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/crash_dump.h>
35 #include <net/pkt_cls.h>
36 #include <linux/mlx5/fs.h>
37 #include <net/vxlan.h>
38 #include <linux/bpf.h>
44 struct mlx5e_rq_param
{
45 u32 rqc
[MLX5_ST_SZ_DW(rqc
)];
46 struct mlx5_wq_param wq
;
49 struct mlx5e_sq_param
{
50 u32 sqc
[MLX5_ST_SZ_DW(sqc
)];
51 struct mlx5_wq_param wq
;
54 struct mlx5e_cq_param
{
55 u32 cqc
[MLX5_ST_SZ_DW(cqc
)];
56 struct mlx5_wq_param wq
;
61 struct mlx5e_channel_param
{
62 struct mlx5e_rq_param rq
;
63 struct mlx5e_sq_param sq
;
64 struct mlx5e_sq_param xdp_sq
;
65 struct mlx5e_sq_param icosq
;
66 struct mlx5e_cq_param rx_cq
;
67 struct mlx5e_cq_param tx_cq
;
68 struct mlx5e_cq_param icosq_cq
;
71 static bool mlx5e_check_fragmented_striding_rq_cap(struct mlx5_core_dev
*mdev
)
73 return MLX5_CAP_GEN(mdev
, striding_rq
) &&
74 MLX5_CAP_GEN(mdev
, umr_ptr_rlky
) &&
75 MLX5_CAP_ETH(mdev
, reg_umr_sq
);
78 void mlx5e_set_rq_type_params(struct mlx5_core_dev
*mdev
,
79 struct mlx5e_params
*params
, u8 rq_type
)
81 params
->rq_wq_type
= rq_type
;
82 params
->lro_wqe_sz
= MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ
;
83 switch (params
->rq_wq_type
) {
84 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ
:
85 params
->log_rq_size
= is_kdump_kernel() ?
86 MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE_MPW
:
87 MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE_MPW
;
88 params
->mpwqe_log_stride_sz
=
89 MLX5E_GET_PFLAG(params
, MLX5E_PFLAG_RX_CQE_COMPRESS
) ?
90 MLX5_MPWRQ_CQE_CMPRS_LOG_STRIDE_SZ(mdev
) :
91 MLX5_MPWRQ_DEF_LOG_STRIDE_SZ(mdev
);
92 params
->mpwqe_log_num_strides
= MLX5_MPWRQ_LOG_WQE_SZ
-
93 params
->mpwqe_log_stride_sz
;
95 default: /* MLX5_WQ_TYPE_LINKED_LIST */
96 params
->log_rq_size
= is_kdump_kernel() ?
97 MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE
:
98 MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE
;
100 /* Extra room needed for build_skb */
101 params
->lro_wqe_sz
-= MLX5_RX_HEADROOM
+
102 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
105 mlx5_core_info(mdev
, "MLX5E: StrdRq(%d) RqSz(%ld) StrdSz(%ld) RxCqeCmprss(%d)\n",
106 params
->rq_wq_type
== MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ
,
107 BIT(params
->log_rq_size
),
108 BIT(params
->mpwqe_log_stride_sz
),
109 MLX5E_GET_PFLAG(params
, MLX5E_PFLAG_RX_CQE_COMPRESS
));
112 static void mlx5e_set_rq_params(struct mlx5_core_dev
*mdev
, struct mlx5e_params
*params
)
114 u8 rq_type
= mlx5e_check_fragmented_striding_rq_cap(mdev
) &&
116 MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ
:
117 MLX5_WQ_TYPE_LINKED_LIST
;
118 mlx5e_set_rq_type_params(mdev
, params
, rq_type
);
121 static void mlx5e_update_carrier(struct mlx5e_priv
*priv
)
123 struct mlx5_core_dev
*mdev
= priv
->mdev
;
126 port_state
= mlx5_query_vport_state(mdev
,
127 MLX5_QUERY_VPORT_STATE_IN_OP_MOD_VNIC_VPORT
, 0);
129 if (port_state
== VPORT_STATE_UP
) {
130 netdev_info(priv
->netdev
, "Link up\n");
131 netif_carrier_on(priv
->netdev
);
133 netdev_info(priv
->netdev
, "Link down\n");
134 netif_carrier_off(priv
->netdev
);
138 static void mlx5e_update_carrier_work(struct work_struct
*work
)
140 struct mlx5e_priv
*priv
= container_of(work
, struct mlx5e_priv
,
141 update_carrier_work
);
143 mutex_lock(&priv
->state_lock
);
144 if (test_bit(MLX5E_STATE_OPENED
, &priv
->state
))
145 mlx5e_update_carrier(priv
);
146 mutex_unlock(&priv
->state_lock
);
149 static void mlx5e_tx_timeout_work(struct work_struct
*work
)
151 struct mlx5e_priv
*priv
= container_of(work
, struct mlx5e_priv
,
156 mutex_lock(&priv
->state_lock
);
157 if (!test_bit(MLX5E_STATE_OPENED
, &priv
->state
))
159 mlx5e_close_locked(priv
->netdev
);
160 err
= mlx5e_open_locked(priv
->netdev
);
162 netdev_err(priv
->netdev
, "mlx5e_open_locked failed recovering from a tx_timeout, err(%d).\n",
165 mutex_unlock(&priv
->state_lock
);
169 static void mlx5e_update_sw_counters(struct mlx5e_priv
*priv
)
171 struct mlx5e_sw_stats
*s
= &priv
->stats
.sw
;
172 struct mlx5e_rq_stats
*rq_stats
;
173 struct mlx5e_sq_stats
*sq_stats
;
174 u64 tx_offload_none
= 0;
177 memset(s
, 0, sizeof(*s
));
178 for (i
= 0; i
< priv
->channels
.num
; i
++) {
179 struct mlx5e_channel
*c
= priv
->channels
.c
[i
];
181 rq_stats
= &c
->rq
.stats
;
183 s
->rx_packets
+= rq_stats
->packets
;
184 s
->rx_bytes
+= rq_stats
->bytes
;
185 s
->rx_lro_packets
+= rq_stats
->lro_packets
;
186 s
->rx_lro_bytes
+= rq_stats
->lro_bytes
;
187 s
->rx_csum_none
+= rq_stats
->csum_none
;
188 s
->rx_csum_complete
+= rq_stats
->csum_complete
;
189 s
->rx_csum_unnecessary_inner
+= rq_stats
->csum_unnecessary_inner
;
190 s
->rx_xdp_drop
+= rq_stats
->xdp_drop
;
191 s
->rx_xdp_tx
+= rq_stats
->xdp_tx
;
192 s
->rx_xdp_tx_full
+= rq_stats
->xdp_tx_full
;
193 s
->rx_wqe_err
+= rq_stats
->wqe_err
;
194 s
->rx_mpwqe_filler
+= rq_stats
->mpwqe_filler
;
195 s
->rx_buff_alloc_err
+= rq_stats
->buff_alloc_err
;
196 s
->rx_cqe_compress_blks
+= rq_stats
->cqe_compress_blks
;
197 s
->rx_cqe_compress_pkts
+= rq_stats
->cqe_compress_pkts
;
198 s
->rx_cache_reuse
+= rq_stats
->cache_reuse
;
199 s
->rx_cache_full
+= rq_stats
->cache_full
;
200 s
->rx_cache_empty
+= rq_stats
->cache_empty
;
201 s
->rx_cache_busy
+= rq_stats
->cache_busy
;
203 for (j
= 0; j
< priv
->channels
.params
.num_tc
; j
++) {
204 sq_stats
= &c
->sq
[j
].stats
;
206 s
->tx_packets
+= sq_stats
->packets
;
207 s
->tx_bytes
+= sq_stats
->bytes
;
208 s
->tx_tso_packets
+= sq_stats
->tso_packets
;
209 s
->tx_tso_bytes
+= sq_stats
->tso_bytes
;
210 s
->tx_tso_inner_packets
+= sq_stats
->tso_inner_packets
;
211 s
->tx_tso_inner_bytes
+= sq_stats
->tso_inner_bytes
;
212 s
->tx_queue_stopped
+= sq_stats
->stopped
;
213 s
->tx_queue_wake
+= sq_stats
->wake
;
214 s
->tx_queue_dropped
+= sq_stats
->dropped
;
215 s
->tx_xmit_more
+= sq_stats
->xmit_more
;
216 s
->tx_csum_partial_inner
+= sq_stats
->csum_partial_inner
;
217 tx_offload_none
+= sq_stats
->csum_none
;
221 /* Update calculated offload counters */
222 s
->tx_csum_partial
= s
->tx_packets
- tx_offload_none
- s
->tx_csum_partial_inner
;
223 s
->rx_csum_unnecessary
= s
->rx_packets
- s
->rx_csum_none
- s
->rx_csum_complete
;
225 s
->link_down_events_phy
= MLX5_GET(ppcnt_reg
,
226 priv
->stats
.pport
.phy_counters
,
227 counter_set
.phys_layer_cntrs
.link_down_events
);
230 static void mlx5e_update_vport_counters(struct mlx5e_priv
*priv
)
232 int outlen
= MLX5_ST_SZ_BYTES(query_vport_counter_out
);
233 u32
*out
= (u32
*)priv
->stats
.vport
.query_vport_out
;
234 u32 in
[MLX5_ST_SZ_DW(query_vport_counter_in
)] = {0};
235 struct mlx5_core_dev
*mdev
= priv
->mdev
;
237 MLX5_SET(query_vport_counter_in
, in
, opcode
,
238 MLX5_CMD_OP_QUERY_VPORT_COUNTER
);
239 MLX5_SET(query_vport_counter_in
, in
, op_mod
, 0);
240 MLX5_SET(query_vport_counter_in
, in
, other_vport
, 0);
242 memset(out
, 0, outlen
);
243 mlx5_cmd_exec(mdev
, in
, sizeof(in
), out
, outlen
);
246 static void mlx5e_update_pport_counters(struct mlx5e_priv
*priv
)
248 struct mlx5e_pport_stats
*pstats
= &priv
->stats
.pport
;
249 struct mlx5_core_dev
*mdev
= priv
->mdev
;
250 int sz
= MLX5_ST_SZ_BYTES(ppcnt_reg
);
255 in
= mlx5_vzalloc(sz
);
259 MLX5_SET(ppcnt_reg
, in
, local_port
, 1);
261 out
= pstats
->IEEE_802_3_counters
;
262 MLX5_SET(ppcnt_reg
, in
, grp
, MLX5_IEEE_802_3_COUNTERS_GROUP
);
263 mlx5_core_access_reg(mdev
, in
, sz
, out
, sz
, MLX5_REG_PPCNT
, 0, 0);
265 out
= pstats
->RFC_2863_counters
;
266 MLX5_SET(ppcnt_reg
, in
, grp
, MLX5_RFC_2863_COUNTERS_GROUP
);
267 mlx5_core_access_reg(mdev
, in
, sz
, out
, sz
, MLX5_REG_PPCNT
, 0, 0);
269 out
= pstats
->RFC_2819_counters
;
270 MLX5_SET(ppcnt_reg
, in
, grp
, MLX5_RFC_2819_COUNTERS_GROUP
);
271 mlx5_core_access_reg(mdev
, in
, sz
, out
, sz
, MLX5_REG_PPCNT
, 0, 0);
273 out
= pstats
->phy_counters
;
274 MLX5_SET(ppcnt_reg
, in
, grp
, MLX5_PHYSICAL_LAYER_COUNTERS_GROUP
);
275 mlx5_core_access_reg(mdev
, in
, sz
, out
, sz
, MLX5_REG_PPCNT
, 0, 0);
277 if (MLX5_CAP_PCAM_FEATURE(mdev
, ppcnt_statistical_group
)) {
278 out
= pstats
->phy_statistical_counters
;
279 MLX5_SET(ppcnt_reg
, in
, grp
, MLX5_PHYSICAL_LAYER_STATISTICAL_GROUP
);
280 mlx5_core_access_reg(mdev
, in
, sz
, out
, sz
, MLX5_REG_PPCNT
, 0, 0);
283 MLX5_SET(ppcnt_reg
, in
, grp
, MLX5_PER_PRIORITY_COUNTERS_GROUP
);
284 for (prio
= 0; prio
< NUM_PPORT_PRIO
; prio
++) {
285 out
= pstats
->per_prio_counters
[prio
];
286 MLX5_SET(ppcnt_reg
, in
, prio_tc
, prio
);
287 mlx5_core_access_reg(mdev
, in
, sz
, out
, sz
,
288 MLX5_REG_PPCNT
, 0, 0);
295 static void mlx5e_update_q_counter(struct mlx5e_priv
*priv
)
297 struct mlx5e_qcounter_stats
*qcnt
= &priv
->stats
.qcnt
;
299 if (!priv
->q_counter
)
302 mlx5_core_query_out_of_buffer(priv
->mdev
, priv
->q_counter
,
303 &qcnt
->rx_out_of_buffer
);
306 static void mlx5e_update_pcie_counters(struct mlx5e_priv
*priv
)
308 struct mlx5e_pcie_stats
*pcie_stats
= &priv
->stats
.pcie
;
309 struct mlx5_core_dev
*mdev
= priv
->mdev
;
310 int sz
= MLX5_ST_SZ_BYTES(mpcnt_reg
);
314 if (!MLX5_CAP_MCAM_FEATURE(mdev
, pcie_performance_group
))
317 in
= mlx5_vzalloc(sz
);
321 out
= pcie_stats
->pcie_perf_counters
;
322 MLX5_SET(mpcnt_reg
, in
, grp
, MLX5_PCIE_PERFORMANCE_COUNTERS_GROUP
);
323 mlx5_core_access_reg(mdev
, in
, sz
, out
, sz
, MLX5_REG_MPCNT
, 0, 0);
328 void mlx5e_update_stats(struct mlx5e_priv
*priv
)
330 mlx5e_update_pcie_counters(priv
);
331 mlx5e_update_pport_counters(priv
);
332 mlx5e_update_vport_counters(priv
);
333 mlx5e_update_q_counter(priv
);
334 mlx5e_update_sw_counters(priv
);
337 void mlx5e_update_stats_work(struct work_struct
*work
)
339 struct delayed_work
*dwork
= to_delayed_work(work
);
340 struct mlx5e_priv
*priv
= container_of(dwork
, struct mlx5e_priv
,
342 mutex_lock(&priv
->state_lock
);
343 if (test_bit(MLX5E_STATE_OPENED
, &priv
->state
)) {
344 priv
->profile
->update_stats(priv
);
345 queue_delayed_work(priv
->wq
, dwork
,
346 msecs_to_jiffies(MLX5E_UPDATE_STATS_INTERVAL
));
348 mutex_unlock(&priv
->state_lock
);
351 static void mlx5e_async_event(struct mlx5_core_dev
*mdev
, void *vpriv
,
352 enum mlx5_dev_event event
, unsigned long param
)
354 struct mlx5e_priv
*priv
= vpriv
;
355 struct ptp_clock_event ptp_event
;
356 struct mlx5_eqe
*eqe
= NULL
;
358 if (!test_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLED
, &priv
->state
))
362 case MLX5_DEV_EVENT_PORT_UP
:
363 case MLX5_DEV_EVENT_PORT_DOWN
:
364 queue_work(priv
->wq
, &priv
->update_carrier_work
);
366 case MLX5_DEV_EVENT_PPS
:
367 eqe
= (struct mlx5_eqe
*)param
;
368 ptp_event
.type
= PTP_CLOCK_EXTTS
;
369 ptp_event
.index
= eqe
->data
.pps
.pin
;
370 ptp_event
.timestamp
=
371 timecounter_cyc2time(&priv
->tstamp
.clock
,
372 be64_to_cpu(eqe
->data
.pps
.time_stamp
));
373 mlx5e_pps_event_handler(vpriv
, &ptp_event
);
380 static void mlx5e_enable_async_events(struct mlx5e_priv
*priv
)
382 set_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLED
, &priv
->state
);
385 static void mlx5e_disable_async_events(struct mlx5e_priv
*priv
)
387 clear_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLED
, &priv
->state
);
388 synchronize_irq(mlx5_get_msix_vec(priv
->mdev
, MLX5_EQ_VEC_ASYNC
));
391 static inline int mlx5e_get_wqe_mtt_sz(void)
393 /* UMR copies MTTs in units of MLX5_UMR_MTT_ALIGNMENT bytes.
394 * To avoid copying garbage after the mtt array, we allocate
397 return ALIGN(MLX5_MPWRQ_PAGES_PER_WQE
* sizeof(__be64
),
398 MLX5_UMR_MTT_ALIGNMENT
);
401 static inline void mlx5e_build_umr_wqe(struct mlx5e_rq
*rq
,
402 struct mlx5e_icosq
*sq
,
403 struct mlx5e_umr_wqe
*wqe
,
406 struct mlx5_wqe_ctrl_seg
*cseg
= &wqe
->ctrl
;
407 struct mlx5_wqe_umr_ctrl_seg
*ucseg
= &wqe
->uctrl
;
408 struct mlx5_wqe_data_seg
*dseg
= &wqe
->data
;
409 struct mlx5e_mpw_info
*wi
= &rq
->mpwqe
.info
[ix
];
410 u8 ds_cnt
= DIV_ROUND_UP(sizeof(*wqe
), MLX5_SEND_WQE_DS
);
411 u32 umr_wqe_mtt_offset
= mlx5e_get_wqe_mtt_offset(rq
, ix
);
413 cseg
->qpn_ds
= cpu_to_be32((sq
->sqn
<< MLX5_WQE_CTRL_QPN_SHIFT
) |
415 cseg
->fm_ce_se
= MLX5_WQE_CTRL_CQ_UPDATE
;
416 cseg
->imm
= rq
->mkey_be
;
418 ucseg
->flags
= MLX5_UMR_TRANSLATION_OFFSET_EN
;
419 ucseg
->xlt_octowords
=
420 cpu_to_be16(MLX5_MTT_OCTW(MLX5_MPWRQ_PAGES_PER_WQE
));
421 ucseg
->bsf_octowords
=
422 cpu_to_be16(MLX5_MTT_OCTW(umr_wqe_mtt_offset
));
423 ucseg
->mkey_mask
= cpu_to_be64(MLX5_MKEY_MASK_FREE
);
425 dseg
->lkey
= sq
->mkey_be
;
426 dseg
->addr
= cpu_to_be64(wi
->umr
.mtt_addr
);
429 static int mlx5e_rq_alloc_mpwqe_info(struct mlx5e_rq
*rq
,
430 struct mlx5e_channel
*c
)
432 int wq_sz
= mlx5_wq_ll_get_size(&rq
->wq
);
433 int mtt_sz
= mlx5e_get_wqe_mtt_sz();
434 int mtt_alloc
= mtt_sz
+ MLX5_UMR_ALIGN
- 1;
437 rq
->mpwqe
.info
= kzalloc_node(wq_sz
* sizeof(*rq
->mpwqe
.info
),
438 GFP_KERNEL
, cpu_to_node(c
->cpu
));
442 /* We allocate more than mtt_sz as we will align the pointer */
443 rq
->mpwqe
.mtt_no_align
= kzalloc_node(mtt_alloc
* wq_sz
, GFP_KERNEL
,
444 cpu_to_node(c
->cpu
));
445 if (unlikely(!rq
->mpwqe
.mtt_no_align
))
446 goto err_free_wqe_info
;
448 for (i
= 0; i
< wq_sz
; i
++) {
449 struct mlx5e_mpw_info
*wi
= &rq
->mpwqe
.info
[i
];
451 wi
->umr
.mtt
= PTR_ALIGN(rq
->mpwqe
.mtt_no_align
+ i
* mtt_alloc
,
453 wi
->umr
.mtt_addr
= dma_map_single(c
->pdev
, wi
->umr
.mtt
, mtt_sz
,
455 if (unlikely(dma_mapping_error(c
->pdev
, wi
->umr
.mtt_addr
)))
458 mlx5e_build_umr_wqe(rq
, &c
->icosq
, &wi
->umr
.wqe
, i
);
465 struct mlx5e_mpw_info
*wi
= &rq
->mpwqe
.info
[i
];
467 dma_unmap_single(c
->pdev
, wi
->umr
.mtt_addr
, mtt_sz
,
470 kfree(rq
->mpwqe
.mtt_no_align
);
472 kfree(rq
->mpwqe
.info
);
478 static void mlx5e_rq_free_mpwqe_info(struct mlx5e_rq
*rq
)
480 int wq_sz
= mlx5_wq_ll_get_size(&rq
->wq
);
481 int mtt_sz
= mlx5e_get_wqe_mtt_sz();
484 for (i
= 0; i
< wq_sz
; i
++) {
485 struct mlx5e_mpw_info
*wi
= &rq
->mpwqe
.info
[i
];
487 dma_unmap_single(rq
->pdev
, wi
->umr
.mtt_addr
, mtt_sz
,
490 kfree(rq
->mpwqe
.mtt_no_align
);
491 kfree(rq
->mpwqe
.info
);
494 static int mlx5e_create_umr_mkey(struct mlx5_core_dev
*mdev
,
495 u64 npages
, u8 page_shift
,
496 struct mlx5_core_mkey
*umr_mkey
)
498 int inlen
= MLX5_ST_SZ_BYTES(create_mkey_in
);
503 if (!MLX5E_VALID_NUM_MTTS(npages
))
506 in
= mlx5_vzalloc(inlen
);
510 mkc
= MLX5_ADDR_OF(create_mkey_in
, in
, memory_key_mkey_entry
);
512 MLX5_SET(mkc
, mkc
, free
, 1);
513 MLX5_SET(mkc
, mkc
, umr_en
, 1);
514 MLX5_SET(mkc
, mkc
, lw
, 1);
515 MLX5_SET(mkc
, mkc
, lr
, 1);
516 MLX5_SET(mkc
, mkc
, access_mode
, MLX5_MKC_ACCESS_MODE_MTT
);
518 MLX5_SET(mkc
, mkc
, qpn
, 0xffffff);
519 MLX5_SET(mkc
, mkc
, pd
, mdev
->mlx5e_res
.pdn
);
520 MLX5_SET64(mkc
, mkc
, len
, npages
<< page_shift
);
521 MLX5_SET(mkc
, mkc
, translations_octword_size
,
522 MLX5_MTT_OCTW(npages
));
523 MLX5_SET(mkc
, mkc
, log_page_size
, page_shift
);
525 err
= mlx5_core_create_mkey(mdev
, umr_mkey
, in
, inlen
);
531 static int mlx5e_create_rq_umr_mkey(struct mlx5_core_dev
*mdev
, struct mlx5e_rq
*rq
)
533 u64 num_mtts
= MLX5E_REQUIRED_MTTS(mlx5_wq_ll_get_size(&rq
->wq
));
535 return mlx5e_create_umr_mkey(mdev
, num_mtts
, PAGE_SHIFT
, &rq
->umr_mkey
);
538 static int mlx5e_alloc_rq(struct mlx5e_channel
*c
,
539 struct mlx5e_params
*params
,
540 struct mlx5e_rq_param
*rqp
,
543 struct mlx5_core_dev
*mdev
= c
->mdev
;
544 void *rqc
= rqp
->rqc
;
545 void *rqc_wq
= MLX5_ADDR_OF(rqc
, rqc
, wq
);
553 rqp
->wq
.db_numa_node
= cpu_to_node(c
->cpu
);
555 err
= mlx5_wq_ll_create(mdev
, &rqp
->wq
, rqc_wq
, &rq
->wq
,
560 rq
->wq
.db
= &rq
->wq
.db
[MLX5_RCV_DBR
];
562 wq_sz
= mlx5_wq_ll_get_size(&rq
->wq
);
564 rq
->wq_type
= params
->rq_wq_type
;
566 rq
->netdev
= c
->netdev
;
567 rq
->tstamp
= c
->tstamp
;
572 rq
->xdp_prog
= params
->xdp_prog
? bpf_prog_inc(params
->xdp_prog
) : NULL
;
573 if (IS_ERR(rq
->xdp_prog
)) {
574 err
= PTR_ERR(rq
->xdp_prog
);
576 goto err_rq_wq_destroy
;
580 rq
->buff
.map_dir
= DMA_BIDIRECTIONAL
;
581 rq
->rx_headroom
= XDP_PACKET_HEADROOM
;
583 rq
->buff
.map_dir
= DMA_FROM_DEVICE
;
584 rq
->rx_headroom
= MLX5_RX_HEADROOM
;
587 switch (rq
->wq_type
) {
588 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ
:
589 if (mlx5e_is_vf_vport_rep(c
->priv
)) {
591 goto err_rq_wq_destroy
;
594 rq
->handle_rx_cqe
= mlx5e_handle_rx_cqe_mpwrq
;
595 rq
->alloc_wqe
= mlx5e_alloc_rx_mpwqe
;
596 rq
->dealloc_wqe
= mlx5e_dealloc_rx_mpwqe
;
598 rq
->mpwqe_stride_sz
= BIT(params
->mpwqe_log_stride_sz
);
599 rq
->mpwqe_num_strides
= BIT(params
->mpwqe_log_num_strides
);
601 rq
->buff
.wqe_sz
= rq
->mpwqe_stride_sz
* rq
->mpwqe_num_strides
;
602 byte_count
= rq
->buff
.wqe_sz
;
604 err
= mlx5e_create_rq_umr_mkey(mdev
, rq
);
606 goto err_rq_wq_destroy
;
607 rq
->mkey_be
= cpu_to_be32(rq
->umr_mkey
.key
);
609 err
= mlx5e_rq_alloc_mpwqe_info(rq
, c
);
611 goto err_destroy_umr_mkey
;
613 default: /* MLX5_WQ_TYPE_LINKED_LIST */
614 rq
->dma_info
= kzalloc_node(wq_sz
* sizeof(*rq
->dma_info
),
615 GFP_KERNEL
, cpu_to_node(c
->cpu
));
618 goto err_rq_wq_destroy
;
621 if (mlx5e_is_vf_vport_rep(c
->priv
))
622 rq
->handle_rx_cqe
= mlx5e_handle_rx_cqe_rep
;
624 rq
->handle_rx_cqe
= mlx5e_handle_rx_cqe
;
626 rq
->alloc_wqe
= mlx5e_alloc_rx_wqe
;
627 rq
->dealloc_wqe
= mlx5e_dealloc_rx_wqe
;
629 rq
->buff
.wqe_sz
= params
->lro_en
?
631 MLX5E_SW2HW_MTU(c
->netdev
->mtu
);
632 byte_count
= rq
->buff
.wqe_sz
;
634 /* calc the required page order */
635 frag_sz
= rq
->rx_headroom
+
636 byte_count
/* packet data */ +
637 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
638 frag_sz
= SKB_DATA_ALIGN(frag_sz
);
640 npages
= DIV_ROUND_UP(frag_sz
, PAGE_SIZE
);
641 rq
->buff
.page_order
= order_base_2(npages
);
643 byte_count
|= MLX5_HW_START_PADDING
;
644 rq
->mkey_be
= c
->mkey_be
;
647 for (i
= 0; i
< wq_sz
; i
++) {
648 struct mlx5e_rx_wqe
*wqe
= mlx5_wq_ll_get_wqe(&rq
->wq
, i
);
650 wqe
->data
.byte_count
= cpu_to_be32(byte_count
);
651 wqe
->data
.lkey
= rq
->mkey_be
;
654 INIT_WORK(&rq
->am
.work
, mlx5e_rx_am_work
);
655 rq
->am
.mode
= params
->rx_cq_period_mode
;
656 rq
->page_cache
.head
= 0;
657 rq
->page_cache
.tail
= 0;
661 err_destroy_umr_mkey
:
662 mlx5_core_destroy_mkey(mdev
, &rq
->umr_mkey
);
666 bpf_prog_put(rq
->xdp_prog
);
667 mlx5_wq_destroy(&rq
->wq_ctrl
);
672 static void mlx5e_free_rq(struct mlx5e_rq
*rq
)
677 bpf_prog_put(rq
->xdp_prog
);
679 switch (rq
->wq_type
) {
680 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ
:
681 mlx5e_rq_free_mpwqe_info(rq
);
682 mlx5_core_destroy_mkey(rq
->mdev
, &rq
->umr_mkey
);
684 default: /* MLX5_WQ_TYPE_LINKED_LIST */
688 for (i
= rq
->page_cache
.head
; i
!= rq
->page_cache
.tail
;
689 i
= (i
+ 1) & (MLX5E_CACHE_SIZE
- 1)) {
690 struct mlx5e_dma_info
*dma_info
= &rq
->page_cache
.page_cache
[i
];
692 mlx5e_page_release(rq
, dma_info
, false);
694 mlx5_wq_destroy(&rq
->wq_ctrl
);
697 static int mlx5e_create_rq(struct mlx5e_rq
*rq
,
698 struct mlx5e_rq_param
*param
)
700 struct mlx5_core_dev
*mdev
= rq
->mdev
;
708 inlen
= MLX5_ST_SZ_BYTES(create_rq_in
) +
709 sizeof(u64
) * rq
->wq_ctrl
.buf
.npages
;
710 in
= mlx5_vzalloc(inlen
);
714 rqc
= MLX5_ADDR_OF(create_rq_in
, in
, ctx
);
715 wq
= MLX5_ADDR_OF(rqc
, rqc
, wq
);
717 memcpy(rqc
, param
->rqc
, sizeof(param
->rqc
));
719 MLX5_SET(rqc
, rqc
, cqn
, rq
->cq
.mcq
.cqn
);
720 MLX5_SET(rqc
, rqc
, state
, MLX5_RQC_STATE_RST
);
721 MLX5_SET(wq
, wq
, log_wq_pg_sz
, rq
->wq_ctrl
.buf
.page_shift
-
722 MLX5_ADAPTER_PAGE_SHIFT
);
723 MLX5_SET64(wq
, wq
, dbr_addr
, rq
->wq_ctrl
.db
.dma
);
725 mlx5_fill_page_array(&rq
->wq_ctrl
.buf
,
726 (__be64
*)MLX5_ADDR_OF(wq
, wq
, pas
));
728 err
= mlx5_core_create_rq(mdev
, in
, inlen
, &rq
->rqn
);
735 static int mlx5e_modify_rq_state(struct mlx5e_rq
*rq
, int curr_state
,
738 struct mlx5e_channel
*c
= rq
->channel
;
739 struct mlx5_core_dev
*mdev
= c
->mdev
;
746 inlen
= MLX5_ST_SZ_BYTES(modify_rq_in
);
747 in
= mlx5_vzalloc(inlen
);
751 rqc
= MLX5_ADDR_OF(modify_rq_in
, in
, ctx
);
753 MLX5_SET(modify_rq_in
, in
, rq_state
, curr_state
);
754 MLX5_SET(rqc
, rqc
, state
, next_state
);
756 err
= mlx5_core_modify_rq(mdev
, rq
->rqn
, in
, inlen
);
763 static int mlx5e_modify_rq_scatter_fcs(struct mlx5e_rq
*rq
, bool enable
)
765 struct mlx5e_channel
*c
= rq
->channel
;
766 struct mlx5e_priv
*priv
= c
->priv
;
767 struct mlx5_core_dev
*mdev
= priv
->mdev
;
774 inlen
= MLX5_ST_SZ_BYTES(modify_rq_in
);
775 in
= mlx5_vzalloc(inlen
);
779 rqc
= MLX5_ADDR_OF(modify_rq_in
, in
, ctx
);
781 MLX5_SET(modify_rq_in
, in
, rq_state
, MLX5_RQC_STATE_RDY
);
782 MLX5_SET64(modify_rq_in
, in
, modify_bitmask
,
783 MLX5_MODIFY_RQ_IN_MODIFY_BITMASK_SCATTER_FCS
);
784 MLX5_SET(rqc
, rqc
, scatter_fcs
, enable
);
785 MLX5_SET(rqc
, rqc
, state
, MLX5_RQC_STATE_RDY
);
787 err
= mlx5_core_modify_rq(mdev
, rq
->rqn
, in
, inlen
);
794 static int mlx5e_modify_rq_vsd(struct mlx5e_rq
*rq
, bool vsd
)
796 struct mlx5e_channel
*c
= rq
->channel
;
797 struct mlx5_core_dev
*mdev
= c
->mdev
;
803 inlen
= MLX5_ST_SZ_BYTES(modify_rq_in
);
804 in
= mlx5_vzalloc(inlen
);
808 rqc
= MLX5_ADDR_OF(modify_rq_in
, in
, ctx
);
810 MLX5_SET(modify_rq_in
, in
, rq_state
, MLX5_RQC_STATE_RDY
);
811 MLX5_SET64(modify_rq_in
, in
, modify_bitmask
,
812 MLX5_MODIFY_RQ_IN_MODIFY_BITMASK_VSD
);
813 MLX5_SET(rqc
, rqc
, vsd
, vsd
);
814 MLX5_SET(rqc
, rqc
, state
, MLX5_RQC_STATE_RDY
);
816 err
= mlx5_core_modify_rq(mdev
, rq
->rqn
, in
, inlen
);
823 static void mlx5e_destroy_rq(struct mlx5e_rq
*rq
)
825 mlx5_core_destroy_rq(rq
->mdev
, rq
->rqn
);
828 static int mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq
*rq
)
830 unsigned long exp_time
= jiffies
+ msecs_to_jiffies(20000);
831 struct mlx5e_channel
*c
= rq
->channel
;
833 struct mlx5_wq_ll
*wq
= &rq
->wq
;
834 u16 min_wqes
= mlx5_min_rx_wqes(rq
->wq_type
, mlx5_wq_ll_get_size(wq
));
836 while (time_before(jiffies
, exp_time
)) {
837 if (wq
->cur_sz
>= min_wqes
)
843 netdev_warn(c
->netdev
, "Failed to get min RX wqes on RQN[0x%x] wq cur_sz(%d) min_rx_wqes(%d)\n",
844 rq
->rqn
, wq
->cur_sz
, min_wqes
);
848 static void mlx5e_free_rx_descs(struct mlx5e_rq
*rq
)
850 struct mlx5_wq_ll
*wq
= &rq
->wq
;
851 struct mlx5e_rx_wqe
*wqe
;
855 /* UMR WQE (if in progress) is always at wq->head */
856 if (test_bit(MLX5E_RQ_STATE_UMR_WQE_IN_PROGRESS
, &rq
->state
))
857 mlx5e_free_rx_mpwqe(rq
, &rq
->mpwqe
.info
[wq
->head
]);
859 while (!mlx5_wq_ll_is_empty(wq
)) {
860 wqe_ix_be
= *wq
->tail_next
;
861 wqe_ix
= be16_to_cpu(wqe_ix_be
);
862 wqe
= mlx5_wq_ll_get_wqe(&rq
->wq
, wqe_ix
);
863 rq
->dealloc_wqe(rq
, wqe_ix
);
864 mlx5_wq_ll_pop(&rq
->wq
, wqe_ix_be
,
865 &wqe
->next
.next_wqe_index
);
869 static int mlx5e_open_rq(struct mlx5e_channel
*c
,
870 struct mlx5e_params
*params
,
871 struct mlx5e_rq_param
*param
,
876 err
= mlx5e_alloc_rq(c
, params
, param
, rq
);
880 err
= mlx5e_create_rq(rq
, param
);
884 err
= mlx5e_modify_rq_state(rq
, MLX5_RQC_STATE_RST
, MLX5_RQC_STATE_RDY
);
888 if (params
->rx_am_enabled
)
889 set_bit(MLX5E_RQ_STATE_AM
, &c
->rq
.state
);
894 mlx5e_destroy_rq(rq
);
901 static void mlx5e_activate_rq(struct mlx5e_rq
*rq
)
903 struct mlx5e_icosq
*sq
= &rq
->channel
->icosq
;
904 u16 pi
= sq
->pc
& sq
->wq
.sz_m1
;
905 struct mlx5e_tx_wqe
*nopwqe
;
907 set_bit(MLX5E_RQ_STATE_ENABLED
, &rq
->state
);
908 sq
->db
.ico_wqe
[pi
].opcode
= MLX5_OPCODE_NOP
;
909 sq
->db
.ico_wqe
[pi
].num_wqebbs
= 1;
910 nopwqe
= mlx5e_post_nop(&sq
->wq
, sq
->sqn
, &sq
->pc
);
911 mlx5e_notify_hw(&sq
->wq
, sq
->pc
, sq
->uar_map
, &nopwqe
->ctrl
);
914 static void mlx5e_deactivate_rq(struct mlx5e_rq
*rq
)
916 clear_bit(MLX5E_RQ_STATE_ENABLED
, &rq
->state
);
917 napi_synchronize(&rq
->channel
->napi
); /* prevent mlx5e_post_rx_wqes */
920 static void mlx5e_close_rq(struct mlx5e_rq
*rq
)
922 cancel_work_sync(&rq
->am
.work
);
923 mlx5e_destroy_rq(rq
);
924 mlx5e_free_rx_descs(rq
);
928 static void mlx5e_free_xdpsq_db(struct mlx5e_xdpsq
*sq
)
933 static int mlx5e_alloc_xdpsq_db(struct mlx5e_xdpsq
*sq
, int numa
)
935 int wq_sz
= mlx5_wq_cyc_get_size(&sq
->wq
);
937 sq
->db
.di
= kzalloc_node(sizeof(*sq
->db
.di
) * wq_sz
,
940 mlx5e_free_xdpsq_db(sq
);
947 static int mlx5e_alloc_xdpsq(struct mlx5e_channel
*c
,
948 struct mlx5e_params
*params
,
949 struct mlx5e_sq_param
*param
,
950 struct mlx5e_xdpsq
*sq
)
952 void *sqc_wq
= MLX5_ADDR_OF(sqc
, param
->sqc
, wq
);
953 struct mlx5_core_dev
*mdev
= c
->mdev
;
957 sq
->mkey_be
= c
->mkey_be
;
959 sq
->uar_map
= mdev
->mlx5e_res
.bfreg
.map
;
960 sq
->min_inline_mode
= params
->tx_min_inline_mode
;
962 param
->wq
.db_numa_node
= cpu_to_node(c
->cpu
);
963 err
= mlx5_wq_cyc_create(mdev
, ¶m
->wq
, sqc_wq
, &sq
->wq
, &sq
->wq_ctrl
);
966 sq
->wq
.db
= &sq
->wq
.db
[MLX5_SND_DBR
];
968 err
= mlx5e_alloc_xdpsq_db(sq
, cpu_to_node(c
->cpu
));
970 goto err_sq_wq_destroy
;
975 mlx5_wq_destroy(&sq
->wq_ctrl
);
980 static void mlx5e_free_xdpsq(struct mlx5e_xdpsq
*sq
)
982 mlx5e_free_xdpsq_db(sq
);
983 mlx5_wq_destroy(&sq
->wq_ctrl
);
986 static void mlx5e_free_icosq_db(struct mlx5e_icosq
*sq
)
988 kfree(sq
->db
.ico_wqe
);
991 static int mlx5e_alloc_icosq_db(struct mlx5e_icosq
*sq
, int numa
)
993 u8 wq_sz
= mlx5_wq_cyc_get_size(&sq
->wq
);
995 sq
->db
.ico_wqe
= kzalloc_node(sizeof(*sq
->db
.ico_wqe
) * wq_sz
,
1003 static int mlx5e_alloc_icosq(struct mlx5e_channel
*c
,
1004 struct mlx5e_sq_param
*param
,
1005 struct mlx5e_icosq
*sq
)
1007 void *sqc_wq
= MLX5_ADDR_OF(sqc
, param
->sqc
, wq
);
1008 struct mlx5_core_dev
*mdev
= c
->mdev
;
1012 sq
->mkey_be
= c
->mkey_be
;
1014 sq
->uar_map
= mdev
->mlx5e_res
.bfreg
.map
;
1016 param
->wq
.db_numa_node
= cpu_to_node(c
->cpu
);
1017 err
= mlx5_wq_cyc_create(mdev
, ¶m
->wq
, sqc_wq
, &sq
->wq
, &sq
->wq_ctrl
);
1020 sq
->wq
.db
= &sq
->wq
.db
[MLX5_SND_DBR
];
1022 err
= mlx5e_alloc_icosq_db(sq
, cpu_to_node(c
->cpu
));
1024 goto err_sq_wq_destroy
;
1026 sq
->edge
= (sq
->wq
.sz_m1
+ 1) - MLX5E_ICOSQ_MAX_WQEBBS
;
1031 mlx5_wq_destroy(&sq
->wq_ctrl
);
1036 static void mlx5e_free_icosq(struct mlx5e_icosq
*sq
)
1038 mlx5e_free_icosq_db(sq
);
1039 mlx5_wq_destroy(&sq
->wq_ctrl
);
1042 static void mlx5e_free_txqsq_db(struct mlx5e_txqsq
*sq
)
1044 kfree(sq
->db
.wqe_info
);
1045 kfree(sq
->db
.dma_fifo
);
1049 static int mlx5e_alloc_txqsq_db(struct mlx5e_txqsq
*sq
, int numa
)
1051 int wq_sz
= mlx5_wq_cyc_get_size(&sq
->wq
);
1052 int df_sz
= wq_sz
* MLX5_SEND_WQEBB_NUM_DS
;
1054 sq
->db
.skb
= kzalloc_node(wq_sz
* sizeof(*sq
->db
.skb
),
1056 sq
->db
.dma_fifo
= kzalloc_node(df_sz
* sizeof(*sq
->db
.dma_fifo
),
1058 sq
->db
.wqe_info
= kzalloc_node(wq_sz
* sizeof(*sq
->db
.wqe_info
),
1060 if (!sq
->db
.skb
|| !sq
->db
.dma_fifo
|| !sq
->db
.wqe_info
) {
1061 mlx5e_free_txqsq_db(sq
);
1065 sq
->dma_fifo_mask
= df_sz
- 1;
1070 static int mlx5e_alloc_txqsq(struct mlx5e_channel
*c
,
1072 struct mlx5e_params
*params
,
1073 struct mlx5e_sq_param
*param
,
1074 struct mlx5e_txqsq
*sq
)
1076 void *sqc_wq
= MLX5_ADDR_OF(sqc
, param
->sqc
, wq
);
1077 struct mlx5_core_dev
*mdev
= c
->mdev
;
1081 sq
->tstamp
= c
->tstamp
;
1082 sq
->mkey_be
= c
->mkey_be
;
1084 sq
->txq_ix
= txq_ix
;
1085 sq
->uar_map
= mdev
->mlx5e_res
.bfreg
.map
;
1086 sq
->max_inline
= params
->tx_max_inline
;
1087 sq
->min_inline_mode
= params
->tx_min_inline_mode
;
1089 param
->wq
.db_numa_node
= cpu_to_node(c
->cpu
);
1090 err
= mlx5_wq_cyc_create(mdev
, ¶m
->wq
, sqc_wq
, &sq
->wq
, &sq
->wq_ctrl
);
1093 sq
->wq
.db
= &sq
->wq
.db
[MLX5_SND_DBR
];
1095 err
= mlx5e_alloc_txqsq_db(sq
, cpu_to_node(c
->cpu
));
1097 goto err_sq_wq_destroy
;
1099 sq
->edge
= (sq
->wq
.sz_m1
+ 1) - MLX5_SEND_WQE_MAX_WQEBBS
;
1104 mlx5_wq_destroy(&sq
->wq_ctrl
);
1109 static void mlx5e_free_txqsq(struct mlx5e_txqsq
*sq
)
1111 mlx5e_free_txqsq_db(sq
);
1112 mlx5_wq_destroy(&sq
->wq_ctrl
);
1115 struct mlx5e_create_sq_param
{
1116 struct mlx5_wq_ctrl
*wq_ctrl
;
1123 static int mlx5e_create_sq(struct mlx5_core_dev
*mdev
,
1124 struct mlx5e_sq_param
*param
,
1125 struct mlx5e_create_sq_param
*csp
,
1134 inlen
= MLX5_ST_SZ_BYTES(create_sq_in
) +
1135 sizeof(u64
) * csp
->wq_ctrl
->buf
.npages
;
1136 in
= mlx5_vzalloc(inlen
);
1140 sqc
= MLX5_ADDR_OF(create_sq_in
, in
, ctx
);
1141 wq
= MLX5_ADDR_OF(sqc
, sqc
, wq
);
1143 memcpy(sqc
, param
->sqc
, sizeof(param
->sqc
));
1144 MLX5_SET(sqc
, sqc
, tis_lst_sz
, csp
->tis_lst_sz
);
1145 MLX5_SET(sqc
, sqc
, tis_num_0
, csp
->tisn
);
1146 MLX5_SET(sqc
, sqc
, cqn
, csp
->cqn
);
1148 if (MLX5_CAP_ETH(mdev
, wqe_inline_mode
) == MLX5_CAP_INLINE_MODE_VPORT_CONTEXT
)
1149 MLX5_SET(sqc
, sqc
, min_wqe_inline_mode
, csp
->min_inline_mode
);
1151 MLX5_SET(sqc
, sqc
, state
, MLX5_SQC_STATE_RST
);
1153 MLX5_SET(wq
, wq
, wq_type
, MLX5_WQ_TYPE_CYCLIC
);
1154 MLX5_SET(wq
, wq
, uar_page
, mdev
->mlx5e_res
.bfreg
.index
);
1155 MLX5_SET(wq
, wq
, log_wq_pg_sz
, csp
->wq_ctrl
->buf
.page_shift
-
1156 MLX5_ADAPTER_PAGE_SHIFT
);
1157 MLX5_SET64(wq
, wq
, dbr_addr
, csp
->wq_ctrl
->db
.dma
);
1159 mlx5_fill_page_array(&csp
->wq_ctrl
->buf
, (__be64
*)MLX5_ADDR_OF(wq
, wq
, pas
));
1161 err
= mlx5_core_create_sq(mdev
, in
, inlen
, sqn
);
1168 struct mlx5e_modify_sq_param
{
1175 static int mlx5e_modify_sq(struct mlx5_core_dev
*mdev
, u32 sqn
,
1176 struct mlx5e_modify_sq_param
*p
)
1183 inlen
= MLX5_ST_SZ_BYTES(modify_sq_in
);
1184 in
= mlx5_vzalloc(inlen
);
1188 sqc
= MLX5_ADDR_OF(modify_sq_in
, in
, ctx
);
1190 MLX5_SET(modify_sq_in
, in
, sq_state
, p
->curr_state
);
1191 MLX5_SET(sqc
, sqc
, state
, p
->next_state
);
1192 if (p
->rl_update
&& p
->next_state
== MLX5_SQC_STATE_RDY
) {
1193 MLX5_SET64(modify_sq_in
, in
, modify_bitmask
, 1);
1194 MLX5_SET(sqc
, sqc
, packet_pacing_rate_limit_index
, p
->rl_index
);
1197 err
= mlx5_core_modify_sq(mdev
, sqn
, in
, inlen
);
1204 static void mlx5e_destroy_sq(struct mlx5_core_dev
*mdev
, u32 sqn
)
1206 mlx5_core_destroy_sq(mdev
, sqn
);
1209 static int mlx5e_create_sq_rdy(struct mlx5_core_dev
*mdev
,
1210 struct mlx5e_sq_param
*param
,
1211 struct mlx5e_create_sq_param
*csp
,
1214 struct mlx5e_modify_sq_param msp
= {0};
1217 err
= mlx5e_create_sq(mdev
, param
, csp
, sqn
);
1221 msp
.curr_state
= MLX5_SQC_STATE_RST
;
1222 msp
.next_state
= MLX5_SQC_STATE_RDY
;
1223 err
= mlx5e_modify_sq(mdev
, *sqn
, &msp
);
1225 mlx5e_destroy_sq(mdev
, *sqn
);
1230 static int mlx5e_set_sq_maxrate(struct net_device
*dev
,
1231 struct mlx5e_txqsq
*sq
, u32 rate
);
1233 static int mlx5e_open_txqsq(struct mlx5e_channel
*c
,
1236 struct mlx5e_params
*params
,
1237 struct mlx5e_sq_param
*param
,
1238 struct mlx5e_txqsq
*sq
)
1240 struct mlx5e_create_sq_param csp
= {};
1244 err
= mlx5e_alloc_txqsq(c
, txq_ix
, params
, param
, sq
);
1250 csp
.cqn
= sq
->cq
.mcq
.cqn
;
1251 csp
.wq_ctrl
= &sq
->wq_ctrl
;
1252 csp
.min_inline_mode
= sq
->min_inline_mode
;
1253 err
= mlx5e_create_sq_rdy(c
->mdev
, param
, &csp
, &sq
->sqn
);
1255 goto err_free_txqsq
;
1257 tx_rate
= c
->priv
->tx_rates
[sq
->txq_ix
];
1259 mlx5e_set_sq_maxrate(c
->netdev
, sq
, tx_rate
);
1264 clear_bit(MLX5E_SQ_STATE_ENABLED
, &sq
->state
);
1265 mlx5e_free_txqsq(sq
);
1270 static void mlx5e_activate_txqsq(struct mlx5e_txqsq
*sq
)
1272 sq
->txq
= netdev_get_tx_queue(sq
->channel
->netdev
, sq
->txq_ix
);
1273 set_bit(MLX5E_SQ_STATE_ENABLED
, &sq
->state
);
1274 netdev_tx_reset_queue(sq
->txq
);
1275 netif_tx_start_queue(sq
->txq
);
1278 static inline void netif_tx_disable_queue(struct netdev_queue
*txq
)
1280 __netif_tx_lock_bh(txq
);
1281 netif_tx_stop_queue(txq
);
1282 __netif_tx_unlock_bh(txq
);
1285 static void mlx5e_deactivate_txqsq(struct mlx5e_txqsq
*sq
)
1287 struct mlx5e_channel
*c
= sq
->channel
;
1289 clear_bit(MLX5E_SQ_STATE_ENABLED
, &sq
->state
);
1290 /* prevent netif_tx_wake_queue */
1291 napi_synchronize(&c
->napi
);
1293 netif_tx_disable_queue(sq
->txq
);
1295 /* last doorbell out, godspeed .. */
1296 if (mlx5e_wqc_has_room_for(&sq
->wq
, sq
->cc
, sq
->pc
, 1)) {
1297 struct mlx5e_tx_wqe
*nop
;
1299 sq
->db
.skb
[(sq
->pc
& sq
->wq
.sz_m1
)] = NULL
;
1300 nop
= mlx5e_post_nop(&sq
->wq
, sq
->sqn
, &sq
->pc
);
1301 mlx5e_notify_hw(&sq
->wq
, sq
->pc
, sq
->uar_map
, &nop
->ctrl
);
1305 static void mlx5e_close_txqsq(struct mlx5e_txqsq
*sq
)
1307 struct mlx5e_channel
*c
= sq
->channel
;
1308 struct mlx5_core_dev
*mdev
= c
->mdev
;
1310 mlx5e_destroy_sq(mdev
, sq
->sqn
);
1312 mlx5_rl_remove_rate(mdev
, sq
->rate_limit
);
1313 mlx5e_free_txqsq_descs(sq
);
1314 mlx5e_free_txqsq(sq
);
1317 static int mlx5e_open_icosq(struct mlx5e_channel
*c
,
1318 struct mlx5e_params
*params
,
1319 struct mlx5e_sq_param
*param
,
1320 struct mlx5e_icosq
*sq
)
1322 struct mlx5e_create_sq_param csp
= {};
1325 err
= mlx5e_alloc_icosq(c
, param
, sq
);
1329 csp
.cqn
= sq
->cq
.mcq
.cqn
;
1330 csp
.wq_ctrl
= &sq
->wq_ctrl
;
1331 csp
.min_inline_mode
= params
->tx_min_inline_mode
;
1332 set_bit(MLX5E_SQ_STATE_ENABLED
, &sq
->state
);
1333 err
= mlx5e_create_sq_rdy(c
->mdev
, param
, &csp
, &sq
->sqn
);
1335 goto err_free_icosq
;
1340 clear_bit(MLX5E_SQ_STATE_ENABLED
, &sq
->state
);
1341 mlx5e_free_icosq(sq
);
1346 static void mlx5e_close_icosq(struct mlx5e_icosq
*sq
)
1348 struct mlx5e_channel
*c
= sq
->channel
;
1350 clear_bit(MLX5E_SQ_STATE_ENABLED
, &sq
->state
);
1351 napi_synchronize(&c
->napi
);
1353 mlx5e_destroy_sq(c
->mdev
, sq
->sqn
);
1354 mlx5e_free_icosq(sq
);
1357 static int mlx5e_open_xdpsq(struct mlx5e_channel
*c
,
1358 struct mlx5e_params
*params
,
1359 struct mlx5e_sq_param
*param
,
1360 struct mlx5e_xdpsq
*sq
)
1362 unsigned int ds_cnt
= MLX5E_XDP_TX_DS_COUNT
;
1363 struct mlx5e_create_sq_param csp
= {};
1364 unsigned int inline_hdr_sz
= 0;
1368 err
= mlx5e_alloc_xdpsq(c
, params
, param
, sq
);
1373 csp
.tisn
= c
->priv
->tisn
[0]; /* tc = 0 */
1374 csp
.cqn
= sq
->cq
.mcq
.cqn
;
1375 csp
.wq_ctrl
= &sq
->wq_ctrl
;
1376 csp
.min_inline_mode
= sq
->min_inline_mode
;
1377 set_bit(MLX5E_SQ_STATE_ENABLED
, &sq
->state
);
1378 err
= mlx5e_create_sq_rdy(c
->mdev
, param
, &csp
, &sq
->sqn
);
1380 goto err_free_xdpsq
;
1382 if (sq
->min_inline_mode
!= MLX5_INLINE_MODE_NONE
) {
1383 inline_hdr_sz
= MLX5E_XDP_MIN_INLINE
;
1387 /* Pre initialize fixed WQE fields */
1388 for (i
= 0; i
< mlx5_wq_cyc_get_size(&sq
->wq
); i
++) {
1389 struct mlx5e_tx_wqe
*wqe
= mlx5_wq_cyc_get_wqe(&sq
->wq
, i
);
1390 struct mlx5_wqe_ctrl_seg
*cseg
= &wqe
->ctrl
;
1391 struct mlx5_wqe_eth_seg
*eseg
= &wqe
->eth
;
1392 struct mlx5_wqe_data_seg
*dseg
;
1394 cseg
->qpn_ds
= cpu_to_be32((sq
->sqn
<< 8) | ds_cnt
);
1395 eseg
->inline_hdr
.sz
= cpu_to_be16(inline_hdr_sz
);
1397 dseg
= (struct mlx5_wqe_data_seg
*)cseg
+ (ds_cnt
- 1);
1398 dseg
->lkey
= sq
->mkey_be
;
1404 clear_bit(MLX5E_SQ_STATE_ENABLED
, &sq
->state
);
1405 mlx5e_free_xdpsq(sq
);
1410 static void mlx5e_close_xdpsq(struct mlx5e_xdpsq
*sq
)
1412 struct mlx5e_channel
*c
= sq
->channel
;
1414 clear_bit(MLX5E_SQ_STATE_ENABLED
, &sq
->state
);
1415 napi_synchronize(&c
->napi
);
1417 mlx5e_destroy_sq(c
->mdev
, sq
->sqn
);
1418 mlx5e_free_xdpsq_descs(sq
);
1419 mlx5e_free_xdpsq(sq
);
1422 static int mlx5e_alloc_cq_common(struct mlx5_core_dev
*mdev
,
1423 struct mlx5e_cq_param
*param
,
1424 struct mlx5e_cq
*cq
)
1426 struct mlx5_core_cq
*mcq
= &cq
->mcq
;
1432 err
= mlx5_cqwq_create(mdev
, ¶m
->wq
, param
->cqc
, &cq
->wq
,
1437 mlx5_vector2eqn(mdev
, param
->eq_ix
, &eqn_not_used
, &irqn
);
1440 mcq
->set_ci_db
= cq
->wq_ctrl
.db
.db
;
1441 mcq
->arm_db
= cq
->wq_ctrl
.db
.db
+ 1;
1442 *mcq
->set_ci_db
= 0;
1444 mcq
->vector
= param
->eq_ix
;
1445 mcq
->comp
= mlx5e_completion_event
;
1446 mcq
->event
= mlx5e_cq_error_event
;
1449 for (i
= 0; i
< mlx5_cqwq_get_size(&cq
->wq
); i
++) {
1450 struct mlx5_cqe64
*cqe
= mlx5_cqwq_get_wqe(&cq
->wq
, i
);
1460 static int mlx5e_alloc_cq(struct mlx5e_channel
*c
,
1461 struct mlx5e_cq_param
*param
,
1462 struct mlx5e_cq
*cq
)
1464 struct mlx5_core_dev
*mdev
= c
->priv
->mdev
;
1467 param
->wq
.buf_numa_node
= cpu_to_node(c
->cpu
);
1468 param
->wq
.db_numa_node
= cpu_to_node(c
->cpu
);
1469 param
->eq_ix
= c
->ix
;
1471 err
= mlx5e_alloc_cq_common(mdev
, param
, cq
);
1473 cq
->napi
= &c
->napi
;
1479 static void mlx5e_free_cq(struct mlx5e_cq
*cq
)
1481 mlx5_cqwq_destroy(&cq
->wq_ctrl
);
1484 static int mlx5e_create_cq(struct mlx5e_cq
*cq
, struct mlx5e_cq_param
*param
)
1486 struct mlx5_core_dev
*mdev
= cq
->mdev
;
1487 struct mlx5_core_cq
*mcq
= &cq
->mcq
;
1492 unsigned int irqn_not_used
;
1496 inlen
= MLX5_ST_SZ_BYTES(create_cq_in
) +
1497 sizeof(u64
) * cq
->wq_ctrl
.frag_buf
.npages
;
1498 in
= mlx5_vzalloc(inlen
);
1502 cqc
= MLX5_ADDR_OF(create_cq_in
, in
, cq_context
);
1504 memcpy(cqc
, param
->cqc
, sizeof(param
->cqc
));
1506 mlx5_fill_page_frag_array(&cq
->wq_ctrl
.frag_buf
,
1507 (__be64
*)MLX5_ADDR_OF(create_cq_in
, in
, pas
));
1509 mlx5_vector2eqn(mdev
, param
->eq_ix
, &eqn
, &irqn_not_used
);
1511 MLX5_SET(cqc
, cqc
, cq_period_mode
, param
->cq_period_mode
);
1512 MLX5_SET(cqc
, cqc
, c_eqn
, eqn
);
1513 MLX5_SET(cqc
, cqc
, uar_page
, mdev
->priv
.uar
->index
);
1514 MLX5_SET(cqc
, cqc
, log_page_size
, cq
->wq_ctrl
.frag_buf
.page_shift
-
1515 MLX5_ADAPTER_PAGE_SHIFT
);
1516 MLX5_SET64(cqc
, cqc
, dbr_addr
, cq
->wq_ctrl
.db
.dma
);
1518 err
= mlx5_core_create_cq(mdev
, mcq
, in
, inlen
);
1530 static void mlx5e_destroy_cq(struct mlx5e_cq
*cq
)
1532 mlx5_core_destroy_cq(cq
->mdev
, &cq
->mcq
);
1535 static int mlx5e_open_cq(struct mlx5e_channel
*c
,
1536 struct mlx5e_cq_moder moder
,
1537 struct mlx5e_cq_param
*param
,
1538 struct mlx5e_cq
*cq
)
1540 struct mlx5_core_dev
*mdev
= c
->mdev
;
1543 err
= mlx5e_alloc_cq(c
, param
, cq
);
1547 err
= mlx5e_create_cq(cq
, param
);
1551 if (MLX5_CAP_GEN(mdev
, cq_moderation
))
1552 mlx5_core_modify_cq_moderation(mdev
, &cq
->mcq
, moder
.usec
, moder
.pkts
);
1561 static void mlx5e_close_cq(struct mlx5e_cq
*cq
)
1563 mlx5e_destroy_cq(cq
);
1567 static int mlx5e_get_cpu(struct mlx5e_priv
*priv
, int ix
)
1569 return cpumask_first(priv
->mdev
->priv
.irq_info
[ix
].mask
);
1572 static int mlx5e_open_tx_cqs(struct mlx5e_channel
*c
,
1573 struct mlx5e_params
*params
,
1574 struct mlx5e_channel_param
*cparam
)
1579 for (tc
= 0; tc
< c
->num_tc
; tc
++) {
1580 err
= mlx5e_open_cq(c
, params
->tx_cq_moderation
,
1581 &cparam
->tx_cq
, &c
->sq
[tc
].cq
);
1583 goto err_close_tx_cqs
;
1589 for (tc
--; tc
>= 0; tc
--)
1590 mlx5e_close_cq(&c
->sq
[tc
].cq
);
1595 static void mlx5e_close_tx_cqs(struct mlx5e_channel
*c
)
1599 for (tc
= 0; tc
< c
->num_tc
; tc
++)
1600 mlx5e_close_cq(&c
->sq
[tc
].cq
);
1603 static int mlx5e_open_sqs(struct mlx5e_channel
*c
,
1604 struct mlx5e_params
*params
,
1605 struct mlx5e_channel_param
*cparam
)
1610 for (tc
= 0; tc
< params
->num_tc
; tc
++) {
1611 int txq_ix
= c
->ix
+ tc
* params
->num_channels
;
1613 err
= mlx5e_open_txqsq(c
, c
->priv
->tisn
[tc
], txq_ix
,
1614 params
, &cparam
->sq
, &c
->sq
[tc
]);
1622 for (tc
--; tc
>= 0; tc
--)
1623 mlx5e_close_txqsq(&c
->sq
[tc
]);
1628 static void mlx5e_close_sqs(struct mlx5e_channel
*c
)
1632 for (tc
= 0; tc
< c
->num_tc
; tc
++)
1633 mlx5e_close_txqsq(&c
->sq
[tc
]);
1636 static int mlx5e_set_sq_maxrate(struct net_device
*dev
,
1637 struct mlx5e_txqsq
*sq
, u32 rate
)
1639 struct mlx5e_priv
*priv
= netdev_priv(dev
);
1640 struct mlx5_core_dev
*mdev
= priv
->mdev
;
1641 struct mlx5e_modify_sq_param msp
= {0};
1645 if (rate
== sq
->rate_limit
)
1650 /* remove current rl index to free space to next ones */
1651 mlx5_rl_remove_rate(mdev
, sq
->rate_limit
);
1656 err
= mlx5_rl_add_rate(mdev
, rate
, &rl_index
);
1658 netdev_err(dev
, "Failed configuring rate %u: %d\n",
1664 msp
.curr_state
= MLX5_SQC_STATE_RDY
;
1665 msp
.next_state
= MLX5_SQC_STATE_RDY
;
1666 msp
.rl_index
= rl_index
;
1667 msp
.rl_update
= true;
1668 err
= mlx5e_modify_sq(mdev
, sq
->sqn
, &msp
);
1670 netdev_err(dev
, "Failed configuring rate %u: %d\n",
1672 /* remove the rate from the table */
1674 mlx5_rl_remove_rate(mdev
, rate
);
1678 sq
->rate_limit
= rate
;
1682 static int mlx5e_set_tx_maxrate(struct net_device
*dev
, int index
, u32 rate
)
1684 struct mlx5e_priv
*priv
= netdev_priv(dev
);
1685 struct mlx5_core_dev
*mdev
= priv
->mdev
;
1686 struct mlx5e_txqsq
*sq
= priv
->txq2sq
[index
];
1689 if (!mlx5_rl_is_supported(mdev
)) {
1690 netdev_err(dev
, "Rate limiting is not supported on this device\n");
1694 /* rate is given in Mb/sec, HW config is in Kb/sec */
1697 /* Check whether rate in valid range, 0 is always valid */
1698 if (rate
&& !mlx5_rl_is_in_range(mdev
, rate
)) {
1699 netdev_err(dev
, "TX rate %u, is not in range\n", rate
);
1703 mutex_lock(&priv
->state_lock
);
1704 if (test_bit(MLX5E_STATE_OPENED
, &priv
->state
))
1705 err
= mlx5e_set_sq_maxrate(dev
, sq
, rate
);
1707 priv
->tx_rates
[index
] = rate
;
1708 mutex_unlock(&priv
->state_lock
);
1713 static inline int mlx5e_get_max_num_channels(struct mlx5_core_dev
*mdev
)
1715 return is_kdump_kernel() ?
1716 MLX5E_MIN_NUM_CHANNELS
:
1717 min_t(int, mdev
->priv
.eq_table
.num_comp_vectors
,
1718 MLX5E_MAX_NUM_CHANNELS
);
1721 static int mlx5e_open_channel(struct mlx5e_priv
*priv
, int ix
,
1722 struct mlx5e_params
*params
,
1723 struct mlx5e_channel_param
*cparam
,
1724 struct mlx5e_channel
**cp
)
1726 struct mlx5e_cq_moder icocq_moder
= {0, 0};
1727 struct net_device
*netdev
= priv
->netdev
;
1728 int cpu
= mlx5e_get_cpu(priv
, ix
);
1729 struct mlx5e_channel
*c
;
1732 c
= kzalloc_node(sizeof(*c
), GFP_KERNEL
, cpu_to_node(cpu
));
1737 c
->mdev
= priv
->mdev
;
1738 c
->tstamp
= &priv
->tstamp
;
1741 c
->pdev
= &priv
->mdev
->pdev
->dev
;
1742 c
->netdev
= priv
->netdev
;
1743 c
->mkey_be
= cpu_to_be32(priv
->mdev
->mlx5e_res
.mkey
.key
);
1744 c
->num_tc
= params
->num_tc
;
1745 c
->xdp
= !!params
->xdp_prog
;
1747 netif_napi_add(netdev
, &c
->napi
, mlx5e_napi_poll
, 64);
1749 err
= mlx5e_open_cq(c
, icocq_moder
, &cparam
->icosq_cq
, &c
->icosq
.cq
);
1753 err
= mlx5e_open_tx_cqs(c
, params
, cparam
);
1755 goto err_close_icosq_cq
;
1757 err
= mlx5e_open_cq(c
, params
->rx_cq_moderation
, &cparam
->rx_cq
, &c
->rq
.cq
);
1759 goto err_close_tx_cqs
;
1761 /* XDP SQ CQ params are same as normal TXQ sq CQ params */
1762 err
= c
->xdp
? mlx5e_open_cq(c
, params
->tx_cq_moderation
,
1763 &cparam
->tx_cq
, &c
->rq
.xdpsq
.cq
) : 0;
1765 goto err_close_rx_cq
;
1767 napi_enable(&c
->napi
);
1769 err
= mlx5e_open_icosq(c
, params
, &cparam
->icosq
, &c
->icosq
);
1771 goto err_disable_napi
;
1773 err
= mlx5e_open_sqs(c
, params
, cparam
);
1775 goto err_close_icosq
;
1777 err
= c
->xdp
? mlx5e_open_xdpsq(c
, params
, &cparam
->xdp_sq
, &c
->rq
.xdpsq
) : 0;
1781 err
= mlx5e_open_rq(c
, params
, &cparam
->rq
, &c
->rq
);
1783 goto err_close_xdp_sq
;
1790 mlx5e_close_xdpsq(&c
->rq
.xdpsq
);
1796 mlx5e_close_icosq(&c
->icosq
);
1799 napi_disable(&c
->napi
);
1801 mlx5e_close_cq(&c
->rq
.xdpsq
.cq
);
1804 mlx5e_close_cq(&c
->rq
.cq
);
1807 mlx5e_close_tx_cqs(c
);
1810 mlx5e_close_cq(&c
->icosq
.cq
);
1813 netif_napi_del(&c
->napi
);
1819 static void mlx5e_activate_channel(struct mlx5e_channel
*c
)
1823 for (tc
= 0; tc
< c
->num_tc
; tc
++)
1824 mlx5e_activate_txqsq(&c
->sq
[tc
]);
1825 mlx5e_activate_rq(&c
->rq
);
1826 netif_set_xps_queue(c
->netdev
, get_cpu_mask(c
->cpu
), c
->ix
);
1829 static void mlx5e_deactivate_channel(struct mlx5e_channel
*c
)
1833 mlx5e_deactivate_rq(&c
->rq
);
1834 for (tc
= 0; tc
< c
->num_tc
; tc
++)
1835 mlx5e_deactivate_txqsq(&c
->sq
[tc
]);
1838 static void mlx5e_close_channel(struct mlx5e_channel
*c
)
1840 mlx5e_close_rq(&c
->rq
);
1842 mlx5e_close_xdpsq(&c
->rq
.xdpsq
);
1844 mlx5e_close_icosq(&c
->icosq
);
1845 napi_disable(&c
->napi
);
1847 mlx5e_close_cq(&c
->rq
.xdpsq
.cq
);
1848 mlx5e_close_cq(&c
->rq
.cq
);
1849 mlx5e_close_tx_cqs(c
);
1850 mlx5e_close_cq(&c
->icosq
.cq
);
1851 netif_napi_del(&c
->napi
);
1856 static void mlx5e_build_rq_param(struct mlx5e_priv
*priv
,
1857 struct mlx5e_params
*params
,
1858 struct mlx5e_rq_param
*param
)
1860 void *rqc
= param
->rqc
;
1861 void *wq
= MLX5_ADDR_OF(rqc
, rqc
, wq
);
1863 switch (params
->rq_wq_type
) {
1864 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ
:
1865 MLX5_SET(wq
, wq
, log_wqe_num_of_strides
, params
->mpwqe_log_num_strides
- 9);
1866 MLX5_SET(wq
, wq
, log_wqe_stride_size
, params
->mpwqe_log_stride_sz
- 6);
1867 MLX5_SET(wq
, wq
, wq_type
, MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ
);
1869 default: /* MLX5_WQ_TYPE_LINKED_LIST */
1870 MLX5_SET(wq
, wq
, wq_type
, MLX5_WQ_TYPE_LINKED_LIST
);
1873 MLX5_SET(wq
, wq
, end_padding_mode
, MLX5_WQ_END_PAD_MODE_ALIGN
);
1874 MLX5_SET(wq
, wq
, log_wq_stride
, ilog2(sizeof(struct mlx5e_rx_wqe
)));
1875 MLX5_SET(wq
, wq
, log_wq_sz
, params
->log_rq_size
);
1876 MLX5_SET(wq
, wq
, pd
, priv
->mdev
->mlx5e_res
.pdn
);
1877 MLX5_SET(rqc
, rqc
, counter_set_id
, priv
->q_counter
);
1878 MLX5_SET(rqc
, rqc
, vsd
, params
->vlan_strip_disable
);
1879 MLX5_SET(rqc
, rqc
, scatter_fcs
, params
->scatter_fcs_en
);
1881 param
->wq
.buf_numa_node
= dev_to_node(&priv
->mdev
->pdev
->dev
);
1882 param
->wq
.linear
= 1;
1885 static void mlx5e_build_drop_rq_param(struct mlx5e_rq_param
*param
)
1887 void *rqc
= param
->rqc
;
1888 void *wq
= MLX5_ADDR_OF(rqc
, rqc
, wq
);
1890 MLX5_SET(wq
, wq
, wq_type
, MLX5_WQ_TYPE_LINKED_LIST
);
1891 MLX5_SET(wq
, wq
, log_wq_stride
, ilog2(sizeof(struct mlx5e_rx_wqe
)));
1894 static void mlx5e_build_sq_param_common(struct mlx5e_priv
*priv
,
1895 struct mlx5e_sq_param
*param
)
1897 void *sqc
= param
->sqc
;
1898 void *wq
= MLX5_ADDR_OF(sqc
, sqc
, wq
);
1900 MLX5_SET(wq
, wq
, log_wq_stride
, ilog2(MLX5_SEND_WQE_BB
));
1901 MLX5_SET(wq
, wq
, pd
, priv
->mdev
->mlx5e_res
.pdn
);
1903 param
->wq
.buf_numa_node
= dev_to_node(&priv
->mdev
->pdev
->dev
);
1906 static void mlx5e_build_sq_param(struct mlx5e_priv
*priv
,
1907 struct mlx5e_params
*params
,
1908 struct mlx5e_sq_param
*param
)
1910 void *sqc
= param
->sqc
;
1911 void *wq
= MLX5_ADDR_OF(sqc
, sqc
, wq
);
1913 mlx5e_build_sq_param_common(priv
, param
);
1914 MLX5_SET(wq
, wq
, log_wq_sz
, params
->log_sq_size
);
1917 static void mlx5e_build_common_cq_param(struct mlx5e_priv
*priv
,
1918 struct mlx5e_cq_param
*param
)
1920 void *cqc
= param
->cqc
;
1922 MLX5_SET(cqc
, cqc
, uar_page
, priv
->mdev
->priv
.uar
->index
);
1925 static void mlx5e_build_rx_cq_param(struct mlx5e_priv
*priv
,
1926 struct mlx5e_params
*params
,
1927 struct mlx5e_cq_param
*param
)
1929 void *cqc
= param
->cqc
;
1932 switch (params
->rq_wq_type
) {
1933 case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ
:
1934 log_cq_size
= params
->log_rq_size
+ params
->mpwqe_log_num_strides
;
1936 default: /* MLX5_WQ_TYPE_LINKED_LIST */
1937 log_cq_size
= params
->log_rq_size
;
1940 MLX5_SET(cqc
, cqc
, log_cq_size
, log_cq_size
);
1941 if (MLX5E_GET_PFLAG(params
, MLX5E_PFLAG_RX_CQE_COMPRESS
)) {
1942 MLX5_SET(cqc
, cqc
, mini_cqe_res_format
, MLX5_CQE_FORMAT_CSUM
);
1943 MLX5_SET(cqc
, cqc
, cqe_comp_en
, 1);
1946 mlx5e_build_common_cq_param(priv
, param
);
1949 static void mlx5e_build_tx_cq_param(struct mlx5e_priv
*priv
,
1950 struct mlx5e_params
*params
,
1951 struct mlx5e_cq_param
*param
)
1953 void *cqc
= param
->cqc
;
1955 MLX5_SET(cqc
, cqc
, log_cq_size
, params
->log_sq_size
);
1957 mlx5e_build_common_cq_param(priv
, param
);
1959 param
->cq_period_mode
= MLX5_CQ_PERIOD_MODE_START_FROM_EQE
;
1962 static void mlx5e_build_ico_cq_param(struct mlx5e_priv
*priv
,
1964 struct mlx5e_cq_param
*param
)
1966 void *cqc
= param
->cqc
;
1968 MLX5_SET(cqc
, cqc
, log_cq_size
, log_wq_size
);
1970 mlx5e_build_common_cq_param(priv
, param
);
1972 param
->cq_period_mode
= MLX5_CQ_PERIOD_MODE_START_FROM_EQE
;
1975 static void mlx5e_build_icosq_param(struct mlx5e_priv
*priv
,
1977 struct mlx5e_sq_param
*param
)
1979 void *sqc
= param
->sqc
;
1980 void *wq
= MLX5_ADDR_OF(sqc
, sqc
, wq
);
1982 mlx5e_build_sq_param_common(priv
, param
);
1984 MLX5_SET(wq
, wq
, log_wq_sz
, log_wq_size
);
1985 MLX5_SET(sqc
, sqc
, reg_umr
, MLX5_CAP_ETH(priv
->mdev
, reg_umr_sq
));
1988 static void mlx5e_build_xdpsq_param(struct mlx5e_priv
*priv
,
1989 struct mlx5e_params
*params
,
1990 struct mlx5e_sq_param
*param
)
1992 void *sqc
= param
->sqc
;
1993 void *wq
= MLX5_ADDR_OF(sqc
, sqc
, wq
);
1995 mlx5e_build_sq_param_common(priv
, param
);
1996 MLX5_SET(wq
, wq
, log_wq_sz
, params
->log_sq_size
);
1999 static void mlx5e_build_channel_param(struct mlx5e_priv
*priv
,
2000 struct mlx5e_params
*params
,
2001 struct mlx5e_channel_param
*cparam
)
2003 u8 icosq_log_wq_sz
= MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE
;
2005 mlx5e_build_rq_param(priv
, params
, &cparam
->rq
);
2006 mlx5e_build_sq_param(priv
, params
, &cparam
->sq
);
2007 mlx5e_build_xdpsq_param(priv
, params
, &cparam
->xdp_sq
);
2008 mlx5e_build_icosq_param(priv
, icosq_log_wq_sz
, &cparam
->icosq
);
2009 mlx5e_build_rx_cq_param(priv
, params
, &cparam
->rx_cq
);
2010 mlx5e_build_tx_cq_param(priv
, params
, &cparam
->tx_cq
);
2011 mlx5e_build_ico_cq_param(priv
, icosq_log_wq_sz
, &cparam
->icosq_cq
);
2014 int mlx5e_open_channels(struct mlx5e_priv
*priv
,
2015 struct mlx5e_channels
*chs
)
2017 struct mlx5e_channel_param
*cparam
;
2021 chs
->num
= chs
->params
.num_channels
;
2023 chs
->c
= kcalloc(chs
->num
, sizeof(struct mlx5e_channel
*), GFP_KERNEL
);
2024 cparam
= kzalloc(sizeof(struct mlx5e_channel_param
), GFP_KERNEL
);
2025 if (!chs
->c
|| !cparam
)
2028 mlx5e_build_channel_param(priv
, &chs
->params
, cparam
);
2029 for (i
= 0; i
< chs
->num
; i
++) {
2030 err
= mlx5e_open_channel(priv
, i
, &chs
->params
, cparam
, &chs
->c
[i
]);
2032 goto err_close_channels
;
2039 for (i
--; i
>= 0; i
--)
2040 mlx5e_close_channel(chs
->c
[i
]);
2049 static void mlx5e_activate_channels(struct mlx5e_channels
*chs
)
2053 for (i
= 0; i
< chs
->num
; i
++)
2054 mlx5e_activate_channel(chs
->c
[i
]);
2057 static int mlx5e_wait_channels_min_rx_wqes(struct mlx5e_channels
*chs
)
2062 for (i
= 0; i
< chs
->num
; i
++) {
2063 err
= mlx5e_wait_for_min_rx_wqes(&chs
->c
[i
]->rq
);
2071 static void mlx5e_deactivate_channels(struct mlx5e_channels
*chs
)
2075 for (i
= 0; i
< chs
->num
; i
++)
2076 mlx5e_deactivate_channel(chs
->c
[i
]);
2079 void mlx5e_close_channels(struct mlx5e_channels
*chs
)
2083 for (i
= 0; i
< chs
->num
; i
++)
2084 mlx5e_close_channel(chs
->c
[i
]);
2091 mlx5e_create_rqt(struct mlx5e_priv
*priv
, int sz
, struct mlx5e_rqt
*rqt
)
2093 struct mlx5_core_dev
*mdev
= priv
->mdev
;
2100 inlen
= MLX5_ST_SZ_BYTES(create_rqt_in
) + sizeof(u32
) * sz
;
2101 in
= mlx5_vzalloc(inlen
);
2105 rqtc
= MLX5_ADDR_OF(create_rqt_in
, in
, rqt_context
);
2107 MLX5_SET(rqtc
, rqtc
, rqt_actual_size
, sz
);
2108 MLX5_SET(rqtc
, rqtc
, rqt_max_size
, sz
);
2110 for (i
= 0; i
< sz
; i
++)
2111 MLX5_SET(rqtc
, rqtc
, rq_num
[i
], priv
->drop_rq
.rqn
);
2113 err
= mlx5_core_create_rqt(mdev
, in
, inlen
, &rqt
->rqtn
);
2115 rqt
->enabled
= true;
2121 void mlx5e_destroy_rqt(struct mlx5e_priv
*priv
, struct mlx5e_rqt
*rqt
)
2123 rqt
->enabled
= false;
2124 mlx5_core_destroy_rqt(priv
->mdev
, rqt
->rqtn
);
2127 static int mlx5e_create_indirect_rqts(struct mlx5e_priv
*priv
)
2129 struct mlx5e_rqt
*rqt
= &priv
->indir_rqt
;
2131 return mlx5e_create_rqt(priv
, MLX5E_INDIR_RQT_SIZE
, rqt
);
2134 int mlx5e_create_direct_rqts(struct mlx5e_priv
*priv
)
2136 struct mlx5e_rqt
*rqt
;
2140 for (ix
= 0; ix
< priv
->profile
->max_nch(priv
->mdev
); ix
++) {
2141 rqt
= &priv
->direct_tir
[ix
].rqt
;
2142 err
= mlx5e_create_rqt(priv
, 1 /*size */, rqt
);
2144 goto err_destroy_rqts
;
2150 for (ix
--; ix
>= 0; ix
--)
2151 mlx5e_destroy_rqt(priv
, &priv
->direct_tir
[ix
].rqt
);
2156 static int mlx5e_rx_hash_fn(int hfunc
)
2158 return (hfunc
== ETH_RSS_HASH_TOP
) ?
2159 MLX5_RX_HASH_FN_TOEPLITZ
:
2160 MLX5_RX_HASH_FN_INVERTED_XOR8
;
2163 static int mlx5e_bits_invert(unsigned long a
, int size
)
2168 for (i
= 0; i
< size
; i
++)
2169 inv
|= (test_bit(size
- i
- 1, &a
) ? 1 : 0) << i
;
2174 static void mlx5e_fill_rqt_rqns(struct mlx5e_priv
*priv
, int sz
,
2175 struct mlx5e_redirect_rqt_param rrp
, void *rqtc
)
2179 for (i
= 0; i
< sz
; i
++) {
2185 if (rrp
.rss
.hfunc
== ETH_RSS_HASH_XOR
)
2186 ix
= mlx5e_bits_invert(i
, ilog2(sz
));
2188 ix
= priv
->channels
.params
.indirection_rqt
[ix
];
2189 rqn
= rrp
.rss
.channels
->c
[ix
]->rq
.rqn
;
2193 MLX5_SET(rqtc
, rqtc
, rq_num
[i
], rqn
);
2197 int mlx5e_redirect_rqt(struct mlx5e_priv
*priv
, u32 rqtn
, int sz
,
2198 struct mlx5e_redirect_rqt_param rrp
)
2200 struct mlx5_core_dev
*mdev
= priv
->mdev
;
2206 inlen
= MLX5_ST_SZ_BYTES(modify_rqt_in
) + sizeof(u32
) * sz
;
2207 in
= mlx5_vzalloc(inlen
);
2211 rqtc
= MLX5_ADDR_OF(modify_rqt_in
, in
, ctx
);
2213 MLX5_SET(rqtc
, rqtc
, rqt_actual_size
, sz
);
2214 MLX5_SET(modify_rqt_in
, in
, bitmask
.rqn_list
, 1);
2215 mlx5e_fill_rqt_rqns(priv
, sz
, rrp
, rqtc
);
2216 err
= mlx5_core_modify_rqt(mdev
, rqtn
, in
, inlen
);
2222 static u32
mlx5e_get_direct_rqn(struct mlx5e_priv
*priv
, int ix
,
2223 struct mlx5e_redirect_rqt_param rrp
)
2228 if (ix
>= rrp
.rss
.channels
->num
)
2229 return priv
->drop_rq
.rqn
;
2231 return rrp
.rss
.channels
->c
[ix
]->rq
.rqn
;
2234 static void mlx5e_redirect_rqts(struct mlx5e_priv
*priv
,
2235 struct mlx5e_redirect_rqt_param rrp
)
2240 if (priv
->indir_rqt
.enabled
) {
2242 rqtn
= priv
->indir_rqt
.rqtn
;
2243 mlx5e_redirect_rqt(priv
, rqtn
, MLX5E_INDIR_RQT_SIZE
, rrp
);
2246 for (ix
= 0; ix
< priv
->profile
->max_nch(priv
->mdev
); ix
++) {
2247 struct mlx5e_redirect_rqt_param direct_rrp
= {
2250 .rqn
= mlx5e_get_direct_rqn(priv
, ix
, rrp
)
2254 /* Direct RQ Tables */
2255 if (!priv
->direct_tir
[ix
].rqt
.enabled
)
2258 rqtn
= priv
->direct_tir
[ix
].rqt
.rqtn
;
2259 mlx5e_redirect_rqt(priv
, rqtn
, 1, direct_rrp
);
2263 static void mlx5e_redirect_rqts_to_channels(struct mlx5e_priv
*priv
,
2264 struct mlx5e_channels
*chs
)
2266 struct mlx5e_redirect_rqt_param rrp
= {
2271 .hfunc
= chs
->params
.rss_hfunc
,
2276 mlx5e_redirect_rqts(priv
, rrp
);
2279 static void mlx5e_redirect_rqts_to_drop(struct mlx5e_priv
*priv
)
2281 struct mlx5e_redirect_rqt_param drop_rrp
= {
2284 .rqn
= priv
->drop_rq
.rqn
,
2288 mlx5e_redirect_rqts(priv
, drop_rrp
);
2291 static void mlx5e_build_tir_ctx_lro(struct mlx5e_params
*params
, void *tirc
)
2293 if (!params
->lro_en
)
2296 #define ROUGH_MAX_L2_L3_HDR_SZ 256
2298 MLX5_SET(tirc
, tirc
, lro_enable_mask
,
2299 MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO
|
2300 MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO
);
2301 MLX5_SET(tirc
, tirc
, lro_max_ip_payload_size
,
2302 (params
->lro_wqe_sz
- ROUGH_MAX_L2_L3_HDR_SZ
) >> 8);
2303 MLX5_SET(tirc
, tirc
, lro_timeout_period_usecs
, params
->lro_timeout
);
2306 void mlx5e_build_indir_tir_ctx_hash(struct mlx5e_params
*params
,
2307 enum mlx5e_traffic_types tt
,
2310 void *hfso
= MLX5_ADDR_OF(tirc
, tirc
, rx_hash_field_selector_outer
);
2312 #define MLX5_HASH_IP (MLX5_HASH_FIELD_SEL_SRC_IP |\
2313 MLX5_HASH_FIELD_SEL_DST_IP)
2315 #define MLX5_HASH_IP_L4PORTS (MLX5_HASH_FIELD_SEL_SRC_IP |\
2316 MLX5_HASH_FIELD_SEL_DST_IP |\
2317 MLX5_HASH_FIELD_SEL_L4_SPORT |\
2318 MLX5_HASH_FIELD_SEL_L4_DPORT)
2320 #define MLX5_HASH_IP_IPSEC_SPI (MLX5_HASH_FIELD_SEL_SRC_IP |\
2321 MLX5_HASH_FIELD_SEL_DST_IP |\
2322 MLX5_HASH_FIELD_SEL_IPSEC_SPI)
2324 MLX5_SET(tirc
, tirc
, rx_hash_fn
, mlx5e_rx_hash_fn(params
->rss_hfunc
));
2325 if (params
->rss_hfunc
== ETH_RSS_HASH_TOP
) {
2326 void *rss_key
= MLX5_ADDR_OF(tirc
, tirc
,
2327 rx_hash_toeplitz_key
);
2328 size_t len
= MLX5_FLD_SZ_BYTES(tirc
,
2329 rx_hash_toeplitz_key
);
2331 MLX5_SET(tirc
, tirc
, rx_hash_symmetric
, 1);
2332 memcpy(rss_key
, params
->toeplitz_hash_key
, len
);
2336 case MLX5E_TT_IPV4_TCP
:
2337 MLX5_SET(rx_hash_field_select
, hfso
, l3_prot_type
,
2338 MLX5_L3_PROT_TYPE_IPV4
);
2339 MLX5_SET(rx_hash_field_select
, hfso
, l4_prot_type
,
2340 MLX5_L4_PROT_TYPE_TCP
);
2341 MLX5_SET(rx_hash_field_select
, hfso
, selected_fields
,
2342 MLX5_HASH_IP_L4PORTS
);
2345 case MLX5E_TT_IPV6_TCP
:
2346 MLX5_SET(rx_hash_field_select
, hfso
, l3_prot_type
,
2347 MLX5_L3_PROT_TYPE_IPV6
);
2348 MLX5_SET(rx_hash_field_select
, hfso
, l4_prot_type
,
2349 MLX5_L4_PROT_TYPE_TCP
);
2350 MLX5_SET(rx_hash_field_select
, hfso
, selected_fields
,
2351 MLX5_HASH_IP_L4PORTS
);
2354 case MLX5E_TT_IPV4_UDP
:
2355 MLX5_SET(rx_hash_field_select
, hfso
, l3_prot_type
,
2356 MLX5_L3_PROT_TYPE_IPV4
);
2357 MLX5_SET(rx_hash_field_select
, hfso
, l4_prot_type
,
2358 MLX5_L4_PROT_TYPE_UDP
);
2359 MLX5_SET(rx_hash_field_select
, hfso
, selected_fields
,
2360 MLX5_HASH_IP_L4PORTS
);
2363 case MLX5E_TT_IPV6_UDP
:
2364 MLX5_SET(rx_hash_field_select
, hfso
, l3_prot_type
,
2365 MLX5_L3_PROT_TYPE_IPV6
);
2366 MLX5_SET(rx_hash_field_select
, hfso
, l4_prot_type
,
2367 MLX5_L4_PROT_TYPE_UDP
);
2368 MLX5_SET(rx_hash_field_select
, hfso
, selected_fields
,
2369 MLX5_HASH_IP_L4PORTS
);
2372 case MLX5E_TT_IPV4_IPSEC_AH
:
2373 MLX5_SET(rx_hash_field_select
, hfso
, l3_prot_type
,
2374 MLX5_L3_PROT_TYPE_IPV4
);
2375 MLX5_SET(rx_hash_field_select
, hfso
, selected_fields
,
2376 MLX5_HASH_IP_IPSEC_SPI
);
2379 case MLX5E_TT_IPV6_IPSEC_AH
:
2380 MLX5_SET(rx_hash_field_select
, hfso
, l3_prot_type
,
2381 MLX5_L3_PROT_TYPE_IPV6
);
2382 MLX5_SET(rx_hash_field_select
, hfso
, selected_fields
,
2383 MLX5_HASH_IP_IPSEC_SPI
);
2386 case MLX5E_TT_IPV4_IPSEC_ESP
:
2387 MLX5_SET(rx_hash_field_select
, hfso
, l3_prot_type
,
2388 MLX5_L3_PROT_TYPE_IPV4
);
2389 MLX5_SET(rx_hash_field_select
, hfso
, selected_fields
,
2390 MLX5_HASH_IP_IPSEC_SPI
);
2393 case MLX5E_TT_IPV6_IPSEC_ESP
:
2394 MLX5_SET(rx_hash_field_select
, hfso
, l3_prot_type
,
2395 MLX5_L3_PROT_TYPE_IPV6
);
2396 MLX5_SET(rx_hash_field_select
, hfso
, selected_fields
,
2397 MLX5_HASH_IP_IPSEC_SPI
);
2401 MLX5_SET(rx_hash_field_select
, hfso
, l3_prot_type
,
2402 MLX5_L3_PROT_TYPE_IPV4
);
2403 MLX5_SET(rx_hash_field_select
, hfso
, selected_fields
,
2408 MLX5_SET(rx_hash_field_select
, hfso
, l3_prot_type
,
2409 MLX5_L3_PROT_TYPE_IPV6
);
2410 MLX5_SET(rx_hash_field_select
, hfso
, selected_fields
,
2414 WARN_ONCE(true, "%s: bad traffic type!\n", __func__
);
2418 static int mlx5e_modify_tirs_lro(struct mlx5e_priv
*priv
)
2420 struct mlx5_core_dev
*mdev
= priv
->mdev
;
2429 inlen
= MLX5_ST_SZ_BYTES(modify_tir_in
);
2430 in
= mlx5_vzalloc(inlen
);
2434 MLX5_SET(modify_tir_in
, in
, bitmask
.lro
, 1);
2435 tirc
= MLX5_ADDR_OF(modify_tir_in
, in
, ctx
);
2437 mlx5e_build_tir_ctx_lro(&priv
->channels
.params
, tirc
);
2439 for (tt
= 0; tt
< MLX5E_NUM_INDIR_TIRS
; tt
++) {
2440 err
= mlx5_core_modify_tir(mdev
, priv
->indir_tir
[tt
].tirn
, in
,
2446 for (ix
= 0; ix
< priv
->profile
->max_nch(priv
->mdev
); ix
++) {
2447 err
= mlx5_core_modify_tir(mdev
, priv
->direct_tir
[ix
].tirn
,
2459 static int mlx5e_set_mtu(struct mlx5e_priv
*priv
, u16 mtu
)
2461 struct mlx5_core_dev
*mdev
= priv
->mdev
;
2462 u16 hw_mtu
= MLX5E_SW2HW_MTU(mtu
);
2465 err
= mlx5_set_port_mtu(mdev
, hw_mtu
, 1);
2469 /* Update vport context MTU */
2470 mlx5_modify_nic_vport_mtu(mdev
, hw_mtu
);
2474 static void mlx5e_query_mtu(struct mlx5e_priv
*priv
, u16
*mtu
)
2476 struct mlx5_core_dev
*mdev
= priv
->mdev
;
2480 err
= mlx5_query_nic_vport_mtu(mdev
, &hw_mtu
);
2481 if (err
|| !hw_mtu
) /* fallback to port oper mtu */
2482 mlx5_query_port_oper_mtu(mdev
, &hw_mtu
, 1);
2484 *mtu
= MLX5E_HW2SW_MTU(hw_mtu
);
2487 static int mlx5e_set_dev_port_mtu(struct mlx5e_priv
*priv
)
2489 struct net_device
*netdev
= priv
->netdev
;
2493 err
= mlx5e_set_mtu(priv
, netdev
->mtu
);
2497 mlx5e_query_mtu(priv
, &mtu
);
2498 if (mtu
!= netdev
->mtu
)
2499 netdev_warn(netdev
, "%s: VPort MTU %d is different than netdev mtu %d\n",
2500 __func__
, mtu
, netdev
->mtu
);
2506 static void mlx5e_netdev_set_tcs(struct net_device
*netdev
)
2508 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
2509 int nch
= priv
->channels
.params
.num_channels
;
2510 int ntc
= priv
->channels
.params
.num_tc
;
2513 netdev_reset_tc(netdev
);
2518 netdev_set_num_tc(netdev
, ntc
);
2520 /* Map netdev TCs to offset 0
2521 * We have our own UP to TXQ mapping for QoS
2523 for (tc
= 0; tc
< ntc
; tc
++)
2524 netdev_set_tc_queue(netdev
, tc
, nch
, 0);
2527 static void mlx5e_build_channels_tx_maps(struct mlx5e_priv
*priv
)
2529 struct mlx5e_channel
*c
;
2530 struct mlx5e_txqsq
*sq
;
2533 for (i
= 0; i
< priv
->channels
.num
; i
++)
2534 for (tc
= 0; tc
< priv
->profile
->max_tc
; tc
++)
2535 priv
->channel_tc2txq
[i
][tc
] = i
+ tc
* priv
->channels
.num
;
2537 for (i
= 0; i
< priv
->channels
.num
; i
++) {
2538 c
= priv
->channels
.c
[i
];
2539 for (tc
= 0; tc
< c
->num_tc
; tc
++) {
2541 priv
->txq2sq
[sq
->txq_ix
] = sq
;
2546 static void mlx5e_activate_priv_channels(struct mlx5e_priv
*priv
)
2548 int num_txqs
= priv
->channels
.num
* priv
->channels
.params
.num_tc
;
2549 struct net_device
*netdev
= priv
->netdev
;
2551 mlx5e_netdev_set_tcs(netdev
);
2552 netif_set_real_num_tx_queues(netdev
, num_txqs
);
2553 netif_set_real_num_rx_queues(netdev
, priv
->channels
.num
);
2555 mlx5e_build_channels_tx_maps(priv
);
2556 mlx5e_activate_channels(&priv
->channels
);
2557 netif_tx_start_all_queues(priv
->netdev
);
2559 if (MLX5_CAP_GEN(priv
->mdev
, vport_group_manager
))
2560 mlx5e_add_sqs_fwd_rules(priv
);
2562 mlx5e_wait_channels_min_rx_wqes(&priv
->channels
);
2563 mlx5e_redirect_rqts_to_channels(priv
, &priv
->channels
);
2566 static void mlx5e_deactivate_priv_channels(struct mlx5e_priv
*priv
)
2568 mlx5e_redirect_rqts_to_drop(priv
);
2570 if (MLX5_CAP_GEN(priv
->mdev
, vport_group_manager
))
2571 mlx5e_remove_sqs_fwd_rules(priv
);
2573 /* FIXME: This is a W/A only for tx timeout watch dog false alarm when
2574 * polling for inactive tx queues.
2576 netif_tx_stop_all_queues(priv
->netdev
);
2577 netif_tx_disable(priv
->netdev
);
2578 mlx5e_deactivate_channels(&priv
->channels
);
2581 void mlx5e_switch_priv_channels(struct mlx5e_priv
*priv
,
2582 struct mlx5e_channels
*new_chs
,
2583 mlx5e_fp_hw_modify hw_modify
)
2585 struct net_device
*netdev
= priv
->netdev
;
2588 new_num_txqs
= new_chs
->num
* new_chs
->params
.num_tc
;
2590 netif_carrier_off(netdev
);
2592 if (new_num_txqs
< netdev
->real_num_tx_queues
)
2593 netif_set_real_num_tx_queues(netdev
, new_num_txqs
);
2595 mlx5e_deactivate_priv_channels(priv
);
2596 mlx5e_close_channels(&priv
->channels
);
2598 priv
->channels
= *new_chs
;
2600 /* New channels are ready to roll, modify HW settings if needed */
2604 mlx5e_refresh_tirs(priv
, false);
2605 mlx5e_activate_priv_channels(priv
);
2607 mlx5e_update_carrier(priv
);
2610 int mlx5e_open_locked(struct net_device
*netdev
)
2612 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
2615 set_bit(MLX5E_STATE_OPENED
, &priv
->state
);
2617 err
= mlx5e_open_channels(priv
, &priv
->channels
);
2619 goto err_clear_state_opened_flag
;
2621 mlx5e_refresh_tirs(priv
, false);
2622 mlx5e_activate_priv_channels(priv
);
2623 mlx5e_update_carrier(priv
);
2624 mlx5e_timestamp_init(priv
);
2626 if (priv
->profile
->update_stats
)
2627 queue_delayed_work(priv
->wq
, &priv
->update_stats_work
, 0);
2631 err_clear_state_opened_flag
:
2632 clear_bit(MLX5E_STATE_OPENED
, &priv
->state
);
2636 int mlx5e_open(struct net_device
*netdev
)
2638 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
2641 mutex_lock(&priv
->state_lock
);
2642 err
= mlx5e_open_locked(netdev
);
2643 mutex_unlock(&priv
->state_lock
);
2648 int mlx5e_close_locked(struct net_device
*netdev
)
2650 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
2652 /* May already be CLOSED in case a previous configuration operation
2653 * (e.g RX/TX queue size change) that involves close&open failed.
2655 if (!test_bit(MLX5E_STATE_OPENED
, &priv
->state
))
2658 clear_bit(MLX5E_STATE_OPENED
, &priv
->state
);
2660 mlx5e_timestamp_cleanup(priv
);
2661 netif_carrier_off(priv
->netdev
);
2662 mlx5e_deactivate_priv_channels(priv
);
2663 mlx5e_close_channels(&priv
->channels
);
2668 int mlx5e_close(struct net_device
*netdev
)
2670 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
2673 if (!netif_device_present(netdev
))
2676 mutex_lock(&priv
->state_lock
);
2677 err
= mlx5e_close_locked(netdev
);
2678 mutex_unlock(&priv
->state_lock
);
2683 static int mlx5e_alloc_drop_rq(struct mlx5_core_dev
*mdev
,
2684 struct mlx5e_rq
*rq
,
2685 struct mlx5e_rq_param
*param
)
2687 void *rqc
= param
->rqc
;
2688 void *rqc_wq
= MLX5_ADDR_OF(rqc
, rqc
, wq
);
2691 param
->wq
.db_numa_node
= param
->wq
.buf_numa_node
;
2693 err
= mlx5_wq_ll_create(mdev
, ¶m
->wq
, rqc_wq
, &rq
->wq
,
2703 static int mlx5e_alloc_drop_cq(struct mlx5_core_dev
*mdev
,
2704 struct mlx5e_cq
*cq
,
2705 struct mlx5e_cq_param
*param
)
2707 return mlx5e_alloc_cq_common(mdev
, param
, cq
);
2710 static int mlx5e_open_drop_rq(struct mlx5_core_dev
*mdev
,
2711 struct mlx5e_rq
*drop_rq
)
2713 struct mlx5e_cq_param cq_param
= {};
2714 struct mlx5e_rq_param rq_param
= {};
2715 struct mlx5e_cq
*cq
= &drop_rq
->cq
;
2718 mlx5e_build_drop_rq_param(&rq_param
);
2720 err
= mlx5e_alloc_drop_cq(mdev
, cq
, &cq_param
);
2724 err
= mlx5e_create_cq(cq
, &cq_param
);
2728 err
= mlx5e_alloc_drop_rq(mdev
, drop_rq
, &rq_param
);
2730 goto err_destroy_cq
;
2732 err
= mlx5e_create_rq(drop_rq
, &rq_param
);
2739 mlx5e_free_rq(drop_rq
);
2742 mlx5e_destroy_cq(cq
);
2750 static void mlx5e_close_drop_rq(struct mlx5e_rq
*drop_rq
)
2752 mlx5e_destroy_rq(drop_rq
);
2753 mlx5e_free_rq(drop_rq
);
2754 mlx5e_destroy_cq(&drop_rq
->cq
);
2755 mlx5e_free_cq(&drop_rq
->cq
);
2758 static int mlx5e_create_tis(struct mlx5e_priv
*priv
, int tc
)
2760 struct mlx5_core_dev
*mdev
= priv
->mdev
;
2761 u32 in
[MLX5_ST_SZ_DW(create_tis_in
)] = {0};
2762 void *tisc
= MLX5_ADDR_OF(create_tis_in
, in
, ctx
);
2764 MLX5_SET(tisc
, tisc
, prio
, tc
<< 1);
2765 MLX5_SET(tisc
, tisc
, transport_domain
, mdev
->mlx5e_res
.td
.tdn
);
2767 if (mlx5_lag_is_lacp_owner(mdev
))
2768 MLX5_SET(tisc
, tisc
, strict_lag_tx_port_affinity
, 1);
2770 return mlx5_core_create_tis(mdev
, in
, sizeof(in
), &priv
->tisn
[tc
]);
2773 static void mlx5e_destroy_tis(struct mlx5e_priv
*priv
, int tc
)
2775 mlx5_core_destroy_tis(priv
->mdev
, priv
->tisn
[tc
]);
2778 int mlx5e_create_tises(struct mlx5e_priv
*priv
)
2783 for (tc
= 0; tc
< priv
->profile
->max_tc
; tc
++) {
2784 err
= mlx5e_create_tis(priv
, tc
);
2786 goto err_close_tises
;
2792 for (tc
--; tc
>= 0; tc
--)
2793 mlx5e_destroy_tis(priv
, tc
);
2798 void mlx5e_cleanup_nic_tx(struct mlx5e_priv
*priv
)
2802 for (tc
= 0; tc
< priv
->profile
->max_tc
; tc
++)
2803 mlx5e_destroy_tis(priv
, tc
);
2806 static void mlx5e_build_indir_tir_ctx(struct mlx5e_priv
*priv
,
2807 enum mlx5e_traffic_types tt
,
2810 MLX5_SET(tirc
, tirc
, transport_domain
, priv
->mdev
->mlx5e_res
.td
.tdn
);
2812 mlx5e_build_tir_ctx_lro(&priv
->channels
.params
, tirc
);
2814 MLX5_SET(tirc
, tirc
, disp_type
, MLX5_TIRC_DISP_TYPE_INDIRECT
);
2815 MLX5_SET(tirc
, tirc
, indirect_table
, priv
->indir_rqt
.rqtn
);
2816 mlx5e_build_indir_tir_ctx_hash(&priv
->channels
.params
, tt
, tirc
);
2819 static void mlx5e_build_direct_tir_ctx(struct mlx5e_priv
*priv
, u32 rqtn
, u32
*tirc
)
2821 MLX5_SET(tirc
, tirc
, transport_domain
, priv
->mdev
->mlx5e_res
.td
.tdn
);
2823 mlx5e_build_tir_ctx_lro(&priv
->channels
.params
, tirc
);
2825 MLX5_SET(tirc
, tirc
, disp_type
, MLX5_TIRC_DISP_TYPE_INDIRECT
);
2826 MLX5_SET(tirc
, tirc
, indirect_table
, rqtn
);
2827 MLX5_SET(tirc
, tirc
, rx_hash_fn
, MLX5_RX_HASH_FN_INVERTED_XOR8
);
2830 static int mlx5e_create_indirect_tirs(struct mlx5e_priv
*priv
)
2832 struct mlx5e_tir
*tir
;
2839 inlen
= MLX5_ST_SZ_BYTES(create_tir_in
);
2840 in
= mlx5_vzalloc(inlen
);
2844 for (tt
= 0; tt
< MLX5E_NUM_INDIR_TIRS
; tt
++) {
2845 memset(in
, 0, inlen
);
2846 tir
= &priv
->indir_tir
[tt
];
2847 tirc
= MLX5_ADDR_OF(create_tir_in
, in
, ctx
);
2848 mlx5e_build_indir_tir_ctx(priv
, tt
, tirc
);
2849 err
= mlx5e_create_tir(priv
->mdev
, tir
, in
, inlen
);
2851 goto err_destroy_tirs
;
2859 for (tt
--; tt
>= 0; tt
--)
2860 mlx5e_destroy_tir(priv
->mdev
, &priv
->indir_tir
[tt
]);
2867 int mlx5e_create_direct_tirs(struct mlx5e_priv
*priv
)
2869 int nch
= priv
->profile
->max_nch(priv
->mdev
);
2870 struct mlx5e_tir
*tir
;
2877 inlen
= MLX5_ST_SZ_BYTES(create_tir_in
);
2878 in
= mlx5_vzalloc(inlen
);
2882 for (ix
= 0; ix
< nch
; ix
++) {
2883 memset(in
, 0, inlen
);
2884 tir
= &priv
->direct_tir
[ix
];
2885 tirc
= MLX5_ADDR_OF(create_tir_in
, in
, ctx
);
2886 mlx5e_build_direct_tir_ctx(priv
, priv
->direct_tir
[ix
].rqt
.rqtn
, tirc
);
2887 err
= mlx5e_create_tir(priv
->mdev
, tir
, in
, inlen
);
2889 goto err_destroy_ch_tirs
;
2896 err_destroy_ch_tirs
:
2897 for (ix
--; ix
>= 0; ix
--)
2898 mlx5e_destroy_tir(priv
->mdev
, &priv
->direct_tir
[ix
]);
2905 static void mlx5e_destroy_indirect_tirs(struct mlx5e_priv
*priv
)
2909 for (i
= 0; i
< MLX5E_NUM_INDIR_TIRS
; i
++)
2910 mlx5e_destroy_tir(priv
->mdev
, &priv
->indir_tir
[i
]);
2913 void mlx5e_destroy_direct_tirs(struct mlx5e_priv
*priv
)
2915 int nch
= priv
->profile
->max_nch(priv
->mdev
);
2918 for (i
= 0; i
< nch
; i
++)
2919 mlx5e_destroy_tir(priv
->mdev
, &priv
->direct_tir
[i
]);
2922 static int mlx5e_modify_channels_scatter_fcs(struct mlx5e_channels
*chs
, bool enable
)
2927 for (i
= 0; i
< chs
->num
; i
++) {
2928 err
= mlx5e_modify_rq_scatter_fcs(&chs
->c
[i
]->rq
, enable
);
2936 static int mlx5e_modify_channels_vsd(struct mlx5e_channels
*chs
, bool vsd
)
2941 for (i
= 0; i
< chs
->num
; i
++) {
2942 err
= mlx5e_modify_rq_vsd(&chs
->c
[i
]->rq
, vsd
);
2950 static int mlx5e_setup_tc(struct net_device
*netdev
, u8 tc
)
2952 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
2953 struct mlx5e_channels new_channels
= {};
2956 if (tc
&& tc
!= MLX5E_MAX_NUM_TC
)
2959 mutex_lock(&priv
->state_lock
);
2961 new_channels
.params
= priv
->channels
.params
;
2962 new_channels
.params
.num_tc
= tc
? tc
: 1;
2964 if (test_bit(MLX5E_STATE_OPENED
, &priv
->state
)) {
2965 priv
->channels
.params
= new_channels
.params
;
2969 err
= mlx5e_open_channels(priv
, &new_channels
);
2973 mlx5e_switch_priv_channels(priv
, &new_channels
, NULL
);
2975 mutex_unlock(&priv
->state_lock
);
2979 static int mlx5e_ndo_setup_tc(struct net_device
*dev
, u32 handle
,
2980 __be16 proto
, struct tc_to_netdev
*tc
)
2982 struct mlx5e_priv
*priv
= netdev_priv(dev
);
2984 if (TC_H_MAJ(handle
) != TC_H_MAJ(TC_H_INGRESS
))
2988 case TC_SETUP_CLSFLOWER
:
2989 switch (tc
->cls_flower
->command
) {
2990 case TC_CLSFLOWER_REPLACE
:
2991 return mlx5e_configure_flower(priv
, proto
, tc
->cls_flower
);
2992 case TC_CLSFLOWER_DESTROY
:
2993 return mlx5e_delete_flower(priv
, tc
->cls_flower
);
2994 case TC_CLSFLOWER_STATS
:
2995 return mlx5e_stats_flower(priv
, tc
->cls_flower
);
3002 if (tc
->type
!= TC_SETUP_MQPRIO
)
3005 tc
->mqprio
->hw
= TC_MQPRIO_HW_OFFLOAD_TCS
;
3007 return mlx5e_setup_tc(dev
, tc
->mqprio
->num_tc
);
3011 mlx5e_get_stats(struct net_device
*dev
, struct rtnl_link_stats64
*stats
)
3013 struct mlx5e_priv
*priv
= netdev_priv(dev
);
3014 struct mlx5e_sw_stats
*sstats
= &priv
->stats
.sw
;
3015 struct mlx5e_vport_stats
*vstats
= &priv
->stats
.vport
;
3016 struct mlx5e_pport_stats
*pstats
= &priv
->stats
.pport
;
3018 if (mlx5e_is_uplink_rep(priv
)) {
3019 stats
->rx_packets
= PPORT_802_3_GET(pstats
, a_frames_received_ok
);
3020 stats
->rx_bytes
= PPORT_802_3_GET(pstats
, a_octets_received_ok
);
3021 stats
->tx_packets
= PPORT_802_3_GET(pstats
, a_frames_transmitted_ok
);
3022 stats
->tx_bytes
= PPORT_802_3_GET(pstats
, a_octets_transmitted_ok
);
3024 stats
->rx_packets
= sstats
->rx_packets
;
3025 stats
->rx_bytes
= sstats
->rx_bytes
;
3026 stats
->tx_packets
= sstats
->tx_packets
;
3027 stats
->tx_bytes
= sstats
->tx_bytes
;
3028 stats
->tx_dropped
= sstats
->tx_queue_dropped
;
3031 stats
->rx_dropped
= priv
->stats
.qcnt
.rx_out_of_buffer
;
3033 stats
->rx_length_errors
=
3034 PPORT_802_3_GET(pstats
, a_in_range_length_errors
) +
3035 PPORT_802_3_GET(pstats
, a_out_of_range_length_field
) +
3036 PPORT_802_3_GET(pstats
, a_frame_too_long_errors
);
3037 stats
->rx_crc_errors
=
3038 PPORT_802_3_GET(pstats
, a_frame_check_sequence_errors
);
3039 stats
->rx_frame_errors
= PPORT_802_3_GET(pstats
, a_alignment_errors
);
3040 stats
->tx_aborted_errors
= PPORT_2863_GET(pstats
, if_out_discards
);
3041 stats
->tx_carrier_errors
=
3042 PPORT_802_3_GET(pstats
, a_symbol_error_during_carrier
);
3043 stats
->rx_errors
= stats
->rx_length_errors
+ stats
->rx_crc_errors
+
3044 stats
->rx_frame_errors
;
3045 stats
->tx_errors
= stats
->tx_aborted_errors
+ stats
->tx_carrier_errors
;
3047 /* vport multicast also counts packets that are dropped due to steering
3048 * or rx out of buffer
3051 VPORT_COUNTER_GET(vstats
, received_eth_multicast
.packets
);
3055 static void mlx5e_set_rx_mode(struct net_device
*dev
)
3057 struct mlx5e_priv
*priv
= netdev_priv(dev
);
3059 queue_work(priv
->wq
, &priv
->set_rx_mode_work
);
3062 static int mlx5e_set_mac(struct net_device
*netdev
, void *addr
)
3064 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3065 struct sockaddr
*saddr
= addr
;
3067 if (!is_valid_ether_addr(saddr
->sa_data
))
3068 return -EADDRNOTAVAIL
;
3070 netif_addr_lock_bh(netdev
);
3071 ether_addr_copy(netdev
->dev_addr
, saddr
->sa_data
);
3072 netif_addr_unlock_bh(netdev
);
3074 queue_work(priv
->wq
, &priv
->set_rx_mode_work
);
3079 #define MLX5E_SET_FEATURE(netdev, feature, enable) \
3082 netdev->features |= feature; \
3084 netdev->features &= ~feature; \
3087 typedef int (*mlx5e_feature_handler
)(struct net_device
*netdev
, bool enable
);
3089 static int set_feature_lro(struct net_device
*netdev
, bool enable
)
3091 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3092 struct mlx5e_channels new_channels
= {};
3096 mutex_lock(&priv
->state_lock
);
3098 reset
= (priv
->channels
.params
.rq_wq_type
== MLX5_WQ_TYPE_LINKED_LIST
);
3099 reset
= reset
&& test_bit(MLX5E_STATE_OPENED
, &priv
->state
);
3101 new_channels
.params
= priv
->channels
.params
;
3102 new_channels
.params
.lro_en
= enable
;
3105 priv
->channels
.params
= new_channels
.params
;
3106 err
= mlx5e_modify_tirs_lro(priv
);
3110 err
= mlx5e_open_channels(priv
, &new_channels
);
3114 mlx5e_switch_priv_channels(priv
, &new_channels
, mlx5e_modify_tirs_lro
);
3116 mutex_unlock(&priv
->state_lock
);
3120 static int set_feature_vlan_filter(struct net_device
*netdev
, bool enable
)
3122 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3125 mlx5e_enable_vlan_filter(priv
);
3127 mlx5e_disable_vlan_filter(priv
);
3132 static int set_feature_tc_num_filters(struct net_device
*netdev
, bool enable
)
3134 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3136 if (!enable
&& mlx5e_tc_num_filters(priv
)) {
3138 "Active offloaded tc filters, can't turn hw_tc_offload off\n");
3145 static int set_feature_rx_all(struct net_device
*netdev
, bool enable
)
3147 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3148 struct mlx5_core_dev
*mdev
= priv
->mdev
;
3150 return mlx5_set_port_fcs(mdev
, !enable
);
3153 static int set_feature_rx_fcs(struct net_device
*netdev
, bool enable
)
3155 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3158 mutex_lock(&priv
->state_lock
);
3160 priv
->channels
.params
.scatter_fcs_en
= enable
;
3161 err
= mlx5e_modify_channels_scatter_fcs(&priv
->channels
, enable
);
3163 priv
->channels
.params
.scatter_fcs_en
= !enable
;
3165 mutex_unlock(&priv
->state_lock
);
3170 static int set_feature_rx_vlan(struct net_device
*netdev
, bool enable
)
3172 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3175 mutex_lock(&priv
->state_lock
);
3177 priv
->channels
.params
.vlan_strip_disable
= !enable
;
3178 if (!test_bit(MLX5E_STATE_OPENED
, &priv
->state
))
3181 err
= mlx5e_modify_channels_vsd(&priv
->channels
, !enable
);
3183 priv
->channels
.params
.vlan_strip_disable
= enable
;
3186 mutex_unlock(&priv
->state_lock
);
3191 #ifdef CONFIG_RFS_ACCEL
3192 static int set_feature_arfs(struct net_device
*netdev
, bool enable
)
3194 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3198 err
= mlx5e_arfs_enable(priv
);
3200 err
= mlx5e_arfs_disable(priv
);
3206 static int mlx5e_handle_feature(struct net_device
*netdev
,
3207 netdev_features_t wanted_features
,
3208 netdev_features_t feature
,
3209 mlx5e_feature_handler feature_handler
)
3211 netdev_features_t changes
= wanted_features
^ netdev
->features
;
3212 bool enable
= !!(wanted_features
& feature
);
3215 if (!(changes
& feature
))
3218 err
= feature_handler(netdev
, enable
);
3220 netdev_err(netdev
, "%s feature 0x%llx failed err %d\n",
3221 enable
? "Enable" : "Disable", feature
, err
);
3225 MLX5E_SET_FEATURE(netdev
, feature
, enable
);
3229 static int mlx5e_set_features(struct net_device
*netdev
,
3230 netdev_features_t features
)
3234 err
= mlx5e_handle_feature(netdev
, features
, NETIF_F_LRO
,
3236 err
|= mlx5e_handle_feature(netdev
, features
,
3237 NETIF_F_HW_VLAN_CTAG_FILTER
,
3238 set_feature_vlan_filter
);
3239 err
|= mlx5e_handle_feature(netdev
, features
, NETIF_F_HW_TC
,
3240 set_feature_tc_num_filters
);
3241 err
|= mlx5e_handle_feature(netdev
, features
, NETIF_F_RXALL
,
3242 set_feature_rx_all
);
3243 err
|= mlx5e_handle_feature(netdev
, features
, NETIF_F_RXFCS
,
3244 set_feature_rx_fcs
);
3245 err
|= mlx5e_handle_feature(netdev
, features
, NETIF_F_HW_VLAN_CTAG_RX
,
3246 set_feature_rx_vlan
);
3247 #ifdef CONFIG_RFS_ACCEL
3248 err
|= mlx5e_handle_feature(netdev
, features
, NETIF_F_NTUPLE
,
3252 return err
? -EINVAL
: 0;
3255 static int mlx5e_change_mtu(struct net_device
*netdev
, int new_mtu
)
3257 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3258 struct mlx5e_channels new_channels
= {};
3263 mutex_lock(&priv
->state_lock
);
3265 reset
= !priv
->channels
.params
.lro_en
&&
3266 (priv
->channels
.params
.rq_wq_type
!=
3267 MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ
);
3269 reset
= reset
&& test_bit(MLX5E_STATE_OPENED
, &priv
->state
);
3271 curr_mtu
= netdev
->mtu
;
3272 netdev
->mtu
= new_mtu
;
3275 mlx5e_set_dev_port_mtu(priv
);
3279 new_channels
.params
= priv
->channels
.params
;
3280 err
= mlx5e_open_channels(priv
, &new_channels
);
3282 netdev
->mtu
= curr_mtu
;
3286 mlx5e_switch_priv_channels(priv
, &new_channels
, mlx5e_set_dev_port_mtu
);
3289 mutex_unlock(&priv
->state_lock
);
3293 static int mlx5e_ioctl(struct net_device
*dev
, struct ifreq
*ifr
, int cmd
)
3297 return mlx5e_hwstamp_set(dev
, ifr
);
3299 return mlx5e_hwstamp_get(dev
, ifr
);
3305 static int mlx5e_set_vf_mac(struct net_device
*dev
, int vf
, u8
*mac
)
3307 struct mlx5e_priv
*priv
= netdev_priv(dev
);
3308 struct mlx5_core_dev
*mdev
= priv
->mdev
;
3310 return mlx5_eswitch_set_vport_mac(mdev
->priv
.eswitch
, vf
+ 1, mac
);
3313 static int mlx5e_set_vf_vlan(struct net_device
*dev
, int vf
, u16 vlan
, u8 qos
,
3316 struct mlx5e_priv
*priv
= netdev_priv(dev
);
3317 struct mlx5_core_dev
*mdev
= priv
->mdev
;
3319 if (vlan_proto
!= htons(ETH_P_8021Q
))
3320 return -EPROTONOSUPPORT
;
3322 return mlx5_eswitch_set_vport_vlan(mdev
->priv
.eswitch
, vf
+ 1,
3326 static int mlx5e_set_vf_spoofchk(struct net_device
*dev
, int vf
, bool setting
)
3328 struct mlx5e_priv
*priv
= netdev_priv(dev
);
3329 struct mlx5_core_dev
*mdev
= priv
->mdev
;
3331 return mlx5_eswitch_set_vport_spoofchk(mdev
->priv
.eswitch
, vf
+ 1, setting
);
3334 static int mlx5e_set_vf_trust(struct net_device
*dev
, int vf
, bool setting
)
3336 struct mlx5e_priv
*priv
= netdev_priv(dev
);
3337 struct mlx5_core_dev
*mdev
= priv
->mdev
;
3339 return mlx5_eswitch_set_vport_trust(mdev
->priv
.eswitch
, vf
+ 1, setting
);
3342 static int mlx5e_set_vf_rate(struct net_device
*dev
, int vf
, int min_tx_rate
,
3345 struct mlx5e_priv
*priv
= netdev_priv(dev
);
3346 struct mlx5_core_dev
*mdev
= priv
->mdev
;
3348 return mlx5_eswitch_set_vport_rate(mdev
->priv
.eswitch
, vf
+ 1,
3349 max_tx_rate
, min_tx_rate
);
3352 static int mlx5_vport_link2ifla(u8 esw_link
)
3355 case MLX5_ESW_VPORT_ADMIN_STATE_DOWN
:
3356 return IFLA_VF_LINK_STATE_DISABLE
;
3357 case MLX5_ESW_VPORT_ADMIN_STATE_UP
:
3358 return IFLA_VF_LINK_STATE_ENABLE
;
3360 return IFLA_VF_LINK_STATE_AUTO
;
3363 static int mlx5_ifla_link2vport(u8 ifla_link
)
3365 switch (ifla_link
) {
3366 case IFLA_VF_LINK_STATE_DISABLE
:
3367 return MLX5_ESW_VPORT_ADMIN_STATE_DOWN
;
3368 case IFLA_VF_LINK_STATE_ENABLE
:
3369 return MLX5_ESW_VPORT_ADMIN_STATE_UP
;
3371 return MLX5_ESW_VPORT_ADMIN_STATE_AUTO
;
3374 static int mlx5e_set_vf_link_state(struct net_device
*dev
, int vf
,
3377 struct mlx5e_priv
*priv
= netdev_priv(dev
);
3378 struct mlx5_core_dev
*mdev
= priv
->mdev
;
3380 return mlx5_eswitch_set_vport_state(mdev
->priv
.eswitch
, vf
+ 1,
3381 mlx5_ifla_link2vport(link_state
));
3384 static int mlx5e_get_vf_config(struct net_device
*dev
,
3385 int vf
, struct ifla_vf_info
*ivi
)
3387 struct mlx5e_priv
*priv
= netdev_priv(dev
);
3388 struct mlx5_core_dev
*mdev
= priv
->mdev
;
3391 err
= mlx5_eswitch_get_vport_config(mdev
->priv
.eswitch
, vf
+ 1, ivi
);
3394 ivi
->linkstate
= mlx5_vport_link2ifla(ivi
->linkstate
);
3398 static int mlx5e_get_vf_stats(struct net_device
*dev
,
3399 int vf
, struct ifla_vf_stats
*vf_stats
)
3401 struct mlx5e_priv
*priv
= netdev_priv(dev
);
3402 struct mlx5_core_dev
*mdev
= priv
->mdev
;
3404 return mlx5_eswitch_get_vport_stats(mdev
->priv
.eswitch
, vf
+ 1,
3408 static void mlx5e_add_vxlan_port(struct net_device
*netdev
,
3409 struct udp_tunnel_info
*ti
)
3411 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3413 if (ti
->type
!= UDP_TUNNEL_TYPE_VXLAN
)
3416 if (!mlx5e_vxlan_allowed(priv
->mdev
))
3419 mlx5e_vxlan_queue_work(priv
, ti
->sa_family
, be16_to_cpu(ti
->port
), 1);
3422 static void mlx5e_del_vxlan_port(struct net_device
*netdev
,
3423 struct udp_tunnel_info
*ti
)
3425 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3427 if (ti
->type
!= UDP_TUNNEL_TYPE_VXLAN
)
3430 if (!mlx5e_vxlan_allowed(priv
->mdev
))
3433 mlx5e_vxlan_queue_work(priv
, ti
->sa_family
, be16_to_cpu(ti
->port
), 0);
3436 static netdev_features_t
mlx5e_vxlan_features_check(struct mlx5e_priv
*priv
,
3437 struct sk_buff
*skb
,
3438 netdev_features_t features
)
3440 struct udphdr
*udph
;
3444 switch (vlan_get_protocol(skb
)) {
3445 case htons(ETH_P_IP
):
3446 proto
= ip_hdr(skb
)->protocol
;
3448 case htons(ETH_P_IPV6
):
3449 proto
= ipv6_hdr(skb
)->nexthdr
;
3455 if (proto
== IPPROTO_UDP
) {
3456 udph
= udp_hdr(skb
);
3457 port
= be16_to_cpu(udph
->dest
);
3460 /* Verify if UDP port is being offloaded by HW */
3461 if (port
&& mlx5e_vxlan_lookup_port(priv
, port
))
3465 /* Disable CSUM and GSO if the udp dport is not offloaded by HW */
3466 return features
& ~(NETIF_F_CSUM_MASK
| NETIF_F_GSO_MASK
);
3469 static netdev_features_t
mlx5e_features_check(struct sk_buff
*skb
,
3470 struct net_device
*netdev
,
3471 netdev_features_t features
)
3473 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3475 features
= vlan_features_check(skb
, features
);
3476 features
= vxlan_features_check(skb
, features
);
3478 /* Validate if the tunneled packet is being offloaded by HW */
3479 if (skb
->encapsulation
&&
3480 (features
& NETIF_F_CSUM_MASK
|| features
& NETIF_F_GSO_MASK
))
3481 return mlx5e_vxlan_features_check(priv
, skb
, features
);
3486 static void mlx5e_tx_timeout(struct net_device
*dev
)
3488 struct mlx5e_priv
*priv
= netdev_priv(dev
);
3489 bool sched_work
= false;
3492 netdev_err(dev
, "TX timeout detected\n");
3494 for (i
= 0; i
< priv
->channels
.num
* priv
->channels
.params
.num_tc
; i
++) {
3495 struct mlx5e_txqsq
*sq
= priv
->txq2sq
[i
];
3497 if (!netif_xmit_stopped(netdev_get_tx_queue(dev
, i
)))
3500 clear_bit(MLX5E_SQ_STATE_ENABLED
, &sq
->state
);
3501 netdev_err(dev
, "TX timeout on queue: %d, SQ: 0x%x, CQ: 0x%x, SQ Cons: 0x%x SQ Prod: 0x%x\n",
3502 i
, sq
->sqn
, sq
->cq
.mcq
.cqn
, sq
->cc
, sq
->pc
);
3505 if (sched_work
&& test_bit(MLX5E_STATE_OPENED
, &priv
->state
))
3506 schedule_work(&priv
->tx_timeout_work
);
3509 static int mlx5e_xdp_set(struct net_device
*netdev
, struct bpf_prog
*prog
)
3511 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3512 struct bpf_prog
*old_prog
;
3514 bool reset
, was_opened
;
3517 mutex_lock(&priv
->state_lock
);
3519 if ((netdev
->features
& NETIF_F_LRO
) && prog
) {
3520 netdev_warn(netdev
, "can't set XDP while LRO is on, disable LRO first\n");
3525 was_opened
= test_bit(MLX5E_STATE_OPENED
, &priv
->state
);
3526 /* no need for full reset when exchanging programs */
3527 reset
= (!priv
->channels
.params
.xdp_prog
|| !prog
);
3529 if (was_opened
&& reset
)
3530 mlx5e_close_locked(netdev
);
3531 if (was_opened
&& !reset
) {
3532 /* num_channels is invariant here, so we can take the
3533 * batched reference right upfront.
3535 prog
= bpf_prog_add(prog
, priv
->channels
.num
);
3537 err
= PTR_ERR(prog
);
3542 /* exchange programs, extra prog reference we got from caller
3543 * as long as we don't fail from this point onwards.
3545 old_prog
= xchg(&priv
->channels
.params
.xdp_prog
, prog
);
3547 bpf_prog_put(old_prog
);
3549 if (reset
) /* change RQ type according to priv->xdp_prog */
3550 mlx5e_set_rq_params(priv
->mdev
, &priv
->channels
.params
);
3552 if (was_opened
&& reset
)
3553 mlx5e_open_locked(netdev
);
3555 if (!test_bit(MLX5E_STATE_OPENED
, &priv
->state
) || reset
)
3558 /* exchanging programs w/o reset, we update ref counts on behalf
3559 * of the channels RQs here.
3561 for (i
= 0; i
< priv
->channels
.num
; i
++) {
3562 struct mlx5e_channel
*c
= priv
->channels
.c
[i
];
3564 clear_bit(MLX5E_RQ_STATE_ENABLED
, &c
->rq
.state
);
3565 napi_synchronize(&c
->napi
);
3566 /* prevent mlx5e_poll_rx_cq from accessing rq->xdp_prog */
3568 old_prog
= xchg(&c
->rq
.xdp_prog
, prog
);
3570 set_bit(MLX5E_RQ_STATE_ENABLED
, &c
->rq
.state
);
3571 /* napi_schedule in case we have missed anything */
3572 set_bit(MLX5E_CHANNEL_NAPI_SCHED
, &c
->flags
);
3573 napi_schedule(&c
->napi
);
3576 bpf_prog_put(old_prog
);
3580 mutex_unlock(&priv
->state_lock
);
3584 static bool mlx5e_xdp_attached(struct net_device
*dev
)
3586 struct mlx5e_priv
*priv
= netdev_priv(dev
);
3588 return !!priv
->channels
.params
.xdp_prog
;
3591 static int mlx5e_xdp(struct net_device
*dev
, struct netdev_xdp
*xdp
)
3593 switch (xdp
->command
) {
3594 case XDP_SETUP_PROG
:
3595 return mlx5e_xdp_set(dev
, xdp
->prog
);
3596 case XDP_QUERY_PROG
:
3597 xdp
->prog_attached
= mlx5e_xdp_attached(dev
);
3604 #ifdef CONFIG_NET_POLL_CONTROLLER
3605 /* Fake "interrupt" called by netpoll (eg netconsole) to send skbs without
3606 * reenabling interrupts.
3608 static void mlx5e_netpoll(struct net_device
*dev
)
3610 struct mlx5e_priv
*priv
= netdev_priv(dev
);
3611 struct mlx5e_channels
*chs
= &priv
->channels
;
3615 for (i
= 0; i
< chs
->num
; i
++)
3616 napi_schedule(&chs
->c
[i
]->napi
);
3620 static const struct net_device_ops mlx5e_netdev_ops_basic
= {
3621 .ndo_open
= mlx5e_open
,
3622 .ndo_stop
= mlx5e_close
,
3623 .ndo_start_xmit
= mlx5e_xmit
,
3624 .ndo_setup_tc
= mlx5e_ndo_setup_tc
,
3625 .ndo_select_queue
= mlx5e_select_queue
,
3626 .ndo_get_stats64
= mlx5e_get_stats
,
3627 .ndo_set_rx_mode
= mlx5e_set_rx_mode
,
3628 .ndo_set_mac_address
= mlx5e_set_mac
,
3629 .ndo_vlan_rx_add_vid
= mlx5e_vlan_rx_add_vid
,
3630 .ndo_vlan_rx_kill_vid
= mlx5e_vlan_rx_kill_vid
,
3631 .ndo_set_features
= mlx5e_set_features
,
3632 .ndo_change_mtu
= mlx5e_change_mtu
,
3633 .ndo_do_ioctl
= mlx5e_ioctl
,
3634 .ndo_set_tx_maxrate
= mlx5e_set_tx_maxrate
,
3635 #ifdef CONFIG_RFS_ACCEL
3636 .ndo_rx_flow_steer
= mlx5e_rx_flow_steer
,
3638 .ndo_tx_timeout
= mlx5e_tx_timeout
,
3639 .ndo_xdp
= mlx5e_xdp
,
3640 #ifdef CONFIG_NET_POLL_CONTROLLER
3641 .ndo_poll_controller
= mlx5e_netpoll
,
3645 static const struct net_device_ops mlx5e_netdev_ops_sriov
= {
3646 .ndo_open
= mlx5e_open
,
3647 .ndo_stop
= mlx5e_close
,
3648 .ndo_start_xmit
= mlx5e_xmit
,
3649 .ndo_setup_tc
= mlx5e_ndo_setup_tc
,
3650 .ndo_select_queue
= mlx5e_select_queue
,
3651 .ndo_get_stats64
= mlx5e_get_stats
,
3652 .ndo_set_rx_mode
= mlx5e_set_rx_mode
,
3653 .ndo_set_mac_address
= mlx5e_set_mac
,
3654 .ndo_vlan_rx_add_vid
= mlx5e_vlan_rx_add_vid
,
3655 .ndo_vlan_rx_kill_vid
= mlx5e_vlan_rx_kill_vid
,
3656 .ndo_set_features
= mlx5e_set_features
,
3657 .ndo_change_mtu
= mlx5e_change_mtu
,
3658 .ndo_do_ioctl
= mlx5e_ioctl
,
3659 .ndo_udp_tunnel_add
= mlx5e_add_vxlan_port
,
3660 .ndo_udp_tunnel_del
= mlx5e_del_vxlan_port
,
3661 .ndo_set_tx_maxrate
= mlx5e_set_tx_maxrate
,
3662 .ndo_features_check
= mlx5e_features_check
,
3663 #ifdef CONFIG_RFS_ACCEL
3664 .ndo_rx_flow_steer
= mlx5e_rx_flow_steer
,
3666 .ndo_set_vf_mac
= mlx5e_set_vf_mac
,
3667 .ndo_set_vf_vlan
= mlx5e_set_vf_vlan
,
3668 .ndo_set_vf_spoofchk
= mlx5e_set_vf_spoofchk
,
3669 .ndo_set_vf_trust
= mlx5e_set_vf_trust
,
3670 .ndo_set_vf_rate
= mlx5e_set_vf_rate
,
3671 .ndo_get_vf_config
= mlx5e_get_vf_config
,
3672 .ndo_set_vf_link_state
= mlx5e_set_vf_link_state
,
3673 .ndo_get_vf_stats
= mlx5e_get_vf_stats
,
3674 .ndo_tx_timeout
= mlx5e_tx_timeout
,
3675 .ndo_xdp
= mlx5e_xdp
,
3676 #ifdef CONFIG_NET_POLL_CONTROLLER
3677 .ndo_poll_controller
= mlx5e_netpoll
,
3679 .ndo_has_offload_stats
= mlx5e_has_offload_stats
,
3680 .ndo_get_offload_stats
= mlx5e_get_offload_stats
,
3683 static int mlx5e_check_required_hca_cap(struct mlx5_core_dev
*mdev
)
3685 if (MLX5_CAP_GEN(mdev
, port_type
) != MLX5_CAP_PORT_TYPE_ETH
)
3687 if (!MLX5_CAP_GEN(mdev
, eth_net_offloads
) ||
3688 !MLX5_CAP_GEN(mdev
, nic_flow_table
) ||
3689 !MLX5_CAP_ETH(mdev
, csum_cap
) ||
3690 !MLX5_CAP_ETH(mdev
, max_lso_cap
) ||
3691 !MLX5_CAP_ETH(mdev
, vlan_cap
) ||
3692 !MLX5_CAP_ETH(mdev
, rss_ind_tbl_cap
) ||
3693 MLX5_CAP_FLOWTABLE(mdev
,
3694 flow_table_properties_nic_receive
.max_ft_level
)
3696 mlx5_core_warn(mdev
,
3697 "Not creating net device, some required device capabilities are missing\n");
3700 if (!MLX5_CAP_ETH(mdev
, self_lb_en_modifiable
))
3701 mlx5_core_warn(mdev
, "Self loop back prevention is not supported\n");
3702 if (!MLX5_CAP_GEN(mdev
, cq_moderation
))
3703 mlx5_core_warn(mdev
, "CQ modiration is not supported\n");
3708 u16
mlx5e_get_max_inline_cap(struct mlx5_core_dev
*mdev
)
3710 int bf_buf_size
= (1 << MLX5_CAP_GEN(mdev
, log_bf_reg_size
)) / 2;
3712 return bf_buf_size
-
3713 sizeof(struct mlx5e_tx_wqe
) +
3714 2 /*sizeof(mlx5e_tx_wqe.inline_hdr_start)*/;
3717 void mlx5e_build_default_indir_rqt(struct mlx5_core_dev
*mdev
,
3718 u32
*indirection_rqt
, int len
,
3721 int node
= mdev
->priv
.numa_node
;
3722 int node_num_of_cores
;
3726 node
= first_online_node
;
3728 node_num_of_cores
= cpumask_weight(cpumask_of_node(node
));
3730 if (node_num_of_cores
)
3731 num_channels
= min_t(int, num_channels
, node_num_of_cores
);
3733 for (i
= 0; i
< len
; i
++)
3734 indirection_rqt
[i
] = i
% num_channels
;
3737 static int mlx5e_get_pci_bw(struct mlx5_core_dev
*mdev
, u32
*pci_bw
)
3739 enum pcie_link_width width
;
3740 enum pci_bus_speed speed
;
3743 err
= pcie_get_minimum_link(mdev
->pdev
, &speed
, &width
);
3747 if (speed
== PCI_SPEED_UNKNOWN
|| width
== PCIE_LNK_WIDTH_UNKNOWN
)
3751 case PCIE_SPEED_2_5GT
:
3752 *pci_bw
= 2500 * width
;
3754 case PCIE_SPEED_5_0GT
:
3755 *pci_bw
= 5000 * width
;
3757 case PCIE_SPEED_8_0GT
:
3758 *pci_bw
= 8000 * width
;
3767 static bool cqe_compress_heuristic(u32 link_speed
, u32 pci_bw
)
3769 return (link_speed
&& pci_bw
&&
3770 (pci_bw
< 40000) && (pci_bw
< link_speed
));
3773 void mlx5e_set_rx_cq_mode_params(struct mlx5e_params
*params
, u8 cq_period_mode
)
3775 params
->rx_cq_period_mode
= cq_period_mode
;
3777 params
->rx_cq_moderation
.pkts
=
3778 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS
;
3779 params
->rx_cq_moderation
.usec
=
3780 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC
;
3782 if (cq_period_mode
== MLX5_CQ_PERIOD_MODE_START_FROM_CQE
)
3783 params
->rx_cq_moderation
.usec
=
3784 MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE
;
3786 if (params
->rx_am_enabled
)
3787 params
->rx_cq_moderation
=
3788 mlx5e_am_get_def_profile(params
->rx_cq_period_mode
);
3790 MLX5E_SET_PFLAG(params
, MLX5E_PFLAG_RX_CQE_BASED_MODER
,
3791 params
->rx_cq_period_mode
== MLX5_CQ_PERIOD_MODE_START_FROM_CQE
);
3794 u32
mlx5e_choose_lro_timeout(struct mlx5_core_dev
*mdev
, u32 wanted_timeout
)
3798 /* The supported periods are organized in ascending order */
3799 for (i
= 0; i
< MLX5E_LRO_TIMEOUT_ARR_SIZE
- 1; i
++)
3800 if (MLX5_CAP_ETH(mdev
, lro_timer_supported_periods
[i
]) >= wanted_timeout
)
3803 return MLX5_CAP_ETH(mdev
, lro_timer_supported_periods
[i
]);
3806 static void mlx5e_build_nic_params(struct mlx5_core_dev
*mdev
,
3807 struct mlx5e_params
*params
,
3810 u8 cq_period_mode
= 0;
3814 params
->num_channels
= max_channels
;
3818 params
->log_sq_size
= is_kdump_kernel() ?
3819 MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE
:
3820 MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE
;
3822 /* set CQE compression */
3823 params
->rx_cqe_compress_def
= false;
3824 if (MLX5_CAP_GEN(mdev
, cqe_compression
) &&
3825 MLX5_CAP_GEN(mdev
, vport_group_manager
)) {
3826 mlx5e_get_max_linkspeed(mdev
, &link_speed
);
3827 mlx5e_get_pci_bw(mdev
, &pci_bw
);
3828 mlx5_core_dbg(mdev
, "Max link speed = %d, PCI BW = %d\n",
3829 link_speed
, pci_bw
);
3830 params
->rx_cqe_compress_def
= cqe_compress_heuristic(link_speed
, pci_bw
);
3832 MLX5E_SET_PFLAG(params
, MLX5E_PFLAG_RX_CQE_COMPRESS
, params
->rx_cqe_compress_def
);
3835 mlx5e_set_rq_params(mdev
, params
);
3838 if (params
->rq_wq_type
== MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ
)
3839 params
->lro_en
= true;
3840 params
->lro_timeout
= mlx5e_choose_lro_timeout(mdev
, MLX5E_DEFAULT_LRO_TIMEOUT
);
3842 /* CQ moderation params */
3843 cq_period_mode
= MLX5_CAP_GEN(mdev
, cq_period_start_from_cqe
) ?
3844 MLX5_CQ_PERIOD_MODE_START_FROM_CQE
:
3845 MLX5_CQ_PERIOD_MODE_START_FROM_EQE
;
3846 params
->rx_am_enabled
= MLX5_CAP_GEN(mdev
, cq_moderation
);
3847 mlx5e_set_rx_cq_mode_params(params
, cq_period_mode
);
3849 params
->tx_cq_moderation
.usec
= MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC
;
3850 params
->tx_cq_moderation
.pkts
= MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS
;
3853 params
->tx_max_inline
= mlx5e_get_max_inline_cap(mdev
);
3854 mlx5_query_min_inline(mdev
, ¶ms
->tx_min_inline_mode
);
3855 if (params
->tx_min_inline_mode
== MLX5_INLINE_MODE_NONE
&&
3856 !MLX5_CAP_ETH(mdev
, wqe_vlan_insert
))
3857 params
->tx_min_inline_mode
= MLX5_INLINE_MODE_L2
;
3860 params
->rss_hfunc
= ETH_RSS_HASH_XOR
;
3861 netdev_rss_key_fill(params
->toeplitz_hash_key
, sizeof(params
->toeplitz_hash_key
));
3862 mlx5e_build_default_indir_rqt(mdev
, params
->indirection_rqt
,
3863 MLX5E_INDIR_RQT_SIZE
, max_channels
);
3866 static void mlx5e_build_nic_netdev_priv(struct mlx5_core_dev
*mdev
,
3867 struct net_device
*netdev
,
3868 const struct mlx5e_profile
*profile
,
3871 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3874 priv
->netdev
= netdev
;
3875 priv
->profile
= profile
;
3876 priv
->ppriv
= ppriv
;
3878 mlx5e_build_nic_params(mdev
, &priv
->channels
.params
, profile
->max_nch(mdev
));
3880 mutex_init(&priv
->state_lock
);
3882 INIT_WORK(&priv
->update_carrier_work
, mlx5e_update_carrier_work
);
3883 INIT_WORK(&priv
->set_rx_mode_work
, mlx5e_set_rx_mode_work
);
3884 INIT_WORK(&priv
->tx_timeout_work
, mlx5e_tx_timeout_work
);
3885 INIT_DELAYED_WORK(&priv
->update_stats_work
, mlx5e_update_stats_work
);
3888 static void mlx5e_set_netdev_dev_addr(struct net_device
*netdev
)
3890 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3892 mlx5_query_nic_vport_mac_address(priv
->mdev
, 0, netdev
->dev_addr
);
3893 if (is_zero_ether_addr(netdev
->dev_addr
) &&
3894 !MLX5_CAP_GEN(priv
->mdev
, vport_group_manager
)) {
3895 eth_hw_addr_random(netdev
);
3896 mlx5_core_info(priv
->mdev
, "Assigned random MAC address %pM\n", netdev
->dev_addr
);
3900 static const struct switchdev_ops mlx5e_switchdev_ops
= {
3901 .switchdev_port_attr_get
= mlx5e_attr_get
,
3904 static void mlx5e_build_nic_netdev(struct net_device
*netdev
)
3906 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
3907 struct mlx5_core_dev
*mdev
= priv
->mdev
;
3911 SET_NETDEV_DEV(netdev
, &mdev
->pdev
->dev
);
3913 if (MLX5_CAP_GEN(mdev
, vport_group_manager
)) {
3914 netdev
->netdev_ops
= &mlx5e_netdev_ops_sriov
;
3915 #ifdef CONFIG_MLX5_CORE_EN_DCB
3916 if (MLX5_CAP_GEN(mdev
, qos
))
3917 netdev
->dcbnl_ops
= &mlx5e_dcbnl_ops
;
3920 netdev
->netdev_ops
= &mlx5e_netdev_ops_basic
;
3923 netdev
->watchdog_timeo
= 15 * HZ
;
3925 netdev
->ethtool_ops
= &mlx5e_ethtool_ops
;
3927 netdev
->vlan_features
|= NETIF_F_SG
;
3928 netdev
->vlan_features
|= NETIF_F_IP_CSUM
;
3929 netdev
->vlan_features
|= NETIF_F_IPV6_CSUM
;
3930 netdev
->vlan_features
|= NETIF_F_GRO
;
3931 netdev
->vlan_features
|= NETIF_F_TSO
;
3932 netdev
->vlan_features
|= NETIF_F_TSO6
;
3933 netdev
->vlan_features
|= NETIF_F_RXCSUM
;
3934 netdev
->vlan_features
|= NETIF_F_RXHASH
;
3936 if (!!MLX5_CAP_ETH(mdev
, lro_cap
))
3937 netdev
->vlan_features
|= NETIF_F_LRO
;
3939 netdev
->hw_features
= netdev
->vlan_features
;
3940 netdev
->hw_features
|= NETIF_F_HW_VLAN_CTAG_TX
;
3941 netdev
->hw_features
|= NETIF_F_HW_VLAN_CTAG_RX
;
3942 netdev
->hw_features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
3944 if (mlx5e_vxlan_allowed(mdev
)) {
3945 netdev
->hw_features
|= NETIF_F_GSO_UDP_TUNNEL
|
3946 NETIF_F_GSO_UDP_TUNNEL_CSUM
|
3947 NETIF_F_GSO_PARTIAL
;
3948 netdev
->hw_enc_features
|= NETIF_F_IP_CSUM
;
3949 netdev
->hw_enc_features
|= NETIF_F_IPV6_CSUM
;
3950 netdev
->hw_enc_features
|= NETIF_F_TSO
;
3951 netdev
->hw_enc_features
|= NETIF_F_TSO6
;
3952 netdev
->hw_enc_features
|= NETIF_F_GSO_UDP_TUNNEL
;
3953 netdev
->hw_enc_features
|= NETIF_F_GSO_UDP_TUNNEL_CSUM
|
3954 NETIF_F_GSO_PARTIAL
;
3955 netdev
->gso_partial_features
= NETIF_F_GSO_UDP_TUNNEL_CSUM
;
3958 mlx5_query_port_fcs(mdev
, &fcs_supported
, &fcs_enabled
);
3961 netdev
->hw_features
|= NETIF_F_RXALL
;
3963 if (MLX5_CAP_ETH(mdev
, scatter_fcs
))
3964 netdev
->hw_features
|= NETIF_F_RXFCS
;
3966 netdev
->features
= netdev
->hw_features
;
3967 if (!priv
->channels
.params
.lro_en
)
3968 netdev
->features
&= ~NETIF_F_LRO
;
3971 netdev
->features
&= ~NETIF_F_RXALL
;
3973 if (!priv
->channels
.params
.scatter_fcs_en
)
3974 netdev
->features
&= ~NETIF_F_RXFCS
;
3976 #define FT_CAP(f) MLX5_CAP_FLOWTABLE(mdev, flow_table_properties_nic_receive.f)
3977 if (FT_CAP(flow_modify_en
) &&
3978 FT_CAP(modify_root
) &&
3979 FT_CAP(identified_miss_table_mode
) &&
3980 FT_CAP(flow_table_modify
)) {
3981 netdev
->hw_features
|= NETIF_F_HW_TC
;
3982 #ifdef CONFIG_RFS_ACCEL
3983 netdev
->hw_features
|= NETIF_F_NTUPLE
;
3987 netdev
->features
|= NETIF_F_HIGHDMA
;
3989 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
3991 mlx5e_set_netdev_dev_addr(netdev
);
3993 #ifdef CONFIG_NET_SWITCHDEV
3994 if (MLX5_CAP_GEN(mdev
, vport_group_manager
))
3995 netdev
->switchdev_ops
= &mlx5e_switchdev_ops
;
3999 static void mlx5e_create_q_counter(struct mlx5e_priv
*priv
)
4001 struct mlx5_core_dev
*mdev
= priv
->mdev
;
4004 err
= mlx5_core_alloc_q_counter(mdev
, &priv
->q_counter
);
4006 mlx5_core_warn(mdev
, "alloc queue counter failed, %d\n", err
);
4007 priv
->q_counter
= 0;
4011 static void mlx5e_destroy_q_counter(struct mlx5e_priv
*priv
)
4013 if (!priv
->q_counter
)
4016 mlx5_core_dealloc_q_counter(priv
->mdev
, priv
->q_counter
);
4019 static void mlx5e_nic_init(struct mlx5_core_dev
*mdev
,
4020 struct net_device
*netdev
,
4021 const struct mlx5e_profile
*profile
,
4024 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
4026 mlx5e_build_nic_netdev_priv(mdev
, netdev
, profile
, ppriv
);
4027 mlx5e_build_nic_netdev(netdev
);
4028 mlx5e_vxlan_init(priv
);
4031 static void mlx5e_nic_cleanup(struct mlx5e_priv
*priv
)
4033 mlx5e_vxlan_cleanup(priv
);
4035 if (priv
->channels
.params
.xdp_prog
)
4036 bpf_prog_put(priv
->channels
.params
.xdp_prog
);
4039 static int mlx5e_init_nic_rx(struct mlx5e_priv
*priv
)
4041 struct mlx5_core_dev
*mdev
= priv
->mdev
;
4045 err
= mlx5e_create_indirect_rqts(priv
);
4047 mlx5_core_warn(mdev
, "create indirect rqts failed, %d\n", err
);
4051 err
= mlx5e_create_direct_rqts(priv
);
4053 mlx5_core_warn(mdev
, "create direct rqts failed, %d\n", err
);
4054 goto err_destroy_indirect_rqts
;
4057 err
= mlx5e_create_indirect_tirs(priv
);
4059 mlx5_core_warn(mdev
, "create indirect tirs failed, %d\n", err
);
4060 goto err_destroy_direct_rqts
;
4063 err
= mlx5e_create_direct_tirs(priv
);
4065 mlx5_core_warn(mdev
, "create direct tirs failed, %d\n", err
);
4066 goto err_destroy_indirect_tirs
;
4069 err
= mlx5e_create_flow_steering(priv
);
4071 mlx5_core_warn(mdev
, "create flow steering failed, %d\n", err
);
4072 goto err_destroy_direct_tirs
;
4075 err
= mlx5e_tc_init(priv
);
4077 goto err_destroy_flow_steering
;
4081 err_destroy_flow_steering
:
4082 mlx5e_destroy_flow_steering(priv
);
4083 err_destroy_direct_tirs
:
4084 mlx5e_destroy_direct_tirs(priv
);
4085 err_destroy_indirect_tirs
:
4086 mlx5e_destroy_indirect_tirs(priv
);
4087 err_destroy_direct_rqts
:
4088 for (i
= 0; i
< priv
->profile
->max_nch(mdev
); i
++)
4089 mlx5e_destroy_rqt(priv
, &priv
->direct_tir
[i
].rqt
);
4090 err_destroy_indirect_rqts
:
4091 mlx5e_destroy_rqt(priv
, &priv
->indir_rqt
);
4095 static void mlx5e_cleanup_nic_rx(struct mlx5e_priv
*priv
)
4099 mlx5e_tc_cleanup(priv
);
4100 mlx5e_destroy_flow_steering(priv
);
4101 mlx5e_destroy_direct_tirs(priv
);
4102 mlx5e_destroy_indirect_tirs(priv
);
4103 for (i
= 0; i
< priv
->profile
->max_nch(priv
->mdev
); i
++)
4104 mlx5e_destroy_rqt(priv
, &priv
->direct_tir
[i
].rqt
);
4105 mlx5e_destroy_rqt(priv
, &priv
->indir_rqt
);
4108 static int mlx5e_init_nic_tx(struct mlx5e_priv
*priv
)
4112 err
= mlx5e_create_tises(priv
);
4114 mlx5_core_warn(priv
->mdev
, "create tises failed, %d\n", err
);
4118 #ifdef CONFIG_MLX5_CORE_EN_DCB
4119 mlx5e_dcbnl_initialize(priv
);
4124 static void mlx5e_nic_enable(struct mlx5e_priv
*priv
)
4126 struct net_device
*netdev
= priv
->netdev
;
4127 struct mlx5_core_dev
*mdev
= priv
->mdev
;
4128 struct mlx5_eswitch
*esw
= mdev
->priv
.eswitch
;
4129 struct mlx5_eswitch_rep rep
;
4131 mlx5_lag_add(mdev
, netdev
);
4133 mlx5e_enable_async_events(priv
);
4135 if (MLX5_CAP_GEN(mdev
, vport_group_manager
)) {
4136 mlx5_query_nic_vport_mac_address(mdev
, 0, rep
.hw_id
);
4137 rep
.load
= mlx5e_nic_rep_load
;
4138 rep
.unload
= mlx5e_nic_rep_unload
;
4139 rep
.vport
= FDB_UPLINK_VPORT
;
4140 rep
.netdev
= netdev
;
4141 mlx5_eswitch_register_vport_rep(esw
, 0, &rep
);
4144 if (netdev
->reg_state
!= NETREG_REGISTERED
)
4147 /* Device already registered: sync netdev system state */
4148 if (mlx5e_vxlan_allowed(mdev
)) {
4150 udp_tunnel_get_rx_info(netdev
);
4154 queue_work(priv
->wq
, &priv
->set_rx_mode_work
);
4157 static void mlx5e_nic_disable(struct mlx5e_priv
*priv
)
4159 struct mlx5_core_dev
*mdev
= priv
->mdev
;
4160 struct mlx5_eswitch
*esw
= mdev
->priv
.eswitch
;
4162 queue_work(priv
->wq
, &priv
->set_rx_mode_work
);
4163 if (MLX5_CAP_GEN(mdev
, vport_group_manager
))
4164 mlx5_eswitch_unregister_vport_rep(esw
, 0);
4165 mlx5e_disable_async_events(priv
);
4166 mlx5_lag_remove(mdev
);
4169 static const struct mlx5e_profile mlx5e_nic_profile
= {
4170 .init
= mlx5e_nic_init
,
4171 .cleanup
= mlx5e_nic_cleanup
,
4172 .init_rx
= mlx5e_init_nic_rx
,
4173 .cleanup_rx
= mlx5e_cleanup_nic_rx
,
4174 .init_tx
= mlx5e_init_nic_tx
,
4175 .cleanup_tx
= mlx5e_cleanup_nic_tx
,
4176 .enable
= mlx5e_nic_enable
,
4177 .disable
= mlx5e_nic_disable
,
4178 .update_stats
= mlx5e_update_stats
,
4179 .max_nch
= mlx5e_get_max_num_channels
,
4180 .max_tc
= MLX5E_MAX_NUM_TC
,
4183 struct net_device
*mlx5e_create_netdev(struct mlx5_core_dev
*mdev
,
4184 const struct mlx5e_profile
*profile
,
4187 int nch
= profile
->max_nch(mdev
);
4188 struct net_device
*netdev
;
4189 struct mlx5e_priv
*priv
;
4191 netdev
= alloc_etherdev_mqs(sizeof(struct mlx5e_priv
),
4192 nch
* profile
->max_tc
,
4195 mlx5_core_err(mdev
, "alloc_etherdev_mqs() failed\n");
4199 #ifdef CONFIG_RFS_ACCEL
4200 netdev
->rx_cpu_rmap
= mdev
->rmap
;
4203 profile
->init(mdev
, netdev
, profile
, ppriv
);
4205 netif_carrier_off(netdev
);
4207 priv
= netdev_priv(netdev
);
4209 priv
->wq
= create_singlethread_workqueue("mlx5e");
4211 goto err_cleanup_nic
;
4216 profile
->cleanup(priv
);
4217 free_netdev(netdev
);
4222 int mlx5e_attach_netdev(struct mlx5_core_dev
*mdev
, struct net_device
*netdev
)
4224 const struct mlx5e_profile
*profile
;
4225 struct mlx5e_priv
*priv
;
4229 priv
= netdev_priv(netdev
);
4230 profile
= priv
->profile
;
4231 clear_bit(MLX5E_STATE_DESTROYING
, &priv
->state
);
4233 err
= profile
->init_tx(priv
);
4237 err
= mlx5e_open_drop_rq(mdev
, &priv
->drop_rq
);
4239 mlx5_core_err(mdev
, "open drop rq failed, %d\n", err
);
4240 goto err_cleanup_tx
;
4243 err
= profile
->init_rx(priv
);
4245 goto err_close_drop_rq
;
4247 mlx5e_create_q_counter(priv
);
4249 mlx5e_init_l2_addr(priv
);
4251 /* MTU range: 68 - hw-specific max */
4252 netdev
->min_mtu
= ETH_MIN_MTU
;
4253 mlx5_query_port_max_mtu(priv
->mdev
, &max_mtu
, 1);
4254 netdev
->max_mtu
= MLX5E_HW2SW_MTU(max_mtu
);
4256 mlx5e_set_dev_port_mtu(priv
);
4258 if (profile
->enable
)
4259 profile
->enable(priv
);
4262 if (netif_running(netdev
))
4264 netif_device_attach(netdev
);
4270 mlx5e_close_drop_rq(&priv
->drop_rq
);
4273 profile
->cleanup_tx(priv
);
4279 static void mlx5e_register_vport_rep(struct mlx5_core_dev
*mdev
)
4281 struct mlx5_eswitch
*esw
= mdev
->priv
.eswitch
;
4282 int total_vfs
= MLX5_TOTAL_VPORTS(mdev
);
4286 if (!MLX5_CAP_GEN(mdev
, vport_group_manager
))
4289 mlx5_query_nic_vport_mac_address(mdev
, 0, mac
);
4291 for (vport
= 1; vport
< total_vfs
; vport
++) {
4292 struct mlx5_eswitch_rep rep
;
4294 rep
.load
= mlx5e_vport_rep_load
;
4295 rep
.unload
= mlx5e_vport_rep_unload
;
4297 ether_addr_copy(rep
.hw_id
, mac
);
4298 mlx5_eswitch_register_vport_rep(esw
, vport
, &rep
);
4302 static void mlx5e_unregister_vport_rep(struct mlx5_core_dev
*mdev
)
4304 struct mlx5_eswitch
*esw
= mdev
->priv
.eswitch
;
4305 int total_vfs
= MLX5_TOTAL_VPORTS(mdev
);
4308 if (!MLX5_CAP_GEN(mdev
, vport_group_manager
))
4311 for (vport
= 1; vport
< total_vfs
; vport
++)
4312 mlx5_eswitch_unregister_vport_rep(esw
, vport
);
4315 void mlx5e_detach_netdev(struct mlx5_core_dev
*mdev
, struct net_device
*netdev
)
4317 struct mlx5e_priv
*priv
= netdev_priv(netdev
);
4318 const struct mlx5e_profile
*profile
= priv
->profile
;
4320 set_bit(MLX5E_STATE_DESTROYING
, &priv
->state
);
4323 if (netif_running(netdev
))
4324 mlx5e_close(netdev
);
4325 netif_device_detach(netdev
);
4328 if (profile
->disable
)
4329 profile
->disable(priv
);
4330 flush_workqueue(priv
->wq
);
4332 mlx5e_destroy_q_counter(priv
);
4333 profile
->cleanup_rx(priv
);
4334 mlx5e_close_drop_rq(&priv
->drop_rq
);
4335 profile
->cleanup_tx(priv
);
4336 cancel_delayed_work_sync(&priv
->update_stats_work
);
4339 /* mlx5e_attach and mlx5e_detach scope should be only creating/destroying
4340 * hardware contexts and to connect it to the current netdev.
4342 static int mlx5e_attach(struct mlx5_core_dev
*mdev
, void *vpriv
)
4344 struct mlx5e_priv
*priv
= vpriv
;
4345 struct net_device
*netdev
= priv
->netdev
;
4348 if (netif_device_present(netdev
))
4351 err
= mlx5e_create_mdev_resources(mdev
);
4355 err
= mlx5e_attach_netdev(mdev
, netdev
);
4357 mlx5e_destroy_mdev_resources(mdev
);
4361 mlx5e_register_vport_rep(mdev
);
4365 static void mlx5e_detach(struct mlx5_core_dev
*mdev
, void *vpriv
)
4367 struct mlx5e_priv
*priv
= vpriv
;
4368 struct net_device
*netdev
= priv
->netdev
;
4370 if (!netif_device_present(netdev
))
4373 mlx5e_unregister_vport_rep(mdev
);
4374 mlx5e_detach_netdev(mdev
, netdev
);
4375 mlx5e_destroy_mdev_resources(mdev
);
4378 static void *mlx5e_add(struct mlx5_core_dev
*mdev
)
4380 struct mlx5_eswitch
*esw
= mdev
->priv
.eswitch
;
4381 int total_vfs
= MLX5_TOTAL_VPORTS(mdev
);
4386 struct net_device
*netdev
;
4388 err
= mlx5e_check_required_hca_cap(mdev
);
4392 if (MLX5_CAP_GEN(mdev
, vport_group_manager
))
4393 ppriv
= &esw
->offloads
.vport_reps
[0];
4395 netdev
= mlx5e_create_netdev(mdev
, &mlx5e_nic_profile
, ppriv
);
4397 mlx5_core_err(mdev
, "mlx5e_create_netdev failed\n");
4398 goto err_unregister_reps
;
4401 priv
= netdev_priv(netdev
);
4403 err
= mlx5e_attach(mdev
, priv
);
4405 mlx5_core_err(mdev
, "mlx5e_attach failed, %d\n", err
);
4406 goto err_destroy_netdev
;
4409 err
= register_netdev(netdev
);
4411 mlx5_core_err(mdev
, "register_netdev failed, %d\n", err
);
4418 mlx5e_detach(mdev
, priv
);
4421 mlx5e_destroy_netdev(mdev
, priv
);
4423 err_unregister_reps
:
4424 for (vport
= 1; vport
< total_vfs
; vport
++)
4425 mlx5_eswitch_unregister_vport_rep(esw
, vport
);
4430 void mlx5e_destroy_netdev(struct mlx5_core_dev
*mdev
, struct mlx5e_priv
*priv
)
4432 const struct mlx5e_profile
*profile
= priv
->profile
;
4433 struct net_device
*netdev
= priv
->netdev
;
4435 destroy_workqueue(priv
->wq
);
4436 if (profile
->cleanup
)
4437 profile
->cleanup(priv
);
4438 free_netdev(netdev
);
4441 static void mlx5e_remove(struct mlx5_core_dev
*mdev
, void *vpriv
)
4443 struct mlx5e_priv
*priv
= vpriv
;
4445 unregister_netdev(priv
->netdev
);
4446 mlx5e_detach(mdev
, vpriv
);
4447 mlx5e_destroy_netdev(mdev
, priv
);
4450 static void *mlx5e_get_netdev(void *vpriv
)
4452 struct mlx5e_priv
*priv
= vpriv
;
4454 return priv
->netdev
;
4457 static struct mlx5_interface mlx5e_interface
= {
4459 .remove
= mlx5e_remove
,
4460 .attach
= mlx5e_attach
,
4461 .detach
= mlx5e_detach
,
4462 .event
= mlx5e_async_event
,
4463 .protocol
= MLX5_INTERFACE_PROTOCOL_ETH
,
4464 .get_dev
= mlx5e_get_netdev
,
4467 void mlx5e_init(void)
4469 mlx5e_build_ptys2ethtool_map();
4470 mlx5_register_interface(&mlx5e_interface
);
4473 void mlx5e_cleanup(void)
4475 mlx5_unregister_interface(&mlx5e_interface
);