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_change_mtu(struct net_device *netdev, int new_mtu);
44 static int mlx5i_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd);
46 static const struct net_device_ops mlx5i_netdev_ops = {
47 .ndo_open = mlx5i_open,
48 .ndo_stop = mlx5i_close,
49 .ndo_init = mlx5i_dev_init,
50 .ndo_uninit = mlx5i_dev_cleanup,
51 .ndo_change_mtu = mlx5i_change_mtu,
52 .ndo_do_ioctl = mlx5i_ioctl,
55 /* IPoIB mlx5 netdev profile */
56 static void mlx5i_build_nic_params(struct mlx5_core_dev *mdev,
57 struct mlx5e_params *params)
59 /* Override RQ params as IPoIB supports only LINKED LIST RQ for now */
60 mlx5e_init_rq_type_params(mdev, params, MLX5_WQ_TYPE_LINKED_LIST);
62 /* RQ size in ipoib by default is 512 */
63 params->log_rq_size = is_kdump_kernel() ?
64 MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE :
65 MLX5I_PARAMS_DEFAULT_LOG_RQ_SIZE;
67 params->lro_en = false;
70 /* Called directly after IPoIB netdevice was created to initialize SW structs */
71 void mlx5i_init(struct mlx5_core_dev *mdev,
72 struct net_device *netdev,
73 const struct mlx5e_profile *profile,
76 struct mlx5e_priv *priv = mlx5i_epriv(netdev);
80 priv->netdev = netdev;
81 priv->profile = profile;
83 priv->hard_mtu = MLX5_IB_GRH_BYTES + MLX5_IPOIB_HARD_LEN;
84 mutex_init(&priv->state_lock);
86 mlx5e_build_nic_params(mdev, &priv->channels.params, profile->max_nch(mdev));
87 mlx5i_build_nic_params(mdev, &priv->channels.params);
89 mlx5e_timestamp_init(priv);
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 int mlx5i_init_underlay_qp(struct mlx5e_priv *priv)
113 struct mlx5_core_dev *mdev = priv->mdev;
114 struct mlx5i_priv *ipriv = priv->ppriv;
115 struct mlx5_core_qp *qp = &ipriv->qp;
116 struct mlx5_qp_context *context;
120 context = kzalloc(sizeof(*context), GFP_KERNEL);
124 context->flags = cpu_to_be32(MLX5_QP_PM_MIGRATED << 11);
125 context->pri_path.port = 1;
126 context->pri_path.pkey_index = cpu_to_be16(ipriv->pkey_index);
127 context->qkey = cpu_to_be32(IB_DEFAULT_Q_KEY);
129 ret = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_RST2INIT_QP, 0, context, qp);
131 mlx5_core_err(mdev, "Failed to modify qp RST2INIT, err: %d\n", ret);
132 goto err_qp_modify_to_err;
134 memset(context, 0, sizeof(*context));
136 ret = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_INIT2RTR_QP, 0, context, qp);
138 mlx5_core_err(mdev, "Failed to modify qp INIT2RTR, err: %d\n", ret);
139 goto err_qp_modify_to_err;
142 ret = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_RTR2RTS_QP, 0, context, qp);
144 mlx5_core_err(mdev, "Failed to modify qp RTR2RTS, err: %d\n", ret);
145 goto err_qp_modify_to_err;
151 err_qp_modify_to_err:
152 mlx5_core_qp_modify(mdev, MLX5_CMD_OP_2ERR_QP, 0, &context, qp);
157 void mlx5i_uninit_underlay_qp(struct mlx5e_priv *priv)
159 struct mlx5i_priv *ipriv = priv->ppriv;
160 struct mlx5_core_dev *mdev = priv->mdev;
161 struct mlx5_qp_context context;
164 err = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_2RST_QP, 0, &context,
167 mlx5_core_err(mdev, "Failed to modify qp 2RST, err: %d\n", err);
170 #define MLX5_QP_ENHANCED_ULP_STATELESS_MODE 2
172 int mlx5i_create_underlay_qp(struct mlx5_core_dev *mdev, struct mlx5_core_qp *qp)
180 inlen = MLX5_ST_SZ_BYTES(create_qp_in);
181 in = kvzalloc(inlen, GFP_KERNEL);
185 qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
186 MLX5_SET(qpc, qpc, st, MLX5_QP_ST_UD);
187 MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
188 MLX5_SET(qpc, qpc, ulp_stateless_offload_mode,
189 MLX5_QP_ENHANCED_ULP_STATELESS_MODE);
191 addr_path = MLX5_ADDR_OF(qpc, qpc, primary_address_path);
192 MLX5_SET(ads, addr_path, port, 1);
193 MLX5_SET(ads, addr_path, grh, 1);
195 ret = mlx5_core_create_qp(mdev, qp, in, inlen);
197 mlx5_core_err(mdev, "Failed creating IPoIB QP err : %d\n", ret);
206 void mlx5i_destroy_underlay_qp(struct mlx5_core_dev *mdev, struct mlx5_core_qp *qp)
208 mlx5_core_destroy_qp(mdev, qp);
211 static int mlx5i_init_tx(struct mlx5e_priv *priv)
213 struct mlx5i_priv *ipriv = priv->ppriv;
216 err = mlx5i_create_underlay_qp(priv->mdev, &ipriv->qp);
218 mlx5_core_warn(priv->mdev, "create underlay QP failed, %d\n", err);
222 err = mlx5e_create_tis(priv->mdev, 0 /* tc */, ipriv->qp.qpn, &priv->tisn[0]);
224 mlx5_core_warn(priv->mdev, "create tis failed, %d\n", err);
225 goto err_destroy_underlay_qp;
230 err_destroy_underlay_qp:
231 mlx5i_destroy_underlay_qp(priv->mdev, &ipriv->qp);
235 static void mlx5i_cleanup_tx(struct mlx5e_priv *priv)
237 struct mlx5i_priv *ipriv = priv->ppriv;
239 mlx5e_destroy_tis(priv->mdev, priv->tisn[0]);
240 mlx5i_destroy_underlay_qp(priv->mdev, &ipriv->qp);
243 static int mlx5i_create_flow_steering(struct mlx5e_priv *priv)
247 priv->fs.ns = mlx5_get_flow_namespace(priv->mdev,
248 MLX5_FLOW_NAMESPACE_KERNEL);
253 err = mlx5e_arfs_create_tables(priv);
255 netdev_err(priv->netdev, "Failed to create arfs tables, err=%d\n",
257 priv->netdev->hw_features &= ~NETIF_F_NTUPLE;
260 err = mlx5e_create_inner_ttc_table(priv);
262 netdev_err(priv->netdev, "Failed to create inner ttc table, err=%d\n",
264 goto err_destroy_arfs_tables;
267 err = mlx5e_create_ttc_table(priv);
269 netdev_err(priv->netdev, "Failed to create ttc table, err=%d\n",
271 goto err_destroy_inner_ttc_table;
276 err_destroy_inner_ttc_table:
277 mlx5e_destroy_inner_ttc_table(priv);
278 err_destroy_arfs_tables:
279 mlx5e_arfs_destroy_tables(priv);
284 static void mlx5i_destroy_flow_steering(struct mlx5e_priv *priv)
286 mlx5e_destroy_ttc_table(priv);
287 mlx5e_destroy_inner_ttc_table(priv);
288 mlx5e_arfs_destroy_tables(priv);
291 static int mlx5i_init_rx(struct mlx5e_priv *priv)
295 err = mlx5e_create_indirect_rqt(priv);
299 err = mlx5e_create_direct_rqts(priv);
301 goto err_destroy_indirect_rqts;
303 err = mlx5e_create_indirect_tirs(priv);
305 goto err_destroy_direct_rqts;
307 err = mlx5e_create_direct_tirs(priv);
309 goto err_destroy_indirect_tirs;
311 err = mlx5i_create_flow_steering(priv);
313 goto err_destroy_direct_tirs;
317 err_destroy_direct_tirs:
318 mlx5e_destroy_direct_tirs(priv);
319 err_destroy_indirect_tirs:
320 mlx5e_destroy_indirect_tirs(priv);
321 err_destroy_direct_rqts:
322 mlx5e_destroy_direct_rqts(priv);
323 err_destroy_indirect_rqts:
324 mlx5e_destroy_rqt(priv, &priv->indir_rqt);
328 static void mlx5i_cleanup_rx(struct mlx5e_priv *priv)
330 mlx5i_destroy_flow_steering(priv);
331 mlx5e_destroy_direct_tirs(priv);
332 mlx5e_destroy_indirect_tirs(priv);
333 mlx5e_destroy_direct_rqts(priv);
334 mlx5e_destroy_rqt(priv, &priv->indir_rqt);
337 static const struct mlx5e_profile mlx5i_nic_profile = {
339 .cleanup = mlx5i_cleanup,
340 .init_tx = mlx5i_init_tx,
341 .cleanup_tx = mlx5i_cleanup_tx,
342 .init_rx = mlx5i_init_rx,
343 .cleanup_rx = mlx5i_cleanup_rx,
344 .enable = NULL, /* mlx5i_enable */
345 .disable = NULL, /* mlx5i_disable */
346 .update_stats = NULL, /* mlx5i_update_stats */
347 .max_nch = mlx5e_get_max_num_channels,
348 .update_carrier = NULL, /* no HW update in IB link */
349 .rx_handlers.handle_rx_cqe = mlx5i_handle_rx_cqe,
350 .rx_handlers.handle_rx_cqe_mpwqe = NULL, /* Not supported */
351 .max_tc = MLX5I_MAX_NUM_TC,
354 /* mlx5i netdev NDos */
356 static int mlx5i_change_mtu(struct net_device *netdev, int new_mtu)
358 struct mlx5e_priv *priv = mlx5i_epriv(netdev);
359 struct mlx5e_channels new_channels = {};
363 mutex_lock(&priv->state_lock);
365 curr_mtu = netdev->mtu;
366 netdev->mtu = new_mtu;
368 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
371 new_channels.params = priv->channels.params;
372 err = mlx5e_open_channels(priv, &new_channels);
374 netdev->mtu = curr_mtu;
378 mlx5e_switch_priv_channels(priv, &new_channels, NULL);
381 mutex_unlock(&priv->state_lock);
385 int mlx5i_dev_init(struct net_device *dev)
387 struct mlx5e_priv *priv = mlx5i_epriv(dev);
388 struct mlx5i_priv *ipriv = priv->ppriv;
390 /* Set dev address using underlay QP */
391 dev->dev_addr[1] = (ipriv->qp.qpn >> 16) & 0xff;
392 dev->dev_addr[2] = (ipriv->qp.qpn >> 8) & 0xff;
393 dev->dev_addr[3] = (ipriv->qp.qpn) & 0xff;
395 /* Add QPN to net-device mapping to HT */
396 mlx5i_pkey_add_qpn(dev ,ipriv->qp.qpn);
401 static int mlx5i_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
403 struct mlx5e_priv *priv = mlx5i_epriv(dev);
407 return mlx5e_hwstamp_set(priv, ifr);
409 return mlx5e_hwstamp_get(priv, ifr);
415 void mlx5i_dev_cleanup(struct net_device *dev)
417 struct mlx5e_priv *priv = mlx5i_epriv(dev);
418 struct mlx5i_priv *ipriv = priv->ppriv;
420 mlx5i_uninit_underlay_qp(priv);
422 /* Delete QPN to net-device mapping from HT */
423 mlx5i_pkey_del_qpn(dev, ipriv->qp.qpn);
426 static int mlx5i_open(struct net_device *netdev)
428 struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
429 struct mlx5i_priv *ipriv = epriv->ppriv;
430 struct mlx5_core_dev *mdev = epriv->mdev;
433 mutex_lock(&epriv->state_lock);
435 set_bit(MLX5E_STATE_OPENED, &epriv->state);
437 err = mlx5i_init_underlay_qp(epriv);
439 mlx5_core_warn(mdev, "prepare underlay qp state failed, %d\n", err);
440 goto err_clear_state_opened_flag;
443 err = mlx5_fs_add_rx_underlay_qpn(mdev, ipriv->qp.qpn);
445 mlx5_core_warn(mdev, "attach underlay qp to ft failed, %d\n", err);
449 err = mlx5e_open_channels(epriv, &epriv->channels);
451 goto err_remove_fs_underlay_qp;
453 mlx5e_refresh_tirs(epriv, false);
454 mlx5e_activate_priv_channels(epriv);
456 mutex_unlock(&epriv->state_lock);
459 err_remove_fs_underlay_qp:
460 mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qp.qpn);
462 mlx5i_uninit_underlay_qp(epriv);
463 err_clear_state_opened_flag:
464 clear_bit(MLX5E_STATE_OPENED, &epriv->state);
465 mutex_unlock(&epriv->state_lock);
469 static int mlx5i_close(struct net_device *netdev)
471 struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
472 struct mlx5i_priv *ipriv = epriv->ppriv;
473 struct mlx5_core_dev *mdev = epriv->mdev;
475 /* May already be CLOSED in case a previous configuration operation
476 * (e.g RX/TX queue size change) that involves close&open failed.
478 mutex_lock(&epriv->state_lock);
480 if (!test_bit(MLX5E_STATE_OPENED, &epriv->state))
483 clear_bit(MLX5E_STATE_OPENED, &epriv->state);
485 netif_carrier_off(epriv->netdev);
486 mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qp.qpn);
487 mlx5i_uninit_underlay_qp(epriv);
488 mlx5e_deactivate_priv_channels(epriv);
489 mlx5e_close_channels(&epriv->channels);;
491 mutex_unlock(&epriv->state_lock);
495 /* IPoIB RDMA netdev callbacks */
496 static int mlx5i_attach_mcast(struct net_device *netdev, struct ib_device *hca,
497 union ib_gid *gid, u16 lid, int set_qkey,
500 struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
501 struct mlx5_core_dev *mdev = epriv->mdev;
502 struct mlx5i_priv *ipriv = epriv->ppriv;
505 mlx5_core_dbg(mdev, "attaching QPN 0x%x, MGID %pI6\n", ipriv->qp.qpn, gid->raw);
506 err = mlx5_core_attach_mcg(mdev, gid, ipriv->qp.qpn);
508 mlx5_core_warn(mdev, "failed attaching QPN 0x%x, MGID %pI6\n",
509 ipriv->qp.qpn, gid->raw);
512 mlx5_core_dbg(mdev, "%s setting qkey 0x%x\n",
520 static int mlx5i_detach_mcast(struct net_device *netdev, struct ib_device *hca,
521 union ib_gid *gid, u16 lid)
523 struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
524 struct mlx5_core_dev *mdev = epriv->mdev;
525 struct mlx5i_priv *ipriv = epriv->ppriv;
528 mlx5_core_dbg(mdev, "detaching QPN 0x%x, MGID %pI6\n", ipriv->qp.qpn, gid->raw);
530 err = mlx5_core_detach_mcg(mdev, gid, ipriv->qp.qpn);
532 mlx5_core_dbg(mdev, "failed dettaching QPN 0x%x, MGID %pI6\n",
533 ipriv->qp.qpn, gid->raw);
538 static int mlx5i_xmit(struct net_device *dev, struct sk_buff *skb,
539 struct ib_ah *address, u32 dqpn)
541 struct mlx5e_priv *epriv = mlx5i_epriv(dev);
542 struct mlx5e_txqsq *sq = epriv->txq2sq[skb_get_queue_mapping(skb)];
543 struct mlx5_ib_ah *mah = to_mah(address);
544 struct mlx5i_priv *ipriv = epriv->ppriv;
546 return mlx5i_sq_xmit(sq, skb, &mah->av, dqpn, ipriv->qkey);
549 static void mlx5i_set_pkey_index(struct net_device *netdev, int id)
551 struct mlx5i_priv *ipriv = netdev_priv(netdev);
553 ipriv->pkey_index = (u16)id;
556 static int mlx5i_check_required_hca_cap(struct mlx5_core_dev *mdev)
558 if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_IB)
561 if (!MLX5_CAP_GEN(mdev, ipoib_enhanced_offloads)) {
562 mlx5_core_warn(mdev, "IPoIB enhanced offloads are not supported\n");
569 struct net_device *mlx5_rdma_netdev_alloc(struct mlx5_core_dev *mdev,
570 struct ib_device *ibdev,
572 void (*setup)(struct net_device *))
574 const struct mlx5e_profile *profile;
575 struct net_device *netdev;
576 struct mlx5i_priv *ipriv;
577 struct mlx5e_priv *epriv;
578 struct rdma_netdev *rn;
583 if (mlx5i_check_required_hca_cap(mdev)) {
584 mlx5_core_warn(mdev, "Accelerated mode is not supported\n");
585 return ERR_PTR(-EOPNOTSUPP);
588 /* TODO: Need to find a better way to check if child device*/
589 sub_interface = (mdev->mlx5e_res.pdn != 0);
592 profile = mlx5i_pkey_get_profile();
594 profile = &mlx5i_nic_profile;
596 nch = profile->max_nch(mdev);
598 netdev = alloc_netdev_mqs(sizeof(struct mlx5i_priv) + sizeof(struct mlx5e_priv),
599 name, NET_NAME_UNKNOWN,
601 nch * MLX5E_MAX_NUM_TC,
604 mlx5_core_warn(mdev, "alloc_netdev_mqs failed\n");
608 ipriv = netdev_priv(netdev);
609 epriv = mlx5i_epriv(netdev);
611 epriv->wq = create_singlethread_workqueue("mlx5i");
613 goto err_free_netdev;
615 ipriv->sub_interface = sub_interface;
616 if (!ipriv->sub_interface) {
617 err = mlx5i_pkey_qpn_ht_init(netdev);
619 mlx5_core_warn(mdev, "allocate qpn_to_netdev ht failed\n");
623 /* This should only be called once per mdev */
624 err = mlx5e_create_mdev_resources(mdev);
629 profile->init(mdev, netdev, profile, ipriv);
631 mlx5e_attach_netdev(epriv);
632 netif_carrier_off(netdev);
634 /* set rdma_netdev func pointers */
637 rn->send = mlx5i_xmit;
638 rn->attach_mcast = mlx5i_attach_mcast;
639 rn->detach_mcast = mlx5i_detach_mcast;
640 rn->set_id = mlx5i_set_pkey_index;
645 mlx5i_pkey_qpn_ht_cleanup(netdev);
647 destroy_workqueue(epriv->wq);
653 EXPORT_SYMBOL(mlx5_rdma_netdev_alloc);
655 void mlx5_rdma_netdev_free(struct net_device *netdev)
657 struct mlx5e_priv *priv = mlx5i_epriv(netdev);
658 struct mlx5i_priv *ipriv = priv->ppriv;
659 const struct mlx5e_profile *profile = priv->profile;
661 mlx5e_detach_netdev(priv);
662 profile->cleanup(priv);
663 destroy_workqueue(priv->wq);
665 if (!ipriv->sub_interface) {
666 mlx5i_pkey_qpn_ht_cleanup(netdev);
667 mlx5e_destroy_mdev_resources(priv->mdev);
671 EXPORT_SYMBOL(mlx5_rdma_netdev_free);