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);
45 static const struct net_device_ops mlx5i_netdev_ops = {
46 .ndo_open = mlx5i_open,
47 .ndo_stop = mlx5i_close,
48 .ndo_init = mlx5i_dev_init,
49 .ndo_uninit = mlx5i_dev_cleanup,
50 .ndo_change_mtu = mlx5i_change_mtu,
51 .ndo_do_ioctl = mlx5i_ioctl,
54 /* IPoIB mlx5 netdev profile */
55 static void mlx5i_build_nic_params(struct mlx5_core_dev *mdev,
56 struct mlx5e_params *params)
58 /* Override RQ params as IPoIB supports only LINKED LIST RQ for now */
59 MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_STRIDING_RQ, false);
60 mlx5e_set_rq_type(mdev, params);
61 mlx5e_init_rq_type_params(mdev, params);
63 /* RQ size in ipoib by default is 512 */
64 params->log_rq_mtu_frames = is_kdump_kernel() ?
65 MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE :
66 MLX5I_PARAMS_DEFAULT_LOG_RQ_SIZE;
68 params->lro_en = false;
69 params->hard_mtu = MLX5_IB_GRH_BYTES + MLX5_IPOIB_HARD_LEN;
72 /* Called directly after IPoIB netdevice was created to initialize SW structs */
73 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 mutex_init(&priv->state_lock);
87 mlx5e_build_nic_params(mdev, &priv->channels.params,
88 profile->max_nch(mdev), netdev->mtu);
89 mlx5i_build_nic_params(mdev, &priv->channels.params);
91 mlx5e_timestamp_init(priv);
94 netdev->hw_features |= NETIF_F_SG;
95 netdev->hw_features |= NETIF_F_IP_CSUM;
96 netdev->hw_features |= NETIF_F_IPV6_CSUM;
97 netdev->hw_features |= NETIF_F_GRO;
98 netdev->hw_features |= NETIF_F_TSO;
99 netdev->hw_features |= NETIF_F_TSO6;
100 netdev->hw_features |= NETIF_F_RXCSUM;
101 netdev->hw_features |= NETIF_F_RXHASH;
103 netdev->netdev_ops = &mlx5i_netdev_ops;
104 netdev->ethtool_ops = &mlx5i_ethtool_ops;
107 /* Called directly before IPoIB netdevice is destroyed to cleanup SW structs */
108 static void mlx5i_cleanup(struct mlx5e_priv *priv)
113 int mlx5i_init_underlay_qp(struct mlx5e_priv *priv)
115 struct mlx5_core_dev *mdev = priv->mdev;
116 struct mlx5i_priv *ipriv = priv->ppriv;
117 struct mlx5_core_qp *qp = &ipriv->qp;
118 struct mlx5_qp_context *context;
122 context = kzalloc(sizeof(*context), GFP_KERNEL);
126 context->flags = cpu_to_be32(MLX5_QP_PM_MIGRATED << 11);
127 context->pri_path.port = 1;
128 context->pri_path.pkey_index = cpu_to_be16(ipriv->pkey_index);
129 context->qkey = cpu_to_be32(IB_DEFAULT_Q_KEY);
131 ret = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_RST2INIT_QP, 0, context, qp);
133 mlx5_core_err(mdev, "Failed to modify qp RST2INIT, err: %d\n", ret);
134 goto err_qp_modify_to_err;
136 memset(context, 0, sizeof(*context));
138 ret = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_INIT2RTR_QP, 0, context, qp);
140 mlx5_core_err(mdev, "Failed to modify qp INIT2RTR, err: %d\n", ret);
141 goto err_qp_modify_to_err;
144 ret = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_RTR2RTS_QP, 0, context, qp);
146 mlx5_core_err(mdev, "Failed to modify qp RTR2RTS, err: %d\n", ret);
147 goto err_qp_modify_to_err;
153 err_qp_modify_to_err:
154 mlx5_core_qp_modify(mdev, MLX5_CMD_OP_2ERR_QP, 0, &context, qp);
159 void mlx5i_uninit_underlay_qp(struct mlx5e_priv *priv)
161 struct mlx5i_priv *ipriv = priv->ppriv;
162 struct mlx5_core_dev *mdev = priv->mdev;
163 struct mlx5_qp_context context;
166 err = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_2RST_QP, 0, &context,
169 mlx5_core_err(mdev, "Failed to modify qp 2RST, err: %d\n", err);
172 #define MLX5_QP_ENHANCED_ULP_STATELESS_MODE 2
174 int mlx5i_create_underlay_qp(struct mlx5_core_dev *mdev, struct mlx5_core_qp *qp)
182 inlen = MLX5_ST_SZ_BYTES(create_qp_in);
183 in = kvzalloc(inlen, GFP_KERNEL);
187 qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
188 MLX5_SET(qpc, qpc, st, MLX5_QP_ST_UD);
189 MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
190 MLX5_SET(qpc, qpc, ulp_stateless_offload_mode,
191 MLX5_QP_ENHANCED_ULP_STATELESS_MODE);
193 addr_path = MLX5_ADDR_OF(qpc, qpc, primary_address_path);
194 MLX5_SET(ads, addr_path, vhca_port_num, 1);
195 MLX5_SET(ads, addr_path, grh, 1);
197 ret = mlx5_core_create_qp(mdev, qp, in, inlen);
199 mlx5_core_err(mdev, "Failed creating IPoIB QP err : %d\n", ret);
208 void mlx5i_destroy_underlay_qp(struct mlx5_core_dev *mdev, struct mlx5_core_qp *qp)
210 mlx5_core_destroy_qp(mdev, qp);
213 static int mlx5i_init_tx(struct mlx5e_priv *priv)
215 struct mlx5i_priv *ipriv = priv->ppriv;
218 err = mlx5i_create_underlay_qp(priv->mdev, &ipriv->qp);
220 mlx5_core_warn(priv->mdev, "create underlay QP failed, %d\n", err);
224 err = mlx5e_create_tis(priv->mdev, 0 /* tc */, ipriv->qp.qpn, &priv->tisn[0]);
226 mlx5_core_warn(priv->mdev, "create tis failed, %d\n", err);
227 goto err_destroy_underlay_qp;
232 err_destroy_underlay_qp:
233 mlx5i_destroy_underlay_qp(priv->mdev, &ipriv->qp);
237 static void mlx5i_cleanup_tx(struct mlx5e_priv *priv)
239 struct mlx5i_priv *ipriv = priv->ppriv;
241 mlx5e_destroy_tis(priv->mdev, priv->tisn[0]);
242 mlx5i_destroy_underlay_qp(priv->mdev, &ipriv->qp);
245 static int mlx5i_create_flow_steering(struct mlx5e_priv *priv)
247 struct ttc_params ttc_params = {};
250 priv->fs.ns = mlx5_get_flow_namespace(priv->mdev,
251 MLX5_FLOW_NAMESPACE_KERNEL);
256 err = mlx5e_arfs_create_tables(priv);
258 netdev_err(priv->netdev, "Failed to create arfs tables, err=%d\n",
260 priv->netdev->hw_features &= ~NETIF_F_NTUPLE;
263 mlx5e_set_ttc_basic_params(priv, &ttc_params);
264 mlx5e_set_inner_ttc_ft_params(&ttc_params);
265 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
266 ttc_params.indir_tirn[tt] = priv->inner_indir_tir[tt].tirn;
268 err = mlx5e_create_inner_ttc_table(priv, &ttc_params, &priv->fs.inner_ttc);
270 netdev_err(priv->netdev, "Failed to create inner ttc table, err=%d\n",
272 goto err_destroy_arfs_tables;
275 mlx5e_set_ttc_ft_params(&ttc_params);
276 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
277 ttc_params.indir_tirn[tt] = priv->indir_tir[tt].tirn;
279 err = mlx5e_create_ttc_table(priv, &ttc_params, &priv->fs.ttc);
281 netdev_err(priv->netdev, "Failed to create ttc table, err=%d\n",
283 goto err_destroy_inner_ttc_table;
288 err_destroy_inner_ttc_table:
289 mlx5e_destroy_inner_ttc_table(priv, &priv->fs.inner_ttc);
290 err_destroy_arfs_tables:
291 mlx5e_arfs_destroy_tables(priv);
296 static void mlx5i_destroy_flow_steering(struct mlx5e_priv *priv)
298 mlx5e_destroy_ttc_table(priv, &priv->fs.ttc);
299 mlx5e_destroy_inner_ttc_table(priv, &priv->fs.inner_ttc);
300 mlx5e_arfs_destroy_tables(priv);
303 static int mlx5i_init_rx(struct mlx5e_priv *priv)
307 err = mlx5e_create_indirect_rqt(priv);
311 err = mlx5e_create_direct_rqts(priv);
313 goto err_destroy_indirect_rqts;
315 err = mlx5e_create_indirect_tirs(priv);
317 goto err_destroy_direct_rqts;
319 err = mlx5e_create_direct_tirs(priv);
321 goto err_destroy_indirect_tirs;
323 err = mlx5i_create_flow_steering(priv);
325 goto err_destroy_direct_tirs;
329 err_destroy_direct_tirs:
330 mlx5e_destroy_direct_tirs(priv);
331 err_destroy_indirect_tirs:
332 mlx5e_destroy_indirect_tirs(priv);
333 err_destroy_direct_rqts:
334 mlx5e_destroy_direct_rqts(priv);
335 err_destroy_indirect_rqts:
336 mlx5e_destroy_rqt(priv, &priv->indir_rqt);
340 static void mlx5i_cleanup_rx(struct mlx5e_priv *priv)
342 mlx5i_destroy_flow_steering(priv);
343 mlx5e_destroy_direct_tirs(priv);
344 mlx5e_destroy_indirect_tirs(priv);
345 mlx5e_destroy_direct_rqts(priv);
346 mlx5e_destroy_rqt(priv, &priv->indir_rqt);
349 static const struct mlx5e_profile mlx5i_nic_profile = {
351 .cleanup = mlx5i_cleanup,
352 .init_tx = mlx5i_init_tx,
353 .cleanup_tx = mlx5i_cleanup_tx,
354 .init_rx = mlx5i_init_rx,
355 .cleanup_rx = mlx5i_cleanup_rx,
356 .enable = NULL, /* mlx5i_enable */
357 .disable = NULL, /* mlx5i_disable */
358 .update_stats = NULL, /* mlx5i_update_stats */
359 .max_nch = mlx5e_get_max_num_channels,
360 .update_carrier = NULL, /* no HW update in IB link */
361 .rx_handlers.handle_rx_cqe = mlx5i_handle_rx_cqe,
362 .rx_handlers.handle_rx_cqe_mpwqe = NULL, /* Not supported */
363 .max_tc = MLX5I_MAX_NUM_TC,
366 /* mlx5i netdev NDos */
368 static int mlx5i_change_mtu(struct net_device *netdev, int new_mtu)
370 struct mlx5e_priv *priv = mlx5i_epriv(netdev);
371 struct mlx5e_channels new_channels = {};
372 struct mlx5e_params *params;
375 mutex_lock(&priv->state_lock);
377 params = &priv->channels.params;
379 if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
380 params->sw_mtu = new_mtu;
381 netdev->mtu = params->sw_mtu;
385 new_channels.params = *params;
386 new_channels.params.sw_mtu = new_mtu;
387 err = mlx5e_open_channels(priv, &new_channels);
391 mlx5e_switch_priv_channels(priv, &new_channels, NULL);
392 netdev->mtu = new_channels.params.sw_mtu;
395 mutex_unlock(&priv->state_lock);
399 int mlx5i_dev_init(struct net_device *dev)
401 struct mlx5e_priv *priv = mlx5i_epriv(dev);
402 struct mlx5i_priv *ipriv = priv->ppriv;
404 /* Set dev address using underlay QP */
405 dev->dev_addr[1] = (ipriv->qp.qpn >> 16) & 0xff;
406 dev->dev_addr[2] = (ipriv->qp.qpn >> 8) & 0xff;
407 dev->dev_addr[3] = (ipriv->qp.qpn) & 0xff;
409 /* Add QPN to net-device mapping to HT */
410 mlx5i_pkey_add_qpn(dev ,ipriv->qp.qpn);
415 int mlx5i_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
417 struct mlx5e_priv *priv = mlx5i_epriv(dev);
421 return mlx5e_hwstamp_set(priv, ifr);
423 return mlx5e_hwstamp_get(priv, ifr);
429 void mlx5i_dev_cleanup(struct net_device *dev)
431 struct mlx5e_priv *priv = mlx5i_epriv(dev);
432 struct mlx5i_priv *ipriv = priv->ppriv;
434 mlx5i_uninit_underlay_qp(priv);
436 /* Delete QPN to net-device mapping from HT */
437 mlx5i_pkey_del_qpn(dev, ipriv->qp.qpn);
440 static int mlx5i_open(struct net_device *netdev)
442 struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
443 struct mlx5i_priv *ipriv = epriv->ppriv;
444 struct mlx5_core_dev *mdev = epriv->mdev;
447 mutex_lock(&epriv->state_lock);
449 set_bit(MLX5E_STATE_OPENED, &epriv->state);
451 err = mlx5i_init_underlay_qp(epriv);
453 mlx5_core_warn(mdev, "prepare underlay qp state failed, %d\n", err);
454 goto err_clear_state_opened_flag;
457 err = mlx5_fs_add_rx_underlay_qpn(mdev, ipriv->qp.qpn);
459 mlx5_core_warn(mdev, "attach underlay qp to ft failed, %d\n", err);
463 err = mlx5e_open_channels(epriv, &epriv->channels);
465 goto err_remove_fs_underlay_qp;
467 mlx5e_refresh_tirs(epriv, false);
468 mlx5e_activate_priv_channels(epriv);
470 mutex_unlock(&epriv->state_lock);
473 err_remove_fs_underlay_qp:
474 mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qp.qpn);
476 mlx5i_uninit_underlay_qp(epriv);
477 err_clear_state_opened_flag:
478 clear_bit(MLX5E_STATE_OPENED, &epriv->state);
479 mutex_unlock(&epriv->state_lock);
483 static int mlx5i_close(struct net_device *netdev)
485 struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
486 struct mlx5i_priv *ipriv = epriv->ppriv;
487 struct mlx5_core_dev *mdev = epriv->mdev;
489 /* May already be CLOSED in case a previous configuration operation
490 * (e.g RX/TX queue size change) that involves close&open failed.
492 mutex_lock(&epriv->state_lock);
494 if (!test_bit(MLX5E_STATE_OPENED, &epriv->state))
497 clear_bit(MLX5E_STATE_OPENED, &epriv->state);
499 netif_carrier_off(epriv->netdev);
500 mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qp.qpn);
501 mlx5i_uninit_underlay_qp(epriv);
502 mlx5e_deactivate_priv_channels(epriv);
503 mlx5e_close_channels(&epriv->channels);
505 mutex_unlock(&epriv->state_lock);
509 /* IPoIB RDMA netdev callbacks */
510 static int mlx5i_attach_mcast(struct net_device *netdev, struct ib_device *hca,
511 union ib_gid *gid, u16 lid, int set_qkey,
514 struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
515 struct mlx5_core_dev *mdev = epriv->mdev;
516 struct mlx5i_priv *ipriv = epriv->ppriv;
519 mlx5_core_dbg(mdev, "attaching QPN 0x%x, MGID %pI6\n", ipriv->qp.qpn, gid->raw);
520 err = mlx5_core_attach_mcg(mdev, gid, ipriv->qp.qpn);
522 mlx5_core_warn(mdev, "failed attaching QPN 0x%x, MGID %pI6\n",
523 ipriv->qp.qpn, gid->raw);
526 mlx5_core_dbg(mdev, "%s setting qkey 0x%x\n",
534 static int mlx5i_detach_mcast(struct net_device *netdev, struct ib_device *hca,
535 union ib_gid *gid, u16 lid)
537 struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
538 struct mlx5_core_dev *mdev = epriv->mdev;
539 struct mlx5i_priv *ipriv = epriv->ppriv;
542 mlx5_core_dbg(mdev, "detaching QPN 0x%x, MGID %pI6\n", ipriv->qp.qpn, gid->raw);
544 err = mlx5_core_detach_mcg(mdev, gid, ipriv->qp.qpn);
546 mlx5_core_dbg(mdev, "failed detaching QPN 0x%x, MGID %pI6\n",
547 ipriv->qp.qpn, gid->raw);
552 static int mlx5i_xmit(struct net_device *dev, struct sk_buff *skb,
553 struct ib_ah *address, u32 dqpn)
555 struct mlx5e_priv *epriv = mlx5i_epriv(dev);
556 struct mlx5e_txqsq *sq = epriv->txq2sq[skb_get_queue_mapping(skb)];
557 struct mlx5_ib_ah *mah = to_mah(address);
558 struct mlx5i_priv *ipriv = epriv->ppriv;
560 return mlx5i_sq_xmit(sq, skb, &mah->av, dqpn, ipriv->qkey);
563 static void mlx5i_set_pkey_index(struct net_device *netdev, int id)
565 struct mlx5i_priv *ipriv = netdev_priv(netdev);
567 ipriv->pkey_index = (u16)id;
570 static int mlx5i_check_required_hca_cap(struct mlx5_core_dev *mdev)
572 if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_IB)
575 if (!MLX5_CAP_GEN(mdev, ipoib_enhanced_offloads)) {
576 mlx5_core_warn(mdev, "IPoIB enhanced offloads are not supported\n");
583 struct net_device *mlx5_rdma_netdev_alloc(struct mlx5_core_dev *mdev,
584 struct ib_device *ibdev,
586 void (*setup)(struct net_device *))
588 const struct mlx5e_profile *profile;
589 struct net_device *netdev;
590 struct mlx5i_priv *ipriv;
591 struct mlx5e_priv *epriv;
592 struct rdma_netdev *rn;
597 if (mlx5i_check_required_hca_cap(mdev)) {
598 mlx5_core_warn(mdev, "Accelerated mode is not supported\n");
599 return ERR_PTR(-EOPNOTSUPP);
602 /* TODO: Need to find a better way to check if child device*/
603 sub_interface = (mdev->mlx5e_res.pdn != 0);
606 profile = mlx5i_pkey_get_profile();
608 profile = &mlx5i_nic_profile;
610 nch = profile->max_nch(mdev);
612 netdev = alloc_netdev_mqs(sizeof(struct mlx5i_priv) + sizeof(struct mlx5e_priv),
613 name, NET_NAME_UNKNOWN,
615 nch * MLX5E_MAX_NUM_TC,
618 mlx5_core_warn(mdev, "alloc_netdev_mqs failed\n");
622 ipriv = netdev_priv(netdev);
623 epriv = mlx5i_epriv(netdev);
625 epriv->wq = create_singlethread_workqueue("mlx5i");
627 goto err_free_netdev;
629 ipriv->sub_interface = sub_interface;
630 if (!ipriv->sub_interface) {
631 err = mlx5i_pkey_qpn_ht_init(netdev);
633 mlx5_core_warn(mdev, "allocate qpn_to_netdev ht failed\n");
637 /* This should only be called once per mdev */
638 err = mlx5e_create_mdev_resources(mdev);
643 profile->init(mdev, netdev, profile, ipriv);
645 mlx5e_attach_netdev(epriv);
646 netif_carrier_off(netdev);
648 /* set rdma_netdev func pointers */
651 rn->send = mlx5i_xmit;
652 rn->attach_mcast = mlx5i_attach_mcast;
653 rn->detach_mcast = mlx5i_detach_mcast;
654 rn->set_id = mlx5i_set_pkey_index;
659 mlx5i_pkey_qpn_ht_cleanup(netdev);
661 destroy_workqueue(epriv->wq);
667 EXPORT_SYMBOL(mlx5_rdma_netdev_alloc);
669 void mlx5_rdma_netdev_free(struct net_device *netdev)
671 struct mlx5e_priv *priv = mlx5i_epriv(netdev);
672 struct mlx5i_priv *ipriv = priv->ppriv;
673 const struct mlx5e_profile *profile = priv->profile;
675 mlx5e_detach_netdev(priv);
676 profile->cleanup(priv);
677 destroy_workqueue(priv->wq);
679 if (!ipriv->sub_interface) {
680 mlx5i_pkey_qpn_ht_cleanup(netdev);
681 mlx5e_destroy_mdev_resources(priv->mdev);
685 EXPORT_SYMBOL(mlx5_rdma_netdev_free);