2 * Copyright (c) 2017, 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 <rdma/ib_verbs.h>
34 #include <linux/mlx5/fs.h>
38 #define IB_DEFAULT_Q_KEY 0xb1b
39 #define MLX5I_PARAMS_DEFAULT_LOG_RQ_SIZE 9
41 static int mlx5i_open(struct net_device
*netdev
);
42 static int mlx5i_close(struct net_device
*netdev
);
43 static int mlx5i_dev_init(struct net_device
*dev
);
44 static void mlx5i_dev_cleanup(struct net_device
*dev
);
45 static int mlx5i_change_mtu(struct net_device
*netdev
, int new_mtu
);
46 static int mlx5i_ioctl(struct net_device
*dev
, struct ifreq
*ifr
, int cmd
);
48 static const struct net_device_ops mlx5i_netdev_ops
= {
49 .ndo_open
= mlx5i_open
,
50 .ndo_stop
= mlx5i_close
,
51 .ndo_init
= mlx5i_dev_init
,
52 .ndo_uninit
= mlx5i_dev_cleanup
,
53 .ndo_change_mtu
= mlx5i_change_mtu
,
54 .ndo_do_ioctl
= mlx5i_ioctl
,
57 /* IPoIB mlx5 netdev profile */
58 static void mlx5i_build_nic_params(struct mlx5_core_dev
*mdev
,
59 struct mlx5e_params
*params
)
61 /* Override RQ params as IPoIB supports only LINKED LIST RQ for now */
62 mlx5e_set_rq_type_params(mdev
, params
, MLX5_WQ_TYPE_LINKED_LIST
);
64 /* RQ size in ipoib by default is 512 */
65 params
->log_rq_size
= is_kdump_kernel() ?
66 MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE
:
67 MLX5I_PARAMS_DEFAULT_LOG_RQ_SIZE
;
69 params
->lro_en
= false;
72 /* Called directly after IPoIB netdevice was created to initialize SW structs */
73 static void mlx5i_init(struct mlx5_core_dev
*mdev
,
74 struct net_device
*netdev
,
75 const struct mlx5e_profile
*profile
,
78 struct mlx5e_priv
*priv
= mlx5i_epriv(netdev
);
82 priv
->netdev
= netdev
;
83 priv
->profile
= profile
;
85 priv
->hard_mtu
= MLX5_IB_GRH_BYTES
+ MLX5_IPOIB_HARD_LEN
;
86 mutex_init(&priv
->state_lock
);
88 mlx5e_build_nic_params(mdev
, &priv
->channels
.params
, profile
->max_nch(mdev
));
89 mlx5i_build_nic_params(mdev
, &priv
->channels
.params
);
92 netdev
->hw_features
|= NETIF_F_SG
;
93 netdev
->hw_features
|= NETIF_F_IP_CSUM
;
94 netdev
->hw_features
|= NETIF_F_IPV6_CSUM
;
95 netdev
->hw_features
|= NETIF_F_GRO
;
96 netdev
->hw_features
|= NETIF_F_TSO
;
97 netdev
->hw_features
|= NETIF_F_TSO6
;
98 netdev
->hw_features
|= NETIF_F_RXCSUM
;
99 netdev
->hw_features
|= NETIF_F_RXHASH
;
101 netdev
->netdev_ops
= &mlx5i_netdev_ops
;
102 netdev
->ethtool_ops
= &mlx5i_ethtool_ops
;
105 /* Called directly before IPoIB netdevice is destroyed to cleanup SW structs */
106 static void mlx5i_cleanup(struct mlx5e_priv
*priv
)
111 #define MLX5_QP_ENHANCED_ULP_STATELESS_MODE 2
113 static int mlx5i_create_underlay_qp(struct mlx5_core_dev
*mdev
, struct mlx5_core_qp
*qp
)
115 struct mlx5_qp_context
*context
= NULL
;
122 inlen
= MLX5_ST_SZ_BYTES(create_qp_in
);
123 in
= kvzalloc(inlen
, GFP_KERNEL
);
127 qpc
= MLX5_ADDR_OF(create_qp_in
, in
, qpc
);
128 MLX5_SET(qpc
, qpc
, st
, MLX5_QP_ST_UD
);
129 MLX5_SET(qpc
, qpc
, pm_state
, MLX5_QP_PM_MIGRATED
);
130 MLX5_SET(qpc
, qpc
, ulp_stateless_offload_mode
,
131 MLX5_QP_ENHANCED_ULP_STATELESS_MODE
);
133 addr_path
= MLX5_ADDR_OF(qpc
, qpc
, primary_address_path
);
134 MLX5_SET(ads
, addr_path
, port
, 1);
135 MLX5_SET(ads
, addr_path
, grh
, 1);
137 ret
= mlx5_core_create_qp(mdev
, qp
, in
, inlen
);
139 mlx5_core_err(mdev
, "Failed creating IPoIB QP err : %d\n", ret
);
144 context
= kzalloc(sizeof(*context
), GFP_KERNEL
);
150 context
->flags
= cpu_to_be32(MLX5_QP_PM_MIGRATED
<< 11);
151 context
->pri_path
.port
= 1;
152 context
->qkey
= cpu_to_be32(IB_DEFAULT_Q_KEY
);
154 ret
= mlx5_core_qp_modify(mdev
, MLX5_CMD_OP_RST2INIT_QP
, 0, context
, qp
);
156 mlx5_core_err(mdev
, "Failed to modify qp RST2INIT, err: %d\n", ret
);
159 memset(context
, 0, sizeof(*context
));
161 ret
= mlx5_core_qp_modify(mdev
, MLX5_CMD_OP_INIT2RTR_QP
, 0, context
, qp
);
163 mlx5_core_err(mdev
, "Failed to modify qp INIT2RTR, err: %d\n", ret
);
167 ret
= mlx5_core_qp_modify(mdev
, MLX5_CMD_OP_RTR2RTS_QP
, 0, context
, qp
);
169 mlx5_core_err(mdev
, "Failed to modify qp RTR2RTS, err: %d\n", ret
);
179 static void mlx5i_destroy_underlay_qp(struct mlx5_core_dev
*mdev
, struct mlx5_core_qp
*qp
)
181 mlx5_core_destroy_qp(mdev
, qp
);
184 static int mlx5i_init_tx(struct mlx5e_priv
*priv
)
186 struct mlx5i_priv
*ipriv
= priv
->ppriv
;
189 err
= mlx5i_create_underlay_qp(priv
->mdev
, &ipriv
->qp
);
191 mlx5_core_warn(priv
->mdev
, "create underlay QP failed, %d\n", err
);
195 err
= mlx5e_create_tis(priv
->mdev
, 0 /* tc */, ipriv
->qp
.qpn
, &priv
->tisn
[0]);
197 mlx5_core_warn(priv
->mdev
, "create tis failed, %d\n", err
);
204 static void mlx5i_cleanup_tx(struct mlx5e_priv
*priv
)
206 struct mlx5i_priv
*ipriv
= priv
->ppriv
;
208 mlx5e_destroy_tis(priv
->mdev
, priv
->tisn
[0]);
209 mlx5i_destroy_underlay_qp(priv
->mdev
, &ipriv
->qp
);
212 static int mlx5i_create_flow_steering(struct mlx5e_priv
*priv
)
216 priv
->fs
.ns
= mlx5_get_flow_namespace(priv
->mdev
,
217 MLX5_FLOW_NAMESPACE_KERNEL
);
222 err
= mlx5e_arfs_create_tables(priv
);
224 netdev_err(priv
->netdev
, "Failed to create arfs tables, err=%d\n",
226 priv
->netdev
->hw_features
&= ~NETIF_F_NTUPLE
;
229 err
= mlx5e_create_ttc_table(priv
);
231 netdev_err(priv
->netdev
, "Failed to create ttc table, err=%d\n",
233 goto err_destroy_arfs_tables
;
238 err_destroy_arfs_tables
:
239 mlx5e_arfs_destroy_tables(priv
);
244 static void mlx5i_destroy_flow_steering(struct mlx5e_priv
*priv
)
246 mlx5e_destroy_ttc_table(priv
);
247 mlx5e_arfs_destroy_tables(priv
);
250 static int mlx5i_init_rx(struct mlx5e_priv
*priv
)
252 struct mlx5i_priv
*ipriv
= priv
->ppriv
;
255 err
= mlx5e_create_indirect_rqt(priv
);
259 err
= mlx5e_create_direct_rqts(priv
);
261 goto err_destroy_indirect_rqts
;
263 err
= mlx5e_create_indirect_tirs(priv
);
265 goto err_destroy_direct_rqts
;
267 err
= mlx5e_create_direct_tirs(priv
);
269 goto err_destroy_indirect_tirs
;
271 err
= mlx5_fs_add_rx_underlay_qpn(priv
->mdev
, ipriv
->qp
.qpn
);
273 goto err_destroy_direct_tirs
;
275 err
= mlx5i_create_flow_steering(priv
);
277 goto err_remove_rx_underlay_qpn
;
281 err_remove_rx_underlay_qpn
:
282 mlx5_fs_remove_rx_underlay_qpn(priv
->mdev
, ipriv
->qp
.qpn
);
283 err_destroy_direct_tirs
:
284 mlx5e_destroy_direct_tirs(priv
);
285 err_destroy_indirect_tirs
:
286 mlx5e_destroy_indirect_tirs(priv
);
287 err_destroy_direct_rqts
:
288 mlx5e_destroy_direct_rqts(priv
);
289 err_destroy_indirect_rqts
:
290 mlx5e_destroy_rqt(priv
, &priv
->indir_rqt
);
294 static void mlx5i_cleanup_rx(struct mlx5e_priv
*priv
)
296 struct mlx5i_priv
*ipriv
= priv
->ppriv
;
298 mlx5_fs_remove_rx_underlay_qpn(priv
->mdev
, ipriv
->qp
.qpn
);
299 mlx5i_destroy_flow_steering(priv
);
300 mlx5e_destroy_direct_tirs(priv
);
301 mlx5e_destroy_indirect_tirs(priv
);
302 mlx5e_destroy_direct_rqts(priv
);
303 mlx5e_destroy_rqt(priv
, &priv
->indir_rqt
);
306 static const struct mlx5e_profile mlx5i_nic_profile
= {
308 .cleanup
= mlx5i_cleanup
,
309 .init_tx
= mlx5i_init_tx
,
310 .cleanup_tx
= mlx5i_cleanup_tx
,
311 .init_rx
= mlx5i_init_rx
,
312 .cleanup_rx
= mlx5i_cleanup_rx
,
313 .enable
= NULL
, /* mlx5i_enable */
314 .disable
= NULL
, /* mlx5i_disable */
315 .update_stats
= NULL
, /* mlx5i_update_stats */
316 .max_nch
= mlx5e_get_max_num_channels
,
317 .update_carrier
= NULL
, /* no HW update in IB link */
318 .rx_handlers
.handle_rx_cqe
= mlx5i_handle_rx_cqe
,
319 .rx_handlers
.handle_rx_cqe_mpwqe
= NULL
, /* Not supported */
320 .max_tc
= MLX5I_MAX_NUM_TC
,
323 /* mlx5i netdev NDos */
325 static int mlx5i_change_mtu(struct net_device
*netdev
, int new_mtu
)
327 struct mlx5e_priv
*priv
= mlx5i_epriv(netdev
);
328 struct mlx5e_channels new_channels
= {};
332 mutex_lock(&priv
->state_lock
);
334 curr_mtu
= netdev
->mtu
;
335 netdev
->mtu
= new_mtu
;
337 if (!test_bit(MLX5E_STATE_OPENED
, &priv
->state
))
340 new_channels
.params
= priv
->channels
.params
;
341 err
= mlx5e_open_channels(priv
, &new_channels
);
343 netdev
->mtu
= curr_mtu
;
347 mlx5e_switch_priv_channels(priv
, &new_channels
, NULL
);
350 mutex_unlock(&priv
->state_lock
);
354 static int mlx5i_dev_init(struct net_device
*dev
)
356 struct mlx5e_priv
*priv
= mlx5i_epriv(dev
);
357 struct mlx5i_priv
*ipriv
= priv
->ppriv
;
359 /* Set dev address using underlay QP */
360 dev
->dev_addr
[1] = (ipriv
->qp
.qpn
>> 16) & 0xff;
361 dev
->dev_addr
[2] = (ipriv
->qp
.qpn
>> 8) & 0xff;
362 dev
->dev_addr
[3] = (ipriv
->qp
.qpn
) & 0xff;
367 static int mlx5i_ioctl(struct net_device
*dev
, struct ifreq
*ifr
, int cmd
)
369 struct mlx5e_priv
*priv
= mlx5i_epriv(dev
);
373 return mlx5e_hwstamp_set(priv
, ifr
);
375 return mlx5e_hwstamp_get(priv
, ifr
);
381 static void mlx5i_dev_cleanup(struct net_device
*dev
)
383 struct mlx5e_priv
*priv
= mlx5i_epriv(dev
);
384 struct mlx5_core_dev
*mdev
= priv
->mdev
;
385 struct mlx5i_priv
*ipriv
= priv
->ppriv
;
386 struct mlx5_qp_context context
;
388 /* detach qp from flow-steering by reset it */
389 mlx5_core_qp_modify(mdev
, MLX5_CMD_OP_2RST_QP
, 0, &context
, &ipriv
->qp
);
392 static int mlx5i_open(struct net_device
*netdev
)
394 struct mlx5e_priv
*priv
= mlx5i_epriv(netdev
);
397 mutex_lock(&priv
->state_lock
);
399 set_bit(MLX5E_STATE_OPENED
, &priv
->state
);
401 err
= mlx5e_open_channels(priv
, &priv
->channels
);
403 goto err_clear_state_opened_flag
;
405 mlx5e_refresh_tirs(priv
, false);
406 mlx5e_activate_priv_channels(priv
);
407 mlx5e_timestamp_init(priv
);
409 mutex_unlock(&priv
->state_lock
);
412 err_clear_state_opened_flag
:
413 clear_bit(MLX5E_STATE_OPENED
, &priv
->state
);
414 mutex_unlock(&priv
->state_lock
);
418 static int mlx5i_close(struct net_device
*netdev
)
420 struct mlx5e_priv
*priv
= mlx5i_epriv(netdev
);
422 /* May already be CLOSED in case a previous configuration operation
423 * (e.g RX/TX queue size change) that involves close&open failed.
425 mutex_lock(&priv
->state_lock
);
427 if (!test_bit(MLX5E_STATE_OPENED
, &priv
->state
))
430 clear_bit(MLX5E_STATE_OPENED
, &priv
->state
);
432 mlx5e_timestamp_cleanup(priv
);
433 netif_carrier_off(priv
->netdev
);
434 mlx5e_deactivate_priv_channels(priv
);
435 mlx5e_close_channels(&priv
->channels
);
437 mutex_unlock(&priv
->state_lock
);
441 /* IPoIB RDMA netdev callbacks */
442 static int mlx5i_attach_mcast(struct net_device
*netdev
, struct ib_device
*hca
,
443 union ib_gid
*gid
, u16 lid
, int set_qkey
,
446 struct mlx5e_priv
*epriv
= mlx5i_epriv(netdev
);
447 struct mlx5_core_dev
*mdev
= epriv
->mdev
;
448 struct mlx5i_priv
*ipriv
= epriv
->ppriv
;
451 mlx5_core_dbg(mdev
, "attaching QPN 0x%x, MGID %pI6\n", ipriv
->qp
.qpn
, gid
->raw
);
452 err
= mlx5_core_attach_mcg(mdev
, gid
, ipriv
->qp
.qpn
);
454 mlx5_core_warn(mdev
, "failed attaching QPN 0x%x, MGID %pI6\n",
455 ipriv
->qp
.qpn
, gid
->raw
);
458 mlx5_core_dbg(mdev
, "%s setting qkey 0x%x\n",
466 static int mlx5i_detach_mcast(struct net_device
*netdev
, struct ib_device
*hca
,
467 union ib_gid
*gid
, u16 lid
)
469 struct mlx5e_priv
*epriv
= mlx5i_epriv(netdev
);
470 struct mlx5_core_dev
*mdev
= epriv
->mdev
;
471 struct mlx5i_priv
*ipriv
= epriv
->ppriv
;
474 mlx5_core_dbg(mdev
, "detaching QPN 0x%x, MGID %pI6\n", ipriv
->qp
.qpn
, gid
->raw
);
476 err
= mlx5_core_detach_mcg(mdev
, gid
, ipriv
->qp
.qpn
);
478 mlx5_core_dbg(mdev
, "failed dettaching QPN 0x%x, MGID %pI6\n",
479 ipriv
->qp
.qpn
, gid
->raw
);
484 static int mlx5i_xmit(struct net_device
*dev
, struct sk_buff
*skb
,
485 struct ib_ah
*address
, u32 dqpn
)
487 struct mlx5e_priv
*epriv
= mlx5i_epriv(dev
);
488 struct mlx5e_txqsq
*sq
= epriv
->txq2sq
[skb_get_queue_mapping(skb
)];
489 struct mlx5_ib_ah
*mah
= to_mah(address
);
490 struct mlx5i_priv
*ipriv
= epriv
->ppriv
;
492 return mlx5i_sq_xmit(sq
, skb
, &mah
->av
, dqpn
, ipriv
->qkey
);
495 static int mlx5i_check_required_hca_cap(struct mlx5_core_dev
*mdev
)
497 if (MLX5_CAP_GEN(mdev
, port_type
) != MLX5_CAP_PORT_TYPE_IB
)
500 if (!MLX5_CAP_GEN(mdev
, ipoib_enhanced_offloads
)) {
501 mlx5_core_warn(mdev
, "IPoIB enhanced offloads are not supported\n");
508 struct net_device
*mlx5_rdma_netdev_alloc(struct mlx5_core_dev
*mdev
,
509 struct ib_device
*ibdev
,
511 void (*setup
)(struct net_device
*))
513 const struct mlx5e_profile
*profile
= &mlx5i_nic_profile
;
514 int nch
= profile
->max_nch(mdev
);
515 struct net_device
*netdev
;
516 struct mlx5i_priv
*ipriv
;
517 struct mlx5e_priv
*epriv
;
518 struct rdma_netdev
*rn
;
521 if (mlx5i_check_required_hca_cap(mdev
)) {
522 mlx5_core_warn(mdev
, "Accelerated mode is not supported\n");
523 return ERR_PTR(-EOPNOTSUPP
);
526 /* This function should only be called once per mdev */
527 err
= mlx5e_create_mdev_resources(mdev
);
531 netdev
= alloc_netdev_mqs(sizeof(struct mlx5i_priv
) + sizeof(struct mlx5e_priv
),
532 name
, NET_NAME_UNKNOWN
,
534 nch
* MLX5E_MAX_NUM_TC
,
537 mlx5_core_warn(mdev
, "alloc_netdev_mqs failed\n");
538 goto free_mdev_resources
;
541 ipriv
= netdev_priv(netdev
);
542 epriv
= mlx5i_epriv(netdev
);
544 epriv
->wq
= create_singlethread_workqueue("mlx5i");
546 goto err_free_netdev
;
548 profile
->init(mdev
, netdev
, profile
, ipriv
);
550 mlx5e_attach_netdev(epriv
);
551 netif_carrier_off(netdev
);
553 /* set rdma_netdev func pointers */
556 rn
->send
= mlx5i_xmit
;
557 rn
->attach_mcast
= mlx5i_attach_mcast
;
558 rn
->detach_mcast
= mlx5i_detach_mcast
;
565 mlx5e_destroy_mdev_resources(mdev
);
569 EXPORT_SYMBOL(mlx5_rdma_netdev_alloc
);
571 void mlx5_rdma_netdev_free(struct net_device
*netdev
)
573 struct mlx5e_priv
*priv
= mlx5i_epriv(netdev
);
574 const struct mlx5e_profile
*profile
= priv
->profile
;
576 mlx5e_detach_netdev(priv
);
577 profile
->cleanup(priv
);
578 destroy_workqueue(priv
->wq
);
581 mlx5e_destroy_mdev_resources(priv
->mdev
);
583 EXPORT_SYMBOL(mlx5_rdma_netdev_free
);