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 = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
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: 7 <="
94 					 " log_num_mgm_entry_size <= 12."
95 					 " To activate device managed"
96 					 " flow steering when available, set to -1");
97 
98 static bool enable_64b_cqe_eqe;
99 module_param(enable_64b_cqe_eqe, bool, 0444);
100 MODULE_PARM_DESC(enable_64b_cqe_eqe,
101 		 "Enable 64 byte CQEs/EQEs when the the FW supports this");
102 
103 #define HCA_GLOBAL_CAP_MASK            0
104 
105 #define PF_CONTEXT_BEHAVIOUR_MASK	MLX4_FUNC_CAP_64B_EQE_CQE
106 
107 static char mlx4_version[] =
108 	DRV_NAME ": Mellanox ConnectX core driver v"
109 	DRV_VERSION " (" DRV_RELDATE ")\n";
110 
111 static struct mlx4_profile default_profile = {
112 	.num_qp		= 1 << 18,
113 	.num_srq	= 1 << 16,
114 	.rdmarc_per_qp	= 1 << 4,
115 	.num_cq		= 1 << 16,
116 	.num_mcg	= 1 << 13,
117 	.num_mpt	= 1 << 19,
118 	.num_mtt	= 1 << 20, /* It is really num mtt segements */
119 };
120 
121 static int log_num_mac = 7;
122 module_param_named(log_num_mac, log_num_mac, int, 0444);
123 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
124 
125 static int log_num_vlan;
126 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
127 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
128 /* Log2 max number of VLANs per ETH port (0-7) */
129 #define MLX4_LOG_NUM_VLANS 7
130 
131 static bool use_prio;
132 module_param_named(use_prio, use_prio, bool, 0444);
133 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports "
134 		  "(0/1, default 0)");
135 
136 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
137 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
138 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
139 
140 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
141 static int arr_argc = 2;
142 module_param_array(port_type_array, int, &arr_argc, 0444);
143 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
144 				"1 for IB, 2 for Ethernet");
145 
146 struct mlx4_port_config {
147 	struct list_head list;
148 	enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
149 	struct pci_dev *pdev;
150 };
151 
152 int mlx4_check_port_params(struct mlx4_dev *dev,
153 			   enum mlx4_port_type *port_type)
154 {
155 	int i;
156 
157 	for (i = 0; i < dev->caps.num_ports - 1; i++) {
158 		if (port_type[i] != port_type[i + 1]) {
159 			if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
160 				mlx4_err(dev, "Only same port types supported "
161 					 "on this HCA, aborting.\n");
162 				return -EINVAL;
163 			}
164 		}
165 	}
166 
167 	for (i = 0; i < dev->caps.num_ports; i++) {
168 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
169 			mlx4_err(dev, "Requested port type for port %d is not "
170 				      "supported on this HCA\n", i + 1);
171 			return -EINVAL;
172 		}
173 	}
174 	return 0;
175 }
176 
177 static void mlx4_set_port_mask(struct mlx4_dev *dev)
178 {
179 	int i;
180 
181 	for (i = 1; i <= dev->caps.num_ports; ++i)
182 		dev->caps.port_mask[i] = dev->caps.port_type[i];
183 }
184 
185 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
186 {
187 	int err;
188 	int i;
189 
190 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
191 	if (err) {
192 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
193 		return err;
194 	}
195 
196 	if (dev_cap->min_page_sz > PAGE_SIZE) {
197 		mlx4_err(dev, "HCA minimum page size of %d bigger than "
198 			 "kernel PAGE_SIZE of %ld, aborting.\n",
199 			 dev_cap->min_page_sz, PAGE_SIZE);
200 		return -ENODEV;
201 	}
202 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
203 		mlx4_err(dev, "HCA has %d ports, but we only support %d, "
204 			 "aborting.\n",
205 			 dev_cap->num_ports, MLX4_MAX_PORTS);
206 		return -ENODEV;
207 	}
208 
209 	if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
210 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than "
211 			 "PCI resource 2 size of 0x%llx, aborting.\n",
212 			 dev_cap->uar_size,
213 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
214 		return -ENODEV;
215 	}
216 
217 	dev->caps.num_ports	     = dev_cap->num_ports;
218 	dev->phys_caps.num_phys_eqs  = MLX4_MAX_EQ_NUM;
219 	for (i = 1; i <= dev->caps.num_ports; ++i) {
220 		dev->caps.vl_cap[i]	    = dev_cap->max_vl[i];
221 		dev->caps.ib_mtu_cap[i]	    = dev_cap->ib_mtu[i];
222 		dev->phys_caps.gid_phys_table_len[i]  = dev_cap->max_gids[i];
223 		dev->phys_caps.pkey_phys_table_len[i] = dev_cap->max_pkeys[i];
224 		/* set gid and pkey table operating lengths by default
225 		 * to non-sriov values */
226 		dev->caps.gid_table_len[i]  = dev_cap->max_gids[i];
227 		dev->caps.pkey_table_len[i] = dev_cap->max_pkeys[i];
228 		dev->caps.port_width_cap[i] = dev_cap->max_port_width[i];
229 		dev->caps.eth_mtu_cap[i]    = dev_cap->eth_mtu[i];
230 		dev->caps.def_mac[i]        = dev_cap->def_mac[i];
231 		dev->caps.supported_type[i] = dev_cap->supported_port_types[i];
232 		dev->caps.suggested_type[i] = dev_cap->suggested_type[i];
233 		dev->caps.default_sense[i] = dev_cap->default_sense[i];
234 		dev->caps.trans_type[i]	    = dev_cap->trans_type[i];
235 		dev->caps.vendor_oui[i]     = dev_cap->vendor_oui[i];
236 		dev->caps.wavelength[i]     = dev_cap->wavelength[i];
237 		dev->caps.trans_code[i]     = dev_cap->trans_code[i];
238 	}
239 
240 	dev->caps.uar_page_size	     = PAGE_SIZE;
241 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
242 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
243 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
244 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
245 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
246 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
247 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
248 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
249 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
250 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
251 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
252 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
253 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
254 	/*
255 	 * Subtract 1 from the limit because we need to allocate a
256 	 * spare CQE so the HCA HW can tell the difference between an
257 	 * empty CQ and a full CQ.
258 	 */
259 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
260 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
261 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
262 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
263 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
264 
265 	/* The first 128 UARs are used for EQ doorbells */
266 	dev->caps.reserved_uars	     = max_t(int, 128, dev_cap->reserved_uars);
267 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
268 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
269 					dev_cap->reserved_xrcds : 0;
270 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
271 					dev_cap->max_xrcds : 0;
272 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
273 
274 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
275 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
276 	dev->caps.flags		     = dev_cap->flags;
277 	dev->caps.flags2	     = dev_cap->flags2;
278 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
279 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
280 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
281 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
282 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
283 
284 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
285 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
286 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
287 	/* Don't do sense port on multifunction devices (for now at least) */
288 	if (mlx4_is_mfunc(dev))
289 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
290 
291 	dev->caps.log_num_macs  = log_num_mac;
292 	dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
293 	dev->caps.log_num_prios = use_prio ? 3 : 0;
294 
295 	for (i = 1; i <= dev->caps.num_ports; ++i) {
296 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
297 		if (dev->caps.supported_type[i]) {
298 			/* if only ETH is supported - assign ETH */
299 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
300 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
301 			/* if only IB is supported, assign IB */
302 			else if (dev->caps.supported_type[i] ==
303 				 MLX4_PORT_TYPE_IB)
304 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
305 			else {
306 				/* if IB and ETH are supported, we set the port
307 				 * type according to user selection of port type;
308 				 * if user 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;
312 				else
313 					dev->caps.port_type[i] = port_type_array[i - 1];
314 			}
315 		}
316 		/*
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
321 		 */
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));
326 
327 		/*
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
331 		 */
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;
338 		} else {
339 			dev->caps.possible_type[i] = dev->caps.port_type[i];
340 		}
341 
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);
347 		}
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);
353 		}
354 	}
355 
356 	dev->caps.max_counters = 1 << ilog2(dev_cap->max_counters);
357 
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) *
364 		dev->caps.num_ports;
365 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
366 
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];
371 
372 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
373 
374 	if (!enable_64b_cqe_eqe) {
375 		if (dev_cap->flags &
376 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
377 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
378 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
379 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
380 		}
381 	}
382 
383 	if ((dev->caps.flags &
384 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
385 	    mlx4_is_master(dev))
386 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
387 
388 	return 0;
389 }
390 /*The function checks if there are live vf, return the num of them*/
391 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
392 {
393 	struct mlx4_priv *priv = mlx4_priv(dev);
394 	struct mlx4_slave_state *s_state;
395 	int i;
396 	int ret = 0;
397 
398 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
399 		s_state = &priv->mfunc.master.slave_state[i];
400 		if (s_state->active && s_state->last_cmd !=
401 		    MLX4_COMM_CMD_RESET) {
402 			mlx4_warn(dev, "%s: slave: %d is still active\n",
403 				  __func__, i);
404 			ret++;
405 		}
406 	}
407 	return ret;
408 }
409 
410 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
411 {
412 	u32 qk = MLX4_RESERVED_QKEY_BASE;
413 
414 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
415 	    qpn < dev->phys_caps.base_proxy_sqpn)
416 		return -EINVAL;
417 
418 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
419 		/* tunnel qp */
420 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
421 	else
422 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
423 	*qkey = qk;
424 	return 0;
425 }
426 EXPORT_SYMBOL(mlx4_get_parav_qkey);
427 
428 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
429 {
430 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
431 
432 	if (!mlx4_is_master(dev))
433 		return;
434 
435 	priv->virt2phys_pkey[slave][port - 1][i] = val;
436 }
437 EXPORT_SYMBOL(mlx4_sync_pkey_table);
438 
439 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
440 {
441 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
442 
443 	if (!mlx4_is_master(dev))
444 		return;
445 
446 	priv->slave_node_guids[slave] = guid;
447 }
448 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
449 
450 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
451 {
452 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
453 
454 	if (!mlx4_is_master(dev))
455 		return 0;
456 
457 	return priv->slave_node_guids[slave];
458 }
459 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
460 
461 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
462 {
463 	struct mlx4_priv *priv = mlx4_priv(dev);
464 	struct mlx4_slave_state *s_slave;
465 
466 	if (!mlx4_is_master(dev))
467 		return 0;
468 
469 	s_slave = &priv->mfunc.master.slave_state[slave];
470 	return !!s_slave->active;
471 }
472 EXPORT_SYMBOL(mlx4_is_slave_active);
473 
474 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
475 				       struct mlx4_dev_cap *dev_cap,
476 				       struct mlx4_init_hca_param *hca_param)
477 {
478 	dev->caps.steering_mode = hca_param->steering_mode;
479 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
480 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
481 		dev->caps.fs_log_max_ucast_qp_range_size =
482 			dev_cap->fs_log_max_ucast_qp_range_size;
483 	} else
484 		dev->caps.num_qp_per_mgm =
485 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
486 
487 	mlx4_dbg(dev, "Steering mode is: %s\n",
488 		 mlx4_steering_mode_str(dev->caps.steering_mode));
489 }
490 
491 static int mlx4_slave_cap(struct mlx4_dev *dev)
492 {
493 	int			   err;
494 	u32			   page_size;
495 	struct mlx4_dev_cap	   dev_cap;
496 	struct mlx4_func_cap	   func_cap;
497 	struct mlx4_init_hca_param hca_param;
498 	int			   i;
499 
500 	memset(&hca_param, 0, sizeof(hca_param));
501 	err = mlx4_QUERY_HCA(dev, &hca_param);
502 	if (err) {
503 		mlx4_err(dev, "QUERY_HCA command failed, aborting.\n");
504 		return err;
505 	}
506 
507 	/*fail if the hca has an unknown capability */
508 	if ((hca_param.global_caps | HCA_GLOBAL_CAP_MASK) !=
509 	    HCA_GLOBAL_CAP_MASK) {
510 		mlx4_err(dev, "Unknown hca global capabilities\n");
511 		return -ENOSYS;
512 	}
513 
514 	mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
515 
516 	memset(&dev_cap, 0, sizeof(dev_cap));
517 	dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
518 	err = mlx4_dev_cap(dev, &dev_cap);
519 	if (err) {
520 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
521 		return err;
522 	}
523 
524 	err = mlx4_QUERY_FW(dev);
525 	if (err)
526 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version.\n");
527 
528 	page_size = ~dev->caps.page_size_cap + 1;
529 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
530 	if (page_size > PAGE_SIZE) {
531 		mlx4_err(dev, "HCA minimum page size of %d bigger than "
532 			 "kernel PAGE_SIZE of %ld, aborting.\n",
533 			 page_size, PAGE_SIZE);
534 		return -ENODEV;
535 	}
536 
537 	/* slave gets uar page size from QUERY_HCA fw command */
538 	dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
539 
540 	/* TODO: relax this assumption */
541 	if (dev->caps.uar_page_size != PAGE_SIZE) {
542 		mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
543 			 dev->caps.uar_page_size, PAGE_SIZE);
544 		return -ENODEV;
545 	}
546 
547 	memset(&func_cap, 0, sizeof(func_cap));
548 	err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
549 	if (err) {
550 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d).\n",
551 			  err);
552 		return err;
553 	}
554 
555 	if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
556 	    PF_CONTEXT_BEHAVIOUR_MASK) {
557 		mlx4_err(dev, "Unknown pf context behaviour\n");
558 		return -ENOSYS;
559 	}
560 
561 	dev->caps.num_ports		= func_cap.num_ports;
562 	dev->caps.num_qps		= func_cap.qp_quota;
563 	dev->caps.num_srqs		= func_cap.srq_quota;
564 	dev->caps.num_cqs		= func_cap.cq_quota;
565 	dev->caps.num_eqs               = func_cap.max_eq;
566 	dev->caps.reserved_eqs          = func_cap.reserved_eq;
567 	dev->caps.num_mpts		= func_cap.mpt_quota;
568 	dev->caps.num_mtts		= func_cap.mtt_quota;
569 	dev->caps.num_pds               = MLX4_NUM_PDS;
570 	dev->caps.num_mgms              = 0;
571 	dev->caps.num_amgms             = 0;
572 
573 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
574 		mlx4_err(dev, "HCA has %d ports, but we only support %d, "
575 			 "aborting.\n", dev->caps.num_ports, MLX4_MAX_PORTS);
576 		return -ENODEV;
577 	}
578 
579 	dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
580 	dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
581 	dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
582 	dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
583 
584 	if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
585 	    !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy) {
586 		err = -ENOMEM;
587 		goto err_mem;
588 	}
589 
590 	for (i = 1; i <= dev->caps.num_ports; ++i) {
591 		err = mlx4_QUERY_FUNC_CAP(dev, (u32) i, &func_cap);
592 		if (err) {
593 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for"
594 				 " port %d, aborting (%d).\n", i, err);
595 			goto err_mem;
596 		}
597 		dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
598 		dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
599 		dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
600 		dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
601 		dev->caps.port_mask[i] = dev->caps.port_type[i];
602 		if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
603 						    &dev->caps.gid_table_len[i],
604 						    &dev->caps.pkey_table_len[i]))
605 			goto err_mem;
606 	}
607 
608 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
609 				       dev->caps.reserved_uars) >
610 				       pci_resource_len(dev->pdev, 2)) {
611 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than "
612 			 "PCI resource 2 size of 0x%llx, aborting.\n",
613 			 dev->caps.uar_page_size * dev->caps.num_uars,
614 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
615 		goto err_mem;
616 	}
617 
618 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
619 		dev->caps.eqe_size   = 64;
620 		dev->caps.eqe_factor = 1;
621 	} else {
622 		dev->caps.eqe_size   = 32;
623 		dev->caps.eqe_factor = 0;
624 	}
625 
626 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
627 		dev->caps.cqe_size   = 64;
628 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_64B_CQE;
629 	} else {
630 		dev->caps.cqe_size   = 32;
631 	}
632 
633 	slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
634 
635 	return 0;
636 
637 err_mem:
638 	kfree(dev->caps.qp0_tunnel);
639 	kfree(dev->caps.qp0_proxy);
640 	kfree(dev->caps.qp1_tunnel);
641 	kfree(dev->caps.qp1_proxy);
642 	dev->caps.qp0_tunnel = dev->caps.qp0_proxy =
643 		dev->caps.qp1_tunnel = dev->caps.qp1_proxy = NULL;
644 
645 	return err;
646 }
647 
648 /*
649  * Change the port configuration of the device.
650  * Every user of this function must hold the port mutex.
651  */
652 int mlx4_change_port_types(struct mlx4_dev *dev,
653 			   enum mlx4_port_type *port_types)
654 {
655 	int err = 0;
656 	int change = 0;
657 	int port;
658 
659 	for (port = 0; port <  dev->caps.num_ports; port++) {
660 		/* Change the port type only if the new type is different
661 		 * from the current, and not set to Auto */
662 		if (port_types[port] != dev->caps.port_type[port + 1])
663 			change = 1;
664 	}
665 	if (change) {
666 		mlx4_unregister_device(dev);
667 		for (port = 1; port <= dev->caps.num_ports; port++) {
668 			mlx4_CLOSE_PORT(dev, port);
669 			dev->caps.port_type[port] = port_types[port - 1];
670 			err = mlx4_SET_PORT(dev, port, -1);
671 			if (err) {
672 				mlx4_err(dev, "Failed to set port %d, "
673 					      "aborting\n", port);
674 				goto out;
675 			}
676 		}
677 		mlx4_set_port_mask(dev);
678 		err = mlx4_register_device(dev);
679 	}
680 
681 out:
682 	return err;
683 }
684 
685 static ssize_t show_port_type(struct device *dev,
686 			      struct device_attribute *attr,
687 			      char *buf)
688 {
689 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
690 						   port_attr);
691 	struct mlx4_dev *mdev = info->dev;
692 	char type[8];
693 
694 	sprintf(type, "%s",
695 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
696 		"ib" : "eth");
697 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
698 		sprintf(buf, "auto (%s)\n", type);
699 	else
700 		sprintf(buf, "%s\n", type);
701 
702 	return strlen(buf);
703 }
704 
705 static ssize_t set_port_type(struct device *dev,
706 			     struct device_attribute *attr,
707 			     const char *buf, size_t count)
708 {
709 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
710 						   port_attr);
711 	struct mlx4_dev *mdev = info->dev;
712 	struct mlx4_priv *priv = mlx4_priv(mdev);
713 	enum mlx4_port_type types[MLX4_MAX_PORTS];
714 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
715 	int i;
716 	int err = 0;
717 
718 	if (!strcmp(buf, "ib\n"))
719 		info->tmp_type = MLX4_PORT_TYPE_IB;
720 	else if (!strcmp(buf, "eth\n"))
721 		info->tmp_type = MLX4_PORT_TYPE_ETH;
722 	else if (!strcmp(buf, "auto\n"))
723 		info->tmp_type = MLX4_PORT_TYPE_AUTO;
724 	else {
725 		mlx4_err(mdev, "%s is not supported port type\n", buf);
726 		return -EINVAL;
727 	}
728 
729 	mlx4_stop_sense(mdev);
730 	mutex_lock(&priv->port_mutex);
731 	/* Possible type is always the one that was delivered */
732 	mdev->caps.possible_type[info->port] = info->tmp_type;
733 
734 	for (i = 0; i < mdev->caps.num_ports; i++) {
735 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
736 					mdev->caps.possible_type[i+1];
737 		if (types[i] == MLX4_PORT_TYPE_AUTO)
738 			types[i] = mdev->caps.port_type[i+1];
739 	}
740 
741 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
742 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
743 		for (i = 1; i <= mdev->caps.num_ports; i++) {
744 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
745 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
746 				err = -EINVAL;
747 			}
748 		}
749 	}
750 	if (err) {
751 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. "
752 			       "Set only 'eth' or 'ib' for both ports "
753 			       "(should be the same)\n");
754 		goto out;
755 	}
756 
757 	mlx4_do_sense_ports(mdev, new_types, types);
758 
759 	err = mlx4_check_port_params(mdev, new_types);
760 	if (err)
761 		goto out;
762 
763 	/* We are about to apply the changes after the configuration
764 	 * was verified, no need to remember the temporary types
765 	 * any more */
766 	for (i = 0; i < mdev->caps.num_ports; i++)
767 		priv->port[i + 1].tmp_type = 0;
768 
769 	err = mlx4_change_port_types(mdev, new_types);
770 
771 out:
772 	mlx4_start_sense(mdev);
773 	mutex_unlock(&priv->port_mutex);
774 	return err ? err : count;
775 }
776 
777 enum ibta_mtu {
778 	IB_MTU_256  = 1,
779 	IB_MTU_512  = 2,
780 	IB_MTU_1024 = 3,
781 	IB_MTU_2048 = 4,
782 	IB_MTU_4096 = 5
783 };
784 
785 static inline int int_to_ibta_mtu(int mtu)
786 {
787 	switch (mtu) {
788 	case 256:  return IB_MTU_256;
789 	case 512:  return IB_MTU_512;
790 	case 1024: return IB_MTU_1024;
791 	case 2048: return IB_MTU_2048;
792 	case 4096: return IB_MTU_4096;
793 	default: return -1;
794 	}
795 }
796 
797 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
798 {
799 	switch (mtu) {
800 	case IB_MTU_256:  return  256;
801 	case IB_MTU_512:  return  512;
802 	case IB_MTU_1024: return 1024;
803 	case IB_MTU_2048: return 2048;
804 	case IB_MTU_4096: return 4096;
805 	default: return -1;
806 	}
807 }
808 
809 static ssize_t show_port_ib_mtu(struct device *dev,
810 			     struct device_attribute *attr,
811 			     char *buf)
812 {
813 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
814 						   port_mtu_attr);
815 	struct mlx4_dev *mdev = info->dev;
816 
817 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
818 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
819 
820 	sprintf(buf, "%d\n",
821 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
822 	return strlen(buf);
823 }
824 
825 static ssize_t set_port_ib_mtu(struct device *dev,
826 			     struct device_attribute *attr,
827 			     const char *buf, size_t count)
828 {
829 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
830 						   port_mtu_attr);
831 	struct mlx4_dev *mdev = info->dev;
832 	struct mlx4_priv *priv = mlx4_priv(mdev);
833 	int err, port, mtu, ibta_mtu = -1;
834 
835 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
836 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
837 		return -EINVAL;
838 	}
839 
840 	err = sscanf(buf, "%d", &mtu);
841 	if (err > 0)
842 		ibta_mtu = int_to_ibta_mtu(mtu);
843 
844 	if (err <= 0 || ibta_mtu < 0) {
845 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
846 		return -EINVAL;
847 	}
848 
849 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
850 
851 	mlx4_stop_sense(mdev);
852 	mutex_lock(&priv->port_mutex);
853 	mlx4_unregister_device(mdev);
854 	for (port = 1; port <= mdev->caps.num_ports; port++) {
855 		mlx4_CLOSE_PORT(mdev, port);
856 		err = mlx4_SET_PORT(mdev, port, -1);
857 		if (err) {
858 			mlx4_err(mdev, "Failed to set port %d, "
859 				      "aborting\n", port);
860 			goto err_set_port;
861 		}
862 	}
863 	err = mlx4_register_device(mdev);
864 err_set_port:
865 	mutex_unlock(&priv->port_mutex);
866 	mlx4_start_sense(mdev);
867 	return err ? err : count;
868 }
869 
870 static int mlx4_load_fw(struct mlx4_dev *dev)
871 {
872 	struct mlx4_priv *priv = mlx4_priv(dev);
873 	int err;
874 
875 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
876 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
877 	if (!priv->fw.fw_icm) {
878 		mlx4_err(dev, "Couldn't allocate FW area, aborting.\n");
879 		return -ENOMEM;
880 	}
881 
882 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
883 	if (err) {
884 		mlx4_err(dev, "MAP_FA command failed, aborting.\n");
885 		goto err_free;
886 	}
887 
888 	err = mlx4_RUN_FW(dev);
889 	if (err) {
890 		mlx4_err(dev, "RUN_FW command failed, aborting.\n");
891 		goto err_unmap_fa;
892 	}
893 
894 	return 0;
895 
896 err_unmap_fa:
897 	mlx4_UNMAP_FA(dev);
898 
899 err_free:
900 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
901 	return err;
902 }
903 
904 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
905 				int cmpt_entry_sz)
906 {
907 	struct mlx4_priv *priv = mlx4_priv(dev);
908 	int err;
909 	int num_eqs;
910 
911 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
912 				  cmpt_base +
913 				  ((u64) (MLX4_CMPT_TYPE_QP *
914 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
915 				  cmpt_entry_sz, dev->caps.num_qps,
916 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
917 				  0, 0);
918 	if (err)
919 		goto err;
920 
921 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
922 				  cmpt_base +
923 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
924 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
925 				  cmpt_entry_sz, dev->caps.num_srqs,
926 				  dev->caps.reserved_srqs, 0, 0);
927 	if (err)
928 		goto err_qp;
929 
930 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
931 				  cmpt_base +
932 				  ((u64) (MLX4_CMPT_TYPE_CQ *
933 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
934 				  cmpt_entry_sz, dev->caps.num_cqs,
935 				  dev->caps.reserved_cqs, 0, 0);
936 	if (err)
937 		goto err_srq;
938 
939 	num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
940 		  dev->caps.num_eqs;
941 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
942 				  cmpt_base +
943 				  ((u64) (MLX4_CMPT_TYPE_EQ *
944 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
945 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
946 	if (err)
947 		goto err_cq;
948 
949 	return 0;
950 
951 err_cq:
952 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
953 
954 err_srq:
955 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
956 
957 err_qp:
958 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
959 
960 err:
961 	return err;
962 }
963 
964 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
965 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
966 {
967 	struct mlx4_priv *priv = mlx4_priv(dev);
968 	u64 aux_pages;
969 	int num_eqs;
970 	int err;
971 
972 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
973 	if (err) {
974 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting.\n");
975 		return err;
976 	}
977 
978 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory.\n",
979 		 (unsigned long long) icm_size >> 10,
980 		 (unsigned long long) aux_pages << 2);
981 
982 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
983 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
984 	if (!priv->fw.aux_icm) {
985 		mlx4_err(dev, "Couldn't allocate aux memory, aborting.\n");
986 		return -ENOMEM;
987 	}
988 
989 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
990 	if (err) {
991 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting.\n");
992 		goto err_free_aux;
993 	}
994 
995 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
996 	if (err) {
997 		mlx4_err(dev, "Failed to map cMPT context memory, aborting.\n");
998 		goto err_unmap_aux;
999 	}
1000 
1001 
1002 	num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
1003 		   dev->caps.num_eqs;
1004 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1005 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1006 				  num_eqs, num_eqs, 0, 0);
1007 	if (err) {
1008 		mlx4_err(dev, "Failed to map EQ context memory, aborting.\n");
1009 		goto err_unmap_cmpt;
1010 	}
1011 
1012 	/*
1013 	 * Reserved MTT entries must be aligned up to a cacheline
1014 	 * boundary, since the FW will write to them, while the driver
1015 	 * writes to all other MTT entries. (The variable
1016 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1017 	 * size, not the raw entry size)
1018 	 */
1019 	dev->caps.reserved_mtts =
1020 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1021 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1022 
1023 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1024 				  init_hca->mtt_base,
1025 				  dev->caps.mtt_entry_sz,
1026 				  dev->caps.num_mtts,
1027 				  dev->caps.reserved_mtts, 1, 0);
1028 	if (err) {
1029 		mlx4_err(dev, "Failed to map MTT context memory, aborting.\n");
1030 		goto err_unmap_eq;
1031 	}
1032 
1033 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1034 				  init_hca->dmpt_base,
1035 				  dev_cap->dmpt_entry_sz,
1036 				  dev->caps.num_mpts,
1037 				  dev->caps.reserved_mrws, 1, 1);
1038 	if (err) {
1039 		mlx4_err(dev, "Failed to map dMPT context memory, aborting.\n");
1040 		goto err_unmap_mtt;
1041 	}
1042 
1043 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1044 				  init_hca->qpc_base,
1045 				  dev_cap->qpc_entry_sz,
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 QP context memory, aborting.\n");
1051 		goto err_unmap_dmpt;
1052 	}
1053 
1054 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1055 				  init_hca->auxc_base,
1056 				  dev_cap->aux_entry_sz,
1057 				  dev->caps.num_qps,
1058 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1059 				  0, 0);
1060 	if (err) {
1061 		mlx4_err(dev, "Failed to map AUXC context memory, aborting.\n");
1062 		goto err_unmap_qp;
1063 	}
1064 
1065 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1066 				  init_hca->altc_base,
1067 				  dev_cap->altc_entry_sz,
1068 				  dev->caps.num_qps,
1069 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1070 				  0, 0);
1071 	if (err) {
1072 		mlx4_err(dev, "Failed to map ALTC context memory, aborting.\n");
1073 		goto err_unmap_auxc;
1074 	}
1075 
1076 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1077 				  init_hca->rdmarc_base,
1078 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1079 				  dev->caps.num_qps,
1080 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1081 				  0, 0);
1082 	if (err) {
1083 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1084 		goto err_unmap_altc;
1085 	}
1086 
1087 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1088 				  init_hca->cqc_base,
1089 				  dev_cap->cqc_entry_sz,
1090 				  dev->caps.num_cqs,
1091 				  dev->caps.reserved_cqs, 0, 0);
1092 	if (err) {
1093 		mlx4_err(dev, "Failed to map CQ context memory, aborting.\n");
1094 		goto err_unmap_rdmarc;
1095 	}
1096 
1097 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1098 				  init_hca->srqc_base,
1099 				  dev_cap->srq_entry_sz,
1100 				  dev->caps.num_srqs,
1101 				  dev->caps.reserved_srqs, 0, 0);
1102 	if (err) {
1103 		mlx4_err(dev, "Failed to map SRQ context memory, aborting.\n");
1104 		goto err_unmap_cq;
1105 	}
1106 
1107 	/*
1108 	 * For flow steering device managed mode it is required to use
1109 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1110 	 * required, but for simplicity just map the whole multicast
1111 	 * group table now.  The table isn't very big and it's a lot
1112 	 * easier than trying to track ref counts.
1113 	 */
1114 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1115 				  init_hca->mc_base,
1116 				  mlx4_get_mgm_entry_size(dev),
1117 				  dev->caps.num_mgms + dev->caps.num_amgms,
1118 				  dev->caps.num_mgms + dev->caps.num_amgms,
1119 				  0, 0);
1120 	if (err) {
1121 		mlx4_err(dev, "Failed to map MCG context memory, aborting.\n");
1122 		goto err_unmap_srq;
1123 	}
1124 
1125 	return 0;
1126 
1127 err_unmap_srq:
1128 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1129 
1130 err_unmap_cq:
1131 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1132 
1133 err_unmap_rdmarc:
1134 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1135 
1136 err_unmap_altc:
1137 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1138 
1139 err_unmap_auxc:
1140 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1141 
1142 err_unmap_qp:
1143 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1144 
1145 err_unmap_dmpt:
1146 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1147 
1148 err_unmap_mtt:
1149 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1150 
1151 err_unmap_eq:
1152 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1153 
1154 err_unmap_cmpt:
1155 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1156 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1157 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1158 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1159 
1160 err_unmap_aux:
1161 	mlx4_UNMAP_ICM_AUX(dev);
1162 
1163 err_free_aux:
1164 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1165 
1166 	return err;
1167 }
1168 
1169 static void mlx4_free_icms(struct mlx4_dev *dev)
1170 {
1171 	struct mlx4_priv *priv = mlx4_priv(dev);
1172 
1173 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1174 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1175 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1176 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1177 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1178 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1179 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1180 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1181 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1182 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1183 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1184 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1185 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1186 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1187 
1188 	mlx4_UNMAP_ICM_AUX(dev);
1189 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1190 }
1191 
1192 static void mlx4_slave_exit(struct mlx4_dev *dev)
1193 {
1194 	struct mlx4_priv *priv = mlx4_priv(dev);
1195 
1196 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1197 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_TIME))
1198 		mlx4_warn(dev, "Failed to close slave function.\n");
1199 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1200 }
1201 
1202 static int map_bf_area(struct mlx4_dev *dev)
1203 {
1204 	struct mlx4_priv *priv = mlx4_priv(dev);
1205 	resource_size_t bf_start;
1206 	resource_size_t bf_len;
1207 	int err = 0;
1208 
1209 	if (!dev->caps.bf_reg_size)
1210 		return -ENXIO;
1211 
1212 	bf_start = pci_resource_start(dev->pdev, 2) +
1213 			(dev->caps.num_uars << PAGE_SHIFT);
1214 	bf_len = pci_resource_len(dev->pdev, 2) -
1215 			(dev->caps.num_uars << PAGE_SHIFT);
1216 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1217 	if (!priv->bf_mapping)
1218 		err = -ENOMEM;
1219 
1220 	return err;
1221 }
1222 
1223 static void unmap_bf_area(struct mlx4_dev *dev)
1224 {
1225 	if (mlx4_priv(dev)->bf_mapping)
1226 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1227 }
1228 
1229 static void mlx4_close_hca(struct mlx4_dev *dev)
1230 {
1231 	unmap_bf_area(dev);
1232 	if (mlx4_is_slave(dev))
1233 		mlx4_slave_exit(dev);
1234 	else {
1235 		mlx4_CLOSE_HCA(dev, 0);
1236 		mlx4_free_icms(dev);
1237 		mlx4_UNMAP_FA(dev);
1238 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1239 	}
1240 }
1241 
1242 static int mlx4_init_slave(struct mlx4_dev *dev)
1243 {
1244 	struct mlx4_priv *priv = mlx4_priv(dev);
1245 	u64 dma = (u64) priv->mfunc.vhcr_dma;
1246 	int num_of_reset_retries = NUM_OF_RESET_RETRIES;
1247 	int ret_from_reset = 0;
1248 	u32 slave_read;
1249 	u32 cmd_channel_ver;
1250 
1251 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1252 	priv->cmd.max_cmds = 1;
1253 	mlx4_warn(dev, "Sending reset\n");
1254 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1255 				       MLX4_COMM_TIME);
1256 	/* if we are in the middle of flr the slave will try
1257 	 * NUM_OF_RESET_RETRIES times before leaving.*/
1258 	if (ret_from_reset) {
1259 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1260 			msleep(SLEEP_TIME_IN_RESET);
1261 			while (ret_from_reset && num_of_reset_retries) {
1262 				mlx4_warn(dev, "slave is currently in the"
1263 					  "middle of FLR. retrying..."
1264 					  "(try num:%d)\n",
1265 					  (NUM_OF_RESET_RETRIES -
1266 					   num_of_reset_retries  + 1));
1267 				ret_from_reset =
1268 					mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET,
1269 						      0, MLX4_COMM_TIME);
1270 				num_of_reset_retries = num_of_reset_retries - 1;
1271 			}
1272 		} else
1273 			goto err;
1274 	}
1275 
1276 	/* check the driver version - the slave I/F revision
1277 	 * must match the master's */
1278 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1279 	cmd_channel_ver = mlx4_comm_get_version();
1280 
1281 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1282 		MLX4_COMM_GET_IF_REV(slave_read)) {
1283 		mlx4_err(dev, "slave driver version is not supported"
1284 			 " by the master\n");
1285 		goto err;
1286 	}
1287 
1288 	mlx4_warn(dev, "Sending vhcr0\n");
1289 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1290 						    MLX4_COMM_TIME))
1291 		goto err;
1292 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1293 						    MLX4_COMM_TIME))
1294 		goto err;
1295 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1296 						    MLX4_COMM_TIME))
1297 		goto err;
1298 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma, MLX4_COMM_TIME))
1299 		goto err;
1300 
1301 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1302 	return 0;
1303 
1304 err:
1305 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, 0);
1306 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1307 	return -EIO;
1308 }
1309 
1310 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1311 {
1312 	int i;
1313 
1314 	for (i = 1; i <= dev->caps.num_ports; i++) {
1315 		dev->caps.gid_table_len[i] = 1;
1316 		dev->caps.pkey_table_len[i] =
1317 			dev->phys_caps.pkey_phys_table_len[i] - 1;
1318 	}
1319 }
1320 
1321 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
1322 {
1323 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
1324 
1325 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
1326 	      i++) {
1327 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
1328 			break;
1329 	}
1330 
1331 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
1332 }
1333 
1334 static void choose_steering_mode(struct mlx4_dev *dev,
1335 				 struct mlx4_dev_cap *dev_cap)
1336 {
1337 	if (mlx4_log_num_mgm_entry_size == -1 &&
1338 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
1339 	    (!mlx4_is_mfunc(dev) ||
1340 	     (dev_cap->fs_max_num_qp_per_entry >= (num_vfs + 1))) &&
1341 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
1342 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
1343 		dev->oper_log_mgm_entry_size =
1344 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
1345 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1346 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
1347 		dev->caps.fs_log_max_ucast_qp_range_size =
1348 			dev_cap->fs_log_max_ucast_qp_range_size;
1349 	} else {
1350 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
1351 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1352 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
1353 		else {
1354 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
1355 
1356 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
1357 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1358 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags "
1359 					  "set to use B0 steering. Falling back to A0 steering mode.\n");
1360 		}
1361 		dev->oper_log_mgm_entry_size =
1362 			mlx4_log_num_mgm_entry_size > 0 ?
1363 			mlx4_log_num_mgm_entry_size :
1364 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
1365 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
1366 	}
1367 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, "
1368 		 "modparam log_num_mgm_entry_size = %d\n",
1369 		 mlx4_steering_mode_str(dev->caps.steering_mode),
1370 		 dev->oper_log_mgm_entry_size,
1371 		 mlx4_log_num_mgm_entry_size);
1372 }
1373 
1374 static int mlx4_init_hca(struct mlx4_dev *dev)
1375 {
1376 	struct mlx4_priv	  *priv = mlx4_priv(dev);
1377 	struct mlx4_adapter	   adapter;
1378 	struct mlx4_dev_cap	   dev_cap;
1379 	struct mlx4_mod_stat_cfg   mlx4_cfg;
1380 	struct mlx4_profile	   profile;
1381 	struct mlx4_init_hca_param init_hca;
1382 	u64 icm_size;
1383 	int err;
1384 
1385 	if (!mlx4_is_slave(dev)) {
1386 		err = mlx4_QUERY_FW(dev);
1387 		if (err) {
1388 			if (err == -EACCES)
1389 				mlx4_info(dev, "non-primary physical function, skipping.\n");
1390 			else
1391 				mlx4_err(dev, "QUERY_FW command failed, aborting.\n");
1392 			return err;
1393 		}
1394 
1395 		err = mlx4_load_fw(dev);
1396 		if (err) {
1397 			mlx4_err(dev, "Failed to start FW, aborting.\n");
1398 			return err;
1399 		}
1400 
1401 		mlx4_cfg.log_pg_sz_m = 1;
1402 		mlx4_cfg.log_pg_sz = 0;
1403 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
1404 		if (err)
1405 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
1406 
1407 		err = mlx4_dev_cap(dev, &dev_cap);
1408 		if (err) {
1409 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
1410 			goto err_stop_fw;
1411 		}
1412 
1413 		choose_steering_mode(dev, &dev_cap);
1414 
1415 		if (mlx4_is_master(dev))
1416 			mlx4_parav_master_pf_caps(dev);
1417 
1418 		profile = default_profile;
1419 		if (dev->caps.steering_mode ==
1420 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
1421 			profile.num_mcg = MLX4_FS_NUM_MCG;
1422 
1423 		icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
1424 					     &init_hca);
1425 		if ((long long) icm_size < 0) {
1426 			err = icm_size;
1427 			goto err_stop_fw;
1428 		}
1429 
1430 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
1431 
1432 		init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
1433 		init_hca.uar_page_sz = PAGE_SHIFT - 12;
1434 
1435 		err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
1436 		if (err)
1437 			goto err_stop_fw;
1438 
1439 		err = mlx4_INIT_HCA(dev, &init_hca);
1440 		if (err) {
1441 			mlx4_err(dev, "INIT_HCA command failed, aborting.\n");
1442 			goto err_free_icm;
1443 		}
1444 	} else {
1445 		err = mlx4_init_slave(dev);
1446 		if (err) {
1447 			mlx4_err(dev, "Failed to initialize slave\n");
1448 			return err;
1449 		}
1450 
1451 		err = mlx4_slave_cap(dev);
1452 		if (err) {
1453 			mlx4_err(dev, "Failed to obtain slave caps\n");
1454 			goto err_close;
1455 		}
1456 	}
1457 
1458 	if (map_bf_area(dev))
1459 		mlx4_dbg(dev, "Failed to map blue flame area\n");
1460 
1461 	/*Only the master set the ports, all the rest got it from it.*/
1462 	if (!mlx4_is_slave(dev))
1463 		mlx4_set_port_mask(dev);
1464 
1465 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
1466 	if (err) {
1467 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting.\n");
1468 		goto unmap_bf;
1469 	}
1470 
1471 	priv->eq_table.inta_pin = adapter.inta_pin;
1472 	memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
1473 
1474 	return 0;
1475 
1476 unmap_bf:
1477 	unmap_bf_area(dev);
1478 
1479 err_close:
1480 	if (mlx4_is_slave(dev))
1481 		mlx4_slave_exit(dev);
1482 	else
1483 		mlx4_CLOSE_HCA(dev, 0);
1484 
1485 err_free_icm:
1486 	if (!mlx4_is_slave(dev))
1487 		mlx4_free_icms(dev);
1488 
1489 err_stop_fw:
1490 	if (!mlx4_is_slave(dev)) {
1491 		mlx4_UNMAP_FA(dev);
1492 		mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1493 	}
1494 	return err;
1495 }
1496 
1497 static int mlx4_init_counters_table(struct mlx4_dev *dev)
1498 {
1499 	struct mlx4_priv *priv = mlx4_priv(dev);
1500 	int nent;
1501 
1502 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1503 		return -ENOENT;
1504 
1505 	nent = dev->caps.max_counters;
1506 	return mlx4_bitmap_init(&priv->counters_bitmap, nent, nent - 1, 0, 0);
1507 }
1508 
1509 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
1510 {
1511 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
1512 }
1513 
1514 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1515 {
1516 	struct mlx4_priv *priv = mlx4_priv(dev);
1517 
1518 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1519 		return -ENOENT;
1520 
1521 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
1522 	if (*idx == -1)
1523 		return -ENOMEM;
1524 
1525 	return 0;
1526 }
1527 
1528 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1529 {
1530 	u64 out_param;
1531 	int err;
1532 
1533 	if (mlx4_is_mfunc(dev)) {
1534 		err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
1535 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
1536 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
1537 		if (!err)
1538 			*idx = get_param_l(&out_param);
1539 
1540 		return err;
1541 	}
1542 	return __mlx4_counter_alloc(dev, idx);
1543 }
1544 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
1545 
1546 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1547 {
1548 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx);
1549 	return;
1550 }
1551 
1552 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1553 {
1554 	u64 in_param;
1555 
1556 	if (mlx4_is_mfunc(dev)) {
1557 		set_param_l(&in_param, idx);
1558 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
1559 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
1560 			 MLX4_CMD_WRAPPED);
1561 		return;
1562 	}
1563 	__mlx4_counter_free(dev, idx);
1564 }
1565 EXPORT_SYMBOL_GPL(mlx4_counter_free);
1566 
1567 static int mlx4_setup_hca(struct mlx4_dev *dev)
1568 {
1569 	struct mlx4_priv *priv = mlx4_priv(dev);
1570 	int err;
1571 	int port;
1572 	__be32 ib_port_default_caps;
1573 
1574 	err = mlx4_init_uar_table(dev);
1575 	if (err) {
1576 		mlx4_err(dev, "Failed to initialize "
1577 			 "user access region table, aborting.\n");
1578 		return err;
1579 	}
1580 
1581 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
1582 	if (err) {
1583 		mlx4_err(dev, "Failed to allocate driver access region, "
1584 			 "aborting.\n");
1585 		goto err_uar_table_free;
1586 	}
1587 
1588 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
1589 	if (!priv->kar) {
1590 		mlx4_err(dev, "Couldn't map kernel access region, "
1591 			 "aborting.\n");
1592 		err = -ENOMEM;
1593 		goto err_uar_free;
1594 	}
1595 
1596 	err = mlx4_init_pd_table(dev);
1597 	if (err) {
1598 		mlx4_err(dev, "Failed to initialize "
1599 			 "protection domain table, aborting.\n");
1600 		goto err_kar_unmap;
1601 	}
1602 
1603 	err = mlx4_init_xrcd_table(dev);
1604 	if (err) {
1605 		mlx4_err(dev, "Failed to initialize "
1606 			 "reliable connection domain table, aborting.\n");
1607 		goto err_pd_table_free;
1608 	}
1609 
1610 	err = mlx4_init_mr_table(dev);
1611 	if (err) {
1612 		mlx4_err(dev, "Failed to initialize "
1613 			 "memory region table, aborting.\n");
1614 		goto err_xrcd_table_free;
1615 	}
1616 
1617 	err = mlx4_init_eq_table(dev);
1618 	if (err) {
1619 		mlx4_err(dev, "Failed to initialize "
1620 			 "event queue table, aborting.\n");
1621 		goto err_mr_table_free;
1622 	}
1623 
1624 	err = mlx4_cmd_use_events(dev);
1625 	if (err) {
1626 		mlx4_err(dev, "Failed to switch to event-driven "
1627 			 "firmware commands, aborting.\n");
1628 		goto err_eq_table_free;
1629 	}
1630 
1631 	err = mlx4_NOP(dev);
1632 	if (err) {
1633 		if (dev->flags & MLX4_FLAG_MSI_X) {
1634 			mlx4_warn(dev, "NOP command failed to generate MSI-X "
1635 				  "interrupt IRQ %d).\n",
1636 				  priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1637 			mlx4_warn(dev, "Trying again without MSI-X.\n");
1638 		} else {
1639 			mlx4_err(dev, "NOP command failed to generate interrupt "
1640 				 "(IRQ %d), aborting.\n",
1641 				 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1642 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
1643 		}
1644 
1645 		goto err_cmd_poll;
1646 	}
1647 
1648 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
1649 
1650 	err = mlx4_init_cq_table(dev);
1651 	if (err) {
1652 		mlx4_err(dev, "Failed to initialize "
1653 			 "completion queue table, aborting.\n");
1654 		goto err_cmd_poll;
1655 	}
1656 
1657 	err = mlx4_init_srq_table(dev);
1658 	if (err) {
1659 		mlx4_err(dev, "Failed to initialize "
1660 			 "shared receive queue table, aborting.\n");
1661 		goto err_cq_table_free;
1662 	}
1663 
1664 	err = mlx4_init_qp_table(dev);
1665 	if (err) {
1666 		mlx4_err(dev, "Failed to initialize "
1667 			 "queue pair table, aborting.\n");
1668 		goto err_srq_table_free;
1669 	}
1670 
1671 	if (!mlx4_is_slave(dev)) {
1672 		err = mlx4_init_mcg_table(dev);
1673 		if (err) {
1674 			mlx4_err(dev, "Failed to initialize "
1675 				 "multicast group table, aborting.\n");
1676 			goto err_qp_table_free;
1677 		}
1678 	}
1679 
1680 	err = mlx4_init_counters_table(dev);
1681 	if (err && err != -ENOENT) {
1682 		mlx4_err(dev, "Failed to initialize counters table, aborting.\n");
1683 		goto err_mcg_table_free;
1684 	}
1685 
1686 	if (!mlx4_is_slave(dev)) {
1687 		for (port = 1; port <= dev->caps.num_ports; port++) {
1688 			ib_port_default_caps = 0;
1689 			err = mlx4_get_port_ib_caps(dev, port,
1690 						    &ib_port_default_caps);
1691 			if (err)
1692 				mlx4_warn(dev, "failed to get port %d default "
1693 					  "ib capabilities (%d). Continuing "
1694 					  "with caps = 0\n", port, err);
1695 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
1696 
1697 			/* initialize per-slave default ib port capabilities */
1698 			if (mlx4_is_master(dev)) {
1699 				int i;
1700 				for (i = 0; i < dev->num_slaves; i++) {
1701 					if (i == mlx4_master_func_num(dev))
1702 						continue;
1703 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
1704 							ib_port_default_caps;
1705 				}
1706 			}
1707 
1708 			if (mlx4_is_mfunc(dev))
1709 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
1710 			else
1711 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
1712 
1713 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
1714 					    dev->caps.pkey_table_len[port] : -1);
1715 			if (err) {
1716 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1717 					port);
1718 				goto err_counters_table_free;
1719 			}
1720 		}
1721 	}
1722 
1723 	return 0;
1724 
1725 err_counters_table_free:
1726 	mlx4_cleanup_counters_table(dev);
1727 
1728 err_mcg_table_free:
1729 	mlx4_cleanup_mcg_table(dev);
1730 
1731 err_qp_table_free:
1732 	mlx4_cleanup_qp_table(dev);
1733 
1734 err_srq_table_free:
1735 	mlx4_cleanup_srq_table(dev);
1736 
1737 err_cq_table_free:
1738 	mlx4_cleanup_cq_table(dev);
1739 
1740 err_cmd_poll:
1741 	mlx4_cmd_use_polling(dev);
1742 
1743 err_eq_table_free:
1744 	mlx4_cleanup_eq_table(dev);
1745 
1746 err_mr_table_free:
1747 	mlx4_cleanup_mr_table(dev);
1748 
1749 err_xrcd_table_free:
1750 	mlx4_cleanup_xrcd_table(dev);
1751 
1752 err_pd_table_free:
1753 	mlx4_cleanup_pd_table(dev);
1754 
1755 err_kar_unmap:
1756 	iounmap(priv->kar);
1757 
1758 err_uar_free:
1759 	mlx4_uar_free(dev, &priv->driver_uar);
1760 
1761 err_uar_table_free:
1762 	mlx4_cleanup_uar_table(dev);
1763 	return err;
1764 }
1765 
1766 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
1767 {
1768 	struct mlx4_priv *priv = mlx4_priv(dev);
1769 	struct msix_entry *entries;
1770 	int nreq = min_t(int, dev->caps.num_ports *
1771 			 min_t(int, netif_get_num_default_rss_queues() + 1,
1772 			       MAX_MSIX_P_PORT) + MSIX_LEGACY_SZ, MAX_MSIX);
1773 	int err;
1774 	int i;
1775 
1776 	if (msi_x) {
1777 		nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
1778 			     nreq);
1779 
1780 		entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
1781 		if (!entries)
1782 			goto no_msi;
1783 
1784 		for (i = 0; i < nreq; ++i)
1785 			entries[i].entry = i;
1786 
1787 	retry:
1788 		err = pci_enable_msix(dev->pdev, entries, nreq);
1789 		if (err) {
1790 			/* Try again if at least 2 vectors are available */
1791 			if (err > 1) {
1792 				mlx4_info(dev, "Requested %d vectors, "
1793 					  "but only %d MSI-X vectors available, "
1794 					  "trying again\n", nreq, err);
1795 				nreq = err;
1796 				goto retry;
1797 			}
1798 			kfree(entries);
1799 			goto no_msi;
1800 		}
1801 
1802 		if (nreq <
1803 		    MSIX_LEGACY_SZ + dev->caps.num_ports * MIN_MSIX_P_PORT) {
1804 			/*Working in legacy mode , all EQ's shared*/
1805 			dev->caps.comp_pool           = 0;
1806 			dev->caps.num_comp_vectors = nreq - 1;
1807 		} else {
1808 			dev->caps.comp_pool           = nreq - MSIX_LEGACY_SZ;
1809 			dev->caps.num_comp_vectors = MSIX_LEGACY_SZ - 1;
1810 		}
1811 		for (i = 0; i < nreq; ++i)
1812 			priv->eq_table.eq[i].irq = entries[i].vector;
1813 
1814 		dev->flags |= MLX4_FLAG_MSI_X;
1815 
1816 		kfree(entries);
1817 		return;
1818 	}
1819 
1820 no_msi:
1821 	dev->caps.num_comp_vectors = 1;
1822 	dev->caps.comp_pool	   = 0;
1823 
1824 	for (i = 0; i < 2; ++i)
1825 		priv->eq_table.eq[i].irq = dev->pdev->irq;
1826 }
1827 
1828 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
1829 {
1830 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
1831 	int err = 0;
1832 
1833 	info->dev = dev;
1834 	info->port = port;
1835 	if (!mlx4_is_slave(dev)) {
1836 		mlx4_init_mac_table(dev, &info->mac_table);
1837 		mlx4_init_vlan_table(dev, &info->vlan_table);
1838 		info->base_qpn = mlx4_get_base_qpn(dev, port);
1839 	}
1840 
1841 	sprintf(info->dev_name, "mlx4_port%d", port);
1842 	info->port_attr.attr.name = info->dev_name;
1843 	if (mlx4_is_mfunc(dev))
1844 		info->port_attr.attr.mode = S_IRUGO;
1845 	else {
1846 		info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
1847 		info->port_attr.store     = set_port_type;
1848 	}
1849 	info->port_attr.show      = show_port_type;
1850 	sysfs_attr_init(&info->port_attr.attr);
1851 
1852 	err = device_create_file(&dev->pdev->dev, &info->port_attr);
1853 	if (err) {
1854 		mlx4_err(dev, "Failed to create file for port %d\n", port);
1855 		info->port = -1;
1856 	}
1857 
1858 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
1859 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
1860 	if (mlx4_is_mfunc(dev))
1861 		info->port_mtu_attr.attr.mode = S_IRUGO;
1862 	else {
1863 		info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
1864 		info->port_mtu_attr.store     = set_port_ib_mtu;
1865 	}
1866 	info->port_mtu_attr.show      = show_port_ib_mtu;
1867 	sysfs_attr_init(&info->port_mtu_attr.attr);
1868 
1869 	err = device_create_file(&dev->pdev->dev, &info->port_mtu_attr);
1870 	if (err) {
1871 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
1872 		device_remove_file(&info->dev->pdev->dev, &info->port_attr);
1873 		info->port = -1;
1874 	}
1875 
1876 	return err;
1877 }
1878 
1879 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
1880 {
1881 	if (info->port < 0)
1882 		return;
1883 
1884 	device_remove_file(&info->dev->pdev->dev, &info->port_attr);
1885 	device_remove_file(&info->dev->pdev->dev, &info->port_mtu_attr);
1886 }
1887 
1888 static int mlx4_init_steering(struct mlx4_dev *dev)
1889 {
1890 	struct mlx4_priv *priv = mlx4_priv(dev);
1891 	int num_entries = dev->caps.num_ports;
1892 	int i, j;
1893 
1894 	priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
1895 	if (!priv->steer)
1896 		return -ENOMEM;
1897 
1898 	for (i = 0; i < num_entries; i++)
1899 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
1900 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
1901 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
1902 		}
1903 	return 0;
1904 }
1905 
1906 static void mlx4_clear_steering(struct mlx4_dev *dev)
1907 {
1908 	struct mlx4_priv *priv = mlx4_priv(dev);
1909 	struct mlx4_steer_index *entry, *tmp_entry;
1910 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
1911 	int num_entries = dev->caps.num_ports;
1912 	int i, j;
1913 
1914 	for (i = 0; i < num_entries; i++) {
1915 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
1916 			list_for_each_entry_safe(pqp, tmp_pqp,
1917 						 &priv->steer[i].promisc_qps[j],
1918 						 list) {
1919 				list_del(&pqp->list);
1920 				kfree(pqp);
1921 			}
1922 			list_for_each_entry_safe(entry, tmp_entry,
1923 						 &priv->steer[i].steer_entries[j],
1924 						 list) {
1925 				list_del(&entry->list);
1926 				list_for_each_entry_safe(pqp, tmp_pqp,
1927 							 &entry->duplicates,
1928 							 list) {
1929 					list_del(&pqp->list);
1930 					kfree(pqp);
1931 				}
1932 				kfree(entry);
1933 			}
1934 		}
1935 	}
1936 	kfree(priv->steer);
1937 }
1938 
1939 static int extended_func_num(struct pci_dev *pdev)
1940 {
1941 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
1942 }
1943 
1944 #define MLX4_OWNER_BASE	0x8069c
1945 #define MLX4_OWNER_SIZE	4
1946 
1947 static int mlx4_get_ownership(struct mlx4_dev *dev)
1948 {
1949 	void __iomem *owner;
1950 	u32 ret;
1951 
1952 	if (pci_channel_offline(dev->pdev))
1953 		return -EIO;
1954 
1955 	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
1956 			MLX4_OWNER_SIZE);
1957 	if (!owner) {
1958 		mlx4_err(dev, "Failed to obtain ownership bit\n");
1959 		return -ENOMEM;
1960 	}
1961 
1962 	ret = readl(owner);
1963 	iounmap(owner);
1964 	return (int) !!ret;
1965 }
1966 
1967 static void mlx4_free_ownership(struct mlx4_dev *dev)
1968 {
1969 	void __iomem *owner;
1970 
1971 	if (pci_channel_offline(dev->pdev))
1972 		return;
1973 
1974 	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
1975 			MLX4_OWNER_SIZE);
1976 	if (!owner) {
1977 		mlx4_err(dev, "Failed to obtain ownership bit\n");
1978 		return;
1979 	}
1980 	writel(0, owner);
1981 	msleep(1000);
1982 	iounmap(owner);
1983 }
1984 
1985 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data)
1986 {
1987 	struct mlx4_priv *priv;
1988 	struct mlx4_dev *dev;
1989 	int err;
1990 	int port;
1991 
1992 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
1993 
1994 	err = pci_enable_device(pdev);
1995 	if (err) {
1996 		dev_err(&pdev->dev, "Cannot enable PCI device, "
1997 			"aborting.\n");
1998 		return err;
1999 	}
2000 	if (num_vfs > MLX4_MAX_NUM_VF) {
2001 		printk(KERN_ERR "There are more VF's (%d) than allowed(%d)\n",
2002 		       num_vfs, MLX4_MAX_NUM_VF);
2003 		return -EINVAL;
2004 	}
2005 	/*
2006 	 * Check for BARs.
2007 	 */
2008 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
2009 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
2010 		dev_err(&pdev->dev, "Missing DCS, aborting."
2011 			"(driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
2012 			pci_dev_data, pci_resource_flags(pdev, 0));
2013 		err = -ENODEV;
2014 		goto err_disable_pdev;
2015 	}
2016 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
2017 		dev_err(&pdev->dev, "Missing UAR, aborting.\n");
2018 		err = -ENODEV;
2019 		goto err_disable_pdev;
2020 	}
2021 
2022 	err = pci_request_regions(pdev, DRV_NAME);
2023 	if (err) {
2024 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
2025 		goto err_disable_pdev;
2026 	}
2027 
2028 	pci_set_master(pdev);
2029 
2030 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2031 	if (err) {
2032 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
2033 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2034 		if (err) {
2035 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
2036 			goto err_release_regions;
2037 		}
2038 	}
2039 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2040 	if (err) {
2041 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
2042 			 "consistent PCI DMA mask.\n");
2043 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2044 		if (err) {
2045 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
2046 				"aborting.\n");
2047 			goto err_release_regions;
2048 		}
2049 	}
2050 
2051 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
2052 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
2053 
2054 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
2055 	if (!priv) {
2056 		err = -ENOMEM;
2057 		goto err_release_regions;
2058 	}
2059 
2060 	dev       = &priv->dev;
2061 	dev->pdev = pdev;
2062 	INIT_LIST_HEAD(&priv->ctx_list);
2063 	spin_lock_init(&priv->ctx_lock);
2064 
2065 	mutex_init(&priv->port_mutex);
2066 
2067 	INIT_LIST_HEAD(&priv->pgdir_list);
2068 	mutex_init(&priv->pgdir_mutex);
2069 
2070 	INIT_LIST_HEAD(&priv->bf_list);
2071 	mutex_init(&priv->bf_mutex);
2072 
2073 	dev->rev_id = pdev->revision;
2074 	/* Detect if this device is a virtual function */
2075 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
2076 		/* When acting as pf, we normally skip vfs unless explicitly
2077 		 * requested to probe them. */
2078 		if (num_vfs && extended_func_num(pdev) > probe_vf) {
2079 			mlx4_warn(dev, "Skipping virtual function:%d\n",
2080 						extended_func_num(pdev));
2081 			err = -ENODEV;
2082 			goto err_free_dev;
2083 		}
2084 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
2085 		dev->flags |= MLX4_FLAG_SLAVE;
2086 	} else {
2087 		/* We reset the device and enable SRIOV only for physical
2088 		 * devices.  Try to claim ownership on the device;
2089 		 * if already taken, skip -- do not allow multiple PFs */
2090 		err = mlx4_get_ownership(dev);
2091 		if (err) {
2092 			if (err < 0)
2093 				goto err_free_dev;
2094 			else {
2095 				mlx4_warn(dev, "Multiple PFs not yet supported."
2096 					  " Skipping PF.\n");
2097 				err = -EINVAL;
2098 				goto err_free_dev;
2099 			}
2100 		}
2101 
2102 		if (num_vfs) {
2103 			mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", num_vfs);
2104 			err = pci_enable_sriov(pdev, num_vfs);
2105 			if (err) {
2106 				mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d).\n",
2107 					 err);
2108 				err = 0;
2109 			} else {
2110 				mlx4_warn(dev, "Running in master mode\n");
2111 				dev->flags |= MLX4_FLAG_SRIOV |
2112 					      MLX4_FLAG_MASTER;
2113 				dev->num_vfs = num_vfs;
2114 			}
2115 		}
2116 
2117 		/*
2118 		 * Now reset the HCA before we touch the PCI capabilities or
2119 		 * attempt a firmware command, since a boot ROM may have left
2120 		 * the HCA in an undefined state.
2121 		 */
2122 		err = mlx4_reset(dev);
2123 		if (err) {
2124 			mlx4_err(dev, "Failed to reset HCA, aborting.\n");
2125 			goto err_rel_own;
2126 		}
2127 	}
2128 
2129 slave_start:
2130 	err = mlx4_cmd_init(dev);
2131 	if (err) {
2132 		mlx4_err(dev, "Failed to init command interface, aborting.\n");
2133 		goto err_sriov;
2134 	}
2135 
2136 	/* In slave functions, the communication channel must be initialized
2137 	 * before posting commands. Also, init num_slaves before calling
2138 	 * mlx4_init_hca */
2139 	if (mlx4_is_mfunc(dev)) {
2140 		if (mlx4_is_master(dev))
2141 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
2142 		else {
2143 			dev->num_slaves = 0;
2144 			err = mlx4_multi_func_init(dev);
2145 			if (err) {
2146 				mlx4_err(dev, "Failed to init slave mfunc"
2147 					 " interface, aborting.\n");
2148 				goto err_cmd;
2149 			}
2150 		}
2151 	}
2152 
2153 	err = mlx4_init_hca(dev);
2154 	if (err) {
2155 		if (err == -EACCES) {
2156 			/* Not primary Physical function
2157 			 * Running in slave mode */
2158 			mlx4_cmd_cleanup(dev);
2159 			dev->flags |= MLX4_FLAG_SLAVE;
2160 			dev->flags &= ~MLX4_FLAG_MASTER;
2161 			goto slave_start;
2162 		} else
2163 			goto err_mfunc;
2164 	}
2165 
2166 	/* In master functions, the communication channel must be initialized
2167 	 * after obtaining its address from fw */
2168 	if (mlx4_is_master(dev)) {
2169 		err = mlx4_multi_func_init(dev);
2170 		if (err) {
2171 			mlx4_err(dev, "Failed to init master mfunc"
2172 				 "interface, aborting.\n");
2173 			goto err_close;
2174 		}
2175 	}
2176 
2177 	err = mlx4_alloc_eq_table(dev);
2178 	if (err)
2179 		goto err_master_mfunc;
2180 
2181 	priv->msix_ctl.pool_bm = 0;
2182 	mutex_init(&priv->msix_ctl.pool_lock);
2183 
2184 	mlx4_enable_msi_x(dev);
2185 	if ((mlx4_is_mfunc(dev)) &&
2186 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
2187 		err = -ENOSYS;
2188 		mlx4_err(dev, "INTx is not supported in multi-function mode."
2189 			 " aborting.\n");
2190 		goto err_free_eq;
2191 	}
2192 
2193 	if (!mlx4_is_slave(dev)) {
2194 		err = mlx4_init_steering(dev);
2195 		if (err)
2196 			goto err_free_eq;
2197 	}
2198 
2199 	err = mlx4_setup_hca(dev);
2200 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
2201 	    !mlx4_is_mfunc(dev)) {
2202 		dev->flags &= ~MLX4_FLAG_MSI_X;
2203 		dev->caps.num_comp_vectors = 1;
2204 		dev->caps.comp_pool	   = 0;
2205 		pci_disable_msix(pdev);
2206 		err = mlx4_setup_hca(dev);
2207 	}
2208 
2209 	if (err)
2210 		goto err_steer;
2211 
2212 	for (port = 1; port <= dev->caps.num_ports; port++) {
2213 		err = mlx4_init_port_info(dev, port);
2214 		if (err)
2215 			goto err_port;
2216 	}
2217 
2218 	err = mlx4_register_device(dev);
2219 	if (err)
2220 		goto err_port;
2221 
2222 	mlx4_sense_init(dev);
2223 	mlx4_start_sense(dev);
2224 
2225 	priv->pci_dev_data = pci_dev_data;
2226 	pci_set_drvdata(pdev, dev);
2227 
2228 	return 0;
2229 
2230 err_port:
2231 	for (--port; port >= 1; --port)
2232 		mlx4_cleanup_port_info(&priv->port[port]);
2233 
2234 	mlx4_cleanup_counters_table(dev);
2235 	mlx4_cleanup_mcg_table(dev);
2236 	mlx4_cleanup_qp_table(dev);
2237 	mlx4_cleanup_srq_table(dev);
2238 	mlx4_cleanup_cq_table(dev);
2239 	mlx4_cmd_use_polling(dev);
2240 	mlx4_cleanup_eq_table(dev);
2241 	mlx4_cleanup_mr_table(dev);
2242 	mlx4_cleanup_xrcd_table(dev);
2243 	mlx4_cleanup_pd_table(dev);
2244 	mlx4_cleanup_uar_table(dev);
2245 
2246 err_steer:
2247 	if (!mlx4_is_slave(dev))
2248 		mlx4_clear_steering(dev);
2249 
2250 err_free_eq:
2251 	mlx4_free_eq_table(dev);
2252 
2253 err_master_mfunc:
2254 	if (mlx4_is_master(dev))
2255 		mlx4_multi_func_cleanup(dev);
2256 
2257 err_close:
2258 	if (dev->flags & MLX4_FLAG_MSI_X)
2259 		pci_disable_msix(pdev);
2260 
2261 	mlx4_close_hca(dev);
2262 
2263 err_mfunc:
2264 	if (mlx4_is_slave(dev))
2265 		mlx4_multi_func_cleanup(dev);
2266 
2267 err_cmd:
2268 	mlx4_cmd_cleanup(dev);
2269 
2270 err_sriov:
2271 	if (dev->flags & MLX4_FLAG_SRIOV)
2272 		pci_disable_sriov(pdev);
2273 
2274 err_rel_own:
2275 	if (!mlx4_is_slave(dev))
2276 		mlx4_free_ownership(dev);
2277 
2278 err_free_dev:
2279 	kfree(priv);
2280 
2281 err_release_regions:
2282 	pci_release_regions(pdev);
2283 
2284 err_disable_pdev:
2285 	pci_disable_device(pdev);
2286 	pci_set_drvdata(pdev, NULL);
2287 	return err;
2288 }
2289 
2290 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
2291 {
2292 	printk_once(KERN_INFO "%s", mlx4_version);
2293 
2294 	return __mlx4_init_one(pdev, id->driver_data);
2295 }
2296 
2297 static void mlx4_remove_one(struct pci_dev *pdev)
2298 {
2299 	struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
2300 	struct mlx4_priv *priv = mlx4_priv(dev);
2301 	int p;
2302 
2303 	if (dev) {
2304 		/* in SRIOV it is not allowed to unload the pf's
2305 		 * driver while there are alive vf's */
2306 		if (mlx4_is_master(dev)) {
2307 			if (mlx4_how_many_lives_vf(dev))
2308 				printk(KERN_ERR "Removing PF when there are assigned VF's !!!\n");
2309 		}
2310 		mlx4_stop_sense(dev);
2311 		mlx4_unregister_device(dev);
2312 
2313 		for (p = 1; p <= dev->caps.num_ports; p++) {
2314 			mlx4_cleanup_port_info(&priv->port[p]);
2315 			mlx4_CLOSE_PORT(dev, p);
2316 		}
2317 
2318 		if (mlx4_is_master(dev))
2319 			mlx4_free_resource_tracker(dev,
2320 						   RES_TR_FREE_SLAVES_ONLY);
2321 
2322 		mlx4_cleanup_counters_table(dev);
2323 		mlx4_cleanup_mcg_table(dev);
2324 		mlx4_cleanup_qp_table(dev);
2325 		mlx4_cleanup_srq_table(dev);
2326 		mlx4_cleanup_cq_table(dev);
2327 		mlx4_cmd_use_polling(dev);
2328 		mlx4_cleanup_eq_table(dev);
2329 		mlx4_cleanup_mr_table(dev);
2330 		mlx4_cleanup_xrcd_table(dev);
2331 		mlx4_cleanup_pd_table(dev);
2332 
2333 		if (mlx4_is_master(dev))
2334 			mlx4_free_resource_tracker(dev,
2335 						   RES_TR_FREE_STRUCTS_ONLY);
2336 
2337 		iounmap(priv->kar);
2338 		mlx4_uar_free(dev, &priv->driver_uar);
2339 		mlx4_cleanup_uar_table(dev);
2340 		if (!mlx4_is_slave(dev))
2341 			mlx4_clear_steering(dev);
2342 		mlx4_free_eq_table(dev);
2343 		if (mlx4_is_master(dev))
2344 			mlx4_multi_func_cleanup(dev);
2345 		mlx4_close_hca(dev);
2346 		if (mlx4_is_slave(dev))
2347 			mlx4_multi_func_cleanup(dev);
2348 		mlx4_cmd_cleanup(dev);
2349 
2350 		if (dev->flags & MLX4_FLAG_MSI_X)
2351 			pci_disable_msix(pdev);
2352 		if (dev->flags & MLX4_FLAG_SRIOV) {
2353 			mlx4_warn(dev, "Disabling SR-IOV\n");
2354 			pci_disable_sriov(pdev);
2355 		}
2356 
2357 		if (!mlx4_is_slave(dev))
2358 			mlx4_free_ownership(dev);
2359 
2360 		kfree(dev->caps.qp0_tunnel);
2361 		kfree(dev->caps.qp0_proxy);
2362 		kfree(dev->caps.qp1_tunnel);
2363 		kfree(dev->caps.qp1_proxy);
2364 
2365 		kfree(priv);
2366 		pci_release_regions(pdev);
2367 		pci_disable_device(pdev);
2368 		pci_set_drvdata(pdev, NULL);
2369 	}
2370 }
2371 
2372 int mlx4_restart_one(struct pci_dev *pdev)
2373 {
2374 	struct mlx4_dev	 *dev  = pci_get_drvdata(pdev);
2375 	struct mlx4_priv *priv = mlx4_priv(dev);
2376 	int		  pci_dev_data;
2377 
2378 	pci_dev_data = priv->pci_dev_data;
2379 	mlx4_remove_one(pdev);
2380 	return __mlx4_init_one(pdev, pci_dev_data);
2381 }
2382 
2383 static DEFINE_PCI_DEVICE_TABLE(mlx4_pci_table) = {
2384 	/* MT25408 "Hermon" SDR */
2385 	{ PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2386 	/* MT25408 "Hermon" DDR */
2387 	{ PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2388 	/* MT25408 "Hermon" QDR */
2389 	{ PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2390 	/* MT25408 "Hermon" DDR PCIe gen2 */
2391 	{ PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2392 	/* MT25408 "Hermon" QDR PCIe gen2 */
2393 	{ PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2394 	/* MT25408 "Hermon" EN 10GigE */
2395 	{ PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2396 	/* MT25408 "Hermon" EN 10GigE PCIe gen2 */
2397 	{ PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2398 	/* MT25458 ConnectX EN 10GBASE-T 10GigE */
2399 	{ PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2400 	/* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
2401 	{ PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2402 	/* MT26468 ConnectX EN 10GigE PCIe gen2*/
2403 	{ PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2404 	/* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
2405 	{ PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2406 	/* MT26478 ConnectX2 40GigE PCIe gen2 */
2407 	{ PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2408 	/* MT25400 Family [ConnectX-2 Virtual Function] */
2409 	{ PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
2410 	/* MT27500 Family [ConnectX-3] */
2411 	{ PCI_VDEVICE(MELLANOX, 0x1003), 0 },
2412 	/* MT27500 Family [ConnectX-3 Virtual Function] */
2413 	{ PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
2414 	{ PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
2415 	{ PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
2416 	{ PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
2417 	{ PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
2418 	{ PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
2419 	{ PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
2420 	{ PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
2421 	{ PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
2422 	{ PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
2423 	{ PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
2424 	{ PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
2425 	{ PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
2426 	{ 0, }
2427 };
2428 
2429 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
2430 
2431 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
2432 					      pci_channel_state_t state)
2433 {
2434 	mlx4_remove_one(pdev);
2435 
2436 	return state == pci_channel_io_perm_failure ?
2437 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
2438 }
2439 
2440 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
2441 {
2442 	int ret = __mlx4_init_one(pdev, 0);
2443 
2444 	return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
2445 }
2446 
2447 static const struct pci_error_handlers mlx4_err_handler = {
2448 	.error_detected = mlx4_pci_err_detected,
2449 	.slot_reset     = mlx4_pci_slot_reset,
2450 };
2451 
2452 static struct pci_driver mlx4_driver = {
2453 	.name		= DRV_NAME,
2454 	.id_table	= mlx4_pci_table,
2455 	.probe		= mlx4_init_one,
2456 	.remove		= mlx4_remove_one,
2457 	.err_handler    = &mlx4_err_handler,
2458 };
2459 
2460 static int __init mlx4_verify_params(void)
2461 {
2462 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
2463 		pr_warning("mlx4_core: bad num_mac: %d\n", log_num_mac);
2464 		return -1;
2465 	}
2466 
2467 	if (log_num_vlan != 0)
2468 		pr_warning("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
2469 			   MLX4_LOG_NUM_VLANS);
2470 
2471 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
2472 		pr_warning("mlx4_core: bad log_mtts_per_seg: %d\n", log_mtts_per_seg);
2473 		return -1;
2474 	}
2475 
2476 	/* Check if module param for ports type has legal combination */
2477 	if (port_type_array[0] == false && port_type_array[1] == true) {
2478 		printk(KERN_WARNING "Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
2479 		port_type_array[0] = true;
2480 	}
2481 
2482 	if (mlx4_log_num_mgm_entry_size != -1 &&
2483 	    (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
2484 	     mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE)) {
2485 		pr_warning("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not "
2486 			   "in legal range (-1 or %d..%d)\n",
2487 			   mlx4_log_num_mgm_entry_size,
2488 			   MLX4_MIN_MGM_LOG_ENTRY_SIZE,
2489 			   MLX4_MAX_MGM_LOG_ENTRY_SIZE);
2490 		return -1;
2491 	}
2492 
2493 	return 0;
2494 }
2495 
2496 static int __init mlx4_init(void)
2497 {
2498 	int ret;
2499 
2500 	if (mlx4_verify_params())
2501 		return -EINVAL;
2502 
2503 	mlx4_catas_init();
2504 
2505 	mlx4_wq = create_singlethread_workqueue("mlx4");
2506 	if (!mlx4_wq)
2507 		return -ENOMEM;
2508 
2509 	ret = pci_register_driver(&mlx4_driver);
2510 	return ret < 0 ? ret : 0;
2511 }
2512 
2513 static void __exit mlx4_cleanup(void)
2514 {
2515 	pci_unregister_driver(&mlx4_driver);
2516 	destroy_workqueue(mlx4_wq);
2517 }
2518 
2519 module_init(mlx4_init);
2520 module_exit(mlx4_cleanup);
2521