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/kernel.h>
38 #include <linux/init.h>
39 #include <linux/errno.h>
40 #include <linux/pci.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/slab.h>
43 #include <linux/io-mapping.h>
44 #include <linux/delay.h>
45 #include <linux/kmod.h>
46 #include <linux/etherdevice.h>
47 #include <net/devlink.h>
48 
49 #include <uapi/rdma/mlx4-abi.h>
50 #include <linux/mlx4/device.h>
51 #include <linux/mlx4/doorbell.h>
52 
53 #include "mlx4.h"
54 #include "fw.h"
55 #include "icm.h"
56 
57 MODULE_AUTHOR("Roland Dreier");
58 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
59 MODULE_LICENSE("Dual BSD/GPL");
60 MODULE_VERSION(DRV_VERSION);
61 
62 struct workqueue_struct *mlx4_wq;
63 
64 #ifdef CONFIG_MLX4_DEBUG
65 
66 int mlx4_debug_level; /* 0 by default */
67 module_param_named(debug_level, mlx4_debug_level, int, 0644);
68 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
69 
70 #endif /* CONFIG_MLX4_DEBUG */
71 
72 #ifdef CONFIG_PCI_MSI
73 
74 static int msi_x = 1;
75 module_param(msi_x, int, 0444);
76 MODULE_PARM_DESC(msi_x, "0 - don't use MSI-X, 1 - use MSI-X, >1 - limit number of MSI-X irqs to msi_x");
77 
78 #else /* CONFIG_PCI_MSI */
79 
80 #define msi_x (0)
81 
82 #endif /* CONFIG_PCI_MSI */
83 
84 static uint8_t num_vfs[3] = {0, 0, 0};
85 static int num_vfs_argc;
86 module_param_array(num_vfs, byte, &num_vfs_argc, 0444);
87 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
88 			  "num_vfs=port1,port2,port1+2");
89 
90 static uint8_t probe_vf[3] = {0, 0, 0};
91 static int probe_vfs_argc;
92 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
93 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
94 			   "probe_vf=port1,port2,port1+2");
95 
96 static int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
97 module_param_named(log_num_mgm_entry_size,
98 			mlx4_log_num_mgm_entry_size, int, 0444);
99 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
100 					 " of qp per mcg, for example:"
101 					 " 10 gives 248.range: 7 <="
102 					 " log_num_mgm_entry_size <= 12."
103 					 " To activate device managed"
104 					 " flow steering when available, set to -1");
105 
106 static bool enable_64b_cqe_eqe = true;
107 module_param(enable_64b_cqe_eqe, bool, 0444);
108 MODULE_PARM_DESC(enable_64b_cqe_eqe,
109 		 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
110 
111 static bool enable_4k_uar;
112 module_param(enable_4k_uar, bool, 0444);
113 MODULE_PARM_DESC(enable_4k_uar,
114 		 "Enable using 4K UAR. Should not be enabled if have VFs which do not support 4K UARs (default: false)");
115 
116 #define PF_CONTEXT_BEHAVIOUR_MASK	(MLX4_FUNC_CAP_64B_EQE_CQE | \
117 					 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
118 					 MLX4_FUNC_CAP_DMFS_A0_STATIC)
119 
120 #define RESET_PERSIST_MASK_FLAGS	(MLX4_FLAG_SRIOV)
121 
122 static char mlx4_version[] =
123 	DRV_NAME ": Mellanox ConnectX core driver v"
124 	DRV_VERSION "\n";
125 
126 static const struct mlx4_profile default_profile = {
127 	.num_qp		= 1 << 18,
128 	.num_srq	= 1 << 16,
129 	.rdmarc_per_qp	= 1 << 4,
130 	.num_cq		= 1 << 16,
131 	.num_mcg	= 1 << 13,
132 	.num_mpt	= 1 << 19,
133 	.num_mtt	= 1 << 20, /* It is really num mtt segements */
134 };
135 
136 static const struct mlx4_profile low_mem_profile = {
137 	.num_qp		= 1 << 17,
138 	.num_srq	= 1 << 6,
139 	.rdmarc_per_qp	= 1 << 4,
140 	.num_cq		= 1 << 8,
141 	.num_mcg	= 1 << 8,
142 	.num_mpt	= 1 << 9,
143 	.num_mtt	= 1 << 7,
144 };
145 
146 static int log_num_mac = 7;
147 module_param_named(log_num_mac, log_num_mac, int, 0444);
148 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
149 
150 static int log_num_vlan;
151 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
152 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
153 /* Log2 max number of VLANs per ETH port (0-7) */
154 #define MLX4_LOG_NUM_VLANS 7
155 #define MLX4_MIN_LOG_NUM_VLANS 0
156 #define MLX4_MIN_LOG_NUM_MAC 1
157 
158 static bool use_prio;
159 module_param_named(use_prio, use_prio, bool, 0444);
160 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
161 
162 int log_mtts_per_seg = ilog2(1);
163 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
164 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment "
165 		 "(0-7) (default: 0)");
166 
167 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
168 static int arr_argc = 2;
169 module_param_array(port_type_array, int, &arr_argc, 0444);
170 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
171 				"1 for IB, 2 for Ethernet");
172 
173 struct mlx4_port_config {
174 	struct list_head list;
175 	enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
176 	struct pci_dev *pdev;
177 };
178 
179 static atomic_t pf_loading = ATOMIC_INIT(0);
180 
181 static int mlx4_devlink_ierr_reset_get(struct devlink *devlink, u32 id,
182 				       struct devlink_param_gset_ctx *ctx)
183 {
184 	ctx->val.vbool = !!mlx4_internal_err_reset;
185 	return 0;
186 }
187 
188 static int mlx4_devlink_ierr_reset_set(struct devlink *devlink, u32 id,
189 				       struct devlink_param_gset_ctx *ctx)
190 {
191 	mlx4_internal_err_reset = ctx->val.vbool;
192 	return 0;
193 }
194 
195 static int mlx4_devlink_crdump_snapshot_get(struct devlink *devlink, u32 id,
196 					    struct devlink_param_gset_ctx *ctx)
197 {
198 	struct mlx4_priv *priv = devlink_priv(devlink);
199 	struct mlx4_dev *dev = &priv->dev;
200 
201 	ctx->val.vbool = dev->persist->crdump.snapshot_enable;
202 	return 0;
203 }
204 
205 static int mlx4_devlink_crdump_snapshot_set(struct devlink *devlink, u32 id,
206 					    struct devlink_param_gset_ctx *ctx)
207 {
208 	struct mlx4_priv *priv = devlink_priv(devlink);
209 	struct mlx4_dev *dev = &priv->dev;
210 
211 	dev->persist->crdump.snapshot_enable = ctx->val.vbool;
212 	return 0;
213 }
214 
215 static int
216 mlx4_devlink_max_macs_validate(struct devlink *devlink, u32 id,
217 			       union devlink_param_value val,
218 			       struct netlink_ext_ack *extack)
219 {
220 	u32 value = val.vu32;
221 
222 	if (value < 1 || value > 128)
223 		return -ERANGE;
224 
225 	if (!is_power_of_2(value)) {
226 		NL_SET_ERR_MSG_MOD(extack, "max_macs supported must be power of 2");
227 		return -EINVAL;
228 	}
229 
230 	return 0;
231 }
232 
233 enum mlx4_devlink_param_id {
234 	MLX4_DEVLINK_PARAM_ID_BASE = DEVLINK_PARAM_GENERIC_ID_MAX,
235 	MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
236 	MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
237 };
238 
239 static const struct devlink_param mlx4_devlink_params[] = {
240 	DEVLINK_PARAM_GENERIC(INT_ERR_RESET,
241 			      BIT(DEVLINK_PARAM_CMODE_RUNTIME) |
242 			      BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
243 			      mlx4_devlink_ierr_reset_get,
244 			      mlx4_devlink_ierr_reset_set, NULL),
245 	DEVLINK_PARAM_GENERIC(MAX_MACS,
246 			      BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
247 			      NULL, NULL, mlx4_devlink_max_macs_validate),
248 	DEVLINK_PARAM_GENERIC(REGION_SNAPSHOT,
249 			      BIT(DEVLINK_PARAM_CMODE_RUNTIME) |
250 			      BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
251 			      mlx4_devlink_crdump_snapshot_get,
252 			      mlx4_devlink_crdump_snapshot_set, NULL),
253 	DEVLINK_PARAM_DRIVER(MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
254 			     "enable_64b_cqe_eqe", DEVLINK_PARAM_TYPE_BOOL,
255 			     BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
256 			     NULL, NULL, NULL),
257 	DEVLINK_PARAM_DRIVER(MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
258 			     "enable_4k_uar", DEVLINK_PARAM_TYPE_BOOL,
259 			     BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
260 			     NULL, NULL, NULL),
261 };
262 
263 static void mlx4_devlink_set_params_init_values(struct devlink *devlink)
264 {
265 	union devlink_param_value value;
266 
267 	value.vbool = !!mlx4_internal_err_reset;
268 	devlink_param_driverinit_value_set(devlink,
269 					   DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET,
270 					   value);
271 
272 	value.vu32 = 1UL << log_num_mac;
273 	devlink_param_driverinit_value_set(devlink,
274 					   DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
275 					   value);
276 
277 	value.vbool = enable_64b_cqe_eqe;
278 	devlink_param_driverinit_value_set(devlink,
279 					   MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
280 					   value);
281 
282 	value.vbool = enable_4k_uar;
283 	devlink_param_driverinit_value_set(devlink,
284 					   MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
285 					   value);
286 
287 	value.vbool = false;
288 	devlink_param_driverinit_value_set(devlink,
289 					   DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT,
290 					   value);
291 }
292 
293 static inline void mlx4_set_num_reserved_uars(struct mlx4_dev *dev,
294 					      struct mlx4_dev_cap *dev_cap)
295 {
296 	/* The reserved_uars is calculated by system page size unit.
297 	 * Therefore, adjustment is added when the uar page size is less
298 	 * than the system page size
299 	 */
300 	dev->caps.reserved_uars	=
301 		max_t(int,
302 		      mlx4_get_num_reserved_uar(dev),
303 		      dev_cap->reserved_uars /
304 			(1 << (PAGE_SHIFT - dev->uar_page_shift)));
305 }
306 
307 int mlx4_check_port_params(struct mlx4_dev *dev,
308 			   enum mlx4_port_type *port_type)
309 {
310 	int i;
311 
312 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
313 		for (i = 0; i < dev->caps.num_ports - 1; i++) {
314 			if (port_type[i] != port_type[i + 1]) {
315 				mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
316 				return -EOPNOTSUPP;
317 			}
318 		}
319 	}
320 
321 	for (i = 0; i < dev->caps.num_ports; i++) {
322 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
323 			mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
324 				 i + 1);
325 			return -EOPNOTSUPP;
326 		}
327 	}
328 	return 0;
329 }
330 
331 static void mlx4_set_port_mask(struct mlx4_dev *dev)
332 {
333 	int i;
334 
335 	for (i = 1; i <= dev->caps.num_ports; ++i)
336 		dev->caps.port_mask[i] = dev->caps.port_type[i];
337 }
338 
339 enum {
340 	MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
341 };
342 
343 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
344 {
345 	int err = 0;
346 	struct mlx4_func func;
347 
348 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
349 		err = mlx4_QUERY_FUNC(dev, &func, 0);
350 		if (err) {
351 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
352 			return err;
353 		}
354 		dev_cap->max_eqs = func.max_eq;
355 		dev_cap->reserved_eqs = func.rsvd_eqs;
356 		dev_cap->reserved_uars = func.rsvd_uars;
357 		err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
358 	}
359 	return err;
360 }
361 
362 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
363 {
364 	struct mlx4_caps *dev_cap = &dev->caps;
365 
366 	/* FW not supporting or cancelled by user */
367 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
368 	    !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
369 		return;
370 
371 	/* Must have 64B CQE_EQE enabled by FW to use bigger stride
372 	 * When FW has NCSI it may decide not to report 64B CQE/EQEs
373 	 */
374 	if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
375 	    !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
376 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
377 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
378 		return;
379 	}
380 
381 	if (cache_line_size() == 128 || cache_line_size() == 256) {
382 		mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
383 		/* Changing the real data inside CQE size to 32B */
384 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
385 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
386 
387 		if (mlx4_is_master(dev))
388 			dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
389 	} else {
390 		if (cache_line_size() != 32  && cache_line_size() != 64)
391 			mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
392 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
393 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
394 	}
395 }
396 
397 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
398 			  struct mlx4_port_cap *port_cap)
399 {
400 	dev->caps.vl_cap[port]	    = port_cap->max_vl;
401 	dev->caps.ib_mtu_cap[port]	    = port_cap->ib_mtu;
402 	dev->phys_caps.gid_phys_table_len[port]  = port_cap->max_gids;
403 	dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
404 	/* set gid and pkey table operating lengths by default
405 	 * to non-sriov values
406 	 */
407 	dev->caps.gid_table_len[port]  = port_cap->max_gids;
408 	dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
409 	dev->caps.port_width_cap[port] = port_cap->max_port_width;
410 	dev->caps.eth_mtu_cap[port]    = port_cap->eth_mtu;
411 	dev->caps.max_tc_eth	       = port_cap->max_tc_eth;
412 	dev->caps.def_mac[port]        = port_cap->def_mac;
413 	dev->caps.supported_type[port] = port_cap->supported_port_types;
414 	dev->caps.suggested_type[port] = port_cap->suggested_type;
415 	dev->caps.default_sense[port] = port_cap->default_sense;
416 	dev->caps.trans_type[port]	    = port_cap->trans_type;
417 	dev->caps.vendor_oui[port]     = port_cap->vendor_oui;
418 	dev->caps.wavelength[port]     = port_cap->wavelength;
419 	dev->caps.trans_code[port]     = port_cap->trans_code;
420 
421 	return 0;
422 }
423 
424 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
425 			 struct mlx4_port_cap *port_cap)
426 {
427 	int err = 0;
428 
429 	err = mlx4_QUERY_PORT(dev, port, port_cap);
430 
431 	if (err)
432 		mlx4_err(dev, "QUERY_PORT command failed.\n");
433 
434 	return err;
435 }
436 
437 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
438 {
439 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
440 		return;
441 
442 	if (mlx4_is_mfunc(dev)) {
443 		mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
444 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
445 		return;
446 	}
447 
448 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
449 		mlx4_dbg(dev,
450 			 "Keep FCS is not supported - Disabling Ignore FCS");
451 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
452 		return;
453 	}
454 }
455 
456 #define MLX4_A0_STEERING_TABLE_SIZE	256
457 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
458 {
459 	int err;
460 	int i;
461 
462 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
463 	if (err) {
464 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
465 		return err;
466 	}
467 	mlx4_dev_cap_dump(dev, dev_cap);
468 
469 	if (dev_cap->min_page_sz > PAGE_SIZE) {
470 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
471 			 dev_cap->min_page_sz, PAGE_SIZE);
472 		return -ENODEV;
473 	}
474 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
475 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
476 			 dev_cap->num_ports, MLX4_MAX_PORTS);
477 		return -ENODEV;
478 	}
479 
480 	if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
481 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
482 			 dev_cap->uar_size,
483 			 (unsigned long long)
484 			 pci_resource_len(dev->persist->pdev, 2));
485 		return -ENODEV;
486 	}
487 
488 	dev->caps.num_ports	     = dev_cap->num_ports;
489 	dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
490 	dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
491 				      dev->caps.num_sys_eqs :
492 				      MLX4_MAX_EQ_NUM;
493 	for (i = 1; i <= dev->caps.num_ports; ++i) {
494 		err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
495 		if (err) {
496 			mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
497 			return err;
498 		}
499 	}
500 
501 	dev->caps.uar_page_size	     = PAGE_SIZE;
502 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
503 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
504 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
505 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
506 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
507 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
508 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
509 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
510 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
511 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
512 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
513 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
514 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
515 	/*
516 	 * Subtract 1 from the limit because we need to allocate a
517 	 * spare CQE to enable resizing the CQ.
518 	 */
519 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
520 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
521 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
522 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
523 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
524 
525 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
526 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
527 					dev_cap->reserved_xrcds : 0;
528 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
529 					dev_cap->max_xrcds : 0;
530 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
531 
532 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
533 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
534 	dev->caps.flags		     = dev_cap->flags;
535 	dev->caps.flags2	     = dev_cap->flags2;
536 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
537 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
538 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
539 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
540 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
541 	dev->caps.wol_port[1]          = dev_cap->wol_port[1];
542 	dev->caps.wol_port[2]          = dev_cap->wol_port[2];
543 	dev->caps.health_buffer_addrs  = dev_cap->health_buffer_addrs;
544 
545 	/* Save uar page shift */
546 	if (!mlx4_is_slave(dev)) {
547 		/* Virtual PCI function needs to determine UAR page size from
548 		 * firmware. Only master PCI function can set the uar page size
549 		 */
550 		if (enable_4k_uar || !dev->persist->num_vfs)
551 			dev->uar_page_shift = DEFAULT_UAR_PAGE_SHIFT;
552 		else
553 			dev->uar_page_shift = PAGE_SHIFT;
554 
555 		mlx4_set_num_reserved_uars(dev, dev_cap);
556 	}
557 
558 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN) {
559 		struct mlx4_init_hca_param hca_param;
560 
561 		memset(&hca_param, 0, sizeof(hca_param));
562 		err = mlx4_QUERY_HCA(dev, &hca_param);
563 		/* Turn off PHV_EN flag in case phv_check_en is set.
564 		 * phv_check_en is a HW check that parse the packet and verify
565 		 * phv bit was reported correctly in the wqe. To allow QinQ
566 		 * PHV_EN flag should be set and phv_check_en must be cleared
567 		 * otherwise QinQ packets will be drop by the HW.
568 		 */
569 		if (err || hca_param.phv_check_en)
570 			dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_PHV_EN;
571 	}
572 
573 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
574 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
575 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
576 	/* Don't do sense port on multifunction devices (for now at least) */
577 	if (mlx4_is_mfunc(dev))
578 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
579 
580 	if (mlx4_low_memory_profile()) {
581 		dev->caps.log_num_macs  = MLX4_MIN_LOG_NUM_MAC;
582 		dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
583 	} else {
584 		dev->caps.log_num_macs  = log_num_mac;
585 		dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
586 	}
587 
588 	for (i = 1; i <= dev->caps.num_ports; ++i) {
589 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
590 		if (dev->caps.supported_type[i]) {
591 			/* if only ETH is supported - assign ETH */
592 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
593 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
594 			/* if only IB is supported, assign IB */
595 			else if (dev->caps.supported_type[i] ==
596 				 MLX4_PORT_TYPE_IB)
597 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
598 			else {
599 				/* if IB and ETH are supported, we set the port
600 				 * type according to user selection of port type;
601 				 * if user selected none, take the FW hint */
602 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
603 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
604 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
605 				else
606 					dev->caps.port_type[i] = port_type_array[i - 1];
607 			}
608 		}
609 		/*
610 		 * Link sensing is allowed on the port if 3 conditions are true:
611 		 * 1. Both protocols are supported on the port.
612 		 * 2. Different types are supported on the port
613 		 * 3. FW declared that it supports link sensing
614 		 */
615 		mlx4_priv(dev)->sense.sense_allowed[i] =
616 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
617 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
618 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
619 
620 		/*
621 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
622 		 * and perform sense_port FW command to try and set the correct
623 		 * port type from beginning
624 		 */
625 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
626 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
627 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
628 			mlx4_SENSE_PORT(dev, i, &sensed_port);
629 			if (sensed_port != MLX4_PORT_TYPE_NONE)
630 				dev->caps.port_type[i] = sensed_port;
631 		} else {
632 			dev->caps.possible_type[i] = dev->caps.port_type[i];
633 		}
634 
635 		if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
636 			dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
637 			mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
638 				  i, 1 << dev->caps.log_num_macs);
639 		}
640 		if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
641 			dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
642 			mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
643 				  i, 1 << dev->caps.log_num_vlans);
644 		}
645 	}
646 
647 	if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
648 	    (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
649 	    (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
650 		mlx4_warn(dev,
651 			  "Granular QoS per VF not supported with IB/Eth configuration\n");
652 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
653 	}
654 
655 	dev->caps.max_counters = dev_cap->max_counters;
656 
657 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
658 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
659 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
660 		(1 << dev->caps.log_num_macs) *
661 		(1 << dev->caps.log_num_vlans) *
662 		dev->caps.num_ports;
663 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
664 
665 	if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
666 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
667 		dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
668 	else
669 		dev->caps.dmfs_high_rate_qpn_base =
670 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
671 
672 	if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
673 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
674 		dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
675 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
676 		dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
677 	} else {
678 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
679 		dev->caps.dmfs_high_rate_qpn_base =
680 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
681 		dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
682 	}
683 
684 	dev->caps.rl_caps = dev_cap->rl_caps;
685 
686 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
687 		dev->caps.dmfs_high_rate_qpn_range;
688 
689 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
690 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
691 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
692 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
693 
694 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
695 
696 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
697 		if (dev_cap->flags &
698 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
699 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
700 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
701 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
702 		}
703 
704 		if (dev_cap->flags2 &
705 		    (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
706 		     MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
707 			mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
708 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
709 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
710 		}
711 	}
712 
713 	if ((dev->caps.flags &
714 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
715 	    mlx4_is_master(dev))
716 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
717 
718 	if (!mlx4_is_slave(dev)) {
719 		mlx4_enable_cqe_eqe_stride(dev);
720 		dev->caps.alloc_res_qp_mask =
721 			(dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
722 			MLX4_RESERVE_A0_QP;
723 
724 		if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
725 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
726 			mlx4_warn(dev, "Old device ETS support detected\n");
727 			mlx4_warn(dev, "Consider upgrading device FW.\n");
728 			dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
729 		}
730 
731 	} else {
732 		dev->caps.alloc_res_qp_mask = 0;
733 	}
734 
735 	mlx4_enable_ignore_fcs(dev);
736 
737 	return 0;
738 }
739 
740 /*The function checks if there are live vf, return the num of them*/
741 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
742 {
743 	struct mlx4_priv *priv = mlx4_priv(dev);
744 	struct mlx4_slave_state *s_state;
745 	int i;
746 	int ret = 0;
747 
748 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
749 		s_state = &priv->mfunc.master.slave_state[i];
750 		if (s_state->active && s_state->last_cmd !=
751 		    MLX4_COMM_CMD_RESET) {
752 			mlx4_warn(dev, "%s: slave: %d is still active\n",
753 				  __func__, i);
754 			ret++;
755 		}
756 	}
757 	return ret;
758 }
759 
760 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
761 {
762 	u32 qk = MLX4_RESERVED_QKEY_BASE;
763 
764 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
765 	    qpn < dev->phys_caps.base_proxy_sqpn)
766 		return -EINVAL;
767 
768 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
769 		/* tunnel qp */
770 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
771 	else
772 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
773 	*qkey = qk;
774 	return 0;
775 }
776 EXPORT_SYMBOL(mlx4_get_parav_qkey);
777 
778 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
779 {
780 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
781 
782 	if (!mlx4_is_master(dev))
783 		return;
784 
785 	priv->virt2phys_pkey[slave][port - 1][i] = val;
786 }
787 EXPORT_SYMBOL(mlx4_sync_pkey_table);
788 
789 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
790 {
791 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
792 
793 	if (!mlx4_is_master(dev))
794 		return;
795 
796 	priv->slave_node_guids[slave] = guid;
797 }
798 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
799 
800 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
801 {
802 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
803 
804 	if (!mlx4_is_master(dev))
805 		return 0;
806 
807 	return priv->slave_node_guids[slave];
808 }
809 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
810 
811 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
812 {
813 	struct mlx4_priv *priv = mlx4_priv(dev);
814 	struct mlx4_slave_state *s_slave;
815 
816 	if (!mlx4_is_master(dev))
817 		return 0;
818 
819 	s_slave = &priv->mfunc.master.slave_state[slave];
820 	return !!s_slave->active;
821 }
822 EXPORT_SYMBOL(mlx4_is_slave_active);
823 
824 void mlx4_handle_eth_header_mcast_prio(struct mlx4_net_trans_rule_hw_ctrl *ctrl,
825 				       struct _rule_hw *eth_header)
826 {
827 	if (is_multicast_ether_addr(eth_header->eth.dst_mac) ||
828 	    is_broadcast_ether_addr(eth_header->eth.dst_mac)) {
829 		struct mlx4_net_trans_rule_hw_eth *eth =
830 			(struct mlx4_net_trans_rule_hw_eth *)eth_header;
831 		struct _rule_hw *next_rule = (struct _rule_hw *)(eth + 1);
832 		bool last_rule = next_rule->size == 0 && next_rule->id == 0 &&
833 			next_rule->rsvd == 0;
834 
835 		if (last_rule)
836 			ctrl->prio = cpu_to_be16(MLX4_DOMAIN_NIC);
837 	}
838 }
839 EXPORT_SYMBOL(mlx4_handle_eth_header_mcast_prio);
840 
841 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
842 				       struct mlx4_dev_cap *dev_cap,
843 				       struct mlx4_init_hca_param *hca_param)
844 {
845 	dev->caps.steering_mode = hca_param->steering_mode;
846 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
847 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
848 		dev->caps.fs_log_max_ucast_qp_range_size =
849 			dev_cap->fs_log_max_ucast_qp_range_size;
850 	} else
851 		dev->caps.num_qp_per_mgm =
852 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
853 
854 	mlx4_dbg(dev, "Steering mode is: %s\n",
855 		 mlx4_steering_mode_str(dev->caps.steering_mode));
856 }
857 
858 static void mlx4_slave_destroy_special_qp_cap(struct mlx4_dev *dev)
859 {
860 	kfree(dev->caps.spec_qps);
861 	dev->caps.spec_qps = NULL;
862 }
863 
864 static int mlx4_slave_special_qp_cap(struct mlx4_dev *dev)
865 {
866 	struct mlx4_func_cap *func_cap = NULL;
867 	struct mlx4_caps *caps = &dev->caps;
868 	int i, err = 0;
869 
870 	func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
871 	caps->spec_qps = kcalloc(caps->num_ports, sizeof(*caps->spec_qps), GFP_KERNEL);
872 
873 	if (!func_cap || !caps->spec_qps) {
874 		mlx4_err(dev, "Failed to allocate memory for special qps cap\n");
875 		err = -ENOMEM;
876 		goto err_mem;
877 	}
878 
879 	for (i = 1; i <= caps->num_ports; ++i) {
880 		err = mlx4_QUERY_FUNC_CAP(dev, i, func_cap);
881 		if (err) {
882 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
883 				 i, err);
884 			goto err_mem;
885 		}
886 		caps->spec_qps[i - 1] = func_cap->spec_qps;
887 		caps->port_mask[i] = caps->port_type[i];
888 		caps->phys_port_id[i] = func_cap->phys_port_id;
889 		err = mlx4_get_slave_pkey_gid_tbl_len(dev, i,
890 						      &caps->gid_table_len[i],
891 						      &caps->pkey_table_len[i]);
892 		if (err) {
893 			mlx4_err(dev, "QUERY_PORT command failed for port %d, aborting (%d)\n",
894 				 i, err);
895 			goto err_mem;
896 		}
897 	}
898 
899 err_mem:
900 	if (err)
901 		mlx4_slave_destroy_special_qp_cap(dev);
902 	kfree(func_cap);
903 	return err;
904 }
905 
906 static int mlx4_slave_cap(struct mlx4_dev *dev)
907 {
908 	int			   err;
909 	u32			   page_size;
910 	struct mlx4_dev_cap	   *dev_cap = NULL;
911 	struct mlx4_func_cap	   *func_cap = NULL;
912 	struct mlx4_init_hca_param *hca_param = NULL;
913 
914 	hca_param = kzalloc(sizeof(*hca_param), GFP_KERNEL);
915 	func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
916 	dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
917 	if (!hca_param || !func_cap || !dev_cap) {
918 		mlx4_err(dev, "Failed to allocate memory for slave_cap\n");
919 		err = -ENOMEM;
920 		goto free_mem;
921 	}
922 
923 	err = mlx4_QUERY_HCA(dev, hca_param);
924 	if (err) {
925 		mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
926 		goto free_mem;
927 	}
928 
929 	/* fail if the hca has an unknown global capability
930 	 * at this time global_caps should be always zeroed
931 	 */
932 	if (hca_param->global_caps) {
933 		mlx4_err(dev, "Unknown hca global capabilities\n");
934 		err = -EINVAL;
935 		goto free_mem;
936 	}
937 
938 	dev->caps.hca_core_clock = hca_param->hca_core_clock;
939 
940 	dev->caps.max_qp_dest_rdma = 1 << hca_param->log_rd_per_qp;
941 	err = mlx4_dev_cap(dev, dev_cap);
942 	if (err) {
943 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
944 		goto free_mem;
945 	}
946 
947 	err = mlx4_QUERY_FW(dev);
948 	if (err)
949 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
950 
951 	page_size = ~dev->caps.page_size_cap + 1;
952 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
953 	if (page_size > PAGE_SIZE) {
954 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
955 			 page_size, PAGE_SIZE);
956 		err = -ENODEV;
957 		goto free_mem;
958 	}
959 
960 	/* Set uar_page_shift for VF */
961 	dev->uar_page_shift = hca_param->uar_page_sz + 12;
962 
963 	/* Make sure the master uar page size is valid */
964 	if (dev->uar_page_shift > PAGE_SHIFT) {
965 		mlx4_err(dev,
966 			 "Invalid configuration: uar page size is larger than system page size\n");
967 		err = -ENODEV;
968 		goto free_mem;
969 	}
970 
971 	/* Set reserved_uars based on the uar_page_shift */
972 	mlx4_set_num_reserved_uars(dev, dev_cap);
973 
974 	/* Although uar page size in FW differs from system page size,
975 	 * upper software layers (mlx4_ib, mlx4_en and part of mlx4_core)
976 	 * still works with assumption that uar page size == system page size
977 	 */
978 	dev->caps.uar_page_size = PAGE_SIZE;
979 
980 	err = mlx4_QUERY_FUNC_CAP(dev, 0, func_cap);
981 	if (err) {
982 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
983 			 err);
984 		goto free_mem;
985 	}
986 
987 	if ((func_cap->pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
988 	    PF_CONTEXT_BEHAVIOUR_MASK) {
989 		mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
990 			 func_cap->pf_context_behaviour,
991 			 PF_CONTEXT_BEHAVIOUR_MASK);
992 		err = -EINVAL;
993 		goto free_mem;
994 	}
995 
996 	dev->caps.num_ports		= func_cap->num_ports;
997 	dev->quotas.qp			= func_cap->qp_quota;
998 	dev->quotas.srq			= func_cap->srq_quota;
999 	dev->quotas.cq			= func_cap->cq_quota;
1000 	dev->quotas.mpt			= func_cap->mpt_quota;
1001 	dev->quotas.mtt			= func_cap->mtt_quota;
1002 	dev->caps.num_qps		= 1 << hca_param->log_num_qps;
1003 	dev->caps.num_srqs		= 1 << hca_param->log_num_srqs;
1004 	dev->caps.num_cqs		= 1 << hca_param->log_num_cqs;
1005 	dev->caps.num_mpts		= 1 << hca_param->log_mpt_sz;
1006 	dev->caps.num_eqs		= func_cap->max_eq;
1007 	dev->caps.reserved_eqs		= func_cap->reserved_eq;
1008 	dev->caps.reserved_lkey		= func_cap->reserved_lkey;
1009 	dev->caps.num_pds               = MLX4_NUM_PDS;
1010 	dev->caps.num_mgms              = 0;
1011 	dev->caps.num_amgms             = 0;
1012 
1013 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
1014 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
1015 			 dev->caps.num_ports, MLX4_MAX_PORTS);
1016 		err = -ENODEV;
1017 		goto free_mem;
1018 	}
1019 
1020 	mlx4_replace_zero_macs(dev);
1021 
1022 	err = mlx4_slave_special_qp_cap(dev);
1023 	if (err) {
1024 		mlx4_err(dev, "Set special QP caps failed. aborting\n");
1025 		goto free_mem;
1026 	}
1027 
1028 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
1029 				       dev->caps.reserved_uars) >
1030 				       pci_resource_len(dev->persist->pdev,
1031 							2)) {
1032 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
1033 			 dev->caps.uar_page_size * dev->caps.num_uars,
1034 			 (unsigned long long)
1035 			 pci_resource_len(dev->persist->pdev, 2));
1036 		err = -ENOMEM;
1037 		goto err_mem;
1038 	}
1039 
1040 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
1041 		dev->caps.eqe_size   = 64;
1042 		dev->caps.eqe_factor = 1;
1043 	} else {
1044 		dev->caps.eqe_size   = 32;
1045 		dev->caps.eqe_factor = 0;
1046 	}
1047 
1048 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
1049 		dev->caps.cqe_size   = 64;
1050 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1051 	} else {
1052 		dev->caps.cqe_size   = 32;
1053 	}
1054 
1055 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
1056 		dev->caps.eqe_size = hca_param->eqe_size;
1057 		dev->caps.eqe_factor = 0;
1058 	}
1059 
1060 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
1061 		dev->caps.cqe_size = hca_param->cqe_size;
1062 		/* User still need to know when CQE > 32B */
1063 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1064 	}
1065 
1066 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1067 	mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
1068 
1069 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_USER_MAC_EN;
1070 	mlx4_dbg(dev, "User MAC FW update is not supported in slave mode\n");
1071 
1072 	slave_adjust_steering_mode(dev, dev_cap, hca_param);
1073 	mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
1074 		 hca_param->rss_ip_frags ? "on" : "off");
1075 
1076 	if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
1077 	    dev->caps.bf_reg_size)
1078 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
1079 
1080 	if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
1081 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
1082 
1083 err_mem:
1084 	if (err)
1085 		mlx4_slave_destroy_special_qp_cap(dev);
1086 free_mem:
1087 	kfree(hca_param);
1088 	kfree(func_cap);
1089 	kfree(dev_cap);
1090 	return err;
1091 }
1092 
1093 static void mlx4_request_modules(struct mlx4_dev *dev)
1094 {
1095 	int port;
1096 	int has_ib_port = false;
1097 	int has_eth_port = false;
1098 #define EN_DRV_NAME	"mlx4_en"
1099 #define IB_DRV_NAME	"mlx4_ib"
1100 
1101 	for (port = 1; port <= dev->caps.num_ports; port++) {
1102 		if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
1103 			has_ib_port = true;
1104 		else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
1105 			has_eth_port = true;
1106 	}
1107 
1108 	if (has_eth_port)
1109 		request_module_nowait(EN_DRV_NAME);
1110 	if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
1111 		request_module_nowait(IB_DRV_NAME);
1112 }
1113 
1114 /*
1115  * Change the port configuration of the device.
1116  * Every user of this function must hold the port mutex.
1117  */
1118 int mlx4_change_port_types(struct mlx4_dev *dev,
1119 			   enum mlx4_port_type *port_types)
1120 {
1121 	int err = 0;
1122 	int change = 0;
1123 	int port;
1124 
1125 	for (port = 0; port <  dev->caps.num_ports; port++) {
1126 		/* Change the port type only if the new type is different
1127 		 * from the current, and not set to Auto */
1128 		if (port_types[port] != dev->caps.port_type[port + 1])
1129 			change = 1;
1130 	}
1131 	if (change) {
1132 		mlx4_unregister_device(dev);
1133 		for (port = 1; port <= dev->caps.num_ports; port++) {
1134 			mlx4_CLOSE_PORT(dev, port);
1135 			dev->caps.port_type[port] = port_types[port - 1];
1136 			err = mlx4_SET_PORT(dev, port, -1);
1137 			if (err) {
1138 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1139 					 port);
1140 				goto out;
1141 			}
1142 		}
1143 		mlx4_set_port_mask(dev);
1144 		err = mlx4_register_device(dev);
1145 		if (err) {
1146 			mlx4_err(dev, "Failed to register device\n");
1147 			goto out;
1148 		}
1149 		mlx4_request_modules(dev);
1150 	}
1151 
1152 out:
1153 	return err;
1154 }
1155 
1156 static ssize_t show_port_type(struct device *dev,
1157 			      struct device_attribute *attr,
1158 			      char *buf)
1159 {
1160 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1161 						   port_attr);
1162 	struct mlx4_dev *mdev = info->dev;
1163 	char type[8];
1164 
1165 	sprintf(type, "%s",
1166 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1167 		"ib" : "eth");
1168 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1169 		sprintf(buf, "auto (%s)\n", type);
1170 	else
1171 		sprintf(buf, "%s\n", type);
1172 
1173 	return strlen(buf);
1174 }
1175 
1176 static int __set_port_type(struct mlx4_port_info *info,
1177 			   enum mlx4_port_type port_type)
1178 {
1179 	struct mlx4_dev *mdev = info->dev;
1180 	struct mlx4_priv *priv = mlx4_priv(mdev);
1181 	enum mlx4_port_type types[MLX4_MAX_PORTS];
1182 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1183 	int i;
1184 	int err = 0;
1185 
1186 	if ((port_type & mdev->caps.supported_type[info->port]) != port_type) {
1187 		mlx4_err(mdev,
1188 			 "Requested port type for port %d is not supported on this HCA\n",
1189 			 info->port);
1190 		return -EOPNOTSUPP;
1191 	}
1192 
1193 	mlx4_stop_sense(mdev);
1194 	mutex_lock(&priv->port_mutex);
1195 	info->tmp_type = port_type;
1196 
1197 	/* Possible type is always the one that was delivered */
1198 	mdev->caps.possible_type[info->port] = info->tmp_type;
1199 
1200 	for (i = 0; i < mdev->caps.num_ports; i++) {
1201 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1202 					mdev->caps.possible_type[i+1];
1203 		if (types[i] == MLX4_PORT_TYPE_AUTO)
1204 			types[i] = mdev->caps.port_type[i+1];
1205 	}
1206 
1207 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1208 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1209 		for (i = 1; i <= mdev->caps.num_ports; i++) {
1210 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1211 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1212 				err = -EOPNOTSUPP;
1213 			}
1214 		}
1215 	}
1216 	if (err) {
1217 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1218 		goto out;
1219 	}
1220 
1221 	mlx4_do_sense_ports(mdev, new_types, types);
1222 
1223 	err = mlx4_check_port_params(mdev, new_types);
1224 	if (err)
1225 		goto out;
1226 
1227 	/* We are about to apply the changes after the configuration
1228 	 * was verified, no need to remember the temporary types
1229 	 * any more */
1230 	for (i = 0; i < mdev->caps.num_ports; i++)
1231 		priv->port[i + 1].tmp_type = 0;
1232 
1233 	err = mlx4_change_port_types(mdev, new_types);
1234 
1235 out:
1236 	mlx4_start_sense(mdev);
1237 	mutex_unlock(&priv->port_mutex);
1238 
1239 	return err;
1240 }
1241 
1242 static ssize_t set_port_type(struct device *dev,
1243 			     struct device_attribute *attr,
1244 			     const char *buf, size_t count)
1245 {
1246 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1247 						   port_attr);
1248 	struct mlx4_dev *mdev = info->dev;
1249 	enum mlx4_port_type port_type;
1250 	static DEFINE_MUTEX(set_port_type_mutex);
1251 	int err;
1252 
1253 	mutex_lock(&set_port_type_mutex);
1254 
1255 	if (!strcmp(buf, "ib\n")) {
1256 		port_type = MLX4_PORT_TYPE_IB;
1257 	} else if (!strcmp(buf, "eth\n")) {
1258 		port_type = MLX4_PORT_TYPE_ETH;
1259 	} else if (!strcmp(buf, "auto\n")) {
1260 		port_type = MLX4_PORT_TYPE_AUTO;
1261 	} else {
1262 		mlx4_err(mdev, "%s is not supported port type\n", buf);
1263 		err = -EINVAL;
1264 		goto err_out;
1265 	}
1266 
1267 	err = __set_port_type(info, port_type);
1268 
1269 err_out:
1270 	mutex_unlock(&set_port_type_mutex);
1271 
1272 	return err ? err : count;
1273 }
1274 
1275 enum ibta_mtu {
1276 	IB_MTU_256  = 1,
1277 	IB_MTU_512  = 2,
1278 	IB_MTU_1024 = 3,
1279 	IB_MTU_2048 = 4,
1280 	IB_MTU_4096 = 5
1281 };
1282 
1283 static inline int int_to_ibta_mtu(int mtu)
1284 {
1285 	switch (mtu) {
1286 	case 256:  return IB_MTU_256;
1287 	case 512:  return IB_MTU_512;
1288 	case 1024: return IB_MTU_1024;
1289 	case 2048: return IB_MTU_2048;
1290 	case 4096: return IB_MTU_4096;
1291 	default: return -1;
1292 	}
1293 }
1294 
1295 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1296 {
1297 	switch (mtu) {
1298 	case IB_MTU_256:  return  256;
1299 	case IB_MTU_512:  return  512;
1300 	case IB_MTU_1024: return 1024;
1301 	case IB_MTU_2048: return 2048;
1302 	case IB_MTU_4096: return 4096;
1303 	default: return -1;
1304 	}
1305 }
1306 
1307 static ssize_t show_port_ib_mtu(struct device *dev,
1308 			     struct device_attribute *attr,
1309 			     char *buf)
1310 {
1311 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1312 						   port_mtu_attr);
1313 	struct mlx4_dev *mdev = info->dev;
1314 
1315 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1316 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1317 
1318 	sprintf(buf, "%d\n",
1319 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1320 	return strlen(buf);
1321 }
1322 
1323 static ssize_t set_port_ib_mtu(struct device *dev,
1324 			     struct device_attribute *attr,
1325 			     const char *buf, size_t count)
1326 {
1327 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1328 						   port_mtu_attr);
1329 	struct mlx4_dev *mdev = info->dev;
1330 	struct mlx4_priv *priv = mlx4_priv(mdev);
1331 	int err, port, mtu, ibta_mtu = -1;
1332 
1333 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1334 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1335 		return -EINVAL;
1336 	}
1337 
1338 	err = kstrtoint(buf, 0, &mtu);
1339 	if (!err)
1340 		ibta_mtu = int_to_ibta_mtu(mtu);
1341 
1342 	if (err || ibta_mtu < 0) {
1343 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1344 		return -EINVAL;
1345 	}
1346 
1347 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1348 
1349 	mlx4_stop_sense(mdev);
1350 	mutex_lock(&priv->port_mutex);
1351 	mlx4_unregister_device(mdev);
1352 	for (port = 1; port <= mdev->caps.num_ports; port++) {
1353 		mlx4_CLOSE_PORT(mdev, port);
1354 		err = mlx4_SET_PORT(mdev, port, -1);
1355 		if (err) {
1356 			mlx4_err(mdev, "Failed to set port %d, aborting\n",
1357 				 port);
1358 			goto err_set_port;
1359 		}
1360 	}
1361 	err = mlx4_register_device(mdev);
1362 err_set_port:
1363 	mutex_unlock(&priv->port_mutex);
1364 	mlx4_start_sense(mdev);
1365 	return err ? err : count;
1366 }
1367 
1368 /* bond for multi-function device */
1369 #define MAX_MF_BOND_ALLOWED_SLAVES 63
1370 static int mlx4_mf_bond(struct mlx4_dev *dev)
1371 {
1372 	int err = 0;
1373 	int nvfs;
1374 	struct mlx4_slaves_pport slaves_port1;
1375 	struct mlx4_slaves_pport slaves_port2;
1376 	DECLARE_BITMAP(slaves_port_1_2, MLX4_MFUNC_MAX);
1377 
1378 	slaves_port1 = mlx4_phys_to_slaves_pport(dev, 1);
1379 	slaves_port2 = mlx4_phys_to_slaves_pport(dev, 2);
1380 	bitmap_and(slaves_port_1_2,
1381 		   slaves_port1.slaves, slaves_port2.slaves,
1382 		   dev->persist->num_vfs + 1);
1383 
1384 	/* only single port vfs are allowed */
1385 	if (bitmap_weight(slaves_port_1_2, dev->persist->num_vfs + 1) > 1) {
1386 		mlx4_warn(dev, "HA mode unsupported for dual ported VFs\n");
1387 		return -EINVAL;
1388 	}
1389 
1390 	/* number of virtual functions is number of total functions minus one
1391 	 * physical function for each port.
1392 	 */
1393 	nvfs = bitmap_weight(slaves_port1.slaves, dev->persist->num_vfs + 1) +
1394 		bitmap_weight(slaves_port2.slaves, dev->persist->num_vfs + 1) - 2;
1395 
1396 	/* limit on maximum allowed VFs */
1397 	if (nvfs > MAX_MF_BOND_ALLOWED_SLAVES) {
1398 		mlx4_warn(dev, "HA mode is not supported for %d VFs (max %d are allowed)\n",
1399 			  nvfs, MAX_MF_BOND_ALLOWED_SLAVES);
1400 		return -EINVAL;
1401 	}
1402 
1403 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) {
1404 		mlx4_warn(dev, "HA mode unsupported for NON DMFS steering\n");
1405 		return -EINVAL;
1406 	}
1407 
1408 	err = mlx4_bond_mac_table(dev);
1409 	if (err)
1410 		return err;
1411 	err = mlx4_bond_vlan_table(dev);
1412 	if (err)
1413 		goto err1;
1414 	err = mlx4_bond_fs_rules(dev);
1415 	if (err)
1416 		goto err2;
1417 
1418 	return 0;
1419 err2:
1420 	(void)mlx4_unbond_vlan_table(dev);
1421 err1:
1422 	(void)mlx4_unbond_mac_table(dev);
1423 	return err;
1424 }
1425 
1426 static int mlx4_mf_unbond(struct mlx4_dev *dev)
1427 {
1428 	int ret, ret1;
1429 
1430 	ret = mlx4_unbond_fs_rules(dev);
1431 	if (ret)
1432 		mlx4_warn(dev, "multifunction unbond for flow rules failed (%d)\n", ret);
1433 	ret1 = mlx4_unbond_mac_table(dev);
1434 	if (ret1) {
1435 		mlx4_warn(dev, "multifunction unbond for MAC table failed (%d)\n", ret1);
1436 		ret = ret1;
1437 	}
1438 	ret1 = mlx4_unbond_vlan_table(dev);
1439 	if (ret1) {
1440 		mlx4_warn(dev, "multifunction unbond for VLAN table failed (%d)\n", ret1);
1441 		ret = ret1;
1442 	}
1443 	return ret;
1444 }
1445 
1446 int mlx4_bond(struct mlx4_dev *dev)
1447 {
1448 	int ret = 0;
1449 	struct mlx4_priv *priv = mlx4_priv(dev);
1450 
1451 	mutex_lock(&priv->bond_mutex);
1452 
1453 	if (!mlx4_is_bonded(dev)) {
1454 		ret = mlx4_do_bond(dev, true);
1455 		if (ret)
1456 			mlx4_err(dev, "Failed to bond device: %d\n", ret);
1457 		if (!ret && mlx4_is_master(dev)) {
1458 			ret = mlx4_mf_bond(dev);
1459 			if (ret) {
1460 				mlx4_err(dev, "bond for multifunction failed\n");
1461 				mlx4_do_bond(dev, false);
1462 			}
1463 		}
1464 	}
1465 
1466 	mutex_unlock(&priv->bond_mutex);
1467 	if (!ret)
1468 		mlx4_dbg(dev, "Device is bonded\n");
1469 
1470 	return ret;
1471 }
1472 EXPORT_SYMBOL_GPL(mlx4_bond);
1473 
1474 int mlx4_unbond(struct mlx4_dev *dev)
1475 {
1476 	int ret = 0;
1477 	struct mlx4_priv *priv = mlx4_priv(dev);
1478 
1479 	mutex_lock(&priv->bond_mutex);
1480 
1481 	if (mlx4_is_bonded(dev)) {
1482 		int ret2 = 0;
1483 
1484 		ret = mlx4_do_bond(dev, false);
1485 		if (ret)
1486 			mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1487 		if (mlx4_is_master(dev))
1488 			ret2 = mlx4_mf_unbond(dev);
1489 		if (ret2) {
1490 			mlx4_warn(dev, "Failed to unbond device for multifunction (%d)\n", ret2);
1491 			ret = ret2;
1492 		}
1493 	}
1494 
1495 	mutex_unlock(&priv->bond_mutex);
1496 	if (!ret)
1497 		mlx4_dbg(dev, "Device is unbonded\n");
1498 
1499 	return ret;
1500 }
1501 EXPORT_SYMBOL_GPL(mlx4_unbond);
1502 
1503 
1504 int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1505 {
1506 	u8 port1 = v2p->port1;
1507 	u8 port2 = v2p->port2;
1508 	struct mlx4_priv *priv = mlx4_priv(dev);
1509 	int err;
1510 
1511 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1512 		return -EOPNOTSUPP;
1513 
1514 	mutex_lock(&priv->bond_mutex);
1515 
1516 	/* zero means keep current mapping for this port */
1517 	if (port1 == 0)
1518 		port1 = priv->v2p.port1;
1519 	if (port2 == 0)
1520 		port2 = priv->v2p.port2;
1521 
1522 	if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1523 	    (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1524 	    (port1 == 2 && port2 == 1)) {
1525 		/* besides boundary checks cross mapping makes
1526 		 * no sense and therefore not allowed */
1527 		err = -EINVAL;
1528 	} else if ((port1 == priv->v2p.port1) &&
1529 		 (port2 == priv->v2p.port2)) {
1530 		err = 0;
1531 	} else {
1532 		err = mlx4_virt2phy_port_map(dev, port1, port2);
1533 		if (!err) {
1534 			mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1535 				 port1, port2);
1536 			priv->v2p.port1 = port1;
1537 			priv->v2p.port2 = port2;
1538 		} else {
1539 			mlx4_err(dev, "Failed to change port mape: %d\n", err);
1540 		}
1541 	}
1542 
1543 	mutex_unlock(&priv->bond_mutex);
1544 	return err;
1545 }
1546 EXPORT_SYMBOL_GPL(mlx4_port_map_set);
1547 
1548 static int mlx4_load_fw(struct mlx4_dev *dev)
1549 {
1550 	struct mlx4_priv *priv = mlx4_priv(dev);
1551 	int err;
1552 
1553 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1554 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
1555 	if (!priv->fw.fw_icm) {
1556 		mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1557 		return -ENOMEM;
1558 	}
1559 
1560 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1561 	if (err) {
1562 		mlx4_err(dev, "MAP_FA command failed, aborting\n");
1563 		goto err_free;
1564 	}
1565 
1566 	err = mlx4_RUN_FW(dev);
1567 	if (err) {
1568 		mlx4_err(dev, "RUN_FW command failed, aborting\n");
1569 		goto err_unmap_fa;
1570 	}
1571 
1572 	return 0;
1573 
1574 err_unmap_fa:
1575 	mlx4_UNMAP_FA(dev);
1576 
1577 err_free:
1578 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1579 	return err;
1580 }
1581 
1582 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1583 				int cmpt_entry_sz)
1584 {
1585 	struct mlx4_priv *priv = mlx4_priv(dev);
1586 	int err;
1587 	int num_eqs;
1588 
1589 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1590 				  cmpt_base +
1591 				  ((u64) (MLX4_CMPT_TYPE_QP *
1592 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1593 				  cmpt_entry_sz, dev->caps.num_qps,
1594 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1595 				  0, 0);
1596 	if (err)
1597 		goto err;
1598 
1599 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1600 				  cmpt_base +
1601 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1602 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1603 				  cmpt_entry_sz, dev->caps.num_srqs,
1604 				  dev->caps.reserved_srqs, 0, 0);
1605 	if (err)
1606 		goto err_qp;
1607 
1608 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1609 				  cmpt_base +
1610 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1611 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1612 				  cmpt_entry_sz, dev->caps.num_cqs,
1613 				  dev->caps.reserved_cqs, 0, 0);
1614 	if (err)
1615 		goto err_srq;
1616 
1617 	num_eqs = dev->phys_caps.num_phys_eqs;
1618 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1619 				  cmpt_base +
1620 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1621 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1622 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1623 	if (err)
1624 		goto err_cq;
1625 
1626 	return 0;
1627 
1628 err_cq:
1629 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1630 
1631 err_srq:
1632 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1633 
1634 err_qp:
1635 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1636 
1637 err:
1638 	return err;
1639 }
1640 
1641 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1642 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1643 {
1644 	struct mlx4_priv *priv = mlx4_priv(dev);
1645 	u64 aux_pages;
1646 	int num_eqs;
1647 	int err;
1648 
1649 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1650 	if (err) {
1651 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1652 		return err;
1653 	}
1654 
1655 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1656 		 (unsigned long long) icm_size >> 10,
1657 		 (unsigned long long) aux_pages << 2);
1658 
1659 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1660 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1661 	if (!priv->fw.aux_icm) {
1662 		mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1663 		return -ENOMEM;
1664 	}
1665 
1666 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1667 	if (err) {
1668 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1669 		goto err_free_aux;
1670 	}
1671 
1672 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1673 	if (err) {
1674 		mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1675 		goto err_unmap_aux;
1676 	}
1677 
1678 
1679 	num_eqs = dev->phys_caps.num_phys_eqs;
1680 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1681 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1682 				  num_eqs, num_eqs, 0, 0);
1683 	if (err) {
1684 		mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1685 		goto err_unmap_cmpt;
1686 	}
1687 
1688 	/*
1689 	 * Reserved MTT entries must be aligned up to a cacheline
1690 	 * boundary, since the FW will write to them, while the driver
1691 	 * writes to all other MTT entries. (The variable
1692 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1693 	 * size, not the raw entry size)
1694 	 */
1695 	dev->caps.reserved_mtts =
1696 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1697 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1698 
1699 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1700 				  init_hca->mtt_base,
1701 				  dev->caps.mtt_entry_sz,
1702 				  dev->caps.num_mtts,
1703 				  dev->caps.reserved_mtts, 1, 0);
1704 	if (err) {
1705 		mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1706 		goto err_unmap_eq;
1707 	}
1708 
1709 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1710 				  init_hca->dmpt_base,
1711 				  dev_cap->dmpt_entry_sz,
1712 				  dev->caps.num_mpts,
1713 				  dev->caps.reserved_mrws, 1, 1);
1714 	if (err) {
1715 		mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1716 		goto err_unmap_mtt;
1717 	}
1718 
1719 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1720 				  init_hca->qpc_base,
1721 				  dev_cap->qpc_entry_sz,
1722 				  dev->caps.num_qps,
1723 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1724 				  0, 0);
1725 	if (err) {
1726 		mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1727 		goto err_unmap_dmpt;
1728 	}
1729 
1730 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1731 				  init_hca->auxc_base,
1732 				  dev_cap->aux_entry_sz,
1733 				  dev->caps.num_qps,
1734 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1735 				  0, 0);
1736 	if (err) {
1737 		mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1738 		goto err_unmap_qp;
1739 	}
1740 
1741 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1742 				  init_hca->altc_base,
1743 				  dev_cap->altc_entry_sz,
1744 				  dev->caps.num_qps,
1745 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1746 				  0, 0);
1747 	if (err) {
1748 		mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1749 		goto err_unmap_auxc;
1750 	}
1751 
1752 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1753 				  init_hca->rdmarc_base,
1754 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1755 				  dev->caps.num_qps,
1756 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1757 				  0, 0);
1758 	if (err) {
1759 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1760 		goto err_unmap_altc;
1761 	}
1762 
1763 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1764 				  init_hca->cqc_base,
1765 				  dev_cap->cqc_entry_sz,
1766 				  dev->caps.num_cqs,
1767 				  dev->caps.reserved_cqs, 0, 0);
1768 	if (err) {
1769 		mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1770 		goto err_unmap_rdmarc;
1771 	}
1772 
1773 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1774 				  init_hca->srqc_base,
1775 				  dev_cap->srq_entry_sz,
1776 				  dev->caps.num_srqs,
1777 				  dev->caps.reserved_srqs, 0, 0);
1778 	if (err) {
1779 		mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1780 		goto err_unmap_cq;
1781 	}
1782 
1783 	/*
1784 	 * For flow steering device managed mode it is required to use
1785 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1786 	 * required, but for simplicity just map the whole multicast
1787 	 * group table now.  The table isn't very big and it's a lot
1788 	 * easier than trying to track ref counts.
1789 	 */
1790 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1791 				  init_hca->mc_base,
1792 				  mlx4_get_mgm_entry_size(dev),
1793 				  dev->caps.num_mgms + dev->caps.num_amgms,
1794 				  dev->caps.num_mgms + dev->caps.num_amgms,
1795 				  0, 0);
1796 	if (err) {
1797 		mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1798 		goto err_unmap_srq;
1799 	}
1800 
1801 	return 0;
1802 
1803 err_unmap_srq:
1804 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1805 
1806 err_unmap_cq:
1807 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1808 
1809 err_unmap_rdmarc:
1810 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1811 
1812 err_unmap_altc:
1813 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1814 
1815 err_unmap_auxc:
1816 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1817 
1818 err_unmap_qp:
1819 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1820 
1821 err_unmap_dmpt:
1822 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1823 
1824 err_unmap_mtt:
1825 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1826 
1827 err_unmap_eq:
1828 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1829 
1830 err_unmap_cmpt:
1831 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1832 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1833 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1834 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1835 
1836 err_unmap_aux:
1837 	mlx4_UNMAP_ICM_AUX(dev);
1838 
1839 err_free_aux:
1840 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1841 
1842 	return err;
1843 }
1844 
1845 static void mlx4_free_icms(struct mlx4_dev *dev)
1846 {
1847 	struct mlx4_priv *priv = mlx4_priv(dev);
1848 
1849 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1850 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1851 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1852 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1853 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1854 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1855 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1856 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1857 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1858 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1859 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1860 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1861 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1862 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1863 
1864 	mlx4_UNMAP_ICM_AUX(dev);
1865 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1866 }
1867 
1868 static void mlx4_slave_exit(struct mlx4_dev *dev)
1869 {
1870 	struct mlx4_priv *priv = mlx4_priv(dev);
1871 
1872 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1873 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1874 			  MLX4_COMM_TIME))
1875 		mlx4_warn(dev, "Failed to close slave function\n");
1876 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1877 }
1878 
1879 static int map_bf_area(struct mlx4_dev *dev)
1880 {
1881 	struct mlx4_priv *priv = mlx4_priv(dev);
1882 	resource_size_t bf_start;
1883 	resource_size_t bf_len;
1884 	int err = 0;
1885 
1886 	if (!dev->caps.bf_reg_size)
1887 		return -ENXIO;
1888 
1889 	bf_start = pci_resource_start(dev->persist->pdev, 2) +
1890 			(dev->caps.num_uars << PAGE_SHIFT);
1891 	bf_len = pci_resource_len(dev->persist->pdev, 2) -
1892 			(dev->caps.num_uars << PAGE_SHIFT);
1893 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1894 	if (!priv->bf_mapping)
1895 		err = -ENOMEM;
1896 
1897 	return err;
1898 }
1899 
1900 static void unmap_bf_area(struct mlx4_dev *dev)
1901 {
1902 	if (mlx4_priv(dev)->bf_mapping)
1903 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1904 }
1905 
1906 u64 mlx4_read_clock(struct mlx4_dev *dev)
1907 {
1908 	u32 clockhi, clocklo, clockhi1;
1909 	u64 cycles;
1910 	int i;
1911 	struct mlx4_priv *priv = mlx4_priv(dev);
1912 
1913 	for (i = 0; i < 10; i++) {
1914 		clockhi = swab32(readl(priv->clock_mapping));
1915 		clocklo = swab32(readl(priv->clock_mapping + 4));
1916 		clockhi1 = swab32(readl(priv->clock_mapping));
1917 		if (clockhi == clockhi1)
1918 			break;
1919 	}
1920 
1921 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1922 
1923 	return cycles;
1924 }
1925 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1926 
1927 
1928 static int map_internal_clock(struct mlx4_dev *dev)
1929 {
1930 	struct mlx4_priv *priv = mlx4_priv(dev);
1931 
1932 	priv->clock_mapping =
1933 		ioremap(pci_resource_start(dev->persist->pdev,
1934 					   priv->fw.clock_bar) +
1935 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1936 
1937 	if (!priv->clock_mapping)
1938 		return -ENOMEM;
1939 
1940 	return 0;
1941 }
1942 
1943 int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1944 				   struct mlx4_clock_params *params)
1945 {
1946 	struct mlx4_priv *priv = mlx4_priv(dev);
1947 
1948 	if (mlx4_is_slave(dev))
1949 		return -EOPNOTSUPP;
1950 
1951 	if (!params)
1952 		return -EINVAL;
1953 
1954 	params->bar = priv->fw.clock_bar;
1955 	params->offset = priv->fw.clock_offset;
1956 	params->size = MLX4_CLOCK_SIZE;
1957 
1958 	return 0;
1959 }
1960 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1961 
1962 static void unmap_internal_clock(struct mlx4_dev *dev)
1963 {
1964 	struct mlx4_priv *priv = mlx4_priv(dev);
1965 
1966 	if (priv->clock_mapping)
1967 		iounmap(priv->clock_mapping);
1968 }
1969 
1970 static void mlx4_close_hca(struct mlx4_dev *dev)
1971 {
1972 	unmap_internal_clock(dev);
1973 	unmap_bf_area(dev);
1974 	if (mlx4_is_slave(dev))
1975 		mlx4_slave_exit(dev);
1976 	else {
1977 		mlx4_CLOSE_HCA(dev, 0);
1978 		mlx4_free_icms(dev);
1979 	}
1980 }
1981 
1982 static void mlx4_close_fw(struct mlx4_dev *dev)
1983 {
1984 	if (!mlx4_is_slave(dev)) {
1985 		mlx4_UNMAP_FA(dev);
1986 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1987 	}
1988 }
1989 
1990 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1991 {
1992 #define COMM_CHAN_OFFLINE_OFFSET 0x09
1993 
1994 	u32 comm_flags;
1995 	u32 offline_bit;
1996 	unsigned long end;
1997 	struct mlx4_priv *priv = mlx4_priv(dev);
1998 
1999 	end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
2000 	while (time_before(jiffies, end)) {
2001 		comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
2002 					  MLX4_COMM_CHAN_FLAGS));
2003 		offline_bit = (comm_flags &
2004 			       (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
2005 		if (!offline_bit)
2006 			return 0;
2007 
2008 		/* If device removal has been requested,
2009 		 * do not continue retrying.
2010 		 */
2011 		if (dev->persist->interface_state &
2012 		    MLX4_INTERFACE_STATE_NOWAIT)
2013 			break;
2014 
2015 		/* There are cases as part of AER/Reset flow that PF needs
2016 		 * around 100 msec to load. We therefore sleep for 100 msec
2017 		 * to allow other tasks to make use of that CPU during this
2018 		 * time interval.
2019 		 */
2020 		msleep(100);
2021 	}
2022 	mlx4_err(dev, "Communication channel is offline.\n");
2023 	return -EIO;
2024 }
2025 
2026 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
2027 {
2028 #define COMM_CHAN_RST_OFFSET 0x1e
2029 
2030 	struct mlx4_priv *priv = mlx4_priv(dev);
2031 	u32 comm_rst;
2032 	u32 comm_caps;
2033 
2034 	comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
2035 				 MLX4_COMM_CHAN_CAPS));
2036 	comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
2037 
2038 	if (comm_rst)
2039 		dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
2040 }
2041 
2042 static int mlx4_init_slave(struct mlx4_dev *dev)
2043 {
2044 	struct mlx4_priv *priv = mlx4_priv(dev);
2045 	u64 dma = (u64) priv->mfunc.vhcr_dma;
2046 	int ret_from_reset = 0;
2047 	u32 slave_read;
2048 	u32 cmd_channel_ver;
2049 
2050 	if (atomic_read(&pf_loading)) {
2051 		mlx4_warn(dev, "PF is not ready - Deferring probe\n");
2052 		return -EPROBE_DEFER;
2053 	}
2054 
2055 	mutex_lock(&priv->cmd.slave_cmd_mutex);
2056 	priv->cmd.max_cmds = 1;
2057 	if (mlx4_comm_check_offline(dev)) {
2058 		mlx4_err(dev, "PF is not responsive, skipping initialization\n");
2059 		goto err_offline;
2060 	}
2061 
2062 	mlx4_reset_vf_support(dev);
2063 	mlx4_warn(dev, "Sending reset\n");
2064 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
2065 				       MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
2066 	/* if we are in the middle of flr the slave will try
2067 	 * NUM_OF_RESET_RETRIES times before leaving.*/
2068 	if (ret_from_reset) {
2069 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
2070 			mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
2071 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
2072 			return -EPROBE_DEFER;
2073 		} else
2074 			goto err;
2075 	}
2076 
2077 	/* check the driver version - the slave I/F revision
2078 	 * must match the master's */
2079 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
2080 	cmd_channel_ver = mlx4_comm_get_version();
2081 
2082 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
2083 		MLX4_COMM_GET_IF_REV(slave_read)) {
2084 		mlx4_err(dev, "slave driver version is not supported by the master\n");
2085 		goto err;
2086 	}
2087 
2088 	mlx4_warn(dev, "Sending vhcr0\n");
2089 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
2090 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2091 		goto err;
2092 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
2093 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2094 		goto err;
2095 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
2096 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2097 		goto err;
2098 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
2099 			  MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2100 		goto err;
2101 
2102 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2103 	return 0;
2104 
2105 err:
2106 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
2107 err_offline:
2108 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2109 	return -EIO;
2110 }
2111 
2112 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
2113 {
2114 	int i;
2115 
2116 	for (i = 1; i <= dev->caps.num_ports; i++) {
2117 		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
2118 			dev->caps.gid_table_len[i] =
2119 				mlx4_get_slave_num_gids(dev, 0, i);
2120 		else
2121 			dev->caps.gid_table_len[i] = 1;
2122 		dev->caps.pkey_table_len[i] =
2123 			dev->phys_caps.pkey_phys_table_len[i] - 1;
2124 	}
2125 }
2126 
2127 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
2128 {
2129 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
2130 
2131 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
2132 	      i++) {
2133 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
2134 			break;
2135 	}
2136 
2137 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
2138 }
2139 
2140 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
2141 {
2142 	switch (dmfs_high_steer_mode) {
2143 	case MLX4_STEERING_DMFS_A0_DEFAULT:
2144 		return "default performance";
2145 
2146 	case MLX4_STEERING_DMFS_A0_DYNAMIC:
2147 		return "dynamic hybrid mode";
2148 
2149 	case MLX4_STEERING_DMFS_A0_STATIC:
2150 		return "performance optimized for limited rule configuration (static)";
2151 
2152 	case MLX4_STEERING_DMFS_A0_DISABLE:
2153 		return "disabled performance optimized steering";
2154 
2155 	case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
2156 		return "performance optimized steering not supported";
2157 
2158 	default:
2159 		return "Unrecognized mode";
2160 	}
2161 }
2162 
2163 #define MLX4_DMFS_A0_STEERING			(1UL << 2)
2164 
2165 static void choose_steering_mode(struct mlx4_dev *dev,
2166 				 struct mlx4_dev_cap *dev_cap)
2167 {
2168 	if (mlx4_log_num_mgm_entry_size <= 0) {
2169 		if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
2170 			if (dev->caps.dmfs_high_steer_mode ==
2171 			    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2172 				mlx4_err(dev, "DMFS high rate mode not supported\n");
2173 			else
2174 				dev->caps.dmfs_high_steer_mode =
2175 					MLX4_STEERING_DMFS_A0_STATIC;
2176 		}
2177 	}
2178 
2179 	if (mlx4_log_num_mgm_entry_size <= 0 &&
2180 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
2181 	    (!mlx4_is_mfunc(dev) ||
2182 	     (dev_cap->fs_max_num_qp_per_entry >=
2183 	     (dev->persist->num_vfs + 1))) &&
2184 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
2185 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
2186 		dev->oper_log_mgm_entry_size =
2187 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
2188 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
2189 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
2190 		dev->caps.fs_log_max_ucast_qp_range_size =
2191 			dev_cap->fs_log_max_ucast_qp_range_size;
2192 	} else {
2193 		if (dev->caps.dmfs_high_steer_mode !=
2194 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2195 			dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
2196 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
2197 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2198 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
2199 		else {
2200 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
2201 
2202 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
2203 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2204 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
2205 		}
2206 		dev->oper_log_mgm_entry_size =
2207 			mlx4_log_num_mgm_entry_size > 0 ?
2208 			mlx4_log_num_mgm_entry_size :
2209 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
2210 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
2211 	}
2212 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
2213 		 mlx4_steering_mode_str(dev->caps.steering_mode),
2214 		 dev->oper_log_mgm_entry_size,
2215 		 mlx4_log_num_mgm_entry_size);
2216 }
2217 
2218 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
2219 				       struct mlx4_dev_cap *dev_cap)
2220 {
2221 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2222 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
2223 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
2224 	else
2225 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
2226 
2227 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
2228 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
2229 }
2230 
2231 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
2232 {
2233 	int i;
2234 	struct mlx4_port_cap port_cap;
2235 
2236 	if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2237 		return -EINVAL;
2238 
2239 	for (i = 1; i <= dev->caps.num_ports; i++) {
2240 		if (mlx4_dev_port(dev, i, &port_cap)) {
2241 			mlx4_err(dev,
2242 				 "QUERY_DEV_CAP command failed, can't verify DMFS high rate steering.\n");
2243 		} else if ((dev->caps.dmfs_high_steer_mode !=
2244 			    MLX4_STEERING_DMFS_A0_DEFAULT) &&
2245 			   (port_cap.dmfs_optimized_state ==
2246 			    !!(dev->caps.dmfs_high_steer_mode ==
2247 			    MLX4_STEERING_DMFS_A0_DISABLE))) {
2248 			mlx4_err(dev,
2249 				 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2250 				 dmfs_high_rate_steering_mode_str(
2251 					dev->caps.dmfs_high_steer_mode),
2252 				 (port_cap.dmfs_optimized_state ?
2253 					"enabled" : "disabled"));
2254 		}
2255 	}
2256 
2257 	return 0;
2258 }
2259 
2260 static int mlx4_init_fw(struct mlx4_dev *dev)
2261 {
2262 	struct mlx4_mod_stat_cfg   mlx4_cfg;
2263 	int err = 0;
2264 
2265 	if (!mlx4_is_slave(dev)) {
2266 		err = mlx4_QUERY_FW(dev);
2267 		if (err) {
2268 			if (err == -EACCES)
2269 				mlx4_info(dev, "non-primary physical function, skipping\n");
2270 			else
2271 				mlx4_err(dev, "QUERY_FW command failed, aborting\n");
2272 			return err;
2273 		}
2274 
2275 		err = mlx4_load_fw(dev);
2276 		if (err) {
2277 			mlx4_err(dev, "Failed to start FW, aborting\n");
2278 			return err;
2279 		}
2280 
2281 		mlx4_cfg.log_pg_sz_m = 1;
2282 		mlx4_cfg.log_pg_sz = 0;
2283 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2284 		if (err)
2285 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2286 	}
2287 
2288 	return err;
2289 }
2290 
2291 static int mlx4_init_hca(struct mlx4_dev *dev)
2292 {
2293 	struct mlx4_priv	  *priv = mlx4_priv(dev);
2294 	struct mlx4_init_hca_param *init_hca = NULL;
2295 	struct mlx4_dev_cap	  *dev_cap = NULL;
2296 	struct mlx4_adapter	   adapter;
2297 	struct mlx4_profile	   profile;
2298 	u64 icm_size;
2299 	struct mlx4_config_dev_params params;
2300 	int err;
2301 
2302 	if (!mlx4_is_slave(dev)) {
2303 		dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
2304 		init_hca = kzalloc(sizeof(*init_hca), GFP_KERNEL);
2305 
2306 		if (!dev_cap || !init_hca) {
2307 			err = -ENOMEM;
2308 			goto out_free;
2309 		}
2310 
2311 		err = mlx4_dev_cap(dev, dev_cap);
2312 		if (err) {
2313 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
2314 			goto out_free;
2315 		}
2316 
2317 		choose_steering_mode(dev, dev_cap);
2318 		choose_tunnel_offload_mode(dev, dev_cap);
2319 
2320 		if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2321 		    mlx4_is_master(dev))
2322 			dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2323 
2324 		err = mlx4_get_phys_port_id(dev);
2325 		if (err)
2326 			mlx4_err(dev, "Fail to get physical port id\n");
2327 
2328 		if (mlx4_is_master(dev))
2329 			mlx4_parav_master_pf_caps(dev);
2330 
2331 		if (mlx4_low_memory_profile()) {
2332 			mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2333 			profile = low_mem_profile;
2334 		} else {
2335 			profile = default_profile;
2336 		}
2337 		if (dev->caps.steering_mode ==
2338 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
2339 			profile.num_mcg = MLX4_FS_NUM_MCG;
2340 
2341 		icm_size = mlx4_make_profile(dev, &profile, dev_cap,
2342 					     init_hca);
2343 		if ((long long) icm_size < 0) {
2344 			err = icm_size;
2345 			goto out_free;
2346 		}
2347 
2348 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
2349 
2350 		if (enable_4k_uar || !dev->persist->num_vfs) {
2351 			init_hca->log_uar_sz = ilog2(dev->caps.num_uars) +
2352 						    PAGE_SHIFT - DEFAULT_UAR_PAGE_SHIFT;
2353 			init_hca->uar_page_sz = DEFAULT_UAR_PAGE_SHIFT - 12;
2354 		} else {
2355 			init_hca->log_uar_sz = ilog2(dev->caps.num_uars);
2356 			init_hca->uar_page_sz = PAGE_SHIFT - 12;
2357 		}
2358 
2359 		init_hca->mw_enabled = 0;
2360 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2361 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2362 			init_hca->mw_enabled = INIT_HCA_TPT_MW_ENABLE;
2363 
2364 		err = mlx4_init_icm(dev, dev_cap, init_hca, icm_size);
2365 		if (err)
2366 			goto out_free;
2367 
2368 		err = mlx4_INIT_HCA(dev, init_hca);
2369 		if (err) {
2370 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
2371 			goto err_free_icm;
2372 		}
2373 
2374 		if (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2375 			err = mlx4_query_func(dev, dev_cap);
2376 			if (err < 0) {
2377 				mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
2378 				goto err_close;
2379 			} else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2380 				dev->caps.num_eqs = dev_cap->max_eqs;
2381 				dev->caps.reserved_eqs = dev_cap->reserved_eqs;
2382 				dev->caps.reserved_uars = dev_cap->reserved_uars;
2383 			}
2384 		}
2385 
2386 		/*
2387 		 * If TS is supported by FW
2388 		 * read HCA frequency by QUERY_HCA command
2389 		 */
2390 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2391 			err = mlx4_QUERY_HCA(dev, init_hca);
2392 			if (err) {
2393 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
2394 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2395 			} else {
2396 				dev->caps.hca_core_clock =
2397 					init_hca->hca_core_clock;
2398 			}
2399 
2400 			/* In case we got HCA frequency 0 - disable timestamping
2401 			 * to avoid dividing by zero
2402 			 */
2403 			if (!dev->caps.hca_core_clock) {
2404 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2405 				mlx4_err(dev,
2406 					 "HCA frequency is 0 - timestamping is not supported\n");
2407 			} else if (map_internal_clock(dev)) {
2408 				/*
2409 				 * Map internal clock,
2410 				 * in case of failure disable timestamping
2411 				 */
2412 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2413 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
2414 			}
2415 		}
2416 
2417 		if (dev->caps.dmfs_high_steer_mode !=
2418 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2419 			if (mlx4_validate_optimized_steering(dev))
2420 				mlx4_warn(dev, "Optimized steering validation failed\n");
2421 
2422 			if (dev->caps.dmfs_high_steer_mode ==
2423 			    MLX4_STEERING_DMFS_A0_DISABLE) {
2424 				dev->caps.dmfs_high_rate_qpn_base =
2425 					dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2426 				dev->caps.dmfs_high_rate_qpn_range =
2427 					MLX4_A0_STEERING_TABLE_SIZE;
2428 			}
2429 
2430 			mlx4_info(dev, "DMFS high rate steer mode is: %s\n",
2431 				  dmfs_high_rate_steering_mode_str(
2432 					dev->caps.dmfs_high_steer_mode));
2433 		}
2434 	} else {
2435 		err = mlx4_init_slave(dev);
2436 		if (err) {
2437 			if (err != -EPROBE_DEFER)
2438 				mlx4_err(dev, "Failed to initialize slave\n");
2439 			return err;
2440 		}
2441 
2442 		err = mlx4_slave_cap(dev);
2443 		if (err) {
2444 			mlx4_err(dev, "Failed to obtain slave caps\n");
2445 			goto err_close;
2446 		}
2447 	}
2448 
2449 	if (map_bf_area(dev))
2450 		mlx4_dbg(dev, "Failed to map blue flame area\n");
2451 
2452 	/*Only the master set the ports, all the rest got it from it.*/
2453 	if (!mlx4_is_slave(dev))
2454 		mlx4_set_port_mask(dev);
2455 
2456 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
2457 	if (err) {
2458 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2459 		goto unmap_bf;
2460 	}
2461 
2462 	/* Query CONFIG_DEV parameters */
2463 	err = mlx4_config_dev_retrieval(dev, &params);
2464 	if (err && err != -EOPNOTSUPP) {
2465 		mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2466 	} else if (!err) {
2467 		dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2468 		dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2469 	}
2470 	priv->eq_table.inta_pin = adapter.inta_pin;
2471 	memcpy(dev->board_id, adapter.board_id, sizeof(dev->board_id));
2472 
2473 	err = 0;
2474 	goto out_free;
2475 
2476 unmap_bf:
2477 	unmap_internal_clock(dev);
2478 	unmap_bf_area(dev);
2479 
2480 	if (mlx4_is_slave(dev))
2481 		mlx4_slave_destroy_special_qp_cap(dev);
2482 
2483 err_close:
2484 	if (mlx4_is_slave(dev))
2485 		mlx4_slave_exit(dev);
2486 	else
2487 		mlx4_CLOSE_HCA(dev, 0);
2488 
2489 err_free_icm:
2490 	if (!mlx4_is_slave(dev))
2491 		mlx4_free_icms(dev);
2492 
2493 out_free:
2494 	kfree(dev_cap);
2495 	kfree(init_hca);
2496 
2497 	return err;
2498 }
2499 
2500 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2501 {
2502 	struct mlx4_priv *priv = mlx4_priv(dev);
2503 	int nent_pow2;
2504 
2505 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2506 		return -ENOENT;
2507 
2508 	if (!dev->caps.max_counters)
2509 		return -ENOSPC;
2510 
2511 	nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2512 	/* reserve last counter index for sink counter */
2513 	return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2514 				nent_pow2 - 1, 0,
2515 				nent_pow2 - dev->caps.max_counters + 1);
2516 }
2517 
2518 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2519 {
2520 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2521 		return;
2522 
2523 	if (!dev->caps.max_counters)
2524 		return;
2525 
2526 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2527 }
2528 
2529 static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2530 {
2531 	struct mlx4_priv *priv = mlx4_priv(dev);
2532 	int port;
2533 
2534 	for (port = 0; port < dev->caps.num_ports; port++)
2535 		if (priv->def_counter[port] != -1)
2536 			mlx4_counter_free(dev,  priv->def_counter[port]);
2537 }
2538 
2539 static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2540 {
2541 	struct mlx4_priv *priv = mlx4_priv(dev);
2542 	int port, err = 0;
2543 	u32 idx;
2544 
2545 	for (port = 0; port < dev->caps.num_ports; port++)
2546 		priv->def_counter[port] = -1;
2547 
2548 	for (port = 0; port < dev->caps.num_ports; port++) {
2549 		err = mlx4_counter_alloc(dev, &idx, MLX4_RES_USAGE_DRIVER);
2550 
2551 		if (!err || err == -ENOSPC) {
2552 			priv->def_counter[port] = idx;
2553 		} else if (err == -ENOENT) {
2554 			err = 0;
2555 			continue;
2556 		} else if (mlx4_is_slave(dev) && err == -EINVAL) {
2557 			priv->def_counter[port] = MLX4_SINK_COUNTER_INDEX(dev);
2558 			mlx4_warn(dev, "can't allocate counter from old PF driver, using index %d\n",
2559 				  MLX4_SINK_COUNTER_INDEX(dev));
2560 			err = 0;
2561 		} else {
2562 			mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2563 				 __func__, port + 1, err);
2564 			mlx4_cleanup_default_counters(dev);
2565 			return err;
2566 		}
2567 
2568 		mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2569 			 __func__, priv->def_counter[port], port + 1);
2570 	}
2571 
2572 	return err;
2573 }
2574 
2575 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2576 {
2577 	struct mlx4_priv *priv = mlx4_priv(dev);
2578 
2579 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2580 		return -ENOENT;
2581 
2582 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2583 	if (*idx == -1) {
2584 		*idx = MLX4_SINK_COUNTER_INDEX(dev);
2585 		return -ENOSPC;
2586 	}
2587 
2588 	return 0;
2589 }
2590 
2591 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx, u8 usage)
2592 {
2593 	u32 in_modifier = RES_COUNTER | (((u32)usage & 3) << 30);
2594 	u64 out_param;
2595 	int err;
2596 
2597 	if (mlx4_is_mfunc(dev)) {
2598 		err = mlx4_cmd_imm(dev, 0, &out_param, in_modifier,
2599 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2600 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2601 		if (!err)
2602 			*idx = get_param_l(&out_param);
2603 
2604 		return err;
2605 	}
2606 	return __mlx4_counter_alloc(dev, idx);
2607 }
2608 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2609 
2610 static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2611 				u8 counter_index)
2612 {
2613 	struct mlx4_cmd_mailbox *if_stat_mailbox;
2614 	int err;
2615 	u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2616 
2617 	if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2618 	if (IS_ERR(if_stat_mailbox))
2619 		return PTR_ERR(if_stat_mailbox);
2620 
2621 	err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2622 			   MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2623 			   MLX4_CMD_NATIVE);
2624 
2625 	mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2626 	return err;
2627 }
2628 
2629 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2630 {
2631 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2632 		return;
2633 
2634 	if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2635 		return;
2636 
2637 	__mlx4_clear_if_stat(dev, idx);
2638 
2639 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2640 	return;
2641 }
2642 
2643 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2644 {
2645 	u64 in_param = 0;
2646 
2647 	if (mlx4_is_mfunc(dev)) {
2648 		set_param_l(&in_param, idx);
2649 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2650 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2651 			 MLX4_CMD_WRAPPED);
2652 		return;
2653 	}
2654 	__mlx4_counter_free(dev, idx);
2655 }
2656 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2657 
2658 int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2659 {
2660 	struct mlx4_priv *priv = mlx4_priv(dev);
2661 
2662 	return priv->def_counter[port - 1];
2663 }
2664 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2665 
2666 void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2667 {
2668 	struct mlx4_priv *priv = mlx4_priv(dev);
2669 
2670 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2671 }
2672 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2673 
2674 __be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2675 {
2676 	struct mlx4_priv *priv = mlx4_priv(dev);
2677 
2678 	return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2679 }
2680 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2681 
2682 void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2683 {
2684 	struct mlx4_priv *priv = mlx4_priv(dev);
2685 	__be64 guid;
2686 
2687 	/* hw GUID */
2688 	if (entry == 0)
2689 		return;
2690 
2691 	get_random_bytes((char *)&guid, sizeof(guid));
2692 	guid &= ~(cpu_to_be64(1ULL << 56));
2693 	guid |= cpu_to_be64(1ULL << 57);
2694 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2695 }
2696 
2697 static int mlx4_setup_hca(struct mlx4_dev *dev)
2698 {
2699 	struct mlx4_priv *priv = mlx4_priv(dev);
2700 	int err;
2701 	int port;
2702 	__be32 ib_port_default_caps;
2703 
2704 	err = mlx4_init_uar_table(dev);
2705 	if (err) {
2706 		mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2707 		return err;
2708 	}
2709 
2710 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
2711 	if (err) {
2712 		mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2713 		goto err_uar_table_free;
2714 	}
2715 
2716 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2717 	if (!priv->kar) {
2718 		mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2719 		err = -ENOMEM;
2720 		goto err_uar_free;
2721 	}
2722 
2723 	err = mlx4_init_pd_table(dev);
2724 	if (err) {
2725 		mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2726 		goto err_kar_unmap;
2727 	}
2728 
2729 	err = mlx4_init_xrcd_table(dev);
2730 	if (err) {
2731 		mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2732 		goto err_pd_table_free;
2733 	}
2734 
2735 	err = mlx4_init_mr_table(dev);
2736 	if (err) {
2737 		mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2738 		goto err_xrcd_table_free;
2739 	}
2740 
2741 	if (!mlx4_is_slave(dev)) {
2742 		err = mlx4_init_mcg_table(dev);
2743 		if (err) {
2744 			mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2745 			goto err_mr_table_free;
2746 		}
2747 		err = mlx4_config_mad_demux(dev);
2748 		if (err) {
2749 			mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2750 			goto err_mcg_table_free;
2751 		}
2752 	}
2753 
2754 	err = mlx4_init_eq_table(dev);
2755 	if (err) {
2756 		mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2757 		goto err_mcg_table_free;
2758 	}
2759 
2760 	err = mlx4_cmd_use_events(dev);
2761 	if (err) {
2762 		mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2763 		goto err_eq_table_free;
2764 	}
2765 
2766 	err = mlx4_NOP(dev);
2767 	if (err) {
2768 		if (dev->flags & MLX4_FLAG_MSI_X) {
2769 			mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2770 				  priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2771 			mlx4_warn(dev, "Trying again without MSI-X\n");
2772 		} else {
2773 			mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2774 				 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2775 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2776 		}
2777 
2778 		goto err_cmd_poll;
2779 	}
2780 
2781 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
2782 
2783 	err = mlx4_init_cq_table(dev);
2784 	if (err) {
2785 		mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2786 		goto err_cmd_poll;
2787 	}
2788 
2789 	err = mlx4_init_srq_table(dev);
2790 	if (err) {
2791 		mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2792 		goto err_cq_table_free;
2793 	}
2794 
2795 	err = mlx4_init_qp_table(dev);
2796 	if (err) {
2797 		mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2798 		goto err_srq_table_free;
2799 	}
2800 
2801 	if (!mlx4_is_slave(dev)) {
2802 		err = mlx4_init_counters_table(dev);
2803 		if (err && err != -ENOENT) {
2804 			mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2805 			goto err_qp_table_free;
2806 		}
2807 	}
2808 
2809 	err = mlx4_allocate_default_counters(dev);
2810 	if (err) {
2811 		mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2812 		goto err_counters_table_free;
2813 	}
2814 
2815 	if (!mlx4_is_slave(dev)) {
2816 		for (port = 1; port <= dev->caps.num_ports; port++) {
2817 			ib_port_default_caps = 0;
2818 			err = mlx4_get_port_ib_caps(dev, port,
2819 						    &ib_port_default_caps);
2820 			if (err)
2821 				mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2822 					  port, err);
2823 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2824 
2825 			/* initialize per-slave default ib port capabilities */
2826 			if (mlx4_is_master(dev)) {
2827 				int i;
2828 				for (i = 0; i < dev->num_slaves; i++) {
2829 					if (i == mlx4_master_func_num(dev))
2830 						continue;
2831 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2832 						ib_port_default_caps;
2833 				}
2834 			}
2835 
2836 			if (mlx4_is_mfunc(dev))
2837 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2838 			else
2839 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2840 
2841 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2842 					    dev->caps.pkey_table_len[port] : -1);
2843 			if (err) {
2844 				mlx4_err(dev, "Failed to set port %d, aborting\n",
2845 					 port);
2846 				goto err_default_countes_free;
2847 			}
2848 		}
2849 	}
2850 
2851 	return 0;
2852 
2853 err_default_countes_free:
2854 	mlx4_cleanup_default_counters(dev);
2855 
2856 err_counters_table_free:
2857 	if (!mlx4_is_slave(dev))
2858 		mlx4_cleanup_counters_table(dev);
2859 
2860 err_qp_table_free:
2861 	mlx4_cleanup_qp_table(dev);
2862 
2863 err_srq_table_free:
2864 	mlx4_cleanup_srq_table(dev);
2865 
2866 err_cq_table_free:
2867 	mlx4_cleanup_cq_table(dev);
2868 
2869 err_cmd_poll:
2870 	mlx4_cmd_use_polling(dev);
2871 
2872 err_eq_table_free:
2873 	mlx4_cleanup_eq_table(dev);
2874 
2875 err_mcg_table_free:
2876 	if (!mlx4_is_slave(dev))
2877 		mlx4_cleanup_mcg_table(dev);
2878 
2879 err_mr_table_free:
2880 	mlx4_cleanup_mr_table(dev);
2881 
2882 err_xrcd_table_free:
2883 	mlx4_cleanup_xrcd_table(dev);
2884 
2885 err_pd_table_free:
2886 	mlx4_cleanup_pd_table(dev);
2887 
2888 err_kar_unmap:
2889 	iounmap(priv->kar);
2890 
2891 err_uar_free:
2892 	mlx4_uar_free(dev, &priv->driver_uar);
2893 
2894 err_uar_table_free:
2895 	mlx4_cleanup_uar_table(dev);
2896 	return err;
2897 }
2898 
2899 static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2900 {
2901 	int requested_cpu = 0;
2902 	struct mlx4_priv *priv = mlx4_priv(dev);
2903 	struct mlx4_eq *eq;
2904 	int off = 0;
2905 	int i;
2906 
2907 	if (eqn > dev->caps.num_comp_vectors)
2908 		return -EINVAL;
2909 
2910 	for (i = 1; i < port; i++)
2911 		off += mlx4_get_eqs_per_port(dev, i);
2912 
2913 	requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2914 
2915 	/* Meaning EQs are shared, and this call comes from the second port */
2916 	if (requested_cpu < 0)
2917 		return 0;
2918 
2919 	eq = &priv->eq_table.eq[eqn];
2920 
2921 	if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2922 		return -ENOMEM;
2923 
2924 	cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2925 
2926 	return 0;
2927 }
2928 
2929 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2930 {
2931 	struct mlx4_priv *priv = mlx4_priv(dev);
2932 	struct msix_entry *entries;
2933 	int i;
2934 	int port = 0;
2935 
2936 	if (msi_x) {
2937 		int nreq = min3(dev->caps.num_ports *
2938 				(int)num_online_cpus() + 1,
2939 				dev->caps.num_eqs - dev->caps.reserved_eqs,
2940 				MAX_MSIX);
2941 
2942 		if (msi_x > 1)
2943 			nreq = min_t(int, nreq, msi_x);
2944 
2945 		entries = kcalloc(nreq, sizeof(*entries), GFP_KERNEL);
2946 		if (!entries)
2947 			goto no_msi;
2948 
2949 		for (i = 0; i < nreq; ++i)
2950 			entries[i].entry = i;
2951 
2952 		nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2953 					     nreq);
2954 
2955 		if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
2956 			kfree(entries);
2957 			goto no_msi;
2958 		}
2959 		/* 1 is reserved for events (asyncrounous EQ) */
2960 		dev->caps.num_comp_vectors = nreq - 1;
2961 
2962 		priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2963 		bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2964 			    dev->caps.num_ports);
2965 
2966 		for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2967 			if (i == MLX4_EQ_ASYNC)
2968 				continue;
2969 
2970 			priv->eq_table.eq[i].irq =
2971 				entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2972 
2973 			if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2974 				bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2975 					    dev->caps.num_ports);
2976 				/* We don't set affinity hint when there
2977 				 * aren't enough EQs
2978 				 */
2979 			} else {
2980 				set_bit(port,
2981 					priv->eq_table.eq[i].actv_ports.ports);
2982 				if (mlx4_init_affinity_hint(dev, port + 1, i))
2983 					mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
2984 						  i);
2985 			}
2986 			/* We divide the Eqs evenly between the two ports.
2987 			 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
2988 			 * refers to the number of Eqs per port
2989 			 * (i.e eqs_per_port). Theoretically, we would like to
2990 			 * write something like (i + 1) % eqs_per_port == 0.
2991 			 * However, since there's an asynchronous Eq, we have
2992 			 * to skip over it by comparing this condition to
2993 			 * !!((i + 1) > MLX4_EQ_ASYNC).
2994 			 */
2995 			if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
2996 			    ((i + 1) %
2997 			     (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
2998 			    !!((i + 1) > MLX4_EQ_ASYNC))
2999 				/* If dev->caps.num_comp_vectors < dev->caps.num_ports,
3000 				 * everything is shared anyway.
3001 				 */
3002 				port++;
3003 		}
3004 
3005 		dev->flags |= MLX4_FLAG_MSI_X;
3006 
3007 		kfree(entries);
3008 		return;
3009 	}
3010 
3011 no_msi:
3012 	dev->caps.num_comp_vectors = 1;
3013 
3014 	BUG_ON(MLX4_EQ_ASYNC >= 2);
3015 	for (i = 0; i < 2; ++i) {
3016 		priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
3017 		if (i != MLX4_EQ_ASYNC) {
3018 			bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
3019 				    dev->caps.num_ports);
3020 		}
3021 	}
3022 }
3023 
3024 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
3025 {
3026 	struct devlink *devlink = priv_to_devlink(mlx4_priv(dev));
3027 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
3028 	int err;
3029 
3030 	err = devlink_port_register(devlink, &info->devlink_port, port);
3031 	if (err)
3032 		return err;
3033 
3034 	info->dev = dev;
3035 	info->port = port;
3036 	if (!mlx4_is_slave(dev)) {
3037 		mlx4_init_mac_table(dev, &info->mac_table);
3038 		mlx4_init_vlan_table(dev, &info->vlan_table);
3039 		mlx4_init_roce_gid_table(dev, &info->gid_table);
3040 		info->base_qpn = mlx4_get_base_qpn(dev, port);
3041 	}
3042 
3043 	sprintf(info->dev_name, "mlx4_port%d", port);
3044 	info->port_attr.attr.name = info->dev_name;
3045 	if (mlx4_is_mfunc(dev)) {
3046 		info->port_attr.attr.mode = 0444;
3047 	} else {
3048 		info->port_attr.attr.mode = 0644;
3049 		info->port_attr.store     = set_port_type;
3050 	}
3051 	info->port_attr.show      = show_port_type;
3052 	sysfs_attr_init(&info->port_attr.attr);
3053 
3054 	err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
3055 	if (err) {
3056 		mlx4_err(dev, "Failed to create file for port %d\n", port);
3057 		devlink_port_unregister(&info->devlink_port);
3058 		info->port = -1;
3059 		return err;
3060 	}
3061 
3062 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
3063 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
3064 	if (mlx4_is_mfunc(dev)) {
3065 		info->port_mtu_attr.attr.mode = 0444;
3066 	} else {
3067 		info->port_mtu_attr.attr.mode = 0644;
3068 		info->port_mtu_attr.store     = set_port_ib_mtu;
3069 	}
3070 	info->port_mtu_attr.show      = show_port_ib_mtu;
3071 	sysfs_attr_init(&info->port_mtu_attr.attr);
3072 
3073 	err = device_create_file(&dev->persist->pdev->dev,
3074 				 &info->port_mtu_attr);
3075 	if (err) {
3076 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
3077 		device_remove_file(&info->dev->persist->pdev->dev,
3078 				   &info->port_attr);
3079 		devlink_port_unregister(&info->devlink_port);
3080 		info->port = -1;
3081 		return err;
3082 	}
3083 
3084 	return 0;
3085 }
3086 
3087 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
3088 {
3089 	if (info->port < 0)
3090 		return;
3091 
3092 	device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
3093 	device_remove_file(&info->dev->persist->pdev->dev,
3094 			   &info->port_mtu_attr);
3095 	devlink_port_unregister(&info->devlink_port);
3096 
3097 #ifdef CONFIG_RFS_ACCEL
3098 	free_irq_cpu_rmap(info->rmap);
3099 	info->rmap = NULL;
3100 #endif
3101 }
3102 
3103 static int mlx4_init_steering(struct mlx4_dev *dev)
3104 {
3105 	struct mlx4_priv *priv = mlx4_priv(dev);
3106 	int num_entries = dev->caps.num_ports;
3107 	int i, j;
3108 
3109 	priv->steer = kcalloc(num_entries, sizeof(struct mlx4_steer),
3110 			      GFP_KERNEL);
3111 	if (!priv->steer)
3112 		return -ENOMEM;
3113 
3114 	for (i = 0; i < num_entries; i++)
3115 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3116 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
3117 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
3118 		}
3119 	return 0;
3120 }
3121 
3122 static void mlx4_clear_steering(struct mlx4_dev *dev)
3123 {
3124 	struct mlx4_priv *priv = mlx4_priv(dev);
3125 	struct mlx4_steer_index *entry, *tmp_entry;
3126 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
3127 	int num_entries = dev->caps.num_ports;
3128 	int i, j;
3129 
3130 	for (i = 0; i < num_entries; i++) {
3131 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3132 			list_for_each_entry_safe(pqp, tmp_pqp,
3133 						 &priv->steer[i].promisc_qps[j],
3134 						 list) {
3135 				list_del(&pqp->list);
3136 				kfree(pqp);
3137 			}
3138 			list_for_each_entry_safe(entry, tmp_entry,
3139 						 &priv->steer[i].steer_entries[j],
3140 						 list) {
3141 				list_del(&entry->list);
3142 				list_for_each_entry_safe(pqp, tmp_pqp,
3143 							 &entry->duplicates,
3144 							 list) {
3145 					list_del(&pqp->list);
3146 					kfree(pqp);
3147 				}
3148 				kfree(entry);
3149 			}
3150 		}
3151 	}
3152 	kfree(priv->steer);
3153 }
3154 
3155 static int extended_func_num(struct pci_dev *pdev)
3156 {
3157 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
3158 }
3159 
3160 #define MLX4_OWNER_BASE	0x8069c
3161 #define MLX4_OWNER_SIZE	4
3162 
3163 static int mlx4_get_ownership(struct mlx4_dev *dev)
3164 {
3165 	void __iomem *owner;
3166 	u32 ret;
3167 
3168 	if (pci_channel_offline(dev->persist->pdev))
3169 		return -EIO;
3170 
3171 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3172 			MLX4_OWNER_BASE,
3173 			MLX4_OWNER_SIZE);
3174 	if (!owner) {
3175 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3176 		return -ENOMEM;
3177 	}
3178 
3179 	ret = readl(owner);
3180 	iounmap(owner);
3181 	return (int) !!ret;
3182 }
3183 
3184 static void mlx4_free_ownership(struct mlx4_dev *dev)
3185 {
3186 	void __iomem *owner;
3187 
3188 	if (pci_channel_offline(dev->persist->pdev))
3189 		return;
3190 
3191 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3192 			MLX4_OWNER_BASE,
3193 			MLX4_OWNER_SIZE);
3194 	if (!owner) {
3195 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3196 		return;
3197 	}
3198 	writel(0, owner);
3199 	msleep(1000);
3200 	iounmap(owner);
3201 }
3202 
3203 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV)	==\
3204 				  !!((flags) & MLX4_FLAG_MASTER))
3205 
3206 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
3207 			     u8 total_vfs, int existing_vfs, int reset_flow)
3208 {
3209 	u64 dev_flags = dev->flags;
3210 	int err = 0;
3211 	int fw_enabled_sriov_vfs = min(pci_sriov_get_totalvfs(pdev),
3212 					MLX4_MAX_NUM_VF);
3213 
3214 	if (reset_flow) {
3215 		dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
3216 				       GFP_KERNEL);
3217 		if (!dev->dev_vfs)
3218 			goto free_mem;
3219 		return dev_flags;
3220 	}
3221 
3222 	atomic_inc(&pf_loading);
3223 	if (dev->flags &  MLX4_FLAG_SRIOV) {
3224 		if (existing_vfs != total_vfs) {
3225 			mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
3226 				 existing_vfs, total_vfs);
3227 			total_vfs = existing_vfs;
3228 		}
3229 	}
3230 
3231 	dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs), GFP_KERNEL);
3232 	if (NULL == dev->dev_vfs) {
3233 		mlx4_err(dev, "Failed to allocate memory for VFs\n");
3234 		goto disable_sriov;
3235 	}
3236 
3237 	if (!(dev->flags &  MLX4_FLAG_SRIOV)) {
3238 		if (total_vfs > fw_enabled_sriov_vfs) {
3239 			mlx4_err(dev, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
3240 				 total_vfs, fw_enabled_sriov_vfs);
3241 			err = -ENOMEM;
3242 			goto disable_sriov;
3243 		}
3244 		mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
3245 		err = pci_enable_sriov(pdev, total_vfs);
3246 	}
3247 	if (err) {
3248 		mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
3249 			 err);
3250 		goto disable_sriov;
3251 	} else {
3252 		mlx4_warn(dev, "Running in master mode\n");
3253 		dev_flags |= MLX4_FLAG_SRIOV |
3254 			MLX4_FLAG_MASTER;
3255 		dev_flags &= ~MLX4_FLAG_SLAVE;
3256 		dev->persist->num_vfs = total_vfs;
3257 	}
3258 	return dev_flags;
3259 
3260 disable_sriov:
3261 	atomic_dec(&pf_loading);
3262 free_mem:
3263 	dev->persist->num_vfs = 0;
3264 	kfree(dev->dev_vfs);
3265 	dev->dev_vfs = NULL;
3266 	return dev_flags & ~MLX4_FLAG_MASTER;
3267 }
3268 
3269 enum {
3270 	MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
3271 };
3272 
3273 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
3274 			      int *nvfs)
3275 {
3276 	int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
3277 	/* Checking for 64 VFs as a limitation of CX2 */
3278 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
3279 	    requested_vfs >= 64) {
3280 		mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
3281 			 requested_vfs);
3282 		return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
3283 	}
3284 	return 0;
3285 }
3286 
3287 static int mlx4_pci_enable_device(struct mlx4_dev *dev)
3288 {
3289 	struct pci_dev *pdev = dev->persist->pdev;
3290 	int err = 0;
3291 
3292 	mutex_lock(&dev->persist->pci_status_mutex);
3293 	if (dev->persist->pci_status == MLX4_PCI_STATUS_DISABLED) {
3294 		err = pci_enable_device(pdev);
3295 		if (!err)
3296 			dev->persist->pci_status = MLX4_PCI_STATUS_ENABLED;
3297 	}
3298 	mutex_unlock(&dev->persist->pci_status_mutex);
3299 
3300 	return err;
3301 }
3302 
3303 static void mlx4_pci_disable_device(struct mlx4_dev *dev)
3304 {
3305 	struct pci_dev *pdev = dev->persist->pdev;
3306 
3307 	mutex_lock(&dev->persist->pci_status_mutex);
3308 	if (dev->persist->pci_status == MLX4_PCI_STATUS_ENABLED) {
3309 		pci_disable_device(pdev);
3310 		dev->persist->pci_status = MLX4_PCI_STATUS_DISABLED;
3311 	}
3312 	mutex_unlock(&dev->persist->pci_status_mutex);
3313 }
3314 
3315 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
3316 			 int total_vfs, int *nvfs, struct mlx4_priv *priv,
3317 			 int reset_flow)
3318 {
3319 	struct mlx4_dev *dev;
3320 	unsigned sum = 0;
3321 	int err;
3322 	int port;
3323 	int i;
3324 	struct mlx4_dev_cap *dev_cap = NULL;
3325 	int existing_vfs = 0;
3326 
3327 	dev = &priv->dev;
3328 
3329 	INIT_LIST_HEAD(&priv->ctx_list);
3330 	spin_lock_init(&priv->ctx_lock);
3331 
3332 	mutex_init(&priv->port_mutex);
3333 	mutex_init(&priv->bond_mutex);
3334 
3335 	INIT_LIST_HEAD(&priv->pgdir_list);
3336 	mutex_init(&priv->pgdir_mutex);
3337 	spin_lock_init(&priv->cmd.context_lock);
3338 
3339 	INIT_LIST_HEAD(&priv->bf_list);
3340 	mutex_init(&priv->bf_mutex);
3341 
3342 	dev->rev_id = pdev->revision;
3343 	dev->numa_node = dev_to_node(&pdev->dev);
3344 
3345 	/* Detect if this device is a virtual function */
3346 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3347 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3348 		dev->flags |= MLX4_FLAG_SLAVE;
3349 	} else {
3350 		/* We reset the device and enable SRIOV only for physical
3351 		 * devices.  Try to claim ownership on the device;
3352 		 * if already taken, skip -- do not allow multiple PFs */
3353 		err = mlx4_get_ownership(dev);
3354 		if (err) {
3355 			if (err < 0)
3356 				return err;
3357 			else {
3358 				mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
3359 				return -EINVAL;
3360 			}
3361 		}
3362 
3363 		atomic_set(&priv->opreq_count, 0);
3364 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3365 
3366 		/*
3367 		 * Now reset the HCA before we touch the PCI capabilities or
3368 		 * attempt a firmware command, since a boot ROM may have left
3369 		 * the HCA in an undefined state.
3370 		 */
3371 		err = mlx4_reset(dev);
3372 		if (err) {
3373 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
3374 			goto err_sriov;
3375 		}
3376 
3377 		if (total_vfs) {
3378 			dev->flags = MLX4_FLAG_MASTER;
3379 			existing_vfs = pci_num_vf(pdev);
3380 			if (existing_vfs)
3381 				dev->flags |= MLX4_FLAG_SRIOV;
3382 			dev->persist->num_vfs = total_vfs;
3383 		}
3384 	}
3385 
3386 	/* on load remove any previous indication of internal error,
3387 	 * device is up.
3388 	 */
3389 	dev->persist->state = MLX4_DEVICE_STATE_UP;
3390 
3391 slave_start:
3392 	err = mlx4_cmd_init(dev);
3393 	if (err) {
3394 		mlx4_err(dev, "Failed to init command interface, aborting\n");
3395 		goto err_sriov;
3396 	}
3397 
3398 	/* In slave functions, the communication channel must be initialized
3399 	 * before posting commands. Also, init num_slaves before calling
3400 	 * mlx4_init_hca */
3401 	if (mlx4_is_mfunc(dev)) {
3402 		if (mlx4_is_master(dev)) {
3403 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
3404 
3405 		} else {
3406 			dev->num_slaves = 0;
3407 			err = mlx4_multi_func_init(dev);
3408 			if (err) {
3409 				mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
3410 				goto err_cmd;
3411 			}
3412 		}
3413 	}
3414 
3415 	err = mlx4_init_fw(dev);
3416 	if (err) {
3417 		mlx4_err(dev, "Failed to init fw, aborting.\n");
3418 		goto err_mfunc;
3419 	}
3420 
3421 	if (mlx4_is_master(dev)) {
3422 		/* when we hit the goto slave_start below, dev_cap already initialized */
3423 		if (!dev_cap) {
3424 			dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3425 
3426 			if (!dev_cap) {
3427 				err = -ENOMEM;
3428 				goto err_fw;
3429 			}
3430 
3431 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3432 			if (err) {
3433 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3434 				goto err_fw;
3435 			}
3436 
3437 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3438 				goto err_fw;
3439 
3440 			if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3441 				u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3442 								  total_vfs,
3443 								  existing_vfs,
3444 								  reset_flow);
3445 
3446 				mlx4_close_fw(dev);
3447 				mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3448 				dev->flags = dev_flags;
3449 				if (!SRIOV_VALID_STATE(dev->flags)) {
3450 					mlx4_err(dev, "Invalid SRIOV state\n");
3451 					goto err_sriov;
3452 				}
3453 				err = mlx4_reset(dev);
3454 				if (err) {
3455 					mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3456 					goto err_sriov;
3457 				}
3458 				goto slave_start;
3459 			}
3460 		} else {
3461 			/* Legacy mode FW requires SRIOV to be enabled before
3462 			 * doing QUERY_DEV_CAP, since max_eq's value is different if
3463 			 * SRIOV is enabled.
3464 			 */
3465 			memset(dev_cap, 0, sizeof(*dev_cap));
3466 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3467 			if (err) {
3468 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3469 				goto err_fw;
3470 			}
3471 
3472 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3473 				goto err_fw;
3474 		}
3475 	}
3476 
3477 	err = mlx4_init_hca(dev);
3478 	if (err) {
3479 		if (err == -EACCES) {
3480 			/* Not primary Physical function
3481 			 * Running in slave mode */
3482 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3483 			/* We're not a PF */
3484 			if (dev->flags & MLX4_FLAG_SRIOV) {
3485 				if (!existing_vfs)
3486 					pci_disable_sriov(pdev);
3487 				if (mlx4_is_master(dev) && !reset_flow)
3488 					atomic_dec(&pf_loading);
3489 				dev->flags &= ~MLX4_FLAG_SRIOV;
3490 			}
3491 			if (!mlx4_is_slave(dev))
3492 				mlx4_free_ownership(dev);
3493 			dev->flags |= MLX4_FLAG_SLAVE;
3494 			dev->flags &= ~MLX4_FLAG_MASTER;
3495 			goto slave_start;
3496 		} else
3497 			goto err_fw;
3498 	}
3499 
3500 	if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3501 		u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3502 						  existing_vfs, reset_flow);
3503 
3504 		if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3505 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3506 			dev->flags = dev_flags;
3507 			err = mlx4_cmd_init(dev);
3508 			if (err) {
3509 				/* Only VHCR is cleaned up, so could still
3510 				 * send FW commands
3511 				 */
3512 				mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3513 				goto err_close;
3514 			}
3515 		} else {
3516 			dev->flags = dev_flags;
3517 		}
3518 
3519 		if (!SRIOV_VALID_STATE(dev->flags)) {
3520 			mlx4_err(dev, "Invalid SRIOV state\n");
3521 			goto err_close;
3522 		}
3523 	}
3524 
3525 	/* check if the device is functioning at its maximum possible speed.
3526 	 * No return code for this call, just warn the user in case of PCI
3527 	 * express device capabilities are under-satisfied by the bus.
3528 	 */
3529 	if (!mlx4_is_slave(dev))
3530 		pcie_print_link_status(dev->persist->pdev);
3531 
3532 	/* In master functions, the communication channel must be initialized
3533 	 * after obtaining its address from fw */
3534 	if (mlx4_is_master(dev)) {
3535 		if (dev->caps.num_ports < 2 &&
3536 		    num_vfs_argc > 1) {
3537 			err = -EINVAL;
3538 			mlx4_err(dev,
3539 				 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3540 				 dev->caps.num_ports);
3541 			goto err_close;
3542 		}
3543 		memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
3544 
3545 		for (i = 0;
3546 		     i < sizeof(dev->persist->nvfs)/
3547 		     sizeof(dev->persist->nvfs[0]); i++) {
3548 			unsigned j;
3549 
3550 			for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
3551 				dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3552 				dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3553 					dev->caps.num_ports;
3554 			}
3555 		}
3556 
3557 		/* In master functions, the communication channel
3558 		 * must be initialized after obtaining its address from fw
3559 		 */
3560 		err = mlx4_multi_func_init(dev);
3561 		if (err) {
3562 			mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3563 			goto err_close;
3564 		}
3565 	}
3566 
3567 	err = mlx4_alloc_eq_table(dev);
3568 	if (err)
3569 		goto err_master_mfunc;
3570 
3571 	bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
3572 	mutex_init(&priv->msix_ctl.pool_lock);
3573 
3574 	mlx4_enable_msi_x(dev);
3575 	if ((mlx4_is_mfunc(dev)) &&
3576 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
3577 		err = -EOPNOTSUPP;
3578 		mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
3579 		goto err_free_eq;
3580 	}
3581 
3582 	if (!mlx4_is_slave(dev)) {
3583 		err = mlx4_init_steering(dev);
3584 		if (err)
3585 			goto err_disable_msix;
3586 	}
3587 
3588 	mlx4_init_quotas(dev);
3589 
3590 	err = mlx4_setup_hca(dev);
3591 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3592 	    !mlx4_is_mfunc(dev)) {
3593 		dev->flags &= ~MLX4_FLAG_MSI_X;
3594 		dev->caps.num_comp_vectors = 1;
3595 		pci_disable_msix(pdev);
3596 		err = mlx4_setup_hca(dev);
3597 	}
3598 
3599 	if (err)
3600 		goto err_steer;
3601 
3602 	/* When PF resources are ready arm its comm channel to enable
3603 	 * getting commands
3604 	 */
3605 	if (mlx4_is_master(dev)) {
3606 		err = mlx4_ARM_COMM_CHANNEL(dev);
3607 		if (err) {
3608 			mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3609 				 err);
3610 			goto err_steer;
3611 		}
3612 	}
3613 
3614 	for (port = 1; port <= dev->caps.num_ports; port++) {
3615 		err = mlx4_init_port_info(dev, port);
3616 		if (err)
3617 			goto err_port;
3618 	}
3619 
3620 	priv->v2p.port1 = 1;
3621 	priv->v2p.port2 = 2;
3622 
3623 	err = mlx4_register_device(dev);
3624 	if (err)
3625 		goto err_port;
3626 
3627 	mlx4_request_modules(dev);
3628 
3629 	mlx4_sense_init(dev);
3630 	mlx4_start_sense(dev);
3631 
3632 	priv->removed = 0;
3633 
3634 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3635 		atomic_dec(&pf_loading);
3636 
3637 	kfree(dev_cap);
3638 	return 0;
3639 
3640 err_port:
3641 	for (--port; port >= 1; --port)
3642 		mlx4_cleanup_port_info(&priv->port[port]);
3643 
3644 	mlx4_cleanup_default_counters(dev);
3645 	if (!mlx4_is_slave(dev))
3646 		mlx4_cleanup_counters_table(dev);
3647 	mlx4_cleanup_qp_table(dev);
3648 	mlx4_cleanup_srq_table(dev);
3649 	mlx4_cleanup_cq_table(dev);
3650 	mlx4_cmd_use_polling(dev);
3651 	mlx4_cleanup_eq_table(dev);
3652 	mlx4_cleanup_mcg_table(dev);
3653 	mlx4_cleanup_mr_table(dev);
3654 	mlx4_cleanup_xrcd_table(dev);
3655 	mlx4_cleanup_pd_table(dev);
3656 	mlx4_cleanup_uar_table(dev);
3657 
3658 err_steer:
3659 	if (!mlx4_is_slave(dev))
3660 		mlx4_clear_steering(dev);
3661 
3662 err_disable_msix:
3663 	if (dev->flags & MLX4_FLAG_MSI_X)
3664 		pci_disable_msix(pdev);
3665 
3666 err_free_eq:
3667 	mlx4_free_eq_table(dev);
3668 
3669 err_master_mfunc:
3670 	if (mlx4_is_master(dev)) {
3671 		mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
3672 		mlx4_multi_func_cleanup(dev);
3673 	}
3674 
3675 	if (mlx4_is_slave(dev))
3676 		mlx4_slave_destroy_special_qp_cap(dev);
3677 
3678 err_close:
3679 	mlx4_close_hca(dev);
3680 
3681 err_fw:
3682 	mlx4_close_fw(dev);
3683 
3684 err_mfunc:
3685 	if (mlx4_is_slave(dev))
3686 		mlx4_multi_func_cleanup(dev);
3687 
3688 err_cmd:
3689 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3690 
3691 err_sriov:
3692 	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3693 		pci_disable_sriov(pdev);
3694 		dev->flags &= ~MLX4_FLAG_SRIOV;
3695 	}
3696 
3697 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3698 		atomic_dec(&pf_loading);
3699 
3700 	kfree(priv->dev.dev_vfs);
3701 
3702 	if (!mlx4_is_slave(dev))
3703 		mlx4_free_ownership(dev);
3704 
3705 	kfree(dev_cap);
3706 	return err;
3707 }
3708 
3709 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3710 			   struct mlx4_priv *priv)
3711 {
3712 	int err;
3713 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3714 	int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3715 	const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3716 		{2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3717 	unsigned total_vfs = 0;
3718 	unsigned int i;
3719 
3720 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3721 
3722 	err = mlx4_pci_enable_device(&priv->dev);
3723 	if (err) {
3724 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3725 		return err;
3726 	}
3727 
3728 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3729 	 * per port, we must limit the number of VFs to 63 (since their are
3730 	 * 128 MACs)
3731 	 */
3732 	for (i = 0; i < ARRAY_SIZE(nvfs) && i < num_vfs_argc;
3733 	     total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3734 		nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3735 		if (nvfs[i] < 0) {
3736 			dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3737 			err = -EINVAL;
3738 			goto err_disable_pdev;
3739 		}
3740 	}
3741 	for (i = 0; i < ARRAY_SIZE(prb_vf) && i < probe_vfs_argc;
3742 	     i++) {
3743 		prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3744 		if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3745 			dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3746 			err = -EINVAL;
3747 			goto err_disable_pdev;
3748 		}
3749 	}
3750 	if (total_vfs > MLX4_MAX_NUM_VF) {
3751 		dev_err(&pdev->dev,
3752 			"Requested more VF's (%d) than allowed by hw (%d)\n",
3753 			total_vfs, MLX4_MAX_NUM_VF);
3754 		err = -EINVAL;
3755 		goto err_disable_pdev;
3756 	}
3757 
3758 	for (i = 0; i < MLX4_MAX_PORTS; i++) {
3759 		if (nvfs[i] + nvfs[2] > MLX4_MAX_NUM_VF_P_PORT) {
3760 			dev_err(&pdev->dev,
3761 				"Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3762 				nvfs[i] + nvfs[2], i + 1,
3763 				MLX4_MAX_NUM_VF_P_PORT);
3764 			err = -EINVAL;
3765 			goto err_disable_pdev;
3766 		}
3767 	}
3768 
3769 	/* Check for BARs. */
3770 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3771 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3772 		dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3773 			pci_dev_data, pci_resource_flags(pdev, 0));
3774 		err = -ENODEV;
3775 		goto err_disable_pdev;
3776 	}
3777 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3778 		dev_err(&pdev->dev, "Missing UAR, aborting\n");
3779 		err = -ENODEV;
3780 		goto err_disable_pdev;
3781 	}
3782 
3783 	err = pci_request_regions(pdev, DRV_NAME);
3784 	if (err) {
3785 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3786 		goto err_disable_pdev;
3787 	}
3788 
3789 	pci_set_master(pdev);
3790 
3791 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3792 	if (err) {
3793 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3794 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3795 		if (err) {
3796 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3797 			goto err_release_regions;
3798 		}
3799 	}
3800 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3801 	if (err) {
3802 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3803 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3804 		if (err) {
3805 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3806 			goto err_release_regions;
3807 		}
3808 	}
3809 
3810 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
3811 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3812 	/* Detect if this device is a virtual function */
3813 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3814 		/* When acting as pf, we normally skip vfs unless explicitly
3815 		 * requested to probe them.
3816 		 */
3817 		if (total_vfs) {
3818 			unsigned vfs_offset = 0;
3819 
3820 			for (i = 0; i < ARRAY_SIZE(nvfs) &&
3821 			     vfs_offset + nvfs[i] < extended_func_num(pdev);
3822 			     vfs_offset += nvfs[i], i++)
3823 				;
3824 			if (i == ARRAY_SIZE(nvfs)) {
3825 				err = -ENODEV;
3826 				goto err_release_regions;
3827 			}
3828 			if ((extended_func_num(pdev) - vfs_offset)
3829 			    > prb_vf[i]) {
3830 				dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3831 					 extended_func_num(pdev));
3832 				err = -ENODEV;
3833 				goto err_release_regions;
3834 			}
3835 		}
3836 	}
3837 
3838 	err = mlx4_crdump_init(&priv->dev);
3839 	if (err)
3840 		goto err_release_regions;
3841 
3842 	err = mlx4_catas_init(&priv->dev);
3843 	if (err)
3844 		goto err_crdump;
3845 
3846 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3847 	if (err)
3848 		goto err_catas;
3849 
3850 	return 0;
3851 
3852 err_catas:
3853 	mlx4_catas_end(&priv->dev);
3854 
3855 err_crdump:
3856 	mlx4_crdump_end(&priv->dev);
3857 
3858 err_release_regions:
3859 	pci_release_regions(pdev);
3860 
3861 err_disable_pdev:
3862 	mlx4_pci_disable_device(&priv->dev);
3863 	return err;
3864 }
3865 
3866 static int mlx4_devlink_port_type_set(struct devlink_port *devlink_port,
3867 				      enum devlink_port_type port_type)
3868 {
3869 	struct mlx4_port_info *info = container_of(devlink_port,
3870 						   struct mlx4_port_info,
3871 						   devlink_port);
3872 	enum mlx4_port_type mlx4_port_type;
3873 
3874 	switch (port_type) {
3875 	case DEVLINK_PORT_TYPE_AUTO:
3876 		mlx4_port_type = MLX4_PORT_TYPE_AUTO;
3877 		break;
3878 	case DEVLINK_PORT_TYPE_ETH:
3879 		mlx4_port_type = MLX4_PORT_TYPE_ETH;
3880 		break;
3881 	case DEVLINK_PORT_TYPE_IB:
3882 		mlx4_port_type = MLX4_PORT_TYPE_IB;
3883 		break;
3884 	default:
3885 		return -EOPNOTSUPP;
3886 	}
3887 
3888 	return __set_port_type(info, mlx4_port_type);
3889 }
3890 
3891 static void mlx4_devlink_param_load_driverinit_values(struct devlink *devlink)
3892 {
3893 	struct mlx4_priv *priv = devlink_priv(devlink);
3894 	struct mlx4_dev *dev = &priv->dev;
3895 	struct mlx4_fw_crdump *crdump = &dev->persist->crdump;
3896 	union devlink_param_value saved_value;
3897 	int err;
3898 
3899 	err = devlink_param_driverinit_value_get(devlink,
3900 						 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET,
3901 						 &saved_value);
3902 	if (!err && mlx4_internal_err_reset != saved_value.vbool) {
3903 		mlx4_internal_err_reset = saved_value.vbool;
3904 		/* Notify on value changed on runtime configuration mode */
3905 		devlink_param_value_changed(devlink,
3906 					    DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET);
3907 	}
3908 	err = devlink_param_driverinit_value_get(devlink,
3909 						 DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
3910 						 &saved_value);
3911 	if (!err)
3912 		log_num_mac = order_base_2(saved_value.vu32);
3913 	err = devlink_param_driverinit_value_get(devlink,
3914 						 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
3915 						 &saved_value);
3916 	if (!err)
3917 		enable_64b_cqe_eqe = saved_value.vbool;
3918 	err = devlink_param_driverinit_value_get(devlink,
3919 						 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
3920 						 &saved_value);
3921 	if (!err)
3922 		enable_4k_uar = saved_value.vbool;
3923 	err = devlink_param_driverinit_value_get(devlink,
3924 						 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT,
3925 						 &saved_value);
3926 	if (!err && crdump->snapshot_enable != saved_value.vbool) {
3927 		crdump->snapshot_enable = saved_value.vbool;
3928 		devlink_param_value_changed(devlink,
3929 					    DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT);
3930 	}
3931 }
3932 
3933 static void mlx4_restart_one_down(struct pci_dev *pdev);
3934 static int mlx4_restart_one_up(struct pci_dev *pdev, bool reload,
3935 			       struct devlink *devlink);
3936 
3937 static int mlx4_devlink_reload_down(struct devlink *devlink, bool netns_change,
3938 				    struct netlink_ext_ack *extack)
3939 {
3940 	struct mlx4_priv *priv = devlink_priv(devlink);
3941 	struct mlx4_dev *dev = &priv->dev;
3942 	struct mlx4_dev_persistent *persist = dev->persist;
3943 
3944 	if (netns_change) {
3945 		NL_SET_ERR_MSG_MOD(extack, "Namespace change is not supported");
3946 		return -EOPNOTSUPP;
3947 	}
3948 	if (persist->num_vfs)
3949 		mlx4_warn(persist->dev, "Reload performed on PF, will cause reset on operating Virtual Functions\n");
3950 	mlx4_restart_one_down(persist->pdev);
3951 	return 0;
3952 }
3953 
3954 static int mlx4_devlink_reload_up(struct devlink *devlink,
3955 				  struct netlink_ext_ack *extack)
3956 {
3957 	struct mlx4_priv *priv = devlink_priv(devlink);
3958 	struct mlx4_dev *dev = &priv->dev;
3959 	struct mlx4_dev_persistent *persist = dev->persist;
3960 	int err;
3961 
3962 	err = mlx4_restart_one_up(persist->pdev, true, devlink);
3963 	if (err)
3964 		mlx4_err(persist->dev, "mlx4_restart_one_up failed, ret=%d\n",
3965 			 err);
3966 
3967 	return err;
3968 }
3969 
3970 static const struct devlink_ops mlx4_devlink_ops = {
3971 	.port_type_set	= mlx4_devlink_port_type_set,
3972 	.reload_down	= mlx4_devlink_reload_down,
3973 	.reload_up	= mlx4_devlink_reload_up,
3974 };
3975 
3976 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3977 {
3978 	struct devlink *devlink;
3979 	struct mlx4_priv *priv;
3980 	struct mlx4_dev *dev;
3981 	int ret;
3982 
3983 	printk_once(KERN_INFO "%s", mlx4_version);
3984 
3985 	devlink = devlink_alloc(&mlx4_devlink_ops, sizeof(*priv));
3986 	if (!devlink)
3987 		return -ENOMEM;
3988 	priv = devlink_priv(devlink);
3989 
3990 	dev       = &priv->dev;
3991 	dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3992 	if (!dev->persist) {
3993 		ret = -ENOMEM;
3994 		goto err_devlink_free;
3995 	}
3996 	dev->persist->pdev = pdev;
3997 	dev->persist->dev = dev;
3998 	pci_set_drvdata(pdev, dev->persist);
3999 	priv->pci_dev_data = id->driver_data;
4000 	mutex_init(&dev->persist->device_state_mutex);
4001 	mutex_init(&dev->persist->interface_state_mutex);
4002 	mutex_init(&dev->persist->pci_status_mutex);
4003 
4004 	ret = devlink_register(devlink, &pdev->dev);
4005 	if (ret)
4006 		goto err_persist_free;
4007 	ret = devlink_params_register(devlink, mlx4_devlink_params,
4008 				      ARRAY_SIZE(mlx4_devlink_params));
4009 	if (ret)
4010 		goto err_devlink_unregister;
4011 	mlx4_devlink_set_params_init_values(devlink);
4012 	ret =  __mlx4_init_one(pdev, id->driver_data, priv);
4013 	if (ret)
4014 		goto err_params_unregister;
4015 
4016 	devlink_params_publish(devlink);
4017 	devlink_reload_enable(devlink);
4018 	pci_save_state(pdev);
4019 	return 0;
4020 
4021 err_params_unregister:
4022 	devlink_params_unregister(devlink, mlx4_devlink_params,
4023 				  ARRAY_SIZE(mlx4_devlink_params));
4024 err_devlink_unregister:
4025 	devlink_unregister(devlink);
4026 err_persist_free:
4027 	kfree(dev->persist);
4028 err_devlink_free:
4029 	devlink_free(devlink);
4030 	return ret;
4031 }
4032 
4033 static void mlx4_clean_dev(struct mlx4_dev *dev)
4034 {
4035 	struct mlx4_dev_persistent *persist = dev->persist;
4036 	struct mlx4_priv *priv = mlx4_priv(dev);
4037 	unsigned long	flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
4038 
4039 	memset(priv, 0, sizeof(*priv));
4040 	priv->dev.persist = persist;
4041 	priv->dev.flags = flags;
4042 }
4043 
4044 static void mlx4_unload_one(struct pci_dev *pdev)
4045 {
4046 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4047 	struct mlx4_dev  *dev  = persist->dev;
4048 	struct mlx4_priv *priv = mlx4_priv(dev);
4049 	int               pci_dev_data;
4050 	int p, i;
4051 
4052 	if (priv->removed)
4053 		return;
4054 
4055 	/* saving current ports type for further use */
4056 	for (i = 0; i < dev->caps.num_ports; i++) {
4057 		dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
4058 		dev->persist->curr_port_poss_type[i] = dev->caps.
4059 						       possible_type[i + 1];
4060 	}
4061 
4062 	pci_dev_data = priv->pci_dev_data;
4063 
4064 	mlx4_stop_sense(dev);
4065 	mlx4_unregister_device(dev);
4066 
4067 	for (p = 1; p <= dev->caps.num_ports; p++) {
4068 		mlx4_cleanup_port_info(&priv->port[p]);
4069 		mlx4_CLOSE_PORT(dev, p);
4070 	}
4071 
4072 	if (mlx4_is_master(dev))
4073 		mlx4_free_resource_tracker(dev,
4074 					   RES_TR_FREE_SLAVES_ONLY);
4075 
4076 	mlx4_cleanup_default_counters(dev);
4077 	if (!mlx4_is_slave(dev))
4078 		mlx4_cleanup_counters_table(dev);
4079 	mlx4_cleanup_qp_table(dev);
4080 	mlx4_cleanup_srq_table(dev);
4081 	mlx4_cleanup_cq_table(dev);
4082 	mlx4_cmd_use_polling(dev);
4083 	mlx4_cleanup_eq_table(dev);
4084 	mlx4_cleanup_mcg_table(dev);
4085 	mlx4_cleanup_mr_table(dev);
4086 	mlx4_cleanup_xrcd_table(dev);
4087 	mlx4_cleanup_pd_table(dev);
4088 
4089 	if (mlx4_is_master(dev))
4090 		mlx4_free_resource_tracker(dev,
4091 					   RES_TR_FREE_STRUCTS_ONLY);
4092 
4093 	iounmap(priv->kar);
4094 	mlx4_uar_free(dev, &priv->driver_uar);
4095 	mlx4_cleanup_uar_table(dev);
4096 	if (!mlx4_is_slave(dev))
4097 		mlx4_clear_steering(dev);
4098 	mlx4_free_eq_table(dev);
4099 	if (mlx4_is_master(dev))
4100 		mlx4_multi_func_cleanup(dev);
4101 	mlx4_close_hca(dev);
4102 	mlx4_close_fw(dev);
4103 	if (mlx4_is_slave(dev))
4104 		mlx4_multi_func_cleanup(dev);
4105 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
4106 
4107 	if (dev->flags & MLX4_FLAG_MSI_X)
4108 		pci_disable_msix(pdev);
4109 
4110 	if (!mlx4_is_slave(dev))
4111 		mlx4_free_ownership(dev);
4112 
4113 	mlx4_slave_destroy_special_qp_cap(dev);
4114 	kfree(dev->dev_vfs);
4115 
4116 	mlx4_clean_dev(dev);
4117 	priv->pci_dev_data = pci_dev_data;
4118 	priv->removed = 1;
4119 }
4120 
4121 static void mlx4_remove_one(struct pci_dev *pdev)
4122 {
4123 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4124 	struct mlx4_dev  *dev  = persist->dev;
4125 	struct mlx4_priv *priv = mlx4_priv(dev);
4126 	struct devlink *devlink = priv_to_devlink(priv);
4127 	int active_vfs = 0;
4128 
4129 	devlink_reload_disable(devlink);
4130 
4131 	if (mlx4_is_slave(dev))
4132 		persist->interface_state |= MLX4_INTERFACE_STATE_NOWAIT;
4133 
4134 	mutex_lock(&persist->interface_state_mutex);
4135 	persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
4136 	mutex_unlock(&persist->interface_state_mutex);
4137 
4138 	/* Disabling SR-IOV is not allowed while there are active vf's */
4139 	if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
4140 		active_vfs = mlx4_how_many_lives_vf(dev);
4141 		if (active_vfs) {
4142 			pr_warn("Removing PF when there are active VF's !!\n");
4143 			pr_warn("Will not disable SR-IOV.\n");
4144 		}
4145 	}
4146 
4147 	/* device marked to be under deletion running now without the lock
4148 	 * letting other tasks to be terminated
4149 	 */
4150 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4151 		mlx4_unload_one(pdev);
4152 	else
4153 		mlx4_info(dev, "%s: interface is down\n", __func__);
4154 	mlx4_catas_end(dev);
4155 	mlx4_crdump_end(dev);
4156 	if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
4157 		mlx4_warn(dev, "Disabling SR-IOV\n");
4158 		pci_disable_sriov(pdev);
4159 	}
4160 
4161 	pci_release_regions(pdev);
4162 	mlx4_pci_disable_device(dev);
4163 	devlink_params_unregister(devlink, mlx4_devlink_params,
4164 				  ARRAY_SIZE(mlx4_devlink_params));
4165 	devlink_unregister(devlink);
4166 	kfree(dev->persist);
4167 	devlink_free(devlink);
4168 }
4169 
4170 static int restore_current_port_types(struct mlx4_dev *dev,
4171 				      enum mlx4_port_type *types,
4172 				      enum mlx4_port_type *poss_types)
4173 {
4174 	struct mlx4_priv *priv = mlx4_priv(dev);
4175 	int err, i;
4176 
4177 	mlx4_stop_sense(dev);
4178 
4179 	mutex_lock(&priv->port_mutex);
4180 	for (i = 0; i < dev->caps.num_ports; i++)
4181 		dev->caps.possible_type[i + 1] = poss_types[i];
4182 	err = mlx4_change_port_types(dev, types);
4183 	mlx4_start_sense(dev);
4184 	mutex_unlock(&priv->port_mutex);
4185 
4186 	return err;
4187 }
4188 
4189 static void mlx4_restart_one_down(struct pci_dev *pdev)
4190 {
4191 	mlx4_unload_one(pdev);
4192 }
4193 
4194 static int mlx4_restart_one_up(struct pci_dev *pdev, bool reload,
4195 			       struct devlink *devlink)
4196 {
4197 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4198 	struct mlx4_dev	 *dev  = persist->dev;
4199 	struct mlx4_priv *priv = mlx4_priv(dev);
4200 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4201 	int pci_dev_data, err, total_vfs;
4202 
4203 	pci_dev_data = priv->pci_dev_data;
4204 	total_vfs = dev->persist->num_vfs;
4205 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4206 
4207 	if (reload)
4208 		mlx4_devlink_param_load_driverinit_values(devlink);
4209 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
4210 	if (err) {
4211 		mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
4212 			 __func__, pci_name(pdev), err);
4213 		return err;
4214 	}
4215 
4216 	err = restore_current_port_types(dev, dev->persist->curr_port_type,
4217 					 dev->persist->curr_port_poss_type);
4218 	if (err)
4219 		mlx4_err(dev, "could not restore original port types (%d)\n",
4220 			 err);
4221 
4222 	return err;
4223 }
4224 
4225 int mlx4_restart_one(struct pci_dev *pdev)
4226 {
4227 	mlx4_restart_one_down(pdev);
4228 	return mlx4_restart_one_up(pdev, false, NULL);
4229 }
4230 
4231 #define MLX_SP(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_FORCE_SENSE_PORT }
4232 #define MLX_VF(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_IS_VF }
4233 #define MLX_GN(id) { PCI_VDEVICE(MELLANOX, id), 0 }
4234 
4235 static const struct pci_device_id mlx4_pci_table[] = {
4236 #ifdef CONFIG_MLX4_CORE_GEN2
4237 	/* MT25408 "Hermon" */
4238 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_SDR),	/* SDR */
4239 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR),	/* DDR */
4240 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR),	/* QDR */
4241 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR_GEN2), /* DDR Gen2 */
4242 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR_GEN2),	/* QDR Gen2 */
4243 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN),	/* EN 10GigE */
4244 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN_GEN2),  /* EN 10GigE Gen2 */
4245 	/* MT25458 ConnectX EN 10GBASE-T */
4246 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN),
4247 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_T_GEN2),	/* Gen2 */
4248 	/* MT26468 ConnectX EN 10GigE PCIe Gen2*/
4249 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_GEN2),
4250 	/* MT26438 ConnectX EN 40GigE PCIe Gen2 5GT/s */
4251 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_5_GEN2),
4252 	/* MT26478 ConnectX2 40GigE PCIe Gen2 */
4253 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX2),
4254 	/* MT25400 Family [ConnectX-2] */
4255 	MLX_VF(0x1002),					/* Virtual Function */
4256 #endif /* CONFIG_MLX4_CORE_GEN2 */
4257 	/* MT27500 Family [ConnectX-3] */
4258 	MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3),
4259 	MLX_VF(0x1004),					/* Virtual Function */
4260 	MLX_GN(0x1005),					/* MT27510 Family */
4261 	MLX_GN(0x1006),					/* MT27511 Family */
4262 	MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3_PRO),	/* MT27520 Family */
4263 	MLX_GN(0x1008),					/* MT27521 Family */
4264 	MLX_GN(0x1009),					/* MT27530 Family */
4265 	MLX_GN(0x100a),					/* MT27531 Family */
4266 	MLX_GN(0x100b),					/* MT27540 Family */
4267 	MLX_GN(0x100c),					/* MT27541 Family */
4268 	MLX_GN(0x100d),					/* MT27550 Family */
4269 	MLX_GN(0x100e),					/* MT27551 Family */
4270 	MLX_GN(0x100f),					/* MT27560 Family */
4271 	MLX_GN(0x1010),					/* MT27561 Family */
4272 
4273 	/*
4274 	 * See the mellanox_check_broken_intx_masking() quirk when
4275 	 * adding devices
4276 	 */
4277 
4278 	{ 0, }
4279 };
4280 
4281 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
4282 
4283 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
4284 					      pci_channel_state_t state)
4285 {
4286 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4287 
4288 	mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
4289 	mlx4_enter_error_state(persist);
4290 
4291 	mutex_lock(&persist->interface_state_mutex);
4292 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4293 		mlx4_unload_one(pdev);
4294 
4295 	mutex_unlock(&persist->interface_state_mutex);
4296 	if (state == pci_channel_io_perm_failure)
4297 		return PCI_ERS_RESULT_DISCONNECT;
4298 
4299 	mlx4_pci_disable_device(persist->dev);
4300 	return PCI_ERS_RESULT_NEED_RESET;
4301 }
4302 
4303 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
4304 {
4305 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4306 	struct mlx4_dev	 *dev  = persist->dev;
4307 	int err;
4308 
4309 	mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
4310 	err = mlx4_pci_enable_device(dev);
4311 	if (err) {
4312 		mlx4_err(dev, "Can not re-enable device, err=%d\n", err);
4313 		return PCI_ERS_RESULT_DISCONNECT;
4314 	}
4315 
4316 	pci_set_master(pdev);
4317 	pci_restore_state(pdev);
4318 	pci_save_state(pdev);
4319 	return PCI_ERS_RESULT_RECOVERED;
4320 }
4321 
4322 static void mlx4_pci_resume(struct pci_dev *pdev)
4323 {
4324 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4325 	struct mlx4_dev	 *dev  = persist->dev;
4326 	struct mlx4_priv *priv = mlx4_priv(dev);
4327 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4328 	int total_vfs;
4329 	int err;
4330 
4331 	mlx4_err(dev, "%s was called\n", __func__);
4332 	total_vfs = dev->persist->num_vfs;
4333 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4334 
4335 	mutex_lock(&persist->interface_state_mutex);
4336 	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4337 		err = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
4338 				    priv, 1);
4339 		if (err) {
4340 			mlx4_err(dev, "%s: mlx4_load_one failed, err=%d\n",
4341 				 __func__,  err);
4342 			goto end;
4343 		}
4344 
4345 		err = restore_current_port_types(dev, dev->persist->
4346 						 curr_port_type, dev->persist->
4347 						 curr_port_poss_type);
4348 		if (err)
4349 			mlx4_err(dev, "could not restore original port types (%d)\n", err);
4350 	}
4351 end:
4352 	mutex_unlock(&persist->interface_state_mutex);
4353 
4354 }
4355 
4356 static void mlx4_shutdown(struct pci_dev *pdev)
4357 {
4358 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4359 
4360 	mlx4_info(persist->dev, "mlx4_shutdown was called\n");
4361 	mutex_lock(&persist->interface_state_mutex);
4362 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4363 		mlx4_unload_one(pdev);
4364 	mutex_unlock(&persist->interface_state_mutex);
4365 }
4366 
4367 static const struct pci_error_handlers mlx4_err_handler = {
4368 	.error_detected = mlx4_pci_err_detected,
4369 	.slot_reset     = mlx4_pci_slot_reset,
4370 	.resume		= mlx4_pci_resume,
4371 };
4372 
4373 static int mlx4_suspend(struct pci_dev *pdev, pm_message_t state)
4374 {
4375 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4376 	struct mlx4_dev	*dev = persist->dev;
4377 
4378 	mlx4_err(dev, "suspend was called\n");
4379 	mutex_lock(&persist->interface_state_mutex);
4380 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4381 		mlx4_unload_one(pdev);
4382 	mutex_unlock(&persist->interface_state_mutex);
4383 
4384 	return 0;
4385 }
4386 
4387 static int mlx4_resume(struct pci_dev *pdev)
4388 {
4389 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4390 	struct mlx4_dev	*dev = persist->dev;
4391 	struct mlx4_priv *priv = mlx4_priv(dev);
4392 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4393 	int total_vfs;
4394 	int ret = 0;
4395 
4396 	mlx4_err(dev, "resume was called\n");
4397 	total_vfs = dev->persist->num_vfs;
4398 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4399 
4400 	mutex_lock(&persist->interface_state_mutex);
4401 	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4402 		ret = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs,
4403 				    nvfs, priv, 1);
4404 		if (!ret) {
4405 			ret = restore_current_port_types(dev,
4406 					dev->persist->curr_port_type,
4407 					dev->persist->curr_port_poss_type);
4408 			if (ret)
4409 				mlx4_err(dev, "resume: could not restore original port types (%d)\n", ret);
4410 		}
4411 	}
4412 	mutex_unlock(&persist->interface_state_mutex);
4413 
4414 	return ret;
4415 }
4416 
4417 static struct pci_driver mlx4_driver = {
4418 	.name		= DRV_NAME,
4419 	.id_table	= mlx4_pci_table,
4420 	.probe		= mlx4_init_one,
4421 	.shutdown	= mlx4_shutdown,
4422 	.remove		= mlx4_remove_one,
4423 	.suspend	= mlx4_suspend,
4424 	.resume		= mlx4_resume,
4425 	.err_handler    = &mlx4_err_handler,
4426 };
4427 
4428 static int __init mlx4_verify_params(void)
4429 {
4430 	if (msi_x < 0) {
4431 		pr_warn("mlx4_core: bad msi_x: %d\n", msi_x);
4432 		return -1;
4433 	}
4434 
4435 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
4436 		pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
4437 		return -1;
4438 	}
4439 
4440 	if (log_num_vlan != 0)
4441 		pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
4442 			MLX4_LOG_NUM_VLANS);
4443 
4444 	if (use_prio != 0)
4445 		pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
4446 
4447 	if ((log_mtts_per_seg < 0) || (log_mtts_per_seg > 7)) {
4448 		pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
4449 			log_mtts_per_seg);
4450 		return -1;
4451 	}
4452 
4453 	/* Check if module param for ports type has legal combination */
4454 	if (port_type_array[0] == false && port_type_array[1] == true) {
4455 		pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
4456 		port_type_array[0] = true;
4457 	}
4458 
4459 	if (mlx4_log_num_mgm_entry_size < -7 ||
4460 	    (mlx4_log_num_mgm_entry_size > 0 &&
4461 	     (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
4462 	      mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
4463 		pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
4464 			mlx4_log_num_mgm_entry_size,
4465 			MLX4_MIN_MGM_LOG_ENTRY_SIZE,
4466 			MLX4_MAX_MGM_LOG_ENTRY_SIZE);
4467 		return -1;
4468 	}
4469 
4470 	return 0;
4471 }
4472 
4473 static int __init mlx4_init(void)
4474 {
4475 	int ret;
4476 
4477 	if (mlx4_verify_params())
4478 		return -EINVAL;
4479 
4480 
4481 	mlx4_wq = create_singlethread_workqueue("mlx4");
4482 	if (!mlx4_wq)
4483 		return -ENOMEM;
4484 
4485 	ret = pci_register_driver(&mlx4_driver);
4486 	if (ret < 0)
4487 		destroy_workqueue(mlx4_wq);
4488 	return ret < 0 ? ret : 0;
4489 }
4490 
4491 static void __exit mlx4_cleanup(void)
4492 {
4493 	pci_unregister_driver(&mlx4_driver);
4494 	destroy_workqueue(mlx4_wq);
4495 }
4496 
4497 module_init(mlx4_init);
4498 module_exit(mlx4_cleanup);
4499