1 /*
2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4  * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
5  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * OpenIB.org BSD license below:
12  *
13  *     Redistribution and use in source and binary forms, with or
14  *     without modification, are permitted provided that the following
15  *     conditions are met:
16  *
17  *      - Redistributions of source code must retain the above
18  *        copyright notice, this list of conditions and the following
19  *        disclaimer.
20  *
21  *      - Redistributions in binary form must reproduce the above
22  *        copyright notice, this list of conditions and the following
23  *        disclaimer in the documentation and/or other materials
24  *        provided with the distribution.
25  *
26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33  * SOFTWARE.
34  */
35 
36 #include <linux/module.h>
37 #include <linux/kernel.h>
38 #include <linux/init.h>
39 #include <linux/errno.h>
40 #include <linux/pci.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/slab.h>
43 #include <linux/io-mapping.h>
44 #include <linux/delay.h>
45 #include <linux/kmod.h>
46 #include <linux/etherdevice.h>
47 #include <net/devlink.h>
48 
49 #include <uapi/rdma/mlx4-abi.h>
50 #include <linux/mlx4/device.h>
51 #include <linux/mlx4/doorbell.h>
52 
53 #include "mlx4.h"
54 #include "fw.h"
55 #include "icm.h"
56 
57 MODULE_AUTHOR("Roland Dreier");
58 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
59 MODULE_LICENSE("Dual BSD/GPL");
60 MODULE_VERSION(DRV_VERSION);
61 
62 struct workqueue_struct *mlx4_wq;
63 
64 #ifdef CONFIG_MLX4_DEBUG
65 
66 int mlx4_debug_level; /* 0 by default */
67 module_param_named(debug_level, mlx4_debug_level, int, 0644);
68 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
69 
70 #endif /* CONFIG_MLX4_DEBUG */
71 
72 #ifdef CONFIG_PCI_MSI
73 
74 static int msi_x = 1;
75 module_param(msi_x, int, 0444);
76 MODULE_PARM_DESC(msi_x, "0 - don't use MSI-X, 1 - use MSI-X, >1 - limit number of MSI-X irqs to msi_x");
77 
78 #else /* CONFIG_PCI_MSI */
79 
80 #define msi_x (0)
81 
82 #endif /* CONFIG_PCI_MSI */
83 
84 static uint8_t num_vfs[3] = {0, 0, 0};
85 static int num_vfs_argc;
86 module_param_array(num_vfs, byte, &num_vfs_argc, 0444);
87 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
88 			  "num_vfs=port1,port2,port1+2");
89 
90 static uint8_t probe_vf[3] = {0, 0, 0};
91 static int probe_vfs_argc;
92 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
93 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
94 			   "probe_vf=port1,port2,port1+2");
95 
96 static int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
97 module_param_named(log_num_mgm_entry_size,
98 			mlx4_log_num_mgm_entry_size, int, 0444);
99 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
100 					 " of qp per mcg, for example:"
101 					 " 10 gives 248.range: 7 <="
102 					 " log_num_mgm_entry_size <= 12."
103 					 " To activate device managed"
104 					 " flow steering when available, set to -1");
105 
106 static bool enable_64b_cqe_eqe = true;
107 module_param(enable_64b_cqe_eqe, bool, 0444);
108 MODULE_PARM_DESC(enable_64b_cqe_eqe,
109 		 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
110 
111 static bool enable_4k_uar;
112 module_param(enable_4k_uar, bool, 0444);
113 MODULE_PARM_DESC(enable_4k_uar,
114 		 "Enable using 4K UAR. Should not be enabled if have VFs which do not support 4K UARs (default: false)");
115 
116 #define PF_CONTEXT_BEHAVIOUR_MASK	(MLX4_FUNC_CAP_64B_EQE_CQE | \
117 					 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
118 					 MLX4_FUNC_CAP_DMFS_A0_STATIC)
119 
120 #define RESET_PERSIST_MASK_FLAGS	(MLX4_FLAG_SRIOV)
121 
122 static char mlx4_version[] =
123 	DRV_NAME ": Mellanox ConnectX core driver v"
124 	DRV_VERSION "\n";
125 
126 static const struct mlx4_profile default_profile = {
127 	.num_qp		= 1 << 18,
128 	.num_srq	= 1 << 16,
129 	.rdmarc_per_qp	= 1 << 4,
130 	.num_cq		= 1 << 16,
131 	.num_mcg	= 1 << 13,
132 	.num_mpt	= 1 << 19,
133 	.num_mtt	= 1 << 20, /* It is really num mtt segements */
134 };
135 
136 static const struct mlx4_profile low_mem_profile = {
137 	.num_qp		= 1 << 17,
138 	.num_srq	= 1 << 6,
139 	.rdmarc_per_qp	= 1 << 4,
140 	.num_cq		= 1 << 8,
141 	.num_mcg	= 1 << 8,
142 	.num_mpt	= 1 << 9,
143 	.num_mtt	= 1 << 7,
144 };
145 
146 static int log_num_mac = 7;
147 module_param_named(log_num_mac, log_num_mac, int, 0444);
148 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
149 
150 static int log_num_vlan;
151 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
152 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
153 /* Log2 max number of VLANs per ETH port (0-7) */
154 #define MLX4_LOG_NUM_VLANS 7
155 #define MLX4_MIN_LOG_NUM_VLANS 0
156 #define MLX4_MIN_LOG_NUM_MAC 1
157 
158 static bool use_prio;
159 module_param_named(use_prio, use_prio, bool, 0444);
160 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
161 
162 int log_mtts_per_seg = ilog2(1);
163 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
164 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment "
165 		 "(0-7) (default: 0)");
166 
167 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
168 static int arr_argc = 2;
169 module_param_array(port_type_array, int, &arr_argc, 0444);
170 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
171 				"1 for IB, 2 for Ethernet");
172 
173 struct mlx4_port_config {
174 	struct list_head list;
175 	enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
176 	struct pci_dev *pdev;
177 };
178 
179 static atomic_t pf_loading = ATOMIC_INIT(0);
180 
181 static int mlx4_devlink_ierr_reset_get(struct devlink *devlink, u32 id,
182 				       struct devlink_param_gset_ctx *ctx)
183 {
184 	ctx->val.vbool = !!mlx4_internal_err_reset;
185 	return 0;
186 }
187 
188 static int mlx4_devlink_ierr_reset_set(struct devlink *devlink, u32 id,
189 				       struct devlink_param_gset_ctx *ctx)
190 {
191 	mlx4_internal_err_reset = ctx->val.vbool;
192 	return 0;
193 }
194 
195 static int mlx4_devlink_crdump_snapshot_get(struct devlink *devlink, u32 id,
196 					    struct devlink_param_gset_ctx *ctx)
197 {
198 	struct mlx4_priv *priv = devlink_priv(devlink);
199 	struct mlx4_dev *dev = &priv->dev;
200 
201 	ctx->val.vbool = dev->persist->crdump.snapshot_enable;
202 	return 0;
203 }
204 
205 static int mlx4_devlink_crdump_snapshot_set(struct devlink *devlink, u32 id,
206 					    struct devlink_param_gset_ctx *ctx)
207 {
208 	struct mlx4_priv *priv = devlink_priv(devlink);
209 	struct mlx4_dev *dev = &priv->dev;
210 
211 	dev->persist->crdump.snapshot_enable = ctx->val.vbool;
212 	return 0;
213 }
214 
215 static int
216 mlx4_devlink_max_macs_validate(struct devlink *devlink, u32 id,
217 			       union devlink_param_value val,
218 			       struct netlink_ext_ack *extack)
219 {
220 	u32 value = val.vu32;
221 
222 	if (value < 1 || value > 128)
223 		return -ERANGE;
224 
225 	if (!is_power_of_2(value)) {
226 		NL_SET_ERR_MSG_MOD(extack, "max_macs supported must be power of 2");
227 		return -EINVAL;
228 	}
229 
230 	return 0;
231 }
232 
233 enum mlx4_devlink_param_id {
234 	MLX4_DEVLINK_PARAM_ID_BASE = DEVLINK_PARAM_GENERIC_ID_MAX,
235 	MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
236 	MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
237 };
238 
239 static const struct devlink_param mlx4_devlink_params[] = {
240 	DEVLINK_PARAM_GENERIC(INT_ERR_RESET,
241 			      BIT(DEVLINK_PARAM_CMODE_RUNTIME) |
242 			      BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
243 			      mlx4_devlink_ierr_reset_get,
244 			      mlx4_devlink_ierr_reset_set, NULL),
245 	DEVLINK_PARAM_GENERIC(MAX_MACS,
246 			      BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
247 			      NULL, NULL, mlx4_devlink_max_macs_validate),
248 	DEVLINK_PARAM_GENERIC(REGION_SNAPSHOT,
249 			      BIT(DEVLINK_PARAM_CMODE_RUNTIME) |
250 			      BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
251 			      mlx4_devlink_crdump_snapshot_get,
252 			      mlx4_devlink_crdump_snapshot_set, NULL),
253 	DEVLINK_PARAM_DRIVER(MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
254 			     "enable_64b_cqe_eqe", DEVLINK_PARAM_TYPE_BOOL,
255 			     BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
256 			     NULL, NULL, NULL),
257 	DEVLINK_PARAM_DRIVER(MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
258 			     "enable_4k_uar", DEVLINK_PARAM_TYPE_BOOL,
259 			     BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
260 			     NULL, NULL, NULL),
261 };
262 
263 static void mlx4_devlink_set_params_init_values(struct devlink *devlink)
264 {
265 	union devlink_param_value value;
266 
267 	value.vbool = !!mlx4_internal_err_reset;
268 	devlink_param_driverinit_value_set(devlink,
269 					   DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET,
270 					   value);
271 
272 	value.vu32 = 1UL << log_num_mac;
273 	devlink_param_driverinit_value_set(devlink,
274 					   DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
275 					   value);
276 
277 	value.vbool = enable_64b_cqe_eqe;
278 	devlink_param_driverinit_value_set(devlink,
279 					   MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
280 					   value);
281 
282 	value.vbool = enable_4k_uar;
283 	devlink_param_driverinit_value_set(devlink,
284 					   MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
285 					   value);
286 
287 	value.vbool = false;
288 	devlink_param_driverinit_value_set(devlink,
289 					   DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT,
290 					   value);
291 }
292 
293 static inline void mlx4_set_num_reserved_uars(struct mlx4_dev *dev,
294 					      struct mlx4_dev_cap *dev_cap)
295 {
296 	/* The reserved_uars is calculated by system page size unit.
297 	 * Therefore, adjustment is added when the uar page size is less
298 	 * than the system page size
299 	 */
300 	dev->caps.reserved_uars	=
301 		max_t(int,
302 		      mlx4_get_num_reserved_uar(dev),
303 		      dev_cap->reserved_uars /
304 			(1 << (PAGE_SHIFT - dev->uar_page_shift)));
305 }
306 
307 int mlx4_check_port_params(struct mlx4_dev *dev,
308 			   enum mlx4_port_type *port_type)
309 {
310 	int i;
311 
312 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
313 		for (i = 0; i < dev->caps.num_ports - 1; i++) {
314 			if (port_type[i] != port_type[i + 1]) {
315 				mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
316 				return -EOPNOTSUPP;
317 			}
318 		}
319 	}
320 
321 	for (i = 0; i < dev->caps.num_ports; i++) {
322 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
323 			mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
324 				 i + 1);
325 			return -EOPNOTSUPP;
326 		}
327 	}
328 	return 0;
329 }
330 
331 static void mlx4_set_port_mask(struct mlx4_dev *dev)
332 {
333 	int i;
334 
335 	for (i = 1; i <= dev->caps.num_ports; ++i)
336 		dev->caps.port_mask[i] = dev->caps.port_type[i];
337 }
338 
339 enum {
340 	MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
341 };
342 
343 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
344 {
345 	int err = 0;
346 	struct mlx4_func func;
347 
348 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
349 		err = mlx4_QUERY_FUNC(dev, &func, 0);
350 		if (err) {
351 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
352 			return err;
353 		}
354 		dev_cap->max_eqs = func.max_eq;
355 		dev_cap->reserved_eqs = func.rsvd_eqs;
356 		dev_cap->reserved_uars = func.rsvd_uars;
357 		err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
358 	}
359 	return err;
360 }
361 
362 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
363 {
364 	struct mlx4_caps *dev_cap = &dev->caps;
365 
366 	/* FW not supporting or cancelled by user */
367 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
368 	    !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
369 		return;
370 
371 	/* Must have 64B CQE_EQE enabled by FW to use bigger stride
372 	 * When FW has NCSI it may decide not to report 64B CQE/EQEs
373 	 */
374 	if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
375 	    !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
376 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
377 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
378 		return;
379 	}
380 
381 	if (cache_line_size() == 128 || cache_line_size() == 256) {
382 		mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
383 		/* Changing the real data inside CQE size to 32B */
384 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
385 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
386 
387 		if (mlx4_is_master(dev))
388 			dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
389 	} else {
390 		if (cache_line_size() != 32  && cache_line_size() != 64)
391 			mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
392 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
393 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
394 	}
395 }
396 
397 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
398 			  struct mlx4_port_cap *port_cap)
399 {
400 	dev->caps.vl_cap[port]	    = port_cap->max_vl;
401 	dev->caps.ib_mtu_cap[port]	    = port_cap->ib_mtu;
402 	dev->phys_caps.gid_phys_table_len[port]  = port_cap->max_gids;
403 	dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
404 	/* set gid and pkey table operating lengths by default
405 	 * to non-sriov values
406 	 */
407 	dev->caps.gid_table_len[port]  = port_cap->max_gids;
408 	dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
409 	dev->caps.port_width_cap[port] = port_cap->max_port_width;
410 	dev->caps.eth_mtu_cap[port]    = port_cap->eth_mtu;
411 	dev->caps.max_tc_eth	       = port_cap->max_tc_eth;
412 	dev->caps.def_mac[port]        = port_cap->def_mac;
413 	dev->caps.supported_type[port] = port_cap->supported_port_types;
414 	dev->caps.suggested_type[port] = port_cap->suggested_type;
415 	dev->caps.default_sense[port] = port_cap->default_sense;
416 	dev->caps.trans_type[port]	    = port_cap->trans_type;
417 	dev->caps.vendor_oui[port]     = port_cap->vendor_oui;
418 	dev->caps.wavelength[port]     = port_cap->wavelength;
419 	dev->caps.trans_code[port]     = port_cap->trans_code;
420 
421 	return 0;
422 }
423 
424 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
425 			 struct mlx4_port_cap *port_cap)
426 {
427 	int err = 0;
428 
429 	err = mlx4_QUERY_PORT(dev, port, port_cap);
430 
431 	if (err)
432 		mlx4_err(dev, "QUERY_PORT command failed.\n");
433 
434 	return err;
435 }
436 
437 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
438 {
439 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
440 		return;
441 
442 	if (mlx4_is_mfunc(dev)) {
443 		mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
444 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
445 		return;
446 	}
447 
448 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
449 		mlx4_dbg(dev,
450 			 "Keep FCS is not supported - Disabling Ignore FCS");
451 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
452 		return;
453 	}
454 }
455 
456 #define MLX4_A0_STEERING_TABLE_SIZE	256
457 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
458 {
459 	int err;
460 	int i;
461 
462 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
463 	if (err) {
464 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
465 		return err;
466 	}
467 	mlx4_dev_cap_dump(dev, dev_cap);
468 
469 	if (dev_cap->min_page_sz > PAGE_SIZE) {
470 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
471 			 dev_cap->min_page_sz, PAGE_SIZE);
472 		return -ENODEV;
473 	}
474 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
475 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
476 			 dev_cap->num_ports, MLX4_MAX_PORTS);
477 		return -ENODEV;
478 	}
479 
480 	if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
481 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
482 			 dev_cap->uar_size,
483 			 (unsigned long long)
484 			 pci_resource_len(dev->persist->pdev, 2));
485 		return -ENODEV;
486 	}
487 
488 	dev->caps.num_ports	     = dev_cap->num_ports;
489 	dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
490 	dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
491 				      dev->caps.num_sys_eqs :
492 				      MLX4_MAX_EQ_NUM;
493 	for (i = 1; i <= dev->caps.num_ports; ++i) {
494 		err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
495 		if (err) {
496 			mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
497 			return err;
498 		}
499 	}
500 
501 	dev->caps.uar_page_size	     = PAGE_SIZE;
502 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
503 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
504 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
505 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
506 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
507 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
508 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
509 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
510 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
511 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
512 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
513 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
514 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
515 	/*
516 	 * Subtract 1 from the limit because we need to allocate a
517 	 * spare CQE so the HCA HW can tell the difference between an
518 	 * empty CQ and a full CQ.
519 	 */
520 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
521 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
522 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
523 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
524 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
525 
526 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
527 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
528 					dev_cap->reserved_xrcds : 0;
529 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
530 					dev_cap->max_xrcds : 0;
531 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
532 
533 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
534 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
535 	dev->caps.flags		     = dev_cap->flags;
536 	dev->caps.flags2	     = dev_cap->flags2;
537 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
538 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
539 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
540 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
541 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
542 	dev->caps.wol_port[1]          = dev_cap->wol_port[1];
543 	dev->caps.wol_port[2]          = dev_cap->wol_port[2];
544 	dev->caps.health_buffer_addrs  = dev_cap->health_buffer_addrs;
545 
546 	/* Save uar page shift */
547 	if (!mlx4_is_slave(dev)) {
548 		/* Virtual PCI function needs to determine UAR page size from
549 		 * firmware. Only master PCI function can set the uar page size
550 		 */
551 		if (enable_4k_uar || !dev->persist->num_vfs)
552 			dev->uar_page_shift = DEFAULT_UAR_PAGE_SHIFT;
553 		else
554 			dev->uar_page_shift = PAGE_SHIFT;
555 
556 		mlx4_set_num_reserved_uars(dev, dev_cap);
557 	}
558 
559 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN) {
560 		struct mlx4_init_hca_param hca_param;
561 
562 		memset(&hca_param, 0, sizeof(hca_param));
563 		err = mlx4_QUERY_HCA(dev, &hca_param);
564 		/* Turn off PHV_EN flag in case phv_check_en is set.
565 		 * phv_check_en is a HW check that parse the packet and verify
566 		 * phv bit was reported correctly in the wqe. To allow QinQ
567 		 * PHV_EN flag should be set and phv_check_en must be cleared
568 		 * otherwise QinQ packets will be drop by the HW.
569 		 */
570 		if (err || hca_param.phv_check_en)
571 			dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_PHV_EN;
572 	}
573 
574 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
575 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
576 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
577 	/* Don't do sense port on multifunction devices (for now at least) */
578 	if (mlx4_is_mfunc(dev))
579 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
580 
581 	if (mlx4_low_memory_profile()) {
582 		dev->caps.log_num_macs  = MLX4_MIN_LOG_NUM_MAC;
583 		dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
584 	} else {
585 		dev->caps.log_num_macs  = log_num_mac;
586 		dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
587 	}
588 
589 	for (i = 1; i <= dev->caps.num_ports; ++i) {
590 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
591 		if (dev->caps.supported_type[i]) {
592 			/* if only ETH is supported - assign ETH */
593 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
594 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
595 			/* if only IB is supported, assign IB */
596 			else if (dev->caps.supported_type[i] ==
597 				 MLX4_PORT_TYPE_IB)
598 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
599 			else {
600 				/* if IB and ETH are supported, we set the port
601 				 * type according to user selection of port type;
602 				 * if user selected none, take the FW hint */
603 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
604 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
605 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
606 				else
607 					dev->caps.port_type[i] = port_type_array[i - 1];
608 			}
609 		}
610 		/*
611 		 * Link sensing is allowed on the port if 3 conditions are true:
612 		 * 1. Both protocols are supported on the port.
613 		 * 2. Different types are supported on the port
614 		 * 3. FW declared that it supports link sensing
615 		 */
616 		mlx4_priv(dev)->sense.sense_allowed[i] =
617 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
618 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
619 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
620 
621 		/*
622 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
623 		 * and perform sense_port FW command to try and set the correct
624 		 * port type from beginning
625 		 */
626 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
627 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
628 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
629 			mlx4_SENSE_PORT(dev, i, &sensed_port);
630 			if (sensed_port != MLX4_PORT_TYPE_NONE)
631 				dev->caps.port_type[i] = sensed_port;
632 		} else {
633 			dev->caps.possible_type[i] = dev->caps.port_type[i];
634 		}
635 
636 		if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
637 			dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
638 			mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
639 				  i, 1 << dev->caps.log_num_macs);
640 		}
641 		if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
642 			dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
643 			mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
644 				  i, 1 << dev->caps.log_num_vlans);
645 		}
646 	}
647 
648 	if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
649 	    (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
650 	    (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
651 		mlx4_warn(dev,
652 			  "Granular QoS per VF not supported with IB/Eth configuration\n");
653 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
654 	}
655 
656 	dev->caps.max_counters = dev_cap->max_counters;
657 
658 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
659 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
660 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
661 		(1 << dev->caps.log_num_macs) *
662 		(1 << dev->caps.log_num_vlans) *
663 		dev->caps.num_ports;
664 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
665 
666 	if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
667 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
668 		dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
669 	else
670 		dev->caps.dmfs_high_rate_qpn_base =
671 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
672 
673 	if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
674 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
675 		dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
676 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
677 		dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
678 	} else {
679 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
680 		dev->caps.dmfs_high_rate_qpn_base =
681 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
682 		dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
683 	}
684 
685 	dev->caps.rl_caps = dev_cap->rl_caps;
686 
687 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
688 		dev->caps.dmfs_high_rate_qpn_range;
689 
690 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
691 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
692 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
693 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
694 
695 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
696 
697 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
698 		if (dev_cap->flags &
699 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
700 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
701 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
702 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
703 		}
704 
705 		if (dev_cap->flags2 &
706 		    (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
707 		     MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
708 			mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
709 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
710 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
711 		}
712 	}
713 
714 	if ((dev->caps.flags &
715 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
716 	    mlx4_is_master(dev))
717 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
718 
719 	if (!mlx4_is_slave(dev)) {
720 		mlx4_enable_cqe_eqe_stride(dev);
721 		dev->caps.alloc_res_qp_mask =
722 			(dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
723 			MLX4_RESERVE_A0_QP;
724 
725 		if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
726 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
727 			mlx4_warn(dev, "Old device ETS support detected\n");
728 			mlx4_warn(dev, "Consider upgrading device FW.\n");
729 			dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
730 		}
731 
732 	} else {
733 		dev->caps.alloc_res_qp_mask = 0;
734 	}
735 
736 	mlx4_enable_ignore_fcs(dev);
737 
738 	return 0;
739 }
740 
741 /*The function checks if there are live vf, return the num of them*/
742 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
743 {
744 	struct mlx4_priv *priv = mlx4_priv(dev);
745 	struct mlx4_slave_state *s_state;
746 	int i;
747 	int ret = 0;
748 
749 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
750 		s_state = &priv->mfunc.master.slave_state[i];
751 		if (s_state->active && s_state->last_cmd !=
752 		    MLX4_COMM_CMD_RESET) {
753 			mlx4_warn(dev, "%s: slave: %d is still active\n",
754 				  __func__, i);
755 			ret++;
756 		}
757 	}
758 	return ret;
759 }
760 
761 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
762 {
763 	u32 qk = MLX4_RESERVED_QKEY_BASE;
764 
765 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
766 	    qpn < dev->phys_caps.base_proxy_sqpn)
767 		return -EINVAL;
768 
769 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
770 		/* tunnel qp */
771 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
772 	else
773 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
774 	*qkey = qk;
775 	return 0;
776 }
777 EXPORT_SYMBOL(mlx4_get_parav_qkey);
778 
779 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
780 {
781 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
782 
783 	if (!mlx4_is_master(dev))
784 		return;
785 
786 	priv->virt2phys_pkey[slave][port - 1][i] = val;
787 }
788 EXPORT_SYMBOL(mlx4_sync_pkey_table);
789 
790 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
791 {
792 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
793 
794 	if (!mlx4_is_master(dev))
795 		return;
796 
797 	priv->slave_node_guids[slave] = guid;
798 }
799 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
800 
801 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
802 {
803 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
804 
805 	if (!mlx4_is_master(dev))
806 		return 0;
807 
808 	return priv->slave_node_guids[slave];
809 }
810 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
811 
812 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
813 {
814 	struct mlx4_priv *priv = mlx4_priv(dev);
815 	struct mlx4_slave_state *s_slave;
816 
817 	if (!mlx4_is_master(dev))
818 		return 0;
819 
820 	s_slave = &priv->mfunc.master.slave_state[slave];
821 	return !!s_slave->active;
822 }
823 EXPORT_SYMBOL(mlx4_is_slave_active);
824 
825 void mlx4_handle_eth_header_mcast_prio(struct mlx4_net_trans_rule_hw_ctrl *ctrl,
826 				       struct _rule_hw *eth_header)
827 {
828 	if (is_multicast_ether_addr(eth_header->eth.dst_mac) ||
829 	    is_broadcast_ether_addr(eth_header->eth.dst_mac)) {
830 		struct mlx4_net_trans_rule_hw_eth *eth =
831 			(struct mlx4_net_trans_rule_hw_eth *)eth_header;
832 		struct _rule_hw *next_rule = (struct _rule_hw *)(eth + 1);
833 		bool last_rule = next_rule->size == 0 && next_rule->id == 0 &&
834 			next_rule->rsvd == 0;
835 
836 		if (last_rule)
837 			ctrl->prio = cpu_to_be16(MLX4_DOMAIN_NIC);
838 	}
839 }
840 EXPORT_SYMBOL(mlx4_handle_eth_header_mcast_prio);
841 
842 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
843 				       struct mlx4_dev_cap *dev_cap,
844 				       struct mlx4_init_hca_param *hca_param)
845 {
846 	dev->caps.steering_mode = hca_param->steering_mode;
847 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
848 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
849 		dev->caps.fs_log_max_ucast_qp_range_size =
850 			dev_cap->fs_log_max_ucast_qp_range_size;
851 	} else
852 		dev->caps.num_qp_per_mgm =
853 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
854 
855 	mlx4_dbg(dev, "Steering mode is: %s\n",
856 		 mlx4_steering_mode_str(dev->caps.steering_mode));
857 }
858 
859 static void mlx4_slave_destroy_special_qp_cap(struct mlx4_dev *dev)
860 {
861 	kfree(dev->caps.spec_qps);
862 	dev->caps.spec_qps = NULL;
863 }
864 
865 static int mlx4_slave_special_qp_cap(struct mlx4_dev *dev)
866 {
867 	struct mlx4_func_cap *func_cap = NULL;
868 	struct mlx4_caps *caps = &dev->caps;
869 	int i, err = 0;
870 
871 	func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
872 	caps->spec_qps = kcalloc(caps->num_ports, sizeof(*caps->spec_qps), GFP_KERNEL);
873 
874 	if (!func_cap || !caps->spec_qps) {
875 		mlx4_err(dev, "Failed to allocate memory for special qps cap\n");
876 		err = -ENOMEM;
877 		goto err_mem;
878 	}
879 
880 	for (i = 1; i <= caps->num_ports; ++i) {
881 		err = mlx4_QUERY_FUNC_CAP(dev, i, func_cap);
882 		if (err) {
883 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
884 				 i, err);
885 			goto err_mem;
886 		}
887 		caps->spec_qps[i - 1] = func_cap->spec_qps;
888 		caps->port_mask[i] = caps->port_type[i];
889 		caps->phys_port_id[i] = func_cap->phys_port_id;
890 		err = mlx4_get_slave_pkey_gid_tbl_len(dev, i,
891 						      &caps->gid_table_len[i],
892 						      &caps->pkey_table_len[i]);
893 		if (err) {
894 			mlx4_err(dev, "QUERY_PORT command failed for port %d, aborting (%d)\n",
895 				 i, err);
896 			goto err_mem;
897 		}
898 	}
899 
900 err_mem:
901 	if (err)
902 		mlx4_slave_destroy_special_qp_cap(dev);
903 	kfree(func_cap);
904 	return err;
905 }
906 
907 static int mlx4_slave_cap(struct mlx4_dev *dev)
908 {
909 	int			   err;
910 	u32			   page_size;
911 	struct mlx4_dev_cap	   *dev_cap = NULL;
912 	struct mlx4_func_cap	   *func_cap = NULL;
913 	struct mlx4_init_hca_param *hca_param = NULL;
914 
915 	hca_param = kzalloc(sizeof(*hca_param), GFP_KERNEL);
916 	func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
917 	dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
918 	if (!hca_param || !func_cap || !dev_cap) {
919 		mlx4_err(dev, "Failed to allocate memory for slave_cap\n");
920 		err = -ENOMEM;
921 		goto free_mem;
922 	}
923 
924 	err = mlx4_QUERY_HCA(dev, hca_param);
925 	if (err) {
926 		mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
927 		goto free_mem;
928 	}
929 
930 	/* fail if the hca has an unknown global capability
931 	 * at this time global_caps should be always zeroed
932 	 */
933 	if (hca_param->global_caps) {
934 		mlx4_err(dev, "Unknown hca global capabilities\n");
935 		err = -EINVAL;
936 		goto free_mem;
937 	}
938 
939 	dev->caps.hca_core_clock = hca_param->hca_core_clock;
940 
941 	dev->caps.max_qp_dest_rdma = 1 << hca_param->log_rd_per_qp;
942 	err = mlx4_dev_cap(dev, dev_cap);
943 	if (err) {
944 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
945 		goto free_mem;
946 	}
947 
948 	err = mlx4_QUERY_FW(dev);
949 	if (err)
950 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
951 
952 	page_size = ~dev->caps.page_size_cap + 1;
953 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
954 	if (page_size > PAGE_SIZE) {
955 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
956 			 page_size, PAGE_SIZE);
957 		err = -ENODEV;
958 		goto free_mem;
959 	}
960 
961 	/* Set uar_page_shift for VF */
962 	dev->uar_page_shift = hca_param->uar_page_sz + 12;
963 
964 	/* Make sure the master uar page size is valid */
965 	if (dev->uar_page_shift > PAGE_SHIFT) {
966 		mlx4_err(dev,
967 			 "Invalid configuration: uar page size is larger than system page size\n");
968 		err = -ENODEV;
969 		goto free_mem;
970 	}
971 
972 	/* Set reserved_uars based on the uar_page_shift */
973 	mlx4_set_num_reserved_uars(dev, dev_cap);
974 
975 	/* Although uar page size in FW differs from system page size,
976 	 * upper software layers (mlx4_ib, mlx4_en and part of mlx4_core)
977 	 * still works with assumption that uar page size == system page size
978 	 */
979 	dev->caps.uar_page_size = PAGE_SIZE;
980 
981 	err = mlx4_QUERY_FUNC_CAP(dev, 0, func_cap);
982 	if (err) {
983 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
984 			 err);
985 		goto free_mem;
986 	}
987 
988 	if ((func_cap->pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
989 	    PF_CONTEXT_BEHAVIOUR_MASK) {
990 		mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
991 			 func_cap->pf_context_behaviour,
992 			 PF_CONTEXT_BEHAVIOUR_MASK);
993 		err = -EINVAL;
994 		goto free_mem;
995 	}
996 
997 	dev->caps.num_ports		= func_cap->num_ports;
998 	dev->quotas.qp			= func_cap->qp_quota;
999 	dev->quotas.srq			= func_cap->srq_quota;
1000 	dev->quotas.cq			= func_cap->cq_quota;
1001 	dev->quotas.mpt			= func_cap->mpt_quota;
1002 	dev->quotas.mtt			= func_cap->mtt_quota;
1003 	dev->caps.num_qps		= 1 << hca_param->log_num_qps;
1004 	dev->caps.num_srqs		= 1 << hca_param->log_num_srqs;
1005 	dev->caps.num_cqs		= 1 << hca_param->log_num_cqs;
1006 	dev->caps.num_mpts		= 1 << hca_param->log_mpt_sz;
1007 	dev->caps.num_eqs		= func_cap->max_eq;
1008 	dev->caps.reserved_eqs		= func_cap->reserved_eq;
1009 	dev->caps.reserved_lkey		= func_cap->reserved_lkey;
1010 	dev->caps.num_pds               = MLX4_NUM_PDS;
1011 	dev->caps.num_mgms              = 0;
1012 	dev->caps.num_amgms             = 0;
1013 
1014 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
1015 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
1016 			 dev->caps.num_ports, MLX4_MAX_PORTS);
1017 		err = -ENODEV;
1018 		goto free_mem;
1019 	}
1020 
1021 	mlx4_replace_zero_macs(dev);
1022 
1023 	err = mlx4_slave_special_qp_cap(dev);
1024 	if (err) {
1025 		mlx4_err(dev, "Set special QP caps failed. aborting\n");
1026 		goto free_mem;
1027 	}
1028 
1029 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
1030 				       dev->caps.reserved_uars) >
1031 				       pci_resource_len(dev->persist->pdev,
1032 							2)) {
1033 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
1034 			 dev->caps.uar_page_size * dev->caps.num_uars,
1035 			 (unsigned long long)
1036 			 pci_resource_len(dev->persist->pdev, 2));
1037 		err = -ENOMEM;
1038 		goto err_mem;
1039 	}
1040 
1041 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
1042 		dev->caps.eqe_size   = 64;
1043 		dev->caps.eqe_factor = 1;
1044 	} else {
1045 		dev->caps.eqe_size   = 32;
1046 		dev->caps.eqe_factor = 0;
1047 	}
1048 
1049 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
1050 		dev->caps.cqe_size   = 64;
1051 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1052 	} else {
1053 		dev->caps.cqe_size   = 32;
1054 	}
1055 
1056 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
1057 		dev->caps.eqe_size = hca_param->eqe_size;
1058 		dev->caps.eqe_factor = 0;
1059 	}
1060 
1061 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
1062 		dev->caps.cqe_size = hca_param->cqe_size;
1063 		/* User still need to know when CQE > 32B */
1064 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1065 	}
1066 
1067 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1068 	mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
1069 
1070 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_USER_MAC_EN;
1071 	mlx4_dbg(dev, "User MAC FW update is not supported in slave mode\n");
1072 
1073 	slave_adjust_steering_mode(dev, dev_cap, hca_param);
1074 	mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
1075 		 hca_param->rss_ip_frags ? "on" : "off");
1076 
1077 	if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
1078 	    dev->caps.bf_reg_size)
1079 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
1080 
1081 	if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
1082 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
1083 
1084 err_mem:
1085 	if (err)
1086 		mlx4_slave_destroy_special_qp_cap(dev);
1087 free_mem:
1088 	kfree(hca_param);
1089 	kfree(func_cap);
1090 	kfree(dev_cap);
1091 	return err;
1092 }
1093 
1094 static void mlx4_request_modules(struct mlx4_dev *dev)
1095 {
1096 	int port;
1097 	int has_ib_port = false;
1098 	int has_eth_port = false;
1099 #define EN_DRV_NAME	"mlx4_en"
1100 #define IB_DRV_NAME	"mlx4_ib"
1101 
1102 	for (port = 1; port <= dev->caps.num_ports; port++) {
1103 		if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
1104 			has_ib_port = true;
1105 		else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
1106 			has_eth_port = true;
1107 	}
1108 
1109 	if (has_eth_port)
1110 		request_module_nowait(EN_DRV_NAME);
1111 	if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
1112 		request_module_nowait(IB_DRV_NAME);
1113 }
1114 
1115 /*
1116  * Change the port configuration of the device.
1117  * Every user of this function must hold the port mutex.
1118  */
1119 int mlx4_change_port_types(struct mlx4_dev *dev,
1120 			   enum mlx4_port_type *port_types)
1121 {
1122 	int err = 0;
1123 	int change = 0;
1124 	int port;
1125 
1126 	for (port = 0; port <  dev->caps.num_ports; port++) {
1127 		/* Change the port type only if the new type is different
1128 		 * from the current, and not set to Auto */
1129 		if (port_types[port] != dev->caps.port_type[port + 1])
1130 			change = 1;
1131 	}
1132 	if (change) {
1133 		mlx4_unregister_device(dev);
1134 		for (port = 1; port <= dev->caps.num_ports; port++) {
1135 			mlx4_CLOSE_PORT(dev, port);
1136 			dev->caps.port_type[port] = port_types[port - 1];
1137 			err = mlx4_SET_PORT(dev, port, -1);
1138 			if (err) {
1139 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1140 					 port);
1141 				goto out;
1142 			}
1143 		}
1144 		mlx4_set_port_mask(dev);
1145 		err = mlx4_register_device(dev);
1146 		if (err) {
1147 			mlx4_err(dev, "Failed to register device\n");
1148 			goto out;
1149 		}
1150 		mlx4_request_modules(dev);
1151 	}
1152 
1153 out:
1154 	return err;
1155 }
1156 
1157 static ssize_t show_port_type(struct device *dev,
1158 			      struct device_attribute *attr,
1159 			      char *buf)
1160 {
1161 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1162 						   port_attr);
1163 	struct mlx4_dev *mdev = info->dev;
1164 	char type[8];
1165 
1166 	sprintf(type, "%s",
1167 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1168 		"ib" : "eth");
1169 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1170 		sprintf(buf, "auto (%s)\n", type);
1171 	else
1172 		sprintf(buf, "%s\n", type);
1173 
1174 	return strlen(buf);
1175 }
1176 
1177 static int __set_port_type(struct mlx4_port_info *info,
1178 			   enum mlx4_port_type port_type)
1179 {
1180 	struct mlx4_dev *mdev = info->dev;
1181 	struct mlx4_priv *priv = mlx4_priv(mdev);
1182 	enum mlx4_port_type types[MLX4_MAX_PORTS];
1183 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1184 	int i;
1185 	int err = 0;
1186 
1187 	if ((port_type & mdev->caps.supported_type[info->port]) != port_type) {
1188 		mlx4_err(mdev,
1189 			 "Requested port type for port %d is not supported on this HCA\n",
1190 			 info->port);
1191 		return -EOPNOTSUPP;
1192 	}
1193 
1194 	mlx4_stop_sense(mdev);
1195 	mutex_lock(&priv->port_mutex);
1196 	info->tmp_type = port_type;
1197 
1198 	/* Possible type is always the one that was delivered */
1199 	mdev->caps.possible_type[info->port] = info->tmp_type;
1200 
1201 	for (i = 0; i < mdev->caps.num_ports; i++) {
1202 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1203 					mdev->caps.possible_type[i+1];
1204 		if (types[i] == MLX4_PORT_TYPE_AUTO)
1205 			types[i] = mdev->caps.port_type[i+1];
1206 	}
1207 
1208 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1209 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1210 		for (i = 1; i <= mdev->caps.num_ports; i++) {
1211 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1212 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1213 				err = -EOPNOTSUPP;
1214 			}
1215 		}
1216 	}
1217 	if (err) {
1218 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1219 		goto out;
1220 	}
1221 
1222 	mlx4_do_sense_ports(mdev, new_types, types);
1223 
1224 	err = mlx4_check_port_params(mdev, new_types);
1225 	if (err)
1226 		goto out;
1227 
1228 	/* We are about to apply the changes after the configuration
1229 	 * was verified, no need to remember the temporary types
1230 	 * any more */
1231 	for (i = 0; i < mdev->caps.num_ports; i++)
1232 		priv->port[i + 1].tmp_type = 0;
1233 
1234 	err = mlx4_change_port_types(mdev, new_types);
1235 
1236 out:
1237 	mlx4_start_sense(mdev);
1238 	mutex_unlock(&priv->port_mutex);
1239 
1240 	return err;
1241 }
1242 
1243 static ssize_t set_port_type(struct device *dev,
1244 			     struct device_attribute *attr,
1245 			     const char *buf, size_t count)
1246 {
1247 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1248 						   port_attr);
1249 	struct mlx4_dev *mdev = info->dev;
1250 	enum mlx4_port_type port_type;
1251 	static DEFINE_MUTEX(set_port_type_mutex);
1252 	int err;
1253 
1254 	mutex_lock(&set_port_type_mutex);
1255 
1256 	if (!strcmp(buf, "ib\n")) {
1257 		port_type = MLX4_PORT_TYPE_IB;
1258 	} else if (!strcmp(buf, "eth\n")) {
1259 		port_type = MLX4_PORT_TYPE_ETH;
1260 	} else if (!strcmp(buf, "auto\n")) {
1261 		port_type = MLX4_PORT_TYPE_AUTO;
1262 	} else {
1263 		mlx4_err(mdev, "%s is not supported port type\n", buf);
1264 		err = -EINVAL;
1265 		goto err_out;
1266 	}
1267 
1268 	err = __set_port_type(info, port_type);
1269 
1270 err_out:
1271 	mutex_unlock(&set_port_type_mutex);
1272 
1273 	return err ? err : count;
1274 }
1275 
1276 enum ibta_mtu {
1277 	IB_MTU_256  = 1,
1278 	IB_MTU_512  = 2,
1279 	IB_MTU_1024 = 3,
1280 	IB_MTU_2048 = 4,
1281 	IB_MTU_4096 = 5
1282 };
1283 
1284 static inline int int_to_ibta_mtu(int mtu)
1285 {
1286 	switch (mtu) {
1287 	case 256:  return IB_MTU_256;
1288 	case 512:  return IB_MTU_512;
1289 	case 1024: return IB_MTU_1024;
1290 	case 2048: return IB_MTU_2048;
1291 	case 4096: return IB_MTU_4096;
1292 	default: return -1;
1293 	}
1294 }
1295 
1296 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1297 {
1298 	switch (mtu) {
1299 	case IB_MTU_256:  return  256;
1300 	case IB_MTU_512:  return  512;
1301 	case IB_MTU_1024: return 1024;
1302 	case IB_MTU_2048: return 2048;
1303 	case IB_MTU_4096: return 4096;
1304 	default: return -1;
1305 	}
1306 }
1307 
1308 static ssize_t show_port_ib_mtu(struct device *dev,
1309 			     struct device_attribute *attr,
1310 			     char *buf)
1311 {
1312 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1313 						   port_mtu_attr);
1314 	struct mlx4_dev *mdev = info->dev;
1315 
1316 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1317 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1318 
1319 	sprintf(buf, "%d\n",
1320 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1321 	return strlen(buf);
1322 }
1323 
1324 static ssize_t set_port_ib_mtu(struct device *dev,
1325 			     struct device_attribute *attr,
1326 			     const char *buf, size_t count)
1327 {
1328 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1329 						   port_mtu_attr);
1330 	struct mlx4_dev *mdev = info->dev;
1331 	struct mlx4_priv *priv = mlx4_priv(mdev);
1332 	int err, port, mtu, ibta_mtu = -1;
1333 
1334 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1335 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1336 		return -EINVAL;
1337 	}
1338 
1339 	err = kstrtoint(buf, 0, &mtu);
1340 	if (!err)
1341 		ibta_mtu = int_to_ibta_mtu(mtu);
1342 
1343 	if (err || ibta_mtu < 0) {
1344 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1345 		return -EINVAL;
1346 	}
1347 
1348 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1349 
1350 	mlx4_stop_sense(mdev);
1351 	mutex_lock(&priv->port_mutex);
1352 	mlx4_unregister_device(mdev);
1353 	for (port = 1; port <= mdev->caps.num_ports; port++) {
1354 		mlx4_CLOSE_PORT(mdev, port);
1355 		err = mlx4_SET_PORT(mdev, port, -1);
1356 		if (err) {
1357 			mlx4_err(mdev, "Failed to set port %d, aborting\n",
1358 				 port);
1359 			goto err_set_port;
1360 		}
1361 	}
1362 	err = mlx4_register_device(mdev);
1363 err_set_port:
1364 	mutex_unlock(&priv->port_mutex);
1365 	mlx4_start_sense(mdev);
1366 	return err ? err : count;
1367 }
1368 
1369 /* bond for multi-function device */
1370 #define MAX_MF_BOND_ALLOWED_SLAVES 63
1371 static int mlx4_mf_bond(struct mlx4_dev *dev)
1372 {
1373 	int err = 0;
1374 	int nvfs;
1375 	struct mlx4_slaves_pport slaves_port1;
1376 	struct mlx4_slaves_pport slaves_port2;
1377 	DECLARE_BITMAP(slaves_port_1_2, MLX4_MFUNC_MAX);
1378 
1379 	slaves_port1 = mlx4_phys_to_slaves_pport(dev, 1);
1380 	slaves_port2 = mlx4_phys_to_slaves_pport(dev, 2);
1381 	bitmap_and(slaves_port_1_2,
1382 		   slaves_port1.slaves, slaves_port2.slaves,
1383 		   dev->persist->num_vfs + 1);
1384 
1385 	/* only single port vfs are allowed */
1386 	if (bitmap_weight(slaves_port_1_2, dev->persist->num_vfs + 1) > 1) {
1387 		mlx4_warn(dev, "HA mode unsupported for dual ported VFs\n");
1388 		return -EINVAL;
1389 	}
1390 
1391 	/* number of virtual functions is number of total functions minus one
1392 	 * physical function for each port.
1393 	 */
1394 	nvfs = bitmap_weight(slaves_port1.slaves, dev->persist->num_vfs + 1) +
1395 		bitmap_weight(slaves_port2.slaves, dev->persist->num_vfs + 1) - 2;
1396 
1397 	/* limit on maximum allowed VFs */
1398 	if (nvfs > MAX_MF_BOND_ALLOWED_SLAVES) {
1399 		mlx4_warn(dev, "HA mode is not supported for %d VFs (max %d are allowed)\n",
1400 			  nvfs, MAX_MF_BOND_ALLOWED_SLAVES);
1401 		return -EINVAL;
1402 	}
1403 
1404 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) {
1405 		mlx4_warn(dev, "HA mode unsupported for NON DMFS steering\n");
1406 		return -EINVAL;
1407 	}
1408 
1409 	err = mlx4_bond_mac_table(dev);
1410 	if (err)
1411 		return err;
1412 	err = mlx4_bond_vlan_table(dev);
1413 	if (err)
1414 		goto err1;
1415 	err = mlx4_bond_fs_rules(dev);
1416 	if (err)
1417 		goto err2;
1418 
1419 	return 0;
1420 err2:
1421 	(void)mlx4_unbond_vlan_table(dev);
1422 err1:
1423 	(void)mlx4_unbond_mac_table(dev);
1424 	return err;
1425 }
1426 
1427 static int mlx4_mf_unbond(struct mlx4_dev *dev)
1428 {
1429 	int ret, ret1;
1430 
1431 	ret = mlx4_unbond_fs_rules(dev);
1432 	if (ret)
1433 		mlx4_warn(dev, "multifunction unbond for flow rules failed (%d)\n", ret);
1434 	ret1 = mlx4_unbond_mac_table(dev);
1435 	if (ret1) {
1436 		mlx4_warn(dev, "multifunction unbond for MAC table failed (%d)\n", ret1);
1437 		ret = ret1;
1438 	}
1439 	ret1 = mlx4_unbond_vlan_table(dev);
1440 	if (ret1) {
1441 		mlx4_warn(dev, "multifunction unbond for VLAN table failed (%d)\n", ret1);
1442 		ret = ret1;
1443 	}
1444 	return ret;
1445 }
1446 
1447 int mlx4_bond(struct mlx4_dev *dev)
1448 {
1449 	int ret = 0;
1450 	struct mlx4_priv *priv = mlx4_priv(dev);
1451 
1452 	mutex_lock(&priv->bond_mutex);
1453 
1454 	if (!mlx4_is_bonded(dev)) {
1455 		ret = mlx4_do_bond(dev, true);
1456 		if (ret)
1457 			mlx4_err(dev, "Failed to bond device: %d\n", ret);
1458 		if (!ret && mlx4_is_master(dev)) {
1459 			ret = mlx4_mf_bond(dev);
1460 			if (ret) {
1461 				mlx4_err(dev, "bond for multifunction failed\n");
1462 				mlx4_do_bond(dev, false);
1463 			}
1464 		}
1465 	}
1466 
1467 	mutex_unlock(&priv->bond_mutex);
1468 	if (!ret)
1469 		mlx4_dbg(dev, "Device is bonded\n");
1470 
1471 	return ret;
1472 }
1473 EXPORT_SYMBOL_GPL(mlx4_bond);
1474 
1475 int mlx4_unbond(struct mlx4_dev *dev)
1476 {
1477 	int ret = 0;
1478 	struct mlx4_priv *priv = mlx4_priv(dev);
1479 
1480 	mutex_lock(&priv->bond_mutex);
1481 
1482 	if (mlx4_is_bonded(dev)) {
1483 		int ret2 = 0;
1484 
1485 		ret = mlx4_do_bond(dev, false);
1486 		if (ret)
1487 			mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1488 		if (mlx4_is_master(dev))
1489 			ret2 = mlx4_mf_unbond(dev);
1490 		if (ret2) {
1491 			mlx4_warn(dev, "Failed to unbond device for multifunction (%d)\n", ret2);
1492 			ret = ret2;
1493 		}
1494 	}
1495 
1496 	mutex_unlock(&priv->bond_mutex);
1497 	if (!ret)
1498 		mlx4_dbg(dev, "Device is unbonded\n");
1499 
1500 	return ret;
1501 }
1502 EXPORT_SYMBOL_GPL(mlx4_unbond);
1503 
1504 
1505 int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1506 {
1507 	u8 port1 = v2p->port1;
1508 	u8 port2 = v2p->port2;
1509 	struct mlx4_priv *priv = mlx4_priv(dev);
1510 	int err;
1511 
1512 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1513 		return -EOPNOTSUPP;
1514 
1515 	mutex_lock(&priv->bond_mutex);
1516 
1517 	/* zero means keep current mapping for this port */
1518 	if (port1 == 0)
1519 		port1 = priv->v2p.port1;
1520 	if (port2 == 0)
1521 		port2 = priv->v2p.port2;
1522 
1523 	if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1524 	    (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1525 	    (port1 == 2 && port2 == 1)) {
1526 		/* besides boundary checks cross mapping makes
1527 		 * no sense and therefore not allowed */
1528 		err = -EINVAL;
1529 	} else if ((port1 == priv->v2p.port1) &&
1530 		 (port2 == priv->v2p.port2)) {
1531 		err = 0;
1532 	} else {
1533 		err = mlx4_virt2phy_port_map(dev, port1, port2);
1534 		if (!err) {
1535 			mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1536 				 port1, port2);
1537 			priv->v2p.port1 = port1;
1538 			priv->v2p.port2 = port2;
1539 		} else {
1540 			mlx4_err(dev, "Failed to change port mape: %d\n", err);
1541 		}
1542 	}
1543 
1544 	mutex_unlock(&priv->bond_mutex);
1545 	return err;
1546 }
1547 EXPORT_SYMBOL_GPL(mlx4_port_map_set);
1548 
1549 static int mlx4_load_fw(struct mlx4_dev *dev)
1550 {
1551 	struct mlx4_priv *priv = mlx4_priv(dev);
1552 	int err;
1553 
1554 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1555 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
1556 	if (!priv->fw.fw_icm) {
1557 		mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1558 		return -ENOMEM;
1559 	}
1560 
1561 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1562 	if (err) {
1563 		mlx4_err(dev, "MAP_FA command failed, aborting\n");
1564 		goto err_free;
1565 	}
1566 
1567 	err = mlx4_RUN_FW(dev);
1568 	if (err) {
1569 		mlx4_err(dev, "RUN_FW command failed, aborting\n");
1570 		goto err_unmap_fa;
1571 	}
1572 
1573 	return 0;
1574 
1575 err_unmap_fa:
1576 	mlx4_UNMAP_FA(dev);
1577 
1578 err_free:
1579 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1580 	return err;
1581 }
1582 
1583 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1584 				int cmpt_entry_sz)
1585 {
1586 	struct mlx4_priv *priv = mlx4_priv(dev);
1587 	int err;
1588 	int num_eqs;
1589 
1590 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1591 				  cmpt_base +
1592 				  ((u64) (MLX4_CMPT_TYPE_QP *
1593 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1594 				  cmpt_entry_sz, dev->caps.num_qps,
1595 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1596 				  0, 0);
1597 	if (err)
1598 		goto err;
1599 
1600 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1601 				  cmpt_base +
1602 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1603 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1604 				  cmpt_entry_sz, dev->caps.num_srqs,
1605 				  dev->caps.reserved_srqs, 0, 0);
1606 	if (err)
1607 		goto err_qp;
1608 
1609 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1610 				  cmpt_base +
1611 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1612 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1613 				  cmpt_entry_sz, dev->caps.num_cqs,
1614 				  dev->caps.reserved_cqs, 0, 0);
1615 	if (err)
1616 		goto err_srq;
1617 
1618 	num_eqs = dev->phys_caps.num_phys_eqs;
1619 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1620 				  cmpt_base +
1621 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1622 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1623 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1624 	if (err)
1625 		goto err_cq;
1626 
1627 	return 0;
1628 
1629 err_cq:
1630 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1631 
1632 err_srq:
1633 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1634 
1635 err_qp:
1636 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1637 
1638 err:
1639 	return err;
1640 }
1641 
1642 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1643 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1644 {
1645 	struct mlx4_priv *priv = mlx4_priv(dev);
1646 	u64 aux_pages;
1647 	int num_eqs;
1648 	int err;
1649 
1650 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1651 	if (err) {
1652 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1653 		return err;
1654 	}
1655 
1656 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1657 		 (unsigned long long) icm_size >> 10,
1658 		 (unsigned long long) aux_pages << 2);
1659 
1660 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1661 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1662 	if (!priv->fw.aux_icm) {
1663 		mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1664 		return -ENOMEM;
1665 	}
1666 
1667 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1668 	if (err) {
1669 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1670 		goto err_free_aux;
1671 	}
1672 
1673 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1674 	if (err) {
1675 		mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1676 		goto err_unmap_aux;
1677 	}
1678 
1679 
1680 	num_eqs = dev->phys_caps.num_phys_eqs;
1681 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1682 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1683 				  num_eqs, num_eqs, 0, 0);
1684 	if (err) {
1685 		mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1686 		goto err_unmap_cmpt;
1687 	}
1688 
1689 	/*
1690 	 * Reserved MTT entries must be aligned up to a cacheline
1691 	 * boundary, since the FW will write to them, while the driver
1692 	 * writes to all other MTT entries. (The variable
1693 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1694 	 * size, not the raw entry size)
1695 	 */
1696 	dev->caps.reserved_mtts =
1697 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1698 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1699 
1700 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1701 				  init_hca->mtt_base,
1702 				  dev->caps.mtt_entry_sz,
1703 				  dev->caps.num_mtts,
1704 				  dev->caps.reserved_mtts, 1, 0);
1705 	if (err) {
1706 		mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1707 		goto err_unmap_eq;
1708 	}
1709 
1710 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1711 				  init_hca->dmpt_base,
1712 				  dev_cap->dmpt_entry_sz,
1713 				  dev->caps.num_mpts,
1714 				  dev->caps.reserved_mrws, 1, 1);
1715 	if (err) {
1716 		mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1717 		goto err_unmap_mtt;
1718 	}
1719 
1720 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1721 				  init_hca->qpc_base,
1722 				  dev_cap->qpc_entry_sz,
1723 				  dev->caps.num_qps,
1724 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1725 				  0, 0);
1726 	if (err) {
1727 		mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1728 		goto err_unmap_dmpt;
1729 	}
1730 
1731 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1732 				  init_hca->auxc_base,
1733 				  dev_cap->aux_entry_sz,
1734 				  dev->caps.num_qps,
1735 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1736 				  0, 0);
1737 	if (err) {
1738 		mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1739 		goto err_unmap_qp;
1740 	}
1741 
1742 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1743 				  init_hca->altc_base,
1744 				  dev_cap->altc_entry_sz,
1745 				  dev->caps.num_qps,
1746 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1747 				  0, 0);
1748 	if (err) {
1749 		mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1750 		goto err_unmap_auxc;
1751 	}
1752 
1753 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1754 				  init_hca->rdmarc_base,
1755 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1756 				  dev->caps.num_qps,
1757 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1758 				  0, 0);
1759 	if (err) {
1760 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1761 		goto err_unmap_altc;
1762 	}
1763 
1764 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1765 				  init_hca->cqc_base,
1766 				  dev_cap->cqc_entry_sz,
1767 				  dev->caps.num_cqs,
1768 				  dev->caps.reserved_cqs, 0, 0);
1769 	if (err) {
1770 		mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1771 		goto err_unmap_rdmarc;
1772 	}
1773 
1774 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1775 				  init_hca->srqc_base,
1776 				  dev_cap->srq_entry_sz,
1777 				  dev->caps.num_srqs,
1778 				  dev->caps.reserved_srqs, 0, 0);
1779 	if (err) {
1780 		mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1781 		goto err_unmap_cq;
1782 	}
1783 
1784 	/*
1785 	 * For flow steering device managed mode it is required to use
1786 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1787 	 * required, but for simplicity just map the whole multicast
1788 	 * group table now.  The table isn't very big and it's a lot
1789 	 * easier than trying to track ref counts.
1790 	 */
1791 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1792 				  init_hca->mc_base,
1793 				  mlx4_get_mgm_entry_size(dev),
1794 				  dev->caps.num_mgms + dev->caps.num_amgms,
1795 				  dev->caps.num_mgms + dev->caps.num_amgms,
1796 				  0, 0);
1797 	if (err) {
1798 		mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1799 		goto err_unmap_srq;
1800 	}
1801 
1802 	return 0;
1803 
1804 err_unmap_srq:
1805 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1806 
1807 err_unmap_cq:
1808 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1809 
1810 err_unmap_rdmarc:
1811 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1812 
1813 err_unmap_altc:
1814 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1815 
1816 err_unmap_auxc:
1817 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1818 
1819 err_unmap_qp:
1820 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1821 
1822 err_unmap_dmpt:
1823 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1824 
1825 err_unmap_mtt:
1826 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1827 
1828 err_unmap_eq:
1829 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1830 
1831 err_unmap_cmpt:
1832 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1833 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1834 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1835 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1836 
1837 err_unmap_aux:
1838 	mlx4_UNMAP_ICM_AUX(dev);
1839 
1840 err_free_aux:
1841 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1842 
1843 	return err;
1844 }
1845 
1846 static void mlx4_free_icms(struct mlx4_dev *dev)
1847 {
1848 	struct mlx4_priv *priv = mlx4_priv(dev);
1849 
1850 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1851 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1852 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1853 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1854 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1855 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1856 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1857 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1858 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1859 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1860 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1861 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1862 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1863 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1864 
1865 	mlx4_UNMAP_ICM_AUX(dev);
1866 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1867 }
1868 
1869 static void mlx4_slave_exit(struct mlx4_dev *dev)
1870 {
1871 	struct mlx4_priv *priv = mlx4_priv(dev);
1872 
1873 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1874 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1875 			  MLX4_COMM_TIME))
1876 		mlx4_warn(dev, "Failed to close slave function\n");
1877 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1878 }
1879 
1880 static int map_bf_area(struct mlx4_dev *dev)
1881 {
1882 	struct mlx4_priv *priv = mlx4_priv(dev);
1883 	resource_size_t bf_start;
1884 	resource_size_t bf_len;
1885 	int err = 0;
1886 
1887 	if (!dev->caps.bf_reg_size)
1888 		return -ENXIO;
1889 
1890 	bf_start = pci_resource_start(dev->persist->pdev, 2) +
1891 			(dev->caps.num_uars << PAGE_SHIFT);
1892 	bf_len = pci_resource_len(dev->persist->pdev, 2) -
1893 			(dev->caps.num_uars << PAGE_SHIFT);
1894 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1895 	if (!priv->bf_mapping)
1896 		err = -ENOMEM;
1897 
1898 	return err;
1899 }
1900 
1901 static void unmap_bf_area(struct mlx4_dev *dev)
1902 {
1903 	if (mlx4_priv(dev)->bf_mapping)
1904 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1905 }
1906 
1907 u64 mlx4_read_clock(struct mlx4_dev *dev)
1908 {
1909 	u32 clockhi, clocklo, clockhi1;
1910 	u64 cycles;
1911 	int i;
1912 	struct mlx4_priv *priv = mlx4_priv(dev);
1913 
1914 	for (i = 0; i < 10; i++) {
1915 		clockhi = swab32(readl(priv->clock_mapping));
1916 		clocklo = swab32(readl(priv->clock_mapping + 4));
1917 		clockhi1 = swab32(readl(priv->clock_mapping));
1918 		if (clockhi == clockhi1)
1919 			break;
1920 	}
1921 
1922 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1923 
1924 	return cycles;
1925 }
1926 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1927 
1928 
1929 static int map_internal_clock(struct mlx4_dev *dev)
1930 {
1931 	struct mlx4_priv *priv = mlx4_priv(dev);
1932 
1933 	priv->clock_mapping =
1934 		ioremap(pci_resource_start(dev->persist->pdev,
1935 					   priv->fw.clock_bar) +
1936 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1937 
1938 	if (!priv->clock_mapping)
1939 		return -ENOMEM;
1940 
1941 	return 0;
1942 }
1943 
1944 int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1945 				   struct mlx4_clock_params *params)
1946 {
1947 	struct mlx4_priv *priv = mlx4_priv(dev);
1948 
1949 	if (mlx4_is_slave(dev))
1950 		return -EOPNOTSUPP;
1951 
1952 	if (!params)
1953 		return -EINVAL;
1954 
1955 	params->bar = priv->fw.clock_bar;
1956 	params->offset = priv->fw.clock_offset;
1957 	params->size = MLX4_CLOCK_SIZE;
1958 
1959 	return 0;
1960 }
1961 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1962 
1963 static void unmap_internal_clock(struct mlx4_dev *dev)
1964 {
1965 	struct mlx4_priv *priv = mlx4_priv(dev);
1966 
1967 	if (priv->clock_mapping)
1968 		iounmap(priv->clock_mapping);
1969 }
1970 
1971 static void mlx4_close_hca(struct mlx4_dev *dev)
1972 {
1973 	unmap_internal_clock(dev);
1974 	unmap_bf_area(dev);
1975 	if (mlx4_is_slave(dev))
1976 		mlx4_slave_exit(dev);
1977 	else {
1978 		mlx4_CLOSE_HCA(dev, 0);
1979 		mlx4_free_icms(dev);
1980 	}
1981 }
1982 
1983 static void mlx4_close_fw(struct mlx4_dev *dev)
1984 {
1985 	if (!mlx4_is_slave(dev)) {
1986 		mlx4_UNMAP_FA(dev);
1987 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1988 	}
1989 }
1990 
1991 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1992 {
1993 #define COMM_CHAN_OFFLINE_OFFSET 0x09
1994 
1995 	u32 comm_flags;
1996 	u32 offline_bit;
1997 	unsigned long end;
1998 	struct mlx4_priv *priv = mlx4_priv(dev);
1999 
2000 	end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
2001 	while (time_before(jiffies, end)) {
2002 		comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
2003 					  MLX4_COMM_CHAN_FLAGS));
2004 		offline_bit = (comm_flags &
2005 			       (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
2006 		if (!offline_bit)
2007 			return 0;
2008 
2009 		/* If device removal has been requested,
2010 		 * do not continue retrying.
2011 		 */
2012 		if (dev->persist->interface_state &
2013 		    MLX4_INTERFACE_STATE_NOWAIT)
2014 			break;
2015 
2016 		/* There are cases as part of AER/Reset flow that PF needs
2017 		 * around 100 msec to load. We therefore sleep for 100 msec
2018 		 * to allow other tasks to make use of that CPU during this
2019 		 * time interval.
2020 		 */
2021 		msleep(100);
2022 	}
2023 	mlx4_err(dev, "Communication channel is offline.\n");
2024 	return -EIO;
2025 }
2026 
2027 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
2028 {
2029 #define COMM_CHAN_RST_OFFSET 0x1e
2030 
2031 	struct mlx4_priv *priv = mlx4_priv(dev);
2032 	u32 comm_rst;
2033 	u32 comm_caps;
2034 
2035 	comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
2036 				 MLX4_COMM_CHAN_CAPS));
2037 	comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
2038 
2039 	if (comm_rst)
2040 		dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
2041 }
2042 
2043 static int mlx4_init_slave(struct mlx4_dev *dev)
2044 {
2045 	struct mlx4_priv *priv = mlx4_priv(dev);
2046 	u64 dma = (u64) priv->mfunc.vhcr_dma;
2047 	int ret_from_reset = 0;
2048 	u32 slave_read;
2049 	u32 cmd_channel_ver;
2050 
2051 	if (atomic_read(&pf_loading)) {
2052 		mlx4_warn(dev, "PF is not ready - Deferring probe\n");
2053 		return -EPROBE_DEFER;
2054 	}
2055 
2056 	mutex_lock(&priv->cmd.slave_cmd_mutex);
2057 	priv->cmd.max_cmds = 1;
2058 	if (mlx4_comm_check_offline(dev)) {
2059 		mlx4_err(dev, "PF is not responsive, skipping initialization\n");
2060 		goto err_offline;
2061 	}
2062 
2063 	mlx4_reset_vf_support(dev);
2064 	mlx4_warn(dev, "Sending reset\n");
2065 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
2066 				       MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
2067 	/* if we are in the middle of flr the slave will try
2068 	 * NUM_OF_RESET_RETRIES times before leaving.*/
2069 	if (ret_from_reset) {
2070 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
2071 			mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
2072 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
2073 			return -EPROBE_DEFER;
2074 		} else
2075 			goto err;
2076 	}
2077 
2078 	/* check the driver version - the slave I/F revision
2079 	 * must match the master's */
2080 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
2081 	cmd_channel_ver = mlx4_comm_get_version();
2082 
2083 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
2084 		MLX4_COMM_GET_IF_REV(slave_read)) {
2085 		mlx4_err(dev, "slave driver version is not supported by the master\n");
2086 		goto err;
2087 	}
2088 
2089 	mlx4_warn(dev, "Sending vhcr0\n");
2090 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
2091 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2092 		goto err;
2093 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
2094 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2095 		goto err;
2096 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
2097 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2098 		goto err;
2099 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
2100 			  MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2101 		goto err;
2102 
2103 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2104 	return 0;
2105 
2106 err:
2107 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
2108 err_offline:
2109 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2110 	return -EIO;
2111 }
2112 
2113 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
2114 {
2115 	int i;
2116 
2117 	for (i = 1; i <= dev->caps.num_ports; i++) {
2118 		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
2119 			dev->caps.gid_table_len[i] =
2120 				mlx4_get_slave_num_gids(dev, 0, i);
2121 		else
2122 			dev->caps.gid_table_len[i] = 1;
2123 		dev->caps.pkey_table_len[i] =
2124 			dev->phys_caps.pkey_phys_table_len[i] - 1;
2125 	}
2126 }
2127 
2128 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
2129 {
2130 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
2131 
2132 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
2133 	      i++) {
2134 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
2135 			break;
2136 	}
2137 
2138 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
2139 }
2140 
2141 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
2142 {
2143 	switch (dmfs_high_steer_mode) {
2144 	case MLX4_STEERING_DMFS_A0_DEFAULT:
2145 		return "default performance";
2146 
2147 	case MLX4_STEERING_DMFS_A0_DYNAMIC:
2148 		return "dynamic hybrid mode";
2149 
2150 	case MLX4_STEERING_DMFS_A0_STATIC:
2151 		return "performance optimized for limited rule configuration (static)";
2152 
2153 	case MLX4_STEERING_DMFS_A0_DISABLE:
2154 		return "disabled performance optimized steering";
2155 
2156 	case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
2157 		return "performance optimized steering not supported";
2158 
2159 	default:
2160 		return "Unrecognized mode";
2161 	}
2162 }
2163 
2164 #define MLX4_DMFS_A0_STEERING			(1UL << 2)
2165 
2166 static void choose_steering_mode(struct mlx4_dev *dev,
2167 				 struct mlx4_dev_cap *dev_cap)
2168 {
2169 	if (mlx4_log_num_mgm_entry_size <= 0) {
2170 		if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
2171 			if (dev->caps.dmfs_high_steer_mode ==
2172 			    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2173 				mlx4_err(dev, "DMFS high rate mode not supported\n");
2174 			else
2175 				dev->caps.dmfs_high_steer_mode =
2176 					MLX4_STEERING_DMFS_A0_STATIC;
2177 		}
2178 	}
2179 
2180 	if (mlx4_log_num_mgm_entry_size <= 0 &&
2181 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
2182 	    (!mlx4_is_mfunc(dev) ||
2183 	     (dev_cap->fs_max_num_qp_per_entry >=
2184 	     (dev->persist->num_vfs + 1))) &&
2185 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
2186 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
2187 		dev->oper_log_mgm_entry_size =
2188 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
2189 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
2190 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
2191 		dev->caps.fs_log_max_ucast_qp_range_size =
2192 			dev_cap->fs_log_max_ucast_qp_range_size;
2193 	} else {
2194 		if (dev->caps.dmfs_high_steer_mode !=
2195 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2196 			dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
2197 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
2198 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2199 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
2200 		else {
2201 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
2202 
2203 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
2204 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2205 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
2206 		}
2207 		dev->oper_log_mgm_entry_size =
2208 			mlx4_log_num_mgm_entry_size > 0 ?
2209 			mlx4_log_num_mgm_entry_size :
2210 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
2211 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
2212 	}
2213 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
2214 		 mlx4_steering_mode_str(dev->caps.steering_mode),
2215 		 dev->oper_log_mgm_entry_size,
2216 		 mlx4_log_num_mgm_entry_size);
2217 }
2218 
2219 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
2220 				       struct mlx4_dev_cap *dev_cap)
2221 {
2222 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2223 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
2224 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
2225 	else
2226 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
2227 
2228 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
2229 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
2230 }
2231 
2232 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
2233 {
2234 	int i;
2235 	struct mlx4_port_cap port_cap;
2236 
2237 	if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2238 		return -EINVAL;
2239 
2240 	for (i = 1; i <= dev->caps.num_ports; i++) {
2241 		if (mlx4_dev_port(dev, i, &port_cap)) {
2242 			mlx4_err(dev,
2243 				 "QUERY_DEV_CAP command failed, can't verify DMFS high rate steering.\n");
2244 		} else if ((dev->caps.dmfs_high_steer_mode !=
2245 			    MLX4_STEERING_DMFS_A0_DEFAULT) &&
2246 			   (port_cap.dmfs_optimized_state ==
2247 			    !!(dev->caps.dmfs_high_steer_mode ==
2248 			    MLX4_STEERING_DMFS_A0_DISABLE))) {
2249 			mlx4_err(dev,
2250 				 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2251 				 dmfs_high_rate_steering_mode_str(
2252 					dev->caps.dmfs_high_steer_mode),
2253 				 (port_cap.dmfs_optimized_state ?
2254 					"enabled" : "disabled"));
2255 		}
2256 	}
2257 
2258 	return 0;
2259 }
2260 
2261 static int mlx4_init_fw(struct mlx4_dev *dev)
2262 {
2263 	struct mlx4_mod_stat_cfg   mlx4_cfg;
2264 	int err = 0;
2265 
2266 	if (!mlx4_is_slave(dev)) {
2267 		err = mlx4_QUERY_FW(dev);
2268 		if (err) {
2269 			if (err == -EACCES)
2270 				mlx4_info(dev, "non-primary physical function, skipping\n");
2271 			else
2272 				mlx4_err(dev, "QUERY_FW command failed, aborting\n");
2273 			return err;
2274 		}
2275 
2276 		err = mlx4_load_fw(dev);
2277 		if (err) {
2278 			mlx4_err(dev, "Failed to start FW, aborting\n");
2279 			return err;
2280 		}
2281 
2282 		mlx4_cfg.log_pg_sz_m = 1;
2283 		mlx4_cfg.log_pg_sz = 0;
2284 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2285 		if (err)
2286 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2287 	}
2288 
2289 	return err;
2290 }
2291 
2292 static int mlx4_init_hca(struct mlx4_dev *dev)
2293 {
2294 	struct mlx4_priv	  *priv = mlx4_priv(dev);
2295 	struct mlx4_init_hca_param *init_hca = NULL;
2296 	struct mlx4_dev_cap	  *dev_cap = NULL;
2297 	struct mlx4_adapter	   adapter;
2298 	struct mlx4_profile	   profile;
2299 	u64 icm_size;
2300 	struct mlx4_config_dev_params params;
2301 	int err;
2302 
2303 	if (!mlx4_is_slave(dev)) {
2304 		dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
2305 		init_hca = kzalloc(sizeof(*init_hca), GFP_KERNEL);
2306 
2307 		if (!dev_cap || !init_hca) {
2308 			err = -ENOMEM;
2309 			goto out_free;
2310 		}
2311 
2312 		err = mlx4_dev_cap(dev, dev_cap);
2313 		if (err) {
2314 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
2315 			goto out_free;
2316 		}
2317 
2318 		choose_steering_mode(dev, dev_cap);
2319 		choose_tunnel_offload_mode(dev, dev_cap);
2320 
2321 		if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2322 		    mlx4_is_master(dev))
2323 			dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2324 
2325 		err = mlx4_get_phys_port_id(dev);
2326 		if (err)
2327 			mlx4_err(dev, "Fail to get physical port id\n");
2328 
2329 		if (mlx4_is_master(dev))
2330 			mlx4_parav_master_pf_caps(dev);
2331 
2332 		if (mlx4_low_memory_profile()) {
2333 			mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2334 			profile = low_mem_profile;
2335 		} else {
2336 			profile = default_profile;
2337 		}
2338 		if (dev->caps.steering_mode ==
2339 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
2340 			profile.num_mcg = MLX4_FS_NUM_MCG;
2341 
2342 		icm_size = mlx4_make_profile(dev, &profile, dev_cap,
2343 					     init_hca);
2344 		if ((long long) icm_size < 0) {
2345 			err = icm_size;
2346 			goto out_free;
2347 		}
2348 
2349 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
2350 
2351 		if (enable_4k_uar || !dev->persist->num_vfs) {
2352 			init_hca->log_uar_sz = ilog2(dev->caps.num_uars) +
2353 						    PAGE_SHIFT - DEFAULT_UAR_PAGE_SHIFT;
2354 			init_hca->uar_page_sz = DEFAULT_UAR_PAGE_SHIFT - 12;
2355 		} else {
2356 			init_hca->log_uar_sz = ilog2(dev->caps.num_uars);
2357 			init_hca->uar_page_sz = PAGE_SHIFT - 12;
2358 		}
2359 
2360 		init_hca->mw_enabled = 0;
2361 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2362 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2363 			init_hca->mw_enabled = INIT_HCA_TPT_MW_ENABLE;
2364 
2365 		err = mlx4_init_icm(dev, dev_cap, init_hca, icm_size);
2366 		if (err)
2367 			goto out_free;
2368 
2369 		err = mlx4_INIT_HCA(dev, init_hca);
2370 		if (err) {
2371 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
2372 			goto err_free_icm;
2373 		}
2374 
2375 		if (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2376 			err = mlx4_query_func(dev, dev_cap);
2377 			if (err < 0) {
2378 				mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
2379 				goto err_close;
2380 			} else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2381 				dev->caps.num_eqs = dev_cap->max_eqs;
2382 				dev->caps.reserved_eqs = dev_cap->reserved_eqs;
2383 				dev->caps.reserved_uars = dev_cap->reserved_uars;
2384 			}
2385 		}
2386 
2387 		/*
2388 		 * If TS is supported by FW
2389 		 * read HCA frequency by QUERY_HCA command
2390 		 */
2391 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2392 			err = mlx4_QUERY_HCA(dev, init_hca);
2393 			if (err) {
2394 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
2395 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2396 			} else {
2397 				dev->caps.hca_core_clock =
2398 					init_hca->hca_core_clock;
2399 			}
2400 
2401 			/* In case we got HCA frequency 0 - disable timestamping
2402 			 * to avoid dividing by zero
2403 			 */
2404 			if (!dev->caps.hca_core_clock) {
2405 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2406 				mlx4_err(dev,
2407 					 "HCA frequency is 0 - timestamping is not supported\n");
2408 			} else if (map_internal_clock(dev)) {
2409 				/*
2410 				 * Map internal clock,
2411 				 * in case of failure disable timestamping
2412 				 */
2413 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2414 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
2415 			}
2416 		}
2417 
2418 		if (dev->caps.dmfs_high_steer_mode !=
2419 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2420 			if (mlx4_validate_optimized_steering(dev))
2421 				mlx4_warn(dev, "Optimized steering validation failed\n");
2422 
2423 			if (dev->caps.dmfs_high_steer_mode ==
2424 			    MLX4_STEERING_DMFS_A0_DISABLE) {
2425 				dev->caps.dmfs_high_rate_qpn_base =
2426 					dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2427 				dev->caps.dmfs_high_rate_qpn_range =
2428 					MLX4_A0_STEERING_TABLE_SIZE;
2429 			}
2430 
2431 			mlx4_info(dev, "DMFS high rate steer mode is: %s\n",
2432 				  dmfs_high_rate_steering_mode_str(
2433 					dev->caps.dmfs_high_steer_mode));
2434 		}
2435 	} else {
2436 		err = mlx4_init_slave(dev);
2437 		if (err) {
2438 			if (err != -EPROBE_DEFER)
2439 				mlx4_err(dev, "Failed to initialize slave\n");
2440 			return err;
2441 		}
2442 
2443 		err = mlx4_slave_cap(dev);
2444 		if (err) {
2445 			mlx4_err(dev, "Failed to obtain slave caps\n");
2446 			goto err_close;
2447 		}
2448 	}
2449 
2450 	if (map_bf_area(dev))
2451 		mlx4_dbg(dev, "Failed to map blue flame area\n");
2452 
2453 	/*Only the master set the ports, all the rest got it from it.*/
2454 	if (!mlx4_is_slave(dev))
2455 		mlx4_set_port_mask(dev);
2456 
2457 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
2458 	if (err) {
2459 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2460 		goto unmap_bf;
2461 	}
2462 
2463 	/* Query CONFIG_DEV parameters */
2464 	err = mlx4_config_dev_retrieval(dev, &params);
2465 	if (err && err != -EOPNOTSUPP) {
2466 		mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2467 	} else if (!err) {
2468 		dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2469 		dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2470 	}
2471 	priv->eq_table.inta_pin = adapter.inta_pin;
2472 	memcpy(dev->board_id, adapter.board_id, sizeof(dev->board_id));
2473 
2474 	err = 0;
2475 	goto out_free;
2476 
2477 unmap_bf:
2478 	unmap_internal_clock(dev);
2479 	unmap_bf_area(dev);
2480 
2481 	if (mlx4_is_slave(dev))
2482 		mlx4_slave_destroy_special_qp_cap(dev);
2483 
2484 err_close:
2485 	if (mlx4_is_slave(dev))
2486 		mlx4_slave_exit(dev);
2487 	else
2488 		mlx4_CLOSE_HCA(dev, 0);
2489 
2490 err_free_icm:
2491 	if (!mlx4_is_slave(dev))
2492 		mlx4_free_icms(dev);
2493 
2494 out_free:
2495 	kfree(dev_cap);
2496 	kfree(init_hca);
2497 
2498 	return err;
2499 }
2500 
2501 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2502 {
2503 	struct mlx4_priv *priv = mlx4_priv(dev);
2504 	int nent_pow2;
2505 
2506 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2507 		return -ENOENT;
2508 
2509 	if (!dev->caps.max_counters)
2510 		return -ENOSPC;
2511 
2512 	nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2513 	/* reserve last counter index for sink counter */
2514 	return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2515 				nent_pow2 - 1, 0,
2516 				nent_pow2 - dev->caps.max_counters + 1);
2517 }
2518 
2519 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2520 {
2521 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2522 		return;
2523 
2524 	if (!dev->caps.max_counters)
2525 		return;
2526 
2527 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2528 }
2529 
2530 static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2531 {
2532 	struct mlx4_priv *priv = mlx4_priv(dev);
2533 	int port;
2534 
2535 	for (port = 0; port < dev->caps.num_ports; port++)
2536 		if (priv->def_counter[port] != -1)
2537 			mlx4_counter_free(dev,  priv->def_counter[port]);
2538 }
2539 
2540 static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2541 {
2542 	struct mlx4_priv *priv = mlx4_priv(dev);
2543 	int port, err = 0;
2544 	u32 idx;
2545 
2546 	for (port = 0; port < dev->caps.num_ports; port++)
2547 		priv->def_counter[port] = -1;
2548 
2549 	for (port = 0; port < dev->caps.num_ports; port++) {
2550 		err = mlx4_counter_alloc(dev, &idx, MLX4_RES_USAGE_DRIVER);
2551 
2552 		if (!err || err == -ENOSPC) {
2553 			priv->def_counter[port] = idx;
2554 		} else if (err == -ENOENT) {
2555 			err = 0;
2556 			continue;
2557 		} else if (mlx4_is_slave(dev) && err == -EINVAL) {
2558 			priv->def_counter[port] = MLX4_SINK_COUNTER_INDEX(dev);
2559 			mlx4_warn(dev, "can't allocate counter from old PF driver, using index %d\n",
2560 				  MLX4_SINK_COUNTER_INDEX(dev));
2561 			err = 0;
2562 		} else {
2563 			mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2564 				 __func__, port + 1, err);
2565 			mlx4_cleanup_default_counters(dev);
2566 			return err;
2567 		}
2568 
2569 		mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2570 			 __func__, priv->def_counter[port], port + 1);
2571 	}
2572 
2573 	return err;
2574 }
2575 
2576 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2577 {
2578 	struct mlx4_priv *priv = mlx4_priv(dev);
2579 
2580 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2581 		return -ENOENT;
2582 
2583 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2584 	if (*idx == -1) {
2585 		*idx = MLX4_SINK_COUNTER_INDEX(dev);
2586 		return -ENOSPC;
2587 	}
2588 
2589 	return 0;
2590 }
2591 
2592 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx, u8 usage)
2593 {
2594 	u32 in_modifier = RES_COUNTER | (((u32)usage & 3) << 30);
2595 	u64 out_param;
2596 	int err;
2597 
2598 	if (mlx4_is_mfunc(dev)) {
2599 		err = mlx4_cmd_imm(dev, 0, &out_param, in_modifier,
2600 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2601 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2602 		if (!err)
2603 			*idx = get_param_l(&out_param);
2604 
2605 		return err;
2606 	}
2607 	return __mlx4_counter_alloc(dev, idx);
2608 }
2609 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2610 
2611 static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2612 				u8 counter_index)
2613 {
2614 	struct mlx4_cmd_mailbox *if_stat_mailbox;
2615 	int err;
2616 	u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2617 
2618 	if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2619 	if (IS_ERR(if_stat_mailbox))
2620 		return PTR_ERR(if_stat_mailbox);
2621 
2622 	err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2623 			   MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2624 			   MLX4_CMD_NATIVE);
2625 
2626 	mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2627 	return err;
2628 }
2629 
2630 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2631 {
2632 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2633 		return;
2634 
2635 	if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2636 		return;
2637 
2638 	__mlx4_clear_if_stat(dev, idx);
2639 
2640 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2641 	return;
2642 }
2643 
2644 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2645 {
2646 	u64 in_param = 0;
2647 
2648 	if (mlx4_is_mfunc(dev)) {
2649 		set_param_l(&in_param, idx);
2650 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2651 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2652 			 MLX4_CMD_WRAPPED);
2653 		return;
2654 	}
2655 	__mlx4_counter_free(dev, idx);
2656 }
2657 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2658 
2659 int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2660 {
2661 	struct mlx4_priv *priv = mlx4_priv(dev);
2662 
2663 	return priv->def_counter[port - 1];
2664 }
2665 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2666 
2667 void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2668 {
2669 	struct mlx4_priv *priv = mlx4_priv(dev);
2670 
2671 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2672 }
2673 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2674 
2675 __be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2676 {
2677 	struct mlx4_priv *priv = mlx4_priv(dev);
2678 
2679 	return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2680 }
2681 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2682 
2683 void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2684 {
2685 	struct mlx4_priv *priv = mlx4_priv(dev);
2686 	__be64 guid;
2687 
2688 	/* hw GUID */
2689 	if (entry == 0)
2690 		return;
2691 
2692 	get_random_bytes((char *)&guid, sizeof(guid));
2693 	guid &= ~(cpu_to_be64(1ULL << 56));
2694 	guid |= cpu_to_be64(1ULL << 57);
2695 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2696 }
2697 
2698 static int mlx4_setup_hca(struct mlx4_dev *dev)
2699 {
2700 	struct mlx4_priv *priv = mlx4_priv(dev);
2701 	int err;
2702 	int port;
2703 	__be32 ib_port_default_caps;
2704 
2705 	err = mlx4_init_uar_table(dev);
2706 	if (err) {
2707 		mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2708 		return err;
2709 	}
2710 
2711 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
2712 	if (err) {
2713 		mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2714 		goto err_uar_table_free;
2715 	}
2716 
2717 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2718 	if (!priv->kar) {
2719 		mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2720 		err = -ENOMEM;
2721 		goto err_uar_free;
2722 	}
2723 
2724 	err = mlx4_init_pd_table(dev);
2725 	if (err) {
2726 		mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2727 		goto err_kar_unmap;
2728 	}
2729 
2730 	err = mlx4_init_xrcd_table(dev);
2731 	if (err) {
2732 		mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2733 		goto err_pd_table_free;
2734 	}
2735 
2736 	err = mlx4_init_mr_table(dev);
2737 	if (err) {
2738 		mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2739 		goto err_xrcd_table_free;
2740 	}
2741 
2742 	if (!mlx4_is_slave(dev)) {
2743 		err = mlx4_init_mcg_table(dev);
2744 		if (err) {
2745 			mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2746 			goto err_mr_table_free;
2747 		}
2748 		err = mlx4_config_mad_demux(dev);
2749 		if (err) {
2750 			mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2751 			goto err_mcg_table_free;
2752 		}
2753 	}
2754 
2755 	err = mlx4_init_eq_table(dev);
2756 	if (err) {
2757 		mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2758 		goto err_mcg_table_free;
2759 	}
2760 
2761 	err = mlx4_cmd_use_events(dev);
2762 	if (err) {
2763 		mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2764 		goto err_eq_table_free;
2765 	}
2766 
2767 	err = mlx4_NOP(dev);
2768 	if (err) {
2769 		if (dev->flags & MLX4_FLAG_MSI_X) {
2770 			mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2771 				  priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2772 			mlx4_warn(dev, "Trying again without MSI-X\n");
2773 		} else {
2774 			mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2775 				 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2776 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2777 		}
2778 
2779 		goto err_cmd_poll;
2780 	}
2781 
2782 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
2783 
2784 	err = mlx4_init_cq_table(dev);
2785 	if (err) {
2786 		mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2787 		goto err_cmd_poll;
2788 	}
2789 
2790 	err = mlx4_init_srq_table(dev);
2791 	if (err) {
2792 		mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2793 		goto err_cq_table_free;
2794 	}
2795 
2796 	err = mlx4_init_qp_table(dev);
2797 	if (err) {
2798 		mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2799 		goto err_srq_table_free;
2800 	}
2801 
2802 	if (!mlx4_is_slave(dev)) {
2803 		err = mlx4_init_counters_table(dev);
2804 		if (err && err != -ENOENT) {
2805 			mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2806 			goto err_qp_table_free;
2807 		}
2808 	}
2809 
2810 	err = mlx4_allocate_default_counters(dev);
2811 	if (err) {
2812 		mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2813 		goto err_counters_table_free;
2814 	}
2815 
2816 	if (!mlx4_is_slave(dev)) {
2817 		for (port = 1; port <= dev->caps.num_ports; port++) {
2818 			ib_port_default_caps = 0;
2819 			err = mlx4_get_port_ib_caps(dev, port,
2820 						    &ib_port_default_caps);
2821 			if (err)
2822 				mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2823 					  port, err);
2824 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2825 
2826 			/* initialize per-slave default ib port capabilities */
2827 			if (mlx4_is_master(dev)) {
2828 				int i;
2829 				for (i = 0; i < dev->num_slaves; i++) {
2830 					if (i == mlx4_master_func_num(dev))
2831 						continue;
2832 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2833 						ib_port_default_caps;
2834 				}
2835 			}
2836 
2837 			if (mlx4_is_mfunc(dev))
2838 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2839 			else
2840 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2841 
2842 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2843 					    dev->caps.pkey_table_len[port] : -1);
2844 			if (err) {
2845 				mlx4_err(dev, "Failed to set port %d, aborting\n",
2846 					 port);
2847 				goto err_default_countes_free;
2848 			}
2849 		}
2850 	}
2851 
2852 	return 0;
2853 
2854 err_default_countes_free:
2855 	mlx4_cleanup_default_counters(dev);
2856 
2857 err_counters_table_free:
2858 	if (!mlx4_is_slave(dev))
2859 		mlx4_cleanup_counters_table(dev);
2860 
2861 err_qp_table_free:
2862 	mlx4_cleanup_qp_table(dev);
2863 
2864 err_srq_table_free:
2865 	mlx4_cleanup_srq_table(dev);
2866 
2867 err_cq_table_free:
2868 	mlx4_cleanup_cq_table(dev);
2869 
2870 err_cmd_poll:
2871 	mlx4_cmd_use_polling(dev);
2872 
2873 err_eq_table_free:
2874 	mlx4_cleanup_eq_table(dev);
2875 
2876 err_mcg_table_free:
2877 	if (!mlx4_is_slave(dev))
2878 		mlx4_cleanup_mcg_table(dev);
2879 
2880 err_mr_table_free:
2881 	mlx4_cleanup_mr_table(dev);
2882 
2883 err_xrcd_table_free:
2884 	mlx4_cleanup_xrcd_table(dev);
2885 
2886 err_pd_table_free:
2887 	mlx4_cleanup_pd_table(dev);
2888 
2889 err_kar_unmap:
2890 	iounmap(priv->kar);
2891 
2892 err_uar_free:
2893 	mlx4_uar_free(dev, &priv->driver_uar);
2894 
2895 err_uar_table_free:
2896 	mlx4_cleanup_uar_table(dev);
2897 	return err;
2898 }
2899 
2900 static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2901 {
2902 	int requested_cpu = 0;
2903 	struct mlx4_priv *priv = mlx4_priv(dev);
2904 	struct mlx4_eq *eq;
2905 	int off = 0;
2906 	int i;
2907 
2908 	if (eqn > dev->caps.num_comp_vectors)
2909 		return -EINVAL;
2910 
2911 	for (i = 1; i < port; i++)
2912 		off += mlx4_get_eqs_per_port(dev, i);
2913 
2914 	requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2915 
2916 	/* Meaning EQs are shared, and this call comes from the second port */
2917 	if (requested_cpu < 0)
2918 		return 0;
2919 
2920 	eq = &priv->eq_table.eq[eqn];
2921 
2922 	if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2923 		return -ENOMEM;
2924 
2925 	cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2926 
2927 	return 0;
2928 }
2929 
2930 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2931 {
2932 	struct mlx4_priv *priv = mlx4_priv(dev);
2933 	struct msix_entry *entries;
2934 	int i;
2935 	int port = 0;
2936 
2937 	if (msi_x) {
2938 		int nreq = min3(dev->caps.num_ports *
2939 				(int)num_online_cpus() + 1,
2940 				dev->caps.num_eqs - dev->caps.reserved_eqs,
2941 				MAX_MSIX);
2942 
2943 		if (msi_x > 1)
2944 			nreq = min_t(int, nreq, msi_x);
2945 
2946 		entries = kcalloc(nreq, sizeof(*entries), GFP_KERNEL);
2947 		if (!entries)
2948 			goto no_msi;
2949 
2950 		for (i = 0; i < nreq; ++i)
2951 			entries[i].entry = i;
2952 
2953 		nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2954 					     nreq);
2955 
2956 		if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
2957 			kfree(entries);
2958 			goto no_msi;
2959 		}
2960 		/* 1 is reserved for events (asyncrounous EQ) */
2961 		dev->caps.num_comp_vectors = nreq - 1;
2962 
2963 		priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2964 		bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2965 			    dev->caps.num_ports);
2966 
2967 		for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2968 			if (i == MLX4_EQ_ASYNC)
2969 				continue;
2970 
2971 			priv->eq_table.eq[i].irq =
2972 				entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2973 
2974 			if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2975 				bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2976 					    dev->caps.num_ports);
2977 				/* We don't set affinity hint when there
2978 				 * aren't enough EQs
2979 				 */
2980 			} else {
2981 				set_bit(port,
2982 					priv->eq_table.eq[i].actv_ports.ports);
2983 				if (mlx4_init_affinity_hint(dev, port + 1, i))
2984 					mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
2985 						  i);
2986 			}
2987 			/* We divide the Eqs evenly between the two ports.
2988 			 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
2989 			 * refers to the number of Eqs per port
2990 			 * (i.e eqs_per_port). Theoretically, we would like to
2991 			 * write something like (i + 1) % eqs_per_port == 0.
2992 			 * However, since there's an asynchronous Eq, we have
2993 			 * to skip over it by comparing this condition to
2994 			 * !!((i + 1) > MLX4_EQ_ASYNC).
2995 			 */
2996 			if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
2997 			    ((i + 1) %
2998 			     (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
2999 			    !!((i + 1) > MLX4_EQ_ASYNC))
3000 				/* If dev->caps.num_comp_vectors < dev->caps.num_ports,
3001 				 * everything is shared anyway.
3002 				 */
3003 				port++;
3004 		}
3005 
3006 		dev->flags |= MLX4_FLAG_MSI_X;
3007 
3008 		kfree(entries);
3009 		return;
3010 	}
3011 
3012 no_msi:
3013 	dev->caps.num_comp_vectors = 1;
3014 
3015 	BUG_ON(MLX4_EQ_ASYNC >= 2);
3016 	for (i = 0; i < 2; ++i) {
3017 		priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
3018 		if (i != MLX4_EQ_ASYNC) {
3019 			bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
3020 				    dev->caps.num_ports);
3021 		}
3022 	}
3023 }
3024 
3025 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
3026 {
3027 	struct devlink *devlink = priv_to_devlink(mlx4_priv(dev));
3028 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
3029 	int err;
3030 
3031 	err = devlink_port_register(devlink, &info->devlink_port, port);
3032 	if (err)
3033 		return err;
3034 
3035 	info->dev = dev;
3036 	info->port = port;
3037 	if (!mlx4_is_slave(dev)) {
3038 		mlx4_init_mac_table(dev, &info->mac_table);
3039 		mlx4_init_vlan_table(dev, &info->vlan_table);
3040 		mlx4_init_roce_gid_table(dev, &info->gid_table);
3041 		info->base_qpn = mlx4_get_base_qpn(dev, port);
3042 	}
3043 
3044 	sprintf(info->dev_name, "mlx4_port%d", port);
3045 	info->port_attr.attr.name = info->dev_name;
3046 	if (mlx4_is_mfunc(dev)) {
3047 		info->port_attr.attr.mode = 0444;
3048 	} else {
3049 		info->port_attr.attr.mode = 0644;
3050 		info->port_attr.store     = set_port_type;
3051 	}
3052 	info->port_attr.show      = show_port_type;
3053 	sysfs_attr_init(&info->port_attr.attr);
3054 
3055 	err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
3056 	if (err) {
3057 		mlx4_err(dev, "Failed to create file for port %d\n", port);
3058 		devlink_port_unregister(&info->devlink_port);
3059 		info->port = -1;
3060 		return err;
3061 	}
3062 
3063 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
3064 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
3065 	if (mlx4_is_mfunc(dev)) {
3066 		info->port_mtu_attr.attr.mode = 0444;
3067 	} else {
3068 		info->port_mtu_attr.attr.mode = 0644;
3069 		info->port_mtu_attr.store     = set_port_ib_mtu;
3070 	}
3071 	info->port_mtu_attr.show      = show_port_ib_mtu;
3072 	sysfs_attr_init(&info->port_mtu_attr.attr);
3073 
3074 	err = device_create_file(&dev->persist->pdev->dev,
3075 				 &info->port_mtu_attr);
3076 	if (err) {
3077 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
3078 		device_remove_file(&info->dev->persist->pdev->dev,
3079 				   &info->port_attr);
3080 		devlink_port_unregister(&info->devlink_port);
3081 		info->port = -1;
3082 		return err;
3083 	}
3084 
3085 	return 0;
3086 }
3087 
3088 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
3089 {
3090 	if (info->port < 0)
3091 		return;
3092 
3093 	device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
3094 	device_remove_file(&info->dev->persist->pdev->dev,
3095 			   &info->port_mtu_attr);
3096 	devlink_port_unregister(&info->devlink_port);
3097 
3098 #ifdef CONFIG_RFS_ACCEL
3099 	free_irq_cpu_rmap(info->rmap);
3100 	info->rmap = NULL;
3101 #endif
3102 }
3103 
3104 static int mlx4_init_steering(struct mlx4_dev *dev)
3105 {
3106 	struct mlx4_priv *priv = mlx4_priv(dev);
3107 	int num_entries = dev->caps.num_ports;
3108 	int i, j;
3109 
3110 	priv->steer = kcalloc(num_entries, sizeof(struct mlx4_steer),
3111 			      GFP_KERNEL);
3112 	if (!priv->steer)
3113 		return -ENOMEM;
3114 
3115 	for (i = 0; i < num_entries; i++)
3116 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3117 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
3118 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
3119 		}
3120 	return 0;
3121 }
3122 
3123 static void mlx4_clear_steering(struct mlx4_dev *dev)
3124 {
3125 	struct mlx4_priv *priv = mlx4_priv(dev);
3126 	struct mlx4_steer_index *entry, *tmp_entry;
3127 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
3128 	int num_entries = dev->caps.num_ports;
3129 	int i, j;
3130 
3131 	for (i = 0; i < num_entries; i++) {
3132 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3133 			list_for_each_entry_safe(pqp, tmp_pqp,
3134 						 &priv->steer[i].promisc_qps[j],
3135 						 list) {
3136 				list_del(&pqp->list);
3137 				kfree(pqp);
3138 			}
3139 			list_for_each_entry_safe(entry, tmp_entry,
3140 						 &priv->steer[i].steer_entries[j],
3141 						 list) {
3142 				list_del(&entry->list);
3143 				list_for_each_entry_safe(pqp, tmp_pqp,
3144 							 &entry->duplicates,
3145 							 list) {
3146 					list_del(&pqp->list);
3147 					kfree(pqp);
3148 				}
3149 				kfree(entry);
3150 			}
3151 		}
3152 	}
3153 	kfree(priv->steer);
3154 }
3155 
3156 static int extended_func_num(struct pci_dev *pdev)
3157 {
3158 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
3159 }
3160 
3161 #define MLX4_OWNER_BASE	0x8069c
3162 #define MLX4_OWNER_SIZE	4
3163 
3164 static int mlx4_get_ownership(struct mlx4_dev *dev)
3165 {
3166 	void __iomem *owner;
3167 	u32 ret;
3168 
3169 	if (pci_channel_offline(dev->persist->pdev))
3170 		return -EIO;
3171 
3172 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3173 			MLX4_OWNER_BASE,
3174 			MLX4_OWNER_SIZE);
3175 	if (!owner) {
3176 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3177 		return -ENOMEM;
3178 	}
3179 
3180 	ret = readl(owner);
3181 	iounmap(owner);
3182 	return (int) !!ret;
3183 }
3184 
3185 static void mlx4_free_ownership(struct mlx4_dev *dev)
3186 {
3187 	void __iomem *owner;
3188 
3189 	if (pci_channel_offline(dev->persist->pdev))
3190 		return;
3191 
3192 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3193 			MLX4_OWNER_BASE,
3194 			MLX4_OWNER_SIZE);
3195 	if (!owner) {
3196 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3197 		return;
3198 	}
3199 	writel(0, owner);
3200 	msleep(1000);
3201 	iounmap(owner);
3202 }
3203 
3204 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV)	==\
3205 				  !!((flags) & MLX4_FLAG_MASTER))
3206 
3207 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
3208 			     u8 total_vfs, int existing_vfs, int reset_flow)
3209 {
3210 	u64 dev_flags = dev->flags;
3211 	int err = 0;
3212 	int fw_enabled_sriov_vfs = min(pci_sriov_get_totalvfs(pdev),
3213 					MLX4_MAX_NUM_VF);
3214 
3215 	if (reset_flow) {
3216 		dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
3217 				       GFP_KERNEL);
3218 		if (!dev->dev_vfs)
3219 			goto free_mem;
3220 		return dev_flags;
3221 	}
3222 
3223 	atomic_inc(&pf_loading);
3224 	if (dev->flags &  MLX4_FLAG_SRIOV) {
3225 		if (existing_vfs != total_vfs) {
3226 			mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
3227 				 existing_vfs, total_vfs);
3228 			total_vfs = existing_vfs;
3229 		}
3230 	}
3231 
3232 	dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs), GFP_KERNEL);
3233 	if (NULL == dev->dev_vfs) {
3234 		mlx4_err(dev, "Failed to allocate memory for VFs\n");
3235 		goto disable_sriov;
3236 	}
3237 
3238 	if (!(dev->flags &  MLX4_FLAG_SRIOV)) {
3239 		if (total_vfs > fw_enabled_sriov_vfs) {
3240 			mlx4_err(dev, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
3241 				 total_vfs, fw_enabled_sriov_vfs);
3242 			err = -ENOMEM;
3243 			goto disable_sriov;
3244 		}
3245 		mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
3246 		err = pci_enable_sriov(pdev, total_vfs);
3247 	}
3248 	if (err) {
3249 		mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
3250 			 err);
3251 		goto disable_sriov;
3252 	} else {
3253 		mlx4_warn(dev, "Running in master mode\n");
3254 		dev_flags |= MLX4_FLAG_SRIOV |
3255 			MLX4_FLAG_MASTER;
3256 		dev_flags &= ~MLX4_FLAG_SLAVE;
3257 		dev->persist->num_vfs = total_vfs;
3258 	}
3259 	return dev_flags;
3260 
3261 disable_sriov:
3262 	atomic_dec(&pf_loading);
3263 free_mem:
3264 	dev->persist->num_vfs = 0;
3265 	kfree(dev->dev_vfs);
3266 	dev->dev_vfs = NULL;
3267 	return dev_flags & ~MLX4_FLAG_MASTER;
3268 }
3269 
3270 enum {
3271 	MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
3272 };
3273 
3274 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
3275 			      int *nvfs)
3276 {
3277 	int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
3278 	/* Checking for 64 VFs as a limitation of CX2 */
3279 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
3280 	    requested_vfs >= 64) {
3281 		mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
3282 			 requested_vfs);
3283 		return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
3284 	}
3285 	return 0;
3286 }
3287 
3288 static int mlx4_pci_enable_device(struct mlx4_dev *dev)
3289 {
3290 	struct pci_dev *pdev = dev->persist->pdev;
3291 	int err = 0;
3292 
3293 	mutex_lock(&dev->persist->pci_status_mutex);
3294 	if (dev->persist->pci_status == MLX4_PCI_STATUS_DISABLED) {
3295 		err = pci_enable_device(pdev);
3296 		if (!err)
3297 			dev->persist->pci_status = MLX4_PCI_STATUS_ENABLED;
3298 	}
3299 	mutex_unlock(&dev->persist->pci_status_mutex);
3300 
3301 	return err;
3302 }
3303 
3304 static void mlx4_pci_disable_device(struct mlx4_dev *dev)
3305 {
3306 	struct pci_dev *pdev = dev->persist->pdev;
3307 
3308 	mutex_lock(&dev->persist->pci_status_mutex);
3309 	if (dev->persist->pci_status == MLX4_PCI_STATUS_ENABLED) {
3310 		pci_disable_device(pdev);
3311 		dev->persist->pci_status = MLX4_PCI_STATUS_DISABLED;
3312 	}
3313 	mutex_unlock(&dev->persist->pci_status_mutex);
3314 }
3315 
3316 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
3317 			 int total_vfs, int *nvfs, struct mlx4_priv *priv,
3318 			 int reset_flow)
3319 {
3320 	struct mlx4_dev *dev;
3321 	unsigned sum = 0;
3322 	int err;
3323 	int port;
3324 	int i;
3325 	struct mlx4_dev_cap *dev_cap = NULL;
3326 	int existing_vfs = 0;
3327 
3328 	dev = &priv->dev;
3329 
3330 	INIT_LIST_HEAD(&priv->ctx_list);
3331 	spin_lock_init(&priv->ctx_lock);
3332 
3333 	mutex_init(&priv->port_mutex);
3334 	mutex_init(&priv->bond_mutex);
3335 
3336 	INIT_LIST_HEAD(&priv->pgdir_list);
3337 	mutex_init(&priv->pgdir_mutex);
3338 	spin_lock_init(&priv->cmd.context_lock);
3339 
3340 	INIT_LIST_HEAD(&priv->bf_list);
3341 	mutex_init(&priv->bf_mutex);
3342 
3343 	dev->rev_id = pdev->revision;
3344 	dev->numa_node = dev_to_node(&pdev->dev);
3345 
3346 	/* Detect if this device is a virtual function */
3347 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3348 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3349 		dev->flags |= MLX4_FLAG_SLAVE;
3350 	} else {
3351 		/* We reset the device and enable SRIOV only for physical
3352 		 * devices.  Try to claim ownership on the device;
3353 		 * if already taken, skip -- do not allow multiple PFs */
3354 		err = mlx4_get_ownership(dev);
3355 		if (err) {
3356 			if (err < 0)
3357 				return err;
3358 			else {
3359 				mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
3360 				return -EINVAL;
3361 			}
3362 		}
3363 
3364 		atomic_set(&priv->opreq_count, 0);
3365 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3366 
3367 		/*
3368 		 * Now reset the HCA before we touch the PCI capabilities or
3369 		 * attempt a firmware command, since a boot ROM may have left
3370 		 * the HCA in an undefined state.
3371 		 */
3372 		err = mlx4_reset(dev);
3373 		if (err) {
3374 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
3375 			goto err_sriov;
3376 		}
3377 
3378 		if (total_vfs) {
3379 			dev->flags = MLX4_FLAG_MASTER;
3380 			existing_vfs = pci_num_vf(pdev);
3381 			if (existing_vfs)
3382 				dev->flags |= MLX4_FLAG_SRIOV;
3383 			dev->persist->num_vfs = total_vfs;
3384 		}
3385 	}
3386 
3387 	/* on load remove any previous indication of internal error,
3388 	 * device is up.
3389 	 */
3390 	dev->persist->state = MLX4_DEVICE_STATE_UP;
3391 
3392 slave_start:
3393 	err = mlx4_cmd_init(dev);
3394 	if (err) {
3395 		mlx4_err(dev, "Failed to init command interface, aborting\n");
3396 		goto err_sriov;
3397 	}
3398 
3399 	/* In slave functions, the communication channel must be initialized
3400 	 * before posting commands. Also, init num_slaves before calling
3401 	 * mlx4_init_hca */
3402 	if (mlx4_is_mfunc(dev)) {
3403 		if (mlx4_is_master(dev)) {
3404 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
3405 
3406 		} else {
3407 			dev->num_slaves = 0;
3408 			err = mlx4_multi_func_init(dev);
3409 			if (err) {
3410 				mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
3411 				goto err_cmd;
3412 			}
3413 		}
3414 	}
3415 
3416 	err = mlx4_init_fw(dev);
3417 	if (err) {
3418 		mlx4_err(dev, "Failed to init fw, aborting.\n");
3419 		goto err_mfunc;
3420 	}
3421 
3422 	if (mlx4_is_master(dev)) {
3423 		/* when we hit the goto slave_start below, dev_cap already initialized */
3424 		if (!dev_cap) {
3425 			dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3426 
3427 			if (!dev_cap) {
3428 				err = -ENOMEM;
3429 				goto err_fw;
3430 			}
3431 
3432 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3433 			if (err) {
3434 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3435 				goto err_fw;
3436 			}
3437 
3438 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3439 				goto err_fw;
3440 
3441 			if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3442 				u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3443 								  total_vfs,
3444 								  existing_vfs,
3445 								  reset_flow);
3446 
3447 				mlx4_close_fw(dev);
3448 				mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3449 				dev->flags = dev_flags;
3450 				if (!SRIOV_VALID_STATE(dev->flags)) {
3451 					mlx4_err(dev, "Invalid SRIOV state\n");
3452 					goto err_sriov;
3453 				}
3454 				err = mlx4_reset(dev);
3455 				if (err) {
3456 					mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3457 					goto err_sriov;
3458 				}
3459 				goto slave_start;
3460 			}
3461 		} else {
3462 			/* Legacy mode FW requires SRIOV to be enabled before
3463 			 * doing QUERY_DEV_CAP, since max_eq's value is different if
3464 			 * SRIOV is enabled.
3465 			 */
3466 			memset(dev_cap, 0, sizeof(*dev_cap));
3467 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3468 			if (err) {
3469 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3470 				goto err_fw;
3471 			}
3472 
3473 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3474 				goto err_fw;
3475 		}
3476 	}
3477 
3478 	err = mlx4_init_hca(dev);
3479 	if (err) {
3480 		if (err == -EACCES) {
3481 			/* Not primary Physical function
3482 			 * Running in slave mode */
3483 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3484 			/* We're not a PF */
3485 			if (dev->flags & MLX4_FLAG_SRIOV) {
3486 				if (!existing_vfs)
3487 					pci_disable_sriov(pdev);
3488 				if (mlx4_is_master(dev) && !reset_flow)
3489 					atomic_dec(&pf_loading);
3490 				dev->flags &= ~MLX4_FLAG_SRIOV;
3491 			}
3492 			if (!mlx4_is_slave(dev))
3493 				mlx4_free_ownership(dev);
3494 			dev->flags |= MLX4_FLAG_SLAVE;
3495 			dev->flags &= ~MLX4_FLAG_MASTER;
3496 			goto slave_start;
3497 		} else
3498 			goto err_fw;
3499 	}
3500 
3501 	if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3502 		u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3503 						  existing_vfs, reset_flow);
3504 
3505 		if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3506 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3507 			dev->flags = dev_flags;
3508 			err = mlx4_cmd_init(dev);
3509 			if (err) {
3510 				/* Only VHCR is cleaned up, so could still
3511 				 * send FW commands
3512 				 */
3513 				mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3514 				goto err_close;
3515 			}
3516 		} else {
3517 			dev->flags = dev_flags;
3518 		}
3519 
3520 		if (!SRIOV_VALID_STATE(dev->flags)) {
3521 			mlx4_err(dev, "Invalid SRIOV state\n");
3522 			goto err_close;
3523 		}
3524 	}
3525 
3526 	/* check if the device is functioning at its maximum possible speed.
3527 	 * No return code for this call, just warn the user in case of PCI
3528 	 * express device capabilities are under-satisfied by the bus.
3529 	 */
3530 	if (!mlx4_is_slave(dev))
3531 		pcie_print_link_status(dev->persist->pdev);
3532 
3533 	/* In master functions, the communication channel must be initialized
3534 	 * after obtaining its address from fw */
3535 	if (mlx4_is_master(dev)) {
3536 		if (dev->caps.num_ports < 2 &&
3537 		    num_vfs_argc > 1) {
3538 			err = -EINVAL;
3539 			mlx4_err(dev,
3540 				 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3541 				 dev->caps.num_ports);
3542 			goto err_close;
3543 		}
3544 		memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
3545 
3546 		for (i = 0;
3547 		     i < sizeof(dev->persist->nvfs)/
3548 		     sizeof(dev->persist->nvfs[0]); i++) {
3549 			unsigned j;
3550 
3551 			for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
3552 				dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3553 				dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3554 					dev->caps.num_ports;
3555 			}
3556 		}
3557 
3558 		/* In master functions, the communication channel
3559 		 * must be initialized after obtaining its address from fw
3560 		 */
3561 		err = mlx4_multi_func_init(dev);
3562 		if (err) {
3563 			mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3564 			goto err_close;
3565 		}
3566 	}
3567 
3568 	err = mlx4_alloc_eq_table(dev);
3569 	if (err)
3570 		goto err_master_mfunc;
3571 
3572 	bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
3573 	mutex_init(&priv->msix_ctl.pool_lock);
3574 
3575 	mlx4_enable_msi_x(dev);
3576 	if ((mlx4_is_mfunc(dev)) &&
3577 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
3578 		err = -EOPNOTSUPP;
3579 		mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
3580 		goto err_free_eq;
3581 	}
3582 
3583 	if (!mlx4_is_slave(dev)) {
3584 		err = mlx4_init_steering(dev);
3585 		if (err)
3586 			goto err_disable_msix;
3587 	}
3588 
3589 	mlx4_init_quotas(dev);
3590 
3591 	err = mlx4_setup_hca(dev);
3592 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3593 	    !mlx4_is_mfunc(dev)) {
3594 		dev->flags &= ~MLX4_FLAG_MSI_X;
3595 		dev->caps.num_comp_vectors = 1;
3596 		pci_disable_msix(pdev);
3597 		err = mlx4_setup_hca(dev);
3598 	}
3599 
3600 	if (err)
3601 		goto err_steer;
3602 
3603 	/* When PF resources are ready arm its comm channel to enable
3604 	 * getting commands
3605 	 */
3606 	if (mlx4_is_master(dev)) {
3607 		err = mlx4_ARM_COMM_CHANNEL(dev);
3608 		if (err) {
3609 			mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3610 				 err);
3611 			goto err_steer;
3612 		}
3613 	}
3614 
3615 	for (port = 1; port <= dev->caps.num_ports; port++) {
3616 		err = mlx4_init_port_info(dev, port);
3617 		if (err)
3618 			goto err_port;
3619 	}
3620 
3621 	priv->v2p.port1 = 1;
3622 	priv->v2p.port2 = 2;
3623 
3624 	err = mlx4_register_device(dev);
3625 	if (err)
3626 		goto err_port;
3627 
3628 	mlx4_request_modules(dev);
3629 
3630 	mlx4_sense_init(dev);
3631 	mlx4_start_sense(dev);
3632 
3633 	priv->removed = 0;
3634 
3635 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3636 		atomic_dec(&pf_loading);
3637 
3638 	kfree(dev_cap);
3639 	return 0;
3640 
3641 err_port:
3642 	for (--port; port >= 1; --port)
3643 		mlx4_cleanup_port_info(&priv->port[port]);
3644 
3645 	mlx4_cleanup_default_counters(dev);
3646 	if (!mlx4_is_slave(dev))
3647 		mlx4_cleanup_counters_table(dev);
3648 	mlx4_cleanup_qp_table(dev);
3649 	mlx4_cleanup_srq_table(dev);
3650 	mlx4_cleanup_cq_table(dev);
3651 	mlx4_cmd_use_polling(dev);
3652 	mlx4_cleanup_eq_table(dev);
3653 	mlx4_cleanup_mcg_table(dev);
3654 	mlx4_cleanup_mr_table(dev);
3655 	mlx4_cleanup_xrcd_table(dev);
3656 	mlx4_cleanup_pd_table(dev);
3657 	mlx4_cleanup_uar_table(dev);
3658 
3659 err_steer:
3660 	if (!mlx4_is_slave(dev))
3661 		mlx4_clear_steering(dev);
3662 
3663 err_disable_msix:
3664 	if (dev->flags & MLX4_FLAG_MSI_X)
3665 		pci_disable_msix(pdev);
3666 
3667 err_free_eq:
3668 	mlx4_free_eq_table(dev);
3669 
3670 err_master_mfunc:
3671 	if (mlx4_is_master(dev)) {
3672 		mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
3673 		mlx4_multi_func_cleanup(dev);
3674 	}
3675 
3676 	if (mlx4_is_slave(dev))
3677 		mlx4_slave_destroy_special_qp_cap(dev);
3678 
3679 err_close:
3680 	mlx4_close_hca(dev);
3681 
3682 err_fw:
3683 	mlx4_close_fw(dev);
3684 
3685 err_mfunc:
3686 	if (mlx4_is_slave(dev))
3687 		mlx4_multi_func_cleanup(dev);
3688 
3689 err_cmd:
3690 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3691 
3692 err_sriov:
3693 	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3694 		pci_disable_sriov(pdev);
3695 		dev->flags &= ~MLX4_FLAG_SRIOV;
3696 	}
3697 
3698 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3699 		atomic_dec(&pf_loading);
3700 
3701 	kfree(priv->dev.dev_vfs);
3702 
3703 	if (!mlx4_is_slave(dev))
3704 		mlx4_free_ownership(dev);
3705 
3706 	kfree(dev_cap);
3707 	return err;
3708 }
3709 
3710 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3711 			   struct mlx4_priv *priv)
3712 {
3713 	int err;
3714 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3715 	int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3716 	const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3717 		{2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3718 	unsigned total_vfs = 0;
3719 	unsigned int i;
3720 
3721 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3722 
3723 	err = mlx4_pci_enable_device(&priv->dev);
3724 	if (err) {
3725 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3726 		return err;
3727 	}
3728 
3729 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3730 	 * per port, we must limit the number of VFs to 63 (since their are
3731 	 * 128 MACs)
3732 	 */
3733 	for (i = 0; i < ARRAY_SIZE(nvfs) && i < num_vfs_argc;
3734 	     total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3735 		nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3736 		if (nvfs[i] < 0) {
3737 			dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3738 			err = -EINVAL;
3739 			goto err_disable_pdev;
3740 		}
3741 	}
3742 	for (i = 0; i < ARRAY_SIZE(prb_vf) && i < probe_vfs_argc;
3743 	     i++) {
3744 		prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3745 		if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3746 			dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3747 			err = -EINVAL;
3748 			goto err_disable_pdev;
3749 		}
3750 	}
3751 	if (total_vfs > MLX4_MAX_NUM_VF) {
3752 		dev_err(&pdev->dev,
3753 			"Requested more VF's (%d) than allowed by hw (%d)\n",
3754 			total_vfs, MLX4_MAX_NUM_VF);
3755 		err = -EINVAL;
3756 		goto err_disable_pdev;
3757 	}
3758 
3759 	for (i = 0; i < MLX4_MAX_PORTS; i++) {
3760 		if (nvfs[i] + nvfs[2] > MLX4_MAX_NUM_VF_P_PORT) {
3761 			dev_err(&pdev->dev,
3762 				"Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3763 				nvfs[i] + nvfs[2], i + 1,
3764 				MLX4_MAX_NUM_VF_P_PORT);
3765 			err = -EINVAL;
3766 			goto err_disable_pdev;
3767 		}
3768 	}
3769 
3770 	/* Check for BARs. */
3771 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3772 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3773 		dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3774 			pci_dev_data, pci_resource_flags(pdev, 0));
3775 		err = -ENODEV;
3776 		goto err_disable_pdev;
3777 	}
3778 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3779 		dev_err(&pdev->dev, "Missing UAR, aborting\n");
3780 		err = -ENODEV;
3781 		goto err_disable_pdev;
3782 	}
3783 
3784 	err = pci_request_regions(pdev, DRV_NAME);
3785 	if (err) {
3786 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3787 		goto err_disable_pdev;
3788 	}
3789 
3790 	pci_set_master(pdev);
3791 
3792 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3793 	if (err) {
3794 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3795 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3796 		if (err) {
3797 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3798 			goto err_release_regions;
3799 		}
3800 	}
3801 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3802 	if (err) {
3803 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3804 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3805 		if (err) {
3806 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3807 			goto err_release_regions;
3808 		}
3809 	}
3810 
3811 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
3812 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3813 	/* Detect if this device is a virtual function */
3814 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3815 		/* When acting as pf, we normally skip vfs unless explicitly
3816 		 * requested to probe them.
3817 		 */
3818 		if (total_vfs) {
3819 			unsigned vfs_offset = 0;
3820 
3821 			for (i = 0; i < ARRAY_SIZE(nvfs) &&
3822 			     vfs_offset + nvfs[i] < extended_func_num(pdev);
3823 			     vfs_offset += nvfs[i], i++)
3824 				;
3825 			if (i == ARRAY_SIZE(nvfs)) {
3826 				err = -ENODEV;
3827 				goto err_release_regions;
3828 			}
3829 			if ((extended_func_num(pdev) - vfs_offset)
3830 			    > prb_vf[i]) {
3831 				dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3832 					 extended_func_num(pdev));
3833 				err = -ENODEV;
3834 				goto err_release_regions;
3835 			}
3836 		}
3837 	}
3838 
3839 	err = mlx4_crdump_init(&priv->dev);
3840 	if (err)
3841 		goto err_release_regions;
3842 
3843 	err = mlx4_catas_init(&priv->dev);
3844 	if (err)
3845 		goto err_crdump;
3846 
3847 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3848 	if (err)
3849 		goto err_catas;
3850 
3851 	return 0;
3852 
3853 err_catas:
3854 	mlx4_catas_end(&priv->dev);
3855 
3856 err_crdump:
3857 	mlx4_crdump_end(&priv->dev);
3858 
3859 err_release_regions:
3860 	pci_release_regions(pdev);
3861 
3862 err_disable_pdev:
3863 	mlx4_pci_disable_device(&priv->dev);
3864 	return err;
3865 }
3866 
3867 static int mlx4_devlink_port_type_set(struct devlink_port *devlink_port,
3868 				      enum devlink_port_type port_type)
3869 {
3870 	struct mlx4_port_info *info = container_of(devlink_port,
3871 						   struct mlx4_port_info,
3872 						   devlink_port);
3873 	enum mlx4_port_type mlx4_port_type;
3874 
3875 	switch (port_type) {
3876 	case DEVLINK_PORT_TYPE_AUTO:
3877 		mlx4_port_type = MLX4_PORT_TYPE_AUTO;
3878 		break;
3879 	case DEVLINK_PORT_TYPE_ETH:
3880 		mlx4_port_type = MLX4_PORT_TYPE_ETH;
3881 		break;
3882 	case DEVLINK_PORT_TYPE_IB:
3883 		mlx4_port_type = MLX4_PORT_TYPE_IB;
3884 		break;
3885 	default:
3886 		return -EOPNOTSUPP;
3887 	}
3888 
3889 	return __set_port_type(info, mlx4_port_type);
3890 }
3891 
3892 static void mlx4_devlink_param_load_driverinit_values(struct devlink *devlink)
3893 {
3894 	struct mlx4_priv *priv = devlink_priv(devlink);
3895 	struct mlx4_dev *dev = &priv->dev;
3896 	struct mlx4_fw_crdump *crdump = &dev->persist->crdump;
3897 	union devlink_param_value saved_value;
3898 	int err;
3899 
3900 	err = devlink_param_driverinit_value_get(devlink,
3901 						 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET,
3902 						 &saved_value);
3903 	if (!err && mlx4_internal_err_reset != saved_value.vbool) {
3904 		mlx4_internal_err_reset = saved_value.vbool;
3905 		/* Notify on value changed on runtime configuration mode */
3906 		devlink_param_value_changed(devlink,
3907 					    DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET);
3908 	}
3909 	err = devlink_param_driverinit_value_get(devlink,
3910 						 DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
3911 						 &saved_value);
3912 	if (!err)
3913 		log_num_mac = order_base_2(saved_value.vu32);
3914 	err = devlink_param_driverinit_value_get(devlink,
3915 						 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
3916 						 &saved_value);
3917 	if (!err)
3918 		enable_64b_cqe_eqe = saved_value.vbool;
3919 	err = devlink_param_driverinit_value_get(devlink,
3920 						 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
3921 						 &saved_value);
3922 	if (!err)
3923 		enable_4k_uar = saved_value.vbool;
3924 	err = devlink_param_driverinit_value_get(devlink,
3925 						 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT,
3926 						 &saved_value);
3927 	if (!err && crdump->snapshot_enable != saved_value.vbool) {
3928 		crdump->snapshot_enable = saved_value.vbool;
3929 		devlink_param_value_changed(devlink,
3930 					    DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT);
3931 	}
3932 }
3933 
3934 static void mlx4_restart_one_down(struct pci_dev *pdev);
3935 static int mlx4_restart_one_up(struct pci_dev *pdev, bool reload,
3936 			       struct devlink *devlink);
3937 
3938 static int mlx4_devlink_reload_down(struct devlink *devlink,
3939 				    struct netlink_ext_ack *extack)
3940 {
3941 	struct mlx4_priv *priv = devlink_priv(devlink);
3942 	struct mlx4_dev *dev = &priv->dev;
3943 	struct mlx4_dev_persistent *persist = dev->persist;
3944 
3945 	if (persist->num_vfs)
3946 		mlx4_warn(persist->dev, "Reload performed on PF, will cause reset on operating Virtual Functions\n");
3947 	mlx4_restart_one_down(persist->pdev);
3948 	return 0;
3949 }
3950 
3951 static int mlx4_devlink_reload_up(struct devlink *devlink,
3952 				  struct netlink_ext_ack *extack)
3953 {
3954 	struct mlx4_priv *priv = devlink_priv(devlink);
3955 	struct mlx4_dev *dev = &priv->dev;
3956 	struct mlx4_dev_persistent *persist = dev->persist;
3957 	int err;
3958 
3959 	err = mlx4_restart_one_up(persist->pdev, true, devlink);
3960 	if (err)
3961 		mlx4_err(persist->dev, "mlx4_restart_one_up failed, ret=%d\n",
3962 			 err);
3963 
3964 	return err;
3965 }
3966 
3967 static const struct devlink_ops mlx4_devlink_ops = {
3968 	.port_type_set	= mlx4_devlink_port_type_set,
3969 	.reload_down	= mlx4_devlink_reload_down,
3970 	.reload_up	= mlx4_devlink_reload_up,
3971 };
3972 
3973 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3974 {
3975 	struct devlink *devlink;
3976 	struct mlx4_priv *priv;
3977 	struct mlx4_dev *dev;
3978 	int ret;
3979 
3980 	printk_once(KERN_INFO "%s", mlx4_version);
3981 
3982 	devlink = devlink_alloc(&mlx4_devlink_ops, sizeof(*priv));
3983 	if (!devlink)
3984 		return -ENOMEM;
3985 	priv = devlink_priv(devlink);
3986 
3987 	dev       = &priv->dev;
3988 	dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3989 	if (!dev->persist) {
3990 		ret = -ENOMEM;
3991 		goto err_devlink_free;
3992 	}
3993 	dev->persist->pdev = pdev;
3994 	dev->persist->dev = dev;
3995 	pci_set_drvdata(pdev, dev->persist);
3996 	priv->pci_dev_data = id->driver_data;
3997 	mutex_init(&dev->persist->device_state_mutex);
3998 	mutex_init(&dev->persist->interface_state_mutex);
3999 	mutex_init(&dev->persist->pci_status_mutex);
4000 
4001 	ret = devlink_register(devlink, &pdev->dev);
4002 	if (ret)
4003 		goto err_persist_free;
4004 	ret = devlink_params_register(devlink, mlx4_devlink_params,
4005 				      ARRAY_SIZE(mlx4_devlink_params));
4006 	if (ret)
4007 		goto err_devlink_unregister;
4008 	mlx4_devlink_set_params_init_values(devlink);
4009 	ret =  __mlx4_init_one(pdev, id->driver_data, priv);
4010 	if (ret)
4011 		goto err_params_unregister;
4012 
4013 	devlink_params_publish(devlink);
4014 	pci_save_state(pdev);
4015 	return 0;
4016 
4017 err_params_unregister:
4018 	devlink_params_unregister(devlink, mlx4_devlink_params,
4019 				  ARRAY_SIZE(mlx4_devlink_params));
4020 err_devlink_unregister:
4021 	devlink_unregister(devlink);
4022 err_persist_free:
4023 	kfree(dev->persist);
4024 err_devlink_free:
4025 	devlink_free(devlink);
4026 	return ret;
4027 }
4028 
4029 static void mlx4_clean_dev(struct mlx4_dev *dev)
4030 {
4031 	struct mlx4_dev_persistent *persist = dev->persist;
4032 	struct mlx4_priv *priv = mlx4_priv(dev);
4033 	unsigned long	flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
4034 
4035 	memset(priv, 0, sizeof(*priv));
4036 	priv->dev.persist = persist;
4037 	priv->dev.flags = flags;
4038 }
4039 
4040 static void mlx4_unload_one(struct pci_dev *pdev)
4041 {
4042 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4043 	struct mlx4_dev  *dev  = persist->dev;
4044 	struct mlx4_priv *priv = mlx4_priv(dev);
4045 	int               pci_dev_data;
4046 	int p, i;
4047 
4048 	if (priv->removed)
4049 		return;
4050 
4051 	/* saving current ports type for further use */
4052 	for (i = 0; i < dev->caps.num_ports; i++) {
4053 		dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
4054 		dev->persist->curr_port_poss_type[i] = dev->caps.
4055 						       possible_type[i + 1];
4056 	}
4057 
4058 	pci_dev_data = priv->pci_dev_data;
4059 
4060 	mlx4_stop_sense(dev);
4061 	mlx4_unregister_device(dev);
4062 
4063 	for (p = 1; p <= dev->caps.num_ports; p++) {
4064 		mlx4_cleanup_port_info(&priv->port[p]);
4065 		mlx4_CLOSE_PORT(dev, p);
4066 	}
4067 
4068 	if (mlx4_is_master(dev))
4069 		mlx4_free_resource_tracker(dev,
4070 					   RES_TR_FREE_SLAVES_ONLY);
4071 
4072 	mlx4_cleanup_default_counters(dev);
4073 	if (!mlx4_is_slave(dev))
4074 		mlx4_cleanup_counters_table(dev);
4075 	mlx4_cleanup_qp_table(dev);
4076 	mlx4_cleanup_srq_table(dev);
4077 	mlx4_cleanup_cq_table(dev);
4078 	mlx4_cmd_use_polling(dev);
4079 	mlx4_cleanup_eq_table(dev);
4080 	mlx4_cleanup_mcg_table(dev);
4081 	mlx4_cleanup_mr_table(dev);
4082 	mlx4_cleanup_xrcd_table(dev);
4083 	mlx4_cleanup_pd_table(dev);
4084 
4085 	if (mlx4_is_master(dev))
4086 		mlx4_free_resource_tracker(dev,
4087 					   RES_TR_FREE_STRUCTS_ONLY);
4088 
4089 	iounmap(priv->kar);
4090 	mlx4_uar_free(dev, &priv->driver_uar);
4091 	mlx4_cleanup_uar_table(dev);
4092 	if (!mlx4_is_slave(dev))
4093 		mlx4_clear_steering(dev);
4094 	mlx4_free_eq_table(dev);
4095 	if (mlx4_is_master(dev))
4096 		mlx4_multi_func_cleanup(dev);
4097 	mlx4_close_hca(dev);
4098 	mlx4_close_fw(dev);
4099 	if (mlx4_is_slave(dev))
4100 		mlx4_multi_func_cleanup(dev);
4101 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
4102 
4103 	if (dev->flags & MLX4_FLAG_MSI_X)
4104 		pci_disable_msix(pdev);
4105 
4106 	if (!mlx4_is_slave(dev))
4107 		mlx4_free_ownership(dev);
4108 
4109 	mlx4_slave_destroy_special_qp_cap(dev);
4110 	kfree(dev->dev_vfs);
4111 
4112 	mlx4_clean_dev(dev);
4113 	priv->pci_dev_data = pci_dev_data;
4114 	priv->removed = 1;
4115 }
4116 
4117 static void mlx4_remove_one(struct pci_dev *pdev)
4118 {
4119 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4120 	struct mlx4_dev  *dev  = persist->dev;
4121 	struct mlx4_priv *priv = mlx4_priv(dev);
4122 	struct devlink *devlink = priv_to_devlink(priv);
4123 	int active_vfs = 0;
4124 
4125 	if (mlx4_is_slave(dev))
4126 		persist->interface_state |= MLX4_INTERFACE_STATE_NOWAIT;
4127 
4128 	mutex_lock(&persist->interface_state_mutex);
4129 	persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
4130 	mutex_unlock(&persist->interface_state_mutex);
4131 
4132 	/* Disabling SR-IOV is not allowed while there are active vf's */
4133 	if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
4134 		active_vfs = mlx4_how_many_lives_vf(dev);
4135 		if (active_vfs) {
4136 			pr_warn("Removing PF when there are active VF's !!\n");
4137 			pr_warn("Will not disable SR-IOV.\n");
4138 		}
4139 	}
4140 
4141 	/* device marked to be under deletion running now without the lock
4142 	 * letting other tasks to be terminated
4143 	 */
4144 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4145 		mlx4_unload_one(pdev);
4146 	else
4147 		mlx4_info(dev, "%s: interface is down\n", __func__);
4148 	mlx4_catas_end(dev);
4149 	mlx4_crdump_end(dev);
4150 	if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
4151 		mlx4_warn(dev, "Disabling SR-IOV\n");
4152 		pci_disable_sriov(pdev);
4153 	}
4154 
4155 	pci_release_regions(pdev);
4156 	mlx4_pci_disable_device(dev);
4157 	devlink_params_unregister(devlink, mlx4_devlink_params,
4158 				  ARRAY_SIZE(mlx4_devlink_params));
4159 	devlink_unregister(devlink);
4160 	kfree(dev->persist);
4161 	devlink_free(devlink);
4162 }
4163 
4164 static int restore_current_port_types(struct mlx4_dev *dev,
4165 				      enum mlx4_port_type *types,
4166 				      enum mlx4_port_type *poss_types)
4167 {
4168 	struct mlx4_priv *priv = mlx4_priv(dev);
4169 	int err, i;
4170 
4171 	mlx4_stop_sense(dev);
4172 
4173 	mutex_lock(&priv->port_mutex);
4174 	for (i = 0; i < dev->caps.num_ports; i++)
4175 		dev->caps.possible_type[i + 1] = poss_types[i];
4176 	err = mlx4_change_port_types(dev, types);
4177 	mlx4_start_sense(dev);
4178 	mutex_unlock(&priv->port_mutex);
4179 
4180 	return err;
4181 }
4182 
4183 static void mlx4_restart_one_down(struct pci_dev *pdev)
4184 {
4185 	mlx4_unload_one(pdev);
4186 }
4187 
4188 static int mlx4_restart_one_up(struct pci_dev *pdev, bool reload,
4189 			       struct devlink *devlink)
4190 {
4191 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4192 	struct mlx4_dev	 *dev  = persist->dev;
4193 	struct mlx4_priv *priv = mlx4_priv(dev);
4194 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4195 	int pci_dev_data, err, total_vfs;
4196 
4197 	pci_dev_data = priv->pci_dev_data;
4198 	total_vfs = dev->persist->num_vfs;
4199 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4200 
4201 	if (reload)
4202 		mlx4_devlink_param_load_driverinit_values(devlink);
4203 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
4204 	if (err) {
4205 		mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
4206 			 __func__, pci_name(pdev), err);
4207 		return err;
4208 	}
4209 
4210 	err = restore_current_port_types(dev, dev->persist->curr_port_type,
4211 					 dev->persist->curr_port_poss_type);
4212 	if (err)
4213 		mlx4_err(dev, "could not restore original port types (%d)\n",
4214 			 err);
4215 
4216 	return err;
4217 }
4218 
4219 int mlx4_restart_one(struct pci_dev *pdev)
4220 {
4221 	mlx4_restart_one_down(pdev);
4222 	return mlx4_restart_one_up(pdev, false, NULL);
4223 }
4224 
4225 #define MLX_SP(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_FORCE_SENSE_PORT }
4226 #define MLX_VF(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_IS_VF }
4227 #define MLX_GN(id) { PCI_VDEVICE(MELLANOX, id), 0 }
4228 
4229 static const struct pci_device_id mlx4_pci_table[] = {
4230 #ifdef CONFIG_MLX4_CORE_GEN2
4231 	/* MT25408 "Hermon" */
4232 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_SDR),	/* SDR */
4233 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR),	/* DDR */
4234 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR),	/* QDR */
4235 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR_GEN2), /* DDR Gen2 */
4236 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR_GEN2),	/* QDR Gen2 */
4237 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN),	/* EN 10GigE */
4238 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN_GEN2),  /* EN 10GigE Gen2 */
4239 	/* MT25458 ConnectX EN 10GBASE-T */
4240 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN),
4241 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_T_GEN2),	/* Gen2 */
4242 	/* MT26468 ConnectX EN 10GigE PCIe Gen2*/
4243 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_GEN2),
4244 	/* MT26438 ConnectX EN 40GigE PCIe Gen2 5GT/s */
4245 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_5_GEN2),
4246 	/* MT26478 ConnectX2 40GigE PCIe Gen2 */
4247 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX2),
4248 	/* MT25400 Family [ConnectX-2] */
4249 	MLX_VF(0x1002),					/* Virtual Function */
4250 #endif /* CONFIG_MLX4_CORE_GEN2 */
4251 	/* MT27500 Family [ConnectX-3] */
4252 	MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3),
4253 	MLX_VF(0x1004),					/* Virtual Function */
4254 	MLX_GN(0x1005),					/* MT27510 Family */
4255 	MLX_GN(0x1006),					/* MT27511 Family */
4256 	MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3_PRO),	/* MT27520 Family */
4257 	MLX_GN(0x1008),					/* MT27521 Family */
4258 	MLX_GN(0x1009),					/* MT27530 Family */
4259 	MLX_GN(0x100a),					/* MT27531 Family */
4260 	MLX_GN(0x100b),					/* MT27540 Family */
4261 	MLX_GN(0x100c),					/* MT27541 Family */
4262 	MLX_GN(0x100d),					/* MT27550 Family */
4263 	MLX_GN(0x100e),					/* MT27551 Family */
4264 	MLX_GN(0x100f),					/* MT27560 Family */
4265 	MLX_GN(0x1010),					/* MT27561 Family */
4266 
4267 	/*
4268 	 * See the mellanox_check_broken_intx_masking() quirk when
4269 	 * adding devices
4270 	 */
4271 
4272 	{ 0, }
4273 };
4274 
4275 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
4276 
4277 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
4278 					      pci_channel_state_t state)
4279 {
4280 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4281 
4282 	mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
4283 	mlx4_enter_error_state(persist);
4284 
4285 	mutex_lock(&persist->interface_state_mutex);
4286 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4287 		mlx4_unload_one(pdev);
4288 
4289 	mutex_unlock(&persist->interface_state_mutex);
4290 	if (state == pci_channel_io_perm_failure)
4291 		return PCI_ERS_RESULT_DISCONNECT;
4292 
4293 	mlx4_pci_disable_device(persist->dev);
4294 	return PCI_ERS_RESULT_NEED_RESET;
4295 }
4296 
4297 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
4298 {
4299 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4300 	struct mlx4_dev	 *dev  = persist->dev;
4301 	int err;
4302 
4303 	mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
4304 	err = mlx4_pci_enable_device(dev);
4305 	if (err) {
4306 		mlx4_err(dev, "Can not re-enable device, err=%d\n", err);
4307 		return PCI_ERS_RESULT_DISCONNECT;
4308 	}
4309 
4310 	pci_set_master(pdev);
4311 	pci_restore_state(pdev);
4312 	pci_save_state(pdev);
4313 	return PCI_ERS_RESULT_RECOVERED;
4314 }
4315 
4316 static void mlx4_pci_resume(struct pci_dev *pdev)
4317 {
4318 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4319 	struct mlx4_dev	 *dev  = persist->dev;
4320 	struct mlx4_priv *priv = mlx4_priv(dev);
4321 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4322 	int total_vfs;
4323 	int err;
4324 
4325 	mlx4_err(dev, "%s was called\n", __func__);
4326 	total_vfs = dev->persist->num_vfs;
4327 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4328 
4329 	mutex_lock(&persist->interface_state_mutex);
4330 	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4331 		err = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
4332 				    priv, 1);
4333 		if (err) {
4334 			mlx4_err(dev, "%s: mlx4_load_one failed, err=%d\n",
4335 				 __func__,  err);
4336 			goto end;
4337 		}
4338 
4339 		err = restore_current_port_types(dev, dev->persist->
4340 						 curr_port_type, dev->persist->
4341 						 curr_port_poss_type);
4342 		if (err)
4343 			mlx4_err(dev, "could not restore original port types (%d)\n", err);
4344 	}
4345 end:
4346 	mutex_unlock(&persist->interface_state_mutex);
4347 
4348 }
4349 
4350 static void mlx4_shutdown(struct pci_dev *pdev)
4351 {
4352 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4353 
4354 	mlx4_info(persist->dev, "mlx4_shutdown was called\n");
4355 	mutex_lock(&persist->interface_state_mutex);
4356 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4357 		mlx4_unload_one(pdev);
4358 	mutex_unlock(&persist->interface_state_mutex);
4359 }
4360 
4361 static const struct pci_error_handlers mlx4_err_handler = {
4362 	.error_detected = mlx4_pci_err_detected,
4363 	.slot_reset     = mlx4_pci_slot_reset,
4364 	.resume		= mlx4_pci_resume,
4365 };
4366 
4367 static int mlx4_suspend(struct pci_dev *pdev, pm_message_t state)
4368 {
4369 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4370 	struct mlx4_dev	*dev = persist->dev;
4371 
4372 	mlx4_err(dev, "suspend was called\n");
4373 	mutex_lock(&persist->interface_state_mutex);
4374 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4375 		mlx4_unload_one(pdev);
4376 	mutex_unlock(&persist->interface_state_mutex);
4377 
4378 	return 0;
4379 }
4380 
4381 static int mlx4_resume(struct pci_dev *pdev)
4382 {
4383 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4384 	struct mlx4_dev	*dev = persist->dev;
4385 	struct mlx4_priv *priv = mlx4_priv(dev);
4386 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4387 	int total_vfs;
4388 	int ret = 0;
4389 
4390 	mlx4_err(dev, "resume was called\n");
4391 	total_vfs = dev->persist->num_vfs;
4392 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4393 
4394 	mutex_lock(&persist->interface_state_mutex);
4395 	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4396 		ret = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs,
4397 				    nvfs, priv, 1);
4398 		if (!ret) {
4399 			ret = restore_current_port_types(dev,
4400 					dev->persist->curr_port_type,
4401 					dev->persist->curr_port_poss_type);
4402 			if (ret)
4403 				mlx4_err(dev, "resume: could not restore original port types (%d)\n", ret);
4404 		}
4405 	}
4406 	mutex_unlock(&persist->interface_state_mutex);
4407 
4408 	return ret;
4409 }
4410 
4411 static struct pci_driver mlx4_driver = {
4412 	.name		= DRV_NAME,
4413 	.id_table	= mlx4_pci_table,
4414 	.probe		= mlx4_init_one,
4415 	.shutdown	= mlx4_shutdown,
4416 	.remove		= mlx4_remove_one,
4417 	.suspend	= mlx4_suspend,
4418 	.resume		= mlx4_resume,
4419 	.err_handler    = &mlx4_err_handler,
4420 };
4421 
4422 static int __init mlx4_verify_params(void)
4423 {
4424 	if (msi_x < 0) {
4425 		pr_warn("mlx4_core: bad msi_x: %d\n", msi_x);
4426 		return -1;
4427 	}
4428 
4429 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
4430 		pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
4431 		return -1;
4432 	}
4433 
4434 	if (log_num_vlan != 0)
4435 		pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
4436 			MLX4_LOG_NUM_VLANS);
4437 
4438 	if (use_prio != 0)
4439 		pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
4440 
4441 	if ((log_mtts_per_seg < 0) || (log_mtts_per_seg > 7)) {
4442 		pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
4443 			log_mtts_per_seg);
4444 		return -1;
4445 	}
4446 
4447 	/* Check if module param for ports type has legal combination */
4448 	if (port_type_array[0] == false && port_type_array[1] == true) {
4449 		pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
4450 		port_type_array[0] = true;
4451 	}
4452 
4453 	if (mlx4_log_num_mgm_entry_size < -7 ||
4454 	    (mlx4_log_num_mgm_entry_size > 0 &&
4455 	     (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
4456 	      mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
4457 		pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
4458 			mlx4_log_num_mgm_entry_size,
4459 			MLX4_MIN_MGM_LOG_ENTRY_SIZE,
4460 			MLX4_MAX_MGM_LOG_ENTRY_SIZE);
4461 		return -1;
4462 	}
4463 
4464 	return 0;
4465 }
4466 
4467 static int __init mlx4_init(void)
4468 {
4469 	int ret;
4470 
4471 	if (mlx4_verify_params())
4472 		return -EINVAL;
4473 
4474 
4475 	mlx4_wq = create_singlethread_workqueue("mlx4");
4476 	if (!mlx4_wq)
4477 		return -ENOMEM;
4478 
4479 	ret = pci_register_driver(&mlx4_driver);
4480 	if (ret < 0)
4481 		destroy_workqueue(mlx4_wq);
4482 	return ret < 0 ? ret : 0;
4483 }
4484 
4485 static void __exit mlx4_cleanup(void)
4486 {
4487 	pci_unregister_driver(&mlx4_driver);
4488 	destroy_workqueue(mlx4_wq);
4489 }
4490 
4491 module_init(mlx4_init);
4492 module_exit(mlx4_cleanup);
4493