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_init_rq_type_params(mdev, params, MLX5_WQ_TYPE_LINKED_LIST);
61 /* RQ size in ipoib by default is 512 */
62 params->log_rq_size = is_kdump_kernel() ?
63 MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE :
64 MLX5I_PARAMS_DEFAULT_LOG_RQ_SIZE;
66 params->lro_en = false;
69 /* Called directly after IPoIB netdevice was created to initialize SW structs */
70 void mlx5i_init(struct mlx5_core_dev *mdev,
71 struct net_device *netdev,
72 const struct mlx5e_profile *profile,
75 struct mlx5e_priv *priv = mlx5i_epriv(netdev);
79 priv->netdev = netdev;
80 priv->profile = profile;
82 priv->hard_mtu = MLX5_IB_GRH_BYTES + MLX5_IPOIB_HARD_LEN;
83 mutex_init(&priv->state_lock);
85 mlx5e_build_nic_params(mdev, &priv->channels.params, profile->max_nch(mdev));
86 mlx5i_build_nic_params(mdev, &priv->channels.params);
88 mlx5e_timestamp_init(priv);
91 netdev->hw_features |= NETIF_F_SG;
92 netdev->hw_features |= NETIF_F_IP_CSUM;
93 netdev->hw_features |= NETIF_F_IPV6_CSUM;
94 netdev->hw_features |= NETIF_F_GRO;
95 netdev->hw_features |= NETIF_F_TSO;
96 netdev->hw_features |= NETIF_F_TSO6;
97 netdev->hw_features |= NETIF_F_RXCSUM;
98 netdev->hw_features |= NETIF_F_RXHASH;
100 netdev->netdev_ops = &mlx5i_netdev_ops;
101 netdev->ethtool_ops = &mlx5i_ethtool_ops;
104 /* Called directly before IPoIB netdevice is destroyed to cleanup SW structs */
105 static void mlx5i_cleanup(struct mlx5e_priv *priv)
110 int mlx5i_init_underlay_qp(struct mlx5e_priv *priv)
112 struct mlx5_core_dev *mdev = priv->mdev;
113 struct mlx5i_priv *ipriv = priv->ppriv;
114 struct mlx5_core_qp *qp = &ipriv->qp;
115 struct mlx5_qp_context *context;
119 context = kzalloc(sizeof(*context), GFP_KERNEL);
123 context->flags = cpu_to_be32(MLX5_QP_PM_MIGRATED << 11);
124 context->pri_path.port = 1;
125 context->pri_path.pkey_index = cpu_to_be16(ipriv->pkey_index);
126 context->qkey = cpu_to_be32(IB_DEFAULT_Q_KEY);
128 ret = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_RST2INIT_QP, 0, context, qp);
130 mlx5_core_err(mdev, "Failed to modify qp RST2INIT, err: %d\n", ret);
131 goto err_qp_modify_to_err;
133 memset(context, 0, sizeof(*context));
135 ret = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_INIT2RTR_QP, 0, context, qp);
137 mlx5_core_err(mdev, "Failed to modify qp INIT2RTR, err: %d\n", ret);
138 goto err_qp_modify_to_err;
141 ret = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_RTR2RTS_QP, 0, context, qp);
143 mlx5_core_err(mdev, "Failed to modify qp RTR2RTS, err: %d\n", ret);
144 goto err_qp_modify_to_err;
150 err_qp_modify_to_err:
151 mlx5_core_qp_modify(mdev, MLX5_CMD_OP_2ERR_QP, 0, &context, qp);
156 void mlx5i_uninit_underlay_qp(struct mlx5e_priv *priv)
158 struct mlx5i_priv *ipriv = priv->ppriv;
159 struct mlx5_core_dev *mdev = priv->mdev;
160 struct mlx5_qp_context context;
163 err = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_2RST_QP, 0, &context,
166 mlx5_core_err(mdev, "Failed to modify qp 2RST, err: %d\n", err);
169 #define MLX5_QP_ENHANCED_ULP_STATELESS_MODE 2
171 int mlx5i_create_underlay_qp(struct mlx5_core_dev *mdev, struct mlx5_core_qp *qp)
179 inlen = MLX5_ST_SZ_BYTES(create_qp_in);
180 in = kvzalloc(inlen, GFP_KERNEL);
184 qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
185 MLX5_SET(qpc, qpc, st, MLX5_QP_ST_UD);
186 MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
187 MLX5_SET(qpc, qpc, ulp_stateless_offload_mode,
188 MLX5_QP_ENHANCED_ULP_STATELESS_MODE);
190 addr_path = MLX5_ADDR_OF(qpc, qpc, primary_address_path);
191 MLX5_SET(ads, addr_path, port, 1);
192 MLX5_SET(ads, addr_path, grh, 1);
194 ret = mlx5_core_create_qp(mdev, qp, in, inlen);
196 mlx5_core_err(mdev, "Failed creating IPoIB QP err : %d\n", ret);
205 void mlx5i_destroy_underlay_qp(struct mlx5_core_dev *mdev, struct mlx5_core_qp *qp)
207 mlx5_core_destroy_qp(mdev, qp);
210 static int mlx5i_init_tx(struct mlx5e_priv *priv)
212 struct mlx5i_priv *ipriv = priv->ppriv;
215 err = mlx5i_create_underlay_qp(priv->mdev, &ipriv->qp);
217 mlx5_core_warn(priv->mdev, "create underlay QP failed, %d\n", err);
221 err = mlx5e_create_tis(priv->mdev, 0 /* tc */, ipriv->qp.qpn, &priv->tisn[0]);
223 mlx5_core_warn(priv->mdev, "create tis failed, %d\n", err);
224 goto err_destroy_underlay_qp;
229 err_destroy_underlay_qp:
230 mlx5i_destroy_underlay_qp(priv->mdev, &ipriv->qp);
234 static void mlx5i_cleanup_tx(struct mlx5e_priv *priv)
236 struct mlx5i_priv *ipriv = priv->ppriv;
238 mlx5e_destroy_tis(priv->mdev, priv->tisn[0]);
239 mlx5i_destroy_underlay_qp(priv->mdev, &ipriv->qp);
242 static int mlx5i_create_flow_steering(struct mlx5e_priv *priv)
244 struct ttc_params ttc_params = {};
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 mlx5e_set_ttc_basic_params(priv, &ttc_params);
261 mlx5e_set_inner_ttc_ft_params(&ttc_params);
262 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
263 ttc_params.indir_tirn[tt] = priv->inner_indir_tir[tt].tirn;
265 err = mlx5e_create_inner_ttc_table(priv, &ttc_params, &priv->fs.inner_ttc);
267 netdev_err(priv->netdev, "Failed to create inner ttc table, err=%d\n",
269 goto err_destroy_arfs_tables;
272 mlx5e_set_ttc_ft_params(&ttc_params);
273 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
274 ttc_params.indir_tirn[tt] = priv->inner_indir_tir[tt].tirn;
276 err = mlx5e_create_ttc_table(priv, &ttc_params, &priv->fs.ttc);
278 netdev_err(priv->netdev, "Failed to create ttc table, err=%d\n",
280 goto err_destroy_inner_ttc_table;
285 err_destroy_inner_ttc_table:
286 mlx5e_destroy_inner_ttc_table(priv, &priv->fs.inner_ttc);
287 err_destroy_arfs_tables:
288 mlx5e_arfs_destroy_tables(priv);
293 static void mlx5i_destroy_flow_steering(struct mlx5e_priv *priv)
295 mlx5e_destroy_ttc_table(priv, &priv->fs.ttc);
296 mlx5e_destroy_inner_ttc_table(priv, &priv->fs.inner_ttc);
297 mlx5e_arfs_destroy_tables(priv);
300 static int mlx5i_init_rx(struct mlx5e_priv *priv)
304 err = mlx5e_create_indirect_rqt(priv);
308 err = mlx5e_create_direct_rqts(priv);
310 goto err_destroy_indirect_rqts;
312 err = mlx5e_create_indirect_tirs(priv);
314 goto err_destroy_direct_rqts;
316 err = mlx5e_create_direct_tirs(priv);
318 goto err_destroy_indirect_tirs;
320 err = mlx5i_create_flow_steering(priv);
322 goto err_destroy_direct_tirs;
326 err_destroy_direct_tirs:
327 mlx5e_destroy_direct_tirs(priv);
328 err_destroy_indirect_tirs:
329 mlx5e_destroy_indirect_tirs(priv);
330 err_destroy_direct_rqts:
331 mlx5e_destroy_direct_rqts(priv);
332 err_destroy_indirect_rqts:
333 mlx5e_destroy_rqt(priv, &priv->indir_rqt);
337 static void mlx5i_cleanup_rx(struct mlx5e_priv *priv)
339 mlx5i_destroy_flow_steering(priv);
340 mlx5e_destroy_direct_tirs(priv);
341 mlx5e_destroy_indirect_tirs(priv);
342 mlx5e_destroy_direct_rqts(priv);
343 mlx5e_destroy_rqt(priv, &priv->indir_rqt);
346 static const struct mlx5e_profile mlx5i_nic_profile = {
348 .cleanup = mlx5i_cleanup,
349 .init_tx = mlx5i_init_tx,
350 .cleanup_tx = mlx5i_cleanup_tx,
351 .init_rx = mlx5i_init_rx,
352 .cleanup_rx = mlx5i_cleanup_rx,
353 .enable = NULL, /* mlx5i_enable */
354 .disable = NULL, /* mlx5i_disable */
355 .update_stats = NULL, /* mlx5i_update_stats */
356 .max_nch = mlx5e_get_max_num_channels,
357 .update_carrier = NULL, /* no HW update in IB link */
358 .rx_handlers.handle_rx_cqe = mlx5i_handle_rx_cqe,
359 .rx_handlers.handle_rx_cqe_mpwqe = NULL, /* Not supported */
360 .max_tc = MLX5I_MAX_NUM_TC,
363 /* mlx5i netdev NDos */
365 static int mlx5i_change_mtu(struct net_device *netdev, int new_mtu)
367 struct mlx5e_priv *priv = mlx5i_epriv(netdev);
368 struct mlx5e_channels new_channels = {};
372 mutex_lock(&priv->state_lock);
374 curr_mtu = netdev->mtu;
375 netdev->mtu = new_mtu;
377 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
380 new_channels.params = priv->channels.params;
381 err = mlx5e_open_channels(priv, &new_channels);
383 netdev->mtu = curr_mtu;
387 mlx5e_switch_priv_channels(priv, &new_channels, NULL);
390 mutex_unlock(&priv->state_lock);
394 int mlx5i_dev_init(struct net_device *dev)
396 struct mlx5e_priv *priv = mlx5i_epriv(dev);
397 struct mlx5i_priv *ipriv = priv->ppriv;
399 /* Set dev address using underlay QP */
400 dev->dev_addr[1] = (ipriv->qp.qpn >> 16) & 0xff;
401 dev->dev_addr[2] = (ipriv->qp.qpn >> 8) & 0xff;
402 dev->dev_addr[3] = (ipriv->qp.qpn) & 0xff;
404 /* Add QPN to net-device mapping to HT */
405 mlx5i_pkey_add_qpn(dev ,ipriv->qp.qpn);
410 int mlx5i_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
412 struct mlx5e_priv *priv = mlx5i_epriv(dev);
416 return mlx5e_hwstamp_set(priv, ifr);
418 return mlx5e_hwstamp_get(priv, ifr);
424 void mlx5i_dev_cleanup(struct net_device *dev)
426 struct mlx5e_priv *priv = mlx5i_epriv(dev);
427 struct mlx5i_priv *ipriv = priv->ppriv;
429 mlx5i_uninit_underlay_qp(priv);
431 /* Delete QPN to net-device mapping from HT */
432 mlx5i_pkey_del_qpn(dev, ipriv->qp.qpn);
435 static int mlx5i_open(struct net_device *netdev)
437 struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
438 struct mlx5i_priv *ipriv = epriv->ppriv;
439 struct mlx5_core_dev *mdev = epriv->mdev;
442 mutex_lock(&epriv->state_lock);
444 set_bit(MLX5E_STATE_OPENED, &epriv->state);
446 err = mlx5i_init_underlay_qp(epriv);
448 mlx5_core_warn(mdev, "prepare underlay qp state failed, %d\n", err);
449 goto err_clear_state_opened_flag;
452 err = mlx5_fs_add_rx_underlay_qpn(mdev, ipriv->qp.qpn);
454 mlx5_core_warn(mdev, "attach underlay qp to ft failed, %d\n", err);
458 err = mlx5e_open_channels(epriv, &epriv->channels);
460 goto err_remove_fs_underlay_qp;
462 mlx5e_refresh_tirs(epriv, false);
463 mlx5e_activate_priv_channels(epriv);
465 mutex_unlock(&epriv->state_lock);
468 err_remove_fs_underlay_qp:
469 mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qp.qpn);
471 mlx5i_uninit_underlay_qp(epriv);
472 err_clear_state_opened_flag:
473 clear_bit(MLX5E_STATE_OPENED, &epriv->state);
474 mutex_unlock(&epriv->state_lock);
478 static int mlx5i_close(struct net_device *netdev)
480 struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
481 struct mlx5i_priv *ipriv = epriv->ppriv;
482 struct mlx5_core_dev *mdev = epriv->mdev;
484 /* May already be CLOSED in case a previous configuration operation
485 * (e.g RX/TX queue size change) that involves close&open failed.
487 mutex_lock(&epriv->state_lock);
489 if (!test_bit(MLX5E_STATE_OPENED, &epriv->state))
492 clear_bit(MLX5E_STATE_OPENED, &epriv->state);
494 netif_carrier_off(epriv->netdev);
495 mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qp.qpn);
496 mlx5i_uninit_underlay_qp(epriv);
497 mlx5e_deactivate_priv_channels(epriv);
498 mlx5e_close_channels(&epriv->channels);
500 mutex_unlock(&epriv->state_lock);
504 /* IPoIB RDMA netdev callbacks */
505 static int mlx5i_attach_mcast(struct net_device *netdev, struct ib_device *hca,
506 union ib_gid *gid, u16 lid, int set_qkey,
509 struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
510 struct mlx5_core_dev *mdev = epriv->mdev;
511 struct mlx5i_priv *ipriv = epriv->ppriv;
514 mlx5_core_dbg(mdev, "attaching QPN 0x%x, MGID %pI6\n", ipriv->qp.qpn, gid->raw);
515 err = mlx5_core_attach_mcg(mdev, gid, ipriv->qp.qpn);
517 mlx5_core_warn(mdev, "failed attaching QPN 0x%x, MGID %pI6\n",
518 ipriv->qp.qpn, gid->raw);
521 mlx5_core_dbg(mdev, "%s setting qkey 0x%x\n",
529 static int mlx5i_detach_mcast(struct net_device *netdev, struct ib_device *hca,
530 union ib_gid *gid, u16 lid)
532 struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
533 struct mlx5_core_dev *mdev = epriv->mdev;
534 struct mlx5i_priv *ipriv = epriv->ppriv;
537 mlx5_core_dbg(mdev, "detaching QPN 0x%x, MGID %pI6\n", ipriv->qp.qpn, gid->raw);
539 err = mlx5_core_detach_mcg(mdev, gid, ipriv->qp.qpn);
541 mlx5_core_dbg(mdev, "failed dettaching QPN 0x%x, MGID %pI6\n",
542 ipriv->qp.qpn, gid->raw);
547 static int mlx5i_xmit(struct net_device *dev, struct sk_buff *skb,
548 struct ib_ah *address, u32 dqpn)
550 struct mlx5e_priv *epriv = mlx5i_epriv(dev);
551 struct mlx5e_txqsq *sq = epriv->txq2sq[skb_get_queue_mapping(skb)];
552 struct mlx5_ib_ah *mah = to_mah(address);
553 struct mlx5i_priv *ipriv = epriv->ppriv;
555 return mlx5i_sq_xmit(sq, skb, &mah->av, dqpn, ipriv->qkey);
558 static void mlx5i_set_pkey_index(struct net_device *netdev, int id)
560 struct mlx5i_priv *ipriv = netdev_priv(netdev);
562 ipriv->pkey_index = (u16)id;
565 static int mlx5i_check_required_hca_cap(struct mlx5_core_dev *mdev)
567 if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_IB)
570 if (!MLX5_CAP_GEN(mdev, ipoib_enhanced_offloads)) {
571 mlx5_core_warn(mdev, "IPoIB enhanced offloads are not supported\n");
578 struct net_device *mlx5_rdma_netdev_alloc(struct mlx5_core_dev *mdev,
579 struct ib_device *ibdev,
581 void (*setup)(struct net_device *))
583 const struct mlx5e_profile *profile;
584 struct net_device *netdev;
585 struct mlx5i_priv *ipriv;
586 struct mlx5e_priv *epriv;
587 struct rdma_netdev *rn;
592 if (mlx5i_check_required_hca_cap(mdev)) {
593 mlx5_core_warn(mdev, "Accelerated mode is not supported\n");
594 return ERR_PTR(-EOPNOTSUPP);
597 /* TODO: Need to find a better way to check if child device*/
598 sub_interface = (mdev->mlx5e_res.pdn != 0);
601 profile = mlx5i_pkey_get_profile();
603 profile = &mlx5i_nic_profile;
605 nch = profile->max_nch(mdev);
607 netdev = alloc_netdev_mqs(sizeof(struct mlx5i_priv) + sizeof(struct mlx5e_priv),
608 name, NET_NAME_UNKNOWN,
610 nch * MLX5E_MAX_NUM_TC,
613 mlx5_core_warn(mdev, "alloc_netdev_mqs failed\n");
617 ipriv = netdev_priv(netdev);
618 epriv = mlx5i_epriv(netdev);
620 epriv->wq = create_singlethread_workqueue("mlx5i");
622 goto err_free_netdev;
624 ipriv->sub_interface = sub_interface;
625 if (!ipriv->sub_interface) {
626 err = mlx5i_pkey_qpn_ht_init(netdev);
628 mlx5_core_warn(mdev, "allocate qpn_to_netdev ht failed\n");
632 /* This should only be called once per mdev */
633 err = mlx5e_create_mdev_resources(mdev);
638 profile->init(mdev, netdev, profile, ipriv);
640 mlx5e_attach_netdev(epriv);
641 netif_carrier_off(netdev);
643 /* set rdma_netdev func pointers */
646 rn->send = mlx5i_xmit;
647 rn->attach_mcast = mlx5i_attach_mcast;
648 rn->detach_mcast = mlx5i_detach_mcast;
649 rn->set_id = mlx5i_set_pkey_index;
654 mlx5i_pkey_qpn_ht_cleanup(netdev);
656 destroy_workqueue(epriv->wq);
662 EXPORT_SYMBOL(mlx5_rdma_netdev_alloc);
664 void mlx5_rdma_netdev_free(struct net_device *netdev)
666 struct mlx5e_priv *priv = mlx5i_epriv(netdev);
667 struct mlx5i_priv *ipriv = priv->ppriv;
668 const struct mlx5e_profile *profile = priv->profile;
670 mlx5e_detach_netdev(priv);
671 profile->cleanup(priv);
672 destroy_workqueue(priv->wq);
674 if (!ipriv->sub_interface) {
675 mlx5i_pkey_qpn_ht_cleanup(netdev);
676 mlx5e_destroy_mdev_resources(priv->mdev);
680 EXPORT_SYMBOL(mlx5_rdma_netdev_free);