• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/slab.h>
37 #include <linux/errno.h>
38 #include <linux/netdevice.h>
39 #include <linux/inetdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/if_vlan.h>
42 #include <linux/sched/mm.h>
43 #include <linux/sched/task.h>
44 
45 #include <net/ipv6.h>
46 #include <net/addrconf.h>
47 #include <net/devlink.h>
48 
49 #include <rdma/ib_smi.h>
50 #include <rdma/ib_user_verbs.h>
51 #include <rdma/ib_addr.h>
52 #include <rdma/ib_cache.h>
53 
54 #include <net/bonding.h>
55 
56 #include <linux/mlx4/driver.h>
57 #include <linux/mlx4/cmd.h>
58 #include <linux/mlx4/qp.h>
59 
60 #include "mlx4_ib.h"
61 #include <rdma/mlx4-abi.h>
62 
63 #define DRV_NAME	MLX4_IB_DRV_NAME
64 #define DRV_VERSION	"4.0-0"
65 
66 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
67 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
68 #define MLX4_IB_CARD_REV_A0   0xA0
69 
70 MODULE_AUTHOR("Roland Dreier");
71 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
72 MODULE_LICENSE("Dual BSD/GPL");
73 
74 int mlx4_ib_sm_guid_assign = 0;
75 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
76 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
77 
78 static const char mlx4_ib_version[] =
79 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
80 	DRV_VERSION "\n";
81 
82 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
83 static enum rdma_link_layer mlx4_ib_port_link_layer(struct ib_device *device,
84 						    u8 port_num);
85 
86 static struct workqueue_struct *wq;
87 
init_query_mad(struct ib_smp * mad)88 static void init_query_mad(struct ib_smp *mad)
89 {
90 	mad->base_version  = 1;
91 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
92 	mad->class_version = 1;
93 	mad->method	   = IB_MGMT_METHOD_GET;
94 }
95 
check_flow_steering_support(struct mlx4_dev * dev)96 static int check_flow_steering_support(struct mlx4_dev *dev)
97 {
98 	int eth_num_ports = 0;
99 	int ib_num_ports = 0;
100 
101 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
102 
103 	if (dmfs) {
104 		int i;
105 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
106 			eth_num_ports++;
107 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
108 			ib_num_ports++;
109 		dmfs &= (!ib_num_ports ||
110 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
111 			(!eth_num_ports ||
112 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
113 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
114 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
115 			dmfs = 0;
116 		}
117 	}
118 	return dmfs;
119 }
120 
num_ib_ports(struct mlx4_dev * dev)121 static int num_ib_ports(struct mlx4_dev *dev)
122 {
123 	int ib_ports = 0;
124 	int i;
125 
126 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
127 		ib_ports++;
128 
129 	return ib_ports;
130 }
131 
mlx4_ib_get_netdev(struct ib_device * device,u8 port_num)132 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
133 {
134 	struct mlx4_ib_dev *ibdev = to_mdev(device);
135 	struct net_device *dev;
136 
137 	rcu_read_lock();
138 	dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
139 
140 	if (dev) {
141 		if (mlx4_is_bonded(ibdev->dev)) {
142 			struct net_device *upper = NULL;
143 
144 			upper = netdev_master_upper_dev_get_rcu(dev);
145 			if (upper) {
146 				struct net_device *active;
147 
148 				active = bond_option_active_slave_get_rcu(netdev_priv(upper));
149 				if (active)
150 					dev = active;
151 			}
152 		}
153 	}
154 	if (dev)
155 		dev_hold(dev);
156 
157 	rcu_read_unlock();
158 	return dev;
159 }
160 
mlx4_ib_update_gids_v1(struct gid_entry * gids,struct mlx4_ib_dev * ibdev,u8 port_num)161 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
162 				  struct mlx4_ib_dev *ibdev,
163 				  u8 port_num)
164 {
165 	struct mlx4_cmd_mailbox *mailbox;
166 	int err;
167 	struct mlx4_dev *dev = ibdev->dev;
168 	int i;
169 	union ib_gid *gid_tbl;
170 
171 	mailbox = mlx4_alloc_cmd_mailbox(dev);
172 	if (IS_ERR(mailbox))
173 		return -ENOMEM;
174 
175 	gid_tbl = mailbox->buf;
176 
177 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
178 		memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
179 
180 	err = mlx4_cmd(dev, mailbox->dma,
181 		       MLX4_SET_PORT_GID_TABLE << 8 | port_num,
182 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
183 		       MLX4_CMD_WRAPPED);
184 	if (mlx4_is_bonded(dev))
185 		err += mlx4_cmd(dev, mailbox->dma,
186 				MLX4_SET_PORT_GID_TABLE << 8 | 2,
187 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
188 				MLX4_CMD_WRAPPED);
189 
190 	mlx4_free_cmd_mailbox(dev, mailbox);
191 	return err;
192 }
193 
mlx4_ib_update_gids_v1_v2(struct gid_entry * gids,struct mlx4_ib_dev * ibdev,u8 port_num)194 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
195 				     struct mlx4_ib_dev *ibdev,
196 				     u8 port_num)
197 {
198 	struct mlx4_cmd_mailbox *mailbox;
199 	int err;
200 	struct mlx4_dev *dev = ibdev->dev;
201 	int i;
202 	struct {
203 		union ib_gid	gid;
204 		__be32		rsrvd1[2];
205 		__be16		rsrvd2;
206 		u8		type;
207 		u8		version;
208 		__be32		rsrvd3;
209 	} *gid_tbl;
210 
211 	mailbox = mlx4_alloc_cmd_mailbox(dev);
212 	if (IS_ERR(mailbox))
213 		return -ENOMEM;
214 
215 	gid_tbl = mailbox->buf;
216 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
217 		memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
218 		if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
219 			gid_tbl[i].version = 2;
220 			if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
221 				gid_tbl[i].type = 1;
222 		}
223 	}
224 
225 	err = mlx4_cmd(dev, mailbox->dma,
226 		       MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
227 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
228 		       MLX4_CMD_WRAPPED);
229 	if (mlx4_is_bonded(dev))
230 		err += mlx4_cmd(dev, mailbox->dma,
231 				MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
232 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
233 				MLX4_CMD_WRAPPED);
234 
235 	mlx4_free_cmd_mailbox(dev, mailbox);
236 	return err;
237 }
238 
mlx4_ib_update_gids(struct gid_entry * gids,struct mlx4_ib_dev * ibdev,u8 port_num)239 static int mlx4_ib_update_gids(struct gid_entry *gids,
240 			       struct mlx4_ib_dev *ibdev,
241 			       u8 port_num)
242 {
243 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
244 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
245 
246 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
247 }
248 
free_gid_entry(struct gid_entry * entry)249 static void free_gid_entry(struct gid_entry *entry)
250 {
251 	memset(&entry->gid, 0, sizeof(entry->gid));
252 	kfree(entry->ctx);
253 	entry->ctx = NULL;
254 }
255 
mlx4_ib_add_gid(const struct ib_gid_attr * attr,void ** context)256 static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
257 {
258 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
259 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
260 	struct mlx4_port_gid_table   *port_gid_table;
261 	int free = -1, found = -1;
262 	int ret = 0;
263 	int hw_update = 0;
264 	int i;
265 	struct gid_entry *gids = NULL;
266 
267 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
268 		return -EINVAL;
269 
270 	if (attr->port_num > MLX4_MAX_PORTS)
271 		return -EINVAL;
272 
273 	if (!context)
274 		return -EINVAL;
275 
276 	port_gid_table = &iboe->gids[attr->port_num - 1];
277 	spin_lock_bh(&iboe->lock);
278 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
279 		if (!memcmp(&port_gid_table->gids[i].gid,
280 			    &attr->gid, sizeof(attr->gid)) &&
281 		    port_gid_table->gids[i].gid_type == attr->gid_type)  {
282 			found = i;
283 			break;
284 		}
285 		if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid))
286 			free = i; /* HW has space */
287 	}
288 
289 	if (found < 0) {
290 		if (free < 0) {
291 			ret = -ENOSPC;
292 		} else {
293 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
294 			if (!port_gid_table->gids[free].ctx) {
295 				ret = -ENOMEM;
296 			} else {
297 				*context = port_gid_table->gids[free].ctx;
298 				memcpy(&port_gid_table->gids[free].gid,
299 				       &attr->gid, sizeof(attr->gid));
300 				port_gid_table->gids[free].gid_type = attr->gid_type;
301 				port_gid_table->gids[free].ctx->real_index = free;
302 				port_gid_table->gids[free].ctx->refcount = 1;
303 				hw_update = 1;
304 			}
305 		}
306 	} else {
307 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
308 		*context = ctx;
309 		ctx->refcount++;
310 	}
311 	if (!ret && hw_update) {
312 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
313 				     GFP_ATOMIC);
314 		if (!gids) {
315 			ret = -ENOMEM;
316 			*context = NULL;
317 			free_gid_entry(&port_gid_table->gids[free]);
318 		} else {
319 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
320 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
321 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
322 			}
323 		}
324 	}
325 	spin_unlock_bh(&iboe->lock);
326 
327 	if (!ret && hw_update) {
328 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
329 		if (ret) {
330 			spin_lock_bh(&iboe->lock);
331 			*context = NULL;
332 			free_gid_entry(&port_gid_table->gids[free]);
333 			spin_unlock_bh(&iboe->lock);
334 		}
335 		kfree(gids);
336 	}
337 
338 	return ret;
339 }
340 
mlx4_ib_del_gid(const struct ib_gid_attr * attr,void ** context)341 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
342 {
343 	struct gid_cache_context *ctx = *context;
344 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
345 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
346 	struct mlx4_port_gid_table   *port_gid_table;
347 	int ret = 0;
348 	int hw_update = 0;
349 	struct gid_entry *gids = NULL;
350 
351 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
352 		return -EINVAL;
353 
354 	if (attr->port_num > MLX4_MAX_PORTS)
355 		return -EINVAL;
356 
357 	port_gid_table = &iboe->gids[attr->port_num - 1];
358 	spin_lock_bh(&iboe->lock);
359 	if (ctx) {
360 		ctx->refcount--;
361 		if (!ctx->refcount) {
362 			unsigned int real_index = ctx->real_index;
363 
364 			free_gid_entry(&port_gid_table->gids[real_index]);
365 			hw_update = 1;
366 		}
367 	}
368 	if (!ret && hw_update) {
369 		int i;
370 
371 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
372 				     GFP_ATOMIC);
373 		if (!gids) {
374 			ret = -ENOMEM;
375 		} else {
376 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
377 				memcpy(&gids[i].gid,
378 				       &port_gid_table->gids[i].gid,
379 				       sizeof(union ib_gid));
380 				gids[i].gid_type =
381 				    port_gid_table->gids[i].gid_type;
382 			}
383 		}
384 	}
385 	spin_unlock_bh(&iboe->lock);
386 
387 	if (!ret && hw_update) {
388 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
389 		kfree(gids);
390 	}
391 	return ret;
392 }
393 
mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev * ibdev,const struct ib_gid_attr * attr)394 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
395 				    const struct ib_gid_attr *attr)
396 {
397 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
398 	struct gid_cache_context *ctx = NULL;
399 	struct mlx4_port_gid_table   *port_gid_table;
400 	int real_index = -EINVAL;
401 	int i;
402 	unsigned long flags;
403 	u8 port_num = attr->port_num;
404 
405 	if (port_num > MLX4_MAX_PORTS)
406 		return -EINVAL;
407 
408 	if (mlx4_is_bonded(ibdev->dev))
409 		port_num = 1;
410 
411 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
412 		return attr->index;
413 
414 	spin_lock_irqsave(&iboe->lock, flags);
415 	port_gid_table = &iboe->gids[port_num - 1];
416 
417 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
418 		if (!memcmp(&port_gid_table->gids[i].gid,
419 			    &attr->gid, sizeof(attr->gid)) &&
420 		    attr->gid_type == port_gid_table->gids[i].gid_type) {
421 			ctx = port_gid_table->gids[i].ctx;
422 			break;
423 		}
424 	if (ctx)
425 		real_index = ctx->real_index;
426 	spin_unlock_irqrestore(&iboe->lock, flags);
427 	return real_index;
428 }
429 
430 #define field_avail(type, fld, sz) (offsetof(type, fld) + \
431 				    sizeof(((type *)0)->fld) <= (sz))
432 
mlx4_ib_query_device(struct ib_device * ibdev,struct ib_device_attr * props,struct ib_udata * uhw)433 static int mlx4_ib_query_device(struct ib_device *ibdev,
434 				struct ib_device_attr *props,
435 				struct ib_udata *uhw)
436 {
437 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
438 	struct ib_smp *in_mad  = NULL;
439 	struct ib_smp *out_mad = NULL;
440 	int err;
441 	int have_ib_ports;
442 	struct mlx4_uverbs_ex_query_device cmd;
443 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
444 	struct mlx4_clock_params clock_params;
445 
446 	if (uhw->inlen) {
447 		if (uhw->inlen < sizeof(cmd))
448 			return -EINVAL;
449 
450 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
451 		if (err)
452 			return err;
453 
454 		if (cmd.comp_mask)
455 			return -EINVAL;
456 
457 		if (cmd.reserved)
458 			return -EINVAL;
459 	}
460 
461 	resp.response_length = offsetof(typeof(resp), response_length) +
462 		sizeof(resp.response_length);
463 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
464 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
465 	err = -ENOMEM;
466 	if (!in_mad || !out_mad)
467 		goto out;
468 
469 	init_query_mad(in_mad);
470 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
471 
472 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
473 			   1, NULL, NULL, in_mad, out_mad);
474 	if (err)
475 		goto out;
476 
477 	memset(props, 0, sizeof *props);
478 
479 	have_ib_ports = num_ib_ports(dev->dev);
480 
481 	props->fw_ver = dev->dev->caps.fw_ver;
482 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
483 		IB_DEVICE_PORT_ACTIVE_EVENT		|
484 		IB_DEVICE_SYS_IMAGE_GUID		|
485 		IB_DEVICE_RC_RNR_NAK_GEN		|
486 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
487 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
488 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
489 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
490 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
491 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
492 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
493 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
494 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
495 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
496 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
497 	if (dev->dev->caps.max_gso_sz &&
498 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
499 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
500 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
501 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
502 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
503 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
504 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
505 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
506 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
507 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
508 		props->device_cap_flags |= IB_DEVICE_XRC;
509 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
510 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
511 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
512 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
513 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
514 		else
515 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
516 	}
517 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
518 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
519 
520 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
521 
522 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
523 		0xffffff;
524 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
525 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
526 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
527 
528 	props->max_mr_size	   = ~0ull;
529 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
530 	props->max_qp		   = dev->dev->quotas.qp;
531 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
532 	props->max_send_sge =
533 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
534 	props->max_recv_sge =
535 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
536 	props->max_sge_rd = MLX4_MAX_SGE_RD;
537 	props->max_cq		   = dev->dev->quotas.cq;
538 	props->max_cqe		   = dev->dev->caps.max_cqes;
539 	props->max_mr		   = dev->dev->quotas.mpt;
540 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
541 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
542 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
543 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
544 	props->max_srq		   = dev->dev->quotas.srq;
545 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
546 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
547 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
548 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
549 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
550 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
551 	props->masked_atomic_cap   = props->atomic_cap;
552 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
553 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
554 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
555 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
556 					   props->max_mcast_grp;
557 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
558 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
559 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
560 	props->max_ah = INT_MAX;
561 
562 	if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
563 	    mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
564 		if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
565 			props->rss_caps.max_rwq_indirection_tables =
566 				props->max_qp;
567 			props->rss_caps.max_rwq_indirection_table_size =
568 				dev->dev->caps.max_rss_tbl_sz;
569 			props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
570 			props->max_wq_type_rq = props->max_qp;
571 		}
572 
573 		if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
574 			props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
575 	}
576 
577 	props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
578 	props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
579 
580 	if (!mlx4_is_slave(dev->dev))
581 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
582 
583 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
584 		resp.response_length += sizeof(resp.hca_core_clock_offset);
585 		if (!err && !mlx4_is_slave(dev->dev)) {
586 			resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
587 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
588 		}
589 	}
590 
591 	if (uhw->outlen >= resp.response_length +
592 	    sizeof(resp.max_inl_recv_sz)) {
593 		resp.response_length += sizeof(resp.max_inl_recv_sz);
594 		resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
595 			sizeof(struct mlx4_wqe_data_seg);
596 	}
597 
598 	if (field_avail(typeof(resp), rss_caps, uhw->outlen)) {
599 		if (props->rss_caps.supported_qpts) {
600 			resp.rss_caps.rx_hash_function =
601 				MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
602 
603 			resp.rss_caps.rx_hash_fields_mask =
604 				MLX4_IB_RX_HASH_SRC_IPV4 |
605 				MLX4_IB_RX_HASH_DST_IPV4 |
606 				MLX4_IB_RX_HASH_SRC_IPV6 |
607 				MLX4_IB_RX_HASH_DST_IPV6 |
608 				MLX4_IB_RX_HASH_SRC_PORT_TCP |
609 				MLX4_IB_RX_HASH_DST_PORT_TCP |
610 				MLX4_IB_RX_HASH_SRC_PORT_UDP |
611 				MLX4_IB_RX_HASH_DST_PORT_UDP;
612 
613 			if (dev->dev->caps.tunnel_offload_mode ==
614 			    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
615 				resp.rss_caps.rx_hash_fields_mask |=
616 					MLX4_IB_RX_HASH_INNER;
617 		}
618 		resp.response_length = offsetof(typeof(resp), rss_caps) +
619 				       sizeof(resp.rss_caps);
620 	}
621 
622 	if (field_avail(typeof(resp), tso_caps, uhw->outlen)) {
623 		if (dev->dev->caps.max_gso_sz &&
624 		    ((mlx4_ib_port_link_layer(ibdev, 1) ==
625 		    IB_LINK_LAYER_ETHERNET) ||
626 		    (mlx4_ib_port_link_layer(ibdev, 2) ==
627 		    IB_LINK_LAYER_ETHERNET))) {
628 			resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
629 			resp.tso_caps.supported_qpts |=
630 				1 << IB_QPT_RAW_PACKET;
631 		}
632 		resp.response_length = offsetof(typeof(resp), tso_caps) +
633 				       sizeof(resp.tso_caps);
634 	}
635 
636 	if (uhw->outlen) {
637 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
638 		if (err)
639 			goto out;
640 	}
641 out:
642 	kfree(in_mad);
643 	kfree(out_mad);
644 
645 	return err;
646 }
647 
648 static enum rdma_link_layer
mlx4_ib_port_link_layer(struct ib_device * device,u8 port_num)649 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
650 {
651 	struct mlx4_dev *dev = to_mdev(device)->dev;
652 
653 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
654 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
655 }
656 
ib_link_query_port(struct ib_device * ibdev,u8 port,struct ib_port_attr * props,int netw_view)657 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
658 			      struct ib_port_attr *props, int netw_view)
659 {
660 	struct ib_smp *in_mad  = NULL;
661 	struct ib_smp *out_mad = NULL;
662 	int ext_active_speed;
663 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
664 	int err = -ENOMEM;
665 
666 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
667 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
668 	if (!in_mad || !out_mad)
669 		goto out;
670 
671 	init_query_mad(in_mad);
672 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
673 	in_mad->attr_mod = cpu_to_be32(port);
674 
675 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
676 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
677 
678 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
679 				in_mad, out_mad);
680 	if (err)
681 		goto out;
682 
683 
684 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
685 	props->lmc		= out_mad->data[34] & 0x7;
686 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
687 	props->sm_sl		= out_mad->data[36] & 0xf;
688 	props->state		= out_mad->data[32] & 0xf;
689 	props->phys_state	= out_mad->data[33] >> 4;
690 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
691 	if (netw_view)
692 		props->gid_tbl_len = out_mad->data[50];
693 	else
694 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
695 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
696 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
697 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
698 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
699 	props->active_width	= out_mad->data[31] & 0xf;
700 	props->active_speed	= out_mad->data[35] >> 4;
701 	props->max_mtu		= out_mad->data[41] & 0xf;
702 	props->active_mtu	= out_mad->data[36] >> 4;
703 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
704 	props->max_vl_num	= out_mad->data[37] >> 4;
705 	props->init_type_reply	= out_mad->data[41] >> 4;
706 
707 	/* Check if extended speeds (EDR/FDR/...) are supported */
708 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
709 		ext_active_speed = out_mad->data[62] >> 4;
710 
711 		switch (ext_active_speed) {
712 		case 1:
713 			props->active_speed = IB_SPEED_FDR;
714 			break;
715 		case 2:
716 			props->active_speed = IB_SPEED_EDR;
717 			break;
718 		}
719 	}
720 
721 	/* If reported active speed is QDR, check if is FDR-10 */
722 	if (props->active_speed == IB_SPEED_QDR) {
723 		init_query_mad(in_mad);
724 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
725 		in_mad->attr_mod = cpu_to_be32(port);
726 
727 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
728 				   NULL, NULL, in_mad, out_mad);
729 		if (err)
730 			goto out;
731 
732 		/* Checking LinkSpeedActive for FDR-10 */
733 		if (out_mad->data[15] & 0x1)
734 			props->active_speed = IB_SPEED_FDR10;
735 	}
736 
737 	/* Avoid wrong speed value returned by FW if the IB link is down. */
738 	if (props->state == IB_PORT_DOWN)
739 		 props->active_speed = IB_SPEED_SDR;
740 
741 out:
742 	kfree(in_mad);
743 	kfree(out_mad);
744 	return err;
745 }
746 
state_to_phys_state(enum ib_port_state state)747 static u8 state_to_phys_state(enum ib_port_state state)
748 {
749 	return state == IB_PORT_ACTIVE ? 5 : 3;
750 }
751 
eth_link_query_port(struct ib_device * ibdev,u8 port,struct ib_port_attr * props)752 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
753 			       struct ib_port_attr *props)
754 {
755 
756 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
757 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
758 	struct net_device *ndev;
759 	enum ib_mtu tmp;
760 	struct mlx4_cmd_mailbox *mailbox;
761 	int err = 0;
762 	int is_bonded = mlx4_is_bonded(mdev->dev);
763 
764 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
765 	if (IS_ERR(mailbox))
766 		return PTR_ERR(mailbox);
767 
768 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
769 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
770 			   MLX4_CMD_WRAPPED);
771 	if (err)
772 		goto out;
773 
774 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
775 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
776 					   IB_WIDTH_4X : IB_WIDTH_1X;
777 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
778 					   IB_SPEED_FDR : IB_SPEED_QDR;
779 	props->port_cap_flags	= IB_PORT_CM_SUP;
780 	props->ip_gids = true;
781 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
782 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
783 	props->pkey_tbl_len	= 1;
784 	props->max_mtu		= IB_MTU_4096;
785 	props->max_vl_num	= 2;
786 	props->state		= IB_PORT_DOWN;
787 	props->phys_state	= state_to_phys_state(props->state);
788 	props->active_mtu	= IB_MTU_256;
789 	spin_lock_bh(&iboe->lock);
790 	ndev = iboe->netdevs[port - 1];
791 	if (ndev && is_bonded) {
792 		rcu_read_lock(); /* required to get upper dev */
793 		ndev = netdev_master_upper_dev_get_rcu(ndev);
794 		rcu_read_unlock();
795 	}
796 	if (!ndev)
797 		goto out_unlock;
798 
799 	tmp = iboe_get_mtu(ndev->mtu);
800 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
801 
802 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
803 					IB_PORT_ACTIVE : IB_PORT_DOWN;
804 	props->phys_state	= state_to_phys_state(props->state);
805 out_unlock:
806 	spin_unlock_bh(&iboe->lock);
807 out:
808 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
809 	return err;
810 }
811 
__mlx4_ib_query_port(struct ib_device * ibdev,u8 port,struct ib_port_attr * props,int netw_view)812 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
813 			 struct ib_port_attr *props, int netw_view)
814 {
815 	int err;
816 
817 	/* props being zeroed by the caller, avoid zeroing it here */
818 
819 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
820 		ib_link_query_port(ibdev, port, props, netw_view) :
821 				eth_link_query_port(ibdev, port, props);
822 
823 	return err;
824 }
825 
mlx4_ib_query_port(struct ib_device * ibdev,u8 port,struct ib_port_attr * props)826 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
827 			      struct ib_port_attr *props)
828 {
829 	/* returns host view */
830 	return __mlx4_ib_query_port(ibdev, port, props, 0);
831 }
832 
__mlx4_ib_query_gid(struct ib_device * ibdev,u8 port,int index,union ib_gid * gid,int netw_view)833 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
834 			union ib_gid *gid, int netw_view)
835 {
836 	struct ib_smp *in_mad  = NULL;
837 	struct ib_smp *out_mad = NULL;
838 	int err = -ENOMEM;
839 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
840 	int clear = 0;
841 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
842 
843 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
844 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
845 	if (!in_mad || !out_mad)
846 		goto out;
847 
848 	init_query_mad(in_mad);
849 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
850 	in_mad->attr_mod = cpu_to_be32(port);
851 
852 	if (mlx4_is_mfunc(dev->dev) && netw_view)
853 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
854 
855 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
856 	if (err)
857 		goto out;
858 
859 	memcpy(gid->raw, out_mad->data + 8, 8);
860 
861 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
862 		if (index) {
863 			/* For any index > 0, return the null guid */
864 			err = 0;
865 			clear = 1;
866 			goto out;
867 		}
868 	}
869 
870 	init_query_mad(in_mad);
871 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
872 	in_mad->attr_mod = cpu_to_be32(index / 8);
873 
874 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
875 			   NULL, NULL, in_mad, out_mad);
876 	if (err)
877 		goto out;
878 
879 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
880 
881 out:
882 	if (clear)
883 		memset(gid->raw + 8, 0, 8);
884 	kfree(in_mad);
885 	kfree(out_mad);
886 	return err;
887 }
888 
mlx4_ib_query_gid(struct ib_device * ibdev,u8 port,int index,union ib_gid * gid)889 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
890 			     union ib_gid *gid)
891 {
892 	if (rdma_protocol_ib(ibdev, port))
893 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
894 	return 0;
895 }
896 
mlx4_ib_query_sl2vl(struct ib_device * ibdev,u8 port,u64 * sl2vl_tbl)897 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
898 {
899 	union sl2vl_tbl_to_u64 sl2vl64;
900 	struct ib_smp *in_mad  = NULL;
901 	struct ib_smp *out_mad = NULL;
902 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
903 	int err = -ENOMEM;
904 	int jj;
905 
906 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
907 		*sl2vl_tbl = 0;
908 		return 0;
909 	}
910 
911 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
912 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
913 	if (!in_mad || !out_mad)
914 		goto out;
915 
916 	init_query_mad(in_mad);
917 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
918 	in_mad->attr_mod = 0;
919 
920 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
921 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
922 
923 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
924 			   in_mad, out_mad);
925 	if (err)
926 		goto out;
927 
928 	for (jj = 0; jj < 8; jj++)
929 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
930 	*sl2vl_tbl = sl2vl64.sl64;
931 
932 out:
933 	kfree(in_mad);
934 	kfree(out_mad);
935 	return err;
936 }
937 
mlx4_init_sl2vl_tbl(struct mlx4_ib_dev * mdev)938 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
939 {
940 	u64 sl2vl;
941 	int i;
942 	int err;
943 
944 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
945 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
946 			continue;
947 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
948 		if (err) {
949 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
950 			       i, err);
951 			sl2vl = 0;
952 		}
953 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
954 	}
955 }
956 
__mlx4_ib_query_pkey(struct ib_device * ibdev,u8 port,u16 index,u16 * pkey,int netw_view)957 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
958 			 u16 *pkey, int netw_view)
959 {
960 	struct ib_smp *in_mad  = NULL;
961 	struct ib_smp *out_mad = NULL;
962 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
963 	int err = -ENOMEM;
964 
965 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
966 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
967 	if (!in_mad || !out_mad)
968 		goto out;
969 
970 	init_query_mad(in_mad);
971 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
972 	in_mad->attr_mod = cpu_to_be32(index / 32);
973 
974 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
975 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
976 
977 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
978 			   in_mad, out_mad);
979 	if (err)
980 		goto out;
981 
982 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
983 
984 out:
985 	kfree(in_mad);
986 	kfree(out_mad);
987 	return err;
988 }
989 
mlx4_ib_query_pkey(struct ib_device * ibdev,u8 port,u16 index,u16 * pkey)990 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
991 {
992 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
993 }
994 
mlx4_ib_modify_device(struct ib_device * ibdev,int mask,struct ib_device_modify * props)995 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
996 				 struct ib_device_modify *props)
997 {
998 	struct mlx4_cmd_mailbox *mailbox;
999 	unsigned long flags;
1000 
1001 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
1002 		return -EOPNOTSUPP;
1003 
1004 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
1005 		return 0;
1006 
1007 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
1008 		return -EOPNOTSUPP;
1009 
1010 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
1011 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1012 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1013 
1014 	/*
1015 	 * If possible, pass node desc to FW, so it can generate
1016 	 * a 144 trap.  If cmd fails, just ignore.
1017 	 */
1018 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1019 	if (IS_ERR(mailbox))
1020 		return 0;
1021 
1022 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1023 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1024 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1025 
1026 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1027 
1028 	return 0;
1029 }
1030 
mlx4_ib_SET_PORT(struct mlx4_ib_dev * dev,u8 port,int reset_qkey_viols,u32 cap_mask)1031 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
1032 			    u32 cap_mask)
1033 {
1034 	struct mlx4_cmd_mailbox *mailbox;
1035 	int err;
1036 
1037 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1038 	if (IS_ERR(mailbox))
1039 		return PTR_ERR(mailbox);
1040 
1041 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1042 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
1043 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1044 	} else {
1045 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1046 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1047 	}
1048 
1049 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1050 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1051 		       MLX4_CMD_WRAPPED);
1052 
1053 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1054 	return err;
1055 }
1056 
mlx4_ib_modify_port(struct ib_device * ibdev,u8 port,int mask,struct ib_port_modify * props)1057 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1058 			       struct ib_port_modify *props)
1059 {
1060 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1061 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1062 	struct ib_port_attr attr;
1063 	u32 cap_mask;
1064 	int err;
1065 
1066 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1067 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1068 	 * violations and port capabilities are not meaningful.
1069 	 */
1070 	if (is_eth)
1071 		return 0;
1072 
1073 	mutex_lock(&mdev->cap_mask_mutex);
1074 
1075 	err = ib_query_port(ibdev, port, &attr);
1076 	if (err)
1077 		goto out;
1078 
1079 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1080 		~props->clr_port_cap_mask;
1081 
1082 	err = mlx4_ib_SET_PORT(mdev, port,
1083 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1084 			       cap_mask);
1085 
1086 out:
1087 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1088 	return err;
1089 }
1090 
mlx4_ib_alloc_ucontext(struct ib_device * ibdev,struct ib_udata * udata)1091 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1092 						  struct ib_udata *udata)
1093 {
1094 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1095 	struct mlx4_ib_ucontext *context;
1096 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1097 	struct mlx4_ib_alloc_ucontext_resp resp;
1098 	int err;
1099 
1100 	if (!dev->ib_active)
1101 		return ERR_PTR(-EAGAIN);
1102 
1103 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1104 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1105 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1106 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1107 	} else {
1108 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1109 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1110 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1111 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1112 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1113 	}
1114 
1115 	context = kzalloc(sizeof(*context), GFP_KERNEL);
1116 	if (!context)
1117 		return ERR_PTR(-ENOMEM);
1118 
1119 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1120 	if (err) {
1121 		kfree(context);
1122 		return ERR_PTR(err);
1123 	}
1124 
1125 	INIT_LIST_HEAD(&context->db_page_list);
1126 	mutex_init(&context->db_page_mutex);
1127 
1128 	INIT_LIST_HEAD(&context->wqn_ranges_list);
1129 	mutex_init(&context->wqn_ranges_mutex);
1130 
1131 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1132 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1133 	else
1134 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1135 
1136 	if (err) {
1137 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1138 		kfree(context);
1139 		return ERR_PTR(-EFAULT);
1140 	}
1141 
1142 	return &context->ibucontext;
1143 }
1144 
mlx4_ib_dealloc_ucontext(struct ib_ucontext * ibcontext)1145 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1146 {
1147 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1148 
1149 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1150 	kfree(context);
1151 
1152 	return 0;
1153 }
1154 
mlx4_ib_vma_open(struct vm_area_struct * area)1155 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1156 {
1157 	/* vma_open is called when a new VMA is created on top of our VMA.
1158 	 * This is done through either mremap flow or split_vma (usually due
1159 	 * to mlock, madvise, munmap, etc.). We do not support a clone of the
1160 	 * vma, as this VMA is strongly hardware related. Therefore we set the
1161 	 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1162 	 * calling us again and trying to do incorrect actions. We assume that
1163 	 * the original vma size is exactly a single page that there will be no
1164 	 * "splitting" operations on.
1165 	 */
1166 	area->vm_ops = NULL;
1167 }
1168 
mlx4_ib_vma_close(struct vm_area_struct * area)1169 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1170 {
1171 	struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1172 
1173 	/* It's guaranteed that all VMAs opened on a FD are closed before the
1174 	 * file itself is closed, therefore no sync is needed with the regular
1175 	 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1176 	 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1177 	 * The close operation is usually called under mm->mmap_sem except when
1178 	 * process is exiting.  The exiting case is handled explicitly as part
1179 	 * of mlx4_ib_disassociate_ucontext.
1180 	 */
1181 	mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1182 				area->vm_private_data;
1183 
1184 	/* set the vma context pointer to null in the mlx4_ib driver's private
1185 	 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1186 	 */
1187 	mlx4_ib_vma_priv_data->vma = NULL;
1188 }
1189 
1190 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1191 	.open = mlx4_ib_vma_open,
1192 	.close = mlx4_ib_vma_close
1193 };
1194 
mlx4_ib_disassociate_ucontext(struct ib_ucontext * ibcontext)1195 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1196 {
1197 	int i;
1198 	struct vm_area_struct *vma;
1199 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1200 
1201 	/* need to protect from a race on closing the vma as part of
1202 	 * mlx4_ib_vma_close().
1203 	 */
1204 	for (i = 0; i < HW_BAR_COUNT; i++) {
1205 		vma = context->hw_bar_info[i].vma;
1206 		if (!vma)
1207 			continue;
1208 
1209 		zap_vma_ptes(context->hw_bar_info[i].vma,
1210 			     context->hw_bar_info[i].vma->vm_start, PAGE_SIZE);
1211 
1212 		context->hw_bar_info[i].vma->vm_flags &=
1213 			~(VM_SHARED | VM_MAYSHARE);
1214 		/* context going to be destroyed, should not access ops any more */
1215 		context->hw_bar_info[i].vma->vm_ops = NULL;
1216 	}
1217 }
1218 
mlx4_ib_set_vma_data(struct vm_area_struct * vma,struct mlx4_ib_vma_private_data * vma_private_data)1219 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1220 				 struct mlx4_ib_vma_private_data *vma_private_data)
1221 {
1222 	vma_private_data->vma = vma;
1223 	vma->vm_private_data = vma_private_data;
1224 	vma->vm_ops =  &mlx4_ib_vm_ops;
1225 }
1226 
mlx4_ib_mmap(struct ib_ucontext * context,struct vm_area_struct * vma)1227 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1228 {
1229 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1230 	struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1231 
1232 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1233 		return -EINVAL;
1234 
1235 	if (vma->vm_pgoff == 0) {
1236 		/* We prevent double mmaping on same context */
1237 		if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1238 			return -EINVAL;
1239 
1240 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1241 
1242 		if (io_remap_pfn_range(vma, vma->vm_start,
1243 				       to_mucontext(context)->uar.pfn,
1244 				       PAGE_SIZE, vma->vm_page_prot))
1245 			return -EAGAIN;
1246 
1247 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1248 
1249 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1250 		/* We prevent double mmaping on same context */
1251 		if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1252 			return -EINVAL;
1253 
1254 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1255 
1256 		if (io_remap_pfn_range(vma, vma->vm_start,
1257 				       to_mucontext(context)->uar.pfn +
1258 				       dev->dev->caps.num_uars,
1259 				       PAGE_SIZE, vma->vm_page_prot))
1260 			return -EAGAIN;
1261 
1262 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1263 
1264 	} else if (vma->vm_pgoff == 3) {
1265 		struct mlx4_clock_params params;
1266 		int ret;
1267 
1268 		/* We prevent double mmaping on same context */
1269 		if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1270 			return -EINVAL;
1271 
1272 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1273 
1274 		if (ret)
1275 			return ret;
1276 
1277 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1278 		if (io_remap_pfn_range(vma, vma->vm_start,
1279 				       (pci_resource_start(dev->dev->persist->pdev,
1280 							   params.bar) +
1281 					params.offset)
1282 				       >> PAGE_SHIFT,
1283 				       PAGE_SIZE, vma->vm_page_prot))
1284 			return -EAGAIN;
1285 
1286 		mlx4_ib_set_vma_data(vma,
1287 				     &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1288 	} else {
1289 		return -EINVAL;
1290 	}
1291 
1292 	return 0;
1293 }
1294 
mlx4_ib_alloc_pd(struct ib_device * ibdev,struct ib_ucontext * context,struct ib_udata * udata)1295 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1296 				      struct ib_ucontext *context,
1297 				      struct ib_udata *udata)
1298 {
1299 	struct mlx4_ib_pd *pd;
1300 	int err;
1301 
1302 	pd = kzalloc(sizeof(*pd), GFP_KERNEL);
1303 	if (!pd)
1304 		return ERR_PTR(-ENOMEM);
1305 
1306 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1307 	if (err) {
1308 		kfree(pd);
1309 		return ERR_PTR(err);
1310 	}
1311 
1312 	if (context)
1313 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1314 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1315 			kfree(pd);
1316 			return ERR_PTR(-EFAULT);
1317 		}
1318 	return &pd->ibpd;
1319 }
1320 
mlx4_ib_dealloc_pd(struct ib_pd * pd)1321 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1322 {
1323 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1324 	kfree(pd);
1325 
1326 	return 0;
1327 }
1328 
mlx4_ib_alloc_xrcd(struct ib_device * ibdev,struct ib_ucontext * context,struct ib_udata * udata)1329 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1330 					  struct ib_ucontext *context,
1331 					  struct ib_udata *udata)
1332 {
1333 	struct mlx4_ib_xrcd *xrcd;
1334 	struct ib_cq_init_attr cq_attr = {};
1335 	int err;
1336 
1337 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1338 		return ERR_PTR(-ENOSYS);
1339 
1340 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1341 	if (!xrcd)
1342 		return ERR_PTR(-ENOMEM);
1343 
1344 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1345 	if (err)
1346 		goto err1;
1347 
1348 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1349 	if (IS_ERR(xrcd->pd)) {
1350 		err = PTR_ERR(xrcd->pd);
1351 		goto err2;
1352 	}
1353 
1354 	cq_attr.cqe = 1;
1355 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1356 	if (IS_ERR(xrcd->cq)) {
1357 		err = PTR_ERR(xrcd->cq);
1358 		goto err3;
1359 	}
1360 
1361 	return &xrcd->ibxrcd;
1362 
1363 err3:
1364 	ib_dealloc_pd(xrcd->pd);
1365 err2:
1366 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1367 err1:
1368 	kfree(xrcd);
1369 	return ERR_PTR(err);
1370 }
1371 
mlx4_ib_dealloc_xrcd(struct ib_xrcd * xrcd)1372 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1373 {
1374 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1375 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1376 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1377 	kfree(xrcd);
1378 
1379 	return 0;
1380 }
1381 
add_gid_entry(struct ib_qp * ibqp,union ib_gid * gid)1382 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1383 {
1384 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1385 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1386 	struct mlx4_ib_gid_entry *ge;
1387 
1388 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1389 	if (!ge)
1390 		return -ENOMEM;
1391 
1392 	ge->gid = *gid;
1393 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1394 		ge->port = mqp->port;
1395 		ge->added = 1;
1396 	}
1397 
1398 	mutex_lock(&mqp->mutex);
1399 	list_add_tail(&ge->list, &mqp->gid_list);
1400 	mutex_unlock(&mqp->mutex);
1401 
1402 	return 0;
1403 }
1404 
mlx4_ib_delete_counters_table(struct mlx4_ib_dev * ibdev,struct mlx4_ib_counters * ctr_table)1405 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1406 					  struct mlx4_ib_counters *ctr_table)
1407 {
1408 	struct counter_index *counter, *tmp_count;
1409 
1410 	mutex_lock(&ctr_table->mutex);
1411 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1412 				 list) {
1413 		if (counter->allocated)
1414 			mlx4_counter_free(ibdev->dev, counter->index);
1415 		list_del(&counter->list);
1416 		kfree(counter);
1417 	}
1418 	mutex_unlock(&ctr_table->mutex);
1419 }
1420 
mlx4_ib_add_mc(struct mlx4_ib_dev * mdev,struct mlx4_ib_qp * mqp,union ib_gid * gid)1421 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1422 		   union ib_gid *gid)
1423 {
1424 	struct net_device *ndev;
1425 	int ret = 0;
1426 
1427 	if (!mqp->port)
1428 		return 0;
1429 
1430 	spin_lock_bh(&mdev->iboe.lock);
1431 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1432 	if (ndev)
1433 		dev_hold(ndev);
1434 	spin_unlock_bh(&mdev->iboe.lock);
1435 
1436 	if (ndev) {
1437 		ret = 1;
1438 		dev_put(ndev);
1439 	}
1440 
1441 	return ret;
1442 }
1443 
1444 struct mlx4_ib_steering {
1445 	struct list_head list;
1446 	struct mlx4_flow_reg_id reg_id;
1447 	union ib_gid gid;
1448 };
1449 
1450 #define LAST_ETH_FIELD vlan_tag
1451 #define LAST_IB_FIELD sl
1452 #define LAST_IPV4_FIELD dst_ip
1453 #define LAST_TCP_UDP_FIELD src_port
1454 
1455 /* Field is the last supported field */
1456 #define FIELDS_NOT_SUPPORTED(filter, field)\
1457 	memchr_inv((void *)&filter.field  +\
1458 		   sizeof(filter.field), 0,\
1459 		   sizeof(filter) -\
1460 		   offsetof(typeof(filter), field) -\
1461 		   sizeof(filter.field))
1462 
parse_flow_attr(struct mlx4_dev * dev,u32 qp_num,union ib_flow_spec * ib_spec,struct _rule_hw * mlx4_spec)1463 static int parse_flow_attr(struct mlx4_dev *dev,
1464 			   u32 qp_num,
1465 			   union ib_flow_spec *ib_spec,
1466 			   struct _rule_hw *mlx4_spec)
1467 {
1468 	enum mlx4_net_trans_rule_id type;
1469 
1470 	switch (ib_spec->type) {
1471 	case IB_FLOW_SPEC_ETH:
1472 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1473 			return -ENOTSUPP;
1474 
1475 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1476 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1477 		       ETH_ALEN);
1478 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1479 		       ETH_ALEN);
1480 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1481 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1482 		break;
1483 	case IB_FLOW_SPEC_IB:
1484 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1485 			return -ENOTSUPP;
1486 
1487 		type = MLX4_NET_TRANS_RULE_ID_IB;
1488 		mlx4_spec->ib.l3_qpn =
1489 			cpu_to_be32(qp_num);
1490 		mlx4_spec->ib.qpn_mask =
1491 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1492 		break;
1493 
1494 
1495 	case IB_FLOW_SPEC_IPV4:
1496 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1497 			return -ENOTSUPP;
1498 
1499 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1500 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1501 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1502 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1503 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1504 		break;
1505 
1506 	case IB_FLOW_SPEC_TCP:
1507 	case IB_FLOW_SPEC_UDP:
1508 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1509 			return -ENOTSUPP;
1510 
1511 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1512 					MLX4_NET_TRANS_RULE_ID_TCP :
1513 					MLX4_NET_TRANS_RULE_ID_UDP;
1514 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1515 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1516 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1517 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1518 		break;
1519 
1520 	default:
1521 		return -EINVAL;
1522 	}
1523 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1524 	    mlx4_hw_rule_sz(dev, type) < 0)
1525 		return -EINVAL;
1526 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1527 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1528 	return mlx4_hw_rule_sz(dev, type);
1529 }
1530 
1531 struct default_rules {
1532 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1533 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1534 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1535 	__u8  link_layer;
1536 };
1537 static const struct default_rules default_table[] = {
1538 	{
1539 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1540 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1541 		.rules_create_list = {IB_FLOW_SPEC_IB},
1542 		.link_layer = IB_LINK_LAYER_INFINIBAND
1543 	}
1544 };
1545 
__mlx4_ib_default_rules_match(struct ib_qp * qp,struct ib_flow_attr * flow_attr)1546 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1547 					 struct ib_flow_attr *flow_attr)
1548 {
1549 	int i, j, k;
1550 	void *ib_flow;
1551 	const struct default_rules *pdefault_rules = default_table;
1552 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1553 
1554 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1555 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1556 		memset(&field_types, 0, sizeof(field_types));
1557 
1558 		if (link_layer != pdefault_rules->link_layer)
1559 			continue;
1560 
1561 		ib_flow = flow_attr + 1;
1562 		/* we assume the specs are sorted */
1563 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1564 		     j < flow_attr->num_of_specs; k++) {
1565 			union ib_flow_spec *current_flow =
1566 				(union ib_flow_spec *)ib_flow;
1567 
1568 			/* same layer but different type */
1569 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1570 			     (pdefault_rules->mandatory_fields[k] &
1571 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1572 			    (current_flow->type !=
1573 			     pdefault_rules->mandatory_fields[k]))
1574 				goto out;
1575 
1576 			/* same layer, try match next one */
1577 			if (current_flow->type ==
1578 			    pdefault_rules->mandatory_fields[k]) {
1579 				j++;
1580 				ib_flow +=
1581 					((union ib_flow_spec *)ib_flow)->size;
1582 			}
1583 		}
1584 
1585 		ib_flow = flow_attr + 1;
1586 		for (j = 0; j < flow_attr->num_of_specs;
1587 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1588 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1589 				/* same layer and same type */
1590 				if (((union ib_flow_spec *)ib_flow)->type ==
1591 				    pdefault_rules->mandatory_not_fields[k])
1592 					goto out;
1593 
1594 		return i;
1595 	}
1596 out:
1597 	return -1;
1598 }
1599 
__mlx4_ib_create_default_rules(struct mlx4_ib_dev * mdev,struct ib_qp * qp,const struct default_rules * pdefault_rules,struct _rule_hw * mlx4_spec)1600 static int __mlx4_ib_create_default_rules(
1601 		struct mlx4_ib_dev *mdev,
1602 		struct ib_qp *qp,
1603 		const struct default_rules *pdefault_rules,
1604 		struct _rule_hw *mlx4_spec) {
1605 	int size = 0;
1606 	int i;
1607 
1608 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1609 		union ib_flow_spec ib_spec = {};
1610 		int ret;
1611 
1612 		switch (pdefault_rules->rules_create_list[i]) {
1613 		case 0:
1614 			/* no rule */
1615 			continue;
1616 		case IB_FLOW_SPEC_IB:
1617 			ib_spec.type = IB_FLOW_SPEC_IB;
1618 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1619 
1620 			break;
1621 		default:
1622 			/* invalid rule */
1623 			return -EINVAL;
1624 		}
1625 		/* We must put empty rule, qpn is being ignored */
1626 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1627 				      mlx4_spec);
1628 		if (ret < 0) {
1629 			pr_info("invalid parsing\n");
1630 			return -EINVAL;
1631 		}
1632 
1633 		mlx4_spec = (void *)mlx4_spec + ret;
1634 		size += ret;
1635 	}
1636 	return size;
1637 }
1638 
__mlx4_ib_create_flow(struct ib_qp * qp,struct ib_flow_attr * flow_attr,int domain,enum mlx4_net_trans_promisc_mode flow_type,u64 * reg_id)1639 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1640 			  int domain,
1641 			  enum mlx4_net_trans_promisc_mode flow_type,
1642 			  u64 *reg_id)
1643 {
1644 	int ret, i;
1645 	int size = 0;
1646 	void *ib_flow;
1647 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1648 	struct mlx4_cmd_mailbox *mailbox;
1649 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1650 	int default_flow;
1651 
1652 	static const u16 __mlx4_domain[] = {
1653 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1654 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1655 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1656 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1657 	};
1658 
1659 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1660 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1661 		return -EINVAL;
1662 	}
1663 
1664 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1665 		pr_err("Invalid domain value %d\n", domain);
1666 		return -EINVAL;
1667 	}
1668 
1669 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1670 		return -EINVAL;
1671 
1672 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1673 	if (IS_ERR(mailbox))
1674 		return PTR_ERR(mailbox);
1675 	ctrl = mailbox->buf;
1676 
1677 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1678 				 flow_attr->priority);
1679 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1680 	ctrl->port = flow_attr->port;
1681 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1682 
1683 	ib_flow = flow_attr + 1;
1684 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1685 	/* Add default flows */
1686 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1687 	if (default_flow >= 0) {
1688 		ret = __mlx4_ib_create_default_rules(
1689 				mdev, qp, default_table + default_flow,
1690 				mailbox->buf + size);
1691 		if (ret < 0) {
1692 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1693 			return -EINVAL;
1694 		}
1695 		size += ret;
1696 	}
1697 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1698 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1699 				      mailbox->buf + size);
1700 		if (ret < 0) {
1701 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1702 			return -EINVAL;
1703 		}
1704 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1705 		size += ret;
1706 	}
1707 
1708 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1709 	    flow_attr->num_of_specs == 1) {
1710 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1711 		enum ib_flow_spec_type header_spec =
1712 			((union ib_flow_spec *)(flow_attr + 1))->type;
1713 
1714 		if (header_spec == IB_FLOW_SPEC_ETH)
1715 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1716 	}
1717 
1718 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1719 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1720 			   MLX4_CMD_NATIVE);
1721 	if (ret == -ENOMEM)
1722 		pr_err("mcg table is full. Fail to register network rule.\n");
1723 	else if (ret == -ENXIO)
1724 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1725 	else if (ret)
1726 		pr_err("Invalid argument. Fail to register network rule.\n");
1727 
1728 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1729 	return ret;
1730 }
1731 
__mlx4_ib_destroy_flow(struct mlx4_dev * dev,u64 reg_id)1732 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1733 {
1734 	int err;
1735 	err = mlx4_cmd(dev, reg_id, 0, 0,
1736 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1737 		       MLX4_CMD_NATIVE);
1738 	if (err)
1739 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1740 		       reg_id);
1741 	return err;
1742 }
1743 
mlx4_ib_tunnel_steer_add(struct ib_qp * qp,struct ib_flow_attr * flow_attr,u64 * reg_id)1744 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1745 				    u64 *reg_id)
1746 {
1747 	void *ib_flow;
1748 	union ib_flow_spec *ib_spec;
1749 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1750 	int err = 0;
1751 
1752 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1753 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1754 		return 0; /* do nothing */
1755 
1756 	ib_flow = flow_attr + 1;
1757 	ib_spec = (union ib_flow_spec *)ib_flow;
1758 
1759 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1760 		return 0; /* do nothing */
1761 
1762 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1763 				    flow_attr->port, qp->qp_num,
1764 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1765 				    reg_id);
1766 	return err;
1767 }
1768 
mlx4_ib_add_dont_trap_rule(struct mlx4_dev * dev,struct ib_flow_attr * flow_attr,enum mlx4_net_trans_promisc_mode * type)1769 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1770 				      struct ib_flow_attr *flow_attr,
1771 				      enum mlx4_net_trans_promisc_mode *type)
1772 {
1773 	int err = 0;
1774 
1775 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1776 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1777 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1778 		return -EOPNOTSUPP;
1779 	}
1780 
1781 	if (flow_attr->num_of_specs == 0) {
1782 		type[0] = MLX4_FS_MC_SNIFFER;
1783 		type[1] = MLX4_FS_UC_SNIFFER;
1784 	} else {
1785 		union ib_flow_spec *ib_spec;
1786 
1787 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1788 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1789 			return -EINVAL;
1790 
1791 		/* if all is zero than MC and UC */
1792 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1793 			type[0] = MLX4_FS_MC_SNIFFER;
1794 			type[1] = MLX4_FS_UC_SNIFFER;
1795 		} else {
1796 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1797 					    ib_spec->eth.mask.dst_mac[1],
1798 					    ib_spec->eth.mask.dst_mac[2],
1799 					    ib_spec->eth.mask.dst_mac[3],
1800 					    ib_spec->eth.mask.dst_mac[4],
1801 					    ib_spec->eth.mask.dst_mac[5]};
1802 
1803 			/* Above xor was only on MC bit, non empty mask is valid
1804 			 * only if this bit is set and rest are zero.
1805 			 */
1806 			if (!is_zero_ether_addr(&mac[0]))
1807 				return -EINVAL;
1808 
1809 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1810 				type[0] = MLX4_FS_MC_SNIFFER;
1811 			else
1812 				type[0] = MLX4_FS_UC_SNIFFER;
1813 		}
1814 	}
1815 
1816 	return err;
1817 }
1818 
mlx4_ib_create_flow(struct ib_qp * qp,struct ib_flow_attr * flow_attr,int domain,struct ib_udata * udata)1819 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1820 				    struct ib_flow_attr *flow_attr,
1821 				    int domain, struct ib_udata *udata)
1822 {
1823 	int err = 0, i = 0, j = 0;
1824 	struct mlx4_ib_flow *mflow;
1825 	enum mlx4_net_trans_promisc_mode type[2];
1826 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1827 	int is_bonded = mlx4_is_bonded(dev);
1828 
1829 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1830 		return ERR_PTR(-EINVAL);
1831 
1832 	if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1833 		return ERR_PTR(-EOPNOTSUPP);
1834 
1835 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1836 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1837 		return ERR_PTR(-EOPNOTSUPP);
1838 
1839 	if (udata &&
1840 	    udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1841 		return ERR_PTR(-EOPNOTSUPP);
1842 
1843 	memset(type, 0, sizeof(type));
1844 
1845 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1846 	if (!mflow) {
1847 		err = -ENOMEM;
1848 		goto err_free;
1849 	}
1850 
1851 	switch (flow_attr->type) {
1852 	case IB_FLOW_ATTR_NORMAL:
1853 		/* If dont trap flag (continue match) is set, under specific
1854 		 * condition traffic be replicated to given qp,
1855 		 * without stealing it
1856 		 */
1857 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1858 			err = mlx4_ib_add_dont_trap_rule(dev,
1859 							 flow_attr,
1860 							 type);
1861 			if (err)
1862 				goto err_free;
1863 		} else {
1864 			type[0] = MLX4_FS_REGULAR;
1865 		}
1866 		break;
1867 
1868 	case IB_FLOW_ATTR_ALL_DEFAULT:
1869 		type[0] = MLX4_FS_ALL_DEFAULT;
1870 		break;
1871 
1872 	case IB_FLOW_ATTR_MC_DEFAULT:
1873 		type[0] = MLX4_FS_MC_DEFAULT;
1874 		break;
1875 
1876 	case IB_FLOW_ATTR_SNIFFER:
1877 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1878 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1879 		break;
1880 
1881 	default:
1882 		err = -EINVAL;
1883 		goto err_free;
1884 	}
1885 
1886 	while (i < ARRAY_SIZE(type) && type[i]) {
1887 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1888 					    &mflow->reg_id[i].id);
1889 		if (err)
1890 			goto err_create_flow;
1891 		if (is_bonded) {
1892 			/* Application always sees one port so the mirror rule
1893 			 * must be on port #2
1894 			 */
1895 			flow_attr->port = 2;
1896 			err = __mlx4_ib_create_flow(qp, flow_attr,
1897 						    domain, type[j],
1898 						    &mflow->reg_id[j].mirror);
1899 			flow_attr->port = 1;
1900 			if (err)
1901 				goto err_create_flow;
1902 			j++;
1903 		}
1904 
1905 		i++;
1906 	}
1907 
1908 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1909 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1910 					       &mflow->reg_id[i].id);
1911 		if (err)
1912 			goto err_create_flow;
1913 
1914 		if (is_bonded) {
1915 			flow_attr->port = 2;
1916 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1917 						       &mflow->reg_id[j].mirror);
1918 			flow_attr->port = 1;
1919 			if (err)
1920 				goto err_create_flow;
1921 			j++;
1922 		}
1923 		/* function to create mirror rule */
1924 		i++;
1925 	}
1926 
1927 	return &mflow->ibflow;
1928 
1929 err_create_flow:
1930 	while (i) {
1931 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1932 					     mflow->reg_id[i].id);
1933 		i--;
1934 	}
1935 
1936 	while (j) {
1937 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1938 					     mflow->reg_id[j].mirror);
1939 		j--;
1940 	}
1941 err_free:
1942 	kfree(mflow);
1943 	return ERR_PTR(err);
1944 }
1945 
mlx4_ib_destroy_flow(struct ib_flow * flow_id)1946 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1947 {
1948 	int err, ret = 0;
1949 	int i = 0;
1950 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1951 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1952 
1953 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1954 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1955 		if (err)
1956 			ret = err;
1957 		if (mflow->reg_id[i].mirror) {
1958 			err = __mlx4_ib_destroy_flow(mdev->dev,
1959 						     mflow->reg_id[i].mirror);
1960 			if (err)
1961 				ret = err;
1962 		}
1963 		i++;
1964 	}
1965 
1966 	kfree(mflow);
1967 	return ret;
1968 }
1969 
mlx4_ib_mcg_attach(struct ib_qp * ibqp,union ib_gid * gid,u16 lid)1970 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1971 {
1972 	int err;
1973 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1974 	struct mlx4_dev	*dev = mdev->dev;
1975 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1976 	struct mlx4_ib_steering *ib_steering = NULL;
1977 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1978 	struct mlx4_flow_reg_id	reg_id;
1979 
1980 	if (mdev->dev->caps.steering_mode ==
1981 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1982 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1983 		if (!ib_steering)
1984 			return -ENOMEM;
1985 	}
1986 
1987 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1988 				    !!(mqp->flags &
1989 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1990 				    prot, &reg_id.id);
1991 	if (err) {
1992 		pr_err("multicast attach op failed, err %d\n", err);
1993 		goto err_malloc;
1994 	}
1995 
1996 	reg_id.mirror = 0;
1997 	if (mlx4_is_bonded(dev)) {
1998 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1999 					    (mqp->port == 1) ? 2 : 1,
2000 					    !!(mqp->flags &
2001 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
2002 					    prot, &reg_id.mirror);
2003 		if (err)
2004 			goto err_add;
2005 	}
2006 
2007 	err = add_gid_entry(ibqp, gid);
2008 	if (err)
2009 		goto err_add;
2010 
2011 	if (ib_steering) {
2012 		memcpy(ib_steering->gid.raw, gid->raw, 16);
2013 		ib_steering->reg_id = reg_id;
2014 		mutex_lock(&mqp->mutex);
2015 		list_add(&ib_steering->list, &mqp->steering_rules);
2016 		mutex_unlock(&mqp->mutex);
2017 	}
2018 	return 0;
2019 
2020 err_add:
2021 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2022 			      prot, reg_id.id);
2023 	if (reg_id.mirror)
2024 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2025 				      prot, reg_id.mirror);
2026 err_malloc:
2027 	kfree(ib_steering);
2028 
2029 	return err;
2030 }
2031 
find_gid_entry(struct mlx4_ib_qp * qp,u8 * raw)2032 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
2033 {
2034 	struct mlx4_ib_gid_entry *ge;
2035 	struct mlx4_ib_gid_entry *tmp;
2036 	struct mlx4_ib_gid_entry *ret = NULL;
2037 
2038 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
2039 		if (!memcmp(raw, ge->gid.raw, 16)) {
2040 			ret = ge;
2041 			break;
2042 		}
2043 	}
2044 
2045 	return ret;
2046 }
2047 
mlx4_ib_mcg_detach(struct ib_qp * ibqp,union ib_gid * gid,u16 lid)2048 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
2049 {
2050 	int err;
2051 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
2052 	struct mlx4_dev *dev = mdev->dev;
2053 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
2054 	struct net_device *ndev;
2055 	struct mlx4_ib_gid_entry *ge;
2056 	struct mlx4_flow_reg_id reg_id = {0, 0};
2057 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
2058 
2059 	if (mdev->dev->caps.steering_mode ==
2060 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
2061 		struct mlx4_ib_steering *ib_steering;
2062 
2063 		mutex_lock(&mqp->mutex);
2064 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
2065 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
2066 				list_del(&ib_steering->list);
2067 				break;
2068 			}
2069 		}
2070 		mutex_unlock(&mqp->mutex);
2071 		if (&ib_steering->list == &mqp->steering_rules) {
2072 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
2073 			return -EINVAL;
2074 		}
2075 		reg_id = ib_steering->reg_id;
2076 		kfree(ib_steering);
2077 	}
2078 
2079 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2080 				    prot, reg_id.id);
2081 	if (err)
2082 		return err;
2083 
2084 	if (mlx4_is_bonded(dev)) {
2085 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2086 					    prot, reg_id.mirror);
2087 		if (err)
2088 			return err;
2089 	}
2090 
2091 	mutex_lock(&mqp->mutex);
2092 	ge = find_gid_entry(mqp, gid->raw);
2093 	if (ge) {
2094 		spin_lock_bh(&mdev->iboe.lock);
2095 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
2096 		if (ndev)
2097 			dev_hold(ndev);
2098 		spin_unlock_bh(&mdev->iboe.lock);
2099 		if (ndev)
2100 			dev_put(ndev);
2101 		list_del(&ge->list);
2102 		kfree(ge);
2103 	} else
2104 		pr_warn("could not find mgid entry\n");
2105 
2106 	mutex_unlock(&mqp->mutex);
2107 
2108 	return 0;
2109 }
2110 
init_node_data(struct mlx4_ib_dev * dev)2111 static int init_node_data(struct mlx4_ib_dev *dev)
2112 {
2113 	struct ib_smp *in_mad  = NULL;
2114 	struct ib_smp *out_mad = NULL;
2115 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2116 	int err = -ENOMEM;
2117 
2118 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2119 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2120 	if (!in_mad || !out_mad)
2121 		goto out;
2122 
2123 	init_query_mad(in_mad);
2124 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2125 	if (mlx4_is_master(dev->dev))
2126 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2127 
2128 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2129 	if (err)
2130 		goto out;
2131 
2132 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2133 
2134 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2135 
2136 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2137 	if (err)
2138 		goto out;
2139 
2140 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2141 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2142 
2143 out:
2144 	kfree(in_mad);
2145 	kfree(out_mad);
2146 	return err;
2147 }
2148 
show_hca(struct device * device,struct device_attribute * attr,char * buf)2149 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2150 			char *buf)
2151 {
2152 	struct mlx4_ib_dev *dev =
2153 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2154 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2155 }
2156 
show_rev(struct device * device,struct device_attribute * attr,char * buf)2157 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2158 			char *buf)
2159 {
2160 	struct mlx4_ib_dev *dev =
2161 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2162 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2163 }
2164 
show_board(struct device * device,struct device_attribute * attr,char * buf)2165 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2166 			  char *buf)
2167 {
2168 	struct mlx4_ib_dev *dev =
2169 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2170 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2171 		       dev->dev->board_id);
2172 }
2173 
2174 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2175 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2176 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2177 
2178 static struct device_attribute *mlx4_class_attributes[] = {
2179 	&dev_attr_hw_rev,
2180 	&dev_attr_hca_type,
2181 	&dev_attr_board_id
2182 };
2183 
2184 struct diag_counter {
2185 	const char *name;
2186 	u32 offset;
2187 };
2188 
2189 #define DIAG_COUNTER(_name, _offset)			\
2190 	{ .name = #_name, .offset = _offset }
2191 
2192 static const struct diag_counter diag_basic[] = {
2193 	DIAG_COUNTER(rq_num_lle, 0x00),
2194 	DIAG_COUNTER(sq_num_lle, 0x04),
2195 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2196 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2197 	DIAG_COUNTER(rq_num_lpe, 0x18),
2198 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2199 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2200 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2201 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2202 	DIAG_COUNTER(sq_num_bre, 0x34),
2203 	DIAG_COUNTER(sq_num_rire, 0x44),
2204 	DIAG_COUNTER(rq_num_rire, 0x48),
2205 	DIAG_COUNTER(sq_num_rae, 0x4C),
2206 	DIAG_COUNTER(rq_num_rae, 0x50),
2207 	DIAG_COUNTER(sq_num_roe, 0x54),
2208 	DIAG_COUNTER(sq_num_tree, 0x5C),
2209 	DIAG_COUNTER(sq_num_rree, 0x64),
2210 	DIAG_COUNTER(rq_num_rnr, 0x68),
2211 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2212 	DIAG_COUNTER(rq_num_oos, 0x100),
2213 	DIAG_COUNTER(sq_num_oos, 0x104),
2214 };
2215 
2216 static const struct diag_counter diag_ext[] = {
2217 	DIAG_COUNTER(rq_num_dup, 0x130),
2218 	DIAG_COUNTER(sq_num_to, 0x134),
2219 };
2220 
2221 static const struct diag_counter diag_device_only[] = {
2222 	DIAG_COUNTER(num_cqovf, 0x1A0),
2223 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2224 };
2225 
mlx4_ib_alloc_hw_stats(struct ib_device * ibdev,u8 port_num)2226 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2227 						    u8 port_num)
2228 {
2229 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2230 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2231 
2232 	if (!diag[!!port_num].name)
2233 		return NULL;
2234 
2235 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2236 					  diag[!!port_num].num_counters,
2237 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2238 }
2239 
mlx4_ib_get_hw_stats(struct ib_device * ibdev,struct rdma_hw_stats * stats,u8 port,int index)2240 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2241 				struct rdma_hw_stats *stats,
2242 				u8 port, int index)
2243 {
2244 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2245 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2246 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2247 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2248 	int ret;
2249 	int i;
2250 
2251 	ret = mlx4_query_diag_counters(dev->dev,
2252 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2253 				       diag[!!port].offset, hw_value,
2254 				       diag[!!port].num_counters, port);
2255 
2256 	if (ret)
2257 		return ret;
2258 
2259 	for (i = 0; i < diag[!!port].num_counters; i++)
2260 		stats->value[i] = hw_value[i];
2261 
2262 	return diag[!!port].num_counters;
2263 }
2264 
__mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev * ibdev,const char *** name,u32 ** offset,u32 * num,bool port)2265 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2266 					 const char ***name,
2267 					 u32 **offset,
2268 					 u32 *num,
2269 					 bool port)
2270 {
2271 	u32 num_counters;
2272 
2273 	num_counters = ARRAY_SIZE(diag_basic);
2274 
2275 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2276 		num_counters += ARRAY_SIZE(diag_ext);
2277 
2278 	if (!port)
2279 		num_counters += ARRAY_SIZE(diag_device_only);
2280 
2281 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2282 	if (!*name)
2283 		return -ENOMEM;
2284 
2285 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2286 	if (!*offset)
2287 		goto err_name;
2288 
2289 	*num = num_counters;
2290 
2291 	return 0;
2292 
2293 err_name:
2294 	kfree(*name);
2295 	return -ENOMEM;
2296 }
2297 
mlx4_ib_fill_diag_counters(struct mlx4_ib_dev * ibdev,const char ** name,u32 * offset,bool port)2298 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2299 				       const char **name,
2300 				       u32 *offset,
2301 				       bool port)
2302 {
2303 	int i;
2304 	int j;
2305 
2306 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2307 		name[i] = diag_basic[i].name;
2308 		offset[i] = diag_basic[i].offset;
2309 	}
2310 
2311 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2312 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2313 			name[j] = diag_ext[i].name;
2314 			offset[j] = diag_ext[i].offset;
2315 		}
2316 	}
2317 
2318 	if (!port) {
2319 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2320 			name[j] = diag_device_only[i].name;
2321 			offset[j] = diag_device_only[i].offset;
2322 		}
2323 	}
2324 }
2325 
mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev * ibdev)2326 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2327 {
2328 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2329 	int i;
2330 	int ret;
2331 	bool per_port = !!(ibdev->dev->caps.flags2 &
2332 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2333 
2334 	if (mlx4_is_slave(ibdev->dev))
2335 		return 0;
2336 
2337 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2338 		/* i == 1 means we are building port counters */
2339 		if (i && !per_port)
2340 			continue;
2341 
2342 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2343 						    &diag[i].offset,
2344 						    &diag[i].num_counters, i);
2345 		if (ret)
2346 			goto err_alloc;
2347 
2348 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2349 					   diag[i].offset, i);
2350 	}
2351 
2352 	ibdev->ib_dev.get_hw_stats	= mlx4_ib_get_hw_stats;
2353 	ibdev->ib_dev.alloc_hw_stats	= mlx4_ib_alloc_hw_stats;
2354 
2355 	return 0;
2356 
2357 err_alloc:
2358 	if (i) {
2359 		kfree(diag[i - 1].name);
2360 		kfree(diag[i - 1].offset);
2361 	}
2362 
2363 	return ret;
2364 }
2365 
mlx4_ib_diag_cleanup(struct mlx4_ib_dev * ibdev)2366 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2367 {
2368 	int i;
2369 
2370 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2371 		kfree(ibdev->diag_counters[i].offset);
2372 		kfree(ibdev->diag_counters[i].name);
2373 	}
2374 }
2375 
2376 #define MLX4_IB_INVALID_MAC	((u64)-1)
mlx4_ib_update_qps(struct mlx4_ib_dev * ibdev,struct net_device * dev,int port)2377 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2378 			       struct net_device *dev,
2379 			       int port)
2380 {
2381 	u64 new_smac = 0;
2382 	u64 release_mac = MLX4_IB_INVALID_MAC;
2383 	struct mlx4_ib_qp *qp;
2384 
2385 	read_lock(&dev_base_lock);
2386 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
2387 	read_unlock(&dev_base_lock);
2388 
2389 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2390 
2391 	/* no need for update QP1 and mac registration in non-SRIOV */
2392 	if (!mlx4_is_mfunc(ibdev->dev))
2393 		return;
2394 
2395 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2396 	qp = ibdev->qp1_proxy[port - 1];
2397 	if (qp) {
2398 		int new_smac_index;
2399 		u64 old_smac;
2400 		struct mlx4_update_qp_params update_params;
2401 
2402 		mutex_lock(&qp->mutex);
2403 		old_smac = qp->pri.smac;
2404 		if (new_smac == old_smac)
2405 			goto unlock;
2406 
2407 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2408 
2409 		if (new_smac_index < 0)
2410 			goto unlock;
2411 
2412 		update_params.smac_index = new_smac_index;
2413 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2414 				   &update_params)) {
2415 			release_mac = new_smac;
2416 			goto unlock;
2417 		}
2418 		/* if old port was zero, no mac was yet registered for this QP */
2419 		if (qp->pri.smac_port)
2420 			release_mac = old_smac;
2421 		qp->pri.smac = new_smac;
2422 		qp->pri.smac_port = port;
2423 		qp->pri.smac_index = new_smac_index;
2424 	}
2425 
2426 unlock:
2427 	if (release_mac != MLX4_IB_INVALID_MAC)
2428 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2429 	if (qp)
2430 		mutex_unlock(&qp->mutex);
2431 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2432 }
2433 
mlx4_ib_scan_netdevs(struct mlx4_ib_dev * ibdev,struct net_device * dev,unsigned long event)2434 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2435 				 struct net_device *dev,
2436 				 unsigned long event)
2437 
2438 {
2439 	struct mlx4_ib_iboe *iboe;
2440 	int update_qps_port = -1;
2441 	int port;
2442 
2443 	ASSERT_RTNL();
2444 
2445 	iboe = &ibdev->iboe;
2446 
2447 	spin_lock_bh(&iboe->lock);
2448 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2449 
2450 		iboe->netdevs[port - 1] =
2451 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2452 
2453 		if (dev == iboe->netdevs[port - 1] &&
2454 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2455 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2456 			update_qps_port = port;
2457 
2458 	}
2459 	spin_unlock_bh(&iboe->lock);
2460 
2461 	if (update_qps_port > 0)
2462 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2463 }
2464 
mlx4_ib_netdev_event(struct notifier_block * this,unsigned long event,void * ptr)2465 static int mlx4_ib_netdev_event(struct notifier_block *this,
2466 				unsigned long event, void *ptr)
2467 {
2468 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2469 	struct mlx4_ib_dev *ibdev;
2470 
2471 	if (!net_eq(dev_net(dev), &init_net))
2472 		return NOTIFY_DONE;
2473 
2474 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2475 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2476 
2477 	return NOTIFY_DONE;
2478 }
2479 
init_pkeys(struct mlx4_ib_dev * ibdev)2480 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2481 {
2482 	int port;
2483 	int slave;
2484 	int i;
2485 
2486 	if (mlx4_is_master(ibdev->dev)) {
2487 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2488 		     ++slave) {
2489 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2490 				for (i = 0;
2491 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2492 				     ++i) {
2493 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2494 					/* master has the identity virt2phys pkey mapping */
2495 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2496 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2497 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2498 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2499 				}
2500 			}
2501 		}
2502 		/* initialize pkey cache */
2503 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2504 			for (i = 0;
2505 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2506 			     ++i)
2507 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2508 					(i) ? 0 : 0xFFFF;
2509 		}
2510 	}
2511 }
2512 
mlx4_ib_alloc_eqs(struct mlx4_dev * dev,struct mlx4_ib_dev * ibdev)2513 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2514 {
2515 	int i, j, eq = 0, total_eqs = 0;
2516 
2517 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2518 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2519 	if (!ibdev->eq_table)
2520 		return;
2521 
2522 	for (i = 1; i <= dev->caps.num_ports; i++) {
2523 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2524 		     j++, total_eqs++) {
2525 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2526 				continue;
2527 			ibdev->eq_table[eq] = total_eqs;
2528 			if (!mlx4_assign_eq(dev, i,
2529 					    &ibdev->eq_table[eq]))
2530 				eq++;
2531 			else
2532 				ibdev->eq_table[eq] = -1;
2533 		}
2534 	}
2535 
2536 	for (i = eq; i < dev->caps.num_comp_vectors;
2537 	     ibdev->eq_table[i++] = -1)
2538 		;
2539 
2540 	/* Advertise the new number of EQs to clients */
2541 	ibdev->ib_dev.num_comp_vectors = eq;
2542 }
2543 
mlx4_ib_free_eqs(struct mlx4_dev * dev,struct mlx4_ib_dev * ibdev)2544 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2545 {
2546 	int i;
2547 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2548 
2549 	/* no eqs were allocated */
2550 	if (!ibdev->eq_table)
2551 		return;
2552 
2553 	/* Reset the advertised EQ number */
2554 	ibdev->ib_dev.num_comp_vectors = 0;
2555 
2556 	for (i = 0; i < total_eqs; i++)
2557 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2558 
2559 	kfree(ibdev->eq_table);
2560 	ibdev->eq_table = NULL;
2561 }
2562 
mlx4_port_immutable(struct ib_device * ibdev,u8 port_num,struct ib_port_immutable * immutable)2563 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2564 			       struct ib_port_immutable *immutable)
2565 {
2566 	struct ib_port_attr attr;
2567 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2568 	int err;
2569 
2570 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2571 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2572 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2573 	} else {
2574 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2575 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2576 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2577 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2578 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2579 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2580 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2581 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2582 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2583 	}
2584 
2585 	err = ib_query_port(ibdev, port_num, &attr);
2586 	if (err)
2587 		return err;
2588 
2589 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2590 	immutable->gid_tbl_len = attr.gid_tbl_len;
2591 
2592 	return 0;
2593 }
2594 
get_fw_ver_str(struct ib_device * device,char * str)2595 static void get_fw_ver_str(struct ib_device *device, char *str)
2596 {
2597 	struct mlx4_ib_dev *dev =
2598 		container_of(device, struct mlx4_ib_dev, ib_dev);
2599 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2600 		 (int) (dev->dev->caps.fw_ver >> 32),
2601 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2602 		 (int) dev->dev->caps.fw_ver & 0xffff);
2603 }
2604 
mlx4_ib_add(struct mlx4_dev * dev)2605 static void *mlx4_ib_add(struct mlx4_dev *dev)
2606 {
2607 	struct mlx4_ib_dev *ibdev;
2608 	int num_ports = 0;
2609 	int i, j;
2610 	int err;
2611 	struct mlx4_ib_iboe *iboe;
2612 	int ib_num_ports = 0;
2613 	int num_req_counters;
2614 	int allocated;
2615 	u32 counter_index;
2616 	struct counter_index *new_counter_index = NULL;
2617 
2618 	pr_info_once("%s", mlx4_ib_version);
2619 
2620 	num_ports = 0;
2621 	mlx4_foreach_ib_transport_port(i, dev)
2622 		num_ports++;
2623 
2624 	/* No point in registering a device with no ports... */
2625 	if (num_ports == 0)
2626 		return NULL;
2627 
2628 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2629 	if (!ibdev) {
2630 		dev_err(&dev->persist->pdev->dev,
2631 			"Device struct alloc failed\n");
2632 		return NULL;
2633 	}
2634 
2635 	iboe = &ibdev->iboe;
2636 
2637 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2638 		goto err_dealloc;
2639 
2640 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2641 		goto err_pd;
2642 
2643 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2644 				 PAGE_SIZE);
2645 	if (!ibdev->uar_map)
2646 		goto err_uar;
2647 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2648 
2649 	ibdev->dev = dev;
2650 	ibdev->bond_next_port	= 0;
2651 
2652 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2653 	ibdev->ib_dev.owner		= THIS_MODULE;
2654 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2655 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2656 	ibdev->num_ports		= num_ports;
2657 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2658 						1 : ibdev->num_ports;
2659 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2660 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2661 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2662 	ibdev->ib_dev.add_gid		= mlx4_ib_add_gid;
2663 	ibdev->ib_dev.del_gid		= mlx4_ib_del_gid;
2664 
2665 	if (dev->caps.userspace_caps)
2666 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2667 	else
2668 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2669 
2670 	ibdev->ib_dev.uverbs_cmd_mask	=
2671 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2672 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2673 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2674 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2675 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2676 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2677 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2678 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2679 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2680 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2681 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2682 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2683 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2684 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2685 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2686 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2687 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2688 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2689 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2690 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2691 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2692 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2693 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2694 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2695 
2696 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2697 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2698 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2699 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2700 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2701 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2702 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2703 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2704 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2705 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2706 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2707 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2708 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2709 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2710 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2711 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2712 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2713 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2714 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2715 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2716 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2717 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2718 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2719 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2720 	ibdev->ib_dev.drain_sq		= mlx4_ib_drain_sq;
2721 	ibdev->ib_dev.drain_rq		= mlx4_ib_drain_rq;
2722 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2723 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2724 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2725 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2726 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2727 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2728 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2729 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2730 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2731 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2732 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2733 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2734 	ibdev->ib_dev.alloc_mr		= mlx4_ib_alloc_mr;
2735 	ibdev->ib_dev.map_mr_sg		= mlx4_ib_map_mr_sg;
2736 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2737 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2738 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2739 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2740 	ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2741 	ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2742 
2743 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2744 		(1ull << IB_USER_VERBS_EX_CMD_MODIFY_CQ);
2745 
2746 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2747 	    ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2748 	    IB_LINK_LAYER_ETHERNET) ||
2749 	    (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2750 	    IB_LINK_LAYER_ETHERNET))) {
2751 		ibdev->ib_dev.create_wq		= mlx4_ib_create_wq;
2752 		ibdev->ib_dev.modify_wq		= mlx4_ib_modify_wq;
2753 		ibdev->ib_dev.destroy_wq	= mlx4_ib_destroy_wq;
2754 		ibdev->ib_dev.create_rwq_ind_table  =
2755 			mlx4_ib_create_rwq_ind_table;
2756 		ibdev->ib_dev.destroy_rwq_ind_table =
2757 			mlx4_ib_destroy_rwq_ind_table;
2758 		ibdev->ib_dev.uverbs_ex_cmd_mask |=
2759 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ)	  |
2760 			(1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ)	  |
2761 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ)	  |
2762 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) |
2763 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
2764 	}
2765 
2766 	if (!mlx4_is_slave(ibdev->dev)) {
2767 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2768 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2769 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2770 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2771 	}
2772 
2773 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2774 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2775 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2776 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2777 
2778 		ibdev->ib_dev.uverbs_cmd_mask |=
2779 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2780 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2781 	}
2782 
2783 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2784 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2785 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2786 		ibdev->ib_dev.uverbs_cmd_mask |=
2787 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2788 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2789 	}
2790 
2791 	if (check_flow_steering_support(dev)) {
2792 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2793 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2794 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2795 
2796 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2797 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2798 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2799 	}
2800 
2801 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2802 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2803 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2804 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2805 
2806 	mlx4_ib_alloc_eqs(dev, ibdev);
2807 
2808 	spin_lock_init(&iboe->lock);
2809 
2810 	if (init_node_data(ibdev))
2811 		goto err_map;
2812 	mlx4_init_sl2vl_tbl(ibdev);
2813 
2814 	for (i = 0; i < ibdev->num_ports; ++i) {
2815 		mutex_init(&ibdev->counters_table[i].mutex);
2816 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2817 	}
2818 
2819 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2820 	for (i = 0; i < num_req_counters; ++i) {
2821 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2822 		allocated = 0;
2823 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2824 						IB_LINK_LAYER_ETHERNET) {
2825 			err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2826 						 MLX4_RES_USAGE_DRIVER);
2827 			/* if failed to allocate a new counter, use default */
2828 			if (err)
2829 				counter_index =
2830 					mlx4_get_default_counter_index(dev,
2831 								       i + 1);
2832 			else
2833 				allocated = 1;
2834 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2835 			counter_index = mlx4_get_default_counter_index(dev,
2836 								       i + 1);
2837 		}
2838 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2839 					    GFP_KERNEL);
2840 		if (!new_counter_index) {
2841 			if (allocated)
2842 				mlx4_counter_free(ibdev->dev, counter_index);
2843 			goto err_counter;
2844 		}
2845 		new_counter_index->index = counter_index;
2846 		new_counter_index->allocated = allocated;
2847 		list_add_tail(&new_counter_index->list,
2848 			      &ibdev->counters_table[i].counters_list);
2849 		ibdev->counters_table[i].default_counter = counter_index;
2850 		pr_info("counter index %d for port %d allocated %d\n",
2851 			counter_index, i + 1, allocated);
2852 	}
2853 	if (mlx4_is_bonded(dev))
2854 		for (i = 1; i < ibdev->num_ports ; ++i) {
2855 			new_counter_index =
2856 					kmalloc(sizeof(struct counter_index),
2857 						GFP_KERNEL);
2858 			if (!new_counter_index)
2859 				goto err_counter;
2860 			new_counter_index->index = counter_index;
2861 			new_counter_index->allocated = 0;
2862 			list_add_tail(&new_counter_index->list,
2863 				      &ibdev->counters_table[i].counters_list);
2864 			ibdev->counters_table[i].default_counter =
2865 								counter_index;
2866 		}
2867 
2868 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2869 		ib_num_ports++;
2870 
2871 	spin_lock_init(&ibdev->sm_lock);
2872 	mutex_init(&ibdev->cap_mask_mutex);
2873 	INIT_LIST_HEAD(&ibdev->qp_list);
2874 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2875 
2876 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2877 	    ib_num_ports) {
2878 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2879 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2880 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2881 					    &ibdev->steer_qpn_base, 0,
2882 					    MLX4_RES_USAGE_DRIVER);
2883 		if (err)
2884 			goto err_counter;
2885 
2886 		ibdev->ib_uc_qpns_bitmap =
2887 			kmalloc_array(BITS_TO_LONGS(ibdev->steer_qpn_count),
2888 				      sizeof(long),
2889 				      GFP_KERNEL);
2890 		if (!ibdev->ib_uc_qpns_bitmap)
2891 			goto err_steer_qp_release;
2892 
2893 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2894 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2895 				    ibdev->steer_qpn_count);
2896 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2897 					dev, ibdev->steer_qpn_base,
2898 					ibdev->steer_qpn_base +
2899 					ibdev->steer_qpn_count - 1);
2900 			if (err)
2901 				goto err_steer_free_bitmap;
2902 		} else {
2903 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2904 				    ibdev->steer_qpn_count);
2905 		}
2906 	}
2907 
2908 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2909 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2910 
2911 	if (mlx4_ib_alloc_diag_counters(ibdev))
2912 		goto err_steer_free_bitmap;
2913 
2914 	ibdev->ib_dev.driver_id = RDMA_DRIVER_MLX4;
2915 	if (ib_register_device(&ibdev->ib_dev, NULL))
2916 		goto err_diag_counters;
2917 
2918 	if (mlx4_ib_mad_init(ibdev))
2919 		goto err_reg;
2920 
2921 	if (mlx4_ib_init_sriov(ibdev))
2922 		goto err_mad;
2923 
2924 	if (!iboe->nb.notifier_call) {
2925 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
2926 		err = register_netdevice_notifier(&iboe->nb);
2927 		if (err) {
2928 			iboe->nb.notifier_call = NULL;
2929 			goto err_notif;
2930 		}
2931 	}
2932 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2933 		err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2934 		if (err)
2935 			goto err_notif;
2936 	}
2937 
2938 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2939 		if (device_create_file(&ibdev->ib_dev.dev,
2940 				       mlx4_class_attributes[j]))
2941 			goto err_notif;
2942 	}
2943 
2944 	ibdev->ib_active = true;
2945 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2946 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2947 					 &ibdev->ib_dev);
2948 
2949 	if (mlx4_is_mfunc(ibdev->dev))
2950 		init_pkeys(ibdev);
2951 
2952 	/* create paravirt contexts for any VFs which are active */
2953 	if (mlx4_is_master(ibdev->dev)) {
2954 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2955 			if (j == mlx4_master_func_num(ibdev->dev))
2956 				continue;
2957 			if (mlx4_is_slave_active(ibdev->dev, j))
2958 				do_slave_init(ibdev, j, 1);
2959 		}
2960 	}
2961 	return ibdev;
2962 
2963 err_notif:
2964 	if (ibdev->iboe.nb.notifier_call) {
2965 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2966 			pr_warn("failure unregistering notifier\n");
2967 		ibdev->iboe.nb.notifier_call = NULL;
2968 	}
2969 	flush_workqueue(wq);
2970 
2971 	mlx4_ib_close_sriov(ibdev);
2972 
2973 err_mad:
2974 	mlx4_ib_mad_cleanup(ibdev);
2975 
2976 err_reg:
2977 	ib_unregister_device(&ibdev->ib_dev);
2978 
2979 err_diag_counters:
2980 	mlx4_ib_diag_cleanup(ibdev);
2981 
2982 err_steer_free_bitmap:
2983 	kfree(ibdev->ib_uc_qpns_bitmap);
2984 
2985 err_steer_qp_release:
2986 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2987 			      ibdev->steer_qpn_count);
2988 err_counter:
2989 	for (i = 0; i < ibdev->num_ports; ++i)
2990 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2991 
2992 err_map:
2993 	mlx4_ib_free_eqs(dev, ibdev);
2994 	iounmap(ibdev->uar_map);
2995 
2996 err_uar:
2997 	mlx4_uar_free(dev, &ibdev->priv_uar);
2998 
2999 err_pd:
3000 	mlx4_pd_free(dev, ibdev->priv_pdn);
3001 
3002 err_dealloc:
3003 	ib_dealloc_device(&ibdev->ib_dev);
3004 
3005 	return NULL;
3006 }
3007 
mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev * dev,int count,int * qpn)3008 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
3009 {
3010 	int offset;
3011 
3012 	WARN_ON(!dev->ib_uc_qpns_bitmap);
3013 
3014 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
3015 					 dev->steer_qpn_count,
3016 					 get_count_order(count));
3017 	if (offset < 0)
3018 		return offset;
3019 
3020 	*qpn = dev->steer_qpn_base + offset;
3021 	return 0;
3022 }
3023 
mlx4_ib_steer_qp_free(struct mlx4_ib_dev * dev,u32 qpn,int count)3024 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
3025 {
3026 	if (!qpn ||
3027 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
3028 		return;
3029 
3030 	if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
3031 		 qpn, dev->steer_qpn_base))
3032 		/* not supposed to be here */
3033 		return;
3034 
3035 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
3036 			      qpn - dev->steer_qpn_base,
3037 			      get_count_order(count));
3038 }
3039 
mlx4_ib_steer_qp_reg(struct mlx4_ib_dev * mdev,struct mlx4_ib_qp * mqp,int is_attach)3040 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
3041 			 int is_attach)
3042 {
3043 	int err;
3044 	size_t flow_size;
3045 	struct ib_flow_attr *flow = NULL;
3046 	struct ib_flow_spec_ib *ib_spec;
3047 
3048 	if (is_attach) {
3049 		flow_size = sizeof(struct ib_flow_attr) +
3050 			    sizeof(struct ib_flow_spec_ib);
3051 		flow = kzalloc(flow_size, GFP_KERNEL);
3052 		if (!flow)
3053 			return -ENOMEM;
3054 		flow->port = mqp->port;
3055 		flow->num_of_specs = 1;
3056 		flow->size = flow_size;
3057 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
3058 		ib_spec->type = IB_FLOW_SPEC_IB;
3059 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
3060 		/* Add an empty rule for IB L2 */
3061 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
3062 
3063 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
3064 					    IB_FLOW_DOMAIN_NIC,
3065 					    MLX4_FS_REGULAR,
3066 					    &mqp->reg_id);
3067 	} else {
3068 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
3069 	}
3070 	kfree(flow);
3071 	return err;
3072 }
3073 
mlx4_ib_remove(struct mlx4_dev * dev,void * ibdev_ptr)3074 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3075 {
3076 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
3077 	int p;
3078 	int i;
3079 
3080 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3081 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3082 	ibdev->ib_active = false;
3083 	flush_workqueue(wq);
3084 
3085 	if (ibdev->iboe.nb.notifier_call) {
3086 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3087 			pr_warn("failure unregistering notifier\n");
3088 		ibdev->iboe.nb.notifier_call = NULL;
3089 	}
3090 
3091 	mlx4_ib_close_sriov(ibdev);
3092 	mlx4_ib_mad_cleanup(ibdev);
3093 	ib_unregister_device(&ibdev->ib_dev);
3094 	mlx4_ib_diag_cleanup(ibdev);
3095 
3096 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3097 			      ibdev->steer_qpn_count);
3098 	kfree(ibdev->ib_uc_qpns_bitmap);
3099 
3100 	iounmap(ibdev->uar_map);
3101 	for (p = 0; p < ibdev->num_ports; ++p)
3102 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3103 
3104 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3105 		mlx4_CLOSE_PORT(dev, p);
3106 
3107 	mlx4_ib_free_eqs(dev, ibdev);
3108 
3109 	mlx4_uar_free(dev, &ibdev->priv_uar);
3110 	mlx4_pd_free(dev, ibdev->priv_pdn);
3111 	ib_dealloc_device(&ibdev->ib_dev);
3112 }
3113 
do_slave_init(struct mlx4_ib_dev * ibdev,int slave,int do_init)3114 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3115 {
3116 	struct mlx4_ib_demux_work **dm = NULL;
3117 	struct mlx4_dev *dev = ibdev->dev;
3118 	int i;
3119 	unsigned long flags;
3120 	struct mlx4_active_ports actv_ports;
3121 	unsigned int ports;
3122 	unsigned int first_port;
3123 
3124 	if (!mlx4_is_master(dev))
3125 		return;
3126 
3127 	actv_ports = mlx4_get_active_ports(dev, slave);
3128 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3129 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3130 
3131 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3132 	if (!dm)
3133 		return;
3134 
3135 	for (i = 0; i < ports; i++) {
3136 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3137 		if (!dm[i]) {
3138 			while (--i >= 0)
3139 				kfree(dm[i]);
3140 			goto out;
3141 		}
3142 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3143 		dm[i]->port = first_port + i + 1;
3144 		dm[i]->slave = slave;
3145 		dm[i]->do_init = do_init;
3146 		dm[i]->dev = ibdev;
3147 	}
3148 	/* initialize or tear down tunnel QPs for the slave */
3149 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3150 	if (!ibdev->sriov.is_going_down) {
3151 		for (i = 0; i < ports; i++)
3152 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3153 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3154 	} else {
3155 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3156 		for (i = 0; i < ports; i++)
3157 			kfree(dm[i]);
3158 	}
3159 out:
3160 	kfree(dm);
3161 	return;
3162 }
3163 
mlx4_ib_handle_catas_error(struct mlx4_ib_dev * ibdev)3164 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3165 {
3166 	struct mlx4_ib_qp *mqp;
3167 	unsigned long flags_qp;
3168 	unsigned long flags_cq;
3169 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3170 	struct list_head    cq_notify_list;
3171 	struct mlx4_cq *mcq;
3172 	unsigned long flags;
3173 
3174 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3175 	INIT_LIST_HEAD(&cq_notify_list);
3176 
3177 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3178 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3179 
3180 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3181 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3182 		if (mqp->sq.tail != mqp->sq.head) {
3183 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3184 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3185 			if (send_mcq->mcq.comp &&
3186 			    mqp->ibqp.send_cq->comp_handler) {
3187 				if (!send_mcq->mcq.reset_notify_added) {
3188 					send_mcq->mcq.reset_notify_added = 1;
3189 					list_add_tail(&send_mcq->mcq.reset_notify,
3190 						      &cq_notify_list);
3191 				}
3192 			}
3193 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3194 		}
3195 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3196 		/* Now, handle the QP's receive queue */
3197 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3198 		/* no handling is needed for SRQ */
3199 		if (!mqp->ibqp.srq) {
3200 			if (mqp->rq.tail != mqp->rq.head) {
3201 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3202 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3203 				if (recv_mcq->mcq.comp &&
3204 				    mqp->ibqp.recv_cq->comp_handler) {
3205 					if (!recv_mcq->mcq.reset_notify_added) {
3206 						recv_mcq->mcq.reset_notify_added = 1;
3207 						list_add_tail(&recv_mcq->mcq.reset_notify,
3208 							      &cq_notify_list);
3209 					}
3210 				}
3211 				spin_unlock_irqrestore(&recv_mcq->lock,
3212 						       flags_cq);
3213 			}
3214 		}
3215 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3216 	}
3217 
3218 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3219 		mcq->comp(mcq);
3220 	}
3221 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3222 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3223 }
3224 
handle_bonded_port_state_event(struct work_struct * work)3225 static void handle_bonded_port_state_event(struct work_struct *work)
3226 {
3227 	struct ib_event_work *ew =
3228 		container_of(work, struct ib_event_work, work);
3229 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3230 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3231 	int i;
3232 	struct ib_event ibev;
3233 
3234 	kfree(ew);
3235 	spin_lock_bh(&ibdev->iboe.lock);
3236 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3237 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3238 		enum ib_port_state curr_port_state;
3239 
3240 		if (!curr_netdev)
3241 			continue;
3242 
3243 		curr_port_state =
3244 			(netif_running(curr_netdev) &&
3245 			 netif_carrier_ok(curr_netdev)) ?
3246 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3247 
3248 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3249 			curr_port_state : IB_PORT_ACTIVE;
3250 	}
3251 	spin_unlock_bh(&ibdev->iboe.lock);
3252 
3253 	ibev.device = &ibdev->ib_dev;
3254 	ibev.element.port_num = 1;
3255 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3256 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3257 
3258 	ib_dispatch_event(&ibev);
3259 }
3260 
mlx4_ib_sl2vl_update(struct mlx4_ib_dev * mdev,int port)3261 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3262 {
3263 	u64 sl2vl;
3264 	int err;
3265 
3266 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3267 	if (err) {
3268 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3269 		       port, err);
3270 		sl2vl = 0;
3271 	}
3272 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3273 }
3274 
ib_sl2vl_update_work(struct work_struct * work)3275 static void ib_sl2vl_update_work(struct work_struct *work)
3276 {
3277 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3278 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3279 	int port = ew->port;
3280 
3281 	mlx4_ib_sl2vl_update(mdev, port);
3282 
3283 	kfree(ew);
3284 }
3285 
mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev * ibdev,int port)3286 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3287 				     int port)
3288 {
3289 	struct ib_event_work *ew;
3290 
3291 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3292 	if (ew) {
3293 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3294 		ew->port = port;
3295 		ew->ib_dev = ibdev;
3296 		queue_work(wq, &ew->work);
3297 	}
3298 }
3299 
mlx4_ib_event(struct mlx4_dev * dev,void * ibdev_ptr,enum mlx4_dev_event event,unsigned long param)3300 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3301 			  enum mlx4_dev_event event, unsigned long param)
3302 {
3303 	struct ib_event ibev;
3304 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3305 	struct mlx4_eqe *eqe = NULL;
3306 	struct ib_event_work *ew;
3307 	int p = 0;
3308 
3309 	if (mlx4_is_bonded(dev) &&
3310 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3311 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3312 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3313 		if (!ew)
3314 			return;
3315 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3316 		ew->ib_dev = ibdev;
3317 		queue_work(wq, &ew->work);
3318 		return;
3319 	}
3320 
3321 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3322 		eqe = (struct mlx4_eqe *)param;
3323 	else
3324 		p = (int) param;
3325 
3326 	switch (event) {
3327 	case MLX4_DEV_EVENT_PORT_UP:
3328 		if (p > ibdev->num_ports)
3329 			return;
3330 		if (!mlx4_is_slave(dev) &&
3331 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3332 			IB_LINK_LAYER_INFINIBAND) {
3333 			if (mlx4_is_master(dev))
3334 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3335 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3336 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3337 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3338 		}
3339 		ibev.event = IB_EVENT_PORT_ACTIVE;
3340 		break;
3341 
3342 	case MLX4_DEV_EVENT_PORT_DOWN:
3343 		if (p > ibdev->num_ports)
3344 			return;
3345 		ibev.event = IB_EVENT_PORT_ERR;
3346 		break;
3347 
3348 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3349 		ibdev->ib_active = false;
3350 		ibev.event = IB_EVENT_DEVICE_FATAL;
3351 		mlx4_ib_handle_catas_error(ibdev);
3352 		break;
3353 
3354 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3355 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3356 		if (!ew)
3357 			break;
3358 
3359 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3360 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3361 		ew->ib_dev = ibdev;
3362 		/* need to queue only for port owner, which uses GEN_EQE */
3363 		if (mlx4_is_master(dev))
3364 			queue_work(wq, &ew->work);
3365 		else
3366 			handle_port_mgmt_change_event(&ew->work);
3367 		return;
3368 
3369 	case MLX4_DEV_EVENT_SLAVE_INIT:
3370 		/* here, p is the slave id */
3371 		do_slave_init(ibdev, p, 1);
3372 		if (mlx4_is_master(dev)) {
3373 			int i;
3374 
3375 			for (i = 1; i <= ibdev->num_ports; i++) {
3376 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3377 					== IB_LINK_LAYER_INFINIBAND)
3378 					mlx4_ib_slave_alias_guid_event(ibdev,
3379 								       p, i,
3380 								       1);
3381 			}
3382 		}
3383 		return;
3384 
3385 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3386 		if (mlx4_is_master(dev)) {
3387 			int i;
3388 
3389 			for (i = 1; i <= ibdev->num_ports; i++) {
3390 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3391 					== IB_LINK_LAYER_INFINIBAND)
3392 					mlx4_ib_slave_alias_guid_event(ibdev,
3393 								       p, i,
3394 								       0);
3395 			}
3396 		}
3397 		/* here, p is the slave id */
3398 		do_slave_init(ibdev, p, 0);
3399 		return;
3400 
3401 	default:
3402 		return;
3403 	}
3404 
3405 	ibev.device	      = ibdev_ptr;
3406 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3407 
3408 	ib_dispatch_event(&ibev);
3409 }
3410 
3411 static struct mlx4_interface mlx4_ib_interface = {
3412 	.add		= mlx4_ib_add,
3413 	.remove		= mlx4_ib_remove,
3414 	.event		= mlx4_ib_event,
3415 	.protocol	= MLX4_PROT_IB_IPV6,
3416 	.flags		= MLX4_INTFF_BONDING
3417 };
3418 
mlx4_ib_init(void)3419 static int __init mlx4_ib_init(void)
3420 {
3421 	int err;
3422 
3423 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3424 	if (!wq)
3425 		return -ENOMEM;
3426 
3427 	err = mlx4_ib_mcg_init();
3428 	if (err)
3429 		goto clean_wq;
3430 
3431 	err = mlx4_register_interface(&mlx4_ib_interface);
3432 	if (err)
3433 		goto clean_mcg;
3434 
3435 	return 0;
3436 
3437 clean_mcg:
3438 	mlx4_ib_mcg_destroy();
3439 
3440 clean_wq:
3441 	destroy_workqueue(wq);
3442 	return err;
3443 }
3444 
mlx4_ib_cleanup(void)3445 static void __exit mlx4_ib_cleanup(void)
3446 {
3447 	mlx4_unregister_interface(&mlx4_ib_interface);
3448 	mlx4_ib_mcg_destroy();
3449 	destroy_workqueue(wq);
3450 }
3451 
3452 module_init(mlx4_ib_init);
3453 module_exit(mlx4_ib_cleanup);
3454