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 <net/devlink.h>
46 
47 #include <linux/mlx4/device.h>
48 #include <linux/mlx4/doorbell.h>
49 
50 #include "mlx4.h"
51 #include "fw.h"
52 #include "icm.h"
53 
54 MODULE_AUTHOR("Roland Dreier");
55 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
56 MODULE_LICENSE("Dual BSD/GPL");
57 MODULE_VERSION(DRV_VERSION);
58 
59 struct workqueue_struct *mlx4_wq;
60 
61 #ifdef CONFIG_MLX4_DEBUG
62 
63 int mlx4_debug_level = 0;
64 module_param_named(debug_level, mlx4_debug_level, int, 0644);
65 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
66 
67 #endif /* CONFIG_MLX4_DEBUG */
68 
69 #ifdef CONFIG_PCI_MSI
70 
71 static int msi_x = 1;
72 module_param(msi_x, int, 0444);
73 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
74 
75 #else /* CONFIG_PCI_MSI */
76 
77 #define msi_x (0)
78 
79 #endif /* CONFIG_PCI_MSI */
80 
81 static uint8_t num_vfs[3] = {0, 0, 0};
82 static int num_vfs_argc;
83 module_param_array(num_vfs, byte , &num_vfs_argc, 0444);
84 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
85 			  "num_vfs=port1,port2,port1+2");
86 
87 static uint8_t probe_vf[3] = {0, 0, 0};
88 static int probe_vfs_argc;
89 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
90 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
91 			   "probe_vf=port1,port2,port1+2");
92 
93 int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
94 module_param_named(log_num_mgm_entry_size,
95 			mlx4_log_num_mgm_entry_size, int, 0444);
96 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
97 					 " of qp per mcg, for example:"
98 					 " 10 gives 248.range: 7 <="
99 					 " log_num_mgm_entry_size <= 12."
100 					 " To activate device managed"
101 					 " flow steering when available, set to -1");
102 
103 static bool enable_64b_cqe_eqe = true;
104 module_param(enable_64b_cqe_eqe, bool, 0444);
105 MODULE_PARM_DESC(enable_64b_cqe_eqe,
106 		 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
107 
108 static bool enable_4k_uar;
109 module_param(enable_4k_uar, bool, 0444);
110 MODULE_PARM_DESC(enable_4k_uar,
111 		 "Enable using 4K UAR. Should not be enabled if have VFs which do not support 4K UARs (default: false)");
112 
113 #define PF_CONTEXT_BEHAVIOUR_MASK	(MLX4_FUNC_CAP_64B_EQE_CQE | \
114 					 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
115 					 MLX4_FUNC_CAP_DMFS_A0_STATIC)
116 
117 #define RESET_PERSIST_MASK_FLAGS	(MLX4_FLAG_SRIOV)
118 
119 static char mlx4_version[] =
120 	DRV_NAME ": Mellanox ConnectX core driver v"
121 	DRV_VERSION " (" DRV_RELDATE ")\n";
122 
123 static struct mlx4_profile default_profile = {
124 	.num_qp		= 1 << 18,
125 	.num_srq	= 1 << 16,
126 	.rdmarc_per_qp	= 1 << 4,
127 	.num_cq		= 1 << 16,
128 	.num_mcg	= 1 << 13,
129 	.num_mpt	= 1 << 19,
130 	.num_mtt	= 1 << 20, /* It is really num mtt segements */
131 };
132 
133 static struct mlx4_profile low_mem_profile = {
134 	.num_qp		= 1 << 17,
135 	.num_srq	= 1 << 6,
136 	.rdmarc_per_qp	= 1 << 4,
137 	.num_cq		= 1 << 8,
138 	.num_mcg	= 1 << 8,
139 	.num_mpt	= 1 << 9,
140 	.num_mtt	= 1 << 7,
141 };
142 
143 static int log_num_mac = 7;
144 module_param_named(log_num_mac, log_num_mac, int, 0444);
145 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
146 
147 static int log_num_vlan;
148 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
149 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
150 /* Log2 max number of VLANs per ETH port (0-7) */
151 #define MLX4_LOG_NUM_VLANS 7
152 #define MLX4_MIN_LOG_NUM_VLANS 0
153 #define MLX4_MIN_LOG_NUM_MAC 1
154 
155 static bool use_prio;
156 module_param_named(use_prio, use_prio, bool, 0444);
157 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
158 
159 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
160 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
161 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
162 
163 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
164 static int arr_argc = 2;
165 module_param_array(port_type_array, int, &arr_argc, 0444);
166 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
167 				"1 for IB, 2 for Ethernet");
168 
169 struct mlx4_port_config {
170 	struct list_head list;
171 	enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
172 	struct pci_dev *pdev;
173 };
174 
175 static atomic_t pf_loading = ATOMIC_INIT(0);
176 
177 static inline void mlx4_set_num_reserved_uars(struct mlx4_dev *dev,
178 					      struct mlx4_dev_cap *dev_cap)
179 {
180 	/* The reserved_uars is calculated by system page size unit.
181 	 * Therefore, adjustment is added when the uar page size is less
182 	 * than the system page size
183 	 */
184 	dev->caps.reserved_uars	=
185 		max_t(int,
186 		      mlx4_get_num_reserved_uar(dev),
187 		      dev_cap->reserved_uars /
188 			(1 << (PAGE_SHIFT - dev->uar_page_shift)));
189 }
190 
191 int mlx4_check_port_params(struct mlx4_dev *dev,
192 			   enum mlx4_port_type *port_type)
193 {
194 	int i;
195 
196 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
197 		for (i = 0; i < dev->caps.num_ports - 1; i++) {
198 			if (port_type[i] != port_type[i + 1]) {
199 				mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
200 				return -EINVAL;
201 			}
202 		}
203 	}
204 
205 	for (i = 0; i < dev->caps.num_ports; i++) {
206 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
207 			mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
208 				 i + 1);
209 			return -EINVAL;
210 		}
211 	}
212 	return 0;
213 }
214 
215 static void mlx4_set_port_mask(struct mlx4_dev *dev)
216 {
217 	int i;
218 
219 	for (i = 1; i <= dev->caps.num_ports; ++i)
220 		dev->caps.port_mask[i] = dev->caps.port_type[i];
221 }
222 
223 enum {
224 	MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
225 };
226 
227 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
228 {
229 	int err = 0;
230 	struct mlx4_func func;
231 
232 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
233 		err = mlx4_QUERY_FUNC(dev, &func, 0);
234 		if (err) {
235 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
236 			return err;
237 		}
238 		dev_cap->max_eqs = func.max_eq;
239 		dev_cap->reserved_eqs = func.rsvd_eqs;
240 		dev_cap->reserved_uars = func.rsvd_uars;
241 		err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
242 	}
243 	return err;
244 }
245 
246 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
247 {
248 	struct mlx4_caps *dev_cap = &dev->caps;
249 
250 	/* FW not supporting or cancelled by user */
251 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
252 	    !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
253 		return;
254 
255 	/* Must have 64B CQE_EQE enabled by FW to use bigger stride
256 	 * When FW has NCSI it may decide not to report 64B CQE/EQEs
257 	 */
258 	if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
259 	    !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
260 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
261 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
262 		return;
263 	}
264 
265 	if (cache_line_size() == 128 || cache_line_size() == 256) {
266 		mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
267 		/* Changing the real data inside CQE size to 32B */
268 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
269 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
270 
271 		if (mlx4_is_master(dev))
272 			dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
273 	} else {
274 		if (cache_line_size() != 32  && cache_line_size() != 64)
275 			mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
276 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
277 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
278 	}
279 }
280 
281 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
282 			  struct mlx4_port_cap *port_cap)
283 {
284 	dev->caps.vl_cap[port]	    = port_cap->max_vl;
285 	dev->caps.ib_mtu_cap[port]	    = port_cap->ib_mtu;
286 	dev->phys_caps.gid_phys_table_len[port]  = port_cap->max_gids;
287 	dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
288 	/* set gid and pkey table operating lengths by default
289 	 * to non-sriov values
290 	 */
291 	dev->caps.gid_table_len[port]  = port_cap->max_gids;
292 	dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
293 	dev->caps.port_width_cap[port] = port_cap->max_port_width;
294 	dev->caps.eth_mtu_cap[port]    = port_cap->eth_mtu;
295 	dev->caps.max_tc_eth	       = port_cap->max_tc_eth;
296 	dev->caps.def_mac[port]        = port_cap->def_mac;
297 	dev->caps.supported_type[port] = port_cap->supported_port_types;
298 	dev->caps.suggested_type[port] = port_cap->suggested_type;
299 	dev->caps.default_sense[port] = port_cap->default_sense;
300 	dev->caps.trans_type[port]	    = port_cap->trans_type;
301 	dev->caps.vendor_oui[port]     = port_cap->vendor_oui;
302 	dev->caps.wavelength[port]     = port_cap->wavelength;
303 	dev->caps.trans_code[port]     = port_cap->trans_code;
304 
305 	return 0;
306 }
307 
308 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
309 			 struct mlx4_port_cap *port_cap)
310 {
311 	int err = 0;
312 
313 	err = mlx4_QUERY_PORT(dev, port, port_cap);
314 
315 	if (err)
316 		mlx4_err(dev, "QUERY_PORT command failed.\n");
317 
318 	return err;
319 }
320 
321 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
322 {
323 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
324 		return;
325 
326 	if (mlx4_is_mfunc(dev)) {
327 		mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
328 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
329 		return;
330 	}
331 
332 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
333 		mlx4_dbg(dev,
334 			 "Keep FCS is not supported - Disabling Ignore FCS");
335 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
336 		return;
337 	}
338 }
339 
340 #define MLX4_A0_STEERING_TABLE_SIZE	256
341 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
342 {
343 	int err;
344 	int i;
345 
346 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
347 	if (err) {
348 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
349 		return err;
350 	}
351 	mlx4_dev_cap_dump(dev, dev_cap);
352 
353 	if (dev_cap->min_page_sz > PAGE_SIZE) {
354 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
355 			 dev_cap->min_page_sz, PAGE_SIZE);
356 		return -ENODEV;
357 	}
358 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
359 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
360 			 dev_cap->num_ports, MLX4_MAX_PORTS);
361 		return -ENODEV;
362 	}
363 
364 	if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
365 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
366 			 dev_cap->uar_size,
367 			 (unsigned long long)
368 			 pci_resource_len(dev->persist->pdev, 2));
369 		return -ENODEV;
370 	}
371 
372 	dev->caps.num_ports	     = dev_cap->num_ports;
373 	dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
374 	dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
375 				      dev->caps.num_sys_eqs :
376 				      MLX4_MAX_EQ_NUM;
377 	for (i = 1; i <= dev->caps.num_ports; ++i) {
378 		err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
379 		if (err) {
380 			mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
381 			return err;
382 		}
383 	}
384 
385 	dev->caps.uar_page_size	     = PAGE_SIZE;
386 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
387 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
388 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
389 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
390 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
391 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
392 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
393 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
394 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
395 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
396 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
397 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
398 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
399 	/*
400 	 * Subtract 1 from the limit because we need to allocate a
401 	 * spare CQE so the HCA HW can tell the difference between an
402 	 * empty CQ and a full CQ.
403 	 */
404 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
405 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
406 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
407 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
408 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
409 
410 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
411 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
412 					dev_cap->reserved_xrcds : 0;
413 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
414 					dev_cap->max_xrcds : 0;
415 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
416 
417 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
418 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
419 	dev->caps.flags		     = dev_cap->flags;
420 	dev->caps.flags2	     = dev_cap->flags2;
421 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
422 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
423 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
424 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
425 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
426 
427 	/* Save uar page shift */
428 	if (!mlx4_is_slave(dev)) {
429 		/* Virtual PCI function needs to determine UAR page size from
430 		 * firmware. Only master PCI function can set the uar page size
431 		 */
432 		if (enable_4k_uar)
433 			dev->uar_page_shift = DEFAULT_UAR_PAGE_SHIFT;
434 		else
435 			dev->uar_page_shift = PAGE_SHIFT;
436 
437 		mlx4_set_num_reserved_uars(dev, dev_cap);
438 	}
439 
440 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN) {
441 		struct mlx4_init_hca_param hca_param;
442 
443 		memset(&hca_param, 0, sizeof(hca_param));
444 		err = mlx4_QUERY_HCA(dev, &hca_param);
445 		/* Turn off PHV_EN flag in case phv_check_en is set.
446 		 * phv_check_en is a HW check that parse the packet and verify
447 		 * phv bit was reported correctly in the wqe. To allow QinQ
448 		 * PHV_EN flag should be set and phv_check_en must be cleared
449 		 * otherwise QinQ packets will be drop by the HW.
450 		 */
451 		if (err || hca_param.phv_check_en)
452 			dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_PHV_EN;
453 	}
454 
455 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
456 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
457 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
458 	/* Don't do sense port on multifunction devices (for now at least) */
459 	if (mlx4_is_mfunc(dev))
460 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
461 
462 	if (mlx4_low_memory_profile()) {
463 		dev->caps.log_num_macs  = MLX4_MIN_LOG_NUM_MAC;
464 		dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
465 	} else {
466 		dev->caps.log_num_macs  = log_num_mac;
467 		dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
468 	}
469 
470 	for (i = 1; i <= dev->caps.num_ports; ++i) {
471 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
472 		if (dev->caps.supported_type[i]) {
473 			/* if only ETH is supported - assign ETH */
474 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
475 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
476 			/* if only IB is supported, assign IB */
477 			else if (dev->caps.supported_type[i] ==
478 				 MLX4_PORT_TYPE_IB)
479 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
480 			else {
481 				/* if IB and ETH are supported, we set the port
482 				 * type according to user selection of port type;
483 				 * if user selected none, take the FW hint */
484 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
485 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
486 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
487 				else
488 					dev->caps.port_type[i] = port_type_array[i - 1];
489 			}
490 		}
491 		/*
492 		 * Link sensing is allowed on the port if 3 conditions are true:
493 		 * 1. Both protocols are supported on the port.
494 		 * 2. Different types are supported on the port
495 		 * 3. FW declared that it supports link sensing
496 		 */
497 		mlx4_priv(dev)->sense.sense_allowed[i] =
498 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
499 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
500 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
501 
502 		/*
503 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
504 		 * and perform sense_port FW command to try and set the correct
505 		 * port type from beginning
506 		 */
507 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
508 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
509 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
510 			mlx4_SENSE_PORT(dev, i, &sensed_port);
511 			if (sensed_port != MLX4_PORT_TYPE_NONE)
512 				dev->caps.port_type[i] = sensed_port;
513 		} else {
514 			dev->caps.possible_type[i] = dev->caps.port_type[i];
515 		}
516 
517 		if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
518 			dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
519 			mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
520 				  i, 1 << dev->caps.log_num_macs);
521 		}
522 		if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
523 			dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
524 			mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
525 				  i, 1 << dev->caps.log_num_vlans);
526 		}
527 	}
528 
529 	if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
530 	    (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
531 	    (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
532 		mlx4_warn(dev,
533 			  "Granular QoS per VF not supported with IB/Eth configuration\n");
534 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
535 	}
536 
537 	dev->caps.max_counters = dev_cap->max_counters;
538 
539 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
540 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
541 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
542 		(1 << dev->caps.log_num_macs) *
543 		(1 << dev->caps.log_num_vlans) *
544 		dev->caps.num_ports;
545 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
546 
547 	if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
548 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
549 		dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
550 	else
551 		dev->caps.dmfs_high_rate_qpn_base =
552 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
553 
554 	if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
555 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
556 		dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
557 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
558 		dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
559 	} else {
560 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
561 		dev->caps.dmfs_high_rate_qpn_base =
562 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
563 		dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
564 	}
565 
566 	dev->caps.rl_caps = dev_cap->rl_caps;
567 
568 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
569 		dev->caps.dmfs_high_rate_qpn_range;
570 
571 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
572 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
573 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
574 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
575 
576 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
577 
578 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
579 		if (dev_cap->flags &
580 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
581 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
582 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
583 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
584 		}
585 
586 		if (dev_cap->flags2 &
587 		    (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
588 		     MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
589 			mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
590 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
591 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
592 		}
593 	}
594 
595 	if ((dev->caps.flags &
596 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
597 	    mlx4_is_master(dev))
598 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
599 
600 	if (!mlx4_is_slave(dev)) {
601 		mlx4_enable_cqe_eqe_stride(dev);
602 		dev->caps.alloc_res_qp_mask =
603 			(dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
604 			MLX4_RESERVE_A0_QP;
605 
606 		if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
607 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
608 			mlx4_warn(dev, "Old device ETS support detected\n");
609 			mlx4_warn(dev, "Consider upgrading device FW.\n");
610 			dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
611 		}
612 
613 	} else {
614 		dev->caps.alloc_res_qp_mask = 0;
615 	}
616 
617 	mlx4_enable_ignore_fcs(dev);
618 
619 	return 0;
620 }
621 
622 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
623 				       enum pci_bus_speed *speed,
624 				       enum pcie_link_width *width)
625 {
626 	u32 lnkcap1, lnkcap2;
627 	int err1, err2;
628 
629 #define  PCIE_MLW_CAP_SHIFT 4	/* start of MLW mask in link capabilities */
630 
631 	*speed = PCI_SPEED_UNKNOWN;
632 	*width = PCIE_LNK_WIDTH_UNKNOWN;
633 
634 	err1 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP,
635 					  &lnkcap1);
636 	err2 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP2,
637 					  &lnkcap2);
638 	if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
639 		if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
640 			*speed = PCIE_SPEED_8_0GT;
641 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
642 			*speed = PCIE_SPEED_5_0GT;
643 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
644 			*speed = PCIE_SPEED_2_5GT;
645 	}
646 	if (!err1) {
647 		*width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
648 		if (!lnkcap2) { /* pre-r3.0 */
649 			if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
650 				*speed = PCIE_SPEED_5_0GT;
651 			else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
652 				*speed = PCIE_SPEED_2_5GT;
653 		}
654 	}
655 
656 	if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
657 		return err1 ? err1 :
658 			err2 ? err2 : -EINVAL;
659 	}
660 	return 0;
661 }
662 
663 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
664 {
665 	enum pcie_link_width width, width_cap;
666 	enum pci_bus_speed speed, speed_cap;
667 	int err;
668 
669 #define PCIE_SPEED_STR(speed) \
670 	(speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
671 	 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
672 	 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
673 	 "Unknown")
674 
675 	err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
676 	if (err) {
677 		mlx4_warn(dev,
678 			  "Unable to determine PCIe device BW capabilities\n");
679 		return;
680 	}
681 
682 	err = pcie_get_minimum_link(dev->persist->pdev, &speed, &width);
683 	if (err || speed == PCI_SPEED_UNKNOWN ||
684 	    width == PCIE_LNK_WIDTH_UNKNOWN) {
685 		mlx4_warn(dev,
686 			  "Unable to determine PCI device chain minimum BW\n");
687 		return;
688 	}
689 
690 	if (width != width_cap || speed != speed_cap)
691 		mlx4_warn(dev,
692 			  "PCIe BW is different than device's capability\n");
693 
694 	mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
695 		  PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
696 	mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
697 		  width, width_cap);
698 	return;
699 }
700 
701 /*The function checks if there are live vf, return the num of them*/
702 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
703 {
704 	struct mlx4_priv *priv = mlx4_priv(dev);
705 	struct mlx4_slave_state *s_state;
706 	int i;
707 	int ret = 0;
708 
709 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
710 		s_state = &priv->mfunc.master.slave_state[i];
711 		if (s_state->active && s_state->last_cmd !=
712 		    MLX4_COMM_CMD_RESET) {
713 			mlx4_warn(dev, "%s: slave: %d is still active\n",
714 				  __func__, i);
715 			ret++;
716 		}
717 	}
718 	return ret;
719 }
720 
721 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
722 {
723 	u32 qk = MLX4_RESERVED_QKEY_BASE;
724 
725 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
726 	    qpn < dev->phys_caps.base_proxy_sqpn)
727 		return -EINVAL;
728 
729 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
730 		/* tunnel qp */
731 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
732 	else
733 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
734 	*qkey = qk;
735 	return 0;
736 }
737 EXPORT_SYMBOL(mlx4_get_parav_qkey);
738 
739 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
740 {
741 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
742 
743 	if (!mlx4_is_master(dev))
744 		return;
745 
746 	priv->virt2phys_pkey[slave][port - 1][i] = val;
747 }
748 EXPORT_SYMBOL(mlx4_sync_pkey_table);
749 
750 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
751 {
752 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
753 
754 	if (!mlx4_is_master(dev))
755 		return;
756 
757 	priv->slave_node_guids[slave] = guid;
758 }
759 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
760 
761 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
762 {
763 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
764 
765 	if (!mlx4_is_master(dev))
766 		return 0;
767 
768 	return priv->slave_node_guids[slave];
769 }
770 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
771 
772 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
773 {
774 	struct mlx4_priv *priv = mlx4_priv(dev);
775 	struct mlx4_slave_state *s_slave;
776 
777 	if (!mlx4_is_master(dev))
778 		return 0;
779 
780 	s_slave = &priv->mfunc.master.slave_state[slave];
781 	return !!s_slave->active;
782 }
783 EXPORT_SYMBOL(mlx4_is_slave_active);
784 
785 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
786 				       struct mlx4_dev_cap *dev_cap,
787 				       struct mlx4_init_hca_param *hca_param)
788 {
789 	dev->caps.steering_mode = hca_param->steering_mode;
790 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
791 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
792 		dev->caps.fs_log_max_ucast_qp_range_size =
793 			dev_cap->fs_log_max_ucast_qp_range_size;
794 	} else
795 		dev->caps.num_qp_per_mgm =
796 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
797 
798 	mlx4_dbg(dev, "Steering mode is: %s\n",
799 		 mlx4_steering_mode_str(dev->caps.steering_mode));
800 }
801 
802 static int mlx4_slave_cap(struct mlx4_dev *dev)
803 {
804 	int			   err;
805 	u32			   page_size;
806 	struct mlx4_dev_cap	   dev_cap;
807 	struct mlx4_func_cap	   func_cap;
808 	struct mlx4_init_hca_param hca_param;
809 	u8			   i;
810 
811 	memset(&hca_param, 0, sizeof(hca_param));
812 	err = mlx4_QUERY_HCA(dev, &hca_param);
813 	if (err) {
814 		mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
815 		return err;
816 	}
817 
818 	/* fail if the hca has an unknown global capability
819 	 * at this time global_caps should be always zeroed
820 	 */
821 	if (hca_param.global_caps) {
822 		mlx4_err(dev, "Unknown hca global capabilities\n");
823 		return -ENOSYS;
824 	}
825 
826 	mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
827 
828 	dev->caps.hca_core_clock = hca_param.hca_core_clock;
829 
830 	memset(&dev_cap, 0, sizeof(dev_cap));
831 	dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
832 	err = mlx4_dev_cap(dev, &dev_cap);
833 	if (err) {
834 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
835 		return err;
836 	}
837 
838 	err = mlx4_QUERY_FW(dev);
839 	if (err)
840 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
841 
842 	page_size = ~dev->caps.page_size_cap + 1;
843 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
844 	if (page_size > PAGE_SIZE) {
845 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
846 			 page_size, PAGE_SIZE);
847 		return -ENODEV;
848 	}
849 
850 	/* Set uar_page_shift for VF */
851 	dev->uar_page_shift = hca_param.uar_page_sz + 12;
852 
853 	/* Make sure the master uar page size is valid */
854 	if (dev->uar_page_shift > PAGE_SHIFT) {
855 		mlx4_err(dev,
856 			 "Invalid configuration: uar page size is larger than system page size\n");
857 		return  -ENODEV;
858 	}
859 
860 	/* Set reserved_uars based on the uar_page_shift */
861 	mlx4_set_num_reserved_uars(dev, &dev_cap);
862 
863 	/* Although uar page size in FW differs from system page size,
864 	 * upper software layers (mlx4_ib, mlx4_en and part of mlx4_core)
865 	 * still works with assumption that uar page size == system page size
866 	 */
867 	dev->caps.uar_page_size = PAGE_SIZE;
868 
869 	memset(&func_cap, 0, sizeof(func_cap));
870 	err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
871 	if (err) {
872 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
873 			 err);
874 		return err;
875 	}
876 
877 	if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
878 	    PF_CONTEXT_BEHAVIOUR_MASK) {
879 		mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
880 			 func_cap.pf_context_behaviour, PF_CONTEXT_BEHAVIOUR_MASK);
881 		return -ENOSYS;
882 	}
883 
884 	dev->caps.num_ports		= func_cap.num_ports;
885 	dev->quotas.qp			= func_cap.qp_quota;
886 	dev->quotas.srq			= func_cap.srq_quota;
887 	dev->quotas.cq			= func_cap.cq_quota;
888 	dev->quotas.mpt			= func_cap.mpt_quota;
889 	dev->quotas.mtt			= func_cap.mtt_quota;
890 	dev->caps.num_qps		= 1 << hca_param.log_num_qps;
891 	dev->caps.num_srqs		= 1 << hca_param.log_num_srqs;
892 	dev->caps.num_cqs		= 1 << hca_param.log_num_cqs;
893 	dev->caps.num_mpts		= 1 << hca_param.log_mpt_sz;
894 	dev->caps.num_eqs		= func_cap.max_eq;
895 	dev->caps.reserved_eqs		= func_cap.reserved_eq;
896 	dev->caps.reserved_lkey		= func_cap.reserved_lkey;
897 	dev->caps.num_pds               = MLX4_NUM_PDS;
898 	dev->caps.num_mgms              = 0;
899 	dev->caps.num_amgms             = 0;
900 
901 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
902 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
903 			 dev->caps.num_ports, MLX4_MAX_PORTS);
904 		return -ENODEV;
905 	}
906 
907 	mlx4_replace_zero_macs(dev);
908 
909 	dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
910 	dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
911 	dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
912 	dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
913 	dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
914 
915 	if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
916 	    !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
917 	    !dev->caps.qp0_qkey) {
918 		err = -ENOMEM;
919 		goto err_mem;
920 	}
921 
922 	for (i = 1; i <= dev->caps.num_ports; ++i) {
923 		err = mlx4_QUERY_FUNC_CAP(dev, i, &func_cap);
924 		if (err) {
925 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
926 				 i, err);
927 			goto err_mem;
928 		}
929 		dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
930 		dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
931 		dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
932 		dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
933 		dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
934 		dev->caps.port_mask[i] = dev->caps.port_type[i];
935 		dev->caps.phys_port_id[i] = func_cap.phys_port_id;
936 		err = mlx4_get_slave_pkey_gid_tbl_len(dev, i,
937 						      &dev->caps.gid_table_len[i],
938 						      &dev->caps.pkey_table_len[i]);
939 		if (err)
940 			goto err_mem;
941 	}
942 
943 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
944 				       dev->caps.reserved_uars) >
945 				       pci_resource_len(dev->persist->pdev,
946 							2)) {
947 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
948 			 dev->caps.uar_page_size * dev->caps.num_uars,
949 			 (unsigned long long)
950 			 pci_resource_len(dev->persist->pdev, 2));
951 		err = -ENOMEM;
952 		goto err_mem;
953 	}
954 
955 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
956 		dev->caps.eqe_size   = 64;
957 		dev->caps.eqe_factor = 1;
958 	} else {
959 		dev->caps.eqe_size   = 32;
960 		dev->caps.eqe_factor = 0;
961 	}
962 
963 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
964 		dev->caps.cqe_size   = 64;
965 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
966 	} else {
967 		dev->caps.cqe_size   = 32;
968 	}
969 
970 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
971 		dev->caps.eqe_size = hca_param.eqe_size;
972 		dev->caps.eqe_factor = 0;
973 	}
974 
975 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
976 		dev->caps.cqe_size = hca_param.cqe_size;
977 		/* User still need to know when CQE > 32B */
978 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
979 	}
980 
981 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
982 	mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
983 
984 	slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
985 	mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
986 		 hca_param.rss_ip_frags ? "on" : "off");
987 
988 	if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
989 	    dev->caps.bf_reg_size)
990 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
991 
992 	if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
993 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
994 
995 	return 0;
996 
997 err_mem:
998 	kfree(dev->caps.qp0_qkey);
999 	kfree(dev->caps.qp0_tunnel);
1000 	kfree(dev->caps.qp0_proxy);
1001 	kfree(dev->caps.qp1_tunnel);
1002 	kfree(dev->caps.qp1_proxy);
1003 	dev->caps.qp0_qkey = NULL;
1004 	dev->caps.qp0_tunnel = NULL;
1005 	dev->caps.qp0_proxy = NULL;
1006 	dev->caps.qp1_tunnel = NULL;
1007 	dev->caps.qp1_proxy = NULL;
1008 
1009 	return err;
1010 }
1011 
1012 static void mlx4_request_modules(struct mlx4_dev *dev)
1013 {
1014 	int port;
1015 	int has_ib_port = false;
1016 	int has_eth_port = false;
1017 #define EN_DRV_NAME	"mlx4_en"
1018 #define IB_DRV_NAME	"mlx4_ib"
1019 
1020 	for (port = 1; port <= dev->caps.num_ports; port++) {
1021 		if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
1022 			has_ib_port = true;
1023 		else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
1024 			has_eth_port = true;
1025 	}
1026 
1027 	if (has_eth_port)
1028 		request_module_nowait(EN_DRV_NAME);
1029 	if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
1030 		request_module_nowait(IB_DRV_NAME);
1031 }
1032 
1033 /*
1034  * Change the port configuration of the device.
1035  * Every user of this function must hold the port mutex.
1036  */
1037 int mlx4_change_port_types(struct mlx4_dev *dev,
1038 			   enum mlx4_port_type *port_types)
1039 {
1040 	int err = 0;
1041 	int change = 0;
1042 	int port;
1043 
1044 	for (port = 0; port <  dev->caps.num_ports; port++) {
1045 		/* Change the port type only if the new type is different
1046 		 * from the current, and not set to Auto */
1047 		if (port_types[port] != dev->caps.port_type[port + 1])
1048 			change = 1;
1049 	}
1050 	if (change) {
1051 		mlx4_unregister_device(dev);
1052 		for (port = 1; port <= dev->caps.num_ports; port++) {
1053 			mlx4_CLOSE_PORT(dev, port);
1054 			dev->caps.port_type[port] = port_types[port - 1];
1055 			err = mlx4_SET_PORT(dev, port, -1);
1056 			if (err) {
1057 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1058 					 port);
1059 				goto out;
1060 			}
1061 		}
1062 		mlx4_set_port_mask(dev);
1063 		err = mlx4_register_device(dev);
1064 		if (err) {
1065 			mlx4_err(dev, "Failed to register device\n");
1066 			goto out;
1067 		}
1068 		mlx4_request_modules(dev);
1069 	}
1070 
1071 out:
1072 	return err;
1073 }
1074 
1075 static ssize_t show_port_type(struct device *dev,
1076 			      struct device_attribute *attr,
1077 			      char *buf)
1078 {
1079 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1080 						   port_attr);
1081 	struct mlx4_dev *mdev = info->dev;
1082 	char type[8];
1083 
1084 	sprintf(type, "%s",
1085 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1086 		"ib" : "eth");
1087 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1088 		sprintf(buf, "auto (%s)\n", type);
1089 	else
1090 		sprintf(buf, "%s\n", type);
1091 
1092 	return strlen(buf);
1093 }
1094 
1095 static int __set_port_type(struct mlx4_port_info *info,
1096 			   enum mlx4_port_type port_type)
1097 {
1098 	struct mlx4_dev *mdev = info->dev;
1099 	struct mlx4_priv *priv = mlx4_priv(mdev);
1100 	enum mlx4_port_type types[MLX4_MAX_PORTS];
1101 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1102 	int i;
1103 	int err = 0;
1104 
1105 	mlx4_stop_sense(mdev);
1106 	mutex_lock(&priv->port_mutex);
1107 	info->tmp_type = port_type;
1108 
1109 	/* Possible type is always the one that was delivered */
1110 	mdev->caps.possible_type[info->port] = info->tmp_type;
1111 
1112 	for (i = 0; i < mdev->caps.num_ports; i++) {
1113 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1114 					mdev->caps.possible_type[i+1];
1115 		if (types[i] == MLX4_PORT_TYPE_AUTO)
1116 			types[i] = mdev->caps.port_type[i+1];
1117 	}
1118 
1119 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1120 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1121 		for (i = 1; i <= mdev->caps.num_ports; i++) {
1122 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1123 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1124 				err = -EINVAL;
1125 			}
1126 		}
1127 	}
1128 	if (err) {
1129 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1130 		goto out;
1131 	}
1132 
1133 	mlx4_do_sense_ports(mdev, new_types, types);
1134 
1135 	err = mlx4_check_port_params(mdev, new_types);
1136 	if (err)
1137 		goto out;
1138 
1139 	/* We are about to apply the changes after the configuration
1140 	 * was verified, no need to remember the temporary types
1141 	 * any more */
1142 	for (i = 0; i < mdev->caps.num_ports; i++)
1143 		priv->port[i + 1].tmp_type = 0;
1144 
1145 	err = mlx4_change_port_types(mdev, new_types);
1146 
1147 out:
1148 	mlx4_start_sense(mdev);
1149 	mutex_unlock(&priv->port_mutex);
1150 
1151 	return err;
1152 }
1153 
1154 static ssize_t set_port_type(struct device *dev,
1155 			     struct device_attribute *attr,
1156 			     const char *buf, size_t count)
1157 {
1158 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1159 						   port_attr);
1160 	struct mlx4_dev *mdev = info->dev;
1161 	enum mlx4_port_type port_type;
1162 	static DEFINE_MUTEX(set_port_type_mutex);
1163 	int err;
1164 
1165 	mutex_lock(&set_port_type_mutex);
1166 
1167 	if (!strcmp(buf, "ib\n")) {
1168 		port_type = MLX4_PORT_TYPE_IB;
1169 	} else if (!strcmp(buf, "eth\n")) {
1170 		port_type = MLX4_PORT_TYPE_ETH;
1171 	} else if (!strcmp(buf, "auto\n")) {
1172 		port_type = MLX4_PORT_TYPE_AUTO;
1173 	} else {
1174 		mlx4_err(mdev, "%s is not supported port type\n", buf);
1175 		err = -EINVAL;
1176 		goto err_out;
1177 	}
1178 
1179 	err = __set_port_type(info, port_type);
1180 
1181 err_out:
1182 	mutex_unlock(&set_port_type_mutex);
1183 
1184 	return err ? err : count;
1185 }
1186 
1187 enum ibta_mtu {
1188 	IB_MTU_256  = 1,
1189 	IB_MTU_512  = 2,
1190 	IB_MTU_1024 = 3,
1191 	IB_MTU_2048 = 4,
1192 	IB_MTU_4096 = 5
1193 };
1194 
1195 static inline int int_to_ibta_mtu(int mtu)
1196 {
1197 	switch (mtu) {
1198 	case 256:  return IB_MTU_256;
1199 	case 512:  return IB_MTU_512;
1200 	case 1024: return IB_MTU_1024;
1201 	case 2048: return IB_MTU_2048;
1202 	case 4096: return IB_MTU_4096;
1203 	default: return -1;
1204 	}
1205 }
1206 
1207 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1208 {
1209 	switch (mtu) {
1210 	case IB_MTU_256:  return  256;
1211 	case IB_MTU_512:  return  512;
1212 	case IB_MTU_1024: return 1024;
1213 	case IB_MTU_2048: return 2048;
1214 	case IB_MTU_4096: return 4096;
1215 	default: return -1;
1216 	}
1217 }
1218 
1219 static ssize_t show_port_ib_mtu(struct device *dev,
1220 			     struct device_attribute *attr,
1221 			     char *buf)
1222 {
1223 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1224 						   port_mtu_attr);
1225 	struct mlx4_dev *mdev = info->dev;
1226 
1227 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1228 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1229 
1230 	sprintf(buf, "%d\n",
1231 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1232 	return strlen(buf);
1233 }
1234 
1235 static ssize_t set_port_ib_mtu(struct device *dev,
1236 			     struct device_attribute *attr,
1237 			     const char *buf, size_t count)
1238 {
1239 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1240 						   port_mtu_attr);
1241 	struct mlx4_dev *mdev = info->dev;
1242 	struct mlx4_priv *priv = mlx4_priv(mdev);
1243 	int err, port, mtu, ibta_mtu = -1;
1244 
1245 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1246 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1247 		return -EINVAL;
1248 	}
1249 
1250 	err = kstrtoint(buf, 0, &mtu);
1251 	if (!err)
1252 		ibta_mtu = int_to_ibta_mtu(mtu);
1253 
1254 	if (err || ibta_mtu < 0) {
1255 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1256 		return -EINVAL;
1257 	}
1258 
1259 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1260 
1261 	mlx4_stop_sense(mdev);
1262 	mutex_lock(&priv->port_mutex);
1263 	mlx4_unregister_device(mdev);
1264 	for (port = 1; port <= mdev->caps.num_ports; port++) {
1265 		mlx4_CLOSE_PORT(mdev, port);
1266 		err = mlx4_SET_PORT(mdev, port, -1);
1267 		if (err) {
1268 			mlx4_err(mdev, "Failed to set port %d, aborting\n",
1269 				 port);
1270 			goto err_set_port;
1271 		}
1272 	}
1273 	err = mlx4_register_device(mdev);
1274 err_set_port:
1275 	mutex_unlock(&priv->port_mutex);
1276 	mlx4_start_sense(mdev);
1277 	return err ? err : count;
1278 }
1279 
1280 /* bond for multi-function device */
1281 #define MAX_MF_BOND_ALLOWED_SLAVES 63
1282 static int mlx4_mf_bond(struct mlx4_dev *dev)
1283 {
1284 	int err = 0;
1285 	int nvfs;
1286 	struct mlx4_slaves_pport slaves_port1;
1287 	struct mlx4_slaves_pport slaves_port2;
1288 	DECLARE_BITMAP(slaves_port_1_2, MLX4_MFUNC_MAX);
1289 
1290 	slaves_port1 = mlx4_phys_to_slaves_pport(dev, 1);
1291 	slaves_port2 = mlx4_phys_to_slaves_pport(dev, 2);
1292 	bitmap_and(slaves_port_1_2,
1293 		   slaves_port1.slaves, slaves_port2.slaves,
1294 		   dev->persist->num_vfs + 1);
1295 
1296 	/* only single port vfs are allowed */
1297 	if (bitmap_weight(slaves_port_1_2, dev->persist->num_vfs + 1) > 1) {
1298 		mlx4_warn(dev, "HA mode unsupported for dual ported VFs\n");
1299 		return -EINVAL;
1300 	}
1301 
1302 	/* number of virtual functions is number of total functions minus one
1303 	 * physical function for each port.
1304 	 */
1305 	nvfs = bitmap_weight(slaves_port1.slaves, dev->persist->num_vfs + 1) +
1306 		bitmap_weight(slaves_port2.slaves, dev->persist->num_vfs + 1) - 2;
1307 
1308 	/* limit on maximum allowed VFs */
1309 	if (nvfs > MAX_MF_BOND_ALLOWED_SLAVES) {
1310 		mlx4_warn(dev, "HA mode is not supported for %d VFs (max %d are allowed)\n",
1311 			  nvfs, MAX_MF_BOND_ALLOWED_SLAVES);
1312 		return -EINVAL;
1313 	}
1314 
1315 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) {
1316 		mlx4_warn(dev, "HA mode unsupported for NON DMFS steering\n");
1317 		return -EINVAL;
1318 	}
1319 
1320 	err = mlx4_bond_mac_table(dev);
1321 	if (err)
1322 		return err;
1323 	err = mlx4_bond_vlan_table(dev);
1324 	if (err)
1325 		goto err1;
1326 	err = mlx4_bond_fs_rules(dev);
1327 	if (err)
1328 		goto err2;
1329 
1330 	return 0;
1331 err2:
1332 	(void)mlx4_unbond_vlan_table(dev);
1333 err1:
1334 	(void)mlx4_unbond_mac_table(dev);
1335 	return err;
1336 }
1337 
1338 static int mlx4_mf_unbond(struct mlx4_dev *dev)
1339 {
1340 	int ret, ret1;
1341 
1342 	ret = mlx4_unbond_fs_rules(dev);
1343 	if (ret)
1344 		mlx4_warn(dev, "multifunction unbond for flow rules failedi (%d)\n", ret);
1345 	ret1 = mlx4_unbond_mac_table(dev);
1346 	if (ret1) {
1347 		mlx4_warn(dev, "multifunction unbond for MAC table failed (%d)\n", ret1);
1348 		ret = ret1;
1349 	}
1350 	ret1 = mlx4_unbond_vlan_table(dev);
1351 	if (ret1) {
1352 		mlx4_warn(dev, "multifunction unbond for VLAN table failed (%d)\n", ret1);
1353 		ret = ret1;
1354 	}
1355 	return ret;
1356 }
1357 
1358 int mlx4_bond(struct mlx4_dev *dev)
1359 {
1360 	int ret = 0;
1361 	struct mlx4_priv *priv = mlx4_priv(dev);
1362 
1363 	mutex_lock(&priv->bond_mutex);
1364 
1365 	if (!mlx4_is_bonded(dev)) {
1366 		ret = mlx4_do_bond(dev, true);
1367 		if (ret)
1368 			mlx4_err(dev, "Failed to bond device: %d\n", ret);
1369 		if (!ret && mlx4_is_master(dev)) {
1370 			ret = mlx4_mf_bond(dev);
1371 			if (ret) {
1372 				mlx4_err(dev, "bond for multifunction failed\n");
1373 				mlx4_do_bond(dev, false);
1374 			}
1375 		}
1376 	}
1377 
1378 	mutex_unlock(&priv->bond_mutex);
1379 	if (!ret)
1380 		mlx4_dbg(dev, "Device is bonded\n");
1381 
1382 	return ret;
1383 }
1384 EXPORT_SYMBOL_GPL(mlx4_bond);
1385 
1386 int mlx4_unbond(struct mlx4_dev *dev)
1387 {
1388 	int ret = 0;
1389 	struct mlx4_priv *priv = mlx4_priv(dev);
1390 
1391 	mutex_lock(&priv->bond_mutex);
1392 
1393 	if (mlx4_is_bonded(dev)) {
1394 		int ret2 = 0;
1395 
1396 		ret = mlx4_do_bond(dev, false);
1397 		if (ret)
1398 			mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1399 		if (mlx4_is_master(dev))
1400 			ret2 = mlx4_mf_unbond(dev);
1401 		if (ret2) {
1402 			mlx4_warn(dev, "Failed to unbond device for multifunction (%d)\n", ret2);
1403 			ret = ret2;
1404 		}
1405 	}
1406 
1407 	mutex_unlock(&priv->bond_mutex);
1408 	if (!ret)
1409 		mlx4_dbg(dev, "Device is unbonded\n");
1410 
1411 	return ret;
1412 }
1413 EXPORT_SYMBOL_GPL(mlx4_unbond);
1414 
1415 
1416 int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1417 {
1418 	u8 port1 = v2p->port1;
1419 	u8 port2 = v2p->port2;
1420 	struct mlx4_priv *priv = mlx4_priv(dev);
1421 	int err;
1422 
1423 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1424 		return -ENOTSUPP;
1425 
1426 	mutex_lock(&priv->bond_mutex);
1427 
1428 	/* zero means keep current mapping for this port */
1429 	if (port1 == 0)
1430 		port1 = priv->v2p.port1;
1431 	if (port2 == 0)
1432 		port2 = priv->v2p.port2;
1433 
1434 	if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1435 	    (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1436 	    (port1 == 2 && port2 == 1)) {
1437 		/* besides boundary checks cross mapping makes
1438 		 * no sense and therefore not allowed */
1439 		err = -EINVAL;
1440 	} else if ((port1 == priv->v2p.port1) &&
1441 		 (port2 == priv->v2p.port2)) {
1442 		err = 0;
1443 	} else {
1444 		err = mlx4_virt2phy_port_map(dev, port1, port2);
1445 		if (!err) {
1446 			mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1447 				 port1, port2);
1448 			priv->v2p.port1 = port1;
1449 			priv->v2p.port2 = port2;
1450 		} else {
1451 			mlx4_err(dev, "Failed to change port mape: %d\n", err);
1452 		}
1453 	}
1454 
1455 	mutex_unlock(&priv->bond_mutex);
1456 	return err;
1457 }
1458 EXPORT_SYMBOL_GPL(mlx4_port_map_set);
1459 
1460 static int mlx4_load_fw(struct mlx4_dev *dev)
1461 {
1462 	struct mlx4_priv *priv = mlx4_priv(dev);
1463 	int err;
1464 
1465 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1466 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
1467 	if (!priv->fw.fw_icm) {
1468 		mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1469 		return -ENOMEM;
1470 	}
1471 
1472 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1473 	if (err) {
1474 		mlx4_err(dev, "MAP_FA command failed, aborting\n");
1475 		goto err_free;
1476 	}
1477 
1478 	err = mlx4_RUN_FW(dev);
1479 	if (err) {
1480 		mlx4_err(dev, "RUN_FW command failed, aborting\n");
1481 		goto err_unmap_fa;
1482 	}
1483 
1484 	return 0;
1485 
1486 err_unmap_fa:
1487 	mlx4_UNMAP_FA(dev);
1488 
1489 err_free:
1490 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1491 	return err;
1492 }
1493 
1494 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1495 				int cmpt_entry_sz)
1496 {
1497 	struct mlx4_priv *priv = mlx4_priv(dev);
1498 	int err;
1499 	int num_eqs;
1500 
1501 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1502 				  cmpt_base +
1503 				  ((u64) (MLX4_CMPT_TYPE_QP *
1504 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1505 				  cmpt_entry_sz, dev->caps.num_qps,
1506 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1507 				  0, 0);
1508 	if (err)
1509 		goto err;
1510 
1511 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1512 				  cmpt_base +
1513 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1514 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1515 				  cmpt_entry_sz, dev->caps.num_srqs,
1516 				  dev->caps.reserved_srqs, 0, 0);
1517 	if (err)
1518 		goto err_qp;
1519 
1520 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1521 				  cmpt_base +
1522 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1523 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1524 				  cmpt_entry_sz, dev->caps.num_cqs,
1525 				  dev->caps.reserved_cqs, 0, 0);
1526 	if (err)
1527 		goto err_srq;
1528 
1529 	num_eqs = dev->phys_caps.num_phys_eqs;
1530 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1531 				  cmpt_base +
1532 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1533 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1534 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1535 	if (err)
1536 		goto err_cq;
1537 
1538 	return 0;
1539 
1540 err_cq:
1541 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1542 
1543 err_srq:
1544 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1545 
1546 err_qp:
1547 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1548 
1549 err:
1550 	return err;
1551 }
1552 
1553 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1554 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1555 {
1556 	struct mlx4_priv *priv = mlx4_priv(dev);
1557 	u64 aux_pages;
1558 	int num_eqs;
1559 	int err;
1560 
1561 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1562 	if (err) {
1563 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1564 		return err;
1565 	}
1566 
1567 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1568 		 (unsigned long long) icm_size >> 10,
1569 		 (unsigned long long) aux_pages << 2);
1570 
1571 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1572 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1573 	if (!priv->fw.aux_icm) {
1574 		mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1575 		return -ENOMEM;
1576 	}
1577 
1578 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1579 	if (err) {
1580 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1581 		goto err_free_aux;
1582 	}
1583 
1584 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1585 	if (err) {
1586 		mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1587 		goto err_unmap_aux;
1588 	}
1589 
1590 
1591 	num_eqs = dev->phys_caps.num_phys_eqs;
1592 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1593 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1594 				  num_eqs, num_eqs, 0, 0);
1595 	if (err) {
1596 		mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1597 		goto err_unmap_cmpt;
1598 	}
1599 
1600 	/*
1601 	 * Reserved MTT entries must be aligned up to a cacheline
1602 	 * boundary, since the FW will write to them, while the driver
1603 	 * writes to all other MTT entries. (The variable
1604 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1605 	 * size, not the raw entry size)
1606 	 */
1607 	dev->caps.reserved_mtts =
1608 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1609 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1610 
1611 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1612 				  init_hca->mtt_base,
1613 				  dev->caps.mtt_entry_sz,
1614 				  dev->caps.num_mtts,
1615 				  dev->caps.reserved_mtts, 1, 0);
1616 	if (err) {
1617 		mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1618 		goto err_unmap_eq;
1619 	}
1620 
1621 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1622 				  init_hca->dmpt_base,
1623 				  dev_cap->dmpt_entry_sz,
1624 				  dev->caps.num_mpts,
1625 				  dev->caps.reserved_mrws, 1, 1);
1626 	if (err) {
1627 		mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1628 		goto err_unmap_mtt;
1629 	}
1630 
1631 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1632 				  init_hca->qpc_base,
1633 				  dev_cap->qpc_entry_sz,
1634 				  dev->caps.num_qps,
1635 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1636 				  0, 0);
1637 	if (err) {
1638 		mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1639 		goto err_unmap_dmpt;
1640 	}
1641 
1642 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1643 				  init_hca->auxc_base,
1644 				  dev_cap->aux_entry_sz,
1645 				  dev->caps.num_qps,
1646 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1647 				  0, 0);
1648 	if (err) {
1649 		mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1650 		goto err_unmap_qp;
1651 	}
1652 
1653 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1654 				  init_hca->altc_base,
1655 				  dev_cap->altc_entry_sz,
1656 				  dev->caps.num_qps,
1657 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1658 				  0, 0);
1659 	if (err) {
1660 		mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1661 		goto err_unmap_auxc;
1662 	}
1663 
1664 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1665 				  init_hca->rdmarc_base,
1666 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1667 				  dev->caps.num_qps,
1668 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1669 				  0, 0);
1670 	if (err) {
1671 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1672 		goto err_unmap_altc;
1673 	}
1674 
1675 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1676 				  init_hca->cqc_base,
1677 				  dev_cap->cqc_entry_sz,
1678 				  dev->caps.num_cqs,
1679 				  dev->caps.reserved_cqs, 0, 0);
1680 	if (err) {
1681 		mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1682 		goto err_unmap_rdmarc;
1683 	}
1684 
1685 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1686 				  init_hca->srqc_base,
1687 				  dev_cap->srq_entry_sz,
1688 				  dev->caps.num_srqs,
1689 				  dev->caps.reserved_srqs, 0, 0);
1690 	if (err) {
1691 		mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1692 		goto err_unmap_cq;
1693 	}
1694 
1695 	/*
1696 	 * For flow steering device managed mode it is required to use
1697 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1698 	 * required, but for simplicity just map the whole multicast
1699 	 * group table now.  The table isn't very big and it's a lot
1700 	 * easier than trying to track ref counts.
1701 	 */
1702 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1703 				  init_hca->mc_base,
1704 				  mlx4_get_mgm_entry_size(dev),
1705 				  dev->caps.num_mgms + dev->caps.num_amgms,
1706 				  dev->caps.num_mgms + dev->caps.num_amgms,
1707 				  0, 0);
1708 	if (err) {
1709 		mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1710 		goto err_unmap_srq;
1711 	}
1712 
1713 	return 0;
1714 
1715 err_unmap_srq:
1716 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1717 
1718 err_unmap_cq:
1719 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1720 
1721 err_unmap_rdmarc:
1722 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1723 
1724 err_unmap_altc:
1725 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1726 
1727 err_unmap_auxc:
1728 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1729 
1730 err_unmap_qp:
1731 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1732 
1733 err_unmap_dmpt:
1734 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1735 
1736 err_unmap_mtt:
1737 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1738 
1739 err_unmap_eq:
1740 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1741 
1742 err_unmap_cmpt:
1743 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1744 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1745 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1746 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1747 
1748 err_unmap_aux:
1749 	mlx4_UNMAP_ICM_AUX(dev);
1750 
1751 err_free_aux:
1752 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1753 
1754 	return err;
1755 }
1756 
1757 static void mlx4_free_icms(struct mlx4_dev *dev)
1758 {
1759 	struct mlx4_priv *priv = mlx4_priv(dev);
1760 
1761 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1762 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1763 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1764 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1765 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1766 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1767 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1768 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1769 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1770 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1771 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1772 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1773 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1774 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1775 
1776 	mlx4_UNMAP_ICM_AUX(dev);
1777 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1778 }
1779 
1780 static void mlx4_slave_exit(struct mlx4_dev *dev)
1781 {
1782 	struct mlx4_priv *priv = mlx4_priv(dev);
1783 
1784 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1785 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1786 			  MLX4_COMM_TIME))
1787 		mlx4_warn(dev, "Failed to close slave function\n");
1788 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1789 }
1790 
1791 static int map_bf_area(struct mlx4_dev *dev)
1792 {
1793 	struct mlx4_priv *priv = mlx4_priv(dev);
1794 	resource_size_t bf_start;
1795 	resource_size_t bf_len;
1796 	int err = 0;
1797 
1798 	if (!dev->caps.bf_reg_size)
1799 		return -ENXIO;
1800 
1801 	bf_start = pci_resource_start(dev->persist->pdev, 2) +
1802 			(dev->caps.num_uars << PAGE_SHIFT);
1803 	bf_len = pci_resource_len(dev->persist->pdev, 2) -
1804 			(dev->caps.num_uars << PAGE_SHIFT);
1805 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1806 	if (!priv->bf_mapping)
1807 		err = -ENOMEM;
1808 
1809 	return err;
1810 }
1811 
1812 static void unmap_bf_area(struct mlx4_dev *dev)
1813 {
1814 	if (mlx4_priv(dev)->bf_mapping)
1815 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1816 }
1817 
1818 cycle_t mlx4_read_clock(struct mlx4_dev *dev)
1819 {
1820 	u32 clockhi, clocklo, clockhi1;
1821 	cycle_t cycles;
1822 	int i;
1823 	struct mlx4_priv *priv = mlx4_priv(dev);
1824 
1825 	for (i = 0; i < 10; i++) {
1826 		clockhi = swab32(readl(priv->clock_mapping));
1827 		clocklo = swab32(readl(priv->clock_mapping + 4));
1828 		clockhi1 = swab32(readl(priv->clock_mapping));
1829 		if (clockhi == clockhi1)
1830 			break;
1831 	}
1832 
1833 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1834 
1835 	return cycles;
1836 }
1837 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1838 
1839 
1840 static int map_internal_clock(struct mlx4_dev *dev)
1841 {
1842 	struct mlx4_priv *priv = mlx4_priv(dev);
1843 
1844 	priv->clock_mapping =
1845 		ioremap(pci_resource_start(dev->persist->pdev,
1846 					   priv->fw.clock_bar) +
1847 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1848 
1849 	if (!priv->clock_mapping)
1850 		return -ENOMEM;
1851 
1852 	return 0;
1853 }
1854 
1855 int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1856 				   struct mlx4_clock_params *params)
1857 {
1858 	struct mlx4_priv *priv = mlx4_priv(dev);
1859 
1860 	if (mlx4_is_slave(dev))
1861 		return -ENOTSUPP;
1862 
1863 	if (!params)
1864 		return -EINVAL;
1865 
1866 	params->bar = priv->fw.clock_bar;
1867 	params->offset = priv->fw.clock_offset;
1868 	params->size = MLX4_CLOCK_SIZE;
1869 
1870 	return 0;
1871 }
1872 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1873 
1874 static void unmap_internal_clock(struct mlx4_dev *dev)
1875 {
1876 	struct mlx4_priv *priv = mlx4_priv(dev);
1877 
1878 	if (priv->clock_mapping)
1879 		iounmap(priv->clock_mapping);
1880 }
1881 
1882 static void mlx4_close_hca(struct mlx4_dev *dev)
1883 {
1884 	unmap_internal_clock(dev);
1885 	unmap_bf_area(dev);
1886 	if (mlx4_is_slave(dev))
1887 		mlx4_slave_exit(dev);
1888 	else {
1889 		mlx4_CLOSE_HCA(dev, 0);
1890 		mlx4_free_icms(dev);
1891 	}
1892 }
1893 
1894 static void mlx4_close_fw(struct mlx4_dev *dev)
1895 {
1896 	if (!mlx4_is_slave(dev)) {
1897 		mlx4_UNMAP_FA(dev);
1898 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1899 	}
1900 }
1901 
1902 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1903 {
1904 #define COMM_CHAN_OFFLINE_OFFSET 0x09
1905 
1906 	u32 comm_flags;
1907 	u32 offline_bit;
1908 	unsigned long end;
1909 	struct mlx4_priv *priv = mlx4_priv(dev);
1910 
1911 	end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
1912 	while (time_before(jiffies, end)) {
1913 		comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
1914 					  MLX4_COMM_CHAN_FLAGS));
1915 		offline_bit = (comm_flags &
1916 			       (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
1917 		if (!offline_bit)
1918 			return 0;
1919 		/* There are cases as part of AER/Reset flow that PF needs
1920 		 * around 100 msec to load. We therefore sleep for 100 msec
1921 		 * to allow other tasks to make use of that CPU during this
1922 		 * time interval.
1923 		 */
1924 		msleep(100);
1925 	}
1926 	mlx4_err(dev, "Communication channel is offline.\n");
1927 	return -EIO;
1928 }
1929 
1930 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
1931 {
1932 #define COMM_CHAN_RST_OFFSET 0x1e
1933 
1934 	struct mlx4_priv *priv = mlx4_priv(dev);
1935 	u32 comm_rst;
1936 	u32 comm_caps;
1937 
1938 	comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
1939 				 MLX4_COMM_CHAN_CAPS));
1940 	comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
1941 
1942 	if (comm_rst)
1943 		dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
1944 }
1945 
1946 static int mlx4_init_slave(struct mlx4_dev *dev)
1947 {
1948 	struct mlx4_priv *priv = mlx4_priv(dev);
1949 	u64 dma = (u64) priv->mfunc.vhcr_dma;
1950 	int ret_from_reset = 0;
1951 	u32 slave_read;
1952 	u32 cmd_channel_ver;
1953 
1954 	if (atomic_read(&pf_loading)) {
1955 		mlx4_warn(dev, "PF is not ready - Deferring probe\n");
1956 		return -EPROBE_DEFER;
1957 	}
1958 
1959 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1960 	priv->cmd.max_cmds = 1;
1961 	if (mlx4_comm_check_offline(dev)) {
1962 		mlx4_err(dev, "PF is not responsive, skipping initialization\n");
1963 		goto err_offline;
1964 	}
1965 
1966 	mlx4_reset_vf_support(dev);
1967 	mlx4_warn(dev, "Sending reset\n");
1968 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1969 				       MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
1970 	/* if we are in the middle of flr the slave will try
1971 	 * NUM_OF_RESET_RETRIES times before leaving.*/
1972 	if (ret_from_reset) {
1973 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1974 			mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
1975 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
1976 			return -EPROBE_DEFER;
1977 		} else
1978 			goto err;
1979 	}
1980 
1981 	/* check the driver version - the slave I/F revision
1982 	 * must match the master's */
1983 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1984 	cmd_channel_ver = mlx4_comm_get_version();
1985 
1986 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1987 		MLX4_COMM_GET_IF_REV(slave_read)) {
1988 		mlx4_err(dev, "slave driver version is not supported by the master\n");
1989 		goto err;
1990 	}
1991 
1992 	mlx4_warn(dev, "Sending vhcr0\n");
1993 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1994 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1995 		goto err;
1996 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1997 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1998 		goto err;
1999 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
2000 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2001 		goto err;
2002 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
2003 			  MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2004 		goto err;
2005 
2006 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2007 	return 0;
2008 
2009 err:
2010 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
2011 err_offline:
2012 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2013 	return -EIO;
2014 }
2015 
2016 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
2017 {
2018 	int i;
2019 
2020 	for (i = 1; i <= dev->caps.num_ports; i++) {
2021 		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
2022 			dev->caps.gid_table_len[i] =
2023 				mlx4_get_slave_num_gids(dev, 0, i);
2024 		else
2025 			dev->caps.gid_table_len[i] = 1;
2026 		dev->caps.pkey_table_len[i] =
2027 			dev->phys_caps.pkey_phys_table_len[i] - 1;
2028 	}
2029 }
2030 
2031 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
2032 {
2033 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
2034 
2035 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
2036 	      i++) {
2037 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
2038 			break;
2039 	}
2040 
2041 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
2042 }
2043 
2044 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
2045 {
2046 	switch (dmfs_high_steer_mode) {
2047 	case MLX4_STEERING_DMFS_A0_DEFAULT:
2048 		return "default performance";
2049 
2050 	case MLX4_STEERING_DMFS_A0_DYNAMIC:
2051 		return "dynamic hybrid mode";
2052 
2053 	case MLX4_STEERING_DMFS_A0_STATIC:
2054 		return "performance optimized for limited rule configuration (static)";
2055 
2056 	case MLX4_STEERING_DMFS_A0_DISABLE:
2057 		return "disabled performance optimized steering";
2058 
2059 	case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
2060 		return "performance optimized steering not supported";
2061 
2062 	default:
2063 		return "Unrecognized mode";
2064 	}
2065 }
2066 
2067 #define MLX4_DMFS_A0_STEERING			(1UL << 2)
2068 
2069 static void choose_steering_mode(struct mlx4_dev *dev,
2070 				 struct mlx4_dev_cap *dev_cap)
2071 {
2072 	if (mlx4_log_num_mgm_entry_size <= 0) {
2073 		if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
2074 			if (dev->caps.dmfs_high_steer_mode ==
2075 			    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2076 				mlx4_err(dev, "DMFS high rate mode not supported\n");
2077 			else
2078 				dev->caps.dmfs_high_steer_mode =
2079 					MLX4_STEERING_DMFS_A0_STATIC;
2080 		}
2081 	}
2082 
2083 	if (mlx4_log_num_mgm_entry_size <= 0 &&
2084 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
2085 	    (!mlx4_is_mfunc(dev) ||
2086 	     (dev_cap->fs_max_num_qp_per_entry >=
2087 	     (dev->persist->num_vfs + 1))) &&
2088 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
2089 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
2090 		dev->oper_log_mgm_entry_size =
2091 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
2092 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
2093 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
2094 		dev->caps.fs_log_max_ucast_qp_range_size =
2095 			dev_cap->fs_log_max_ucast_qp_range_size;
2096 	} else {
2097 		if (dev->caps.dmfs_high_steer_mode !=
2098 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2099 			dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
2100 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
2101 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2102 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
2103 		else {
2104 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
2105 
2106 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
2107 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2108 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
2109 		}
2110 		dev->oper_log_mgm_entry_size =
2111 			mlx4_log_num_mgm_entry_size > 0 ?
2112 			mlx4_log_num_mgm_entry_size :
2113 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
2114 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
2115 	}
2116 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
2117 		 mlx4_steering_mode_str(dev->caps.steering_mode),
2118 		 dev->oper_log_mgm_entry_size,
2119 		 mlx4_log_num_mgm_entry_size);
2120 }
2121 
2122 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
2123 				       struct mlx4_dev_cap *dev_cap)
2124 {
2125 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2126 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
2127 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
2128 	else
2129 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
2130 
2131 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
2132 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
2133 }
2134 
2135 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
2136 {
2137 	int i;
2138 	struct mlx4_port_cap port_cap;
2139 
2140 	if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2141 		return -EINVAL;
2142 
2143 	for (i = 1; i <= dev->caps.num_ports; i++) {
2144 		if (mlx4_dev_port(dev, i, &port_cap)) {
2145 			mlx4_err(dev,
2146 				 "QUERY_DEV_CAP command failed, can't veify DMFS high rate steering.\n");
2147 		} else if ((dev->caps.dmfs_high_steer_mode !=
2148 			    MLX4_STEERING_DMFS_A0_DEFAULT) &&
2149 			   (port_cap.dmfs_optimized_state ==
2150 			    !!(dev->caps.dmfs_high_steer_mode ==
2151 			    MLX4_STEERING_DMFS_A0_DISABLE))) {
2152 			mlx4_err(dev,
2153 				 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2154 				 dmfs_high_rate_steering_mode_str(
2155 					dev->caps.dmfs_high_steer_mode),
2156 				 (port_cap.dmfs_optimized_state ?
2157 					"enabled" : "disabled"));
2158 		}
2159 	}
2160 
2161 	return 0;
2162 }
2163 
2164 static int mlx4_init_fw(struct mlx4_dev *dev)
2165 {
2166 	struct mlx4_mod_stat_cfg   mlx4_cfg;
2167 	int err = 0;
2168 
2169 	if (!mlx4_is_slave(dev)) {
2170 		err = mlx4_QUERY_FW(dev);
2171 		if (err) {
2172 			if (err == -EACCES)
2173 				mlx4_info(dev, "non-primary physical function, skipping\n");
2174 			else
2175 				mlx4_err(dev, "QUERY_FW command failed, aborting\n");
2176 			return err;
2177 		}
2178 
2179 		err = mlx4_load_fw(dev);
2180 		if (err) {
2181 			mlx4_err(dev, "Failed to start FW, aborting\n");
2182 			return err;
2183 		}
2184 
2185 		mlx4_cfg.log_pg_sz_m = 1;
2186 		mlx4_cfg.log_pg_sz = 0;
2187 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2188 		if (err)
2189 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2190 	}
2191 
2192 	return err;
2193 }
2194 
2195 static int mlx4_init_hca(struct mlx4_dev *dev)
2196 {
2197 	struct mlx4_priv	  *priv = mlx4_priv(dev);
2198 	struct mlx4_adapter	   adapter;
2199 	struct mlx4_dev_cap	   dev_cap;
2200 	struct mlx4_profile	   profile;
2201 	struct mlx4_init_hca_param init_hca;
2202 	u64 icm_size;
2203 	struct mlx4_config_dev_params params;
2204 	int err;
2205 
2206 	if (!mlx4_is_slave(dev)) {
2207 		err = mlx4_dev_cap(dev, &dev_cap);
2208 		if (err) {
2209 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
2210 			return err;
2211 		}
2212 
2213 		choose_steering_mode(dev, &dev_cap);
2214 		choose_tunnel_offload_mode(dev, &dev_cap);
2215 
2216 		if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2217 		    mlx4_is_master(dev))
2218 			dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2219 
2220 		err = mlx4_get_phys_port_id(dev);
2221 		if (err)
2222 			mlx4_err(dev, "Fail to get physical port id\n");
2223 
2224 		if (mlx4_is_master(dev))
2225 			mlx4_parav_master_pf_caps(dev);
2226 
2227 		if (mlx4_low_memory_profile()) {
2228 			mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2229 			profile = low_mem_profile;
2230 		} else {
2231 			profile = default_profile;
2232 		}
2233 		if (dev->caps.steering_mode ==
2234 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
2235 			profile.num_mcg = MLX4_FS_NUM_MCG;
2236 
2237 		icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
2238 					     &init_hca);
2239 		if ((long long) icm_size < 0) {
2240 			err = icm_size;
2241 			return err;
2242 		}
2243 
2244 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
2245 
2246 		if (enable_4k_uar) {
2247 			init_hca.log_uar_sz = ilog2(dev->caps.num_uars) +
2248 						    PAGE_SHIFT - DEFAULT_UAR_PAGE_SHIFT;
2249 			init_hca.uar_page_sz = DEFAULT_UAR_PAGE_SHIFT - 12;
2250 		} else {
2251 			init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
2252 			init_hca.uar_page_sz = PAGE_SHIFT - 12;
2253 		}
2254 
2255 		init_hca.mw_enabled = 0;
2256 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2257 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2258 			init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
2259 
2260 		err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
2261 		if (err)
2262 			return err;
2263 
2264 		err = mlx4_INIT_HCA(dev, &init_hca);
2265 		if (err) {
2266 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
2267 			goto err_free_icm;
2268 		}
2269 
2270 		if (dev_cap.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2271 			err = mlx4_query_func(dev, &dev_cap);
2272 			if (err < 0) {
2273 				mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
2274 				goto err_close;
2275 			} else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2276 				dev->caps.num_eqs = dev_cap.max_eqs;
2277 				dev->caps.reserved_eqs = dev_cap.reserved_eqs;
2278 				dev->caps.reserved_uars = dev_cap.reserved_uars;
2279 			}
2280 		}
2281 
2282 		/*
2283 		 * If TS is supported by FW
2284 		 * read HCA frequency by QUERY_HCA command
2285 		 */
2286 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2287 			memset(&init_hca, 0, sizeof(init_hca));
2288 			err = mlx4_QUERY_HCA(dev, &init_hca);
2289 			if (err) {
2290 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
2291 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2292 			} else {
2293 				dev->caps.hca_core_clock =
2294 					init_hca.hca_core_clock;
2295 			}
2296 
2297 			/* In case we got HCA frequency 0 - disable timestamping
2298 			 * to avoid dividing by zero
2299 			 */
2300 			if (!dev->caps.hca_core_clock) {
2301 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2302 				mlx4_err(dev,
2303 					 "HCA frequency is 0 - timestamping is not supported\n");
2304 			} else if (map_internal_clock(dev)) {
2305 				/*
2306 				 * Map internal clock,
2307 				 * in case of failure disable timestamping
2308 				 */
2309 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2310 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
2311 			}
2312 		}
2313 
2314 		if (dev->caps.dmfs_high_steer_mode !=
2315 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2316 			if (mlx4_validate_optimized_steering(dev))
2317 				mlx4_warn(dev, "Optimized steering validation failed\n");
2318 
2319 			if (dev->caps.dmfs_high_steer_mode ==
2320 			    MLX4_STEERING_DMFS_A0_DISABLE) {
2321 				dev->caps.dmfs_high_rate_qpn_base =
2322 					dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2323 				dev->caps.dmfs_high_rate_qpn_range =
2324 					MLX4_A0_STEERING_TABLE_SIZE;
2325 			}
2326 
2327 			mlx4_dbg(dev, "DMFS high rate steer mode is: %s\n",
2328 				 dmfs_high_rate_steering_mode_str(
2329 					dev->caps.dmfs_high_steer_mode));
2330 		}
2331 	} else {
2332 		err = mlx4_init_slave(dev);
2333 		if (err) {
2334 			if (err != -EPROBE_DEFER)
2335 				mlx4_err(dev, "Failed to initialize slave\n");
2336 			return err;
2337 		}
2338 
2339 		err = mlx4_slave_cap(dev);
2340 		if (err) {
2341 			mlx4_err(dev, "Failed to obtain slave caps\n");
2342 			goto err_close;
2343 		}
2344 	}
2345 
2346 	if (map_bf_area(dev))
2347 		mlx4_dbg(dev, "Failed to map blue flame area\n");
2348 
2349 	/*Only the master set the ports, all the rest got it from it.*/
2350 	if (!mlx4_is_slave(dev))
2351 		mlx4_set_port_mask(dev);
2352 
2353 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
2354 	if (err) {
2355 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2356 		goto unmap_bf;
2357 	}
2358 
2359 	/* Query CONFIG_DEV parameters */
2360 	err = mlx4_config_dev_retrieval(dev, &params);
2361 	if (err && err != -ENOTSUPP) {
2362 		mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2363 	} else if (!err) {
2364 		dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2365 		dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2366 	}
2367 	priv->eq_table.inta_pin = adapter.inta_pin;
2368 	memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
2369 
2370 	return 0;
2371 
2372 unmap_bf:
2373 	unmap_internal_clock(dev);
2374 	unmap_bf_area(dev);
2375 
2376 	if (mlx4_is_slave(dev)) {
2377 		kfree(dev->caps.qp0_qkey);
2378 		kfree(dev->caps.qp0_tunnel);
2379 		kfree(dev->caps.qp0_proxy);
2380 		kfree(dev->caps.qp1_tunnel);
2381 		kfree(dev->caps.qp1_proxy);
2382 	}
2383 
2384 err_close:
2385 	if (mlx4_is_slave(dev))
2386 		mlx4_slave_exit(dev);
2387 	else
2388 		mlx4_CLOSE_HCA(dev, 0);
2389 
2390 err_free_icm:
2391 	if (!mlx4_is_slave(dev))
2392 		mlx4_free_icms(dev);
2393 
2394 	return err;
2395 }
2396 
2397 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2398 {
2399 	struct mlx4_priv *priv = mlx4_priv(dev);
2400 	int nent_pow2;
2401 
2402 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2403 		return -ENOENT;
2404 
2405 	if (!dev->caps.max_counters)
2406 		return -ENOSPC;
2407 
2408 	nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2409 	/* reserve last counter index for sink counter */
2410 	return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2411 				nent_pow2 - 1, 0,
2412 				nent_pow2 - dev->caps.max_counters + 1);
2413 }
2414 
2415 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2416 {
2417 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2418 		return;
2419 
2420 	if (!dev->caps.max_counters)
2421 		return;
2422 
2423 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2424 }
2425 
2426 static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2427 {
2428 	struct mlx4_priv *priv = mlx4_priv(dev);
2429 	int port;
2430 
2431 	for (port = 0; port < dev->caps.num_ports; port++)
2432 		if (priv->def_counter[port] != -1)
2433 			mlx4_counter_free(dev,  priv->def_counter[port]);
2434 }
2435 
2436 static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2437 {
2438 	struct mlx4_priv *priv = mlx4_priv(dev);
2439 	int port, err = 0;
2440 	u32 idx;
2441 
2442 	for (port = 0; port < dev->caps.num_ports; port++)
2443 		priv->def_counter[port] = -1;
2444 
2445 	for (port = 0; port < dev->caps.num_ports; port++) {
2446 		err = mlx4_counter_alloc(dev, &idx);
2447 
2448 		if (!err || err == -ENOSPC) {
2449 			priv->def_counter[port] = idx;
2450 		} else if (err == -ENOENT) {
2451 			err = 0;
2452 			continue;
2453 		} else if (mlx4_is_slave(dev) && err == -EINVAL) {
2454 			priv->def_counter[port] = MLX4_SINK_COUNTER_INDEX(dev);
2455 			mlx4_warn(dev, "can't allocate counter from old PF driver, using index %d\n",
2456 				  MLX4_SINK_COUNTER_INDEX(dev));
2457 			err = 0;
2458 		} else {
2459 			mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2460 				 __func__, port + 1, err);
2461 			mlx4_cleanup_default_counters(dev);
2462 			return err;
2463 		}
2464 
2465 		mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2466 			 __func__, priv->def_counter[port], port + 1);
2467 	}
2468 
2469 	return err;
2470 }
2471 
2472 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2473 {
2474 	struct mlx4_priv *priv = mlx4_priv(dev);
2475 
2476 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2477 		return -ENOENT;
2478 
2479 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2480 	if (*idx == -1) {
2481 		*idx = MLX4_SINK_COUNTER_INDEX(dev);
2482 		return -ENOSPC;
2483 	}
2484 
2485 	return 0;
2486 }
2487 
2488 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2489 {
2490 	u64 out_param;
2491 	int err;
2492 
2493 	if (mlx4_is_mfunc(dev)) {
2494 		err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
2495 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2496 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2497 		if (!err)
2498 			*idx = get_param_l(&out_param);
2499 
2500 		return err;
2501 	}
2502 	return __mlx4_counter_alloc(dev, idx);
2503 }
2504 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2505 
2506 static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2507 				u8 counter_index)
2508 {
2509 	struct mlx4_cmd_mailbox *if_stat_mailbox;
2510 	int err;
2511 	u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2512 
2513 	if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2514 	if (IS_ERR(if_stat_mailbox))
2515 		return PTR_ERR(if_stat_mailbox);
2516 
2517 	err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2518 			   MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2519 			   MLX4_CMD_NATIVE);
2520 
2521 	mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2522 	return err;
2523 }
2524 
2525 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2526 {
2527 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2528 		return;
2529 
2530 	if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2531 		return;
2532 
2533 	__mlx4_clear_if_stat(dev, idx);
2534 
2535 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2536 	return;
2537 }
2538 
2539 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2540 {
2541 	u64 in_param = 0;
2542 
2543 	if (mlx4_is_mfunc(dev)) {
2544 		set_param_l(&in_param, idx);
2545 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2546 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2547 			 MLX4_CMD_WRAPPED);
2548 		return;
2549 	}
2550 	__mlx4_counter_free(dev, idx);
2551 }
2552 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2553 
2554 int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2555 {
2556 	struct mlx4_priv *priv = mlx4_priv(dev);
2557 
2558 	return priv->def_counter[port - 1];
2559 }
2560 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2561 
2562 void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2563 {
2564 	struct mlx4_priv *priv = mlx4_priv(dev);
2565 
2566 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2567 }
2568 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2569 
2570 __be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2571 {
2572 	struct mlx4_priv *priv = mlx4_priv(dev);
2573 
2574 	return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2575 }
2576 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2577 
2578 void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2579 {
2580 	struct mlx4_priv *priv = mlx4_priv(dev);
2581 	__be64 guid;
2582 
2583 	/* hw GUID */
2584 	if (entry == 0)
2585 		return;
2586 
2587 	get_random_bytes((char *)&guid, sizeof(guid));
2588 	guid &= ~(cpu_to_be64(1ULL << 56));
2589 	guid |= cpu_to_be64(1ULL << 57);
2590 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2591 }
2592 
2593 static int mlx4_setup_hca(struct mlx4_dev *dev)
2594 {
2595 	struct mlx4_priv *priv = mlx4_priv(dev);
2596 	int err;
2597 	int port;
2598 	__be32 ib_port_default_caps;
2599 
2600 	err = mlx4_init_uar_table(dev);
2601 	if (err) {
2602 		mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2603 		return err;
2604 	}
2605 
2606 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
2607 	if (err) {
2608 		mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2609 		goto err_uar_table_free;
2610 	}
2611 
2612 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2613 	if (!priv->kar) {
2614 		mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2615 		err = -ENOMEM;
2616 		goto err_uar_free;
2617 	}
2618 
2619 	err = mlx4_init_pd_table(dev);
2620 	if (err) {
2621 		mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2622 		goto err_kar_unmap;
2623 	}
2624 
2625 	err = mlx4_init_xrcd_table(dev);
2626 	if (err) {
2627 		mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2628 		goto err_pd_table_free;
2629 	}
2630 
2631 	err = mlx4_init_mr_table(dev);
2632 	if (err) {
2633 		mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2634 		goto err_xrcd_table_free;
2635 	}
2636 
2637 	if (!mlx4_is_slave(dev)) {
2638 		err = mlx4_init_mcg_table(dev);
2639 		if (err) {
2640 			mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2641 			goto err_mr_table_free;
2642 		}
2643 		err = mlx4_config_mad_demux(dev);
2644 		if (err) {
2645 			mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2646 			goto err_mcg_table_free;
2647 		}
2648 	}
2649 
2650 	err = mlx4_init_eq_table(dev);
2651 	if (err) {
2652 		mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2653 		goto err_mcg_table_free;
2654 	}
2655 
2656 	err = mlx4_cmd_use_events(dev);
2657 	if (err) {
2658 		mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2659 		goto err_eq_table_free;
2660 	}
2661 
2662 	err = mlx4_NOP(dev);
2663 	if (err) {
2664 		if (dev->flags & MLX4_FLAG_MSI_X) {
2665 			mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2666 				  priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2667 			mlx4_warn(dev, "Trying again without MSI-X\n");
2668 		} else {
2669 			mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2670 				 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2671 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2672 		}
2673 
2674 		goto err_cmd_poll;
2675 	}
2676 
2677 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
2678 
2679 	err = mlx4_init_cq_table(dev);
2680 	if (err) {
2681 		mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2682 		goto err_cmd_poll;
2683 	}
2684 
2685 	err = mlx4_init_srq_table(dev);
2686 	if (err) {
2687 		mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2688 		goto err_cq_table_free;
2689 	}
2690 
2691 	err = mlx4_init_qp_table(dev);
2692 	if (err) {
2693 		mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2694 		goto err_srq_table_free;
2695 	}
2696 
2697 	if (!mlx4_is_slave(dev)) {
2698 		err = mlx4_init_counters_table(dev);
2699 		if (err && err != -ENOENT) {
2700 			mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2701 			goto err_qp_table_free;
2702 		}
2703 	}
2704 
2705 	err = mlx4_allocate_default_counters(dev);
2706 	if (err) {
2707 		mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2708 		goto err_counters_table_free;
2709 	}
2710 
2711 	if (!mlx4_is_slave(dev)) {
2712 		for (port = 1; port <= dev->caps.num_ports; port++) {
2713 			ib_port_default_caps = 0;
2714 			err = mlx4_get_port_ib_caps(dev, port,
2715 						    &ib_port_default_caps);
2716 			if (err)
2717 				mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2718 					  port, err);
2719 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2720 
2721 			/* initialize per-slave default ib port capabilities */
2722 			if (mlx4_is_master(dev)) {
2723 				int i;
2724 				for (i = 0; i < dev->num_slaves; i++) {
2725 					if (i == mlx4_master_func_num(dev))
2726 						continue;
2727 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2728 						ib_port_default_caps;
2729 				}
2730 			}
2731 
2732 			if (mlx4_is_mfunc(dev))
2733 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2734 			else
2735 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2736 
2737 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2738 					    dev->caps.pkey_table_len[port] : -1);
2739 			if (err) {
2740 				mlx4_err(dev, "Failed to set port %d, aborting\n",
2741 					 port);
2742 				goto err_default_countes_free;
2743 			}
2744 		}
2745 	}
2746 
2747 	return 0;
2748 
2749 err_default_countes_free:
2750 	mlx4_cleanup_default_counters(dev);
2751 
2752 err_counters_table_free:
2753 	if (!mlx4_is_slave(dev))
2754 		mlx4_cleanup_counters_table(dev);
2755 
2756 err_qp_table_free:
2757 	mlx4_cleanup_qp_table(dev);
2758 
2759 err_srq_table_free:
2760 	mlx4_cleanup_srq_table(dev);
2761 
2762 err_cq_table_free:
2763 	mlx4_cleanup_cq_table(dev);
2764 
2765 err_cmd_poll:
2766 	mlx4_cmd_use_polling(dev);
2767 
2768 err_eq_table_free:
2769 	mlx4_cleanup_eq_table(dev);
2770 
2771 err_mcg_table_free:
2772 	if (!mlx4_is_slave(dev))
2773 		mlx4_cleanup_mcg_table(dev);
2774 
2775 err_mr_table_free:
2776 	mlx4_cleanup_mr_table(dev);
2777 
2778 err_xrcd_table_free:
2779 	mlx4_cleanup_xrcd_table(dev);
2780 
2781 err_pd_table_free:
2782 	mlx4_cleanup_pd_table(dev);
2783 
2784 err_kar_unmap:
2785 	iounmap(priv->kar);
2786 
2787 err_uar_free:
2788 	mlx4_uar_free(dev, &priv->driver_uar);
2789 
2790 err_uar_table_free:
2791 	mlx4_cleanup_uar_table(dev);
2792 	return err;
2793 }
2794 
2795 static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2796 {
2797 	int requested_cpu = 0;
2798 	struct mlx4_priv *priv = mlx4_priv(dev);
2799 	struct mlx4_eq *eq;
2800 	int off = 0;
2801 	int i;
2802 
2803 	if (eqn > dev->caps.num_comp_vectors)
2804 		return -EINVAL;
2805 
2806 	for (i = 1; i < port; i++)
2807 		off += mlx4_get_eqs_per_port(dev, i);
2808 
2809 	requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2810 
2811 	/* Meaning EQs are shared, and this call comes from the second port */
2812 	if (requested_cpu < 0)
2813 		return 0;
2814 
2815 	eq = &priv->eq_table.eq[eqn];
2816 
2817 	if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2818 		return -ENOMEM;
2819 
2820 	cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2821 
2822 	return 0;
2823 }
2824 
2825 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2826 {
2827 	struct mlx4_priv *priv = mlx4_priv(dev);
2828 	struct msix_entry *entries;
2829 	int i;
2830 	int port = 0;
2831 
2832 	if (msi_x) {
2833 		int nreq = dev->caps.num_ports * num_online_cpus() + 1;
2834 
2835 		nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
2836 			     nreq);
2837 		if (nreq > MAX_MSIX)
2838 			nreq = MAX_MSIX;
2839 
2840 		entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
2841 		if (!entries)
2842 			goto no_msi;
2843 
2844 		for (i = 0; i < nreq; ++i)
2845 			entries[i].entry = i;
2846 
2847 		nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2848 					     nreq);
2849 
2850 		if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
2851 			kfree(entries);
2852 			goto no_msi;
2853 		}
2854 		/* 1 is reserved for events (asyncrounous EQ) */
2855 		dev->caps.num_comp_vectors = nreq - 1;
2856 
2857 		priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2858 		bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2859 			    dev->caps.num_ports);
2860 
2861 		for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2862 			if (i == MLX4_EQ_ASYNC)
2863 				continue;
2864 
2865 			priv->eq_table.eq[i].irq =
2866 				entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2867 
2868 			if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2869 				bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2870 					    dev->caps.num_ports);
2871 				/* We don't set affinity hint when there
2872 				 * aren't enough EQs
2873 				 */
2874 			} else {
2875 				set_bit(port,
2876 					priv->eq_table.eq[i].actv_ports.ports);
2877 				if (mlx4_init_affinity_hint(dev, port + 1, i))
2878 					mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
2879 						  i);
2880 			}
2881 			/* We divide the Eqs evenly between the two ports.
2882 			 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
2883 			 * refers to the number of Eqs per port
2884 			 * (i.e eqs_per_port). Theoretically, we would like to
2885 			 * write something like (i + 1) % eqs_per_port == 0.
2886 			 * However, since there's an asynchronous Eq, we have
2887 			 * to skip over it by comparing this condition to
2888 			 * !!((i + 1) > MLX4_EQ_ASYNC).
2889 			 */
2890 			if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
2891 			    ((i + 1) %
2892 			     (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
2893 			    !!((i + 1) > MLX4_EQ_ASYNC))
2894 				/* If dev->caps.num_comp_vectors < dev->caps.num_ports,
2895 				 * everything is shared anyway.
2896 				 */
2897 				port++;
2898 		}
2899 
2900 		dev->flags |= MLX4_FLAG_MSI_X;
2901 
2902 		kfree(entries);
2903 		return;
2904 	}
2905 
2906 no_msi:
2907 	dev->caps.num_comp_vectors = 1;
2908 
2909 	BUG_ON(MLX4_EQ_ASYNC >= 2);
2910 	for (i = 0; i < 2; ++i) {
2911 		priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
2912 		if (i != MLX4_EQ_ASYNC) {
2913 			bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2914 				    dev->caps.num_ports);
2915 		}
2916 	}
2917 }
2918 
2919 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2920 {
2921 	struct devlink *devlink = priv_to_devlink(mlx4_priv(dev));
2922 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2923 	int err;
2924 
2925 	err = devlink_port_register(devlink, &info->devlink_port, port);
2926 	if (err)
2927 		return err;
2928 
2929 	info->dev = dev;
2930 	info->port = port;
2931 	if (!mlx4_is_slave(dev)) {
2932 		mlx4_init_mac_table(dev, &info->mac_table);
2933 		mlx4_init_vlan_table(dev, &info->vlan_table);
2934 		mlx4_init_roce_gid_table(dev, &info->gid_table);
2935 		info->base_qpn = mlx4_get_base_qpn(dev, port);
2936 	}
2937 
2938 	sprintf(info->dev_name, "mlx4_port%d", port);
2939 	info->port_attr.attr.name = info->dev_name;
2940 	if (mlx4_is_mfunc(dev))
2941 		info->port_attr.attr.mode = S_IRUGO;
2942 	else {
2943 		info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2944 		info->port_attr.store     = set_port_type;
2945 	}
2946 	info->port_attr.show      = show_port_type;
2947 	sysfs_attr_init(&info->port_attr.attr);
2948 
2949 	err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
2950 	if (err) {
2951 		mlx4_err(dev, "Failed to create file for port %d\n", port);
2952 		devlink_port_unregister(&info->devlink_port);
2953 		info->port = -1;
2954 	}
2955 
2956 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2957 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
2958 	if (mlx4_is_mfunc(dev))
2959 		info->port_mtu_attr.attr.mode = S_IRUGO;
2960 	else {
2961 		info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2962 		info->port_mtu_attr.store     = set_port_ib_mtu;
2963 	}
2964 	info->port_mtu_attr.show      = show_port_ib_mtu;
2965 	sysfs_attr_init(&info->port_mtu_attr.attr);
2966 
2967 	err = device_create_file(&dev->persist->pdev->dev,
2968 				 &info->port_mtu_attr);
2969 	if (err) {
2970 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
2971 		device_remove_file(&info->dev->persist->pdev->dev,
2972 				   &info->port_attr);
2973 		info->port = -1;
2974 	}
2975 
2976 	return err;
2977 }
2978 
2979 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
2980 {
2981 	if (info->port < 0)
2982 		return;
2983 
2984 	device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
2985 	device_remove_file(&info->dev->persist->pdev->dev,
2986 			   &info->port_mtu_attr);
2987 #ifdef CONFIG_RFS_ACCEL
2988 	free_irq_cpu_rmap(info->rmap);
2989 	info->rmap = NULL;
2990 #endif
2991 }
2992 
2993 static int mlx4_init_steering(struct mlx4_dev *dev)
2994 {
2995 	struct mlx4_priv *priv = mlx4_priv(dev);
2996 	int num_entries = dev->caps.num_ports;
2997 	int i, j;
2998 
2999 	priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
3000 	if (!priv->steer)
3001 		return -ENOMEM;
3002 
3003 	for (i = 0; i < num_entries; i++)
3004 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3005 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
3006 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
3007 		}
3008 	return 0;
3009 }
3010 
3011 static void mlx4_clear_steering(struct mlx4_dev *dev)
3012 {
3013 	struct mlx4_priv *priv = mlx4_priv(dev);
3014 	struct mlx4_steer_index *entry, *tmp_entry;
3015 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
3016 	int num_entries = dev->caps.num_ports;
3017 	int i, j;
3018 
3019 	for (i = 0; i < num_entries; i++) {
3020 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3021 			list_for_each_entry_safe(pqp, tmp_pqp,
3022 						 &priv->steer[i].promisc_qps[j],
3023 						 list) {
3024 				list_del(&pqp->list);
3025 				kfree(pqp);
3026 			}
3027 			list_for_each_entry_safe(entry, tmp_entry,
3028 						 &priv->steer[i].steer_entries[j],
3029 						 list) {
3030 				list_del(&entry->list);
3031 				list_for_each_entry_safe(pqp, tmp_pqp,
3032 							 &entry->duplicates,
3033 							 list) {
3034 					list_del(&pqp->list);
3035 					kfree(pqp);
3036 				}
3037 				kfree(entry);
3038 			}
3039 		}
3040 	}
3041 	kfree(priv->steer);
3042 }
3043 
3044 static int extended_func_num(struct pci_dev *pdev)
3045 {
3046 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
3047 }
3048 
3049 #define MLX4_OWNER_BASE	0x8069c
3050 #define MLX4_OWNER_SIZE	4
3051 
3052 static int mlx4_get_ownership(struct mlx4_dev *dev)
3053 {
3054 	void __iomem *owner;
3055 	u32 ret;
3056 
3057 	if (pci_channel_offline(dev->persist->pdev))
3058 		return -EIO;
3059 
3060 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3061 			MLX4_OWNER_BASE,
3062 			MLX4_OWNER_SIZE);
3063 	if (!owner) {
3064 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3065 		return -ENOMEM;
3066 	}
3067 
3068 	ret = readl(owner);
3069 	iounmap(owner);
3070 	return (int) !!ret;
3071 }
3072 
3073 static void mlx4_free_ownership(struct mlx4_dev *dev)
3074 {
3075 	void __iomem *owner;
3076 
3077 	if (pci_channel_offline(dev->persist->pdev))
3078 		return;
3079 
3080 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3081 			MLX4_OWNER_BASE,
3082 			MLX4_OWNER_SIZE);
3083 	if (!owner) {
3084 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3085 		return;
3086 	}
3087 	writel(0, owner);
3088 	msleep(1000);
3089 	iounmap(owner);
3090 }
3091 
3092 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV)	==\
3093 				  !!((flags) & MLX4_FLAG_MASTER))
3094 
3095 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
3096 			     u8 total_vfs, int existing_vfs, int reset_flow)
3097 {
3098 	u64 dev_flags = dev->flags;
3099 	int err = 0;
3100 	int fw_enabled_sriov_vfs = min(pci_sriov_get_totalvfs(pdev),
3101 					MLX4_MAX_NUM_VF);
3102 
3103 	if (reset_flow) {
3104 		dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
3105 				       GFP_KERNEL);
3106 		if (!dev->dev_vfs)
3107 			goto free_mem;
3108 		return dev_flags;
3109 	}
3110 
3111 	atomic_inc(&pf_loading);
3112 	if (dev->flags &  MLX4_FLAG_SRIOV) {
3113 		if (existing_vfs != total_vfs) {
3114 			mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
3115 				 existing_vfs, total_vfs);
3116 			total_vfs = existing_vfs;
3117 		}
3118 	}
3119 
3120 	dev->dev_vfs = kzalloc(total_vfs * sizeof(*dev->dev_vfs), GFP_KERNEL);
3121 	if (NULL == dev->dev_vfs) {
3122 		mlx4_err(dev, "Failed to allocate memory for VFs\n");
3123 		goto disable_sriov;
3124 	}
3125 
3126 	if (!(dev->flags &  MLX4_FLAG_SRIOV)) {
3127 		if (total_vfs > fw_enabled_sriov_vfs) {
3128 			mlx4_err(dev, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
3129 				 total_vfs, fw_enabled_sriov_vfs);
3130 			err = -ENOMEM;
3131 			goto disable_sriov;
3132 		}
3133 		mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
3134 		err = pci_enable_sriov(pdev, total_vfs);
3135 	}
3136 	if (err) {
3137 		mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
3138 			 err);
3139 		goto disable_sriov;
3140 	} else {
3141 		mlx4_warn(dev, "Running in master mode\n");
3142 		dev_flags |= MLX4_FLAG_SRIOV |
3143 			MLX4_FLAG_MASTER;
3144 		dev_flags &= ~MLX4_FLAG_SLAVE;
3145 		dev->persist->num_vfs = total_vfs;
3146 	}
3147 	return dev_flags;
3148 
3149 disable_sriov:
3150 	atomic_dec(&pf_loading);
3151 free_mem:
3152 	dev->persist->num_vfs = 0;
3153 	kfree(dev->dev_vfs);
3154         dev->dev_vfs = NULL;
3155 	return dev_flags & ~MLX4_FLAG_MASTER;
3156 }
3157 
3158 enum {
3159 	MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
3160 };
3161 
3162 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
3163 			      int *nvfs)
3164 {
3165 	int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
3166 	/* Checking for 64 VFs as a limitation of CX2 */
3167 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
3168 	    requested_vfs >= 64) {
3169 		mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
3170 			 requested_vfs);
3171 		return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
3172 	}
3173 	return 0;
3174 }
3175 
3176 static int mlx4_pci_enable_device(struct mlx4_dev *dev)
3177 {
3178 	struct pci_dev *pdev = dev->persist->pdev;
3179 	int err = 0;
3180 
3181 	mutex_lock(&dev->persist->pci_status_mutex);
3182 	if (dev->persist->pci_status == MLX4_PCI_STATUS_DISABLED) {
3183 		err = pci_enable_device(pdev);
3184 		if (!err)
3185 			dev->persist->pci_status = MLX4_PCI_STATUS_ENABLED;
3186 	}
3187 	mutex_unlock(&dev->persist->pci_status_mutex);
3188 
3189 	return err;
3190 }
3191 
3192 static void mlx4_pci_disable_device(struct mlx4_dev *dev)
3193 {
3194 	struct pci_dev *pdev = dev->persist->pdev;
3195 
3196 	mutex_lock(&dev->persist->pci_status_mutex);
3197 	if (dev->persist->pci_status == MLX4_PCI_STATUS_ENABLED) {
3198 		pci_disable_device(pdev);
3199 		dev->persist->pci_status = MLX4_PCI_STATUS_DISABLED;
3200 	}
3201 	mutex_unlock(&dev->persist->pci_status_mutex);
3202 }
3203 
3204 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
3205 			 int total_vfs, int *nvfs, struct mlx4_priv *priv,
3206 			 int reset_flow)
3207 {
3208 	struct mlx4_dev *dev;
3209 	unsigned sum = 0;
3210 	int err;
3211 	int port;
3212 	int i;
3213 	struct mlx4_dev_cap *dev_cap = NULL;
3214 	int existing_vfs = 0;
3215 
3216 	dev = &priv->dev;
3217 
3218 	INIT_LIST_HEAD(&priv->ctx_list);
3219 	spin_lock_init(&priv->ctx_lock);
3220 
3221 	mutex_init(&priv->port_mutex);
3222 	mutex_init(&priv->bond_mutex);
3223 
3224 	INIT_LIST_HEAD(&priv->pgdir_list);
3225 	mutex_init(&priv->pgdir_mutex);
3226 	spin_lock_init(&priv->cmd.context_lock);
3227 
3228 	INIT_LIST_HEAD(&priv->bf_list);
3229 	mutex_init(&priv->bf_mutex);
3230 
3231 	dev->rev_id = pdev->revision;
3232 	dev->numa_node = dev_to_node(&pdev->dev);
3233 
3234 	/* Detect if this device is a virtual function */
3235 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3236 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3237 		dev->flags |= MLX4_FLAG_SLAVE;
3238 	} else {
3239 		/* We reset the device and enable SRIOV only for physical
3240 		 * devices.  Try to claim ownership on the device;
3241 		 * if already taken, skip -- do not allow multiple PFs */
3242 		err = mlx4_get_ownership(dev);
3243 		if (err) {
3244 			if (err < 0)
3245 				return err;
3246 			else {
3247 				mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
3248 				return -EINVAL;
3249 			}
3250 		}
3251 
3252 		atomic_set(&priv->opreq_count, 0);
3253 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3254 
3255 		/*
3256 		 * Now reset the HCA before we touch the PCI capabilities or
3257 		 * attempt a firmware command, since a boot ROM may have left
3258 		 * the HCA in an undefined state.
3259 		 */
3260 		err = mlx4_reset(dev);
3261 		if (err) {
3262 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
3263 			goto err_sriov;
3264 		}
3265 
3266 		if (total_vfs) {
3267 			dev->flags = MLX4_FLAG_MASTER;
3268 			existing_vfs = pci_num_vf(pdev);
3269 			if (existing_vfs)
3270 				dev->flags |= MLX4_FLAG_SRIOV;
3271 			dev->persist->num_vfs = total_vfs;
3272 		}
3273 	}
3274 
3275 	/* on load remove any previous indication of internal error,
3276 	 * device is up.
3277 	 */
3278 	dev->persist->state = MLX4_DEVICE_STATE_UP;
3279 
3280 slave_start:
3281 	err = mlx4_cmd_init(dev);
3282 	if (err) {
3283 		mlx4_err(dev, "Failed to init command interface, aborting\n");
3284 		goto err_sriov;
3285 	}
3286 
3287 	/* In slave functions, the communication channel must be initialized
3288 	 * before posting commands. Also, init num_slaves before calling
3289 	 * mlx4_init_hca */
3290 	if (mlx4_is_mfunc(dev)) {
3291 		if (mlx4_is_master(dev)) {
3292 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
3293 
3294 		} else {
3295 			dev->num_slaves = 0;
3296 			err = mlx4_multi_func_init(dev);
3297 			if (err) {
3298 				mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
3299 				goto err_cmd;
3300 			}
3301 		}
3302 	}
3303 
3304 	err = mlx4_init_fw(dev);
3305 	if (err) {
3306 		mlx4_err(dev, "Failed to init fw, aborting.\n");
3307 		goto err_mfunc;
3308 	}
3309 
3310 	if (mlx4_is_master(dev)) {
3311 		/* when we hit the goto slave_start below, dev_cap already initialized */
3312 		if (!dev_cap) {
3313 			dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3314 
3315 			if (!dev_cap) {
3316 				err = -ENOMEM;
3317 				goto err_fw;
3318 			}
3319 
3320 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3321 			if (err) {
3322 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3323 				goto err_fw;
3324 			}
3325 
3326 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3327 				goto err_fw;
3328 
3329 			if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3330 				u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3331 								  total_vfs,
3332 								  existing_vfs,
3333 								  reset_flow);
3334 
3335 				mlx4_close_fw(dev);
3336 				mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3337 				dev->flags = dev_flags;
3338 				if (!SRIOV_VALID_STATE(dev->flags)) {
3339 					mlx4_err(dev, "Invalid SRIOV state\n");
3340 					goto err_sriov;
3341 				}
3342 				err = mlx4_reset(dev);
3343 				if (err) {
3344 					mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3345 					goto err_sriov;
3346 				}
3347 				goto slave_start;
3348 			}
3349 		} else {
3350 			/* Legacy mode FW requires SRIOV to be enabled before
3351 			 * doing QUERY_DEV_CAP, since max_eq's value is different if
3352 			 * SRIOV is enabled.
3353 			 */
3354 			memset(dev_cap, 0, sizeof(*dev_cap));
3355 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3356 			if (err) {
3357 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3358 				goto err_fw;
3359 			}
3360 
3361 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3362 				goto err_fw;
3363 		}
3364 	}
3365 
3366 	err = mlx4_init_hca(dev);
3367 	if (err) {
3368 		if (err == -EACCES) {
3369 			/* Not primary Physical function
3370 			 * Running in slave mode */
3371 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3372 			/* We're not a PF */
3373 			if (dev->flags & MLX4_FLAG_SRIOV) {
3374 				if (!existing_vfs)
3375 					pci_disable_sriov(pdev);
3376 				if (mlx4_is_master(dev) && !reset_flow)
3377 					atomic_dec(&pf_loading);
3378 				dev->flags &= ~MLX4_FLAG_SRIOV;
3379 			}
3380 			if (!mlx4_is_slave(dev))
3381 				mlx4_free_ownership(dev);
3382 			dev->flags |= MLX4_FLAG_SLAVE;
3383 			dev->flags &= ~MLX4_FLAG_MASTER;
3384 			goto slave_start;
3385 		} else
3386 			goto err_fw;
3387 	}
3388 
3389 	if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3390 		u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3391 						  existing_vfs, reset_flow);
3392 
3393 		if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3394 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3395 			dev->flags = dev_flags;
3396 			err = mlx4_cmd_init(dev);
3397 			if (err) {
3398 				/* Only VHCR is cleaned up, so could still
3399 				 * send FW commands
3400 				 */
3401 				mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3402 				goto err_close;
3403 			}
3404 		} else {
3405 			dev->flags = dev_flags;
3406 		}
3407 
3408 		if (!SRIOV_VALID_STATE(dev->flags)) {
3409 			mlx4_err(dev, "Invalid SRIOV state\n");
3410 			goto err_close;
3411 		}
3412 	}
3413 
3414 	/* check if the device is functioning at its maximum possible speed.
3415 	 * No return code for this call, just warn the user in case of PCI
3416 	 * express device capabilities are under-satisfied by the bus.
3417 	 */
3418 	if (!mlx4_is_slave(dev))
3419 		mlx4_check_pcie_caps(dev);
3420 
3421 	/* In master functions, the communication channel must be initialized
3422 	 * after obtaining its address from fw */
3423 	if (mlx4_is_master(dev)) {
3424 		if (dev->caps.num_ports < 2 &&
3425 		    num_vfs_argc > 1) {
3426 			err = -EINVAL;
3427 			mlx4_err(dev,
3428 				 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3429 				 dev->caps.num_ports);
3430 			goto err_close;
3431 		}
3432 		memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
3433 
3434 		for (i = 0;
3435 		     i < sizeof(dev->persist->nvfs)/
3436 		     sizeof(dev->persist->nvfs[0]); i++) {
3437 			unsigned j;
3438 
3439 			for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
3440 				dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3441 				dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3442 					dev->caps.num_ports;
3443 			}
3444 		}
3445 
3446 		/* In master functions, the communication channel
3447 		 * must be initialized after obtaining its address from fw
3448 		 */
3449 		err = mlx4_multi_func_init(dev);
3450 		if (err) {
3451 			mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3452 			goto err_close;
3453 		}
3454 	}
3455 
3456 	err = mlx4_alloc_eq_table(dev);
3457 	if (err)
3458 		goto err_master_mfunc;
3459 
3460 	bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
3461 	mutex_init(&priv->msix_ctl.pool_lock);
3462 
3463 	mlx4_enable_msi_x(dev);
3464 	if ((mlx4_is_mfunc(dev)) &&
3465 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
3466 		err = -ENOSYS;
3467 		mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
3468 		goto err_free_eq;
3469 	}
3470 
3471 	if (!mlx4_is_slave(dev)) {
3472 		err = mlx4_init_steering(dev);
3473 		if (err)
3474 			goto err_disable_msix;
3475 	}
3476 
3477 	err = mlx4_setup_hca(dev);
3478 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3479 	    !mlx4_is_mfunc(dev)) {
3480 		dev->flags &= ~MLX4_FLAG_MSI_X;
3481 		dev->caps.num_comp_vectors = 1;
3482 		pci_disable_msix(pdev);
3483 		err = mlx4_setup_hca(dev);
3484 	}
3485 
3486 	if (err)
3487 		goto err_steer;
3488 
3489 	mlx4_init_quotas(dev);
3490 	/* When PF resources are ready arm its comm channel to enable
3491 	 * getting commands
3492 	 */
3493 	if (mlx4_is_master(dev)) {
3494 		err = mlx4_ARM_COMM_CHANNEL(dev);
3495 		if (err) {
3496 			mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3497 				 err);
3498 			goto err_steer;
3499 		}
3500 	}
3501 
3502 	for (port = 1; port <= dev->caps.num_ports; port++) {
3503 		err = mlx4_init_port_info(dev, port);
3504 		if (err)
3505 			goto err_port;
3506 	}
3507 
3508 	priv->v2p.port1 = 1;
3509 	priv->v2p.port2 = 2;
3510 
3511 	err = mlx4_register_device(dev);
3512 	if (err)
3513 		goto err_port;
3514 
3515 	mlx4_request_modules(dev);
3516 
3517 	mlx4_sense_init(dev);
3518 	mlx4_start_sense(dev);
3519 
3520 	priv->removed = 0;
3521 
3522 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3523 		atomic_dec(&pf_loading);
3524 
3525 	kfree(dev_cap);
3526 	return 0;
3527 
3528 err_port:
3529 	for (--port; port >= 1; --port)
3530 		mlx4_cleanup_port_info(&priv->port[port]);
3531 
3532 	mlx4_cleanup_default_counters(dev);
3533 	if (!mlx4_is_slave(dev))
3534 		mlx4_cleanup_counters_table(dev);
3535 	mlx4_cleanup_qp_table(dev);
3536 	mlx4_cleanup_srq_table(dev);
3537 	mlx4_cleanup_cq_table(dev);
3538 	mlx4_cmd_use_polling(dev);
3539 	mlx4_cleanup_eq_table(dev);
3540 	mlx4_cleanup_mcg_table(dev);
3541 	mlx4_cleanup_mr_table(dev);
3542 	mlx4_cleanup_xrcd_table(dev);
3543 	mlx4_cleanup_pd_table(dev);
3544 	mlx4_cleanup_uar_table(dev);
3545 
3546 err_steer:
3547 	if (!mlx4_is_slave(dev))
3548 		mlx4_clear_steering(dev);
3549 
3550 err_disable_msix:
3551 	if (dev->flags & MLX4_FLAG_MSI_X)
3552 		pci_disable_msix(pdev);
3553 
3554 err_free_eq:
3555 	mlx4_free_eq_table(dev);
3556 
3557 err_master_mfunc:
3558 	if (mlx4_is_master(dev)) {
3559 		mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
3560 		mlx4_multi_func_cleanup(dev);
3561 	}
3562 
3563 	if (mlx4_is_slave(dev)) {
3564 		kfree(dev->caps.qp0_qkey);
3565 		kfree(dev->caps.qp0_tunnel);
3566 		kfree(dev->caps.qp0_proxy);
3567 		kfree(dev->caps.qp1_tunnel);
3568 		kfree(dev->caps.qp1_proxy);
3569 	}
3570 
3571 err_close:
3572 	mlx4_close_hca(dev);
3573 
3574 err_fw:
3575 	mlx4_close_fw(dev);
3576 
3577 err_mfunc:
3578 	if (mlx4_is_slave(dev))
3579 		mlx4_multi_func_cleanup(dev);
3580 
3581 err_cmd:
3582 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3583 
3584 err_sriov:
3585 	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3586 		pci_disable_sriov(pdev);
3587 		dev->flags &= ~MLX4_FLAG_SRIOV;
3588 	}
3589 
3590 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3591 		atomic_dec(&pf_loading);
3592 
3593 	kfree(priv->dev.dev_vfs);
3594 
3595 	if (!mlx4_is_slave(dev))
3596 		mlx4_free_ownership(dev);
3597 
3598 	kfree(dev_cap);
3599 	return err;
3600 }
3601 
3602 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3603 			   struct mlx4_priv *priv)
3604 {
3605 	int err;
3606 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3607 	int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3608 	const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3609 		{2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3610 	unsigned total_vfs = 0;
3611 	unsigned int i;
3612 
3613 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3614 
3615 	err = mlx4_pci_enable_device(&priv->dev);
3616 	if (err) {
3617 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3618 		return err;
3619 	}
3620 
3621 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3622 	 * per port, we must limit the number of VFs to 63 (since their are
3623 	 * 128 MACs)
3624 	 */
3625 	for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
3626 	     total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3627 		nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3628 		if (nvfs[i] < 0) {
3629 			dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3630 			err = -EINVAL;
3631 			goto err_disable_pdev;
3632 		}
3633 	}
3634 	for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
3635 	     i++) {
3636 		prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3637 		if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3638 			dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3639 			err = -EINVAL;
3640 			goto err_disable_pdev;
3641 		}
3642 	}
3643 	if (total_vfs > MLX4_MAX_NUM_VF) {
3644 		dev_err(&pdev->dev,
3645 			"Requested more VF's (%d) than allowed by hw (%d)\n",
3646 			total_vfs, MLX4_MAX_NUM_VF);
3647 		err = -EINVAL;
3648 		goto err_disable_pdev;
3649 	}
3650 
3651 	for (i = 0; i < MLX4_MAX_PORTS; i++) {
3652 		if (nvfs[i] + nvfs[2] > MLX4_MAX_NUM_VF_P_PORT) {
3653 			dev_err(&pdev->dev,
3654 				"Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3655 				nvfs[i] + nvfs[2], i + 1,
3656 				MLX4_MAX_NUM_VF_P_PORT);
3657 			err = -EINVAL;
3658 			goto err_disable_pdev;
3659 		}
3660 	}
3661 
3662 	/* Check for BARs. */
3663 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3664 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3665 		dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3666 			pci_dev_data, pci_resource_flags(pdev, 0));
3667 		err = -ENODEV;
3668 		goto err_disable_pdev;
3669 	}
3670 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3671 		dev_err(&pdev->dev, "Missing UAR, aborting\n");
3672 		err = -ENODEV;
3673 		goto err_disable_pdev;
3674 	}
3675 
3676 	err = pci_request_regions(pdev, DRV_NAME);
3677 	if (err) {
3678 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3679 		goto err_disable_pdev;
3680 	}
3681 
3682 	pci_set_master(pdev);
3683 
3684 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3685 	if (err) {
3686 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3687 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3688 		if (err) {
3689 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3690 			goto err_release_regions;
3691 		}
3692 	}
3693 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3694 	if (err) {
3695 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3696 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3697 		if (err) {
3698 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3699 			goto err_release_regions;
3700 		}
3701 	}
3702 
3703 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
3704 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3705 	/* Detect if this device is a virtual function */
3706 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3707 		/* When acting as pf, we normally skip vfs unless explicitly
3708 		 * requested to probe them.
3709 		 */
3710 		if (total_vfs) {
3711 			unsigned vfs_offset = 0;
3712 
3713 			for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
3714 			     vfs_offset + nvfs[i] < extended_func_num(pdev);
3715 			     vfs_offset += nvfs[i], i++)
3716 				;
3717 			if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
3718 				err = -ENODEV;
3719 				goto err_release_regions;
3720 			}
3721 			if ((extended_func_num(pdev) - vfs_offset)
3722 			    > prb_vf[i]) {
3723 				dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3724 					 extended_func_num(pdev));
3725 				err = -ENODEV;
3726 				goto err_release_regions;
3727 			}
3728 		}
3729 	}
3730 
3731 	err = mlx4_catas_init(&priv->dev);
3732 	if (err)
3733 		goto err_release_regions;
3734 
3735 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3736 	if (err)
3737 		goto err_catas;
3738 
3739 	return 0;
3740 
3741 err_catas:
3742 	mlx4_catas_end(&priv->dev);
3743 
3744 err_release_regions:
3745 	pci_release_regions(pdev);
3746 
3747 err_disable_pdev:
3748 	mlx4_pci_disable_device(&priv->dev);
3749 	pci_set_drvdata(pdev, NULL);
3750 	return err;
3751 }
3752 
3753 static int mlx4_devlink_port_type_set(struct devlink_port *devlink_port,
3754 				      enum devlink_port_type port_type)
3755 {
3756 	struct mlx4_port_info *info = container_of(devlink_port,
3757 						   struct mlx4_port_info,
3758 						   devlink_port);
3759 	enum mlx4_port_type mlx4_port_type;
3760 
3761 	switch (port_type) {
3762 	case DEVLINK_PORT_TYPE_AUTO:
3763 		mlx4_port_type = MLX4_PORT_TYPE_AUTO;
3764 		break;
3765 	case DEVLINK_PORT_TYPE_ETH:
3766 		mlx4_port_type = MLX4_PORT_TYPE_ETH;
3767 		break;
3768 	case DEVLINK_PORT_TYPE_IB:
3769 		mlx4_port_type = MLX4_PORT_TYPE_IB;
3770 		break;
3771 	default:
3772 		return -EOPNOTSUPP;
3773 	}
3774 
3775 	return __set_port_type(info, mlx4_port_type);
3776 }
3777 
3778 static const struct devlink_ops mlx4_devlink_ops = {
3779 	.port_type_set	= mlx4_devlink_port_type_set,
3780 };
3781 
3782 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3783 {
3784 	struct devlink *devlink;
3785 	struct mlx4_priv *priv;
3786 	struct mlx4_dev *dev;
3787 	int ret;
3788 
3789 	printk_once(KERN_INFO "%s", mlx4_version);
3790 
3791 	devlink = devlink_alloc(&mlx4_devlink_ops, sizeof(*priv));
3792 	if (!devlink)
3793 		return -ENOMEM;
3794 	priv = devlink_priv(devlink);
3795 
3796 	dev       = &priv->dev;
3797 	dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3798 	if (!dev->persist) {
3799 		ret = -ENOMEM;
3800 		goto err_devlink_free;
3801 	}
3802 	dev->persist->pdev = pdev;
3803 	dev->persist->dev = dev;
3804 	pci_set_drvdata(pdev, dev->persist);
3805 	priv->pci_dev_data = id->driver_data;
3806 	mutex_init(&dev->persist->device_state_mutex);
3807 	mutex_init(&dev->persist->interface_state_mutex);
3808 	mutex_init(&dev->persist->pci_status_mutex);
3809 
3810 	ret = devlink_register(devlink, &pdev->dev);
3811 	if (ret)
3812 		goto err_persist_free;
3813 
3814 	ret =  __mlx4_init_one(pdev, id->driver_data, priv);
3815 	if (ret)
3816 		goto err_devlink_unregister;
3817 
3818 	pci_save_state(pdev);
3819 	return 0;
3820 
3821 err_devlink_unregister:
3822 	devlink_unregister(devlink);
3823 err_persist_free:
3824 	kfree(dev->persist);
3825 err_devlink_free:
3826 	devlink_free(devlink);
3827 	return ret;
3828 }
3829 
3830 static void mlx4_clean_dev(struct mlx4_dev *dev)
3831 {
3832 	struct mlx4_dev_persistent *persist = dev->persist;
3833 	struct mlx4_priv *priv = mlx4_priv(dev);
3834 	unsigned long	flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
3835 
3836 	memset(priv, 0, sizeof(*priv));
3837 	priv->dev.persist = persist;
3838 	priv->dev.flags = flags;
3839 }
3840 
3841 static void mlx4_unload_one(struct pci_dev *pdev)
3842 {
3843 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3844 	struct mlx4_dev  *dev  = persist->dev;
3845 	struct mlx4_priv *priv = mlx4_priv(dev);
3846 	int               pci_dev_data;
3847 	int p, i;
3848 
3849 	if (priv->removed)
3850 		return;
3851 
3852 	/* saving current ports type for further use */
3853 	for (i = 0; i < dev->caps.num_ports; i++) {
3854 		dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
3855 		dev->persist->curr_port_poss_type[i] = dev->caps.
3856 						       possible_type[i + 1];
3857 	}
3858 
3859 	pci_dev_data = priv->pci_dev_data;
3860 
3861 	mlx4_stop_sense(dev);
3862 	mlx4_unregister_device(dev);
3863 
3864 	for (p = 1; p <= dev->caps.num_ports; p++) {
3865 		mlx4_cleanup_port_info(&priv->port[p]);
3866 		mlx4_CLOSE_PORT(dev, p);
3867 	}
3868 
3869 	if (mlx4_is_master(dev))
3870 		mlx4_free_resource_tracker(dev,
3871 					   RES_TR_FREE_SLAVES_ONLY);
3872 
3873 	mlx4_cleanup_default_counters(dev);
3874 	if (!mlx4_is_slave(dev))
3875 		mlx4_cleanup_counters_table(dev);
3876 	mlx4_cleanup_qp_table(dev);
3877 	mlx4_cleanup_srq_table(dev);
3878 	mlx4_cleanup_cq_table(dev);
3879 	mlx4_cmd_use_polling(dev);
3880 	mlx4_cleanup_eq_table(dev);
3881 	mlx4_cleanup_mcg_table(dev);
3882 	mlx4_cleanup_mr_table(dev);
3883 	mlx4_cleanup_xrcd_table(dev);
3884 	mlx4_cleanup_pd_table(dev);
3885 
3886 	if (mlx4_is_master(dev))
3887 		mlx4_free_resource_tracker(dev,
3888 					   RES_TR_FREE_STRUCTS_ONLY);
3889 
3890 	iounmap(priv->kar);
3891 	mlx4_uar_free(dev, &priv->driver_uar);
3892 	mlx4_cleanup_uar_table(dev);
3893 	if (!mlx4_is_slave(dev))
3894 		mlx4_clear_steering(dev);
3895 	mlx4_free_eq_table(dev);
3896 	if (mlx4_is_master(dev))
3897 		mlx4_multi_func_cleanup(dev);
3898 	mlx4_close_hca(dev);
3899 	mlx4_close_fw(dev);
3900 	if (mlx4_is_slave(dev))
3901 		mlx4_multi_func_cleanup(dev);
3902 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3903 
3904 	if (dev->flags & MLX4_FLAG_MSI_X)
3905 		pci_disable_msix(pdev);
3906 
3907 	if (!mlx4_is_slave(dev))
3908 		mlx4_free_ownership(dev);
3909 
3910 	kfree(dev->caps.qp0_qkey);
3911 	kfree(dev->caps.qp0_tunnel);
3912 	kfree(dev->caps.qp0_proxy);
3913 	kfree(dev->caps.qp1_tunnel);
3914 	kfree(dev->caps.qp1_proxy);
3915 	kfree(dev->dev_vfs);
3916 
3917 	mlx4_clean_dev(dev);
3918 	priv->pci_dev_data = pci_dev_data;
3919 	priv->removed = 1;
3920 }
3921 
3922 static void mlx4_remove_one(struct pci_dev *pdev)
3923 {
3924 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3925 	struct mlx4_dev  *dev  = persist->dev;
3926 	struct mlx4_priv *priv = mlx4_priv(dev);
3927 	struct devlink *devlink = priv_to_devlink(priv);
3928 	int active_vfs = 0;
3929 
3930 	mutex_lock(&persist->interface_state_mutex);
3931 	persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
3932 	mutex_unlock(&persist->interface_state_mutex);
3933 
3934 	/* Disabling SR-IOV is not allowed while there are active vf's */
3935 	if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
3936 		active_vfs = mlx4_how_many_lives_vf(dev);
3937 		if (active_vfs) {
3938 			pr_warn("Removing PF when there are active VF's !!\n");
3939 			pr_warn("Will not disable SR-IOV.\n");
3940 		}
3941 	}
3942 
3943 	/* device marked to be under deletion running now without the lock
3944 	 * letting other tasks to be terminated
3945 	 */
3946 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3947 		mlx4_unload_one(pdev);
3948 	else
3949 		mlx4_info(dev, "%s: interface is down\n", __func__);
3950 	mlx4_catas_end(dev);
3951 	if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
3952 		mlx4_warn(dev, "Disabling SR-IOV\n");
3953 		pci_disable_sriov(pdev);
3954 	}
3955 
3956 	pci_release_regions(pdev);
3957 	mlx4_pci_disable_device(dev);
3958 	devlink_unregister(devlink);
3959 	kfree(dev->persist);
3960 	devlink_free(devlink);
3961 	pci_set_drvdata(pdev, NULL);
3962 }
3963 
3964 static int restore_current_port_types(struct mlx4_dev *dev,
3965 				      enum mlx4_port_type *types,
3966 				      enum mlx4_port_type *poss_types)
3967 {
3968 	struct mlx4_priv *priv = mlx4_priv(dev);
3969 	int err, i;
3970 
3971 	mlx4_stop_sense(dev);
3972 
3973 	mutex_lock(&priv->port_mutex);
3974 	for (i = 0; i < dev->caps.num_ports; i++)
3975 		dev->caps.possible_type[i + 1] = poss_types[i];
3976 	err = mlx4_change_port_types(dev, types);
3977 	mlx4_start_sense(dev);
3978 	mutex_unlock(&priv->port_mutex);
3979 
3980 	return err;
3981 }
3982 
3983 int mlx4_restart_one(struct pci_dev *pdev)
3984 {
3985 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3986 	struct mlx4_dev	 *dev  = persist->dev;
3987 	struct mlx4_priv *priv = mlx4_priv(dev);
3988 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3989 	int pci_dev_data, err, total_vfs;
3990 
3991 	pci_dev_data = priv->pci_dev_data;
3992 	total_vfs = dev->persist->num_vfs;
3993 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
3994 
3995 	mlx4_unload_one(pdev);
3996 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
3997 	if (err) {
3998 		mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
3999 			 __func__, pci_name(pdev), err);
4000 		return err;
4001 	}
4002 
4003 	err = restore_current_port_types(dev, dev->persist->curr_port_type,
4004 					 dev->persist->curr_port_poss_type);
4005 	if (err)
4006 		mlx4_err(dev, "could not restore original port types (%d)\n",
4007 			 err);
4008 
4009 	return err;
4010 }
4011 
4012 static const struct pci_device_id mlx4_pci_table[] = {
4013 	/* MT25408 "Hermon" SDR */
4014 	{ PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4015 	/* MT25408 "Hermon" DDR */
4016 	{ PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4017 	/* MT25408 "Hermon" QDR */
4018 	{ PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4019 	/* MT25408 "Hermon" DDR PCIe gen2 */
4020 	{ PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4021 	/* MT25408 "Hermon" QDR PCIe gen2 */
4022 	{ PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4023 	/* MT25408 "Hermon" EN 10GigE */
4024 	{ PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4025 	/* MT25408 "Hermon" EN 10GigE PCIe gen2 */
4026 	{ PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4027 	/* MT25458 ConnectX EN 10GBASE-T 10GigE */
4028 	{ PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4029 	/* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
4030 	{ PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4031 	/* MT26468 ConnectX EN 10GigE PCIe gen2*/
4032 	{ PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4033 	/* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
4034 	{ PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4035 	/* MT26478 ConnectX2 40GigE PCIe gen2 */
4036 	{ PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4037 	/* MT25400 Family [ConnectX-2 Virtual Function] */
4038 	{ PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
4039 	/* MT27500 Family [ConnectX-3] */
4040 	{ PCI_VDEVICE(MELLANOX, 0x1003), 0 },
4041 	/* MT27500 Family [ConnectX-3 Virtual Function] */
4042 	{ PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
4043 	{ PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
4044 	{ PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
4045 	{ PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
4046 	{ PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
4047 	{ PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
4048 	{ PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
4049 	{ PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
4050 	{ PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
4051 	{ PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
4052 	{ PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
4053 	{ PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
4054 	{ PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
4055 	{ 0, }
4056 };
4057 
4058 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
4059 
4060 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
4061 					      pci_channel_state_t state)
4062 {
4063 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4064 
4065 	mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
4066 	mlx4_enter_error_state(persist);
4067 
4068 	mutex_lock(&persist->interface_state_mutex);
4069 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4070 		mlx4_unload_one(pdev);
4071 
4072 	mutex_unlock(&persist->interface_state_mutex);
4073 	if (state == pci_channel_io_perm_failure)
4074 		return PCI_ERS_RESULT_DISCONNECT;
4075 
4076 	mlx4_pci_disable_device(persist->dev);
4077 	return PCI_ERS_RESULT_NEED_RESET;
4078 }
4079 
4080 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
4081 {
4082 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4083 	struct mlx4_dev	 *dev  = persist->dev;
4084 	int err;
4085 
4086 	mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
4087 	err = mlx4_pci_enable_device(dev);
4088 	if (err) {
4089 		mlx4_err(dev, "Can not re-enable device, err=%d\n", err);
4090 		return PCI_ERS_RESULT_DISCONNECT;
4091 	}
4092 
4093 	pci_set_master(pdev);
4094 	pci_restore_state(pdev);
4095 	pci_save_state(pdev);
4096 	return PCI_ERS_RESULT_RECOVERED;
4097 }
4098 
4099 static void mlx4_pci_resume(struct pci_dev *pdev)
4100 {
4101 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4102 	struct mlx4_dev	 *dev  = persist->dev;
4103 	struct mlx4_priv *priv = mlx4_priv(dev);
4104 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4105 	int total_vfs;
4106 	int err;
4107 
4108 	mlx4_err(dev, "%s was called\n", __func__);
4109 	total_vfs = dev->persist->num_vfs;
4110 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4111 
4112 	mutex_lock(&persist->interface_state_mutex);
4113 	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4114 		err = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
4115 				    priv, 1);
4116 		if (err) {
4117 			mlx4_err(dev, "%s: mlx4_load_one failed, err=%d\n",
4118 				 __func__,  err);
4119 			goto end;
4120 		}
4121 
4122 		err = restore_current_port_types(dev, dev->persist->
4123 						 curr_port_type, dev->persist->
4124 						 curr_port_poss_type);
4125 		if (err)
4126 			mlx4_err(dev, "could not restore original port types (%d)\n", err);
4127 	}
4128 end:
4129 	mutex_unlock(&persist->interface_state_mutex);
4130 
4131 }
4132 
4133 static void mlx4_shutdown(struct pci_dev *pdev)
4134 {
4135 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4136 
4137 	mlx4_info(persist->dev, "mlx4_shutdown was called\n");
4138 	mutex_lock(&persist->interface_state_mutex);
4139 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP) {
4140 		/* Notify mlx4 clients that the kernel is being shut down */
4141 		persist->interface_state |= MLX4_INTERFACE_STATE_SHUTDOWN;
4142 		mlx4_unload_one(pdev);
4143 	}
4144 	mutex_unlock(&persist->interface_state_mutex);
4145 }
4146 
4147 static const struct pci_error_handlers mlx4_err_handler = {
4148 	.error_detected = mlx4_pci_err_detected,
4149 	.slot_reset     = mlx4_pci_slot_reset,
4150 	.resume		= mlx4_pci_resume,
4151 };
4152 
4153 static struct pci_driver mlx4_driver = {
4154 	.name		= DRV_NAME,
4155 	.id_table	= mlx4_pci_table,
4156 	.probe		= mlx4_init_one,
4157 	.shutdown	= mlx4_shutdown,
4158 	.remove		= mlx4_remove_one,
4159 	.err_handler    = &mlx4_err_handler,
4160 };
4161 
4162 static int __init mlx4_verify_params(void)
4163 {
4164 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
4165 		pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
4166 		return -1;
4167 	}
4168 
4169 	if (log_num_vlan != 0)
4170 		pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
4171 			MLX4_LOG_NUM_VLANS);
4172 
4173 	if (use_prio != 0)
4174 		pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
4175 
4176 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
4177 		pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
4178 			log_mtts_per_seg);
4179 		return -1;
4180 	}
4181 
4182 	/* Check if module param for ports type has legal combination */
4183 	if (port_type_array[0] == false && port_type_array[1] == true) {
4184 		pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
4185 		port_type_array[0] = true;
4186 	}
4187 
4188 	if (mlx4_log_num_mgm_entry_size < -7 ||
4189 	    (mlx4_log_num_mgm_entry_size > 0 &&
4190 	     (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
4191 	      mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
4192 		pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
4193 			mlx4_log_num_mgm_entry_size,
4194 			MLX4_MIN_MGM_LOG_ENTRY_SIZE,
4195 			MLX4_MAX_MGM_LOG_ENTRY_SIZE);
4196 		return -1;
4197 	}
4198 
4199 	return 0;
4200 }
4201 
4202 static int __init mlx4_init(void)
4203 {
4204 	int ret;
4205 
4206 	if (mlx4_verify_params())
4207 		return -EINVAL;
4208 
4209 
4210 	mlx4_wq = create_singlethread_workqueue("mlx4");
4211 	if (!mlx4_wq)
4212 		return -ENOMEM;
4213 
4214 	ret = pci_register_driver(&mlx4_driver);
4215 	if (ret < 0)
4216 		destroy_workqueue(mlx4_wq);
4217 	return ret < 0 ? ret : 0;
4218 }
4219 
4220 static void __exit mlx4_cleanup(void)
4221 {
4222 	pci_unregister_driver(&mlx4_driver);
4223 	destroy_workqueue(mlx4_wq);
4224 }
4225 
4226 module_init(mlx4_init);
4227 module_exit(mlx4_cleanup);
4228