1 /*
2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4  * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
5  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * OpenIB.org BSD license below:
12  *
13  *     Redistribution and use in source and binary forms, with or
14  *     without modification, are permitted provided that the following
15  *     conditions are met:
16  *
17  *      - Redistributions of source code must retain the above
18  *        copyright notice, this list of conditions and the following
19  *        disclaimer.
20  *
21  *      - Redistributions in binary form must reproduce the above
22  *        copyright notice, this list of conditions and the following
23  *        disclaimer in the documentation and/or other materials
24  *        provided with the distribution.
25  *
26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33  * SOFTWARE.
34  */
35 
36 #include <linux/module.h>
37 #include <linux/init.h>
38 #include <linux/errno.h>
39 #include <linux/pci.h>
40 #include <linux/dma-mapping.h>
41 #include <linux/slab.h>
42 #include <linux/io-mapping.h>
43 #include <linux/delay.h>
44 #include <linux/kmod.h>
45 #include <net/devlink.h>
46 
47 #include <linux/mlx4/device.h>
48 #include <linux/mlx4/doorbell.h>
49 
50 #include "mlx4.h"
51 #include "fw.h"
52 #include "icm.h"
53 
54 MODULE_AUTHOR("Roland Dreier");
55 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
56 MODULE_LICENSE("Dual BSD/GPL");
57 MODULE_VERSION(DRV_VERSION);
58 
59 struct workqueue_struct *mlx4_wq;
60 
61 #ifdef CONFIG_MLX4_DEBUG
62 
63 int mlx4_debug_level = 0;
64 module_param_named(debug_level, mlx4_debug_level, int, 0644);
65 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
66 
67 #endif /* CONFIG_MLX4_DEBUG */
68 
69 #ifdef CONFIG_PCI_MSI
70 
71 static int msi_x = 1;
72 module_param(msi_x, int, 0444);
73 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
74 
75 #else /* CONFIG_PCI_MSI */
76 
77 #define msi_x (0)
78 
79 #endif /* CONFIG_PCI_MSI */
80 
81 static uint8_t num_vfs[3] = {0, 0, 0};
82 static int num_vfs_argc;
83 module_param_array(num_vfs, byte , &num_vfs_argc, 0444);
84 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
85 			  "num_vfs=port1,port2,port1+2");
86 
87 static uint8_t probe_vf[3] = {0, 0, 0};
88 static int probe_vfs_argc;
89 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
90 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
91 			   "probe_vf=port1,port2,port1+2");
92 
93 int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
94 module_param_named(log_num_mgm_entry_size,
95 			mlx4_log_num_mgm_entry_size, int, 0444);
96 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
97 					 " of qp per mcg, for example:"
98 					 " 10 gives 248.range: 7 <="
99 					 " log_num_mgm_entry_size <= 12."
100 					 " To activate device managed"
101 					 " flow steering when available, set to -1");
102 
103 static bool enable_64b_cqe_eqe = true;
104 module_param(enable_64b_cqe_eqe, bool, 0444);
105 MODULE_PARM_DESC(enable_64b_cqe_eqe,
106 		 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
107 
108 static bool enable_4k_uar;
109 module_param(enable_4k_uar, bool, 0444);
110 MODULE_PARM_DESC(enable_4k_uar,
111 		 "Enable using 4K UAR. Should not be enabled if have VFs which do not support 4K UARs (default: false)");
112 
113 #define PF_CONTEXT_BEHAVIOUR_MASK	(MLX4_FUNC_CAP_64B_EQE_CQE | \
114 					 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
115 					 MLX4_FUNC_CAP_DMFS_A0_STATIC)
116 
117 #define RESET_PERSIST_MASK_FLAGS	(MLX4_FLAG_SRIOV)
118 
119 static char mlx4_version[] =
120 	DRV_NAME ": Mellanox ConnectX core driver v"
121 	DRV_VERSION " (" DRV_RELDATE ")\n";
122 
123 static struct mlx4_profile default_profile = {
124 	.num_qp		= 1 << 18,
125 	.num_srq	= 1 << 16,
126 	.rdmarc_per_qp	= 1 << 4,
127 	.num_cq		= 1 << 16,
128 	.num_mcg	= 1 << 13,
129 	.num_mpt	= 1 << 19,
130 	.num_mtt	= 1 << 20, /* It is really num mtt segements */
131 };
132 
133 static struct mlx4_profile low_mem_profile = {
134 	.num_qp		= 1 << 17,
135 	.num_srq	= 1 << 6,
136 	.rdmarc_per_qp	= 1 << 4,
137 	.num_cq		= 1 << 8,
138 	.num_mcg	= 1 << 8,
139 	.num_mpt	= 1 << 9,
140 	.num_mtt	= 1 << 7,
141 };
142 
143 static int log_num_mac = 7;
144 module_param_named(log_num_mac, log_num_mac, int, 0444);
145 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
146 
147 static int log_num_vlan;
148 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
149 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
150 /* Log2 max number of VLANs per ETH port (0-7) */
151 #define MLX4_LOG_NUM_VLANS 7
152 #define MLX4_MIN_LOG_NUM_VLANS 0
153 #define MLX4_MIN_LOG_NUM_MAC 1
154 
155 static bool use_prio;
156 module_param_named(use_prio, use_prio, bool, 0444);
157 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
158 
159 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
160 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
161 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
162 
163 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
164 static int arr_argc = 2;
165 module_param_array(port_type_array, int, &arr_argc, 0444);
166 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
167 				"1 for IB, 2 for Ethernet");
168 
169 struct mlx4_port_config {
170 	struct list_head list;
171 	enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
172 	struct pci_dev *pdev;
173 };
174 
175 static atomic_t pf_loading = ATOMIC_INIT(0);
176 
177 static inline void mlx4_set_num_reserved_uars(struct mlx4_dev *dev,
178 					      struct mlx4_dev_cap *dev_cap)
179 {
180 	/* The reserved_uars is calculated by system page size unit.
181 	 * Therefore, adjustment is added when the uar page size is less
182 	 * than the system page size
183 	 */
184 	dev->caps.reserved_uars	=
185 		max_t(int,
186 		      mlx4_get_num_reserved_uar(dev),
187 		      dev_cap->reserved_uars /
188 			(1 << (PAGE_SHIFT - dev->uar_page_shift)));
189 }
190 
191 int mlx4_check_port_params(struct mlx4_dev *dev,
192 			   enum mlx4_port_type *port_type)
193 {
194 	int i;
195 
196 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
197 		for (i = 0; i < dev->caps.num_ports - 1; i++) {
198 			if (port_type[i] != port_type[i + 1]) {
199 				mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
200 				return -EINVAL;
201 			}
202 		}
203 	}
204 
205 	for (i = 0; i < dev->caps.num_ports; i++) {
206 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
207 			mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
208 				 i + 1);
209 			return -EINVAL;
210 		}
211 	}
212 	return 0;
213 }
214 
215 static void mlx4_set_port_mask(struct mlx4_dev *dev)
216 {
217 	int i;
218 
219 	for (i = 1; i <= dev->caps.num_ports; ++i)
220 		dev->caps.port_mask[i] = dev->caps.port_type[i];
221 }
222 
223 enum {
224 	MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
225 };
226 
227 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
228 {
229 	int err = 0;
230 	struct mlx4_func func;
231 
232 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
233 		err = mlx4_QUERY_FUNC(dev, &func, 0);
234 		if (err) {
235 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
236 			return err;
237 		}
238 		dev_cap->max_eqs = func.max_eq;
239 		dev_cap->reserved_eqs = func.rsvd_eqs;
240 		dev_cap->reserved_uars = func.rsvd_uars;
241 		err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
242 	}
243 	return err;
244 }
245 
246 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
247 {
248 	struct mlx4_caps *dev_cap = &dev->caps;
249 
250 	/* FW not supporting or cancelled by user */
251 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
252 	    !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
253 		return;
254 
255 	/* Must have 64B CQE_EQE enabled by FW to use bigger stride
256 	 * When FW has NCSI it may decide not to report 64B CQE/EQEs
257 	 */
258 	if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
259 	    !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
260 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
261 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
262 		return;
263 	}
264 
265 	if (cache_line_size() == 128 || cache_line_size() == 256) {
266 		mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
267 		/* Changing the real data inside CQE size to 32B */
268 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
269 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
270 
271 		if (mlx4_is_master(dev))
272 			dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
273 	} else {
274 		if (cache_line_size() != 32  && cache_line_size() != 64)
275 			mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
276 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
277 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
278 	}
279 }
280 
281 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
282 			  struct mlx4_port_cap *port_cap)
283 {
284 	dev->caps.vl_cap[port]	    = port_cap->max_vl;
285 	dev->caps.ib_mtu_cap[port]	    = port_cap->ib_mtu;
286 	dev->phys_caps.gid_phys_table_len[port]  = port_cap->max_gids;
287 	dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
288 	/* set gid and pkey table operating lengths by default
289 	 * to non-sriov values
290 	 */
291 	dev->caps.gid_table_len[port]  = port_cap->max_gids;
292 	dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
293 	dev->caps.port_width_cap[port] = port_cap->max_port_width;
294 	dev->caps.eth_mtu_cap[port]    = port_cap->eth_mtu;
295 	dev->caps.def_mac[port]        = port_cap->def_mac;
296 	dev->caps.supported_type[port] = port_cap->supported_port_types;
297 	dev->caps.suggested_type[port] = port_cap->suggested_type;
298 	dev->caps.default_sense[port] = port_cap->default_sense;
299 	dev->caps.trans_type[port]	    = port_cap->trans_type;
300 	dev->caps.vendor_oui[port]     = port_cap->vendor_oui;
301 	dev->caps.wavelength[port]     = port_cap->wavelength;
302 	dev->caps.trans_code[port]     = port_cap->trans_code;
303 
304 	return 0;
305 }
306 
307 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
308 			 struct mlx4_port_cap *port_cap)
309 {
310 	int err = 0;
311 
312 	err = mlx4_QUERY_PORT(dev, port, port_cap);
313 
314 	if (err)
315 		mlx4_err(dev, "QUERY_PORT command failed.\n");
316 
317 	return err;
318 }
319 
320 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
321 {
322 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
323 		return;
324 
325 	if (mlx4_is_mfunc(dev)) {
326 		mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
327 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
328 		return;
329 	}
330 
331 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
332 		mlx4_dbg(dev,
333 			 "Keep FCS is not supported - Disabling Ignore FCS");
334 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
335 		return;
336 	}
337 }
338 
339 #define MLX4_A0_STEERING_TABLE_SIZE	256
340 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
341 {
342 	int err;
343 	int i;
344 
345 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
346 	if (err) {
347 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
348 		return err;
349 	}
350 	mlx4_dev_cap_dump(dev, dev_cap);
351 
352 	if (dev_cap->min_page_sz > PAGE_SIZE) {
353 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
354 			 dev_cap->min_page_sz, PAGE_SIZE);
355 		return -ENODEV;
356 	}
357 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
358 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
359 			 dev_cap->num_ports, MLX4_MAX_PORTS);
360 		return -ENODEV;
361 	}
362 
363 	if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
364 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
365 			 dev_cap->uar_size,
366 			 (unsigned long long)
367 			 pci_resource_len(dev->persist->pdev, 2));
368 		return -ENODEV;
369 	}
370 
371 	dev->caps.num_ports	     = dev_cap->num_ports;
372 	dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
373 	dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
374 				      dev->caps.num_sys_eqs :
375 				      MLX4_MAX_EQ_NUM;
376 	for (i = 1; i <= dev->caps.num_ports; ++i) {
377 		err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
378 		if (err) {
379 			mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
380 			return err;
381 		}
382 	}
383 
384 	dev->caps.uar_page_size	     = PAGE_SIZE;
385 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
386 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
387 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
388 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
389 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
390 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
391 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
392 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
393 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
394 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
395 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
396 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
397 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
398 	/*
399 	 * Subtract 1 from the limit because we need to allocate a
400 	 * spare CQE so the HCA HW can tell the difference between an
401 	 * empty CQ and a full CQ.
402 	 */
403 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
404 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
405 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
406 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
407 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
408 
409 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
410 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
411 					dev_cap->reserved_xrcds : 0;
412 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
413 					dev_cap->max_xrcds : 0;
414 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
415 
416 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
417 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
418 	dev->caps.flags		     = dev_cap->flags;
419 	dev->caps.flags2	     = dev_cap->flags2;
420 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
421 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
422 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
423 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
424 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
425 
426 	/* Save uar page shift */
427 	if (!mlx4_is_slave(dev)) {
428 		/* Virtual PCI function needs to determine UAR page size from
429 		 * firmware. Only master PCI function can set the uar page size
430 		 */
431 		if (enable_4k_uar)
432 			dev->uar_page_shift = DEFAULT_UAR_PAGE_SHIFT;
433 		else
434 			dev->uar_page_shift = PAGE_SHIFT;
435 
436 		mlx4_set_num_reserved_uars(dev, dev_cap);
437 	}
438 
439 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN) {
440 		struct mlx4_init_hca_param hca_param;
441 
442 		memset(&hca_param, 0, sizeof(hca_param));
443 		err = mlx4_QUERY_HCA(dev, &hca_param);
444 		/* Turn off PHV_EN flag in case phv_check_en is set.
445 		 * phv_check_en is a HW check that parse the packet and verify
446 		 * phv bit was reported correctly in the wqe. To allow QinQ
447 		 * PHV_EN flag should be set and phv_check_en must be cleared
448 		 * otherwise QinQ packets will be drop by the HW.
449 		 */
450 		if (err || hca_param.phv_check_en)
451 			dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_PHV_EN;
452 	}
453 
454 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
455 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
456 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
457 	/* Don't do sense port on multifunction devices (for now at least) */
458 	if (mlx4_is_mfunc(dev))
459 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
460 
461 	if (mlx4_low_memory_profile()) {
462 		dev->caps.log_num_macs  = MLX4_MIN_LOG_NUM_MAC;
463 		dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
464 	} else {
465 		dev->caps.log_num_macs  = log_num_mac;
466 		dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
467 	}
468 
469 	for (i = 1; i <= dev->caps.num_ports; ++i) {
470 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
471 		if (dev->caps.supported_type[i]) {
472 			/* if only ETH is supported - assign ETH */
473 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
474 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
475 			/* if only IB is supported, assign IB */
476 			else if (dev->caps.supported_type[i] ==
477 				 MLX4_PORT_TYPE_IB)
478 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
479 			else {
480 				/* if IB and ETH are supported, we set the port
481 				 * type according to user selection of port type;
482 				 * if user selected none, take the FW hint */
483 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
484 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
485 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
486 				else
487 					dev->caps.port_type[i] = port_type_array[i - 1];
488 			}
489 		}
490 		/*
491 		 * Link sensing is allowed on the port if 3 conditions are true:
492 		 * 1. Both protocols are supported on the port.
493 		 * 2. Different types are supported on the port
494 		 * 3. FW declared that it supports link sensing
495 		 */
496 		mlx4_priv(dev)->sense.sense_allowed[i] =
497 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
498 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
499 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
500 
501 		/*
502 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
503 		 * and perform sense_port FW command to try and set the correct
504 		 * port type from beginning
505 		 */
506 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
507 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
508 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
509 			mlx4_SENSE_PORT(dev, i, &sensed_port);
510 			if (sensed_port != MLX4_PORT_TYPE_NONE)
511 				dev->caps.port_type[i] = sensed_port;
512 		} else {
513 			dev->caps.possible_type[i] = dev->caps.port_type[i];
514 		}
515 
516 		if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
517 			dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
518 			mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
519 				  i, 1 << dev->caps.log_num_macs);
520 		}
521 		if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
522 			dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
523 			mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
524 				  i, 1 << dev->caps.log_num_vlans);
525 		}
526 	}
527 
528 	if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
529 	    (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
530 	    (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
531 		mlx4_warn(dev,
532 			  "Granular QoS per VF not supported with IB/Eth configuration\n");
533 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
534 	}
535 
536 	dev->caps.max_counters = dev_cap->max_counters;
537 
538 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
539 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
540 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
541 		(1 << dev->caps.log_num_macs) *
542 		(1 << dev->caps.log_num_vlans) *
543 		dev->caps.num_ports;
544 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
545 
546 	if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
547 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
548 		dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
549 	else
550 		dev->caps.dmfs_high_rate_qpn_base =
551 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
552 
553 	if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
554 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
555 		dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
556 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
557 		dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
558 	} else {
559 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
560 		dev->caps.dmfs_high_rate_qpn_base =
561 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
562 		dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
563 	}
564 
565 	dev->caps.rl_caps = dev_cap->rl_caps;
566 
567 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
568 		dev->caps.dmfs_high_rate_qpn_range;
569 
570 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
571 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
572 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
573 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
574 
575 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
576 
577 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
578 		if (dev_cap->flags &
579 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
580 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
581 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
582 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
583 		}
584 
585 		if (dev_cap->flags2 &
586 		    (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
587 		     MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
588 			mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
589 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
590 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
591 		}
592 	}
593 
594 	if ((dev->caps.flags &
595 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
596 	    mlx4_is_master(dev))
597 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
598 
599 	if (!mlx4_is_slave(dev)) {
600 		mlx4_enable_cqe_eqe_stride(dev);
601 		dev->caps.alloc_res_qp_mask =
602 			(dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
603 			MLX4_RESERVE_A0_QP;
604 
605 		if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
606 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
607 			mlx4_warn(dev, "Old device ETS support detected\n");
608 			mlx4_warn(dev, "Consider upgrading device FW.\n");
609 			dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
610 		}
611 
612 	} else {
613 		dev->caps.alloc_res_qp_mask = 0;
614 	}
615 
616 	mlx4_enable_ignore_fcs(dev);
617 
618 	return 0;
619 }
620 
621 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
622 				       enum pci_bus_speed *speed,
623 				       enum pcie_link_width *width)
624 {
625 	u32 lnkcap1, lnkcap2;
626 	int err1, err2;
627 
628 #define  PCIE_MLW_CAP_SHIFT 4	/* start of MLW mask in link capabilities */
629 
630 	*speed = PCI_SPEED_UNKNOWN;
631 	*width = PCIE_LNK_WIDTH_UNKNOWN;
632 
633 	err1 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP,
634 					  &lnkcap1);
635 	err2 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP2,
636 					  &lnkcap2);
637 	if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
638 		if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
639 			*speed = PCIE_SPEED_8_0GT;
640 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
641 			*speed = PCIE_SPEED_5_0GT;
642 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
643 			*speed = PCIE_SPEED_2_5GT;
644 	}
645 	if (!err1) {
646 		*width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
647 		if (!lnkcap2) { /* pre-r3.0 */
648 			if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
649 				*speed = PCIE_SPEED_5_0GT;
650 			else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
651 				*speed = PCIE_SPEED_2_5GT;
652 		}
653 	}
654 
655 	if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
656 		return err1 ? err1 :
657 			err2 ? err2 : -EINVAL;
658 	}
659 	return 0;
660 }
661 
662 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
663 {
664 	enum pcie_link_width width, width_cap;
665 	enum pci_bus_speed speed, speed_cap;
666 	int err;
667 
668 #define PCIE_SPEED_STR(speed) \
669 	(speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
670 	 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
671 	 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
672 	 "Unknown")
673 
674 	err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
675 	if (err) {
676 		mlx4_warn(dev,
677 			  "Unable to determine PCIe device BW capabilities\n");
678 		return;
679 	}
680 
681 	err = pcie_get_minimum_link(dev->persist->pdev, &speed, &width);
682 	if (err || speed == PCI_SPEED_UNKNOWN ||
683 	    width == PCIE_LNK_WIDTH_UNKNOWN) {
684 		mlx4_warn(dev,
685 			  "Unable to determine PCI device chain minimum BW\n");
686 		return;
687 	}
688 
689 	if (width != width_cap || speed != speed_cap)
690 		mlx4_warn(dev,
691 			  "PCIe BW is different than device's capability\n");
692 
693 	mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
694 		  PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
695 	mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
696 		  width, width_cap);
697 	return;
698 }
699 
700 /*The function checks if there are live vf, return the num of them*/
701 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
702 {
703 	struct mlx4_priv *priv = mlx4_priv(dev);
704 	struct mlx4_slave_state *s_state;
705 	int i;
706 	int ret = 0;
707 
708 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
709 		s_state = &priv->mfunc.master.slave_state[i];
710 		if (s_state->active && s_state->last_cmd !=
711 		    MLX4_COMM_CMD_RESET) {
712 			mlx4_warn(dev, "%s: slave: %d is still active\n",
713 				  __func__, i);
714 			ret++;
715 		}
716 	}
717 	return ret;
718 }
719 
720 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
721 {
722 	u32 qk = MLX4_RESERVED_QKEY_BASE;
723 
724 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
725 	    qpn < dev->phys_caps.base_proxy_sqpn)
726 		return -EINVAL;
727 
728 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
729 		/* tunnel qp */
730 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
731 	else
732 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
733 	*qkey = qk;
734 	return 0;
735 }
736 EXPORT_SYMBOL(mlx4_get_parav_qkey);
737 
738 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
739 {
740 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
741 
742 	if (!mlx4_is_master(dev))
743 		return;
744 
745 	priv->virt2phys_pkey[slave][port - 1][i] = val;
746 }
747 EXPORT_SYMBOL(mlx4_sync_pkey_table);
748 
749 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
750 {
751 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
752 
753 	if (!mlx4_is_master(dev))
754 		return;
755 
756 	priv->slave_node_guids[slave] = guid;
757 }
758 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
759 
760 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
761 {
762 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
763 
764 	if (!mlx4_is_master(dev))
765 		return 0;
766 
767 	return priv->slave_node_guids[slave];
768 }
769 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
770 
771 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
772 {
773 	struct mlx4_priv *priv = mlx4_priv(dev);
774 	struct mlx4_slave_state *s_slave;
775 
776 	if (!mlx4_is_master(dev))
777 		return 0;
778 
779 	s_slave = &priv->mfunc.master.slave_state[slave];
780 	return !!s_slave->active;
781 }
782 EXPORT_SYMBOL(mlx4_is_slave_active);
783 
784 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
785 				       struct mlx4_dev_cap *dev_cap,
786 				       struct mlx4_init_hca_param *hca_param)
787 {
788 	dev->caps.steering_mode = hca_param->steering_mode;
789 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
790 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
791 		dev->caps.fs_log_max_ucast_qp_range_size =
792 			dev_cap->fs_log_max_ucast_qp_range_size;
793 	} else
794 		dev->caps.num_qp_per_mgm =
795 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
796 
797 	mlx4_dbg(dev, "Steering mode is: %s\n",
798 		 mlx4_steering_mode_str(dev->caps.steering_mode));
799 }
800 
801 static int mlx4_slave_cap(struct mlx4_dev *dev)
802 {
803 	int			   err;
804 	u32			   page_size;
805 	struct mlx4_dev_cap	   dev_cap;
806 	struct mlx4_func_cap	   func_cap;
807 	struct mlx4_init_hca_param hca_param;
808 	u8			   i;
809 
810 	memset(&hca_param, 0, sizeof(hca_param));
811 	err = mlx4_QUERY_HCA(dev, &hca_param);
812 	if (err) {
813 		mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
814 		return err;
815 	}
816 
817 	/* fail if the hca has an unknown global capability
818 	 * at this time global_caps should be always zeroed
819 	 */
820 	if (hca_param.global_caps) {
821 		mlx4_err(dev, "Unknown hca global capabilities\n");
822 		return -ENOSYS;
823 	}
824 
825 	mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
826 
827 	dev->caps.hca_core_clock = hca_param.hca_core_clock;
828 
829 	memset(&dev_cap, 0, sizeof(dev_cap));
830 	dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
831 	err = mlx4_dev_cap(dev, &dev_cap);
832 	if (err) {
833 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
834 		return err;
835 	}
836 
837 	err = mlx4_QUERY_FW(dev);
838 	if (err)
839 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
840 
841 	page_size = ~dev->caps.page_size_cap + 1;
842 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
843 	if (page_size > PAGE_SIZE) {
844 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
845 			 page_size, PAGE_SIZE);
846 		return -ENODEV;
847 	}
848 
849 	/* Set uar_page_shift for VF */
850 	dev->uar_page_shift = hca_param.uar_page_sz + 12;
851 
852 	/* Make sure the master uar page size is valid */
853 	if (dev->uar_page_shift > PAGE_SHIFT) {
854 		mlx4_err(dev,
855 			 "Invalid configuration: uar page size is larger than system page size\n");
856 		return  -ENODEV;
857 	}
858 
859 	/* Set reserved_uars based on the uar_page_shift */
860 	mlx4_set_num_reserved_uars(dev, &dev_cap);
861 
862 	/* Although uar page size in FW differs from system page size,
863 	 * upper software layers (mlx4_ib, mlx4_en and part of mlx4_core)
864 	 * still works with assumption that uar page size == system page size
865 	 */
866 	dev->caps.uar_page_size = PAGE_SIZE;
867 
868 	memset(&func_cap, 0, sizeof(func_cap));
869 	err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
870 	if (err) {
871 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
872 			 err);
873 		return err;
874 	}
875 
876 	if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
877 	    PF_CONTEXT_BEHAVIOUR_MASK) {
878 		mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
879 			 func_cap.pf_context_behaviour, PF_CONTEXT_BEHAVIOUR_MASK);
880 		return -ENOSYS;
881 	}
882 
883 	dev->caps.num_ports		= func_cap.num_ports;
884 	dev->quotas.qp			= func_cap.qp_quota;
885 	dev->quotas.srq			= func_cap.srq_quota;
886 	dev->quotas.cq			= func_cap.cq_quota;
887 	dev->quotas.mpt			= func_cap.mpt_quota;
888 	dev->quotas.mtt			= func_cap.mtt_quota;
889 	dev->caps.num_qps		= 1 << hca_param.log_num_qps;
890 	dev->caps.num_srqs		= 1 << hca_param.log_num_srqs;
891 	dev->caps.num_cqs		= 1 << hca_param.log_num_cqs;
892 	dev->caps.num_mpts		= 1 << hca_param.log_mpt_sz;
893 	dev->caps.num_eqs		= func_cap.max_eq;
894 	dev->caps.reserved_eqs		= func_cap.reserved_eq;
895 	dev->caps.reserved_lkey		= func_cap.reserved_lkey;
896 	dev->caps.num_pds               = MLX4_NUM_PDS;
897 	dev->caps.num_mgms              = 0;
898 	dev->caps.num_amgms             = 0;
899 
900 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
901 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
902 			 dev->caps.num_ports, MLX4_MAX_PORTS);
903 		return -ENODEV;
904 	}
905 
906 	mlx4_replace_zero_macs(dev);
907 
908 	dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
909 	dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
910 	dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
911 	dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
912 	dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
913 
914 	if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
915 	    !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
916 	    !dev->caps.qp0_qkey) {
917 		err = -ENOMEM;
918 		goto err_mem;
919 	}
920 
921 	for (i = 1; i <= dev->caps.num_ports; ++i) {
922 		err = mlx4_QUERY_FUNC_CAP(dev, i, &func_cap);
923 		if (err) {
924 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
925 				 i, err);
926 			goto err_mem;
927 		}
928 		dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
929 		dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
930 		dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
931 		dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
932 		dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
933 		dev->caps.port_mask[i] = dev->caps.port_type[i];
934 		dev->caps.phys_port_id[i] = func_cap.phys_port_id;
935 		err = mlx4_get_slave_pkey_gid_tbl_len(dev, i,
936 						      &dev->caps.gid_table_len[i],
937 						      &dev->caps.pkey_table_len[i]);
938 		if (err)
939 			goto err_mem;
940 	}
941 
942 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
943 				       dev->caps.reserved_uars) >
944 				       pci_resource_len(dev->persist->pdev,
945 							2)) {
946 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
947 			 dev->caps.uar_page_size * dev->caps.num_uars,
948 			 (unsigned long long)
949 			 pci_resource_len(dev->persist->pdev, 2));
950 		err = -ENOMEM;
951 		goto err_mem;
952 	}
953 
954 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
955 		dev->caps.eqe_size   = 64;
956 		dev->caps.eqe_factor = 1;
957 	} else {
958 		dev->caps.eqe_size   = 32;
959 		dev->caps.eqe_factor = 0;
960 	}
961 
962 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
963 		dev->caps.cqe_size   = 64;
964 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
965 	} else {
966 		dev->caps.cqe_size   = 32;
967 	}
968 
969 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
970 		dev->caps.eqe_size = hca_param.eqe_size;
971 		dev->caps.eqe_factor = 0;
972 	}
973 
974 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
975 		dev->caps.cqe_size = hca_param.cqe_size;
976 		/* User still need to know when CQE > 32B */
977 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
978 	}
979 
980 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
981 	mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
982 
983 	slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
984 	mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
985 		 hca_param.rss_ip_frags ? "on" : "off");
986 
987 	if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
988 	    dev->caps.bf_reg_size)
989 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
990 
991 	if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
992 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
993 
994 	return 0;
995 
996 err_mem:
997 	kfree(dev->caps.qp0_qkey);
998 	kfree(dev->caps.qp0_tunnel);
999 	kfree(dev->caps.qp0_proxy);
1000 	kfree(dev->caps.qp1_tunnel);
1001 	kfree(dev->caps.qp1_proxy);
1002 	dev->caps.qp0_qkey = NULL;
1003 	dev->caps.qp0_tunnel = NULL;
1004 	dev->caps.qp0_proxy = NULL;
1005 	dev->caps.qp1_tunnel = NULL;
1006 	dev->caps.qp1_proxy = NULL;
1007 
1008 	return err;
1009 }
1010 
1011 static void mlx4_request_modules(struct mlx4_dev *dev)
1012 {
1013 	int port;
1014 	int has_ib_port = false;
1015 	int has_eth_port = false;
1016 #define EN_DRV_NAME	"mlx4_en"
1017 #define IB_DRV_NAME	"mlx4_ib"
1018 
1019 	for (port = 1; port <= dev->caps.num_ports; port++) {
1020 		if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
1021 			has_ib_port = true;
1022 		else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
1023 			has_eth_port = true;
1024 	}
1025 
1026 	if (has_eth_port)
1027 		request_module_nowait(EN_DRV_NAME);
1028 	if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
1029 		request_module_nowait(IB_DRV_NAME);
1030 }
1031 
1032 /*
1033  * Change the port configuration of the device.
1034  * Every user of this function must hold the port mutex.
1035  */
1036 int mlx4_change_port_types(struct mlx4_dev *dev,
1037 			   enum mlx4_port_type *port_types)
1038 {
1039 	int err = 0;
1040 	int change = 0;
1041 	int port;
1042 
1043 	for (port = 0; port <  dev->caps.num_ports; port++) {
1044 		/* Change the port type only if the new type is different
1045 		 * from the current, and not set to Auto */
1046 		if (port_types[port] != dev->caps.port_type[port + 1])
1047 			change = 1;
1048 	}
1049 	if (change) {
1050 		mlx4_unregister_device(dev);
1051 		for (port = 1; port <= dev->caps.num_ports; port++) {
1052 			mlx4_CLOSE_PORT(dev, port);
1053 			dev->caps.port_type[port] = port_types[port - 1];
1054 			err = mlx4_SET_PORT(dev, port, -1);
1055 			if (err) {
1056 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1057 					 port);
1058 				goto out;
1059 			}
1060 		}
1061 		mlx4_set_port_mask(dev);
1062 		err = mlx4_register_device(dev);
1063 		if (err) {
1064 			mlx4_err(dev, "Failed to register device\n");
1065 			goto out;
1066 		}
1067 		mlx4_request_modules(dev);
1068 	}
1069 
1070 out:
1071 	return err;
1072 }
1073 
1074 static ssize_t show_port_type(struct device *dev,
1075 			      struct device_attribute *attr,
1076 			      char *buf)
1077 {
1078 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1079 						   port_attr);
1080 	struct mlx4_dev *mdev = info->dev;
1081 	char type[8];
1082 
1083 	sprintf(type, "%s",
1084 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1085 		"ib" : "eth");
1086 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1087 		sprintf(buf, "auto (%s)\n", type);
1088 	else
1089 		sprintf(buf, "%s\n", type);
1090 
1091 	return strlen(buf);
1092 }
1093 
1094 static int __set_port_type(struct mlx4_port_info *info,
1095 			   enum mlx4_port_type port_type)
1096 {
1097 	struct mlx4_dev *mdev = info->dev;
1098 	struct mlx4_priv *priv = mlx4_priv(mdev);
1099 	enum mlx4_port_type types[MLX4_MAX_PORTS];
1100 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1101 	int i;
1102 	int err = 0;
1103 
1104 	mlx4_stop_sense(mdev);
1105 	mutex_lock(&priv->port_mutex);
1106 	info->tmp_type = port_type;
1107 
1108 	/* Possible type is always the one that was delivered */
1109 	mdev->caps.possible_type[info->port] = info->tmp_type;
1110 
1111 	for (i = 0; i < mdev->caps.num_ports; i++) {
1112 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1113 					mdev->caps.possible_type[i+1];
1114 		if (types[i] == MLX4_PORT_TYPE_AUTO)
1115 			types[i] = mdev->caps.port_type[i+1];
1116 	}
1117 
1118 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1119 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1120 		for (i = 1; i <= mdev->caps.num_ports; i++) {
1121 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1122 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1123 				err = -EINVAL;
1124 			}
1125 		}
1126 	}
1127 	if (err) {
1128 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1129 		goto out;
1130 	}
1131 
1132 	mlx4_do_sense_ports(mdev, new_types, types);
1133 
1134 	err = mlx4_check_port_params(mdev, new_types);
1135 	if (err)
1136 		goto out;
1137 
1138 	/* We are about to apply the changes after the configuration
1139 	 * was verified, no need to remember the temporary types
1140 	 * any more */
1141 	for (i = 0; i < mdev->caps.num_ports; i++)
1142 		priv->port[i + 1].tmp_type = 0;
1143 
1144 	err = mlx4_change_port_types(mdev, new_types);
1145 
1146 out:
1147 	mlx4_start_sense(mdev);
1148 	mutex_unlock(&priv->port_mutex);
1149 
1150 	return err;
1151 }
1152 
1153 static ssize_t set_port_type(struct device *dev,
1154 			     struct device_attribute *attr,
1155 			     const char *buf, size_t count)
1156 {
1157 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1158 						   port_attr);
1159 	struct mlx4_dev *mdev = info->dev;
1160 	enum mlx4_port_type port_type;
1161 	static DEFINE_MUTEX(set_port_type_mutex);
1162 	int err;
1163 
1164 	mutex_lock(&set_port_type_mutex);
1165 
1166 	if (!strcmp(buf, "ib\n")) {
1167 		port_type = MLX4_PORT_TYPE_IB;
1168 	} else if (!strcmp(buf, "eth\n")) {
1169 		port_type = MLX4_PORT_TYPE_ETH;
1170 	} else if (!strcmp(buf, "auto\n")) {
1171 		port_type = MLX4_PORT_TYPE_AUTO;
1172 	} else {
1173 		mlx4_err(mdev, "%s is not supported port type\n", buf);
1174 		err = -EINVAL;
1175 		goto err_out;
1176 	}
1177 
1178 	err = __set_port_type(info, port_type);
1179 
1180 err_out:
1181 	mutex_unlock(&set_port_type_mutex);
1182 
1183 	return err ? err : count;
1184 }
1185 
1186 enum ibta_mtu {
1187 	IB_MTU_256  = 1,
1188 	IB_MTU_512  = 2,
1189 	IB_MTU_1024 = 3,
1190 	IB_MTU_2048 = 4,
1191 	IB_MTU_4096 = 5
1192 };
1193 
1194 static inline int int_to_ibta_mtu(int mtu)
1195 {
1196 	switch (mtu) {
1197 	case 256:  return IB_MTU_256;
1198 	case 512:  return IB_MTU_512;
1199 	case 1024: return IB_MTU_1024;
1200 	case 2048: return IB_MTU_2048;
1201 	case 4096: return IB_MTU_4096;
1202 	default: return -1;
1203 	}
1204 }
1205 
1206 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1207 {
1208 	switch (mtu) {
1209 	case IB_MTU_256:  return  256;
1210 	case IB_MTU_512:  return  512;
1211 	case IB_MTU_1024: return 1024;
1212 	case IB_MTU_2048: return 2048;
1213 	case IB_MTU_4096: return 4096;
1214 	default: return -1;
1215 	}
1216 }
1217 
1218 static ssize_t show_port_ib_mtu(struct device *dev,
1219 			     struct device_attribute *attr,
1220 			     char *buf)
1221 {
1222 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1223 						   port_mtu_attr);
1224 	struct mlx4_dev *mdev = info->dev;
1225 
1226 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1227 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1228 
1229 	sprintf(buf, "%d\n",
1230 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1231 	return strlen(buf);
1232 }
1233 
1234 static ssize_t set_port_ib_mtu(struct device *dev,
1235 			     struct device_attribute *attr,
1236 			     const char *buf, size_t count)
1237 {
1238 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1239 						   port_mtu_attr);
1240 	struct mlx4_dev *mdev = info->dev;
1241 	struct mlx4_priv *priv = mlx4_priv(mdev);
1242 	int err, port, mtu, ibta_mtu = -1;
1243 
1244 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1245 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1246 		return -EINVAL;
1247 	}
1248 
1249 	err = kstrtoint(buf, 0, &mtu);
1250 	if (!err)
1251 		ibta_mtu = int_to_ibta_mtu(mtu);
1252 
1253 	if (err || ibta_mtu < 0) {
1254 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1255 		return -EINVAL;
1256 	}
1257 
1258 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1259 
1260 	mlx4_stop_sense(mdev);
1261 	mutex_lock(&priv->port_mutex);
1262 	mlx4_unregister_device(mdev);
1263 	for (port = 1; port <= mdev->caps.num_ports; port++) {
1264 		mlx4_CLOSE_PORT(mdev, port);
1265 		err = mlx4_SET_PORT(mdev, port, -1);
1266 		if (err) {
1267 			mlx4_err(mdev, "Failed to set port %d, aborting\n",
1268 				 port);
1269 			goto err_set_port;
1270 		}
1271 	}
1272 	err = mlx4_register_device(mdev);
1273 err_set_port:
1274 	mutex_unlock(&priv->port_mutex);
1275 	mlx4_start_sense(mdev);
1276 	return err ? err : count;
1277 }
1278 
1279 /* bond for multi-function device */
1280 #define MAX_MF_BOND_ALLOWED_SLAVES 63
1281 static int mlx4_mf_bond(struct mlx4_dev *dev)
1282 {
1283 	int err = 0;
1284 	int nvfs;
1285 	struct mlx4_slaves_pport slaves_port1;
1286 	struct mlx4_slaves_pport slaves_port2;
1287 	DECLARE_BITMAP(slaves_port_1_2, MLX4_MFUNC_MAX);
1288 
1289 	slaves_port1 = mlx4_phys_to_slaves_pport(dev, 1);
1290 	slaves_port2 = mlx4_phys_to_slaves_pport(dev, 2);
1291 	bitmap_and(slaves_port_1_2,
1292 		   slaves_port1.slaves, slaves_port2.slaves,
1293 		   dev->persist->num_vfs + 1);
1294 
1295 	/* only single port vfs are allowed */
1296 	if (bitmap_weight(slaves_port_1_2, dev->persist->num_vfs + 1) > 1) {
1297 		mlx4_warn(dev, "HA mode unsupported for dual ported VFs\n");
1298 		return -EINVAL;
1299 	}
1300 
1301 	/* number of virtual functions is number of total functions minus one
1302 	 * physical function for each port.
1303 	 */
1304 	nvfs = bitmap_weight(slaves_port1.slaves, dev->persist->num_vfs + 1) +
1305 		bitmap_weight(slaves_port2.slaves, dev->persist->num_vfs + 1) - 2;
1306 
1307 	/* limit on maximum allowed VFs */
1308 	if (nvfs > MAX_MF_BOND_ALLOWED_SLAVES) {
1309 		mlx4_warn(dev, "HA mode is not supported for %d VFs (max %d are allowed)\n",
1310 			  nvfs, MAX_MF_BOND_ALLOWED_SLAVES);
1311 		return -EINVAL;
1312 	}
1313 
1314 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) {
1315 		mlx4_warn(dev, "HA mode unsupported for NON DMFS steering\n");
1316 		return -EINVAL;
1317 	}
1318 
1319 	err = mlx4_bond_mac_table(dev);
1320 	if (err)
1321 		return err;
1322 	err = mlx4_bond_vlan_table(dev);
1323 	if (err)
1324 		goto err1;
1325 	err = mlx4_bond_fs_rules(dev);
1326 	if (err)
1327 		goto err2;
1328 
1329 	return 0;
1330 err2:
1331 	(void)mlx4_unbond_vlan_table(dev);
1332 err1:
1333 	(void)mlx4_unbond_mac_table(dev);
1334 	return err;
1335 }
1336 
1337 static int mlx4_mf_unbond(struct mlx4_dev *dev)
1338 {
1339 	int ret, ret1;
1340 
1341 	ret = mlx4_unbond_fs_rules(dev);
1342 	if (ret)
1343 		mlx4_warn(dev, "multifunction unbond for flow rules failedi (%d)\n", ret);
1344 	ret1 = mlx4_unbond_mac_table(dev);
1345 	if (ret1) {
1346 		mlx4_warn(dev, "multifunction unbond for MAC table failed (%d)\n", ret1);
1347 		ret = ret1;
1348 	}
1349 	ret1 = mlx4_unbond_vlan_table(dev);
1350 	if (ret1) {
1351 		mlx4_warn(dev, "multifunction unbond for VLAN table failed (%d)\n", ret1);
1352 		ret = ret1;
1353 	}
1354 	return ret;
1355 }
1356 
1357 int mlx4_bond(struct mlx4_dev *dev)
1358 {
1359 	int ret = 0;
1360 	struct mlx4_priv *priv = mlx4_priv(dev);
1361 
1362 	mutex_lock(&priv->bond_mutex);
1363 
1364 	if (!mlx4_is_bonded(dev)) {
1365 		ret = mlx4_do_bond(dev, true);
1366 		if (ret)
1367 			mlx4_err(dev, "Failed to bond device: %d\n", ret);
1368 		if (!ret && mlx4_is_master(dev)) {
1369 			ret = mlx4_mf_bond(dev);
1370 			if (ret) {
1371 				mlx4_err(dev, "bond for multifunction failed\n");
1372 				mlx4_do_bond(dev, false);
1373 			}
1374 		}
1375 	}
1376 
1377 	mutex_unlock(&priv->bond_mutex);
1378 	if (!ret)
1379 		mlx4_dbg(dev, "Device is bonded\n");
1380 
1381 	return ret;
1382 }
1383 EXPORT_SYMBOL_GPL(mlx4_bond);
1384 
1385 int mlx4_unbond(struct mlx4_dev *dev)
1386 {
1387 	int ret = 0;
1388 	struct mlx4_priv *priv = mlx4_priv(dev);
1389 
1390 	mutex_lock(&priv->bond_mutex);
1391 
1392 	if (mlx4_is_bonded(dev)) {
1393 		int ret2 = 0;
1394 
1395 		ret = mlx4_do_bond(dev, false);
1396 		if (ret)
1397 			mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1398 		if (mlx4_is_master(dev))
1399 			ret2 = mlx4_mf_unbond(dev);
1400 		if (ret2) {
1401 			mlx4_warn(dev, "Failed to unbond device for multifunction (%d)\n", ret2);
1402 			ret = ret2;
1403 		}
1404 	}
1405 
1406 	mutex_unlock(&priv->bond_mutex);
1407 	if (!ret)
1408 		mlx4_dbg(dev, "Device is unbonded\n");
1409 
1410 	return ret;
1411 }
1412 EXPORT_SYMBOL_GPL(mlx4_unbond);
1413 
1414 
1415 int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1416 {
1417 	u8 port1 = v2p->port1;
1418 	u8 port2 = v2p->port2;
1419 	struct mlx4_priv *priv = mlx4_priv(dev);
1420 	int err;
1421 
1422 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1423 		return -ENOTSUPP;
1424 
1425 	mutex_lock(&priv->bond_mutex);
1426 
1427 	/* zero means keep current mapping for this port */
1428 	if (port1 == 0)
1429 		port1 = priv->v2p.port1;
1430 	if (port2 == 0)
1431 		port2 = priv->v2p.port2;
1432 
1433 	if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1434 	    (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1435 	    (port1 == 2 && port2 == 1)) {
1436 		/* besides boundary checks cross mapping makes
1437 		 * no sense and therefore not allowed */
1438 		err = -EINVAL;
1439 	} else if ((port1 == priv->v2p.port1) &&
1440 		 (port2 == priv->v2p.port2)) {
1441 		err = 0;
1442 	} else {
1443 		err = mlx4_virt2phy_port_map(dev, port1, port2);
1444 		if (!err) {
1445 			mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1446 				 port1, port2);
1447 			priv->v2p.port1 = port1;
1448 			priv->v2p.port2 = port2;
1449 		} else {
1450 			mlx4_err(dev, "Failed to change port mape: %d\n", err);
1451 		}
1452 	}
1453 
1454 	mutex_unlock(&priv->bond_mutex);
1455 	return err;
1456 }
1457 EXPORT_SYMBOL_GPL(mlx4_port_map_set);
1458 
1459 static int mlx4_load_fw(struct mlx4_dev *dev)
1460 {
1461 	struct mlx4_priv *priv = mlx4_priv(dev);
1462 	int err;
1463 
1464 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1465 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
1466 	if (!priv->fw.fw_icm) {
1467 		mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1468 		return -ENOMEM;
1469 	}
1470 
1471 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1472 	if (err) {
1473 		mlx4_err(dev, "MAP_FA command failed, aborting\n");
1474 		goto err_free;
1475 	}
1476 
1477 	err = mlx4_RUN_FW(dev);
1478 	if (err) {
1479 		mlx4_err(dev, "RUN_FW command failed, aborting\n");
1480 		goto err_unmap_fa;
1481 	}
1482 
1483 	return 0;
1484 
1485 err_unmap_fa:
1486 	mlx4_UNMAP_FA(dev);
1487 
1488 err_free:
1489 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1490 	return err;
1491 }
1492 
1493 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1494 				int cmpt_entry_sz)
1495 {
1496 	struct mlx4_priv *priv = mlx4_priv(dev);
1497 	int err;
1498 	int num_eqs;
1499 
1500 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1501 				  cmpt_base +
1502 				  ((u64) (MLX4_CMPT_TYPE_QP *
1503 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1504 				  cmpt_entry_sz, dev->caps.num_qps,
1505 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1506 				  0, 0);
1507 	if (err)
1508 		goto err;
1509 
1510 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1511 				  cmpt_base +
1512 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1513 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1514 				  cmpt_entry_sz, dev->caps.num_srqs,
1515 				  dev->caps.reserved_srqs, 0, 0);
1516 	if (err)
1517 		goto err_qp;
1518 
1519 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1520 				  cmpt_base +
1521 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1522 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1523 				  cmpt_entry_sz, dev->caps.num_cqs,
1524 				  dev->caps.reserved_cqs, 0, 0);
1525 	if (err)
1526 		goto err_srq;
1527 
1528 	num_eqs = dev->phys_caps.num_phys_eqs;
1529 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1530 				  cmpt_base +
1531 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1532 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1533 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1534 	if (err)
1535 		goto err_cq;
1536 
1537 	return 0;
1538 
1539 err_cq:
1540 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1541 
1542 err_srq:
1543 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1544 
1545 err_qp:
1546 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1547 
1548 err:
1549 	return err;
1550 }
1551 
1552 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1553 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1554 {
1555 	struct mlx4_priv *priv = mlx4_priv(dev);
1556 	u64 aux_pages;
1557 	int num_eqs;
1558 	int err;
1559 
1560 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1561 	if (err) {
1562 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1563 		return err;
1564 	}
1565 
1566 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1567 		 (unsigned long long) icm_size >> 10,
1568 		 (unsigned long long) aux_pages << 2);
1569 
1570 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1571 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1572 	if (!priv->fw.aux_icm) {
1573 		mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1574 		return -ENOMEM;
1575 	}
1576 
1577 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1578 	if (err) {
1579 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1580 		goto err_free_aux;
1581 	}
1582 
1583 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1584 	if (err) {
1585 		mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1586 		goto err_unmap_aux;
1587 	}
1588 
1589 
1590 	num_eqs = dev->phys_caps.num_phys_eqs;
1591 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1592 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1593 				  num_eqs, num_eqs, 0, 0);
1594 	if (err) {
1595 		mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1596 		goto err_unmap_cmpt;
1597 	}
1598 
1599 	/*
1600 	 * Reserved MTT entries must be aligned up to a cacheline
1601 	 * boundary, since the FW will write to them, while the driver
1602 	 * writes to all other MTT entries. (The variable
1603 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1604 	 * size, not the raw entry size)
1605 	 */
1606 	dev->caps.reserved_mtts =
1607 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1608 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1609 
1610 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1611 				  init_hca->mtt_base,
1612 				  dev->caps.mtt_entry_sz,
1613 				  dev->caps.num_mtts,
1614 				  dev->caps.reserved_mtts, 1, 0);
1615 	if (err) {
1616 		mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1617 		goto err_unmap_eq;
1618 	}
1619 
1620 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1621 				  init_hca->dmpt_base,
1622 				  dev_cap->dmpt_entry_sz,
1623 				  dev->caps.num_mpts,
1624 				  dev->caps.reserved_mrws, 1, 1);
1625 	if (err) {
1626 		mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1627 		goto err_unmap_mtt;
1628 	}
1629 
1630 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1631 				  init_hca->qpc_base,
1632 				  dev_cap->qpc_entry_sz,
1633 				  dev->caps.num_qps,
1634 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1635 				  0, 0);
1636 	if (err) {
1637 		mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1638 		goto err_unmap_dmpt;
1639 	}
1640 
1641 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1642 				  init_hca->auxc_base,
1643 				  dev_cap->aux_entry_sz,
1644 				  dev->caps.num_qps,
1645 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1646 				  0, 0);
1647 	if (err) {
1648 		mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1649 		goto err_unmap_qp;
1650 	}
1651 
1652 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1653 				  init_hca->altc_base,
1654 				  dev_cap->altc_entry_sz,
1655 				  dev->caps.num_qps,
1656 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1657 				  0, 0);
1658 	if (err) {
1659 		mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1660 		goto err_unmap_auxc;
1661 	}
1662 
1663 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1664 				  init_hca->rdmarc_base,
1665 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1666 				  dev->caps.num_qps,
1667 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1668 				  0, 0);
1669 	if (err) {
1670 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1671 		goto err_unmap_altc;
1672 	}
1673 
1674 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1675 				  init_hca->cqc_base,
1676 				  dev_cap->cqc_entry_sz,
1677 				  dev->caps.num_cqs,
1678 				  dev->caps.reserved_cqs, 0, 0);
1679 	if (err) {
1680 		mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1681 		goto err_unmap_rdmarc;
1682 	}
1683 
1684 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1685 				  init_hca->srqc_base,
1686 				  dev_cap->srq_entry_sz,
1687 				  dev->caps.num_srqs,
1688 				  dev->caps.reserved_srqs, 0, 0);
1689 	if (err) {
1690 		mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1691 		goto err_unmap_cq;
1692 	}
1693 
1694 	/*
1695 	 * For flow steering device managed mode it is required to use
1696 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1697 	 * required, but for simplicity just map the whole multicast
1698 	 * group table now.  The table isn't very big and it's a lot
1699 	 * easier than trying to track ref counts.
1700 	 */
1701 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1702 				  init_hca->mc_base,
1703 				  mlx4_get_mgm_entry_size(dev),
1704 				  dev->caps.num_mgms + dev->caps.num_amgms,
1705 				  dev->caps.num_mgms + dev->caps.num_amgms,
1706 				  0, 0);
1707 	if (err) {
1708 		mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1709 		goto err_unmap_srq;
1710 	}
1711 
1712 	return 0;
1713 
1714 err_unmap_srq:
1715 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1716 
1717 err_unmap_cq:
1718 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1719 
1720 err_unmap_rdmarc:
1721 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1722 
1723 err_unmap_altc:
1724 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1725 
1726 err_unmap_auxc:
1727 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1728 
1729 err_unmap_qp:
1730 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1731 
1732 err_unmap_dmpt:
1733 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1734 
1735 err_unmap_mtt:
1736 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1737 
1738 err_unmap_eq:
1739 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1740 
1741 err_unmap_cmpt:
1742 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1743 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1744 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1745 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1746 
1747 err_unmap_aux:
1748 	mlx4_UNMAP_ICM_AUX(dev);
1749 
1750 err_free_aux:
1751 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1752 
1753 	return err;
1754 }
1755 
1756 static void mlx4_free_icms(struct mlx4_dev *dev)
1757 {
1758 	struct mlx4_priv *priv = mlx4_priv(dev);
1759 
1760 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1761 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1762 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1763 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1764 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1765 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1766 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1767 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1768 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1769 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1770 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1771 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1772 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1773 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1774 
1775 	mlx4_UNMAP_ICM_AUX(dev);
1776 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1777 }
1778 
1779 static void mlx4_slave_exit(struct mlx4_dev *dev)
1780 {
1781 	struct mlx4_priv *priv = mlx4_priv(dev);
1782 
1783 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1784 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1785 			  MLX4_COMM_TIME))
1786 		mlx4_warn(dev, "Failed to close slave function\n");
1787 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1788 }
1789 
1790 static int map_bf_area(struct mlx4_dev *dev)
1791 {
1792 	struct mlx4_priv *priv = mlx4_priv(dev);
1793 	resource_size_t bf_start;
1794 	resource_size_t bf_len;
1795 	int err = 0;
1796 
1797 	if (!dev->caps.bf_reg_size)
1798 		return -ENXIO;
1799 
1800 	bf_start = pci_resource_start(dev->persist->pdev, 2) +
1801 			(dev->caps.num_uars << PAGE_SHIFT);
1802 	bf_len = pci_resource_len(dev->persist->pdev, 2) -
1803 			(dev->caps.num_uars << PAGE_SHIFT);
1804 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1805 	if (!priv->bf_mapping)
1806 		err = -ENOMEM;
1807 
1808 	return err;
1809 }
1810 
1811 static void unmap_bf_area(struct mlx4_dev *dev)
1812 {
1813 	if (mlx4_priv(dev)->bf_mapping)
1814 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1815 }
1816 
1817 cycle_t mlx4_read_clock(struct mlx4_dev *dev)
1818 {
1819 	u32 clockhi, clocklo, clockhi1;
1820 	cycle_t cycles;
1821 	int i;
1822 	struct mlx4_priv *priv = mlx4_priv(dev);
1823 
1824 	for (i = 0; i < 10; i++) {
1825 		clockhi = swab32(readl(priv->clock_mapping));
1826 		clocklo = swab32(readl(priv->clock_mapping + 4));
1827 		clockhi1 = swab32(readl(priv->clock_mapping));
1828 		if (clockhi == clockhi1)
1829 			break;
1830 	}
1831 
1832 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1833 
1834 	return cycles;
1835 }
1836 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1837 
1838 
1839 static int map_internal_clock(struct mlx4_dev *dev)
1840 {
1841 	struct mlx4_priv *priv = mlx4_priv(dev);
1842 
1843 	priv->clock_mapping =
1844 		ioremap(pci_resource_start(dev->persist->pdev,
1845 					   priv->fw.clock_bar) +
1846 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1847 
1848 	if (!priv->clock_mapping)
1849 		return -ENOMEM;
1850 
1851 	return 0;
1852 }
1853 
1854 int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1855 				   struct mlx4_clock_params *params)
1856 {
1857 	struct mlx4_priv *priv = mlx4_priv(dev);
1858 
1859 	if (mlx4_is_slave(dev))
1860 		return -ENOTSUPP;
1861 
1862 	if (!params)
1863 		return -EINVAL;
1864 
1865 	params->bar = priv->fw.clock_bar;
1866 	params->offset = priv->fw.clock_offset;
1867 	params->size = MLX4_CLOCK_SIZE;
1868 
1869 	return 0;
1870 }
1871 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1872 
1873 static void unmap_internal_clock(struct mlx4_dev *dev)
1874 {
1875 	struct mlx4_priv *priv = mlx4_priv(dev);
1876 
1877 	if (priv->clock_mapping)
1878 		iounmap(priv->clock_mapping);
1879 }
1880 
1881 static void mlx4_close_hca(struct mlx4_dev *dev)
1882 {
1883 	unmap_internal_clock(dev);
1884 	unmap_bf_area(dev);
1885 	if (mlx4_is_slave(dev))
1886 		mlx4_slave_exit(dev);
1887 	else {
1888 		mlx4_CLOSE_HCA(dev, 0);
1889 		mlx4_free_icms(dev);
1890 	}
1891 }
1892 
1893 static void mlx4_close_fw(struct mlx4_dev *dev)
1894 {
1895 	if (!mlx4_is_slave(dev)) {
1896 		mlx4_UNMAP_FA(dev);
1897 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1898 	}
1899 }
1900 
1901 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1902 {
1903 #define COMM_CHAN_OFFLINE_OFFSET 0x09
1904 
1905 	u32 comm_flags;
1906 	u32 offline_bit;
1907 	unsigned long end;
1908 	struct mlx4_priv *priv = mlx4_priv(dev);
1909 
1910 	end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
1911 	while (time_before(jiffies, end)) {
1912 		comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
1913 					  MLX4_COMM_CHAN_FLAGS));
1914 		offline_bit = (comm_flags &
1915 			       (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
1916 		if (!offline_bit)
1917 			return 0;
1918 		/* There are cases as part of AER/Reset flow that PF needs
1919 		 * around 100 msec to load. We therefore sleep for 100 msec
1920 		 * to allow other tasks to make use of that CPU during this
1921 		 * time interval.
1922 		 */
1923 		msleep(100);
1924 	}
1925 	mlx4_err(dev, "Communication channel is offline.\n");
1926 	return -EIO;
1927 }
1928 
1929 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
1930 {
1931 #define COMM_CHAN_RST_OFFSET 0x1e
1932 
1933 	struct mlx4_priv *priv = mlx4_priv(dev);
1934 	u32 comm_rst;
1935 	u32 comm_caps;
1936 
1937 	comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
1938 				 MLX4_COMM_CHAN_CAPS));
1939 	comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
1940 
1941 	if (comm_rst)
1942 		dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
1943 }
1944 
1945 static int mlx4_init_slave(struct mlx4_dev *dev)
1946 {
1947 	struct mlx4_priv *priv = mlx4_priv(dev);
1948 	u64 dma = (u64) priv->mfunc.vhcr_dma;
1949 	int ret_from_reset = 0;
1950 	u32 slave_read;
1951 	u32 cmd_channel_ver;
1952 
1953 	if (atomic_read(&pf_loading)) {
1954 		mlx4_warn(dev, "PF is not ready - Deferring probe\n");
1955 		return -EPROBE_DEFER;
1956 	}
1957 
1958 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1959 	priv->cmd.max_cmds = 1;
1960 	if (mlx4_comm_check_offline(dev)) {
1961 		mlx4_err(dev, "PF is not responsive, skipping initialization\n");
1962 		goto err_offline;
1963 	}
1964 
1965 	mlx4_reset_vf_support(dev);
1966 	mlx4_warn(dev, "Sending reset\n");
1967 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1968 				       MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
1969 	/* if we are in the middle of flr the slave will try
1970 	 * NUM_OF_RESET_RETRIES times before leaving.*/
1971 	if (ret_from_reset) {
1972 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1973 			mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
1974 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
1975 			return -EPROBE_DEFER;
1976 		} else
1977 			goto err;
1978 	}
1979 
1980 	/* check the driver version - the slave I/F revision
1981 	 * must match the master's */
1982 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1983 	cmd_channel_ver = mlx4_comm_get_version();
1984 
1985 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1986 		MLX4_COMM_GET_IF_REV(slave_read)) {
1987 		mlx4_err(dev, "slave driver version is not supported by the master\n");
1988 		goto err;
1989 	}
1990 
1991 	mlx4_warn(dev, "Sending vhcr0\n");
1992 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1993 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1994 		goto err;
1995 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1996 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1997 		goto err;
1998 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1999 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2000 		goto err;
2001 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
2002 			  MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2003 		goto err;
2004 
2005 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2006 	return 0;
2007 
2008 err:
2009 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
2010 err_offline:
2011 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2012 	return -EIO;
2013 }
2014 
2015 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
2016 {
2017 	int i;
2018 
2019 	for (i = 1; i <= dev->caps.num_ports; i++) {
2020 		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
2021 			dev->caps.gid_table_len[i] =
2022 				mlx4_get_slave_num_gids(dev, 0, i);
2023 		else
2024 			dev->caps.gid_table_len[i] = 1;
2025 		dev->caps.pkey_table_len[i] =
2026 			dev->phys_caps.pkey_phys_table_len[i] - 1;
2027 	}
2028 }
2029 
2030 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
2031 {
2032 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
2033 
2034 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
2035 	      i++) {
2036 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
2037 			break;
2038 	}
2039 
2040 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
2041 }
2042 
2043 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
2044 {
2045 	switch (dmfs_high_steer_mode) {
2046 	case MLX4_STEERING_DMFS_A0_DEFAULT:
2047 		return "default performance";
2048 
2049 	case MLX4_STEERING_DMFS_A0_DYNAMIC:
2050 		return "dynamic hybrid mode";
2051 
2052 	case MLX4_STEERING_DMFS_A0_STATIC:
2053 		return "performance optimized for limited rule configuration (static)";
2054 
2055 	case MLX4_STEERING_DMFS_A0_DISABLE:
2056 		return "disabled performance optimized steering";
2057 
2058 	case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
2059 		return "performance optimized steering not supported";
2060 
2061 	default:
2062 		return "Unrecognized mode";
2063 	}
2064 }
2065 
2066 #define MLX4_DMFS_A0_STEERING			(1UL << 2)
2067 
2068 static void choose_steering_mode(struct mlx4_dev *dev,
2069 				 struct mlx4_dev_cap *dev_cap)
2070 {
2071 	if (mlx4_log_num_mgm_entry_size <= 0) {
2072 		if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
2073 			if (dev->caps.dmfs_high_steer_mode ==
2074 			    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2075 				mlx4_err(dev, "DMFS high rate mode not supported\n");
2076 			else
2077 				dev->caps.dmfs_high_steer_mode =
2078 					MLX4_STEERING_DMFS_A0_STATIC;
2079 		}
2080 	}
2081 
2082 	if (mlx4_log_num_mgm_entry_size <= 0 &&
2083 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
2084 	    (!mlx4_is_mfunc(dev) ||
2085 	     (dev_cap->fs_max_num_qp_per_entry >=
2086 	     (dev->persist->num_vfs + 1))) &&
2087 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
2088 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
2089 		dev->oper_log_mgm_entry_size =
2090 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
2091 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
2092 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
2093 		dev->caps.fs_log_max_ucast_qp_range_size =
2094 			dev_cap->fs_log_max_ucast_qp_range_size;
2095 	} else {
2096 		if (dev->caps.dmfs_high_steer_mode !=
2097 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2098 			dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
2099 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
2100 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2101 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
2102 		else {
2103 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
2104 
2105 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
2106 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2107 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
2108 		}
2109 		dev->oper_log_mgm_entry_size =
2110 			mlx4_log_num_mgm_entry_size > 0 ?
2111 			mlx4_log_num_mgm_entry_size :
2112 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
2113 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
2114 	}
2115 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
2116 		 mlx4_steering_mode_str(dev->caps.steering_mode),
2117 		 dev->oper_log_mgm_entry_size,
2118 		 mlx4_log_num_mgm_entry_size);
2119 }
2120 
2121 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
2122 				       struct mlx4_dev_cap *dev_cap)
2123 {
2124 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2125 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
2126 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
2127 	else
2128 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
2129 
2130 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
2131 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
2132 }
2133 
2134 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
2135 {
2136 	int i;
2137 	struct mlx4_port_cap port_cap;
2138 
2139 	if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2140 		return -EINVAL;
2141 
2142 	for (i = 1; i <= dev->caps.num_ports; i++) {
2143 		if (mlx4_dev_port(dev, i, &port_cap)) {
2144 			mlx4_err(dev,
2145 				 "QUERY_DEV_CAP command failed, can't veify DMFS high rate steering.\n");
2146 		} else if ((dev->caps.dmfs_high_steer_mode !=
2147 			    MLX4_STEERING_DMFS_A0_DEFAULT) &&
2148 			   (port_cap.dmfs_optimized_state ==
2149 			    !!(dev->caps.dmfs_high_steer_mode ==
2150 			    MLX4_STEERING_DMFS_A0_DISABLE))) {
2151 			mlx4_err(dev,
2152 				 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2153 				 dmfs_high_rate_steering_mode_str(
2154 					dev->caps.dmfs_high_steer_mode),
2155 				 (port_cap.dmfs_optimized_state ?
2156 					"enabled" : "disabled"));
2157 		}
2158 	}
2159 
2160 	return 0;
2161 }
2162 
2163 static int mlx4_init_fw(struct mlx4_dev *dev)
2164 {
2165 	struct mlx4_mod_stat_cfg   mlx4_cfg;
2166 	int err = 0;
2167 
2168 	if (!mlx4_is_slave(dev)) {
2169 		err = mlx4_QUERY_FW(dev);
2170 		if (err) {
2171 			if (err == -EACCES)
2172 				mlx4_info(dev, "non-primary physical function, skipping\n");
2173 			else
2174 				mlx4_err(dev, "QUERY_FW command failed, aborting\n");
2175 			return err;
2176 		}
2177 
2178 		err = mlx4_load_fw(dev);
2179 		if (err) {
2180 			mlx4_err(dev, "Failed to start FW, aborting\n");
2181 			return err;
2182 		}
2183 
2184 		mlx4_cfg.log_pg_sz_m = 1;
2185 		mlx4_cfg.log_pg_sz = 0;
2186 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2187 		if (err)
2188 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2189 	}
2190 
2191 	return err;
2192 }
2193 
2194 static int mlx4_init_hca(struct mlx4_dev *dev)
2195 {
2196 	struct mlx4_priv	  *priv = mlx4_priv(dev);
2197 	struct mlx4_adapter	   adapter;
2198 	struct mlx4_dev_cap	   dev_cap;
2199 	struct mlx4_profile	   profile;
2200 	struct mlx4_init_hca_param init_hca;
2201 	u64 icm_size;
2202 	struct mlx4_config_dev_params params;
2203 	int err;
2204 
2205 	if (!mlx4_is_slave(dev)) {
2206 		err = mlx4_dev_cap(dev, &dev_cap);
2207 		if (err) {
2208 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
2209 			return err;
2210 		}
2211 
2212 		choose_steering_mode(dev, &dev_cap);
2213 		choose_tunnel_offload_mode(dev, &dev_cap);
2214 
2215 		if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2216 		    mlx4_is_master(dev))
2217 			dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2218 
2219 		err = mlx4_get_phys_port_id(dev);
2220 		if (err)
2221 			mlx4_err(dev, "Fail to get physical port id\n");
2222 
2223 		if (mlx4_is_master(dev))
2224 			mlx4_parav_master_pf_caps(dev);
2225 
2226 		if (mlx4_low_memory_profile()) {
2227 			mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2228 			profile = low_mem_profile;
2229 		} else {
2230 			profile = default_profile;
2231 		}
2232 		if (dev->caps.steering_mode ==
2233 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
2234 			profile.num_mcg = MLX4_FS_NUM_MCG;
2235 
2236 		icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
2237 					     &init_hca);
2238 		if ((long long) icm_size < 0) {
2239 			err = icm_size;
2240 			return err;
2241 		}
2242 
2243 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
2244 
2245 		if (enable_4k_uar) {
2246 			init_hca.log_uar_sz = ilog2(dev->caps.num_uars) +
2247 						    PAGE_SHIFT - DEFAULT_UAR_PAGE_SHIFT;
2248 			init_hca.uar_page_sz = DEFAULT_UAR_PAGE_SHIFT - 12;
2249 		} else {
2250 			init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
2251 			init_hca.uar_page_sz = PAGE_SHIFT - 12;
2252 		}
2253 
2254 		init_hca.mw_enabled = 0;
2255 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2256 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2257 			init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
2258 
2259 		err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
2260 		if (err)
2261 			return err;
2262 
2263 		err = mlx4_INIT_HCA(dev, &init_hca);
2264 		if (err) {
2265 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
2266 			goto err_free_icm;
2267 		}
2268 
2269 		if (dev_cap.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2270 			err = mlx4_query_func(dev, &dev_cap);
2271 			if (err < 0) {
2272 				mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
2273 				goto err_close;
2274 			} else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2275 				dev->caps.num_eqs = dev_cap.max_eqs;
2276 				dev->caps.reserved_eqs = dev_cap.reserved_eqs;
2277 				dev->caps.reserved_uars = dev_cap.reserved_uars;
2278 			}
2279 		}
2280 
2281 		/*
2282 		 * If TS is supported by FW
2283 		 * read HCA frequency by QUERY_HCA command
2284 		 */
2285 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2286 			memset(&init_hca, 0, sizeof(init_hca));
2287 			err = mlx4_QUERY_HCA(dev, &init_hca);
2288 			if (err) {
2289 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
2290 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2291 			} else {
2292 				dev->caps.hca_core_clock =
2293 					init_hca.hca_core_clock;
2294 			}
2295 
2296 			/* In case we got HCA frequency 0 - disable timestamping
2297 			 * to avoid dividing by zero
2298 			 */
2299 			if (!dev->caps.hca_core_clock) {
2300 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2301 				mlx4_err(dev,
2302 					 "HCA frequency is 0 - timestamping is not supported\n");
2303 			} else if (map_internal_clock(dev)) {
2304 				/*
2305 				 * Map internal clock,
2306 				 * in case of failure disable timestamping
2307 				 */
2308 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2309 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
2310 			}
2311 		}
2312 
2313 		if (dev->caps.dmfs_high_steer_mode !=
2314 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2315 			if (mlx4_validate_optimized_steering(dev))
2316 				mlx4_warn(dev, "Optimized steering validation failed\n");
2317 
2318 			if (dev->caps.dmfs_high_steer_mode ==
2319 			    MLX4_STEERING_DMFS_A0_DISABLE) {
2320 				dev->caps.dmfs_high_rate_qpn_base =
2321 					dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2322 				dev->caps.dmfs_high_rate_qpn_range =
2323 					MLX4_A0_STEERING_TABLE_SIZE;
2324 			}
2325 
2326 			mlx4_dbg(dev, "DMFS high rate steer mode is: %s\n",
2327 				 dmfs_high_rate_steering_mode_str(
2328 					dev->caps.dmfs_high_steer_mode));
2329 		}
2330 	} else {
2331 		err = mlx4_init_slave(dev);
2332 		if (err) {
2333 			if (err != -EPROBE_DEFER)
2334 				mlx4_err(dev, "Failed to initialize slave\n");
2335 			return err;
2336 		}
2337 
2338 		err = mlx4_slave_cap(dev);
2339 		if (err) {
2340 			mlx4_err(dev, "Failed to obtain slave caps\n");
2341 			goto err_close;
2342 		}
2343 	}
2344 
2345 	if (map_bf_area(dev))
2346 		mlx4_dbg(dev, "Failed to map blue flame area\n");
2347 
2348 	/*Only the master set the ports, all the rest got it from it.*/
2349 	if (!mlx4_is_slave(dev))
2350 		mlx4_set_port_mask(dev);
2351 
2352 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
2353 	if (err) {
2354 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2355 		goto unmap_bf;
2356 	}
2357 
2358 	/* Query CONFIG_DEV parameters */
2359 	err = mlx4_config_dev_retrieval(dev, &params);
2360 	if (err && err != -ENOTSUPP) {
2361 		mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2362 	} else if (!err) {
2363 		dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2364 		dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2365 	}
2366 	priv->eq_table.inta_pin = adapter.inta_pin;
2367 	memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
2368 
2369 	return 0;
2370 
2371 unmap_bf:
2372 	unmap_internal_clock(dev);
2373 	unmap_bf_area(dev);
2374 
2375 	if (mlx4_is_slave(dev)) {
2376 		kfree(dev->caps.qp0_qkey);
2377 		kfree(dev->caps.qp0_tunnel);
2378 		kfree(dev->caps.qp0_proxy);
2379 		kfree(dev->caps.qp1_tunnel);
2380 		kfree(dev->caps.qp1_proxy);
2381 	}
2382 
2383 err_close:
2384 	if (mlx4_is_slave(dev))
2385 		mlx4_slave_exit(dev);
2386 	else
2387 		mlx4_CLOSE_HCA(dev, 0);
2388 
2389 err_free_icm:
2390 	if (!mlx4_is_slave(dev))
2391 		mlx4_free_icms(dev);
2392 
2393 	return err;
2394 }
2395 
2396 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2397 {
2398 	struct mlx4_priv *priv = mlx4_priv(dev);
2399 	int nent_pow2;
2400 
2401 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2402 		return -ENOENT;
2403 
2404 	if (!dev->caps.max_counters)
2405 		return -ENOSPC;
2406 
2407 	nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2408 	/* reserve last counter index for sink counter */
2409 	return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2410 				nent_pow2 - 1, 0,
2411 				nent_pow2 - dev->caps.max_counters + 1);
2412 }
2413 
2414 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2415 {
2416 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2417 		return;
2418 
2419 	if (!dev->caps.max_counters)
2420 		return;
2421 
2422 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2423 }
2424 
2425 static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2426 {
2427 	struct mlx4_priv *priv = mlx4_priv(dev);
2428 	int port;
2429 
2430 	for (port = 0; port < dev->caps.num_ports; port++)
2431 		if (priv->def_counter[port] != -1)
2432 			mlx4_counter_free(dev,  priv->def_counter[port]);
2433 }
2434 
2435 static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2436 {
2437 	struct mlx4_priv *priv = mlx4_priv(dev);
2438 	int port, err = 0;
2439 	u32 idx;
2440 
2441 	for (port = 0; port < dev->caps.num_ports; port++)
2442 		priv->def_counter[port] = -1;
2443 
2444 	for (port = 0; port < dev->caps.num_ports; port++) {
2445 		err = mlx4_counter_alloc(dev, &idx);
2446 
2447 		if (!err || err == -ENOSPC) {
2448 			priv->def_counter[port] = idx;
2449 		} else if (err == -ENOENT) {
2450 			err = 0;
2451 			continue;
2452 		} else if (mlx4_is_slave(dev) && err == -EINVAL) {
2453 			priv->def_counter[port] = MLX4_SINK_COUNTER_INDEX(dev);
2454 			mlx4_warn(dev, "can't allocate counter from old PF driver, using index %d\n",
2455 				  MLX4_SINK_COUNTER_INDEX(dev));
2456 			err = 0;
2457 		} else {
2458 			mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2459 				 __func__, port + 1, err);
2460 			mlx4_cleanup_default_counters(dev);
2461 			return err;
2462 		}
2463 
2464 		mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2465 			 __func__, priv->def_counter[port], port + 1);
2466 	}
2467 
2468 	return err;
2469 }
2470 
2471 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2472 {
2473 	struct mlx4_priv *priv = mlx4_priv(dev);
2474 
2475 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2476 		return -ENOENT;
2477 
2478 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2479 	if (*idx == -1) {
2480 		*idx = MLX4_SINK_COUNTER_INDEX(dev);
2481 		return -ENOSPC;
2482 	}
2483 
2484 	return 0;
2485 }
2486 
2487 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2488 {
2489 	u64 out_param;
2490 	int err;
2491 
2492 	if (mlx4_is_mfunc(dev)) {
2493 		err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
2494 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2495 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2496 		if (!err)
2497 			*idx = get_param_l(&out_param);
2498 
2499 		return err;
2500 	}
2501 	return __mlx4_counter_alloc(dev, idx);
2502 }
2503 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2504 
2505 static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2506 				u8 counter_index)
2507 {
2508 	struct mlx4_cmd_mailbox *if_stat_mailbox;
2509 	int err;
2510 	u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2511 
2512 	if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2513 	if (IS_ERR(if_stat_mailbox))
2514 		return PTR_ERR(if_stat_mailbox);
2515 
2516 	err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2517 			   MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2518 			   MLX4_CMD_NATIVE);
2519 
2520 	mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2521 	return err;
2522 }
2523 
2524 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2525 {
2526 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2527 		return;
2528 
2529 	if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2530 		return;
2531 
2532 	__mlx4_clear_if_stat(dev, idx);
2533 
2534 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2535 	return;
2536 }
2537 
2538 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2539 {
2540 	u64 in_param = 0;
2541 
2542 	if (mlx4_is_mfunc(dev)) {
2543 		set_param_l(&in_param, idx);
2544 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2545 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2546 			 MLX4_CMD_WRAPPED);
2547 		return;
2548 	}
2549 	__mlx4_counter_free(dev, idx);
2550 }
2551 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2552 
2553 int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2554 {
2555 	struct mlx4_priv *priv = mlx4_priv(dev);
2556 
2557 	return priv->def_counter[port - 1];
2558 }
2559 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2560 
2561 void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2562 {
2563 	struct mlx4_priv *priv = mlx4_priv(dev);
2564 
2565 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2566 }
2567 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2568 
2569 __be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2570 {
2571 	struct mlx4_priv *priv = mlx4_priv(dev);
2572 
2573 	return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2574 }
2575 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2576 
2577 void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2578 {
2579 	struct mlx4_priv *priv = mlx4_priv(dev);
2580 	__be64 guid;
2581 
2582 	/* hw GUID */
2583 	if (entry == 0)
2584 		return;
2585 
2586 	get_random_bytes((char *)&guid, sizeof(guid));
2587 	guid &= ~(cpu_to_be64(1ULL << 56));
2588 	guid |= cpu_to_be64(1ULL << 57);
2589 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2590 }
2591 
2592 static int mlx4_setup_hca(struct mlx4_dev *dev)
2593 {
2594 	struct mlx4_priv *priv = mlx4_priv(dev);
2595 	int err;
2596 	int port;
2597 	__be32 ib_port_default_caps;
2598 
2599 	err = mlx4_init_uar_table(dev);
2600 	if (err) {
2601 		mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2602 		 return err;
2603 	}
2604 
2605 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
2606 	if (err) {
2607 		mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2608 		goto err_uar_table_free;
2609 	}
2610 
2611 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2612 	if (!priv->kar) {
2613 		mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2614 		err = -ENOMEM;
2615 		goto err_uar_free;
2616 	}
2617 
2618 	err = mlx4_init_pd_table(dev);
2619 	if (err) {
2620 		mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2621 		goto err_kar_unmap;
2622 	}
2623 
2624 	err = mlx4_init_xrcd_table(dev);
2625 	if (err) {
2626 		mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2627 		goto err_pd_table_free;
2628 	}
2629 
2630 	err = mlx4_init_mr_table(dev);
2631 	if (err) {
2632 		mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2633 		goto err_xrcd_table_free;
2634 	}
2635 
2636 	if (!mlx4_is_slave(dev)) {
2637 		err = mlx4_init_mcg_table(dev);
2638 		if (err) {
2639 			mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2640 			goto err_mr_table_free;
2641 		}
2642 		err = mlx4_config_mad_demux(dev);
2643 		if (err) {
2644 			mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2645 			goto err_mcg_table_free;
2646 		}
2647 	}
2648 
2649 	err = mlx4_init_eq_table(dev);
2650 	if (err) {
2651 		mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2652 		goto err_mcg_table_free;
2653 	}
2654 
2655 	err = mlx4_cmd_use_events(dev);
2656 	if (err) {
2657 		mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2658 		goto err_eq_table_free;
2659 	}
2660 
2661 	err = mlx4_NOP(dev);
2662 	if (err) {
2663 		if (dev->flags & MLX4_FLAG_MSI_X) {
2664 			mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2665 				  priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2666 			mlx4_warn(dev, "Trying again without MSI-X\n");
2667 		} else {
2668 			mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2669 				 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2670 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2671 		}
2672 
2673 		goto err_cmd_poll;
2674 	}
2675 
2676 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
2677 
2678 	err = mlx4_init_cq_table(dev);
2679 	if (err) {
2680 		mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2681 		goto err_cmd_poll;
2682 	}
2683 
2684 	err = mlx4_init_srq_table(dev);
2685 	if (err) {
2686 		mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2687 		goto err_cq_table_free;
2688 	}
2689 
2690 	err = mlx4_init_qp_table(dev);
2691 	if (err) {
2692 		mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2693 		goto err_srq_table_free;
2694 	}
2695 
2696 	if (!mlx4_is_slave(dev)) {
2697 		err = mlx4_init_counters_table(dev);
2698 		if (err && err != -ENOENT) {
2699 			mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2700 			goto err_qp_table_free;
2701 		}
2702 	}
2703 
2704 	err = mlx4_allocate_default_counters(dev);
2705 	if (err) {
2706 		mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2707 		goto err_counters_table_free;
2708 	}
2709 
2710 	if (!mlx4_is_slave(dev)) {
2711 		for (port = 1; port <= dev->caps.num_ports; port++) {
2712 			ib_port_default_caps = 0;
2713 			err = mlx4_get_port_ib_caps(dev, port,
2714 						    &ib_port_default_caps);
2715 			if (err)
2716 				mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2717 					  port, err);
2718 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2719 
2720 			/* initialize per-slave default ib port capabilities */
2721 			if (mlx4_is_master(dev)) {
2722 				int i;
2723 				for (i = 0; i < dev->num_slaves; i++) {
2724 					if (i == mlx4_master_func_num(dev))
2725 						continue;
2726 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2727 						ib_port_default_caps;
2728 				}
2729 			}
2730 
2731 			if (mlx4_is_mfunc(dev))
2732 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2733 			else
2734 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2735 
2736 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2737 					    dev->caps.pkey_table_len[port] : -1);
2738 			if (err) {
2739 				mlx4_err(dev, "Failed to set port %d, aborting\n",
2740 					 port);
2741 				goto err_default_countes_free;
2742 			}
2743 		}
2744 	}
2745 
2746 	return 0;
2747 
2748 err_default_countes_free:
2749 	mlx4_cleanup_default_counters(dev);
2750 
2751 err_counters_table_free:
2752 	if (!mlx4_is_slave(dev))
2753 		mlx4_cleanup_counters_table(dev);
2754 
2755 err_qp_table_free:
2756 	mlx4_cleanup_qp_table(dev);
2757 
2758 err_srq_table_free:
2759 	mlx4_cleanup_srq_table(dev);
2760 
2761 err_cq_table_free:
2762 	mlx4_cleanup_cq_table(dev);
2763 
2764 err_cmd_poll:
2765 	mlx4_cmd_use_polling(dev);
2766 
2767 err_eq_table_free:
2768 	mlx4_cleanup_eq_table(dev);
2769 
2770 err_mcg_table_free:
2771 	if (!mlx4_is_slave(dev))
2772 		mlx4_cleanup_mcg_table(dev);
2773 
2774 err_mr_table_free:
2775 	mlx4_cleanup_mr_table(dev);
2776 
2777 err_xrcd_table_free:
2778 	mlx4_cleanup_xrcd_table(dev);
2779 
2780 err_pd_table_free:
2781 	mlx4_cleanup_pd_table(dev);
2782 
2783 err_kar_unmap:
2784 	iounmap(priv->kar);
2785 
2786 err_uar_free:
2787 	mlx4_uar_free(dev, &priv->driver_uar);
2788 
2789 err_uar_table_free:
2790 	mlx4_cleanup_uar_table(dev);
2791 	return err;
2792 }
2793 
2794 static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2795 {
2796 	int requested_cpu = 0;
2797 	struct mlx4_priv *priv = mlx4_priv(dev);
2798 	struct mlx4_eq *eq;
2799 	int off = 0;
2800 	int i;
2801 
2802 	if (eqn > dev->caps.num_comp_vectors)
2803 		return -EINVAL;
2804 
2805 	for (i = 1; i < port; i++)
2806 		off += mlx4_get_eqs_per_port(dev, i);
2807 
2808 	requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2809 
2810 	/* Meaning EQs are shared, and this call comes from the second port */
2811 	if (requested_cpu < 0)
2812 		return 0;
2813 
2814 	eq = &priv->eq_table.eq[eqn];
2815 
2816 	if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2817 		return -ENOMEM;
2818 
2819 	cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2820 
2821 	return 0;
2822 }
2823 
2824 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2825 {
2826 	struct mlx4_priv *priv = mlx4_priv(dev);
2827 	struct msix_entry *entries;
2828 	int i;
2829 	int port = 0;
2830 
2831 	if (msi_x) {
2832 		int nreq = dev->caps.num_ports * num_online_cpus() + 1;
2833 
2834 		nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
2835 			     nreq);
2836 		if (nreq > MAX_MSIX)
2837 			nreq = MAX_MSIX;
2838 
2839 		entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
2840 		if (!entries)
2841 			goto no_msi;
2842 
2843 		for (i = 0; i < nreq; ++i)
2844 			entries[i].entry = i;
2845 
2846 		nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2847 					     nreq);
2848 
2849 		if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
2850 			kfree(entries);
2851 			goto no_msi;
2852 		}
2853 		/* 1 is reserved for events (asyncrounous EQ) */
2854 		dev->caps.num_comp_vectors = nreq - 1;
2855 
2856 		priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2857 		bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2858 			    dev->caps.num_ports);
2859 
2860 		for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2861 			if (i == MLX4_EQ_ASYNC)
2862 				continue;
2863 
2864 			priv->eq_table.eq[i].irq =
2865 				entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2866 
2867 			if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2868 				bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2869 					    dev->caps.num_ports);
2870 				/* We don't set affinity hint when there
2871 				 * aren't enough EQs
2872 				 */
2873 			} else {
2874 				set_bit(port,
2875 					priv->eq_table.eq[i].actv_ports.ports);
2876 				if (mlx4_init_affinity_hint(dev, port + 1, i))
2877 					mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
2878 						  i);
2879 			}
2880 			/* We divide the Eqs evenly between the two ports.
2881 			 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
2882 			 * refers to the number of Eqs per port
2883 			 * (i.e eqs_per_port). Theoretically, we would like to
2884 			 * write something like (i + 1) % eqs_per_port == 0.
2885 			 * However, since there's an asynchronous Eq, we have
2886 			 * to skip over it by comparing this condition to
2887 			 * !!((i + 1) > MLX4_EQ_ASYNC).
2888 			 */
2889 			if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
2890 			    ((i + 1) %
2891 			     (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
2892 			    !!((i + 1) > MLX4_EQ_ASYNC))
2893 				/* If dev->caps.num_comp_vectors < dev->caps.num_ports,
2894 				 * everything is shared anyway.
2895 				 */
2896 				port++;
2897 		}
2898 
2899 		dev->flags |= MLX4_FLAG_MSI_X;
2900 
2901 		kfree(entries);
2902 		return;
2903 	}
2904 
2905 no_msi:
2906 	dev->caps.num_comp_vectors = 1;
2907 
2908 	BUG_ON(MLX4_EQ_ASYNC >= 2);
2909 	for (i = 0; i < 2; ++i) {
2910 		priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
2911 		if (i != MLX4_EQ_ASYNC) {
2912 			bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2913 				    dev->caps.num_ports);
2914 		}
2915 	}
2916 }
2917 
2918 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2919 {
2920 	struct devlink *devlink = priv_to_devlink(mlx4_priv(dev));
2921 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2922 	int err;
2923 
2924 	err = devlink_port_register(devlink, &info->devlink_port, port);
2925 	if (err)
2926 		return err;
2927 
2928 	info->dev = dev;
2929 	info->port = port;
2930 	if (!mlx4_is_slave(dev)) {
2931 		mlx4_init_mac_table(dev, &info->mac_table);
2932 		mlx4_init_vlan_table(dev, &info->vlan_table);
2933 		mlx4_init_roce_gid_table(dev, &info->gid_table);
2934 		info->base_qpn = mlx4_get_base_qpn(dev, port);
2935 	}
2936 
2937 	sprintf(info->dev_name, "mlx4_port%d", port);
2938 	info->port_attr.attr.name = info->dev_name;
2939 	if (mlx4_is_mfunc(dev))
2940 		info->port_attr.attr.mode = S_IRUGO;
2941 	else {
2942 		info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2943 		info->port_attr.store     = set_port_type;
2944 	}
2945 	info->port_attr.show      = show_port_type;
2946 	sysfs_attr_init(&info->port_attr.attr);
2947 
2948 	err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
2949 	if (err) {
2950 		mlx4_err(dev, "Failed to create file for port %d\n", port);
2951 		devlink_port_unregister(&info->devlink_port);
2952 		info->port = -1;
2953 	}
2954 
2955 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2956 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
2957 	if (mlx4_is_mfunc(dev))
2958 		info->port_mtu_attr.attr.mode = S_IRUGO;
2959 	else {
2960 		info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2961 		info->port_mtu_attr.store     = set_port_ib_mtu;
2962 	}
2963 	info->port_mtu_attr.show      = show_port_ib_mtu;
2964 	sysfs_attr_init(&info->port_mtu_attr.attr);
2965 
2966 	err = device_create_file(&dev->persist->pdev->dev,
2967 				 &info->port_mtu_attr);
2968 	if (err) {
2969 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
2970 		device_remove_file(&info->dev->persist->pdev->dev,
2971 				   &info->port_attr);
2972 		info->port = -1;
2973 	}
2974 
2975 	return err;
2976 }
2977 
2978 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
2979 {
2980 	if (info->port < 0)
2981 		return;
2982 
2983 	device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
2984 	device_remove_file(&info->dev->persist->pdev->dev,
2985 			   &info->port_mtu_attr);
2986 #ifdef CONFIG_RFS_ACCEL
2987 	free_irq_cpu_rmap(info->rmap);
2988 	info->rmap = NULL;
2989 #endif
2990 }
2991 
2992 static int mlx4_init_steering(struct mlx4_dev *dev)
2993 {
2994 	struct mlx4_priv *priv = mlx4_priv(dev);
2995 	int num_entries = dev->caps.num_ports;
2996 	int i, j;
2997 
2998 	priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
2999 	if (!priv->steer)
3000 		return -ENOMEM;
3001 
3002 	for (i = 0; i < num_entries; i++)
3003 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3004 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
3005 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
3006 		}
3007 	return 0;
3008 }
3009 
3010 static void mlx4_clear_steering(struct mlx4_dev *dev)
3011 {
3012 	struct mlx4_priv *priv = mlx4_priv(dev);
3013 	struct mlx4_steer_index *entry, *tmp_entry;
3014 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
3015 	int num_entries = dev->caps.num_ports;
3016 	int i, j;
3017 
3018 	for (i = 0; i < num_entries; i++) {
3019 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3020 			list_for_each_entry_safe(pqp, tmp_pqp,
3021 						 &priv->steer[i].promisc_qps[j],
3022 						 list) {
3023 				list_del(&pqp->list);
3024 				kfree(pqp);
3025 			}
3026 			list_for_each_entry_safe(entry, tmp_entry,
3027 						 &priv->steer[i].steer_entries[j],
3028 						 list) {
3029 				list_del(&entry->list);
3030 				list_for_each_entry_safe(pqp, tmp_pqp,
3031 							 &entry->duplicates,
3032 							 list) {
3033 					list_del(&pqp->list);
3034 					kfree(pqp);
3035 				}
3036 				kfree(entry);
3037 			}
3038 		}
3039 	}
3040 	kfree(priv->steer);
3041 }
3042 
3043 static int extended_func_num(struct pci_dev *pdev)
3044 {
3045 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
3046 }
3047 
3048 #define MLX4_OWNER_BASE	0x8069c
3049 #define MLX4_OWNER_SIZE	4
3050 
3051 static int mlx4_get_ownership(struct mlx4_dev *dev)
3052 {
3053 	void __iomem *owner;
3054 	u32 ret;
3055 
3056 	if (pci_channel_offline(dev->persist->pdev))
3057 		return -EIO;
3058 
3059 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3060 			MLX4_OWNER_BASE,
3061 			MLX4_OWNER_SIZE);
3062 	if (!owner) {
3063 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3064 		return -ENOMEM;
3065 	}
3066 
3067 	ret = readl(owner);
3068 	iounmap(owner);
3069 	return (int) !!ret;
3070 }
3071 
3072 static void mlx4_free_ownership(struct mlx4_dev *dev)
3073 {
3074 	void __iomem *owner;
3075 
3076 	if (pci_channel_offline(dev->persist->pdev))
3077 		return;
3078 
3079 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3080 			MLX4_OWNER_BASE,
3081 			MLX4_OWNER_SIZE);
3082 	if (!owner) {
3083 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3084 		return;
3085 	}
3086 	writel(0, owner);
3087 	msleep(1000);
3088 	iounmap(owner);
3089 }
3090 
3091 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV)	==\
3092 				  !!((flags) & MLX4_FLAG_MASTER))
3093 
3094 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
3095 			     u8 total_vfs, int existing_vfs, int reset_flow)
3096 {
3097 	u64 dev_flags = dev->flags;
3098 	int err = 0;
3099 	int fw_enabled_sriov_vfs = min(pci_sriov_get_totalvfs(pdev),
3100 					MLX4_MAX_NUM_VF);
3101 
3102 	if (reset_flow) {
3103 		dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
3104 				       GFP_KERNEL);
3105 		if (!dev->dev_vfs)
3106 			goto free_mem;
3107 		return dev_flags;
3108 	}
3109 
3110 	atomic_inc(&pf_loading);
3111 	if (dev->flags &  MLX4_FLAG_SRIOV) {
3112 		if (existing_vfs != total_vfs) {
3113 			mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
3114 				 existing_vfs, total_vfs);
3115 			total_vfs = existing_vfs;
3116 		}
3117 	}
3118 
3119 	dev->dev_vfs = kzalloc(total_vfs * sizeof(*dev->dev_vfs), GFP_KERNEL);
3120 	if (NULL == dev->dev_vfs) {
3121 		mlx4_err(dev, "Failed to allocate memory for VFs\n");
3122 		goto disable_sriov;
3123 	}
3124 
3125 	if (!(dev->flags &  MLX4_FLAG_SRIOV)) {
3126 		if (total_vfs > fw_enabled_sriov_vfs) {
3127 			mlx4_err(dev, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
3128 				 total_vfs, fw_enabled_sriov_vfs);
3129 			err = -ENOMEM;
3130 			goto disable_sriov;
3131 		}
3132 		mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
3133 		err = pci_enable_sriov(pdev, total_vfs);
3134 	}
3135 	if (err) {
3136 		mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
3137 			 err);
3138 		goto disable_sriov;
3139 	} else {
3140 		mlx4_warn(dev, "Running in master mode\n");
3141 		dev_flags |= MLX4_FLAG_SRIOV |
3142 			MLX4_FLAG_MASTER;
3143 		dev_flags &= ~MLX4_FLAG_SLAVE;
3144 		dev->persist->num_vfs = total_vfs;
3145 	}
3146 	return dev_flags;
3147 
3148 disable_sriov:
3149 	atomic_dec(&pf_loading);
3150 free_mem:
3151 	dev->persist->num_vfs = 0;
3152 	kfree(dev->dev_vfs);
3153         dev->dev_vfs = NULL;
3154 	return dev_flags & ~MLX4_FLAG_MASTER;
3155 }
3156 
3157 enum {
3158 	MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
3159 };
3160 
3161 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
3162 			      int *nvfs)
3163 {
3164 	int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
3165 	/* Checking for 64 VFs as a limitation of CX2 */
3166 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
3167 	    requested_vfs >= 64) {
3168 		mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
3169 			 requested_vfs);
3170 		return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
3171 	}
3172 	return 0;
3173 }
3174 
3175 static int mlx4_pci_enable_device(struct mlx4_dev *dev)
3176 {
3177 	struct pci_dev *pdev = dev->persist->pdev;
3178 	int err = 0;
3179 
3180 	mutex_lock(&dev->persist->pci_status_mutex);
3181 	if (dev->persist->pci_status == MLX4_PCI_STATUS_DISABLED) {
3182 		err = pci_enable_device(pdev);
3183 		if (!err)
3184 			dev->persist->pci_status = MLX4_PCI_STATUS_ENABLED;
3185 	}
3186 	mutex_unlock(&dev->persist->pci_status_mutex);
3187 
3188 	return err;
3189 }
3190 
3191 static void mlx4_pci_disable_device(struct mlx4_dev *dev)
3192 {
3193 	struct pci_dev *pdev = dev->persist->pdev;
3194 
3195 	mutex_lock(&dev->persist->pci_status_mutex);
3196 	if (dev->persist->pci_status == MLX4_PCI_STATUS_ENABLED) {
3197 		pci_disable_device(pdev);
3198 		dev->persist->pci_status = MLX4_PCI_STATUS_DISABLED;
3199 	}
3200 	mutex_unlock(&dev->persist->pci_status_mutex);
3201 }
3202 
3203 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
3204 			 int total_vfs, int *nvfs, struct mlx4_priv *priv,
3205 			 int reset_flow)
3206 {
3207 	struct mlx4_dev *dev;
3208 	unsigned sum = 0;
3209 	int err;
3210 	int port;
3211 	int i;
3212 	struct mlx4_dev_cap *dev_cap = NULL;
3213 	int existing_vfs = 0;
3214 
3215 	dev = &priv->dev;
3216 
3217 	INIT_LIST_HEAD(&priv->ctx_list);
3218 	spin_lock_init(&priv->ctx_lock);
3219 
3220 	mutex_init(&priv->port_mutex);
3221 	mutex_init(&priv->bond_mutex);
3222 
3223 	INIT_LIST_HEAD(&priv->pgdir_list);
3224 	mutex_init(&priv->pgdir_mutex);
3225 
3226 	INIT_LIST_HEAD(&priv->bf_list);
3227 	mutex_init(&priv->bf_mutex);
3228 
3229 	dev->rev_id = pdev->revision;
3230 	dev->numa_node = dev_to_node(&pdev->dev);
3231 
3232 	/* Detect if this device is a virtual function */
3233 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3234 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3235 		dev->flags |= MLX4_FLAG_SLAVE;
3236 	} else {
3237 		/* We reset the device and enable SRIOV only for physical
3238 		 * devices.  Try to claim ownership on the device;
3239 		 * if already taken, skip -- do not allow multiple PFs */
3240 		err = mlx4_get_ownership(dev);
3241 		if (err) {
3242 			if (err < 0)
3243 				return err;
3244 			else {
3245 				mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
3246 				return -EINVAL;
3247 			}
3248 		}
3249 
3250 		atomic_set(&priv->opreq_count, 0);
3251 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3252 
3253 		/*
3254 		 * Now reset the HCA before we touch the PCI capabilities or
3255 		 * attempt a firmware command, since a boot ROM may have left
3256 		 * the HCA in an undefined state.
3257 		 */
3258 		err = mlx4_reset(dev);
3259 		if (err) {
3260 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
3261 			goto err_sriov;
3262 		}
3263 
3264 		if (total_vfs) {
3265 			dev->flags = MLX4_FLAG_MASTER;
3266 			existing_vfs = pci_num_vf(pdev);
3267 			if (existing_vfs)
3268 				dev->flags |= MLX4_FLAG_SRIOV;
3269 			dev->persist->num_vfs = total_vfs;
3270 		}
3271 	}
3272 
3273 	/* on load remove any previous indication of internal error,
3274 	 * device is up.
3275 	 */
3276 	dev->persist->state = MLX4_DEVICE_STATE_UP;
3277 
3278 slave_start:
3279 	err = mlx4_cmd_init(dev);
3280 	if (err) {
3281 		mlx4_err(dev, "Failed to init command interface, aborting\n");
3282 		goto err_sriov;
3283 	}
3284 
3285 	/* In slave functions, the communication channel must be initialized
3286 	 * before posting commands. Also, init num_slaves before calling
3287 	 * mlx4_init_hca */
3288 	if (mlx4_is_mfunc(dev)) {
3289 		if (mlx4_is_master(dev)) {
3290 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
3291 
3292 		} else {
3293 			dev->num_slaves = 0;
3294 			err = mlx4_multi_func_init(dev);
3295 			if (err) {
3296 				mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
3297 				goto err_cmd;
3298 			}
3299 		}
3300 	}
3301 
3302 	err = mlx4_init_fw(dev);
3303 	if (err) {
3304 		mlx4_err(dev, "Failed to init fw, aborting.\n");
3305 		goto err_mfunc;
3306 	}
3307 
3308 	if (mlx4_is_master(dev)) {
3309 		/* when we hit the goto slave_start below, dev_cap already initialized */
3310 		if (!dev_cap) {
3311 			dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3312 
3313 			if (!dev_cap) {
3314 				err = -ENOMEM;
3315 				goto err_fw;
3316 			}
3317 
3318 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3319 			if (err) {
3320 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3321 				goto err_fw;
3322 			}
3323 
3324 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3325 				goto err_fw;
3326 
3327 			if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3328 				u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3329 								  total_vfs,
3330 								  existing_vfs,
3331 								  reset_flow);
3332 
3333 				mlx4_close_fw(dev);
3334 				mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3335 				dev->flags = dev_flags;
3336 				if (!SRIOV_VALID_STATE(dev->flags)) {
3337 					mlx4_err(dev, "Invalid SRIOV state\n");
3338 					goto err_sriov;
3339 				}
3340 				err = mlx4_reset(dev);
3341 				if (err) {
3342 					mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3343 					goto err_sriov;
3344 				}
3345 				goto slave_start;
3346 			}
3347 		} else {
3348 			/* Legacy mode FW requires SRIOV to be enabled before
3349 			 * doing QUERY_DEV_CAP, since max_eq's value is different if
3350 			 * SRIOV is enabled.
3351 			 */
3352 			memset(dev_cap, 0, sizeof(*dev_cap));
3353 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3354 			if (err) {
3355 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3356 				goto err_fw;
3357 			}
3358 
3359 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3360 				goto err_fw;
3361 		}
3362 	}
3363 
3364 	err = mlx4_init_hca(dev);
3365 	if (err) {
3366 		if (err == -EACCES) {
3367 			/* Not primary Physical function
3368 			 * Running in slave mode */
3369 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3370 			/* We're not a PF */
3371 			if (dev->flags & MLX4_FLAG_SRIOV) {
3372 				if (!existing_vfs)
3373 					pci_disable_sriov(pdev);
3374 				if (mlx4_is_master(dev) && !reset_flow)
3375 					atomic_dec(&pf_loading);
3376 				dev->flags &= ~MLX4_FLAG_SRIOV;
3377 			}
3378 			if (!mlx4_is_slave(dev))
3379 				mlx4_free_ownership(dev);
3380 			dev->flags |= MLX4_FLAG_SLAVE;
3381 			dev->flags &= ~MLX4_FLAG_MASTER;
3382 			goto slave_start;
3383 		} else
3384 			goto err_fw;
3385 	}
3386 
3387 	if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3388 		u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3389 						  existing_vfs, reset_flow);
3390 
3391 		if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3392 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3393 			dev->flags = dev_flags;
3394 			err = mlx4_cmd_init(dev);
3395 			if (err) {
3396 				/* Only VHCR is cleaned up, so could still
3397 				 * send FW commands
3398 				 */
3399 				mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3400 				goto err_close;
3401 			}
3402 		} else {
3403 			dev->flags = dev_flags;
3404 		}
3405 
3406 		if (!SRIOV_VALID_STATE(dev->flags)) {
3407 			mlx4_err(dev, "Invalid SRIOV state\n");
3408 			goto err_close;
3409 		}
3410 	}
3411 
3412 	/* check if the device is functioning at its maximum possible speed.
3413 	 * No return code for this call, just warn the user in case of PCI
3414 	 * express device capabilities are under-satisfied by the bus.
3415 	 */
3416 	if (!mlx4_is_slave(dev))
3417 		mlx4_check_pcie_caps(dev);
3418 
3419 	/* In master functions, the communication channel must be initialized
3420 	 * after obtaining its address from fw */
3421 	if (mlx4_is_master(dev)) {
3422 		if (dev->caps.num_ports < 2 &&
3423 		    num_vfs_argc > 1) {
3424 			err = -EINVAL;
3425 			mlx4_err(dev,
3426 				 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3427 				 dev->caps.num_ports);
3428 			goto err_close;
3429 		}
3430 		memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
3431 
3432 		for (i = 0;
3433 		     i < sizeof(dev->persist->nvfs)/
3434 		     sizeof(dev->persist->nvfs[0]); i++) {
3435 			unsigned j;
3436 
3437 			for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
3438 				dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3439 				dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3440 					dev->caps.num_ports;
3441 			}
3442 		}
3443 
3444 		/* In master functions, the communication channel
3445 		 * must be initialized after obtaining its address from fw
3446 		 */
3447 		err = mlx4_multi_func_init(dev);
3448 		if (err) {
3449 			mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3450 			goto err_close;
3451 		}
3452 	}
3453 
3454 	err = mlx4_alloc_eq_table(dev);
3455 	if (err)
3456 		goto err_master_mfunc;
3457 
3458 	bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
3459 	mutex_init(&priv->msix_ctl.pool_lock);
3460 
3461 	mlx4_enable_msi_x(dev);
3462 	if ((mlx4_is_mfunc(dev)) &&
3463 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
3464 		err = -ENOSYS;
3465 		mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
3466 		goto err_free_eq;
3467 	}
3468 
3469 	if (!mlx4_is_slave(dev)) {
3470 		err = mlx4_init_steering(dev);
3471 		if (err)
3472 			goto err_disable_msix;
3473 	}
3474 
3475 	err = mlx4_setup_hca(dev);
3476 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3477 	    !mlx4_is_mfunc(dev)) {
3478 		dev->flags &= ~MLX4_FLAG_MSI_X;
3479 		dev->caps.num_comp_vectors = 1;
3480 		pci_disable_msix(pdev);
3481 		err = mlx4_setup_hca(dev);
3482 	}
3483 
3484 	if (err)
3485 		goto err_steer;
3486 
3487 	mlx4_init_quotas(dev);
3488 	/* When PF resources are ready arm its comm channel to enable
3489 	 * getting commands
3490 	 */
3491 	if (mlx4_is_master(dev)) {
3492 		err = mlx4_ARM_COMM_CHANNEL(dev);
3493 		if (err) {
3494 			mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3495 				 err);
3496 			goto err_steer;
3497 		}
3498 	}
3499 
3500 	for (port = 1; port <= dev->caps.num_ports; port++) {
3501 		err = mlx4_init_port_info(dev, port);
3502 		if (err)
3503 			goto err_port;
3504 	}
3505 
3506 	priv->v2p.port1 = 1;
3507 	priv->v2p.port2 = 2;
3508 
3509 	err = mlx4_register_device(dev);
3510 	if (err)
3511 		goto err_port;
3512 
3513 	mlx4_request_modules(dev);
3514 
3515 	mlx4_sense_init(dev);
3516 	mlx4_start_sense(dev);
3517 
3518 	priv->removed = 0;
3519 
3520 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3521 		atomic_dec(&pf_loading);
3522 
3523 	kfree(dev_cap);
3524 	return 0;
3525 
3526 err_port:
3527 	for (--port; port >= 1; --port)
3528 		mlx4_cleanup_port_info(&priv->port[port]);
3529 
3530 	mlx4_cleanup_default_counters(dev);
3531 	if (!mlx4_is_slave(dev))
3532 		mlx4_cleanup_counters_table(dev);
3533 	mlx4_cleanup_qp_table(dev);
3534 	mlx4_cleanup_srq_table(dev);
3535 	mlx4_cleanup_cq_table(dev);
3536 	mlx4_cmd_use_polling(dev);
3537 	mlx4_cleanup_eq_table(dev);
3538 	mlx4_cleanup_mcg_table(dev);
3539 	mlx4_cleanup_mr_table(dev);
3540 	mlx4_cleanup_xrcd_table(dev);
3541 	mlx4_cleanup_pd_table(dev);
3542 	mlx4_cleanup_uar_table(dev);
3543 
3544 err_steer:
3545 	if (!mlx4_is_slave(dev))
3546 		mlx4_clear_steering(dev);
3547 
3548 err_disable_msix:
3549 	if (dev->flags & MLX4_FLAG_MSI_X)
3550 		pci_disable_msix(pdev);
3551 
3552 err_free_eq:
3553 	mlx4_free_eq_table(dev);
3554 
3555 err_master_mfunc:
3556 	if (mlx4_is_master(dev)) {
3557 		mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
3558 		mlx4_multi_func_cleanup(dev);
3559 	}
3560 
3561 	if (mlx4_is_slave(dev)) {
3562 		kfree(dev->caps.qp0_qkey);
3563 		kfree(dev->caps.qp0_tunnel);
3564 		kfree(dev->caps.qp0_proxy);
3565 		kfree(dev->caps.qp1_tunnel);
3566 		kfree(dev->caps.qp1_proxy);
3567 	}
3568 
3569 err_close:
3570 	mlx4_close_hca(dev);
3571 
3572 err_fw:
3573 	mlx4_close_fw(dev);
3574 
3575 err_mfunc:
3576 	if (mlx4_is_slave(dev))
3577 		mlx4_multi_func_cleanup(dev);
3578 
3579 err_cmd:
3580 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3581 
3582 err_sriov:
3583 	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3584 		pci_disable_sriov(pdev);
3585 		dev->flags &= ~MLX4_FLAG_SRIOV;
3586 	}
3587 
3588 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3589 		atomic_dec(&pf_loading);
3590 
3591 	kfree(priv->dev.dev_vfs);
3592 
3593 	if (!mlx4_is_slave(dev))
3594 		mlx4_free_ownership(dev);
3595 
3596 	kfree(dev_cap);
3597 	return err;
3598 }
3599 
3600 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3601 			   struct mlx4_priv *priv)
3602 {
3603 	int err;
3604 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3605 	int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3606 	const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3607 		{2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3608 	unsigned total_vfs = 0;
3609 	unsigned int i;
3610 
3611 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3612 
3613 	err = mlx4_pci_enable_device(&priv->dev);
3614 	if (err) {
3615 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3616 		return err;
3617 	}
3618 
3619 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3620 	 * per port, we must limit the number of VFs to 63 (since their are
3621 	 * 128 MACs)
3622 	 */
3623 	for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
3624 	     total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3625 		nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3626 		if (nvfs[i] < 0) {
3627 			dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3628 			err = -EINVAL;
3629 			goto err_disable_pdev;
3630 		}
3631 	}
3632 	for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
3633 	     i++) {
3634 		prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3635 		if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3636 			dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3637 			err = -EINVAL;
3638 			goto err_disable_pdev;
3639 		}
3640 	}
3641 	if (total_vfs > MLX4_MAX_NUM_VF) {
3642 		dev_err(&pdev->dev,
3643 			"Requested more VF's (%d) than allowed by hw (%d)\n",
3644 			total_vfs, MLX4_MAX_NUM_VF);
3645 		err = -EINVAL;
3646 		goto err_disable_pdev;
3647 	}
3648 
3649 	for (i = 0; i < MLX4_MAX_PORTS; i++) {
3650 		if (nvfs[i] + nvfs[2] > MLX4_MAX_NUM_VF_P_PORT) {
3651 			dev_err(&pdev->dev,
3652 				"Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3653 				nvfs[i] + nvfs[2], i + 1,
3654 				MLX4_MAX_NUM_VF_P_PORT);
3655 			err = -EINVAL;
3656 			goto err_disable_pdev;
3657 		}
3658 	}
3659 
3660 	/* Check for BARs. */
3661 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3662 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3663 		dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3664 			pci_dev_data, pci_resource_flags(pdev, 0));
3665 		err = -ENODEV;
3666 		goto err_disable_pdev;
3667 	}
3668 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3669 		dev_err(&pdev->dev, "Missing UAR, aborting\n");
3670 		err = -ENODEV;
3671 		goto err_disable_pdev;
3672 	}
3673 
3674 	err = pci_request_regions(pdev, DRV_NAME);
3675 	if (err) {
3676 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3677 		goto err_disable_pdev;
3678 	}
3679 
3680 	pci_set_master(pdev);
3681 
3682 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3683 	if (err) {
3684 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3685 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3686 		if (err) {
3687 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3688 			goto err_release_regions;
3689 		}
3690 	}
3691 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3692 	if (err) {
3693 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3694 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3695 		if (err) {
3696 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3697 			goto err_release_regions;
3698 		}
3699 	}
3700 
3701 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
3702 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3703 	/* Detect if this device is a virtual function */
3704 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3705 		/* When acting as pf, we normally skip vfs unless explicitly
3706 		 * requested to probe them.
3707 		 */
3708 		if (total_vfs) {
3709 			unsigned vfs_offset = 0;
3710 
3711 			for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
3712 			     vfs_offset + nvfs[i] < extended_func_num(pdev);
3713 			     vfs_offset += nvfs[i], i++)
3714 				;
3715 			if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
3716 				err = -ENODEV;
3717 				goto err_release_regions;
3718 			}
3719 			if ((extended_func_num(pdev) - vfs_offset)
3720 			    > prb_vf[i]) {
3721 				dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3722 					 extended_func_num(pdev));
3723 				err = -ENODEV;
3724 				goto err_release_regions;
3725 			}
3726 		}
3727 	}
3728 
3729 	err = mlx4_catas_init(&priv->dev);
3730 	if (err)
3731 		goto err_release_regions;
3732 
3733 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3734 	if (err)
3735 		goto err_catas;
3736 
3737 	return 0;
3738 
3739 err_catas:
3740 	mlx4_catas_end(&priv->dev);
3741 
3742 err_release_regions:
3743 	pci_release_regions(pdev);
3744 
3745 err_disable_pdev:
3746 	mlx4_pci_disable_device(&priv->dev);
3747 	pci_set_drvdata(pdev, NULL);
3748 	return err;
3749 }
3750 
3751 static int mlx4_devlink_port_type_set(struct devlink_port *devlink_port,
3752 				      enum devlink_port_type port_type)
3753 {
3754 	struct mlx4_port_info *info = container_of(devlink_port,
3755 						   struct mlx4_port_info,
3756 						   devlink_port);
3757 	enum mlx4_port_type mlx4_port_type;
3758 
3759 	switch (port_type) {
3760 	case DEVLINK_PORT_TYPE_AUTO:
3761 		mlx4_port_type = MLX4_PORT_TYPE_AUTO;
3762 		break;
3763 	case DEVLINK_PORT_TYPE_ETH:
3764 		mlx4_port_type = MLX4_PORT_TYPE_ETH;
3765 		break;
3766 	case DEVLINK_PORT_TYPE_IB:
3767 		mlx4_port_type = MLX4_PORT_TYPE_IB;
3768 		break;
3769 	default:
3770 		return -EOPNOTSUPP;
3771 	}
3772 
3773 	return __set_port_type(info, mlx4_port_type);
3774 }
3775 
3776 static const struct devlink_ops mlx4_devlink_ops = {
3777 	.port_type_set	= mlx4_devlink_port_type_set,
3778 };
3779 
3780 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3781 {
3782 	struct devlink *devlink;
3783 	struct mlx4_priv *priv;
3784 	struct mlx4_dev *dev;
3785 	int ret;
3786 
3787 	printk_once(KERN_INFO "%s", mlx4_version);
3788 
3789 	devlink = devlink_alloc(&mlx4_devlink_ops, sizeof(*priv));
3790 	if (!devlink)
3791 		return -ENOMEM;
3792 	priv = devlink_priv(devlink);
3793 
3794 	dev       = &priv->dev;
3795 	dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3796 	if (!dev->persist) {
3797 		ret = -ENOMEM;
3798 		goto err_devlink_free;
3799 	}
3800 	dev->persist->pdev = pdev;
3801 	dev->persist->dev = dev;
3802 	pci_set_drvdata(pdev, dev->persist);
3803 	priv->pci_dev_data = id->driver_data;
3804 	mutex_init(&dev->persist->device_state_mutex);
3805 	mutex_init(&dev->persist->interface_state_mutex);
3806 	mutex_init(&dev->persist->pci_status_mutex);
3807 
3808 	ret = devlink_register(devlink, &pdev->dev);
3809 	if (ret)
3810 		goto err_persist_free;
3811 
3812 	ret =  __mlx4_init_one(pdev, id->driver_data, priv);
3813 	if (ret)
3814 		goto err_devlink_unregister;
3815 
3816 	pci_save_state(pdev);
3817 	return 0;
3818 
3819 err_devlink_unregister:
3820 	devlink_unregister(devlink);
3821 err_persist_free:
3822 	kfree(dev->persist);
3823 err_devlink_free:
3824 	devlink_free(devlink);
3825 	return ret;
3826 }
3827 
3828 static void mlx4_clean_dev(struct mlx4_dev *dev)
3829 {
3830 	struct mlx4_dev_persistent *persist = dev->persist;
3831 	struct mlx4_priv *priv = mlx4_priv(dev);
3832 	unsigned long	flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
3833 
3834 	memset(priv, 0, sizeof(*priv));
3835 	priv->dev.persist = persist;
3836 	priv->dev.flags = flags;
3837 }
3838 
3839 static void mlx4_unload_one(struct pci_dev *pdev)
3840 {
3841 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3842 	struct mlx4_dev  *dev  = persist->dev;
3843 	struct mlx4_priv *priv = mlx4_priv(dev);
3844 	int               pci_dev_data;
3845 	int p, i;
3846 
3847 	if (priv->removed)
3848 		return;
3849 
3850 	/* saving current ports type for further use */
3851 	for (i = 0; i < dev->caps.num_ports; i++) {
3852 		dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
3853 		dev->persist->curr_port_poss_type[i] = dev->caps.
3854 						       possible_type[i + 1];
3855 	}
3856 
3857 	pci_dev_data = priv->pci_dev_data;
3858 
3859 	mlx4_stop_sense(dev);
3860 	mlx4_unregister_device(dev);
3861 
3862 	for (p = 1; p <= dev->caps.num_ports; p++) {
3863 		mlx4_cleanup_port_info(&priv->port[p]);
3864 		mlx4_CLOSE_PORT(dev, p);
3865 	}
3866 
3867 	if (mlx4_is_master(dev))
3868 		mlx4_free_resource_tracker(dev,
3869 					   RES_TR_FREE_SLAVES_ONLY);
3870 
3871 	mlx4_cleanup_default_counters(dev);
3872 	if (!mlx4_is_slave(dev))
3873 		mlx4_cleanup_counters_table(dev);
3874 	mlx4_cleanup_qp_table(dev);
3875 	mlx4_cleanup_srq_table(dev);
3876 	mlx4_cleanup_cq_table(dev);
3877 	mlx4_cmd_use_polling(dev);
3878 	mlx4_cleanup_eq_table(dev);
3879 	mlx4_cleanup_mcg_table(dev);
3880 	mlx4_cleanup_mr_table(dev);
3881 	mlx4_cleanup_xrcd_table(dev);
3882 	mlx4_cleanup_pd_table(dev);
3883 
3884 	if (mlx4_is_master(dev))
3885 		mlx4_free_resource_tracker(dev,
3886 					   RES_TR_FREE_STRUCTS_ONLY);
3887 
3888 	iounmap(priv->kar);
3889 	mlx4_uar_free(dev, &priv->driver_uar);
3890 	mlx4_cleanup_uar_table(dev);
3891 	if (!mlx4_is_slave(dev))
3892 		mlx4_clear_steering(dev);
3893 	mlx4_free_eq_table(dev);
3894 	if (mlx4_is_master(dev))
3895 		mlx4_multi_func_cleanup(dev);
3896 	mlx4_close_hca(dev);
3897 	mlx4_close_fw(dev);
3898 	if (mlx4_is_slave(dev))
3899 		mlx4_multi_func_cleanup(dev);
3900 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3901 
3902 	if (dev->flags & MLX4_FLAG_MSI_X)
3903 		pci_disable_msix(pdev);
3904 
3905 	if (!mlx4_is_slave(dev))
3906 		mlx4_free_ownership(dev);
3907 
3908 	kfree(dev->caps.qp0_qkey);
3909 	kfree(dev->caps.qp0_tunnel);
3910 	kfree(dev->caps.qp0_proxy);
3911 	kfree(dev->caps.qp1_tunnel);
3912 	kfree(dev->caps.qp1_proxy);
3913 	kfree(dev->dev_vfs);
3914 
3915 	mlx4_clean_dev(dev);
3916 	priv->pci_dev_data = pci_dev_data;
3917 	priv->removed = 1;
3918 }
3919 
3920 static void mlx4_remove_one(struct pci_dev *pdev)
3921 {
3922 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3923 	struct mlx4_dev  *dev  = persist->dev;
3924 	struct mlx4_priv *priv = mlx4_priv(dev);
3925 	struct devlink *devlink = priv_to_devlink(priv);
3926 	int active_vfs = 0;
3927 
3928 	mutex_lock(&persist->interface_state_mutex);
3929 	persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
3930 	mutex_unlock(&persist->interface_state_mutex);
3931 
3932 	/* Disabling SR-IOV is not allowed while there are active vf's */
3933 	if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
3934 		active_vfs = mlx4_how_many_lives_vf(dev);
3935 		if (active_vfs) {
3936 			pr_warn("Removing PF when there are active VF's !!\n");
3937 			pr_warn("Will not disable SR-IOV.\n");
3938 		}
3939 	}
3940 
3941 	/* device marked to be under deletion running now without the lock
3942 	 * letting other tasks to be terminated
3943 	 */
3944 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3945 		mlx4_unload_one(pdev);
3946 	else
3947 		mlx4_info(dev, "%s: interface is down\n", __func__);
3948 	mlx4_catas_end(dev);
3949 	if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
3950 		mlx4_warn(dev, "Disabling SR-IOV\n");
3951 		pci_disable_sriov(pdev);
3952 	}
3953 
3954 	pci_release_regions(pdev);
3955 	mlx4_pci_disable_device(dev);
3956 	devlink_unregister(devlink);
3957 	kfree(dev->persist);
3958 	devlink_free(devlink);
3959 	pci_set_drvdata(pdev, NULL);
3960 }
3961 
3962 static int restore_current_port_types(struct mlx4_dev *dev,
3963 				      enum mlx4_port_type *types,
3964 				      enum mlx4_port_type *poss_types)
3965 {
3966 	struct mlx4_priv *priv = mlx4_priv(dev);
3967 	int err, i;
3968 
3969 	mlx4_stop_sense(dev);
3970 
3971 	mutex_lock(&priv->port_mutex);
3972 	for (i = 0; i < dev->caps.num_ports; i++)
3973 		dev->caps.possible_type[i + 1] = poss_types[i];
3974 	err = mlx4_change_port_types(dev, types);
3975 	mlx4_start_sense(dev);
3976 	mutex_unlock(&priv->port_mutex);
3977 
3978 	return err;
3979 }
3980 
3981 int mlx4_restart_one(struct pci_dev *pdev)
3982 {
3983 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3984 	struct mlx4_dev	 *dev  = persist->dev;
3985 	struct mlx4_priv *priv = mlx4_priv(dev);
3986 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3987 	int pci_dev_data, err, total_vfs;
3988 
3989 	pci_dev_data = priv->pci_dev_data;
3990 	total_vfs = dev->persist->num_vfs;
3991 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
3992 
3993 	mlx4_unload_one(pdev);
3994 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
3995 	if (err) {
3996 		mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
3997 			 __func__, pci_name(pdev), err);
3998 		return err;
3999 	}
4000 
4001 	err = restore_current_port_types(dev, dev->persist->curr_port_type,
4002 					 dev->persist->curr_port_poss_type);
4003 	if (err)
4004 		mlx4_err(dev, "could not restore original port types (%d)\n",
4005 			 err);
4006 
4007 	return err;
4008 }
4009 
4010 static const struct pci_device_id mlx4_pci_table[] = {
4011 	/* MT25408 "Hermon" SDR */
4012 	{ PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4013 	/* MT25408 "Hermon" DDR */
4014 	{ PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4015 	/* MT25408 "Hermon" QDR */
4016 	{ PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4017 	/* MT25408 "Hermon" DDR PCIe gen2 */
4018 	{ PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4019 	/* MT25408 "Hermon" QDR PCIe gen2 */
4020 	{ PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4021 	/* MT25408 "Hermon" EN 10GigE */
4022 	{ PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4023 	/* MT25408 "Hermon" EN 10GigE PCIe gen2 */
4024 	{ PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4025 	/* MT25458 ConnectX EN 10GBASE-T 10GigE */
4026 	{ PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4027 	/* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
4028 	{ PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4029 	/* MT26468 ConnectX EN 10GigE PCIe gen2*/
4030 	{ PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4031 	/* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
4032 	{ PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4033 	/* MT26478 ConnectX2 40GigE PCIe gen2 */
4034 	{ PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
4035 	/* MT25400 Family [ConnectX-2 Virtual Function] */
4036 	{ PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
4037 	/* MT27500 Family [ConnectX-3] */
4038 	{ PCI_VDEVICE(MELLANOX, 0x1003), 0 },
4039 	/* MT27500 Family [ConnectX-3 Virtual Function] */
4040 	{ PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
4041 	{ PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
4042 	{ PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
4043 	{ PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
4044 	{ PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
4045 	{ PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
4046 	{ PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
4047 	{ PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
4048 	{ PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
4049 	{ PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
4050 	{ PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
4051 	{ PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
4052 	{ PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
4053 	{ 0, }
4054 };
4055 
4056 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
4057 
4058 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
4059 					      pci_channel_state_t state)
4060 {
4061 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4062 
4063 	mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
4064 	mlx4_enter_error_state(persist);
4065 
4066 	mutex_lock(&persist->interface_state_mutex);
4067 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4068 		mlx4_unload_one(pdev);
4069 
4070 	mutex_unlock(&persist->interface_state_mutex);
4071 	if (state == pci_channel_io_perm_failure)
4072 		return PCI_ERS_RESULT_DISCONNECT;
4073 
4074 	mlx4_pci_disable_device(persist->dev);
4075 	return PCI_ERS_RESULT_NEED_RESET;
4076 }
4077 
4078 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
4079 {
4080 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4081 	struct mlx4_dev	 *dev  = persist->dev;
4082 	int err;
4083 
4084 	mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
4085 	err = mlx4_pci_enable_device(dev);
4086 	if (err) {
4087 		mlx4_err(dev, "Can not re-enable device, err=%d\n", err);
4088 		return PCI_ERS_RESULT_DISCONNECT;
4089 	}
4090 
4091 	pci_set_master(pdev);
4092 	pci_restore_state(pdev);
4093 	pci_save_state(pdev);
4094 	return PCI_ERS_RESULT_RECOVERED;
4095 }
4096 
4097 static void mlx4_pci_resume(struct pci_dev *pdev)
4098 {
4099 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4100 	struct mlx4_dev	 *dev  = persist->dev;
4101 	struct mlx4_priv *priv = mlx4_priv(dev);
4102 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4103 	int total_vfs;
4104 	int err;
4105 
4106 	mlx4_err(dev, "%s was called\n", __func__);
4107 	total_vfs = dev->persist->num_vfs;
4108 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4109 
4110 	mutex_lock(&persist->interface_state_mutex);
4111 	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4112 		err = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
4113 				    priv, 1);
4114 		if (err) {
4115 			mlx4_err(dev, "%s: mlx4_load_one failed, err=%d\n",
4116 				 __func__,  err);
4117 			goto end;
4118 		}
4119 
4120 		err = restore_current_port_types(dev, dev->persist->
4121 						 curr_port_type, dev->persist->
4122 						 curr_port_poss_type);
4123 		if (err)
4124 			mlx4_err(dev, "could not restore original port types (%d)\n", err);
4125 	}
4126 end:
4127 	mutex_unlock(&persist->interface_state_mutex);
4128 
4129 }
4130 
4131 static void mlx4_shutdown(struct pci_dev *pdev)
4132 {
4133 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4134 
4135 	mlx4_info(persist->dev, "mlx4_shutdown was called\n");
4136 	mutex_lock(&persist->interface_state_mutex);
4137 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4138 		mlx4_unload_one(pdev);
4139 	mutex_unlock(&persist->interface_state_mutex);
4140 }
4141 
4142 static const struct pci_error_handlers mlx4_err_handler = {
4143 	.error_detected = mlx4_pci_err_detected,
4144 	.slot_reset     = mlx4_pci_slot_reset,
4145 	.resume		= mlx4_pci_resume,
4146 };
4147 
4148 static struct pci_driver mlx4_driver = {
4149 	.name		= DRV_NAME,
4150 	.id_table	= mlx4_pci_table,
4151 	.probe		= mlx4_init_one,
4152 	.shutdown	= mlx4_shutdown,
4153 	.remove		= mlx4_remove_one,
4154 	.err_handler    = &mlx4_err_handler,
4155 };
4156 
4157 static int __init mlx4_verify_params(void)
4158 {
4159 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
4160 		pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
4161 		return -1;
4162 	}
4163 
4164 	if (log_num_vlan != 0)
4165 		pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
4166 			MLX4_LOG_NUM_VLANS);
4167 
4168 	if (use_prio != 0)
4169 		pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
4170 
4171 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
4172 		pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
4173 			log_mtts_per_seg);
4174 		return -1;
4175 	}
4176 
4177 	/* Check if module param for ports type has legal combination */
4178 	if (port_type_array[0] == false && port_type_array[1] == true) {
4179 		pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
4180 		port_type_array[0] = true;
4181 	}
4182 
4183 	if (mlx4_log_num_mgm_entry_size < -7 ||
4184 	    (mlx4_log_num_mgm_entry_size > 0 &&
4185 	     (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
4186 	      mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
4187 		pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
4188 			mlx4_log_num_mgm_entry_size,
4189 			MLX4_MIN_MGM_LOG_ENTRY_SIZE,
4190 			MLX4_MAX_MGM_LOG_ENTRY_SIZE);
4191 		return -1;
4192 	}
4193 
4194 	return 0;
4195 }
4196 
4197 static int __init mlx4_init(void)
4198 {
4199 	int ret;
4200 
4201 	if (mlx4_verify_params())
4202 		return -EINVAL;
4203 
4204 
4205 	mlx4_wq = create_singlethread_workqueue("mlx4");
4206 	if (!mlx4_wq)
4207 		return -ENOMEM;
4208 
4209 	ret = pci_register_driver(&mlx4_driver);
4210 	if (ret < 0)
4211 		destroy_workqueue(mlx4_wq);
4212 	return ret < 0 ? ret : 0;
4213 }
4214 
4215 static void __exit mlx4_cleanup(void)
4216 {
4217 	pci_unregister_driver(&mlx4_driver);
4218 	destroy_workqueue(mlx4_wq);
4219 }
4220 
4221 module_init(mlx4_init);
4222 module_exit(mlx4_cleanup);
4223