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 <net/addrconf.h>
45 #include <rdma/ib_smi.h>
46 #include <rdma/ib_user_verbs.h>
47 #include <rdma/ib_addr.h>
49 #include <linux/mlx4/driver.h>
50 #include <linux/mlx4/cmd.h>
51 #include <linux/mlx4/qp.h>
56 #define DRV_NAME MLX4_IB_DRV_NAME
57 #define DRV_VERSION "2.2-1"
58 #define DRV_RELDATE "Feb 2014"
60 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
61 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
62 #define MLX4_IB_CARD_REV_A0 0xA0
64 MODULE_AUTHOR("Roland Dreier");
65 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
66 MODULE_LICENSE("Dual BSD/GPL");
67 MODULE_VERSION(DRV_VERSION
);
69 int mlx4_ib_sm_guid_assign
= 0;
70 module_param_named(sm_guid_assign
, mlx4_ib_sm_guid_assign
, int, 0444);
71 MODULE_PARM_DESC(sm_guid_assign
, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
73 static const char mlx4_ib_version
[] =
74 DRV_NAME
": Mellanox ConnectX InfiniBand driver v"
75 DRV_VERSION
" (" DRV_RELDATE
")\n";
77 struct update_gid_work
{
78 struct work_struct work
;
79 union ib_gid gids
[128];
80 struct mlx4_ib_dev
*dev
;
84 static void do_slave_init(struct mlx4_ib_dev
*ibdev
, int slave
, int do_init
);
86 static struct workqueue_struct
*wq
;
88 static void init_query_mad(struct ib_smp
*mad
)
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
;
96 static union ib_gid zgid
;
98 static int check_flow_steering_support(struct mlx4_dev
*dev
)
100 int eth_num_ports
= 0;
101 int ib_num_ports
= 0;
103 int dmfs
= dev
->caps
.steering_mode
== MLX4_STEERING_MODE_DEVICE_MANAGED
;
107 mlx4_foreach_port(i
, dev
, MLX4_PORT_TYPE_ETH
)
109 mlx4_foreach_port(i
, dev
, MLX4_PORT_TYPE_IB
)
111 dmfs
&= (!ib_num_ports
||
112 (dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_DMFS_IPOIB
)) &&
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");
123 static int num_ib_ports(struct mlx4_dev
*dev
)
128 mlx4_foreach_port(i
, dev
, MLX4_PORT_TYPE_IB
)
134 static int mlx4_ib_query_device(struct ib_device
*ibdev
,
135 struct ib_device_attr
*props
)
137 struct mlx4_ib_dev
*dev
= to_mdev(ibdev
);
138 struct ib_smp
*in_mad
= NULL
;
139 struct ib_smp
*out_mad
= NULL
;
143 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
144 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
145 if (!in_mad
|| !out_mad
)
148 init_query_mad(in_mad
);
149 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
151 err
= mlx4_MAD_IFC(to_mdev(ibdev
), MLX4_MAD_IFC_IGNORE_KEYS
,
152 1, NULL
, NULL
, in_mad
, out_mad
);
156 memset(props
, 0, sizeof *props
);
158 have_ib_ports
= num_ib_ports(dev
->dev
);
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
;
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
;
199 props
->vendor_id
= be32_to_cpup((__be32
*) (out_mad
->data
+ 36)) &
201 props
->vendor_part_id
= dev
->dev
->persist
->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);
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
;
240 static enum rdma_link_layer
241 mlx4_ib_port_link_layer(struct ib_device
*device
, u8 port_num
)
243 struct mlx4_dev
*dev
= to_mdev(device
)->dev
;
245 return dev
->caps
.port_mask
[port_num
] == MLX4_PORT_TYPE_IB
?
246 IB_LINK_LAYER_INFINIBAND
: IB_LINK_LAYER_ETHERNET
;
249 static int ib_link_query_port(struct ib_device
*ibdev
, u8 port
,
250 struct ib_port_attr
*props
, int netw_view
)
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
;
258 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
259 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
260 if (!in_mad
|| !out_mad
)
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
);
267 if (mlx4_is_mfunc(to_mdev(ibdev
)->dev
) && netw_view
)
268 mad_ifc_flags
|= MLX4_MAD_IFC_NET_VIEW
;
270 err
= mlx4_MAD_IFC(to_mdev(ibdev
), mad_ifc_flags
, port
, NULL
, NULL
,
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));
284 props
->gid_tbl_len
= out_mad
->data
[50];
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;
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;
303 switch (ext_active_speed
) {
305 props
->active_speed
= IB_SPEED_FDR
;
308 props
->active_speed
= IB_SPEED_EDR
;
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
);
319 err
= mlx4_MAD_IFC(to_mdev(ibdev
), mad_ifc_flags
, port
,
320 NULL
, NULL
, in_mad
, out_mad
);
324 /* Checking LinkSpeedActive for FDR-10 */
325 if (out_mad
->data
[15] & 0x1)
326 props
->active_speed
= IB_SPEED_FDR10
;
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
;
339 static u8
state_to_phys_state(enum ib_port_state state
)
341 return state
== IB_PORT_ACTIVE
? 5 : 3;
344 static int eth_link_query_port(struct ib_device
*ibdev
, u8 port
,
345 struct ib_port_attr
*props
, int netw_view
)
348 struct mlx4_ib_dev
*mdev
= to_mdev(ibdev
);
349 struct mlx4_ib_iboe
*iboe
= &mdev
->iboe
;
350 struct net_device
*ndev
;
352 struct mlx4_cmd_mailbox
*mailbox
;
354 int is_bonded
= mlx4_is_bonded(mdev
->dev
);
356 mailbox
= mlx4_alloc_cmd_mailbox(mdev
->dev
);
358 return PTR_ERR(mailbox
);
360 err
= mlx4_cmd_box(mdev
->dev
, 0, mailbox
->dma
, port
, 0,
361 MLX4_CMD_QUERY_PORT
, MLX4_CMD_TIME_CLASS_B
,
366 props
->active_width
= (((u8
*)mailbox
->buf
)[5] == 0x40) ?
367 IB_WIDTH_4X
: IB_WIDTH_1X
;
368 props
->active_speed
= IB_SPEED_QDR
;
369 props
->port_cap_flags
= IB_PORT_CM_SUP
| IB_PORT_IP_BASED_GIDS
;
370 props
->gid_tbl_len
= mdev
->dev
->caps
.gid_table_len
[port
];
371 props
->max_msg_sz
= mdev
->dev
->caps
.max_msg_sz
;
372 props
->pkey_tbl_len
= 1;
373 props
->max_mtu
= IB_MTU_4096
;
374 props
->max_vl_num
= 2;
375 props
->state
= IB_PORT_DOWN
;
376 props
->phys_state
= state_to_phys_state(props
->state
);
377 props
->active_mtu
= IB_MTU_256
;
379 rtnl_lock(); /* required to get upper dev */
380 spin_lock_bh(&iboe
->lock
);
381 ndev
= iboe
->netdevs
[port
- 1];
382 if (ndev
&& is_bonded
)
383 ndev
= netdev_master_upper_dev_get(ndev
);
387 tmp
= iboe_get_mtu(ndev
->mtu
);
388 props
->active_mtu
= tmp
? min(props
->max_mtu
, tmp
) : IB_MTU_256
;
390 props
->state
= (netif_running(ndev
) && netif_carrier_ok(ndev
)) ?
391 IB_PORT_ACTIVE
: IB_PORT_DOWN
;
392 props
->phys_state
= state_to_phys_state(props
->state
);
394 spin_unlock_bh(&iboe
->lock
);
398 mlx4_free_cmd_mailbox(mdev
->dev
, mailbox
);
402 int __mlx4_ib_query_port(struct ib_device
*ibdev
, u8 port
,
403 struct ib_port_attr
*props
, int netw_view
)
407 memset(props
, 0, sizeof *props
);
409 err
= mlx4_ib_port_link_layer(ibdev
, port
) == IB_LINK_LAYER_INFINIBAND
?
410 ib_link_query_port(ibdev
, port
, props
, netw_view
) :
411 eth_link_query_port(ibdev
, port
, props
, netw_view
);
416 static int mlx4_ib_query_port(struct ib_device
*ibdev
, u8 port
,
417 struct ib_port_attr
*props
)
419 /* returns host view */
420 return __mlx4_ib_query_port(ibdev
, port
, props
, 0);
423 static enum rdma_protocol_type
424 mlx4_ib_query_protocol(struct ib_device
*device
, u8 port_num
)
426 struct mlx4_dev
*dev
= to_mdev(device
)->dev
;
428 return dev
->caps
.port_mask
[port_num
] == MLX4_PORT_TYPE_IB
?
429 RDMA_PROTOCOL_IB
: RDMA_PROTOCOL_IBOE
;
432 int __mlx4_ib_query_gid(struct ib_device
*ibdev
, u8 port
, int index
,
433 union ib_gid
*gid
, int netw_view
)
435 struct ib_smp
*in_mad
= NULL
;
436 struct ib_smp
*out_mad
= NULL
;
438 struct mlx4_ib_dev
*dev
= to_mdev(ibdev
);
440 int mad_ifc_flags
= MLX4_MAD_IFC_IGNORE_KEYS
;
442 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
443 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
444 if (!in_mad
|| !out_mad
)
447 init_query_mad(in_mad
);
448 in_mad
->attr_id
= IB_SMP_ATTR_PORT_INFO
;
449 in_mad
->attr_mod
= cpu_to_be32(port
);
451 if (mlx4_is_mfunc(dev
->dev
) && netw_view
)
452 mad_ifc_flags
|= MLX4_MAD_IFC_NET_VIEW
;
454 err
= mlx4_MAD_IFC(dev
, mad_ifc_flags
, port
, NULL
, NULL
, in_mad
, out_mad
);
458 memcpy(gid
->raw
, out_mad
->data
+ 8, 8);
460 if (mlx4_is_mfunc(dev
->dev
) && !netw_view
) {
462 /* For any index > 0, return the null guid */
469 init_query_mad(in_mad
);
470 in_mad
->attr_id
= IB_SMP_ATTR_GUID_INFO
;
471 in_mad
->attr_mod
= cpu_to_be32(index
/ 8);
473 err
= mlx4_MAD_IFC(dev
, mad_ifc_flags
, port
,
474 NULL
, NULL
, in_mad
, out_mad
);
478 memcpy(gid
->raw
+ 8, out_mad
->data
+ (index
% 8) * 8, 8);
482 memset(gid
->raw
+ 8, 0, 8);
488 static int iboe_query_gid(struct ib_device
*ibdev
, u8 port
, int index
,
491 struct mlx4_ib_dev
*dev
= to_mdev(ibdev
);
493 *gid
= dev
->iboe
.gid_table
[port
- 1][index
];
498 static int mlx4_ib_query_gid(struct ib_device
*ibdev
, u8 port
, int index
,
501 if (rdma_port_get_link_layer(ibdev
, port
) == IB_LINK_LAYER_INFINIBAND
)
502 return __mlx4_ib_query_gid(ibdev
, port
, index
, gid
, 0);
504 return iboe_query_gid(ibdev
, port
, index
, gid
);
507 int __mlx4_ib_query_pkey(struct ib_device
*ibdev
, u8 port
, u16 index
,
508 u16
*pkey
, int netw_view
)
510 struct ib_smp
*in_mad
= NULL
;
511 struct ib_smp
*out_mad
= NULL
;
512 int mad_ifc_flags
= MLX4_MAD_IFC_IGNORE_KEYS
;
515 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
516 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
517 if (!in_mad
|| !out_mad
)
520 init_query_mad(in_mad
);
521 in_mad
->attr_id
= IB_SMP_ATTR_PKEY_TABLE
;
522 in_mad
->attr_mod
= cpu_to_be32(index
/ 32);
524 if (mlx4_is_mfunc(to_mdev(ibdev
)->dev
) && netw_view
)
525 mad_ifc_flags
|= MLX4_MAD_IFC_NET_VIEW
;
527 err
= mlx4_MAD_IFC(to_mdev(ibdev
), mad_ifc_flags
, port
, NULL
, NULL
,
532 *pkey
= be16_to_cpu(((__be16
*) out_mad
->data
)[index
% 32]);
540 static int mlx4_ib_query_pkey(struct ib_device
*ibdev
, u8 port
, u16 index
, u16
*pkey
)
542 return __mlx4_ib_query_pkey(ibdev
, port
, index
, pkey
, 0);
545 static int mlx4_ib_modify_device(struct ib_device
*ibdev
, int mask
,
546 struct ib_device_modify
*props
)
548 struct mlx4_cmd_mailbox
*mailbox
;
551 if (mask
& ~IB_DEVICE_MODIFY_NODE_DESC
)
554 if (!(mask
& IB_DEVICE_MODIFY_NODE_DESC
))
557 if (mlx4_is_slave(to_mdev(ibdev
)->dev
))
560 spin_lock_irqsave(&to_mdev(ibdev
)->sm_lock
, flags
);
561 memcpy(ibdev
->node_desc
, props
->node_desc
, 64);
562 spin_unlock_irqrestore(&to_mdev(ibdev
)->sm_lock
, flags
);
565 * If possible, pass node desc to FW, so it can generate
566 * a 144 trap. If cmd fails, just ignore.
568 mailbox
= mlx4_alloc_cmd_mailbox(to_mdev(ibdev
)->dev
);
572 memcpy(mailbox
->buf
, props
->node_desc
, 64);
573 mlx4_cmd(to_mdev(ibdev
)->dev
, mailbox
->dma
, 1, 0,
574 MLX4_CMD_SET_NODE
, MLX4_CMD_TIME_CLASS_A
, MLX4_CMD_NATIVE
);
576 mlx4_free_cmd_mailbox(to_mdev(ibdev
)->dev
, mailbox
);
581 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev
*dev
, u8 port
, int reset_qkey_viols
,
584 struct mlx4_cmd_mailbox
*mailbox
;
587 mailbox
= mlx4_alloc_cmd_mailbox(dev
->dev
);
589 return PTR_ERR(mailbox
);
591 if (dev
->dev
->flags
& MLX4_FLAG_OLD_PORT_CMDS
) {
592 *(u8
*) mailbox
->buf
= !!reset_qkey_viols
<< 6;
593 ((__be32
*) mailbox
->buf
)[2] = cpu_to_be32(cap_mask
);
595 ((u8
*) mailbox
->buf
)[3] = !!reset_qkey_viols
;
596 ((__be32
*) mailbox
->buf
)[1] = cpu_to_be32(cap_mask
);
599 err
= mlx4_cmd(dev
->dev
, mailbox
->dma
, port
, MLX4_SET_PORT_IB_OPCODE
,
600 MLX4_CMD_SET_PORT
, MLX4_CMD_TIME_CLASS_B
,
603 mlx4_free_cmd_mailbox(dev
->dev
, mailbox
);
607 static int mlx4_ib_modify_port(struct ib_device
*ibdev
, u8 port
, int mask
,
608 struct ib_port_modify
*props
)
610 struct mlx4_ib_dev
*mdev
= to_mdev(ibdev
);
611 u8 is_eth
= mdev
->dev
->caps
.port_type
[port
] == MLX4_PORT_TYPE_ETH
;
612 struct ib_port_attr attr
;
616 /* return OK if this is RoCE. CM calls ib_modify_port() regardless
617 * of whether port link layer is ETH or IB. For ETH ports, qkey
618 * violations and port capabilities are not meaningful.
623 mutex_lock(&mdev
->cap_mask_mutex
);
625 err
= mlx4_ib_query_port(ibdev
, port
, &attr
);
629 cap_mask
= (attr
.port_cap_flags
| props
->set_port_cap_mask
) &
630 ~props
->clr_port_cap_mask
;
632 err
= mlx4_ib_SET_PORT(mdev
, port
,
633 !!(mask
& IB_PORT_RESET_QKEY_CNTR
),
637 mutex_unlock(&to_mdev(ibdev
)->cap_mask_mutex
);
641 static struct ib_ucontext
*mlx4_ib_alloc_ucontext(struct ib_device
*ibdev
,
642 struct ib_udata
*udata
)
644 struct mlx4_ib_dev
*dev
= to_mdev(ibdev
);
645 struct mlx4_ib_ucontext
*context
;
646 struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3
;
647 struct mlx4_ib_alloc_ucontext_resp resp
;
651 return ERR_PTR(-EAGAIN
);
653 if (ibdev
->uverbs_abi_ver
== MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION
) {
654 resp_v3
.qp_tab_size
= dev
->dev
->caps
.num_qps
;
655 resp_v3
.bf_reg_size
= dev
->dev
->caps
.bf_reg_size
;
656 resp_v3
.bf_regs_per_page
= dev
->dev
->caps
.bf_regs_per_page
;
658 resp
.dev_caps
= dev
->dev
->caps
.userspace_caps
;
659 resp
.qp_tab_size
= dev
->dev
->caps
.num_qps
;
660 resp
.bf_reg_size
= dev
->dev
->caps
.bf_reg_size
;
661 resp
.bf_regs_per_page
= dev
->dev
->caps
.bf_regs_per_page
;
662 resp
.cqe_size
= dev
->dev
->caps
.cqe_size
;
665 context
= kmalloc(sizeof *context
, GFP_KERNEL
);
667 return ERR_PTR(-ENOMEM
);
669 err
= mlx4_uar_alloc(to_mdev(ibdev
)->dev
, &context
->uar
);
675 INIT_LIST_HEAD(&context
->db_page_list
);
676 mutex_init(&context
->db_page_mutex
);
678 if (ibdev
->uverbs_abi_ver
== MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION
)
679 err
= ib_copy_to_udata(udata
, &resp_v3
, sizeof(resp_v3
));
681 err
= ib_copy_to_udata(udata
, &resp
, sizeof(resp
));
684 mlx4_uar_free(to_mdev(ibdev
)->dev
, &context
->uar
);
686 return ERR_PTR(-EFAULT
);
689 return &context
->ibucontext
;
692 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext
*ibcontext
)
694 struct mlx4_ib_ucontext
*context
= to_mucontext(ibcontext
);
696 mlx4_uar_free(to_mdev(ibcontext
->device
)->dev
, &context
->uar
);
702 static int mlx4_ib_mmap(struct ib_ucontext
*context
, struct vm_area_struct
*vma
)
704 struct mlx4_ib_dev
*dev
= to_mdev(context
->device
);
706 if (vma
->vm_end
- vma
->vm_start
!= PAGE_SIZE
)
709 if (vma
->vm_pgoff
== 0) {
710 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
712 if (io_remap_pfn_range(vma
, vma
->vm_start
,
713 to_mucontext(context
)->uar
.pfn
,
714 PAGE_SIZE
, vma
->vm_page_prot
))
716 } else if (vma
->vm_pgoff
== 1 && dev
->dev
->caps
.bf_reg_size
!= 0) {
717 vma
->vm_page_prot
= pgprot_writecombine(vma
->vm_page_prot
);
719 if (io_remap_pfn_range(vma
, vma
->vm_start
,
720 to_mucontext(context
)->uar
.pfn
+
721 dev
->dev
->caps
.num_uars
,
722 PAGE_SIZE
, vma
->vm_page_prot
))
730 static struct ib_pd
*mlx4_ib_alloc_pd(struct ib_device
*ibdev
,
731 struct ib_ucontext
*context
,
732 struct ib_udata
*udata
)
734 struct mlx4_ib_pd
*pd
;
737 pd
= kmalloc(sizeof *pd
, GFP_KERNEL
);
739 return ERR_PTR(-ENOMEM
);
741 err
= mlx4_pd_alloc(to_mdev(ibdev
)->dev
, &pd
->pdn
);
748 if (ib_copy_to_udata(udata
, &pd
->pdn
, sizeof (__u32
))) {
749 mlx4_pd_free(to_mdev(ibdev
)->dev
, pd
->pdn
);
751 return ERR_PTR(-EFAULT
);
757 static int mlx4_ib_dealloc_pd(struct ib_pd
*pd
)
759 mlx4_pd_free(to_mdev(pd
->device
)->dev
, to_mpd(pd
)->pdn
);
765 static struct ib_xrcd
*mlx4_ib_alloc_xrcd(struct ib_device
*ibdev
,
766 struct ib_ucontext
*context
,
767 struct ib_udata
*udata
)
769 struct mlx4_ib_xrcd
*xrcd
;
772 if (!(to_mdev(ibdev
)->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_XRC
))
773 return ERR_PTR(-ENOSYS
);
775 xrcd
= kmalloc(sizeof *xrcd
, GFP_KERNEL
);
777 return ERR_PTR(-ENOMEM
);
779 err
= mlx4_xrcd_alloc(to_mdev(ibdev
)->dev
, &xrcd
->xrcdn
);
783 xrcd
->pd
= ib_alloc_pd(ibdev
);
784 if (IS_ERR(xrcd
->pd
)) {
785 err
= PTR_ERR(xrcd
->pd
);
789 xrcd
->cq
= ib_create_cq(ibdev
, NULL
, NULL
, xrcd
, 1, 0);
790 if (IS_ERR(xrcd
->cq
)) {
791 err
= PTR_ERR(xrcd
->cq
);
795 return &xrcd
->ibxrcd
;
798 ib_dealloc_pd(xrcd
->pd
);
800 mlx4_xrcd_free(to_mdev(ibdev
)->dev
, xrcd
->xrcdn
);
806 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd
*xrcd
)
808 ib_destroy_cq(to_mxrcd(xrcd
)->cq
);
809 ib_dealloc_pd(to_mxrcd(xrcd
)->pd
);
810 mlx4_xrcd_free(to_mdev(xrcd
->device
)->dev
, to_mxrcd(xrcd
)->xrcdn
);
816 static int add_gid_entry(struct ib_qp
*ibqp
, union ib_gid
*gid
)
818 struct mlx4_ib_qp
*mqp
= to_mqp(ibqp
);
819 struct mlx4_ib_dev
*mdev
= to_mdev(ibqp
->device
);
820 struct mlx4_ib_gid_entry
*ge
;
822 ge
= kzalloc(sizeof *ge
, GFP_KERNEL
);
827 if (mlx4_ib_add_mc(mdev
, mqp
, gid
)) {
828 ge
->port
= mqp
->port
;
832 mutex_lock(&mqp
->mutex
);
833 list_add_tail(&ge
->list
, &mqp
->gid_list
);
834 mutex_unlock(&mqp
->mutex
);
839 int mlx4_ib_add_mc(struct mlx4_ib_dev
*mdev
, struct mlx4_ib_qp
*mqp
,
842 struct net_device
*ndev
;
848 spin_lock_bh(&mdev
->iboe
.lock
);
849 ndev
= mdev
->iboe
.netdevs
[mqp
->port
- 1];
852 spin_unlock_bh(&mdev
->iboe
.lock
);
862 struct mlx4_ib_steering
{
863 struct list_head list
;
864 struct mlx4_flow_reg_id reg_id
;
868 static int parse_flow_attr(struct mlx4_dev
*dev
,
870 union ib_flow_spec
*ib_spec
,
871 struct _rule_hw
*mlx4_spec
)
873 enum mlx4_net_trans_rule_id type
;
875 switch (ib_spec
->type
) {
876 case IB_FLOW_SPEC_ETH
:
877 type
= MLX4_NET_TRANS_RULE_ID_ETH
;
878 memcpy(mlx4_spec
->eth
.dst_mac
, ib_spec
->eth
.val
.dst_mac
,
880 memcpy(mlx4_spec
->eth
.dst_mac_msk
, ib_spec
->eth
.mask
.dst_mac
,
882 mlx4_spec
->eth
.vlan_tag
= ib_spec
->eth
.val
.vlan_tag
;
883 mlx4_spec
->eth
.vlan_tag_msk
= ib_spec
->eth
.mask
.vlan_tag
;
885 case IB_FLOW_SPEC_IB
:
886 type
= MLX4_NET_TRANS_RULE_ID_IB
;
887 mlx4_spec
->ib
.l3_qpn
=
889 mlx4_spec
->ib
.qpn_mask
=
890 cpu_to_be32(MLX4_IB_FLOW_QPN_MASK
);
894 case IB_FLOW_SPEC_IPV4
:
895 type
= MLX4_NET_TRANS_RULE_ID_IPV4
;
896 mlx4_spec
->ipv4
.src_ip
= ib_spec
->ipv4
.val
.src_ip
;
897 mlx4_spec
->ipv4
.src_ip_msk
= ib_spec
->ipv4
.mask
.src_ip
;
898 mlx4_spec
->ipv4
.dst_ip
= ib_spec
->ipv4
.val
.dst_ip
;
899 mlx4_spec
->ipv4
.dst_ip_msk
= ib_spec
->ipv4
.mask
.dst_ip
;
902 case IB_FLOW_SPEC_TCP
:
903 case IB_FLOW_SPEC_UDP
:
904 type
= ib_spec
->type
== IB_FLOW_SPEC_TCP
?
905 MLX4_NET_TRANS_RULE_ID_TCP
:
906 MLX4_NET_TRANS_RULE_ID_UDP
;
907 mlx4_spec
->tcp_udp
.dst_port
= ib_spec
->tcp_udp
.val
.dst_port
;
908 mlx4_spec
->tcp_udp
.dst_port_msk
= ib_spec
->tcp_udp
.mask
.dst_port
;
909 mlx4_spec
->tcp_udp
.src_port
= ib_spec
->tcp_udp
.val
.src_port
;
910 mlx4_spec
->tcp_udp
.src_port_msk
= ib_spec
->tcp_udp
.mask
.src_port
;
916 if (mlx4_map_sw_to_hw_steering_id(dev
, type
) < 0 ||
917 mlx4_hw_rule_sz(dev
, type
) < 0)
919 mlx4_spec
->id
= cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev
, type
));
920 mlx4_spec
->size
= mlx4_hw_rule_sz(dev
, type
) >> 2;
921 return mlx4_hw_rule_sz(dev
, type
);
924 struct default_rules
{
925 __u32 mandatory_fields
[IB_FLOW_SPEC_SUPPORT_LAYERS
];
926 __u32 mandatory_not_fields
[IB_FLOW_SPEC_SUPPORT_LAYERS
];
927 __u32 rules_create_list
[IB_FLOW_SPEC_SUPPORT_LAYERS
];
930 static const struct default_rules default_table
[] = {
932 .mandatory_fields
= {IB_FLOW_SPEC_IPV4
},
933 .mandatory_not_fields
= {IB_FLOW_SPEC_ETH
},
934 .rules_create_list
= {IB_FLOW_SPEC_IB
},
935 .link_layer
= IB_LINK_LAYER_INFINIBAND
939 static int __mlx4_ib_default_rules_match(struct ib_qp
*qp
,
940 struct ib_flow_attr
*flow_attr
)
944 const struct default_rules
*pdefault_rules
= default_table
;
945 u8 link_layer
= rdma_port_get_link_layer(qp
->device
, flow_attr
->port
);
947 for (i
= 0; i
< ARRAY_SIZE(default_table
); i
++, pdefault_rules
++) {
948 __u32 field_types
[IB_FLOW_SPEC_SUPPORT_LAYERS
];
949 memset(&field_types
, 0, sizeof(field_types
));
951 if (link_layer
!= pdefault_rules
->link_layer
)
954 ib_flow
= flow_attr
+ 1;
955 /* we assume the specs are sorted */
956 for (j
= 0, k
= 0; k
< IB_FLOW_SPEC_SUPPORT_LAYERS
&&
957 j
< flow_attr
->num_of_specs
; k
++) {
958 union ib_flow_spec
*current_flow
=
959 (union ib_flow_spec
*)ib_flow
;
961 /* same layer but different type */
962 if (((current_flow
->type
& IB_FLOW_SPEC_LAYER_MASK
) ==
963 (pdefault_rules
->mandatory_fields
[k
] &
964 IB_FLOW_SPEC_LAYER_MASK
)) &&
965 (current_flow
->type
!=
966 pdefault_rules
->mandatory_fields
[k
]))
969 /* same layer, try match next one */
970 if (current_flow
->type
==
971 pdefault_rules
->mandatory_fields
[k
]) {
974 ((union ib_flow_spec
*)ib_flow
)->size
;
978 ib_flow
= flow_attr
+ 1;
979 for (j
= 0; j
< flow_attr
->num_of_specs
;
980 j
++, ib_flow
+= ((union ib_flow_spec
*)ib_flow
)->size
)
981 for (k
= 0; k
< IB_FLOW_SPEC_SUPPORT_LAYERS
; k
++)
982 /* same layer and same type */
983 if (((union ib_flow_spec
*)ib_flow
)->type
==
984 pdefault_rules
->mandatory_not_fields
[k
])
993 static int __mlx4_ib_create_default_rules(
994 struct mlx4_ib_dev
*mdev
,
996 const struct default_rules
*pdefault_rules
,
997 struct _rule_hw
*mlx4_spec
) {
1001 for (i
= 0; i
< ARRAY_SIZE(pdefault_rules
->rules_create_list
); i
++) {
1003 union ib_flow_spec ib_spec
;
1004 switch (pdefault_rules
->rules_create_list
[i
]) {
1008 case IB_FLOW_SPEC_IB
:
1009 ib_spec
.type
= IB_FLOW_SPEC_IB
;
1010 ib_spec
.size
= sizeof(struct ib_flow_spec_ib
);
1017 /* We must put empty rule, qpn is being ignored */
1018 ret
= parse_flow_attr(mdev
->dev
, 0, &ib_spec
,
1021 pr_info("invalid parsing\n");
1025 mlx4_spec
= (void *)mlx4_spec
+ ret
;
1031 static int __mlx4_ib_create_flow(struct ib_qp
*qp
, struct ib_flow_attr
*flow_attr
,
1033 enum mlx4_net_trans_promisc_mode flow_type
,
1039 struct mlx4_ib_dev
*mdev
= to_mdev(qp
->device
);
1040 struct mlx4_cmd_mailbox
*mailbox
;
1041 struct mlx4_net_trans_rule_hw_ctrl
*ctrl
;
1044 static const u16 __mlx4_domain
[] = {
1045 [IB_FLOW_DOMAIN_USER
] = MLX4_DOMAIN_UVERBS
,
1046 [IB_FLOW_DOMAIN_ETHTOOL
] = MLX4_DOMAIN_ETHTOOL
,
1047 [IB_FLOW_DOMAIN_RFS
] = MLX4_DOMAIN_RFS
,
1048 [IB_FLOW_DOMAIN_NIC
] = MLX4_DOMAIN_NIC
,
1051 if (flow_attr
->priority
> MLX4_IB_FLOW_MAX_PRIO
) {
1052 pr_err("Invalid priority value %d\n", flow_attr
->priority
);
1056 if (domain
>= IB_FLOW_DOMAIN_NUM
) {
1057 pr_err("Invalid domain value %d\n", domain
);
1061 if (mlx4_map_sw_to_hw_steering_mode(mdev
->dev
, flow_type
) < 0)
1064 mailbox
= mlx4_alloc_cmd_mailbox(mdev
->dev
);
1065 if (IS_ERR(mailbox
))
1066 return PTR_ERR(mailbox
);
1067 ctrl
= mailbox
->buf
;
1069 ctrl
->prio
= cpu_to_be16(__mlx4_domain
[domain
] |
1070 flow_attr
->priority
);
1071 ctrl
->type
= mlx4_map_sw_to_hw_steering_mode(mdev
->dev
, flow_type
);
1072 ctrl
->port
= flow_attr
->port
;
1073 ctrl
->qpn
= cpu_to_be32(qp
->qp_num
);
1075 ib_flow
= flow_attr
+ 1;
1076 size
+= sizeof(struct mlx4_net_trans_rule_hw_ctrl
);
1077 /* Add default flows */
1078 default_flow
= __mlx4_ib_default_rules_match(qp
, flow_attr
);
1079 if (default_flow
>= 0) {
1080 ret
= __mlx4_ib_create_default_rules(
1081 mdev
, qp
, default_table
+ default_flow
,
1082 mailbox
->buf
+ size
);
1084 mlx4_free_cmd_mailbox(mdev
->dev
, mailbox
);
1089 for (i
= 0; i
< flow_attr
->num_of_specs
; i
++) {
1090 ret
= parse_flow_attr(mdev
->dev
, qp
->qp_num
, ib_flow
,
1091 mailbox
->buf
+ size
);
1093 mlx4_free_cmd_mailbox(mdev
->dev
, mailbox
);
1096 ib_flow
+= ((union ib_flow_spec
*) ib_flow
)->size
;
1100 ret
= mlx4_cmd_imm(mdev
->dev
, mailbox
->dma
, reg_id
, size
>> 2, 0,
1101 MLX4_QP_FLOW_STEERING_ATTACH
, MLX4_CMD_TIME_CLASS_A
,
1104 pr_err("mcg table is full. Fail to register network rule.\n");
1105 else if (ret
== -ENXIO
)
1106 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1108 pr_err("Invalid argumant. Fail to register network rule.\n");
1110 mlx4_free_cmd_mailbox(mdev
->dev
, mailbox
);
1114 static int __mlx4_ib_destroy_flow(struct mlx4_dev
*dev
, u64 reg_id
)
1117 err
= mlx4_cmd(dev
, reg_id
, 0, 0,
1118 MLX4_QP_FLOW_STEERING_DETACH
, MLX4_CMD_TIME_CLASS_A
,
1121 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1126 static int mlx4_ib_tunnel_steer_add(struct ib_qp
*qp
, struct ib_flow_attr
*flow_attr
,
1130 union ib_flow_spec
*ib_spec
;
1131 struct mlx4_dev
*dev
= to_mdev(qp
->device
)->dev
;
1134 if (dev
->caps
.tunnel_offload_mode
!= MLX4_TUNNEL_OFFLOAD_MODE_VXLAN
||
1135 dev
->caps
.dmfs_high_steer_mode
== MLX4_STEERING_DMFS_A0_STATIC
)
1136 return 0; /* do nothing */
1138 ib_flow
= flow_attr
+ 1;
1139 ib_spec
= (union ib_flow_spec
*)ib_flow
;
1141 if (ib_spec
->type
!= IB_FLOW_SPEC_ETH
|| flow_attr
->num_of_specs
!= 1)
1142 return 0; /* do nothing */
1144 err
= mlx4_tunnel_steer_add(to_mdev(qp
->device
)->dev
, ib_spec
->eth
.val
.dst_mac
,
1145 flow_attr
->port
, qp
->qp_num
,
1146 MLX4_DOMAIN_UVERBS
| (flow_attr
->priority
& 0xff),
1151 static struct ib_flow
*mlx4_ib_create_flow(struct ib_qp
*qp
,
1152 struct ib_flow_attr
*flow_attr
,
1155 int err
= 0, i
= 0, j
= 0;
1156 struct mlx4_ib_flow
*mflow
;
1157 enum mlx4_net_trans_promisc_mode type
[2];
1158 struct mlx4_dev
*dev
= (to_mdev(qp
->device
))->dev
;
1159 int is_bonded
= mlx4_is_bonded(dev
);
1161 memset(type
, 0, sizeof(type
));
1163 mflow
= kzalloc(sizeof(*mflow
), GFP_KERNEL
);
1169 switch (flow_attr
->type
) {
1170 case IB_FLOW_ATTR_NORMAL
:
1171 type
[0] = MLX4_FS_REGULAR
;
1174 case IB_FLOW_ATTR_ALL_DEFAULT
:
1175 type
[0] = MLX4_FS_ALL_DEFAULT
;
1178 case IB_FLOW_ATTR_MC_DEFAULT
:
1179 type
[0] = MLX4_FS_MC_DEFAULT
;
1182 case IB_FLOW_ATTR_SNIFFER
:
1183 type
[0] = MLX4_FS_UC_SNIFFER
;
1184 type
[1] = MLX4_FS_MC_SNIFFER
;
1192 while (i
< ARRAY_SIZE(type
) && type
[i
]) {
1193 err
= __mlx4_ib_create_flow(qp
, flow_attr
, domain
, type
[i
],
1194 &mflow
->reg_id
[i
].id
);
1196 goto err_create_flow
;
1199 /* Application always sees one port so the mirror rule
1200 * must be on port #2
1202 flow_attr
->port
= 2;
1203 err
= __mlx4_ib_create_flow(qp
, flow_attr
,
1205 &mflow
->reg_id
[j
].mirror
);
1206 flow_attr
->port
= 1;
1208 goto err_create_flow
;
1214 if (i
< ARRAY_SIZE(type
) && flow_attr
->type
== IB_FLOW_ATTR_NORMAL
) {
1215 err
= mlx4_ib_tunnel_steer_add(qp
, flow_attr
,
1216 &mflow
->reg_id
[i
].id
);
1218 goto err_create_flow
;
1221 flow_attr
->port
= 2;
1222 err
= mlx4_ib_tunnel_steer_add(qp
, flow_attr
,
1223 &mflow
->reg_id
[j
].mirror
);
1224 flow_attr
->port
= 1;
1226 goto err_create_flow
;
1229 /* function to create mirror rule */
1232 return &mflow
->ibflow
;
1236 (void)__mlx4_ib_destroy_flow(to_mdev(qp
->device
)->dev
,
1237 mflow
->reg_id
[i
].id
);
1242 (void)__mlx4_ib_destroy_flow(to_mdev(qp
->device
)->dev
,
1243 mflow
->reg_id
[j
].mirror
);
1248 return ERR_PTR(err
);
1251 static int mlx4_ib_destroy_flow(struct ib_flow
*flow_id
)
1255 struct mlx4_ib_dev
*mdev
= to_mdev(flow_id
->qp
->device
);
1256 struct mlx4_ib_flow
*mflow
= to_mflow(flow_id
);
1258 while (i
< ARRAY_SIZE(mflow
->reg_id
) && mflow
->reg_id
[i
].id
) {
1259 err
= __mlx4_ib_destroy_flow(mdev
->dev
, mflow
->reg_id
[i
].id
);
1262 if (mflow
->reg_id
[i
].mirror
) {
1263 err
= __mlx4_ib_destroy_flow(mdev
->dev
,
1264 mflow
->reg_id
[i
].mirror
);
1275 static int mlx4_ib_mcg_attach(struct ib_qp
*ibqp
, union ib_gid
*gid
, u16 lid
)
1278 struct mlx4_ib_dev
*mdev
= to_mdev(ibqp
->device
);
1279 struct mlx4_dev
*dev
= mdev
->dev
;
1280 struct mlx4_ib_qp
*mqp
= to_mqp(ibqp
);
1281 struct mlx4_ib_steering
*ib_steering
= NULL
;
1282 enum mlx4_protocol prot
= MLX4_PROT_IB_IPV6
;
1283 struct mlx4_flow_reg_id reg_id
;
1285 if (mdev
->dev
->caps
.steering_mode
==
1286 MLX4_STEERING_MODE_DEVICE_MANAGED
) {
1287 ib_steering
= kmalloc(sizeof(*ib_steering
), GFP_KERNEL
);
1292 err
= mlx4_multicast_attach(mdev
->dev
, &mqp
->mqp
, gid
->raw
, mqp
->port
,
1294 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK
),
1297 pr_err("multicast attach op failed, err %d\n", err
);
1302 if (mlx4_is_bonded(dev
)) {
1303 err
= mlx4_multicast_attach(mdev
->dev
, &mqp
->mqp
, gid
->raw
,
1304 (mqp
->port
== 1) ? 2 : 1,
1306 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK
),
1307 prot
, ®_id
.mirror
);
1312 err
= add_gid_entry(ibqp
, gid
);
1317 memcpy(ib_steering
->gid
.raw
, gid
->raw
, 16);
1318 ib_steering
->reg_id
= reg_id
;
1319 mutex_lock(&mqp
->mutex
);
1320 list_add(&ib_steering
->list
, &mqp
->steering_rules
);
1321 mutex_unlock(&mqp
->mutex
);
1326 mlx4_multicast_detach(mdev
->dev
, &mqp
->mqp
, gid
->raw
,
1329 mlx4_multicast_detach(mdev
->dev
, &mqp
->mqp
, gid
->raw
,
1330 prot
, reg_id
.mirror
);
1337 static struct mlx4_ib_gid_entry
*find_gid_entry(struct mlx4_ib_qp
*qp
, u8
*raw
)
1339 struct mlx4_ib_gid_entry
*ge
;
1340 struct mlx4_ib_gid_entry
*tmp
;
1341 struct mlx4_ib_gid_entry
*ret
= NULL
;
1343 list_for_each_entry_safe(ge
, tmp
, &qp
->gid_list
, list
) {
1344 if (!memcmp(raw
, ge
->gid
.raw
, 16)) {
1353 static int mlx4_ib_mcg_detach(struct ib_qp
*ibqp
, union ib_gid
*gid
, u16 lid
)
1356 struct mlx4_ib_dev
*mdev
= to_mdev(ibqp
->device
);
1357 struct mlx4_dev
*dev
= mdev
->dev
;
1358 struct mlx4_ib_qp
*mqp
= to_mqp(ibqp
);
1359 struct net_device
*ndev
;
1360 struct mlx4_ib_gid_entry
*ge
;
1361 struct mlx4_flow_reg_id reg_id
= {0, 0};
1362 enum mlx4_protocol prot
= MLX4_PROT_IB_IPV6
;
1364 if (mdev
->dev
->caps
.steering_mode
==
1365 MLX4_STEERING_MODE_DEVICE_MANAGED
) {
1366 struct mlx4_ib_steering
*ib_steering
;
1368 mutex_lock(&mqp
->mutex
);
1369 list_for_each_entry(ib_steering
, &mqp
->steering_rules
, list
) {
1370 if (!memcmp(ib_steering
->gid
.raw
, gid
->raw
, 16)) {
1371 list_del(&ib_steering
->list
);
1375 mutex_unlock(&mqp
->mutex
);
1376 if (&ib_steering
->list
== &mqp
->steering_rules
) {
1377 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1380 reg_id
= ib_steering
->reg_id
;
1384 err
= mlx4_multicast_detach(mdev
->dev
, &mqp
->mqp
, gid
->raw
,
1389 if (mlx4_is_bonded(dev
)) {
1390 err
= mlx4_multicast_detach(mdev
->dev
, &mqp
->mqp
, gid
->raw
,
1391 prot
, reg_id
.mirror
);
1396 mutex_lock(&mqp
->mutex
);
1397 ge
= find_gid_entry(mqp
, gid
->raw
);
1399 spin_lock_bh(&mdev
->iboe
.lock
);
1400 ndev
= ge
->added
? mdev
->iboe
.netdevs
[ge
->port
- 1] : NULL
;
1403 spin_unlock_bh(&mdev
->iboe
.lock
);
1406 list_del(&ge
->list
);
1409 pr_warn("could not find mgid entry\n");
1411 mutex_unlock(&mqp
->mutex
);
1416 static int init_node_data(struct mlx4_ib_dev
*dev
)
1418 struct ib_smp
*in_mad
= NULL
;
1419 struct ib_smp
*out_mad
= NULL
;
1420 int mad_ifc_flags
= MLX4_MAD_IFC_IGNORE_KEYS
;
1423 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
1424 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
1425 if (!in_mad
|| !out_mad
)
1428 init_query_mad(in_mad
);
1429 in_mad
->attr_id
= IB_SMP_ATTR_NODE_DESC
;
1430 if (mlx4_is_master(dev
->dev
))
1431 mad_ifc_flags
|= MLX4_MAD_IFC_NET_VIEW
;
1433 err
= mlx4_MAD_IFC(dev
, mad_ifc_flags
, 1, NULL
, NULL
, in_mad
, out_mad
);
1437 memcpy(dev
->ib_dev
.node_desc
, out_mad
->data
, 64);
1439 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
1441 err
= mlx4_MAD_IFC(dev
, mad_ifc_flags
, 1, NULL
, NULL
, in_mad
, out_mad
);
1445 dev
->dev
->rev_id
= be32_to_cpup((__be32
*) (out_mad
->data
+ 32));
1446 memcpy(&dev
->ib_dev
.node_guid
, out_mad
->data
+ 12, 8);
1454 static ssize_t
show_hca(struct device
*device
, struct device_attribute
*attr
,
1457 struct mlx4_ib_dev
*dev
=
1458 container_of(device
, struct mlx4_ib_dev
, ib_dev
.dev
);
1459 return sprintf(buf
, "MT%d\n", dev
->dev
->persist
->pdev
->device
);
1462 static ssize_t
show_fw_ver(struct device
*device
, struct device_attribute
*attr
,
1465 struct mlx4_ib_dev
*dev
=
1466 container_of(device
, struct mlx4_ib_dev
, ib_dev
.dev
);
1467 return sprintf(buf
, "%d.%d.%d\n", (int) (dev
->dev
->caps
.fw_ver
>> 32),
1468 (int) (dev
->dev
->caps
.fw_ver
>> 16) & 0xffff,
1469 (int) dev
->dev
->caps
.fw_ver
& 0xffff);
1472 static ssize_t
show_rev(struct device
*device
, struct device_attribute
*attr
,
1475 struct mlx4_ib_dev
*dev
=
1476 container_of(device
, struct mlx4_ib_dev
, ib_dev
.dev
);
1477 return sprintf(buf
, "%x\n", dev
->dev
->rev_id
);
1480 static ssize_t
show_board(struct device
*device
, struct device_attribute
*attr
,
1483 struct mlx4_ib_dev
*dev
=
1484 container_of(device
, struct mlx4_ib_dev
, ib_dev
.dev
);
1485 return sprintf(buf
, "%.*s\n", MLX4_BOARD_ID_LEN
,
1486 dev
->dev
->board_id
);
1489 static DEVICE_ATTR(hw_rev
, S_IRUGO
, show_rev
, NULL
);
1490 static DEVICE_ATTR(fw_ver
, S_IRUGO
, show_fw_ver
, NULL
);
1491 static DEVICE_ATTR(hca_type
, S_IRUGO
, show_hca
, NULL
);
1492 static DEVICE_ATTR(board_id
, S_IRUGO
, show_board
, NULL
);
1494 static struct device_attribute
*mlx4_class_attributes
[] = {
1501 static void mlx4_addrconf_ifid_eui48(u8
*eui
, u16 vlan_id
,
1502 struct net_device
*dev
)
1504 memcpy(eui
, dev
->dev_addr
, 3);
1505 memcpy(eui
+ 5, dev
->dev_addr
+ 3, 3);
1506 if (vlan_id
< 0x1000) {
1507 eui
[3] = vlan_id
>> 8;
1508 eui
[4] = vlan_id
& 0xff;
1516 static void update_gids_task(struct work_struct
*work
)
1518 struct update_gid_work
*gw
= container_of(work
, struct update_gid_work
, work
);
1519 struct mlx4_cmd_mailbox
*mailbox
;
1522 struct mlx4_dev
*dev
= gw
->dev
->dev
;
1523 int is_bonded
= mlx4_is_bonded(dev
);
1525 if (!gw
->dev
->ib_active
)
1528 mailbox
= mlx4_alloc_cmd_mailbox(dev
);
1529 if (IS_ERR(mailbox
)) {
1530 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox
));
1534 gids
= mailbox
->buf
;
1535 memcpy(gids
, gw
->gids
, sizeof gw
->gids
);
1537 err
= mlx4_cmd(dev
, mailbox
->dma
, MLX4_SET_PORT_GID_TABLE
<< 8 | gw
->port
,
1538 MLX4_SET_PORT_ETH_OPCODE
, MLX4_CMD_SET_PORT
,
1539 MLX4_CMD_TIME_CLASS_B
, MLX4_CMD_WRAPPED
);
1541 pr_warn("set port command failed\n");
1543 if ((gw
->port
== 1) || !is_bonded
)
1544 mlx4_ib_dispatch_event(gw
->dev
,
1545 is_bonded
? 1 : gw
->port
,
1546 IB_EVENT_GID_CHANGE
);
1548 mlx4_free_cmd_mailbox(dev
, mailbox
);
1552 static void reset_gids_task(struct work_struct
*work
)
1554 struct update_gid_work
*gw
=
1555 container_of(work
, struct update_gid_work
, work
);
1556 struct mlx4_cmd_mailbox
*mailbox
;
1559 struct mlx4_dev
*dev
= gw
->dev
->dev
;
1561 if (!gw
->dev
->ib_active
)
1564 mailbox
= mlx4_alloc_cmd_mailbox(dev
);
1565 if (IS_ERR(mailbox
)) {
1566 pr_warn("reset gid table failed\n");
1570 gids
= mailbox
->buf
;
1571 memcpy(gids
, gw
->gids
, sizeof(gw
->gids
));
1573 if (mlx4_ib_port_link_layer(&gw
->dev
->ib_dev
, gw
->port
) ==
1574 IB_LINK_LAYER_ETHERNET
) {
1575 err
= mlx4_cmd(dev
, mailbox
->dma
,
1576 MLX4_SET_PORT_GID_TABLE
<< 8 | gw
->port
,
1577 MLX4_SET_PORT_ETH_OPCODE
, MLX4_CMD_SET_PORT
,
1578 MLX4_CMD_TIME_CLASS_B
,
1581 pr_warn("set port %d command failed\n", gw
->port
);
1584 mlx4_free_cmd_mailbox(dev
, mailbox
);
1589 static int update_gid_table(struct mlx4_ib_dev
*dev
, int port
,
1590 union ib_gid
*gid
, int clear
,
1593 struct update_gid_work
*work
;
1595 int need_update
= 0;
1603 max_gids
= dev
->dev
->caps
.gid_table_len
[port
];
1604 for (i
= 1; i
< max_gids
; ++i
) {
1605 if (!memcmp(&dev
->iboe
.gid_table
[port
- 1][i
], gid
,
1612 dev
->iboe
.gid_table
[port
- 1][found
] =
1621 !memcmp(&dev
->iboe
.gid_table
[port
- 1][i
],
1622 &zgid
, sizeof(*gid
)))
1628 if (found
== -1 && !clear
&& free
>= 0) {
1629 dev
->iboe
.gid_table
[port
- 1][free
] = *gid
;
1636 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
1640 memcpy(work
->gids
, dev
->iboe
.gid_table
[port
- 1], sizeof(work
->gids
));
1641 INIT_WORK(&work
->work
, update_gids_task
);
1644 queue_work(wq
, &work
->work
);
1649 static void mlx4_make_default_gid(struct net_device
*dev
, union ib_gid
*gid
)
1651 gid
->global
.subnet_prefix
= cpu_to_be64(0xfe80000000000000LL
);
1652 mlx4_addrconf_ifid_eui48(&gid
->raw
[8], 0xffff, dev
);
1656 static int reset_gid_table(struct mlx4_ib_dev
*dev
, u8 port
)
1658 struct update_gid_work
*work
;
1660 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
1664 memset(dev
->iboe
.gid_table
[port
- 1], 0, sizeof(work
->gids
));
1665 memset(work
->gids
, 0, sizeof(work
->gids
));
1666 INIT_WORK(&work
->work
, reset_gids_task
);
1669 queue_work(wq
, &work
->work
);
1673 static int mlx4_ib_addr_event(int event
, struct net_device
*event_netdev
,
1674 struct mlx4_ib_dev
*ibdev
, union ib_gid
*gid
)
1676 struct mlx4_ib_iboe
*iboe
;
1678 struct net_device
*real_dev
= rdma_vlan_dev_real_dev(event_netdev
) ?
1679 rdma_vlan_dev_real_dev(event_netdev
) :
1681 union ib_gid default_gid
;
1683 mlx4_make_default_gid(real_dev
, &default_gid
);
1685 if (!memcmp(gid
, &default_gid
, sizeof(*gid
)))
1688 if (event
!= NETDEV_DOWN
&& event
!= NETDEV_UP
)
1691 if ((real_dev
!= event_netdev
) &&
1692 (event
== NETDEV_DOWN
) &&
1693 rdma_link_local_addr((struct in6_addr
*)gid
))
1696 iboe
= &ibdev
->iboe
;
1697 spin_lock_bh(&iboe
->lock
);
1699 for (port
= 1; port
<= ibdev
->dev
->caps
.num_ports
; ++port
)
1700 if ((netif_is_bond_master(real_dev
) &&
1701 (real_dev
== iboe
->masters
[port
- 1])) ||
1702 (!netif_is_bond_master(real_dev
) &&
1703 (real_dev
== iboe
->netdevs
[port
- 1])))
1704 update_gid_table(ibdev
, port
, gid
,
1705 event
== NETDEV_DOWN
, 0);
1707 spin_unlock_bh(&iboe
->lock
);
1712 static u8
mlx4_ib_get_dev_port(struct net_device
*dev
,
1713 struct mlx4_ib_dev
*ibdev
)
1716 struct mlx4_ib_iboe
*iboe
;
1717 struct net_device
*real_dev
= rdma_vlan_dev_real_dev(dev
) ?
1718 rdma_vlan_dev_real_dev(dev
) : dev
;
1720 iboe
= &ibdev
->iboe
;
1722 for (port
= 1; port
<= ibdev
->dev
->caps
.num_ports
; ++port
)
1723 if ((netif_is_bond_master(real_dev
) &&
1724 (real_dev
== iboe
->masters
[port
- 1])) ||
1725 (!netif_is_bond_master(real_dev
) &&
1726 (real_dev
== iboe
->netdevs
[port
- 1])))
1729 if ((port
== 0) || (port
> ibdev
->dev
->caps
.num_ports
))
1735 static int mlx4_ib_inet_event(struct notifier_block
*this, unsigned long event
,
1738 struct mlx4_ib_dev
*ibdev
;
1739 struct in_ifaddr
*ifa
= ptr
;
1741 struct net_device
*event_netdev
= ifa
->ifa_dev
->dev
;
1743 ipv6_addr_set_v4mapped(ifa
->ifa_address
, (struct in6_addr
*)&gid
);
1745 ibdev
= container_of(this, struct mlx4_ib_dev
, iboe
.nb_inet
);
1747 mlx4_ib_addr_event(event
, event_netdev
, ibdev
, &gid
);
1751 #if IS_ENABLED(CONFIG_IPV6)
1752 static int mlx4_ib_inet6_event(struct notifier_block
*this, unsigned long event
,
1755 struct mlx4_ib_dev
*ibdev
;
1756 struct inet6_ifaddr
*ifa
= ptr
;
1757 union ib_gid
*gid
= (union ib_gid
*)&ifa
->addr
;
1758 struct net_device
*event_netdev
= ifa
->idev
->dev
;
1760 ibdev
= container_of(this, struct mlx4_ib_dev
, iboe
.nb_inet6
);
1762 mlx4_ib_addr_event(event
, event_netdev
, ibdev
, gid
);
1767 #define MLX4_IB_INVALID_MAC ((u64)-1)
1768 static void mlx4_ib_update_qps(struct mlx4_ib_dev
*ibdev
,
1769 struct net_device
*dev
,
1773 u64 release_mac
= MLX4_IB_INVALID_MAC
;
1774 struct mlx4_ib_qp
*qp
;
1776 read_lock(&dev_base_lock
);
1777 new_smac
= mlx4_mac_to_u64(dev
->dev_addr
);
1778 read_unlock(&dev_base_lock
);
1780 atomic64_set(&ibdev
->iboe
.mac
[port
- 1], new_smac
);
1782 /* no need for update QP1 and mac registration in non-SRIOV */
1783 if (!mlx4_is_mfunc(ibdev
->dev
))
1786 mutex_lock(&ibdev
->qp1_proxy_lock
[port
- 1]);
1787 qp
= ibdev
->qp1_proxy
[port
- 1];
1791 struct mlx4_update_qp_params update_params
;
1793 mutex_lock(&qp
->mutex
);
1794 old_smac
= qp
->pri
.smac
;
1795 if (new_smac
== old_smac
)
1798 new_smac_index
= mlx4_register_mac(ibdev
->dev
, port
, new_smac
);
1800 if (new_smac_index
< 0)
1803 update_params
.smac_index
= new_smac_index
;
1804 if (mlx4_update_qp(ibdev
->dev
, qp
->mqp
.qpn
, MLX4_UPDATE_QP_SMAC
,
1806 release_mac
= new_smac
;
1809 /* if old port was zero, no mac was yet registered for this QP */
1810 if (qp
->pri
.smac_port
)
1811 release_mac
= old_smac
;
1812 qp
->pri
.smac
= new_smac
;
1813 qp
->pri
.smac_port
= port
;
1814 qp
->pri
.smac_index
= new_smac_index
;
1818 if (release_mac
!= MLX4_IB_INVALID_MAC
)
1819 mlx4_unregister_mac(ibdev
->dev
, port
, release_mac
);
1821 mutex_unlock(&qp
->mutex
);
1822 mutex_unlock(&ibdev
->qp1_proxy_lock
[port
- 1]);
1825 static void mlx4_ib_get_dev_addr(struct net_device
*dev
,
1826 struct mlx4_ib_dev
*ibdev
, u8 port
)
1828 struct in_device
*in_dev
;
1829 #if IS_ENABLED(CONFIG_IPV6)
1830 struct inet6_dev
*in6_dev
;
1832 struct inet6_ifaddr
*ifp
;
1833 union ib_gid default_gid
;
1838 if ((port
== 0) || (port
> ibdev
->dev
->caps
.num_ports
))
1842 in_dev
= in_dev_get(dev
);
1845 /*ifa->ifa_address;*/
1846 ipv6_addr_set_v4mapped(ifa
->ifa_address
,
1847 (struct in6_addr
*)&gid
);
1848 update_gid_table(ibdev
, port
, &gid
, 0, 0);
1853 #if IS_ENABLED(CONFIG_IPV6)
1854 mlx4_make_default_gid(dev
, &default_gid
);
1856 in6_dev
= in6_dev_get(dev
);
1858 read_lock_bh(&in6_dev
->lock
);
1859 list_for_each_entry(ifp
, &in6_dev
->addr_list
, if_list
) {
1860 pgid
= (union ib_gid
*)&ifp
->addr
;
1861 if (!memcmp(pgid
, &default_gid
, sizeof(*pgid
)))
1863 update_gid_table(ibdev
, port
, pgid
, 0, 0);
1865 read_unlock_bh(&in6_dev
->lock
);
1866 in6_dev_put(in6_dev
);
1871 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev
*ibdev
,
1872 struct net_device
*dev
, u8 port
)
1875 mlx4_make_default_gid(dev
, &gid
);
1876 update_gid_table(ibdev
, port
, &gid
, 0, 1);
1879 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev
*ibdev
)
1881 struct net_device
*dev
;
1882 struct mlx4_ib_iboe
*iboe
= &ibdev
->iboe
;
1886 for (i
= 1; i
<= ibdev
->num_ports
; ++i
) {
1887 if (rdma_port_get_link_layer(&ibdev
->ib_dev
, i
) ==
1888 IB_LINK_LAYER_ETHERNET
) {
1889 err
= reset_gid_table(ibdev
, i
);
1895 read_lock(&dev_base_lock
);
1896 spin_lock_bh(&iboe
->lock
);
1898 for_each_netdev(&init_net
, dev
) {
1899 u8 port
= mlx4_ib_get_dev_port(dev
, ibdev
);
1900 /* port will be non-zero only for ETH ports */
1902 mlx4_ib_set_default_gid(ibdev
, dev
, port
);
1903 mlx4_ib_get_dev_addr(dev
, ibdev
, port
);
1907 spin_unlock_bh(&iboe
->lock
);
1908 read_unlock(&dev_base_lock
);
1913 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev
*ibdev
,
1914 struct net_device
*dev
,
1915 unsigned long event
)
1918 struct mlx4_ib_iboe
*iboe
;
1919 int update_qps_port
= -1;
1922 iboe
= &ibdev
->iboe
;
1924 spin_lock_bh(&iboe
->lock
);
1925 mlx4_foreach_ib_transport_port(port
, ibdev
->dev
) {
1926 enum ib_port_state port_state
= IB_PORT_NOP
;
1927 struct net_device
*old_master
= iboe
->masters
[port
- 1];
1928 struct net_device
*curr_netdev
;
1929 struct net_device
*curr_master
;
1931 iboe
->netdevs
[port
- 1] =
1932 mlx4_get_protocol_dev(ibdev
->dev
, MLX4_PROT_ETH
, port
);
1933 if (iboe
->netdevs
[port
- 1])
1934 mlx4_ib_set_default_gid(ibdev
,
1935 iboe
->netdevs
[port
- 1], port
);
1936 curr_netdev
= iboe
->netdevs
[port
- 1];
1938 if (iboe
->netdevs
[port
- 1] &&
1939 netif_is_bond_slave(iboe
->netdevs
[port
- 1])) {
1940 iboe
->masters
[port
- 1] = netdev_master_upper_dev_get(
1941 iboe
->netdevs
[port
- 1]);
1943 iboe
->masters
[port
- 1] = NULL
;
1945 curr_master
= iboe
->masters
[port
- 1];
1947 if (dev
== iboe
->netdevs
[port
- 1] &&
1948 (event
== NETDEV_CHANGEADDR
|| event
== NETDEV_REGISTER
||
1949 event
== NETDEV_UP
|| event
== NETDEV_CHANGE
))
1950 update_qps_port
= port
;
1953 port_state
= (netif_running(curr_netdev
) && netif_carrier_ok(curr_netdev
)) ?
1954 IB_PORT_ACTIVE
: IB_PORT_DOWN
;
1955 mlx4_ib_set_default_gid(ibdev
, curr_netdev
, port
);
1957 /* if using bonding/team and a slave port is down, we
1958 * don't want the bond IP based gids in the table since
1959 * flows that select port by gid may get the down port.
1961 if (port_state
== IB_PORT_DOWN
&&
1962 !mlx4_is_bonded(ibdev
->dev
)) {
1963 reset_gid_table(ibdev
, port
);
1964 mlx4_ib_set_default_gid(ibdev
,
1968 /* gids from the upper dev (bond/team)
1969 * should appear in port's gid table
1971 mlx4_ib_get_dev_addr(curr_master
,
1975 /* if bonding is used it is possible that we add it to
1976 * masters only after IP address is assigned to the
1977 * net bonding interface.
1979 if (curr_master
&& (old_master
!= curr_master
)) {
1980 reset_gid_table(ibdev
, port
);
1981 mlx4_ib_set_default_gid(ibdev
,
1983 mlx4_ib_get_dev_addr(curr_master
, ibdev
, port
);
1986 if (!curr_master
&& (old_master
!= curr_master
)) {
1987 reset_gid_table(ibdev
, port
);
1988 mlx4_ib_set_default_gid(ibdev
,
1990 mlx4_ib_get_dev_addr(curr_netdev
, ibdev
, port
);
1993 reset_gid_table(ibdev
, port
);
1997 spin_unlock_bh(&iboe
->lock
);
1999 if (update_qps_port
> 0)
2000 mlx4_ib_update_qps(ibdev
, dev
, update_qps_port
);
2003 static int mlx4_ib_netdev_event(struct notifier_block
*this,
2004 unsigned long event
, void *ptr
)
2006 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
2007 struct mlx4_ib_dev
*ibdev
;
2009 if (!net_eq(dev_net(dev
), &init_net
))
2012 ibdev
= container_of(this, struct mlx4_ib_dev
, iboe
.nb
);
2013 mlx4_ib_scan_netdevs(ibdev
, dev
, event
);
2018 static void init_pkeys(struct mlx4_ib_dev
*ibdev
)
2024 if (mlx4_is_master(ibdev
->dev
)) {
2025 for (slave
= 0; slave
<= ibdev
->dev
->persist
->num_vfs
;
2027 for (port
= 1; port
<= ibdev
->dev
->caps
.num_ports
; ++port
) {
2029 i
< ibdev
->dev
->phys_caps
.pkey_phys_table_len
[port
];
2031 ibdev
->pkeys
.virt2phys_pkey
[slave
][port
- 1][i
] =
2032 /* master has the identity virt2phys pkey mapping */
2033 (slave
== mlx4_master_func_num(ibdev
->dev
) || !i
) ? i
:
2034 ibdev
->dev
->phys_caps
.pkey_phys_table_len
[port
] - 1;
2035 mlx4_sync_pkey_table(ibdev
->dev
, slave
, port
, i
,
2036 ibdev
->pkeys
.virt2phys_pkey
[slave
][port
- 1][i
]);
2040 /* initialize pkey cache */
2041 for (port
= 1; port
<= ibdev
->dev
->caps
.num_ports
; ++port
) {
2043 i
< ibdev
->dev
->phys_caps
.pkey_phys_table_len
[port
];
2045 ibdev
->pkeys
.phys_pkey_cache
[port
-1][i
] =
2051 static void mlx4_ib_alloc_eqs(struct mlx4_dev
*dev
, struct mlx4_ib_dev
*ibdev
)
2054 int eq_per_port
= 0;
2059 /* Legacy mode or comp_pool is not large enough */
2060 if (dev
->caps
.comp_pool
== 0 ||
2061 dev
->caps
.num_ports
> dev
->caps
.comp_pool
)
2064 eq_per_port
= dev
->caps
.comp_pool
/ dev
->caps
.num_ports
;
2068 mlx4_foreach_port(i
, dev
, MLX4_PORT_TYPE_IB
)
2069 added_eqs
+= eq_per_port
;
2071 total_eqs
= dev
->caps
.num_comp_vectors
+ added_eqs
;
2073 ibdev
->eq_table
= kzalloc(total_eqs
* sizeof(int), GFP_KERNEL
);
2074 if (!ibdev
->eq_table
)
2077 ibdev
->eq_added
= added_eqs
;
2080 mlx4_foreach_port(i
, dev
, MLX4_PORT_TYPE_IB
) {
2081 for (j
= 0; j
< eq_per_port
; j
++) {
2082 snprintf(name
, sizeof(name
), "mlx4-ib-%d-%d@%s",
2083 i
, j
, dev
->persist
->pdev
->bus
->name
);
2084 /* Set IRQ for specific name (per ring) */
2085 if (mlx4_assign_eq(dev
, name
, NULL
,
2086 &ibdev
->eq_table
[eq
])) {
2087 /* Use legacy (same as mlx4_en driver) */
2088 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq
);
2089 ibdev
->eq_table
[eq
] =
2090 (eq
% dev
->caps
.num_comp_vectors
);
2096 /* Fill the reset of the vector with legacy EQ */
2097 for (i
= 0, eq
= added_eqs
; i
< dev
->caps
.num_comp_vectors
; i
++)
2098 ibdev
->eq_table
[eq
++] = i
;
2100 /* Advertise the new number of EQs to clients */
2101 ibdev
->ib_dev
.num_comp_vectors
= total_eqs
;
2104 static void mlx4_ib_free_eqs(struct mlx4_dev
*dev
, struct mlx4_ib_dev
*ibdev
)
2108 /* no additional eqs were added */
2109 if (!ibdev
->eq_table
)
2112 /* Reset the advertised EQ number */
2113 ibdev
->ib_dev
.num_comp_vectors
= dev
->caps
.num_comp_vectors
;
2115 /* Free only the added eqs */
2116 for (i
= 0; i
< ibdev
->eq_added
; i
++) {
2117 /* Don't free legacy eqs if used */
2118 if (ibdev
->eq_table
[i
] <= dev
->caps
.num_comp_vectors
)
2120 mlx4_release_eq(dev
, ibdev
->eq_table
[i
]);
2123 kfree(ibdev
->eq_table
);
2126 static int mlx4_port_immutable(struct ib_device
*ibdev
, u8 port_num
,
2127 struct ib_port_immutable
*immutable
)
2129 struct ib_port_attr attr
;
2132 err
= mlx4_ib_query_port(ibdev
, port_num
, &attr
);
2136 immutable
->pkey_tbl_len
= attr
.pkey_tbl_len
;
2137 immutable
->gid_tbl_len
= attr
.gid_tbl_len
;
2142 static void *mlx4_ib_add(struct mlx4_dev
*dev
)
2144 struct mlx4_ib_dev
*ibdev
;
2148 struct mlx4_ib_iboe
*iboe
;
2149 int ib_num_ports
= 0;
2150 int num_req_counters
;
2152 pr_info_once("%s", mlx4_ib_version
);
2155 mlx4_foreach_ib_transport_port(i
, dev
)
2158 /* No point in registering a device with no ports... */
2162 ibdev
= (struct mlx4_ib_dev
*) ib_alloc_device(sizeof *ibdev
);
2164 dev_err(&dev
->persist
->pdev
->dev
,
2165 "Device struct alloc failed\n");
2169 iboe
= &ibdev
->iboe
;
2171 if (mlx4_pd_alloc(dev
, &ibdev
->priv_pdn
))
2174 if (mlx4_uar_alloc(dev
, &ibdev
->priv_uar
))
2177 ibdev
->uar_map
= ioremap((phys_addr_t
) ibdev
->priv_uar
.pfn
<< PAGE_SHIFT
,
2179 if (!ibdev
->uar_map
)
2181 MLX4_INIT_DOORBELL_LOCK(&ibdev
->uar_lock
);
2184 ibdev
->bond_next_port
= 0;
2186 strlcpy(ibdev
->ib_dev
.name
, "mlx4_%d", IB_DEVICE_NAME_MAX
);
2187 ibdev
->ib_dev
.owner
= THIS_MODULE
;
2188 ibdev
->ib_dev
.node_type
= RDMA_NODE_IB_CA
;
2189 ibdev
->ib_dev
.local_dma_lkey
= dev
->caps
.reserved_lkey
;
2190 ibdev
->num_ports
= num_ports
;
2191 ibdev
->ib_dev
.phys_port_cnt
= mlx4_is_bonded(dev
) ?
2192 1 : ibdev
->num_ports
;
2193 ibdev
->ib_dev
.num_comp_vectors
= dev
->caps
.num_comp_vectors
;
2194 ibdev
->ib_dev
.dma_device
= &dev
->persist
->pdev
->dev
;
2196 if (dev
->caps
.userspace_caps
)
2197 ibdev
->ib_dev
.uverbs_abi_ver
= MLX4_IB_UVERBS_ABI_VERSION
;
2199 ibdev
->ib_dev
.uverbs_abi_ver
= MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION
;
2201 ibdev
->ib_dev
.uverbs_cmd_mask
=
2202 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT
) |
2203 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE
) |
2204 (1ull << IB_USER_VERBS_CMD_QUERY_PORT
) |
2205 (1ull << IB_USER_VERBS_CMD_ALLOC_PD
) |
2206 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD
) |
2207 (1ull << IB_USER_VERBS_CMD_REG_MR
) |
2208 (1ull << IB_USER_VERBS_CMD_REREG_MR
) |
2209 (1ull << IB_USER_VERBS_CMD_DEREG_MR
) |
2210 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL
) |
2211 (1ull << IB_USER_VERBS_CMD_CREATE_CQ
) |
2212 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ
) |
2213 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ
) |
2214 (1ull << IB_USER_VERBS_CMD_CREATE_QP
) |
2215 (1ull << IB_USER_VERBS_CMD_MODIFY_QP
) |
2216 (1ull << IB_USER_VERBS_CMD_QUERY_QP
) |
2217 (1ull << IB_USER_VERBS_CMD_DESTROY_QP
) |
2218 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST
) |
2219 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST
) |
2220 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ
) |
2221 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ
) |
2222 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ
) |
2223 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ
) |
2224 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ
) |
2225 (1ull << IB_USER_VERBS_CMD_OPEN_QP
);
2227 ibdev
->ib_dev
.query_device
= mlx4_ib_query_device
;
2228 ibdev
->ib_dev
.query_port
= mlx4_ib_query_port
;
2229 ibdev
->ib_dev
.query_protocol
= mlx4_ib_query_protocol
;
2230 ibdev
->ib_dev
.get_link_layer
= mlx4_ib_port_link_layer
;
2231 ibdev
->ib_dev
.query_gid
= mlx4_ib_query_gid
;
2232 ibdev
->ib_dev
.query_pkey
= mlx4_ib_query_pkey
;
2233 ibdev
->ib_dev
.modify_device
= mlx4_ib_modify_device
;
2234 ibdev
->ib_dev
.modify_port
= mlx4_ib_modify_port
;
2235 ibdev
->ib_dev
.alloc_ucontext
= mlx4_ib_alloc_ucontext
;
2236 ibdev
->ib_dev
.dealloc_ucontext
= mlx4_ib_dealloc_ucontext
;
2237 ibdev
->ib_dev
.mmap
= mlx4_ib_mmap
;
2238 ibdev
->ib_dev
.alloc_pd
= mlx4_ib_alloc_pd
;
2239 ibdev
->ib_dev
.dealloc_pd
= mlx4_ib_dealloc_pd
;
2240 ibdev
->ib_dev
.create_ah
= mlx4_ib_create_ah
;
2241 ibdev
->ib_dev
.query_ah
= mlx4_ib_query_ah
;
2242 ibdev
->ib_dev
.destroy_ah
= mlx4_ib_destroy_ah
;
2243 ibdev
->ib_dev
.create_srq
= mlx4_ib_create_srq
;
2244 ibdev
->ib_dev
.modify_srq
= mlx4_ib_modify_srq
;
2245 ibdev
->ib_dev
.query_srq
= mlx4_ib_query_srq
;
2246 ibdev
->ib_dev
.destroy_srq
= mlx4_ib_destroy_srq
;
2247 ibdev
->ib_dev
.post_srq_recv
= mlx4_ib_post_srq_recv
;
2248 ibdev
->ib_dev
.create_qp
= mlx4_ib_create_qp
;
2249 ibdev
->ib_dev
.modify_qp
= mlx4_ib_modify_qp
;
2250 ibdev
->ib_dev
.query_qp
= mlx4_ib_query_qp
;
2251 ibdev
->ib_dev
.destroy_qp
= mlx4_ib_destroy_qp
;
2252 ibdev
->ib_dev
.post_send
= mlx4_ib_post_send
;
2253 ibdev
->ib_dev
.post_recv
= mlx4_ib_post_recv
;
2254 ibdev
->ib_dev
.create_cq
= mlx4_ib_create_cq
;
2255 ibdev
->ib_dev
.modify_cq
= mlx4_ib_modify_cq
;
2256 ibdev
->ib_dev
.resize_cq
= mlx4_ib_resize_cq
;
2257 ibdev
->ib_dev
.destroy_cq
= mlx4_ib_destroy_cq
;
2258 ibdev
->ib_dev
.poll_cq
= mlx4_ib_poll_cq
;
2259 ibdev
->ib_dev
.req_notify_cq
= mlx4_ib_arm_cq
;
2260 ibdev
->ib_dev
.get_dma_mr
= mlx4_ib_get_dma_mr
;
2261 ibdev
->ib_dev
.reg_user_mr
= mlx4_ib_reg_user_mr
;
2262 ibdev
->ib_dev
.rereg_user_mr
= mlx4_ib_rereg_user_mr
;
2263 ibdev
->ib_dev
.dereg_mr
= mlx4_ib_dereg_mr
;
2264 ibdev
->ib_dev
.alloc_fast_reg_mr
= mlx4_ib_alloc_fast_reg_mr
;
2265 ibdev
->ib_dev
.alloc_fast_reg_page_list
= mlx4_ib_alloc_fast_reg_page_list
;
2266 ibdev
->ib_dev
.free_fast_reg_page_list
= mlx4_ib_free_fast_reg_page_list
;
2267 ibdev
->ib_dev
.attach_mcast
= mlx4_ib_mcg_attach
;
2268 ibdev
->ib_dev
.detach_mcast
= mlx4_ib_mcg_detach
;
2269 ibdev
->ib_dev
.process_mad
= mlx4_ib_process_mad
;
2270 ibdev
->ib_dev
.get_port_immutable
= mlx4_port_immutable
;
2272 if (!mlx4_is_slave(ibdev
->dev
)) {
2273 ibdev
->ib_dev
.alloc_fmr
= mlx4_ib_fmr_alloc
;
2274 ibdev
->ib_dev
.map_phys_fmr
= mlx4_ib_map_phys_fmr
;
2275 ibdev
->ib_dev
.unmap_fmr
= mlx4_ib_unmap_fmr
;
2276 ibdev
->ib_dev
.dealloc_fmr
= mlx4_ib_fmr_dealloc
;
2279 if (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_MEM_WINDOW
||
2280 dev
->caps
.bmme_flags
& MLX4_BMME_FLAG_TYPE_2_WIN
) {
2281 ibdev
->ib_dev
.alloc_mw
= mlx4_ib_alloc_mw
;
2282 ibdev
->ib_dev
.bind_mw
= mlx4_ib_bind_mw
;
2283 ibdev
->ib_dev
.dealloc_mw
= mlx4_ib_dealloc_mw
;
2285 ibdev
->ib_dev
.uverbs_cmd_mask
|=
2286 (1ull << IB_USER_VERBS_CMD_ALLOC_MW
) |
2287 (1ull << IB_USER_VERBS_CMD_DEALLOC_MW
);
2290 if (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_XRC
) {
2291 ibdev
->ib_dev
.alloc_xrcd
= mlx4_ib_alloc_xrcd
;
2292 ibdev
->ib_dev
.dealloc_xrcd
= mlx4_ib_dealloc_xrcd
;
2293 ibdev
->ib_dev
.uverbs_cmd_mask
|=
2294 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD
) |
2295 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD
);
2298 if (check_flow_steering_support(dev
)) {
2299 ibdev
->steering_support
= MLX4_STEERING_MODE_DEVICE_MANAGED
;
2300 ibdev
->ib_dev
.create_flow
= mlx4_ib_create_flow
;
2301 ibdev
->ib_dev
.destroy_flow
= mlx4_ib_destroy_flow
;
2303 ibdev
->ib_dev
.uverbs_ex_cmd_mask
|=
2304 (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW
) |
2305 (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW
);
2308 mlx4_ib_alloc_eqs(dev
, ibdev
);
2310 spin_lock_init(&iboe
->lock
);
2312 if (init_node_data(ibdev
))
2315 num_req_counters
= mlx4_is_bonded(dev
) ? 1 : ibdev
->num_ports
;
2316 for (i
= 0; i
< num_req_counters
; ++i
) {
2317 mutex_init(&ibdev
->qp1_proxy_lock
[i
]);
2318 if (mlx4_ib_port_link_layer(&ibdev
->ib_dev
, i
+ 1) ==
2319 IB_LINK_LAYER_ETHERNET
) {
2320 err
= mlx4_counter_alloc(ibdev
->dev
, &ibdev
->counters
[i
]);
2322 ibdev
->counters
[i
] = -1;
2324 ibdev
->counters
[i
] = -1;
2327 if (mlx4_is_bonded(dev
))
2328 for (i
= 1; i
< ibdev
->num_ports
; ++i
)
2329 ibdev
->counters
[i
] = ibdev
->counters
[0];
2332 mlx4_foreach_port(i
, dev
, MLX4_PORT_TYPE_IB
)
2335 spin_lock_init(&ibdev
->sm_lock
);
2336 mutex_init(&ibdev
->cap_mask_mutex
);
2337 INIT_LIST_HEAD(&ibdev
->qp_list
);
2338 spin_lock_init(&ibdev
->reset_flow_resource_lock
);
2340 if (ibdev
->steering_support
== MLX4_STEERING_MODE_DEVICE_MANAGED
&&
2342 ibdev
->steer_qpn_count
= MLX4_IB_UC_MAX_NUM_QPS
;
2343 err
= mlx4_qp_reserve_range(dev
, ibdev
->steer_qpn_count
,
2344 MLX4_IB_UC_STEER_QPN_ALIGN
,
2345 &ibdev
->steer_qpn_base
, 0);
2349 ibdev
->ib_uc_qpns_bitmap
=
2350 kmalloc(BITS_TO_LONGS(ibdev
->steer_qpn_count
) *
2353 if (!ibdev
->ib_uc_qpns_bitmap
) {
2354 dev_err(&dev
->persist
->pdev
->dev
,
2355 "bit map alloc failed\n");
2356 goto err_steer_qp_release
;
2359 bitmap_zero(ibdev
->ib_uc_qpns_bitmap
, ibdev
->steer_qpn_count
);
2361 err
= mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2362 dev
, ibdev
->steer_qpn_base
,
2363 ibdev
->steer_qpn_base
+
2364 ibdev
->steer_qpn_count
- 1);
2366 goto err_steer_free_bitmap
;
2369 for (j
= 1; j
<= ibdev
->dev
->caps
.num_ports
; j
++)
2370 atomic64_set(&iboe
->mac
[j
- 1], ibdev
->dev
->caps
.def_mac
[j
]);
2372 if (ib_register_device(&ibdev
->ib_dev
, NULL
))
2373 goto err_steer_free_bitmap
;
2375 if (mlx4_ib_mad_init(ibdev
))
2378 if (mlx4_ib_init_sriov(ibdev
))
2381 if (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_IBOE
) {
2382 if (!iboe
->nb
.notifier_call
) {
2383 iboe
->nb
.notifier_call
= mlx4_ib_netdev_event
;
2384 err
= register_netdevice_notifier(&iboe
->nb
);
2386 iboe
->nb
.notifier_call
= NULL
;
2390 if (!iboe
->nb_inet
.notifier_call
) {
2391 iboe
->nb_inet
.notifier_call
= mlx4_ib_inet_event
;
2392 err
= register_inetaddr_notifier(&iboe
->nb_inet
);
2394 iboe
->nb_inet
.notifier_call
= NULL
;
2398 #if IS_ENABLED(CONFIG_IPV6)
2399 if (!iboe
->nb_inet6
.notifier_call
) {
2400 iboe
->nb_inet6
.notifier_call
= mlx4_ib_inet6_event
;
2401 err
= register_inet6addr_notifier(&iboe
->nb_inet6
);
2403 iboe
->nb_inet6
.notifier_call
= NULL
;
2408 if (mlx4_ib_init_gid_table(ibdev
))
2412 for (j
= 0; j
< ARRAY_SIZE(mlx4_class_attributes
); ++j
) {
2413 if (device_create_file(&ibdev
->ib_dev
.dev
,
2414 mlx4_class_attributes
[j
]))
2418 ibdev
->ib_active
= true;
2420 if (mlx4_is_mfunc(ibdev
->dev
))
2423 /* create paravirt contexts for any VFs which are active */
2424 if (mlx4_is_master(ibdev
->dev
)) {
2425 for (j
= 0; j
< MLX4_MFUNC_MAX
; j
++) {
2426 if (j
== mlx4_master_func_num(ibdev
->dev
))
2428 if (mlx4_is_slave_active(ibdev
->dev
, j
))
2429 do_slave_init(ibdev
, j
, 1);
2435 if (ibdev
->iboe
.nb
.notifier_call
) {
2436 if (unregister_netdevice_notifier(&ibdev
->iboe
.nb
))
2437 pr_warn("failure unregistering notifier\n");
2438 ibdev
->iboe
.nb
.notifier_call
= NULL
;
2440 if (ibdev
->iboe
.nb_inet
.notifier_call
) {
2441 if (unregister_inetaddr_notifier(&ibdev
->iboe
.nb_inet
))
2442 pr_warn("failure unregistering notifier\n");
2443 ibdev
->iboe
.nb_inet
.notifier_call
= NULL
;
2445 #if IS_ENABLED(CONFIG_IPV6)
2446 if (ibdev
->iboe
.nb_inet6
.notifier_call
) {
2447 if (unregister_inet6addr_notifier(&ibdev
->iboe
.nb_inet6
))
2448 pr_warn("failure unregistering notifier\n");
2449 ibdev
->iboe
.nb_inet6
.notifier_call
= NULL
;
2452 flush_workqueue(wq
);
2454 mlx4_ib_close_sriov(ibdev
);
2457 mlx4_ib_mad_cleanup(ibdev
);
2460 ib_unregister_device(&ibdev
->ib_dev
);
2462 err_steer_free_bitmap
:
2463 kfree(ibdev
->ib_uc_qpns_bitmap
);
2465 err_steer_qp_release
:
2466 if (ibdev
->steering_support
== MLX4_STEERING_MODE_DEVICE_MANAGED
)
2467 mlx4_qp_release_range(dev
, ibdev
->steer_qpn_base
,
2468 ibdev
->steer_qpn_count
);
2471 if (ibdev
->counters
[i
- 1] != -1)
2472 mlx4_counter_free(ibdev
->dev
, ibdev
->counters
[i
- 1]);
2475 iounmap(ibdev
->uar_map
);
2478 mlx4_uar_free(dev
, &ibdev
->priv_uar
);
2481 mlx4_pd_free(dev
, ibdev
->priv_pdn
);
2484 ib_dealloc_device(&ibdev
->ib_dev
);
2489 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev
*dev
, int count
, int *qpn
)
2493 WARN_ON(!dev
->ib_uc_qpns_bitmap
);
2495 offset
= bitmap_find_free_region(dev
->ib_uc_qpns_bitmap
,
2496 dev
->steer_qpn_count
,
2497 get_count_order(count
));
2501 *qpn
= dev
->steer_qpn_base
+ offset
;
2505 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev
*dev
, u32 qpn
, int count
)
2508 dev
->steering_support
!= MLX4_STEERING_MODE_DEVICE_MANAGED
)
2511 BUG_ON(qpn
< dev
->steer_qpn_base
);
2513 bitmap_release_region(dev
->ib_uc_qpns_bitmap
,
2514 qpn
- dev
->steer_qpn_base
,
2515 get_count_order(count
));
2518 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev
*mdev
, struct mlx4_ib_qp
*mqp
,
2523 struct ib_flow_attr
*flow
= NULL
;
2524 struct ib_flow_spec_ib
*ib_spec
;
2527 flow_size
= sizeof(struct ib_flow_attr
) +
2528 sizeof(struct ib_flow_spec_ib
);
2529 flow
= kzalloc(flow_size
, GFP_KERNEL
);
2532 flow
->port
= mqp
->port
;
2533 flow
->num_of_specs
= 1;
2534 flow
->size
= flow_size
;
2535 ib_spec
= (struct ib_flow_spec_ib
*)(flow
+ 1);
2536 ib_spec
->type
= IB_FLOW_SPEC_IB
;
2537 ib_spec
->size
= sizeof(struct ib_flow_spec_ib
);
2538 /* Add an empty rule for IB L2 */
2539 memset(&ib_spec
->mask
, 0, sizeof(ib_spec
->mask
));
2541 err
= __mlx4_ib_create_flow(&mqp
->ibqp
, flow
,
2546 err
= __mlx4_ib_destroy_flow(mdev
->dev
, mqp
->reg_id
);
2552 static void mlx4_ib_remove(struct mlx4_dev
*dev
, void *ibdev_ptr
)
2554 struct mlx4_ib_dev
*ibdev
= ibdev_ptr
;
2557 ibdev
->ib_active
= false;
2558 flush_workqueue(wq
);
2560 mlx4_ib_close_sriov(ibdev
);
2561 mlx4_ib_mad_cleanup(ibdev
);
2562 ib_unregister_device(&ibdev
->ib_dev
);
2563 if (ibdev
->iboe
.nb
.notifier_call
) {
2564 if (unregister_netdevice_notifier(&ibdev
->iboe
.nb
))
2565 pr_warn("failure unregistering notifier\n");
2566 ibdev
->iboe
.nb
.notifier_call
= NULL
;
2569 if (ibdev
->steering_support
== MLX4_STEERING_MODE_DEVICE_MANAGED
) {
2570 mlx4_qp_release_range(dev
, ibdev
->steer_qpn_base
,
2571 ibdev
->steer_qpn_count
);
2572 kfree(ibdev
->ib_uc_qpns_bitmap
);
2575 if (ibdev
->iboe
.nb_inet
.notifier_call
) {
2576 if (unregister_inetaddr_notifier(&ibdev
->iboe
.nb_inet
))
2577 pr_warn("failure unregistering notifier\n");
2578 ibdev
->iboe
.nb_inet
.notifier_call
= NULL
;
2580 #if IS_ENABLED(CONFIG_IPV6)
2581 if (ibdev
->iboe
.nb_inet6
.notifier_call
) {
2582 if (unregister_inet6addr_notifier(&ibdev
->iboe
.nb_inet6
))
2583 pr_warn("failure unregistering notifier\n");
2584 ibdev
->iboe
.nb_inet6
.notifier_call
= NULL
;
2588 iounmap(ibdev
->uar_map
);
2589 for (p
= 0; p
< ibdev
->num_ports
; ++p
)
2590 if (ibdev
->counters
[p
] != -1)
2591 mlx4_counter_free(ibdev
->dev
, ibdev
->counters
[p
]);
2592 mlx4_foreach_port(p
, dev
, MLX4_PORT_TYPE_IB
)
2593 mlx4_CLOSE_PORT(dev
, p
);
2595 mlx4_ib_free_eqs(dev
, ibdev
);
2597 mlx4_uar_free(dev
, &ibdev
->priv_uar
);
2598 mlx4_pd_free(dev
, ibdev
->priv_pdn
);
2599 ib_dealloc_device(&ibdev
->ib_dev
);
2602 static void do_slave_init(struct mlx4_ib_dev
*ibdev
, int slave
, int do_init
)
2604 struct mlx4_ib_demux_work
**dm
= NULL
;
2605 struct mlx4_dev
*dev
= ibdev
->dev
;
2607 unsigned long flags
;
2608 struct mlx4_active_ports actv_ports
;
2610 unsigned int first_port
;
2612 if (!mlx4_is_master(dev
))
2615 actv_ports
= mlx4_get_active_ports(dev
, slave
);
2616 ports
= bitmap_weight(actv_ports
.ports
, dev
->caps
.num_ports
);
2617 first_port
= find_first_bit(actv_ports
.ports
, dev
->caps
.num_ports
);
2619 dm
= kcalloc(ports
, sizeof(*dm
), GFP_ATOMIC
);
2621 pr_err("failed to allocate memory for tunneling qp update\n");
2625 for (i
= 0; i
< ports
; i
++) {
2626 dm
[i
] = kmalloc(sizeof (struct mlx4_ib_demux_work
), GFP_ATOMIC
);
2628 pr_err("failed to allocate memory for tunneling qp update work struct\n");
2629 for (i
= 0; i
< dev
->caps
.num_ports
; i
++) {
2636 /* initialize or tear down tunnel QPs for the slave */
2637 for (i
= 0; i
< ports
; i
++) {
2638 INIT_WORK(&dm
[i
]->work
, mlx4_ib_tunnels_update_work
);
2639 dm
[i
]->port
= first_port
+ i
+ 1;
2640 dm
[i
]->slave
= slave
;
2641 dm
[i
]->do_init
= do_init
;
2643 spin_lock_irqsave(&ibdev
->sriov
.going_down_lock
, flags
);
2644 if (!ibdev
->sriov
.is_going_down
)
2645 queue_work(ibdev
->sriov
.demux
[i
].ud_wq
, &dm
[i
]->work
);
2646 spin_unlock_irqrestore(&ibdev
->sriov
.going_down_lock
, flags
);
2653 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev
*ibdev
)
2655 struct mlx4_ib_qp
*mqp
;
2656 unsigned long flags_qp
;
2657 unsigned long flags_cq
;
2658 struct mlx4_ib_cq
*send_mcq
, *recv_mcq
;
2659 struct list_head cq_notify_list
;
2660 struct mlx4_cq
*mcq
;
2661 unsigned long flags
;
2663 pr_warn("mlx4_ib_handle_catas_error was started\n");
2664 INIT_LIST_HEAD(&cq_notify_list
);
2666 /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2667 spin_lock_irqsave(&ibdev
->reset_flow_resource_lock
, flags
);
2669 list_for_each_entry(mqp
, &ibdev
->qp_list
, qps_list
) {
2670 spin_lock_irqsave(&mqp
->sq
.lock
, flags_qp
);
2671 if (mqp
->sq
.tail
!= mqp
->sq
.head
) {
2672 send_mcq
= to_mcq(mqp
->ibqp
.send_cq
);
2673 spin_lock_irqsave(&send_mcq
->lock
, flags_cq
);
2674 if (send_mcq
->mcq
.comp
&&
2675 mqp
->ibqp
.send_cq
->comp_handler
) {
2676 if (!send_mcq
->mcq
.reset_notify_added
) {
2677 send_mcq
->mcq
.reset_notify_added
= 1;
2678 list_add_tail(&send_mcq
->mcq
.reset_notify
,
2682 spin_unlock_irqrestore(&send_mcq
->lock
, flags_cq
);
2684 spin_unlock_irqrestore(&mqp
->sq
.lock
, flags_qp
);
2685 /* Now, handle the QP's receive queue */
2686 spin_lock_irqsave(&mqp
->rq
.lock
, flags_qp
);
2687 /* no handling is needed for SRQ */
2688 if (!mqp
->ibqp
.srq
) {
2689 if (mqp
->rq
.tail
!= mqp
->rq
.head
) {
2690 recv_mcq
= to_mcq(mqp
->ibqp
.recv_cq
);
2691 spin_lock_irqsave(&recv_mcq
->lock
, flags_cq
);
2692 if (recv_mcq
->mcq
.comp
&&
2693 mqp
->ibqp
.recv_cq
->comp_handler
) {
2694 if (!recv_mcq
->mcq
.reset_notify_added
) {
2695 recv_mcq
->mcq
.reset_notify_added
= 1;
2696 list_add_tail(&recv_mcq
->mcq
.reset_notify
,
2700 spin_unlock_irqrestore(&recv_mcq
->lock
,
2704 spin_unlock_irqrestore(&mqp
->rq
.lock
, flags_qp
);
2707 list_for_each_entry(mcq
, &cq_notify_list
, reset_notify
) {
2710 spin_unlock_irqrestore(&ibdev
->reset_flow_resource_lock
, flags
);
2711 pr_warn("mlx4_ib_handle_catas_error ended\n");
2714 static void handle_bonded_port_state_event(struct work_struct
*work
)
2716 struct ib_event_work
*ew
=
2717 container_of(work
, struct ib_event_work
, work
);
2718 struct mlx4_ib_dev
*ibdev
= ew
->ib_dev
;
2719 enum ib_port_state bonded_port_state
= IB_PORT_NOP
;
2721 struct ib_event ibev
;
2724 spin_lock_bh(&ibdev
->iboe
.lock
);
2725 for (i
= 0; i
< MLX4_MAX_PORTS
; ++i
) {
2726 struct net_device
*curr_netdev
= ibdev
->iboe
.netdevs
[i
];
2727 enum ib_port_state curr_port_state
;
2733 (netif_running(curr_netdev
) &&
2734 netif_carrier_ok(curr_netdev
)) ?
2735 IB_PORT_ACTIVE
: IB_PORT_DOWN
;
2737 bonded_port_state
= (bonded_port_state
!= IB_PORT_ACTIVE
) ?
2738 curr_port_state
: IB_PORT_ACTIVE
;
2740 spin_unlock_bh(&ibdev
->iboe
.lock
);
2742 ibev
.device
= &ibdev
->ib_dev
;
2743 ibev
.element
.port_num
= 1;
2744 ibev
.event
= (bonded_port_state
== IB_PORT_ACTIVE
) ?
2745 IB_EVENT_PORT_ACTIVE
: IB_EVENT_PORT_ERR
;
2747 ib_dispatch_event(&ibev
);
2750 static void mlx4_ib_event(struct mlx4_dev
*dev
, void *ibdev_ptr
,
2751 enum mlx4_dev_event event
, unsigned long param
)
2753 struct ib_event ibev
;
2754 struct mlx4_ib_dev
*ibdev
= to_mdev((struct ib_device
*) ibdev_ptr
);
2755 struct mlx4_eqe
*eqe
= NULL
;
2756 struct ib_event_work
*ew
;
2759 if (mlx4_is_bonded(dev
) &&
2760 ((event
== MLX4_DEV_EVENT_PORT_UP
) ||
2761 (event
== MLX4_DEV_EVENT_PORT_DOWN
))) {
2762 ew
= kmalloc(sizeof(*ew
), GFP_ATOMIC
);
2765 INIT_WORK(&ew
->work
, handle_bonded_port_state_event
);
2767 queue_work(wq
, &ew
->work
);
2771 if (event
== MLX4_DEV_EVENT_PORT_MGMT_CHANGE
)
2772 eqe
= (struct mlx4_eqe
*)param
;
2777 case MLX4_DEV_EVENT_PORT_UP
:
2778 if (p
> ibdev
->num_ports
)
2780 if (mlx4_is_master(dev
) &&
2781 rdma_port_get_link_layer(&ibdev
->ib_dev
, p
) ==
2782 IB_LINK_LAYER_INFINIBAND
) {
2783 mlx4_ib_invalidate_all_guid_record(ibdev
, p
);
2785 ibev
.event
= IB_EVENT_PORT_ACTIVE
;
2788 case MLX4_DEV_EVENT_PORT_DOWN
:
2789 if (p
> ibdev
->num_ports
)
2791 ibev
.event
= IB_EVENT_PORT_ERR
;
2794 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR
:
2795 ibdev
->ib_active
= false;
2796 ibev
.event
= IB_EVENT_DEVICE_FATAL
;
2797 mlx4_ib_handle_catas_error(ibdev
);
2800 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE
:
2801 ew
= kmalloc(sizeof *ew
, GFP_ATOMIC
);
2803 pr_err("failed to allocate memory for events work\n");
2807 INIT_WORK(&ew
->work
, handle_port_mgmt_change_event
);
2808 memcpy(&ew
->ib_eqe
, eqe
, sizeof *eqe
);
2810 /* need to queue only for port owner, which uses GEN_EQE */
2811 if (mlx4_is_master(dev
))
2812 queue_work(wq
, &ew
->work
);
2814 handle_port_mgmt_change_event(&ew
->work
);
2817 case MLX4_DEV_EVENT_SLAVE_INIT
:
2818 /* here, p is the slave id */
2819 do_slave_init(ibdev
, p
, 1);
2820 if (mlx4_is_master(dev
)) {
2823 for (i
= 1; i
<= ibdev
->num_ports
; i
++) {
2824 if (rdma_port_get_link_layer(&ibdev
->ib_dev
, i
)
2825 == IB_LINK_LAYER_INFINIBAND
)
2826 mlx4_ib_slave_alias_guid_event(ibdev
,
2833 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN
:
2834 if (mlx4_is_master(dev
)) {
2837 for (i
= 1; i
<= ibdev
->num_ports
; i
++) {
2838 if (rdma_port_get_link_layer(&ibdev
->ib_dev
, i
)
2839 == IB_LINK_LAYER_INFINIBAND
)
2840 mlx4_ib_slave_alias_guid_event(ibdev
,
2845 /* here, p is the slave id */
2846 do_slave_init(ibdev
, p
, 0);
2853 ibev
.device
= ibdev_ptr
;
2854 ibev
.element
.port_num
= mlx4_is_bonded(ibdev
->dev
) ? 1 : (u8
)p
;
2856 ib_dispatch_event(&ibev
);
2859 static struct mlx4_interface mlx4_ib_interface
= {
2861 .remove
= mlx4_ib_remove
,
2862 .event
= mlx4_ib_event
,
2863 .protocol
= MLX4_PROT_IB_IPV6
,
2864 .flags
= MLX4_INTFF_BONDING
2867 static int __init
mlx4_ib_init(void)
2871 wq
= create_singlethread_workqueue("mlx4_ib");
2875 err
= mlx4_ib_mcg_init();
2879 err
= mlx4_register_interface(&mlx4_ib_interface
);
2886 mlx4_ib_mcg_destroy();
2889 destroy_workqueue(wq
);
2893 static void __exit
mlx4_ib_cleanup(void)
2895 mlx4_unregister_interface(&mlx4_ib_interface
);
2896 mlx4_ib_mcg_destroy();
2897 destroy_workqueue(wq
);
2900 module_init(mlx4_ib_init
);
2901 module_exit(mlx4_ib_cleanup
);