xref: /linux/drivers/net/ethernet/mellanox/mlx4/main.c (revision 43dfc13ca972988e620a6edb72956981b75ab6b0)
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/etherdevice.h>
46 #include <net/devlink.h>
47 
48 #include <uapi/rdma/mlx4-abi.h>
49 #include <linux/mlx4/device.h>
50 #include <linux/mlx4/doorbell.h>
51 
52 #include <rdma/ib_verbs.h>
53 
54 #include "mlx4.h"
55 #include "fw.h"
56 #include "icm.h"
57 
58 MODULE_AUTHOR("Roland Dreier");
59 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
60 MODULE_LICENSE("Dual BSD/GPL");
61 MODULE_VERSION(DRV_VERSION);
62 
63 struct workqueue_struct *mlx4_wq;
64 
65 #ifdef CONFIG_MLX4_DEBUG
66 
67 int mlx4_debug_level; /* 0 by default */
68 module_param_named(debug_level, mlx4_debug_level, int, 0644);
69 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
70 
71 #endif /* CONFIG_MLX4_DEBUG */
72 
73 #ifdef CONFIG_PCI_MSI
74 
75 static int msi_x = 1;
76 module_param(msi_x, int, 0444);
77 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");
78 
79 #else /* CONFIG_PCI_MSI */
80 
81 #define msi_x (0)
82 
83 #endif /* CONFIG_PCI_MSI */
84 
85 static uint8_t num_vfs[3] = {0, 0, 0};
86 static int num_vfs_argc;
87 module_param_array(num_vfs, byte, &num_vfs_argc, 0444);
88 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
89 			  "num_vfs=port1,port2,port1+2");
90 
91 static uint8_t probe_vf[3] = {0, 0, 0};
92 static int probe_vfs_argc;
93 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
94 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
95 			   "probe_vf=port1,port2,port1+2");
96 
97 static int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
98 module_param_named(log_num_mgm_entry_size,
99 			mlx4_log_num_mgm_entry_size, int, 0444);
100 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
101 					 " of qp per mcg, for example:"
102 					 " 10 gives 248.range: 7 <="
103 					 " log_num_mgm_entry_size <= 12."
104 					 " To activate device managed"
105 					 " flow steering when available, set to -1");
106 
107 static bool enable_64b_cqe_eqe = true;
108 module_param(enable_64b_cqe_eqe, bool, 0444);
109 MODULE_PARM_DESC(enable_64b_cqe_eqe,
110 		 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
111 
112 static bool enable_4k_uar;
113 module_param(enable_4k_uar, bool, 0444);
114 MODULE_PARM_DESC(enable_4k_uar,
115 		 "Enable using 4K UAR. Should not be enabled if have VFs which do not support 4K UARs (default: false)");
116 
117 #define PF_CONTEXT_BEHAVIOUR_MASK	(MLX4_FUNC_CAP_64B_EQE_CQE | \
118 					 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
119 					 MLX4_FUNC_CAP_DMFS_A0_STATIC)
120 
121 #define RESET_PERSIST_MASK_FLAGS	(MLX4_FLAG_SRIOV)
122 
123 static char mlx4_version[] =
124 	DRV_NAME ": Mellanox ConnectX core driver v"
125 	DRV_VERSION "\n";
126 
127 static const struct mlx4_profile default_profile = {
128 	.num_qp		= 1 << 18,
129 	.num_srq	= 1 << 16,
130 	.rdmarc_per_qp	= 1 << 4,
131 	.num_cq		= 1 << 16,
132 	.num_mcg	= 1 << 13,
133 	.num_mpt	= 1 << 19,
134 	.num_mtt	= 1 << 20, /* It is really num mtt segments */
135 };
136 
137 static const struct mlx4_profile low_mem_profile = {
138 	.num_qp		= 1 << 17,
139 	.num_srq	= 1 << 6,
140 	.rdmarc_per_qp	= 1 << 4,
141 	.num_cq		= 1 << 8,
142 	.num_mcg	= 1 << 8,
143 	.num_mpt	= 1 << 9,
144 	.num_mtt	= 1 << 7,
145 };
146 
147 static int log_num_mac = 7;
148 module_param_named(log_num_mac, log_num_mac, int, 0444);
149 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
150 
151 static int log_num_vlan;
152 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
153 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
154 /* Log2 max number of VLANs per ETH port (0-7) */
155 #define MLX4_LOG_NUM_VLANS 7
156 #define MLX4_MIN_LOG_NUM_VLANS 0
157 #define MLX4_MIN_LOG_NUM_MAC 1
158 
159 static bool use_prio;
160 module_param_named(use_prio, use_prio, bool, 0444);
161 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
162 
163 int log_mtts_per_seg = ilog2(1);
164 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
165 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment "
166 		 "(0-7) (default: 0)");
167 
168 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
169 static int arr_argc = 2;
170 module_param_array(port_type_array, int, &arr_argc, 0444);
171 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
172 				"1 for IB, 2 for Ethernet");
173 
174 static atomic_t pf_loading = ATOMIC_INIT(0);
175 
176 static int mlx4_devlink_ierr_reset_get(struct devlink *devlink, u32 id,
177 				       struct devlink_param_gset_ctx *ctx,
178 				       struct netlink_ext_ack *extack)
179 {
180 	ctx->val.vbool = !!mlx4_internal_err_reset;
181 	return 0;
182 }
183 
184 static int mlx4_devlink_ierr_reset_set(struct devlink *devlink, u32 id,
185 				       struct devlink_param_gset_ctx *ctx,
186 				       struct netlink_ext_ack *extack)
187 {
188 	mlx4_internal_err_reset = ctx->val.vbool;
189 	return 0;
190 }
191 
192 static int mlx4_devlink_crdump_snapshot_get(struct devlink *devlink, u32 id,
193 					    struct devlink_param_gset_ctx *ctx,
194 					    struct netlink_ext_ack *extack)
195 {
196 	struct mlx4_priv *priv = devlink_priv(devlink);
197 	struct mlx4_dev *dev = &priv->dev;
198 
199 	ctx->val.vbool = dev->persist->crdump.snapshot_enable;
200 	return 0;
201 }
202 
203 static int mlx4_devlink_crdump_snapshot_set(struct devlink *devlink, u32 id,
204 					    struct devlink_param_gset_ctx *ctx,
205 					    struct netlink_ext_ack *extack)
206 {
207 	struct mlx4_priv *priv = devlink_priv(devlink);
208 	struct mlx4_dev *dev = &priv->dev;
209 
210 	dev->persist->crdump.snapshot_enable = ctx->val.vbool;
211 	return 0;
212 }
213 
214 static int
215 mlx4_devlink_max_macs_validate(struct devlink *devlink, u32 id,
216 			       union devlink_param_value val,
217 			       struct netlink_ext_ack *extack)
218 {
219 	u32 value = val.vu32;
220 
221 	if (value < 1 || value > 128)
222 		return -ERANGE;
223 
224 	if (!is_power_of_2(value)) {
225 		NL_SET_ERR_MSG_MOD(extack, "max_macs supported must be power of 2");
226 		return -EINVAL;
227 	}
228 
229 	return 0;
230 }
231 
232 enum mlx4_devlink_param_id {
233 	MLX4_DEVLINK_PARAM_ID_BASE = DEVLINK_PARAM_GENERIC_ID_MAX,
234 	MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
235 	MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
236 };
237 
238 static const struct devlink_param mlx4_devlink_params[] = {
239 	DEVLINK_PARAM_GENERIC(INT_ERR_RESET,
240 			      BIT(DEVLINK_PARAM_CMODE_RUNTIME) |
241 			      BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
242 			      mlx4_devlink_ierr_reset_get,
243 			      mlx4_devlink_ierr_reset_set, NULL),
244 	DEVLINK_PARAM_GENERIC(MAX_MACS,
245 			      BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
246 			      NULL, NULL, mlx4_devlink_max_macs_validate),
247 	DEVLINK_PARAM_GENERIC(REGION_SNAPSHOT,
248 			      BIT(DEVLINK_PARAM_CMODE_RUNTIME) |
249 			      BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
250 			      mlx4_devlink_crdump_snapshot_get,
251 			      mlx4_devlink_crdump_snapshot_set, NULL),
252 	DEVLINK_PARAM_DRIVER(MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
253 			     "enable_64b_cqe_eqe", DEVLINK_PARAM_TYPE_BOOL,
254 			     BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
255 			     NULL, NULL, NULL),
256 	DEVLINK_PARAM_DRIVER(MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
257 			     "enable_4k_uar", DEVLINK_PARAM_TYPE_BOOL,
258 			     BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
259 			     NULL, NULL, NULL),
260 };
261 
262 static void mlx4_devlink_set_params_init_values(struct devlink *devlink)
263 {
264 	union devlink_param_value value;
265 
266 	value.vbool = !!mlx4_internal_err_reset;
267 	devl_param_driverinit_value_set(devlink,
268 					DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET,
269 					value);
270 
271 	value.vu32 = 1UL << log_num_mac;
272 	devl_param_driverinit_value_set(devlink,
273 					DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
274 					value);
275 
276 	value.vbool = enable_64b_cqe_eqe;
277 	devl_param_driverinit_value_set(devlink,
278 					MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
279 					value);
280 
281 	value.vbool = enable_4k_uar;
282 	devl_param_driverinit_value_set(devlink,
283 					MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
284 					value);
285 
286 	value.vbool = false;
287 	devl_param_driverinit_value_set(devlink,
288 					DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT,
289 					value);
290 }
291 
292 static inline void mlx4_set_num_reserved_uars(struct mlx4_dev *dev,
293 					      struct mlx4_dev_cap *dev_cap)
294 {
295 	/* The reserved_uars is calculated by system page size unit.
296 	 * Therefore, adjustment is added when the uar page size is less
297 	 * than the system page size
298 	 */
299 	dev->caps.reserved_uars	=
300 		max_t(int,
301 		      mlx4_get_num_reserved_uar(dev),
302 		      dev_cap->reserved_uars /
303 			(1 << (PAGE_SHIFT - dev->uar_page_shift)));
304 }
305 
306 int mlx4_check_port_params(struct mlx4_dev *dev,
307 			   enum mlx4_port_type *port_type)
308 {
309 	int i;
310 
311 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
312 		for (i = 0; i < dev->caps.num_ports - 1; i++) {
313 			if (port_type[i] != port_type[i + 1]) {
314 				mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
315 				return -EOPNOTSUPP;
316 			}
317 		}
318 	}
319 
320 	for (i = 0; i < dev->caps.num_ports; i++) {
321 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
322 			mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
323 				 i + 1);
324 			return -EOPNOTSUPP;
325 		}
326 	}
327 	return 0;
328 }
329 
330 static void mlx4_set_port_mask(struct mlx4_dev *dev)
331 {
332 	int i;
333 
334 	for (i = 1; i <= dev->caps.num_ports; ++i)
335 		dev->caps.port_mask[i] = dev->caps.port_type[i];
336 }
337 
338 enum {
339 	MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
340 };
341 
342 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
343 {
344 	int err = 0;
345 	struct mlx4_func func;
346 
347 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
348 		err = mlx4_QUERY_FUNC(dev, &func, 0);
349 		if (err) {
350 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
351 			return err;
352 		}
353 		dev_cap->max_eqs = func.max_eq;
354 		dev_cap->reserved_eqs = func.rsvd_eqs;
355 		dev_cap->reserved_uars = func.rsvd_uars;
356 		err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
357 	}
358 	return err;
359 }
360 
361 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
362 {
363 	struct mlx4_caps *dev_cap = &dev->caps;
364 
365 	/* FW not supporting or cancelled by user */
366 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
367 	    !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
368 		return;
369 
370 	/* Must have 64B CQE_EQE enabled by FW to use bigger stride
371 	 * When FW has NCSI it may decide not to report 64B CQE/EQEs
372 	 */
373 	if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
374 	    !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
375 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
376 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
377 		return;
378 	}
379 
380 	if (cache_line_size() == 128 || cache_line_size() == 256) {
381 		mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
382 		/* Changing the real data inside CQE size to 32B */
383 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
384 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
385 
386 		if (mlx4_is_master(dev))
387 			dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
388 	} else {
389 		if (cache_line_size() != 32  && cache_line_size() != 64)
390 			mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
391 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
392 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
393 	}
394 }
395 
396 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
397 			  struct mlx4_port_cap *port_cap)
398 {
399 	dev->caps.vl_cap[port]	    = port_cap->max_vl;
400 	dev->caps.ib_mtu_cap[port]	    = port_cap->ib_mtu;
401 	dev->phys_caps.gid_phys_table_len[port]  = port_cap->max_gids;
402 	dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
403 	/* set gid and pkey table operating lengths by default
404 	 * to non-sriov values
405 	 */
406 	dev->caps.gid_table_len[port]  = port_cap->max_gids;
407 	dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
408 	dev->caps.port_width_cap[port] = port_cap->max_port_width;
409 	dev->caps.eth_mtu_cap[port]    = port_cap->eth_mtu;
410 	dev->caps.max_tc_eth	       = port_cap->max_tc_eth;
411 	dev->caps.def_mac[port]        = port_cap->def_mac;
412 	dev->caps.supported_type[port] = port_cap->supported_port_types;
413 	dev->caps.suggested_type[port] = port_cap->suggested_type;
414 	dev->caps.default_sense[port] = port_cap->default_sense;
415 	dev->caps.trans_type[port]	    = port_cap->trans_type;
416 	dev->caps.vendor_oui[port]     = port_cap->vendor_oui;
417 	dev->caps.wavelength[port]     = port_cap->wavelength;
418 	dev->caps.trans_code[port]     = port_cap->trans_code;
419 
420 	return 0;
421 }
422 
423 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
424 			 struct mlx4_port_cap *port_cap)
425 {
426 	int err = 0;
427 
428 	err = mlx4_QUERY_PORT(dev, port, port_cap);
429 
430 	if (err)
431 		mlx4_err(dev, "QUERY_PORT command failed.\n");
432 
433 	return err;
434 }
435 
436 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
437 {
438 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
439 		return;
440 
441 	if (mlx4_is_mfunc(dev)) {
442 		mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
443 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
444 		return;
445 	}
446 
447 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
448 		mlx4_dbg(dev,
449 			 "Keep FCS is not supported - Disabling Ignore FCS");
450 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
451 		return;
452 	}
453 }
454 
455 #define MLX4_A0_STEERING_TABLE_SIZE	256
456 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
457 {
458 	int err;
459 	int i;
460 
461 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
462 	if (err) {
463 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
464 		return err;
465 	}
466 	mlx4_dev_cap_dump(dev, dev_cap);
467 
468 	if (dev_cap->min_page_sz > PAGE_SIZE) {
469 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
470 			 dev_cap->min_page_sz, PAGE_SIZE);
471 		return -ENODEV;
472 	}
473 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
474 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
475 			 dev_cap->num_ports, MLX4_MAX_PORTS);
476 		return -ENODEV;
477 	}
478 
479 	if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
480 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
481 			 dev_cap->uar_size,
482 			 (unsigned long long)
483 			 pci_resource_len(dev->persist->pdev, 2));
484 		return -ENODEV;
485 	}
486 
487 	dev->caps.num_ports	     = dev_cap->num_ports;
488 	dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
489 	dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
490 				      dev->caps.num_sys_eqs :
491 				      MLX4_MAX_EQ_NUM;
492 	for (i = 1; i <= dev->caps.num_ports; ++i) {
493 		err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
494 		if (err) {
495 			mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
496 			return err;
497 		}
498 	}
499 
500 	dev->caps.map_clock_to_user  = dev_cap->map_clock_to_user;
501 	dev->caps.uar_page_size	     = PAGE_SIZE;
502 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
503 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
504 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
505 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
506 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
507 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
508 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
509 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
510 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
511 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
512 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
513 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
514 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
515 	/*
516 	 * Subtract 1 from the limit because we need to allocate a
517 	 * spare CQE to enable resizing the CQ.
518 	 */
519 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
520 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
521 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
522 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
523 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
524 
525 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
526 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
527 					dev_cap->reserved_xrcds : 0;
528 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
529 					dev_cap->max_xrcds : 0;
530 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
531 
532 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
533 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
534 	dev->caps.flags		     = dev_cap->flags;
535 	dev->caps.flags2	     = dev_cap->flags2;
536 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
537 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
538 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
539 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
540 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
541 	dev->caps.wol_port[1]          = dev_cap->wol_port[1];
542 	dev->caps.wol_port[2]          = dev_cap->wol_port[2];
543 	dev->caps.health_buffer_addrs  = dev_cap->health_buffer_addrs;
544 
545 	/* Save uar page shift */
546 	if (!mlx4_is_slave(dev)) {
547 		/* Virtual PCI function needs to determine UAR page size from
548 		 * firmware. Only master PCI function can set the uar page size
549 		 */
550 		if (enable_4k_uar || !dev->persist->num_vfs)
551 			dev->uar_page_shift = DEFAULT_UAR_PAGE_SHIFT;
552 		else
553 			dev->uar_page_shift = PAGE_SHIFT;
554 
555 		mlx4_set_num_reserved_uars(dev, dev_cap);
556 	}
557 
558 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN) {
559 		struct mlx4_init_hca_param hca_param;
560 
561 		memset(&hca_param, 0, sizeof(hca_param));
562 		err = mlx4_QUERY_HCA(dev, &hca_param);
563 		/* Turn off PHV_EN flag in case phv_check_en is set.
564 		 * phv_check_en is a HW check that parse the packet and verify
565 		 * phv bit was reported correctly in the wqe. To allow QinQ
566 		 * PHV_EN flag should be set and phv_check_en must be cleared
567 		 * otherwise QinQ packets will be drop by the HW.
568 		 */
569 		if (err || hca_param.phv_check_en)
570 			dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_PHV_EN;
571 	}
572 
573 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
574 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
575 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
576 	/* Don't do sense port on multifunction devices (for now at least) */
577 	if (mlx4_is_mfunc(dev))
578 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
579 
580 	if (mlx4_low_memory_profile()) {
581 		dev->caps.log_num_macs  = MLX4_MIN_LOG_NUM_MAC;
582 		dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
583 	} else {
584 		dev->caps.log_num_macs  = log_num_mac;
585 		dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
586 	}
587 
588 	for (i = 1; i <= dev->caps.num_ports; ++i) {
589 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
590 		if (dev->caps.supported_type[i]) {
591 			/* if only ETH is supported - assign ETH */
592 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
593 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
594 			/* if only IB is supported, assign IB */
595 			else if (dev->caps.supported_type[i] ==
596 				 MLX4_PORT_TYPE_IB)
597 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
598 			else {
599 				/* if IB and ETH are supported, we set the port
600 				 * type according to user selection of port type;
601 				 * if user selected none, take the FW hint */
602 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
603 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
604 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
605 				else
606 					dev->caps.port_type[i] = port_type_array[i - 1];
607 			}
608 		}
609 		/*
610 		 * Link sensing is allowed on the port if 3 conditions are true:
611 		 * 1. Both protocols are supported on the port.
612 		 * 2. Different types are supported on the port
613 		 * 3. FW declared that it supports link sensing
614 		 */
615 		mlx4_priv(dev)->sense.sense_allowed[i] =
616 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
617 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
618 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
619 
620 		/*
621 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
622 		 * and perform sense_port FW command to try and set the correct
623 		 * port type from beginning
624 		 */
625 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
626 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
627 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
628 			mlx4_SENSE_PORT(dev, i, &sensed_port);
629 			if (sensed_port != MLX4_PORT_TYPE_NONE)
630 				dev->caps.port_type[i] = sensed_port;
631 		} else {
632 			dev->caps.possible_type[i] = dev->caps.port_type[i];
633 		}
634 
635 		if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
636 			dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
637 			mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
638 				  i, 1 << dev->caps.log_num_macs);
639 		}
640 		if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
641 			dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
642 			mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
643 				  i, 1 << dev->caps.log_num_vlans);
644 		}
645 	}
646 
647 	if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
648 	    (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
649 	    (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
650 		mlx4_warn(dev,
651 			  "Granular QoS per VF not supported with IB/Eth configuration\n");
652 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
653 	}
654 
655 	dev->caps.max_counters = dev_cap->max_counters;
656 
657 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
658 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
659 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
660 		(1 << dev->caps.log_num_macs) *
661 		(1 << dev->caps.log_num_vlans) *
662 		dev->caps.num_ports;
663 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
664 
665 	if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
666 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
667 		dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
668 	else
669 		dev->caps.dmfs_high_rate_qpn_base =
670 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
671 
672 	if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
673 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
674 		dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
675 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
676 		dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
677 	} else {
678 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
679 		dev->caps.dmfs_high_rate_qpn_base =
680 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
681 		dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
682 	}
683 
684 	dev->caps.rl_caps = dev_cap->rl_caps;
685 
686 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
687 		dev->caps.dmfs_high_rate_qpn_range;
688 
689 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
690 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
691 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
692 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
693 
694 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
695 
696 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
697 		if (dev_cap->flags &
698 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
699 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
700 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
701 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
702 		}
703 
704 		if (dev_cap->flags2 &
705 		    (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
706 		     MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
707 			mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
708 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
709 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
710 		}
711 	}
712 
713 	if ((dev->caps.flags &
714 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
715 	    mlx4_is_master(dev))
716 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
717 
718 	if (!mlx4_is_slave(dev)) {
719 		mlx4_enable_cqe_eqe_stride(dev);
720 		dev->caps.alloc_res_qp_mask =
721 			(dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
722 			MLX4_RESERVE_A0_QP;
723 
724 		if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
725 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
726 			mlx4_warn(dev, "Old device ETS support detected\n");
727 			mlx4_warn(dev, "Consider upgrading device FW.\n");
728 			dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
729 		}
730 
731 	} else {
732 		dev->caps.alloc_res_qp_mask = 0;
733 	}
734 
735 	mlx4_enable_ignore_fcs(dev);
736 
737 	return 0;
738 }
739 
740 /*The function checks if there are live vf, return the num of them*/
741 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
742 {
743 	struct mlx4_priv *priv = mlx4_priv(dev);
744 	struct mlx4_slave_state *s_state;
745 	int i;
746 	int ret = 0;
747 
748 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
749 		s_state = &priv->mfunc.master.slave_state[i];
750 		if (s_state->active && s_state->last_cmd !=
751 		    MLX4_COMM_CMD_RESET) {
752 			mlx4_warn(dev, "%s: slave: %d is still active\n",
753 				  __func__, i);
754 			ret++;
755 		}
756 	}
757 	return ret;
758 }
759 
760 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
761 {
762 	u32 qk = MLX4_RESERVED_QKEY_BASE;
763 
764 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
765 	    qpn < dev->phys_caps.base_proxy_sqpn)
766 		return -EINVAL;
767 
768 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
769 		/* tunnel qp */
770 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
771 	else
772 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
773 	*qkey = qk;
774 	return 0;
775 }
776 EXPORT_SYMBOL(mlx4_get_parav_qkey);
777 
778 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
779 {
780 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
781 
782 	if (!mlx4_is_master(dev))
783 		return;
784 
785 	priv->virt2phys_pkey[slave][port - 1][i] = val;
786 }
787 EXPORT_SYMBOL(mlx4_sync_pkey_table);
788 
789 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
790 {
791 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
792 
793 	if (!mlx4_is_master(dev))
794 		return;
795 
796 	priv->slave_node_guids[slave] = guid;
797 }
798 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
799 
800 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
801 {
802 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
803 
804 	if (!mlx4_is_master(dev))
805 		return 0;
806 
807 	return priv->slave_node_guids[slave];
808 }
809 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
810 
811 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
812 {
813 	struct mlx4_priv *priv = mlx4_priv(dev);
814 	struct mlx4_slave_state *s_slave;
815 
816 	if (!mlx4_is_master(dev))
817 		return 0;
818 
819 	s_slave = &priv->mfunc.master.slave_state[slave];
820 	return !!s_slave->active;
821 }
822 EXPORT_SYMBOL(mlx4_is_slave_active);
823 
824 void mlx4_handle_eth_header_mcast_prio(struct mlx4_net_trans_rule_hw_ctrl *ctrl,
825 				       struct _rule_hw *eth_header)
826 {
827 	if (is_multicast_ether_addr(eth_header->eth.dst_mac) ||
828 	    is_broadcast_ether_addr(eth_header->eth.dst_mac)) {
829 		struct mlx4_net_trans_rule_hw_eth *eth =
830 			(struct mlx4_net_trans_rule_hw_eth *)eth_header;
831 		struct _rule_hw *next_rule = (struct _rule_hw *)(eth + 1);
832 		bool last_rule = next_rule->size == 0 && next_rule->id == 0 &&
833 			next_rule->rsvd == 0;
834 
835 		if (last_rule)
836 			ctrl->prio = cpu_to_be16(MLX4_DOMAIN_NIC);
837 	}
838 }
839 EXPORT_SYMBOL(mlx4_handle_eth_header_mcast_prio);
840 
841 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
842 				       struct mlx4_dev_cap *dev_cap,
843 				       struct mlx4_init_hca_param *hca_param)
844 {
845 	dev->caps.steering_mode = hca_param->steering_mode;
846 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
847 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
848 		dev->caps.fs_log_max_ucast_qp_range_size =
849 			dev_cap->fs_log_max_ucast_qp_range_size;
850 	} else
851 		dev->caps.num_qp_per_mgm =
852 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
853 
854 	mlx4_dbg(dev, "Steering mode is: %s\n",
855 		 mlx4_steering_mode_str(dev->caps.steering_mode));
856 }
857 
858 static void mlx4_slave_destroy_special_qp_cap(struct mlx4_dev *dev)
859 {
860 	kfree(dev->caps.spec_qps);
861 	dev->caps.spec_qps = NULL;
862 }
863 
864 static int mlx4_slave_special_qp_cap(struct mlx4_dev *dev)
865 {
866 	struct mlx4_func_cap *func_cap;
867 	struct mlx4_caps *caps = &dev->caps;
868 	int i, err = 0;
869 
870 	func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
871 	caps->spec_qps = kcalloc(caps->num_ports, sizeof(*caps->spec_qps), GFP_KERNEL);
872 
873 	if (!func_cap || !caps->spec_qps) {
874 		mlx4_err(dev, "Failed to allocate memory for special qps cap\n");
875 		err = -ENOMEM;
876 		goto err_mem;
877 	}
878 
879 	for (i = 1; i <= caps->num_ports; ++i) {
880 		err = mlx4_QUERY_FUNC_CAP(dev, i, func_cap);
881 		if (err) {
882 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
883 				 i, err);
884 			goto err_mem;
885 		}
886 		caps->spec_qps[i - 1] = func_cap->spec_qps;
887 		caps->port_mask[i] = caps->port_type[i];
888 		caps->phys_port_id[i] = func_cap->phys_port_id;
889 		err = mlx4_get_slave_pkey_gid_tbl_len(dev, i,
890 						      &caps->gid_table_len[i],
891 						      &caps->pkey_table_len[i]);
892 		if (err) {
893 			mlx4_err(dev, "QUERY_PORT command failed for port %d, aborting (%d)\n",
894 				 i, err);
895 			goto err_mem;
896 		}
897 	}
898 
899 err_mem:
900 	if (err)
901 		mlx4_slave_destroy_special_qp_cap(dev);
902 	kfree(func_cap);
903 	return err;
904 }
905 
906 static int mlx4_slave_cap(struct mlx4_dev *dev)
907 {
908 	int			   err;
909 	u32			   page_size;
910 	struct mlx4_dev_cap	   *dev_cap;
911 	struct mlx4_func_cap	   *func_cap;
912 	struct mlx4_init_hca_param *hca_param;
913 
914 	hca_param = kzalloc(sizeof(*hca_param), GFP_KERNEL);
915 	func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
916 	dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
917 	if (!hca_param || !func_cap || !dev_cap) {
918 		mlx4_err(dev, "Failed to allocate memory for slave_cap\n");
919 		err = -ENOMEM;
920 		goto free_mem;
921 	}
922 
923 	err = mlx4_QUERY_HCA(dev, hca_param);
924 	if (err) {
925 		mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
926 		goto free_mem;
927 	}
928 
929 	/* fail if the hca has an unknown global capability
930 	 * at this time global_caps should be always zeroed
931 	 */
932 	if (hca_param->global_caps) {
933 		mlx4_err(dev, "Unknown hca global capabilities\n");
934 		err = -EINVAL;
935 		goto free_mem;
936 	}
937 
938 	dev->caps.hca_core_clock = hca_param->hca_core_clock;
939 
940 	dev->caps.max_qp_dest_rdma = 1 << hca_param->log_rd_per_qp;
941 	err = mlx4_dev_cap(dev, dev_cap);
942 	if (err) {
943 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
944 		goto free_mem;
945 	}
946 
947 	err = mlx4_QUERY_FW(dev);
948 	if (err)
949 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
950 
951 	page_size = ~dev->caps.page_size_cap + 1;
952 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
953 	if (page_size > PAGE_SIZE) {
954 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
955 			 page_size, PAGE_SIZE);
956 		err = -ENODEV;
957 		goto free_mem;
958 	}
959 
960 	/* Set uar_page_shift for VF */
961 	dev->uar_page_shift = hca_param->uar_page_sz + 12;
962 
963 	/* Make sure the master uar page size is valid */
964 	if (dev->uar_page_shift > PAGE_SHIFT) {
965 		mlx4_err(dev,
966 			 "Invalid configuration: uar page size is larger than system page size\n");
967 		err = -ENODEV;
968 		goto free_mem;
969 	}
970 
971 	/* Set reserved_uars based on the uar_page_shift */
972 	mlx4_set_num_reserved_uars(dev, dev_cap);
973 
974 	/* Although uar page size in FW differs from system page size,
975 	 * upper software layers (mlx4_ib, mlx4_en and part of mlx4_core)
976 	 * still works with assumption that uar page size == system page size
977 	 */
978 	dev->caps.uar_page_size = PAGE_SIZE;
979 
980 	err = mlx4_QUERY_FUNC_CAP(dev, 0, func_cap);
981 	if (err) {
982 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
983 			 err);
984 		goto free_mem;
985 	}
986 
987 	if ((func_cap->pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
988 	    PF_CONTEXT_BEHAVIOUR_MASK) {
989 		mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
990 			 func_cap->pf_context_behaviour,
991 			 PF_CONTEXT_BEHAVIOUR_MASK);
992 		err = -EINVAL;
993 		goto free_mem;
994 	}
995 
996 	dev->caps.num_ports		= func_cap->num_ports;
997 	dev->quotas.qp			= func_cap->qp_quota;
998 	dev->quotas.srq			= func_cap->srq_quota;
999 	dev->quotas.cq			= func_cap->cq_quota;
1000 	dev->quotas.mpt			= func_cap->mpt_quota;
1001 	dev->quotas.mtt			= func_cap->mtt_quota;
1002 	dev->caps.num_qps		= 1 << hca_param->log_num_qps;
1003 	dev->caps.num_srqs		= 1 << hca_param->log_num_srqs;
1004 	dev->caps.num_cqs		= 1 << hca_param->log_num_cqs;
1005 	dev->caps.num_mpts		= 1 << hca_param->log_mpt_sz;
1006 	dev->caps.num_eqs		= func_cap->max_eq;
1007 	dev->caps.reserved_eqs		= func_cap->reserved_eq;
1008 	dev->caps.reserved_lkey		= func_cap->reserved_lkey;
1009 	dev->caps.num_pds               = MLX4_NUM_PDS;
1010 	dev->caps.num_mgms              = 0;
1011 	dev->caps.num_amgms             = 0;
1012 
1013 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
1014 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
1015 			 dev->caps.num_ports, MLX4_MAX_PORTS);
1016 		err = -ENODEV;
1017 		goto free_mem;
1018 	}
1019 
1020 	mlx4_replace_zero_macs(dev);
1021 
1022 	err = mlx4_slave_special_qp_cap(dev);
1023 	if (err) {
1024 		mlx4_err(dev, "Set special QP caps failed. aborting\n");
1025 		goto free_mem;
1026 	}
1027 
1028 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
1029 				       dev->caps.reserved_uars) >
1030 				       pci_resource_len(dev->persist->pdev,
1031 							2)) {
1032 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
1033 			 dev->caps.uar_page_size * dev->caps.num_uars,
1034 			 (unsigned long long)
1035 			 pci_resource_len(dev->persist->pdev, 2));
1036 		err = -ENOMEM;
1037 		goto err_mem;
1038 	}
1039 
1040 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
1041 		dev->caps.eqe_size   = 64;
1042 		dev->caps.eqe_factor = 1;
1043 	} else {
1044 		dev->caps.eqe_size   = 32;
1045 		dev->caps.eqe_factor = 0;
1046 	}
1047 
1048 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
1049 		dev->caps.cqe_size   = 64;
1050 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1051 	} else {
1052 		dev->caps.cqe_size   = 32;
1053 	}
1054 
1055 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
1056 		dev->caps.eqe_size = hca_param->eqe_size;
1057 		dev->caps.eqe_factor = 0;
1058 	}
1059 
1060 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
1061 		dev->caps.cqe_size = hca_param->cqe_size;
1062 		/* User still need to know when CQE > 32B */
1063 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1064 	}
1065 
1066 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1067 	mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
1068 
1069 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_USER_MAC_EN;
1070 	mlx4_dbg(dev, "User MAC FW update is not supported in slave mode\n");
1071 
1072 	slave_adjust_steering_mode(dev, dev_cap, hca_param);
1073 	mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
1074 		 hca_param->rss_ip_frags ? "on" : "off");
1075 
1076 	if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
1077 	    dev->caps.bf_reg_size)
1078 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
1079 
1080 	if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
1081 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
1082 
1083 err_mem:
1084 	if (err)
1085 		mlx4_slave_destroy_special_qp_cap(dev);
1086 free_mem:
1087 	kfree(hca_param);
1088 	kfree(func_cap);
1089 	kfree(dev_cap);
1090 	return err;
1091 }
1092 
1093 /*
1094  * Change the port configuration of the device.
1095  * Every user of this function must hold the port mutex.
1096  */
1097 int mlx4_change_port_types(struct mlx4_dev *dev,
1098 			   enum mlx4_port_type *port_types)
1099 {
1100 	int err = 0;
1101 	int change = 0;
1102 	int port;
1103 
1104 	for (port = 0; port <  dev->caps.num_ports; port++) {
1105 		/* Change the port type only if the new type is different
1106 		 * from the current, and not set to Auto */
1107 		if (port_types[port] != dev->caps.port_type[port + 1])
1108 			change = 1;
1109 	}
1110 	if (change) {
1111 		mlx4_unregister_device(dev);
1112 		for (port = 1; port <= dev->caps.num_ports; port++) {
1113 			mlx4_CLOSE_PORT(dev, port);
1114 			dev->caps.port_type[port] = port_types[port - 1];
1115 			err = mlx4_SET_PORT(dev, port, -1);
1116 			if (err) {
1117 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1118 					 port);
1119 				goto out;
1120 			}
1121 		}
1122 		mlx4_set_port_mask(dev);
1123 		err = mlx4_register_device(dev);
1124 		if (err) {
1125 			mlx4_err(dev, "Failed to register device\n");
1126 			goto out;
1127 		}
1128 	}
1129 
1130 out:
1131 	return err;
1132 }
1133 
1134 static ssize_t show_port_type(struct device *dev,
1135 			      struct device_attribute *attr,
1136 			      char *buf)
1137 {
1138 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1139 						   port_attr);
1140 	struct mlx4_dev *mdev = info->dev;
1141 	char type[8];
1142 
1143 	sprintf(type, "%s",
1144 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1145 		"ib" : "eth");
1146 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1147 		sprintf(buf, "auto (%s)\n", type);
1148 	else
1149 		sprintf(buf, "%s\n", type);
1150 
1151 	return strlen(buf);
1152 }
1153 
1154 static int __set_port_type(struct mlx4_port_info *info,
1155 			   enum mlx4_port_type port_type)
1156 {
1157 	struct mlx4_dev *mdev = info->dev;
1158 	struct mlx4_priv *priv = mlx4_priv(mdev);
1159 	enum mlx4_port_type types[MLX4_MAX_PORTS];
1160 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1161 	int i;
1162 	int err = 0;
1163 
1164 	if ((port_type & mdev->caps.supported_type[info->port]) != port_type) {
1165 		mlx4_err(mdev,
1166 			 "Requested port type for port %d is not supported on this HCA\n",
1167 			 info->port);
1168 		return -EOPNOTSUPP;
1169 	}
1170 
1171 	mlx4_stop_sense(mdev);
1172 	mutex_lock(&priv->port_mutex);
1173 	info->tmp_type = port_type;
1174 
1175 	/* Possible type is always the one that was delivered */
1176 	mdev->caps.possible_type[info->port] = info->tmp_type;
1177 
1178 	for (i = 0; i < mdev->caps.num_ports; i++) {
1179 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1180 					mdev->caps.possible_type[i+1];
1181 		if (types[i] == MLX4_PORT_TYPE_AUTO)
1182 			types[i] = mdev->caps.port_type[i+1];
1183 	}
1184 
1185 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1186 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1187 		for (i = 1; i <= mdev->caps.num_ports; i++) {
1188 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1189 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1190 				err = -EOPNOTSUPP;
1191 			}
1192 		}
1193 	}
1194 	if (err) {
1195 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1196 		goto out;
1197 	}
1198 
1199 	mlx4_do_sense_ports(mdev, new_types, types);
1200 
1201 	err = mlx4_check_port_params(mdev, new_types);
1202 	if (err)
1203 		goto out;
1204 
1205 	/* We are about to apply the changes after the configuration
1206 	 * was verified, no need to remember the temporary types
1207 	 * any more */
1208 	for (i = 0; i < mdev->caps.num_ports; i++)
1209 		priv->port[i + 1].tmp_type = 0;
1210 
1211 	err = mlx4_change_port_types(mdev, new_types);
1212 
1213 out:
1214 	mlx4_start_sense(mdev);
1215 	mutex_unlock(&priv->port_mutex);
1216 
1217 	return err;
1218 }
1219 
1220 static ssize_t set_port_type(struct device *dev,
1221 			     struct device_attribute *attr,
1222 			     const char *buf, size_t count)
1223 {
1224 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1225 						   port_attr);
1226 	struct mlx4_dev *mdev = info->dev;
1227 	enum mlx4_port_type port_type;
1228 	static DEFINE_MUTEX(set_port_type_mutex);
1229 	int err;
1230 
1231 	mutex_lock(&set_port_type_mutex);
1232 
1233 	if (!strcmp(buf, "ib\n")) {
1234 		port_type = MLX4_PORT_TYPE_IB;
1235 	} else if (!strcmp(buf, "eth\n")) {
1236 		port_type = MLX4_PORT_TYPE_ETH;
1237 	} else if (!strcmp(buf, "auto\n")) {
1238 		port_type = MLX4_PORT_TYPE_AUTO;
1239 	} else {
1240 		mlx4_err(mdev, "%s is not supported port type\n", buf);
1241 		err = -EINVAL;
1242 		goto err_out;
1243 	}
1244 
1245 	err = __set_port_type(info, port_type);
1246 
1247 err_out:
1248 	mutex_unlock(&set_port_type_mutex);
1249 
1250 	return err ? err : count;
1251 }
1252 
1253 static inline int int_to_ibta_mtu(int mtu)
1254 {
1255 	switch (mtu) {
1256 	case 256:  return IB_MTU_256;
1257 	case 512:  return IB_MTU_512;
1258 	case 1024: return IB_MTU_1024;
1259 	case 2048: return IB_MTU_2048;
1260 	case 4096: return IB_MTU_4096;
1261 	default: return -1;
1262 	}
1263 }
1264 
1265 static inline int ibta_mtu_to_int(enum ib_mtu mtu)
1266 {
1267 	switch (mtu) {
1268 	case IB_MTU_256:  return  256;
1269 	case IB_MTU_512:  return  512;
1270 	case IB_MTU_1024: return 1024;
1271 	case IB_MTU_2048: return 2048;
1272 	case IB_MTU_4096: return 4096;
1273 	default: return -1;
1274 	}
1275 }
1276 
1277 static ssize_t show_port_ib_mtu(struct device *dev,
1278 			     struct device_attribute *attr,
1279 			     char *buf)
1280 {
1281 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1282 						   port_mtu_attr);
1283 	struct mlx4_dev *mdev = info->dev;
1284 
1285 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1286 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1287 
1288 	sprintf(buf, "%d\n",
1289 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1290 	return strlen(buf);
1291 }
1292 
1293 static ssize_t set_port_ib_mtu(struct device *dev,
1294 			     struct device_attribute *attr,
1295 			     const char *buf, size_t count)
1296 {
1297 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1298 						   port_mtu_attr);
1299 	struct mlx4_dev *mdev = info->dev;
1300 	struct mlx4_priv *priv = mlx4_priv(mdev);
1301 	int err, port, mtu, ibta_mtu = -1;
1302 
1303 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1304 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1305 		return -EINVAL;
1306 	}
1307 
1308 	err = kstrtoint(buf, 0, &mtu);
1309 	if (!err)
1310 		ibta_mtu = int_to_ibta_mtu(mtu);
1311 
1312 	if (err || ibta_mtu < 0) {
1313 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1314 		return -EINVAL;
1315 	}
1316 
1317 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1318 
1319 	mlx4_stop_sense(mdev);
1320 	mutex_lock(&priv->port_mutex);
1321 	mlx4_unregister_device(mdev);
1322 	for (port = 1; port <= mdev->caps.num_ports; port++) {
1323 		mlx4_CLOSE_PORT(mdev, port);
1324 		err = mlx4_SET_PORT(mdev, port, -1);
1325 		if (err) {
1326 			mlx4_err(mdev, "Failed to set port %d, aborting\n",
1327 				 port);
1328 			goto err_set_port;
1329 		}
1330 	}
1331 	err = mlx4_register_device(mdev);
1332 err_set_port:
1333 	mutex_unlock(&priv->port_mutex);
1334 	mlx4_start_sense(mdev);
1335 	return err ? err : count;
1336 }
1337 
1338 /* bond for multi-function device */
1339 #define MAX_MF_BOND_ALLOWED_SLAVES 63
1340 static int mlx4_mf_bond(struct mlx4_dev *dev)
1341 {
1342 	int err = 0;
1343 	int nvfs;
1344 	struct mlx4_slaves_pport slaves_port1;
1345 	struct mlx4_slaves_pport slaves_port2;
1346 
1347 	slaves_port1 = mlx4_phys_to_slaves_pport(dev, 1);
1348 	slaves_port2 = mlx4_phys_to_slaves_pport(dev, 2);
1349 
1350 	/* only single port vfs are allowed */
1351 	if (bitmap_weight_and(slaves_port1.slaves, slaves_port2.slaves,
1352 			      dev->persist->num_vfs + 1) > 1) {
1353 		mlx4_warn(dev, "HA mode unsupported for dual ported VFs\n");
1354 		return -EINVAL;
1355 	}
1356 
1357 	/* number of virtual functions is number of total functions minus one
1358 	 * physical function for each port.
1359 	 */
1360 	nvfs = bitmap_weight(slaves_port1.slaves, dev->persist->num_vfs + 1) +
1361 		bitmap_weight(slaves_port2.slaves, dev->persist->num_vfs + 1) - 2;
1362 
1363 	/* limit on maximum allowed VFs */
1364 	if (nvfs > MAX_MF_BOND_ALLOWED_SLAVES) {
1365 		mlx4_warn(dev, "HA mode is not supported for %d VFs (max %d are allowed)\n",
1366 			  nvfs, MAX_MF_BOND_ALLOWED_SLAVES);
1367 		return -EINVAL;
1368 	}
1369 
1370 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) {
1371 		mlx4_warn(dev, "HA mode unsupported for NON DMFS steering\n");
1372 		return -EINVAL;
1373 	}
1374 
1375 	err = mlx4_bond_mac_table(dev);
1376 	if (err)
1377 		return err;
1378 	err = mlx4_bond_vlan_table(dev);
1379 	if (err)
1380 		goto err1;
1381 	err = mlx4_bond_fs_rules(dev);
1382 	if (err)
1383 		goto err2;
1384 
1385 	return 0;
1386 err2:
1387 	(void)mlx4_unbond_vlan_table(dev);
1388 err1:
1389 	(void)mlx4_unbond_mac_table(dev);
1390 	return err;
1391 }
1392 
1393 static int mlx4_mf_unbond(struct mlx4_dev *dev)
1394 {
1395 	int ret, ret1;
1396 
1397 	ret = mlx4_unbond_fs_rules(dev);
1398 	if (ret)
1399 		mlx4_warn(dev, "multifunction unbond for flow rules failed (%d)\n", ret);
1400 	ret1 = mlx4_unbond_mac_table(dev);
1401 	if (ret1) {
1402 		mlx4_warn(dev, "multifunction unbond for MAC table failed (%d)\n", ret1);
1403 		ret = ret1;
1404 	}
1405 	ret1 = mlx4_unbond_vlan_table(dev);
1406 	if (ret1) {
1407 		mlx4_warn(dev, "multifunction unbond for VLAN table failed (%d)\n", ret1);
1408 		ret = ret1;
1409 	}
1410 	return ret;
1411 }
1412 
1413 static int mlx4_bond(struct mlx4_dev *dev)
1414 {
1415 	int ret = 0;
1416 	struct mlx4_priv *priv = mlx4_priv(dev);
1417 
1418 	mutex_lock(&priv->bond_mutex);
1419 
1420 	if (!mlx4_is_bonded(dev)) {
1421 		ret = mlx4_do_bond(dev, true);
1422 		if (ret)
1423 			mlx4_err(dev, "Failed to bond device: %d\n", ret);
1424 		if (!ret && mlx4_is_master(dev)) {
1425 			ret = mlx4_mf_bond(dev);
1426 			if (ret) {
1427 				mlx4_err(dev, "bond for multifunction failed\n");
1428 				mlx4_do_bond(dev, false);
1429 			}
1430 		}
1431 	}
1432 
1433 	mutex_unlock(&priv->bond_mutex);
1434 	if (!ret)
1435 		mlx4_dbg(dev, "Device is bonded\n");
1436 
1437 	return ret;
1438 }
1439 
1440 static int mlx4_unbond(struct mlx4_dev *dev)
1441 {
1442 	int ret = 0;
1443 	struct mlx4_priv *priv = mlx4_priv(dev);
1444 
1445 	mutex_lock(&priv->bond_mutex);
1446 
1447 	if (mlx4_is_bonded(dev)) {
1448 		int ret2 = 0;
1449 
1450 		ret = mlx4_do_bond(dev, false);
1451 		if (ret)
1452 			mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1453 		if (mlx4_is_master(dev))
1454 			ret2 = mlx4_mf_unbond(dev);
1455 		if (ret2) {
1456 			mlx4_warn(dev, "Failed to unbond device for multifunction (%d)\n", ret2);
1457 			ret = ret2;
1458 		}
1459 	}
1460 
1461 	mutex_unlock(&priv->bond_mutex);
1462 	if (!ret)
1463 		mlx4_dbg(dev, "Device is unbonded\n");
1464 
1465 	return ret;
1466 }
1467 
1468 static int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1469 {
1470 	u8 port1 = v2p->port1;
1471 	u8 port2 = v2p->port2;
1472 	struct mlx4_priv *priv = mlx4_priv(dev);
1473 	int err;
1474 
1475 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1476 		return -EOPNOTSUPP;
1477 
1478 	mutex_lock(&priv->bond_mutex);
1479 
1480 	/* zero means keep current mapping for this port */
1481 	if (port1 == 0)
1482 		port1 = priv->v2p.port1;
1483 	if (port2 == 0)
1484 		port2 = priv->v2p.port2;
1485 
1486 	if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1487 	    (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1488 	    (port1 == 2 && port2 == 1)) {
1489 		/* besides boundary checks cross mapping makes
1490 		 * no sense and therefore not allowed */
1491 		err = -EINVAL;
1492 	} else if ((port1 == priv->v2p.port1) &&
1493 		 (port2 == priv->v2p.port2)) {
1494 		err = 0;
1495 	} else {
1496 		err = mlx4_virt2phy_port_map(dev, port1, port2);
1497 		if (!err) {
1498 			mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1499 				 port1, port2);
1500 			priv->v2p.port1 = port1;
1501 			priv->v2p.port2 = port2;
1502 		} else {
1503 			mlx4_err(dev, "Failed to change port map: %d\n", err);
1504 		}
1505 	}
1506 
1507 	mutex_unlock(&priv->bond_mutex);
1508 	return err;
1509 }
1510 
1511 struct mlx4_bond {
1512 	struct work_struct work;
1513 	struct mlx4_dev *dev;
1514 	int is_bonded;
1515 	struct mlx4_port_map port_map;
1516 };
1517 
1518 static void mlx4_bond_work(struct work_struct *work)
1519 {
1520 	struct mlx4_bond *bond = container_of(work, struct mlx4_bond, work);
1521 	int err = 0;
1522 
1523 	if (bond->is_bonded) {
1524 		if (!mlx4_is_bonded(bond->dev)) {
1525 			err = mlx4_bond(bond->dev);
1526 			if (err)
1527 				mlx4_err(bond->dev, "Fail to bond device\n");
1528 		}
1529 		if (!err) {
1530 			err = mlx4_port_map_set(bond->dev, &bond->port_map);
1531 			if (err)
1532 				mlx4_err(bond->dev,
1533 					 "Fail to set port map [%d][%d]: %d\n",
1534 					 bond->port_map.port1,
1535 					 bond->port_map.port2, err);
1536 		}
1537 	} else if (mlx4_is_bonded(bond->dev)) {
1538 		err = mlx4_unbond(bond->dev);
1539 		if (err)
1540 			mlx4_err(bond->dev, "Fail to unbond device\n");
1541 	}
1542 	put_device(&bond->dev->persist->pdev->dev);
1543 	kfree(bond);
1544 }
1545 
1546 int mlx4_queue_bond_work(struct mlx4_dev *dev, int is_bonded, u8 v2p_p1,
1547 			 u8 v2p_p2)
1548 {
1549 	struct mlx4_bond *bond;
1550 
1551 	bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
1552 	if (!bond)
1553 		return -ENOMEM;
1554 
1555 	INIT_WORK(&bond->work, mlx4_bond_work);
1556 	get_device(&dev->persist->pdev->dev);
1557 	bond->dev = dev;
1558 	bond->is_bonded = is_bonded;
1559 	bond->port_map.port1 = v2p_p1;
1560 	bond->port_map.port2 = v2p_p2;
1561 	queue_work(mlx4_wq, &bond->work);
1562 	return 0;
1563 }
1564 EXPORT_SYMBOL(mlx4_queue_bond_work);
1565 
1566 static int mlx4_load_fw(struct mlx4_dev *dev)
1567 {
1568 	struct mlx4_priv *priv = mlx4_priv(dev);
1569 	int err;
1570 
1571 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1572 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
1573 	if (!priv->fw.fw_icm) {
1574 		mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1575 		return -ENOMEM;
1576 	}
1577 
1578 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1579 	if (err) {
1580 		mlx4_err(dev, "MAP_FA command failed, aborting\n");
1581 		goto err_free;
1582 	}
1583 
1584 	err = mlx4_RUN_FW(dev);
1585 	if (err) {
1586 		mlx4_err(dev, "RUN_FW command failed, aborting\n");
1587 		goto err_unmap_fa;
1588 	}
1589 
1590 	return 0;
1591 
1592 err_unmap_fa:
1593 	mlx4_UNMAP_FA(dev);
1594 
1595 err_free:
1596 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1597 	return err;
1598 }
1599 
1600 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1601 				int cmpt_entry_sz)
1602 {
1603 	struct mlx4_priv *priv = mlx4_priv(dev);
1604 	int err;
1605 	int num_eqs;
1606 
1607 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1608 				  cmpt_base +
1609 				  ((u64) (MLX4_CMPT_TYPE_QP *
1610 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1611 				  cmpt_entry_sz, dev->caps.num_qps,
1612 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1613 				  0, 0);
1614 	if (err)
1615 		goto err;
1616 
1617 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1618 				  cmpt_base +
1619 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1620 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1621 				  cmpt_entry_sz, dev->caps.num_srqs,
1622 				  dev->caps.reserved_srqs, 0, 0);
1623 	if (err)
1624 		goto err_qp;
1625 
1626 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1627 				  cmpt_base +
1628 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1629 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1630 				  cmpt_entry_sz, dev->caps.num_cqs,
1631 				  dev->caps.reserved_cqs, 0, 0);
1632 	if (err)
1633 		goto err_srq;
1634 
1635 	num_eqs = dev->phys_caps.num_phys_eqs;
1636 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1637 				  cmpt_base +
1638 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1639 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1640 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1641 	if (err)
1642 		goto err_cq;
1643 
1644 	return 0;
1645 
1646 err_cq:
1647 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1648 
1649 err_srq:
1650 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1651 
1652 err_qp:
1653 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1654 
1655 err:
1656 	return err;
1657 }
1658 
1659 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1660 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1661 {
1662 	struct mlx4_priv *priv = mlx4_priv(dev);
1663 	u64 aux_pages;
1664 	int num_eqs;
1665 	int err;
1666 
1667 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1668 	if (err) {
1669 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1670 		return err;
1671 	}
1672 
1673 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1674 		 (unsigned long long) icm_size >> 10,
1675 		 (unsigned long long) aux_pages << 2);
1676 
1677 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1678 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1679 	if (!priv->fw.aux_icm) {
1680 		mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1681 		return -ENOMEM;
1682 	}
1683 
1684 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1685 	if (err) {
1686 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1687 		goto err_free_aux;
1688 	}
1689 
1690 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1691 	if (err) {
1692 		mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1693 		goto err_unmap_aux;
1694 	}
1695 
1696 
1697 	num_eqs = dev->phys_caps.num_phys_eqs;
1698 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1699 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1700 				  num_eqs, num_eqs, 0, 0);
1701 	if (err) {
1702 		mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1703 		goto err_unmap_cmpt;
1704 	}
1705 
1706 	/*
1707 	 * Reserved MTT entries must be aligned up to a cacheline
1708 	 * boundary, since the FW will write to them, while the driver
1709 	 * writes to all other MTT entries. (The variable
1710 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1711 	 * size, not the raw entry size)
1712 	 */
1713 	dev->caps.reserved_mtts =
1714 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1715 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1716 
1717 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1718 				  init_hca->mtt_base,
1719 				  dev->caps.mtt_entry_sz,
1720 				  dev->caps.num_mtts,
1721 				  dev->caps.reserved_mtts, 1, 0);
1722 	if (err) {
1723 		mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1724 		goto err_unmap_eq;
1725 	}
1726 
1727 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1728 				  init_hca->dmpt_base,
1729 				  dev_cap->dmpt_entry_sz,
1730 				  dev->caps.num_mpts,
1731 				  dev->caps.reserved_mrws, 1, 1);
1732 	if (err) {
1733 		mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1734 		goto err_unmap_mtt;
1735 	}
1736 
1737 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1738 				  init_hca->qpc_base,
1739 				  dev_cap->qpc_entry_sz,
1740 				  dev->caps.num_qps,
1741 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1742 				  0, 0);
1743 	if (err) {
1744 		mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1745 		goto err_unmap_dmpt;
1746 	}
1747 
1748 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1749 				  init_hca->auxc_base,
1750 				  dev_cap->aux_entry_sz,
1751 				  dev->caps.num_qps,
1752 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1753 				  0, 0);
1754 	if (err) {
1755 		mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1756 		goto err_unmap_qp;
1757 	}
1758 
1759 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1760 				  init_hca->altc_base,
1761 				  dev_cap->altc_entry_sz,
1762 				  dev->caps.num_qps,
1763 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1764 				  0, 0);
1765 	if (err) {
1766 		mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1767 		goto err_unmap_auxc;
1768 	}
1769 
1770 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1771 				  init_hca->rdmarc_base,
1772 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1773 				  dev->caps.num_qps,
1774 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1775 				  0, 0);
1776 	if (err) {
1777 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1778 		goto err_unmap_altc;
1779 	}
1780 
1781 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1782 				  init_hca->cqc_base,
1783 				  dev_cap->cqc_entry_sz,
1784 				  dev->caps.num_cqs,
1785 				  dev->caps.reserved_cqs, 0, 0);
1786 	if (err) {
1787 		mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1788 		goto err_unmap_rdmarc;
1789 	}
1790 
1791 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1792 				  init_hca->srqc_base,
1793 				  dev_cap->srq_entry_sz,
1794 				  dev->caps.num_srqs,
1795 				  dev->caps.reserved_srqs, 0, 0);
1796 	if (err) {
1797 		mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1798 		goto err_unmap_cq;
1799 	}
1800 
1801 	/*
1802 	 * For flow steering device managed mode it is required to use
1803 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1804 	 * required, but for simplicity just map the whole multicast
1805 	 * group table now.  The table isn't very big and it's a lot
1806 	 * easier than trying to track ref counts.
1807 	 */
1808 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1809 				  init_hca->mc_base,
1810 				  mlx4_get_mgm_entry_size(dev),
1811 				  dev->caps.num_mgms + dev->caps.num_amgms,
1812 				  dev->caps.num_mgms + dev->caps.num_amgms,
1813 				  0, 0);
1814 	if (err) {
1815 		mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1816 		goto err_unmap_srq;
1817 	}
1818 
1819 	return 0;
1820 
1821 err_unmap_srq:
1822 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1823 
1824 err_unmap_cq:
1825 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1826 
1827 err_unmap_rdmarc:
1828 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1829 
1830 err_unmap_altc:
1831 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1832 
1833 err_unmap_auxc:
1834 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1835 
1836 err_unmap_qp:
1837 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1838 
1839 err_unmap_dmpt:
1840 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1841 
1842 err_unmap_mtt:
1843 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1844 
1845 err_unmap_eq:
1846 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1847 
1848 err_unmap_cmpt:
1849 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1850 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1851 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1852 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1853 
1854 err_unmap_aux:
1855 	mlx4_UNMAP_ICM_AUX(dev);
1856 
1857 err_free_aux:
1858 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1859 
1860 	return err;
1861 }
1862 
1863 static void mlx4_free_icms(struct mlx4_dev *dev)
1864 {
1865 	struct mlx4_priv *priv = mlx4_priv(dev);
1866 
1867 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1868 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1869 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1870 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1871 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1872 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1873 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1874 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1875 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1876 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1877 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1878 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1879 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1880 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1881 
1882 	mlx4_UNMAP_ICM_AUX(dev);
1883 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1884 }
1885 
1886 static void mlx4_slave_exit(struct mlx4_dev *dev)
1887 {
1888 	struct mlx4_priv *priv = mlx4_priv(dev);
1889 
1890 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1891 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1892 			  MLX4_COMM_TIME))
1893 		mlx4_warn(dev, "Failed to close slave function\n");
1894 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1895 }
1896 
1897 static int map_bf_area(struct mlx4_dev *dev)
1898 {
1899 	struct mlx4_priv *priv = mlx4_priv(dev);
1900 	resource_size_t bf_start;
1901 	resource_size_t bf_len;
1902 	int err = 0;
1903 
1904 	if (!dev->caps.bf_reg_size)
1905 		return -ENXIO;
1906 
1907 	bf_start = pci_resource_start(dev->persist->pdev, 2) +
1908 			(dev->caps.num_uars << PAGE_SHIFT);
1909 	bf_len = pci_resource_len(dev->persist->pdev, 2) -
1910 			(dev->caps.num_uars << PAGE_SHIFT);
1911 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1912 	if (!priv->bf_mapping)
1913 		err = -ENOMEM;
1914 
1915 	return err;
1916 }
1917 
1918 static void unmap_bf_area(struct mlx4_dev *dev)
1919 {
1920 	if (mlx4_priv(dev)->bf_mapping)
1921 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1922 }
1923 
1924 u64 mlx4_read_clock(struct mlx4_dev *dev)
1925 {
1926 	u32 clockhi, clocklo, clockhi1;
1927 	u64 cycles;
1928 	int i;
1929 	struct mlx4_priv *priv = mlx4_priv(dev);
1930 
1931 	for (i = 0; i < 10; i++) {
1932 		clockhi = swab32(readl(priv->clock_mapping));
1933 		clocklo = swab32(readl(priv->clock_mapping + 4));
1934 		clockhi1 = swab32(readl(priv->clock_mapping));
1935 		if (clockhi == clockhi1)
1936 			break;
1937 	}
1938 
1939 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1940 
1941 	return cycles;
1942 }
1943 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1944 
1945 
1946 static int map_internal_clock(struct mlx4_dev *dev)
1947 {
1948 	struct mlx4_priv *priv = mlx4_priv(dev);
1949 
1950 	priv->clock_mapping =
1951 		ioremap(pci_resource_start(dev->persist->pdev,
1952 					   priv->fw.clock_bar) +
1953 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1954 
1955 	if (!priv->clock_mapping)
1956 		return -ENOMEM;
1957 
1958 	return 0;
1959 }
1960 
1961 int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1962 				   struct mlx4_clock_params *params)
1963 {
1964 	struct mlx4_priv *priv = mlx4_priv(dev);
1965 
1966 	if (mlx4_is_slave(dev))
1967 		return -EOPNOTSUPP;
1968 
1969 	if (!dev->caps.map_clock_to_user) {
1970 		mlx4_dbg(dev, "Map clock to user is not supported.\n");
1971 		return -EOPNOTSUPP;
1972 	}
1973 
1974 	if (!params)
1975 		return -EINVAL;
1976 
1977 	params->bar = priv->fw.clock_bar;
1978 	params->offset = priv->fw.clock_offset;
1979 	params->size = MLX4_CLOCK_SIZE;
1980 
1981 	return 0;
1982 }
1983 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1984 
1985 static void unmap_internal_clock(struct mlx4_dev *dev)
1986 {
1987 	struct mlx4_priv *priv = mlx4_priv(dev);
1988 
1989 	if (priv->clock_mapping)
1990 		iounmap(priv->clock_mapping);
1991 }
1992 
1993 static void mlx4_close_hca(struct mlx4_dev *dev)
1994 {
1995 	unmap_internal_clock(dev);
1996 	unmap_bf_area(dev);
1997 	if (mlx4_is_slave(dev))
1998 		mlx4_slave_exit(dev);
1999 	else {
2000 		mlx4_CLOSE_HCA(dev, 0);
2001 		mlx4_free_icms(dev);
2002 	}
2003 }
2004 
2005 static void mlx4_close_fw(struct mlx4_dev *dev)
2006 {
2007 	if (!mlx4_is_slave(dev)) {
2008 		mlx4_UNMAP_FA(dev);
2009 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
2010 	}
2011 }
2012 
2013 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
2014 {
2015 #define COMM_CHAN_OFFLINE_OFFSET 0x09
2016 
2017 	u32 comm_flags;
2018 	u32 offline_bit;
2019 	unsigned long end;
2020 	struct mlx4_priv *priv = mlx4_priv(dev);
2021 
2022 	end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
2023 	while (time_before(jiffies, end)) {
2024 		comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
2025 					  MLX4_COMM_CHAN_FLAGS));
2026 		offline_bit = (comm_flags &
2027 			       (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
2028 		if (!offline_bit)
2029 			return 0;
2030 
2031 		/* If device removal has been requested,
2032 		 * do not continue retrying.
2033 		 */
2034 		if (dev->persist->interface_state &
2035 		    MLX4_INTERFACE_STATE_NOWAIT)
2036 			break;
2037 
2038 		/* There are cases as part of AER/Reset flow that PF needs
2039 		 * around 100 msec to load. We therefore sleep for 100 msec
2040 		 * to allow other tasks to make use of that CPU during this
2041 		 * time interval.
2042 		 */
2043 		msleep(100);
2044 	}
2045 	mlx4_err(dev, "Communication channel is offline.\n");
2046 	return -EIO;
2047 }
2048 
2049 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
2050 {
2051 #define COMM_CHAN_RST_OFFSET 0x1e
2052 
2053 	struct mlx4_priv *priv = mlx4_priv(dev);
2054 	u32 comm_rst;
2055 	u32 comm_caps;
2056 
2057 	comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
2058 				 MLX4_COMM_CHAN_CAPS));
2059 	comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
2060 
2061 	if (comm_rst)
2062 		dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
2063 }
2064 
2065 static int mlx4_init_slave(struct mlx4_dev *dev)
2066 {
2067 	struct mlx4_priv *priv = mlx4_priv(dev);
2068 	u64 dma = (u64) priv->mfunc.vhcr_dma;
2069 	int ret_from_reset = 0;
2070 	u32 slave_read;
2071 	u32 cmd_channel_ver;
2072 
2073 	if (atomic_read(&pf_loading)) {
2074 		mlx4_warn(dev, "PF is not ready - Deferring probe\n");
2075 		return -EPROBE_DEFER;
2076 	}
2077 
2078 	mutex_lock(&priv->cmd.slave_cmd_mutex);
2079 	priv->cmd.max_cmds = 1;
2080 	if (mlx4_comm_check_offline(dev)) {
2081 		mlx4_err(dev, "PF is not responsive, skipping initialization\n");
2082 		goto err_offline;
2083 	}
2084 
2085 	mlx4_reset_vf_support(dev);
2086 	mlx4_warn(dev, "Sending reset\n");
2087 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
2088 				       MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
2089 	/* if we are in the middle of flr the slave will try
2090 	 * NUM_OF_RESET_RETRIES times before leaving.*/
2091 	if (ret_from_reset) {
2092 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
2093 			mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
2094 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
2095 			return -EPROBE_DEFER;
2096 		} else
2097 			goto err;
2098 	}
2099 
2100 	/* check the driver version - the slave I/F revision
2101 	 * must match the master's */
2102 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
2103 	cmd_channel_ver = mlx4_comm_get_version();
2104 
2105 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
2106 		MLX4_COMM_GET_IF_REV(slave_read)) {
2107 		mlx4_err(dev, "slave driver version is not supported by the master\n");
2108 		goto err;
2109 	}
2110 
2111 	mlx4_warn(dev, "Sending vhcr0\n");
2112 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
2113 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2114 		goto err;
2115 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
2116 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2117 		goto err;
2118 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
2119 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2120 		goto err;
2121 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
2122 			  MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2123 		goto err;
2124 
2125 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2126 	return 0;
2127 
2128 err:
2129 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
2130 err_offline:
2131 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2132 	return -EIO;
2133 }
2134 
2135 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
2136 {
2137 	int i;
2138 
2139 	for (i = 1; i <= dev->caps.num_ports; i++) {
2140 		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
2141 			dev->caps.gid_table_len[i] =
2142 				mlx4_get_slave_num_gids(dev, 0, i);
2143 		else
2144 			dev->caps.gid_table_len[i] = 1;
2145 		dev->caps.pkey_table_len[i] =
2146 			dev->phys_caps.pkey_phys_table_len[i] - 1;
2147 	}
2148 }
2149 
2150 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
2151 {
2152 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
2153 
2154 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
2155 	      i++) {
2156 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
2157 			break;
2158 	}
2159 
2160 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
2161 }
2162 
2163 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
2164 {
2165 	switch (dmfs_high_steer_mode) {
2166 	case MLX4_STEERING_DMFS_A0_DEFAULT:
2167 		return "default performance";
2168 
2169 	case MLX4_STEERING_DMFS_A0_DYNAMIC:
2170 		return "dynamic hybrid mode";
2171 
2172 	case MLX4_STEERING_DMFS_A0_STATIC:
2173 		return "performance optimized for limited rule configuration (static)";
2174 
2175 	case MLX4_STEERING_DMFS_A0_DISABLE:
2176 		return "disabled performance optimized steering";
2177 
2178 	case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
2179 		return "performance optimized steering not supported";
2180 
2181 	default:
2182 		return "Unrecognized mode";
2183 	}
2184 }
2185 
2186 #define MLX4_DMFS_A0_STEERING			(1UL << 2)
2187 
2188 static void choose_steering_mode(struct mlx4_dev *dev,
2189 				 struct mlx4_dev_cap *dev_cap)
2190 {
2191 	if (mlx4_log_num_mgm_entry_size <= 0) {
2192 		if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
2193 			if (dev->caps.dmfs_high_steer_mode ==
2194 			    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2195 				mlx4_err(dev, "DMFS high rate mode not supported\n");
2196 			else
2197 				dev->caps.dmfs_high_steer_mode =
2198 					MLX4_STEERING_DMFS_A0_STATIC;
2199 		}
2200 	}
2201 
2202 	if (mlx4_log_num_mgm_entry_size <= 0 &&
2203 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
2204 	    (!mlx4_is_mfunc(dev) ||
2205 	     (dev_cap->fs_max_num_qp_per_entry >=
2206 	     (dev->persist->num_vfs + 1))) &&
2207 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
2208 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
2209 		dev->oper_log_mgm_entry_size =
2210 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
2211 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
2212 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
2213 		dev->caps.fs_log_max_ucast_qp_range_size =
2214 			dev_cap->fs_log_max_ucast_qp_range_size;
2215 	} else {
2216 		if (dev->caps.dmfs_high_steer_mode !=
2217 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2218 			dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
2219 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
2220 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2221 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
2222 		else {
2223 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
2224 
2225 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
2226 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2227 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
2228 		}
2229 		dev->oper_log_mgm_entry_size =
2230 			mlx4_log_num_mgm_entry_size > 0 ?
2231 			mlx4_log_num_mgm_entry_size :
2232 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
2233 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
2234 	}
2235 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
2236 		 mlx4_steering_mode_str(dev->caps.steering_mode),
2237 		 dev->oper_log_mgm_entry_size,
2238 		 mlx4_log_num_mgm_entry_size);
2239 }
2240 
2241 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
2242 				       struct mlx4_dev_cap *dev_cap)
2243 {
2244 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2245 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
2246 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
2247 	else
2248 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
2249 
2250 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
2251 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
2252 }
2253 
2254 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
2255 {
2256 	int i;
2257 	struct mlx4_port_cap port_cap;
2258 
2259 	if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2260 		return -EINVAL;
2261 
2262 	for (i = 1; i <= dev->caps.num_ports; i++) {
2263 		if (mlx4_dev_port(dev, i, &port_cap)) {
2264 			mlx4_err(dev,
2265 				 "QUERY_DEV_CAP command failed, can't verify DMFS high rate steering.\n");
2266 		} else if ((dev->caps.dmfs_high_steer_mode !=
2267 			    MLX4_STEERING_DMFS_A0_DEFAULT) &&
2268 			   (port_cap.dmfs_optimized_state ==
2269 			    !!(dev->caps.dmfs_high_steer_mode ==
2270 			    MLX4_STEERING_DMFS_A0_DISABLE))) {
2271 			mlx4_err(dev,
2272 				 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2273 				 dmfs_high_rate_steering_mode_str(
2274 					dev->caps.dmfs_high_steer_mode),
2275 				 (port_cap.dmfs_optimized_state ?
2276 					"enabled" : "disabled"));
2277 		}
2278 	}
2279 
2280 	return 0;
2281 }
2282 
2283 static int mlx4_init_fw(struct mlx4_dev *dev)
2284 {
2285 	struct mlx4_mod_stat_cfg   mlx4_cfg;
2286 	int err = 0;
2287 
2288 	if (!mlx4_is_slave(dev)) {
2289 		err = mlx4_QUERY_FW(dev);
2290 		if (err) {
2291 			if (err == -EACCES)
2292 				mlx4_info(dev, "non-primary physical function, skipping\n");
2293 			else
2294 				mlx4_err(dev, "QUERY_FW command failed, aborting\n");
2295 			return err;
2296 		}
2297 
2298 		err = mlx4_load_fw(dev);
2299 		if (err) {
2300 			mlx4_err(dev, "Failed to start FW, aborting\n");
2301 			return err;
2302 		}
2303 
2304 		mlx4_cfg.log_pg_sz_m = 1;
2305 		mlx4_cfg.log_pg_sz = 0;
2306 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2307 		if (err)
2308 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2309 	}
2310 
2311 	return err;
2312 }
2313 
2314 static int mlx4_init_hca(struct mlx4_dev *dev)
2315 {
2316 	struct mlx4_priv	  *priv = mlx4_priv(dev);
2317 	struct mlx4_init_hca_param *init_hca = NULL;
2318 	struct mlx4_dev_cap	  *dev_cap = NULL;
2319 	struct mlx4_adapter	   adapter;
2320 	struct mlx4_profile	   profile;
2321 	u64 icm_size;
2322 	struct mlx4_config_dev_params params;
2323 	int err;
2324 
2325 	if (!mlx4_is_slave(dev)) {
2326 		dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
2327 		init_hca = kzalloc(sizeof(*init_hca), GFP_KERNEL);
2328 
2329 		if (!dev_cap || !init_hca) {
2330 			err = -ENOMEM;
2331 			goto out_free;
2332 		}
2333 
2334 		err = mlx4_dev_cap(dev, dev_cap);
2335 		if (err) {
2336 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
2337 			goto out_free;
2338 		}
2339 
2340 		choose_steering_mode(dev, dev_cap);
2341 		choose_tunnel_offload_mode(dev, dev_cap);
2342 
2343 		if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2344 		    mlx4_is_master(dev))
2345 			dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2346 
2347 		err = mlx4_get_phys_port_id(dev);
2348 		if (err)
2349 			mlx4_err(dev, "Fail to get physical port id\n");
2350 
2351 		if (mlx4_is_master(dev))
2352 			mlx4_parav_master_pf_caps(dev);
2353 
2354 		if (mlx4_low_memory_profile()) {
2355 			mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2356 			profile = low_mem_profile;
2357 		} else {
2358 			profile = default_profile;
2359 		}
2360 		if (dev->caps.steering_mode ==
2361 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
2362 			profile.num_mcg = MLX4_FS_NUM_MCG;
2363 
2364 		icm_size = mlx4_make_profile(dev, &profile, dev_cap,
2365 					     init_hca);
2366 		if ((long long) icm_size < 0) {
2367 			err = icm_size;
2368 			goto out_free;
2369 		}
2370 
2371 		if (enable_4k_uar || !dev->persist->num_vfs) {
2372 			init_hca->log_uar_sz = ilog2(dev->caps.num_uars) +
2373 						    PAGE_SHIFT - DEFAULT_UAR_PAGE_SHIFT;
2374 			init_hca->uar_page_sz = DEFAULT_UAR_PAGE_SHIFT - 12;
2375 		} else {
2376 			init_hca->log_uar_sz = ilog2(dev->caps.num_uars);
2377 			init_hca->uar_page_sz = PAGE_SHIFT - 12;
2378 		}
2379 
2380 		init_hca->mw_enabled = 0;
2381 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2382 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2383 			init_hca->mw_enabled = INIT_HCA_TPT_MW_ENABLE;
2384 
2385 		err = mlx4_init_icm(dev, dev_cap, init_hca, icm_size);
2386 		if (err)
2387 			goto out_free;
2388 
2389 		err = mlx4_INIT_HCA(dev, init_hca);
2390 		if (err) {
2391 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
2392 			goto err_free_icm;
2393 		}
2394 
2395 		if (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2396 			err = mlx4_query_func(dev, dev_cap);
2397 			if (err < 0) {
2398 				mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
2399 				goto err_close;
2400 			} else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2401 				dev->caps.num_eqs = dev_cap->max_eqs;
2402 				dev->caps.reserved_eqs = dev_cap->reserved_eqs;
2403 				dev->caps.reserved_uars = dev_cap->reserved_uars;
2404 			}
2405 		}
2406 
2407 		/*
2408 		 * If TS is supported by FW
2409 		 * read HCA frequency by QUERY_HCA command
2410 		 */
2411 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2412 			err = mlx4_QUERY_HCA(dev, init_hca);
2413 			if (err) {
2414 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
2415 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2416 			} else {
2417 				dev->caps.hca_core_clock =
2418 					init_hca->hca_core_clock;
2419 			}
2420 
2421 			/* In case we got HCA frequency 0 - disable timestamping
2422 			 * to avoid dividing by zero
2423 			 */
2424 			if (!dev->caps.hca_core_clock) {
2425 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2426 				mlx4_err(dev,
2427 					 "HCA frequency is 0 - timestamping is not supported\n");
2428 			} else if (map_internal_clock(dev)) {
2429 				/*
2430 				 * Map internal clock,
2431 				 * in case of failure disable timestamping
2432 				 */
2433 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2434 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
2435 			}
2436 		}
2437 
2438 		if (dev->caps.dmfs_high_steer_mode !=
2439 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2440 			if (mlx4_validate_optimized_steering(dev))
2441 				mlx4_warn(dev, "Optimized steering validation failed\n");
2442 
2443 			if (dev->caps.dmfs_high_steer_mode ==
2444 			    MLX4_STEERING_DMFS_A0_DISABLE) {
2445 				dev->caps.dmfs_high_rate_qpn_base =
2446 					dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2447 				dev->caps.dmfs_high_rate_qpn_range =
2448 					MLX4_A0_STEERING_TABLE_SIZE;
2449 			}
2450 
2451 			mlx4_info(dev, "DMFS high rate steer mode is: %s\n",
2452 				  dmfs_high_rate_steering_mode_str(
2453 					dev->caps.dmfs_high_steer_mode));
2454 		}
2455 	} else {
2456 		err = mlx4_init_slave(dev);
2457 		if (err) {
2458 			if (err != -EPROBE_DEFER)
2459 				mlx4_err(dev, "Failed to initialize slave\n");
2460 			return err;
2461 		}
2462 
2463 		err = mlx4_slave_cap(dev);
2464 		if (err) {
2465 			mlx4_err(dev, "Failed to obtain slave caps\n");
2466 			goto err_close;
2467 		}
2468 	}
2469 
2470 	if (map_bf_area(dev))
2471 		mlx4_dbg(dev, "Failed to map blue flame area\n");
2472 
2473 	/*Only the master set the ports, all the rest got it from it.*/
2474 	if (!mlx4_is_slave(dev))
2475 		mlx4_set_port_mask(dev);
2476 
2477 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
2478 	if (err) {
2479 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2480 		goto unmap_bf;
2481 	}
2482 
2483 	/* Query CONFIG_DEV parameters */
2484 	err = mlx4_config_dev_retrieval(dev, &params);
2485 	if (err && err != -EOPNOTSUPP) {
2486 		mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2487 	} else if (!err) {
2488 		dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2489 		dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2490 	}
2491 	priv->eq_table.inta_pin = adapter.inta_pin;
2492 	memcpy(dev->board_id, adapter.board_id, sizeof(dev->board_id));
2493 
2494 	err = 0;
2495 	goto out_free;
2496 
2497 unmap_bf:
2498 	unmap_internal_clock(dev);
2499 	unmap_bf_area(dev);
2500 
2501 	if (mlx4_is_slave(dev))
2502 		mlx4_slave_destroy_special_qp_cap(dev);
2503 
2504 err_close:
2505 	if (mlx4_is_slave(dev))
2506 		mlx4_slave_exit(dev);
2507 	else
2508 		mlx4_CLOSE_HCA(dev, 0);
2509 
2510 err_free_icm:
2511 	if (!mlx4_is_slave(dev))
2512 		mlx4_free_icms(dev);
2513 
2514 out_free:
2515 	kfree(dev_cap);
2516 	kfree(init_hca);
2517 
2518 	return err;
2519 }
2520 
2521 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2522 {
2523 	struct mlx4_priv *priv = mlx4_priv(dev);
2524 	int nent_pow2;
2525 
2526 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2527 		return -ENOENT;
2528 
2529 	if (!dev->caps.max_counters)
2530 		return -ENOSPC;
2531 
2532 	nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2533 	/* reserve last counter index for sink counter */
2534 	return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2535 				nent_pow2 - 1, 0,
2536 				nent_pow2 - dev->caps.max_counters + 1);
2537 }
2538 
2539 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2540 {
2541 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2542 		return;
2543 
2544 	if (!dev->caps.max_counters)
2545 		return;
2546 
2547 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2548 }
2549 
2550 static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2551 {
2552 	struct mlx4_priv *priv = mlx4_priv(dev);
2553 	int port;
2554 
2555 	for (port = 0; port < dev->caps.num_ports; port++)
2556 		if (priv->def_counter[port] != -1)
2557 			mlx4_counter_free(dev,  priv->def_counter[port]);
2558 }
2559 
2560 static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2561 {
2562 	struct mlx4_priv *priv = mlx4_priv(dev);
2563 	int port, err = 0;
2564 	u32 idx;
2565 
2566 	for (port = 0; port < dev->caps.num_ports; port++)
2567 		priv->def_counter[port] = -1;
2568 
2569 	for (port = 0; port < dev->caps.num_ports; port++) {
2570 		err = mlx4_counter_alloc(dev, &idx, MLX4_RES_USAGE_DRIVER);
2571 
2572 		if (!err || err == -ENOSPC) {
2573 			priv->def_counter[port] = idx;
2574 			err = 0;
2575 		} else if (err == -ENOENT) {
2576 			err = 0;
2577 			continue;
2578 		} else if (mlx4_is_slave(dev) && err == -EINVAL) {
2579 			priv->def_counter[port] = MLX4_SINK_COUNTER_INDEX(dev);
2580 			mlx4_warn(dev, "can't allocate counter from old PF driver, using index %d\n",
2581 				  MLX4_SINK_COUNTER_INDEX(dev));
2582 			err = 0;
2583 		} else {
2584 			mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2585 				 __func__, port + 1, err);
2586 			mlx4_cleanup_default_counters(dev);
2587 			return err;
2588 		}
2589 
2590 		mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2591 			 __func__, priv->def_counter[port], port + 1);
2592 	}
2593 
2594 	return err;
2595 }
2596 
2597 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2598 {
2599 	struct mlx4_priv *priv = mlx4_priv(dev);
2600 
2601 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2602 		return -ENOENT;
2603 
2604 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2605 	if (*idx == -1) {
2606 		*idx = MLX4_SINK_COUNTER_INDEX(dev);
2607 		return -ENOSPC;
2608 	}
2609 
2610 	return 0;
2611 }
2612 
2613 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx, u8 usage)
2614 {
2615 	u32 in_modifier = RES_COUNTER | (((u32)usage & 3) << 30);
2616 	u64 out_param;
2617 	int err;
2618 
2619 	if (mlx4_is_mfunc(dev)) {
2620 		err = mlx4_cmd_imm(dev, 0, &out_param, in_modifier,
2621 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2622 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2623 		if (!err)
2624 			*idx = get_param_l(&out_param);
2625 		if (WARN_ON(err == -ENOSPC))
2626 			err = -EINVAL;
2627 		return err;
2628 	}
2629 	return __mlx4_counter_alloc(dev, idx);
2630 }
2631 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2632 
2633 static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2634 				u8 counter_index)
2635 {
2636 	struct mlx4_cmd_mailbox *if_stat_mailbox;
2637 	int err;
2638 	u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2639 
2640 	if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2641 	if (IS_ERR(if_stat_mailbox))
2642 		return PTR_ERR(if_stat_mailbox);
2643 
2644 	err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2645 			   MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2646 			   MLX4_CMD_NATIVE);
2647 
2648 	mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2649 	return err;
2650 }
2651 
2652 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2653 {
2654 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2655 		return;
2656 
2657 	if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2658 		return;
2659 
2660 	__mlx4_clear_if_stat(dev, idx);
2661 
2662 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2663 	return;
2664 }
2665 
2666 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2667 {
2668 	u64 in_param = 0;
2669 
2670 	if (mlx4_is_mfunc(dev)) {
2671 		set_param_l(&in_param, idx);
2672 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2673 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2674 			 MLX4_CMD_WRAPPED);
2675 		return;
2676 	}
2677 	__mlx4_counter_free(dev, idx);
2678 }
2679 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2680 
2681 int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2682 {
2683 	struct mlx4_priv *priv = mlx4_priv(dev);
2684 
2685 	return priv->def_counter[port - 1];
2686 }
2687 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2688 
2689 void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2690 {
2691 	struct mlx4_priv *priv = mlx4_priv(dev);
2692 
2693 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2694 }
2695 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2696 
2697 __be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2698 {
2699 	struct mlx4_priv *priv = mlx4_priv(dev);
2700 
2701 	return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2702 }
2703 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2704 
2705 void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2706 {
2707 	struct mlx4_priv *priv = mlx4_priv(dev);
2708 	__be64 guid;
2709 
2710 	/* hw GUID */
2711 	if (entry == 0)
2712 		return;
2713 
2714 	get_random_bytes((char *)&guid, sizeof(guid));
2715 	guid &= ~(cpu_to_be64(1ULL << 56));
2716 	guid |= cpu_to_be64(1ULL << 57);
2717 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2718 }
2719 
2720 static int mlx4_setup_hca(struct mlx4_dev *dev)
2721 {
2722 	struct mlx4_priv *priv = mlx4_priv(dev);
2723 	int err;
2724 	int port;
2725 	__be32 ib_port_default_caps;
2726 
2727 	err = mlx4_init_uar_table(dev);
2728 	if (err) {
2729 		mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2730 		return err;
2731 	}
2732 
2733 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
2734 	if (err) {
2735 		mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2736 		goto err_uar_table_free;
2737 	}
2738 
2739 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2740 	if (!priv->kar) {
2741 		mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2742 		err = -ENOMEM;
2743 		goto err_uar_free;
2744 	}
2745 
2746 	err = mlx4_init_pd_table(dev);
2747 	if (err) {
2748 		mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2749 		goto err_kar_unmap;
2750 	}
2751 
2752 	err = mlx4_init_xrcd_table(dev);
2753 	if (err) {
2754 		mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2755 		goto err_pd_table_free;
2756 	}
2757 
2758 	err = mlx4_init_mr_table(dev);
2759 	if (err) {
2760 		mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2761 		goto err_xrcd_table_free;
2762 	}
2763 
2764 	if (!mlx4_is_slave(dev)) {
2765 		err = mlx4_init_mcg_table(dev);
2766 		if (err) {
2767 			mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2768 			goto err_mr_table_free;
2769 		}
2770 		err = mlx4_config_mad_demux(dev);
2771 		if (err) {
2772 			mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2773 			goto err_mcg_table_free;
2774 		}
2775 	}
2776 
2777 	err = mlx4_init_eq_table(dev);
2778 	if (err) {
2779 		mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2780 		goto err_mcg_table_free;
2781 	}
2782 
2783 	err = mlx4_cmd_use_events(dev);
2784 	if (err) {
2785 		mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2786 		goto err_eq_table_free;
2787 	}
2788 
2789 	err = mlx4_NOP(dev);
2790 	if (err) {
2791 		if (dev->flags & MLX4_FLAG_MSI_X) {
2792 			mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2793 				  priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2794 			mlx4_warn(dev, "Trying again without MSI-X\n");
2795 		} else {
2796 			mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2797 				 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2798 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2799 		}
2800 
2801 		goto err_cmd_poll;
2802 	}
2803 
2804 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
2805 
2806 	err = mlx4_init_cq_table(dev);
2807 	if (err) {
2808 		mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2809 		goto err_cmd_poll;
2810 	}
2811 
2812 	err = mlx4_init_srq_table(dev);
2813 	if (err) {
2814 		mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2815 		goto err_cq_table_free;
2816 	}
2817 
2818 	err = mlx4_init_qp_table(dev);
2819 	if (err) {
2820 		mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2821 		goto err_srq_table_free;
2822 	}
2823 
2824 	if (!mlx4_is_slave(dev)) {
2825 		err = mlx4_init_counters_table(dev);
2826 		if (err && err != -ENOENT) {
2827 			mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2828 			goto err_qp_table_free;
2829 		}
2830 	}
2831 
2832 	err = mlx4_allocate_default_counters(dev);
2833 	if (err) {
2834 		mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2835 		goto err_counters_table_free;
2836 	}
2837 
2838 	if (!mlx4_is_slave(dev)) {
2839 		for (port = 1; port <= dev->caps.num_ports; port++) {
2840 			ib_port_default_caps = 0;
2841 			err = mlx4_get_port_ib_caps(dev, port,
2842 						    &ib_port_default_caps);
2843 			if (err)
2844 				mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2845 					  port, err);
2846 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2847 
2848 			/* initialize per-slave default ib port capabilities */
2849 			if (mlx4_is_master(dev)) {
2850 				int i;
2851 				for (i = 0; i < dev->num_slaves; i++) {
2852 					if (i == mlx4_master_func_num(dev))
2853 						continue;
2854 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2855 						ib_port_default_caps;
2856 				}
2857 			}
2858 
2859 			if (mlx4_is_mfunc(dev))
2860 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2861 			else
2862 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2863 
2864 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2865 					    dev->caps.pkey_table_len[port] : -1);
2866 			if (err) {
2867 				mlx4_err(dev, "Failed to set port %d, aborting\n",
2868 					 port);
2869 				goto err_default_countes_free;
2870 			}
2871 		}
2872 	}
2873 
2874 	return 0;
2875 
2876 err_default_countes_free:
2877 	mlx4_cleanup_default_counters(dev);
2878 
2879 err_counters_table_free:
2880 	if (!mlx4_is_slave(dev))
2881 		mlx4_cleanup_counters_table(dev);
2882 
2883 err_qp_table_free:
2884 	mlx4_cleanup_qp_table(dev);
2885 
2886 err_srq_table_free:
2887 	mlx4_cleanup_srq_table(dev);
2888 
2889 err_cq_table_free:
2890 	mlx4_cleanup_cq_table(dev);
2891 
2892 err_cmd_poll:
2893 	mlx4_cmd_use_polling(dev);
2894 
2895 err_eq_table_free:
2896 	mlx4_cleanup_eq_table(dev);
2897 
2898 err_mcg_table_free:
2899 	if (!mlx4_is_slave(dev))
2900 		mlx4_cleanup_mcg_table(dev);
2901 
2902 err_mr_table_free:
2903 	mlx4_cleanup_mr_table(dev);
2904 
2905 err_xrcd_table_free:
2906 	mlx4_cleanup_xrcd_table(dev);
2907 
2908 err_pd_table_free:
2909 	mlx4_cleanup_pd_table(dev);
2910 
2911 err_kar_unmap:
2912 	iounmap(priv->kar);
2913 
2914 err_uar_free:
2915 	mlx4_uar_free(dev, &priv->driver_uar);
2916 
2917 err_uar_table_free:
2918 	mlx4_cleanup_uar_table(dev);
2919 	return err;
2920 }
2921 
2922 static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2923 {
2924 	int requested_cpu = 0;
2925 	struct mlx4_priv *priv = mlx4_priv(dev);
2926 	struct mlx4_eq *eq;
2927 	int off = 0;
2928 	int i;
2929 
2930 	if (eqn > dev->caps.num_comp_vectors)
2931 		return -EINVAL;
2932 
2933 	for (i = 1; i < port; i++)
2934 		off += mlx4_get_eqs_per_port(dev, i);
2935 
2936 	requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2937 
2938 	/* Meaning EQs are shared, and this call comes from the second port */
2939 	if (requested_cpu < 0)
2940 		return 0;
2941 
2942 	eq = &priv->eq_table.eq[eqn];
2943 
2944 	if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2945 		return -ENOMEM;
2946 
2947 	cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2948 
2949 	return 0;
2950 }
2951 
2952 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2953 {
2954 	struct mlx4_priv *priv = mlx4_priv(dev);
2955 	struct msix_entry *entries;
2956 	int i;
2957 	int port = 0;
2958 
2959 	if (msi_x) {
2960 		int nreq = min3(dev->caps.num_ports *
2961 				(int)num_online_cpus() + 1,
2962 				dev->caps.num_eqs - dev->caps.reserved_eqs,
2963 				MAX_MSIX);
2964 
2965 		if (msi_x > 1)
2966 			nreq = min_t(int, nreq, msi_x);
2967 
2968 		entries = kcalloc(nreq, sizeof(*entries), GFP_KERNEL);
2969 		if (!entries)
2970 			goto no_msi;
2971 
2972 		for (i = 0; i < nreq; ++i)
2973 			entries[i].entry = i;
2974 
2975 		nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2976 					     nreq);
2977 
2978 		if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
2979 			kfree(entries);
2980 			goto no_msi;
2981 		}
2982 		/* 1 is reserved for events (asyncrounous EQ) */
2983 		dev->caps.num_comp_vectors = nreq - 1;
2984 
2985 		priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2986 		bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2987 			    dev->caps.num_ports);
2988 
2989 		for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2990 			if (i == MLX4_EQ_ASYNC)
2991 				continue;
2992 
2993 			priv->eq_table.eq[i].irq =
2994 				entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2995 
2996 			if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2997 				bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2998 					    dev->caps.num_ports);
2999 				/* We don't set affinity hint when there
3000 				 * aren't enough EQs
3001 				 */
3002 			} else {
3003 				set_bit(port,
3004 					priv->eq_table.eq[i].actv_ports.ports);
3005 				if (mlx4_init_affinity_hint(dev, port + 1, i))
3006 					mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
3007 						  i);
3008 			}
3009 			/* We divide the Eqs evenly between the two ports.
3010 			 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
3011 			 * refers to the number of Eqs per port
3012 			 * (i.e eqs_per_port). Theoretically, we would like to
3013 			 * write something like (i + 1) % eqs_per_port == 0.
3014 			 * However, since there's an asynchronous Eq, we have
3015 			 * to skip over it by comparing this condition to
3016 			 * !!((i + 1) > MLX4_EQ_ASYNC).
3017 			 */
3018 			if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
3019 			    ((i + 1) %
3020 			     (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
3021 			    !!((i + 1) > MLX4_EQ_ASYNC))
3022 				/* If dev->caps.num_comp_vectors < dev->caps.num_ports,
3023 				 * everything is shared anyway.
3024 				 */
3025 				port++;
3026 		}
3027 
3028 		dev->flags |= MLX4_FLAG_MSI_X;
3029 
3030 		kfree(entries);
3031 		return;
3032 	}
3033 
3034 no_msi:
3035 	dev->caps.num_comp_vectors = 1;
3036 
3037 	BUG_ON(MLX4_EQ_ASYNC >= 2);
3038 	for (i = 0; i < 2; ++i) {
3039 		priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
3040 		if (i != MLX4_EQ_ASYNC) {
3041 			bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
3042 				    dev->caps.num_ports);
3043 		}
3044 	}
3045 }
3046 
3047 static int mlx4_devlink_port_type_set(struct devlink_port *devlink_port,
3048 				      enum devlink_port_type port_type)
3049 {
3050 	struct mlx4_port_info *info = container_of(devlink_port,
3051 						   struct mlx4_port_info,
3052 						   devlink_port);
3053 	enum mlx4_port_type mlx4_port_type;
3054 
3055 	switch (port_type) {
3056 	case DEVLINK_PORT_TYPE_AUTO:
3057 		mlx4_port_type = MLX4_PORT_TYPE_AUTO;
3058 		break;
3059 	case DEVLINK_PORT_TYPE_ETH:
3060 		mlx4_port_type = MLX4_PORT_TYPE_ETH;
3061 		break;
3062 	case DEVLINK_PORT_TYPE_IB:
3063 		mlx4_port_type = MLX4_PORT_TYPE_IB;
3064 		break;
3065 	default:
3066 		return -EOPNOTSUPP;
3067 	}
3068 
3069 	return __set_port_type(info, mlx4_port_type);
3070 }
3071 
3072 static const struct devlink_port_ops mlx4_devlink_port_ops = {
3073 	.port_type_set = mlx4_devlink_port_type_set,
3074 };
3075 
3076 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
3077 {
3078 	struct devlink *devlink = priv_to_devlink(mlx4_priv(dev));
3079 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
3080 	int err;
3081 
3082 	err = devl_port_register_with_ops(devlink, &info->devlink_port, port,
3083 					  &mlx4_devlink_port_ops);
3084 	if (err)
3085 		return err;
3086 
3087 	/* Ethernet and IB drivers will normally set the port type,
3088 	 * but if they are not built set the type now to prevent
3089 	 * devlink_port_type_warn() from firing.
3090 	 */
3091 	if (!IS_ENABLED(CONFIG_MLX4_EN) &&
3092 	    dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
3093 		devlink_port_type_eth_set(&info->devlink_port);
3094 	else if (!IS_ENABLED(CONFIG_MLX4_INFINIBAND) &&
3095 		 dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
3096 		devlink_port_type_ib_set(&info->devlink_port, NULL);
3097 
3098 	info->dev = dev;
3099 	info->port = port;
3100 	if (!mlx4_is_slave(dev)) {
3101 		mlx4_init_mac_table(dev, &info->mac_table);
3102 		mlx4_init_vlan_table(dev, &info->vlan_table);
3103 		mlx4_init_roce_gid_table(dev, &info->gid_table);
3104 		info->base_qpn = mlx4_get_base_qpn(dev, port);
3105 	}
3106 
3107 	sprintf(info->dev_name, "mlx4_port%d", port);
3108 	info->port_attr.attr.name = info->dev_name;
3109 	if (mlx4_is_mfunc(dev)) {
3110 		info->port_attr.attr.mode = 0444;
3111 	} else {
3112 		info->port_attr.attr.mode = 0644;
3113 		info->port_attr.store     = set_port_type;
3114 	}
3115 	info->port_attr.show      = show_port_type;
3116 	sysfs_attr_init(&info->port_attr.attr);
3117 
3118 	err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
3119 	if (err) {
3120 		mlx4_err(dev, "Failed to create file for port %d\n", port);
3121 		devlink_port_type_clear(&info->devlink_port);
3122 		devl_port_unregister(&info->devlink_port);
3123 		info->port = -1;
3124 		return err;
3125 	}
3126 
3127 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
3128 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
3129 	if (mlx4_is_mfunc(dev)) {
3130 		info->port_mtu_attr.attr.mode = 0444;
3131 	} else {
3132 		info->port_mtu_attr.attr.mode = 0644;
3133 		info->port_mtu_attr.store     = set_port_ib_mtu;
3134 	}
3135 	info->port_mtu_attr.show      = show_port_ib_mtu;
3136 	sysfs_attr_init(&info->port_mtu_attr.attr);
3137 
3138 	err = device_create_file(&dev->persist->pdev->dev,
3139 				 &info->port_mtu_attr);
3140 	if (err) {
3141 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
3142 		device_remove_file(&info->dev->persist->pdev->dev,
3143 				   &info->port_attr);
3144 		devlink_port_type_clear(&info->devlink_port);
3145 		devl_port_unregister(&info->devlink_port);
3146 		info->port = -1;
3147 		return err;
3148 	}
3149 
3150 	return 0;
3151 }
3152 
3153 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
3154 {
3155 	if (info->port < 0)
3156 		return;
3157 
3158 	device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
3159 	device_remove_file(&info->dev->persist->pdev->dev,
3160 			   &info->port_mtu_attr);
3161 	devlink_port_type_clear(&info->devlink_port);
3162 	devl_port_unregister(&info->devlink_port);
3163 
3164 #ifdef CONFIG_RFS_ACCEL
3165 	free_irq_cpu_rmap(info->rmap);
3166 	info->rmap = NULL;
3167 #endif
3168 }
3169 
3170 static int mlx4_init_steering(struct mlx4_dev *dev)
3171 {
3172 	struct mlx4_priv *priv = mlx4_priv(dev);
3173 	int num_entries = dev->caps.num_ports;
3174 	int i, j;
3175 
3176 	priv->steer = kcalloc(num_entries, sizeof(struct mlx4_steer),
3177 			      GFP_KERNEL);
3178 	if (!priv->steer)
3179 		return -ENOMEM;
3180 
3181 	for (i = 0; i < num_entries; i++)
3182 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3183 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
3184 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
3185 		}
3186 	return 0;
3187 }
3188 
3189 static void mlx4_clear_steering(struct mlx4_dev *dev)
3190 {
3191 	struct mlx4_priv *priv = mlx4_priv(dev);
3192 	struct mlx4_steer_index *entry, *tmp_entry;
3193 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
3194 	int num_entries = dev->caps.num_ports;
3195 	int i, j;
3196 
3197 	for (i = 0; i < num_entries; i++) {
3198 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3199 			list_for_each_entry_safe(pqp, tmp_pqp,
3200 						 &priv->steer[i].promisc_qps[j],
3201 						 list) {
3202 				list_del(&pqp->list);
3203 				kfree(pqp);
3204 			}
3205 			list_for_each_entry_safe(entry, tmp_entry,
3206 						 &priv->steer[i].steer_entries[j],
3207 						 list) {
3208 				list_del(&entry->list);
3209 				list_for_each_entry_safe(pqp, tmp_pqp,
3210 							 &entry->duplicates,
3211 							 list) {
3212 					list_del(&pqp->list);
3213 					kfree(pqp);
3214 				}
3215 				kfree(entry);
3216 			}
3217 		}
3218 	}
3219 	kfree(priv->steer);
3220 }
3221 
3222 static int extended_func_num(struct pci_dev *pdev)
3223 {
3224 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
3225 }
3226 
3227 #define MLX4_OWNER_BASE	0x8069c
3228 #define MLX4_OWNER_SIZE	4
3229 
3230 static int mlx4_get_ownership(struct mlx4_dev *dev)
3231 {
3232 	void __iomem *owner;
3233 	u32 ret;
3234 
3235 	if (pci_channel_offline(dev->persist->pdev))
3236 		return -EIO;
3237 
3238 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3239 			MLX4_OWNER_BASE,
3240 			MLX4_OWNER_SIZE);
3241 	if (!owner) {
3242 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3243 		return -ENOMEM;
3244 	}
3245 
3246 	ret = readl(owner);
3247 	iounmap(owner);
3248 	return (int) !!ret;
3249 }
3250 
3251 static void mlx4_free_ownership(struct mlx4_dev *dev)
3252 {
3253 	void __iomem *owner;
3254 
3255 	if (pci_channel_offline(dev->persist->pdev))
3256 		return;
3257 
3258 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3259 			MLX4_OWNER_BASE,
3260 			MLX4_OWNER_SIZE);
3261 	if (!owner) {
3262 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3263 		return;
3264 	}
3265 	writel(0, owner);
3266 	msleep(1000);
3267 	iounmap(owner);
3268 }
3269 
3270 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV)	==\
3271 				  !!((flags) & MLX4_FLAG_MASTER))
3272 
3273 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
3274 			     u8 total_vfs, int existing_vfs, int reset_flow)
3275 {
3276 	u64 dev_flags = dev->flags;
3277 	int err = 0;
3278 	int fw_enabled_sriov_vfs = min(pci_sriov_get_totalvfs(pdev),
3279 					MLX4_MAX_NUM_VF);
3280 
3281 	if (reset_flow) {
3282 		dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
3283 				       GFP_KERNEL);
3284 		if (!dev->dev_vfs)
3285 			goto free_mem;
3286 		return dev_flags;
3287 	}
3288 
3289 	atomic_inc(&pf_loading);
3290 	if (dev->flags &  MLX4_FLAG_SRIOV) {
3291 		if (existing_vfs != total_vfs) {
3292 			mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
3293 				 existing_vfs, total_vfs);
3294 			total_vfs = existing_vfs;
3295 		}
3296 	}
3297 
3298 	dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs), GFP_KERNEL);
3299 	if (NULL == dev->dev_vfs) {
3300 		mlx4_err(dev, "Failed to allocate memory for VFs\n");
3301 		goto disable_sriov;
3302 	}
3303 
3304 	if (!(dev->flags &  MLX4_FLAG_SRIOV)) {
3305 		if (total_vfs > fw_enabled_sriov_vfs) {
3306 			mlx4_err(dev, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
3307 				 total_vfs, fw_enabled_sriov_vfs);
3308 			err = -ENOMEM;
3309 			goto disable_sriov;
3310 		}
3311 		mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
3312 		err = pci_enable_sriov(pdev, total_vfs);
3313 	}
3314 	if (err) {
3315 		mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
3316 			 err);
3317 		goto disable_sriov;
3318 	} else {
3319 		mlx4_warn(dev, "Running in master mode\n");
3320 		dev_flags |= MLX4_FLAG_SRIOV |
3321 			MLX4_FLAG_MASTER;
3322 		dev_flags &= ~MLX4_FLAG_SLAVE;
3323 		dev->persist->num_vfs = total_vfs;
3324 	}
3325 	return dev_flags;
3326 
3327 disable_sriov:
3328 	atomic_dec(&pf_loading);
3329 free_mem:
3330 	dev->persist->num_vfs = 0;
3331 	kfree(dev->dev_vfs);
3332 	dev->dev_vfs = NULL;
3333 	return dev_flags & ~MLX4_FLAG_MASTER;
3334 }
3335 
3336 enum {
3337 	MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
3338 };
3339 
3340 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
3341 			      int *nvfs)
3342 {
3343 	int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
3344 	/* Checking for 64 VFs as a limitation of CX2 */
3345 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
3346 	    requested_vfs >= 64) {
3347 		mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
3348 			 requested_vfs);
3349 		return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
3350 	}
3351 	return 0;
3352 }
3353 
3354 static int mlx4_pci_enable_device(struct mlx4_dev *dev)
3355 {
3356 	struct pci_dev *pdev = dev->persist->pdev;
3357 	int err = 0;
3358 
3359 	mutex_lock(&dev->persist->pci_status_mutex);
3360 	if (dev->persist->pci_status == MLX4_PCI_STATUS_DISABLED) {
3361 		err = pci_enable_device(pdev);
3362 		if (!err)
3363 			dev->persist->pci_status = MLX4_PCI_STATUS_ENABLED;
3364 	}
3365 	mutex_unlock(&dev->persist->pci_status_mutex);
3366 
3367 	return err;
3368 }
3369 
3370 static void mlx4_pci_disable_device(struct mlx4_dev *dev)
3371 {
3372 	struct pci_dev *pdev = dev->persist->pdev;
3373 
3374 	mutex_lock(&dev->persist->pci_status_mutex);
3375 	if (dev->persist->pci_status == MLX4_PCI_STATUS_ENABLED) {
3376 		pci_disable_device(pdev);
3377 		dev->persist->pci_status = MLX4_PCI_STATUS_DISABLED;
3378 	}
3379 	mutex_unlock(&dev->persist->pci_status_mutex);
3380 }
3381 
3382 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
3383 			 int total_vfs, int *nvfs, struct mlx4_priv *priv,
3384 			 int reset_flow)
3385 {
3386 	struct devlink *devlink = priv_to_devlink(priv);
3387 	struct mlx4_dev *dev;
3388 	unsigned sum = 0;
3389 	int err;
3390 	int port;
3391 	int i;
3392 	struct mlx4_dev_cap *dev_cap = NULL;
3393 	int existing_vfs = 0;
3394 
3395 	devl_assert_locked(devlink);
3396 	dev = &priv->dev;
3397 
3398 	err = mlx4_adev_init(dev);
3399 	if (err)
3400 		return err;
3401 
3402 	ATOMIC_INIT_NOTIFIER_HEAD(&priv->event_nh);
3403 
3404 	mutex_init(&priv->port_mutex);
3405 	mutex_init(&priv->bond_mutex);
3406 
3407 	INIT_LIST_HEAD(&priv->pgdir_list);
3408 	mutex_init(&priv->pgdir_mutex);
3409 	spin_lock_init(&priv->cmd.context_lock);
3410 
3411 	INIT_LIST_HEAD(&priv->bf_list);
3412 	mutex_init(&priv->bf_mutex);
3413 
3414 	dev->rev_id = pdev->revision;
3415 	dev->numa_node = dev_to_node(&pdev->dev);
3416 
3417 	/* Detect if this device is a virtual function */
3418 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3419 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3420 		dev->flags |= MLX4_FLAG_SLAVE;
3421 	} else {
3422 		/* We reset the device and enable SRIOV only for physical
3423 		 * devices.  Try to claim ownership on the device;
3424 		 * if already taken, skip -- do not allow multiple PFs */
3425 		err = mlx4_get_ownership(dev);
3426 		if (err) {
3427 			if (err < 0)
3428 				goto err_adev;
3429 			else {
3430 				mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
3431 				err = -EINVAL;
3432 				goto err_adev;
3433 			}
3434 		}
3435 
3436 		atomic_set(&priv->opreq_count, 0);
3437 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3438 
3439 		/*
3440 		 * Now reset the HCA before we touch the PCI capabilities or
3441 		 * attempt a firmware command, since a boot ROM may have left
3442 		 * the HCA in an undefined state.
3443 		 */
3444 		err = mlx4_reset(dev);
3445 		if (err) {
3446 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
3447 			goto err_sriov;
3448 		}
3449 
3450 		if (total_vfs) {
3451 			dev->flags = MLX4_FLAG_MASTER;
3452 			existing_vfs = pci_num_vf(pdev);
3453 			if (existing_vfs)
3454 				dev->flags |= MLX4_FLAG_SRIOV;
3455 			dev->persist->num_vfs = total_vfs;
3456 		}
3457 	}
3458 
3459 	/* on load remove any previous indication of internal error,
3460 	 * device is up.
3461 	 */
3462 	dev->persist->state = MLX4_DEVICE_STATE_UP;
3463 
3464 slave_start:
3465 	err = mlx4_cmd_init(dev);
3466 	if (err) {
3467 		mlx4_err(dev, "Failed to init command interface, aborting\n");
3468 		goto err_sriov;
3469 	}
3470 
3471 	/* In slave functions, the communication channel must be initialized
3472 	 * before posting commands. Also, init num_slaves before calling
3473 	 * mlx4_init_hca */
3474 	if (mlx4_is_mfunc(dev)) {
3475 		if (mlx4_is_master(dev)) {
3476 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
3477 
3478 		} else {
3479 			dev->num_slaves = 0;
3480 			err = mlx4_multi_func_init(dev);
3481 			if (err) {
3482 				mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
3483 				goto err_cmd;
3484 			}
3485 		}
3486 	}
3487 
3488 	err = mlx4_init_fw(dev);
3489 	if (err) {
3490 		mlx4_err(dev, "Failed to init fw, aborting.\n");
3491 		goto err_mfunc;
3492 	}
3493 
3494 	if (mlx4_is_master(dev)) {
3495 		/* when we hit the goto slave_start below, dev_cap already initialized */
3496 		if (!dev_cap) {
3497 			dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3498 
3499 			if (!dev_cap) {
3500 				err = -ENOMEM;
3501 				goto err_fw;
3502 			}
3503 
3504 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3505 			if (err) {
3506 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3507 				goto err_fw;
3508 			}
3509 
3510 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3511 				goto err_fw;
3512 
3513 			if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3514 				u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3515 								  total_vfs,
3516 								  existing_vfs,
3517 								  reset_flow);
3518 
3519 				mlx4_close_fw(dev);
3520 				mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3521 				dev->flags = dev_flags;
3522 				if (!SRIOV_VALID_STATE(dev->flags)) {
3523 					mlx4_err(dev, "Invalid SRIOV state\n");
3524 					goto err_sriov;
3525 				}
3526 				err = mlx4_reset(dev);
3527 				if (err) {
3528 					mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3529 					goto err_sriov;
3530 				}
3531 				goto slave_start;
3532 			}
3533 		} else {
3534 			/* Legacy mode FW requires SRIOV to be enabled before
3535 			 * doing QUERY_DEV_CAP, since max_eq's value is different if
3536 			 * SRIOV is enabled.
3537 			 */
3538 			memset(dev_cap, 0, sizeof(*dev_cap));
3539 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3540 			if (err) {
3541 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3542 				goto err_fw;
3543 			}
3544 
3545 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3546 				goto err_fw;
3547 		}
3548 	}
3549 
3550 	err = mlx4_init_hca(dev);
3551 	if (err) {
3552 		if (err == -EACCES) {
3553 			/* Not primary Physical function
3554 			 * Running in slave mode */
3555 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3556 			/* We're not a PF */
3557 			if (dev->flags & MLX4_FLAG_SRIOV) {
3558 				if (!existing_vfs)
3559 					pci_disable_sriov(pdev);
3560 				if (mlx4_is_master(dev) && !reset_flow)
3561 					atomic_dec(&pf_loading);
3562 				dev->flags &= ~MLX4_FLAG_SRIOV;
3563 			}
3564 			if (!mlx4_is_slave(dev))
3565 				mlx4_free_ownership(dev);
3566 			dev->flags |= MLX4_FLAG_SLAVE;
3567 			dev->flags &= ~MLX4_FLAG_MASTER;
3568 			goto slave_start;
3569 		} else
3570 			goto err_fw;
3571 	}
3572 
3573 	if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3574 		u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3575 						  existing_vfs, reset_flow);
3576 
3577 		if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3578 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3579 			dev->flags = dev_flags;
3580 			err = mlx4_cmd_init(dev);
3581 			if (err) {
3582 				/* Only VHCR is cleaned up, so could still
3583 				 * send FW commands
3584 				 */
3585 				mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3586 				goto err_close;
3587 			}
3588 		} else {
3589 			dev->flags = dev_flags;
3590 		}
3591 
3592 		if (!SRIOV_VALID_STATE(dev->flags)) {
3593 			mlx4_err(dev, "Invalid SRIOV state\n");
3594 			err = -EINVAL;
3595 			goto err_close;
3596 		}
3597 	}
3598 
3599 	/* check if the device is functioning at its maximum possible speed.
3600 	 * No return code for this call, just warn the user in case of PCI
3601 	 * express device capabilities are under-satisfied by the bus.
3602 	 */
3603 	if (!mlx4_is_slave(dev))
3604 		pcie_print_link_status(dev->persist->pdev);
3605 
3606 	/* In master functions, the communication channel must be initialized
3607 	 * after obtaining its address from fw */
3608 	if (mlx4_is_master(dev)) {
3609 		if (dev->caps.num_ports < 2 &&
3610 		    num_vfs_argc > 1) {
3611 			err = -EINVAL;
3612 			mlx4_err(dev,
3613 				 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3614 				 dev->caps.num_ports);
3615 			goto err_close;
3616 		}
3617 		memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
3618 
3619 		for (i = 0;
3620 		     i < sizeof(dev->persist->nvfs)/
3621 		     sizeof(dev->persist->nvfs[0]); i++) {
3622 			unsigned j;
3623 
3624 			for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
3625 				dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3626 				dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3627 					dev->caps.num_ports;
3628 			}
3629 		}
3630 
3631 		/* In master functions, the communication channel
3632 		 * must be initialized after obtaining its address from fw
3633 		 */
3634 		err = mlx4_multi_func_init(dev);
3635 		if (err) {
3636 			mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3637 			goto err_close;
3638 		}
3639 	}
3640 
3641 	err = mlx4_alloc_eq_table(dev);
3642 	if (err)
3643 		goto err_master_mfunc;
3644 
3645 	bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
3646 	mutex_init(&priv->msix_ctl.pool_lock);
3647 
3648 	mlx4_enable_msi_x(dev);
3649 	if ((mlx4_is_mfunc(dev)) &&
3650 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
3651 		err = -EOPNOTSUPP;
3652 		mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
3653 		goto err_free_eq;
3654 	}
3655 
3656 	if (!mlx4_is_slave(dev)) {
3657 		err = mlx4_init_steering(dev);
3658 		if (err)
3659 			goto err_disable_msix;
3660 	}
3661 
3662 	mlx4_init_quotas(dev);
3663 
3664 	err = mlx4_setup_hca(dev);
3665 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3666 	    !mlx4_is_mfunc(dev)) {
3667 		dev->flags &= ~MLX4_FLAG_MSI_X;
3668 		dev->caps.num_comp_vectors = 1;
3669 		pci_disable_msix(pdev);
3670 		err = mlx4_setup_hca(dev);
3671 	}
3672 
3673 	if (err)
3674 		goto err_steer;
3675 
3676 	/* When PF resources are ready arm its comm channel to enable
3677 	 * getting commands
3678 	 */
3679 	if (mlx4_is_master(dev)) {
3680 		err = mlx4_ARM_COMM_CHANNEL(dev);
3681 		if (err) {
3682 			mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3683 				 err);
3684 			goto err_steer;
3685 		}
3686 	}
3687 
3688 	for (port = 1; port <= dev->caps.num_ports; port++) {
3689 		err = mlx4_init_port_info(dev, port);
3690 		if (err)
3691 			goto err_port;
3692 	}
3693 
3694 	priv->v2p.port1 = 1;
3695 	priv->v2p.port2 = 2;
3696 
3697 	err = mlx4_register_device(dev);
3698 	if (err)
3699 		goto err_port;
3700 
3701 	mlx4_sense_init(dev);
3702 	mlx4_start_sense(dev);
3703 
3704 	priv->removed = 0;
3705 
3706 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3707 		atomic_dec(&pf_loading);
3708 
3709 	kfree(dev_cap);
3710 	return 0;
3711 
3712 err_port:
3713 	for (--port; port >= 1; --port)
3714 		mlx4_cleanup_port_info(&priv->port[port]);
3715 
3716 	mlx4_cleanup_default_counters(dev);
3717 	if (!mlx4_is_slave(dev))
3718 		mlx4_cleanup_counters_table(dev);
3719 	mlx4_cleanup_qp_table(dev);
3720 	mlx4_cleanup_srq_table(dev);
3721 	mlx4_cleanup_cq_table(dev);
3722 	mlx4_cmd_use_polling(dev);
3723 	mlx4_cleanup_eq_table(dev);
3724 	mlx4_cleanup_mcg_table(dev);
3725 	mlx4_cleanup_mr_table(dev);
3726 	mlx4_cleanup_xrcd_table(dev);
3727 	mlx4_cleanup_pd_table(dev);
3728 	mlx4_cleanup_uar_table(dev);
3729 
3730 err_steer:
3731 	if (!mlx4_is_slave(dev))
3732 		mlx4_clear_steering(dev);
3733 
3734 err_disable_msix:
3735 	if (dev->flags & MLX4_FLAG_MSI_X)
3736 		pci_disable_msix(pdev);
3737 
3738 err_free_eq:
3739 	mlx4_free_eq_table(dev);
3740 
3741 err_master_mfunc:
3742 	if (mlx4_is_master(dev)) {
3743 		mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
3744 		mlx4_multi_func_cleanup(dev);
3745 	}
3746 
3747 	if (mlx4_is_slave(dev))
3748 		mlx4_slave_destroy_special_qp_cap(dev);
3749 
3750 err_close:
3751 	mlx4_close_hca(dev);
3752 
3753 err_fw:
3754 	mlx4_close_fw(dev);
3755 
3756 err_mfunc:
3757 	if (mlx4_is_slave(dev))
3758 		mlx4_multi_func_cleanup(dev);
3759 
3760 err_cmd:
3761 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3762 
3763 err_sriov:
3764 	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3765 		pci_disable_sriov(pdev);
3766 		dev->flags &= ~MLX4_FLAG_SRIOV;
3767 	}
3768 
3769 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3770 		atomic_dec(&pf_loading);
3771 
3772 	kfree(priv->dev.dev_vfs);
3773 
3774 	if (!mlx4_is_slave(dev))
3775 		mlx4_free_ownership(dev);
3776 
3777 	kfree(dev_cap);
3778 
3779 err_adev:
3780 	mlx4_adev_cleanup(dev);
3781 	return err;
3782 }
3783 
3784 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3785 			   struct mlx4_priv *priv)
3786 {
3787 	int err;
3788 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3789 	int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3790 	const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3791 		{2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3792 	unsigned total_vfs = 0;
3793 	unsigned int i;
3794 
3795 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3796 
3797 	err = mlx4_pci_enable_device(&priv->dev);
3798 	if (err) {
3799 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3800 		return err;
3801 	}
3802 
3803 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3804 	 * per port, we must limit the number of VFs to 63 (since their are
3805 	 * 128 MACs)
3806 	 */
3807 	for (i = 0; i < ARRAY_SIZE(nvfs) && i < num_vfs_argc;
3808 	     total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3809 		nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3810 		if (nvfs[i] < 0) {
3811 			dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3812 			err = -EINVAL;
3813 			goto err_disable_pdev;
3814 		}
3815 	}
3816 	for (i = 0; i < ARRAY_SIZE(prb_vf) && i < probe_vfs_argc;
3817 	     i++) {
3818 		prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3819 		if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3820 			dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3821 			err = -EINVAL;
3822 			goto err_disable_pdev;
3823 		}
3824 	}
3825 	if (total_vfs > MLX4_MAX_NUM_VF) {
3826 		dev_err(&pdev->dev,
3827 			"Requested more VF's (%d) than allowed by hw (%d)\n",
3828 			total_vfs, MLX4_MAX_NUM_VF);
3829 		err = -EINVAL;
3830 		goto err_disable_pdev;
3831 	}
3832 
3833 	for (i = 0; i < MLX4_MAX_PORTS; i++) {
3834 		if (nvfs[i] + nvfs[2] > MLX4_MAX_NUM_VF_P_PORT) {
3835 			dev_err(&pdev->dev,
3836 				"Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3837 				nvfs[i] + nvfs[2], i + 1,
3838 				MLX4_MAX_NUM_VF_P_PORT);
3839 			err = -EINVAL;
3840 			goto err_disable_pdev;
3841 		}
3842 	}
3843 
3844 	/* Check for BARs. */
3845 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3846 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3847 		dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3848 			pci_dev_data, pci_resource_flags(pdev, 0));
3849 		err = -ENODEV;
3850 		goto err_disable_pdev;
3851 	}
3852 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3853 		dev_err(&pdev->dev, "Missing UAR, aborting\n");
3854 		err = -ENODEV;
3855 		goto err_disable_pdev;
3856 	}
3857 
3858 	err = pci_request_regions(pdev, DRV_NAME);
3859 	if (err) {
3860 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3861 		goto err_disable_pdev;
3862 	}
3863 
3864 	pci_set_master(pdev);
3865 
3866 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3867 	if (err) {
3868 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3869 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3870 		if (err) {
3871 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3872 			goto err_release_regions;
3873 		}
3874 	}
3875 
3876 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
3877 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3878 	/* Detect if this device is a virtual function */
3879 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3880 		/* When acting as pf, we normally skip vfs unless explicitly
3881 		 * requested to probe them.
3882 		 */
3883 		if (total_vfs) {
3884 			unsigned vfs_offset = 0;
3885 
3886 			for (i = 0; i < ARRAY_SIZE(nvfs) &&
3887 			     vfs_offset + nvfs[i] < extended_func_num(pdev);
3888 			     vfs_offset += nvfs[i], i++)
3889 				;
3890 			if (i == ARRAY_SIZE(nvfs)) {
3891 				err = -ENODEV;
3892 				goto err_release_regions;
3893 			}
3894 			if ((extended_func_num(pdev) - vfs_offset)
3895 			    > prb_vf[i]) {
3896 				dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3897 					 extended_func_num(pdev));
3898 				err = -ENODEV;
3899 				goto err_release_regions;
3900 			}
3901 		}
3902 	}
3903 
3904 	err = mlx4_crdump_init(&priv->dev);
3905 	if (err)
3906 		goto err_release_regions;
3907 
3908 	err = mlx4_catas_init(&priv->dev);
3909 	if (err)
3910 		goto err_crdump;
3911 
3912 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3913 	if (err)
3914 		goto err_catas;
3915 
3916 	return 0;
3917 
3918 err_catas:
3919 	mlx4_catas_end(&priv->dev);
3920 
3921 err_crdump:
3922 	mlx4_crdump_end(&priv->dev);
3923 
3924 err_release_regions:
3925 	pci_release_regions(pdev);
3926 
3927 err_disable_pdev:
3928 	mlx4_pci_disable_device(&priv->dev);
3929 	return err;
3930 }
3931 
3932 static void mlx4_devlink_param_load_driverinit_values(struct devlink *devlink)
3933 {
3934 	struct mlx4_priv *priv = devlink_priv(devlink);
3935 	struct mlx4_dev *dev = &priv->dev;
3936 	struct mlx4_fw_crdump *crdump = &dev->persist->crdump;
3937 	union devlink_param_value saved_value;
3938 	int err;
3939 
3940 	err = devl_param_driverinit_value_get(devlink,
3941 					      DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET,
3942 					      &saved_value);
3943 	if (!err && mlx4_internal_err_reset != saved_value.vbool) {
3944 		mlx4_internal_err_reset = saved_value.vbool;
3945 		/* Notify on value changed on runtime configuration mode */
3946 		devl_param_value_changed(devlink,
3947 					 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET);
3948 	}
3949 	err = devl_param_driverinit_value_get(devlink,
3950 					      DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
3951 					      &saved_value);
3952 	if (!err)
3953 		log_num_mac = order_base_2(saved_value.vu32);
3954 	err = devl_param_driverinit_value_get(devlink,
3955 					      MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
3956 					      &saved_value);
3957 	if (!err)
3958 		enable_64b_cqe_eqe = saved_value.vbool;
3959 	err = devl_param_driverinit_value_get(devlink,
3960 					      MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
3961 					      &saved_value);
3962 	if (!err)
3963 		enable_4k_uar = saved_value.vbool;
3964 	err = devl_param_driverinit_value_get(devlink,
3965 					      DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT,
3966 					      &saved_value);
3967 	if (!err && crdump->snapshot_enable != saved_value.vbool) {
3968 		crdump->snapshot_enable = saved_value.vbool;
3969 		devl_param_value_changed(devlink,
3970 					 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT);
3971 	}
3972 }
3973 
3974 static void mlx4_restart_one_down(struct pci_dev *pdev);
3975 static int mlx4_restart_one_up(struct pci_dev *pdev, bool reload,
3976 			       struct devlink *devlink);
3977 
3978 static int mlx4_devlink_reload_down(struct devlink *devlink, bool netns_change,
3979 				    enum devlink_reload_action action,
3980 				    enum devlink_reload_limit limit,
3981 				    struct netlink_ext_ack *extack)
3982 {
3983 	struct mlx4_priv *priv = devlink_priv(devlink);
3984 	struct mlx4_dev *dev = &priv->dev;
3985 	struct mlx4_dev_persistent *persist = dev->persist;
3986 
3987 	if (netns_change) {
3988 		NL_SET_ERR_MSG_MOD(extack, "Namespace change is not supported");
3989 		return -EOPNOTSUPP;
3990 	}
3991 	if (persist->num_vfs)
3992 		mlx4_warn(persist->dev, "Reload performed on PF, will cause reset on operating Virtual Functions\n");
3993 	mlx4_restart_one_down(persist->pdev);
3994 	return 0;
3995 }
3996 
3997 static int mlx4_devlink_reload_up(struct devlink *devlink, enum devlink_reload_action action,
3998 				  enum devlink_reload_limit limit, u32 *actions_performed,
3999 				  struct netlink_ext_ack *extack)
4000 {
4001 	struct mlx4_priv *priv = devlink_priv(devlink);
4002 	struct mlx4_dev *dev = &priv->dev;
4003 	struct mlx4_dev_persistent *persist = dev->persist;
4004 	int err;
4005 
4006 	*actions_performed = BIT(DEVLINK_RELOAD_ACTION_DRIVER_REINIT);
4007 	err = mlx4_restart_one_up(persist->pdev, true, devlink);
4008 	if (err)
4009 		mlx4_err(persist->dev, "mlx4_restart_one_up failed, ret=%d\n",
4010 			 err);
4011 
4012 	return err;
4013 }
4014 
4015 static const struct devlink_ops mlx4_devlink_ops = {
4016 	.reload_actions = BIT(DEVLINK_RELOAD_ACTION_DRIVER_REINIT),
4017 	.reload_down	= mlx4_devlink_reload_down,
4018 	.reload_up	= mlx4_devlink_reload_up,
4019 };
4020 
4021 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
4022 {
4023 	struct devlink *devlink;
4024 	struct mlx4_priv *priv;
4025 	struct mlx4_dev *dev;
4026 	int ret;
4027 
4028 	printk_once(KERN_INFO "%s", mlx4_version);
4029 
4030 	devlink = devlink_alloc(&mlx4_devlink_ops, sizeof(*priv), &pdev->dev);
4031 	if (!devlink)
4032 		return -ENOMEM;
4033 	devl_lock(devlink);
4034 	priv = devlink_priv(devlink);
4035 
4036 	dev       = &priv->dev;
4037 	dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
4038 	if (!dev->persist) {
4039 		ret = -ENOMEM;
4040 		goto err_devlink_free;
4041 	}
4042 	dev->persist->pdev = pdev;
4043 	dev->persist->dev = dev;
4044 	pci_set_drvdata(pdev, dev->persist);
4045 	priv->pci_dev_data = id->driver_data;
4046 	mutex_init(&dev->persist->device_state_mutex);
4047 	mutex_init(&dev->persist->interface_state_mutex);
4048 	mutex_init(&dev->persist->pci_status_mutex);
4049 
4050 	ret = devl_params_register(devlink, mlx4_devlink_params,
4051 				   ARRAY_SIZE(mlx4_devlink_params));
4052 	if (ret)
4053 		goto err_devlink_unregister;
4054 	mlx4_devlink_set_params_init_values(devlink);
4055 	ret =  __mlx4_init_one(pdev, id->driver_data, priv);
4056 	if (ret)
4057 		goto err_params_unregister;
4058 
4059 	pci_save_state(pdev);
4060 	devl_unlock(devlink);
4061 	devlink_register(devlink);
4062 	return 0;
4063 
4064 err_params_unregister:
4065 	devl_params_unregister(devlink, mlx4_devlink_params,
4066 			       ARRAY_SIZE(mlx4_devlink_params));
4067 err_devlink_unregister:
4068 	kfree(dev->persist);
4069 err_devlink_free:
4070 	devl_unlock(devlink);
4071 	devlink_free(devlink);
4072 	return ret;
4073 }
4074 
4075 static void mlx4_clean_dev(struct mlx4_dev *dev)
4076 {
4077 	struct mlx4_dev_persistent *persist = dev->persist;
4078 	struct mlx4_priv *priv = mlx4_priv(dev);
4079 	unsigned long	flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
4080 
4081 	memset(priv, 0, sizeof(*priv));
4082 	priv->dev.persist = persist;
4083 	priv->dev.flags = flags;
4084 }
4085 
4086 static void mlx4_unload_one(struct pci_dev *pdev)
4087 {
4088 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4089 	struct mlx4_dev  *dev  = persist->dev;
4090 	struct mlx4_priv *priv = mlx4_priv(dev);
4091 	int               pci_dev_data;
4092 	struct devlink *devlink;
4093 	int p, i;
4094 
4095 	devlink = priv_to_devlink(priv);
4096 	devl_assert_locked(devlink);
4097 	if (priv->removed)
4098 		return;
4099 
4100 	/* saving current ports type for further use */
4101 	for (i = 0; i < dev->caps.num_ports; i++) {
4102 		dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
4103 		dev->persist->curr_port_poss_type[i] = dev->caps.
4104 						       possible_type[i + 1];
4105 	}
4106 
4107 	pci_dev_data = priv->pci_dev_data;
4108 
4109 	mlx4_stop_sense(dev);
4110 	mlx4_unregister_device(dev);
4111 
4112 	for (p = 1; p <= dev->caps.num_ports; p++) {
4113 		mlx4_cleanup_port_info(&priv->port[p]);
4114 		mlx4_CLOSE_PORT(dev, p);
4115 	}
4116 
4117 	if (mlx4_is_master(dev))
4118 		mlx4_free_resource_tracker(dev,
4119 					   RES_TR_FREE_SLAVES_ONLY);
4120 
4121 	mlx4_cleanup_default_counters(dev);
4122 	if (!mlx4_is_slave(dev))
4123 		mlx4_cleanup_counters_table(dev);
4124 	mlx4_cleanup_qp_table(dev);
4125 	mlx4_cleanup_srq_table(dev);
4126 	mlx4_cleanup_cq_table(dev);
4127 	mlx4_cmd_use_polling(dev);
4128 	mlx4_cleanup_eq_table(dev);
4129 	mlx4_cleanup_mcg_table(dev);
4130 	mlx4_cleanup_mr_table(dev);
4131 	mlx4_cleanup_xrcd_table(dev);
4132 	mlx4_cleanup_pd_table(dev);
4133 
4134 	if (mlx4_is_master(dev))
4135 		mlx4_free_resource_tracker(dev,
4136 					   RES_TR_FREE_STRUCTS_ONLY);
4137 
4138 	iounmap(priv->kar);
4139 	mlx4_uar_free(dev, &priv->driver_uar);
4140 	mlx4_cleanup_uar_table(dev);
4141 	if (!mlx4_is_slave(dev))
4142 		mlx4_clear_steering(dev);
4143 	mlx4_free_eq_table(dev);
4144 	if (mlx4_is_master(dev))
4145 		mlx4_multi_func_cleanup(dev);
4146 	mlx4_close_hca(dev);
4147 	mlx4_close_fw(dev);
4148 	if (mlx4_is_slave(dev))
4149 		mlx4_multi_func_cleanup(dev);
4150 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
4151 
4152 	if (dev->flags & MLX4_FLAG_MSI_X)
4153 		pci_disable_msix(pdev);
4154 
4155 	if (!mlx4_is_slave(dev))
4156 		mlx4_free_ownership(dev);
4157 
4158 	mlx4_slave_destroy_special_qp_cap(dev);
4159 	kfree(dev->dev_vfs);
4160 
4161 	mlx4_adev_cleanup(dev);
4162 
4163 	mlx4_clean_dev(dev);
4164 	priv->pci_dev_data = pci_dev_data;
4165 	priv->removed = 1;
4166 }
4167 
4168 static void mlx4_remove_one(struct pci_dev *pdev)
4169 {
4170 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4171 	struct mlx4_dev  *dev  = persist->dev;
4172 	struct mlx4_priv *priv = mlx4_priv(dev);
4173 	struct devlink *devlink = priv_to_devlink(priv);
4174 	int active_vfs = 0;
4175 
4176 	devlink_unregister(devlink);
4177 
4178 	devl_lock(devlink);
4179 	if (mlx4_is_slave(dev))
4180 		persist->interface_state |= MLX4_INTERFACE_STATE_NOWAIT;
4181 
4182 	mutex_lock(&persist->interface_state_mutex);
4183 	persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
4184 	mutex_unlock(&persist->interface_state_mutex);
4185 
4186 	/* Disabling SR-IOV is not allowed while there are active vf's */
4187 	if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
4188 		active_vfs = mlx4_how_many_lives_vf(dev);
4189 		if (active_vfs) {
4190 			pr_warn("Removing PF when there are active VF's !!\n");
4191 			pr_warn("Will not disable SR-IOV.\n");
4192 		}
4193 	}
4194 
4195 	/* device marked to be under deletion running now without the lock
4196 	 * letting other tasks to be terminated
4197 	 */
4198 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4199 		mlx4_unload_one(pdev);
4200 	else
4201 		mlx4_info(dev, "%s: interface is down\n", __func__);
4202 	mlx4_catas_end(dev);
4203 	mlx4_crdump_end(dev);
4204 	if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
4205 		mlx4_warn(dev, "Disabling SR-IOV\n");
4206 		pci_disable_sriov(pdev);
4207 	}
4208 
4209 	pci_release_regions(pdev);
4210 	mlx4_pci_disable_device(dev);
4211 	devl_params_unregister(devlink, mlx4_devlink_params,
4212 			       ARRAY_SIZE(mlx4_devlink_params));
4213 	kfree(dev->persist);
4214 	devl_unlock(devlink);
4215 	devlink_free(devlink);
4216 }
4217 
4218 static int restore_current_port_types(struct mlx4_dev *dev,
4219 				      enum mlx4_port_type *types,
4220 				      enum mlx4_port_type *poss_types)
4221 {
4222 	struct mlx4_priv *priv = mlx4_priv(dev);
4223 	int err, i;
4224 
4225 	mlx4_stop_sense(dev);
4226 
4227 	mutex_lock(&priv->port_mutex);
4228 	for (i = 0; i < dev->caps.num_ports; i++)
4229 		dev->caps.possible_type[i + 1] = poss_types[i];
4230 	err = mlx4_change_port_types(dev, types);
4231 	mlx4_start_sense(dev);
4232 	mutex_unlock(&priv->port_mutex);
4233 
4234 	return err;
4235 }
4236 
4237 static void mlx4_restart_one_down(struct pci_dev *pdev)
4238 {
4239 	mlx4_unload_one(pdev);
4240 }
4241 
4242 static int mlx4_restart_one_up(struct pci_dev *pdev, bool reload,
4243 			       struct devlink *devlink)
4244 {
4245 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4246 	struct mlx4_dev	 *dev  = persist->dev;
4247 	struct mlx4_priv *priv = mlx4_priv(dev);
4248 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4249 	int pci_dev_data, err, total_vfs;
4250 
4251 	pci_dev_data = priv->pci_dev_data;
4252 	total_vfs = dev->persist->num_vfs;
4253 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4254 
4255 	if (reload)
4256 		mlx4_devlink_param_load_driverinit_values(devlink);
4257 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
4258 	if (err) {
4259 		mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
4260 			 __func__, pci_name(pdev), err);
4261 		return err;
4262 	}
4263 
4264 	err = restore_current_port_types(dev, dev->persist->curr_port_type,
4265 					 dev->persist->curr_port_poss_type);
4266 	if (err)
4267 		mlx4_err(dev, "could not restore original port types (%d)\n",
4268 			 err);
4269 
4270 	return err;
4271 }
4272 
4273 int mlx4_restart_one(struct pci_dev *pdev)
4274 {
4275 	mlx4_restart_one_down(pdev);
4276 	return mlx4_restart_one_up(pdev, false, NULL);
4277 }
4278 
4279 #define MLX_SP(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_FORCE_SENSE_PORT }
4280 #define MLX_VF(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_IS_VF }
4281 #define MLX_GN(id) { PCI_VDEVICE(MELLANOX, id), 0 }
4282 
4283 static const struct pci_device_id mlx4_pci_table[] = {
4284 #ifdef CONFIG_MLX4_CORE_GEN2
4285 	/* MT25408 "Hermon" */
4286 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_SDR),	/* SDR */
4287 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR),	/* DDR */
4288 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR),	/* QDR */
4289 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR_GEN2), /* DDR Gen2 */
4290 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR_GEN2),	/* QDR Gen2 */
4291 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN),	/* EN 10GigE */
4292 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN_GEN2),  /* EN 10GigE Gen2 */
4293 	/* MT25458 ConnectX EN 10GBASE-T */
4294 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN),
4295 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_T_GEN2),	/* Gen2 */
4296 	/* MT26468 ConnectX EN 10GigE PCIe Gen2*/
4297 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_GEN2),
4298 	/* MT26438 ConnectX EN 40GigE PCIe Gen2 5GT/s */
4299 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_5_GEN2),
4300 	/* MT26478 ConnectX2 40GigE PCIe Gen2 */
4301 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX2),
4302 	/* MT25400 Family [ConnectX-2] */
4303 	MLX_VF(0x1002),					/* Virtual Function */
4304 #endif /* CONFIG_MLX4_CORE_GEN2 */
4305 	/* MT27500 Family [ConnectX-3] */
4306 	MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3),
4307 	MLX_VF(0x1004),					/* Virtual Function */
4308 	MLX_GN(0x1005),					/* MT27510 Family */
4309 	MLX_GN(0x1006),					/* MT27511 Family */
4310 	MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3_PRO),	/* MT27520 Family */
4311 	MLX_GN(0x1008),					/* MT27521 Family */
4312 	MLX_GN(0x1009),					/* MT27530 Family */
4313 	MLX_GN(0x100a),					/* MT27531 Family */
4314 	MLX_GN(0x100b),					/* MT27540 Family */
4315 	MLX_GN(0x100c),					/* MT27541 Family */
4316 	MLX_GN(0x100d),					/* MT27550 Family */
4317 	MLX_GN(0x100e),					/* MT27551 Family */
4318 	MLX_GN(0x100f),					/* MT27560 Family */
4319 	MLX_GN(0x1010),					/* MT27561 Family */
4320 
4321 	/*
4322 	 * See the mellanox_check_broken_intx_masking() quirk when
4323 	 * adding devices
4324 	 */
4325 
4326 	{ 0, }
4327 };
4328 
4329 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
4330 
4331 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
4332 					      pci_channel_state_t state)
4333 {
4334 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4335 	struct mlx4_dev *dev = persist->dev;
4336 	struct devlink *devlink;
4337 
4338 	mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
4339 	mlx4_enter_error_state(persist);
4340 
4341 	devlink = priv_to_devlink(mlx4_priv(dev));
4342 	devl_lock(devlink);
4343 	mutex_lock(&persist->interface_state_mutex);
4344 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4345 		mlx4_unload_one(pdev);
4346 
4347 	mutex_unlock(&persist->interface_state_mutex);
4348 	devl_unlock(devlink);
4349 	if (state == pci_channel_io_perm_failure)
4350 		return PCI_ERS_RESULT_DISCONNECT;
4351 
4352 	mlx4_pci_disable_device(persist->dev);
4353 	return PCI_ERS_RESULT_NEED_RESET;
4354 }
4355 
4356 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
4357 {
4358 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4359 	struct mlx4_dev	 *dev  = persist->dev;
4360 	int err;
4361 
4362 	mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
4363 	err = mlx4_pci_enable_device(dev);
4364 	if (err) {
4365 		mlx4_err(dev, "Can not re-enable device, err=%d\n", err);
4366 		return PCI_ERS_RESULT_DISCONNECT;
4367 	}
4368 
4369 	pci_set_master(pdev);
4370 	pci_restore_state(pdev);
4371 	return PCI_ERS_RESULT_RECOVERED;
4372 }
4373 
4374 static void mlx4_pci_resume(struct pci_dev *pdev)
4375 {
4376 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4377 	struct mlx4_dev	 *dev  = persist->dev;
4378 	struct mlx4_priv *priv = mlx4_priv(dev);
4379 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4380 	struct devlink *devlink;
4381 	int total_vfs;
4382 	int err;
4383 
4384 	mlx4_err(dev, "%s was called\n", __func__);
4385 	total_vfs = dev->persist->num_vfs;
4386 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4387 
4388 	devlink = priv_to_devlink(priv);
4389 	devl_lock(devlink);
4390 	mutex_lock(&persist->interface_state_mutex);
4391 	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4392 		err = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
4393 				    priv, 1);
4394 		if (err) {
4395 			mlx4_err(dev, "%s: mlx4_load_one failed, err=%d\n",
4396 				 __func__,  err);
4397 			goto end;
4398 		}
4399 
4400 		err = restore_current_port_types(dev, dev->persist->
4401 						 curr_port_type, dev->persist->
4402 						 curr_port_poss_type);
4403 		if (err)
4404 			mlx4_err(dev, "could not restore original port types (%d)\n", err);
4405 	}
4406 end:
4407 	mutex_unlock(&persist->interface_state_mutex);
4408 	devl_unlock(devlink);
4409 }
4410 
4411 static void mlx4_shutdown(struct pci_dev *pdev)
4412 {
4413 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4414 	struct mlx4_dev *dev = persist->dev;
4415 	struct devlink *devlink;
4416 
4417 	mlx4_info(persist->dev, "mlx4_shutdown was called\n");
4418 	devlink = priv_to_devlink(mlx4_priv(dev));
4419 	devl_lock(devlink);
4420 	mutex_lock(&persist->interface_state_mutex);
4421 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4422 		mlx4_unload_one(pdev);
4423 	mutex_unlock(&persist->interface_state_mutex);
4424 	devl_unlock(devlink);
4425 	mlx4_pci_disable_device(dev);
4426 }
4427 
4428 static const struct pci_error_handlers mlx4_err_handler = {
4429 	.error_detected = mlx4_pci_err_detected,
4430 	.slot_reset     = mlx4_pci_slot_reset,
4431 	.resume		= mlx4_pci_resume,
4432 };
4433 
4434 static int __maybe_unused mlx4_suspend(struct device *dev_d)
4435 {
4436 	struct pci_dev *pdev = to_pci_dev(dev_d);
4437 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4438 	struct mlx4_dev	*dev = persist->dev;
4439 	struct devlink *devlink;
4440 
4441 	mlx4_err(dev, "suspend was called\n");
4442 	devlink = priv_to_devlink(mlx4_priv(dev));
4443 	devl_lock(devlink);
4444 	mutex_lock(&persist->interface_state_mutex);
4445 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4446 		mlx4_unload_one(pdev);
4447 	mutex_unlock(&persist->interface_state_mutex);
4448 	devl_unlock(devlink);
4449 
4450 	return 0;
4451 }
4452 
4453 static int __maybe_unused mlx4_resume(struct device *dev_d)
4454 {
4455 	struct pci_dev *pdev = to_pci_dev(dev_d);
4456 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4457 	struct mlx4_dev	*dev = persist->dev;
4458 	struct mlx4_priv *priv = mlx4_priv(dev);
4459 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4460 	struct devlink *devlink;
4461 	int total_vfs;
4462 	int ret = 0;
4463 
4464 	mlx4_err(dev, "resume was called\n");
4465 	total_vfs = dev->persist->num_vfs;
4466 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4467 
4468 	devlink = priv_to_devlink(priv);
4469 	devl_lock(devlink);
4470 	mutex_lock(&persist->interface_state_mutex);
4471 	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4472 		ret = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs,
4473 				    nvfs, priv, 1);
4474 		if (!ret) {
4475 			ret = restore_current_port_types(dev,
4476 					dev->persist->curr_port_type,
4477 					dev->persist->curr_port_poss_type);
4478 			if (ret)
4479 				mlx4_err(dev, "resume: could not restore original port types (%d)\n", ret);
4480 		}
4481 	}
4482 	mutex_unlock(&persist->interface_state_mutex);
4483 	devl_unlock(devlink);
4484 
4485 	return ret;
4486 }
4487 
4488 static SIMPLE_DEV_PM_OPS(mlx4_pm_ops, mlx4_suspend, mlx4_resume);
4489 
4490 static struct pci_driver mlx4_driver = {
4491 	.name		= DRV_NAME,
4492 	.id_table	= mlx4_pci_table,
4493 	.probe		= mlx4_init_one,
4494 	.shutdown	= mlx4_shutdown,
4495 	.remove		= mlx4_remove_one,
4496 	.driver.pm	= &mlx4_pm_ops,
4497 	.err_handler    = &mlx4_err_handler,
4498 };
4499 
4500 static int __init mlx4_verify_params(void)
4501 {
4502 	if (msi_x < 0) {
4503 		pr_warn("mlx4_core: bad msi_x: %d\n", msi_x);
4504 		return -1;
4505 	}
4506 
4507 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
4508 		pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
4509 		return -1;
4510 	}
4511 
4512 	if (log_num_vlan != 0)
4513 		pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
4514 			MLX4_LOG_NUM_VLANS);
4515 
4516 	if (use_prio != 0)
4517 		pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
4518 
4519 	if ((log_mtts_per_seg < 0) || (log_mtts_per_seg > 7)) {
4520 		pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
4521 			log_mtts_per_seg);
4522 		return -1;
4523 	}
4524 
4525 	/* Check if module param for ports type has legal combination */
4526 	if (port_type_array[0] == false && port_type_array[1] == true) {
4527 		pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
4528 		port_type_array[0] = true;
4529 	}
4530 
4531 	if (mlx4_log_num_mgm_entry_size < -7 ||
4532 	    (mlx4_log_num_mgm_entry_size > 0 &&
4533 	     (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
4534 	      mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
4535 		pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
4536 			mlx4_log_num_mgm_entry_size,
4537 			MLX4_MIN_MGM_LOG_ENTRY_SIZE,
4538 			MLX4_MAX_MGM_LOG_ENTRY_SIZE);
4539 		return -1;
4540 	}
4541 
4542 	return 0;
4543 }
4544 
4545 static int __init mlx4_init(void)
4546 {
4547 	int ret;
4548 
4549 	WARN_ONCE(strcmp(MLX4_ADEV_NAME, KBUILD_MODNAME),
4550 		  "mlx4_core name not in sync with kernel module name");
4551 
4552 	if (mlx4_verify_params())
4553 		return -EINVAL;
4554 
4555 
4556 	mlx4_wq = create_singlethread_workqueue("mlx4");
4557 	if (!mlx4_wq)
4558 		return -ENOMEM;
4559 
4560 	ret = pci_register_driver(&mlx4_driver);
4561 	if (ret < 0)
4562 		destroy_workqueue(mlx4_wq);
4563 	return ret < 0 ? ret : 0;
4564 }
4565 
4566 static void __exit mlx4_cleanup(void)
4567 {
4568 	pci_unregister_driver(&mlx4_driver);
4569 	destroy_workqueue(mlx4_wq);
4570 }
4571 
4572 module_init(mlx4_init);
4573 module_exit(mlx4_cleanup);
4574