net/mlx5e: IPoIB, Xmit flow
[linux-2.6/btrfs-unstable.git] / drivers / net / ethernet / mellanox / mlx5 / core / ipoib.c
blobc468aaedf0a66d70817115d2cf33e317eb7e5b1d
1 /*
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
12 * conditions are met:
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
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
30 * SOFTWARE.
33 #include <linux/mlx5/fs.h>
34 #include "en.h"
35 #include "ipoib.h"
37 #define IB_DEFAULT_Q_KEY 0xb1b
39 static int mlx5i_open(struct net_device *netdev);
40 static int mlx5i_close(struct net_device *netdev);
41 static int mlx5i_dev_init(struct net_device *dev);
42 static void mlx5i_dev_cleanup(struct net_device *dev);
44 static const struct net_device_ops mlx5i_netdev_ops = {
45 .ndo_open = mlx5i_open,
46 .ndo_stop = mlx5i_close,
47 .ndo_init = mlx5i_dev_init,
48 .ndo_uninit = mlx5i_dev_cleanup,
51 /* IPoIB mlx5 netdev profile */
53 /* Called directly after IPoIB netdevice was created to initialize SW structs */
54 static void mlx5i_init(struct mlx5_core_dev *mdev,
55 struct net_device *netdev,
56 const struct mlx5e_profile *profile,
57 void *ppriv)
59 struct mlx5e_priv *priv = mlx5i_epriv(netdev);
61 priv->mdev = mdev;
62 priv->netdev = netdev;
63 priv->profile = profile;
64 priv->ppriv = ppriv;
66 mlx5e_build_nic_params(mdev, &priv->channels.params, profile->max_nch(mdev));
68 mutex_init(&priv->state_lock);
70 netdev->hw_features |= NETIF_F_SG;
71 netdev->hw_features |= NETIF_F_IP_CSUM;
72 netdev->hw_features |= NETIF_F_IPV6_CSUM;
73 netdev->hw_features |= NETIF_F_GRO;
74 netdev->hw_features |= NETIF_F_TSO;
75 netdev->hw_features |= NETIF_F_TSO6;
76 netdev->hw_features |= NETIF_F_RXCSUM;
77 netdev->hw_features |= NETIF_F_RXHASH;
79 netdev->netdev_ops = &mlx5i_netdev_ops;
82 /* Called directly before IPoIB netdevice is destroyed to cleanup SW structs */
83 static void mlx5i_cleanup(struct mlx5e_priv *priv)
85 /* Do nothing .. */
88 #define MLX5_QP_ENHANCED_ULP_STATELESS_MODE 2
90 static int mlx5i_create_underlay_qp(struct mlx5_core_dev *mdev, struct mlx5_core_qp *qp)
92 struct mlx5_qp_context *context = NULL;
93 u32 *in = NULL;
94 void *addr_path;
95 int ret = 0;
96 int inlen;
97 void *qpc;
99 inlen = MLX5_ST_SZ_BYTES(create_qp_in);
100 in = mlx5_vzalloc(inlen);
101 if (!in)
102 return -ENOMEM;
104 qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
105 MLX5_SET(qpc, qpc, st, MLX5_QP_ST_UD);
106 MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
107 MLX5_SET(qpc, qpc, ulp_stateless_offload_mode,
108 MLX5_QP_ENHANCED_ULP_STATELESS_MODE);
110 addr_path = MLX5_ADDR_OF(qpc, qpc, primary_address_path);
111 MLX5_SET(ads, addr_path, port, 1);
112 MLX5_SET(ads, addr_path, grh, 1);
114 ret = mlx5_core_create_qp(mdev, qp, in, inlen);
115 if (ret) {
116 mlx5_core_err(mdev, "Failed creating IPoIB QP err : %d\n", ret);
117 goto out;
120 /* QP states */
121 context = kzalloc(sizeof(*context), GFP_KERNEL);
122 if (!context) {
123 ret = -ENOMEM;
124 goto out;
127 context->flags = cpu_to_be32(MLX5_QP_PM_MIGRATED << 11);
128 context->pri_path.port = 1;
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);
132 if (ret) {
133 mlx5_core_err(mdev, "Failed to modify qp RST2INIT, err: %d\n", ret);
134 goto out;
136 memset(context, 0, sizeof(*context));
138 ret = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_INIT2RTR_QP, 0, context, qp);
139 if (ret) {
140 mlx5_core_err(mdev, "Failed to modify qp INIT2RTR, err: %d\n", ret);
141 goto out;
144 ret = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_RTR2RTS_QP, 0, context, qp);
145 if (ret) {
146 mlx5_core_err(mdev, "Failed to modify qp RTR2RTS, err: %d\n", ret);
147 goto out;
150 out:
151 kfree(context);
152 kvfree(in);
153 return ret;
156 static void mlx5i_destroy_underlay_qp(struct mlx5_core_dev *mdev, struct mlx5_core_qp *qp)
158 mlx5_core_destroy_qp(mdev, qp);
161 static int mlx5i_init_tx(struct mlx5e_priv *priv)
163 struct mlx5i_priv *ipriv = priv->ppriv;
164 int err;
166 err = mlx5i_create_underlay_qp(priv->mdev, &ipriv->qp);
167 if (err) {
168 mlx5_core_warn(priv->mdev, "create underlay QP failed, %d\n", err);
169 return err;
172 err = mlx5e_create_tis(priv->mdev, 0 /* tc */, ipriv->qp.qpn, &priv->tisn[0]);
173 if (err) {
174 mlx5_core_warn(priv->mdev, "create tis failed, %d\n", err);
175 return err;
178 return 0;
181 void mlx5i_cleanup_tx(struct mlx5e_priv *priv)
183 struct mlx5i_priv *ipriv = priv->ppriv;
185 mlx5e_destroy_tis(priv->mdev, priv->tisn[0]);
186 mlx5i_destroy_underlay_qp(priv->mdev, &ipriv->qp);
189 static int mlx5i_create_flow_steering(struct mlx5e_priv *priv)
191 struct mlx5i_priv *ipriv = priv->ppriv;
192 int err;
194 priv->fs.ns = mlx5_get_flow_namespace(priv->mdev,
195 MLX5_FLOW_NAMESPACE_KERNEL);
197 if (!priv->fs.ns)
198 return -EINVAL;
200 err = mlx5e_arfs_create_tables(priv);
201 if (err) {
202 netdev_err(priv->netdev, "Failed to create arfs tables, err=%d\n",
203 err);
204 priv->netdev->hw_features &= ~NETIF_F_NTUPLE;
207 err = mlx5e_create_ttc_table(priv, ipriv->qp.qpn);
208 if (err) {
209 netdev_err(priv->netdev, "Failed to create ttc table, err=%d\n",
210 err);
211 goto err_destroy_arfs_tables;
214 return 0;
216 err_destroy_arfs_tables:
217 mlx5e_arfs_destroy_tables(priv);
219 return err;
222 static void mlx5i_destroy_flow_steering(struct mlx5e_priv *priv)
224 mlx5e_destroy_ttc_table(priv);
225 mlx5e_arfs_destroy_tables(priv);
228 static int mlx5i_init_rx(struct mlx5e_priv *priv)
230 int err;
232 err = mlx5e_create_indirect_rqt(priv);
233 if (err)
234 return err;
236 err = mlx5e_create_direct_rqts(priv);
237 if (err)
238 goto err_destroy_indirect_rqts;
240 err = mlx5e_create_indirect_tirs(priv);
241 if (err)
242 goto err_destroy_direct_rqts;
244 err = mlx5e_create_direct_tirs(priv);
245 if (err)
246 goto err_destroy_indirect_tirs;
248 err = mlx5i_create_flow_steering(priv);
249 if (err)
250 goto err_destroy_direct_tirs;
252 return 0;
254 err_destroy_direct_tirs:
255 mlx5e_destroy_direct_tirs(priv);
256 err_destroy_indirect_tirs:
257 mlx5e_destroy_indirect_tirs(priv);
258 err_destroy_direct_rqts:
259 mlx5e_destroy_direct_rqts(priv);
260 err_destroy_indirect_rqts:
261 mlx5e_destroy_rqt(priv, &priv->indir_rqt);
262 return err;
265 static void mlx5i_cleanup_rx(struct mlx5e_priv *priv)
267 mlx5i_destroy_flow_steering(priv);
268 mlx5e_destroy_direct_tirs(priv);
269 mlx5e_destroy_indirect_tirs(priv);
270 mlx5e_destroy_direct_rqts(priv);
271 mlx5e_destroy_rqt(priv, &priv->indir_rqt);
274 static const struct mlx5e_profile mlx5i_nic_profile = {
275 .init = mlx5i_init,
276 .cleanup = mlx5i_cleanup,
277 .init_tx = mlx5i_init_tx,
278 .cleanup_tx = mlx5i_cleanup_tx,
279 .init_rx = mlx5i_init_rx,
280 .cleanup_rx = mlx5i_cleanup_rx,
281 .enable = NULL, /* mlx5i_enable */
282 .disable = NULL, /* mlx5i_disable */
283 .update_stats = NULL, /* mlx5i_update_stats */
284 .max_nch = mlx5e_get_max_num_channels,
285 .max_tc = MLX5I_MAX_NUM_TC,
288 /* mlx5i netdev NDos */
290 static int mlx5i_dev_init(struct net_device *dev)
292 struct mlx5e_priv *priv = mlx5i_epriv(dev);
293 struct mlx5i_priv *ipriv = priv->ppriv;
295 /* Set dev address using underlay QP */
296 dev->dev_addr[1] = (ipriv->qp.qpn >> 16) & 0xff;
297 dev->dev_addr[2] = (ipriv->qp.qpn >> 8) & 0xff;
298 dev->dev_addr[3] = (ipriv->qp.qpn) & 0xff;
300 return 0;
303 static void mlx5i_dev_cleanup(struct net_device *dev)
305 struct mlx5e_priv *priv = mlx5i_epriv(dev);
306 struct mlx5_core_dev *mdev = priv->mdev;
307 struct mlx5i_priv *ipriv = priv->ppriv;
308 struct mlx5_qp_context context;
310 /* detach qp from flow-steering by reset it */
311 mlx5_core_qp_modify(mdev, MLX5_CMD_OP_2RST_QP, 0, &context, &ipriv->qp);
314 static int mlx5i_open(struct net_device *netdev)
316 struct mlx5e_priv *priv = mlx5i_epriv(netdev);
317 int err;
319 mutex_lock(&priv->state_lock);
321 set_bit(MLX5E_STATE_OPENED, &priv->state);
323 err = mlx5e_open_channels(priv, &priv->channels);
324 if (err)
325 goto err_clear_state_opened_flag;
327 mlx5e_refresh_tirs(priv, false);
328 mlx5e_activate_priv_channels(priv);
329 mutex_unlock(&priv->state_lock);
330 return 0;
332 err_clear_state_opened_flag:
333 clear_bit(MLX5E_STATE_OPENED, &priv->state);
334 mutex_unlock(&priv->state_lock);
335 return err;
338 static int mlx5i_close(struct net_device *netdev)
340 struct mlx5e_priv *priv = mlx5i_epriv(netdev);
342 /* May already be CLOSED in case a previous configuration operation
343 * (e.g RX/TX queue size change) that involves close&open failed.
345 mutex_lock(&priv->state_lock);
347 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
348 goto unlock;
350 clear_bit(MLX5E_STATE_OPENED, &priv->state);
352 netif_carrier_off(priv->netdev);
353 mlx5e_deactivate_priv_channels(priv);
354 mlx5e_close_channels(&priv->channels);
355 unlock:
356 mutex_unlock(&priv->state_lock);
357 return 0;
360 /* IPoIB RDMA netdev callbacks */
361 int mlx5i_attach_mcast(struct net_device *netdev, struct ib_device *hca,
362 union ib_gid *gid, u16 lid, int set_qkey)
364 struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
365 struct mlx5_core_dev *mdev = epriv->mdev;
366 struct mlx5i_priv *ipriv = epriv->ppriv;
367 int err;
369 mlx5_core_dbg(mdev, "attaching QPN 0x%x, MGID %pI6\n", ipriv->qp.qpn, gid->raw);
370 err = mlx5_core_attach_mcg(mdev, gid, ipriv->qp.qpn);
371 if (err)
372 mlx5_core_warn(mdev, "failed attaching QPN 0x%x, MGID %pI6\n",
373 ipriv->qp.qpn, gid->raw);
375 return err;
378 int mlx5i_detach_mcast(struct net_device *netdev, struct ib_device *hca,
379 union ib_gid *gid, u16 lid)
381 struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
382 struct mlx5_core_dev *mdev = epriv->mdev;
383 struct mlx5i_priv *ipriv = epriv->ppriv;
384 int err;
386 mlx5_core_dbg(mdev, "detaching QPN 0x%x, MGID %pI6\n", ipriv->qp.qpn, gid->raw);
388 err = mlx5_core_detach_mcg(mdev, gid, ipriv->qp.qpn);
389 if (err)
390 mlx5_core_dbg(mdev, "failed dettaching QPN 0x%x, MGID %pI6\n",
391 ipriv->qp.qpn, gid->raw);
393 return err;
396 int mlx5i_xmit(struct net_device *dev, struct sk_buff *skb,
397 struct ib_ah *address, u32 dqpn, u32 dqkey)
399 struct mlx5e_priv *epriv = mlx5i_epriv(dev);
400 struct mlx5e_txqsq *sq = epriv->txq2sq[skb_get_queue_mapping(skb)];
401 struct mlx5_ib_ah *mah = to_mah(address);
403 return mlx5i_sq_xmit(sq, skb, &mah->av, dqpn, dqkey);
406 static int mlx5i_check_required_hca_cap(struct mlx5_core_dev *mdev)
408 if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_IB)
409 return -EOPNOTSUPP;
411 if (!MLX5_CAP_GEN(mdev, ipoib_enhanced_offloads)) {
412 mlx5_core_warn(mdev, "IPoIB enhanced offloads are not supported\n");
413 return -ENOTSUPP;
416 return 0;
419 struct net_device *mlx5_rdma_netdev_alloc(struct mlx5_core_dev *mdev,
420 struct ib_device *ibdev,
421 const char *name,
422 void (*setup)(struct net_device *))
424 const struct mlx5e_profile *profile = &mlx5i_nic_profile;
425 int nch = profile->max_nch(mdev);
426 struct net_device *netdev;
427 struct mlx5i_priv *ipriv;
428 struct mlx5e_priv *epriv;
429 int err;
431 if (mlx5i_check_required_hca_cap(mdev)) {
432 mlx5_core_warn(mdev, "Accelerated mode is not supported\n");
433 return ERR_PTR(-EOPNOTSUPP);
436 /* This function should only be called once per mdev */
437 err = mlx5e_create_mdev_resources(mdev);
438 if (err)
439 return NULL;
441 netdev = alloc_netdev_mqs(sizeof(struct mlx5i_priv) + sizeof(struct mlx5e_priv),
442 name, NET_NAME_UNKNOWN,
443 setup,
444 nch * MLX5E_MAX_NUM_TC,
445 nch);
446 if (!netdev) {
447 mlx5_core_warn(mdev, "alloc_netdev_mqs failed\n");
448 goto free_mdev_resources;
451 ipriv = netdev_priv(netdev);
452 epriv = mlx5i_epriv(netdev);
454 epriv->wq = create_singlethread_workqueue("mlx5i");
455 if (!epriv->wq)
456 goto err_free_netdev;
458 profile->init(mdev, netdev, profile, ipriv);
460 mlx5e_attach_netdev(epriv);
461 netif_carrier_off(netdev);
463 /* TODO: set rdma_netdev func pointers
464 * rn = &ipriv->rn;
465 * rn->hca = ibdev;
466 * rn->send = mlx5i_xmit;
467 * rn->attach_mcast = mlx5i_attach_mcast;
468 * rn->detach_mcast = mlx5i_detach_mcast;
470 return netdev;
472 free_mdev_resources:
473 mlx5e_destroy_mdev_resources(mdev);
474 err_free_netdev:
475 free_netdev(netdev);
476 return NULL;
478 EXPORT_SYMBOL(mlx5_rdma_netdev_alloc);
480 void mlx5_rdma_netdev_free(struct net_device *netdev)
482 struct mlx5e_priv *priv = mlx5i_epriv(netdev);
483 const struct mlx5e_profile *profile = priv->profile;
485 mlx5e_detach_netdev(priv);
486 profile->cleanup(priv);
487 destroy_workqueue(priv->wq);
488 free_netdev(netdev);
490 mlx5e_destroy_mdev_resources(priv->mdev);
492 EXPORT_SYMBOL(mlx5_rdma_netdev_free);