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/errno.h>
37#include <linux/netdevice.h>
38#include <linux/inetdevice.h>
39#include <linux/rtnetlink.h>
40#include <linux/if_vlan.h>
41
42#include <rdma/ib_smi.h>
43#include <rdma/ib_user_verbs.h>
44#include <rdma/ib_addr.h>
45
46#include <linux/mlx4/driver.h>
47#include <linux/mlx4/cmd.h>
48
49#include "mlx4_ib.h"
50#include "user.h"
51#include "wc.h"
52
53#define DRV_NAME	MLX4_IB_DRV_NAME
54#define DRV_VERSION	"1.0-ofed1.5.2"
55#define DRV_RELDATE	"August 4, 2010"
56
57MODULE_AUTHOR("Roland Dreier");
58MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
59MODULE_LICENSE("Dual BSD/GPL");
60MODULE_VERSION(DRV_VERSION);
61
62#ifdef CONFIG_MLX4_DEBUG
63
64int mlx4_ib_debug_level = 0;
65module_param_named(debug_level, mlx4_ib_debug_level, int, 0644);
66MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
67
68#endif /* CONFIG_MLX4_DEBUG */
69
70static const char mlx4_ib_version[] =
71	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
72	DRV_VERSION " (" DRV_RELDATE ")\n";
73
74static void *get_ibdev(struct mlx4_dev *dev, void *ctx, u8 port)
75{
76       struct mlx4_ib_dev *mlxibdev = ctx;
77       return &mlxibdev->ib_dev;
78}
79
80struct update_gid_work {
81	struct work_struct work;
82	union ib_gid gids[128];
83	int port;
84	struct mlx4_ib_dev *dev;
85};
86
87static struct workqueue_struct *wq;
88
89static void init_query_mad(struct ib_smp *mad)
90{
91	mad->base_version  = 1;
92	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
93	mad->class_version = 1;
94	mad->method	   = IB_MGMT_METHOD_GET;
95}
96
97static union ib_gid zgid;
98
99static int mlx4_ib_query_device(struct ib_device *ibdev,
100				struct ib_device_attr *props)
101{
102	struct mlx4_ib_dev *dev = to_mdev(ibdev);
103	struct ib_smp *in_mad  = NULL;
104	struct ib_smp *out_mad = NULL;
105	int err = -ENOMEM;
106
107	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
108	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
109	if (!in_mad || !out_mad)
110		goto out;
111
112	init_query_mad(in_mad);
113	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
114
115	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, 1, NULL, NULL, in_mad, out_mad);
116	if (err)
117		goto out;
118
119	memset(props, 0, sizeof *props);
120
121	props->fw_ver = dev->dev->caps.fw_ver;
122	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
123		IB_DEVICE_PORT_ACTIVE_EVENT		|
124		IB_DEVICE_SYS_IMAGE_GUID		|
125		IB_DEVICE_RC_RNR_NAK_GEN		|
126		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
127	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
128		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
129	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
130		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
131	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
132		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
133	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
134		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
135	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
136		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
137	if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
138		props->device_cap_flags |= IB_DEVICE_UD_TSO;
139	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
140		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
141	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
142	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
143	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
144		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
145	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
146		props->device_cap_flags |= IB_DEVICE_XRC;
147	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_RAW_ETY)
148		props->max_raw_ethy_qp = dev->ib_dev.phys_port_cnt;
149
150	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
151		0xffffff;
152	props->vendor_part_id	   = be16_to_cpup((__be16 *) (out_mad->data + 30));
153	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
154	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
155
156	props->max_mr_size	   = ~0ull;
157	props->page_size_cap	   = dev->dev->caps.page_size_cap;
158	props->max_qp		   = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
159	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
160	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
161					 dev->dev->caps.max_rq_sg);
162	props->max_cq		   = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
163	props->max_cqe		   = dev->dev->caps.max_cqes;
164	props->max_mr		   = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
165	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
166	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
167	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
168	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
169	props->max_srq		   = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
170	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
171	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
172	props->max_fast_reg_page_list_len = MAX_FAST_REG_PAGES;
173	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
174	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
175		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
176	props->masked_atomic_cap   = IB_ATOMIC_HCA;
177	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
178	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
179	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
180	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
181					   props->max_mcast_grp;
182	props->max_map_per_fmr = (1 << (32 - ilog2(dev->dev->caps.num_mpts))) - 1;
183
184out:
185	kfree(in_mad);
186	kfree(out_mad);
187
188	return err;
189}
190
191static enum rdma_link_layer
192mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
193{
194	struct mlx4_dev *dev = to_mdev(device)->dev;
195
196	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
197		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
198}
199
200static void ib_link_query_port(struct ib_device *ibdev, u8 port,
201			       struct ib_port_attr *props,
202			       struct ib_smp *out_mad)
203{
204	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
205	props->lmc		= out_mad->data[34] & 0x7;
206	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
207	props->sm_sl		= out_mad->data[36] & 0xf;
208	props->state		= out_mad->data[32] & 0xf;
209	props->phys_state	= out_mad->data[33] >> 4;
210	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
211	props->gid_tbl_len	= to_mdev(ibdev)->dev->caps.gid_table_len[port];
212	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
213	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
214	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
215	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
216	props->active_width	= out_mad->data[31] & 0xf;
217	props->active_speed	= out_mad->data[35] >> 4;
218	props->max_mtu		= out_mad->data[41] & 0xf;
219	props->active_mtu	= out_mad->data[36] >> 4;
220	props->subnet_timeout	= out_mad->data[51] & 0x1f;
221	props->max_vl_num	= out_mad->data[37] >> 4;
222	props->init_type_reply	= out_mad->data[41] >> 4;
223	props->link_layer	= IB_LINK_LAYER_INFINIBAND;
224}
225
226#ifdef notyet
227static int eth_to_ib_width(int w)
228{
229	switch (w) {
230	case 4:
231		return IB_WIDTH_4X;
232	case 8:
233	case 16:
234		return IB_WIDTH_8X;
235	case 32:
236		return IB_WIDTH_12X;
237	default:
238		return IB_WIDTH_1X;
239	}
240}
241
242static int eth_to_ib_speed(int s)
243{
244	switch (s) {
245	case 256:
246		return 1;
247	case 512:
248		return 2;
249	case 1024:
250		return 4;
251	default:
252		return 1;
253	}
254}
255#endif
256
257static u8 state_to_phys_state(enum ib_port_state state)
258{
259	return state == IB_PORT_ACTIVE ? 5 : 3;
260}
261
262static int eth_link_query_port(struct ib_device *ibdev, u8 port,
263			       struct ib_port_attr *props,
264			       struct ib_smp *out_mad)
265{
266	struct mlx4_ib_iboe *iboe = &to_mdev(ibdev)->iboe;
267	struct net_device *ndev;
268	enum ib_mtu tmp;
269
270	props->active_width	= IB_WIDTH_4X;
271	props->active_speed	= 1;
272	props->port_cap_flags	= IB_PORT_CM_SUP;
273	props->gid_tbl_len	= to_mdev(ibdev)->dev->caps.gid_table_len[port];
274	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
275	props->pkey_tbl_len	= 1;
276	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
277	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
278	props->max_mtu		= IB_MTU_2048;
279	props->subnet_timeout	= 0;
280	props->max_vl_num	= out_mad->data[37] >> 4;
281	props->init_type_reply	= 0;
282	props->link_layer	= IB_LINK_LAYER_ETHERNET;
283	props->state		= IB_PORT_DOWN;
284	props->phys_state	= state_to_phys_state(props->state);
285	props->active_mtu	= IB_MTU_256;
286	spin_lock(&iboe->lock);
287	ndev = iboe->netdevs[port - 1];
288	if (!ndev)
289		goto out;
290
291#ifdef __linux__
292	tmp = iboe_get_mtu(ndev->mtu);
293#else
294	tmp = iboe_get_mtu(ndev->if_mtu);
295#endif
296	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
297	props->state		= netif_carrier_ok(ndev) &&  netif_oper_up(ndev) ?
298					IB_PORT_ACTIVE : IB_PORT_DOWN;
299	props->phys_state	= state_to_phys_state(props->state);
300
301out:
302	spin_unlock(&iboe->lock);
303	return 0;
304}
305
306static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
307			      struct ib_port_attr *props)
308{
309	struct ib_smp *in_mad  = NULL;
310	struct ib_smp *out_mad = NULL;
311	int err = -ENOMEM;
312
313	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
314	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
315	if (!in_mad || !out_mad)
316		goto out;
317
318	memset(props, 0, sizeof *props);
319
320	init_query_mad(in_mad);
321	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
322	in_mad->attr_mod = cpu_to_be32(port);
323
324	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
325	if (err)
326		goto out;
327
328	mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
329		ib_link_query_port(ibdev, port, props, out_mad) :
330		eth_link_query_port(ibdev, port, props, out_mad);
331
332out:
333	kfree(in_mad);
334	kfree(out_mad);
335
336	return err;
337}
338
339static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
340			       union ib_gid *gid)
341{
342	struct ib_smp *in_mad  = NULL;
343	struct ib_smp *out_mad = NULL;
344	int err = -ENOMEM;
345
346	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
347	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
348	if (!in_mad || !out_mad)
349		goto out;
350
351	init_query_mad(in_mad);
352	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
353	in_mad->attr_mod = cpu_to_be32(port);
354
355	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
356	if (err)
357		goto out;
358
359	memcpy(gid->raw, out_mad->data + 8, 8);
360
361	init_query_mad(in_mad);
362	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
363	in_mad->attr_mod = cpu_to_be32(index / 8);
364
365	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
366	if (err)
367		goto out;
368
369	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
370
371out:
372	kfree(in_mad);
373	kfree(out_mad);
374	return err;
375}
376
377static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
378			    union ib_gid *gid)
379{
380	struct mlx4_ib_dev *dev = to_mdev(ibdev);
381
382	*gid = dev->iboe.gid_table[port - 1][index];
383
384	return 0;
385}
386
387static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
388			     union ib_gid *gid)
389{
390	if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
391		return __mlx4_ib_query_gid(ibdev, port, index, gid);
392	else
393		return iboe_query_gid(ibdev, port, index, gid);
394}
395
396static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
397			      u16 *pkey)
398{
399	struct ib_smp *in_mad  = NULL;
400	struct ib_smp *out_mad = NULL;
401	int err = -ENOMEM;
402
403	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
404	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
405	if (!in_mad || !out_mad)
406		goto out;
407
408	init_query_mad(in_mad);
409	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
410	in_mad->attr_mod = cpu_to_be32(index / 32);
411
412	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
413	if (err)
414		goto out;
415
416	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
417
418out:
419	kfree(in_mad);
420	kfree(out_mad);
421	return err;
422}
423
424static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
425				 struct ib_device_modify *props)
426{
427	struct mlx4_cmd_mailbox *mailbox;
428	int err;
429
430	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
431		return -EOPNOTSUPP;
432
433	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
434		return 0;
435
436	spin_lock(&to_mdev(ibdev)->sm_lock);
437	memcpy(ibdev->node_desc, props->node_desc, 64);
438	spin_unlock(&to_mdev(ibdev)->sm_lock);
439
440	/* if possible, pass node desc to FW, so it can generate
441	 * a 144 trap. If cmd fails, just ignore.
442	 */
443	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
444	if (IS_ERR(mailbox))
445		return 0;
446
447	memset(mailbox->buf, 0, 256);
448	memcpy(mailbox->buf, props->node_desc, 64);
449	err = mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
450		       MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A);
451	if (err)
452		mlx4_ib_dbg("SET_NODE command failed (%d)", err);
453
454	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
455
456	return 0;
457}
458
459static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
460			 u32 cap_mask)
461{
462	struct mlx4_cmd_mailbox *mailbox;
463	int err;
464	u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
465
466	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
467	if (IS_ERR(mailbox))
468		return PTR_ERR(mailbox);
469
470	memset(mailbox->buf, 0, 256);
471
472	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
473		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
474		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
475	} else {
476		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
477		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
478	}
479
480	err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
481		       MLX4_CMD_TIME_CLASS_B);
482
483	mlx4_free_cmd_mailbox(dev->dev, mailbox);
484	return err;
485}
486
487static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
488			       struct ib_port_modify *props)
489{
490	struct ib_port_attr attr;
491	u32 cap_mask;
492	int err;
493
494	mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
495
496	err = mlx4_ib_query_port(ibdev, port, &attr);
497	if (err)
498		goto out;
499
500	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
501		~props->clr_port_cap_mask;
502
503	err = mlx4_SET_PORT(to_mdev(ibdev), port,
504			    !!(mask & IB_PORT_RESET_QKEY_CNTR),
505			    cap_mask);
506
507out:
508	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
509	return err;
510}
511
512static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
513						  struct ib_udata *udata)
514{
515	struct mlx4_ib_dev *dev = to_mdev(ibdev);
516	struct mlx4_ib_ucontext *context;
517	struct mlx4_ib_alloc_ucontext_resp resp;
518	int err;
519
520	if (!dev->ib_active)
521		return ERR_PTR(-EAGAIN);
522
523	resp.qp_tab_size      = dev->dev->caps.num_qps;
524
525	if (mlx4_wc_enabled()) {
526		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
527		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
528	} else {
529		resp.bf_reg_size      = 0;
530		resp.bf_regs_per_page = 0;
531	}
532
533	context = kzalloc(sizeof *context, GFP_KERNEL);
534	if (!context)
535		return ERR_PTR(-ENOMEM);
536
537	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
538	if (err) {
539		kfree(context);
540		return ERR_PTR(err);
541	}
542
543	INIT_LIST_HEAD(&context->db_page_list);
544	mutex_init(&context->db_page_mutex);
545
546	err = ib_copy_to_udata(udata, &resp, sizeof resp);
547	if (err) {
548		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
549		kfree(context);
550		return ERR_PTR(-EFAULT);
551	}
552
553	return &context->ibucontext;
554}
555
556static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
557{
558	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
559
560	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
561	kfree(context);
562
563	return 0;
564}
565
566static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
567{
568	struct mlx4_ib_dev *dev = to_mdev(context->device);
569
570	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
571		return -EINVAL;
572
573	if (vma->vm_pgoff == 0) {
574		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
575
576		if (io_remap_pfn_range(vma, vma->vm_start,
577				       to_mucontext(context)->uar.pfn,
578				       PAGE_SIZE, vma->vm_page_prot))
579			return -EAGAIN;
580	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
581		vma->vm_page_prot = pgprot_wc(vma->vm_page_prot);
582
583		if (io_remap_pfn_range(vma, vma->vm_start,
584				       to_mucontext(context)->uar.pfn +
585				       dev->dev->caps.num_uars,
586				       PAGE_SIZE, vma->vm_page_prot))
587			return -EAGAIN;
588	} else
589		return -EINVAL;
590
591	return 0;
592}
593
594static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
595				      struct ib_ucontext *context,
596				      struct ib_udata *udata)
597{
598	struct mlx4_ib_pd *pd;
599	int err;
600
601	pd = kzalloc(sizeof *pd, GFP_KERNEL);
602	if (!pd)
603		return ERR_PTR(-ENOMEM);
604
605	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
606	if (err) {
607		kfree(pd);
608		return ERR_PTR(err);
609	}
610
611	if (context)
612		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
613			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
614			kfree(pd);
615			return ERR_PTR(-EFAULT);
616		}
617
618	return &pd->ibpd;
619}
620
621static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
622{
623	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
624	kfree(pd);
625
626	return 0;
627}
628
629static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
630{
631	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
632	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
633	struct gid_entry *ge;
634
635	ge = kzalloc(sizeof *ge, GFP_KERNEL);
636	if (!ge)
637		return -ENOMEM;
638
639	ge->gid = *gid;
640	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
641		ge->port = mqp->port;
642		ge->added = 1;
643	}
644
645	mutex_lock(&mqp->mutex);
646	list_add_tail(&ge->list, &mqp->gid_list);
647	mutex_unlock(&mqp->mutex);
648
649	return 0;
650}
651
652int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
653		   union ib_gid *gid)
654{
655	u8 mac[6];
656	struct net_device *ndev;
657	int ret = 0;
658
659	if (!mqp->port)
660		return 0;
661	spin_lock(&mdev->iboe.lock);
662	ndev = mdev->iboe.netdevs[mqp->port - 1];
663	if (ndev)
664		dev_hold(ndev);
665	spin_unlock(&mdev->iboe.lock);
666	if (ndev) {
667		rdma_get_mcast_mac((struct in6_addr *)gid, mac);
668		rtnl_lock();
669		dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac, 6, 0);
670		ret = 1;
671		rtnl_unlock();
672		dev_put(ndev);
673	}
674
675	return ret;
676}
677
678static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
679{
680	int err;
681	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
682	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
683
684	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, !!(mqp->flags &
685				MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
686				(ibqp->qp_type == IB_QPT_RAW_ETH) ?
687				MLX4_MCAST_PROT_EN : MLX4_MCAST_PROT_IB);
688	if (err)
689		return err;
690
691	err = add_gid_entry(ibqp, gid);
692	if (err)
693		goto err_add;
694
695	return 0;
696
697err_add:
698	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
699				(ibqp->qp_type == IB_QPT_RAW_ETH) ?
700				MLX4_MCAST_PROT_EN : MLX4_MCAST_PROT_IB);
701	return err;
702}
703
704static struct gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
705{
706	struct gid_entry *ge;
707	struct gid_entry *tmp;
708	struct gid_entry *ret = NULL;
709
710	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
711		if (!memcmp(raw, ge->gid.raw, 16)) {
712			ret = ge;
713			break;
714		}
715	}
716
717	return ret;
718}
719
720static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
721{
722	int err;
723	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
724	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
725	u8 mac[6];
726	struct net_device *ndev;
727	struct gid_entry *ge;
728
729	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
730				(ibqp->qp_type == IB_QPT_RAW_ETH) ?
731				MLX4_MCAST_PROT_EN : MLX4_MCAST_PROT_IB);
732	if (err)
733		return err;
734
735	mutex_lock(&mqp->mutex);
736	ge = find_gid_entry(mqp, gid->raw);
737	if (ge) {
738		spin_lock(&mdev->iboe.lock);
739		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
740		if (ndev)
741			dev_hold(ndev);
742		spin_unlock(&mdev->iboe.lock);
743		rdma_get_mcast_mac((struct in6_addr *)gid, mac);
744		if (ndev) {
745			rtnl_lock();
746			dev_mc_delete(mdev->iboe.netdevs[ge->port - 1], mac, 6, 0);
747			rtnl_unlock();
748			dev_put(ndev);
749		}
750		list_del(&ge->list);
751		kfree(ge);
752	} else
753		printk(KERN_WARNING "could not find mgid entry\n");
754
755	mutex_unlock(&mqp->mutex);
756
757	return 0;
758}
759
760static void mlx4_dummy_comp_handler(struct ib_cq *cq, void *cq_context)
761{
762}
763
764static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
765					  struct ib_ucontext *context,
766					  struct ib_udata *udata)
767{
768	struct mlx4_ib_xrcd *xrcd;
769	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
770	struct ib_pd *pd;
771	struct ib_cq *cq;
772	int err;
773
774	if (!(mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
775		return ERR_PTR(-ENOSYS);
776
777	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
778	if (!xrcd)
779		return ERR_PTR(-ENOMEM);
780
781	err = mlx4_xrcd_alloc(mdev->dev, &xrcd->xrcdn);
782	if (err)
783		goto err_xrcd;
784
785	pd = mlx4_ib_alloc_pd(ibdev, NULL, NULL);
786	if (IS_ERR(pd)) {
787		err = PTR_ERR(pd);
788		goto err_pd;
789	}
790	pd->device  = ibdev;
791
792	cq = mlx4_ib_create_cq(ibdev, 1, 0, NULL, NULL);
793	if (IS_ERR(cq)) {
794		err = PTR_ERR(cq);
795		goto err_cq;
796	}
797	cq->device        = ibdev;
798	cq->comp_handler  = mlx4_dummy_comp_handler;
799
800	if (context)
801		if (ib_copy_to_udata(udata, &xrcd->xrcdn, sizeof(__u32))) {
802			err = -EFAULT;
803			goto err_copy;
804		}
805
806	xrcd->cq = cq;
807	xrcd->pd = pd;
808	return &xrcd->ibxrcd;
809
810err_copy:
811	mlx4_ib_destroy_cq(cq);
812err_cq:
813	mlx4_ib_dealloc_pd(pd);
814err_pd:
815	mlx4_xrcd_free(mdev->dev, xrcd->xrcdn);
816err_xrcd:
817	kfree(xrcd);
818	return ERR_PTR(err);
819}
820
821static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
822{
823	struct mlx4_ib_xrcd *mxrcd = to_mxrcd(xrcd);
824
825	mlx4_ib_destroy_cq(mxrcd->cq);
826	mlx4_ib_dealloc_pd(mxrcd->pd);
827	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
828	kfree(xrcd);
829
830	return 0;
831}
832
833
834static int init_node_data(struct mlx4_ib_dev *dev)
835{
836	struct ib_smp *in_mad  = NULL;
837	struct ib_smp *out_mad = NULL;
838	int err = -ENOMEM;
839
840	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
841	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
842	if (!in_mad || !out_mad)
843		goto out;
844
845	init_query_mad(in_mad);
846	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
847
848	err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
849	if (err)
850		goto out;
851
852	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
853
854	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
855
856	err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
857	if (err)
858		goto out;
859
860	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
861	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
862
863out:
864	kfree(in_mad);
865	kfree(out_mad);
866	return err;
867}
868
869static ssize_t show_hca(struct device *device, struct device_attribute *attr,
870			char *buf)
871{
872	struct mlx4_ib_dev *dev =
873		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
874	return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
875}
876
877static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
878			   char *buf)
879{
880	struct mlx4_ib_dev *dev =
881		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
882	return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
883		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
884		       (int) dev->dev->caps.fw_ver & 0xffff);
885}
886
887static ssize_t show_rev(struct device *device, struct device_attribute *attr,
888			char *buf)
889{
890	struct mlx4_ib_dev *dev =
891		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
892	return sprintf(buf, "%x\n", dev->dev->rev_id);
893}
894
895static ssize_t show_board(struct device *device, struct device_attribute *attr,
896			  char *buf)
897{
898	struct mlx4_ib_dev *dev =
899		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
900	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
901		       dev->dev->board_id);
902}
903
904static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
905static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
906static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
907static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
908
909static struct device_attribute *mlx4_class_attributes[] = {
910	&dev_attr_hw_rev,
911	&dev_attr_fw_ver,
912	&dev_attr_hca_type,
913	&dev_attr_board_id
914};
915
916/*
917 * create show function and a device_attribute struct pointing to
918 * the function for _name
919 */
920#define DEVICE_DIAG_RPRT_ATTR(_name, _offset, _op_mod)		\
921static ssize_t show_rprt_##_name(struct device *dev,		\
922				 struct device_attribute *attr,	\
923				 char *buf){			\
924	return show_diag_rprt(dev, buf, _offset, _op_mod);	\
925}								\
926static DEVICE_ATTR(_name, S_IRUGO, show_rprt_##_name, NULL);
927
928#define MLX4_DIAG_RPRT_CLEAR_DIAGS 3
929
930static size_t show_diag_rprt(struct device *device, char *buf,
931                              u32 offset, u8 op_modifier)
932{
933	size_t ret;
934	u32 counter_offset = offset;
935	u32 diag_counter = 0;
936	struct mlx4_ib_dev *dev = container_of(device, struct mlx4_ib_dev,
937					       ib_dev.dev);
938
939	ret = mlx4_query_diag_counters(dev->dev, 1, op_modifier,
940				       &counter_offset, &diag_counter);
941	if (ret)
942		return ret;
943
944	return sprintf(buf,"%d\n", diag_counter);
945}
946
947static ssize_t clear_diag_counters(struct device *device,
948				   struct device_attribute *attr,
949				   const char *buf, size_t length)
950{
951	size_t ret;
952	struct mlx4_ib_dev *dev = container_of(device, struct mlx4_ib_dev,
953					       ib_dev.dev);
954
955	ret = mlx4_query_diag_counters(dev->dev, 0, MLX4_DIAG_RPRT_CLEAR_DIAGS,
956				       NULL, NULL);
957	if (ret)
958		return ret;
959
960	return length;
961}
962
963DEVICE_DIAG_RPRT_ATTR(rq_num_lle	, 0x00, 2);
964DEVICE_DIAG_RPRT_ATTR(sq_num_lle	, 0x04, 2);
965DEVICE_DIAG_RPRT_ATTR(rq_num_lqpoe	, 0x08, 2);
966DEVICE_DIAG_RPRT_ATTR(sq_num_lqpoe 	, 0x0C, 2);
967DEVICE_DIAG_RPRT_ATTR(rq_num_leeoe	, 0x10, 2);
968DEVICE_DIAG_RPRT_ATTR(sq_num_leeoe	, 0x14, 2);
969DEVICE_DIAG_RPRT_ATTR(rq_num_lpe	, 0x18, 2);
970DEVICE_DIAG_RPRT_ATTR(sq_num_lpe	, 0x1C, 2);
971DEVICE_DIAG_RPRT_ATTR(rq_num_wrfe	, 0x20, 2);
972DEVICE_DIAG_RPRT_ATTR(sq_num_wrfe	, 0x24, 2);
973DEVICE_DIAG_RPRT_ATTR(sq_num_mwbe	, 0x2C, 2);
974DEVICE_DIAG_RPRT_ATTR(sq_num_bre	, 0x34, 2);
975DEVICE_DIAG_RPRT_ATTR(rq_num_lae	, 0x38, 2);
976DEVICE_DIAG_RPRT_ATTR(sq_num_rire	, 0x44, 2);
977DEVICE_DIAG_RPRT_ATTR(rq_num_rire	, 0x48, 2);
978DEVICE_DIAG_RPRT_ATTR(sq_num_rae	, 0x4C, 2);
979DEVICE_DIAG_RPRT_ATTR(rq_num_rae	, 0x50, 2);
980DEVICE_DIAG_RPRT_ATTR(sq_num_roe	, 0x54, 2);
981DEVICE_DIAG_RPRT_ATTR(sq_num_tree	, 0x5C, 2);
982DEVICE_DIAG_RPRT_ATTR(sq_num_rree	, 0x64, 2);
983DEVICE_DIAG_RPRT_ATTR(rq_num_rnr	, 0x68, 2);
984DEVICE_DIAG_RPRT_ATTR(sq_num_rnr	, 0x6C, 2);
985DEVICE_DIAG_RPRT_ATTR(sq_num_rabrte	, 0x7C, 2);
986DEVICE_DIAG_RPRT_ATTR(sq_num_ieecne	, 0x84, 2);
987DEVICE_DIAG_RPRT_ATTR(sq_num_ieecse	, 0x8C, 2);
988DEVICE_DIAG_RPRT_ATTR(rq_num_oos	, 0x100, 2);
989DEVICE_DIAG_RPRT_ATTR(sq_num_oos	, 0x104, 2);
990DEVICE_DIAG_RPRT_ATTR(rq_num_mce	, 0x108, 2);
991DEVICE_DIAG_RPRT_ATTR(rq_num_rsync	, 0x110, 2);
992DEVICE_DIAG_RPRT_ATTR(sq_num_rsync	, 0x114, 2);
993DEVICE_DIAG_RPRT_ATTR(rq_num_udsdprd	, 0x118, 2);
994DEVICE_DIAG_RPRT_ATTR(rq_num_ucsdprd	, 0x120, 2);
995DEVICE_DIAG_RPRT_ATTR(num_cqovf		, 0x1A0, 2);
996DEVICE_DIAG_RPRT_ATTR(num_eqovf		, 0x1A4, 2);
997DEVICE_DIAG_RPRT_ATTR(num_baddb		, 0x1A8, 2);
998
999static DEVICE_ATTR(clear_diag, S_IWUGO, NULL, clear_diag_counters);
1000
1001static struct attribute *diag_rprt_attrs[] = {
1002	&dev_attr_rq_num_lle.attr,
1003	&dev_attr_sq_num_lle.attr,
1004	&dev_attr_rq_num_lqpoe.attr,
1005	&dev_attr_sq_num_lqpoe.attr,
1006	&dev_attr_rq_num_leeoe.attr,
1007	&dev_attr_sq_num_leeoe.attr,
1008	&dev_attr_rq_num_lpe.attr,
1009	&dev_attr_sq_num_lpe.attr,
1010	&dev_attr_rq_num_wrfe.attr,
1011	&dev_attr_sq_num_wrfe.attr,
1012	&dev_attr_sq_num_mwbe.attr,
1013	&dev_attr_sq_num_bre.attr,
1014	&dev_attr_rq_num_lae.attr,
1015	&dev_attr_sq_num_rire.attr,
1016	&dev_attr_rq_num_rire.attr,
1017	&dev_attr_sq_num_rae.attr,
1018	&dev_attr_rq_num_rae.attr,
1019	&dev_attr_sq_num_roe.attr,
1020	&dev_attr_sq_num_tree.attr,
1021	&dev_attr_sq_num_rree.attr,
1022	&dev_attr_rq_num_rnr.attr,
1023	&dev_attr_sq_num_rnr.attr,
1024	&dev_attr_sq_num_rabrte.attr,
1025	&dev_attr_sq_num_ieecne.attr,
1026	&dev_attr_sq_num_ieecse.attr,
1027	&dev_attr_rq_num_oos.attr,
1028	&dev_attr_sq_num_oos.attr,
1029	&dev_attr_rq_num_mce.attr,
1030	&dev_attr_rq_num_rsync.attr,
1031	&dev_attr_sq_num_rsync.attr,
1032	&dev_attr_rq_num_udsdprd.attr,
1033	&dev_attr_rq_num_ucsdprd.attr,
1034	&dev_attr_num_cqovf.attr,
1035	&dev_attr_num_eqovf.attr,
1036	&dev_attr_num_baddb.attr,
1037	&dev_attr_clear_diag.attr,
1038	NULL
1039};
1040
1041struct attribute_group diag_counters_group = {
1042	.name  = "diag_counters",
1043	.attrs  = diag_rprt_attrs
1044};
1045
1046static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
1047{
1048#ifdef __linux__
1049	memcpy(eui, dev->dev_addr, 3);
1050	memcpy(eui + 5, dev->dev_addr + 3, 3);
1051#else
1052	memcpy(eui, IF_LLADDR(dev), 3);
1053	memcpy(eui + 5, IF_LLADDR(dev) + 3, 3);
1054#endif
1055	if (vlan_id < 0x1000) {
1056		eui[3] = vlan_id >> 8;
1057		eui[4] = vlan_id & 0xff;
1058	} else {
1059		eui[3] = 0xff;
1060		eui[4] = 0xfe;
1061	}
1062	eui[0] ^= 2;
1063}
1064
1065static void update_gids_task(struct work_struct *work)
1066{
1067	struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1068	struct mlx4_cmd_mailbox *mailbox;
1069	union ib_gid *gids;
1070	int err;
1071	struct mlx4_dev	*dev = gw->dev->dev;
1072	struct ib_event event;
1073
1074	mailbox = mlx4_alloc_cmd_mailbox(dev);
1075	if (IS_ERR(mailbox)) {
1076		printk(KERN_WARNING "update gid table failed %ld\n", PTR_ERR(mailbox));
1077		return;
1078	}
1079
1080	gids = mailbox->buf;
1081	memcpy(gids, gw->gids, sizeof gw->gids);
1082
1083	err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1084		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B);
1085	if (err)
1086		printk(KERN_WARNING "set port command failed\n");
1087	else {
1088		memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
1089		event.device = &gw->dev->ib_dev;
1090		event.element.port_num = gw->port;
1091		event.event    = IB_EVENT_GID_CHANGE;
1092		ib_dispatch_event(&event);
1093	}
1094
1095	mlx4_free_cmd_mailbox(dev, mailbox);
1096	kfree(gw);
1097}
1098
1099enum {
1100	MLX4_MAX_EFF_VLANS = 128 - MLX4_VLAN_REGULAR,
1101};
1102
1103static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
1104{
1105	struct net_device *ndev = dev->iboe.netdevs[port - 1];
1106	struct update_gid_work *work;
1107	struct net_device *tmp;
1108	int i;
1109	u8 *hits;
1110	int ret;
1111	union ib_gid gid;
1112	int tofree;
1113	int found;
1114	int need_update = 0;
1115	u16 vid;
1116
1117	work = kzalloc(sizeof *work, GFP_ATOMIC);
1118	if (!work)
1119		return -ENOMEM;
1120
1121	hits = kzalloc(MLX4_MAX_EFF_VLANS + 1, GFP_ATOMIC);
1122	if (!hits) {
1123		ret = -ENOMEM;
1124		goto out;
1125	}
1126
1127#ifdef __linux__
1128	read_lock(&dev_base_lock);
1129	for_each_netdev(&init_net, tmp) {
1130#else
1131	IFNET_RLOCK();
1132	TAILQ_FOREACH(tmp, &V_ifnet, if_link) {
1133#endif
1134		if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
1135			gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1136			vid = rdma_vlan_dev_vlan_id(tmp);
1137			mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
1138			found = 0;
1139			tofree = -1;
1140			for (i = 0; i < MLX4_MAX_EFF_VLANS + 1; ++i) {
1141				if (tofree < 0 &&
1142				    !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1143					tofree = i;
1144				if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
1145					hits[i] = 1;
1146					found = 1;
1147					break;
1148				}
1149			}
1150
1151			if (!found) {
1152				if (tmp == ndev  && (memcmp(&dev->iboe.gid_table[port - 1][0], &gid, sizeof gid) || !memcmp(&dev->iboe.gid_table[port - 1][0], &zgid, sizeof gid))) {
1153					dev->iboe.gid_table[port - 1][0] = gid;
1154					++need_update;
1155					hits[0] = 1;
1156				} else if (tofree >= 0) {
1157					dev->iboe.gid_table[port - 1][tofree] = gid;
1158					hits[tofree] = 1;
1159					++need_update;
1160				}
1161			}
1162		}
1163#ifdef __linux__
1164	}
1165	read_unlock(&dev_base_lock);
1166#else
1167	}
1168	IFNET_RUNLOCK();
1169#endif
1170
1171	for (i = 0; i < MLX4_MAX_EFF_VLANS + 1; ++i)
1172		if (!hits[i]) {
1173			if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1174				++need_update;
1175			dev->iboe.gid_table[port - 1][i] = zgid;
1176		}
1177
1178
1179	if (need_update) {
1180		memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
1181		INIT_WORK(&work->work, update_gids_task);
1182		work->port = port;
1183		work->dev = dev;
1184		queue_work(wq, &work->work);
1185	} else
1186		kfree(work);
1187
1188	kfree(hits);
1189	return 0;
1190
1191out:
1192	kfree(work);
1193	return ret;
1194}
1195
1196static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
1197{
1198	switch (event) {
1199	case NETDEV_UP:
1200#ifdef __linux__
1201	case NETDEV_CHANGEADDR:
1202#endif
1203		update_ipv6_gids(dev, port, 0);
1204		break;
1205
1206	case NETDEV_DOWN:
1207		update_ipv6_gids(dev, port, 1);
1208		dev->iboe.netdevs[port - 1] = NULL;
1209	}
1210}
1211
1212static void netdev_added(struct mlx4_ib_dev *dev, int port)
1213{
1214	update_ipv6_gids(dev, port, 0);
1215}
1216
1217static void netdev_removed(struct mlx4_ib_dev *dev, int port)
1218{
1219	update_ipv6_gids(dev, port, 1);
1220}
1221
1222static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1223				void *ptr)
1224{
1225	struct net_device *dev = ptr;
1226	struct mlx4_ib_dev *ibdev;
1227	struct net_device *oldnd;
1228	struct mlx4_ib_iboe *iboe;
1229	int port;
1230
1231#ifdef __linux__
1232	if (!net_eq(dev_net(dev), &init_net))
1233		return NOTIFY_DONE;
1234#endif
1235
1236	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1237	iboe = &ibdev->iboe;
1238
1239	spin_lock(&iboe->lock);
1240	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1241		oldnd = iboe->netdevs[port - 1];
1242		iboe->netdevs[port - 1] = mlx4_get_prot_dev(ibdev->dev, MLX4_PROT_EN, port);
1243		if (oldnd != iboe->netdevs[port - 1]) {
1244			if (iboe->netdevs[port - 1])
1245				netdev_added(ibdev, port);
1246			else
1247				netdev_removed(ibdev, port);
1248		}
1249	}
1250
1251	if (dev == iboe->netdevs[0] ||
1252	    (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
1253		handle_en_event(ibdev, 1, event);
1254	else if (dev == iboe->netdevs[1]
1255		 || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
1256		handle_en_event(ibdev, 2, event);
1257
1258	spin_unlock(&iboe->lock);
1259
1260	return NOTIFY_DONE;
1261}
1262
1263static void *mlx4_ib_add(struct mlx4_dev *dev)
1264{
1265	static int mlx4_ib_version_printed;
1266	struct mlx4_ib_dev *ibdev;
1267	int num_ports = 0;
1268	int i;
1269	int err;
1270	struct mlx4_ib_iboe *iboe;
1271	int k;
1272
1273	if (!mlx4_ib_version_printed) {
1274		printk(KERN_INFO "%s", mlx4_ib_version);
1275		++mlx4_ib_version_printed;
1276	}
1277
1278	mlx4_foreach_ib_transport_port(i, dev)
1279		num_ports++;
1280
1281	/* No point in registering a device with no ports... */
1282	if (num_ports == 0)
1283		return NULL;
1284
1285	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1286	if (!ibdev) {
1287		dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1288		return NULL;
1289	}
1290
1291	iboe = &ibdev->iboe;
1292
1293	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1294		goto err_dealloc;
1295
1296	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1297		goto err_pd;
1298
1299	ibdev->priv_uar.map = ioremap(ibdev->priv_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
1300	if (!ibdev->priv_uar.map)
1301		goto err_uar;
1302	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1303
1304	ibdev->dev = dev;
1305
1306	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1307	ibdev->ib_dev.owner		= THIS_MODULE;
1308	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
1309	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
1310	ibdev->num_ports		= num_ports;
1311	ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
1312	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
1313	ibdev->ib_dev.dma_device	= &dev->pdev->dev;
1314
1315	ibdev->ib_dev.uverbs_abi_ver	= MLX4_IB_UVERBS_ABI_VERSION;
1316	ibdev->ib_dev.uverbs_cmd_mask	=
1317		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
1318		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
1319		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
1320		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
1321		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
1322		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
1323		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
1324		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
1325		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
1326		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
1327		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
1328		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
1329		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
1330		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
1331		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
1332		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
1333		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
1334		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
1335		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
1336		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
1337		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
1338
1339	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
1340	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
1341	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
1342	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
1343	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
1344	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
1345	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
1346	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
1347	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
1348	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
1349	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
1350	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
1351	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
1352	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
1353	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
1354	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
1355	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
1356	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
1357	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
1358	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
1359	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
1360	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
1361	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
1362	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
1363	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
1364	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
1365	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
1366	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
1367	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
1368	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
1369	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
1370	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
1371	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
1372	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
1373	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
1374	ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1375	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1376	ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
1377	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
1378	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
1379	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
1380
1381	ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
1382	ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
1383	ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
1384	ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
1385	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1386		ibdev->ib_dev.create_xrc_srq = mlx4_ib_create_xrc_srq;
1387		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1388		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1389		ibdev->ib_dev.create_xrc_rcv_qp = mlx4_ib_create_xrc_rcv_qp;
1390		ibdev->ib_dev.modify_xrc_rcv_qp = mlx4_ib_modify_xrc_rcv_qp;
1391		ibdev->ib_dev.query_xrc_rcv_qp = mlx4_ib_query_xrc_rcv_qp;
1392		ibdev->ib_dev.reg_xrc_rcv_qp = mlx4_ib_reg_xrc_rcv_qp;
1393		ibdev->ib_dev.unreg_xrc_rcv_qp = mlx4_ib_unreg_xrc_rcv_qp;
1394		ibdev->ib_dev.uverbs_cmd_mask |=
1395			(1ull << IB_USER_VERBS_CMD_CREATE_XRC_SRQ)	|
1396			(1ull << IB_USER_VERBS_CMD_OPEN_XRC_DOMAIN)	|
1397			(1ull << IB_USER_VERBS_CMD_CLOSE_XRC_DOMAIN)	|
1398			(1ull << IB_USER_VERBS_CMD_CREATE_XRC_RCV_QP)	|
1399			(1ull << IB_USER_VERBS_CMD_MODIFY_XRC_RCV_QP)	|
1400			(1ull << IB_USER_VERBS_CMD_QUERY_XRC_RCV_QP)	|
1401			(1ull << IB_USER_VERBS_CMD_REG_XRC_RCV_QP)	|
1402			(1ull << IB_USER_VERBS_CMD_UNREG_XRC_RCV_QP);
1403	}
1404
1405
1406	spin_lock_init(&iboe->lock);
1407	if (init_node_data(ibdev))
1408		goto err_map;
1409
1410	for (k = 0; k < ibdev->num_ports; ++k) {
1411		err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[k]);
1412		if (err)
1413			ibdev->counters[k] = -1;
1414		else
1415			mlx4_set_iboe_counter(dev, ibdev->counters[k], k + 1);
1416	}
1417
1418	spin_lock_init(&ibdev->sm_lock);
1419	mutex_init(&ibdev->cap_mask_mutex);
1420	mutex_init(&ibdev->xrc_reg_mutex);
1421
1422	if (ib_register_device(&ibdev->ib_dev))
1423		goto err_counter;
1424
1425	if (mlx4_ib_mad_init(ibdev))
1426		goto err_reg;
1427	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1428		iboe->nb.notifier_call = mlx4_ib_netdev_event;
1429		err = register_netdevice_notifier(&iboe->nb);
1430		if (err)
1431			goto err_reg;
1432	}
1433	for (i = 0; i < ARRAY_SIZE(mlx4_class_attributes); ++i) {
1434		if (device_create_file(&ibdev->ib_dev.dev,
1435				       mlx4_class_attributes[i]))
1436			goto err_notif;
1437	}
1438
1439	if(sysfs_create_group(&ibdev->ib_dev.dev.kobj, &diag_counters_group))
1440		goto err_notif;
1441
1442	ibdev->ib_active = 1;
1443
1444	return ibdev;
1445
1446err_notif:
1447	if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1448		printk(KERN_WARNING "failure unregistering notifier\n");
1449	flush_workqueue(wq);
1450
1451err_reg:
1452	ib_unregister_device(&ibdev->ib_dev);
1453
1454err_counter:
1455	for (; k; --k)
1456		mlx4_counter_free(ibdev->dev, ibdev->counters[k - 1]);
1457
1458err_map:
1459	iounmap(ibdev->priv_uar.map);
1460
1461err_uar:
1462	mlx4_uar_free(dev, &ibdev->priv_uar);
1463
1464err_pd:
1465	mlx4_pd_free(dev, ibdev->priv_pdn);
1466
1467err_dealloc:
1468	ib_dealloc_device(&ibdev->ib_dev);
1469
1470	return NULL;
1471}
1472
1473static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1474{
1475	struct mlx4_ib_dev *ibdev = ibdev_ptr;
1476	int p;
1477	int k;
1478
1479	sysfs_remove_group(&ibdev->ib_dev.dev.kobj, &diag_counters_group);
1480
1481	mlx4_ib_mad_cleanup(ibdev);
1482	ib_unregister_device(&ibdev->ib_dev);
1483	for (k = 0; k < ibdev->num_ports; ++k)
1484		mlx4_counter_free(ibdev->dev, ibdev->counters[k]);
1485
1486	if (ibdev->iboe.nb.notifier_call) {
1487		unregister_netdevice_notifier(&ibdev->iboe.nb);
1488		flush_workqueue(wq);
1489		ibdev->iboe.nb.notifier_call = NULL;
1490	}
1491	iounmap(ibdev->priv_uar.map);
1492
1493	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
1494		mlx4_CLOSE_PORT(dev, p);
1495
1496	mlx4_uar_free(dev, &ibdev->priv_uar);
1497	mlx4_pd_free(dev, ibdev->priv_pdn);
1498	ib_dealloc_device(&ibdev->ib_dev);
1499}
1500
1501static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
1502			  enum mlx4_dev_event event, int port)
1503{
1504	struct ib_event ibev;
1505	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
1506
1507	if (port > ibdev->num_ports)
1508		return;
1509
1510	switch (event) {
1511	case MLX4_DEV_EVENT_PORT_UP:
1512		ibev.event = IB_EVENT_PORT_ACTIVE;
1513		break;
1514
1515	case MLX4_DEV_EVENT_PORT_DOWN:
1516		ibev.event = IB_EVENT_PORT_ERR;
1517		break;
1518
1519	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
1520		ibdev->ib_active = 0;
1521		ibev.event = IB_EVENT_DEVICE_FATAL;
1522		break;
1523
1524	default:
1525		return;
1526	}
1527
1528	ibev.device	      = ibdev_ptr;
1529	ibev.element.port_num = port;
1530
1531	ib_dispatch_event(&ibev);
1532}
1533
1534static struct mlx4_interface mlx4_ib_interface = {
1535	.add	= mlx4_ib_add,
1536	.remove	= mlx4_ib_remove,
1537       .event  = mlx4_ib_event,
1538       .get_prot_dev = get_ibdev,
1539       .protocol     = MLX4_PROT_IB,
1540};
1541
1542static int __init mlx4_ib_init(void)
1543{
1544	int err;
1545
1546	wq = create_singlethread_workqueue("mlx4_ib");
1547	if (!wq)
1548		return -ENOMEM;
1549
1550	err = mlx4_register_interface(&mlx4_ib_interface);
1551	if (err) {
1552		destroy_workqueue(wq);
1553		return err;
1554	}
1555
1556	return 0;
1557}
1558
1559static void __exit mlx4_ib_cleanup(void)
1560{
1561	mlx4_unregister_interface(&mlx4_ib_interface);
1562	destroy_workqueue(wq);
1563}
1564
1565module_init_order(mlx4_ib_init, SI_ORDER_MIDDLE);
1566module_exit(mlx4_ib_cleanup);
1567
1568#undef MODULE_VERSION
1569#include <sys/module.h>
1570static int
1571mlx4ib_evhand(module_t mod, int event, void *arg)
1572{
1573        return (0);
1574}
1575static moduledata_t mlx4ib_mod = {
1576        .name = "mlx4ib",
1577        .evhand = mlx4ib_evhand,
1578};
1579DECLARE_MODULE(mlx4ib, mlx4ib_mod, SI_SUB_SMP, SI_ORDER_ANY);
1580MODULE_DEPEND(mlx4ib, mlx4, 1, 1, 1);
1581