_mkey, in, inlen);
441 static int mlx5e_create_umr_klm_mkey(struct mlx5_core_dev *mdev,
450 inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
452 in = kvzalloc(inlen, GFP_KERNEL);
456 mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
458 MLX5_SET(mkc, mkc, free, 1);
459 MLX5_SET(mkc, mkc, umr_en, 1);
460 MLX5_SET(mkc, mkc, lw, 1);
461 MLX5_SET(mkc, mkc, lr, 1);
462 MLX5_SET(mkc, mkc, access_mode_1_0, MLX5_MKC_ACCESS_MODE_KLMS);
463 mlx5e_mkey_set_relaxed_ordering(mdev, mkc);
464 MLX5_SET(mkc, mkc, qpn, 0xffffff);
465 MLX5_SET(mkc, mkc, pd, mdev->mlx5e_res.hw_objs.pdn);
466 MLX5_SET(mkc, mkc, translations_octword_size, nentries);
467 MLX5_SET(mkc, mkc, length64, 1);
468 err = mlx5_core_create_mkey(mdev, umr_mkey, in, inlen);
474 static int mlx5e_create_rq_umr_mkey(struct mlx5_core_dev *mdev, struct mlx5e_rq *rq)
476 u32 xsk_chunk_size = rq->xsk_pool ? rq->xsk_pool->chunk_size : 0;
477 u32 wq_size = mlx5_wq_ll_get_size(&rq->mpwqe.wq);
478 u32 num_entries, max_num_entries;
482 max_num_entries = mlx5e_mpwrq_max_num_entries(mdev, rq->mpwqe.umr_mode);
484 /* Shouldn't overflow, the result is at most MLX5E_MAX_RQ_NUM_MTTS. */
485 if (WARN_ON_ONCE(check_mul_overflow(wq_size, (u32)rq->mpwqe.mtts_per_wqe,
487 num_entries > max_num_entries))
488 mlx5_core_err(mdev, "%s: multiplication overflow: %u * %u > %u\n",
489 __func__, wq_size, rq->mpwqe.mtts_per_wqe,
492 err = mlx5e_create_umr_mkey(mdev, num_entries, rq->mpwqe.page_shift,
493 &umr_mkey, rq->wqe_overflow.addr,
494 rq->mpwqe.umr_mode, xsk_chunk_size);
495 rq->mpwqe.umr_mkey_be = cpu_to_be32(umr_mkey);
499 static int mlx5e_create_rq_hd_umr_mkey(struct mlx5_core_dev *mdev,
502 u32 max_klm_size = BIT(MLX5_CAP_GEN(mdev, log_max_klm_list_size));
504 if (max_klm_size < rq->mpwqe.shampo->hd_per_wq) {
505 mlx5_core_err(mdev, "max klm list size 0x%x is smaller than shampo header buffer list size 0x%x\n",
506 max_klm_size, rq->mpwqe.shampo->hd_per_wq);
509 return mlx5e_create_umr_klm_mkey(mdev, rq->mpwqe.shampo->hd_per_wq,
510 &rq->mpwqe.shampo->mkey);
513 static void mlx5e_init_frags_partition(struct mlx5e_rq *rq)
515 struct mlx5e_wqe_frag_info next_frag = {};
516 struct mlx5e_wqe_frag_info *prev = NULL;
519 WARN_ON(rq->xsk_pool);
521 next_frag.frag_page = &rq->wqe.alloc_units->frag_pages[0];
523 /* Skip first release due to deferred release. */
524 next_frag.flags = BIT(MLX5E_WQE_FRAG_SKIP_RELEASE);
526 for (i = 0; i < mlx5_wq_cyc_get_size(&rq->wqe.wq); i++) {
527 struct mlx5e_rq_frag_info *frag_info = &rq->wqe.info.arr[0];
528 struct mlx5e_wqe_frag_info *frag =
529 &rq->wqe.frags[i << rq->wqe.info.log_num_frags];
532 for (f = 0; f < rq->wqe.info.num_frags; f++, frag++) {
533 if (next_frag.offset + frag_info[f].frag_stride > PAGE_SIZE) {
534 /* Pages are assigned at runtime. */
535 next_frag.frag_page++;
536 next_frag.offset = 0;
538 prev->flags |= BIT(MLX5E_WQE_FRAG_LAST_IN_PAGE);
543 next_frag.offset += frag_info[f].frag_stride;
549 prev->flags |= BIT(MLX5E_WQE_FRAG_LAST_IN_PAGE);
552 static void mlx5e_init_xsk_buffs(struct mlx5e_rq *rq)
556 /* Assumptions used by XSK batched allocator. */
557 WARN_ON(rq->wqe.info.num_frags != 1);
558 WARN_ON(rq->wqe.info.log_num_frags != 0);
559 WARN_ON(rq->wqe.info.arr[0].frag_stride != PAGE_SIZE);
561 /* Considering the above assumptions a fragment maps to a single
564 for (i = 0; i < mlx5_wq_cyc_get_size(&rq->wqe.wq); i++) {
565 rq->wqe.frags[i].xskp = &rq->wqe.alloc_units->xsk_buffs[i];
567 /* Skip first release due to deferred release as WQES are
570 rq->wqe.frags[i].flags |= BIT(MLX5E_WQE_FRAG_SKIP_RELEASE);
574 static int mlx5e_init_wqe_alloc_info(struct mlx5e_rq *rq, int node)
576 int wq_sz = mlx5_wq_cyc_get_size(&rq->wqe.wq);
577 int len = wq_sz << rq->wqe.info.log_num_frags;
578 struct mlx5e_wqe_frag_info *frags;
579 union mlx5e_alloc_units *aus;
583 aus_sz = sizeof(*aus->xsk_buffs);
585 aus_sz = sizeof(*aus->frag_pages);
587 aus = kvzalloc_node(array_size(len, aus_sz), GFP_KERNEL, node);
591 frags = kvzalloc_node(array_size(len, sizeof(*frags)), GFP_KERNEL, node);
597 rq->wqe.alloc_units = aus;
598 rq->wqe.frags = frags;
601 mlx5e_init_xsk_buffs(rq);
603 mlx5e_init_frags_partition(rq);
608 static void mlx5e_free_wqe_alloc_info(struct mlx5e_rq *rq)
610 kvfree(rq->wqe.frags);
611 kvfree(rq->wqe.alloc_units);
614 static void mlx5e_rq_err_cqe_work(struct work_struct *recover_work)
616 struct mlx5e_rq *rq = container_of(recover_work, struct mlx5e_rq, recover_work);
618 mlx5e_reporter_rq_cqe_err(rq);
621 static int mlx5e_alloc_mpwqe_rq_drop_page(struct mlx5e_rq *rq)
623 rq->wqe_overflow.page = alloc_page(GFP_KERNEL);
624 if (!rq->wqe_overflow.page)
627 rq->wqe_overflow.addr = dma_map_page(rq->pdev, rq->wqe_overflow.page, 0,
628 PAGE_SIZE, rq->buff.map_dir);
629 if (dma_mapping_error(rq->pdev, rq->wqe_overflow.addr)) {
630 __free_page(rq->wqe_overflow.page);
636 static void mlx5e_free_mpwqe_rq_drop_page(struct mlx5e_rq *rq)
638 dma_unmap_page(rq->pdev, rq->wqe_overflow.addr, PAGE_SIZE,
640 __free_page(rq->wqe_overflow.page);
643 static int mlx5e_init_rxq_rq(struct mlx5e_channel *c, struct mlx5e_params *params,
644 u32 xdp_frag_size, struct mlx5e_rq *rq)
646 struct mlx5_core_dev *mdev = c->mdev;
649 rq->wq_type = params->rq_wq_type;
651 rq->netdev = c->netdev;
653 rq->tstamp = c->tstamp;
654 rq->clock = &mdev->clock;
655 rq->icosq = &c->icosq;
660 MLX5E_SW2HW_MTU(params, params->sw_mtu) - ETH_FCS_LEN * !params->scatter_fcs_en;
661 rq->xdpsq = &c->rq_xdpsq;
662 rq->stats = &c->priv->channel_stats[c->ix]->rq;
663 rq->ptp_cyc2time = mlx5_rq_ts_translator(mdev);
664 err = mlx5e_rq_set_handlers(rq, params, NULL);
668 return __xdp_rxq_info_reg(&rq->xdp_rxq, rq->netdev, rq->ix, c->napi.napi_id,
672 static int mlx5_rq_shampo_alloc(struct mlx5_core_dev *mdev,
673 struct mlx5e_params *params,
674 struct mlx5e_rq_param *rqp,
679 void *wqc = MLX5_ADDR_OF(rqc, rqp->rqc, wq);
683 if (!test_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state))
685 err = mlx5e_rq_shampo_hd_alloc(rq, node);
688 rq->mpwqe.shampo->hd_per_wq =
689 mlx5e_shampo_hd_per_wq(mdev, params, rqp);
690 err = mlx5e_create_rq_hd_umr_mkey(mdev, rq);
693 err = mlx5e_rq_shampo_hd_info_alloc(rq, node);
695 goto err_shampo_info;
696 rq->hw_gro_data = kvzalloc_node(sizeof(*rq->hw_gro_data), GFP_KERNEL, node);
697 if (!rq->hw_gro_data) {
699 goto err_hw_gro_data;
701 rq->mpwqe.shampo->key =
702 cpu_to_be32(rq->mpwqe.shampo->mkey);
703 rq->mpwqe.shampo->hd_per_wqe =
704 mlx5e_shampo_hd_per_wqe(mdev, params, rqp);
705 wq_size = BIT(MLX5_GET(wq, wqc, log_wq_sz));
706 *pool_size += (rq->mpwqe.shampo->hd_per_wqe * wq_size) /
707 MLX5E_SHAMPO_WQ_HEADER_PER_PAGE;
711 mlx5e_rq_shampo_hd_info_free(rq);
713 mlx5_core_destroy_mkey(mdev, rq->mpwqe.shampo->mkey);
715 mlx5e_rq_shampo_hd_free(rq);
720 static void mlx5e_rq_free_shampo(struct mlx5e_rq *rq)
722 if (!test_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state))
725 kvfree(rq->hw_gro_data);
726 mlx5e_rq_shampo_hd_info_free(rq);
727 mlx5_core_destroy_mkey(rq->mdev, rq->mpwqe.shampo->mkey);
728 mlx5e_rq_shampo_hd_free(rq);
731 static int mlx5e_alloc_rq(struct mlx5e_params *params,
732 struct mlx5e_xsk_param *xsk,
733 struct mlx5e_rq_param *rqp,
734 int node, struct mlx5e_rq *rq)
736 struct mlx5_core_dev *mdev = rq->mdev;
737 void *rqc = rqp->rqc;
738 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
744 rqp->wq.db_numa_node = node;
745 INIT_WORK(&rq->recover_work, mlx5e_rq_err_cqe_work);
747 if (params->xdp_prog)
748 bpf_prog_inc(params->xdp_prog);
749 RCU_INIT_POINTER(rq->xdp_prog, params->xdp_prog);
751 rq->buff.map_dir = params->xdp_prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
752 rq->buff.headroom = mlx5e_get_rq_headroom(mdev, params, xsk);
753 pool_size = 1 << params->log_rq_mtu_frames;
755 rq->mkey_be = cpu_to_be32(mdev->mlx5e_res.hw_objs.mkey);
757 switch (rq->wq_type) {
758 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
759 err = mlx5_wq_ll_create(mdev, &rqp->wq, rqc_wq, &rq->mpwqe.wq,
762 goto err_rq_xdp_prog;
764 err = mlx5e_alloc_mpwqe_rq_drop_page(rq);
766 goto err_rq_wq_destroy;
768 rq->mpwqe.wq.db = &rq->mpwqe.wq.db[MLX5_RCV_DBR];
770 wq_sz = mlx5_wq_ll_get_size(&rq->mpwqe.wq);
772 rq->mpwqe.page_shift = mlx5e_mpwrq_page_shift(mdev, xsk);
773 rq->mpwqe.umr_mode = mlx5e_mpwrq_umr_mode(mdev, xsk);
774 rq->mpwqe.pages_per_wqe =
775 mlx5e_mpwrq_pages_per_wqe(mdev, rq->mpwqe.page_shift,
777 rq->mpwqe.umr_wqebbs =
778 mlx5e_mpwrq_umr_wqebbs(mdev, rq->mpwqe.page_shift,
780 rq->mpwqe.mtts_per_wqe =
781 mlx5e_mpwrq_mtts_per_wqe(mdev, rq->mpwqe.page_shift,
784 pool_size = rq->mpwqe.pages_per_wqe <<
785 mlx5e_mpwqe_get_log_rq_size(mdev, params, xsk);
787 if (!mlx5e_rx_mpwqe_is_linear_skb(mdev, params, xsk) && params->xdp_prog)
788 pool_size *= 2; /* additional page per packet for the linear part */
790 rq->mpwqe.log_stride_sz = mlx5e_mpwqe_get_log_stride_size(mdev, params, xsk);
791 rq->mpwqe.num_strides =
792 BIT(mlx5e_mpwqe_get_log_num_strides(mdev, params, xsk));
793 rq->mpwqe.min_wqe_bulk = mlx5e_mpwqe_get_min_wqe_bulk(wq_sz);
795 rq->buff.frame0_sz = (1 << rq->mpwqe.log_stride_sz);
797 err = mlx5e_create_rq_umr_mkey(mdev, rq);
799 goto err_rq_drop_page;
801 err = mlx5e_rq_alloc_mpwqe_info(rq, node);
805 err = mlx5_rq_shampo_alloc(mdev, params, rqp, rq, &pool_size, node);
807 goto err_free_mpwqe_info;
810 default: /* MLX5_WQ_TYPE_CYCLIC */
811 err = mlx5_wq_cyc_create(mdev, &rqp->wq, rqc_wq, &rq->wqe.wq,
814 goto err_rq_xdp_prog;
816 rq->wqe.wq.db = &rq->wqe.wq.db[MLX5_RCV_DBR];
818 wq_sz = mlx5_wq_cyc_get_size(&rq->wqe.wq);
820 rq->wqe.info = rqp->frags_info;
821 rq->buff.frame0_sz = rq->wqe.info.arr[0].frag_stride;
823 err = mlx5e_init_wqe_alloc_info(rq, node);
825 goto err_rq_wq_destroy;
829 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
830 MEM_TYPE_XSK_BUFF_POOL, NULL);
831 xsk_pool_set_rxq_info(rq->xsk_pool, &rq->xdp_rxq);
833 /* Create a page_pool and register it with rxq */
834 struct page_pool_params pp_params = { 0 };
837 pp_params.flags = PP_FLAG_DMA_MAP | PP_FLAG_DMA_SYNC_DEV | PP_FLAG_PAGE_FRAG;
838 pp_params.pool_size = pool_size;
839 pp_params.nid = node;
840 pp_params.dev = rq->pdev;
841 pp_params.napi = rq->cq.napi;
842 pp_params.dma_dir = rq->buff.map_dir;
843 pp_params.max_len = PAGE_SIZE;
845 /* page_pool can be used even when there is no rq->xdp_prog,
846 * given page_pool does not handle DMA mapping there is no
847 * required state to clear. And page_pool gracefully handle
850 rq->page_pool = page_pool_create(&pp_params);
851 if (IS_ERR(rq->page_pool)) {
852 err = PTR_ERR(rq->page_pool);
853 rq->page_pool = NULL;
854 goto err_free_by_rq_type;
856 if (xdp_rxq_info_is_reg(&rq->xdp_rxq))
857 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
858 MEM_TYPE_PAGE_POOL, rq->page_pool);
861 goto err_destroy_page_pool;
863 for (i = 0; i < wq_sz; i++) {
864 if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
865 struct mlx5e_rx_wqe_ll *wqe =
866 mlx5_wq_ll_get_wqe(&rq->mpwqe.wq, i);
868 rq->mpwqe.num_strides << rq->mpwqe.log_stride_sz;
869 u64 dma_offset = mul_u32_u32(i, rq->mpwqe.mtts_per_wqe) <<
870 rq->mpwqe.page_shift;
871 u16 headroom = test_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state) ?
872 0 : rq->buff.headroom;
874 wqe->data[0].addr = cpu_to_be64(dma_offset + headroom);
875 wqe->data[0].byte_count = cpu_to_be32(byte_count);
876 wqe->data[0].lkey = rq->mpwqe.umr_mkey_be;
878 struct mlx5e_rx_wqe_cyc *wqe =
879 mlx5_wq_cyc_get_wqe(&rq->wqe.wq, i);
882 for (f = 0; f < rq->wqe.info.num_frags; f++) {
883 u32 frag_size = rq->wqe.info.arr[f].frag_size |
884 MLX5_HW_START_PADDING;
886 wqe->data[f].byte_count = cpu_to_be32(frag_size);
887 wqe->data[f].lkey = rq->mkey_be;
889 /* check if num_frags is not a pow of two */
890 if (rq->wqe.info.num_frags < (1 << rq->wqe.info.log_num_frags)) {
891 wqe->data[f].byte_count = 0;
892 wqe->data[f].lkey = params->terminate_lkey_be;
893 wqe->data[f].addr = 0;
898 INIT_WORK(&rq->dim.work, mlx5e_rx_dim_work);
900 switch (params->rx_cq_moderation.cq_period_mode) {
901 case MLX5_CQ_PERIOD_MODE_START_FROM_CQE:
902 rq->dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_CQE;
904 case MLX5_CQ_PERIOD_MODE_START_FROM_EQE:
906 rq->dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
911 err_destroy_page_pool:
912 page_pool_destroy(rq->page_pool);
914 switch (rq->wq_type) {
915 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
916 mlx5e_rq_free_shampo(rq);
918 kvfree(rq->mpwqe.info);
920 mlx5_core_destroy_mkey(mdev, be32_to_cpu(rq->mpwqe.umr_mkey_be));
922 mlx5e_free_mpwqe_rq_drop_page(rq);
924 default: /* MLX5_WQ_TYPE_CYCLIC */
925 mlx5e_free_wqe_alloc_info(rq);
928 mlx5_wq_destroy(&rq->wq_ctrl);
930 if (params->xdp_prog)
931 bpf_prog_put(params->xdp_prog);
936 static void mlx5e_free_rq(struct mlx5e_rq *rq)
938 struct bpf_prog *old_prog;
940 if (xdp_rxq_info_is_reg(&rq->xdp_rxq)) {
941 old_prog = rcu_dereference_protected(rq->xdp_prog,
942 lockdep_is_held(&rq->priv->state_lock));
944 bpf_prog_put(old_prog);
947 switch (rq->wq_type) {
948 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
949 kvfree(rq->mpwqe.info);
950 mlx5_core_destroy_mkey(rq->mdev, be32_to_cpu(rq->mpwqe.umr_mkey_be));
951 mlx5e_free_mpwqe_rq_drop_page(rq);
952 mlx5e_rq_free_shampo(rq);
954 default: /* MLX5_WQ_TYPE_CYCLIC */
955 mlx5e_free_wqe_alloc_info(rq);
958 xdp_rxq_info_unreg(&rq->xdp_rxq);
959 page_pool_destroy(rq->page_pool);
960 mlx5_wq_destroy(&rq->wq_ctrl);
963 int mlx5e_create_rq(struct mlx5e_rq *rq, struct mlx5e_rq_param *param)
965 struct mlx5_core_dev *mdev = rq->mdev;
973 inlen = MLX5_ST_SZ_BYTES(create_rq_in) +
974 sizeof(u64) * rq->wq_ctrl.buf.npages;
975 in = kvzalloc(inlen, GFP_KERNEL);
979 ts_format = mlx5_is_real_time_rq(mdev) ?
980 MLX5_TIMESTAMP_FORMAT_REAL_TIME :
981 MLX5_TIMESTAMP_FORMAT_FREE_RUNNING;
982 rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
983 wq = MLX5_ADDR_OF(rqc, rqc, wq);
985 memcpy(rqc, param->rqc, sizeof(param->rqc));
987 MLX5_SET(rqc, rqc, cqn, rq->cq.mcq.cqn);
988 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST);
989 MLX5_SET(rqc, rqc, ts_format, ts_format);
990 MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift -
991 MLX5_ADAPTER_PAGE_SHIFT);
992 MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma);
994 if (test_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state)) {
995 MLX5_SET(wq, wq, log_headers_buffer_entry_num,
996 order_base_2(rq->mpwqe.shampo->hd_per_wq));
997 MLX5_SET(wq, wq, headers_mkey, rq->mpwqe.shampo->mkey);
1000 mlx5_fill_page_frag_array(&rq->wq_ctrl.buf,
1001 (__be64 *)MLX5_ADDR_OF(wq, wq, pas));
1003 err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn);
1010 static int mlx5e_modify_rq_state(struct mlx5e_rq *rq, int curr_state, int next_state)
1012 struct mlx5_core_dev *mdev = rq->mdev;
1019 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
1020 in = kvzalloc(inlen, GFP_KERNEL);
1024 if (curr_state == MLX5_RQC_STATE_RST && next_state == MLX5_RQC_STATE_RDY)
1025 mlx5e_rqwq_reset(rq);
1027 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
1029 MLX5_SET(modify_rq_in, in, rq_state, curr_state);
1030 MLX5_SET(rqc, rqc, state, next_state);
1032 err = mlx5_core_modify_rq(mdev, rq->rqn, in);
1039 static void mlx5e_flush_rq_cq(struct mlx5e_rq *rq)
1041 struct mlx5_cqwq *cqwq = &rq->cq.wq;
1042 struct mlx5_cqe64 *cqe;
1044 if (test_bit(MLX5E_RQ_STATE_MINI_CQE_ENHANCED, &rq->state)) {
1045 while ((cqe = mlx5_cqwq_get_cqe_enahnced_comp(cqwq)))
1046 mlx5_cqwq_pop(cqwq);
1048 while ((cqe = mlx5_cqwq_get_cqe(cqwq)))
1049 mlx5_cqwq_pop(cqwq);
1052 mlx5_cqwq_update_db_record(cqwq);
1055 int mlx5e_flush_rq(struct mlx5e_rq *rq, int curr_state)
1057 struct net_device *dev = rq->netdev;
1060 err = mlx5e_modify_rq_state(rq, curr_state, MLX5_RQC_STATE_RST);
1062 netdev_err(dev, "Failed to move rq 0x%x to reset\n", rq->rqn);
1066 mlx5e_free_rx_descs(rq);
1067 mlx5e_flush_rq_cq(rq);
1069 err = mlx5e_modify_rq_state(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
1071 netdev_err(dev, "Failed to move rq 0x%x to ready\n", rq->rqn);
1078 static int mlx5e_modify_rq_vsd(struct mlx5e_rq *rq, bool vsd)
1080 struct mlx5_core_dev *mdev = rq->mdev;
1086 inlen = MLX5_ST_SZ_BYTES(modify_rq_in);
1087 in = kvzalloc(inlen, GFP_KERNEL);
1091 rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx);
1093 MLX5_SET(modify_rq_in, in, rq_state, MLX5_RQC_STATE_RDY);
1094 MLX5_SET64(modify_rq_in, in, modify_bitmask,
1095 MLX5_MODIFY_RQ_IN_MODIFY_BITMASK_VSD);
1096 MLX5_SET(rqc, rqc, vsd, vsd);
1097 MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RDY);
1099 err = mlx5_core_modify_rq(mdev, rq->rqn, in);
1106 void mlx5e_destroy_rq(struct mlx5e_rq *rq)
1108 mlx5_core_destroy_rq(rq->mdev, rq->rqn);
1111 int mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq, int wait_time)
1113 unsigned long exp_time = jiffies + msecs_to_jiffies(wait_time);
1115 u16 min_wqes = mlx5_min_rx_wqes(rq->wq_type, mlx5e_rqwq_get_size(rq));
1118 if (mlx5e_rqwq_get_cur_sz(rq) >= min_wqes)
1122 } while (time_before(jiffies, exp_time));
1124 netdev_warn(rq->netdev, "Failed to get min RX wqes on Channel[%d] RQN[0x%x] wq cur_sz(%d) min_rx_wqes(%d)\n",
1125 rq->ix, rq->rqn, mlx5e_rqwq_get_cur_sz(rq), min_wqes);
1127 mlx5e_reporter_rx_timeout(rq);
1131 void mlx5e_free_rx_missing_descs(struct mlx5e_rq *rq)
1133 struct mlx5_wq_ll *wq;
1137 if (rq->wq_type != MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ)
1143 /* Release WQEs that are in missing state: they have been
1144 * popped from the list after completion but were not freed
1145 * due to deferred release.
1146 * Also free the linked-list reserved entry, hence the "+ 1".
1148 for (i = 0; i < mlx5_wq_ll_missing(wq) + 1; i++) {
1149 rq->dealloc_wqe(rq, head);
1150 head = mlx5_wq_ll_get_wqe_next_ix(wq, head);
1153 if (test_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state)) {
1156 len = (rq->mpwqe.shampo->pi - rq->mpwqe.shampo->ci) &
1157 (rq->mpwqe.shampo->hd_per_wq - 1);
1158 mlx5e_shampo_dealloc_hd(rq, len, rq->mpwqe.shampo->ci, false);
1159 rq->mpwqe.shampo->pi = rq->mpwqe.shampo->ci;
1162 rq->mpwqe.actual_wq_head = wq->head;
1163 rq->mpwqe.umr_in_progress = 0;
1164 rq->mpwqe.umr_completed = 0;
1167 void mlx5e_free_rx_descs(struct mlx5e_rq *rq)
1172 if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
1173 struct mlx5_wq_ll *wq = &rq->mpwqe.wq;
1175 mlx5e_free_rx_missing_descs(rq);
1177 while (!mlx5_wq_ll_is_empty(wq)) {
1178 struct mlx5e_rx_wqe_ll *wqe;
1180 wqe_ix_be = *wq->tail_next;
1181 wqe_ix = be16_to_cpu(wqe_ix_be);
1182 wqe = mlx5_wq_ll_get_wqe(wq, wqe_ix);
1183 rq->dealloc_wqe(rq, wqe_ix);
1184 mlx5_wq_ll_pop(wq, wqe_ix_be,
1185 &wqe->next.next_wqe_index);
1188 if (test_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state))
1189 mlx5e_shampo_dealloc_hd(rq, rq->mpwqe.shampo->hd_per_wq,
1192 struct mlx5_wq_cyc *wq = &rq->wqe.wq;
1193 u16 missing = mlx5_wq_cyc_missing(wq);
1194 u16 head = mlx5_wq_cyc_get_head(wq);
1196 while (!mlx5_wq_cyc_is_empty(wq)) {
1197 wqe_ix = mlx5_wq_cyc_get_tail(wq);
1198 rq->dealloc_wqe(rq, wqe_ix);
1199 mlx5_wq_cyc_pop(wq);
1201 /* Missing slots might also contain unreleased pages due to
1205 wqe_ix = mlx5_wq_cyc_ctr2ix(wq, head++);
1206 rq->dealloc_wqe(rq, wqe_ix);
1212 int mlx5e_open_rq(struct mlx5e_params *params, struct mlx5e_rq_param *param,
1213 struct mlx5e_xsk_param *xsk, int node,
1214 struct mlx5e_rq *rq)
1216 struct mlx5_core_dev *mdev = rq->mdev;
1219 if (params->packet_merge.type == MLX5E_PACKET_MERGE_SHAMPO)
1220 __set_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state);
1222 err = mlx5e_alloc_rq(params, xsk, param, node, rq);
1226 err = mlx5e_create_rq(rq, param);
1230 err = mlx5e_modify_rq_state(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
1232 goto err_destroy_rq;
1234 if (MLX5_CAP_ETH(mdev, cqe_checksum_full))
1235 __set_bit(MLX5E_RQ_STATE_CSUM_FULL, &rq->state);
1237 if (params->rx_dim_enabled)
1238 __set_bit(MLX5E_RQ_STATE_DIM, &rq->state);
1240 /* We disable csum_complete when XDP is enabled since
1241 * XDP programs might manipulate packets which will render
1242 * skb->checksum incorrect.
1244 if (MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_NO_CSUM_COMPLETE) || params->xdp_prog)
1245 __set_bit(MLX5E_RQ_STATE_NO_CSUM_COMPLETE, &rq->state);
1247 /* For CQE compression on striding RQ, use stride index provided by
1248 * HW if capability is supported.
1250 if (MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_STRIDING_RQ) &&
1251 MLX5_CAP_GEN(mdev, mini_cqe_resp_stride_index))
1252 __set_bit(MLX5E_RQ_STATE_MINI_CQE_HW_STRIDX, &rq->state);
1254 /* For enhanced CQE compression packet processing. decompress
1255 * session according to the enhanced layout.
1257 if (MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS) &&
1258 MLX5_CAP_GEN(mdev, enhanced_cqe_compression))
1259 __set_bit(MLX5E_RQ_STATE_MINI_CQE_ENHANCED, &rq->state);
1264 mlx5e_destroy_rq(rq);
1271 void mlx5e_activate_rq(struct mlx5e_rq *rq)
1273 set_bit(MLX5E_RQ_STATE_ENABLED, &rq->state);
1276 void mlx5e_deactivate_rq(struct mlx5e_rq *rq)
1278 clear_bit(MLX5E_RQ_STATE_ENABLED, &rq->state);
1279 synchronize_net(); /* Sync with NAPI to prevent mlx5e_post_rx_wqes. */
1282 void mlx5e_close_rq(struct mlx5e_rq *rq)
1284 cancel_work_sync(&rq->dim.work);
1285 cancel_work_sync(&rq->recover_work);
1286 mlx5e_destroy_rq(rq);
1287 mlx5e_free_rx_descs(rq);
1291 static void mlx5e_free_xdpsq_db(struct mlx5e_xdpsq *sq)
1293 kvfree(sq->db.xdpi_fifo.xi);
1294 kvfree(sq->db.wqe_info);
1297 static int mlx5e_alloc_xdpsq_fifo(struct mlx5e_xdpsq *sq, int numa)
1299 struct mlx5e_xdp_info_fifo *xdpi_fifo = &sq->db.xdpi_fifo;
1300 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1304 /* upper bound for maximum num of entries of all xmit_modes. */
1305 entries = roundup_pow_of_two(wq_sz * MLX5_SEND_WQEBB_NUM_DS *
1306 MLX5E_XDP_FIFO_ENTRIES2DS_MAX_RATIO);
1308 size = array_size(sizeof(*xdpi_fifo->xi), entries);
1309 xdpi_fifo->xi = kvzalloc_node(size, GFP_KERNEL, numa);
1313 xdpi_fifo->pc = &sq->xdpi_fifo_pc;
1314 xdpi_fifo->cc = &sq->xdpi_fifo_cc;
1315 xdpi_fifo->mask = entries - 1;
1320 static int mlx5e_alloc_xdpsq_db(struct mlx5e_xdpsq *sq, int numa)
1322 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1326 size = array_size(sizeof(*sq->db.wqe_info), wq_sz);
1327 sq->db.wqe_info = kvzalloc_node(size, GFP_KERNEL, numa);
1328 if (!sq->db.wqe_info)
1331 err = mlx5e_alloc_xdpsq_fifo(sq, numa);
1333 mlx5e_free_xdpsq_db(sq);
1340 static int mlx5e_alloc_xdpsq(struct mlx5e_channel *c,
1341 struct mlx5e_params *params,
1342 struct xsk_buff_pool *xsk_pool,
1343 struct mlx5e_sq_param *param,
1344 struct mlx5e_xdpsq *sq,
1347 void *sqc_wq = MLX5_ADDR_OF(sqc, param->sqc, wq);
1348 struct mlx5_core_dev *mdev = c->mdev;
1349 struct mlx5_wq_cyc *wq = &sq->wq;
1353 sq->mkey_be = c->mkey_be;
1355 sq->uar_map = mdev->mlx5e_res.hw_objs.bfreg.map;
1356 sq->min_inline_mode = params->tx_min_inline_mode;
1357 sq->hw_mtu = MLX5E_SW2HW_MTU(params, params->sw_mtu) - ETH_FCS_LEN;
1358 sq->xsk_pool = xsk_pool;
1360 sq->stats = sq->xsk_pool ?
1361 &c->priv->channel_stats[c->ix]->xsksq :
1363 &c->priv->channel_stats[c->ix]->xdpsq :
1364 &c->priv->channel_stats[c->ix]->rq_xdpsq;
1365 sq->stop_room = param->is_mpw ? mlx5e_stop_room_for_mpwqe(mdev) :
1366 mlx5e_stop_room_for_max_wqe(mdev);
1367 sq->max_sq_mpw_wqebbs = mlx5e_get_max_sq_aligned_wqebbs(mdev);
1369 param->wq.db_numa_node = cpu_to_node(c->cpu);
1370 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, wq, &sq->wq_ctrl);
1373 wq->db = &wq->db[MLX5_SND_DBR];
1375 err = mlx5e_alloc_xdpsq_db(sq, cpu_to_node(c->cpu));
1377 goto err_sq_wq_destroy;
1382 mlx5_wq_destroy(&sq->wq_ctrl);
1387 static void mlx5e_free_xdpsq(struct mlx5e_xdpsq *sq)
1389 mlx5e_free_xdpsq_db(sq);
1390 mlx5_wq_destroy(&sq->wq_ctrl);
1393 static void mlx5e_free_icosq_db(struct mlx5e_icosq *sq)
1395 kvfree(sq->db.wqe_info);
1398 static int mlx5e_alloc_icosq_db(struct mlx5e_icosq *sq, int numa)
1400 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1403 size = array_size(wq_sz, sizeof(*sq->db.wqe_info));
1404 sq->db.wqe_info = kvzalloc_node(size, GFP_KERNEL, numa);
1405 if (!sq->db.wqe_info)
1411 static void mlx5e_icosq_err_cqe_work(struct work_struct *recover_work)
1413 struct mlx5e_icosq *sq = container_of(recover_work, struct mlx5e_icosq,
1416 mlx5e_reporter_icosq_cqe_err(sq);
1419 static void mlx5e_async_icosq_err_cqe_work(struct work_struct *recover_work)
1421 struct mlx5e_icosq *sq = container_of(recover_work, struct mlx5e_icosq,
1424 /* Not implemented yet. */
1426 netdev_warn(sq->channel->netdev, "async_icosq recovery is not implemented\n");
1429 static int mlx5e_alloc_icosq(struct mlx5e_channel *c,
1430 struct mlx5e_sq_param *param,
1431 struct mlx5e_icosq *sq,
1432 work_func_t recover_work_func)
1434 void *sqc_wq = MLX5_ADDR_OF(sqc, param->sqc, wq);
1435 struct mlx5_core_dev *mdev = c->mdev;
1436 struct mlx5_wq_cyc *wq = &sq->wq;
1440 sq->uar_map = mdev->mlx5e_res.hw_objs.bfreg.map;
1441 sq->reserved_room = param->stop_room;
1443 param->wq.db_numa_node = cpu_to_node(c->cpu);
1444 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, wq, &sq->wq_ctrl);
1447 wq->db = &wq->db[MLX5_SND_DBR];
1449 err = mlx5e_alloc_icosq_db(sq, cpu_to_node(c->cpu));
1451 goto err_sq_wq_destroy;
1453 INIT_WORK(&sq->recover_work, recover_work_func);
1458 mlx5_wq_destroy(&sq->wq_ctrl);
1463 static void mlx5e_free_icosq(struct mlx5e_icosq *sq)
1465 mlx5e_free_icosq_db(sq);
1466 mlx5_wq_destroy(&sq->wq_ctrl);
1469 void mlx5e_free_txqsq_db(struct mlx5e_txqsq *sq)
1471 kvfree(sq->db.wqe_info);
1472 kvfree(sq->db.skb_fifo.fifo);
1473 kvfree(sq->db.dma_fifo);
1476 int mlx5e_alloc_txqsq_db(struct mlx5e_txqsq *sq, int numa)
1478 int wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
1479 int df_sz = wq_sz * MLX5_SEND_WQEBB_NUM_DS;
1481 sq->db.dma_fifo = kvzalloc_node(array_size(df_sz,
1482 sizeof(*sq->db.dma_fifo)),
1484 sq->db.skb_fifo.fifo = kvzalloc_node(array_size(df_sz,
1485 sizeof(*sq->db.skb_fifo.fifo)),
1487 sq->db.wqe_info = kvzalloc_node(array_size(wq_sz,
1488 sizeof(*sq->db.wqe_info)),
1490 if (!sq->db.dma_fifo || !sq->db.skb_fifo.fifo || !sq->db.wqe_info) {
1491 mlx5e_free_txqsq_db(sq);
1495 sq->dma_fifo_mask = df_sz - 1;
1497 sq->db.skb_fifo.pc = &sq->skb_fifo_pc;
1498 sq->db.skb_fifo.cc = &sq->skb_fifo_cc;
1499 sq->db.skb_fifo.mask = df_sz - 1;
1504 static int mlx5e_alloc_txqsq(struct mlx5e_channel *c,
1506 struct mlx5e_params *params,
1507 struct mlx5e_sq_param *param,
1508 struct mlx5e_txqsq *sq,
1511 void *sqc_wq = MLX5_ADDR_OF(sqc, param->sqc, wq);
1512 struct mlx5_core_dev *mdev = c->mdev;
1513 struct mlx5_wq_cyc *wq = &sq->wq;
1517 sq->clock = &mdev->clock;
1518 sq->mkey_be = c->mkey_be;
1519 sq->netdev = c->netdev;
1524 sq->txq_ix = txq_ix;
1525 sq->uar_map = mdev->mlx5e_res.hw_objs.bfreg.map;
1526 sq->min_inline_mode = params->tx_min_inline_mode;
1527 sq->hw_mtu = MLX5E_SW2HW_MTU(params, params->sw_mtu);
1528 sq->max_sq_mpw_wqebbs = mlx5e_get_max_sq_aligned_wqebbs(mdev);
1529 INIT_WORK(&sq->recover_work, mlx5e_tx_err_cqe_work);
1530 if (!MLX5_CAP_ETH(mdev, wqe_vlan_insert))
1531 set_bit(MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE, &sq->state);
1532 if (mlx5_ipsec_device_caps(c->priv->mdev))
1533 set_bit(MLX5E_SQ_STATE_IPSEC, &sq->state);
1535 set_bit(MLX5E_SQ_STATE_MPWQE, &sq->state);
1536 sq->stop_room = param->stop_room;
1537 sq->ptp_cyc2time = mlx5_sq_ts_translator(mdev);
1539 param->wq.db_numa_node = cpu_to_node(c->cpu);
1540 err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, wq, &sq->wq_ctrl);
1543 wq->db = &wq->db[MLX5_SND_DBR];
1545 err = mlx5e_alloc_txqsq_db(sq, cpu_to_node(c->cpu));
1547 goto err_sq_wq_destroy;
1549 INIT_WORK(&sq->dim.work, mlx5e_tx_dim_work);
1550 sq->dim.mode = params->tx_cq_moderation.cq_period_mode;
1555 mlx5_wq_destroy(&sq->wq_ctrl);
1560 void mlx5e_free_txqsq(struct mlx5e_txqsq *sq)
1562 mlx5e_free_txqsq_db(sq);
1563 mlx5_wq_destroy(&sq->wq_ctrl);
1566 static int mlx5e_create_sq(struct mlx5_core_dev *mdev,
1567 struct mlx5e_sq_param *param,
1568 struct mlx5e_create_sq_param *csp,
1578 inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
1579 sizeof(u64) * csp->wq_ctrl->buf.npages;
1580 in = kvzalloc(inlen, GFP_KERNEL);
1584 ts_format = mlx5_is_real_time_sq(mdev) ?
1585 MLX5_TIMESTAMP_FORMAT_REAL_TIME :
1586 MLX5_TIMESTAMP_FORMAT_FREE_RUNNING;
1587 sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
1588 wq = MLX5_ADDR_OF(sqc, sqc, wq);
1590 memcpy(sqc, param->sqc, sizeof(param->sqc));
1591 MLX5_SET(sqc, sqc, tis_lst_sz, csp->tis_lst_sz);
1592 MLX5_SET(sqc, sqc, tis_num_0, csp->tisn);
1593 MLX5_SET(sqc, sqc, cqn, csp->cqn);
1594 MLX5_SET(sqc, sqc, ts_cqe_to_dest_cqn, csp->ts_cqe_to_dest_cqn);
1595 MLX5_SET(sqc, sqc, ts_format, ts_format);
1598 if (MLX5_CAP_ETH(mdev, wqe_inline_mode) == MLX5_CAP_INLINE_MODE_VPORT_CONTEXT)
1599 MLX5_SET(sqc, sqc, min_wqe_inline_mode, csp->min_inline_mode);
1601 MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
1602 MLX5_SET(sqc, sqc, flush_in_error_en, 1);
1604 MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
1605 MLX5_SET(wq, wq, uar_page, mdev->mlx5e_res.hw_objs.bfreg.index);
1606 MLX5_SET(wq, wq, log_wq_pg_sz, csp->wq_ctrl->buf.page_shift -
1607 MLX5_ADAPTER_PAGE_SHIFT);
1608 MLX5_SET64(wq, wq, dbr_addr, csp->wq_ctrl->db.dma);
1610 mlx5_fill_page_frag_array(&csp->wq_ctrl->buf,
1611 (__be64 *)MLX5_ADDR_OF(wq, wq, pas));
1613 err = mlx5_core_create_sq(mdev, in, inlen, sqn);
1620 int mlx5e_modify_sq(struct mlx5_core_dev *mdev, u32 sqn,
1621 struct mlx5e_modify_sq_param *p)
1629 inlen = MLX5_ST_SZ_BYTES(modify_sq_in);
1630 in = kvzalloc(inlen, GFP_KERNEL);
1634 sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx);
1636 MLX5_SET(modify_sq_in, in, sq_state, p->curr_state);
1637 MLX5_SET(sqc, sqc, state, p->next_state);
1638 if (p->rl_update && p->next_state == MLX5_SQC_STATE_RDY) {
1640 MLX5_SET(sqc, sqc, packet_pacing_rate_limit_index, p->rl_index);
1642 if (p->qos_update && p->next_state == MLX5_SQC_STATE_RDY) {
1644 MLX5_SET(sqc, sqc, qos_queue_group_id, p->qos_queue_group_id);
1646 MLX5_SET64(modify_sq_in, in, modify_bitmask, bitmask);
1648 err = mlx5_core_modify_sq(mdev, sqn, in);
1655 static void mlx5e_destroy_sq(struct mlx5_core_dev *mdev, u32 sqn)
1657 mlx5_core_destroy_sq(mdev, sqn);
1660 int mlx5e_create_sq_rdy(struct mlx5_core_dev *mdev,
1661 struct mlx5e_sq_param *param,
1662 struct mlx5e_create_sq_param *csp,
1663 u16 qos_queue_group_id,
1666 struct mlx5e_modify_sq_param msp = {0};
1669 err = mlx5e_create_sq(mdev, param, csp, sqn);
1673 msp.curr_state = MLX5_SQC_STATE_RST;
1674 msp.next_state = MLX5_SQC_STATE_RDY;
1675 if (qos_queue_group_id) {
1676 msp.qos_update = true;
1677 msp.qos_queue_group_id = qos_queue_group_id;
1679 err = mlx5e_modify_sq(mdev, *sqn, &msp);
1681 mlx5e_destroy_sq(mdev, *sqn);
1686 static int mlx5e_set_sq_maxrate(struct net_device *dev,
1687 struct mlx5e_txqsq *sq, u32 rate);
1689 int mlx5e_open_txqsq(struct mlx5e_channel *c, u32 tisn, int txq_ix,
1690 struct mlx5e_params *params, struct mlx5e_sq_param *param,
1691 struct mlx5e_txqsq *sq, int tc, u16 qos_queue_group_id,
1692 struct mlx5e_sq_stats *sq_stats)
1694 struct mlx5e_create_sq_param csp = {};
1698 err = mlx5e_alloc_txqsq(c, txq_ix, params, param, sq, tc);
1702 sq->stats = sq_stats;
1706 csp.cqn = sq->cq.mcq.cqn;
1707 csp.wq_ctrl = &sq->wq_ctrl;
1708 csp.min_inline_mode = sq->min_inline_mode;
1709 err = mlx5e_create_sq_rdy(c->mdev, param, &csp, qos_queue_group_id, &sq->sqn);
1711 goto err_free_txqsq;
1713 tx_rate = c->priv->tx_rates[sq->txq_ix];
1715 mlx5e_set_sq_maxrate(c->netdev, sq, tx_rate);
1717 if (params->tx_dim_enabled)
1718 sq->state |= BIT(MLX5E_SQ_STATE_DIM);
1723 mlx5e_free_txqsq(sq);
1728 void mlx5e_activate_txqsq(struct mlx5e_txqsq *sq)
1730 sq->txq = netdev_get_tx_queue(sq->netdev, sq->txq_ix);
1731 set_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1732 netdev_tx_reset_queue(sq->txq);
1733 netif_tx_start_queue(sq->txq);
1736 void mlx5e_tx_disable_queue(struct netdev_queue *txq)
1738 __netif_tx_lock_bh(txq);
1739 netif_tx_stop_queue(txq);
1740 __netif_tx_unlock_bh(txq);
1743 void mlx5e_deactivate_txqsq(struct mlx5e_txqsq *sq)
1745 struct mlx5_wq_cyc *wq = &sq->wq;
1747 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1748 synchronize_net(); /* Sync with NAPI to prevent netif_tx_wake_queue. */
1750 mlx5e_tx_disable_queue(sq->txq);
1752 /* last doorbell out, godspeed .. */
1753 if (mlx5e_wqc_has_room_for(wq, sq->cc, sq->pc, 1)) {
1754 u16 pi = mlx5_wq_cyc_ctr2ix(wq, sq->pc);
1755 struct mlx5e_tx_wqe *nop;
1757 sq->db.wqe_info[pi] = (struct mlx5e_tx_wqe_info) {
1761 nop = mlx5e_post_nop(wq, sq->sqn, &sq->pc);
1762 mlx5e_notify_hw(wq, sq->pc, sq->uar_map, &nop->ctrl);
1766 void mlx5e_close_txqsq(struct mlx5e_txqsq *sq)
1768 struct mlx5_core_dev *mdev = sq->mdev;
1769 struct mlx5_rate_limit rl = {0};
1771 cancel_work_sync(&sq->dim.work);
1772 cancel_work_sync(&sq->recover_work);
1773 mlx5e_destroy_sq(mdev, sq->sqn);
1774 if (sq->rate_limit) {
1775 rl.rate = sq->rate_limit;
1776 mlx5_rl_remove_rate(mdev, &rl);
1778 mlx5e_free_txqsq_descs(sq);
1779 mlx5e_free_txqsq(sq);
1782 void mlx5e_tx_err_cqe_work(struct work_struct *recover_work)
1784 struct mlx5e_txqsq *sq = container_of(recover_work, struct mlx5e_txqsq,
1787 mlx5e_reporter_tx_err_cqe(sq);
1790 static int mlx5e_open_icosq(struct mlx5e_channel *c, struct mlx5e_params *params,
1791 struct mlx5e_sq_param *param, struct mlx5e_icosq *sq,
1792 work_func_t recover_work_func)
1794 struct mlx5e_create_sq_param csp = {};
1797 err = mlx5e_alloc_icosq(c, param, sq, recover_work_func);
1801 csp.cqn = sq->cq.mcq.cqn;
1802 csp.wq_ctrl = &sq->wq_ctrl;
1803 csp.min_inline_mode = params->tx_min_inline_mode;
1804 err = mlx5e_create_sq_rdy(c->mdev, param, &csp, 0, &sq->sqn);
1806 goto err_free_icosq;
1808 if (param->is_tls) {
1809 sq->ktls_resync = mlx5e_ktls_rx_resync_create_resp_list();
1810 if (IS_ERR(sq->ktls_resync)) {
1811 err = PTR_ERR(sq->ktls_resync);
1812 goto err_destroy_icosq;
1818 mlx5e_destroy_sq(c->mdev, sq->sqn);
1820 mlx5e_free_icosq(sq);
1825 void mlx5e_activate_icosq(struct mlx5e_icosq *icosq)
1827 set_bit(MLX5E_SQ_STATE_ENABLED, &icosq->state);
1830 void mlx5e_deactivate_icosq(struct mlx5e_icosq *icosq)
1832 clear_bit(MLX5E_SQ_STATE_ENABLED, &icosq->state);
1833 synchronize_net(); /* Sync with NAPI. */
1836 static void mlx5e_close_icosq(struct mlx5e_icosq *sq)
1838 struct mlx5e_channel *c = sq->channel;
1840 if (sq->ktls_resync)
1841 mlx5e_ktls_rx_resync_destroy_resp_list(sq->ktls_resync);
1842 mlx5e_destroy_sq(c->mdev, sq->sqn);
1843 mlx5e_free_icosq_descs(sq);
1844 mlx5e_free_icosq(sq);
1847 int mlx5e_open_xdpsq(struct mlx5e_channel *c, struct mlx5e_params *params,
1848 struct mlx5e_sq_param *param, struct xsk_buff_pool *xsk_pool,
1849 struct mlx5e_xdpsq *sq, bool is_redirect)
1851 struct mlx5e_create_sq_param csp = {};
1854 err = mlx5e_alloc_xdpsq(c, params, xsk_pool, param, sq, is_redirect);
1859 csp.tisn = c->priv->tisn[c->lag_port][0]; /* tc = 0 */
1860 csp.cqn = sq->cq.mcq.cqn;
1861 csp.wq_ctrl = &sq->wq_ctrl;
1862 csp.min_inline_mode = sq->min_inline_mode;
1863 set_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1865 if (param->is_xdp_mb)
1866 set_bit(MLX5E_SQ_STATE_XDP_MULTIBUF, &sq->state);
1868 err = mlx5e_create_sq_rdy(c->mdev, param, &csp, 0, &sq->sqn);
1870 goto err_free_xdpsq;
1872 mlx5e_set_xmit_fp(sq, param->is_mpw);
1874 if (!param->is_mpw && !test_bit(MLX5E_SQ_STATE_XDP_MULTIBUF, &sq->state)) {
1875 unsigned int ds_cnt = MLX5E_TX_WQE_EMPTY_DS_COUNT + 1;
1876 unsigned int inline_hdr_sz = 0;
1879 if (sq->min_inline_mode != MLX5_INLINE_MODE_NONE) {
1880 inline_hdr_sz = MLX5E_XDP_MIN_INLINE;
1884 /* Pre initialize fixed WQE fields */
1885 for (i = 0; i < mlx5_wq_cyc_get_size(&sq->wq); i++) {
1886 struct mlx5e_tx_wqe *wqe = mlx5_wq_cyc_get_wqe(&sq->wq, i);
1887 struct mlx5_wqe_ctrl_seg *cseg = &wqe->ctrl;
1888 struct mlx5_wqe_eth_seg *eseg = &wqe->eth;
1890 sq->db.wqe_info[i] = (struct mlx5e_xdp_wqe_info) {
1895 cseg->qpn_ds = cpu_to_be32((sq->sqn << 8) | ds_cnt);
1896 eseg->inline_hdr.sz = cpu_to_be16(inline_hdr_sz);
1903 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1904 mlx5e_free_xdpsq(sq);
1909 void mlx5e_close_xdpsq(struct mlx5e_xdpsq *sq)
1911 struct mlx5e_channel *c = sq->channel;
1913 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
1914 synchronize_net(); /* Sync with NAPI. */
1916 mlx5e_destroy_sq(c->mdev, sq->sqn);
1917 mlx5e_free_xdpsq_descs(sq);
1918 mlx5e_free_xdpsq(sq);
1921 static int mlx5e_alloc_cq_common(struct mlx5e_priv *priv,
1922 struct mlx5e_cq_param *param,
1923 struct mlx5e_cq *cq)
1925 struct mlx5_core_dev *mdev = priv->mdev;
1926 struct mlx5_core_cq *mcq = &cq->mcq;
1930 err = mlx5_cqwq_create(mdev, ¶m->wq, param->cqc, &cq->wq,
1936 mcq->set_ci_db = cq->wq_ctrl.db.db;
1937 mcq->arm_db = cq->wq_ctrl.db.db + 1;
1938 *mcq->set_ci_db = 0;
1940 mcq->vector = param->eq_ix;
1941 mcq->comp = mlx5e_completion_event;
1942 mcq->event = mlx5e_cq_error_event;
1944 for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) {
1945 struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i);
1948 cqe->validity_iteration_count = 0xff;
1952 cq->netdev = priv->netdev;
1958 static int mlx5e_alloc_cq(struct mlx5e_priv *priv,
1959 struct mlx5e_cq_param *param,
1960 struct mlx5e_create_cq_param *ccp,
1961 struct mlx5e_cq *cq)
1965 param->wq.buf_numa_node = ccp->node;
1966 param->wq.db_numa_node = ccp->node;
1967 param->eq_ix = ccp->ix;
1969 err = mlx5e_alloc_cq_common(priv, param, cq);
1971 cq->napi = ccp->napi;
1972 cq->ch_stats = ccp->ch_stats;
1977 static void mlx5e_free_cq(struct mlx5e_cq *cq)
1979 mlx5_wq_destroy(&cq->wq_ctrl);
1982 static int mlx5e_create_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param)
1984 u32 out[MLX5_ST_SZ_DW(create_cq_out)];
1985 struct mlx5_core_dev *mdev = cq->mdev;
1986 struct mlx5_core_cq *mcq = &cq->mcq;
1994 err = mlx5_comp_eqn_get(mdev, param->eq_ix, &eqn);
1998 inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
1999 sizeof(u64) * cq->wq_ctrl.buf.npages;
2000 in = kvzalloc(inlen, GFP_KERNEL);
2004 cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
2006 memcpy(cqc, param->cqc, sizeof(param->cqc));
2008 mlx5_fill_page_frag_array(&cq->wq_ctrl.buf,
2009 (__be64 *)MLX5_ADDR_OF(create_cq_in, in, pas));
2011 MLX5_SET(cqc, cqc, cq_period_mode, param->cq_period_mode);
2012 MLX5_SET(cqc, cqc, c_eqn_or_apu_element, eqn);
2013 MLX5_SET(cqc, cqc, uar_page, mdev->priv.uar->index);
2014 MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.buf.page_shift -
2015 MLX5_ADAPTER_PAGE_SHIFT);
2016 MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
2018 err = mlx5_core_create_cq(mdev, mcq, in, inlen, out, sizeof(out));
2030 static void mlx5e_destroy_cq(struct mlx5e_cq *cq)
2032 mlx5_core_destroy_cq(cq->mdev, &cq->mcq);
2035 int mlx5e_open_cq(struct mlx5e_priv *priv, struct dim_cq_moder moder,
2036 struct mlx5e_cq_param *param, struct mlx5e_create_cq_param *ccp,
2037 struct mlx5e_cq *cq)
2039 struct mlx5_core_dev *mdev = priv->mdev;
2042 err = mlx5e_alloc_cq(priv, param, ccp, cq);
2046 err = mlx5e_create_cq(cq, param);
2050 if (MLX5_CAP_GEN(mdev, cq_moderation))
2051 mlx5_core_modify_cq_moderation(mdev, &cq->mcq, moder.usec, moder.pkts);
2060 void mlx5e_close_cq(struct mlx5e_cq *cq)
2062 mlx5e_destroy_cq(cq);
2066 static int mlx5e_open_tx_cqs(struct mlx5e_channel *c,
2067 struct mlx5e_params *params,
2068 struct mlx5e_create_cq_param *ccp,
2069 struct mlx5e_channel_param *cparam)
2074 for (tc = 0; tc < c->num_tc; tc++) {
2075 err = mlx5e_open_cq(c->priv, params->tx_cq_moderation, &cparam->txq_sq.cqp,
2076 ccp, &c->sq[tc].cq);
2078 goto err_close_tx_cqs;
2084 for (tc--; tc >= 0; tc--)
2085 mlx5e_close_cq(&c->sq[tc].cq);
2090 static void mlx5e_close_tx_cqs(struct mlx5e_channel *c)
2094 for (tc = 0; tc < c->num_tc; tc++)
2095 mlx5e_close_cq(&c->sq[tc].cq);
2098 static int mlx5e_mqprio_txq_to_tc(struct netdev_tc_txq *tc_to_txq, unsigned int txq)
2102 for (tc = 0; tc < TC_MAX_QUEUE; tc++)
2103 if (txq - tc_to_txq[tc].offset < tc_to_txq[tc].count)
2106 WARN(1, "Unexpected TCs configuration. No match found for txq %u", txq);
2110 static int mlx5e_txq_get_qos_node_hw_id(struct mlx5e_params *params, int txq_ix,
2115 if (params->mqprio.mode != TC_MQPRIO_MODE_CHANNEL) {
2120 tc = mlx5e_mqprio_txq_to_tc(params->mqprio.tc_to_txq, txq_ix);
2124 if (tc >= params->mqprio.num_tc) {
2125 WARN(1, "Unexpected TCs configuration. tc %d is out of range of %u",
2126 tc, params->mqprio.num_tc);
2130 *hw_id = params->mqprio.channel.hw_id[tc];
2134 static int mlx5e_open_sqs(struct mlx5e_channel *c,
2135 struct mlx5e_params *params,
2136 struct mlx5e_channel_param *cparam)
2140 for (tc = 0; tc < mlx5e_get_dcb_num_tc(params); tc++) {
2141 int txq_ix = c->ix + tc * params->num_channels;
2142 u32 qos_queue_group_id;
2144 err = mlx5e_txq_get_qos_node_hw_id(params, txq_ix, &qos_queue_group_id);
2148 err = mlx5e_open_txqsq(c, c->priv->tisn[c->lag_port][tc], txq_ix,
2149 params, &cparam->txq_sq, &c->sq[tc], tc,
2151 &c->priv->channel_stats[c->ix]->sq[tc]);
2159 for (tc--; tc >= 0; tc--)
2160 mlx5e_close_txqsq(&c->sq[tc]);
2165 static void mlx5e_close_sqs(struct mlx5e_channel *c)
2169 for (tc = 0; tc < c->num_tc; tc++)
2170 mlx5e_close_txqsq(&c->sq[tc]);
2173 static int mlx5e_set_sq_maxrate(struct net_device *dev,
2174 struct mlx5e_txqsq *sq, u32 rate)
2176 struct mlx5e_priv *priv = netdev_priv(dev);
2177 struct mlx5_core_dev *mdev = priv->mdev;
2178 struct mlx5e_modify_sq_param msp = {0};
2179 struct mlx5_rate_limit rl = {0};
2183 if (rate == sq->rate_limit)
2187 if (sq->rate_limit) {
2188 rl.rate = sq->rate_limit;
2189 /* remove current rl index to free space to next ones */
2190 mlx5_rl_remove_rate(mdev, &rl);
2197 err = mlx5_rl_add_rate(mdev, &rl_index, &rl);
2199 netdev_err(dev, "Failed configuring rate %u: %d\n",
2205 msp.curr_state = MLX5_SQC_STATE_RDY;
2206 msp.next_state = MLX5_SQC_STATE_RDY;
2207 msp.rl_index = rl_index;
2208 msp.rl_update = true;
2209 err = mlx5e_modify_sq(mdev, sq->sqn, &msp);
2211 netdev_err(dev, "Failed configuring rate %u: %d\n",
2213 /* remove the rate from the table */
2215 mlx5_rl_remove_rate(mdev, &rl);
2219 sq->rate_limit = rate;
2223 static int mlx5e_set_tx_maxrate(struct net_device *dev, int index, u32 rate)
2225 struct mlx5e_priv *priv = netdev_priv(dev);
2226 struct mlx5_core_dev *mdev = priv->mdev;
2227 struct mlx5e_txqsq *sq = priv->txq2sq[index];
2230 if (!mlx5_rl_is_supported(mdev)) {
2231 netdev_err(dev, "Rate limiting is not supported on this device\n");
2235 /* rate is given in Mb/sec, HW config is in Kb/sec */
2238 /* Check whether rate in valid range, 0 is always valid */
2239 if (rate && !mlx5_rl_is_in_range(mdev, rate)) {
2240 netdev_err(dev, "TX rate %u, is not in range\n", rate);
2244 mutex_lock(&priv->state_lock);
2245 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
2246 err = mlx5e_set_sq_maxrate(dev, sq, rate);
2248 priv->tx_rates[index] = rate;
2249 mutex_unlock(&priv->state_lock);
2254 static int mlx5e_open_rxq_rq(struct mlx5e_channel *c, struct mlx5e_params *params,
2255 struct mlx5e_rq_param *rq_params)
2259 err = mlx5e_init_rxq_rq(c, params, rq_params->xdp_frag_size, &c->rq);
2263 return mlx5e_open_rq(params, rq_params, NULL, cpu_to_node(c->cpu), &c->rq);
2266 static int mlx5e_open_queues(struct mlx5e_channel *c,
2267 struct mlx5e_params *params,
2268 struct mlx5e_channel_param *cparam)
2270 struct dim_cq_moder icocq_moder = {0, 0};
2271 struct mlx5e_create_cq_param ccp;
2274 mlx5e_build_create_cq_param(&ccp, c);
2276 err = mlx5e_open_cq(c->priv, icocq_moder, &cparam->async_icosq.cqp, &ccp,
2277 &c->async_icosq.cq);
2281 err = mlx5e_open_cq(c->priv, icocq_moder, &cparam->icosq.cqp, &ccp,
2284 goto err_close_async_icosq_cq;
2286 err = mlx5e_open_tx_cqs(c, params, &ccp, cparam);
2288 goto err_close_icosq_cq;
2290 err = mlx5e_open_cq(c->priv, params->tx_cq_moderation, &cparam->xdp_sq.cqp, &ccp,
2293 goto err_close_tx_cqs;
2295 err = mlx5e_open_cq(c->priv, params->rx_cq_moderation, &cparam->rq.cqp, &ccp,
2298 goto err_close_xdp_tx_cqs;
2300 err = c->xdp ? mlx5e_open_cq(c->priv, params->tx_cq_moderation, &cparam->xdp_sq.cqp,
2301 &ccp, &c->rq_xdpsq.cq) : 0;
2303 goto err_close_rx_cq;
2305 spin_lock_init(&c->async_icosq_lock);
2307 err = mlx5e_open_icosq(c, params, &cparam->async_icosq, &c->async_icosq,
2308 mlx5e_async_icosq_err_cqe_work);
2310 goto err_close_xdpsq_cq;
2312 mutex_init(&c->icosq_recovery_lock);
2314 err = mlx5e_open_icosq(c, params, &cparam->icosq, &c->icosq,
2315 mlx5e_icosq_err_cqe_work);
2317 goto err_close_async_icosq;
2319 err = mlx5e_open_sqs(c, params, cparam);
2321 goto err_close_icosq;
2323 err = mlx5e_open_rxq_rq(c, params, &cparam->rq);
2328 err = mlx5e_open_xdpsq(c, params, &cparam->xdp_sq, NULL,
2329 &c->rq_xdpsq, false);
2334 err = mlx5e_open_xdpsq(c, params, &cparam->xdp_sq, NULL, &c->xdpsq, true);
2336 goto err_close_xdp_sq;
2342 mlx5e_close_xdpsq(&c->rq_xdpsq);
2345 mlx5e_close_rq(&c->rq);
2351 mlx5e_close_icosq(&c->icosq);
2353 err_close_async_icosq:
2354 mlx5e_close_icosq(&c->async_icosq);
2358 mlx5e_close_cq(&c->rq_xdpsq.cq);
2361 mlx5e_close_cq(&c->rq.cq);
2363 err_close_xdp_tx_cqs:
2364 mlx5e_close_cq(&c->xdpsq.cq);
2367 mlx5e_close_tx_cqs(c);
2370 mlx5e_close_cq(&c->icosq.cq);
2372 err_close_async_icosq_cq:
2373 mlx5e_close_cq(&c->async_icosq.cq);
2378 static void mlx5e_close_queues(struct mlx5e_channel *c)
2380 mlx5e_close_xdpsq(&c->xdpsq);
2382 mlx5e_close_xdpsq(&c->rq_xdpsq);
2383 /* The same ICOSQ is used for UMRs for both RQ and XSKRQ. */
2384 cancel_work_sync(&c->icosq.recover_work);
2385 mlx5e_close_rq(&c->rq);
2387 mlx5e_close_icosq(&c->icosq);
2388 mutex_destroy(&c->icosq_recovery_lock);
2389 mlx5e_close_icosq(&c->async_icosq);
2391 mlx5e_close_cq(&c->rq_xdpsq.cq);
2392 mlx5e_close_cq(&c->rq.cq);
2393 mlx5e_close_cq(&c->xdpsq.cq);
2394 mlx5e_close_tx_cqs(c);
2395 mlx5e_close_cq(&c->icosq.cq);
2396 mlx5e_close_cq(&c->async_icosq.cq);
2399 static u8 mlx5e_enumerate_lag_port(struct mlx5_core_dev *mdev, int ix)
2401 u16 port_aff_bias = mlx5_core_is_pf(mdev) ? 0 : MLX5_CAP_GEN(mdev, vhca_id);
2403 return (ix + port_aff_bias) % mlx5e_get_num_lag_ports(mdev);
2406 static int mlx5e_channel_stats_alloc(struct mlx5e_priv *priv, int ix, int cpu)
2408 if (ix > priv->stats_nch) {
2409 netdev_warn(priv->netdev, "Unexpected channel stats index %d > %d\n", ix,
2414 if (priv->channel_stats[ix])
2417 /* Asymmetric dynamic memory allocation.
2418 * Freed in mlx5e_priv_arrays_free, not on channel closure.
2420 netdev_dbg(priv->netdev, "Creating channel stats %d\n", ix);
2421 priv->channel_stats[ix] = kvzalloc_node(sizeof(**priv->channel_stats),
2422 GFP_KERNEL, cpu_to_node(cpu));
2423 if (!priv->channel_stats[ix])
2430 void mlx5e_trigger_napi_icosq(struct mlx5e_channel *c)
2432 spin_lock_bh(&c->async_icosq_lock);
2433 mlx5e_trigger_irq(&c->async_icosq);
2434 spin_unlock_bh(&c->async_icosq_lock);
2437 void mlx5e_trigger_napi_sched(struct napi_struct *napi)
2440 napi_schedule(napi);
2444 static int mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
2445 struct mlx5e_params *params,
2446 struct mlx5e_channel_param *cparam,
2447 struct xsk_buff_pool *xsk_pool,
2448 struct mlx5e_channel **cp)
2450 int cpu = mlx5_comp_vector_get_cpu(priv->mdev, ix);
2451 struct net_device *netdev = priv->netdev;
2452 struct mlx5e_xsk_param xsk;
2453 struct mlx5e_channel *c;
2457 err = mlx5_comp_irqn_get(priv->mdev, ix, &irq);
2461 err = mlx5e_channel_stats_alloc(priv, ix, cpu);
2465 c = kvzalloc_node(sizeof(*c), GFP_KERNEL, cpu_to_node(cpu));
2470 c->mdev = priv->mdev;
2471 c->tstamp = &priv->tstamp;
2474 c->pdev = mlx5_core_dma_dev(priv->mdev);
2475 c->netdev = priv->netdev;
2476 c->mkey_be = cpu_to_be32(priv->mdev->mlx5e_res.hw_objs.mkey);
2477 c->num_tc = mlx5e_get_dcb_num_tc(params);
2478 c->xdp = !!params->xdp_prog;
2479 c->stats = &priv->channel_stats[ix]->ch;
2480 c->aff_mask = irq_get_effective_affinity_mask(irq);
2481 c->lag_port = mlx5e_enumerate_lag_port(priv->mdev, ix);
2483 netif_napi_add(netdev, &c->napi, mlx5e_napi_poll);
2485 err = mlx5e_open_queues(c, params, cparam);
2490 mlx5e_build_xsk_param(xsk_pool, &xsk);
2491 err = mlx5e_open_xsk(priv, params, &xsk, xsk_pool, c);
2493 goto err_close_queues;
2501 mlx5e_close_queues(c);
2504 netif_napi_del(&c->napi);
2511 static void mlx5e_activate_channel(struct mlx5e_channel *c)
2515 napi_enable(&c->napi);
2517 for (tc = 0; tc < c->num_tc; tc++)
2518 mlx5e_activate_txqsq(&c->sq[tc]);
2519 mlx5e_activate_icosq(&c->icosq);
2520 mlx5e_activate_icosq(&c->async_icosq);
2522 if (test_bit(MLX5E_CHANNEL_STATE_XSK, c->state))
2523 mlx5e_activate_xsk(c);
2525 mlx5e_activate_rq(&c->rq);
2528 static void mlx5e_deactivate_channel(struct mlx5e_channel *c)
2532 if (test_bit(MLX5E_CHANNEL_STATE_XSK, c->state))
2533 mlx5e_deactivate_xsk(c);
2535 mlx5e_deactivate_rq(&c->rq);
2537 mlx5e_deactivate_icosq(&c->async_icosq);
2538 mlx5e_deactivate_icosq(&c->icosq);
2539 for (tc = 0; tc < c->num_tc; tc++)
2540 mlx5e_deactivate_txqsq(&c->sq[tc]);
2541 mlx5e_qos_deactivate_queues(c);
2543 napi_disable(&c->napi);
2546 static void mlx5e_close_channel(struct mlx5e_channel *c)
2548 if (test_bit(MLX5E_CHANNEL_STATE_XSK, c->state))
2550 mlx5e_close_queues(c);
2551 mlx5e_qos_close_queues(c);
2552 netif_napi_del(&c->napi);
2557 int mlx5e_open_channels(struct mlx5e_priv *priv,
2558 struct mlx5e_channels *chs)
2560 struct mlx5e_channel_param *cparam;
2564 chs->num = chs->params.num_channels;
2566 chs->c = kcalloc(chs->num, sizeof(struct mlx5e_channel *), GFP_KERNEL);
2567 cparam = kvzalloc(sizeof(struct mlx5e_channel_param), GFP_KERNEL);
2568 if (!chs->c || !cparam)
2571 err = mlx5e_build_channel_param(priv->mdev, &chs->params, priv->q_counter, cparam);
2575 for (i = 0; i < chs->num; i++) {
2576 struct xsk_buff_pool *xsk_pool = NULL;
2578 if (chs->params.xdp_prog)
2579 xsk_pool = mlx5e_xsk_get_pool(&chs->params, chs->params.xsk, i);
2581 err = mlx5e_open_channel(priv, i, &chs->params, cparam, xsk_pool, &chs->c[i]);
2583 goto err_close_channels;
2586 if (MLX5E_GET_PFLAG(&chs->params, MLX5E_PFLAG_TX_PORT_TS) || chs->params.ptp_rx) {
2587 err = mlx5e_ptp_open(priv, &chs->params, chs->c[0]->lag_port, &chs->ptp);
2589 goto err_close_channels;
2593 err = mlx5e_qos_open_queues(priv, chs);
2598 mlx5e_health_channels_update(priv);
2604 mlx5e_ptp_close(chs->ptp);
2607 for (i--; i >= 0; i--)
2608 mlx5e_close_channel(chs->c[i]);
2617 static void mlx5e_activate_channels(struct mlx5e_priv *priv, struct mlx5e_channels *chs)
2621 for (i = 0; i < chs->num; i++)
2622 mlx5e_activate_channel(chs->c[i]);
2625 mlx5e_qos_activate_queues(priv);
2627 for (i = 0; i < chs->num; i++)
2628 mlx5e_trigger_napi_icosq(chs->c[i]);
2631 mlx5e_ptp_activate_channel(chs->ptp);
2634 static int mlx5e_wait_channels_min_rx_wqes(struct mlx5e_channels *chs)
2639 for (i = 0; i < chs->num; i++) {
2640 int timeout = err ? 0 : MLX5E_RQ_WQES_TIMEOUT;
2641 struct mlx5e_channel *c = chs->c[i];
2643 if (test_bit(MLX5E_CHANNEL_STATE_XSK, c->state))
2646 err |= mlx5e_wait_for_min_rx_wqes(&c->rq, timeout);
2648 /* Don't wait on the XSK RQ, because the newer xdpsock sample
2649 * doesn't provide any Fill Ring entries at the setup stage.
2653 return err ? -ETIMEDOUT : 0;
2656 static void mlx5e_deactivate_channels(struct mlx5e_channels *chs)
2661 mlx5e_ptp_deactivate_channel(chs->ptp);
2663 for (i = 0; i < chs->num; i++)
2664 mlx5e_deactivate_channel(chs->c[i]);
2667 void mlx5e_close_channels(struct mlx5e_channels *chs)
2672 mlx5e_ptp_close(chs->ptp);
2675 for (i = 0; i < chs->num; i++)
2676 mlx5e_close_channel(chs->c[i]);
2682 static int mlx5e_modify_tirs_packet_merge(struct mlx5e_priv *priv)
2684 struct mlx5e_rx_res *res = priv->rx_res;
2686 return mlx5e_rx_res_packet_merge_set_param(res, &priv->channels.params.packet_merge);
2689 static MLX5E_DEFINE_PREACTIVATE_WRAPPER_CTX(mlx5e_modify_tirs_packet_merge);
2691 static int mlx5e_set_mtu(struct mlx5_core_dev *mdev,
2692 struct mlx5e_params *params, u16 mtu)
2694 u16 hw_mtu = MLX5E_SW2HW_MTU(params, mtu);
2697 err = mlx5_set_port_mtu(mdev, hw_mtu, 1);
2701 /* Update vport context MTU */
2702 mlx5_modify_nic_vport_mtu(mdev, hw_mtu);
2706 static void mlx5e_query_mtu(struct mlx5_core_dev *mdev,
2707 struct mlx5e_params *params, u16 *mtu)
2712 err = mlx5_query_nic_vport_mtu(mdev, &hw_mtu);
2713 if (err || !hw_mtu) /* fallback to port oper mtu */
2714 mlx5_query_port_oper_mtu(mdev, &hw_mtu, 1);
2716 *mtu = MLX5E_HW2SW_MTU(params, hw_mtu);
2719 int mlx5e_set_dev_port_mtu(struct mlx5e_priv *priv)
2721 struct mlx5e_params *params = &priv->channels.params;
2722 struct net_device *netdev = priv->netdev;
2723 struct mlx5_core_dev *mdev = priv->mdev;
2727 err = mlx5e_set_mtu(mdev, params, params->sw_mtu);
2731 mlx5e_query_mtu(mdev, params, &mtu);
2732 if (mtu != params->sw_mtu)
2733 netdev_warn(netdev, "%s: VPort MTU %d is different than netdev mtu %d\n",
2734 __func__, mtu, params->sw_mtu);
2736 params->sw_mtu = mtu;
2740 MLX5E_DEFINE_PREACTIVATE_WRAPPER_CTX(mlx5e_set_dev_port_mtu);
2742 void mlx5e_set_netdev_mtu_boundaries(struct mlx5e_priv *priv)
2744 struct mlx5e_params *params = &priv->channels.params;
2745 struct net_device *netdev = priv->netdev;
2746 struct mlx5_core_dev *mdev = priv->mdev;
2749 /* MTU range: 68 - hw-specific max */
2750 netdev->min_mtu = ETH_MIN_MTU;
2752 mlx5_query_port_max_mtu(mdev, &max_mtu, 1);
2753 netdev->max_mtu = min_t(unsigned int, MLX5E_HW2SW_MTU(params, max_mtu),
2757 static int mlx5e_netdev_set_tcs(struct net_device *netdev, u16 nch, u8 ntc,
2758 struct netdev_tc_txq *tc_to_txq)
2762 netdev_reset_tc(netdev);
2767 err = netdev_set_num_tc(netdev, ntc);
2769 netdev_WARN(netdev, "netdev_set_num_tc failed (%d), ntc = %d\n", err, ntc);
2773 for (tc = 0; tc < ntc; tc++) {
2776 count = tc_to_txq[tc].count;
2777 offset = tc_to_txq[tc].offset;
2778 netdev_set_tc_queue(netdev, tc, count, offset);
2784 int mlx5e_update_tx_netdev_queues(struct mlx5e_priv *priv)
2786 int nch, ntc, num_txqs, err;
2790 qos_queues = mlx5e_htb_cur_leaf_nodes(priv->htb);
2792 nch = priv->channels.params.num_channels;
2793 ntc = mlx5e_get_dcb_num_tc(&priv->channels.params);
2794 num_txqs = nch * ntc + qos_queues;
2795 if (MLX5E_GET_PFLAG(&priv->channels.params, MLX5E_PFLAG_TX_PORT_TS))
2798 netdev_dbg(priv->netdev, "Setting num_txqs %d\n", num_txqs);
2799 err = netif_set_real_num_tx_queues(priv->netdev, num_txqs);
2801 netdev_warn(priv->netdev, "netif_set_real_num_tx_queues failed, %d\n", err);
2806 static int mlx5e_update_netdev_queues(struct mlx5e_priv *priv)
2808 struct netdev_tc_txq old_tc_to_txq[TC_MAX_QUEUE], *tc_to_txq;
2809 struct net_device *netdev = priv->netdev;
2810 int old_num_txqs, old_ntc;
2815 old_num_txqs = netdev->real_num_tx_queues;
2816 old_ntc = netdev->num_tc ? : 1;
2817 for (i = 0; i < ARRAY_SIZE(old_tc_to_txq); i++)
2818 old_tc_to_txq[i] = netdev->tc_to_txq[i];
2820 nch = priv->channels.params.num_channels;
2821 ntc = priv->channels.params.mqprio.num_tc;
2822 tc_to_txq = priv->channels.params.mqprio.tc_to_txq;
2824 err = mlx5e_netdev_set_tcs(netdev, nch, ntc, tc_to_txq);
2827 err = mlx5e_update_tx_netdev_queues(priv);
2830 err = netif_set_real_num_rx_queues(netdev, nch);
2832 netdev_warn(netdev, "netif_set_real_num_rx_queues failed, %d\n", err);
2839 /* netif_set_real_num_rx_queues could fail only when nch increased. Only
2840 * one of nch and ntc is changed in this function. That means, the call
2841 * to netif_set_real_num_tx_queues below should not fail, because it
2842 * decreases the number of TX queues.
2844 WARN_ON_ONCE(netif_set_real_num_tx_queues(netdev, old_num_txqs));
2847 WARN_ON_ONCE(mlx5e_netdev_set_tcs(netdev, old_num_txqs / old_ntc, old_ntc,
2853 static MLX5E_DEFINE_PREACTIVATE_WRAPPER_CTX(mlx5e_update_netdev_queues);
2855 static void mlx5e_set_default_xps_cpumasks(struct mlx5e_priv *priv,
2856 struct mlx5e_params *params)
2858 struct mlx5_core_dev *mdev = priv->mdev;
2859 int num_comp_vectors, ix, irq;
2861 num_comp_vectors = mlx5_comp_vectors_max(mdev);
2863 for (ix = 0; ix < params->num_channels; ix++) {
2864 cpumask_clear(priv->scratchpad.cpumask);
2866 for (irq = ix; irq < num_comp_vectors; irq += params->num_channels) {
2867 int cpu = mlx5_comp_vector_get_cpu(mdev, irq);
2869 cpumask_set_cpu(cpu, priv->scratchpad.cpumask);
2872 netif_set_xps_queue(priv->netdev, priv->scratchpad.cpumask, ix);
2876 static int mlx5e_num_channels_changed(struct mlx5e_priv *priv)
2878 u16 count = priv->channels.params.num_channels;
2881 err = mlx5e_update_netdev_queues(priv);
2885 mlx5e_set_default_xps_cpumasks(priv, &priv->channels.params);
2887 /* This function may be called on attach, before priv->rx_res is created. */
2888 if (!netif_is_rxfh_configured(priv->netdev) && priv->rx_res)
2889 mlx5e_rx_res_rss_set_indir_uniform(priv->rx_res, count);
2894 MLX5E_DEFINE_PREACTIVATE_WRAPPER_CTX(mlx5e_num_channels_changed);
2896 static void mlx5e_build_txq_maps(struct mlx5e_priv *priv)
2898 int i, ch, tc, num_tc;
2900 ch = priv->channels.num;
2901 num_tc = mlx5e_get_dcb_num_tc(&priv->channels.params);
2903 for (i = 0; i < ch; i++) {
2904 for (tc = 0; tc < num_tc; tc++) {
2905 struct mlx5e_channel *c = priv->channels.c[i];
2906 struct mlx5e_txqsq *sq = &c->sq[tc];
2908 priv->txq2sq[sq->txq_ix] = sq;
2912 if (!priv->channels.ptp)
2915 if (!test_bit(MLX5E_PTP_STATE_TX, priv->channels.ptp->state))
2918 for (tc = 0; tc < num_tc; tc++) {
2919 struct mlx5e_ptp *c = priv->channels.ptp;
2920 struct mlx5e_txqsq *sq = &c->ptpsq[tc].txqsq;
2922 priv->txq2sq[sq->txq_ix] = sq;
2926 /* Make the change to txq2sq visible before the queue is started.
2927 * As mlx5e_xmit runs under a spinlock, there is an implicit ACQUIRE,
2928 * which pairs with this barrier.
2933 void mlx5e_activate_priv_channels(struct mlx5e_priv *priv)
2935 mlx5e_build_txq_maps(priv);
2936 mlx5e_activate_channels(priv, &priv->channels);
2937 mlx5e_xdp_tx_enable(priv);
2939 /* dev_watchdog() wants all TX queues to be started when the carrier is
2940 * OK, including the ones in range real_num_tx_queues..num_tx_queues-1.
2941 * Make it happy to avoid TX timeout false alarms.
2943 netif_tx_start_all_queues(priv->netdev);
2945 if (mlx5e_is_vport_rep(priv))
2946 mlx5e_rep_activate_channels(priv);
2948 mlx5e_wait_channels_min_rx_wqes(&priv->channels);
2951 mlx5e_rx_res_channels_activate(priv->rx_res, &priv->channels);
2954 void mlx5e_deactivate_priv_channels(struct mlx5e_priv *priv)
2957 mlx5e_rx_res_channels_deactivate(priv->rx_res);
2959 if (mlx5e_is_vport_rep(priv))
2960 mlx5e_rep_deactivate_channels(priv);
2962 /* The results of ndo_select_queue are unreliable, while netdev config
2963 * is being changed (real_num_tx_queues, num_tc). Stop all queues to
2964 * prevent ndo_start_xmit from being called, so that it can assume that
2965 * the selected queue is always valid.
2967 netif_tx_disable(priv->netdev);
2969 mlx5e_xdp_tx_disable(priv);
2970 mlx5e_deactivate_channels(&priv->channels);
2973 static int mlx5e_switch_priv_params(struct mlx5e_priv *priv,
2974 struct mlx5e_params *new_params,
2975 mlx5e_fp_preactivate preactivate,
2978 struct mlx5e_params old_params;
2980 old_params = priv->channels.params;
2981 priv->channels.params = *new_params;
2986 err = preactivate(priv, context);
2988 priv->channels.params = old_params;
2996 static int mlx5e_switch_priv_channels(struct mlx5e_priv *priv,
2997 struct mlx5e_channels *new_chs,
2998 mlx5e_fp_preactivate preactivate,
3001 struct net_device *netdev = priv->netdev;
3002 struct mlx5e_channels old_chs;
3006 carrier_ok = netif_carrier_ok(netdev);
3007 netif_carrier_off(netdev);
3009 mlx5e_deactivate_priv_channels(priv);
3011 old_chs = priv->channels;
3012 priv->channels = *new_chs;
3014 /* New channels are ready to roll, call the preactivate hook if needed
3015 * to modify HW settings or update kernel parameters.
3018 err = preactivate(priv, context);
3020 priv->channels = old_chs;
3025 mlx5e_close_channels(&old_chs);
3026 priv->profile->update_rx(priv);
3028 mlx5e_selq_apply(&priv->selq);
3030 mlx5e_activate_priv_channels(priv);
3032 /* return carrier back if needed */
3034 netif_carrier_on(netdev);
3039 int mlx5e_safe_switch_params(struct mlx5e_priv *priv,
3040 struct mlx5e_params *params,
3041 mlx5e_fp_preactivate preactivate,
3042 void *context, bool reset)
3044 struct mlx5e_channels *new_chs;
3047 reset &= test_bit(MLX5E_STATE_OPENED, &priv->state);
3049 return mlx5e_switch_priv_params(priv, params, preactivate, context);
3051 new_chs = kzalloc(sizeof(*new_chs), GFP_KERNEL);
3054 new_chs->params = *params;
3056 mlx5e_selq_prepare_params(&priv->selq, &new_chs->params);
3058 err = mlx5e_open_channels(priv, new_chs);
3060 goto err_cancel_selq;
3062 err = mlx5e_switch_priv_channels(priv, new_chs, preactivate, context);
3070 mlx5e_close_channels(new_chs);
3073 mlx5e_selq_cancel(&priv->selq);
3078 int mlx5e_safe_reopen_channels(struct mlx5e_priv *priv)
3080 return mlx5e_safe_switch_params(priv, &priv->channels.params, NULL, NULL, true);
3083 void mlx5e_timestamp_init(struct mlx5e_priv *priv)
3085 priv->tstamp.tx_type = HWTSTAMP_TX_OFF;
3086 priv->tstamp.rx_filter = HWTSTAMP_FILTER_NONE;
3089 static void mlx5e_modify_admin_state(struct mlx5_core_dev *mdev,
3090 enum mlx5_port_status state)
3092 struct mlx5_eswitch *esw = mdev->priv.eswitch;
3093 int vport_admin_state;
3095 mlx5_set_port_admin_status(mdev, state);
3097 if (mlx5_eswitch_mode(mdev) == MLX5_ESWITCH_OFFLOADS ||
3098 !MLX5_CAP_GEN(mdev, uplink_follow))
3101 if (state == MLX5_PORT_UP)
3102 vport_admin_state = MLX5_VPORT_ADMIN_STATE_AUTO;
3104 vport_admin_state = MLX5_VPORT_ADMIN_STATE_DOWN;
3106 mlx5_eswitch_set_vport_state(esw, MLX5_VPORT_UPLINK, vport_admin_state);
3109 int mlx5e_open_locked(struct net_device *netdev)
3111 struct mlx5e_priv *priv = netdev_priv(netdev);
3114 mlx5e_selq_prepare_params(&priv->selq, &priv->channels.params);
3116 set_bit(MLX5E_STATE_OPENED, &priv->state);
3118 err = mlx5e_open_channels(priv, &priv->channels);
3120 goto err_clear_state_opened_flag;
3122 err = priv->profile->update_rx(priv);
3124 goto err_close_channels;
3126 mlx5e_selq_apply(&priv->selq);
3127 mlx5e_activate_priv_channels(priv);
3128 mlx5e_apply_traps(priv, true);
3129 if (priv->profile->update_carrier)
3130 priv->profile->update_carrier(priv);
3132 mlx5e_queue_update_stats(priv);
3136 mlx5e_close_channels(&priv->channels);
3137 err_clear_state_opened_flag:
3138 clear_bit(MLX5E_STATE_OPENED, &priv->state);
3139 mlx5e_selq_cancel(&priv->selq);
3143 int mlx5e_open(struct net_device *netdev)
3145 struct mlx5e_priv *priv = netdev_priv(netdev);
3148 mutex_lock(&priv->state_lock);
3149 err = mlx5e_open_locked(netdev);
3151 mlx5e_modify_admin_state(priv->mdev, MLX5_PORT_UP);
3152 mutex_unlock(&priv->state_lock);
3157 int mlx5e_close_locked(struct net_device *netdev)
3159 struct mlx5e_priv *priv = netdev_priv(netdev);
3161 /* May already be CLOSED in case a previous configuration operation
3162 * (e.g RX/TX queue size change) that involves close&open failed.
3164 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
3167 mlx5e_apply_traps(priv, false);
3168 clear_bit(MLX5E_STATE_OPENED, &priv->state);
3170 netif_carrier_off(priv->netdev);
3171 mlx5e_deactivate_priv_channels(priv);
3172 mlx5e_close_channels(&priv->channels);
3177 int mlx5e_close(struct net_device *netdev)
3179 struct mlx5e_priv *priv = netdev_priv(netdev);
3182 if (!netif_device_present(netdev))
3185 mutex_lock(&priv->state_lock);
3186 mlx5e_modify_admin_state(priv->mdev, MLX5_PORT_DOWN);
3187 err = mlx5e_close_locked(netdev);
3188 mutex_unlock(&priv->state_lock);
3193 static void mlx5e_free_drop_rq(struct mlx5e_rq *rq)
3195 mlx5_wq_destroy(&rq->wq_ctrl);
3198 static int mlx5e_alloc_drop_rq(struct mlx5_core_dev *mdev,
3199 struct mlx5e_rq *rq,
3200 struct mlx5e_rq_param *param)
3202 void *rqc = param->rqc;
3203 void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
3206 param->wq.db_numa_node = param->wq.buf_numa_node;
3208 err = mlx5_wq_cyc_create(mdev, ¶m->wq, rqc_wq, &rq->wqe.wq,
3213 /* Mark as unused given "Drop-RQ" packets never reach XDP */
3214 xdp_rxq_info_unused(&rq->xdp_rxq);
3221 static int mlx5e_alloc_drop_cq(struct mlx5e_priv *priv,
3222 struct mlx5e_cq *cq,
3223 struct mlx5e_cq_param *param)
3225 struct mlx5_core_dev *mdev = priv->mdev;
3227 param->wq.buf_numa_node = dev_to_node(mlx5_core_dma_dev(mdev));
3228 param->wq.db_numa_node = dev_to_node(mlx5_core_dma_dev(mdev));
3230 return mlx5e_alloc_cq_common(priv, param, cq);
3233 int mlx5e_open_drop_rq(struct mlx5e_priv *priv,
3234 struct mlx5e_rq *drop_rq)
3236 struct mlx5_core_dev *mdev = priv->mdev;
3237 struct mlx5e_cq_param cq_param = {};
3238 struct mlx5e_rq_param rq_param = {};
3239 struct mlx5e_cq *cq = &drop_rq->cq;
3242 mlx5e_build_drop_rq_param(mdev, priv->drop_rq_q_counter, &rq_param);
3244 err = mlx5e_alloc_drop_cq(priv, cq, &cq_param);
3248 err = mlx5e_create_cq(cq, &cq_param);
3252 err = mlx5e_alloc_drop_rq(mdev, drop_rq, &rq_param);
3254 goto err_destroy_cq;
3256 err = mlx5e_create_rq(drop_rq, &rq_param);
3260 err = mlx5e_modify_rq_state(drop_rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
3262 mlx5_core_warn(priv->mdev, "modify_rq_state failed, rx_if_down_packets won't be counted %d\n", err);
3267 mlx5e_free_drop_rq(drop_rq);
3270 mlx5e_destroy_cq(cq);
3278 void mlx5e_close_drop_rq(struct mlx5e_rq *drop_rq)
3280 mlx5e_destroy_rq(drop_rq);
3281 mlx5e_free_drop_rq(drop_rq);
3282 mlx5e_destroy_cq(&drop_rq->cq);
3283 mlx5e_free_cq(&drop_rq->cq);
3286 int mlx5e_create_tis(struct mlx5_core_dev *mdev, void *in, u32 *tisn)
3288 void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
3290 MLX5_SET(tisc, tisc, transport_domain, mdev->mlx5e_res.hw_objs.td.tdn);
3292 if (MLX5_GET(tisc, tisc, tls_en))
3293 MLX5_SET(tisc, tisc, pd, mdev->mlx5e_res.hw_objs.pdn);
3295 if (mlx5_lag_is_lacp_owner(mdev))
3296 MLX5_SET(tisc, tisc, strict_lag_tx_port_affinity, 1);
3298 return mlx5_core_create_tis(mdev, in, tisn);
3301 void mlx5e_destroy_tis(struct mlx5_core_dev *mdev, u32 tisn)
3303 mlx5_core_destroy_tis(mdev, tisn);
3306 void mlx5e_destroy_tises(struct mlx5e_priv *priv)
3310 for (i = 0; i < mlx5e_get_num_lag_ports(priv->mdev); i++)
3311 for (tc = 0; tc < priv->profile->max_tc; tc++)
3312 mlx5e_destroy_tis(priv->mdev, priv->tisn[i][tc]);
3315 static bool mlx5e_lag_should_assign_affinity(struct mlx5_core_dev *mdev)
3317 return MLX5_CAP_GEN(mdev, lag_tx_port_affinity) && mlx5e_get_num_lag_ports(mdev) > 1;
3320 int mlx5e_create_tises(struct mlx5e_priv *priv)
3325 for (i = 0; i < mlx5e_get_num_lag_ports(priv->mdev); i++) {
3326 for (tc = 0; tc < priv->profile->max_tc; tc++) {
3327 u32 in[MLX5_ST_SZ_DW(create_tis_in)] = {};
3330 tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
3332 MLX5_SET(tisc, tisc, prio, tc << 1);
3334 if (mlx5e_lag_should_assign_affinity(priv->mdev))
3335 MLX5_SET(tisc, tisc, lag_tx_port_affinity, i + 1);
3337 err = mlx5e_create_tis(priv->mdev, in, &priv->tisn[i][tc]);
3339 goto err_close_tises;
3346 for (; i >= 0; i--) {
3347 for (tc--; tc >= 0; tc--)
3348 mlx5e_destroy_tis(priv->mdev, priv->tisn[i][tc]);
3349 tc = priv->profile->max_tc;
3355 static void mlx5e_cleanup_nic_tx(struct mlx5e_priv *priv)
3357 if (priv->mqprio_rl) {
3358 mlx5e_mqprio_rl_cleanup(priv->mqprio_rl);
3359 mlx5e_mqprio_rl_free(priv->mqprio_rl);
3360 priv->mqprio_rl = NULL;
3362 mlx5e_accel_cleanup_tx(priv);
3363 mlx5e_destroy_tises(priv);
3366 static int mlx5e_modify_channels_vsd(struct mlx5e_channels *chs, bool vsd)
3371 for (i = 0; i < chs->num; i++) {
3372 err = mlx5e_modify_rq_vsd(&chs->c[i]->rq, vsd);
3376 if (chs->ptp && test_bit(MLX5E_PTP_STATE_RX, chs->ptp->state))
3377 return mlx5e_modify_rq_vsd(&chs->ptp->rq, vsd);
3382 static void mlx5e_mqprio_build_default_tc_to_txq(struct netdev_tc_txq *tc_to_txq,
3387 memset(tc_to_txq, 0, sizeof(*tc_to_txq) * TC_MAX_QUEUE);
3389 /* Map netdev TCs to offset 0.
3390 * We have our own UP to TXQ mapping for DCB mode of QoS
3392 for (tc = 0; tc < ntc; tc++) {
3393 tc_to_txq[tc] = (struct netdev_tc_txq) {
3400 static void mlx5e_mqprio_build_tc_to_txq(struct netdev_tc_txq *tc_to_txq,
3401 struct tc_mqprio_qopt *qopt)
3405 for (tc = 0; tc < TC_MAX_QUEUE; tc++) {
3406 tc_to_txq[tc] = (struct netdev_tc_txq) {
3407 .count = qopt->count[tc],
3408 .offset = qopt->offset[tc],
3413 static void mlx5e_params_mqprio_dcb_set(struct mlx5e_params *params, u8 num_tc)
3415 params->mqprio.mode = TC_MQPRIO_MODE_DCB;
3416 params->mqprio.num_tc = num_tc;
3417 mlx5e_mqprio_build_default_tc_to_txq(params->mqprio.tc_to_txq, num_tc,
3418 params->num_channels);
3421 static void mlx5e_mqprio_rl_update_params(struct mlx5e_params *params,
3422 struct mlx5e_mqprio_rl *rl)
3426 for (tc = 0; tc < TC_MAX_QUEUE; tc++) {
3430 mlx5e_mqprio_rl_get_node_hw_id(rl, tc, &hw_id);
3431 params->mqprio.channel.hw_id[tc] = hw_id;
3435 static void mlx5e_params_mqprio_channel_set(struct mlx5e_params *params,
3436 struct tc_mqprio_qopt_offload *mqprio,
3437 struct mlx5e_mqprio_rl *rl)
3441 params->mqprio.mode = TC_MQPRIO_MODE_CHANNEL;
3442 params->mqprio.num_tc = mqprio->qopt.num_tc;
3444 for (tc = 0; tc < TC_MAX_QUEUE; tc++)
3445 params->mqprio.channel.max_rate[tc] = mqprio->max_rate[tc];
3447 mlx5e_mqprio_rl_update_params(params, rl);
3448 mlx5e_mqprio_build_tc_to_txq(params->mqprio.tc_to_txq, &mqprio->qopt);
3451 static void mlx5e_params_mqprio_reset(struct mlx5e_params *params)
3453 mlx5e_params_mqprio_dcb_set(params, 1);
3456 static int mlx5e_setup_tc_mqprio_dcb(struct mlx5e_priv *priv,
3457 struct tc_mqprio_qopt *mqprio)
3459 struct mlx5e_params new_params;
3460 u8 tc = mqprio->num_tc;
3463 mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
3465 if (tc && tc != MLX5E_MAX_NUM_TC)
3468 new_params = priv->channels.params;
3469 mlx5e_params_mqprio_dcb_set(&new_params, tc ? tc : 1);
3471 err = mlx5e_safe_switch_params(priv, &new_params,
3472 mlx5e_num_channels_changed_ctx, NULL, true);
3474 if (!err && priv->mqprio_rl) {
3475 mlx5e_mqprio_rl_cleanup(priv->mqprio_rl);
3476 mlx5e_mqprio_rl_free(priv->mqprio_rl);
3477 priv->mqprio_rl = NULL;
3480 priv->max_opened_tc = max_t(u8, priv->max_opened_tc,
3481 mlx5e_get_dcb_num_tc(&priv->channels.params));
3485 static int mlx5e_mqprio_channel_validate(struct mlx5e_priv *priv,
3486 struct tc_mqprio_qopt_offload *mqprio)
3488 struct net_device *netdev = priv->netdev;
3489 struct mlx5e_ptp *ptp_channel;
3493 ptp_channel = priv->channels.ptp;
3494 if (ptp_channel && test_bit(MLX5E_PTP_STATE_TX, ptp_channel->state)) {
3496 "Cannot activate MQPRIO mode channel since it conflicts with TX port TS\n");
3500 if (mqprio->qopt.offset[0] != 0 || mqprio->qopt.num_tc < 1 ||
3501 mqprio->qopt.num_tc > MLX5E_MAX_NUM_MQPRIO_CH_TC)
3504 for (i = 0; i < mqprio->qopt.num_tc; i++) {
3505 if (!mqprio->qopt.count[i]) {
3506 netdev_err(netdev, "Zero size for queue-group (%d) is not supported\n", i);
3509 if (mqprio->min_rate[i]) {
3510 netdev_err(netdev, "Min tx rate is not supported\n");
3514 if (mqprio->max_rate[i]) {
3517 err = mlx5e_qos_bytes_rate_check(priv->mdev, mqprio->max_rate[i]);
3522 if (mqprio->qopt.offset[i] != agg_count) {
3523 netdev_err(netdev, "Discontinuous queues config is not supported\n");
3526 agg_count += mqprio->qopt.count[i];
3529 if (priv->channels.params.num_channels != agg_count) {
3530 netdev_err(netdev, "Num of queues (%d) does not match available (%d)\n",
3531 agg_count, priv->channels.params.num_channels);
3538 static bool mlx5e_mqprio_rate_limit(u8 num_tc, u64 max_rate[])
3542 for (tc = 0; tc < num_tc; tc++)
3548 static struct mlx5e_mqprio_rl *mlx5e_mqprio_rl_create(struct mlx5_core_dev *mdev,
3549 u8 num_tc, u64 max_rate[])
3551 struct mlx5e_mqprio_rl *rl;
3554 if (!mlx5e_mqprio_rate_limit(num_tc, max_rate))
3557 rl = mlx5e_mqprio_rl_alloc();
3559 return ERR_PTR(-ENOMEM);
3561 err = mlx5e_mqprio_rl_init(rl, mdev, num_tc, max_rate);
3563 mlx5e_mqprio_rl_free(rl);
3564 return ERR_PTR(err);
3570 static int mlx5e_setup_tc_mqprio_channel(struct mlx5e_priv *priv,
3571 struct tc_mqprio_qopt_offload *mqprio)
3573 mlx5e_fp_preactivate preactivate;
3574 struct mlx5e_params new_params;
3575 struct mlx5e_mqprio_rl *rl;
3579 err = mlx5e_mqprio_channel_validate(priv, mqprio);
3583 rl = mlx5e_mqprio_rl_create(priv->mdev, mqprio->qopt.num_tc, mqprio->max_rate);
3587 new_params = priv->channels.params;
3588 mlx5e_params_mqprio_channel_set(&new_params, mqprio, rl);
3590 nch_changed = mlx5e_get_dcb_num_tc(&priv->channels.params) > 1;
3591 preactivate = nch_changed ? mlx5e_num_channels_changed_ctx :
3592 mlx5e_update_netdev_queues_ctx;
3593 err = mlx5e_safe_switch_params(priv, &new_params, preactivate, NULL, true);
3596 mlx5e_mqprio_rl_cleanup(rl);
3597 mlx5e_mqprio_rl_free(rl);
3602 if (priv->mqprio_rl) {
3603 mlx5e_mqprio_rl_cleanup(priv->mqprio_rl);
3604 mlx5e_mqprio_rl_free(priv->mqprio_rl);
3606 priv->mqprio_rl = rl;
3611 static int mlx5e_setup_tc_mqprio(struct mlx5e_priv *priv,
3612 struct tc_mqprio_qopt_offload *mqprio)
3614 /* MQPRIO is another toplevel qdisc that can't be attached
3615 * simultaneously with the offloaded HTB.
3617 if (WARN_ON(mlx5e_selq_is_htb_enabled(&priv->selq)))
3620 switch (mqprio->mode) {
3621 case TC_MQPRIO_MODE_DCB:
3622 return mlx5e_setup_tc_mqprio_dcb(priv, &mqprio->qopt);
3623 case TC_MQPRIO_MODE_CHANNEL:
3624 return mlx5e_setup_tc_mqprio_channel(priv, mqprio);
3630 static LIST_HEAD(mlx5e_block_cb_list);
3632 static int mlx5e_setup_tc(struct net_device *dev, enum tc_setup_type type,
3635 struct mlx5e_priv *priv = netdev_priv(dev);
3636 bool tc_unbind = false;
3639 if (type == TC_SETUP_BLOCK &&
3640 ((struct flow_block_offload *)type_data)->command == FLOW_BLOCK_UNBIND)
3643 if (!netif_device_present(dev) && !tc_unbind)
3647 case TC_SETUP_BLOCK: {
3648 struct flow_block_offload *f = type_data;
3650 f->unlocked_driver_cb = true;
3651 return flow_block_cb_setup_simple(type_data,
3652 &mlx5e_block_cb_list,
3653 mlx5e_setup_tc_block_cb,
3656 case TC_SETUP_QDISC_MQPRIO:
3657 mutex_lock(&priv->state_lock);
3658 err = mlx5e_setup_tc_mqprio(priv, type_data);
3659 mutex_unlock(&priv->state_lock);
3661 case TC_SETUP_QDISC_HTB:
3662 mutex_lock(&priv->state_lock);
3663 err = mlx5e_htb_setup_tc(priv, type_data);
3664 mutex_unlock(&priv->state_lock);
3671 void mlx5e_fold_sw_stats64(struct mlx5e_priv *priv, struct rtnl_link_stats64 *s)
3675 for (i = 0; i < priv->stats_nch; i++) {
3676 struct mlx5e_channel_stats *channel_stats = priv->channel_stats[i];
3677 struct mlx5e_rq_stats *xskrq_stats = &channel_stats->xskrq;
3678 struct mlx5e_rq_stats *rq_stats = &channel_stats->rq;
3681 s->rx_packets += rq_stats->packets + xskrq_stats->packets;
3682 s->rx_bytes += rq_stats->bytes + xskrq_stats->bytes;
3683 s->multicast += rq_stats->mcast_packets + xskrq_stats->mcast_packets;
3685 for (j = 0; j < priv->max_opened_tc; j++) {
3686 struct mlx5e_sq_stats *sq_stats = &channel_stats->sq[j];
3688 s->tx_packets += sq_stats->packets;
3689 s->tx_bytes += sq_stats->bytes;
3690 s->tx_dropped += sq_stats->dropped;
3693 if (priv->tx_ptp_opened) {
3694 for (i = 0; i < priv->max_opened_tc; i++) {
3695 struct mlx5e_sq_stats *sq_stats = &priv->ptp_stats.sq[i];
3697 s->tx_packets += sq_stats->packets;
3698 s->tx_bytes += sq_stats->bytes;
3699 s->tx_dropped += sq_stats->dropped;
3702 if (priv->rx_ptp_opened) {
3703 struct mlx5e_rq_stats *rq_stats = &priv->ptp_stats.rq;
3705 s->rx_packets += rq_stats->packets;
3706 s->rx_bytes += rq_stats->bytes;
3707 s->multicast += rq_stats->mcast_packets;
3712 mlx5e_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats)
3714 struct mlx5e_priv *priv = netdev_priv(dev);
3715 struct mlx5e_pport_stats *pstats = &priv->stats.pport;
3717 if (!netif_device_present(dev))
3720 /* In switchdev mode, monitor counters doesn't monitor
3721 * rx/tx stats of 802_3. The update stats mechanism
3722 * should keep the 802_3 layout counters updated
3724 if (!mlx5e_monitor_counter_supported(priv) ||
3725 mlx5e_is_uplink_rep(priv)) {
3726 /* update HW stats in background for next time */
3727 mlx5e_queue_update_stats(priv);
3730 if (mlx5e_is_uplink_rep(priv)) {
3731 struct mlx5e_vport_stats *vstats = &priv->stats.vport;
3733 stats->rx_packets = PPORT_802_3_GET(pstats, a_frames_received_ok);
3734 stats->rx_bytes = PPORT_802_3_GET(pstats, a_octets_received_ok);
3735 stats->tx_packets = PPORT_802_3_GET(pstats, a_frames_transmitted_ok);
3736 stats->tx_bytes = PPORT_802_3_GET(pstats, a_octets_transmitted_ok);
3738 /* vport multicast also counts packets that are dropped due to steering
3739 * or rx out of buffer
3741 stats->multicast = VPORT_COUNTER_GET(vstats, received_eth_multicast.packets);
3743 mlx5e_fold_sw_stats64(priv, stats);
3746 stats->rx_dropped = priv->stats.qcnt.rx_out_of_buffer;
3748 stats->rx_length_errors =
3749 PPORT_802_3_GET(pstats, a_in_range_length_errors) +
3750 PPORT_802_3_GET(pstats, a_out_of_range_length_field) +
3751 PPORT_802_3_GET(pstats, a_frame_too_long_errors) +
3752 VNIC_ENV_GET(&priv->stats.vnic, eth_wqe_too_small);
3753 stats->rx_crc_errors =
3754 PPORT_802_3_GET(pstats, a_frame_check_sequence_errors);
3755 stats->rx_frame_errors = PPORT_802_3_GET(pstats, a_alignment_errors);
3756 stats->tx_aborted_errors = PPORT_2863_GET(pstats, if_out_discards);
3757 stats->rx_errors = stats->rx_length_errors + stats->rx_crc_errors +
3758 stats->rx_frame_errors;
3759 stats->tx_errors = stats->tx_aborted_errors + stats->tx_carrier_errors;
3762 static void mlx5e_nic_set_rx_mode(struct mlx5e_priv *priv)
3764 if (mlx5e_is_uplink_rep(priv))
3765 return; /* no rx mode for uplink rep */
3767 queue_work(priv->wq, &priv->set_rx_mode_work);
3770 static void mlx5e_set_rx_mode(struct net_device *dev)
3772 struct mlx5e_priv *priv = netdev_priv(dev);
3774 mlx5e_nic_set_rx_mode(priv);
3777 static int mlx5e_set_mac(struct net_device *netdev, void *addr)
3779 struct mlx5e_priv *priv = netdev_priv(netdev);
3780 struct sockaddr *saddr = addr;
3782 if (!is_valid_ether_addr(saddr->sa_data))
3783 return -EADDRNOTAVAIL;
3785 netif_addr_lock_bh(netdev);
3786 eth_hw_addr_set(netdev, saddr->sa_data);
3787 netif_addr_unlock_bh(netdev);
3789 mlx5e_nic_set_rx_mode(priv);
3794 #define MLX5E_SET_FEATURE(features, feature, enable) \
3797 *features |= feature; \
3799 *features &= ~feature; \
3802 typedef int (*mlx5e_feature_handler)(struct net_device *netdev, bool enable);
3804 static int set_feature_lro(struct net_device *netdev, bool enable)
3806 struct mlx5e_priv *priv = netdev_priv(netdev);
3807 struct mlx5_core_dev *mdev = priv->mdev;
3808 struct mlx5e_params *cur_params;
3809 struct mlx5e_params new_params;
3813 mutex_lock(&priv->state_lock);
3815 cur_params = &priv->channels.params;
3816 new_params = *cur_params;
3819 new_params.packet_merge.type = MLX5E_PACKET_MERGE_LRO;
3820 else if (new_params.packet_merge.type == MLX5E_PACKET_MERGE_LRO)
3821 new_params.packet_merge.type = MLX5E_PACKET_MERGE_NONE;
3825 if (!(cur_params->packet_merge.type == MLX5E_PACKET_MERGE_SHAMPO &&
3826 new_params.packet_merge.type == MLX5E_PACKET_MERGE_LRO)) {
3827 if (cur_params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
3828 if (mlx5e_rx_mpwqe_is_linear_skb(mdev, cur_params, NULL) ==
3829 mlx5e_rx_mpwqe_is_linear_skb(mdev, &new_params, NULL))
3834 err = mlx5e_safe_switch_params(priv, &new_params,
3835 mlx5e_modify_tirs_packet_merge_ctx, NULL, reset);
3837 mutex_unlock(&priv->state_lock);
3841 static int set_feature_hw_gro(struct net_device *netdev, bool enable)
3843 struct mlx5e_priv *priv = netdev_priv(netdev);
3844 struct mlx5e_params new_params;
3848 mutex_lock(&priv->state_lock);
3849 new_params = priv->channels.params;
3852 new_params.packet_merge.type = MLX5E_PACKET_MERGE_SHAMPO;
3853 new_params.packet_merge.shampo.match_criteria_type =
3854 MLX5_RQC_SHAMPO_MATCH_CRITERIA_TYPE_EXTENDED;
3855 new_params.packet_merge.shampo.alignment_granularity =
3856 MLX5_RQC_SHAMPO_NO_MATCH_ALIGNMENT_GRANULARITY_STRIDE;
3857 } else if (new_params.packet_merge.type == MLX5E_PACKET_MERGE_SHAMPO) {
3858 new_params.packet_merge.type = MLX5E_PACKET_MERGE_NONE;
3863 err = mlx5e_safe_switch_params(priv, &new_params, NULL, NULL, reset);
3865 mutex_unlock(&priv->state_lock);
3869 static int set_feature_cvlan_filter(struct net_device *netdev, bool enable)
3871 struct mlx5e_priv *priv = netdev_priv(netdev);
3874 mlx5e_enable_cvlan_filter(priv->fs,
3875 !!(priv->netdev->flags & IFF_PROMISC));
3877 mlx5e_disable_cvlan_filter(priv->fs,
3878 !!(priv->netdev->flags & IFF_PROMISC));
3883 static int set_feature_hw_tc(struct net_device *netdev, bool enable)
3885 struct mlx5e_priv *priv = netdev_priv(netdev);
3888 #if IS_ENABLED(CONFIG_MLX5_CLS_ACT)
3889 int tc_flag = mlx5e_is_uplink_rep(priv) ? MLX5_TC_FLAG(ESW_OFFLOAD) :
3890 MLX5_TC_FLAG(NIC_OFFLOAD);
3891 if (!enable && mlx5e_tc_num_filters(priv, tc_flag)) {
3893 "Active offloaded tc filters, can't turn hw_tc_offload off\n");
3898 mutex_lock(&priv->state_lock);
3899 if (!enable && mlx5e_selq_is_htb_enabled(&priv->selq)) {
3900 netdev_err(netdev, "Active HTB offload, can't turn hw_tc_offload off\n");
3903 mutex_unlock(&priv->state_lock);
3908 static int set_feature_rx_all(struct net_device *netdev, bool enable)
3910 struct mlx5e_priv *priv = netdev_priv(netdev);
3911 struct mlx5_core_dev *mdev = priv->mdev;
3913 return mlx5_set_port_fcs(mdev, !enable);
3916 static int mlx5e_set_rx_port_ts(struct mlx5_core_dev *mdev, bool enable)
3918 u32 in[MLX5_ST_SZ_DW(pcmr_reg)] = {};
3919 bool supported, curr_state;
3922 if (!MLX5_CAP_GEN(mdev, ports_check))
3925 err = mlx5_query_ports_check(mdev, in, sizeof(in));
3929 supported = MLX5_GET(pcmr_reg, in, rx_ts_over_crc_cap);
3930 curr_state = MLX5_GET(pcmr_reg, in, rx_ts_over_crc);
3932 if (!supported || enable == curr_state)
3935 MLX5_SET(pcmr_reg, in, local_port, 1);
3936 MLX5_SET(pcmr_reg, in, rx_ts_over_crc, enable);
3938 return mlx5_set_ports_check(mdev, in, sizeof(in));
3941 static int mlx5e_set_rx_port_ts_wrap(struct mlx5e_priv *priv, void *ctx)
3943 struct mlx5_core_dev *mdev = priv->mdev;
3944 bool enable = *(bool *)ctx;
3946 return mlx5e_set_rx_port_ts(mdev, enable);
3949 static int set_feature_rx_fcs(struct net_device *netdev, bool enable)
3951 struct mlx5e_priv *priv = netdev_priv(netdev);
3952 struct mlx5e_channels *chs = &priv->channels;
3953 struct mlx5e_params new_params;
3955 bool rx_ts_over_crc = !enable;
3957 mutex_lock(&priv->state_lock);
3959 new_params = chs->params;
3960 new_params.scatter_fcs_en = enable;
3961 err = mlx5e_safe_switch_params(priv, &new_params, mlx5e_set_rx_port_ts_wrap,
3962 &rx_ts_over_crc, true);
3963 mutex_unlock(&priv->state_lock);
3967 static int set_feature_rx_vlan(struct net_device *netdev, bool enable)
3969 struct mlx5e_priv *priv = netdev_priv(netdev);
3972 mutex_lock(&priv->state_lock);
3974 mlx5e_fs_set_vlan_strip_disable(priv->fs, !enable);
3975 priv->channels.params.vlan_strip_disable = !enable;
3977 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
3980 err = mlx5e_modify_channels_vsd(&priv->channels, !enable);
3982 mlx5e_fs_set_vlan_strip_disable(priv->fs, enable);
3983 priv->channels.params.vlan_strip_disable = enable;
3986 mutex_unlock(&priv->state_lock);
3991 int mlx5e_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
3993 struct mlx5e_priv *priv = netdev_priv(dev);
3994 struct mlx5e_flow_steering *fs = priv->fs;
3996 if (mlx5e_is_uplink_rep(priv))
3997 return 0; /* no vlan table for uplink rep */
3999 return mlx5e_fs_vlan_rx_add_vid(fs, dev, proto, vid);
4002 int mlx5e_vlan_rx_kill_vid(struct net_device *dev, __be16 proto, u16 vid)
4004 struct mlx5e_priv *priv = netdev_priv(dev);
4005 struct mlx5e_flow_steering *fs = priv->fs;
4007 if (mlx5e_is_uplink_rep(priv))
4008 return 0; /* no vlan table for uplink rep */
4010 return mlx5e_fs_vlan_rx_kill_vid(fs, dev, proto, vid);
4013 #ifdef CONFIG_MLX5_EN_ARFS
4014 static int set_feature_arfs(struct net_device *netdev, bool enable)
4016 struct mlx5e_priv *priv = netdev_priv(netdev);
4020 err = mlx5e_arfs_enable(priv->fs);
4022 err = mlx5e_arfs_disable(priv->fs);
4028 static int mlx5e_handle_feature(struct net_device *netdev,
4029 netdev_features_t *features,
4030 netdev_features_t feature,
4031 mlx5e_feature_handler feature_handler)
4033 netdev_features_t changes = *features ^ netdev->features;
4034 bool enable = !!(*features & feature);
4037 if (!(changes & feature))
4040 err = feature_handler(netdev, enable);
4042 MLX5E_SET_FEATURE(features, feature, !enable);
4043 netdev_err(netdev, "%s feature %pNF failed, err %d\n",
4044 enable ? "Enable" : "Disable", &feature, err);
4051 void mlx5e_set_xdp_feature(struct net_device *netdev)
4053 struct mlx5e_priv *priv = netdev_priv(netdev);
4054 struct mlx5e_params *params = &priv->channels.params;
4057 if (params->packet_merge.type != MLX5E_PACKET_MERGE_NONE) {
4058 xdp_clear_features_flag(netdev);
4062 val = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
4063 NETDEV_XDP_ACT_XSK_ZEROCOPY |
4064 NETDEV_XDP_ACT_RX_SG |
4065 NETDEV_XDP_ACT_NDO_XMIT |
4066 NETDEV_XDP_ACT_NDO_XMIT_SG;
4067 xdp_set_features_flag(netdev, val);
4070 int mlx5e_set_features(struct net_device *netdev, netdev_features_t features)
4072 netdev_features_t oper_features = features;
4075 #define MLX5E_HANDLE_FEATURE(feature, handler) \
4076 mlx5e_handle_feature(netdev, &oper_features, feature, handler)
4078 err |= MLX5E_HANDLE_FEATURE(NETIF_F_LRO, set_feature_lro);
4079 err |= MLX5E_HANDLE_FEATURE(NETIF_F_GRO_HW, set_feature_hw_gro);
4080 err |= MLX5E_HANDLE_FEATURE(NETIF_F_HW_VLAN_CTAG_FILTER,
4081 set_feature_cvlan_filter);
4082 err |= MLX5E_HANDLE_FEATURE(NETIF_F_HW_TC, set_feature_hw_tc);
4083 err |= MLX5E_HANDLE_FEATURE(NETIF_F_RXALL, set_feature_rx_all);
4084 err |= MLX5E_HANDLE_FEATURE(NETIF_F_RXFCS, set_feature_rx_fcs);
4085 err |= MLX5E_HANDLE_FEATURE(NETIF_F_HW_VLAN_CTAG_RX, set_feature_rx_vlan);
4086 #ifdef CONFIG_MLX5_EN_ARFS
4087 err |= MLX5E_HANDLE_FEATURE(NETIF_F_NTUPLE, set_feature_arfs);
4089 err |= MLX5E_HANDLE_FEATURE(NETIF_F_HW_TLS_RX, mlx5e_ktls_set_feature_rx);
4092 netdev->features = oper_features;
4096 /* update XDP supported features */
4097 mlx5e_set_xdp_feature(netdev);
4102 static netdev_features_t mlx5e_fix_uplink_rep_features(struct net_device *netdev,
4103 netdev_features_t features)
4105 features &= ~NETIF_F_HW_TLS_RX;
4106 if (netdev->features & NETIF_F_HW_TLS_RX)
4107 netdev_warn(netdev, "Disabling hw_tls_rx, not supported in switchdev mode\n");
4109 features &= ~NETIF_F_HW_TLS_TX;
4110 if (netdev->features & NETIF_F_HW_TLS_TX)
4111 netdev_warn(netdev, "Disabling hw_tls_tx, not supported in switchdev mode\n");
4113 features &= ~NETIF_F_NTUPLE;
4114 if (netdev->features & NETIF_F_NTUPLE)
4115 netdev_warn(netdev, "Disabling ntuple, not supported in switchdev mode\n");
4117 features &= ~NETIF_F_GRO_HW;
4118 if (netdev->features & NETIF_F_GRO_HW)
4119 netdev_warn(netdev, "Disabling HW_GRO, not supported in switchdev mode\n");
4121 features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4122 if (netdev->features & NETIF_F_HW_VLAN_CTAG_FILTER)
4123 netdev_warn(netdev, "Disabling HW_VLAN CTAG FILTERING, not supported in switchdev mode\n");
4128 static netdev_features_t mlx5e_fix_features(struct net_device *netdev,
4129 netdev_features_t features)
4131 struct mlx5e_priv *priv = netdev_priv(netdev);
4132 struct mlx5e_vlan_table *vlan;
4133 struct mlx5e_params *params;
4135 if (!netif_device_present(netdev))
4138 vlan = mlx5e_fs_get_vlan(priv->fs);
4139 mutex_lock(&priv->state_lock);
4140 params = &priv->channels.params;
4142 !bitmap_empty(mlx5e_vlan_get_active_svlans(vlan), VLAN_N_VID)) {
4143 /* HW strips the outer C-tag header, this is a problem
4144 * for S-tag traffic.
4146 features &= ~NETIF_F_HW_VLAN_CTAG_RX;
4147 if (!params->vlan_strip_disable)
4148 netdev_warn(netdev, "Dropping C-tag vlan stripping offload due to S-tag vlan\n");
4151 if (!MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_STRIDING_RQ)) {
4152 if (features & NETIF_F_LRO) {
4153 netdev_warn(netdev, "Disabling LRO, not supported in legacy RQ\n");
4154 features &= ~NETIF_F_LRO;
4156 if (features & NETIF_F_GRO_HW) {
4157 netdev_warn(netdev, "Disabling HW-GRO, not supported in legacy RQ\n");
4158 features &= ~NETIF_F_GRO_HW;
4162 if (params->xdp_prog) {
4163 if (features & NETIF_F_LRO) {
4164 netdev_warn(netdev, "LRO is incompatible with XDP\n");
4165 features &= ~NETIF_F_LRO;
4167 if (features & NETIF_F_GRO_HW) {
4168 netdev_warn(netdev, "HW GRO is incompatible with XDP\n");
4169 features &= ~NETIF_F_GRO_HW;
4173 if (priv->xsk.refcnt) {
4174 if (features & NETIF_F_LRO) {
4175 netdev_warn(netdev, "LRO is incompatible with AF_XDP (%u XSKs are active)\n",
4177 features &= ~NETIF_F_LRO;
4179 if (features & NETIF_F_GRO_HW) {
4180 netdev_warn(netdev, "HW GRO is incompatible with AF_XDP (%u XSKs are active)\n",
4182 features &= ~NETIF_F_GRO_HW;
4186 if (MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS)) {
4187 features &= ~NETIF_F_RXHASH;
4188 if (netdev->features & NETIF_F_RXHASH)
4189 netdev_warn(netdev, "Disabling rxhash, not supported when CQE compress is active\n");
4191 if (features & NETIF_F_GRO_HW) {
4192 netdev_warn(netdev, "Disabling HW-GRO, not supported when CQE compress is active\n");
4193 features &= ~NETIF_F_GRO_HW;
4197 if (mlx5e_is_uplink_rep(priv)) {
4198 features = mlx5e_fix_uplink_rep_features(netdev, features);
4199 features |= NETIF_F_NETNS_LOCAL;
4201 features &= ~NETIF_F_NETNS_LOCAL;
4204 mutex_unlock(&priv->state_lock);
4209 static bool mlx5e_xsk_validate_mtu(struct net_device *netdev,
4210 struct mlx5e_channels *chs,
4211 struct mlx5e_params *new_params,
4212 struct mlx5_core_dev *mdev)
4216 for (ix = 0; ix < chs->params.num_channels; ix++) {
4217 struct xsk_buff_pool *xsk_pool =
4218 mlx5e_xsk_get_pool(&chs->params, chs->params.xsk, ix);
4219 struct mlx5e_xsk_param xsk;
4225 mlx5e_build_xsk_param(xsk_pool, &xsk);
4226 max_xdp_mtu = mlx5e_xdp_max_mtu(new_params, &xsk);
4228 /* Validate XSK params and XDP MTU in advance */
4229 if (!mlx5e_validate_xsk_param(new_params, &xsk, mdev) ||
4230 new_params->sw_mtu > max_xdp_mtu) {
4231 u32 hr = mlx5e_get_linear_rq_headroom(new_params, &xsk);
4232 int max_mtu_frame, max_mtu_page, max_mtu;
4234 /* Two criteria must be met:
4235 * 1. HW MTU + all headrooms <= XSK frame size.
4236 * 2. Size of SKBs allocated on XDP_PASS <= PAGE_SIZE.
4238 max_mtu_frame = MLX5E_HW2SW_MTU(new_params, xsk.chunk_size - hr);
4239 max_mtu_page = MLX5E_HW2SW_MTU(new_params, SKB_MAX_HEAD(0));
4240 max_mtu = min3(max_mtu_frame, max_mtu_page, max_xdp_mtu);
4242 netdev_err(netdev, "MTU %d is too big for an XSK running on channel %u or its redirection XDP program. Try MTU <= %d\n",
4243 new_params->sw_mtu, ix, max_mtu);
4251 static bool mlx5e_params_validate_xdp(struct net_device *netdev,
4252 struct mlx5_core_dev *mdev,
4253 struct mlx5e_params *params)
4257 /* No XSK params: AF_XDP can't be enabled yet at the point of setting
4260 is_linear = params->rq_wq_type == MLX5_WQ_TYPE_CYCLIC ?
4261 mlx5e_rx_is_linear_skb(mdev, params, NULL) :
4262 mlx5e_rx_mpwqe_is_linear_skb(mdev, params, NULL);
4265 if (!params->xdp_prog->aux->xdp_has_frags) {
4266 netdev_warn(netdev, "MTU(%d) > %d, too big for an XDP program not aware of multi buffer\n",
4268 mlx5e_xdp_max_mtu(params, NULL));
4271 if (params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ &&
4272 !mlx5e_verify_params_rx_mpwqe_strides(mdev, params, NULL)) {
4273 netdev_warn(netdev, "XDP is not allowed with striding RQ and MTU(%d) > %d\n",
4275 mlx5e_xdp_max_mtu(params, NULL));
4283 int mlx5e_change_mtu(struct net_device *netdev, int new_mtu,
4284 mlx5e_fp_preactivate preactivate)
4286 struct mlx5e_priv *priv = netdev_priv(netdev);
4287 struct mlx5e_params new_params;
4288 struct mlx5e_params *params;
4292 mutex_lock(&priv->state_lock);
4294 params = &priv->channels.params;
4296 new_params = *params;
4297 new_params.sw_mtu = new_mtu;
4298 err = mlx5e_validate_params(priv->mdev, &new_params);
4302 if (new_params.xdp_prog && !mlx5e_params_validate_xdp(netdev, priv->mdev,
4308 if (priv->xsk.refcnt &&
4309 !mlx5e_xsk_validate_mtu(netdev, &priv->channels,
4310 &new_params, priv->mdev)) {
4315 if (params->packet_merge.type == MLX5E_PACKET_MERGE_LRO)
4318 if (params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ &&
4319 params->packet_merge.type != MLX5E_PACKET_MERGE_SHAMPO) {
4320 bool is_linear_old = mlx5e_rx_mpwqe_is_linear_skb(priv->mdev, params, NULL);
4321 bool is_linear_new = mlx5e_rx_mpwqe_is_linear_skb(priv->mdev,
4323 u8 sz_old = mlx5e_mpwqe_get_log_rq_size(priv->mdev, params, NULL);
4324 u8 sz_new = mlx5e_mpwqe_get_log_rq_size(priv->mdev, &new_params, NULL);
4326 /* Always reset in linear mode - hw_mtu is used in data path.
4327 * Check that the mode was non-linear and didn't change.
4328 * If XSK is active, XSK RQs are linear.
4329 * Reset if the RQ size changed, even if it's non-linear.
4331 if (!is_linear_old && !is_linear_new && !priv->xsk.refcnt &&
4336 err = mlx5e_safe_switch_params(priv, &new_params, preactivate, NULL, reset);
4339 netdev->mtu = params->sw_mtu;
4340 mutex_unlock(&priv->state_lock);
4344 static int mlx5e_change_nic_mtu(struct net_device *netdev, int new_mtu)
4346 return mlx5e_change_mtu(netdev, new_mtu, mlx5e_set_dev_port_mtu_ctx);
4349 int mlx5e_ptp_rx_manage_fs_ctx(struct mlx5e_priv *priv, void *ctx)
4351 bool set = *(bool *)ctx;
4353 return mlx5e_ptp_rx_manage_fs(priv, set);
4356 static int mlx5e_hwstamp_config_no_ptp_rx(struct mlx5e_priv *priv, bool rx_filter)
4358 bool rx_cqe_compress_def = priv->channels.params.rx_cqe_compress_def;
4362 /* Reset CQE compression to Admin default */
4363 return mlx5e_modify_rx_cqe_compression_locked(priv, rx_cqe_compress_def, false);
4365 if (!MLX5E_GET_PFLAG(&priv->channels.params, MLX5E_PFLAG_RX_CQE_COMPRESS))
4368 /* Disable CQE compression */
4369 netdev_warn(priv->netdev, "Disabling RX cqe compression\n");
4370 err = mlx5e_modify_rx_cqe_compression_locked(priv, false, true);
4372 netdev_err(priv->netdev, "Failed disabling cqe compression err=%d\n", err);
4377 static int mlx5e_hwstamp_config_ptp_rx(struct mlx5e_priv *priv, bool ptp_rx)
4379 struct mlx5e_params new_params;
4381 if (ptp_rx == priv->channels.params.ptp_rx)
4384 new_params = priv->channels.params;
4385 new_params.ptp_rx = ptp_rx;
4386 return mlx5e_safe_switch_params(priv, &new_params, mlx5e_ptp_rx_manage_fs_ctx,
4387 &new_params.ptp_rx, true);
4390 int mlx5e_hwstamp_set(struct mlx5e_priv *priv, struct ifreq *ifr)
4392 struct hwtstamp_config config;
4393 bool rx_cqe_compress_def;
4397 if (!MLX5_CAP_GEN(priv->mdev, device_frequency_khz) ||
4398 (mlx5_clock_get_ptp_index(priv->mdev) == -1))
4401 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
4404 /* TX HW timestamp */
4405 switch (config.tx_type) {
4406 case HWTSTAMP_TX_OFF:
4407 case HWTSTAMP_TX_ON:
4413 mutex_lock(&priv->state_lock);
4414 rx_cqe_compress_def = priv->channels.params.rx_cqe_compress_def;
4416 /* RX HW timestamp */
4417 switch (config.rx_filter) {
4418 case HWTSTAMP_FILTER_NONE:
4421 case HWTSTAMP_FILTER_ALL:
4422 case HWTSTAMP_FILTER_SOME:
4423 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4424 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4425 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4426 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4427 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4428 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4429 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4430 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4431 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4432 case HWTSTAMP_FILTER_PTP_V2_EVENT:
4433 case HWTSTAMP_FILTER_PTP_V2_SYNC:
4434 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4435 case HWTSTAMP_FILTER_NTP_ALL:
4436 config.rx_filter = HWTSTAMP_FILTER_ALL;
4437 /* ptp_rx is set if both HW TS is set and CQE
4438 * compression is set
4440 ptp_rx = rx_cqe_compress_def;
4447 if (!mlx5e_profile_feature_cap(priv->profile, PTP_RX))
4448 err = mlx5e_hwstamp_config_no_ptp_rx(priv,
4449 config.rx_filter != HWTSTAMP_FILTER_NONE);
4451 err = mlx5e_hwstamp_config_ptp_rx(priv, ptp_rx);
4455 memcpy(&priv->tstamp, &config, sizeof(config));
4456 mutex_unlock(&priv->state_lock);
4458 /* might need to fix some features */
4459 netdev_update_features(priv->netdev);
4461 return copy_to_user(ifr->ifr_data, &config,
4462 sizeof(config)) ? -EFAULT : 0;
4464 mutex_unlock(&priv->state_lock);
4468 int mlx5e_hwstamp_get(struct mlx5e_priv *priv, struct ifreq *ifr)
4470 struct hwtstamp_config *cfg = &priv->tstamp;
4472 if (!MLX5_CAP_GEN(priv->mdev, device_frequency_khz))
4475 return copy_to_user(ifr->ifr_data, cfg, sizeof(*cfg)) ? -EFAULT : 0;
4478 static int mlx5e_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
4480 struct mlx5e_priv *priv = netdev_priv(dev);
4484 return mlx5e_hwstamp_set(priv, ifr);
4486 return mlx5e_hwstamp_get(priv, ifr);
4492 #ifdef CONFIG_MLX5_ESWITCH
4493 int mlx5e_set_vf_mac(struct net_device *dev, int vf, u8 *mac)
4495 struct mlx5e_priv *priv = netdev_priv(dev);
4496 struct mlx5_core_dev *mdev = priv->mdev;
4498 return mlx5_eswitch_set_vport_mac(mdev->priv.eswitch, vf + 1, mac);
4501 static int mlx5e_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
4504 struct mlx5e_priv *priv = netdev_priv(dev);
4505 struct mlx5_core_dev *mdev = priv->mdev;
4507 if (vlan_proto != htons(ETH_P_8021Q))
4508 return -EPROTONOSUPPORT;
4510 return mlx5_eswitch_set_vport_vlan(mdev->priv.eswitch, vf + 1,
4514 static int mlx5e_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
4516 struct mlx5e_priv *priv = netdev_priv(dev);
4517 struct mlx5_core_dev *mdev = priv->mdev;
4519 return mlx5_eswitch_set_vport_spoofchk(mdev->priv.eswitch, vf + 1, setting);
4522 static int mlx5e_set_vf_trust(struct net_device *dev, int vf, bool setting)
4524 struct mlx5e_priv *priv = netdev_priv(dev);
4525 struct mlx5_core_dev *mdev = priv->mdev;
4527 return mlx5_eswitch_set_vport_trust(mdev->priv.eswitch, vf + 1, setting);
4530 int mlx5e_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
4533 struct mlx5e_priv *priv = netdev_priv(dev);
4534 struct mlx5_core_dev *mdev = priv->mdev;
4536 return mlx5_eswitch_set_vport_rate(mdev->priv.eswitch, vf + 1,
4537 max_tx_rate, min_tx_rate);
4540 static int mlx5_vport_link2ifla(u8 esw_link)
4543 case MLX5_VPORT_ADMIN_STATE_DOWN:
4544 return IFLA_VF_LINK_STATE_DISABLE;
4545 case MLX5_VPORT_ADMIN_STATE_UP:
4546 return IFLA_VF_LINK_STATE_ENABLE;
4548 return IFLA_VF_LINK_STATE_AUTO;
4551 static int mlx5_ifla_link2vport(u8 ifla_link)
4553 switch (ifla_link) {
4554 case IFLA_VF_LINK_STATE_DISABLE:
4555 return MLX5_VPORT_ADMIN_STATE_DOWN;
4556 case IFLA_VF_LINK_STATE_ENABLE:
4557 return MLX5_VPORT_ADMIN_STATE_UP;
4559 return MLX5_VPORT_ADMIN_STATE_AUTO;
4562 static int mlx5e_set_vf_link_state(struct net_device *dev, int vf,
4565 struct mlx5e_priv *priv = netdev_priv(dev);
4566 struct mlx5_core_dev *mdev = priv->mdev;
4568 if (mlx5e_is_uplink_rep(priv))
4571 return mlx5_eswitch_set_vport_state(mdev->priv.eswitch, vf + 1,
4572 mlx5_ifla_link2vport(link_state));
4575 int mlx5e_get_vf_config(struct net_device *dev,
4576 int vf, struct ifla_vf_info *ivi)
4578 struct mlx5e_priv *priv = netdev_priv(dev);
4579 struct mlx5_core_dev *mdev = priv->mdev;
4582 if (!netif_device_present(dev))
4585 err = mlx5_eswitch_get_vport_config(mdev->priv.eswitch, vf + 1, ivi);
4588 ivi->linkstate = mlx5_vport_link2ifla(ivi->linkstate);
4592 int mlx5e_get_vf_stats(struct net_device *dev,
4593 int vf, struct ifla_vf_stats *vf_stats)
4595 struct mlx5e_priv *priv = netdev_priv(dev);
4596 struct mlx5_core_dev *mdev = priv->mdev;
4598 return mlx5_eswitch_get_vport_stats(mdev->priv.eswitch, vf + 1,
4603 mlx5e_has_offload_stats(const struct net_device *dev, int attr_id)
4605 struct mlx5e_priv *priv = netdev_priv(dev);
4607 if (!netif_device_present(dev))
4610 if (!mlx5e_is_uplink_rep(priv))
4613 return mlx5e_rep_has_offload_stats(dev, attr_id);
4617 mlx5e_get_offload_stats(int attr_id, const struct net_device *dev,
4620 struct mlx5e_priv *priv = netdev_priv(dev);
4622 if (!mlx5e_is_uplink_rep(priv))
4625 return mlx5e_rep_get_offload_stats(attr_id, dev, sp);
4629 static bool mlx5e_tunnel_proto_supported_tx(struct mlx5_core_dev *mdev, u8 proto_type)
4631 switch (proto_type) {
4633 return MLX5_CAP_ETH(mdev, tunnel_stateless_gre);
4636 return (MLX5_CAP_ETH(mdev, tunnel_stateless_ip_over_ip) ||
4637 MLX5_CAP_ETH(mdev, tunnel_stateless_ip_over_ip_tx));
4643 static bool mlx5e_gre_tunnel_inner_proto_offload_supported(struct mlx5_core_dev *mdev,
4644 struct sk_buff *skb)
4646 switch (skb->inner_protocol) {
4647 case htons(ETH_P_IP):
4648 case htons(ETH_P_IPV6):
4649 case htons(ETH_P_TEB):
4651 case htons(ETH_P_MPLS_UC):
4652 case htons(ETH_P_MPLS_MC):
4653 return MLX5_CAP_ETH(mdev, tunnel_stateless_mpls_over_gre);
4658 static netdev_features_t mlx5e_tunnel_features_check(struct mlx5e_priv *priv,
4659 struct sk_buff *skb,
4660 netdev_features_t features)
4662 unsigned int offset = 0;
4663 struct udphdr *udph;
4667 switch (vlan_get_protocol(skb)) {
4668 case htons(ETH_P_IP):
4669 proto = ip_hdr(skb)->protocol;
4671 case htons(ETH_P_IPV6):
4672 proto = ipv6_find_hdr(skb, &offset, -1, NULL, NULL);
4680 if (mlx5e_gre_tunnel_inner_proto_offload_supported(priv->mdev, skb))
4685 if (mlx5e_tunnel_proto_supported_tx(priv->mdev, IPPROTO_IPIP))
4689 udph = udp_hdr(skb);
4690 port = be16_to_cpu(udph->dest);
4692 /* Verify if UDP port is being offloaded by HW */
4693 if (mlx5_vxlan_lookup_port(priv->mdev->vxlan, port))
4696 #if IS_ENABLED(CONFIG_GENEVE)
4697 /* Support Geneve offload for default UDP port */
4698 if (port == GENEVE_UDP_PORT && mlx5_geneve_tx_allowed(priv->mdev))
4702 #ifdef CONFIG_MLX5_EN_IPSEC
4704 return mlx5e_ipsec_feature_check(skb, features);
4709 /* Disable CSUM and GSO if the udp dport is not offloaded by HW */
4710 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
4713 netdev_features_t mlx5e_features_check(struct sk_buff *skb,
4714 struct net_device *netdev,
4715 netdev_features_t features)
4717 struct mlx5e_priv *priv = netdev_priv(netdev);
4719 features = vlan_features_check(skb, features);
4720 features = vxlan_features_check(skb, features);
4722 /* Validate if the tunneled packet is being offloaded by HW */
4723 if (skb->encapsulation &&
4724 (features & NETIF_F_CSUM_MASK || features & NETIF_F_GSO_MASK))
4725 return mlx5e_tunnel_features_check(priv, skb, features);
4730 static void mlx5e_tx_timeout_work(struct work_struct *work)
4732 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
4734 struct net_device *netdev = priv->netdev;
4738 mutex_lock(&priv->state_lock);
4740 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
4743 for (i = 0; i < netdev->real_num_tx_queues; i++) {
4744 struct netdev_queue *dev_queue =
4745 netdev_get_tx_queue(netdev, i);
4746 struct mlx5e_txqsq *sq = priv->txq2sq[i];
4748 if (!netif_xmit_stopped(dev_queue))
4751 if (mlx5e_reporter_tx_timeout(sq))
4752 /* break if tried to reopened channels */
4757 mutex_unlock(&priv->state_lock);
4761 static void mlx5e_tx_timeout(struct net_device *dev, unsigned int txqueue)
4763 struct mlx5e_priv *priv = netdev_priv(dev);
4765 netdev_err(dev, "TX timeout detected\n");
4766 queue_work(priv->wq, &priv->tx_timeout_work);
4769 static int mlx5e_xdp_allowed(struct net_device *netdev, struct mlx5_core_dev *mdev,
4770 struct mlx5e_params *params)
4772 if (params->packet_merge.type != MLX5E_PACKET_MERGE_NONE) {
4773 netdev_warn(netdev, "can't set XDP while HW-GRO/LRO is on, disable them first\n");
4777 if (!mlx5e_params_validate_xdp(netdev, mdev, params))
4783 static void mlx5e_rq_replace_xdp_prog(struct mlx5e_rq *rq, struct bpf_prog *prog)
4785 struct bpf_prog *old_prog;
4787 old_prog = rcu_replace_pointer(rq->xdp_prog, prog,
4788 lockdep_is_held(&rq->priv->state_lock));
4790 bpf_prog_put(old_prog);
4793 static int mlx5e_xdp_set(struct net_device *netdev, struct bpf_prog *prog)
4795 struct mlx5e_priv *priv = netdev_priv(netdev);
4796 struct mlx5e_params new_params;
4797 struct bpf_prog *old_prog;
4802 mutex_lock(&priv->state_lock);
4804 new_params = priv->channels.params;
4805 new_params.xdp_prog = prog;
4808 err = mlx5e_xdp_allowed(netdev, priv->mdev, &new_params);
4813 /* no need for full reset when exchanging programs */
4814 reset = (!priv->channels.params.xdp_prog || !prog);
4816 old_prog = priv->channels.params.xdp_prog;
4818 err = mlx5e_safe_switch_params(priv, &new_params, NULL, NULL, reset);
4823 bpf_prog_put(old_prog);
4825 if (!test_bit(MLX5E_STATE_OPENED, &priv->state) || reset)
4828 /* exchanging programs w/o reset, we update ref counts on behalf
4829 * of the channels RQs here.
4831 bpf_prog_add(prog, priv->channels.num);
4832 for (i = 0; i < priv->channels.num; i++) {
4833 struct mlx5e_channel *c = priv->channels.c[i];
4835 mlx5e_rq_replace_xdp_prog(&c->rq, prog);
4836 if (test_bit(MLX5E_CHANNEL_STATE_XSK, c->state)) {
4838 mlx5e_rq_replace_xdp_prog(&c->xskrq, prog);
4843 mutex_unlock(&priv->state_lock);
4845 /* Need to fix some features. */
4847 netdev_update_features(netdev);
4852 static int mlx5e_xdp(struct net_device *dev, struct netdev_bpf *xdp)
4854 switch (xdp->command) {
4855 case XDP_SETUP_PROG:
4856 return mlx5e_xdp_set(dev, xdp->prog);
4857 case XDP_SETUP_XSK_POOL:
4858 return mlx5e_xsk_setup_pool(dev, xdp->xsk.pool,
4865 #ifdef CONFIG_MLX5_ESWITCH
4866 static int mlx5e_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4867 struct net_device *dev, u32 filter_mask,
4870 struct mlx5e_priv *priv = netdev_priv(dev);
4871 struct mlx5_core_dev *mdev = priv->mdev;
4875 err = mlx5_eswitch_get_vepa(mdev->priv.eswitch, &setting);
4878 mode = setting ? BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB;
4879 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4881 0, 0, nlflags, filter_mask, NULL);
4884 static int mlx5e_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4885 u16 flags, struct netlink_ext_ack *extack)
4887 struct mlx5e_priv *priv = netdev_priv(dev);
4888 struct mlx5_core_dev *mdev = priv->mdev;
4889 struct nlattr *attr, *br_spec;
4890 u16 mode = BRIDGE_MODE_UNDEF;
4894 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4898 nla_for_each_nested(attr, br_spec, rem) {
4899 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4902 mode = nla_get_u16(attr);
4903 if (mode > BRIDGE_MODE_VEPA)
4909 if (mode == BRIDGE_MODE_UNDEF)
4912 setting = (mode == BRIDGE_MODE_VEPA) ? 1 : 0;
4913 return mlx5_eswitch_set_vepa(mdev->priv.eswitch, setting);
4917 const struct net_device_ops mlx5e_netdev_ops = {
4918 .ndo_open = mlx5e_open,
4919 .ndo_stop = mlx5e_close,
4920 .ndo_start_xmit = mlx5e_xmit,
4921 .ndo_setup_tc = mlx5e_setup_tc,
4922 .ndo_select_queue = mlx5e_select_queue,
4923 .ndo_get_stats64 = mlx5e_get_stats,
4924 .ndo_set_rx_mode = mlx5e_set_rx_mode,
4925 .ndo_set_mac_address = mlx5e_set_mac,
4926 .ndo_vlan_rx_add_vid = mlx5e_vlan_rx_add_vid,
4927 .ndo_vlan_rx_kill_vid = mlx5e_vlan_rx_kill_vid,
4928 .ndo_set_features = mlx5e_set_features,
4929 .ndo_fix_features = mlx5e_fix_features,
4930 .ndo_change_mtu = mlx5e_change_nic_mtu,
4931 .ndo_eth_ioctl = mlx5e_ioctl,
4932 .ndo_set_tx_maxrate = mlx5e_set_tx_maxrate,
4933 .ndo_features_check = mlx5e_features_check,
4934 .ndo_tx_timeout = mlx5e_tx_timeout,
4935 .ndo_bpf = mlx5e_xdp,
4936 .ndo_xdp_xmit = mlx5e_xdp_xmit,
4937 .ndo_xsk_wakeup = mlx5e_xsk_wakeup,
4938 #ifdef CONFIG_MLX5_EN_ARFS
4939 .ndo_rx_flow_steer = mlx5e_rx_flow_steer,
4941 #ifdef CONFIG_MLX5_ESWITCH
4942 .ndo_bridge_setlink = mlx5e_bridge_setlink,
4943 .ndo_bridge_getlink = mlx5e_bridge_getlink,
4945 /* SRIOV E-Switch NDOs */
4946 .ndo_set_vf_mac = mlx5e_set_vf_mac,
4947 .ndo_set_vf_vlan = mlx5e_set_vf_vlan,
4948 .ndo_set_vf_spoofchk = mlx5e_set_vf_spoofchk,
4949 .ndo_set_vf_trust = mlx5e_set_vf_trust,
4950 .ndo_set_vf_rate = mlx5e_set_vf_rate,
4951 .ndo_get_vf_config = mlx5e_get_vf_config,
4952 .ndo_set_vf_link_state = mlx5e_set_vf_link_state,
4953 .ndo_get_vf_stats = mlx5e_get_vf_stats,
4954 .ndo_has_offload_stats = mlx5e_has_offload_stats,
4955 .ndo_get_offload_stats = mlx5e_get_offload_stats,
4959 static u32 mlx5e_choose_lro_timeout(struct mlx5_core_dev *mdev, u32 wanted_timeout)
4963 /* The supported periods are organized in ascending order */
4964 for (i = 0; i < MLX5E_LRO_TIMEOUT_ARR_SIZE - 1; i++)
4965 if (MLX5_CAP_ETH(mdev, lro_timer_supported_periods[i]) >= wanted_timeout)
4968 return MLX5_CAP_ETH(mdev, lro_timer_supported_periods[i]);
4971 void mlx5e_build_nic_params(struct mlx5e_priv *priv, struct mlx5e_xsk *xsk, u16 mtu)
4973 struct mlx5e_params *params = &priv->channels.params;
4974 struct mlx5_core_dev *mdev = priv->mdev;
4975 u8 rx_cq_period_mode;
4977 params->sw_mtu = mtu;
4978 params->hard_mtu = MLX5E_ETH_HARD_MTU;
4979 params->num_channels = min_t(unsigned int, MLX5E_MAX_NUM_CHANNELS / 2,
4981 mlx5e_params_mqprio_reset(params);
4984 params->log_sq_size = is_kdump_kernel() ?
4985 MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE :
4986 MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
4987 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_SKB_TX_MPWQE, mlx5e_tx_mpwqe_supported(mdev));
4990 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_XDP_TX_MPWQE, mlx5e_tx_mpwqe_supported(mdev));
4992 /* set CQE compression */
4993 params->rx_cqe_compress_def = false;
4994 if (MLX5_CAP_GEN(mdev, cqe_compression) &&
4995 MLX5_CAP_GEN(mdev, vport_group_manager))
4996 params->rx_cqe_compress_def = slow_pci_heuristic(mdev);
4998 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS, params->rx_cqe_compress_def);
4999 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_NO_CSUM_COMPLETE, false);
5002 mlx5e_build_rq_params(mdev, params);
5004 params->terminate_lkey_be = mlx5_core_get_terminate_scatter_list_mkey(mdev);
5006 params->packet_merge.timeout = mlx5e_choose_lro_timeout(mdev, MLX5E_DEFAULT_LRO_TIMEOUT);
5008 /* CQ moderation params */
5009 rx_cq_period_mode = MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
5010 MLX5_CQ_PERIOD_MODE_START_FROM_CQE :
5011 MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
5012 params->rx_dim_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
5013 params->tx_dim_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
5014 mlx5e_set_rx_cq_mode_params(params, rx_cq_period_mode);
5015 mlx5e_set_tx_cq_mode_params(params, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
5018 mlx5_query_min_inline(mdev, ¶ms->tx_min_inline_mode);
5023 /* Do not update netdev->features directly in here
5024 * on mlx5e_attach_netdev() we will call mlx5e_update_features()
5025 * To update netdev->features please modify mlx5e_fix_features()
5029 static void mlx5e_set_netdev_dev_addr(struct net_device *netdev)
5031 struct mlx5e_priv *priv = netdev_priv(netdev);
5034 mlx5_query_mac_address(priv->mdev, addr);
5035 if (is_zero_ether_addr(addr) &&
5036 !MLX5_CAP_GEN(priv->mdev, vport_group_manager)) {
5037 eth_hw_addr_random(netdev);
5038 mlx5_core_info(priv->mdev, "Assigned random MAC address %pM\n", netdev->dev_addr);
5042 eth_hw_addr_set(netdev, addr);
5045 static int mlx5e_vxlan_set_port(struct net_device *netdev, unsigned int table,
5046 unsigned int entry, struct udp_tunnel_info *ti)
5048 struct mlx5e_priv *priv = netdev_priv(netdev);
5050 return mlx5_vxlan_add_port(priv->mdev->vxlan, ntohs(ti->port));
5053 static int mlx5e_vxlan_unset_port(struct net_device *netdev, unsigned int table,
5054 unsigned int entry, struct udp_tunnel_info *ti)
5056 struct mlx5e_priv *priv = netdev_priv(netdev);
5058 return mlx5_vxlan_del_port(priv->mdev->vxlan, ntohs(ti->port));
5061 void mlx5e_vxlan_set_netdev_info(struct mlx5e_priv *priv)
5063 if (!mlx5_vxlan_allowed(priv->mdev->vxlan))
5066 priv->nic_info.set_port = mlx5e_vxlan_set_port;
5067 priv->nic_info.unset_port = mlx5e_vxlan_unset_port;
5068 priv->nic_info.flags = UDP_TUNNEL_NIC_INFO_MAY_SLEEP |
5069 UDP_TUNNEL_NIC_INFO_STATIC_IANA_VXLAN;
5070 priv->nic_info.tables[0].tunnel_types = UDP_TUNNEL_TYPE_VXLAN;
5071 /* Don't count the space hard-coded to the IANA port */
5072 priv->nic_info.tables[0].n_entries =
5073 mlx5_vxlan_max_udp_ports(priv->mdev) - 1;
5075 priv->netdev->udp_tunnel_nic_info = &priv->nic_info;
5078 static bool mlx5e_tunnel_any_tx_proto_supported(struct mlx5_core_dev *mdev)
5082 for (tt = 0; tt < MLX5_NUM_TUNNEL_TT; tt++) {
5083 if (mlx5e_tunnel_proto_supported_tx(mdev, mlx5_get_proto_by_tunnel_type(tt)))
5086 return (mlx5_vxlan_allowed(mdev->vxlan) || mlx5_geneve_tx_allowed(mdev));
5089 static void mlx5e_build_nic_netdev(struct net_device *netdev)
5091 struct mlx5e_priv *priv = netdev_priv(netdev);
5092 struct mlx5_core_dev *mdev = priv->mdev;
5096 SET_NETDEV_DEV(netdev, mdev->device);
5098 netdev->netdev_ops = &mlx5e_netdev_ops;
5099 netdev->xdp_metadata_ops = &mlx5e_xdp_metadata_ops;
5101 mlx5e_dcbnl_build_netdev(netdev);
5103 netdev->watchdog_timeo = 15 * HZ;
5105 netdev->ethtool_ops = &mlx5e_ethtool_ops;
5107 netdev->vlan_features |= NETIF_F_SG;
5108 netdev->vlan_features |= NETIF_F_HW_CSUM;
5109 netdev->vlan_features |= NETIF_F_HW_MACSEC;
5110 netdev->vlan_features |= NETIF_F_GRO;
5111 netdev->vlan_features |= NETIF_F_TSO;
5112 netdev->vlan_features |= NETIF_F_TSO6;
5113 netdev->vlan_features |= NETIF_F_RXCSUM;
5114 netdev->vlan_features |= NETIF_F_RXHASH;
5115 netdev->vlan_features |= NETIF_F_GSO_PARTIAL;
5117 netdev->mpls_features |= NETIF_F_SG;
5118 netdev->mpls_features |= NETIF_F_HW_CSUM;
5119 netdev->mpls_features |= NETIF_F_TSO;
5120 netdev->mpls_features |= NETIF_F_TSO6;
5122 netdev->hw_enc_features |= NETIF_F_HW_VLAN_CTAG_TX;
5123 netdev->hw_enc_features |= NETIF_F_HW_VLAN_CTAG_RX;
5125 /* Tunneled LRO is not supported in the driver, and the same RQs are
5126 * shared between inner and outer TIRs, so the driver can't disable LRO
5127 * for inner TIRs while having it enabled for outer TIRs. Due to this,
5128 * block LRO altogether if the firmware declares tunneled LRO support.
5130 if (!!MLX5_CAP_ETH(mdev, lro_cap) &&
5131 !MLX5_CAP_ETH(mdev, tunnel_lro_vxlan) &&
5132 !MLX5_CAP_ETH(mdev, tunnel_lro_gre) &&
5133 mlx5e_check_fragmented_striding_rq_cap(mdev, PAGE_SHIFT,
5134 MLX5E_MPWRQ_UMR_MODE_ALIGNED))
5135 netdev->vlan_features |= NETIF_F_LRO;
5137 netdev->hw_features = netdev->vlan_features;
5138 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
5139 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
5140 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
5141 netdev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
5143 if (mlx5e_tunnel_any_tx_proto_supported(mdev)) {
5144 netdev->hw_enc_features |= NETIF_F_HW_CSUM;
5145 netdev->hw_enc_features |= NETIF_F_TSO;
5146 netdev->hw_enc_features |= NETIF_F_TSO6;
5147 netdev->hw_enc_features |= NETIF_F_GSO_PARTIAL;
5150 if (mlx5_vxlan_allowed(mdev->vxlan) || mlx5_geneve_tx_allowed(mdev)) {
5151 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
5152 NETIF_F_GSO_UDP_TUNNEL_CSUM;
5153 netdev->hw_enc_features |= NETIF_F_GSO_UDP_TUNNEL |
5154 NETIF_F_GSO_UDP_TUNNEL_CSUM;
5155 netdev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
5156 netdev->vlan_features |= NETIF_F_GSO_UDP_TUNNEL |
5157 NETIF_F_GSO_UDP_TUNNEL_CSUM;
5160 if (mlx5e_tunnel_proto_supported_tx(mdev, IPPROTO_GRE)) {
5161 netdev->hw_features |= NETIF_F_GSO_GRE |
5162 NETIF_F_GSO_GRE_CSUM;
5163 netdev->hw_enc_features |= NETIF_F_GSO_GRE |
5164 NETIF_F_GSO_GRE_CSUM;
5165 netdev->gso_partial_features |= NETIF_F_GSO_GRE |
5166 NETIF_F_GSO_GRE_CSUM;
5169 if (mlx5e_tunnel_proto_supported_tx(mdev, IPPROTO_IPIP)) {
5170 netdev->hw_features |= NETIF_F_GSO_IPXIP4 |
5172 netdev->hw_enc_features |= NETIF_F_GSO_IPXIP4 |
5174 netdev->gso_partial_features |= NETIF_F_GSO_IPXIP4 |
5178 netdev->gso_partial_features |= NETIF_F_GSO_UDP_L4;
5179 netdev->hw_features |= NETIF_F_GSO_UDP_L4;
5180 netdev->features |= NETIF_F_GSO_UDP_L4;
5182 mlx5_query_port_fcs(mdev, &fcs_supported, &fcs_enabled);
5185 netdev->hw_features |= NETIF_F_RXALL;
5187 if (MLX5_CAP_ETH(mdev, scatter_fcs))
5188 netdev->hw_features |= NETIF_F_RXFCS;
5190 if (mlx5_qos_is_supported(mdev))
5191 netdev->hw_features |= NETIF_F_HW_TC;
5193 netdev->features = netdev->hw_features;
5197 netdev->features &= ~NETIF_F_RXALL;
5198 netdev->features &= ~NETIF_F_LRO;
5199 netdev->features &= ~NETIF_F_GRO_HW;
5200 netdev->features &= ~NETIF_F_RXFCS;
5202 #define FT_CAP(f) MLX5_CAP_FLOWTABLE(mdev, flow_table_properties_nic_receive.f)
5203 if (FT_CAP(flow_modify_en) &&
5204 FT_CAP(modify_root) &&
5205 FT_CAP(identified_miss_table_mode) &&
5206 FT_CAP(flow_table_modify)) {
5207 #if IS_ENABLED(CONFIG_MLX5_CLS_ACT)
5208 netdev->hw_features |= NETIF_F_HW_TC;
5210 #ifdef CONFIG_MLX5_EN_ARFS
5211 netdev->hw_features |= NETIF_F_NTUPLE;
5215 netdev->features |= NETIF_F_HIGHDMA;
5216 netdev->features |= NETIF_F_HW_VLAN_STAG_FILTER;
5218 netdev->priv_flags |= IFF_UNICAST_FLT;
5220 netif_set_tso_max_size(netdev, GSO_MAX_SIZE);
5221 mlx5e_set_xdp_feature(netdev);
5222 mlx5e_set_netdev_dev_addr(netdev);
5223 mlx5e_macsec_build_netdev(priv);
5224 mlx5e_ipsec_build_netdev(priv);
5225 mlx5e_ktls_build_netdev(priv);
5228 void mlx5e_create_q_counters(struct mlx5e_priv *priv)
5230 u32 out[MLX5_ST_SZ_DW(alloc_q_counter_out)] = {};
5231 u32 in[MLX5_ST_SZ_DW(alloc_q_counter_in)] = {};
5232 struct mlx5_core_dev *mdev = priv->mdev;
5235 MLX5_SET(alloc_q_counter_in, in, opcode, MLX5_CMD_OP_ALLOC_Q_COUNTER);
5236 err = mlx5_cmd_exec_inout(mdev, alloc_q_counter, in, out);
5239 MLX5_GET(alloc_q_counter_out, out, counter_set_id);
5241 err = mlx5_cmd_exec_inout(mdev, alloc_q_counter, in, out);
5243 priv->drop_rq_q_counter =
5244 MLX5_GET(alloc_q_counter_out, out, counter_set_id);
5247 void mlx5e_destroy_q_counters(struct mlx5e_priv *priv)
5249 u32 in[MLX5_ST_SZ_DW(dealloc_q_counter_in)] = {};
5251 MLX5_SET(dealloc_q_counter_in, in, opcode,
5252 MLX5_CMD_OP_DEALLOC_Q_COUNTER);
5253 if (priv->q_counter) {
5254 MLX5_SET(dealloc_q_counter_in, in, counter_set_id,
5256 mlx5_cmd_exec_in(priv->mdev, dealloc_q_counter, in);
5259 if (priv->drop_rq_q_counter) {
5260 MLX5_SET(dealloc_q_counter_in, in, counter_set_id,
5261 priv->drop_rq_q_counter);
5262 mlx5_cmd_exec_in(priv->mdev, dealloc_q_counter, in);
5266 static int mlx5e_nic_init(struct mlx5_core_dev *mdev,
5267 struct net_device *netdev)
5269 const bool take_rtnl = netdev->reg_state == NETREG_REGISTERED;
5270 struct mlx5e_priv *priv = netdev_priv(netdev);
5271 struct mlx5e_flow_steering *fs;
5274 mlx5e_build_nic_params(priv, &priv->xsk, netdev->mtu);
5275 mlx5e_vxlan_set_netdev_info(priv);
5277 mlx5e_timestamp_init(priv);
5279 priv->dfs_root = debugfs_create_dir("nic",
5280 mlx5_debugfs_get_dev_root(mdev));
5282 fs = mlx5e_fs_init(priv->profile, mdev,
5283 !test_bit(MLX5E_STATE_DESTROYING, &priv->state),
5287 mlx5_core_err(mdev, "FS initialization failed, %d\n", err);
5288 debugfs_remove_recursive(priv->dfs_root);
5293 err = mlx5e_ktls_init(priv);
5295 mlx5_core_err(mdev, "TLS initialization failed, %d\n", err);
5297 mlx5e_health_create_reporters(priv);
5299 /* If netdev is already registered (e.g. move from uplink to nic profile),
5300 * RTNL lock must be held before triggering netdev notifiers.
5305 /* update XDP supported features */
5306 mlx5e_set_xdp_feature(netdev);
5314 static void mlx5e_nic_cleanup(struct mlx5e_priv *priv)
5316 mlx5e_health_destroy_reporters(priv);
5317 mlx5e_ktls_cleanup(priv);
5318 mlx5e_fs_cleanup(priv->fs);
5319 debugfs_remove_recursive(priv->dfs_root);
5323 static int mlx5e_init_nic_rx(struct mlx5e_priv *priv)
5325 struct mlx5_core_dev *mdev = priv->mdev;
5326 enum mlx5e_rx_res_features features;
5329 priv->rx_res = mlx5e_rx_res_alloc();
5333 mlx5e_create_q_counters(priv);
5335 err = mlx5e_open_drop_rq(priv, &priv->drop_rq);
5337 mlx5_core_err(mdev, "open drop rq failed, %d\n", err);
5338 goto err_destroy_q_counters;
5341 features = MLX5E_RX_RES_FEATURE_PTP;
5342 if (mlx5_tunnel_inner_ft_supported(mdev))
5343 features |= MLX5E_RX_RES_FEATURE_INNER_FT;
5344 err = mlx5e_rx_res_init(priv->rx_res, priv->mdev, features,
5345 priv->max_nch, priv->drop_rq.rqn,
5346 &priv->channels.params.packet_merge,
5347 priv->channels.params.num_channels);
5349 goto err_close_drop_rq;
5351 err = mlx5e_create_flow_steering(priv->fs, priv->rx_res, priv->profile,
5354 mlx5_core_warn(mdev, "create flow steering failed, %d\n", err);
5355 goto err_destroy_rx_res;
5358 err = mlx5e_tc_nic_init(priv);
5360 goto err_destroy_flow_steering;
5362 err = mlx5e_accel_init_rx(priv);
5364 goto err_tc_nic_cleanup;
5366 #ifdef CONFIG_MLX5_EN_ARFS
5367 priv->netdev->rx_cpu_rmap = mlx5_eq_table_get_rmap(priv->mdev);
5373 mlx5e_tc_nic_cleanup(priv);
5374 err_destroy_flow_steering:
5375 mlx5e_destroy_flow_steering(priv->fs, !!(priv->netdev->hw_features & NETIF_F_NTUPLE),
5378 mlx5e_rx_res_destroy(priv->rx_res);
5380 mlx5e_close_drop_rq(&priv->drop_rq);
5381 err_destroy_q_counters:
5382 mlx5e_destroy_q_counters(priv);
5383 mlx5e_rx_res_free(priv->rx_res);
5384 priv->rx_res = NULL;
5388 static void mlx5e_cleanup_nic_rx(struct mlx5e_priv *priv)
5390 mlx5e_accel_cleanup_rx(priv);
5391 mlx5e_tc_nic_cleanup(priv);
5392 mlx5e_destroy_flow_steering(priv->fs, !!(priv->netdev->hw_features & NETIF_F_NTUPLE),
5394 mlx5e_rx_res_destroy(priv->rx_res);
5395 mlx5e_close_drop_rq(&priv->drop_rq);
5396 mlx5e_destroy_q_counters(priv);
5397 mlx5e_rx_res_free(priv->rx_res);
5398 priv->rx_res = NULL;
5401 static void mlx5e_set_mqprio_rl(struct mlx5e_priv *priv)
5403 struct mlx5e_params *params;
5404 struct mlx5e_mqprio_rl *rl;
5406 params = &priv->channels.params;
5407 if (params->mqprio.mode != TC_MQPRIO_MODE_CHANNEL)
5410 rl = mlx5e_mqprio_rl_create(priv->mdev, params->mqprio.num_tc,
5411 params->mqprio.channel.max_rate);
5414 priv->mqprio_rl = rl;
5415 mlx5e_mqprio_rl_update_params(params, rl);
5418 static int mlx5e_init_nic_tx(struct mlx5e_priv *priv)
5422 err = mlx5e_create_tises(priv);
5424 mlx5_core_warn(priv->mdev, "create tises failed, %d\n", err);
5428 err = mlx5e_accel_init_tx(priv);
5430 goto err_destroy_tises;
5432 mlx5e_set_mqprio_rl(priv);
5433 mlx5e_dcbnl_initialize(priv);
5437 mlx5e_destroy_tises(priv);
5441 static void mlx5e_nic_enable(struct mlx5e_priv *priv)
5443 struct net_device *netdev = priv->netdev;
5444 struct mlx5_core_dev *mdev = priv->mdev;
5447 mlx5e_fs_init_l2_addr(priv->fs, netdev);
5448 mlx5e_ipsec_init(priv);
5450 err = mlx5e_macsec_init(priv);
5452 mlx5_core_err(mdev, "MACsec initialization failed, %d\n", err);
5454 /* Marking the link as currently not needed by the Driver */
5455 if (!netif_running(netdev))
5456 mlx5e_modify_admin_state(mdev, MLX5_PORT_DOWN);
5458 mlx5e_set_netdev_mtu_boundaries(priv);
5459 mlx5e_set_dev_port_mtu(priv);
5461 mlx5_lag_add_netdev(mdev, netdev);
5463 mlx5e_enable_async_events(priv);
5464 mlx5e_enable_blocking_events(priv);
5465 if (mlx5e_monitor_counter_supported(priv))
5466 mlx5e_monitor_counter_init(priv);
5468 mlx5e_hv_vhca_stats_create(priv);
5469 if (netdev->reg_state != NETREG_REGISTERED)
5471 mlx5e_dcbnl_init_app(priv);
5473 mlx5e_nic_set_rx_mode(priv);
5476 if (netif_running(netdev))
5478 udp_tunnel_nic_reset_ntf(priv->netdev);
5479 netif_device_attach(netdev);
5483 static void mlx5e_nic_disable(struct mlx5e_priv *priv)
5485 struct mlx5_core_dev *mdev = priv->mdev;
5487 if (priv->netdev->reg_state == NETREG_REGISTERED)
5488 mlx5e_dcbnl_delete_app(priv);
5491 if (netif_running(priv->netdev))
5492 mlx5e_close(priv->netdev);
5493 netif_device_detach(priv->netdev);
5496 mlx5e_nic_set_rx_mode(priv);
5498 mlx5e_hv_vhca_stats_destroy(priv);
5499 if (mlx5e_monitor_counter_supported(priv))
5500 mlx5e_monitor_counter_cleanup(priv);
5502 mlx5e_disable_blocking_events(priv);
5503 if (priv->en_trap) {
5504 mlx5e_deactivate_trap(priv);
5505 mlx5e_close_trap(priv->en_trap);
5506 priv->en_trap = NULL;
5508 mlx5e_disable_async_events(priv);
5509 mlx5_lag_remove_netdev(mdev, priv->netdev);
5510 mlx5_vxlan_reset_to_default(mdev->vxlan);
5511 mlx5e_macsec_cleanup(priv);
5512 mlx5e_ipsec_cleanup(priv);
5515 int mlx5e_update_nic_rx(struct mlx5e_priv *priv)
5517 return mlx5e_refresh_tirs(priv, false, false);
5520 static const struct mlx5e_profile mlx5e_nic_profile = {
5521 .init = mlx5e_nic_init,
5522 .cleanup = mlx5e_nic_cleanup,
5523 .init_rx = mlx5e_init_nic_rx,
5524 .cleanup_rx = mlx5e_cleanup_nic_rx,
5525 .init_tx = mlx5e_init_nic_tx,
5526 .cleanup_tx = mlx5e_cleanup_nic_tx,
5527 .enable = mlx5e_nic_enable,
5528 .disable = mlx5e_nic_disable,
5529 .update_rx = mlx5e_update_nic_rx,
5530 .update_stats = mlx5e_stats_update_ndo_stats,
5531 .update_carrier = mlx5e_update_carrier,
5532 .rx_handlers = &mlx5e_rx_handlers_nic,
5533 .max_tc = MLX5E_MAX_NUM_TC,
5534 .stats_grps = mlx5e_nic_stats_grps,
5535 .stats_grps_num = mlx5e_nic_stats_grps_num,
5536 .features = BIT(MLX5E_PROFILE_FEATURE_PTP_RX) |
5537 BIT(MLX5E_PROFILE_FEATURE_PTP_TX) |
5538 BIT(MLX5E_PROFILE_FEATURE_QOS_HTB) |
5539 BIT(MLX5E_PROFILE_FEATURE_FS_VLAN) |
5540 BIT(MLX5E_PROFILE_FEATURE_FS_TC),
5543 static int mlx5e_profile_max_num_channels(struct mlx5_core_dev *mdev,
5544 const struct mlx5e_profile *profile)
5548 nch = mlx5e_get_max_num_channels(mdev);
5550 if (profile->max_nch_limit)
5551 nch = min_t(int, nch, profile->max_nch_limit(mdev));
5556 mlx5e_calc_max_nch(struct mlx5_core_dev *mdev, struct net_device *netdev,
5557 const struct mlx5e_profile *profile)
5560 unsigned int max_nch, tmp;
5562 /* core resources */
5563 max_nch = mlx5e_profile_max_num_channels(mdev, profile);
5565 /* netdev rx queues */
5566 max_nch = min_t(unsigned int, max_nch, netdev->num_rx_queues);
5568 /* netdev tx queues */
5569 tmp = netdev->num_tx_queues;
5570 if (mlx5_qos_is_supported(mdev))
5571 tmp -= mlx5e_qos_max_leaf_nodes(mdev);
5572 if (MLX5_CAP_GEN(mdev, ts_cqe_to_dest_cqn))
5573 tmp -= profile->max_tc;
5574 tmp = tmp / profile->max_tc;
5575 max_nch = min_t(unsigned int, max_nch, tmp);
5580 int mlx5e_get_pf_num_tirs(struct mlx5_core_dev *mdev)
5582 /* Indirect TIRS: 2 sets of TTCs (inner + outer steering)
5583 * and 1 set of direct TIRS
5585 return 2 * MLX5E_NUM_INDIR_TIRS
5586 + mlx5e_profile_max_num_channels(mdev, &mlx5e_nic_profile);
5589 void mlx5e_set_rx_mode_work(struct work_struct *work)
5591 struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
5594 return mlx5e_fs_set_rx_mode_work(priv->fs, priv->netdev);
5597 /* mlx5e generic netdev management API (move to en_common.c) */
5598 int mlx5e_priv_init(struct mlx5e_priv *priv,
5599 const struct mlx5e_profile *profile,
5600 struct net_device *netdev,
5601 struct mlx5_core_dev *mdev)
5603 int nch, num_txqs, node;
5606 num_txqs = netdev->num_tx_queues;
5607 nch = mlx5e_calc_max_nch(mdev, netdev, profile);
5608 node = dev_to_node(mlx5_core_dma_dev(mdev));
5612 priv->netdev = netdev;
5613 priv->max_nch = nch;
5614 priv->max_opened_tc = 1;
5616 if (!alloc_cpumask_var(&priv->scratchpad.cpumask, GFP_KERNEL))
5619 mutex_init(&priv->state_lock);
5621 err = mlx5e_selq_init(&priv->selq, &priv->state_lock);
5623 goto err_free_cpumask;
5625 INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
5626 INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
5627 INIT_WORK(&priv->tx_timeout_work, mlx5e_tx_timeout_work);
5628 INIT_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
5630 priv->wq = create_singlethread_workqueue("mlx5e");
5634 priv->txq2sq = kcalloc_node(num_txqs, sizeof(*priv->txq2sq), GFP_KERNEL, node);
5636 goto err_destroy_workqueue;
5638 priv->tx_rates = kcalloc_node(num_txqs, sizeof(*priv->tx_rates), GFP_KERNEL, node);
5639 if (!priv->tx_rates)
5640 goto err_free_txq2sq;
5642 priv->channel_stats =
5643 kcalloc_node(nch, sizeof(*priv->channel_stats), GFP_KERNEL, node);
5644 if (!priv->channel_stats)
5645 goto err_free_tx_rates;
5650 kfree(priv->tx_rates);
5652 kfree(priv->txq2sq);
5653 err_destroy_workqueue:
5654 destroy_workqueue(priv->wq);
5656 mlx5e_selq_cleanup(&priv->selq);
5658 free_cpumask_var(priv->scratchpad.cpumask);
5662 void mlx5e_priv_cleanup(struct mlx5e_priv *priv)
5666 /* bail if change profile failed and also rollback failed */
5670 for (i = 0; i < priv->stats_nch; i++)
5671 kvfree(priv->channel_stats[i]);
5672 kfree(priv->channel_stats);
5673 kfree(priv->tx_rates);
5674 kfree(priv->txq2sq);
5675 destroy_workqueue(priv->wq);
5676 mutex_lock(&priv->state_lock);
5677 mlx5e_selq_cleanup(&priv->selq);
5678 mutex_unlock(&priv->state_lock);
5679 free_cpumask_var(priv->scratchpad.cpumask);
5681 for (i = 0; i < priv->htb_max_qos_sqs; i++)
5682 kfree(priv->htb_qos_sq_stats[i]);
5683 kvfree(priv->htb_qos_sq_stats);
5685 memset(priv, 0, sizeof(*priv));
5688 static unsigned int mlx5e_get_max_num_txqs(struct mlx5_core_dev *mdev,
5689 const struct mlx5e_profile *profile)
5691 unsigned int nch, ptp_txqs, qos_txqs;
5693 nch = mlx5e_profile_max_num_channels(mdev, profile);
5695 ptp_txqs = MLX5_CAP_GEN(mdev, ts_cqe_to_dest_cqn) &&
5696 mlx5e_profile_feature_cap(profile, PTP_TX) ?
5697 profile->max_tc : 0;
5699 qos_txqs = mlx5_qos_is_supported(mdev) &&
5700 mlx5e_profile_feature_cap(profile, QOS_HTB) ?
5701 mlx5e_qos_max_leaf_nodes(mdev) : 0;
5703 return nch * profile->max_tc + ptp_txqs + qos_txqs;
5706 static unsigned int mlx5e_get_max_num_rxqs(struct mlx5_core_dev *mdev,
5707 const struct mlx5e_profile *profile)
5709 return mlx5e_profile_max_num_channels(mdev, profile);
5713 mlx5e_create_netdev(struct mlx5_core_dev *mdev, const struct mlx5e_profile *profile)
5715 struct net_device *netdev;
5716 unsigned int txqs, rxqs;
5719 txqs = mlx5e_get_max_num_txqs(mdev, profile);
5720 rxqs = mlx5e_get_max_num_rxqs(mdev, profile);
5722 netdev = alloc_etherdev_mqs(sizeof(struct mlx5e_priv), txqs, rxqs);
5724 mlx5_core_err(mdev, "alloc_etherdev_mqs() failed\n");
5728 err = mlx5e_priv_init(netdev_priv(netdev), profile, netdev, mdev);
5730 mlx5_core_err(mdev, "mlx5e_priv_init failed, err=%d\n", err);
5731 goto err_free_netdev;
5734 netif_carrier_off(netdev);
5735 netif_tx_disable(netdev);
5736 dev_net_set(netdev, mlx5_core_net(mdev));
5741 free_netdev(netdev);
5746 static void mlx5e_update_features(struct net_device *netdev)
5748 if (netdev->reg_state != NETREG_REGISTERED)
5749 return; /* features will be updated on netdev registration */
5752 netdev_update_features(netdev);
5756 static void mlx5e_reset_channels(struct net_device *netdev)
5758 netdev_reset_tc(netdev);
5761 int mlx5e_attach_netdev(struct mlx5e_priv *priv)
5763 const bool take_rtnl = priv->netdev->reg_state == NETREG_REGISTERED;
5764 const struct mlx5e_profile *profile = priv->profile;
5768 clear_bit(MLX5E_STATE_DESTROYING, &priv->state);
5770 mlx5e_fs_set_state_destroy(priv->fs,
5771 !test_bit(MLX5E_STATE_DESTROYING, &priv->state));
5773 /* Validate the max_wqe_size_sq capability. */
5774 if (WARN_ON_ONCE(mlx5e_get_max_sq_wqebbs(priv->mdev) < MLX5E_MAX_TX_WQEBBS)) {
5775 mlx5_core_warn(priv->mdev, "MLX5E: Max SQ WQEBBs firmware capability: %u, needed %u\n",
5776 mlx5e_get_max_sq_wqebbs(priv->mdev), (unsigned int)MLX5E_MAX_TX_WQEBBS);
5780 /* max number of channels may have changed */
5781 max_nch = mlx5e_calc_max_nch(priv->mdev, priv->netdev, profile);
5782 if (priv->channels.params.num_channels > max_nch) {
5783 mlx5_core_warn(priv->mdev, "MLX5E: Reducing number of channels to %d\n", max_nch);
5784 /* Reducing the number of channels - RXFH has to be reset, and
5785 * mlx5e_num_channels_changed below will build the RQT.
5787 priv->netdev->priv_flags &= ~IFF_RXFH_CONFIGURED;
5788 priv->channels.params.num_channels = max_nch;
5789 if (priv->channels.params.mqprio.mode == TC_MQPRIO_MODE_CHANNEL) {
5790 mlx5_core_warn(priv->mdev, "MLX5E: Disabling MQPRIO channel mode\n");
5791 mlx5e_params_mqprio_reset(&priv->channels.params);
5794 if (max_nch != priv->max_nch) {
5795 mlx5_core_warn(priv->mdev,
5796 "MLX5E: Updating max number of channels from %u to %u\n",
5797 priv->max_nch, max_nch);
5798 priv->max_nch = max_nch;
5801 /* 1. Set the real number of queues in the kernel the first time.
5802 * 2. Set our default XPS cpumask.
5805 * rtnl_lock is required by netif_set_real_num_*_queues in case the
5806 * netdev has been registered by this point (if this function was called
5807 * in the reload or resume flow).
5811 err = mlx5e_num_channels_changed(priv);
5817 err = profile->init_tx(priv);
5821 err = profile->init_rx(priv);
5823 goto err_cleanup_tx;
5825 if (profile->enable)
5826 profile->enable(priv);
5828 mlx5e_update_features(priv->netdev);
5833 profile->cleanup_tx(priv);
5836 mlx5e_reset_channels(priv->netdev);
5837 set_bit(MLX5E_STATE_DESTROYING, &priv->state);
5839 mlx5e_fs_set_state_destroy(priv->fs,
5840 !test_bit(MLX5E_STATE_DESTROYING, &priv->state));
5841 cancel_work_sync(&priv->update_stats_work);
5845 void mlx5e_detach_netdev(struct mlx5e_priv *priv)
5847 const struct mlx5e_profile *profile = priv->profile;
5849 set_bit(MLX5E_STATE_DESTROYING, &priv->state);
5851 mlx5e_fs_set_state_destroy(priv->fs,
5852 !test_bit(MLX5E_STATE_DESTROYING, &priv->state));
5854 if (profile->disable)
5855 profile->disable(priv);
5856 flush_workqueue(priv->wq);
5858 profile->cleanup_rx(priv);
5859 profile->cleanup_tx(priv);
5860 mlx5e_reset_channels(priv->netdev);
5861 cancel_work_sync(&priv->update_stats_work);
5865 mlx5e_netdev_init_profile(struct net_device *netdev, struct mlx5_core_dev *mdev,
5866 const struct mlx5e_profile *new_profile, void *new_ppriv)
5868 struct mlx5e_priv *priv = netdev_priv(netdev);
5871 err = mlx5e_priv_init(priv, new_profile, netdev, mdev);
5873 mlx5_core_err(mdev, "mlx5e_priv_init failed, err=%d\n", err);
5876 netif_carrier_off(netdev);
5877 priv->profile = new_profile;
5878 priv->ppriv = new_ppriv;
5879 err = new_profile->init(priv->mdev, priv->netdev);
5886 mlx5e_priv_cleanup(priv);
5891 mlx5e_netdev_attach_profile(struct net_device *netdev, struct mlx5_core_dev *mdev,
5892 const struct mlx5e_profile *new_profile, void *new_ppriv)
5894 struct mlx5e_priv *priv = netdev_priv(netdev);
5897 err = mlx5e_netdev_init_profile(netdev, mdev, new_profile, new_ppriv);
5901 err = mlx5e_attach_netdev(priv);
5903 goto profile_cleanup;
5907 new_profile->cleanup(priv);
5908 mlx5e_priv_cleanup(priv);
5912 int mlx5e_netdev_change_profile(struct mlx5e_priv *priv,
5913 const struct mlx5e_profile *new_profile, void *new_ppriv)
5915 const struct mlx5e_profile *orig_profile = priv->profile;
5916 struct net_device *netdev = priv->netdev;
5917 struct mlx5_core_dev *mdev = priv->mdev;
5918 void *orig_ppriv = priv->ppriv;
5919 int err, rollback_err;
5921 /* cleanup old profile */
5922 mlx5e_detach_netdev(priv);
5923 priv->profile->cleanup(priv);
5924 mlx5e_priv_cleanup(priv);
5926 if (mdev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
5927 mlx5e_netdev_init_profile(netdev, mdev, new_profile, new_ppriv);
5928 set_bit(MLX5E_STATE_DESTROYING, &priv->state);
5932 err = mlx5e_netdev_attach_profile(netdev, mdev, new_profile, new_ppriv);
5933 if (err) { /* roll back to original profile */
5934 netdev_warn(netdev, "%s: new profile init failed, %d\n", __func__, err);
5941 rollback_err = mlx5e_netdev_attach_profile(netdev, mdev, orig_profile, orig_ppriv);
5943 netdev_err(netdev, "%s: failed to rollback to orig profile, %d\n",
5944 __func__, rollback_err);
5948 void mlx5e_netdev_attach_nic_profile(struct mlx5e_priv *priv)
5950 mlx5e_netdev_change_profile(priv, &mlx5e_nic_profile, NULL);
5953 void mlx5e_destroy_netdev(struct mlx5e_priv *priv)
5955 struct net_device *netdev = priv->netdev;
5957 mlx5e_priv_cleanup(priv);
5958 free_netdev(netdev);
5961 static int mlx5e_resume(struct auxiliary_device *adev)
5963 struct mlx5_adev *edev = container_of(adev, struct mlx5_adev, adev);
5964 struct mlx5e_dev *mlx5e_dev = auxiliary_get_drvdata(adev);
5965 struct mlx5e_priv *priv = mlx5e_dev->priv;
5966 struct net_device *netdev = priv->netdev;
5967 struct mlx5_core_dev *mdev = edev->mdev;
5970 if (netif_device_present(netdev))
5973 err = mlx5e_create_mdev_resources(mdev);
5977 err = mlx5e_attach_netdev(priv);
5979 mlx5e_destroy_mdev_resources(mdev);
5986 static int mlx5e_suspend(struct auxiliary_device *adev, pm_message_t state)
5988 struct mlx5e_dev *mlx5e_dev = auxiliary_get_drvdata(adev);
5989 struct mlx5e_priv *priv = mlx5e_dev->priv;
5990 struct net_device *netdev = priv->netdev;
5991 struct mlx5_core_dev *mdev = priv->mdev;
5993 if (!netif_device_present(netdev)) {
5994 if (test_bit(MLX5E_STATE_DESTROYING, &priv->state))
5995 mlx5e_destroy_mdev_resources(mdev);
5999 mlx5e_detach_netdev(priv);
6000 mlx5e_destroy_mdev_resources(mdev);
6004 static int mlx5e_probe(struct auxiliary_device *adev,
6005 const struct auxiliary_device_id *id)
6007 struct mlx5_adev *edev = container_of(adev, struct mlx5_adev, adev);
6008 const struct mlx5e_profile *profile = &mlx5e_nic_profile;
6009 struct mlx5_core_dev *mdev = edev->mdev;
6010 struct mlx5e_dev *mlx5e_dev;
6011 struct net_device *netdev;
6012 pm_message_t state = {};
6013 struct mlx5e_priv *priv;
6016 mlx5e_dev = mlx5e_create_devlink(&adev->dev, mdev);
6017 if (IS_ERR(mlx5e_dev))
6018 return PTR_ERR(mlx5e_dev);
6019 auxiliary_set_drvdata(adev, mlx5e_dev);
6021 err = mlx5e_devlink_port_register(mlx5e_dev, mdev);
6023 mlx5_core_err(mdev, "mlx5e_devlink_port_register failed, %d\n", err);
6024 goto err_devlink_unregister;
6027 netdev = mlx5e_create_netdev(mdev, profile);
6029 mlx5_core_err(mdev, "mlx5e_create_netdev failed\n");
6031 goto err_devlink_port_unregister;
6033 SET_NETDEV_DEVLINK_PORT(netdev, &mlx5e_dev->dl_port);
6035 mlx5e_build_nic_netdev(netdev);
6037 priv = netdev_priv(netdev);
6038 mlx5e_dev->priv = priv;
6040 priv->profile = profile;
6043 err = profile->init(mdev, netdev);
6045 mlx5_core_err(mdev, "mlx5e_nic_profile init failed, %d\n", err);
6046 goto err_destroy_netdev;
6049 err = mlx5e_resume(adev);
6051 mlx5_core_err(mdev, "mlx5e_resume failed, %d\n", err);
6052 goto err_profile_cleanup;
6055 err = register_netdev(netdev);
6057 mlx5_core_err(mdev, "register_netdev failed, %d\n", err);
6061 mlx5e_dcbnl_init_app(priv);
6062 mlx5_core_uplink_netdev_set(mdev, netdev);
6063 mlx5e_params_print_info(mdev, &priv->channels.params);
6067 mlx5e_suspend(adev, state);
6068 err_profile_cleanup:
6069 profile->cleanup(priv);
6071 mlx5e_destroy_netdev(priv);
6072 err_devlink_port_unregister:
6073 mlx5e_devlink_port_unregister(mlx5e_dev);
6074 err_devlink_unregister:
6075 mlx5e_destroy_devlink(mlx5e_dev);
6079 static void mlx5e_remove(struct auxiliary_device *adev)
6081 struct mlx5e_dev *mlx5e_dev = auxiliary_get_drvdata(adev);
6082 struct mlx5e_priv *priv = mlx5e_dev->priv;
6083 pm_message_t state = {};
6085 mlx5_core_uplink_netdev_set(priv->mdev, NULL);
6086 mlx5e_dcbnl_delete_app(priv);
6087 unregister_netdev(priv->netdev);
6088 mlx5e_suspend(adev, state);
6089 priv->profile->cleanup(priv);
6090 mlx5e_destroy_netdev(priv);
6091 mlx5e_devlink_port_unregister(mlx5e_dev);
6092 mlx5e_destroy_devlink(mlx5e_dev);
6095 static const struct auxiliary_device_id mlx5e_id_table[] = {
6096 { .name = MLX5_ADEV_NAME ".eth", },
6100 MODULE_DEVICE_TABLE(auxiliary, mlx5e_id_table);
6102 static struct auxiliary_driver mlx5e_driver = {
6104 .probe = mlx5e_probe,
6105 .remove = mlx5e_remove,
6106 .suspend = mlx5e_suspend,
6107 .resume = mlx5e_resume,
6108 .id_table = mlx5e_id_table,
6111 int mlx5e_init(void)
6115 mlx5e_build_ptys2ethtool_map();
6116 ret = auxiliary_driver_register(&mlx5e_driver);
6120 ret = mlx5e_rep_init();
6122 auxiliary_driver_unregister(&mlx5e_driver);
6126 void mlx5e_cleanup(void)
6128 mlx5e_rep_cleanup();
6129 auxiliary_driver_unregister(&mlx5e_driver);