xref: /linux/drivers/net/ethernet/mellanox/mlx4/main.c (revision 9cfc5c90ad38c8fc11bfd39de42a107da00871ba)
1 /*
2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4  * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
5  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * OpenIB.org BSD license below:
12  *
13  *     Redistribution and use in source and binary forms, with or
14  *     without modification, are permitted provided that the following
15  *     conditions are met:
16  *
17  *      - Redistributions of source code must retain the above
18  *        copyright notice, this list of conditions and the following
19  *        disclaimer.
20  *
21  *      - Redistributions in binary form must reproduce the above
22  *        copyright notice, this list of conditions and the following
23  *        disclaimer in the documentation and/or other materials
24  *        provided with the distribution.
25  *
26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33  * SOFTWARE.
34  */
35 
36 #include <linux/module.h>
37 #include <linux/init.h>
38 #include <linux/errno.h>
39 #include <linux/pci.h>
40 #include <linux/dma-mapping.h>
41 #include <linux/slab.h>
42 #include <linux/io-mapping.h>
43 #include <linux/delay.h>
44 #include <linux/kmod.h>
45 
46 #include <linux/mlx4/device.h>
47 #include <linux/mlx4/doorbell.h>
48 
49 #include "mlx4.h"
50 #include "fw.h"
51 #include "icm.h"
52 
53 MODULE_AUTHOR("Roland Dreier");
54 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
55 MODULE_LICENSE("Dual BSD/GPL");
56 MODULE_VERSION(DRV_VERSION);
57 
58 struct workqueue_struct *mlx4_wq;
59 
60 #ifdef CONFIG_MLX4_DEBUG
61 
62 int mlx4_debug_level = 0;
63 module_param_named(debug_level, mlx4_debug_level, int, 0644);
64 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
65 
66 #endif /* CONFIG_MLX4_DEBUG */
67 
68 #ifdef CONFIG_PCI_MSI
69 
70 static int msi_x = 1;
71 module_param(msi_x, int, 0444);
72 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
73 
74 #else /* CONFIG_PCI_MSI */
75 
76 #define msi_x (0)
77 
78 #endif /* CONFIG_PCI_MSI */
79 
80 static uint8_t num_vfs[3] = {0, 0, 0};
81 static int num_vfs_argc;
82 module_param_array(num_vfs, byte , &num_vfs_argc, 0444);
83 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
84 			  "num_vfs=port1,port2,port1+2");
85 
86 static uint8_t probe_vf[3] = {0, 0, 0};
87 static int probe_vfs_argc;
88 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
89 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
90 			   "probe_vf=port1,port2,port1+2");
91 
92 int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
93 module_param_named(log_num_mgm_entry_size,
94 			mlx4_log_num_mgm_entry_size, int, 0444);
95 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
96 					 " of qp per mcg, for example:"
97 					 " 10 gives 248.range: 7 <="
98 					 " log_num_mgm_entry_size <= 12."
99 					 " To activate device managed"
100 					 " flow steering when available, set to -1");
101 
102 static bool enable_64b_cqe_eqe = true;
103 module_param(enable_64b_cqe_eqe, bool, 0444);
104 MODULE_PARM_DESC(enable_64b_cqe_eqe,
105 		 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
106 
107 #define PF_CONTEXT_BEHAVIOUR_MASK	(MLX4_FUNC_CAP_64B_EQE_CQE | \
108 					 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
109 					 MLX4_FUNC_CAP_DMFS_A0_STATIC)
110 
111 #define RESET_PERSIST_MASK_FLAGS	(MLX4_FLAG_SRIOV)
112 
113 static char mlx4_version[] =
114 	DRV_NAME ": Mellanox ConnectX core driver v"
115 	DRV_VERSION " (" DRV_RELDATE ")\n";
116 
117 static struct mlx4_profile default_profile = {
118 	.num_qp		= 1 << 18,
119 	.num_srq	= 1 << 16,
120 	.rdmarc_per_qp	= 1 << 4,
121 	.num_cq		= 1 << 16,
122 	.num_mcg	= 1 << 13,
123 	.num_mpt	= 1 << 19,
124 	.num_mtt	= 1 << 20, /* It is really num mtt segements */
125 };
126 
127 static struct mlx4_profile low_mem_profile = {
128 	.num_qp		= 1 << 17,
129 	.num_srq	= 1 << 6,
130 	.rdmarc_per_qp	= 1 << 4,
131 	.num_cq		= 1 << 8,
132 	.num_mcg	= 1 << 8,
133 	.num_mpt	= 1 << 9,
134 	.num_mtt	= 1 << 7,
135 };
136 
137 static int log_num_mac = 7;
138 module_param_named(log_num_mac, log_num_mac, int, 0444);
139 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
140 
141 static int log_num_vlan;
142 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
143 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
144 /* Log2 max number of VLANs per ETH port (0-7) */
145 #define MLX4_LOG_NUM_VLANS 7
146 #define MLX4_MIN_LOG_NUM_VLANS 0
147 #define MLX4_MIN_LOG_NUM_MAC 1
148 
149 static bool use_prio;
150 module_param_named(use_prio, use_prio, bool, 0444);
151 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
152 
153 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
154 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
155 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
156 
157 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
158 static int arr_argc = 2;
159 module_param_array(port_type_array, int, &arr_argc, 0444);
160 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
161 				"1 for IB, 2 for Ethernet");
162 
163 struct mlx4_port_config {
164 	struct list_head list;
165 	enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
166 	struct pci_dev *pdev;
167 };
168 
169 static atomic_t pf_loading = ATOMIC_INIT(0);
170 
171 int mlx4_check_port_params(struct mlx4_dev *dev,
172 			   enum mlx4_port_type *port_type)
173 {
174 	int i;
175 
176 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
177 		for (i = 0; i < dev->caps.num_ports - 1; i++) {
178 			if (port_type[i] != port_type[i + 1]) {
179 				mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
180 				return -EINVAL;
181 			}
182 		}
183 	}
184 
185 	for (i = 0; i < dev->caps.num_ports; i++) {
186 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
187 			mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
188 				 i + 1);
189 			return -EINVAL;
190 		}
191 	}
192 	return 0;
193 }
194 
195 static void mlx4_set_port_mask(struct mlx4_dev *dev)
196 {
197 	int i;
198 
199 	for (i = 1; i <= dev->caps.num_ports; ++i)
200 		dev->caps.port_mask[i] = dev->caps.port_type[i];
201 }
202 
203 enum {
204 	MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
205 };
206 
207 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
208 {
209 	int err = 0;
210 	struct mlx4_func func;
211 
212 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
213 		err = mlx4_QUERY_FUNC(dev, &func, 0);
214 		if (err) {
215 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
216 			return err;
217 		}
218 		dev_cap->max_eqs = func.max_eq;
219 		dev_cap->reserved_eqs = func.rsvd_eqs;
220 		dev_cap->reserved_uars = func.rsvd_uars;
221 		err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
222 	}
223 	return err;
224 }
225 
226 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
227 {
228 	struct mlx4_caps *dev_cap = &dev->caps;
229 
230 	/* FW not supporting or cancelled by user */
231 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
232 	    !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
233 		return;
234 
235 	/* Must have 64B CQE_EQE enabled by FW to use bigger stride
236 	 * When FW has NCSI it may decide not to report 64B CQE/EQEs
237 	 */
238 	if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
239 	    !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
240 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
241 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
242 		return;
243 	}
244 
245 	if (cache_line_size() == 128 || cache_line_size() == 256) {
246 		mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
247 		/* Changing the real data inside CQE size to 32B */
248 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
249 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
250 
251 		if (mlx4_is_master(dev))
252 			dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
253 	} else {
254 		if (cache_line_size() != 32  && cache_line_size() != 64)
255 			mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
256 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
257 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
258 	}
259 }
260 
261 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
262 			  struct mlx4_port_cap *port_cap)
263 {
264 	dev->caps.vl_cap[port]	    = port_cap->max_vl;
265 	dev->caps.ib_mtu_cap[port]	    = port_cap->ib_mtu;
266 	dev->phys_caps.gid_phys_table_len[port]  = port_cap->max_gids;
267 	dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
268 	/* set gid and pkey table operating lengths by default
269 	 * to non-sriov values
270 	 */
271 	dev->caps.gid_table_len[port]  = port_cap->max_gids;
272 	dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
273 	dev->caps.port_width_cap[port] = port_cap->max_port_width;
274 	dev->caps.eth_mtu_cap[port]    = port_cap->eth_mtu;
275 	dev->caps.def_mac[port]        = port_cap->def_mac;
276 	dev->caps.supported_type[port] = port_cap->supported_port_types;
277 	dev->caps.suggested_type[port] = port_cap->suggested_type;
278 	dev->caps.default_sense[port] = port_cap->default_sense;
279 	dev->caps.trans_type[port]	    = port_cap->trans_type;
280 	dev->caps.vendor_oui[port]     = port_cap->vendor_oui;
281 	dev->caps.wavelength[port]     = port_cap->wavelength;
282 	dev->caps.trans_code[port]     = port_cap->trans_code;
283 
284 	return 0;
285 }
286 
287 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
288 			 struct mlx4_port_cap *port_cap)
289 {
290 	int err = 0;
291 
292 	err = mlx4_QUERY_PORT(dev, port, port_cap);
293 
294 	if (err)
295 		mlx4_err(dev, "QUERY_PORT command failed.\n");
296 
297 	return err;
298 }
299 
300 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
301 {
302 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
303 		return;
304 
305 	if (mlx4_is_mfunc(dev)) {
306 		mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
307 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
308 		return;
309 	}
310 
311 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
312 		mlx4_dbg(dev,
313 			 "Keep FCS is not supported - Disabling Ignore FCS");
314 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
315 		return;
316 	}
317 }
318 
319 #define MLX4_A0_STEERING_TABLE_SIZE	256
320 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
321 {
322 	int err;
323 	int i;
324 
325 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
326 	if (err) {
327 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
328 		return err;
329 	}
330 	mlx4_dev_cap_dump(dev, dev_cap);
331 
332 	if (dev_cap->min_page_sz > PAGE_SIZE) {
333 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
334 			 dev_cap->min_page_sz, PAGE_SIZE);
335 		return -ENODEV;
336 	}
337 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
338 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
339 			 dev_cap->num_ports, MLX4_MAX_PORTS);
340 		return -ENODEV;
341 	}
342 
343 	if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
344 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
345 			 dev_cap->uar_size,
346 			 (unsigned long long)
347 			 pci_resource_len(dev->persist->pdev, 2));
348 		return -ENODEV;
349 	}
350 
351 	dev->caps.num_ports	     = dev_cap->num_ports;
352 	dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
353 	dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
354 				      dev->caps.num_sys_eqs :
355 				      MLX4_MAX_EQ_NUM;
356 	for (i = 1; i <= dev->caps.num_ports; ++i) {
357 		err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
358 		if (err) {
359 			mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
360 			return err;
361 		}
362 	}
363 
364 	dev->caps.uar_page_size	     = PAGE_SIZE;
365 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
366 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
367 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
368 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
369 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
370 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
371 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
372 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
373 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
374 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
375 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
376 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
377 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
378 	/*
379 	 * Subtract 1 from the limit because we need to allocate a
380 	 * spare CQE so the HCA HW can tell the difference between an
381 	 * empty CQ and a full CQ.
382 	 */
383 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
384 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
385 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
386 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
387 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
388 
389 	/* The first 128 UARs are used for EQ doorbells */
390 	dev->caps.reserved_uars	     = max_t(int, 128, dev_cap->reserved_uars);
391 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
392 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
393 					dev_cap->reserved_xrcds : 0;
394 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
395 					dev_cap->max_xrcds : 0;
396 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
397 
398 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
399 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
400 	dev->caps.flags		     = dev_cap->flags;
401 	dev->caps.flags2	     = dev_cap->flags2;
402 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
403 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
404 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
405 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
406 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
407 
408 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN) {
409 		struct mlx4_init_hca_param hca_param;
410 
411 		memset(&hca_param, 0, sizeof(hca_param));
412 		err = mlx4_QUERY_HCA(dev, &hca_param);
413 		/* Turn off PHV_EN flag in case phv_check_en is set.
414 		 * phv_check_en is a HW check that parse the packet and verify
415 		 * phv bit was reported correctly in the wqe. To allow QinQ
416 		 * PHV_EN flag should be set and phv_check_en must be cleared
417 		 * otherwise QinQ packets will be drop by the HW.
418 		 */
419 		if (err || hca_param.phv_check_en)
420 			dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_PHV_EN;
421 	}
422 
423 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
424 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
425 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
426 	/* Don't do sense port on multifunction devices (for now at least) */
427 	if (mlx4_is_mfunc(dev))
428 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
429 
430 	if (mlx4_low_memory_profile()) {
431 		dev->caps.log_num_macs  = MLX4_MIN_LOG_NUM_MAC;
432 		dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
433 	} else {
434 		dev->caps.log_num_macs  = log_num_mac;
435 		dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
436 	}
437 
438 	for (i = 1; i <= dev->caps.num_ports; ++i) {
439 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
440 		if (dev->caps.supported_type[i]) {
441 			/* if only ETH is supported - assign ETH */
442 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
443 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
444 			/* if only IB is supported, assign IB */
445 			else if (dev->caps.supported_type[i] ==
446 				 MLX4_PORT_TYPE_IB)
447 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
448 			else {
449 				/* if IB and ETH are supported, we set the port
450 				 * type according to user selection of port type;
451 				 * if user selected none, take the FW hint */
452 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
453 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
454 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
455 				else
456 					dev->caps.port_type[i] = port_type_array[i - 1];
457 			}
458 		}
459 		/*
460 		 * Link sensing is allowed on the port if 3 conditions are true:
461 		 * 1. Both protocols are supported on the port.
462 		 * 2. Different types are supported on the port
463 		 * 3. FW declared that it supports link sensing
464 		 */
465 		mlx4_priv(dev)->sense.sense_allowed[i] =
466 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
467 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
468 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
469 
470 		/*
471 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
472 		 * and perform sense_port FW command to try and set the correct
473 		 * port type from beginning
474 		 */
475 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
476 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
477 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
478 			mlx4_SENSE_PORT(dev, i, &sensed_port);
479 			if (sensed_port != MLX4_PORT_TYPE_NONE)
480 				dev->caps.port_type[i] = sensed_port;
481 		} else {
482 			dev->caps.possible_type[i] = dev->caps.port_type[i];
483 		}
484 
485 		if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
486 			dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
487 			mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
488 				  i, 1 << dev->caps.log_num_macs);
489 		}
490 		if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
491 			dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
492 			mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
493 				  i, 1 << dev->caps.log_num_vlans);
494 		}
495 	}
496 
497 	if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
498 	    (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
499 	    (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
500 		mlx4_warn(dev,
501 			  "Granular QoS per VF not supported with IB/Eth configuration\n");
502 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
503 	}
504 
505 	dev->caps.max_counters = dev_cap->max_counters;
506 
507 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
508 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
509 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
510 		(1 << dev->caps.log_num_macs) *
511 		(1 << dev->caps.log_num_vlans) *
512 		dev->caps.num_ports;
513 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
514 
515 	if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
516 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
517 		dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
518 	else
519 		dev->caps.dmfs_high_rate_qpn_base =
520 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
521 
522 	if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
523 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
524 		dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
525 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
526 		dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
527 	} else {
528 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
529 		dev->caps.dmfs_high_rate_qpn_base =
530 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
531 		dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
532 	}
533 
534 	dev->caps.rl_caps = dev_cap->rl_caps;
535 
536 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
537 		dev->caps.dmfs_high_rate_qpn_range;
538 
539 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
540 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
541 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
542 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
543 
544 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
545 
546 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
547 		if (dev_cap->flags &
548 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
549 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
550 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
551 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
552 		}
553 
554 		if (dev_cap->flags2 &
555 		    (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
556 		     MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
557 			mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
558 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
559 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
560 		}
561 	}
562 
563 	if ((dev->caps.flags &
564 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
565 	    mlx4_is_master(dev))
566 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
567 
568 	if (!mlx4_is_slave(dev)) {
569 		mlx4_enable_cqe_eqe_stride(dev);
570 		dev->caps.alloc_res_qp_mask =
571 			(dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
572 			MLX4_RESERVE_A0_QP;
573 
574 		if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
575 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
576 			mlx4_warn(dev, "Old device ETS support detected\n");
577 			mlx4_warn(dev, "Consider upgrading device FW.\n");
578 			dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
579 		}
580 
581 	} else {
582 		dev->caps.alloc_res_qp_mask = 0;
583 	}
584 
585 	mlx4_enable_ignore_fcs(dev);
586 
587 	return 0;
588 }
589 
590 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
591 				       enum pci_bus_speed *speed,
592 				       enum pcie_link_width *width)
593 {
594 	u32 lnkcap1, lnkcap2;
595 	int err1, err2;
596 
597 #define  PCIE_MLW_CAP_SHIFT 4	/* start of MLW mask in link capabilities */
598 
599 	*speed = PCI_SPEED_UNKNOWN;
600 	*width = PCIE_LNK_WIDTH_UNKNOWN;
601 
602 	err1 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP,
603 					  &lnkcap1);
604 	err2 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP2,
605 					  &lnkcap2);
606 	if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
607 		if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
608 			*speed = PCIE_SPEED_8_0GT;
609 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
610 			*speed = PCIE_SPEED_5_0GT;
611 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
612 			*speed = PCIE_SPEED_2_5GT;
613 	}
614 	if (!err1) {
615 		*width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
616 		if (!lnkcap2) { /* pre-r3.0 */
617 			if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
618 				*speed = PCIE_SPEED_5_0GT;
619 			else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
620 				*speed = PCIE_SPEED_2_5GT;
621 		}
622 	}
623 
624 	if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
625 		return err1 ? err1 :
626 			err2 ? err2 : -EINVAL;
627 	}
628 	return 0;
629 }
630 
631 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
632 {
633 	enum pcie_link_width width, width_cap;
634 	enum pci_bus_speed speed, speed_cap;
635 	int err;
636 
637 #define PCIE_SPEED_STR(speed) \
638 	(speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
639 	 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
640 	 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
641 	 "Unknown")
642 
643 	err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
644 	if (err) {
645 		mlx4_warn(dev,
646 			  "Unable to determine PCIe device BW capabilities\n");
647 		return;
648 	}
649 
650 	err = pcie_get_minimum_link(dev->persist->pdev, &speed, &width);
651 	if (err || speed == PCI_SPEED_UNKNOWN ||
652 	    width == PCIE_LNK_WIDTH_UNKNOWN) {
653 		mlx4_warn(dev,
654 			  "Unable to determine PCI device chain minimum BW\n");
655 		return;
656 	}
657 
658 	if (width != width_cap || speed != speed_cap)
659 		mlx4_warn(dev,
660 			  "PCIe BW is different than device's capability\n");
661 
662 	mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
663 		  PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
664 	mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
665 		  width, width_cap);
666 	return;
667 }
668 
669 /*The function checks if there are live vf, return the num of them*/
670 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
671 {
672 	struct mlx4_priv *priv = mlx4_priv(dev);
673 	struct mlx4_slave_state *s_state;
674 	int i;
675 	int ret = 0;
676 
677 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
678 		s_state = &priv->mfunc.master.slave_state[i];
679 		if (s_state->active && s_state->last_cmd !=
680 		    MLX4_COMM_CMD_RESET) {
681 			mlx4_warn(dev, "%s: slave: %d is still active\n",
682 				  __func__, i);
683 			ret++;
684 		}
685 	}
686 	return ret;
687 }
688 
689 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
690 {
691 	u32 qk = MLX4_RESERVED_QKEY_BASE;
692 
693 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
694 	    qpn < dev->phys_caps.base_proxy_sqpn)
695 		return -EINVAL;
696 
697 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
698 		/* tunnel qp */
699 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
700 	else
701 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
702 	*qkey = qk;
703 	return 0;
704 }
705 EXPORT_SYMBOL(mlx4_get_parav_qkey);
706 
707 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
708 {
709 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
710 
711 	if (!mlx4_is_master(dev))
712 		return;
713 
714 	priv->virt2phys_pkey[slave][port - 1][i] = val;
715 }
716 EXPORT_SYMBOL(mlx4_sync_pkey_table);
717 
718 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
719 {
720 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
721 
722 	if (!mlx4_is_master(dev))
723 		return;
724 
725 	priv->slave_node_guids[slave] = guid;
726 }
727 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
728 
729 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
730 {
731 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
732 
733 	if (!mlx4_is_master(dev))
734 		return 0;
735 
736 	return priv->slave_node_guids[slave];
737 }
738 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
739 
740 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
741 {
742 	struct mlx4_priv *priv = mlx4_priv(dev);
743 	struct mlx4_slave_state *s_slave;
744 
745 	if (!mlx4_is_master(dev))
746 		return 0;
747 
748 	s_slave = &priv->mfunc.master.slave_state[slave];
749 	return !!s_slave->active;
750 }
751 EXPORT_SYMBOL(mlx4_is_slave_active);
752 
753 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
754 				       struct mlx4_dev_cap *dev_cap,
755 				       struct mlx4_init_hca_param *hca_param)
756 {
757 	dev->caps.steering_mode = hca_param->steering_mode;
758 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
759 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
760 		dev->caps.fs_log_max_ucast_qp_range_size =
761 			dev_cap->fs_log_max_ucast_qp_range_size;
762 	} else
763 		dev->caps.num_qp_per_mgm =
764 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
765 
766 	mlx4_dbg(dev, "Steering mode is: %s\n",
767 		 mlx4_steering_mode_str(dev->caps.steering_mode));
768 }
769 
770 static int mlx4_slave_cap(struct mlx4_dev *dev)
771 {
772 	int			   err;
773 	u32			   page_size;
774 	struct mlx4_dev_cap	   dev_cap;
775 	struct mlx4_func_cap	   func_cap;
776 	struct mlx4_init_hca_param hca_param;
777 	u8			   i;
778 
779 	memset(&hca_param, 0, sizeof(hca_param));
780 	err = mlx4_QUERY_HCA(dev, &hca_param);
781 	if (err) {
782 		mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
783 		return err;
784 	}
785 
786 	/* fail if the hca has an unknown global capability
787 	 * at this time global_caps should be always zeroed
788 	 */
789 	if (hca_param.global_caps) {
790 		mlx4_err(dev, "Unknown hca global capabilities\n");
791 		return -ENOSYS;
792 	}
793 
794 	mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
795 
796 	dev->caps.hca_core_clock = hca_param.hca_core_clock;
797 
798 	memset(&dev_cap, 0, sizeof(dev_cap));
799 	dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
800 	err = mlx4_dev_cap(dev, &dev_cap);
801 	if (err) {
802 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
803 		return err;
804 	}
805 
806 	err = mlx4_QUERY_FW(dev);
807 	if (err)
808 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
809 
810 	page_size = ~dev->caps.page_size_cap + 1;
811 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
812 	if (page_size > PAGE_SIZE) {
813 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
814 			 page_size, PAGE_SIZE);
815 		return -ENODEV;
816 	}
817 
818 	/* slave gets uar page size from QUERY_HCA fw command */
819 	dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
820 
821 	/* TODO: relax this assumption */
822 	if (dev->caps.uar_page_size != PAGE_SIZE) {
823 		mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
824 			 dev->caps.uar_page_size, PAGE_SIZE);
825 		return -ENODEV;
826 	}
827 
828 	memset(&func_cap, 0, sizeof(func_cap));
829 	err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
830 	if (err) {
831 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
832 			 err);
833 		return err;
834 	}
835 
836 	if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
837 	    PF_CONTEXT_BEHAVIOUR_MASK) {
838 		mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
839 			 func_cap.pf_context_behaviour, PF_CONTEXT_BEHAVIOUR_MASK);
840 		return -ENOSYS;
841 	}
842 
843 	dev->caps.num_ports		= func_cap.num_ports;
844 	dev->quotas.qp			= func_cap.qp_quota;
845 	dev->quotas.srq			= func_cap.srq_quota;
846 	dev->quotas.cq			= func_cap.cq_quota;
847 	dev->quotas.mpt			= func_cap.mpt_quota;
848 	dev->quotas.mtt			= func_cap.mtt_quota;
849 	dev->caps.num_qps		= 1 << hca_param.log_num_qps;
850 	dev->caps.num_srqs		= 1 << hca_param.log_num_srqs;
851 	dev->caps.num_cqs		= 1 << hca_param.log_num_cqs;
852 	dev->caps.num_mpts		= 1 << hca_param.log_mpt_sz;
853 	dev->caps.num_eqs		= func_cap.max_eq;
854 	dev->caps.reserved_eqs		= func_cap.reserved_eq;
855 	dev->caps.reserved_lkey		= func_cap.reserved_lkey;
856 	dev->caps.num_pds               = MLX4_NUM_PDS;
857 	dev->caps.num_mgms              = 0;
858 	dev->caps.num_amgms             = 0;
859 
860 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
861 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
862 			 dev->caps.num_ports, MLX4_MAX_PORTS);
863 		return -ENODEV;
864 	}
865 
866 	mlx4_replace_zero_macs(dev);
867 
868 	dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
869 	dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
870 	dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
871 	dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
872 	dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
873 
874 	if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
875 	    !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
876 	    !dev->caps.qp0_qkey) {
877 		err = -ENOMEM;
878 		goto err_mem;
879 	}
880 
881 	for (i = 1; i <= dev->caps.num_ports; ++i) {
882 		err = mlx4_QUERY_FUNC_CAP(dev, i, &func_cap);
883 		if (err) {
884 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
885 				 i, err);
886 			goto err_mem;
887 		}
888 		dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
889 		dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
890 		dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
891 		dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
892 		dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
893 		dev->caps.port_mask[i] = dev->caps.port_type[i];
894 		dev->caps.phys_port_id[i] = func_cap.phys_port_id;
895 		if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
896 						    &dev->caps.gid_table_len[i],
897 						    &dev->caps.pkey_table_len[i]))
898 			goto err_mem;
899 	}
900 
901 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
902 				       dev->caps.reserved_uars) >
903 				       pci_resource_len(dev->persist->pdev,
904 							2)) {
905 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
906 			 dev->caps.uar_page_size * dev->caps.num_uars,
907 			 (unsigned long long)
908 			 pci_resource_len(dev->persist->pdev, 2));
909 		goto err_mem;
910 	}
911 
912 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
913 		dev->caps.eqe_size   = 64;
914 		dev->caps.eqe_factor = 1;
915 	} else {
916 		dev->caps.eqe_size   = 32;
917 		dev->caps.eqe_factor = 0;
918 	}
919 
920 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
921 		dev->caps.cqe_size   = 64;
922 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
923 	} else {
924 		dev->caps.cqe_size   = 32;
925 	}
926 
927 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
928 		dev->caps.eqe_size = hca_param.eqe_size;
929 		dev->caps.eqe_factor = 0;
930 	}
931 
932 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
933 		dev->caps.cqe_size = hca_param.cqe_size;
934 		/* User still need to know when CQE > 32B */
935 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
936 	}
937 
938 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
939 	mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
940 
941 	slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
942 	mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
943 		 hca_param.rss_ip_frags ? "on" : "off");
944 
945 	if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
946 	    dev->caps.bf_reg_size)
947 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
948 
949 	if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
950 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
951 
952 	return 0;
953 
954 err_mem:
955 	kfree(dev->caps.qp0_qkey);
956 	kfree(dev->caps.qp0_tunnel);
957 	kfree(dev->caps.qp0_proxy);
958 	kfree(dev->caps.qp1_tunnel);
959 	kfree(dev->caps.qp1_proxy);
960 	dev->caps.qp0_qkey = NULL;
961 	dev->caps.qp0_tunnel = NULL;
962 	dev->caps.qp0_proxy = NULL;
963 	dev->caps.qp1_tunnel = NULL;
964 	dev->caps.qp1_proxy = NULL;
965 
966 	return err;
967 }
968 
969 static void mlx4_request_modules(struct mlx4_dev *dev)
970 {
971 	int port;
972 	int has_ib_port = false;
973 	int has_eth_port = false;
974 #define EN_DRV_NAME	"mlx4_en"
975 #define IB_DRV_NAME	"mlx4_ib"
976 
977 	for (port = 1; port <= dev->caps.num_ports; port++) {
978 		if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
979 			has_ib_port = true;
980 		else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
981 			has_eth_port = true;
982 	}
983 
984 	if (has_eth_port)
985 		request_module_nowait(EN_DRV_NAME);
986 	if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
987 		request_module_nowait(IB_DRV_NAME);
988 }
989 
990 /*
991  * Change the port configuration of the device.
992  * Every user of this function must hold the port mutex.
993  */
994 int mlx4_change_port_types(struct mlx4_dev *dev,
995 			   enum mlx4_port_type *port_types)
996 {
997 	int err = 0;
998 	int change = 0;
999 	int port;
1000 
1001 	for (port = 0; port <  dev->caps.num_ports; port++) {
1002 		/* Change the port type only if the new type is different
1003 		 * from the current, and not set to Auto */
1004 		if (port_types[port] != dev->caps.port_type[port + 1])
1005 			change = 1;
1006 	}
1007 	if (change) {
1008 		mlx4_unregister_device(dev);
1009 		for (port = 1; port <= dev->caps.num_ports; port++) {
1010 			mlx4_CLOSE_PORT(dev, port);
1011 			dev->caps.port_type[port] = port_types[port - 1];
1012 			err = mlx4_SET_PORT(dev, port, -1);
1013 			if (err) {
1014 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1015 					 port);
1016 				goto out;
1017 			}
1018 		}
1019 		mlx4_set_port_mask(dev);
1020 		err = mlx4_register_device(dev);
1021 		if (err) {
1022 			mlx4_err(dev, "Failed to register device\n");
1023 			goto out;
1024 		}
1025 		mlx4_request_modules(dev);
1026 	}
1027 
1028 out:
1029 	return err;
1030 }
1031 
1032 static ssize_t show_port_type(struct device *dev,
1033 			      struct device_attribute *attr,
1034 			      char *buf)
1035 {
1036 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1037 						   port_attr);
1038 	struct mlx4_dev *mdev = info->dev;
1039 	char type[8];
1040 
1041 	sprintf(type, "%s",
1042 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1043 		"ib" : "eth");
1044 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1045 		sprintf(buf, "auto (%s)\n", type);
1046 	else
1047 		sprintf(buf, "%s\n", type);
1048 
1049 	return strlen(buf);
1050 }
1051 
1052 static ssize_t set_port_type(struct device *dev,
1053 			     struct device_attribute *attr,
1054 			     const char *buf, size_t count)
1055 {
1056 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1057 						   port_attr);
1058 	struct mlx4_dev *mdev = info->dev;
1059 	struct mlx4_priv *priv = mlx4_priv(mdev);
1060 	enum mlx4_port_type types[MLX4_MAX_PORTS];
1061 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1062 	static DEFINE_MUTEX(set_port_type_mutex);
1063 	int i;
1064 	int err = 0;
1065 
1066 	mutex_lock(&set_port_type_mutex);
1067 
1068 	if (!strcmp(buf, "ib\n"))
1069 		info->tmp_type = MLX4_PORT_TYPE_IB;
1070 	else if (!strcmp(buf, "eth\n"))
1071 		info->tmp_type = MLX4_PORT_TYPE_ETH;
1072 	else if (!strcmp(buf, "auto\n"))
1073 		info->tmp_type = MLX4_PORT_TYPE_AUTO;
1074 	else {
1075 		mlx4_err(mdev, "%s is not supported port type\n", buf);
1076 		err = -EINVAL;
1077 		goto err_out;
1078 	}
1079 
1080 	mlx4_stop_sense(mdev);
1081 	mutex_lock(&priv->port_mutex);
1082 	/* Possible type is always the one that was delivered */
1083 	mdev->caps.possible_type[info->port] = info->tmp_type;
1084 
1085 	for (i = 0; i < mdev->caps.num_ports; i++) {
1086 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1087 					mdev->caps.possible_type[i+1];
1088 		if (types[i] == MLX4_PORT_TYPE_AUTO)
1089 			types[i] = mdev->caps.port_type[i+1];
1090 	}
1091 
1092 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1093 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1094 		for (i = 1; i <= mdev->caps.num_ports; i++) {
1095 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1096 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1097 				err = -EINVAL;
1098 			}
1099 		}
1100 	}
1101 	if (err) {
1102 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1103 		goto out;
1104 	}
1105 
1106 	mlx4_do_sense_ports(mdev, new_types, types);
1107 
1108 	err = mlx4_check_port_params(mdev, new_types);
1109 	if (err)
1110 		goto out;
1111 
1112 	/* We are about to apply the changes after the configuration
1113 	 * was verified, no need to remember the temporary types
1114 	 * any more */
1115 	for (i = 0; i < mdev->caps.num_ports; i++)
1116 		priv->port[i + 1].tmp_type = 0;
1117 
1118 	err = mlx4_change_port_types(mdev, new_types);
1119 
1120 out:
1121 	mlx4_start_sense(mdev);
1122 	mutex_unlock(&priv->port_mutex);
1123 err_out:
1124 	mutex_unlock(&set_port_type_mutex);
1125 
1126 	return err ? err : count;
1127 }
1128 
1129 enum ibta_mtu {
1130 	IB_MTU_256  = 1,
1131 	IB_MTU_512  = 2,
1132 	IB_MTU_1024 = 3,
1133 	IB_MTU_2048 = 4,
1134 	IB_MTU_4096 = 5
1135 };
1136 
1137 static inline int int_to_ibta_mtu(int mtu)
1138 {
1139 	switch (mtu) {
1140 	case 256:  return IB_MTU_256;
1141 	case 512:  return IB_MTU_512;
1142 	case 1024: return IB_MTU_1024;
1143 	case 2048: return IB_MTU_2048;
1144 	case 4096: return IB_MTU_4096;
1145 	default: return -1;
1146 	}
1147 }
1148 
1149 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1150 {
1151 	switch (mtu) {
1152 	case IB_MTU_256:  return  256;
1153 	case IB_MTU_512:  return  512;
1154 	case IB_MTU_1024: return 1024;
1155 	case IB_MTU_2048: return 2048;
1156 	case IB_MTU_4096: return 4096;
1157 	default: return -1;
1158 	}
1159 }
1160 
1161 static ssize_t show_port_ib_mtu(struct device *dev,
1162 			     struct device_attribute *attr,
1163 			     char *buf)
1164 {
1165 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1166 						   port_mtu_attr);
1167 	struct mlx4_dev *mdev = info->dev;
1168 
1169 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1170 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1171 
1172 	sprintf(buf, "%d\n",
1173 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1174 	return strlen(buf);
1175 }
1176 
1177 static ssize_t set_port_ib_mtu(struct device *dev,
1178 			     struct device_attribute *attr,
1179 			     const char *buf, size_t count)
1180 {
1181 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1182 						   port_mtu_attr);
1183 	struct mlx4_dev *mdev = info->dev;
1184 	struct mlx4_priv *priv = mlx4_priv(mdev);
1185 	int err, port, mtu, ibta_mtu = -1;
1186 
1187 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1188 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1189 		return -EINVAL;
1190 	}
1191 
1192 	err = kstrtoint(buf, 0, &mtu);
1193 	if (!err)
1194 		ibta_mtu = int_to_ibta_mtu(mtu);
1195 
1196 	if (err || ibta_mtu < 0) {
1197 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1198 		return -EINVAL;
1199 	}
1200 
1201 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1202 
1203 	mlx4_stop_sense(mdev);
1204 	mutex_lock(&priv->port_mutex);
1205 	mlx4_unregister_device(mdev);
1206 	for (port = 1; port <= mdev->caps.num_ports; port++) {
1207 		mlx4_CLOSE_PORT(mdev, port);
1208 		err = mlx4_SET_PORT(mdev, port, -1);
1209 		if (err) {
1210 			mlx4_err(mdev, "Failed to set port %d, aborting\n",
1211 				 port);
1212 			goto err_set_port;
1213 		}
1214 	}
1215 	err = mlx4_register_device(mdev);
1216 err_set_port:
1217 	mutex_unlock(&priv->port_mutex);
1218 	mlx4_start_sense(mdev);
1219 	return err ? err : count;
1220 }
1221 
1222 int mlx4_bond(struct mlx4_dev *dev)
1223 {
1224 	int ret = 0;
1225 	struct mlx4_priv *priv = mlx4_priv(dev);
1226 
1227 	mutex_lock(&priv->bond_mutex);
1228 
1229 	if (!mlx4_is_bonded(dev))
1230 		ret = mlx4_do_bond(dev, true);
1231 	else
1232 		ret = 0;
1233 
1234 	mutex_unlock(&priv->bond_mutex);
1235 	if (ret)
1236 		mlx4_err(dev, "Failed to bond device: %d\n", ret);
1237 	else
1238 		mlx4_dbg(dev, "Device is bonded\n");
1239 	return ret;
1240 }
1241 EXPORT_SYMBOL_GPL(mlx4_bond);
1242 
1243 int mlx4_unbond(struct mlx4_dev *dev)
1244 {
1245 	int ret = 0;
1246 	struct mlx4_priv *priv = mlx4_priv(dev);
1247 
1248 	mutex_lock(&priv->bond_mutex);
1249 
1250 	if (mlx4_is_bonded(dev))
1251 		ret = mlx4_do_bond(dev, false);
1252 
1253 	mutex_unlock(&priv->bond_mutex);
1254 	if (ret)
1255 		mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1256 	else
1257 		mlx4_dbg(dev, "Device is unbonded\n");
1258 	return ret;
1259 }
1260 EXPORT_SYMBOL_GPL(mlx4_unbond);
1261 
1262 
1263 int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1264 {
1265 	u8 port1 = v2p->port1;
1266 	u8 port2 = v2p->port2;
1267 	struct mlx4_priv *priv = mlx4_priv(dev);
1268 	int err;
1269 
1270 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1271 		return -ENOTSUPP;
1272 
1273 	mutex_lock(&priv->bond_mutex);
1274 
1275 	/* zero means keep current mapping for this port */
1276 	if (port1 == 0)
1277 		port1 = priv->v2p.port1;
1278 	if (port2 == 0)
1279 		port2 = priv->v2p.port2;
1280 
1281 	if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1282 	    (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1283 	    (port1 == 2 && port2 == 1)) {
1284 		/* besides boundary checks cross mapping makes
1285 		 * no sense and therefore not allowed */
1286 		err = -EINVAL;
1287 	} else if ((port1 == priv->v2p.port1) &&
1288 		 (port2 == priv->v2p.port2)) {
1289 		err = 0;
1290 	} else {
1291 		err = mlx4_virt2phy_port_map(dev, port1, port2);
1292 		if (!err) {
1293 			mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1294 				 port1, port2);
1295 			priv->v2p.port1 = port1;
1296 			priv->v2p.port2 = port2;
1297 		} else {
1298 			mlx4_err(dev, "Failed to change port mape: %d\n", err);
1299 		}
1300 	}
1301 
1302 	mutex_unlock(&priv->bond_mutex);
1303 	return err;
1304 }
1305 EXPORT_SYMBOL_GPL(mlx4_port_map_set);
1306 
1307 static int mlx4_load_fw(struct mlx4_dev *dev)
1308 {
1309 	struct mlx4_priv *priv = mlx4_priv(dev);
1310 	int err;
1311 
1312 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1313 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
1314 	if (!priv->fw.fw_icm) {
1315 		mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1316 		return -ENOMEM;
1317 	}
1318 
1319 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1320 	if (err) {
1321 		mlx4_err(dev, "MAP_FA command failed, aborting\n");
1322 		goto err_free;
1323 	}
1324 
1325 	err = mlx4_RUN_FW(dev);
1326 	if (err) {
1327 		mlx4_err(dev, "RUN_FW command failed, aborting\n");
1328 		goto err_unmap_fa;
1329 	}
1330 
1331 	return 0;
1332 
1333 err_unmap_fa:
1334 	mlx4_UNMAP_FA(dev);
1335 
1336 err_free:
1337 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1338 	return err;
1339 }
1340 
1341 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1342 				int cmpt_entry_sz)
1343 {
1344 	struct mlx4_priv *priv = mlx4_priv(dev);
1345 	int err;
1346 	int num_eqs;
1347 
1348 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1349 				  cmpt_base +
1350 				  ((u64) (MLX4_CMPT_TYPE_QP *
1351 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1352 				  cmpt_entry_sz, dev->caps.num_qps,
1353 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1354 				  0, 0);
1355 	if (err)
1356 		goto err;
1357 
1358 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1359 				  cmpt_base +
1360 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1361 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1362 				  cmpt_entry_sz, dev->caps.num_srqs,
1363 				  dev->caps.reserved_srqs, 0, 0);
1364 	if (err)
1365 		goto err_qp;
1366 
1367 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1368 				  cmpt_base +
1369 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1370 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1371 				  cmpt_entry_sz, dev->caps.num_cqs,
1372 				  dev->caps.reserved_cqs, 0, 0);
1373 	if (err)
1374 		goto err_srq;
1375 
1376 	num_eqs = dev->phys_caps.num_phys_eqs;
1377 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1378 				  cmpt_base +
1379 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1380 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1381 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1382 	if (err)
1383 		goto err_cq;
1384 
1385 	return 0;
1386 
1387 err_cq:
1388 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1389 
1390 err_srq:
1391 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1392 
1393 err_qp:
1394 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1395 
1396 err:
1397 	return err;
1398 }
1399 
1400 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1401 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1402 {
1403 	struct mlx4_priv *priv = mlx4_priv(dev);
1404 	u64 aux_pages;
1405 	int num_eqs;
1406 	int err;
1407 
1408 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1409 	if (err) {
1410 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1411 		return err;
1412 	}
1413 
1414 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1415 		 (unsigned long long) icm_size >> 10,
1416 		 (unsigned long long) aux_pages << 2);
1417 
1418 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1419 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1420 	if (!priv->fw.aux_icm) {
1421 		mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1422 		return -ENOMEM;
1423 	}
1424 
1425 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1426 	if (err) {
1427 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1428 		goto err_free_aux;
1429 	}
1430 
1431 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1432 	if (err) {
1433 		mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1434 		goto err_unmap_aux;
1435 	}
1436 
1437 
1438 	num_eqs = dev->phys_caps.num_phys_eqs;
1439 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1440 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1441 				  num_eqs, num_eqs, 0, 0);
1442 	if (err) {
1443 		mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1444 		goto err_unmap_cmpt;
1445 	}
1446 
1447 	/*
1448 	 * Reserved MTT entries must be aligned up to a cacheline
1449 	 * boundary, since the FW will write to them, while the driver
1450 	 * writes to all other MTT entries. (The variable
1451 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1452 	 * size, not the raw entry size)
1453 	 */
1454 	dev->caps.reserved_mtts =
1455 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1456 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1457 
1458 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1459 				  init_hca->mtt_base,
1460 				  dev->caps.mtt_entry_sz,
1461 				  dev->caps.num_mtts,
1462 				  dev->caps.reserved_mtts, 1, 0);
1463 	if (err) {
1464 		mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1465 		goto err_unmap_eq;
1466 	}
1467 
1468 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1469 				  init_hca->dmpt_base,
1470 				  dev_cap->dmpt_entry_sz,
1471 				  dev->caps.num_mpts,
1472 				  dev->caps.reserved_mrws, 1, 1);
1473 	if (err) {
1474 		mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1475 		goto err_unmap_mtt;
1476 	}
1477 
1478 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1479 				  init_hca->qpc_base,
1480 				  dev_cap->qpc_entry_sz,
1481 				  dev->caps.num_qps,
1482 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1483 				  0, 0);
1484 	if (err) {
1485 		mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1486 		goto err_unmap_dmpt;
1487 	}
1488 
1489 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1490 				  init_hca->auxc_base,
1491 				  dev_cap->aux_entry_sz,
1492 				  dev->caps.num_qps,
1493 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1494 				  0, 0);
1495 	if (err) {
1496 		mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1497 		goto err_unmap_qp;
1498 	}
1499 
1500 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1501 				  init_hca->altc_base,
1502 				  dev_cap->altc_entry_sz,
1503 				  dev->caps.num_qps,
1504 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1505 				  0, 0);
1506 	if (err) {
1507 		mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1508 		goto err_unmap_auxc;
1509 	}
1510 
1511 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1512 				  init_hca->rdmarc_base,
1513 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1514 				  dev->caps.num_qps,
1515 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1516 				  0, 0);
1517 	if (err) {
1518 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1519 		goto err_unmap_altc;
1520 	}
1521 
1522 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1523 				  init_hca->cqc_base,
1524 				  dev_cap->cqc_entry_sz,
1525 				  dev->caps.num_cqs,
1526 				  dev->caps.reserved_cqs, 0, 0);
1527 	if (err) {
1528 		mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1529 		goto err_unmap_rdmarc;
1530 	}
1531 
1532 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1533 				  init_hca->srqc_base,
1534 				  dev_cap->srq_entry_sz,
1535 				  dev->caps.num_srqs,
1536 				  dev->caps.reserved_srqs, 0, 0);
1537 	if (err) {
1538 		mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1539 		goto err_unmap_cq;
1540 	}
1541 
1542 	/*
1543 	 * For flow steering device managed mode it is required to use
1544 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1545 	 * required, but for simplicity just map the whole multicast
1546 	 * group table now.  The table isn't very big and it's a lot
1547 	 * easier than trying to track ref counts.
1548 	 */
1549 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1550 				  init_hca->mc_base,
1551 				  mlx4_get_mgm_entry_size(dev),
1552 				  dev->caps.num_mgms + dev->caps.num_amgms,
1553 				  dev->caps.num_mgms + dev->caps.num_amgms,
1554 				  0, 0);
1555 	if (err) {
1556 		mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1557 		goto err_unmap_srq;
1558 	}
1559 
1560 	return 0;
1561 
1562 err_unmap_srq:
1563 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1564 
1565 err_unmap_cq:
1566 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1567 
1568 err_unmap_rdmarc:
1569 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1570 
1571 err_unmap_altc:
1572 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1573 
1574 err_unmap_auxc:
1575 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1576 
1577 err_unmap_qp:
1578 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1579 
1580 err_unmap_dmpt:
1581 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1582 
1583 err_unmap_mtt:
1584 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1585 
1586 err_unmap_eq:
1587 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1588 
1589 err_unmap_cmpt:
1590 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1591 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1592 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1593 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1594 
1595 err_unmap_aux:
1596 	mlx4_UNMAP_ICM_AUX(dev);
1597 
1598 err_free_aux:
1599 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1600 
1601 	return err;
1602 }
1603 
1604 static void mlx4_free_icms(struct mlx4_dev *dev)
1605 {
1606 	struct mlx4_priv *priv = mlx4_priv(dev);
1607 
1608 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1609 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1610 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1611 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1612 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1613 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1614 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1615 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1616 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1617 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1618 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1619 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1620 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1621 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1622 
1623 	mlx4_UNMAP_ICM_AUX(dev);
1624 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1625 }
1626 
1627 static void mlx4_slave_exit(struct mlx4_dev *dev)
1628 {
1629 	struct mlx4_priv *priv = mlx4_priv(dev);
1630 
1631 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1632 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1633 			  MLX4_COMM_TIME))
1634 		mlx4_warn(dev, "Failed to close slave function\n");
1635 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1636 }
1637 
1638 static int map_bf_area(struct mlx4_dev *dev)
1639 {
1640 	struct mlx4_priv *priv = mlx4_priv(dev);
1641 	resource_size_t bf_start;
1642 	resource_size_t bf_len;
1643 	int err = 0;
1644 
1645 	if (!dev->caps.bf_reg_size)
1646 		return -ENXIO;
1647 
1648 	bf_start = pci_resource_start(dev->persist->pdev, 2) +
1649 			(dev->caps.num_uars << PAGE_SHIFT);
1650 	bf_len = pci_resource_len(dev->persist->pdev, 2) -
1651 			(dev->caps.num_uars << PAGE_SHIFT);
1652 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1653 	if (!priv->bf_mapping)
1654 		err = -ENOMEM;
1655 
1656 	return err;
1657 }
1658 
1659 static void unmap_bf_area(struct mlx4_dev *dev)
1660 {
1661 	if (mlx4_priv(dev)->bf_mapping)
1662 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1663 }
1664 
1665 cycle_t mlx4_read_clock(struct mlx4_dev *dev)
1666 {
1667 	u32 clockhi, clocklo, clockhi1;
1668 	cycle_t cycles;
1669 	int i;
1670 	struct mlx4_priv *priv = mlx4_priv(dev);
1671 
1672 	for (i = 0; i < 10; i++) {
1673 		clockhi = swab32(readl(priv->clock_mapping));
1674 		clocklo = swab32(readl(priv->clock_mapping + 4));
1675 		clockhi1 = swab32(readl(priv->clock_mapping));
1676 		if (clockhi == clockhi1)
1677 			break;
1678 	}
1679 
1680 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1681 
1682 	return cycles;
1683 }
1684 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1685 
1686 
1687 static int map_internal_clock(struct mlx4_dev *dev)
1688 {
1689 	struct mlx4_priv *priv = mlx4_priv(dev);
1690 
1691 	priv->clock_mapping =
1692 		ioremap(pci_resource_start(dev->persist->pdev,
1693 					   priv->fw.clock_bar) +
1694 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1695 
1696 	if (!priv->clock_mapping)
1697 		return -ENOMEM;
1698 
1699 	return 0;
1700 }
1701 
1702 int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1703 				   struct mlx4_clock_params *params)
1704 {
1705 	struct mlx4_priv *priv = mlx4_priv(dev);
1706 
1707 	if (mlx4_is_slave(dev))
1708 		return -ENOTSUPP;
1709 
1710 	if (!params)
1711 		return -EINVAL;
1712 
1713 	params->bar = priv->fw.clock_bar;
1714 	params->offset = priv->fw.clock_offset;
1715 	params->size = MLX4_CLOCK_SIZE;
1716 
1717 	return 0;
1718 }
1719 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1720 
1721 static void unmap_internal_clock(struct mlx4_dev *dev)
1722 {
1723 	struct mlx4_priv *priv = mlx4_priv(dev);
1724 
1725 	if (priv->clock_mapping)
1726 		iounmap(priv->clock_mapping);
1727 }
1728 
1729 static void mlx4_close_hca(struct mlx4_dev *dev)
1730 {
1731 	unmap_internal_clock(dev);
1732 	unmap_bf_area(dev);
1733 	if (mlx4_is_slave(dev))
1734 		mlx4_slave_exit(dev);
1735 	else {
1736 		mlx4_CLOSE_HCA(dev, 0);
1737 		mlx4_free_icms(dev);
1738 	}
1739 }
1740 
1741 static void mlx4_close_fw(struct mlx4_dev *dev)
1742 {
1743 	if (!mlx4_is_slave(dev)) {
1744 		mlx4_UNMAP_FA(dev);
1745 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1746 	}
1747 }
1748 
1749 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1750 {
1751 #define COMM_CHAN_OFFLINE_OFFSET 0x09
1752 
1753 	u32 comm_flags;
1754 	u32 offline_bit;
1755 	unsigned long end;
1756 	struct mlx4_priv *priv = mlx4_priv(dev);
1757 
1758 	end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
1759 	while (time_before(jiffies, end)) {
1760 		comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
1761 					  MLX4_COMM_CHAN_FLAGS));
1762 		offline_bit = (comm_flags &
1763 			       (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
1764 		if (!offline_bit)
1765 			return 0;
1766 		/* There are cases as part of AER/Reset flow that PF needs
1767 		 * around 100 msec to load. We therefore sleep for 100 msec
1768 		 * to allow other tasks to make use of that CPU during this
1769 		 * time interval.
1770 		 */
1771 		msleep(100);
1772 	}
1773 	mlx4_err(dev, "Communication channel is offline.\n");
1774 	return -EIO;
1775 }
1776 
1777 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
1778 {
1779 #define COMM_CHAN_RST_OFFSET 0x1e
1780 
1781 	struct mlx4_priv *priv = mlx4_priv(dev);
1782 	u32 comm_rst;
1783 	u32 comm_caps;
1784 
1785 	comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
1786 				 MLX4_COMM_CHAN_CAPS));
1787 	comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
1788 
1789 	if (comm_rst)
1790 		dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
1791 }
1792 
1793 static int mlx4_init_slave(struct mlx4_dev *dev)
1794 {
1795 	struct mlx4_priv *priv = mlx4_priv(dev);
1796 	u64 dma = (u64) priv->mfunc.vhcr_dma;
1797 	int ret_from_reset = 0;
1798 	u32 slave_read;
1799 	u32 cmd_channel_ver;
1800 
1801 	if (atomic_read(&pf_loading)) {
1802 		mlx4_warn(dev, "PF is not ready - Deferring probe\n");
1803 		return -EPROBE_DEFER;
1804 	}
1805 
1806 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1807 	priv->cmd.max_cmds = 1;
1808 	if (mlx4_comm_check_offline(dev)) {
1809 		mlx4_err(dev, "PF is not responsive, skipping initialization\n");
1810 		goto err_offline;
1811 	}
1812 
1813 	mlx4_reset_vf_support(dev);
1814 	mlx4_warn(dev, "Sending reset\n");
1815 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1816 				       MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
1817 	/* if we are in the middle of flr the slave will try
1818 	 * NUM_OF_RESET_RETRIES times before leaving.*/
1819 	if (ret_from_reset) {
1820 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1821 			mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
1822 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
1823 			return -EPROBE_DEFER;
1824 		} else
1825 			goto err;
1826 	}
1827 
1828 	/* check the driver version - the slave I/F revision
1829 	 * must match the master's */
1830 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1831 	cmd_channel_ver = mlx4_comm_get_version();
1832 
1833 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1834 		MLX4_COMM_GET_IF_REV(slave_read)) {
1835 		mlx4_err(dev, "slave driver version is not supported by the master\n");
1836 		goto err;
1837 	}
1838 
1839 	mlx4_warn(dev, "Sending vhcr0\n");
1840 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1841 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1842 		goto err;
1843 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1844 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1845 		goto err;
1846 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1847 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1848 		goto err;
1849 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
1850 			  MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1851 		goto err;
1852 
1853 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1854 	return 0;
1855 
1856 err:
1857 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
1858 err_offline:
1859 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1860 	return -EIO;
1861 }
1862 
1863 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1864 {
1865 	int i;
1866 
1867 	for (i = 1; i <= dev->caps.num_ports; i++) {
1868 		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
1869 			dev->caps.gid_table_len[i] =
1870 				mlx4_get_slave_num_gids(dev, 0, i);
1871 		else
1872 			dev->caps.gid_table_len[i] = 1;
1873 		dev->caps.pkey_table_len[i] =
1874 			dev->phys_caps.pkey_phys_table_len[i] - 1;
1875 	}
1876 }
1877 
1878 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
1879 {
1880 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
1881 
1882 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
1883 	      i++) {
1884 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
1885 			break;
1886 	}
1887 
1888 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
1889 }
1890 
1891 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
1892 {
1893 	switch (dmfs_high_steer_mode) {
1894 	case MLX4_STEERING_DMFS_A0_DEFAULT:
1895 		return "default performance";
1896 
1897 	case MLX4_STEERING_DMFS_A0_DYNAMIC:
1898 		return "dynamic hybrid mode";
1899 
1900 	case MLX4_STEERING_DMFS_A0_STATIC:
1901 		return "performance optimized for limited rule configuration (static)";
1902 
1903 	case MLX4_STEERING_DMFS_A0_DISABLE:
1904 		return "disabled performance optimized steering";
1905 
1906 	case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
1907 		return "performance optimized steering not supported";
1908 
1909 	default:
1910 		return "Unrecognized mode";
1911 	}
1912 }
1913 
1914 #define MLX4_DMFS_A0_STEERING			(1UL << 2)
1915 
1916 static void choose_steering_mode(struct mlx4_dev *dev,
1917 				 struct mlx4_dev_cap *dev_cap)
1918 {
1919 	if (mlx4_log_num_mgm_entry_size <= 0) {
1920 		if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
1921 			if (dev->caps.dmfs_high_steer_mode ==
1922 			    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1923 				mlx4_err(dev, "DMFS high rate mode not supported\n");
1924 			else
1925 				dev->caps.dmfs_high_steer_mode =
1926 					MLX4_STEERING_DMFS_A0_STATIC;
1927 		}
1928 	}
1929 
1930 	if (mlx4_log_num_mgm_entry_size <= 0 &&
1931 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
1932 	    (!mlx4_is_mfunc(dev) ||
1933 	     (dev_cap->fs_max_num_qp_per_entry >=
1934 	     (dev->persist->num_vfs + 1))) &&
1935 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
1936 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
1937 		dev->oper_log_mgm_entry_size =
1938 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
1939 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1940 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
1941 		dev->caps.fs_log_max_ucast_qp_range_size =
1942 			dev_cap->fs_log_max_ucast_qp_range_size;
1943 	} else {
1944 		if (dev->caps.dmfs_high_steer_mode !=
1945 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1946 			dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
1947 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
1948 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1949 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
1950 		else {
1951 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
1952 
1953 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
1954 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1955 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
1956 		}
1957 		dev->oper_log_mgm_entry_size =
1958 			mlx4_log_num_mgm_entry_size > 0 ?
1959 			mlx4_log_num_mgm_entry_size :
1960 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
1961 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
1962 	}
1963 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
1964 		 mlx4_steering_mode_str(dev->caps.steering_mode),
1965 		 dev->oper_log_mgm_entry_size,
1966 		 mlx4_log_num_mgm_entry_size);
1967 }
1968 
1969 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
1970 				       struct mlx4_dev_cap *dev_cap)
1971 {
1972 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
1973 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
1974 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
1975 	else
1976 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
1977 
1978 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
1979 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
1980 }
1981 
1982 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
1983 {
1984 	int i;
1985 	struct mlx4_port_cap port_cap;
1986 
1987 	if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1988 		return -EINVAL;
1989 
1990 	for (i = 1; i <= dev->caps.num_ports; i++) {
1991 		if (mlx4_dev_port(dev, i, &port_cap)) {
1992 			mlx4_err(dev,
1993 				 "QUERY_DEV_CAP command failed, can't veify DMFS high rate steering.\n");
1994 		} else if ((dev->caps.dmfs_high_steer_mode !=
1995 			    MLX4_STEERING_DMFS_A0_DEFAULT) &&
1996 			   (port_cap.dmfs_optimized_state ==
1997 			    !!(dev->caps.dmfs_high_steer_mode ==
1998 			    MLX4_STEERING_DMFS_A0_DISABLE))) {
1999 			mlx4_err(dev,
2000 				 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2001 				 dmfs_high_rate_steering_mode_str(
2002 					dev->caps.dmfs_high_steer_mode),
2003 				 (port_cap.dmfs_optimized_state ?
2004 					"enabled" : "disabled"));
2005 		}
2006 	}
2007 
2008 	return 0;
2009 }
2010 
2011 static int mlx4_init_fw(struct mlx4_dev *dev)
2012 {
2013 	struct mlx4_mod_stat_cfg   mlx4_cfg;
2014 	int err = 0;
2015 
2016 	if (!mlx4_is_slave(dev)) {
2017 		err = mlx4_QUERY_FW(dev);
2018 		if (err) {
2019 			if (err == -EACCES)
2020 				mlx4_info(dev, "non-primary physical function, skipping\n");
2021 			else
2022 				mlx4_err(dev, "QUERY_FW command failed, aborting\n");
2023 			return err;
2024 		}
2025 
2026 		err = mlx4_load_fw(dev);
2027 		if (err) {
2028 			mlx4_err(dev, "Failed to start FW, aborting\n");
2029 			return err;
2030 		}
2031 
2032 		mlx4_cfg.log_pg_sz_m = 1;
2033 		mlx4_cfg.log_pg_sz = 0;
2034 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2035 		if (err)
2036 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2037 	}
2038 
2039 	return err;
2040 }
2041 
2042 static int mlx4_init_hca(struct mlx4_dev *dev)
2043 {
2044 	struct mlx4_priv	  *priv = mlx4_priv(dev);
2045 	struct mlx4_adapter	   adapter;
2046 	struct mlx4_dev_cap	   dev_cap;
2047 	struct mlx4_profile	   profile;
2048 	struct mlx4_init_hca_param init_hca;
2049 	u64 icm_size;
2050 	struct mlx4_config_dev_params params;
2051 	int err;
2052 
2053 	if (!mlx4_is_slave(dev)) {
2054 		err = mlx4_dev_cap(dev, &dev_cap);
2055 		if (err) {
2056 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
2057 			return err;
2058 		}
2059 
2060 		choose_steering_mode(dev, &dev_cap);
2061 		choose_tunnel_offload_mode(dev, &dev_cap);
2062 
2063 		if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2064 		    mlx4_is_master(dev))
2065 			dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2066 
2067 		err = mlx4_get_phys_port_id(dev);
2068 		if (err)
2069 			mlx4_err(dev, "Fail to get physical port id\n");
2070 
2071 		if (mlx4_is_master(dev))
2072 			mlx4_parav_master_pf_caps(dev);
2073 
2074 		if (mlx4_low_memory_profile()) {
2075 			mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2076 			profile = low_mem_profile;
2077 		} else {
2078 			profile = default_profile;
2079 		}
2080 		if (dev->caps.steering_mode ==
2081 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
2082 			profile.num_mcg = MLX4_FS_NUM_MCG;
2083 
2084 		icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
2085 					     &init_hca);
2086 		if ((long long) icm_size < 0) {
2087 			err = icm_size;
2088 			return err;
2089 		}
2090 
2091 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
2092 
2093 		init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
2094 		init_hca.uar_page_sz = PAGE_SHIFT - 12;
2095 		init_hca.mw_enabled = 0;
2096 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2097 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2098 			init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
2099 
2100 		err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
2101 		if (err)
2102 			return err;
2103 
2104 		err = mlx4_INIT_HCA(dev, &init_hca);
2105 		if (err) {
2106 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
2107 			goto err_free_icm;
2108 		}
2109 
2110 		if (dev_cap.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2111 			err = mlx4_query_func(dev, &dev_cap);
2112 			if (err < 0) {
2113 				mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
2114 				goto err_close;
2115 			} else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2116 				dev->caps.num_eqs = dev_cap.max_eqs;
2117 				dev->caps.reserved_eqs = dev_cap.reserved_eqs;
2118 				dev->caps.reserved_uars = dev_cap.reserved_uars;
2119 			}
2120 		}
2121 
2122 		/*
2123 		 * If TS is supported by FW
2124 		 * read HCA frequency by QUERY_HCA command
2125 		 */
2126 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2127 			memset(&init_hca, 0, sizeof(init_hca));
2128 			err = mlx4_QUERY_HCA(dev, &init_hca);
2129 			if (err) {
2130 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
2131 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2132 			} else {
2133 				dev->caps.hca_core_clock =
2134 					init_hca.hca_core_clock;
2135 			}
2136 
2137 			/* In case we got HCA frequency 0 - disable timestamping
2138 			 * to avoid dividing by zero
2139 			 */
2140 			if (!dev->caps.hca_core_clock) {
2141 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2142 				mlx4_err(dev,
2143 					 "HCA frequency is 0 - timestamping is not supported\n");
2144 			} else if (map_internal_clock(dev)) {
2145 				/*
2146 				 * Map internal clock,
2147 				 * in case of failure disable timestamping
2148 				 */
2149 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2150 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
2151 			}
2152 		}
2153 
2154 		if (dev->caps.dmfs_high_steer_mode !=
2155 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2156 			if (mlx4_validate_optimized_steering(dev))
2157 				mlx4_warn(dev, "Optimized steering validation failed\n");
2158 
2159 			if (dev->caps.dmfs_high_steer_mode ==
2160 			    MLX4_STEERING_DMFS_A0_DISABLE) {
2161 				dev->caps.dmfs_high_rate_qpn_base =
2162 					dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2163 				dev->caps.dmfs_high_rate_qpn_range =
2164 					MLX4_A0_STEERING_TABLE_SIZE;
2165 			}
2166 
2167 			mlx4_dbg(dev, "DMFS high rate steer mode is: %s\n",
2168 				 dmfs_high_rate_steering_mode_str(
2169 					dev->caps.dmfs_high_steer_mode));
2170 		}
2171 	} else {
2172 		err = mlx4_init_slave(dev);
2173 		if (err) {
2174 			if (err != -EPROBE_DEFER)
2175 				mlx4_err(dev, "Failed to initialize slave\n");
2176 			return err;
2177 		}
2178 
2179 		err = mlx4_slave_cap(dev);
2180 		if (err) {
2181 			mlx4_err(dev, "Failed to obtain slave caps\n");
2182 			goto err_close;
2183 		}
2184 	}
2185 
2186 	if (map_bf_area(dev))
2187 		mlx4_dbg(dev, "Failed to map blue flame area\n");
2188 
2189 	/*Only the master set the ports, all the rest got it from it.*/
2190 	if (!mlx4_is_slave(dev))
2191 		mlx4_set_port_mask(dev);
2192 
2193 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
2194 	if (err) {
2195 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2196 		goto unmap_bf;
2197 	}
2198 
2199 	/* Query CONFIG_DEV parameters */
2200 	err = mlx4_config_dev_retrieval(dev, &params);
2201 	if (err && err != -ENOTSUPP) {
2202 		mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2203 	} else if (!err) {
2204 		dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2205 		dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2206 	}
2207 	priv->eq_table.inta_pin = adapter.inta_pin;
2208 	memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
2209 
2210 	return 0;
2211 
2212 unmap_bf:
2213 	unmap_internal_clock(dev);
2214 	unmap_bf_area(dev);
2215 
2216 	if (mlx4_is_slave(dev)) {
2217 		kfree(dev->caps.qp0_qkey);
2218 		kfree(dev->caps.qp0_tunnel);
2219 		kfree(dev->caps.qp0_proxy);
2220 		kfree(dev->caps.qp1_tunnel);
2221 		kfree(dev->caps.qp1_proxy);
2222 	}
2223 
2224 err_close:
2225 	if (mlx4_is_slave(dev))
2226 		mlx4_slave_exit(dev);
2227 	else
2228 		mlx4_CLOSE_HCA(dev, 0);
2229 
2230 err_free_icm:
2231 	if (!mlx4_is_slave(dev))
2232 		mlx4_free_icms(dev);
2233 
2234 	return err;
2235 }
2236 
2237 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2238 {
2239 	struct mlx4_priv *priv = mlx4_priv(dev);
2240 	int nent_pow2;
2241 
2242 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2243 		return -ENOENT;
2244 
2245 	if (!dev->caps.max_counters)
2246 		return -ENOSPC;
2247 
2248 	nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2249 	/* reserve last counter index for sink counter */
2250 	return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2251 				nent_pow2 - 1, 0,
2252 				nent_pow2 - dev->caps.max_counters + 1);
2253 }
2254 
2255 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2256 {
2257 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2258 		return;
2259 
2260 	if (!dev->caps.max_counters)
2261 		return;
2262 
2263 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2264 }
2265 
2266 static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2267 {
2268 	struct mlx4_priv *priv = mlx4_priv(dev);
2269 	int port;
2270 
2271 	for (port = 0; port < dev->caps.num_ports; port++)
2272 		if (priv->def_counter[port] != -1)
2273 			mlx4_counter_free(dev,  priv->def_counter[port]);
2274 }
2275 
2276 static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2277 {
2278 	struct mlx4_priv *priv = mlx4_priv(dev);
2279 	int port, err = 0;
2280 	u32 idx;
2281 
2282 	for (port = 0; port < dev->caps.num_ports; port++)
2283 		priv->def_counter[port] = -1;
2284 
2285 	for (port = 0; port < dev->caps.num_ports; port++) {
2286 		err = mlx4_counter_alloc(dev, &idx);
2287 
2288 		if (!err || err == -ENOSPC) {
2289 			priv->def_counter[port] = idx;
2290 		} else if (err == -ENOENT) {
2291 			err = 0;
2292 			continue;
2293 		} else if (mlx4_is_slave(dev) && err == -EINVAL) {
2294 			priv->def_counter[port] = MLX4_SINK_COUNTER_INDEX(dev);
2295 			mlx4_warn(dev, "can't allocate counter from old PF driver, using index %d\n",
2296 				  MLX4_SINK_COUNTER_INDEX(dev));
2297 			err = 0;
2298 		} else {
2299 			mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2300 				 __func__, port + 1, err);
2301 			mlx4_cleanup_default_counters(dev);
2302 			return err;
2303 		}
2304 
2305 		mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2306 			 __func__, priv->def_counter[port], port + 1);
2307 	}
2308 
2309 	return err;
2310 }
2311 
2312 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2313 {
2314 	struct mlx4_priv *priv = mlx4_priv(dev);
2315 
2316 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2317 		return -ENOENT;
2318 
2319 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2320 	if (*idx == -1) {
2321 		*idx = MLX4_SINK_COUNTER_INDEX(dev);
2322 		return -ENOSPC;
2323 	}
2324 
2325 	return 0;
2326 }
2327 
2328 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2329 {
2330 	u64 out_param;
2331 	int err;
2332 
2333 	if (mlx4_is_mfunc(dev)) {
2334 		err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
2335 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2336 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2337 		if (!err)
2338 			*idx = get_param_l(&out_param);
2339 
2340 		return err;
2341 	}
2342 	return __mlx4_counter_alloc(dev, idx);
2343 }
2344 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2345 
2346 static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2347 				u8 counter_index)
2348 {
2349 	struct mlx4_cmd_mailbox *if_stat_mailbox;
2350 	int err;
2351 	u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2352 
2353 	if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2354 	if (IS_ERR(if_stat_mailbox))
2355 		return PTR_ERR(if_stat_mailbox);
2356 
2357 	err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2358 			   MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2359 			   MLX4_CMD_NATIVE);
2360 
2361 	mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2362 	return err;
2363 }
2364 
2365 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2366 {
2367 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2368 		return;
2369 
2370 	if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2371 		return;
2372 
2373 	__mlx4_clear_if_stat(dev, idx);
2374 
2375 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2376 	return;
2377 }
2378 
2379 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2380 {
2381 	u64 in_param = 0;
2382 
2383 	if (mlx4_is_mfunc(dev)) {
2384 		set_param_l(&in_param, idx);
2385 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2386 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2387 			 MLX4_CMD_WRAPPED);
2388 		return;
2389 	}
2390 	__mlx4_counter_free(dev, idx);
2391 }
2392 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2393 
2394 int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2395 {
2396 	struct mlx4_priv *priv = mlx4_priv(dev);
2397 
2398 	return priv->def_counter[port - 1];
2399 }
2400 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2401 
2402 void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2403 {
2404 	struct mlx4_priv *priv = mlx4_priv(dev);
2405 
2406 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2407 }
2408 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2409 
2410 __be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2411 {
2412 	struct mlx4_priv *priv = mlx4_priv(dev);
2413 
2414 	return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2415 }
2416 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2417 
2418 void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2419 {
2420 	struct mlx4_priv *priv = mlx4_priv(dev);
2421 	__be64 guid;
2422 
2423 	/* hw GUID */
2424 	if (entry == 0)
2425 		return;
2426 
2427 	get_random_bytes((char *)&guid, sizeof(guid));
2428 	guid &= ~(cpu_to_be64(1ULL << 56));
2429 	guid |= cpu_to_be64(1ULL << 57);
2430 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2431 }
2432 
2433 static int mlx4_setup_hca(struct mlx4_dev *dev)
2434 {
2435 	struct mlx4_priv *priv = mlx4_priv(dev);
2436 	int err;
2437 	int port;
2438 	__be32 ib_port_default_caps;
2439 
2440 	err = mlx4_init_uar_table(dev);
2441 	if (err) {
2442 		mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2443 		 return err;
2444 	}
2445 
2446 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
2447 	if (err) {
2448 		mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2449 		goto err_uar_table_free;
2450 	}
2451 
2452 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2453 	if (!priv->kar) {
2454 		mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2455 		err = -ENOMEM;
2456 		goto err_uar_free;
2457 	}
2458 
2459 	err = mlx4_init_pd_table(dev);
2460 	if (err) {
2461 		mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2462 		goto err_kar_unmap;
2463 	}
2464 
2465 	err = mlx4_init_xrcd_table(dev);
2466 	if (err) {
2467 		mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2468 		goto err_pd_table_free;
2469 	}
2470 
2471 	err = mlx4_init_mr_table(dev);
2472 	if (err) {
2473 		mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2474 		goto err_xrcd_table_free;
2475 	}
2476 
2477 	if (!mlx4_is_slave(dev)) {
2478 		err = mlx4_init_mcg_table(dev);
2479 		if (err) {
2480 			mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2481 			goto err_mr_table_free;
2482 		}
2483 		err = mlx4_config_mad_demux(dev);
2484 		if (err) {
2485 			mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2486 			goto err_mcg_table_free;
2487 		}
2488 	}
2489 
2490 	err = mlx4_init_eq_table(dev);
2491 	if (err) {
2492 		mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2493 		goto err_mcg_table_free;
2494 	}
2495 
2496 	err = mlx4_cmd_use_events(dev);
2497 	if (err) {
2498 		mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2499 		goto err_eq_table_free;
2500 	}
2501 
2502 	err = mlx4_NOP(dev);
2503 	if (err) {
2504 		if (dev->flags & MLX4_FLAG_MSI_X) {
2505 			mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2506 				  priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2507 			mlx4_warn(dev, "Trying again without MSI-X\n");
2508 		} else {
2509 			mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2510 				 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2511 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2512 		}
2513 
2514 		goto err_cmd_poll;
2515 	}
2516 
2517 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
2518 
2519 	err = mlx4_init_cq_table(dev);
2520 	if (err) {
2521 		mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2522 		goto err_cmd_poll;
2523 	}
2524 
2525 	err = mlx4_init_srq_table(dev);
2526 	if (err) {
2527 		mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2528 		goto err_cq_table_free;
2529 	}
2530 
2531 	err = mlx4_init_qp_table(dev);
2532 	if (err) {
2533 		mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2534 		goto err_srq_table_free;
2535 	}
2536 
2537 	if (!mlx4_is_slave(dev)) {
2538 		err = mlx4_init_counters_table(dev);
2539 		if (err && err != -ENOENT) {
2540 			mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2541 			goto err_qp_table_free;
2542 		}
2543 	}
2544 
2545 	err = mlx4_allocate_default_counters(dev);
2546 	if (err) {
2547 		mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2548 		goto err_counters_table_free;
2549 	}
2550 
2551 	if (!mlx4_is_slave(dev)) {
2552 		for (port = 1; port <= dev->caps.num_ports; port++) {
2553 			ib_port_default_caps = 0;
2554 			err = mlx4_get_port_ib_caps(dev, port,
2555 						    &ib_port_default_caps);
2556 			if (err)
2557 				mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2558 					  port, err);
2559 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2560 
2561 			/* initialize per-slave default ib port capabilities */
2562 			if (mlx4_is_master(dev)) {
2563 				int i;
2564 				for (i = 0; i < dev->num_slaves; i++) {
2565 					if (i == mlx4_master_func_num(dev))
2566 						continue;
2567 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2568 						ib_port_default_caps;
2569 				}
2570 			}
2571 
2572 			if (mlx4_is_mfunc(dev))
2573 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2574 			else
2575 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2576 
2577 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2578 					    dev->caps.pkey_table_len[port] : -1);
2579 			if (err) {
2580 				mlx4_err(dev, "Failed to set port %d, aborting\n",
2581 					 port);
2582 				goto err_default_countes_free;
2583 			}
2584 		}
2585 	}
2586 
2587 	return 0;
2588 
2589 err_default_countes_free:
2590 	mlx4_cleanup_default_counters(dev);
2591 
2592 err_counters_table_free:
2593 	if (!mlx4_is_slave(dev))
2594 		mlx4_cleanup_counters_table(dev);
2595 
2596 err_qp_table_free:
2597 	mlx4_cleanup_qp_table(dev);
2598 
2599 err_srq_table_free:
2600 	mlx4_cleanup_srq_table(dev);
2601 
2602 err_cq_table_free:
2603 	mlx4_cleanup_cq_table(dev);
2604 
2605 err_cmd_poll:
2606 	mlx4_cmd_use_polling(dev);
2607 
2608 err_eq_table_free:
2609 	mlx4_cleanup_eq_table(dev);
2610 
2611 err_mcg_table_free:
2612 	if (!mlx4_is_slave(dev))
2613 		mlx4_cleanup_mcg_table(dev);
2614 
2615 err_mr_table_free:
2616 	mlx4_cleanup_mr_table(dev);
2617 
2618 err_xrcd_table_free:
2619 	mlx4_cleanup_xrcd_table(dev);
2620 
2621 err_pd_table_free:
2622 	mlx4_cleanup_pd_table(dev);
2623 
2624 err_kar_unmap:
2625 	iounmap(priv->kar);
2626 
2627 err_uar_free:
2628 	mlx4_uar_free(dev, &priv->driver_uar);
2629 
2630 err_uar_table_free:
2631 	mlx4_cleanup_uar_table(dev);
2632 	return err;
2633 }
2634 
2635 static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2636 {
2637 	int requested_cpu = 0;
2638 	struct mlx4_priv *priv = mlx4_priv(dev);
2639 	struct mlx4_eq *eq;
2640 	int off = 0;
2641 	int i;
2642 
2643 	if (eqn > dev->caps.num_comp_vectors)
2644 		return -EINVAL;
2645 
2646 	for (i = 1; i < port; i++)
2647 		off += mlx4_get_eqs_per_port(dev, i);
2648 
2649 	requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2650 
2651 	/* Meaning EQs are shared, and this call comes from the second port */
2652 	if (requested_cpu < 0)
2653 		return 0;
2654 
2655 	eq = &priv->eq_table.eq[eqn];
2656 
2657 	if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2658 		return -ENOMEM;
2659 
2660 	cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2661 
2662 	return 0;
2663 }
2664 
2665 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2666 {
2667 	struct mlx4_priv *priv = mlx4_priv(dev);
2668 	struct msix_entry *entries;
2669 	int i;
2670 	int port = 0;
2671 
2672 	if (msi_x) {
2673 		int nreq = dev->caps.num_ports * num_online_cpus() + 1;
2674 
2675 		nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
2676 			     nreq);
2677 		if (nreq > MAX_MSIX)
2678 			nreq = MAX_MSIX;
2679 
2680 		entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
2681 		if (!entries)
2682 			goto no_msi;
2683 
2684 		for (i = 0; i < nreq; ++i)
2685 			entries[i].entry = i;
2686 
2687 		nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2688 					     nreq);
2689 
2690 		if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
2691 			kfree(entries);
2692 			goto no_msi;
2693 		}
2694 		/* 1 is reserved for events (asyncrounous EQ) */
2695 		dev->caps.num_comp_vectors = nreq - 1;
2696 
2697 		priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2698 		bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2699 			    dev->caps.num_ports);
2700 
2701 		for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2702 			if (i == MLX4_EQ_ASYNC)
2703 				continue;
2704 
2705 			priv->eq_table.eq[i].irq =
2706 				entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2707 
2708 			if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2709 				bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2710 					    dev->caps.num_ports);
2711 				/* We don't set affinity hint when there
2712 				 * aren't enough EQs
2713 				 */
2714 			} else {
2715 				set_bit(port,
2716 					priv->eq_table.eq[i].actv_ports.ports);
2717 				if (mlx4_init_affinity_hint(dev, port + 1, i))
2718 					mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
2719 						  i);
2720 			}
2721 			/* We divide the Eqs evenly between the two ports.
2722 			 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
2723 			 * refers to the number of Eqs per port
2724 			 * (i.e eqs_per_port). Theoretically, we would like to
2725 			 * write something like (i + 1) % eqs_per_port == 0.
2726 			 * However, since there's an asynchronous Eq, we have
2727 			 * to skip over it by comparing this condition to
2728 			 * !!((i + 1) > MLX4_EQ_ASYNC).
2729 			 */
2730 			if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
2731 			    ((i + 1) %
2732 			     (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
2733 			    !!((i + 1) > MLX4_EQ_ASYNC))
2734 				/* If dev->caps.num_comp_vectors < dev->caps.num_ports,
2735 				 * everything is shared anyway.
2736 				 */
2737 				port++;
2738 		}
2739 
2740 		dev->flags |= MLX4_FLAG_MSI_X;
2741 
2742 		kfree(entries);
2743 		return;
2744 	}
2745 
2746 no_msi:
2747 	dev->caps.num_comp_vectors = 1;
2748 
2749 	BUG_ON(MLX4_EQ_ASYNC >= 2);
2750 	for (i = 0; i < 2; ++i) {
2751 		priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
2752 		if (i != MLX4_EQ_ASYNC) {
2753 			bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2754 				    dev->caps.num_ports);
2755 		}
2756 	}
2757 }
2758 
2759 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2760 {
2761 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2762 	int err = 0;
2763 
2764 	info->dev = dev;
2765 	info->port = port;
2766 	if (!mlx4_is_slave(dev)) {
2767 		mlx4_init_mac_table(dev, &info->mac_table);
2768 		mlx4_init_vlan_table(dev, &info->vlan_table);
2769 		mlx4_init_roce_gid_table(dev, &info->gid_table);
2770 		info->base_qpn = mlx4_get_base_qpn(dev, port);
2771 	}
2772 
2773 	sprintf(info->dev_name, "mlx4_port%d", port);
2774 	info->port_attr.attr.name = info->dev_name;
2775 	if (mlx4_is_mfunc(dev))
2776 		info->port_attr.attr.mode = S_IRUGO;
2777 	else {
2778 		info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2779 		info->port_attr.store     = set_port_type;
2780 	}
2781 	info->port_attr.show      = show_port_type;
2782 	sysfs_attr_init(&info->port_attr.attr);
2783 
2784 	err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
2785 	if (err) {
2786 		mlx4_err(dev, "Failed to create file for port %d\n", port);
2787 		info->port = -1;
2788 	}
2789 
2790 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2791 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
2792 	if (mlx4_is_mfunc(dev))
2793 		info->port_mtu_attr.attr.mode = S_IRUGO;
2794 	else {
2795 		info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2796 		info->port_mtu_attr.store     = set_port_ib_mtu;
2797 	}
2798 	info->port_mtu_attr.show      = show_port_ib_mtu;
2799 	sysfs_attr_init(&info->port_mtu_attr.attr);
2800 
2801 	err = device_create_file(&dev->persist->pdev->dev,
2802 				 &info->port_mtu_attr);
2803 	if (err) {
2804 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
2805 		device_remove_file(&info->dev->persist->pdev->dev,
2806 				   &info->port_attr);
2807 		info->port = -1;
2808 	}
2809 
2810 	return err;
2811 }
2812 
2813 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
2814 {
2815 	if (info->port < 0)
2816 		return;
2817 
2818 	device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
2819 	device_remove_file(&info->dev->persist->pdev->dev,
2820 			   &info->port_mtu_attr);
2821 #ifdef CONFIG_RFS_ACCEL
2822 	free_irq_cpu_rmap(info->rmap);
2823 	info->rmap = NULL;
2824 #endif
2825 }
2826 
2827 static int mlx4_init_steering(struct mlx4_dev *dev)
2828 {
2829 	struct mlx4_priv *priv = mlx4_priv(dev);
2830 	int num_entries = dev->caps.num_ports;
2831 	int i, j;
2832 
2833 	priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
2834 	if (!priv->steer)
2835 		return -ENOMEM;
2836 
2837 	for (i = 0; i < num_entries; i++)
2838 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
2839 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
2840 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
2841 		}
2842 	return 0;
2843 }
2844 
2845 static void mlx4_clear_steering(struct mlx4_dev *dev)
2846 {
2847 	struct mlx4_priv *priv = mlx4_priv(dev);
2848 	struct mlx4_steer_index *entry, *tmp_entry;
2849 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
2850 	int num_entries = dev->caps.num_ports;
2851 	int i, j;
2852 
2853 	for (i = 0; i < num_entries; i++) {
2854 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
2855 			list_for_each_entry_safe(pqp, tmp_pqp,
2856 						 &priv->steer[i].promisc_qps[j],
2857 						 list) {
2858 				list_del(&pqp->list);
2859 				kfree(pqp);
2860 			}
2861 			list_for_each_entry_safe(entry, tmp_entry,
2862 						 &priv->steer[i].steer_entries[j],
2863 						 list) {
2864 				list_del(&entry->list);
2865 				list_for_each_entry_safe(pqp, tmp_pqp,
2866 							 &entry->duplicates,
2867 							 list) {
2868 					list_del(&pqp->list);
2869 					kfree(pqp);
2870 				}
2871 				kfree(entry);
2872 			}
2873 		}
2874 	}
2875 	kfree(priv->steer);
2876 }
2877 
2878 static int extended_func_num(struct pci_dev *pdev)
2879 {
2880 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
2881 }
2882 
2883 #define MLX4_OWNER_BASE	0x8069c
2884 #define MLX4_OWNER_SIZE	4
2885 
2886 static int mlx4_get_ownership(struct mlx4_dev *dev)
2887 {
2888 	void __iomem *owner;
2889 	u32 ret;
2890 
2891 	if (pci_channel_offline(dev->persist->pdev))
2892 		return -EIO;
2893 
2894 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
2895 			MLX4_OWNER_BASE,
2896 			MLX4_OWNER_SIZE);
2897 	if (!owner) {
2898 		mlx4_err(dev, "Failed to obtain ownership bit\n");
2899 		return -ENOMEM;
2900 	}
2901 
2902 	ret = readl(owner);
2903 	iounmap(owner);
2904 	return (int) !!ret;
2905 }
2906 
2907 static void mlx4_free_ownership(struct mlx4_dev *dev)
2908 {
2909 	void __iomem *owner;
2910 
2911 	if (pci_channel_offline(dev->persist->pdev))
2912 		return;
2913 
2914 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
2915 			MLX4_OWNER_BASE,
2916 			MLX4_OWNER_SIZE);
2917 	if (!owner) {
2918 		mlx4_err(dev, "Failed to obtain ownership bit\n");
2919 		return;
2920 	}
2921 	writel(0, owner);
2922 	msleep(1000);
2923 	iounmap(owner);
2924 }
2925 
2926 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV)	==\
2927 				  !!((flags) & MLX4_FLAG_MASTER))
2928 
2929 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
2930 			     u8 total_vfs, int existing_vfs, int reset_flow)
2931 {
2932 	u64 dev_flags = dev->flags;
2933 	int err = 0;
2934 	int fw_enabled_sriov_vfs = min(pci_sriov_get_totalvfs(pdev),
2935 					MLX4_MAX_NUM_VF);
2936 
2937 	if (reset_flow) {
2938 		dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
2939 				       GFP_KERNEL);
2940 		if (!dev->dev_vfs)
2941 			goto free_mem;
2942 		return dev_flags;
2943 	}
2944 
2945 	atomic_inc(&pf_loading);
2946 	if (dev->flags &  MLX4_FLAG_SRIOV) {
2947 		if (existing_vfs != total_vfs) {
2948 			mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
2949 				 existing_vfs, total_vfs);
2950 			total_vfs = existing_vfs;
2951 		}
2952 	}
2953 
2954 	dev->dev_vfs = kzalloc(total_vfs * sizeof(*dev->dev_vfs), GFP_KERNEL);
2955 	if (NULL == dev->dev_vfs) {
2956 		mlx4_err(dev, "Failed to allocate memory for VFs\n");
2957 		goto disable_sriov;
2958 	}
2959 
2960 	if (!(dev->flags &  MLX4_FLAG_SRIOV)) {
2961 		if (total_vfs > fw_enabled_sriov_vfs) {
2962 			mlx4_err(dev, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
2963 				 total_vfs, fw_enabled_sriov_vfs);
2964 			err = -ENOMEM;
2965 			goto disable_sriov;
2966 		}
2967 		mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
2968 		err = pci_enable_sriov(pdev, total_vfs);
2969 	}
2970 	if (err) {
2971 		mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
2972 			 err);
2973 		goto disable_sriov;
2974 	} else {
2975 		mlx4_warn(dev, "Running in master mode\n");
2976 		dev_flags |= MLX4_FLAG_SRIOV |
2977 			MLX4_FLAG_MASTER;
2978 		dev_flags &= ~MLX4_FLAG_SLAVE;
2979 		dev->persist->num_vfs = total_vfs;
2980 	}
2981 	return dev_flags;
2982 
2983 disable_sriov:
2984 	atomic_dec(&pf_loading);
2985 free_mem:
2986 	dev->persist->num_vfs = 0;
2987 	kfree(dev->dev_vfs);
2988         dev->dev_vfs = NULL;
2989 	return dev_flags & ~MLX4_FLAG_MASTER;
2990 }
2991 
2992 enum {
2993 	MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
2994 };
2995 
2996 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
2997 			      int *nvfs)
2998 {
2999 	int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
3000 	/* Checking for 64 VFs as a limitation of CX2 */
3001 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
3002 	    requested_vfs >= 64) {
3003 		mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
3004 			 requested_vfs);
3005 		return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
3006 	}
3007 	return 0;
3008 }
3009 
3010 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
3011 			 int total_vfs, int *nvfs, struct mlx4_priv *priv,
3012 			 int reset_flow)
3013 {
3014 	struct mlx4_dev *dev;
3015 	unsigned sum = 0;
3016 	int err;
3017 	int port;
3018 	int i;
3019 	struct mlx4_dev_cap *dev_cap = NULL;
3020 	int existing_vfs = 0;
3021 
3022 	dev = &priv->dev;
3023 
3024 	INIT_LIST_HEAD(&priv->ctx_list);
3025 	spin_lock_init(&priv->ctx_lock);
3026 
3027 	mutex_init(&priv->port_mutex);
3028 	mutex_init(&priv->bond_mutex);
3029 
3030 	INIT_LIST_HEAD(&priv->pgdir_list);
3031 	mutex_init(&priv->pgdir_mutex);
3032 
3033 	INIT_LIST_HEAD(&priv->bf_list);
3034 	mutex_init(&priv->bf_mutex);
3035 
3036 	dev->rev_id = pdev->revision;
3037 	dev->numa_node = dev_to_node(&pdev->dev);
3038 
3039 	/* Detect if this device is a virtual function */
3040 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3041 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3042 		dev->flags |= MLX4_FLAG_SLAVE;
3043 	} else {
3044 		/* We reset the device and enable SRIOV only for physical
3045 		 * devices.  Try to claim ownership on the device;
3046 		 * if already taken, skip -- do not allow multiple PFs */
3047 		err = mlx4_get_ownership(dev);
3048 		if (err) {
3049 			if (err < 0)
3050 				return err;
3051 			else {
3052 				mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
3053 				return -EINVAL;
3054 			}
3055 		}
3056 
3057 		atomic_set(&priv->opreq_count, 0);
3058 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3059 
3060 		/*
3061 		 * Now reset the HCA before we touch the PCI capabilities or
3062 		 * attempt a firmware command, since a boot ROM may have left
3063 		 * the HCA in an undefined state.
3064 		 */
3065 		err = mlx4_reset(dev);
3066 		if (err) {
3067 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
3068 			goto err_sriov;
3069 		}
3070 
3071 		if (total_vfs) {
3072 			dev->flags = MLX4_FLAG_MASTER;
3073 			existing_vfs = pci_num_vf(pdev);
3074 			if (existing_vfs)
3075 				dev->flags |= MLX4_FLAG_SRIOV;
3076 			dev->persist->num_vfs = total_vfs;
3077 		}
3078 	}
3079 
3080 	/* on load remove any previous indication of internal error,
3081 	 * device is up.
3082 	 */
3083 	dev->persist->state = MLX4_DEVICE_STATE_UP;
3084 
3085 slave_start:
3086 	err = mlx4_cmd_init(dev);
3087 	if (err) {
3088 		mlx4_err(dev, "Failed to init command interface, aborting\n");
3089 		goto err_sriov;
3090 	}
3091 
3092 	/* In slave functions, the communication channel must be initialized
3093 	 * before posting commands. Also, init num_slaves before calling
3094 	 * mlx4_init_hca */
3095 	if (mlx4_is_mfunc(dev)) {
3096 		if (mlx4_is_master(dev)) {
3097 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
3098 
3099 		} else {
3100 			dev->num_slaves = 0;
3101 			err = mlx4_multi_func_init(dev);
3102 			if (err) {
3103 				mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
3104 				goto err_cmd;
3105 			}
3106 		}
3107 	}
3108 
3109 	err = mlx4_init_fw(dev);
3110 	if (err) {
3111 		mlx4_err(dev, "Failed to init fw, aborting.\n");
3112 		goto err_mfunc;
3113 	}
3114 
3115 	if (mlx4_is_master(dev)) {
3116 		/* when we hit the goto slave_start below, dev_cap already initialized */
3117 		if (!dev_cap) {
3118 			dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3119 
3120 			if (!dev_cap) {
3121 				err = -ENOMEM;
3122 				goto err_fw;
3123 			}
3124 
3125 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3126 			if (err) {
3127 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3128 				goto err_fw;
3129 			}
3130 
3131 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3132 				goto err_fw;
3133 
3134 			if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3135 				u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3136 								  total_vfs,
3137 								  existing_vfs,
3138 								  reset_flow);
3139 
3140 				mlx4_close_fw(dev);
3141 				mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3142 				dev->flags = dev_flags;
3143 				if (!SRIOV_VALID_STATE(dev->flags)) {
3144 					mlx4_err(dev, "Invalid SRIOV state\n");
3145 					goto err_sriov;
3146 				}
3147 				err = mlx4_reset(dev);
3148 				if (err) {
3149 					mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3150 					goto err_sriov;
3151 				}
3152 				goto slave_start;
3153 			}
3154 		} else {
3155 			/* Legacy mode FW requires SRIOV to be enabled before
3156 			 * doing QUERY_DEV_CAP, since max_eq's value is different if
3157 			 * SRIOV is enabled.
3158 			 */
3159 			memset(dev_cap, 0, sizeof(*dev_cap));
3160 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3161 			if (err) {
3162 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3163 				goto err_fw;
3164 			}
3165 
3166 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3167 				goto err_fw;
3168 		}
3169 	}
3170 
3171 	err = mlx4_init_hca(dev);
3172 	if (err) {
3173 		if (err == -EACCES) {
3174 			/* Not primary Physical function
3175 			 * Running in slave mode */
3176 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3177 			/* We're not a PF */
3178 			if (dev->flags & MLX4_FLAG_SRIOV) {
3179 				if (!existing_vfs)
3180 					pci_disable_sriov(pdev);
3181 				if (mlx4_is_master(dev) && !reset_flow)
3182 					atomic_dec(&pf_loading);
3183 				dev->flags &= ~MLX4_FLAG_SRIOV;
3184 			}
3185 			if (!mlx4_is_slave(dev))
3186 				mlx4_free_ownership(dev);
3187 			dev->flags |= MLX4_FLAG_SLAVE;
3188 			dev->flags &= ~MLX4_FLAG_MASTER;
3189 			goto slave_start;
3190 		} else
3191 			goto err_fw;
3192 	}
3193 
3194 	if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3195 		u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3196 						  existing_vfs, reset_flow);
3197 
3198 		if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3199 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3200 			dev->flags = dev_flags;
3201 			err = mlx4_cmd_init(dev);
3202 			if (err) {
3203 				/* Only VHCR is cleaned up, so could still
3204 				 * send FW commands
3205 				 */
3206 				mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3207 				goto err_close;
3208 			}
3209 		} else {
3210 			dev->flags = dev_flags;
3211 		}
3212 
3213 		if (!SRIOV_VALID_STATE(dev->flags)) {
3214 			mlx4_err(dev, "Invalid SRIOV state\n");
3215 			goto err_close;
3216 		}
3217 	}
3218 
3219 	/* check if the device is functioning at its maximum possible speed.
3220 	 * No return code for this call, just warn the user in case of PCI
3221 	 * express device capabilities are under-satisfied by the bus.
3222 	 */
3223 	if (!mlx4_is_slave(dev))
3224 		mlx4_check_pcie_caps(dev);
3225 
3226 	/* In master functions, the communication channel must be initialized
3227 	 * after obtaining its address from fw */
3228 	if (mlx4_is_master(dev)) {
3229 		if (dev->caps.num_ports < 2 &&
3230 		    num_vfs_argc > 1) {
3231 			err = -EINVAL;
3232 			mlx4_err(dev,
3233 				 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3234 				 dev->caps.num_ports);
3235 			goto err_close;
3236 		}
3237 		memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
3238 
3239 		for (i = 0;
3240 		     i < sizeof(dev->persist->nvfs)/
3241 		     sizeof(dev->persist->nvfs[0]); i++) {
3242 			unsigned j;
3243 
3244 			for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
3245 				dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3246 				dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3247 					dev->caps.num_ports;
3248 			}
3249 		}
3250 
3251 		/* In master functions, the communication channel
3252 		 * must be initialized after obtaining its address from fw
3253 		 */
3254 		err = mlx4_multi_func_init(dev);
3255 		if (err) {
3256 			mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3257 			goto err_close;
3258 		}
3259 	}
3260 
3261 	err = mlx4_alloc_eq_table(dev);
3262 	if (err)
3263 		goto err_master_mfunc;
3264 
3265 	bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
3266 	mutex_init(&priv->msix_ctl.pool_lock);
3267 
3268 	mlx4_enable_msi_x(dev);
3269 	if ((mlx4_is_mfunc(dev)) &&
3270 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
3271 		err = -ENOSYS;
3272 		mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
3273 		goto err_free_eq;
3274 	}
3275 
3276 	if (!mlx4_is_slave(dev)) {
3277 		err = mlx4_init_steering(dev);
3278 		if (err)
3279 			goto err_disable_msix;
3280 	}
3281 
3282 	err = mlx4_setup_hca(dev);
3283 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3284 	    !mlx4_is_mfunc(dev)) {
3285 		dev->flags &= ~MLX4_FLAG_MSI_X;
3286 		dev->caps.num_comp_vectors = 1;
3287 		pci_disable_msix(pdev);
3288 		err = mlx4_setup_hca(dev);
3289 	}
3290 
3291 	if (err)
3292 		goto err_steer;
3293 
3294 	mlx4_init_quotas(dev);
3295 	/* When PF resources are ready arm its comm channel to enable
3296 	 * getting commands
3297 	 */
3298 	if (mlx4_is_master(dev)) {
3299 		err = mlx4_ARM_COMM_CHANNEL(dev);
3300 		if (err) {
3301 			mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3302 				 err);
3303 			goto err_steer;
3304 		}
3305 	}
3306 
3307 	for (port = 1; port <= dev->caps.num_ports; port++) {
3308 		err = mlx4_init_port_info(dev, port);
3309 		if (err)
3310 			goto err_port;
3311 	}
3312 
3313 	priv->v2p.port1 = 1;
3314 	priv->v2p.port2 = 2;
3315 
3316 	err = mlx4_register_device(dev);
3317 	if (err)
3318 		goto err_port;
3319 
3320 	mlx4_request_modules(dev);
3321 
3322 	mlx4_sense_init(dev);
3323 	mlx4_start_sense(dev);
3324 
3325 	priv->removed = 0;
3326 
3327 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3328 		atomic_dec(&pf_loading);
3329 
3330 	kfree(dev_cap);
3331 	return 0;
3332 
3333 err_port:
3334 	for (--port; port >= 1; --port)
3335 		mlx4_cleanup_port_info(&priv->port[port]);
3336 
3337 	mlx4_cleanup_default_counters(dev);
3338 	if (!mlx4_is_slave(dev))
3339 		mlx4_cleanup_counters_table(dev);
3340 	mlx4_cleanup_qp_table(dev);
3341 	mlx4_cleanup_srq_table(dev);
3342 	mlx4_cleanup_cq_table(dev);
3343 	mlx4_cmd_use_polling(dev);
3344 	mlx4_cleanup_eq_table(dev);
3345 	mlx4_cleanup_mcg_table(dev);
3346 	mlx4_cleanup_mr_table(dev);
3347 	mlx4_cleanup_xrcd_table(dev);
3348 	mlx4_cleanup_pd_table(dev);
3349 	mlx4_cleanup_uar_table(dev);
3350 
3351 err_steer:
3352 	if (!mlx4_is_slave(dev))
3353 		mlx4_clear_steering(dev);
3354 
3355 err_disable_msix:
3356 	if (dev->flags & MLX4_FLAG_MSI_X)
3357 		pci_disable_msix(pdev);
3358 
3359 err_free_eq:
3360 	mlx4_free_eq_table(dev);
3361 
3362 err_master_mfunc:
3363 	if (mlx4_is_master(dev)) {
3364 		mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
3365 		mlx4_multi_func_cleanup(dev);
3366 	}
3367 
3368 	if (mlx4_is_slave(dev)) {
3369 		kfree(dev->caps.qp0_qkey);
3370 		kfree(dev->caps.qp0_tunnel);
3371 		kfree(dev->caps.qp0_proxy);
3372 		kfree(dev->caps.qp1_tunnel);
3373 		kfree(dev->caps.qp1_proxy);
3374 	}
3375 
3376 err_close:
3377 	mlx4_close_hca(dev);
3378 
3379 err_fw:
3380 	mlx4_close_fw(dev);
3381 
3382 err_mfunc:
3383 	if (mlx4_is_slave(dev))
3384 		mlx4_multi_func_cleanup(dev);
3385 
3386 err_cmd:
3387 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3388 
3389 err_sriov:
3390 	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3391 		pci_disable_sriov(pdev);
3392 		dev->flags &= ~MLX4_FLAG_SRIOV;
3393 	}
3394 
3395 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3396 		atomic_dec(&pf_loading);
3397 
3398 	kfree(priv->dev.dev_vfs);
3399 
3400 	if (!mlx4_is_slave(dev))
3401 		mlx4_free_ownership(dev);
3402 
3403 	kfree(dev_cap);
3404 	return err;
3405 }
3406 
3407 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3408 			   struct mlx4_priv *priv)
3409 {
3410 	int err;
3411 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3412 	int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3413 	const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3414 		{2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3415 	unsigned total_vfs = 0;
3416 	unsigned int i;
3417 
3418 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3419 
3420 	err = pci_enable_device(pdev);
3421 	if (err) {
3422 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3423 		return err;
3424 	}
3425 
3426 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3427 	 * per port, we must limit the number of VFs to 63 (since their are
3428 	 * 128 MACs)
3429 	 */
3430 	for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
3431 	     total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3432 		nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3433 		if (nvfs[i] < 0) {
3434 			dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3435 			err = -EINVAL;
3436 			goto err_disable_pdev;
3437 		}
3438 	}
3439 	for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
3440 	     i++) {
3441 		prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3442 		if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3443 			dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3444 			err = -EINVAL;
3445 			goto err_disable_pdev;
3446 		}
3447 	}
3448 	if (total_vfs > MLX4_MAX_NUM_VF) {
3449 		dev_err(&pdev->dev,
3450 			"Requested more VF's (%d) than allowed by hw (%d)\n",
3451 			total_vfs, MLX4_MAX_NUM_VF);
3452 		err = -EINVAL;
3453 		goto err_disable_pdev;
3454 	}
3455 
3456 	for (i = 0; i < MLX4_MAX_PORTS; i++) {
3457 		if (nvfs[i] + nvfs[2] > MLX4_MAX_NUM_VF_P_PORT) {
3458 			dev_err(&pdev->dev,
3459 				"Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3460 				nvfs[i] + nvfs[2], i + 1,
3461 				MLX4_MAX_NUM_VF_P_PORT);
3462 			err = -EINVAL;
3463 			goto err_disable_pdev;
3464 		}
3465 	}
3466 
3467 	/* Check for BARs. */
3468 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3469 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3470 		dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3471 			pci_dev_data, pci_resource_flags(pdev, 0));
3472 		err = -ENODEV;
3473 		goto err_disable_pdev;
3474 	}
3475 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3476 		dev_err(&pdev->dev, "Missing UAR, aborting\n");
3477 		err = -ENODEV;
3478 		goto err_disable_pdev;
3479 	}
3480 
3481 	err = pci_request_regions(pdev, DRV_NAME);
3482 	if (err) {
3483 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3484 		goto err_disable_pdev;
3485 	}
3486 
3487 	pci_set_master(pdev);
3488 
3489 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3490 	if (err) {
3491 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3492 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3493 		if (err) {
3494 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3495 			goto err_release_regions;
3496 		}
3497 	}
3498 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3499 	if (err) {
3500 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3501 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3502 		if (err) {
3503 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3504 			goto err_release_regions;
3505 		}
3506 	}
3507 
3508 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
3509 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3510 	/* Detect if this device is a virtual function */
3511 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3512 		/* When acting as pf, we normally skip vfs unless explicitly
3513 		 * requested to probe them.
3514 		 */
3515 		if (total_vfs) {
3516 			unsigned vfs_offset = 0;
3517 
3518 			for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
3519 			     vfs_offset + nvfs[i] < extended_func_num(pdev);
3520 			     vfs_offset += nvfs[i], i++)
3521 				;
3522 			if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
3523 				err = -ENODEV;
3524 				goto err_release_regions;
3525 			}
3526 			if ((extended_func_num(pdev) - vfs_offset)
3527 			    > prb_vf[i]) {
3528 				dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3529 					 extended_func_num(pdev));
3530 				err = -ENODEV;
3531 				goto err_release_regions;
3532 			}
3533 		}
3534 	}
3535 
3536 	err = mlx4_catas_init(&priv->dev);
3537 	if (err)
3538 		goto err_release_regions;
3539 
3540 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3541 	if (err)
3542 		goto err_catas;
3543 
3544 	return 0;
3545 
3546 err_catas:
3547 	mlx4_catas_end(&priv->dev);
3548 
3549 err_release_regions:
3550 	pci_release_regions(pdev);
3551 
3552 err_disable_pdev:
3553 	pci_disable_device(pdev);
3554 	pci_set_drvdata(pdev, NULL);
3555 	return err;
3556 }
3557 
3558 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3559 {
3560 	struct mlx4_priv *priv;
3561 	struct mlx4_dev *dev;
3562 	int ret;
3563 
3564 	printk_once(KERN_INFO "%s", mlx4_version);
3565 
3566 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
3567 	if (!priv)
3568 		return -ENOMEM;
3569 
3570 	dev       = &priv->dev;
3571 	dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3572 	if (!dev->persist) {
3573 		kfree(priv);
3574 		return -ENOMEM;
3575 	}
3576 	dev->persist->pdev = pdev;
3577 	dev->persist->dev = dev;
3578 	pci_set_drvdata(pdev, dev->persist);
3579 	priv->pci_dev_data = id->driver_data;
3580 	mutex_init(&dev->persist->device_state_mutex);
3581 	mutex_init(&dev->persist->interface_state_mutex);
3582 
3583 	ret =  __mlx4_init_one(pdev, id->driver_data, priv);
3584 	if (ret) {
3585 		kfree(dev->persist);
3586 		kfree(priv);
3587 	} else {
3588 		pci_save_state(pdev);
3589 	}
3590 
3591 	return ret;
3592 }
3593 
3594 static void mlx4_clean_dev(struct mlx4_dev *dev)
3595 {
3596 	struct mlx4_dev_persistent *persist = dev->persist;
3597 	struct mlx4_priv *priv = mlx4_priv(dev);
3598 	unsigned long	flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
3599 
3600 	memset(priv, 0, sizeof(*priv));
3601 	priv->dev.persist = persist;
3602 	priv->dev.flags = flags;
3603 }
3604 
3605 static void mlx4_unload_one(struct pci_dev *pdev)
3606 {
3607 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3608 	struct mlx4_dev  *dev  = persist->dev;
3609 	struct mlx4_priv *priv = mlx4_priv(dev);
3610 	int               pci_dev_data;
3611 	int p, i;
3612 
3613 	if (priv->removed)
3614 		return;
3615 
3616 	/* saving current ports type for further use */
3617 	for (i = 0; i < dev->caps.num_ports; i++) {
3618 		dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
3619 		dev->persist->curr_port_poss_type[i] = dev->caps.
3620 						       possible_type[i + 1];
3621 	}
3622 
3623 	pci_dev_data = priv->pci_dev_data;
3624 
3625 	mlx4_stop_sense(dev);
3626 	mlx4_unregister_device(dev);
3627 
3628 	for (p = 1; p <= dev->caps.num_ports; p++) {
3629 		mlx4_cleanup_port_info(&priv->port[p]);
3630 		mlx4_CLOSE_PORT(dev, p);
3631 	}
3632 
3633 	if (mlx4_is_master(dev))
3634 		mlx4_free_resource_tracker(dev,
3635 					   RES_TR_FREE_SLAVES_ONLY);
3636 
3637 	mlx4_cleanup_default_counters(dev);
3638 	if (!mlx4_is_slave(dev))
3639 		mlx4_cleanup_counters_table(dev);
3640 	mlx4_cleanup_qp_table(dev);
3641 	mlx4_cleanup_srq_table(dev);
3642 	mlx4_cleanup_cq_table(dev);
3643 	mlx4_cmd_use_polling(dev);
3644 	mlx4_cleanup_eq_table(dev);
3645 	mlx4_cleanup_mcg_table(dev);
3646 	mlx4_cleanup_mr_table(dev);
3647 	mlx4_cleanup_xrcd_table(dev);
3648 	mlx4_cleanup_pd_table(dev);
3649 
3650 	if (mlx4_is_master(dev))
3651 		mlx4_free_resource_tracker(dev,
3652 					   RES_TR_FREE_STRUCTS_ONLY);
3653 
3654 	iounmap(priv->kar);
3655 	mlx4_uar_free(dev, &priv->driver_uar);
3656 	mlx4_cleanup_uar_table(dev);
3657 	if (!mlx4_is_slave(dev))
3658 		mlx4_clear_steering(dev);
3659 	mlx4_free_eq_table(dev);
3660 	if (mlx4_is_master(dev))
3661 		mlx4_multi_func_cleanup(dev);
3662 	mlx4_close_hca(dev);
3663 	mlx4_close_fw(dev);
3664 	if (mlx4_is_slave(dev))
3665 		mlx4_multi_func_cleanup(dev);
3666 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3667 
3668 	if (dev->flags & MLX4_FLAG_MSI_X)
3669 		pci_disable_msix(pdev);
3670 
3671 	if (!mlx4_is_slave(dev))
3672 		mlx4_free_ownership(dev);
3673 
3674 	kfree(dev->caps.qp0_qkey);
3675 	kfree(dev->caps.qp0_tunnel);
3676 	kfree(dev->caps.qp0_proxy);
3677 	kfree(dev->caps.qp1_tunnel);
3678 	kfree(dev->caps.qp1_proxy);
3679 	kfree(dev->dev_vfs);
3680 
3681 	mlx4_clean_dev(dev);
3682 	priv->pci_dev_data = pci_dev_data;
3683 	priv->removed = 1;
3684 }
3685 
3686 static void mlx4_remove_one(struct pci_dev *pdev)
3687 {
3688 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3689 	struct mlx4_dev  *dev  = persist->dev;
3690 	struct mlx4_priv *priv = mlx4_priv(dev);
3691 	int active_vfs = 0;
3692 
3693 	mutex_lock(&persist->interface_state_mutex);
3694 	persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
3695 	mutex_unlock(&persist->interface_state_mutex);
3696 
3697 	/* Disabling SR-IOV is not allowed while there are active vf's */
3698 	if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
3699 		active_vfs = mlx4_how_many_lives_vf(dev);
3700 		if (active_vfs) {
3701 			pr_warn("Removing PF when there are active VF's !!\n");
3702 			pr_warn("Will not disable SR-IOV.\n");
3703 		}
3704 	}
3705 
3706 	/* device marked to be under deletion running now without the lock
3707 	 * letting other tasks to be terminated
3708 	 */
3709 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3710 		mlx4_unload_one(pdev);
3711 	else
3712 		mlx4_info(dev, "%s: interface is down\n", __func__);
3713 	mlx4_catas_end(dev);
3714 	if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
3715 		mlx4_warn(dev, "Disabling SR-IOV\n");
3716 		pci_disable_sriov(pdev);
3717 	}
3718 
3719 	pci_release_regions(pdev);
3720 	pci_disable_device(pdev);
3721 	kfree(dev->persist);
3722 	kfree(priv);
3723 	pci_set_drvdata(pdev, NULL);
3724 }
3725 
3726 static int restore_current_port_types(struct mlx4_dev *dev,
3727 				      enum mlx4_port_type *types,
3728 				      enum mlx4_port_type *poss_types)
3729 {
3730 	struct mlx4_priv *priv = mlx4_priv(dev);
3731 	int err, i;
3732 
3733 	mlx4_stop_sense(dev);
3734 
3735 	mutex_lock(&priv->port_mutex);
3736 	for (i = 0; i < dev->caps.num_ports; i++)
3737 		dev->caps.possible_type[i + 1] = poss_types[i];
3738 	err = mlx4_change_port_types(dev, types);
3739 	mlx4_start_sense(dev);
3740 	mutex_unlock(&priv->port_mutex);
3741 
3742 	return err;
3743 }
3744 
3745 int mlx4_restart_one(struct pci_dev *pdev)
3746 {
3747 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3748 	struct mlx4_dev	 *dev  = persist->dev;
3749 	struct mlx4_priv *priv = mlx4_priv(dev);
3750 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3751 	int pci_dev_data, err, total_vfs;
3752 
3753 	pci_dev_data = priv->pci_dev_data;
3754 	total_vfs = dev->persist->num_vfs;
3755 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
3756 
3757 	mlx4_unload_one(pdev);
3758 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
3759 	if (err) {
3760 		mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
3761 			 __func__, pci_name(pdev), err);
3762 		return err;
3763 	}
3764 
3765 	err = restore_current_port_types(dev, dev->persist->curr_port_type,
3766 					 dev->persist->curr_port_poss_type);
3767 	if (err)
3768 		mlx4_err(dev, "could not restore original port types (%d)\n",
3769 			 err);
3770 
3771 	return err;
3772 }
3773 
3774 static const struct pci_device_id mlx4_pci_table[] = {
3775 	/* MT25408 "Hermon" SDR */
3776 	{ PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3777 	/* MT25408 "Hermon" DDR */
3778 	{ PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3779 	/* MT25408 "Hermon" QDR */
3780 	{ PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3781 	/* MT25408 "Hermon" DDR PCIe gen2 */
3782 	{ PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3783 	/* MT25408 "Hermon" QDR PCIe gen2 */
3784 	{ PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3785 	/* MT25408 "Hermon" EN 10GigE */
3786 	{ PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3787 	/* MT25408 "Hermon" EN 10GigE PCIe gen2 */
3788 	{ PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3789 	/* MT25458 ConnectX EN 10GBASE-T 10GigE */
3790 	{ PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3791 	/* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
3792 	{ PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3793 	/* MT26468 ConnectX EN 10GigE PCIe gen2*/
3794 	{ PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3795 	/* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
3796 	{ PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3797 	/* MT26478 ConnectX2 40GigE PCIe gen2 */
3798 	{ PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3799 	/* MT25400 Family [ConnectX-2 Virtual Function] */
3800 	{ PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
3801 	/* MT27500 Family [ConnectX-3] */
3802 	{ PCI_VDEVICE(MELLANOX, 0x1003), 0 },
3803 	/* MT27500 Family [ConnectX-3 Virtual Function] */
3804 	{ PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
3805 	{ PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
3806 	{ PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
3807 	{ PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
3808 	{ PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
3809 	{ PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
3810 	{ PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
3811 	{ PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
3812 	{ PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
3813 	{ PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
3814 	{ PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
3815 	{ PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
3816 	{ PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
3817 	{ 0, }
3818 };
3819 
3820 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
3821 
3822 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
3823 					      pci_channel_state_t state)
3824 {
3825 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3826 
3827 	mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
3828 	mlx4_enter_error_state(persist);
3829 
3830 	mutex_lock(&persist->interface_state_mutex);
3831 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3832 		mlx4_unload_one(pdev);
3833 
3834 	mutex_unlock(&persist->interface_state_mutex);
3835 	if (state == pci_channel_io_perm_failure)
3836 		return PCI_ERS_RESULT_DISCONNECT;
3837 
3838 	pci_disable_device(pdev);
3839 	return PCI_ERS_RESULT_NEED_RESET;
3840 }
3841 
3842 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
3843 {
3844 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3845 	struct mlx4_dev	 *dev  = persist->dev;
3846 	struct mlx4_priv *priv = mlx4_priv(dev);
3847 	int               ret;
3848 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3849 	int total_vfs;
3850 
3851 	mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
3852 	ret = pci_enable_device(pdev);
3853 	if (ret) {
3854 		mlx4_err(dev, "Can not re-enable device, ret=%d\n", ret);
3855 		return PCI_ERS_RESULT_DISCONNECT;
3856 	}
3857 
3858 	pci_set_master(pdev);
3859 	pci_restore_state(pdev);
3860 	pci_save_state(pdev);
3861 
3862 	total_vfs = dev->persist->num_vfs;
3863 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
3864 
3865 	mutex_lock(&persist->interface_state_mutex);
3866 	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
3867 		ret = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
3868 				    priv, 1);
3869 		if (ret) {
3870 			mlx4_err(dev, "%s: mlx4_load_one failed, ret=%d\n",
3871 				 __func__,  ret);
3872 			goto end;
3873 		}
3874 
3875 		ret = restore_current_port_types(dev, dev->persist->
3876 						 curr_port_type, dev->persist->
3877 						 curr_port_poss_type);
3878 		if (ret)
3879 			mlx4_err(dev, "could not restore original port types (%d)\n", ret);
3880 	}
3881 end:
3882 	mutex_unlock(&persist->interface_state_mutex);
3883 
3884 	return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
3885 }
3886 
3887 static void mlx4_shutdown(struct pci_dev *pdev)
3888 {
3889 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3890 
3891 	mlx4_info(persist->dev, "mlx4_shutdown was called\n");
3892 	mutex_lock(&persist->interface_state_mutex);
3893 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3894 		mlx4_unload_one(pdev);
3895 	mutex_unlock(&persist->interface_state_mutex);
3896 }
3897 
3898 static const struct pci_error_handlers mlx4_err_handler = {
3899 	.error_detected = mlx4_pci_err_detected,
3900 	.slot_reset     = mlx4_pci_slot_reset,
3901 };
3902 
3903 static struct pci_driver mlx4_driver = {
3904 	.name		= DRV_NAME,
3905 	.id_table	= mlx4_pci_table,
3906 	.probe		= mlx4_init_one,
3907 	.shutdown	= mlx4_shutdown,
3908 	.remove		= mlx4_remove_one,
3909 	.err_handler    = &mlx4_err_handler,
3910 };
3911 
3912 static int __init mlx4_verify_params(void)
3913 {
3914 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
3915 		pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
3916 		return -1;
3917 	}
3918 
3919 	if (log_num_vlan != 0)
3920 		pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
3921 			MLX4_LOG_NUM_VLANS);
3922 
3923 	if (use_prio != 0)
3924 		pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
3925 
3926 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
3927 		pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
3928 			log_mtts_per_seg);
3929 		return -1;
3930 	}
3931 
3932 	/* Check if module param for ports type has legal combination */
3933 	if (port_type_array[0] == false && port_type_array[1] == true) {
3934 		pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
3935 		port_type_array[0] = true;
3936 	}
3937 
3938 	if (mlx4_log_num_mgm_entry_size < -7 ||
3939 	    (mlx4_log_num_mgm_entry_size > 0 &&
3940 	     (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
3941 	      mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
3942 		pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
3943 			mlx4_log_num_mgm_entry_size,
3944 			MLX4_MIN_MGM_LOG_ENTRY_SIZE,
3945 			MLX4_MAX_MGM_LOG_ENTRY_SIZE);
3946 		return -1;
3947 	}
3948 
3949 	return 0;
3950 }
3951 
3952 static int __init mlx4_init(void)
3953 {
3954 	int ret;
3955 
3956 	if (mlx4_verify_params())
3957 		return -EINVAL;
3958 
3959 
3960 	mlx4_wq = create_singlethread_workqueue("mlx4");
3961 	if (!mlx4_wq)
3962 		return -ENOMEM;
3963 
3964 	ret = pci_register_driver(&mlx4_driver);
3965 	if (ret < 0)
3966 		destroy_workqueue(mlx4_wq);
3967 	return ret < 0 ? ret : 0;
3968 }
3969 
3970 static void __exit mlx4_cleanup(void)
3971 {
3972 	pci_unregister_driver(&mlx4_driver);
3973 	destroy_workqueue(mlx4_wq);
3974 }
3975 
3976 module_init(mlx4_init);
3977 module_exit(mlx4_cleanup);
3978