2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
5 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
7 * This software is available to you under a choice of one of two
8 * licenses. You may choose to be licensed under the terms of the GNU
9 * General Public License (GPL) Version 2, available from the file
10 * COPYING in the main directory of this source tree, or the
11 * OpenIB.org BSD license below:
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
21 * - Redistributions in binary form must reproduce the above
22 * copyright notice, this list of conditions and the following
23 * disclaimer in the documentation and/or other materials
24 * provided with the distribution.
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 #include <linux/module.h>
37 #include <linux/kernel.h>
38 #include <linux/init.h>
39 #include <linux/errno.h>
40 #include <linux/pci.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/slab.h>
43 #include <linux/io-mapping.h>
44 #include <linux/delay.h>
45 #include <linux/kmod.h>
46 #include <linux/etherdevice.h>
47 #include <net/devlink.h>
49 #include <uapi/rdma/mlx4-abi.h>
50 #include <linux/mlx4/device.h>
51 #include <linux/mlx4/doorbell.h>
57 MODULE_AUTHOR("Roland Dreier");
58 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
59 MODULE_LICENSE("Dual BSD/GPL");
60 MODULE_VERSION(DRV_VERSION
);
62 struct workqueue_struct
*mlx4_wq
;
64 #ifdef CONFIG_MLX4_DEBUG
66 int mlx4_debug_level
; /* 0 by default */
67 module_param_named(debug_level
, mlx4_debug_level
, int, 0644);
68 MODULE_PARM_DESC(debug_level
, "Enable debug tracing if > 0");
70 #endif /* CONFIG_MLX4_DEBUG */
75 module_param(msi_x
, int, 0444);
76 MODULE_PARM_DESC(msi_x
, "0 - don't use MSI-X, 1 - use MSI-X, >1 - limit number of MSI-X irqs to msi_x");
78 #else /* CONFIG_PCI_MSI */
82 #endif /* CONFIG_PCI_MSI */
84 static uint8_t num_vfs
[3] = {0, 0, 0};
85 static int num_vfs_argc
;
86 module_param_array(num_vfs
, byte
, &num_vfs_argc
, 0444);
87 MODULE_PARM_DESC(num_vfs
, "enable #num_vfs functions if num_vfs > 0\n"
88 "num_vfs=port1,port2,port1+2");
90 static uint8_t probe_vf
[3] = {0, 0, 0};
91 static int probe_vfs_argc
;
92 module_param_array(probe_vf
, byte
, &probe_vfs_argc
, 0444);
93 MODULE_PARM_DESC(probe_vf
, "number of vfs to probe by pf driver (num_vfs > 0)\n"
94 "probe_vf=port1,port2,port1+2");
96 static int mlx4_log_num_mgm_entry_size
= MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE
;
97 module_param_named(log_num_mgm_entry_size
,
98 mlx4_log_num_mgm_entry_size
, int, 0444);
99 MODULE_PARM_DESC(log_num_mgm_entry_size
, "log mgm size, that defines the num"
100 " of qp per mcg, for example:"
101 " 10 gives 248.range: 7 <="
102 " log_num_mgm_entry_size <= 12."
103 " To activate device managed"
104 " flow steering when available, set to -1");
106 static bool enable_64b_cqe_eqe
= true;
107 module_param(enable_64b_cqe_eqe
, bool, 0444);
108 MODULE_PARM_DESC(enable_64b_cqe_eqe
,
109 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
111 static bool enable_4k_uar
;
112 module_param(enable_4k_uar
, bool, 0444);
113 MODULE_PARM_DESC(enable_4k_uar
,
114 "Enable using 4K UAR. Should not be enabled if have VFs which do not support 4K UARs (default: false)");
116 #define PF_CONTEXT_BEHAVIOUR_MASK (MLX4_FUNC_CAP_64B_EQE_CQE | \
117 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
118 MLX4_FUNC_CAP_DMFS_A0_STATIC)
120 #define RESET_PERSIST_MASK_FLAGS (MLX4_FLAG_SRIOV)
122 static char mlx4_version
[] =
123 DRV_NAME
": Mellanox ConnectX core driver v"
126 static const struct mlx4_profile default_profile
= {
129 .rdmarc_per_qp
= 1 << 4,
133 .num_mtt
= 1 << 20, /* It is really num mtt segements */
136 static const struct mlx4_profile low_mem_profile
= {
139 .rdmarc_per_qp
= 1 << 4,
146 static int log_num_mac
= 7;
147 module_param_named(log_num_mac
, log_num_mac
, int, 0444);
148 MODULE_PARM_DESC(log_num_mac
, "Log2 max number of MACs per ETH port (1-7)");
150 static int log_num_vlan
;
151 module_param_named(log_num_vlan
, log_num_vlan
, int, 0444);
152 MODULE_PARM_DESC(log_num_vlan
, "Log2 max number of VLANs per ETH port (0-7)");
153 /* Log2 max number of VLANs per ETH port (0-7) */
154 #define MLX4_LOG_NUM_VLANS 7
155 #define MLX4_MIN_LOG_NUM_VLANS 0
156 #define MLX4_MIN_LOG_NUM_MAC 1
158 static bool use_prio
;
159 module_param_named(use_prio
, use_prio
, bool, 0444);
160 MODULE_PARM_DESC(use_prio
, "Enable steering by VLAN priority on ETH ports (deprecated)");
162 int log_mtts_per_seg
= ilog2(1);
163 module_param_named(log_mtts_per_seg
, log_mtts_per_seg
, int, 0444);
164 MODULE_PARM_DESC(log_mtts_per_seg
, "Log2 number of MTT entries per segment "
165 "(0-7) (default: 0)");
167 static int port_type_array
[2] = {MLX4_PORT_TYPE_NONE
, MLX4_PORT_TYPE_NONE
};
168 static int arr_argc
= 2;
169 module_param_array(port_type_array
, int, &arr_argc
, 0444);
170 MODULE_PARM_DESC(port_type_array
, "Array of port types: HW_DEFAULT (0) is default "
171 "1 for IB, 2 for Ethernet");
173 struct mlx4_port_config
{
174 struct list_head list
;
175 enum mlx4_port_type port_type
[MLX4_MAX_PORTS
+ 1];
176 struct pci_dev
*pdev
;
179 static atomic_t pf_loading
= ATOMIC_INIT(0);
181 static int mlx4_devlink_ierr_reset_get(struct devlink
*devlink
, u32 id
,
182 struct devlink_param_gset_ctx
*ctx
)
184 ctx
->val
.vbool
= !!mlx4_internal_err_reset
;
188 static int mlx4_devlink_ierr_reset_set(struct devlink
*devlink
, u32 id
,
189 struct devlink_param_gset_ctx
*ctx
)
191 mlx4_internal_err_reset
= ctx
->val
.vbool
;
195 static int mlx4_devlink_crdump_snapshot_get(struct devlink
*devlink
, u32 id
,
196 struct devlink_param_gset_ctx
*ctx
)
198 struct mlx4_priv
*priv
= devlink_priv(devlink
);
199 struct mlx4_dev
*dev
= &priv
->dev
;
201 ctx
->val
.vbool
= dev
->persist
->crdump
.snapshot_enable
;
205 static int mlx4_devlink_crdump_snapshot_set(struct devlink
*devlink
, u32 id
,
206 struct devlink_param_gset_ctx
*ctx
)
208 struct mlx4_priv
*priv
= devlink_priv(devlink
);
209 struct mlx4_dev
*dev
= &priv
->dev
;
211 dev
->persist
->crdump
.snapshot_enable
= ctx
->val
.vbool
;
216 mlx4_devlink_max_macs_validate(struct devlink
*devlink
, u32 id
,
217 union devlink_param_value val
,
218 struct netlink_ext_ack
*extack
)
220 u32 value
= val
.vu32
;
222 if (value
< 1 || value
> 128)
225 if (!is_power_of_2(value
)) {
226 NL_SET_ERR_MSG_MOD(extack
, "max_macs supported must be power of 2");
233 enum mlx4_devlink_param_id
{
234 MLX4_DEVLINK_PARAM_ID_BASE
= DEVLINK_PARAM_GENERIC_ID_MAX
,
235 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE
,
236 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR
,
239 static const struct devlink_param mlx4_devlink_params
[] = {
240 DEVLINK_PARAM_GENERIC(INT_ERR_RESET
,
241 BIT(DEVLINK_PARAM_CMODE_RUNTIME
) |
242 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT
),
243 mlx4_devlink_ierr_reset_get
,
244 mlx4_devlink_ierr_reset_set
, NULL
),
245 DEVLINK_PARAM_GENERIC(MAX_MACS
,
246 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT
),
247 NULL
, NULL
, mlx4_devlink_max_macs_validate
),
248 DEVLINK_PARAM_GENERIC(REGION_SNAPSHOT
,
249 BIT(DEVLINK_PARAM_CMODE_RUNTIME
) |
250 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT
),
251 mlx4_devlink_crdump_snapshot_get
,
252 mlx4_devlink_crdump_snapshot_set
, NULL
),
253 DEVLINK_PARAM_DRIVER(MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE
,
254 "enable_64b_cqe_eqe", DEVLINK_PARAM_TYPE_BOOL
,
255 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT
),
257 DEVLINK_PARAM_DRIVER(MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR
,
258 "enable_4k_uar", DEVLINK_PARAM_TYPE_BOOL
,
259 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT
),
263 static void mlx4_devlink_set_params_init_values(struct devlink
*devlink
)
265 union devlink_param_value value
;
267 value
.vbool
= !!mlx4_internal_err_reset
;
268 devlink_param_driverinit_value_set(devlink
,
269 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET
,
272 value
.vu32
= 1UL << log_num_mac
;
273 devlink_param_driverinit_value_set(devlink
,
274 DEVLINK_PARAM_GENERIC_ID_MAX_MACS
,
277 value
.vbool
= enable_64b_cqe_eqe
;
278 devlink_param_driverinit_value_set(devlink
,
279 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE
,
282 value
.vbool
= enable_4k_uar
;
283 devlink_param_driverinit_value_set(devlink
,
284 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR
,
288 devlink_param_driverinit_value_set(devlink
,
289 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT
,
293 static inline void mlx4_set_num_reserved_uars(struct mlx4_dev
*dev
,
294 struct mlx4_dev_cap
*dev_cap
)
296 /* The reserved_uars is calculated by system page size unit.
297 * Therefore, adjustment is added when the uar page size is less
298 * than the system page size
300 dev
->caps
.reserved_uars
=
302 mlx4_get_num_reserved_uar(dev
),
303 dev_cap
->reserved_uars
/
304 (1 << (PAGE_SHIFT
- dev
->uar_page_shift
)));
307 int mlx4_check_port_params(struct mlx4_dev
*dev
,
308 enum mlx4_port_type
*port_type
)
312 if (!(dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_DPDP
)) {
313 for (i
= 0; i
< dev
->caps
.num_ports
- 1; i
++) {
314 if (port_type
[i
] != port_type
[i
+ 1]) {
315 mlx4_err(dev
, "Only same port types supported on this HCA, aborting\n");
321 for (i
= 0; i
< dev
->caps
.num_ports
; i
++) {
322 if (!(port_type
[i
] & dev
->caps
.supported_type
[i
+1])) {
323 mlx4_err(dev
, "Requested port type for port %d is not supported on this HCA\n",
331 static void mlx4_set_port_mask(struct mlx4_dev
*dev
)
335 for (i
= 1; i
<= dev
->caps
.num_ports
; ++i
)
336 dev
->caps
.port_mask
[i
] = dev
->caps
.port_type
[i
];
340 MLX4_QUERY_FUNC_NUM_SYS_EQS
= 1 << 0,
343 static int mlx4_query_func(struct mlx4_dev
*dev
, struct mlx4_dev_cap
*dev_cap
)
346 struct mlx4_func func
;
348 if (dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_SYS_EQS
) {
349 err
= mlx4_QUERY_FUNC(dev
, &func
, 0);
351 mlx4_err(dev
, "QUERY_DEV_CAP command failed, aborting.\n");
354 dev_cap
->max_eqs
= func
.max_eq
;
355 dev_cap
->reserved_eqs
= func
.rsvd_eqs
;
356 dev_cap
->reserved_uars
= func
.rsvd_uars
;
357 err
|= MLX4_QUERY_FUNC_NUM_SYS_EQS
;
362 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev
*dev
)
364 struct mlx4_caps
*dev_cap
= &dev
->caps
;
366 /* FW not supporting or cancelled by user */
367 if (!(dev_cap
->flags2
& MLX4_DEV_CAP_FLAG2_EQE_STRIDE
) ||
368 !(dev_cap
->flags2
& MLX4_DEV_CAP_FLAG2_CQE_STRIDE
))
371 /* Must have 64B CQE_EQE enabled by FW to use bigger stride
372 * When FW has NCSI it may decide not to report 64B CQE/EQEs
374 if (!(dev_cap
->flags
& MLX4_DEV_CAP_FLAG_64B_EQE
) ||
375 !(dev_cap
->flags
& MLX4_DEV_CAP_FLAG_64B_CQE
)) {
376 dev_cap
->flags2
&= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE
;
377 dev_cap
->flags2
&= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE
;
381 if (cache_line_size() == 128 || cache_line_size() == 256) {
382 mlx4_dbg(dev
, "Enabling CQE stride cacheLine supported\n");
383 /* Changing the real data inside CQE size to 32B */
384 dev_cap
->flags
&= ~MLX4_DEV_CAP_FLAG_64B_CQE
;
385 dev_cap
->flags
&= ~MLX4_DEV_CAP_FLAG_64B_EQE
;
387 if (mlx4_is_master(dev
))
388 dev_cap
->function_caps
|= MLX4_FUNC_CAP_EQE_CQE_STRIDE
;
390 if (cache_line_size() != 32 && cache_line_size() != 64)
391 mlx4_dbg(dev
, "Disabling CQE stride, cacheLine size unsupported\n");
392 dev_cap
->flags2
&= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE
;
393 dev_cap
->flags2
&= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE
;
397 static int _mlx4_dev_port(struct mlx4_dev
*dev
, int port
,
398 struct mlx4_port_cap
*port_cap
)
400 dev
->caps
.vl_cap
[port
] = port_cap
->max_vl
;
401 dev
->caps
.ib_mtu_cap
[port
] = port_cap
->ib_mtu
;
402 dev
->phys_caps
.gid_phys_table_len
[port
] = port_cap
->max_gids
;
403 dev
->phys_caps
.pkey_phys_table_len
[port
] = port_cap
->max_pkeys
;
404 /* set gid and pkey table operating lengths by default
405 * to non-sriov values
407 dev
->caps
.gid_table_len
[port
] = port_cap
->max_gids
;
408 dev
->caps
.pkey_table_len
[port
] = port_cap
->max_pkeys
;
409 dev
->caps
.port_width_cap
[port
] = port_cap
->max_port_width
;
410 dev
->caps
.eth_mtu_cap
[port
] = port_cap
->eth_mtu
;
411 dev
->caps
.max_tc_eth
= port_cap
->max_tc_eth
;
412 dev
->caps
.def_mac
[port
] = port_cap
->def_mac
;
413 dev
->caps
.supported_type
[port
] = port_cap
->supported_port_types
;
414 dev
->caps
.suggested_type
[port
] = port_cap
->suggested_type
;
415 dev
->caps
.default_sense
[port
] = port_cap
->default_sense
;
416 dev
->caps
.trans_type
[port
] = port_cap
->trans_type
;
417 dev
->caps
.vendor_oui
[port
] = port_cap
->vendor_oui
;
418 dev
->caps
.wavelength
[port
] = port_cap
->wavelength
;
419 dev
->caps
.trans_code
[port
] = port_cap
->trans_code
;
424 static int mlx4_dev_port(struct mlx4_dev
*dev
, int port
,
425 struct mlx4_port_cap
*port_cap
)
429 err
= mlx4_QUERY_PORT(dev
, port
, port_cap
);
432 mlx4_err(dev
, "QUERY_PORT command failed.\n");
437 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev
*dev
)
439 if (!(dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_IGNORE_FCS
))
442 if (mlx4_is_mfunc(dev
)) {
443 mlx4_dbg(dev
, "SRIOV mode - Disabling Ignore FCS");
444 dev
->caps
.flags2
&= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS
;
448 if (!(dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_FCS_KEEP
)) {
450 "Keep FCS is not supported - Disabling Ignore FCS");
451 dev
->caps
.flags2
&= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS
;
456 #define MLX4_A0_STEERING_TABLE_SIZE 256
457 static int mlx4_dev_cap(struct mlx4_dev
*dev
, struct mlx4_dev_cap
*dev_cap
)
462 err
= mlx4_QUERY_DEV_CAP(dev
, dev_cap
);
464 mlx4_err(dev
, "QUERY_DEV_CAP command failed, aborting\n");
467 mlx4_dev_cap_dump(dev
, dev_cap
);
469 if (dev_cap
->min_page_sz
> PAGE_SIZE
) {
470 mlx4_err(dev
, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
471 dev_cap
->min_page_sz
, PAGE_SIZE
);
474 if (dev_cap
->num_ports
> MLX4_MAX_PORTS
) {
475 mlx4_err(dev
, "HCA has %d ports, but we only support %d, aborting\n",
476 dev_cap
->num_ports
, MLX4_MAX_PORTS
);
480 if (dev_cap
->uar_size
> pci_resource_len(dev
->persist
->pdev
, 2)) {
481 mlx4_err(dev
, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
484 pci_resource_len(dev
->persist
->pdev
, 2));
488 dev
->caps
.num_ports
= dev_cap
->num_ports
;
489 dev
->caps
.num_sys_eqs
= dev_cap
->num_sys_eqs
;
490 dev
->phys_caps
.num_phys_eqs
= dev_cap
->flags2
& MLX4_DEV_CAP_FLAG2_SYS_EQS
?
491 dev
->caps
.num_sys_eqs
:
493 for (i
= 1; i
<= dev
->caps
.num_ports
; ++i
) {
494 err
= _mlx4_dev_port(dev
, i
, dev_cap
->port_cap
+ i
);
496 mlx4_err(dev
, "QUERY_PORT command failed, aborting\n");
501 dev
->caps
.uar_page_size
= PAGE_SIZE
;
502 dev
->caps
.num_uars
= dev_cap
->uar_size
/ PAGE_SIZE
;
503 dev
->caps
.local_ca_ack_delay
= dev_cap
->local_ca_ack_delay
;
504 dev
->caps
.bf_reg_size
= dev_cap
->bf_reg_size
;
505 dev
->caps
.bf_regs_per_page
= dev_cap
->bf_regs_per_page
;
506 dev
->caps
.max_sq_sg
= dev_cap
->max_sq_sg
;
507 dev
->caps
.max_rq_sg
= dev_cap
->max_rq_sg
;
508 dev
->caps
.max_wqes
= dev_cap
->max_qp_sz
;
509 dev
->caps
.max_qp_init_rdma
= dev_cap
->max_requester_per_qp
;
510 dev
->caps
.max_srq_wqes
= dev_cap
->max_srq_sz
;
511 dev
->caps
.max_srq_sge
= dev_cap
->max_rq_sg
- 1;
512 dev
->caps
.reserved_srqs
= dev_cap
->reserved_srqs
;
513 dev
->caps
.max_sq_desc_sz
= dev_cap
->max_sq_desc_sz
;
514 dev
->caps
.max_rq_desc_sz
= dev_cap
->max_rq_desc_sz
;
516 * Subtract 1 from the limit because we need to allocate a
517 * spare CQE to enable resizing the CQ.
519 dev
->caps
.max_cqes
= dev_cap
->max_cq_sz
- 1;
520 dev
->caps
.reserved_cqs
= dev_cap
->reserved_cqs
;
521 dev
->caps
.reserved_eqs
= dev_cap
->reserved_eqs
;
522 dev
->caps
.reserved_mtts
= dev_cap
->reserved_mtts
;
523 dev
->caps
.reserved_mrws
= dev_cap
->reserved_mrws
;
525 dev
->caps
.reserved_pds
= dev_cap
->reserved_pds
;
526 dev
->caps
.reserved_xrcds
= (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_XRC
) ?
527 dev_cap
->reserved_xrcds
: 0;
528 dev
->caps
.max_xrcds
= (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_XRC
) ?
529 dev_cap
->max_xrcds
: 0;
530 dev
->caps
.mtt_entry_sz
= dev_cap
->mtt_entry_sz
;
532 dev
->caps
.max_msg_sz
= dev_cap
->max_msg_sz
;
533 dev
->caps
.page_size_cap
= ~(u32
) (dev_cap
->min_page_sz
- 1);
534 dev
->caps
.flags
= dev_cap
->flags
;
535 dev
->caps
.flags2
= dev_cap
->flags2
;
536 dev
->caps
.bmme_flags
= dev_cap
->bmme_flags
;
537 dev
->caps
.reserved_lkey
= dev_cap
->reserved_lkey
;
538 dev
->caps
.stat_rate_support
= dev_cap
->stat_rate_support
;
539 dev
->caps
.max_gso_sz
= dev_cap
->max_gso_sz
;
540 dev
->caps
.max_rss_tbl_sz
= dev_cap
->max_rss_tbl_sz
;
541 dev
->caps
.wol_port
[1] = dev_cap
->wol_port
[1];
542 dev
->caps
.wol_port
[2] = dev_cap
->wol_port
[2];
543 dev
->caps
.health_buffer_addrs
= dev_cap
->health_buffer_addrs
;
545 /* Save uar page shift */
546 if (!mlx4_is_slave(dev
)) {
547 /* Virtual PCI function needs to determine UAR page size from
548 * firmware. Only master PCI function can set the uar page size
550 if (enable_4k_uar
|| !dev
->persist
->num_vfs
)
551 dev
->uar_page_shift
= DEFAULT_UAR_PAGE_SHIFT
;
553 dev
->uar_page_shift
= PAGE_SHIFT
;
555 mlx4_set_num_reserved_uars(dev
, dev_cap
);
558 if (dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_PHV_EN
) {
559 struct mlx4_init_hca_param hca_param
;
561 memset(&hca_param
, 0, sizeof(hca_param
));
562 err
= mlx4_QUERY_HCA(dev
, &hca_param
);
563 /* Turn off PHV_EN flag in case phv_check_en is set.
564 * phv_check_en is a HW check that parse the packet and verify
565 * phv bit was reported correctly in the wqe. To allow QinQ
566 * PHV_EN flag should be set and phv_check_en must be cleared
567 * otherwise QinQ packets will be drop by the HW.
569 if (err
|| hca_param
.phv_check_en
)
570 dev
->caps
.flags2
&= ~MLX4_DEV_CAP_FLAG2_PHV_EN
;
573 /* Sense port always allowed on supported devices for ConnectX-1 and -2 */
574 if (mlx4_priv(dev
)->pci_dev_data
& MLX4_PCI_DEV_FORCE_SENSE_PORT
)
575 dev
->caps
.flags
|= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT
;
576 /* Don't do sense port on multifunction devices (for now at least) */
577 if (mlx4_is_mfunc(dev
))
578 dev
->caps
.flags
&= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT
;
580 if (mlx4_low_memory_profile()) {
581 dev
->caps
.log_num_macs
= MLX4_MIN_LOG_NUM_MAC
;
582 dev
->caps
.log_num_vlans
= MLX4_MIN_LOG_NUM_VLANS
;
584 dev
->caps
.log_num_macs
= log_num_mac
;
585 dev
->caps
.log_num_vlans
= MLX4_LOG_NUM_VLANS
;
588 for (i
= 1; i
<= dev
->caps
.num_ports
; ++i
) {
589 dev
->caps
.port_type
[i
] = MLX4_PORT_TYPE_NONE
;
590 if (dev
->caps
.supported_type
[i
]) {
591 /* if only ETH is supported - assign ETH */
592 if (dev
->caps
.supported_type
[i
] == MLX4_PORT_TYPE_ETH
)
593 dev
->caps
.port_type
[i
] = MLX4_PORT_TYPE_ETH
;
594 /* if only IB is supported, assign IB */
595 else if (dev
->caps
.supported_type
[i
] ==
597 dev
->caps
.port_type
[i
] = MLX4_PORT_TYPE_IB
;
599 /* if IB and ETH are supported, we set the port
600 * type according to user selection of port type;
601 * if user selected none, take the FW hint */
602 if (port_type_array
[i
- 1] == MLX4_PORT_TYPE_NONE
)
603 dev
->caps
.port_type
[i
] = dev
->caps
.suggested_type
[i
] ?
604 MLX4_PORT_TYPE_ETH
: MLX4_PORT_TYPE_IB
;
606 dev
->caps
.port_type
[i
] = port_type_array
[i
- 1];
610 * Link sensing is allowed on the port if 3 conditions are true:
611 * 1. Both protocols are supported on the port.
612 * 2. Different types are supported on the port
613 * 3. FW declared that it supports link sensing
615 mlx4_priv(dev
)->sense
.sense_allowed
[i
] =
616 ((dev
->caps
.supported_type
[i
] == MLX4_PORT_TYPE_AUTO
) &&
617 (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_DPDP
) &&
618 (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_SENSE_SUPPORT
));
621 * If "default_sense" bit is set, we move the port to "AUTO" mode
622 * and perform sense_port FW command to try and set the correct
623 * port type from beginning
625 if (mlx4_priv(dev
)->sense
.sense_allowed
[i
] && dev
->caps
.default_sense
[i
]) {
626 enum mlx4_port_type sensed_port
= MLX4_PORT_TYPE_NONE
;
627 dev
->caps
.possible_type
[i
] = MLX4_PORT_TYPE_AUTO
;
628 mlx4_SENSE_PORT(dev
, i
, &sensed_port
);
629 if (sensed_port
!= MLX4_PORT_TYPE_NONE
)
630 dev
->caps
.port_type
[i
] = sensed_port
;
632 dev
->caps
.possible_type
[i
] = dev
->caps
.port_type
[i
];
635 if (dev
->caps
.log_num_macs
> dev_cap
->port_cap
[i
].log_max_macs
) {
636 dev
->caps
.log_num_macs
= dev_cap
->port_cap
[i
].log_max_macs
;
637 mlx4_warn(dev
, "Requested number of MACs is too much for port %d, reducing to %d\n",
638 i
, 1 << dev
->caps
.log_num_macs
);
640 if (dev
->caps
.log_num_vlans
> dev_cap
->port_cap
[i
].log_max_vlans
) {
641 dev
->caps
.log_num_vlans
= dev_cap
->port_cap
[i
].log_max_vlans
;
642 mlx4_warn(dev
, "Requested number of VLANs is too much for port %d, reducing to %d\n",
643 i
, 1 << dev
->caps
.log_num_vlans
);
647 if (mlx4_is_master(dev
) && (dev
->caps
.num_ports
== 2) &&
648 (port_type_array
[0] == MLX4_PORT_TYPE_IB
) &&
649 (port_type_array
[1] == MLX4_PORT_TYPE_ETH
)) {
651 "Granular QoS per VF not supported with IB/Eth configuration\n");
652 dev
->caps
.flags2
&= ~MLX4_DEV_CAP_FLAG2_QOS_VPP
;
655 dev
->caps
.max_counters
= dev_cap
->max_counters
;
657 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
] = dev_cap
->reserved_qps
;
658 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_ETH_ADDR
] =
659 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FC_ADDR
] =
660 (1 << dev
->caps
.log_num_macs
) *
661 (1 << dev
->caps
.log_num_vlans
) *
663 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FC_EXCH
] = MLX4_NUM_FEXCH
;
665 if (dev_cap
->dmfs_high_rate_qpn_base
> 0 &&
666 dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_FS_EN
)
667 dev
->caps
.dmfs_high_rate_qpn_base
= dev_cap
->dmfs_high_rate_qpn_base
;
669 dev
->caps
.dmfs_high_rate_qpn_base
=
670 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
];
672 if (dev_cap
->dmfs_high_rate_qpn_range
> 0 &&
673 dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_FS_EN
) {
674 dev
->caps
.dmfs_high_rate_qpn_range
= dev_cap
->dmfs_high_rate_qpn_range
;
675 dev
->caps
.dmfs_high_steer_mode
= MLX4_STEERING_DMFS_A0_DEFAULT
;
676 dev
->caps
.flags2
|= MLX4_DEV_CAP_FLAG2_FS_A0
;
678 dev
->caps
.dmfs_high_steer_mode
= MLX4_STEERING_DMFS_A0_NOT_SUPPORTED
;
679 dev
->caps
.dmfs_high_rate_qpn_base
=
680 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
];
681 dev
->caps
.dmfs_high_rate_qpn_range
= MLX4_A0_STEERING_TABLE_SIZE
;
684 dev
->caps
.rl_caps
= dev_cap
->rl_caps
;
686 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_RSS_RAW_ETH
] =
687 dev
->caps
.dmfs_high_rate_qpn_range
;
689 dev
->caps
.reserved_qps
= dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
] +
690 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_ETH_ADDR
] +
691 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FC_ADDR
] +
692 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FC_EXCH
];
694 dev
->caps
.sqp_demux
= (mlx4_is_master(dev
)) ? MLX4_MAX_NUM_SLAVES
: 0;
696 if (!enable_64b_cqe_eqe
&& !mlx4_is_slave(dev
)) {
698 (MLX4_DEV_CAP_FLAG_64B_CQE
| MLX4_DEV_CAP_FLAG_64B_EQE
)) {
699 mlx4_warn(dev
, "64B EQEs/CQEs supported by the device but not enabled\n");
700 dev
->caps
.flags
&= ~MLX4_DEV_CAP_FLAG_64B_CQE
;
701 dev
->caps
.flags
&= ~MLX4_DEV_CAP_FLAG_64B_EQE
;
704 if (dev_cap
->flags2
&
705 (MLX4_DEV_CAP_FLAG2_CQE_STRIDE
|
706 MLX4_DEV_CAP_FLAG2_EQE_STRIDE
)) {
707 mlx4_warn(dev
, "Disabling EQE/CQE stride per user request\n");
708 dev_cap
->flags2
&= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE
;
709 dev_cap
->flags2
&= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE
;
713 if ((dev
->caps
.flags
&
714 (MLX4_DEV_CAP_FLAG_64B_CQE
| MLX4_DEV_CAP_FLAG_64B_EQE
)) &&
716 dev
->caps
.function_caps
|= MLX4_FUNC_CAP_64B_EQE_CQE
;
718 if (!mlx4_is_slave(dev
)) {
719 mlx4_enable_cqe_eqe_stride(dev
);
720 dev
->caps
.alloc_res_qp_mask
=
721 (dev
->caps
.bf_reg_size
? MLX4_RESERVE_ETH_BF_QP
: 0) |
724 if (!(dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_ETS_CFG
) &&
725 dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_SET_ETH_SCHED
) {
726 mlx4_warn(dev
, "Old device ETS support detected\n");
727 mlx4_warn(dev
, "Consider upgrading device FW.\n");
728 dev
->caps
.flags2
|= MLX4_DEV_CAP_FLAG2_ETS_CFG
;
732 dev
->caps
.alloc_res_qp_mask
= 0;
735 mlx4_enable_ignore_fcs(dev
);
740 /*The function checks if there are live vf, return the num of them*/
741 static int mlx4_how_many_lives_vf(struct mlx4_dev
*dev
)
743 struct mlx4_priv
*priv
= mlx4_priv(dev
);
744 struct mlx4_slave_state
*s_state
;
748 for (i
= 1/*the ppf is 0*/; i
< dev
->num_slaves
; ++i
) {
749 s_state
= &priv
->mfunc
.master
.slave_state
[i
];
750 if (s_state
->active
&& s_state
->last_cmd
!=
751 MLX4_COMM_CMD_RESET
) {
752 mlx4_warn(dev
, "%s: slave: %d is still active\n",
760 int mlx4_get_parav_qkey(struct mlx4_dev
*dev
, u32 qpn
, u32
*qkey
)
762 u32 qk
= MLX4_RESERVED_QKEY_BASE
;
764 if (qpn
>= dev
->phys_caps
.base_tunnel_sqpn
+ 8 * MLX4_MFUNC_MAX
||
765 qpn
< dev
->phys_caps
.base_proxy_sqpn
)
768 if (qpn
>= dev
->phys_caps
.base_tunnel_sqpn
)
770 qk
+= qpn
- dev
->phys_caps
.base_tunnel_sqpn
;
772 qk
+= qpn
- dev
->phys_caps
.base_proxy_sqpn
;
776 EXPORT_SYMBOL(mlx4_get_parav_qkey
);
778 void mlx4_sync_pkey_table(struct mlx4_dev
*dev
, int slave
, int port
, int i
, int val
)
780 struct mlx4_priv
*priv
= container_of(dev
, struct mlx4_priv
, dev
);
782 if (!mlx4_is_master(dev
))
785 priv
->virt2phys_pkey
[slave
][port
- 1][i
] = val
;
787 EXPORT_SYMBOL(mlx4_sync_pkey_table
);
789 void mlx4_put_slave_node_guid(struct mlx4_dev
*dev
, int slave
, __be64 guid
)
791 struct mlx4_priv
*priv
= container_of(dev
, struct mlx4_priv
, dev
);
793 if (!mlx4_is_master(dev
))
796 priv
->slave_node_guids
[slave
] = guid
;
798 EXPORT_SYMBOL(mlx4_put_slave_node_guid
);
800 __be64
mlx4_get_slave_node_guid(struct mlx4_dev
*dev
, int slave
)
802 struct mlx4_priv
*priv
= container_of(dev
, struct mlx4_priv
, dev
);
804 if (!mlx4_is_master(dev
))
807 return priv
->slave_node_guids
[slave
];
809 EXPORT_SYMBOL(mlx4_get_slave_node_guid
);
811 int mlx4_is_slave_active(struct mlx4_dev
*dev
, int slave
)
813 struct mlx4_priv
*priv
= mlx4_priv(dev
);
814 struct mlx4_slave_state
*s_slave
;
816 if (!mlx4_is_master(dev
))
819 s_slave
= &priv
->mfunc
.master
.slave_state
[slave
];
820 return !!s_slave
->active
;
822 EXPORT_SYMBOL(mlx4_is_slave_active
);
824 void mlx4_handle_eth_header_mcast_prio(struct mlx4_net_trans_rule_hw_ctrl
*ctrl
,
825 struct _rule_hw
*eth_header
)
827 if (is_multicast_ether_addr(eth_header
->eth
.dst_mac
) ||
828 is_broadcast_ether_addr(eth_header
->eth
.dst_mac
)) {
829 struct mlx4_net_trans_rule_hw_eth
*eth
=
830 (struct mlx4_net_trans_rule_hw_eth
*)eth_header
;
831 struct _rule_hw
*next_rule
= (struct _rule_hw
*)(eth
+ 1);
832 bool last_rule
= next_rule
->size
== 0 && next_rule
->id
== 0 &&
833 next_rule
->rsvd
== 0;
836 ctrl
->prio
= cpu_to_be16(MLX4_DOMAIN_NIC
);
839 EXPORT_SYMBOL(mlx4_handle_eth_header_mcast_prio
);
841 static void slave_adjust_steering_mode(struct mlx4_dev
*dev
,
842 struct mlx4_dev_cap
*dev_cap
,
843 struct mlx4_init_hca_param
*hca_param
)
845 dev
->caps
.steering_mode
= hca_param
->steering_mode
;
846 if (dev
->caps
.steering_mode
== MLX4_STEERING_MODE_DEVICE_MANAGED
) {
847 dev
->caps
.num_qp_per_mgm
= dev_cap
->fs_max_num_qp_per_entry
;
848 dev
->caps
.fs_log_max_ucast_qp_range_size
=
849 dev_cap
->fs_log_max_ucast_qp_range_size
;
851 dev
->caps
.num_qp_per_mgm
=
852 4 * ((1 << hca_param
->log_mc_entry_sz
)/16 - 2);
854 mlx4_dbg(dev
, "Steering mode is: %s\n",
855 mlx4_steering_mode_str(dev
->caps
.steering_mode
));
858 static void mlx4_slave_destroy_special_qp_cap(struct mlx4_dev
*dev
)
860 kfree(dev
->caps
.spec_qps
);
861 dev
->caps
.spec_qps
= NULL
;
864 static int mlx4_slave_special_qp_cap(struct mlx4_dev
*dev
)
866 struct mlx4_func_cap
*func_cap
= NULL
;
867 struct mlx4_caps
*caps
= &dev
->caps
;
870 func_cap
= kzalloc(sizeof(*func_cap
), GFP_KERNEL
);
871 caps
->spec_qps
= kcalloc(caps
->num_ports
, sizeof(*caps
->spec_qps
), GFP_KERNEL
);
873 if (!func_cap
|| !caps
->spec_qps
) {
874 mlx4_err(dev
, "Failed to allocate memory for special qps cap\n");
879 for (i
= 1; i
<= caps
->num_ports
; ++i
) {
880 err
= mlx4_QUERY_FUNC_CAP(dev
, i
, func_cap
);
882 mlx4_err(dev
, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
886 caps
->spec_qps
[i
- 1] = func_cap
->spec_qps
;
887 caps
->port_mask
[i
] = caps
->port_type
[i
];
888 caps
->phys_port_id
[i
] = func_cap
->phys_port_id
;
889 err
= mlx4_get_slave_pkey_gid_tbl_len(dev
, i
,
890 &caps
->gid_table_len
[i
],
891 &caps
->pkey_table_len
[i
]);
893 mlx4_err(dev
, "QUERY_PORT command failed for port %d, aborting (%d)\n",
901 mlx4_slave_destroy_special_qp_cap(dev
);
906 static int mlx4_slave_cap(struct mlx4_dev
*dev
)
910 struct mlx4_dev_cap
*dev_cap
= NULL
;
911 struct mlx4_func_cap
*func_cap
= NULL
;
912 struct mlx4_init_hca_param
*hca_param
= NULL
;
914 hca_param
= kzalloc(sizeof(*hca_param
), GFP_KERNEL
);
915 func_cap
= kzalloc(sizeof(*func_cap
), GFP_KERNEL
);
916 dev_cap
= kzalloc(sizeof(*dev_cap
), GFP_KERNEL
);
917 if (!hca_param
|| !func_cap
|| !dev_cap
) {
918 mlx4_err(dev
, "Failed to allocate memory for slave_cap\n");
923 err
= mlx4_QUERY_HCA(dev
, hca_param
);
925 mlx4_err(dev
, "QUERY_HCA command failed, aborting\n");
929 /* fail if the hca has an unknown global capability
930 * at this time global_caps should be always zeroed
932 if (hca_param
->global_caps
) {
933 mlx4_err(dev
, "Unknown hca global capabilities\n");
938 dev
->caps
.hca_core_clock
= hca_param
->hca_core_clock
;
940 dev
->caps
.max_qp_dest_rdma
= 1 << hca_param
->log_rd_per_qp
;
941 err
= mlx4_dev_cap(dev
, dev_cap
);
943 mlx4_err(dev
, "QUERY_DEV_CAP command failed, aborting\n");
947 err
= mlx4_QUERY_FW(dev
);
949 mlx4_err(dev
, "QUERY_FW command failed: could not get FW version\n");
951 page_size
= ~dev
->caps
.page_size_cap
+ 1;
952 mlx4_warn(dev
, "HCA minimum page size:%d\n", page_size
);
953 if (page_size
> PAGE_SIZE
) {
954 mlx4_err(dev
, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
955 page_size
, PAGE_SIZE
);
960 /* Set uar_page_shift for VF */
961 dev
->uar_page_shift
= hca_param
->uar_page_sz
+ 12;
963 /* Make sure the master uar page size is valid */
964 if (dev
->uar_page_shift
> PAGE_SHIFT
) {
966 "Invalid configuration: uar page size is larger than system page size\n");
971 /* Set reserved_uars based on the uar_page_shift */
972 mlx4_set_num_reserved_uars(dev
, dev_cap
);
974 /* Although uar page size in FW differs from system page size,
975 * upper software layers (mlx4_ib, mlx4_en and part of mlx4_core)
976 * still works with assumption that uar page size == system page size
978 dev
->caps
.uar_page_size
= PAGE_SIZE
;
980 err
= mlx4_QUERY_FUNC_CAP(dev
, 0, func_cap
);
982 mlx4_err(dev
, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
987 if ((func_cap
->pf_context_behaviour
| PF_CONTEXT_BEHAVIOUR_MASK
) !=
988 PF_CONTEXT_BEHAVIOUR_MASK
) {
989 mlx4_err(dev
, "Unknown pf context behaviour %x known flags %x\n",
990 func_cap
->pf_context_behaviour
,
991 PF_CONTEXT_BEHAVIOUR_MASK
);
996 dev
->caps
.num_ports
= func_cap
->num_ports
;
997 dev
->quotas
.qp
= func_cap
->qp_quota
;
998 dev
->quotas
.srq
= func_cap
->srq_quota
;
999 dev
->quotas
.cq
= func_cap
->cq_quota
;
1000 dev
->quotas
.mpt
= func_cap
->mpt_quota
;
1001 dev
->quotas
.mtt
= func_cap
->mtt_quota
;
1002 dev
->caps
.num_qps
= 1 << hca_param
->log_num_qps
;
1003 dev
->caps
.num_srqs
= 1 << hca_param
->log_num_srqs
;
1004 dev
->caps
.num_cqs
= 1 << hca_param
->log_num_cqs
;
1005 dev
->caps
.num_mpts
= 1 << hca_param
->log_mpt_sz
;
1006 dev
->caps
.num_eqs
= func_cap
->max_eq
;
1007 dev
->caps
.reserved_eqs
= func_cap
->reserved_eq
;
1008 dev
->caps
.reserved_lkey
= func_cap
->reserved_lkey
;
1009 dev
->caps
.num_pds
= MLX4_NUM_PDS
;
1010 dev
->caps
.num_mgms
= 0;
1011 dev
->caps
.num_amgms
= 0;
1013 if (dev
->caps
.num_ports
> MLX4_MAX_PORTS
) {
1014 mlx4_err(dev
, "HCA has %d ports, but we only support %d, aborting\n",
1015 dev
->caps
.num_ports
, MLX4_MAX_PORTS
);
1020 mlx4_replace_zero_macs(dev
);
1022 err
= mlx4_slave_special_qp_cap(dev
);
1024 mlx4_err(dev
, "Set special QP caps failed. aborting\n");
1028 if (dev
->caps
.uar_page_size
* (dev
->caps
.num_uars
-
1029 dev
->caps
.reserved_uars
) >
1030 pci_resource_len(dev
->persist
->pdev
,
1032 mlx4_err(dev
, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
1033 dev
->caps
.uar_page_size
* dev
->caps
.num_uars
,
1034 (unsigned long long)
1035 pci_resource_len(dev
->persist
->pdev
, 2));
1040 if (hca_param
->dev_cap_enabled
& MLX4_DEV_CAP_64B_EQE_ENABLED
) {
1041 dev
->caps
.eqe_size
= 64;
1042 dev
->caps
.eqe_factor
= 1;
1044 dev
->caps
.eqe_size
= 32;
1045 dev
->caps
.eqe_factor
= 0;
1048 if (hca_param
->dev_cap_enabled
& MLX4_DEV_CAP_64B_CQE_ENABLED
) {
1049 dev
->caps
.cqe_size
= 64;
1050 dev
->caps
.userspace_caps
|= MLX4_USER_DEV_CAP_LARGE_CQE
;
1052 dev
->caps
.cqe_size
= 32;
1055 if (hca_param
->dev_cap_enabled
& MLX4_DEV_CAP_EQE_STRIDE_ENABLED
) {
1056 dev
->caps
.eqe_size
= hca_param
->eqe_size
;
1057 dev
->caps
.eqe_factor
= 0;
1060 if (hca_param
->dev_cap_enabled
& MLX4_DEV_CAP_CQE_STRIDE_ENABLED
) {
1061 dev
->caps
.cqe_size
= hca_param
->cqe_size
;
1062 /* User still need to know when CQE > 32B */
1063 dev
->caps
.userspace_caps
|= MLX4_USER_DEV_CAP_LARGE_CQE
;
1066 dev
->caps
.flags2
&= ~MLX4_DEV_CAP_FLAG2_TS
;
1067 mlx4_warn(dev
, "Timestamping is not supported in slave mode\n");
1069 dev
->caps
.flags2
&= ~MLX4_DEV_CAP_FLAG2_USER_MAC_EN
;
1070 mlx4_dbg(dev
, "User MAC FW update is not supported in slave mode\n");
1072 slave_adjust_steering_mode(dev
, dev_cap
, hca_param
);
1073 mlx4_dbg(dev
, "RSS support for IP fragments is %s\n",
1074 hca_param
->rss_ip_frags
? "on" : "off");
1076 if (func_cap
->extra_flags
& MLX4_QUERY_FUNC_FLAGS_BF_RES_QP
&&
1077 dev
->caps
.bf_reg_size
)
1078 dev
->caps
.alloc_res_qp_mask
|= MLX4_RESERVE_ETH_BF_QP
;
1080 if (func_cap
->extra_flags
& MLX4_QUERY_FUNC_FLAGS_A0_RES_QP
)
1081 dev
->caps
.alloc_res_qp_mask
|= MLX4_RESERVE_A0_QP
;
1085 mlx4_slave_destroy_special_qp_cap(dev
);
1093 static void mlx4_request_modules(struct mlx4_dev
*dev
)
1096 int has_ib_port
= false;
1097 int has_eth_port
= false;
1098 #define EN_DRV_NAME "mlx4_en"
1099 #define IB_DRV_NAME "mlx4_ib"
1101 for (port
= 1; port
<= dev
->caps
.num_ports
; port
++) {
1102 if (dev
->caps
.port_type
[port
] == MLX4_PORT_TYPE_IB
)
1104 else if (dev
->caps
.port_type
[port
] == MLX4_PORT_TYPE_ETH
)
1105 has_eth_port
= true;
1109 request_module_nowait(EN_DRV_NAME
);
1110 if (has_ib_port
|| (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_IBOE
))
1111 request_module_nowait(IB_DRV_NAME
);
1115 * Change the port configuration of the device.
1116 * Every user of this function must hold the port mutex.
1118 int mlx4_change_port_types(struct mlx4_dev
*dev
,
1119 enum mlx4_port_type
*port_types
)
1125 for (port
= 0; port
< dev
->caps
.num_ports
; port
++) {
1126 /* Change the port type only if the new type is different
1127 * from the current, and not set to Auto */
1128 if (port_types
[port
] != dev
->caps
.port_type
[port
+ 1])
1132 mlx4_unregister_device(dev
);
1133 for (port
= 1; port
<= dev
->caps
.num_ports
; port
++) {
1134 mlx4_CLOSE_PORT(dev
, port
);
1135 dev
->caps
.port_type
[port
] = port_types
[port
- 1];
1136 err
= mlx4_SET_PORT(dev
, port
, -1);
1138 mlx4_err(dev
, "Failed to set port %d, aborting\n",
1143 mlx4_set_port_mask(dev
);
1144 err
= mlx4_register_device(dev
);
1146 mlx4_err(dev
, "Failed to register device\n");
1149 mlx4_request_modules(dev
);
1156 static ssize_t
show_port_type(struct device
*dev
,
1157 struct device_attribute
*attr
,
1160 struct mlx4_port_info
*info
= container_of(attr
, struct mlx4_port_info
,
1162 struct mlx4_dev
*mdev
= info
->dev
;
1166 (mdev
->caps
.port_type
[info
->port
] == MLX4_PORT_TYPE_IB
) ?
1168 if (mdev
->caps
.possible_type
[info
->port
] == MLX4_PORT_TYPE_AUTO
)
1169 sprintf(buf
, "auto (%s)\n", type
);
1171 sprintf(buf
, "%s\n", type
);
1176 static int __set_port_type(struct mlx4_port_info
*info
,
1177 enum mlx4_port_type port_type
)
1179 struct mlx4_dev
*mdev
= info
->dev
;
1180 struct mlx4_priv
*priv
= mlx4_priv(mdev
);
1181 enum mlx4_port_type types
[MLX4_MAX_PORTS
];
1182 enum mlx4_port_type new_types
[MLX4_MAX_PORTS
];
1186 if ((port_type
& mdev
->caps
.supported_type
[info
->port
]) != port_type
) {
1188 "Requested port type for port %d is not supported on this HCA\n",
1193 mlx4_stop_sense(mdev
);
1194 mutex_lock(&priv
->port_mutex
);
1195 info
->tmp_type
= port_type
;
1197 /* Possible type is always the one that was delivered */
1198 mdev
->caps
.possible_type
[info
->port
] = info
->tmp_type
;
1200 for (i
= 0; i
< mdev
->caps
.num_ports
; i
++) {
1201 types
[i
] = priv
->port
[i
+1].tmp_type
? priv
->port
[i
+1].tmp_type
:
1202 mdev
->caps
.possible_type
[i
+1];
1203 if (types
[i
] == MLX4_PORT_TYPE_AUTO
)
1204 types
[i
] = mdev
->caps
.port_type
[i
+1];
1207 if (!(mdev
->caps
.flags
& MLX4_DEV_CAP_FLAG_DPDP
) &&
1208 !(mdev
->caps
.flags
& MLX4_DEV_CAP_FLAG_SENSE_SUPPORT
)) {
1209 for (i
= 1; i
<= mdev
->caps
.num_ports
; i
++) {
1210 if (mdev
->caps
.possible_type
[i
] == MLX4_PORT_TYPE_AUTO
) {
1211 mdev
->caps
.possible_type
[i
] = mdev
->caps
.port_type
[i
];
1217 mlx4_err(mdev
, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1221 mlx4_do_sense_ports(mdev
, new_types
, types
);
1223 err
= mlx4_check_port_params(mdev
, new_types
);
1227 /* We are about to apply the changes after the configuration
1228 * was verified, no need to remember the temporary types
1230 for (i
= 0; i
< mdev
->caps
.num_ports
; i
++)
1231 priv
->port
[i
+ 1].tmp_type
= 0;
1233 err
= mlx4_change_port_types(mdev
, new_types
);
1236 mlx4_start_sense(mdev
);
1237 mutex_unlock(&priv
->port_mutex
);
1242 static ssize_t
set_port_type(struct device
*dev
,
1243 struct device_attribute
*attr
,
1244 const char *buf
, size_t count
)
1246 struct mlx4_port_info
*info
= container_of(attr
, struct mlx4_port_info
,
1248 struct mlx4_dev
*mdev
= info
->dev
;
1249 enum mlx4_port_type port_type
;
1250 static DEFINE_MUTEX(set_port_type_mutex
);
1253 mutex_lock(&set_port_type_mutex
);
1255 if (!strcmp(buf
, "ib\n")) {
1256 port_type
= MLX4_PORT_TYPE_IB
;
1257 } else if (!strcmp(buf
, "eth\n")) {
1258 port_type
= MLX4_PORT_TYPE_ETH
;
1259 } else if (!strcmp(buf
, "auto\n")) {
1260 port_type
= MLX4_PORT_TYPE_AUTO
;
1262 mlx4_err(mdev
, "%s is not supported port type\n", buf
);
1267 err
= __set_port_type(info
, port_type
);
1270 mutex_unlock(&set_port_type_mutex
);
1272 return err
? err
: count
;
1283 static inline int int_to_ibta_mtu(int mtu
)
1286 case 256: return IB_MTU_256
;
1287 case 512: return IB_MTU_512
;
1288 case 1024: return IB_MTU_1024
;
1289 case 2048: return IB_MTU_2048
;
1290 case 4096: return IB_MTU_4096
;
1295 static inline int ibta_mtu_to_int(enum ibta_mtu mtu
)
1298 case IB_MTU_256
: return 256;
1299 case IB_MTU_512
: return 512;
1300 case IB_MTU_1024
: return 1024;
1301 case IB_MTU_2048
: return 2048;
1302 case IB_MTU_4096
: return 4096;
1307 static ssize_t
show_port_ib_mtu(struct device
*dev
,
1308 struct device_attribute
*attr
,
1311 struct mlx4_port_info
*info
= container_of(attr
, struct mlx4_port_info
,
1313 struct mlx4_dev
*mdev
= info
->dev
;
1315 if (mdev
->caps
.port_type
[info
->port
] == MLX4_PORT_TYPE_ETH
)
1316 mlx4_warn(mdev
, "port level mtu is only used for IB ports\n");
1318 sprintf(buf
, "%d\n",
1319 ibta_mtu_to_int(mdev
->caps
.port_ib_mtu
[info
->port
]));
1323 static ssize_t
set_port_ib_mtu(struct device
*dev
,
1324 struct device_attribute
*attr
,
1325 const char *buf
, size_t count
)
1327 struct mlx4_port_info
*info
= container_of(attr
, struct mlx4_port_info
,
1329 struct mlx4_dev
*mdev
= info
->dev
;
1330 struct mlx4_priv
*priv
= mlx4_priv(mdev
);
1331 int err
, port
, mtu
, ibta_mtu
= -1;
1333 if (mdev
->caps
.port_type
[info
->port
] == MLX4_PORT_TYPE_ETH
) {
1334 mlx4_warn(mdev
, "port level mtu is only used for IB ports\n");
1338 err
= kstrtoint(buf
, 0, &mtu
);
1340 ibta_mtu
= int_to_ibta_mtu(mtu
);
1342 if (err
|| ibta_mtu
< 0) {
1343 mlx4_err(mdev
, "%s is invalid IBTA mtu\n", buf
);
1347 mdev
->caps
.port_ib_mtu
[info
->port
] = ibta_mtu
;
1349 mlx4_stop_sense(mdev
);
1350 mutex_lock(&priv
->port_mutex
);
1351 mlx4_unregister_device(mdev
);
1352 for (port
= 1; port
<= mdev
->caps
.num_ports
; port
++) {
1353 mlx4_CLOSE_PORT(mdev
, port
);
1354 err
= mlx4_SET_PORT(mdev
, port
, -1);
1356 mlx4_err(mdev
, "Failed to set port %d, aborting\n",
1361 err
= mlx4_register_device(mdev
);
1363 mutex_unlock(&priv
->port_mutex
);
1364 mlx4_start_sense(mdev
);
1365 return err
? err
: count
;
1368 /* bond for multi-function device */
1369 #define MAX_MF_BOND_ALLOWED_SLAVES 63
1370 static int mlx4_mf_bond(struct mlx4_dev
*dev
)
1374 struct mlx4_slaves_pport slaves_port1
;
1375 struct mlx4_slaves_pport slaves_port2
;
1376 DECLARE_BITMAP(slaves_port_1_2
, MLX4_MFUNC_MAX
);
1378 slaves_port1
= mlx4_phys_to_slaves_pport(dev
, 1);
1379 slaves_port2
= mlx4_phys_to_slaves_pport(dev
, 2);
1380 bitmap_and(slaves_port_1_2
,
1381 slaves_port1
.slaves
, slaves_port2
.slaves
,
1382 dev
->persist
->num_vfs
+ 1);
1384 /* only single port vfs are allowed */
1385 if (bitmap_weight(slaves_port_1_2
, dev
->persist
->num_vfs
+ 1) > 1) {
1386 mlx4_warn(dev
, "HA mode unsupported for dual ported VFs\n");
1390 /* number of virtual functions is number of total functions minus one
1391 * physical function for each port.
1393 nvfs
= bitmap_weight(slaves_port1
.slaves
, dev
->persist
->num_vfs
+ 1) +
1394 bitmap_weight(slaves_port2
.slaves
, dev
->persist
->num_vfs
+ 1) - 2;
1396 /* limit on maximum allowed VFs */
1397 if (nvfs
> MAX_MF_BOND_ALLOWED_SLAVES
) {
1398 mlx4_warn(dev
, "HA mode is not supported for %d VFs (max %d are allowed)\n",
1399 nvfs
, MAX_MF_BOND_ALLOWED_SLAVES
);
1403 if (dev
->caps
.steering_mode
!= MLX4_STEERING_MODE_DEVICE_MANAGED
) {
1404 mlx4_warn(dev
, "HA mode unsupported for NON DMFS steering\n");
1408 err
= mlx4_bond_mac_table(dev
);
1411 err
= mlx4_bond_vlan_table(dev
);
1414 err
= mlx4_bond_fs_rules(dev
);
1420 (void)mlx4_unbond_vlan_table(dev
);
1422 (void)mlx4_unbond_mac_table(dev
);
1426 static int mlx4_mf_unbond(struct mlx4_dev
*dev
)
1430 ret
= mlx4_unbond_fs_rules(dev
);
1432 mlx4_warn(dev
, "multifunction unbond for flow rules failed (%d)\n", ret
);
1433 ret1
= mlx4_unbond_mac_table(dev
);
1435 mlx4_warn(dev
, "multifunction unbond for MAC table failed (%d)\n", ret1
);
1438 ret1
= mlx4_unbond_vlan_table(dev
);
1440 mlx4_warn(dev
, "multifunction unbond for VLAN table failed (%d)\n", ret1
);
1446 int mlx4_bond(struct mlx4_dev
*dev
)
1449 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1451 mutex_lock(&priv
->bond_mutex
);
1453 if (!mlx4_is_bonded(dev
)) {
1454 ret
= mlx4_do_bond(dev
, true);
1456 mlx4_err(dev
, "Failed to bond device: %d\n", ret
);
1457 if (!ret
&& mlx4_is_master(dev
)) {
1458 ret
= mlx4_mf_bond(dev
);
1460 mlx4_err(dev
, "bond for multifunction failed\n");
1461 mlx4_do_bond(dev
, false);
1466 mutex_unlock(&priv
->bond_mutex
);
1468 mlx4_dbg(dev
, "Device is bonded\n");
1472 EXPORT_SYMBOL_GPL(mlx4_bond
);
1474 int mlx4_unbond(struct mlx4_dev
*dev
)
1477 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1479 mutex_lock(&priv
->bond_mutex
);
1481 if (mlx4_is_bonded(dev
)) {
1484 ret
= mlx4_do_bond(dev
, false);
1486 mlx4_err(dev
, "Failed to unbond device: %d\n", ret
);
1487 if (mlx4_is_master(dev
))
1488 ret2
= mlx4_mf_unbond(dev
);
1490 mlx4_warn(dev
, "Failed to unbond device for multifunction (%d)\n", ret2
);
1495 mutex_unlock(&priv
->bond_mutex
);
1497 mlx4_dbg(dev
, "Device is unbonded\n");
1501 EXPORT_SYMBOL_GPL(mlx4_unbond
);
1504 int mlx4_port_map_set(struct mlx4_dev
*dev
, struct mlx4_port_map
*v2p
)
1506 u8 port1
= v2p
->port1
;
1507 u8 port2
= v2p
->port2
;
1508 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1511 if (!(dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_PORT_REMAP
))
1514 mutex_lock(&priv
->bond_mutex
);
1516 /* zero means keep current mapping for this port */
1518 port1
= priv
->v2p
.port1
;
1520 port2
= priv
->v2p
.port2
;
1522 if ((port1
< 1) || (port1
> MLX4_MAX_PORTS
) ||
1523 (port2
< 1) || (port2
> MLX4_MAX_PORTS
) ||
1524 (port1
== 2 && port2
== 1)) {
1525 /* besides boundary checks cross mapping makes
1526 * no sense and therefore not allowed */
1528 } else if ((port1
== priv
->v2p
.port1
) &&
1529 (port2
== priv
->v2p
.port2
)) {
1532 err
= mlx4_virt2phy_port_map(dev
, port1
, port2
);
1534 mlx4_dbg(dev
, "port map changed: [%d][%d]\n",
1536 priv
->v2p
.port1
= port1
;
1537 priv
->v2p
.port2
= port2
;
1539 mlx4_err(dev
, "Failed to change port mape: %d\n", err
);
1543 mutex_unlock(&priv
->bond_mutex
);
1546 EXPORT_SYMBOL_GPL(mlx4_port_map_set
);
1548 static int mlx4_load_fw(struct mlx4_dev
*dev
)
1550 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1553 priv
->fw
.fw_icm
= mlx4_alloc_icm(dev
, priv
->fw
.fw_pages
,
1554 GFP_HIGHUSER
| __GFP_NOWARN
, 0);
1555 if (!priv
->fw
.fw_icm
) {
1556 mlx4_err(dev
, "Couldn't allocate FW area, aborting\n");
1560 err
= mlx4_MAP_FA(dev
, priv
->fw
.fw_icm
);
1562 mlx4_err(dev
, "MAP_FA command failed, aborting\n");
1566 err
= mlx4_RUN_FW(dev
);
1568 mlx4_err(dev
, "RUN_FW command failed, aborting\n");
1578 mlx4_free_icm(dev
, priv
->fw
.fw_icm
, 0);
1582 static int mlx4_init_cmpt_table(struct mlx4_dev
*dev
, u64 cmpt_base
,
1585 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1589 err
= mlx4_init_icm_table(dev
, &priv
->qp_table
.cmpt_table
,
1591 ((u64
) (MLX4_CMPT_TYPE_QP
*
1592 cmpt_entry_sz
) << MLX4_CMPT_SHIFT
),
1593 cmpt_entry_sz
, dev
->caps
.num_qps
,
1594 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
],
1599 err
= mlx4_init_icm_table(dev
, &priv
->srq_table
.cmpt_table
,
1601 ((u64
) (MLX4_CMPT_TYPE_SRQ
*
1602 cmpt_entry_sz
) << MLX4_CMPT_SHIFT
),
1603 cmpt_entry_sz
, dev
->caps
.num_srqs
,
1604 dev
->caps
.reserved_srqs
, 0, 0);
1608 err
= mlx4_init_icm_table(dev
, &priv
->cq_table
.cmpt_table
,
1610 ((u64
) (MLX4_CMPT_TYPE_CQ
*
1611 cmpt_entry_sz
) << MLX4_CMPT_SHIFT
),
1612 cmpt_entry_sz
, dev
->caps
.num_cqs
,
1613 dev
->caps
.reserved_cqs
, 0, 0);
1617 num_eqs
= dev
->phys_caps
.num_phys_eqs
;
1618 err
= mlx4_init_icm_table(dev
, &priv
->eq_table
.cmpt_table
,
1620 ((u64
) (MLX4_CMPT_TYPE_EQ
*
1621 cmpt_entry_sz
) << MLX4_CMPT_SHIFT
),
1622 cmpt_entry_sz
, num_eqs
, num_eqs
, 0, 0);
1629 mlx4_cleanup_icm_table(dev
, &priv
->cq_table
.cmpt_table
);
1632 mlx4_cleanup_icm_table(dev
, &priv
->srq_table
.cmpt_table
);
1635 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.cmpt_table
);
1641 static int mlx4_init_icm(struct mlx4_dev
*dev
, struct mlx4_dev_cap
*dev_cap
,
1642 struct mlx4_init_hca_param
*init_hca
, u64 icm_size
)
1644 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1649 err
= mlx4_SET_ICM_SIZE(dev
, icm_size
, &aux_pages
);
1651 mlx4_err(dev
, "SET_ICM_SIZE command failed, aborting\n");
1655 mlx4_dbg(dev
, "%lld KB of HCA context requires %lld KB aux memory\n",
1656 (unsigned long long) icm_size
>> 10,
1657 (unsigned long long) aux_pages
<< 2);
1659 priv
->fw
.aux_icm
= mlx4_alloc_icm(dev
, aux_pages
,
1660 GFP_HIGHUSER
| __GFP_NOWARN
, 0);
1661 if (!priv
->fw
.aux_icm
) {
1662 mlx4_err(dev
, "Couldn't allocate aux memory, aborting\n");
1666 err
= mlx4_MAP_ICM_AUX(dev
, priv
->fw
.aux_icm
);
1668 mlx4_err(dev
, "MAP_ICM_AUX command failed, aborting\n");
1672 err
= mlx4_init_cmpt_table(dev
, init_hca
->cmpt_base
, dev_cap
->cmpt_entry_sz
);
1674 mlx4_err(dev
, "Failed to map cMPT context memory, aborting\n");
1679 num_eqs
= dev
->phys_caps
.num_phys_eqs
;
1680 err
= mlx4_init_icm_table(dev
, &priv
->eq_table
.table
,
1681 init_hca
->eqc_base
, dev_cap
->eqc_entry_sz
,
1682 num_eqs
, num_eqs
, 0, 0);
1684 mlx4_err(dev
, "Failed to map EQ context memory, aborting\n");
1685 goto err_unmap_cmpt
;
1689 * Reserved MTT entries must be aligned up to a cacheline
1690 * boundary, since the FW will write to them, while the driver
1691 * writes to all other MTT entries. (The variable
1692 * dev->caps.mtt_entry_sz below is really the MTT segment
1693 * size, not the raw entry size)
1695 dev
->caps
.reserved_mtts
=
1696 ALIGN(dev
->caps
.reserved_mtts
* dev
->caps
.mtt_entry_sz
,
1697 dma_get_cache_alignment()) / dev
->caps
.mtt_entry_sz
;
1699 err
= mlx4_init_icm_table(dev
, &priv
->mr_table
.mtt_table
,
1701 dev
->caps
.mtt_entry_sz
,
1703 dev
->caps
.reserved_mtts
, 1, 0);
1705 mlx4_err(dev
, "Failed to map MTT context memory, aborting\n");
1709 err
= mlx4_init_icm_table(dev
, &priv
->mr_table
.dmpt_table
,
1710 init_hca
->dmpt_base
,
1711 dev_cap
->dmpt_entry_sz
,
1713 dev
->caps
.reserved_mrws
, 1, 1);
1715 mlx4_err(dev
, "Failed to map dMPT context memory, aborting\n");
1719 err
= mlx4_init_icm_table(dev
, &priv
->qp_table
.qp_table
,
1721 dev_cap
->qpc_entry_sz
,
1723 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
],
1726 mlx4_err(dev
, "Failed to map QP context memory, aborting\n");
1727 goto err_unmap_dmpt
;
1730 err
= mlx4_init_icm_table(dev
, &priv
->qp_table
.auxc_table
,
1731 init_hca
->auxc_base
,
1732 dev_cap
->aux_entry_sz
,
1734 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
],
1737 mlx4_err(dev
, "Failed to map AUXC context memory, aborting\n");
1741 err
= mlx4_init_icm_table(dev
, &priv
->qp_table
.altc_table
,
1742 init_hca
->altc_base
,
1743 dev_cap
->altc_entry_sz
,
1745 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
],
1748 mlx4_err(dev
, "Failed to map ALTC context memory, aborting\n");
1749 goto err_unmap_auxc
;
1752 err
= mlx4_init_icm_table(dev
, &priv
->qp_table
.rdmarc_table
,
1753 init_hca
->rdmarc_base
,
1754 dev_cap
->rdmarc_entry_sz
<< priv
->qp_table
.rdmarc_shift
,
1756 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
],
1759 mlx4_err(dev
, "Failed to map RDMARC context memory, aborting\n");
1760 goto err_unmap_altc
;
1763 err
= mlx4_init_icm_table(dev
, &priv
->cq_table
.table
,
1765 dev_cap
->cqc_entry_sz
,
1767 dev
->caps
.reserved_cqs
, 0, 0);
1769 mlx4_err(dev
, "Failed to map CQ context memory, aborting\n");
1770 goto err_unmap_rdmarc
;
1773 err
= mlx4_init_icm_table(dev
, &priv
->srq_table
.table
,
1774 init_hca
->srqc_base
,
1775 dev_cap
->srq_entry_sz
,
1777 dev
->caps
.reserved_srqs
, 0, 0);
1779 mlx4_err(dev
, "Failed to map SRQ context memory, aborting\n");
1784 * For flow steering device managed mode it is required to use
1785 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1786 * required, but for simplicity just map the whole multicast
1787 * group table now. The table isn't very big and it's a lot
1788 * easier than trying to track ref counts.
1790 err
= mlx4_init_icm_table(dev
, &priv
->mcg_table
.table
,
1792 mlx4_get_mgm_entry_size(dev
),
1793 dev
->caps
.num_mgms
+ dev
->caps
.num_amgms
,
1794 dev
->caps
.num_mgms
+ dev
->caps
.num_amgms
,
1797 mlx4_err(dev
, "Failed to map MCG context memory, aborting\n");
1804 mlx4_cleanup_icm_table(dev
, &priv
->srq_table
.table
);
1807 mlx4_cleanup_icm_table(dev
, &priv
->cq_table
.table
);
1810 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.rdmarc_table
);
1813 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.altc_table
);
1816 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.auxc_table
);
1819 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.qp_table
);
1822 mlx4_cleanup_icm_table(dev
, &priv
->mr_table
.dmpt_table
);
1825 mlx4_cleanup_icm_table(dev
, &priv
->mr_table
.mtt_table
);
1828 mlx4_cleanup_icm_table(dev
, &priv
->eq_table
.table
);
1831 mlx4_cleanup_icm_table(dev
, &priv
->eq_table
.cmpt_table
);
1832 mlx4_cleanup_icm_table(dev
, &priv
->cq_table
.cmpt_table
);
1833 mlx4_cleanup_icm_table(dev
, &priv
->srq_table
.cmpt_table
);
1834 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.cmpt_table
);
1837 mlx4_UNMAP_ICM_AUX(dev
);
1840 mlx4_free_icm(dev
, priv
->fw
.aux_icm
, 0);
1845 static void mlx4_free_icms(struct mlx4_dev
*dev
)
1847 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1849 mlx4_cleanup_icm_table(dev
, &priv
->mcg_table
.table
);
1850 mlx4_cleanup_icm_table(dev
, &priv
->srq_table
.table
);
1851 mlx4_cleanup_icm_table(dev
, &priv
->cq_table
.table
);
1852 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.rdmarc_table
);
1853 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.altc_table
);
1854 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.auxc_table
);
1855 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.qp_table
);
1856 mlx4_cleanup_icm_table(dev
, &priv
->mr_table
.dmpt_table
);
1857 mlx4_cleanup_icm_table(dev
, &priv
->mr_table
.mtt_table
);
1858 mlx4_cleanup_icm_table(dev
, &priv
->eq_table
.table
);
1859 mlx4_cleanup_icm_table(dev
, &priv
->eq_table
.cmpt_table
);
1860 mlx4_cleanup_icm_table(dev
, &priv
->cq_table
.cmpt_table
);
1861 mlx4_cleanup_icm_table(dev
, &priv
->srq_table
.cmpt_table
);
1862 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.cmpt_table
);
1864 mlx4_UNMAP_ICM_AUX(dev
);
1865 mlx4_free_icm(dev
, priv
->fw
.aux_icm
, 0);
1868 static void mlx4_slave_exit(struct mlx4_dev
*dev
)
1870 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1872 mutex_lock(&priv
->cmd
.slave_cmd_mutex
);
1873 if (mlx4_comm_cmd(dev
, MLX4_COMM_CMD_RESET
, 0, MLX4_COMM_CMD_NA_OP
,
1875 mlx4_warn(dev
, "Failed to close slave function\n");
1876 mutex_unlock(&priv
->cmd
.slave_cmd_mutex
);
1879 static int map_bf_area(struct mlx4_dev
*dev
)
1881 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1882 resource_size_t bf_start
;
1883 resource_size_t bf_len
;
1886 if (!dev
->caps
.bf_reg_size
)
1889 bf_start
= pci_resource_start(dev
->persist
->pdev
, 2) +
1890 (dev
->caps
.num_uars
<< PAGE_SHIFT
);
1891 bf_len
= pci_resource_len(dev
->persist
->pdev
, 2) -
1892 (dev
->caps
.num_uars
<< PAGE_SHIFT
);
1893 priv
->bf_mapping
= io_mapping_create_wc(bf_start
, bf_len
);
1894 if (!priv
->bf_mapping
)
1900 static void unmap_bf_area(struct mlx4_dev
*dev
)
1902 if (mlx4_priv(dev
)->bf_mapping
)
1903 io_mapping_free(mlx4_priv(dev
)->bf_mapping
);
1906 u64
mlx4_read_clock(struct mlx4_dev
*dev
)
1908 u32 clockhi
, clocklo
, clockhi1
;
1911 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1913 for (i
= 0; i
< 10; i
++) {
1914 clockhi
= swab32(readl(priv
->clock_mapping
));
1915 clocklo
= swab32(readl(priv
->clock_mapping
+ 4));
1916 clockhi1
= swab32(readl(priv
->clock_mapping
));
1917 if (clockhi
== clockhi1
)
1921 cycles
= (u64
) clockhi
<< 32 | (u64
) clocklo
;
1925 EXPORT_SYMBOL_GPL(mlx4_read_clock
);
1928 static int map_internal_clock(struct mlx4_dev
*dev
)
1930 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1932 priv
->clock_mapping
=
1933 ioremap(pci_resource_start(dev
->persist
->pdev
,
1934 priv
->fw
.clock_bar
) +
1935 priv
->fw
.clock_offset
, MLX4_CLOCK_SIZE
);
1937 if (!priv
->clock_mapping
)
1943 int mlx4_get_internal_clock_params(struct mlx4_dev
*dev
,
1944 struct mlx4_clock_params
*params
)
1946 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1948 if (mlx4_is_slave(dev
))
1954 params
->bar
= priv
->fw
.clock_bar
;
1955 params
->offset
= priv
->fw
.clock_offset
;
1956 params
->size
= MLX4_CLOCK_SIZE
;
1960 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params
);
1962 static void unmap_internal_clock(struct mlx4_dev
*dev
)
1964 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1966 if (priv
->clock_mapping
)
1967 iounmap(priv
->clock_mapping
);
1970 static void mlx4_close_hca(struct mlx4_dev
*dev
)
1972 unmap_internal_clock(dev
);
1974 if (mlx4_is_slave(dev
))
1975 mlx4_slave_exit(dev
);
1977 mlx4_CLOSE_HCA(dev
, 0);
1978 mlx4_free_icms(dev
);
1982 static void mlx4_close_fw(struct mlx4_dev
*dev
)
1984 if (!mlx4_is_slave(dev
)) {
1986 mlx4_free_icm(dev
, mlx4_priv(dev
)->fw
.fw_icm
, 0);
1990 static int mlx4_comm_check_offline(struct mlx4_dev
*dev
)
1992 #define COMM_CHAN_OFFLINE_OFFSET 0x09
1997 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1999 end
= msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT
) + jiffies
;
2000 while (time_before(jiffies
, end
)) {
2001 comm_flags
= swab32(readl((__iomem
char *)priv
->mfunc
.comm
+
2002 MLX4_COMM_CHAN_FLAGS
));
2003 offline_bit
= (comm_flags
&
2004 (u32
)(1 << COMM_CHAN_OFFLINE_OFFSET
));
2008 /* If device removal has been requested,
2009 * do not continue retrying.
2011 if (dev
->persist
->interface_state
&
2012 MLX4_INTERFACE_STATE_NOWAIT
)
2015 /* There are cases as part of AER/Reset flow that PF needs
2016 * around 100 msec to load. We therefore sleep for 100 msec
2017 * to allow other tasks to make use of that CPU during this
2022 mlx4_err(dev
, "Communication channel is offline.\n");
2026 static void mlx4_reset_vf_support(struct mlx4_dev
*dev
)
2028 #define COMM_CHAN_RST_OFFSET 0x1e
2030 struct mlx4_priv
*priv
= mlx4_priv(dev
);
2034 comm_caps
= swab32(readl((__iomem
char *)priv
->mfunc
.comm
+
2035 MLX4_COMM_CHAN_CAPS
));
2036 comm_rst
= (comm_caps
& (u32
)(1 << COMM_CHAN_RST_OFFSET
));
2039 dev
->caps
.vf_caps
|= MLX4_VF_CAP_FLAG_RESET
;
2042 static int mlx4_init_slave(struct mlx4_dev
*dev
)
2044 struct mlx4_priv
*priv
= mlx4_priv(dev
);
2045 u64 dma
= (u64
) priv
->mfunc
.vhcr_dma
;
2046 int ret_from_reset
= 0;
2048 u32 cmd_channel_ver
;
2050 if (atomic_read(&pf_loading
)) {
2051 mlx4_warn(dev
, "PF is not ready - Deferring probe\n");
2052 return -EPROBE_DEFER
;
2055 mutex_lock(&priv
->cmd
.slave_cmd_mutex
);
2056 priv
->cmd
.max_cmds
= 1;
2057 if (mlx4_comm_check_offline(dev
)) {
2058 mlx4_err(dev
, "PF is not responsive, skipping initialization\n");
2062 mlx4_reset_vf_support(dev
);
2063 mlx4_warn(dev
, "Sending reset\n");
2064 ret_from_reset
= mlx4_comm_cmd(dev
, MLX4_COMM_CMD_RESET
, 0,
2065 MLX4_COMM_CMD_NA_OP
, MLX4_COMM_TIME
);
2066 /* if we are in the middle of flr the slave will try
2067 * NUM_OF_RESET_RETRIES times before leaving.*/
2068 if (ret_from_reset
) {
2069 if (MLX4_DELAY_RESET_SLAVE
== ret_from_reset
) {
2070 mlx4_warn(dev
, "slave is currently in the middle of FLR - Deferring probe\n");
2071 mutex_unlock(&priv
->cmd
.slave_cmd_mutex
);
2072 return -EPROBE_DEFER
;
2077 /* check the driver version - the slave I/F revision
2078 * must match the master's */
2079 slave_read
= swab32(readl(&priv
->mfunc
.comm
->slave_read
));
2080 cmd_channel_ver
= mlx4_comm_get_version();
2082 if (MLX4_COMM_GET_IF_REV(cmd_channel_ver
) !=
2083 MLX4_COMM_GET_IF_REV(slave_read
)) {
2084 mlx4_err(dev
, "slave driver version is not supported by the master\n");
2088 mlx4_warn(dev
, "Sending vhcr0\n");
2089 if (mlx4_comm_cmd(dev
, MLX4_COMM_CMD_VHCR0
, dma
>> 48,
2090 MLX4_COMM_CMD_NA_OP
, MLX4_COMM_TIME
))
2092 if (mlx4_comm_cmd(dev
, MLX4_COMM_CMD_VHCR1
, dma
>> 32,
2093 MLX4_COMM_CMD_NA_OP
, MLX4_COMM_TIME
))
2095 if (mlx4_comm_cmd(dev
, MLX4_COMM_CMD_VHCR2
, dma
>> 16,
2096 MLX4_COMM_CMD_NA_OP
, MLX4_COMM_TIME
))
2098 if (mlx4_comm_cmd(dev
, MLX4_COMM_CMD_VHCR_EN
, dma
,
2099 MLX4_COMM_CMD_NA_OP
, MLX4_COMM_TIME
))
2102 mutex_unlock(&priv
->cmd
.slave_cmd_mutex
);
2106 mlx4_comm_cmd(dev
, MLX4_COMM_CMD_RESET
, 0, MLX4_COMM_CMD_NA_OP
, 0);
2108 mutex_unlock(&priv
->cmd
.slave_cmd_mutex
);
2112 static void mlx4_parav_master_pf_caps(struct mlx4_dev
*dev
)
2116 for (i
= 1; i
<= dev
->caps
.num_ports
; i
++) {
2117 if (dev
->caps
.port_type
[i
] == MLX4_PORT_TYPE_ETH
)
2118 dev
->caps
.gid_table_len
[i
] =
2119 mlx4_get_slave_num_gids(dev
, 0, i
);
2121 dev
->caps
.gid_table_len
[i
] = 1;
2122 dev
->caps
.pkey_table_len
[i
] =
2123 dev
->phys_caps
.pkey_phys_table_len
[i
] - 1;
2127 static int choose_log_fs_mgm_entry_size(int qp_per_entry
)
2129 int i
= MLX4_MIN_MGM_LOG_ENTRY_SIZE
;
2131 for (i
= MLX4_MIN_MGM_LOG_ENTRY_SIZE
; i
<= MLX4_MAX_MGM_LOG_ENTRY_SIZE
;
2133 if (qp_per_entry
<= 4 * ((1 << i
) / 16 - 2))
2137 return (i
<= MLX4_MAX_MGM_LOG_ENTRY_SIZE
) ? i
: -1;
2140 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode
)
2142 switch (dmfs_high_steer_mode
) {
2143 case MLX4_STEERING_DMFS_A0_DEFAULT
:
2144 return "default performance";
2146 case MLX4_STEERING_DMFS_A0_DYNAMIC
:
2147 return "dynamic hybrid mode";
2149 case MLX4_STEERING_DMFS_A0_STATIC
:
2150 return "performance optimized for limited rule configuration (static)";
2152 case MLX4_STEERING_DMFS_A0_DISABLE
:
2153 return "disabled performance optimized steering";
2155 case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED
:
2156 return "performance optimized steering not supported";
2159 return "Unrecognized mode";
2163 #define MLX4_DMFS_A0_STEERING (1UL << 2)
2165 static void choose_steering_mode(struct mlx4_dev
*dev
,
2166 struct mlx4_dev_cap
*dev_cap
)
2168 if (mlx4_log_num_mgm_entry_size
<= 0) {
2169 if ((-mlx4_log_num_mgm_entry_size
) & MLX4_DMFS_A0_STEERING
) {
2170 if (dev
->caps
.dmfs_high_steer_mode
==
2171 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED
)
2172 mlx4_err(dev
, "DMFS high rate mode not supported\n");
2174 dev
->caps
.dmfs_high_steer_mode
=
2175 MLX4_STEERING_DMFS_A0_STATIC
;
2179 if (mlx4_log_num_mgm_entry_size
<= 0 &&
2180 dev_cap
->flags2
& MLX4_DEV_CAP_FLAG2_FS_EN
&&
2181 (!mlx4_is_mfunc(dev
) ||
2182 (dev_cap
->fs_max_num_qp_per_entry
>=
2183 (dev
->persist
->num_vfs
+ 1))) &&
2184 choose_log_fs_mgm_entry_size(dev_cap
->fs_max_num_qp_per_entry
) >=
2185 MLX4_MIN_MGM_LOG_ENTRY_SIZE
) {
2186 dev
->oper_log_mgm_entry_size
=
2187 choose_log_fs_mgm_entry_size(dev_cap
->fs_max_num_qp_per_entry
);
2188 dev
->caps
.steering_mode
= MLX4_STEERING_MODE_DEVICE_MANAGED
;
2189 dev
->caps
.num_qp_per_mgm
= dev_cap
->fs_max_num_qp_per_entry
;
2190 dev
->caps
.fs_log_max_ucast_qp_range_size
=
2191 dev_cap
->fs_log_max_ucast_qp_range_size
;
2193 if (dev
->caps
.dmfs_high_steer_mode
!=
2194 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED
)
2195 dev
->caps
.dmfs_high_steer_mode
= MLX4_STEERING_DMFS_A0_DISABLE
;
2196 if (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_VEP_UC_STEER
&&
2197 dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_VEP_MC_STEER
)
2198 dev
->caps
.steering_mode
= MLX4_STEERING_MODE_B0
;
2200 dev
->caps
.steering_mode
= MLX4_STEERING_MODE_A0
;
2202 if (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_VEP_UC_STEER
||
2203 dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_VEP_MC_STEER
)
2204 mlx4_warn(dev
, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
2206 dev
->oper_log_mgm_entry_size
=
2207 mlx4_log_num_mgm_entry_size
> 0 ?
2208 mlx4_log_num_mgm_entry_size
:
2209 MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE
;
2210 dev
->caps
.num_qp_per_mgm
= mlx4_get_qp_per_mgm(dev
);
2212 mlx4_dbg(dev
, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
2213 mlx4_steering_mode_str(dev
->caps
.steering_mode
),
2214 dev
->oper_log_mgm_entry_size
,
2215 mlx4_log_num_mgm_entry_size
);
2218 static void choose_tunnel_offload_mode(struct mlx4_dev
*dev
,
2219 struct mlx4_dev_cap
*dev_cap
)
2221 if (dev
->caps
.steering_mode
== MLX4_STEERING_MODE_DEVICE_MANAGED
&&
2222 dev_cap
->flags2
& MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS
)
2223 dev
->caps
.tunnel_offload_mode
= MLX4_TUNNEL_OFFLOAD_MODE_VXLAN
;
2225 dev
->caps
.tunnel_offload_mode
= MLX4_TUNNEL_OFFLOAD_MODE_NONE
;
2227 mlx4_dbg(dev
, "Tunneling offload mode is: %s\n", (dev
->caps
.tunnel_offload_mode
2228 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN
) ? "vxlan" : "none");
2231 static int mlx4_validate_optimized_steering(struct mlx4_dev
*dev
)
2234 struct mlx4_port_cap port_cap
;
2236 if (dev
->caps
.dmfs_high_steer_mode
== MLX4_STEERING_DMFS_A0_NOT_SUPPORTED
)
2239 for (i
= 1; i
<= dev
->caps
.num_ports
; i
++) {
2240 if (mlx4_dev_port(dev
, i
, &port_cap
)) {
2242 "QUERY_DEV_CAP command failed, can't verify DMFS high rate steering.\n");
2243 } else if ((dev
->caps
.dmfs_high_steer_mode
!=
2244 MLX4_STEERING_DMFS_A0_DEFAULT
) &&
2245 (port_cap
.dmfs_optimized_state
==
2246 !!(dev
->caps
.dmfs_high_steer_mode
==
2247 MLX4_STEERING_DMFS_A0_DISABLE
))) {
2249 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2250 dmfs_high_rate_steering_mode_str(
2251 dev
->caps
.dmfs_high_steer_mode
),
2252 (port_cap
.dmfs_optimized_state
?
2253 "enabled" : "disabled"));
2260 static int mlx4_init_fw(struct mlx4_dev
*dev
)
2262 struct mlx4_mod_stat_cfg mlx4_cfg
;
2265 if (!mlx4_is_slave(dev
)) {
2266 err
= mlx4_QUERY_FW(dev
);
2269 mlx4_info(dev
, "non-primary physical function, skipping\n");
2271 mlx4_err(dev
, "QUERY_FW command failed, aborting\n");
2275 err
= mlx4_load_fw(dev
);
2277 mlx4_err(dev
, "Failed to start FW, aborting\n");
2281 mlx4_cfg
.log_pg_sz_m
= 1;
2282 mlx4_cfg
.log_pg_sz
= 0;
2283 err
= mlx4_MOD_STAT_CFG(dev
, &mlx4_cfg
);
2285 mlx4_warn(dev
, "Failed to override log_pg_sz parameter\n");
2291 static int mlx4_init_hca(struct mlx4_dev
*dev
)
2293 struct mlx4_priv
*priv
= mlx4_priv(dev
);
2294 struct mlx4_init_hca_param
*init_hca
= NULL
;
2295 struct mlx4_dev_cap
*dev_cap
= NULL
;
2296 struct mlx4_adapter adapter
;
2297 struct mlx4_profile profile
;
2299 struct mlx4_config_dev_params params
;
2302 if (!mlx4_is_slave(dev
)) {
2303 dev_cap
= kzalloc(sizeof(*dev_cap
), GFP_KERNEL
);
2304 init_hca
= kzalloc(sizeof(*init_hca
), GFP_KERNEL
);
2306 if (!dev_cap
|| !init_hca
) {
2311 err
= mlx4_dev_cap(dev
, dev_cap
);
2313 mlx4_err(dev
, "QUERY_DEV_CAP command failed, aborting\n");
2317 choose_steering_mode(dev
, dev_cap
);
2318 choose_tunnel_offload_mode(dev
, dev_cap
);
2320 if (dev
->caps
.dmfs_high_steer_mode
== MLX4_STEERING_DMFS_A0_STATIC
&&
2321 mlx4_is_master(dev
))
2322 dev
->caps
.function_caps
|= MLX4_FUNC_CAP_DMFS_A0_STATIC
;
2324 err
= mlx4_get_phys_port_id(dev
);
2326 mlx4_err(dev
, "Fail to get physical port id\n");
2328 if (mlx4_is_master(dev
))
2329 mlx4_parav_master_pf_caps(dev
);
2331 if (mlx4_low_memory_profile()) {
2332 mlx4_info(dev
, "Running from within kdump kernel. Using low memory profile\n");
2333 profile
= low_mem_profile
;
2335 profile
= default_profile
;
2337 if (dev
->caps
.steering_mode
==
2338 MLX4_STEERING_MODE_DEVICE_MANAGED
)
2339 profile
.num_mcg
= MLX4_FS_NUM_MCG
;
2341 icm_size
= mlx4_make_profile(dev
, &profile
, dev_cap
,
2343 if ((long long) icm_size
< 0) {
2348 dev
->caps
.max_fmr_maps
= (1 << (32 - ilog2(dev
->caps
.num_mpts
))) - 1;
2350 if (enable_4k_uar
|| !dev
->persist
->num_vfs
) {
2351 init_hca
->log_uar_sz
= ilog2(dev
->caps
.num_uars
) +
2352 PAGE_SHIFT
- DEFAULT_UAR_PAGE_SHIFT
;
2353 init_hca
->uar_page_sz
= DEFAULT_UAR_PAGE_SHIFT
- 12;
2355 init_hca
->log_uar_sz
= ilog2(dev
->caps
.num_uars
);
2356 init_hca
->uar_page_sz
= PAGE_SHIFT
- 12;
2359 init_hca
->mw_enabled
= 0;
2360 if (dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_MEM_WINDOW
||
2361 dev
->caps
.bmme_flags
& MLX4_BMME_FLAG_TYPE_2_WIN
)
2362 init_hca
->mw_enabled
= INIT_HCA_TPT_MW_ENABLE
;
2364 err
= mlx4_init_icm(dev
, dev_cap
, init_hca
, icm_size
);
2368 err
= mlx4_INIT_HCA(dev
, init_hca
);
2370 mlx4_err(dev
, "INIT_HCA command failed, aborting\n");
2374 if (dev_cap
->flags2
& MLX4_DEV_CAP_FLAG2_SYS_EQS
) {
2375 err
= mlx4_query_func(dev
, dev_cap
);
2377 mlx4_err(dev
, "QUERY_FUNC command failed, aborting.\n");
2379 } else if (err
& MLX4_QUERY_FUNC_NUM_SYS_EQS
) {
2380 dev
->caps
.num_eqs
= dev_cap
->max_eqs
;
2381 dev
->caps
.reserved_eqs
= dev_cap
->reserved_eqs
;
2382 dev
->caps
.reserved_uars
= dev_cap
->reserved_uars
;
2387 * If TS is supported by FW
2388 * read HCA frequency by QUERY_HCA command
2390 if (dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_TS
) {
2391 err
= mlx4_QUERY_HCA(dev
, init_hca
);
2393 mlx4_err(dev
, "QUERY_HCA command failed, disable timestamp\n");
2394 dev
->caps
.flags2
&= ~MLX4_DEV_CAP_FLAG2_TS
;
2396 dev
->caps
.hca_core_clock
=
2397 init_hca
->hca_core_clock
;
2400 /* In case we got HCA frequency 0 - disable timestamping
2401 * to avoid dividing by zero
2403 if (!dev
->caps
.hca_core_clock
) {
2404 dev
->caps
.flags2
&= ~MLX4_DEV_CAP_FLAG2_TS
;
2406 "HCA frequency is 0 - timestamping is not supported\n");
2407 } else if (map_internal_clock(dev
)) {
2409 * Map internal clock,
2410 * in case of failure disable timestamping
2412 dev
->caps
.flags2
&= ~MLX4_DEV_CAP_FLAG2_TS
;
2413 mlx4_err(dev
, "Failed to map internal clock. Timestamping is not supported\n");
2417 if (dev
->caps
.dmfs_high_steer_mode
!=
2418 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED
) {
2419 if (mlx4_validate_optimized_steering(dev
))
2420 mlx4_warn(dev
, "Optimized steering validation failed\n");
2422 if (dev
->caps
.dmfs_high_steer_mode
==
2423 MLX4_STEERING_DMFS_A0_DISABLE
) {
2424 dev
->caps
.dmfs_high_rate_qpn_base
=
2425 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
];
2426 dev
->caps
.dmfs_high_rate_qpn_range
=
2427 MLX4_A0_STEERING_TABLE_SIZE
;
2430 mlx4_info(dev
, "DMFS high rate steer mode is: %s\n",
2431 dmfs_high_rate_steering_mode_str(
2432 dev
->caps
.dmfs_high_steer_mode
));
2435 err
= mlx4_init_slave(dev
);
2437 if (err
!= -EPROBE_DEFER
)
2438 mlx4_err(dev
, "Failed to initialize slave\n");
2442 err
= mlx4_slave_cap(dev
);
2444 mlx4_err(dev
, "Failed to obtain slave caps\n");
2449 if (map_bf_area(dev
))
2450 mlx4_dbg(dev
, "Failed to map blue flame area\n");
2452 /*Only the master set the ports, all the rest got it from it.*/
2453 if (!mlx4_is_slave(dev
))
2454 mlx4_set_port_mask(dev
);
2456 err
= mlx4_QUERY_ADAPTER(dev
, &adapter
);
2458 mlx4_err(dev
, "QUERY_ADAPTER command failed, aborting\n");
2462 /* Query CONFIG_DEV parameters */
2463 err
= mlx4_config_dev_retrieval(dev
, ¶ms
);
2464 if (err
&& err
!= -EOPNOTSUPP
) {
2465 mlx4_err(dev
, "Failed to query CONFIG_DEV parameters\n");
2467 dev
->caps
.rx_checksum_flags_port
[1] = params
.rx_csum_flags_port_1
;
2468 dev
->caps
.rx_checksum_flags_port
[2] = params
.rx_csum_flags_port_2
;
2470 priv
->eq_table
.inta_pin
= adapter
.inta_pin
;
2471 memcpy(dev
->board_id
, adapter
.board_id
, sizeof(dev
->board_id
));
2477 unmap_internal_clock(dev
);
2480 if (mlx4_is_slave(dev
))
2481 mlx4_slave_destroy_special_qp_cap(dev
);
2484 if (mlx4_is_slave(dev
))
2485 mlx4_slave_exit(dev
);
2487 mlx4_CLOSE_HCA(dev
, 0);
2490 if (!mlx4_is_slave(dev
))
2491 mlx4_free_icms(dev
);
2500 static int mlx4_init_counters_table(struct mlx4_dev
*dev
)
2502 struct mlx4_priv
*priv
= mlx4_priv(dev
);
2505 if (!(dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_COUNTERS
))
2508 if (!dev
->caps
.max_counters
)
2511 nent_pow2
= roundup_pow_of_two(dev
->caps
.max_counters
);
2512 /* reserve last counter index for sink counter */
2513 return mlx4_bitmap_init(&priv
->counters_bitmap
, nent_pow2
,
2515 nent_pow2
- dev
->caps
.max_counters
+ 1);
2518 static void mlx4_cleanup_counters_table(struct mlx4_dev
*dev
)
2520 if (!(dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_COUNTERS
))
2523 if (!dev
->caps
.max_counters
)
2526 mlx4_bitmap_cleanup(&mlx4_priv(dev
)->counters_bitmap
);
2529 static void mlx4_cleanup_default_counters(struct mlx4_dev
*dev
)
2531 struct mlx4_priv
*priv
= mlx4_priv(dev
);
2534 for (port
= 0; port
< dev
->caps
.num_ports
; port
++)
2535 if (priv
->def_counter
[port
] != -1)
2536 mlx4_counter_free(dev
, priv
->def_counter
[port
]);
2539 static int mlx4_allocate_default_counters(struct mlx4_dev
*dev
)
2541 struct mlx4_priv
*priv
= mlx4_priv(dev
);
2545 for (port
= 0; port
< dev
->caps
.num_ports
; port
++)
2546 priv
->def_counter
[port
] = -1;
2548 for (port
= 0; port
< dev
->caps
.num_ports
; port
++) {
2549 err
= mlx4_counter_alloc(dev
, &idx
, MLX4_RES_USAGE_DRIVER
);
2551 if (!err
|| err
== -ENOSPC
) {
2552 priv
->def_counter
[port
] = idx
;
2554 } else if (err
== -ENOENT
) {
2557 } else if (mlx4_is_slave(dev
) && err
== -EINVAL
) {
2558 priv
->def_counter
[port
] = MLX4_SINK_COUNTER_INDEX(dev
);
2559 mlx4_warn(dev
, "can't allocate counter from old PF driver, using index %d\n",
2560 MLX4_SINK_COUNTER_INDEX(dev
));
2563 mlx4_err(dev
, "%s: failed to allocate default counter port %d err %d\n",
2564 __func__
, port
+ 1, err
);
2565 mlx4_cleanup_default_counters(dev
);
2569 mlx4_dbg(dev
, "%s: default counter index %d for port %d\n",
2570 __func__
, priv
->def_counter
[port
], port
+ 1);
2576 int __mlx4_counter_alloc(struct mlx4_dev
*dev
, u32
*idx
)
2578 struct mlx4_priv
*priv
= mlx4_priv(dev
);
2580 if (!(dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_COUNTERS
))
2583 *idx
= mlx4_bitmap_alloc(&priv
->counters_bitmap
);
2585 *idx
= MLX4_SINK_COUNTER_INDEX(dev
);
2592 int mlx4_counter_alloc(struct mlx4_dev
*dev
, u32
*idx
, u8 usage
)
2594 u32 in_modifier
= RES_COUNTER
| (((u32
)usage
& 3) << 30);
2598 if (mlx4_is_mfunc(dev
)) {
2599 err
= mlx4_cmd_imm(dev
, 0, &out_param
, in_modifier
,
2600 RES_OP_RESERVE
, MLX4_CMD_ALLOC_RES
,
2601 MLX4_CMD_TIME_CLASS_A
, MLX4_CMD_WRAPPED
);
2603 *idx
= get_param_l(&out_param
);
2604 if (WARN_ON(err
== -ENOSPC
))
2608 return __mlx4_counter_alloc(dev
, idx
);
2610 EXPORT_SYMBOL_GPL(mlx4_counter_alloc
);
2612 static int __mlx4_clear_if_stat(struct mlx4_dev
*dev
,
2615 struct mlx4_cmd_mailbox
*if_stat_mailbox
;
2617 u32 if_stat_in_mod
= (counter_index
& 0xff) | MLX4_QUERY_IF_STAT_RESET
;
2619 if_stat_mailbox
= mlx4_alloc_cmd_mailbox(dev
);
2620 if (IS_ERR(if_stat_mailbox
))
2621 return PTR_ERR(if_stat_mailbox
);
2623 err
= mlx4_cmd_box(dev
, 0, if_stat_mailbox
->dma
, if_stat_in_mod
, 0,
2624 MLX4_CMD_QUERY_IF_STAT
, MLX4_CMD_TIME_CLASS_C
,
2627 mlx4_free_cmd_mailbox(dev
, if_stat_mailbox
);
2631 void __mlx4_counter_free(struct mlx4_dev
*dev
, u32 idx
)
2633 if (!(dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_COUNTERS
))
2636 if (idx
== MLX4_SINK_COUNTER_INDEX(dev
))
2639 __mlx4_clear_if_stat(dev
, idx
);
2641 mlx4_bitmap_free(&mlx4_priv(dev
)->counters_bitmap
, idx
, MLX4_USE_RR
);
2645 void mlx4_counter_free(struct mlx4_dev
*dev
, u32 idx
)
2649 if (mlx4_is_mfunc(dev
)) {
2650 set_param_l(&in_param
, idx
);
2651 mlx4_cmd(dev
, in_param
, RES_COUNTER
, RES_OP_RESERVE
,
2652 MLX4_CMD_FREE_RES
, MLX4_CMD_TIME_CLASS_A
,
2656 __mlx4_counter_free(dev
, idx
);
2658 EXPORT_SYMBOL_GPL(mlx4_counter_free
);
2660 int mlx4_get_default_counter_index(struct mlx4_dev
*dev
, int port
)
2662 struct mlx4_priv
*priv
= mlx4_priv(dev
);
2664 return priv
->def_counter
[port
- 1];
2666 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index
);
2668 void mlx4_set_admin_guid(struct mlx4_dev
*dev
, __be64 guid
, int entry
, int port
)
2670 struct mlx4_priv
*priv
= mlx4_priv(dev
);
2672 priv
->mfunc
.master
.vf_admin
[entry
].vport
[port
].guid
= guid
;
2674 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid
);
2676 __be64
mlx4_get_admin_guid(struct mlx4_dev
*dev
, int entry
, int port
)
2678 struct mlx4_priv
*priv
= mlx4_priv(dev
);
2680 return priv
->mfunc
.master
.vf_admin
[entry
].vport
[port
].guid
;
2682 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid
);
2684 void mlx4_set_random_admin_guid(struct mlx4_dev
*dev
, int entry
, int port
)
2686 struct mlx4_priv
*priv
= mlx4_priv(dev
);
2693 get_random_bytes((char *)&guid
, sizeof(guid
));
2694 guid
&= ~(cpu_to_be64(1ULL << 56));
2695 guid
|= cpu_to_be64(1ULL << 57);
2696 priv
->mfunc
.master
.vf_admin
[entry
].vport
[port
].guid
= guid
;
2699 static int mlx4_setup_hca(struct mlx4_dev
*dev
)
2701 struct mlx4_priv
*priv
= mlx4_priv(dev
);
2704 __be32 ib_port_default_caps
;
2706 err
= mlx4_init_uar_table(dev
);
2708 mlx4_err(dev
, "Failed to initialize user access region table, aborting\n");
2712 err
= mlx4_uar_alloc(dev
, &priv
->driver_uar
);
2714 mlx4_err(dev
, "Failed to allocate driver access region, aborting\n");
2715 goto err_uar_table_free
;
2718 priv
->kar
= ioremap((phys_addr_t
) priv
->driver_uar
.pfn
<< PAGE_SHIFT
, PAGE_SIZE
);
2720 mlx4_err(dev
, "Couldn't map kernel access region, aborting\n");
2725 err
= mlx4_init_pd_table(dev
);
2727 mlx4_err(dev
, "Failed to initialize protection domain table, aborting\n");
2731 err
= mlx4_init_xrcd_table(dev
);
2733 mlx4_err(dev
, "Failed to initialize reliable connection domain table, aborting\n");
2734 goto err_pd_table_free
;
2737 err
= mlx4_init_mr_table(dev
);
2739 mlx4_err(dev
, "Failed to initialize memory region table, aborting\n");
2740 goto err_xrcd_table_free
;
2743 if (!mlx4_is_slave(dev
)) {
2744 err
= mlx4_init_mcg_table(dev
);
2746 mlx4_err(dev
, "Failed to initialize multicast group table, aborting\n");
2747 goto err_mr_table_free
;
2749 err
= mlx4_config_mad_demux(dev
);
2751 mlx4_err(dev
, "Failed in config_mad_demux, aborting\n");
2752 goto err_mcg_table_free
;
2756 err
= mlx4_init_eq_table(dev
);
2758 mlx4_err(dev
, "Failed to initialize event queue table, aborting\n");
2759 goto err_mcg_table_free
;
2762 err
= mlx4_cmd_use_events(dev
);
2764 mlx4_err(dev
, "Failed to switch to event-driven firmware commands, aborting\n");
2765 goto err_eq_table_free
;
2768 err
= mlx4_NOP(dev
);
2770 if (dev
->flags
& MLX4_FLAG_MSI_X
) {
2771 mlx4_warn(dev
, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2772 priv
->eq_table
.eq
[MLX4_EQ_ASYNC
].irq
);
2773 mlx4_warn(dev
, "Trying again without MSI-X\n");
2775 mlx4_err(dev
, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2776 priv
->eq_table
.eq
[MLX4_EQ_ASYNC
].irq
);
2777 mlx4_err(dev
, "BIOS or ACPI interrupt routing problem?\n");
2783 mlx4_dbg(dev
, "NOP command IRQ test passed\n");
2785 err
= mlx4_init_cq_table(dev
);
2787 mlx4_err(dev
, "Failed to initialize completion queue table, aborting\n");
2791 err
= mlx4_init_srq_table(dev
);
2793 mlx4_err(dev
, "Failed to initialize shared receive queue table, aborting\n");
2794 goto err_cq_table_free
;
2797 err
= mlx4_init_qp_table(dev
);
2799 mlx4_err(dev
, "Failed to initialize queue pair table, aborting\n");
2800 goto err_srq_table_free
;
2803 if (!mlx4_is_slave(dev
)) {
2804 err
= mlx4_init_counters_table(dev
);
2805 if (err
&& err
!= -ENOENT
) {
2806 mlx4_err(dev
, "Failed to initialize counters table, aborting\n");
2807 goto err_qp_table_free
;
2811 err
= mlx4_allocate_default_counters(dev
);
2813 mlx4_err(dev
, "Failed to allocate default counters, aborting\n");
2814 goto err_counters_table_free
;
2817 if (!mlx4_is_slave(dev
)) {
2818 for (port
= 1; port
<= dev
->caps
.num_ports
; port
++) {
2819 ib_port_default_caps
= 0;
2820 err
= mlx4_get_port_ib_caps(dev
, port
,
2821 &ib_port_default_caps
);
2823 mlx4_warn(dev
, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2825 dev
->caps
.ib_port_def_cap
[port
] = ib_port_default_caps
;
2827 /* initialize per-slave default ib port capabilities */
2828 if (mlx4_is_master(dev
)) {
2830 for (i
= 0; i
< dev
->num_slaves
; i
++) {
2831 if (i
== mlx4_master_func_num(dev
))
2833 priv
->mfunc
.master
.slave_state
[i
].ib_cap_mask
[port
] =
2834 ib_port_default_caps
;
2838 if (mlx4_is_mfunc(dev
))
2839 dev
->caps
.port_ib_mtu
[port
] = IB_MTU_2048
;
2841 dev
->caps
.port_ib_mtu
[port
] = IB_MTU_4096
;
2843 err
= mlx4_SET_PORT(dev
, port
, mlx4_is_master(dev
) ?
2844 dev
->caps
.pkey_table_len
[port
] : -1);
2846 mlx4_err(dev
, "Failed to set port %d, aborting\n",
2848 goto err_default_countes_free
;
2855 err_default_countes_free
:
2856 mlx4_cleanup_default_counters(dev
);
2858 err_counters_table_free
:
2859 if (!mlx4_is_slave(dev
))
2860 mlx4_cleanup_counters_table(dev
);
2863 mlx4_cleanup_qp_table(dev
);
2866 mlx4_cleanup_srq_table(dev
);
2869 mlx4_cleanup_cq_table(dev
);
2872 mlx4_cmd_use_polling(dev
);
2875 mlx4_cleanup_eq_table(dev
);
2878 if (!mlx4_is_slave(dev
))
2879 mlx4_cleanup_mcg_table(dev
);
2882 mlx4_cleanup_mr_table(dev
);
2884 err_xrcd_table_free
:
2885 mlx4_cleanup_xrcd_table(dev
);
2888 mlx4_cleanup_pd_table(dev
);
2894 mlx4_uar_free(dev
, &priv
->driver_uar
);
2897 mlx4_cleanup_uar_table(dev
);
2901 static int mlx4_init_affinity_hint(struct mlx4_dev
*dev
, int port
, int eqn
)
2903 int requested_cpu
= 0;
2904 struct mlx4_priv
*priv
= mlx4_priv(dev
);
2909 if (eqn
> dev
->caps
.num_comp_vectors
)
2912 for (i
= 1; i
< port
; i
++)
2913 off
+= mlx4_get_eqs_per_port(dev
, i
);
2915 requested_cpu
= eqn
- off
- !!(eqn
> MLX4_EQ_ASYNC
);
2917 /* Meaning EQs are shared, and this call comes from the second port */
2918 if (requested_cpu
< 0)
2921 eq
= &priv
->eq_table
.eq
[eqn
];
2923 if (!zalloc_cpumask_var(&eq
->affinity_mask
, GFP_KERNEL
))
2926 cpumask_set_cpu(requested_cpu
, eq
->affinity_mask
);
2931 static void mlx4_enable_msi_x(struct mlx4_dev
*dev
)
2933 struct mlx4_priv
*priv
= mlx4_priv(dev
);
2934 struct msix_entry
*entries
;
2939 int nreq
= min3(dev
->caps
.num_ports
*
2940 (int)num_online_cpus() + 1,
2941 dev
->caps
.num_eqs
- dev
->caps
.reserved_eqs
,
2945 nreq
= min_t(int, nreq
, msi_x
);
2947 entries
= kcalloc(nreq
, sizeof(*entries
), GFP_KERNEL
);
2951 for (i
= 0; i
< nreq
; ++i
)
2952 entries
[i
].entry
= i
;
2954 nreq
= pci_enable_msix_range(dev
->persist
->pdev
, entries
, 2,
2957 if (nreq
< 0 || nreq
< MLX4_EQ_ASYNC
) {
2961 /* 1 is reserved for events (asyncrounous EQ) */
2962 dev
->caps
.num_comp_vectors
= nreq
- 1;
2964 priv
->eq_table
.eq
[MLX4_EQ_ASYNC
].irq
= entries
[0].vector
;
2965 bitmap_zero(priv
->eq_table
.eq
[MLX4_EQ_ASYNC
].actv_ports
.ports
,
2966 dev
->caps
.num_ports
);
2968 for (i
= 0; i
< dev
->caps
.num_comp_vectors
+ 1; i
++) {
2969 if (i
== MLX4_EQ_ASYNC
)
2972 priv
->eq_table
.eq
[i
].irq
=
2973 entries
[i
+ 1 - !!(i
> MLX4_EQ_ASYNC
)].vector
;
2975 if (MLX4_IS_LEGACY_EQ_MODE(dev
->caps
)) {
2976 bitmap_fill(priv
->eq_table
.eq
[i
].actv_ports
.ports
,
2977 dev
->caps
.num_ports
);
2978 /* We don't set affinity hint when there
2983 priv
->eq_table
.eq
[i
].actv_ports
.ports
);
2984 if (mlx4_init_affinity_hint(dev
, port
+ 1, i
))
2985 mlx4_warn(dev
, "Couldn't init hint cpumask for EQ %d\n",
2988 /* We divide the Eqs evenly between the two ports.
2989 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
2990 * refers to the number of Eqs per port
2991 * (i.e eqs_per_port). Theoretically, we would like to
2992 * write something like (i + 1) % eqs_per_port == 0.
2993 * However, since there's an asynchronous Eq, we have
2994 * to skip over it by comparing this condition to
2995 * !!((i + 1) > MLX4_EQ_ASYNC).
2997 if ((dev
->caps
.num_comp_vectors
> dev
->caps
.num_ports
) &&
2999 (dev
->caps
.num_comp_vectors
/ dev
->caps
.num_ports
)) ==
3000 !!((i
+ 1) > MLX4_EQ_ASYNC
))
3001 /* If dev->caps.num_comp_vectors < dev->caps.num_ports,
3002 * everything is shared anyway.
3007 dev
->flags
|= MLX4_FLAG_MSI_X
;
3014 dev
->caps
.num_comp_vectors
= 1;
3016 BUG_ON(MLX4_EQ_ASYNC
>= 2);
3017 for (i
= 0; i
< 2; ++i
) {
3018 priv
->eq_table
.eq
[i
].irq
= dev
->persist
->pdev
->irq
;
3019 if (i
!= MLX4_EQ_ASYNC
) {
3020 bitmap_fill(priv
->eq_table
.eq
[i
].actv_ports
.ports
,
3021 dev
->caps
.num_ports
);
3026 static int mlx4_init_port_info(struct mlx4_dev
*dev
, int port
)
3028 struct devlink
*devlink
= priv_to_devlink(mlx4_priv(dev
));
3029 struct mlx4_port_info
*info
= &mlx4_priv(dev
)->port
[port
];
3032 err
= devlink_port_register(devlink
, &info
->devlink_port
, port
);
3038 if (!mlx4_is_slave(dev
)) {
3039 mlx4_init_mac_table(dev
, &info
->mac_table
);
3040 mlx4_init_vlan_table(dev
, &info
->vlan_table
);
3041 mlx4_init_roce_gid_table(dev
, &info
->gid_table
);
3042 info
->base_qpn
= mlx4_get_base_qpn(dev
, port
);
3045 sprintf(info
->dev_name
, "mlx4_port%d", port
);
3046 info
->port_attr
.attr
.name
= info
->dev_name
;
3047 if (mlx4_is_mfunc(dev
)) {
3048 info
->port_attr
.attr
.mode
= 0444;
3050 info
->port_attr
.attr
.mode
= 0644;
3051 info
->port_attr
.store
= set_port_type
;
3053 info
->port_attr
.show
= show_port_type
;
3054 sysfs_attr_init(&info
->port_attr
.attr
);
3056 err
= device_create_file(&dev
->persist
->pdev
->dev
, &info
->port_attr
);
3058 mlx4_err(dev
, "Failed to create file for port %d\n", port
);
3059 devlink_port_unregister(&info
->devlink_port
);
3064 sprintf(info
->dev_mtu_name
, "mlx4_port%d_mtu", port
);
3065 info
->port_mtu_attr
.attr
.name
= info
->dev_mtu_name
;
3066 if (mlx4_is_mfunc(dev
)) {
3067 info
->port_mtu_attr
.attr
.mode
= 0444;
3069 info
->port_mtu_attr
.attr
.mode
= 0644;
3070 info
->port_mtu_attr
.store
= set_port_ib_mtu
;
3072 info
->port_mtu_attr
.show
= show_port_ib_mtu
;
3073 sysfs_attr_init(&info
->port_mtu_attr
.attr
);
3075 err
= device_create_file(&dev
->persist
->pdev
->dev
,
3076 &info
->port_mtu_attr
);
3078 mlx4_err(dev
, "Failed to create mtu file for port %d\n", port
);
3079 device_remove_file(&info
->dev
->persist
->pdev
->dev
,
3081 devlink_port_unregister(&info
->devlink_port
);
3089 static void mlx4_cleanup_port_info(struct mlx4_port_info
*info
)
3094 device_remove_file(&info
->dev
->persist
->pdev
->dev
, &info
->port_attr
);
3095 device_remove_file(&info
->dev
->persist
->pdev
->dev
,
3096 &info
->port_mtu_attr
);
3097 devlink_port_unregister(&info
->devlink_port
);
3099 #ifdef CONFIG_RFS_ACCEL
3100 free_irq_cpu_rmap(info
->rmap
);
3105 static int mlx4_init_steering(struct mlx4_dev
*dev
)
3107 struct mlx4_priv
*priv
= mlx4_priv(dev
);
3108 int num_entries
= dev
->caps
.num_ports
;
3111 priv
->steer
= kcalloc(num_entries
, sizeof(struct mlx4_steer
),
3116 for (i
= 0; i
< num_entries
; i
++)
3117 for (j
= 0; j
< MLX4_NUM_STEERS
; j
++) {
3118 INIT_LIST_HEAD(&priv
->steer
[i
].promisc_qps
[j
]);
3119 INIT_LIST_HEAD(&priv
->steer
[i
].steer_entries
[j
]);
3124 static void mlx4_clear_steering(struct mlx4_dev
*dev
)
3126 struct mlx4_priv
*priv
= mlx4_priv(dev
);
3127 struct mlx4_steer_index
*entry
, *tmp_entry
;
3128 struct mlx4_promisc_qp
*pqp
, *tmp_pqp
;
3129 int num_entries
= dev
->caps
.num_ports
;
3132 for (i
= 0; i
< num_entries
; i
++) {
3133 for (j
= 0; j
< MLX4_NUM_STEERS
; j
++) {
3134 list_for_each_entry_safe(pqp
, tmp_pqp
,
3135 &priv
->steer
[i
].promisc_qps
[j
],
3137 list_del(&pqp
->list
);
3140 list_for_each_entry_safe(entry
, tmp_entry
,
3141 &priv
->steer
[i
].steer_entries
[j
],
3143 list_del(&entry
->list
);
3144 list_for_each_entry_safe(pqp
, tmp_pqp
,
3147 list_del(&pqp
->list
);
3157 static int extended_func_num(struct pci_dev
*pdev
)
3159 return PCI_SLOT(pdev
->devfn
) * 8 + PCI_FUNC(pdev
->devfn
);
3162 #define MLX4_OWNER_BASE 0x8069c
3163 #define MLX4_OWNER_SIZE 4
3165 static int mlx4_get_ownership(struct mlx4_dev
*dev
)
3167 void __iomem
*owner
;
3170 if (pci_channel_offline(dev
->persist
->pdev
))
3173 owner
= ioremap(pci_resource_start(dev
->persist
->pdev
, 0) +
3177 mlx4_err(dev
, "Failed to obtain ownership bit\n");
3186 static void mlx4_free_ownership(struct mlx4_dev
*dev
)
3188 void __iomem
*owner
;
3190 if (pci_channel_offline(dev
->persist
->pdev
))
3193 owner
= ioremap(pci_resource_start(dev
->persist
->pdev
, 0) +
3197 mlx4_err(dev
, "Failed to obtain ownership bit\n");
3205 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV) ==\
3206 !!((flags) & MLX4_FLAG_MASTER))
3208 static u64
mlx4_enable_sriov(struct mlx4_dev
*dev
, struct pci_dev
*pdev
,
3209 u8 total_vfs
, int existing_vfs
, int reset_flow
)
3211 u64 dev_flags
= dev
->flags
;
3213 int fw_enabled_sriov_vfs
= min(pci_sriov_get_totalvfs(pdev
),
3217 dev
->dev_vfs
= kcalloc(total_vfs
, sizeof(*dev
->dev_vfs
),
3224 atomic_inc(&pf_loading
);
3225 if (dev
->flags
& MLX4_FLAG_SRIOV
) {
3226 if (existing_vfs
!= total_vfs
) {
3227 mlx4_err(dev
, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
3228 existing_vfs
, total_vfs
);
3229 total_vfs
= existing_vfs
;
3233 dev
->dev_vfs
= kcalloc(total_vfs
, sizeof(*dev
->dev_vfs
), GFP_KERNEL
);
3234 if (NULL
== dev
->dev_vfs
) {
3235 mlx4_err(dev
, "Failed to allocate memory for VFs\n");
3239 if (!(dev
->flags
& MLX4_FLAG_SRIOV
)) {
3240 if (total_vfs
> fw_enabled_sriov_vfs
) {
3241 mlx4_err(dev
, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
3242 total_vfs
, fw_enabled_sriov_vfs
);
3246 mlx4_warn(dev
, "Enabling SR-IOV with %d VFs\n", total_vfs
);
3247 err
= pci_enable_sriov(pdev
, total_vfs
);
3250 mlx4_err(dev
, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
3254 mlx4_warn(dev
, "Running in master mode\n");
3255 dev_flags
|= MLX4_FLAG_SRIOV
|
3257 dev_flags
&= ~MLX4_FLAG_SLAVE
;
3258 dev
->persist
->num_vfs
= total_vfs
;
3263 atomic_dec(&pf_loading
);
3265 dev
->persist
->num_vfs
= 0;
3266 kfree(dev
->dev_vfs
);
3267 dev
->dev_vfs
= NULL
;
3268 return dev_flags
& ~MLX4_FLAG_MASTER
;
3272 MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64
= -1,
3275 static int mlx4_check_dev_cap(struct mlx4_dev
*dev
, struct mlx4_dev_cap
*dev_cap
,
3278 int requested_vfs
= nvfs
[0] + nvfs
[1] + nvfs
[2];
3279 /* Checking for 64 VFs as a limitation of CX2 */
3280 if (!(dev_cap
->flags2
& MLX4_DEV_CAP_FLAG2_80_VFS
) &&
3281 requested_vfs
>= 64) {
3282 mlx4_err(dev
, "Requested %d VFs, but FW does not support more than 64\n",
3284 return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64
;
3289 static int mlx4_pci_enable_device(struct mlx4_dev
*dev
)
3291 struct pci_dev
*pdev
= dev
->persist
->pdev
;
3294 mutex_lock(&dev
->persist
->pci_status_mutex
);
3295 if (dev
->persist
->pci_status
== MLX4_PCI_STATUS_DISABLED
) {
3296 err
= pci_enable_device(pdev
);
3298 dev
->persist
->pci_status
= MLX4_PCI_STATUS_ENABLED
;
3300 mutex_unlock(&dev
->persist
->pci_status_mutex
);
3305 static void mlx4_pci_disable_device(struct mlx4_dev
*dev
)
3307 struct pci_dev
*pdev
= dev
->persist
->pdev
;
3309 mutex_lock(&dev
->persist
->pci_status_mutex
);
3310 if (dev
->persist
->pci_status
== MLX4_PCI_STATUS_ENABLED
) {
3311 pci_disable_device(pdev
);
3312 dev
->persist
->pci_status
= MLX4_PCI_STATUS_DISABLED
;
3314 mutex_unlock(&dev
->persist
->pci_status_mutex
);
3317 static int mlx4_load_one(struct pci_dev
*pdev
, int pci_dev_data
,
3318 int total_vfs
, int *nvfs
, struct mlx4_priv
*priv
,
3321 struct mlx4_dev
*dev
;
3326 struct mlx4_dev_cap
*dev_cap
= NULL
;
3327 int existing_vfs
= 0;
3331 INIT_LIST_HEAD(&priv
->ctx_list
);
3332 spin_lock_init(&priv
->ctx_lock
);
3334 mutex_init(&priv
->port_mutex
);
3335 mutex_init(&priv
->bond_mutex
);
3337 INIT_LIST_HEAD(&priv
->pgdir_list
);
3338 mutex_init(&priv
->pgdir_mutex
);
3339 spin_lock_init(&priv
->cmd
.context_lock
);
3341 INIT_LIST_HEAD(&priv
->bf_list
);
3342 mutex_init(&priv
->bf_mutex
);
3344 dev
->rev_id
= pdev
->revision
;
3345 dev
->numa_node
= dev_to_node(&pdev
->dev
);
3347 /* Detect if this device is a virtual function */
3348 if (pci_dev_data
& MLX4_PCI_DEV_IS_VF
) {
3349 mlx4_warn(dev
, "Detected virtual function - running in slave mode\n");
3350 dev
->flags
|= MLX4_FLAG_SLAVE
;
3352 /* We reset the device and enable SRIOV only for physical
3353 * devices. Try to claim ownership on the device;
3354 * if already taken, skip -- do not allow multiple PFs */
3355 err
= mlx4_get_ownership(dev
);
3360 mlx4_warn(dev
, "Multiple PFs not yet supported - Skipping PF\n");
3365 atomic_set(&priv
->opreq_count
, 0);
3366 INIT_WORK(&priv
->opreq_task
, mlx4_opreq_action
);
3369 * Now reset the HCA before we touch the PCI capabilities or
3370 * attempt a firmware command, since a boot ROM may have left
3371 * the HCA in an undefined state.
3373 err
= mlx4_reset(dev
);
3375 mlx4_err(dev
, "Failed to reset HCA, aborting\n");
3380 dev
->flags
= MLX4_FLAG_MASTER
;
3381 existing_vfs
= pci_num_vf(pdev
);
3383 dev
->flags
|= MLX4_FLAG_SRIOV
;
3384 dev
->persist
->num_vfs
= total_vfs
;
3388 /* on load remove any previous indication of internal error,
3391 dev
->persist
->state
= MLX4_DEVICE_STATE_UP
;
3394 err
= mlx4_cmd_init(dev
);
3396 mlx4_err(dev
, "Failed to init command interface, aborting\n");
3400 /* In slave functions, the communication channel must be initialized
3401 * before posting commands. Also, init num_slaves before calling
3403 if (mlx4_is_mfunc(dev
)) {
3404 if (mlx4_is_master(dev
)) {
3405 dev
->num_slaves
= MLX4_MAX_NUM_SLAVES
;
3408 dev
->num_slaves
= 0;
3409 err
= mlx4_multi_func_init(dev
);
3411 mlx4_err(dev
, "Failed to init slave mfunc interface, aborting\n");
3417 err
= mlx4_init_fw(dev
);
3419 mlx4_err(dev
, "Failed to init fw, aborting.\n");
3423 if (mlx4_is_master(dev
)) {
3424 /* when we hit the goto slave_start below, dev_cap already initialized */
3426 dev_cap
= kzalloc(sizeof(*dev_cap
), GFP_KERNEL
);
3433 err
= mlx4_QUERY_DEV_CAP(dev
, dev_cap
);
3435 mlx4_err(dev
, "QUERY_DEV_CAP command failed, aborting.\n");
3439 if (mlx4_check_dev_cap(dev
, dev_cap
, nvfs
))
3442 if (!(dev_cap
->flags2
& MLX4_DEV_CAP_FLAG2_SYS_EQS
)) {
3443 u64 dev_flags
= mlx4_enable_sriov(dev
, pdev
,
3449 mlx4_cmd_cleanup(dev
, MLX4_CMD_CLEANUP_ALL
);
3450 dev
->flags
= dev_flags
;
3451 if (!SRIOV_VALID_STATE(dev
->flags
)) {
3452 mlx4_err(dev
, "Invalid SRIOV state\n");
3455 err
= mlx4_reset(dev
);
3457 mlx4_err(dev
, "Failed to reset HCA, aborting.\n");
3463 /* Legacy mode FW requires SRIOV to be enabled before
3464 * doing QUERY_DEV_CAP, since max_eq's value is different if
3467 memset(dev_cap
, 0, sizeof(*dev_cap
));
3468 err
= mlx4_QUERY_DEV_CAP(dev
, dev_cap
);
3470 mlx4_err(dev
, "QUERY_DEV_CAP command failed, aborting.\n");
3474 if (mlx4_check_dev_cap(dev
, dev_cap
, nvfs
))
3479 err
= mlx4_init_hca(dev
);
3481 if (err
== -EACCES
) {
3482 /* Not primary Physical function
3483 * Running in slave mode */
3484 mlx4_cmd_cleanup(dev
, MLX4_CMD_CLEANUP_ALL
);
3485 /* We're not a PF */
3486 if (dev
->flags
& MLX4_FLAG_SRIOV
) {
3488 pci_disable_sriov(pdev
);
3489 if (mlx4_is_master(dev
) && !reset_flow
)
3490 atomic_dec(&pf_loading
);
3491 dev
->flags
&= ~MLX4_FLAG_SRIOV
;
3493 if (!mlx4_is_slave(dev
))
3494 mlx4_free_ownership(dev
);
3495 dev
->flags
|= MLX4_FLAG_SLAVE
;
3496 dev
->flags
&= ~MLX4_FLAG_MASTER
;
3502 if (mlx4_is_master(dev
) && (dev_cap
->flags2
& MLX4_DEV_CAP_FLAG2_SYS_EQS
)) {
3503 u64 dev_flags
= mlx4_enable_sriov(dev
, pdev
, total_vfs
,
3504 existing_vfs
, reset_flow
);
3506 if ((dev
->flags
^ dev_flags
) & (MLX4_FLAG_MASTER
| MLX4_FLAG_SLAVE
)) {
3507 mlx4_cmd_cleanup(dev
, MLX4_CMD_CLEANUP_VHCR
);
3508 dev
->flags
= dev_flags
;
3509 err
= mlx4_cmd_init(dev
);
3511 /* Only VHCR is cleaned up, so could still
3514 mlx4_err(dev
, "Failed to init VHCR command interface, aborting\n");
3518 dev
->flags
= dev_flags
;
3521 if (!SRIOV_VALID_STATE(dev
->flags
)) {
3522 mlx4_err(dev
, "Invalid SRIOV state\n");
3527 /* check if the device is functioning at its maximum possible speed.
3528 * No return code for this call, just warn the user in case of PCI
3529 * express device capabilities are under-satisfied by the bus.
3531 if (!mlx4_is_slave(dev
))
3532 pcie_print_link_status(dev
->persist
->pdev
);
3534 /* In master functions, the communication channel must be initialized
3535 * after obtaining its address from fw */
3536 if (mlx4_is_master(dev
)) {
3537 if (dev
->caps
.num_ports
< 2 &&
3541 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3542 dev
->caps
.num_ports
);
3545 memcpy(dev
->persist
->nvfs
, nvfs
, sizeof(dev
->persist
->nvfs
));
3548 i
< sizeof(dev
->persist
->nvfs
)/
3549 sizeof(dev
->persist
->nvfs
[0]); i
++) {
3552 for (j
= 0; j
< dev
->persist
->nvfs
[i
]; ++sum
, ++j
) {
3553 dev
->dev_vfs
[sum
].min_port
= i
< 2 ? i
+ 1 : 1;
3554 dev
->dev_vfs
[sum
].n_ports
= i
< 2 ? 1 :
3555 dev
->caps
.num_ports
;
3559 /* In master functions, the communication channel
3560 * must be initialized after obtaining its address from fw
3562 err
= mlx4_multi_func_init(dev
);
3564 mlx4_err(dev
, "Failed to init master mfunc interface, aborting.\n");
3569 err
= mlx4_alloc_eq_table(dev
);
3571 goto err_master_mfunc
;
3573 bitmap_zero(priv
->msix_ctl
.pool_bm
, MAX_MSIX
);
3574 mutex_init(&priv
->msix_ctl
.pool_lock
);
3576 mlx4_enable_msi_x(dev
);
3577 if ((mlx4_is_mfunc(dev
)) &&
3578 !(dev
->flags
& MLX4_FLAG_MSI_X
)) {
3580 mlx4_err(dev
, "INTx is not supported in multi-function mode, aborting\n");
3584 if (!mlx4_is_slave(dev
)) {
3585 err
= mlx4_init_steering(dev
);
3587 goto err_disable_msix
;
3590 mlx4_init_quotas(dev
);
3592 err
= mlx4_setup_hca(dev
);
3593 if (err
== -EBUSY
&& (dev
->flags
& MLX4_FLAG_MSI_X
) &&
3594 !mlx4_is_mfunc(dev
)) {
3595 dev
->flags
&= ~MLX4_FLAG_MSI_X
;
3596 dev
->caps
.num_comp_vectors
= 1;
3597 pci_disable_msix(pdev
);
3598 err
= mlx4_setup_hca(dev
);
3604 /* When PF resources are ready arm its comm channel to enable
3607 if (mlx4_is_master(dev
)) {
3608 err
= mlx4_ARM_COMM_CHANNEL(dev
);
3610 mlx4_err(dev
, " Failed to arm comm channel eq: %x\n",
3616 for (port
= 1; port
<= dev
->caps
.num_ports
; port
++) {
3617 err
= mlx4_init_port_info(dev
, port
);
3622 priv
->v2p
.port1
= 1;
3623 priv
->v2p
.port2
= 2;
3625 err
= mlx4_register_device(dev
);
3629 mlx4_request_modules(dev
);
3631 mlx4_sense_init(dev
);
3632 mlx4_start_sense(dev
);
3636 if (mlx4_is_master(dev
) && dev
->persist
->num_vfs
&& !reset_flow
)
3637 atomic_dec(&pf_loading
);
3643 for (--port
; port
>= 1; --port
)
3644 mlx4_cleanup_port_info(&priv
->port
[port
]);
3646 mlx4_cleanup_default_counters(dev
);
3647 if (!mlx4_is_slave(dev
))
3648 mlx4_cleanup_counters_table(dev
);
3649 mlx4_cleanup_qp_table(dev
);
3650 mlx4_cleanup_srq_table(dev
);
3651 mlx4_cleanup_cq_table(dev
);
3652 mlx4_cmd_use_polling(dev
);
3653 mlx4_cleanup_eq_table(dev
);
3654 mlx4_cleanup_mcg_table(dev
);
3655 mlx4_cleanup_mr_table(dev
);
3656 mlx4_cleanup_xrcd_table(dev
);
3657 mlx4_cleanup_pd_table(dev
);
3658 mlx4_cleanup_uar_table(dev
);
3661 if (!mlx4_is_slave(dev
))
3662 mlx4_clear_steering(dev
);
3665 if (dev
->flags
& MLX4_FLAG_MSI_X
)
3666 pci_disable_msix(pdev
);
3669 mlx4_free_eq_table(dev
);
3672 if (mlx4_is_master(dev
)) {
3673 mlx4_free_resource_tracker(dev
, RES_TR_FREE_STRUCTS_ONLY
);
3674 mlx4_multi_func_cleanup(dev
);
3677 if (mlx4_is_slave(dev
))
3678 mlx4_slave_destroy_special_qp_cap(dev
);
3681 mlx4_close_hca(dev
);
3687 if (mlx4_is_slave(dev
))
3688 mlx4_multi_func_cleanup(dev
);
3691 mlx4_cmd_cleanup(dev
, MLX4_CMD_CLEANUP_ALL
);
3694 if (dev
->flags
& MLX4_FLAG_SRIOV
&& !existing_vfs
) {
3695 pci_disable_sriov(pdev
);
3696 dev
->flags
&= ~MLX4_FLAG_SRIOV
;
3699 if (mlx4_is_master(dev
) && dev
->persist
->num_vfs
&& !reset_flow
)
3700 atomic_dec(&pf_loading
);
3702 kfree(priv
->dev
.dev_vfs
);
3704 if (!mlx4_is_slave(dev
))
3705 mlx4_free_ownership(dev
);
3711 static int __mlx4_init_one(struct pci_dev
*pdev
, int pci_dev_data
,
3712 struct mlx4_priv
*priv
)
3715 int nvfs
[MLX4_MAX_PORTS
+ 1] = {0, 0, 0};
3716 int prb_vf
[MLX4_MAX_PORTS
+ 1] = {0, 0, 0};
3717 const int param_map
[MLX4_MAX_PORTS
+ 1][MLX4_MAX_PORTS
+ 1] = {
3718 {2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3719 unsigned total_vfs
= 0;
3722 pr_info(DRV_NAME
": Initializing %s\n", pci_name(pdev
));
3724 err
= mlx4_pci_enable_device(&priv
->dev
);
3726 dev_err(&pdev
->dev
, "Cannot enable PCI device, aborting\n");
3730 /* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3731 * per port, we must limit the number of VFs to 63 (since their are
3734 for (i
= 0; i
< ARRAY_SIZE(nvfs
) && i
< num_vfs_argc
;
3735 total_vfs
+= nvfs
[param_map
[num_vfs_argc
- 1][i
]], i
++) {
3736 nvfs
[param_map
[num_vfs_argc
- 1][i
]] = num_vfs
[i
];
3738 dev_err(&pdev
->dev
, "num_vfs module parameter cannot be negative\n");
3740 goto err_disable_pdev
;
3743 for (i
= 0; i
< ARRAY_SIZE(prb_vf
) && i
< probe_vfs_argc
;
3745 prb_vf
[param_map
[probe_vfs_argc
- 1][i
]] = probe_vf
[i
];
3746 if (prb_vf
[i
] < 0 || prb_vf
[i
] > nvfs
[i
]) {
3747 dev_err(&pdev
->dev
, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3749 goto err_disable_pdev
;
3752 if (total_vfs
> MLX4_MAX_NUM_VF
) {
3754 "Requested more VF's (%d) than allowed by hw (%d)\n",
3755 total_vfs
, MLX4_MAX_NUM_VF
);
3757 goto err_disable_pdev
;
3760 for (i
= 0; i
< MLX4_MAX_PORTS
; i
++) {
3761 if (nvfs
[i
] + nvfs
[2] > MLX4_MAX_NUM_VF_P_PORT
) {
3763 "Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3764 nvfs
[i
] + nvfs
[2], i
+ 1,
3765 MLX4_MAX_NUM_VF_P_PORT
);
3767 goto err_disable_pdev
;
3771 /* Check for BARs. */
3772 if (!(pci_dev_data
& MLX4_PCI_DEV_IS_VF
) &&
3773 !(pci_resource_flags(pdev
, 0) & IORESOURCE_MEM
)) {
3774 dev_err(&pdev
->dev
, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3775 pci_dev_data
, pci_resource_flags(pdev
, 0));
3777 goto err_disable_pdev
;
3779 if (!(pci_resource_flags(pdev
, 2) & IORESOURCE_MEM
)) {
3780 dev_err(&pdev
->dev
, "Missing UAR, aborting\n");
3782 goto err_disable_pdev
;
3785 err
= pci_request_regions(pdev
, DRV_NAME
);
3787 dev_err(&pdev
->dev
, "Couldn't get PCI resources, aborting\n");
3788 goto err_disable_pdev
;
3791 pci_set_master(pdev
);
3793 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(64));
3795 dev_warn(&pdev
->dev
, "Warning: couldn't set 64-bit PCI DMA mask\n");
3796 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(32));
3798 dev_err(&pdev
->dev
, "Can't set PCI DMA mask, aborting\n");
3799 goto err_release_regions
;
3802 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(64));
3804 dev_warn(&pdev
->dev
, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3805 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(32));
3807 dev_err(&pdev
->dev
, "Can't set consistent PCI DMA mask, aborting\n");
3808 goto err_release_regions
;
3812 /* Allow large DMA segments, up to the firmware limit of 1 GB */
3813 dma_set_max_seg_size(&pdev
->dev
, 1024 * 1024 * 1024);
3814 /* Detect if this device is a virtual function */
3815 if (pci_dev_data
& MLX4_PCI_DEV_IS_VF
) {
3816 /* When acting as pf, we normally skip vfs unless explicitly
3817 * requested to probe them.
3820 unsigned vfs_offset
= 0;
3822 for (i
= 0; i
< ARRAY_SIZE(nvfs
) &&
3823 vfs_offset
+ nvfs
[i
] < extended_func_num(pdev
);
3824 vfs_offset
+= nvfs
[i
], i
++)
3826 if (i
== ARRAY_SIZE(nvfs
)) {
3828 goto err_release_regions
;
3830 if ((extended_func_num(pdev
) - vfs_offset
)
3832 dev_warn(&pdev
->dev
, "Skipping virtual function:%d\n",
3833 extended_func_num(pdev
));
3835 goto err_release_regions
;
3840 err
= mlx4_crdump_init(&priv
->dev
);
3842 goto err_release_regions
;
3844 err
= mlx4_catas_init(&priv
->dev
);
3848 err
= mlx4_load_one(pdev
, pci_dev_data
, total_vfs
, nvfs
, priv
, 0);
3855 mlx4_catas_end(&priv
->dev
);
3858 mlx4_crdump_end(&priv
->dev
);
3860 err_release_regions
:
3861 pci_release_regions(pdev
);
3864 mlx4_pci_disable_device(&priv
->dev
);
3868 static int mlx4_devlink_port_type_set(struct devlink_port
*devlink_port
,
3869 enum devlink_port_type port_type
)
3871 struct mlx4_port_info
*info
= container_of(devlink_port
,
3872 struct mlx4_port_info
,
3874 enum mlx4_port_type mlx4_port_type
;
3876 switch (port_type
) {
3877 case DEVLINK_PORT_TYPE_AUTO
:
3878 mlx4_port_type
= MLX4_PORT_TYPE_AUTO
;
3880 case DEVLINK_PORT_TYPE_ETH
:
3881 mlx4_port_type
= MLX4_PORT_TYPE_ETH
;
3883 case DEVLINK_PORT_TYPE_IB
:
3884 mlx4_port_type
= MLX4_PORT_TYPE_IB
;
3890 return __set_port_type(info
, mlx4_port_type
);
3893 static void mlx4_devlink_param_load_driverinit_values(struct devlink
*devlink
)
3895 struct mlx4_priv
*priv
= devlink_priv(devlink
);
3896 struct mlx4_dev
*dev
= &priv
->dev
;
3897 struct mlx4_fw_crdump
*crdump
= &dev
->persist
->crdump
;
3898 union devlink_param_value saved_value
;
3901 err
= devlink_param_driverinit_value_get(devlink
,
3902 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET
,
3904 if (!err
&& mlx4_internal_err_reset
!= saved_value
.vbool
) {
3905 mlx4_internal_err_reset
= saved_value
.vbool
;
3906 /* Notify on value changed on runtime configuration mode */
3907 devlink_param_value_changed(devlink
,
3908 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET
);
3910 err
= devlink_param_driverinit_value_get(devlink
,
3911 DEVLINK_PARAM_GENERIC_ID_MAX_MACS
,
3914 log_num_mac
= order_base_2(saved_value
.vu32
);
3915 err
= devlink_param_driverinit_value_get(devlink
,
3916 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE
,
3919 enable_64b_cqe_eqe
= saved_value
.vbool
;
3920 err
= devlink_param_driverinit_value_get(devlink
,
3921 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR
,
3924 enable_4k_uar
= saved_value
.vbool
;
3925 err
= devlink_param_driverinit_value_get(devlink
,
3926 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT
,
3928 if (!err
&& crdump
->snapshot_enable
!= saved_value
.vbool
) {
3929 crdump
->snapshot_enable
= saved_value
.vbool
;
3930 devlink_param_value_changed(devlink
,
3931 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT
);
3935 static void mlx4_restart_one_down(struct pci_dev
*pdev
);
3936 static int mlx4_restart_one_up(struct pci_dev
*pdev
, bool reload
,
3937 struct devlink
*devlink
);
3939 static int mlx4_devlink_reload_down(struct devlink
*devlink
, bool netns_change
,
3940 struct netlink_ext_ack
*extack
)
3942 struct mlx4_priv
*priv
= devlink_priv(devlink
);
3943 struct mlx4_dev
*dev
= &priv
->dev
;
3944 struct mlx4_dev_persistent
*persist
= dev
->persist
;
3947 NL_SET_ERR_MSG_MOD(extack
, "Namespace change is not supported");
3950 if (persist
->num_vfs
)
3951 mlx4_warn(persist
->dev
, "Reload performed on PF, will cause reset on operating Virtual Functions\n");
3952 mlx4_restart_one_down(persist
->pdev
);
3956 static int mlx4_devlink_reload_up(struct devlink
*devlink
,
3957 struct netlink_ext_ack
*extack
)
3959 struct mlx4_priv
*priv
= devlink_priv(devlink
);
3960 struct mlx4_dev
*dev
= &priv
->dev
;
3961 struct mlx4_dev_persistent
*persist
= dev
->persist
;
3964 err
= mlx4_restart_one_up(persist
->pdev
, true, devlink
);
3966 mlx4_err(persist
->dev
, "mlx4_restart_one_up failed, ret=%d\n",
3972 static const struct devlink_ops mlx4_devlink_ops
= {
3973 .port_type_set
= mlx4_devlink_port_type_set
,
3974 .reload_down
= mlx4_devlink_reload_down
,
3975 .reload_up
= mlx4_devlink_reload_up
,
3978 static int mlx4_init_one(struct pci_dev
*pdev
, const struct pci_device_id
*id
)
3980 struct devlink
*devlink
;
3981 struct mlx4_priv
*priv
;
3982 struct mlx4_dev
*dev
;
3985 printk_once(KERN_INFO
"%s", mlx4_version
);
3987 devlink
= devlink_alloc(&mlx4_devlink_ops
, sizeof(*priv
));
3990 priv
= devlink_priv(devlink
);
3993 dev
->persist
= kzalloc(sizeof(*dev
->persist
), GFP_KERNEL
);
3994 if (!dev
->persist
) {
3996 goto err_devlink_free
;
3998 dev
->persist
->pdev
= pdev
;
3999 dev
->persist
->dev
= dev
;
4000 pci_set_drvdata(pdev
, dev
->persist
);
4001 priv
->pci_dev_data
= id
->driver_data
;
4002 mutex_init(&dev
->persist
->device_state_mutex
);
4003 mutex_init(&dev
->persist
->interface_state_mutex
);
4004 mutex_init(&dev
->persist
->pci_status_mutex
);
4006 ret
= devlink_register(devlink
, &pdev
->dev
);
4008 goto err_persist_free
;
4009 ret
= devlink_params_register(devlink
, mlx4_devlink_params
,
4010 ARRAY_SIZE(mlx4_devlink_params
));
4012 goto err_devlink_unregister
;
4013 mlx4_devlink_set_params_init_values(devlink
);
4014 ret
= __mlx4_init_one(pdev
, id
->driver_data
, priv
);
4016 goto err_params_unregister
;
4018 devlink_params_publish(devlink
);
4019 devlink_reload_enable(devlink
);
4020 pci_save_state(pdev
);
4023 err_params_unregister
:
4024 devlink_params_unregister(devlink
, mlx4_devlink_params
,
4025 ARRAY_SIZE(mlx4_devlink_params
));
4026 err_devlink_unregister
:
4027 devlink_unregister(devlink
);
4029 kfree(dev
->persist
);
4031 devlink_free(devlink
);
4035 static void mlx4_clean_dev(struct mlx4_dev
*dev
)
4037 struct mlx4_dev_persistent
*persist
= dev
->persist
;
4038 struct mlx4_priv
*priv
= mlx4_priv(dev
);
4039 unsigned long flags
= (dev
->flags
& RESET_PERSIST_MASK_FLAGS
);
4041 memset(priv
, 0, sizeof(*priv
));
4042 priv
->dev
.persist
= persist
;
4043 priv
->dev
.flags
= flags
;
4046 static void mlx4_unload_one(struct pci_dev
*pdev
)
4048 struct mlx4_dev_persistent
*persist
= pci_get_drvdata(pdev
);
4049 struct mlx4_dev
*dev
= persist
->dev
;
4050 struct mlx4_priv
*priv
= mlx4_priv(dev
);
4057 /* saving current ports type for further use */
4058 for (i
= 0; i
< dev
->caps
.num_ports
; i
++) {
4059 dev
->persist
->curr_port_type
[i
] = dev
->caps
.port_type
[i
+ 1];
4060 dev
->persist
->curr_port_poss_type
[i
] = dev
->caps
.
4061 possible_type
[i
+ 1];
4064 pci_dev_data
= priv
->pci_dev_data
;
4066 mlx4_stop_sense(dev
);
4067 mlx4_unregister_device(dev
);
4069 for (p
= 1; p
<= dev
->caps
.num_ports
; p
++) {
4070 mlx4_cleanup_port_info(&priv
->port
[p
]);
4071 mlx4_CLOSE_PORT(dev
, p
);
4074 if (mlx4_is_master(dev
))
4075 mlx4_free_resource_tracker(dev
,
4076 RES_TR_FREE_SLAVES_ONLY
);
4078 mlx4_cleanup_default_counters(dev
);
4079 if (!mlx4_is_slave(dev
))
4080 mlx4_cleanup_counters_table(dev
);
4081 mlx4_cleanup_qp_table(dev
);
4082 mlx4_cleanup_srq_table(dev
);
4083 mlx4_cleanup_cq_table(dev
);
4084 mlx4_cmd_use_polling(dev
);
4085 mlx4_cleanup_eq_table(dev
);
4086 mlx4_cleanup_mcg_table(dev
);
4087 mlx4_cleanup_mr_table(dev
);
4088 mlx4_cleanup_xrcd_table(dev
);
4089 mlx4_cleanup_pd_table(dev
);
4091 if (mlx4_is_master(dev
))
4092 mlx4_free_resource_tracker(dev
,
4093 RES_TR_FREE_STRUCTS_ONLY
);
4096 mlx4_uar_free(dev
, &priv
->driver_uar
);
4097 mlx4_cleanup_uar_table(dev
);
4098 if (!mlx4_is_slave(dev
))
4099 mlx4_clear_steering(dev
);
4100 mlx4_free_eq_table(dev
);
4101 if (mlx4_is_master(dev
))
4102 mlx4_multi_func_cleanup(dev
);
4103 mlx4_close_hca(dev
);
4105 if (mlx4_is_slave(dev
))
4106 mlx4_multi_func_cleanup(dev
);
4107 mlx4_cmd_cleanup(dev
, MLX4_CMD_CLEANUP_ALL
);
4109 if (dev
->flags
& MLX4_FLAG_MSI_X
)
4110 pci_disable_msix(pdev
);
4112 if (!mlx4_is_slave(dev
))
4113 mlx4_free_ownership(dev
);
4115 mlx4_slave_destroy_special_qp_cap(dev
);
4116 kfree(dev
->dev_vfs
);
4118 mlx4_clean_dev(dev
);
4119 priv
->pci_dev_data
= pci_dev_data
;
4123 static void mlx4_remove_one(struct pci_dev
*pdev
)
4125 struct mlx4_dev_persistent
*persist
= pci_get_drvdata(pdev
);
4126 struct mlx4_dev
*dev
= persist
->dev
;
4127 struct mlx4_priv
*priv
= mlx4_priv(dev
);
4128 struct devlink
*devlink
= priv_to_devlink(priv
);
4131 devlink_reload_disable(devlink
);
4133 if (mlx4_is_slave(dev
))
4134 persist
->interface_state
|= MLX4_INTERFACE_STATE_NOWAIT
;
4136 mutex_lock(&persist
->interface_state_mutex
);
4137 persist
->interface_state
|= MLX4_INTERFACE_STATE_DELETION
;
4138 mutex_unlock(&persist
->interface_state_mutex
);
4140 /* Disabling SR-IOV is not allowed while there are active vf's */
4141 if (mlx4_is_master(dev
) && dev
->flags
& MLX4_FLAG_SRIOV
) {
4142 active_vfs
= mlx4_how_many_lives_vf(dev
);
4144 pr_warn("Removing PF when there are active VF's !!\n");
4145 pr_warn("Will not disable SR-IOV.\n");
4149 /* device marked to be under deletion running now without the lock
4150 * letting other tasks to be terminated
4152 if (persist
->interface_state
& MLX4_INTERFACE_STATE_UP
)
4153 mlx4_unload_one(pdev
);
4155 mlx4_info(dev
, "%s: interface is down\n", __func__
);
4156 mlx4_catas_end(dev
);
4157 mlx4_crdump_end(dev
);
4158 if (dev
->flags
& MLX4_FLAG_SRIOV
&& !active_vfs
) {
4159 mlx4_warn(dev
, "Disabling SR-IOV\n");
4160 pci_disable_sriov(pdev
);
4163 pci_release_regions(pdev
);
4164 mlx4_pci_disable_device(dev
);
4165 devlink_params_unregister(devlink
, mlx4_devlink_params
,
4166 ARRAY_SIZE(mlx4_devlink_params
));
4167 devlink_unregister(devlink
);
4168 kfree(dev
->persist
);
4169 devlink_free(devlink
);
4172 static int restore_current_port_types(struct mlx4_dev
*dev
,
4173 enum mlx4_port_type
*types
,
4174 enum mlx4_port_type
*poss_types
)
4176 struct mlx4_priv
*priv
= mlx4_priv(dev
);
4179 mlx4_stop_sense(dev
);
4181 mutex_lock(&priv
->port_mutex
);
4182 for (i
= 0; i
< dev
->caps
.num_ports
; i
++)
4183 dev
->caps
.possible_type
[i
+ 1] = poss_types
[i
];
4184 err
= mlx4_change_port_types(dev
, types
);
4185 mlx4_start_sense(dev
);
4186 mutex_unlock(&priv
->port_mutex
);
4191 static void mlx4_restart_one_down(struct pci_dev
*pdev
)
4193 mlx4_unload_one(pdev
);
4196 static int mlx4_restart_one_up(struct pci_dev
*pdev
, bool reload
,
4197 struct devlink
*devlink
)
4199 struct mlx4_dev_persistent
*persist
= pci_get_drvdata(pdev
);
4200 struct mlx4_dev
*dev
= persist
->dev
;
4201 struct mlx4_priv
*priv
= mlx4_priv(dev
);
4202 int nvfs
[MLX4_MAX_PORTS
+ 1] = {0, 0, 0};
4203 int pci_dev_data
, err
, total_vfs
;
4205 pci_dev_data
= priv
->pci_dev_data
;
4206 total_vfs
= dev
->persist
->num_vfs
;
4207 memcpy(nvfs
, dev
->persist
->nvfs
, sizeof(dev
->persist
->nvfs
));
4210 mlx4_devlink_param_load_driverinit_values(devlink
);
4211 err
= mlx4_load_one(pdev
, pci_dev_data
, total_vfs
, nvfs
, priv
, 1);
4213 mlx4_err(dev
, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
4214 __func__
, pci_name(pdev
), err
);
4218 err
= restore_current_port_types(dev
, dev
->persist
->curr_port_type
,
4219 dev
->persist
->curr_port_poss_type
);
4221 mlx4_err(dev
, "could not restore original port types (%d)\n",
4227 int mlx4_restart_one(struct pci_dev
*pdev
)
4229 mlx4_restart_one_down(pdev
);
4230 return mlx4_restart_one_up(pdev
, false, NULL
);
4233 #define MLX_SP(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_FORCE_SENSE_PORT }
4234 #define MLX_VF(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_IS_VF }
4235 #define MLX_GN(id) { PCI_VDEVICE(MELLANOX, id), 0 }
4237 static const struct pci_device_id mlx4_pci_table
[] = {
4238 #ifdef CONFIG_MLX4_CORE_GEN2
4239 /* MT25408 "Hermon" */
4240 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_SDR
), /* SDR */
4241 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR
), /* DDR */
4242 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR
), /* QDR */
4243 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR_GEN2
), /* DDR Gen2 */
4244 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR_GEN2
), /* QDR Gen2 */
4245 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN
), /* EN 10GigE */
4246 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN_GEN2
), /* EN 10GigE Gen2 */
4247 /* MT25458 ConnectX EN 10GBASE-T */
4248 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN
),
4249 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_T_GEN2
), /* Gen2 */
4250 /* MT26468 ConnectX EN 10GigE PCIe Gen2*/
4251 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_GEN2
),
4252 /* MT26438 ConnectX EN 40GigE PCIe Gen2 5GT/s */
4253 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_5_GEN2
),
4254 /* MT26478 ConnectX2 40GigE PCIe Gen2 */
4255 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX2
),
4256 /* MT25400 Family [ConnectX-2] */
4257 MLX_VF(0x1002), /* Virtual Function */
4258 #endif /* CONFIG_MLX4_CORE_GEN2 */
4259 /* MT27500 Family [ConnectX-3] */
4260 MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3
),
4261 MLX_VF(0x1004), /* Virtual Function */
4262 MLX_GN(0x1005), /* MT27510 Family */
4263 MLX_GN(0x1006), /* MT27511 Family */
4264 MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3_PRO
), /* MT27520 Family */
4265 MLX_GN(0x1008), /* MT27521 Family */
4266 MLX_GN(0x1009), /* MT27530 Family */
4267 MLX_GN(0x100a), /* MT27531 Family */
4268 MLX_GN(0x100b), /* MT27540 Family */
4269 MLX_GN(0x100c), /* MT27541 Family */
4270 MLX_GN(0x100d), /* MT27550 Family */
4271 MLX_GN(0x100e), /* MT27551 Family */
4272 MLX_GN(0x100f), /* MT27560 Family */
4273 MLX_GN(0x1010), /* MT27561 Family */
4276 * See the mellanox_check_broken_intx_masking() quirk when
4283 MODULE_DEVICE_TABLE(pci
, mlx4_pci_table
);
4285 static pci_ers_result_t
mlx4_pci_err_detected(struct pci_dev
*pdev
,
4286 pci_channel_state_t state
)
4288 struct mlx4_dev_persistent
*persist
= pci_get_drvdata(pdev
);
4290 mlx4_err(persist
->dev
, "mlx4_pci_err_detected was called\n");
4291 mlx4_enter_error_state(persist
);
4293 mutex_lock(&persist
->interface_state_mutex
);
4294 if (persist
->interface_state
& MLX4_INTERFACE_STATE_UP
)
4295 mlx4_unload_one(pdev
);
4297 mutex_unlock(&persist
->interface_state_mutex
);
4298 if (state
== pci_channel_io_perm_failure
)
4299 return PCI_ERS_RESULT_DISCONNECT
;
4301 mlx4_pci_disable_device(persist
->dev
);
4302 return PCI_ERS_RESULT_NEED_RESET
;
4305 static pci_ers_result_t
mlx4_pci_slot_reset(struct pci_dev
*pdev
)
4307 struct mlx4_dev_persistent
*persist
= pci_get_drvdata(pdev
);
4308 struct mlx4_dev
*dev
= persist
->dev
;
4311 mlx4_err(dev
, "mlx4_pci_slot_reset was called\n");
4312 err
= mlx4_pci_enable_device(dev
);
4314 mlx4_err(dev
, "Can not re-enable device, err=%d\n", err
);
4315 return PCI_ERS_RESULT_DISCONNECT
;
4318 pci_set_master(pdev
);
4319 pci_restore_state(pdev
);
4320 pci_save_state(pdev
);
4321 return PCI_ERS_RESULT_RECOVERED
;
4324 static void mlx4_pci_resume(struct pci_dev
*pdev
)
4326 struct mlx4_dev_persistent
*persist
= pci_get_drvdata(pdev
);
4327 struct mlx4_dev
*dev
= persist
->dev
;
4328 struct mlx4_priv
*priv
= mlx4_priv(dev
);
4329 int nvfs
[MLX4_MAX_PORTS
+ 1] = {0, 0, 0};
4333 mlx4_err(dev
, "%s was called\n", __func__
);
4334 total_vfs
= dev
->persist
->num_vfs
;
4335 memcpy(nvfs
, dev
->persist
->nvfs
, sizeof(dev
->persist
->nvfs
));
4337 mutex_lock(&persist
->interface_state_mutex
);
4338 if (!(persist
->interface_state
& MLX4_INTERFACE_STATE_UP
)) {
4339 err
= mlx4_load_one(pdev
, priv
->pci_dev_data
, total_vfs
, nvfs
,
4342 mlx4_err(dev
, "%s: mlx4_load_one failed, err=%d\n",
4347 err
= restore_current_port_types(dev
, dev
->persist
->
4348 curr_port_type
, dev
->persist
->
4349 curr_port_poss_type
);
4351 mlx4_err(dev
, "could not restore original port types (%d)\n", err
);
4354 mutex_unlock(&persist
->interface_state_mutex
);
4358 static void mlx4_shutdown(struct pci_dev
*pdev
)
4360 struct mlx4_dev_persistent
*persist
= pci_get_drvdata(pdev
);
4362 mlx4_info(persist
->dev
, "mlx4_shutdown was called\n");
4363 mutex_lock(&persist
->interface_state_mutex
);
4364 if (persist
->interface_state
& MLX4_INTERFACE_STATE_UP
)
4365 mlx4_unload_one(pdev
);
4366 mutex_unlock(&persist
->interface_state_mutex
);
4369 static const struct pci_error_handlers mlx4_err_handler
= {
4370 .error_detected
= mlx4_pci_err_detected
,
4371 .slot_reset
= mlx4_pci_slot_reset
,
4372 .resume
= mlx4_pci_resume
,
4375 static int mlx4_suspend(struct pci_dev
*pdev
, pm_message_t state
)
4377 struct mlx4_dev_persistent
*persist
= pci_get_drvdata(pdev
);
4378 struct mlx4_dev
*dev
= persist
->dev
;
4380 mlx4_err(dev
, "suspend was called\n");
4381 mutex_lock(&persist
->interface_state_mutex
);
4382 if (persist
->interface_state
& MLX4_INTERFACE_STATE_UP
)
4383 mlx4_unload_one(pdev
);
4384 mutex_unlock(&persist
->interface_state_mutex
);
4389 static int mlx4_resume(struct pci_dev
*pdev
)
4391 struct mlx4_dev_persistent
*persist
= pci_get_drvdata(pdev
);
4392 struct mlx4_dev
*dev
= persist
->dev
;
4393 struct mlx4_priv
*priv
= mlx4_priv(dev
);
4394 int nvfs
[MLX4_MAX_PORTS
+ 1] = {0, 0, 0};
4398 mlx4_err(dev
, "resume was called\n");
4399 total_vfs
= dev
->persist
->num_vfs
;
4400 memcpy(nvfs
, dev
->persist
->nvfs
, sizeof(dev
->persist
->nvfs
));
4402 mutex_lock(&persist
->interface_state_mutex
);
4403 if (!(persist
->interface_state
& MLX4_INTERFACE_STATE_UP
)) {
4404 ret
= mlx4_load_one(pdev
, priv
->pci_dev_data
, total_vfs
,
4407 ret
= restore_current_port_types(dev
,
4408 dev
->persist
->curr_port_type
,
4409 dev
->persist
->curr_port_poss_type
);
4411 mlx4_err(dev
, "resume: could not restore original port types (%d)\n", ret
);
4414 mutex_unlock(&persist
->interface_state_mutex
);
4419 static struct pci_driver mlx4_driver
= {
4421 .id_table
= mlx4_pci_table
,
4422 .probe
= mlx4_init_one
,
4423 .shutdown
= mlx4_shutdown
,
4424 .remove
= mlx4_remove_one
,
4425 .suspend
= mlx4_suspend
,
4426 .resume
= mlx4_resume
,
4427 .err_handler
= &mlx4_err_handler
,
4430 static int __init
mlx4_verify_params(void)
4433 pr_warn("mlx4_core: bad msi_x: %d\n", msi_x
);
4437 if ((log_num_mac
< 0) || (log_num_mac
> 7)) {
4438 pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac
);
4442 if (log_num_vlan
!= 0)
4443 pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
4444 MLX4_LOG_NUM_VLANS
);
4447 pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
4449 if ((log_mtts_per_seg
< 0) || (log_mtts_per_seg
> 7)) {
4450 pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
4455 /* Check if module param for ports type has legal combination */
4456 if (port_type_array
[0] == false && port_type_array
[1] == true) {
4457 pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
4458 port_type_array
[0] = true;
4461 if (mlx4_log_num_mgm_entry_size
< -7 ||
4462 (mlx4_log_num_mgm_entry_size
> 0 &&
4463 (mlx4_log_num_mgm_entry_size
< MLX4_MIN_MGM_LOG_ENTRY_SIZE
||
4464 mlx4_log_num_mgm_entry_size
> MLX4_MAX_MGM_LOG_ENTRY_SIZE
))) {
4465 pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
4466 mlx4_log_num_mgm_entry_size
,
4467 MLX4_MIN_MGM_LOG_ENTRY_SIZE
,
4468 MLX4_MAX_MGM_LOG_ENTRY_SIZE
);
4475 static int __init
mlx4_init(void)
4479 if (mlx4_verify_params())
4483 mlx4_wq
= create_singlethread_workqueue("mlx4");
4487 ret
= pci_register_driver(&mlx4_driver
);
4489 destroy_workqueue(mlx4_wq
);
4490 return ret
< 0 ? ret
: 0;
4493 static void __exit
mlx4_cleanup(void)
4495 pci_unregister_driver(&mlx4_driver
);
4496 destroy_workqueue(mlx4_wq
);
4499 module_init(mlx4_init
);
4500 module_exit(mlx4_cleanup
);