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