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/kmod.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 uint8_t num_vfs[3] = {0, 0, 0};
81 static int num_vfs_argc;
82 module_param_array(num_vfs, byte , &num_vfs_argc, 0444);
83 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
84 			  "num_vfs=port1,port2,port1+2");
85 
86 static uint8_t probe_vf[3] = {0, 0, 0};
87 static int probe_vfs_argc;
88 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
89 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
90 			   "probe_vf=port1,port2,port1+2");
91 
92 int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
93 module_param_named(log_num_mgm_entry_size,
94 			mlx4_log_num_mgm_entry_size, int, 0444);
95 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
96 					 " of qp per mcg, for example:"
97 					 " 10 gives 248.range: 7 <="
98 					 " log_num_mgm_entry_size <= 12."
99 					 " To activate device managed"
100 					 " flow steering when available, set to -1");
101 
102 static bool enable_64b_cqe_eqe = true;
103 module_param(enable_64b_cqe_eqe, bool, 0444);
104 MODULE_PARM_DESC(enable_64b_cqe_eqe,
105 		 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
106 
107 #define PF_CONTEXT_BEHAVIOUR_MASK	(MLX4_FUNC_CAP_64B_EQE_CQE | \
108 					 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
109 					 MLX4_FUNC_CAP_DMFS_A0_STATIC)
110 
111 static char mlx4_version[] =
112 	DRV_NAME ": Mellanox ConnectX core driver v"
113 	DRV_VERSION " (" DRV_RELDATE ")\n";
114 
115 static struct mlx4_profile default_profile = {
116 	.num_qp		= 1 << 18,
117 	.num_srq	= 1 << 16,
118 	.rdmarc_per_qp	= 1 << 4,
119 	.num_cq		= 1 << 16,
120 	.num_mcg	= 1 << 13,
121 	.num_mpt	= 1 << 19,
122 	.num_mtt	= 1 << 20, /* It is really num mtt segements */
123 };
124 
125 static struct mlx4_profile low_mem_profile = {
126 	.num_qp		= 1 << 17,
127 	.num_srq	= 1 << 6,
128 	.rdmarc_per_qp	= 1 << 4,
129 	.num_cq		= 1 << 8,
130 	.num_mcg	= 1 << 8,
131 	.num_mpt	= 1 << 9,
132 	.num_mtt	= 1 << 7,
133 };
134 
135 static int log_num_mac = 7;
136 module_param_named(log_num_mac, log_num_mac, int, 0444);
137 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
138 
139 static int log_num_vlan;
140 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
141 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
142 /* Log2 max number of VLANs per ETH port (0-7) */
143 #define MLX4_LOG_NUM_VLANS 7
144 #define MLX4_MIN_LOG_NUM_VLANS 0
145 #define MLX4_MIN_LOG_NUM_MAC 1
146 
147 static bool use_prio;
148 module_param_named(use_prio, use_prio, bool, 0444);
149 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
150 
151 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
152 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
153 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
154 
155 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
156 static int arr_argc = 2;
157 module_param_array(port_type_array, int, &arr_argc, 0444);
158 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
159 				"1 for IB, 2 for Ethernet");
160 
161 struct mlx4_port_config {
162 	struct list_head list;
163 	enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
164 	struct pci_dev *pdev;
165 };
166 
167 static atomic_t pf_loading = ATOMIC_INIT(0);
168 
169 int mlx4_check_port_params(struct mlx4_dev *dev,
170 			   enum mlx4_port_type *port_type)
171 {
172 	int i;
173 
174 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
175 		for (i = 0; i < dev->caps.num_ports - 1; i++) {
176 			if (port_type[i] != port_type[i + 1]) {
177 				mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
178 				return -EINVAL;
179 			}
180 		}
181 	}
182 
183 	for (i = 0; i < dev->caps.num_ports; i++) {
184 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
185 			mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
186 				 i + 1);
187 			return -EINVAL;
188 		}
189 	}
190 	return 0;
191 }
192 
193 static void mlx4_set_port_mask(struct mlx4_dev *dev)
194 {
195 	int i;
196 
197 	for (i = 1; i <= dev->caps.num_ports; ++i)
198 		dev->caps.port_mask[i] = dev->caps.port_type[i];
199 }
200 
201 enum {
202 	MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
203 };
204 
205 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
206 {
207 	int err = 0;
208 	struct mlx4_func func;
209 
210 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
211 		err = mlx4_QUERY_FUNC(dev, &func, 0);
212 		if (err) {
213 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
214 			return err;
215 		}
216 		dev_cap->max_eqs = func.max_eq;
217 		dev_cap->reserved_eqs = func.rsvd_eqs;
218 		dev_cap->reserved_uars = func.rsvd_uars;
219 		err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
220 	}
221 	return err;
222 }
223 
224 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
225 {
226 	struct mlx4_caps *dev_cap = &dev->caps;
227 
228 	/* FW not supporting or cancelled by user */
229 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
230 	    !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
231 		return;
232 
233 	/* Must have 64B CQE_EQE enabled by FW to use bigger stride
234 	 * When FW has NCSI it may decide not to report 64B CQE/EQEs
235 	 */
236 	if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
237 	    !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
238 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
239 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
240 		return;
241 	}
242 
243 	if (cache_line_size() == 128 || cache_line_size() == 256) {
244 		mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
245 		/* Changing the real data inside CQE size to 32B */
246 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
247 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
248 
249 		if (mlx4_is_master(dev))
250 			dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
251 	} else {
252 		mlx4_dbg(dev, "Disabling CQE stride cacheLine unsupported\n");
253 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
254 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
255 	}
256 }
257 
258 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
259 			  struct mlx4_port_cap *port_cap)
260 {
261 	dev->caps.vl_cap[port]	    = port_cap->max_vl;
262 	dev->caps.ib_mtu_cap[port]	    = port_cap->ib_mtu;
263 	dev->phys_caps.gid_phys_table_len[port]  = port_cap->max_gids;
264 	dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
265 	/* set gid and pkey table operating lengths by default
266 	 * to non-sriov values
267 	 */
268 	dev->caps.gid_table_len[port]  = port_cap->max_gids;
269 	dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
270 	dev->caps.port_width_cap[port] = port_cap->max_port_width;
271 	dev->caps.eth_mtu_cap[port]    = port_cap->eth_mtu;
272 	dev->caps.def_mac[port]        = port_cap->def_mac;
273 	dev->caps.supported_type[port] = port_cap->supported_port_types;
274 	dev->caps.suggested_type[port] = port_cap->suggested_type;
275 	dev->caps.default_sense[port] = port_cap->default_sense;
276 	dev->caps.trans_type[port]	    = port_cap->trans_type;
277 	dev->caps.vendor_oui[port]     = port_cap->vendor_oui;
278 	dev->caps.wavelength[port]     = port_cap->wavelength;
279 	dev->caps.trans_code[port]     = port_cap->trans_code;
280 
281 	return 0;
282 }
283 
284 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
285 			 struct mlx4_port_cap *port_cap)
286 {
287 	int err = 0;
288 
289 	err = mlx4_QUERY_PORT(dev, port, port_cap);
290 
291 	if (err)
292 		mlx4_err(dev, "QUERY_PORT command failed.\n");
293 
294 	return err;
295 }
296 
297 #define MLX4_A0_STEERING_TABLE_SIZE	256
298 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
299 {
300 	int err;
301 	int i;
302 
303 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
304 	if (err) {
305 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
306 		return err;
307 	}
308 	mlx4_dev_cap_dump(dev, dev_cap);
309 
310 	if (dev_cap->min_page_sz > PAGE_SIZE) {
311 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
312 			 dev_cap->min_page_sz, PAGE_SIZE);
313 		return -ENODEV;
314 	}
315 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
316 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
317 			 dev_cap->num_ports, MLX4_MAX_PORTS);
318 		return -ENODEV;
319 	}
320 
321 	if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
322 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
323 			 dev_cap->uar_size,
324 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
325 		return -ENODEV;
326 	}
327 
328 	dev->caps.num_ports	     = dev_cap->num_ports;
329 	dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
330 	dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
331 				      dev->caps.num_sys_eqs :
332 				      MLX4_MAX_EQ_NUM;
333 	for (i = 1; i <= dev->caps.num_ports; ++i) {
334 		err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
335 		if (err) {
336 			mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
337 			return err;
338 		}
339 	}
340 
341 	dev->caps.uar_page_size	     = PAGE_SIZE;
342 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
343 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
344 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
345 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
346 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
347 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
348 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
349 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
350 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
351 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
352 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
353 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
354 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
355 	/*
356 	 * Subtract 1 from the limit because we need to allocate a
357 	 * spare CQE so the HCA HW can tell the difference between an
358 	 * empty CQ and a full CQ.
359 	 */
360 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
361 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
362 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
363 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
364 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
365 
366 	/* The first 128 UARs are used for EQ doorbells */
367 	dev->caps.reserved_uars	     = max_t(int, 128, dev_cap->reserved_uars);
368 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
369 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
370 					dev_cap->reserved_xrcds : 0;
371 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
372 					dev_cap->max_xrcds : 0;
373 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
374 
375 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
376 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
377 	dev->caps.flags		     = dev_cap->flags;
378 	dev->caps.flags2	     = dev_cap->flags2;
379 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
380 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
381 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
382 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
383 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
384 
385 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
386 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
387 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
388 	/* Don't do sense port on multifunction devices (for now at least) */
389 	if (mlx4_is_mfunc(dev))
390 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
391 
392 	if (mlx4_low_memory_profile()) {
393 		dev->caps.log_num_macs  = MLX4_MIN_LOG_NUM_MAC;
394 		dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
395 	} else {
396 		dev->caps.log_num_macs  = log_num_mac;
397 		dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
398 	}
399 
400 	for (i = 1; i <= dev->caps.num_ports; ++i) {
401 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
402 		if (dev->caps.supported_type[i]) {
403 			/* if only ETH is supported - assign ETH */
404 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
405 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
406 			/* if only IB is supported, assign IB */
407 			else if (dev->caps.supported_type[i] ==
408 				 MLX4_PORT_TYPE_IB)
409 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
410 			else {
411 				/* if IB and ETH are supported, we set the port
412 				 * type according to user selection of port type;
413 				 * if user selected none, take the FW hint */
414 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
415 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
416 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
417 				else
418 					dev->caps.port_type[i] = port_type_array[i - 1];
419 			}
420 		}
421 		/*
422 		 * Link sensing is allowed on the port if 3 conditions are true:
423 		 * 1. Both protocols are supported on the port.
424 		 * 2. Different types are supported on the port
425 		 * 3. FW declared that it supports link sensing
426 		 */
427 		mlx4_priv(dev)->sense.sense_allowed[i] =
428 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
429 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
430 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
431 
432 		/*
433 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
434 		 * and perform sense_port FW command to try and set the correct
435 		 * port type from beginning
436 		 */
437 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
438 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
439 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
440 			mlx4_SENSE_PORT(dev, i, &sensed_port);
441 			if (sensed_port != MLX4_PORT_TYPE_NONE)
442 				dev->caps.port_type[i] = sensed_port;
443 		} else {
444 			dev->caps.possible_type[i] = dev->caps.port_type[i];
445 		}
446 
447 		if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
448 			dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
449 			mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
450 				  i, 1 << dev->caps.log_num_macs);
451 		}
452 		if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
453 			dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
454 			mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
455 				  i, 1 << dev->caps.log_num_vlans);
456 		}
457 	}
458 
459 	dev->caps.max_counters = 1 << ilog2(dev_cap->max_counters);
460 
461 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
462 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
463 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
464 		(1 << dev->caps.log_num_macs) *
465 		(1 << dev->caps.log_num_vlans) *
466 		dev->caps.num_ports;
467 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
468 
469 	if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
470 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
471 		dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
472 	else
473 		dev->caps.dmfs_high_rate_qpn_base =
474 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
475 
476 	if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
477 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
478 		dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
479 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
480 		dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
481 	} else {
482 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
483 		dev->caps.dmfs_high_rate_qpn_base =
484 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
485 		dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
486 	}
487 
488 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
489 		dev->caps.dmfs_high_rate_qpn_range;
490 
491 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
492 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
493 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
494 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
495 
496 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
497 
498 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
499 		if (dev_cap->flags &
500 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
501 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
502 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
503 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
504 		}
505 
506 		if (dev_cap->flags2 &
507 		    (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
508 		     MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
509 			mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
510 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
511 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
512 		}
513 	}
514 
515 	if ((dev->caps.flags &
516 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
517 	    mlx4_is_master(dev))
518 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
519 
520 	if (!mlx4_is_slave(dev)) {
521 		mlx4_enable_cqe_eqe_stride(dev);
522 		dev->caps.alloc_res_qp_mask =
523 			(dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
524 			MLX4_RESERVE_A0_QP;
525 	} else {
526 		dev->caps.alloc_res_qp_mask = 0;
527 	}
528 
529 	return 0;
530 }
531 
532 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
533 				       enum pci_bus_speed *speed,
534 				       enum pcie_link_width *width)
535 {
536 	u32 lnkcap1, lnkcap2;
537 	int err1, err2;
538 
539 #define  PCIE_MLW_CAP_SHIFT 4	/* start of MLW mask in link capabilities */
540 
541 	*speed = PCI_SPEED_UNKNOWN;
542 	*width = PCIE_LNK_WIDTH_UNKNOWN;
543 
544 	err1 = pcie_capability_read_dword(dev->pdev, PCI_EXP_LNKCAP, &lnkcap1);
545 	err2 = pcie_capability_read_dword(dev->pdev, PCI_EXP_LNKCAP2, &lnkcap2);
546 	if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
547 		if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
548 			*speed = PCIE_SPEED_8_0GT;
549 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
550 			*speed = PCIE_SPEED_5_0GT;
551 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
552 			*speed = PCIE_SPEED_2_5GT;
553 	}
554 	if (!err1) {
555 		*width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
556 		if (!lnkcap2) { /* pre-r3.0 */
557 			if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
558 				*speed = PCIE_SPEED_5_0GT;
559 			else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
560 				*speed = PCIE_SPEED_2_5GT;
561 		}
562 	}
563 
564 	if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
565 		return err1 ? err1 :
566 			err2 ? err2 : -EINVAL;
567 	}
568 	return 0;
569 }
570 
571 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
572 {
573 	enum pcie_link_width width, width_cap;
574 	enum pci_bus_speed speed, speed_cap;
575 	int err;
576 
577 #define PCIE_SPEED_STR(speed) \
578 	(speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
579 	 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
580 	 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
581 	 "Unknown")
582 
583 	err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
584 	if (err) {
585 		mlx4_warn(dev,
586 			  "Unable to determine PCIe device BW capabilities\n");
587 		return;
588 	}
589 
590 	err = pcie_get_minimum_link(dev->pdev, &speed, &width);
591 	if (err || speed == PCI_SPEED_UNKNOWN ||
592 	    width == PCIE_LNK_WIDTH_UNKNOWN) {
593 		mlx4_warn(dev,
594 			  "Unable to determine PCI device chain minimum BW\n");
595 		return;
596 	}
597 
598 	if (width != width_cap || speed != speed_cap)
599 		mlx4_warn(dev,
600 			  "PCIe BW is different than device's capability\n");
601 
602 	mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
603 		  PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
604 	mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
605 		  width, width_cap);
606 	return;
607 }
608 
609 /*The function checks if there are live vf, return the num of them*/
610 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
611 {
612 	struct mlx4_priv *priv = mlx4_priv(dev);
613 	struct mlx4_slave_state *s_state;
614 	int i;
615 	int ret = 0;
616 
617 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
618 		s_state = &priv->mfunc.master.slave_state[i];
619 		if (s_state->active && s_state->last_cmd !=
620 		    MLX4_COMM_CMD_RESET) {
621 			mlx4_warn(dev, "%s: slave: %d is still active\n",
622 				  __func__, i);
623 			ret++;
624 		}
625 	}
626 	return ret;
627 }
628 
629 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
630 {
631 	u32 qk = MLX4_RESERVED_QKEY_BASE;
632 
633 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
634 	    qpn < dev->phys_caps.base_proxy_sqpn)
635 		return -EINVAL;
636 
637 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
638 		/* tunnel qp */
639 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
640 	else
641 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
642 	*qkey = qk;
643 	return 0;
644 }
645 EXPORT_SYMBOL(mlx4_get_parav_qkey);
646 
647 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
648 {
649 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
650 
651 	if (!mlx4_is_master(dev))
652 		return;
653 
654 	priv->virt2phys_pkey[slave][port - 1][i] = val;
655 }
656 EXPORT_SYMBOL(mlx4_sync_pkey_table);
657 
658 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
659 {
660 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
661 
662 	if (!mlx4_is_master(dev))
663 		return;
664 
665 	priv->slave_node_guids[slave] = guid;
666 }
667 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
668 
669 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
670 {
671 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
672 
673 	if (!mlx4_is_master(dev))
674 		return 0;
675 
676 	return priv->slave_node_guids[slave];
677 }
678 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
679 
680 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
681 {
682 	struct mlx4_priv *priv = mlx4_priv(dev);
683 	struct mlx4_slave_state *s_slave;
684 
685 	if (!mlx4_is_master(dev))
686 		return 0;
687 
688 	s_slave = &priv->mfunc.master.slave_state[slave];
689 	return !!s_slave->active;
690 }
691 EXPORT_SYMBOL(mlx4_is_slave_active);
692 
693 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
694 				       struct mlx4_dev_cap *dev_cap,
695 				       struct mlx4_init_hca_param *hca_param)
696 {
697 	dev->caps.steering_mode = hca_param->steering_mode;
698 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
699 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
700 		dev->caps.fs_log_max_ucast_qp_range_size =
701 			dev_cap->fs_log_max_ucast_qp_range_size;
702 	} else
703 		dev->caps.num_qp_per_mgm =
704 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
705 
706 	mlx4_dbg(dev, "Steering mode is: %s\n",
707 		 mlx4_steering_mode_str(dev->caps.steering_mode));
708 }
709 
710 static int mlx4_slave_cap(struct mlx4_dev *dev)
711 {
712 	int			   err;
713 	u32			   page_size;
714 	struct mlx4_dev_cap	   dev_cap;
715 	struct mlx4_func_cap	   func_cap;
716 	struct mlx4_init_hca_param hca_param;
717 	u8			   i;
718 
719 	memset(&hca_param, 0, sizeof(hca_param));
720 	err = mlx4_QUERY_HCA(dev, &hca_param);
721 	if (err) {
722 		mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
723 		return err;
724 	}
725 
726 	/* fail if the hca has an unknown global capability
727 	 * at this time global_caps should be always zeroed
728 	 */
729 	if (hca_param.global_caps) {
730 		mlx4_err(dev, "Unknown hca global capabilities\n");
731 		return -ENOSYS;
732 	}
733 
734 	mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
735 
736 	dev->caps.hca_core_clock = hca_param.hca_core_clock;
737 
738 	memset(&dev_cap, 0, sizeof(dev_cap));
739 	dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
740 	err = mlx4_dev_cap(dev, &dev_cap);
741 	if (err) {
742 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
743 		return err;
744 	}
745 
746 	err = mlx4_QUERY_FW(dev);
747 	if (err)
748 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
749 
750 	page_size = ~dev->caps.page_size_cap + 1;
751 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
752 	if (page_size > PAGE_SIZE) {
753 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
754 			 page_size, PAGE_SIZE);
755 		return -ENODEV;
756 	}
757 
758 	/* slave gets uar page size from QUERY_HCA fw command */
759 	dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
760 
761 	/* TODO: relax this assumption */
762 	if (dev->caps.uar_page_size != PAGE_SIZE) {
763 		mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
764 			 dev->caps.uar_page_size, PAGE_SIZE);
765 		return -ENODEV;
766 	}
767 
768 	memset(&func_cap, 0, sizeof(func_cap));
769 	err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
770 	if (err) {
771 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
772 			 err);
773 		return err;
774 	}
775 
776 	if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
777 	    PF_CONTEXT_BEHAVIOUR_MASK) {
778 		mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
779 			 func_cap.pf_context_behaviour, PF_CONTEXT_BEHAVIOUR_MASK);
780 		return -ENOSYS;
781 	}
782 
783 	dev->caps.num_ports		= func_cap.num_ports;
784 	dev->quotas.qp			= func_cap.qp_quota;
785 	dev->quotas.srq			= func_cap.srq_quota;
786 	dev->quotas.cq			= func_cap.cq_quota;
787 	dev->quotas.mpt			= func_cap.mpt_quota;
788 	dev->quotas.mtt			= func_cap.mtt_quota;
789 	dev->caps.num_qps		= 1 << hca_param.log_num_qps;
790 	dev->caps.num_srqs		= 1 << hca_param.log_num_srqs;
791 	dev->caps.num_cqs		= 1 << hca_param.log_num_cqs;
792 	dev->caps.num_mpts		= 1 << hca_param.log_mpt_sz;
793 	dev->caps.num_eqs		= func_cap.max_eq;
794 	dev->caps.reserved_eqs		= func_cap.reserved_eq;
795 	dev->caps.num_pds               = MLX4_NUM_PDS;
796 	dev->caps.num_mgms              = 0;
797 	dev->caps.num_amgms             = 0;
798 
799 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
800 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
801 			 dev->caps.num_ports, MLX4_MAX_PORTS);
802 		return -ENODEV;
803 	}
804 
805 	dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
806 	dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
807 	dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
808 	dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
809 	dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
810 
811 	if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
812 	    !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
813 	    !dev->caps.qp0_qkey) {
814 		err = -ENOMEM;
815 		goto err_mem;
816 	}
817 
818 	for (i = 1; i <= dev->caps.num_ports; ++i) {
819 		err = mlx4_QUERY_FUNC_CAP(dev, i, &func_cap);
820 		if (err) {
821 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
822 				 i, err);
823 			goto err_mem;
824 		}
825 		dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
826 		dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
827 		dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
828 		dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
829 		dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
830 		dev->caps.port_mask[i] = dev->caps.port_type[i];
831 		dev->caps.phys_port_id[i] = func_cap.phys_port_id;
832 		if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
833 						    &dev->caps.gid_table_len[i],
834 						    &dev->caps.pkey_table_len[i]))
835 			goto err_mem;
836 	}
837 
838 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
839 				       dev->caps.reserved_uars) >
840 				       pci_resource_len(dev->pdev, 2)) {
841 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
842 			 dev->caps.uar_page_size * dev->caps.num_uars,
843 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
844 		goto err_mem;
845 	}
846 
847 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
848 		dev->caps.eqe_size   = 64;
849 		dev->caps.eqe_factor = 1;
850 	} else {
851 		dev->caps.eqe_size   = 32;
852 		dev->caps.eqe_factor = 0;
853 	}
854 
855 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
856 		dev->caps.cqe_size   = 64;
857 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
858 	} else {
859 		dev->caps.cqe_size   = 32;
860 	}
861 
862 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
863 		dev->caps.eqe_size = hca_param.eqe_size;
864 		dev->caps.eqe_factor = 0;
865 	}
866 
867 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
868 		dev->caps.cqe_size = hca_param.cqe_size;
869 		/* User still need to know when CQE > 32B */
870 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
871 	}
872 
873 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
874 	mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
875 
876 	slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
877 
878 	if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
879 	    dev->caps.bf_reg_size)
880 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
881 
882 	if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
883 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
884 
885 	return 0;
886 
887 err_mem:
888 	kfree(dev->caps.qp0_qkey);
889 	kfree(dev->caps.qp0_tunnel);
890 	kfree(dev->caps.qp0_proxy);
891 	kfree(dev->caps.qp1_tunnel);
892 	kfree(dev->caps.qp1_proxy);
893 	dev->caps.qp0_qkey = NULL;
894 	dev->caps.qp0_tunnel = NULL;
895 	dev->caps.qp0_proxy = NULL;
896 	dev->caps.qp1_tunnel = NULL;
897 	dev->caps.qp1_proxy = NULL;
898 
899 	return err;
900 }
901 
902 static void mlx4_request_modules(struct mlx4_dev *dev)
903 {
904 	int port;
905 	int has_ib_port = false;
906 	int has_eth_port = false;
907 #define EN_DRV_NAME	"mlx4_en"
908 #define IB_DRV_NAME	"mlx4_ib"
909 
910 	for (port = 1; port <= dev->caps.num_ports; port++) {
911 		if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
912 			has_ib_port = true;
913 		else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
914 			has_eth_port = true;
915 	}
916 
917 	if (has_eth_port)
918 		request_module_nowait(EN_DRV_NAME);
919 	if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
920 		request_module_nowait(IB_DRV_NAME);
921 }
922 
923 /*
924  * Change the port configuration of the device.
925  * Every user of this function must hold the port mutex.
926  */
927 int mlx4_change_port_types(struct mlx4_dev *dev,
928 			   enum mlx4_port_type *port_types)
929 {
930 	int err = 0;
931 	int change = 0;
932 	int port;
933 
934 	for (port = 0; port <  dev->caps.num_ports; port++) {
935 		/* Change the port type only if the new type is different
936 		 * from the current, and not set to Auto */
937 		if (port_types[port] != dev->caps.port_type[port + 1])
938 			change = 1;
939 	}
940 	if (change) {
941 		mlx4_unregister_device(dev);
942 		for (port = 1; port <= dev->caps.num_ports; port++) {
943 			mlx4_CLOSE_PORT(dev, port);
944 			dev->caps.port_type[port] = port_types[port - 1];
945 			err = mlx4_SET_PORT(dev, port, -1);
946 			if (err) {
947 				mlx4_err(dev, "Failed to set port %d, aborting\n",
948 					 port);
949 				goto out;
950 			}
951 		}
952 		mlx4_set_port_mask(dev);
953 		err = mlx4_register_device(dev);
954 		if (err) {
955 			mlx4_err(dev, "Failed to register device\n");
956 			goto out;
957 		}
958 		mlx4_request_modules(dev);
959 	}
960 
961 out:
962 	return err;
963 }
964 
965 static ssize_t show_port_type(struct device *dev,
966 			      struct device_attribute *attr,
967 			      char *buf)
968 {
969 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
970 						   port_attr);
971 	struct mlx4_dev *mdev = info->dev;
972 	char type[8];
973 
974 	sprintf(type, "%s",
975 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
976 		"ib" : "eth");
977 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
978 		sprintf(buf, "auto (%s)\n", type);
979 	else
980 		sprintf(buf, "%s\n", type);
981 
982 	return strlen(buf);
983 }
984 
985 static ssize_t set_port_type(struct device *dev,
986 			     struct device_attribute *attr,
987 			     const char *buf, size_t count)
988 {
989 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
990 						   port_attr);
991 	struct mlx4_dev *mdev = info->dev;
992 	struct mlx4_priv *priv = mlx4_priv(mdev);
993 	enum mlx4_port_type types[MLX4_MAX_PORTS];
994 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
995 	static DEFINE_MUTEX(set_port_type_mutex);
996 	int i;
997 	int err = 0;
998 
999 	mutex_lock(&set_port_type_mutex);
1000 
1001 	if (!strcmp(buf, "ib\n"))
1002 		info->tmp_type = MLX4_PORT_TYPE_IB;
1003 	else if (!strcmp(buf, "eth\n"))
1004 		info->tmp_type = MLX4_PORT_TYPE_ETH;
1005 	else if (!strcmp(buf, "auto\n"))
1006 		info->tmp_type = MLX4_PORT_TYPE_AUTO;
1007 	else {
1008 		mlx4_err(mdev, "%s is not supported port type\n", buf);
1009 		err = -EINVAL;
1010 		goto err_out;
1011 	}
1012 
1013 	mlx4_stop_sense(mdev);
1014 	mutex_lock(&priv->port_mutex);
1015 	/* Possible type is always the one that was delivered */
1016 	mdev->caps.possible_type[info->port] = info->tmp_type;
1017 
1018 	for (i = 0; i < mdev->caps.num_ports; i++) {
1019 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1020 					mdev->caps.possible_type[i+1];
1021 		if (types[i] == MLX4_PORT_TYPE_AUTO)
1022 			types[i] = mdev->caps.port_type[i+1];
1023 	}
1024 
1025 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1026 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1027 		for (i = 1; i <= mdev->caps.num_ports; i++) {
1028 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1029 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1030 				err = -EINVAL;
1031 			}
1032 		}
1033 	}
1034 	if (err) {
1035 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1036 		goto out;
1037 	}
1038 
1039 	mlx4_do_sense_ports(mdev, new_types, types);
1040 
1041 	err = mlx4_check_port_params(mdev, new_types);
1042 	if (err)
1043 		goto out;
1044 
1045 	/* We are about to apply the changes after the configuration
1046 	 * was verified, no need to remember the temporary types
1047 	 * any more */
1048 	for (i = 0; i < mdev->caps.num_ports; i++)
1049 		priv->port[i + 1].tmp_type = 0;
1050 
1051 	err = mlx4_change_port_types(mdev, new_types);
1052 
1053 out:
1054 	mlx4_start_sense(mdev);
1055 	mutex_unlock(&priv->port_mutex);
1056 err_out:
1057 	mutex_unlock(&set_port_type_mutex);
1058 
1059 	return err ? err : count;
1060 }
1061 
1062 enum ibta_mtu {
1063 	IB_MTU_256  = 1,
1064 	IB_MTU_512  = 2,
1065 	IB_MTU_1024 = 3,
1066 	IB_MTU_2048 = 4,
1067 	IB_MTU_4096 = 5
1068 };
1069 
1070 static inline int int_to_ibta_mtu(int mtu)
1071 {
1072 	switch (mtu) {
1073 	case 256:  return IB_MTU_256;
1074 	case 512:  return IB_MTU_512;
1075 	case 1024: return IB_MTU_1024;
1076 	case 2048: return IB_MTU_2048;
1077 	case 4096: return IB_MTU_4096;
1078 	default: return -1;
1079 	}
1080 }
1081 
1082 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1083 {
1084 	switch (mtu) {
1085 	case IB_MTU_256:  return  256;
1086 	case IB_MTU_512:  return  512;
1087 	case IB_MTU_1024: return 1024;
1088 	case IB_MTU_2048: return 2048;
1089 	case IB_MTU_4096: return 4096;
1090 	default: return -1;
1091 	}
1092 }
1093 
1094 static ssize_t show_port_ib_mtu(struct device *dev,
1095 			     struct device_attribute *attr,
1096 			     char *buf)
1097 {
1098 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1099 						   port_mtu_attr);
1100 	struct mlx4_dev *mdev = info->dev;
1101 
1102 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1103 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1104 
1105 	sprintf(buf, "%d\n",
1106 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1107 	return strlen(buf);
1108 }
1109 
1110 static ssize_t set_port_ib_mtu(struct device *dev,
1111 			     struct device_attribute *attr,
1112 			     const char *buf, size_t count)
1113 {
1114 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1115 						   port_mtu_attr);
1116 	struct mlx4_dev *mdev = info->dev;
1117 	struct mlx4_priv *priv = mlx4_priv(mdev);
1118 	int err, port, mtu, ibta_mtu = -1;
1119 
1120 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1121 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1122 		return -EINVAL;
1123 	}
1124 
1125 	err = kstrtoint(buf, 0, &mtu);
1126 	if (!err)
1127 		ibta_mtu = int_to_ibta_mtu(mtu);
1128 
1129 	if (err || ibta_mtu < 0) {
1130 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1131 		return -EINVAL;
1132 	}
1133 
1134 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1135 
1136 	mlx4_stop_sense(mdev);
1137 	mutex_lock(&priv->port_mutex);
1138 	mlx4_unregister_device(mdev);
1139 	for (port = 1; port <= mdev->caps.num_ports; port++) {
1140 		mlx4_CLOSE_PORT(mdev, port);
1141 		err = mlx4_SET_PORT(mdev, port, -1);
1142 		if (err) {
1143 			mlx4_err(mdev, "Failed to set port %d, aborting\n",
1144 				 port);
1145 			goto err_set_port;
1146 		}
1147 	}
1148 	err = mlx4_register_device(mdev);
1149 err_set_port:
1150 	mutex_unlock(&priv->port_mutex);
1151 	mlx4_start_sense(mdev);
1152 	return err ? err : count;
1153 }
1154 
1155 static int mlx4_load_fw(struct mlx4_dev *dev)
1156 {
1157 	struct mlx4_priv *priv = mlx4_priv(dev);
1158 	int err;
1159 
1160 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1161 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
1162 	if (!priv->fw.fw_icm) {
1163 		mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1164 		return -ENOMEM;
1165 	}
1166 
1167 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1168 	if (err) {
1169 		mlx4_err(dev, "MAP_FA command failed, aborting\n");
1170 		goto err_free;
1171 	}
1172 
1173 	err = mlx4_RUN_FW(dev);
1174 	if (err) {
1175 		mlx4_err(dev, "RUN_FW command failed, aborting\n");
1176 		goto err_unmap_fa;
1177 	}
1178 
1179 	return 0;
1180 
1181 err_unmap_fa:
1182 	mlx4_UNMAP_FA(dev);
1183 
1184 err_free:
1185 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1186 	return err;
1187 }
1188 
1189 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1190 				int cmpt_entry_sz)
1191 {
1192 	struct mlx4_priv *priv = mlx4_priv(dev);
1193 	int err;
1194 	int num_eqs;
1195 
1196 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1197 				  cmpt_base +
1198 				  ((u64) (MLX4_CMPT_TYPE_QP *
1199 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1200 				  cmpt_entry_sz, dev->caps.num_qps,
1201 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1202 				  0, 0);
1203 	if (err)
1204 		goto err;
1205 
1206 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1207 				  cmpt_base +
1208 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1209 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1210 				  cmpt_entry_sz, dev->caps.num_srqs,
1211 				  dev->caps.reserved_srqs, 0, 0);
1212 	if (err)
1213 		goto err_qp;
1214 
1215 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1216 				  cmpt_base +
1217 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1218 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1219 				  cmpt_entry_sz, dev->caps.num_cqs,
1220 				  dev->caps.reserved_cqs, 0, 0);
1221 	if (err)
1222 		goto err_srq;
1223 
1224 	num_eqs = dev->phys_caps.num_phys_eqs;
1225 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1226 				  cmpt_base +
1227 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1228 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1229 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1230 	if (err)
1231 		goto err_cq;
1232 
1233 	return 0;
1234 
1235 err_cq:
1236 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1237 
1238 err_srq:
1239 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1240 
1241 err_qp:
1242 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1243 
1244 err:
1245 	return err;
1246 }
1247 
1248 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1249 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1250 {
1251 	struct mlx4_priv *priv = mlx4_priv(dev);
1252 	u64 aux_pages;
1253 	int num_eqs;
1254 	int err;
1255 
1256 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1257 	if (err) {
1258 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1259 		return err;
1260 	}
1261 
1262 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1263 		 (unsigned long long) icm_size >> 10,
1264 		 (unsigned long long) aux_pages << 2);
1265 
1266 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1267 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1268 	if (!priv->fw.aux_icm) {
1269 		mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1270 		return -ENOMEM;
1271 	}
1272 
1273 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1274 	if (err) {
1275 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1276 		goto err_free_aux;
1277 	}
1278 
1279 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1280 	if (err) {
1281 		mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1282 		goto err_unmap_aux;
1283 	}
1284 
1285 
1286 	num_eqs = dev->phys_caps.num_phys_eqs;
1287 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1288 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1289 				  num_eqs, num_eqs, 0, 0);
1290 	if (err) {
1291 		mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1292 		goto err_unmap_cmpt;
1293 	}
1294 
1295 	/*
1296 	 * Reserved MTT entries must be aligned up to a cacheline
1297 	 * boundary, since the FW will write to them, while the driver
1298 	 * writes to all other MTT entries. (The variable
1299 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1300 	 * size, not the raw entry size)
1301 	 */
1302 	dev->caps.reserved_mtts =
1303 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1304 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1305 
1306 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1307 				  init_hca->mtt_base,
1308 				  dev->caps.mtt_entry_sz,
1309 				  dev->caps.num_mtts,
1310 				  dev->caps.reserved_mtts, 1, 0);
1311 	if (err) {
1312 		mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1313 		goto err_unmap_eq;
1314 	}
1315 
1316 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1317 				  init_hca->dmpt_base,
1318 				  dev_cap->dmpt_entry_sz,
1319 				  dev->caps.num_mpts,
1320 				  dev->caps.reserved_mrws, 1, 1);
1321 	if (err) {
1322 		mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1323 		goto err_unmap_mtt;
1324 	}
1325 
1326 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1327 				  init_hca->qpc_base,
1328 				  dev_cap->qpc_entry_sz,
1329 				  dev->caps.num_qps,
1330 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1331 				  0, 0);
1332 	if (err) {
1333 		mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1334 		goto err_unmap_dmpt;
1335 	}
1336 
1337 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1338 				  init_hca->auxc_base,
1339 				  dev_cap->aux_entry_sz,
1340 				  dev->caps.num_qps,
1341 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1342 				  0, 0);
1343 	if (err) {
1344 		mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1345 		goto err_unmap_qp;
1346 	}
1347 
1348 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1349 				  init_hca->altc_base,
1350 				  dev_cap->altc_entry_sz,
1351 				  dev->caps.num_qps,
1352 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1353 				  0, 0);
1354 	if (err) {
1355 		mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1356 		goto err_unmap_auxc;
1357 	}
1358 
1359 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1360 				  init_hca->rdmarc_base,
1361 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1362 				  dev->caps.num_qps,
1363 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1364 				  0, 0);
1365 	if (err) {
1366 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1367 		goto err_unmap_altc;
1368 	}
1369 
1370 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1371 				  init_hca->cqc_base,
1372 				  dev_cap->cqc_entry_sz,
1373 				  dev->caps.num_cqs,
1374 				  dev->caps.reserved_cqs, 0, 0);
1375 	if (err) {
1376 		mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1377 		goto err_unmap_rdmarc;
1378 	}
1379 
1380 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1381 				  init_hca->srqc_base,
1382 				  dev_cap->srq_entry_sz,
1383 				  dev->caps.num_srqs,
1384 				  dev->caps.reserved_srqs, 0, 0);
1385 	if (err) {
1386 		mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1387 		goto err_unmap_cq;
1388 	}
1389 
1390 	/*
1391 	 * For flow steering device managed mode it is required to use
1392 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1393 	 * required, but for simplicity just map the whole multicast
1394 	 * group table now.  The table isn't very big and it's a lot
1395 	 * easier than trying to track ref counts.
1396 	 */
1397 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1398 				  init_hca->mc_base,
1399 				  mlx4_get_mgm_entry_size(dev),
1400 				  dev->caps.num_mgms + dev->caps.num_amgms,
1401 				  dev->caps.num_mgms + dev->caps.num_amgms,
1402 				  0, 0);
1403 	if (err) {
1404 		mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1405 		goto err_unmap_srq;
1406 	}
1407 
1408 	return 0;
1409 
1410 err_unmap_srq:
1411 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1412 
1413 err_unmap_cq:
1414 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1415 
1416 err_unmap_rdmarc:
1417 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1418 
1419 err_unmap_altc:
1420 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1421 
1422 err_unmap_auxc:
1423 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1424 
1425 err_unmap_qp:
1426 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1427 
1428 err_unmap_dmpt:
1429 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1430 
1431 err_unmap_mtt:
1432 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1433 
1434 err_unmap_eq:
1435 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1436 
1437 err_unmap_cmpt:
1438 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1439 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1440 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1441 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1442 
1443 err_unmap_aux:
1444 	mlx4_UNMAP_ICM_AUX(dev);
1445 
1446 err_free_aux:
1447 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1448 
1449 	return err;
1450 }
1451 
1452 static void mlx4_free_icms(struct mlx4_dev *dev)
1453 {
1454 	struct mlx4_priv *priv = mlx4_priv(dev);
1455 
1456 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1457 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1458 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1459 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1460 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1461 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1462 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1463 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1464 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1465 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1466 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1467 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1468 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1469 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1470 
1471 	mlx4_UNMAP_ICM_AUX(dev);
1472 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1473 }
1474 
1475 static void mlx4_slave_exit(struct mlx4_dev *dev)
1476 {
1477 	struct mlx4_priv *priv = mlx4_priv(dev);
1478 
1479 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1480 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_TIME))
1481 		mlx4_warn(dev, "Failed to close slave function\n");
1482 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1483 }
1484 
1485 static int map_bf_area(struct mlx4_dev *dev)
1486 {
1487 	struct mlx4_priv *priv = mlx4_priv(dev);
1488 	resource_size_t bf_start;
1489 	resource_size_t bf_len;
1490 	int err = 0;
1491 
1492 	if (!dev->caps.bf_reg_size)
1493 		return -ENXIO;
1494 
1495 	bf_start = pci_resource_start(dev->pdev, 2) +
1496 			(dev->caps.num_uars << PAGE_SHIFT);
1497 	bf_len = pci_resource_len(dev->pdev, 2) -
1498 			(dev->caps.num_uars << PAGE_SHIFT);
1499 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1500 	if (!priv->bf_mapping)
1501 		err = -ENOMEM;
1502 
1503 	return err;
1504 }
1505 
1506 static void unmap_bf_area(struct mlx4_dev *dev)
1507 {
1508 	if (mlx4_priv(dev)->bf_mapping)
1509 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1510 }
1511 
1512 cycle_t mlx4_read_clock(struct mlx4_dev *dev)
1513 {
1514 	u32 clockhi, clocklo, clockhi1;
1515 	cycle_t cycles;
1516 	int i;
1517 	struct mlx4_priv *priv = mlx4_priv(dev);
1518 
1519 	for (i = 0; i < 10; i++) {
1520 		clockhi = swab32(readl(priv->clock_mapping));
1521 		clocklo = swab32(readl(priv->clock_mapping + 4));
1522 		clockhi1 = swab32(readl(priv->clock_mapping));
1523 		if (clockhi == clockhi1)
1524 			break;
1525 	}
1526 
1527 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1528 
1529 	return cycles;
1530 }
1531 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1532 
1533 
1534 static int map_internal_clock(struct mlx4_dev *dev)
1535 {
1536 	struct mlx4_priv *priv = mlx4_priv(dev);
1537 
1538 	priv->clock_mapping =
1539 		ioremap(pci_resource_start(dev->pdev, priv->fw.clock_bar) +
1540 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1541 
1542 	if (!priv->clock_mapping)
1543 		return -ENOMEM;
1544 
1545 	return 0;
1546 }
1547 
1548 static void unmap_internal_clock(struct mlx4_dev *dev)
1549 {
1550 	struct mlx4_priv *priv = mlx4_priv(dev);
1551 
1552 	if (priv->clock_mapping)
1553 		iounmap(priv->clock_mapping);
1554 }
1555 
1556 static void mlx4_close_hca(struct mlx4_dev *dev)
1557 {
1558 	unmap_internal_clock(dev);
1559 	unmap_bf_area(dev);
1560 	if (mlx4_is_slave(dev))
1561 		mlx4_slave_exit(dev);
1562 	else {
1563 		mlx4_CLOSE_HCA(dev, 0);
1564 		mlx4_free_icms(dev);
1565 	}
1566 }
1567 
1568 static void mlx4_close_fw(struct mlx4_dev *dev)
1569 {
1570 	if (!mlx4_is_slave(dev)) {
1571 		mlx4_UNMAP_FA(dev);
1572 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1573 	}
1574 }
1575 
1576 static int mlx4_init_slave(struct mlx4_dev *dev)
1577 {
1578 	struct mlx4_priv *priv = mlx4_priv(dev);
1579 	u64 dma = (u64) priv->mfunc.vhcr_dma;
1580 	int ret_from_reset = 0;
1581 	u32 slave_read;
1582 	u32 cmd_channel_ver;
1583 
1584 	if (atomic_read(&pf_loading)) {
1585 		mlx4_warn(dev, "PF is not ready - Deferring probe\n");
1586 		return -EPROBE_DEFER;
1587 	}
1588 
1589 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1590 	priv->cmd.max_cmds = 1;
1591 	mlx4_warn(dev, "Sending reset\n");
1592 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1593 				       MLX4_COMM_TIME);
1594 	/* if we are in the middle of flr the slave will try
1595 	 * NUM_OF_RESET_RETRIES times before leaving.*/
1596 	if (ret_from_reset) {
1597 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1598 			mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
1599 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
1600 			return -EPROBE_DEFER;
1601 		} else
1602 			goto err;
1603 	}
1604 
1605 	/* check the driver version - the slave I/F revision
1606 	 * must match the master's */
1607 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1608 	cmd_channel_ver = mlx4_comm_get_version();
1609 
1610 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1611 		MLX4_COMM_GET_IF_REV(slave_read)) {
1612 		mlx4_err(dev, "slave driver version is not supported by the master\n");
1613 		goto err;
1614 	}
1615 
1616 	mlx4_warn(dev, "Sending vhcr0\n");
1617 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1618 						    MLX4_COMM_TIME))
1619 		goto err;
1620 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1621 						    MLX4_COMM_TIME))
1622 		goto err;
1623 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1624 						    MLX4_COMM_TIME))
1625 		goto err;
1626 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma, MLX4_COMM_TIME))
1627 		goto err;
1628 
1629 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1630 	return 0;
1631 
1632 err:
1633 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, 0);
1634 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1635 	return -EIO;
1636 }
1637 
1638 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1639 {
1640 	int i;
1641 
1642 	for (i = 1; i <= dev->caps.num_ports; i++) {
1643 		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
1644 			dev->caps.gid_table_len[i] =
1645 				mlx4_get_slave_num_gids(dev, 0, i);
1646 		else
1647 			dev->caps.gid_table_len[i] = 1;
1648 		dev->caps.pkey_table_len[i] =
1649 			dev->phys_caps.pkey_phys_table_len[i] - 1;
1650 	}
1651 }
1652 
1653 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
1654 {
1655 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
1656 
1657 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
1658 	      i++) {
1659 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
1660 			break;
1661 	}
1662 
1663 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
1664 }
1665 
1666 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
1667 {
1668 	switch (dmfs_high_steer_mode) {
1669 	case MLX4_STEERING_DMFS_A0_DEFAULT:
1670 		return "default performance";
1671 
1672 	case MLX4_STEERING_DMFS_A0_DYNAMIC:
1673 		return "dynamic hybrid mode";
1674 
1675 	case MLX4_STEERING_DMFS_A0_STATIC:
1676 		return "performance optimized for limited rule configuration (static)";
1677 
1678 	case MLX4_STEERING_DMFS_A0_DISABLE:
1679 		return "disabled performance optimized steering";
1680 
1681 	case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
1682 		return "performance optimized steering not supported";
1683 
1684 	default:
1685 		return "Unrecognized mode";
1686 	}
1687 }
1688 
1689 #define MLX4_DMFS_A0_STEERING			(1UL << 2)
1690 
1691 static void choose_steering_mode(struct mlx4_dev *dev,
1692 				 struct mlx4_dev_cap *dev_cap)
1693 {
1694 	if (mlx4_log_num_mgm_entry_size <= 0) {
1695 		if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
1696 			if (dev->caps.dmfs_high_steer_mode ==
1697 			    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1698 				mlx4_err(dev, "DMFS high rate mode not supported\n");
1699 			else
1700 				dev->caps.dmfs_high_steer_mode =
1701 					MLX4_STEERING_DMFS_A0_STATIC;
1702 		}
1703 	}
1704 
1705 	if (mlx4_log_num_mgm_entry_size <= 0 &&
1706 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
1707 	    (!mlx4_is_mfunc(dev) ||
1708 	     (dev_cap->fs_max_num_qp_per_entry >= (dev->num_vfs + 1))) &&
1709 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
1710 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
1711 		dev->oper_log_mgm_entry_size =
1712 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
1713 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1714 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
1715 		dev->caps.fs_log_max_ucast_qp_range_size =
1716 			dev_cap->fs_log_max_ucast_qp_range_size;
1717 	} else {
1718 		if (dev->caps.dmfs_high_steer_mode !=
1719 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1720 			dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
1721 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
1722 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1723 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
1724 		else {
1725 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
1726 
1727 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
1728 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1729 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
1730 		}
1731 		dev->oper_log_mgm_entry_size =
1732 			mlx4_log_num_mgm_entry_size > 0 ?
1733 			mlx4_log_num_mgm_entry_size :
1734 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
1735 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
1736 	}
1737 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
1738 		 mlx4_steering_mode_str(dev->caps.steering_mode),
1739 		 dev->oper_log_mgm_entry_size,
1740 		 mlx4_log_num_mgm_entry_size);
1741 }
1742 
1743 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
1744 				       struct mlx4_dev_cap *dev_cap)
1745 {
1746 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
1747 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS &&
1748 	    dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
1749 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
1750 	else
1751 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
1752 
1753 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
1754 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
1755 }
1756 
1757 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
1758 {
1759 	int i;
1760 	struct mlx4_port_cap port_cap;
1761 
1762 	if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1763 		return -EINVAL;
1764 
1765 	for (i = 1; i <= dev->caps.num_ports; i++) {
1766 		if (mlx4_dev_port(dev, i, &port_cap)) {
1767 			mlx4_err(dev,
1768 				 "QUERY_DEV_CAP command failed, can't veify DMFS high rate steering.\n");
1769 		} else if ((dev->caps.dmfs_high_steer_mode !=
1770 			    MLX4_STEERING_DMFS_A0_DEFAULT) &&
1771 			   (port_cap.dmfs_optimized_state ==
1772 			    !!(dev->caps.dmfs_high_steer_mode ==
1773 			    MLX4_STEERING_DMFS_A0_DISABLE))) {
1774 			mlx4_err(dev,
1775 				 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
1776 				 dmfs_high_rate_steering_mode_str(
1777 					dev->caps.dmfs_high_steer_mode),
1778 				 (port_cap.dmfs_optimized_state ?
1779 					"enabled" : "disabled"));
1780 		}
1781 	}
1782 
1783 	return 0;
1784 }
1785 
1786 static int mlx4_init_fw(struct mlx4_dev *dev)
1787 {
1788 	struct mlx4_mod_stat_cfg   mlx4_cfg;
1789 	int err = 0;
1790 
1791 	if (!mlx4_is_slave(dev)) {
1792 		err = mlx4_QUERY_FW(dev);
1793 		if (err) {
1794 			if (err == -EACCES)
1795 				mlx4_info(dev, "non-primary physical function, skipping\n");
1796 			else
1797 				mlx4_err(dev, "QUERY_FW command failed, aborting\n");
1798 			return err;
1799 		}
1800 
1801 		err = mlx4_load_fw(dev);
1802 		if (err) {
1803 			mlx4_err(dev, "Failed to start FW, aborting\n");
1804 			return err;
1805 		}
1806 
1807 		mlx4_cfg.log_pg_sz_m = 1;
1808 		mlx4_cfg.log_pg_sz = 0;
1809 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
1810 		if (err)
1811 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
1812 	}
1813 
1814 	return err;
1815 }
1816 
1817 static int mlx4_init_hca(struct mlx4_dev *dev)
1818 {
1819 	struct mlx4_priv	  *priv = mlx4_priv(dev);
1820 	struct mlx4_adapter	   adapter;
1821 	struct mlx4_dev_cap	   dev_cap;
1822 	struct mlx4_profile	   profile;
1823 	struct mlx4_init_hca_param init_hca;
1824 	u64 icm_size;
1825 	struct mlx4_config_dev_params params;
1826 	int err;
1827 
1828 	if (!mlx4_is_slave(dev)) {
1829 		err = mlx4_dev_cap(dev, &dev_cap);
1830 		if (err) {
1831 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
1832 			goto err_stop_fw;
1833 		}
1834 
1835 		choose_steering_mode(dev, &dev_cap);
1836 		choose_tunnel_offload_mode(dev, &dev_cap);
1837 
1838 		if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
1839 		    mlx4_is_master(dev))
1840 			dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
1841 
1842 		err = mlx4_get_phys_port_id(dev);
1843 		if (err)
1844 			mlx4_err(dev, "Fail to get physical port id\n");
1845 
1846 		if (mlx4_is_master(dev))
1847 			mlx4_parav_master_pf_caps(dev);
1848 
1849 		if (mlx4_low_memory_profile()) {
1850 			mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
1851 			profile = low_mem_profile;
1852 		} else {
1853 			profile = default_profile;
1854 		}
1855 		if (dev->caps.steering_mode ==
1856 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
1857 			profile.num_mcg = MLX4_FS_NUM_MCG;
1858 
1859 		icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
1860 					     &init_hca);
1861 		if ((long long) icm_size < 0) {
1862 			err = icm_size;
1863 			goto err_stop_fw;
1864 		}
1865 
1866 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
1867 
1868 		init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
1869 		init_hca.uar_page_sz = PAGE_SHIFT - 12;
1870 		init_hca.mw_enabled = 0;
1871 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1872 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
1873 			init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
1874 
1875 		err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
1876 		if (err)
1877 			goto err_stop_fw;
1878 
1879 		err = mlx4_INIT_HCA(dev, &init_hca);
1880 		if (err) {
1881 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
1882 			goto err_free_icm;
1883 		}
1884 
1885 		if (dev_cap.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
1886 			err = mlx4_query_func(dev, &dev_cap);
1887 			if (err < 0) {
1888 				mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
1889 				goto err_stop_fw;
1890 			} else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
1891 				dev->caps.num_eqs = dev_cap.max_eqs;
1892 				dev->caps.reserved_eqs = dev_cap.reserved_eqs;
1893 				dev->caps.reserved_uars = dev_cap.reserved_uars;
1894 			}
1895 		}
1896 
1897 		/*
1898 		 * If TS is supported by FW
1899 		 * read HCA frequency by QUERY_HCA command
1900 		 */
1901 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
1902 			memset(&init_hca, 0, sizeof(init_hca));
1903 			err = mlx4_QUERY_HCA(dev, &init_hca);
1904 			if (err) {
1905 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
1906 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1907 			} else {
1908 				dev->caps.hca_core_clock =
1909 					init_hca.hca_core_clock;
1910 			}
1911 
1912 			/* In case we got HCA frequency 0 - disable timestamping
1913 			 * to avoid dividing by zero
1914 			 */
1915 			if (!dev->caps.hca_core_clock) {
1916 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1917 				mlx4_err(dev,
1918 					 "HCA frequency is 0 - timestamping is not supported\n");
1919 			} else if (map_internal_clock(dev)) {
1920 				/*
1921 				 * Map internal clock,
1922 				 * in case of failure disable timestamping
1923 				 */
1924 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1925 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
1926 			}
1927 		}
1928 
1929 		if (dev->caps.dmfs_high_steer_mode !=
1930 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
1931 			if (mlx4_validate_optimized_steering(dev))
1932 				mlx4_warn(dev, "Optimized steering validation failed\n");
1933 
1934 			if (dev->caps.dmfs_high_steer_mode ==
1935 			    MLX4_STEERING_DMFS_A0_DISABLE) {
1936 				dev->caps.dmfs_high_rate_qpn_base =
1937 					dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
1938 				dev->caps.dmfs_high_rate_qpn_range =
1939 					MLX4_A0_STEERING_TABLE_SIZE;
1940 			}
1941 
1942 			mlx4_dbg(dev, "DMFS high rate steer mode is: %s\n",
1943 				 dmfs_high_rate_steering_mode_str(
1944 					dev->caps.dmfs_high_steer_mode));
1945 		}
1946 	} else {
1947 		err = mlx4_init_slave(dev);
1948 		if (err) {
1949 			if (err != -EPROBE_DEFER)
1950 				mlx4_err(dev, "Failed to initialize slave\n");
1951 			return err;
1952 		}
1953 
1954 		err = mlx4_slave_cap(dev);
1955 		if (err) {
1956 			mlx4_err(dev, "Failed to obtain slave caps\n");
1957 			goto err_close;
1958 		}
1959 	}
1960 
1961 	if (map_bf_area(dev))
1962 		mlx4_dbg(dev, "Failed to map blue flame area\n");
1963 
1964 	/*Only the master set the ports, all the rest got it from it.*/
1965 	if (!mlx4_is_slave(dev))
1966 		mlx4_set_port_mask(dev);
1967 
1968 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
1969 	if (err) {
1970 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
1971 		goto unmap_bf;
1972 	}
1973 
1974 	/* Query CONFIG_DEV parameters */
1975 	err = mlx4_config_dev_retrieval(dev, &params);
1976 	if (err && err != -ENOTSUPP) {
1977 		mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
1978 	} else if (!err) {
1979 		dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
1980 		dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
1981 	}
1982 	priv->eq_table.inta_pin = adapter.inta_pin;
1983 	memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
1984 
1985 	return 0;
1986 
1987 unmap_bf:
1988 	unmap_internal_clock(dev);
1989 	unmap_bf_area(dev);
1990 
1991 	if (mlx4_is_slave(dev)) {
1992 		kfree(dev->caps.qp0_qkey);
1993 		kfree(dev->caps.qp0_tunnel);
1994 		kfree(dev->caps.qp0_proxy);
1995 		kfree(dev->caps.qp1_tunnel);
1996 		kfree(dev->caps.qp1_proxy);
1997 	}
1998 
1999 err_close:
2000 	if (mlx4_is_slave(dev))
2001 		mlx4_slave_exit(dev);
2002 	else
2003 		mlx4_CLOSE_HCA(dev, 0);
2004 
2005 err_free_icm:
2006 	if (!mlx4_is_slave(dev))
2007 		mlx4_free_icms(dev);
2008 
2009 err_stop_fw:
2010 	if (!mlx4_is_slave(dev)) {
2011 		mlx4_UNMAP_FA(dev);
2012 		mlx4_free_icm(dev, priv->fw.fw_icm, 0);
2013 	}
2014 	return err;
2015 }
2016 
2017 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2018 {
2019 	struct mlx4_priv *priv = mlx4_priv(dev);
2020 	int nent;
2021 
2022 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2023 		return -ENOENT;
2024 
2025 	nent = dev->caps.max_counters;
2026 	return mlx4_bitmap_init(&priv->counters_bitmap, nent, nent - 1, 0, 0);
2027 }
2028 
2029 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2030 {
2031 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2032 }
2033 
2034 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2035 {
2036 	struct mlx4_priv *priv = mlx4_priv(dev);
2037 
2038 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2039 		return -ENOENT;
2040 
2041 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2042 	if (*idx == -1)
2043 		return -ENOMEM;
2044 
2045 	return 0;
2046 }
2047 
2048 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2049 {
2050 	u64 out_param;
2051 	int err;
2052 
2053 	if (mlx4_is_mfunc(dev)) {
2054 		err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
2055 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2056 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2057 		if (!err)
2058 			*idx = get_param_l(&out_param);
2059 
2060 		return err;
2061 	}
2062 	return __mlx4_counter_alloc(dev, idx);
2063 }
2064 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2065 
2066 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2067 {
2068 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2069 	return;
2070 }
2071 
2072 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2073 {
2074 	u64 in_param = 0;
2075 
2076 	if (mlx4_is_mfunc(dev)) {
2077 		set_param_l(&in_param, idx);
2078 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2079 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2080 			 MLX4_CMD_WRAPPED);
2081 		return;
2082 	}
2083 	__mlx4_counter_free(dev, idx);
2084 }
2085 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2086 
2087 static int mlx4_setup_hca(struct mlx4_dev *dev)
2088 {
2089 	struct mlx4_priv *priv = mlx4_priv(dev);
2090 	int err;
2091 	int port;
2092 	__be32 ib_port_default_caps;
2093 
2094 	err = mlx4_init_uar_table(dev);
2095 	if (err) {
2096 		mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2097 		 return err;
2098 	}
2099 
2100 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
2101 	if (err) {
2102 		mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2103 		goto err_uar_table_free;
2104 	}
2105 
2106 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2107 	if (!priv->kar) {
2108 		mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2109 		err = -ENOMEM;
2110 		goto err_uar_free;
2111 	}
2112 
2113 	err = mlx4_init_pd_table(dev);
2114 	if (err) {
2115 		mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2116 		goto err_kar_unmap;
2117 	}
2118 
2119 	err = mlx4_init_xrcd_table(dev);
2120 	if (err) {
2121 		mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2122 		goto err_pd_table_free;
2123 	}
2124 
2125 	err = mlx4_init_mr_table(dev);
2126 	if (err) {
2127 		mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2128 		goto err_xrcd_table_free;
2129 	}
2130 
2131 	if (!mlx4_is_slave(dev)) {
2132 		err = mlx4_init_mcg_table(dev);
2133 		if (err) {
2134 			mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2135 			goto err_mr_table_free;
2136 		}
2137 		err = mlx4_config_mad_demux(dev);
2138 		if (err) {
2139 			mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2140 			goto err_mcg_table_free;
2141 		}
2142 	}
2143 
2144 	err = mlx4_init_eq_table(dev);
2145 	if (err) {
2146 		mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2147 		goto err_mcg_table_free;
2148 	}
2149 
2150 	err = mlx4_cmd_use_events(dev);
2151 	if (err) {
2152 		mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2153 		goto err_eq_table_free;
2154 	}
2155 
2156 	err = mlx4_NOP(dev);
2157 	if (err) {
2158 		if (dev->flags & MLX4_FLAG_MSI_X) {
2159 			mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2160 				  priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
2161 			mlx4_warn(dev, "Trying again without MSI-X\n");
2162 		} else {
2163 			mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2164 				 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
2165 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2166 		}
2167 
2168 		goto err_cmd_poll;
2169 	}
2170 
2171 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
2172 
2173 	err = mlx4_init_cq_table(dev);
2174 	if (err) {
2175 		mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2176 		goto err_cmd_poll;
2177 	}
2178 
2179 	err = mlx4_init_srq_table(dev);
2180 	if (err) {
2181 		mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2182 		goto err_cq_table_free;
2183 	}
2184 
2185 	err = mlx4_init_qp_table(dev);
2186 	if (err) {
2187 		mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2188 		goto err_srq_table_free;
2189 	}
2190 
2191 	err = mlx4_init_counters_table(dev);
2192 	if (err && err != -ENOENT) {
2193 		mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2194 		goto err_qp_table_free;
2195 	}
2196 
2197 	if (!mlx4_is_slave(dev)) {
2198 		for (port = 1; port <= dev->caps.num_ports; port++) {
2199 			ib_port_default_caps = 0;
2200 			err = mlx4_get_port_ib_caps(dev, port,
2201 						    &ib_port_default_caps);
2202 			if (err)
2203 				mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2204 					  port, err);
2205 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2206 
2207 			/* initialize per-slave default ib port capabilities */
2208 			if (mlx4_is_master(dev)) {
2209 				int i;
2210 				for (i = 0; i < dev->num_slaves; i++) {
2211 					if (i == mlx4_master_func_num(dev))
2212 						continue;
2213 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2214 						ib_port_default_caps;
2215 				}
2216 			}
2217 
2218 			if (mlx4_is_mfunc(dev))
2219 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2220 			else
2221 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2222 
2223 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2224 					    dev->caps.pkey_table_len[port] : -1);
2225 			if (err) {
2226 				mlx4_err(dev, "Failed to set port %d, aborting\n",
2227 					 port);
2228 				goto err_counters_table_free;
2229 			}
2230 		}
2231 	}
2232 
2233 	return 0;
2234 
2235 err_counters_table_free:
2236 	mlx4_cleanup_counters_table(dev);
2237 
2238 err_qp_table_free:
2239 	mlx4_cleanup_qp_table(dev);
2240 
2241 err_srq_table_free:
2242 	mlx4_cleanup_srq_table(dev);
2243 
2244 err_cq_table_free:
2245 	mlx4_cleanup_cq_table(dev);
2246 
2247 err_cmd_poll:
2248 	mlx4_cmd_use_polling(dev);
2249 
2250 err_eq_table_free:
2251 	mlx4_cleanup_eq_table(dev);
2252 
2253 err_mcg_table_free:
2254 	if (!mlx4_is_slave(dev))
2255 		mlx4_cleanup_mcg_table(dev);
2256 
2257 err_mr_table_free:
2258 	mlx4_cleanup_mr_table(dev);
2259 
2260 err_xrcd_table_free:
2261 	mlx4_cleanup_xrcd_table(dev);
2262 
2263 err_pd_table_free:
2264 	mlx4_cleanup_pd_table(dev);
2265 
2266 err_kar_unmap:
2267 	iounmap(priv->kar);
2268 
2269 err_uar_free:
2270 	mlx4_uar_free(dev, &priv->driver_uar);
2271 
2272 err_uar_table_free:
2273 	mlx4_cleanup_uar_table(dev);
2274 	return err;
2275 }
2276 
2277 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2278 {
2279 	struct mlx4_priv *priv = mlx4_priv(dev);
2280 	struct msix_entry *entries;
2281 	int i;
2282 
2283 	if (msi_x) {
2284 		int nreq = dev->caps.num_ports * num_online_cpus() + MSIX_LEGACY_SZ;
2285 
2286 		nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
2287 			     nreq);
2288 
2289 		entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
2290 		if (!entries)
2291 			goto no_msi;
2292 
2293 		for (i = 0; i < nreq; ++i)
2294 			entries[i].entry = i;
2295 
2296 		nreq = pci_enable_msix_range(dev->pdev, entries, 2, nreq);
2297 
2298 		if (nreq < 0) {
2299 			kfree(entries);
2300 			goto no_msi;
2301 		} else if (nreq < MSIX_LEGACY_SZ +
2302 			   dev->caps.num_ports * MIN_MSIX_P_PORT) {
2303 			/*Working in legacy mode , all EQ's shared*/
2304 			dev->caps.comp_pool           = 0;
2305 			dev->caps.num_comp_vectors = nreq - 1;
2306 		} else {
2307 			dev->caps.comp_pool           = nreq - MSIX_LEGACY_SZ;
2308 			dev->caps.num_comp_vectors = MSIX_LEGACY_SZ - 1;
2309 		}
2310 		for (i = 0; i < nreq; ++i)
2311 			priv->eq_table.eq[i].irq = entries[i].vector;
2312 
2313 		dev->flags |= MLX4_FLAG_MSI_X;
2314 
2315 		kfree(entries);
2316 		return;
2317 	}
2318 
2319 no_msi:
2320 	dev->caps.num_comp_vectors = 1;
2321 	dev->caps.comp_pool	   = 0;
2322 
2323 	for (i = 0; i < 2; ++i)
2324 		priv->eq_table.eq[i].irq = dev->pdev->irq;
2325 }
2326 
2327 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2328 {
2329 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2330 	int err = 0;
2331 
2332 	info->dev = dev;
2333 	info->port = port;
2334 	if (!mlx4_is_slave(dev)) {
2335 		mlx4_init_mac_table(dev, &info->mac_table);
2336 		mlx4_init_vlan_table(dev, &info->vlan_table);
2337 		mlx4_init_roce_gid_table(dev, &info->gid_table);
2338 		info->base_qpn = mlx4_get_base_qpn(dev, port);
2339 	}
2340 
2341 	sprintf(info->dev_name, "mlx4_port%d", port);
2342 	info->port_attr.attr.name = info->dev_name;
2343 	if (mlx4_is_mfunc(dev))
2344 		info->port_attr.attr.mode = S_IRUGO;
2345 	else {
2346 		info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2347 		info->port_attr.store     = set_port_type;
2348 	}
2349 	info->port_attr.show      = show_port_type;
2350 	sysfs_attr_init(&info->port_attr.attr);
2351 
2352 	err = device_create_file(&dev->pdev->dev, &info->port_attr);
2353 	if (err) {
2354 		mlx4_err(dev, "Failed to create file for port %d\n", port);
2355 		info->port = -1;
2356 	}
2357 
2358 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2359 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
2360 	if (mlx4_is_mfunc(dev))
2361 		info->port_mtu_attr.attr.mode = S_IRUGO;
2362 	else {
2363 		info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2364 		info->port_mtu_attr.store     = set_port_ib_mtu;
2365 	}
2366 	info->port_mtu_attr.show      = show_port_ib_mtu;
2367 	sysfs_attr_init(&info->port_mtu_attr.attr);
2368 
2369 	err = device_create_file(&dev->pdev->dev, &info->port_mtu_attr);
2370 	if (err) {
2371 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
2372 		device_remove_file(&info->dev->pdev->dev, &info->port_attr);
2373 		info->port = -1;
2374 	}
2375 
2376 	return err;
2377 }
2378 
2379 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
2380 {
2381 	if (info->port < 0)
2382 		return;
2383 
2384 	device_remove_file(&info->dev->pdev->dev, &info->port_attr);
2385 	device_remove_file(&info->dev->pdev->dev, &info->port_mtu_attr);
2386 }
2387 
2388 static int mlx4_init_steering(struct mlx4_dev *dev)
2389 {
2390 	struct mlx4_priv *priv = mlx4_priv(dev);
2391 	int num_entries = dev->caps.num_ports;
2392 	int i, j;
2393 
2394 	priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
2395 	if (!priv->steer)
2396 		return -ENOMEM;
2397 
2398 	for (i = 0; i < num_entries; i++)
2399 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
2400 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
2401 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
2402 		}
2403 	return 0;
2404 }
2405 
2406 static void mlx4_clear_steering(struct mlx4_dev *dev)
2407 {
2408 	struct mlx4_priv *priv = mlx4_priv(dev);
2409 	struct mlx4_steer_index *entry, *tmp_entry;
2410 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
2411 	int num_entries = dev->caps.num_ports;
2412 	int i, j;
2413 
2414 	for (i = 0; i < num_entries; i++) {
2415 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
2416 			list_for_each_entry_safe(pqp, tmp_pqp,
2417 						 &priv->steer[i].promisc_qps[j],
2418 						 list) {
2419 				list_del(&pqp->list);
2420 				kfree(pqp);
2421 			}
2422 			list_for_each_entry_safe(entry, tmp_entry,
2423 						 &priv->steer[i].steer_entries[j],
2424 						 list) {
2425 				list_del(&entry->list);
2426 				list_for_each_entry_safe(pqp, tmp_pqp,
2427 							 &entry->duplicates,
2428 							 list) {
2429 					list_del(&pqp->list);
2430 					kfree(pqp);
2431 				}
2432 				kfree(entry);
2433 			}
2434 		}
2435 	}
2436 	kfree(priv->steer);
2437 }
2438 
2439 static int extended_func_num(struct pci_dev *pdev)
2440 {
2441 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
2442 }
2443 
2444 #define MLX4_OWNER_BASE	0x8069c
2445 #define MLX4_OWNER_SIZE	4
2446 
2447 static int mlx4_get_ownership(struct mlx4_dev *dev)
2448 {
2449 	void __iomem *owner;
2450 	u32 ret;
2451 
2452 	if (pci_channel_offline(dev->pdev))
2453 		return -EIO;
2454 
2455 	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
2456 			MLX4_OWNER_SIZE);
2457 	if (!owner) {
2458 		mlx4_err(dev, "Failed to obtain ownership bit\n");
2459 		return -ENOMEM;
2460 	}
2461 
2462 	ret = readl(owner);
2463 	iounmap(owner);
2464 	return (int) !!ret;
2465 }
2466 
2467 static void mlx4_free_ownership(struct mlx4_dev *dev)
2468 {
2469 	void __iomem *owner;
2470 
2471 	if (pci_channel_offline(dev->pdev))
2472 		return;
2473 
2474 	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
2475 			MLX4_OWNER_SIZE);
2476 	if (!owner) {
2477 		mlx4_err(dev, "Failed to obtain ownership bit\n");
2478 		return;
2479 	}
2480 	writel(0, owner);
2481 	msleep(1000);
2482 	iounmap(owner);
2483 }
2484 
2485 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV)	==\
2486 				  !!((flags) & MLX4_FLAG_MASTER))
2487 
2488 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
2489 			     u8 total_vfs, int existing_vfs)
2490 {
2491 	u64 dev_flags = dev->flags;
2492 	int err = 0;
2493 
2494 	atomic_inc(&pf_loading);
2495 	if (dev->flags &  MLX4_FLAG_SRIOV) {
2496 		if (existing_vfs != total_vfs) {
2497 			mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
2498 				 existing_vfs, total_vfs);
2499 			total_vfs = existing_vfs;
2500 		}
2501 	}
2502 
2503 	dev->dev_vfs = kzalloc(total_vfs * sizeof(*dev->dev_vfs), GFP_KERNEL);
2504 	if (NULL == dev->dev_vfs) {
2505 		mlx4_err(dev, "Failed to allocate memory for VFs\n");
2506 		goto disable_sriov;
2507 	}
2508 
2509 	if (!(dev->flags &  MLX4_FLAG_SRIOV)) {
2510 		mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
2511 		err = pci_enable_sriov(pdev, total_vfs);
2512 	}
2513 	if (err) {
2514 		mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
2515 			 err);
2516 		goto disable_sriov;
2517 	} else {
2518 		mlx4_warn(dev, "Running in master mode\n");
2519 		dev_flags |= MLX4_FLAG_SRIOV |
2520 			MLX4_FLAG_MASTER;
2521 		dev_flags &= ~MLX4_FLAG_SLAVE;
2522 		dev->num_vfs = total_vfs;
2523 	}
2524 	return dev_flags;
2525 
2526 disable_sriov:
2527 	atomic_dec(&pf_loading);
2528 	dev->num_vfs = 0;
2529 	kfree(dev->dev_vfs);
2530 	return dev_flags & ~MLX4_FLAG_MASTER;
2531 }
2532 
2533 enum {
2534 	MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
2535 };
2536 
2537 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
2538 			      int *nvfs)
2539 {
2540 	int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
2541 	/* Checking for 64 VFs as a limitation of CX2 */
2542 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
2543 	    requested_vfs >= 64) {
2544 		mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
2545 			 requested_vfs);
2546 		return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
2547 	}
2548 	return 0;
2549 }
2550 
2551 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
2552 			 int total_vfs, int *nvfs, struct mlx4_priv *priv)
2553 {
2554 	struct mlx4_dev *dev;
2555 	unsigned sum = 0;
2556 	int err;
2557 	int port;
2558 	int i;
2559 	struct mlx4_dev_cap *dev_cap = NULL;
2560 	int existing_vfs = 0;
2561 
2562 	dev = &priv->dev;
2563 
2564 	INIT_LIST_HEAD(&priv->ctx_list);
2565 	spin_lock_init(&priv->ctx_lock);
2566 
2567 	mutex_init(&priv->port_mutex);
2568 
2569 	INIT_LIST_HEAD(&priv->pgdir_list);
2570 	mutex_init(&priv->pgdir_mutex);
2571 
2572 	INIT_LIST_HEAD(&priv->bf_list);
2573 	mutex_init(&priv->bf_mutex);
2574 
2575 	dev->rev_id = pdev->revision;
2576 	dev->numa_node = dev_to_node(&pdev->dev);
2577 
2578 	/* Detect if this device is a virtual function */
2579 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
2580 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
2581 		dev->flags |= MLX4_FLAG_SLAVE;
2582 	} else {
2583 		/* We reset the device and enable SRIOV only for physical
2584 		 * devices.  Try to claim ownership on the device;
2585 		 * if already taken, skip -- do not allow multiple PFs */
2586 		err = mlx4_get_ownership(dev);
2587 		if (err) {
2588 			if (err < 0)
2589 				return err;
2590 			else {
2591 				mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
2592 				return -EINVAL;
2593 			}
2594 		}
2595 
2596 		atomic_set(&priv->opreq_count, 0);
2597 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
2598 
2599 		/*
2600 		 * Now reset the HCA before we touch the PCI capabilities or
2601 		 * attempt a firmware command, since a boot ROM may have left
2602 		 * the HCA in an undefined state.
2603 		 */
2604 		err = mlx4_reset(dev);
2605 		if (err) {
2606 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
2607 			goto err_sriov;
2608 		}
2609 
2610 		if (total_vfs) {
2611 			dev->flags = MLX4_FLAG_MASTER;
2612 			existing_vfs = pci_num_vf(pdev);
2613 			if (existing_vfs)
2614 				dev->flags |= MLX4_FLAG_SRIOV;
2615 			dev->num_vfs = total_vfs;
2616 		}
2617 	}
2618 
2619 slave_start:
2620 	err = mlx4_cmd_init(dev);
2621 	if (err) {
2622 		mlx4_err(dev, "Failed to init command interface, aborting\n");
2623 		goto err_sriov;
2624 	}
2625 
2626 	/* In slave functions, the communication channel must be initialized
2627 	 * before posting commands. Also, init num_slaves before calling
2628 	 * mlx4_init_hca */
2629 	if (mlx4_is_mfunc(dev)) {
2630 		if (mlx4_is_master(dev)) {
2631 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
2632 
2633 		} else {
2634 			dev->num_slaves = 0;
2635 			err = mlx4_multi_func_init(dev);
2636 			if (err) {
2637 				mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
2638 				goto err_cmd;
2639 			}
2640 		}
2641 	}
2642 
2643 	err = mlx4_init_fw(dev);
2644 	if (err) {
2645 		mlx4_err(dev, "Failed to init fw, aborting.\n");
2646 		goto err_mfunc;
2647 	}
2648 
2649 	if (mlx4_is_master(dev)) {
2650 		/* when we hit the goto slave_start below, dev_cap already initialized */
2651 		if (!dev_cap) {
2652 			dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
2653 
2654 			if (!dev_cap) {
2655 				err = -ENOMEM;
2656 				goto err_fw;
2657 			}
2658 
2659 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
2660 			if (err) {
2661 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
2662 				goto err_fw;
2663 			}
2664 
2665 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
2666 				goto err_fw;
2667 
2668 			if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
2669 				u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
2670 								  existing_vfs);
2671 
2672 				mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
2673 				dev->flags = dev_flags;
2674 				if (!SRIOV_VALID_STATE(dev->flags)) {
2675 					mlx4_err(dev, "Invalid SRIOV state\n");
2676 					goto err_sriov;
2677 				}
2678 				err = mlx4_reset(dev);
2679 				if (err) {
2680 					mlx4_err(dev, "Failed to reset HCA, aborting.\n");
2681 					goto err_sriov;
2682 				}
2683 				goto slave_start;
2684 			}
2685 		} else {
2686 			/* Legacy mode FW requires SRIOV to be enabled before
2687 			 * doing QUERY_DEV_CAP, since max_eq's value is different if
2688 			 * SRIOV is enabled.
2689 			 */
2690 			memset(dev_cap, 0, sizeof(*dev_cap));
2691 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
2692 			if (err) {
2693 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
2694 				goto err_fw;
2695 			}
2696 
2697 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
2698 				goto err_fw;
2699 		}
2700 	}
2701 
2702 	err = mlx4_init_hca(dev);
2703 	if (err) {
2704 		if (err == -EACCES) {
2705 			/* Not primary Physical function
2706 			 * Running in slave mode */
2707 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
2708 			/* We're not a PF */
2709 			if (dev->flags & MLX4_FLAG_SRIOV) {
2710 				if (!existing_vfs)
2711 					pci_disable_sriov(pdev);
2712 				if (mlx4_is_master(dev))
2713 					atomic_dec(&pf_loading);
2714 				dev->flags &= ~MLX4_FLAG_SRIOV;
2715 			}
2716 			if (!mlx4_is_slave(dev))
2717 				mlx4_free_ownership(dev);
2718 			dev->flags |= MLX4_FLAG_SLAVE;
2719 			dev->flags &= ~MLX4_FLAG_MASTER;
2720 			goto slave_start;
2721 		} else
2722 			goto err_fw;
2723 	}
2724 
2725 	if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
2726 		u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs, existing_vfs);
2727 
2728 		if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
2729 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
2730 			dev->flags = dev_flags;
2731 			err = mlx4_cmd_init(dev);
2732 			if (err) {
2733 				/* Only VHCR is cleaned up, so could still
2734 				 * send FW commands
2735 				 */
2736 				mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
2737 				goto err_close;
2738 			}
2739 		} else {
2740 			dev->flags = dev_flags;
2741 		}
2742 
2743 		if (!SRIOV_VALID_STATE(dev->flags)) {
2744 			mlx4_err(dev, "Invalid SRIOV state\n");
2745 			goto err_close;
2746 		}
2747 	}
2748 
2749 	/* check if the device is functioning at its maximum possible speed.
2750 	 * No return code for this call, just warn the user in case of PCI
2751 	 * express device capabilities are under-satisfied by the bus.
2752 	 */
2753 	if (!mlx4_is_slave(dev))
2754 		mlx4_check_pcie_caps(dev);
2755 
2756 	/* In master functions, the communication channel must be initialized
2757 	 * after obtaining its address from fw */
2758 	if (mlx4_is_master(dev)) {
2759 		int ib_ports = 0;
2760 
2761 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2762 			ib_ports++;
2763 
2764 		if (ib_ports &&
2765 		    (num_vfs_argc > 1 || probe_vfs_argc > 1)) {
2766 			mlx4_err(dev,
2767 				 "Invalid syntax of num_vfs/probe_vfs with IB port - single port VFs syntax is only supported when all ports are configured as ethernet\n");
2768 			err = -EINVAL;
2769 			goto err_close;
2770 		}
2771 		if (dev->caps.num_ports < 2 &&
2772 		    num_vfs_argc > 1) {
2773 			err = -EINVAL;
2774 			mlx4_err(dev,
2775 				 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
2776 				 dev->caps.num_ports);
2777 			goto err_close;
2778 		}
2779 		memcpy(dev->nvfs, nvfs, sizeof(dev->nvfs));
2780 
2781 		for (i = 0; i < sizeof(dev->nvfs)/sizeof(dev->nvfs[0]); i++) {
2782 			unsigned j;
2783 
2784 			for (j = 0; j < dev->nvfs[i]; ++sum, ++j) {
2785 				dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
2786 				dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
2787 					dev->caps.num_ports;
2788 			}
2789 		}
2790 
2791 		/* In master functions, the communication channel
2792 		 * must be initialized after obtaining its address from fw
2793 		 */
2794 		err = mlx4_multi_func_init(dev);
2795 		if (err) {
2796 			mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
2797 			goto err_close;
2798 		}
2799 	}
2800 
2801 	err = mlx4_alloc_eq_table(dev);
2802 	if (err)
2803 		goto err_master_mfunc;
2804 
2805 	priv->msix_ctl.pool_bm = 0;
2806 	mutex_init(&priv->msix_ctl.pool_lock);
2807 
2808 	mlx4_enable_msi_x(dev);
2809 	if ((mlx4_is_mfunc(dev)) &&
2810 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
2811 		err = -ENOSYS;
2812 		mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
2813 		goto err_free_eq;
2814 	}
2815 
2816 	if (!mlx4_is_slave(dev)) {
2817 		err = mlx4_init_steering(dev);
2818 		if (err)
2819 			goto err_disable_msix;
2820 	}
2821 
2822 	err = mlx4_setup_hca(dev);
2823 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
2824 	    !mlx4_is_mfunc(dev)) {
2825 		dev->flags &= ~MLX4_FLAG_MSI_X;
2826 		dev->caps.num_comp_vectors = 1;
2827 		dev->caps.comp_pool	   = 0;
2828 		pci_disable_msix(pdev);
2829 		err = mlx4_setup_hca(dev);
2830 	}
2831 
2832 	if (err)
2833 		goto err_steer;
2834 
2835 	mlx4_init_quotas(dev);
2836 
2837 	for (port = 1; port <= dev->caps.num_ports; port++) {
2838 		err = mlx4_init_port_info(dev, port);
2839 		if (err)
2840 			goto err_port;
2841 	}
2842 
2843 	err = mlx4_register_device(dev);
2844 	if (err)
2845 		goto err_port;
2846 
2847 	mlx4_request_modules(dev);
2848 
2849 	mlx4_sense_init(dev);
2850 	mlx4_start_sense(dev);
2851 
2852 	priv->removed = 0;
2853 
2854 	if (mlx4_is_master(dev) && dev->num_vfs)
2855 		atomic_dec(&pf_loading);
2856 
2857 	kfree(dev_cap);
2858 	return 0;
2859 
2860 err_port:
2861 	for (--port; port >= 1; --port)
2862 		mlx4_cleanup_port_info(&priv->port[port]);
2863 
2864 	mlx4_cleanup_counters_table(dev);
2865 	mlx4_cleanup_qp_table(dev);
2866 	mlx4_cleanup_srq_table(dev);
2867 	mlx4_cleanup_cq_table(dev);
2868 	mlx4_cmd_use_polling(dev);
2869 	mlx4_cleanup_eq_table(dev);
2870 	mlx4_cleanup_mcg_table(dev);
2871 	mlx4_cleanup_mr_table(dev);
2872 	mlx4_cleanup_xrcd_table(dev);
2873 	mlx4_cleanup_pd_table(dev);
2874 	mlx4_cleanup_uar_table(dev);
2875 
2876 err_steer:
2877 	if (!mlx4_is_slave(dev))
2878 		mlx4_clear_steering(dev);
2879 
2880 err_disable_msix:
2881 	if (dev->flags & MLX4_FLAG_MSI_X)
2882 		pci_disable_msix(pdev);
2883 
2884 err_free_eq:
2885 	mlx4_free_eq_table(dev);
2886 
2887 err_master_mfunc:
2888 	if (mlx4_is_master(dev))
2889 		mlx4_multi_func_cleanup(dev);
2890 
2891 	if (mlx4_is_slave(dev)) {
2892 		kfree(dev->caps.qp0_qkey);
2893 		kfree(dev->caps.qp0_tunnel);
2894 		kfree(dev->caps.qp0_proxy);
2895 		kfree(dev->caps.qp1_tunnel);
2896 		kfree(dev->caps.qp1_proxy);
2897 	}
2898 
2899 err_close:
2900 	mlx4_close_hca(dev);
2901 
2902 err_fw:
2903 	mlx4_close_fw(dev);
2904 
2905 err_mfunc:
2906 	if (mlx4_is_slave(dev))
2907 		mlx4_multi_func_cleanup(dev);
2908 
2909 err_cmd:
2910 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
2911 
2912 err_sriov:
2913 	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs)
2914 		pci_disable_sriov(pdev);
2915 
2916 	if (mlx4_is_master(dev) && dev->num_vfs)
2917 		atomic_dec(&pf_loading);
2918 
2919 	kfree(priv->dev.dev_vfs);
2920 
2921 	if (!mlx4_is_slave(dev))
2922 		mlx4_free_ownership(dev);
2923 
2924 	kfree(dev_cap);
2925 	return err;
2926 }
2927 
2928 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
2929 			   struct mlx4_priv *priv)
2930 {
2931 	int err;
2932 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
2933 	int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
2934 	const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
2935 		{2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
2936 	unsigned total_vfs = 0;
2937 	unsigned int i;
2938 
2939 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
2940 
2941 	err = pci_enable_device(pdev);
2942 	if (err) {
2943 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
2944 		return err;
2945 	}
2946 
2947 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
2948 	 * per port, we must limit the number of VFs to 63 (since their are
2949 	 * 128 MACs)
2950 	 */
2951 	for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
2952 	     total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
2953 		nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
2954 		if (nvfs[i] < 0) {
2955 			dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
2956 			err = -EINVAL;
2957 			goto err_disable_pdev;
2958 		}
2959 	}
2960 	for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
2961 	     i++) {
2962 		prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
2963 		if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
2964 			dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
2965 			err = -EINVAL;
2966 			goto err_disable_pdev;
2967 		}
2968 	}
2969 	if (total_vfs >= MLX4_MAX_NUM_VF) {
2970 		dev_err(&pdev->dev,
2971 			"Requested more VF's (%d) than allowed (%d)\n",
2972 			total_vfs, MLX4_MAX_NUM_VF - 1);
2973 		err = -EINVAL;
2974 		goto err_disable_pdev;
2975 	}
2976 
2977 	for (i = 0; i < MLX4_MAX_PORTS; i++) {
2978 		if (nvfs[i] + nvfs[2] >= MLX4_MAX_NUM_VF_P_PORT) {
2979 			dev_err(&pdev->dev,
2980 				"Requested more VF's (%d) for port (%d) than allowed (%d)\n",
2981 				nvfs[i] + nvfs[2], i + 1,
2982 				MLX4_MAX_NUM_VF_P_PORT - 1);
2983 			err = -EINVAL;
2984 			goto err_disable_pdev;
2985 		}
2986 	}
2987 
2988 	/* Check for BARs. */
2989 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
2990 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
2991 		dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
2992 			pci_dev_data, pci_resource_flags(pdev, 0));
2993 		err = -ENODEV;
2994 		goto err_disable_pdev;
2995 	}
2996 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
2997 		dev_err(&pdev->dev, "Missing UAR, aborting\n");
2998 		err = -ENODEV;
2999 		goto err_disable_pdev;
3000 	}
3001 
3002 	err = pci_request_regions(pdev, DRV_NAME);
3003 	if (err) {
3004 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3005 		goto err_disable_pdev;
3006 	}
3007 
3008 	pci_set_master(pdev);
3009 
3010 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3011 	if (err) {
3012 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3013 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3014 		if (err) {
3015 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3016 			goto err_release_regions;
3017 		}
3018 	}
3019 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3020 	if (err) {
3021 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3022 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3023 		if (err) {
3024 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3025 			goto err_release_regions;
3026 		}
3027 	}
3028 
3029 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
3030 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3031 	/* Detect if this device is a virtual function */
3032 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3033 		/* When acting as pf, we normally skip vfs unless explicitly
3034 		 * requested to probe them.
3035 		 */
3036 		if (total_vfs) {
3037 			unsigned vfs_offset = 0;
3038 
3039 			for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
3040 			     vfs_offset + nvfs[i] < extended_func_num(pdev);
3041 			     vfs_offset += nvfs[i], i++)
3042 				;
3043 			if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
3044 				err = -ENODEV;
3045 				goto err_release_regions;
3046 			}
3047 			if ((extended_func_num(pdev) - vfs_offset)
3048 			    > prb_vf[i]) {
3049 				dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3050 					 extended_func_num(pdev));
3051 				err = -ENODEV;
3052 				goto err_release_regions;
3053 			}
3054 		}
3055 	}
3056 
3057 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv);
3058 	if (err)
3059 		goto err_release_regions;
3060 	return 0;
3061 
3062 err_release_regions:
3063 	pci_release_regions(pdev);
3064 
3065 err_disable_pdev:
3066 	pci_disable_device(pdev);
3067 	pci_set_drvdata(pdev, NULL);
3068 	return err;
3069 }
3070 
3071 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3072 {
3073 	struct mlx4_priv *priv;
3074 	struct mlx4_dev *dev;
3075 	int ret;
3076 
3077 	printk_once(KERN_INFO "%s", mlx4_version);
3078 
3079 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
3080 	if (!priv)
3081 		return -ENOMEM;
3082 
3083 	dev       = &priv->dev;
3084 	dev->pdev = pdev;
3085 	pci_set_drvdata(pdev, dev);
3086 	priv->pci_dev_data = id->driver_data;
3087 
3088 	ret =  __mlx4_init_one(pdev, id->driver_data, priv);
3089 	if (ret)
3090 		kfree(priv);
3091 
3092 	return ret;
3093 }
3094 
3095 static void mlx4_unload_one(struct pci_dev *pdev)
3096 {
3097 	struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
3098 	struct mlx4_priv *priv = mlx4_priv(dev);
3099 	int               pci_dev_data;
3100 	int p;
3101 	int active_vfs = 0;
3102 
3103 	if (priv->removed)
3104 		return;
3105 
3106 	pci_dev_data = priv->pci_dev_data;
3107 
3108 	/* Disabling SR-IOV is not allowed while there are active vf's */
3109 	if (mlx4_is_master(dev)) {
3110 		active_vfs = mlx4_how_many_lives_vf(dev);
3111 		if (active_vfs) {
3112 			pr_warn("Removing PF when there are active VF's !!\n");
3113 			pr_warn("Will not disable SR-IOV.\n");
3114 		}
3115 	}
3116 	mlx4_stop_sense(dev);
3117 	mlx4_unregister_device(dev);
3118 
3119 	for (p = 1; p <= dev->caps.num_ports; p++) {
3120 		mlx4_cleanup_port_info(&priv->port[p]);
3121 		mlx4_CLOSE_PORT(dev, p);
3122 	}
3123 
3124 	if (mlx4_is_master(dev))
3125 		mlx4_free_resource_tracker(dev,
3126 					   RES_TR_FREE_SLAVES_ONLY);
3127 
3128 	mlx4_cleanup_counters_table(dev);
3129 	mlx4_cleanup_qp_table(dev);
3130 	mlx4_cleanup_srq_table(dev);
3131 	mlx4_cleanup_cq_table(dev);
3132 	mlx4_cmd_use_polling(dev);
3133 	mlx4_cleanup_eq_table(dev);
3134 	mlx4_cleanup_mcg_table(dev);
3135 	mlx4_cleanup_mr_table(dev);
3136 	mlx4_cleanup_xrcd_table(dev);
3137 	mlx4_cleanup_pd_table(dev);
3138 
3139 	if (mlx4_is_master(dev))
3140 		mlx4_free_resource_tracker(dev,
3141 					   RES_TR_FREE_STRUCTS_ONLY);
3142 
3143 	iounmap(priv->kar);
3144 	mlx4_uar_free(dev, &priv->driver_uar);
3145 	mlx4_cleanup_uar_table(dev);
3146 	if (!mlx4_is_slave(dev))
3147 		mlx4_clear_steering(dev);
3148 	mlx4_free_eq_table(dev);
3149 	if (mlx4_is_master(dev))
3150 		mlx4_multi_func_cleanup(dev);
3151 	mlx4_close_hca(dev);
3152 	mlx4_close_fw(dev);
3153 	if (mlx4_is_slave(dev))
3154 		mlx4_multi_func_cleanup(dev);
3155 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3156 
3157 	if (dev->flags & MLX4_FLAG_MSI_X)
3158 		pci_disable_msix(pdev);
3159 	if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
3160 		mlx4_warn(dev, "Disabling SR-IOV\n");
3161 		pci_disable_sriov(pdev);
3162 		dev->flags &= ~MLX4_FLAG_SRIOV;
3163 		dev->num_vfs = 0;
3164 	}
3165 
3166 	if (!mlx4_is_slave(dev))
3167 		mlx4_free_ownership(dev);
3168 
3169 	kfree(dev->caps.qp0_qkey);
3170 	kfree(dev->caps.qp0_tunnel);
3171 	kfree(dev->caps.qp0_proxy);
3172 	kfree(dev->caps.qp1_tunnel);
3173 	kfree(dev->caps.qp1_proxy);
3174 	kfree(dev->dev_vfs);
3175 
3176 	memset(priv, 0, sizeof(*priv));
3177 	priv->pci_dev_data = pci_dev_data;
3178 	priv->removed = 1;
3179 }
3180 
3181 static void mlx4_remove_one(struct pci_dev *pdev)
3182 {
3183 	struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
3184 	struct mlx4_priv *priv = mlx4_priv(dev);
3185 
3186 	mlx4_unload_one(pdev);
3187 	pci_release_regions(pdev);
3188 	pci_disable_device(pdev);
3189 	kfree(priv);
3190 	pci_set_drvdata(pdev, NULL);
3191 }
3192 
3193 int mlx4_restart_one(struct pci_dev *pdev)
3194 {
3195 	struct mlx4_dev	 *dev  = pci_get_drvdata(pdev);
3196 	struct mlx4_priv *priv = mlx4_priv(dev);
3197 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3198 	int pci_dev_data, err, total_vfs;
3199 
3200 	pci_dev_data = priv->pci_dev_data;
3201 	total_vfs = dev->num_vfs;
3202 	memcpy(nvfs, dev->nvfs, sizeof(dev->nvfs));
3203 
3204 	mlx4_unload_one(pdev);
3205 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv);
3206 	if (err) {
3207 		mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
3208 			 __func__, pci_name(pdev), err);
3209 		return err;
3210 	}
3211 
3212 	return err;
3213 }
3214 
3215 static const struct pci_device_id mlx4_pci_table[] = {
3216 	/* MT25408 "Hermon" SDR */
3217 	{ PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3218 	/* MT25408 "Hermon" DDR */
3219 	{ PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3220 	/* MT25408 "Hermon" QDR */
3221 	{ PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3222 	/* MT25408 "Hermon" DDR PCIe gen2 */
3223 	{ PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3224 	/* MT25408 "Hermon" QDR PCIe gen2 */
3225 	{ PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3226 	/* MT25408 "Hermon" EN 10GigE */
3227 	{ PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3228 	/* MT25408 "Hermon" EN 10GigE PCIe gen2 */
3229 	{ PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3230 	/* MT25458 ConnectX EN 10GBASE-T 10GigE */
3231 	{ PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3232 	/* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
3233 	{ PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3234 	/* MT26468 ConnectX EN 10GigE PCIe gen2*/
3235 	{ PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3236 	/* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
3237 	{ PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3238 	/* MT26478 ConnectX2 40GigE PCIe gen2 */
3239 	{ PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3240 	/* MT25400 Family [ConnectX-2 Virtual Function] */
3241 	{ PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
3242 	/* MT27500 Family [ConnectX-3] */
3243 	{ PCI_VDEVICE(MELLANOX, 0x1003), 0 },
3244 	/* MT27500 Family [ConnectX-3 Virtual Function] */
3245 	{ PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
3246 	{ PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
3247 	{ PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
3248 	{ PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
3249 	{ PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
3250 	{ PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
3251 	{ PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
3252 	{ PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
3253 	{ PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
3254 	{ PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
3255 	{ PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
3256 	{ PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
3257 	{ PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
3258 	{ 0, }
3259 };
3260 
3261 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
3262 
3263 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
3264 					      pci_channel_state_t state)
3265 {
3266 	mlx4_unload_one(pdev);
3267 
3268 	return state == pci_channel_io_perm_failure ?
3269 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
3270 }
3271 
3272 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
3273 {
3274 	struct mlx4_dev	 *dev  = pci_get_drvdata(pdev);
3275 	struct mlx4_priv *priv = mlx4_priv(dev);
3276 	int               ret;
3277 
3278 	ret = __mlx4_init_one(pdev, priv->pci_dev_data, priv);
3279 
3280 	return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
3281 }
3282 
3283 static const struct pci_error_handlers mlx4_err_handler = {
3284 	.error_detected = mlx4_pci_err_detected,
3285 	.slot_reset     = mlx4_pci_slot_reset,
3286 };
3287 
3288 static struct pci_driver mlx4_driver = {
3289 	.name		= DRV_NAME,
3290 	.id_table	= mlx4_pci_table,
3291 	.probe		= mlx4_init_one,
3292 	.shutdown	= mlx4_unload_one,
3293 	.remove		= mlx4_remove_one,
3294 	.err_handler    = &mlx4_err_handler,
3295 };
3296 
3297 static int __init mlx4_verify_params(void)
3298 {
3299 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
3300 		pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
3301 		return -1;
3302 	}
3303 
3304 	if (log_num_vlan != 0)
3305 		pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
3306 			MLX4_LOG_NUM_VLANS);
3307 
3308 	if (use_prio != 0)
3309 		pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
3310 
3311 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
3312 		pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
3313 			log_mtts_per_seg);
3314 		return -1;
3315 	}
3316 
3317 	/* Check if module param for ports type has legal combination */
3318 	if (port_type_array[0] == false && port_type_array[1] == true) {
3319 		pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
3320 		port_type_array[0] = true;
3321 	}
3322 
3323 	if (mlx4_log_num_mgm_entry_size < -7 ||
3324 	    (mlx4_log_num_mgm_entry_size > 0 &&
3325 	     (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
3326 	      mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
3327 		pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
3328 			mlx4_log_num_mgm_entry_size,
3329 			MLX4_MIN_MGM_LOG_ENTRY_SIZE,
3330 			MLX4_MAX_MGM_LOG_ENTRY_SIZE);
3331 		return -1;
3332 	}
3333 
3334 	return 0;
3335 }
3336 
3337 static int __init mlx4_init(void)
3338 {
3339 	int ret;
3340 
3341 	if (mlx4_verify_params())
3342 		return -EINVAL;
3343 
3344 	mlx4_catas_init();
3345 
3346 	mlx4_wq = create_singlethread_workqueue("mlx4");
3347 	if (!mlx4_wq)
3348 		return -ENOMEM;
3349 
3350 	ret = pci_register_driver(&mlx4_driver);
3351 	if (ret < 0)
3352 		destroy_workqueue(mlx4_wq);
3353 	return ret < 0 ? ret : 0;
3354 }
3355 
3356 static void __exit mlx4_cleanup(void)
3357 {
3358 	pci_unregister_driver(&mlx4_driver);
3359 	destroy_workqueue(mlx4_wq);
3360 }
3361 
3362 module_init(mlx4_init);
3363 module_exit(mlx4_cleanup);
3364