Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[pandora-kernel.git] / drivers / net / mlx4 / main.c
1 /*
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.
6  *
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:
12  *
13  *     Redistribution and use in source and binary forms, with or
14  *     without modification, are permitted provided that the following
15  *     conditions are met:
16  *
17  *      - Redistributions of source code must retain the above
18  *        copyright notice, this list of conditions and the following
19  *        disclaimer.
20  *
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.
25  *
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
33  * SOFTWARE.
34  */
35
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>
42 #include <linux/io-mapping.h>
43
44 #include <linux/mlx4/device.h>
45 #include <linux/mlx4/doorbell.h>
46
47 #include "mlx4.h"
48 #include "fw.h"
49 #include "icm.h"
50
51 MODULE_AUTHOR("Roland Dreier");
52 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
53 MODULE_LICENSE("Dual BSD/GPL");
54 MODULE_VERSION(DRV_VERSION);
55
56 struct workqueue_struct *mlx4_wq;
57
58 #ifdef CONFIG_MLX4_DEBUG
59
60 int mlx4_debug_level = 0;
61 module_param_named(debug_level, mlx4_debug_level, int, 0644);
62 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
63
64 #endif /* CONFIG_MLX4_DEBUG */
65
66 #ifdef CONFIG_PCI_MSI
67
68 static int msi_x = 1;
69 module_param(msi_x, int, 0444);
70 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
71
72 #else /* CONFIG_PCI_MSI */
73
74 #define msi_x (0)
75
76 #endif /* CONFIG_PCI_MSI */
77
78 static char mlx4_version[] __devinitdata =
79         DRV_NAME ": Mellanox ConnectX core driver v"
80         DRV_VERSION " (" DRV_RELDATE ")\n";
81
82 static struct mlx4_profile default_profile = {
83         .num_qp         = 1 << 17,
84         .num_srq        = 1 << 16,
85         .rdmarc_per_qp  = 1 << 4,
86         .num_cq         = 1 << 16,
87         .num_mcg        = 1 << 13,
88         .num_mpt        = 1 << 17,
89         .num_mtt        = 1 << 20,
90 };
91
92 static int log_num_mac = 2;
93 module_param_named(log_num_mac, log_num_mac, int, 0444);
94 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
95
96 static int log_num_vlan;
97 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
98 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
99
100 static int use_prio;
101 module_param_named(use_prio, use_prio, bool, 0444);
102 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports "
103                   "(0/1, default 0)");
104
105 static int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
106 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
107 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
108
109 int mlx4_check_port_params(struct mlx4_dev *dev,
110                            enum mlx4_port_type *port_type)
111 {
112         int i;
113
114         for (i = 0; i < dev->caps.num_ports - 1; i++) {
115                 if (port_type[i] != port_type[i + 1]) {
116                         if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
117                                 mlx4_err(dev, "Only same port types supported "
118                                          "on this HCA, aborting.\n");
119                                 return -EINVAL;
120                         }
121                         if (port_type[i] == MLX4_PORT_TYPE_ETH &&
122                             port_type[i + 1] == MLX4_PORT_TYPE_IB)
123                                 return -EINVAL;
124                 }
125         }
126
127         for (i = 0; i < dev->caps.num_ports; i++) {
128                 if (!(port_type[i] & dev->caps.supported_type[i+1])) {
129                         mlx4_err(dev, "Requested port type for port %d is not "
130                                       "supported on this HCA\n", i + 1);
131                         return -EINVAL;
132                 }
133         }
134         return 0;
135 }
136
137 static void mlx4_set_port_mask(struct mlx4_dev *dev)
138 {
139         int i;
140
141         dev->caps.port_mask = 0;
142         for (i = 1; i <= dev->caps.num_ports; ++i)
143                 if (dev->caps.port_type[i] == MLX4_PORT_TYPE_IB)
144                         dev->caps.port_mask |= 1 << (i - 1);
145 }
146 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
147 {
148         int err;
149         int i;
150
151         err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
152         if (err) {
153                 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
154                 return err;
155         }
156
157         if (dev_cap->min_page_sz > PAGE_SIZE) {
158                 mlx4_err(dev, "HCA minimum page size of %d bigger than "
159                          "kernel PAGE_SIZE of %ld, aborting.\n",
160                          dev_cap->min_page_sz, PAGE_SIZE);
161                 return -ENODEV;
162         }
163         if (dev_cap->num_ports > MLX4_MAX_PORTS) {
164                 mlx4_err(dev, "HCA has %d ports, but we only support %d, "
165                          "aborting.\n",
166                          dev_cap->num_ports, MLX4_MAX_PORTS);
167                 return -ENODEV;
168         }
169
170         if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
171                 mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than "
172                          "PCI resource 2 size of 0x%llx, aborting.\n",
173                          dev_cap->uar_size,
174                          (unsigned long long) pci_resource_len(dev->pdev, 2));
175                 return -ENODEV;
176         }
177
178         dev->caps.num_ports          = dev_cap->num_ports;
179         for (i = 1; i <= dev->caps.num_ports; ++i) {
180                 dev->caps.vl_cap[i]         = dev_cap->max_vl[i];
181                 dev->caps.ib_mtu_cap[i]     = dev_cap->ib_mtu[i];
182                 dev->caps.gid_table_len[i]  = dev_cap->max_gids[i];
183                 dev->caps.pkey_table_len[i] = dev_cap->max_pkeys[i];
184                 dev->caps.port_width_cap[i] = dev_cap->max_port_width[i];
185                 dev->caps.eth_mtu_cap[i]    = dev_cap->eth_mtu[i];
186                 dev->caps.def_mac[i]        = dev_cap->def_mac[i];
187                 dev->caps.supported_type[i] = dev_cap->supported_port_types[i];
188                 dev->caps.trans_type[i]     = dev_cap->trans_type[i];
189                 dev->caps.vendor_oui[i]     = dev_cap->vendor_oui[i];
190                 dev->caps.wavelength[i]     = dev_cap->wavelength[i];
191                 dev->caps.trans_code[i]     = dev_cap->trans_code[i];
192         }
193
194         dev->caps.num_uars           = dev_cap->uar_size / PAGE_SIZE;
195         dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
196         dev->caps.bf_reg_size        = dev_cap->bf_reg_size;
197         dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
198         dev->caps.max_sq_sg          = dev_cap->max_sq_sg;
199         dev->caps.max_rq_sg          = dev_cap->max_rq_sg;
200         dev->caps.max_wqes           = dev_cap->max_qp_sz;
201         dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
202         dev->caps.max_srq_wqes       = dev_cap->max_srq_sz;
203         dev->caps.max_srq_sge        = dev_cap->max_rq_sg - 1;
204         dev->caps.reserved_srqs      = dev_cap->reserved_srqs;
205         dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
206         dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
207         dev->caps.num_qp_per_mgm     = MLX4_QP_PER_MGM;
208         /*
209          * Subtract 1 from the limit because we need to allocate a
210          * spare CQE so the HCA HW can tell the difference between an
211          * empty CQ and a full CQ.
212          */
213         dev->caps.max_cqes           = dev_cap->max_cq_sz - 1;
214         dev->caps.reserved_cqs       = dev_cap->reserved_cqs;
215         dev->caps.reserved_eqs       = dev_cap->reserved_eqs;
216         dev->caps.mtts_per_seg       = 1 << log_mtts_per_seg;
217         dev->caps.reserved_mtts      = DIV_ROUND_UP(dev_cap->reserved_mtts,
218                                                     dev->caps.mtts_per_seg);
219         dev->caps.reserved_mrws      = dev_cap->reserved_mrws;
220         dev->caps.reserved_uars      = dev_cap->reserved_uars;
221         dev->caps.reserved_pds       = dev_cap->reserved_pds;
222         dev->caps.mtt_entry_sz       = dev->caps.mtts_per_seg * dev_cap->mtt_entry_sz;
223         dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
224         dev->caps.page_size_cap      = ~(u32) (dev_cap->min_page_sz - 1);
225         dev->caps.flags              = dev_cap->flags;
226         dev->caps.bmme_flags         = dev_cap->bmme_flags;
227         dev->caps.reserved_lkey      = dev_cap->reserved_lkey;
228         dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
229         dev->caps.udp_rss            = dev_cap->udp_rss;
230         dev->caps.loopback_support   = dev_cap->loopback_support;
231         dev->caps.vep_uc_steering    = dev_cap->vep_uc_steering;
232         dev->caps.vep_mc_steering    = dev_cap->vep_mc_steering;
233         dev->caps.wol                = dev_cap->wol;
234         dev->caps.max_gso_sz         = dev_cap->max_gso_sz;
235
236         dev->caps.log_num_macs  = log_num_mac;
237         dev->caps.log_num_vlans = log_num_vlan;
238         dev->caps.log_num_prios = use_prio ? 3 : 0;
239
240         for (i = 1; i <= dev->caps.num_ports; ++i) {
241                 if (dev->caps.supported_type[i] != MLX4_PORT_TYPE_ETH)
242                         dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
243                 else
244                         dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
245                 dev->caps.possible_type[i] = dev->caps.port_type[i];
246                 mlx4_priv(dev)->sense.sense_allowed[i] =
247                         dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO;
248
249                 if (dev->caps.log_num_macs > dev_cap->log_max_macs[i]) {
250                         dev->caps.log_num_macs = dev_cap->log_max_macs[i];
251                         mlx4_warn(dev, "Requested number of MACs is too much "
252                                   "for port %d, reducing to %d.\n",
253                                   i, 1 << dev->caps.log_num_macs);
254                 }
255                 if (dev->caps.log_num_vlans > dev_cap->log_max_vlans[i]) {
256                         dev->caps.log_num_vlans = dev_cap->log_max_vlans[i];
257                         mlx4_warn(dev, "Requested number of VLANs is too much "
258                                   "for port %d, reducing to %d.\n",
259                                   i, 1 << dev->caps.log_num_vlans);
260                 }
261         }
262
263         mlx4_set_port_mask(dev);
264
265         dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
266         dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
267                 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
268                 (1 << dev->caps.log_num_macs) *
269                 (1 << dev->caps.log_num_vlans) *
270                 (1 << dev->caps.log_num_prios) *
271                 dev->caps.num_ports;
272         dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
273
274         dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
275                 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
276                 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
277                 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
278
279         return 0;
280 }
281
282 /*
283  * Change the port configuration of the device.
284  * Every user of this function must hold the port mutex.
285  */
286 int mlx4_change_port_types(struct mlx4_dev *dev,
287                            enum mlx4_port_type *port_types)
288 {
289         int err = 0;
290         int change = 0;
291         int port;
292
293         for (port = 0; port <  dev->caps.num_ports; port++) {
294                 /* Change the port type only if the new type is different
295                  * from the current, and not set to Auto */
296                 if (port_types[port] != dev->caps.port_type[port + 1]) {
297                         change = 1;
298                         dev->caps.port_type[port + 1] = port_types[port];
299                 }
300         }
301         if (change) {
302                 mlx4_unregister_device(dev);
303                 for (port = 1; port <= dev->caps.num_ports; port++) {
304                         mlx4_CLOSE_PORT(dev, port);
305                         err = mlx4_SET_PORT(dev, port);
306                         if (err) {
307                                 mlx4_err(dev, "Failed to set port %d, "
308                                               "aborting\n", port);
309                                 goto out;
310                         }
311                 }
312                 mlx4_set_port_mask(dev);
313                 err = mlx4_register_device(dev);
314         }
315
316 out:
317         return err;
318 }
319
320 static ssize_t show_port_type(struct device *dev,
321                               struct device_attribute *attr,
322                               char *buf)
323 {
324         struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
325                                                    port_attr);
326         struct mlx4_dev *mdev = info->dev;
327         char type[8];
328
329         sprintf(type, "%s",
330                 (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
331                 "ib" : "eth");
332         if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
333                 sprintf(buf, "auto (%s)\n", type);
334         else
335                 sprintf(buf, "%s\n", type);
336
337         return strlen(buf);
338 }
339
340 static ssize_t set_port_type(struct device *dev,
341                              struct device_attribute *attr,
342                              const char *buf, size_t count)
343 {
344         struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
345                                                    port_attr);
346         struct mlx4_dev *mdev = info->dev;
347         struct mlx4_priv *priv = mlx4_priv(mdev);
348         enum mlx4_port_type types[MLX4_MAX_PORTS];
349         enum mlx4_port_type new_types[MLX4_MAX_PORTS];
350         int i;
351         int err = 0;
352
353         if (!strcmp(buf, "ib\n"))
354                 info->tmp_type = MLX4_PORT_TYPE_IB;
355         else if (!strcmp(buf, "eth\n"))
356                 info->tmp_type = MLX4_PORT_TYPE_ETH;
357         else if (!strcmp(buf, "auto\n"))
358                 info->tmp_type = MLX4_PORT_TYPE_AUTO;
359         else {
360                 mlx4_err(mdev, "%s is not supported port type\n", buf);
361                 return -EINVAL;
362         }
363
364         mlx4_stop_sense(mdev);
365         mutex_lock(&priv->port_mutex);
366         /* Possible type is always the one that was delivered */
367         mdev->caps.possible_type[info->port] = info->tmp_type;
368
369         for (i = 0; i < mdev->caps.num_ports; i++) {
370                 types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
371                                         mdev->caps.possible_type[i+1];
372                 if (types[i] == MLX4_PORT_TYPE_AUTO)
373                         types[i] = mdev->caps.port_type[i+1];
374         }
375
376         if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
377                 for (i = 1; i <= mdev->caps.num_ports; i++) {
378                         if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
379                                 mdev->caps.possible_type[i] = mdev->caps.port_type[i];
380                                 err = -EINVAL;
381                         }
382                 }
383         }
384         if (err) {
385                 mlx4_err(mdev, "Auto sensing is not supported on this HCA. "
386                                "Set only 'eth' or 'ib' for both ports "
387                                "(should be the same)\n");
388                 goto out;
389         }
390
391         mlx4_do_sense_ports(mdev, new_types, types);
392
393         err = mlx4_check_port_params(mdev, new_types);
394         if (err)
395                 goto out;
396
397         /* We are about to apply the changes after the configuration
398          * was verified, no need to remember the temporary types
399          * any more */
400         for (i = 0; i < mdev->caps.num_ports; i++)
401                 priv->port[i + 1].tmp_type = 0;
402
403         err = mlx4_change_port_types(mdev, new_types);
404
405 out:
406         mlx4_start_sense(mdev);
407         mutex_unlock(&priv->port_mutex);
408         return err ? err : count;
409 }
410
411 static int mlx4_load_fw(struct mlx4_dev *dev)
412 {
413         struct mlx4_priv *priv = mlx4_priv(dev);
414         int err;
415
416         priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
417                                          GFP_HIGHUSER | __GFP_NOWARN, 0);
418         if (!priv->fw.fw_icm) {
419                 mlx4_err(dev, "Couldn't allocate FW area, aborting.\n");
420                 return -ENOMEM;
421         }
422
423         err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
424         if (err) {
425                 mlx4_err(dev, "MAP_FA command failed, aborting.\n");
426                 goto err_free;
427         }
428
429         err = mlx4_RUN_FW(dev);
430         if (err) {
431                 mlx4_err(dev, "RUN_FW command failed, aborting.\n");
432                 goto err_unmap_fa;
433         }
434
435         return 0;
436
437 err_unmap_fa:
438         mlx4_UNMAP_FA(dev);
439
440 err_free:
441         mlx4_free_icm(dev, priv->fw.fw_icm, 0);
442         return err;
443 }
444
445 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
446                                 int cmpt_entry_sz)
447 {
448         struct mlx4_priv *priv = mlx4_priv(dev);
449         int err;
450
451         err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
452                                   cmpt_base +
453                                   ((u64) (MLX4_CMPT_TYPE_QP *
454                                           cmpt_entry_sz) << MLX4_CMPT_SHIFT),
455                                   cmpt_entry_sz, dev->caps.num_qps,
456                                   dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
457                                   0, 0);
458         if (err)
459                 goto err;
460
461         err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
462                                   cmpt_base +
463                                   ((u64) (MLX4_CMPT_TYPE_SRQ *
464                                           cmpt_entry_sz) << MLX4_CMPT_SHIFT),
465                                   cmpt_entry_sz, dev->caps.num_srqs,
466                                   dev->caps.reserved_srqs, 0, 0);
467         if (err)
468                 goto err_qp;
469
470         err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
471                                   cmpt_base +
472                                   ((u64) (MLX4_CMPT_TYPE_CQ *
473                                           cmpt_entry_sz) << MLX4_CMPT_SHIFT),
474                                   cmpt_entry_sz, dev->caps.num_cqs,
475                                   dev->caps.reserved_cqs, 0, 0);
476         if (err)
477                 goto err_srq;
478
479         err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
480                                   cmpt_base +
481                                   ((u64) (MLX4_CMPT_TYPE_EQ *
482                                           cmpt_entry_sz) << MLX4_CMPT_SHIFT),
483                                   cmpt_entry_sz,
484                                   dev->caps.num_eqs, dev->caps.num_eqs, 0, 0);
485         if (err)
486                 goto err_cq;
487
488         return 0;
489
490 err_cq:
491         mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
492
493 err_srq:
494         mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
495
496 err_qp:
497         mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
498
499 err:
500         return err;
501 }
502
503 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
504                          struct mlx4_init_hca_param *init_hca, u64 icm_size)
505 {
506         struct mlx4_priv *priv = mlx4_priv(dev);
507         u64 aux_pages;
508         int err;
509
510         err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
511         if (err) {
512                 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting.\n");
513                 return err;
514         }
515
516         mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory.\n",
517                  (unsigned long long) icm_size >> 10,
518                  (unsigned long long) aux_pages << 2);
519
520         priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
521                                           GFP_HIGHUSER | __GFP_NOWARN, 0);
522         if (!priv->fw.aux_icm) {
523                 mlx4_err(dev, "Couldn't allocate aux memory, aborting.\n");
524                 return -ENOMEM;
525         }
526
527         err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
528         if (err) {
529                 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting.\n");
530                 goto err_free_aux;
531         }
532
533         err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
534         if (err) {
535                 mlx4_err(dev, "Failed to map cMPT context memory, aborting.\n");
536                 goto err_unmap_aux;
537         }
538
539         err = mlx4_init_icm_table(dev, &priv->eq_table.table,
540                                   init_hca->eqc_base, dev_cap->eqc_entry_sz,
541                                   dev->caps.num_eqs, dev->caps.num_eqs,
542                                   0, 0);
543         if (err) {
544                 mlx4_err(dev, "Failed to map EQ context memory, aborting.\n");
545                 goto err_unmap_cmpt;
546         }
547
548         /*
549          * Reserved MTT entries must be aligned up to a cacheline
550          * boundary, since the FW will write to them, while the driver
551          * writes to all other MTT entries. (The variable
552          * dev->caps.mtt_entry_sz below is really the MTT segment
553          * size, not the raw entry size)
554          */
555         dev->caps.reserved_mtts =
556                 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
557                       dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
558
559         err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
560                                   init_hca->mtt_base,
561                                   dev->caps.mtt_entry_sz,
562                                   dev->caps.num_mtt_segs,
563                                   dev->caps.reserved_mtts, 1, 0);
564         if (err) {
565                 mlx4_err(dev, "Failed to map MTT context memory, aborting.\n");
566                 goto err_unmap_eq;
567         }
568
569         err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
570                                   init_hca->dmpt_base,
571                                   dev_cap->dmpt_entry_sz,
572                                   dev->caps.num_mpts,
573                                   dev->caps.reserved_mrws, 1, 1);
574         if (err) {
575                 mlx4_err(dev, "Failed to map dMPT context memory, aborting.\n");
576                 goto err_unmap_mtt;
577         }
578
579         err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
580                                   init_hca->qpc_base,
581                                   dev_cap->qpc_entry_sz,
582                                   dev->caps.num_qps,
583                                   dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
584                                   0, 0);
585         if (err) {
586                 mlx4_err(dev, "Failed to map QP context memory, aborting.\n");
587                 goto err_unmap_dmpt;
588         }
589
590         err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
591                                   init_hca->auxc_base,
592                                   dev_cap->aux_entry_sz,
593                                   dev->caps.num_qps,
594                                   dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
595                                   0, 0);
596         if (err) {
597                 mlx4_err(dev, "Failed to map AUXC context memory, aborting.\n");
598                 goto err_unmap_qp;
599         }
600
601         err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
602                                   init_hca->altc_base,
603                                   dev_cap->altc_entry_sz,
604                                   dev->caps.num_qps,
605                                   dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
606                                   0, 0);
607         if (err) {
608                 mlx4_err(dev, "Failed to map ALTC context memory, aborting.\n");
609                 goto err_unmap_auxc;
610         }
611
612         err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
613                                   init_hca->rdmarc_base,
614                                   dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
615                                   dev->caps.num_qps,
616                                   dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
617                                   0, 0);
618         if (err) {
619                 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
620                 goto err_unmap_altc;
621         }
622
623         err = mlx4_init_icm_table(dev, &priv->cq_table.table,
624                                   init_hca->cqc_base,
625                                   dev_cap->cqc_entry_sz,
626                                   dev->caps.num_cqs,
627                                   dev->caps.reserved_cqs, 0, 0);
628         if (err) {
629                 mlx4_err(dev, "Failed to map CQ context memory, aborting.\n");
630                 goto err_unmap_rdmarc;
631         }
632
633         err = mlx4_init_icm_table(dev, &priv->srq_table.table,
634                                   init_hca->srqc_base,
635                                   dev_cap->srq_entry_sz,
636                                   dev->caps.num_srqs,
637                                   dev->caps.reserved_srqs, 0, 0);
638         if (err) {
639                 mlx4_err(dev, "Failed to map SRQ context memory, aborting.\n");
640                 goto err_unmap_cq;
641         }
642
643         /*
644          * It's not strictly required, but for simplicity just map the
645          * whole multicast group table now.  The table isn't very big
646          * and it's a lot easier than trying to track ref counts.
647          */
648         err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
649                                   init_hca->mc_base, MLX4_MGM_ENTRY_SIZE,
650                                   dev->caps.num_mgms + dev->caps.num_amgms,
651                                   dev->caps.num_mgms + dev->caps.num_amgms,
652                                   0, 0);
653         if (err) {
654                 mlx4_err(dev, "Failed to map MCG context memory, aborting.\n");
655                 goto err_unmap_srq;
656         }
657
658         return 0;
659
660 err_unmap_srq:
661         mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
662
663 err_unmap_cq:
664         mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
665
666 err_unmap_rdmarc:
667         mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
668
669 err_unmap_altc:
670         mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
671
672 err_unmap_auxc:
673         mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
674
675 err_unmap_qp:
676         mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
677
678 err_unmap_dmpt:
679         mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
680
681 err_unmap_mtt:
682         mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
683
684 err_unmap_eq:
685         mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
686
687 err_unmap_cmpt:
688         mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
689         mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
690         mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
691         mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
692
693 err_unmap_aux:
694         mlx4_UNMAP_ICM_AUX(dev);
695
696 err_free_aux:
697         mlx4_free_icm(dev, priv->fw.aux_icm, 0);
698
699         return err;
700 }
701
702 static void mlx4_free_icms(struct mlx4_dev *dev)
703 {
704         struct mlx4_priv *priv = mlx4_priv(dev);
705
706         mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
707         mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
708         mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
709         mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
710         mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
711         mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
712         mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
713         mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
714         mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
715         mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
716         mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
717         mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
718         mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
719         mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
720
721         mlx4_UNMAP_ICM_AUX(dev);
722         mlx4_free_icm(dev, priv->fw.aux_icm, 0);
723 }
724
725 static int map_bf_area(struct mlx4_dev *dev)
726 {
727         struct mlx4_priv *priv = mlx4_priv(dev);
728         resource_size_t bf_start;
729         resource_size_t bf_len;
730         int err = 0;
731
732         bf_start = pci_resource_start(dev->pdev, 2) + (dev->caps.num_uars << PAGE_SHIFT);
733         bf_len = pci_resource_len(dev->pdev, 2) - (dev->caps.num_uars << PAGE_SHIFT);
734         priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
735         if (!priv->bf_mapping)
736                 err = -ENOMEM;
737
738         return err;
739 }
740
741 static void unmap_bf_area(struct mlx4_dev *dev)
742 {
743         if (mlx4_priv(dev)->bf_mapping)
744                 io_mapping_free(mlx4_priv(dev)->bf_mapping);
745 }
746
747 static void mlx4_close_hca(struct mlx4_dev *dev)
748 {
749         unmap_bf_area(dev);
750         mlx4_CLOSE_HCA(dev, 0);
751         mlx4_free_icms(dev);
752         mlx4_UNMAP_FA(dev);
753         mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
754 }
755
756 static int mlx4_init_hca(struct mlx4_dev *dev)
757 {
758         struct mlx4_priv          *priv = mlx4_priv(dev);
759         struct mlx4_adapter        adapter;
760         struct mlx4_dev_cap        dev_cap;
761         struct mlx4_mod_stat_cfg   mlx4_cfg;
762         struct mlx4_profile        profile;
763         struct mlx4_init_hca_param init_hca;
764         u64 icm_size;
765         int err;
766
767         err = mlx4_QUERY_FW(dev);
768         if (err) {
769                 if (err == -EACCES)
770                         mlx4_info(dev, "non-primary physical function, skipping.\n");
771                 else
772                         mlx4_err(dev, "QUERY_FW command failed, aborting.\n");
773                 return err;
774         }
775
776         err = mlx4_load_fw(dev);
777         if (err) {
778                 mlx4_err(dev, "Failed to start FW, aborting.\n");
779                 return err;
780         }
781
782         mlx4_cfg.log_pg_sz_m = 1;
783         mlx4_cfg.log_pg_sz = 0;
784         err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
785         if (err)
786                 mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
787
788         err = mlx4_dev_cap(dev, &dev_cap);
789         if (err) {
790                 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
791                 goto err_stop_fw;
792         }
793
794         profile = default_profile;
795
796         icm_size = mlx4_make_profile(dev, &profile, &dev_cap, &init_hca);
797         if ((long long) icm_size < 0) {
798                 err = icm_size;
799                 goto err_stop_fw;
800         }
801
802         if (map_bf_area(dev))
803                 mlx4_dbg(dev, "Failed to map blue flame area\n");
804
805         init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
806
807         err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
808         if (err)
809                 goto err_stop_fw;
810
811         err = mlx4_INIT_HCA(dev, &init_hca);
812         if (err) {
813                 mlx4_err(dev, "INIT_HCA command failed, aborting.\n");
814                 goto err_free_icm;
815         }
816
817         err = mlx4_QUERY_ADAPTER(dev, &adapter);
818         if (err) {
819                 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting.\n");
820                 goto err_close;
821         }
822
823         priv->eq_table.inta_pin = adapter.inta_pin;
824         memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
825
826         return 0;
827
828 err_close:
829         mlx4_CLOSE_HCA(dev, 0);
830
831 err_free_icm:
832         mlx4_free_icms(dev);
833
834 err_stop_fw:
835         unmap_bf_area(dev);
836         mlx4_UNMAP_FA(dev);
837         mlx4_free_icm(dev, priv->fw.fw_icm, 0);
838
839         return err;
840 }
841
842 static int mlx4_setup_hca(struct mlx4_dev *dev)
843 {
844         struct mlx4_priv *priv = mlx4_priv(dev);
845         int err;
846         int port;
847         __be32 ib_port_default_caps;
848
849         err = mlx4_init_uar_table(dev);
850         if (err) {
851                 mlx4_err(dev, "Failed to initialize "
852                          "user access region table, aborting.\n");
853                 return err;
854         }
855
856         err = mlx4_uar_alloc(dev, &priv->driver_uar);
857         if (err) {
858                 mlx4_err(dev, "Failed to allocate driver access region, "
859                          "aborting.\n");
860                 goto err_uar_table_free;
861         }
862
863         priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
864         if (!priv->kar) {
865                 mlx4_err(dev, "Couldn't map kernel access region, "
866                          "aborting.\n");
867                 err = -ENOMEM;
868                 goto err_uar_free;
869         }
870
871         err = mlx4_init_pd_table(dev);
872         if (err) {
873                 mlx4_err(dev, "Failed to initialize "
874                          "protection domain table, aborting.\n");
875                 goto err_kar_unmap;
876         }
877
878         err = mlx4_init_mr_table(dev);
879         if (err) {
880                 mlx4_err(dev, "Failed to initialize "
881                          "memory region table, aborting.\n");
882                 goto err_pd_table_free;
883         }
884
885         err = mlx4_init_eq_table(dev);
886         if (err) {
887                 mlx4_err(dev, "Failed to initialize "
888                          "event queue table, aborting.\n");
889                 goto err_mr_table_free;
890         }
891
892         err = mlx4_cmd_use_events(dev);
893         if (err) {
894                 mlx4_err(dev, "Failed to switch to event-driven "
895                          "firmware commands, aborting.\n");
896                 goto err_eq_table_free;
897         }
898
899         err = mlx4_NOP(dev);
900         if (err) {
901                 if (dev->flags & MLX4_FLAG_MSI_X) {
902                         mlx4_warn(dev, "NOP command failed to generate MSI-X "
903                                   "interrupt IRQ %d).\n",
904                                   priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
905                         mlx4_warn(dev, "Trying again without MSI-X.\n");
906                 } else {
907                         mlx4_err(dev, "NOP command failed to generate interrupt "
908                                  "(IRQ %d), aborting.\n",
909                                  priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
910                         mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
911                 }
912
913                 goto err_cmd_poll;
914         }
915
916         mlx4_dbg(dev, "NOP command IRQ test passed\n");
917
918         err = mlx4_init_cq_table(dev);
919         if (err) {
920                 mlx4_err(dev, "Failed to initialize "
921                          "completion queue table, aborting.\n");
922                 goto err_cmd_poll;
923         }
924
925         err = mlx4_init_srq_table(dev);
926         if (err) {
927                 mlx4_err(dev, "Failed to initialize "
928                          "shared receive queue table, aborting.\n");
929                 goto err_cq_table_free;
930         }
931
932         err = mlx4_init_qp_table(dev);
933         if (err) {
934                 mlx4_err(dev, "Failed to initialize "
935                          "queue pair table, aborting.\n");
936                 goto err_srq_table_free;
937         }
938
939         err = mlx4_init_mcg_table(dev);
940         if (err) {
941                 mlx4_err(dev, "Failed to initialize "
942                          "multicast group table, aborting.\n");
943                 goto err_qp_table_free;
944         }
945
946         for (port = 1; port <= dev->caps.num_ports; port++) {
947                 ib_port_default_caps = 0;
948                 err = mlx4_get_port_ib_caps(dev, port, &ib_port_default_caps);
949                 if (err)
950                         mlx4_warn(dev, "failed to get port %d default "
951                                   "ib capabilities (%d). Continuing with "
952                                   "caps = 0\n", port, err);
953                 dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
954                 err = mlx4_SET_PORT(dev, port);
955                 if (err) {
956                         mlx4_err(dev, "Failed to set port %d, aborting\n",
957                                 port);
958                         goto err_mcg_table_free;
959                 }
960         }
961
962         return 0;
963
964 err_mcg_table_free:
965         mlx4_cleanup_mcg_table(dev);
966
967 err_qp_table_free:
968         mlx4_cleanup_qp_table(dev);
969
970 err_srq_table_free:
971         mlx4_cleanup_srq_table(dev);
972
973 err_cq_table_free:
974         mlx4_cleanup_cq_table(dev);
975
976 err_cmd_poll:
977         mlx4_cmd_use_polling(dev);
978
979 err_eq_table_free:
980         mlx4_cleanup_eq_table(dev);
981
982 err_mr_table_free:
983         mlx4_cleanup_mr_table(dev);
984
985 err_pd_table_free:
986         mlx4_cleanup_pd_table(dev);
987
988 err_kar_unmap:
989         iounmap(priv->kar);
990
991 err_uar_free:
992         mlx4_uar_free(dev, &priv->driver_uar);
993
994 err_uar_table_free:
995         mlx4_cleanup_uar_table(dev);
996         return err;
997 }
998
999 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
1000 {
1001         struct mlx4_priv *priv = mlx4_priv(dev);
1002         struct msix_entry *entries;
1003         int nreq = min_t(int, dev->caps.num_ports *
1004                          min_t(int, num_online_cpus() + 1, MAX_MSIX_P_PORT)
1005                                 + MSIX_LEGACY_SZ, MAX_MSIX);
1006         int err;
1007         int i;
1008
1009         if (msi_x) {
1010                 nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
1011                              nreq);
1012                 entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
1013                 if (!entries)
1014                         goto no_msi;
1015
1016                 for (i = 0; i < nreq; ++i)
1017                         entries[i].entry = i;
1018
1019         retry:
1020                 err = pci_enable_msix(dev->pdev, entries, nreq);
1021                 if (err) {
1022                         /* Try again if at least 2 vectors are available */
1023                         if (err > 1) {
1024                                 mlx4_info(dev, "Requested %d vectors, "
1025                                           "but only %d MSI-X vectors available, "
1026                                           "trying again\n", nreq, err);
1027                                 nreq = err;
1028                                 goto retry;
1029                         }
1030                         kfree(entries);
1031                         goto no_msi;
1032                 }
1033
1034                 if (nreq <
1035                     MSIX_LEGACY_SZ + dev->caps.num_ports * MIN_MSIX_P_PORT) {
1036                         /*Working in legacy mode , all EQ's shared*/
1037                         dev->caps.comp_pool           = 0;
1038                         dev->caps.num_comp_vectors = nreq - 1;
1039                 } else {
1040                         dev->caps.comp_pool           = nreq - MSIX_LEGACY_SZ;
1041                         dev->caps.num_comp_vectors = MSIX_LEGACY_SZ - 1;
1042                 }
1043                 for (i = 0; i < nreq; ++i)
1044                         priv->eq_table.eq[i].irq = entries[i].vector;
1045
1046                 dev->flags |= MLX4_FLAG_MSI_X;
1047
1048                 kfree(entries);
1049                 return;
1050         }
1051
1052 no_msi:
1053         dev->caps.num_comp_vectors = 1;
1054         dev->caps.comp_pool        = 0;
1055
1056         for (i = 0; i < 2; ++i)
1057                 priv->eq_table.eq[i].irq = dev->pdev->irq;
1058 }
1059
1060 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
1061 {
1062         struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
1063         int err = 0;
1064
1065         info->dev = dev;
1066         info->port = port;
1067         mlx4_init_mac_table(dev, &info->mac_table);
1068         mlx4_init_vlan_table(dev, &info->vlan_table);
1069
1070         sprintf(info->dev_name, "mlx4_port%d", port);
1071         info->port_attr.attr.name = info->dev_name;
1072         info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
1073         info->port_attr.show      = show_port_type;
1074         info->port_attr.store     = set_port_type;
1075         sysfs_attr_init(&info->port_attr.attr);
1076
1077         err = device_create_file(&dev->pdev->dev, &info->port_attr);
1078         if (err) {
1079                 mlx4_err(dev, "Failed to create file for port %d\n", port);
1080                 info->port = -1;
1081         }
1082
1083         return err;
1084 }
1085
1086 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
1087 {
1088         if (info->port < 0)
1089                 return;
1090
1091         device_remove_file(&info->dev->pdev->dev, &info->port_attr);
1092 }
1093
1094 static int mlx4_init_steering(struct mlx4_dev *dev)
1095 {
1096         struct mlx4_priv *priv = mlx4_priv(dev);
1097         int num_entries = dev->caps.num_ports;
1098         int i, j;
1099
1100         priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
1101         if (!priv->steer)
1102                 return -ENOMEM;
1103
1104         for (i = 0; i < num_entries; i++) {
1105                 for (j = 0; j < MLX4_NUM_STEERS; j++) {
1106                         INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
1107                         INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
1108                 }
1109                 INIT_LIST_HEAD(&priv->steer[i].high_prios);
1110         }
1111         return 0;
1112 }
1113
1114 static void mlx4_clear_steering(struct mlx4_dev *dev)
1115 {
1116         struct mlx4_priv *priv = mlx4_priv(dev);
1117         struct mlx4_steer_index *entry, *tmp_entry;
1118         struct mlx4_promisc_qp *pqp, *tmp_pqp;
1119         int num_entries = dev->caps.num_ports;
1120         int i, j;
1121
1122         for (i = 0; i < num_entries; i++) {
1123                 for (j = 0; j < MLX4_NUM_STEERS; j++) {
1124                         list_for_each_entry_safe(pqp, tmp_pqp,
1125                                                  &priv->steer[i].promisc_qps[j],
1126                                                  list) {
1127                                 list_del(&pqp->list);
1128                                 kfree(pqp);
1129                         }
1130                         list_for_each_entry_safe(entry, tmp_entry,
1131                                                  &priv->steer[i].steer_entries[j],
1132                                                  list) {
1133                                 list_del(&entry->list);
1134                                 list_for_each_entry_safe(pqp, tmp_pqp,
1135                                                          &entry->duplicates,
1136                                                          list) {
1137                                         list_del(&pqp->list);
1138                                         kfree(pqp);
1139                                 }
1140                                 kfree(entry);
1141                         }
1142                 }
1143         }
1144         kfree(priv->steer);
1145 }
1146
1147 static int __mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
1148 {
1149         struct mlx4_priv *priv;
1150         struct mlx4_dev *dev;
1151         int err;
1152         int port;
1153
1154         pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
1155
1156         err = pci_enable_device(pdev);
1157         if (err) {
1158                 dev_err(&pdev->dev, "Cannot enable PCI device, "
1159                         "aborting.\n");
1160                 return err;
1161         }
1162
1163         /*
1164          * Check for BARs.  We expect 0: 1MB
1165          */
1166         if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
1167             pci_resource_len(pdev, 0) != 1 << 20) {
1168                 dev_err(&pdev->dev, "Missing DCS, aborting.\n");
1169                 err = -ENODEV;
1170                 goto err_disable_pdev;
1171         }
1172         if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
1173                 dev_err(&pdev->dev, "Missing UAR, aborting.\n");
1174                 err = -ENODEV;
1175                 goto err_disable_pdev;
1176         }
1177
1178         err = pci_request_regions(pdev, DRV_NAME);
1179         if (err) {
1180                 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
1181                 goto err_disable_pdev;
1182         }
1183
1184         pci_set_master(pdev);
1185
1186         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1187         if (err) {
1188                 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
1189                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1190                 if (err) {
1191                         dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
1192                         goto err_release_regions;
1193                 }
1194         }
1195         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1196         if (err) {
1197                 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
1198                          "consistent PCI DMA mask.\n");
1199                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1200                 if (err) {
1201                         dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
1202                                 "aborting.\n");
1203                         goto err_release_regions;
1204                 }
1205         }
1206
1207         /* Allow large DMA segments, up to the firmware limit of 1 GB */
1208         dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
1209
1210         priv = kzalloc(sizeof *priv, GFP_KERNEL);
1211         if (!priv) {
1212                 dev_err(&pdev->dev, "Device struct alloc failed, "
1213                         "aborting.\n");
1214                 err = -ENOMEM;
1215                 goto err_release_regions;
1216         }
1217
1218         dev       = &priv->dev;
1219         dev->pdev = pdev;
1220         INIT_LIST_HEAD(&priv->ctx_list);
1221         spin_lock_init(&priv->ctx_lock);
1222
1223         mutex_init(&priv->port_mutex);
1224
1225         INIT_LIST_HEAD(&priv->pgdir_list);
1226         mutex_init(&priv->pgdir_mutex);
1227
1228         pci_read_config_byte(pdev, PCI_REVISION_ID, &dev->rev_id);
1229
1230         INIT_LIST_HEAD(&priv->bf_list);
1231         mutex_init(&priv->bf_mutex);
1232
1233         /*
1234          * Now reset the HCA before we touch the PCI capabilities or
1235          * attempt a firmware command, since a boot ROM may have left
1236          * the HCA in an undefined state.
1237          */
1238         err = mlx4_reset(dev);
1239         if (err) {
1240                 mlx4_err(dev, "Failed to reset HCA, aborting.\n");
1241                 goto err_free_dev;
1242         }
1243
1244         if (mlx4_cmd_init(dev)) {
1245                 mlx4_err(dev, "Failed to init command interface, aborting.\n");
1246                 goto err_free_dev;
1247         }
1248
1249         err = mlx4_init_hca(dev);
1250         if (err)
1251                 goto err_cmd;
1252
1253         err = mlx4_alloc_eq_table(dev);
1254         if (err)
1255                 goto err_close;
1256
1257         priv->msix_ctl.pool_bm = 0;
1258         spin_lock_init(&priv->msix_ctl.pool_lock);
1259
1260         mlx4_enable_msi_x(dev);
1261
1262         err = mlx4_init_steering(dev);
1263         if (err)
1264                 goto err_free_eq;
1265
1266         err = mlx4_setup_hca(dev);
1267         if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X)) {
1268                 dev->flags &= ~MLX4_FLAG_MSI_X;
1269                 pci_disable_msix(pdev);
1270                 err = mlx4_setup_hca(dev);
1271         }
1272
1273         if (err)
1274                 goto err_steer;
1275
1276         for (port = 1; port <= dev->caps.num_ports; port++) {
1277                 err = mlx4_init_port_info(dev, port);
1278                 if (err)
1279                         goto err_port;
1280         }
1281
1282         err = mlx4_register_device(dev);
1283         if (err)
1284                 goto err_port;
1285
1286         mlx4_sense_init(dev);
1287         mlx4_start_sense(dev);
1288
1289         pci_set_drvdata(pdev, dev);
1290
1291         return 0;
1292
1293 err_port:
1294         for (--port; port >= 1; --port)
1295                 mlx4_cleanup_port_info(&priv->port[port]);
1296
1297         mlx4_cleanup_mcg_table(dev);
1298         mlx4_cleanup_qp_table(dev);
1299         mlx4_cleanup_srq_table(dev);
1300         mlx4_cleanup_cq_table(dev);
1301         mlx4_cmd_use_polling(dev);
1302         mlx4_cleanup_eq_table(dev);
1303         mlx4_cleanup_mr_table(dev);
1304         mlx4_cleanup_pd_table(dev);
1305         mlx4_cleanup_uar_table(dev);
1306
1307 err_steer:
1308         mlx4_clear_steering(dev);
1309
1310 err_free_eq:
1311         mlx4_free_eq_table(dev);
1312
1313 err_close:
1314         if (dev->flags & MLX4_FLAG_MSI_X)
1315                 pci_disable_msix(pdev);
1316
1317         mlx4_close_hca(dev);
1318
1319 err_cmd:
1320         mlx4_cmd_cleanup(dev);
1321
1322 err_free_dev:
1323         kfree(priv);
1324
1325 err_release_regions:
1326         pci_release_regions(pdev);
1327
1328 err_disable_pdev:
1329         pci_disable_device(pdev);
1330         pci_set_drvdata(pdev, NULL);
1331         return err;
1332 }
1333
1334 static int __devinit mlx4_init_one(struct pci_dev *pdev,
1335                                    const struct pci_device_id *id)
1336 {
1337         printk_once(KERN_INFO "%s", mlx4_version);
1338
1339         return __mlx4_init_one(pdev, id);
1340 }
1341
1342 static void mlx4_remove_one(struct pci_dev *pdev)
1343 {
1344         struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
1345         struct mlx4_priv *priv = mlx4_priv(dev);
1346         int p;
1347
1348         if (dev) {
1349                 mlx4_stop_sense(dev);
1350                 mlx4_unregister_device(dev);
1351
1352                 for (p = 1; p <= dev->caps.num_ports; p++) {
1353                         mlx4_cleanup_port_info(&priv->port[p]);
1354                         mlx4_CLOSE_PORT(dev, p);
1355                 }
1356
1357                 mlx4_cleanup_mcg_table(dev);
1358                 mlx4_cleanup_qp_table(dev);
1359                 mlx4_cleanup_srq_table(dev);
1360                 mlx4_cleanup_cq_table(dev);
1361                 mlx4_cmd_use_polling(dev);
1362                 mlx4_cleanup_eq_table(dev);
1363                 mlx4_cleanup_mr_table(dev);
1364                 mlx4_cleanup_pd_table(dev);
1365
1366                 iounmap(priv->kar);
1367                 mlx4_uar_free(dev, &priv->driver_uar);
1368                 mlx4_cleanup_uar_table(dev);
1369                 mlx4_clear_steering(dev);
1370                 mlx4_free_eq_table(dev);
1371                 mlx4_close_hca(dev);
1372                 mlx4_cmd_cleanup(dev);
1373
1374                 if (dev->flags & MLX4_FLAG_MSI_X)
1375                         pci_disable_msix(pdev);
1376
1377                 kfree(priv);
1378                 pci_release_regions(pdev);
1379                 pci_disable_device(pdev);
1380                 pci_set_drvdata(pdev, NULL);
1381         }
1382 }
1383
1384 int mlx4_restart_one(struct pci_dev *pdev)
1385 {
1386         mlx4_remove_one(pdev);
1387         return __mlx4_init_one(pdev, NULL);
1388 }
1389
1390 static DEFINE_PCI_DEVICE_TABLE(mlx4_pci_table) = {
1391         { PCI_VDEVICE(MELLANOX, 0x6340) }, /* MT25408 "Hermon" SDR */
1392         { PCI_VDEVICE(MELLANOX, 0x634a) }, /* MT25408 "Hermon" DDR */
1393         { PCI_VDEVICE(MELLANOX, 0x6354) }, /* MT25408 "Hermon" QDR */
1394         { PCI_VDEVICE(MELLANOX, 0x6732) }, /* MT25408 "Hermon" DDR PCIe gen2 */
1395         { PCI_VDEVICE(MELLANOX, 0x673c) }, /* MT25408 "Hermon" QDR PCIe gen2 */
1396         { PCI_VDEVICE(MELLANOX, 0x6368) }, /* MT25408 "Hermon" EN 10GigE */
1397         { PCI_VDEVICE(MELLANOX, 0x6750) }, /* MT25408 "Hermon" EN 10GigE PCIe gen2 */
1398         { PCI_VDEVICE(MELLANOX, 0x6372) }, /* MT25458 ConnectX EN 10GBASE-T 10GigE */
1399         { PCI_VDEVICE(MELLANOX, 0x675a) }, /* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
1400         { PCI_VDEVICE(MELLANOX, 0x6764) }, /* MT26468 ConnectX EN 10GigE PCIe gen2*/
1401         { PCI_VDEVICE(MELLANOX, 0x6746) }, /* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
1402         { PCI_VDEVICE(MELLANOX, 0x676e) }, /* MT26478 ConnectX2 40GigE PCIe gen2 */
1403         { PCI_VDEVICE(MELLANOX, 0x1002) }, /* MT25400 Family [ConnectX-2 Virtual Function] */
1404         { PCI_VDEVICE(MELLANOX, 0x1003) }, /* MT27500 Family [ConnectX-3] */
1405         { PCI_VDEVICE(MELLANOX, 0x1004) }, /* MT27500 Family [ConnectX-3 Virtual Function] */
1406         { PCI_VDEVICE(MELLANOX, 0x1005) }, /* MT27510 Family */
1407         { PCI_VDEVICE(MELLANOX, 0x1006) }, /* MT27511 Family */
1408         { PCI_VDEVICE(MELLANOX, 0x1007) }, /* MT27520 Family */
1409         { PCI_VDEVICE(MELLANOX, 0x1008) }, /* MT27521 Family */
1410         { PCI_VDEVICE(MELLANOX, 0x1009) }, /* MT27530 Family */
1411         { PCI_VDEVICE(MELLANOX, 0x100a) }, /* MT27531 Family */
1412         { PCI_VDEVICE(MELLANOX, 0x100b) }, /* MT27540 Family */
1413         { PCI_VDEVICE(MELLANOX, 0x100c) }, /* MT27541 Family */
1414         { PCI_VDEVICE(MELLANOX, 0x100d) }, /* MT27550 Family */
1415         { PCI_VDEVICE(MELLANOX, 0x100e) }, /* MT27551 Family */
1416         { PCI_VDEVICE(MELLANOX, 0x100f) }, /* MT27560 Family */
1417         { PCI_VDEVICE(MELLANOX, 0x1010) }, /* MT27561 Family */
1418         { 0, }
1419 };
1420
1421 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
1422
1423 static struct pci_driver mlx4_driver = {
1424         .name           = DRV_NAME,
1425         .id_table       = mlx4_pci_table,
1426         .probe          = mlx4_init_one,
1427         .remove         = __devexit_p(mlx4_remove_one)
1428 };
1429
1430 static int __init mlx4_verify_params(void)
1431 {
1432         if ((log_num_mac < 0) || (log_num_mac > 7)) {
1433                 pr_warning("mlx4_core: bad num_mac: %d\n", log_num_mac);
1434                 return -1;
1435         }
1436
1437         if ((log_num_vlan < 0) || (log_num_vlan > 7)) {
1438                 pr_warning("mlx4_core: bad num_vlan: %d\n", log_num_vlan);
1439                 return -1;
1440         }
1441
1442         if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
1443                 pr_warning("mlx4_core: bad log_mtts_per_seg: %d\n", log_mtts_per_seg);
1444                 return -1;
1445         }
1446
1447         return 0;
1448 }
1449
1450 static int __init mlx4_init(void)
1451 {
1452         int ret;
1453
1454         if (mlx4_verify_params())
1455                 return -EINVAL;
1456
1457         mlx4_catas_init();
1458
1459         mlx4_wq = create_singlethread_workqueue("mlx4");
1460         if (!mlx4_wq)
1461                 return -ENOMEM;
1462
1463         ret = pci_register_driver(&mlx4_driver);
1464         return ret < 0 ? ret : 0;
1465 }
1466
1467 static void __exit mlx4_cleanup(void)
1468 {
1469         pci_unregister_driver(&mlx4_driver);
1470         destroy_workqueue(mlx4_wq);
1471 }
1472
1473 module_init(mlx4_init);
1474 module_exit(mlx4_cleanup);