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;
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_init_value(struct devlink *devlink, u32 param_id,
264 					union devlink_param_value init_val)
265 {
266 	struct mlx4_priv *priv = devlink_priv(devlink);
267 	struct mlx4_dev *dev = &priv->dev;
268 	int err;
269 
270 	err = devlink_param_driverinit_value_set(devlink, param_id, init_val);
271 	if (err)
272 		mlx4_warn(dev,
273 			  "devlink set parameter %u value failed (err = %d)",
274 			  param_id, err);
275 }
276 
277 static void mlx4_devlink_set_params_init_values(struct devlink *devlink)
278 {
279 	union devlink_param_value value;
280 
281 	value.vbool = !!mlx4_internal_err_reset;
282 	mlx4_devlink_set_init_value(devlink,
283 				    DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET,
284 				    value);
285 
286 	value.vu32 = 1UL << log_num_mac;
287 	mlx4_devlink_set_init_value(devlink,
288 				    DEVLINK_PARAM_GENERIC_ID_MAX_MACS, value);
289 
290 	value.vbool = enable_64b_cqe_eqe;
291 	mlx4_devlink_set_init_value(devlink,
292 				    MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
293 				    value);
294 
295 	value.vbool = enable_4k_uar;
296 	mlx4_devlink_set_init_value(devlink,
297 				    MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
298 				    value);
299 
300 	value.vbool = false;
301 	mlx4_devlink_set_init_value(devlink,
302 				    DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT,
303 				    value);
304 }
305 
306 static inline void mlx4_set_num_reserved_uars(struct mlx4_dev *dev,
307 					      struct mlx4_dev_cap *dev_cap)
308 {
309 	/* The reserved_uars is calculated by system page size unit.
310 	 * Therefore, adjustment is added when the uar page size is less
311 	 * than the system page size
312 	 */
313 	dev->caps.reserved_uars	=
314 		max_t(int,
315 		      mlx4_get_num_reserved_uar(dev),
316 		      dev_cap->reserved_uars /
317 			(1 << (PAGE_SHIFT - dev->uar_page_shift)));
318 }
319 
320 int mlx4_check_port_params(struct mlx4_dev *dev,
321 			   enum mlx4_port_type *port_type)
322 {
323 	int i;
324 
325 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
326 		for (i = 0; i < dev->caps.num_ports - 1; i++) {
327 			if (port_type[i] != port_type[i + 1]) {
328 				mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
329 				return -EINVAL;
330 			}
331 		}
332 	}
333 
334 	for (i = 0; i < dev->caps.num_ports; i++) {
335 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
336 			mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
337 				 i + 1);
338 			return -EINVAL;
339 		}
340 	}
341 	return 0;
342 }
343 
344 static void mlx4_set_port_mask(struct mlx4_dev *dev)
345 {
346 	int i;
347 
348 	for (i = 1; i <= dev->caps.num_ports; ++i)
349 		dev->caps.port_mask[i] = dev->caps.port_type[i];
350 }
351 
352 enum {
353 	MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
354 };
355 
356 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
357 {
358 	int err = 0;
359 	struct mlx4_func func;
360 
361 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
362 		err = mlx4_QUERY_FUNC(dev, &func, 0);
363 		if (err) {
364 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
365 			return err;
366 		}
367 		dev_cap->max_eqs = func.max_eq;
368 		dev_cap->reserved_eqs = func.rsvd_eqs;
369 		dev_cap->reserved_uars = func.rsvd_uars;
370 		err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
371 	}
372 	return err;
373 }
374 
375 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
376 {
377 	struct mlx4_caps *dev_cap = &dev->caps;
378 
379 	/* FW not supporting or cancelled by user */
380 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
381 	    !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
382 		return;
383 
384 	/* Must have 64B CQE_EQE enabled by FW to use bigger stride
385 	 * When FW has NCSI it may decide not to report 64B CQE/EQEs
386 	 */
387 	if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
388 	    !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
389 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
390 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
391 		return;
392 	}
393 
394 	if (cache_line_size() == 128 || cache_line_size() == 256) {
395 		mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
396 		/* Changing the real data inside CQE size to 32B */
397 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
398 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
399 
400 		if (mlx4_is_master(dev))
401 			dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
402 	} else {
403 		if (cache_line_size() != 32  && cache_line_size() != 64)
404 			mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
405 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
406 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
407 	}
408 }
409 
410 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
411 			  struct mlx4_port_cap *port_cap)
412 {
413 	dev->caps.vl_cap[port]	    = port_cap->max_vl;
414 	dev->caps.ib_mtu_cap[port]	    = port_cap->ib_mtu;
415 	dev->phys_caps.gid_phys_table_len[port]  = port_cap->max_gids;
416 	dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
417 	/* set gid and pkey table operating lengths by default
418 	 * to non-sriov values
419 	 */
420 	dev->caps.gid_table_len[port]  = port_cap->max_gids;
421 	dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
422 	dev->caps.port_width_cap[port] = port_cap->max_port_width;
423 	dev->caps.eth_mtu_cap[port]    = port_cap->eth_mtu;
424 	dev->caps.max_tc_eth	       = port_cap->max_tc_eth;
425 	dev->caps.def_mac[port]        = port_cap->def_mac;
426 	dev->caps.supported_type[port] = port_cap->supported_port_types;
427 	dev->caps.suggested_type[port] = port_cap->suggested_type;
428 	dev->caps.default_sense[port] = port_cap->default_sense;
429 	dev->caps.trans_type[port]	    = port_cap->trans_type;
430 	dev->caps.vendor_oui[port]     = port_cap->vendor_oui;
431 	dev->caps.wavelength[port]     = port_cap->wavelength;
432 	dev->caps.trans_code[port]     = port_cap->trans_code;
433 
434 	return 0;
435 }
436 
437 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
438 			 struct mlx4_port_cap *port_cap)
439 {
440 	int err = 0;
441 
442 	err = mlx4_QUERY_PORT(dev, port, port_cap);
443 
444 	if (err)
445 		mlx4_err(dev, "QUERY_PORT command failed.\n");
446 
447 	return err;
448 }
449 
450 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
451 {
452 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
453 		return;
454 
455 	if (mlx4_is_mfunc(dev)) {
456 		mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
457 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
458 		return;
459 	}
460 
461 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
462 		mlx4_dbg(dev,
463 			 "Keep FCS is not supported - Disabling Ignore FCS");
464 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
465 		return;
466 	}
467 }
468 
469 #define MLX4_A0_STEERING_TABLE_SIZE	256
470 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
471 {
472 	int err;
473 	int i;
474 
475 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
476 	if (err) {
477 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
478 		return err;
479 	}
480 	mlx4_dev_cap_dump(dev, dev_cap);
481 
482 	if (dev_cap->min_page_sz > PAGE_SIZE) {
483 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
484 			 dev_cap->min_page_sz, PAGE_SIZE);
485 		return -ENODEV;
486 	}
487 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
488 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
489 			 dev_cap->num_ports, MLX4_MAX_PORTS);
490 		return -ENODEV;
491 	}
492 
493 	if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
494 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
495 			 dev_cap->uar_size,
496 			 (unsigned long long)
497 			 pci_resource_len(dev->persist->pdev, 2));
498 		return -ENODEV;
499 	}
500 
501 	dev->caps.num_ports	     = dev_cap->num_ports;
502 	dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
503 	dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
504 				      dev->caps.num_sys_eqs :
505 				      MLX4_MAX_EQ_NUM;
506 	for (i = 1; i <= dev->caps.num_ports; ++i) {
507 		err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
508 		if (err) {
509 			mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
510 			return err;
511 		}
512 	}
513 
514 	dev->caps.uar_page_size	     = PAGE_SIZE;
515 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
516 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
517 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
518 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
519 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
520 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
521 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
522 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
523 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
524 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
525 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
526 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
527 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
528 	/*
529 	 * Subtract 1 from the limit because we need to allocate a
530 	 * spare CQE so the HCA HW can tell the difference between an
531 	 * empty CQ and a full CQ.
532 	 */
533 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
534 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
535 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
536 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
537 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
538 
539 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
540 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
541 					dev_cap->reserved_xrcds : 0;
542 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
543 					dev_cap->max_xrcds : 0;
544 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
545 
546 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
547 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
548 	dev->caps.flags		     = dev_cap->flags;
549 	dev->caps.flags2	     = dev_cap->flags2;
550 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
551 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
552 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
553 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
554 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
555 	dev->caps.wol_port[1]          = dev_cap->wol_port[1];
556 	dev->caps.wol_port[2]          = dev_cap->wol_port[2];
557 	dev->caps.health_buffer_addrs  = dev_cap->health_buffer_addrs;
558 
559 	/* Save uar page shift */
560 	if (!mlx4_is_slave(dev)) {
561 		/* Virtual PCI function needs to determine UAR page size from
562 		 * firmware. Only master PCI function can set the uar page size
563 		 */
564 		if (enable_4k_uar || !dev->persist->num_vfs)
565 			dev->uar_page_shift = DEFAULT_UAR_PAGE_SHIFT;
566 		else
567 			dev->uar_page_shift = PAGE_SHIFT;
568 
569 		mlx4_set_num_reserved_uars(dev, dev_cap);
570 	}
571 
572 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN) {
573 		struct mlx4_init_hca_param hca_param;
574 
575 		memset(&hca_param, 0, sizeof(hca_param));
576 		err = mlx4_QUERY_HCA(dev, &hca_param);
577 		/* Turn off PHV_EN flag in case phv_check_en is set.
578 		 * phv_check_en is a HW check that parse the packet and verify
579 		 * phv bit was reported correctly in the wqe. To allow QinQ
580 		 * PHV_EN flag should be set and phv_check_en must be cleared
581 		 * otherwise QinQ packets will be drop by the HW.
582 		 */
583 		if (err || hca_param.phv_check_en)
584 			dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_PHV_EN;
585 	}
586 
587 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
588 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
589 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
590 	/* Don't do sense port on multifunction devices (for now at least) */
591 	if (mlx4_is_mfunc(dev))
592 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
593 
594 	if (mlx4_low_memory_profile()) {
595 		dev->caps.log_num_macs  = MLX4_MIN_LOG_NUM_MAC;
596 		dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
597 	} else {
598 		dev->caps.log_num_macs  = log_num_mac;
599 		dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
600 	}
601 
602 	for (i = 1; i <= dev->caps.num_ports; ++i) {
603 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
604 		if (dev->caps.supported_type[i]) {
605 			/* if only ETH is supported - assign ETH */
606 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
607 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
608 			/* if only IB is supported, assign IB */
609 			else if (dev->caps.supported_type[i] ==
610 				 MLX4_PORT_TYPE_IB)
611 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
612 			else {
613 				/* if IB and ETH are supported, we set the port
614 				 * type according to user selection of port type;
615 				 * if user selected none, take the FW hint */
616 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
617 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
618 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
619 				else
620 					dev->caps.port_type[i] = port_type_array[i - 1];
621 			}
622 		}
623 		/*
624 		 * Link sensing is allowed on the port if 3 conditions are true:
625 		 * 1. Both protocols are supported on the port.
626 		 * 2. Different types are supported on the port
627 		 * 3. FW declared that it supports link sensing
628 		 */
629 		mlx4_priv(dev)->sense.sense_allowed[i] =
630 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
631 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
632 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
633 
634 		/*
635 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
636 		 * and perform sense_port FW command to try and set the correct
637 		 * port type from beginning
638 		 */
639 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
640 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
641 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
642 			mlx4_SENSE_PORT(dev, i, &sensed_port);
643 			if (sensed_port != MLX4_PORT_TYPE_NONE)
644 				dev->caps.port_type[i] = sensed_port;
645 		} else {
646 			dev->caps.possible_type[i] = dev->caps.port_type[i];
647 		}
648 
649 		if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
650 			dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
651 			mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
652 				  i, 1 << dev->caps.log_num_macs);
653 		}
654 		if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
655 			dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
656 			mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
657 				  i, 1 << dev->caps.log_num_vlans);
658 		}
659 	}
660 
661 	if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
662 	    (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
663 	    (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
664 		mlx4_warn(dev,
665 			  "Granular QoS per VF not supported with IB/Eth configuration\n");
666 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
667 	}
668 
669 	dev->caps.max_counters = dev_cap->max_counters;
670 
671 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
672 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
673 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
674 		(1 << dev->caps.log_num_macs) *
675 		(1 << dev->caps.log_num_vlans) *
676 		dev->caps.num_ports;
677 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
678 
679 	if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
680 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
681 		dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
682 	else
683 		dev->caps.dmfs_high_rate_qpn_base =
684 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
685 
686 	if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
687 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
688 		dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
689 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
690 		dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
691 	} else {
692 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
693 		dev->caps.dmfs_high_rate_qpn_base =
694 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
695 		dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
696 	}
697 
698 	dev->caps.rl_caps = dev_cap->rl_caps;
699 
700 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
701 		dev->caps.dmfs_high_rate_qpn_range;
702 
703 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
704 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
705 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
706 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
707 
708 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
709 
710 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
711 		if (dev_cap->flags &
712 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
713 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
714 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
715 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
716 		}
717 
718 		if (dev_cap->flags2 &
719 		    (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
720 		     MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
721 			mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
722 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
723 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
724 		}
725 	}
726 
727 	if ((dev->caps.flags &
728 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
729 	    mlx4_is_master(dev))
730 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
731 
732 	if (!mlx4_is_slave(dev)) {
733 		mlx4_enable_cqe_eqe_stride(dev);
734 		dev->caps.alloc_res_qp_mask =
735 			(dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
736 			MLX4_RESERVE_A0_QP;
737 
738 		if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
739 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
740 			mlx4_warn(dev, "Old device ETS support detected\n");
741 			mlx4_warn(dev, "Consider upgrading device FW.\n");
742 			dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
743 		}
744 
745 	} else {
746 		dev->caps.alloc_res_qp_mask = 0;
747 	}
748 
749 	mlx4_enable_ignore_fcs(dev);
750 
751 	return 0;
752 }
753 
754 /*The function checks if there are live vf, return the num of them*/
755 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
756 {
757 	struct mlx4_priv *priv = mlx4_priv(dev);
758 	struct mlx4_slave_state *s_state;
759 	int i;
760 	int ret = 0;
761 
762 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
763 		s_state = &priv->mfunc.master.slave_state[i];
764 		if (s_state->active && s_state->last_cmd !=
765 		    MLX4_COMM_CMD_RESET) {
766 			mlx4_warn(dev, "%s: slave: %d is still active\n",
767 				  __func__, i);
768 			ret++;
769 		}
770 	}
771 	return ret;
772 }
773 
774 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
775 {
776 	u32 qk = MLX4_RESERVED_QKEY_BASE;
777 
778 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
779 	    qpn < dev->phys_caps.base_proxy_sqpn)
780 		return -EINVAL;
781 
782 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
783 		/* tunnel qp */
784 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
785 	else
786 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
787 	*qkey = qk;
788 	return 0;
789 }
790 EXPORT_SYMBOL(mlx4_get_parav_qkey);
791 
792 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
793 {
794 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
795 
796 	if (!mlx4_is_master(dev))
797 		return;
798 
799 	priv->virt2phys_pkey[slave][port - 1][i] = val;
800 }
801 EXPORT_SYMBOL(mlx4_sync_pkey_table);
802 
803 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
804 {
805 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
806 
807 	if (!mlx4_is_master(dev))
808 		return;
809 
810 	priv->slave_node_guids[slave] = guid;
811 }
812 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
813 
814 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
815 {
816 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
817 
818 	if (!mlx4_is_master(dev))
819 		return 0;
820 
821 	return priv->slave_node_guids[slave];
822 }
823 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
824 
825 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
826 {
827 	struct mlx4_priv *priv = mlx4_priv(dev);
828 	struct mlx4_slave_state *s_slave;
829 
830 	if (!mlx4_is_master(dev))
831 		return 0;
832 
833 	s_slave = &priv->mfunc.master.slave_state[slave];
834 	return !!s_slave->active;
835 }
836 EXPORT_SYMBOL(mlx4_is_slave_active);
837 
838 void mlx4_handle_eth_header_mcast_prio(struct mlx4_net_trans_rule_hw_ctrl *ctrl,
839 				       struct _rule_hw *eth_header)
840 {
841 	if (is_multicast_ether_addr(eth_header->eth.dst_mac) ||
842 	    is_broadcast_ether_addr(eth_header->eth.dst_mac)) {
843 		struct mlx4_net_trans_rule_hw_eth *eth =
844 			(struct mlx4_net_trans_rule_hw_eth *)eth_header;
845 		struct _rule_hw *next_rule = (struct _rule_hw *)(eth + 1);
846 		bool last_rule = next_rule->size == 0 && next_rule->id == 0 &&
847 			next_rule->rsvd == 0;
848 
849 		if (last_rule)
850 			ctrl->prio = cpu_to_be16(MLX4_DOMAIN_NIC);
851 	}
852 }
853 EXPORT_SYMBOL(mlx4_handle_eth_header_mcast_prio);
854 
855 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
856 				       struct mlx4_dev_cap *dev_cap,
857 				       struct mlx4_init_hca_param *hca_param)
858 {
859 	dev->caps.steering_mode = hca_param->steering_mode;
860 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
861 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
862 		dev->caps.fs_log_max_ucast_qp_range_size =
863 			dev_cap->fs_log_max_ucast_qp_range_size;
864 	} else
865 		dev->caps.num_qp_per_mgm =
866 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
867 
868 	mlx4_dbg(dev, "Steering mode is: %s\n",
869 		 mlx4_steering_mode_str(dev->caps.steering_mode));
870 }
871 
872 static void mlx4_slave_destroy_special_qp_cap(struct mlx4_dev *dev)
873 {
874 	kfree(dev->caps.spec_qps);
875 	dev->caps.spec_qps = NULL;
876 }
877 
878 static int mlx4_slave_special_qp_cap(struct mlx4_dev *dev)
879 {
880 	struct mlx4_func_cap *func_cap = NULL;
881 	struct mlx4_caps *caps = &dev->caps;
882 	int i, err = 0;
883 
884 	func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
885 	caps->spec_qps = kcalloc(caps->num_ports, sizeof(*caps->spec_qps), GFP_KERNEL);
886 
887 	if (!func_cap || !caps->spec_qps) {
888 		mlx4_err(dev, "Failed to allocate memory for special qps cap\n");
889 		err = -ENOMEM;
890 		goto err_mem;
891 	}
892 
893 	for (i = 1; i <= caps->num_ports; ++i) {
894 		err = mlx4_QUERY_FUNC_CAP(dev, i, func_cap);
895 		if (err) {
896 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
897 				 i, err);
898 			goto err_mem;
899 		}
900 		caps->spec_qps[i - 1] = func_cap->spec_qps;
901 		caps->port_mask[i] = caps->port_type[i];
902 		caps->phys_port_id[i] = func_cap->phys_port_id;
903 		err = mlx4_get_slave_pkey_gid_tbl_len(dev, i,
904 						      &caps->gid_table_len[i],
905 						      &caps->pkey_table_len[i]);
906 		if (err) {
907 			mlx4_err(dev, "QUERY_PORT command failed for port %d, aborting (%d)\n",
908 				 i, err);
909 			goto err_mem;
910 		}
911 	}
912 
913 err_mem:
914 	if (err)
915 		mlx4_slave_destroy_special_qp_cap(dev);
916 	kfree(func_cap);
917 	return err;
918 }
919 
920 static int mlx4_slave_cap(struct mlx4_dev *dev)
921 {
922 	int			   err;
923 	u32			   page_size;
924 	struct mlx4_dev_cap	   *dev_cap = NULL;
925 	struct mlx4_func_cap	   *func_cap = NULL;
926 	struct mlx4_init_hca_param *hca_param = NULL;
927 
928 	hca_param = kzalloc(sizeof(*hca_param), GFP_KERNEL);
929 	func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
930 	dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
931 	if (!hca_param || !func_cap || !dev_cap) {
932 		mlx4_err(dev, "Failed to allocate memory for slave_cap\n");
933 		err = -ENOMEM;
934 		goto free_mem;
935 	}
936 
937 	err = mlx4_QUERY_HCA(dev, hca_param);
938 	if (err) {
939 		mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
940 		goto free_mem;
941 	}
942 
943 	/* fail if the hca has an unknown global capability
944 	 * at this time global_caps should be always zeroed
945 	 */
946 	if (hca_param->global_caps) {
947 		mlx4_err(dev, "Unknown hca global capabilities\n");
948 		err = -EINVAL;
949 		goto free_mem;
950 	}
951 
952 	dev->caps.hca_core_clock = hca_param->hca_core_clock;
953 
954 	dev->caps.max_qp_dest_rdma = 1 << hca_param->log_rd_per_qp;
955 	err = mlx4_dev_cap(dev, dev_cap);
956 	if (err) {
957 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
958 		goto free_mem;
959 	}
960 
961 	err = mlx4_QUERY_FW(dev);
962 	if (err)
963 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
964 
965 	page_size = ~dev->caps.page_size_cap + 1;
966 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
967 	if (page_size > PAGE_SIZE) {
968 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
969 			 page_size, PAGE_SIZE);
970 		err = -ENODEV;
971 		goto free_mem;
972 	}
973 
974 	/* Set uar_page_shift for VF */
975 	dev->uar_page_shift = hca_param->uar_page_sz + 12;
976 
977 	/* Make sure the master uar page size is valid */
978 	if (dev->uar_page_shift > PAGE_SHIFT) {
979 		mlx4_err(dev,
980 			 "Invalid configuration: uar page size is larger than system page size\n");
981 		err = -ENODEV;
982 		goto free_mem;
983 	}
984 
985 	/* Set reserved_uars based on the uar_page_shift */
986 	mlx4_set_num_reserved_uars(dev, dev_cap);
987 
988 	/* Although uar page size in FW differs from system page size,
989 	 * upper software layers (mlx4_ib, mlx4_en and part of mlx4_core)
990 	 * still works with assumption that uar page size == system page size
991 	 */
992 	dev->caps.uar_page_size = PAGE_SIZE;
993 
994 	err = mlx4_QUERY_FUNC_CAP(dev, 0, func_cap);
995 	if (err) {
996 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
997 			 err);
998 		goto free_mem;
999 	}
1000 
1001 	if ((func_cap->pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
1002 	    PF_CONTEXT_BEHAVIOUR_MASK) {
1003 		mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
1004 			 func_cap->pf_context_behaviour,
1005 			 PF_CONTEXT_BEHAVIOUR_MASK);
1006 		err = -EINVAL;
1007 		goto free_mem;
1008 	}
1009 
1010 	dev->caps.num_ports		= func_cap->num_ports;
1011 	dev->quotas.qp			= func_cap->qp_quota;
1012 	dev->quotas.srq			= func_cap->srq_quota;
1013 	dev->quotas.cq			= func_cap->cq_quota;
1014 	dev->quotas.mpt			= func_cap->mpt_quota;
1015 	dev->quotas.mtt			= func_cap->mtt_quota;
1016 	dev->caps.num_qps		= 1 << hca_param->log_num_qps;
1017 	dev->caps.num_srqs		= 1 << hca_param->log_num_srqs;
1018 	dev->caps.num_cqs		= 1 << hca_param->log_num_cqs;
1019 	dev->caps.num_mpts		= 1 << hca_param->log_mpt_sz;
1020 	dev->caps.num_eqs		= func_cap->max_eq;
1021 	dev->caps.reserved_eqs		= func_cap->reserved_eq;
1022 	dev->caps.reserved_lkey		= func_cap->reserved_lkey;
1023 	dev->caps.num_pds               = MLX4_NUM_PDS;
1024 	dev->caps.num_mgms              = 0;
1025 	dev->caps.num_amgms             = 0;
1026 
1027 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
1028 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
1029 			 dev->caps.num_ports, MLX4_MAX_PORTS);
1030 		err = -ENODEV;
1031 		goto free_mem;
1032 	}
1033 
1034 	mlx4_replace_zero_macs(dev);
1035 
1036 	err = mlx4_slave_special_qp_cap(dev);
1037 	if (err) {
1038 		mlx4_err(dev, "Set special QP caps failed. aborting\n");
1039 		goto free_mem;
1040 	}
1041 
1042 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
1043 				       dev->caps.reserved_uars) >
1044 				       pci_resource_len(dev->persist->pdev,
1045 							2)) {
1046 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
1047 			 dev->caps.uar_page_size * dev->caps.num_uars,
1048 			 (unsigned long long)
1049 			 pci_resource_len(dev->persist->pdev, 2));
1050 		err = -ENOMEM;
1051 		goto err_mem;
1052 	}
1053 
1054 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
1055 		dev->caps.eqe_size   = 64;
1056 		dev->caps.eqe_factor = 1;
1057 	} else {
1058 		dev->caps.eqe_size   = 32;
1059 		dev->caps.eqe_factor = 0;
1060 	}
1061 
1062 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
1063 		dev->caps.cqe_size   = 64;
1064 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1065 	} else {
1066 		dev->caps.cqe_size   = 32;
1067 	}
1068 
1069 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
1070 		dev->caps.eqe_size = hca_param->eqe_size;
1071 		dev->caps.eqe_factor = 0;
1072 	}
1073 
1074 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
1075 		dev->caps.cqe_size = hca_param->cqe_size;
1076 		/* User still need to know when CQE > 32B */
1077 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1078 	}
1079 
1080 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1081 	mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
1082 
1083 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_USER_MAC_EN;
1084 	mlx4_dbg(dev, "User MAC FW update is not supported in slave mode\n");
1085 
1086 	slave_adjust_steering_mode(dev, dev_cap, hca_param);
1087 	mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
1088 		 hca_param->rss_ip_frags ? "on" : "off");
1089 
1090 	if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
1091 	    dev->caps.bf_reg_size)
1092 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
1093 
1094 	if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
1095 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
1096 
1097 err_mem:
1098 	if (err)
1099 		mlx4_slave_destroy_special_qp_cap(dev);
1100 free_mem:
1101 	kfree(hca_param);
1102 	kfree(func_cap);
1103 	kfree(dev_cap);
1104 	return err;
1105 }
1106 
1107 static void mlx4_request_modules(struct mlx4_dev *dev)
1108 {
1109 	int port;
1110 	int has_ib_port = false;
1111 	int has_eth_port = false;
1112 #define EN_DRV_NAME	"mlx4_en"
1113 #define IB_DRV_NAME	"mlx4_ib"
1114 
1115 	for (port = 1; port <= dev->caps.num_ports; port++) {
1116 		if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
1117 			has_ib_port = true;
1118 		else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
1119 			has_eth_port = true;
1120 	}
1121 
1122 	if (has_eth_port)
1123 		request_module_nowait(EN_DRV_NAME);
1124 	if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
1125 		request_module_nowait(IB_DRV_NAME);
1126 }
1127 
1128 /*
1129  * Change the port configuration of the device.
1130  * Every user of this function must hold the port mutex.
1131  */
1132 int mlx4_change_port_types(struct mlx4_dev *dev,
1133 			   enum mlx4_port_type *port_types)
1134 {
1135 	int err = 0;
1136 	int change = 0;
1137 	int port;
1138 
1139 	for (port = 0; port <  dev->caps.num_ports; port++) {
1140 		/* Change the port type only if the new type is different
1141 		 * from the current, and not set to Auto */
1142 		if (port_types[port] != dev->caps.port_type[port + 1])
1143 			change = 1;
1144 	}
1145 	if (change) {
1146 		mlx4_unregister_device(dev);
1147 		for (port = 1; port <= dev->caps.num_ports; port++) {
1148 			mlx4_CLOSE_PORT(dev, port);
1149 			dev->caps.port_type[port] = port_types[port - 1];
1150 			err = mlx4_SET_PORT(dev, port, -1);
1151 			if (err) {
1152 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1153 					 port);
1154 				goto out;
1155 			}
1156 		}
1157 		mlx4_set_port_mask(dev);
1158 		err = mlx4_register_device(dev);
1159 		if (err) {
1160 			mlx4_err(dev, "Failed to register device\n");
1161 			goto out;
1162 		}
1163 		mlx4_request_modules(dev);
1164 	}
1165 
1166 out:
1167 	return err;
1168 }
1169 
1170 static ssize_t show_port_type(struct device *dev,
1171 			      struct device_attribute *attr,
1172 			      char *buf)
1173 {
1174 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1175 						   port_attr);
1176 	struct mlx4_dev *mdev = info->dev;
1177 	char type[8];
1178 
1179 	sprintf(type, "%s",
1180 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1181 		"ib" : "eth");
1182 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1183 		sprintf(buf, "auto (%s)\n", type);
1184 	else
1185 		sprintf(buf, "%s\n", type);
1186 
1187 	return strlen(buf);
1188 }
1189 
1190 static int __set_port_type(struct mlx4_port_info *info,
1191 			   enum mlx4_port_type port_type)
1192 {
1193 	struct mlx4_dev *mdev = info->dev;
1194 	struct mlx4_priv *priv = mlx4_priv(mdev);
1195 	enum mlx4_port_type types[MLX4_MAX_PORTS];
1196 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1197 	int i;
1198 	int err = 0;
1199 
1200 	if ((port_type & mdev->caps.supported_type[info->port]) != port_type) {
1201 		mlx4_err(mdev,
1202 			 "Requested port type for port %d is not supported on this HCA\n",
1203 			 info->port);
1204 		err = -EINVAL;
1205 		goto err_sup;
1206 	}
1207 
1208 	mlx4_stop_sense(mdev);
1209 	mutex_lock(&priv->port_mutex);
1210 	info->tmp_type = port_type;
1211 
1212 	/* Possible type is always the one that was delivered */
1213 	mdev->caps.possible_type[info->port] = info->tmp_type;
1214 
1215 	for (i = 0; i < mdev->caps.num_ports; i++) {
1216 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1217 					mdev->caps.possible_type[i+1];
1218 		if (types[i] == MLX4_PORT_TYPE_AUTO)
1219 			types[i] = mdev->caps.port_type[i+1];
1220 	}
1221 
1222 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1223 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1224 		for (i = 1; i <= mdev->caps.num_ports; i++) {
1225 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1226 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1227 				err = -EINVAL;
1228 			}
1229 		}
1230 	}
1231 	if (err) {
1232 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1233 		goto out;
1234 	}
1235 
1236 	mlx4_do_sense_ports(mdev, new_types, types);
1237 
1238 	err = mlx4_check_port_params(mdev, new_types);
1239 	if (err)
1240 		goto out;
1241 
1242 	/* We are about to apply the changes after the configuration
1243 	 * was verified, no need to remember the temporary types
1244 	 * any more */
1245 	for (i = 0; i < mdev->caps.num_ports; i++)
1246 		priv->port[i + 1].tmp_type = 0;
1247 
1248 	err = mlx4_change_port_types(mdev, new_types);
1249 
1250 out:
1251 	mlx4_start_sense(mdev);
1252 	mutex_unlock(&priv->port_mutex);
1253 err_sup:
1254 	return err;
1255 }
1256 
1257 static ssize_t set_port_type(struct device *dev,
1258 			     struct device_attribute *attr,
1259 			     const char *buf, size_t count)
1260 {
1261 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1262 						   port_attr);
1263 	struct mlx4_dev *mdev = info->dev;
1264 	enum mlx4_port_type port_type;
1265 	static DEFINE_MUTEX(set_port_type_mutex);
1266 	int err;
1267 
1268 	mutex_lock(&set_port_type_mutex);
1269 
1270 	if (!strcmp(buf, "ib\n")) {
1271 		port_type = MLX4_PORT_TYPE_IB;
1272 	} else if (!strcmp(buf, "eth\n")) {
1273 		port_type = MLX4_PORT_TYPE_ETH;
1274 	} else if (!strcmp(buf, "auto\n")) {
1275 		port_type = MLX4_PORT_TYPE_AUTO;
1276 	} else {
1277 		mlx4_err(mdev, "%s is not supported port type\n", buf);
1278 		err = -EINVAL;
1279 		goto err_out;
1280 	}
1281 
1282 	err = __set_port_type(info, port_type);
1283 
1284 err_out:
1285 	mutex_unlock(&set_port_type_mutex);
1286 
1287 	return err ? err : count;
1288 }
1289 
1290 enum ibta_mtu {
1291 	IB_MTU_256  = 1,
1292 	IB_MTU_512  = 2,
1293 	IB_MTU_1024 = 3,
1294 	IB_MTU_2048 = 4,
1295 	IB_MTU_4096 = 5
1296 };
1297 
1298 static inline int int_to_ibta_mtu(int mtu)
1299 {
1300 	switch (mtu) {
1301 	case 256:  return IB_MTU_256;
1302 	case 512:  return IB_MTU_512;
1303 	case 1024: return IB_MTU_1024;
1304 	case 2048: return IB_MTU_2048;
1305 	case 4096: return IB_MTU_4096;
1306 	default: return -1;
1307 	}
1308 }
1309 
1310 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1311 {
1312 	switch (mtu) {
1313 	case IB_MTU_256:  return  256;
1314 	case IB_MTU_512:  return  512;
1315 	case IB_MTU_1024: return 1024;
1316 	case IB_MTU_2048: return 2048;
1317 	case IB_MTU_4096: return 4096;
1318 	default: return -1;
1319 	}
1320 }
1321 
1322 static ssize_t show_port_ib_mtu(struct device *dev,
1323 			     struct device_attribute *attr,
1324 			     char *buf)
1325 {
1326 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1327 						   port_mtu_attr);
1328 	struct mlx4_dev *mdev = info->dev;
1329 
1330 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1331 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1332 
1333 	sprintf(buf, "%d\n",
1334 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1335 	return strlen(buf);
1336 }
1337 
1338 static ssize_t set_port_ib_mtu(struct device *dev,
1339 			     struct device_attribute *attr,
1340 			     const char *buf, size_t count)
1341 {
1342 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1343 						   port_mtu_attr);
1344 	struct mlx4_dev *mdev = info->dev;
1345 	struct mlx4_priv *priv = mlx4_priv(mdev);
1346 	int err, port, mtu, ibta_mtu = -1;
1347 
1348 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1349 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1350 		return -EINVAL;
1351 	}
1352 
1353 	err = kstrtoint(buf, 0, &mtu);
1354 	if (!err)
1355 		ibta_mtu = int_to_ibta_mtu(mtu);
1356 
1357 	if (err || ibta_mtu < 0) {
1358 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1359 		return -EINVAL;
1360 	}
1361 
1362 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1363 
1364 	mlx4_stop_sense(mdev);
1365 	mutex_lock(&priv->port_mutex);
1366 	mlx4_unregister_device(mdev);
1367 	for (port = 1; port <= mdev->caps.num_ports; port++) {
1368 		mlx4_CLOSE_PORT(mdev, port);
1369 		err = mlx4_SET_PORT(mdev, port, -1);
1370 		if (err) {
1371 			mlx4_err(mdev, "Failed to set port %d, aborting\n",
1372 				 port);
1373 			goto err_set_port;
1374 		}
1375 	}
1376 	err = mlx4_register_device(mdev);
1377 err_set_port:
1378 	mutex_unlock(&priv->port_mutex);
1379 	mlx4_start_sense(mdev);
1380 	return err ? err : count;
1381 }
1382 
1383 /* bond for multi-function device */
1384 #define MAX_MF_BOND_ALLOWED_SLAVES 63
1385 static int mlx4_mf_bond(struct mlx4_dev *dev)
1386 {
1387 	int err = 0;
1388 	int nvfs;
1389 	struct mlx4_slaves_pport slaves_port1;
1390 	struct mlx4_slaves_pport slaves_port2;
1391 	DECLARE_BITMAP(slaves_port_1_2, MLX4_MFUNC_MAX);
1392 
1393 	slaves_port1 = mlx4_phys_to_slaves_pport(dev, 1);
1394 	slaves_port2 = mlx4_phys_to_slaves_pport(dev, 2);
1395 	bitmap_and(slaves_port_1_2,
1396 		   slaves_port1.slaves, slaves_port2.slaves,
1397 		   dev->persist->num_vfs + 1);
1398 
1399 	/* only single port vfs are allowed */
1400 	if (bitmap_weight(slaves_port_1_2, dev->persist->num_vfs + 1) > 1) {
1401 		mlx4_warn(dev, "HA mode unsupported for dual ported VFs\n");
1402 		return -EINVAL;
1403 	}
1404 
1405 	/* number of virtual functions is number of total functions minus one
1406 	 * physical function for each port.
1407 	 */
1408 	nvfs = bitmap_weight(slaves_port1.slaves, dev->persist->num_vfs + 1) +
1409 		bitmap_weight(slaves_port2.slaves, dev->persist->num_vfs + 1) - 2;
1410 
1411 	/* limit on maximum allowed VFs */
1412 	if (nvfs > MAX_MF_BOND_ALLOWED_SLAVES) {
1413 		mlx4_warn(dev, "HA mode is not supported for %d VFs (max %d are allowed)\n",
1414 			  nvfs, MAX_MF_BOND_ALLOWED_SLAVES);
1415 		return -EINVAL;
1416 	}
1417 
1418 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) {
1419 		mlx4_warn(dev, "HA mode unsupported for NON DMFS steering\n");
1420 		return -EINVAL;
1421 	}
1422 
1423 	err = mlx4_bond_mac_table(dev);
1424 	if (err)
1425 		return err;
1426 	err = mlx4_bond_vlan_table(dev);
1427 	if (err)
1428 		goto err1;
1429 	err = mlx4_bond_fs_rules(dev);
1430 	if (err)
1431 		goto err2;
1432 
1433 	return 0;
1434 err2:
1435 	(void)mlx4_unbond_vlan_table(dev);
1436 err1:
1437 	(void)mlx4_unbond_mac_table(dev);
1438 	return err;
1439 }
1440 
1441 static int mlx4_mf_unbond(struct mlx4_dev *dev)
1442 {
1443 	int ret, ret1;
1444 
1445 	ret = mlx4_unbond_fs_rules(dev);
1446 	if (ret)
1447 		mlx4_warn(dev, "multifunction unbond for flow rules failed (%d)\n", ret);
1448 	ret1 = mlx4_unbond_mac_table(dev);
1449 	if (ret1) {
1450 		mlx4_warn(dev, "multifunction unbond for MAC table failed (%d)\n", ret1);
1451 		ret = ret1;
1452 	}
1453 	ret1 = mlx4_unbond_vlan_table(dev);
1454 	if (ret1) {
1455 		mlx4_warn(dev, "multifunction unbond for VLAN table failed (%d)\n", ret1);
1456 		ret = ret1;
1457 	}
1458 	return ret;
1459 }
1460 
1461 int mlx4_bond(struct mlx4_dev *dev)
1462 {
1463 	int ret = 0;
1464 	struct mlx4_priv *priv = mlx4_priv(dev);
1465 
1466 	mutex_lock(&priv->bond_mutex);
1467 
1468 	if (!mlx4_is_bonded(dev)) {
1469 		ret = mlx4_do_bond(dev, true);
1470 		if (ret)
1471 			mlx4_err(dev, "Failed to bond device: %d\n", ret);
1472 		if (!ret && mlx4_is_master(dev)) {
1473 			ret = mlx4_mf_bond(dev);
1474 			if (ret) {
1475 				mlx4_err(dev, "bond for multifunction failed\n");
1476 				mlx4_do_bond(dev, false);
1477 			}
1478 		}
1479 	}
1480 
1481 	mutex_unlock(&priv->bond_mutex);
1482 	if (!ret)
1483 		mlx4_dbg(dev, "Device is bonded\n");
1484 
1485 	return ret;
1486 }
1487 EXPORT_SYMBOL_GPL(mlx4_bond);
1488 
1489 int mlx4_unbond(struct mlx4_dev *dev)
1490 {
1491 	int ret = 0;
1492 	struct mlx4_priv *priv = mlx4_priv(dev);
1493 
1494 	mutex_lock(&priv->bond_mutex);
1495 
1496 	if (mlx4_is_bonded(dev)) {
1497 		int ret2 = 0;
1498 
1499 		ret = mlx4_do_bond(dev, false);
1500 		if (ret)
1501 			mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1502 		if (mlx4_is_master(dev))
1503 			ret2 = mlx4_mf_unbond(dev);
1504 		if (ret2) {
1505 			mlx4_warn(dev, "Failed to unbond device for multifunction (%d)\n", ret2);
1506 			ret = ret2;
1507 		}
1508 	}
1509 
1510 	mutex_unlock(&priv->bond_mutex);
1511 	if (!ret)
1512 		mlx4_dbg(dev, "Device is unbonded\n");
1513 
1514 	return ret;
1515 }
1516 EXPORT_SYMBOL_GPL(mlx4_unbond);
1517 
1518 
1519 int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1520 {
1521 	u8 port1 = v2p->port1;
1522 	u8 port2 = v2p->port2;
1523 	struct mlx4_priv *priv = mlx4_priv(dev);
1524 	int err;
1525 
1526 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1527 		return -EOPNOTSUPP;
1528 
1529 	mutex_lock(&priv->bond_mutex);
1530 
1531 	/* zero means keep current mapping for this port */
1532 	if (port1 == 0)
1533 		port1 = priv->v2p.port1;
1534 	if (port2 == 0)
1535 		port2 = priv->v2p.port2;
1536 
1537 	if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1538 	    (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1539 	    (port1 == 2 && port2 == 1)) {
1540 		/* besides boundary checks cross mapping makes
1541 		 * no sense and therefore not allowed */
1542 		err = -EINVAL;
1543 	} else if ((port1 == priv->v2p.port1) &&
1544 		 (port2 == priv->v2p.port2)) {
1545 		err = 0;
1546 	} else {
1547 		err = mlx4_virt2phy_port_map(dev, port1, port2);
1548 		if (!err) {
1549 			mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1550 				 port1, port2);
1551 			priv->v2p.port1 = port1;
1552 			priv->v2p.port2 = port2;
1553 		} else {
1554 			mlx4_err(dev, "Failed to change port mape: %d\n", err);
1555 		}
1556 	}
1557 
1558 	mutex_unlock(&priv->bond_mutex);
1559 	return err;
1560 }
1561 EXPORT_SYMBOL_GPL(mlx4_port_map_set);
1562 
1563 static int mlx4_load_fw(struct mlx4_dev *dev)
1564 {
1565 	struct mlx4_priv *priv = mlx4_priv(dev);
1566 	int err;
1567 
1568 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1569 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
1570 	if (!priv->fw.fw_icm) {
1571 		mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1572 		return -ENOMEM;
1573 	}
1574 
1575 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1576 	if (err) {
1577 		mlx4_err(dev, "MAP_FA command failed, aborting\n");
1578 		goto err_free;
1579 	}
1580 
1581 	err = mlx4_RUN_FW(dev);
1582 	if (err) {
1583 		mlx4_err(dev, "RUN_FW command failed, aborting\n");
1584 		goto err_unmap_fa;
1585 	}
1586 
1587 	return 0;
1588 
1589 err_unmap_fa:
1590 	mlx4_UNMAP_FA(dev);
1591 
1592 err_free:
1593 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1594 	return err;
1595 }
1596 
1597 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1598 				int cmpt_entry_sz)
1599 {
1600 	struct mlx4_priv *priv = mlx4_priv(dev);
1601 	int err;
1602 	int num_eqs;
1603 
1604 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1605 				  cmpt_base +
1606 				  ((u64) (MLX4_CMPT_TYPE_QP *
1607 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1608 				  cmpt_entry_sz, dev->caps.num_qps,
1609 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1610 				  0, 0);
1611 	if (err)
1612 		goto err;
1613 
1614 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1615 				  cmpt_base +
1616 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1617 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1618 				  cmpt_entry_sz, dev->caps.num_srqs,
1619 				  dev->caps.reserved_srqs, 0, 0);
1620 	if (err)
1621 		goto err_qp;
1622 
1623 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1624 				  cmpt_base +
1625 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1626 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1627 				  cmpt_entry_sz, dev->caps.num_cqs,
1628 				  dev->caps.reserved_cqs, 0, 0);
1629 	if (err)
1630 		goto err_srq;
1631 
1632 	num_eqs = dev->phys_caps.num_phys_eqs;
1633 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1634 				  cmpt_base +
1635 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1636 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1637 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1638 	if (err)
1639 		goto err_cq;
1640 
1641 	return 0;
1642 
1643 err_cq:
1644 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1645 
1646 err_srq:
1647 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1648 
1649 err_qp:
1650 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1651 
1652 err:
1653 	return err;
1654 }
1655 
1656 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1657 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1658 {
1659 	struct mlx4_priv *priv = mlx4_priv(dev);
1660 	u64 aux_pages;
1661 	int num_eqs;
1662 	int err;
1663 
1664 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1665 	if (err) {
1666 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1667 		return err;
1668 	}
1669 
1670 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1671 		 (unsigned long long) icm_size >> 10,
1672 		 (unsigned long long) aux_pages << 2);
1673 
1674 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1675 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1676 	if (!priv->fw.aux_icm) {
1677 		mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1678 		return -ENOMEM;
1679 	}
1680 
1681 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1682 	if (err) {
1683 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1684 		goto err_free_aux;
1685 	}
1686 
1687 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1688 	if (err) {
1689 		mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1690 		goto err_unmap_aux;
1691 	}
1692 
1693 
1694 	num_eqs = dev->phys_caps.num_phys_eqs;
1695 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1696 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1697 				  num_eqs, num_eqs, 0, 0);
1698 	if (err) {
1699 		mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1700 		goto err_unmap_cmpt;
1701 	}
1702 
1703 	/*
1704 	 * Reserved MTT entries must be aligned up to a cacheline
1705 	 * boundary, since the FW will write to them, while the driver
1706 	 * writes to all other MTT entries. (The variable
1707 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1708 	 * size, not the raw entry size)
1709 	 */
1710 	dev->caps.reserved_mtts =
1711 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1712 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1713 
1714 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1715 				  init_hca->mtt_base,
1716 				  dev->caps.mtt_entry_sz,
1717 				  dev->caps.num_mtts,
1718 				  dev->caps.reserved_mtts, 1, 0);
1719 	if (err) {
1720 		mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1721 		goto err_unmap_eq;
1722 	}
1723 
1724 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1725 				  init_hca->dmpt_base,
1726 				  dev_cap->dmpt_entry_sz,
1727 				  dev->caps.num_mpts,
1728 				  dev->caps.reserved_mrws, 1, 1);
1729 	if (err) {
1730 		mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1731 		goto err_unmap_mtt;
1732 	}
1733 
1734 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1735 				  init_hca->qpc_base,
1736 				  dev_cap->qpc_entry_sz,
1737 				  dev->caps.num_qps,
1738 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1739 				  0, 0);
1740 	if (err) {
1741 		mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1742 		goto err_unmap_dmpt;
1743 	}
1744 
1745 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1746 				  init_hca->auxc_base,
1747 				  dev_cap->aux_entry_sz,
1748 				  dev->caps.num_qps,
1749 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1750 				  0, 0);
1751 	if (err) {
1752 		mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1753 		goto err_unmap_qp;
1754 	}
1755 
1756 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1757 				  init_hca->altc_base,
1758 				  dev_cap->altc_entry_sz,
1759 				  dev->caps.num_qps,
1760 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1761 				  0, 0);
1762 	if (err) {
1763 		mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1764 		goto err_unmap_auxc;
1765 	}
1766 
1767 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1768 				  init_hca->rdmarc_base,
1769 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1770 				  dev->caps.num_qps,
1771 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1772 				  0, 0);
1773 	if (err) {
1774 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1775 		goto err_unmap_altc;
1776 	}
1777 
1778 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1779 				  init_hca->cqc_base,
1780 				  dev_cap->cqc_entry_sz,
1781 				  dev->caps.num_cqs,
1782 				  dev->caps.reserved_cqs, 0, 0);
1783 	if (err) {
1784 		mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1785 		goto err_unmap_rdmarc;
1786 	}
1787 
1788 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1789 				  init_hca->srqc_base,
1790 				  dev_cap->srq_entry_sz,
1791 				  dev->caps.num_srqs,
1792 				  dev->caps.reserved_srqs, 0, 0);
1793 	if (err) {
1794 		mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1795 		goto err_unmap_cq;
1796 	}
1797 
1798 	/*
1799 	 * For flow steering device managed mode it is required to use
1800 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1801 	 * required, but for simplicity just map the whole multicast
1802 	 * group table now.  The table isn't very big and it's a lot
1803 	 * easier than trying to track ref counts.
1804 	 */
1805 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1806 				  init_hca->mc_base,
1807 				  mlx4_get_mgm_entry_size(dev),
1808 				  dev->caps.num_mgms + dev->caps.num_amgms,
1809 				  dev->caps.num_mgms + dev->caps.num_amgms,
1810 				  0, 0);
1811 	if (err) {
1812 		mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1813 		goto err_unmap_srq;
1814 	}
1815 
1816 	return 0;
1817 
1818 err_unmap_srq:
1819 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1820 
1821 err_unmap_cq:
1822 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1823 
1824 err_unmap_rdmarc:
1825 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1826 
1827 err_unmap_altc:
1828 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1829 
1830 err_unmap_auxc:
1831 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1832 
1833 err_unmap_qp:
1834 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1835 
1836 err_unmap_dmpt:
1837 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1838 
1839 err_unmap_mtt:
1840 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1841 
1842 err_unmap_eq:
1843 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1844 
1845 err_unmap_cmpt:
1846 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1847 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1848 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1849 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1850 
1851 err_unmap_aux:
1852 	mlx4_UNMAP_ICM_AUX(dev);
1853 
1854 err_free_aux:
1855 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1856 
1857 	return err;
1858 }
1859 
1860 static void mlx4_free_icms(struct mlx4_dev *dev)
1861 {
1862 	struct mlx4_priv *priv = mlx4_priv(dev);
1863 
1864 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1865 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1866 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1867 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1868 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1869 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1870 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1871 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1872 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1873 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1874 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1875 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1876 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1877 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1878 
1879 	mlx4_UNMAP_ICM_AUX(dev);
1880 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1881 }
1882 
1883 static void mlx4_slave_exit(struct mlx4_dev *dev)
1884 {
1885 	struct mlx4_priv *priv = mlx4_priv(dev);
1886 
1887 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1888 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1889 			  MLX4_COMM_TIME))
1890 		mlx4_warn(dev, "Failed to close slave function\n");
1891 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1892 }
1893 
1894 static int map_bf_area(struct mlx4_dev *dev)
1895 {
1896 	struct mlx4_priv *priv = mlx4_priv(dev);
1897 	resource_size_t bf_start;
1898 	resource_size_t bf_len;
1899 	int err = 0;
1900 
1901 	if (!dev->caps.bf_reg_size)
1902 		return -ENXIO;
1903 
1904 	bf_start = pci_resource_start(dev->persist->pdev, 2) +
1905 			(dev->caps.num_uars << PAGE_SHIFT);
1906 	bf_len = pci_resource_len(dev->persist->pdev, 2) -
1907 			(dev->caps.num_uars << PAGE_SHIFT);
1908 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1909 	if (!priv->bf_mapping)
1910 		err = -ENOMEM;
1911 
1912 	return err;
1913 }
1914 
1915 static void unmap_bf_area(struct mlx4_dev *dev)
1916 {
1917 	if (mlx4_priv(dev)->bf_mapping)
1918 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1919 }
1920 
1921 u64 mlx4_read_clock(struct mlx4_dev *dev)
1922 {
1923 	u32 clockhi, clocklo, clockhi1;
1924 	u64 cycles;
1925 	int i;
1926 	struct mlx4_priv *priv = mlx4_priv(dev);
1927 
1928 	for (i = 0; i < 10; i++) {
1929 		clockhi = swab32(readl(priv->clock_mapping));
1930 		clocklo = swab32(readl(priv->clock_mapping + 4));
1931 		clockhi1 = swab32(readl(priv->clock_mapping));
1932 		if (clockhi == clockhi1)
1933 			break;
1934 	}
1935 
1936 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1937 
1938 	return cycles;
1939 }
1940 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1941 
1942 
1943 static int map_internal_clock(struct mlx4_dev *dev)
1944 {
1945 	struct mlx4_priv *priv = mlx4_priv(dev);
1946 
1947 	priv->clock_mapping =
1948 		ioremap(pci_resource_start(dev->persist->pdev,
1949 					   priv->fw.clock_bar) +
1950 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1951 
1952 	if (!priv->clock_mapping)
1953 		return -ENOMEM;
1954 
1955 	return 0;
1956 }
1957 
1958 int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1959 				   struct mlx4_clock_params *params)
1960 {
1961 	struct mlx4_priv *priv = mlx4_priv(dev);
1962 
1963 	if (mlx4_is_slave(dev))
1964 		return -EOPNOTSUPP;
1965 
1966 	if (!params)
1967 		return -EINVAL;
1968 
1969 	params->bar = priv->fw.clock_bar;
1970 	params->offset = priv->fw.clock_offset;
1971 	params->size = MLX4_CLOCK_SIZE;
1972 
1973 	return 0;
1974 }
1975 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1976 
1977 static void unmap_internal_clock(struct mlx4_dev *dev)
1978 {
1979 	struct mlx4_priv *priv = mlx4_priv(dev);
1980 
1981 	if (priv->clock_mapping)
1982 		iounmap(priv->clock_mapping);
1983 }
1984 
1985 static void mlx4_close_hca(struct mlx4_dev *dev)
1986 {
1987 	unmap_internal_clock(dev);
1988 	unmap_bf_area(dev);
1989 	if (mlx4_is_slave(dev))
1990 		mlx4_slave_exit(dev);
1991 	else {
1992 		mlx4_CLOSE_HCA(dev, 0);
1993 		mlx4_free_icms(dev);
1994 	}
1995 }
1996 
1997 static void mlx4_close_fw(struct mlx4_dev *dev)
1998 {
1999 	if (!mlx4_is_slave(dev)) {
2000 		mlx4_UNMAP_FA(dev);
2001 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
2002 	}
2003 }
2004 
2005 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
2006 {
2007 #define COMM_CHAN_OFFLINE_OFFSET 0x09
2008 
2009 	u32 comm_flags;
2010 	u32 offline_bit;
2011 	unsigned long end;
2012 	struct mlx4_priv *priv = mlx4_priv(dev);
2013 
2014 	end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
2015 	while (time_before(jiffies, end)) {
2016 		comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
2017 					  MLX4_COMM_CHAN_FLAGS));
2018 		offline_bit = (comm_flags &
2019 			       (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
2020 		if (!offline_bit)
2021 			return 0;
2022 
2023 		/* If device removal has been requested,
2024 		 * do not continue retrying.
2025 		 */
2026 		if (dev->persist->interface_state &
2027 		    MLX4_INTERFACE_STATE_NOWAIT)
2028 			break;
2029 
2030 		/* There are cases as part of AER/Reset flow that PF needs
2031 		 * around 100 msec to load. We therefore sleep for 100 msec
2032 		 * to allow other tasks to make use of that CPU during this
2033 		 * time interval.
2034 		 */
2035 		msleep(100);
2036 	}
2037 	mlx4_err(dev, "Communication channel is offline.\n");
2038 	return -EIO;
2039 }
2040 
2041 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
2042 {
2043 #define COMM_CHAN_RST_OFFSET 0x1e
2044 
2045 	struct mlx4_priv *priv = mlx4_priv(dev);
2046 	u32 comm_rst;
2047 	u32 comm_caps;
2048 
2049 	comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
2050 				 MLX4_COMM_CHAN_CAPS));
2051 	comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
2052 
2053 	if (comm_rst)
2054 		dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
2055 }
2056 
2057 static int mlx4_init_slave(struct mlx4_dev *dev)
2058 {
2059 	struct mlx4_priv *priv = mlx4_priv(dev);
2060 	u64 dma = (u64) priv->mfunc.vhcr_dma;
2061 	int ret_from_reset = 0;
2062 	u32 slave_read;
2063 	u32 cmd_channel_ver;
2064 
2065 	if (atomic_read(&pf_loading)) {
2066 		mlx4_warn(dev, "PF is not ready - Deferring probe\n");
2067 		return -EPROBE_DEFER;
2068 	}
2069 
2070 	mutex_lock(&priv->cmd.slave_cmd_mutex);
2071 	priv->cmd.max_cmds = 1;
2072 	if (mlx4_comm_check_offline(dev)) {
2073 		mlx4_err(dev, "PF is not responsive, skipping initialization\n");
2074 		goto err_offline;
2075 	}
2076 
2077 	mlx4_reset_vf_support(dev);
2078 	mlx4_warn(dev, "Sending reset\n");
2079 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
2080 				       MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
2081 	/* if we are in the middle of flr the slave will try
2082 	 * NUM_OF_RESET_RETRIES times before leaving.*/
2083 	if (ret_from_reset) {
2084 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
2085 			mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
2086 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
2087 			return -EPROBE_DEFER;
2088 		} else
2089 			goto err;
2090 	}
2091 
2092 	/* check the driver version - the slave I/F revision
2093 	 * must match the master's */
2094 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
2095 	cmd_channel_ver = mlx4_comm_get_version();
2096 
2097 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
2098 		MLX4_COMM_GET_IF_REV(slave_read)) {
2099 		mlx4_err(dev, "slave driver version is not supported by the master\n");
2100 		goto err;
2101 	}
2102 
2103 	mlx4_warn(dev, "Sending vhcr0\n");
2104 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
2105 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2106 		goto err;
2107 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
2108 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2109 		goto err;
2110 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
2111 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2112 		goto err;
2113 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
2114 			  MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2115 		goto err;
2116 
2117 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2118 	return 0;
2119 
2120 err:
2121 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
2122 err_offline:
2123 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2124 	return -EIO;
2125 }
2126 
2127 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
2128 {
2129 	int i;
2130 
2131 	for (i = 1; i <= dev->caps.num_ports; i++) {
2132 		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
2133 			dev->caps.gid_table_len[i] =
2134 				mlx4_get_slave_num_gids(dev, 0, i);
2135 		else
2136 			dev->caps.gid_table_len[i] = 1;
2137 		dev->caps.pkey_table_len[i] =
2138 			dev->phys_caps.pkey_phys_table_len[i] - 1;
2139 	}
2140 }
2141 
2142 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
2143 {
2144 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
2145 
2146 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
2147 	      i++) {
2148 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
2149 			break;
2150 	}
2151 
2152 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
2153 }
2154 
2155 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
2156 {
2157 	switch (dmfs_high_steer_mode) {
2158 	case MLX4_STEERING_DMFS_A0_DEFAULT:
2159 		return "default performance";
2160 
2161 	case MLX4_STEERING_DMFS_A0_DYNAMIC:
2162 		return "dynamic hybrid mode";
2163 
2164 	case MLX4_STEERING_DMFS_A0_STATIC:
2165 		return "performance optimized for limited rule configuration (static)";
2166 
2167 	case MLX4_STEERING_DMFS_A0_DISABLE:
2168 		return "disabled performance optimized steering";
2169 
2170 	case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
2171 		return "performance optimized steering not supported";
2172 
2173 	default:
2174 		return "Unrecognized mode";
2175 	}
2176 }
2177 
2178 #define MLX4_DMFS_A0_STEERING			(1UL << 2)
2179 
2180 static void choose_steering_mode(struct mlx4_dev *dev,
2181 				 struct mlx4_dev_cap *dev_cap)
2182 {
2183 	if (mlx4_log_num_mgm_entry_size <= 0) {
2184 		if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
2185 			if (dev->caps.dmfs_high_steer_mode ==
2186 			    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2187 				mlx4_err(dev, "DMFS high rate mode not supported\n");
2188 			else
2189 				dev->caps.dmfs_high_steer_mode =
2190 					MLX4_STEERING_DMFS_A0_STATIC;
2191 		}
2192 	}
2193 
2194 	if (mlx4_log_num_mgm_entry_size <= 0 &&
2195 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
2196 	    (!mlx4_is_mfunc(dev) ||
2197 	     (dev_cap->fs_max_num_qp_per_entry >=
2198 	     (dev->persist->num_vfs + 1))) &&
2199 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
2200 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
2201 		dev->oper_log_mgm_entry_size =
2202 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
2203 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
2204 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
2205 		dev->caps.fs_log_max_ucast_qp_range_size =
2206 			dev_cap->fs_log_max_ucast_qp_range_size;
2207 	} else {
2208 		if (dev->caps.dmfs_high_steer_mode !=
2209 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2210 			dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
2211 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
2212 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2213 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
2214 		else {
2215 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
2216 
2217 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
2218 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2219 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
2220 		}
2221 		dev->oper_log_mgm_entry_size =
2222 			mlx4_log_num_mgm_entry_size > 0 ?
2223 			mlx4_log_num_mgm_entry_size :
2224 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
2225 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
2226 	}
2227 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
2228 		 mlx4_steering_mode_str(dev->caps.steering_mode),
2229 		 dev->oper_log_mgm_entry_size,
2230 		 mlx4_log_num_mgm_entry_size);
2231 }
2232 
2233 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
2234 				       struct mlx4_dev_cap *dev_cap)
2235 {
2236 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2237 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
2238 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
2239 	else
2240 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
2241 
2242 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
2243 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
2244 }
2245 
2246 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
2247 {
2248 	int i;
2249 	struct mlx4_port_cap port_cap;
2250 
2251 	if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2252 		return -EINVAL;
2253 
2254 	for (i = 1; i <= dev->caps.num_ports; i++) {
2255 		if (mlx4_dev_port(dev, i, &port_cap)) {
2256 			mlx4_err(dev,
2257 				 "QUERY_DEV_CAP command failed, can't veify DMFS high rate steering.\n");
2258 		} else if ((dev->caps.dmfs_high_steer_mode !=
2259 			    MLX4_STEERING_DMFS_A0_DEFAULT) &&
2260 			   (port_cap.dmfs_optimized_state ==
2261 			    !!(dev->caps.dmfs_high_steer_mode ==
2262 			    MLX4_STEERING_DMFS_A0_DISABLE))) {
2263 			mlx4_err(dev,
2264 				 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2265 				 dmfs_high_rate_steering_mode_str(
2266 					dev->caps.dmfs_high_steer_mode),
2267 				 (port_cap.dmfs_optimized_state ?
2268 					"enabled" : "disabled"));
2269 		}
2270 	}
2271 
2272 	return 0;
2273 }
2274 
2275 static int mlx4_init_fw(struct mlx4_dev *dev)
2276 {
2277 	struct mlx4_mod_stat_cfg   mlx4_cfg;
2278 	int err = 0;
2279 
2280 	if (!mlx4_is_slave(dev)) {
2281 		err = mlx4_QUERY_FW(dev);
2282 		if (err) {
2283 			if (err == -EACCES)
2284 				mlx4_info(dev, "non-primary physical function, skipping\n");
2285 			else
2286 				mlx4_err(dev, "QUERY_FW command failed, aborting\n");
2287 			return err;
2288 		}
2289 
2290 		err = mlx4_load_fw(dev);
2291 		if (err) {
2292 			mlx4_err(dev, "Failed to start FW, aborting\n");
2293 			return err;
2294 		}
2295 
2296 		mlx4_cfg.log_pg_sz_m = 1;
2297 		mlx4_cfg.log_pg_sz = 0;
2298 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2299 		if (err)
2300 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2301 	}
2302 
2303 	return err;
2304 }
2305 
2306 static int mlx4_init_hca(struct mlx4_dev *dev)
2307 {
2308 	struct mlx4_priv	  *priv = mlx4_priv(dev);
2309 	struct mlx4_adapter	   adapter;
2310 	struct mlx4_dev_cap	   dev_cap;
2311 	struct mlx4_profile	   profile;
2312 	struct mlx4_init_hca_param init_hca;
2313 	u64 icm_size;
2314 	struct mlx4_config_dev_params params;
2315 	int err;
2316 
2317 	if (!mlx4_is_slave(dev)) {
2318 		err = mlx4_dev_cap(dev, &dev_cap);
2319 		if (err) {
2320 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
2321 			return err;
2322 		}
2323 
2324 		choose_steering_mode(dev, &dev_cap);
2325 		choose_tunnel_offload_mode(dev, &dev_cap);
2326 
2327 		if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2328 		    mlx4_is_master(dev))
2329 			dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2330 
2331 		err = mlx4_get_phys_port_id(dev);
2332 		if (err)
2333 			mlx4_err(dev, "Fail to get physical port id\n");
2334 
2335 		if (mlx4_is_master(dev))
2336 			mlx4_parav_master_pf_caps(dev);
2337 
2338 		if (mlx4_low_memory_profile()) {
2339 			mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2340 			profile = low_mem_profile;
2341 		} else {
2342 			profile = default_profile;
2343 		}
2344 		if (dev->caps.steering_mode ==
2345 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
2346 			profile.num_mcg = MLX4_FS_NUM_MCG;
2347 
2348 		icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
2349 					     &init_hca);
2350 		if ((long long) icm_size < 0) {
2351 			err = icm_size;
2352 			return err;
2353 		}
2354 
2355 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
2356 
2357 		if (enable_4k_uar || !dev->persist->num_vfs) {
2358 			init_hca.log_uar_sz = ilog2(dev->caps.num_uars) +
2359 						    PAGE_SHIFT - DEFAULT_UAR_PAGE_SHIFT;
2360 			init_hca.uar_page_sz = DEFAULT_UAR_PAGE_SHIFT - 12;
2361 		} else {
2362 			init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
2363 			init_hca.uar_page_sz = PAGE_SHIFT - 12;
2364 		}
2365 
2366 		init_hca.mw_enabled = 0;
2367 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2368 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2369 			init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
2370 
2371 		err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
2372 		if (err)
2373 			return err;
2374 
2375 		err = mlx4_INIT_HCA(dev, &init_hca);
2376 		if (err) {
2377 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
2378 			goto err_free_icm;
2379 		}
2380 
2381 		if (dev_cap.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2382 			err = mlx4_query_func(dev, &dev_cap);
2383 			if (err < 0) {
2384 				mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
2385 				goto err_close;
2386 			} else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2387 				dev->caps.num_eqs = dev_cap.max_eqs;
2388 				dev->caps.reserved_eqs = dev_cap.reserved_eqs;
2389 				dev->caps.reserved_uars = dev_cap.reserved_uars;
2390 			}
2391 		}
2392 
2393 		/*
2394 		 * If TS is supported by FW
2395 		 * read HCA frequency by QUERY_HCA command
2396 		 */
2397 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2398 			memset(&init_hca, 0, sizeof(init_hca));
2399 			err = mlx4_QUERY_HCA(dev, &init_hca);
2400 			if (err) {
2401 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
2402 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2403 			} else {
2404 				dev->caps.hca_core_clock =
2405 					init_hca.hca_core_clock;
2406 			}
2407 
2408 			/* In case we got HCA frequency 0 - disable timestamping
2409 			 * to avoid dividing by zero
2410 			 */
2411 			if (!dev->caps.hca_core_clock) {
2412 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2413 				mlx4_err(dev,
2414 					 "HCA frequency is 0 - timestamping is not supported\n");
2415 			} else if (map_internal_clock(dev)) {
2416 				/*
2417 				 * Map internal clock,
2418 				 * in case of failure disable timestamping
2419 				 */
2420 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2421 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
2422 			}
2423 		}
2424 
2425 		if (dev->caps.dmfs_high_steer_mode !=
2426 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2427 			if (mlx4_validate_optimized_steering(dev))
2428 				mlx4_warn(dev, "Optimized steering validation failed\n");
2429 
2430 			if (dev->caps.dmfs_high_steer_mode ==
2431 			    MLX4_STEERING_DMFS_A0_DISABLE) {
2432 				dev->caps.dmfs_high_rate_qpn_base =
2433 					dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2434 				dev->caps.dmfs_high_rate_qpn_range =
2435 					MLX4_A0_STEERING_TABLE_SIZE;
2436 			}
2437 
2438 			mlx4_info(dev, "DMFS high rate steer mode is: %s\n",
2439 				  dmfs_high_rate_steering_mode_str(
2440 					dev->caps.dmfs_high_steer_mode));
2441 		}
2442 	} else {
2443 		err = mlx4_init_slave(dev);
2444 		if (err) {
2445 			if (err != -EPROBE_DEFER)
2446 				mlx4_err(dev, "Failed to initialize slave\n");
2447 			return err;
2448 		}
2449 
2450 		err = mlx4_slave_cap(dev);
2451 		if (err) {
2452 			mlx4_err(dev, "Failed to obtain slave caps\n");
2453 			goto err_close;
2454 		}
2455 	}
2456 
2457 	if (map_bf_area(dev))
2458 		mlx4_dbg(dev, "Failed to map blue flame area\n");
2459 
2460 	/*Only the master set the ports, all the rest got it from it.*/
2461 	if (!mlx4_is_slave(dev))
2462 		mlx4_set_port_mask(dev);
2463 
2464 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
2465 	if (err) {
2466 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2467 		goto unmap_bf;
2468 	}
2469 
2470 	/* Query CONFIG_DEV parameters */
2471 	err = mlx4_config_dev_retrieval(dev, &params);
2472 	if (err && err != -EOPNOTSUPP) {
2473 		mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2474 	} else if (!err) {
2475 		dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2476 		dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2477 	}
2478 	priv->eq_table.inta_pin = adapter.inta_pin;
2479 	memcpy(dev->board_id, adapter.board_id, sizeof(dev->board_id));
2480 
2481 	return 0;
2482 
2483 unmap_bf:
2484 	unmap_internal_clock(dev);
2485 	unmap_bf_area(dev);
2486 
2487 	if (mlx4_is_slave(dev))
2488 		mlx4_slave_destroy_special_qp_cap(dev);
2489 
2490 err_close:
2491 	if (mlx4_is_slave(dev))
2492 		mlx4_slave_exit(dev);
2493 	else
2494 		mlx4_CLOSE_HCA(dev, 0);
2495 
2496 err_free_icm:
2497 	if (!mlx4_is_slave(dev))
2498 		mlx4_free_icms(dev);
2499 
2500 	return err;
2501 }
2502 
2503 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2504 {
2505 	struct mlx4_priv *priv = mlx4_priv(dev);
2506 	int nent_pow2;
2507 
2508 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2509 		return -ENOENT;
2510 
2511 	if (!dev->caps.max_counters)
2512 		return -ENOSPC;
2513 
2514 	nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2515 	/* reserve last counter index for sink counter */
2516 	return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2517 				nent_pow2 - 1, 0,
2518 				nent_pow2 - dev->caps.max_counters + 1);
2519 }
2520 
2521 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2522 {
2523 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2524 		return;
2525 
2526 	if (!dev->caps.max_counters)
2527 		return;
2528 
2529 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2530 }
2531 
2532 static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2533 {
2534 	struct mlx4_priv *priv = mlx4_priv(dev);
2535 	int port;
2536 
2537 	for (port = 0; port < dev->caps.num_ports; port++)
2538 		if (priv->def_counter[port] != -1)
2539 			mlx4_counter_free(dev,  priv->def_counter[port]);
2540 }
2541 
2542 static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2543 {
2544 	struct mlx4_priv *priv = mlx4_priv(dev);
2545 	int port, err = 0;
2546 	u32 idx;
2547 
2548 	for (port = 0; port < dev->caps.num_ports; port++)
2549 		priv->def_counter[port] = -1;
2550 
2551 	for (port = 0; port < dev->caps.num_ports; port++) {
2552 		err = mlx4_counter_alloc(dev, &idx, MLX4_RES_USAGE_DRIVER);
2553 
2554 		if (!err || err == -ENOSPC) {
2555 			priv->def_counter[port] = idx;
2556 		} else if (err == -ENOENT) {
2557 			err = 0;
2558 			continue;
2559 		} else if (mlx4_is_slave(dev) && err == -EINVAL) {
2560 			priv->def_counter[port] = MLX4_SINK_COUNTER_INDEX(dev);
2561 			mlx4_warn(dev, "can't allocate counter from old PF driver, using index %d\n",
2562 				  MLX4_SINK_COUNTER_INDEX(dev));
2563 			err = 0;
2564 		} else {
2565 			mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2566 				 __func__, port + 1, err);
2567 			mlx4_cleanup_default_counters(dev);
2568 			return err;
2569 		}
2570 
2571 		mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2572 			 __func__, priv->def_counter[port], port + 1);
2573 	}
2574 
2575 	return err;
2576 }
2577 
2578 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2579 {
2580 	struct mlx4_priv *priv = mlx4_priv(dev);
2581 
2582 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2583 		return -ENOENT;
2584 
2585 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2586 	if (*idx == -1) {
2587 		*idx = MLX4_SINK_COUNTER_INDEX(dev);
2588 		return -ENOSPC;
2589 	}
2590 
2591 	return 0;
2592 }
2593 
2594 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx, u8 usage)
2595 {
2596 	u32 in_modifier = RES_COUNTER | (((u32)usage & 3) << 30);
2597 	u64 out_param;
2598 	int err;
2599 
2600 	if (mlx4_is_mfunc(dev)) {
2601 		err = mlx4_cmd_imm(dev, 0, &out_param, in_modifier,
2602 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2603 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2604 		if (!err)
2605 			*idx = get_param_l(&out_param);
2606 
2607 		return err;
2608 	}
2609 	return __mlx4_counter_alloc(dev, idx);
2610 }
2611 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2612 
2613 static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2614 				u8 counter_index)
2615 {
2616 	struct mlx4_cmd_mailbox *if_stat_mailbox;
2617 	int err;
2618 	u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2619 
2620 	if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2621 	if (IS_ERR(if_stat_mailbox))
2622 		return PTR_ERR(if_stat_mailbox);
2623 
2624 	err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2625 			   MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2626 			   MLX4_CMD_NATIVE);
2627 
2628 	mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2629 	return err;
2630 }
2631 
2632 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2633 {
2634 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2635 		return;
2636 
2637 	if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2638 		return;
2639 
2640 	__mlx4_clear_if_stat(dev, idx);
2641 
2642 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2643 	return;
2644 }
2645 
2646 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2647 {
2648 	u64 in_param = 0;
2649 
2650 	if (mlx4_is_mfunc(dev)) {
2651 		set_param_l(&in_param, idx);
2652 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2653 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2654 			 MLX4_CMD_WRAPPED);
2655 		return;
2656 	}
2657 	__mlx4_counter_free(dev, idx);
2658 }
2659 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2660 
2661 int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2662 {
2663 	struct mlx4_priv *priv = mlx4_priv(dev);
2664 
2665 	return priv->def_counter[port - 1];
2666 }
2667 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2668 
2669 void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2670 {
2671 	struct mlx4_priv *priv = mlx4_priv(dev);
2672 
2673 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2674 }
2675 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2676 
2677 __be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2678 {
2679 	struct mlx4_priv *priv = mlx4_priv(dev);
2680 
2681 	return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2682 }
2683 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2684 
2685 void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2686 {
2687 	struct mlx4_priv *priv = mlx4_priv(dev);
2688 	__be64 guid;
2689 
2690 	/* hw GUID */
2691 	if (entry == 0)
2692 		return;
2693 
2694 	get_random_bytes((char *)&guid, sizeof(guid));
2695 	guid &= ~(cpu_to_be64(1ULL << 56));
2696 	guid |= cpu_to_be64(1ULL << 57);
2697 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2698 }
2699 
2700 static int mlx4_setup_hca(struct mlx4_dev *dev)
2701 {
2702 	struct mlx4_priv *priv = mlx4_priv(dev);
2703 	int err;
2704 	int port;
2705 	__be32 ib_port_default_caps;
2706 
2707 	err = mlx4_init_uar_table(dev);
2708 	if (err) {
2709 		mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2710 		return err;
2711 	}
2712 
2713 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
2714 	if (err) {
2715 		mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2716 		goto err_uar_table_free;
2717 	}
2718 
2719 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2720 	if (!priv->kar) {
2721 		mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2722 		err = -ENOMEM;
2723 		goto err_uar_free;
2724 	}
2725 
2726 	err = mlx4_init_pd_table(dev);
2727 	if (err) {
2728 		mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2729 		goto err_kar_unmap;
2730 	}
2731 
2732 	err = mlx4_init_xrcd_table(dev);
2733 	if (err) {
2734 		mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2735 		goto err_pd_table_free;
2736 	}
2737 
2738 	err = mlx4_init_mr_table(dev);
2739 	if (err) {
2740 		mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2741 		goto err_xrcd_table_free;
2742 	}
2743 
2744 	if (!mlx4_is_slave(dev)) {
2745 		err = mlx4_init_mcg_table(dev);
2746 		if (err) {
2747 			mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2748 			goto err_mr_table_free;
2749 		}
2750 		err = mlx4_config_mad_demux(dev);
2751 		if (err) {
2752 			mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2753 			goto err_mcg_table_free;
2754 		}
2755 	}
2756 
2757 	err = mlx4_init_eq_table(dev);
2758 	if (err) {
2759 		mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2760 		goto err_mcg_table_free;
2761 	}
2762 
2763 	err = mlx4_cmd_use_events(dev);
2764 	if (err) {
2765 		mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2766 		goto err_eq_table_free;
2767 	}
2768 
2769 	err = mlx4_NOP(dev);
2770 	if (err) {
2771 		if (dev->flags & MLX4_FLAG_MSI_X) {
2772 			mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2773 				  priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2774 			mlx4_warn(dev, "Trying again without MSI-X\n");
2775 		} else {
2776 			mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2777 				 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2778 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2779 		}
2780 
2781 		goto err_cmd_poll;
2782 	}
2783 
2784 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
2785 
2786 	err = mlx4_init_cq_table(dev);
2787 	if (err) {
2788 		mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2789 		goto err_cmd_poll;
2790 	}
2791 
2792 	err = mlx4_init_srq_table(dev);
2793 	if (err) {
2794 		mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2795 		goto err_cq_table_free;
2796 	}
2797 
2798 	err = mlx4_init_qp_table(dev);
2799 	if (err) {
2800 		mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2801 		goto err_srq_table_free;
2802 	}
2803 
2804 	if (!mlx4_is_slave(dev)) {
2805 		err = mlx4_init_counters_table(dev);
2806 		if (err && err != -ENOENT) {
2807 			mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2808 			goto err_qp_table_free;
2809 		}
2810 	}
2811 
2812 	err = mlx4_allocate_default_counters(dev);
2813 	if (err) {
2814 		mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2815 		goto err_counters_table_free;
2816 	}
2817 
2818 	if (!mlx4_is_slave(dev)) {
2819 		for (port = 1; port <= dev->caps.num_ports; port++) {
2820 			ib_port_default_caps = 0;
2821 			err = mlx4_get_port_ib_caps(dev, port,
2822 						    &ib_port_default_caps);
2823 			if (err)
2824 				mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2825 					  port, err);
2826 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2827 
2828 			/* initialize per-slave default ib port capabilities */
2829 			if (mlx4_is_master(dev)) {
2830 				int i;
2831 				for (i = 0; i < dev->num_slaves; i++) {
2832 					if (i == mlx4_master_func_num(dev))
2833 						continue;
2834 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2835 						ib_port_default_caps;
2836 				}
2837 			}
2838 
2839 			if (mlx4_is_mfunc(dev))
2840 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2841 			else
2842 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2843 
2844 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2845 					    dev->caps.pkey_table_len[port] : -1);
2846 			if (err) {
2847 				mlx4_err(dev, "Failed to set port %d, aborting\n",
2848 					 port);
2849 				goto err_default_countes_free;
2850 			}
2851 		}
2852 	}
2853 
2854 	return 0;
2855 
2856 err_default_countes_free:
2857 	mlx4_cleanup_default_counters(dev);
2858 
2859 err_counters_table_free:
2860 	if (!mlx4_is_slave(dev))
2861 		mlx4_cleanup_counters_table(dev);
2862 
2863 err_qp_table_free:
2864 	mlx4_cleanup_qp_table(dev);
2865 
2866 err_srq_table_free:
2867 	mlx4_cleanup_srq_table(dev);
2868 
2869 err_cq_table_free:
2870 	mlx4_cleanup_cq_table(dev);
2871 
2872 err_cmd_poll:
2873 	mlx4_cmd_use_polling(dev);
2874 
2875 err_eq_table_free:
2876 	mlx4_cleanup_eq_table(dev);
2877 
2878 err_mcg_table_free:
2879 	if (!mlx4_is_slave(dev))
2880 		mlx4_cleanup_mcg_table(dev);
2881 
2882 err_mr_table_free:
2883 	mlx4_cleanup_mr_table(dev);
2884 
2885 err_xrcd_table_free:
2886 	mlx4_cleanup_xrcd_table(dev);
2887 
2888 err_pd_table_free:
2889 	mlx4_cleanup_pd_table(dev);
2890 
2891 err_kar_unmap:
2892 	iounmap(priv->kar);
2893 
2894 err_uar_free:
2895 	mlx4_uar_free(dev, &priv->driver_uar);
2896 
2897 err_uar_table_free:
2898 	mlx4_cleanup_uar_table(dev);
2899 	return err;
2900 }
2901 
2902 static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2903 {
2904 	int requested_cpu = 0;
2905 	struct mlx4_priv *priv = mlx4_priv(dev);
2906 	struct mlx4_eq *eq;
2907 	int off = 0;
2908 	int i;
2909 
2910 	if (eqn > dev->caps.num_comp_vectors)
2911 		return -EINVAL;
2912 
2913 	for (i = 1; i < port; i++)
2914 		off += mlx4_get_eqs_per_port(dev, i);
2915 
2916 	requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2917 
2918 	/* Meaning EQs are shared, and this call comes from the second port */
2919 	if (requested_cpu < 0)
2920 		return 0;
2921 
2922 	eq = &priv->eq_table.eq[eqn];
2923 
2924 	if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2925 		return -ENOMEM;
2926 
2927 	cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2928 
2929 	return 0;
2930 }
2931 
2932 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2933 {
2934 	struct mlx4_priv *priv = mlx4_priv(dev);
2935 	struct msix_entry *entries;
2936 	int i;
2937 	int port = 0;
2938 
2939 	if (msi_x) {
2940 		int nreq = min3(dev->caps.num_ports *
2941 				(int)num_online_cpus() + 1,
2942 				dev->caps.num_eqs - dev->caps.reserved_eqs,
2943 				MAX_MSIX);
2944 
2945 		if (msi_x > 1)
2946 			nreq = min_t(int, nreq, msi_x);
2947 
2948 		entries = kcalloc(nreq, sizeof(*entries), GFP_KERNEL);
2949 		if (!entries)
2950 			goto no_msi;
2951 
2952 		for (i = 0; i < nreq; ++i)
2953 			entries[i].entry = i;
2954 
2955 		nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2956 					     nreq);
2957 
2958 		if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
2959 			kfree(entries);
2960 			goto no_msi;
2961 		}
2962 		/* 1 is reserved for events (asyncrounous EQ) */
2963 		dev->caps.num_comp_vectors = nreq - 1;
2964 
2965 		priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2966 		bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2967 			    dev->caps.num_ports);
2968 
2969 		for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2970 			if (i == MLX4_EQ_ASYNC)
2971 				continue;
2972 
2973 			priv->eq_table.eq[i].irq =
2974 				entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2975 
2976 			if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2977 				bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2978 					    dev->caps.num_ports);
2979 				/* We don't set affinity hint when there
2980 				 * aren't enough EQs
2981 				 */
2982 			} else {
2983 				set_bit(port,
2984 					priv->eq_table.eq[i].actv_ports.ports);
2985 				if (mlx4_init_affinity_hint(dev, port + 1, i))
2986 					mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
2987 						  i);
2988 			}
2989 			/* We divide the Eqs evenly between the two ports.
2990 			 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
2991 			 * refers to the number of Eqs per port
2992 			 * (i.e eqs_per_port). Theoretically, we would like to
2993 			 * write something like (i + 1) % eqs_per_port == 0.
2994 			 * However, since there's an asynchronous Eq, we have
2995 			 * to skip over it by comparing this condition to
2996 			 * !!((i + 1) > MLX4_EQ_ASYNC).
2997 			 */
2998 			if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
2999 			    ((i + 1) %
3000 			     (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
3001 			    !!((i + 1) > MLX4_EQ_ASYNC))
3002 				/* If dev->caps.num_comp_vectors < dev->caps.num_ports,
3003 				 * everything is shared anyway.
3004 				 */
3005 				port++;
3006 		}
3007 
3008 		dev->flags |= MLX4_FLAG_MSI_X;
3009 
3010 		kfree(entries);
3011 		return;
3012 	}
3013 
3014 no_msi:
3015 	dev->caps.num_comp_vectors = 1;
3016 
3017 	BUG_ON(MLX4_EQ_ASYNC >= 2);
3018 	for (i = 0; i < 2; ++i) {
3019 		priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
3020 		if (i != MLX4_EQ_ASYNC) {
3021 			bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
3022 				    dev->caps.num_ports);
3023 		}
3024 	}
3025 }
3026 
3027 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
3028 {
3029 	struct devlink *devlink = priv_to_devlink(mlx4_priv(dev));
3030 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
3031 	int err;
3032 
3033 	err = devlink_port_register(devlink, &info->devlink_port, port);
3034 	if (err)
3035 		return err;
3036 
3037 	info->dev = dev;
3038 	info->port = port;
3039 	if (!mlx4_is_slave(dev)) {
3040 		mlx4_init_mac_table(dev, &info->mac_table);
3041 		mlx4_init_vlan_table(dev, &info->vlan_table);
3042 		mlx4_init_roce_gid_table(dev, &info->gid_table);
3043 		info->base_qpn = mlx4_get_base_qpn(dev, port);
3044 	}
3045 
3046 	sprintf(info->dev_name, "mlx4_port%d", port);
3047 	info->port_attr.attr.name = info->dev_name;
3048 	if (mlx4_is_mfunc(dev)) {
3049 		info->port_attr.attr.mode = 0444;
3050 	} else {
3051 		info->port_attr.attr.mode = 0644;
3052 		info->port_attr.store     = set_port_type;
3053 	}
3054 	info->port_attr.show      = show_port_type;
3055 	sysfs_attr_init(&info->port_attr.attr);
3056 
3057 	err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
3058 	if (err) {
3059 		mlx4_err(dev, "Failed to create file for port %d\n", port);
3060 		devlink_port_unregister(&info->devlink_port);
3061 		info->port = -1;
3062 		return err;
3063 	}
3064 
3065 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
3066 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
3067 	if (mlx4_is_mfunc(dev)) {
3068 		info->port_mtu_attr.attr.mode = 0444;
3069 	} else {
3070 		info->port_mtu_attr.attr.mode = 0644;
3071 		info->port_mtu_attr.store     = set_port_ib_mtu;
3072 	}
3073 	info->port_mtu_attr.show      = show_port_ib_mtu;
3074 	sysfs_attr_init(&info->port_mtu_attr.attr);
3075 
3076 	err = device_create_file(&dev->persist->pdev->dev,
3077 				 &info->port_mtu_attr);
3078 	if (err) {
3079 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
3080 		device_remove_file(&info->dev->persist->pdev->dev,
3081 				   &info->port_attr);
3082 		devlink_port_unregister(&info->devlink_port);
3083 		info->port = -1;
3084 		return err;
3085 	}
3086 
3087 	return 0;
3088 }
3089 
3090 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
3091 {
3092 	if (info->port < 0)
3093 		return;
3094 
3095 	device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
3096 	device_remove_file(&info->dev->persist->pdev->dev,
3097 			   &info->port_mtu_attr);
3098 	devlink_port_unregister(&info->devlink_port);
3099 
3100 #ifdef CONFIG_RFS_ACCEL
3101 	free_irq_cpu_rmap(info->rmap);
3102 	info->rmap = NULL;
3103 #endif
3104 }
3105 
3106 static int mlx4_init_steering(struct mlx4_dev *dev)
3107 {
3108 	struct mlx4_priv *priv = mlx4_priv(dev);
3109 	int num_entries = dev->caps.num_ports;
3110 	int i, j;
3111 
3112 	priv->steer = kcalloc(num_entries, sizeof(struct mlx4_steer),
3113 			      GFP_KERNEL);
3114 	if (!priv->steer)
3115 		return -ENOMEM;
3116 
3117 	for (i = 0; i < num_entries; i++)
3118 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3119 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
3120 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
3121 		}
3122 	return 0;
3123 }
3124 
3125 static void mlx4_clear_steering(struct mlx4_dev *dev)
3126 {
3127 	struct mlx4_priv *priv = mlx4_priv(dev);
3128 	struct mlx4_steer_index *entry, *tmp_entry;
3129 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
3130 	int num_entries = dev->caps.num_ports;
3131 	int i, j;
3132 
3133 	for (i = 0; i < num_entries; i++) {
3134 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3135 			list_for_each_entry_safe(pqp, tmp_pqp,
3136 						 &priv->steer[i].promisc_qps[j],
3137 						 list) {
3138 				list_del(&pqp->list);
3139 				kfree(pqp);
3140 			}
3141 			list_for_each_entry_safe(entry, tmp_entry,
3142 						 &priv->steer[i].steer_entries[j],
3143 						 list) {
3144 				list_del(&entry->list);
3145 				list_for_each_entry_safe(pqp, tmp_pqp,
3146 							 &entry->duplicates,
3147 							 list) {
3148 					list_del(&pqp->list);
3149 					kfree(pqp);
3150 				}
3151 				kfree(entry);
3152 			}
3153 		}
3154 	}
3155 	kfree(priv->steer);
3156 }
3157 
3158 static int extended_func_num(struct pci_dev *pdev)
3159 {
3160 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
3161 }
3162 
3163 #define MLX4_OWNER_BASE	0x8069c
3164 #define MLX4_OWNER_SIZE	4
3165 
3166 static int mlx4_get_ownership(struct mlx4_dev *dev)
3167 {
3168 	void __iomem *owner;
3169 	u32 ret;
3170 
3171 	if (pci_channel_offline(dev->persist->pdev))
3172 		return -EIO;
3173 
3174 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3175 			MLX4_OWNER_BASE,
3176 			MLX4_OWNER_SIZE);
3177 	if (!owner) {
3178 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3179 		return -ENOMEM;
3180 	}
3181 
3182 	ret = readl(owner);
3183 	iounmap(owner);
3184 	return (int) !!ret;
3185 }
3186 
3187 static void mlx4_free_ownership(struct mlx4_dev *dev)
3188 {
3189 	void __iomem *owner;
3190 
3191 	if (pci_channel_offline(dev->persist->pdev))
3192 		return;
3193 
3194 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3195 			MLX4_OWNER_BASE,
3196 			MLX4_OWNER_SIZE);
3197 	if (!owner) {
3198 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3199 		return;
3200 	}
3201 	writel(0, owner);
3202 	msleep(1000);
3203 	iounmap(owner);
3204 }
3205 
3206 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV)	==\
3207 				  !!((flags) & MLX4_FLAG_MASTER))
3208 
3209 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
3210 			     u8 total_vfs, int existing_vfs, int reset_flow)
3211 {
3212 	u64 dev_flags = dev->flags;
3213 	int err = 0;
3214 	int fw_enabled_sriov_vfs = min(pci_sriov_get_totalvfs(pdev),
3215 					MLX4_MAX_NUM_VF);
3216 
3217 	if (reset_flow) {
3218 		dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
3219 				       GFP_KERNEL);
3220 		if (!dev->dev_vfs)
3221 			goto free_mem;
3222 		return dev_flags;
3223 	}
3224 
3225 	atomic_inc(&pf_loading);
3226 	if (dev->flags &  MLX4_FLAG_SRIOV) {
3227 		if (existing_vfs != total_vfs) {
3228 			mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
3229 				 existing_vfs, total_vfs);
3230 			total_vfs = existing_vfs;
3231 		}
3232 	}
3233 
3234 	dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs), GFP_KERNEL);
3235 	if (NULL == dev->dev_vfs) {
3236 		mlx4_err(dev, "Failed to allocate memory for VFs\n");
3237 		goto disable_sriov;
3238 	}
3239 
3240 	if (!(dev->flags &  MLX4_FLAG_SRIOV)) {
3241 		if (total_vfs > fw_enabled_sriov_vfs) {
3242 			mlx4_err(dev, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
3243 				 total_vfs, fw_enabled_sriov_vfs);
3244 			err = -ENOMEM;
3245 			goto disable_sriov;
3246 		}
3247 		mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
3248 		err = pci_enable_sriov(pdev, total_vfs);
3249 	}
3250 	if (err) {
3251 		mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
3252 			 err);
3253 		goto disable_sriov;
3254 	} else {
3255 		mlx4_warn(dev, "Running in master mode\n");
3256 		dev_flags |= MLX4_FLAG_SRIOV |
3257 			MLX4_FLAG_MASTER;
3258 		dev_flags &= ~MLX4_FLAG_SLAVE;
3259 		dev->persist->num_vfs = total_vfs;
3260 	}
3261 	return dev_flags;
3262 
3263 disable_sriov:
3264 	atomic_dec(&pf_loading);
3265 free_mem:
3266 	dev->persist->num_vfs = 0;
3267 	kfree(dev->dev_vfs);
3268         dev->dev_vfs = NULL;
3269 	return dev_flags & ~MLX4_FLAG_MASTER;
3270 }
3271 
3272 enum {
3273 	MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
3274 };
3275 
3276 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
3277 			      int *nvfs)
3278 {
3279 	int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
3280 	/* Checking for 64 VFs as a limitation of CX2 */
3281 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
3282 	    requested_vfs >= 64) {
3283 		mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
3284 			 requested_vfs);
3285 		return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
3286 	}
3287 	return 0;
3288 }
3289 
3290 static int mlx4_pci_enable_device(struct mlx4_dev *dev)
3291 {
3292 	struct pci_dev *pdev = dev->persist->pdev;
3293 	int err = 0;
3294 
3295 	mutex_lock(&dev->persist->pci_status_mutex);
3296 	if (dev->persist->pci_status == MLX4_PCI_STATUS_DISABLED) {
3297 		err = pci_enable_device(pdev);
3298 		if (!err)
3299 			dev->persist->pci_status = MLX4_PCI_STATUS_ENABLED;
3300 	}
3301 	mutex_unlock(&dev->persist->pci_status_mutex);
3302 
3303 	return err;
3304 }
3305 
3306 static void mlx4_pci_disable_device(struct mlx4_dev *dev)
3307 {
3308 	struct pci_dev *pdev = dev->persist->pdev;
3309 
3310 	mutex_lock(&dev->persist->pci_status_mutex);
3311 	if (dev->persist->pci_status == MLX4_PCI_STATUS_ENABLED) {
3312 		pci_disable_device(pdev);
3313 		dev->persist->pci_status = MLX4_PCI_STATUS_DISABLED;
3314 	}
3315 	mutex_unlock(&dev->persist->pci_status_mutex);
3316 }
3317 
3318 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
3319 			 int total_vfs, int *nvfs, struct mlx4_priv *priv,
3320 			 int reset_flow)
3321 {
3322 	struct mlx4_dev *dev;
3323 	unsigned sum = 0;
3324 	int err;
3325 	int port;
3326 	int i;
3327 	struct mlx4_dev_cap *dev_cap = NULL;
3328 	int existing_vfs = 0;
3329 
3330 	dev = &priv->dev;
3331 
3332 	INIT_LIST_HEAD(&priv->ctx_list);
3333 	spin_lock_init(&priv->ctx_lock);
3334 
3335 	mutex_init(&priv->port_mutex);
3336 	mutex_init(&priv->bond_mutex);
3337 
3338 	INIT_LIST_HEAD(&priv->pgdir_list);
3339 	mutex_init(&priv->pgdir_mutex);
3340 	spin_lock_init(&priv->cmd.context_lock);
3341 
3342 	INIT_LIST_HEAD(&priv->bf_list);
3343 	mutex_init(&priv->bf_mutex);
3344 
3345 	dev->rev_id = pdev->revision;
3346 	dev->numa_node = dev_to_node(&pdev->dev);
3347 
3348 	/* Detect if this device is a virtual function */
3349 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3350 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3351 		dev->flags |= MLX4_FLAG_SLAVE;
3352 	} else {
3353 		/* We reset the device and enable SRIOV only for physical
3354 		 * devices.  Try to claim ownership on the device;
3355 		 * if already taken, skip -- do not allow multiple PFs */
3356 		err = mlx4_get_ownership(dev);
3357 		if (err) {
3358 			if (err < 0)
3359 				return err;
3360 			else {
3361 				mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
3362 				return -EINVAL;
3363 			}
3364 		}
3365 
3366 		atomic_set(&priv->opreq_count, 0);
3367 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3368 
3369 		/*
3370 		 * Now reset the HCA before we touch the PCI capabilities or
3371 		 * attempt a firmware command, since a boot ROM may have left
3372 		 * the HCA in an undefined state.
3373 		 */
3374 		err = mlx4_reset(dev);
3375 		if (err) {
3376 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
3377 			goto err_sriov;
3378 		}
3379 
3380 		if (total_vfs) {
3381 			dev->flags = MLX4_FLAG_MASTER;
3382 			existing_vfs = pci_num_vf(pdev);
3383 			if (existing_vfs)
3384 				dev->flags |= MLX4_FLAG_SRIOV;
3385 			dev->persist->num_vfs = total_vfs;
3386 		}
3387 	}
3388 
3389 	/* on load remove any previous indication of internal error,
3390 	 * device is up.
3391 	 */
3392 	dev->persist->state = MLX4_DEVICE_STATE_UP;
3393 
3394 slave_start:
3395 	err = mlx4_cmd_init(dev);
3396 	if (err) {
3397 		mlx4_err(dev, "Failed to init command interface, aborting\n");
3398 		goto err_sriov;
3399 	}
3400 
3401 	/* In slave functions, the communication channel must be initialized
3402 	 * before posting commands. Also, init num_slaves before calling
3403 	 * mlx4_init_hca */
3404 	if (mlx4_is_mfunc(dev)) {
3405 		if (mlx4_is_master(dev)) {
3406 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
3407 
3408 		} else {
3409 			dev->num_slaves = 0;
3410 			err = mlx4_multi_func_init(dev);
3411 			if (err) {
3412 				mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
3413 				goto err_cmd;
3414 			}
3415 		}
3416 	}
3417 
3418 	err = mlx4_init_fw(dev);
3419 	if (err) {
3420 		mlx4_err(dev, "Failed to init fw, aborting.\n");
3421 		goto err_mfunc;
3422 	}
3423 
3424 	if (mlx4_is_master(dev)) {
3425 		/* when we hit the goto slave_start below, dev_cap already initialized */
3426 		if (!dev_cap) {
3427 			dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3428 
3429 			if (!dev_cap) {
3430 				err = -ENOMEM;
3431 				goto err_fw;
3432 			}
3433 
3434 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3435 			if (err) {
3436 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3437 				goto err_fw;
3438 			}
3439 
3440 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3441 				goto err_fw;
3442 
3443 			if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3444 				u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3445 								  total_vfs,
3446 								  existing_vfs,
3447 								  reset_flow);
3448 
3449 				mlx4_close_fw(dev);
3450 				mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3451 				dev->flags = dev_flags;
3452 				if (!SRIOV_VALID_STATE(dev->flags)) {
3453 					mlx4_err(dev, "Invalid SRIOV state\n");
3454 					goto err_sriov;
3455 				}
3456 				err = mlx4_reset(dev);
3457 				if (err) {
3458 					mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3459 					goto err_sriov;
3460 				}
3461 				goto slave_start;
3462 			}
3463 		} else {
3464 			/* Legacy mode FW requires SRIOV to be enabled before
3465 			 * doing QUERY_DEV_CAP, since max_eq's value is different if
3466 			 * SRIOV is enabled.
3467 			 */
3468 			memset(dev_cap, 0, sizeof(*dev_cap));
3469 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3470 			if (err) {
3471 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3472 				goto err_fw;
3473 			}
3474 
3475 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3476 				goto err_fw;
3477 		}
3478 	}
3479 
3480 	err = mlx4_init_hca(dev);
3481 	if (err) {
3482 		if (err == -EACCES) {
3483 			/* Not primary Physical function
3484 			 * Running in slave mode */
3485 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3486 			/* We're not a PF */
3487 			if (dev->flags & MLX4_FLAG_SRIOV) {
3488 				if (!existing_vfs)
3489 					pci_disable_sriov(pdev);
3490 				if (mlx4_is_master(dev) && !reset_flow)
3491 					atomic_dec(&pf_loading);
3492 				dev->flags &= ~MLX4_FLAG_SRIOV;
3493 			}
3494 			if (!mlx4_is_slave(dev))
3495 				mlx4_free_ownership(dev);
3496 			dev->flags |= MLX4_FLAG_SLAVE;
3497 			dev->flags &= ~MLX4_FLAG_MASTER;
3498 			goto slave_start;
3499 		} else
3500 			goto err_fw;
3501 	}
3502 
3503 	if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3504 		u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3505 						  existing_vfs, reset_flow);
3506 
3507 		if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3508 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3509 			dev->flags = dev_flags;
3510 			err = mlx4_cmd_init(dev);
3511 			if (err) {
3512 				/* Only VHCR is cleaned up, so could still
3513 				 * send FW commands
3514 				 */
3515 				mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3516 				goto err_close;
3517 			}
3518 		} else {
3519 			dev->flags = dev_flags;
3520 		}
3521 
3522 		if (!SRIOV_VALID_STATE(dev->flags)) {
3523 			mlx4_err(dev, "Invalid SRIOV state\n");
3524 			goto err_close;
3525 		}
3526 	}
3527 
3528 	/* check if the device is functioning at its maximum possible speed.
3529 	 * No return code for this call, just warn the user in case of PCI
3530 	 * express device capabilities are under-satisfied by the bus.
3531 	 */
3532 	if (!mlx4_is_slave(dev))
3533 		pcie_print_link_status(dev->persist->pdev);
3534 
3535 	/* In master functions, the communication channel must be initialized
3536 	 * after obtaining its address from fw */
3537 	if (mlx4_is_master(dev)) {
3538 		if (dev->caps.num_ports < 2 &&
3539 		    num_vfs_argc > 1) {
3540 			err = -EINVAL;
3541 			mlx4_err(dev,
3542 				 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3543 				 dev->caps.num_ports);
3544 			goto err_close;
3545 		}
3546 		memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
3547 
3548 		for (i = 0;
3549 		     i < sizeof(dev->persist->nvfs)/
3550 		     sizeof(dev->persist->nvfs[0]); i++) {
3551 			unsigned j;
3552 
3553 			for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
3554 				dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3555 				dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3556 					dev->caps.num_ports;
3557 			}
3558 		}
3559 
3560 		/* In master functions, the communication channel
3561 		 * must be initialized after obtaining its address from fw
3562 		 */
3563 		err = mlx4_multi_func_init(dev);
3564 		if (err) {
3565 			mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3566 			goto err_close;
3567 		}
3568 	}
3569 
3570 	err = mlx4_alloc_eq_table(dev);
3571 	if (err)
3572 		goto err_master_mfunc;
3573 
3574 	bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
3575 	mutex_init(&priv->msix_ctl.pool_lock);
3576 
3577 	mlx4_enable_msi_x(dev);
3578 	if ((mlx4_is_mfunc(dev)) &&
3579 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
3580 		err = -EOPNOTSUPP;
3581 		mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
3582 		goto err_free_eq;
3583 	}
3584 
3585 	if (!mlx4_is_slave(dev)) {
3586 		err = mlx4_init_steering(dev);
3587 		if (err)
3588 			goto err_disable_msix;
3589 	}
3590 
3591 	mlx4_init_quotas(dev);
3592 
3593 	err = mlx4_setup_hca(dev);
3594 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3595 	    !mlx4_is_mfunc(dev)) {
3596 		dev->flags &= ~MLX4_FLAG_MSI_X;
3597 		dev->caps.num_comp_vectors = 1;
3598 		pci_disable_msix(pdev);
3599 		err = mlx4_setup_hca(dev);
3600 	}
3601 
3602 	if (err)
3603 		goto err_steer;
3604 
3605 	/* When PF resources are ready arm its comm channel to enable
3606 	 * getting commands
3607 	 */
3608 	if (mlx4_is_master(dev)) {
3609 		err = mlx4_ARM_COMM_CHANNEL(dev);
3610 		if (err) {
3611 			mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3612 				 err);
3613 			goto err_steer;
3614 		}
3615 	}
3616 
3617 	for (port = 1; port <= dev->caps.num_ports; port++) {
3618 		err = mlx4_init_port_info(dev, port);
3619 		if (err)
3620 			goto err_port;
3621 	}
3622 
3623 	priv->v2p.port1 = 1;
3624 	priv->v2p.port2 = 2;
3625 
3626 	err = mlx4_register_device(dev);
3627 	if (err)
3628 		goto err_port;
3629 
3630 	mlx4_request_modules(dev);
3631 
3632 	mlx4_sense_init(dev);
3633 	mlx4_start_sense(dev);
3634 
3635 	priv->removed = 0;
3636 
3637 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3638 		atomic_dec(&pf_loading);
3639 
3640 	kfree(dev_cap);
3641 	return 0;
3642 
3643 err_port:
3644 	for (--port; port >= 1; --port)
3645 		mlx4_cleanup_port_info(&priv->port[port]);
3646 
3647 	mlx4_cleanup_default_counters(dev);
3648 	if (!mlx4_is_slave(dev))
3649 		mlx4_cleanup_counters_table(dev);
3650 	mlx4_cleanup_qp_table(dev);
3651 	mlx4_cleanup_srq_table(dev);
3652 	mlx4_cleanup_cq_table(dev);
3653 	mlx4_cmd_use_polling(dev);
3654 	mlx4_cleanup_eq_table(dev);
3655 	mlx4_cleanup_mcg_table(dev);
3656 	mlx4_cleanup_mr_table(dev);
3657 	mlx4_cleanup_xrcd_table(dev);
3658 	mlx4_cleanup_pd_table(dev);
3659 	mlx4_cleanup_uar_table(dev);
3660 
3661 err_steer:
3662 	if (!mlx4_is_slave(dev))
3663 		mlx4_clear_steering(dev);
3664 
3665 err_disable_msix:
3666 	if (dev->flags & MLX4_FLAG_MSI_X)
3667 		pci_disable_msix(pdev);
3668 
3669 err_free_eq:
3670 	mlx4_free_eq_table(dev);
3671 
3672 err_master_mfunc:
3673 	if (mlx4_is_master(dev)) {
3674 		mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
3675 		mlx4_multi_func_cleanup(dev);
3676 	}
3677 
3678 	if (mlx4_is_slave(dev))
3679 		mlx4_slave_destroy_special_qp_cap(dev);
3680 
3681 err_close:
3682 	mlx4_close_hca(dev);
3683 
3684 err_fw:
3685 	mlx4_close_fw(dev);
3686 
3687 err_mfunc:
3688 	if (mlx4_is_slave(dev))
3689 		mlx4_multi_func_cleanup(dev);
3690 
3691 err_cmd:
3692 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3693 
3694 err_sriov:
3695 	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3696 		pci_disable_sriov(pdev);
3697 		dev->flags &= ~MLX4_FLAG_SRIOV;
3698 	}
3699 
3700 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3701 		atomic_dec(&pf_loading);
3702 
3703 	kfree(priv->dev.dev_vfs);
3704 
3705 	if (!mlx4_is_slave(dev))
3706 		mlx4_free_ownership(dev);
3707 
3708 	kfree(dev_cap);
3709 	return err;
3710 }
3711 
3712 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3713 			   struct mlx4_priv *priv)
3714 {
3715 	int err;
3716 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3717 	int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3718 	const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3719 		{2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3720 	unsigned total_vfs = 0;
3721 	unsigned int i;
3722 
3723 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3724 
3725 	err = mlx4_pci_enable_device(&priv->dev);
3726 	if (err) {
3727 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3728 		return err;
3729 	}
3730 
3731 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3732 	 * per port, we must limit the number of VFs to 63 (since their are
3733 	 * 128 MACs)
3734 	 */
3735 	for (i = 0; i < ARRAY_SIZE(nvfs) && i < num_vfs_argc;
3736 	     total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3737 		nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3738 		if (nvfs[i] < 0) {
3739 			dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3740 			err = -EINVAL;
3741 			goto err_disable_pdev;
3742 		}
3743 	}
3744 	for (i = 0; i < ARRAY_SIZE(prb_vf) && i < probe_vfs_argc;
3745 	     i++) {
3746 		prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3747 		if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3748 			dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3749 			err = -EINVAL;
3750 			goto err_disable_pdev;
3751 		}
3752 	}
3753 	if (total_vfs > MLX4_MAX_NUM_VF) {
3754 		dev_err(&pdev->dev,
3755 			"Requested more VF's (%d) than allowed by hw (%d)\n",
3756 			total_vfs, MLX4_MAX_NUM_VF);
3757 		err = -EINVAL;
3758 		goto err_disable_pdev;
3759 	}
3760 
3761 	for (i = 0; i < MLX4_MAX_PORTS; i++) {
3762 		if (nvfs[i] + nvfs[2] > MLX4_MAX_NUM_VF_P_PORT) {
3763 			dev_err(&pdev->dev,
3764 				"Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3765 				nvfs[i] + nvfs[2], i + 1,
3766 				MLX4_MAX_NUM_VF_P_PORT);
3767 			err = -EINVAL;
3768 			goto err_disable_pdev;
3769 		}
3770 	}
3771 
3772 	/* Check for BARs. */
3773 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3774 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3775 		dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3776 			pci_dev_data, pci_resource_flags(pdev, 0));
3777 		err = -ENODEV;
3778 		goto err_disable_pdev;
3779 	}
3780 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3781 		dev_err(&pdev->dev, "Missing UAR, aborting\n");
3782 		err = -ENODEV;
3783 		goto err_disable_pdev;
3784 	}
3785 
3786 	err = pci_request_regions(pdev, DRV_NAME);
3787 	if (err) {
3788 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3789 		goto err_disable_pdev;
3790 	}
3791 
3792 	pci_set_master(pdev);
3793 
3794 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3795 	if (err) {
3796 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3797 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3798 		if (err) {
3799 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3800 			goto err_release_regions;
3801 		}
3802 	}
3803 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3804 	if (err) {
3805 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3806 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3807 		if (err) {
3808 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3809 			goto err_release_regions;
3810 		}
3811 	}
3812 
3813 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
3814 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3815 	/* Detect if this device is a virtual function */
3816 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3817 		/* When acting as pf, we normally skip vfs unless explicitly
3818 		 * requested to probe them.
3819 		 */
3820 		if (total_vfs) {
3821 			unsigned vfs_offset = 0;
3822 
3823 			for (i = 0; i < ARRAY_SIZE(nvfs) &&
3824 			     vfs_offset + nvfs[i] < extended_func_num(pdev);
3825 			     vfs_offset += nvfs[i], i++)
3826 				;
3827 			if (i == ARRAY_SIZE(nvfs)) {
3828 				err = -ENODEV;
3829 				goto err_release_regions;
3830 			}
3831 			if ((extended_func_num(pdev) - vfs_offset)
3832 			    > prb_vf[i]) {
3833 				dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3834 					 extended_func_num(pdev));
3835 				err = -ENODEV;
3836 				goto err_release_regions;
3837 			}
3838 		}
3839 	}
3840 
3841 	err = mlx4_crdump_init(&priv->dev);
3842 	if (err)
3843 		goto err_release_regions;
3844 
3845 	err = mlx4_catas_init(&priv->dev);
3846 	if (err)
3847 		goto err_crdump;
3848 
3849 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3850 	if (err)
3851 		goto err_catas;
3852 
3853 	return 0;
3854 
3855 err_catas:
3856 	mlx4_catas_end(&priv->dev);
3857 
3858 err_crdump:
3859 	mlx4_crdump_end(&priv->dev);
3860 
3861 err_release_regions:
3862 	pci_release_regions(pdev);
3863 
3864 err_disable_pdev:
3865 	mlx4_pci_disable_device(&priv->dev);
3866 	return err;
3867 }
3868 
3869 static int mlx4_devlink_port_type_set(struct devlink_port *devlink_port,
3870 				      enum devlink_port_type port_type)
3871 {
3872 	struct mlx4_port_info *info = container_of(devlink_port,
3873 						   struct mlx4_port_info,
3874 						   devlink_port);
3875 	enum mlx4_port_type mlx4_port_type;
3876 
3877 	switch (port_type) {
3878 	case DEVLINK_PORT_TYPE_AUTO:
3879 		mlx4_port_type = MLX4_PORT_TYPE_AUTO;
3880 		break;
3881 	case DEVLINK_PORT_TYPE_ETH:
3882 		mlx4_port_type = MLX4_PORT_TYPE_ETH;
3883 		break;
3884 	case DEVLINK_PORT_TYPE_IB:
3885 		mlx4_port_type = MLX4_PORT_TYPE_IB;
3886 		break;
3887 	default:
3888 		return -EOPNOTSUPP;
3889 	}
3890 
3891 	return __set_port_type(info, mlx4_port_type);
3892 }
3893 
3894 static void mlx4_devlink_param_load_driverinit_values(struct devlink *devlink)
3895 {
3896 	struct mlx4_priv *priv = devlink_priv(devlink);
3897 	struct mlx4_dev *dev = &priv->dev;
3898 	struct mlx4_fw_crdump *crdump = &dev->persist->crdump;
3899 	union devlink_param_value saved_value;
3900 	int err;
3901 
3902 	err = devlink_param_driverinit_value_get(devlink,
3903 						 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET,
3904 						 &saved_value);
3905 	if (!err && mlx4_internal_err_reset != saved_value.vbool) {
3906 		mlx4_internal_err_reset = saved_value.vbool;
3907 		/* Notify on value changed on runtime configuration mode */
3908 		devlink_param_value_changed(devlink,
3909 					    DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET);
3910 	}
3911 	err = devlink_param_driverinit_value_get(devlink,
3912 						 DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
3913 						 &saved_value);
3914 	if (!err)
3915 		log_num_mac = order_base_2(saved_value.vu32);
3916 	err = devlink_param_driverinit_value_get(devlink,
3917 						 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
3918 						 &saved_value);
3919 	if (!err)
3920 		enable_64b_cqe_eqe = saved_value.vbool;
3921 	err = devlink_param_driverinit_value_get(devlink,
3922 						 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
3923 						 &saved_value);
3924 	if (!err)
3925 		enable_4k_uar = saved_value.vbool;
3926 	err = devlink_param_driverinit_value_get(devlink,
3927 						 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT,
3928 						 &saved_value);
3929 	if (!err && crdump->snapshot_enable != saved_value.vbool) {
3930 		crdump->snapshot_enable = saved_value.vbool;
3931 		devlink_param_value_changed(devlink,
3932 					    DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT);
3933 	}
3934 }
3935 
3936 static int mlx4_devlink_reload(struct devlink *devlink,
3937 			       struct netlink_ext_ack *extack)
3938 {
3939 	struct mlx4_priv *priv = devlink_priv(devlink);
3940 	struct mlx4_dev *dev = &priv->dev;
3941 	struct mlx4_dev_persistent *persist = dev->persist;
3942 	int err;
3943 
3944 	if (persist->num_vfs)
3945 		mlx4_warn(persist->dev, "Reload performed on PF, will cause reset on operating Virtual Functions\n");
3946 	err = mlx4_restart_one(persist->pdev, true, devlink);
3947 	if (err)
3948 		mlx4_err(persist->dev, "mlx4_restart_one failed, ret=%d\n", err);
3949 
3950 	return err;
3951 }
3952 
3953 static const struct devlink_ops mlx4_devlink_ops = {
3954 	.port_type_set	= mlx4_devlink_port_type_set,
3955 	.reload		= mlx4_devlink_reload,
3956 };
3957 
3958 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3959 {
3960 	struct devlink *devlink;
3961 	struct mlx4_priv *priv;
3962 	struct mlx4_dev *dev;
3963 	int ret;
3964 
3965 	printk_once(KERN_INFO "%s", mlx4_version);
3966 
3967 	devlink = devlink_alloc(&mlx4_devlink_ops, sizeof(*priv));
3968 	if (!devlink)
3969 		return -ENOMEM;
3970 	priv = devlink_priv(devlink);
3971 
3972 	dev       = &priv->dev;
3973 	dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3974 	if (!dev->persist) {
3975 		ret = -ENOMEM;
3976 		goto err_devlink_free;
3977 	}
3978 	dev->persist->pdev = pdev;
3979 	dev->persist->dev = dev;
3980 	pci_set_drvdata(pdev, dev->persist);
3981 	priv->pci_dev_data = id->driver_data;
3982 	mutex_init(&dev->persist->device_state_mutex);
3983 	mutex_init(&dev->persist->interface_state_mutex);
3984 	mutex_init(&dev->persist->pci_status_mutex);
3985 
3986 	ret = devlink_register(devlink, &pdev->dev);
3987 	if (ret)
3988 		goto err_persist_free;
3989 	ret = devlink_params_register(devlink, mlx4_devlink_params,
3990 				      ARRAY_SIZE(mlx4_devlink_params));
3991 	if (ret)
3992 		goto err_devlink_unregister;
3993 	mlx4_devlink_set_params_init_values(devlink);
3994 	ret =  __mlx4_init_one(pdev, id->driver_data, priv);
3995 	if (ret)
3996 		goto err_params_unregister;
3997 
3998 	pci_save_state(pdev);
3999 	return 0;
4000 
4001 err_params_unregister:
4002 	devlink_params_unregister(devlink, mlx4_devlink_params,
4003 				  ARRAY_SIZE(mlx4_devlink_params));
4004 err_devlink_unregister:
4005 	devlink_unregister(devlink);
4006 err_persist_free:
4007 	kfree(dev->persist);
4008 err_devlink_free:
4009 	devlink_free(devlink);
4010 	return ret;
4011 }
4012 
4013 static void mlx4_clean_dev(struct mlx4_dev *dev)
4014 {
4015 	struct mlx4_dev_persistent *persist = dev->persist;
4016 	struct mlx4_priv *priv = mlx4_priv(dev);
4017 	unsigned long	flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
4018 
4019 	memset(priv, 0, sizeof(*priv));
4020 	priv->dev.persist = persist;
4021 	priv->dev.flags = flags;
4022 }
4023 
4024 static void mlx4_unload_one(struct pci_dev *pdev)
4025 {
4026 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4027 	struct mlx4_dev  *dev  = persist->dev;
4028 	struct mlx4_priv *priv = mlx4_priv(dev);
4029 	int               pci_dev_data;
4030 	int p, i;
4031 
4032 	if (priv->removed)
4033 		return;
4034 
4035 	/* saving current ports type for further use */
4036 	for (i = 0; i < dev->caps.num_ports; i++) {
4037 		dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
4038 		dev->persist->curr_port_poss_type[i] = dev->caps.
4039 						       possible_type[i + 1];
4040 	}
4041 
4042 	pci_dev_data = priv->pci_dev_data;
4043 
4044 	mlx4_stop_sense(dev);
4045 	mlx4_unregister_device(dev);
4046 
4047 	for (p = 1; p <= dev->caps.num_ports; p++) {
4048 		mlx4_cleanup_port_info(&priv->port[p]);
4049 		mlx4_CLOSE_PORT(dev, p);
4050 	}
4051 
4052 	if (mlx4_is_master(dev))
4053 		mlx4_free_resource_tracker(dev,
4054 					   RES_TR_FREE_SLAVES_ONLY);
4055 
4056 	mlx4_cleanup_default_counters(dev);
4057 	if (!mlx4_is_slave(dev))
4058 		mlx4_cleanup_counters_table(dev);
4059 	mlx4_cleanup_qp_table(dev);
4060 	mlx4_cleanup_srq_table(dev);
4061 	mlx4_cleanup_cq_table(dev);
4062 	mlx4_cmd_use_polling(dev);
4063 	mlx4_cleanup_eq_table(dev);
4064 	mlx4_cleanup_mcg_table(dev);
4065 	mlx4_cleanup_mr_table(dev);
4066 	mlx4_cleanup_xrcd_table(dev);
4067 	mlx4_cleanup_pd_table(dev);
4068 
4069 	if (mlx4_is_master(dev))
4070 		mlx4_free_resource_tracker(dev,
4071 					   RES_TR_FREE_STRUCTS_ONLY);
4072 
4073 	iounmap(priv->kar);
4074 	mlx4_uar_free(dev, &priv->driver_uar);
4075 	mlx4_cleanup_uar_table(dev);
4076 	if (!mlx4_is_slave(dev))
4077 		mlx4_clear_steering(dev);
4078 	mlx4_free_eq_table(dev);
4079 	if (mlx4_is_master(dev))
4080 		mlx4_multi_func_cleanup(dev);
4081 	mlx4_close_hca(dev);
4082 	mlx4_close_fw(dev);
4083 	if (mlx4_is_slave(dev))
4084 		mlx4_multi_func_cleanup(dev);
4085 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
4086 
4087 	if (dev->flags & MLX4_FLAG_MSI_X)
4088 		pci_disable_msix(pdev);
4089 
4090 	if (!mlx4_is_slave(dev))
4091 		mlx4_free_ownership(dev);
4092 
4093 	mlx4_slave_destroy_special_qp_cap(dev);
4094 	kfree(dev->dev_vfs);
4095 
4096 	mlx4_clean_dev(dev);
4097 	priv->pci_dev_data = pci_dev_data;
4098 	priv->removed = 1;
4099 }
4100 
4101 static void mlx4_remove_one(struct pci_dev *pdev)
4102 {
4103 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4104 	struct mlx4_dev  *dev  = persist->dev;
4105 	struct mlx4_priv *priv = mlx4_priv(dev);
4106 	struct devlink *devlink = priv_to_devlink(priv);
4107 	int active_vfs = 0;
4108 
4109 	if (mlx4_is_slave(dev))
4110 		persist->interface_state |= MLX4_INTERFACE_STATE_NOWAIT;
4111 
4112 	mutex_lock(&persist->interface_state_mutex);
4113 	persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
4114 	mutex_unlock(&persist->interface_state_mutex);
4115 
4116 	/* Disabling SR-IOV is not allowed while there are active vf's */
4117 	if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
4118 		active_vfs = mlx4_how_many_lives_vf(dev);
4119 		if (active_vfs) {
4120 			pr_warn("Removing PF when there are active VF's !!\n");
4121 			pr_warn("Will not disable SR-IOV.\n");
4122 		}
4123 	}
4124 
4125 	/* device marked to be under deletion running now without the lock
4126 	 * letting other tasks to be terminated
4127 	 */
4128 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4129 		mlx4_unload_one(pdev);
4130 	else
4131 		mlx4_info(dev, "%s: interface is down\n", __func__);
4132 	mlx4_catas_end(dev);
4133 	mlx4_crdump_end(dev);
4134 	if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
4135 		mlx4_warn(dev, "Disabling SR-IOV\n");
4136 		pci_disable_sriov(pdev);
4137 	}
4138 
4139 	pci_release_regions(pdev);
4140 	mlx4_pci_disable_device(dev);
4141 	devlink_params_unregister(devlink, mlx4_devlink_params,
4142 				  ARRAY_SIZE(mlx4_devlink_params));
4143 	devlink_unregister(devlink);
4144 	kfree(dev->persist);
4145 	devlink_free(devlink);
4146 }
4147 
4148 static int restore_current_port_types(struct mlx4_dev *dev,
4149 				      enum mlx4_port_type *types,
4150 				      enum mlx4_port_type *poss_types)
4151 {
4152 	struct mlx4_priv *priv = mlx4_priv(dev);
4153 	int err, i;
4154 
4155 	mlx4_stop_sense(dev);
4156 
4157 	mutex_lock(&priv->port_mutex);
4158 	for (i = 0; i < dev->caps.num_ports; i++)
4159 		dev->caps.possible_type[i + 1] = poss_types[i];
4160 	err = mlx4_change_port_types(dev, types);
4161 	mlx4_start_sense(dev);
4162 	mutex_unlock(&priv->port_mutex);
4163 
4164 	return err;
4165 }
4166 
4167 int mlx4_restart_one(struct pci_dev *pdev, bool reload, struct devlink *devlink)
4168 {
4169 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4170 	struct mlx4_dev	 *dev  = persist->dev;
4171 	struct mlx4_priv *priv = mlx4_priv(dev);
4172 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4173 	int pci_dev_data, err, total_vfs;
4174 
4175 	pci_dev_data = priv->pci_dev_data;
4176 	total_vfs = dev->persist->num_vfs;
4177 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4178 
4179 	mlx4_unload_one(pdev);
4180 	if (reload)
4181 		mlx4_devlink_param_load_driverinit_values(devlink);
4182 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
4183 	if (err) {
4184 		mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
4185 			 __func__, pci_name(pdev), err);
4186 		return err;
4187 	}
4188 
4189 	err = restore_current_port_types(dev, dev->persist->curr_port_type,
4190 					 dev->persist->curr_port_poss_type);
4191 	if (err)
4192 		mlx4_err(dev, "could not restore original port types (%d)\n",
4193 			 err);
4194 
4195 	return err;
4196 }
4197 
4198 #define MLX_SP(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_FORCE_SENSE_PORT }
4199 #define MLX_VF(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_IS_VF }
4200 #define MLX_GN(id) { PCI_VDEVICE(MELLANOX, id), 0 }
4201 
4202 static const struct pci_device_id mlx4_pci_table[] = {
4203 #ifdef CONFIG_MLX4_CORE_GEN2
4204 	/* MT25408 "Hermon" */
4205 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_SDR),	/* SDR */
4206 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR),	/* DDR */
4207 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR),	/* QDR */
4208 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR_GEN2), /* DDR Gen2 */
4209 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR_GEN2),	/* QDR Gen2 */
4210 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN),	/* EN 10GigE */
4211 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN_GEN2),  /* EN 10GigE Gen2 */
4212 	/* MT25458 ConnectX EN 10GBASE-T */
4213 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN),
4214 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_T_GEN2),	/* Gen2 */
4215 	/* MT26468 ConnectX EN 10GigE PCIe Gen2*/
4216 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_GEN2),
4217 	/* MT26438 ConnectX EN 40GigE PCIe Gen2 5GT/s */
4218 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_5_GEN2),
4219 	/* MT26478 ConnectX2 40GigE PCIe Gen2 */
4220 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX2),
4221 	/* MT25400 Family [ConnectX-2] */
4222 	MLX_VF(0x1002),					/* Virtual Function */
4223 #endif /* CONFIG_MLX4_CORE_GEN2 */
4224 	/* MT27500 Family [ConnectX-3] */
4225 	MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3),
4226 	MLX_VF(0x1004),					/* Virtual Function */
4227 	MLX_GN(0x1005),					/* MT27510 Family */
4228 	MLX_GN(0x1006),					/* MT27511 Family */
4229 	MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3_PRO),	/* MT27520 Family */
4230 	MLX_GN(0x1008),					/* MT27521 Family */
4231 	MLX_GN(0x1009),					/* MT27530 Family */
4232 	MLX_GN(0x100a),					/* MT27531 Family */
4233 	MLX_GN(0x100b),					/* MT27540 Family */
4234 	MLX_GN(0x100c),					/* MT27541 Family */
4235 	MLX_GN(0x100d),					/* MT27550 Family */
4236 	MLX_GN(0x100e),					/* MT27551 Family */
4237 	MLX_GN(0x100f),					/* MT27560 Family */
4238 	MLX_GN(0x1010),					/* MT27561 Family */
4239 
4240 	/*
4241 	 * See the mellanox_check_broken_intx_masking() quirk when
4242 	 * adding devices
4243 	 */
4244 
4245 	{ 0, }
4246 };
4247 
4248 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
4249 
4250 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
4251 					      pci_channel_state_t state)
4252 {
4253 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4254 
4255 	mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
4256 	mlx4_enter_error_state(persist);
4257 
4258 	mutex_lock(&persist->interface_state_mutex);
4259 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4260 		mlx4_unload_one(pdev);
4261 
4262 	mutex_unlock(&persist->interface_state_mutex);
4263 	if (state == pci_channel_io_perm_failure)
4264 		return PCI_ERS_RESULT_DISCONNECT;
4265 
4266 	mlx4_pci_disable_device(persist->dev);
4267 	return PCI_ERS_RESULT_NEED_RESET;
4268 }
4269 
4270 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
4271 {
4272 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4273 	struct mlx4_dev	 *dev  = persist->dev;
4274 	int err;
4275 
4276 	mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
4277 	err = mlx4_pci_enable_device(dev);
4278 	if (err) {
4279 		mlx4_err(dev, "Can not re-enable device, err=%d\n", err);
4280 		return PCI_ERS_RESULT_DISCONNECT;
4281 	}
4282 
4283 	pci_set_master(pdev);
4284 	pci_restore_state(pdev);
4285 	pci_save_state(pdev);
4286 	return PCI_ERS_RESULT_RECOVERED;
4287 }
4288 
4289 static void mlx4_pci_resume(struct pci_dev *pdev)
4290 {
4291 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4292 	struct mlx4_dev	 *dev  = persist->dev;
4293 	struct mlx4_priv *priv = mlx4_priv(dev);
4294 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4295 	int total_vfs;
4296 	int err;
4297 
4298 	mlx4_err(dev, "%s was called\n", __func__);
4299 	total_vfs = dev->persist->num_vfs;
4300 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4301 
4302 	mutex_lock(&persist->interface_state_mutex);
4303 	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4304 		err = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
4305 				    priv, 1);
4306 		if (err) {
4307 			mlx4_err(dev, "%s: mlx4_load_one failed, err=%d\n",
4308 				 __func__,  err);
4309 			goto end;
4310 		}
4311 
4312 		err = restore_current_port_types(dev, dev->persist->
4313 						 curr_port_type, dev->persist->
4314 						 curr_port_poss_type);
4315 		if (err)
4316 			mlx4_err(dev, "could not restore original port types (%d)\n", err);
4317 	}
4318 end:
4319 	mutex_unlock(&persist->interface_state_mutex);
4320 
4321 }
4322 
4323 static void mlx4_shutdown(struct pci_dev *pdev)
4324 {
4325 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4326 
4327 	mlx4_info(persist->dev, "mlx4_shutdown was called\n");
4328 	mutex_lock(&persist->interface_state_mutex);
4329 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4330 		mlx4_unload_one(pdev);
4331 	mutex_unlock(&persist->interface_state_mutex);
4332 }
4333 
4334 static const struct pci_error_handlers mlx4_err_handler = {
4335 	.error_detected = mlx4_pci_err_detected,
4336 	.slot_reset     = mlx4_pci_slot_reset,
4337 	.resume		= mlx4_pci_resume,
4338 };
4339 
4340 static int mlx4_suspend(struct pci_dev *pdev, pm_message_t state)
4341 {
4342 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4343 	struct mlx4_dev	*dev = persist->dev;
4344 
4345 	mlx4_err(dev, "suspend was called\n");
4346 	mutex_lock(&persist->interface_state_mutex);
4347 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4348 		mlx4_unload_one(pdev);
4349 	mutex_unlock(&persist->interface_state_mutex);
4350 
4351 	return 0;
4352 }
4353 
4354 static int mlx4_resume(struct pci_dev *pdev)
4355 {
4356 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4357 	struct mlx4_dev	*dev = persist->dev;
4358 	struct mlx4_priv *priv = mlx4_priv(dev);
4359 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4360 	int total_vfs;
4361 	int ret = 0;
4362 
4363 	mlx4_err(dev, "resume was called\n");
4364 	total_vfs = dev->persist->num_vfs;
4365 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4366 
4367 	mutex_lock(&persist->interface_state_mutex);
4368 	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4369 		ret = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs,
4370 				    nvfs, priv, 1);
4371 		if (!ret) {
4372 			ret = restore_current_port_types(dev,
4373 					dev->persist->curr_port_type,
4374 					dev->persist->curr_port_poss_type);
4375 			if (ret)
4376 				mlx4_err(dev, "resume: could not restore original port types (%d)\n", ret);
4377 		}
4378 	}
4379 	mutex_unlock(&persist->interface_state_mutex);
4380 
4381 	return ret;
4382 }
4383 
4384 static struct pci_driver mlx4_driver = {
4385 	.name		= DRV_NAME,
4386 	.id_table	= mlx4_pci_table,
4387 	.probe		= mlx4_init_one,
4388 	.shutdown	= mlx4_shutdown,
4389 	.remove		= mlx4_remove_one,
4390 	.suspend	= mlx4_suspend,
4391 	.resume		= mlx4_resume,
4392 	.err_handler    = &mlx4_err_handler,
4393 };
4394 
4395 static int __init mlx4_verify_params(void)
4396 {
4397 	if (msi_x < 0) {
4398 		pr_warn("mlx4_core: bad msi_x: %d\n", msi_x);
4399 		return -1;
4400 	}
4401 
4402 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
4403 		pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
4404 		return -1;
4405 	}
4406 
4407 	if (log_num_vlan != 0)
4408 		pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
4409 			MLX4_LOG_NUM_VLANS);
4410 
4411 	if (use_prio != 0)
4412 		pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
4413 
4414 	if ((log_mtts_per_seg < 0) || (log_mtts_per_seg > 7)) {
4415 		pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
4416 			log_mtts_per_seg);
4417 		return -1;
4418 	}
4419 
4420 	/* Check if module param for ports type has legal combination */
4421 	if (port_type_array[0] == false && port_type_array[1] == true) {
4422 		pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
4423 		port_type_array[0] = true;
4424 	}
4425 
4426 	if (mlx4_log_num_mgm_entry_size < -7 ||
4427 	    (mlx4_log_num_mgm_entry_size > 0 &&
4428 	     (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
4429 	      mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
4430 		pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
4431 			mlx4_log_num_mgm_entry_size,
4432 			MLX4_MIN_MGM_LOG_ENTRY_SIZE,
4433 			MLX4_MAX_MGM_LOG_ENTRY_SIZE);
4434 		return -1;
4435 	}
4436 
4437 	return 0;
4438 }
4439 
4440 static int __init mlx4_init(void)
4441 {
4442 	int ret;
4443 
4444 	if (mlx4_verify_params())
4445 		return -EINVAL;
4446 
4447 
4448 	mlx4_wq = create_singlethread_workqueue("mlx4");
4449 	if (!mlx4_wq)
4450 		return -ENOMEM;
4451 
4452 	ret = pci_register_driver(&mlx4_driver);
4453 	if (ret < 0)
4454 		destroy_workqueue(mlx4_wq);
4455 	return ret < 0 ? ret : 0;
4456 }
4457 
4458 static void __exit mlx4_cleanup(void)
4459 {
4460 	pci_unregister_driver(&mlx4_driver);
4461 	destroy_workqueue(mlx4_wq);
4462 }
4463 
4464 module_init(mlx4_init);
4465 module_exit(mlx4_cleanup);
4466