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 #include <linux/delay.h>
44 #include <linux/netdevice.h>
45 
46 #include <linux/mlx4/device.h>
47 #include <linux/mlx4/doorbell.h>
48 
49 #include "mlx4.h"
50 #include "fw.h"
51 #include "icm.h"
52 
53 MODULE_AUTHOR("Roland Dreier");
54 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
55 MODULE_LICENSE("Dual BSD/GPL");
56 MODULE_VERSION(DRV_VERSION);
57 
58 struct workqueue_struct *mlx4_wq;
59 
60 #ifdef CONFIG_MLX4_DEBUG
61 
62 int mlx4_debug_level = 0;
63 module_param_named(debug_level, mlx4_debug_level, int, 0644);
64 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
65 
66 #endif /* CONFIG_MLX4_DEBUG */
67 
68 #ifdef CONFIG_PCI_MSI
69 
70 static int msi_x = 1;
71 module_param(msi_x, int, 0444);
72 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
73 
74 #else /* CONFIG_PCI_MSI */
75 
76 #define msi_x (0)
77 
78 #endif /* CONFIG_PCI_MSI */
79 
80 static int num_vfs;
81 module_param(num_vfs, int, 0444);
82 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0");
83 
84 static int probe_vf;
85 module_param(probe_vf, int, 0644);
86 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)");
87 
88 int mlx4_log_num_mgm_entry_size = 10;
89 module_param_named(log_num_mgm_entry_size,
90 			mlx4_log_num_mgm_entry_size, int, 0444);
91 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
92 					 " of qp per mcg, for example:"
93 					 " 10 gives 248.range: 9<="
94 					 " log_num_mgm_entry_size <= 12."
95 					 " Not in use with device managed"
96 					 " flow steering");
97 
98 #define HCA_GLOBAL_CAP_MASK            0
99 #define PF_CONTEXT_BEHAVIOUR_MASK      0
100 
101 static char mlx4_version[] __devinitdata =
102 	DRV_NAME ": Mellanox ConnectX core driver v"
103 	DRV_VERSION " (" DRV_RELDATE ")\n";
104 
105 static struct mlx4_profile default_profile = {
106 	.num_qp		= 1 << 18,
107 	.num_srq	= 1 << 16,
108 	.rdmarc_per_qp	= 1 << 4,
109 	.num_cq		= 1 << 16,
110 	.num_mcg	= 1 << 13,
111 	.num_mpt	= 1 << 19,
112 	.num_mtt	= 1 << 20, /* It is really num mtt segements */
113 };
114 
115 static int log_num_mac = 7;
116 module_param_named(log_num_mac, log_num_mac, int, 0444);
117 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
118 
119 static int log_num_vlan;
120 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
121 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
122 /* Log2 max number of VLANs per ETH port (0-7) */
123 #define MLX4_LOG_NUM_VLANS 7
124 
125 static bool use_prio;
126 module_param_named(use_prio, use_prio, bool, 0444);
127 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports "
128 		  "(0/1, default 0)");
129 
130 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
131 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
132 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
133 
134 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
135 static int arr_argc = 2;
136 module_param_array(port_type_array, int, &arr_argc, 0444);
137 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
138 				"1 for IB, 2 for Ethernet");
139 
140 struct mlx4_port_config {
141 	struct list_head list;
142 	enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
143 	struct pci_dev *pdev;
144 };
145 
146 int mlx4_check_port_params(struct mlx4_dev *dev,
147 			   enum mlx4_port_type *port_type)
148 {
149 	int i;
150 
151 	for (i = 0; i < dev->caps.num_ports - 1; i++) {
152 		if (port_type[i] != port_type[i + 1]) {
153 			if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
154 				mlx4_err(dev, "Only same port types supported "
155 					 "on this HCA, aborting.\n");
156 				return -EINVAL;
157 			}
158 		}
159 	}
160 
161 	for (i = 0; i < dev->caps.num_ports; i++) {
162 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
163 			mlx4_err(dev, "Requested port type for port %d is not "
164 				      "supported on this HCA\n", i + 1);
165 			return -EINVAL;
166 		}
167 	}
168 	return 0;
169 }
170 
171 static void mlx4_set_port_mask(struct mlx4_dev *dev)
172 {
173 	int i;
174 
175 	for (i = 1; i <= dev->caps.num_ports; ++i)
176 		dev->caps.port_mask[i] = dev->caps.port_type[i];
177 }
178 
179 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
180 {
181 	int err;
182 	int i;
183 
184 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
185 	if (err) {
186 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
187 		return err;
188 	}
189 
190 	if (dev_cap->min_page_sz > PAGE_SIZE) {
191 		mlx4_err(dev, "HCA minimum page size of %d bigger than "
192 			 "kernel PAGE_SIZE of %ld, aborting.\n",
193 			 dev_cap->min_page_sz, PAGE_SIZE);
194 		return -ENODEV;
195 	}
196 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
197 		mlx4_err(dev, "HCA has %d ports, but we only support %d, "
198 			 "aborting.\n",
199 			 dev_cap->num_ports, MLX4_MAX_PORTS);
200 		return -ENODEV;
201 	}
202 
203 	if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
204 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than "
205 			 "PCI resource 2 size of 0x%llx, aborting.\n",
206 			 dev_cap->uar_size,
207 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
208 		return -ENODEV;
209 	}
210 
211 	dev->caps.num_ports	     = dev_cap->num_ports;
212 	dev->phys_caps.num_phys_eqs  = MLX4_MAX_EQ_NUM;
213 	for (i = 1; i <= dev->caps.num_ports; ++i) {
214 		dev->caps.vl_cap[i]	    = dev_cap->max_vl[i];
215 		dev->caps.ib_mtu_cap[i]	    = dev_cap->ib_mtu[i];
216 		dev->phys_caps.gid_phys_table_len[i]  = dev_cap->max_gids[i];
217 		dev->phys_caps.pkey_phys_table_len[i] = dev_cap->max_pkeys[i];
218 		/* set gid and pkey table operating lengths by default
219 		 * to non-sriov values */
220 		dev->caps.gid_table_len[i]  = dev_cap->max_gids[i];
221 		dev->caps.pkey_table_len[i] = dev_cap->max_pkeys[i];
222 		dev->caps.port_width_cap[i] = dev_cap->max_port_width[i];
223 		dev->caps.eth_mtu_cap[i]    = dev_cap->eth_mtu[i];
224 		dev->caps.def_mac[i]        = dev_cap->def_mac[i];
225 		dev->caps.supported_type[i] = dev_cap->supported_port_types[i];
226 		dev->caps.suggested_type[i] = dev_cap->suggested_type[i];
227 		dev->caps.default_sense[i] = dev_cap->default_sense[i];
228 		dev->caps.trans_type[i]	    = dev_cap->trans_type[i];
229 		dev->caps.vendor_oui[i]     = dev_cap->vendor_oui[i];
230 		dev->caps.wavelength[i]     = dev_cap->wavelength[i];
231 		dev->caps.trans_code[i]     = dev_cap->trans_code[i];
232 	}
233 
234 	dev->caps.uar_page_size	     = PAGE_SIZE;
235 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
236 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
237 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
238 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
239 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
240 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
241 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
242 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
243 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
244 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
245 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
246 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
247 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
248 	/*
249 	 * Subtract 1 from the limit because we need to allocate a
250 	 * spare CQE so the HCA HW can tell the difference between an
251 	 * empty CQ and a full CQ.
252 	 */
253 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
254 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
255 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
256 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
257 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
258 
259 	/* The first 128 UARs are used for EQ doorbells */
260 	dev->caps.reserved_uars	     = max_t(int, 128, dev_cap->reserved_uars);
261 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
262 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
263 					dev_cap->reserved_xrcds : 0;
264 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
265 					dev_cap->max_xrcds : 0;
266 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
267 
268 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
269 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
270 	dev->caps.flags		     = dev_cap->flags;
271 	dev->caps.flags2	     = dev_cap->flags2;
272 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
273 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
274 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
275 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
276 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
277 
278 	if (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
279 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
280 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
281 		dev->caps.fs_log_max_ucast_qp_range_size =
282 			dev_cap->fs_log_max_ucast_qp_range_size;
283 	} else {
284 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
285 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER) {
286 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
287 		} else {
288 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
289 
290 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
291 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
292 				mlx4_warn(dev, "Must have UC_STEER and MC_STEER flags "
293 						"set to use B0 steering. Falling back to A0 steering mode.\n");
294 		}
295 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
296 	}
297 	mlx4_dbg(dev, "Steering mode is: %s\n",
298 		 mlx4_steering_mode_str(dev->caps.steering_mode));
299 
300 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
301 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
302 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
303 	/* Don't do sense port on multifunction devices (for now at least) */
304 	if (mlx4_is_mfunc(dev))
305 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
306 
307 	dev->caps.log_num_macs  = log_num_mac;
308 	dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
309 	dev->caps.log_num_prios = use_prio ? 3 : 0;
310 
311 	for (i = 1; i <= dev->caps.num_ports; ++i) {
312 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
313 		if (dev->caps.supported_type[i]) {
314 			/* if only ETH is supported - assign ETH */
315 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
316 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
317 			/* if only IB is supported, assign IB */
318 			else if (dev->caps.supported_type[i] ==
319 				 MLX4_PORT_TYPE_IB)
320 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
321 			else {
322 				/* if IB and ETH are supported, we set the port
323 				 * type according to user selection of port type;
324 				 * if user selected none, take the FW hint */
325 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
326 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
327 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
328 				else
329 					dev->caps.port_type[i] = port_type_array[i - 1];
330 			}
331 		}
332 		/*
333 		 * Link sensing is allowed on the port if 3 conditions are true:
334 		 * 1. Both protocols are supported on the port.
335 		 * 2. Different types are supported on the port
336 		 * 3. FW declared that it supports link sensing
337 		 */
338 		mlx4_priv(dev)->sense.sense_allowed[i] =
339 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
340 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
341 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
342 
343 		/*
344 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
345 		 * and perform sense_port FW command to try and set the correct
346 		 * port type from beginning
347 		 */
348 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
349 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
350 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
351 			mlx4_SENSE_PORT(dev, i, &sensed_port);
352 			if (sensed_port != MLX4_PORT_TYPE_NONE)
353 				dev->caps.port_type[i] = sensed_port;
354 		} else {
355 			dev->caps.possible_type[i] = dev->caps.port_type[i];
356 		}
357 
358 		if (dev->caps.log_num_macs > dev_cap->log_max_macs[i]) {
359 			dev->caps.log_num_macs = dev_cap->log_max_macs[i];
360 			mlx4_warn(dev, "Requested number of MACs is too much "
361 				  "for port %d, reducing to %d.\n",
362 				  i, 1 << dev->caps.log_num_macs);
363 		}
364 		if (dev->caps.log_num_vlans > dev_cap->log_max_vlans[i]) {
365 			dev->caps.log_num_vlans = dev_cap->log_max_vlans[i];
366 			mlx4_warn(dev, "Requested number of VLANs is too much "
367 				  "for port %d, reducing to %d.\n",
368 				  i, 1 << dev->caps.log_num_vlans);
369 		}
370 	}
371 
372 	dev->caps.max_counters = 1 << ilog2(dev_cap->max_counters);
373 
374 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
375 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
376 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
377 		(1 << dev->caps.log_num_macs) *
378 		(1 << dev->caps.log_num_vlans) *
379 		(1 << dev->caps.log_num_prios) *
380 		dev->caps.num_ports;
381 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
382 
383 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
384 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
385 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
386 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
387 
388 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
389 	return 0;
390 }
391 /*The function checks if there are live vf, return the num of them*/
392 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
393 {
394 	struct mlx4_priv *priv = mlx4_priv(dev);
395 	struct mlx4_slave_state *s_state;
396 	int i;
397 	int ret = 0;
398 
399 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
400 		s_state = &priv->mfunc.master.slave_state[i];
401 		if (s_state->active && s_state->last_cmd !=
402 		    MLX4_COMM_CMD_RESET) {
403 			mlx4_warn(dev, "%s: slave: %d is still active\n",
404 				  __func__, i);
405 			ret++;
406 		}
407 	}
408 	return ret;
409 }
410 
411 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
412 {
413 	u32 qk = MLX4_RESERVED_QKEY_BASE;
414 
415 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
416 	    qpn < dev->phys_caps.base_proxy_sqpn)
417 		return -EINVAL;
418 
419 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
420 		/* tunnel qp */
421 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
422 	else
423 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
424 	*qkey = qk;
425 	return 0;
426 }
427 EXPORT_SYMBOL(mlx4_get_parav_qkey);
428 
429 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
430 {
431 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
432 
433 	if (!mlx4_is_master(dev))
434 		return;
435 
436 	priv->virt2phys_pkey[slave][port - 1][i] = val;
437 }
438 EXPORT_SYMBOL(mlx4_sync_pkey_table);
439 
440 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
441 {
442 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
443 
444 	if (!mlx4_is_master(dev))
445 		return;
446 
447 	priv->slave_node_guids[slave] = guid;
448 }
449 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
450 
451 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
452 {
453 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
454 
455 	if (!mlx4_is_master(dev))
456 		return 0;
457 
458 	return priv->slave_node_guids[slave];
459 }
460 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
461 
462 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
463 {
464 	struct mlx4_priv *priv = mlx4_priv(dev);
465 	struct mlx4_slave_state *s_slave;
466 
467 	if (!mlx4_is_master(dev))
468 		return 0;
469 
470 	s_slave = &priv->mfunc.master.slave_state[slave];
471 	return !!s_slave->active;
472 }
473 EXPORT_SYMBOL(mlx4_is_slave_active);
474 
475 static int mlx4_slave_cap(struct mlx4_dev *dev)
476 {
477 	int			   err;
478 	u32			   page_size;
479 	struct mlx4_dev_cap	   dev_cap;
480 	struct mlx4_func_cap	   func_cap;
481 	struct mlx4_init_hca_param hca_param;
482 	int			   i;
483 
484 	memset(&hca_param, 0, sizeof(hca_param));
485 	err = mlx4_QUERY_HCA(dev, &hca_param);
486 	if (err) {
487 		mlx4_err(dev, "QUERY_HCA command failed, aborting.\n");
488 		return err;
489 	}
490 
491 	/*fail if the hca has an unknown capability */
492 	if ((hca_param.global_caps | HCA_GLOBAL_CAP_MASK) !=
493 	    HCA_GLOBAL_CAP_MASK) {
494 		mlx4_err(dev, "Unknown hca global capabilities\n");
495 		return -ENOSYS;
496 	}
497 
498 	mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
499 
500 	memset(&dev_cap, 0, sizeof(dev_cap));
501 	dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
502 	err = mlx4_dev_cap(dev, &dev_cap);
503 	if (err) {
504 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
505 		return err;
506 	}
507 
508 	err = mlx4_QUERY_FW(dev);
509 	if (err)
510 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version.\n");
511 
512 	page_size = ~dev->caps.page_size_cap + 1;
513 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
514 	if (page_size > PAGE_SIZE) {
515 		mlx4_err(dev, "HCA minimum page size of %d bigger than "
516 			 "kernel PAGE_SIZE of %ld, aborting.\n",
517 			 page_size, PAGE_SIZE);
518 		return -ENODEV;
519 	}
520 
521 	/* slave gets uar page size from QUERY_HCA fw command */
522 	dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
523 
524 	/* TODO: relax this assumption */
525 	if (dev->caps.uar_page_size != PAGE_SIZE) {
526 		mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
527 			 dev->caps.uar_page_size, PAGE_SIZE);
528 		return -ENODEV;
529 	}
530 
531 	memset(&func_cap, 0, sizeof(func_cap));
532 	err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
533 	if (err) {
534 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d).\n",
535 			  err);
536 		return err;
537 	}
538 
539 	if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
540 	    PF_CONTEXT_BEHAVIOUR_MASK) {
541 		mlx4_err(dev, "Unknown pf context behaviour\n");
542 		return -ENOSYS;
543 	}
544 
545 	dev->caps.num_ports		= func_cap.num_ports;
546 	dev->caps.num_qps		= func_cap.qp_quota;
547 	dev->caps.num_srqs		= func_cap.srq_quota;
548 	dev->caps.num_cqs		= func_cap.cq_quota;
549 	dev->caps.num_eqs               = func_cap.max_eq;
550 	dev->caps.reserved_eqs          = func_cap.reserved_eq;
551 	dev->caps.num_mpts		= func_cap.mpt_quota;
552 	dev->caps.num_mtts		= func_cap.mtt_quota;
553 	dev->caps.num_pds               = MLX4_NUM_PDS;
554 	dev->caps.num_mgms              = 0;
555 	dev->caps.num_amgms             = 0;
556 
557 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
558 		mlx4_err(dev, "HCA has %d ports, but we only support %d, "
559 			 "aborting.\n", dev->caps.num_ports, MLX4_MAX_PORTS);
560 		return -ENODEV;
561 	}
562 
563 	dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
564 	dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
565 	dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
566 	dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
567 
568 	if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
569 	    !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy) {
570 		err = -ENOMEM;
571 		goto err_mem;
572 	}
573 
574 	for (i = 1; i <= dev->caps.num_ports; ++i) {
575 		err = mlx4_QUERY_FUNC_CAP(dev, (u32) i, &func_cap);
576 		if (err) {
577 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for"
578 				 " port %d, aborting (%d).\n", i, err);
579 			goto err_mem;
580 		}
581 		dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
582 		dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
583 		dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
584 		dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
585 		dev->caps.port_mask[i] = dev->caps.port_type[i];
586 		if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
587 						    &dev->caps.gid_table_len[i],
588 						    &dev->caps.pkey_table_len[i]))
589 			goto err_mem;
590 	}
591 
592 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
593 				       dev->caps.reserved_uars) >
594 				       pci_resource_len(dev->pdev, 2)) {
595 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than "
596 			 "PCI resource 2 size of 0x%llx, aborting.\n",
597 			 dev->caps.uar_page_size * dev->caps.num_uars,
598 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
599 		goto err_mem;
600 	}
601 
602 	return 0;
603 
604 err_mem:
605 	kfree(dev->caps.qp0_tunnel);
606 	kfree(dev->caps.qp0_proxy);
607 	kfree(dev->caps.qp1_tunnel);
608 	kfree(dev->caps.qp1_proxy);
609 	dev->caps.qp0_tunnel = dev->caps.qp0_proxy =
610 		dev->caps.qp1_tunnel = dev->caps.qp1_proxy = NULL;
611 
612 	return err;
613 }
614 
615 /*
616  * Change the port configuration of the device.
617  * Every user of this function must hold the port mutex.
618  */
619 int mlx4_change_port_types(struct mlx4_dev *dev,
620 			   enum mlx4_port_type *port_types)
621 {
622 	int err = 0;
623 	int change = 0;
624 	int port;
625 
626 	for (port = 0; port <  dev->caps.num_ports; port++) {
627 		/* Change the port type only if the new type is different
628 		 * from the current, and not set to Auto */
629 		if (port_types[port] != dev->caps.port_type[port + 1])
630 			change = 1;
631 	}
632 	if (change) {
633 		mlx4_unregister_device(dev);
634 		for (port = 1; port <= dev->caps.num_ports; port++) {
635 			mlx4_CLOSE_PORT(dev, port);
636 			dev->caps.port_type[port] = port_types[port - 1];
637 			err = mlx4_SET_PORT(dev, port, -1);
638 			if (err) {
639 				mlx4_err(dev, "Failed to set port %d, "
640 					      "aborting\n", port);
641 				goto out;
642 			}
643 		}
644 		mlx4_set_port_mask(dev);
645 		err = mlx4_register_device(dev);
646 	}
647 
648 out:
649 	return err;
650 }
651 
652 static ssize_t show_port_type(struct device *dev,
653 			      struct device_attribute *attr,
654 			      char *buf)
655 {
656 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
657 						   port_attr);
658 	struct mlx4_dev *mdev = info->dev;
659 	char type[8];
660 
661 	sprintf(type, "%s",
662 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
663 		"ib" : "eth");
664 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
665 		sprintf(buf, "auto (%s)\n", type);
666 	else
667 		sprintf(buf, "%s\n", type);
668 
669 	return strlen(buf);
670 }
671 
672 static ssize_t set_port_type(struct device *dev,
673 			     struct device_attribute *attr,
674 			     const char *buf, size_t count)
675 {
676 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
677 						   port_attr);
678 	struct mlx4_dev *mdev = info->dev;
679 	struct mlx4_priv *priv = mlx4_priv(mdev);
680 	enum mlx4_port_type types[MLX4_MAX_PORTS];
681 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
682 	int i;
683 	int err = 0;
684 
685 	if (!strcmp(buf, "ib\n"))
686 		info->tmp_type = MLX4_PORT_TYPE_IB;
687 	else if (!strcmp(buf, "eth\n"))
688 		info->tmp_type = MLX4_PORT_TYPE_ETH;
689 	else if (!strcmp(buf, "auto\n"))
690 		info->tmp_type = MLX4_PORT_TYPE_AUTO;
691 	else {
692 		mlx4_err(mdev, "%s is not supported port type\n", buf);
693 		return -EINVAL;
694 	}
695 
696 	mlx4_stop_sense(mdev);
697 	mutex_lock(&priv->port_mutex);
698 	/* Possible type is always the one that was delivered */
699 	mdev->caps.possible_type[info->port] = info->tmp_type;
700 
701 	for (i = 0; i < mdev->caps.num_ports; i++) {
702 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
703 					mdev->caps.possible_type[i+1];
704 		if (types[i] == MLX4_PORT_TYPE_AUTO)
705 			types[i] = mdev->caps.port_type[i+1];
706 	}
707 
708 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
709 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
710 		for (i = 1; i <= mdev->caps.num_ports; i++) {
711 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
712 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
713 				err = -EINVAL;
714 			}
715 		}
716 	}
717 	if (err) {
718 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. "
719 			       "Set only 'eth' or 'ib' for both ports "
720 			       "(should be the same)\n");
721 		goto out;
722 	}
723 
724 	mlx4_do_sense_ports(mdev, new_types, types);
725 
726 	err = mlx4_check_port_params(mdev, new_types);
727 	if (err)
728 		goto out;
729 
730 	/* We are about to apply the changes after the configuration
731 	 * was verified, no need to remember the temporary types
732 	 * any more */
733 	for (i = 0; i < mdev->caps.num_ports; i++)
734 		priv->port[i + 1].tmp_type = 0;
735 
736 	err = mlx4_change_port_types(mdev, new_types);
737 
738 out:
739 	mlx4_start_sense(mdev);
740 	mutex_unlock(&priv->port_mutex);
741 	return err ? err : count;
742 }
743 
744 enum ibta_mtu {
745 	IB_MTU_256  = 1,
746 	IB_MTU_512  = 2,
747 	IB_MTU_1024 = 3,
748 	IB_MTU_2048 = 4,
749 	IB_MTU_4096 = 5
750 };
751 
752 static inline int int_to_ibta_mtu(int mtu)
753 {
754 	switch (mtu) {
755 	case 256:  return IB_MTU_256;
756 	case 512:  return IB_MTU_512;
757 	case 1024: return IB_MTU_1024;
758 	case 2048: return IB_MTU_2048;
759 	case 4096: return IB_MTU_4096;
760 	default: return -1;
761 	}
762 }
763 
764 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
765 {
766 	switch (mtu) {
767 	case IB_MTU_256:  return  256;
768 	case IB_MTU_512:  return  512;
769 	case IB_MTU_1024: return 1024;
770 	case IB_MTU_2048: return 2048;
771 	case IB_MTU_4096: return 4096;
772 	default: return -1;
773 	}
774 }
775 
776 static ssize_t show_port_ib_mtu(struct device *dev,
777 			     struct device_attribute *attr,
778 			     char *buf)
779 {
780 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
781 						   port_mtu_attr);
782 	struct mlx4_dev *mdev = info->dev;
783 
784 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
785 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
786 
787 	sprintf(buf, "%d\n",
788 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
789 	return strlen(buf);
790 }
791 
792 static ssize_t set_port_ib_mtu(struct device *dev,
793 			     struct device_attribute *attr,
794 			     const char *buf, size_t count)
795 {
796 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
797 						   port_mtu_attr);
798 	struct mlx4_dev *mdev = info->dev;
799 	struct mlx4_priv *priv = mlx4_priv(mdev);
800 	int err, port, mtu, ibta_mtu = -1;
801 
802 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
803 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
804 		return -EINVAL;
805 	}
806 
807 	err = sscanf(buf, "%d", &mtu);
808 	if (err > 0)
809 		ibta_mtu = int_to_ibta_mtu(mtu);
810 
811 	if (err <= 0 || ibta_mtu < 0) {
812 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
813 		return -EINVAL;
814 	}
815 
816 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
817 
818 	mlx4_stop_sense(mdev);
819 	mutex_lock(&priv->port_mutex);
820 	mlx4_unregister_device(mdev);
821 	for (port = 1; port <= mdev->caps.num_ports; port++) {
822 		mlx4_CLOSE_PORT(mdev, port);
823 		err = mlx4_SET_PORT(mdev, port, -1);
824 		if (err) {
825 			mlx4_err(mdev, "Failed to set port %d, "
826 				      "aborting\n", port);
827 			goto err_set_port;
828 		}
829 	}
830 	err = mlx4_register_device(mdev);
831 err_set_port:
832 	mutex_unlock(&priv->port_mutex);
833 	mlx4_start_sense(mdev);
834 	return err ? err : count;
835 }
836 
837 static int mlx4_load_fw(struct mlx4_dev *dev)
838 {
839 	struct mlx4_priv *priv = mlx4_priv(dev);
840 	int err;
841 
842 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
843 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
844 	if (!priv->fw.fw_icm) {
845 		mlx4_err(dev, "Couldn't allocate FW area, aborting.\n");
846 		return -ENOMEM;
847 	}
848 
849 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
850 	if (err) {
851 		mlx4_err(dev, "MAP_FA command failed, aborting.\n");
852 		goto err_free;
853 	}
854 
855 	err = mlx4_RUN_FW(dev);
856 	if (err) {
857 		mlx4_err(dev, "RUN_FW command failed, aborting.\n");
858 		goto err_unmap_fa;
859 	}
860 
861 	return 0;
862 
863 err_unmap_fa:
864 	mlx4_UNMAP_FA(dev);
865 
866 err_free:
867 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
868 	return err;
869 }
870 
871 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
872 				int cmpt_entry_sz)
873 {
874 	struct mlx4_priv *priv = mlx4_priv(dev);
875 	int err;
876 	int num_eqs;
877 
878 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
879 				  cmpt_base +
880 				  ((u64) (MLX4_CMPT_TYPE_QP *
881 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
882 				  cmpt_entry_sz, dev->caps.num_qps,
883 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
884 				  0, 0);
885 	if (err)
886 		goto err;
887 
888 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
889 				  cmpt_base +
890 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
891 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
892 				  cmpt_entry_sz, dev->caps.num_srqs,
893 				  dev->caps.reserved_srqs, 0, 0);
894 	if (err)
895 		goto err_qp;
896 
897 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
898 				  cmpt_base +
899 				  ((u64) (MLX4_CMPT_TYPE_CQ *
900 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
901 				  cmpt_entry_sz, dev->caps.num_cqs,
902 				  dev->caps.reserved_cqs, 0, 0);
903 	if (err)
904 		goto err_srq;
905 
906 	num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
907 		  dev->caps.num_eqs;
908 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
909 				  cmpt_base +
910 				  ((u64) (MLX4_CMPT_TYPE_EQ *
911 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
912 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
913 	if (err)
914 		goto err_cq;
915 
916 	return 0;
917 
918 err_cq:
919 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
920 
921 err_srq:
922 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
923 
924 err_qp:
925 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
926 
927 err:
928 	return err;
929 }
930 
931 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
932 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
933 {
934 	struct mlx4_priv *priv = mlx4_priv(dev);
935 	u64 aux_pages;
936 	int num_eqs;
937 	int err;
938 
939 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
940 	if (err) {
941 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting.\n");
942 		return err;
943 	}
944 
945 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory.\n",
946 		 (unsigned long long) icm_size >> 10,
947 		 (unsigned long long) aux_pages << 2);
948 
949 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
950 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
951 	if (!priv->fw.aux_icm) {
952 		mlx4_err(dev, "Couldn't allocate aux memory, aborting.\n");
953 		return -ENOMEM;
954 	}
955 
956 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
957 	if (err) {
958 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting.\n");
959 		goto err_free_aux;
960 	}
961 
962 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
963 	if (err) {
964 		mlx4_err(dev, "Failed to map cMPT context memory, aborting.\n");
965 		goto err_unmap_aux;
966 	}
967 
968 
969 	num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
970 		   dev->caps.num_eqs;
971 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
972 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
973 				  num_eqs, num_eqs, 0, 0);
974 	if (err) {
975 		mlx4_err(dev, "Failed to map EQ context memory, aborting.\n");
976 		goto err_unmap_cmpt;
977 	}
978 
979 	/*
980 	 * Reserved MTT entries must be aligned up to a cacheline
981 	 * boundary, since the FW will write to them, while the driver
982 	 * writes to all other MTT entries. (The variable
983 	 * dev->caps.mtt_entry_sz below is really the MTT segment
984 	 * size, not the raw entry size)
985 	 */
986 	dev->caps.reserved_mtts =
987 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
988 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
989 
990 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
991 				  init_hca->mtt_base,
992 				  dev->caps.mtt_entry_sz,
993 				  dev->caps.num_mtts,
994 				  dev->caps.reserved_mtts, 1, 0);
995 	if (err) {
996 		mlx4_err(dev, "Failed to map MTT context memory, aborting.\n");
997 		goto err_unmap_eq;
998 	}
999 
1000 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1001 				  init_hca->dmpt_base,
1002 				  dev_cap->dmpt_entry_sz,
1003 				  dev->caps.num_mpts,
1004 				  dev->caps.reserved_mrws, 1, 1);
1005 	if (err) {
1006 		mlx4_err(dev, "Failed to map dMPT context memory, aborting.\n");
1007 		goto err_unmap_mtt;
1008 	}
1009 
1010 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1011 				  init_hca->qpc_base,
1012 				  dev_cap->qpc_entry_sz,
1013 				  dev->caps.num_qps,
1014 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1015 				  0, 0);
1016 	if (err) {
1017 		mlx4_err(dev, "Failed to map QP context memory, aborting.\n");
1018 		goto err_unmap_dmpt;
1019 	}
1020 
1021 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1022 				  init_hca->auxc_base,
1023 				  dev_cap->aux_entry_sz,
1024 				  dev->caps.num_qps,
1025 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1026 				  0, 0);
1027 	if (err) {
1028 		mlx4_err(dev, "Failed to map AUXC context memory, aborting.\n");
1029 		goto err_unmap_qp;
1030 	}
1031 
1032 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1033 				  init_hca->altc_base,
1034 				  dev_cap->altc_entry_sz,
1035 				  dev->caps.num_qps,
1036 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1037 				  0, 0);
1038 	if (err) {
1039 		mlx4_err(dev, "Failed to map ALTC context memory, aborting.\n");
1040 		goto err_unmap_auxc;
1041 	}
1042 
1043 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1044 				  init_hca->rdmarc_base,
1045 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1046 				  dev->caps.num_qps,
1047 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1048 				  0, 0);
1049 	if (err) {
1050 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1051 		goto err_unmap_altc;
1052 	}
1053 
1054 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1055 				  init_hca->cqc_base,
1056 				  dev_cap->cqc_entry_sz,
1057 				  dev->caps.num_cqs,
1058 				  dev->caps.reserved_cqs, 0, 0);
1059 	if (err) {
1060 		mlx4_err(dev, "Failed to map CQ context memory, aborting.\n");
1061 		goto err_unmap_rdmarc;
1062 	}
1063 
1064 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1065 				  init_hca->srqc_base,
1066 				  dev_cap->srq_entry_sz,
1067 				  dev->caps.num_srqs,
1068 				  dev->caps.reserved_srqs, 0, 0);
1069 	if (err) {
1070 		mlx4_err(dev, "Failed to map SRQ context memory, aborting.\n");
1071 		goto err_unmap_cq;
1072 	}
1073 
1074 	/*
1075 	 * For flow steering device managed mode it is required to use
1076 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1077 	 * required, but for simplicity just map the whole multicast
1078 	 * group table now.  The table isn't very big and it's a lot
1079 	 * easier than trying to track ref counts.
1080 	 */
1081 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1082 				  init_hca->mc_base,
1083 				  mlx4_get_mgm_entry_size(dev),
1084 				  dev->caps.num_mgms + dev->caps.num_amgms,
1085 				  dev->caps.num_mgms + dev->caps.num_amgms,
1086 				  0, 0);
1087 	if (err) {
1088 		mlx4_err(dev, "Failed to map MCG context memory, aborting.\n");
1089 		goto err_unmap_srq;
1090 	}
1091 
1092 	return 0;
1093 
1094 err_unmap_srq:
1095 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1096 
1097 err_unmap_cq:
1098 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1099 
1100 err_unmap_rdmarc:
1101 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1102 
1103 err_unmap_altc:
1104 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1105 
1106 err_unmap_auxc:
1107 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1108 
1109 err_unmap_qp:
1110 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1111 
1112 err_unmap_dmpt:
1113 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1114 
1115 err_unmap_mtt:
1116 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1117 
1118 err_unmap_eq:
1119 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1120 
1121 err_unmap_cmpt:
1122 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1123 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1124 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1125 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1126 
1127 err_unmap_aux:
1128 	mlx4_UNMAP_ICM_AUX(dev);
1129 
1130 err_free_aux:
1131 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1132 
1133 	return err;
1134 }
1135 
1136 static void mlx4_free_icms(struct mlx4_dev *dev)
1137 {
1138 	struct mlx4_priv *priv = mlx4_priv(dev);
1139 
1140 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1141 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1142 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1143 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1144 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1145 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1146 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1147 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1148 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1149 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1150 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1151 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1152 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1153 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1154 
1155 	mlx4_UNMAP_ICM_AUX(dev);
1156 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1157 }
1158 
1159 static void mlx4_slave_exit(struct mlx4_dev *dev)
1160 {
1161 	struct mlx4_priv *priv = mlx4_priv(dev);
1162 
1163 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1164 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_TIME))
1165 		mlx4_warn(dev, "Failed to close slave function.\n");
1166 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1167 }
1168 
1169 static int map_bf_area(struct mlx4_dev *dev)
1170 {
1171 	struct mlx4_priv *priv = mlx4_priv(dev);
1172 	resource_size_t bf_start;
1173 	resource_size_t bf_len;
1174 	int err = 0;
1175 
1176 	if (!dev->caps.bf_reg_size)
1177 		return -ENXIO;
1178 
1179 	bf_start = pci_resource_start(dev->pdev, 2) +
1180 			(dev->caps.num_uars << PAGE_SHIFT);
1181 	bf_len = pci_resource_len(dev->pdev, 2) -
1182 			(dev->caps.num_uars << PAGE_SHIFT);
1183 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1184 	if (!priv->bf_mapping)
1185 		err = -ENOMEM;
1186 
1187 	return err;
1188 }
1189 
1190 static void unmap_bf_area(struct mlx4_dev *dev)
1191 {
1192 	if (mlx4_priv(dev)->bf_mapping)
1193 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1194 }
1195 
1196 static void mlx4_close_hca(struct mlx4_dev *dev)
1197 {
1198 	unmap_bf_area(dev);
1199 	if (mlx4_is_slave(dev))
1200 		mlx4_slave_exit(dev);
1201 	else {
1202 		mlx4_CLOSE_HCA(dev, 0);
1203 		mlx4_free_icms(dev);
1204 		mlx4_UNMAP_FA(dev);
1205 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1206 	}
1207 }
1208 
1209 static int mlx4_init_slave(struct mlx4_dev *dev)
1210 {
1211 	struct mlx4_priv *priv = mlx4_priv(dev);
1212 	u64 dma = (u64) priv->mfunc.vhcr_dma;
1213 	int num_of_reset_retries = NUM_OF_RESET_RETRIES;
1214 	int ret_from_reset = 0;
1215 	u32 slave_read;
1216 	u32 cmd_channel_ver;
1217 
1218 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1219 	priv->cmd.max_cmds = 1;
1220 	mlx4_warn(dev, "Sending reset\n");
1221 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1222 				       MLX4_COMM_TIME);
1223 	/* if we are in the middle of flr the slave will try
1224 	 * NUM_OF_RESET_RETRIES times before leaving.*/
1225 	if (ret_from_reset) {
1226 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1227 			msleep(SLEEP_TIME_IN_RESET);
1228 			while (ret_from_reset && num_of_reset_retries) {
1229 				mlx4_warn(dev, "slave is currently in the"
1230 					  "middle of FLR. retrying..."
1231 					  "(try num:%d)\n",
1232 					  (NUM_OF_RESET_RETRIES -
1233 					   num_of_reset_retries  + 1));
1234 				ret_from_reset =
1235 					mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET,
1236 						      0, MLX4_COMM_TIME);
1237 				num_of_reset_retries = num_of_reset_retries - 1;
1238 			}
1239 		} else
1240 			goto err;
1241 	}
1242 
1243 	/* check the driver version - the slave I/F revision
1244 	 * must match the master's */
1245 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1246 	cmd_channel_ver = mlx4_comm_get_version();
1247 
1248 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1249 		MLX4_COMM_GET_IF_REV(slave_read)) {
1250 		mlx4_err(dev, "slave driver version is not supported"
1251 			 " by the master\n");
1252 		goto err;
1253 	}
1254 
1255 	mlx4_warn(dev, "Sending vhcr0\n");
1256 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1257 						    MLX4_COMM_TIME))
1258 		goto err;
1259 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1260 						    MLX4_COMM_TIME))
1261 		goto err;
1262 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1263 						    MLX4_COMM_TIME))
1264 		goto err;
1265 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma, MLX4_COMM_TIME))
1266 		goto err;
1267 
1268 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1269 	return 0;
1270 
1271 err:
1272 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, 0);
1273 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1274 	return -EIO;
1275 }
1276 
1277 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1278 {
1279 	int i;
1280 
1281 	for (i = 1; i <= dev->caps.num_ports; i++) {
1282 		dev->caps.gid_table_len[i] = 1;
1283 		dev->caps.pkey_table_len[i] =
1284 			dev->phys_caps.pkey_phys_table_len[i] - 1;
1285 	}
1286 }
1287 
1288 static int mlx4_init_hca(struct mlx4_dev *dev)
1289 {
1290 	struct mlx4_priv	  *priv = mlx4_priv(dev);
1291 	struct mlx4_adapter	   adapter;
1292 	struct mlx4_dev_cap	   dev_cap;
1293 	struct mlx4_mod_stat_cfg   mlx4_cfg;
1294 	struct mlx4_profile	   profile;
1295 	struct mlx4_init_hca_param init_hca;
1296 	u64 icm_size;
1297 	int err;
1298 
1299 	if (!mlx4_is_slave(dev)) {
1300 		err = mlx4_QUERY_FW(dev);
1301 		if (err) {
1302 			if (err == -EACCES)
1303 				mlx4_info(dev, "non-primary physical function, skipping.\n");
1304 			else
1305 				mlx4_err(dev, "QUERY_FW command failed, aborting.\n");
1306 			return err;
1307 		}
1308 
1309 		err = mlx4_load_fw(dev);
1310 		if (err) {
1311 			mlx4_err(dev, "Failed to start FW, aborting.\n");
1312 			return err;
1313 		}
1314 
1315 		mlx4_cfg.log_pg_sz_m = 1;
1316 		mlx4_cfg.log_pg_sz = 0;
1317 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
1318 		if (err)
1319 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
1320 
1321 		err = mlx4_dev_cap(dev, &dev_cap);
1322 		if (err) {
1323 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
1324 			goto err_stop_fw;
1325 		}
1326 
1327 		if (mlx4_is_master(dev))
1328 			mlx4_parav_master_pf_caps(dev);
1329 
1330 		priv->fs_hash_mode = MLX4_FS_L2_HASH;
1331 
1332 		switch (priv->fs_hash_mode) {
1333 		case MLX4_FS_L2_HASH:
1334 			init_hca.fs_hash_enable_bits = 0;
1335 			break;
1336 
1337 		case MLX4_FS_L2_L3_L4_HASH:
1338 			/* Enable flow steering with
1339 			 * udp unicast and tcp unicast
1340 			 */
1341 			init_hca.fs_hash_enable_bits =
1342 				MLX4_FS_UDP_UC_EN | MLX4_FS_TCP_UC_EN;
1343 			break;
1344 		}
1345 
1346 		profile = default_profile;
1347 		if (dev->caps.steering_mode ==
1348 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
1349 			profile.num_mcg = MLX4_FS_NUM_MCG;
1350 
1351 		icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
1352 					     &init_hca);
1353 		if ((long long) icm_size < 0) {
1354 			err = icm_size;
1355 			goto err_stop_fw;
1356 		}
1357 
1358 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
1359 
1360 		init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
1361 		init_hca.uar_page_sz = PAGE_SHIFT - 12;
1362 
1363 		err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
1364 		if (err)
1365 			goto err_stop_fw;
1366 
1367 		err = mlx4_INIT_HCA(dev, &init_hca);
1368 		if (err) {
1369 			mlx4_err(dev, "INIT_HCA command failed, aborting.\n");
1370 			goto err_free_icm;
1371 		}
1372 	} else {
1373 		err = mlx4_init_slave(dev);
1374 		if (err) {
1375 			mlx4_err(dev, "Failed to initialize slave\n");
1376 			return err;
1377 		}
1378 
1379 		err = mlx4_slave_cap(dev);
1380 		if (err) {
1381 			mlx4_err(dev, "Failed to obtain slave caps\n");
1382 			goto err_close;
1383 		}
1384 	}
1385 
1386 	if (map_bf_area(dev))
1387 		mlx4_dbg(dev, "Failed to map blue flame area\n");
1388 
1389 	/*Only the master set the ports, all the rest got it from it.*/
1390 	if (!mlx4_is_slave(dev))
1391 		mlx4_set_port_mask(dev);
1392 
1393 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
1394 	if (err) {
1395 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting.\n");
1396 		goto unmap_bf;
1397 	}
1398 
1399 	priv->eq_table.inta_pin = adapter.inta_pin;
1400 	memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
1401 
1402 	return 0;
1403 
1404 unmap_bf:
1405 	unmap_bf_area(dev);
1406 
1407 err_close:
1408 	if (mlx4_is_slave(dev))
1409 		mlx4_slave_exit(dev);
1410 	else
1411 		mlx4_CLOSE_HCA(dev, 0);
1412 
1413 err_free_icm:
1414 	if (!mlx4_is_slave(dev))
1415 		mlx4_free_icms(dev);
1416 
1417 err_stop_fw:
1418 	if (!mlx4_is_slave(dev)) {
1419 		mlx4_UNMAP_FA(dev);
1420 		mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1421 	}
1422 	return err;
1423 }
1424 
1425 static int mlx4_init_counters_table(struct mlx4_dev *dev)
1426 {
1427 	struct mlx4_priv *priv = mlx4_priv(dev);
1428 	int nent;
1429 
1430 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1431 		return -ENOENT;
1432 
1433 	nent = dev->caps.max_counters;
1434 	return mlx4_bitmap_init(&priv->counters_bitmap, nent, nent - 1, 0, 0);
1435 }
1436 
1437 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
1438 {
1439 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
1440 }
1441 
1442 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1443 {
1444 	struct mlx4_priv *priv = mlx4_priv(dev);
1445 
1446 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1447 		return -ENOENT;
1448 
1449 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
1450 	if (*idx == -1)
1451 		return -ENOMEM;
1452 
1453 	return 0;
1454 }
1455 
1456 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1457 {
1458 	u64 out_param;
1459 	int err;
1460 
1461 	if (mlx4_is_mfunc(dev)) {
1462 		err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
1463 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
1464 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
1465 		if (!err)
1466 			*idx = get_param_l(&out_param);
1467 
1468 		return err;
1469 	}
1470 	return __mlx4_counter_alloc(dev, idx);
1471 }
1472 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
1473 
1474 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1475 {
1476 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx);
1477 	return;
1478 }
1479 
1480 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1481 {
1482 	u64 in_param;
1483 
1484 	if (mlx4_is_mfunc(dev)) {
1485 		set_param_l(&in_param, idx);
1486 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
1487 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
1488 			 MLX4_CMD_WRAPPED);
1489 		return;
1490 	}
1491 	__mlx4_counter_free(dev, idx);
1492 }
1493 EXPORT_SYMBOL_GPL(mlx4_counter_free);
1494 
1495 static int mlx4_setup_hca(struct mlx4_dev *dev)
1496 {
1497 	struct mlx4_priv *priv = mlx4_priv(dev);
1498 	int err;
1499 	int port;
1500 	__be32 ib_port_default_caps;
1501 
1502 	err = mlx4_init_uar_table(dev);
1503 	if (err) {
1504 		mlx4_err(dev, "Failed to initialize "
1505 			 "user access region table, aborting.\n");
1506 		return err;
1507 	}
1508 
1509 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
1510 	if (err) {
1511 		mlx4_err(dev, "Failed to allocate driver access region, "
1512 			 "aborting.\n");
1513 		goto err_uar_table_free;
1514 	}
1515 
1516 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
1517 	if (!priv->kar) {
1518 		mlx4_err(dev, "Couldn't map kernel access region, "
1519 			 "aborting.\n");
1520 		err = -ENOMEM;
1521 		goto err_uar_free;
1522 	}
1523 
1524 	err = mlx4_init_pd_table(dev);
1525 	if (err) {
1526 		mlx4_err(dev, "Failed to initialize "
1527 			 "protection domain table, aborting.\n");
1528 		goto err_kar_unmap;
1529 	}
1530 
1531 	err = mlx4_init_xrcd_table(dev);
1532 	if (err) {
1533 		mlx4_err(dev, "Failed to initialize "
1534 			 "reliable connection domain table, aborting.\n");
1535 		goto err_pd_table_free;
1536 	}
1537 
1538 	err = mlx4_init_mr_table(dev);
1539 	if (err) {
1540 		mlx4_err(dev, "Failed to initialize "
1541 			 "memory region table, aborting.\n");
1542 		goto err_xrcd_table_free;
1543 	}
1544 
1545 	err = mlx4_init_eq_table(dev);
1546 	if (err) {
1547 		mlx4_err(dev, "Failed to initialize "
1548 			 "event queue table, aborting.\n");
1549 		goto err_mr_table_free;
1550 	}
1551 
1552 	err = mlx4_cmd_use_events(dev);
1553 	if (err) {
1554 		mlx4_err(dev, "Failed to switch to event-driven "
1555 			 "firmware commands, aborting.\n");
1556 		goto err_eq_table_free;
1557 	}
1558 
1559 	err = mlx4_NOP(dev);
1560 	if (err) {
1561 		if (dev->flags & MLX4_FLAG_MSI_X) {
1562 			mlx4_warn(dev, "NOP command failed to generate MSI-X "
1563 				  "interrupt IRQ %d).\n",
1564 				  priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1565 			mlx4_warn(dev, "Trying again without MSI-X.\n");
1566 		} else {
1567 			mlx4_err(dev, "NOP command failed to generate interrupt "
1568 				 "(IRQ %d), aborting.\n",
1569 				 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1570 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
1571 		}
1572 
1573 		goto err_cmd_poll;
1574 	}
1575 
1576 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
1577 
1578 	err = mlx4_init_cq_table(dev);
1579 	if (err) {
1580 		mlx4_err(dev, "Failed to initialize "
1581 			 "completion queue table, aborting.\n");
1582 		goto err_cmd_poll;
1583 	}
1584 
1585 	err = mlx4_init_srq_table(dev);
1586 	if (err) {
1587 		mlx4_err(dev, "Failed to initialize "
1588 			 "shared receive queue table, aborting.\n");
1589 		goto err_cq_table_free;
1590 	}
1591 
1592 	err = mlx4_init_qp_table(dev);
1593 	if (err) {
1594 		mlx4_err(dev, "Failed to initialize "
1595 			 "queue pair table, aborting.\n");
1596 		goto err_srq_table_free;
1597 	}
1598 
1599 	if (!mlx4_is_slave(dev)) {
1600 		err = mlx4_init_mcg_table(dev);
1601 		if (err) {
1602 			mlx4_err(dev, "Failed to initialize "
1603 				 "multicast group table, aborting.\n");
1604 			goto err_qp_table_free;
1605 		}
1606 	}
1607 
1608 	err = mlx4_init_counters_table(dev);
1609 	if (err && err != -ENOENT) {
1610 		mlx4_err(dev, "Failed to initialize counters table, aborting.\n");
1611 		goto err_mcg_table_free;
1612 	}
1613 
1614 	if (!mlx4_is_slave(dev)) {
1615 		for (port = 1; port <= dev->caps.num_ports; port++) {
1616 			ib_port_default_caps = 0;
1617 			err = mlx4_get_port_ib_caps(dev, port,
1618 						    &ib_port_default_caps);
1619 			if (err)
1620 				mlx4_warn(dev, "failed to get port %d default "
1621 					  "ib capabilities (%d). Continuing "
1622 					  "with caps = 0\n", port, err);
1623 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
1624 
1625 			/* initialize per-slave default ib port capabilities */
1626 			if (mlx4_is_master(dev)) {
1627 				int i;
1628 				for (i = 0; i < dev->num_slaves; i++) {
1629 					if (i == mlx4_master_func_num(dev))
1630 						continue;
1631 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
1632 							ib_port_default_caps;
1633 				}
1634 			}
1635 
1636 			if (mlx4_is_mfunc(dev))
1637 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
1638 			else
1639 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
1640 
1641 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
1642 					    dev->caps.pkey_table_len[port] : -1);
1643 			if (err) {
1644 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1645 					port);
1646 				goto err_counters_table_free;
1647 			}
1648 		}
1649 	}
1650 
1651 	return 0;
1652 
1653 err_counters_table_free:
1654 	mlx4_cleanup_counters_table(dev);
1655 
1656 err_mcg_table_free:
1657 	mlx4_cleanup_mcg_table(dev);
1658 
1659 err_qp_table_free:
1660 	mlx4_cleanup_qp_table(dev);
1661 
1662 err_srq_table_free:
1663 	mlx4_cleanup_srq_table(dev);
1664 
1665 err_cq_table_free:
1666 	mlx4_cleanup_cq_table(dev);
1667 
1668 err_cmd_poll:
1669 	mlx4_cmd_use_polling(dev);
1670 
1671 err_eq_table_free:
1672 	mlx4_cleanup_eq_table(dev);
1673 
1674 err_mr_table_free:
1675 	mlx4_cleanup_mr_table(dev);
1676 
1677 err_xrcd_table_free:
1678 	mlx4_cleanup_xrcd_table(dev);
1679 
1680 err_pd_table_free:
1681 	mlx4_cleanup_pd_table(dev);
1682 
1683 err_kar_unmap:
1684 	iounmap(priv->kar);
1685 
1686 err_uar_free:
1687 	mlx4_uar_free(dev, &priv->driver_uar);
1688 
1689 err_uar_table_free:
1690 	mlx4_cleanup_uar_table(dev);
1691 	return err;
1692 }
1693 
1694 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
1695 {
1696 	struct mlx4_priv *priv = mlx4_priv(dev);
1697 	struct msix_entry *entries;
1698 	int nreq = min_t(int, dev->caps.num_ports *
1699 			 min_t(int, netif_get_num_default_rss_queues() + 1,
1700 			       MAX_MSIX_P_PORT) + MSIX_LEGACY_SZ, MAX_MSIX);
1701 	int err;
1702 	int i;
1703 
1704 	if (msi_x) {
1705 		/* In multifunction mode each function gets 2 msi-X vectors
1706 		 * one for data path completions anf the other for asynch events
1707 		 * or command completions */
1708 		if (mlx4_is_mfunc(dev)) {
1709 			nreq = 2;
1710 		} else {
1711 			nreq = min_t(int, dev->caps.num_eqs -
1712 				     dev->caps.reserved_eqs, nreq);
1713 		}
1714 
1715 		entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
1716 		if (!entries)
1717 			goto no_msi;
1718 
1719 		for (i = 0; i < nreq; ++i)
1720 			entries[i].entry = i;
1721 
1722 	retry:
1723 		err = pci_enable_msix(dev->pdev, entries, nreq);
1724 		if (err) {
1725 			/* Try again if at least 2 vectors are available */
1726 			if (err > 1) {
1727 				mlx4_info(dev, "Requested %d vectors, "
1728 					  "but only %d MSI-X vectors available, "
1729 					  "trying again\n", nreq, err);
1730 				nreq = err;
1731 				goto retry;
1732 			}
1733 			kfree(entries);
1734 			goto no_msi;
1735 		}
1736 
1737 		if (nreq <
1738 		    MSIX_LEGACY_SZ + dev->caps.num_ports * MIN_MSIX_P_PORT) {
1739 			/*Working in legacy mode , all EQ's shared*/
1740 			dev->caps.comp_pool           = 0;
1741 			dev->caps.num_comp_vectors = nreq - 1;
1742 		} else {
1743 			dev->caps.comp_pool           = nreq - MSIX_LEGACY_SZ;
1744 			dev->caps.num_comp_vectors = MSIX_LEGACY_SZ - 1;
1745 		}
1746 		for (i = 0; i < nreq; ++i)
1747 			priv->eq_table.eq[i].irq = entries[i].vector;
1748 
1749 		dev->flags |= MLX4_FLAG_MSI_X;
1750 
1751 		kfree(entries);
1752 		return;
1753 	}
1754 
1755 no_msi:
1756 	dev->caps.num_comp_vectors = 1;
1757 	dev->caps.comp_pool	   = 0;
1758 
1759 	for (i = 0; i < 2; ++i)
1760 		priv->eq_table.eq[i].irq = dev->pdev->irq;
1761 }
1762 
1763 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
1764 {
1765 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
1766 	int err = 0;
1767 
1768 	info->dev = dev;
1769 	info->port = port;
1770 	if (!mlx4_is_slave(dev)) {
1771 		INIT_RADIX_TREE(&info->mac_tree, GFP_KERNEL);
1772 		mlx4_init_mac_table(dev, &info->mac_table);
1773 		mlx4_init_vlan_table(dev, &info->vlan_table);
1774 		info->base_qpn =
1775 			dev->caps.reserved_qps_base[MLX4_QP_REGION_ETH_ADDR] +
1776 			(port - 1) * (1 << log_num_mac);
1777 	}
1778 
1779 	sprintf(info->dev_name, "mlx4_port%d", port);
1780 	info->port_attr.attr.name = info->dev_name;
1781 	if (mlx4_is_mfunc(dev))
1782 		info->port_attr.attr.mode = S_IRUGO;
1783 	else {
1784 		info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
1785 		info->port_attr.store     = set_port_type;
1786 	}
1787 	info->port_attr.show      = show_port_type;
1788 	sysfs_attr_init(&info->port_attr.attr);
1789 
1790 	err = device_create_file(&dev->pdev->dev, &info->port_attr);
1791 	if (err) {
1792 		mlx4_err(dev, "Failed to create file for port %d\n", port);
1793 		info->port = -1;
1794 	}
1795 
1796 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
1797 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
1798 	if (mlx4_is_mfunc(dev))
1799 		info->port_mtu_attr.attr.mode = S_IRUGO;
1800 	else {
1801 		info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
1802 		info->port_mtu_attr.store     = set_port_ib_mtu;
1803 	}
1804 	info->port_mtu_attr.show      = show_port_ib_mtu;
1805 	sysfs_attr_init(&info->port_mtu_attr.attr);
1806 
1807 	err = device_create_file(&dev->pdev->dev, &info->port_mtu_attr);
1808 	if (err) {
1809 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
1810 		device_remove_file(&info->dev->pdev->dev, &info->port_attr);
1811 		info->port = -1;
1812 	}
1813 
1814 	return err;
1815 }
1816 
1817 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
1818 {
1819 	if (info->port < 0)
1820 		return;
1821 
1822 	device_remove_file(&info->dev->pdev->dev, &info->port_attr);
1823 	device_remove_file(&info->dev->pdev->dev, &info->port_mtu_attr);
1824 }
1825 
1826 static int mlx4_init_steering(struct mlx4_dev *dev)
1827 {
1828 	struct mlx4_priv *priv = mlx4_priv(dev);
1829 	int num_entries = dev->caps.num_ports;
1830 	int i, j;
1831 
1832 	priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
1833 	if (!priv->steer)
1834 		return -ENOMEM;
1835 
1836 	for (i = 0; i < num_entries; i++)
1837 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
1838 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
1839 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
1840 		}
1841 	return 0;
1842 }
1843 
1844 static void mlx4_clear_steering(struct mlx4_dev *dev)
1845 {
1846 	struct mlx4_priv *priv = mlx4_priv(dev);
1847 	struct mlx4_steer_index *entry, *tmp_entry;
1848 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
1849 	int num_entries = dev->caps.num_ports;
1850 	int i, j;
1851 
1852 	for (i = 0; i < num_entries; i++) {
1853 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
1854 			list_for_each_entry_safe(pqp, tmp_pqp,
1855 						 &priv->steer[i].promisc_qps[j],
1856 						 list) {
1857 				list_del(&pqp->list);
1858 				kfree(pqp);
1859 			}
1860 			list_for_each_entry_safe(entry, tmp_entry,
1861 						 &priv->steer[i].steer_entries[j],
1862 						 list) {
1863 				list_del(&entry->list);
1864 				list_for_each_entry_safe(pqp, tmp_pqp,
1865 							 &entry->duplicates,
1866 							 list) {
1867 					list_del(&pqp->list);
1868 					kfree(pqp);
1869 				}
1870 				kfree(entry);
1871 			}
1872 		}
1873 	}
1874 	kfree(priv->steer);
1875 }
1876 
1877 static int extended_func_num(struct pci_dev *pdev)
1878 {
1879 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
1880 }
1881 
1882 #define MLX4_OWNER_BASE	0x8069c
1883 #define MLX4_OWNER_SIZE	4
1884 
1885 static int mlx4_get_ownership(struct mlx4_dev *dev)
1886 {
1887 	void __iomem *owner;
1888 	u32 ret;
1889 
1890 	if (pci_channel_offline(dev->pdev))
1891 		return -EIO;
1892 
1893 	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
1894 			MLX4_OWNER_SIZE);
1895 	if (!owner) {
1896 		mlx4_err(dev, "Failed to obtain ownership bit\n");
1897 		return -ENOMEM;
1898 	}
1899 
1900 	ret = readl(owner);
1901 	iounmap(owner);
1902 	return (int) !!ret;
1903 }
1904 
1905 static void mlx4_free_ownership(struct mlx4_dev *dev)
1906 {
1907 	void __iomem *owner;
1908 
1909 	if (pci_channel_offline(dev->pdev))
1910 		return;
1911 
1912 	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
1913 			MLX4_OWNER_SIZE);
1914 	if (!owner) {
1915 		mlx4_err(dev, "Failed to obtain ownership bit\n");
1916 		return;
1917 	}
1918 	writel(0, owner);
1919 	msleep(1000);
1920 	iounmap(owner);
1921 }
1922 
1923 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data)
1924 {
1925 	struct mlx4_priv *priv;
1926 	struct mlx4_dev *dev;
1927 	int err;
1928 	int port;
1929 
1930 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
1931 
1932 	err = pci_enable_device(pdev);
1933 	if (err) {
1934 		dev_err(&pdev->dev, "Cannot enable PCI device, "
1935 			"aborting.\n");
1936 		return err;
1937 	}
1938 	if (num_vfs > MLX4_MAX_NUM_VF) {
1939 		printk(KERN_ERR "There are more VF's (%d) than allowed(%d)\n",
1940 		       num_vfs, MLX4_MAX_NUM_VF);
1941 		return -EINVAL;
1942 	}
1943 	/*
1944 	 * Check for BARs.
1945 	 */
1946 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
1947 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
1948 		dev_err(&pdev->dev, "Missing DCS, aborting."
1949 			"(driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
1950 			pci_dev_data, pci_resource_flags(pdev, 0));
1951 		err = -ENODEV;
1952 		goto err_disable_pdev;
1953 	}
1954 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
1955 		dev_err(&pdev->dev, "Missing UAR, aborting.\n");
1956 		err = -ENODEV;
1957 		goto err_disable_pdev;
1958 	}
1959 
1960 	err = pci_request_regions(pdev, DRV_NAME);
1961 	if (err) {
1962 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
1963 		goto err_disable_pdev;
1964 	}
1965 
1966 	pci_set_master(pdev);
1967 
1968 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1969 	if (err) {
1970 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
1971 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1972 		if (err) {
1973 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
1974 			goto err_release_regions;
1975 		}
1976 	}
1977 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1978 	if (err) {
1979 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
1980 			 "consistent PCI DMA mask.\n");
1981 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1982 		if (err) {
1983 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
1984 				"aborting.\n");
1985 			goto err_release_regions;
1986 		}
1987 	}
1988 
1989 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
1990 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
1991 
1992 	priv = kzalloc(sizeof *priv, GFP_KERNEL);
1993 	if (!priv) {
1994 		dev_err(&pdev->dev, "Device struct alloc failed, "
1995 			"aborting.\n");
1996 		err = -ENOMEM;
1997 		goto err_release_regions;
1998 	}
1999 
2000 	dev       = &priv->dev;
2001 	dev->pdev = pdev;
2002 	INIT_LIST_HEAD(&priv->ctx_list);
2003 	spin_lock_init(&priv->ctx_lock);
2004 
2005 	mutex_init(&priv->port_mutex);
2006 
2007 	INIT_LIST_HEAD(&priv->pgdir_list);
2008 	mutex_init(&priv->pgdir_mutex);
2009 
2010 	INIT_LIST_HEAD(&priv->bf_list);
2011 	mutex_init(&priv->bf_mutex);
2012 
2013 	dev->rev_id = pdev->revision;
2014 	/* Detect if this device is a virtual function */
2015 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
2016 		/* When acting as pf, we normally skip vfs unless explicitly
2017 		 * requested to probe them. */
2018 		if (num_vfs && extended_func_num(pdev) > probe_vf) {
2019 			mlx4_warn(dev, "Skipping virtual function:%d\n",
2020 						extended_func_num(pdev));
2021 			err = -ENODEV;
2022 			goto err_free_dev;
2023 		}
2024 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
2025 		dev->flags |= MLX4_FLAG_SLAVE;
2026 	} else {
2027 		/* We reset the device and enable SRIOV only for physical
2028 		 * devices.  Try to claim ownership on the device;
2029 		 * if already taken, skip -- do not allow multiple PFs */
2030 		err = mlx4_get_ownership(dev);
2031 		if (err) {
2032 			if (err < 0)
2033 				goto err_free_dev;
2034 			else {
2035 				mlx4_warn(dev, "Multiple PFs not yet supported."
2036 					  " Skipping PF.\n");
2037 				err = -EINVAL;
2038 				goto err_free_dev;
2039 			}
2040 		}
2041 
2042 		if (num_vfs) {
2043 			mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", num_vfs);
2044 			err = pci_enable_sriov(pdev, num_vfs);
2045 			if (err) {
2046 				mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d).\n",
2047 					 err);
2048 				err = 0;
2049 			} else {
2050 				mlx4_warn(dev, "Running in master mode\n");
2051 				dev->flags |= MLX4_FLAG_SRIOV |
2052 					      MLX4_FLAG_MASTER;
2053 				dev->num_vfs = num_vfs;
2054 			}
2055 		}
2056 
2057 		/*
2058 		 * Now reset the HCA before we touch the PCI capabilities or
2059 		 * attempt a firmware command, since a boot ROM may have left
2060 		 * the HCA in an undefined state.
2061 		 */
2062 		err = mlx4_reset(dev);
2063 		if (err) {
2064 			mlx4_err(dev, "Failed to reset HCA, aborting.\n");
2065 			goto err_rel_own;
2066 		}
2067 	}
2068 
2069 slave_start:
2070 	err = mlx4_cmd_init(dev);
2071 	if (err) {
2072 		mlx4_err(dev, "Failed to init command interface, aborting.\n");
2073 		goto err_sriov;
2074 	}
2075 
2076 	/* In slave functions, the communication channel must be initialized
2077 	 * before posting commands. Also, init num_slaves before calling
2078 	 * mlx4_init_hca */
2079 	if (mlx4_is_mfunc(dev)) {
2080 		if (mlx4_is_master(dev))
2081 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
2082 		else {
2083 			dev->num_slaves = 0;
2084 			if (mlx4_multi_func_init(dev)) {
2085 				mlx4_err(dev, "Failed to init slave mfunc"
2086 					 " interface, aborting.\n");
2087 				goto err_cmd;
2088 			}
2089 		}
2090 	}
2091 
2092 	err = mlx4_init_hca(dev);
2093 	if (err) {
2094 		if (err == -EACCES) {
2095 			/* Not primary Physical function
2096 			 * Running in slave mode */
2097 			mlx4_cmd_cleanup(dev);
2098 			dev->flags |= MLX4_FLAG_SLAVE;
2099 			dev->flags &= ~MLX4_FLAG_MASTER;
2100 			goto slave_start;
2101 		} else
2102 			goto err_mfunc;
2103 	}
2104 
2105 	/* In master functions, the communication channel must be initialized
2106 	 * after obtaining its address from fw */
2107 	if (mlx4_is_master(dev)) {
2108 		if (mlx4_multi_func_init(dev)) {
2109 			mlx4_err(dev, "Failed to init master mfunc"
2110 				 "interface, aborting.\n");
2111 			goto err_close;
2112 		}
2113 	}
2114 
2115 	err = mlx4_alloc_eq_table(dev);
2116 	if (err)
2117 		goto err_master_mfunc;
2118 
2119 	priv->msix_ctl.pool_bm = 0;
2120 	mutex_init(&priv->msix_ctl.pool_lock);
2121 
2122 	mlx4_enable_msi_x(dev);
2123 	if ((mlx4_is_mfunc(dev)) &&
2124 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
2125 		mlx4_err(dev, "INTx is not supported in multi-function mode."
2126 			 " aborting.\n");
2127 		goto err_free_eq;
2128 	}
2129 
2130 	if (!mlx4_is_slave(dev)) {
2131 		err = mlx4_init_steering(dev);
2132 		if (err)
2133 			goto err_free_eq;
2134 	}
2135 
2136 	err = mlx4_setup_hca(dev);
2137 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
2138 	    !mlx4_is_mfunc(dev)) {
2139 		dev->flags &= ~MLX4_FLAG_MSI_X;
2140 		dev->caps.num_comp_vectors = 1;
2141 		dev->caps.comp_pool	   = 0;
2142 		pci_disable_msix(pdev);
2143 		err = mlx4_setup_hca(dev);
2144 	}
2145 
2146 	if (err)
2147 		goto err_steer;
2148 
2149 	for (port = 1; port <= dev->caps.num_ports; port++) {
2150 		err = mlx4_init_port_info(dev, port);
2151 		if (err)
2152 			goto err_port;
2153 	}
2154 
2155 	err = mlx4_register_device(dev);
2156 	if (err)
2157 		goto err_port;
2158 
2159 	mlx4_sense_init(dev);
2160 	mlx4_start_sense(dev);
2161 
2162 	priv->pci_dev_data = pci_dev_data;
2163 	pci_set_drvdata(pdev, dev);
2164 
2165 	return 0;
2166 
2167 err_port:
2168 	for (--port; port >= 1; --port)
2169 		mlx4_cleanup_port_info(&priv->port[port]);
2170 
2171 	mlx4_cleanup_counters_table(dev);
2172 	mlx4_cleanup_mcg_table(dev);
2173 	mlx4_cleanup_qp_table(dev);
2174 	mlx4_cleanup_srq_table(dev);
2175 	mlx4_cleanup_cq_table(dev);
2176 	mlx4_cmd_use_polling(dev);
2177 	mlx4_cleanup_eq_table(dev);
2178 	mlx4_cleanup_mr_table(dev);
2179 	mlx4_cleanup_xrcd_table(dev);
2180 	mlx4_cleanup_pd_table(dev);
2181 	mlx4_cleanup_uar_table(dev);
2182 
2183 err_steer:
2184 	if (!mlx4_is_slave(dev))
2185 		mlx4_clear_steering(dev);
2186 
2187 err_free_eq:
2188 	mlx4_free_eq_table(dev);
2189 
2190 err_master_mfunc:
2191 	if (mlx4_is_master(dev))
2192 		mlx4_multi_func_cleanup(dev);
2193 
2194 err_close:
2195 	if (dev->flags & MLX4_FLAG_MSI_X)
2196 		pci_disable_msix(pdev);
2197 
2198 	mlx4_close_hca(dev);
2199 
2200 err_mfunc:
2201 	if (mlx4_is_slave(dev))
2202 		mlx4_multi_func_cleanup(dev);
2203 
2204 err_cmd:
2205 	mlx4_cmd_cleanup(dev);
2206 
2207 err_sriov:
2208 	if (dev->flags & MLX4_FLAG_SRIOV)
2209 		pci_disable_sriov(pdev);
2210 
2211 err_rel_own:
2212 	if (!mlx4_is_slave(dev))
2213 		mlx4_free_ownership(dev);
2214 
2215 err_free_dev:
2216 	kfree(priv);
2217 
2218 err_release_regions:
2219 	pci_release_regions(pdev);
2220 
2221 err_disable_pdev:
2222 	pci_disable_device(pdev);
2223 	pci_set_drvdata(pdev, NULL);
2224 	return err;
2225 }
2226 
2227 static int __devinit mlx4_init_one(struct pci_dev *pdev,
2228 				   const struct pci_device_id *id)
2229 {
2230 	printk_once(KERN_INFO "%s", mlx4_version);
2231 
2232 	return __mlx4_init_one(pdev, id->driver_data);
2233 }
2234 
2235 static void mlx4_remove_one(struct pci_dev *pdev)
2236 {
2237 	struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
2238 	struct mlx4_priv *priv = mlx4_priv(dev);
2239 	int p;
2240 
2241 	if (dev) {
2242 		/* in SRIOV it is not allowed to unload the pf's
2243 		 * driver while there are alive vf's */
2244 		if (mlx4_is_master(dev)) {
2245 			if (mlx4_how_many_lives_vf(dev))
2246 				printk(KERN_ERR "Removing PF when there are assigned VF's !!!\n");
2247 		}
2248 		mlx4_stop_sense(dev);
2249 		mlx4_unregister_device(dev);
2250 
2251 		for (p = 1; p <= dev->caps.num_ports; p++) {
2252 			mlx4_cleanup_port_info(&priv->port[p]);
2253 			mlx4_CLOSE_PORT(dev, p);
2254 		}
2255 
2256 		if (mlx4_is_master(dev))
2257 			mlx4_free_resource_tracker(dev,
2258 						   RES_TR_FREE_SLAVES_ONLY);
2259 
2260 		mlx4_cleanup_counters_table(dev);
2261 		mlx4_cleanup_mcg_table(dev);
2262 		mlx4_cleanup_qp_table(dev);
2263 		mlx4_cleanup_srq_table(dev);
2264 		mlx4_cleanup_cq_table(dev);
2265 		mlx4_cmd_use_polling(dev);
2266 		mlx4_cleanup_eq_table(dev);
2267 		mlx4_cleanup_mr_table(dev);
2268 		mlx4_cleanup_xrcd_table(dev);
2269 		mlx4_cleanup_pd_table(dev);
2270 
2271 		if (mlx4_is_master(dev))
2272 			mlx4_free_resource_tracker(dev,
2273 						   RES_TR_FREE_STRUCTS_ONLY);
2274 
2275 		iounmap(priv->kar);
2276 		mlx4_uar_free(dev, &priv->driver_uar);
2277 		mlx4_cleanup_uar_table(dev);
2278 		if (!mlx4_is_slave(dev))
2279 			mlx4_clear_steering(dev);
2280 		mlx4_free_eq_table(dev);
2281 		if (mlx4_is_master(dev))
2282 			mlx4_multi_func_cleanup(dev);
2283 		mlx4_close_hca(dev);
2284 		if (mlx4_is_slave(dev))
2285 			mlx4_multi_func_cleanup(dev);
2286 		mlx4_cmd_cleanup(dev);
2287 
2288 		if (dev->flags & MLX4_FLAG_MSI_X)
2289 			pci_disable_msix(pdev);
2290 		if (dev->flags & MLX4_FLAG_SRIOV) {
2291 			mlx4_warn(dev, "Disabling SR-IOV\n");
2292 			pci_disable_sriov(pdev);
2293 		}
2294 
2295 		if (!mlx4_is_slave(dev))
2296 			mlx4_free_ownership(dev);
2297 
2298 		kfree(dev->caps.qp0_tunnel);
2299 		kfree(dev->caps.qp0_proxy);
2300 		kfree(dev->caps.qp1_tunnel);
2301 		kfree(dev->caps.qp1_proxy);
2302 
2303 		kfree(priv);
2304 		pci_release_regions(pdev);
2305 		pci_disable_device(pdev);
2306 		pci_set_drvdata(pdev, NULL);
2307 	}
2308 }
2309 
2310 int mlx4_restart_one(struct pci_dev *pdev)
2311 {
2312 	struct mlx4_dev	 *dev  = pci_get_drvdata(pdev);
2313 	struct mlx4_priv *priv = mlx4_priv(dev);
2314 	int		  pci_dev_data;
2315 
2316 	pci_dev_data = priv->pci_dev_data;
2317 	mlx4_remove_one(pdev);
2318 	return __mlx4_init_one(pdev, pci_dev_data);
2319 }
2320 
2321 static DEFINE_PCI_DEVICE_TABLE(mlx4_pci_table) = {
2322 	/* MT25408 "Hermon" SDR */
2323 	{ PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2324 	/* MT25408 "Hermon" DDR */
2325 	{ PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2326 	/* MT25408 "Hermon" QDR */
2327 	{ PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2328 	/* MT25408 "Hermon" DDR PCIe gen2 */
2329 	{ PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2330 	/* MT25408 "Hermon" QDR PCIe gen2 */
2331 	{ PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2332 	/* MT25408 "Hermon" EN 10GigE */
2333 	{ PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2334 	/* MT25408 "Hermon" EN 10GigE PCIe gen2 */
2335 	{ PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2336 	/* MT25458 ConnectX EN 10GBASE-T 10GigE */
2337 	{ PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2338 	/* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
2339 	{ PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2340 	/* MT26468 ConnectX EN 10GigE PCIe gen2*/
2341 	{ PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2342 	/* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
2343 	{ PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2344 	/* MT26478 ConnectX2 40GigE PCIe gen2 */
2345 	{ PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2346 	/* MT25400 Family [ConnectX-2 Virtual Function] */
2347 	{ PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
2348 	/* MT27500 Family [ConnectX-3] */
2349 	{ PCI_VDEVICE(MELLANOX, 0x1003), 0 },
2350 	/* MT27500 Family [ConnectX-3 Virtual Function] */
2351 	{ PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
2352 	{ PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
2353 	{ PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
2354 	{ PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
2355 	{ PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
2356 	{ PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
2357 	{ PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
2358 	{ PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
2359 	{ PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
2360 	{ PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
2361 	{ PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
2362 	{ PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
2363 	{ PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
2364 	{ 0, }
2365 };
2366 
2367 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
2368 
2369 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
2370 					      pci_channel_state_t state)
2371 {
2372 	mlx4_remove_one(pdev);
2373 
2374 	return state == pci_channel_io_perm_failure ?
2375 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
2376 }
2377 
2378 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
2379 {
2380 	int ret = __mlx4_init_one(pdev, 0);
2381 
2382 	return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
2383 }
2384 
2385 static const struct pci_error_handlers mlx4_err_handler = {
2386 	.error_detected = mlx4_pci_err_detected,
2387 	.slot_reset     = mlx4_pci_slot_reset,
2388 };
2389 
2390 static struct pci_driver mlx4_driver = {
2391 	.name		= DRV_NAME,
2392 	.id_table	= mlx4_pci_table,
2393 	.probe		= mlx4_init_one,
2394 	.remove		= __devexit_p(mlx4_remove_one),
2395 	.err_handler    = &mlx4_err_handler,
2396 };
2397 
2398 static int __init mlx4_verify_params(void)
2399 {
2400 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
2401 		pr_warning("mlx4_core: bad num_mac: %d\n", log_num_mac);
2402 		return -1;
2403 	}
2404 
2405 	if (log_num_vlan != 0)
2406 		pr_warning("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
2407 			   MLX4_LOG_NUM_VLANS);
2408 
2409 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
2410 		pr_warning("mlx4_core: bad log_mtts_per_seg: %d\n", log_mtts_per_seg);
2411 		return -1;
2412 	}
2413 
2414 	/* Check if module param for ports type has legal combination */
2415 	if (port_type_array[0] == false && port_type_array[1] == true) {
2416 		printk(KERN_WARNING "Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
2417 		port_type_array[0] = true;
2418 	}
2419 
2420 	return 0;
2421 }
2422 
2423 static int __init mlx4_init(void)
2424 {
2425 	int ret;
2426 
2427 	if (mlx4_verify_params())
2428 		return -EINVAL;
2429 
2430 	mlx4_catas_init();
2431 
2432 	mlx4_wq = create_singlethread_workqueue("mlx4");
2433 	if (!mlx4_wq)
2434 		return -ENOMEM;
2435 
2436 	ret = pci_register_driver(&mlx4_driver);
2437 	return ret < 0 ? ret : 0;
2438 }
2439 
2440 static void __exit mlx4_cleanup(void)
2441 {
2442 	pci_unregister_driver(&mlx4_driver);
2443 	destroy_workqueue(mlx4_wq);
2444 }
2445 
2446 module_init(mlx4_init);
2447 module_exit(mlx4_cleanup);
2448