1 /*
2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4  * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
5  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * OpenIB.org BSD license below:
12  *
13  *     Redistribution and use in source and binary forms, with or
14  *     without modification, are permitted provided that the following
15  *     conditions are met:
16  *
17  *      - Redistributions of source code must retain the above
18  *        copyright notice, this list of conditions and the following
19  *        disclaimer.
20  *
21  *      - Redistributions in binary form must reproduce the above
22  *        copyright notice, this list of conditions and the following
23  *        disclaimer in the documentation and/or other materials
24  *        provided with the distribution.
25  *
26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33  * SOFTWARE.
34  */
35 
36 #include <linux/module.h>
37 #include <linux/init.h>
38 #include <linux/errno.h>
39 #include <linux/pci.h>
40 #include <linux/dma-mapping.h>
41 #include <linux/slab.h>
42 #include <linux/io-mapping.h>
43 #include <linux/delay.h>
44 #include <linux/kmod.h>
45 #include <linux/etherdevice.h>
46 #include <net/devlink.h>
47 
48 #include <linux/mlx4/device.h>
49 #include <linux/mlx4/doorbell.h>
50 
51 #include "mlx4.h"
52 #include "fw.h"
53 #include "icm.h"
54 
55 MODULE_AUTHOR("Roland Dreier");
56 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
57 MODULE_LICENSE("Dual BSD/GPL");
58 MODULE_VERSION(DRV_VERSION);
59 
60 struct workqueue_struct *mlx4_wq;
61 
62 #ifdef CONFIG_MLX4_DEBUG
63 
64 int mlx4_debug_level = 0;
65 module_param_named(debug_level, mlx4_debug_level, int, 0644);
66 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
67 
68 #endif /* CONFIG_MLX4_DEBUG */
69 
70 #ifdef CONFIG_PCI_MSI
71 
72 static int msi_x = 1;
73 module_param(msi_x, int, 0444);
74 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
75 
76 #else /* CONFIG_PCI_MSI */
77 
78 #define msi_x (0)
79 
80 #endif /* CONFIG_PCI_MSI */
81 
82 static uint8_t num_vfs[3] = {0, 0, 0};
83 static int num_vfs_argc;
84 module_param_array(num_vfs, byte , &num_vfs_argc, 0444);
85 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
86 			  "num_vfs=port1,port2,port1+2");
87 
88 static uint8_t probe_vf[3] = {0, 0, 0};
89 static int probe_vfs_argc;
90 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
91 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
92 			   "probe_vf=port1,port2,port1+2");
93 
94 int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
95 module_param_named(log_num_mgm_entry_size,
96 			mlx4_log_num_mgm_entry_size, int, 0444);
97 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
98 					 " of qp per mcg, for example:"
99 					 " 10 gives 248.range: 7 <="
100 					 " log_num_mgm_entry_size <= 12."
101 					 " To activate device managed"
102 					 " flow steering when available, set to -1");
103 
104 static bool enable_64b_cqe_eqe = true;
105 module_param(enable_64b_cqe_eqe, bool, 0444);
106 MODULE_PARM_DESC(enable_64b_cqe_eqe,
107 		 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
108 
109 static bool enable_4k_uar;
110 module_param(enable_4k_uar, bool, 0444);
111 MODULE_PARM_DESC(enable_4k_uar,
112 		 "Enable using 4K UAR. Should not be enabled if have VFs which do not support 4K UARs (default: false)");
113 
114 #define PF_CONTEXT_BEHAVIOUR_MASK	(MLX4_FUNC_CAP_64B_EQE_CQE | \
115 					 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
116 					 MLX4_FUNC_CAP_DMFS_A0_STATIC)
117 
118 #define RESET_PERSIST_MASK_FLAGS	(MLX4_FLAG_SRIOV)
119 
120 static char mlx4_version[] =
121 	DRV_NAME ": Mellanox ConnectX core driver v"
122 	DRV_VERSION " (" DRV_RELDATE ")\n";
123 
124 static struct mlx4_profile default_profile = {
125 	.num_qp		= 1 << 18,
126 	.num_srq	= 1 << 16,
127 	.rdmarc_per_qp	= 1 << 4,
128 	.num_cq		= 1 << 16,
129 	.num_mcg	= 1 << 13,
130 	.num_mpt	= 1 << 19,
131 	.num_mtt	= 1 << 20, /* It is really num mtt segements */
132 };
133 
134 static struct mlx4_profile low_mem_profile = {
135 	.num_qp		= 1 << 17,
136 	.num_srq	= 1 << 6,
137 	.rdmarc_per_qp	= 1 << 4,
138 	.num_cq		= 1 << 8,
139 	.num_mcg	= 1 << 8,
140 	.num_mpt	= 1 << 9,
141 	.num_mtt	= 1 << 7,
142 };
143 
144 static int log_num_mac = 7;
145 module_param_named(log_num_mac, log_num_mac, int, 0444);
146 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
147 
148 static int log_num_vlan;
149 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
150 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
151 /* Log2 max number of VLANs per ETH port (0-7) */
152 #define MLX4_LOG_NUM_VLANS 7
153 #define MLX4_MIN_LOG_NUM_VLANS 0
154 #define MLX4_MIN_LOG_NUM_MAC 1
155 
156 static bool use_prio;
157 module_param_named(use_prio, use_prio, bool, 0444);
158 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
159 
160 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
161 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
162 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
163 
164 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
165 static int arr_argc = 2;
166 module_param_array(port_type_array, int, &arr_argc, 0444);
167 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
168 				"1 for IB, 2 for Ethernet");
169 
170 struct mlx4_port_config {
171 	struct list_head list;
172 	enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
173 	struct pci_dev *pdev;
174 };
175 
176 static atomic_t pf_loading = ATOMIC_INIT(0);
177 
178 static inline void mlx4_set_num_reserved_uars(struct mlx4_dev *dev,
179 					      struct mlx4_dev_cap *dev_cap)
180 {
181 	/* The reserved_uars is calculated by system page size unit.
182 	 * Therefore, adjustment is added when the uar page size is less
183 	 * than the system page size
184 	 */
185 	dev->caps.reserved_uars	=
186 		max_t(int,
187 		      mlx4_get_num_reserved_uar(dev),
188 		      dev_cap->reserved_uars /
189 			(1 << (PAGE_SHIFT - dev->uar_page_shift)));
190 }
191 
192 int mlx4_check_port_params(struct mlx4_dev *dev,
193 			   enum mlx4_port_type *port_type)
194 {
195 	int i;
196 
197 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
198 		for (i = 0; i < dev->caps.num_ports - 1; i++) {
199 			if (port_type[i] != port_type[i + 1]) {
200 				mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
201 				return -EINVAL;
202 			}
203 		}
204 	}
205 
206 	for (i = 0; i < dev->caps.num_ports; i++) {
207 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
208 			mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
209 				 i + 1);
210 			return -EINVAL;
211 		}
212 	}
213 	return 0;
214 }
215 
216 static void mlx4_set_port_mask(struct mlx4_dev *dev)
217 {
218 	int i;
219 
220 	for (i = 1; i <= dev->caps.num_ports; ++i)
221 		dev->caps.port_mask[i] = dev->caps.port_type[i];
222 }
223 
224 enum {
225 	MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
226 };
227 
228 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
229 {
230 	int err = 0;
231 	struct mlx4_func func;
232 
233 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
234 		err = mlx4_QUERY_FUNC(dev, &func, 0);
235 		if (err) {
236 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
237 			return err;
238 		}
239 		dev_cap->max_eqs = func.max_eq;
240 		dev_cap->reserved_eqs = func.rsvd_eqs;
241 		dev_cap->reserved_uars = func.rsvd_uars;
242 		err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
243 	}
244 	return err;
245 }
246 
247 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
248 {
249 	struct mlx4_caps *dev_cap = &dev->caps;
250 
251 	/* FW not supporting or cancelled by user */
252 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
253 	    !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
254 		return;
255 
256 	/* Must have 64B CQE_EQE enabled by FW to use bigger stride
257 	 * When FW has NCSI it may decide not to report 64B CQE/EQEs
258 	 */
259 	if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
260 	    !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
261 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
262 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
263 		return;
264 	}
265 
266 	if (cache_line_size() == 128 || cache_line_size() == 256) {
267 		mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
268 		/* Changing the real data inside CQE size to 32B */
269 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
270 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
271 
272 		if (mlx4_is_master(dev))
273 			dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
274 	} else {
275 		if (cache_line_size() != 32  && cache_line_size() != 64)
276 			mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
277 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
278 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
279 	}
280 }
281 
282 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
283 			  struct mlx4_port_cap *port_cap)
284 {
285 	dev->caps.vl_cap[port]	    = port_cap->max_vl;
286 	dev->caps.ib_mtu_cap[port]	    = port_cap->ib_mtu;
287 	dev->phys_caps.gid_phys_table_len[port]  = port_cap->max_gids;
288 	dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
289 	/* set gid and pkey table operating lengths by default
290 	 * to non-sriov values
291 	 */
292 	dev->caps.gid_table_len[port]  = port_cap->max_gids;
293 	dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
294 	dev->caps.port_width_cap[port] = port_cap->max_port_width;
295 	dev->caps.eth_mtu_cap[port]    = port_cap->eth_mtu;
296 	dev->caps.max_tc_eth	       = port_cap->max_tc_eth;
297 	dev->caps.def_mac[port]        = port_cap->def_mac;
298 	dev->caps.supported_type[port] = port_cap->supported_port_types;
299 	dev->caps.suggested_type[port] = port_cap->suggested_type;
300 	dev->caps.default_sense[port] = port_cap->default_sense;
301 	dev->caps.trans_type[port]	    = port_cap->trans_type;
302 	dev->caps.vendor_oui[port]     = port_cap->vendor_oui;
303 	dev->caps.wavelength[port]     = port_cap->wavelength;
304 	dev->caps.trans_code[port]     = port_cap->trans_code;
305 
306 	return 0;
307 }
308 
309 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
310 			 struct mlx4_port_cap *port_cap)
311 {
312 	int err = 0;
313 
314 	err = mlx4_QUERY_PORT(dev, port, port_cap);
315 
316 	if (err)
317 		mlx4_err(dev, "QUERY_PORT command failed.\n");
318 
319 	return err;
320 }
321 
322 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
323 {
324 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
325 		return;
326 
327 	if (mlx4_is_mfunc(dev)) {
328 		mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
329 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
330 		return;
331 	}
332 
333 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
334 		mlx4_dbg(dev,
335 			 "Keep FCS is not supported - Disabling Ignore FCS");
336 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
337 		return;
338 	}
339 }
340 
341 #define MLX4_A0_STEERING_TABLE_SIZE	256
342 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
343 {
344 	int err;
345 	int i;
346 
347 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
348 	if (err) {
349 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
350 		return err;
351 	}
352 	mlx4_dev_cap_dump(dev, dev_cap);
353 
354 	if (dev_cap->min_page_sz > PAGE_SIZE) {
355 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
356 			 dev_cap->min_page_sz, PAGE_SIZE);
357 		return -ENODEV;
358 	}
359 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
360 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
361 			 dev_cap->num_ports, MLX4_MAX_PORTS);
362 		return -ENODEV;
363 	}
364 
365 	if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
366 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
367 			 dev_cap->uar_size,
368 			 (unsigned long long)
369 			 pci_resource_len(dev->persist->pdev, 2));
370 		return -ENODEV;
371 	}
372 
373 	dev->caps.num_ports	     = dev_cap->num_ports;
374 	dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
375 	dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
376 				      dev->caps.num_sys_eqs :
377 				      MLX4_MAX_EQ_NUM;
378 	for (i = 1; i <= dev->caps.num_ports; ++i) {
379 		err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
380 		if (err) {
381 			mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
382 			return err;
383 		}
384 	}
385 
386 	dev->caps.uar_page_size	     = PAGE_SIZE;
387 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
388 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
389 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
390 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
391 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
392 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
393 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
394 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
395 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
396 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
397 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
398 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
399 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
400 	/*
401 	 * Subtract 1 from the limit because we need to allocate a
402 	 * spare CQE so the HCA HW can tell the difference between an
403 	 * empty CQ and a full CQ.
404 	 */
405 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
406 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
407 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
408 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
409 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
410 
411 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
412 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
413 					dev_cap->reserved_xrcds : 0;
414 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
415 					dev_cap->max_xrcds : 0;
416 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
417 
418 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
419 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
420 	dev->caps.flags		     = dev_cap->flags;
421 	dev->caps.flags2	     = dev_cap->flags2;
422 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
423 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
424 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
425 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
426 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
427 
428 	/* Save uar page shift */
429 	if (!mlx4_is_slave(dev)) {
430 		/* Virtual PCI function needs to determine UAR page size from
431 		 * firmware. Only master PCI function can set the uar page size
432 		 */
433 		if (enable_4k_uar)
434 			dev->uar_page_shift = DEFAULT_UAR_PAGE_SHIFT;
435 		else
436 			dev->uar_page_shift = PAGE_SHIFT;
437 
438 		mlx4_set_num_reserved_uars(dev, dev_cap);
439 	}
440 
441 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN) {
442 		struct mlx4_init_hca_param hca_param;
443 
444 		memset(&hca_param, 0, sizeof(hca_param));
445 		err = mlx4_QUERY_HCA(dev, &hca_param);
446 		/* Turn off PHV_EN flag in case phv_check_en is set.
447 		 * phv_check_en is a HW check that parse the packet and verify
448 		 * phv bit was reported correctly in the wqe. To allow QinQ
449 		 * PHV_EN flag should be set and phv_check_en must be cleared
450 		 * otherwise QinQ packets will be drop by the HW.
451 		 */
452 		if (err || hca_param.phv_check_en)
453 			dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_PHV_EN;
454 	}
455 
456 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
457 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
458 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
459 	/* Don't do sense port on multifunction devices (for now at least) */
460 	if (mlx4_is_mfunc(dev))
461 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
462 
463 	if (mlx4_low_memory_profile()) {
464 		dev->caps.log_num_macs  = MLX4_MIN_LOG_NUM_MAC;
465 		dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
466 	} else {
467 		dev->caps.log_num_macs  = log_num_mac;
468 		dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
469 	}
470 
471 	for (i = 1; i <= dev->caps.num_ports; ++i) {
472 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
473 		if (dev->caps.supported_type[i]) {
474 			/* if only ETH is supported - assign ETH */
475 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
476 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
477 			/* if only IB is supported, assign IB */
478 			else if (dev->caps.supported_type[i] ==
479 				 MLX4_PORT_TYPE_IB)
480 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
481 			else {
482 				/* if IB and ETH are supported, we set the port
483 				 * type according to user selection of port type;
484 				 * if user selected none, take the FW hint */
485 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
486 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
487 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
488 				else
489 					dev->caps.port_type[i] = port_type_array[i - 1];
490 			}
491 		}
492 		/*
493 		 * Link sensing is allowed on the port if 3 conditions are true:
494 		 * 1. Both protocols are supported on the port.
495 		 * 2. Different types are supported on the port
496 		 * 3. FW declared that it supports link sensing
497 		 */
498 		mlx4_priv(dev)->sense.sense_allowed[i] =
499 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
500 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
501 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
502 
503 		/*
504 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
505 		 * and perform sense_port FW command to try and set the correct
506 		 * port type from beginning
507 		 */
508 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
509 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
510 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
511 			mlx4_SENSE_PORT(dev, i, &sensed_port);
512 			if (sensed_port != MLX4_PORT_TYPE_NONE)
513 				dev->caps.port_type[i] = sensed_port;
514 		} else {
515 			dev->caps.possible_type[i] = dev->caps.port_type[i];
516 		}
517 
518 		if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
519 			dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
520 			mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
521 				  i, 1 << dev->caps.log_num_macs);
522 		}
523 		if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
524 			dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
525 			mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
526 				  i, 1 << dev->caps.log_num_vlans);
527 		}
528 	}
529 
530 	if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
531 	    (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
532 	    (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
533 		mlx4_warn(dev,
534 			  "Granular QoS per VF not supported with IB/Eth configuration\n");
535 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
536 	}
537 
538 	dev->caps.max_counters = dev_cap->max_counters;
539 
540 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
541 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
542 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
543 		(1 << dev->caps.log_num_macs) *
544 		(1 << dev->caps.log_num_vlans) *
545 		dev->caps.num_ports;
546 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
547 
548 	if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
549 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
550 		dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
551 	else
552 		dev->caps.dmfs_high_rate_qpn_base =
553 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
554 
555 	if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
556 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
557 		dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
558 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
559 		dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
560 	} else {
561 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
562 		dev->caps.dmfs_high_rate_qpn_base =
563 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
564 		dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
565 	}
566 
567 	dev->caps.rl_caps = dev_cap->rl_caps;
568 
569 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
570 		dev->caps.dmfs_high_rate_qpn_range;
571 
572 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
573 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
574 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
575 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
576 
577 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
578 
579 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
580 		if (dev_cap->flags &
581 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
582 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
583 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
584 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
585 		}
586 
587 		if (dev_cap->flags2 &
588 		    (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
589 		     MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
590 			mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
591 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
592 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
593 		}
594 	}
595 
596 	if ((dev->caps.flags &
597 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
598 	    mlx4_is_master(dev))
599 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
600 
601 	if (!mlx4_is_slave(dev)) {
602 		mlx4_enable_cqe_eqe_stride(dev);
603 		dev->caps.alloc_res_qp_mask =
604 			(dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
605 			MLX4_RESERVE_A0_QP;
606 
607 		if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
608 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
609 			mlx4_warn(dev, "Old device ETS support detected\n");
610 			mlx4_warn(dev, "Consider upgrading device FW.\n");
611 			dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
612 		}
613 
614 	} else {
615 		dev->caps.alloc_res_qp_mask = 0;
616 	}
617 
618 	mlx4_enable_ignore_fcs(dev);
619 
620 	return 0;
621 }
622 
623 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
624 				       enum pci_bus_speed *speed,
625 				       enum pcie_link_width *width)
626 {
627 	u32 lnkcap1, lnkcap2;
628 	int err1, err2;
629 
630 #define  PCIE_MLW_CAP_SHIFT 4	/* start of MLW mask in link capabilities */
631 
632 	*speed = PCI_SPEED_UNKNOWN;
633 	*width = PCIE_LNK_WIDTH_UNKNOWN;
634 
635 	err1 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP,
636 					  &lnkcap1);
637 	err2 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP2,
638 					  &lnkcap2);
639 	if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
640 		if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
641 			*speed = PCIE_SPEED_8_0GT;
642 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
643 			*speed = PCIE_SPEED_5_0GT;
644 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
645 			*speed = PCIE_SPEED_2_5GT;
646 	}
647 	if (!err1) {
648 		*width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
649 		if (!lnkcap2) { /* pre-r3.0 */
650 			if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
651 				*speed = PCIE_SPEED_5_0GT;
652 			else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
653 				*speed = PCIE_SPEED_2_5GT;
654 		}
655 	}
656 
657 	if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
658 		return err1 ? err1 :
659 			err2 ? err2 : -EINVAL;
660 	}
661 	return 0;
662 }
663 
664 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
665 {
666 	enum pcie_link_width width, width_cap;
667 	enum pci_bus_speed speed, speed_cap;
668 	int err;
669 
670 #define PCIE_SPEED_STR(speed) \
671 	(speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
672 	 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
673 	 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
674 	 "Unknown")
675 
676 	err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
677 	if (err) {
678 		mlx4_warn(dev,
679 			  "Unable to determine PCIe device BW capabilities\n");
680 		return;
681 	}
682 
683 	err = pcie_get_minimum_link(dev->persist->pdev, &speed, &width);
684 	if (err || speed == PCI_SPEED_UNKNOWN ||
685 	    width == PCIE_LNK_WIDTH_UNKNOWN) {
686 		mlx4_warn(dev,
687 			  "Unable to determine PCI device chain minimum BW\n");
688 		return;
689 	}
690 
691 	if (width != width_cap || speed != speed_cap)
692 		mlx4_warn(dev,
693 			  "PCIe BW is different than device's capability\n");
694 
695 	mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
696 		  PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
697 	mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
698 		  width, width_cap);
699 	return;
700 }
701 
702 /*The function checks if there are live vf, return the num of them*/
703 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
704 {
705 	struct mlx4_priv *priv = mlx4_priv(dev);
706 	struct mlx4_slave_state *s_state;
707 	int i;
708 	int ret = 0;
709 
710 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
711 		s_state = &priv->mfunc.master.slave_state[i];
712 		if (s_state->active && s_state->last_cmd !=
713 		    MLX4_COMM_CMD_RESET) {
714 			mlx4_warn(dev, "%s: slave: %d is still active\n",
715 				  __func__, i);
716 			ret++;
717 		}
718 	}
719 	return ret;
720 }
721 
722 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
723 {
724 	u32 qk = MLX4_RESERVED_QKEY_BASE;
725 
726 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
727 	    qpn < dev->phys_caps.base_proxy_sqpn)
728 		return -EINVAL;
729 
730 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
731 		/* tunnel qp */
732 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
733 	else
734 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
735 	*qkey = qk;
736 	return 0;
737 }
738 EXPORT_SYMBOL(mlx4_get_parav_qkey);
739 
740 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
741 {
742 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
743 
744 	if (!mlx4_is_master(dev))
745 		return;
746 
747 	priv->virt2phys_pkey[slave][port - 1][i] = val;
748 }
749 EXPORT_SYMBOL(mlx4_sync_pkey_table);
750 
751 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
752 {
753 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
754 
755 	if (!mlx4_is_master(dev))
756 		return;
757 
758 	priv->slave_node_guids[slave] = guid;
759 }
760 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
761 
762 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
763 {
764 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
765 
766 	if (!mlx4_is_master(dev))
767 		return 0;
768 
769 	return priv->slave_node_guids[slave];
770 }
771 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
772 
773 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
774 {
775 	struct mlx4_priv *priv = mlx4_priv(dev);
776 	struct mlx4_slave_state *s_slave;
777 
778 	if (!mlx4_is_master(dev))
779 		return 0;
780 
781 	s_slave = &priv->mfunc.master.slave_state[slave];
782 	return !!s_slave->active;
783 }
784 EXPORT_SYMBOL(mlx4_is_slave_active);
785 
786 void mlx4_handle_eth_header_mcast_prio(struct mlx4_net_trans_rule_hw_ctrl *ctrl,
787 				       struct _rule_hw *eth_header)
788 {
789 	if (is_multicast_ether_addr(eth_header->eth.dst_mac) ||
790 	    is_broadcast_ether_addr(eth_header->eth.dst_mac)) {
791 		struct mlx4_net_trans_rule_hw_eth *eth =
792 			(struct mlx4_net_trans_rule_hw_eth *)eth_header;
793 		struct _rule_hw *next_rule = (struct _rule_hw *)(eth + 1);
794 		bool last_rule = next_rule->size == 0 && next_rule->id == 0 &&
795 			next_rule->rsvd == 0;
796 
797 		if (last_rule)
798 			ctrl->prio = cpu_to_be16(MLX4_DOMAIN_NIC);
799 	}
800 }
801 EXPORT_SYMBOL(mlx4_handle_eth_header_mcast_prio);
802 
803 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
804 				       struct mlx4_dev_cap *dev_cap,
805 				       struct mlx4_init_hca_param *hca_param)
806 {
807 	dev->caps.steering_mode = hca_param->steering_mode;
808 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
809 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
810 		dev->caps.fs_log_max_ucast_qp_range_size =
811 			dev_cap->fs_log_max_ucast_qp_range_size;
812 	} else
813 		dev->caps.num_qp_per_mgm =
814 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
815 
816 	mlx4_dbg(dev, "Steering mode is: %s\n",
817 		 mlx4_steering_mode_str(dev->caps.steering_mode));
818 }
819 
820 static int mlx4_slave_cap(struct mlx4_dev *dev)
821 {
822 	int			   err;
823 	u32			   page_size;
824 	struct mlx4_dev_cap	   dev_cap;
825 	struct mlx4_func_cap	   func_cap;
826 	struct mlx4_init_hca_param hca_param;
827 	u8			   i;
828 
829 	memset(&hca_param, 0, sizeof(hca_param));
830 	err = mlx4_QUERY_HCA(dev, &hca_param);
831 	if (err) {
832 		mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
833 		return err;
834 	}
835 
836 	/* fail if the hca has an unknown global capability
837 	 * at this time global_caps should be always zeroed
838 	 */
839 	if (hca_param.global_caps) {
840 		mlx4_err(dev, "Unknown hca global capabilities\n");
841 		return -ENOSYS;
842 	}
843 
844 	mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
845 
846 	dev->caps.hca_core_clock = hca_param.hca_core_clock;
847 
848 	memset(&dev_cap, 0, sizeof(dev_cap));
849 	dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
850 	err = mlx4_dev_cap(dev, &dev_cap);
851 	if (err) {
852 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
853 		return err;
854 	}
855 
856 	err = mlx4_QUERY_FW(dev);
857 	if (err)
858 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
859 
860 	page_size = ~dev->caps.page_size_cap + 1;
861 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
862 	if (page_size > PAGE_SIZE) {
863 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
864 			 page_size, PAGE_SIZE);
865 		return -ENODEV;
866 	}
867 
868 	/* Set uar_page_shift for VF */
869 	dev->uar_page_shift = hca_param.uar_page_sz + 12;
870 
871 	/* Make sure the master uar page size is valid */
872 	if (dev->uar_page_shift > PAGE_SHIFT) {
873 		mlx4_err(dev,
874 			 "Invalid configuration: uar page size is larger than system page size\n");
875 		return  -ENODEV;
876 	}
877 
878 	/* Set reserved_uars based on the uar_page_shift */
879 	mlx4_set_num_reserved_uars(dev, &dev_cap);
880 
881 	/* Although uar page size in FW differs from system page size,
882 	 * upper software layers (mlx4_ib, mlx4_en and part of mlx4_core)
883 	 * still works with assumption that uar page size == system page size
884 	 */
885 	dev->caps.uar_page_size = PAGE_SIZE;
886 
887 	memset(&func_cap, 0, sizeof(func_cap));
888 	err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
889 	if (err) {
890 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
891 			 err);
892 		return err;
893 	}
894 
895 	if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
896 	    PF_CONTEXT_BEHAVIOUR_MASK) {
897 		mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
898 			 func_cap.pf_context_behaviour, PF_CONTEXT_BEHAVIOUR_MASK);
899 		return -ENOSYS;
900 	}
901 
902 	dev->caps.num_ports		= func_cap.num_ports;
903 	dev->quotas.qp			= func_cap.qp_quota;
904 	dev->quotas.srq			= func_cap.srq_quota;
905 	dev->quotas.cq			= func_cap.cq_quota;
906 	dev->quotas.mpt			= func_cap.mpt_quota;
907 	dev->quotas.mtt			= func_cap.mtt_quota;
908 	dev->caps.num_qps		= 1 << hca_param.log_num_qps;
909 	dev->caps.num_srqs		= 1 << hca_param.log_num_srqs;
910 	dev->caps.num_cqs		= 1 << hca_param.log_num_cqs;
911 	dev->caps.num_mpts		= 1 << hca_param.log_mpt_sz;
912 	dev->caps.num_eqs		= func_cap.max_eq;
913 	dev->caps.reserved_eqs		= func_cap.reserved_eq;
914 	dev->caps.reserved_lkey		= func_cap.reserved_lkey;
915 	dev->caps.num_pds               = MLX4_NUM_PDS;
916 	dev->caps.num_mgms              = 0;
917 	dev->caps.num_amgms             = 0;
918 
919 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
920 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
921 			 dev->caps.num_ports, MLX4_MAX_PORTS);
922 		return -ENODEV;
923 	}
924 
925 	mlx4_replace_zero_macs(dev);
926 
927 	dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
928 	dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
929 	dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
930 	dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
931 	dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
932 
933 	if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
934 	    !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
935 	    !dev->caps.qp0_qkey) {
936 		err = -ENOMEM;
937 		goto err_mem;
938 	}
939 
940 	for (i = 1; i <= dev->caps.num_ports; ++i) {
941 		err = mlx4_QUERY_FUNC_CAP(dev, i, &func_cap);
942 		if (err) {
943 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
944 				 i, err);
945 			goto err_mem;
946 		}
947 		dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
948 		dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
949 		dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
950 		dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
951 		dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
952 		dev->caps.port_mask[i] = dev->caps.port_type[i];
953 		dev->caps.phys_port_id[i] = func_cap.phys_port_id;
954 		err = mlx4_get_slave_pkey_gid_tbl_len(dev, i,
955 						      &dev->caps.gid_table_len[i],
956 						      &dev->caps.pkey_table_len[i]);
957 		if (err)
958 			goto err_mem;
959 	}
960 
961 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
962 				       dev->caps.reserved_uars) >
963 				       pci_resource_len(dev->persist->pdev,
964 							2)) {
965 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
966 			 dev->caps.uar_page_size * dev->caps.num_uars,
967 			 (unsigned long long)
968 			 pci_resource_len(dev->persist->pdev, 2));
969 		err = -ENOMEM;
970 		goto err_mem;
971 	}
972 
973 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
974 		dev->caps.eqe_size   = 64;
975 		dev->caps.eqe_factor = 1;
976 	} else {
977 		dev->caps.eqe_size   = 32;
978 		dev->caps.eqe_factor = 0;
979 	}
980 
981 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
982 		dev->caps.cqe_size   = 64;
983 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
984 	} else {
985 		dev->caps.cqe_size   = 32;
986 	}
987 
988 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
989 		dev->caps.eqe_size = hca_param.eqe_size;
990 		dev->caps.eqe_factor = 0;
991 	}
992 
993 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
994 		dev->caps.cqe_size = hca_param.cqe_size;
995 		/* User still need to know when CQE > 32B */
996 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
997 	}
998 
999 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1000 	mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
1001 
1002 	slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
1003 	mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
1004 		 hca_param.rss_ip_frags ? "on" : "off");
1005 
1006 	if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
1007 	    dev->caps.bf_reg_size)
1008 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
1009 
1010 	if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
1011 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
1012 
1013 	return 0;
1014 
1015 err_mem:
1016 	kfree(dev->caps.qp0_qkey);
1017 	kfree(dev->caps.qp0_tunnel);
1018 	kfree(dev->caps.qp0_proxy);
1019 	kfree(dev->caps.qp1_tunnel);
1020 	kfree(dev->caps.qp1_proxy);
1021 	dev->caps.qp0_qkey = NULL;
1022 	dev->caps.qp0_tunnel = NULL;
1023 	dev->caps.qp0_proxy = NULL;
1024 	dev->caps.qp1_tunnel = NULL;
1025 	dev->caps.qp1_proxy = NULL;
1026 
1027 	return err;
1028 }
1029 
1030 static void mlx4_request_modules(struct mlx4_dev *dev)
1031 {
1032 	int port;
1033 	int has_ib_port = false;
1034 	int has_eth_port = false;
1035 #define EN_DRV_NAME	"mlx4_en"
1036 #define IB_DRV_NAME	"mlx4_ib"
1037 
1038 	for (port = 1; port <= dev->caps.num_ports; port++) {
1039 		if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
1040 			has_ib_port = true;
1041 		else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
1042 			has_eth_port = true;
1043 	}
1044 
1045 	if (has_eth_port)
1046 		request_module_nowait(EN_DRV_NAME);
1047 	if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
1048 		request_module_nowait(IB_DRV_NAME);
1049 }
1050 
1051 /*
1052  * Change the port configuration of the device.
1053  * Every user of this function must hold the port mutex.
1054  */
1055 int mlx4_change_port_types(struct mlx4_dev *dev,
1056 			   enum mlx4_port_type *port_types)
1057 {
1058 	int err = 0;
1059 	int change = 0;
1060 	int port;
1061 
1062 	for (port = 0; port <  dev->caps.num_ports; port++) {
1063 		/* Change the port type only if the new type is different
1064 		 * from the current, and not set to Auto */
1065 		if (port_types[port] != dev->caps.port_type[port + 1])
1066 			change = 1;
1067 	}
1068 	if (change) {
1069 		mlx4_unregister_device(dev);
1070 		for (port = 1; port <= dev->caps.num_ports; port++) {
1071 			mlx4_CLOSE_PORT(dev, port);
1072 			dev->caps.port_type[port] = port_types[port - 1];
1073 			err = mlx4_SET_PORT(dev, port, -1);
1074 			if (err) {
1075 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1076 					 port);
1077 				goto out;
1078 			}
1079 		}
1080 		mlx4_set_port_mask(dev);
1081 		err = mlx4_register_device(dev);
1082 		if (err) {
1083 			mlx4_err(dev, "Failed to register device\n");
1084 			goto out;
1085 		}
1086 		mlx4_request_modules(dev);
1087 	}
1088 
1089 out:
1090 	return err;
1091 }
1092 
1093 static ssize_t show_port_type(struct device *dev,
1094 			      struct device_attribute *attr,
1095 			      char *buf)
1096 {
1097 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1098 						   port_attr);
1099 	struct mlx4_dev *mdev = info->dev;
1100 	char type[8];
1101 
1102 	sprintf(type, "%s",
1103 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1104 		"ib" : "eth");
1105 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1106 		sprintf(buf, "auto (%s)\n", type);
1107 	else
1108 		sprintf(buf, "%s\n", type);
1109 
1110 	return strlen(buf);
1111 }
1112 
1113 static int __set_port_type(struct mlx4_port_info *info,
1114 			   enum mlx4_port_type port_type)
1115 {
1116 	struct mlx4_dev *mdev = info->dev;
1117 	struct mlx4_priv *priv = mlx4_priv(mdev);
1118 	enum mlx4_port_type types[MLX4_MAX_PORTS];
1119 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1120 	int i;
1121 	int err = 0;
1122 
1123 	if ((port_type & mdev->caps.supported_type[info->port]) != port_type) {
1124 		mlx4_err(mdev,
1125 			 "Requested port type for port %d is not supported on this HCA\n",
1126 			 info->port);
1127 		err = -EINVAL;
1128 		goto err_sup;
1129 	}
1130 
1131 	mlx4_stop_sense(mdev);
1132 	mutex_lock(&priv->port_mutex);
1133 	info->tmp_type = port_type;
1134 
1135 	/* Possible type is always the one that was delivered */
1136 	mdev->caps.possible_type[info->port] = info->tmp_type;
1137 
1138 	for (i = 0; i < mdev->caps.num_ports; i++) {
1139 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1140 					mdev->caps.possible_type[i+1];
1141 		if (types[i] == MLX4_PORT_TYPE_AUTO)
1142 			types[i] = mdev->caps.port_type[i+1];
1143 	}
1144 
1145 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1146 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1147 		for (i = 1; i <= mdev->caps.num_ports; i++) {
1148 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1149 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1150 				err = -EINVAL;
1151 			}
1152 		}
1153 	}
1154 	if (err) {
1155 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1156 		goto out;
1157 	}
1158 
1159 	mlx4_do_sense_ports(mdev, new_types, types);
1160 
1161 	err = mlx4_check_port_params(mdev, new_types);
1162 	if (err)
1163 		goto out;
1164 
1165 	/* We are about to apply the changes after the configuration
1166 	 * was verified, no need to remember the temporary types
1167 	 * any more */
1168 	for (i = 0; i < mdev->caps.num_ports; i++)
1169 		priv->port[i + 1].tmp_type = 0;
1170 
1171 	err = mlx4_change_port_types(mdev, new_types);
1172 
1173 out:
1174 	mlx4_start_sense(mdev);
1175 	mutex_unlock(&priv->port_mutex);
1176 err_sup:
1177 	return err;
1178 }
1179 
1180 static ssize_t set_port_type(struct device *dev,
1181 			     struct device_attribute *attr,
1182 			     const char *buf, size_t count)
1183 {
1184 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1185 						   port_attr);
1186 	struct mlx4_dev *mdev = info->dev;
1187 	enum mlx4_port_type port_type;
1188 	static DEFINE_MUTEX(set_port_type_mutex);
1189 	int err;
1190 
1191 	mutex_lock(&set_port_type_mutex);
1192 
1193 	if (!strcmp(buf, "ib\n")) {
1194 		port_type = MLX4_PORT_TYPE_IB;
1195 	} else if (!strcmp(buf, "eth\n")) {
1196 		port_type = MLX4_PORT_TYPE_ETH;
1197 	} else if (!strcmp(buf, "auto\n")) {
1198 		port_type = MLX4_PORT_TYPE_AUTO;
1199 	} else {
1200 		mlx4_err(mdev, "%s is not supported port type\n", buf);
1201 		err = -EINVAL;
1202 		goto err_out;
1203 	}
1204 
1205 	err = __set_port_type(info, port_type);
1206 
1207 err_out:
1208 	mutex_unlock(&set_port_type_mutex);
1209 
1210 	return err ? err : count;
1211 }
1212 
1213 enum ibta_mtu {
1214 	IB_MTU_256  = 1,
1215 	IB_MTU_512  = 2,
1216 	IB_MTU_1024 = 3,
1217 	IB_MTU_2048 = 4,
1218 	IB_MTU_4096 = 5
1219 };
1220 
1221 static inline int int_to_ibta_mtu(int mtu)
1222 {
1223 	switch (mtu) {
1224 	case 256:  return IB_MTU_256;
1225 	case 512:  return IB_MTU_512;
1226 	case 1024: return IB_MTU_1024;
1227 	case 2048: return IB_MTU_2048;
1228 	case 4096: return IB_MTU_4096;
1229 	default: return -1;
1230 	}
1231 }
1232 
1233 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1234 {
1235 	switch (mtu) {
1236 	case IB_MTU_256:  return  256;
1237 	case IB_MTU_512:  return  512;
1238 	case IB_MTU_1024: return 1024;
1239 	case IB_MTU_2048: return 2048;
1240 	case IB_MTU_4096: return 4096;
1241 	default: return -1;
1242 	}
1243 }
1244 
1245 static ssize_t show_port_ib_mtu(struct device *dev,
1246 			     struct device_attribute *attr,
1247 			     char *buf)
1248 {
1249 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1250 						   port_mtu_attr);
1251 	struct mlx4_dev *mdev = info->dev;
1252 
1253 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1254 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1255 
1256 	sprintf(buf, "%d\n",
1257 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1258 	return strlen(buf);
1259 }
1260 
1261 static ssize_t set_port_ib_mtu(struct device *dev,
1262 			     struct device_attribute *attr,
1263 			     const char *buf, size_t count)
1264 {
1265 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1266 						   port_mtu_attr);
1267 	struct mlx4_dev *mdev = info->dev;
1268 	struct mlx4_priv *priv = mlx4_priv(mdev);
1269 	int err, port, mtu, ibta_mtu = -1;
1270 
1271 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1272 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1273 		return -EINVAL;
1274 	}
1275 
1276 	err = kstrtoint(buf, 0, &mtu);
1277 	if (!err)
1278 		ibta_mtu = int_to_ibta_mtu(mtu);
1279 
1280 	if (err || ibta_mtu < 0) {
1281 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1282 		return -EINVAL;
1283 	}
1284 
1285 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1286 
1287 	mlx4_stop_sense(mdev);
1288 	mutex_lock(&priv->port_mutex);
1289 	mlx4_unregister_device(mdev);
1290 	for (port = 1; port <= mdev->caps.num_ports; port++) {
1291 		mlx4_CLOSE_PORT(mdev, port);
1292 		err = mlx4_SET_PORT(mdev, port, -1);
1293 		if (err) {
1294 			mlx4_err(mdev, "Failed to set port %d, aborting\n",
1295 				 port);
1296 			goto err_set_port;
1297 		}
1298 	}
1299 	err = mlx4_register_device(mdev);
1300 err_set_port:
1301 	mutex_unlock(&priv->port_mutex);
1302 	mlx4_start_sense(mdev);
1303 	return err ? err : count;
1304 }
1305 
1306 /* bond for multi-function device */
1307 #define MAX_MF_BOND_ALLOWED_SLAVES 63
1308 static int mlx4_mf_bond(struct mlx4_dev *dev)
1309 {
1310 	int err = 0;
1311 	int nvfs;
1312 	struct mlx4_slaves_pport slaves_port1;
1313 	struct mlx4_slaves_pport slaves_port2;
1314 	DECLARE_BITMAP(slaves_port_1_2, MLX4_MFUNC_MAX);
1315 
1316 	slaves_port1 = mlx4_phys_to_slaves_pport(dev, 1);
1317 	slaves_port2 = mlx4_phys_to_slaves_pport(dev, 2);
1318 	bitmap_and(slaves_port_1_2,
1319 		   slaves_port1.slaves, slaves_port2.slaves,
1320 		   dev->persist->num_vfs + 1);
1321 
1322 	/* only single port vfs are allowed */
1323 	if (bitmap_weight(slaves_port_1_2, dev->persist->num_vfs + 1) > 1) {
1324 		mlx4_warn(dev, "HA mode unsupported for dual ported VFs\n");
1325 		return -EINVAL;
1326 	}
1327 
1328 	/* number of virtual functions is number of total functions minus one
1329 	 * physical function for each port.
1330 	 */
1331 	nvfs = bitmap_weight(slaves_port1.slaves, dev->persist->num_vfs + 1) +
1332 		bitmap_weight(slaves_port2.slaves, dev->persist->num_vfs + 1) - 2;
1333 
1334 	/* limit on maximum allowed VFs */
1335 	if (nvfs > MAX_MF_BOND_ALLOWED_SLAVES) {
1336 		mlx4_warn(dev, "HA mode is not supported for %d VFs (max %d are allowed)\n",
1337 			  nvfs, MAX_MF_BOND_ALLOWED_SLAVES);
1338 		return -EINVAL;
1339 	}
1340 
1341 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) {
1342 		mlx4_warn(dev, "HA mode unsupported for NON DMFS steering\n");
1343 		return -EINVAL;
1344 	}
1345 
1346 	err = mlx4_bond_mac_table(dev);
1347 	if (err)
1348 		return err;
1349 	err = mlx4_bond_vlan_table(dev);
1350 	if (err)
1351 		goto err1;
1352 	err = mlx4_bond_fs_rules(dev);
1353 	if (err)
1354 		goto err2;
1355 
1356 	return 0;
1357 err2:
1358 	(void)mlx4_unbond_vlan_table(dev);
1359 err1:
1360 	(void)mlx4_unbond_mac_table(dev);
1361 	return err;
1362 }
1363 
1364 static int mlx4_mf_unbond(struct mlx4_dev *dev)
1365 {
1366 	int ret, ret1;
1367 
1368 	ret = mlx4_unbond_fs_rules(dev);
1369 	if (ret)
1370 		mlx4_warn(dev, "multifunction unbond for flow rules failedi (%d)\n", ret);
1371 	ret1 = mlx4_unbond_mac_table(dev);
1372 	if (ret1) {
1373 		mlx4_warn(dev, "multifunction unbond for MAC table failed (%d)\n", ret1);
1374 		ret = ret1;
1375 	}
1376 	ret1 = mlx4_unbond_vlan_table(dev);
1377 	if (ret1) {
1378 		mlx4_warn(dev, "multifunction unbond for VLAN table failed (%d)\n", ret1);
1379 		ret = ret1;
1380 	}
1381 	return ret;
1382 }
1383 
1384 int mlx4_bond(struct mlx4_dev *dev)
1385 {
1386 	int ret = 0;
1387 	struct mlx4_priv *priv = mlx4_priv(dev);
1388 
1389 	mutex_lock(&priv->bond_mutex);
1390 
1391 	if (!mlx4_is_bonded(dev)) {
1392 		ret = mlx4_do_bond(dev, true);
1393 		if (ret)
1394 			mlx4_err(dev, "Failed to bond device: %d\n", ret);
1395 		if (!ret && mlx4_is_master(dev)) {
1396 			ret = mlx4_mf_bond(dev);
1397 			if (ret) {
1398 				mlx4_err(dev, "bond for multifunction failed\n");
1399 				mlx4_do_bond(dev, false);
1400 			}
1401 		}
1402 	}
1403 
1404 	mutex_unlock(&priv->bond_mutex);
1405 	if (!ret)
1406 		mlx4_dbg(dev, "Device is bonded\n");
1407 
1408 	return ret;
1409 }
1410 EXPORT_SYMBOL_GPL(mlx4_bond);
1411 
1412 int mlx4_unbond(struct mlx4_dev *dev)
1413 {
1414 	int ret = 0;
1415 	struct mlx4_priv *priv = mlx4_priv(dev);
1416 
1417 	mutex_lock(&priv->bond_mutex);
1418 
1419 	if (mlx4_is_bonded(dev)) {
1420 		int ret2 = 0;
1421 
1422 		ret = mlx4_do_bond(dev, false);
1423 		if (ret)
1424 			mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1425 		if (mlx4_is_master(dev))
1426 			ret2 = mlx4_mf_unbond(dev);
1427 		if (ret2) {
1428 			mlx4_warn(dev, "Failed to unbond device for multifunction (%d)\n", ret2);
1429 			ret = ret2;
1430 		}
1431 	}
1432 
1433 	mutex_unlock(&priv->bond_mutex);
1434 	if (!ret)
1435 		mlx4_dbg(dev, "Device is unbonded\n");
1436 
1437 	return ret;
1438 }
1439 EXPORT_SYMBOL_GPL(mlx4_unbond);
1440 
1441 
1442 int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1443 {
1444 	u8 port1 = v2p->port1;
1445 	u8 port2 = v2p->port2;
1446 	struct mlx4_priv *priv = mlx4_priv(dev);
1447 	int err;
1448 
1449 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1450 		return -ENOTSUPP;
1451 
1452 	mutex_lock(&priv->bond_mutex);
1453 
1454 	/* zero means keep current mapping for this port */
1455 	if (port1 == 0)
1456 		port1 = priv->v2p.port1;
1457 	if (port2 == 0)
1458 		port2 = priv->v2p.port2;
1459 
1460 	if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1461 	    (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1462 	    (port1 == 2 && port2 == 1)) {
1463 		/* besides boundary checks cross mapping makes
1464 		 * no sense and therefore not allowed */
1465 		err = -EINVAL;
1466 	} else if ((port1 == priv->v2p.port1) &&
1467 		 (port2 == priv->v2p.port2)) {
1468 		err = 0;
1469 	} else {
1470 		err = mlx4_virt2phy_port_map(dev, port1, port2);
1471 		if (!err) {
1472 			mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1473 				 port1, port2);
1474 			priv->v2p.port1 = port1;
1475 			priv->v2p.port2 = port2;
1476 		} else {
1477 			mlx4_err(dev, "Failed to change port mape: %d\n", err);
1478 		}
1479 	}
1480 
1481 	mutex_unlock(&priv->bond_mutex);
1482 	return err;
1483 }
1484 EXPORT_SYMBOL_GPL(mlx4_port_map_set);
1485 
1486 static int mlx4_load_fw(struct mlx4_dev *dev)
1487 {
1488 	struct mlx4_priv *priv = mlx4_priv(dev);
1489 	int err;
1490 
1491 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1492 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
1493 	if (!priv->fw.fw_icm) {
1494 		mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1495 		return -ENOMEM;
1496 	}
1497 
1498 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1499 	if (err) {
1500 		mlx4_err(dev, "MAP_FA command failed, aborting\n");
1501 		goto err_free;
1502 	}
1503 
1504 	err = mlx4_RUN_FW(dev);
1505 	if (err) {
1506 		mlx4_err(dev, "RUN_FW command failed, aborting\n");
1507 		goto err_unmap_fa;
1508 	}
1509 
1510 	return 0;
1511 
1512 err_unmap_fa:
1513 	mlx4_UNMAP_FA(dev);
1514 
1515 err_free:
1516 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1517 	return err;
1518 }
1519 
1520 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1521 				int cmpt_entry_sz)
1522 {
1523 	struct mlx4_priv *priv = mlx4_priv(dev);
1524 	int err;
1525 	int num_eqs;
1526 
1527 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1528 				  cmpt_base +
1529 				  ((u64) (MLX4_CMPT_TYPE_QP *
1530 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1531 				  cmpt_entry_sz, dev->caps.num_qps,
1532 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1533 				  0, 0);
1534 	if (err)
1535 		goto err;
1536 
1537 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1538 				  cmpt_base +
1539 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1540 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1541 				  cmpt_entry_sz, dev->caps.num_srqs,
1542 				  dev->caps.reserved_srqs, 0, 0);
1543 	if (err)
1544 		goto err_qp;
1545 
1546 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1547 				  cmpt_base +
1548 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1549 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1550 				  cmpt_entry_sz, dev->caps.num_cqs,
1551 				  dev->caps.reserved_cqs, 0, 0);
1552 	if (err)
1553 		goto err_srq;
1554 
1555 	num_eqs = dev->phys_caps.num_phys_eqs;
1556 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1557 				  cmpt_base +
1558 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1559 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1560 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1561 	if (err)
1562 		goto err_cq;
1563 
1564 	return 0;
1565 
1566 err_cq:
1567 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1568 
1569 err_srq:
1570 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1571 
1572 err_qp:
1573 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1574 
1575 err:
1576 	return err;
1577 }
1578 
1579 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1580 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1581 {
1582 	struct mlx4_priv *priv = mlx4_priv(dev);
1583 	u64 aux_pages;
1584 	int num_eqs;
1585 	int err;
1586 
1587 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1588 	if (err) {
1589 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1590 		return err;
1591 	}
1592 
1593 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1594 		 (unsigned long long) icm_size >> 10,
1595 		 (unsigned long long) aux_pages << 2);
1596 
1597 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1598 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1599 	if (!priv->fw.aux_icm) {
1600 		mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1601 		return -ENOMEM;
1602 	}
1603 
1604 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1605 	if (err) {
1606 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1607 		goto err_free_aux;
1608 	}
1609 
1610 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1611 	if (err) {
1612 		mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1613 		goto err_unmap_aux;
1614 	}
1615 
1616 
1617 	num_eqs = dev->phys_caps.num_phys_eqs;
1618 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1619 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1620 				  num_eqs, num_eqs, 0, 0);
1621 	if (err) {
1622 		mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1623 		goto err_unmap_cmpt;
1624 	}
1625 
1626 	/*
1627 	 * Reserved MTT entries must be aligned up to a cacheline
1628 	 * boundary, since the FW will write to them, while the driver
1629 	 * writes to all other MTT entries. (The variable
1630 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1631 	 * size, not the raw entry size)
1632 	 */
1633 	dev->caps.reserved_mtts =
1634 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1635 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1636 
1637 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1638 				  init_hca->mtt_base,
1639 				  dev->caps.mtt_entry_sz,
1640 				  dev->caps.num_mtts,
1641 				  dev->caps.reserved_mtts, 1, 0);
1642 	if (err) {
1643 		mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1644 		goto err_unmap_eq;
1645 	}
1646 
1647 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1648 				  init_hca->dmpt_base,
1649 				  dev_cap->dmpt_entry_sz,
1650 				  dev->caps.num_mpts,
1651 				  dev->caps.reserved_mrws, 1, 1);
1652 	if (err) {
1653 		mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1654 		goto err_unmap_mtt;
1655 	}
1656 
1657 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1658 				  init_hca->qpc_base,
1659 				  dev_cap->qpc_entry_sz,
1660 				  dev->caps.num_qps,
1661 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1662 				  0, 0);
1663 	if (err) {
1664 		mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1665 		goto err_unmap_dmpt;
1666 	}
1667 
1668 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1669 				  init_hca->auxc_base,
1670 				  dev_cap->aux_entry_sz,
1671 				  dev->caps.num_qps,
1672 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1673 				  0, 0);
1674 	if (err) {
1675 		mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1676 		goto err_unmap_qp;
1677 	}
1678 
1679 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1680 				  init_hca->altc_base,
1681 				  dev_cap->altc_entry_sz,
1682 				  dev->caps.num_qps,
1683 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1684 				  0, 0);
1685 	if (err) {
1686 		mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1687 		goto err_unmap_auxc;
1688 	}
1689 
1690 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1691 				  init_hca->rdmarc_base,
1692 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1693 				  dev->caps.num_qps,
1694 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1695 				  0, 0);
1696 	if (err) {
1697 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1698 		goto err_unmap_altc;
1699 	}
1700 
1701 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1702 				  init_hca->cqc_base,
1703 				  dev_cap->cqc_entry_sz,
1704 				  dev->caps.num_cqs,
1705 				  dev->caps.reserved_cqs, 0, 0);
1706 	if (err) {
1707 		mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1708 		goto err_unmap_rdmarc;
1709 	}
1710 
1711 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1712 				  init_hca->srqc_base,
1713 				  dev_cap->srq_entry_sz,
1714 				  dev->caps.num_srqs,
1715 				  dev->caps.reserved_srqs, 0, 0);
1716 	if (err) {
1717 		mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1718 		goto err_unmap_cq;
1719 	}
1720 
1721 	/*
1722 	 * For flow steering device managed mode it is required to use
1723 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1724 	 * required, but for simplicity just map the whole multicast
1725 	 * group table now.  The table isn't very big and it's a lot
1726 	 * easier than trying to track ref counts.
1727 	 */
1728 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1729 				  init_hca->mc_base,
1730 				  mlx4_get_mgm_entry_size(dev),
1731 				  dev->caps.num_mgms + dev->caps.num_amgms,
1732 				  dev->caps.num_mgms + dev->caps.num_amgms,
1733 				  0, 0);
1734 	if (err) {
1735 		mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1736 		goto err_unmap_srq;
1737 	}
1738 
1739 	return 0;
1740 
1741 err_unmap_srq:
1742 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1743 
1744 err_unmap_cq:
1745 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1746 
1747 err_unmap_rdmarc:
1748 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1749 
1750 err_unmap_altc:
1751 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1752 
1753 err_unmap_auxc:
1754 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1755 
1756 err_unmap_qp:
1757 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1758 
1759 err_unmap_dmpt:
1760 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1761 
1762 err_unmap_mtt:
1763 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1764 
1765 err_unmap_eq:
1766 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1767 
1768 err_unmap_cmpt:
1769 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1770 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1771 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1772 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1773 
1774 err_unmap_aux:
1775 	mlx4_UNMAP_ICM_AUX(dev);
1776 
1777 err_free_aux:
1778 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1779 
1780 	return err;
1781 }
1782 
1783 static void mlx4_free_icms(struct mlx4_dev *dev)
1784 {
1785 	struct mlx4_priv *priv = mlx4_priv(dev);
1786 
1787 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1788 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1789 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1790 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1791 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1792 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1793 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1794 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1795 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1796 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1797 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1798 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1799 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1800 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1801 
1802 	mlx4_UNMAP_ICM_AUX(dev);
1803 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1804 }
1805 
1806 static void mlx4_slave_exit(struct mlx4_dev *dev)
1807 {
1808 	struct mlx4_priv *priv = mlx4_priv(dev);
1809 
1810 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1811 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1812 			  MLX4_COMM_TIME))
1813 		mlx4_warn(dev, "Failed to close slave function\n");
1814 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1815 }
1816 
1817 static int map_bf_area(struct mlx4_dev *dev)
1818 {
1819 	struct mlx4_priv *priv = mlx4_priv(dev);
1820 	resource_size_t bf_start;
1821 	resource_size_t bf_len;
1822 	int err = 0;
1823 
1824 	if (!dev->caps.bf_reg_size)
1825 		return -ENXIO;
1826 
1827 	bf_start = pci_resource_start(dev->persist->pdev, 2) +
1828 			(dev->caps.num_uars << PAGE_SHIFT);
1829 	bf_len = pci_resource_len(dev->persist->pdev, 2) -
1830 			(dev->caps.num_uars << PAGE_SHIFT);
1831 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1832 	if (!priv->bf_mapping)
1833 		err = -ENOMEM;
1834 
1835 	return err;
1836 }
1837 
1838 static void unmap_bf_area(struct mlx4_dev *dev)
1839 {
1840 	if (mlx4_priv(dev)->bf_mapping)
1841 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1842 }
1843 
1844 u64 mlx4_read_clock(struct mlx4_dev *dev)
1845 {
1846 	u32 clockhi, clocklo, clockhi1;
1847 	u64 cycles;
1848 	int i;
1849 	struct mlx4_priv *priv = mlx4_priv(dev);
1850 
1851 	for (i = 0; i < 10; i++) {
1852 		clockhi = swab32(readl(priv->clock_mapping));
1853 		clocklo = swab32(readl(priv->clock_mapping + 4));
1854 		clockhi1 = swab32(readl(priv->clock_mapping));
1855 		if (clockhi == clockhi1)
1856 			break;
1857 	}
1858 
1859 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1860 
1861 	return cycles;
1862 }
1863 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1864 
1865 
1866 static int map_internal_clock(struct mlx4_dev *dev)
1867 {
1868 	struct mlx4_priv *priv = mlx4_priv(dev);
1869 
1870 	priv->clock_mapping =
1871 		ioremap(pci_resource_start(dev->persist->pdev,
1872 					   priv->fw.clock_bar) +
1873 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1874 
1875 	if (!priv->clock_mapping)
1876 		return -ENOMEM;
1877 
1878 	return 0;
1879 }
1880 
1881 int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1882 				   struct mlx4_clock_params *params)
1883 {
1884 	struct mlx4_priv *priv = mlx4_priv(dev);
1885 
1886 	if (mlx4_is_slave(dev))
1887 		return -ENOTSUPP;
1888 
1889 	if (!params)
1890 		return -EINVAL;
1891 
1892 	params->bar = priv->fw.clock_bar;
1893 	params->offset = priv->fw.clock_offset;
1894 	params->size = MLX4_CLOCK_SIZE;
1895 
1896 	return 0;
1897 }
1898 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1899 
1900 static void unmap_internal_clock(struct mlx4_dev *dev)
1901 {
1902 	struct mlx4_priv *priv = mlx4_priv(dev);
1903 
1904 	if (priv->clock_mapping)
1905 		iounmap(priv->clock_mapping);
1906 }
1907 
1908 static void mlx4_close_hca(struct mlx4_dev *dev)
1909 {
1910 	unmap_internal_clock(dev);
1911 	unmap_bf_area(dev);
1912 	if (mlx4_is_slave(dev))
1913 		mlx4_slave_exit(dev);
1914 	else {
1915 		mlx4_CLOSE_HCA(dev, 0);
1916 		mlx4_free_icms(dev);
1917 	}
1918 }
1919 
1920 static void mlx4_close_fw(struct mlx4_dev *dev)
1921 {
1922 	if (!mlx4_is_slave(dev)) {
1923 		mlx4_UNMAP_FA(dev);
1924 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1925 	}
1926 }
1927 
1928 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1929 {
1930 #define COMM_CHAN_OFFLINE_OFFSET 0x09
1931 
1932 	u32 comm_flags;
1933 	u32 offline_bit;
1934 	unsigned long end;
1935 	struct mlx4_priv *priv = mlx4_priv(dev);
1936 
1937 	end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
1938 	while (time_before(jiffies, end)) {
1939 		comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
1940 					  MLX4_COMM_CHAN_FLAGS));
1941 		offline_bit = (comm_flags &
1942 			       (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
1943 		if (!offline_bit)
1944 			return 0;
1945 		/* There are cases as part of AER/Reset flow that PF needs
1946 		 * around 100 msec to load. We therefore sleep for 100 msec
1947 		 * to allow other tasks to make use of that CPU during this
1948 		 * time interval.
1949 		 */
1950 		msleep(100);
1951 	}
1952 	mlx4_err(dev, "Communication channel is offline.\n");
1953 	return -EIO;
1954 }
1955 
1956 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
1957 {
1958 #define COMM_CHAN_RST_OFFSET 0x1e
1959 
1960 	struct mlx4_priv *priv = mlx4_priv(dev);
1961 	u32 comm_rst;
1962 	u32 comm_caps;
1963 
1964 	comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
1965 				 MLX4_COMM_CHAN_CAPS));
1966 	comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
1967 
1968 	if (comm_rst)
1969 		dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
1970 }
1971 
1972 static int mlx4_init_slave(struct mlx4_dev *dev)
1973 {
1974 	struct mlx4_priv *priv = mlx4_priv(dev);
1975 	u64 dma = (u64) priv->mfunc.vhcr_dma;
1976 	int ret_from_reset = 0;
1977 	u32 slave_read;
1978 	u32 cmd_channel_ver;
1979 
1980 	if (atomic_read(&pf_loading)) {
1981 		mlx4_warn(dev, "PF is not ready - Deferring probe\n");
1982 		return -EPROBE_DEFER;
1983 	}
1984 
1985 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1986 	priv->cmd.max_cmds = 1;
1987 	if (mlx4_comm_check_offline(dev)) {
1988 		mlx4_err(dev, "PF is not responsive, skipping initialization\n");
1989 		goto err_offline;
1990 	}
1991 
1992 	mlx4_reset_vf_support(dev);
1993 	mlx4_warn(dev, "Sending reset\n");
1994 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1995 				       MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
1996 	/* if we are in the middle of flr the slave will try
1997 	 * NUM_OF_RESET_RETRIES times before leaving.*/
1998 	if (ret_from_reset) {
1999 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
2000 			mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
2001 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
2002 			return -EPROBE_DEFER;
2003 		} else
2004 			goto err;
2005 	}
2006 
2007 	/* check the driver version - the slave I/F revision
2008 	 * must match the master's */
2009 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
2010 	cmd_channel_ver = mlx4_comm_get_version();
2011 
2012 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
2013 		MLX4_COMM_GET_IF_REV(slave_read)) {
2014 		mlx4_err(dev, "slave driver version is not supported by the master\n");
2015 		goto err;
2016 	}
2017 
2018 	mlx4_warn(dev, "Sending vhcr0\n");
2019 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
2020 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2021 		goto err;
2022 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
2023 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2024 		goto err;
2025 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
2026 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2027 		goto err;
2028 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
2029 			  MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2030 		goto err;
2031 
2032 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2033 	return 0;
2034 
2035 err:
2036 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
2037 err_offline:
2038 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2039 	return -EIO;
2040 }
2041 
2042 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
2043 {
2044 	int i;
2045 
2046 	for (i = 1; i <= dev->caps.num_ports; i++) {
2047 		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
2048 			dev->caps.gid_table_len[i] =
2049 				mlx4_get_slave_num_gids(dev, 0, i);
2050 		else
2051 			dev->caps.gid_table_len[i] = 1;
2052 		dev->caps.pkey_table_len[i] =
2053 			dev->phys_caps.pkey_phys_table_len[i] - 1;
2054 	}
2055 }
2056 
2057 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
2058 {
2059 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
2060 
2061 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
2062 	      i++) {
2063 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
2064 			break;
2065 	}
2066 
2067 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
2068 }
2069 
2070 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
2071 {
2072 	switch (dmfs_high_steer_mode) {
2073 	case MLX4_STEERING_DMFS_A0_DEFAULT:
2074 		return "default performance";
2075 
2076 	case MLX4_STEERING_DMFS_A0_DYNAMIC:
2077 		return "dynamic hybrid mode";
2078 
2079 	case MLX4_STEERING_DMFS_A0_STATIC:
2080 		return "performance optimized for limited rule configuration (static)";
2081 
2082 	case MLX4_STEERING_DMFS_A0_DISABLE:
2083 		return "disabled performance optimized steering";
2084 
2085 	case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
2086 		return "performance optimized steering not supported";
2087 
2088 	default:
2089 		return "Unrecognized mode";
2090 	}
2091 }
2092 
2093 #define MLX4_DMFS_A0_STEERING			(1UL << 2)
2094 
2095 static void choose_steering_mode(struct mlx4_dev *dev,
2096 				 struct mlx4_dev_cap *dev_cap)
2097 {
2098 	if (mlx4_log_num_mgm_entry_size <= 0) {
2099 		if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
2100 			if (dev->caps.dmfs_high_steer_mode ==
2101 			    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2102 				mlx4_err(dev, "DMFS high rate mode not supported\n");
2103 			else
2104 				dev->caps.dmfs_high_steer_mode =
2105 					MLX4_STEERING_DMFS_A0_STATIC;
2106 		}
2107 	}
2108 
2109 	if (mlx4_log_num_mgm_entry_size <= 0 &&
2110 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
2111 	    (!mlx4_is_mfunc(dev) ||
2112 	     (dev_cap->fs_max_num_qp_per_entry >=
2113 	     (dev->persist->num_vfs + 1))) &&
2114 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
2115 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
2116 		dev->oper_log_mgm_entry_size =
2117 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
2118 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
2119 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
2120 		dev->caps.fs_log_max_ucast_qp_range_size =
2121 			dev_cap->fs_log_max_ucast_qp_range_size;
2122 	} else {
2123 		if (dev->caps.dmfs_high_steer_mode !=
2124 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2125 			dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
2126 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
2127 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2128 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
2129 		else {
2130 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
2131 
2132 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
2133 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2134 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
2135 		}
2136 		dev->oper_log_mgm_entry_size =
2137 			mlx4_log_num_mgm_entry_size > 0 ?
2138 			mlx4_log_num_mgm_entry_size :
2139 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
2140 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
2141 	}
2142 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
2143 		 mlx4_steering_mode_str(dev->caps.steering_mode),
2144 		 dev->oper_log_mgm_entry_size,
2145 		 mlx4_log_num_mgm_entry_size);
2146 }
2147 
2148 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
2149 				       struct mlx4_dev_cap *dev_cap)
2150 {
2151 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2152 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
2153 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
2154 	else
2155 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
2156 
2157 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
2158 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
2159 }
2160 
2161 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
2162 {
2163 	int i;
2164 	struct mlx4_port_cap port_cap;
2165 
2166 	if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2167 		return -EINVAL;
2168 
2169 	for (i = 1; i <= dev->caps.num_ports; i++) {
2170 		if (mlx4_dev_port(dev, i, &port_cap)) {
2171 			mlx4_err(dev,
2172 				 "QUERY_DEV_CAP command failed, can't veify DMFS high rate steering.\n");
2173 		} else if ((dev->caps.dmfs_high_steer_mode !=
2174 			    MLX4_STEERING_DMFS_A0_DEFAULT) &&
2175 			   (port_cap.dmfs_optimized_state ==
2176 			    !!(dev->caps.dmfs_high_steer_mode ==
2177 			    MLX4_STEERING_DMFS_A0_DISABLE))) {
2178 			mlx4_err(dev,
2179 				 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2180 				 dmfs_high_rate_steering_mode_str(
2181 					dev->caps.dmfs_high_steer_mode),
2182 				 (port_cap.dmfs_optimized_state ?
2183 					"enabled" : "disabled"));
2184 		}
2185 	}
2186 
2187 	return 0;
2188 }
2189 
2190 static int mlx4_init_fw(struct mlx4_dev *dev)
2191 {
2192 	struct mlx4_mod_stat_cfg   mlx4_cfg;
2193 	int err = 0;
2194 
2195 	if (!mlx4_is_slave(dev)) {
2196 		err = mlx4_QUERY_FW(dev);
2197 		if (err) {
2198 			if (err == -EACCES)
2199 				mlx4_info(dev, "non-primary physical function, skipping\n");
2200 			else
2201 				mlx4_err(dev, "QUERY_FW command failed, aborting\n");
2202 			return err;
2203 		}
2204 
2205 		err = mlx4_load_fw(dev);
2206 		if (err) {
2207 			mlx4_err(dev, "Failed to start FW, aborting\n");
2208 			return err;
2209 		}
2210 
2211 		mlx4_cfg.log_pg_sz_m = 1;
2212 		mlx4_cfg.log_pg_sz = 0;
2213 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2214 		if (err)
2215 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2216 	}
2217 
2218 	return err;
2219 }
2220 
2221 static int mlx4_init_hca(struct mlx4_dev *dev)
2222 {
2223 	struct mlx4_priv	  *priv = mlx4_priv(dev);
2224 	struct mlx4_adapter	   adapter;
2225 	struct mlx4_dev_cap	   dev_cap;
2226 	struct mlx4_profile	   profile;
2227 	struct mlx4_init_hca_param init_hca;
2228 	u64 icm_size;
2229 	struct mlx4_config_dev_params params;
2230 	int err;
2231 
2232 	if (!mlx4_is_slave(dev)) {
2233 		err = mlx4_dev_cap(dev, &dev_cap);
2234 		if (err) {
2235 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
2236 			return err;
2237 		}
2238 
2239 		choose_steering_mode(dev, &dev_cap);
2240 		choose_tunnel_offload_mode(dev, &dev_cap);
2241 
2242 		if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2243 		    mlx4_is_master(dev))
2244 			dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2245 
2246 		err = mlx4_get_phys_port_id(dev);
2247 		if (err)
2248 			mlx4_err(dev, "Fail to get physical port id\n");
2249 
2250 		if (mlx4_is_master(dev))
2251 			mlx4_parav_master_pf_caps(dev);
2252 
2253 		if (mlx4_low_memory_profile()) {
2254 			mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2255 			profile = low_mem_profile;
2256 		} else {
2257 			profile = default_profile;
2258 		}
2259 		if (dev->caps.steering_mode ==
2260 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
2261 			profile.num_mcg = MLX4_FS_NUM_MCG;
2262 
2263 		icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
2264 					     &init_hca);
2265 		if ((long long) icm_size < 0) {
2266 			err = icm_size;
2267 			return err;
2268 		}
2269 
2270 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
2271 
2272 		if (enable_4k_uar) {
2273 			init_hca.log_uar_sz = ilog2(dev->caps.num_uars) +
2274 						    PAGE_SHIFT - DEFAULT_UAR_PAGE_SHIFT;
2275 			init_hca.uar_page_sz = DEFAULT_UAR_PAGE_SHIFT - 12;
2276 		} else {
2277 			init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
2278 			init_hca.uar_page_sz = PAGE_SHIFT - 12;
2279 		}
2280 
2281 		init_hca.mw_enabled = 0;
2282 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2283 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2284 			init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
2285 
2286 		err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
2287 		if (err)
2288 			return err;
2289 
2290 		err = mlx4_INIT_HCA(dev, &init_hca);
2291 		if (err) {
2292 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
2293 			goto err_free_icm;
2294 		}
2295 
2296 		if (dev_cap.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2297 			err = mlx4_query_func(dev, &dev_cap);
2298 			if (err < 0) {
2299 				mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
2300 				goto err_close;
2301 			} else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2302 				dev->caps.num_eqs = dev_cap.max_eqs;
2303 				dev->caps.reserved_eqs = dev_cap.reserved_eqs;
2304 				dev->caps.reserved_uars = dev_cap.reserved_uars;
2305 			}
2306 		}
2307 
2308 		/*
2309 		 * If TS is supported by FW
2310 		 * read HCA frequency by QUERY_HCA command
2311 		 */
2312 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2313 			memset(&init_hca, 0, sizeof(init_hca));
2314 			err = mlx4_QUERY_HCA(dev, &init_hca);
2315 			if (err) {
2316 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
2317 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2318 			} else {
2319 				dev->caps.hca_core_clock =
2320 					init_hca.hca_core_clock;
2321 			}
2322 
2323 			/* In case we got HCA frequency 0 - disable timestamping
2324 			 * to avoid dividing by zero
2325 			 */
2326 			if (!dev->caps.hca_core_clock) {
2327 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2328 				mlx4_err(dev,
2329 					 "HCA frequency is 0 - timestamping is not supported\n");
2330 			} else if (map_internal_clock(dev)) {
2331 				/*
2332 				 * Map internal clock,
2333 				 * in case of failure disable timestamping
2334 				 */
2335 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2336 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
2337 			}
2338 		}
2339 
2340 		if (dev->caps.dmfs_high_steer_mode !=
2341 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2342 			if (mlx4_validate_optimized_steering(dev))
2343 				mlx4_warn(dev, "Optimized steering validation failed\n");
2344 
2345 			if (dev->caps.dmfs_high_steer_mode ==
2346 			    MLX4_STEERING_DMFS_A0_DISABLE) {
2347 				dev->caps.dmfs_high_rate_qpn_base =
2348 					dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2349 				dev->caps.dmfs_high_rate_qpn_range =
2350 					MLX4_A0_STEERING_TABLE_SIZE;
2351 			}
2352 
2353 			mlx4_dbg(dev, "DMFS high rate steer mode is: %s\n",
2354 				 dmfs_high_rate_steering_mode_str(
2355 					dev->caps.dmfs_high_steer_mode));
2356 		}
2357 	} else {
2358 		err = mlx4_init_slave(dev);
2359 		if (err) {
2360 			if (err != -EPROBE_DEFER)
2361 				mlx4_err(dev, "Failed to initialize slave\n");
2362 			return err;
2363 		}
2364 
2365 		err = mlx4_slave_cap(dev);
2366 		if (err) {
2367 			mlx4_err(dev, "Failed to obtain slave caps\n");
2368 			goto err_close;
2369 		}
2370 	}
2371 
2372 	if (map_bf_area(dev))
2373 		mlx4_dbg(dev, "Failed to map blue flame area\n");
2374 
2375 	/*Only the master set the ports, all the rest got it from it.*/
2376 	if (!mlx4_is_slave(dev))
2377 		mlx4_set_port_mask(dev);
2378 
2379 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
2380 	if (err) {
2381 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2382 		goto unmap_bf;
2383 	}
2384 
2385 	/* Query CONFIG_DEV parameters */
2386 	err = mlx4_config_dev_retrieval(dev, &params);
2387 	if (err && err != -ENOTSUPP) {
2388 		mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2389 	} else if (!err) {
2390 		dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2391 		dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2392 	}
2393 	priv->eq_table.inta_pin = adapter.inta_pin;
2394 	memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
2395 
2396 	return 0;
2397 
2398 unmap_bf:
2399 	unmap_internal_clock(dev);
2400 	unmap_bf_area(dev);
2401 
2402 	if (mlx4_is_slave(dev)) {
2403 		kfree(dev->caps.qp0_qkey);
2404 		kfree(dev->caps.qp0_tunnel);
2405 		kfree(dev->caps.qp0_proxy);
2406 		kfree(dev->caps.qp1_tunnel);
2407 		kfree(dev->caps.qp1_proxy);
2408 	}
2409 
2410 err_close:
2411 	if (mlx4_is_slave(dev))
2412 		mlx4_slave_exit(dev);
2413 	else
2414 		mlx4_CLOSE_HCA(dev, 0);
2415 
2416 err_free_icm:
2417 	if (!mlx4_is_slave(dev))
2418 		mlx4_free_icms(dev);
2419 
2420 	return err;
2421 }
2422 
2423 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2424 {
2425 	struct mlx4_priv *priv = mlx4_priv(dev);
2426 	int nent_pow2;
2427 
2428 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2429 		return -ENOENT;
2430 
2431 	if (!dev->caps.max_counters)
2432 		return -ENOSPC;
2433 
2434 	nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2435 	/* reserve last counter index for sink counter */
2436 	return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2437 				nent_pow2 - 1, 0,
2438 				nent_pow2 - dev->caps.max_counters + 1);
2439 }
2440 
2441 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2442 {
2443 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2444 		return;
2445 
2446 	if (!dev->caps.max_counters)
2447 		return;
2448 
2449 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2450 }
2451 
2452 static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2453 {
2454 	struct mlx4_priv *priv = mlx4_priv(dev);
2455 	int port;
2456 
2457 	for (port = 0; port < dev->caps.num_ports; port++)
2458 		if (priv->def_counter[port] != -1)
2459 			mlx4_counter_free(dev,  priv->def_counter[port]);
2460 }
2461 
2462 static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2463 {
2464 	struct mlx4_priv *priv = mlx4_priv(dev);
2465 	int port, err = 0;
2466 	u32 idx;
2467 
2468 	for (port = 0; port < dev->caps.num_ports; port++)
2469 		priv->def_counter[port] = -1;
2470 
2471 	for (port = 0; port < dev->caps.num_ports; port++) {
2472 		err = mlx4_counter_alloc(dev, &idx);
2473 
2474 		if (!err || err == -ENOSPC) {
2475 			priv->def_counter[port] = idx;
2476 		} else if (err == -ENOENT) {
2477 			err = 0;
2478 			continue;
2479 		} else if (mlx4_is_slave(dev) && err == -EINVAL) {
2480 			priv->def_counter[port] = MLX4_SINK_COUNTER_INDEX(dev);
2481 			mlx4_warn(dev, "can't allocate counter from old PF driver, using index %d\n",
2482 				  MLX4_SINK_COUNTER_INDEX(dev));
2483 			err = 0;
2484 		} else {
2485 			mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2486 				 __func__, port + 1, err);
2487 			mlx4_cleanup_default_counters(dev);
2488 			return err;
2489 		}
2490 
2491 		mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2492 			 __func__, priv->def_counter[port], port + 1);
2493 	}
2494 
2495 	return err;
2496 }
2497 
2498 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2499 {
2500 	struct mlx4_priv *priv = mlx4_priv(dev);
2501 
2502 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2503 		return -ENOENT;
2504 
2505 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2506 	if (*idx == -1) {
2507 		*idx = MLX4_SINK_COUNTER_INDEX(dev);
2508 		return -ENOSPC;
2509 	}
2510 
2511 	return 0;
2512 }
2513 
2514 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2515 {
2516 	u64 out_param;
2517 	int err;
2518 
2519 	if (mlx4_is_mfunc(dev)) {
2520 		err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
2521 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2522 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2523 		if (!err)
2524 			*idx = get_param_l(&out_param);
2525 
2526 		return err;
2527 	}
2528 	return __mlx4_counter_alloc(dev, idx);
2529 }
2530 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2531 
2532 static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2533 				u8 counter_index)
2534 {
2535 	struct mlx4_cmd_mailbox *if_stat_mailbox;
2536 	int err;
2537 	u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2538 
2539 	if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2540 	if (IS_ERR(if_stat_mailbox))
2541 		return PTR_ERR(if_stat_mailbox);
2542 
2543 	err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2544 			   MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2545 			   MLX4_CMD_NATIVE);
2546 
2547 	mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2548 	return err;
2549 }
2550 
2551 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2552 {
2553 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2554 		return;
2555 
2556 	if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2557 		return;
2558 
2559 	__mlx4_clear_if_stat(dev, idx);
2560 
2561 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2562 	return;
2563 }
2564 
2565 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2566 {
2567 	u64 in_param = 0;
2568 
2569 	if (mlx4_is_mfunc(dev)) {
2570 		set_param_l(&in_param, idx);
2571 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2572 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2573 			 MLX4_CMD_WRAPPED);
2574 		return;
2575 	}
2576 	__mlx4_counter_free(dev, idx);
2577 }
2578 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2579 
2580 int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2581 {
2582 	struct mlx4_priv *priv = mlx4_priv(dev);
2583 
2584 	return priv->def_counter[port - 1];
2585 }
2586 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2587 
2588 void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2589 {
2590 	struct mlx4_priv *priv = mlx4_priv(dev);
2591 
2592 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2593 }
2594 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2595 
2596 __be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2597 {
2598 	struct mlx4_priv *priv = mlx4_priv(dev);
2599 
2600 	return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2601 }
2602 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2603 
2604 void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2605 {
2606 	struct mlx4_priv *priv = mlx4_priv(dev);
2607 	__be64 guid;
2608 
2609 	/* hw GUID */
2610 	if (entry == 0)
2611 		return;
2612 
2613 	get_random_bytes((char *)&guid, sizeof(guid));
2614 	guid &= ~(cpu_to_be64(1ULL << 56));
2615 	guid |= cpu_to_be64(1ULL << 57);
2616 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2617 }
2618 
2619 static int mlx4_setup_hca(struct mlx4_dev *dev)
2620 {
2621 	struct mlx4_priv *priv = mlx4_priv(dev);
2622 	int err;
2623 	int port;
2624 	__be32 ib_port_default_caps;
2625 
2626 	err = mlx4_init_uar_table(dev);
2627 	if (err) {
2628 		mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2629 		return err;
2630 	}
2631 
2632 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
2633 	if (err) {
2634 		mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2635 		goto err_uar_table_free;
2636 	}
2637 
2638 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2639 	if (!priv->kar) {
2640 		mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2641 		err = -ENOMEM;
2642 		goto err_uar_free;
2643 	}
2644 
2645 	err = mlx4_init_pd_table(dev);
2646 	if (err) {
2647 		mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2648 		goto err_kar_unmap;
2649 	}
2650 
2651 	err = mlx4_init_xrcd_table(dev);
2652 	if (err) {
2653 		mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2654 		goto err_pd_table_free;
2655 	}
2656 
2657 	err = mlx4_init_mr_table(dev);
2658 	if (err) {
2659 		mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2660 		goto err_xrcd_table_free;
2661 	}
2662 
2663 	if (!mlx4_is_slave(dev)) {
2664 		err = mlx4_init_mcg_table(dev);
2665 		if (err) {
2666 			mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2667 			goto err_mr_table_free;
2668 		}
2669 		err = mlx4_config_mad_demux(dev);
2670 		if (err) {
2671 			mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2672 			goto err_mcg_table_free;
2673 		}
2674 	}
2675 
2676 	err = mlx4_init_eq_table(dev);
2677 	if (err) {
2678 		mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2679 		goto err_mcg_table_free;
2680 	}
2681 
2682 	err = mlx4_cmd_use_events(dev);
2683 	if (err) {
2684 		mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2685 		goto err_eq_table_free;
2686 	}
2687 
2688 	err = mlx4_NOP(dev);
2689 	if (err) {
2690 		if (dev->flags & MLX4_FLAG_MSI_X) {
2691 			mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2692 				  priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2693 			mlx4_warn(dev, "Trying again without MSI-X\n");
2694 		} else {
2695 			mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2696 				 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2697 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2698 		}
2699 
2700 		goto err_cmd_poll;
2701 	}
2702 
2703 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
2704 
2705 	err = mlx4_init_cq_table(dev);
2706 	if (err) {
2707 		mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2708 		goto err_cmd_poll;
2709 	}
2710 
2711 	err = mlx4_init_srq_table(dev);
2712 	if (err) {
2713 		mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2714 		goto err_cq_table_free;
2715 	}
2716 
2717 	err = mlx4_init_qp_table(dev);
2718 	if (err) {
2719 		mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2720 		goto err_srq_table_free;
2721 	}
2722 
2723 	if (!mlx4_is_slave(dev)) {
2724 		err = mlx4_init_counters_table(dev);
2725 		if (err && err != -ENOENT) {
2726 			mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2727 			goto err_qp_table_free;
2728 		}
2729 	}
2730 
2731 	err = mlx4_allocate_default_counters(dev);
2732 	if (err) {
2733 		mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2734 		goto err_counters_table_free;
2735 	}
2736 
2737 	if (!mlx4_is_slave(dev)) {
2738 		for (port = 1; port <= dev->caps.num_ports; port++) {
2739 			ib_port_default_caps = 0;
2740 			err = mlx4_get_port_ib_caps(dev, port,
2741 						    &ib_port_default_caps);
2742 			if (err)
2743 				mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2744 					  port, err);
2745 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2746 
2747 			/* initialize per-slave default ib port capabilities */
2748 			if (mlx4_is_master(dev)) {
2749 				int i;
2750 				for (i = 0; i < dev->num_slaves; i++) {
2751 					if (i == mlx4_master_func_num(dev))
2752 						continue;
2753 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2754 						ib_port_default_caps;
2755 				}
2756 			}
2757 
2758 			if (mlx4_is_mfunc(dev))
2759 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2760 			else
2761 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2762 
2763 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2764 					    dev->caps.pkey_table_len[port] : -1);
2765 			if (err) {
2766 				mlx4_err(dev, "Failed to set port %d, aborting\n",
2767 					 port);
2768 				goto err_default_countes_free;
2769 			}
2770 		}
2771 	}
2772 
2773 	return 0;
2774 
2775 err_default_countes_free:
2776 	mlx4_cleanup_default_counters(dev);
2777 
2778 err_counters_table_free:
2779 	if (!mlx4_is_slave(dev))
2780 		mlx4_cleanup_counters_table(dev);
2781 
2782 err_qp_table_free:
2783 	mlx4_cleanup_qp_table(dev);
2784 
2785 err_srq_table_free:
2786 	mlx4_cleanup_srq_table(dev);
2787 
2788 err_cq_table_free:
2789 	mlx4_cleanup_cq_table(dev);
2790 
2791 err_cmd_poll:
2792 	mlx4_cmd_use_polling(dev);
2793 
2794 err_eq_table_free:
2795 	mlx4_cleanup_eq_table(dev);
2796 
2797 err_mcg_table_free:
2798 	if (!mlx4_is_slave(dev))
2799 		mlx4_cleanup_mcg_table(dev);
2800 
2801 err_mr_table_free:
2802 	mlx4_cleanup_mr_table(dev);
2803 
2804 err_xrcd_table_free:
2805 	mlx4_cleanup_xrcd_table(dev);
2806 
2807 err_pd_table_free:
2808 	mlx4_cleanup_pd_table(dev);
2809 
2810 err_kar_unmap:
2811 	iounmap(priv->kar);
2812 
2813 err_uar_free:
2814 	mlx4_uar_free(dev, &priv->driver_uar);
2815 
2816 err_uar_table_free:
2817 	mlx4_cleanup_uar_table(dev);
2818 	return err;
2819 }
2820 
2821 static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2822 {
2823 	int requested_cpu = 0;
2824 	struct mlx4_priv *priv = mlx4_priv(dev);
2825 	struct mlx4_eq *eq;
2826 	int off = 0;
2827 	int i;
2828 
2829 	if (eqn > dev->caps.num_comp_vectors)
2830 		return -EINVAL;
2831 
2832 	for (i = 1; i < port; i++)
2833 		off += mlx4_get_eqs_per_port(dev, i);
2834 
2835 	requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2836 
2837 	/* Meaning EQs are shared, and this call comes from the second port */
2838 	if (requested_cpu < 0)
2839 		return 0;
2840 
2841 	eq = &priv->eq_table.eq[eqn];
2842 
2843 	if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2844 		return -ENOMEM;
2845 
2846 	cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2847 
2848 	return 0;
2849 }
2850 
2851 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2852 {
2853 	struct mlx4_priv *priv = mlx4_priv(dev);
2854 	struct msix_entry *entries;
2855 	int i;
2856 	int port = 0;
2857 
2858 	if (msi_x) {
2859 		int nreq = dev->caps.num_ports * num_online_cpus() + 1;
2860 
2861 		nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
2862 			     nreq);
2863 		if (nreq > MAX_MSIX)
2864 			nreq = MAX_MSIX;
2865 
2866 		entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
2867 		if (!entries)
2868 			goto no_msi;
2869 
2870 		for (i = 0; i < nreq; ++i)
2871 			entries[i].entry = i;
2872 
2873 		nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2874 					     nreq);
2875 
2876 		if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
2877 			kfree(entries);
2878 			goto no_msi;
2879 		}
2880 		/* 1 is reserved for events (asyncrounous EQ) */
2881 		dev->caps.num_comp_vectors = nreq - 1;
2882 
2883 		priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2884 		bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2885 			    dev->caps.num_ports);
2886 
2887 		for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2888 			if (i == MLX4_EQ_ASYNC)
2889 				continue;
2890 
2891 			priv->eq_table.eq[i].irq =
2892 				entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2893 
2894 			if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2895 				bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2896 					    dev->caps.num_ports);
2897 				/* We don't set affinity hint when there
2898 				 * aren't enough EQs
2899 				 */
2900 			} else {
2901 				set_bit(port,
2902 					priv->eq_table.eq[i].actv_ports.ports);
2903 				if (mlx4_init_affinity_hint(dev, port + 1, i))
2904 					mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
2905 						  i);
2906 			}
2907 			/* We divide the Eqs evenly between the two ports.
2908 			 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
2909 			 * refers to the number of Eqs per port
2910 			 * (i.e eqs_per_port). Theoretically, we would like to
2911 			 * write something like (i + 1) % eqs_per_port == 0.
2912 			 * However, since there's an asynchronous Eq, we have
2913 			 * to skip over it by comparing this condition to
2914 			 * !!((i + 1) > MLX4_EQ_ASYNC).
2915 			 */
2916 			if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
2917 			    ((i + 1) %
2918 			     (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
2919 			    !!((i + 1) > MLX4_EQ_ASYNC))
2920 				/* If dev->caps.num_comp_vectors < dev->caps.num_ports,
2921 				 * everything is shared anyway.
2922 				 */
2923 				port++;
2924 		}
2925 
2926 		dev->flags |= MLX4_FLAG_MSI_X;
2927 
2928 		kfree(entries);
2929 		return;
2930 	}
2931 
2932 no_msi:
2933 	dev->caps.num_comp_vectors = 1;
2934 
2935 	BUG_ON(MLX4_EQ_ASYNC >= 2);
2936 	for (i = 0; i < 2; ++i) {
2937 		priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
2938 		if (i != MLX4_EQ_ASYNC) {
2939 			bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2940 				    dev->caps.num_ports);
2941 		}
2942 	}
2943 }
2944 
2945 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2946 {
2947 	struct devlink *devlink = priv_to_devlink(mlx4_priv(dev));
2948 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2949 	int err;
2950 
2951 	err = devlink_port_register(devlink, &info->devlink_port, port);
2952 	if (err)
2953 		return err;
2954 
2955 	info->dev = dev;
2956 	info->port = port;
2957 	if (!mlx4_is_slave(dev)) {
2958 		mlx4_init_mac_table(dev, &info->mac_table);
2959 		mlx4_init_vlan_table(dev, &info->vlan_table);
2960 		mlx4_init_roce_gid_table(dev, &info->gid_table);
2961 		info->base_qpn = mlx4_get_base_qpn(dev, port);
2962 	}
2963 
2964 	sprintf(info->dev_name, "mlx4_port%d", port);
2965 	info->port_attr.attr.name = info->dev_name;
2966 	if (mlx4_is_mfunc(dev))
2967 		info->port_attr.attr.mode = S_IRUGO;
2968 	else {
2969 		info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2970 		info->port_attr.store     = set_port_type;
2971 	}
2972 	info->port_attr.show      = show_port_type;
2973 	sysfs_attr_init(&info->port_attr.attr);
2974 
2975 	err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
2976 	if (err) {
2977 		mlx4_err(dev, "Failed to create file for port %d\n", port);
2978 		devlink_port_unregister(&info->devlink_port);
2979 		info->port = -1;
2980 	}
2981 
2982 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2983 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
2984 	if (mlx4_is_mfunc(dev))
2985 		info->port_mtu_attr.attr.mode = S_IRUGO;
2986 	else {
2987 		info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2988 		info->port_mtu_attr.store     = set_port_ib_mtu;
2989 	}
2990 	info->port_mtu_attr.show      = show_port_ib_mtu;
2991 	sysfs_attr_init(&info->port_mtu_attr.attr);
2992 
2993 	err = device_create_file(&dev->persist->pdev->dev,
2994 				 &info->port_mtu_attr);
2995 	if (err) {
2996 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
2997 		device_remove_file(&info->dev->persist->pdev->dev,
2998 				   &info->port_attr);
2999 		devlink_port_unregister(&info->devlink_port);
3000 		info->port = -1;
3001 	}
3002 
3003 	return err;
3004 }
3005 
3006 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
3007 {
3008 	if (info->port < 0)
3009 		return;
3010 
3011 	device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
3012 	device_remove_file(&info->dev->persist->pdev->dev,
3013 			   &info->port_mtu_attr);
3014 	devlink_port_unregister(&info->devlink_port);
3015 
3016 #ifdef CONFIG_RFS_ACCEL
3017 	free_irq_cpu_rmap(info->rmap);
3018 	info->rmap = NULL;
3019 #endif
3020 }
3021 
3022 static int mlx4_init_steering(struct mlx4_dev *dev)
3023 {
3024 	struct mlx4_priv *priv = mlx4_priv(dev);
3025 	int num_entries = dev->caps.num_ports;
3026 	int i, j;
3027 
3028 	priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
3029 	if (!priv->steer)
3030 		return -ENOMEM;
3031 
3032 	for (i = 0; i < num_entries; i++)
3033 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3034 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
3035 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
3036 		}
3037 	return 0;
3038 }
3039 
3040 static void mlx4_clear_steering(struct mlx4_dev *dev)
3041 {
3042 	struct mlx4_priv *priv = mlx4_priv(dev);
3043 	struct mlx4_steer_index *entry, *tmp_entry;
3044 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
3045 	int num_entries = dev->caps.num_ports;
3046 	int i, j;
3047 
3048 	for (i = 0; i < num_entries; i++) {
3049 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3050 			list_for_each_entry_safe(pqp, tmp_pqp,
3051 						 &priv->steer[i].promisc_qps[j],
3052 						 list) {
3053 				list_del(&pqp->list);
3054 				kfree(pqp);
3055 			}
3056 			list_for_each_entry_safe(entry, tmp_entry,
3057 						 &priv->steer[i].steer_entries[j],
3058 						 list) {
3059 				list_del(&entry->list);
3060 				list_for_each_entry_safe(pqp, tmp_pqp,
3061 							 &entry->duplicates,
3062 							 list) {
3063 					list_del(&pqp->list);
3064 					kfree(pqp);
3065 				}
3066 				kfree(entry);
3067 			}
3068 		}
3069 	}
3070 	kfree(priv->steer);
3071 }
3072 
3073 static int extended_func_num(struct pci_dev *pdev)
3074 {
3075 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
3076 }
3077 
3078 #define MLX4_OWNER_BASE	0x8069c
3079 #define MLX4_OWNER_SIZE	4
3080 
3081 static int mlx4_get_ownership(struct mlx4_dev *dev)
3082 {
3083 	void __iomem *owner;
3084 	u32 ret;
3085 
3086 	if (pci_channel_offline(dev->persist->pdev))
3087 		return -EIO;
3088 
3089 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3090 			MLX4_OWNER_BASE,
3091 			MLX4_OWNER_SIZE);
3092 	if (!owner) {
3093 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3094 		return -ENOMEM;
3095 	}
3096 
3097 	ret = readl(owner);
3098 	iounmap(owner);
3099 	return (int) !!ret;
3100 }
3101 
3102 static void mlx4_free_ownership(struct mlx4_dev *dev)
3103 {
3104 	void __iomem *owner;
3105 
3106 	if (pci_channel_offline(dev->persist->pdev))
3107 		return;
3108 
3109 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3110 			MLX4_OWNER_BASE,
3111 			MLX4_OWNER_SIZE);
3112 	if (!owner) {
3113 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3114 		return;
3115 	}
3116 	writel(0, owner);
3117 	msleep(1000);
3118 	iounmap(owner);
3119 }
3120 
3121 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV)	==\
3122 				  !!((flags) & MLX4_FLAG_MASTER))
3123 
3124 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
3125 			     u8 total_vfs, int existing_vfs, int reset_flow)
3126 {
3127 	u64 dev_flags = dev->flags;
3128 	int err = 0;
3129 	int fw_enabled_sriov_vfs = min(pci_sriov_get_totalvfs(pdev),
3130 					MLX4_MAX_NUM_VF);
3131 
3132 	if (reset_flow) {
3133 		dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
3134 				       GFP_KERNEL);
3135 		if (!dev->dev_vfs)
3136 			goto free_mem;
3137 		return dev_flags;
3138 	}
3139 
3140 	atomic_inc(&pf_loading);
3141 	if (dev->flags &  MLX4_FLAG_SRIOV) {
3142 		if (existing_vfs != total_vfs) {
3143 			mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
3144 				 existing_vfs, total_vfs);
3145 			total_vfs = existing_vfs;
3146 		}
3147 	}
3148 
3149 	dev->dev_vfs = kzalloc(total_vfs * sizeof(*dev->dev_vfs), GFP_KERNEL);
3150 	if (NULL == dev->dev_vfs) {
3151 		mlx4_err(dev, "Failed to allocate memory for VFs\n");
3152 		goto disable_sriov;
3153 	}
3154 
3155 	if (!(dev->flags &  MLX4_FLAG_SRIOV)) {
3156 		if (total_vfs > fw_enabled_sriov_vfs) {
3157 			mlx4_err(dev, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
3158 				 total_vfs, fw_enabled_sriov_vfs);
3159 			err = -ENOMEM;
3160 			goto disable_sriov;
3161 		}
3162 		mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
3163 		err = pci_enable_sriov(pdev, total_vfs);
3164 	}
3165 	if (err) {
3166 		mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
3167 			 err);
3168 		goto disable_sriov;
3169 	} else {
3170 		mlx4_warn(dev, "Running in master mode\n");
3171 		dev_flags |= MLX4_FLAG_SRIOV |
3172 			MLX4_FLAG_MASTER;
3173 		dev_flags &= ~MLX4_FLAG_SLAVE;
3174 		dev->persist->num_vfs = total_vfs;
3175 	}
3176 	return dev_flags;
3177 
3178 disable_sriov:
3179 	atomic_dec(&pf_loading);
3180 free_mem:
3181 	dev->persist->num_vfs = 0;
3182 	kfree(dev->dev_vfs);
3183         dev->dev_vfs = NULL;
3184 	return dev_flags & ~MLX4_FLAG_MASTER;
3185 }
3186 
3187 enum {
3188 	MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
3189 };
3190 
3191 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
3192 			      int *nvfs)
3193 {
3194 	int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
3195 	/* Checking for 64 VFs as a limitation of CX2 */
3196 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
3197 	    requested_vfs >= 64) {
3198 		mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
3199 			 requested_vfs);
3200 		return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
3201 	}
3202 	return 0;
3203 }
3204 
3205 static int mlx4_pci_enable_device(struct mlx4_dev *dev)
3206 {
3207 	struct pci_dev *pdev = dev->persist->pdev;
3208 	int err = 0;
3209 
3210 	mutex_lock(&dev->persist->pci_status_mutex);
3211 	if (dev->persist->pci_status == MLX4_PCI_STATUS_DISABLED) {
3212 		err = pci_enable_device(pdev);
3213 		if (!err)
3214 			dev->persist->pci_status = MLX4_PCI_STATUS_ENABLED;
3215 	}
3216 	mutex_unlock(&dev->persist->pci_status_mutex);
3217 
3218 	return err;
3219 }
3220 
3221 static void mlx4_pci_disable_device(struct mlx4_dev *dev)
3222 {
3223 	struct pci_dev *pdev = dev->persist->pdev;
3224 
3225 	mutex_lock(&dev->persist->pci_status_mutex);
3226 	if (dev->persist->pci_status == MLX4_PCI_STATUS_ENABLED) {
3227 		pci_disable_device(pdev);
3228 		dev->persist->pci_status = MLX4_PCI_STATUS_DISABLED;
3229 	}
3230 	mutex_unlock(&dev->persist->pci_status_mutex);
3231 }
3232 
3233 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
3234 			 int total_vfs, int *nvfs, struct mlx4_priv *priv,
3235 			 int reset_flow)
3236 {
3237 	struct mlx4_dev *dev;
3238 	unsigned sum = 0;
3239 	int err;
3240 	int port;
3241 	int i;
3242 	struct mlx4_dev_cap *dev_cap = NULL;
3243 	int existing_vfs = 0;
3244 
3245 	dev = &priv->dev;
3246 
3247 	INIT_LIST_HEAD(&priv->ctx_list);
3248 	spin_lock_init(&priv->ctx_lock);
3249 
3250 	mutex_init(&priv->port_mutex);
3251 	mutex_init(&priv->bond_mutex);
3252 
3253 	INIT_LIST_HEAD(&priv->pgdir_list);
3254 	mutex_init(&priv->pgdir_mutex);
3255 	spin_lock_init(&priv->cmd.context_lock);
3256 
3257 	INIT_LIST_HEAD(&priv->bf_list);
3258 	mutex_init(&priv->bf_mutex);
3259 
3260 	dev->rev_id = pdev->revision;
3261 	dev->numa_node = dev_to_node(&pdev->dev);
3262 
3263 	/* Detect if this device is a virtual function */
3264 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3265 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3266 		dev->flags |= MLX4_FLAG_SLAVE;
3267 	} else {
3268 		/* We reset the device and enable SRIOV only for physical
3269 		 * devices.  Try to claim ownership on the device;
3270 		 * if already taken, skip -- do not allow multiple PFs */
3271 		err = mlx4_get_ownership(dev);
3272 		if (err) {
3273 			if (err < 0)
3274 				return err;
3275 			else {
3276 				mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
3277 				return -EINVAL;
3278 			}
3279 		}
3280 
3281 		atomic_set(&priv->opreq_count, 0);
3282 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3283 
3284 		/*
3285 		 * Now reset the HCA before we touch the PCI capabilities or
3286 		 * attempt a firmware command, since a boot ROM may have left
3287 		 * the HCA in an undefined state.
3288 		 */
3289 		err = mlx4_reset(dev);
3290 		if (err) {
3291 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
3292 			goto err_sriov;
3293 		}
3294 
3295 		if (total_vfs) {
3296 			dev->flags = MLX4_FLAG_MASTER;
3297 			existing_vfs = pci_num_vf(pdev);
3298 			if (existing_vfs)
3299 				dev->flags |= MLX4_FLAG_SRIOV;
3300 			dev->persist->num_vfs = total_vfs;
3301 		}
3302 	}
3303 
3304 	/* on load remove any previous indication of internal error,
3305 	 * device is up.
3306 	 */
3307 	dev->persist->state = MLX4_DEVICE_STATE_UP;
3308 
3309 slave_start:
3310 	err = mlx4_cmd_init(dev);
3311 	if (err) {
3312 		mlx4_err(dev, "Failed to init command interface, aborting\n");
3313 		goto err_sriov;
3314 	}
3315 
3316 	/* In slave functions, the communication channel must be initialized
3317 	 * before posting commands. Also, init num_slaves before calling
3318 	 * mlx4_init_hca */
3319 	if (mlx4_is_mfunc(dev)) {
3320 		if (mlx4_is_master(dev)) {
3321 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
3322 
3323 		} else {
3324 			dev->num_slaves = 0;
3325 			err = mlx4_multi_func_init(dev);
3326 			if (err) {
3327 				mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
3328 				goto err_cmd;
3329 			}
3330 		}
3331 	}
3332 
3333 	err = mlx4_init_fw(dev);
3334 	if (err) {
3335 		mlx4_err(dev, "Failed to init fw, aborting.\n");
3336 		goto err_mfunc;
3337 	}
3338 
3339 	if (mlx4_is_master(dev)) {
3340 		/* when we hit the goto slave_start below, dev_cap already initialized */
3341 		if (!dev_cap) {
3342 			dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3343 
3344 			if (!dev_cap) {
3345 				err = -ENOMEM;
3346 				goto err_fw;
3347 			}
3348 
3349 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3350 			if (err) {
3351 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3352 				goto err_fw;
3353 			}
3354 
3355 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3356 				goto err_fw;
3357 
3358 			if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3359 				u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3360 								  total_vfs,
3361 								  existing_vfs,
3362 								  reset_flow);
3363 
3364 				mlx4_close_fw(dev);
3365 				mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3366 				dev->flags = dev_flags;
3367 				if (!SRIOV_VALID_STATE(dev->flags)) {
3368 					mlx4_err(dev, "Invalid SRIOV state\n");
3369 					goto err_sriov;
3370 				}
3371 				err = mlx4_reset(dev);
3372 				if (err) {
3373 					mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3374 					goto err_sriov;
3375 				}
3376 				goto slave_start;
3377 			}
3378 		} else {
3379 			/* Legacy mode FW requires SRIOV to be enabled before
3380 			 * doing QUERY_DEV_CAP, since max_eq's value is different if
3381 			 * SRIOV is enabled.
3382 			 */
3383 			memset(dev_cap, 0, sizeof(*dev_cap));
3384 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3385 			if (err) {
3386 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3387 				goto err_fw;
3388 			}
3389 
3390 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3391 				goto err_fw;
3392 		}
3393 	}
3394 
3395 	err = mlx4_init_hca(dev);
3396 	if (err) {
3397 		if (err == -EACCES) {
3398 			/* Not primary Physical function
3399 			 * Running in slave mode */
3400 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3401 			/* We're not a PF */
3402 			if (dev->flags & MLX4_FLAG_SRIOV) {
3403 				if (!existing_vfs)
3404 					pci_disable_sriov(pdev);
3405 				if (mlx4_is_master(dev) && !reset_flow)
3406 					atomic_dec(&pf_loading);
3407 				dev->flags &= ~MLX4_FLAG_SRIOV;
3408 			}
3409 			if (!mlx4_is_slave(dev))
3410 				mlx4_free_ownership(dev);
3411 			dev->flags |= MLX4_FLAG_SLAVE;
3412 			dev->flags &= ~MLX4_FLAG_MASTER;
3413 			goto slave_start;
3414 		} else
3415 			goto err_fw;
3416 	}
3417 
3418 	if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3419 		u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3420 						  existing_vfs, reset_flow);
3421 
3422 		if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3423 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3424 			dev->flags = dev_flags;
3425 			err = mlx4_cmd_init(dev);
3426 			if (err) {
3427 				/* Only VHCR is cleaned up, so could still
3428 				 * send FW commands
3429 				 */
3430 				mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3431 				goto err_close;
3432 			}
3433 		} else {
3434 			dev->flags = dev_flags;
3435 		}
3436 
3437 		if (!SRIOV_VALID_STATE(dev->flags)) {
3438 			mlx4_err(dev, "Invalid SRIOV state\n");
3439 			goto err_close;
3440 		}
3441 	}
3442 
3443 	/* check if the device is functioning at its maximum possible speed.
3444 	 * No return code for this call, just warn the user in case of PCI
3445 	 * express device capabilities are under-satisfied by the bus.
3446 	 */
3447 	if (!mlx4_is_slave(dev))
3448 		mlx4_check_pcie_caps(dev);
3449 
3450 	/* In master functions, the communication channel must be initialized
3451 	 * after obtaining its address from fw */
3452 	if (mlx4_is_master(dev)) {
3453 		if (dev->caps.num_ports < 2 &&
3454 		    num_vfs_argc > 1) {
3455 			err = -EINVAL;
3456 			mlx4_err(dev,
3457 				 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3458 				 dev->caps.num_ports);
3459 			goto err_close;
3460 		}
3461 		memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
3462 
3463 		for (i = 0;
3464 		     i < sizeof(dev->persist->nvfs)/
3465 		     sizeof(dev->persist->nvfs[0]); i++) {
3466 			unsigned j;
3467 
3468 			for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
3469 				dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3470 				dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3471 					dev->caps.num_ports;
3472 			}
3473 		}
3474 
3475 		/* In master functions, the communication channel
3476 		 * must be initialized after obtaining its address from fw
3477 		 */
3478 		err = mlx4_multi_func_init(dev);
3479 		if (err) {
3480 			mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3481 			goto err_close;
3482 		}
3483 	}
3484 
3485 	err = mlx4_alloc_eq_table(dev);
3486 	if (err)
3487 		goto err_master_mfunc;
3488 
3489 	bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
3490 	mutex_init(&priv->msix_ctl.pool_lock);
3491 
3492 	mlx4_enable_msi_x(dev);
3493 	if ((mlx4_is_mfunc(dev)) &&
3494 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
3495 		err = -ENOSYS;
3496 		mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
3497 		goto err_free_eq;
3498 	}
3499 
3500 	if (!mlx4_is_slave(dev)) {
3501 		err = mlx4_init_steering(dev);
3502 		if (err)
3503 			goto err_disable_msix;
3504 	}
3505 
3506 	err = mlx4_setup_hca(dev);
3507 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3508 	    !mlx4_is_mfunc(dev)) {
3509 		dev->flags &= ~MLX4_FLAG_MSI_X;
3510 		dev->caps.num_comp_vectors = 1;
3511 		pci_disable_msix(pdev);
3512 		err = mlx4_setup_hca(dev);
3513 	}
3514 
3515 	if (err)
3516 		goto err_steer;
3517 
3518 	mlx4_init_quotas(dev);
3519 	/* When PF resources are ready arm its comm channel to enable
3520 	 * getting commands
3521 	 */
3522 	if (mlx4_is_master(dev)) {
3523 		err = mlx4_ARM_COMM_CHANNEL(dev);
3524 		if (err) {
3525 			mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3526 				 err);
3527 			goto err_steer;
3528 		}
3529 	}
3530 
3531 	for (port = 1; port <= dev->caps.num_ports; port++) {
3532 		err = mlx4_init_port_info(dev, port);
3533 		if (err)
3534 			goto err_port;
3535 	}
3536 
3537 	priv->v2p.port1 = 1;
3538 	priv->v2p.port2 = 2;
3539 
3540 	err = mlx4_register_device(dev);
3541 	if (err)
3542 		goto err_port;
3543 
3544 	mlx4_request_modules(dev);
3545 
3546 	mlx4_sense_init(dev);
3547 	mlx4_start_sense(dev);
3548 
3549 	priv->removed = 0;
3550 
3551 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3552 		atomic_dec(&pf_loading);
3553 
3554 	kfree(dev_cap);
3555 	return 0;
3556 
3557 err_port:
3558 	for (--port; port >= 1; --port)
3559 		mlx4_cleanup_port_info(&priv->port[port]);
3560 
3561 	mlx4_cleanup_default_counters(dev);
3562 	if (!mlx4_is_slave(dev))
3563 		mlx4_cleanup_counters_table(dev);
3564 	mlx4_cleanup_qp_table(dev);
3565 	mlx4_cleanup_srq_table(dev);
3566 	mlx4_cleanup_cq_table(dev);
3567 	mlx4_cmd_use_polling(dev);
3568 	mlx4_cleanup_eq_table(dev);
3569 	mlx4_cleanup_mcg_table(dev);
3570 	mlx4_cleanup_mr_table(dev);
3571 	mlx4_cleanup_xrcd_table(dev);
3572 	mlx4_cleanup_pd_table(dev);
3573 	mlx4_cleanup_uar_table(dev);
3574 
3575 err_steer:
3576 	if (!mlx4_is_slave(dev))
3577 		mlx4_clear_steering(dev);
3578 
3579 err_disable_msix:
3580 	if (dev->flags & MLX4_FLAG_MSI_X)
3581 		pci_disable_msix(pdev);
3582 
3583 err_free_eq:
3584 	mlx4_free_eq_table(dev);
3585 
3586 err_master_mfunc:
3587 	if (mlx4_is_master(dev)) {
3588 		mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
3589 		mlx4_multi_func_cleanup(dev);
3590 	}
3591 
3592 	if (mlx4_is_slave(dev)) {
3593 		kfree(dev->caps.qp0_qkey);
3594 		kfree(dev->caps.qp0_tunnel);
3595 		kfree(dev->caps.qp0_proxy);
3596 		kfree(dev->caps.qp1_tunnel);
3597 		kfree(dev->caps.qp1_proxy);
3598 	}
3599 
3600 err_close:
3601 	mlx4_close_hca(dev);
3602 
3603 err_fw:
3604 	mlx4_close_fw(dev);
3605 
3606 err_mfunc:
3607 	if (mlx4_is_slave(dev))
3608 		mlx4_multi_func_cleanup(dev);
3609 
3610 err_cmd:
3611 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3612 
3613 err_sriov:
3614 	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3615 		pci_disable_sriov(pdev);
3616 		dev->flags &= ~MLX4_FLAG_SRIOV;
3617 	}
3618 
3619 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3620 		atomic_dec(&pf_loading);
3621 
3622 	kfree(priv->dev.dev_vfs);
3623 
3624 	if (!mlx4_is_slave(dev))
3625 		mlx4_free_ownership(dev);
3626 
3627 	kfree(dev_cap);
3628 	return err;
3629 }
3630 
3631 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3632 			   struct mlx4_priv *priv)
3633 {
3634 	int err;
3635 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3636 	int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3637 	const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3638 		{2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3639 	unsigned total_vfs = 0;
3640 	unsigned int i;
3641 
3642 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3643 
3644 	err = mlx4_pci_enable_device(&priv->dev);
3645 	if (err) {
3646 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3647 		return err;
3648 	}
3649 
3650 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3651 	 * per port, we must limit the number of VFs to 63 (since their are
3652 	 * 128 MACs)
3653 	 */
3654 	for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
3655 	     total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3656 		nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3657 		if (nvfs[i] < 0) {
3658 			dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3659 			err = -EINVAL;
3660 			goto err_disable_pdev;
3661 		}
3662 	}
3663 	for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
3664 	     i++) {
3665 		prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3666 		if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3667 			dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3668 			err = -EINVAL;
3669 			goto err_disable_pdev;
3670 		}
3671 	}
3672 	if (total_vfs > MLX4_MAX_NUM_VF) {
3673 		dev_err(&pdev->dev,
3674 			"Requested more VF's (%d) than allowed by hw (%d)\n",
3675 			total_vfs, MLX4_MAX_NUM_VF);
3676 		err = -EINVAL;
3677 		goto err_disable_pdev;
3678 	}
3679 
3680 	for (i = 0; i < MLX4_MAX_PORTS; i++) {
3681 		if (nvfs[i] + nvfs[2] > MLX4_MAX_NUM_VF_P_PORT) {
3682 			dev_err(&pdev->dev,
3683 				"Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3684 				nvfs[i] + nvfs[2], i + 1,
3685 				MLX4_MAX_NUM_VF_P_PORT);
3686 			err = -EINVAL;
3687 			goto err_disable_pdev;
3688 		}
3689 	}
3690 
3691 	/* Check for BARs. */
3692 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3693 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3694 		dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3695 			pci_dev_data, pci_resource_flags(pdev, 0));
3696 		err = -ENODEV;
3697 		goto err_disable_pdev;
3698 	}
3699 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3700 		dev_err(&pdev->dev, "Missing UAR, aborting\n");
3701 		err = -ENODEV;
3702 		goto err_disable_pdev;
3703 	}
3704 
3705 	err = pci_request_regions(pdev, DRV_NAME);
3706 	if (err) {
3707 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3708 		goto err_disable_pdev;
3709 	}
3710 
3711 	pci_set_master(pdev);
3712 
3713 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3714 	if (err) {
3715 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3716 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3717 		if (err) {
3718 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3719 			goto err_release_regions;
3720 		}
3721 	}
3722 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3723 	if (err) {
3724 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3725 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3726 		if (err) {
3727 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3728 			goto err_release_regions;
3729 		}
3730 	}
3731 
3732 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
3733 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3734 	/* Detect if this device is a virtual function */
3735 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3736 		/* When acting as pf, we normally skip vfs unless explicitly
3737 		 * requested to probe them.
3738 		 */
3739 		if (total_vfs) {
3740 			unsigned vfs_offset = 0;
3741 
3742 			for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
3743 			     vfs_offset + nvfs[i] < extended_func_num(pdev);
3744 			     vfs_offset += nvfs[i], i++)
3745 				;
3746 			if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
3747 				err = -ENODEV;
3748 				goto err_release_regions;
3749 			}
3750 			if ((extended_func_num(pdev) - vfs_offset)
3751 			    > prb_vf[i]) {
3752 				dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3753 					 extended_func_num(pdev));
3754 				err = -ENODEV;
3755 				goto err_release_regions;
3756 			}
3757 		}
3758 	}
3759 
3760 	err = mlx4_catas_init(&priv->dev);
3761 	if (err)
3762 		goto err_release_regions;
3763 
3764 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3765 	if (err)
3766 		goto err_catas;
3767 
3768 	return 0;
3769 
3770 err_catas:
3771 	mlx4_catas_end(&priv->dev);
3772 
3773 err_release_regions:
3774 	pci_release_regions(pdev);
3775 
3776 err_disable_pdev:
3777 	mlx4_pci_disable_device(&priv->dev);
3778 	pci_set_drvdata(pdev, NULL);
3779 	return err;
3780 }
3781 
3782 static int mlx4_devlink_port_type_set(struct devlink_port *devlink_port,
3783 				      enum devlink_port_type port_type)
3784 {
3785 	struct mlx4_port_info *info = container_of(devlink_port,
3786 						   struct mlx4_port_info,
3787 						   devlink_port);
3788 	enum mlx4_port_type mlx4_port_type;
3789 
3790 	switch (port_type) {
3791 	case DEVLINK_PORT_TYPE_AUTO:
3792 		mlx4_port_type = MLX4_PORT_TYPE_AUTO;
3793 		break;
3794 	case DEVLINK_PORT_TYPE_ETH:
3795 		mlx4_port_type = MLX4_PORT_TYPE_ETH;
3796 		break;
3797 	case DEVLINK_PORT_TYPE_IB:
3798 		mlx4_port_type = MLX4_PORT_TYPE_IB;
3799 		break;
3800 	default:
3801 		return -EOPNOTSUPP;
3802 	}
3803 
3804 	return __set_port_type(info, mlx4_port_type);
3805 }
3806 
3807 static const struct devlink_ops mlx4_devlink_ops = {
3808 	.port_type_set	= mlx4_devlink_port_type_set,
3809 };
3810 
3811 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3812 {
3813 	struct devlink *devlink;
3814 	struct mlx4_priv *priv;
3815 	struct mlx4_dev *dev;
3816 	int ret;
3817 
3818 	printk_once(KERN_INFO "%s", mlx4_version);
3819 
3820 	devlink = devlink_alloc(&mlx4_devlink_ops, sizeof(*priv));
3821 	if (!devlink)
3822 		return -ENOMEM;
3823 	priv = devlink_priv(devlink);
3824 
3825 	dev       = &priv->dev;
3826 	dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3827 	if (!dev->persist) {
3828 		ret = -ENOMEM;
3829 		goto err_devlink_free;
3830 	}
3831 	dev->persist->pdev = pdev;
3832 	dev->persist->dev = dev;
3833 	pci_set_drvdata(pdev, dev->persist);
3834 	priv->pci_dev_data = id->driver_data;
3835 	mutex_init(&dev->persist->device_state_mutex);
3836 	mutex_init(&dev->persist->interface_state_mutex);
3837 	mutex_init(&dev->persist->pci_status_mutex);
3838 
3839 	ret = devlink_register(devlink, &pdev->dev);
3840 	if (ret)
3841 		goto err_persist_free;
3842 
3843 	ret =  __mlx4_init_one(pdev, id->driver_data, priv);
3844 	if (ret)
3845 		goto err_devlink_unregister;
3846 
3847 	pci_save_state(pdev);
3848 	return 0;
3849 
3850 err_devlink_unregister:
3851 	devlink_unregister(devlink);
3852 err_persist_free:
3853 	kfree(dev->persist);
3854 err_devlink_free:
3855 	devlink_free(devlink);
3856 	return ret;
3857 }
3858 
3859 static void mlx4_clean_dev(struct mlx4_dev *dev)
3860 {
3861 	struct mlx4_dev_persistent *persist = dev->persist;
3862 	struct mlx4_priv *priv = mlx4_priv(dev);
3863 	unsigned long	flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
3864 
3865 	memset(priv, 0, sizeof(*priv));
3866 	priv->dev.persist = persist;
3867 	priv->dev.flags = flags;
3868 }
3869 
3870 static void mlx4_unload_one(struct pci_dev *pdev)
3871 {
3872 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3873 	struct mlx4_dev  *dev  = persist->dev;
3874 	struct mlx4_priv *priv = mlx4_priv(dev);
3875 	int               pci_dev_data;
3876 	int p, i;
3877 
3878 	if (priv->removed)
3879 		return;
3880 
3881 	/* saving current ports type for further use */
3882 	for (i = 0; i < dev->caps.num_ports; i++) {
3883 		dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
3884 		dev->persist->curr_port_poss_type[i] = dev->caps.
3885 						       possible_type[i + 1];
3886 	}
3887 
3888 	pci_dev_data = priv->pci_dev_data;
3889 
3890 	mlx4_stop_sense(dev);
3891 	mlx4_unregister_device(dev);
3892 
3893 	for (p = 1; p <= dev->caps.num_ports; p++) {
3894 		mlx4_cleanup_port_info(&priv->port[p]);
3895 		mlx4_CLOSE_PORT(dev, p);
3896 	}
3897 
3898 	if (mlx4_is_master(dev))
3899 		mlx4_free_resource_tracker(dev,
3900 					   RES_TR_FREE_SLAVES_ONLY);
3901 
3902 	mlx4_cleanup_default_counters(dev);
3903 	if (!mlx4_is_slave(dev))
3904 		mlx4_cleanup_counters_table(dev);
3905 	mlx4_cleanup_qp_table(dev);
3906 	mlx4_cleanup_srq_table(dev);
3907 	mlx4_cleanup_cq_table(dev);
3908 	mlx4_cmd_use_polling(dev);
3909 	mlx4_cleanup_eq_table(dev);
3910 	mlx4_cleanup_mcg_table(dev);
3911 	mlx4_cleanup_mr_table(dev);
3912 	mlx4_cleanup_xrcd_table(dev);
3913 	mlx4_cleanup_pd_table(dev);
3914 
3915 	if (mlx4_is_master(dev))
3916 		mlx4_free_resource_tracker(dev,
3917 					   RES_TR_FREE_STRUCTS_ONLY);
3918 
3919 	iounmap(priv->kar);
3920 	mlx4_uar_free(dev, &priv->driver_uar);
3921 	mlx4_cleanup_uar_table(dev);
3922 	if (!mlx4_is_slave(dev))
3923 		mlx4_clear_steering(dev);
3924 	mlx4_free_eq_table(dev);
3925 	if (mlx4_is_master(dev))
3926 		mlx4_multi_func_cleanup(dev);
3927 	mlx4_close_hca(dev);
3928 	mlx4_close_fw(dev);
3929 	if (mlx4_is_slave(dev))
3930 		mlx4_multi_func_cleanup(dev);
3931 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3932 
3933 	if (dev->flags & MLX4_FLAG_MSI_X)
3934 		pci_disable_msix(pdev);
3935 
3936 	if (!mlx4_is_slave(dev))
3937 		mlx4_free_ownership(dev);
3938 
3939 	kfree(dev->caps.qp0_qkey);
3940 	kfree(dev->caps.qp0_tunnel);
3941 	kfree(dev->caps.qp0_proxy);
3942 	kfree(dev->caps.qp1_tunnel);
3943 	kfree(dev->caps.qp1_proxy);
3944 	kfree(dev->dev_vfs);
3945 
3946 	mlx4_clean_dev(dev);
3947 	priv->pci_dev_data = pci_dev_data;
3948 	priv->removed = 1;
3949 }
3950 
3951 static void mlx4_remove_one(struct pci_dev *pdev)
3952 {
3953 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3954 	struct mlx4_dev  *dev  = persist->dev;
3955 	struct mlx4_priv *priv = mlx4_priv(dev);
3956 	struct devlink *devlink = priv_to_devlink(priv);
3957 	int active_vfs = 0;
3958 
3959 	mutex_lock(&persist->interface_state_mutex);
3960 	persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
3961 	mutex_unlock(&persist->interface_state_mutex);
3962 
3963 	/* Disabling SR-IOV is not allowed while there are active vf's */
3964 	if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
3965 		active_vfs = mlx4_how_many_lives_vf(dev);
3966 		if (active_vfs) {
3967 			pr_warn("Removing PF when there are active VF's !!\n");
3968 			pr_warn("Will not disable SR-IOV.\n");
3969 		}
3970 	}
3971 
3972 	/* device marked to be under deletion running now without the lock
3973 	 * letting other tasks to be terminated
3974 	 */
3975 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3976 		mlx4_unload_one(pdev);
3977 	else
3978 		mlx4_info(dev, "%s: interface is down\n", __func__);
3979 	mlx4_catas_end(dev);
3980 	if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
3981 		mlx4_warn(dev, "Disabling SR-IOV\n");
3982 		pci_disable_sriov(pdev);
3983 	}
3984 
3985 	pci_release_regions(pdev);
3986 	mlx4_pci_disable_device(dev);
3987 	devlink_unregister(devlink);
3988 	kfree(dev->persist);
3989 	devlink_free(devlink);
3990 	pci_set_drvdata(pdev, NULL);
3991 }
3992 
3993 static int restore_current_port_types(struct mlx4_dev *dev,
3994 				      enum mlx4_port_type *types,
3995 				      enum mlx4_port_type *poss_types)
3996 {
3997 	struct mlx4_priv *priv = mlx4_priv(dev);
3998 	int err, i;
3999 
4000 	mlx4_stop_sense(dev);
4001 
4002 	mutex_lock(&priv->port_mutex);
4003 	for (i = 0; i < dev->caps.num_ports; i++)
4004 		dev->caps.possible_type[i + 1] = poss_types[i];
4005 	err = mlx4_change_port_types(dev, types);
4006 	mlx4_start_sense(dev);
4007 	mutex_unlock(&priv->port_mutex);
4008 
4009 	return err;
4010 }
4011 
4012 int mlx4_restart_one(struct pci_dev *pdev)
4013 {
4014 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4015 	struct mlx4_dev	 *dev  = persist->dev;
4016 	struct mlx4_priv *priv = mlx4_priv(dev);
4017 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4018 	int pci_dev_data, err, total_vfs;
4019 
4020 	pci_dev_data = priv->pci_dev_data;
4021 	total_vfs = dev->persist->num_vfs;
4022 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4023 
4024 	mlx4_unload_one(pdev);
4025 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
4026 	if (err) {
4027 		mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
4028 			 __func__, pci_name(pdev), err);
4029 		return err;
4030 	}
4031 
4032 	err = restore_current_port_types(dev, dev->persist->curr_port_type,
4033 					 dev->persist->curr_port_poss_type);
4034 	if (err)
4035 		mlx4_err(dev, "could not restore original port types (%d)\n",
4036 			 err);
4037 
4038 	return err;
4039 }
4040 
4041 #define MLX_SP(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_FORCE_SENSE_PORT }
4042 #define MLX_VF(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_IS_VF }
4043 #define MLX_GN(id) { PCI_VDEVICE(MELLANOX, id), 0 }
4044 
4045 static const struct pci_device_id mlx4_pci_table[] = {
4046 	/* MT25408 "Hermon" */
4047 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_SDR),	/* SDR */
4048 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR),	/* DDR */
4049 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR),	/* QDR */
4050 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR_GEN2), /* DDR Gen2 */
4051 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR_GEN2),	/* QDR Gen2 */
4052 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN),	/* EN 10GigE */
4053 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN_GEN2),  /* EN 10GigE Gen2 */
4054 	/* MT25458 ConnectX EN 10GBASE-T */
4055 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN),
4056 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_T_GEN2),	/* Gen2 */
4057 	/* MT26468 ConnectX EN 10GigE PCIe Gen2*/
4058 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_GEN2),
4059 	/* MT26438 ConnectX EN 40GigE PCIe Gen2 5GT/s */
4060 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_5_GEN2),
4061 	/* MT26478 ConnectX2 40GigE PCIe Gen2 */
4062 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX2),
4063 	/* MT25400 Family [ConnectX-2] */
4064 	MLX_VF(0x1002),					/* Virtual Function */
4065 	/* MT27500 Family [ConnectX-3] */
4066 	MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3),
4067 	MLX_VF(0x1004),					/* Virtual Function */
4068 	MLX_GN(0x1005),					/* MT27510 Family */
4069 	MLX_GN(0x1006),					/* MT27511 Family */
4070 	MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3_PRO),	/* MT27520 Family */
4071 	MLX_GN(0x1008),					/* MT27521 Family */
4072 	MLX_GN(0x1009),					/* MT27530 Family */
4073 	MLX_GN(0x100a),					/* MT27531 Family */
4074 	MLX_GN(0x100b),					/* MT27540 Family */
4075 	MLX_GN(0x100c),					/* MT27541 Family */
4076 	MLX_GN(0x100d),					/* MT27550 Family */
4077 	MLX_GN(0x100e),					/* MT27551 Family */
4078 	MLX_GN(0x100f),					/* MT27560 Family */
4079 	MLX_GN(0x1010),					/* MT27561 Family */
4080 
4081 	/*
4082 	 * See the mellanox_check_broken_intx_masking() quirk when
4083 	 * adding devices
4084 	 */
4085 
4086 	{ 0, }
4087 };
4088 
4089 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
4090 
4091 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
4092 					      pci_channel_state_t state)
4093 {
4094 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4095 
4096 	mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
4097 	mlx4_enter_error_state(persist);
4098 
4099 	mutex_lock(&persist->interface_state_mutex);
4100 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4101 		mlx4_unload_one(pdev);
4102 
4103 	mutex_unlock(&persist->interface_state_mutex);
4104 	if (state == pci_channel_io_perm_failure)
4105 		return PCI_ERS_RESULT_DISCONNECT;
4106 
4107 	mlx4_pci_disable_device(persist->dev);
4108 	return PCI_ERS_RESULT_NEED_RESET;
4109 }
4110 
4111 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
4112 {
4113 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4114 	struct mlx4_dev	 *dev  = persist->dev;
4115 	int err;
4116 
4117 	mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
4118 	err = mlx4_pci_enable_device(dev);
4119 	if (err) {
4120 		mlx4_err(dev, "Can not re-enable device, err=%d\n", err);
4121 		return PCI_ERS_RESULT_DISCONNECT;
4122 	}
4123 
4124 	pci_set_master(pdev);
4125 	pci_restore_state(pdev);
4126 	pci_save_state(pdev);
4127 	return PCI_ERS_RESULT_RECOVERED;
4128 }
4129 
4130 static void mlx4_pci_resume(struct pci_dev *pdev)
4131 {
4132 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4133 	struct mlx4_dev	 *dev  = persist->dev;
4134 	struct mlx4_priv *priv = mlx4_priv(dev);
4135 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4136 	int total_vfs;
4137 	int err;
4138 
4139 	mlx4_err(dev, "%s was called\n", __func__);
4140 	total_vfs = dev->persist->num_vfs;
4141 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4142 
4143 	mutex_lock(&persist->interface_state_mutex);
4144 	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4145 		err = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
4146 				    priv, 1);
4147 		if (err) {
4148 			mlx4_err(dev, "%s: mlx4_load_one failed, err=%d\n",
4149 				 __func__,  err);
4150 			goto end;
4151 		}
4152 
4153 		err = restore_current_port_types(dev, dev->persist->
4154 						 curr_port_type, dev->persist->
4155 						 curr_port_poss_type);
4156 		if (err)
4157 			mlx4_err(dev, "could not restore original port types (%d)\n", err);
4158 	}
4159 end:
4160 	mutex_unlock(&persist->interface_state_mutex);
4161 
4162 }
4163 
4164 static void mlx4_shutdown(struct pci_dev *pdev)
4165 {
4166 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4167 
4168 	mlx4_info(persist->dev, "mlx4_shutdown was called\n");
4169 	mutex_lock(&persist->interface_state_mutex);
4170 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4171 		mlx4_unload_one(pdev);
4172 	mutex_unlock(&persist->interface_state_mutex);
4173 }
4174 
4175 static const struct pci_error_handlers mlx4_err_handler = {
4176 	.error_detected = mlx4_pci_err_detected,
4177 	.slot_reset     = mlx4_pci_slot_reset,
4178 	.resume		= mlx4_pci_resume,
4179 };
4180 
4181 static struct pci_driver mlx4_driver = {
4182 	.name		= DRV_NAME,
4183 	.id_table	= mlx4_pci_table,
4184 	.probe		= mlx4_init_one,
4185 	.shutdown	= mlx4_shutdown,
4186 	.remove		= mlx4_remove_one,
4187 	.err_handler    = &mlx4_err_handler,
4188 };
4189 
4190 static int __init mlx4_verify_params(void)
4191 {
4192 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
4193 		pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
4194 		return -1;
4195 	}
4196 
4197 	if (log_num_vlan != 0)
4198 		pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
4199 			MLX4_LOG_NUM_VLANS);
4200 
4201 	if (use_prio != 0)
4202 		pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
4203 
4204 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
4205 		pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
4206 			log_mtts_per_seg);
4207 		return -1;
4208 	}
4209 
4210 	/* Check if module param for ports type has legal combination */
4211 	if (port_type_array[0] == false && port_type_array[1] == true) {
4212 		pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
4213 		port_type_array[0] = true;
4214 	}
4215 
4216 	if (mlx4_log_num_mgm_entry_size < -7 ||
4217 	    (mlx4_log_num_mgm_entry_size > 0 &&
4218 	     (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
4219 	      mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
4220 		pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
4221 			mlx4_log_num_mgm_entry_size,
4222 			MLX4_MIN_MGM_LOG_ENTRY_SIZE,
4223 			MLX4_MAX_MGM_LOG_ENTRY_SIZE);
4224 		return -1;
4225 	}
4226 
4227 	return 0;
4228 }
4229 
4230 static int __init mlx4_init(void)
4231 {
4232 	int ret;
4233 
4234 	if (mlx4_verify_params())
4235 		return -EINVAL;
4236 
4237 
4238 	mlx4_wq = create_singlethread_workqueue("mlx4");
4239 	if (!mlx4_wq)
4240 		return -ENOMEM;
4241 
4242 	ret = pci_register_driver(&mlx4_driver);
4243 	if (ret < 0)
4244 		destroy_workqueue(mlx4_wq);
4245 	return ret < 0 ? ret : 0;
4246 }
4247 
4248 static void __exit mlx4_cleanup(void)
4249 {
4250 	pci_unregister_driver(&mlx4_driver);
4251 	destroy_workqueue(mlx4_wq);
4252 }
4253 
4254 module_init(mlx4_init);
4255 module_exit(mlx4_cleanup);
4256