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>
44 #include <net/devlink.h>
46 #include <rdma/ib_smi.h>
47 #include <rdma/ib_user_verbs.h>
48 #include <rdma/ib_addr.h>
49 #include <rdma/ib_cache.h>
51 #include <net/bonding.h>
53 #include <linux/mlx4/driver.h>
54 #include <linux/mlx4/cmd.h>
55 #include <linux/mlx4/qp.h>
60 #define DRV_NAME MLX4_IB_DRV_NAME
61 #define DRV_VERSION "2.2-1"
62 #define DRV_RELDATE "Feb 2014"
64 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
65 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
66 #define MLX4_IB_CARD_REV_A0 0xA0
68 MODULE_AUTHOR("Roland Dreier");
69 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
70 MODULE_LICENSE("Dual BSD/GPL");
71 MODULE_VERSION(DRV_VERSION
);
73 int mlx4_ib_sm_guid_assign
= 0;
74 module_param_named(sm_guid_assign
, mlx4_ib_sm_guid_assign
, int, 0444);
75 MODULE_PARM_DESC(sm_guid_assign
, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
77 static const char mlx4_ib_version
[] =
78 DRV_NAME
": Mellanox ConnectX InfiniBand driver v"
79 DRV_VERSION
" (" DRV_RELDATE
")\n";
81 static void do_slave_init(struct mlx4_ib_dev
*ibdev
, int slave
, int do_init
);
83 static struct workqueue_struct
*wq
;
85 static void init_query_mad(struct ib_smp
*mad
)
87 mad
->base_version
= 1;
88 mad
->mgmt_class
= IB_MGMT_CLASS_SUBN_LID_ROUTED
;
89 mad
->class_version
= 1;
90 mad
->method
= IB_MGMT_METHOD_GET
;
93 static int check_flow_steering_support(struct mlx4_dev
*dev
)
95 int eth_num_ports
= 0;
98 int dmfs
= dev
->caps
.steering_mode
== MLX4_STEERING_MODE_DEVICE_MANAGED
;
102 mlx4_foreach_port(i
, dev
, MLX4_PORT_TYPE_ETH
)
104 mlx4_foreach_port(i
, dev
, MLX4_PORT_TYPE_IB
)
106 dmfs
&= (!ib_num_ports
||
107 (dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_DMFS_IPOIB
)) &&
109 (dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_FS_EN
));
110 if (ib_num_ports
&& mlx4_is_mfunc(dev
)) {
111 pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
118 static int num_ib_ports(struct mlx4_dev
*dev
)
123 mlx4_foreach_port(i
, dev
, MLX4_PORT_TYPE_IB
)
129 static struct net_device
*mlx4_ib_get_netdev(struct ib_device
*device
, u8 port_num
)
131 struct mlx4_ib_dev
*ibdev
= to_mdev(device
);
132 struct net_device
*dev
;
135 dev
= mlx4_get_protocol_dev(ibdev
->dev
, MLX4_PROT_ETH
, port_num
);
138 if (mlx4_is_bonded(ibdev
->dev
)) {
139 struct net_device
*upper
= NULL
;
141 upper
= netdev_master_upper_dev_get_rcu(dev
);
143 struct net_device
*active
;
145 active
= bond_option_active_slave_get_rcu(netdev_priv(upper
));
158 static int mlx4_ib_update_gids_v1(struct gid_entry
*gids
,
159 struct mlx4_ib_dev
*ibdev
,
162 struct mlx4_cmd_mailbox
*mailbox
;
164 struct mlx4_dev
*dev
= ibdev
->dev
;
166 union ib_gid
*gid_tbl
;
168 mailbox
= mlx4_alloc_cmd_mailbox(dev
);
172 gid_tbl
= mailbox
->buf
;
174 for (i
= 0; i
< MLX4_MAX_PORT_GIDS
; ++i
)
175 memcpy(&gid_tbl
[i
], &gids
[i
].gid
, sizeof(union ib_gid
));
177 err
= mlx4_cmd(dev
, mailbox
->dma
,
178 MLX4_SET_PORT_GID_TABLE
<< 8 | port_num
,
179 1, MLX4_CMD_SET_PORT
, MLX4_CMD_TIME_CLASS_B
,
181 if (mlx4_is_bonded(dev
))
182 err
+= mlx4_cmd(dev
, mailbox
->dma
,
183 MLX4_SET_PORT_GID_TABLE
<< 8 | 2,
184 1, MLX4_CMD_SET_PORT
, MLX4_CMD_TIME_CLASS_B
,
187 mlx4_free_cmd_mailbox(dev
, mailbox
);
191 static int mlx4_ib_update_gids_v1_v2(struct gid_entry
*gids
,
192 struct mlx4_ib_dev
*ibdev
,
195 struct mlx4_cmd_mailbox
*mailbox
;
197 struct mlx4_dev
*dev
= ibdev
->dev
;
208 mailbox
= mlx4_alloc_cmd_mailbox(dev
);
212 gid_tbl
= mailbox
->buf
;
213 for (i
= 0; i
< MLX4_MAX_PORT_GIDS
; ++i
) {
214 memcpy(&gid_tbl
[i
].gid
, &gids
[i
].gid
, sizeof(union ib_gid
));
215 if (gids
[i
].gid_type
== IB_GID_TYPE_ROCE_UDP_ENCAP
) {
216 gid_tbl
[i
].version
= 2;
217 if (!ipv6_addr_v4mapped((struct in6_addr
*)&gids
[i
].gid
))
220 memset(&gid_tbl
[i
].gid
, 0, 12);
224 err
= mlx4_cmd(dev
, mailbox
->dma
,
225 MLX4_SET_PORT_ROCE_ADDR
<< 8 | port_num
,
226 1, MLX4_CMD_SET_PORT
, MLX4_CMD_TIME_CLASS_B
,
228 if (mlx4_is_bonded(dev
))
229 err
+= mlx4_cmd(dev
, mailbox
->dma
,
230 MLX4_SET_PORT_ROCE_ADDR
<< 8 | 2,
231 1, MLX4_CMD_SET_PORT
, MLX4_CMD_TIME_CLASS_B
,
234 mlx4_free_cmd_mailbox(dev
, mailbox
);
238 static int mlx4_ib_update_gids(struct gid_entry
*gids
,
239 struct mlx4_ib_dev
*ibdev
,
242 if (ibdev
->dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_ROCE_V1_V2
)
243 return mlx4_ib_update_gids_v1_v2(gids
, ibdev
, port_num
);
245 return mlx4_ib_update_gids_v1(gids
, ibdev
, port_num
);
248 static int mlx4_ib_add_gid(struct ib_device
*device
,
251 const union ib_gid
*gid
,
252 const struct ib_gid_attr
*attr
,
255 struct mlx4_ib_dev
*ibdev
= to_mdev(device
);
256 struct mlx4_ib_iboe
*iboe
= &ibdev
->iboe
;
257 struct mlx4_port_gid_table
*port_gid_table
;
258 int free
= -1, found
= -1;
262 struct gid_entry
*gids
= NULL
;
264 if (!rdma_cap_roce_gid_table(device
, port_num
))
267 if (port_num
> MLX4_MAX_PORTS
)
273 port_gid_table
= &iboe
->gids
[port_num
- 1];
274 spin_lock_bh(&iboe
->lock
);
275 for (i
= 0; i
< MLX4_MAX_PORT_GIDS
; ++i
) {
276 if (!memcmp(&port_gid_table
->gids
[i
].gid
, gid
, sizeof(*gid
)) &&
277 (port_gid_table
->gids
[i
].gid_type
== attr
->gid_type
)) {
281 if (free
< 0 && !memcmp(&port_gid_table
->gids
[i
].gid
, &zgid
, sizeof(*gid
)))
282 free
= i
; /* HW has space */
289 port_gid_table
->gids
[free
].ctx
= kmalloc(sizeof(*port_gid_table
->gids
[free
].ctx
), GFP_ATOMIC
);
290 if (!port_gid_table
->gids
[free
].ctx
) {
293 *context
= port_gid_table
->gids
[free
].ctx
;
294 memcpy(&port_gid_table
->gids
[free
].gid
, gid
, sizeof(*gid
));
295 port_gid_table
->gids
[free
].gid_type
= attr
->gid_type
;
296 port_gid_table
->gids
[free
].ctx
->real_index
= free
;
297 port_gid_table
->gids
[free
].ctx
->refcount
= 1;
302 struct gid_cache_context
*ctx
= port_gid_table
->gids
[found
].ctx
;
306 if (!ret
&& hw_update
) {
307 gids
= kmalloc(sizeof(*gids
) * MLX4_MAX_PORT_GIDS
, GFP_ATOMIC
);
311 for (i
= 0; i
< MLX4_MAX_PORT_GIDS
; i
++) {
312 memcpy(&gids
[i
].gid
, &port_gid_table
->gids
[i
].gid
, sizeof(union ib_gid
));
313 gids
[i
].gid_type
= port_gid_table
->gids
[i
].gid_type
;
317 spin_unlock_bh(&iboe
->lock
);
319 if (!ret
&& hw_update
) {
320 ret
= mlx4_ib_update_gids(gids
, ibdev
, port_num
);
327 static int mlx4_ib_del_gid(struct ib_device
*device
,
332 struct gid_cache_context
*ctx
= *context
;
333 struct mlx4_ib_dev
*ibdev
= to_mdev(device
);
334 struct mlx4_ib_iboe
*iboe
= &ibdev
->iboe
;
335 struct mlx4_port_gid_table
*port_gid_table
;
338 struct gid_entry
*gids
= NULL
;
340 if (!rdma_cap_roce_gid_table(device
, port_num
))
343 if (port_num
> MLX4_MAX_PORTS
)
346 port_gid_table
= &iboe
->gids
[port_num
- 1];
347 spin_lock_bh(&iboe
->lock
);
350 if (!ctx
->refcount
) {
351 unsigned int real_index
= ctx
->real_index
;
353 memcpy(&port_gid_table
->gids
[real_index
].gid
, &zgid
, sizeof(zgid
));
354 kfree(port_gid_table
->gids
[real_index
].ctx
);
355 port_gid_table
->gids
[real_index
].ctx
= NULL
;
359 if (!ret
&& hw_update
) {
362 gids
= kmalloc(sizeof(*gids
) * MLX4_MAX_PORT_GIDS
, GFP_ATOMIC
);
366 for (i
= 0; i
< MLX4_MAX_PORT_GIDS
; i
++)
367 memcpy(&gids
[i
].gid
, &port_gid_table
->gids
[i
].gid
, sizeof(union ib_gid
));
370 spin_unlock_bh(&iboe
->lock
);
372 if (!ret
&& hw_update
) {
373 ret
= mlx4_ib_update_gids(gids
, ibdev
, port_num
);
379 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev
*ibdev
,
380 u8 port_num
, int index
)
382 struct mlx4_ib_iboe
*iboe
= &ibdev
->iboe
;
383 struct gid_cache_context
*ctx
= NULL
;
385 struct mlx4_port_gid_table
*port_gid_table
;
386 int real_index
= -EINVAL
;
390 struct ib_gid_attr attr
;
392 if (port_num
> MLX4_MAX_PORTS
)
395 if (mlx4_is_bonded(ibdev
->dev
))
398 if (!rdma_cap_roce_gid_table(&ibdev
->ib_dev
, port_num
))
401 ret
= ib_get_cached_gid(&ibdev
->ib_dev
, port_num
, index
, &gid
, &attr
);
408 if (!memcmp(&gid
, &zgid
, sizeof(gid
)))
411 spin_lock_irqsave(&iboe
->lock
, flags
);
412 port_gid_table
= &iboe
->gids
[port_num
- 1];
414 for (i
= 0; i
< MLX4_MAX_PORT_GIDS
; ++i
)
415 if (!memcmp(&port_gid_table
->gids
[i
].gid
, &gid
, sizeof(gid
)) &&
416 attr
.gid_type
== port_gid_table
->gids
[i
].gid_type
) {
417 ctx
= port_gid_table
->gids
[i
].ctx
;
421 real_index
= ctx
->real_index
;
422 spin_unlock_irqrestore(&iboe
->lock
, flags
);
426 static int mlx4_ib_query_device(struct ib_device
*ibdev
,
427 struct ib_device_attr
*props
,
428 struct ib_udata
*uhw
)
430 struct mlx4_ib_dev
*dev
= to_mdev(ibdev
);
431 struct ib_smp
*in_mad
= NULL
;
432 struct ib_smp
*out_mad
= NULL
;
435 struct mlx4_uverbs_ex_query_device cmd
;
436 struct mlx4_uverbs_ex_query_device_resp resp
= {.comp_mask
= 0};
437 struct mlx4_clock_params clock_params
;
440 if (uhw
->inlen
< sizeof(cmd
))
443 err
= ib_copy_from_udata(&cmd
, uhw
, sizeof(cmd
));
454 resp
.response_length
= offsetof(typeof(resp
), response_length
) +
455 sizeof(resp
.response_length
);
456 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
457 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
458 if (!in_mad
|| !out_mad
)
461 init_query_mad(in_mad
);
462 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
464 err
= mlx4_MAD_IFC(to_mdev(ibdev
), MLX4_MAD_IFC_IGNORE_KEYS
,
465 1, NULL
, NULL
, in_mad
, out_mad
);
469 memset(props
, 0, sizeof *props
);
471 have_ib_ports
= num_ib_ports(dev
->dev
);
473 props
->fw_ver
= dev
->dev
->caps
.fw_ver
;
474 props
->device_cap_flags
= IB_DEVICE_CHANGE_PHY_PORT
|
475 IB_DEVICE_PORT_ACTIVE_EVENT
|
476 IB_DEVICE_SYS_IMAGE_GUID
|
477 IB_DEVICE_RC_RNR_NAK_GEN
|
478 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK
;
479 if (dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR
)
480 props
->device_cap_flags
|= IB_DEVICE_BAD_PKEY_CNTR
;
481 if (dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR
)
482 props
->device_cap_flags
|= IB_DEVICE_BAD_QKEY_CNTR
;
483 if (dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_APM
&& have_ib_ports
)
484 props
->device_cap_flags
|= IB_DEVICE_AUTO_PATH_MIG
;
485 if (dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_UD_AV_PORT
)
486 props
->device_cap_flags
|= IB_DEVICE_UD_AV_PORT_ENFORCE
;
487 if (dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_IPOIB_CSUM
)
488 props
->device_cap_flags
|= IB_DEVICE_UD_IP_CSUM
;
489 if (dev
->dev
->caps
.max_gso_sz
&&
490 (dev
->dev
->rev_id
!= MLX4_IB_CARD_REV_A0
) &&
491 (dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_BLH
))
492 props
->device_cap_flags
|= IB_DEVICE_UD_TSO
;
493 if (dev
->dev
->caps
.bmme_flags
& MLX4_BMME_FLAG_RESERVED_LKEY
)
494 props
->device_cap_flags
|= IB_DEVICE_LOCAL_DMA_LKEY
;
495 if ((dev
->dev
->caps
.bmme_flags
& MLX4_BMME_FLAG_LOCAL_INV
) &&
496 (dev
->dev
->caps
.bmme_flags
& MLX4_BMME_FLAG_REMOTE_INV
) &&
497 (dev
->dev
->caps
.bmme_flags
& MLX4_BMME_FLAG_FAST_REG_WR
))
498 props
->device_cap_flags
|= IB_DEVICE_MEM_MGT_EXTENSIONS
;
499 if (dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_XRC
)
500 props
->device_cap_flags
|= IB_DEVICE_XRC
;
501 if (dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_MEM_WINDOW
)
502 props
->device_cap_flags
|= IB_DEVICE_MEM_WINDOW
;
503 if (dev
->dev
->caps
.bmme_flags
& MLX4_BMME_FLAG_TYPE_2_WIN
) {
504 if (dev
->dev
->caps
.bmme_flags
& MLX4_BMME_FLAG_WIN_TYPE_2B
)
505 props
->device_cap_flags
|= IB_DEVICE_MEM_WINDOW_TYPE_2B
;
507 props
->device_cap_flags
|= IB_DEVICE_MEM_WINDOW_TYPE_2A
;
509 if (dev
->steering_support
== MLX4_STEERING_MODE_DEVICE_MANAGED
)
510 props
->device_cap_flags
|= IB_DEVICE_MANAGED_FLOW_STEERING
;
512 props
->device_cap_flags
|= IB_DEVICE_RAW_IP_CSUM
;
514 props
->vendor_id
= be32_to_cpup((__be32
*) (out_mad
->data
+ 36)) &
516 props
->vendor_part_id
= dev
->dev
->persist
->pdev
->device
;
517 props
->hw_ver
= be32_to_cpup((__be32
*) (out_mad
->data
+ 32));
518 memcpy(&props
->sys_image_guid
, out_mad
->data
+ 4, 8);
520 props
->max_mr_size
= ~0ull;
521 props
->page_size_cap
= dev
->dev
->caps
.page_size_cap
;
522 props
->max_qp
= dev
->dev
->quotas
.qp
;
523 props
->max_qp_wr
= dev
->dev
->caps
.max_wqes
- MLX4_IB_SQ_MAX_SPARE
;
524 props
->max_sge
= min(dev
->dev
->caps
.max_sq_sg
,
525 dev
->dev
->caps
.max_rq_sg
);
526 props
->max_sge_rd
= MLX4_MAX_SGE_RD
;
527 props
->max_cq
= dev
->dev
->quotas
.cq
;
528 props
->max_cqe
= dev
->dev
->caps
.max_cqes
;
529 props
->max_mr
= dev
->dev
->quotas
.mpt
;
530 props
->max_pd
= dev
->dev
->caps
.num_pds
- dev
->dev
->caps
.reserved_pds
;
531 props
->max_qp_rd_atom
= dev
->dev
->caps
.max_qp_dest_rdma
;
532 props
->max_qp_init_rd_atom
= dev
->dev
->caps
.max_qp_init_rdma
;
533 props
->max_res_rd_atom
= props
->max_qp_rd_atom
* props
->max_qp
;
534 props
->max_srq
= dev
->dev
->quotas
.srq
;
535 props
->max_srq_wr
= dev
->dev
->caps
.max_srq_wqes
- 1;
536 props
->max_srq_sge
= dev
->dev
->caps
.max_srq_sge
;
537 props
->max_fast_reg_page_list_len
= MLX4_MAX_FAST_REG_PAGES
;
538 props
->local_ca_ack_delay
= dev
->dev
->caps
.local_ca_ack_delay
;
539 props
->atomic_cap
= dev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_ATOMIC
?
540 IB_ATOMIC_HCA
: IB_ATOMIC_NONE
;
541 props
->masked_atomic_cap
= props
->atomic_cap
;
542 props
->max_pkeys
= dev
->dev
->caps
.pkey_table_len
[1];
543 props
->max_mcast_grp
= dev
->dev
->caps
.num_mgms
+ dev
->dev
->caps
.num_amgms
;
544 props
->max_mcast_qp_attach
= dev
->dev
->caps
.num_qp_per_mgm
;
545 props
->max_total_mcast_qp_attach
= props
->max_mcast_qp_attach
*
546 props
->max_mcast_grp
;
547 props
->max_map_per_fmr
= dev
->dev
->caps
.max_fmr_maps
;
548 props
->hca_core_clock
= dev
->dev
->caps
.hca_core_clock
* 1000UL;
549 props
->timestamp_mask
= 0xFFFFFFFFFFFFULL
;
551 if (!mlx4_is_slave(dev
->dev
))
552 err
= mlx4_get_internal_clock_params(dev
->dev
, &clock_params
);
554 if (uhw
->outlen
>= resp
.response_length
+ sizeof(resp
.hca_core_clock_offset
)) {
555 resp
.response_length
+= sizeof(resp
.hca_core_clock_offset
);
556 if (!err
&& !mlx4_is_slave(dev
->dev
)) {
557 resp
.comp_mask
|= QUERY_DEVICE_RESP_MASK_TIMESTAMP
;
558 resp
.hca_core_clock_offset
= clock_params
.offset
% PAGE_SIZE
;
563 err
= ib_copy_to_udata(uhw
, &resp
, resp
.response_length
);
574 static enum rdma_link_layer
575 mlx4_ib_port_link_layer(struct ib_device
*device
, u8 port_num
)
577 struct mlx4_dev
*dev
= to_mdev(device
)->dev
;
579 return dev
->caps
.port_mask
[port_num
] == MLX4_PORT_TYPE_IB
?
580 IB_LINK_LAYER_INFINIBAND
: IB_LINK_LAYER_ETHERNET
;
583 static int ib_link_query_port(struct ib_device
*ibdev
, u8 port
,
584 struct ib_port_attr
*props
, int netw_view
)
586 struct ib_smp
*in_mad
= NULL
;
587 struct ib_smp
*out_mad
= NULL
;
588 int ext_active_speed
;
589 int mad_ifc_flags
= MLX4_MAD_IFC_IGNORE_KEYS
;
592 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
593 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
594 if (!in_mad
|| !out_mad
)
597 init_query_mad(in_mad
);
598 in_mad
->attr_id
= IB_SMP_ATTR_PORT_INFO
;
599 in_mad
->attr_mod
= cpu_to_be32(port
);
601 if (mlx4_is_mfunc(to_mdev(ibdev
)->dev
) && netw_view
)
602 mad_ifc_flags
|= MLX4_MAD_IFC_NET_VIEW
;
604 err
= mlx4_MAD_IFC(to_mdev(ibdev
), mad_ifc_flags
, port
, NULL
, NULL
,
610 props
->lid
= be16_to_cpup((__be16
*) (out_mad
->data
+ 16));
611 props
->lmc
= out_mad
->data
[34] & 0x7;
612 props
->sm_lid
= be16_to_cpup((__be16
*) (out_mad
->data
+ 18));
613 props
->sm_sl
= out_mad
->data
[36] & 0xf;
614 props
->state
= out_mad
->data
[32] & 0xf;
615 props
->phys_state
= out_mad
->data
[33] >> 4;
616 props
->port_cap_flags
= be32_to_cpup((__be32
*) (out_mad
->data
+ 20));
618 props
->gid_tbl_len
= out_mad
->data
[50];
620 props
->gid_tbl_len
= to_mdev(ibdev
)->dev
->caps
.gid_table_len
[port
];
621 props
->max_msg_sz
= to_mdev(ibdev
)->dev
->caps
.max_msg_sz
;
622 props
->pkey_tbl_len
= to_mdev(ibdev
)->dev
->caps
.pkey_table_len
[port
];
623 props
->bad_pkey_cntr
= be16_to_cpup((__be16
*) (out_mad
->data
+ 46));
624 props
->qkey_viol_cntr
= be16_to_cpup((__be16
*) (out_mad
->data
+ 48));
625 props
->active_width
= out_mad
->data
[31] & 0xf;
626 props
->active_speed
= out_mad
->data
[35] >> 4;
627 props
->max_mtu
= out_mad
->data
[41] & 0xf;
628 props
->active_mtu
= out_mad
->data
[36] >> 4;
629 props
->subnet_timeout
= out_mad
->data
[51] & 0x1f;
630 props
->max_vl_num
= out_mad
->data
[37] >> 4;
631 props
->init_type_reply
= out_mad
->data
[41] >> 4;
633 /* Check if extended speeds (EDR/FDR/...) are supported */
634 if (props
->port_cap_flags
& IB_PORT_EXTENDED_SPEEDS_SUP
) {
635 ext_active_speed
= out_mad
->data
[62] >> 4;
637 switch (ext_active_speed
) {
639 props
->active_speed
= IB_SPEED_FDR
;
642 props
->active_speed
= IB_SPEED_EDR
;
647 /* If reported active speed is QDR, check if is FDR-10 */
648 if (props
->active_speed
== IB_SPEED_QDR
) {
649 init_query_mad(in_mad
);
650 in_mad
->attr_id
= MLX4_ATTR_EXTENDED_PORT_INFO
;
651 in_mad
->attr_mod
= cpu_to_be32(port
);
653 err
= mlx4_MAD_IFC(to_mdev(ibdev
), mad_ifc_flags
, port
,
654 NULL
, NULL
, in_mad
, out_mad
);
658 /* Checking LinkSpeedActive for FDR-10 */
659 if (out_mad
->data
[15] & 0x1)
660 props
->active_speed
= IB_SPEED_FDR10
;
663 /* Avoid wrong speed value returned by FW if the IB link is down. */
664 if (props
->state
== IB_PORT_DOWN
)
665 props
->active_speed
= IB_SPEED_SDR
;
673 static u8
state_to_phys_state(enum ib_port_state state
)
675 return state
== IB_PORT_ACTIVE
? 5 : 3;
678 static int eth_link_query_port(struct ib_device
*ibdev
, u8 port
,
679 struct ib_port_attr
*props
, int netw_view
)
682 struct mlx4_ib_dev
*mdev
= to_mdev(ibdev
);
683 struct mlx4_ib_iboe
*iboe
= &mdev
->iboe
;
684 struct net_device
*ndev
;
686 struct mlx4_cmd_mailbox
*mailbox
;
688 int is_bonded
= mlx4_is_bonded(mdev
->dev
);
690 mailbox
= mlx4_alloc_cmd_mailbox(mdev
->dev
);
692 return PTR_ERR(mailbox
);
694 err
= mlx4_cmd_box(mdev
->dev
, 0, mailbox
->dma
, port
, 0,
695 MLX4_CMD_QUERY_PORT
, MLX4_CMD_TIME_CLASS_B
,
700 props
->active_width
= (((u8
*)mailbox
->buf
)[5] == 0x40) ?
701 IB_WIDTH_4X
: IB_WIDTH_1X
;
702 props
->active_speed
= IB_SPEED_QDR
;
703 props
->port_cap_flags
= IB_PORT_CM_SUP
| IB_PORT_IP_BASED_GIDS
;
704 props
->gid_tbl_len
= mdev
->dev
->caps
.gid_table_len
[port
];
705 props
->max_msg_sz
= mdev
->dev
->caps
.max_msg_sz
;
706 props
->pkey_tbl_len
= 1;
707 props
->max_mtu
= IB_MTU_4096
;
708 props
->max_vl_num
= 2;
709 props
->state
= IB_PORT_DOWN
;
710 props
->phys_state
= state_to_phys_state(props
->state
);
711 props
->active_mtu
= IB_MTU_256
;
712 spin_lock_bh(&iboe
->lock
);
713 ndev
= iboe
->netdevs
[port
- 1];
714 if (ndev
&& is_bonded
) {
715 rcu_read_lock(); /* required to get upper dev */
716 ndev
= netdev_master_upper_dev_get_rcu(ndev
);
722 tmp
= iboe_get_mtu(ndev
->mtu
);
723 props
->active_mtu
= tmp
? min(props
->max_mtu
, tmp
) : IB_MTU_256
;
725 props
->state
= (netif_running(ndev
) && netif_carrier_ok(ndev
)) ?
726 IB_PORT_ACTIVE
: IB_PORT_DOWN
;
727 props
->phys_state
= state_to_phys_state(props
->state
);
729 spin_unlock_bh(&iboe
->lock
);
731 mlx4_free_cmd_mailbox(mdev
->dev
, mailbox
);
735 int __mlx4_ib_query_port(struct ib_device
*ibdev
, u8 port
,
736 struct ib_port_attr
*props
, int netw_view
)
740 memset(props
, 0, sizeof *props
);
742 err
= mlx4_ib_port_link_layer(ibdev
, port
) == IB_LINK_LAYER_INFINIBAND
?
743 ib_link_query_port(ibdev
, port
, props
, netw_view
) :
744 eth_link_query_port(ibdev
, port
, props
, netw_view
);
749 static int mlx4_ib_query_port(struct ib_device
*ibdev
, u8 port
,
750 struct ib_port_attr
*props
)
752 /* returns host view */
753 return __mlx4_ib_query_port(ibdev
, port
, props
, 0);
756 int __mlx4_ib_query_gid(struct ib_device
*ibdev
, u8 port
, int index
,
757 union ib_gid
*gid
, int netw_view
)
759 struct ib_smp
*in_mad
= NULL
;
760 struct ib_smp
*out_mad
= NULL
;
762 struct mlx4_ib_dev
*dev
= to_mdev(ibdev
);
764 int mad_ifc_flags
= MLX4_MAD_IFC_IGNORE_KEYS
;
766 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
767 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
768 if (!in_mad
|| !out_mad
)
771 init_query_mad(in_mad
);
772 in_mad
->attr_id
= IB_SMP_ATTR_PORT_INFO
;
773 in_mad
->attr_mod
= cpu_to_be32(port
);
775 if (mlx4_is_mfunc(dev
->dev
) && netw_view
)
776 mad_ifc_flags
|= MLX4_MAD_IFC_NET_VIEW
;
778 err
= mlx4_MAD_IFC(dev
, mad_ifc_flags
, port
, NULL
, NULL
, in_mad
, out_mad
);
782 memcpy(gid
->raw
, out_mad
->data
+ 8, 8);
784 if (mlx4_is_mfunc(dev
->dev
) && !netw_view
) {
786 /* For any index > 0, return the null guid */
793 init_query_mad(in_mad
);
794 in_mad
->attr_id
= IB_SMP_ATTR_GUID_INFO
;
795 in_mad
->attr_mod
= cpu_to_be32(index
/ 8);
797 err
= mlx4_MAD_IFC(dev
, mad_ifc_flags
, port
,
798 NULL
, NULL
, in_mad
, out_mad
);
802 memcpy(gid
->raw
+ 8, out_mad
->data
+ (index
% 8) * 8, 8);
806 memset(gid
->raw
+ 8, 0, 8);
812 static int mlx4_ib_query_gid(struct ib_device
*ibdev
, u8 port
, int index
,
817 if (rdma_protocol_ib(ibdev
, port
))
818 return __mlx4_ib_query_gid(ibdev
, port
, index
, gid
, 0);
820 if (!rdma_protocol_roce(ibdev
, port
))
823 if (!rdma_cap_roce_gid_table(ibdev
, port
))
826 ret
= ib_get_cached_gid(ibdev
, port
, index
, gid
, NULL
);
827 if (ret
== -EAGAIN
) {
828 memcpy(gid
, &zgid
, sizeof(*gid
));
835 int __mlx4_ib_query_pkey(struct ib_device
*ibdev
, u8 port
, u16 index
,
836 u16
*pkey
, int netw_view
)
838 struct ib_smp
*in_mad
= NULL
;
839 struct ib_smp
*out_mad
= NULL
;
840 int mad_ifc_flags
= MLX4_MAD_IFC_IGNORE_KEYS
;
843 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
844 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
845 if (!in_mad
|| !out_mad
)
848 init_query_mad(in_mad
);
849 in_mad
->attr_id
= IB_SMP_ATTR_PKEY_TABLE
;
850 in_mad
->attr_mod
= cpu_to_be32(index
/ 32);
852 if (mlx4_is_mfunc(to_mdev(ibdev
)->dev
) && netw_view
)
853 mad_ifc_flags
|= MLX4_MAD_IFC_NET_VIEW
;
855 err
= mlx4_MAD_IFC(to_mdev(ibdev
), mad_ifc_flags
, port
, NULL
, NULL
,
860 *pkey
= be16_to_cpu(((__be16
*) out_mad
->data
)[index
% 32]);
868 static int mlx4_ib_query_pkey(struct ib_device
*ibdev
, u8 port
, u16 index
, u16
*pkey
)
870 return __mlx4_ib_query_pkey(ibdev
, port
, index
, pkey
, 0);
873 static int mlx4_ib_modify_device(struct ib_device
*ibdev
, int mask
,
874 struct ib_device_modify
*props
)
876 struct mlx4_cmd_mailbox
*mailbox
;
879 if (mask
& ~IB_DEVICE_MODIFY_NODE_DESC
)
882 if (!(mask
& IB_DEVICE_MODIFY_NODE_DESC
))
885 if (mlx4_is_slave(to_mdev(ibdev
)->dev
))
888 spin_lock_irqsave(&to_mdev(ibdev
)->sm_lock
, flags
);
889 memcpy(ibdev
->node_desc
, props
->node_desc
, 64);
890 spin_unlock_irqrestore(&to_mdev(ibdev
)->sm_lock
, flags
);
893 * If possible, pass node desc to FW, so it can generate
894 * a 144 trap. If cmd fails, just ignore.
896 mailbox
= mlx4_alloc_cmd_mailbox(to_mdev(ibdev
)->dev
);
900 memcpy(mailbox
->buf
, props
->node_desc
, 64);
901 mlx4_cmd(to_mdev(ibdev
)->dev
, mailbox
->dma
, 1, 0,
902 MLX4_CMD_SET_NODE
, MLX4_CMD_TIME_CLASS_A
, MLX4_CMD_NATIVE
);
904 mlx4_free_cmd_mailbox(to_mdev(ibdev
)->dev
, mailbox
);
909 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev
*dev
, u8 port
, int reset_qkey_viols
,
912 struct mlx4_cmd_mailbox
*mailbox
;
915 mailbox
= mlx4_alloc_cmd_mailbox(dev
->dev
);
917 return PTR_ERR(mailbox
);
919 if (dev
->dev
->flags
& MLX4_FLAG_OLD_PORT_CMDS
) {
920 *(u8
*) mailbox
->buf
= !!reset_qkey_viols
<< 6;
921 ((__be32
*) mailbox
->buf
)[2] = cpu_to_be32(cap_mask
);
923 ((u8
*) mailbox
->buf
)[3] = !!reset_qkey_viols
;
924 ((__be32
*) mailbox
->buf
)[1] = cpu_to_be32(cap_mask
);
927 err
= mlx4_cmd(dev
->dev
, mailbox
->dma
, port
, MLX4_SET_PORT_IB_OPCODE
,
928 MLX4_CMD_SET_PORT
, MLX4_CMD_TIME_CLASS_B
,
931 mlx4_free_cmd_mailbox(dev
->dev
, mailbox
);
935 static int mlx4_ib_modify_port(struct ib_device
*ibdev
, u8 port
, int mask
,
936 struct ib_port_modify
*props
)
938 struct mlx4_ib_dev
*mdev
= to_mdev(ibdev
);
939 u8 is_eth
= mdev
->dev
->caps
.port_type
[port
] == MLX4_PORT_TYPE_ETH
;
940 struct ib_port_attr attr
;
944 /* return OK if this is RoCE. CM calls ib_modify_port() regardless
945 * of whether port link layer is ETH or IB. For ETH ports, qkey
946 * violations and port capabilities are not meaningful.
951 mutex_lock(&mdev
->cap_mask_mutex
);
953 err
= mlx4_ib_query_port(ibdev
, port
, &attr
);
957 cap_mask
= (attr
.port_cap_flags
| props
->set_port_cap_mask
) &
958 ~props
->clr_port_cap_mask
;
960 err
= mlx4_ib_SET_PORT(mdev
, port
,
961 !!(mask
& IB_PORT_RESET_QKEY_CNTR
),
965 mutex_unlock(&to_mdev(ibdev
)->cap_mask_mutex
);
969 static struct ib_ucontext
*mlx4_ib_alloc_ucontext(struct ib_device
*ibdev
,
970 struct ib_udata
*udata
)
972 struct mlx4_ib_dev
*dev
= to_mdev(ibdev
);
973 struct mlx4_ib_ucontext
*context
;
974 struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3
;
975 struct mlx4_ib_alloc_ucontext_resp resp
;
979 return ERR_PTR(-EAGAIN
);
981 if (ibdev
->uverbs_abi_ver
== MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION
) {
982 resp_v3
.qp_tab_size
= dev
->dev
->caps
.num_qps
;
983 resp_v3
.bf_reg_size
= dev
->dev
->caps
.bf_reg_size
;
984 resp_v3
.bf_regs_per_page
= dev
->dev
->caps
.bf_regs_per_page
;
986 resp
.dev_caps
= dev
->dev
->caps
.userspace_caps
;
987 resp
.qp_tab_size
= dev
->dev
->caps
.num_qps
;
988 resp
.bf_reg_size
= dev
->dev
->caps
.bf_reg_size
;
989 resp
.bf_regs_per_page
= dev
->dev
->caps
.bf_regs_per_page
;
990 resp
.cqe_size
= dev
->dev
->caps
.cqe_size
;
993 context
= kzalloc(sizeof(*context
), GFP_KERNEL
);
995 return ERR_PTR(-ENOMEM
);
997 err
= mlx4_uar_alloc(to_mdev(ibdev
)->dev
, &context
->uar
);
1000 return ERR_PTR(err
);
1003 INIT_LIST_HEAD(&context
->db_page_list
);
1004 mutex_init(&context
->db_page_mutex
);
1006 if (ibdev
->uverbs_abi_ver
== MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION
)
1007 err
= ib_copy_to_udata(udata
, &resp_v3
, sizeof(resp_v3
));
1009 err
= ib_copy_to_udata(udata
, &resp
, sizeof(resp
));
1012 mlx4_uar_free(to_mdev(ibdev
)->dev
, &context
->uar
);
1014 return ERR_PTR(-EFAULT
);
1017 return &context
->ibucontext
;
1020 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext
*ibcontext
)
1022 struct mlx4_ib_ucontext
*context
= to_mucontext(ibcontext
);
1024 mlx4_uar_free(to_mdev(ibcontext
->device
)->dev
, &context
->uar
);
1030 static void mlx4_ib_vma_open(struct vm_area_struct
*area
)
1032 /* vma_open is called when a new VMA is created on top of our VMA.
1033 * This is done through either mremap flow or split_vma (usually due
1034 * to mlock, madvise, munmap, etc.). We do not support a clone of the
1035 * vma, as this VMA is strongly hardware related. Therefore we set the
1036 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1037 * calling us again and trying to do incorrect actions. We assume that
1038 * the original vma size is exactly a single page that there will be no
1039 * "splitting" operations on.
1041 area
->vm_ops
= NULL
;
1044 static void mlx4_ib_vma_close(struct vm_area_struct
*area
)
1046 struct mlx4_ib_vma_private_data
*mlx4_ib_vma_priv_data
;
1048 /* It's guaranteed that all VMAs opened on a FD are closed before the
1049 * file itself is closed, therefore no sync is needed with the regular
1050 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1051 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1052 * The close operation is usually called under mm->mmap_sem except when
1053 * process is exiting. The exiting case is handled explicitly as part
1054 * of mlx4_ib_disassociate_ucontext.
1056 mlx4_ib_vma_priv_data
= (struct mlx4_ib_vma_private_data
*)
1057 area
->vm_private_data
;
1059 /* set the vma context pointer to null in the mlx4_ib driver's private
1060 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1062 mlx4_ib_vma_priv_data
->vma
= NULL
;
1065 static const struct vm_operations_struct mlx4_ib_vm_ops
= {
1066 .open
= mlx4_ib_vma_open
,
1067 .close
= mlx4_ib_vma_close
1070 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext
*ibcontext
)
1074 struct vm_area_struct
*vma
;
1075 struct mlx4_ib_ucontext
*context
= to_mucontext(ibcontext
);
1076 struct task_struct
*owning_process
= NULL
;
1077 struct mm_struct
*owning_mm
= NULL
;
1079 owning_process
= get_pid_task(ibcontext
->tgid
, PIDTYPE_PID
);
1080 if (!owning_process
)
1083 owning_mm
= get_task_mm(owning_process
);
1085 pr_info("no mm, disassociate ucontext is pending task termination\n");
1087 /* make sure that task is dead before returning, it may
1088 * prevent a rare case of module down in parallel to a
1089 * call to mlx4_ib_vma_close.
1091 put_task_struct(owning_process
);
1093 owning_process
= get_pid_task(ibcontext
->tgid
,
1095 if (!owning_process
||
1096 owning_process
->state
== TASK_DEAD
) {
1097 pr_info("disassociate ucontext done, task was terminated\n");
1098 /* in case task was dead need to release the task struct */
1100 put_task_struct(owning_process
);
1106 /* need to protect from a race on closing the vma as part of
1107 * mlx4_ib_vma_close().
1109 down_read(&owning_mm
->mmap_sem
);
1110 for (i
= 0; i
< HW_BAR_COUNT
; i
++) {
1111 vma
= context
->hw_bar_info
[i
].vma
;
1115 ret
= zap_vma_ptes(context
->hw_bar_info
[i
].vma
,
1116 context
->hw_bar_info
[i
].vma
->vm_start
,
1119 pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i
, ret
);
1123 /* context going to be destroyed, should not access ops any more */
1124 context
->hw_bar_info
[i
].vma
->vm_ops
= NULL
;
1127 up_read(&owning_mm
->mmap_sem
);
1129 put_task_struct(owning_process
);
1132 static void mlx4_ib_set_vma_data(struct vm_area_struct
*vma
,
1133 struct mlx4_ib_vma_private_data
*vma_private_data
)
1135 vma_private_data
->vma
= vma
;
1136 vma
->vm_private_data
= vma_private_data
;
1137 vma
->vm_ops
= &mlx4_ib_vm_ops
;
1140 static int mlx4_ib_mmap(struct ib_ucontext
*context
, struct vm_area_struct
*vma
)
1142 struct mlx4_ib_dev
*dev
= to_mdev(context
->device
);
1143 struct mlx4_ib_ucontext
*mucontext
= to_mucontext(context
);
1145 if (vma
->vm_end
- vma
->vm_start
!= PAGE_SIZE
)
1148 if (vma
->vm_pgoff
== 0) {
1149 /* We prevent double mmaping on same context */
1150 if (mucontext
->hw_bar_info
[HW_BAR_DB
].vma
)
1153 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
1155 if (io_remap_pfn_range(vma
, vma
->vm_start
,
1156 to_mucontext(context
)->uar
.pfn
,
1157 PAGE_SIZE
, vma
->vm_page_prot
))
1160 mlx4_ib_set_vma_data(vma
, &mucontext
->hw_bar_info
[HW_BAR_DB
]);
1162 } else if (vma
->vm_pgoff
== 1 && dev
->dev
->caps
.bf_reg_size
!= 0) {
1163 /* We prevent double mmaping on same context */
1164 if (mucontext
->hw_bar_info
[HW_BAR_BF
].vma
)
1167 vma
->vm_page_prot
= pgprot_writecombine(vma
->vm_page_prot
);
1169 if (io_remap_pfn_range(vma
, vma
->vm_start
,
1170 to_mucontext(context
)->uar
.pfn
+
1171 dev
->dev
->caps
.num_uars
,
1172 PAGE_SIZE
, vma
->vm_page_prot
))
1175 mlx4_ib_set_vma_data(vma
, &mucontext
->hw_bar_info
[HW_BAR_BF
]);
1177 } else if (vma
->vm_pgoff
== 3) {
1178 struct mlx4_clock_params params
;
1181 /* We prevent double mmaping on same context */
1182 if (mucontext
->hw_bar_info
[HW_BAR_CLOCK
].vma
)
1185 ret
= mlx4_get_internal_clock_params(dev
->dev
, ¶ms
);
1190 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
1191 if (io_remap_pfn_range(vma
, vma
->vm_start
,
1192 (pci_resource_start(dev
->dev
->persist
->pdev
,
1196 PAGE_SIZE
, vma
->vm_page_prot
))
1199 mlx4_ib_set_vma_data(vma
,
1200 &mucontext
->hw_bar_info
[HW_BAR_CLOCK
]);
1208 static struct ib_pd
*mlx4_ib_alloc_pd(struct ib_device
*ibdev
,
1209 struct ib_ucontext
*context
,
1210 struct ib_udata
*udata
)
1212 struct mlx4_ib_pd
*pd
;
1215 pd
= kmalloc(sizeof *pd
, GFP_KERNEL
);
1217 return ERR_PTR(-ENOMEM
);
1219 err
= mlx4_pd_alloc(to_mdev(ibdev
)->dev
, &pd
->pdn
);
1222 return ERR_PTR(err
);
1226 if (ib_copy_to_udata(udata
, &pd
->pdn
, sizeof (__u32
))) {
1227 mlx4_pd_free(to_mdev(ibdev
)->dev
, pd
->pdn
);
1229 return ERR_PTR(-EFAULT
);
1235 static int mlx4_ib_dealloc_pd(struct ib_pd
*pd
)
1237 mlx4_pd_free(to_mdev(pd
->device
)->dev
, to_mpd(pd
)->pdn
);
1243 static struct ib_xrcd
*mlx4_ib_alloc_xrcd(struct ib_device
*ibdev
,
1244 struct ib_ucontext
*context
,
1245 struct ib_udata
*udata
)
1247 struct mlx4_ib_xrcd
*xrcd
;
1248 struct ib_cq_init_attr cq_attr
= {};
1251 if (!(to_mdev(ibdev
)->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_XRC
))
1252 return ERR_PTR(-ENOSYS
);
1254 xrcd
= kmalloc(sizeof *xrcd
, GFP_KERNEL
);
1256 return ERR_PTR(-ENOMEM
);
1258 err
= mlx4_xrcd_alloc(to_mdev(ibdev
)->dev
, &xrcd
->xrcdn
);
1262 xrcd
->pd
= ib_alloc_pd(ibdev
);
1263 if (IS_ERR(xrcd
->pd
)) {
1264 err
= PTR_ERR(xrcd
->pd
);
1269 xrcd
->cq
= ib_create_cq(ibdev
, NULL
, NULL
, xrcd
, &cq_attr
);
1270 if (IS_ERR(xrcd
->cq
)) {
1271 err
= PTR_ERR(xrcd
->cq
);
1275 return &xrcd
->ibxrcd
;
1278 ib_dealloc_pd(xrcd
->pd
);
1280 mlx4_xrcd_free(to_mdev(ibdev
)->dev
, xrcd
->xrcdn
);
1283 return ERR_PTR(err
);
1286 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd
*xrcd
)
1288 ib_destroy_cq(to_mxrcd(xrcd
)->cq
);
1289 ib_dealloc_pd(to_mxrcd(xrcd
)->pd
);
1290 mlx4_xrcd_free(to_mdev(xrcd
->device
)->dev
, to_mxrcd(xrcd
)->xrcdn
);
1296 static int add_gid_entry(struct ib_qp
*ibqp
, union ib_gid
*gid
)
1298 struct mlx4_ib_qp
*mqp
= to_mqp(ibqp
);
1299 struct mlx4_ib_dev
*mdev
= to_mdev(ibqp
->device
);
1300 struct mlx4_ib_gid_entry
*ge
;
1302 ge
= kzalloc(sizeof *ge
, GFP_KERNEL
);
1307 if (mlx4_ib_add_mc(mdev
, mqp
, gid
)) {
1308 ge
->port
= mqp
->port
;
1312 mutex_lock(&mqp
->mutex
);
1313 list_add_tail(&ge
->list
, &mqp
->gid_list
);
1314 mutex_unlock(&mqp
->mutex
);
1319 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev
*ibdev
,
1320 struct mlx4_ib_counters
*ctr_table
)
1322 struct counter_index
*counter
, *tmp_count
;
1324 mutex_lock(&ctr_table
->mutex
);
1325 list_for_each_entry_safe(counter
, tmp_count
, &ctr_table
->counters_list
,
1327 if (counter
->allocated
)
1328 mlx4_counter_free(ibdev
->dev
, counter
->index
);
1329 list_del(&counter
->list
);
1332 mutex_unlock(&ctr_table
->mutex
);
1335 int mlx4_ib_add_mc(struct mlx4_ib_dev
*mdev
, struct mlx4_ib_qp
*mqp
,
1338 struct net_device
*ndev
;
1344 spin_lock_bh(&mdev
->iboe
.lock
);
1345 ndev
= mdev
->iboe
.netdevs
[mqp
->port
- 1];
1348 spin_unlock_bh(&mdev
->iboe
.lock
);
1358 struct mlx4_ib_steering
{
1359 struct list_head list
;
1360 struct mlx4_flow_reg_id reg_id
;
1364 static int parse_flow_attr(struct mlx4_dev
*dev
,
1366 union ib_flow_spec
*ib_spec
,
1367 struct _rule_hw
*mlx4_spec
)
1369 enum mlx4_net_trans_rule_id type
;
1371 switch (ib_spec
->type
) {
1372 case IB_FLOW_SPEC_ETH
:
1373 type
= MLX4_NET_TRANS_RULE_ID_ETH
;
1374 memcpy(mlx4_spec
->eth
.dst_mac
, ib_spec
->eth
.val
.dst_mac
,
1376 memcpy(mlx4_spec
->eth
.dst_mac_msk
, ib_spec
->eth
.mask
.dst_mac
,
1378 mlx4_spec
->eth
.vlan_tag
= ib_spec
->eth
.val
.vlan_tag
;
1379 mlx4_spec
->eth
.vlan_tag_msk
= ib_spec
->eth
.mask
.vlan_tag
;
1381 case IB_FLOW_SPEC_IB
:
1382 type
= MLX4_NET_TRANS_RULE_ID_IB
;
1383 mlx4_spec
->ib
.l3_qpn
=
1384 cpu_to_be32(qp_num
);
1385 mlx4_spec
->ib
.qpn_mask
=
1386 cpu_to_be32(MLX4_IB_FLOW_QPN_MASK
);
1390 case IB_FLOW_SPEC_IPV4
:
1391 type
= MLX4_NET_TRANS_RULE_ID_IPV4
;
1392 mlx4_spec
->ipv4
.src_ip
= ib_spec
->ipv4
.val
.src_ip
;
1393 mlx4_spec
->ipv4
.src_ip_msk
= ib_spec
->ipv4
.mask
.src_ip
;
1394 mlx4_spec
->ipv4
.dst_ip
= ib_spec
->ipv4
.val
.dst_ip
;
1395 mlx4_spec
->ipv4
.dst_ip_msk
= ib_spec
->ipv4
.mask
.dst_ip
;
1398 case IB_FLOW_SPEC_TCP
:
1399 case IB_FLOW_SPEC_UDP
:
1400 type
= ib_spec
->type
== IB_FLOW_SPEC_TCP
?
1401 MLX4_NET_TRANS_RULE_ID_TCP
:
1402 MLX4_NET_TRANS_RULE_ID_UDP
;
1403 mlx4_spec
->tcp_udp
.dst_port
= ib_spec
->tcp_udp
.val
.dst_port
;
1404 mlx4_spec
->tcp_udp
.dst_port_msk
= ib_spec
->tcp_udp
.mask
.dst_port
;
1405 mlx4_spec
->tcp_udp
.src_port
= ib_spec
->tcp_udp
.val
.src_port
;
1406 mlx4_spec
->tcp_udp
.src_port_msk
= ib_spec
->tcp_udp
.mask
.src_port
;
1412 if (mlx4_map_sw_to_hw_steering_id(dev
, type
) < 0 ||
1413 mlx4_hw_rule_sz(dev
, type
) < 0)
1415 mlx4_spec
->id
= cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev
, type
));
1416 mlx4_spec
->size
= mlx4_hw_rule_sz(dev
, type
) >> 2;
1417 return mlx4_hw_rule_sz(dev
, type
);
1420 struct default_rules
{
1421 __u32 mandatory_fields
[IB_FLOW_SPEC_SUPPORT_LAYERS
];
1422 __u32 mandatory_not_fields
[IB_FLOW_SPEC_SUPPORT_LAYERS
];
1423 __u32 rules_create_list
[IB_FLOW_SPEC_SUPPORT_LAYERS
];
1426 static const struct default_rules default_table
[] = {
1428 .mandatory_fields
= {IB_FLOW_SPEC_IPV4
},
1429 .mandatory_not_fields
= {IB_FLOW_SPEC_ETH
},
1430 .rules_create_list
= {IB_FLOW_SPEC_IB
},
1431 .link_layer
= IB_LINK_LAYER_INFINIBAND
1435 static int __mlx4_ib_default_rules_match(struct ib_qp
*qp
,
1436 struct ib_flow_attr
*flow_attr
)
1440 const struct default_rules
*pdefault_rules
= default_table
;
1441 u8 link_layer
= rdma_port_get_link_layer(qp
->device
, flow_attr
->port
);
1443 for (i
= 0; i
< ARRAY_SIZE(default_table
); i
++, pdefault_rules
++) {
1444 __u32 field_types
[IB_FLOW_SPEC_SUPPORT_LAYERS
];
1445 memset(&field_types
, 0, sizeof(field_types
));
1447 if (link_layer
!= pdefault_rules
->link_layer
)
1450 ib_flow
= flow_attr
+ 1;
1451 /* we assume the specs are sorted */
1452 for (j
= 0, k
= 0; k
< IB_FLOW_SPEC_SUPPORT_LAYERS
&&
1453 j
< flow_attr
->num_of_specs
; k
++) {
1454 union ib_flow_spec
*current_flow
=
1455 (union ib_flow_spec
*)ib_flow
;
1457 /* same layer but different type */
1458 if (((current_flow
->type
& IB_FLOW_SPEC_LAYER_MASK
) ==
1459 (pdefault_rules
->mandatory_fields
[k
] &
1460 IB_FLOW_SPEC_LAYER_MASK
)) &&
1461 (current_flow
->type
!=
1462 pdefault_rules
->mandatory_fields
[k
]))
1465 /* same layer, try match next one */
1466 if (current_flow
->type
==
1467 pdefault_rules
->mandatory_fields
[k
]) {
1470 ((union ib_flow_spec
*)ib_flow
)->size
;
1474 ib_flow
= flow_attr
+ 1;
1475 for (j
= 0; j
< flow_attr
->num_of_specs
;
1476 j
++, ib_flow
+= ((union ib_flow_spec
*)ib_flow
)->size
)
1477 for (k
= 0; k
< IB_FLOW_SPEC_SUPPORT_LAYERS
; k
++)
1478 /* same layer and same type */
1479 if (((union ib_flow_spec
*)ib_flow
)->type
==
1480 pdefault_rules
->mandatory_not_fields
[k
])
1489 static int __mlx4_ib_create_default_rules(
1490 struct mlx4_ib_dev
*mdev
,
1492 const struct default_rules
*pdefault_rules
,
1493 struct _rule_hw
*mlx4_spec
) {
1497 for (i
= 0; i
< ARRAY_SIZE(pdefault_rules
->rules_create_list
); i
++) {
1499 union ib_flow_spec ib_spec
;
1500 switch (pdefault_rules
->rules_create_list
[i
]) {
1504 case IB_FLOW_SPEC_IB
:
1505 ib_spec
.type
= IB_FLOW_SPEC_IB
;
1506 ib_spec
.size
= sizeof(struct ib_flow_spec_ib
);
1513 /* We must put empty rule, qpn is being ignored */
1514 ret
= parse_flow_attr(mdev
->dev
, 0, &ib_spec
,
1517 pr_info("invalid parsing\n");
1521 mlx4_spec
= (void *)mlx4_spec
+ ret
;
1527 static int __mlx4_ib_create_flow(struct ib_qp
*qp
, struct ib_flow_attr
*flow_attr
,
1529 enum mlx4_net_trans_promisc_mode flow_type
,
1535 struct mlx4_ib_dev
*mdev
= to_mdev(qp
->device
);
1536 struct mlx4_cmd_mailbox
*mailbox
;
1537 struct mlx4_net_trans_rule_hw_ctrl
*ctrl
;
1540 static const u16 __mlx4_domain
[] = {
1541 [IB_FLOW_DOMAIN_USER
] = MLX4_DOMAIN_UVERBS
,
1542 [IB_FLOW_DOMAIN_ETHTOOL
] = MLX4_DOMAIN_ETHTOOL
,
1543 [IB_FLOW_DOMAIN_RFS
] = MLX4_DOMAIN_RFS
,
1544 [IB_FLOW_DOMAIN_NIC
] = MLX4_DOMAIN_NIC
,
1547 if (flow_attr
->priority
> MLX4_IB_FLOW_MAX_PRIO
) {
1548 pr_err("Invalid priority value %d\n", flow_attr
->priority
);
1552 if (domain
>= IB_FLOW_DOMAIN_NUM
) {
1553 pr_err("Invalid domain value %d\n", domain
);
1557 if (mlx4_map_sw_to_hw_steering_mode(mdev
->dev
, flow_type
) < 0)
1560 mailbox
= mlx4_alloc_cmd_mailbox(mdev
->dev
);
1561 if (IS_ERR(mailbox
))
1562 return PTR_ERR(mailbox
);
1563 ctrl
= mailbox
->buf
;
1565 ctrl
->prio
= cpu_to_be16(__mlx4_domain
[domain
] |
1566 flow_attr
->priority
);
1567 ctrl
->type
= mlx4_map_sw_to_hw_steering_mode(mdev
->dev
, flow_type
);
1568 ctrl
->port
= flow_attr
->port
;
1569 ctrl
->qpn
= cpu_to_be32(qp
->qp_num
);
1571 ib_flow
= flow_attr
+ 1;
1572 size
+= sizeof(struct mlx4_net_trans_rule_hw_ctrl
);
1573 /* Add default flows */
1574 default_flow
= __mlx4_ib_default_rules_match(qp
, flow_attr
);
1575 if (default_flow
>= 0) {
1576 ret
= __mlx4_ib_create_default_rules(
1577 mdev
, qp
, default_table
+ default_flow
,
1578 mailbox
->buf
+ size
);
1580 mlx4_free_cmd_mailbox(mdev
->dev
, mailbox
);
1585 for (i
= 0; i
< flow_attr
->num_of_specs
; i
++) {
1586 ret
= parse_flow_attr(mdev
->dev
, qp
->qp_num
, ib_flow
,
1587 mailbox
->buf
+ size
);
1589 mlx4_free_cmd_mailbox(mdev
->dev
, mailbox
);
1592 ib_flow
+= ((union ib_flow_spec
*) ib_flow
)->size
;
1596 ret
= mlx4_cmd_imm(mdev
->dev
, mailbox
->dma
, reg_id
, size
>> 2, 0,
1597 MLX4_QP_FLOW_STEERING_ATTACH
, MLX4_CMD_TIME_CLASS_A
,
1600 pr_err("mcg table is full. Fail to register network rule.\n");
1601 else if (ret
== -ENXIO
)
1602 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1604 pr_err("Invalid argument. Fail to register network rule.\n");
1606 mlx4_free_cmd_mailbox(mdev
->dev
, mailbox
);
1610 static int __mlx4_ib_destroy_flow(struct mlx4_dev
*dev
, u64 reg_id
)
1613 err
= mlx4_cmd(dev
, reg_id
, 0, 0,
1614 MLX4_QP_FLOW_STEERING_DETACH
, MLX4_CMD_TIME_CLASS_A
,
1617 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1622 static int mlx4_ib_tunnel_steer_add(struct ib_qp
*qp
, struct ib_flow_attr
*flow_attr
,
1626 union ib_flow_spec
*ib_spec
;
1627 struct mlx4_dev
*dev
= to_mdev(qp
->device
)->dev
;
1630 if (dev
->caps
.tunnel_offload_mode
!= MLX4_TUNNEL_OFFLOAD_MODE_VXLAN
||
1631 dev
->caps
.dmfs_high_steer_mode
== MLX4_STEERING_DMFS_A0_STATIC
)
1632 return 0; /* do nothing */
1634 ib_flow
= flow_attr
+ 1;
1635 ib_spec
= (union ib_flow_spec
*)ib_flow
;
1637 if (ib_spec
->type
!= IB_FLOW_SPEC_ETH
|| flow_attr
->num_of_specs
!= 1)
1638 return 0; /* do nothing */
1640 err
= mlx4_tunnel_steer_add(to_mdev(qp
->device
)->dev
, ib_spec
->eth
.val
.dst_mac
,
1641 flow_attr
->port
, qp
->qp_num
,
1642 MLX4_DOMAIN_UVERBS
| (flow_attr
->priority
& 0xff),
1647 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev
*dev
,
1648 struct ib_flow_attr
*flow_attr
,
1649 enum mlx4_net_trans_promisc_mode
*type
)
1653 if (!(dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER
) ||
1654 (dev
->caps
.dmfs_high_steer_mode
== MLX4_STEERING_DMFS_A0_STATIC
) ||
1655 (flow_attr
->num_of_specs
> 1) || (flow_attr
->priority
!= 0)) {
1659 if (flow_attr
->num_of_specs
== 0) {
1660 type
[0] = MLX4_FS_MC_SNIFFER
;
1661 type
[1] = MLX4_FS_UC_SNIFFER
;
1663 union ib_flow_spec
*ib_spec
;
1665 ib_spec
= (union ib_flow_spec
*)(flow_attr
+ 1);
1666 if (ib_spec
->type
!= IB_FLOW_SPEC_ETH
)
1669 /* if all is zero than MC and UC */
1670 if (is_zero_ether_addr(ib_spec
->eth
.mask
.dst_mac
)) {
1671 type
[0] = MLX4_FS_MC_SNIFFER
;
1672 type
[1] = MLX4_FS_UC_SNIFFER
;
1674 u8 mac
[ETH_ALEN
] = {ib_spec
->eth
.mask
.dst_mac
[0] ^ 0x01,
1675 ib_spec
->eth
.mask
.dst_mac
[1],
1676 ib_spec
->eth
.mask
.dst_mac
[2],
1677 ib_spec
->eth
.mask
.dst_mac
[3],
1678 ib_spec
->eth
.mask
.dst_mac
[4],
1679 ib_spec
->eth
.mask
.dst_mac
[5]};
1681 /* Above xor was only on MC bit, non empty mask is valid
1682 * only if this bit is set and rest are zero.
1684 if (!is_zero_ether_addr(&mac
[0]))
1687 if (is_multicast_ether_addr(ib_spec
->eth
.val
.dst_mac
))
1688 type
[0] = MLX4_FS_MC_SNIFFER
;
1690 type
[0] = MLX4_FS_UC_SNIFFER
;
1697 static struct ib_flow
*mlx4_ib_create_flow(struct ib_qp
*qp
,
1698 struct ib_flow_attr
*flow_attr
,
1701 int err
= 0, i
= 0, j
= 0;
1702 struct mlx4_ib_flow
*mflow
;
1703 enum mlx4_net_trans_promisc_mode type
[2];
1704 struct mlx4_dev
*dev
= (to_mdev(qp
->device
))->dev
;
1705 int is_bonded
= mlx4_is_bonded(dev
);
1707 if (flow_attr
->port
< 1 || flow_attr
->port
> qp
->device
->phys_port_cnt
)
1708 return ERR_PTR(-EINVAL
);
1710 if ((flow_attr
->flags
& IB_FLOW_ATTR_FLAGS_DONT_TRAP
) &&
1711 (flow_attr
->type
!= IB_FLOW_ATTR_NORMAL
))
1712 return ERR_PTR(-EOPNOTSUPP
);
1714 memset(type
, 0, sizeof(type
));
1716 mflow
= kzalloc(sizeof(*mflow
), GFP_KERNEL
);
1722 switch (flow_attr
->type
) {
1723 case IB_FLOW_ATTR_NORMAL
:
1724 /* If dont trap flag (continue match) is set, under specific
1725 * condition traffic be replicated to given qp,
1726 * without stealing it
1728 if (unlikely(flow_attr
->flags
& IB_FLOW_ATTR_FLAGS_DONT_TRAP
)) {
1729 err
= mlx4_ib_add_dont_trap_rule(dev
,
1735 type
[0] = MLX4_FS_REGULAR
;
1739 case IB_FLOW_ATTR_ALL_DEFAULT
:
1740 type
[0] = MLX4_FS_ALL_DEFAULT
;
1743 case IB_FLOW_ATTR_MC_DEFAULT
:
1744 type
[0] = MLX4_FS_MC_DEFAULT
;
1747 case IB_FLOW_ATTR_SNIFFER
:
1748 type
[0] = MLX4_FS_MIRROR_RX_PORT
;
1749 type
[1] = MLX4_FS_MIRROR_SX_PORT
;
1757 while (i
< ARRAY_SIZE(type
) && type
[i
]) {
1758 err
= __mlx4_ib_create_flow(qp
, flow_attr
, domain
, type
[i
],
1759 &mflow
->reg_id
[i
].id
);
1761 goto err_create_flow
;
1763 /* Application always sees one port so the mirror rule
1764 * must be on port #2
1766 flow_attr
->port
= 2;
1767 err
= __mlx4_ib_create_flow(qp
, flow_attr
,
1769 &mflow
->reg_id
[j
].mirror
);
1770 flow_attr
->port
= 1;
1772 goto err_create_flow
;
1779 if (i
< ARRAY_SIZE(type
) && flow_attr
->type
== IB_FLOW_ATTR_NORMAL
) {
1780 err
= mlx4_ib_tunnel_steer_add(qp
, flow_attr
,
1781 &mflow
->reg_id
[i
].id
);
1783 goto err_create_flow
;
1786 flow_attr
->port
= 2;
1787 err
= mlx4_ib_tunnel_steer_add(qp
, flow_attr
,
1788 &mflow
->reg_id
[j
].mirror
);
1789 flow_attr
->port
= 1;
1791 goto err_create_flow
;
1794 /* function to create mirror rule */
1798 return &mflow
->ibflow
;
1802 (void)__mlx4_ib_destroy_flow(to_mdev(qp
->device
)->dev
,
1803 mflow
->reg_id
[i
].id
);
1808 (void)__mlx4_ib_destroy_flow(to_mdev(qp
->device
)->dev
,
1809 mflow
->reg_id
[j
].mirror
);
1814 return ERR_PTR(err
);
1817 static int mlx4_ib_destroy_flow(struct ib_flow
*flow_id
)
1821 struct mlx4_ib_dev
*mdev
= to_mdev(flow_id
->qp
->device
);
1822 struct mlx4_ib_flow
*mflow
= to_mflow(flow_id
);
1824 while (i
< ARRAY_SIZE(mflow
->reg_id
) && mflow
->reg_id
[i
].id
) {
1825 err
= __mlx4_ib_destroy_flow(mdev
->dev
, mflow
->reg_id
[i
].id
);
1828 if (mflow
->reg_id
[i
].mirror
) {
1829 err
= __mlx4_ib_destroy_flow(mdev
->dev
,
1830 mflow
->reg_id
[i
].mirror
);
1841 static int mlx4_ib_mcg_attach(struct ib_qp
*ibqp
, union ib_gid
*gid
, u16 lid
)
1844 struct mlx4_ib_dev
*mdev
= to_mdev(ibqp
->device
);
1845 struct mlx4_dev
*dev
= mdev
->dev
;
1846 struct mlx4_ib_qp
*mqp
= to_mqp(ibqp
);
1847 struct mlx4_ib_steering
*ib_steering
= NULL
;
1848 enum mlx4_protocol prot
= MLX4_PROT_IB_IPV6
;
1849 struct mlx4_flow_reg_id reg_id
;
1851 if (mdev
->dev
->caps
.steering_mode
==
1852 MLX4_STEERING_MODE_DEVICE_MANAGED
) {
1853 ib_steering
= kmalloc(sizeof(*ib_steering
), GFP_KERNEL
);
1858 err
= mlx4_multicast_attach(mdev
->dev
, &mqp
->mqp
, gid
->raw
, mqp
->port
,
1860 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK
),
1863 pr_err("multicast attach op failed, err %d\n", err
);
1868 if (mlx4_is_bonded(dev
)) {
1869 err
= mlx4_multicast_attach(mdev
->dev
, &mqp
->mqp
, gid
->raw
,
1870 (mqp
->port
== 1) ? 2 : 1,
1872 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK
),
1873 prot
, ®_id
.mirror
);
1878 err
= add_gid_entry(ibqp
, gid
);
1883 memcpy(ib_steering
->gid
.raw
, gid
->raw
, 16);
1884 ib_steering
->reg_id
= reg_id
;
1885 mutex_lock(&mqp
->mutex
);
1886 list_add(&ib_steering
->list
, &mqp
->steering_rules
);
1887 mutex_unlock(&mqp
->mutex
);
1892 mlx4_multicast_detach(mdev
->dev
, &mqp
->mqp
, gid
->raw
,
1895 mlx4_multicast_detach(mdev
->dev
, &mqp
->mqp
, gid
->raw
,
1896 prot
, reg_id
.mirror
);
1903 static struct mlx4_ib_gid_entry
*find_gid_entry(struct mlx4_ib_qp
*qp
, u8
*raw
)
1905 struct mlx4_ib_gid_entry
*ge
;
1906 struct mlx4_ib_gid_entry
*tmp
;
1907 struct mlx4_ib_gid_entry
*ret
= NULL
;
1909 list_for_each_entry_safe(ge
, tmp
, &qp
->gid_list
, list
) {
1910 if (!memcmp(raw
, ge
->gid
.raw
, 16)) {
1919 static int mlx4_ib_mcg_detach(struct ib_qp
*ibqp
, union ib_gid
*gid
, u16 lid
)
1922 struct mlx4_ib_dev
*mdev
= to_mdev(ibqp
->device
);
1923 struct mlx4_dev
*dev
= mdev
->dev
;
1924 struct mlx4_ib_qp
*mqp
= to_mqp(ibqp
);
1925 struct net_device
*ndev
;
1926 struct mlx4_ib_gid_entry
*ge
;
1927 struct mlx4_flow_reg_id reg_id
= {0, 0};
1928 enum mlx4_protocol prot
= MLX4_PROT_IB_IPV6
;
1930 if (mdev
->dev
->caps
.steering_mode
==
1931 MLX4_STEERING_MODE_DEVICE_MANAGED
) {
1932 struct mlx4_ib_steering
*ib_steering
;
1934 mutex_lock(&mqp
->mutex
);
1935 list_for_each_entry(ib_steering
, &mqp
->steering_rules
, list
) {
1936 if (!memcmp(ib_steering
->gid
.raw
, gid
->raw
, 16)) {
1937 list_del(&ib_steering
->list
);
1941 mutex_unlock(&mqp
->mutex
);
1942 if (&ib_steering
->list
== &mqp
->steering_rules
) {
1943 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1946 reg_id
= ib_steering
->reg_id
;
1950 err
= mlx4_multicast_detach(mdev
->dev
, &mqp
->mqp
, gid
->raw
,
1955 if (mlx4_is_bonded(dev
)) {
1956 err
= mlx4_multicast_detach(mdev
->dev
, &mqp
->mqp
, gid
->raw
,
1957 prot
, reg_id
.mirror
);
1962 mutex_lock(&mqp
->mutex
);
1963 ge
= find_gid_entry(mqp
, gid
->raw
);
1965 spin_lock_bh(&mdev
->iboe
.lock
);
1966 ndev
= ge
->added
? mdev
->iboe
.netdevs
[ge
->port
- 1] : NULL
;
1969 spin_unlock_bh(&mdev
->iboe
.lock
);
1972 list_del(&ge
->list
);
1975 pr_warn("could not find mgid entry\n");
1977 mutex_unlock(&mqp
->mutex
);
1982 static int init_node_data(struct mlx4_ib_dev
*dev
)
1984 struct ib_smp
*in_mad
= NULL
;
1985 struct ib_smp
*out_mad
= NULL
;
1986 int mad_ifc_flags
= MLX4_MAD_IFC_IGNORE_KEYS
;
1989 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
1990 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
1991 if (!in_mad
|| !out_mad
)
1994 init_query_mad(in_mad
);
1995 in_mad
->attr_id
= IB_SMP_ATTR_NODE_DESC
;
1996 if (mlx4_is_master(dev
->dev
))
1997 mad_ifc_flags
|= MLX4_MAD_IFC_NET_VIEW
;
1999 err
= mlx4_MAD_IFC(dev
, mad_ifc_flags
, 1, NULL
, NULL
, in_mad
, out_mad
);
2003 memcpy(dev
->ib_dev
.node_desc
, out_mad
->data
, 64);
2005 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
2007 err
= mlx4_MAD_IFC(dev
, mad_ifc_flags
, 1, NULL
, NULL
, in_mad
, out_mad
);
2011 dev
->dev
->rev_id
= be32_to_cpup((__be32
*) (out_mad
->data
+ 32));
2012 memcpy(&dev
->ib_dev
.node_guid
, out_mad
->data
+ 12, 8);
2020 static ssize_t
show_hca(struct device
*device
, struct device_attribute
*attr
,
2023 struct mlx4_ib_dev
*dev
=
2024 container_of(device
, struct mlx4_ib_dev
, ib_dev
.dev
);
2025 return sprintf(buf
, "MT%d\n", dev
->dev
->persist
->pdev
->device
);
2028 static ssize_t
show_fw_ver(struct device
*device
, struct device_attribute
*attr
,
2031 struct mlx4_ib_dev
*dev
=
2032 container_of(device
, struct mlx4_ib_dev
, ib_dev
.dev
);
2033 return sprintf(buf
, "%d.%d.%d\n", (int) (dev
->dev
->caps
.fw_ver
>> 32),
2034 (int) (dev
->dev
->caps
.fw_ver
>> 16) & 0xffff,
2035 (int) dev
->dev
->caps
.fw_ver
& 0xffff);
2038 static ssize_t
show_rev(struct device
*device
, struct device_attribute
*attr
,
2041 struct mlx4_ib_dev
*dev
=
2042 container_of(device
, struct mlx4_ib_dev
, ib_dev
.dev
);
2043 return sprintf(buf
, "%x\n", dev
->dev
->rev_id
);
2046 static ssize_t
show_board(struct device
*device
, struct device_attribute
*attr
,
2049 struct mlx4_ib_dev
*dev
=
2050 container_of(device
, struct mlx4_ib_dev
, ib_dev
.dev
);
2051 return sprintf(buf
, "%.*s\n", MLX4_BOARD_ID_LEN
,
2052 dev
->dev
->board_id
);
2055 static DEVICE_ATTR(hw_rev
, S_IRUGO
, show_rev
, NULL
);
2056 static DEVICE_ATTR(fw_ver
, S_IRUGO
, show_fw_ver
, NULL
);
2057 static DEVICE_ATTR(hca_type
, S_IRUGO
, show_hca
, NULL
);
2058 static DEVICE_ATTR(board_id
, S_IRUGO
, show_board
, NULL
);
2060 static struct device_attribute
*mlx4_class_attributes
[] = {
2067 #define MLX4_IB_INVALID_MAC ((u64)-1)
2068 static void mlx4_ib_update_qps(struct mlx4_ib_dev
*ibdev
,
2069 struct net_device
*dev
,
2073 u64 release_mac
= MLX4_IB_INVALID_MAC
;
2074 struct mlx4_ib_qp
*qp
;
2076 read_lock(&dev_base_lock
);
2077 new_smac
= mlx4_mac_to_u64(dev
->dev_addr
);
2078 read_unlock(&dev_base_lock
);
2080 atomic64_set(&ibdev
->iboe
.mac
[port
- 1], new_smac
);
2082 /* no need for update QP1 and mac registration in non-SRIOV */
2083 if (!mlx4_is_mfunc(ibdev
->dev
))
2086 mutex_lock(&ibdev
->qp1_proxy_lock
[port
- 1]);
2087 qp
= ibdev
->qp1_proxy
[port
- 1];
2091 struct mlx4_update_qp_params update_params
;
2093 mutex_lock(&qp
->mutex
);
2094 old_smac
= qp
->pri
.smac
;
2095 if (new_smac
== old_smac
)
2098 new_smac_index
= mlx4_register_mac(ibdev
->dev
, port
, new_smac
);
2100 if (new_smac_index
< 0)
2103 update_params
.smac_index
= new_smac_index
;
2104 if (mlx4_update_qp(ibdev
->dev
, qp
->mqp
.qpn
, MLX4_UPDATE_QP_SMAC
,
2106 release_mac
= new_smac
;
2109 /* if old port was zero, no mac was yet registered for this QP */
2110 if (qp
->pri
.smac_port
)
2111 release_mac
= old_smac
;
2112 qp
->pri
.smac
= new_smac
;
2113 qp
->pri
.smac_port
= port
;
2114 qp
->pri
.smac_index
= new_smac_index
;
2118 if (release_mac
!= MLX4_IB_INVALID_MAC
)
2119 mlx4_unregister_mac(ibdev
->dev
, port
, release_mac
);
2121 mutex_unlock(&qp
->mutex
);
2122 mutex_unlock(&ibdev
->qp1_proxy_lock
[port
- 1]);
2125 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev
*ibdev
,
2126 struct net_device
*dev
,
2127 unsigned long event
)
2130 struct mlx4_ib_iboe
*iboe
;
2131 int update_qps_port
= -1;
2136 iboe
= &ibdev
->iboe
;
2138 spin_lock_bh(&iboe
->lock
);
2139 mlx4_foreach_ib_transport_port(port
, ibdev
->dev
) {
2141 iboe
->netdevs
[port
- 1] =
2142 mlx4_get_protocol_dev(ibdev
->dev
, MLX4_PROT_ETH
, port
);
2144 if (dev
== iboe
->netdevs
[port
- 1] &&
2145 (event
== NETDEV_CHANGEADDR
|| event
== NETDEV_REGISTER
||
2146 event
== NETDEV_UP
|| event
== NETDEV_CHANGE
))
2147 update_qps_port
= port
;
2150 spin_unlock_bh(&iboe
->lock
);
2152 if (update_qps_port
> 0)
2153 mlx4_ib_update_qps(ibdev
, dev
, update_qps_port
);
2156 static int mlx4_ib_netdev_event(struct notifier_block
*this,
2157 unsigned long event
, void *ptr
)
2159 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
2160 struct mlx4_ib_dev
*ibdev
;
2162 if (!net_eq(dev_net(dev
), &init_net
))
2165 ibdev
= container_of(this, struct mlx4_ib_dev
, iboe
.nb
);
2166 mlx4_ib_scan_netdevs(ibdev
, dev
, event
);
2171 static void init_pkeys(struct mlx4_ib_dev
*ibdev
)
2177 if (mlx4_is_master(ibdev
->dev
)) {
2178 for (slave
= 0; slave
<= ibdev
->dev
->persist
->num_vfs
;
2180 for (port
= 1; port
<= ibdev
->dev
->caps
.num_ports
; ++port
) {
2182 i
< ibdev
->dev
->phys_caps
.pkey_phys_table_len
[port
];
2184 ibdev
->pkeys
.virt2phys_pkey
[slave
][port
- 1][i
] =
2185 /* master has the identity virt2phys pkey mapping */
2186 (slave
== mlx4_master_func_num(ibdev
->dev
) || !i
) ? i
:
2187 ibdev
->dev
->phys_caps
.pkey_phys_table_len
[port
] - 1;
2188 mlx4_sync_pkey_table(ibdev
->dev
, slave
, port
, i
,
2189 ibdev
->pkeys
.virt2phys_pkey
[slave
][port
- 1][i
]);
2193 /* initialize pkey cache */
2194 for (port
= 1; port
<= ibdev
->dev
->caps
.num_ports
; ++port
) {
2196 i
< ibdev
->dev
->phys_caps
.pkey_phys_table_len
[port
];
2198 ibdev
->pkeys
.phys_pkey_cache
[port
-1][i
] =
2204 static void mlx4_ib_alloc_eqs(struct mlx4_dev
*dev
, struct mlx4_ib_dev
*ibdev
)
2206 int i
, j
, eq
= 0, total_eqs
= 0;
2208 ibdev
->eq_table
= kcalloc(dev
->caps
.num_comp_vectors
,
2209 sizeof(ibdev
->eq_table
[0]), GFP_KERNEL
);
2210 if (!ibdev
->eq_table
)
2213 for (i
= 1; i
<= dev
->caps
.num_ports
; i
++) {
2214 for (j
= 0; j
< mlx4_get_eqs_per_port(dev
, i
);
2216 if (i
> 1 && mlx4_is_eq_shared(dev
, total_eqs
))
2218 ibdev
->eq_table
[eq
] = total_eqs
;
2219 if (!mlx4_assign_eq(dev
, i
,
2220 &ibdev
->eq_table
[eq
]))
2223 ibdev
->eq_table
[eq
] = -1;
2227 for (i
= eq
; i
< dev
->caps
.num_comp_vectors
;
2228 ibdev
->eq_table
[i
++] = -1)
2231 /* Advertise the new number of EQs to clients */
2232 ibdev
->ib_dev
.num_comp_vectors
= eq
;
2235 static void mlx4_ib_free_eqs(struct mlx4_dev
*dev
, struct mlx4_ib_dev
*ibdev
)
2238 int total_eqs
= ibdev
->ib_dev
.num_comp_vectors
;
2240 /* no eqs were allocated */
2241 if (!ibdev
->eq_table
)
2244 /* Reset the advertised EQ number */
2245 ibdev
->ib_dev
.num_comp_vectors
= 0;
2247 for (i
= 0; i
< total_eqs
; i
++)
2248 mlx4_release_eq(dev
, ibdev
->eq_table
[i
]);
2250 kfree(ibdev
->eq_table
);
2251 ibdev
->eq_table
= NULL
;
2254 static int mlx4_port_immutable(struct ib_device
*ibdev
, u8 port_num
,
2255 struct ib_port_immutable
*immutable
)
2257 struct ib_port_attr attr
;
2258 struct mlx4_ib_dev
*mdev
= to_mdev(ibdev
);
2261 err
= mlx4_ib_query_port(ibdev
, port_num
, &attr
);
2265 immutable
->pkey_tbl_len
= attr
.pkey_tbl_len
;
2266 immutable
->gid_tbl_len
= attr
.gid_tbl_len
;
2268 if (mlx4_ib_port_link_layer(ibdev
, port_num
) == IB_LINK_LAYER_INFINIBAND
) {
2269 immutable
->core_cap_flags
= RDMA_CORE_PORT_IBA_IB
;
2271 if (mdev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_IBOE
)
2272 immutable
->core_cap_flags
= RDMA_CORE_PORT_IBA_ROCE
;
2273 if (mdev
->dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_ROCE_V1_V2
)
2274 immutable
->core_cap_flags
= RDMA_CORE_PORT_IBA_ROCE
|
2275 RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP
;
2278 immutable
->max_mad_size
= IB_MGMT_MAD_SIZE
;
2283 static void *mlx4_ib_add(struct mlx4_dev
*dev
)
2285 struct mlx4_ib_dev
*ibdev
;
2289 struct mlx4_ib_iboe
*iboe
;
2290 int ib_num_ports
= 0;
2291 int num_req_counters
;
2294 struct counter_index
*new_counter_index
= NULL
;
2296 pr_info_once("%s", mlx4_ib_version
);
2299 mlx4_foreach_ib_transport_port(i
, dev
)
2302 /* No point in registering a device with no ports... */
2306 ibdev
= (struct mlx4_ib_dev
*) ib_alloc_device(sizeof *ibdev
);
2308 dev_err(&dev
->persist
->pdev
->dev
,
2309 "Device struct alloc failed\n");
2313 iboe
= &ibdev
->iboe
;
2315 if (mlx4_pd_alloc(dev
, &ibdev
->priv_pdn
))
2318 if (mlx4_uar_alloc(dev
, &ibdev
->priv_uar
))
2321 ibdev
->uar_map
= ioremap((phys_addr_t
) ibdev
->priv_uar
.pfn
<< PAGE_SHIFT
,
2323 if (!ibdev
->uar_map
)
2325 MLX4_INIT_DOORBELL_LOCK(&ibdev
->uar_lock
);
2328 ibdev
->bond_next_port
= 0;
2330 strlcpy(ibdev
->ib_dev
.name
, "mlx4_%d", IB_DEVICE_NAME_MAX
);
2331 ibdev
->ib_dev
.owner
= THIS_MODULE
;
2332 ibdev
->ib_dev
.node_type
= RDMA_NODE_IB_CA
;
2333 ibdev
->ib_dev
.local_dma_lkey
= dev
->caps
.reserved_lkey
;
2334 ibdev
->num_ports
= num_ports
;
2335 ibdev
->ib_dev
.phys_port_cnt
= mlx4_is_bonded(dev
) ?
2336 1 : ibdev
->num_ports
;
2337 ibdev
->ib_dev
.num_comp_vectors
= dev
->caps
.num_comp_vectors
;
2338 ibdev
->ib_dev
.dma_device
= &dev
->persist
->pdev
->dev
;
2339 ibdev
->ib_dev
.get_netdev
= mlx4_ib_get_netdev
;
2340 ibdev
->ib_dev
.add_gid
= mlx4_ib_add_gid
;
2341 ibdev
->ib_dev
.del_gid
= mlx4_ib_del_gid
;
2343 if (dev
->caps
.userspace_caps
)
2344 ibdev
->ib_dev
.uverbs_abi_ver
= MLX4_IB_UVERBS_ABI_VERSION
;
2346 ibdev
->ib_dev
.uverbs_abi_ver
= MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION
;
2348 ibdev
->ib_dev
.uverbs_cmd_mask
=
2349 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT
) |
2350 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE
) |
2351 (1ull << IB_USER_VERBS_CMD_QUERY_PORT
) |
2352 (1ull << IB_USER_VERBS_CMD_ALLOC_PD
) |
2353 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD
) |
2354 (1ull << IB_USER_VERBS_CMD_REG_MR
) |
2355 (1ull << IB_USER_VERBS_CMD_REREG_MR
) |
2356 (1ull << IB_USER_VERBS_CMD_DEREG_MR
) |
2357 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL
) |
2358 (1ull << IB_USER_VERBS_CMD_CREATE_CQ
) |
2359 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ
) |
2360 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ
) |
2361 (1ull << IB_USER_VERBS_CMD_CREATE_QP
) |
2362 (1ull << IB_USER_VERBS_CMD_MODIFY_QP
) |
2363 (1ull << IB_USER_VERBS_CMD_QUERY_QP
) |
2364 (1ull << IB_USER_VERBS_CMD_DESTROY_QP
) |
2365 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST
) |
2366 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST
) |
2367 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ
) |
2368 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ
) |
2369 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ
) |
2370 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ
) |
2371 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ
) |
2372 (1ull << IB_USER_VERBS_CMD_OPEN_QP
);
2374 ibdev
->ib_dev
.query_device
= mlx4_ib_query_device
;
2375 ibdev
->ib_dev
.query_port
= mlx4_ib_query_port
;
2376 ibdev
->ib_dev
.get_link_layer
= mlx4_ib_port_link_layer
;
2377 ibdev
->ib_dev
.query_gid
= mlx4_ib_query_gid
;
2378 ibdev
->ib_dev
.query_pkey
= mlx4_ib_query_pkey
;
2379 ibdev
->ib_dev
.modify_device
= mlx4_ib_modify_device
;
2380 ibdev
->ib_dev
.modify_port
= mlx4_ib_modify_port
;
2381 ibdev
->ib_dev
.alloc_ucontext
= mlx4_ib_alloc_ucontext
;
2382 ibdev
->ib_dev
.dealloc_ucontext
= mlx4_ib_dealloc_ucontext
;
2383 ibdev
->ib_dev
.mmap
= mlx4_ib_mmap
;
2384 ibdev
->ib_dev
.alloc_pd
= mlx4_ib_alloc_pd
;
2385 ibdev
->ib_dev
.dealloc_pd
= mlx4_ib_dealloc_pd
;
2386 ibdev
->ib_dev
.create_ah
= mlx4_ib_create_ah
;
2387 ibdev
->ib_dev
.query_ah
= mlx4_ib_query_ah
;
2388 ibdev
->ib_dev
.destroy_ah
= mlx4_ib_destroy_ah
;
2389 ibdev
->ib_dev
.create_srq
= mlx4_ib_create_srq
;
2390 ibdev
->ib_dev
.modify_srq
= mlx4_ib_modify_srq
;
2391 ibdev
->ib_dev
.query_srq
= mlx4_ib_query_srq
;
2392 ibdev
->ib_dev
.destroy_srq
= mlx4_ib_destroy_srq
;
2393 ibdev
->ib_dev
.post_srq_recv
= mlx4_ib_post_srq_recv
;
2394 ibdev
->ib_dev
.create_qp
= mlx4_ib_create_qp
;
2395 ibdev
->ib_dev
.modify_qp
= mlx4_ib_modify_qp
;
2396 ibdev
->ib_dev
.query_qp
= mlx4_ib_query_qp
;
2397 ibdev
->ib_dev
.destroy_qp
= mlx4_ib_destroy_qp
;
2398 ibdev
->ib_dev
.post_send
= mlx4_ib_post_send
;
2399 ibdev
->ib_dev
.post_recv
= mlx4_ib_post_recv
;
2400 ibdev
->ib_dev
.create_cq
= mlx4_ib_create_cq
;
2401 ibdev
->ib_dev
.modify_cq
= mlx4_ib_modify_cq
;
2402 ibdev
->ib_dev
.resize_cq
= mlx4_ib_resize_cq
;
2403 ibdev
->ib_dev
.destroy_cq
= mlx4_ib_destroy_cq
;
2404 ibdev
->ib_dev
.poll_cq
= mlx4_ib_poll_cq
;
2405 ibdev
->ib_dev
.req_notify_cq
= mlx4_ib_arm_cq
;
2406 ibdev
->ib_dev
.get_dma_mr
= mlx4_ib_get_dma_mr
;
2407 ibdev
->ib_dev
.reg_user_mr
= mlx4_ib_reg_user_mr
;
2408 ibdev
->ib_dev
.rereg_user_mr
= mlx4_ib_rereg_user_mr
;
2409 ibdev
->ib_dev
.dereg_mr
= mlx4_ib_dereg_mr
;
2410 ibdev
->ib_dev
.alloc_mr
= mlx4_ib_alloc_mr
;
2411 ibdev
->ib_dev
.map_mr_sg
= mlx4_ib_map_mr_sg
;
2412 ibdev
->ib_dev
.attach_mcast
= mlx4_ib_mcg_attach
;
2413 ibdev
->ib_dev
.detach_mcast
= mlx4_ib_mcg_detach
;
2414 ibdev
->ib_dev
.process_mad
= mlx4_ib_process_mad
;
2415 ibdev
->ib_dev
.get_port_immutable
= mlx4_port_immutable
;
2416 ibdev
->ib_dev
.disassociate_ucontext
= mlx4_ib_disassociate_ucontext
;
2418 if (!mlx4_is_slave(ibdev
->dev
)) {
2419 ibdev
->ib_dev
.alloc_fmr
= mlx4_ib_fmr_alloc
;
2420 ibdev
->ib_dev
.map_phys_fmr
= mlx4_ib_map_phys_fmr
;
2421 ibdev
->ib_dev
.unmap_fmr
= mlx4_ib_unmap_fmr
;
2422 ibdev
->ib_dev
.dealloc_fmr
= mlx4_ib_fmr_dealloc
;
2425 if (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_MEM_WINDOW
||
2426 dev
->caps
.bmme_flags
& MLX4_BMME_FLAG_TYPE_2_WIN
) {
2427 ibdev
->ib_dev
.alloc_mw
= mlx4_ib_alloc_mw
;
2428 ibdev
->ib_dev
.dealloc_mw
= mlx4_ib_dealloc_mw
;
2430 ibdev
->ib_dev
.uverbs_cmd_mask
|=
2431 (1ull << IB_USER_VERBS_CMD_ALLOC_MW
) |
2432 (1ull << IB_USER_VERBS_CMD_DEALLOC_MW
);
2435 if (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_XRC
) {
2436 ibdev
->ib_dev
.alloc_xrcd
= mlx4_ib_alloc_xrcd
;
2437 ibdev
->ib_dev
.dealloc_xrcd
= mlx4_ib_dealloc_xrcd
;
2438 ibdev
->ib_dev
.uverbs_cmd_mask
|=
2439 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD
) |
2440 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD
);
2443 if (check_flow_steering_support(dev
)) {
2444 ibdev
->steering_support
= MLX4_STEERING_MODE_DEVICE_MANAGED
;
2445 ibdev
->ib_dev
.create_flow
= mlx4_ib_create_flow
;
2446 ibdev
->ib_dev
.destroy_flow
= mlx4_ib_destroy_flow
;
2448 ibdev
->ib_dev
.uverbs_ex_cmd_mask
|=
2449 (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW
) |
2450 (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW
);
2453 ibdev
->ib_dev
.uverbs_ex_cmd_mask
|=
2454 (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE
) |
2455 (1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ
) |
2456 (1ull << IB_USER_VERBS_EX_CMD_CREATE_QP
);
2458 mlx4_ib_alloc_eqs(dev
, ibdev
);
2460 spin_lock_init(&iboe
->lock
);
2462 if (init_node_data(ibdev
))
2465 for (i
= 0; i
< ibdev
->num_ports
; ++i
) {
2466 mutex_init(&ibdev
->counters_table
[i
].mutex
);
2467 INIT_LIST_HEAD(&ibdev
->counters_table
[i
].counters_list
);
2470 num_req_counters
= mlx4_is_bonded(dev
) ? 1 : ibdev
->num_ports
;
2471 for (i
= 0; i
< num_req_counters
; ++i
) {
2472 mutex_init(&ibdev
->qp1_proxy_lock
[i
]);
2474 if (mlx4_ib_port_link_layer(&ibdev
->ib_dev
, i
+ 1) ==
2475 IB_LINK_LAYER_ETHERNET
) {
2476 err
= mlx4_counter_alloc(ibdev
->dev
, &counter_index
);
2477 /* if failed to allocate a new counter, use default */
2480 mlx4_get_default_counter_index(dev
,
2484 } else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2485 counter_index
= mlx4_get_default_counter_index(dev
,
2488 new_counter_index
= kmalloc(sizeof(*new_counter_index
),
2490 if (!new_counter_index
) {
2492 mlx4_counter_free(ibdev
->dev
, counter_index
);
2495 new_counter_index
->index
= counter_index
;
2496 new_counter_index
->allocated
= allocated
;
2497 list_add_tail(&new_counter_index
->list
,
2498 &ibdev
->counters_table
[i
].counters_list
);
2499 ibdev
->counters_table
[i
].default_counter
= counter_index
;
2500 pr_info("counter index %d for port %d allocated %d\n",
2501 counter_index
, i
+ 1, allocated
);
2503 if (mlx4_is_bonded(dev
))
2504 for (i
= 1; i
< ibdev
->num_ports
; ++i
) {
2506 kmalloc(sizeof(struct counter_index
),
2508 if (!new_counter_index
)
2510 new_counter_index
->index
= counter_index
;
2511 new_counter_index
->allocated
= 0;
2512 list_add_tail(&new_counter_index
->list
,
2513 &ibdev
->counters_table
[i
].counters_list
);
2514 ibdev
->counters_table
[i
].default_counter
=
2518 mlx4_foreach_port(i
, dev
, MLX4_PORT_TYPE_IB
)
2521 spin_lock_init(&ibdev
->sm_lock
);
2522 mutex_init(&ibdev
->cap_mask_mutex
);
2523 INIT_LIST_HEAD(&ibdev
->qp_list
);
2524 spin_lock_init(&ibdev
->reset_flow_resource_lock
);
2526 if (ibdev
->steering_support
== MLX4_STEERING_MODE_DEVICE_MANAGED
&&
2528 ibdev
->steer_qpn_count
= MLX4_IB_UC_MAX_NUM_QPS
;
2529 err
= mlx4_qp_reserve_range(dev
, ibdev
->steer_qpn_count
,
2530 MLX4_IB_UC_STEER_QPN_ALIGN
,
2531 &ibdev
->steer_qpn_base
, 0);
2535 ibdev
->ib_uc_qpns_bitmap
=
2536 kmalloc(BITS_TO_LONGS(ibdev
->steer_qpn_count
) *
2539 if (!ibdev
->ib_uc_qpns_bitmap
) {
2540 dev_err(&dev
->persist
->pdev
->dev
,
2541 "bit map alloc failed\n");
2542 goto err_steer_qp_release
;
2545 bitmap_zero(ibdev
->ib_uc_qpns_bitmap
, ibdev
->steer_qpn_count
);
2547 err
= mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2548 dev
, ibdev
->steer_qpn_base
,
2549 ibdev
->steer_qpn_base
+
2550 ibdev
->steer_qpn_count
- 1);
2552 goto err_steer_free_bitmap
;
2555 for (j
= 1; j
<= ibdev
->dev
->caps
.num_ports
; j
++)
2556 atomic64_set(&iboe
->mac
[j
- 1], ibdev
->dev
->caps
.def_mac
[j
]);
2558 if (ib_register_device(&ibdev
->ib_dev
, NULL
))
2559 goto err_steer_free_bitmap
;
2561 if (mlx4_ib_mad_init(ibdev
))
2564 if (mlx4_ib_init_sriov(ibdev
))
2567 if (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_IBOE
||
2568 dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_ROCE_V1_V2
) {
2569 if (!iboe
->nb
.notifier_call
) {
2570 iboe
->nb
.notifier_call
= mlx4_ib_netdev_event
;
2571 err
= register_netdevice_notifier(&iboe
->nb
);
2573 iboe
->nb
.notifier_call
= NULL
;
2577 if (dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_ROCE_V1_V2
) {
2578 err
= mlx4_config_roce_v2_port(dev
, ROCE_V2_UDP_DPORT
);
2585 for (j
= 0; j
< ARRAY_SIZE(mlx4_class_attributes
); ++j
) {
2586 if (device_create_file(&ibdev
->ib_dev
.dev
,
2587 mlx4_class_attributes
[j
]))
2591 ibdev
->ib_active
= true;
2592 mlx4_foreach_port(i
, dev
, MLX4_PORT_TYPE_IB
)
2593 devlink_port_type_ib_set(mlx4_get_devlink_port(dev
, i
),
2596 if (mlx4_is_mfunc(ibdev
->dev
))
2599 /* create paravirt contexts for any VFs which are active */
2600 if (mlx4_is_master(ibdev
->dev
)) {
2601 for (j
= 0; j
< MLX4_MFUNC_MAX
; j
++) {
2602 if (j
== mlx4_master_func_num(ibdev
->dev
))
2604 if (mlx4_is_slave_active(ibdev
->dev
, j
))
2605 do_slave_init(ibdev
, j
, 1);
2611 if (ibdev
->iboe
.nb
.notifier_call
) {
2612 if (unregister_netdevice_notifier(&ibdev
->iboe
.nb
))
2613 pr_warn("failure unregistering notifier\n");
2614 ibdev
->iboe
.nb
.notifier_call
= NULL
;
2616 flush_workqueue(wq
);
2618 mlx4_ib_close_sriov(ibdev
);
2621 mlx4_ib_mad_cleanup(ibdev
);
2624 ib_unregister_device(&ibdev
->ib_dev
);
2626 err_steer_free_bitmap
:
2627 kfree(ibdev
->ib_uc_qpns_bitmap
);
2629 err_steer_qp_release
:
2630 if (ibdev
->steering_support
== MLX4_STEERING_MODE_DEVICE_MANAGED
)
2631 mlx4_qp_release_range(dev
, ibdev
->steer_qpn_base
,
2632 ibdev
->steer_qpn_count
);
2634 for (i
= 0; i
< ibdev
->num_ports
; ++i
)
2635 mlx4_ib_delete_counters_table(ibdev
, &ibdev
->counters_table
[i
]);
2638 iounmap(ibdev
->uar_map
);
2641 mlx4_uar_free(dev
, &ibdev
->priv_uar
);
2644 mlx4_pd_free(dev
, ibdev
->priv_pdn
);
2647 ib_dealloc_device(&ibdev
->ib_dev
);
2652 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev
*dev
, int count
, int *qpn
)
2656 WARN_ON(!dev
->ib_uc_qpns_bitmap
);
2658 offset
= bitmap_find_free_region(dev
->ib_uc_qpns_bitmap
,
2659 dev
->steer_qpn_count
,
2660 get_count_order(count
));
2664 *qpn
= dev
->steer_qpn_base
+ offset
;
2668 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev
*dev
, u32 qpn
, int count
)
2671 dev
->steering_support
!= MLX4_STEERING_MODE_DEVICE_MANAGED
)
2674 BUG_ON(qpn
< dev
->steer_qpn_base
);
2676 bitmap_release_region(dev
->ib_uc_qpns_bitmap
,
2677 qpn
- dev
->steer_qpn_base
,
2678 get_count_order(count
));
2681 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev
*mdev
, struct mlx4_ib_qp
*mqp
,
2686 struct ib_flow_attr
*flow
= NULL
;
2687 struct ib_flow_spec_ib
*ib_spec
;
2690 flow_size
= sizeof(struct ib_flow_attr
) +
2691 sizeof(struct ib_flow_spec_ib
);
2692 flow
= kzalloc(flow_size
, GFP_KERNEL
);
2695 flow
->port
= mqp
->port
;
2696 flow
->num_of_specs
= 1;
2697 flow
->size
= flow_size
;
2698 ib_spec
= (struct ib_flow_spec_ib
*)(flow
+ 1);
2699 ib_spec
->type
= IB_FLOW_SPEC_IB
;
2700 ib_spec
->size
= sizeof(struct ib_flow_spec_ib
);
2701 /* Add an empty rule for IB L2 */
2702 memset(&ib_spec
->mask
, 0, sizeof(ib_spec
->mask
));
2704 err
= __mlx4_ib_create_flow(&mqp
->ibqp
, flow
,
2709 err
= __mlx4_ib_destroy_flow(mdev
->dev
, mqp
->reg_id
);
2715 static void mlx4_ib_remove(struct mlx4_dev
*dev
, void *ibdev_ptr
)
2717 struct mlx4_ib_dev
*ibdev
= ibdev_ptr
;
2721 mlx4_foreach_port(i
, dev
, MLX4_PORT_TYPE_IB
)
2722 devlink_port_type_clear(mlx4_get_devlink_port(dev
, i
));
2723 ibdev
->ib_active
= false;
2724 flush_workqueue(wq
);
2726 mlx4_ib_close_sriov(ibdev
);
2727 mlx4_ib_mad_cleanup(ibdev
);
2728 ib_unregister_device(&ibdev
->ib_dev
);
2729 if (ibdev
->iboe
.nb
.notifier_call
) {
2730 if (unregister_netdevice_notifier(&ibdev
->iboe
.nb
))
2731 pr_warn("failure unregistering notifier\n");
2732 ibdev
->iboe
.nb
.notifier_call
= NULL
;
2735 if (ibdev
->steering_support
== MLX4_STEERING_MODE_DEVICE_MANAGED
) {
2736 mlx4_qp_release_range(dev
, ibdev
->steer_qpn_base
,
2737 ibdev
->steer_qpn_count
);
2738 kfree(ibdev
->ib_uc_qpns_bitmap
);
2741 iounmap(ibdev
->uar_map
);
2742 for (p
= 0; p
< ibdev
->num_ports
; ++p
)
2743 mlx4_ib_delete_counters_table(ibdev
, &ibdev
->counters_table
[p
]);
2745 mlx4_foreach_port(p
, dev
, MLX4_PORT_TYPE_IB
)
2746 mlx4_CLOSE_PORT(dev
, p
);
2748 mlx4_ib_free_eqs(dev
, ibdev
);
2750 mlx4_uar_free(dev
, &ibdev
->priv_uar
);
2751 mlx4_pd_free(dev
, ibdev
->priv_pdn
);
2752 ib_dealloc_device(&ibdev
->ib_dev
);
2755 static void do_slave_init(struct mlx4_ib_dev
*ibdev
, int slave
, int do_init
)
2757 struct mlx4_ib_demux_work
**dm
= NULL
;
2758 struct mlx4_dev
*dev
= ibdev
->dev
;
2760 unsigned long flags
;
2761 struct mlx4_active_ports actv_ports
;
2763 unsigned int first_port
;
2765 if (!mlx4_is_master(dev
))
2768 actv_ports
= mlx4_get_active_ports(dev
, slave
);
2769 ports
= bitmap_weight(actv_ports
.ports
, dev
->caps
.num_ports
);
2770 first_port
= find_first_bit(actv_ports
.ports
, dev
->caps
.num_ports
);
2772 dm
= kcalloc(ports
, sizeof(*dm
), GFP_ATOMIC
);
2774 pr_err("failed to allocate memory for tunneling qp update\n");
2778 for (i
= 0; i
< ports
; i
++) {
2779 dm
[i
] = kmalloc(sizeof (struct mlx4_ib_demux_work
), GFP_ATOMIC
);
2781 pr_err("failed to allocate memory for tunneling qp update work struct\n");
2786 INIT_WORK(&dm
[i
]->work
, mlx4_ib_tunnels_update_work
);
2787 dm
[i
]->port
= first_port
+ i
+ 1;
2788 dm
[i
]->slave
= slave
;
2789 dm
[i
]->do_init
= do_init
;
2792 /* initialize or tear down tunnel QPs for the slave */
2793 spin_lock_irqsave(&ibdev
->sriov
.going_down_lock
, flags
);
2794 if (!ibdev
->sriov
.is_going_down
) {
2795 for (i
= 0; i
< ports
; i
++)
2796 queue_work(ibdev
->sriov
.demux
[i
].ud_wq
, &dm
[i
]->work
);
2797 spin_unlock_irqrestore(&ibdev
->sriov
.going_down_lock
, flags
);
2799 spin_unlock_irqrestore(&ibdev
->sriov
.going_down_lock
, flags
);
2800 for (i
= 0; i
< ports
; i
++)
2808 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev
*ibdev
)
2810 struct mlx4_ib_qp
*mqp
;
2811 unsigned long flags_qp
;
2812 unsigned long flags_cq
;
2813 struct mlx4_ib_cq
*send_mcq
, *recv_mcq
;
2814 struct list_head cq_notify_list
;
2815 struct mlx4_cq
*mcq
;
2816 unsigned long flags
;
2818 pr_warn("mlx4_ib_handle_catas_error was started\n");
2819 INIT_LIST_HEAD(&cq_notify_list
);
2821 /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2822 spin_lock_irqsave(&ibdev
->reset_flow_resource_lock
, flags
);
2824 list_for_each_entry(mqp
, &ibdev
->qp_list
, qps_list
) {
2825 spin_lock_irqsave(&mqp
->sq
.lock
, flags_qp
);
2826 if (mqp
->sq
.tail
!= mqp
->sq
.head
) {
2827 send_mcq
= to_mcq(mqp
->ibqp
.send_cq
);
2828 spin_lock_irqsave(&send_mcq
->lock
, flags_cq
);
2829 if (send_mcq
->mcq
.comp
&&
2830 mqp
->ibqp
.send_cq
->comp_handler
) {
2831 if (!send_mcq
->mcq
.reset_notify_added
) {
2832 send_mcq
->mcq
.reset_notify_added
= 1;
2833 list_add_tail(&send_mcq
->mcq
.reset_notify
,
2837 spin_unlock_irqrestore(&send_mcq
->lock
, flags_cq
);
2839 spin_unlock_irqrestore(&mqp
->sq
.lock
, flags_qp
);
2840 /* Now, handle the QP's receive queue */
2841 spin_lock_irqsave(&mqp
->rq
.lock
, flags_qp
);
2842 /* no handling is needed for SRQ */
2843 if (!mqp
->ibqp
.srq
) {
2844 if (mqp
->rq
.tail
!= mqp
->rq
.head
) {
2845 recv_mcq
= to_mcq(mqp
->ibqp
.recv_cq
);
2846 spin_lock_irqsave(&recv_mcq
->lock
, flags_cq
);
2847 if (recv_mcq
->mcq
.comp
&&
2848 mqp
->ibqp
.recv_cq
->comp_handler
) {
2849 if (!recv_mcq
->mcq
.reset_notify_added
) {
2850 recv_mcq
->mcq
.reset_notify_added
= 1;
2851 list_add_tail(&recv_mcq
->mcq
.reset_notify
,
2855 spin_unlock_irqrestore(&recv_mcq
->lock
,
2859 spin_unlock_irqrestore(&mqp
->rq
.lock
, flags_qp
);
2862 list_for_each_entry(mcq
, &cq_notify_list
, reset_notify
) {
2865 spin_unlock_irqrestore(&ibdev
->reset_flow_resource_lock
, flags
);
2866 pr_warn("mlx4_ib_handle_catas_error ended\n");
2869 static void handle_bonded_port_state_event(struct work_struct
*work
)
2871 struct ib_event_work
*ew
=
2872 container_of(work
, struct ib_event_work
, work
);
2873 struct mlx4_ib_dev
*ibdev
= ew
->ib_dev
;
2874 enum ib_port_state bonded_port_state
= IB_PORT_NOP
;
2876 struct ib_event ibev
;
2879 spin_lock_bh(&ibdev
->iboe
.lock
);
2880 for (i
= 0; i
< MLX4_MAX_PORTS
; ++i
) {
2881 struct net_device
*curr_netdev
= ibdev
->iboe
.netdevs
[i
];
2882 enum ib_port_state curr_port_state
;
2888 (netif_running(curr_netdev
) &&
2889 netif_carrier_ok(curr_netdev
)) ?
2890 IB_PORT_ACTIVE
: IB_PORT_DOWN
;
2892 bonded_port_state
= (bonded_port_state
!= IB_PORT_ACTIVE
) ?
2893 curr_port_state
: IB_PORT_ACTIVE
;
2895 spin_unlock_bh(&ibdev
->iboe
.lock
);
2897 ibev
.device
= &ibdev
->ib_dev
;
2898 ibev
.element
.port_num
= 1;
2899 ibev
.event
= (bonded_port_state
== IB_PORT_ACTIVE
) ?
2900 IB_EVENT_PORT_ACTIVE
: IB_EVENT_PORT_ERR
;
2902 ib_dispatch_event(&ibev
);
2905 static void mlx4_ib_event(struct mlx4_dev
*dev
, void *ibdev_ptr
,
2906 enum mlx4_dev_event event
, unsigned long param
)
2908 struct ib_event ibev
;
2909 struct mlx4_ib_dev
*ibdev
= to_mdev((struct ib_device
*) ibdev_ptr
);
2910 struct mlx4_eqe
*eqe
= NULL
;
2911 struct ib_event_work
*ew
;
2914 if (mlx4_is_bonded(dev
) &&
2915 ((event
== MLX4_DEV_EVENT_PORT_UP
) ||
2916 (event
== MLX4_DEV_EVENT_PORT_DOWN
))) {
2917 ew
= kmalloc(sizeof(*ew
), GFP_ATOMIC
);
2920 INIT_WORK(&ew
->work
, handle_bonded_port_state_event
);
2922 queue_work(wq
, &ew
->work
);
2926 if (event
== MLX4_DEV_EVENT_PORT_MGMT_CHANGE
)
2927 eqe
= (struct mlx4_eqe
*)param
;
2932 case MLX4_DEV_EVENT_PORT_UP
:
2933 if (p
> ibdev
->num_ports
)
2935 if (mlx4_is_master(dev
) &&
2936 rdma_port_get_link_layer(&ibdev
->ib_dev
, p
) ==
2937 IB_LINK_LAYER_INFINIBAND
) {
2938 mlx4_ib_invalidate_all_guid_record(ibdev
, p
);
2940 ibev
.event
= IB_EVENT_PORT_ACTIVE
;
2943 case MLX4_DEV_EVENT_PORT_DOWN
:
2944 if (p
> ibdev
->num_ports
)
2946 ibev
.event
= IB_EVENT_PORT_ERR
;
2949 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR
:
2950 ibdev
->ib_active
= false;
2951 ibev
.event
= IB_EVENT_DEVICE_FATAL
;
2952 mlx4_ib_handle_catas_error(ibdev
);
2955 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE
:
2956 ew
= kmalloc(sizeof *ew
, GFP_ATOMIC
);
2958 pr_err("failed to allocate memory for events work\n");
2962 INIT_WORK(&ew
->work
, handle_port_mgmt_change_event
);
2963 memcpy(&ew
->ib_eqe
, eqe
, sizeof *eqe
);
2965 /* need to queue only for port owner, which uses GEN_EQE */
2966 if (mlx4_is_master(dev
))
2967 queue_work(wq
, &ew
->work
);
2969 handle_port_mgmt_change_event(&ew
->work
);
2972 case MLX4_DEV_EVENT_SLAVE_INIT
:
2973 /* here, p is the slave id */
2974 do_slave_init(ibdev
, p
, 1);
2975 if (mlx4_is_master(dev
)) {
2978 for (i
= 1; i
<= ibdev
->num_ports
; i
++) {
2979 if (rdma_port_get_link_layer(&ibdev
->ib_dev
, i
)
2980 == IB_LINK_LAYER_INFINIBAND
)
2981 mlx4_ib_slave_alias_guid_event(ibdev
,
2988 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN
:
2989 if (mlx4_is_master(dev
)) {
2992 for (i
= 1; i
<= ibdev
->num_ports
; i
++) {
2993 if (rdma_port_get_link_layer(&ibdev
->ib_dev
, i
)
2994 == IB_LINK_LAYER_INFINIBAND
)
2995 mlx4_ib_slave_alias_guid_event(ibdev
,
3000 /* here, p is the slave id */
3001 do_slave_init(ibdev
, p
, 0);
3008 ibev
.device
= ibdev_ptr
;
3009 ibev
.element
.port_num
= mlx4_is_bonded(ibdev
->dev
) ? 1 : (u8
)p
;
3011 ib_dispatch_event(&ibev
);
3014 static struct mlx4_interface mlx4_ib_interface
= {
3016 .remove
= mlx4_ib_remove
,
3017 .event
= mlx4_ib_event
,
3018 .protocol
= MLX4_PROT_IB_IPV6
,
3019 .flags
= MLX4_INTFF_BONDING
3022 static int __init
mlx4_ib_init(void)
3026 wq
= create_singlethread_workqueue("mlx4_ib");
3030 err
= mlx4_ib_mcg_init();
3034 err
= mlx4_register_interface(&mlx4_ib_interface
);
3041 mlx4_ib_mcg_destroy();
3044 destroy_workqueue(wq
);
3048 static void __exit
mlx4_ib_cleanup(void)
3050 mlx4_unregister_interface(&mlx4_ib_interface
);
3051 mlx4_ib_mcg_destroy();
3052 destroy_workqueue(wq
);
3055 module_init(mlx4_ib_init
);
3056 module_exit(mlx4_ib_cleanup
);