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/init.h>
38 #include <linux/errno.h>
39 #include <linux/pci.h>
40 #include <linux/dma-mapping.h>
41 #include <linux/slab.h>
43 #include <linux/mlx4/device.h>
44 #include <linux/mlx4/doorbell.h>
50 MODULE_AUTHOR("Roland Dreier");
51 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
52 MODULE_LICENSE("Dual BSD/GPL");
53 MODULE_VERSION(DRV_VERSION
);
55 struct workqueue_struct
*mlx4_wq
;
57 #ifdef CONFIG_MLX4_DEBUG
59 int mlx4_debug_level
= 0;
60 module_param_named(debug_level
, mlx4_debug_level
, int, 0644);
61 MODULE_PARM_DESC(debug_level
, "Enable debug tracing if > 0");
63 #endif /* CONFIG_MLX4_DEBUG */
68 module_param(msi_x
, int, 0444);
69 MODULE_PARM_DESC(msi_x
, "attempt to use MSI-X if nonzero");
71 #else /* CONFIG_PCI_MSI */
75 #endif /* CONFIG_PCI_MSI */
77 static char mlx4_version
[] __devinitdata
=
78 DRV_NAME
": Mellanox ConnectX core driver v"
79 DRV_VERSION
" (" DRV_RELDATE
")\n";
81 static struct mlx4_profile default_profile
= {
84 .rdmarc_per_qp
= 1 << 4,
91 static int log_num_mac
= 2;
92 module_param_named(log_num_mac
, log_num_mac
, int, 0444);
93 MODULE_PARM_DESC(log_num_mac
, "Log2 max number of MACs per ETH port (1-7)");
95 static int log_num_vlan
;
96 module_param_named(log_num_vlan
, log_num_vlan
, int, 0444);
97 MODULE_PARM_DESC(log_num_vlan
, "Log2 max number of VLANs per ETH port (0-7)");
100 module_param_named(use_prio
, use_prio
, bool, 0444);
101 MODULE_PARM_DESC(use_prio
, "Enable steering by VLAN priority on ETH ports "
104 static int log_mtts_per_seg
= ilog2(MLX4_MTT_ENTRY_PER_SEG
);
105 module_param_named(log_mtts_per_seg
, log_mtts_per_seg
, int, 0444);
106 MODULE_PARM_DESC(log_mtts_per_seg
, "Log2 number of MTT entries per segment (1-5)");
108 int mlx4_check_port_params(struct mlx4_dev
*dev
,
109 enum mlx4_port_type
*port_type
)
113 for (i
= 0; i
< dev
->caps
.num_ports
- 1; i
++) {
114 if (port_type
[i
] != port_type
[i
+ 1]) {
115 if (!(dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_DPDP
)) {
116 mlx4_err(dev
, "Only same port types supported "
117 "on this HCA, aborting.\n");
120 if (port_type
[i
] == MLX4_PORT_TYPE_ETH
&&
121 port_type
[i
+ 1] == MLX4_PORT_TYPE_IB
)
126 for (i
= 0; i
< dev
->caps
.num_ports
; i
++) {
127 if (!(port_type
[i
] & dev
->caps
.supported_type
[i
+1])) {
128 mlx4_err(dev
, "Requested port type for port %d is not "
129 "supported on this HCA\n", i
+ 1);
136 static void mlx4_set_port_mask(struct mlx4_dev
*dev
)
140 dev
->caps
.port_mask
= 0;
141 for (i
= 1; i
<= dev
->caps
.num_ports
; ++i
)
142 if (dev
->caps
.port_type
[i
] == MLX4_PORT_TYPE_IB
)
143 dev
->caps
.port_mask
|= 1 << (i
- 1);
145 static int mlx4_dev_cap(struct mlx4_dev
*dev
, struct mlx4_dev_cap
*dev_cap
)
150 err
= mlx4_QUERY_DEV_CAP(dev
, dev_cap
);
152 mlx4_err(dev
, "QUERY_DEV_CAP command failed, aborting.\n");
156 if (dev_cap
->min_page_sz
> PAGE_SIZE
) {
157 mlx4_err(dev
, "HCA minimum page size of %d bigger than "
158 "kernel PAGE_SIZE of %ld, aborting.\n",
159 dev_cap
->min_page_sz
, PAGE_SIZE
);
162 if (dev_cap
->num_ports
> MLX4_MAX_PORTS
) {
163 mlx4_err(dev
, "HCA has %d ports, but we only support %d, "
165 dev_cap
->num_ports
, MLX4_MAX_PORTS
);
169 if (dev_cap
->uar_size
> pci_resource_len(dev
->pdev
, 2)) {
170 mlx4_err(dev
, "HCA reported UAR size of 0x%x bigger than "
171 "PCI resource 2 size of 0x%llx, aborting.\n",
173 (unsigned long long) pci_resource_len(dev
->pdev
, 2));
177 dev
->caps
.num_ports
= dev_cap
->num_ports
;
178 for (i
= 1; i
<= dev
->caps
.num_ports
; ++i
) {
179 dev
->caps
.vl_cap
[i
] = dev_cap
->max_vl
[i
];
180 dev
->caps
.ib_mtu_cap
[i
] = dev_cap
->ib_mtu
[i
];
181 dev
->caps
.gid_table_len
[i
] = dev_cap
->max_gids
[i
];
182 dev
->caps
.pkey_table_len
[i
] = dev_cap
->max_pkeys
[i
];
183 dev
->caps
.port_width_cap
[i
] = dev_cap
->max_port_width
[i
];
184 dev
->caps
.eth_mtu_cap
[i
] = dev_cap
->eth_mtu
[i
];
185 dev
->caps
.def_mac
[i
] = dev_cap
->def_mac
[i
];
186 dev
->caps
.supported_type
[i
] = dev_cap
->supported_port_types
[i
];
189 dev
->caps
.num_uars
= dev_cap
->uar_size
/ PAGE_SIZE
;
190 dev
->caps
.local_ca_ack_delay
= dev_cap
->local_ca_ack_delay
;
191 dev
->caps
.bf_reg_size
= dev_cap
->bf_reg_size
;
192 dev
->caps
.bf_regs_per_page
= dev_cap
->bf_regs_per_page
;
193 dev
->caps
.max_sq_sg
= dev_cap
->max_sq_sg
;
194 dev
->caps
.max_rq_sg
= dev_cap
->max_rq_sg
;
195 dev
->caps
.max_wqes
= dev_cap
->max_qp_sz
;
196 dev
->caps
.max_qp_init_rdma
= dev_cap
->max_requester_per_qp
;
197 dev
->caps
.max_srq_wqes
= dev_cap
->max_srq_sz
;
198 dev
->caps
.max_srq_sge
= dev_cap
->max_rq_sg
- 1;
199 dev
->caps
.reserved_srqs
= dev_cap
->reserved_srqs
;
200 dev
->caps
.max_sq_desc_sz
= dev_cap
->max_sq_desc_sz
;
201 dev
->caps
.max_rq_desc_sz
= dev_cap
->max_rq_desc_sz
;
202 dev
->caps
.num_qp_per_mgm
= MLX4_QP_PER_MGM
;
204 * Subtract 1 from the limit because we need to allocate a
205 * spare CQE so the HCA HW can tell the difference between an
206 * empty CQ and a full CQ.
208 dev
->caps
.max_cqes
= dev_cap
->max_cq_sz
- 1;
209 dev
->caps
.reserved_cqs
= dev_cap
->reserved_cqs
;
210 dev
->caps
.reserved_eqs
= dev_cap
->reserved_eqs
;
211 dev
->caps
.mtts_per_seg
= 1 << log_mtts_per_seg
;
212 dev
->caps
.reserved_mtts
= DIV_ROUND_UP(dev_cap
->reserved_mtts
,
213 dev
->caps
.mtts_per_seg
);
214 dev
->caps
.reserved_mrws
= dev_cap
->reserved_mrws
;
215 dev
->caps
.reserved_uars
= dev_cap
->reserved_uars
;
216 dev
->caps
.reserved_pds
= dev_cap
->reserved_pds
;
217 dev
->caps
.mtt_entry_sz
= dev
->caps
.mtts_per_seg
* dev_cap
->mtt_entry_sz
;
218 dev
->caps
.max_msg_sz
= dev_cap
->max_msg_sz
;
219 dev
->caps
.page_size_cap
= ~(u32
) (dev_cap
->min_page_sz
- 1);
220 dev
->caps
.flags
= dev_cap
->flags
;
221 dev
->caps
.bmme_flags
= dev_cap
->bmme_flags
;
222 dev
->caps
.reserved_lkey
= dev_cap
->reserved_lkey
;
223 dev
->caps
.stat_rate_support
= dev_cap
->stat_rate_support
;
224 dev
->caps
.max_gso_sz
= dev_cap
->max_gso_sz
;
226 dev
->caps
.log_num_macs
= log_num_mac
;
227 dev
->caps
.log_num_vlans
= log_num_vlan
;
228 dev
->caps
.log_num_prios
= use_prio
? 3 : 0;
230 for (i
= 1; i
<= dev
->caps
.num_ports
; ++i
) {
231 if (dev
->caps
.supported_type
[i
] != MLX4_PORT_TYPE_ETH
)
232 dev
->caps
.port_type
[i
] = MLX4_PORT_TYPE_IB
;
234 dev
->caps
.port_type
[i
] = MLX4_PORT_TYPE_ETH
;
235 dev
->caps
.possible_type
[i
] = dev
->caps
.port_type
[i
];
236 mlx4_priv(dev
)->sense
.sense_allowed
[i
] =
237 dev
->caps
.supported_type
[i
] == MLX4_PORT_TYPE_AUTO
;
239 if (dev
->caps
.log_num_macs
> dev_cap
->log_max_macs
[i
]) {
240 dev
->caps
.log_num_macs
= dev_cap
->log_max_macs
[i
];
241 mlx4_warn(dev
, "Requested number of MACs is too much "
242 "for port %d, reducing to %d.\n",
243 i
, 1 << dev
->caps
.log_num_macs
);
245 if (dev
->caps
.log_num_vlans
> dev_cap
->log_max_vlans
[i
]) {
246 dev
->caps
.log_num_vlans
= dev_cap
->log_max_vlans
[i
];
247 mlx4_warn(dev
, "Requested number of VLANs is too much "
248 "for port %d, reducing to %d.\n",
249 i
, 1 << dev
->caps
.log_num_vlans
);
253 mlx4_set_port_mask(dev
);
255 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
] = dev_cap
->reserved_qps
;
256 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_ETH_ADDR
] =
257 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FC_ADDR
] =
258 (1 << dev
->caps
.log_num_macs
) *
259 (1 << dev
->caps
.log_num_vlans
) *
260 (1 << dev
->caps
.log_num_prios
) *
262 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FC_EXCH
] = MLX4_NUM_FEXCH
;
264 dev
->caps
.reserved_qps
= dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
] +
265 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_ETH_ADDR
] +
266 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FC_ADDR
] +
267 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FC_EXCH
];
273 * Change the port configuration of the device.
274 * Every user of this function must hold the port mutex.
276 int mlx4_change_port_types(struct mlx4_dev
*dev
,
277 enum mlx4_port_type
*port_types
)
283 for (port
= 0; port
< dev
->caps
.num_ports
; port
++) {
284 /* Change the port type only if the new type is different
285 * from the current, and not set to Auto */
286 if (port_types
[port
] != dev
->caps
.port_type
[port
+ 1]) {
288 dev
->caps
.port_type
[port
+ 1] = port_types
[port
];
292 mlx4_unregister_device(dev
);
293 for (port
= 1; port
<= dev
->caps
.num_ports
; port
++) {
294 mlx4_CLOSE_PORT(dev
, port
);
295 err
= mlx4_SET_PORT(dev
, port
);
297 mlx4_err(dev
, "Failed to set port %d, "
302 mlx4_set_port_mask(dev
);
303 err
= mlx4_register_device(dev
);
310 static ssize_t
show_port_type(struct device
*dev
,
311 struct device_attribute
*attr
,
314 struct mlx4_port_info
*info
= container_of(attr
, struct mlx4_port_info
,
316 struct mlx4_dev
*mdev
= info
->dev
;
320 (mdev
->caps
.port_type
[info
->port
] == MLX4_PORT_TYPE_IB
) ?
322 if (mdev
->caps
.possible_type
[info
->port
] == MLX4_PORT_TYPE_AUTO
)
323 sprintf(buf
, "auto (%s)\n", type
);
325 sprintf(buf
, "%s\n", type
);
330 static ssize_t
set_port_type(struct device
*dev
,
331 struct device_attribute
*attr
,
332 const char *buf
, size_t count
)
334 struct mlx4_port_info
*info
= container_of(attr
, struct mlx4_port_info
,
336 struct mlx4_dev
*mdev
= info
->dev
;
337 struct mlx4_priv
*priv
= mlx4_priv(mdev
);
338 enum mlx4_port_type types
[MLX4_MAX_PORTS
];
339 enum mlx4_port_type new_types
[MLX4_MAX_PORTS
];
343 if (!strcmp(buf
, "ib\n"))
344 info
->tmp_type
= MLX4_PORT_TYPE_IB
;
345 else if (!strcmp(buf
, "eth\n"))
346 info
->tmp_type
= MLX4_PORT_TYPE_ETH
;
347 else if (!strcmp(buf
, "auto\n"))
348 info
->tmp_type
= MLX4_PORT_TYPE_AUTO
;
350 mlx4_err(mdev
, "%s is not supported port type\n", buf
);
354 mlx4_stop_sense(mdev
);
355 mutex_lock(&priv
->port_mutex
);
356 /* Possible type is always the one that was delivered */
357 mdev
->caps
.possible_type
[info
->port
] = info
->tmp_type
;
359 for (i
= 0; i
< mdev
->caps
.num_ports
; i
++) {
360 types
[i
] = priv
->port
[i
+1].tmp_type
? priv
->port
[i
+1].tmp_type
:
361 mdev
->caps
.possible_type
[i
+1];
362 if (types
[i
] == MLX4_PORT_TYPE_AUTO
)
363 types
[i
] = mdev
->caps
.port_type
[i
+1];
366 if (!(mdev
->caps
.flags
& MLX4_DEV_CAP_FLAG_DPDP
)) {
367 for (i
= 1; i
<= mdev
->caps
.num_ports
; i
++) {
368 if (mdev
->caps
.possible_type
[i
] == MLX4_PORT_TYPE_AUTO
) {
369 mdev
->caps
.possible_type
[i
] = mdev
->caps
.port_type
[i
];
375 mlx4_err(mdev
, "Auto sensing is not supported on this HCA. "
376 "Set only 'eth' or 'ib' for both ports "
377 "(should be the same)\n");
381 mlx4_do_sense_ports(mdev
, new_types
, types
);
383 err
= mlx4_check_port_params(mdev
, new_types
);
387 /* We are about to apply the changes after the configuration
388 * was verified, no need to remember the temporary types
390 for (i
= 0; i
< mdev
->caps
.num_ports
; i
++)
391 priv
->port
[i
+ 1].tmp_type
= 0;
393 err
= mlx4_change_port_types(mdev
, new_types
);
396 mlx4_start_sense(mdev
);
397 mutex_unlock(&priv
->port_mutex
);
398 return err
? err
: count
;
401 static int mlx4_load_fw(struct mlx4_dev
*dev
)
403 struct mlx4_priv
*priv
= mlx4_priv(dev
);
406 priv
->fw
.fw_icm
= mlx4_alloc_icm(dev
, priv
->fw
.fw_pages
,
407 GFP_HIGHUSER
| __GFP_NOWARN
, 0);
408 if (!priv
->fw
.fw_icm
) {
409 mlx4_err(dev
, "Couldn't allocate FW area, aborting.\n");
413 err
= mlx4_MAP_FA(dev
, priv
->fw
.fw_icm
);
415 mlx4_err(dev
, "MAP_FA command failed, aborting.\n");
419 err
= mlx4_RUN_FW(dev
);
421 mlx4_err(dev
, "RUN_FW command failed, aborting.\n");
431 mlx4_free_icm(dev
, priv
->fw
.fw_icm
, 0);
435 static int mlx4_init_cmpt_table(struct mlx4_dev
*dev
, u64 cmpt_base
,
438 struct mlx4_priv
*priv
= mlx4_priv(dev
);
441 err
= mlx4_init_icm_table(dev
, &priv
->qp_table
.cmpt_table
,
443 ((u64
) (MLX4_CMPT_TYPE_QP
*
444 cmpt_entry_sz
) << MLX4_CMPT_SHIFT
),
445 cmpt_entry_sz
, dev
->caps
.num_qps
,
446 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
],
451 err
= mlx4_init_icm_table(dev
, &priv
->srq_table
.cmpt_table
,
453 ((u64
) (MLX4_CMPT_TYPE_SRQ
*
454 cmpt_entry_sz
) << MLX4_CMPT_SHIFT
),
455 cmpt_entry_sz
, dev
->caps
.num_srqs
,
456 dev
->caps
.reserved_srqs
, 0, 0);
460 err
= mlx4_init_icm_table(dev
, &priv
->cq_table
.cmpt_table
,
462 ((u64
) (MLX4_CMPT_TYPE_CQ
*
463 cmpt_entry_sz
) << MLX4_CMPT_SHIFT
),
464 cmpt_entry_sz
, dev
->caps
.num_cqs
,
465 dev
->caps
.reserved_cqs
, 0, 0);
469 err
= mlx4_init_icm_table(dev
, &priv
->eq_table
.cmpt_table
,
471 ((u64
) (MLX4_CMPT_TYPE_EQ
*
472 cmpt_entry_sz
) << MLX4_CMPT_SHIFT
),
474 dev
->caps
.num_eqs
, dev
->caps
.num_eqs
, 0, 0);
481 mlx4_cleanup_icm_table(dev
, &priv
->cq_table
.cmpt_table
);
484 mlx4_cleanup_icm_table(dev
, &priv
->srq_table
.cmpt_table
);
487 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.cmpt_table
);
493 static int mlx4_init_icm(struct mlx4_dev
*dev
, struct mlx4_dev_cap
*dev_cap
,
494 struct mlx4_init_hca_param
*init_hca
, u64 icm_size
)
496 struct mlx4_priv
*priv
= mlx4_priv(dev
);
500 err
= mlx4_SET_ICM_SIZE(dev
, icm_size
, &aux_pages
);
502 mlx4_err(dev
, "SET_ICM_SIZE command failed, aborting.\n");
506 mlx4_dbg(dev
, "%lld KB of HCA context requires %lld KB aux memory.\n",
507 (unsigned long long) icm_size
>> 10,
508 (unsigned long long) aux_pages
<< 2);
510 priv
->fw
.aux_icm
= mlx4_alloc_icm(dev
, aux_pages
,
511 GFP_HIGHUSER
| __GFP_NOWARN
, 0);
512 if (!priv
->fw
.aux_icm
) {
513 mlx4_err(dev
, "Couldn't allocate aux memory, aborting.\n");
517 err
= mlx4_MAP_ICM_AUX(dev
, priv
->fw
.aux_icm
);
519 mlx4_err(dev
, "MAP_ICM_AUX command failed, aborting.\n");
523 err
= mlx4_init_cmpt_table(dev
, init_hca
->cmpt_base
, dev_cap
->cmpt_entry_sz
);
525 mlx4_err(dev
, "Failed to map cMPT context memory, aborting.\n");
529 err
= mlx4_init_icm_table(dev
, &priv
->eq_table
.table
,
530 init_hca
->eqc_base
, dev_cap
->eqc_entry_sz
,
531 dev
->caps
.num_eqs
, dev
->caps
.num_eqs
,
534 mlx4_err(dev
, "Failed to map EQ context memory, aborting.\n");
539 * Reserved MTT entries must be aligned up to a cacheline
540 * boundary, since the FW will write to them, while the driver
541 * writes to all other MTT entries. (The variable
542 * dev->caps.mtt_entry_sz below is really the MTT segment
543 * size, not the raw entry size)
545 dev
->caps
.reserved_mtts
=
546 ALIGN(dev
->caps
.reserved_mtts
* dev
->caps
.mtt_entry_sz
,
547 dma_get_cache_alignment()) / dev
->caps
.mtt_entry_sz
;
549 err
= mlx4_init_icm_table(dev
, &priv
->mr_table
.mtt_table
,
551 dev
->caps
.mtt_entry_sz
,
552 dev
->caps
.num_mtt_segs
,
553 dev
->caps
.reserved_mtts
, 1, 0);
555 mlx4_err(dev
, "Failed to map MTT context memory, aborting.\n");
559 err
= mlx4_init_icm_table(dev
, &priv
->mr_table
.dmpt_table
,
561 dev_cap
->dmpt_entry_sz
,
563 dev
->caps
.reserved_mrws
, 1, 1);
565 mlx4_err(dev
, "Failed to map dMPT context memory, aborting.\n");
569 err
= mlx4_init_icm_table(dev
, &priv
->qp_table
.qp_table
,
571 dev_cap
->qpc_entry_sz
,
573 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
],
576 mlx4_err(dev
, "Failed to map QP context memory, aborting.\n");
580 err
= mlx4_init_icm_table(dev
, &priv
->qp_table
.auxc_table
,
582 dev_cap
->aux_entry_sz
,
584 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
],
587 mlx4_err(dev
, "Failed to map AUXC context memory, aborting.\n");
591 err
= mlx4_init_icm_table(dev
, &priv
->qp_table
.altc_table
,
593 dev_cap
->altc_entry_sz
,
595 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
],
598 mlx4_err(dev
, "Failed to map ALTC context memory, aborting.\n");
602 err
= mlx4_init_icm_table(dev
, &priv
->qp_table
.rdmarc_table
,
603 init_hca
->rdmarc_base
,
604 dev_cap
->rdmarc_entry_sz
<< priv
->qp_table
.rdmarc_shift
,
606 dev
->caps
.reserved_qps_cnt
[MLX4_QP_REGION_FW
],
609 mlx4_err(dev
, "Failed to map RDMARC context memory, aborting\n");
613 err
= mlx4_init_icm_table(dev
, &priv
->cq_table
.table
,
615 dev_cap
->cqc_entry_sz
,
617 dev
->caps
.reserved_cqs
, 0, 0);
619 mlx4_err(dev
, "Failed to map CQ context memory, aborting.\n");
620 goto err_unmap_rdmarc
;
623 err
= mlx4_init_icm_table(dev
, &priv
->srq_table
.table
,
625 dev_cap
->srq_entry_sz
,
627 dev
->caps
.reserved_srqs
, 0, 0);
629 mlx4_err(dev
, "Failed to map SRQ context memory, aborting.\n");
634 * It's not strictly required, but for simplicity just map the
635 * whole multicast group table now. The table isn't very big
636 * and it's a lot easier than trying to track ref counts.
638 err
= mlx4_init_icm_table(dev
, &priv
->mcg_table
.table
,
639 init_hca
->mc_base
, MLX4_MGM_ENTRY_SIZE
,
640 dev
->caps
.num_mgms
+ dev
->caps
.num_amgms
,
641 dev
->caps
.num_mgms
+ dev
->caps
.num_amgms
,
644 mlx4_err(dev
, "Failed to map MCG context memory, aborting.\n");
651 mlx4_cleanup_icm_table(dev
, &priv
->srq_table
.table
);
654 mlx4_cleanup_icm_table(dev
, &priv
->cq_table
.table
);
657 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.rdmarc_table
);
660 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.altc_table
);
663 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.auxc_table
);
666 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.qp_table
);
669 mlx4_cleanup_icm_table(dev
, &priv
->mr_table
.dmpt_table
);
672 mlx4_cleanup_icm_table(dev
, &priv
->mr_table
.mtt_table
);
675 mlx4_cleanup_icm_table(dev
, &priv
->eq_table
.table
);
678 mlx4_cleanup_icm_table(dev
, &priv
->eq_table
.cmpt_table
);
679 mlx4_cleanup_icm_table(dev
, &priv
->cq_table
.cmpt_table
);
680 mlx4_cleanup_icm_table(dev
, &priv
->srq_table
.cmpt_table
);
681 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.cmpt_table
);
684 mlx4_UNMAP_ICM_AUX(dev
);
687 mlx4_free_icm(dev
, priv
->fw
.aux_icm
, 0);
692 static void mlx4_free_icms(struct mlx4_dev
*dev
)
694 struct mlx4_priv
*priv
= mlx4_priv(dev
);
696 mlx4_cleanup_icm_table(dev
, &priv
->mcg_table
.table
);
697 mlx4_cleanup_icm_table(dev
, &priv
->srq_table
.table
);
698 mlx4_cleanup_icm_table(dev
, &priv
->cq_table
.table
);
699 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.rdmarc_table
);
700 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.altc_table
);
701 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.auxc_table
);
702 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.qp_table
);
703 mlx4_cleanup_icm_table(dev
, &priv
->mr_table
.dmpt_table
);
704 mlx4_cleanup_icm_table(dev
, &priv
->mr_table
.mtt_table
);
705 mlx4_cleanup_icm_table(dev
, &priv
->eq_table
.table
);
706 mlx4_cleanup_icm_table(dev
, &priv
->eq_table
.cmpt_table
);
707 mlx4_cleanup_icm_table(dev
, &priv
->cq_table
.cmpt_table
);
708 mlx4_cleanup_icm_table(dev
, &priv
->srq_table
.cmpt_table
);
709 mlx4_cleanup_icm_table(dev
, &priv
->qp_table
.cmpt_table
);
711 mlx4_UNMAP_ICM_AUX(dev
);
712 mlx4_free_icm(dev
, priv
->fw
.aux_icm
, 0);
715 static void mlx4_close_hca(struct mlx4_dev
*dev
)
717 mlx4_CLOSE_HCA(dev
, 0);
720 mlx4_free_icm(dev
, mlx4_priv(dev
)->fw
.fw_icm
, 0);
723 static int mlx4_init_hca(struct mlx4_dev
*dev
)
725 struct mlx4_priv
*priv
= mlx4_priv(dev
);
726 struct mlx4_adapter adapter
;
727 struct mlx4_dev_cap dev_cap
;
728 struct mlx4_mod_stat_cfg mlx4_cfg
;
729 struct mlx4_profile profile
;
730 struct mlx4_init_hca_param init_hca
;
734 err
= mlx4_QUERY_FW(dev
);
737 mlx4_info(dev
, "non-primary physical function, skipping.\n");
739 mlx4_err(dev
, "QUERY_FW command failed, aborting.\n");
743 err
= mlx4_load_fw(dev
);
745 mlx4_err(dev
, "Failed to start FW, aborting.\n");
749 mlx4_cfg
.log_pg_sz_m
= 1;
750 mlx4_cfg
.log_pg_sz
= 0;
751 err
= mlx4_MOD_STAT_CFG(dev
, &mlx4_cfg
);
753 mlx4_warn(dev
, "Failed to override log_pg_sz parameter\n");
755 err
= mlx4_dev_cap(dev
, &dev_cap
);
757 mlx4_err(dev
, "QUERY_DEV_CAP command failed, aborting.\n");
761 profile
= default_profile
;
763 icm_size
= mlx4_make_profile(dev
, &profile
, &dev_cap
, &init_hca
);
764 if ((long long) icm_size
< 0) {
769 init_hca
.log_uar_sz
= ilog2(dev
->caps
.num_uars
);
771 err
= mlx4_init_icm(dev
, &dev_cap
, &init_hca
, icm_size
);
775 err
= mlx4_INIT_HCA(dev
, &init_hca
);
777 mlx4_err(dev
, "INIT_HCA command failed, aborting.\n");
781 err
= mlx4_QUERY_ADAPTER(dev
, &adapter
);
783 mlx4_err(dev
, "QUERY_ADAPTER command failed, aborting.\n");
787 priv
->eq_table
.inta_pin
= adapter
.inta_pin
;
788 memcpy(dev
->board_id
, adapter
.board_id
, sizeof dev
->board_id
);
793 mlx4_CLOSE_HCA(dev
, 0);
800 mlx4_free_icm(dev
, priv
->fw
.fw_icm
, 0);
805 static int mlx4_setup_hca(struct mlx4_dev
*dev
)
807 struct mlx4_priv
*priv
= mlx4_priv(dev
);
810 __be32 ib_port_default_caps
;
812 err
= mlx4_init_uar_table(dev
);
814 mlx4_err(dev
, "Failed to initialize "
815 "user access region table, aborting.\n");
819 err
= mlx4_uar_alloc(dev
, &priv
->driver_uar
);
821 mlx4_err(dev
, "Failed to allocate driver access region, "
823 goto err_uar_table_free
;
826 priv
->kar
= ioremap(priv
->driver_uar
.pfn
<< PAGE_SHIFT
, PAGE_SIZE
);
828 mlx4_err(dev
, "Couldn't map kernel access region, "
834 err
= mlx4_init_pd_table(dev
);
836 mlx4_err(dev
, "Failed to initialize "
837 "protection domain table, aborting.\n");
841 err
= mlx4_init_mr_table(dev
);
843 mlx4_err(dev
, "Failed to initialize "
844 "memory region table, aborting.\n");
845 goto err_pd_table_free
;
848 err
= mlx4_init_eq_table(dev
);
850 mlx4_err(dev
, "Failed to initialize "
851 "event queue table, aborting.\n");
852 goto err_mr_table_free
;
855 err
= mlx4_cmd_use_events(dev
);
857 mlx4_err(dev
, "Failed to switch to event-driven "
858 "firmware commands, aborting.\n");
859 goto err_eq_table_free
;
864 if (dev
->flags
& MLX4_FLAG_MSI_X
) {
865 mlx4_warn(dev
, "NOP command failed to generate MSI-X "
866 "interrupt IRQ %d).\n",
867 priv
->eq_table
.eq
[dev
->caps
.num_comp_vectors
].irq
);
868 mlx4_warn(dev
, "Trying again without MSI-X.\n");
870 mlx4_err(dev
, "NOP command failed to generate interrupt "
871 "(IRQ %d), aborting.\n",
872 priv
->eq_table
.eq
[dev
->caps
.num_comp_vectors
].irq
);
873 mlx4_err(dev
, "BIOS or ACPI interrupt routing problem?\n");
879 mlx4_dbg(dev
, "NOP command IRQ test passed\n");
881 err
= mlx4_init_cq_table(dev
);
883 mlx4_err(dev
, "Failed to initialize "
884 "completion queue table, aborting.\n");
888 err
= mlx4_init_srq_table(dev
);
890 mlx4_err(dev
, "Failed to initialize "
891 "shared receive queue table, aborting.\n");
892 goto err_cq_table_free
;
895 err
= mlx4_init_qp_table(dev
);
897 mlx4_err(dev
, "Failed to initialize "
898 "queue pair table, aborting.\n");
899 goto err_srq_table_free
;
902 err
= mlx4_init_mcg_table(dev
);
904 mlx4_err(dev
, "Failed to initialize "
905 "multicast group table, aborting.\n");
906 goto err_qp_table_free
;
909 for (port
= 1; port
<= dev
->caps
.num_ports
; port
++) {
910 ib_port_default_caps
= 0;
911 err
= mlx4_get_port_ib_caps(dev
, port
, &ib_port_default_caps
);
913 mlx4_warn(dev
, "failed to get port %d default "
914 "ib capabilities (%d). Continuing with "
915 "caps = 0\n", port
, err
);
916 dev
->caps
.ib_port_def_cap
[port
] = ib_port_default_caps
;
917 err
= mlx4_SET_PORT(dev
, port
);
919 mlx4_err(dev
, "Failed to set port %d, aborting\n",
921 goto err_mcg_table_free
;
928 mlx4_cleanup_mcg_table(dev
);
931 mlx4_cleanup_qp_table(dev
);
934 mlx4_cleanup_srq_table(dev
);
937 mlx4_cleanup_cq_table(dev
);
940 mlx4_cmd_use_polling(dev
);
943 mlx4_cleanup_eq_table(dev
);
946 mlx4_cleanup_mr_table(dev
);
949 mlx4_cleanup_pd_table(dev
);
955 mlx4_uar_free(dev
, &priv
->driver_uar
);
958 mlx4_cleanup_uar_table(dev
);
962 static void mlx4_enable_msi_x(struct mlx4_dev
*dev
)
964 struct mlx4_priv
*priv
= mlx4_priv(dev
);
965 struct msix_entry
*entries
;
971 nreq
= min_t(int, dev
->caps
.num_eqs
- dev
->caps
.reserved_eqs
,
972 num_possible_cpus() + 1);
973 entries
= kcalloc(nreq
, sizeof *entries
, GFP_KERNEL
);
977 for (i
= 0; i
< nreq
; ++i
)
978 entries
[i
].entry
= i
;
981 err
= pci_enable_msix(dev
->pdev
, entries
, nreq
);
983 /* Try again if at least 2 vectors are available */
985 mlx4_info(dev
, "Requested %d vectors, "
986 "but only %d MSI-X vectors available, "
987 "trying again\n", nreq
, err
);
995 dev
->caps
.num_comp_vectors
= nreq
- 1;
996 for (i
= 0; i
< nreq
; ++i
)
997 priv
->eq_table
.eq
[i
].irq
= entries
[i
].vector
;
999 dev
->flags
|= MLX4_FLAG_MSI_X
;
1006 dev
->caps
.num_comp_vectors
= 1;
1008 for (i
= 0; i
< 2; ++i
)
1009 priv
->eq_table
.eq
[i
].irq
= dev
->pdev
->irq
;
1012 static int mlx4_init_port_info(struct mlx4_dev
*dev
, int port
)
1014 struct mlx4_port_info
*info
= &mlx4_priv(dev
)->port
[port
];
1019 mlx4_init_mac_table(dev
, &info
->mac_table
);
1020 mlx4_init_vlan_table(dev
, &info
->vlan_table
);
1022 sprintf(info
->dev_name
, "mlx4_port%d", port
);
1023 info
->port_attr
.attr
.name
= info
->dev_name
;
1024 info
->port_attr
.attr
.mode
= S_IRUGO
| S_IWUSR
;
1025 info
->port_attr
.show
= show_port_type
;
1026 info
->port_attr
.store
= set_port_type
;
1027 sysfs_attr_init(&info
->port_attr
.attr
);
1029 err
= device_create_file(&dev
->pdev
->dev
, &info
->port_attr
);
1031 mlx4_err(dev
, "Failed to create file for port %d\n", port
);
1038 static void mlx4_cleanup_port_info(struct mlx4_port_info
*info
)
1043 device_remove_file(&info
->dev
->pdev
->dev
, &info
->port_attr
);
1046 static int __mlx4_init_one(struct pci_dev
*pdev
, const struct pci_device_id
*id
)
1048 struct mlx4_priv
*priv
;
1049 struct mlx4_dev
*dev
;
1053 printk(KERN_INFO PFX
"Initializing %s\n",
1056 err
= pci_enable_device(pdev
);
1058 dev_err(&pdev
->dev
, "Cannot enable PCI device, "
1064 * Check for BARs. We expect 0: 1MB
1066 if (!(pci_resource_flags(pdev
, 0) & IORESOURCE_MEM
) ||
1067 pci_resource_len(pdev
, 0) != 1 << 20) {
1068 dev_err(&pdev
->dev
, "Missing DCS, aborting.\n");
1070 goto err_disable_pdev
;
1072 if (!(pci_resource_flags(pdev
, 2) & IORESOURCE_MEM
)) {
1073 dev_err(&pdev
->dev
, "Missing UAR, aborting.\n");
1075 goto err_disable_pdev
;
1078 err
= pci_request_regions(pdev
, DRV_NAME
);
1080 dev_err(&pdev
->dev
, "Couldn't get PCI resources, aborting\n");
1081 goto err_disable_pdev
;
1084 pci_set_master(pdev
);
1086 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(64));
1088 dev_warn(&pdev
->dev
, "Warning: couldn't set 64-bit PCI DMA mask.\n");
1089 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(32));
1091 dev_err(&pdev
->dev
, "Can't set PCI DMA mask, aborting.\n");
1092 goto err_release_regions
;
1095 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(64));
1097 dev_warn(&pdev
->dev
, "Warning: couldn't set 64-bit "
1098 "consistent PCI DMA mask.\n");
1099 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(32));
1101 dev_err(&pdev
->dev
, "Can't set consistent PCI DMA mask, "
1103 goto err_release_regions
;
1107 priv
= kzalloc(sizeof *priv
, GFP_KERNEL
);
1109 dev_err(&pdev
->dev
, "Device struct alloc failed, "
1112 goto err_release_regions
;
1117 INIT_LIST_HEAD(&priv
->ctx_list
);
1118 spin_lock_init(&priv
->ctx_lock
);
1120 mutex_init(&priv
->port_mutex
);
1122 INIT_LIST_HEAD(&priv
->pgdir_list
);
1123 mutex_init(&priv
->pgdir_mutex
);
1126 * Now reset the HCA before we touch the PCI capabilities or
1127 * attempt a firmware command, since a boot ROM may have left
1128 * the HCA in an undefined state.
1130 err
= mlx4_reset(dev
);
1132 mlx4_err(dev
, "Failed to reset HCA, aborting.\n");
1136 if (mlx4_cmd_init(dev
)) {
1137 mlx4_err(dev
, "Failed to init command interface, aborting.\n");
1141 err
= mlx4_init_hca(dev
);
1145 err
= mlx4_alloc_eq_table(dev
);
1149 mlx4_enable_msi_x(dev
);
1151 err
= mlx4_setup_hca(dev
);
1152 if (err
== -EBUSY
&& (dev
->flags
& MLX4_FLAG_MSI_X
)) {
1153 dev
->flags
&= ~MLX4_FLAG_MSI_X
;
1154 pci_disable_msix(pdev
);
1155 err
= mlx4_setup_hca(dev
);
1161 for (port
= 1; port
<= dev
->caps
.num_ports
; port
++) {
1162 err
= mlx4_init_port_info(dev
, port
);
1167 err
= mlx4_register_device(dev
);
1171 mlx4_sense_init(dev
);
1172 mlx4_start_sense(dev
);
1174 pci_set_drvdata(pdev
, dev
);
1179 for (--port
; port
>= 1; --port
)
1180 mlx4_cleanup_port_info(&priv
->port
[port
]);
1182 mlx4_cleanup_mcg_table(dev
);
1183 mlx4_cleanup_qp_table(dev
);
1184 mlx4_cleanup_srq_table(dev
);
1185 mlx4_cleanup_cq_table(dev
);
1186 mlx4_cmd_use_polling(dev
);
1187 mlx4_cleanup_eq_table(dev
);
1188 mlx4_cleanup_mr_table(dev
);
1189 mlx4_cleanup_pd_table(dev
);
1190 mlx4_cleanup_uar_table(dev
);
1193 mlx4_free_eq_table(dev
);
1196 if (dev
->flags
& MLX4_FLAG_MSI_X
)
1197 pci_disable_msix(pdev
);
1199 mlx4_close_hca(dev
);
1202 mlx4_cmd_cleanup(dev
);
1207 err_release_regions
:
1208 pci_release_regions(pdev
);
1211 pci_disable_device(pdev
);
1212 pci_set_drvdata(pdev
, NULL
);
1216 static int __devinit
mlx4_init_one(struct pci_dev
*pdev
,
1217 const struct pci_device_id
*id
)
1219 static int mlx4_version_printed
;
1221 if (!mlx4_version_printed
) {
1222 printk(KERN_INFO
"%s", mlx4_version
);
1223 ++mlx4_version_printed
;
1226 return __mlx4_init_one(pdev
, id
);
1229 static void mlx4_remove_one(struct pci_dev
*pdev
)
1231 struct mlx4_dev
*dev
= pci_get_drvdata(pdev
);
1232 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1236 mlx4_stop_sense(dev
);
1237 mlx4_unregister_device(dev
);
1239 for (p
= 1; p
<= dev
->caps
.num_ports
; p
++) {
1240 mlx4_cleanup_port_info(&priv
->port
[p
]);
1241 mlx4_CLOSE_PORT(dev
, p
);
1244 mlx4_cleanup_mcg_table(dev
);
1245 mlx4_cleanup_qp_table(dev
);
1246 mlx4_cleanup_srq_table(dev
);
1247 mlx4_cleanup_cq_table(dev
);
1248 mlx4_cmd_use_polling(dev
);
1249 mlx4_cleanup_eq_table(dev
);
1250 mlx4_cleanup_mr_table(dev
);
1251 mlx4_cleanup_pd_table(dev
);
1254 mlx4_uar_free(dev
, &priv
->driver_uar
);
1255 mlx4_cleanup_uar_table(dev
);
1256 mlx4_free_eq_table(dev
);
1257 mlx4_close_hca(dev
);
1258 mlx4_cmd_cleanup(dev
);
1260 if (dev
->flags
& MLX4_FLAG_MSI_X
)
1261 pci_disable_msix(pdev
);
1264 pci_release_regions(pdev
);
1265 pci_disable_device(pdev
);
1266 pci_set_drvdata(pdev
, NULL
);
1270 int mlx4_restart_one(struct pci_dev
*pdev
)
1272 mlx4_remove_one(pdev
);
1273 return __mlx4_init_one(pdev
, NULL
);
1276 static DEFINE_PCI_DEVICE_TABLE(mlx4_pci_table
) = {
1277 { PCI_VDEVICE(MELLANOX
, 0x6340) }, /* MT25408 "Hermon" SDR */
1278 { PCI_VDEVICE(MELLANOX
, 0x634a) }, /* MT25408 "Hermon" DDR */
1279 { PCI_VDEVICE(MELLANOX
, 0x6354) }, /* MT25408 "Hermon" QDR */
1280 { PCI_VDEVICE(MELLANOX
, 0x6732) }, /* MT25408 "Hermon" DDR PCIe gen2 */
1281 { PCI_VDEVICE(MELLANOX
, 0x673c) }, /* MT25408 "Hermon" QDR PCIe gen2 */
1282 { PCI_VDEVICE(MELLANOX
, 0x6368) }, /* MT25408 "Hermon" EN 10GigE */
1283 { PCI_VDEVICE(MELLANOX
, 0x6750) }, /* MT25408 "Hermon" EN 10GigE PCIe gen2 */
1284 { PCI_VDEVICE(MELLANOX
, 0x6372) }, /* MT25458 ConnectX EN 10GBASE-T 10GigE */
1285 { PCI_VDEVICE(MELLANOX
, 0x675a) }, /* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
1286 { PCI_VDEVICE(MELLANOX
, 0x6764) }, /* MT26468 ConnectX EN 10GigE PCIe gen2*/
1287 { PCI_VDEVICE(MELLANOX
, 0x6746) }, /* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
1288 { PCI_VDEVICE(MELLANOX
, 0x676e) }, /* MT26478 ConnectX2 40GigE PCIe gen2 */
1292 MODULE_DEVICE_TABLE(pci
, mlx4_pci_table
);
1294 static struct pci_driver mlx4_driver
= {
1296 .id_table
= mlx4_pci_table
,
1297 .probe
= mlx4_init_one
,
1298 .remove
= __devexit_p(mlx4_remove_one
)
1301 static int __init
mlx4_verify_params(void)
1303 if ((log_num_mac
< 0) || (log_num_mac
> 7)) {
1304 printk(KERN_WARNING
"mlx4_core: bad num_mac: %d\n", log_num_mac
);
1308 if ((log_num_vlan
< 0) || (log_num_vlan
> 7)) {
1309 printk(KERN_WARNING
"mlx4_core: bad num_vlan: %d\n", log_num_vlan
);
1313 if ((log_mtts_per_seg
< 1) || (log_mtts_per_seg
> 5)) {
1314 printk(KERN_WARNING
"mlx4_core: bad log_mtts_per_seg: %d\n", log_mtts_per_seg
);
1321 static int __init
mlx4_init(void)
1325 if (mlx4_verify_params())
1330 mlx4_wq
= create_singlethread_workqueue("mlx4");
1334 ret
= pci_register_driver(&mlx4_driver
);
1335 return ret
< 0 ? ret
: 0;
1338 static void __exit
mlx4_cleanup(void)
1340 pci_unregister_driver(&mlx4_driver
);
1341 destroy_workqueue(mlx4_wq
);
1344 module_init(mlx4_init
);
1345 module_exit(mlx4_cleanup
);