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/kernel.h>
38 #include <linux/init.h>
39 #include <linux/errno.h>
40 #include <linux/pci.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/slab.h>
43 #include <linux/io-mapping.h>
44 #include <linux/delay.h>
45 #include <linux/kmod.h>
46 #include <linux/etherdevice.h>
47 #include <net/devlink.h>
48 
49 #include <linux/mlx4/device.h>
50 #include <linux/mlx4/doorbell.h>
51 
52 #include "mlx4.h"
53 #include "fw.h"
54 #include "icm.h"
55 
56 MODULE_AUTHOR("Roland Dreier");
57 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
58 MODULE_LICENSE("Dual BSD/GPL");
59 MODULE_VERSION(DRV_VERSION);
60 
61 struct workqueue_struct *mlx4_wq;
62 
63 #ifdef CONFIG_MLX4_DEBUG
64 
65 int mlx4_debug_level = 0;
66 module_param_named(debug_level, mlx4_debug_level, int, 0644);
67 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
68 
69 #endif /* CONFIG_MLX4_DEBUG */
70 
71 #ifdef CONFIG_PCI_MSI
72 
73 static int msi_x = 1;
74 module_param(msi_x, int, 0444);
75 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
76 
77 #else /* CONFIG_PCI_MSI */
78 
79 #define msi_x (0)
80 
81 #endif /* CONFIG_PCI_MSI */
82 
83 static uint8_t num_vfs[3] = {0, 0, 0};
84 static int num_vfs_argc;
85 module_param_array(num_vfs, byte , &num_vfs_argc, 0444);
86 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
87 			  "num_vfs=port1,port2,port1+2");
88 
89 static uint8_t probe_vf[3] = {0, 0, 0};
90 static int probe_vfs_argc;
91 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
92 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
93 			   "probe_vf=port1,port2,port1+2");
94 
95 static int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
96 module_param_named(log_num_mgm_entry_size,
97 			mlx4_log_num_mgm_entry_size, int, 0444);
98 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
99 					 " of qp per mcg, for example:"
100 					 " 10 gives 248.range: 7 <="
101 					 " log_num_mgm_entry_size <= 12."
102 					 " To activate device managed"
103 					 " flow steering when available, set to -1");
104 
105 static bool enable_64b_cqe_eqe = true;
106 module_param(enable_64b_cqe_eqe, bool, 0444);
107 MODULE_PARM_DESC(enable_64b_cqe_eqe,
108 		 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
109 
110 static bool enable_4k_uar;
111 module_param(enable_4k_uar, bool, 0444);
112 MODULE_PARM_DESC(enable_4k_uar,
113 		 "Enable using 4K UAR. Should not be enabled if have VFs which do not support 4K UARs (default: false)");
114 
115 #define PF_CONTEXT_BEHAVIOUR_MASK	(MLX4_FUNC_CAP_64B_EQE_CQE | \
116 					 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
117 					 MLX4_FUNC_CAP_DMFS_A0_STATIC)
118 
119 #define RESET_PERSIST_MASK_FLAGS	(MLX4_FLAG_SRIOV)
120 
121 static char mlx4_version[] =
122 	DRV_NAME ": Mellanox ConnectX core driver v"
123 	DRV_VERSION "\n";
124 
125 static const struct mlx4_profile default_profile = {
126 	.num_qp		= 1 << 18,
127 	.num_srq	= 1 << 16,
128 	.rdmarc_per_qp	= 1 << 4,
129 	.num_cq		= 1 << 16,
130 	.num_mcg	= 1 << 13,
131 	.num_mpt	= 1 << 19,
132 	.num_mtt	= 1 << 20, /* It is really num mtt segements */
133 };
134 
135 static const struct mlx4_profile low_mem_profile = {
136 	.num_qp		= 1 << 17,
137 	.num_srq	= 1 << 6,
138 	.rdmarc_per_qp	= 1 << 4,
139 	.num_cq		= 1 << 8,
140 	.num_mcg	= 1 << 8,
141 	.num_mpt	= 1 << 9,
142 	.num_mtt	= 1 << 7,
143 };
144 
145 static int log_num_mac = 7;
146 module_param_named(log_num_mac, log_num_mac, int, 0444);
147 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
148 
149 static int log_num_vlan;
150 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
151 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
152 /* Log2 max number of VLANs per ETH port (0-7) */
153 #define MLX4_LOG_NUM_VLANS 7
154 #define MLX4_MIN_LOG_NUM_VLANS 0
155 #define MLX4_MIN_LOG_NUM_MAC 1
156 
157 static bool use_prio;
158 module_param_named(use_prio, use_prio, bool, 0444);
159 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
160 
161 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
162 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
163 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
164 
165 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
166 static int arr_argc = 2;
167 module_param_array(port_type_array, int, &arr_argc, 0444);
168 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
169 				"1 for IB, 2 for Ethernet");
170 
171 struct mlx4_port_config {
172 	struct list_head list;
173 	enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
174 	struct pci_dev *pdev;
175 };
176 
177 static atomic_t pf_loading = ATOMIC_INIT(0);
178 
179 static inline void mlx4_set_num_reserved_uars(struct mlx4_dev *dev,
180 					      struct mlx4_dev_cap *dev_cap)
181 {
182 	/* The reserved_uars is calculated by system page size unit.
183 	 * Therefore, adjustment is added when the uar page size is less
184 	 * than the system page size
185 	 */
186 	dev->caps.reserved_uars	=
187 		max_t(int,
188 		      mlx4_get_num_reserved_uar(dev),
189 		      dev_cap->reserved_uars /
190 			(1 << (PAGE_SHIFT - dev->uar_page_shift)));
191 }
192 
193 int mlx4_check_port_params(struct mlx4_dev *dev,
194 			   enum mlx4_port_type *port_type)
195 {
196 	int i;
197 
198 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
199 		for (i = 0; i < dev->caps.num_ports - 1; i++) {
200 			if (port_type[i] != port_type[i + 1]) {
201 				mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
202 				return -EINVAL;
203 			}
204 		}
205 	}
206 
207 	for (i = 0; i < dev->caps.num_ports; i++) {
208 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
209 			mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
210 				 i + 1);
211 			return -EINVAL;
212 		}
213 	}
214 	return 0;
215 }
216 
217 static void mlx4_set_port_mask(struct mlx4_dev *dev)
218 {
219 	int i;
220 
221 	for (i = 1; i <= dev->caps.num_ports; ++i)
222 		dev->caps.port_mask[i] = dev->caps.port_type[i];
223 }
224 
225 enum {
226 	MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
227 };
228 
229 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
230 {
231 	int err = 0;
232 	struct mlx4_func func;
233 
234 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
235 		err = mlx4_QUERY_FUNC(dev, &func, 0);
236 		if (err) {
237 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
238 			return err;
239 		}
240 		dev_cap->max_eqs = func.max_eq;
241 		dev_cap->reserved_eqs = func.rsvd_eqs;
242 		dev_cap->reserved_uars = func.rsvd_uars;
243 		err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
244 	}
245 	return err;
246 }
247 
248 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
249 {
250 	struct mlx4_caps *dev_cap = &dev->caps;
251 
252 	/* FW not supporting or cancelled by user */
253 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
254 	    !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
255 		return;
256 
257 	/* Must have 64B CQE_EQE enabled by FW to use bigger stride
258 	 * When FW has NCSI it may decide not to report 64B CQE/EQEs
259 	 */
260 	if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
261 	    !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
262 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
263 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
264 		return;
265 	}
266 
267 	if (cache_line_size() == 128 || cache_line_size() == 256) {
268 		mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
269 		/* Changing the real data inside CQE size to 32B */
270 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
271 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
272 
273 		if (mlx4_is_master(dev))
274 			dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
275 	} else {
276 		if (cache_line_size() != 32  && cache_line_size() != 64)
277 			mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
278 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
279 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
280 	}
281 }
282 
283 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
284 			  struct mlx4_port_cap *port_cap)
285 {
286 	dev->caps.vl_cap[port]	    = port_cap->max_vl;
287 	dev->caps.ib_mtu_cap[port]	    = port_cap->ib_mtu;
288 	dev->phys_caps.gid_phys_table_len[port]  = port_cap->max_gids;
289 	dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
290 	/* set gid and pkey table operating lengths by default
291 	 * to non-sriov values
292 	 */
293 	dev->caps.gid_table_len[port]  = port_cap->max_gids;
294 	dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
295 	dev->caps.port_width_cap[port] = port_cap->max_port_width;
296 	dev->caps.eth_mtu_cap[port]    = port_cap->eth_mtu;
297 	dev->caps.max_tc_eth	       = port_cap->max_tc_eth;
298 	dev->caps.def_mac[port]        = port_cap->def_mac;
299 	dev->caps.supported_type[port] = port_cap->supported_port_types;
300 	dev->caps.suggested_type[port] = port_cap->suggested_type;
301 	dev->caps.default_sense[port] = port_cap->default_sense;
302 	dev->caps.trans_type[port]	    = port_cap->trans_type;
303 	dev->caps.vendor_oui[port]     = port_cap->vendor_oui;
304 	dev->caps.wavelength[port]     = port_cap->wavelength;
305 	dev->caps.trans_code[port]     = port_cap->trans_code;
306 
307 	return 0;
308 }
309 
310 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
311 			 struct mlx4_port_cap *port_cap)
312 {
313 	int err = 0;
314 
315 	err = mlx4_QUERY_PORT(dev, port, port_cap);
316 
317 	if (err)
318 		mlx4_err(dev, "QUERY_PORT command failed.\n");
319 
320 	return err;
321 }
322 
323 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
324 {
325 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
326 		return;
327 
328 	if (mlx4_is_mfunc(dev)) {
329 		mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
330 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
331 		return;
332 	}
333 
334 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
335 		mlx4_dbg(dev,
336 			 "Keep FCS is not supported - Disabling Ignore FCS");
337 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
338 		return;
339 	}
340 }
341 
342 #define MLX4_A0_STEERING_TABLE_SIZE	256
343 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
344 {
345 	int err;
346 	int i;
347 
348 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
349 	if (err) {
350 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
351 		return err;
352 	}
353 	mlx4_dev_cap_dump(dev, dev_cap);
354 
355 	if (dev_cap->min_page_sz > PAGE_SIZE) {
356 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
357 			 dev_cap->min_page_sz, PAGE_SIZE);
358 		return -ENODEV;
359 	}
360 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
361 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
362 			 dev_cap->num_ports, MLX4_MAX_PORTS);
363 		return -ENODEV;
364 	}
365 
366 	if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
367 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
368 			 dev_cap->uar_size,
369 			 (unsigned long long)
370 			 pci_resource_len(dev->persist->pdev, 2));
371 		return -ENODEV;
372 	}
373 
374 	dev->caps.num_ports	     = dev_cap->num_ports;
375 	dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
376 	dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
377 				      dev->caps.num_sys_eqs :
378 				      MLX4_MAX_EQ_NUM;
379 	for (i = 1; i <= dev->caps.num_ports; ++i) {
380 		err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
381 		if (err) {
382 			mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
383 			return err;
384 		}
385 	}
386 
387 	dev->caps.uar_page_size	     = PAGE_SIZE;
388 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
389 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
390 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
391 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
392 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
393 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
394 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
395 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
396 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
397 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
398 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
399 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
400 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
401 	/*
402 	 * Subtract 1 from the limit because we need to allocate a
403 	 * spare CQE so the HCA HW can tell the difference between an
404 	 * empty CQ and a full CQ.
405 	 */
406 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
407 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
408 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
409 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
410 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
411 
412 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
413 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
414 					dev_cap->reserved_xrcds : 0;
415 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
416 					dev_cap->max_xrcds : 0;
417 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
418 
419 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
420 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
421 	dev->caps.flags		     = dev_cap->flags;
422 	dev->caps.flags2	     = dev_cap->flags2;
423 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
424 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
425 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
426 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
427 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
428 	dev->caps.wol_port[1]          = dev_cap->wol_port[1];
429 	dev->caps.wol_port[2]          = dev_cap->wol_port[2];
430 
431 	/* Save uar page shift */
432 	if (!mlx4_is_slave(dev)) {
433 		/* Virtual PCI function needs to determine UAR page size from
434 		 * firmware. Only master PCI function can set the uar page size
435 		 */
436 		if (enable_4k_uar || !dev->persist->num_vfs)
437 			dev->uar_page_shift = DEFAULT_UAR_PAGE_SHIFT;
438 		else
439 			dev->uar_page_shift = PAGE_SHIFT;
440 
441 		mlx4_set_num_reserved_uars(dev, dev_cap);
442 	}
443 
444 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN) {
445 		struct mlx4_init_hca_param hca_param;
446 
447 		memset(&hca_param, 0, sizeof(hca_param));
448 		err = mlx4_QUERY_HCA(dev, &hca_param);
449 		/* Turn off PHV_EN flag in case phv_check_en is set.
450 		 * phv_check_en is a HW check that parse the packet and verify
451 		 * phv bit was reported correctly in the wqe. To allow QinQ
452 		 * PHV_EN flag should be set and phv_check_en must be cleared
453 		 * otherwise QinQ packets will be drop by the HW.
454 		 */
455 		if (err || hca_param.phv_check_en)
456 			dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_PHV_EN;
457 	}
458 
459 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
460 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
461 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
462 	/* Don't do sense port on multifunction devices (for now at least) */
463 	if (mlx4_is_mfunc(dev))
464 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
465 
466 	if (mlx4_low_memory_profile()) {
467 		dev->caps.log_num_macs  = MLX4_MIN_LOG_NUM_MAC;
468 		dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
469 	} else {
470 		dev->caps.log_num_macs  = log_num_mac;
471 		dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
472 	}
473 
474 	for (i = 1; i <= dev->caps.num_ports; ++i) {
475 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
476 		if (dev->caps.supported_type[i]) {
477 			/* if only ETH is supported - assign ETH */
478 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
479 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
480 			/* if only IB is supported, assign IB */
481 			else if (dev->caps.supported_type[i] ==
482 				 MLX4_PORT_TYPE_IB)
483 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
484 			else {
485 				/* if IB and ETH are supported, we set the port
486 				 * type according to user selection of port type;
487 				 * if user selected none, take the FW hint */
488 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
489 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
490 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
491 				else
492 					dev->caps.port_type[i] = port_type_array[i - 1];
493 			}
494 		}
495 		/*
496 		 * Link sensing is allowed on the port if 3 conditions are true:
497 		 * 1. Both protocols are supported on the port.
498 		 * 2. Different types are supported on the port
499 		 * 3. FW declared that it supports link sensing
500 		 */
501 		mlx4_priv(dev)->sense.sense_allowed[i] =
502 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
503 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
504 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
505 
506 		/*
507 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
508 		 * and perform sense_port FW command to try and set the correct
509 		 * port type from beginning
510 		 */
511 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
512 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
513 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
514 			mlx4_SENSE_PORT(dev, i, &sensed_port);
515 			if (sensed_port != MLX4_PORT_TYPE_NONE)
516 				dev->caps.port_type[i] = sensed_port;
517 		} else {
518 			dev->caps.possible_type[i] = dev->caps.port_type[i];
519 		}
520 
521 		if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
522 			dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
523 			mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
524 				  i, 1 << dev->caps.log_num_macs);
525 		}
526 		if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
527 			dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
528 			mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
529 				  i, 1 << dev->caps.log_num_vlans);
530 		}
531 	}
532 
533 	if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
534 	    (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
535 	    (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
536 		mlx4_warn(dev,
537 			  "Granular QoS per VF not supported with IB/Eth configuration\n");
538 		dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
539 	}
540 
541 	dev->caps.max_counters = dev_cap->max_counters;
542 
543 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
544 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
545 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
546 		(1 << dev->caps.log_num_macs) *
547 		(1 << dev->caps.log_num_vlans) *
548 		dev->caps.num_ports;
549 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
550 
551 	if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
552 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
553 		dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
554 	else
555 		dev->caps.dmfs_high_rate_qpn_base =
556 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
557 
558 	if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
559 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
560 		dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
561 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
562 		dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
563 	} else {
564 		dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
565 		dev->caps.dmfs_high_rate_qpn_base =
566 			dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
567 		dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
568 	}
569 
570 	dev->caps.rl_caps = dev_cap->rl_caps;
571 
572 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
573 		dev->caps.dmfs_high_rate_qpn_range;
574 
575 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
576 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
577 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
578 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
579 
580 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
581 
582 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
583 		if (dev_cap->flags &
584 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
585 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
586 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
587 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
588 		}
589 
590 		if (dev_cap->flags2 &
591 		    (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
592 		     MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
593 			mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
594 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
595 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
596 		}
597 	}
598 
599 	if ((dev->caps.flags &
600 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
601 	    mlx4_is_master(dev))
602 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
603 
604 	if (!mlx4_is_slave(dev)) {
605 		mlx4_enable_cqe_eqe_stride(dev);
606 		dev->caps.alloc_res_qp_mask =
607 			(dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
608 			MLX4_RESERVE_A0_QP;
609 
610 		if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
611 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
612 			mlx4_warn(dev, "Old device ETS support detected\n");
613 			mlx4_warn(dev, "Consider upgrading device FW.\n");
614 			dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
615 		}
616 
617 	} else {
618 		dev->caps.alloc_res_qp_mask = 0;
619 	}
620 
621 	mlx4_enable_ignore_fcs(dev);
622 
623 	return 0;
624 }
625 
626 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
627 				       enum pci_bus_speed *speed,
628 				       enum pcie_link_width *width)
629 {
630 	u32 lnkcap1, lnkcap2;
631 	int err1, err2;
632 
633 #define  PCIE_MLW_CAP_SHIFT 4	/* start of MLW mask in link capabilities */
634 
635 	*speed = PCI_SPEED_UNKNOWN;
636 	*width = PCIE_LNK_WIDTH_UNKNOWN;
637 
638 	err1 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP,
639 					  &lnkcap1);
640 	err2 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP2,
641 					  &lnkcap2);
642 	if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
643 		if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
644 			*speed = PCIE_SPEED_8_0GT;
645 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
646 			*speed = PCIE_SPEED_5_0GT;
647 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
648 			*speed = PCIE_SPEED_2_5GT;
649 	}
650 	if (!err1) {
651 		*width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
652 		if (!lnkcap2) { /* pre-r3.0 */
653 			if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
654 				*speed = PCIE_SPEED_5_0GT;
655 			else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
656 				*speed = PCIE_SPEED_2_5GT;
657 		}
658 	}
659 
660 	if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
661 		return err1 ? err1 :
662 			err2 ? err2 : -EINVAL;
663 	}
664 	return 0;
665 }
666 
667 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
668 {
669 	enum pcie_link_width width, width_cap;
670 	enum pci_bus_speed speed, speed_cap;
671 	int err;
672 
673 #define PCIE_SPEED_STR(speed) \
674 	(speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
675 	 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
676 	 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
677 	 "Unknown")
678 
679 	err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
680 	if (err) {
681 		mlx4_warn(dev,
682 			  "Unable to determine PCIe device BW capabilities\n");
683 		return;
684 	}
685 
686 	err = pcie_get_minimum_link(dev->persist->pdev, &speed, &width);
687 	if (err || speed == PCI_SPEED_UNKNOWN ||
688 	    width == PCIE_LNK_WIDTH_UNKNOWN) {
689 		mlx4_warn(dev,
690 			  "Unable to determine PCI device chain minimum BW\n");
691 		return;
692 	}
693 
694 	if (width != width_cap || speed != speed_cap)
695 		mlx4_warn(dev,
696 			  "PCIe BW is different than device's capability\n");
697 
698 	mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
699 		  PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
700 	mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
701 		  width, width_cap);
702 	return;
703 }
704 
705 /*The function checks if there are live vf, return the num of them*/
706 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
707 {
708 	struct mlx4_priv *priv = mlx4_priv(dev);
709 	struct mlx4_slave_state *s_state;
710 	int i;
711 	int ret = 0;
712 
713 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
714 		s_state = &priv->mfunc.master.slave_state[i];
715 		if (s_state->active && s_state->last_cmd !=
716 		    MLX4_COMM_CMD_RESET) {
717 			mlx4_warn(dev, "%s: slave: %d is still active\n",
718 				  __func__, i);
719 			ret++;
720 		}
721 	}
722 	return ret;
723 }
724 
725 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
726 {
727 	u32 qk = MLX4_RESERVED_QKEY_BASE;
728 
729 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
730 	    qpn < dev->phys_caps.base_proxy_sqpn)
731 		return -EINVAL;
732 
733 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
734 		/* tunnel qp */
735 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
736 	else
737 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
738 	*qkey = qk;
739 	return 0;
740 }
741 EXPORT_SYMBOL(mlx4_get_parav_qkey);
742 
743 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
744 {
745 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
746 
747 	if (!mlx4_is_master(dev))
748 		return;
749 
750 	priv->virt2phys_pkey[slave][port - 1][i] = val;
751 }
752 EXPORT_SYMBOL(mlx4_sync_pkey_table);
753 
754 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
755 {
756 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
757 
758 	if (!mlx4_is_master(dev))
759 		return;
760 
761 	priv->slave_node_guids[slave] = guid;
762 }
763 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
764 
765 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
766 {
767 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
768 
769 	if (!mlx4_is_master(dev))
770 		return 0;
771 
772 	return priv->slave_node_guids[slave];
773 }
774 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
775 
776 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
777 {
778 	struct mlx4_priv *priv = mlx4_priv(dev);
779 	struct mlx4_slave_state *s_slave;
780 
781 	if (!mlx4_is_master(dev))
782 		return 0;
783 
784 	s_slave = &priv->mfunc.master.slave_state[slave];
785 	return !!s_slave->active;
786 }
787 EXPORT_SYMBOL(mlx4_is_slave_active);
788 
789 void mlx4_handle_eth_header_mcast_prio(struct mlx4_net_trans_rule_hw_ctrl *ctrl,
790 				       struct _rule_hw *eth_header)
791 {
792 	if (is_multicast_ether_addr(eth_header->eth.dst_mac) ||
793 	    is_broadcast_ether_addr(eth_header->eth.dst_mac)) {
794 		struct mlx4_net_trans_rule_hw_eth *eth =
795 			(struct mlx4_net_trans_rule_hw_eth *)eth_header;
796 		struct _rule_hw *next_rule = (struct _rule_hw *)(eth + 1);
797 		bool last_rule = next_rule->size == 0 && next_rule->id == 0 &&
798 			next_rule->rsvd == 0;
799 
800 		if (last_rule)
801 			ctrl->prio = cpu_to_be16(MLX4_DOMAIN_NIC);
802 	}
803 }
804 EXPORT_SYMBOL(mlx4_handle_eth_header_mcast_prio);
805 
806 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
807 				       struct mlx4_dev_cap *dev_cap,
808 				       struct mlx4_init_hca_param *hca_param)
809 {
810 	dev->caps.steering_mode = hca_param->steering_mode;
811 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
812 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
813 		dev->caps.fs_log_max_ucast_qp_range_size =
814 			dev_cap->fs_log_max_ucast_qp_range_size;
815 	} else
816 		dev->caps.num_qp_per_mgm =
817 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
818 
819 	mlx4_dbg(dev, "Steering mode is: %s\n",
820 		 mlx4_steering_mode_str(dev->caps.steering_mode));
821 }
822 
823 static void mlx4_slave_destroy_special_qp_cap(struct mlx4_dev *dev)
824 {
825 	kfree(dev->caps.spec_qps);
826 	dev->caps.spec_qps = NULL;
827 }
828 
829 static int mlx4_slave_special_qp_cap(struct mlx4_dev *dev)
830 {
831 	struct mlx4_func_cap *func_cap = NULL;
832 	struct mlx4_caps *caps = &dev->caps;
833 	int i, err = 0;
834 
835 	func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
836 	caps->spec_qps = kcalloc(caps->num_ports, sizeof(*caps->spec_qps), GFP_KERNEL);
837 
838 	if (!func_cap || !caps->spec_qps) {
839 		mlx4_err(dev, "Failed to allocate memory for special qps cap\n");
840 		err = -ENOMEM;
841 		goto err_mem;
842 	}
843 
844 	for (i = 1; i <= caps->num_ports; ++i) {
845 		err = mlx4_QUERY_FUNC_CAP(dev, i, func_cap);
846 		if (err) {
847 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
848 				 i, err);
849 			goto err_mem;
850 		}
851 		caps->spec_qps[i - 1] = func_cap->spec_qps;
852 		caps->port_mask[i] = caps->port_type[i];
853 		caps->phys_port_id[i] = func_cap->phys_port_id;
854 		err = mlx4_get_slave_pkey_gid_tbl_len(dev, i,
855 						      &caps->gid_table_len[i],
856 						      &caps->pkey_table_len[i]);
857 		if (err) {
858 			mlx4_err(dev, "QUERY_PORT command failed for port %d, aborting (%d)\n",
859 				 i, err);
860 			goto err_mem;
861 		}
862 	}
863 
864 err_mem:
865 	if (err)
866 		mlx4_slave_destroy_special_qp_cap(dev);
867 	kfree(func_cap);
868 	return err;
869 }
870 
871 static int mlx4_slave_cap(struct mlx4_dev *dev)
872 {
873 	int			   err;
874 	u32			   page_size;
875 	struct mlx4_dev_cap	   *dev_cap = NULL;
876 	struct mlx4_func_cap	   *func_cap = NULL;
877 	struct mlx4_init_hca_param *hca_param = NULL;
878 
879 	hca_param = kzalloc(sizeof(*hca_param), GFP_KERNEL);
880 	func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
881 	dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
882 	if (!hca_param || !func_cap || !dev_cap) {
883 		mlx4_err(dev, "Failed to allocate memory for slave_cap\n");
884 		err = -ENOMEM;
885 		goto free_mem;
886 	}
887 
888 	err = mlx4_QUERY_HCA(dev, hca_param);
889 	if (err) {
890 		mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
891 		goto free_mem;
892 	}
893 
894 	/* fail if the hca has an unknown global capability
895 	 * at this time global_caps should be always zeroed
896 	 */
897 	if (hca_param->global_caps) {
898 		mlx4_err(dev, "Unknown hca global capabilities\n");
899 		err = -EINVAL;
900 		goto free_mem;
901 	}
902 
903 	dev->caps.hca_core_clock = hca_param->hca_core_clock;
904 
905 	dev->caps.max_qp_dest_rdma = 1 << hca_param->log_rd_per_qp;
906 	err = mlx4_dev_cap(dev, dev_cap);
907 	if (err) {
908 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
909 		goto free_mem;
910 	}
911 
912 	err = mlx4_QUERY_FW(dev);
913 	if (err)
914 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
915 
916 	page_size = ~dev->caps.page_size_cap + 1;
917 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
918 	if (page_size > PAGE_SIZE) {
919 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
920 			 page_size, PAGE_SIZE);
921 		err = -ENODEV;
922 		goto free_mem;
923 	}
924 
925 	/* Set uar_page_shift for VF */
926 	dev->uar_page_shift = hca_param->uar_page_sz + 12;
927 
928 	/* Make sure the master uar page size is valid */
929 	if (dev->uar_page_shift > PAGE_SHIFT) {
930 		mlx4_err(dev,
931 			 "Invalid configuration: uar page size is larger than system page size\n");
932 		err = -ENODEV;
933 		goto free_mem;
934 	}
935 
936 	/* Set reserved_uars based on the uar_page_shift */
937 	mlx4_set_num_reserved_uars(dev, dev_cap);
938 
939 	/* Although uar page size in FW differs from system page size,
940 	 * upper software layers (mlx4_ib, mlx4_en and part of mlx4_core)
941 	 * still works with assumption that uar page size == system page size
942 	 */
943 	dev->caps.uar_page_size = PAGE_SIZE;
944 
945 	err = mlx4_QUERY_FUNC_CAP(dev, 0, func_cap);
946 	if (err) {
947 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
948 			 err);
949 		goto free_mem;
950 	}
951 
952 	if ((func_cap->pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
953 	    PF_CONTEXT_BEHAVIOUR_MASK) {
954 		mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
955 			 func_cap->pf_context_behaviour,
956 			 PF_CONTEXT_BEHAVIOUR_MASK);
957 		err = -EINVAL;
958 		goto free_mem;
959 	}
960 
961 	dev->caps.num_ports		= func_cap->num_ports;
962 	dev->quotas.qp			= func_cap->qp_quota;
963 	dev->quotas.srq			= func_cap->srq_quota;
964 	dev->quotas.cq			= func_cap->cq_quota;
965 	dev->quotas.mpt			= func_cap->mpt_quota;
966 	dev->quotas.mtt			= func_cap->mtt_quota;
967 	dev->caps.num_qps		= 1 << hca_param->log_num_qps;
968 	dev->caps.num_srqs		= 1 << hca_param->log_num_srqs;
969 	dev->caps.num_cqs		= 1 << hca_param->log_num_cqs;
970 	dev->caps.num_mpts		= 1 << hca_param->log_mpt_sz;
971 	dev->caps.num_eqs		= func_cap->max_eq;
972 	dev->caps.reserved_eqs		= func_cap->reserved_eq;
973 	dev->caps.reserved_lkey		= func_cap->reserved_lkey;
974 	dev->caps.num_pds               = MLX4_NUM_PDS;
975 	dev->caps.num_mgms              = 0;
976 	dev->caps.num_amgms             = 0;
977 
978 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
979 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
980 			 dev->caps.num_ports, MLX4_MAX_PORTS);
981 		err = -ENODEV;
982 		goto free_mem;
983 	}
984 
985 	mlx4_replace_zero_macs(dev);
986 
987 	err = mlx4_slave_special_qp_cap(dev);
988 	if (err) {
989 		mlx4_err(dev, "Set special QP caps failed. aborting\n");
990 		goto free_mem;
991 	}
992 
993 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
994 				       dev->caps.reserved_uars) >
995 				       pci_resource_len(dev->persist->pdev,
996 							2)) {
997 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
998 			 dev->caps.uar_page_size * dev->caps.num_uars,
999 			 (unsigned long long)
1000 			 pci_resource_len(dev->persist->pdev, 2));
1001 		err = -ENOMEM;
1002 		goto err_mem;
1003 	}
1004 
1005 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
1006 		dev->caps.eqe_size   = 64;
1007 		dev->caps.eqe_factor = 1;
1008 	} else {
1009 		dev->caps.eqe_size   = 32;
1010 		dev->caps.eqe_factor = 0;
1011 	}
1012 
1013 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
1014 		dev->caps.cqe_size   = 64;
1015 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1016 	} else {
1017 		dev->caps.cqe_size   = 32;
1018 	}
1019 
1020 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
1021 		dev->caps.eqe_size = hca_param->eqe_size;
1022 		dev->caps.eqe_factor = 0;
1023 	}
1024 
1025 	if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
1026 		dev->caps.cqe_size = hca_param->cqe_size;
1027 		/* User still need to know when CQE > 32B */
1028 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1029 	}
1030 
1031 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1032 	mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
1033 
1034 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_USER_MAC_EN;
1035 	mlx4_dbg(dev, "User MAC FW update is not supported in slave mode\n");
1036 
1037 	slave_adjust_steering_mode(dev, dev_cap, hca_param);
1038 	mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
1039 		 hca_param->rss_ip_frags ? "on" : "off");
1040 
1041 	if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
1042 	    dev->caps.bf_reg_size)
1043 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
1044 
1045 	if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
1046 		dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
1047 
1048 err_mem:
1049 	if (err)
1050 		mlx4_slave_destroy_special_qp_cap(dev);
1051 free_mem:
1052 	kfree(hca_param);
1053 	kfree(func_cap);
1054 	kfree(dev_cap);
1055 	return err;
1056 }
1057 
1058 static void mlx4_request_modules(struct mlx4_dev *dev)
1059 {
1060 	int port;
1061 	int has_ib_port = false;
1062 	int has_eth_port = false;
1063 #define EN_DRV_NAME	"mlx4_en"
1064 #define IB_DRV_NAME	"mlx4_ib"
1065 
1066 	for (port = 1; port <= dev->caps.num_ports; port++) {
1067 		if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
1068 			has_ib_port = true;
1069 		else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
1070 			has_eth_port = true;
1071 	}
1072 
1073 	if (has_eth_port)
1074 		request_module_nowait(EN_DRV_NAME);
1075 	if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
1076 		request_module_nowait(IB_DRV_NAME);
1077 }
1078 
1079 /*
1080  * Change the port configuration of the device.
1081  * Every user of this function must hold the port mutex.
1082  */
1083 int mlx4_change_port_types(struct mlx4_dev *dev,
1084 			   enum mlx4_port_type *port_types)
1085 {
1086 	int err = 0;
1087 	int change = 0;
1088 	int port;
1089 
1090 	for (port = 0; port <  dev->caps.num_ports; port++) {
1091 		/* Change the port type only if the new type is different
1092 		 * from the current, and not set to Auto */
1093 		if (port_types[port] != dev->caps.port_type[port + 1])
1094 			change = 1;
1095 	}
1096 	if (change) {
1097 		mlx4_unregister_device(dev);
1098 		for (port = 1; port <= dev->caps.num_ports; port++) {
1099 			mlx4_CLOSE_PORT(dev, port);
1100 			dev->caps.port_type[port] = port_types[port - 1];
1101 			err = mlx4_SET_PORT(dev, port, -1);
1102 			if (err) {
1103 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1104 					 port);
1105 				goto out;
1106 			}
1107 		}
1108 		mlx4_set_port_mask(dev);
1109 		err = mlx4_register_device(dev);
1110 		if (err) {
1111 			mlx4_err(dev, "Failed to register device\n");
1112 			goto out;
1113 		}
1114 		mlx4_request_modules(dev);
1115 	}
1116 
1117 out:
1118 	return err;
1119 }
1120 
1121 static ssize_t show_port_type(struct device *dev,
1122 			      struct device_attribute *attr,
1123 			      char *buf)
1124 {
1125 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1126 						   port_attr);
1127 	struct mlx4_dev *mdev = info->dev;
1128 	char type[8];
1129 
1130 	sprintf(type, "%s",
1131 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1132 		"ib" : "eth");
1133 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1134 		sprintf(buf, "auto (%s)\n", type);
1135 	else
1136 		sprintf(buf, "%s\n", type);
1137 
1138 	return strlen(buf);
1139 }
1140 
1141 static int __set_port_type(struct mlx4_port_info *info,
1142 			   enum mlx4_port_type port_type)
1143 {
1144 	struct mlx4_dev *mdev = info->dev;
1145 	struct mlx4_priv *priv = mlx4_priv(mdev);
1146 	enum mlx4_port_type types[MLX4_MAX_PORTS];
1147 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1148 	int i;
1149 	int err = 0;
1150 
1151 	if ((port_type & mdev->caps.supported_type[info->port]) != port_type) {
1152 		mlx4_err(mdev,
1153 			 "Requested port type for port %d is not supported on this HCA\n",
1154 			 info->port);
1155 		err = -EINVAL;
1156 		goto err_sup;
1157 	}
1158 
1159 	mlx4_stop_sense(mdev);
1160 	mutex_lock(&priv->port_mutex);
1161 	info->tmp_type = port_type;
1162 
1163 	/* Possible type is always the one that was delivered */
1164 	mdev->caps.possible_type[info->port] = info->tmp_type;
1165 
1166 	for (i = 0; i < mdev->caps.num_ports; i++) {
1167 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1168 					mdev->caps.possible_type[i+1];
1169 		if (types[i] == MLX4_PORT_TYPE_AUTO)
1170 			types[i] = mdev->caps.port_type[i+1];
1171 	}
1172 
1173 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1174 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1175 		for (i = 1; i <= mdev->caps.num_ports; i++) {
1176 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1177 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1178 				err = -EINVAL;
1179 			}
1180 		}
1181 	}
1182 	if (err) {
1183 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1184 		goto out;
1185 	}
1186 
1187 	mlx4_do_sense_ports(mdev, new_types, types);
1188 
1189 	err = mlx4_check_port_params(mdev, new_types);
1190 	if (err)
1191 		goto out;
1192 
1193 	/* We are about to apply the changes after the configuration
1194 	 * was verified, no need to remember the temporary types
1195 	 * any more */
1196 	for (i = 0; i < mdev->caps.num_ports; i++)
1197 		priv->port[i + 1].tmp_type = 0;
1198 
1199 	err = mlx4_change_port_types(mdev, new_types);
1200 
1201 out:
1202 	mlx4_start_sense(mdev);
1203 	mutex_unlock(&priv->port_mutex);
1204 err_sup:
1205 	return err;
1206 }
1207 
1208 static ssize_t set_port_type(struct device *dev,
1209 			     struct device_attribute *attr,
1210 			     const char *buf, size_t count)
1211 {
1212 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1213 						   port_attr);
1214 	struct mlx4_dev *mdev = info->dev;
1215 	enum mlx4_port_type port_type;
1216 	static DEFINE_MUTEX(set_port_type_mutex);
1217 	int err;
1218 
1219 	mutex_lock(&set_port_type_mutex);
1220 
1221 	if (!strcmp(buf, "ib\n")) {
1222 		port_type = MLX4_PORT_TYPE_IB;
1223 	} else if (!strcmp(buf, "eth\n")) {
1224 		port_type = MLX4_PORT_TYPE_ETH;
1225 	} else if (!strcmp(buf, "auto\n")) {
1226 		port_type = MLX4_PORT_TYPE_AUTO;
1227 	} else {
1228 		mlx4_err(mdev, "%s is not supported port type\n", buf);
1229 		err = -EINVAL;
1230 		goto err_out;
1231 	}
1232 
1233 	err = __set_port_type(info, port_type);
1234 
1235 err_out:
1236 	mutex_unlock(&set_port_type_mutex);
1237 
1238 	return err ? err : count;
1239 }
1240 
1241 enum ibta_mtu {
1242 	IB_MTU_256  = 1,
1243 	IB_MTU_512  = 2,
1244 	IB_MTU_1024 = 3,
1245 	IB_MTU_2048 = 4,
1246 	IB_MTU_4096 = 5
1247 };
1248 
1249 static inline int int_to_ibta_mtu(int mtu)
1250 {
1251 	switch (mtu) {
1252 	case 256:  return IB_MTU_256;
1253 	case 512:  return IB_MTU_512;
1254 	case 1024: return IB_MTU_1024;
1255 	case 2048: return IB_MTU_2048;
1256 	case 4096: return IB_MTU_4096;
1257 	default: return -1;
1258 	}
1259 }
1260 
1261 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1262 {
1263 	switch (mtu) {
1264 	case IB_MTU_256:  return  256;
1265 	case IB_MTU_512:  return  512;
1266 	case IB_MTU_1024: return 1024;
1267 	case IB_MTU_2048: return 2048;
1268 	case IB_MTU_4096: return 4096;
1269 	default: return -1;
1270 	}
1271 }
1272 
1273 static ssize_t show_port_ib_mtu(struct device *dev,
1274 			     struct device_attribute *attr,
1275 			     char *buf)
1276 {
1277 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1278 						   port_mtu_attr);
1279 	struct mlx4_dev *mdev = info->dev;
1280 
1281 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1282 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1283 
1284 	sprintf(buf, "%d\n",
1285 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1286 	return strlen(buf);
1287 }
1288 
1289 static ssize_t set_port_ib_mtu(struct device *dev,
1290 			     struct device_attribute *attr,
1291 			     const char *buf, size_t count)
1292 {
1293 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1294 						   port_mtu_attr);
1295 	struct mlx4_dev *mdev = info->dev;
1296 	struct mlx4_priv *priv = mlx4_priv(mdev);
1297 	int err, port, mtu, ibta_mtu = -1;
1298 
1299 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1300 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1301 		return -EINVAL;
1302 	}
1303 
1304 	err = kstrtoint(buf, 0, &mtu);
1305 	if (!err)
1306 		ibta_mtu = int_to_ibta_mtu(mtu);
1307 
1308 	if (err || ibta_mtu < 0) {
1309 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1310 		return -EINVAL;
1311 	}
1312 
1313 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1314 
1315 	mlx4_stop_sense(mdev);
1316 	mutex_lock(&priv->port_mutex);
1317 	mlx4_unregister_device(mdev);
1318 	for (port = 1; port <= mdev->caps.num_ports; port++) {
1319 		mlx4_CLOSE_PORT(mdev, port);
1320 		err = mlx4_SET_PORT(mdev, port, -1);
1321 		if (err) {
1322 			mlx4_err(mdev, "Failed to set port %d, aborting\n",
1323 				 port);
1324 			goto err_set_port;
1325 		}
1326 	}
1327 	err = mlx4_register_device(mdev);
1328 err_set_port:
1329 	mutex_unlock(&priv->port_mutex);
1330 	mlx4_start_sense(mdev);
1331 	return err ? err : count;
1332 }
1333 
1334 /* bond for multi-function device */
1335 #define MAX_MF_BOND_ALLOWED_SLAVES 63
1336 static int mlx4_mf_bond(struct mlx4_dev *dev)
1337 {
1338 	int err = 0;
1339 	int nvfs;
1340 	struct mlx4_slaves_pport slaves_port1;
1341 	struct mlx4_slaves_pport slaves_port2;
1342 	DECLARE_BITMAP(slaves_port_1_2, MLX4_MFUNC_MAX);
1343 
1344 	slaves_port1 = mlx4_phys_to_slaves_pport(dev, 1);
1345 	slaves_port2 = mlx4_phys_to_slaves_pport(dev, 2);
1346 	bitmap_and(slaves_port_1_2,
1347 		   slaves_port1.slaves, slaves_port2.slaves,
1348 		   dev->persist->num_vfs + 1);
1349 
1350 	/* only single port vfs are allowed */
1351 	if (bitmap_weight(slaves_port_1_2, dev->persist->num_vfs + 1) > 1) {
1352 		mlx4_warn(dev, "HA mode unsupported for dual ported VFs\n");
1353 		return -EINVAL;
1354 	}
1355 
1356 	/* number of virtual functions is number of total functions minus one
1357 	 * physical function for each port.
1358 	 */
1359 	nvfs = bitmap_weight(slaves_port1.slaves, dev->persist->num_vfs + 1) +
1360 		bitmap_weight(slaves_port2.slaves, dev->persist->num_vfs + 1) - 2;
1361 
1362 	/* limit on maximum allowed VFs */
1363 	if (nvfs > MAX_MF_BOND_ALLOWED_SLAVES) {
1364 		mlx4_warn(dev, "HA mode is not supported for %d VFs (max %d are allowed)\n",
1365 			  nvfs, MAX_MF_BOND_ALLOWED_SLAVES);
1366 		return -EINVAL;
1367 	}
1368 
1369 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) {
1370 		mlx4_warn(dev, "HA mode unsupported for NON DMFS steering\n");
1371 		return -EINVAL;
1372 	}
1373 
1374 	err = mlx4_bond_mac_table(dev);
1375 	if (err)
1376 		return err;
1377 	err = mlx4_bond_vlan_table(dev);
1378 	if (err)
1379 		goto err1;
1380 	err = mlx4_bond_fs_rules(dev);
1381 	if (err)
1382 		goto err2;
1383 
1384 	return 0;
1385 err2:
1386 	(void)mlx4_unbond_vlan_table(dev);
1387 err1:
1388 	(void)mlx4_unbond_mac_table(dev);
1389 	return err;
1390 }
1391 
1392 static int mlx4_mf_unbond(struct mlx4_dev *dev)
1393 {
1394 	int ret, ret1;
1395 
1396 	ret = mlx4_unbond_fs_rules(dev);
1397 	if (ret)
1398 		mlx4_warn(dev, "multifunction unbond for flow rules failedi (%d)\n", ret);
1399 	ret1 = mlx4_unbond_mac_table(dev);
1400 	if (ret1) {
1401 		mlx4_warn(dev, "multifunction unbond for MAC table failed (%d)\n", ret1);
1402 		ret = ret1;
1403 	}
1404 	ret1 = mlx4_unbond_vlan_table(dev);
1405 	if (ret1) {
1406 		mlx4_warn(dev, "multifunction unbond for VLAN table failed (%d)\n", ret1);
1407 		ret = ret1;
1408 	}
1409 	return ret;
1410 }
1411 
1412 int mlx4_bond(struct mlx4_dev *dev)
1413 {
1414 	int ret = 0;
1415 	struct mlx4_priv *priv = mlx4_priv(dev);
1416 
1417 	mutex_lock(&priv->bond_mutex);
1418 
1419 	if (!mlx4_is_bonded(dev)) {
1420 		ret = mlx4_do_bond(dev, true);
1421 		if (ret)
1422 			mlx4_err(dev, "Failed to bond device: %d\n", ret);
1423 		if (!ret && mlx4_is_master(dev)) {
1424 			ret = mlx4_mf_bond(dev);
1425 			if (ret) {
1426 				mlx4_err(dev, "bond for multifunction failed\n");
1427 				mlx4_do_bond(dev, false);
1428 			}
1429 		}
1430 	}
1431 
1432 	mutex_unlock(&priv->bond_mutex);
1433 	if (!ret)
1434 		mlx4_dbg(dev, "Device is bonded\n");
1435 
1436 	return ret;
1437 }
1438 EXPORT_SYMBOL_GPL(mlx4_bond);
1439 
1440 int mlx4_unbond(struct mlx4_dev *dev)
1441 {
1442 	int ret = 0;
1443 	struct mlx4_priv *priv = mlx4_priv(dev);
1444 
1445 	mutex_lock(&priv->bond_mutex);
1446 
1447 	if (mlx4_is_bonded(dev)) {
1448 		int ret2 = 0;
1449 
1450 		ret = mlx4_do_bond(dev, false);
1451 		if (ret)
1452 			mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1453 		if (mlx4_is_master(dev))
1454 			ret2 = mlx4_mf_unbond(dev);
1455 		if (ret2) {
1456 			mlx4_warn(dev, "Failed to unbond device for multifunction (%d)\n", ret2);
1457 			ret = ret2;
1458 		}
1459 	}
1460 
1461 	mutex_unlock(&priv->bond_mutex);
1462 	if (!ret)
1463 		mlx4_dbg(dev, "Device is unbonded\n");
1464 
1465 	return ret;
1466 }
1467 EXPORT_SYMBOL_GPL(mlx4_unbond);
1468 
1469 
1470 int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1471 {
1472 	u8 port1 = v2p->port1;
1473 	u8 port2 = v2p->port2;
1474 	struct mlx4_priv *priv = mlx4_priv(dev);
1475 	int err;
1476 
1477 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1478 		return -EOPNOTSUPP;
1479 
1480 	mutex_lock(&priv->bond_mutex);
1481 
1482 	/* zero means keep current mapping for this port */
1483 	if (port1 == 0)
1484 		port1 = priv->v2p.port1;
1485 	if (port2 == 0)
1486 		port2 = priv->v2p.port2;
1487 
1488 	if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1489 	    (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1490 	    (port1 == 2 && port2 == 1)) {
1491 		/* besides boundary checks cross mapping makes
1492 		 * no sense and therefore not allowed */
1493 		err = -EINVAL;
1494 	} else if ((port1 == priv->v2p.port1) &&
1495 		 (port2 == priv->v2p.port2)) {
1496 		err = 0;
1497 	} else {
1498 		err = mlx4_virt2phy_port_map(dev, port1, port2);
1499 		if (!err) {
1500 			mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1501 				 port1, port2);
1502 			priv->v2p.port1 = port1;
1503 			priv->v2p.port2 = port2;
1504 		} else {
1505 			mlx4_err(dev, "Failed to change port mape: %d\n", err);
1506 		}
1507 	}
1508 
1509 	mutex_unlock(&priv->bond_mutex);
1510 	return err;
1511 }
1512 EXPORT_SYMBOL_GPL(mlx4_port_map_set);
1513 
1514 static int mlx4_load_fw(struct mlx4_dev *dev)
1515 {
1516 	struct mlx4_priv *priv = mlx4_priv(dev);
1517 	int err;
1518 
1519 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1520 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
1521 	if (!priv->fw.fw_icm) {
1522 		mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1523 		return -ENOMEM;
1524 	}
1525 
1526 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1527 	if (err) {
1528 		mlx4_err(dev, "MAP_FA command failed, aborting\n");
1529 		goto err_free;
1530 	}
1531 
1532 	err = mlx4_RUN_FW(dev);
1533 	if (err) {
1534 		mlx4_err(dev, "RUN_FW command failed, aborting\n");
1535 		goto err_unmap_fa;
1536 	}
1537 
1538 	return 0;
1539 
1540 err_unmap_fa:
1541 	mlx4_UNMAP_FA(dev);
1542 
1543 err_free:
1544 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1545 	return err;
1546 }
1547 
1548 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1549 				int cmpt_entry_sz)
1550 {
1551 	struct mlx4_priv *priv = mlx4_priv(dev);
1552 	int err;
1553 	int num_eqs;
1554 
1555 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1556 				  cmpt_base +
1557 				  ((u64) (MLX4_CMPT_TYPE_QP *
1558 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1559 				  cmpt_entry_sz, dev->caps.num_qps,
1560 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1561 				  0, 0);
1562 	if (err)
1563 		goto err;
1564 
1565 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1566 				  cmpt_base +
1567 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1568 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1569 				  cmpt_entry_sz, dev->caps.num_srqs,
1570 				  dev->caps.reserved_srqs, 0, 0);
1571 	if (err)
1572 		goto err_qp;
1573 
1574 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1575 				  cmpt_base +
1576 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1577 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1578 				  cmpt_entry_sz, dev->caps.num_cqs,
1579 				  dev->caps.reserved_cqs, 0, 0);
1580 	if (err)
1581 		goto err_srq;
1582 
1583 	num_eqs = dev->phys_caps.num_phys_eqs;
1584 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1585 				  cmpt_base +
1586 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1587 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1588 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1589 	if (err)
1590 		goto err_cq;
1591 
1592 	return 0;
1593 
1594 err_cq:
1595 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1596 
1597 err_srq:
1598 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1599 
1600 err_qp:
1601 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1602 
1603 err:
1604 	return err;
1605 }
1606 
1607 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1608 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1609 {
1610 	struct mlx4_priv *priv = mlx4_priv(dev);
1611 	u64 aux_pages;
1612 	int num_eqs;
1613 	int err;
1614 
1615 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1616 	if (err) {
1617 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1618 		return err;
1619 	}
1620 
1621 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1622 		 (unsigned long long) icm_size >> 10,
1623 		 (unsigned long long) aux_pages << 2);
1624 
1625 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1626 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1627 	if (!priv->fw.aux_icm) {
1628 		mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1629 		return -ENOMEM;
1630 	}
1631 
1632 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1633 	if (err) {
1634 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1635 		goto err_free_aux;
1636 	}
1637 
1638 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1639 	if (err) {
1640 		mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1641 		goto err_unmap_aux;
1642 	}
1643 
1644 
1645 	num_eqs = dev->phys_caps.num_phys_eqs;
1646 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1647 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1648 				  num_eqs, num_eqs, 0, 0);
1649 	if (err) {
1650 		mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1651 		goto err_unmap_cmpt;
1652 	}
1653 
1654 	/*
1655 	 * Reserved MTT entries must be aligned up to a cacheline
1656 	 * boundary, since the FW will write to them, while the driver
1657 	 * writes to all other MTT entries. (The variable
1658 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1659 	 * size, not the raw entry size)
1660 	 */
1661 	dev->caps.reserved_mtts =
1662 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1663 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1664 
1665 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1666 				  init_hca->mtt_base,
1667 				  dev->caps.mtt_entry_sz,
1668 				  dev->caps.num_mtts,
1669 				  dev->caps.reserved_mtts, 1, 0);
1670 	if (err) {
1671 		mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1672 		goto err_unmap_eq;
1673 	}
1674 
1675 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1676 				  init_hca->dmpt_base,
1677 				  dev_cap->dmpt_entry_sz,
1678 				  dev->caps.num_mpts,
1679 				  dev->caps.reserved_mrws, 1, 1);
1680 	if (err) {
1681 		mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1682 		goto err_unmap_mtt;
1683 	}
1684 
1685 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1686 				  init_hca->qpc_base,
1687 				  dev_cap->qpc_entry_sz,
1688 				  dev->caps.num_qps,
1689 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1690 				  0, 0);
1691 	if (err) {
1692 		mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1693 		goto err_unmap_dmpt;
1694 	}
1695 
1696 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1697 				  init_hca->auxc_base,
1698 				  dev_cap->aux_entry_sz,
1699 				  dev->caps.num_qps,
1700 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1701 				  0, 0);
1702 	if (err) {
1703 		mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1704 		goto err_unmap_qp;
1705 	}
1706 
1707 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1708 				  init_hca->altc_base,
1709 				  dev_cap->altc_entry_sz,
1710 				  dev->caps.num_qps,
1711 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1712 				  0, 0);
1713 	if (err) {
1714 		mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1715 		goto err_unmap_auxc;
1716 	}
1717 
1718 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1719 				  init_hca->rdmarc_base,
1720 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1721 				  dev->caps.num_qps,
1722 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1723 				  0, 0);
1724 	if (err) {
1725 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1726 		goto err_unmap_altc;
1727 	}
1728 
1729 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1730 				  init_hca->cqc_base,
1731 				  dev_cap->cqc_entry_sz,
1732 				  dev->caps.num_cqs,
1733 				  dev->caps.reserved_cqs, 0, 0);
1734 	if (err) {
1735 		mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1736 		goto err_unmap_rdmarc;
1737 	}
1738 
1739 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1740 				  init_hca->srqc_base,
1741 				  dev_cap->srq_entry_sz,
1742 				  dev->caps.num_srqs,
1743 				  dev->caps.reserved_srqs, 0, 0);
1744 	if (err) {
1745 		mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1746 		goto err_unmap_cq;
1747 	}
1748 
1749 	/*
1750 	 * For flow steering device managed mode it is required to use
1751 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1752 	 * required, but for simplicity just map the whole multicast
1753 	 * group table now.  The table isn't very big and it's a lot
1754 	 * easier than trying to track ref counts.
1755 	 */
1756 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1757 				  init_hca->mc_base,
1758 				  mlx4_get_mgm_entry_size(dev),
1759 				  dev->caps.num_mgms + dev->caps.num_amgms,
1760 				  dev->caps.num_mgms + dev->caps.num_amgms,
1761 				  0, 0);
1762 	if (err) {
1763 		mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1764 		goto err_unmap_srq;
1765 	}
1766 
1767 	return 0;
1768 
1769 err_unmap_srq:
1770 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1771 
1772 err_unmap_cq:
1773 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1774 
1775 err_unmap_rdmarc:
1776 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1777 
1778 err_unmap_altc:
1779 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1780 
1781 err_unmap_auxc:
1782 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1783 
1784 err_unmap_qp:
1785 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1786 
1787 err_unmap_dmpt:
1788 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1789 
1790 err_unmap_mtt:
1791 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1792 
1793 err_unmap_eq:
1794 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1795 
1796 err_unmap_cmpt:
1797 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1798 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1799 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1800 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1801 
1802 err_unmap_aux:
1803 	mlx4_UNMAP_ICM_AUX(dev);
1804 
1805 err_free_aux:
1806 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1807 
1808 	return err;
1809 }
1810 
1811 static void mlx4_free_icms(struct mlx4_dev *dev)
1812 {
1813 	struct mlx4_priv *priv = mlx4_priv(dev);
1814 
1815 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1816 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1817 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1818 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1819 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1820 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1821 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1822 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1823 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1824 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1825 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1826 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1827 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1828 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1829 
1830 	mlx4_UNMAP_ICM_AUX(dev);
1831 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1832 }
1833 
1834 static void mlx4_slave_exit(struct mlx4_dev *dev)
1835 {
1836 	struct mlx4_priv *priv = mlx4_priv(dev);
1837 
1838 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1839 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1840 			  MLX4_COMM_TIME))
1841 		mlx4_warn(dev, "Failed to close slave function\n");
1842 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1843 }
1844 
1845 static int map_bf_area(struct mlx4_dev *dev)
1846 {
1847 	struct mlx4_priv *priv = mlx4_priv(dev);
1848 	resource_size_t bf_start;
1849 	resource_size_t bf_len;
1850 	int err = 0;
1851 
1852 	if (!dev->caps.bf_reg_size)
1853 		return -ENXIO;
1854 
1855 	bf_start = pci_resource_start(dev->persist->pdev, 2) +
1856 			(dev->caps.num_uars << PAGE_SHIFT);
1857 	bf_len = pci_resource_len(dev->persist->pdev, 2) -
1858 			(dev->caps.num_uars << PAGE_SHIFT);
1859 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1860 	if (!priv->bf_mapping)
1861 		err = -ENOMEM;
1862 
1863 	return err;
1864 }
1865 
1866 static void unmap_bf_area(struct mlx4_dev *dev)
1867 {
1868 	if (mlx4_priv(dev)->bf_mapping)
1869 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1870 }
1871 
1872 u64 mlx4_read_clock(struct mlx4_dev *dev)
1873 {
1874 	u32 clockhi, clocklo, clockhi1;
1875 	u64 cycles;
1876 	int i;
1877 	struct mlx4_priv *priv = mlx4_priv(dev);
1878 
1879 	for (i = 0; i < 10; i++) {
1880 		clockhi = swab32(readl(priv->clock_mapping));
1881 		clocklo = swab32(readl(priv->clock_mapping + 4));
1882 		clockhi1 = swab32(readl(priv->clock_mapping));
1883 		if (clockhi == clockhi1)
1884 			break;
1885 	}
1886 
1887 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1888 
1889 	return cycles;
1890 }
1891 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1892 
1893 
1894 static int map_internal_clock(struct mlx4_dev *dev)
1895 {
1896 	struct mlx4_priv *priv = mlx4_priv(dev);
1897 
1898 	priv->clock_mapping =
1899 		ioremap(pci_resource_start(dev->persist->pdev,
1900 					   priv->fw.clock_bar) +
1901 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1902 
1903 	if (!priv->clock_mapping)
1904 		return -ENOMEM;
1905 
1906 	return 0;
1907 }
1908 
1909 int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1910 				   struct mlx4_clock_params *params)
1911 {
1912 	struct mlx4_priv *priv = mlx4_priv(dev);
1913 
1914 	if (mlx4_is_slave(dev))
1915 		return -EOPNOTSUPP;
1916 
1917 	if (!params)
1918 		return -EINVAL;
1919 
1920 	params->bar = priv->fw.clock_bar;
1921 	params->offset = priv->fw.clock_offset;
1922 	params->size = MLX4_CLOCK_SIZE;
1923 
1924 	return 0;
1925 }
1926 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1927 
1928 static void unmap_internal_clock(struct mlx4_dev *dev)
1929 {
1930 	struct mlx4_priv *priv = mlx4_priv(dev);
1931 
1932 	if (priv->clock_mapping)
1933 		iounmap(priv->clock_mapping);
1934 }
1935 
1936 static void mlx4_close_hca(struct mlx4_dev *dev)
1937 {
1938 	unmap_internal_clock(dev);
1939 	unmap_bf_area(dev);
1940 	if (mlx4_is_slave(dev))
1941 		mlx4_slave_exit(dev);
1942 	else {
1943 		mlx4_CLOSE_HCA(dev, 0);
1944 		mlx4_free_icms(dev);
1945 	}
1946 }
1947 
1948 static void mlx4_close_fw(struct mlx4_dev *dev)
1949 {
1950 	if (!mlx4_is_slave(dev)) {
1951 		mlx4_UNMAP_FA(dev);
1952 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1953 	}
1954 }
1955 
1956 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1957 {
1958 #define COMM_CHAN_OFFLINE_OFFSET 0x09
1959 
1960 	u32 comm_flags;
1961 	u32 offline_bit;
1962 	unsigned long end;
1963 	struct mlx4_priv *priv = mlx4_priv(dev);
1964 
1965 	end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
1966 	while (time_before(jiffies, end)) {
1967 		comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
1968 					  MLX4_COMM_CHAN_FLAGS));
1969 		offline_bit = (comm_flags &
1970 			       (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
1971 		if (!offline_bit)
1972 			return 0;
1973 
1974 		/* If device removal has been requested,
1975 		 * do not continue retrying.
1976 		 */
1977 		if (dev->persist->interface_state &
1978 		    MLX4_INTERFACE_STATE_NOWAIT)
1979 			break;
1980 
1981 		/* There are cases as part of AER/Reset flow that PF needs
1982 		 * around 100 msec to load. We therefore sleep for 100 msec
1983 		 * to allow other tasks to make use of that CPU during this
1984 		 * time interval.
1985 		 */
1986 		msleep(100);
1987 	}
1988 	mlx4_err(dev, "Communication channel is offline.\n");
1989 	return -EIO;
1990 }
1991 
1992 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
1993 {
1994 #define COMM_CHAN_RST_OFFSET 0x1e
1995 
1996 	struct mlx4_priv *priv = mlx4_priv(dev);
1997 	u32 comm_rst;
1998 	u32 comm_caps;
1999 
2000 	comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
2001 				 MLX4_COMM_CHAN_CAPS));
2002 	comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
2003 
2004 	if (comm_rst)
2005 		dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
2006 }
2007 
2008 static int mlx4_init_slave(struct mlx4_dev *dev)
2009 {
2010 	struct mlx4_priv *priv = mlx4_priv(dev);
2011 	u64 dma = (u64) priv->mfunc.vhcr_dma;
2012 	int ret_from_reset = 0;
2013 	u32 slave_read;
2014 	u32 cmd_channel_ver;
2015 
2016 	if (atomic_read(&pf_loading)) {
2017 		mlx4_warn(dev, "PF is not ready - Deferring probe\n");
2018 		return -EPROBE_DEFER;
2019 	}
2020 
2021 	mutex_lock(&priv->cmd.slave_cmd_mutex);
2022 	priv->cmd.max_cmds = 1;
2023 	if (mlx4_comm_check_offline(dev)) {
2024 		mlx4_err(dev, "PF is not responsive, skipping initialization\n");
2025 		goto err_offline;
2026 	}
2027 
2028 	mlx4_reset_vf_support(dev);
2029 	mlx4_warn(dev, "Sending reset\n");
2030 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
2031 				       MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
2032 	/* if we are in the middle of flr the slave will try
2033 	 * NUM_OF_RESET_RETRIES times before leaving.*/
2034 	if (ret_from_reset) {
2035 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
2036 			mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
2037 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
2038 			return -EPROBE_DEFER;
2039 		} else
2040 			goto err;
2041 	}
2042 
2043 	/* check the driver version - the slave I/F revision
2044 	 * must match the master's */
2045 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
2046 	cmd_channel_ver = mlx4_comm_get_version();
2047 
2048 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
2049 		MLX4_COMM_GET_IF_REV(slave_read)) {
2050 		mlx4_err(dev, "slave driver version is not supported by the master\n");
2051 		goto err;
2052 	}
2053 
2054 	mlx4_warn(dev, "Sending vhcr0\n");
2055 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
2056 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2057 		goto err;
2058 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
2059 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2060 		goto err;
2061 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
2062 			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2063 		goto err;
2064 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
2065 			  MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2066 		goto err;
2067 
2068 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2069 	return 0;
2070 
2071 err:
2072 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
2073 err_offline:
2074 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
2075 	return -EIO;
2076 }
2077 
2078 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
2079 {
2080 	int i;
2081 
2082 	for (i = 1; i <= dev->caps.num_ports; i++) {
2083 		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
2084 			dev->caps.gid_table_len[i] =
2085 				mlx4_get_slave_num_gids(dev, 0, i);
2086 		else
2087 			dev->caps.gid_table_len[i] = 1;
2088 		dev->caps.pkey_table_len[i] =
2089 			dev->phys_caps.pkey_phys_table_len[i] - 1;
2090 	}
2091 }
2092 
2093 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
2094 {
2095 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
2096 
2097 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
2098 	      i++) {
2099 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
2100 			break;
2101 	}
2102 
2103 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
2104 }
2105 
2106 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
2107 {
2108 	switch (dmfs_high_steer_mode) {
2109 	case MLX4_STEERING_DMFS_A0_DEFAULT:
2110 		return "default performance";
2111 
2112 	case MLX4_STEERING_DMFS_A0_DYNAMIC:
2113 		return "dynamic hybrid mode";
2114 
2115 	case MLX4_STEERING_DMFS_A0_STATIC:
2116 		return "performance optimized for limited rule configuration (static)";
2117 
2118 	case MLX4_STEERING_DMFS_A0_DISABLE:
2119 		return "disabled performance optimized steering";
2120 
2121 	case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
2122 		return "performance optimized steering not supported";
2123 
2124 	default:
2125 		return "Unrecognized mode";
2126 	}
2127 }
2128 
2129 #define MLX4_DMFS_A0_STEERING			(1UL << 2)
2130 
2131 static void choose_steering_mode(struct mlx4_dev *dev,
2132 				 struct mlx4_dev_cap *dev_cap)
2133 {
2134 	if (mlx4_log_num_mgm_entry_size <= 0) {
2135 		if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
2136 			if (dev->caps.dmfs_high_steer_mode ==
2137 			    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2138 				mlx4_err(dev, "DMFS high rate mode not supported\n");
2139 			else
2140 				dev->caps.dmfs_high_steer_mode =
2141 					MLX4_STEERING_DMFS_A0_STATIC;
2142 		}
2143 	}
2144 
2145 	if (mlx4_log_num_mgm_entry_size <= 0 &&
2146 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
2147 	    (!mlx4_is_mfunc(dev) ||
2148 	     (dev_cap->fs_max_num_qp_per_entry >=
2149 	     (dev->persist->num_vfs + 1))) &&
2150 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
2151 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
2152 		dev->oper_log_mgm_entry_size =
2153 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
2154 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
2155 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
2156 		dev->caps.fs_log_max_ucast_qp_range_size =
2157 			dev_cap->fs_log_max_ucast_qp_range_size;
2158 	} else {
2159 		if (dev->caps.dmfs_high_steer_mode !=
2160 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2161 			dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
2162 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
2163 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2164 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
2165 		else {
2166 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
2167 
2168 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
2169 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2170 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
2171 		}
2172 		dev->oper_log_mgm_entry_size =
2173 			mlx4_log_num_mgm_entry_size > 0 ?
2174 			mlx4_log_num_mgm_entry_size :
2175 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
2176 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
2177 	}
2178 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
2179 		 mlx4_steering_mode_str(dev->caps.steering_mode),
2180 		 dev->oper_log_mgm_entry_size,
2181 		 mlx4_log_num_mgm_entry_size);
2182 }
2183 
2184 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
2185 				       struct mlx4_dev_cap *dev_cap)
2186 {
2187 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2188 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
2189 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
2190 	else
2191 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
2192 
2193 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
2194 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
2195 }
2196 
2197 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
2198 {
2199 	int i;
2200 	struct mlx4_port_cap port_cap;
2201 
2202 	if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2203 		return -EINVAL;
2204 
2205 	for (i = 1; i <= dev->caps.num_ports; i++) {
2206 		if (mlx4_dev_port(dev, i, &port_cap)) {
2207 			mlx4_err(dev,
2208 				 "QUERY_DEV_CAP command failed, can't veify DMFS high rate steering.\n");
2209 		} else if ((dev->caps.dmfs_high_steer_mode !=
2210 			    MLX4_STEERING_DMFS_A0_DEFAULT) &&
2211 			   (port_cap.dmfs_optimized_state ==
2212 			    !!(dev->caps.dmfs_high_steer_mode ==
2213 			    MLX4_STEERING_DMFS_A0_DISABLE))) {
2214 			mlx4_err(dev,
2215 				 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2216 				 dmfs_high_rate_steering_mode_str(
2217 					dev->caps.dmfs_high_steer_mode),
2218 				 (port_cap.dmfs_optimized_state ?
2219 					"enabled" : "disabled"));
2220 		}
2221 	}
2222 
2223 	return 0;
2224 }
2225 
2226 static int mlx4_init_fw(struct mlx4_dev *dev)
2227 {
2228 	struct mlx4_mod_stat_cfg   mlx4_cfg;
2229 	int err = 0;
2230 
2231 	if (!mlx4_is_slave(dev)) {
2232 		err = mlx4_QUERY_FW(dev);
2233 		if (err) {
2234 			if (err == -EACCES)
2235 				mlx4_info(dev, "non-primary physical function, skipping\n");
2236 			else
2237 				mlx4_err(dev, "QUERY_FW command failed, aborting\n");
2238 			return err;
2239 		}
2240 
2241 		err = mlx4_load_fw(dev);
2242 		if (err) {
2243 			mlx4_err(dev, "Failed to start FW, aborting\n");
2244 			return err;
2245 		}
2246 
2247 		mlx4_cfg.log_pg_sz_m = 1;
2248 		mlx4_cfg.log_pg_sz = 0;
2249 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2250 		if (err)
2251 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2252 	}
2253 
2254 	return err;
2255 }
2256 
2257 static int mlx4_init_hca(struct mlx4_dev *dev)
2258 {
2259 	struct mlx4_priv	  *priv = mlx4_priv(dev);
2260 	struct mlx4_adapter	   adapter;
2261 	struct mlx4_dev_cap	   dev_cap;
2262 	struct mlx4_profile	   profile;
2263 	struct mlx4_init_hca_param init_hca;
2264 	u64 icm_size;
2265 	struct mlx4_config_dev_params params;
2266 	int err;
2267 
2268 	if (!mlx4_is_slave(dev)) {
2269 		err = mlx4_dev_cap(dev, &dev_cap);
2270 		if (err) {
2271 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
2272 			return err;
2273 		}
2274 
2275 		choose_steering_mode(dev, &dev_cap);
2276 		choose_tunnel_offload_mode(dev, &dev_cap);
2277 
2278 		if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2279 		    mlx4_is_master(dev))
2280 			dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2281 
2282 		err = mlx4_get_phys_port_id(dev);
2283 		if (err)
2284 			mlx4_err(dev, "Fail to get physical port id\n");
2285 
2286 		if (mlx4_is_master(dev))
2287 			mlx4_parav_master_pf_caps(dev);
2288 
2289 		if (mlx4_low_memory_profile()) {
2290 			mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2291 			profile = low_mem_profile;
2292 		} else {
2293 			profile = default_profile;
2294 		}
2295 		if (dev->caps.steering_mode ==
2296 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
2297 			profile.num_mcg = MLX4_FS_NUM_MCG;
2298 
2299 		icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
2300 					     &init_hca);
2301 		if ((long long) icm_size < 0) {
2302 			err = icm_size;
2303 			return err;
2304 		}
2305 
2306 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
2307 
2308 		if (enable_4k_uar || !dev->persist->num_vfs) {
2309 			init_hca.log_uar_sz = ilog2(dev->caps.num_uars) +
2310 						    PAGE_SHIFT - DEFAULT_UAR_PAGE_SHIFT;
2311 			init_hca.uar_page_sz = DEFAULT_UAR_PAGE_SHIFT - 12;
2312 		} else {
2313 			init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
2314 			init_hca.uar_page_sz = PAGE_SHIFT - 12;
2315 		}
2316 
2317 		init_hca.mw_enabled = 0;
2318 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2319 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2320 			init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
2321 
2322 		err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
2323 		if (err)
2324 			return err;
2325 
2326 		err = mlx4_INIT_HCA(dev, &init_hca);
2327 		if (err) {
2328 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
2329 			goto err_free_icm;
2330 		}
2331 
2332 		if (dev_cap.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2333 			err = mlx4_query_func(dev, &dev_cap);
2334 			if (err < 0) {
2335 				mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
2336 				goto err_close;
2337 			} else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2338 				dev->caps.num_eqs = dev_cap.max_eqs;
2339 				dev->caps.reserved_eqs = dev_cap.reserved_eqs;
2340 				dev->caps.reserved_uars = dev_cap.reserved_uars;
2341 			}
2342 		}
2343 
2344 		/*
2345 		 * If TS is supported by FW
2346 		 * read HCA frequency by QUERY_HCA command
2347 		 */
2348 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2349 			memset(&init_hca, 0, sizeof(init_hca));
2350 			err = mlx4_QUERY_HCA(dev, &init_hca);
2351 			if (err) {
2352 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
2353 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2354 			} else {
2355 				dev->caps.hca_core_clock =
2356 					init_hca.hca_core_clock;
2357 			}
2358 
2359 			/* In case we got HCA frequency 0 - disable timestamping
2360 			 * to avoid dividing by zero
2361 			 */
2362 			if (!dev->caps.hca_core_clock) {
2363 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2364 				mlx4_err(dev,
2365 					 "HCA frequency is 0 - timestamping is not supported\n");
2366 			} else if (map_internal_clock(dev)) {
2367 				/*
2368 				 * Map internal clock,
2369 				 * in case of failure disable timestamping
2370 				 */
2371 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2372 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
2373 			}
2374 		}
2375 
2376 		if (dev->caps.dmfs_high_steer_mode !=
2377 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2378 			if (mlx4_validate_optimized_steering(dev))
2379 				mlx4_warn(dev, "Optimized steering validation failed\n");
2380 
2381 			if (dev->caps.dmfs_high_steer_mode ==
2382 			    MLX4_STEERING_DMFS_A0_DISABLE) {
2383 				dev->caps.dmfs_high_rate_qpn_base =
2384 					dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2385 				dev->caps.dmfs_high_rate_qpn_range =
2386 					MLX4_A0_STEERING_TABLE_SIZE;
2387 			}
2388 
2389 			mlx4_info(dev, "DMFS high rate steer mode is: %s\n",
2390 				  dmfs_high_rate_steering_mode_str(
2391 					dev->caps.dmfs_high_steer_mode));
2392 		}
2393 	} else {
2394 		err = mlx4_init_slave(dev);
2395 		if (err) {
2396 			if (err != -EPROBE_DEFER)
2397 				mlx4_err(dev, "Failed to initialize slave\n");
2398 			return err;
2399 		}
2400 
2401 		err = mlx4_slave_cap(dev);
2402 		if (err) {
2403 			mlx4_err(dev, "Failed to obtain slave caps\n");
2404 			goto err_close;
2405 		}
2406 	}
2407 
2408 	if (map_bf_area(dev))
2409 		mlx4_dbg(dev, "Failed to map blue flame area\n");
2410 
2411 	/*Only the master set the ports, all the rest got it from it.*/
2412 	if (!mlx4_is_slave(dev))
2413 		mlx4_set_port_mask(dev);
2414 
2415 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
2416 	if (err) {
2417 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2418 		goto unmap_bf;
2419 	}
2420 
2421 	/* Query CONFIG_DEV parameters */
2422 	err = mlx4_config_dev_retrieval(dev, &params);
2423 	if (err && err != -EOPNOTSUPP) {
2424 		mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2425 	} else if (!err) {
2426 		dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2427 		dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2428 	}
2429 	priv->eq_table.inta_pin = adapter.inta_pin;
2430 	memcpy(dev->board_id, adapter.board_id, sizeof(dev->board_id));
2431 
2432 	return 0;
2433 
2434 unmap_bf:
2435 	unmap_internal_clock(dev);
2436 	unmap_bf_area(dev);
2437 
2438 	if (mlx4_is_slave(dev))
2439 		mlx4_slave_destroy_special_qp_cap(dev);
2440 
2441 err_close:
2442 	if (mlx4_is_slave(dev))
2443 		mlx4_slave_exit(dev);
2444 	else
2445 		mlx4_CLOSE_HCA(dev, 0);
2446 
2447 err_free_icm:
2448 	if (!mlx4_is_slave(dev))
2449 		mlx4_free_icms(dev);
2450 
2451 	return err;
2452 }
2453 
2454 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2455 {
2456 	struct mlx4_priv *priv = mlx4_priv(dev);
2457 	int nent_pow2;
2458 
2459 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2460 		return -ENOENT;
2461 
2462 	if (!dev->caps.max_counters)
2463 		return -ENOSPC;
2464 
2465 	nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2466 	/* reserve last counter index for sink counter */
2467 	return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2468 				nent_pow2 - 1, 0,
2469 				nent_pow2 - dev->caps.max_counters + 1);
2470 }
2471 
2472 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2473 {
2474 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2475 		return;
2476 
2477 	if (!dev->caps.max_counters)
2478 		return;
2479 
2480 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2481 }
2482 
2483 static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2484 {
2485 	struct mlx4_priv *priv = mlx4_priv(dev);
2486 	int port;
2487 
2488 	for (port = 0; port < dev->caps.num_ports; port++)
2489 		if (priv->def_counter[port] != -1)
2490 			mlx4_counter_free(dev,  priv->def_counter[port]);
2491 }
2492 
2493 static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2494 {
2495 	struct mlx4_priv *priv = mlx4_priv(dev);
2496 	int port, err = 0;
2497 	u32 idx;
2498 
2499 	for (port = 0; port < dev->caps.num_ports; port++)
2500 		priv->def_counter[port] = -1;
2501 
2502 	for (port = 0; port < dev->caps.num_ports; port++) {
2503 		err = mlx4_counter_alloc(dev, &idx, MLX4_RES_USAGE_DRIVER);
2504 
2505 		if (!err || err == -ENOSPC) {
2506 			priv->def_counter[port] = idx;
2507 		} else if (err == -ENOENT) {
2508 			err = 0;
2509 			continue;
2510 		} else if (mlx4_is_slave(dev) && err == -EINVAL) {
2511 			priv->def_counter[port] = MLX4_SINK_COUNTER_INDEX(dev);
2512 			mlx4_warn(dev, "can't allocate counter from old PF driver, using index %d\n",
2513 				  MLX4_SINK_COUNTER_INDEX(dev));
2514 			err = 0;
2515 		} else {
2516 			mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2517 				 __func__, port + 1, err);
2518 			mlx4_cleanup_default_counters(dev);
2519 			return err;
2520 		}
2521 
2522 		mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2523 			 __func__, priv->def_counter[port], port + 1);
2524 	}
2525 
2526 	return err;
2527 }
2528 
2529 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2530 {
2531 	struct mlx4_priv *priv = mlx4_priv(dev);
2532 
2533 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2534 		return -ENOENT;
2535 
2536 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2537 	if (*idx == -1) {
2538 		*idx = MLX4_SINK_COUNTER_INDEX(dev);
2539 		return -ENOSPC;
2540 	}
2541 
2542 	return 0;
2543 }
2544 
2545 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx, u8 usage)
2546 {
2547 	u32 in_modifier = RES_COUNTER | (((u32)usage & 3) << 30);
2548 	u64 out_param;
2549 	int err;
2550 
2551 	if (mlx4_is_mfunc(dev)) {
2552 		err = mlx4_cmd_imm(dev, 0, &out_param, in_modifier,
2553 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2554 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2555 		if (!err)
2556 			*idx = get_param_l(&out_param);
2557 
2558 		return err;
2559 	}
2560 	return __mlx4_counter_alloc(dev, idx);
2561 }
2562 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2563 
2564 static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2565 				u8 counter_index)
2566 {
2567 	struct mlx4_cmd_mailbox *if_stat_mailbox;
2568 	int err;
2569 	u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2570 
2571 	if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2572 	if (IS_ERR(if_stat_mailbox))
2573 		return PTR_ERR(if_stat_mailbox);
2574 
2575 	err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2576 			   MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2577 			   MLX4_CMD_NATIVE);
2578 
2579 	mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2580 	return err;
2581 }
2582 
2583 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2584 {
2585 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2586 		return;
2587 
2588 	if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2589 		return;
2590 
2591 	__mlx4_clear_if_stat(dev, idx);
2592 
2593 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2594 	return;
2595 }
2596 
2597 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2598 {
2599 	u64 in_param = 0;
2600 
2601 	if (mlx4_is_mfunc(dev)) {
2602 		set_param_l(&in_param, idx);
2603 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2604 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2605 			 MLX4_CMD_WRAPPED);
2606 		return;
2607 	}
2608 	__mlx4_counter_free(dev, idx);
2609 }
2610 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2611 
2612 int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2613 {
2614 	struct mlx4_priv *priv = mlx4_priv(dev);
2615 
2616 	return priv->def_counter[port - 1];
2617 }
2618 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2619 
2620 void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2621 {
2622 	struct mlx4_priv *priv = mlx4_priv(dev);
2623 
2624 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2625 }
2626 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2627 
2628 __be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2629 {
2630 	struct mlx4_priv *priv = mlx4_priv(dev);
2631 
2632 	return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2633 }
2634 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2635 
2636 void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2637 {
2638 	struct mlx4_priv *priv = mlx4_priv(dev);
2639 	__be64 guid;
2640 
2641 	/* hw GUID */
2642 	if (entry == 0)
2643 		return;
2644 
2645 	get_random_bytes((char *)&guid, sizeof(guid));
2646 	guid &= ~(cpu_to_be64(1ULL << 56));
2647 	guid |= cpu_to_be64(1ULL << 57);
2648 	priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2649 }
2650 
2651 static int mlx4_setup_hca(struct mlx4_dev *dev)
2652 {
2653 	struct mlx4_priv *priv = mlx4_priv(dev);
2654 	int err;
2655 	int port;
2656 	__be32 ib_port_default_caps;
2657 
2658 	err = mlx4_init_uar_table(dev);
2659 	if (err) {
2660 		mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2661 		return err;
2662 	}
2663 
2664 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
2665 	if (err) {
2666 		mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2667 		goto err_uar_table_free;
2668 	}
2669 
2670 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2671 	if (!priv->kar) {
2672 		mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2673 		err = -ENOMEM;
2674 		goto err_uar_free;
2675 	}
2676 
2677 	err = mlx4_init_pd_table(dev);
2678 	if (err) {
2679 		mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2680 		goto err_kar_unmap;
2681 	}
2682 
2683 	err = mlx4_init_xrcd_table(dev);
2684 	if (err) {
2685 		mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2686 		goto err_pd_table_free;
2687 	}
2688 
2689 	err = mlx4_init_mr_table(dev);
2690 	if (err) {
2691 		mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2692 		goto err_xrcd_table_free;
2693 	}
2694 
2695 	if (!mlx4_is_slave(dev)) {
2696 		err = mlx4_init_mcg_table(dev);
2697 		if (err) {
2698 			mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2699 			goto err_mr_table_free;
2700 		}
2701 		err = mlx4_config_mad_demux(dev);
2702 		if (err) {
2703 			mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2704 			goto err_mcg_table_free;
2705 		}
2706 	}
2707 
2708 	err = mlx4_init_eq_table(dev);
2709 	if (err) {
2710 		mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2711 		goto err_mcg_table_free;
2712 	}
2713 
2714 	err = mlx4_cmd_use_events(dev);
2715 	if (err) {
2716 		mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2717 		goto err_eq_table_free;
2718 	}
2719 
2720 	err = mlx4_NOP(dev);
2721 	if (err) {
2722 		if (dev->flags & MLX4_FLAG_MSI_X) {
2723 			mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2724 				  priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2725 			mlx4_warn(dev, "Trying again without MSI-X\n");
2726 		} else {
2727 			mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2728 				 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2729 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2730 		}
2731 
2732 		goto err_cmd_poll;
2733 	}
2734 
2735 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
2736 
2737 	err = mlx4_init_cq_table(dev);
2738 	if (err) {
2739 		mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2740 		goto err_cmd_poll;
2741 	}
2742 
2743 	err = mlx4_init_srq_table(dev);
2744 	if (err) {
2745 		mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2746 		goto err_cq_table_free;
2747 	}
2748 
2749 	err = mlx4_init_qp_table(dev);
2750 	if (err) {
2751 		mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2752 		goto err_srq_table_free;
2753 	}
2754 
2755 	if (!mlx4_is_slave(dev)) {
2756 		err = mlx4_init_counters_table(dev);
2757 		if (err && err != -ENOENT) {
2758 			mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2759 			goto err_qp_table_free;
2760 		}
2761 	}
2762 
2763 	err = mlx4_allocate_default_counters(dev);
2764 	if (err) {
2765 		mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2766 		goto err_counters_table_free;
2767 	}
2768 
2769 	if (!mlx4_is_slave(dev)) {
2770 		for (port = 1; port <= dev->caps.num_ports; port++) {
2771 			ib_port_default_caps = 0;
2772 			err = mlx4_get_port_ib_caps(dev, port,
2773 						    &ib_port_default_caps);
2774 			if (err)
2775 				mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2776 					  port, err);
2777 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2778 
2779 			/* initialize per-slave default ib port capabilities */
2780 			if (mlx4_is_master(dev)) {
2781 				int i;
2782 				for (i = 0; i < dev->num_slaves; i++) {
2783 					if (i == mlx4_master_func_num(dev))
2784 						continue;
2785 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2786 						ib_port_default_caps;
2787 				}
2788 			}
2789 
2790 			if (mlx4_is_mfunc(dev))
2791 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2792 			else
2793 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2794 
2795 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2796 					    dev->caps.pkey_table_len[port] : -1);
2797 			if (err) {
2798 				mlx4_err(dev, "Failed to set port %d, aborting\n",
2799 					 port);
2800 				goto err_default_countes_free;
2801 			}
2802 		}
2803 	}
2804 
2805 	return 0;
2806 
2807 err_default_countes_free:
2808 	mlx4_cleanup_default_counters(dev);
2809 
2810 err_counters_table_free:
2811 	if (!mlx4_is_slave(dev))
2812 		mlx4_cleanup_counters_table(dev);
2813 
2814 err_qp_table_free:
2815 	mlx4_cleanup_qp_table(dev);
2816 
2817 err_srq_table_free:
2818 	mlx4_cleanup_srq_table(dev);
2819 
2820 err_cq_table_free:
2821 	mlx4_cleanup_cq_table(dev);
2822 
2823 err_cmd_poll:
2824 	mlx4_cmd_use_polling(dev);
2825 
2826 err_eq_table_free:
2827 	mlx4_cleanup_eq_table(dev);
2828 
2829 err_mcg_table_free:
2830 	if (!mlx4_is_slave(dev))
2831 		mlx4_cleanup_mcg_table(dev);
2832 
2833 err_mr_table_free:
2834 	mlx4_cleanup_mr_table(dev);
2835 
2836 err_xrcd_table_free:
2837 	mlx4_cleanup_xrcd_table(dev);
2838 
2839 err_pd_table_free:
2840 	mlx4_cleanup_pd_table(dev);
2841 
2842 err_kar_unmap:
2843 	iounmap(priv->kar);
2844 
2845 err_uar_free:
2846 	mlx4_uar_free(dev, &priv->driver_uar);
2847 
2848 err_uar_table_free:
2849 	mlx4_cleanup_uar_table(dev);
2850 	return err;
2851 }
2852 
2853 static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2854 {
2855 	int requested_cpu = 0;
2856 	struct mlx4_priv *priv = mlx4_priv(dev);
2857 	struct mlx4_eq *eq;
2858 	int off = 0;
2859 	int i;
2860 
2861 	if (eqn > dev->caps.num_comp_vectors)
2862 		return -EINVAL;
2863 
2864 	for (i = 1; i < port; i++)
2865 		off += mlx4_get_eqs_per_port(dev, i);
2866 
2867 	requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2868 
2869 	/* Meaning EQs are shared, and this call comes from the second port */
2870 	if (requested_cpu < 0)
2871 		return 0;
2872 
2873 	eq = &priv->eq_table.eq[eqn];
2874 
2875 	if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2876 		return -ENOMEM;
2877 
2878 	cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2879 
2880 	return 0;
2881 }
2882 
2883 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2884 {
2885 	struct mlx4_priv *priv = mlx4_priv(dev);
2886 	struct msix_entry *entries;
2887 	int i;
2888 	int port = 0;
2889 
2890 	if (msi_x) {
2891 		int nreq = min3(dev->caps.num_ports *
2892 				(int)num_online_cpus() + 1,
2893 				dev->caps.num_eqs - dev->caps.reserved_eqs,
2894 				MAX_MSIX);
2895 
2896 		entries = kcalloc(nreq, sizeof(*entries), GFP_KERNEL);
2897 		if (!entries)
2898 			goto no_msi;
2899 
2900 		for (i = 0; i < nreq; ++i)
2901 			entries[i].entry = i;
2902 
2903 		nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2904 					     nreq);
2905 
2906 		if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
2907 			kfree(entries);
2908 			goto no_msi;
2909 		}
2910 		/* 1 is reserved for events (asyncrounous EQ) */
2911 		dev->caps.num_comp_vectors = nreq - 1;
2912 
2913 		priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2914 		bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2915 			    dev->caps.num_ports);
2916 
2917 		for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2918 			if (i == MLX4_EQ_ASYNC)
2919 				continue;
2920 
2921 			priv->eq_table.eq[i].irq =
2922 				entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2923 
2924 			if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2925 				bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2926 					    dev->caps.num_ports);
2927 				/* We don't set affinity hint when there
2928 				 * aren't enough EQs
2929 				 */
2930 			} else {
2931 				set_bit(port,
2932 					priv->eq_table.eq[i].actv_ports.ports);
2933 				if (mlx4_init_affinity_hint(dev, port + 1, i))
2934 					mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
2935 						  i);
2936 			}
2937 			/* We divide the Eqs evenly between the two ports.
2938 			 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
2939 			 * refers to the number of Eqs per port
2940 			 * (i.e eqs_per_port). Theoretically, we would like to
2941 			 * write something like (i + 1) % eqs_per_port == 0.
2942 			 * However, since there's an asynchronous Eq, we have
2943 			 * to skip over it by comparing this condition to
2944 			 * !!((i + 1) > MLX4_EQ_ASYNC).
2945 			 */
2946 			if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
2947 			    ((i + 1) %
2948 			     (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
2949 			    !!((i + 1) > MLX4_EQ_ASYNC))
2950 				/* If dev->caps.num_comp_vectors < dev->caps.num_ports,
2951 				 * everything is shared anyway.
2952 				 */
2953 				port++;
2954 		}
2955 
2956 		dev->flags |= MLX4_FLAG_MSI_X;
2957 
2958 		kfree(entries);
2959 		return;
2960 	}
2961 
2962 no_msi:
2963 	dev->caps.num_comp_vectors = 1;
2964 
2965 	BUG_ON(MLX4_EQ_ASYNC >= 2);
2966 	for (i = 0; i < 2; ++i) {
2967 		priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
2968 		if (i != MLX4_EQ_ASYNC) {
2969 			bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2970 				    dev->caps.num_ports);
2971 		}
2972 	}
2973 }
2974 
2975 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2976 {
2977 	struct devlink *devlink = priv_to_devlink(mlx4_priv(dev));
2978 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2979 	int err;
2980 
2981 	err = devlink_port_register(devlink, &info->devlink_port, port);
2982 	if (err)
2983 		return err;
2984 
2985 	info->dev = dev;
2986 	info->port = port;
2987 	if (!mlx4_is_slave(dev)) {
2988 		mlx4_init_mac_table(dev, &info->mac_table);
2989 		mlx4_init_vlan_table(dev, &info->vlan_table);
2990 		mlx4_init_roce_gid_table(dev, &info->gid_table);
2991 		info->base_qpn = mlx4_get_base_qpn(dev, port);
2992 	}
2993 
2994 	sprintf(info->dev_name, "mlx4_port%d", port);
2995 	info->port_attr.attr.name = info->dev_name;
2996 	if (mlx4_is_mfunc(dev))
2997 		info->port_attr.attr.mode = S_IRUGO;
2998 	else {
2999 		info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
3000 		info->port_attr.store     = set_port_type;
3001 	}
3002 	info->port_attr.show      = show_port_type;
3003 	sysfs_attr_init(&info->port_attr.attr);
3004 
3005 	err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
3006 	if (err) {
3007 		mlx4_err(dev, "Failed to create file for port %d\n", port);
3008 		devlink_port_unregister(&info->devlink_port);
3009 		info->port = -1;
3010 	}
3011 
3012 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
3013 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
3014 	if (mlx4_is_mfunc(dev))
3015 		info->port_mtu_attr.attr.mode = S_IRUGO;
3016 	else {
3017 		info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
3018 		info->port_mtu_attr.store     = set_port_ib_mtu;
3019 	}
3020 	info->port_mtu_attr.show      = show_port_ib_mtu;
3021 	sysfs_attr_init(&info->port_mtu_attr.attr);
3022 
3023 	err = device_create_file(&dev->persist->pdev->dev,
3024 				 &info->port_mtu_attr);
3025 	if (err) {
3026 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
3027 		device_remove_file(&info->dev->persist->pdev->dev,
3028 				   &info->port_attr);
3029 		devlink_port_unregister(&info->devlink_port);
3030 		info->port = -1;
3031 	}
3032 
3033 	return err;
3034 }
3035 
3036 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
3037 {
3038 	if (info->port < 0)
3039 		return;
3040 
3041 	device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
3042 	device_remove_file(&info->dev->persist->pdev->dev,
3043 			   &info->port_mtu_attr);
3044 	devlink_port_unregister(&info->devlink_port);
3045 
3046 #ifdef CONFIG_RFS_ACCEL
3047 	free_irq_cpu_rmap(info->rmap);
3048 	info->rmap = NULL;
3049 #endif
3050 }
3051 
3052 static int mlx4_init_steering(struct mlx4_dev *dev)
3053 {
3054 	struct mlx4_priv *priv = mlx4_priv(dev);
3055 	int num_entries = dev->caps.num_ports;
3056 	int i, j;
3057 
3058 	priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
3059 	if (!priv->steer)
3060 		return -ENOMEM;
3061 
3062 	for (i = 0; i < num_entries; i++)
3063 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3064 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
3065 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
3066 		}
3067 	return 0;
3068 }
3069 
3070 static void mlx4_clear_steering(struct mlx4_dev *dev)
3071 {
3072 	struct mlx4_priv *priv = mlx4_priv(dev);
3073 	struct mlx4_steer_index *entry, *tmp_entry;
3074 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
3075 	int num_entries = dev->caps.num_ports;
3076 	int i, j;
3077 
3078 	for (i = 0; i < num_entries; i++) {
3079 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
3080 			list_for_each_entry_safe(pqp, tmp_pqp,
3081 						 &priv->steer[i].promisc_qps[j],
3082 						 list) {
3083 				list_del(&pqp->list);
3084 				kfree(pqp);
3085 			}
3086 			list_for_each_entry_safe(entry, tmp_entry,
3087 						 &priv->steer[i].steer_entries[j],
3088 						 list) {
3089 				list_del(&entry->list);
3090 				list_for_each_entry_safe(pqp, tmp_pqp,
3091 							 &entry->duplicates,
3092 							 list) {
3093 					list_del(&pqp->list);
3094 					kfree(pqp);
3095 				}
3096 				kfree(entry);
3097 			}
3098 		}
3099 	}
3100 	kfree(priv->steer);
3101 }
3102 
3103 static int extended_func_num(struct pci_dev *pdev)
3104 {
3105 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
3106 }
3107 
3108 #define MLX4_OWNER_BASE	0x8069c
3109 #define MLX4_OWNER_SIZE	4
3110 
3111 static int mlx4_get_ownership(struct mlx4_dev *dev)
3112 {
3113 	void __iomem *owner;
3114 	u32 ret;
3115 
3116 	if (pci_channel_offline(dev->persist->pdev))
3117 		return -EIO;
3118 
3119 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3120 			MLX4_OWNER_BASE,
3121 			MLX4_OWNER_SIZE);
3122 	if (!owner) {
3123 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3124 		return -ENOMEM;
3125 	}
3126 
3127 	ret = readl(owner);
3128 	iounmap(owner);
3129 	return (int) !!ret;
3130 }
3131 
3132 static void mlx4_free_ownership(struct mlx4_dev *dev)
3133 {
3134 	void __iomem *owner;
3135 
3136 	if (pci_channel_offline(dev->persist->pdev))
3137 		return;
3138 
3139 	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3140 			MLX4_OWNER_BASE,
3141 			MLX4_OWNER_SIZE);
3142 	if (!owner) {
3143 		mlx4_err(dev, "Failed to obtain ownership bit\n");
3144 		return;
3145 	}
3146 	writel(0, owner);
3147 	msleep(1000);
3148 	iounmap(owner);
3149 }
3150 
3151 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV)	==\
3152 				  !!((flags) & MLX4_FLAG_MASTER))
3153 
3154 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
3155 			     u8 total_vfs, int existing_vfs, int reset_flow)
3156 {
3157 	u64 dev_flags = dev->flags;
3158 	int err = 0;
3159 	int fw_enabled_sriov_vfs = min(pci_sriov_get_totalvfs(pdev),
3160 					MLX4_MAX_NUM_VF);
3161 
3162 	if (reset_flow) {
3163 		dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
3164 				       GFP_KERNEL);
3165 		if (!dev->dev_vfs)
3166 			goto free_mem;
3167 		return dev_flags;
3168 	}
3169 
3170 	atomic_inc(&pf_loading);
3171 	if (dev->flags &  MLX4_FLAG_SRIOV) {
3172 		if (existing_vfs != total_vfs) {
3173 			mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
3174 				 existing_vfs, total_vfs);
3175 			total_vfs = existing_vfs;
3176 		}
3177 	}
3178 
3179 	dev->dev_vfs = kzalloc(total_vfs * sizeof(*dev->dev_vfs), GFP_KERNEL);
3180 	if (NULL == dev->dev_vfs) {
3181 		mlx4_err(dev, "Failed to allocate memory for VFs\n");
3182 		goto disable_sriov;
3183 	}
3184 
3185 	if (!(dev->flags &  MLX4_FLAG_SRIOV)) {
3186 		if (total_vfs > fw_enabled_sriov_vfs) {
3187 			mlx4_err(dev, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
3188 				 total_vfs, fw_enabled_sriov_vfs);
3189 			err = -ENOMEM;
3190 			goto disable_sriov;
3191 		}
3192 		mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
3193 		err = pci_enable_sriov(pdev, total_vfs);
3194 	}
3195 	if (err) {
3196 		mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
3197 			 err);
3198 		goto disable_sriov;
3199 	} else {
3200 		mlx4_warn(dev, "Running in master mode\n");
3201 		dev_flags |= MLX4_FLAG_SRIOV |
3202 			MLX4_FLAG_MASTER;
3203 		dev_flags &= ~MLX4_FLAG_SLAVE;
3204 		dev->persist->num_vfs = total_vfs;
3205 	}
3206 	return dev_flags;
3207 
3208 disable_sriov:
3209 	atomic_dec(&pf_loading);
3210 free_mem:
3211 	dev->persist->num_vfs = 0;
3212 	kfree(dev->dev_vfs);
3213         dev->dev_vfs = NULL;
3214 	return dev_flags & ~MLX4_FLAG_MASTER;
3215 }
3216 
3217 enum {
3218 	MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
3219 };
3220 
3221 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
3222 			      int *nvfs)
3223 {
3224 	int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
3225 	/* Checking for 64 VFs as a limitation of CX2 */
3226 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
3227 	    requested_vfs >= 64) {
3228 		mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
3229 			 requested_vfs);
3230 		return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
3231 	}
3232 	return 0;
3233 }
3234 
3235 static int mlx4_pci_enable_device(struct mlx4_dev *dev)
3236 {
3237 	struct pci_dev *pdev = dev->persist->pdev;
3238 	int err = 0;
3239 
3240 	mutex_lock(&dev->persist->pci_status_mutex);
3241 	if (dev->persist->pci_status == MLX4_PCI_STATUS_DISABLED) {
3242 		err = pci_enable_device(pdev);
3243 		if (!err)
3244 			dev->persist->pci_status = MLX4_PCI_STATUS_ENABLED;
3245 	}
3246 	mutex_unlock(&dev->persist->pci_status_mutex);
3247 
3248 	return err;
3249 }
3250 
3251 static void mlx4_pci_disable_device(struct mlx4_dev *dev)
3252 {
3253 	struct pci_dev *pdev = dev->persist->pdev;
3254 
3255 	mutex_lock(&dev->persist->pci_status_mutex);
3256 	if (dev->persist->pci_status == MLX4_PCI_STATUS_ENABLED) {
3257 		pci_disable_device(pdev);
3258 		dev->persist->pci_status = MLX4_PCI_STATUS_DISABLED;
3259 	}
3260 	mutex_unlock(&dev->persist->pci_status_mutex);
3261 }
3262 
3263 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
3264 			 int total_vfs, int *nvfs, struct mlx4_priv *priv,
3265 			 int reset_flow)
3266 {
3267 	struct mlx4_dev *dev;
3268 	unsigned sum = 0;
3269 	int err;
3270 	int port;
3271 	int i;
3272 	struct mlx4_dev_cap *dev_cap = NULL;
3273 	int existing_vfs = 0;
3274 
3275 	dev = &priv->dev;
3276 
3277 	INIT_LIST_HEAD(&priv->ctx_list);
3278 	spin_lock_init(&priv->ctx_lock);
3279 
3280 	mutex_init(&priv->port_mutex);
3281 	mutex_init(&priv->bond_mutex);
3282 
3283 	INIT_LIST_HEAD(&priv->pgdir_list);
3284 	mutex_init(&priv->pgdir_mutex);
3285 	spin_lock_init(&priv->cmd.context_lock);
3286 
3287 	INIT_LIST_HEAD(&priv->bf_list);
3288 	mutex_init(&priv->bf_mutex);
3289 
3290 	dev->rev_id = pdev->revision;
3291 	dev->numa_node = dev_to_node(&pdev->dev);
3292 
3293 	/* Detect if this device is a virtual function */
3294 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3295 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3296 		dev->flags |= MLX4_FLAG_SLAVE;
3297 	} else {
3298 		/* We reset the device and enable SRIOV only for physical
3299 		 * devices.  Try to claim ownership on the device;
3300 		 * if already taken, skip -- do not allow multiple PFs */
3301 		err = mlx4_get_ownership(dev);
3302 		if (err) {
3303 			if (err < 0)
3304 				return err;
3305 			else {
3306 				mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
3307 				return -EINVAL;
3308 			}
3309 		}
3310 
3311 		atomic_set(&priv->opreq_count, 0);
3312 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3313 
3314 		/*
3315 		 * Now reset the HCA before we touch the PCI capabilities or
3316 		 * attempt a firmware command, since a boot ROM may have left
3317 		 * the HCA in an undefined state.
3318 		 */
3319 		err = mlx4_reset(dev);
3320 		if (err) {
3321 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
3322 			goto err_sriov;
3323 		}
3324 
3325 		if (total_vfs) {
3326 			dev->flags = MLX4_FLAG_MASTER;
3327 			existing_vfs = pci_num_vf(pdev);
3328 			if (existing_vfs)
3329 				dev->flags |= MLX4_FLAG_SRIOV;
3330 			dev->persist->num_vfs = total_vfs;
3331 		}
3332 	}
3333 
3334 	/* on load remove any previous indication of internal error,
3335 	 * device is up.
3336 	 */
3337 	dev->persist->state = MLX4_DEVICE_STATE_UP;
3338 
3339 slave_start:
3340 	err = mlx4_cmd_init(dev);
3341 	if (err) {
3342 		mlx4_err(dev, "Failed to init command interface, aborting\n");
3343 		goto err_sriov;
3344 	}
3345 
3346 	/* In slave functions, the communication channel must be initialized
3347 	 * before posting commands. Also, init num_slaves before calling
3348 	 * mlx4_init_hca */
3349 	if (mlx4_is_mfunc(dev)) {
3350 		if (mlx4_is_master(dev)) {
3351 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
3352 
3353 		} else {
3354 			dev->num_slaves = 0;
3355 			err = mlx4_multi_func_init(dev);
3356 			if (err) {
3357 				mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
3358 				goto err_cmd;
3359 			}
3360 		}
3361 	}
3362 
3363 	err = mlx4_init_fw(dev);
3364 	if (err) {
3365 		mlx4_err(dev, "Failed to init fw, aborting.\n");
3366 		goto err_mfunc;
3367 	}
3368 
3369 	if (mlx4_is_master(dev)) {
3370 		/* when we hit the goto slave_start below, dev_cap already initialized */
3371 		if (!dev_cap) {
3372 			dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3373 
3374 			if (!dev_cap) {
3375 				err = -ENOMEM;
3376 				goto err_fw;
3377 			}
3378 
3379 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3380 			if (err) {
3381 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3382 				goto err_fw;
3383 			}
3384 
3385 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3386 				goto err_fw;
3387 
3388 			if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3389 				u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3390 								  total_vfs,
3391 								  existing_vfs,
3392 								  reset_flow);
3393 
3394 				mlx4_close_fw(dev);
3395 				mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3396 				dev->flags = dev_flags;
3397 				if (!SRIOV_VALID_STATE(dev->flags)) {
3398 					mlx4_err(dev, "Invalid SRIOV state\n");
3399 					goto err_sriov;
3400 				}
3401 				err = mlx4_reset(dev);
3402 				if (err) {
3403 					mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3404 					goto err_sriov;
3405 				}
3406 				goto slave_start;
3407 			}
3408 		} else {
3409 			/* Legacy mode FW requires SRIOV to be enabled before
3410 			 * doing QUERY_DEV_CAP, since max_eq's value is different if
3411 			 * SRIOV is enabled.
3412 			 */
3413 			memset(dev_cap, 0, sizeof(*dev_cap));
3414 			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3415 			if (err) {
3416 				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3417 				goto err_fw;
3418 			}
3419 
3420 			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3421 				goto err_fw;
3422 		}
3423 	}
3424 
3425 	err = mlx4_init_hca(dev);
3426 	if (err) {
3427 		if (err == -EACCES) {
3428 			/* Not primary Physical function
3429 			 * Running in slave mode */
3430 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3431 			/* We're not a PF */
3432 			if (dev->flags & MLX4_FLAG_SRIOV) {
3433 				if (!existing_vfs)
3434 					pci_disable_sriov(pdev);
3435 				if (mlx4_is_master(dev) && !reset_flow)
3436 					atomic_dec(&pf_loading);
3437 				dev->flags &= ~MLX4_FLAG_SRIOV;
3438 			}
3439 			if (!mlx4_is_slave(dev))
3440 				mlx4_free_ownership(dev);
3441 			dev->flags |= MLX4_FLAG_SLAVE;
3442 			dev->flags &= ~MLX4_FLAG_MASTER;
3443 			goto slave_start;
3444 		} else
3445 			goto err_fw;
3446 	}
3447 
3448 	if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3449 		u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3450 						  existing_vfs, reset_flow);
3451 
3452 		if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3453 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3454 			dev->flags = dev_flags;
3455 			err = mlx4_cmd_init(dev);
3456 			if (err) {
3457 				/* Only VHCR is cleaned up, so could still
3458 				 * send FW commands
3459 				 */
3460 				mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3461 				goto err_close;
3462 			}
3463 		} else {
3464 			dev->flags = dev_flags;
3465 		}
3466 
3467 		if (!SRIOV_VALID_STATE(dev->flags)) {
3468 			mlx4_err(dev, "Invalid SRIOV state\n");
3469 			goto err_close;
3470 		}
3471 	}
3472 
3473 	/* check if the device is functioning at its maximum possible speed.
3474 	 * No return code for this call, just warn the user in case of PCI
3475 	 * express device capabilities are under-satisfied by the bus.
3476 	 */
3477 	if (!mlx4_is_slave(dev))
3478 		mlx4_check_pcie_caps(dev);
3479 
3480 	/* In master functions, the communication channel must be initialized
3481 	 * after obtaining its address from fw */
3482 	if (mlx4_is_master(dev)) {
3483 		if (dev->caps.num_ports < 2 &&
3484 		    num_vfs_argc > 1) {
3485 			err = -EINVAL;
3486 			mlx4_err(dev,
3487 				 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3488 				 dev->caps.num_ports);
3489 			goto err_close;
3490 		}
3491 		memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
3492 
3493 		for (i = 0;
3494 		     i < sizeof(dev->persist->nvfs)/
3495 		     sizeof(dev->persist->nvfs[0]); i++) {
3496 			unsigned j;
3497 
3498 			for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
3499 				dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3500 				dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3501 					dev->caps.num_ports;
3502 			}
3503 		}
3504 
3505 		/* In master functions, the communication channel
3506 		 * must be initialized after obtaining its address from fw
3507 		 */
3508 		err = mlx4_multi_func_init(dev);
3509 		if (err) {
3510 			mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3511 			goto err_close;
3512 		}
3513 	}
3514 
3515 	err = mlx4_alloc_eq_table(dev);
3516 	if (err)
3517 		goto err_master_mfunc;
3518 
3519 	bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
3520 	mutex_init(&priv->msix_ctl.pool_lock);
3521 
3522 	mlx4_enable_msi_x(dev);
3523 	if ((mlx4_is_mfunc(dev)) &&
3524 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
3525 		err = -EOPNOTSUPP;
3526 		mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
3527 		goto err_free_eq;
3528 	}
3529 
3530 	if (!mlx4_is_slave(dev)) {
3531 		err = mlx4_init_steering(dev);
3532 		if (err)
3533 			goto err_disable_msix;
3534 	}
3535 
3536 	mlx4_init_quotas(dev);
3537 
3538 	err = mlx4_setup_hca(dev);
3539 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3540 	    !mlx4_is_mfunc(dev)) {
3541 		dev->flags &= ~MLX4_FLAG_MSI_X;
3542 		dev->caps.num_comp_vectors = 1;
3543 		pci_disable_msix(pdev);
3544 		err = mlx4_setup_hca(dev);
3545 	}
3546 
3547 	if (err)
3548 		goto err_steer;
3549 
3550 	/* When PF resources are ready arm its comm channel to enable
3551 	 * getting commands
3552 	 */
3553 	if (mlx4_is_master(dev)) {
3554 		err = mlx4_ARM_COMM_CHANNEL(dev);
3555 		if (err) {
3556 			mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3557 				 err);
3558 			goto err_steer;
3559 		}
3560 	}
3561 
3562 	for (port = 1; port <= dev->caps.num_ports; port++) {
3563 		err = mlx4_init_port_info(dev, port);
3564 		if (err)
3565 			goto err_port;
3566 	}
3567 
3568 	priv->v2p.port1 = 1;
3569 	priv->v2p.port2 = 2;
3570 
3571 	err = mlx4_register_device(dev);
3572 	if (err)
3573 		goto err_port;
3574 
3575 	mlx4_request_modules(dev);
3576 
3577 	mlx4_sense_init(dev);
3578 	mlx4_start_sense(dev);
3579 
3580 	priv->removed = 0;
3581 
3582 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3583 		atomic_dec(&pf_loading);
3584 
3585 	kfree(dev_cap);
3586 	return 0;
3587 
3588 err_port:
3589 	for (--port; port >= 1; --port)
3590 		mlx4_cleanup_port_info(&priv->port[port]);
3591 
3592 	mlx4_cleanup_default_counters(dev);
3593 	if (!mlx4_is_slave(dev))
3594 		mlx4_cleanup_counters_table(dev);
3595 	mlx4_cleanup_qp_table(dev);
3596 	mlx4_cleanup_srq_table(dev);
3597 	mlx4_cleanup_cq_table(dev);
3598 	mlx4_cmd_use_polling(dev);
3599 	mlx4_cleanup_eq_table(dev);
3600 	mlx4_cleanup_mcg_table(dev);
3601 	mlx4_cleanup_mr_table(dev);
3602 	mlx4_cleanup_xrcd_table(dev);
3603 	mlx4_cleanup_pd_table(dev);
3604 	mlx4_cleanup_uar_table(dev);
3605 
3606 err_steer:
3607 	if (!mlx4_is_slave(dev))
3608 		mlx4_clear_steering(dev);
3609 
3610 err_disable_msix:
3611 	if (dev->flags & MLX4_FLAG_MSI_X)
3612 		pci_disable_msix(pdev);
3613 
3614 err_free_eq:
3615 	mlx4_free_eq_table(dev);
3616 
3617 err_master_mfunc:
3618 	if (mlx4_is_master(dev)) {
3619 		mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
3620 		mlx4_multi_func_cleanup(dev);
3621 	}
3622 
3623 	if (mlx4_is_slave(dev))
3624 		mlx4_slave_destroy_special_qp_cap(dev);
3625 
3626 err_close:
3627 	mlx4_close_hca(dev);
3628 
3629 err_fw:
3630 	mlx4_close_fw(dev);
3631 
3632 err_mfunc:
3633 	if (mlx4_is_slave(dev))
3634 		mlx4_multi_func_cleanup(dev);
3635 
3636 err_cmd:
3637 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3638 
3639 err_sriov:
3640 	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3641 		pci_disable_sriov(pdev);
3642 		dev->flags &= ~MLX4_FLAG_SRIOV;
3643 	}
3644 
3645 	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3646 		atomic_dec(&pf_loading);
3647 
3648 	kfree(priv->dev.dev_vfs);
3649 
3650 	if (!mlx4_is_slave(dev))
3651 		mlx4_free_ownership(dev);
3652 
3653 	kfree(dev_cap);
3654 	return err;
3655 }
3656 
3657 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3658 			   struct mlx4_priv *priv)
3659 {
3660 	int err;
3661 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3662 	int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3663 	const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3664 		{2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3665 	unsigned total_vfs = 0;
3666 	unsigned int i;
3667 
3668 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3669 
3670 	err = mlx4_pci_enable_device(&priv->dev);
3671 	if (err) {
3672 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3673 		return err;
3674 	}
3675 
3676 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3677 	 * per port, we must limit the number of VFs to 63 (since their are
3678 	 * 128 MACs)
3679 	 */
3680 	for (i = 0; i < ARRAY_SIZE(nvfs) && i < num_vfs_argc;
3681 	     total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3682 		nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3683 		if (nvfs[i] < 0) {
3684 			dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3685 			err = -EINVAL;
3686 			goto err_disable_pdev;
3687 		}
3688 	}
3689 	for (i = 0; i < ARRAY_SIZE(prb_vf) && i < probe_vfs_argc;
3690 	     i++) {
3691 		prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3692 		if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3693 			dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3694 			err = -EINVAL;
3695 			goto err_disable_pdev;
3696 		}
3697 	}
3698 	if (total_vfs > MLX4_MAX_NUM_VF) {
3699 		dev_err(&pdev->dev,
3700 			"Requested more VF's (%d) than allowed by hw (%d)\n",
3701 			total_vfs, MLX4_MAX_NUM_VF);
3702 		err = -EINVAL;
3703 		goto err_disable_pdev;
3704 	}
3705 
3706 	for (i = 0; i < MLX4_MAX_PORTS; i++) {
3707 		if (nvfs[i] + nvfs[2] > MLX4_MAX_NUM_VF_P_PORT) {
3708 			dev_err(&pdev->dev,
3709 				"Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3710 				nvfs[i] + nvfs[2], i + 1,
3711 				MLX4_MAX_NUM_VF_P_PORT);
3712 			err = -EINVAL;
3713 			goto err_disable_pdev;
3714 		}
3715 	}
3716 
3717 	/* Check for BARs. */
3718 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3719 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3720 		dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3721 			pci_dev_data, pci_resource_flags(pdev, 0));
3722 		err = -ENODEV;
3723 		goto err_disable_pdev;
3724 	}
3725 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3726 		dev_err(&pdev->dev, "Missing UAR, aborting\n");
3727 		err = -ENODEV;
3728 		goto err_disable_pdev;
3729 	}
3730 
3731 	err = pci_request_regions(pdev, DRV_NAME);
3732 	if (err) {
3733 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3734 		goto err_disable_pdev;
3735 	}
3736 
3737 	pci_set_master(pdev);
3738 
3739 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3740 	if (err) {
3741 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3742 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3743 		if (err) {
3744 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3745 			goto err_release_regions;
3746 		}
3747 	}
3748 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3749 	if (err) {
3750 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3751 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3752 		if (err) {
3753 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3754 			goto err_release_regions;
3755 		}
3756 	}
3757 
3758 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
3759 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3760 	/* Detect if this device is a virtual function */
3761 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3762 		/* When acting as pf, we normally skip vfs unless explicitly
3763 		 * requested to probe them.
3764 		 */
3765 		if (total_vfs) {
3766 			unsigned vfs_offset = 0;
3767 
3768 			for (i = 0; i < ARRAY_SIZE(nvfs) &&
3769 			     vfs_offset + nvfs[i] < extended_func_num(pdev);
3770 			     vfs_offset += nvfs[i], i++)
3771 				;
3772 			if (i == ARRAY_SIZE(nvfs)) {
3773 				err = -ENODEV;
3774 				goto err_release_regions;
3775 			}
3776 			if ((extended_func_num(pdev) - vfs_offset)
3777 			    > prb_vf[i]) {
3778 				dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3779 					 extended_func_num(pdev));
3780 				err = -ENODEV;
3781 				goto err_release_regions;
3782 			}
3783 		}
3784 	}
3785 
3786 	err = mlx4_catas_init(&priv->dev);
3787 	if (err)
3788 		goto err_release_regions;
3789 
3790 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3791 	if (err)
3792 		goto err_catas;
3793 
3794 	return 0;
3795 
3796 err_catas:
3797 	mlx4_catas_end(&priv->dev);
3798 
3799 err_release_regions:
3800 	pci_release_regions(pdev);
3801 
3802 err_disable_pdev:
3803 	mlx4_pci_disable_device(&priv->dev);
3804 	return err;
3805 }
3806 
3807 static int mlx4_devlink_port_type_set(struct devlink_port *devlink_port,
3808 				      enum devlink_port_type port_type)
3809 {
3810 	struct mlx4_port_info *info = container_of(devlink_port,
3811 						   struct mlx4_port_info,
3812 						   devlink_port);
3813 	enum mlx4_port_type mlx4_port_type;
3814 
3815 	switch (port_type) {
3816 	case DEVLINK_PORT_TYPE_AUTO:
3817 		mlx4_port_type = MLX4_PORT_TYPE_AUTO;
3818 		break;
3819 	case DEVLINK_PORT_TYPE_ETH:
3820 		mlx4_port_type = MLX4_PORT_TYPE_ETH;
3821 		break;
3822 	case DEVLINK_PORT_TYPE_IB:
3823 		mlx4_port_type = MLX4_PORT_TYPE_IB;
3824 		break;
3825 	default:
3826 		return -EOPNOTSUPP;
3827 	}
3828 
3829 	return __set_port_type(info, mlx4_port_type);
3830 }
3831 
3832 static const struct devlink_ops mlx4_devlink_ops = {
3833 	.port_type_set	= mlx4_devlink_port_type_set,
3834 };
3835 
3836 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3837 {
3838 	struct devlink *devlink;
3839 	struct mlx4_priv *priv;
3840 	struct mlx4_dev *dev;
3841 	int ret;
3842 
3843 	printk_once(KERN_INFO "%s", mlx4_version);
3844 
3845 	devlink = devlink_alloc(&mlx4_devlink_ops, sizeof(*priv));
3846 	if (!devlink)
3847 		return -ENOMEM;
3848 	priv = devlink_priv(devlink);
3849 
3850 	dev       = &priv->dev;
3851 	dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3852 	if (!dev->persist) {
3853 		ret = -ENOMEM;
3854 		goto err_devlink_free;
3855 	}
3856 	dev->persist->pdev = pdev;
3857 	dev->persist->dev = dev;
3858 	pci_set_drvdata(pdev, dev->persist);
3859 	priv->pci_dev_data = id->driver_data;
3860 	mutex_init(&dev->persist->device_state_mutex);
3861 	mutex_init(&dev->persist->interface_state_mutex);
3862 	mutex_init(&dev->persist->pci_status_mutex);
3863 
3864 	ret = devlink_register(devlink, &pdev->dev);
3865 	if (ret)
3866 		goto err_persist_free;
3867 
3868 	ret =  __mlx4_init_one(pdev, id->driver_data, priv);
3869 	if (ret)
3870 		goto err_devlink_unregister;
3871 
3872 	pci_save_state(pdev);
3873 	return 0;
3874 
3875 err_devlink_unregister:
3876 	devlink_unregister(devlink);
3877 err_persist_free:
3878 	kfree(dev->persist);
3879 err_devlink_free:
3880 	devlink_free(devlink);
3881 	return ret;
3882 }
3883 
3884 static void mlx4_clean_dev(struct mlx4_dev *dev)
3885 {
3886 	struct mlx4_dev_persistent *persist = dev->persist;
3887 	struct mlx4_priv *priv = mlx4_priv(dev);
3888 	unsigned long	flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
3889 
3890 	memset(priv, 0, sizeof(*priv));
3891 	priv->dev.persist = persist;
3892 	priv->dev.flags = flags;
3893 }
3894 
3895 static void mlx4_unload_one(struct pci_dev *pdev)
3896 {
3897 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3898 	struct mlx4_dev  *dev  = persist->dev;
3899 	struct mlx4_priv *priv = mlx4_priv(dev);
3900 	int               pci_dev_data;
3901 	int p, i;
3902 
3903 	if (priv->removed)
3904 		return;
3905 
3906 	/* saving current ports type for further use */
3907 	for (i = 0; i < dev->caps.num_ports; i++) {
3908 		dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
3909 		dev->persist->curr_port_poss_type[i] = dev->caps.
3910 						       possible_type[i + 1];
3911 	}
3912 
3913 	pci_dev_data = priv->pci_dev_data;
3914 
3915 	mlx4_stop_sense(dev);
3916 	mlx4_unregister_device(dev);
3917 
3918 	for (p = 1; p <= dev->caps.num_ports; p++) {
3919 		mlx4_cleanup_port_info(&priv->port[p]);
3920 		mlx4_CLOSE_PORT(dev, p);
3921 	}
3922 
3923 	if (mlx4_is_master(dev))
3924 		mlx4_free_resource_tracker(dev,
3925 					   RES_TR_FREE_SLAVES_ONLY);
3926 
3927 	mlx4_cleanup_default_counters(dev);
3928 	if (!mlx4_is_slave(dev))
3929 		mlx4_cleanup_counters_table(dev);
3930 	mlx4_cleanup_qp_table(dev);
3931 	mlx4_cleanup_srq_table(dev);
3932 	mlx4_cleanup_cq_table(dev);
3933 	mlx4_cmd_use_polling(dev);
3934 	mlx4_cleanup_eq_table(dev);
3935 	mlx4_cleanup_mcg_table(dev);
3936 	mlx4_cleanup_mr_table(dev);
3937 	mlx4_cleanup_xrcd_table(dev);
3938 	mlx4_cleanup_pd_table(dev);
3939 
3940 	if (mlx4_is_master(dev))
3941 		mlx4_free_resource_tracker(dev,
3942 					   RES_TR_FREE_STRUCTS_ONLY);
3943 
3944 	iounmap(priv->kar);
3945 	mlx4_uar_free(dev, &priv->driver_uar);
3946 	mlx4_cleanup_uar_table(dev);
3947 	if (!mlx4_is_slave(dev))
3948 		mlx4_clear_steering(dev);
3949 	mlx4_free_eq_table(dev);
3950 	if (mlx4_is_master(dev))
3951 		mlx4_multi_func_cleanup(dev);
3952 	mlx4_close_hca(dev);
3953 	mlx4_close_fw(dev);
3954 	if (mlx4_is_slave(dev))
3955 		mlx4_multi_func_cleanup(dev);
3956 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3957 
3958 	if (dev->flags & MLX4_FLAG_MSI_X)
3959 		pci_disable_msix(pdev);
3960 
3961 	if (!mlx4_is_slave(dev))
3962 		mlx4_free_ownership(dev);
3963 
3964 	mlx4_slave_destroy_special_qp_cap(dev);
3965 	kfree(dev->dev_vfs);
3966 
3967 	mlx4_clean_dev(dev);
3968 	priv->pci_dev_data = pci_dev_data;
3969 	priv->removed = 1;
3970 }
3971 
3972 static void mlx4_remove_one(struct pci_dev *pdev)
3973 {
3974 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3975 	struct mlx4_dev  *dev  = persist->dev;
3976 	struct mlx4_priv *priv = mlx4_priv(dev);
3977 	struct devlink *devlink = priv_to_devlink(priv);
3978 	int active_vfs = 0;
3979 
3980 	if (mlx4_is_slave(dev))
3981 		persist->interface_state |= MLX4_INTERFACE_STATE_NOWAIT;
3982 
3983 	mutex_lock(&persist->interface_state_mutex);
3984 	persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
3985 	mutex_unlock(&persist->interface_state_mutex);
3986 
3987 	/* Disabling SR-IOV is not allowed while there are active vf's */
3988 	if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
3989 		active_vfs = mlx4_how_many_lives_vf(dev);
3990 		if (active_vfs) {
3991 			pr_warn("Removing PF when there are active VF's !!\n");
3992 			pr_warn("Will not disable SR-IOV.\n");
3993 		}
3994 	}
3995 
3996 	/* device marked to be under deletion running now without the lock
3997 	 * letting other tasks to be terminated
3998 	 */
3999 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4000 		mlx4_unload_one(pdev);
4001 	else
4002 		mlx4_info(dev, "%s: interface is down\n", __func__);
4003 	mlx4_catas_end(dev);
4004 	if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
4005 		mlx4_warn(dev, "Disabling SR-IOV\n");
4006 		pci_disable_sriov(pdev);
4007 	}
4008 
4009 	pci_release_regions(pdev);
4010 	mlx4_pci_disable_device(dev);
4011 	devlink_unregister(devlink);
4012 	kfree(dev->persist);
4013 	devlink_free(devlink);
4014 }
4015 
4016 static int restore_current_port_types(struct mlx4_dev *dev,
4017 				      enum mlx4_port_type *types,
4018 				      enum mlx4_port_type *poss_types)
4019 {
4020 	struct mlx4_priv *priv = mlx4_priv(dev);
4021 	int err, i;
4022 
4023 	mlx4_stop_sense(dev);
4024 
4025 	mutex_lock(&priv->port_mutex);
4026 	for (i = 0; i < dev->caps.num_ports; i++)
4027 		dev->caps.possible_type[i + 1] = poss_types[i];
4028 	err = mlx4_change_port_types(dev, types);
4029 	mlx4_start_sense(dev);
4030 	mutex_unlock(&priv->port_mutex);
4031 
4032 	return err;
4033 }
4034 
4035 int mlx4_restart_one(struct pci_dev *pdev)
4036 {
4037 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4038 	struct mlx4_dev	 *dev  = persist->dev;
4039 	struct mlx4_priv *priv = mlx4_priv(dev);
4040 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4041 	int pci_dev_data, err, total_vfs;
4042 
4043 	pci_dev_data = priv->pci_dev_data;
4044 	total_vfs = dev->persist->num_vfs;
4045 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4046 
4047 	mlx4_unload_one(pdev);
4048 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
4049 	if (err) {
4050 		mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
4051 			 __func__, pci_name(pdev), err);
4052 		return err;
4053 	}
4054 
4055 	err = restore_current_port_types(dev, dev->persist->curr_port_type,
4056 					 dev->persist->curr_port_poss_type);
4057 	if (err)
4058 		mlx4_err(dev, "could not restore original port types (%d)\n",
4059 			 err);
4060 
4061 	return err;
4062 }
4063 
4064 #define MLX_SP(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_FORCE_SENSE_PORT }
4065 #define MLX_VF(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_IS_VF }
4066 #define MLX_GN(id) { PCI_VDEVICE(MELLANOX, id), 0 }
4067 
4068 static const struct pci_device_id mlx4_pci_table[] = {
4069 #ifdef CONFIG_MLX4_CORE_GEN2
4070 	/* MT25408 "Hermon" */
4071 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_SDR),	/* SDR */
4072 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR),	/* DDR */
4073 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR),	/* QDR */
4074 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR_GEN2), /* DDR Gen2 */
4075 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR_GEN2),	/* QDR Gen2 */
4076 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN),	/* EN 10GigE */
4077 	MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN_GEN2),  /* EN 10GigE Gen2 */
4078 	/* MT25458 ConnectX EN 10GBASE-T */
4079 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN),
4080 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_T_GEN2),	/* Gen2 */
4081 	/* MT26468 ConnectX EN 10GigE PCIe Gen2*/
4082 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_GEN2),
4083 	/* MT26438 ConnectX EN 40GigE PCIe Gen2 5GT/s */
4084 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_5_GEN2),
4085 	/* MT26478 ConnectX2 40GigE PCIe Gen2 */
4086 	MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX2),
4087 	/* MT25400 Family [ConnectX-2] */
4088 	MLX_VF(0x1002),					/* Virtual Function */
4089 #endif /* CONFIG_MLX4_CORE_GEN2 */
4090 	/* MT27500 Family [ConnectX-3] */
4091 	MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3),
4092 	MLX_VF(0x1004),					/* Virtual Function */
4093 	MLX_GN(0x1005),					/* MT27510 Family */
4094 	MLX_GN(0x1006),					/* MT27511 Family */
4095 	MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3_PRO),	/* MT27520 Family */
4096 	MLX_GN(0x1008),					/* MT27521 Family */
4097 	MLX_GN(0x1009),					/* MT27530 Family */
4098 	MLX_GN(0x100a),					/* MT27531 Family */
4099 	MLX_GN(0x100b),					/* MT27540 Family */
4100 	MLX_GN(0x100c),					/* MT27541 Family */
4101 	MLX_GN(0x100d),					/* MT27550 Family */
4102 	MLX_GN(0x100e),					/* MT27551 Family */
4103 	MLX_GN(0x100f),					/* MT27560 Family */
4104 	MLX_GN(0x1010),					/* MT27561 Family */
4105 
4106 	/*
4107 	 * See the mellanox_check_broken_intx_masking() quirk when
4108 	 * adding devices
4109 	 */
4110 
4111 	{ 0, }
4112 };
4113 
4114 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
4115 
4116 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
4117 					      pci_channel_state_t state)
4118 {
4119 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4120 
4121 	mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
4122 	mlx4_enter_error_state(persist);
4123 
4124 	mutex_lock(&persist->interface_state_mutex);
4125 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4126 		mlx4_unload_one(pdev);
4127 
4128 	mutex_unlock(&persist->interface_state_mutex);
4129 	if (state == pci_channel_io_perm_failure)
4130 		return PCI_ERS_RESULT_DISCONNECT;
4131 
4132 	mlx4_pci_disable_device(persist->dev);
4133 	return PCI_ERS_RESULT_NEED_RESET;
4134 }
4135 
4136 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
4137 {
4138 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4139 	struct mlx4_dev	 *dev  = persist->dev;
4140 	int err;
4141 
4142 	mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
4143 	err = mlx4_pci_enable_device(dev);
4144 	if (err) {
4145 		mlx4_err(dev, "Can not re-enable device, err=%d\n", err);
4146 		return PCI_ERS_RESULT_DISCONNECT;
4147 	}
4148 
4149 	pci_set_master(pdev);
4150 	pci_restore_state(pdev);
4151 	pci_save_state(pdev);
4152 	return PCI_ERS_RESULT_RECOVERED;
4153 }
4154 
4155 static void mlx4_pci_resume(struct pci_dev *pdev)
4156 {
4157 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4158 	struct mlx4_dev	 *dev  = persist->dev;
4159 	struct mlx4_priv *priv = mlx4_priv(dev);
4160 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4161 	int total_vfs;
4162 	int err;
4163 
4164 	mlx4_err(dev, "%s was called\n", __func__);
4165 	total_vfs = dev->persist->num_vfs;
4166 	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4167 
4168 	mutex_lock(&persist->interface_state_mutex);
4169 	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4170 		err = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
4171 				    priv, 1);
4172 		if (err) {
4173 			mlx4_err(dev, "%s: mlx4_load_one failed, err=%d\n",
4174 				 __func__,  err);
4175 			goto end;
4176 		}
4177 
4178 		err = restore_current_port_types(dev, dev->persist->
4179 						 curr_port_type, dev->persist->
4180 						 curr_port_poss_type);
4181 		if (err)
4182 			mlx4_err(dev, "could not restore original port types (%d)\n", err);
4183 	}
4184 end:
4185 	mutex_unlock(&persist->interface_state_mutex);
4186 
4187 }
4188 
4189 static void mlx4_shutdown(struct pci_dev *pdev)
4190 {
4191 	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4192 
4193 	mlx4_info(persist->dev, "mlx4_shutdown was called\n");
4194 	mutex_lock(&persist->interface_state_mutex);
4195 	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4196 		mlx4_unload_one(pdev);
4197 	mutex_unlock(&persist->interface_state_mutex);
4198 }
4199 
4200 static const struct pci_error_handlers mlx4_err_handler = {
4201 	.error_detected = mlx4_pci_err_detected,
4202 	.slot_reset     = mlx4_pci_slot_reset,
4203 	.resume		= mlx4_pci_resume,
4204 };
4205 
4206 static struct pci_driver mlx4_driver = {
4207 	.name		= DRV_NAME,
4208 	.id_table	= mlx4_pci_table,
4209 	.probe		= mlx4_init_one,
4210 	.shutdown	= mlx4_shutdown,
4211 	.remove		= mlx4_remove_one,
4212 	.err_handler    = &mlx4_err_handler,
4213 };
4214 
4215 static int __init mlx4_verify_params(void)
4216 {
4217 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
4218 		pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
4219 		return -1;
4220 	}
4221 
4222 	if (log_num_vlan != 0)
4223 		pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
4224 			MLX4_LOG_NUM_VLANS);
4225 
4226 	if (use_prio != 0)
4227 		pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
4228 
4229 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
4230 		pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
4231 			log_mtts_per_seg);
4232 		return -1;
4233 	}
4234 
4235 	/* Check if module param for ports type has legal combination */
4236 	if (port_type_array[0] == false && port_type_array[1] == true) {
4237 		pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
4238 		port_type_array[0] = true;
4239 	}
4240 
4241 	if (mlx4_log_num_mgm_entry_size < -7 ||
4242 	    (mlx4_log_num_mgm_entry_size > 0 &&
4243 	     (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
4244 	      mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
4245 		pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
4246 			mlx4_log_num_mgm_entry_size,
4247 			MLX4_MIN_MGM_LOG_ENTRY_SIZE,
4248 			MLX4_MAX_MGM_LOG_ENTRY_SIZE);
4249 		return -1;
4250 	}
4251 
4252 	return 0;
4253 }
4254 
4255 static int __init mlx4_init(void)
4256 {
4257 	int ret;
4258 
4259 	if (mlx4_verify_params())
4260 		return -EINVAL;
4261 
4262 
4263 	mlx4_wq = create_singlethread_workqueue("mlx4");
4264 	if (!mlx4_wq)
4265 		return -ENOMEM;
4266 
4267 	ret = pci_register_driver(&mlx4_driver);
4268 	if (ret < 0)
4269 		destroy_workqueue(mlx4_wq);
4270 	return ret < 0 ? ret : 0;
4271 }
4272 
4273 static void __exit mlx4_cleanup(void)
4274 {
4275 	pci_unregister_driver(&mlx4_driver);
4276 	destroy_workqueue(mlx4_wq);
4277 }
4278 
4279 module_init(mlx4_init);
4280 module_exit(mlx4_cleanup);
4281