2 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
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:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
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.
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
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>
43 #include <rdma/ib_smi.h>
44 #include <rdma/ib_user_verbs.h>
45 #include <rdma/ib_addr.h>
47 #include <linux/mlx4/driver.h>
48 #include <linux/mlx4/cmd.h>
53 #define DRV_NAME MLX4_IB_DRV_NAME
54 #define DRV_VERSION "1.0"
55 #define DRV_RELDATE "April 4, 2008"
57 MODULE_AUTHOR("Roland Dreier");
58 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
59 MODULE_LICENSE("Dual BSD/GPL");
60 MODULE_VERSION(DRV_VERSION
);
62 int mlx4_ib_sm_guid_assign
= 1;
63 module_param_named(sm_guid_assign
, mlx4_ib_sm_guid_assign
, int, 0444);
64 MODULE_PARM_DESC(sm_guid_assign
, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
66 static const char mlx4_ib_version
[] =
67 DRV_NAME
": Mellanox ConnectX InfiniBand driver v"
68 DRV_VERSION
" (" DRV_RELDATE
")\n";
70 struct update_gid_work
{
71 struct work_struct work
;
72 union ib_gid gids
[128];
73 struct mlx4_ib_dev
*dev
;
77 static void do_slave_init(struct mlx4_ib_dev
*ibdev
, int slave
, int do_init
);
79 static struct workqueue_struct
*wq
;
81 static void init_query_mad(struct ib_smp
*mad
)
83 mad
->base_version
= 1;
84 mad
->mgmt_class
= IB_MGMT_CLASS_SUBN_LID_ROUTED
;
85 mad
->class_version
= 1;
86 mad
->method
= IB_MGMT_METHOD_GET
;
89 static union ib_gid zgid
;
91 static int mlx4_ib_query_device(struct ib_device
*ibdev
,
92 struct ib_device_attr
*props
)
94 struct mlx4_ib_dev
*dev
= to_mdev(ibdev
);
95 struct ib_smp
*in_mad
= NULL
;
96 struct ib_smp
*out_mad
= NULL
;
99 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
100 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
101 if (!in_mad
|| !out_mad
)
104 init_query_mad(in_mad
);
105 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
107 err
= mlx4_MAD_IFC(to_mdev(ibdev
), MLX4_MAD_IFC_IGNORE_KEYS
,
108 1, NULL
, NULL
, in_mad
, out_mad
);
112 memset(props
, 0, sizeof *props
);
114 props
->fw_ver
= dev
->dev
->caps
.fw_ver
;
115 props
->device_cap_flags
= IB_DEVICE_CHANGE_PHY_PORT
|
116 IB_DEVICE_PORT_ACTIVE_EVENT
|
117 IB_DEVICE_SYS_IMAGE_GUID
|
118 IB_DEVICE_RC_RNR_NAK_GEN
|
119 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK
;
120 if (dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR
)
121 props
->device_cap_flags
|= IB_DEVICE_BAD_PKEY_CNTR
;
122 if (dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR
)
123 props
->device_cap_flags
|= IB_DEVICE_BAD_QKEY_CNTR
;
124 if (dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_APM
)
125 props
->device_cap_flags
|= IB_DEVICE_AUTO_PATH_MIG
;
126 if (dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_UD_AV_PORT
)
127 props
->device_cap_flags
|= IB_DEVICE_UD_AV_PORT_ENFORCE
;
128 if (dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_IPOIB_CSUM
)
129 props
->device_cap_flags
|= IB_DEVICE_UD_IP_CSUM
;
130 if (dev
->dev
->caps
.max_gso_sz
&& dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_BLH
)
131 props
->device_cap_flags
|= IB_DEVICE_UD_TSO
;
132 if (dev
->dev
->caps
.bmme_flags
& MLX4_BMME_FLAG_RESERVED_LKEY
)
133 props
->device_cap_flags
|= IB_DEVICE_LOCAL_DMA_LKEY
;
134 if ((dev
->dev
->caps
.bmme_flags
& MLX4_BMME_FLAG_LOCAL_INV
) &&
135 (dev
->dev
->caps
.bmme_flags
& MLX4_BMME_FLAG_REMOTE_INV
) &&
136 (dev
->dev
->caps
.bmme_flags
& MLX4_BMME_FLAG_FAST_REG_WR
))
137 props
->device_cap_flags
|= IB_DEVICE_MEM_MGT_EXTENSIONS
;
138 if (dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_XRC
)
139 props
->device_cap_flags
|= IB_DEVICE_XRC
;
141 props
->vendor_id
= be32_to_cpup((__be32
*) (out_mad
->data
+ 36)) &
143 props
->vendor_part_id
= dev
->dev
->pdev
->device
;
144 props
->hw_ver
= be32_to_cpup((__be32
*) (out_mad
->data
+ 32));
145 memcpy(&props
->sys_image_guid
, out_mad
->data
+ 4, 8);
147 props
->max_mr_size
= ~0ull;
148 props
->page_size_cap
= dev
->dev
->caps
.page_size_cap
;
149 props
->max_qp
= dev
->dev
->caps
.num_qps
- dev
->dev
->caps
.reserved_qps
;
150 props
->max_qp_wr
= dev
->dev
->caps
.max_wqes
- MLX4_IB_SQ_MAX_SPARE
;
151 props
->max_sge
= min(dev
->dev
->caps
.max_sq_sg
,
152 dev
->dev
->caps
.max_rq_sg
);
153 props
->max_cq
= dev
->dev
->caps
.num_cqs
- dev
->dev
->caps
.reserved_cqs
;
154 props
->max_cqe
= dev
->dev
->caps
.max_cqes
;
155 props
->max_mr
= dev
->dev
->caps
.num_mpts
- dev
->dev
->caps
.reserved_mrws
;
156 props
->max_pd
= dev
->dev
->caps
.num_pds
- dev
->dev
->caps
.reserved_pds
;
157 props
->max_qp_rd_atom
= dev
->dev
->caps
.max_qp_dest_rdma
;
158 props
->max_qp_init_rd_atom
= dev
->dev
->caps
.max_qp_init_rdma
;
159 props
->max_res_rd_atom
= props
->max_qp_rd_atom
* props
->max_qp
;
160 props
->max_srq
= dev
->dev
->caps
.num_srqs
- dev
->dev
->caps
.reserved_srqs
;
161 props
->max_srq_wr
= dev
->dev
->caps
.max_srq_wqes
- 1;
162 props
->max_srq_sge
= dev
->dev
->caps
.max_srq_sge
;
163 props
->max_fast_reg_page_list_len
= MLX4_MAX_FAST_REG_PAGES
;
164 props
->local_ca_ack_delay
= dev
->dev
->caps
.local_ca_ack_delay
;
165 props
->atomic_cap
= dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_ATOMIC
?
166 IB_ATOMIC_HCA
: IB_ATOMIC_NONE
;
167 props
->masked_atomic_cap
= props
->atomic_cap
;
168 props
->max_pkeys
= dev
->dev
->caps
.pkey_table_len
[1];
169 props
->max_mcast_grp
= dev
->dev
->caps
.num_mgms
+ dev
->dev
->caps
.num_amgms
;
170 props
->max_mcast_qp_attach
= dev
->dev
->caps
.num_qp_per_mgm
;
171 props
->max_total_mcast_qp_attach
= props
->max_mcast_qp_attach
*
172 props
->max_mcast_grp
;
173 props
->max_map_per_fmr
= dev
->dev
->caps
.max_fmr_maps
;
182 static enum rdma_link_layer
183 mlx4_ib_port_link_layer(struct ib_device
*device
, u8 port_num
)
185 struct mlx4_dev
*dev
= to_mdev(device
)->dev
;
187 return dev
->caps
.port_mask
[port_num
] == MLX4_PORT_TYPE_IB
?
188 IB_LINK_LAYER_INFINIBAND
: IB_LINK_LAYER_ETHERNET
;
191 static int ib_link_query_port(struct ib_device
*ibdev
, u8 port
,
192 struct ib_port_attr
*props
, int netw_view
)
194 struct ib_smp
*in_mad
= NULL
;
195 struct ib_smp
*out_mad
= NULL
;
196 int ext_active_speed
;
197 int mad_ifc_flags
= MLX4_MAD_IFC_IGNORE_KEYS
;
200 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
201 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
202 if (!in_mad
|| !out_mad
)
205 init_query_mad(in_mad
);
206 in_mad
->attr_id
= IB_SMP_ATTR_PORT_INFO
;
207 in_mad
->attr_mod
= cpu_to_be32(port
);
209 if (mlx4_is_mfunc(to_mdev(ibdev
)->dev
) && netw_view
)
210 mad_ifc_flags
|= MLX4_MAD_IFC_NET_VIEW
;
212 err
= mlx4_MAD_IFC(to_mdev(ibdev
), mad_ifc_flags
, port
, NULL
, NULL
,
218 props
->lid
= be16_to_cpup((__be16
*) (out_mad
->data
+ 16));
219 props
->lmc
= out_mad
->data
[34] & 0x7;
220 props
->sm_lid
= be16_to_cpup((__be16
*) (out_mad
->data
+ 18));
221 props
->sm_sl
= out_mad
->data
[36] & 0xf;
222 props
->state
= out_mad
->data
[32] & 0xf;
223 props
->phys_state
= out_mad
->data
[33] >> 4;
224 props
->port_cap_flags
= be32_to_cpup((__be32
*) (out_mad
->data
+ 20));
226 props
->gid_tbl_len
= out_mad
->data
[50];
228 props
->gid_tbl_len
= to_mdev(ibdev
)->dev
->caps
.gid_table_len
[port
];
229 props
->max_msg_sz
= to_mdev(ibdev
)->dev
->caps
.max_msg_sz
;
230 props
->pkey_tbl_len
= to_mdev(ibdev
)->dev
->caps
.pkey_table_len
[port
];
231 props
->bad_pkey_cntr
= be16_to_cpup((__be16
*) (out_mad
->data
+ 46));
232 props
->qkey_viol_cntr
= be16_to_cpup((__be16
*) (out_mad
->data
+ 48));
233 props
->active_width
= out_mad
->data
[31] & 0xf;
234 props
->active_speed
= out_mad
->data
[35] >> 4;
235 props
->max_mtu
= out_mad
->data
[41] & 0xf;
236 props
->active_mtu
= out_mad
->data
[36] >> 4;
237 props
->subnet_timeout
= out_mad
->data
[51] & 0x1f;
238 props
->max_vl_num
= out_mad
->data
[37] >> 4;
239 props
->init_type_reply
= out_mad
->data
[41] >> 4;
241 /* Check if extended speeds (EDR/FDR/...) are supported */
242 if (props
->port_cap_flags
& IB_PORT_EXTENDED_SPEEDS_SUP
) {
243 ext_active_speed
= out_mad
->data
[62] >> 4;
245 switch (ext_active_speed
) {
247 props
->active_speed
= IB_SPEED_FDR
;
250 props
->active_speed
= IB_SPEED_EDR
;
255 /* If reported active speed is QDR, check if is FDR-10 */
256 if (props
->active_speed
== IB_SPEED_QDR
) {
257 init_query_mad(in_mad
);
258 in_mad
->attr_id
= MLX4_ATTR_EXTENDED_PORT_INFO
;
259 in_mad
->attr_mod
= cpu_to_be32(port
);
261 err
= mlx4_MAD_IFC(to_mdev(ibdev
), mad_ifc_flags
, port
,
262 NULL
, NULL
, in_mad
, out_mad
);
266 /* Checking LinkSpeedActive for FDR-10 */
267 if (out_mad
->data
[15] & 0x1)
268 props
->active_speed
= IB_SPEED_FDR10
;
271 /* Avoid wrong speed value returned by FW if the IB link is down. */
272 if (props
->state
== IB_PORT_DOWN
)
273 props
->active_speed
= IB_SPEED_SDR
;
281 static u8
state_to_phys_state(enum ib_port_state state
)
283 return state
== IB_PORT_ACTIVE
? 5 : 3;
286 static int eth_link_query_port(struct ib_device
*ibdev
, u8 port
,
287 struct ib_port_attr
*props
, int netw_view
)
290 struct mlx4_ib_dev
*mdev
= to_mdev(ibdev
);
291 struct mlx4_ib_iboe
*iboe
= &mdev
->iboe
;
292 struct net_device
*ndev
;
294 struct mlx4_cmd_mailbox
*mailbox
;
297 mailbox
= mlx4_alloc_cmd_mailbox(mdev
->dev
);
299 return PTR_ERR(mailbox
);
301 err
= mlx4_cmd_box(mdev
->dev
, 0, mailbox
->dma
, port
, 0,
302 MLX4_CMD_QUERY_PORT
, MLX4_CMD_TIME_CLASS_B
,
307 props
->active_width
= (((u8
*)mailbox
->buf
)[5] == 0x40) ?
308 IB_WIDTH_4X
: IB_WIDTH_1X
;
309 props
->active_speed
= IB_SPEED_QDR
;
310 props
->port_cap_flags
= IB_PORT_CM_SUP
;
311 props
->gid_tbl_len
= mdev
->dev
->caps
.gid_table_len
[port
];
312 props
->max_msg_sz
= mdev
->dev
->caps
.max_msg_sz
;
313 props
->pkey_tbl_len
= 1;
314 props
->max_mtu
= IB_MTU_4096
;
315 props
->max_vl_num
= 2;
316 props
->state
= IB_PORT_DOWN
;
317 props
->phys_state
= state_to_phys_state(props
->state
);
318 props
->active_mtu
= IB_MTU_256
;
319 spin_lock(&iboe
->lock
);
320 ndev
= iboe
->netdevs
[port
- 1];
324 tmp
= iboe_get_mtu(ndev
->mtu
);
325 props
->active_mtu
= tmp
? min(props
->max_mtu
, tmp
) : IB_MTU_256
;
327 props
->state
= (netif_running(ndev
) && netif_carrier_ok(ndev
)) ?
328 IB_PORT_ACTIVE
: IB_PORT_DOWN
;
329 props
->phys_state
= state_to_phys_state(props
->state
);
331 spin_unlock(&iboe
->lock
);
333 mlx4_free_cmd_mailbox(mdev
->dev
, mailbox
);
337 int __mlx4_ib_query_port(struct ib_device
*ibdev
, u8 port
,
338 struct ib_port_attr
*props
, int netw_view
)
342 memset(props
, 0, sizeof *props
);
344 err
= mlx4_ib_port_link_layer(ibdev
, port
) == IB_LINK_LAYER_INFINIBAND
?
345 ib_link_query_port(ibdev
, port
, props
, netw_view
) :
346 eth_link_query_port(ibdev
, port
, props
, netw_view
);
351 static int mlx4_ib_query_port(struct ib_device
*ibdev
, u8 port
,
352 struct ib_port_attr
*props
)
354 /* returns host view */
355 return __mlx4_ib_query_port(ibdev
, port
, props
, 0);
358 int __mlx4_ib_query_gid(struct ib_device
*ibdev
, u8 port
, int index
,
359 union ib_gid
*gid
, int netw_view
)
361 struct ib_smp
*in_mad
= NULL
;
362 struct ib_smp
*out_mad
= NULL
;
364 struct mlx4_ib_dev
*dev
= to_mdev(ibdev
);
366 int mad_ifc_flags
= MLX4_MAD_IFC_IGNORE_KEYS
;
368 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
369 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
370 if (!in_mad
|| !out_mad
)
373 init_query_mad(in_mad
);
374 in_mad
->attr_id
= IB_SMP_ATTR_PORT_INFO
;
375 in_mad
->attr_mod
= cpu_to_be32(port
);
377 if (mlx4_is_mfunc(dev
->dev
) && netw_view
)
378 mad_ifc_flags
|= MLX4_MAD_IFC_NET_VIEW
;
380 err
= mlx4_MAD_IFC(dev
, mad_ifc_flags
, port
, NULL
, NULL
, in_mad
, out_mad
);
384 memcpy(gid
->raw
, out_mad
->data
+ 8, 8);
386 if (mlx4_is_mfunc(dev
->dev
) && !netw_view
) {
388 /* For any index > 0, return the null guid */
395 init_query_mad(in_mad
);
396 in_mad
->attr_id
= IB_SMP_ATTR_GUID_INFO
;
397 in_mad
->attr_mod
= cpu_to_be32(index
/ 8);
399 err
= mlx4_MAD_IFC(dev
, mad_ifc_flags
, port
,
400 NULL
, NULL
, in_mad
, out_mad
);
404 memcpy(gid
->raw
+ 8, out_mad
->data
+ (index
% 8) * 8, 8);
408 memset(gid
->raw
+ 8, 0, 8);
414 static int iboe_query_gid(struct ib_device
*ibdev
, u8 port
, int index
,
417 struct mlx4_ib_dev
*dev
= to_mdev(ibdev
);
419 *gid
= dev
->iboe
.gid_table
[port
- 1][index
];
424 static int mlx4_ib_query_gid(struct ib_device
*ibdev
, u8 port
, int index
,
427 if (rdma_port_get_link_layer(ibdev
, port
) == IB_LINK_LAYER_INFINIBAND
)
428 return __mlx4_ib_query_gid(ibdev
, port
, index
, gid
, 0);
430 return iboe_query_gid(ibdev
, port
, index
, gid
);
433 int __mlx4_ib_query_pkey(struct ib_device
*ibdev
, u8 port
, u16 index
,
434 u16
*pkey
, int netw_view
)
436 struct ib_smp
*in_mad
= NULL
;
437 struct ib_smp
*out_mad
= NULL
;
438 int mad_ifc_flags
= MLX4_MAD_IFC_IGNORE_KEYS
;
441 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
442 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
443 if (!in_mad
|| !out_mad
)
446 init_query_mad(in_mad
);
447 in_mad
->attr_id
= IB_SMP_ATTR_PKEY_TABLE
;
448 in_mad
->attr_mod
= cpu_to_be32(index
/ 32);
450 if (mlx4_is_mfunc(to_mdev(ibdev
)->dev
) && netw_view
)
451 mad_ifc_flags
|= MLX4_MAD_IFC_NET_VIEW
;
453 err
= mlx4_MAD_IFC(to_mdev(ibdev
), mad_ifc_flags
, port
, NULL
, NULL
,
458 *pkey
= be16_to_cpu(((__be16
*) out_mad
->data
)[index
% 32]);
466 static int mlx4_ib_query_pkey(struct ib_device
*ibdev
, u8 port
, u16 index
, u16
*pkey
)
468 return __mlx4_ib_query_pkey(ibdev
, port
, index
, pkey
, 0);
471 static int mlx4_ib_modify_device(struct ib_device
*ibdev
, int mask
,
472 struct ib_device_modify
*props
)
474 struct mlx4_cmd_mailbox
*mailbox
;
477 if (mask
& ~IB_DEVICE_MODIFY_NODE_DESC
)
480 if (!(mask
& IB_DEVICE_MODIFY_NODE_DESC
))
483 if (mlx4_is_slave(to_mdev(ibdev
)->dev
))
486 spin_lock_irqsave(&to_mdev(ibdev
)->sm_lock
, flags
);
487 memcpy(ibdev
->node_desc
, props
->node_desc
, 64);
488 spin_unlock_irqrestore(&to_mdev(ibdev
)->sm_lock
, flags
);
491 * If possible, pass node desc to FW, so it can generate
492 * a 144 trap. If cmd fails, just ignore.
494 mailbox
= mlx4_alloc_cmd_mailbox(to_mdev(ibdev
)->dev
);
498 memset(mailbox
->buf
, 0, 256);
499 memcpy(mailbox
->buf
, props
->node_desc
, 64);
500 mlx4_cmd(to_mdev(ibdev
)->dev
, mailbox
->dma
, 1, 0,
501 MLX4_CMD_SET_NODE
, MLX4_CMD_TIME_CLASS_A
, MLX4_CMD_NATIVE
);
503 mlx4_free_cmd_mailbox(to_mdev(ibdev
)->dev
, mailbox
);
508 static int mlx4_SET_PORT(struct mlx4_ib_dev
*dev
, u8 port
, int reset_qkey_viols
,
511 struct mlx4_cmd_mailbox
*mailbox
;
513 u8 is_eth
= dev
->dev
->caps
.port_type
[port
] == MLX4_PORT_TYPE_ETH
;
515 mailbox
= mlx4_alloc_cmd_mailbox(dev
->dev
);
517 return PTR_ERR(mailbox
);
519 memset(mailbox
->buf
, 0, 256);
521 if (dev
->dev
->flags
& MLX4_FLAG_OLD_PORT_CMDS
) {
522 *(u8
*) mailbox
->buf
= !!reset_qkey_viols
<< 6;
523 ((__be32
*) mailbox
->buf
)[2] = cpu_to_be32(cap_mask
);
525 ((u8
*) mailbox
->buf
)[3] = !!reset_qkey_viols
;
526 ((__be32
*) mailbox
->buf
)[1] = cpu_to_be32(cap_mask
);
529 err
= mlx4_cmd(dev
->dev
, mailbox
->dma
, port
, is_eth
, MLX4_CMD_SET_PORT
,
530 MLX4_CMD_TIME_CLASS_B
, MLX4_CMD_NATIVE
);
532 mlx4_free_cmd_mailbox(dev
->dev
, mailbox
);
536 static int mlx4_ib_modify_port(struct ib_device
*ibdev
, u8 port
, int mask
,
537 struct ib_port_modify
*props
)
539 struct ib_port_attr attr
;
543 mutex_lock(&to_mdev(ibdev
)->cap_mask_mutex
);
545 err
= mlx4_ib_query_port(ibdev
, port
, &attr
);
549 cap_mask
= (attr
.port_cap_flags
| props
->set_port_cap_mask
) &
550 ~props
->clr_port_cap_mask
;
552 err
= mlx4_SET_PORT(to_mdev(ibdev
), port
,
553 !!(mask
& IB_PORT_RESET_QKEY_CNTR
),
557 mutex_unlock(&to_mdev(ibdev
)->cap_mask_mutex
);
561 static struct ib_ucontext
*mlx4_ib_alloc_ucontext(struct ib_device
*ibdev
,
562 struct ib_udata
*udata
)
564 struct mlx4_ib_dev
*dev
= to_mdev(ibdev
);
565 struct mlx4_ib_ucontext
*context
;
566 struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3
;
567 struct mlx4_ib_alloc_ucontext_resp resp
;
571 return ERR_PTR(-EAGAIN
);
573 if (ibdev
->uverbs_abi_ver
== MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION
) {
574 resp_v3
.qp_tab_size
= dev
->dev
->caps
.num_qps
;
575 resp_v3
.bf_reg_size
= dev
->dev
->caps
.bf_reg_size
;
576 resp_v3
.bf_regs_per_page
= dev
->dev
->caps
.bf_regs_per_page
;
578 resp
.dev_caps
= dev
->dev
->caps
.userspace_caps
;
579 resp
.qp_tab_size
= dev
->dev
->caps
.num_qps
;
580 resp
.bf_reg_size
= dev
->dev
->caps
.bf_reg_size
;
581 resp
.bf_regs_per_page
= dev
->dev
->caps
.bf_regs_per_page
;
582 resp
.cqe_size
= dev
->dev
->caps
.cqe_size
;
585 context
= kmalloc(sizeof *context
, GFP_KERNEL
);
587 return ERR_PTR(-ENOMEM
);
589 err
= mlx4_uar_alloc(to_mdev(ibdev
)->dev
, &context
->uar
);
595 INIT_LIST_HEAD(&context
->db_page_list
);
596 mutex_init(&context
->db_page_mutex
);
598 if (ibdev
->uverbs_abi_ver
== MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION
)
599 err
= ib_copy_to_udata(udata
, &resp_v3
, sizeof(resp_v3
));
601 err
= ib_copy_to_udata(udata
, &resp
, sizeof(resp
));
604 mlx4_uar_free(to_mdev(ibdev
)->dev
, &context
->uar
);
606 return ERR_PTR(-EFAULT
);
609 return &context
->ibucontext
;
612 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext
*ibcontext
)
614 struct mlx4_ib_ucontext
*context
= to_mucontext(ibcontext
);
616 mlx4_uar_free(to_mdev(ibcontext
->device
)->dev
, &context
->uar
);
622 static int mlx4_ib_mmap(struct ib_ucontext
*context
, struct vm_area_struct
*vma
)
624 struct mlx4_ib_dev
*dev
= to_mdev(context
->device
);
626 if (vma
->vm_end
- vma
->vm_start
!= PAGE_SIZE
)
629 if (vma
->vm_pgoff
== 0) {
630 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
632 if (io_remap_pfn_range(vma
, vma
->vm_start
,
633 to_mucontext(context
)->uar
.pfn
,
634 PAGE_SIZE
, vma
->vm_page_prot
))
636 } else if (vma
->vm_pgoff
== 1 && dev
->dev
->caps
.bf_reg_size
!= 0) {
637 vma
->vm_page_prot
= pgprot_writecombine(vma
->vm_page_prot
);
639 if (io_remap_pfn_range(vma
, vma
->vm_start
,
640 to_mucontext(context
)->uar
.pfn
+
641 dev
->dev
->caps
.num_uars
,
642 PAGE_SIZE
, vma
->vm_page_prot
))
650 static struct ib_pd
*mlx4_ib_alloc_pd(struct ib_device
*ibdev
,
651 struct ib_ucontext
*context
,
652 struct ib_udata
*udata
)
654 struct mlx4_ib_pd
*pd
;
657 pd
= kmalloc(sizeof *pd
, GFP_KERNEL
);
659 return ERR_PTR(-ENOMEM
);
661 err
= mlx4_pd_alloc(to_mdev(ibdev
)->dev
, &pd
->pdn
);
668 if (ib_copy_to_udata(udata
, &pd
->pdn
, sizeof (__u32
))) {
669 mlx4_pd_free(to_mdev(ibdev
)->dev
, pd
->pdn
);
671 return ERR_PTR(-EFAULT
);
677 static int mlx4_ib_dealloc_pd(struct ib_pd
*pd
)
679 mlx4_pd_free(to_mdev(pd
->device
)->dev
, to_mpd(pd
)->pdn
);
685 static struct ib_xrcd
*mlx4_ib_alloc_xrcd(struct ib_device
*ibdev
,
686 struct ib_ucontext
*context
,
687 struct ib_udata
*udata
)
689 struct mlx4_ib_xrcd
*xrcd
;
692 if (!(to_mdev(ibdev
)->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_XRC
))
693 return ERR_PTR(-ENOSYS
);
695 xrcd
= kmalloc(sizeof *xrcd
, GFP_KERNEL
);
697 return ERR_PTR(-ENOMEM
);
699 err
= mlx4_xrcd_alloc(to_mdev(ibdev
)->dev
, &xrcd
->xrcdn
);
703 xrcd
->pd
= ib_alloc_pd(ibdev
);
704 if (IS_ERR(xrcd
->pd
)) {
705 err
= PTR_ERR(xrcd
->pd
);
709 xrcd
->cq
= ib_create_cq(ibdev
, NULL
, NULL
, xrcd
, 1, 0);
710 if (IS_ERR(xrcd
->cq
)) {
711 err
= PTR_ERR(xrcd
->cq
);
715 return &xrcd
->ibxrcd
;
718 ib_dealloc_pd(xrcd
->pd
);
720 mlx4_xrcd_free(to_mdev(ibdev
)->dev
, xrcd
->xrcdn
);
726 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd
*xrcd
)
728 ib_destroy_cq(to_mxrcd(xrcd
)->cq
);
729 ib_dealloc_pd(to_mxrcd(xrcd
)->pd
);
730 mlx4_xrcd_free(to_mdev(xrcd
->device
)->dev
, to_mxrcd(xrcd
)->xrcdn
);
736 static int add_gid_entry(struct ib_qp
*ibqp
, union ib_gid
*gid
)
738 struct mlx4_ib_qp
*mqp
= to_mqp(ibqp
);
739 struct mlx4_ib_dev
*mdev
= to_mdev(ibqp
->device
);
740 struct mlx4_ib_gid_entry
*ge
;
742 ge
= kzalloc(sizeof *ge
, GFP_KERNEL
);
747 if (mlx4_ib_add_mc(mdev
, mqp
, gid
)) {
748 ge
->port
= mqp
->port
;
752 mutex_lock(&mqp
->mutex
);
753 list_add_tail(&ge
->list
, &mqp
->gid_list
);
754 mutex_unlock(&mqp
->mutex
);
759 int mlx4_ib_add_mc(struct mlx4_ib_dev
*mdev
, struct mlx4_ib_qp
*mqp
,
763 struct net_device
*ndev
;
769 spin_lock(&mdev
->iboe
.lock
);
770 ndev
= mdev
->iboe
.netdevs
[mqp
->port
- 1];
773 spin_unlock(&mdev
->iboe
.lock
);
776 rdma_get_mcast_mac((struct in6_addr
*)gid
, mac
);
778 dev_mc_add(mdev
->iboe
.netdevs
[mqp
->port
- 1], mac
);
787 struct mlx4_ib_steering
{
788 struct list_head list
;
793 static int mlx4_ib_mcg_attach(struct ib_qp
*ibqp
, union ib_gid
*gid
, u16 lid
)
796 struct mlx4_ib_dev
*mdev
= to_mdev(ibqp
->device
);
797 struct mlx4_ib_qp
*mqp
= to_mqp(ibqp
);
799 struct mlx4_ib_steering
*ib_steering
= NULL
;
801 if (mdev
->dev
->caps
.steering_mode
==
802 MLX4_STEERING_MODE_DEVICE_MANAGED
) {
803 ib_steering
= kmalloc(sizeof(*ib_steering
), GFP_KERNEL
);
808 err
= mlx4_multicast_attach(mdev
->dev
, &mqp
->mqp
, gid
->raw
, mqp
->port
,
810 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK
),
811 MLX4_PROT_IB_IPV6
, ®_id
);
815 err
= add_gid_entry(ibqp
, gid
);
820 memcpy(ib_steering
->gid
.raw
, gid
->raw
, 16);
821 ib_steering
->reg_id
= reg_id
;
822 mutex_lock(&mqp
->mutex
);
823 list_add(&ib_steering
->list
, &mqp
->steering_rules
);
824 mutex_unlock(&mqp
->mutex
);
829 mlx4_multicast_detach(mdev
->dev
, &mqp
->mqp
, gid
->raw
,
830 MLX4_PROT_IB_IPV6
, reg_id
);
837 static struct mlx4_ib_gid_entry
*find_gid_entry(struct mlx4_ib_qp
*qp
, u8
*raw
)
839 struct mlx4_ib_gid_entry
*ge
;
840 struct mlx4_ib_gid_entry
*tmp
;
841 struct mlx4_ib_gid_entry
*ret
= NULL
;
843 list_for_each_entry_safe(ge
, tmp
, &qp
->gid_list
, list
) {
844 if (!memcmp(raw
, ge
->gid
.raw
, 16)) {
853 static int mlx4_ib_mcg_detach(struct ib_qp
*ibqp
, union ib_gid
*gid
, u16 lid
)
856 struct mlx4_ib_dev
*mdev
= to_mdev(ibqp
->device
);
857 struct mlx4_ib_qp
*mqp
= to_mqp(ibqp
);
859 struct net_device
*ndev
;
860 struct mlx4_ib_gid_entry
*ge
;
863 if (mdev
->dev
->caps
.steering_mode
==
864 MLX4_STEERING_MODE_DEVICE_MANAGED
) {
865 struct mlx4_ib_steering
*ib_steering
;
867 mutex_lock(&mqp
->mutex
);
868 list_for_each_entry(ib_steering
, &mqp
->steering_rules
, list
) {
869 if (!memcmp(ib_steering
->gid
.raw
, gid
->raw
, 16)) {
870 list_del(&ib_steering
->list
);
874 mutex_unlock(&mqp
->mutex
);
875 if (&ib_steering
->list
== &mqp
->steering_rules
) {
876 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
879 reg_id
= ib_steering
->reg_id
;
883 err
= mlx4_multicast_detach(mdev
->dev
, &mqp
->mqp
, gid
->raw
,
884 MLX4_PROT_IB_IPV6
, reg_id
);
888 mutex_lock(&mqp
->mutex
);
889 ge
= find_gid_entry(mqp
, gid
->raw
);
891 spin_lock(&mdev
->iboe
.lock
);
892 ndev
= ge
->added
? mdev
->iboe
.netdevs
[ge
->port
- 1] : NULL
;
895 spin_unlock(&mdev
->iboe
.lock
);
896 rdma_get_mcast_mac((struct in6_addr
*)gid
, mac
);
899 dev_mc_del(mdev
->iboe
.netdevs
[ge
->port
- 1], mac
);
906 pr_warn("could not find mgid entry\n");
908 mutex_unlock(&mqp
->mutex
);
913 static int init_node_data(struct mlx4_ib_dev
*dev
)
915 struct ib_smp
*in_mad
= NULL
;
916 struct ib_smp
*out_mad
= NULL
;
917 int mad_ifc_flags
= MLX4_MAD_IFC_IGNORE_KEYS
;
920 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
921 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
922 if (!in_mad
|| !out_mad
)
925 init_query_mad(in_mad
);
926 in_mad
->attr_id
= IB_SMP_ATTR_NODE_DESC
;
927 if (mlx4_is_master(dev
->dev
))
928 mad_ifc_flags
|= MLX4_MAD_IFC_NET_VIEW
;
930 err
= mlx4_MAD_IFC(dev
, mad_ifc_flags
, 1, NULL
, NULL
, in_mad
, out_mad
);
934 memcpy(dev
->ib_dev
.node_desc
, out_mad
->data
, 64);
936 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
938 err
= mlx4_MAD_IFC(dev
, mad_ifc_flags
, 1, NULL
, NULL
, in_mad
, out_mad
);
942 dev
->dev
->rev_id
= be32_to_cpup((__be32
*) (out_mad
->data
+ 32));
943 memcpy(&dev
->ib_dev
.node_guid
, out_mad
->data
+ 12, 8);
951 static ssize_t
show_hca(struct device
*device
, struct device_attribute
*attr
,
954 struct mlx4_ib_dev
*dev
=
955 container_of(device
, struct mlx4_ib_dev
, ib_dev
.dev
);
956 return sprintf(buf
, "MT%d\n", dev
->dev
->pdev
->device
);
959 static ssize_t
show_fw_ver(struct device
*device
, struct device_attribute
*attr
,
962 struct mlx4_ib_dev
*dev
=
963 container_of(device
, struct mlx4_ib_dev
, ib_dev
.dev
);
964 return sprintf(buf
, "%d.%d.%d\n", (int) (dev
->dev
->caps
.fw_ver
>> 32),
965 (int) (dev
->dev
->caps
.fw_ver
>> 16) & 0xffff,
966 (int) dev
->dev
->caps
.fw_ver
& 0xffff);
969 static ssize_t
show_rev(struct device
*device
, struct device_attribute
*attr
,
972 struct mlx4_ib_dev
*dev
=
973 container_of(device
, struct mlx4_ib_dev
, ib_dev
.dev
);
974 return sprintf(buf
, "%x\n", dev
->dev
->rev_id
);
977 static ssize_t
show_board(struct device
*device
, struct device_attribute
*attr
,
980 struct mlx4_ib_dev
*dev
=
981 container_of(device
, struct mlx4_ib_dev
, ib_dev
.dev
);
982 return sprintf(buf
, "%.*s\n", MLX4_BOARD_ID_LEN
,
986 static DEVICE_ATTR(hw_rev
, S_IRUGO
, show_rev
, NULL
);
987 static DEVICE_ATTR(fw_ver
, S_IRUGO
, show_fw_ver
, NULL
);
988 static DEVICE_ATTR(hca_type
, S_IRUGO
, show_hca
, NULL
);
989 static DEVICE_ATTR(board_id
, S_IRUGO
, show_board
, NULL
);
991 static struct device_attribute
*mlx4_class_attributes
[] = {
998 static void mlx4_addrconf_ifid_eui48(u8
*eui
, u16 vlan_id
, struct net_device
*dev
)
1000 memcpy(eui
, dev
->dev_addr
, 3);
1001 memcpy(eui
+ 5, dev
->dev_addr
+ 3, 3);
1002 if (vlan_id
< 0x1000) {
1003 eui
[3] = vlan_id
>> 8;
1004 eui
[4] = vlan_id
& 0xff;
1012 static void update_gids_task(struct work_struct
*work
)
1014 struct update_gid_work
*gw
= container_of(work
, struct update_gid_work
, work
);
1015 struct mlx4_cmd_mailbox
*mailbox
;
1018 struct mlx4_dev
*dev
= gw
->dev
->dev
;
1020 mailbox
= mlx4_alloc_cmd_mailbox(dev
);
1021 if (IS_ERR(mailbox
)) {
1022 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox
));
1026 gids
= mailbox
->buf
;
1027 memcpy(gids
, gw
->gids
, sizeof gw
->gids
);
1029 err
= mlx4_cmd(dev
, mailbox
->dma
, MLX4_SET_PORT_GID_TABLE
<< 8 | gw
->port
,
1030 1, MLX4_CMD_SET_PORT
, MLX4_CMD_TIME_CLASS_B
,
1033 pr_warn("set port command failed\n");
1035 memcpy(gw
->dev
->iboe
.gid_table
[gw
->port
- 1], gw
->gids
, sizeof gw
->gids
);
1036 mlx4_ib_dispatch_event(gw
->dev
, gw
->port
, IB_EVENT_GID_CHANGE
);
1039 mlx4_free_cmd_mailbox(dev
, mailbox
);
1043 static int update_ipv6_gids(struct mlx4_ib_dev
*dev
, int port
, int clear
)
1045 struct net_device
*ndev
= dev
->iboe
.netdevs
[port
- 1];
1046 struct update_gid_work
*work
;
1047 struct net_device
*tmp
;
1054 int need_update
= 0;
1057 work
= kzalloc(sizeof *work
, GFP_ATOMIC
);
1061 hits
= kzalloc(128, GFP_ATOMIC
);
1068 for_each_netdev_rcu(&init_net
, tmp
) {
1069 if (ndev
&& (tmp
== ndev
|| rdma_vlan_dev_real_dev(tmp
) == ndev
)) {
1070 gid
.global
.subnet_prefix
= cpu_to_be64(0xfe80000000000000LL
);
1071 vid
= rdma_vlan_dev_vlan_id(tmp
);
1072 mlx4_addrconf_ifid_eui48(&gid
.raw
[8], vid
, ndev
);
1075 for (i
= 0; i
< 128; ++i
) {
1077 !memcmp(&dev
->iboe
.gid_table
[port
- 1][i
], &zgid
, sizeof zgid
))
1079 if (!memcmp(&dev
->iboe
.gid_table
[port
- 1][i
], &gid
, sizeof gid
)) {
1088 (memcmp(&dev
->iboe
.gid_table
[port
- 1][0],
1089 &gid
, sizeof gid
) ||
1090 !memcmp(&dev
->iboe
.gid_table
[port
- 1][0],
1091 &zgid
, sizeof gid
))) {
1092 dev
->iboe
.gid_table
[port
- 1][0] = gid
;
1095 } else if (free
>= 0) {
1096 dev
->iboe
.gid_table
[port
- 1][free
] = gid
;
1105 for (i
= 0; i
< 128; ++i
)
1107 if (memcmp(&dev
->iboe
.gid_table
[port
- 1][i
], &zgid
, sizeof zgid
))
1109 dev
->iboe
.gid_table
[port
- 1][i
] = zgid
;
1113 memcpy(work
->gids
, dev
->iboe
.gid_table
[port
- 1], sizeof work
->gids
);
1114 INIT_WORK(&work
->work
, update_gids_task
);
1117 queue_work(wq
, &work
->work
);
1129 static void handle_en_event(struct mlx4_ib_dev
*dev
, int port
, unsigned long event
)
1133 case NETDEV_CHANGEADDR
:
1134 update_ipv6_gids(dev
, port
, 0);
1138 update_ipv6_gids(dev
, port
, 1);
1139 dev
->iboe
.netdevs
[port
- 1] = NULL
;
1143 static void netdev_added(struct mlx4_ib_dev
*dev
, int port
)
1145 update_ipv6_gids(dev
, port
, 0);
1148 static void netdev_removed(struct mlx4_ib_dev
*dev
, int port
)
1150 update_ipv6_gids(dev
, port
, 1);
1153 static int mlx4_ib_netdev_event(struct notifier_block
*this, unsigned long event
,
1156 struct net_device
*dev
= ptr
;
1157 struct mlx4_ib_dev
*ibdev
;
1158 struct net_device
*oldnd
;
1159 struct mlx4_ib_iboe
*iboe
;
1162 if (!net_eq(dev_net(dev
), &init_net
))
1165 ibdev
= container_of(this, struct mlx4_ib_dev
, iboe
.nb
);
1166 iboe
= &ibdev
->iboe
;
1168 spin_lock(&iboe
->lock
);
1169 mlx4_foreach_ib_transport_port(port
, ibdev
->dev
) {
1170 oldnd
= iboe
->netdevs
[port
- 1];
1171 iboe
->netdevs
[port
- 1] =
1172 mlx4_get_protocol_dev(ibdev
->dev
, MLX4_PROT_ETH
, port
);
1173 if (oldnd
!= iboe
->netdevs
[port
- 1]) {
1174 if (iboe
->netdevs
[port
- 1])
1175 netdev_added(ibdev
, port
);
1177 netdev_removed(ibdev
, port
);
1181 if (dev
== iboe
->netdevs
[0] ||
1182 (iboe
->netdevs
[0] && rdma_vlan_dev_real_dev(dev
) == iboe
->netdevs
[0]))
1183 handle_en_event(ibdev
, 1, event
);
1184 else if (dev
== iboe
->netdevs
[1]
1185 || (iboe
->netdevs
[1] && rdma_vlan_dev_real_dev(dev
) == iboe
->netdevs
[1]))
1186 handle_en_event(ibdev
, 2, event
);
1188 spin_unlock(&iboe
->lock
);
1193 static void init_pkeys(struct mlx4_ib_dev
*ibdev
)
1199 if (mlx4_is_master(ibdev
->dev
)) {
1200 for (slave
= 0; slave
<= ibdev
->dev
->num_vfs
; ++slave
) {
1201 for (port
= 1; port
<= ibdev
->dev
->caps
.num_ports
; ++port
) {
1203 i
< ibdev
->dev
->phys_caps
.pkey_phys_table_len
[port
];
1205 ibdev
->pkeys
.virt2phys_pkey
[slave
][port
- 1][i
] =
1206 /* master has the identity virt2phys pkey mapping */
1207 (slave
== mlx4_master_func_num(ibdev
->dev
) || !i
) ? i
:
1208 ibdev
->dev
->phys_caps
.pkey_phys_table_len
[port
] - 1;
1209 mlx4_sync_pkey_table(ibdev
->dev
, slave
, port
, i
,
1210 ibdev
->pkeys
.virt2phys_pkey
[slave
][port
- 1][i
]);
1214 /* initialize pkey cache */
1215 for (port
= 1; port
<= ibdev
->dev
->caps
.num_ports
; ++port
) {
1217 i
< ibdev
->dev
->phys_caps
.pkey_phys_table_len
[port
];
1219 ibdev
->pkeys
.phys_pkey_cache
[port
-1][i
] =
1225 static void mlx4_ib_alloc_eqs(struct mlx4_dev
*dev
, struct mlx4_ib_dev
*ibdev
)
1228 int eq_per_port
= 0;
1233 /* Legacy mode or comp_pool is not large enough */
1234 if (dev
->caps
.comp_pool
== 0 ||
1235 dev
->caps
.num_ports
> dev
->caps
.comp_pool
)
1238 eq_per_port
= rounddown_pow_of_two(dev
->caps
.comp_pool
/
1239 dev
->caps
.num_ports
);
1243 mlx4_foreach_port(i
, dev
, MLX4_PORT_TYPE_IB
)
1244 added_eqs
+= eq_per_port
;
1246 total_eqs
= dev
->caps
.num_comp_vectors
+ added_eqs
;
1248 ibdev
->eq_table
= kzalloc(total_eqs
* sizeof(int), GFP_KERNEL
);
1249 if (!ibdev
->eq_table
)
1252 ibdev
->eq_added
= added_eqs
;
1255 mlx4_foreach_port(i
, dev
, MLX4_PORT_TYPE_IB
) {
1256 for (j
= 0; j
< eq_per_port
; j
++) {
1257 sprintf(name
, "mlx4-ib-%d-%d@%s",
1258 i
, j
, dev
->pdev
->bus
->name
);
1259 /* Set IRQ for specific name (per ring) */
1260 if (mlx4_assign_eq(dev
, name
, NULL
,
1261 &ibdev
->eq_table
[eq
])) {
1262 /* Use legacy (same as mlx4_en driver) */
1263 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq
);
1264 ibdev
->eq_table
[eq
] =
1265 (eq
% dev
->caps
.num_comp_vectors
);
1271 /* Fill the reset of the vector with legacy EQ */
1272 for (i
= 0, eq
= added_eqs
; i
< dev
->caps
.num_comp_vectors
; i
++)
1273 ibdev
->eq_table
[eq
++] = i
;
1275 /* Advertise the new number of EQs to clients */
1276 ibdev
->ib_dev
.num_comp_vectors
= total_eqs
;
1279 static void mlx4_ib_free_eqs(struct mlx4_dev
*dev
, struct mlx4_ib_dev
*ibdev
)
1283 /* no additional eqs were added */
1284 if (!ibdev
->eq_table
)
1287 /* Reset the advertised EQ number */
1288 ibdev
->ib_dev
.num_comp_vectors
= dev
->caps
.num_comp_vectors
;
1290 /* Free only the added eqs */
1291 for (i
= 0; i
< ibdev
->eq_added
; i
++) {
1292 /* Don't free legacy eqs if used */
1293 if (ibdev
->eq_table
[i
] <= dev
->caps
.num_comp_vectors
)
1295 mlx4_release_eq(dev
, ibdev
->eq_table
[i
]);
1298 kfree(ibdev
->eq_table
);
1301 static void *mlx4_ib_add(struct mlx4_dev
*dev
)
1303 struct mlx4_ib_dev
*ibdev
;
1307 struct mlx4_ib_iboe
*iboe
;
1309 pr_info_once("%s", mlx4_ib_version
);
1311 mlx4_foreach_non_ib_transport_port(i
, dev
)
1314 if (mlx4_is_mfunc(dev
) && num_ports
) {
1315 dev_err(&dev
->pdev
->dev
, "RoCE is not supported over SRIOV as yet\n");
1320 mlx4_foreach_ib_transport_port(i
, dev
)
1323 /* No point in registering a device with no ports... */
1327 ibdev
= (struct mlx4_ib_dev
*) ib_alloc_device(sizeof *ibdev
);
1329 dev_err(&dev
->pdev
->dev
, "Device struct alloc failed\n");
1333 iboe
= &ibdev
->iboe
;
1335 if (mlx4_pd_alloc(dev
, &ibdev
->priv_pdn
))
1338 if (mlx4_uar_alloc(dev
, &ibdev
->priv_uar
))
1341 ibdev
->uar_map
= ioremap((phys_addr_t
) ibdev
->priv_uar
.pfn
<< PAGE_SHIFT
,
1343 if (!ibdev
->uar_map
)
1345 MLX4_INIT_DOORBELL_LOCK(&ibdev
->uar_lock
);
1349 strlcpy(ibdev
->ib_dev
.name
, "mlx4_%d", IB_DEVICE_NAME_MAX
);
1350 ibdev
->ib_dev
.owner
= THIS_MODULE
;
1351 ibdev
->ib_dev
.node_type
= RDMA_NODE_IB_CA
;
1352 ibdev
->ib_dev
.local_dma_lkey
= dev
->caps
.reserved_lkey
;
1353 ibdev
->num_ports
= num_ports
;
1354 ibdev
->ib_dev
.phys_port_cnt
= ibdev
->num_ports
;
1355 ibdev
->ib_dev
.num_comp_vectors
= dev
->caps
.num_comp_vectors
;
1356 ibdev
->ib_dev
.dma_device
= &dev
->pdev
->dev
;
1358 if (dev
->caps
.userspace_caps
)
1359 ibdev
->ib_dev
.uverbs_abi_ver
= MLX4_IB_UVERBS_ABI_VERSION
;
1361 ibdev
->ib_dev
.uverbs_abi_ver
= MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION
;
1363 ibdev
->ib_dev
.uverbs_cmd_mask
=
1364 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT
) |
1365 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE
) |
1366 (1ull << IB_USER_VERBS_CMD_QUERY_PORT
) |
1367 (1ull << IB_USER_VERBS_CMD_ALLOC_PD
) |
1368 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD
) |
1369 (1ull << IB_USER_VERBS_CMD_REG_MR
) |
1370 (1ull << IB_USER_VERBS_CMD_DEREG_MR
) |
1371 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL
) |
1372 (1ull << IB_USER_VERBS_CMD_CREATE_CQ
) |
1373 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ
) |
1374 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ
) |
1375 (1ull << IB_USER_VERBS_CMD_CREATE_QP
) |
1376 (1ull << IB_USER_VERBS_CMD_MODIFY_QP
) |
1377 (1ull << IB_USER_VERBS_CMD_QUERY_QP
) |
1378 (1ull << IB_USER_VERBS_CMD_DESTROY_QP
) |
1379 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST
) |
1380 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST
) |
1381 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ
) |
1382 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ
) |
1383 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ
) |
1384 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ
) |
1385 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ
) |
1386 (1ull << IB_USER_VERBS_CMD_OPEN_QP
);
1388 ibdev
->ib_dev
.query_device
= mlx4_ib_query_device
;
1389 ibdev
->ib_dev
.query_port
= mlx4_ib_query_port
;
1390 ibdev
->ib_dev
.get_link_layer
= mlx4_ib_port_link_layer
;
1391 ibdev
->ib_dev
.query_gid
= mlx4_ib_query_gid
;
1392 ibdev
->ib_dev
.query_pkey
= mlx4_ib_query_pkey
;
1393 ibdev
->ib_dev
.modify_device
= mlx4_ib_modify_device
;
1394 ibdev
->ib_dev
.modify_port
= mlx4_ib_modify_port
;
1395 ibdev
->ib_dev
.alloc_ucontext
= mlx4_ib_alloc_ucontext
;
1396 ibdev
->ib_dev
.dealloc_ucontext
= mlx4_ib_dealloc_ucontext
;
1397 ibdev
->ib_dev
.mmap
= mlx4_ib_mmap
;
1398 ibdev
->ib_dev
.alloc_pd
= mlx4_ib_alloc_pd
;
1399 ibdev
->ib_dev
.dealloc_pd
= mlx4_ib_dealloc_pd
;
1400 ibdev
->ib_dev
.create_ah
= mlx4_ib_create_ah
;
1401 ibdev
->ib_dev
.query_ah
= mlx4_ib_query_ah
;
1402 ibdev
->ib_dev
.destroy_ah
= mlx4_ib_destroy_ah
;
1403 ibdev
->ib_dev
.create_srq
= mlx4_ib_create_srq
;
1404 ibdev
->ib_dev
.modify_srq
= mlx4_ib_modify_srq
;
1405 ibdev
->ib_dev
.query_srq
= mlx4_ib_query_srq
;
1406 ibdev
->ib_dev
.destroy_srq
= mlx4_ib_destroy_srq
;
1407 ibdev
->ib_dev
.post_srq_recv
= mlx4_ib_post_srq_recv
;
1408 ibdev
->ib_dev
.create_qp
= mlx4_ib_create_qp
;
1409 ibdev
->ib_dev
.modify_qp
= mlx4_ib_modify_qp
;
1410 ibdev
->ib_dev
.query_qp
= mlx4_ib_query_qp
;
1411 ibdev
->ib_dev
.destroy_qp
= mlx4_ib_destroy_qp
;
1412 ibdev
->ib_dev
.post_send
= mlx4_ib_post_send
;
1413 ibdev
->ib_dev
.post_recv
= mlx4_ib_post_recv
;
1414 ibdev
->ib_dev
.create_cq
= mlx4_ib_create_cq
;
1415 ibdev
->ib_dev
.modify_cq
= mlx4_ib_modify_cq
;
1416 ibdev
->ib_dev
.resize_cq
= mlx4_ib_resize_cq
;
1417 ibdev
->ib_dev
.destroy_cq
= mlx4_ib_destroy_cq
;
1418 ibdev
->ib_dev
.poll_cq
= mlx4_ib_poll_cq
;
1419 ibdev
->ib_dev
.req_notify_cq
= mlx4_ib_arm_cq
;
1420 ibdev
->ib_dev
.get_dma_mr
= mlx4_ib_get_dma_mr
;
1421 ibdev
->ib_dev
.reg_user_mr
= mlx4_ib_reg_user_mr
;
1422 ibdev
->ib_dev
.dereg_mr
= mlx4_ib_dereg_mr
;
1423 ibdev
->ib_dev
.alloc_fast_reg_mr
= mlx4_ib_alloc_fast_reg_mr
;
1424 ibdev
->ib_dev
.alloc_fast_reg_page_list
= mlx4_ib_alloc_fast_reg_page_list
;
1425 ibdev
->ib_dev
.free_fast_reg_page_list
= mlx4_ib_free_fast_reg_page_list
;
1426 ibdev
->ib_dev
.attach_mcast
= mlx4_ib_mcg_attach
;
1427 ibdev
->ib_dev
.detach_mcast
= mlx4_ib_mcg_detach
;
1428 ibdev
->ib_dev
.process_mad
= mlx4_ib_process_mad
;
1430 if (!mlx4_is_slave(ibdev
->dev
)) {
1431 ibdev
->ib_dev
.alloc_fmr
= mlx4_ib_fmr_alloc
;
1432 ibdev
->ib_dev
.map_phys_fmr
= mlx4_ib_map_phys_fmr
;
1433 ibdev
->ib_dev
.unmap_fmr
= mlx4_ib_unmap_fmr
;
1434 ibdev
->ib_dev
.dealloc_fmr
= mlx4_ib_fmr_dealloc
;
1437 if (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_XRC
) {
1438 ibdev
->ib_dev
.alloc_xrcd
= mlx4_ib_alloc_xrcd
;
1439 ibdev
->ib_dev
.dealloc_xrcd
= mlx4_ib_dealloc_xrcd
;
1440 ibdev
->ib_dev
.uverbs_cmd_mask
|=
1441 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD
) |
1442 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD
);
1445 mlx4_ib_alloc_eqs(dev
, ibdev
);
1447 spin_lock_init(&iboe
->lock
);
1449 if (init_node_data(ibdev
))
1452 for (i
= 0; i
< ibdev
->num_ports
; ++i
) {
1453 if (mlx4_ib_port_link_layer(&ibdev
->ib_dev
, i
+ 1) ==
1454 IB_LINK_LAYER_ETHERNET
) {
1455 err
= mlx4_counter_alloc(ibdev
->dev
, &ibdev
->counters
[i
]);
1457 ibdev
->counters
[i
] = -1;
1459 ibdev
->counters
[i
] = -1;
1462 spin_lock_init(&ibdev
->sm_lock
);
1463 mutex_init(&ibdev
->cap_mask_mutex
);
1465 if (ib_register_device(&ibdev
->ib_dev
, NULL
))
1468 if (mlx4_ib_mad_init(ibdev
))
1471 if (mlx4_ib_init_sriov(ibdev
))
1474 if (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_IBOE
&& !iboe
->nb
.notifier_call
) {
1475 iboe
->nb
.notifier_call
= mlx4_ib_netdev_event
;
1476 err
= register_netdevice_notifier(&iboe
->nb
);
1481 for (j
= 0; j
< ARRAY_SIZE(mlx4_class_attributes
); ++j
) {
1482 if (device_create_file(&ibdev
->ib_dev
.dev
,
1483 mlx4_class_attributes
[j
]))
1487 ibdev
->ib_active
= true;
1489 if (mlx4_is_mfunc(ibdev
->dev
))
1492 /* create paravirt contexts for any VFs which are active */
1493 if (mlx4_is_master(ibdev
->dev
)) {
1494 for (j
= 0; j
< MLX4_MFUNC_MAX
; j
++) {
1495 if (j
== mlx4_master_func_num(ibdev
->dev
))
1497 if (mlx4_is_slave_active(ibdev
->dev
, j
))
1498 do_slave_init(ibdev
, j
, 1);
1504 if (unregister_netdevice_notifier(&ibdev
->iboe
.nb
))
1505 pr_warn("failure unregistering notifier\n");
1506 flush_workqueue(wq
);
1509 mlx4_ib_close_sriov(ibdev
);
1512 mlx4_ib_mad_cleanup(ibdev
);
1515 ib_unregister_device(&ibdev
->ib_dev
);
1519 if (ibdev
->counters
[i
- 1] != -1)
1520 mlx4_counter_free(ibdev
->dev
, ibdev
->counters
[i
- 1]);
1523 iounmap(ibdev
->uar_map
);
1526 mlx4_uar_free(dev
, &ibdev
->priv_uar
);
1529 mlx4_pd_free(dev
, ibdev
->priv_pdn
);
1532 ib_dealloc_device(&ibdev
->ib_dev
);
1537 static void mlx4_ib_remove(struct mlx4_dev
*dev
, void *ibdev_ptr
)
1539 struct mlx4_ib_dev
*ibdev
= ibdev_ptr
;
1542 mlx4_ib_close_sriov(ibdev
);
1543 mlx4_ib_mad_cleanup(ibdev
);
1544 ib_unregister_device(&ibdev
->ib_dev
);
1545 if (ibdev
->iboe
.nb
.notifier_call
) {
1546 if (unregister_netdevice_notifier(&ibdev
->iboe
.nb
))
1547 pr_warn("failure unregistering notifier\n");
1548 ibdev
->iboe
.nb
.notifier_call
= NULL
;
1550 iounmap(ibdev
->uar_map
);
1551 for (p
= 0; p
< ibdev
->num_ports
; ++p
)
1552 if (ibdev
->counters
[p
] != -1)
1553 mlx4_counter_free(ibdev
->dev
, ibdev
->counters
[p
]);
1554 mlx4_foreach_port(p
, dev
, MLX4_PORT_TYPE_IB
)
1555 mlx4_CLOSE_PORT(dev
, p
);
1557 mlx4_ib_free_eqs(dev
, ibdev
);
1559 mlx4_uar_free(dev
, &ibdev
->priv_uar
);
1560 mlx4_pd_free(dev
, ibdev
->priv_pdn
);
1561 ib_dealloc_device(&ibdev
->ib_dev
);
1564 static void do_slave_init(struct mlx4_ib_dev
*ibdev
, int slave
, int do_init
)
1566 struct mlx4_ib_demux_work
**dm
= NULL
;
1567 struct mlx4_dev
*dev
= ibdev
->dev
;
1569 unsigned long flags
;
1571 if (!mlx4_is_master(dev
))
1574 dm
= kcalloc(dev
->caps
.num_ports
, sizeof *dm
, GFP_ATOMIC
);
1576 pr_err("failed to allocate memory for tunneling qp update\n");
1580 for (i
= 0; i
< dev
->caps
.num_ports
; i
++) {
1581 dm
[i
] = kmalloc(sizeof (struct mlx4_ib_demux_work
), GFP_ATOMIC
);
1583 pr_err("failed to allocate memory for tunneling qp update work struct\n");
1584 for (i
= 0; i
< dev
->caps
.num_ports
; i
++) {
1591 /* initialize or tear down tunnel QPs for the slave */
1592 for (i
= 0; i
< dev
->caps
.num_ports
; i
++) {
1593 INIT_WORK(&dm
[i
]->work
, mlx4_ib_tunnels_update_work
);
1594 dm
[i
]->port
= i
+ 1;
1595 dm
[i
]->slave
= slave
;
1596 dm
[i
]->do_init
= do_init
;
1598 spin_lock_irqsave(&ibdev
->sriov
.going_down_lock
, flags
);
1599 if (!ibdev
->sriov
.is_going_down
)
1600 queue_work(ibdev
->sriov
.demux
[i
].ud_wq
, &dm
[i
]->work
);
1601 spin_unlock_irqrestore(&ibdev
->sriov
.going_down_lock
, flags
);
1609 static void mlx4_ib_event(struct mlx4_dev
*dev
, void *ibdev_ptr
,
1610 enum mlx4_dev_event event
, unsigned long param
)
1612 struct ib_event ibev
;
1613 struct mlx4_ib_dev
*ibdev
= to_mdev((struct ib_device
*) ibdev_ptr
);
1614 struct mlx4_eqe
*eqe
= NULL
;
1615 struct ib_event_work
*ew
;
1618 if (event
== MLX4_DEV_EVENT_PORT_MGMT_CHANGE
)
1619 eqe
= (struct mlx4_eqe
*)param
;
1624 case MLX4_DEV_EVENT_PORT_UP
:
1625 if (p
> ibdev
->num_ports
)
1627 if (mlx4_is_master(dev
) &&
1628 rdma_port_get_link_layer(&ibdev
->ib_dev
, p
) ==
1629 IB_LINK_LAYER_INFINIBAND
) {
1630 mlx4_ib_invalidate_all_guid_record(ibdev
, p
);
1632 ibev
.event
= IB_EVENT_PORT_ACTIVE
;
1635 case MLX4_DEV_EVENT_PORT_DOWN
:
1636 if (p
> ibdev
->num_ports
)
1638 ibev
.event
= IB_EVENT_PORT_ERR
;
1641 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR
:
1642 ibdev
->ib_active
= false;
1643 ibev
.event
= IB_EVENT_DEVICE_FATAL
;
1646 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE
:
1647 ew
= kmalloc(sizeof *ew
, GFP_ATOMIC
);
1649 pr_err("failed to allocate memory for events work\n");
1653 INIT_WORK(&ew
->work
, handle_port_mgmt_change_event
);
1654 memcpy(&ew
->ib_eqe
, eqe
, sizeof *eqe
);
1656 /* need to queue only for port owner, which uses GEN_EQE */
1657 if (mlx4_is_master(dev
))
1658 queue_work(wq
, &ew
->work
);
1660 handle_port_mgmt_change_event(&ew
->work
);
1663 case MLX4_DEV_EVENT_SLAVE_INIT
:
1664 /* here, p is the slave id */
1665 do_slave_init(ibdev
, p
, 1);
1668 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN
:
1669 /* here, p is the slave id */
1670 do_slave_init(ibdev
, p
, 0);
1677 ibev
.device
= ibdev_ptr
;
1678 ibev
.element
.port_num
= (u8
) p
;
1680 ib_dispatch_event(&ibev
);
1683 static struct mlx4_interface mlx4_ib_interface
= {
1685 .remove
= mlx4_ib_remove
,
1686 .event
= mlx4_ib_event
,
1687 .protocol
= MLX4_PROT_IB_IPV6
1690 static int __init
mlx4_ib_init(void)
1694 wq
= create_singlethread_workqueue("mlx4_ib");
1698 err
= mlx4_ib_mcg_init();
1702 err
= mlx4_register_interface(&mlx4_ib_interface
);
1709 mlx4_ib_mcg_destroy();
1712 destroy_workqueue(wq
);
1716 static void __exit
mlx4_ib_cleanup(void)
1718 mlx4_unregister_interface(&mlx4_ib_interface
);
1719 mlx4_ib_mcg_destroy();
1720 destroy_workqueue(wq
);
1723 module_init(mlx4_ib_init
);
1724 module_exit(mlx4_ib_cleanup
);