xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision b34081f1)
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 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
58 
59 MODULE_AUTHOR("Roland Dreier");
60 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
61 MODULE_LICENSE("Dual BSD/GPL");
62 MODULE_VERSION(DRV_VERSION);
63 
64 int mlx4_ib_sm_guid_assign = 1;
65 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
66 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
67 
68 static const char mlx4_ib_version[] =
69 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
70 	DRV_VERSION " (" DRV_RELDATE ")\n";
71 
72 struct update_gid_work {
73 	struct work_struct	work;
74 	union ib_gid		gids[128];
75 	struct mlx4_ib_dev     *dev;
76 	int			port;
77 };
78 
79 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
80 
81 static struct workqueue_struct *wq;
82 
83 static void init_query_mad(struct ib_smp *mad)
84 {
85 	mad->base_version  = 1;
86 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
87 	mad->class_version = 1;
88 	mad->method	   = IB_MGMT_METHOD_GET;
89 }
90 
91 static union ib_gid zgid;
92 
93 static int check_flow_steering_support(struct mlx4_dev *dev)
94 {
95 	int ib_num_ports = 0;
96 	int i;
97 
98 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
99 		ib_num_ports++;
100 
101 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
102 		if (ib_num_ports || mlx4_is_mfunc(dev)) {
103 			pr_warn("Device managed flow steering is unavailable "
104 				"for IB ports or in multifunction env.\n");
105 			return 0;
106 		}
107 		return 1;
108 	}
109 	return 0;
110 }
111 
112 static int mlx4_ib_query_device(struct ib_device *ibdev,
113 				struct ib_device_attr *props)
114 {
115 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
116 	struct ib_smp *in_mad  = NULL;
117 	struct ib_smp *out_mad = NULL;
118 	int err = -ENOMEM;
119 
120 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
121 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
122 	if (!in_mad || !out_mad)
123 		goto out;
124 
125 	init_query_mad(in_mad);
126 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
127 
128 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
129 			   1, NULL, NULL, in_mad, out_mad);
130 	if (err)
131 		goto out;
132 
133 	memset(props, 0, sizeof *props);
134 
135 	props->fw_ver = dev->dev->caps.fw_ver;
136 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
137 		IB_DEVICE_PORT_ACTIVE_EVENT		|
138 		IB_DEVICE_SYS_IMAGE_GUID		|
139 		IB_DEVICE_RC_RNR_NAK_GEN		|
140 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
141 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
142 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
143 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
144 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
145 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
146 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
147 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
148 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
149 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
150 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
151 	if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
152 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
153 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
154 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
155 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
156 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
157 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
158 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
159 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
160 		props->device_cap_flags |= IB_DEVICE_XRC;
161 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
162 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
163 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
164 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
165 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
166 		else
167 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
168 	if (check_flow_steering_support(dev->dev))
169 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
170 	}
171 
172 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
173 		0xffffff;
174 	props->vendor_part_id	   = dev->dev->pdev->device;
175 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
176 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
177 
178 	props->max_mr_size	   = ~0ull;
179 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
180 	props->max_qp		   = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
181 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
182 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
183 					 dev->dev->caps.max_rq_sg);
184 	props->max_cq		   = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
185 	props->max_cqe		   = dev->dev->caps.max_cqes;
186 	props->max_mr		   = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
187 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
188 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
189 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
190 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
191 	props->max_srq		   = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
192 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
193 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
194 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
195 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
196 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
197 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
198 	props->masked_atomic_cap   = props->atomic_cap;
199 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
200 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
201 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
202 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
203 					   props->max_mcast_grp;
204 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
205 
206 out:
207 	kfree(in_mad);
208 	kfree(out_mad);
209 
210 	return err;
211 }
212 
213 static enum rdma_link_layer
214 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
215 {
216 	struct mlx4_dev *dev = to_mdev(device)->dev;
217 
218 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
219 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
220 }
221 
222 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
223 			      struct ib_port_attr *props, int netw_view)
224 {
225 	struct ib_smp *in_mad  = NULL;
226 	struct ib_smp *out_mad = NULL;
227 	int ext_active_speed;
228 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
229 	int err = -ENOMEM;
230 
231 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
232 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
233 	if (!in_mad || !out_mad)
234 		goto out;
235 
236 	init_query_mad(in_mad);
237 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
238 	in_mad->attr_mod = cpu_to_be32(port);
239 
240 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
241 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
242 
243 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
244 				in_mad, out_mad);
245 	if (err)
246 		goto out;
247 
248 
249 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
250 	props->lmc		= out_mad->data[34] & 0x7;
251 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
252 	props->sm_sl		= out_mad->data[36] & 0xf;
253 	props->state		= out_mad->data[32] & 0xf;
254 	props->phys_state	= out_mad->data[33] >> 4;
255 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
256 	if (netw_view)
257 		props->gid_tbl_len = out_mad->data[50];
258 	else
259 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
260 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
261 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
262 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
263 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
264 	props->active_width	= out_mad->data[31] & 0xf;
265 	props->active_speed	= out_mad->data[35] >> 4;
266 	props->max_mtu		= out_mad->data[41] & 0xf;
267 	props->active_mtu	= out_mad->data[36] >> 4;
268 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
269 	props->max_vl_num	= out_mad->data[37] >> 4;
270 	props->init_type_reply	= out_mad->data[41] >> 4;
271 
272 	/* Check if extended speeds (EDR/FDR/...) are supported */
273 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
274 		ext_active_speed = out_mad->data[62] >> 4;
275 
276 		switch (ext_active_speed) {
277 		case 1:
278 			props->active_speed = IB_SPEED_FDR;
279 			break;
280 		case 2:
281 			props->active_speed = IB_SPEED_EDR;
282 			break;
283 		}
284 	}
285 
286 	/* If reported active speed is QDR, check if is FDR-10 */
287 	if (props->active_speed == IB_SPEED_QDR) {
288 		init_query_mad(in_mad);
289 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
290 		in_mad->attr_mod = cpu_to_be32(port);
291 
292 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
293 				   NULL, NULL, in_mad, out_mad);
294 		if (err)
295 			goto out;
296 
297 		/* Checking LinkSpeedActive for FDR-10 */
298 		if (out_mad->data[15] & 0x1)
299 			props->active_speed = IB_SPEED_FDR10;
300 	}
301 
302 	/* Avoid wrong speed value returned by FW if the IB link is down. */
303 	if (props->state == IB_PORT_DOWN)
304 		 props->active_speed = IB_SPEED_SDR;
305 
306 out:
307 	kfree(in_mad);
308 	kfree(out_mad);
309 	return err;
310 }
311 
312 static u8 state_to_phys_state(enum ib_port_state state)
313 {
314 	return state == IB_PORT_ACTIVE ? 5 : 3;
315 }
316 
317 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
318 			       struct ib_port_attr *props, int netw_view)
319 {
320 
321 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
322 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
323 	struct net_device *ndev;
324 	enum ib_mtu tmp;
325 	struct mlx4_cmd_mailbox *mailbox;
326 	int err = 0;
327 
328 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
329 	if (IS_ERR(mailbox))
330 		return PTR_ERR(mailbox);
331 
332 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
333 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
334 			   MLX4_CMD_WRAPPED);
335 	if (err)
336 		goto out;
337 
338 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
339 						IB_WIDTH_4X : IB_WIDTH_1X;
340 	props->active_speed	= IB_SPEED_QDR;
341 	props->port_cap_flags	= IB_PORT_CM_SUP;
342 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
343 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
344 	props->pkey_tbl_len	= 1;
345 	props->max_mtu		= IB_MTU_4096;
346 	props->max_vl_num	= 2;
347 	props->state		= IB_PORT_DOWN;
348 	props->phys_state	= state_to_phys_state(props->state);
349 	props->active_mtu	= IB_MTU_256;
350 	spin_lock(&iboe->lock);
351 	ndev = iboe->netdevs[port - 1];
352 	if (!ndev)
353 		goto out_unlock;
354 
355 	tmp = iboe_get_mtu(ndev->mtu);
356 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
357 
358 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
359 					IB_PORT_ACTIVE : IB_PORT_DOWN;
360 	props->phys_state	= state_to_phys_state(props->state);
361 out_unlock:
362 	spin_unlock(&iboe->lock);
363 out:
364 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
365 	return err;
366 }
367 
368 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
369 			 struct ib_port_attr *props, int netw_view)
370 {
371 	int err;
372 
373 	memset(props, 0, sizeof *props);
374 
375 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
376 		ib_link_query_port(ibdev, port, props, netw_view) :
377 				eth_link_query_port(ibdev, port, props, netw_view);
378 
379 	return err;
380 }
381 
382 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
383 			      struct ib_port_attr *props)
384 {
385 	/* returns host view */
386 	return __mlx4_ib_query_port(ibdev, port, props, 0);
387 }
388 
389 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
390 			union ib_gid *gid, int netw_view)
391 {
392 	struct ib_smp *in_mad  = NULL;
393 	struct ib_smp *out_mad = NULL;
394 	int err = -ENOMEM;
395 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
396 	int clear = 0;
397 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
398 
399 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
400 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
401 	if (!in_mad || !out_mad)
402 		goto out;
403 
404 	init_query_mad(in_mad);
405 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
406 	in_mad->attr_mod = cpu_to_be32(port);
407 
408 	if (mlx4_is_mfunc(dev->dev) && netw_view)
409 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
410 
411 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
412 	if (err)
413 		goto out;
414 
415 	memcpy(gid->raw, out_mad->data + 8, 8);
416 
417 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
418 		if (index) {
419 			/* For any index > 0, return the null guid */
420 			err = 0;
421 			clear = 1;
422 			goto out;
423 		}
424 	}
425 
426 	init_query_mad(in_mad);
427 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
428 	in_mad->attr_mod = cpu_to_be32(index / 8);
429 
430 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
431 			   NULL, NULL, in_mad, out_mad);
432 	if (err)
433 		goto out;
434 
435 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
436 
437 out:
438 	if (clear)
439 		memset(gid->raw + 8, 0, 8);
440 	kfree(in_mad);
441 	kfree(out_mad);
442 	return err;
443 }
444 
445 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
446 			  union ib_gid *gid)
447 {
448 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
449 
450 	*gid = dev->iboe.gid_table[port - 1][index];
451 
452 	return 0;
453 }
454 
455 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
456 			     union ib_gid *gid)
457 {
458 	if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
459 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
460 	else
461 		return iboe_query_gid(ibdev, port, index, gid);
462 }
463 
464 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
465 			 u16 *pkey, int netw_view)
466 {
467 	struct ib_smp *in_mad  = NULL;
468 	struct ib_smp *out_mad = NULL;
469 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
470 	int err = -ENOMEM;
471 
472 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
473 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
474 	if (!in_mad || !out_mad)
475 		goto out;
476 
477 	init_query_mad(in_mad);
478 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
479 	in_mad->attr_mod = cpu_to_be32(index / 32);
480 
481 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
482 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
483 
484 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
485 			   in_mad, out_mad);
486 	if (err)
487 		goto out;
488 
489 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
490 
491 out:
492 	kfree(in_mad);
493 	kfree(out_mad);
494 	return err;
495 }
496 
497 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
498 {
499 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
500 }
501 
502 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
503 				 struct ib_device_modify *props)
504 {
505 	struct mlx4_cmd_mailbox *mailbox;
506 	unsigned long flags;
507 
508 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
509 		return -EOPNOTSUPP;
510 
511 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
512 		return 0;
513 
514 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
515 		return -EOPNOTSUPP;
516 
517 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
518 	memcpy(ibdev->node_desc, props->node_desc, 64);
519 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
520 
521 	/*
522 	 * If possible, pass node desc to FW, so it can generate
523 	 * a 144 trap.  If cmd fails, just ignore.
524 	 */
525 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
526 	if (IS_ERR(mailbox))
527 		return 0;
528 
529 	memset(mailbox->buf, 0, 256);
530 	memcpy(mailbox->buf, props->node_desc, 64);
531 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
532 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
533 
534 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
535 
536 	return 0;
537 }
538 
539 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
540 			 u32 cap_mask)
541 {
542 	struct mlx4_cmd_mailbox *mailbox;
543 	int err;
544 	u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
545 
546 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
547 	if (IS_ERR(mailbox))
548 		return PTR_ERR(mailbox);
549 
550 	memset(mailbox->buf, 0, 256);
551 
552 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
553 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
554 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
555 	} else {
556 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
557 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
558 	}
559 
560 	err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
561 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
562 
563 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
564 	return err;
565 }
566 
567 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
568 			       struct ib_port_modify *props)
569 {
570 	struct ib_port_attr attr;
571 	u32 cap_mask;
572 	int err;
573 
574 	mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
575 
576 	err = mlx4_ib_query_port(ibdev, port, &attr);
577 	if (err)
578 		goto out;
579 
580 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
581 		~props->clr_port_cap_mask;
582 
583 	err = mlx4_SET_PORT(to_mdev(ibdev), port,
584 			    !!(mask & IB_PORT_RESET_QKEY_CNTR),
585 			    cap_mask);
586 
587 out:
588 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
589 	return err;
590 }
591 
592 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
593 						  struct ib_udata *udata)
594 {
595 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
596 	struct mlx4_ib_ucontext *context;
597 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
598 	struct mlx4_ib_alloc_ucontext_resp resp;
599 	int err;
600 
601 	if (!dev->ib_active)
602 		return ERR_PTR(-EAGAIN);
603 
604 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
605 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
606 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
607 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
608 	} else {
609 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
610 		resp.qp_tab_size      = dev->dev->caps.num_qps;
611 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
612 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
613 		resp.cqe_size	      = dev->dev->caps.cqe_size;
614 	}
615 
616 	context = kmalloc(sizeof *context, GFP_KERNEL);
617 	if (!context)
618 		return ERR_PTR(-ENOMEM);
619 
620 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
621 	if (err) {
622 		kfree(context);
623 		return ERR_PTR(err);
624 	}
625 
626 	INIT_LIST_HEAD(&context->db_page_list);
627 	mutex_init(&context->db_page_mutex);
628 
629 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
630 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
631 	else
632 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
633 
634 	if (err) {
635 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
636 		kfree(context);
637 		return ERR_PTR(-EFAULT);
638 	}
639 
640 	return &context->ibucontext;
641 }
642 
643 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
644 {
645 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
646 
647 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
648 	kfree(context);
649 
650 	return 0;
651 }
652 
653 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
654 {
655 	struct mlx4_ib_dev *dev = to_mdev(context->device);
656 
657 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
658 		return -EINVAL;
659 
660 	if (vma->vm_pgoff == 0) {
661 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
662 
663 		if (io_remap_pfn_range(vma, vma->vm_start,
664 				       to_mucontext(context)->uar.pfn,
665 				       PAGE_SIZE, vma->vm_page_prot))
666 			return -EAGAIN;
667 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
668 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
669 
670 		if (io_remap_pfn_range(vma, vma->vm_start,
671 				       to_mucontext(context)->uar.pfn +
672 				       dev->dev->caps.num_uars,
673 				       PAGE_SIZE, vma->vm_page_prot))
674 			return -EAGAIN;
675 	} else
676 		return -EINVAL;
677 
678 	return 0;
679 }
680 
681 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
682 				      struct ib_ucontext *context,
683 				      struct ib_udata *udata)
684 {
685 	struct mlx4_ib_pd *pd;
686 	int err;
687 
688 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
689 	if (!pd)
690 		return ERR_PTR(-ENOMEM);
691 
692 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
693 	if (err) {
694 		kfree(pd);
695 		return ERR_PTR(err);
696 	}
697 
698 	if (context)
699 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
700 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
701 			kfree(pd);
702 			return ERR_PTR(-EFAULT);
703 		}
704 
705 	return &pd->ibpd;
706 }
707 
708 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
709 {
710 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
711 	kfree(pd);
712 
713 	return 0;
714 }
715 
716 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
717 					  struct ib_ucontext *context,
718 					  struct ib_udata *udata)
719 {
720 	struct mlx4_ib_xrcd *xrcd;
721 	int err;
722 
723 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
724 		return ERR_PTR(-ENOSYS);
725 
726 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
727 	if (!xrcd)
728 		return ERR_PTR(-ENOMEM);
729 
730 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
731 	if (err)
732 		goto err1;
733 
734 	xrcd->pd = ib_alloc_pd(ibdev);
735 	if (IS_ERR(xrcd->pd)) {
736 		err = PTR_ERR(xrcd->pd);
737 		goto err2;
738 	}
739 
740 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
741 	if (IS_ERR(xrcd->cq)) {
742 		err = PTR_ERR(xrcd->cq);
743 		goto err3;
744 	}
745 
746 	return &xrcd->ibxrcd;
747 
748 err3:
749 	ib_dealloc_pd(xrcd->pd);
750 err2:
751 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
752 err1:
753 	kfree(xrcd);
754 	return ERR_PTR(err);
755 }
756 
757 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
758 {
759 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
760 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
761 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
762 	kfree(xrcd);
763 
764 	return 0;
765 }
766 
767 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
768 {
769 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
770 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
771 	struct mlx4_ib_gid_entry *ge;
772 
773 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
774 	if (!ge)
775 		return -ENOMEM;
776 
777 	ge->gid = *gid;
778 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
779 		ge->port = mqp->port;
780 		ge->added = 1;
781 	}
782 
783 	mutex_lock(&mqp->mutex);
784 	list_add_tail(&ge->list, &mqp->gid_list);
785 	mutex_unlock(&mqp->mutex);
786 
787 	return 0;
788 }
789 
790 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
791 		   union ib_gid *gid)
792 {
793 	u8 mac[6];
794 	struct net_device *ndev;
795 	int ret = 0;
796 
797 	if (!mqp->port)
798 		return 0;
799 
800 	spin_lock(&mdev->iboe.lock);
801 	ndev = mdev->iboe.netdevs[mqp->port - 1];
802 	if (ndev)
803 		dev_hold(ndev);
804 	spin_unlock(&mdev->iboe.lock);
805 
806 	if (ndev) {
807 		rdma_get_mcast_mac((struct in6_addr *)gid, mac);
808 		rtnl_lock();
809 		dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac);
810 		ret = 1;
811 		rtnl_unlock();
812 		dev_put(ndev);
813 	}
814 
815 	return ret;
816 }
817 
818 struct mlx4_ib_steering {
819 	struct list_head list;
820 	u64 reg_id;
821 	union ib_gid gid;
822 };
823 
824 static int parse_flow_attr(struct mlx4_dev *dev,
825 			   union ib_flow_spec *ib_spec,
826 			   struct _rule_hw *mlx4_spec)
827 {
828 	enum mlx4_net_trans_rule_id type;
829 
830 	switch (ib_spec->type) {
831 	case IB_FLOW_SPEC_ETH:
832 		type = MLX4_NET_TRANS_RULE_ID_ETH;
833 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
834 		       ETH_ALEN);
835 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
836 		       ETH_ALEN);
837 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
838 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
839 		break;
840 
841 	case IB_FLOW_SPEC_IPV4:
842 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
843 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
844 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
845 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
846 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
847 		break;
848 
849 	case IB_FLOW_SPEC_TCP:
850 	case IB_FLOW_SPEC_UDP:
851 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
852 					MLX4_NET_TRANS_RULE_ID_TCP :
853 					MLX4_NET_TRANS_RULE_ID_UDP;
854 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
855 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
856 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
857 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
858 		break;
859 
860 	default:
861 		return -EINVAL;
862 	}
863 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
864 	    mlx4_hw_rule_sz(dev, type) < 0)
865 		return -EINVAL;
866 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
867 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
868 	return mlx4_hw_rule_sz(dev, type);
869 }
870 
871 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
872 			  int domain,
873 			  enum mlx4_net_trans_promisc_mode flow_type,
874 			  u64 *reg_id)
875 {
876 	int ret, i;
877 	int size = 0;
878 	void *ib_flow;
879 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
880 	struct mlx4_cmd_mailbox *mailbox;
881 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
882 	size_t rule_size = sizeof(struct mlx4_net_trans_rule_hw_ctrl) +
883 			   (sizeof(struct _rule_hw) * flow_attr->num_of_specs);
884 
885 	static const u16 __mlx4_domain[] = {
886 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
887 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
888 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
889 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
890 	};
891 
892 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
893 		pr_err("Invalid priority value %d\n", flow_attr->priority);
894 		return -EINVAL;
895 	}
896 
897 	if (domain >= IB_FLOW_DOMAIN_NUM) {
898 		pr_err("Invalid domain value %d\n", domain);
899 		return -EINVAL;
900 	}
901 
902 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
903 		return -EINVAL;
904 
905 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
906 	if (IS_ERR(mailbox))
907 		return PTR_ERR(mailbox);
908 	memset(mailbox->buf, 0, rule_size);
909 	ctrl = mailbox->buf;
910 
911 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
912 				 flow_attr->priority);
913 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
914 	ctrl->port = flow_attr->port;
915 	ctrl->qpn = cpu_to_be32(qp->qp_num);
916 
917 	ib_flow = flow_attr + 1;
918 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
919 	for (i = 0; i < flow_attr->num_of_specs; i++) {
920 		ret = parse_flow_attr(mdev->dev, ib_flow, mailbox->buf + size);
921 		if (ret < 0) {
922 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
923 			return -EINVAL;
924 		}
925 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
926 		size += ret;
927 	}
928 
929 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
930 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
931 			   MLX4_CMD_NATIVE);
932 	if (ret == -ENOMEM)
933 		pr_err("mcg table is full. Fail to register network rule.\n");
934 	else if (ret == -ENXIO)
935 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
936 	else if (ret)
937 		pr_err("Invalid argumant. Fail to register network rule.\n");
938 
939 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
940 	return ret;
941 }
942 
943 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
944 {
945 	int err;
946 	err = mlx4_cmd(dev, reg_id, 0, 0,
947 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
948 		       MLX4_CMD_NATIVE);
949 	if (err)
950 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
951 		       reg_id);
952 	return err;
953 }
954 
955 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
956 				    struct ib_flow_attr *flow_attr,
957 				    int domain)
958 {
959 	int err = 0, i = 0;
960 	struct mlx4_ib_flow *mflow;
961 	enum mlx4_net_trans_promisc_mode type[2];
962 
963 	memset(type, 0, sizeof(type));
964 
965 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
966 	if (!mflow) {
967 		err = -ENOMEM;
968 		goto err_free;
969 	}
970 
971 	switch (flow_attr->type) {
972 	case IB_FLOW_ATTR_NORMAL:
973 		type[0] = MLX4_FS_REGULAR;
974 		break;
975 
976 	case IB_FLOW_ATTR_ALL_DEFAULT:
977 		type[0] = MLX4_FS_ALL_DEFAULT;
978 		break;
979 
980 	case IB_FLOW_ATTR_MC_DEFAULT:
981 		type[0] = MLX4_FS_MC_DEFAULT;
982 		break;
983 
984 	case IB_FLOW_ATTR_SNIFFER:
985 		type[0] = MLX4_FS_UC_SNIFFER;
986 		type[1] = MLX4_FS_MC_SNIFFER;
987 		break;
988 
989 	default:
990 		err = -EINVAL;
991 		goto err_free;
992 	}
993 
994 	while (i < ARRAY_SIZE(type) && type[i]) {
995 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
996 					    &mflow->reg_id[i]);
997 		if (err)
998 			goto err_free;
999 		i++;
1000 	}
1001 
1002 	return &mflow->ibflow;
1003 
1004 err_free:
1005 	kfree(mflow);
1006 	return ERR_PTR(err);
1007 }
1008 
1009 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1010 {
1011 	int err, ret = 0;
1012 	int i = 0;
1013 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1014 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1015 
1016 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) {
1017 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]);
1018 		if (err)
1019 			ret = err;
1020 		i++;
1021 	}
1022 
1023 	kfree(mflow);
1024 	return ret;
1025 }
1026 
1027 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1028 {
1029 	int err;
1030 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1031 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1032 	u64 reg_id;
1033 	struct mlx4_ib_steering *ib_steering = NULL;
1034 
1035 	if (mdev->dev->caps.steering_mode ==
1036 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1037 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1038 		if (!ib_steering)
1039 			return -ENOMEM;
1040 	}
1041 
1042 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1043 				    !!(mqp->flags &
1044 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1045 				    MLX4_PROT_IB_IPV6, &reg_id);
1046 	if (err)
1047 		goto err_malloc;
1048 
1049 	err = add_gid_entry(ibqp, gid);
1050 	if (err)
1051 		goto err_add;
1052 
1053 	if (ib_steering) {
1054 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1055 		ib_steering->reg_id = reg_id;
1056 		mutex_lock(&mqp->mutex);
1057 		list_add(&ib_steering->list, &mqp->steering_rules);
1058 		mutex_unlock(&mqp->mutex);
1059 	}
1060 	return 0;
1061 
1062 err_add:
1063 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1064 			      MLX4_PROT_IB_IPV6, reg_id);
1065 err_malloc:
1066 	kfree(ib_steering);
1067 
1068 	return err;
1069 }
1070 
1071 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1072 {
1073 	struct mlx4_ib_gid_entry *ge;
1074 	struct mlx4_ib_gid_entry *tmp;
1075 	struct mlx4_ib_gid_entry *ret = NULL;
1076 
1077 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1078 		if (!memcmp(raw, ge->gid.raw, 16)) {
1079 			ret = ge;
1080 			break;
1081 		}
1082 	}
1083 
1084 	return ret;
1085 }
1086 
1087 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1088 {
1089 	int err;
1090 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1091 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1092 	u8 mac[6];
1093 	struct net_device *ndev;
1094 	struct mlx4_ib_gid_entry *ge;
1095 	u64 reg_id = 0;
1096 
1097 	if (mdev->dev->caps.steering_mode ==
1098 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1099 		struct mlx4_ib_steering *ib_steering;
1100 
1101 		mutex_lock(&mqp->mutex);
1102 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1103 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1104 				list_del(&ib_steering->list);
1105 				break;
1106 			}
1107 		}
1108 		mutex_unlock(&mqp->mutex);
1109 		if (&ib_steering->list == &mqp->steering_rules) {
1110 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1111 			return -EINVAL;
1112 		}
1113 		reg_id = ib_steering->reg_id;
1114 		kfree(ib_steering);
1115 	}
1116 
1117 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1118 				    MLX4_PROT_IB_IPV6, reg_id);
1119 	if (err)
1120 		return err;
1121 
1122 	mutex_lock(&mqp->mutex);
1123 	ge = find_gid_entry(mqp, gid->raw);
1124 	if (ge) {
1125 		spin_lock(&mdev->iboe.lock);
1126 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1127 		if (ndev)
1128 			dev_hold(ndev);
1129 		spin_unlock(&mdev->iboe.lock);
1130 		rdma_get_mcast_mac((struct in6_addr *)gid, mac);
1131 		if (ndev) {
1132 			rtnl_lock();
1133 			dev_mc_del(mdev->iboe.netdevs[ge->port - 1], mac);
1134 			rtnl_unlock();
1135 			dev_put(ndev);
1136 		}
1137 		list_del(&ge->list);
1138 		kfree(ge);
1139 	} else
1140 		pr_warn("could not find mgid entry\n");
1141 
1142 	mutex_unlock(&mqp->mutex);
1143 
1144 	return 0;
1145 }
1146 
1147 static int init_node_data(struct mlx4_ib_dev *dev)
1148 {
1149 	struct ib_smp *in_mad  = NULL;
1150 	struct ib_smp *out_mad = NULL;
1151 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1152 	int err = -ENOMEM;
1153 
1154 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1155 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1156 	if (!in_mad || !out_mad)
1157 		goto out;
1158 
1159 	init_query_mad(in_mad);
1160 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1161 	if (mlx4_is_master(dev->dev))
1162 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1163 
1164 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1165 	if (err)
1166 		goto out;
1167 
1168 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1169 
1170 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1171 
1172 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1173 	if (err)
1174 		goto out;
1175 
1176 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1177 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1178 
1179 out:
1180 	kfree(in_mad);
1181 	kfree(out_mad);
1182 	return err;
1183 }
1184 
1185 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1186 			char *buf)
1187 {
1188 	struct mlx4_ib_dev *dev =
1189 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1190 	return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
1191 }
1192 
1193 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1194 			   char *buf)
1195 {
1196 	struct mlx4_ib_dev *dev =
1197 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1198 	return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1199 		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1200 		       (int) dev->dev->caps.fw_ver & 0xffff);
1201 }
1202 
1203 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1204 			char *buf)
1205 {
1206 	struct mlx4_ib_dev *dev =
1207 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1208 	return sprintf(buf, "%x\n", dev->dev->rev_id);
1209 }
1210 
1211 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1212 			  char *buf)
1213 {
1214 	struct mlx4_ib_dev *dev =
1215 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1216 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1217 		       dev->dev->board_id);
1218 }
1219 
1220 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1221 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1222 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1223 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1224 
1225 static struct device_attribute *mlx4_class_attributes[] = {
1226 	&dev_attr_hw_rev,
1227 	&dev_attr_fw_ver,
1228 	&dev_attr_hca_type,
1229 	&dev_attr_board_id
1230 };
1231 
1232 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
1233 {
1234 	memcpy(eui, dev->dev_addr, 3);
1235 	memcpy(eui + 5, dev->dev_addr + 3, 3);
1236 	if (vlan_id < 0x1000) {
1237 		eui[3] = vlan_id >> 8;
1238 		eui[4] = vlan_id & 0xff;
1239 	} else {
1240 		eui[3] = 0xff;
1241 		eui[4] = 0xfe;
1242 	}
1243 	eui[0] ^= 2;
1244 }
1245 
1246 static void update_gids_task(struct work_struct *work)
1247 {
1248 	struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1249 	struct mlx4_cmd_mailbox *mailbox;
1250 	union ib_gid *gids;
1251 	int err;
1252 	struct mlx4_dev	*dev = gw->dev->dev;
1253 
1254 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1255 	if (IS_ERR(mailbox)) {
1256 		pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1257 		return;
1258 	}
1259 
1260 	gids = mailbox->buf;
1261 	memcpy(gids, gw->gids, sizeof gw->gids);
1262 
1263 	err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1264 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1265 		       MLX4_CMD_WRAPPED);
1266 	if (err)
1267 		pr_warn("set port command failed\n");
1268 	else {
1269 		memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
1270 		mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1271 	}
1272 
1273 	mlx4_free_cmd_mailbox(dev, mailbox);
1274 	kfree(gw);
1275 }
1276 
1277 static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
1278 {
1279 	struct net_device *ndev = dev->iboe.netdevs[port - 1];
1280 	struct update_gid_work *work;
1281 	struct net_device *tmp;
1282 	int i;
1283 	u8 *hits;
1284 	int ret;
1285 	union ib_gid gid;
1286 	int free;
1287 	int found;
1288 	int need_update = 0;
1289 	u16 vid;
1290 
1291 	work = kzalloc(sizeof *work, GFP_ATOMIC);
1292 	if (!work)
1293 		return -ENOMEM;
1294 
1295 	hits = kzalloc(128, GFP_ATOMIC);
1296 	if (!hits) {
1297 		ret = -ENOMEM;
1298 		goto out;
1299 	}
1300 
1301 	rcu_read_lock();
1302 	for_each_netdev_rcu(&init_net, tmp) {
1303 		if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
1304 			gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1305 			vid = rdma_vlan_dev_vlan_id(tmp);
1306 			mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
1307 			found = 0;
1308 			free = -1;
1309 			for (i = 0; i < 128; ++i) {
1310 				if (free < 0 &&
1311 				    !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1312 					free = i;
1313 				if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
1314 					hits[i] = 1;
1315 					found = 1;
1316 					break;
1317 				}
1318 			}
1319 
1320 			if (!found) {
1321 				if (tmp == ndev &&
1322 				    (memcmp(&dev->iboe.gid_table[port - 1][0],
1323 					    &gid, sizeof gid) ||
1324 				     !memcmp(&dev->iboe.gid_table[port - 1][0],
1325 					     &zgid, sizeof gid))) {
1326 					dev->iboe.gid_table[port - 1][0] = gid;
1327 					++need_update;
1328 					hits[0] = 1;
1329 				} else if (free >= 0) {
1330 					dev->iboe.gid_table[port - 1][free] = gid;
1331 					hits[free] = 1;
1332 					++need_update;
1333 				}
1334 			}
1335 		}
1336 	}
1337 	rcu_read_unlock();
1338 
1339 	for (i = 0; i < 128; ++i)
1340 		if (!hits[i]) {
1341 			if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1342 				++need_update;
1343 			dev->iboe.gid_table[port - 1][i] = zgid;
1344 		}
1345 
1346 	if (need_update) {
1347 		memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
1348 		INIT_WORK(&work->work, update_gids_task);
1349 		work->port = port;
1350 		work->dev = dev;
1351 		queue_work(wq, &work->work);
1352 	} else
1353 		kfree(work);
1354 
1355 	kfree(hits);
1356 	return 0;
1357 
1358 out:
1359 	kfree(work);
1360 	return ret;
1361 }
1362 
1363 static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
1364 {
1365 	switch (event) {
1366 	case NETDEV_UP:
1367 	case NETDEV_CHANGEADDR:
1368 		update_ipv6_gids(dev, port, 0);
1369 		break;
1370 
1371 	case NETDEV_DOWN:
1372 		update_ipv6_gids(dev, port, 1);
1373 		dev->iboe.netdevs[port - 1] = NULL;
1374 	}
1375 }
1376 
1377 static void netdev_added(struct mlx4_ib_dev *dev, int port)
1378 {
1379 	update_ipv6_gids(dev, port, 0);
1380 }
1381 
1382 static void netdev_removed(struct mlx4_ib_dev *dev, int port)
1383 {
1384 	update_ipv6_gids(dev, port, 1);
1385 }
1386 
1387 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1388 				void *ptr)
1389 {
1390 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1391 	struct mlx4_ib_dev *ibdev;
1392 	struct net_device *oldnd;
1393 	struct mlx4_ib_iboe *iboe;
1394 	int port;
1395 
1396 	if (!net_eq(dev_net(dev), &init_net))
1397 		return NOTIFY_DONE;
1398 
1399 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1400 	iboe = &ibdev->iboe;
1401 
1402 	spin_lock(&iboe->lock);
1403 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1404 		oldnd = iboe->netdevs[port - 1];
1405 		iboe->netdevs[port - 1] =
1406 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1407 		if (oldnd != iboe->netdevs[port - 1]) {
1408 			if (iboe->netdevs[port - 1])
1409 				netdev_added(ibdev, port);
1410 			else
1411 				netdev_removed(ibdev, port);
1412 		}
1413 	}
1414 
1415 	if (dev == iboe->netdevs[0] ||
1416 	    (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
1417 		handle_en_event(ibdev, 1, event);
1418 	else if (dev == iboe->netdevs[1]
1419 		 || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
1420 		handle_en_event(ibdev, 2, event);
1421 
1422 	spin_unlock(&iboe->lock);
1423 
1424 	return NOTIFY_DONE;
1425 }
1426 
1427 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1428 {
1429 	int port;
1430 	int slave;
1431 	int i;
1432 
1433 	if (mlx4_is_master(ibdev->dev)) {
1434 		for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1435 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1436 				for (i = 0;
1437 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1438 				     ++i) {
1439 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1440 					/* master has the identity virt2phys pkey mapping */
1441 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1442 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1443 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1444 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1445 				}
1446 			}
1447 		}
1448 		/* initialize pkey cache */
1449 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1450 			for (i = 0;
1451 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1452 			     ++i)
1453 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
1454 					(i) ? 0 : 0xFFFF;
1455 		}
1456 	}
1457 }
1458 
1459 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1460 {
1461 	char name[32];
1462 	int eq_per_port = 0;
1463 	int added_eqs = 0;
1464 	int total_eqs = 0;
1465 	int i, j, eq;
1466 
1467 	/* Legacy mode or comp_pool is not large enough */
1468 	if (dev->caps.comp_pool == 0 ||
1469 	    dev->caps.num_ports > dev->caps.comp_pool)
1470 		return;
1471 
1472 	eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1473 					dev->caps.num_ports);
1474 
1475 	/* Init eq table */
1476 	added_eqs = 0;
1477 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1478 		added_eqs += eq_per_port;
1479 
1480 	total_eqs = dev->caps.num_comp_vectors + added_eqs;
1481 
1482 	ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1483 	if (!ibdev->eq_table)
1484 		return;
1485 
1486 	ibdev->eq_added = added_eqs;
1487 
1488 	eq = 0;
1489 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1490 		for (j = 0; j < eq_per_port; j++) {
1491 			sprintf(name, "mlx4-ib-%d-%d@%s",
1492 				i, j, dev->pdev->bus->name);
1493 			/* Set IRQ for specific name (per ring) */
1494 			if (mlx4_assign_eq(dev, name, NULL,
1495 					   &ibdev->eq_table[eq])) {
1496 				/* Use legacy (same as mlx4_en driver) */
1497 				pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1498 				ibdev->eq_table[eq] =
1499 					(eq % dev->caps.num_comp_vectors);
1500 			}
1501 			eq++;
1502 		}
1503 	}
1504 
1505 	/* Fill the reset of the vector with legacy EQ */
1506 	for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1507 		ibdev->eq_table[eq++] = i;
1508 
1509 	/* Advertise the new number of EQs to clients */
1510 	ibdev->ib_dev.num_comp_vectors = total_eqs;
1511 }
1512 
1513 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1514 {
1515 	int i;
1516 
1517 	/* no additional eqs were added */
1518 	if (!ibdev->eq_table)
1519 		return;
1520 
1521 	/* Reset the advertised EQ number */
1522 	ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1523 
1524 	/* Free only the added eqs */
1525 	for (i = 0; i < ibdev->eq_added; i++) {
1526 		/* Don't free legacy eqs if used */
1527 		if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1528 			continue;
1529 		mlx4_release_eq(dev, ibdev->eq_table[i]);
1530 	}
1531 
1532 	kfree(ibdev->eq_table);
1533 }
1534 
1535 static void *mlx4_ib_add(struct mlx4_dev *dev)
1536 {
1537 	struct mlx4_ib_dev *ibdev;
1538 	int num_ports = 0;
1539 	int i, j;
1540 	int err;
1541 	struct mlx4_ib_iboe *iboe;
1542 
1543 	pr_info_once("%s", mlx4_ib_version);
1544 
1545 	mlx4_foreach_non_ib_transport_port(i, dev)
1546 		num_ports++;
1547 
1548 	if (mlx4_is_mfunc(dev) && num_ports) {
1549 		dev_err(&dev->pdev->dev, "RoCE is not supported over SRIOV as yet\n");
1550 		return NULL;
1551 	}
1552 
1553 	num_ports = 0;
1554 	mlx4_foreach_ib_transport_port(i, dev)
1555 		num_ports++;
1556 
1557 	/* No point in registering a device with no ports... */
1558 	if (num_ports == 0)
1559 		return NULL;
1560 
1561 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1562 	if (!ibdev) {
1563 		dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1564 		return NULL;
1565 	}
1566 
1567 	iboe = &ibdev->iboe;
1568 
1569 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1570 		goto err_dealloc;
1571 
1572 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1573 		goto err_pd;
1574 
1575 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1576 				 PAGE_SIZE);
1577 	if (!ibdev->uar_map)
1578 		goto err_uar;
1579 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1580 
1581 	ibdev->dev = dev;
1582 
1583 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1584 	ibdev->ib_dev.owner		= THIS_MODULE;
1585 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
1586 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
1587 	ibdev->num_ports		= num_ports;
1588 	ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
1589 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
1590 	ibdev->ib_dev.dma_device	= &dev->pdev->dev;
1591 
1592 	if (dev->caps.userspace_caps)
1593 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
1594 	else
1595 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
1596 
1597 	ibdev->ib_dev.uverbs_cmd_mask	=
1598 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
1599 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
1600 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
1601 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
1602 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
1603 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
1604 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
1605 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
1606 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
1607 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
1608 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
1609 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
1610 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
1611 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
1612 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
1613 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
1614 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
1615 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
1616 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
1617 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
1618 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
1619 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
1620 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
1621 
1622 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
1623 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
1624 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
1625 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
1626 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
1627 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
1628 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
1629 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
1630 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
1631 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
1632 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
1633 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
1634 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
1635 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
1636 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
1637 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
1638 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
1639 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
1640 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
1641 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
1642 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
1643 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
1644 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
1645 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
1646 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
1647 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
1648 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
1649 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
1650 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
1651 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
1652 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
1653 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
1654 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
1655 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
1656 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
1657 	ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1658 	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1659 	ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
1660 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
1661 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
1662 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
1663 
1664 	if (!mlx4_is_slave(ibdev->dev)) {
1665 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
1666 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
1667 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
1668 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
1669 	}
1670 
1671 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1672 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
1673 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
1674 		ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
1675 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
1676 
1677 		ibdev->ib_dev.uverbs_cmd_mask |=
1678 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
1679 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
1680 	}
1681 
1682 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1683 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1684 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1685 		ibdev->ib_dev.uverbs_cmd_mask |=
1686 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1687 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1688 	}
1689 
1690 	if (check_flow_steering_support(dev)) {
1691 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
1692 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
1693 
1694 		ibdev->ib_dev.uverbs_cmd_mask	|=
1695 			(1ull << IB_USER_VERBS_CMD_CREATE_FLOW) |
1696 			(1ull << IB_USER_VERBS_CMD_DESTROY_FLOW);
1697 	}
1698 
1699 	mlx4_ib_alloc_eqs(dev, ibdev);
1700 
1701 	spin_lock_init(&iboe->lock);
1702 
1703 	if (init_node_data(ibdev))
1704 		goto err_map;
1705 
1706 	for (i = 0; i < ibdev->num_ports; ++i) {
1707 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1708 						IB_LINK_LAYER_ETHERNET) {
1709 			err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1710 			if (err)
1711 				ibdev->counters[i] = -1;
1712 		} else
1713 				ibdev->counters[i] = -1;
1714 	}
1715 
1716 	spin_lock_init(&ibdev->sm_lock);
1717 	mutex_init(&ibdev->cap_mask_mutex);
1718 
1719 	if (ib_register_device(&ibdev->ib_dev, NULL))
1720 		goto err_counter;
1721 
1722 	if (mlx4_ib_mad_init(ibdev))
1723 		goto err_reg;
1724 
1725 	if (mlx4_ib_init_sriov(ibdev))
1726 		goto err_mad;
1727 
1728 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1729 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
1730 		err = register_netdevice_notifier(&iboe->nb);
1731 		if (err)
1732 			goto err_sriov;
1733 	}
1734 
1735 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
1736 		if (device_create_file(&ibdev->ib_dev.dev,
1737 				       mlx4_class_attributes[j]))
1738 			goto err_notif;
1739 	}
1740 
1741 	ibdev->ib_active = true;
1742 
1743 	if (mlx4_is_mfunc(ibdev->dev))
1744 		init_pkeys(ibdev);
1745 
1746 	/* create paravirt contexts for any VFs which are active */
1747 	if (mlx4_is_master(ibdev->dev)) {
1748 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
1749 			if (j == mlx4_master_func_num(ibdev->dev))
1750 				continue;
1751 			if (mlx4_is_slave_active(ibdev->dev, j))
1752 				do_slave_init(ibdev, j, 1);
1753 		}
1754 	}
1755 	return ibdev;
1756 
1757 err_notif:
1758 	if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1759 		pr_warn("failure unregistering notifier\n");
1760 	flush_workqueue(wq);
1761 
1762 err_sriov:
1763 	mlx4_ib_close_sriov(ibdev);
1764 
1765 err_mad:
1766 	mlx4_ib_mad_cleanup(ibdev);
1767 
1768 err_reg:
1769 	ib_unregister_device(&ibdev->ib_dev);
1770 
1771 err_counter:
1772 	for (; i; --i)
1773 		if (ibdev->counters[i - 1] != -1)
1774 			mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
1775 
1776 err_map:
1777 	iounmap(ibdev->uar_map);
1778 
1779 err_uar:
1780 	mlx4_uar_free(dev, &ibdev->priv_uar);
1781 
1782 err_pd:
1783 	mlx4_pd_free(dev, ibdev->priv_pdn);
1784 
1785 err_dealloc:
1786 	ib_dealloc_device(&ibdev->ib_dev);
1787 
1788 	return NULL;
1789 }
1790 
1791 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1792 {
1793 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
1794 	int p;
1795 
1796 	mlx4_ib_close_sriov(ibdev);
1797 	mlx4_ib_mad_cleanup(ibdev);
1798 	ib_unregister_device(&ibdev->ib_dev);
1799 	if (ibdev->iboe.nb.notifier_call) {
1800 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1801 			pr_warn("failure unregistering notifier\n");
1802 		ibdev->iboe.nb.notifier_call = NULL;
1803 	}
1804 	iounmap(ibdev->uar_map);
1805 	for (p = 0; p < ibdev->num_ports; ++p)
1806 		if (ibdev->counters[p] != -1)
1807 			mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
1808 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
1809 		mlx4_CLOSE_PORT(dev, p);
1810 
1811 	mlx4_ib_free_eqs(dev, ibdev);
1812 
1813 	mlx4_uar_free(dev, &ibdev->priv_uar);
1814 	mlx4_pd_free(dev, ibdev->priv_pdn);
1815 	ib_dealloc_device(&ibdev->ib_dev);
1816 }
1817 
1818 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
1819 {
1820 	struct mlx4_ib_demux_work **dm = NULL;
1821 	struct mlx4_dev *dev = ibdev->dev;
1822 	int i;
1823 	unsigned long flags;
1824 
1825 	if (!mlx4_is_master(dev))
1826 		return;
1827 
1828 	dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC);
1829 	if (!dm) {
1830 		pr_err("failed to allocate memory for tunneling qp update\n");
1831 		goto out;
1832 	}
1833 
1834 	for (i = 0; i < dev->caps.num_ports; i++) {
1835 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
1836 		if (!dm[i]) {
1837 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
1838 			for (i = 0; i < dev->caps.num_ports; i++) {
1839 				if (dm[i])
1840 					kfree(dm[i]);
1841 			}
1842 			goto out;
1843 		}
1844 	}
1845 	/* initialize or tear down tunnel QPs for the slave */
1846 	for (i = 0; i < dev->caps.num_ports; i++) {
1847 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
1848 		dm[i]->port = i + 1;
1849 		dm[i]->slave = slave;
1850 		dm[i]->do_init = do_init;
1851 		dm[i]->dev = ibdev;
1852 		spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
1853 		if (!ibdev->sriov.is_going_down)
1854 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
1855 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
1856 	}
1857 out:
1858 	kfree(dm);
1859 	return;
1860 }
1861 
1862 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
1863 			  enum mlx4_dev_event event, unsigned long param)
1864 {
1865 	struct ib_event ibev;
1866 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
1867 	struct mlx4_eqe *eqe = NULL;
1868 	struct ib_event_work *ew;
1869 	int p = 0;
1870 
1871 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
1872 		eqe = (struct mlx4_eqe *)param;
1873 	else
1874 		p = (int) param;
1875 
1876 	switch (event) {
1877 	case MLX4_DEV_EVENT_PORT_UP:
1878 		if (p > ibdev->num_ports)
1879 			return;
1880 		if (mlx4_is_master(dev) &&
1881 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
1882 			IB_LINK_LAYER_INFINIBAND) {
1883 			mlx4_ib_invalidate_all_guid_record(ibdev, p);
1884 		}
1885 		ibev.event = IB_EVENT_PORT_ACTIVE;
1886 		break;
1887 
1888 	case MLX4_DEV_EVENT_PORT_DOWN:
1889 		if (p > ibdev->num_ports)
1890 			return;
1891 		ibev.event = IB_EVENT_PORT_ERR;
1892 		break;
1893 
1894 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
1895 		ibdev->ib_active = false;
1896 		ibev.event = IB_EVENT_DEVICE_FATAL;
1897 		break;
1898 
1899 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
1900 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
1901 		if (!ew) {
1902 			pr_err("failed to allocate memory for events work\n");
1903 			break;
1904 		}
1905 
1906 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
1907 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
1908 		ew->ib_dev = ibdev;
1909 		/* need to queue only for port owner, which uses GEN_EQE */
1910 		if (mlx4_is_master(dev))
1911 			queue_work(wq, &ew->work);
1912 		else
1913 			handle_port_mgmt_change_event(&ew->work);
1914 		return;
1915 
1916 	case MLX4_DEV_EVENT_SLAVE_INIT:
1917 		/* here, p is the slave id */
1918 		do_slave_init(ibdev, p, 1);
1919 		return;
1920 
1921 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
1922 		/* here, p is the slave id */
1923 		do_slave_init(ibdev, p, 0);
1924 		return;
1925 
1926 	default:
1927 		return;
1928 	}
1929 
1930 	ibev.device	      = ibdev_ptr;
1931 	ibev.element.port_num = (u8) p;
1932 
1933 	ib_dispatch_event(&ibev);
1934 }
1935 
1936 static struct mlx4_interface mlx4_ib_interface = {
1937 	.add		= mlx4_ib_add,
1938 	.remove		= mlx4_ib_remove,
1939 	.event		= mlx4_ib_event,
1940 	.protocol	= MLX4_PROT_IB_IPV6
1941 };
1942 
1943 static int __init mlx4_ib_init(void)
1944 {
1945 	int err;
1946 
1947 	wq = create_singlethread_workqueue("mlx4_ib");
1948 	if (!wq)
1949 		return -ENOMEM;
1950 
1951 	err = mlx4_ib_mcg_init();
1952 	if (err)
1953 		goto clean_wq;
1954 
1955 	err = mlx4_register_interface(&mlx4_ib_interface);
1956 	if (err)
1957 		goto clean_mcg;
1958 
1959 	return 0;
1960 
1961 clean_mcg:
1962 	mlx4_ib_mcg_destroy();
1963 
1964 clean_wq:
1965 	destroy_workqueue(wq);
1966 	return err;
1967 }
1968 
1969 static void __exit mlx4_ib_cleanup(void)
1970 {
1971 	mlx4_unregister_interface(&mlx4_ib_interface);
1972 	mlx4_ib_mcg_destroy();
1973 	destroy_workqueue(wq);
1974 }
1975 
1976 module_init(mlx4_ib_init);
1977 module_exit(mlx4_ib_cleanup);
1978