1 /*
2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4  * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
5  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * OpenIB.org BSD license below:
12  *
13  *     Redistribution and use in source and binary forms, with or
14  *     without modification, are permitted provided that the following
15  *     conditions are met:
16  *
17  *      - Redistributions of source code must retain the above
18  *        copyright notice, this list of conditions and the following
19  *        disclaimer.
20  *
21  *      - Redistributions in binary form must reproduce the above
22  *        copyright notice, this list of conditions and the following
23  *        disclaimer in the documentation and/or other materials
24  *        provided with the distribution.
25  *
26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33  * SOFTWARE.
34  */
35 
36 #include <linux/module.h>
37 #include <linux/init.h>
38 #include <linux/errno.h>
39 #include <linux/pci.h>
40 #include <linux/dma-mapping.h>
41 #include <linux/slab.h>
42 #include <linux/io-mapping.h>
43 #include <linux/delay.h>
44 #include <linux/kmod.h>
45 
46 #include <linux/mlx4/device.h>
47 #include <linux/mlx4/doorbell.h>
48 
49 #include "mlx4.h"
50 #include "fw.h"
51 #include "icm.h"
52 
53 MODULE_AUTHOR("Roland Dreier");
54 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
55 MODULE_LICENSE("Dual BSD/GPL");
56 MODULE_VERSION(DRV_VERSION);
57 
58 struct workqueue_struct *mlx4_wq;
59 
60 #ifdef CONFIG_MLX4_DEBUG
61 
62 int mlx4_debug_level = 0;
63 module_param_named(debug_level, mlx4_debug_level, int, 0644);
64 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
65 
66 #endif /* CONFIG_MLX4_DEBUG */
67 
68 #ifdef CONFIG_PCI_MSI
69 
70 static int msi_x = 1;
71 module_param(msi_x, int, 0444);
72 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
73 
74 #else /* CONFIG_PCI_MSI */
75 
76 #define msi_x (0)
77 
78 #endif /* CONFIG_PCI_MSI */
79 
80 static uint8_t num_vfs[3] = {0, 0, 0};
81 static int num_vfs_argc;
82 module_param_array(num_vfs, byte , &num_vfs_argc, 0444);
83 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
84 			  "num_vfs=port1,port2,port1+2");
85 
86 static uint8_t probe_vf[3] = {0, 0, 0};
87 static int probe_vfs_argc;
88 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
89 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
90 			   "probe_vf=port1,port2,port1+2");
91 
92 int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
93 module_param_named(log_num_mgm_entry_size,
94 			mlx4_log_num_mgm_entry_size, int, 0444);
95 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
96 					 " of qp per mcg, for example:"
97 					 " 10 gives 248.range: 7 <="
98 					 " log_num_mgm_entry_size <= 12."
99 					 " To activate device managed"
100 					 " flow steering when available, set to -1");
101 
102 static bool enable_64b_cqe_eqe = true;
103 module_param(enable_64b_cqe_eqe, bool, 0444);
104 MODULE_PARM_DESC(enable_64b_cqe_eqe,
105 		 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
106 
107 #define PF_CONTEXT_BEHAVIOUR_MASK	(MLX4_FUNC_CAP_64B_EQE_CQE | \
108 					 MLX4_FUNC_CAP_EQE_CQE_STRIDE)
109 
110 static char mlx4_version[] =
111 	DRV_NAME ": Mellanox ConnectX core driver v"
112 	DRV_VERSION " (" DRV_RELDATE ")\n";
113 
114 static struct mlx4_profile default_profile = {
115 	.num_qp		= 1 << 18,
116 	.num_srq	= 1 << 16,
117 	.rdmarc_per_qp	= 1 << 4,
118 	.num_cq		= 1 << 16,
119 	.num_mcg	= 1 << 13,
120 	.num_mpt	= 1 << 19,
121 	.num_mtt	= 1 << 20, /* It is really num mtt segements */
122 };
123 
124 static struct mlx4_profile low_mem_profile = {
125 	.num_qp		= 1 << 17,
126 	.num_srq	= 1 << 6,
127 	.rdmarc_per_qp	= 1 << 4,
128 	.num_cq		= 1 << 8,
129 	.num_mcg	= 1 << 8,
130 	.num_mpt	= 1 << 9,
131 	.num_mtt	= 1 << 7,
132 };
133 
134 static int log_num_mac = 7;
135 module_param_named(log_num_mac, log_num_mac, int, 0444);
136 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
137 
138 static int log_num_vlan;
139 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
140 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
141 /* Log2 max number of VLANs per ETH port (0-7) */
142 #define MLX4_LOG_NUM_VLANS 7
143 #define MLX4_MIN_LOG_NUM_VLANS 0
144 #define MLX4_MIN_LOG_NUM_MAC 1
145 
146 static bool use_prio;
147 module_param_named(use_prio, use_prio, bool, 0444);
148 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
149 
150 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
151 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
152 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
153 
154 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
155 static int arr_argc = 2;
156 module_param_array(port_type_array, int, &arr_argc, 0444);
157 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
158 				"1 for IB, 2 for Ethernet");
159 
160 struct mlx4_port_config {
161 	struct list_head list;
162 	enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
163 	struct pci_dev *pdev;
164 };
165 
166 static atomic_t pf_loading = ATOMIC_INIT(0);
167 
168 int mlx4_check_port_params(struct mlx4_dev *dev,
169 			   enum mlx4_port_type *port_type)
170 {
171 	int i;
172 
173 	for (i = 0; i < dev->caps.num_ports - 1; i++) {
174 		if (port_type[i] != port_type[i + 1]) {
175 			if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
176 				mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
177 				return -EINVAL;
178 			}
179 		}
180 	}
181 
182 	for (i = 0; i < dev->caps.num_ports; i++) {
183 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
184 			mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
185 				 i + 1);
186 			return -EINVAL;
187 		}
188 	}
189 	return 0;
190 }
191 
192 static void mlx4_set_port_mask(struct mlx4_dev *dev)
193 {
194 	int i;
195 
196 	for (i = 1; i <= dev->caps.num_ports; ++i)
197 		dev->caps.port_mask[i] = dev->caps.port_type[i];
198 }
199 
200 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
201 {
202 	struct mlx4_caps *dev_cap = &dev->caps;
203 
204 	/* FW not supporting or cancelled by user */
205 	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
206 	    !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
207 		return;
208 
209 	/* Must have 64B CQE_EQE enabled by FW to use bigger stride
210 	 * When FW has NCSI it may decide not to report 64B CQE/EQEs
211 	 */
212 	if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
213 	    !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
214 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
215 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
216 		return;
217 	}
218 
219 	if (cache_line_size() == 128 || cache_line_size() == 256) {
220 		mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
221 		/* Changing the real data inside CQE size to 32B */
222 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
223 		dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
224 
225 		if (mlx4_is_master(dev))
226 			dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
227 	} else {
228 		mlx4_dbg(dev, "Disabling CQE stride cacheLine unsupported\n");
229 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
230 		dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
231 	}
232 }
233 
234 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
235 {
236 	int err;
237 	int i;
238 
239 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
240 	if (err) {
241 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
242 		return err;
243 	}
244 
245 	if (dev_cap->min_page_sz > PAGE_SIZE) {
246 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
247 			 dev_cap->min_page_sz, PAGE_SIZE);
248 		return -ENODEV;
249 	}
250 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
251 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
252 			 dev_cap->num_ports, MLX4_MAX_PORTS);
253 		return -ENODEV;
254 	}
255 
256 	if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
257 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
258 			 dev_cap->uar_size,
259 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
260 		return -ENODEV;
261 	}
262 
263 	dev->caps.num_ports	     = dev_cap->num_ports;
264 	dev->phys_caps.num_phys_eqs  = MLX4_MAX_EQ_NUM;
265 	for (i = 1; i <= dev->caps.num_ports; ++i) {
266 		dev->caps.vl_cap[i]	    = dev_cap->max_vl[i];
267 		dev->caps.ib_mtu_cap[i]	    = dev_cap->ib_mtu[i];
268 		dev->phys_caps.gid_phys_table_len[i]  = dev_cap->max_gids[i];
269 		dev->phys_caps.pkey_phys_table_len[i] = dev_cap->max_pkeys[i];
270 		/* set gid and pkey table operating lengths by default
271 		 * to non-sriov values */
272 		dev->caps.gid_table_len[i]  = dev_cap->max_gids[i];
273 		dev->caps.pkey_table_len[i] = dev_cap->max_pkeys[i];
274 		dev->caps.port_width_cap[i] = dev_cap->max_port_width[i];
275 		dev->caps.eth_mtu_cap[i]    = dev_cap->eth_mtu[i];
276 		dev->caps.def_mac[i]        = dev_cap->def_mac[i];
277 		dev->caps.supported_type[i] = dev_cap->supported_port_types[i];
278 		dev->caps.suggested_type[i] = dev_cap->suggested_type[i];
279 		dev->caps.default_sense[i] = dev_cap->default_sense[i];
280 		dev->caps.trans_type[i]	    = dev_cap->trans_type[i];
281 		dev->caps.vendor_oui[i]     = dev_cap->vendor_oui[i];
282 		dev->caps.wavelength[i]     = dev_cap->wavelength[i];
283 		dev->caps.trans_code[i]     = dev_cap->trans_code[i];
284 	}
285 
286 	dev->caps.uar_page_size	     = PAGE_SIZE;
287 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
288 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
289 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
290 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
291 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
292 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
293 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
294 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
295 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
296 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
297 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
298 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
299 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
300 	/*
301 	 * Subtract 1 from the limit because we need to allocate a
302 	 * spare CQE so the HCA HW can tell the difference between an
303 	 * empty CQ and a full CQ.
304 	 */
305 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
306 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
307 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
308 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
309 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
310 
311 	/* The first 128 UARs are used for EQ doorbells */
312 	dev->caps.reserved_uars	     = max_t(int, 128, dev_cap->reserved_uars);
313 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
314 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
315 					dev_cap->reserved_xrcds : 0;
316 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
317 					dev_cap->max_xrcds : 0;
318 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
319 
320 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
321 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
322 	dev->caps.flags		     = dev_cap->flags;
323 	dev->caps.flags2	     = dev_cap->flags2;
324 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
325 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
326 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
327 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
328 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
329 
330 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
331 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
332 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
333 	/* Don't do sense port on multifunction devices (for now at least) */
334 	if (mlx4_is_mfunc(dev))
335 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
336 
337 	if (mlx4_low_memory_profile()) {
338 		dev->caps.log_num_macs  = MLX4_MIN_LOG_NUM_MAC;
339 		dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
340 	} else {
341 		dev->caps.log_num_macs  = log_num_mac;
342 		dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
343 	}
344 
345 	for (i = 1; i <= dev->caps.num_ports; ++i) {
346 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
347 		if (dev->caps.supported_type[i]) {
348 			/* if only ETH is supported - assign ETH */
349 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
350 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
351 			/* if only IB is supported, assign IB */
352 			else if (dev->caps.supported_type[i] ==
353 				 MLX4_PORT_TYPE_IB)
354 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
355 			else {
356 				/* if IB and ETH are supported, we set the port
357 				 * type according to user selection of port type;
358 				 * if user selected none, take the FW hint */
359 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
360 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
361 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
362 				else
363 					dev->caps.port_type[i] = port_type_array[i - 1];
364 			}
365 		}
366 		/*
367 		 * Link sensing is allowed on the port if 3 conditions are true:
368 		 * 1. Both protocols are supported on the port.
369 		 * 2. Different types are supported on the port
370 		 * 3. FW declared that it supports link sensing
371 		 */
372 		mlx4_priv(dev)->sense.sense_allowed[i] =
373 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
374 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
375 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
376 
377 		/*
378 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
379 		 * and perform sense_port FW command to try and set the correct
380 		 * port type from beginning
381 		 */
382 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
383 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
384 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
385 			mlx4_SENSE_PORT(dev, i, &sensed_port);
386 			if (sensed_port != MLX4_PORT_TYPE_NONE)
387 				dev->caps.port_type[i] = sensed_port;
388 		} else {
389 			dev->caps.possible_type[i] = dev->caps.port_type[i];
390 		}
391 
392 		if (dev->caps.log_num_macs > dev_cap->log_max_macs[i]) {
393 			dev->caps.log_num_macs = dev_cap->log_max_macs[i];
394 			mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
395 				  i, 1 << dev->caps.log_num_macs);
396 		}
397 		if (dev->caps.log_num_vlans > dev_cap->log_max_vlans[i]) {
398 			dev->caps.log_num_vlans = dev_cap->log_max_vlans[i];
399 			mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
400 				  i, 1 << dev->caps.log_num_vlans);
401 		}
402 	}
403 
404 	dev->caps.max_counters = 1 << ilog2(dev_cap->max_counters);
405 
406 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
407 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
408 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
409 		(1 << dev->caps.log_num_macs) *
410 		(1 << dev->caps.log_num_vlans) *
411 		dev->caps.num_ports;
412 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
413 
414 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
415 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
416 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
417 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
418 
419 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
420 
421 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
422 		if (dev_cap->flags &
423 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
424 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
425 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
426 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
427 		}
428 
429 		if (dev_cap->flags2 &
430 		    (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
431 		     MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
432 			mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
433 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
434 			dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
435 		}
436 	}
437 
438 	if ((dev->caps.flags &
439 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
440 	    mlx4_is_master(dev))
441 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
442 
443 	if (!mlx4_is_slave(dev))
444 		mlx4_enable_cqe_eqe_stride(dev);
445 
446 	return 0;
447 }
448 
449 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
450 				       enum pci_bus_speed *speed,
451 				       enum pcie_link_width *width)
452 {
453 	u32 lnkcap1, lnkcap2;
454 	int err1, err2;
455 
456 #define  PCIE_MLW_CAP_SHIFT 4	/* start of MLW mask in link capabilities */
457 
458 	*speed = PCI_SPEED_UNKNOWN;
459 	*width = PCIE_LNK_WIDTH_UNKNOWN;
460 
461 	err1 = pcie_capability_read_dword(dev->pdev, PCI_EXP_LNKCAP, &lnkcap1);
462 	err2 = pcie_capability_read_dword(dev->pdev, PCI_EXP_LNKCAP2, &lnkcap2);
463 	if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
464 		if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
465 			*speed = PCIE_SPEED_8_0GT;
466 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
467 			*speed = PCIE_SPEED_5_0GT;
468 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
469 			*speed = PCIE_SPEED_2_5GT;
470 	}
471 	if (!err1) {
472 		*width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
473 		if (!lnkcap2) { /* pre-r3.0 */
474 			if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
475 				*speed = PCIE_SPEED_5_0GT;
476 			else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
477 				*speed = PCIE_SPEED_2_5GT;
478 		}
479 	}
480 
481 	if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
482 		return err1 ? err1 :
483 			err2 ? err2 : -EINVAL;
484 	}
485 	return 0;
486 }
487 
488 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
489 {
490 	enum pcie_link_width width, width_cap;
491 	enum pci_bus_speed speed, speed_cap;
492 	int err;
493 
494 #define PCIE_SPEED_STR(speed) \
495 	(speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
496 	 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
497 	 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
498 	 "Unknown")
499 
500 	err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
501 	if (err) {
502 		mlx4_warn(dev,
503 			  "Unable to determine PCIe device BW capabilities\n");
504 		return;
505 	}
506 
507 	err = pcie_get_minimum_link(dev->pdev, &speed, &width);
508 	if (err || speed == PCI_SPEED_UNKNOWN ||
509 	    width == PCIE_LNK_WIDTH_UNKNOWN) {
510 		mlx4_warn(dev,
511 			  "Unable to determine PCI device chain minimum BW\n");
512 		return;
513 	}
514 
515 	if (width != width_cap || speed != speed_cap)
516 		mlx4_warn(dev,
517 			  "PCIe BW is different than device's capability\n");
518 
519 	mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
520 		  PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
521 	mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
522 		  width, width_cap);
523 	return;
524 }
525 
526 /*The function checks if there are live vf, return the num of them*/
527 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
528 {
529 	struct mlx4_priv *priv = mlx4_priv(dev);
530 	struct mlx4_slave_state *s_state;
531 	int i;
532 	int ret = 0;
533 
534 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
535 		s_state = &priv->mfunc.master.slave_state[i];
536 		if (s_state->active && s_state->last_cmd !=
537 		    MLX4_COMM_CMD_RESET) {
538 			mlx4_warn(dev, "%s: slave: %d is still active\n",
539 				  __func__, i);
540 			ret++;
541 		}
542 	}
543 	return ret;
544 }
545 
546 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
547 {
548 	u32 qk = MLX4_RESERVED_QKEY_BASE;
549 
550 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
551 	    qpn < dev->phys_caps.base_proxy_sqpn)
552 		return -EINVAL;
553 
554 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
555 		/* tunnel qp */
556 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
557 	else
558 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
559 	*qkey = qk;
560 	return 0;
561 }
562 EXPORT_SYMBOL(mlx4_get_parav_qkey);
563 
564 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
565 {
566 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
567 
568 	if (!mlx4_is_master(dev))
569 		return;
570 
571 	priv->virt2phys_pkey[slave][port - 1][i] = val;
572 }
573 EXPORT_SYMBOL(mlx4_sync_pkey_table);
574 
575 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
576 {
577 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
578 
579 	if (!mlx4_is_master(dev))
580 		return;
581 
582 	priv->slave_node_guids[slave] = guid;
583 }
584 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
585 
586 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
587 {
588 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
589 
590 	if (!mlx4_is_master(dev))
591 		return 0;
592 
593 	return priv->slave_node_guids[slave];
594 }
595 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
596 
597 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
598 {
599 	struct mlx4_priv *priv = mlx4_priv(dev);
600 	struct mlx4_slave_state *s_slave;
601 
602 	if (!mlx4_is_master(dev))
603 		return 0;
604 
605 	s_slave = &priv->mfunc.master.slave_state[slave];
606 	return !!s_slave->active;
607 }
608 EXPORT_SYMBOL(mlx4_is_slave_active);
609 
610 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
611 				       struct mlx4_dev_cap *dev_cap,
612 				       struct mlx4_init_hca_param *hca_param)
613 {
614 	dev->caps.steering_mode = hca_param->steering_mode;
615 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
616 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
617 		dev->caps.fs_log_max_ucast_qp_range_size =
618 			dev_cap->fs_log_max_ucast_qp_range_size;
619 	} else
620 		dev->caps.num_qp_per_mgm =
621 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
622 
623 	mlx4_dbg(dev, "Steering mode is: %s\n",
624 		 mlx4_steering_mode_str(dev->caps.steering_mode));
625 }
626 
627 static int mlx4_slave_cap(struct mlx4_dev *dev)
628 {
629 	int			   err;
630 	u32			   page_size;
631 	struct mlx4_dev_cap	   dev_cap;
632 	struct mlx4_func_cap	   func_cap;
633 	struct mlx4_init_hca_param hca_param;
634 	int			   i;
635 
636 	memset(&hca_param, 0, sizeof(hca_param));
637 	err = mlx4_QUERY_HCA(dev, &hca_param);
638 	if (err) {
639 		mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
640 		return err;
641 	}
642 
643 	/* fail if the hca has an unknown global capability
644 	 * at this time global_caps should be always zeroed
645 	 */
646 	if (hca_param.global_caps) {
647 		mlx4_err(dev, "Unknown hca global capabilities\n");
648 		return -ENOSYS;
649 	}
650 
651 	mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
652 
653 	dev->caps.hca_core_clock = hca_param.hca_core_clock;
654 
655 	memset(&dev_cap, 0, sizeof(dev_cap));
656 	dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
657 	err = mlx4_dev_cap(dev, &dev_cap);
658 	if (err) {
659 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
660 		return err;
661 	}
662 
663 	err = mlx4_QUERY_FW(dev);
664 	if (err)
665 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
666 
667 	page_size = ~dev->caps.page_size_cap + 1;
668 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
669 	if (page_size > PAGE_SIZE) {
670 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
671 			 page_size, PAGE_SIZE);
672 		return -ENODEV;
673 	}
674 
675 	/* slave gets uar page size from QUERY_HCA fw command */
676 	dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
677 
678 	/* TODO: relax this assumption */
679 	if (dev->caps.uar_page_size != PAGE_SIZE) {
680 		mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
681 			 dev->caps.uar_page_size, PAGE_SIZE);
682 		return -ENODEV;
683 	}
684 
685 	memset(&func_cap, 0, sizeof(func_cap));
686 	err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
687 	if (err) {
688 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
689 			 err);
690 		return err;
691 	}
692 
693 	if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
694 	    PF_CONTEXT_BEHAVIOUR_MASK) {
695 		mlx4_err(dev, "Unknown pf context behaviour\n");
696 		return -ENOSYS;
697 	}
698 
699 	dev->caps.num_ports		= func_cap.num_ports;
700 	dev->quotas.qp			= func_cap.qp_quota;
701 	dev->quotas.srq			= func_cap.srq_quota;
702 	dev->quotas.cq			= func_cap.cq_quota;
703 	dev->quotas.mpt			= func_cap.mpt_quota;
704 	dev->quotas.mtt			= func_cap.mtt_quota;
705 	dev->caps.num_qps		= 1 << hca_param.log_num_qps;
706 	dev->caps.num_srqs		= 1 << hca_param.log_num_srqs;
707 	dev->caps.num_cqs		= 1 << hca_param.log_num_cqs;
708 	dev->caps.num_mpts		= 1 << hca_param.log_mpt_sz;
709 	dev->caps.num_eqs		= func_cap.max_eq;
710 	dev->caps.reserved_eqs		= func_cap.reserved_eq;
711 	dev->caps.num_pds               = MLX4_NUM_PDS;
712 	dev->caps.num_mgms              = 0;
713 	dev->caps.num_amgms             = 0;
714 
715 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
716 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
717 			 dev->caps.num_ports, MLX4_MAX_PORTS);
718 		return -ENODEV;
719 	}
720 
721 	dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
722 	dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
723 	dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
724 	dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
725 	dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
726 
727 	if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
728 	    !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
729 	    !dev->caps.qp0_qkey) {
730 		err = -ENOMEM;
731 		goto err_mem;
732 	}
733 
734 	for (i = 1; i <= dev->caps.num_ports; ++i) {
735 		err = mlx4_QUERY_FUNC_CAP(dev, (u32) i, &func_cap);
736 		if (err) {
737 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
738 				 i, err);
739 			goto err_mem;
740 		}
741 		dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
742 		dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
743 		dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
744 		dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
745 		dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
746 		dev->caps.port_mask[i] = dev->caps.port_type[i];
747 		dev->caps.phys_port_id[i] = func_cap.phys_port_id;
748 		if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
749 						    &dev->caps.gid_table_len[i],
750 						    &dev->caps.pkey_table_len[i]))
751 			goto err_mem;
752 	}
753 
754 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
755 				       dev->caps.reserved_uars) >
756 				       pci_resource_len(dev->pdev, 2)) {
757 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
758 			 dev->caps.uar_page_size * dev->caps.num_uars,
759 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
760 		goto err_mem;
761 	}
762 
763 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
764 		dev->caps.eqe_size   = 64;
765 		dev->caps.eqe_factor = 1;
766 	} else {
767 		dev->caps.eqe_size   = 32;
768 		dev->caps.eqe_factor = 0;
769 	}
770 
771 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
772 		dev->caps.cqe_size   = 64;
773 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
774 	} else {
775 		dev->caps.cqe_size   = 32;
776 	}
777 
778 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
779 		dev->caps.eqe_size = hca_param.eqe_size;
780 		dev->caps.eqe_factor = 0;
781 	}
782 
783 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
784 		dev->caps.cqe_size = hca_param.cqe_size;
785 		/* User still need to know when CQE > 32B */
786 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
787 	}
788 
789 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
790 	mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
791 
792 	slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
793 
794 	return 0;
795 
796 err_mem:
797 	kfree(dev->caps.qp0_qkey);
798 	kfree(dev->caps.qp0_tunnel);
799 	kfree(dev->caps.qp0_proxy);
800 	kfree(dev->caps.qp1_tunnel);
801 	kfree(dev->caps.qp1_proxy);
802 	dev->caps.qp0_qkey = NULL;
803 	dev->caps.qp0_tunnel = NULL;
804 	dev->caps.qp0_proxy = NULL;
805 	dev->caps.qp1_tunnel = NULL;
806 	dev->caps.qp1_proxy = NULL;
807 
808 	return err;
809 }
810 
811 static void mlx4_request_modules(struct mlx4_dev *dev)
812 {
813 	int port;
814 	int has_ib_port = false;
815 	int has_eth_port = false;
816 #define EN_DRV_NAME	"mlx4_en"
817 #define IB_DRV_NAME	"mlx4_ib"
818 
819 	for (port = 1; port <= dev->caps.num_ports; port++) {
820 		if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
821 			has_ib_port = true;
822 		else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
823 			has_eth_port = true;
824 	}
825 
826 	if (has_eth_port)
827 		request_module_nowait(EN_DRV_NAME);
828 	if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
829 		request_module_nowait(IB_DRV_NAME);
830 }
831 
832 /*
833  * Change the port configuration of the device.
834  * Every user of this function must hold the port mutex.
835  */
836 int mlx4_change_port_types(struct mlx4_dev *dev,
837 			   enum mlx4_port_type *port_types)
838 {
839 	int err = 0;
840 	int change = 0;
841 	int port;
842 
843 	for (port = 0; port <  dev->caps.num_ports; port++) {
844 		/* Change the port type only if the new type is different
845 		 * from the current, and not set to Auto */
846 		if (port_types[port] != dev->caps.port_type[port + 1])
847 			change = 1;
848 	}
849 	if (change) {
850 		mlx4_unregister_device(dev);
851 		for (port = 1; port <= dev->caps.num_ports; port++) {
852 			mlx4_CLOSE_PORT(dev, port);
853 			dev->caps.port_type[port] = port_types[port - 1];
854 			err = mlx4_SET_PORT(dev, port, -1);
855 			if (err) {
856 				mlx4_err(dev, "Failed to set port %d, aborting\n",
857 					 port);
858 				goto out;
859 			}
860 		}
861 		mlx4_set_port_mask(dev);
862 		err = mlx4_register_device(dev);
863 		if (err) {
864 			mlx4_err(dev, "Failed to register device\n");
865 			goto out;
866 		}
867 		mlx4_request_modules(dev);
868 	}
869 
870 out:
871 	return err;
872 }
873 
874 static ssize_t show_port_type(struct device *dev,
875 			      struct device_attribute *attr,
876 			      char *buf)
877 {
878 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
879 						   port_attr);
880 	struct mlx4_dev *mdev = info->dev;
881 	char type[8];
882 
883 	sprintf(type, "%s",
884 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
885 		"ib" : "eth");
886 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
887 		sprintf(buf, "auto (%s)\n", type);
888 	else
889 		sprintf(buf, "%s\n", type);
890 
891 	return strlen(buf);
892 }
893 
894 static ssize_t set_port_type(struct device *dev,
895 			     struct device_attribute *attr,
896 			     const char *buf, size_t count)
897 {
898 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
899 						   port_attr);
900 	struct mlx4_dev *mdev = info->dev;
901 	struct mlx4_priv *priv = mlx4_priv(mdev);
902 	enum mlx4_port_type types[MLX4_MAX_PORTS];
903 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
904 	int i;
905 	int err = 0;
906 
907 	if (!strcmp(buf, "ib\n"))
908 		info->tmp_type = MLX4_PORT_TYPE_IB;
909 	else if (!strcmp(buf, "eth\n"))
910 		info->tmp_type = MLX4_PORT_TYPE_ETH;
911 	else if (!strcmp(buf, "auto\n"))
912 		info->tmp_type = MLX4_PORT_TYPE_AUTO;
913 	else {
914 		mlx4_err(mdev, "%s is not supported port type\n", buf);
915 		return -EINVAL;
916 	}
917 
918 	mlx4_stop_sense(mdev);
919 	mutex_lock(&priv->port_mutex);
920 	/* Possible type is always the one that was delivered */
921 	mdev->caps.possible_type[info->port] = info->tmp_type;
922 
923 	for (i = 0; i < mdev->caps.num_ports; i++) {
924 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
925 					mdev->caps.possible_type[i+1];
926 		if (types[i] == MLX4_PORT_TYPE_AUTO)
927 			types[i] = mdev->caps.port_type[i+1];
928 	}
929 
930 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
931 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
932 		for (i = 1; i <= mdev->caps.num_ports; i++) {
933 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
934 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
935 				err = -EINVAL;
936 			}
937 		}
938 	}
939 	if (err) {
940 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
941 		goto out;
942 	}
943 
944 	mlx4_do_sense_ports(mdev, new_types, types);
945 
946 	err = mlx4_check_port_params(mdev, new_types);
947 	if (err)
948 		goto out;
949 
950 	/* We are about to apply the changes after the configuration
951 	 * was verified, no need to remember the temporary types
952 	 * any more */
953 	for (i = 0; i < mdev->caps.num_ports; i++)
954 		priv->port[i + 1].tmp_type = 0;
955 
956 	err = mlx4_change_port_types(mdev, new_types);
957 
958 out:
959 	mlx4_start_sense(mdev);
960 	mutex_unlock(&priv->port_mutex);
961 	return err ? err : count;
962 }
963 
964 enum ibta_mtu {
965 	IB_MTU_256  = 1,
966 	IB_MTU_512  = 2,
967 	IB_MTU_1024 = 3,
968 	IB_MTU_2048 = 4,
969 	IB_MTU_4096 = 5
970 };
971 
972 static inline int int_to_ibta_mtu(int mtu)
973 {
974 	switch (mtu) {
975 	case 256:  return IB_MTU_256;
976 	case 512:  return IB_MTU_512;
977 	case 1024: return IB_MTU_1024;
978 	case 2048: return IB_MTU_2048;
979 	case 4096: return IB_MTU_4096;
980 	default: return -1;
981 	}
982 }
983 
984 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
985 {
986 	switch (mtu) {
987 	case IB_MTU_256:  return  256;
988 	case IB_MTU_512:  return  512;
989 	case IB_MTU_1024: return 1024;
990 	case IB_MTU_2048: return 2048;
991 	case IB_MTU_4096: return 4096;
992 	default: return -1;
993 	}
994 }
995 
996 static ssize_t show_port_ib_mtu(struct device *dev,
997 			     struct device_attribute *attr,
998 			     char *buf)
999 {
1000 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1001 						   port_mtu_attr);
1002 	struct mlx4_dev *mdev = info->dev;
1003 
1004 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1005 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1006 
1007 	sprintf(buf, "%d\n",
1008 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1009 	return strlen(buf);
1010 }
1011 
1012 static ssize_t set_port_ib_mtu(struct device *dev,
1013 			     struct device_attribute *attr,
1014 			     const char *buf, size_t count)
1015 {
1016 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1017 						   port_mtu_attr);
1018 	struct mlx4_dev *mdev = info->dev;
1019 	struct mlx4_priv *priv = mlx4_priv(mdev);
1020 	int err, port, mtu, ibta_mtu = -1;
1021 
1022 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1023 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1024 		return -EINVAL;
1025 	}
1026 
1027 	err = kstrtoint(buf, 0, &mtu);
1028 	if (!err)
1029 		ibta_mtu = int_to_ibta_mtu(mtu);
1030 
1031 	if (err || ibta_mtu < 0) {
1032 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1033 		return -EINVAL;
1034 	}
1035 
1036 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1037 
1038 	mlx4_stop_sense(mdev);
1039 	mutex_lock(&priv->port_mutex);
1040 	mlx4_unregister_device(mdev);
1041 	for (port = 1; port <= mdev->caps.num_ports; port++) {
1042 		mlx4_CLOSE_PORT(mdev, port);
1043 		err = mlx4_SET_PORT(mdev, port, -1);
1044 		if (err) {
1045 			mlx4_err(mdev, "Failed to set port %d, aborting\n",
1046 				 port);
1047 			goto err_set_port;
1048 		}
1049 	}
1050 	err = mlx4_register_device(mdev);
1051 err_set_port:
1052 	mutex_unlock(&priv->port_mutex);
1053 	mlx4_start_sense(mdev);
1054 	return err ? err : count;
1055 }
1056 
1057 static int mlx4_load_fw(struct mlx4_dev *dev)
1058 {
1059 	struct mlx4_priv *priv = mlx4_priv(dev);
1060 	int err;
1061 
1062 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1063 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
1064 	if (!priv->fw.fw_icm) {
1065 		mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1066 		return -ENOMEM;
1067 	}
1068 
1069 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1070 	if (err) {
1071 		mlx4_err(dev, "MAP_FA command failed, aborting\n");
1072 		goto err_free;
1073 	}
1074 
1075 	err = mlx4_RUN_FW(dev);
1076 	if (err) {
1077 		mlx4_err(dev, "RUN_FW command failed, aborting\n");
1078 		goto err_unmap_fa;
1079 	}
1080 
1081 	return 0;
1082 
1083 err_unmap_fa:
1084 	mlx4_UNMAP_FA(dev);
1085 
1086 err_free:
1087 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1088 	return err;
1089 }
1090 
1091 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1092 				int cmpt_entry_sz)
1093 {
1094 	struct mlx4_priv *priv = mlx4_priv(dev);
1095 	int err;
1096 	int num_eqs;
1097 
1098 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1099 				  cmpt_base +
1100 				  ((u64) (MLX4_CMPT_TYPE_QP *
1101 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1102 				  cmpt_entry_sz, dev->caps.num_qps,
1103 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1104 				  0, 0);
1105 	if (err)
1106 		goto err;
1107 
1108 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1109 				  cmpt_base +
1110 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1111 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1112 				  cmpt_entry_sz, dev->caps.num_srqs,
1113 				  dev->caps.reserved_srqs, 0, 0);
1114 	if (err)
1115 		goto err_qp;
1116 
1117 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1118 				  cmpt_base +
1119 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1120 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1121 				  cmpt_entry_sz, dev->caps.num_cqs,
1122 				  dev->caps.reserved_cqs, 0, 0);
1123 	if (err)
1124 		goto err_srq;
1125 
1126 	num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
1127 		  dev->caps.num_eqs;
1128 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1129 				  cmpt_base +
1130 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1131 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1132 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1133 	if (err)
1134 		goto err_cq;
1135 
1136 	return 0;
1137 
1138 err_cq:
1139 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1140 
1141 err_srq:
1142 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1143 
1144 err_qp:
1145 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1146 
1147 err:
1148 	return err;
1149 }
1150 
1151 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1152 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1153 {
1154 	struct mlx4_priv *priv = mlx4_priv(dev);
1155 	u64 aux_pages;
1156 	int num_eqs;
1157 	int err;
1158 
1159 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1160 	if (err) {
1161 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1162 		return err;
1163 	}
1164 
1165 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1166 		 (unsigned long long) icm_size >> 10,
1167 		 (unsigned long long) aux_pages << 2);
1168 
1169 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1170 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1171 	if (!priv->fw.aux_icm) {
1172 		mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1173 		return -ENOMEM;
1174 	}
1175 
1176 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1177 	if (err) {
1178 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1179 		goto err_free_aux;
1180 	}
1181 
1182 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1183 	if (err) {
1184 		mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1185 		goto err_unmap_aux;
1186 	}
1187 
1188 
1189 	num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
1190 		   dev->caps.num_eqs;
1191 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1192 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1193 				  num_eqs, num_eqs, 0, 0);
1194 	if (err) {
1195 		mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1196 		goto err_unmap_cmpt;
1197 	}
1198 
1199 	/*
1200 	 * Reserved MTT entries must be aligned up to a cacheline
1201 	 * boundary, since the FW will write to them, while the driver
1202 	 * writes to all other MTT entries. (The variable
1203 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1204 	 * size, not the raw entry size)
1205 	 */
1206 	dev->caps.reserved_mtts =
1207 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1208 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1209 
1210 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1211 				  init_hca->mtt_base,
1212 				  dev->caps.mtt_entry_sz,
1213 				  dev->caps.num_mtts,
1214 				  dev->caps.reserved_mtts, 1, 0);
1215 	if (err) {
1216 		mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1217 		goto err_unmap_eq;
1218 	}
1219 
1220 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1221 				  init_hca->dmpt_base,
1222 				  dev_cap->dmpt_entry_sz,
1223 				  dev->caps.num_mpts,
1224 				  dev->caps.reserved_mrws, 1, 1);
1225 	if (err) {
1226 		mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1227 		goto err_unmap_mtt;
1228 	}
1229 
1230 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1231 				  init_hca->qpc_base,
1232 				  dev_cap->qpc_entry_sz,
1233 				  dev->caps.num_qps,
1234 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1235 				  0, 0);
1236 	if (err) {
1237 		mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1238 		goto err_unmap_dmpt;
1239 	}
1240 
1241 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1242 				  init_hca->auxc_base,
1243 				  dev_cap->aux_entry_sz,
1244 				  dev->caps.num_qps,
1245 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1246 				  0, 0);
1247 	if (err) {
1248 		mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1249 		goto err_unmap_qp;
1250 	}
1251 
1252 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1253 				  init_hca->altc_base,
1254 				  dev_cap->altc_entry_sz,
1255 				  dev->caps.num_qps,
1256 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1257 				  0, 0);
1258 	if (err) {
1259 		mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1260 		goto err_unmap_auxc;
1261 	}
1262 
1263 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1264 				  init_hca->rdmarc_base,
1265 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1266 				  dev->caps.num_qps,
1267 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1268 				  0, 0);
1269 	if (err) {
1270 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1271 		goto err_unmap_altc;
1272 	}
1273 
1274 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1275 				  init_hca->cqc_base,
1276 				  dev_cap->cqc_entry_sz,
1277 				  dev->caps.num_cqs,
1278 				  dev->caps.reserved_cqs, 0, 0);
1279 	if (err) {
1280 		mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1281 		goto err_unmap_rdmarc;
1282 	}
1283 
1284 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1285 				  init_hca->srqc_base,
1286 				  dev_cap->srq_entry_sz,
1287 				  dev->caps.num_srqs,
1288 				  dev->caps.reserved_srqs, 0, 0);
1289 	if (err) {
1290 		mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1291 		goto err_unmap_cq;
1292 	}
1293 
1294 	/*
1295 	 * For flow steering device managed mode it is required to use
1296 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1297 	 * required, but for simplicity just map the whole multicast
1298 	 * group table now.  The table isn't very big and it's a lot
1299 	 * easier than trying to track ref counts.
1300 	 */
1301 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1302 				  init_hca->mc_base,
1303 				  mlx4_get_mgm_entry_size(dev),
1304 				  dev->caps.num_mgms + dev->caps.num_amgms,
1305 				  dev->caps.num_mgms + dev->caps.num_amgms,
1306 				  0, 0);
1307 	if (err) {
1308 		mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1309 		goto err_unmap_srq;
1310 	}
1311 
1312 	return 0;
1313 
1314 err_unmap_srq:
1315 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1316 
1317 err_unmap_cq:
1318 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1319 
1320 err_unmap_rdmarc:
1321 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1322 
1323 err_unmap_altc:
1324 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1325 
1326 err_unmap_auxc:
1327 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1328 
1329 err_unmap_qp:
1330 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1331 
1332 err_unmap_dmpt:
1333 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1334 
1335 err_unmap_mtt:
1336 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1337 
1338 err_unmap_eq:
1339 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1340 
1341 err_unmap_cmpt:
1342 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1343 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1344 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1345 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1346 
1347 err_unmap_aux:
1348 	mlx4_UNMAP_ICM_AUX(dev);
1349 
1350 err_free_aux:
1351 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1352 
1353 	return err;
1354 }
1355 
1356 static void mlx4_free_icms(struct mlx4_dev *dev)
1357 {
1358 	struct mlx4_priv *priv = mlx4_priv(dev);
1359 
1360 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1361 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1362 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1363 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1364 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1365 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1366 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1367 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1368 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1369 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1370 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1371 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1372 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1373 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1374 
1375 	mlx4_UNMAP_ICM_AUX(dev);
1376 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1377 }
1378 
1379 static void mlx4_slave_exit(struct mlx4_dev *dev)
1380 {
1381 	struct mlx4_priv *priv = mlx4_priv(dev);
1382 
1383 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1384 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_TIME))
1385 		mlx4_warn(dev, "Failed to close slave function\n");
1386 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1387 }
1388 
1389 static int map_bf_area(struct mlx4_dev *dev)
1390 {
1391 	struct mlx4_priv *priv = mlx4_priv(dev);
1392 	resource_size_t bf_start;
1393 	resource_size_t bf_len;
1394 	int err = 0;
1395 
1396 	if (!dev->caps.bf_reg_size)
1397 		return -ENXIO;
1398 
1399 	bf_start = pci_resource_start(dev->pdev, 2) +
1400 			(dev->caps.num_uars << PAGE_SHIFT);
1401 	bf_len = pci_resource_len(dev->pdev, 2) -
1402 			(dev->caps.num_uars << PAGE_SHIFT);
1403 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1404 	if (!priv->bf_mapping)
1405 		err = -ENOMEM;
1406 
1407 	return err;
1408 }
1409 
1410 static void unmap_bf_area(struct mlx4_dev *dev)
1411 {
1412 	if (mlx4_priv(dev)->bf_mapping)
1413 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1414 }
1415 
1416 cycle_t mlx4_read_clock(struct mlx4_dev *dev)
1417 {
1418 	u32 clockhi, clocklo, clockhi1;
1419 	cycle_t cycles;
1420 	int i;
1421 	struct mlx4_priv *priv = mlx4_priv(dev);
1422 
1423 	for (i = 0; i < 10; i++) {
1424 		clockhi = swab32(readl(priv->clock_mapping));
1425 		clocklo = swab32(readl(priv->clock_mapping + 4));
1426 		clockhi1 = swab32(readl(priv->clock_mapping));
1427 		if (clockhi == clockhi1)
1428 			break;
1429 	}
1430 
1431 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1432 
1433 	return cycles;
1434 }
1435 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1436 
1437 
1438 static int map_internal_clock(struct mlx4_dev *dev)
1439 {
1440 	struct mlx4_priv *priv = mlx4_priv(dev);
1441 
1442 	priv->clock_mapping =
1443 		ioremap(pci_resource_start(dev->pdev, priv->fw.clock_bar) +
1444 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1445 
1446 	if (!priv->clock_mapping)
1447 		return -ENOMEM;
1448 
1449 	return 0;
1450 }
1451 
1452 static void unmap_internal_clock(struct mlx4_dev *dev)
1453 {
1454 	struct mlx4_priv *priv = mlx4_priv(dev);
1455 
1456 	if (priv->clock_mapping)
1457 		iounmap(priv->clock_mapping);
1458 }
1459 
1460 static void mlx4_close_hca(struct mlx4_dev *dev)
1461 {
1462 	unmap_internal_clock(dev);
1463 	unmap_bf_area(dev);
1464 	if (mlx4_is_slave(dev))
1465 		mlx4_slave_exit(dev);
1466 	else {
1467 		mlx4_CLOSE_HCA(dev, 0);
1468 		mlx4_free_icms(dev);
1469 		mlx4_UNMAP_FA(dev);
1470 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1471 	}
1472 }
1473 
1474 static int mlx4_init_slave(struct mlx4_dev *dev)
1475 {
1476 	struct mlx4_priv *priv = mlx4_priv(dev);
1477 	u64 dma = (u64) priv->mfunc.vhcr_dma;
1478 	int ret_from_reset = 0;
1479 	u32 slave_read;
1480 	u32 cmd_channel_ver;
1481 
1482 	if (atomic_read(&pf_loading)) {
1483 		mlx4_warn(dev, "PF is not ready - Deferring probe\n");
1484 		return -EPROBE_DEFER;
1485 	}
1486 
1487 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1488 	priv->cmd.max_cmds = 1;
1489 	mlx4_warn(dev, "Sending reset\n");
1490 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1491 				       MLX4_COMM_TIME);
1492 	/* if we are in the middle of flr the slave will try
1493 	 * NUM_OF_RESET_RETRIES times before leaving.*/
1494 	if (ret_from_reset) {
1495 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1496 			mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
1497 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
1498 			return -EPROBE_DEFER;
1499 		} else
1500 			goto err;
1501 	}
1502 
1503 	/* check the driver version - the slave I/F revision
1504 	 * must match the master's */
1505 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1506 	cmd_channel_ver = mlx4_comm_get_version();
1507 
1508 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1509 		MLX4_COMM_GET_IF_REV(slave_read)) {
1510 		mlx4_err(dev, "slave driver version is not supported by the master\n");
1511 		goto err;
1512 	}
1513 
1514 	mlx4_warn(dev, "Sending vhcr0\n");
1515 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1516 						    MLX4_COMM_TIME))
1517 		goto err;
1518 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1519 						    MLX4_COMM_TIME))
1520 		goto err;
1521 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1522 						    MLX4_COMM_TIME))
1523 		goto err;
1524 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma, MLX4_COMM_TIME))
1525 		goto err;
1526 
1527 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1528 	return 0;
1529 
1530 err:
1531 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, 0);
1532 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1533 	return -EIO;
1534 }
1535 
1536 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1537 {
1538 	int i;
1539 
1540 	for (i = 1; i <= dev->caps.num_ports; i++) {
1541 		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
1542 			dev->caps.gid_table_len[i] =
1543 				mlx4_get_slave_num_gids(dev, 0, i);
1544 		else
1545 			dev->caps.gid_table_len[i] = 1;
1546 		dev->caps.pkey_table_len[i] =
1547 			dev->phys_caps.pkey_phys_table_len[i] - 1;
1548 	}
1549 }
1550 
1551 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
1552 {
1553 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
1554 
1555 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
1556 	      i++) {
1557 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
1558 			break;
1559 	}
1560 
1561 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
1562 }
1563 
1564 static void choose_steering_mode(struct mlx4_dev *dev,
1565 				 struct mlx4_dev_cap *dev_cap)
1566 {
1567 	if (mlx4_log_num_mgm_entry_size == -1 &&
1568 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
1569 	    (!mlx4_is_mfunc(dev) ||
1570 	     (dev_cap->fs_max_num_qp_per_entry >= (dev->num_vfs + 1))) &&
1571 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
1572 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
1573 		dev->oper_log_mgm_entry_size =
1574 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
1575 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1576 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
1577 		dev->caps.fs_log_max_ucast_qp_range_size =
1578 			dev_cap->fs_log_max_ucast_qp_range_size;
1579 	} else {
1580 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
1581 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1582 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
1583 		else {
1584 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
1585 
1586 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
1587 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1588 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
1589 		}
1590 		dev->oper_log_mgm_entry_size =
1591 			mlx4_log_num_mgm_entry_size > 0 ?
1592 			mlx4_log_num_mgm_entry_size :
1593 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
1594 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
1595 	}
1596 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
1597 		 mlx4_steering_mode_str(dev->caps.steering_mode),
1598 		 dev->oper_log_mgm_entry_size,
1599 		 mlx4_log_num_mgm_entry_size);
1600 }
1601 
1602 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
1603 				       struct mlx4_dev_cap *dev_cap)
1604 {
1605 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
1606 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
1607 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
1608 	else
1609 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
1610 
1611 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
1612 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
1613 }
1614 
1615 static int mlx4_init_hca(struct mlx4_dev *dev)
1616 {
1617 	struct mlx4_priv	  *priv = mlx4_priv(dev);
1618 	struct mlx4_adapter	   adapter;
1619 	struct mlx4_dev_cap	   dev_cap;
1620 	struct mlx4_mod_stat_cfg   mlx4_cfg;
1621 	struct mlx4_profile	   profile;
1622 	struct mlx4_init_hca_param init_hca;
1623 	u64 icm_size;
1624 	int err;
1625 
1626 	if (!mlx4_is_slave(dev)) {
1627 		err = mlx4_QUERY_FW(dev);
1628 		if (err) {
1629 			if (err == -EACCES)
1630 				mlx4_info(dev, "non-primary physical function, skipping\n");
1631 			else
1632 				mlx4_err(dev, "QUERY_FW command failed, aborting\n");
1633 			return err;
1634 		}
1635 
1636 		err = mlx4_load_fw(dev);
1637 		if (err) {
1638 			mlx4_err(dev, "Failed to start FW, aborting\n");
1639 			return err;
1640 		}
1641 
1642 		mlx4_cfg.log_pg_sz_m = 1;
1643 		mlx4_cfg.log_pg_sz = 0;
1644 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
1645 		if (err)
1646 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
1647 
1648 		err = mlx4_dev_cap(dev, &dev_cap);
1649 		if (err) {
1650 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
1651 			goto err_stop_fw;
1652 		}
1653 
1654 		choose_steering_mode(dev, &dev_cap);
1655 		choose_tunnel_offload_mode(dev, &dev_cap);
1656 
1657 		err = mlx4_get_phys_port_id(dev);
1658 		if (err)
1659 			mlx4_err(dev, "Fail to get physical port id\n");
1660 
1661 		if (mlx4_is_master(dev))
1662 			mlx4_parav_master_pf_caps(dev);
1663 
1664 		if (mlx4_low_memory_profile()) {
1665 			mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
1666 			profile = low_mem_profile;
1667 		} else {
1668 			profile = default_profile;
1669 		}
1670 		if (dev->caps.steering_mode ==
1671 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
1672 			profile.num_mcg = MLX4_FS_NUM_MCG;
1673 
1674 		icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
1675 					     &init_hca);
1676 		if ((long long) icm_size < 0) {
1677 			err = icm_size;
1678 			goto err_stop_fw;
1679 		}
1680 
1681 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
1682 
1683 		init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
1684 		init_hca.uar_page_sz = PAGE_SHIFT - 12;
1685 		init_hca.mw_enabled = 0;
1686 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1687 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
1688 			init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
1689 
1690 		err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
1691 		if (err)
1692 			goto err_stop_fw;
1693 
1694 		err = mlx4_INIT_HCA(dev, &init_hca);
1695 		if (err) {
1696 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
1697 			goto err_free_icm;
1698 		}
1699 		/*
1700 		 * If TS is supported by FW
1701 		 * read HCA frequency by QUERY_HCA command
1702 		 */
1703 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
1704 			memset(&init_hca, 0, sizeof(init_hca));
1705 			err = mlx4_QUERY_HCA(dev, &init_hca);
1706 			if (err) {
1707 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
1708 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1709 			} else {
1710 				dev->caps.hca_core_clock =
1711 					init_hca.hca_core_clock;
1712 			}
1713 
1714 			/* In case we got HCA frequency 0 - disable timestamping
1715 			 * to avoid dividing by zero
1716 			 */
1717 			if (!dev->caps.hca_core_clock) {
1718 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1719 				mlx4_err(dev,
1720 					 "HCA frequency is 0 - timestamping is not supported\n");
1721 			} else if (map_internal_clock(dev)) {
1722 				/*
1723 				 * Map internal clock,
1724 				 * in case of failure disable timestamping
1725 				 */
1726 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1727 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
1728 			}
1729 		}
1730 	} else {
1731 		err = mlx4_init_slave(dev);
1732 		if (err) {
1733 			if (err != -EPROBE_DEFER)
1734 				mlx4_err(dev, "Failed to initialize slave\n");
1735 			return err;
1736 		}
1737 
1738 		err = mlx4_slave_cap(dev);
1739 		if (err) {
1740 			mlx4_err(dev, "Failed to obtain slave caps\n");
1741 			goto err_close;
1742 		}
1743 	}
1744 
1745 	if (map_bf_area(dev))
1746 		mlx4_dbg(dev, "Failed to map blue flame area\n");
1747 
1748 	/*Only the master set the ports, all the rest got it from it.*/
1749 	if (!mlx4_is_slave(dev))
1750 		mlx4_set_port_mask(dev);
1751 
1752 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
1753 	if (err) {
1754 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
1755 		goto unmap_bf;
1756 	}
1757 
1758 	priv->eq_table.inta_pin = adapter.inta_pin;
1759 	memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
1760 
1761 	return 0;
1762 
1763 unmap_bf:
1764 	unmap_internal_clock(dev);
1765 	unmap_bf_area(dev);
1766 
1767 	if (mlx4_is_slave(dev)) {
1768 		kfree(dev->caps.qp0_qkey);
1769 		kfree(dev->caps.qp0_tunnel);
1770 		kfree(dev->caps.qp0_proxy);
1771 		kfree(dev->caps.qp1_tunnel);
1772 		kfree(dev->caps.qp1_proxy);
1773 	}
1774 
1775 err_close:
1776 	if (mlx4_is_slave(dev))
1777 		mlx4_slave_exit(dev);
1778 	else
1779 		mlx4_CLOSE_HCA(dev, 0);
1780 
1781 err_free_icm:
1782 	if (!mlx4_is_slave(dev))
1783 		mlx4_free_icms(dev);
1784 
1785 err_stop_fw:
1786 	if (!mlx4_is_slave(dev)) {
1787 		mlx4_UNMAP_FA(dev);
1788 		mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1789 	}
1790 	return err;
1791 }
1792 
1793 static int mlx4_init_counters_table(struct mlx4_dev *dev)
1794 {
1795 	struct mlx4_priv *priv = mlx4_priv(dev);
1796 	int nent;
1797 
1798 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1799 		return -ENOENT;
1800 
1801 	nent = dev->caps.max_counters;
1802 	return mlx4_bitmap_init(&priv->counters_bitmap, nent, nent - 1, 0, 0);
1803 }
1804 
1805 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
1806 {
1807 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
1808 }
1809 
1810 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1811 {
1812 	struct mlx4_priv *priv = mlx4_priv(dev);
1813 
1814 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1815 		return -ENOENT;
1816 
1817 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
1818 	if (*idx == -1)
1819 		return -ENOMEM;
1820 
1821 	return 0;
1822 }
1823 
1824 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1825 {
1826 	u64 out_param;
1827 	int err;
1828 
1829 	if (mlx4_is_mfunc(dev)) {
1830 		err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
1831 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
1832 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
1833 		if (!err)
1834 			*idx = get_param_l(&out_param);
1835 
1836 		return err;
1837 	}
1838 	return __mlx4_counter_alloc(dev, idx);
1839 }
1840 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
1841 
1842 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1843 {
1844 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
1845 	return;
1846 }
1847 
1848 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1849 {
1850 	u64 in_param = 0;
1851 
1852 	if (mlx4_is_mfunc(dev)) {
1853 		set_param_l(&in_param, idx);
1854 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
1855 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
1856 			 MLX4_CMD_WRAPPED);
1857 		return;
1858 	}
1859 	__mlx4_counter_free(dev, idx);
1860 }
1861 EXPORT_SYMBOL_GPL(mlx4_counter_free);
1862 
1863 static int mlx4_setup_hca(struct mlx4_dev *dev)
1864 {
1865 	struct mlx4_priv *priv = mlx4_priv(dev);
1866 	int err;
1867 	int port;
1868 	__be32 ib_port_default_caps;
1869 
1870 	err = mlx4_init_uar_table(dev);
1871 	if (err) {
1872 		mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
1873 		 return err;
1874 	}
1875 
1876 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
1877 	if (err) {
1878 		mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
1879 		goto err_uar_table_free;
1880 	}
1881 
1882 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
1883 	if (!priv->kar) {
1884 		mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
1885 		err = -ENOMEM;
1886 		goto err_uar_free;
1887 	}
1888 
1889 	err = mlx4_init_pd_table(dev);
1890 	if (err) {
1891 		mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
1892 		goto err_kar_unmap;
1893 	}
1894 
1895 	err = mlx4_init_xrcd_table(dev);
1896 	if (err) {
1897 		mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
1898 		goto err_pd_table_free;
1899 	}
1900 
1901 	err = mlx4_init_mr_table(dev);
1902 	if (err) {
1903 		mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
1904 		goto err_xrcd_table_free;
1905 	}
1906 
1907 	if (!mlx4_is_slave(dev)) {
1908 		err = mlx4_init_mcg_table(dev);
1909 		if (err) {
1910 			mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
1911 			goto err_mr_table_free;
1912 		}
1913 		err = mlx4_config_mad_demux(dev);
1914 		if (err) {
1915 			mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
1916 			goto err_mcg_table_free;
1917 		}
1918 	}
1919 
1920 	err = mlx4_init_eq_table(dev);
1921 	if (err) {
1922 		mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
1923 		goto err_mcg_table_free;
1924 	}
1925 
1926 	err = mlx4_cmd_use_events(dev);
1927 	if (err) {
1928 		mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
1929 		goto err_eq_table_free;
1930 	}
1931 
1932 	err = mlx4_NOP(dev);
1933 	if (err) {
1934 		if (dev->flags & MLX4_FLAG_MSI_X) {
1935 			mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
1936 				  priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1937 			mlx4_warn(dev, "Trying again without MSI-X\n");
1938 		} else {
1939 			mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
1940 				 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1941 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
1942 		}
1943 
1944 		goto err_cmd_poll;
1945 	}
1946 
1947 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
1948 
1949 	err = mlx4_init_cq_table(dev);
1950 	if (err) {
1951 		mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
1952 		goto err_cmd_poll;
1953 	}
1954 
1955 	err = mlx4_init_srq_table(dev);
1956 	if (err) {
1957 		mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
1958 		goto err_cq_table_free;
1959 	}
1960 
1961 	err = mlx4_init_qp_table(dev);
1962 	if (err) {
1963 		mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
1964 		goto err_srq_table_free;
1965 	}
1966 
1967 	err = mlx4_init_counters_table(dev);
1968 	if (err && err != -ENOENT) {
1969 		mlx4_err(dev, "Failed to initialize counters table, aborting\n");
1970 		goto err_qp_table_free;
1971 	}
1972 
1973 	if (!mlx4_is_slave(dev)) {
1974 		for (port = 1; port <= dev->caps.num_ports; port++) {
1975 			ib_port_default_caps = 0;
1976 			err = mlx4_get_port_ib_caps(dev, port,
1977 						    &ib_port_default_caps);
1978 			if (err)
1979 				mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
1980 					  port, err);
1981 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
1982 
1983 			/* initialize per-slave default ib port capabilities */
1984 			if (mlx4_is_master(dev)) {
1985 				int i;
1986 				for (i = 0; i < dev->num_slaves; i++) {
1987 					if (i == mlx4_master_func_num(dev))
1988 						continue;
1989 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
1990 						ib_port_default_caps;
1991 				}
1992 			}
1993 
1994 			if (mlx4_is_mfunc(dev))
1995 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
1996 			else
1997 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
1998 
1999 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2000 					    dev->caps.pkey_table_len[port] : -1);
2001 			if (err) {
2002 				mlx4_err(dev, "Failed to set port %d, aborting\n",
2003 					 port);
2004 				goto err_counters_table_free;
2005 			}
2006 		}
2007 	}
2008 
2009 	return 0;
2010 
2011 err_counters_table_free:
2012 	mlx4_cleanup_counters_table(dev);
2013 
2014 err_qp_table_free:
2015 	mlx4_cleanup_qp_table(dev);
2016 
2017 err_srq_table_free:
2018 	mlx4_cleanup_srq_table(dev);
2019 
2020 err_cq_table_free:
2021 	mlx4_cleanup_cq_table(dev);
2022 
2023 err_cmd_poll:
2024 	mlx4_cmd_use_polling(dev);
2025 
2026 err_eq_table_free:
2027 	mlx4_cleanup_eq_table(dev);
2028 
2029 err_mcg_table_free:
2030 	if (!mlx4_is_slave(dev))
2031 		mlx4_cleanup_mcg_table(dev);
2032 
2033 err_mr_table_free:
2034 	mlx4_cleanup_mr_table(dev);
2035 
2036 err_xrcd_table_free:
2037 	mlx4_cleanup_xrcd_table(dev);
2038 
2039 err_pd_table_free:
2040 	mlx4_cleanup_pd_table(dev);
2041 
2042 err_kar_unmap:
2043 	iounmap(priv->kar);
2044 
2045 err_uar_free:
2046 	mlx4_uar_free(dev, &priv->driver_uar);
2047 
2048 err_uar_table_free:
2049 	mlx4_cleanup_uar_table(dev);
2050 	return err;
2051 }
2052 
2053 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2054 {
2055 	struct mlx4_priv *priv = mlx4_priv(dev);
2056 	struct msix_entry *entries;
2057 	int nreq = min_t(int, dev->caps.num_ports *
2058 			 min_t(int, num_online_cpus() + 1,
2059 			       MAX_MSIX_P_PORT) + MSIX_LEGACY_SZ, MAX_MSIX);
2060 	int i;
2061 
2062 	if (msi_x) {
2063 		nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
2064 			     nreq);
2065 
2066 		entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
2067 		if (!entries)
2068 			goto no_msi;
2069 
2070 		for (i = 0; i < nreq; ++i)
2071 			entries[i].entry = i;
2072 
2073 		nreq = pci_enable_msix_range(dev->pdev, entries, 2, nreq);
2074 
2075 		if (nreq < 0) {
2076 			kfree(entries);
2077 			goto no_msi;
2078 		} else if (nreq < MSIX_LEGACY_SZ +
2079 			   dev->caps.num_ports * MIN_MSIX_P_PORT) {
2080 			/*Working in legacy mode , all EQ's shared*/
2081 			dev->caps.comp_pool           = 0;
2082 			dev->caps.num_comp_vectors = nreq - 1;
2083 		} else {
2084 			dev->caps.comp_pool           = nreq - MSIX_LEGACY_SZ;
2085 			dev->caps.num_comp_vectors = MSIX_LEGACY_SZ - 1;
2086 		}
2087 		for (i = 0; i < nreq; ++i)
2088 			priv->eq_table.eq[i].irq = entries[i].vector;
2089 
2090 		dev->flags |= MLX4_FLAG_MSI_X;
2091 
2092 		kfree(entries);
2093 		return;
2094 	}
2095 
2096 no_msi:
2097 	dev->caps.num_comp_vectors = 1;
2098 	dev->caps.comp_pool	   = 0;
2099 
2100 	for (i = 0; i < 2; ++i)
2101 		priv->eq_table.eq[i].irq = dev->pdev->irq;
2102 }
2103 
2104 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2105 {
2106 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2107 	int err = 0;
2108 
2109 	info->dev = dev;
2110 	info->port = port;
2111 	if (!mlx4_is_slave(dev)) {
2112 		mlx4_init_mac_table(dev, &info->mac_table);
2113 		mlx4_init_vlan_table(dev, &info->vlan_table);
2114 		mlx4_init_roce_gid_table(dev, &info->gid_table);
2115 		info->base_qpn = mlx4_get_base_qpn(dev, port);
2116 	}
2117 
2118 	sprintf(info->dev_name, "mlx4_port%d", port);
2119 	info->port_attr.attr.name = info->dev_name;
2120 	if (mlx4_is_mfunc(dev))
2121 		info->port_attr.attr.mode = S_IRUGO;
2122 	else {
2123 		info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2124 		info->port_attr.store     = set_port_type;
2125 	}
2126 	info->port_attr.show      = show_port_type;
2127 	sysfs_attr_init(&info->port_attr.attr);
2128 
2129 	err = device_create_file(&dev->pdev->dev, &info->port_attr);
2130 	if (err) {
2131 		mlx4_err(dev, "Failed to create file for port %d\n", port);
2132 		info->port = -1;
2133 	}
2134 
2135 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2136 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
2137 	if (mlx4_is_mfunc(dev))
2138 		info->port_mtu_attr.attr.mode = S_IRUGO;
2139 	else {
2140 		info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2141 		info->port_mtu_attr.store     = set_port_ib_mtu;
2142 	}
2143 	info->port_mtu_attr.show      = show_port_ib_mtu;
2144 	sysfs_attr_init(&info->port_mtu_attr.attr);
2145 
2146 	err = device_create_file(&dev->pdev->dev, &info->port_mtu_attr);
2147 	if (err) {
2148 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
2149 		device_remove_file(&info->dev->pdev->dev, &info->port_attr);
2150 		info->port = -1;
2151 	}
2152 
2153 	return err;
2154 }
2155 
2156 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
2157 {
2158 	if (info->port < 0)
2159 		return;
2160 
2161 	device_remove_file(&info->dev->pdev->dev, &info->port_attr);
2162 	device_remove_file(&info->dev->pdev->dev, &info->port_mtu_attr);
2163 }
2164 
2165 static int mlx4_init_steering(struct mlx4_dev *dev)
2166 {
2167 	struct mlx4_priv *priv = mlx4_priv(dev);
2168 	int num_entries = dev->caps.num_ports;
2169 	int i, j;
2170 
2171 	priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
2172 	if (!priv->steer)
2173 		return -ENOMEM;
2174 
2175 	for (i = 0; i < num_entries; i++)
2176 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
2177 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
2178 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
2179 		}
2180 	return 0;
2181 }
2182 
2183 static void mlx4_clear_steering(struct mlx4_dev *dev)
2184 {
2185 	struct mlx4_priv *priv = mlx4_priv(dev);
2186 	struct mlx4_steer_index *entry, *tmp_entry;
2187 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
2188 	int num_entries = dev->caps.num_ports;
2189 	int i, j;
2190 
2191 	for (i = 0; i < num_entries; i++) {
2192 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
2193 			list_for_each_entry_safe(pqp, tmp_pqp,
2194 						 &priv->steer[i].promisc_qps[j],
2195 						 list) {
2196 				list_del(&pqp->list);
2197 				kfree(pqp);
2198 			}
2199 			list_for_each_entry_safe(entry, tmp_entry,
2200 						 &priv->steer[i].steer_entries[j],
2201 						 list) {
2202 				list_del(&entry->list);
2203 				list_for_each_entry_safe(pqp, tmp_pqp,
2204 							 &entry->duplicates,
2205 							 list) {
2206 					list_del(&pqp->list);
2207 					kfree(pqp);
2208 				}
2209 				kfree(entry);
2210 			}
2211 		}
2212 	}
2213 	kfree(priv->steer);
2214 }
2215 
2216 static int extended_func_num(struct pci_dev *pdev)
2217 {
2218 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
2219 }
2220 
2221 #define MLX4_OWNER_BASE	0x8069c
2222 #define MLX4_OWNER_SIZE	4
2223 
2224 static int mlx4_get_ownership(struct mlx4_dev *dev)
2225 {
2226 	void __iomem *owner;
2227 	u32 ret;
2228 
2229 	if (pci_channel_offline(dev->pdev))
2230 		return -EIO;
2231 
2232 	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
2233 			MLX4_OWNER_SIZE);
2234 	if (!owner) {
2235 		mlx4_err(dev, "Failed to obtain ownership bit\n");
2236 		return -ENOMEM;
2237 	}
2238 
2239 	ret = readl(owner);
2240 	iounmap(owner);
2241 	return (int) !!ret;
2242 }
2243 
2244 static void mlx4_free_ownership(struct mlx4_dev *dev)
2245 {
2246 	void __iomem *owner;
2247 
2248 	if (pci_channel_offline(dev->pdev))
2249 		return;
2250 
2251 	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
2252 			MLX4_OWNER_SIZE);
2253 	if (!owner) {
2254 		mlx4_err(dev, "Failed to obtain ownership bit\n");
2255 		return;
2256 	}
2257 	writel(0, owner);
2258 	msleep(1000);
2259 	iounmap(owner);
2260 }
2261 
2262 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
2263 			 int total_vfs, int *nvfs, struct mlx4_priv *priv)
2264 {
2265 	struct mlx4_dev *dev;
2266 	unsigned sum = 0;
2267 	int err;
2268 	int port;
2269 	int i;
2270 	int existing_vfs = 0;
2271 
2272 	dev = &priv->dev;
2273 
2274 	INIT_LIST_HEAD(&priv->ctx_list);
2275 	spin_lock_init(&priv->ctx_lock);
2276 
2277 	mutex_init(&priv->port_mutex);
2278 
2279 	INIT_LIST_HEAD(&priv->pgdir_list);
2280 	mutex_init(&priv->pgdir_mutex);
2281 
2282 	INIT_LIST_HEAD(&priv->bf_list);
2283 	mutex_init(&priv->bf_mutex);
2284 
2285 	dev->rev_id = pdev->revision;
2286 	dev->numa_node = dev_to_node(&pdev->dev);
2287 
2288 	/* Detect if this device is a virtual function */
2289 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
2290 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
2291 		dev->flags |= MLX4_FLAG_SLAVE;
2292 	} else {
2293 		/* We reset the device and enable SRIOV only for physical
2294 		 * devices.  Try to claim ownership on the device;
2295 		 * if already taken, skip -- do not allow multiple PFs */
2296 		err = mlx4_get_ownership(dev);
2297 		if (err) {
2298 			if (err < 0)
2299 				return err;
2300 			else {
2301 				mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
2302 				return -EINVAL;
2303 			}
2304 		}
2305 
2306 		if (total_vfs) {
2307 			mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n",
2308 				  total_vfs);
2309 			dev->dev_vfs = kzalloc(
2310 				total_vfs * sizeof(*dev->dev_vfs),
2311 				GFP_KERNEL);
2312 			if (NULL == dev->dev_vfs) {
2313 				mlx4_err(dev, "Failed to allocate memory for VFs\n");
2314 				err = -ENOMEM;
2315 				goto err_free_own;
2316 			} else {
2317 				atomic_inc(&pf_loading);
2318 				existing_vfs = pci_num_vf(pdev);
2319 				if (existing_vfs) {
2320 					err = 0;
2321 					if (existing_vfs != total_vfs)
2322 						mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
2323 							 existing_vfs, total_vfs);
2324 				} else {
2325 					err = pci_enable_sriov(pdev, total_vfs);
2326 				}
2327 				if (err) {
2328 					mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
2329 						 err);
2330 					atomic_dec(&pf_loading);
2331 				} else {
2332 					mlx4_warn(dev, "Running in master mode\n");
2333 					dev->flags |= MLX4_FLAG_SRIOV |
2334 						MLX4_FLAG_MASTER;
2335 					dev->num_vfs = total_vfs;
2336 				}
2337 			}
2338 		}
2339 
2340 		atomic_set(&priv->opreq_count, 0);
2341 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
2342 
2343 		/*
2344 		 * Now reset the HCA before we touch the PCI capabilities or
2345 		 * attempt a firmware command, since a boot ROM may have left
2346 		 * the HCA in an undefined state.
2347 		 */
2348 		err = mlx4_reset(dev);
2349 		if (err) {
2350 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
2351 			goto err_sriov;
2352 		}
2353 	}
2354 
2355 slave_start:
2356 	err = mlx4_cmd_init(dev);
2357 	if (err) {
2358 		mlx4_err(dev, "Failed to init command interface, aborting\n");
2359 		goto err_sriov;
2360 	}
2361 
2362 	/* In slave functions, the communication channel must be initialized
2363 	 * before posting commands. Also, init num_slaves before calling
2364 	 * mlx4_init_hca */
2365 	if (mlx4_is_mfunc(dev)) {
2366 		if (mlx4_is_master(dev))
2367 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
2368 		else {
2369 			dev->num_slaves = 0;
2370 			err = mlx4_multi_func_init(dev);
2371 			if (err) {
2372 				mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
2373 				goto err_cmd;
2374 			}
2375 		}
2376 	}
2377 
2378 	err = mlx4_init_hca(dev);
2379 	if (err) {
2380 		if (err == -EACCES) {
2381 			/* Not primary Physical function
2382 			 * Running in slave mode */
2383 			mlx4_cmd_cleanup(dev);
2384 			dev->flags |= MLX4_FLAG_SLAVE;
2385 			dev->flags &= ~MLX4_FLAG_MASTER;
2386 			goto slave_start;
2387 		} else
2388 			goto err_mfunc;
2389 	}
2390 
2391 	/* check if the device is functioning at its maximum possible speed.
2392 	 * No return code for this call, just warn the user in case of PCI
2393 	 * express device capabilities are under-satisfied by the bus.
2394 	 */
2395 	if (!mlx4_is_slave(dev))
2396 		mlx4_check_pcie_caps(dev);
2397 
2398 	/* In master functions, the communication channel must be initialized
2399 	 * after obtaining its address from fw */
2400 	if (mlx4_is_master(dev)) {
2401 		int ib_ports = 0;
2402 
2403 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2404 			ib_ports++;
2405 
2406 		if (ib_ports &&
2407 		    (num_vfs_argc > 1 || probe_vfs_argc > 1)) {
2408 			mlx4_err(dev,
2409 				 "Invalid syntax of num_vfs/probe_vfs with IB port - single port VFs syntax is only supported when all ports are configured as ethernet\n");
2410 			err = -EINVAL;
2411 			goto err_close;
2412 		}
2413 		if (dev->caps.num_ports < 2 &&
2414 		    num_vfs_argc > 1) {
2415 			err = -EINVAL;
2416 			mlx4_err(dev,
2417 				 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
2418 				 dev->caps.num_ports);
2419 			goto err_close;
2420 		}
2421 		memcpy(dev->nvfs, nvfs, sizeof(dev->nvfs));
2422 
2423 		for (i = 0; i < sizeof(dev->nvfs)/sizeof(dev->nvfs[0]); i++) {
2424 			unsigned j;
2425 
2426 			for (j = 0; j < dev->nvfs[i]; ++sum, ++j) {
2427 				dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
2428 				dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
2429 					dev->caps.num_ports;
2430 			}
2431 		}
2432 
2433 		/* In master functions, the communication channel
2434 		 * must be initialized after obtaining its address from fw
2435 		 */
2436 		err = mlx4_multi_func_init(dev);
2437 		if (err) {
2438 			mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
2439 			goto err_close;
2440 		}
2441 	}
2442 
2443 	err = mlx4_alloc_eq_table(dev);
2444 	if (err)
2445 		goto err_master_mfunc;
2446 
2447 	priv->msix_ctl.pool_bm = 0;
2448 	mutex_init(&priv->msix_ctl.pool_lock);
2449 
2450 	mlx4_enable_msi_x(dev);
2451 	if ((mlx4_is_mfunc(dev)) &&
2452 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
2453 		err = -ENOSYS;
2454 		mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
2455 		goto err_free_eq;
2456 	}
2457 
2458 	if (!mlx4_is_slave(dev)) {
2459 		err = mlx4_init_steering(dev);
2460 		if (err)
2461 			goto err_disable_msix;
2462 	}
2463 
2464 	err = mlx4_setup_hca(dev);
2465 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
2466 	    !mlx4_is_mfunc(dev)) {
2467 		dev->flags &= ~MLX4_FLAG_MSI_X;
2468 		dev->caps.num_comp_vectors = 1;
2469 		dev->caps.comp_pool	   = 0;
2470 		pci_disable_msix(pdev);
2471 		err = mlx4_setup_hca(dev);
2472 	}
2473 
2474 	if (err)
2475 		goto err_steer;
2476 
2477 	mlx4_init_quotas(dev);
2478 
2479 	for (port = 1; port <= dev->caps.num_ports; port++) {
2480 		err = mlx4_init_port_info(dev, port);
2481 		if (err)
2482 			goto err_port;
2483 	}
2484 
2485 	err = mlx4_register_device(dev);
2486 	if (err)
2487 		goto err_port;
2488 
2489 	mlx4_request_modules(dev);
2490 
2491 	mlx4_sense_init(dev);
2492 	mlx4_start_sense(dev);
2493 
2494 	priv->removed = 0;
2495 
2496 	if (mlx4_is_master(dev) && dev->num_vfs)
2497 		atomic_dec(&pf_loading);
2498 
2499 	return 0;
2500 
2501 err_port:
2502 	for (--port; port >= 1; --port)
2503 		mlx4_cleanup_port_info(&priv->port[port]);
2504 
2505 	mlx4_cleanup_counters_table(dev);
2506 	mlx4_cleanup_qp_table(dev);
2507 	mlx4_cleanup_srq_table(dev);
2508 	mlx4_cleanup_cq_table(dev);
2509 	mlx4_cmd_use_polling(dev);
2510 	mlx4_cleanup_eq_table(dev);
2511 	mlx4_cleanup_mcg_table(dev);
2512 	mlx4_cleanup_mr_table(dev);
2513 	mlx4_cleanup_xrcd_table(dev);
2514 	mlx4_cleanup_pd_table(dev);
2515 	mlx4_cleanup_uar_table(dev);
2516 
2517 err_steer:
2518 	if (!mlx4_is_slave(dev))
2519 		mlx4_clear_steering(dev);
2520 
2521 err_disable_msix:
2522 	if (dev->flags & MLX4_FLAG_MSI_X)
2523 		pci_disable_msix(pdev);
2524 
2525 err_free_eq:
2526 	mlx4_free_eq_table(dev);
2527 
2528 err_master_mfunc:
2529 	if (mlx4_is_master(dev))
2530 		mlx4_multi_func_cleanup(dev);
2531 
2532 	if (mlx4_is_slave(dev)) {
2533 		kfree(dev->caps.qp0_qkey);
2534 		kfree(dev->caps.qp0_tunnel);
2535 		kfree(dev->caps.qp0_proxy);
2536 		kfree(dev->caps.qp1_tunnel);
2537 		kfree(dev->caps.qp1_proxy);
2538 	}
2539 
2540 err_close:
2541 	mlx4_close_hca(dev);
2542 
2543 err_mfunc:
2544 	if (mlx4_is_slave(dev))
2545 		mlx4_multi_func_cleanup(dev);
2546 
2547 err_cmd:
2548 	mlx4_cmd_cleanup(dev);
2549 
2550 err_sriov:
2551 	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs)
2552 		pci_disable_sriov(pdev);
2553 
2554 	if (mlx4_is_master(dev) && dev->num_vfs)
2555 		atomic_dec(&pf_loading);
2556 
2557 	kfree(priv->dev.dev_vfs);
2558 
2559 err_free_own:
2560 	if (!mlx4_is_slave(dev))
2561 		mlx4_free_ownership(dev);
2562 
2563 	return err;
2564 }
2565 
2566 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
2567 			   struct mlx4_priv *priv)
2568 {
2569 	int err;
2570 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
2571 	int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
2572 	const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
2573 		{2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
2574 	unsigned total_vfs = 0;
2575 	unsigned int i;
2576 
2577 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
2578 
2579 	err = pci_enable_device(pdev);
2580 	if (err) {
2581 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
2582 		return err;
2583 	}
2584 
2585 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
2586 	 * per port, we must limit the number of VFs to 63 (since their are
2587 	 * 128 MACs)
2588 	 */
2589 	for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
2590 	     total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
2591 		nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
2592 		if (nvfs[i] < 0) {
2593 			dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
2594 			err = -EINVAL;
2595 			goto err_disable_pdev;
2596 		}
2597 	}
2598 	for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
2599 	     i++) {
2600 		prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
2601 		if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
2602 			dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
2603 			err = -EINVAL;
2604 			goto err_disable_pdev;
2605 		}
2606 	}
2607 	if (total_vfs >= MLX4_MAX_NUM_VF) {
2608 		dev_err(&pdev->dev,
2609 			"Requested more VF's (%d) than allowed (%d)\n",
2610 			total_vfs, MLX4_MAX_NUM_VF - 1);
2611 		err = -EINVAL;
2612 		goto err_disable_pdev;
2613 	}
2614 
2615 	for (i = 0; i < MLX4_MAX_PORTS; i++) {
2616 		if (nvfs[i] + nvfs[2] >= MLX4_MAX_NUM_VF_P_PORT) {
2617 			dev_err(&pdev->dev,
2618 				"Requested more VF's (%d) for port (%d) than allowed (%d)\n",
2619 				nvfs[i] + nvfs[2], i + 1,
2620 				MLX4_MAX_NUM_VF_P_PORT - 1);
2621 			err = -EINVAL;
2622 			goto err_disable_pdev;
2623 		}
2624 	}
2625 
2626 	/* Check for BARs. */
2627 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
2628 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
2629 		dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
2630 			pci_dev_data, pci_resource_flags(pdev, 0));
2631 		err = -ENODEV;
2632 		goto err_disable_pdev;
2633 	}
2634 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
2635 		dev_err(&pdev->dev, "Missing UAR, aborting\n");
2636 		err = -ENODEV;
2637 		goto err_disable_pdev;
2638 	}
2639 
2640 	err = pci_request_regions(pdev, DRV_NAME);
2641 	if (err) {
2642 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
2643 		goto err_disable_pdev;
2644 	}
2645 
2646 	pci_set_master(pdev);
2647 
2648 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2649 	if (err) {
2650 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
2651 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2652 		if (err) {
2653 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
2654 			goto err_release_regions;
2655 		}
2656 	}
2657 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2658 	if (err) {
2659 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
2660 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2661 		if (err) {
2662 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
2663 			goto err_release_regions;
2664 		}
2665 	}
2666 
2667 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
2668 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
2669 	/* Detect if this device is a virtual function */
2670 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
2671 		/* When acting as pf, we normally skip vfs unless explicitly
2672 		 * requested to probe them.
2673 		 */
2674 		if (total_vfs) {
2675 			unsigned vfs_offset = 0;
2676 
2677 			for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
2678 			     vfs_offset + nvfs[i] < extended_func_num(pdev);
2679 			     vfs_offset += nvfs[i], i++)
2680 				;
2681 			if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
2682 				err = -ENODEV;
2683 				goto err_release_regions;
2684 			}
2685 			if ((extended_func_num(pdev) - vfs_offset)
2686 			    > prb_vf[i]) {
2687 				dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
2688 					 extended_func_num(pdev));
2689 				err = -ENODEV;
2690 				goto err_release_regions;
2691 			}
2692 		}
2693 	}
2694 
2695 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv);
2696 	if (err)
2697 		goto err_release_regions;
2698 	return 0;
2699 
2700 err_release_regions:
2701 	pci_release_regions(pdev);
2702 
2703 err_disable_pdev:
2704 	pci_disable_device(pdev);
2705 	pci_set_drvdata(pdev, NULL);
2706 	return err;
2707 }
2708 
2709 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
2710 {
2711 	struct mlx4_priv *priv;
2712 	struct mlx4_dev *dev;
2713 	int ret;
2714 
2715 	printk_once(KERN_INFO "%s", mlx4_version);
2716 
2717 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
2718 	if (!priv)
2719 		return -ENOMEM;
2720 
2721 	dev       = &priv->dev;
2722 	dev->pdev = pdev;
2723 	pci_set_drvdata(pdev, dev);
2724 	priv->pci_dev_data = id->driver_data;
2725 
2726 	ret =  __mlx4_init_one(pdev, id->driver_data, priv);
2727 	if (ret)
2728 		kfree(priv);
2729 
2730 	return ret;
2731 }
2732 
2733 static void mlx4_unload_one(struct pci_dev *pdev)
2734 {
2735 	struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
2736 	struct mlx4_priv *priv = mlx4_priv(dev);
2737 	int               pci_dev_data;
2738 	int p;
2739 	int active_vfs = 0;
2740 
2741 	if (priv->removed)
2742 		return;
2743 
2744 	pci_dev_data = priv->pci_dev_data;
2745 
2746 	/* Disabling SR-IOV is not allowed while there are active vf's */
2747 	if (mlx4_is_master(dev)) {
2748 		active_vfs = mlx4_how_many_lives_vf(dev);
2749 		if (active_vfs) {
2750 			pr_warn("Removing PF when there are active VF's !!\n");
2751 			pr_warn("Will not disable SR-IOV.\n");
2752 		}
2753 	}
2754 	mlx4_stop_sense(dev);
2755 	mlx4_unregister_device(dev);
2756 
2757 	for (p = 1; p <= dev->caps.num_ports; p++) {
2758 		mlx4_cleanup_port_info(&priv->port[p]);
2759 		mlx4_CLOSE_PORT(dev, p);
2760 	}
2761 
2762 	if (mlx4_is_master(dev))
2763 		mlx4_free_resource_tracker(dev,
2764 					   RES_TR_FREE_SLAVES_ONLY);
2765 
2766 	mlx4_cleanup_counters_table(dev);
2767 	mlx4_cleanup_qp_table(dev);
2768 	mlx4_cleanup_srq_table(dev);
2769 	mlx4_cleanup_cq_table(dev);
2770 	mlx4_cmd_use_polling(dev);
2771 	mlx4_cleanup_eq_table(dev);
2772 	mlx4_cleanup_mcg_table(dev);
2773 	mlx4_cleanup_mr_table(dev);
2774 	mlx4_cleanup_xrcd_table(dev);
2775 	mlx4_cleanup_pd_table(dev);
2776 
2777 	if (mlx4_is_master(dev))
2778 		mlx4_free_resource_tracker(dev,
2779 					   RES_TR_FREE_STRUCTS_ONLY);
2780 
2781 	iounmap(priv->kar);
2782 	mlx4_uar_free(dev, &priv->driver_uar);
2783 	mlx4_cleanup_uar_table(dev);
2784 	if (!mlx4_is_slave(dev))
2785 		mlx4_clear_steering(dev);
2786 	mlx4_free_eq_table(dev);
2787 	if (mlx4_is_master(dev))
2788 		mlx4_multi_func_cleanup(dev);
2789 	mlx4_close_hca(dev);
2790 	if (mlx4_is_slave(dev))
2791 		mlx4_multi_func_cleanup(dev);
2792 	mlx4_cmd_cleanup(dev);
2793 
2794 	if (dev->flags & MLX4_FLAG_MSI_X)
2795 		pci_disable_msix(pdev);
2796 	if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
2797 		mlx4_warn(dev, "Disabling SR-IOV\n");
2798 		pci_disable_sriov(pdev);
2799 		dev->num_vfs = 0;
2800 	}
2801 
2802 	if (!mlx4_is_slave(dev))
2803 		mlx4_free_ownership(dev);
2804 
2805 	kfree(dev->caps.qp0_qkey);
2806 	kfree(dev->caps.qp0_tunnel);
2807 	kfree(dev->caps.qp0_proxy);
2808 	kfree(dev->caps.qp1_tunnel);
2809 	kfree(dev->caps.qp1_proxy);
2810 	kfree(dev->dev_vfs);
2811 
2812 	memset(priv, 0, sizeof(*priv));
2813 	priv->pci_dev_data = pci_dev_data;
2814 	priv->removed = 1;
2815 }
2816 
2817 static void mlx4_remove_one(struct pci_dev *pdev)
2818 {
2819 	struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
2820 	struct mlx4_priv *priv = mlx4_priv(dev);
2821 
2822 	mlx4_unload_one(pdev);
2823 	pci_release_regions(pdev);
2824 	pci_disable_device(pdev);
2825 	kfree(priv);
2826 	pci_set_drvdata(pdev, NULL);
2827 }
2828 
2829 int mlx4_restart_one(struct pci_dev *pdev)
2830 {
2831 	struct mlx4_dev	 *dev  = pci_get_drvdata(pdev);
2832 	struct mlx4_priv *priv = mlx4_priv(dev);
2833 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
2834 	int pci_dev_data, err, total_vfs;
2835 
2836 	pci_dev_data = priv->pci_dev_data;
2837 	total_vfs = dev->num_vfs;
2838 	memcpy(nvfs, dev->nvfs, sizeof(dev->nvfs));
2839 
2840 	mlx4_unload_one(pdev);
2841 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv);
2842 	if (err) {
2843 		mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
2844 			 __func__, pci_name(pdev), err);
2845 		return err;
2846 	}
2847 
2848 	return err;
2849 }
2850 
2851 static const struct pci_device_id mlx4_pci_table[] = {
2852 	/* MT25408 "Hermon" SDR */
2853 	{ PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2854 	/* MT25408 "Hermon" DDR */
2855 	{ PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2856 	/* MT25408 "Hermon" QDR */
2857 	{ PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2858 	/* MT25408 "Hermon" DDR PCIe gen2 */
2859 	{ PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2860 	/* MT25408 "Hermon" QDR PCIe gen2 */
2861 	{ PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2862 	/* MT25408 "Hermon" EN 10GigE */
2863 	{ PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2864 	/* MT25408 "Hermon" EN 10GigE PCIe gen2 */
2865 	{ PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2866 	/* MT25458 ConnectX EN 10GBASE-T 10GigE */
2867 	{ PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2868 	/* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
2869 	{ PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2870 	/* MT26468 ConnectX EN 10GigE PCIe gen2*/
2871 	{ PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2872 	/* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
2873 	{ PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2874 	/* MT26478 ConnectX2 40GigE PCIe gen2 */
2875 	{ PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2876 	/* MT25400 Family [ConnectX-2 Virtual Function] */
2877 	{ PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
2878 	/* MT27500 Family [ConnectX-3] */
2879 	{ PCI_VDEVICE(MELLANOX, 0x1003), 0 },
2880 	/* MT27500 Family [ConnectX-3 Virtual Function] */
2881 	{ PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
2882 	{ PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
2883 	{ PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
2884 	{ PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
2885 	{ PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
2886 	{ PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
2887 	{ PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
2888 	{ PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
2889 	{ PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
2890 	{ PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
2891 	{ PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
2892 	{ PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
2893 	{ PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
2894 	{ 0, }
2895 };
2896 
2897 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
2898 
2899 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
2900 					      pci_channel_state_t state)
2901 {
2902 	mlx4_unload_one(pdev);
2903 
2904 	return state == pci_channel_io_perm_failure ?
2905 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
2906 }
2907 
2908 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
2909 {
2910 	struct mlx4_dev	 *dev  = pci_get_drvdata(pdev);
2911 	struct mlx4_priv *priv = mlx4_priv(dev);
2912 	int               ret;
2913 
2914 	ret = __mlx4_init_one(pdev, priv->pci_dev_data, priv);
2915 
2916 	return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
2917 }
2918 
2919 static const struct pci_error_handlers mlx4_err_handler = {
2920 	.error_detected = mlx4_pci_err_detected,
2921 	.slot_reset     = mlx4_pci_slot_reset,
2922 };
2923 
2924 static struct pci_driver mlx4_driver = {
2925 	.name		= DRV_NAME,
2926 	.id_table	= mlx4_pci_table,
2927 	.probe		= mlx4_init_one,
2928 	.shutdown	= mlx4_unload_one,
2929 	.remove		= mlx4_remove_one,
2930 	.err_handler    = &mlx4_err_handler,
2931 };
2932 
2933 static int __init mlx4_verify_params(void)
2934 {
2935 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
2936 		pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
2937 		return -1;
2938 	}
2939 
2940 	if (log_num_vlan != 0)
2941 		pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
2942 			MLX4_LOG_NUM_VLANS);
2943 
2944 	if (use_prio != 0)
2945 		pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
2946 
2947 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
2948 		pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
2949 			log_mtts_per_seg);
2950 		return -1;
2951 	}
2952 
2953 	/* Check if module param for ports type has legal combination */
2954 	if (port_type_array[0] == false && port_type_array[1] == true) {
2955 		pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
2956 		port_type_array[0] = true;
2957 	}
2958 
2959 	if (mlx4_log_num_mgm_entry_size != -1 &&
2960 	    (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
2961 	     mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE)) {
2962 		pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-1 or %d..%d)\n",
2963 			mlx4_log_num_mgm_entry_size,
2964 			MLX4_MIN_MGM_LOG_ENTRY_SIZE,
2965 			MLX4_MAX_MGM_LOG_ENTRY_SIZE);
2966 		return -1;
2967 	}
2968 
2969 	return 0;
2970 }
2971 
2972 static int __init mlx4_init(void)
2973 {
2974 	int ret;
2975 
2976 	if (mlx4_verify_params())
2977 		return -EINVAL;
2978 
2979 	mlx4_catas_init();
2980 
2981 	mlx4_wq = create_singlethread_workqueue("mlx4");
2982 	if (!mlx4_wq)
2983 		return -ENOMEM;
2984 
2985 	ret = pci_register_driver(&mlx4_driver);
2986 	if (ret < 0)
2987 		destroy_workqueue(mlx4_wq);
2988 	return ret < 0 ? ret : 0;
2989 }
2990 
2991 static void __exit mlx4_cleanup(void)
2992 {
2993 	pci_unregister_driver(&mlx4_driver);
2994 	destroy_workqueue(mlx4_wq);
2995 }
2996 
2997 module_init(mlx4_init);
2998 module_exit(mlx4_cleanup);
2999