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>
42 #include <linux/io-mapping.h>
43 #include <linux/delay.h>
45 #include <linux/mlx4/device.h>
46 #include <linux/mlx4/doorbell.h>
52 MODULE_AUTHOR("Roland Dreier");
53 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
54 MODULE_LICENSE("Dual BSD/GPL");
55 MODULE_VERSION(DRV_VERSION);
57 struct workqueue_struct *mlx4_wq;
59 #ifdef CONFIG_MLX4_DEBUG
61 int mlx4_debug_level = 0;
62 module_param_named(debug_level, mlx4_debug_level, int, 0644);
63 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
65 #endif /* CONFIG_MLX4_DEBUG */
70 module_param(msi_x, int, 0444);
71 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
73 #else /* CONFIG_PCI_MSI */
77 #endif /* CONFIG_PCI_MSI */
80 module_param(num_vfs, int, 0444);
81 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0");
84 module_param(probe_vf, int, 0644);
85 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)");
87 int mlx4_log_num_mgm_entry_size = 10;
88 module_param_named(log_num_mgm_entry_size,
89 mlx4_log_num_mgm_entry_size, int, 0444);
90 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
91 " of qp per mcg, for example:"
92 " 10 gives 248.range: 9<="
93 " log_num_mgm_entry_size <= 12");
95 #define MLX4_VF (1 << 0)
97 #define HCA_GLOBAL_CAP_MASK 0
98 #define PF_CONTEXT_BEHAVIOUR_MASK 0
100 static char mlx4_version[] __devinitdata =
101 DRV_NAME ": Mellanox ConnectX core driver v"
102 DRV_VERSION " (" DRV_RELDATE ")\n";
104 static struct mlx4_profile default_profile = {
107 .rdmarc_per_qp = 1 << 4,
111 .num_mtt = 1 << 20, /* It is really num mtt segements */
114 static int log_num_mac = 7;
115 module_param_named(log_num_mac, log_num_mac, int, 0444);
116 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
118 static int log_num_vlan;
119 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
120 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
121 /* Log2 max number of VLANs per ETH port (0-7) */
122 #define MLX4_LOG_NUM_VLANS 7
124 static bool use_prio;
125 module_param_named(use_prio, use_prio, bool, 0444);
126 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports "
129 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
130 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
131 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
133 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
134 static int arr_argc = 2;
135 module_param_array(port_type_array, int, &arr_argc, 0444);
136 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
137 "1 for IB, 2 for Ethernet");
139 struct mlx4_port_config {
140 struct list_head list;
141 enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
142 struct pci_dev *pdev;
145 int mlx4_check_port_params(struct mlx4_dev *dev,
146 enum mlx4_port_type *port_type)
150 for (i = 0; i < dev->caps.num_ports - 1; i++) {
151 if (port_type[i] != port_type[i + 1]) {
152 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
153 mlx4_err(dev, "Only same port types supported "
154 "on this HCA, aborting.\n");
157 if (port_type[i] == MLX4_PORT_TYPE_ETH &&
158 port_type[i + 1] == MLX4_PORT_TYPE_IB)
163 for (i = 0; i < dev->caps.num_ports; i++) {
164 if (!(port_type[i] & dev->caps.supported_type[i+1])) {
165 mlx4_err(dev, "Requested port type for port %d is not "
166 "supported on this HCA\n", i + 1);
173 static void mlx4_set_port_mask(struct mlx4_dev *dev)
177 for (i = 1; i <= dev->caps.num_ports; ++i)
178 dev->caps.port_mask[i] = dev->caps.port_type[i];
181 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
186 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
188 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
192 if (dev_cap->min_page_sz > PAGE_SIZE) {
193 mlx4_err(dev, "HCA minimum page size of %d bigger than "
194 "kernel PAGE_SIZE of %ld, aborting.\n",
195 dev_cap->min_page_sz, PAGE_SIZE);
198 if (dev_cap->num_ports > MLX4_MAX_PORTS) {
199 mlx4_err(dev, "HCA has %d ports, but we only support %d, "
201 dev_cap->num_ports, MLX4_MAX_PORTS);
205 if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
206 mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than "
207 "PCI resource 2 size of 0x%llx, aborting.\n",
209 (unsigned long long) pci_resource_len(dev->pdev, 2));
213 dev->caps.num_ports = dev_cap->num_ports;
214 dev->phys_caps.num_phys_eqs = MLX4_MAX_EQ_NUM;
215 for (i = 1; i <= dev->caps.num_ports; ++i) {
216 dev->caps.vl_cap[i] = dev_cap->max_vl[i];
217 dev->caps.ib_mtu_cap[i] = dev_cap->ib_mtu[i];
218 dev->caps.gid_table_len[i] = dev_cap->max_gids[i];
219 dev->caps.pkey_table_len[i] = dev_cap->max_pkeys[i];
220 dev->caps.port_width_cap[i] = dev_cap->max_port_width[i];
221 dev->caps.eth_mtu_cap[i] = dev_cap->eth_mtu[i];
222 dev->caps.def_mac[i] = dev_cap->def_mac[i];
223 dev->caps.supported_type[i] = dev_cap->supported_port_types[i];
224 dev->caps.suggested_type[i] = dev_cap->suggested_type[i];
225 dev->caps.default_sense[i] = dev_cap->default_sense[i];
226 dev->caps.trans_type[i] = dev_cap->trans_type[i];
227 dev->caps.vendor_oui[i] = dev_cap->vendor_oui[i];
228 dev->caps.wavelength[i] = dev_cap->wavelength[i];
229 dev->caps.trans_code[i] = dev_cap->trans_code[i];
232 dev->caps.uar_page_size = PAGE_SIZE;
233 dev->caps.num_uars = dev_cap->uar_size / PAGE_SIZE;
234 dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
235 dev->caps.bf_reg_size = dev_cap->bf_reg_size;
236 dev->caps.bf_regs_per_page = dev_cap->bf_regs_per_page;
237 dev->caps.max_sq_sg = dev_cap->max_sq_sg;
238 dev->caps.max_rq_sg = dev_cap->max_rq_sg;
239 dev->caps.max_wqes = dev_cap->max_qp_sz;
240 dev->caps.max_qp_init_rdma = dev_cap->max_requester_per_qp;
241 dev->caps.max_srq_wqes = dev_cap->max_srq_sz;
242 dev->caps.max_srq_sge = dev_cap->max_rq_sg - 1;
243 dev->caps.reserved_srqs = dev_cap->reserved_srqs;
244 dev->caps.max_sq_desc_sz = dev_cap->max_sq_desc_sz;
245 dev->caps.max_rq_desc_sz = dev_cap->max_rq_desc_sz;
246 dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
248 * Subtract 1 from the limit because we need to allocate a
249 * spare CQE so the HCA HW can tell the difference between an
250 * empty CQ and a full CQ.
252 dev->caps.max_cqes = dev_cap->max_cq_sz - 1;
253 dev->caps.reserved_cqs = dev_cap->reserved_cqs;
254 dev->caps.reserved_eqs = dev_cap->reserved_eqs;
255 dev->caps.reserved_mtts = dev_cap->reserved_mtts;
256 dev->caps.reserved_mrws = dev_cap->reserved_mrws;
258 /* The first 128 UARs are used for EQ doorbells */
259 dev->caps.reserved_uars = max_t(int, 128, dev_cap->reserved_uars);
260 dev->caps.reserved_pds = dev_cap->reserved_pds;
261 dev->caps.reserved_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
262 dev_cap->reserved_xrcds : 0;
263 dev->caps.max_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
264 dev_cap->max_xrcds : 0;
265 dev->caps.mtt_entry_sz = dev_cap->mtt_entry_sz;
267 dev->caps.max_msg_sz = dev_cap->max_msg_sz;
268 dev->caps.page_size_cap = ~(u32) (dev_cap->min_page_sz - 1);
269 dev->caps.flags = dev_cap->flags;
270 dev->caps.flags2 = dev_cap->flags2;
271 dev->caps.bmme_flags = dev_cap->bmme_flags;
272 dev->caps.reserved_lkey = dev_cap->reserved_lkey;
273 dev->caps.stat_rate_support = dev_cap->stat_rate_support;
274 dev->caps.max_gso_sz = dev_cap->max_gso_sz;
275 dev->caps.max_rss_tbl_sz = dev_cap->max_rss_tbl_sz;
277 /* Sense port always allowed on supported devices for ConnectX1 and 2 */
278 if (dev->pdev->device != 0x1003)
279 dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
281 dev->caps.log_num_macs = log_num_mac;
282 dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
283 dev->caps.log_num_prios = use_prio ? 3 : 0;
285 for (i = 1; i <= dev->caps.num_ports; ++i) {
286 dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
287 if (dev->caps.supported_type[i]) {
288 /* if only ETH is supported - assign ETH */
289 if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
290 dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
291 /* if only IB is supported,
292 * assign IB only if SRIOV is off*/
293 else if (dev->caps.supported_type[i] ==
295 if (dev->flags & MLX4_FLAG_SRIOV)
296 dev->caps.port_type[i] =
299 dev->caps.port_type[i] =
301 /* if IB and ETH are supported,
302 * first of all check if SRIOV is on */
303 } else if (dev->flags & MLX4_FLAG_SRIOV)
304 dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
306 /* In non-SRIOV mode, we set the port type
307 * according to user selection of port type,
308 * if usere selected none, take the FW hint */
309 if (port_type_array[i-1] == MLX4_PORT_TYPE_NONE)
310 dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
311 MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
313 dev->caps.port_type[i] = port_type_array[i-1];
317 * Link sensing is allowed on the port if 3 conditions are true:
318 * 1. Both protocols are supported on the port.
319 * 2. Different types are supported on the port
320 * 3. FW declared that it supports link sensing
322 mlx4_priv(dev)->sense.sense_allowed[i] =
323 ((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
324 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
325 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
328 * If "default_sense" bit is set, we move the port to "AUTO" mode
329 * and perform sense_port FW command to try and set the correct
330 * port type from beginning
332 if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
333 enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
334 dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
335 mlx4_SENSE_PORT(dev, i, &sensed_port);
336 if (sensed_port != MLX4_PORT_TYPE_NONE)
337 dev->caps.port_type[i] = sensed_port;
339 dev->caps.possible_type[i] = dev->caps.port_type[i];
342 if (dev->caps.log_num_macs > dev_cap->log_max_macs[i]) {
343 dev->caps.log_num_macs = dev_cap->log_max_macs[i];
344 mlx4_warn(dev, "Requested number of MACs is too much "
345 "for port %d, reducing to %d.\n",
346 i, 1 << dev->caps.log_num_macs);
348 if (dev->caps.log_num_vlans > dev_cap->log_max_vlans[i]) {
349 dev->caps.log_num_vlans = dev_cap->log_max_vlans[i];
350 mlx4_warn(dev, "Requested number of VLANs is too much "
351 "for port %d, reducing to %d.\n",
352 i, 1 << dev->caps.log_num_vlans);
356 dev->caps.max_counters = 1 << ilog2(dev_cap->max_counters);
358 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
359 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
360 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
361 (1 << dev->caps.log_num_macs) *
362 (1 << dev->caps.log_num_vlans) *
363 (1 << dev->caps.log_num_prios) *
365 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
367 dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
368 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
369 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
370 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
374 /*The function checks if there are live vf, return the num of them*/
375 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
377 struct mlx4_priv *priv = mlx4_priv(dev);
378 struct mlx4_slave_state *s_state;
382 for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
383 s_state = &priv->mfunc.master.slave_state[i];
384 if (s_state->active && s_state->last_cmd !=
385 MLX4_COMM_CMD_RESET) {
386 mlx4_warn(dev, "%s: slave: %d is still active\n",
394 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
396 struct mlx4_priv *priv = mlx4_priv(dev);
397 struct mlx4_slave_state *s_slave;
399 if (!mlx4_is_master(dev))
402 s_slave = &priv->mfunc.master.slave_state[slave];
403 return !!s_slave->active;
405 EXPORT_SYMBOL(mlx4_is_slave_active);
407 static int mlx4_slave_cap(struct mlx4_dev *dev)
411 struct mlx4_dev_cap dev_cap;
412 struct mlx4_func_cap func_cap;
413 struct mlx4_init_hca_param hca_param;
416 memset(&hca_param, 0, sizeof(hca_param));
417 err = mlx4_QUERY_HCA(dev, &hca_param);
419 mlx4_err(dev, "QUERY_HCA command failed, aborting.\n");
423 /*fail if the hca has an unknown capability */
424 if ((hca_param.global_caps | HCA_GLOBAL_CAP_MASK) !=
425 HCA_GLOBAL_CAP_MASK) {
426 mlx4_err(dev, "Unknown hca global capabilities\n");
430 mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
432 memset(&dev_cap, 0, sizeof(dev_cap));
433 err = mlx4_dev_cap(dev, &dev_cap);
435 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
439 page_size = ~dev->caps.page_size_cap + 1;
440 mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
441 if (page_size > PAGE_SIZE) {
442 mlx4_err(dev, "HCA minimum page size of %d bigger than "
443 "kernel PAGE_SIZE of %ld, aborting.\n",
444 page_size, PAGE_SIZE);
448 /* slave gets uar page size from QUERY_HCA fw command */
449 dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
451 /* TODO: relax this assumption */
452 if (dev->caps.uar_page_size != PAGE_SIZE) {
453 mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
454 dev->caps.uar_page_size, PAGE_SIZE);
458 memset(&func_cap, 0, sizeof(func_cap));
459 err = mlx4_QUERY_FUNC_CAP(dev, &func_cap);
461 mlx4_err(dev, "QUERY_FUNC_CAP command failed, aborting.\n");
465 if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
466 PF_CONTEXT_BEHAVIOUR_MASK) {
467 mlx4_err(dev, "Unknown pf context behaviour\n");
471 dev->caps.num_ports = func_cap.num_ports;
472 dev->caps.num_qps = func_cap.qp_quota;
473 dev->caps.num_srqs = func_cap.srq_quota;
474 dev->caps.num_cqs = func_cap.cq_quota;
475 dev->caps.num_eqs = func_cap.max_eq;
476 dev->caps.reserved_eqs = func_cap.reserved_eq;
477 dev->caps.num_mpts = func_cap.mpt_quota;
478 dev->caps.num_mtts = func_cap.mtt_quota;
479 dev->caps.num_pds = MLX4_NUM_PDS;
480 dev->caps.num_mgms = 0;
481 dev->caps.num_amgms = 0;
483 for (i = 1; i <= dev->caps.num_ports; ++i)
484 dev->caps.port_mask[i] = dev->caps.port_type[i];
486 if (dev->caps.num_ports > MLX4_MAX_PORTS) {
487 mlx4_err(dev, "HCA has %d ports, but we only support %d, "
488 "aborting.\n", dev->caps.num_ports, MLX4_MAX_PORTS);
492 if (dev->caps.uar_page_size * (dev->caps.num_uars -
493 dev->caps.reserved_uars) >
494 pci_resource_len(dev->pdev, 2)) {
495 mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than "
496 "PCI resource 2 size of 0x%llx, aborting.\n",
497 dev->caps.uar_page_size * dev->caps.num_uars,
498 (unsigned long long) pci_resource_len(dev->pdev, 2));
503 mlx4_warn(dev, "sqp_demux:%d\n", dev->caps.sqp_demux);
504 mlx4_warn(dev, "num_uars:%d reserved_uars:%d uar region:0x%x bar2:0x%llx\n",
505 dev->caps.num_uars, dev->caps.reserved_uars,
506 dev->caps.uar_page_size * dev->caps.num_uars,
507 pci_resource_len(dev->pdev, 2));
508 mlx4_warn(dev, "num_eqs:%d reserved_eqs:%d\n", dev->caps.num_eqs,
509 dev->caps.reserved_eqs);
510 mlx4_warn(dev, "num_pds:%d reserved_pds:%d slave_pd_shift:%d pd_base:%d\n",
511 dev->caps.num_pds, dev->caps.reserved_pds,
512 dev->caps.slave_pd_shift, dev->caps.pd_base);
518 * Change the port configuration of the device.
519 * Every user of this function must hold the port mutex.
521 int mlx4_change_port_types(struct mlx4_dev *dev,
522 enum mlx4_port_type *port_types)
528 for (port = 0; port < dev->caps.num_ports; port++) {
529 /* Change the port type only if the new type is different
530 * from the current, and not set to Auto */
531 if (port_types[port] != dev->caps.port_type[port + 1])
535 mlx4_unregister_device(dev);
536 for (port = 1; port <= dev->caps.num_ports; port++) {
537 mlx4_CLOSE_PORT(dev, port);
538 dev->caps.port_type[port] = port_types[port - 1];
539 err = mlx4_SET_PORT(dev, port);
541 mlx4_err(dev, "Failed to set port %d, "
546 mlx4_set_port_mask(dev);
547 err = mlx4_register_device(dev);
554 static ssize_t show_port_type(struct device *dev,
555 struct device_attribute *attr,
558 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
560 struct mlx4_dev *mdev = info->dev;
564 (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
566 if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
567 sprintf(buf, "auto (%s)\n", type);
569 sprintf(buf, "%s\n", type);
574 static ssize_t set_port_type(struct device *dev,
575 struct device_attribute *attr,
576 const char *buf, size_t count)
578 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
580 struct mlx4_dev *mdev = info->dev;
581 struct mlx4_priv *priv = mlx4_priv(mdev);
582 enum mlx4_port_type types[MLX4_MAX_PORTS];
583 enum mlx4_port_type new_types[MLX4_MAX_PORTS];
587 if (!strcmp(buf, "ib\n"))
588 info->tmp_type = MLX4_PORT_TYPE_IB;
589 else if (!strcmp(buf, "eth\n"))
590 info->tmp_type = MLX4_PORT_TYPE_ETH;
591 else if (!strcmp(buf, "auto\n"))
592 info->tmp_type = MLX4_PORT_TYPE_AUTO;
594 mlx4_err(mdev, "%s is not supported port type\n", buf);
598 mlx4_stop_sense(mdev);
599 mutex_lock(&priv->port_mutex);
600 /* Possible type is always the one that was delivered */
601 mdev->caps.possible_type[info->port] = info->tmp_type;
603 for (i = 0; i < mdev->caps.num_ports; i++) {
604 types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
605 mdev->caps.possible_type[i+1];
606 if (types[i] == MLX4_PORT_TYPE_AUTO)
607 types[i] = mdev->caps.port_type[i+1];
610 if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
611 !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
612 for (i = 1; i <= mdev->caps.num_ports; i++) {
613 if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
614 mdev->caps.possible_type[i] = mdev->caps.port_type[i];
620 mlx4_err(mdev, "Auto sensing is not supported on this HCA. "
621 "Set only 'eth' or 'ib' for both ports "
622 "(should be the same)\n");
626 mlx4_do_sense_ports(mdev, new_types, types);
628 err = mlx4_check_port_params(mdev, new_types);
632 /* We are about to apply the changes after the configuration
633 * was verified, no need to remember the temporary types
635 for (i = 0; i < mdev->caps.num_ports; i++)
636 priv->port[i + 1].tmp_type = 0;
638 err = mlx4_change_port_types(mdev, new_types);
641 mlx4_start_sense(mdev);
642 mutex_unlock(&priv->port_mutex);
643 return err ? err : count;
654 static inline int int_to_ibta_mtu(int mtu)
657 case 256: return IB_MTU_256;
658 case 512: return IB_MTU_512;
659 case 1024: return IB_MTU_1024;
660 case 2048: return IB_MTU_2048;
661 case 4096: return IB_MTU_4096;
666 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
669 case IB_MTU_256: return 256;
670 case IB_MTU_512: return 512;
671 case IB_MTU_1024: return 1024;
672 case IB_MTU_2048: return 2048;
673 case IB_MTU_4096: return 4096;
678 static ssize_t show_port_ib_mtu(struct device *dev,
679 struct device_attribute *attr,
682 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
684 struct mlx4_dev *mdev = info->dev;
686 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
687 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
690 ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
694 static ssize_t set_port_ib_mtu(struct device *dev,
695 struct device_attribute *attr,
696 const char *buf, size_t count)
698 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
700 struct mlx4_dev *mdev = info->dev;
701 struct mlx4_priv *priv = mlx4_priv(mdev);
702 int err, port, mtu, ibta_mtu = -1;
704 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
705 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
709 err = sscanf(buf, "%d", &mtu);
711 ibta_mtu = int_to_ibta_mtu(mtu);
713 if (err <= 0 || ibta_mtu < 0) {
714 mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
718 mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
720 mlx4_stop_sense(mdev);
721 mutex_lock(&priv->port_mutex);
722 mlx4_unregister_device(mdev);
723 for (port = 1; port <= mdev->caps.num_ports; port++) {
724 mlx4_CLOSE_PORT(mdev, port);
725 err = mlx4_SET_PORT(mdev, port);
727 mlx4_err(mdev, "Failed to set port %d, "
732 err = mlx4_register_device(mdev);
734 mutex_unlock(&priv->port_mutex);
735 mlx4_start_sense(mdev);
736 return err ? err : count;
739 static int mlx4_load_fw(struct mlx4_dev *dev)
741 struct mlx4_priv *priv = mlx4_priv(dev);
744 priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
745 GFP_HIGHUSER | __GFP_NOWARN, 0);
746 if (!priv->fw.fw_icm) {
747 mlx4_err(dev, "Couldn't allocate FW area, aborting.\n");
751 err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
753 mlx4_err(dev, "MAP_FA command failed, aborting.\n");
757 err = mlx4_RUN_FW(dev);
759 mlx4_err(dev, "RUN_FW command failed, aborting.\n");
769 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
773 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
776 struct mlx4_priv *priv = mlx4_priv(dev);
780 err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
782 ((u64) (MLX4_CMPT_TYPE_QP *
783 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
784 cmpt_entry_sz, dev->caps.num_qps,
785 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
790 err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
792 ((u64) (MLX4_CMPT_TYPE_SRQ *
793 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
794 cmpt_entry_sz, dev->caps.num_srqs,
795 dev->caps.reserved_srqs, 0, 0);
799 err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
801 ((u64) (MLX4_CMPT_TYPE_CQ *
802 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
803 cmpt_entry_sz, dev->caps.num_cqs,
804 dev->caps.reserved_cqs, 0, 0);
808 num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
810 err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
812 ((u64) (MLX4_CMPT_TYPE_EQ *
813 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
814 cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
821 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
824 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
827 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
833 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
834 struct mlx4_init_hca_param *init_hca, u64 icm_size)
836 struct mlx4_priv *priv = mlx4_priv(dev);
841 err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
843 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting.\n");
847 mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory.\n",
848 (unsigned long long) icm_size >> 10,
849 (unsigned long long) aux_pages << 2);
851 priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
852 GFP_HIGHUSER | __GFP_NOWARN, 0);
853 if (!priv->fw.aux_icm) {
854 mlx4_err(dev, "Couldn't allocate aux memory, aborting.\n");
858 err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
860 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting.\n");
864 err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
866 mlx4_err(dev, "Failed to map cMPT context memory, aborting.\n");
871 num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
873 err = mlx4_init_icm_table(dev, &priv->eq_table.table,
874 init_hca->eqc_base, dev_cap->eqc_entry_sz,
875 num_eqs, num_eqs, 0, 0);
877 mlx4_err(dev, "Failed to map EQ context memory, aborting.\n");
882 * Reserved MTT entries must be aligned up to a cacheline
883 * boundary, since the FW will write to them, while the driver
884 * writes to all other MTT entries. (The variable
885 * dev->caps.mtt_entry_sz below is really the MTT segment
886 * size, not the raw entry size)
888 dev->caps.reserved_mtts =
889 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
890 dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
892 err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
894 dev->caps.mtt_entry_sz,
896 dev->caps.reserved_mtts, 1, 0);
898 mlx4_err(dev, "Failed to map MTT context memory, aborting.\n");
902 err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
904 dev_cap->dmpt_entry_sz,
906 dev->caps.reserved_mrws, 1, 1);
908 mlx4_err(dev, "Failed to map dMPT context memory, aborting.\n");
912 err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
914 dev_cap->qpc_entry_sz,
916 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
919 mlx4_err(dev, "Failed to map QP context memory, aborting.\n");
923 err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
925 dev_cap->aux_entry_sz,
927 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
930 mlx4_err(dev, "Failed to map AUXC context memory, aborting.\n");
934 err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
936 dev_cap->altc_entry_sz,
938 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
941 mlx4_err(dev, "Failed to map ALTC context memory, aborting.\n");
945 err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
946 init_hca->rdmarc_base,
947 dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
949 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
952 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
956 err = mlx4_init_icm_table(dev, &priv->cq_table.table,
958 dev_cap->cqc_entry_sz,
960 dev->caps.reserved_cqs, 0, 0);
962 mlx4_err(dev, "Failed to map CQ context memory, aborting.\n");
963 goto err_unmap_rdmarc;
966 err = mlx4_init_icm_table(dev, &priv->srq_table.table,
968 dev_cap->srq_entry_sz,
970 dev->caps.reserved_srqs, 0, 0);
972 mlx4_err(dev, "Failed to map SRQ context memory, aborting.\n");
977 * It's not strictly required, but for simplicity just map the
978 * whole multicast group table now. The table isn't very big
979 * and it's a lot easier than trying to track ref counts.
981 err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
983 mlx4_get_mgm_entry_size(dev),
984 dev->caps.num_mgms + dev->caps.num_amgms,
985 dev->caps.num_mgms + dev->caps.num_amgms,
988 mlx4_err(dev, "Failed to map MCG context memory, aborting.\n");
995 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
998 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1001 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1004 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1007 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1010 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1013 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1016 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1019 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1022 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1023 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1024 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1025 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1028 mlx4_UNMAP_ICM_AUX(dev);
1031 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1036 static void mlx4_free_icms(struct mlx4_dev *dev)
1038 struct mlx4_priv *priv = mlx4_priv(dev);
1040 mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1041 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1042 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1043 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1044 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1045 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1046 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1047 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1048 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1049 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1050 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1051 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1052 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1053 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1055 mlx4_UNMAP_ICM_AUX(dev);
1056 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1059 static void mlx4_slave_exit(struct mlx4_dev *dev)
1061 struct mlx4_priv *priv = mlx4_priv(dev);
1063 down(&priv->cmd.slave_sem);
1064 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_TIME))
1065 mlx4_warn(dev, "Failed to close slave function.\n");
1066 up(&priv->cmd.slave_sem);
1069 static int map_bf_area(struct mlx4_dev *dev)
1071 struct mlx4_priv *priv = mlx4_priv(dev);
1072 resource_size_t bf_start;
1073 resource_size_t bf_len;
1076 if (!dev->caps.bf_reg_size)
1079 bf_start = pci_resource_start(dev->pdev, 2) +
1080 (dev->caps.num_uars << PAGE_SHIFT);
1081 bf_len = pci_resource_len(dev->pdev, 2) -
1082 (dev->caps.num_uars << PAGE_SHIFT);
1083 priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1084 if (!priv->bf_mapping)
1090 static void unmap_bf_area(struct mlx4_dev *dev)
1092 if (mlx4_priv(dev)->bf_mapping)
1093 io_mapping_free(mlx4_priv(dev)->bf_mapping);
1096 static void mlx4_close_hca(struct mlx4_dev *dev)
1099 if (mlx4_is_slave(dev))
1100 mlx4_slave_exit(dev);
1102 mlx4_CLOSE_HCA(dev, 0);
1103 mlx4_free_icms(dev);
1105 mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1109 static int mlx4_init_slave(struct mlx4_dev *dev)
1111 struct mlx4_priv *priv = mlx4_priv(dev);
1112 u64 dma = (u64) priv->mfunc.vhcr_dma;
1113 int num_of_reset_retries = NUM_OF_RESET_RETRIES;
1114 int ret_from_reset = 0;
1116 u32 cmd_channel_ver;
1118 down(&priv->cmd.slave_sem);
1119 priv->cmd.max_cmds = 1;
1120 mlx4_warn(dev, "Sending reset\n");
1121 ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1123 /* if we are in the middle of flr the slave will try
1124 * NUM_OF_RESET_RETRIES times before leaving.*/
1125 if (ret_from_reset) {
1126 if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1127 msleep(SLEEP_TIME_IN_RESET);
1128 while (ret_from_reset && num_of_reset_retries) {
1129 mlx4_warn(dev, "slave is currently in the"
1130 "middle of FLR. retrying..."
1132 (NUM_OF_RESET_RETRIES -
1133 num_of_reset_retries + 1));
1135 mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET,
1137 num_of_reset_retries = num_of_reset_retries - 1;
1143 /* check the driver version - the slave I/F revision
1144 * must match the master's */
1145 slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1146 cmd_channel_ver = mlx4_comm_get_version();
1148 if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1149 MLX4_COMM_GET_IF_REV(slave_read)) {
1150 mlx4_err(dev, "slave driver version is not supported"
1151 " by the master\n");
1155 mlx4_warn(dev, "Sending vhcr0\n");
1156 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1159 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1162 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1165 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma, MLX4_COMM_TIME))
1167 up(&priv->cmd.slave_sem);
1171 mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, 0);
1172 up(&priv->cmd.slave_sem);
1176 static int mlx4_init_hca(struct mlx4_dev *dev)
1178 struct mlx4_priv *priv = mlx4_priv(dev);
1179 struct mlx4_adapter adapter;
1180 struct mlx4_dev_cap dev_cap;
1181 struct mlx4_mod_stat_cfg mlx4_cfg;
1182 struct mlx4_profile profile;
1183 struct mlx4_init_hca_param init_hca;
1187 if (!mlx4_is_slave(dev)) {
1188 err = mlx4_QUERY_FW(dev);
1191 mlx4_info(dev, "non-primary physical function, skipping.\n");
1193 mlx4_err(dev, "QUERY_FW command failed, aborting.\n");
1197 err = mlx4_load_fw(dev);
1199 mlx4_err(dev, "Failed to start FW, aborting.\n");
1203 mlx4_cfg.log_pg_sz_m = 1;
1204 mlx4_cfg.log_pg_sz = 0;
1205 err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
1207 mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
1209 err = mlx4_dev_cap(dev, &dev_cap);
1211 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
1215 profile = default_profile;
1217 icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
1219 if ((long long) icm_size < 0) {
1224 dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
1226 init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
1227 init_hca.uar_page_sz = PAGE_SHIFT - 12;
1229 err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
1233 err = mlx4_INIT_HCA(dev, &init_hca);
1235 mlx4_err(dev, "INIT_HCA command failed, aborting.\n");
1239 err = mlx4_init_slave(dev);
1241 mlx4_err(dev, "Failed to initialize slave\n");
1245 err = mlx4_slave_cap(dev);
1247 mlx4_err(dev, "Failed to obtain slave caps\n");
1252 if (map_bf_area(dev))
1253 mlx4_dbg(dev, "Failed to map blue flame area\n");
1255 /*Only the master set the ports, all the rest got it from it.*/
1256 if (!mlx4_is_slave(dev))
1257 mlx4_set_port_mask(dev);
1259 err = mlx4_QUERY_ADAPTER(dev, &adapter);
1261 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting.\n");
1265 priv->eq_table.inta_pin = adapter.inta_pin;
1266 memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
1271 mlx4_close_hca(dev);
1274 if (!mlx4_is_slave(dev))
1275 mlx4_free_icms(dev);
1278 if (!mlx4_is_slave(dev)) {
1280 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1287 static int mlx4_init_counters_table(struct mlx4_dev *dev)
1289 struct mlx4_priv *priv = mlx4_priv(dev);
1292 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1295 nent = dev->caps.max_counters;
1296 return mlx4_bitmap_init(&priv->counters_bitmap, nent, nent - 1, 0, 0);
1299 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
1301 mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
1304 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1306 struct mlx4_priv *priv = mlx4_priv(dev);
1308 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1311 *idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
1318 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1323 if (mlx4_is_mfunc(dev)) {
1324 err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
1325 RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
1326 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
1328 *idx = get_param_l(&out_param);
1332 return __mlx4_counter_alloc(dev, idx);
1334 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
1336 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1338 mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx);
1342 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1346 if (mlx4_is_mfunc(dev)) {
1347 set_param_l(&in_param, idx);
1348 mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
1349 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
1353 __mlx4_counter_free(dev, idx);
1355 EXPORT_SYMBOL_GPL(mlx4_counter_free);
1357 static int mlx4_setup_hca(struct mlx4_dev *dev)
1359 struct mlx4_priv *priv = mlx4_priv(dev);
1362 __be32 ib_port_default_caps;
1364 err = mlx4_init_uar_table(dev);
1366 mlx4_err(dev, "Failed to initialize "
1367 "user access region table, aborting.\n");
1371 err = mlx4_uar_alloc(dev, &priv->driver_uar);
1373 mlx4_err(dev, "Failed to allocate driver access region, "
1375 goto err_uar_table_free;
1378 priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
1380 mlx4_err(dev, "Couldn't map kernel access region, "
1386 err = mlx4_init_pd_table(dev);
1388 mlx4_err(dev, "Failed to initialize "
1389 "protection domain table, aborting.\n");
1393 err = mlx4_init_xrcd_table(dev);
1395 mlx4_err(dev, "Failed to initialize "
1396 "reliable connection domain table, aborting.\n");
1397 goto err_pd_table_free;
1400 err = mlx4_init_mr_table(dev);
1402 mlx4_err(dev, "Failed to initialize "
1403 "memory region table, aborting.\n");
1404 goto err_xrcd_table_free;
1407 err = mlx4_init_eq_table(dev);
1409 mlx4_err(dev, "Failed to initialize "
1410 "event queue table, aborting.\n");
1411 goto err_mr_table_free;
1414 err = mlx4_cmd_use_events(dev);
1416 mlx4_err(dev, "Failed to switch to event-driven "
1417 "firmware commands, aborting.\n");
1418 goto err_eq_table_free;
1421 err = mlx4_NOP(dev);
1423 if (dev->flags & MLX4_FLAG_MSI_X) {
1424 mlx4_warn(dev, "NOP command failed to generate MSI-X "
1425 "interrupt IRQ %d).\n",
1426 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1427 mlx4_warn(dev, "Trying again without MSI-X.\n");
1429 mlx4_err(dev, "NOP command failed to generate interrupt "
1430 "(IRQ %d), aborting.\n",
1431 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1432 mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
1438 mlx4_dbg(dev, "NOP command IRQ test passed\n");
1440 err = mlx4_init_cq_table(dev);
1442 mlx4_err(dev, "Failed to initialize "
1443 "completion queue table, aborting.\n");
1447 err = mlx4_init_srq_table(dev);
1449 mlx4_err(dev, "Failed to initialize "
1450 "shared receive queue table, aborting.\n");
1451 goto err_cq_table_free;
1454 err = mlx4_init_qp_table(dev);
1456 mlx4_err(dev, "Failed to initialize "
1457 "queue pair table, aborting.\n");
1458 goto err_srq_table_free;
1461 if (!mlx4_is_slave(dev)) {
1462 err = mlx4_init_mcg_table(dev);
1464 mlx4_err(dev, "Failed to initialize "
1465 "multicast group table, aborting.\n");
1466 goto err_qp_table_free;
1470 err = mlx4_init_counters_table(dev);
1471 if (err && err != -ENOENT) {
1472 mlx4_err(dev, "Failed to initialize counters table, aborting.\n");
1473 goto err_mcg_table_free;
1476 if (!mlx4_is_slave(dev)) {
1477 for (port = 1; port <= dev->caps.num_ports; port++) {
1478 ib_port_default_caps = 0;
1479 err = mlx4_get_port_ib_caps(dev, port,
1480 &ib_port_default_caps);
1482 mlx4_warn(dev, "failed to get port %d default "
1483 "ib capabilities (%d). Continuing "
1484 "with caps = 0\n", port, err);
1485 dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
1487 if (mlx4_is_mfunc(dev))
1488 dev->caps.port_ib_mtu[port] = IB_MTU_2048;
1490 dev->caps.port_ib_mtu[port] = IB_MTU_4096;
1492 err = mlx4_SET_PORT(dev, port);
1494 mlx4_err(dev, "Failed to set port %d, aborting\n",
1496 goto err_counters_table_free;
1503 err_counters_table_free:
1504 mlx4_cleanup_counters_table(dev);
1507 mlx4_cleanup_mcg_table(dev);
1510 mlx4_cleanup_qp_table(dev);
1513 mlx4_cleanup_srq_table(dev);
1516 mlx4_cleanup_cq_table(dev);
1519 mlx4_cmd_use_polling(dev);
1522 mlx4_cleanup_eq_table(dev);
1525 mlx4_cleanup_mr_table(dev);
1527 err_xrcd_table_free:
1528 mlx4_cleanup_xrcd_table(dev);
1531 mlx4_cleanup_pd_table(dev);
1537 mlx4_uar_free(dev, &priv->driver_uar);
1540 mlx4_cleanup_uar_table(dev);
1544 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
1546 struct mlx4_priv *priv = mlx4_priv(dev);
1547 struct msix_entry *entries;
1548 int nreq = min_t(int, dev->caps.num_ports *
1549 min_t(int, num_online_cpus() + 1, MAX_MSIX_P_PORT)
1550 + MSIX_LEGACY_SZ, MAX_MSIX);
1555 /* In multifunction mode each function gets 2 msi-X vectors
1556 * one for data path completions anf the other for asynch events
1557 * or command completions */
1558 if (mlx4_is_mfunc(dev)) {
1561 nreq = min_t(int, dev->caps.num_eqs -
1562 dev->caps.reserved_eqs, nreq);
1565 entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
1569 for (i = 0; i < nreq; ++i)
1570 entries[i].entry = i;
1573 err = pci_enable_msix(dev->pdev, entries, nreq);
1575 /* Try again if at least 2 vectors are available */
1577 mlx4_info(dev, "Requested %d vectors, "
1578 "but only %d MSI-X vectors available, "
1579 "trying again\n", nreq, err);
1588 MSIX_LEGACY_SZ + dev->caps.num_ports * MIN_MSIX_P_PORT) {
1589 /*Working in legacy mode , all EQ's shared*/
1590 dev->caps.comp_pool = 0;
1591 dev->caps.num_comp_vectors = nreq - 1;
1593 dev->caps.comp_pool = nreq - MSIX_LEGACY_SZ;
1594 dev->caps.num_comp_vectors = MSIX_LEGACY_SZ - 1;
1596 for (i = 0; i < nreq; ++i)
1597 priv->eq_table.eq[i].irq = entries[i].vector;
1599 dev->flags |= MLX4_FLAG_MSI_X;
1606 dev->caps.num_comp_vectors = 1;
1607 dev->caps.comp_pool = 0;
1609 for (i = 0; i < 2; ++i)
1610 priv->eq_table.eq[i].irq = dev->pdev->irq;
1613 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
1615 struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
1620 if (!mlx4_is_slave(dev)) {
1621 INIT_RADIX_TREE(&info->mac_tree, GFP_KERNEL);
1622 mlx4_init_mac_table(dev, &info->mac_table);
1623 mlx4_init_vlan_table(dev, &info->vlan_table);
1625 dev->caps.reserved_qps_base[MLX4_QP_REGION_ETH_ADDR] +
1626 (port - 1) * (1 << log_num_mac);
1629 sprintf(info->dev_name, "mlx4_port%d", port);
1630 info->port_attr.attr.name = info->dev_name;
1631 if (mlx4_is_mfunc(dev))
1632 info->port_attr.attr.mode = S_IRUGO;
1634 info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
1635 info->port_attr.store = set_port_type;
1637 info->port_attr.show = show_port_type;
1638 sysfs_attr_init(&info->port_attr.attr);
1640 err = device_create_file(&dev->pdev->dev, &info->port_attr);
1642 mlx4_err(dev, "Failed to create file for port %d\n", port);
1646 sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
1647 info->port_mtu_attr.attr.name = info->dev_mtu_name;
1648 if (mlx4_is_mfunc(dev))
1649 info->port_mtu_attr.attr.mode = S_IRUGO;
1651 info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
1652 info->port_mtu_attr.store = set_port_ib_mtu;
1654 info->port_mtu_attr.show = show_port_ib_mtu;
1655 sysfs_attr_init(&info->port_mtu_attr.attr);
1657 err = device_create_file(&dev->pdev->dev, &info->port_mtu_attr);
1659 mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
1660 device_remove_file(&info->dev->pdev->dev, &info->port_attr);
1667 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
1672 device_remove_file(&info->dev->pdev->dev, &info->port_attr);
1673 device_remove_file(&info->dev->pdev->dev, &info->port_mtu_attr);
1676 static int mlx4_init_steering(struct mlx4_dev *dev)
1678 struct mlx4_priv *priv = mlx4_priv(dev);
1679 int num_entries = dev->caps.num_ports;
1682 priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
1686 for (i = 0; i < num_entries; i++)
1687 for (j = 0; j < MLX4_NUM_STEERS; j++) {
1688 INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
1689 INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
1694 static void mlx4_clear_steering(struct mlx4_dev *dev)
1696 struct mlx4_priv *priv = mlx4_priv(dev);
1697 struct mlx4_steer_index *entry, *tmp_entry;
1698 struct mlx4_promisc_qp *pqp, *tmp_pqp;
1699 int num_entries = dev->caps.num_ports;
1702 for (i = 0; i < num_entries; i++) {
1703 for (j = 0; j < MLX4_NUM_STEERS; j++) {
1704 list_for_each_entry_safe(pqp, tmp_pqp,
1705 &priv->steer[i].promisc_qps[j],
1707 list_del(&pqp->list);
1710 list_for_each_entry_safe(entry, tmp_entry,
1711 &priv->steer[i].steer_entries[j],
1713 list_del(&entry->list);
1714 list_for_each_entry_safe(pqp, tmp_pqp,
1717 list_del(&pqp->list);
1727 static int extended_func_num(struct pci_dev *pdev)
1729 return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
1732 #define MLX4_OWNER_BASE 0x8069c
1733 #define MLX4_OWNER_SIZE 4
1735 static int mlx4_get_ownership(struct mlx4_dev *dev)
1737 void __iomem *owner;
1740 owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
1743 mlx4_err(dev, "Failed to obtain ownership bit\n");
1752 static void mlx4_free_ownership(struct mlx4_dev *dev)
1754 void __iomem *owner;
1756 owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
1759 mlx4_err(dev, "Failed to obtain ownership bit\n");
1767 static int __mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
1769 struct mlx4_priv *priv;
1770 struct mlx4_dev *dev;
1774 pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
1776 err = pci_enable_device(pdev);
1778 dev_err(&pdev->dev, "Cannot enable PCI device, "
1782 if (num_vfs > MLX4_MAX_NUM_VF) {
1783 printk(KERN_ERR "There are more VF's (%d) than allowed(%d)\n",
1784 num_vfs, MLX4_MAX_NUM_VF);
1790 if (((id == NULL) || !(id->driver_data & MLX4_VF)) &&
1791 !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
1792 dev_err(&pdev->dev, "Missing DCS, aborting."
1793 "(id == 0X%p, id->driver_data: 0x%lx,"
1794 " pci_resource_flags(pdev, 0):0x%lx)\n", id,
1795 id ? id->driver_data : 0, pci_resource_flags(pdev, 0));
1797 goto err_disable_pdev;
1799 if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
1800 dev_err(&pdev->dev, "Missing UAR, aborting.\n");
1802 goto err_disable_pdev;
1805 err = pci_request_regions(pdev, DRV_NAME);
1807 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
1808 goto err_disable_pdev;
1811 pci_set_master(pdev);
1813 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1815 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
1816 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1818 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
1819 goto err_release_regions;
1822 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1824 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
1825 "consistent PCI DMA mask.\n");
1826 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1828 dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
1830 goto err_release_regions;
1834 /* Allow large DMA segments, up to the firmware limit of 1 GB */
1835 dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
1837 priv = kzalloc(sizeof *priv, GFP_KERNEL);
1839 dev_err(&pdev->dev, "Device struct alloc failed, "
1842 goto err_release_regions;
1847 INIT_LIST_HEAD(&priv->ctx_list);
1848 spin_lock_init(&priv->ctx_lock);
1850 mutex_init(&priv->port_mutex);
1852 INIT_LIST_HEAD(&priv->pgdir_list);
1853 mutex_init(&priv->pgdir_mutex);
1855 INIT_LIST_HEAD(&priv->bf_list);
1856 mutex_init(&priv->bf_mutex);
1858 dev->rev_id = pdev->revision;
1859 /* Detect if this device is a virtual function */
1860 if (id && id->driver_data & MLX4_VF) {
1861 /* When acting as pf, we normally skip vfs unless explicitly
1862 * requested to probe them. */
1863 if (num_vfs && extended_func_num(pdev) > probe_vf) {
1864 mlx4_warn(dev, "Skipping virtual function:%d\n",
1865 extended_func_num(pdev));
1869 mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
1870 dev->flags |= MLX4_FLAG_SLAVE;
1872 /* We reset the device and enable SRIOV only for physical
1873 * devices. Try to claim ownership on the device;
1874 * if already taken, skip -- do not allow multiple PFs */
1875 err = mlx4_get_ownership(dev);
1880 mlx4_warn(dev, "Multiple PFs not yet supported."
1888 mlx4_warn(dev, "Enabling sriov with:%d vfs\n", num_vfs);
1889 err = pci_enable_sriov(pdev, num_vfs);
1891 mlx4_err(dev, "Failed to enable sriov,"
1892 "continuing without sriov enabled"
1893 " (err = %d).\n", err);
1896 mlx4_warn(dev, "Running in master mode\n");
1897 dev->flags |= MLX4_FLAG_SRIOV |
1899 dev->num_vfs = num_vfs;
1904 * Now reset the HCA before we touch the PCI capabilities or
1905 * attempt a firmware command, since a boot ROM may have left
1906 * the HCA in an undefined state.
1908 err = mlx4_reset(dev);
1910 mlx4_err(dev, "Failed to reset HCA, aborting.\n");
1916 if (mlx4_cmd_init(dev)) {
1917 mlx4_err(dev, "Failed to init command interface, aborting.\n");
1921 /* In slave functions, the communication channel must be initialized
1922 * before posting commands. Also, init num_slaves before calling
1924 if (mlx4_is_mfunc(dev)) {
1925 if (mlx4_is_master(dev))
1926 dev->num_slaves = MLX4_MAX_NUM_SLAVES;
1928 dev->num_slaves = 0;
1929 if (mlx4_multi_func_init(dev)) {
1930 mlx4_err(dev, "Failed to init slave mfunc"
1931 " interface, aborting.\n");
1937 err = mlx4_init_hca(dev);
1939 if (err == -EACCES) {
1940 /* Not primary Physical function
1941 * Running in slave mode */
1942 mlx4_cmd_cleanup(dev);
1943 dev->flags |= MLX4_FLAG_SLAVE;
1944 dev->flags &= ~MLX4_FLAG_MASTER;
1950 /* In master functions, the communication channel must be initialized
1951 * after obtaining its address from fw */
1952 if (mlx4_is_master(dev)) {
1953 if (mlx4_multi_func_init(dev)) {
1954 mlx4_err(dev, "Failed to init master mfunc"
1955 "interface, aborting.\n");
1960 err = mlx4_alloc_eq_table(dev);
1962 goto err_master_mfunc;
1964 priv->msix_ctl.pool_bm = 0;
1965 mutex_init(&priv->msix_ctl.pool_lock);
1967 mlx4_enable_msi_x(dev);
1968 if ((mlx4_is_mfunc(dev)) &&
1969 !(dev->flags & MLX4_FLAG_MSI_X)) {
1970 mlx4_err(dev, "INTx is not supported in multi-function mode."
1975 if (!mlx4_is_slave(dev)) {
1976 err = mlx4_init_steering(dev);
1981 err = mlx4_setup_hca(dev);
1982 if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
1983 !mlx4_is_mfunc(dev)) {
1984 dev->flags &= ~MLX4_FLAG_MSI_X;
1985 pci_disable_msix(pdev);
1986 err = mlx4_setup_hca(dev);
1992 for (port = 1; port <= dev->caps.num_ports; port++) {
1993 err = mlx4_init_port_info(dev, port);
1998 err = mlx4_register_device(dev);
2002 mlx4_sense_init(dev);
2003 mlx4_start_sense(dev);
2005 pci_set_drvdata(pdev, dev);
2010 for (--port; port >= 1; --port)
2011 mlx4_cleanup_port_info(&priv->port[port]);
2013 mlx4_cleanup_counters_table(dev);
2014 mlx4_cleanup_mcg_table(dev);
2015 mlx4_cleanup_qp_table(dev);
2016 mlx4_cleanup_srq_table(dev);
2017 mlx4_cleanup_cq_table(dev);
2018 mlx4_cmd_use_polling(dev);
2019 mlx4_cleanup_eq_table(dev);
2020 mlx4_cleanup_mr_table(dev);
2021 mlx4_cleanup_xrcd_table(dev);
2022 mlx4_cleanup_pd_table(dev);
2023 mlx4_cleanup_uar_table(dev);
2026 if (!mlx4_is_slave(dev))
2027 mlx4_clear_steering(dev);
2030 mlx4_free_eq_table(dev);
2033 if (mlx4_is_master(dev))
2034 mlx4_multi_func_cleanup(dev);
2037 if (dev->flags & MLX4_FLAG_MSI_X)
2038 pci_disable_msix(pdev);
2040 mlx4_close_hca(dev);
2043 if (mlx4_is_slave(dev))
2044 mlx4_multi_func_cleanup(dev);
2047 mlx4_cmd_cleanup(dev);
2050 if (dev->flags & MLX4_FLAG_SRIOV)
2051 pci_disable_sriov(pdev);
2054 if (!mlx4_is_slave(dev))
2055 mlx4_free_ownership(dev);
2060 err_release_regions:
2061 pci_release_regions(pdev);
2064 pci_disable_device(pdev);
2065 pci_set_drvdata(pdev, NULL);
2069 static int __devinit mlx4_init_one(struct pci_dev *pdev,
2070 const struct pci_device_id *id)
2072 printk_once(KERN_INFO "%s", mlx4_version);
2074 return __mlx4_init_one(pdev, id);
2077 static void mlx4_remove_one(struct pci_dev *pdev)
2079 struct mlx4_dev *dev = pci_get_drvdata(pdev);
2080 struct mlx4_priv *priv = mlx4_priv(dev);
2084 /* in SRIOV it is not allowed to unload the pf's
2085 * driver while there are alive vf's */
2086 if (mlx4_is_master(dev)) {
2087 if (mlx4_how_many_lives_vf(dev))
2088 printk(KERN_ERR "Removing PF when there are assigned VF's !!!\n");
2090 mlx4_stop_sense(dev);
2091 mlx4_unregister_device(dev);
2093 for (p = 1; p <= dev->caps.num_ports; p++) {
2094 mlx4_cleanup_port_info(&priv->port[p]);
2095 mlx4_CLOSE_PORT(dev, p);
2098 if (mlx4_is_master(dev))
2099 mlx4_free_resource_tracker(dev,
2100 RES_TR_FREE_SLAVES_ONLY);
2102 mlx4_cleanup_counters_table(dev);
2103 mlx4_cleanup_mcg_table(dev);
2104 mlx4_cleanup_qp_table(dev);
2105 mlx4_cleanup_srq_table(dev);
2106 mlx4_cleanup_cq_table(dev);
2107 mlx4_cmd_use_polling(dev);
2108 mlx4_cleanup_eq_table(dev);
2109 mlx4_cleanup_mr_table(dev);
2110 mlx4_cleanup_xrcd_table(dev);
2111 mlx4_cleanup_pd_table(dev);
2113 if (mlx4_is_master(dev))
2114 mlx4_free_resource_tracker(dev,
2115 RES_TR_FREE_STRUCTS_ONLY);
2118 mlx4_uar_free(dev, &priv->driver_uar);
2119 mlx4_cleanup_uar_table(dev);
2120 if (!mlx4_is_slave(dev))
2121 mlx4_clear_steering(dev);
2122 mlx4_free_eq_table(dev);
2123 if (mlx4_is_master(dev))
2124 mlx4_multi_func_cleanup(dev);
2125 mlx4_close_hca(dev);
2126 if (mlx4_is_slave(dev))
2127 mlx4_multi_func_cleanup(dev);
2128 mlx4_cmd_cleanup(dev);
2130 if (dev->flags & MLX4_FLAG_MSI_X)
2131 pci_disable_msix(pdev);
2132 if (dev->flags & MLX4_FLAG_SRIOV) {
2133 mlx4_warn(dev, "Disabling sriov\n");
2134 pci_disable_sriov(pdev);
2137 if (!mlx4_is_slave(dev))
2138 mlx4_free_ownership(dev);
2140 pci_release_regions(pdev);
2141 pci_disable_device(pdev);
2142 pci_set_drvdata(pdev, NULL);
2146 int mlx4_restart_one(struct pci_dev *pdev)
2148 mlx4_remove_one(pdev);
2149 return __mlx4_init_one(pdev, NULL);
2152 static DEFINE_PCI_DEVICE_TABLE(mlx4_pci_table) = {
2153 /* MT25408 "Hermon" SDR */
2154 { PCI_VDEVICE(MELLANOX, 0x6340), 0 },
2155 /* MT25408 "Hermon" DDR */
2156 { PCI_VDEVICE(MELLANOX, 0x634a), 0 },
2157 /* MT25408 "Hermon" QDR */
2158 { PCI_VDEVICE(MELLANOX, 0x6354), 0 },
2159 /* MT25408 "Hermon" DDR PCIe gen2 */
2160 { PCI_VDEVICE(MELLANOX, 0x6732), 0 },
2161 /* MT25408 "Hermon" QDR PCIe gen2 */
2162 { PCI_VDEVICE(MELLANOX, 0x673c), 0 },
2163 /* MT25408 "Hermon" EN 10GigE */
2164 { PCI_VDEVICE(MELLANOX, 0x6368), 0 },
2165 /* MT25408 "Hermon" EN 10GigE PCIe gen2 */
2166 { PCI_VDEVICE(MELLANOX, 0x6750), 0 },
2167 /* MT25458 ConnectX EN 10GBASE-T 10GigE */
2168 { PCI_VDEVICE(MELLANOX, 0x6372), 0 },
2169 /* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
2170 { PCI_VDEVICE(MELLANOX, 0x675a), 0 },
2171 /* MT26468 ConnectX EN 10GigE PCIe gen2*/
2172 { PCI_VDEVICE(MELLANOX, 0x6764), 0 },
2173 /* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
2174 { PCI_VDEVICE(MELLANOX, 0x6746), 0 },
2175 /* MT26478 ConnectX2 40GigE PCIe gen2 */
2176 { PCI_VDEVICE(MELLANOX, 0x676e), 0 },
2177 /* MT25400 Family [ConnectX-2 Virtual Function] */
2178 { PCI_VDEVICE(MELLANOX, 0x1002), MLX4_VF },
2179 /* MT27500 Family [ConnectX-3] */
2180 { PCI_VDEVICE(MELLANOX, 0x1003), 0 },
2181 /* MT27500 Family [ConnectX-3 Virtual Function] */
2182 { PCI_VDEVICE(MELLANOX, 0x1004), MLX4_VF },
2183 { PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
2184 { PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
2185 { PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
2186 { PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
2187 { PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
2188 { PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
2189 { PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
2190 { PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
2191 { PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
2192 { PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
2193 { PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
2194 { PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
2198 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
2200 static struct pci_driver mlx4_driver = {
2202 .id_table = mlx4_pci_table,
2203 .probe = mlx4_init_one,
2204 .remove = __devexit_p(mlx4_remove_one)
2207 static int __init mlx4_verify_params(void)
2209 if ((log_num_mac < 0) || (log_num_mac > 7)) {
2210 pr_warning("mlx4_core: bad num_mac: %d\n", log_num_mac);
2214 if (log_num_vlan != 0)
2215 pr_warning("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
2216 MLX4_LOG_NUM_VLANS);
2218 if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
2219 pr_warning("mlx4_core: bad log_mtts_per_seg: %d\n", log_mtts_per_seg);
2223 /* Check if module param for ports type has legal combination */
2224 if (port_type_array[0] == false && port_type_array[1] == true) {
2225 printk(KERN_WARNING "Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
2226 port_type_array[0] = true;
2232 static int __init mlx4_init(void)
2236 if (mlx4_verify_params())
2241 mlx4_wq = create_singlethread_workqueue("mlx4");
2245 ret = pci_register_driver(&mlx4_driver);
2246 return ret < 0 ? ret : 0;
2249 static void __exit mlx4_cleanup(void)
2251 pci_unregister_driver(&mlx4_driver);
2252 destroy_workqueue(mlx4_wq);
2255 module_init(mlx4_init);
2256 module_exit(mlx4_cleanup);