xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision 95e9fd10)
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 
43 #include <rdma/ib_smi.h>
44 #include <rdma/ib_user_verbs.h>
45 #include <rdma/ib_addr.h>
46 
47 #include <linux/mlx4/driver.h>
48 #include <linux/mlx4/cmd.h>
49 
50 #include "mlx4_ib.h"
51 #include "user.h"
52 
53 #define DRV_NAME	MLX4_IB_DRV_NAME
54 #define DRV_VERSION	"1.0"
55 #define DRV_RELDATE	"April 4, 2008"
56 
57 MODULE_AUTHOR("Roland Dreier");
58 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
59 MODULE_LICENSE("Dual BSD/GPL");
60 MODULE_VERSION(DRV_VERSION);
61 
62 static const char mlx4_ib_version[] =
63 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
64 	DRV_VERSION " (" DRV_RELDATE ")\n";
65 
66 struct update_gid_work {
67 	struct work_struct	work;
68 	union ib_gid		gids[128];
69 	struct mlx4_ib_dev     *dev;
70 	int			port;
71 };
72 
73 static struct workqueue_struct *wq;
74 
75 static void init_query_mad(struct ib_smp *mad)
76 {
77 	mad->base_version  = 1;
78 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
79 	mad->class_version = 1;
80 	mad->method	   = IB_MGMT_METHOD_GET;
81 }
82 
83 static union ib_gid zgid;
84 
85 static int mlx4_ib_query_device(struct ib_device *ibdev,
86 				struct ib_device_attr *props)
87 {
88 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
89 	struct ib_smp *in_mad  = NULL;
90 	struct ib_smp *out_mad = NULL;
91 	int err = -ENOMEM;
92 
93 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
94 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
95 	if (!in_mad || !out_mad)
96 		goto out;
97 
98 	init_query_mad(in_mad);
99 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
100 
101 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, 1, NULL, NULL, in_mad, out_mad);
102 	if (err)
103 		goto out;
104 
105 	memset(props, 0, sizeof *props);
106 
107 	props->fw_ver = dev->dev->caps.fw_ver;
108 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
109 		IB_DEVICE_PORT_ACTIVE_EVENT		|
110 		IB_DEVICE_SYS_IMAGE_GUID		|
111 		IB_DEVICE_RC_RNR_NAK_GEN		|
112 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
113 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
114 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
115 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
116 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
117 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
118 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
119 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
120 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
121 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
122 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
123 	if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
124 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
125 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
126 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
127 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
128 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
129 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
130 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
131 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
132 		props->device_cap_flags |= IB_DEVICE_XRC;
133 
134 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
135 		0xffffff;
136 	props->vendor_part_id	   = be16_to_cpup((__be16 *) (out_mad->data + 30));
137 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
138 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
139 
140 	props->max_mr_size	   = ~0ull;
141 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
142 	props->max_qp		   = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
143 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
144 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
145 					 dev->dev->caps.max_rq_sg);
146 	props->max_cq		   = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
147 	props->max_cqe		   = dev->dev->caps.max_cqes;
148 	props->max_mr		   = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
149 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
150 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
151 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
152 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
153 	props->max_srq		   = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
154 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
155 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
156 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
157 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
158 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
159 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
160 	props->masked_atomic_cap   = props->atomic_cap;
161 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
162 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
163 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
164 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
165 					   props->max_mcast_grp;
166 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
167 
168 out:
169 	kfree(in_mad);
170 	kfree(out_mad);
171 
172 	return err;
173 }
174 
175 static enum rdma_link_layer
176 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
177 {
178 	struct mlx4_dev *dev = to_mdev(device)->dev;
179 
180 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
181 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
182 }
183 
184 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
185 			      struct ib_port_attr *props)
186 {
187 	struct ib_smp *in_mad  = NULL;
188 	struct ib_smp *out_mad = NULL;
189 	int ext_active_speed;
190 	int err = -ENOMEM;
191 
192 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
193 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
194 	if (!in_mad || !out_mad)
195 		goto out;
196 
197 	init_query_mad(in_mad);
198 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
199 	in_mad->attr_mod = cpu_to_be32(port);
200 
201 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL,
202 				in_mad, out_mad);
203 	if (err)
204 		goto out;
205 
206 
207 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
208 	props->lmc		= out_mad->data[34] & 0x7;
209 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
210 	props->sm_sl		= out_mad->data[36] & 0xf;
211 	props->state		= out_mad->data[32] & 0xf;
212 	props->phys_state	= out_mad->data[33] >> 4;
213 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
214 	props->gid_tbl_len	= to_mdev(ibdev)->dev->caps.gid_table_len[port];
215 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
216 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
217 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
218 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
219 	props->active_width	= out_mad->data[31] & 0xf;
220 	props->active_speed	= out_mad->data[35] >> 4;
221 	props->max_mtu		= out_mad->data[41] & 0xf;
222 	props->active_mtu	= out_mad->data[36] >> 4;
223 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
224 	props->max_vl_num	= out_mad->data[37] >> 4;
225 	props->init_type_reply	= out_mad->data[41] >> 4;
226 
227 	/* Check if extended speeds (EDR/FDR/...) are supported */
228 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
229 		ext_active_speed = out_mad->data[62] >> 4;
230 
231 		switch (ext_active_speed) {
232 		case 1:
233 			props->active_speed = IB_SPEED_FDR;
234 			break;
235 		case 2:
236 			props->active_speed = IB_SPEED_EDR;
237 			break;
238 		}
239 	}
240 
241 	/* If reported active speed is QDR, check if is FDR-10 */
242 	if (props->active_speed == IB_SPEED_QDR) {
243 		init_query_mad(in_mad);
244 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
245 		in_mad->attr_mod = cpu_to_be32(port);
246 
247 		err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port,
248 				   NULL, NULL, in_mad, out_mad);
249 		if (err)
250 			goto out;
251 
252 		/* Checking LinkSpeedActive for FDR-10 */
253 		if (out_mad->data[15] & 0x1)
254 			props->active_speed = IB_SPEED_FDR10;
255 	}
256 
257 	/* Avoid wrong speed value returned by FW if the IB link is down. */
258 	if (props->state == IB_PORT_DOWN)
259 		 props->active_speed = IB_SPEED_SDR;
260 
261 out:
262 	kfree(in_mad);
263 	kfree(out_mad);
264 	return err;
265 }
266 
267 static u8 state_to_phys_state(enum ib_port_state state)
268 {
269 	return state == IB_PORT_ACTIVE ? 5 : 3;
270 }
271 
272 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
273 			       struct ib_port_attr *props)
274 {
275 
276 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
277 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
278 	struct net_device *ndev;
279 	enum ib_mtu tmp;
280 	struct mlx4_cmd_mailbox *mailbox;
281 	int err = 0;
282 
283 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
284 	if (IS_ERR(mailbox))
285 		return PTR_ERR(mailbox);
286 
287 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
288 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
289 			   MLX4_CMD_WRAPPED);
290 	if (err)
291 		goto out;
292 
293 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
294 						IB_WIDTH_4X : IB_WIDTH_1X;
295 	props->active_speed	= IB_SPEED_QDR;
296 	props->port_cap_flags	= IB_PORT_CM_SUP;
297 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
298 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
299 	props->pkey_tbl_len	= 1;
300 	props->max_mtu		= IB_MTU_4096;
301 	props->max_vl_num	= 2;
302 	props->state		= IB_PORT_DOWN;
303 	props->phys_state	= state_to_phys_state(props->state);
304 	props->active_mtu	= IB_MTU_256;
305 	spin_lock(&iboe->lock);
306 	ndev = iboe->netdevs[port - 1];
307 	if (!ndev)
308 		goto out_unlock;
309 
310 	tmp = iboe_get_mtu(ndev->mtu);
311 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
312 
313 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
314 					IB_PORT_ACTIVE : IB_PORT_DOWN;
315 	props->phys_state	= state_to_phys_state(props->state);
316 out_unlock:
317 	spin_unlock(&iboe->lock);
318 out:
319 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
320 	return err;
321 }
322 
323 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
324 			      struct ib_port_attr *props)
325 {
326 	int err;
327 
328 	memset(props, 0, sizeof *props);
329 
330 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
331 		ib_link_query_port(ibdev, port, props) :
332 				eth_link_query_port(ibdev, port, props);
333 
334 	return err;
335 }
336 
337 static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
338 			       union ib_gid *gid)
339 {
340 	struct ib_smp *in_mad  = NULL;
341 	struct ib_smp *out_mad = NULL;
342 	int err = -ENOMEM;
343 
344 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
345 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
346 	if (!in_mad || !out_mad)
347 		goto out;
348 
349 	init_query_mad(in_mad);
350 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
351 	in_mad->attr_mod = cpu_to_be32(port);
352 
353 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
354 	if (err)
355 		goto out;
356 
357 	memcpy(gid->raw, out_mad->data + 8, 8);
358 
359 	init_query_mad(in_mad);
360 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
361 	in_mad->attr_mod = cpu_to_be32(index / 8);
362 
363 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
364 	if (err)
365 		goto out;
366 
367 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
368 
369 out:
370 	kfree(in_mad);
371 	kfree(out_mad);
372 	return err;
373 }
374 
375 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
376 			  union ib_gid *gid)
377 {
378 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
379 
380 	*gid = dev->iboe.gid_table[port - 1][index];
381 
382 	return 0;
383 }
384 
385 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
386 			     union ib_gid *gid)
387 {
388 	if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
389 		return __mlx4_ib_query_gid(ibdev, port, index, gid);
390 	else
391 		return iboe_query_gid(ibdev, port, index, gid);
392 }
393 
394 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
395 			      u16 *pkey)
396 {
397 	struct ib_smp *in_mad  = NULL;
398 	struct ib_smp *out_mad = NULL;
399 	int err = -ENOMEM;
400 
401 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
402 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
403 	if (!in_mad || !out_mad)
404 		goto out;
405 
406 	init_query_mad(in_mad);
407 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
408 	in_mad->attr_mod = cpu_to_be32(index / 32);
409 
410 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
411 	if (err)
412 		goto out;
413 
414 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
415 
416 out:
417 	kfree(in_mad);
418 	kfree(out_mad);
419 	return err;
420 }
421 
422 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
423 				 struct ib_device_modify *props)
424 {
425 	struct mlx4_cmd_mailbox *mailbox;
426 	unsigned long flags;
427 
428 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
429 		return -EOPNOTSUPP;
430 
431 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
432 		return 0;
433 
434 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
435 	memcpy(ibdev->node_desc, props->node_desc, 64);
436 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
437 
438 	/*
439 	 * If possible, pass node desc to FW, so it can generate
440 	 * a 144 trap.  If cmd fails, just ignore.
441 	 */
442 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
443 	if (IS_ERR(mailbox))
444 		return 0;
445 
446 	memset(mailbox->buf, 0, 256);
447 	memcpy(mailbox->buf, props->node_desc, 64);
448 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
449 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
450 
451 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
452 
453 	return 0;
454 }
455 
456 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
457 			 u32 cap_mask)
458 {
459 	struct mlx4_cmd_mailbox *mailbox;
460 	int err;
461 	u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
462 
463 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
464 	if (IS_ERR(mailbox))
465 		return PTR_ERR(mailbox);
466 
467 	memset(mailbox->buf, 0, 256);
468 
469 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
470 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
471 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
472 	} else {
473 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
474 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
475 	}
476 
477 	err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
478 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
479 
480 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
481 	return err;
482 }
483 
484 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
485 			       struct ib_port_modify *props)
486 {
487 	struct ib_port_attr attr;
488 	u32 cap_mask;
489 	int err;
490 
491 	mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
492 
493 	err = mlx4_ib_query_port(ibdev, port, &attr);
494 	if (err)
495 		goto out;
496 
497 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
498 		~props->clr_port_cap_mask;
499 
500 	err = mlx4_SET_PORT(to_mdev(ibdev), port,
501 			    !!(mask & IB_PORT_RESET_QKEY_CNTR),
502 			    cap_mask);
503 
504 out:
505 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
506 	return err;
507 }
508 
509 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
510 						  struct ib_udata *udata)
511 {
512 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
513 	struct mlx4_ib_ucontext *context;
514 	struct mlx4_ib_alloc_ucontext_resp resp;
515 	int err;
516 
517 	if (!dev->ib_active)
518 		return ERR_PTR(-EAGAIN);
519 
520 	resp.qp_tab_size      = dev->dev->caps.num_qps;
521 	resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
522 	resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
523 
524 	context = kmalloc(sizeof *context, GFP_KERNEL);
525 	if (!context)
526 		return ERR_PTR(-ENOMEM);
527 
528 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
529 	if (err) {
530 		kfree(context);
531 		return ERR_PTR(err);
532 	}
533 
534 	INIT_LIST_HEAD(&context->db_page_list);
535 	mutex_init(&context->db_page_mutex);
536 
537 	err = ib_copy_to_udata(udata, &resp, sizeof resp);
538 	if (err) {
539 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
540 		kfree(context);
541 		return ERR_PTR(-EFAULT);
542 	}
543 
544 	return &context->ibucontext;
545 }
546 
547 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
548 {
549 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
550 
551 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
552 	kfree(context);
553 
554 	return 0;
555 }
556 
557 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
558 {
559 	struct mlx4_ib_dev *dev = to_mdev(context->device);
560 
561 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
562 		return -EINVAL;
563 
564 	if (vma->vm_pgoff == 0) {
565 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
566 
567 		if (io_remap_pfn_range(vma, vma->vm_start,
568 				       to_mucontext(context)->uar.pfn,
569 				       PAGE_SIZE, vma->vm_page_prot))
570 			return -EAGAIN;
571 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
572 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
573 
574 		if (io_remap_pfn_range(vma, vma->vm_start,
575 				       to_mucontext(context)->uar.pfn +
576 				       dev->dev->caps.num_uars,
577 				       PAGE_SIZE, vma->vm_page_prot))
578 			return -EAGAIN;
579 	} else
580 		return -EINVAL;
581 
582 	return 0;
583 }
584 
585 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
586 				      struct ib_ucontext *context,
587 				      struct ib_udata *udata)
588 {
589 	struct mlx4_ib_pd *pd;
590 	int err;
591 
592 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
593 	if (!pd)
594 		return ERR_PTR(-ENOMEM);
595 
596 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
597 	if (err) {
598 		kfree(pd);
599 		return ERR_PTR(err);
600 	}
601 
602 	if (context)
603 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
604 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
605 			kfree(pd);
606 			return ERR_PTR(-EFAULT);
607 		}
608 
609 	return &pd->ibpd;
610 }
611 
612 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
613 {
614 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
615 	kfree(pd);
616 
617 	return 0;
618 }
619 
620 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
621 					  struct ib_ucontext *context,
622 					  struct ib_udata *udata)
623 {
624 	struct mlx4_ib_xrcd *xrcd;
625 	int err;
626 
627 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
628 		return ERR_PTR(-ENOSYS);
629 
630 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
631 	if (!xrcd)
632 		return ERR_PTR(-ENOMEM);
633 
634 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
635 	if (err)
636 		goto err1;
637 
638 	xrcd->pd = ib_alloc_pd(ibdev);
639 	if (IS_ERR(xrcd->pd)) {
640 		err = PTR_ERR(xrcd->pd);
641 		goto err2;
642 	}
643 
644 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
645 	if (IS_ERR(xrcd->cq)) {
646 		err = PTR_ERR(xrcd->cq);
647 		goto err3;
648 	}
649 
650 	return &xrcd->ibxrcd;
651 
652 err3:
653 	ib_dealloc_pd(xrcd->pd);
654 err2:
655 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
656 err1:
657 	kfree(xrcd);
658 	return ERR_PTR(err);
659 }
660 
661 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
662 {
663 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
664 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
665 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
666 	kfree(xrcd);
667 
668 	return 0;
669 }
670 
671 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
672 {
673 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
674 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
675 	struct mlx4_ib_gid_entry *ge;
676 
677 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
678 	if (!ge)
679 		return -ENOMEM;
680 
681 	ge->gid = *gid;
682 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
683 		ge->port = mqp->port;
684 		ge->added = 1;
685 	}
686 
687 	mutex_lock(&mqp->mutex);
688 	list_add_tail(&ge->list, &mqp->gid_list);
689 	mutex_unlock(&mqp->mutex);
690 
691 	return 0;
692 }
693 
694 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
695 		   union ib_gid *gid)
696 {
697 	u8 mac[6];
698 	struct net_device *ndev;
699 	int ret = 0;
700 
701 	if (!mqp->port)
702 		return 0;
703 
704 	spin_lock(&mdev->iboe.lock);
705 	ndev = mdev->iboe.netdevs[mqp->port - 1];
706 	if (ndev)
707 		dev_hold(ndev);
708 	spin_unlock(&mdev->iboe.lock);
709 
710 	if (ndev) {
711 		rdma_get_mcast_mac((struct in6_addr *)gid, mac);
712 		rtnl_lock();
713 		dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac);
714 		ret = 1;
715 		rtnl_unlock();
716 		dev_put(ndev);
717 	}
718 
719 	return ret;
720 }
721 
722 struct mlx4_ib_steering {
723 	struct list_head list;
724 	u64 reg_id;
725 	union ib_gid gid;
726 };
727 
728 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
729 {
730 	int err;
731 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
732 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
733 	u64 reg_id;
734 	struct mlx4_ib_steering *ib_steering = NULL;
735 
736 	if (mdev->dev->caps.steering_mode ==
737 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
738 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
739 		if (!ib_steering)
740 			return -ENOMEM;
741 	}
742 
743 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
744 				    !!(mqp->flags &
745 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
746 				    MLX4_PROT_IB_IPV6, &reg_id);
747 	if (err)
748 		goto err_malloc;
749 
750 	err = add_gid_entry(ibqp, gid);
751 	if (err)
752 		goto err_add;
753 
754 	if (ib_steering) {
755 		memcpy(ib_steering->gid.raw, gid->raw, 16);
756 		ib_steering->reg_id = reg_id;
757 		mutex_lock(&mqp->mutex);
758 		list_add(&ib_steering->list, &mqp->steering_rules);
759 		mutex_unlock(&mqp->mutex);
760 	}
761 	return 0;
762 
763 err_add:
764 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
765 			      MLX4_PROT_IB_IPV6, reg_id);
766 err_malloc:
767 	kfree(ib_steering);
768 
769 	return err;
770 }
771 
772 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
773 {
774 	struct mlx4_ib_gid_entry *ge;
775 	struct mlx4_ib_gid_entry *tmp;
776 	struct mlx4_ib_gid_entry *ret = NULL;
777 
778 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
779 		if (!memcmp(raw, ge->gid.raw, 16)) {
780 			ret = ge;
781 			break;
782 		}
783 	}
784 
785 	return ret;
786 }
787 
788 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
789 {
790 	int err;
791 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
792 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
793 	u8 mac[6];
794 	struct net_device *ndev;
795 	struct mlx4_ib_gid_entry *ge;
796 	u64 reg_id = 0;
797 
798 	if (mdev->dev->caps.steering_mode ==
799 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
800 		struct mlx4_ib_steering *ib_steering;
801 
802 		mutex_lock(&mqp->mutex);
803 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
804 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
805 				list_del(&ib_steering->list);
806 				break;
807 			}
808 		}
809 		mutex_unlock(&mqp->mutex);
810 		if (&ib_steering->list == &mqp->steering_rules) {
811 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
812 			return -EINVAL;
813 		}
814 		reg_id = ib_steering->reg_id;
815 		kfree(ib_steering);
816 	}
817 
818 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
819 				    MLX4_PROT_IB_IPV6, reg_id);
820 	if (err)
821 		return err;
822 
823 	mutex_lock(&mqp->mutex);
824 	ge = find_gid_entry(mqp, gid->raw);
825 	if (ge) {
826 		spin_lock(&mdev->iboe.lock);
827 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
828 		if (ndev)
829 			dev_hold(ndev);
830 		spin_unlock(&mdev->iboe.lock);
831 		rdma_get_mcast_mac((struct in6_addr *)gid, mac);
832 		if (ndev) {
833 			rtnl_lock();
834 			dev_mc_del(mdev->iboe.netdevs[ge->port - 1], mac);
835 			rtnl_unlock();
836 			dev_put(ndev);
837 		}
838 		list_del(&ge->list);
839 		kfree(ge);
840 	} else
841 		pr_warn("could not find mgid entry\n");
842 
843 	mutex_unlock(&mqp->mutex);
844 
845 	return 0;
846 }
847 
848 static int init_node_data(struct mlx4_ib_dev *dev)
849 {
850 	struct ib_smp *in_mad  = NULL;
851 	struct ib_smp *out_mad = NULL;
852 	int err = -ENOMEM;
853 
854 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
855 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
856 	if (!in_mad || !out_mad)
857 		goto out;
858 
859 	init_query_mad(in_mad);
860 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
861 
862 	err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
863 	if (err)
864 		goto out;
865 
866 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
867 
868 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
869 
870 	err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
871 	if (err)
872 		goto out;
873 
874 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
875 
876 out:
877 	kfree(in_mad);
878 	kfree(out_mad);
879 	return err;
880 }
881 
882 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
883 			char *buf)
884 {
885 	struct mlx4_ib_dev *dev =
886 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
887 	return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
888 }
889 
890 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
891 			   char *buf)
892 {
893 	struct mlx4_ib_dev *dev =
894 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
895 	return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
896 		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
897 		       (int) dev->dev->caps.fw_ver & 0xffff);
898 }
899 
900 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
901 			char *buf)
902 {
903 	struct mlx4_ib_dev *dev =
904 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
905 	return sprintf(buf, "%x\n", dev->dev->rev_id);
906 }
907 
908 static ssize_t show_board(struct device *device, struct device_attribute *attr,
909 			  char *buf)
910 {
911 	struct mlx4_ib_dev *dev =
912 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
913 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
914 		       dev->dev->board_id);
915 }
916 
917 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
918 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
919 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
920 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
921 
922 static struct device_attribute *mlx4_class_attributes[] = {
923 	&dev_attr_hw_rev,
924 	&dev_attr_fw_ver,
925 	&dev_attr_hca_type,
926 	&dev_attr_board_id
927 };
928 
929 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
930 {
931 	memcpy(eui, dev->dev_addr, 3);
932 	memcpy(eui + 5, dev->dev_addr + 3, 3);
933 	if (vlan_id < 0x1000) {
934 		eui[3] = vlan_id >> 8;
935 		eui[4] = vlan_id & 0xff;
936 	} else {
937 		eui[3] = 0xff;
938 		eui[4] = 0xfe;
939 	}
940 	eui[0] ^= 2;
941 }
942 
943 static void update_gids_task(struct work_struct *work)
944 {
945 	struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
946 	struct mlx4_cmd_mailbox *mailbox;
947 	union ib_gid *gids;
948 	int err;
949 	struct mlx4_dev	*dev = gw->dev->dev;
950 
951 	mailbox = mlx4_alloc_cmd_mailbox(dev);
952 	if (IS_ERR(mailbox)) {
953 		pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
954 		return;
955 	}
956 
957 	gids = mailbox->buf;
958 	memcpy(gids, gw->gids, sizeof gw->gids);
959 
960 	err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
961 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
962 		       MLX4_CMD_NATIVE);
963 	if (err)
964 		pr_warn("set port command failed\n");
965 	else {
966 		memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
967 		mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
968 	}
969 
970 	mlx4_free_cmd_mailbox(dev, mailbox);
971 	kfree(gw);
972 }
973 
974 static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
975 {
976 	struct net_device *ndev = dev->iboe.netdevs[port - 1];
977 	struct update_gid_work *work;
978 	struct net_device *tmp;
979 	int i;
980 	u8 *hits;
981 	int ret;
982 	union ib_gid gid;
983 	int free;
984 	int found;
985 	int need_update = 0;
986 	u16 vid;
987 
988 	work = kzalloc(sizeof *work, GFP_ATOMIC);
989 	if (!work)
990 		return -ENOMEM;
991 
992 	hits = kzalloc(128, GFP_ATOMIC);
993 	if (!hits) {
994 		ret = -ENOMEM;
995 		goto out;
996 	}
997 
998 	rcu_read_lock();
999 	for_each_netdev_rcu(&init_net, tmp) {
1000 		if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
1001 			gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1002 			vid = rdma_vlan_dev_vlan_id(tmp);
1003 			mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
1004 			found = 0;
1005 			free = -1;
1006 			for (i = 0; i < 128; ++i) {
1007 				if (free < 0 &&
1008 				    !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1009 					free = i;
1010 				if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
1011 					hits[i] = 1;
1012 					found = 1;
1013 					break;
1014 				}
1015 			}
1016 
1017 			if (!found) {
1018 				if (tmp == ndev &&
1019 				    (memcmp(&dev->iboe.gid_table[port - 1][0],
1020 					    &gid, sizeof gid) ||
1021 				     !memcmp(&dev->iboe.gid_table[port - 1][0],
1022 					     &zgid, sizeof gid))) {
1023 					dev->iboe.gid_table[port - 1][0] = gid;
1024 					++need_update;
1025 					hits[0] = 1;
1026 				} else if (free >= 0) {
1027 					dev->iboe.gid_table[port - 1][free] = gid;
1028 					hits[free] = 1;
1029 					++need_update;
1030 				}
1031 			}
1032 		}
1033 	}
1034 	rcu_read_unlock();
1035 
1036 	for (i = 0; i < 128; ++i)
1037 		if (!hits[i]) {
1038 			if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1039 				++need_update;
1040 			dev->iboe.gid_table[port - 1][i] = zgid;
1041 		}
1042 
1043 	if (need_update) {
1044 		memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
1045 		INIT_WORK(&work->work, update_gids_task);
1046 		work->port = port;
1047 		work->dev = dev;
1048 		queue_work(wq, &work->work);
1049 	} else
1050 		kfree(work);
1051 
1052 	kfree(hits);
1053 	return 0;
1054 
1055 out:
1056 	kfree(work);
1057 	return ret;
1058 }
1059 
1060 static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
1061 {
1062 	switch (event) {
1063 	case NETDEV_UP:
1064 	case NETDEV_CHANGEADDR:
1065 		update_ipv6_gids(dev, port, 0);
1066 		break;
1067 
1068 	case NETDEV_DOWN:
1069 		update_ipv6_gids(dev, port, 1);
1070 		dev->iboe.netdevs[port - 1] = NULL;
1071 	}
1072 }
1073 
1074 static void netdev_added(struct mlx4_ib_dev *dev, int port)
1075 {
1076 	update_ipv6_gids(dev, port, 0);
1077 }
1078 
1079 static void netdev_removed(struct mlx4_ib_dev *dev, int port)
1080 {
1081 	update_ipv6_gids(dev, port, 1);
1082 }
1083 
1084 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1085 				void *ptr)
1086 {
1087 	struct net_device *dev = ptr;
1088 	struct mlx4_ib_dev *ibdev;
1089 	struct net_device *oldnd;
1090 	struct mlx4_ib_iboe *iboe;
1091 	int port;
1092 
1093 	if (!net_eq(dev_net(dev), &init_net))
1094 		return NOTIFY_DONE;
1095 
1096 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1097 	iboe = &ibdev->iboe;
1098 
1099 	spin_lock(&iboe->lock);
1100 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1101 		oldnd = iboe->netdevs[port - 1];
1102 		iboe->netdevs[port - 1] =
1103 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1104 		if (oldnd != iboe->netdevs[port - 1]) {
1105 			if (iboe->netdevs[port - 1])
1106 				netdev_added(ibdev, port);
1107 			else
1108 				netdev_removed(ibdev, port);
1109 		}
1110 	}
1111 
1112 	if (dev == iboe->netdevs[0] ||
1113 	    (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
1114 		handle_en_event(ibdev, 1, event);
1115 	else if (dev == iboe->netdevs[1]
1116 		 || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
1117 		handle_en_event(ibdev, 2, event);
1118 
1119 	spin_unlock(&iboe->lock);
1120 
1121 	return NOTIFY_DONE;
1122 }
1123 
1124 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1125 {
1126 	char name[32];
1127 	int eq_per_port = 0;
1128 	int added_eqs = 0;
1129 	int total_eqs = 0;
1130 	int i, j, eq;
1131 
1132 	/* Legacy mode or comp_pool is not large enough */
1133 	if (dev->caps.comp_pool == 0 ||
1134 	    dev->caps.num_ports > dev->caps.comp_pool)
1135 		return;
1136 
1137 	eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1138 					dev->caps.num_ports);
1139 
1140 	/* Init eq table */
1141 	added_eqs = 0;
1142 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1143 		added_eqs += eq_per_port;
1144 
1145 	total_eqs = dev->caps.num_comp_vectors + added_eqs;
1146 
1147 	ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1148 	if (!ibdev->eq_table)
1149 		return;
1150 
1151 	ibdev->eq_added = added_eqs;
1152 
1153 	eq = 0;
1154 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1155 		for (j = 0; j < eq_per_port; j++) {
1156 			sprintf(name, "mlx4-ib-%d-%d@%s",
1157 				i, j, dev->pdev->bus->name);
1158 			/* Set IRQ for specific name (per ring) */
1159 			if (mlx4_assign_eq(dev, name, NULL,
1160 					   &ibdev->eq_table[eq])) {
1161 				/* Use legacy (same as mlx4_en driver) */
1162 				pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1163 				ibdev->eq_table[eq] =
1164 					(eq % dev->caps.num_comp_vectors);
1165 			}
1166 			eq++;
1167 		}
1168 	}
1169 
1170 	/* Fill the reset of the vector with legacy EQ */
1171 	for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1172 		ibdev->eq_table[eq++] = i;
1173 
1174 	/* Advertise the new number of EQs to clients */
1175 	ibdev->ib_dev.num_comp_vectors = total_eqs;
1176 }
1177 
1178 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1179 {
1180 	int i;
1181 
1182 	/* no additional eqs were added */
1183 	if (!ibdev->eq_table)
1184 		return;
1185 
1186 	/* Reset the advertised EQ number */
1187 	ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1188 
1189 	/* Free only the added eqs */
1190 	for (i = 0; i < ibdev->eq_added; i++) {
1191 		/* Don't free legacy eqs if used */
1192 		if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1193 			continue;
1194 		mlx4_release_eq(dev, ibdev->eq_table[i]);
1195 	}
1196 
1197 	kfree(ibdev->eq_table);
1198 }
1199 
1200 static void *mlx4_ib_add(struct mlx4_dev *dev)
1201 {
1202 	struct mlx4_ib_dev *ibdev;
1203 	int num_ports = 0;
1204 	int i, j;
1205 	int err;
1206 	struct mlx4_ib_iboe *iboe;
1207 
1208 	pr_info_once("%s", mlx4_ib_version);
1209 
1210 	if (mlx4_is_mfunc(dev)) {
1211 		pr_warn("IB not yet supported in SRIOV\n");
1212 		return NULL;
1213 	}
1214 
1215 	mlx4_foreach_ib_transport_port(i, dev)
1216 		num_ports++;
1217 
1218 	/* No point in registering a device with no ports... */
1219 	if (num_ports == 0)
1220 		return NULL;
1221 
1222 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1223 	if (!ibdev) {
1224 		dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1225 		return NULL;
1226 	}
1227 
1228 	iboe = &ibdev->iboe;
1229 
1230 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1231 		goto err_dealloc;
1232 
1233 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1234 		goto err_pd;
1235 
1236 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1237 				 PAGE_SIZE);
1238 	if (!ibdev->uar_map)
1239 		goto err_uar;
1240 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1241 
1242 	ibdev->dev = dev;
1243 
1244 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1245 	ibdev->ib_dev.owner		= THIS_MODULE;
1246 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
1247 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
1248 	ibdev->num_ports		= num_ports;
1249 	ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
1250 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
1251 	ibdev->ib_dev.dma_device	= &dev->pdev->dev;
1252 
1253 	ibdev->ib_dev.uverbs_abi_ver	= MLX4_IB_UVERBS_ABI_VERSION;
1254 	ibdev->ib_dev.uverbs_cmd_mask	=
1255 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
1256 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
1257 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
1258 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
1259 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
1260 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
1261 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
1262 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
1263 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
1264 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
1265 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
1266 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
1267 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
1268 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
1269 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
1270 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
1271 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
1272 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
1273 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
1274 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
1275 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
1276 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
1277 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
1278 
1279 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
1280 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
1281 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
1282 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
1283 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
1284 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
1285 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
1286 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
1287 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
1288 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
1289 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
1290 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
1291 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
1292 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
1293 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
1294 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
1295 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
1296 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
1297 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
1298 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
1299 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
1300 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
1301 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
1302 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
1303 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
1304 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
1305 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
1306 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
1307 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
1308 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
1309 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
1310 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
1311 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
1312 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
1313 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
1314 	ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1315 	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1316 	ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
1317 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
1318 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
1319 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
1320 
1321 	ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
1322 	ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
1323 	ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
1324 	ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
1325 
1326 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1327 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1328 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1329 		ibdev->ib_dev.uverbs_cmd_mask |=
1330 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1331 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1332 	}
1333 
1334 	mlx4_ib_alloc_eqs(dev, ibdev);
1335 
1336 	spin_lock_init(&iboe->lock);
1337 
1338 	if (init_node_data(ibdev))
1339 		goto err_map;
1340 
1341 	for (i = 0; i < ibdev->num_ports; ++i) {
1342 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1343 						IB_LINK_LAYER_ETHERNET) {
1344 			err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1345 			if (err)
1346 				ibdev->counters[i] = -1;
1347 		} else
1348 				ibdev->counters[i] = -1;
1349 	}
1350 
1351 	spin_lock_init(&ibdev->sm_lock);
1352 	mutex_init(&ibdev->cap_mask_mutex);
1353 
1354 	if (ib_register_device(&ibdev->ib_dev, NULL))
1355 		goto err_counter;
1356 
1357 	if (mlx4_ib_mad_init(ibdev))
1358 		goto err_reg;
1359 
1360 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1361 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
1362 		err = register_netdevice_notifier(&iboe->nb);
1363 		if (err)
1364 			goto err_reg;
1365 	}
1366 
1367 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
1368 		if (device_create_file(&ibdev->ib_dev.dev,
1369 				       mlx4_class_attributes[j]))
1370 			goto err_notif;
1371 	}
1372 
1373 	ibdev->ib_active = true;
1374 
1375 	return ibdev;
1376 
1377 err_notif:
1378 	if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1379 		pr_warn("failure unregistering notifier\n");
1380 	flush_workqueue(wq);
1381 
1382 err_reg:
1383 	ib_unregister_device(&ibdev->ib_dev);
1384 
1385 err_counter:
1386 	for (; i; --i)
1387 		if (ibdev->counters[i - 1] != -1)
1388 			mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
1389 
1390 err_map:
1391 	iounmap(ibdev->uar_map);
1392 
1393 err_uar:
1394 	mlx4_uar_free(dev, &ibdev->priv_uar);
1395 
1396 err_pd:
1397 	mlx4_pd_free(dev, ibdev->priv_pdn);
1398 
1399 err_dealloc:
1400 	ib_dealloc_device(&ibdev->ib_dev);
1401 
1402 	return NULL;
1403 }
1404 
1405 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1406 {
1407 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
1408 	int p;
1409 
1410 	mlx4_ib_mad_cleanup(ibdev);
1411 	ib_unregister_device(&ibdev->ib_dev);
1412 	if (ibdev->iboe.nb.notifier_call) {
1413 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1414 			pr_warn("failure unregistering notifier\n");
1415 		ibdev->iboe.nb.notifier_call = NULL;
1416 	}
1417 	iounmap(ibdev->uar_map);
1418 	for (p = 0; p < ibdev->num_ports; ++p)
1419 		if (ibdev->counters[p] != -1)
1420 			mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
1421 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
1422 		mlx4_CLOSE_PORT(dev, p);
1423 
1424 	mlx4_ib_free_eqs(dev, ibdev);
1425 
1426 	mlx4_uar_free(dev, &ibdev->priv_uar);
1427 	mlx4_pd_free(dev, ibdev->priv_pdn);
1428 	ib_dealloc_device(&ibdev->ib_dev);
1429 }
1430 
1431 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
1432 			  enum mlx4_dev_event event, unsigned long param)
1433 {
1434 	struct ib_event ibev;
1435 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
1436 	struct mlx4_eqe *eqe = NULL;
1437 	struct ib_event_work *ew;
1438 	int port = 0;
1439 
1440 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
1441 		eqe = (struct mlx4_eqe *)param;
1442 	else
1443 		port = (u8)param;
1444 
1445 	if (port > ibdev->num_ports)
1446 		return;
1447 
1448 	switch (event) {
1449 	case MLX4_DEV_EVENT_PORT_UP:
1450 		ibev.event = IB_EVENT_PORT_ACTIVE;
1451 		break;
1452 
1453 	case MLX4_DEV_EVENT_PORT_DOWN:
1454 		ibev.event = IB_EVENT_PORT_ERR;
1455 		break;
1456 
1457 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
1458 		ibdev->ib_active = false;
1459 		ibev.event = IB_EVENT_DEVICE_FATAL;
1460 		break;
1461 
1462 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
1463 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
1464 		if (!ew) {
1465 			pr_err("failed to allocate memory for events work\n");
1466 			break;
1467 		}
1468 
1469 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
1470 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
1471 		ew->ib_dev = ibdev;
1472 		handle_port_mgmt_change_event(&ew->work);
1473 		return;
1474 
1475 	default:
1476 		return;
1477 	}
1478 
1479 	ibev.device	      = ibdev_ptr;
1480 	ibev.element.port_num = port;
1481 
1482 	ib_dispatch_event(&ibev);
1483 }
1484 
1485 static struct mlx4_interface mlx4_ib_interface = {
1486 	.add		= mlx4_ib_add,
1487 	.remove		= mlx4_ib_remove,
1488 	.event		= mlx4_ib_event,
1489 	.protocol	= MLX4_PROT_IB_IPV6
1490 };
1491 
1492 static int __init mlx4_ib_init(void)
1493 {
1494 	int err;
1495 
1496 	wq = create_singlethread_workqueue("mlx4_ib");
1497 	if (!wq)
1498 		return -ENOMEM;
1499 
1500 	err = mlx4_register_interface(&mlx4_ib_interface);
1501 	if (err) {
1502 		destroy_workqueue(wq);
1503 		return err;
1504 	}
1505 
1506 	return 0;
1507 }
1508 
1509 static void __exit mlx4_ib_cleanup(void)
1510 {
1511 	mlx4_unregister_interface(&mlx4_ib_interface);
1512 	destroy_workqueue(wq);
1513 }
1514 
1515 module_init(mlx4_ib_init);
1516 module_exit(mlx4_ib_cleanup);
1517