[go: nahoru, domu]

1/*
2 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses.  You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 *     Redistribution and use in source and binary forms, with or
12 *     without modification, are permitted provided that the following
13 *     conditions are met:
14 *
15 *      - Redistributions of source code must retain the above
16 *        copyright notice, this list of conditions and the following
17 *        disclaimer.
18 *
19 *      - Redistributions in binary form must reproduce the above
20 *        copyright notice, this list of conditions and the following
21 *        disclaimer in the documentation and/or other materials
22 *        provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33
34#include <linux/module.h>
35#include <linux/init.h>
36#include <linux/slab.h>
37#include <linux/errno.h>
38#include <linux/netdevice.h>
39#include <linux/inetdevice.h>
40#include <linux/rtnetlink.h>
41#include <linux/if_vlan.h>
42#include <net/ipv6.h>
43#include <net/addrconf.h>
44
45#include <rdma/ib_smi.h>
46#include <rdma/ib_user_verbs.h>
47#include <rdma/ib_addr.h>
48
49#include <linux/mlx4/driver.h>
50#include <linux/mlx4/cmd.h>
51#include <linux/mlx4/qp.h>
52
53#include "mlx4_ib.h"
54#include "user.h"
55
56#define DRV_NAME	MLX4_IB_DRV_NAME
57#define DRV_VERSION	"2.2-1"
58#define DRV_RELDATE	"Feb 2014"
59
60#define MLX4_IB_FLOW_MAX_PRIO 0xFFF
61#define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
62#define MLX4_IB_CARD_REV_A0   0xA0
63
64MODULE_AUTHOR("Roland Dreier");
65MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
66MODULE_LICENSE("Dual BSD/GPL");
67MODULE_VERSION(DRV_VERSION);
68
69int mlx4_ib_sm_guid_assign = 1;
70module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
71MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
72
73static const char mlx4_ib_version[] =
74	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
75	DRV_VERSION " (" DRV_RELDATE ")\n";
76
77struct update_gid_work {
78	struct work_struct	work;
79	union ib_gid		gids[128];
80	struct mlx4_ib_dev     *dev;
81	int			port;
82};
83
84static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
85
86static struct workqueue_struct *wq;
87
88static void init_query_mad(struct ib_smp *mad)
89{
90	mad->base_version  = 1;
91	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
92	mad->class_version = 1;
93	mad->method	   = IB_MGMT_METHOD_GET;
94}
95
96static union ib_gid zgid;
97
98static int check_flow_steering_support(struct mlx4_dev *dev)
99{
100	int eth_num_ports = 0;
101	int ib_num_ports = 0;
102
103	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
104
105	if (dmfs) {
106		int i;
107		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
108			eth_num_ports++;
109		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
110			ib_num_ports++;
111		dmfs &= (!ib_num_ports ||
112			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
113			(!eth_num_ports ||
114			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
115		if (ib_num_ports && mlx4_is_mfunc(dev)) {
116			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
117			dmfs = 0;
118		}
119	}
120	return dmfs;
121}
122
123static int num_ib_ports(struct mlx4_dev *dev)
124{
125	int ib_ports = 0;
126	int i;
127
128	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
129		ib_ports++;
130
131	return ib_ports;
132}
133
134static int mlx4_ib_query_device(struct ib_device *ibdev,
135				struct ib_device_attr *props)
136{
137	struct mlx4_ib_dev *dev = to_mdev(ibdev);
138	struct ib_smp *in_mad  = NULL;
139	struct ib_smp *out_mad = NULL;
140	int err = -ENOMEM;
141	int have_ib_ports;
142
143	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
144	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
145	if (!in_mad || !out_mad)
146		goto out;
147
148	init_query_mad(in_mad);
149	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
150
151	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
152			   1, NULL, NULL, in_mad, out_mad);
153	if (err)
154		goto out;
155
156	memset(props, 0, sizeof *props);
157
158	have_ib_ports = num_ib_ports(dev->dev);
159
160	props->fw_ver = dev->dev->caps.fw_ver;
161	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
162		IB_DEVICE_PORT_ACTIVE_EVENT		|
163		IB_DEVICE_SYS_IMAGE_GUID		|
164		IB_DEVICE_RC_RNR_NAK_GEN		|
165		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
166	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
167		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
168	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
169		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
170	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
171		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
172	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
173		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
174	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
175		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
176	if (dev->dev->caps.max_gso_sz &&
177	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
178	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
179		props->device_cap_flags |= IB_DEVICE_UD_TSO;
180	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
181		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
182	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
183	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
184	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
185		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
186	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
187		props->device_cap_flags |= IB_DEVICE_XRC;
188	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
189		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
190	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
191		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
192			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
193		else
194			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
195	if (dev->steering_support ==  MLX4_STEERING_MODE_DEVICE_MANAGED)
196		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
197	}
198
199	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
200		0xffffff;
201	props->vendor_part_id	   = dev->dev->pdev->device;
202	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
203	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
204
205	props->max_mr_size	   = ~0ull;
206	props->page_size_cap	   = dev->dev->caps.page_size_cap;
207	props->max_qp		   = dev->dev->quotas.qp;
208	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
209	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
210					 dev->dev->caps.max_rq_sg);
211	props->max_cq		   = dev->dev->quotas.cq;
212	props->max_cqe		   = dev->dev->caps.max_cqes;
213	props->max_mr		   = dev->dev->quotas.mpt;
214	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
215	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
216	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
217	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
218	props->max_srq		   = dev->dev->quotas.srq;
219	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
220	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
221	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
222	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
223	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
224		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
225	props->masked_atomic_cap   = props->atomic_cap;
226	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
227	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
228	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
229	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
230					   props->max_mcast_grp;
231	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
232
233out:
234	kfree(in_mad);
235	kfree(out_mad);
236
237	return err;
238}
239
240static enum rdma_link_layer
241mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
242{
243	struct mlx4_dev *dev = to_mdev(device)->dev;
244
245	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
246		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
247}
248
249static int ib_link_query_port(struct ib_device *ibdev, u8 port,
250			      struct ib_port_attr *props, int netw_view)
251{
252	struct ib_smp *in_mad  = NULL;
253	struct ib_smp *out_mad = NULL;
254	int ext_active_speed;
255	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
256	int err = -ENOMEM;
257
258	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
259	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
260	if (!in_mad || !out_mad)
261		goto out;
262
263	init_query_mad(in_mad);
264	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
265	in_mad->attr_mod = cpu_to_be32(port);
266
267	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
268		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
269
270	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
271				in_mad, out_mad);
272	if (err)
273		goto out;
274
275
276	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
277	props->lmc		= out_mad->data[34] & 0x7;
278	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
279	props->sm_sl		= out_mad->data[36] & 0xf;
280	props->state		= out_mad->data[32] & 0xf;
281	props->phys_state	= out_mad->data[33] >> 4;
282	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
283	if (netw_view)
284		props->gid_tbl_len = out_mad->data[50];
285	else
286		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
287	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
288	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
289	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
290	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
291	props->active_width	= out_mad->data[31] & 0xf;
292	props->active_speed	= out_mad->data[35] >> 4;
293	props->max_mtu		= out_mad->data[41] & 0xf;
294	props->active_mtu	= out_mad->data[36] >> 4;
295	props->subnet_timeout	= out_mad->data[51] & 0x1f;
296	props->max_vl_num	= out_mad->data[37] >> 4;
297	props->init_type_reply	= out_mad->data[41] >> 4;
298
299	/* Check if extended speeds (EDR/FDR/...) are supported */
300	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
301		ext_active_speed = out_mad->data[62] >> 4;
302
303		switch (ext_active_speed) {
304		case 1:
305			props->active_speed = IB_SPEED_FDR;
306			break;
307		case 2:
308			props->active_speed = IB_SPEED_EDR;
309			break;
310		}
311	}
312
313	/* If reported active speed is QDR, check if is FDR-10 */
314	if (props->active_speed == IB_SPEED_QDR) {
315		init_query_mad(in_mad);
316		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
317		in_mad->attr_mod = cpu_to_be32(port);
318
319		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
320				   NULL, NULL, in_mad, out_mad);
321		if (err)
322			goto out;
323
324		/* Checking LinkSpeedActive for FDR-10 */
325		if (out_mad->data[15] & 0x1)
326			props->active_speed = IB_SPEED_FDR10;
327	}
328
329	/* Avoid wrong speed value returned by FW if the IB link is down. */
330	if (props->state == IB_PORT_DOWN)
331		 props->active_speed = IB_SPEED_SDR;
332
333out:
334	kfree(in_mad);
335	kfree(out_mad);
336	return err;
337}
338
339static u8 state_to_phys_state(enum ib_port_state state)
340{
341	return state == IB_PORT_ACTIVE ? 5 : 3;
342}
343
344static int eth_link_query_port(struct ib_device *ibdev, u8 port,
345			       struct ib_port_attr *props, int netw_view)
346{
347
348	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
349	struct mlx4_ib_iboe *iboe = &mdev->iboe;
350	struct net_device *ndev;
351	enum ib_mtu tmp;
352	struct mlx4_cmd_mailbox *mailbox;
353	int err = 0;
354
355	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
356	if (IS_ERR(mailbox))
357		return PTR_ERR(mailbox);
358
359	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
360			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
361			   MLX4_CMD_WRAPPED);
362	if (err)
363		goto out;
364
365	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
366						IB_WIDTH_4X : IB_WIDTH_1X;
367	props->active_speed	= IB_SPEED_QDR;
368	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
369	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
370	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
371	props->pkey_tbl_len	= 1;
372	props->max_mtu		= IB_MTU_4096;
373	props->max_vl_num	= 2;
374	props->state		= IB_PORT_DOWN;
375	props->phys_state	= state_to_phys_state(props->state);
376	props->active_mtu	= IB_MTU_256;
377	spin_lock_bh(&iboe->lock);
378	ndev = iboe->netdevs[port - 1];
379	if (!ndev)
380		goto out_unlock;
381
382	tmp = iboe_get_mtu(ndev->mtu);
383	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
384
385	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
386					IB_PORT_ACTIVE : IB_PORT_DOWN;
387	props->phys_state	= state_to_phys_state(props->state);
388out_unlock:
389	spin_unlock_bh(&iboe->lock);
390out:
391	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
392	return err;
393}
394
395int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
396			 struct ib_port_attr *props, int netw_view)
397{
398	int err;
399
400	memset(props, 0, sizeof *props);
401
402	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
403		ib_link_query_port(ibdev, port, props, netw_view) :
404				eth_link_query_port(ibdev, port, props, netw_view);
405
406	return err;
407}
408
409static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
410			      struct ib_port_attr *props)
411{
412	/* returns host view */
413	return __mlx4_ib_query_port(ibdev, port, props, 0);
414}
415
416int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
417			union ib_gid *gid, int netw_view)
418{
419	struct ib_smp *in_mad  = NULL;
420	struct ib_smp *out_mad = NULL;
421	int err = -ENOMEM;
422	struct mlx4_ib_dev *dev = to_mdev(ibdev);
423	int clear = 0;
424	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
425
426	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
427	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
428	if (!in_mad || !out_mad)
429		goto out;
430
431	init_query_mad(in_mad);
432	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
433	in_mad->attr_mod = cpu_to_be32(port);
434
435	if (mlx4_is_mfunc(dev->dev) && netw_view)
436		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
437
438	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
439	if (err)
440		goto out;
441
442	memcpy(gid->raw, out_mad->data + 8, 8);
443
444	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
445		if (index) {
446			/* For any index > 0, return the null guid */
447			err = 0;
448			clear = 1;
449			goto out;
450		}
451	}
452
453	init_query_mad(in_mad);
454	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
455	in_mad->attr_mod = cpu_to_be32(index / 8);
456
457	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
458			   NULL, NULL, in_mad, out_mad);
459	if (err)
460		goto out;
461
462	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
463
464out:
465	if (clear)
466		memset(gid->raw + 8, 0, 8);
467	kfree(in_mad);
468	kfree(out_mad);
469	return err;
470}
471
472static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
473			  union ib_gid *gid)
474{
475	struct mlx4_ib_dev *dev = to_mdev(ibdev);
476
477	*gid = dev->iboe.gid_table[port - 1][index];
478
479	return 0;
480}
481
482static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
483			     union ib_gid *gid)
484{
485	if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
486		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
487	else
488		return iboe_query_gid(ibdev, port, index, gid);
489}
490
491int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
492			 u16 *pkey, int netw_view)
493{
494	struct ib_smp *in_mad  = NULL;
495	struct ib_smp *out_mad = NULL;
496	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
497	int err = -ENOMEM;
498
499	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
500	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
501	if (!in_mad || !out_mad)
502		goto out;
503
504	init_query_mad(in_mad);
505	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
506	in_mad->attr_mod = cpu_to_be32(index / 32);
507
508	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
509		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
510
511	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
512			   in_mad, out_mad);
513	if (err)
514		goto out;
515
516	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
517
518out:
519	kfree(in_mad);
520	kfree(out_mad);
521	return err;
522}
523
524static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
525{
526	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
527}
528
529static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
530				 struct ib_device_modify *props)
531{
532	struct mlx4_cmd_mailbox *mailbox;
533	unsigned long flags;
534
535	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
536		return -EOPNOTSUPP;
537
538	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
539		return 0;
540
541	if (mlx4_is_slave(to_mdev(ibdev)->dev))
542		return -EOPNOTSUPP;
543
544	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
545	memcpy(ibdev->node_desc, props->node_desc, 64);
546	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
547
548	/*
549	 * If possible, pass node desc to FW, so it can generate
550	 * a 144 trap.  If cmd fails, just ignore.
551	 */
552	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
553	if (IS_ERR(mailbox))
554		return 0;
555
556	memcpy(mailbox->buf, props->node_desc, 64);
557	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
558		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
559
560	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
561
562	return 0;
563}
564
565static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
566			    u32 cap_mask)
567{
568	struct mlx4_cmd_mailbox *mailbox;
569	int err;
570
571	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
572	if (IS_ERR(mailbox))
573		return PTR_ERR(mailbox);
574
575	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
576		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
577		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
578	} else {
579		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
580		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
581	}
582
583	err = mlx4_cmd(dev->dev, mailbox->dma, port, 0, MLX4_CMD_SET_PORT,
584		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
585
586	mlx4_free_cmd_mailbox(dev->dev, mailbox);
587	return err;
588}
589
590static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
591			       struct ib_port_modify *props)
592{
593	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
594	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
595	struct ib_port_attr attr;
596	u32 cap_mask;
597	int err;
598
599	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
600	 * of whether port link layer is ETH or IB. For ETH ports, qkey
601	 * violations and port capabilities are not meaningful.
602	 */
603	if (is_eth)
604		return 0;
605
606	mutex_lock(&mdev->cap_mask_mutex);
607
608	err = mlx4_ib_query_port(ibdev, port, &attr);
609	if (err)
610		goto out;
611
612	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
613		~props->clr_port_cap_mask;
614
615	err = mlx4_ib_SET_PORT(mdev, port,
616			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
617			       cap_mask);
618
619out:
620	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
621	return err;
622}
623
624static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
625						  struct ib_udata *udata)
626{
627	struct mlx4_ib_dev *dev = to_mdev(ibdev);
628	struct mlx4_ib_ucontext *context;
629	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
630	struct mlx4_ib_alloc_ucontext_resp resp;
631	int err;
632
633	if (!dev->ib_active)
634		return ERR_PTR(-EAGAIN);
635
636	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
637		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
638		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
639		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
640	} else {
641		resp.dev_caps	      = dev->dev->caps.userspace_caps;
642		resp.qp_tab_size      = dev->dev->caps.num_qps;
643		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
644		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
645		resp.cqe_size	      = dev->dev->caps.cqe_size;
646	}
647
648	context = kmalloc(sizeof *context, GFP_KERNEL);
649	if (!context)
650		return ERR_PTR(-ENOMEM);
651
652	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
653	if (err) {
654		kfree(context);
655		return ERR_PTR(err);
656	}
657
658	INIT_LIST_HEAD(&context->db_page_list);
659	mutex_init(&context->db_page_mutex);
660
661	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
662		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
663	else
664		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
665
666	if (err) {
667		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
668		kfree(context);
669		return ERR_PTR(-EFAULT);
670	}
671
672	return &context->ibucontext;
673}
674
675static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
676{
677	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
678
679	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
680	kfree(context);
681
682	return 0;
683}
684
685static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
686{
687	struct mlx4_ib_dev *dev = to_mdev(context->device);
688
689	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
690		return -EINVAL;
691
692	if (vma->vm_pgoff == 0) {
693		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
694
695		if (io_remap_pfn_range(vma, vma->vm_start,
696				       to_mucontext(context)->uar.pfn,
697				       PAGE_SIZE, vma->vm_page_prot))
698			return -EAGAIN;
699	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
700		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
701
702		if (io_remap_pfn_range(vma, vma->vm_start,
703				       to_mucontext(context)->uar.pfn +
704				       dev->dev->caps.num_uars,
705				       PAGE_SIZE, vma->vm_page_prot))
706			return -EAGAIN;
707	} else
708		return -EINVAL;
709
710	return 0;
711}
712
713static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
714				      struct ib_ucontext *context,
715				      struct ib_udata *udata)
716{
717	struct mlx4_ib_pd *pd;
718	int err;
719
720	pd = kmalloc(sizeof *pd, GFP_KERNEL);
721	if (!pd)
722		return ERR_PTR(-ENOMEM);
723
724	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
725	if (err) {
726		kfree(pd);
727		return ERR_PTR(err);
728	}
729
730	if (context)
731		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
732			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
733			kfree(pd);
734			return ERR_PTR(-EFAULT);
735		}
736
737	return &pd->ibpd;
738}
739
740static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
741{
742	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
743	kfree(pd);
744
745	return 0;
746}
747
748static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
749					  struct ib_ucontext *context,
750					  struct ib_udata *udata)
751{
752	struct mlx4_ib_xrcd *xrcd;
753	int err;
754
755	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
756		return ERR_PTR(-ENOSYS);
757
758	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
759	if (!xrcd)
760		return ERR_PTR(-ENOMEM);
761
762	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
763	if (err)
764		goto err1;
765
766	xrcd->pd = ib_alloc_pd(ibdev);
767	if (IS_ERR(xrcd->pd)) {
768		err = PTR_ERR(xrcd->pd);
769		goto err2;
770	}
771
772	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
773	if (IS_ERR(xrcd->cq)) {
774		err = PTR_ERR(xrcd->cq);
775		goto err3;
776	}
777
778	return &xrcd->ibxrcd;
779
780err3:
781	ib_dealloc_pd(xrcd->pd);
782err2:
783	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
784err1:
785	kfree(xrcd);
786	return ERR_PTR(err);
787}
788
789static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
790{
791	ib_destroy_cq(to_mxrcd(xrcd)->cq);
792	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
793	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
794	kfree(xrcd);
795
796	return 0;
797}
798
799static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
800{
801	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
802	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
803	struct mlx4_ib_gid_entry *ge;
804
805	ge = kzalloc(sizeof *ge, GFP_KERNEL);
806	if (!ge)
807		return -ENOMEM;
808
809	ge->gid = *gid;
810	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
811		ge->port = mqp->port;
812		ge->added = 1;
813	}
814
815	mutex_lock(&mqp->mutex);
816	list_add_tail(&ge->list, &mqp->gid_list);
817	mutex_unlock(&mqp->mutex);
818
819	return 0;
820}
821
822int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
823		   union ib_gid *gid)
824{
825	struct net_device *ndev;
826	int ret = 0;
827
828	if (!mqp->port)
829		return 0;
830
831	spin_lock_bh(&mdev->iboe.lock);
832	ndev = mdev->iboe.netdevs[mqp->port - 1];
833	if (ndev)
834		dev_hold(ndev);
835	spin_unlock_bh(&mdev->iboe.lock);
836
837	if (ndev) {
838		ret = 1;
839		dev_put(ndev);
840	}
841
842	return ret;
843}
844
845struct mlx4_ib_steering {
846	struct list_head list;
847	u64 reg_id;
848	union ib_gid gid;
849};
850
851static int parse_flow_attr(struct mlx4_dev *dev,
852			   u32 qp_num,
853			   union ib_flow_spec *ib_spec,
854			   struct _rule_hw *mlx4_spec)
855{
856	enum mlx4_net_trans_rule_id type;
857
858	switch (ib_spec->type) {
859	case IB_FLOW_SPEC_ETH:
860		type = MLX4_NET_TRANS_RULE_ID_ETH;
861		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
862		       ETH_ALEN);
863		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
864		       ETH_ALEN);
865		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
866		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
867		break;
868	case IB_FLOW_SPEC_IB:
869		type = MLX4_NET_TRANS_RULE_ID_IB;
870		mlx4_spec->ib.l3_qpn =
871			cpu_to_be32(qp_num);
872		mlx4_spec->ib.qpn_mask =
873			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
874		break;
875
876
877	case IB_FLOW_SPEC_IPV4:
878		type = MLX4_NET_TRANS_RULE_ID_IPV4;
879		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
880		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
881		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
882		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
883		break;
884
885	case IB_FLOW_SPEC_TCP:
886	case IB_FLOW_SPEC_UDP:
887		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
888					MLX4_NET_TRANS_RULE_ID_TCP :
889					MLX4_NET_TRANS_RULE_ID_UDP;
890		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
891		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
892		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
893		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
894		break;
895
896	default:
897		return -EINVAL;
898	}
899	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
900	    mlx4_hw_rule_sz(dev, type) < 0)
901		return -EINVAL;
902	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
903	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
904	return mlx4_hw_rule_sz(dev, type);
905}
906
907struct default_rules {
908	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
909	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
910	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
911	__u8  link_layer;
912};
913static const struct default_rules default_table[] = {
914	{
915		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
916		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
917		.rules_create_list = {IB_FLOW_SPEC_IB},
918		.link_layer = IB_LINK_LAYER_INFINIBAND
919	}
920};
921
922static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
923					 struct ib_flow_attr *flow_attr)
924{
925	int i, j, k;
926	void *ib_flow;
927	const struct default_rules *pdefault_rules = default_table;
928	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
929
930	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
931		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
932		memset(&field_types, 0, sizeof(field_types));
933
934		if (link_layer != pdefault_rules->link_layer)
935			continue;
936
937		ib_flow = flow_attr + 1;
938		/* we assume the specs are sorted */
939		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
940		     j < flow_attr->num_of_specs; k++) {
941			union ib_flow_spec *current_flow =
942				(union ib_flow_spec *)ib_flow;
943
944			/* same layer but different type */
945			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
946			     (pdefault_rules->mandatory_fields[k] &
947			      IB_FLOW_SPEC_LAYER_MASK)) &&
948			    (current_flow->type !=
949			     pdefault_rules->mandatory_fields[k]))
950				goto out;
951
952			/* same layer, try match next one */
953			if (current_flow->type ==
954			    pdefault_rules->mandatory_fields[k]) {
955				j++;
956				ib_flow +=
957					((union ib_flow_spec *)ib_flow)->size;
958			}
959		}
960
961		ib_flow = flow_attr + 1;
962		for (j = 0; j < flow_attr->num_of_specs;
963		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
964			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
965				/* same layer and same type */
966				if (((union ib_flow_spec *)ib_flow)->type ==
967				    pdefault_rules->mandatory_not_fields[k])
968					goto out;
969
970		return i;
971	}
972out:
973	return -1;
974}
975
976static int __mlx4_ib_create_default_rules(
977		struct mlx4_ib_dev *mdev,
978		struct ib_qp *qp,
979		const struct default_rules *pdefault_rules,
980		struct _rule_hw *mlx4_spec) {
981	int size = 0;
982	int i;
983
984	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
985		int ret;
986		union ib_flow_spec ib_spec;
987		switch (pdefault_rules->rules_create_list[i]) {
988		case 0:
989			/* no rule */
990			continue;
991		case IB_FLOW_SPEC_IB:
992			ib_spec.type = IB_FLOW_SPEC_IB;
993			ib_spec.size = sizeof(struct ib_flow_spec_ib);
994
995			break;
996		default:
997			/* invalid rule */
998			return -EINVAL;
999		}
1000		/* We must put empty rule, qpn is being ignored */
1001		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1002				      mlx4_spec);
1003		if (ret < 0) {
1004			pr_info("invalid parsing\n");
1005			return -EINVAL;
1006		}
1007
1008		mlx4_spec = (void *)mlx4_spec + ret;
1009		size += ret;
1010	}
1011	return size;
1012}
1013
1014static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1015			  int domain,
1016			  enum mlx4_net_trans_promisc_mode flow_type,
1017			  u64 *reg_id)
1018{
1019	int ret, i;
1020	int size = 0;
1021	void *ib_flow;
1022	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1023	struct mlx4_cmd_mailbox *mailbox;
1024	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1025	int default_flow;
1026
1027	static const u16 __mlx4_domain[] = {
1028		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1029		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1030		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1031		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1032	};
1033
1034	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1035		pr_err("Invalid priority value %d\n", flow_attr->priority);
1036		return -EINVAL;
1037	}
1038
1039	if (domain >= IB_FLOW_DOMAIN_NUM) {
1040		pr_err("Invalid domain value %d\n", domain);
1041		return -EINVAL;
1042	}
1043
1044	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1045		return -EINVAL;
1046
1047	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1048	if (IS_ERR(mailbox))
1049		return PTR_ERR(mailbox);
1050	ctrl = mailbox->buf;
1051
1052	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1053				 flow_attr->priority);
1054	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1055	ctrl->port = flow_attr->port;
1056	ctrl->qpn = cpu_to_be32(qp->qp_num);
1057
1058	ib_flow = flow_attr + 1;
1059	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1060	/* Add default flows */
1061	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1062	if (default_flow >= 0) {
1063		ret = __mlx4_ib_create_default_rules(
1064				mdev, qp, default_table + default_flow,
1065				mailbox->buf + size);
1066		if (ret < 0) {
1067			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1068			return -EINVAL;
1069		}
1070		size += ret;
1071	}
1072	for (i = 0; i < flow_attr->num_of_specs; i++) {
1073		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1074				      mailbox->buf + size);
1075		if (ret < 0) {
1076			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1077			return -EINVAL;
1078		}
1079		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1080		size += ret;
1081	}
1082
1083	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1084			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1085			   MLX4_CMD_NATIVE);
1086	if (ret == -ENOMEM)
1087		pr_err("mcg table is full. Fail to register network rule.\n");
1088	else if (ret == -ENXIO)
1089		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1090	else if (ret)
1091		pr_err("Invalid argumant. Fail to register network rule.\n");
1092
1093	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1094	return ret;
1095}
1096
1097static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1098{
1099	int err;
1100	err = mlx4_cmd(dev, reg_id, 0, 0,
1101		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1102		       MLX4_CMD_NATIVE);
1103	if (err)
1104		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1105		       reg_id);
1106	return err;
1107}
1108
1109static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1110				    u64 *reg_id)
1111{
1112	void *ib_flow;
1113	union ib_flow_spec *ib_spec;
1114	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1115	int err = 0;
1116
1117	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1118		return 0; /* do nothing */
1119
1120	ib_flow = flow_attr + 1;
1121	ib_spec = (union ib_flow_spec *)ib_flow;
1122
1123	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1124		return 0; /* do nothing */
1125
1126	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1127				    flow_attr->port, qp->qp_num,
1128				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1129				    reg_id);
1130	return err;
1131}
1132
1133static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1134				    struct ib_flow_attr *flow_attr,
1135				    int domain)
1136{
1137	int err = 0, i = 0;
1138	struct mlx4_ib_flow *mflow;
1139	enum mlx4_net_trans_promisc_mode type[2];
1140
1141	memset(type, 0, sizeof(type));
1142
1143	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1144	if (!mflow) {
1145		err = -ENOMEM;
1146		goto err_free;
1147	}
1148
1149	switch (flow_attr->type) {
1150	case IB_FLOW_ATTR_NORMAL:
1151		type[0] = MLX4_FS_REGULAR;
1152		break;
1153
1154	case IB_FLOW_ATTR_ALL_DEFAULT:
1155		type[0] = MLX4_FS_ALL_DEFAULT;
1156		break;
1157
1158	case IB_FLOW_ATTR_MC_DEFAULT:
1159		type[0] = MLX4_FS_MC_DEFAULT;
1160		break;
1161
1162	case IB_FLOW_ATTR_SNIFFER:
1163		type[0] = MLX4_FS_UC_SNIFFER;
1164		type[1] = MLX4_FS_MC_SNIFFER;
1165		break;
1166
1167	default:
1168		err = -EINVAL;
1169		goto err_free;
1170	}
1171
1172	while (i < ARRAY_SIZE(type) && type[i]) {
1173		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1174					    &mflow->reg_id[i]);
1175		if (err)
1176			goto err_create_flow;
1177		i++;
1178	}
1179
1180	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1181		err = mlx4_ib_tunnel_steer_add(qp, flow_attr, &mflow->reg_id[i]);
1182		if (err)
1183			goto err_create_flow;
1184		i++;
1185	}
1186
1187	return &mflow->ibflow;
1188
1189err_create_flow:
1190	while (i) {
1191		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev, mflow->reg_id[i]);
1192		i--;
1193	}
1194err_free:
1195	kfree(mflow);
1196	return ERR_PTR(err);
1197}
1198
1199static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1200{
1201	int err, ret = 0;
1202	int i = 0;
1203	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1204	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1205
1206	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) {
1207		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]);
1208		if (err)
1209			ret = err;
1210		i++;
1211	}
1212
1213	kfree(mflow);
1214	return ret;
1215}
1216
1217static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1218{
1219	int err;
1220	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1221	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1222	u64 reg_id;
1223	struct mlx4_ib_steering *ib_steering = NULL;
1224	enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1225		MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1226
1227	if (mdev->dev->caps.steering_mode ==
1228	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1229		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1230		if (!ib_steering)
1231			return -ENOMEM;
1232	}
1233
1234	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1235				    !!(mqp->flags &
1236				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1237				    prot, &reg_id);
1238	if (err)
1239		goto err_malloc;
1240
1241	err = add_gid_entry(ibqp, gid);
1242	if (err)
1243		goto err_add;
1244
1245	if (ib_steering) {
1246		memcpy(ib_steering->gid.raw, gid->raw, 16);
1247		ib_steering->reg_id = reg_id;
1248		mutex_lock(&mqp->mutex);
1249		list_add(&ib_steering->list, &mqp->steering_rules);
1250		mutex_unlock(&mqp->mutex);
1251	}
1252	return 0;
1253
1254err_add:
1255	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1256			      prot, reg_id);
1257err_malloc:
1258	kfree(ib_steering);
1259
1260	return err;
1261}
1262
1263static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1264{
1265	struct mlx4_ib_gid_entry *ge;
1266	struct mlx4_ib_gid_entry *tmp;
1267	struct mlx4_ib_gid_entry *ret = NULL;
1268
1269	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1270		if (!memcmp(raw, ge->gid.raw, 16)) {
1271			ret = ge;
1272			break;
1273		}
1274	}
1275
1276	return ret;
1277}
1278
1279static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1280{
1281	int err;
1282	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1283	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1284	struct net_device *ndev;
1285	struct mlx4_ib_gid_entry *ge;
1286	u64 reg_id = 0;
1287	enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1288		MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1289
1290	if (mdev->dev->caps.steering_mode ==
1291	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1292		struct mlx4_ib_steering *ib_steering;
1293
1294		mutex_lock(&mqp->mutex);
1295		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1296			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1297				list_del(&ib_steering->list);
1298				break;
1299			}
1300		}
1301		mutex_unlock(&mqp->mutex);
1302		if (&ib_steering->list == &mqp->steering_rules) {
1303			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1304			return -EINVAL;
1305		}
1306		reg_id = ib_steering->reg_id;
1307		kfree(ib_steering);
1308	}
1309
1310	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1311				    prot, reg_id);
1312	if (err)
1313		return err;
1314
1315	mutex_lock(&mqp->mutex);
1316	ge = find_gid_entry(mqp, gid->raw);
1317	if (ge) {
1318		spin_lock_bh(&mdev->iboe.lock);
1319		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1320		if (ndev)
1321			dev_hold(ndev);
1322		spin_unlock_bh(&mdev->iboe.lock);
1323		if (ndev)
1324			dev_put(ndev);
1325		list_del(&ge->list);
1326		kfree(ge);
1327	} else
1328		pr_warn("could not find mgid entry\n");
1329
1330	mutex_unlock(&mqp->mutex);
1331
1332	return 0;
1333}
1334
1335static int init_node_data(struct mlx4_ib_dev *dev)
1336{
1337	struct ib_smp *in_mad  = NULL;
1338	struct ib_smp *out_mad = NULL;
1339	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1340	int err = -ENOMEM;
1341
1342	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1343	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1344	if (!in_mad || !out_mad)
1345		goto out;
1346
1347	init_query_mad(in_mad);
1348	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1349	if (mlx4_is_master(dev->dev))
1350		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1351
1352	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1353	if (err)
1354		goto out;
1355
1356	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1357
1358	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1359
1360	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1361	if (err)
1362		goto out;
1363
1364	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1365	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1366
1367out:
1368	kfree(in_mad);
1369	kfree(out_mad);
1370	return err;
1371}
1372
1373static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1374			char *buf)
1375{
1376	struct mlx4_ib_dev *dev =
1377		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1378	return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
1379}
1380
1381static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1382			   char *buf)
1383{
1384	struct mlx4_ib_dev *dev =
1385		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1386	return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1387		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1388		       (int) dev->dev->caps.fw_ver & 0xffff);
1389}
1390
1391static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1392			char *buf)
1393{
1394	struct mlx4_ib_dev *dev =
1395		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1396	return sprintf(buf, "%x\n", dev->dev->rev_id);
1397}
1398
1399static ssize_t show_board(struct device *device, struct device_attribute *attr,
1400			  char *buf)
1401{
1402	struct mlx4_ib_dev *dev =
1403		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1404	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1405		       dev->dev->board_id);
1406}
1407
1408static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1409static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1410static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1411static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1412
1413static struct device_attribute *mlx4_class_attributes[] = {
1414	&dev_attr_hw_rev,
1415	&dev_attr_fw_ver,
1416	&dev_attr_hca_type,
1417	&dev_attr_board_id
1418};
1419
1420static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1421				     struct net_device *dev)
1422{
1423	memcpy(eui, dev->dev_addr, 3);
1424	memcpy(eui + 5, dev->dev_addr + 3, 3);
1425	if (vlan_id < 0x1000) {
1426		eui[3] = vlan_id >> 8;
1427		eui[4] = vlan_id & 0xff;
1428	} else {
1429		eui[3] = 0xff;
1430		eui[4] = 0xfe;
1431	}
1432	eui[0] ^= 2;
1433}
1434
1435static void update_gids_task(struct work_struct *work)
1436{
1437	struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1438	struct mlx4_cmd_mailbox *mailbox;
1439	union ib_gid *gids;
1440	int err;
1441	struct mlx4_dev	*dev = gw->dev->dev;
1442
1443	if (!gw->dev->ib_active)
1444		return;
1445
1446	mailbox = mlx4_alloc_cmd_mailbox(dev);
1447	if (IS_ERR(mailbox)) {
1448		pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1449		return;
1450	}
1451
1452	gids = mailbox->buf;
1453	memcpy(gids, gw->gids, sizeof gw->gids);
1454
1455	err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1456		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1457		       MLX4_CMD_WRAPPED);
1458	if (err)
1459		pr_warn("set port command failed\n");
1460	else
1461		mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1462
1463	mlx4_free_cmd_mailbox(dev, mailbox);
1464	kfree(gw);
1465}
1466
1467static void reset_gids_task(struct work_struct *work)
1468{
1469	struct update_gid_work *gw =
1470			container_of(work, struct update_gid_work, work);
1471	struct mlx4_cmd_mailbox *mailbox;
1472	union ib_gid *gids;
1473	int err;
1474	struct mlx4_dev	*dev = gw->dev->dev;
1475
1476	if (!gw->dev->ib_active)
1477		return;
1478
1479	mailbox = mlx4_alloc_cmd_mailbox(dev);
1480	if (IS_ERR(mailbox)) {
1481		pr_warn("reset gid table failed\n");
1482		goto free;
1483	}
1484
1485	gids = mailbox->buf;
1486	memcpy(gids, gw->gids, sizeof(gw->gids));
1487
1488	if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1489				    IB_LINK_LAYER_ETHERNET) {
1490		err = mlx4_cmd(dev, mailbox->dma,
1491			       MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1492			       1, MLX4_CMD_SET_PORT,
1493			       MLX4_CMD_TIME_CLASS_B,
1494			       MLX4_CMD_WRAPPED);
1495		if (err)
1496			pr_warn(KERN_WARNING
1497				"set port %d command failed\n", gw->port);
1498	}
1499
1500	mlx4_free_cmd_mailbox(dev, mailbox);
1501free:
1502	kfree(gw);
1503}
1504
1505static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1506			    union ib_gid *gid, int clear,
1507			    int default_gid)
1508{
1509	struct update_gid_work *work;
1510	int i;
1511	int need_update = 0;
1512	int free = -1;
1513	int found = -1;
1514	int max_gids;
1515
1516	if (default_gid) {
1517		free = 0;
1518	} else {
1519		max_gids = dev->dev->caps.gid_table_len[port];
1520		for (i = 1; i < max_gids; ++i) {
1521			if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1522				    sizeof(*gid)))
1523				found = i;
1524
1525			if (clear) {
1526				if (found >= 0) {
1527					need_update = 1;
1528					dev->iboe.gid_table[port - 1][found] =
1529						zgid;
1530					break;
1531				}
1532			} else {
1533				if (found >= 0)
1534					break;
1535
1536				if (free < 0 &&
1537				    !memcmp(&dev->iboe.gid_table[port - 1][i],
1538					    &zgid, sizeof(*gid)))
1539					free = i;
1540			}
1541		}
1542	}
1543
1544	if (found == -1 && !clear && free >= 0) {
1545		dev->iboe.gid_table[port - 1][free] = *gid;
1546		need_update = 1;
1547	}
1548
1549	if (!need_update)
1550		return 0;
1551
1552	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1553	if (!work)
1554		return -ENOMEM;
1555
1556	memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1557	INIT_WORK(&work->work, update_gids_task);
1558	work->port = port;
1559	work->dev = dev;
1560	queue_work(wq, &work->work);
1561
1562	return 0;
1563}
1564
1565static void mlx4_make_default_gid(struct  net_device *dev, union ib_gid *gid)
1566{
1567	gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1568	mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1569}
1570
1571
1572static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1573{
1574	struct update_gid_work *work;
1575
1576	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1577	if (!work)
1578		return -ENOMEM;
1579
1580	memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1581	memset(work->gids, 0, sizeof(work->gids));
1582	INIT_WORK(&work->work, reset_gids_task);
1583	work->dev = dev;
1584	work->port = port;
1585	queue_work(wq, &work->work);
1586	return 0;
1587}
1588
1589static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1590			      struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1591{
1592	struct mlx4_ib_iboe *iboe;
1593	int port = 0;
1594	struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1595				rdma_vlan_dev_real_dev(event_netdev) :
1596				event_netdev;
1597	union ib_gid default_gid;
1598
1599	mlx4_make_default_gid(real_dev, &default_gid);
1600
1601	if (!memcmp(gid, &default_gid, sizeof(*gid)))
1602		return 0;
1603
1604	if (event != NETDEV_DOWN && event != NETDEV_UP)
1605		return 0;
1606
1607	if ((real_dev != event_netdev) &&
1608	    (event == NETDEV_DOWN) &&
1609	    rdma_link_local_addr((struct in6_addr *)gid))
1610		return 0;
1611
1612	iboe = &ibdev->iboe;
1613	spin_lock_bh(&iboe->lock);
1614
1615	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1616		if ((netif_is_bond_master(real_dev) &&
1617		     (real_dev == iboe->masters[port - 1])) ||
1618		     (!netif_is_bond_master(real_dev) &&
1619		     (real_dev == iboe->netdevs[port - 1])))
1620			update_gid_table(ibdev, port, gid,
1621					 event == NETDEV_DOWN, 0);
1622
1623	spin_unlock_bh(&iboe->lock);
1624	return 0;
1625
1626}
1627
1628static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1629			       struct mlx4_ib_dev *ibdev)
1630{
1631	u8 port = 0;
1632	struct mlx4_ib_iboe *iboe;
1633	struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1634				rdma_vlan_dev_real_dev(dev) : dev;
1635
1636	iboe = &ibdev->iboe;
1637
1638	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1639		if ((netif_is_bond_master(real_dev) &&
1640		     (real_dev == iboe->masters[port - 1])) ||
1641		     (!netif_is_bond_master(real_dev) &&
1642		     (real_dev == iboe->netdevs[port - 1])))
1643			break;
1644
1645	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1646		return 0;
1647	else
1648		return port;
1649}
1650
1651static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1652				void *ptr)
1653{
1654	struct mlx4_ib_dev *ibdev;
1655	struct in_ifaddr *ifa = ptr;
1656	union ib_gid gid;
1657	struct net_device *event_netdev = ifa->ifa_dev->dev;
1658
1659	ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1660
1661	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1662
1663	mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1664	return NOTIFY_DONE;
1665}
1666
1667#if IS_ENABLED(CONFIG_IPV6)
1668static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1669				void *ptr)
1670{
1671	struct mlx4_ib_dev *ibdev;
1672	struct inet6_ifaddr *ifa = ptr;
1673	union  ib_gid *gid = (union ib_gid *)&ifa->addr;
1674	struct net_device *event_netdev = ifa->idev->dev;
1675
1676	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1677
1678	mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1679	return NOTIFY_DONE;
1680}
1681#endif
1682
1683#define MLX4_IB_INVALID_MAC	((u64)-1)
1684static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1685			       struct net_device *dev,
1686			       int port)
1687{
1688	u64 new_smac = 0;
1689	u64 release_mac = MLX4_IB_INVALID_MAC;
1690	struct mlx4_ib_qp *qp;
1691
1692	read_lock(&dev_base_lock);
1693	new_smac = mlx4_mac_to_u64(dev->dev_addr);
1694	read_unlock(&dev_base_lock);
1695
1696	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
1697
1698	/* no need for update QP1 and mac registration in non-SRIOV */
1699	if (!mlx4_is_mfunc(ibdev->dev))
1700		return;
1701
1702	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1703	qp = ibdev->qp1_proxy[port - 1];
1704	if (qp) {
1705		int new_smac_index;
1706		u64 old_smac;
1707		struct mlx4_update_qp_params update_params;
1708
1709		mutex_lock(&qp->mutex);
1710		old_smac = qp->pri.smac;
1711		if (new_smac == old_smac)
1712			goto unlock;
1713
1714		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1715
1716		if (new_smac_index < 0)
1717			goto unlock;
1718
1719		update_params.smac_index = new_smac_index;
1720		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
1721				   &update_params)) {
1722			release_mac = new_smac;
1723			goto unlock;
1724		}
1725		/* if old port was zero, no mac was yet registered for this QP */
1726		if (qp->pri.smac_port)
1727			release_mac = old_smac;
1728		qp->pri.smac = new_smac;
1729		qp->pri.smac_port = port;
1730		qp->pri.smac_index = new_smac_index;
1731	}
1732
1733unlock:
1734	if (release_mac != MLX4_IB_INVALID_MAC)
1735		mlx4_unregister_mac(ibdev->dev, port, release_mac);
1736	if (qp)
1737		mutex_unlock(&qp->mutex);
1738	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1739}
1740
1741static void mlx4_ib_get_dev_addr(struct net_device *dev,
1742				 struct mlx4_ib_dev *ibdev, u8 port)
1743{
1744	struct in_device *in_dev;
1745#if IS_ENABLED(CONFIG_IPV6)
1746	struct inet6_dev *in6_dev;
1747	union ib_gid  *pgid;
1748	struct inet6_ifaddr *ifp;
1749	union ib_gid default_gid;
1750#endif
1751	union ib_gid gid;
1752
1753
1754	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1755		return;
1756
1757	/* IPv4 gids */
1758	in_dev = in_dev_get(dev);
1759	if (in_dev) {
1760		for_ifa(in_dev) {
1761			/*ifa->ifa_address;*/
1762			ipv6_addr_set_v4mapped(ifa->ifa_address,
1763					       (struct in6_addr *)&gid);
1764			update_gid_table(ibdev, port, &gid, 0, 0);
1765		}
1766		endfor_ifa(in_dev);
1767		in_dev_put(in_dev);
1768	}
1769#if IS_ENABLED(CONFIG_IPV6)
1770	mlx4_make_default_gid(dev, &default_gid);
1771	/* IPv6 gids */
1772	in6_dev = in6_dev_get(dev);
1773	if (in6_dev) {
1774		read_lock_bh(&in6_dev->lock);
1775		list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1776			pgid = (union ib_gid *)&ifp->addr;
1777			if (!memcmp(pgid, &default_gid, sizeof(*pgid)))
1778				continue;
1779			update_gid_table(ibdev, port, pgid, 0, 0);
1780		}
1781		read_unlock_bh(&in6_dev->lock);
1782		in6_dev_put(in6_dev);
1783	}
1784#endif
1785}
1786
1787static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1788				 struct  net_device *dev, u8 port)
1789{
1790	union ib_gid gid;
1791	mlx4_make_default_gid(dev, &gid);
1792	update_gid_table(ibdev, port, &gid, 0, 1);
1793}
1794
1795static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1796{
1797	struct	net_device *dev;
1798	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1799	int i;
1800	int err = 0;
1801
1802	for (i = 1; i <= ibdev->num_ports; ++i) {
1803		if (rdma_port_get_link_layer(&ibdev->ib_dev, i) ==
1804		    IB_LINK_LAYER_ETHERNET) {
1805			err = reset_gid_table(ibdev, i);
1806			if (err)
1807				goto out;
1808		}
1809	}
1810
1811	read_lock(&dev_base_lock);
1812	spin_lock_bh(&iboe->lock);
1813
1814	for_each_netdev(&init_net, dev) {
1815		u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1816		/* port will be non-zero only for ETH ports */
1817		if (port) {
1818			mlx4_ib_set_default_gid(ibdev, dev, port);
1819			mlx4_ib_get_dev_addr(dev, ibdev, port);
1820		}
1821	}
1822
1823	spin_unlock_bh(&iboe->lock);
1824	read_unlock(&dev_base_lock);
1825out:
1826	return err;
1827}
1828
1829static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1830				 struct net_device *dev,
1831				 unsigned long event)
1832
1833{
1834	struct mlx4_ib_iboe *iboe;
1835	int update_qps_port = -1;
1836	int port;
1837
1838	iboe = &ibdev->iboe;
1839
1840	spin_lock_bh(&iboe->lock);
1841	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1842		enum ib_port_state	port_state = IB_PORT_NOP;
1843		struct net_device *old_master = iboe->masters[port - 1];
1844		struct net_device *curr_netdev;
1845		struct net_device *curr_master;
1846
1847		iboe->netdevs[port - 1] =
1848			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1849		if (iboe->netdevs[port - 1])
1850			mlx4_ib_set_default_gid(ibdev,
1851						iboe->netdevs[port - 1], port);
1852		curr_netdev = iboe->netdevs[port - 1];
1853
1854		if (iboe->netdevs[port - 1] &&
1855		    netif_is_bond_slave(iboe->netdevs[port - 1])) {
1856			iboe->masters[port - 1] = netdev_master_upper_dev_get(
1857				iboe->netdevs[port - 1]);
1858		} else {
1859			iboe->masters[port - 1] = NULL;
1860		}
1861		curr_master = iboe->masters[port - 1];
1862
1863		if (dev == iboe->netdevs[port - 1] &&
1864		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1865		     event == NETDEV_UP || event == NETDEV_CHANGE))
1866			update_qps_port = port;
1867
1868		if (curr_netdev) {
1869			port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
1870						IB_PORT_ACTIVE : IB_PORT_DOWN;
1871			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1872			if (curr_master) {
1873				/* if using bonding/team and a slave port is down, we
1874				 * don't want the bond IP based gids in the table since
1875				 * flows that select port by gid may get the down port.
1876				*/
1877				if (port_state == IB_PORT_DOWN) {
1878					reset_gid_table(ibdev, port);
1879					mlx4_ib_set_default_gid(ibdev,
1880								curr_netdev,
1881								port);
1882				} else {
1883					/* gids from the upper dev (bond/team)
1884					 * should appear in port's gid table
1885					*/
1886					mlx4_ib_get_dev_addr(curr_master,
1887							     ibdev, port);
1888				}
1889			}
1890			/* if bonding is used it is possible that we add it to
1891			 * masters only after IP address is assigned to the
1892			 * net bonding interface.
1893			*/
1894			if (curr_master && (old_master != curr_master)) {
1895				reset_gid_table(ibdev, port);
1896				mlx4_ib_set_default_gid(ibdev,
1897							curr_netdev, port);
1898				mlx4_ib_get_dev_addr(curr_master, ibdev, port);
1899			}
1900
1901			if (!curr_master && (old_master != curr_master)) {
1902				reset_gid_table(ibdev, port);
1903				mlx4_ib_set_default_gid(ibdev,
1904							curr_netdev, port);
1905				mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
1906			}
1907		} else {
1908			reset_gid_table(ibdev, port);
1909		}
1910	}
1911
1912	spin_unlock_bh(&iboe->lock);
1913
1914	if (update_qps_port > 0)
1915		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
1916}
1917
1918static int mlx4_ib_netdev_event(struct notifier_block *this,
1919				unsigned long event, void *ptr)
1920{
1921	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1922	struct mlx4_ib_dev *ibdev;
1923
1924	if (!net_eq(dev_net(dev), &init_net))
1925		return NOTIFY_DONE;
1926
1927	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1928	mlx4_ib_scan_netdevs(ibdev, dev, event);
1929
1930	return NOTIFY_DONE;
1931}
1932
1933static void init_pkeys(struct mlx4_ib_dev *ibdev)
1934{
1935	int port;
1936	int slave;
1937	int i;
1938
1939	if (mlx4_is_master(ibdev->dev)) {
1940		for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1941			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1942				for (i = 0;
1943				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1944				     ++i) {
1945					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1946					/* master has the identity virt2phys pkey mapping */
1947						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1948							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1949					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1950							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1951				}
1952			}
1953		}
1954		/* initialize pkey cache */
1955		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1956			for (i = 0;
1957			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1958			     ++i)
1959				ibdev->pkeys.phys_pkey_cache[port-1][i] =
1960					(i) ? 0 : 0xFFFF;
1961		}
1962	}
1963}
1964
1965static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1966{
1967	char name[80];
1968	int eq_per_port = 0;
1969	int added_eqs = 0;
1970	int total_eqs = 0;
1971	int i, j, eq;
1972
1973	/* Legacy mode or comp_pool is not large enough */
1974	if (dev->caps.comp_pool == 0 ||
1975	    dev->caps.num_ports > dev->caps.comp_pool)
1976		return;
1977
1978	eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1979					dev->caps.num_ports);
1980
1981	/* Init eq table */
1982	added_eqs = 0;
1983	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1984		added_eqs += eq_per_port;
1985
1986	total_eqs = dev->caps.num_comp_vectors + added_eqs;
1987
1988	ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1989	if (!ibdev->eq_table)
1990		return;
1991
1992	ibdev->eq_added = added_eqs;
1993
1994	eq = 0;
1995	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1996		for (j = 0; j < eq_per_port; j++) {
1997			snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%s",
1998				 i, j, dev->pdev->bus->name);
1999			/* Set IRQ for specific name (per ring) */
2000			if (mlx4_assign_eq(dev, name, NULL,
2001					   &ibdev->eq_table[eq])) {
2002				/* Use legacy (same as mlx4_en driver) */
2003				pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
2004				ibdev->eq_table[eq] =
2005					(eq % dev->caps.num_comp_vectors);
2006			}
2007			eq++;
2008		}
2009	}
2010
2011	/* Fill the reset of the vector with legacy EQ */
2012	for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
2013		ibdev->eq_table[eq++] = i;
2014
2015	/* Advertise the new number of EQs to clients */
2016	ibdev->ib_dev.num_comp_vectors = total_eqs;
2017}
2018
2019static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2020{
2021	int i;
2022
2023	/* no additional eqs were added */
2024	if (!ibdev->eq_table)
2025		return;
2026
2027	/* Reset the advertised EQ number */
2028	ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
2029
2030	/* Free only the added eqs */
2031	for (i = 0; i < ibdev->eq_added; i++) {
2032		/* Don't free legacy eqs if used */
2033		if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
2034			continue;
2035		mlx4_release_eq(dev, ibdev->eq_table[i]);
2036	}
2037
2038	kfree(ibdev->eq_table);
2039}
2040
2041static void *mlx4_ib_add(struct mlx4_dev *dev)
2042{
2043	struct mlx4_ib_dev *ibdev;
2044	int num_ports = 0;
2045	int i, j;
2046	int err;
2047	struct mlx4_ib_iboe *iboe;
2048	int ib_num_ports = 0;
2049
2050	pr_info_once("%s", mlx4_ib_version);
2051
2052	num_ports = 0;
2053	mlx4_foreach_ib_transport_port(i, dev)
2054		num_ports++;
2055
2056	/* No point in registering a device with no ports... */
2057	if (num_ports == 0)
2058		return NULL;
2059
2060	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2061	if (!ibdev) {
2062		dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
2063		return NULL;
2064	}
2065
2066	iboe = &ibdev->iboe;
2067
2068	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2069		goto err_dealloc;
2070
2071	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2072		goto err_pd;
2073
2074	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2075				 PAGE_SIZE);
2076	if (!ibdev->uar_map)
2077		goto err_uar;
2078	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2079
2080	ibdev->dev = dev;
2081
2082	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2083	ibdev->ib_dev.owner		= THIS_MODULE;
2084	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2085	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2086	ibdev->num_ports		= num_ports;
2087	ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
2088	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2089	ibdev->ib_dev.dma_device	= &dev->pdev->dev;
2090
2091	if (dev->caps.userspace_caps)
2092		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2093	else
2094		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2095
2096	ibdev->ib_dev.uverbs_cmd_mask	=
2097		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2098		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2099		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2100		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2101		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2102		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2103		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2104		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2105		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2106		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2107		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2108		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2109		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2110		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2111		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2112		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2113		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2114		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2115		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2116		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2117		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2118		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2119		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2120		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2121
2122	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2123	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2124	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2125	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2126	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2127	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2128	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2129	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2130	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2131	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2132	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2133	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2134	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2135	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2136	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2137	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2138	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2139	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2140	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2141	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2142	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2143	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2144	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2145	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2146	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2147	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2148	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2149	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2150	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2151	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2152	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2153	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2154	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2155	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2156	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2157	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2158	ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2159	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2160	ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
2161	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2162	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2163	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2164
2165	if (!mlx4_is_slave(ibdev->dev)) {
2166		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2167		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2168		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2169		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2170	}
2171
2172	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2173	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2174		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2175		ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2176		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2177
2178		ibdev->ib_dev.uverbs_cmd_mask |=
2179			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2180			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2181	}
2182
2183	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2184		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2185		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2186		ibdev->ib_dev.uverbs_cmd_mask |=
2187			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2188			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2189	}
2190
2191	if (check_flow_steering_support(dev)) {
2192		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2193		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2194		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2195
2196		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2197			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2198			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2199	}
2200
2201	mlx4_ib_alloc_eqs(dev, ibdev);
2202
2203	spin_lock_init(&iboe->lock);
2204
2205	if (init_node_data(ibdev))
2206		goto err_map;
2207
2208	for (i = 0; i < ibdev->num_ports; ++i) {
2209		mutex_init(&ibdev->qp1_proxy_lock[i]);
2210		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2211						IB_LINK_LAYER_ETHERNET) {
2212			err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
2213			if (err)
2214				ibdev->counters[i] = -1;
2215		} else {
2216			ibdev->counters[i] = -1;
2217		}
2218	}
2219
2220	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2221		ib_num_ports++;
2222
2223	spin_lock_init(&ibdev->sm_lock);
2224	mutex_init(&ibdev->cap_mask_mutex);
2225
2226	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2227	    ib_num_ports) {
2228		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2229		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2230					    MLX4_IB_UC_STEER_QPN_ALIGN,
2231					    &ibdev->steer_qpn_base);
2232		if (err)
2233			goto err_counter;
2234
2235		ibdev->ib_uc_qpns_bitmap =
2236			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2237				sizeof(long),
2238				GFP_KERNEL);
2239		if (!ibdev->ib_uc_qpns_bitmap) {
2240			dev_err(&dev->pdev->dev, "bit map alloc failed\n");
2241			goto err_steer_qp_release;
2242		}
2243
2244		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2245
2246		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2247				dev, ibdev->steer_qpn_base,
2248				ibdev->steer_qpn_base +
2249				ibdev->steer_qpn_count - 1);
2250		if (err)
2251			goto err_steer_free_bitmap;
2252	}
2253
2254	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2255		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2256
2257	if (ib_register_device(&ibdev->ib_dev, NULL))
2258		goto err_steer_free_bitmap;
2259
2260	if (mlx4_ib_mad_init(ibdev))
2261		goto err_reg;
2262
2263	if (mlx4_ib_init_sriov(ibdev))
2264		goto err_mad;
2265
2266	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2267		if (!iboe->nb.notifier_call) {
2268			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2269			err = register_netdevice_notifier(&iboe->nb);
2270			if (err) {
2271				iboe->nb.notifier_call = NULL;
2272				goto err_notif;
2273			}
2274		}
2275		if (!iboe->nb_inet.notifier_call) {
2276			iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2277			err = register_inetaddr_notifier(&iboe->nb_inet);
2278			if (err) {
2279				iboe->nb_inet.notifier_call = NULL;
2280				goto err_notif;
2281			}
2282		}
2283#if IS_ENABLED(CONFIG_IPV6)
2284		if (!iboe->nb_inet6.notifier_call) {
2285			iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2286			err = register_inet6addr_notifier(&iboe->nb_inet6);
2287			if (err) {
2288				iboe->nb_inet6.notifier_call = NULL;
2289				goto err_notif;
2290			}
2291		}
2292#endif
2293		if (mlx4_ib_init_gid_table(ibdev))
2294			goto err_notif;
2295	}
2296
2297	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2298		if (device_create_file(&ibdev->ib_dev.dev,
2299				       mlx4_class_attributes[j]))
2300			goto err_notif;
2301	}
2302
2303	ibdev->ib_active = true;
2304
2305	if (mlx4_is_mfunc(ibdev->dev))
2306		init_pkeys(ibdev);
2307
2308	/* create paravirt contexts for any VFs which are active */
2309	if (mlx4_is_master(ibdev->dev)) {
2310		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2311			if (j == mlx4_master_func_num(ibdev->dev))
2312				continue;
2313			if (mlx4_is_slave_active(ibdev->dev, j))
2314				do_slave_init(ibdev, j, 1);
2315		}
2316	}
2317	return ibdev;
2318
2319err_notif:
2320	if (ibdev->iboe.nb.notifier_call) {
2321		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2322			pr_warn("failure unregistering notifier\n");
2323		ibdev->iboe.nb.notifier_call = NULL;
2324	}
2325	if (ibdev->iboe.nb_inet.notifier_call) {
2326		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2327			pr_warn("failure unregistering notifier\n");
2328		ibdev->iboe.nb_inet.notifier_call = NULL;
2329	}
2330#if IS_ENABLED(CONFIG_IPV6)
2331	if (ibdev->iboe.nb_inet6.notifier_call) {
2332		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2333			pr_warn("failure unregistering notifier\n");
2334		ibdev->iboe.nb_inet6.notifier_call = NULL;
2335	}
2336#endif
2337	flush_workqueue(wq);
2338
2339	mlx4_ib_close_sriov(ibdev);
2340
2341err_mad:
2342	mlx4_ib_mad_cleanup(ibdev);
2343
2344err_reg:
2345	ib_unregister_device(&ibdev->ib_dev);
2346
2347err_steer_free_bitmap:
2348	kfree(ibdev->ib_uc_qpns_bitmap);
2349
2350err_steer_qp_release:
2351	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2352		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2353				      ibdev->steer_qpn_count);
2354err_counter:
2355	for (; i; --i)
2356		if (ibdev->counters[i - 1] != -1)
2357			mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2358
2359err_map:
2360	iounmap(ibdev->uar_map);
2361
2362err_uar:
2363	mlx4_uar_free(dev, &ibdev->priv_uar);
2364
2365err_pd:
2366	mlx4_pd_free(dev, ibdev->priv_pdn);
2367
2368err_dealloc:
2369	ib_dealloc_device(&ibdev->ib_dev);
2370
2371	return NULL;
2372}
2373
2374int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2375{
2376	int offset;
2377
2378	WARN_ON(!dev->ib_uc_qpns_bitmap);
2379
2380	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2381					 dev->steer_qpn_count,
2382					 get_count_order(count));
2383	if (offset < 0)
2384		return offset;
2385
2386	*qpn = dev->steer_qpn_base + offset;
2387	return 0;
2388}
2389
2390void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2391{
2392	if (!qpn ||
2393	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2394		return;
2395
2396	BUG_ON(qpn < dev->steer_qpn_base);
2397
2398	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2399			      qpn - dev->steer_qpn_base,
2400			      get_count_order(count));
2401}
2402
2403int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2404			 int is_attach)
2405{
2406	int err;
2407	size_t flow_size;
2408	struct ib_flow_attr *flow = NULL;
2409	struct ib_flow_spec_ib *ib_spec;
2410
2411	if (is_attach) {
2412		flow_size = sizeof(struct ib_flow_attr) +
2413			    sizeof(struct ib_flow_spec_ib);
2414		flow = kzalloc(flow_size, GFP_KERNEL);
2415		if (!flow)
2416			return -ENOMEM;
2417		flow->port = mqp->port;
2418		flow->num_of_specs = 1;
2419		flow->size = flow_size;
2420		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2421		ib_spec->type = IB_FLOW_SPEC_IB;
2422		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2423		/* Add an empty rule for IB L2 */
2424		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2425
2426		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2427					    IB_FLOW_DOMAIN_NIC,
2428					    MLX4_FS_REGULAR,
2429					    &mqp->reg_id);
2430	} else {
2431		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2432	}
2433	kfree(flow);
2434	return err;
2435}
2436
2437static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2438{
2439	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2440	int p;
2441
2442	ibdev->ib_active = false;
2443	flush_workqueue(wq);
2444
2445	mlx4_ib_close_sriov(ibdev);
2446	mlx4_ib_mad_cleanup(ibdev);
2447	ib_unregister_device(&ibdev->ib_dev);
2448	if (ibdev->iboe.nb.notifier_call) {
2449		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2450			pr_warn("failure unregistering notifier\n");
2451		ibdev->iboe.nb.notifier_call = NULL;
2452	}
2453
2454	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2455		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2456				      ibdev->steer_qpn_count);
2457		kfree(ibdev->ib_uc_qpns_bitmap);
2458	}
2459
2460	if (ibdev->iboe.nb_inet.notifier_call) {
2461		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2462			pr_warn("failure unregistering notifier\n");
2463		ibdev->iboe.nb_inet.notifier_call = NULL;
2464	}
2465#if IS_ENABLED(CONFIG_IPV6)
2466	if (ibdev->iboe.nb_inet6.notifier_call) {
2467		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2468			pr_warn("failure unregistering notifier\n");
2469		ibdev->iboe.nb_inet6.notifier_call = NULL;
2470	}
2471#endif
2472
2473	iounmap(ibdev->uar_map);
2474	for (p = 0; p < ibdev->num_ports; ++p)
2475		if (ibdev->counters[p] != -1)
2476			mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2477	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2478		mlx4_CLOSE_PORT(dev, p);
2479
2480	mlx4_ib_free_eqs(dev, ibdev);
2481
2482	mlx4_uar_free(dev, &ibdev->priv_uar);
2483	mlx4_pd_free(dev, ibdev->priv_pdn);
2484	ib_dealloc_device(&ibdev->ib_dev);
2485}
2486
2487static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2488{
2489	struct mlx4_ib_demux_work **dm = NULL;
2490	struct mlx4_dev *dev = ibdev->dev;
2491	int i;
2492	unsigned long flags;
2493	struct mlx4_active_ports actv_ports;
2494	unsigned int ports;
2495	unsigned int first_port;
2496
2497	if (!mlx4_is_master(dev))
2498		return;
2499
2500	actv_ports = mlx4_get_active_ports(dev, slave);
2501	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2502	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2503
2504	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2505	if (!dm) {
2506		pr_err("failed to allocate memory for tunneling qp update\n");
2507		goto out;
2508	}
2509
2510	for (i = 0; i < ports; i++) {
2511		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2512		if (!dm[i]) {
2513			pr_err("failed to allocate memory for tunneling qp update work struct\n");
2514			for (i = 0; i < dev->caps.num_ports; i++) {
2515				if (dm[i])
2516					kfree(dm[i]);
2517			}
2518			goto out;
2519		}
2520	}
2521	/* initialize or tear down tunnel QPs for the slave */
2522	for (i = 0; i < ports; i++) {
2523		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2524		dm[i]->port = first_port + i + 1;
2525		dm[i]->slave = slave;
2526		dm[i]->do_init = do_init;
2527		dm[i]->dev = ibdev;
2528		spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2529		if (!ibdev->sriov.is_going_down)
2530			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2531		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2532	}
2533out:
2534	kfree(dm);
2535	return;
2536}
2537
2538static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2539			  enum mlx4_dev_event event, unsigned long param)
2540{
2541	struct ib_event ibev;
2542	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2543	struct mlx4_eqe *eqe = NULL;
2544	struct ib_event_work *ew;
2545	int p = 0;
2546
2547	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2548		eqe = (struct mlx4_eqe *)param;
2549	else
2550		p = (int) param;
2551
2552	switch (event) {
2553	case MLX4_DEV_EVENT_PORT_UP:
2554		if (p > ibdev->num_ports)
2555			return;
2556		if (mlx4_is_master(dev) &&
2557		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2558			IB_LINK_LAYER_INFINIBAND) {
2559			mlx4_ib_invalidate_all_guid_record(ibdev, p);
2560		}
2561		ibev.event = IB_EVENT_PORT_ACTIVE;
2562		break;
2563
2564	case MLX4_DEV_EVENT_PORT_DOWN:
2565		if (p > ibdev->num_ports)
2566			return;
2567		ibev.event = IB_EVENT_PORT_ERR;
2568		break;
2569
2570	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2571		ibdev->ib_active = false;
2572		ibev.event = IB_EVENT_DEVICE_FATAL;
2573		break;
2574
2575	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2576		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2577		if (!ew) {
2578			pr_err("failed to allocate memory for events work\n");
2579			break;
2580		}
2581
2582		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2583		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2584		ew->ib_dev = ibdev;
2585		/* need to queue only for port owner, which uses GEN_EQE */
2586		if (mlx4_is_master(dev))
2587			queue_work(wq, &ew->work);
2588		else
2589			handle_port_mgmt_change_event(&ew->work);
2590		return;
2591
2592	case MLX4_DEV_EVENT_SLAVE_INIT:
2593		/* here, p is the slave id */
2594		do_slave_init(ibdev, p, 1);
2595		return;
2596
2597	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2598		/* here, p is the slave id */
2599		do_slave_init(ibdev, p, 0);
2600		return;
2601
2602	default:
2603		return;
2604	}
2605
2606	ibev.device	      = ibdev_ptr;
2607	ibev.element.port_num = (u8) p;
2608
2609	ib_dispatch_event(&ibev);
2610}
2611
2612static struct mlx4_interface mlx4_ib_interface = {
2613	.add		= mlx4_ib_add,
2614	.remove		= mlx4_ib_remove,
2615	.event		= mlx4_ib_event,
2616	.protocol	= MLX4_PROT_IB_IPV6
2617};
2618
2619static int __init mlx4_ib_init(void)
2620{
2621	int err;
2622
2623	wq = create_singlethread_workqueue("mlx4_ib");
2624	if (!wq)
2625		return -ENOMEM;
2626
2627	err = mlx4_ib_mcg_init();
2628	if (err)
2629		goto clean_wq;
2630
2631	err = mlx4_register_interface(&mlx4_ib_interface);
2632	if (err)
2633		goto clean_mcg;
2634
2635	return 0;
2636
2637clean_mcg:
2638	mlx4_ib_mcg_destroy();
2639
2640clean_wq:
2641	destroy_workqueue(wq);
2642	return err;
2643}
2644
2645static void __exit mlx4_ib_cleanup(void)
2646{
2647	mlx4_unregister_interface(&mlx4_ib_interface);
2648	mlx4_ib_mcg_destroy();
2649	destroy_workqueue(wq);
2650}
2651
2652module_init(mlx4_ib_init);
2653module_exit(mlx4_ib_cleanup);
2654