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 #include <linux/etherdevice.h>
46 #include <net/devlink.h>
47 
48 #include <linux/mlx4/device.h>
49 #include <linux/mlx4/doorbell.h>
50 
51 #include "mlx4.h"
52 #include "fw.h"
53 #include "icm.h"
54 
55 MODULE_AUTHOR("Roland Dreier");
56 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
57 MODULE_LICENSE("Dual BSD/GPL");
58 MODULE_VERSION(DRV_VERSION);
59 
60 struct workqueue_struct *mlx4_wq;
61 
62 #ifdef CONFIG_MLX4_DEBUG
63 
64 int mlx4_debug_level = 0;
65 module_param_named(debug_level, mlx4_debug_level, int, 0644);
66 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
67 
68 #endif /* CONFIG_MLX4_DEBUG */
69 
70 #ifdef CONFIG_PCI_MSI
71 
72 static int msi_x = 1;
73 module_param(msi_x, int, 0444);
74 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
75 
76 #else /* CONFIG_PCI_MSI */
77 
78 #define msi_x (0)
79 
80 #endif /* CONFIG_PCI_MSI */
81 
82 static uint8_t num_vfs[3] = {0, 0, 0};
83 static int num_vfs_argc;
84 module_param_array(num_vfs, byte , &num_vfs_argc, 0444);
85 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
86 			  "num_vfs=port1,port2,port1+2");
87 
88 static uint8_t probe_vf[3] = {0, 0, 0};
89 static int probe_vfs_argc;
90 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
91 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
92 			   "probe_vf=port1,port2,port1+2");
93 
94 int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
95 module_param_named(log_num_mgm_entry_size,
96 			mlx4_log_num_mgm_entry_size, int, 0444);
97 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
98 					 " of qp per mcg, for example:"
99 					 " 10 gives 248.range: 7 <="
100 					 " log_num_mgm_entry_size <= 12."
101 					 " To activate device managed"
102 					 " flow steering when available, set to -1");
103 
104 static bool enable_64b_cqe_eqe = true;
105 module_param(enable_64b_cqe_eqe, bool, 0444);
106 MODULE_PARM_DESC(enable_64b_cqe_eqe,
107 		 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
108 
109 static bool enable_4k_uar;
110 module_param(enable_4k_uar, bool, 0444);
111 MODULE_PARM_DESC(enable_4k_uar,
112 		 "Enable using 4K UAR. Should not be enabled if have VFs which do not support 4K UARs (default: false)");
113 
114 #define PF_CONTEXT_BEHAVIOUR_MASK	(MLX4_FUNC_CAP_64B_EQE_CQE | \
115 					 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
116 					 MLX4_FUNC_CAP_DMFS_A0_STATIC)
117 
118 #define RESET_PERSIST_MASK_FLAGS	(MLX4_FLAG_SRIOV)
119 
120 static char mlx4_version[] =
121 	DRV_NAME ": Mellanox ConnectX core driver v"
122 	DRV_VERSION " (" DRV_RELDATE ")\n";
123 
124 static struct mlx4_profile default_profile = {
125 	.num_qp		= 1 << 18,
126 	.num_srq	= 1 << 16,
127 	.rdmarc_per_qp	= 1 << 4,
128 	.num_cq		= 1 << 16,
129 	.num_mcg	= 1 << 13,
130 	.num_mpt	= 1 << 19,
131 	.num_mtt	= 1 << 20, /* It is really num mtt segements */
132 };
133 
134 static struct mlx4_profile low_mem_profile = {
135 	.num_qp		= 1 << 17,
136 	.num_srq	= 1 << 6,
137 	.rdmarc_per_qp	= 1 << 4,
138 	.num_cq		= 1 << 8,
139 	.num_mcg	= 1 << 8,
140 	.num_mpt	= 1 << 9,
141 	.num_mtt	= 1 << 7,
142 };
143 
144 static int log_num_mac = 7;
145 module_param_named(log_num_mac, log_num_mac, int, 0444);
146 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
147 
148 static int log_num_vlan;
149 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
150 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
151 /* Log2 max number of VLANs per ETH port (0-7) */
152 #define MLX4_LOG_NUM_VLANS 7
153 #define MLX4_MIN_LOG_NUM_VLANS 0
154 #define MLX4_MIN_LOG_NUM_MAC 1
155 
156 static bool use_prio;
157 module_param_named(use_prio, use_prio, bool, 0444);
158 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
159 
160 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
161 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
162 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
163 
164 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
165 static int arr_argc = 2;
166 module_param_array(port_type_array, int, &arr_argc, 0444);
167 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
168 				"1 for IB, 2 for Ethernet");
169 
170 struct mlx4_port_config {
171 	struct list_head list;
172 	enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
173 	struct pci_dev *pdev;
174 };
175 
176 static atomic_t pf_loading = ATOMIC_INIT(0);
177 
178 static inline void mlx4_set_num_reserved_uars(struct mlx4_dev *dev,
179 					      struct mlx4_dev_cap *dev_cap)
180 {
181 	/* The reserved_uars is calculated by system page size unit.
182 	 * Therefore, adjustment is added when the uar page size is less
183 	 * than the system page size
184 	 */
185 	dev->caps.reserved_uars	=
186 		max_t(int,
187 		      mlx4_get_num_reserved_uar(dev),
188 		      dev_cap->reserved_uars /
189 			(1 << (PAGE_SHIFT - dev->uar_page_shift)));
190 }
191 
192 int mlx4_check_port_params(struct mlx4_dev *dev,
193 			   enum mlx4_port_type *port_type)
194 {
195 	int i;
196 
197 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
198 		for (i = 0; i < dev->caps.num_ports - 1; i++) {
199 			if (port_type[i] != port_type[i + 1]) {
200 				mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
201 				return -EINVAL;
202 			}
203 		}
204 	}
205 
206 	for (i = 0; i < dev->caps.num_ports; i++) {
207 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
208 			mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
209 				 i + 1);
210 			return -EINVAL;
211 		}
212 	}
213 	return 0;
214 }
215 
216 static void mlx4_set_port_mask(struct mlx4_dev *dev)
217 {
218 	int i;
219 
220 	for (i = 1; i <= dev->caps.num_ports; ++i)
221 		dev->caps.port_mask[i] = dev->caps.port_type[i];
222 }
223 
224 enum {
225 	MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
226 };
227 
228 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
229 {
230 	int err = 0;
231 	struct mlx4_func func;
232 
233 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
234 		err = mlx4_QUERY_FUNC(dev, &func, 0);
235 		if (err) {
236 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
237 			return err;
238 		}
239 		dev_cap->max_eqs = func.max_eq;
240 		dev_cap->reserved_eqs = func.rsvd_eqs;
241 		dev_cap->reserved_uars = func.rsvd_uars;
242 		err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
243 	}
244 	return err;
245 }
246 
247 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
248 {
249 	struct mlx4_caps *dev_cap = &dev->caps;
250 
251 	/* FW not supporting or cancelled by user */
252 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
253 	    !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
254 		return;
255 
256 	/* Must have 64B CQE_EQE enabled by FW to use bigger stride
257 	 * When FW has NCSI it may decide not to report 64B CQE/EQEs
258 	 */
259 	if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
260 	    !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
261 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
262 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
263 		return;
264 	}
265 
266 	if (cache_line_size() == 128 || cache_line_size() == 256) {
267 		mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
268 		/* Changing the real data inside CQE size to 32B */
269 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
270 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
271 
272 		if (mlx4_is_master(dev))
273 			dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
274 	} else {
275 		if (cache_line_size() != 32  && cache_line_size() != 64)
276 			mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
277 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
278 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
279 	}
280 }
281 
282 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
283 			  struct mlx4_port_cap *port_cap)
284 {
285 	dev->caps.vl_cap[port]	    = port_cap->max_vl;
286 	dev->caps.ib_mtu_cap[port]	    = port_cap->ib_mtu;
287 	dev->phys_caps.gid_phys_table_len[port]  = port_cap->max_gids;
288 	dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
289 	/* set gid and pkey table operating lengths by default
290 	 * to non-sriov values
291 	 */
292 	dev->caps.gid_table_len[port]  = port_cap->max_gids;
293 	dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
294 	dev->caps.port_width_cap[port] = port_cap->max_port_width;
295 	dev->caps.eth_mtu_cap[port]    = port_cap->eth_mtu;
296 	dev->caps.max_tc_eth	       = port_cap->max_tc_eth;
297 	dev->caps.def_mac[port]        = port_cap->def_mac;
298 	dev->caps.supported_type[port] = port_cap->supported_port_types;
299 	dev->caps.suggested_type[port] = port_cap->suggested_type;
300 	dev->caps.default_sense[port] = port_cap->default_sense;
301 	dev->caps.trans_type[port]	    = port_cap->trans_type;
302 	dev->caps.vendor_oui[port]     = port_cap->vendor_oui;
303 	dev->caps.wavelength[port]     = port_cap->wavelength;
304 	dev->caps.trans_code[port]     = port_cap->trans_code;
305 
306 	return 0;
307 }
308 
309 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
310 			 struct mlx4_port_cap *port_cap)
311 {
312 	int err = 0;
313 
314 	err = mlx4_QUERY_PORT(dev, port, port_cap);
315 
316 	if (err)
317 		mlx4_err(dev, "QUERY_PORT command failed.\n");
318 
319 	return err;
320 }
321 
322 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
323 {
324 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
325 		return;
326 
327 	if (mlx4_is_mfunc(dev)) {
328 		mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
329 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
330 		return;
331 	}
332 
333 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
334 		mlx4_dbg(dev,
335 			 "Keep FCS is not supported - Disabling Ignore FCS");
336 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
337 		return;
338 	}
339 }
340 
341 #define MLX4_A0_STEERING_TABLE_SIZE	256
342 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
343 {
344 	int err;
345 	int i;
346 
347 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
348 	if (err) {
349 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
350 		return err;
351 	}
352 	mlx4_dev_cap_dump(dev, dev_cap);
353 
354 	if (dev_cap->min_page_sz > PAGE_SIZE) {
355 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
356 			 dev_cap->min_page_sz, PAGE_SIZE);
357 		return -ENODEV;
358 	}
359 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
360 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
361 			 dev_cap->num_ports, MLX4_MAX_PORTS);
362 		return -ENODEV;
363 	}
364 
365 	if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
366 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
367 			 dev_cap->uar_size,
368 			 (unsigned long long)
369 			 pci_resource_len(dev->persist->pdev, 2));
370 		return -ENODEV;
371 	}
372 
373 	dev->caps.num_ports	     = dev_cap->num_ports;
374 	dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
375 	dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
376 				      dev->caps.num_sys_eqs :
377 				      MLX4_MAX_EQ_NUM;
378 	for (i = 1; i <= dev->caps.num_ports; ++i) {
379 		err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
380 		if (err) {
381 			mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
382 			return err;
383 		}
384 	}
385 
386 	dev->caps.uar_page_size	     = PAGE_SIZE;
387 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
388 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
389 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
390 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
391 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
392 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
393 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
394 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
395 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
396 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
397 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
398 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
399 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
400 	/*
401 	 * Subtract 1 from the limit because we need to allocate a
402 	 * spare CQE so the HCA HW can tell the difference between an
403 	 * empty CQ and a full CQ.
404 	 */
405 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
406 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
407 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
408 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
409 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
410 
411 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
412 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
413 					dev_cap->reserved_xrcds : 0;
414 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
415 					dev_cap->max_xrcds : 0;
416 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
417 
418 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
419 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
420 	dev->caps.flags		     = dev_cap->flags;
421 	dev->caps.flags2	     = dev_cap->flags2;
422 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
423 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
424 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
425 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
426 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
427 
428 	/* Save uar page shift */
429 	if (!mlx4_is_slave(dev)) {
430 		/* Virtual PCI function needs to determine UAR page size from
431 		 * firmware. Only master PCI function can set the uar page size
432 		 */
433 		if (enable_4k_uar)
434 			dev->uar_page_shift = DEFAULT_UAR_PAGE_SHIFT;
435 		else
436 			dev->uar_page_shift = PAGE_SHIFT;
437 
438 		mlx4_set_num_reserved_uars(dev, dev_cap);
439 	}
440 
441 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN) {
442 		struct mlx4_init_hca_param hca_param;
443 
444 		memset(&hca_param, 0, sizeof(hca_param));
445 		err = mlx4_QUERY_HCA(dev, &hca_param);
446 		/* Turn off PHV_EN flag in case phv_check_en is set.
447 		 * phv_check_en is a HW check that parse the packet and verify
448 		 * phv bit was reported correctly in the wqe. To allow QinQ
449 		 * PHV_EN flag should be set and phv_check_en must be cleared
450 		 * otherwise QinQ packets will be drop by the HW.
451 		 */
452 		if (err || hca_param.phv_check_en)
453 			dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_PHV_EN;
454 	}
455 
456 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
457 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
458 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
459 	/* Don't do sense port on multifunction devices (for now at least) */
460 	if (mlx4_is_mfunc(dev))
461 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
462 
463 	if (mlx4_low_memory_profile()) {
464 		dev->caps.log_num_macs  = MLX4_MIN_LOG_NUM_MAC;
465 		dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
466 	} else {
467 		dev->caps.log_num_macs  = log_num_mac;
468 		dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
469 	}
470 
471 	for (i = 1; i <= dev->caps.num_ports; ++i) {
472 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
473 		if (dev->caps.supported_type[i]) {
474 			/* if only ETH is supported - assign ETH */
475 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
476 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
477 			/* if only IB is supported, assign IB */
478 			else if (dev->caps.supported_type[i] ==
479 				 MLX4_PORT_TYPE_IB)
480 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
481 			else {
482 				/* if IB and ETH are supported, we set the port
483 				 * type according to user selection of port type;
484 				 * if user selected none, take the FW hint */
485 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
486 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
487 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
488 				else
489 					dev->caps.port_type[i] = port_type_array[i - 1];
490 			}
491 		}
492 		/*
493 		 * Link sensing is allowed on the port if 3 conditions are true:
494 		 * 1. Both protocols are supported on the port.
495 		 * 2. Different types are supported on the port
496 		 * 3. FW declared that it supports link sensing
497 		 */
498 		mlx4_priv(dev)->sense.sense_allowed[i] =
499 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
500 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
501 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
502 
503 		/*
504 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
505 		 * and perform sense_port FW command to try and set the correct
506 		 * port type from beginning
507 		 */
508 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
509 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
510 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
511 			mlx4_SENSE_PORT(dev, i, &sensed_port);
512 			if (sensed_port != MLX4_PORT_TYPE_NONE)
513 				dev->caps.port_type[i] = sensed_port;
514 		} else {
515 			dev->caps.possible_type[i] = dev->caps.port_type[i];
516 		}
517 
518 		if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
519 			dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
520 			mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
521 				  i, 1 << dev->caps.log_num_macs);
522 		}
523 		if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
524 			dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
525 			mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
526 				  i, 1 << dev->caps.log_num_vlans);
527 		}
528 	}
529 
530 	if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
531 	    (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
532 	    (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
533 		mlx4_warn(dev,
534 			  "Granular QoS per VF not supported with IB/Eth configuration\n");
535 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
536 	}
537 
538 	dev->caps.max_counters = dev_cap->max_counters;
539 
540 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
541 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
542 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
543 		(1 << dev->caps.log_num_macs) *
544 		(1 << dev->caps.log_num_vlans) *
545 		dev->caps.num_ports;
546 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
547 
548 	if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
549 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
550 		dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
551 	else
552 		dev->caps.dmfs_high_rate_qpn_base =
553 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
554 
555 	if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
556 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
557 		dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
558 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
559 		dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
560 	} else {
561 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
562 		dev->caps.dmfs_high_rate_qpn_base =
563 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
564 		dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
565 	}
566 
567 	dev->caps.rl_caps = dev_cap->rl_caps;
568 
569 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
570 		dev->caps.dmfs_high_rate_qpn_range;
571 
572 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
573 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
574 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
575 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
576 
577 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
578 
579 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
580 		if (dev_cap->flags &
581 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
582 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
583 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
584 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
585 		}
586 
587 		if (dev_cap->flags2 &
588 		    (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
589 		     MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
590 			mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
591 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
592 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
593 		}
594 	}
595 
596 	if ((dev->caps.flags &
597 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
598 	    mlx4_is_master(dev))
599 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
600 
601 	if (!mlx4_is_slave(dev)) {
602 		mlx4_enable_cqe_eqe_stride(dev);
603 		dev->caps.alloc_res_qp_mask =
604 			(dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
605 			MLX4_RESERVE_A0_QP;
606 
607 		if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
608 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
609 			mlx4_warn(dev, "Old device ETS support detected\n");
610 			mlx4_warn(dev, "Consider upgrading device FW.\n");
611 			dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
612 		}
613 
614 	} else {
615 		dev->caps.alloc_res_qp_mask = 0;
616 	}
617 
618 	mlx4_enable_ignore_fcs(dev);
619 
620 	return 0;
621 }
622 
623 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
624 				       enum pci_bus_speed *speed,
625 				       enum pcie_link_width *width)
626 {
627 	u32 lnkcap1, lnkcap2;
628 	int err1, err2;
629 
630 #define  PCIE_MLW_CAP_SHIFT 4	/* start of MLW mask in link capabilities */
631 
632 	*speed = PCI_SPEED_UNKNOWN;
633 	*width = PCIE_LNK_WIDTH_UNKNOWN;
634 
635 	err1 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP,
636 					  &lnkcap1);
637 	err2 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP2,
638 					  &lnkcap2);
639 	if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
640 		if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
641 			*speed = PCIE_SPEED_8_0GT;
642 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
643 			*speed = PCIE_SPEED_5_0GT;
644 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
645 			*speed = PCIE_SPEED_2_5GT;
646 	}
647 	if (!err1) {
648 		*width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
649 		if (!lnkcap2) { /* pre-r3.0 */
650 			if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
651 				*speed = PCIE_SPEED_5_0GT;
652 			else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
653 				*speed = PCIE_SPEED_2_5GT;
654 		}
655 	}
656 
657 	if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
658 		return err1 ? err1 :
659 			err2 ? err2 : -EINVAL;
660 	}
661 	return 0;
662 }
663 
664 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
665 {
666 	enum pcie_link_width width, width_cap;
667 	enum pci_bus_speed speed, speed_cap;
668 	int err;
669 
670 #define PCIE_SPEED_STR(speed) \
671 	(speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
672 	 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
673 	 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
674 	 "Unknown")
675 
676 	err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
677 	if (err) {
678 		mlx4_warn(dev,
679 			  "Unable to determine PCIe device BW capabilities\n");
680 		return;
681 	}
682 
683 	err = pcie_get_minimum_link(dev->persist->pdev, &speed, &width);
684 	if (err || speed == PCI_SPEED_UNKNOWN ||
685 	    width == PCIE_LNK_WIDTH_UNKNOWN) {
686 		mlx4_warn(dev,
687 			  "Unable to determine PCI device chain minimum BW\n");
688 		return;
689 	}
690 
691 	if (width != width_cap || speed != speed_cap)
692 		mlx4_warn(dev,
693 			  "PCIe BW is different than device's capability\n");
694 
695 	mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
696 		  PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
697 	mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
698 		  width, width_cap);
699 	return;
700 }
701 
702 /*The function checks if there are live vf, return the num of them*/
703 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
704 {
705 	struct mlx4_priv *priv = mlx4_priv(dev);
706 	struct mlx4_slave_state *s_state;
707 	int i;
708 	int ret = 0;
709 
710 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
711 		s_state = &priv->mfunc.master.slave_state[i];
712 		if (s_state->active && s_state->last_cmd !=
713 		    MLX4_COMM_CMD_RESET) {
714 			mlx4_warn(dev, "%s: slave: %d is still active\n",
715 				  __func__, i);
716 			ret++;
717 		}
718 	}
719 	return ret;
720 }
721 
722 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
723 {
724 	u32 qk = MLX4_RESERVED_QKEY_BASE;
725 
726 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
727 	    qpn < dev->phys_caps.base_proxy_sqpn)
728 		return -EINVAL;
729 
730 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
731 		/* tunnel qp */
732 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
733 	else
734 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
735 	*qkey = qk;
736 	return 0;
737 }
738 EXPORT_SYMBOL(mlx4_get_parav_qkey);
739 
740 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
741 {
742 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
743 
744 	if (!mlx4_is_master(dev))
745 		return;
746 
747 	priv->virt2phys_pkey[slave][port - 1][i] = val;
748 }
749 EXPORT_SYMBOL(mlx4_sync_pkey_table);
750 
751 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
752 {
753 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
754 
755 	if (!mlx4_is_master(dev))
756 		return;
757 
758 	priv->slave_node_guids[slave] = guid;
759 }
760 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
761 
762 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
763 {
764 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
765 
766 	if (!mlx4_is_master(dev))
767 		return 0;
768 
769 	return priv->slave_node_guids[slave];
770 }
771 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
772 
773 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
774 {
775 	struct mlx4_priv *priv = mlx4_priv(dev);
776 	struct mlx4_slave_state *s_slave;
777 
778 	if (!mlx4_is_master(dev))
779 		return 0;
780 
781 	s_slave = &priv->mfunc.master.slave_state[slave];
782 	return !!s_slave->active;
783 }
784 EXPORT_SYMBOL(mlx4_is_slave_active);
785 
786 void mlx4_handle_eth_header_mcast_prio(struct mlx4_net_trans_rule_hw_ctrl *ctrl,
787 				       struct _rule_hw *eth_header)
788 {
789 	if (is_multicast_ether_addr(eth_header->eth.dst_mac) ||
790 	    is_broadcast_ether_addr(eth_header->eth.dst_mac)) {
791 		struct mlx4_net_trans_rule_hw_eth *eth =
792 			(struct mlx4_net_trans_rule_hw_eth *)eth_header;
793 		struct _rule_hw *next_rule = (struct _rule_hw *)(eth + 1);
794 		bool last_rule = next_rule->size == 0 && next_rule->id == 0 &&
795 			next_rule->rsvd == 0;
796 
797 		if (last_rule)
798 			ctrl->prio = cpu_to_be16(MLX4_DOMAIN_NIC);
799 	}
800 }
801 EXPORT_SYMBOL(mlx4_handle_eth_header_mcast_prio);
802 
803 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
804 				       struct mlx4_dev_cap *dev_cap,
805 				       struct mlx4_init_hca_param *hca_param)
806 {
807 	dev->caps.steering_mode = hca_param->steering_mode;
808 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
809 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
810 		dev->caps.fs_log_max_ucast_qp_range_size =
811 			dev_cap->fs_log_max_ucast_qp_range_size;
812 	} else
813 		dev->caps.num_qp_per_mgm =
814 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
815 
816 	mlx4_dbg(dev, "Steering mode is: %s\n",
817 		 mlx4_steering_mode_str(dev->caps.steering_mode));
818 }
819 
820 static int mlx4_slave_cap(struct mlx4_dev *dev)
821 {
822 	int			   err;
823 	u32			   page_size;
824 	struct mlx4_dev_cap	   dev_cap;
825 	struct mlx4_func_cap	   func_cap;
826 	struct mlx4_init_hca_param hca_param;
827 	u8			   i;
828 
829 	memset(&hca_param, 0, sizeof(hca_param));
830 	err = mlx4_QUERY_HCA(dev, &hca_param);
831 	if (err) {
832 		mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
833 		return err;
834 	}
835 
836 	/* fail if the hca has an unknown global capability
837 	 * at this time global_caps should be always zeroed
838 	 */
839 	if (hca_param.global_caps) {
840 		mlx4_err(dev, "Unknown hca global capabilities\n");
841 		return -EINVAL;
842 	}
843 
844 	dev->caps.hca_core_clock = hca_param.hca_core_clock;
845 
846 	memset(&dev_cap, 0, sizeof(dev_cap));
847 	dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
848 	err = mlx4_dev_cap(dev, &dev_cap);
849 	if (err) {
850 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
851 		return err;
852 	}
853 
854 	err = mlx4_QUERY_FW(dev);
855 	if (err)
856 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
857 
858 	page_size = ~dev->caps.page_size_cap + 1;
859 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
860 	if (page_size > PAGE_SIZE) {
861 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
862 			 page_size, PAGE_SIZE);
863 		return -ENODEV;
864 	}
865 
866 	/* Set uar_page_shift for VF */
867 	dev->uar_page_shift = hca_param.uar_page_sz + 12;
868 
869 	/* Make sure the master uar page size is valid */
870 	if (dev->uar_page_shift > PAGE_SHIFT) {
871 		mlx4_err(dev,
872 			 "Invalid configuration: uar page size is larger than system page size\n");
873 		return  -ENODEV;
874 	}
875 
876 	/* Set reserved_uars based on the uar_page_shift */
877 	mlx4_set_num_reserved_uars(dev, &dev_cap);
878 
879 	/* Although uar page size in FW differs from system page size,
880 	 * upper software layers (mlx4_ib, mlx4_en and part of mlx4_core)
881 	 * still works with assumption that uar page size == system page size
882 	 */
883 	dev->caps.uar_page_size = PAGE_SIZE;
884 
885 	memset(&func_cap, 0, sizeof(func_cap));
886 	err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
887 	if (err) {
888 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
889 			 err);
890 		return err;
891 	}
892 
893 	if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
894 	    PF_CONTEXT_BEHAVIOUR_MASK) {
895 		mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
896 			 func_cap.pf_context_behaviour, PF_CONTEXT_BEHAVIOUR_MASK);
897 		return -EINVAL;
898 	}
899 
900 	dev->caps.num_ports		= func_cap.num_ports;
901 	dev->quotas.qp			= func_cap.qp_quota;
902 	dev->quotas.srq			= func_cap.srq_quota;
903 	dev->quotas.cq			= func_cap.cq_quota;
904 	dev->quotas.mpt			= func_cap.mpt_quota;
905 	dev->quotas.mtt			= func_cap.mtt_quota;
906 	dev->caps.num_qps		= 1 << hca_param.log_num_qps;
907 	dev->caps.num_srqs		= 1 << hca_param.log_num_srqs;
908 	dev->caps.num_cqs		= 1 << hca_param.log_num_cqs;
909 	dev->caps.num_mpts		= 1 << hca_param.log_mpt_sz;
910 	dev->caps.num_eqs		= func_cap.max_eq;
911 	dev->caps.reserved_eqs		= func_cap.reserved_eq;
912 	dev->caps.reserved_lkey		= func_cap.reserved_lkey;
913 	dev->caps.num_pds               = MLX4_NUM_PDS;
914 	dev->caps.num_mgms              = 0;
915 	dev->caps.num_amgms             = 0;
916 
917 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
918 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
919 			 dev->caps.num_ports, MLX4_MAX_PORTS);
920 		return -ENODEV;
921 	}
922 
923 	mlx4_replace_zero_macs(dev);
924 
925 	dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
926 	dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
927 	dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
928 	dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
929 	dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
930 
931 	if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
932 	    !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
933 	    !dev->caps.qp0_qkey) {
934 		err = -ENOMEM;
935 		goto err_mem;
936 	}
937 
938 	for (i = 1; i <= dev->caps.num_ports; ++i) {
939 		err = mlx4_QUERY_FUNC_CAP(dev, i, &func_cap);
940 		if (err) {
941 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
942 				 i, err);
943 			goto err_mem;
944 		}
945 		dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
946 		dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
947 		dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
948 		dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
949 		dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
950 		dev->caps.port_mask[i] = dev->caps.port_type[i];
951 		dev->caps.phys_port_id[i] = func_cap.phys_port_id;
952 		err = mlx4_get_slave_pkey_gid_tbl_len(dev, i,
953 						      &dev->caps.gid_table_len[i],
954 						      &dev->caps.pkey_table_len[i]);
955 		if (err)
956 			goto err_mem;
957 	}
958 
959 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
960 				       dev->caps.reserved_uars) >
961 				       pci_resource_len(dev->persist->pdev,
962 							2)) {
963 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
964 			 dev->caps.uar_page_size * dev->caps.num_uars,
965 			 (unsigned long long)
966 			 pci_resource_len(dev->persist->pdev, 2));
967 		err = -ENOMEM;
968 		goto err_mem;
969 	}
970 
971 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
972 		dev->caps.eqe_size   = 64;
973 		dev->caps.eqe_factor = 1;
974 	} else {
975 		dev->caps.eqe_size   = 32;
976 		dev->caps.eqe_factor = 0;
977 	}
978 
979 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
980 		dev->caps.cqe_size   = 64;
981 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
982 	} else {
983 		dev->caps.cqe_size   = 32;
984 	}
985 
986 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
987 		dev->caps.eqe_size = hca_param.eqe_size;
988 		dev->caps.eqe_factor = 0;
989 	}
990 
991 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
992 		dev->caps.cqe_size = hca_param.cqe_size;
993 		/* User still need to know when CQE > 32B */
994 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
995 	}
996 
997 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
998 	mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
999 
1000 	slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
1001 	mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
1002 		 hca_param.rss_ip_frags ? "on" : "off");
1003 
1004 	if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
1005 	    dev->caps.bf_reg_size)
1006 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
1007 
1008 	if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
1009 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
1010 
1011 	return 0;
1012 
1013 err_mem:
1014 	kfree(dev->caps.qp0_qkey);
1015 	kfree(dev->caps.qp0_tunnel);
1016 	kfree(dev->caps.qp0_proxy);
1017 	kfree(dev->caps.qp1_tunnel);
1018 	kfree(dev->caps.qp1_proxy);
1019 	dev->caps.qp0_qkey = NULL;
1020 	dev->caps.qp0_tunnel = NULL;
1021 	dev->caps.qp0_proxy = NULL;
1022 	dev->caps.qp1_tunnel = NULL;
1023 	dev->caps.qp1_proxy = NULL;
1024 
1025 	return err;
1026 }
1027 
1028 static void mlx4_request_modules(struct mlx4_dev *dev)
1029 {
1030 	int port;
1031 	int has_ib_port = false;
1032 	int has_eth_port = false;
1033 #define EN_DRV_NAME	"mlx4_en"
1034 #define IB_DRV_NAME	"mlx4_ib"
1035 
1036 	for (port = 1; port <= dev->caps.num_ports; port++) {
1037 		if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
1038 			has_ib_port = true;
1039 		else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
1040 			has_eth_port = true;
1041 	}
1042 
1043 	if (has_eth_port)
1044 		request_module_nowait(EN_DRV_NAME);
1045 	if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
1046 		request_module_nowait(IB_DRV_NAME);
1047 }
1048 
1049 /*
1050  * Change the port configuration of the device.
1051  * Every user of this function must hold the port mutex.
1052  */
1053 int mlx4_change_port_types(struct mlx4_dev *dev,
1054 			   enum mlx4_port_type *port_types)
1055 {
1056 	int err = 0;
1057 	int change = 0;
1058 	int port;
1059 
1060 	for (port = 0; port <  dev->caps.num_ports; port++) {
1061 		/* Change the port type only if the new type is different
1062 		 * from the current, and not set to Auto */
1063 		if (port_types[port] != dev->caps.port_type[port + 1])
1064 			change = 1;
1065 	}
1066 	if (change) {
1067 		mlx4_unregister_device(dev);
1068 		for (port = 1; port <= dev->caps.num_ports; port++) {
1069 			mlx4_CLOSE_PORT(dev, port);
1070 			dev->caps.port_type[port] = port_types[port - 1];
1071 			err = mlx4_SET_PORT(dev, port, -1);
1072 			if (err) {
1073 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1074 					 port);
1075 				goto out;
1076 			}
1077 		}
1078 		mlx4_set_port_mask(dev);
1079 		err = mlx4_register_device(dev);
1080 		if (err) {
1081 			mlx4_err(dev, "Failed to register device\n");
1082 			goto out;
1083 		}
1084 		mlx4_request_modules(dev);
1085 	}
1086 
1087 out:
1088 	return err;
1089 }
1090 
1091 static ssize_t show_port_type(struct device *dev,
1092 			      struct device_attribute *attr,
1093 			      char *buf)
1094 {
1095 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1096 						   port_attr);
1097 	struct mlx4_dev *mdev = info->dev;
1098 	char type[8];
1099 
1100 	sprintf(type, "%s",
1101 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1102 		"ib" : "eth");
1103 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1104 		sprintf(buf, "auto (%s)\n", type);
1105 	else
1106 		sprintf(buf, "%s\n", type);
1107 
1108 	return strlen(buf);
1109 }
1110 
1111 static int __set_port_type(struct mlx4_port_info *info,
1112 			   enum mlx4_port_type port_type)
1113 {
1114 	struct mlx4_dev *mdev = info->dev;
1115 	struct mlx4_priv *priv = mlx4_priv(mdev);
1116 	enum mlx4_port_type types[MLX4_MAX_PORTS];
1117 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1118 	int i;
1119 	int err = 0;
1120 
1121 	if ((port_type & mdev->caps.supported_type[info->port]) != port_type) {
1122 		mlx4_err(mdev,
1123 			 "Requested port type for port %d is not supported on this HCA\n",
1124 			 info->port);
1125 		err = -EINVAL;
1126 		goto err_sup;
1127 	}
1128 
1129 	mlx4_stop_sense(mdev);
1130 	mutex_lock(&priv->port_mutex);
1131 	info->tmp_type = port_type;
1132 
1133 	/* Possible type is always the one that was delivered */
1134 	mdev->caps.possible_type[info->port] = info->tmp_type;
1135 
1136 	for (i = 0; i < mdev->caps.num_ports; i++) {
1137 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1138 					mdev->caps.possible_type[i+1];
1139 		if (types[i] == MLX4_PORT_TYPE_AUTO)
1140 			types[i] = mdev->caps.port_type[i+1];
1141 	}
1142 
1143 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1144 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1145 		for (i = 1; i <= mdev->caps.num_ports; i++) {
1146 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1147 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1148 				err = -EINVAL;
1149 			}
1150 		}
1151 	}
1152 	if (err) {
1153 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1154 		goto out;
1155 	}
1156 
1157 	mlx4_do_sense_ports(mdev, new_types, types);
1158 
1159 	err = mlx4_check_port_params(mdev, new_types);
1160 	if (err)
1161 		goto out;
1162 
1163 	/* We are about to apply the changes after the configuration
1164 	 * was verified, no need to remember the temporary types
1165 	 * any more */
1166 	for (i = 0; i < mdev->caps.num_ports; i++)
1167 		priv->port[i + 1].tmp_type = 0;
1168 
1169 	err = mlx4_change_port_types(mdev, new_types);
1170 
1171 out:
1172 	mlx4_start_sense(mdev);
1173 	mutex_unlock(&priv->port_mutex);
1174 err_sup:
1175 	return err;
1176 }
1177 
1178 static ssize_t set_port_type(struct device *dev,
1179 			     struct device_attribute *attr,
1180 			     const char *buf, size_t count)
1181 {
1182 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1183 						   port_attr);
1184 	struct mlx4_dev *mdev = info->dev;
1185 	enum mlx4_port_type port_type;
1186 	static DEFINE_MUTEX(set_port_type_mutex);
1187 	int err;
1188 
1189 	mutex_lock(&set_port_type_mutex);
1190 
1191 	if (!strcmp(buf, "ib\n")) {
1192 		port_type = MLX4_PORT_TYPE_IB;
1193 	} else if (!strcmp(buf, "eth\n")) {
1194 		port_type = MLX4_PORT_TYPE_ETH;
1195 	} else if (!strcmp(buf, "auto\n")) {
1196 		port_type = MLX4_PORT_TYPE_AUTO;
1197 	} else {
1198 		mlx4_err(mdev, "%s is not supported port type\n", buf);
1199 		err = -EINVAL;
1200 		goto err_out;
1201 	}
1202 
1203 	err = __set_port_type(info, port_type);
1204 
1205 err_out:
1206 	mutex_unlock(&set_port_type_mutex);
1207 
1208 	return err ? err : count;
1209 }
1210 
1211 enum ibta_mtu {
1212 	IB_MTU_256  = 1,
1213 	IB_MTU_512  = 2,
1214 	IB_MTU_1024 = 3,
1215 	IB_MTU_2048 = 4,
1216 	IB_MTU_4096 = 5
1217 };
1218 
1219 static inline int int_to_ibta_mtu(int mtu)
1220 {
1221 	switch (mtu) {
1222 	case 256:  return IB_MTU_256;
1223 	case 512:  return IB_MTU_512;
1224 	case 1024: return IB_MTU_1024;
1225 	case 2048: return IB_MTU_2048;
1226 	case 4096: return IB_MTU_4096;
1227 	default: return -1;
1228 	}
1229 }
1230 
1231 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1232 {
1233 	switch (mtu) {
1234 	case IB_MTU_256:  return  256;
1235 	case IB_MTU_512:  return  512;
1236 	case IB_MTU_1024: return 1024;
1237 	case IB_MTU_2048: return 2048;
1238 	case IB_MTU_4096: return 4096;
1239 	default: return -1;
1240 	}
1241 }
1242 
1243 static ssize_t show_port_ib_mtu(struct device *dev,
1244 			     struct device_attribute *attr,
1245 			     char *buf)
1246 {
1247 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1248 						   port_mtu_attr);
1249 	struct mlx4_dev *mdev = info->dev;
1250 
1251 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1252 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1253 
1254 	sprintf(buf, "%d\n",
1255 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1256 	return strlen(buf);
1257 }
1258 
1259 static ssize_t set_port_ib_mtu(struct device *dev,
1260 			     struct device_attribute *attr,
1261 			     const char *buf, size_t count)
1262 {
1263 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1264 						   port_mtu_attr);
1265 	struct mlx4_dev *mdev = info->dev;
1266 	struct mlx4_priv *priv = mlx4_priv(mdev);
1267 	int err, port, mtu, ibta_mtu = -1;
1268 
1269 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1270 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1271 		return -EINVAL;
1272 	}
1273 
1274 	err = kstrtoint(buf, 0, &mtu);
1275 	if (!err)
1276 		ibta_mtu = int_to_ibta_mtu(mtu);
1277 
1278 	if (err || ibta_mtu < 0) {
1279 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1280 		return -EINVAL;
1281 	}
1282 
1283 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1284 
1285 	mlx4_stop_sense(mdev);
1286 	mutex_lock(&priv->port_mutex);
1287 	mlx4_unregister_device(mdev);
1288 	for (port = 1; port <= mdev->caps.num_ports; port++) {
1289 		mlx4_CLOSE_PORT(mdev, port);
1290 		err = mlx4_SET_PORT(mdev, port, -1);
1291 		if (err) {
1292 			mlx4_err(mdev, "Failed to set port %d, aborting\n",
1293 				 port);
1294 			goto err_set_port;
1295 		}
1296 	}
1297 	err = mlx4_register_device(mdev);
1298 err_set_port:
1299 	mutex_unlock(&priv->port_mutex);
1300 	mlx4_start_sense(mdev);
1301 	return err ? err : count;
1302 }
1303 
1304 /* bond for multi-function device */
1305 #define MAX_MF_BOND_ALLOWED_SLAVES 63
1306 static int mlx4_mf_bond(struct mlx4_dev *dev)
1307 {
1308 	int err = 0;
1309 	int nvfs;
1310 	struct mlx4_slaves_pport slaves_port1;
1311 	struct mlx4_slaves_pport slaves_port2;
1312 	DECLARE_BITMAP(slaves_port_1_2, MLX4_MFUNC_MAX);
1313 
1314 	slaves_port1 = mlx4_phys_to_slaves_pport(dev, 1);
1315 	slaves_port2 = mlx4_phys_to_slaves_pport(dev, 2);
1316 	bitmap_and(slaves_port_1_2,
1317 		   slaves_port1.slaves, slaves_port2.slaves,
1318 		   dev->persist->num_vfs + 1);
1319 
1320 	/* only single port vfs are allowed */
1321 	if (bitmap_weight(slaves_port_1_2, dev->persist->num_vfs + 1) > 1) {
1322 		mlx4_warn(dev, "HA mode unsupported for dual ported VFs\n");
1323 		return -EINVAL;
1324 	}
1325 
1326 	/* number of virtual functions is number of total functions minus one
1327 	 * physical function for each port.
1328 	 */
1329 	nvfs = bitmap_weight(slaves_port1.slaves, dev->persist->num_vfs + 1) +
1330 		bitmap_weight(slaves_port2.slaves, dev->persist->num_vfs + 1) - 2;
1331 
1332 	/* limit on maximum allowed VFs */
1333 	if (nvfs > MAX_MF_BOND_ALLOWED_SLAVES) {
1334 		mlx4_warn(dev, "HA mode is not supported for %d VFs (max %d are allowed)\n",
1335 			  nvfs, MAX_MF_BOND_ALLOWED_SLAVES);
1336 		return -EINVAL;
1337 	}
1338 
1339 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) {
1340 		mlx4_warn(dev, "HA mode unsupported for NON DMFS steering\n");
1341 		return -EINVAL;
1342 	}
1343 
1344 	err = mlx4_bond_mac_table(dev);
1345 	if (err)
1346 		return err;
1347 	err = mlx4_bond_vlan_table(dev);
1348 	if (err)
1349 		goto err1;
1350 	err = mlx4_bond_fs_rules(dev);
1351 	if (err)
1352 		goto err2;
1353 
1354 	return 0;
1355 err2:
1356 	(void)mlx4_unbond_vlan_table(dev);
1357 err1:
1358 	(void)mlx4_unbond_mac_table(dev);
1359 	return err;
1360 }
1361 
1362 static int mlx4_mf_unbond(struct mlx4_dev *dev)
1363 {
1364 	int ret, ret1;
1365 
1366 	ret = mlx4_unbond_fs_rules(dev);
1367 	if (ret)
1368 		mlx4_warn(dev, "multifunction unbond for flow rules failedi (%d)\n", ret);
1369 	ret1 = mlx4_unbond_mac_table(dev);
1370 	if (ret1) {
1371 		mlx4_warn(dev, "multifunction unbond for MAC table failed (%d)\n", ret1);
1372 		ret = ret1;
1373 	}
1374 	ret1 = mlx4_unbond_vlan_table(dev);
1375 	if (ret1) {
1376 		mlx4_warn(dev, "multifunction unbond for VLAN table failed (%d)\n", ret1);
1377 		ret = ret1;
1378 	}
1379 	return ret;
1380 }
1381 
1382 int mlx4_bond(struct mlx4_dev *dev)
1383 {
1384 	int ret = 0;
1385 	struct mlx4_priv *priv = mlx4_priv(dev);
1386 
1387 	mutex_lock(&priv->bond_mutex);
1388 
1389 	if (!mlx4_is_bonded(dev)) {
1390 		ret = mlx4_do_bond(dev, true);
1391 		if (ret)
1392 			mlx4_err(dev, "Failed to bond device: %d\n", ret);
1393 		if (!ret && mlx4_is_master(dev)) {
1394 			ret = mlx4_mf_bond(dev);
1395 			if (ret) {
1396 				mlx4_err(dev, "bond for multifunction failed\n");
1397 				mlx4_do_bond(dev, false);
1398 			}
1399 		}
1400 	}
1401 
1402 	mutex_unlock(&priv->bond_mutex);
1403 	if (!ret)
1404 		mlx4_dbg(dev, "Device is bonded\n");
1405 
1406 	return ret;
1407 }
1408 EXPORT_SYMBOL_GPL(mlx4_bond);
1409 
1410 int mlx4_unbond(struct mlx4_dev *dev)
1411 {
1412 	int ret = 0;
1413 	struct mlx4_priv *priv = mlx4_priv(dev);
1414 
1415 	mutex_lock(&priv->bond_mutex);
1416 
1417 	if (mlx4_is_bonded(dev)) {
1418 		int ret2 = 0;
1419 
1420 		ret = mlx4_do_bond(dev, false);
1421 		if (ret)
1422 			mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1423 		if (mlx4_is_master(dev))
1424 			ret2 = mlx4_mf_unbond(dev);
1425 		if (ret2) {
1426 			mlx4_warn(dev, "Failed to unbond device for multifunction (%d)\n", ret2);
1427 			ret = ret2;
1428 		}
1429 	}
1430 
1431 	mutex_unlock(&priv->bond_mutex);
1432 	if (!ret)
1433 		mlx4_dbg(dev, "Device is unbonded\n");
1434 
1435 	return ret;
1436 }
1437 EXPORT_SYMBOL_GPL(mlx4_unbond);
1438 
1439 
1440 int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1441 {
1442 	u8 port1 = v2p->port1;
1443 	u8 port2 = v2p->port2;
1444 	struct mlx4_priv *priv = mlx4_priv(dev);
1445 	int err;
1446 
1447 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1448 		return -EOPNOTSUPP;
1449 
1450 	mutex_lock(&priv->bond_mutex);
1451 
1452 	/* zero means keep current mapping for this port */
1453 	if (port1 == 0)
1454 		port1 = priv->v2p.port1;
1455 	if (port2 == 0)
1456 		port2 = priv->v2p.port2;
1457 
1458 	if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1459 	    (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1460 	    (port1 == 2 && port2 == 1)) {
1461 		/* besides boundary checks cross mapping makes
1462 		 * no sense and therefore not allowed */
1463 		err = -EINVAL;
1464 	} else if ((port1 == priv->v2p.port1) &&
1465 		 (port2 == priv->v2p.port2)) {
1466 		err = 0;
1467 	} else {
1468 		err = mlx4_virt2phy_port_map(dev, port1, port2);
1469 		if (!err) {
1470 			mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1471 				 port1, port2);
1472 			priv->v2p.port1 = port1;
1473 			priv->v2p.port2 = port2;
1474 		} else {
1475 			mlx4_err(dev, "Failed to change port mape: %d\n", err);
1476 		}
1477 	}
1478 
1479 	mutex_unlock(&priv->bond_mutex);
1480 	return err;
1481 }
1482 EXPORT_SYMBOL_GPL(mlx4_port_map_set);
1483 
1484 static int mlx4_load_fw(struct mlx4_dev *dev)
1485 {
1486 	struct mlx4_priv *priv = mlx4_priv(dev);
1487 	int err;
1488 
1489 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1490 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
1491 	if (!priv->fw.fw_icm) {
1492 		mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1493 		return -ENOMEM;
1494 	}
1495 
1496 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1497 	if (err) {
1498 		mlx4_err(dev, "MAP_FA command failed, aborting\n");
1499 		goto err_free;
1500 	}
1501 
1502 	err = mlx4_RUN_FW(dev);
1503 	if (err) {
1504 		mlx4_err(dev, "RUN_FW command failed, aborting\n");
1505 		goto err_unmap_fa;
1506 	}
1507 
1508 	return 0;
1509 
1510 err_unmap_fa:
1511 	mlx4_UNMAP_FA(dev);
1512 
1513 err_free:
1514 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1515 	return err;
1516 }
1517 
1518 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1519 				int cmpt_entry_sz)
1520 {
1521 	struct mlx4_priv *priv = mlx4_priv(dev);
1522 	int err;
1523 	int num_eqs;
1524 
1525 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1526 				  cmpt_base +
1527 				  ((u64) (MLX4_CMPT_TYPE_QP *
1528 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1529 				  cmpt_entry_sz, dev->caps.num_qps,
1530 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1531 				  0, 0);
1532 	if (err)
1533 		goto err;
1534 
1535 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1536 				  cmpt_base +
1537 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1538 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1539 				  cmpt_entry_sz, dev->caps.num_srqs,
1540 				  dev->caps.reserved_srqs, 0, 0);
1541 	if (err)
1542 		goto err_qp;
1543 
1544 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1545 				  cmpt_base +
1546 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1547 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1548 				  cmpt_entry_sz, dev->caps.num_cqs,
1549 				  dev->caps.reserved_cqs, 0, 0);
1550 	if (err)
1551 		goto err_srq;
1552 
1553 	num_eqs = dev->phys_caps.num_phys_eqs;
1554 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1555 				  cmpt_base +
1556 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1557 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1558 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1559 	if (err)
1560 		goto err_cq;
1561 
1562 	return 0;
1563 
1564 err_cq:
1565 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1566 
1567 err_srq:
1568 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1569 
1570 err_qp:
1571 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1572 
1573 err:
1574 	return err;
1575 }
1576 
1577 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1578 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1579 {
1580 	struct mlx4_priv *priv = mlx4_priv(dev);
1581 	u64 aux_pages;
1582 	int num_eqs;
1583 	int err;
1584 
1585 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1586 	if (err) {
1587 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1588 		return err;
1589 	}
1590 
1591 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1592 		 (unsigned long long) icm_size >> 10,
1593 		 (unsigned long long) aux_pages << 2);
1594 
1595 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1596 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1597 	if (!priv->fw.aux_icm) {
1598 		mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1599 		return -ENOMEM;
1600 	}
1601 
1602 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1603 	if (err) {
1604 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1605 		goto err_free_aux;
1606 	}
1607 
1608 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1609 	if (err) {
1610 		mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1611 		goto err_unmap_aux;
1612 	}
1613 
1614 
1615 	num_eqs = dev->phys_caps.num_phys_eqs;
1616 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1617 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1618 				  num_eqs, num_eqs, 0, 0);
1619 	if (err) {
1620 		mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1621 		goto err_unmap_cmpt;
1622 	}
1623 
1624 	/*
1625 	 * Reserved MTT entries must be aligned up to a cacheline
1626 	 * boundary, since the FW will write to them, while the driver
1627 	 * writes to all other MTT entries. (The variable
1628 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1629 	 * size, not the raw entry size)
1630 	 */
1631 	dev->caps.reserved_mtts =
1632 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1633 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1634 
1635 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1636 				  init_hca->mtt_base,
1637 				  dev->caps.mtt_entry_sz,
1638 				  dev->caps.num_mtts,
1639 				  dev->caps.reserved_mtts, 1, 0);
1640 	if (err) {
1641 		mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1642 		goto err_unmap_eq;
1643 	}
1644 
1645 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1646 				  init_hca->dmpt_base,
1647 				  dev_cap->dmpt_entry_sz,
1648 				  dev->caps.num_mpts,
1649 				  dev->caps.reserved_mrws, 1, 1);
1650 	if (err) {
1651 		mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1652 		goto err_unmap_mtt;
1653 	}
1654 
1655 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1656 				  init_hca->qpc_base,
1657 				  dev_cap->qpc_entry_sz,
1658 				  dev->caps.num_qps,
1659 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1660 				  0, 0);
1661 	if (err) {
1662 		mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1663 		goto err_unmap_dmpt;
1664 	}
1665 
1666 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1667 				  init_hca->auxc_base,
1668 				  dev_cap->aux_entry_sz,
1669 				  dev->caps.num_qps,
1670 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1671 				  0, 0);
1672 	if (err) {
1673 		mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1674 		goto err_unmap_qp;
1675 	}
1676 
1677 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1678 				  init_hca->altc_base,
1679 				  dev_cap->altc_entry_sz,
1680 				  dev->caps.num_qps,
1681 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1682 				  0, 0);
1683 	if (err) {
1684 		mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1685 		goto err_unmap_auxc;
1686 	}
1687 
1688 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1689 				  init_hca->rdmarc_base,
1690 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1691 				  dev->caps.num_qps,
1692 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1693 				  0, 0);
1694 	if (err) {
1695 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1696 		goto err_unmap_altc;
1697 	}
1698 
1699 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1700 				  init_hca->cqc_base,
1701 				  dev_cap->cqc_entry_sz,
1702 				  dev->caps.num_cqs,
1703 				  dev->caps.reserved_cqs, 0, 0);
1704 	if (err) {
1705 		mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1706 		goto err_unmap_rdmarc;
1707 	}
1708 
1709 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1710 				  init_hca->srqc_base,
1711 				  dev_cap->srq_entry_sz,
1712 				  dev->caps.num_srqs,
1713 				  dev->caps.reserved_srqs, 0, 0);
1714 	if (err) {
1715 		mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1716 		goto err_unmap_cq;
1717 	}
1718 
1719 	/*
1720 	 * For flow steering device managed mode it is required to use
1721 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1722 	 * required, but for simplicity just map the whole multicast
1723 	 * group table now.  The table isn't very big and it's a lot
1724 	 * easier than trying to track ref counts.
1725 	 */
1726 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1727 				  init_hca->mc_base,
1728 				  mlx4_get_mgm_entry_size(dev),
1729 				  dev->caps.num_mgms + dev->caps.num_amgms,
1730 				  dev->caps.num_mgms + dev->caps.num_amgms,
1731 				  0, 0);
1732 	if (err) {
1733 		mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1734 		goto err_unmap_srq;
1735 	}
1736 
1737 	return 0;
1738 
1739 err_unmap_srq:
1740 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1741 
1742 err_unmap_cq:
1743 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1744 
1745 err_unmap_rdmarc:
1746 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1747 
1748 err_unmap_altc:
1749 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1750 
1751 err_unmap_auxc:
1752 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1753 
1754 err_unmap_qp:
1755 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1756 
1757 err_unmap_dmpt:
1758 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1759 
1760 err_unmap_mtt:
1761 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1762 
1763 err_unmap_eq:
1764 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1765 
1766 err_unmap_cmpt:
1767 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1768 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1769 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1770 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1771 
1772 err_unmap_aux:
1773 	mlx4_UNMAP_ICM_AUX(dev);
1774 
1775 err_free_aux:
1776 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1777 
1778 	return err;
1779 }
1780 
1781 static void mlx4_free_icms(struct mlx4_dev *dev)
1782 {
1783 	struct mlx4_priv *priv = mlx4_priv(dev);
1784 
1785 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1786 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1787 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1788 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1789 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1790 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1791 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1792 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1793 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1794 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1795 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1796 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1797 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1798 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1799 
1800 	mlx4_UNMAP_ICM_AUX(dev);
1801 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1802 }
1803 
1804 static void mlx4_slave_exit(struct mlx4_dev *dev)
1805 {
1806 	struct mlx4_priv *priv = mlx4_priv(dev);
1807 
1808 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1809 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1810 			  MLX4_COMM_TIME))
1811 		mlx4_warn(dev, "Failed to close slave function\n");
1812 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1813 }
1814 
1815 static int map_bf_area(struct mlx4_dev *dev)
1816 {
1817 	struct mlx4_priv *priv = mlx4_priv(dev);
1818 	resource_size_t bf_start;
1819 	resource_size_t bf_len;
1820 	int err = 0;
1821 
1822 	if (!dev->caps.bf_reg_size)
1823 		return -ENXIO;
1824 
1825 	bf_start = pci_resource_start(dev->persist->pdev, 2) +
1826 			(dev->caps.num_uars << PAGE_SHIFT);
1827 	bf_len = pci_resource_len(dev->persist->pdev, 2) -
1828 			(dev->caps.num_uars << PAGE_SHIFT);
1829 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1830 	if (!priv->bf_mapping)
1831 		err = -ENOMEM;
1832 
1833 	return err;
1834 }
1835 
1836 static void unmap_bf_area(struct mlx4_dev *dev)
1837 {
1838 	if (mlx4_priv(dev)->bf_mapping)
1839 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1840 }
1841 
1842 u64 mlx4_read_clock(struct mlx4_dev *dev)
1843 {
1844 	u32 clockhi, clocklo, clockhi1;
1845 	u64 cycles;
1846 	int i;
1847 	struct mlx4_priv *priv = mlx4_priv(dev);
1848 
1849 	for (i = 0; i < 10; i++) {
1850 		clockhi = swab32(readl(priv->clock_mapping));
1851 		clocklo = swab32(readl(priv->clock_mapping + 4));
1852 		clockhi1 = swab32(readl(priv->clock_mapping));
1853 		if (clockhi == clockhi1)
1854 			break;
1855 	}
1856 
1857 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1858 
1859 	return cycles;
1860 }
1861 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1862 
1863 
1864 static int map_internal_clock(struct mlx4_dev *dev)
1865 {
1866 	struct mlx4_priv *priv = mlx4_priv(dev);
1867 
1868 	priv->clock_mapping =
1869 		ioremap(pci_resource_start(dev->persist->pdev,
1870 					   priv->fw.clock_bar) +
1871 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1872 
1873 	if (!priv->clock_mapping)
1874 		return -ENOMEM;
1875 
1876 	return 0;
1877 }
1878 
1879 int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1880 				   struct mlx4_clock_params *params)
1881 {
1882 	struct mlx4_priv *priv = mlx4_priv(dev);
1883 
1884 	if (mlx4_is_slave(dev))
1885 		return -EOPNOTSUPP;
1886 
1887 	if (!params)
1888 		return -EINVAL;
1889 
1890 	params->bar = priv->fw.clock_bar;
1891 	params->offset = priv->fw.clock_offset;
1892 	params->size = MLX4_CLOCK_SIZE;
1893 
1894 	return 0;
1895 }
1896 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1897 
1898 static void unmap_internal_clock(struct mlx4_dev *dev)
1899 {
1900 	struct mlx4_priv *priv = mlx4_priv(dev);
1901 
1902 	if (priv->clock_mapping)
1903 		iounmap(priv->clock_mapping);
1904 }
1905 
1906 static void mlx4_close_hca(struct mlx4_dev *dev)
1907 {
1908 	unmap_internal_clock(dev);
1909 	unmap_bf_area(dev);
1910 	if (mlx4_is_slave(dev))
1911 		mlx4_slave_exit(dev);
1912 	else {
1913 		mlx4_CLOSE_HCA(dev, 0);
1914 		mlx4_free_icms(dev);
1915 	}
1916 }
1917 
1918 static void mlx4_close_fw(struct mlx4_dev *dev)
1919 {
1920 	if (!mlx4_is_slave(dev)) {
1921 		mlx4_UNMAP_FA(dev);
1922 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1923 	}
1924 }
1925 
1926 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1927 {
1928 #define COMM_CHAN_OFFLINE_OFFSET 0x09
1929 
1930 	u32 comm_flags;
1931 	u32 offline_bit;
1932 	unsigned long end;
1933 	struct mlx4_priv *priv = mlx4_priv(dev);
1934 
1935 	end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
1936 	while (time_before(jiffies, end)) {
1937 		comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
1938 					  MLX4_COMM_CHAN_FLAGS));
1939 		offline_bit = (comm_flags &
1940 			       (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
1941 		if (!offline_bit)
1942 			return 0;
1943 		/* There are cases as part of AER/Reset flow that PF needs
1944 		 * around 100 msec to load. We therefore sleep for 100 msec
1945 		 * to allow other tasks to make use of that CPU during this
1946 		 * time interval.
1947 		 */
1948 		msleep(100);
1949 	}
1950 	mlx4_err(dev, "Communication channel is offline.\n");
1951 	return -EIO;
1952 }
1953 
1954 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
1955 {
1956 #define COMM_CHAN_RST_OFFSET 0x1e
1957 
1958 	struct mlx4_priv *priv = mlx4_priv(dev);
1959 	u32 comm_rst;
1960 	u32 comm_caps;
1961 
1962 	comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
1963 				 MLX4_COMM_CHAN_CAPS));
1964 	comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
1965 
1966 	if (comm_rst)
1967 		dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
1968 }
1969 
1970 static int mlx4_init_slave(struct mlx4_dev *dev)
1971 {
1972 	struct mlx4_priv *priv = mlx4_priv(dev);
1973 	u64 dma = (u64) priv->mfunc.vhcr_dma;
1974 	int ret_from_reset = 0;
1975 	u32 slave_read;
1976 	u32 cmd_channel_ver;
1977 
1978 	if (atomic_read(&pf_loading)) {
1979 		mlx4_warn(dev, "PF is not ready - Deferring probe\n");
1980 		return -EPROBE_DEFER;
1981 	}
1982 
1983 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1984 	priv->cmd.max_cmds = 1;
1985 	if (mlx4_comm_check_offline(dev)) {
1986 		mlx4_err(dev, "PF is not responsive, skipping initialization\n");
1987 		goto err_offline;
1988 	}
1989 
1990 	mlx4_reset_vf_support(dev);
1991 	mlx4_warn(dev, "Sending reset\n");
1992 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1993 				       MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
1994 	/* if we are in the middle of flr the slave will try
1995 	 * NUM_OF_RESET_RETRIES times before leaving.*/
1996 	if (ret_from_reset) {
1997 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1998 			mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
1999 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
2000 			return -EPROBE_DEFER;
2001 		} else
2002 			goto err;
2003 	}
2004 
2005 	/* check the driver version - the slave I/F revision
2006 	 * must match the master's */
2007 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
2008 	cmd_channel_ver = mlx4_comm_get_version();
2009 
2010 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
2011 		MLX4_COMM_GET_IF_REV(slave_read)) {
2012 		mlx4_err(dev, "slave driver version is not supported by the master\n");
2013 		goto err;
2014 	}
2015 
2016 	mlx4_warn(dev, "Sending vhcr0\n");
2017 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
2018 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2019 		goto err;
2020 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
2021 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2022 		goto err;
2023 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
2024 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2025 		goto err;
2026 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
2027 			  MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2028 		goto err;
2029 
2030 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2031 	return 0;
2032 
2033 err:
2034 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
2035 err_offline:
2036 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2037 	return -EIO;
2038 }
2039 
2040 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
2041 {
2042 	int i;
2043 
2044 	for (i = 1; i <= dev->caps.num_ports; i++) {
2045 		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
2046 			dev->caps.gid_table_len[i] =
2047 				mlx4_get_slave_num_gids(dev, 0, i);
2048 		else
2049 			dev->caps.gid_table_len[i] = 1;
2050 		dev->caps.pkey_table_len[i] =
2051 			dev->phys_caps.pkey_phys_table_len[i] - 1;
2052 	}
2053 }
2054 
2055 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
2056 {
2057 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
2058 
2059 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
2060 	      i++) {
2061 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
2062 			break;
2063 	}
2064 
2065 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
2066 }
2067 
2068 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
2069 {
2070 	switch (dmfs_high_steer_mode) {
2071 	case MLX4_STEERING_DMFS_A0_DEFAULT:
2072 		return "default performance";
2073 
2074 	case MLX4_STEERING_DMFS_A0_DYNAMIC:
2075 		return "dynamic hybrid mode";
2076 
2077 	case MLX4_STEERING_DMFS_A0_STATIC:
2078 		return "performance optimized for limited rule configuration (static)";
2079 
2080 	case MLX4_STEERING_DMFS_A0_DISABLE:
2081 		return "disabled performance optimized steering";
2082 
2083 	case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
2084 		return "performance optimized steering not supported";
2085 
2086 	default:
2087 		return "Unrecognized mode";
2088 	}
2089 }
2090 
2091 #define MLX4_DMFS_A0_STEERING			(1UL << 2)
2092 
2093 static void choose_steering_mode(struct mlx4_dev *dev,
2094 				 struct mlx4_dev_cap *dev_cap)
2095 {
2096 	if (mlx4_log_num_mgm_entry_size <= 0) {
2097 		if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
2098 			if (dev->caps.dmfs_high_steer_mode ==
2099 			    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2100 				mlx4_err(dev, "DMFS high rate mode not supported\n");
2101 			else
2102 				dev->caps.dmfs_high_steer_mode =
2103 					MLX4_STEERING_DMFS_A0_STATIC;
2104 		}
2105 	}
2106 
2107 	if (mlx4_log_num_mgm_entry_size <= 0 &&
2108 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
2109 	    (!mlx4_is_mfunc(dev) ||
2110 	     (dev_cap->fs_max_num_qp_per_entry >=
2111 	     (dev->persist->num_vfs + 1))) &&
2112 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
2113 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
2114 		dev->oper_log_mgm_entry_size =
2115 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
2116 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
2117 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
2118 		dev->caps.fs_log_max_ucast_qp_range_size =
2119 			dev_cap->fs_log_max_ucast_qp_range_size;
2120 	} else {
2121 		if (dev->caps.dmfs_high_steer_mode !=
2122 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2123 			dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
2124 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
2125 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2126 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
2127 		else {
2128 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
2129 
2130 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
2131 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2132 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
2133 		}
2134 		dev->oper_log_mgm_entry_size =
2135 			mlx4_log_num_mgm_entry_size > 0 ?
2136 			mlx4_log_num_mgm_entry_size :
2137 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
2138 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
2139 	}
2140 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
2141 		 mlx4_steering_mode_str(dev->caps.steering_mode),
2142 		 dev->oper_log_mgm_entry_size,
2143 		 mlx4_log_num_mgm_entry_size);
2144 }
2145 
2146 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
2147 				       struct mlx4_dev_cap *dev_cap)
2148 {
2149 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2150 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
2151 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
2152 	else
2153 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
2154 
2155 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
2156 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
2157 }
2158 
2159 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
2160 {
2161 	int i;
2162 	struct mlx4_port_cap port_cap;
2163 
2164 	if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2165 		return -EINVAL;
2166 
2167 	for (i = 1; i <= dev->caps.num_ports; i++) {
2168 		if (mlx4_dev_port(dev, i, &port_cap)) {
2169 			mlx4_err(dev,
2170 				 "QUERY_DEV_CAP command failed, can't veify DMFS high rate steering.\n");
2171 		} else if ((dev->caps.dmfs_high_steer_mode !=
2172 			    MLX4_STEERING_DMFS_A0_DEFAULT) &&
2173 			   (port_cap.dmfs_optimized_state ==
2174 			    !!(dev->caps.dmfs_high_steer_mode ==
2175 			    MLX4_STEERING_DMFS_A0_DISABLE))) {
2176 			mlx4_err(dev,
2177 				 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2178 				 dmfs_high_rate_steering_mode_str(
2179 					dev->caps.dmfs_high_steer_mode),
2180 				 (port_cap.dmfs_optimized_state ?
2181 					"enabled" : "disabled"));
2182 		}
2183 	}
2184 
2185 	return 0;
2186 }
2187 
2188 static int mlx4_init_fw(struct mlx4_dev *dev)
2189 {
2190 	struct mlx4_mod_stat_cfg   mlx4_cfg;
2191 	int err = 0;
2192 
2193 	if (!mlx4_is_slave(dev)) {
2194 		err = mlx4_QUERY_FW(dev);
2195 		if (err) {
2196 			if (err == -EACCES)
2197 				mlx4_info(dev, "non-primary physical function, skipping\n");
2198 			else
2199 				mlx4_err(dev, "QUERY_FW command failed, aborting\n");
2200 			return err;
2201 		}
2202 
2203 		err = mlx4_load_fw(dev);
2204 		if (err) {
2205 			mlx4_err(dev, "Failed to start FW, aborting\n");
2206 			return err;
2207 		}
2208 
2209 		mlx4_cfg.log_pg_sz_m = 1;
2210 		mlx4_cfg.log_pg_sz = 0;
2211 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2212 		if (err)
2213 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2214 	}
2215 
2216 	return err;
2217 }
2218 
2219 static int mlx4_init_hca(struct mlx4_dev *dev)
2220 {
2221 	struct mlx4_priv	  *priv = mlx4_priv(dev);
2222 	struct mlx4_adapter	   adapter;
2223 	struct mlx4_dev_cap	   dev_cap;
2224 	struct mlx4_profile	   profile;
2225 	struct mlx4_init_hca_param init_hca;
2226 	u64 icm_size;
2227 	struct mlx4_config_dev_params params;
2228 	int err;
2229 
2230 	if (!mlx4_is_slave(dev)) {
2231 		err = mlx4_dev_cap(dev, &dev_cap);
2232 		if (err) {
2233 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
2234 			return err;
2235 		}
2236 
2237 		choose_steering_mode(dev, &dev_cap);
2238 		choose_tunnel_offload_mode(dev, &dev_cap);
2239 
2240 		if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2241 		    mlx4_is_master(dev))
2242 			dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2243 
2244 		err = mlx4_get_phys_port_id(dev);
2245 		if (err)
2246 			mlx4_err(dev, "Fail to get physical port id\n");
2247 
2248 		if (mlx4_is_master(dev))
2249 			mlx4_parav_master_pf_caps(dev);
2250 
2251 		if (mlx4_low_memory_profile()) {
2252 			mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2253 			profile = low_mem_profile;
2254 		} else {
2255 			profile = default_profile;
2256 		}
2257 		if (dev->caps.steering_mode ==
2258 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
2259 			profile.num_mcg = MLX4_FS_NUM_MCG;
2260 
2261 		icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
2262 					     &init_hca);
2263 		if ((long long) icm_size < 0) {
2264 			err = icm_size;
2265 			return err;
2266 		}
2267 
2268 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
2269 
2270 		if (enable_4k_uar) {
2271 			init_hca.log_uar_sz = ilog2(dev->caps.num_uars) +
2272 						    PAGE_SHIFT - DEFAULT_UAR_PAGE_SHIFT;
2273 			init_hca.uar_page_sz = DEFAULT_UAR_PAGE_SHIFT - 12;
2274 		} else {
2275 			init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
2276 			init_hca.uar_page_sz = PAGE_SHIFT - 12;
2277 		}
2278 
2279 		init_hca.mw_enabled = 0;
2280 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2281 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2282 			init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
2283 
2284 		err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
2285 		if (err)
2286 			return err;
2287 
2288 		err = mlx4_INIT_HCA(dev, &init_hca);
2289 		if (err) {
2290 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
2291 			goto err_free_icm;
2292 		}
2293 
2294 		if (dev_cap.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2295 			err = mlx4_query_func(dev, &dev_cap);
2296 			if (err < 0) {
2297 				mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
2298 				goto err_close;
2299 			} else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2300 				dev->caps.num_eqs = dev_cap.max_eqs;
2301 				dev->caps.reserved_eqs = dev_cap.reserved_eqs;
2302 				dev->caps.reserved_uars = dev_cap.reserved_uars;
2303 			}
2304 		}
2305 
2306 		/*
2307 		 * If TS is supported by FW
2308 		 * read HCA frequency by QUERY_HCA command
2309 		 */
2310 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2311 			memset(&init_hca, 0, sizeof(init_hca));
2312 			err = mlx4_QUERY_HCA(dev, &init_hca);
2313 			if (err) {
2314 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
2315 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2316 			} else {
2317 				dev->caps.hca_core_clock =
2318 					init_hca.hca_core_clock;
2319 			}
2320 
2321 			/* In case we got HCA frequency 0 - disable timestamping
2322 			 * to avoid dividing by zero
2323 			 */
2324 			if (!dev->caps.hca_core_clock) {
2325 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2326 				mlx4_err(dev,
2327 					 "HCA frequency is 0 - timestamping is not supported\n");
2328 			} else if (map_internal_clock(dev)) {
2329 				/*
2330 				 * Map internal clock,
2331 				 * in case of failure disable timestamping
2332 				 */
2333 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2334 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
2335 			}
2336 		}
2337 
2338 		if (dev->caps.dmfs_high_steer_mode !=
2339 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2340 			if (mlx4_validate_optimized_steering(dev))
2341 				mlx4_warn(dev, "Optimized steering validation failed\n");
2342 
2343 			if (dev->caps.dmfs_high_steer_mode ==
2344 			    MLX4_STEERING_DMFS_A0_DISABLE) {
2345 				dev->caps.dmfs_high_rate_qpn_base =
2346 					dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2347 				dev->caps.dmfs_high_rate_qpn_range =
2348 					MLX4_A0_STEERING_TABLE_SIZE;
2349 			}
2350 
2351 			mlx4_dbg(dev, "DMFS high rate steer mode is: %s\n",
2352 				 dmfs_high_rate_steering_mode_str(
2353 					dev->caps.dmfs_high_steer_mode));
2354 		}
2355 	} else {
2356 		err = mlx4_init_slave(dev);
2357 		if (err) {
2358 			if (err != -EPROBE_DEFER)
2359 				mlx4_err(dev, "Failed to initialize slave\n");
2360 			return err;
2361 		}
2362 
2363 		err = mlx4_slave_cap(dev);
2364 		if (err) {
2365 			mlx4_err(dev, "Failed to obtain slave caps\n");
2366 			goto err_close;
2367 		}
2368 	}
2369 
2370 	if (map_bf_area(dev))
2371 		mlx4_dbg(dev, "Failed to map blue flame area\n");
2372 
2373 	/*Only the master set the ports, all the rest got it from it.*/
2374 	if (!mlx4_is_slave(dev))
2375 		mlx4_set_port_mask(dev);
2376 
2377 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
2378 	if (err) {
2379 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2380 		goto unmap_bf;
2381 	}
2382 
2383 	/* Query CONFIG_DEV parameters */
2384 	err = mlx4_config_dev_retrieval(dev, &params);
2385 	if (err && err != -EOPNOTSUPP) {
2386 		mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2387 	} else if (!err) {
2388 		dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2389 		dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2390 	}
2391 	priv->eq_table.inta_pin = adapter.inta_pin;
2392 	memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
2393 
2394 	return 0;
2395 
2396 unmap_bf:
2397 	unmap_internal_clock(dev);
2398 	unmap_bf_area(dev);
2399 
2400 	if (mlx4_is_slave(dev)) {
2401 		kfree(dev->caps.qp0_qkey);
2402 		kfree(dev->caps.qp0_tunnel);
2403 		kfree(dev->caps.qp0_proxy);
2404 		kfree(dev->caps.qp1_tunnel);
2405 		kfree(dev->caps.qp1_proxy);
2406 	}
2407 
2408 err_close:
2409 	if (mlx4_is_slave(dev))
2410 		mlx4_slave_exit(dev);
2411 	else
2412 		mlx4_CLOSE_HCA(dev, 0);
2413 
2414 err_free_icm:
2415 	if (!mlx4_is_slave(dev))
2416 		mlx4_free_icms(dev);
2417 
2418 	return err;
2419 }
2420 
2421 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2422 {
2423 	struct mlx4_priv *priv = mlx4_priv(dev);
2424 	int nent_pow2;
2425 
2426 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2427 		return -ENOENT;
2428 
2429 	if (!dev->caps.max_counters)
2430 		return -ENOSPC;
2431 
2432 	nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2433 	/* reserve last counter index for sink counter */
2434 	return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2435 				nent_pow2 - 1, 0,
2436 				nent_pow2 - dev->caps.max_counters + 1);
2437 }
2438 
2439 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2440 {
2441 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2442 		return;
2443 
2444 	if (!dev->caps.max_counters)
2445 		return;
2446 
2447 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2448 }
2449 
2450 static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2451 {
2452 	struct mlx4_priv *priv = mlx4_priv(dev);
2453 	int port;
2454 
2455 	for (port = 0; port < dev->caps.num_ports; port++)
2456 		if (priv->def_counter[port] != -1)
2457 			mlx4_counter_free(dev,  priv->def_counter[port]);
2458 }
2459 
2460 static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2461 {
2462 	struct mlx4_priv *priv = mlx4_priv(dev);
2463 	int port, err = 0;
2464 	u32 idx;
2465 
2466 	for (port = 0; port < dev->caps.num_ports; port++)
2467 		priv->def_counter[port] = -1;
2468 
2469 	for (port = 0; port < dev->caps.num_ports; port++) {
2470 		err = mlx4_counter_alloc(dev, &idx);
2471 
2472 		if (!err || err == -ENOSPC) {
2473 			priv->def_counter[port] = idx;
2474 		} else if (err == -ENOENT) {
2475 			err = 0;
2476 			continue;
2477 		} else if (mlx4_is_slave(dev) && err == -EINVAL) {
2478 			priv->def_counter[port] = MLX4_SINK_COUNTER_INDEX(dev);
2479 			mlx4_warn(dev, "can't allocate counter from old PF driver, using index %d\n",
2480 				  MLX4_SINK_COUNTER_INDEX(dev));
2481 			err = 0;
2482 		} else {
2483 			mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2484 				 __func__, port + 1, err);
2485 			mlx4_cleanup_default_counters(dev);
2486 			return err;
2487 		}
2488 
2489 		mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2490 			 __func__, priv->def_counter[port], port + 1);
2491 	}
2492 
2493 	return err;
2494 }
2495 
2496 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2497 {
2498 	struct mlx4_priv *priv = mlx4_priv(dev);
2499 
2500 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2501 		return -ENOENT;
2502 
2503 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2504 	if (*idx == -1) {
2505 		*idx = MLX4_SINK_COUNTER_INDEX(dev);
2506 		return -ENOSPC;
2507 	}
2508 
2509 	return 0;
2510 }
2511 
2512 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2513 {
2514 	u64 out_param;
2515 	int err;
2516 
2517 	if (mlx4_is_mfunc(dev)) {
2518 		err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
2519 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2520 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2521 		if (!err)
2522 			*idx = get_param_l(&out_param);
2523 
2524 		return err;
2525 	}
2526 	return __mlx4_counter_alloc(dev, idx);
2527 }
2528 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2529 
2530 static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2531 				u8 counter_index)
2532 {
2533 	struct mlx4_cmd_mailbox *if_stat_mailbox;
2534 	int err;
2535 	u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2536 
2537 	if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2538 	if (IS_ERR(if_stat_mailbox))
2539 		return PTR_ERR(if_stat_mailbox);
2540 
2541 	err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2542 			   MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2543 			   MLX4_CMD_NATIVE);
2544 
2545 	mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2546 	return err;
2547 }
2548 
2549 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2550 {
2551 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2552 		return;
2553 
2554 	if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2555 		return;
2556 
2557 	__mlx4_clear_if_stat(dev, idx);
2558 
2559 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2560 	return;
2561 }
2562 
2563 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2564 {
2565 	u64 in_param = 0;
2566 
2567 	if (mlx4_is_mfunc(dev)) {
2568 		set_param_l(&in_param, idx);
2569 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2570 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2571 			 MLX4_CMD_WRAPPED);
2572 		return;
2573 	}
2574 	__mlx4_counter_free(dev, idx);
2575 }
2576 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2577 
2578 int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2579 {
2580 	struct mlx4_priv *priv = mlx4_priv(dev);
2581 
2582 	return priv->def_counter[port - 1];
2583 }
2584 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2585 
2586 void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2587 {
2588 	struct mlx4_priv *priv = mlx4_priv(dev);
2589 
2590 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2591 }
2592 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2593 
2594 __be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2595 {
2596 	struct mlx4_priv *priv = mlx4_priv(dev);
2597 
2598 	return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2599 }
2600 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2601 
2602 void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2603 {
2604 	struct mlx4_priv *priv = mlx4_priv(dev);
2605 	__be64 guid;
2606 
2607 	/* hw GUID */
2608 	if (entry == 0)
2609 		return;
2610 
2611 	get_random_bytes((char *)&guid, sizeof(guid));
2612 	guid &= ~(cpu_to_be64(1ULL << 56));
2613 	guid |= cpu_to_be64(1ULL << 57);
2614 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2615 }
2616 
2617 static int mlx4_setup_hca(struct mlx4_dev *dev)
2618 {
2619 	struct mlx4_priv *priv = mlx4_priv(dev);
2620 	int err;
2621 	int port;
2622 	__be32 ib_port_default_caps;
2623 
2624 	err = mlx4_init_uar_table(dev);
2625 	if (err) {
2626 		mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2627 		return err;
2628 	}
2629 
2630 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
2631 	if (err) {
2632 		mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2633 		goto err_uar_table_free;
2634 	}
2635 
2636 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2637 	if (!priv->kar) {
2638 		mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2639 		err = -ENOMEM;
2640 		goto err_uar_free;
2641 	}
2642 
2643 	err = mlx4_init_pd_table(dev);
2644 	if (err) {
2645 		mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2646 		goto err_kar_unmap;
2647 	}
2648 
2649 	err = mlx4_init_xrcd_table(dev);
2650 	if (err) {
2651 		mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2652 		goto err_pd_table_free;
2653 	}
2654 
2655 	err = mlx4_init_mr_table(dev);
2656 	if (err) {
2657 		mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2658 		goto err_xrcd_table_free;
2659 	}
2660 
2661 	if (!mlx4_is_slave(dev)) {
2662 		err = mlx4_init_mcg_table(dev);
2663 		if (err) {
2664 			mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2665 			goto err_mr_table_free;
2666 		}
2667 		err = mlx4_config_mad_demux(dev);
2668 		if (err) {
2669 			mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2670 			goto err_mcg_table_free;
2671 		}
2672 	}
2673 
2674 	err = mlx4_init_eq_table(dev);
2675 	if (err) {
2676 		mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2677 		goto err_mcg_table_free;
2678 	}
2679 
2680 	err = mlx4_cmd_use_events(dev);
2681 	if (err) {
2682 		mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2683 		goto err_eq_table_free;
2684 	}
2685 
2686 	err = mlx4_NOP(dev);
2687 	if (err) {
2688 		if (dev->flags & MLX4_FLAG_MSI_X) {
2689 			mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2690 				  priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2691 			mlx4_warn(dev, "Trying again without MSI-X\n");
2692 		} else {
2693 			mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2694 				 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2695 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2696 		}
2697 
2698 		goto err_cmd_poll;
2699 	}
2700 
2701 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
2702 
2703 	err = mlx4_init_cq_table(dev);
2704 	if (err) {
2705 		mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2706 		goto err_cmd_poll;
2707 	}
2708 
2709 	err = mlx4_init_srq_table(dev);
2710 	if (err) {
2711 		mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2712 		goto err_cq_table_free;
2713 	}
2714 
2715 	err = mlx4_init_qp_table(dev);
2716 	if (err) {
2717 		mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2718 		goto err_srq_table_free;
2719 	}
2720 
2721 	if (!mlx4_is_slave(dev)) {
2722 		err = mlx4_init_counters_table(dev);
2723 		if (err && err != -ENOENT) {
2724 			mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2725 			goto err_qp_table_free;
2726 		}
2727 	}
2728 
2729 	err = mlx4_allocate_default_counters(dev);
2730 	if (err) {
2731 		mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2732 		goto err_counters_table_free;
2733 	}
2734 
2735 	if (!mlx4_is_slave(dev)) {
2736 		for (port = 1; port <= dev->caps.num_ports; port++) {
2737 			ib_port_default_caps = 0;
2738 			err = mlx4_get_port_ib_caps(dev, port,
2739 						    &ib_port_default_caps);
2740 			if (err)
2741 				mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2742 					  port, err);
2743 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2744 
2745 			/* initialize per-slave default ib port capabilities */
2746 			if (mlx4_is_master(dev)) {
2747 				int i;
2748 				for (i = 0; i < dev->num_slaves; i++) {
2749 					if (i == mlx4_master_func_num(dev))
2750 						continue;
2751 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2752 						ib_port_default_caps;
2753 				}
2754 			}
2755 
2756 			if (mlx4_is_mfunc(dev))
2757 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2758 			else
2759 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2760 
2761 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2762 					    dev->caps.pkey_table_len[port] : -1);
2763 			if (err) {
2764 				mlx4_err(dev, "Failed to set port %d, aborting\n",
2765 					 port);
2766 				goto err_default_countes_free;
2767 			}
2768 		}
2769 	}
2770 
2771 	return 0;
2772 
2773 err_default_countes_free:
2774 	mlx4_cleanup_default_counters(dev);
2775 
2776 err_counters_table_free:
2777 	if (!mlx4_is_slave(dev))
2778 		mlx4_cleanup_counters_table(dev);
2779 
2780 err_qp_table_free:
2781 	mlx4_cleanup_qp_table(dev);
2782 
2783 err_srq_table_free:
2784 	mlx4_cleanup_srq_table(dev);
2785 
2786 err_cq_table_free:
2787 	mlx4_cleanup_cq_table(dev);
2788 
2789 err_cmd_poll:
2790 	mlx4_cmd_use_polling(dev);
2791 
2792 err_eq_table_free:
2793 	mlx4_cleanup_eq_table(dev);
2794 
2795 err_mcg_table_free:
2796 	if (!mlx4_is_slave(dev))
2797 		mlx4_cleanup_mcg_table(dev);
2798 
2799 err_mr_table_free:
2800 	mlx4_cleanup_mr_table(dev);
2801 
2802 err_xrcd_table_free:
2803 	mlx4_cleanup_xrcd_table(dev);
2804 
2805 err_pd_table_free:
2806 	mlx4_cleanup_pd_table(dev);
2807 
2808 err_kar_unmap:
2809 	iounmap(priv->kar);
2810 
2811 err_uar_free:
2812 	mlx4_uar_free(dev, &priv->driver_uar);
2813 
2814 err_uar_table_free:
2815 	mlx4_cleanup_uar_table(dev);
2816 	return err;
2817 }
2818 
2819 static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2820 {
2821 	int requested_cpu = 0;
2822 	struct mlx4_priv *priv = mlx4_priv(dev);
2823 	struct mlx4_eq *eq;
2824 	int off = 0;
2825 	int i;
2826 
2827 	if (eqn > dev->caps.num_comp_vectors)
2828 		return -EINVAL;
2829 
2830 	for (i = 1; i < port; i++)
2831 		off += mlx4_get_eqs_per_port(dev, i);
2832 
2833 	requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2834 
2835 	/* Meaning EQs are shared, and this call comes from the second port */
2836 	if (requested_cpu < 0)
2837 		return 0;
2838 
2839 	eq = &priv->eq_table.eq[eqn];
2840 
2841 	if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2842 		return -ENOMEM;
2843 
2844 	cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2845 
2846 	return 0;
2847 }
2848 
2849 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2850 {
2851 	struct mlx4_priv *priv = mlx4_priv(dev);
2852 	struct msix_entry *entries;
2853 	int i;
2854 	int port = 0;
2855 
2856 	if (msi_x) {
2857 		int nreq = dev->caps.num_ports * num_online_cpus() + 1;
2858 
2859 		nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
2860 			     nreq);
2861 		if (nreq > MAX_MSIX)
2862 			nreq = MAX_MSIX;
2863 
2864 		entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
2865 		if (!entries)
2866 			goto no_msi;
2867 
2868 		for (i = 0; i < nreq; ++i)
2869 			entries[i].entry = i;
2870 
2871 		nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2872 					     nreq);
2873 
2874 		if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
2875 			kfree(entries);
2876 			goto no_msi;
2877 		}
2878 		/* 1 is reserved for events (asyncrounous EQ) */
2879 		dev->caps.num_comp_vectors = nreq - 1;
2880 
2881 		priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2882 		bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2883 			    dev->caps.num_ports);
2884 
2885 		for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2886 			if (i == MLX4_EQ_ASYNC)
2887 				continue;
2888 
2889 			priv->eq_table.eq[i].irq =
2890 				entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2891 
2892 			if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2893 				bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2894 					    dev->caps.num_ports);
2895 				/* We don't set affinity hint when there
2896 				 * aren't enough EQs
2897 				 */
2898 			} else {
2899 				set_bit(port,
2900 					priv->eq_table.eq[i].actv_ports.ports);
2901 				if (mlx4_init_affinity_hint(dev, port + 1, i))
2902 					mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
2903 						  i);
2904 			}
2905 			/* We divide the Eqs evenly between the two ports.
2906 			 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
2907 			 * refers to the number of Eqs per port
2908 			 * (i.e eqs_per_port). Theoretically, we would like to
2909 			 * write something like (i + 1) % eqs_per_port == 0.
2910 			 * However, since there's an asynchronous Eq, we have
2911 			 * to skip over it by comparing this condition to
2912 			 * !!((i + 1) > MLX4_EQ_ASYNC).
2913 			 */
2914 			if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
2915 			    ((i + 1) %
2916 			     (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
2917 			    !!((i + 1) > MLX4_EQ_ASYNC))
2918 				/* If dev->caps.num_comp_vectors < dev->caps.num_ports,
2919 				 * everything is shared anyway.
2920 				 */
2921 				port++;
2922 		}
2923 
2924 		dev->flags |= MLX4_FLAG_MSI_X;
2925 
2926 		kfree(entries);
2927 		return;
2928 	}
2929 
2930 no_msi:
2931 	dev->caps.num_comp_vectors = 1;
2932 
2933 	BUG_ON(MLX4_EQ_ASYNC >= 2);
2934 	for (i = 0; i < 2; ++i) {
2935 		priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
2936 		if (i != MLX4_EQ_ASYNC) {
2937 			bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2938 				    dev->caps.num_ports);
2939 		}
2940 	}
2941 }
2942 
2943 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2944 {
2945 	struct devlink *devlink = priv_to_devlink(mlx4_priv(dev));
2946 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2947 	int err;
2948 
2949 	err = devlink_port_register(devlink, &info->devlink_port, port);
2950 	if (err)
2951 		return err;
2952 
2953 	info->dev = dev;
2954 	info->port = port;
2955 	if (!mlx4_is_slave(dev)) {
2956 		mlx4_init_mac_table(dev, &info->mac_table);
2957 		mlx4_init_vlan_table(dev, &info->vlan_table);
2958 		mlx4_init_roce_gid_table(dev, &info->gid_table);
2959 		info->base_qpn = mlx4_get_base_qpn(dev, port);
2960 	}
2961 
2962 	sprintf(info->dev_name, "mlx4_port%d", port);
2963 	info->port_attr.attr.name = info->dev_name;
2964 	if (mlx4_is_mfunc(dev))
2965 		info->port_attr.attr.mode = S_IRUGO;
2966 	else {
2967 		info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2968 		info->port_attr.store     = set_port_type;
2969 	}
2970 	info->port_attr.show      = show_port_type;
2971 	sysfs_attr_init(&info->port_attr.attr);
2972 
2973 	err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
2974 	if (err) {
2975 		mlx4_err(dev, "Failed to create file for port %d\n", port);
2976 		devlink_port_unregister(&info->devlink_port);
2977 		info->port = -1;
2978 	}
2979 
2980 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2981 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
2982 	if (mlx4_is_mfunc(dev))
2983 		info->port_mtu_attr.attr.mode = S_IRUGO;
2984 	else {
2985 		info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2986 		info->port_mtu_attr.store     = set_port_ib_mtu;
2987 	}
2988 	info->port_mtu_attr.show      = show_port_ib_mtu;
2989 	sysfs_attr_init(&info->port_mtu_attr.attr);
2990 
2991 	err = device_create_file(&dev->persist->pdev->dev,
2992 				 &info->port_mtu_attr);
2993 	if (err) {
2994 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
2995 		device_remove_file(&info->dev->persist->pdev->dev,
2996 				   &info->port_attr);
2997 		devlink_port_unregister(&info->devlink_port);
2998 		info->port = -1;
2999 	}
3000 
3001 	return err;
3002 }
3003 
3004 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
3005 {
3006 	if (info->port < 0)
3007 		return;
3008 
3009 	device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
3010 	device_remove_file(&info->dev->persist->pdev->dev,
3011 			   &info->port_mtu_attr);
3012 	devlink_port_unregister(&info->devlink_port);
3013 
3014 #ifdef CONFIG_RFS_ACCEL
3015 	free_irq_cpu_rmap(info->rmap);
3016 	info->rmap = NULL;
3017 #endif
3018 }
3019 
3020 static int mlx4_init_steering(struct mlx4_dev *dev)
3021 {
3022 	struct mlx4_priv *priv = mlx4_priv(dev);
3023 	int num_entries = dev->caps.num_ports;
3024 	int i, j;
3025 
3026 	priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
3027 	if (!priv->steer)
3028 		return -ENOMEM;
3029 
3030 	for (i = 0; i < num_entries; i++)
3031 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3032 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
3033 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
3034 		}
3035 	return 0;
3036 }
3037 
3038 static void mlx4_clear_steering(struct mlx4_dev *dev)
3039 {
3040 	struct mlx4_priv *priv = mlx4_priv(dev);
3041 	struct mlx4_steer_index *entry, *tmp_entry;
3042 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
3043 	int num_entries = dev->caps.num_ports;
3044 	int i, j;
3045 
3046 	for (i = 0; i < num_entries; i++) {
3047 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3048 			list_for_each_entry_safe(pqp, tmp_pqp,
3049 						 &priv->steer[i].promisc_qps[j],
3050 						 list) {
3051 				list_del(&pqp->list);
3052 				kfree(pqp);
3053 			}
3054 			list_for_each_entry_safe(entry, tmp_entry,
3055 						 &priv->steer[i].steer_entries[j],
3056 						 list) {
3057 				list_del(&entry->list);
3058 				list_for_each_entry_safe(pqp, tmp_pqp,
3059 							 &entry->duplicates,
3060 							 list) {
3061 					list_del(&pqp->list);
3062 					kfree(pqp);
3063 				}
3064 				kfree(entry);
3065 			}
3066 		}
3067 	}
3068 	kfree(priv->steer);
3069 }
3070 
3071 static int extended_func_num(struct pci_dev *pdev)
3072 {
3073 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
3074 }
3075 
3076 #define MLX4_OWNER_BASE	0x8069c
3077 #define MLX4_OWNER_SIZE	4
3078 
3079 static int mlx4_get_ownership(struct mlx4_dev *dev)
3080 {
3081 	void __iomem *owner;
3082 	u32 ret;
3083 
3084 	if (pci_channel_offline(dev->persist->pdev))
3085 		return -EIO;
3086 
3087 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3088 			MLX4_OWNER_BASE,
3089 			MLX4_OWNER_SIZE);
3090 	if (!owner) {
3091 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3092 		return -ENOMEM;
3093 	}
3094 
3095 	ret = readl(owner);
3096 	iounmap(owner);
3097 	return (int) !!ret;
3098 }
3099 
3100 static void mlx4_free_ownership(struct mlx4_dev *dev)
3101 {
3102 	void __iomem *owner;
3103 
3104 	if (pci_channel_offline(dev->persist->pdev))
3105 		return;
3106 
3107 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3108 			MLX4_OWNER_BASE,
3109 			MLX4_OWNER_SIZE);
3110 	if (!owner) {
3111 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3112 		return;
3113 	}
3114 	writel(0, owner);
3115 	msleep(1000);
3116 	iounmap(owner);
3117 }
3118 
3119 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV)	==\
3120 				  !!((flags) & MLX4_FLAG_MASTER))
3121 
3122 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
3123 			     u8 total_vfs, int existing_vfs, int reset_flow)
3124 {
3125 	u64 dev_flags = dev->flags;
3126 	int err = 0;
3127 	int fw_enabled_sriov_vfs = min(pci_sriov_get_totalvfs(pdev),
3128 					MLX4_MAX_NUM_VF);
3129 
3130 	if (reset_flow) {
3131 		dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
3132 				       GFP_KERNEL);
3133 		if (!dev->dev_vfs)
3134 			goto free_mem;
3135 		return dev_flags;
3136 	}
3137 
3138 	atomic_inc(&pf_loading);
3139 	if (dev->flags &  MLX4_FLAG_SRIOV) {
3140 		if (existing_vfs != total_vfs) {
3141 			mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
3142 				 existing_vfs, total_vfs);
3143 			total_vfs = existing_vfs;
3144 		}
3145 	}
3146 
3147 	dev->dev_vfs = kzalloc(total_vfs * sizeof(*dev->dev_vfs), GFP_KERNEL);
3148 	if (NULL == dev->dev_vfs) {
3149 		mlx4_err(dev, "Failed to allocate memory for VFs\n");
3150 		goto disable_sriov;
3151 	}
3152 
3153 	if (!(dev->flags &  MLX4_FLAG_SRIOV)) {
3154 		if (total_vfs > fw_enabled_sriov_vfs) {
3155 			mlx4_err(dev, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
3156 				 total_vfs, fw_enabled_sriov_vfs);
3157 			err = -ENOMEM;
3158 			goto disable_sriov;
3159 		}
3160 		mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
3161 		err = pci_enable_sriov(pdev, total_vfs);
3162 	}
3163 	if (err) {
3164 		mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
3165 			 err);
3166 		goto disable_sriov;
3167 	} else {
3168 		mlx4_warn(dev, "Running in master mode\n");
3169 		dev_flags |= MLX4_FLAG_SRIOV |
3170 			MLX4_FLAG_MASTER;
3171 		dev_flags &= ~MLX4_FLAG_SLAVE;
3172 		dev->persist->num_vfs = total_vfs;
3173 	}
3174 	return dev_flags;
3175 
3176 disable_sriov:
3177 	atomic_dec(&pf_loading);
3178 free_mem:
3179 	dev->persist->num_vfs = 0;
3180 	kfree(dev->dev_vfs);
3181         dev->dev_vfs = NULL;
3182 	return dev_flags & ~MLX4_FLAG_MASTER;
3183 }
3184 
3185 enum {
3186 	MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
3187 };
3188 
3189 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
3190 			      int *nvfs)
3191 {
3192 	int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
3193 	/* Checking for 64 VFs as a limitation of CX2 */
3194 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
3195 	    requested_vfs >= 64) {
3196 		mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
3197 			 requested_vfs);
3198 		return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
3199 	}
3200 	return 0;
3201 }
3202 
3203 static int mlx4_pci_enable_device(struct mlx4_dev *dev)
3204 {
3205 	struct pci_dev *pdev = dev->persist->pdev;
3206 	int err = 0;
3207 
3208 	mutex_lock(&dev->persist->pci_status_mutex);
3209 	if (dev->persist->pci_status == MLX4_PCI_STATUS_DISABLED) {
3210 		err = pci_enable_device(pdev);
3211 		if (!err)
3212 			dev->persist->pci_status = MLX4_PCI_STATUS_ENABLED;
3213 	}
3214 	mutex_unlock(&dev->persist->pci_status_mutex);
3215 
3216 	return err;
3217 }
3218 
3219 static void mlx4_pci_disable_device(struct mlx4_dev *dev)
3220 {
3221 	struct pci_dev *pdev = dev->persist->pdev;
3222 
3223 	mutex_lock(&dev->persist->pci_status_mutex);
3224 	if (dev->persist->pci_status == MLX4_PCI_STATUS_ENABLED) {
3225 		pci_disable_device(pdev);
3226 		dev->persist->pci_status = MLX4_PCI_STATUS_DISABLED;
3227 	}
3228 	mutex_unlock(&dev->persist->pci_status_mutex);
3229 }
3230 
3231 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
3232 			 int total_vfs, int *nvfs, struct mlx4_priv *priv,
3233 			 int reset_flow)
3234 {
3235 	struct mlx4_dev *dev;
3236 	unsigned sum = 0;
3237 	int err;
3238 	int port;
3239 	int i;
3240 	struct mlx4_dev_cap *dev_cap = NULL;
3241 	int existing_vfs = 0;
3242 
3243 	dev = &priv->dev;
3244 
3245 	INIT_LIST_HEAD(&priv->ctx_list);
3246 	spin_lock_init(&priv->ctx_lock);
3247 
3248 	mutex_init(&priv->port_mutex);
3249 	mutex_init(&priv->bond_mutex);
3250 
3251 	INIT_LIST_HEAD(&priv->pgdir_list);
3252 	mutex_init(&priv->pgdir_mutex);
3253 	spin_lock_init(&priv->cmd.context_lock);
3254 
3255 	INIT_LIST_HEAD(&priv->bf_list);
3256 	mutex_init(&priv->bf_mutex);
3257 
3258 	dev->rev_id = pdev->revision;
3259 	dev->numa_node = dev_to_node(&pdev->dev);
3260 
3261 	/* Detect if this device is a virtual function */
3262 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3263 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3264 		dev->flags |= MLX4_FLAG_SLAVE;
3265 	} else {
3266 		/* We reset the device and enable SRIOV only for physical
3267 		 * devices.  Try to claim ownership on the device;
3268 		 * if already taken, skip -- do not allow multiple PFs */
3269 		err = mlx4_get_ownership(dev);
3270 		if (err) {
3271 			if (err < 0)
3272 				return err;
3273 			else {
3274 				mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
3275 				return -EINVAL;
3276 			}
3277 		}
3278 
3279 		atomic_set(&priv->opreq_count, 0);
3280 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3281 
3282 		/*
3283 		 * Now reset the HCA before we touch the PCI capabilities or
3284 		 * attempt a firmware command, since a boot ROM may have left
3285 		 * the HCA in an undefined state.
3286 		 */
3287 		err = mlx4_reset(dev);
3288 		if (err) {
3289 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
3290 			goto err_sriov;
3291 		}
3292 
3293 		if (total_vfs) {
3294 			dev->flags = MLX4_FLAG_MASTER;
3295 			existing_vfs = pci_num_vf(pdev);
3296 			if (existing_vfs)
3297 				dev->flags |= MLX4_FLAG_SRIOV;
3298 			dev->persist->num_vfs = total_vfs;
3299 		}
3300 	}
3301 
3302 	/* on load remove any previous indication of internal error,
3303 	 * device is up.
3304 	 */
3305 	dev->persist->state = MLX4_DEVICE_STATE_UP;
3306 
3307 slave_start:
3308 	err = mlx4_cmd_init(dev);
3309 	if (err) {
3310 		mlx4_err(dev, "Failed to init command interface, aborting\n");
3311 		goto err_sriov;
3312 	}
3313 
3314 	/* In slave functions, the communication channel must be initialized
3315 	 * before posting commands. Also, init num_slaves before calling
3316 	 * mlx4_init_hca */
3317 	if (mlx4_is_mfunc(dev)) {
3318 		if (mlx4_is_master(dev)) {
3319 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
3320 
3321 		} else {
3322 			dev->num_slaves = 0;
3323 			err = mlx4_multi_func_init(dev);
3324 			if (err) {
3325 				mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
3326 				goto err_cmd;
3327 			}
3328 		}
3329 	}
3330 
3331 	err = mlx4_init_fw(dev);
3332 	if (err) {
3333 		mlx4_err(dev, "Failed to init fw, aborting.\n");
3334 		goto err_mfunc;
3335 	}
3336 
3337 	if (mlx4_is_master(dev)) {
3338 		/* when we hit the goto slave_start below, dev_cap already initialized */
3339 		if (!dev_cap) {
3340 			dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3341 
3342 			if (!dev_cap) {
3343 				err = -ENOMEM;
3344 				goto err_fw;
3345 			}
3346 
3347 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3348 			if (err) {
3349 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3350 				goto err_fw;
3351 			}
3352 
3353 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3354 				goto err_fw;
3355 
3356 			if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3357 				u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3358 								  total_vfs,
3359 								  existing_vfs,
3360 								  reset_flow);
3361 
3362 				mlx4_close_fw(dev);
3363 				mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3364 				dev->flags = dev_flags;
3365 				if (!SRIOV_VALID_STATE(dev->flags)) {
3366 					mlx4_err(dev, "Invalid SRIOV state\n");
3367 					goto err_sriov;
3368 				}
3369 				err = mlx4_reset(dev);
3370 				if (err) {
3371 					mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3372 					goto err_sriov;
3373 				}
3374 				goto slave_start;
3375 			}
3376 		} else {
3377 			/* Legacy mode FW requires SRIOV to be enabled before
3378 			 * doing QUERY_DEV_CAP, since max_eq's value is different if
3379 			 * SRIOV is enabled.
3380 			 */
3381 			memset(dev_cap, 0, sizeof(*dev_cap));
3382 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3383 			if (err) {
3384 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3385 				goto err_fw;
3386 			}
3387 
3388 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3389 				goto err_fw;
3390 		}
3391 	}
3392 
3393 	err = mlx4_init_hca(dev);
3394 	if (err) {
3395 		if (err == -EACCES) {
3396 			/* Not primary Physical function
3397 			 * Running in slave mode */
3398 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3399 			/* We're not a PF */
3400 			if (dev->flags & MLX4_FLAG_SRIOV) {
3401 				if (!existing_vfs)
3402 					pci_disable_sriov(pdev);
3403 				if (mlx4_is_master(dev) && !reset_flow)
3404 					atomic_dec(&pf_loading);
3405 				dev->flags &= ~MLX4_FLAG_SRIOV;
3406 			}
3407 			if (!mlx4_is_slave(dev))
3408 				mlx4_free_ownership(dev);
3409 			dev->flags |= MLX4_FLAG_SLAVE;
3410 			dev->flags &= ~MLX4_FLAG_MASTER;
3411 			goto slave_start;
3412 		} else
3413 			goto err_fw;
3414 	}
3415 
3416 	if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3417 		u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3418 						  existing_vfs, reset_flow);
3419 
3420 		if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3421 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3422 			dev->flags = dev_flags;
3423 			err = mlx4_cmd_init(dev);
3424 			if (err) {
3425 				/* Only VHCR is cleaned up, so could still
3426 				 * send FW commands
3427 				 */
3428 				mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3429 				goto err_close;
3430 			}
3431 		} else {
3432 			dev->flags = dev_flags;
3433 		}
3434 
3435 		if (!SRIOV_VALID_STATE(dev->flags)) {
3436 			mlx4_err(dev, "Invalid SRIOV state\n");
3437 			goto err_close;
3438 		}
3439 	}
3440 
3441 	/* check if the device is functioning at its maximum possible speed.
3442 	 * No return code for this call, just warn the user in case of PCI
3443 	 * express device capabilities are under-satisfied by the bus.
3444 	 */
3445 	if (!mlx4_is_slave(dev))
3446 		mlx4_check_pcie_caps(dev);
3447 
3448 	/* In master functions, the communication channel must be initialized
3449 	 * after obtaining its address from fw */
3450 	if (mlx4_is_master(dev)) {
3451 		if (dev->caps.num_ports < 2 &&
3452 		    num_vfs_argc > 1) {
3453 			err = -EINVAL;
3454 			mlx4_err(dev,
3455 				 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3456 				 dev->caps.num_ports);
3457 			goto err_close;
3458 		}
3459 		memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
3460 
3461 		for (i = 0;
3462 		     i < sizeof(dev->persist->nvfs)/
3463 		     sizeof(dev->persist->nvfs[0]); i++) {
3464 			unsigned j;
3465 
3466 			for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
3467 				dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3468 				dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3469 					dev->caps.num_ports;
3470 			}
3471 		}
3472 
3473 		/* In master functions, the communication channel
3474 		 * must be initialized after obtaining its address from fw
3475 		 */
3476 		err = mlx4_multi_func_init(dev);
3477 		if (err) {
3478 			mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3479 			goto err_close;
3480 		}
3481 	}
3482 
3483 	err = mlx4_alloc_eq_table(dev);
3484 	if (err)
3485 		goto err_master_mfunc;
3486 
3487 	bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
3488 	mutex_init(&priv->msix_ctl.pool_lock);
3489 
3490 	mlx4_enable_msi_x(dev);
3491 	if ((mlx4_is_mfunc(dev)) &&
3492 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
3493 		err = -EOPNOTSUPP;
3494 		mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
3495 		goto err_free_eq;
3496 	}
3497 
3498 	if (!mlx4_is_slave(dev)) {
3499 		err = mlx4_init_steering(dev);
3500 		if (err)
3501 			goto err_disable_msix;
3502 	}
3503 
3504 	mlx4_init_quotas(dev);
3505 
3506 	err = mlx4_setup_hca(dev);
3507 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3508 	    !mlx4_is_mfunc(dev)) {
3509 		dev->flags &= ~MLX4_FLAG_MSI_X;
3510 		dev->caps.num_comp_vectors = 1;
3511 		pci_disable_msix(pdev);
3512 		err = mlx4_setup_hca(dev);
3513 	}
3514 
3515 	if (err)
3516 		goto err_steer;
3517 
3518 	/* When PF resources are ready arm its comm channel to enable
3519 	 * getting commands
3520 	 */
3521 	if (mlx4_is_master(dev)) {
3522 		err = mlx4_ARM_COMM_CHANNEL(dev);
3523 		if (err) {
3524 			mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3525 				 err);
3526 			goto err_steer;
3527 		}
3528 	}
3529 
3530 	for (port = 1; port <= dev->caps.num_ports; port++) {
3531 		err = mlx4_init_port_info(dev, port);
3532 		if (err)
3533 			goto err_port;
3534 	}
3535 
3536 	priv->v2p.port1 = 1;
3537 	priv->v2p.port2 = 2;
3538 
3539 	err = mlx4_register_device(dev);
3540 	if (err)
3541 		goto err_port;
3542 
3543 	mlx4_request_modules(dev);
3544 
3545 	mlx4_sense_init(dev);
3546 	mlx4_start_sense(dev);
3547 
3548 	priv->removed = 0;
3549 
3550 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3551 		atomic_dec(&pf_loading);
3552 
3553 	kfree(dev_cap);
3554 	return 0;
3555 
3556 err_port:
3557 	for (--port; port >= 1; --port)
3558 		mlx4_cleanup_port_info(&priv->port[port]);
3559 
3560 	mlx4_cleanup_default_counters(dev);
3561 	if (!mlx4_is_slave(dev))
3562 		mlx4_cleanup_counters_table(dev);
3563 	mlx4_cleanup_qp_table(dev);
3564 	mlx4_cleanup_srq_table(dev);
3565 	mlx4_cleanup_cq_table(dev);
3566 	mlx4_cmd_use_polling(dev);
3567 	mlx4_cleanup_eq_table(dev);
3568 	mlx4_cleanup_mcg_table(dev);
3569 	mlx4_cleanup_mr_table(dev);
3570 	mlx4_cleanup_xrcd_table(dev);
3571 	mlx4_cleanup_pd_table(dev);
3572 	mlx4_cleanup_uar_table(dev);
3573 
3574 err_steer:
3575 	if (!mlx4_is_slave(dev))
3576 		mlx4_clear_steering(dev);
3577 
3578 err_disable_msix:
3579 	if (dev->flags & MLX4_FLAG_MSI_X)
3580 		pci_disable_msix(pdev);
3581 
3582 err_free_eq:
3583 	mlx4_free_eq_table(dev);
3584 
3585 err_master_mfunc:
3586 	if (mlx4_is_master(dev)) {
3587 		mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
3588 		mlx4_multi_func_cleanup(dev);
3589 	}
3590 
3591 	if (mlx4_is_slave(dev)) {
3592 		kfree(dev->caps.qp0_qkey);
3593 		kfree(dev->caps.qp0_tunnel);
3594 		kfree(dev->caps.qp0_proxy);
3595 		kfree(dev->caps.qp1_tunnel);
3596 		kfree(dev->caps.qp1_proxy);
3597 	}
3598 
3599 err_close:
3600 	mlx4_close_hca(dev);
3601 
3602 err_fw:
3603 	mlx4_close_fw(dev);
3604 
3605 err_mfunc:
3606 	if (mlx4_is_slave(dev))
3607 		mlx4_multi_func_cleanup(dev);
3608 
3609 err_cmd:
3610 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3611 
3612 err_sriov:
3613 	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3614 		pci_disable_sriov(pdev);
3615 		dev->flags &= ~MLX4_FLAG_SRIOV;
3616 	}
3617 
3618 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3619 		atomic_dec(&pf_loading);
3620 
3621 	kfree(priv->dev.dev_vfs);
3622 
3623 	if (!mlx4_is_slave(dev))
3624 		mlx4_free_ownership(dev);
3625 
3626 	kfree(dev_cap);
3627 	return err;
3628 }
3629 
3630 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3631 			   struct mlx4_priv *priv)
3632 {
3633 	int err;
3634 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3635 	int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3636 	const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3637 		{2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3638 	unsigned total_vfs = 0;
3639 	unsigned int i;
3640 
3641 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3642 
3643 	err = mlx4_pci_enable_device(&priv->dev);
3644 	if (err) {
3645 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3646 		return err;
3647 	}
3648 
3649 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3650 	 * per port, we must limit the number of VFs to 63 (since their are
3651 	 * 128 MACs)
3652 	 */
3653 	for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
3654 	     total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3655 		nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3656 		if (nvfs[i] < 0) {
3657 			dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3658 			err = -EINVAL;
3659 			goto err_disable_pdev;
3660 		}
3661 	}
3662 	for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
3663 	     i++) {
3664 		prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3665 		if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3666 			dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3667 			err = -EINVAL;
3668 			goto err_disable_pdev;
3669 		}
3670 	}
3671 	if (total_vfs > MLX4_MAX_NUM_VF) {
3672 		dev_err(&pdev->dev,
3673 			"Requested more VF's (%d) than allowed by hw (%d)\n",
3674 			total_vfs, MLX4_MAX_NUM_VF);
3675 		err = -EINVAL;
3676 		goto err_disable_pdev;
3677 	}
3678 
3679 	for (i = 0; i < MLX4_MAX_PORTS; i++) {
3680 		if (nvfs[i] + nvfs[2] > MLX4_MAX_NUM_VF_P_PORT) {
3681 			dev_err(&pdev->dev,
3682 				"Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3683 				nvfs[i] + nvfs[2], i + 1,
3684 				MLX4_MAX_NUM_VF_P_PORT);
3685 			err = -EINVAL;
3686 			goto err_disable_pdev;
3687 		}
3688 	}
3689 
3690 	/* Check for BARs. */
3691 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3692 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3693 		dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3694 			pci_dev_data, pci_resource_flags(pdev, 0));
3695 		err = -ENODEV;
3696 		goto err_disable_pdev;
3697 	}
3698 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3699 		dev_err(&pdev->dev, "Missing UAR, aborting\n");
3700 		err = -ENODEV;
3701 		goto err_disable_pdev;
3702 	}
3703 
3704 	err = pci_request_regions(pdev, DRV_NAME);
3705 	if (err) {
3706 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3707 		goto err_disable_pdev;
3708 	}
3709 
3710 	pci_set_master(pdev);
3711 
3712 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3713 	if (err) {
3714 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3715 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3716 		if (err) {
3717 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3718 			goto err_release_regions;
3719 		}
3720 	}
3721 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3722 	if (err) {
3723 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3724 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3725 		if (err) {
3726 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3727 			goto err_release_regions;
3728 		}
3729 	}
3730 
3731 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
3732 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3733 	/* Detect if this device is a virtual function */
3734 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3735 		/* When acting as pf, we normally skip vfs unless explicitly
3736 		 * requested to probe them.
3737 		 */
3738 		if (total_vfs) {
3739 			unsigned vfs_offset = 0;
3740 
3741 			for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
3742 			     vfs_offset + nvfs[i] < extended_func_num(pdev);
3743 			     vfs_offset += nvfs[i], i++)
3744 				;
3745 			if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
3746 				err = -ENODEV;
3747 				goto err_release_regions;
3748 			}
3749 			if ((extended_func_num(pdev) - vfs_offset)
3750 			    > prb_vf[i]) {
3751 				dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3752 					 extended_func_num(pdev));
3753 				err = -ENODEV;
3754 				goto err_release_regions;
3755 			}
3756 		}
3757 	}
3758 
3759 	err = mlx4_catas_init(&priv->dev);
3760 	if (err)
3761 		goto err_release_regions;
3762 
3763 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3764 	if (err)
3765 		goto err_catas;
3766 
3767 	return 0;
3768 
3769 err_catas:
3770 	mlx4_catas_end(&priv->dev);
3771 
3772 err_release_regions:
3773 	pci_release_regions(pdev);
3774 
3775 err_disable_pdev:
3776 	mlx4_pci_disable_device(&priv->dev);
3777 	pci_set_drvdata(pdev, NULL);
3778 	return err;
3779 }
3780 
3781 static int mlx4_devlink_port_type_set(struct devlink_port *devlink_port,
3782 				      enum devlink_port_type port_type)
3783 {
3784 	struct mlx4_port_info *info = container_of(devlink_port,
3785 						   struct mlx4_port_info,
3786 						   devlink_port);
3787 	enum mlx4_port_type mlx4_port_type;
3788 
3789 	switch (port_type) {
3790 	case DEVLINK_PORT_TYPE_AUTO:
3791 		mlx4_port_type = MLX4_PORT_TYPE_AUTO;
3792 		break;
3793 	case DEVLINK_PORT_TYPE_ETH:
3794 		mlx4_port_type = MLX4_PORT_TYPE_ETH;
3795 		break;
3796 	case DEVLINK_PORT_TYPE_IB:
3797 		mlx4_port_type = MLX4_PORT_TYPE_IB;
3798 		break;
3799 	default:
3800 		return -EOPNOTSUPP;
3801 	}
3802 
3803 	return __set_port_type(info, mlx4_port_type);
3804 }
3805 
3806 static const struct devlink_ops mlx4_devlink_ops = {
3807 	.port_type_set	= mlx4_devlink_port_type_set,
3808 };
3809 
3810 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3811 {
3812 	struct devlink *devlink;
3813 	struct mlx4_priv *priv;
3814 	struct mlx4_dev *dev;
3815 	int ret;
3816 
3817 	printk_once(KERN_INFO "%s", mlx4_version);
3818 
3819 	devlink = devlink_alloc(&mlx4_devlink_ops, sizeof(*priv));
3820 	if (!devlink)
3821 		return -ENOMEM;
3822 	priv = devlink_priv(devlink);
3823 
3824 	dev       = &priv->dev;
3825 	dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3826 	if (!dev->persist) {
3827 		ret = -ENOMEM;
3828 		goto err_devlink_free;
3829 	}
3830 	dev->persist->pdev = pdev;
3831 	dev->persist->dev = dev;
3832 	pci_set_drvdata(pdev, dev->persist);
3833 	priv->pci_dev_data = id->driver_data;
3834 	mutex_init(&dev->persist->device_state_mutex);
3835 	mutex_init(&dev->persist->interface_state_mutex);
3836 	mutex_init(&dev->persist->pci_status_mutex);
3837 
3838 	ret = devlink_register(devlink, &pdev->dev);
3839 	if (ret)
3840 		goto err_persist_free;
3841 
3842 	ret =  __mlx4_init_one(pdev, id->driver_data, priv);
3843 	if (ret)
3844 		goto err_devlink_unregister;
3845 
3846 	pci_save_state(pdev);
3847 	return 0;
3848 
3849 err_devlink_unregister:
3850 	devlink_unregister(devlink);
3851 err_persist_free:
3852 	kfree(dev->persist);
3853 err_devlink_free:
3854 	devlink_free(devlink);
3855 	return ret;
3856 }
3857 
3858 static void mlx4_clean_dev(struct mlx4_dev *dev)
3859 {
3860 	struct mlx4_dev_persistent *persist = dev->persist;
3861 	struct mlx4_priv *priv = mlx4_priv(dev);
3862 	unsigned long	flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
3863 
3864 	memset(priv, 0, sizeof(*priv));
3865 	priv->dev.persist = persist;
3866 	priv->dev.flags = flags;
3867 }
3868 
3869 static void mlx4_unload_one(struct pci_dev *pdev)
3870 {
3871 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3872 	struct mlx4_dev  *dev  = persist->dev;
3873 	struct mlx4_priv *priv = mlx4_priv(dev);
3874 	int               pci_dev_data;
3875 	int p, i;
3876 
3877 	if (priv->removed)
3878 		return;
3879 
3880 	/* saving current ports type for further use */
3881 	for (i = 0; i < dev->caps.num_ports; i++) {
3882 		dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
3883 		dev->persist->curr_port_poss_type[i] = dev->caps.
3884 						       possible_type[i + 1];
3885 	}
3886 
3887 	pci_dev_data = priv->pci_dev_data;
3888 
3889 	mlx4_stop_sense(dev);
3890 	mlx4_unregister_device(dev);
3891 
3892 	for (p = 1; p <= dev->caps.num_ports; p++) {
3893 		mlx4_cleanup_port_info(&priv->port[p]);
3894 		mlx4_CLOSE_PORT(dev, p);
3895 	}
3896 
3897 	if (mlx4_is_master(dev))
3898 		mlx4_free_resource_tracker(dev,
3899 					   RES_TR_FREE_SLAVES_ONLY);
3900 
3901 	mlx4_cleanup_default_counters(dev);
3902 	if (!mlx4_is_slave(dev))
3903 		mlx4_cleanup_counters_table(dev);
3904 	mlx4_cleanup_qp_table(dev);
3905 	mlx4_cleanup_srq_table(dev);
3906 	mlx4_cleanup_cq_table(dev);
3907 	mlx4_cmd_use_polling(dev);
3908 	mlx4_cleanup_eq_table(dev);
3909 	mlx4_cleanup_mcg_table(dev);
3910 	mlx4_cleanup_mr_table(dev);
3911 	mlx4_cleanup_xrcd_table(dev);
3912 	mlx4_cleanup_pd_table(dev);
3913 
3914 	if (mlx4_is_master(dev))
3915 		mlx4_free_resource_tracker(dev,
3916 					   RES_TR_FREE_STRUCTS_ONLY);
3917 
3918 	iounmap(priv->kar);
3919 	mlx4_uar_free(dev, &priv->driver_uar);
3920 	mlx4_cleanup_uar_table(dev);
3921 	if (!mlx4_is_slave(dev))
3922 		mlx4_clear_steering(dev);
3923 	mlx4_free_eq_table(dev);
3924 	if (mlx4_is_master(dev))
3925 		mlx4_multi_func_cleanup(dev);
3926 	mlx4_close_hca(dev);
3927 	mlx4_close_fw(dev);
3928 	if (mlx4_is_slave(dev))
3929 		mlx4_multi_func_cleanup(dev);
3930 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3931 
3932 	if (dev->flags & MLX4_FLAG_MSI_X)
3933 		pci_disable_msix(pdev);
3934 
3935 	if (!mlx4_is_slave(dev))
3936 		mlx4_free_ownership(dev);
3937 
3938 	kfree(dev->caps.qp0_qkey);
3939 	kfree(dev->caps.qp0_tunnel);
3940 	kfree(dev->caps.qp0_proxy);
3941 	kfree(dev->caps.qp1_tunnel);
3942 	kfree(dev->caps.qp1_proxy);
3943 	kfree(dev->dev_vfs);
3944 
3945 	mlx4_clean_dev(dev);
3946 	priv->pci_dev_data = pci_dev_data;
3947 	priv->removed = 1;
3948 }
3949 
3950 static void mlx4_remove_one(struct pci_dev *pdev)
3951 {
3952 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3953 	struct mlx4_dev  *dev  = persist->dev;
3954 	struct mlx4_priv *priv = mlx4_priv(dev);
3955 	struct devlink *devlink = priv_to_devlink(priv);
3956 	int active_vfs = 0;
3957 
3958 	mutex_lock(&persist->interface_state_mutex);
3959 	persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
3960 	mutex_unlock(&persist->interface_state_mutex);
3961 
3962 	/* Disabling SR-IOV is not allowed while there are active vf's */
3963 	if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
3964 		active_vfs = mlx4_how_many_lives_vf(dev);
3965 		if (active_vfs) {
3966 			pr_warn("Removing PF when there are active VF's !!\n");
3967 			pr_warn("Will not disable SR-IOV.\n");
3968 		}
3969 	}
3970 
3971 	/* device marked to be under deletion running now without the lock
3972 	 * letting other tasks to be terminated
3973 	 */
3974 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3975 		mlx4_unload_one(pdev);
3976 	else
3977 		mlx4_info(dev, "%s: interface is down\n", __func__);
3978 	mlx4_catas_end(dev);
3979 	if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
3980 		mlx4_warn(dev, "Disabling SR-IOV\n");
3981 		pci_disable_sriov(pdev);
3982 	}
3983 
3984 	pci_release_regions(pdev);
3985 	mlx4_pci_disable_device(dev);
3986 	devlink_unregister(devlink);
3987 	kfree(dev->persist);
3988 	devlink_free(devlink);
3989 	pci_set_drvdata(pdev, NULL);
3990 }
3991 
3992 static int restore_current_port_types(struct mlx4_dev *dev,
3993 				      enum mlx4_port_type *types,
3994 				      enum mlx4_port_type *poss_types)
3995 {
3996 	struct mlx4_priv *priv = mlx4_priv(dev);
3997 	int err, i;
3998 
3999 	mlx4_stop_sense(dev);
4000 
4001 	mutex_lock(&priv->port_mutex);
4002 	for (i = 0; i < dev->caps.num_ports; i++)
4003 		dev->caps.possible_type[i + 1] = poss_types[i];
4004 	err = mlx4_change_port_types(dev, types);
4005 	mlx4_start_sense(dev);
4006 	mutex_unlock(&priv->port_mutex);
4007 
4008 	return err;
4009 }
4010 
4011 int mlx4_restart_one(struct pci_dev *pdev)
4012 {
4013 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4014 	struct mlx4_dev	 *dev  = persist->dev;
4015 	struct mlx4_priv *priv = mlx4_priv(dev);
4016 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4017 	int pci_dev_data, err, total_vfs;
4018 
4019 	pci_dev_data = priv->pci_dev_data;
4020 	total_vfs = dev->persist->num_vfs;
4021 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4022 
4023 	mlx4_unload_one(pdev);
4024 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
4025 	if (err) {
4026 		mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
4027 			 __func__, pci_name(pdev), err);
4028 		return err;
4029 	}
4030 
4031 	err = restore_current_port_types(dev, dev->persist->curr_port_type,
4032 					 dev->persist->curr_port_poss_type);
4033 	if (err)
4034 		mlx4_err(dev, "could not restore original port types (%d)\n",
4035 			 err);
4036 
4037 	return err;
4038 }
4039 
4040 #define MLX_SP(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_FORCE_SENSE_PORT }
4041 #define MLX_VF(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_IS_VF }
4042 #define MLX_GN(id) { PCI_VDEVICE(MELLANOX, id), 0 }
4043 
4044 static const struct pci_device_id mlx4_pci_table[] = {
4045 	/* MT25408 "Hermon" */
4046 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_SDR),	/* SDR */
4047 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR),	/* DDR */
4048 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR),	/* QDR */
4049 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR_GEN2), /* DDR Gen2 */
4050 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR_GEN2),	/* QDR Gen2 */
4051 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN),	/* EN 10GigE */
4052 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN_GEN2),  /* EN 10GigE Gen2 */
4053 	/* MT25458 ConnectX EN 10GBASE-T */
4054 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN),
4055 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_T_GEN2),	/* Gen2 */
4056 	/* MT26468 ConnectX EN 10GigE PCIe Gen2*/
4057 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_GEN2),
4058 	/* MT26438 ConnectX EN 40GigE PCIe Gen2 5GT/s */
4059 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_5_GEN2),
4060 	/* MT26478 ConnectX2 40GigE PCIe Gen2 */
4061 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX2),
4062 	/* MT25400 Family [ConnectX-2] */
4063 	MLX_VF(0x1002),					/* Virtual Function */
4064 	/* MT27500 Family [ConnectX-3] */
4065 	MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3),
4066 	MLX_VF(0x1004),					/* Virtual Function */
4067 	MLX_GN(0x1005),					/* MT27510 Family */
4068 	MLX_GN(0x1006),					/* MT27511 Family */
4069 	MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3_PRO),	/* MT27520 Family */
4070 	MLX_GN(0x1008),					/* MT27521 Family */
4071 	MLX_GN(0x1009),					/* MT27530 Family */
4072 	MLX_GN(0x100a),					/* MT27531 Family */
4073 	MLX_GN(0x100b),					/* MT27540 Family */
4074 	MLX_GN(0x100c),					/* MT27541 Family */
4075 	MLX_GN(0x100d),					/* MT27550 Family */
4076 	MLX_GN(0x100e),					/* MT27551 Family */
4077 	MLX_GN(0x100f),					/* MT27560 Family */
4078 	MLX_GN(0x1010),					/* MT27561 Family */
4079 
4080 	/*
4081 	 * See the mellanox_check_broken_intx_masking() quirk when
4082 	 * adding devices
4083 	 */
4084 
4085 	{ 0, }
4086 };
4087 
4088 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
4089 
4090 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
4091 					      pci_channel_state_t state)
4092 {
4093 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4094 
4095 	mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
4096 	mlx4_enter_error_state(persist);
4097 
4098 	mutex_lock(&persist->interface_state_mutex);
4099 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4100 		mlx4_unload_one(pdev);
4101 
4102 	mutex_unlock(&persist->interface_state_mutex);
4103 	if (state == pci_channel_io_perm_failure)
4104 		return PCI_ERS_RESULT_DISCONNECT;
4105 
4106 	mlx4_pci_disable_device(persist->dev);
4107 	return PCI_ERS_RESULT_NEED_RESET;
4108 }
4109 
4110 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
4111 {
4112 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4113 	struct mlx4_dev	 *dev  = persist->dev;
4114 	int err;
4115 
4116 	mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
4117 	err = mlx4_pci_enable_device(dev);
4118 	if (err) {
4119 		mlx4_err(dev, "Can not re-enable device, err=%d\n", err);
4120 		return PCI_ERS_RESULT_DISCONNECT;
4121 	}
4122 
4123 	pci_set_master(pdev);
4124 	pci_restore_state(pdev);
4125 	pci_save_state(pdev);
4126 	return PCI_ERS_RESULT_RECOVERED;
4127 }
4128 
4129 static void mlx4_pci_resume(struct pci_dev *pdev)
4130 {
4131 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4132 	struct mlx4_dev	 *dev  = persist->dev;
4133 	struct mlx4_priv *priv = mlx4_priv(dev);
4134 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4135 	int total_vfs;
4136 	int err;
4137 
4138 	mlx4_err(dev, "%s was called\n", __func__);
4139 	total_vfs = dev->persist->num_vfs;
4140 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4141 
4142 	mutex_lock(&persist->interface_state_mutex);
4143 	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4144 		err = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
4145 				    priv, 1);
4146 		if (err) {
4147 			mlx4_err(dev, "%s: mlx4_load_one failed, err=%d\n",
4148 				 __func__,  err);
4149 			goto end;
4150 		}
4151 
4152 		err = restore_current_port_types(dev, dev->persist->
4153 						 curr_port_type, dev->persist->
4154 						 curr_port_poss_type);
4155 		if (err)
4156 			mlx4_err(dev, "could not restore original port types (%d)\n", err);
4157 	}
4158 end:
4159 	mutex_unlock(&persist->interface_state_mutex);
4160 
4161 }
4162 
4163 static void mlx4_shutdown(struct pci_dev *pdev)
4164 {
4165 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4166 
4167 	mlx4_info(persist->dev, "mlx4_shutdown was called\n");
4168 	mutex_lock(&persist->interface_state_mutex);
4169 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4170 		mlx4_unload_one(pdev);
4171 	mutex_unlock(&persist->interface_state_mutex);
4172 }
4173 
4174 static const struct pci_error_handlers mlx4_err_handler = {
4175 	.error_detected = mlx4_pci_err_detected,
4176 	.slot_reset     = mlx4_pci_slot_reset,
4177 	.resume		= mlx4_pci_resume,
4178 };
4179 
4180 static struct pci_driver mlx4_driver = {
4181 	.name		= DRV_NAME,
4182 	.id_table	= mlx4_pci_table,
4183 	.probe		= mlx4_init_one,
4184 	.shutdown	= mlx4_shutdown,
4185 	.remove		= mlx4_remove_one,
4186 	.err_handler    = &mlx4_err_handler,
4187 };
4188 
4189 static int __init mlx4_verify_params(void)
4190 {
4191 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
4192 		pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
4193 		return -1;
4194 	}
4195 
4196 	if (log_num_vlan != 0)
4197 		pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
4198 			MLX4_LOG_NUM_VLANS);
4199 
4200 	if (use_prio != 0)
4201 		pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
4202 
4203 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
4204 		pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
4205 			log_mtts_per_seg);
4206 		return -1;
4207 	}
4208 
4209 	/* Check if module param for ports type has legal combination */
4210 	if (port_type_array[0] == false && port_type_array[1] == true) {
4211 		pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
4212 		port_type_array[0] = true;
4213 	}
4214 
4215 	if (mlx4_log_num_mgm_entry_size < -7 ||
4216 	    (mlx4_log_num_mgm_entry_size > 0 &&
4217 	     (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
4218 	      mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
4219 		pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
4220 			mlx4_log_num_mgm_entry_size,
4221 			MLX4_MIN_MGM_LOG_ENTRY_SIZE,
4222 			MLX4_MAX_MGM_LOG_ENTRY_SIZE);
4223 		return -1;
4224 	}
4225 
4226 	return 0;
4227 }
4228 
4229 static int __init mlx4_init(void)
4230 {
4231 	int ret;
4232 
4233 	if (mlx4_verify_params())
4234 		return -EINVAL;
4235 
4236 
4237 	mlx4_wq = create_singlethread_workqueue("mlx4");
4238 	if (!mlx4_wq)
4239 		return -ENOMEM;
4240 
4241 	ret = pci_register_driver(&mlx4_driver);
4242 	if (ret < 0)
4243 		destroy_workqueue(mlx4_wq);
4244 	return ret < 0 ? ret : 0;
4245 }
4246 
4247 static void __exit mlx4_cleanup(void)
4248 {
4249 	pci_unregister_driver(&mlx4_driver);
4250 	destroy_workqueue(mlx4_wq);
4251 }
4252 
4253 module_init(mlx4_init);
4254 module_exit(mlx4_cleanup);
4255