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/netdevice.h>
45 #include <linux/kmod.h>
46 
47 #include <linux/mlx4/device.h>
48 #include <linux/mlx4/doorbell.h>
49 
50 #include "mlx4.h"
51 #include "fw.h"
52 #include "icm.h"
53 
54 MODULE_AUTHOR("Roland Dreier");
55 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
56 MODULE_LICENSE("Dual BSD/GPL");
57 MODULE_VERSION(DRV_VERSION);
58 
59 struct workqueue_struct *mlx4_wq;
60 
61 #ifdef CONFIG_MLX4_DEBUG
62 
63 int mlx4_debug_level = 0;
64 module_param_named(debug_level, mlx4_debug_level, int, 0644);
65 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
66 
67 #endif /* CONFIG_MLX4_DEBUG */
68 
69 #ifdef CONFIG_PCI_MSI
70 
71 static int msi_x = 1;
72 module_param(msi_x, int, 0444);
73 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
74 
75 #else /* CONFIG_PCI_MSI */
76 
77 #define msi_x (0)
78 
79 #endif /* CONFIG_PCI_MSI */
80 
81 static int num_vfs;
82 module_param(num_vfs, int, 0444);
83 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0");
84 
85 static int probe_vf;
86 module_param(probe_vf, int, 0644);
87 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)");
88 
89 int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
90 module_param_named(log_num_mgm_entry_size,
91 			mlx4_log_num_mgm_entry_size, int, 0444);
92 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
93 					 " of qp per mcg, for example:"
94 					 " 10 gives 248.range: 7 <="
95 					 " log_num_mgm_entry_size <= 12."
96 					 " To activate device managed"
97 					 " flow steering when available, set to -1");
98 
99 static bool enable_64b_cqe_eqe = true;
100 module_param(enable_64b_cqe_eqe, bool, 0444);
101 MODULE_PARM_DESC(enable_64b_cqe_eqe,
102 		 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
103 
104 #define HCA_GLOBAL_CAP_MASK            0
105 
106 #define PF_CONTEXT_BEHAVIOUR_MASK	MLX4_FUNC_CAP_64B_EQE_CQE
107 
108 static char mlx4_version[] =
109 	DRV_NAME ": Mellanox ConnectX core driver v"
110 	DRV_VERSION " (" DRV_RELDATE ")\n";
111 
112 static struct mlx4_profile default_profile = {
113 	.num_qp		= 1 << 18,
114 	.num_srq	= 1 << 16,
115 	.rdmarc_per_qp	= 1 << 4,
116 	.num_cq		= 1 << 16,
117 	.num_mcg	= 1 << 13,
118 	.num_mpt	= 1 << 19,
119 	.num_mtt	= 1 << 20, /* It is really num mtt segements */
120 };
121 
122 static int log_num_mac = 7;
123 module_param_named(log_num_mac, log_num_mac, int, 0444);
124 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
125 
126 static int log_num_vlan;
127 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
128 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
129 /* Log2 max number of VLANs per ETH port (0-7) */
130 #define MLX4_LOG_NUM_VLANS 7
131 
132 static bool use_prio;
133 module_param_named(use_prio, use_prio, bool, 0444);
134 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports "
135 		  "(0/1, default 0)");
136 
137 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
138 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
139 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
140 
141 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
142 static int arr_argc = 2;
143 module_param_array(port_type_array, int, &arr_argc, 0444);
144 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
145 				"1 for IB, 2 for Ethernet");
146 
147 struct mlx4_port_config {
148 	struct list_head list;
149 	enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
150 	struct pci_dev *pdev;
151 };
152 
153 int mlx4_check_port_params(struct mlx4_dev *dev,
154 			   enum mlx4_port_type *port_type)
155 {
156 	int i;
157 
158 	for (i = 0; i < dev->caps.num_ports - 1; i++) {
159 		if (port_type[i] != port_type[i + 1]) {
160 			if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
161 				mlx4_err(dev, "Only same port types supported "
162 					 "on this HCA, aborting.\n");
163 				return -EINVAL;
164 			}
165 		}
166 	}
167 
168 	for (i = 0; i < dev->caps.num_ports; i++) {
169 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
170 			mlx4_err(dev, "Requested port type for port %d is not "
171 				      "supported on this HCA\n", i + 1);
172 			return -EINVAL;
173 		}
174 	}
175 	return 0;
176 }
177 
178 static void mlx4_set_port_mask(struct mlx4_dev *dev)
179 {
180 	int i;
181 
182 	for (i = 1; i <= dev->caps.num_ports; ++i)
183 		dev->caps.port_mask[i] = dev->caps.port_type[i];
184 }
185 
186 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
187 {
188 	int err;
189 	int i;
190 
191 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
192 	if (err) {
193 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
194 		return err;
195 	}
196 
197 	if (dev_cap->min_page_sz > PAGE_SIZE) {
198 		mlx4_err(dev, "HCA minimum page size of %d bigger than "
199 			 "kernel PAGE_SIZE of %ld, aborting.\n",
200 			 dev_cap->min_page_sz, PAGE_SIZE);
201 		return -ENODEV;
202 	}
203 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
204 		mlx4_err(dev, "HCA has %d ports, but we only support %d, "
205 			 "aborting.\n",
206 			 dev_cap->num_ports, MLX4_MAX_PORTS);
207 		return -ENODEV;
208 	}
209 
210 	if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
211 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than "
212 			 "PCI resource 2 size of 0x%llx, aborting.\n",
213 			 dev_cap->uar_size,
214 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
215 		return -ENODEV;
216 	}
217 
218 	dev->caps.num_ports	     = dev_cap->num_ports;
219 	dev->phys_caps.num_phys_eqs  = MLX4_MAX_EQ_NUM;
220 	for (i = 1; i <= dev->caps.num_ports; ++i) {
221 		dev->caps.vl_cap[i]	    = dev_cap->max_vl[i];
222 		dev->caps.ib_mtu_cap[i]	    = dev_cap->ib_mtu[i];
223 		dev->phys_caps.gid_phys_table_len[i]  = dev_cap->max_gids[i];
224 		dev->phys_caps.pkey_phys_table_len[i] = dev_cap->max_pkeys[i];
225 		/* set gid and pkey table operating lengths by default
226 		 * to non-sriov values */
227 		dev->caps.gid_table_len[i]  = dev_cap->max_gids[i];
228 		dev->caps.pkey_table_len[i] = dev_cap->max_pkeys[i];
229 		dev->caps.port_width_cap[i] = dev_cap->max_port_width[i];
230 		dev->caps.eth_mtu_cap[i]    = dev_cap->eth_mtu[i];
231 		dev->caps.def_mac[i]        = dev_cap->def_mac[i];
232 		dev->caps.supported_type[i] = dev_cap->supported_port_types[i];
233 		dev->caps.suggested_type[i] = dev_cap->suggested_type[i];
234 		dev->caps.default_sense[i] = dev_cap->default_sense[i];
235 		dev->caps.trans_type[i]	    = dev_cap->trans_type[i];
236 		dev->caps.vendor_oui[i]     = dev_cap->vendor_oui[i];
237 		dev->caps.wavelength[i]     = dev_cap->wavelength[i];
238 		dev->caps.trans_code[i]     = dev_cap->trans_code[i];
239 	}
240 
241 	dev->caps.uar_page_size	     = PAGE_SIZE;
242 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
243 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
244 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
245 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
246 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
247 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
248 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
249 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
250 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
251 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
252 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
253 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
254 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
255 	/*
256 	 * Subtract 1 from the limit because we need to allocate a
257 	 * spare CQE so the HCA HW can tell the difference between an
258 	 * empty CQ and a full CQ.
259 	 */
260 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
261 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
262 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
263 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
264 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
265 
266 	/* The first 128 UARs are used for EQ doorbells */
267 	dev->caps.reserved_uars	     = max_t(int, 128, dev_cap->reserved_uars);
268 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
269 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
270 					dev_cap->reserved_xrcds : 0;
271 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
272 					dev_cap->max_xrcds : 0;
273 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
274 
275 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
276 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
277 	dev->caps.flags		     = dev_cap->flags;
278 	dev->caps.flags2	     = dev_cap->flags2;
279 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
280 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
281 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
282 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
283 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
284 
285 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
286 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
287 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
288 	/* Don't do sense port on multifunction devices (for now at least) */
289 	if (mlx4_is_mfunc(dev))
290 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
291 
292 	dev->caps.log_num_macs  = log_num_mac;
293 	dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
294 	dev->caps.log_num_prios = use_prio ? 3 : 0;
295 
296 	for (i = 1; i <= dev->caps.num_ports; ++i) {
297 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
298 		if (dev->caps.supported_type[i]) {
299 			/* if only ETH is supported - assign ETH */
300 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
301 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
302 			/* if only IB is supported, assign IB */
303 			else if (dev->caps.supported_type[i] ==
304 				 MLX4_PORT_TYPE_IB)
305 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
306 			else {
307 				/* if IB and ETH are supported, we set the port
308 				 * type according to user selection of port type;
309 				 * if user selected none, take the FW hint */
310 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
311 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
312 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
313 				else
314 					dev->caps.port_type[i] = port_type_array[i - 1];
315 			}
316 		}
317 		/*
318 		 * Link sensing is allowed on the port if 3 conditions are true:
319 		 * 1. Both protocols are supported on the port.
320 		 * 2. Different types are supported on the port
321 		 * 3. FW declared that it supports link sensing
322 		 */
323 		mlx4_priv(dev)->sense.sense_allowed[i] =
324 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
325 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
326 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
327 
328 		/*
329 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
330 		 * and perform sense_port FW command to try and set the correct
331 		 * port type from beginning
332 		 */
333 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
334 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
335 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
336 			mlx4_SENSE_PORT(dev, i, &sensed_port);
337 			if (sensed_port != MLX4_PORT_TYPE_NONE)
338 				dev->caps.port_type[i] = sensed_port;
339 		} else {
340 			dev->caps.possible_type[i] = dev->caps.port_type[i];
341 		}
342 
343 		if (dev->caps.log_num_macs > dev_cap->log_max_macs[i]) {
344 			dev->caps.log_num_macs = dev_cap->log_max_macs[i];
345 			mlx4_warn(dev, "Requested number of MACs is too much "
346 				  "for port %d, reducing to %d.\n",
347 				  i, 1 << dev->caps.log_num_macs);
348 		}
349 		if (dev->caps.log_num_vlans > dev_cap->log_max_vlans[i]) {
350 			dev->caps.log_num_vlans = dev_cap->log_max_vlans[i];
351 			mlx4_warn(dev, "Requested number of VLANs is too much "
352 				  "for port %d, reducing to %d.\n",
353 				  i, 1 << dev->caps.log_num_vlans);
354 		}
355 	}
356 
357 	dev->caps.max_counters = 1 << ilog2(dev_cap->max_counters);
358 
359 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
360 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
361 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
362 		(1 << dev->caps.log_num_macs) *
363 		(1 << dev->caps.log_num_vlans) *
364 		(1 << dev->caps.log_num_prios) *
365 		dev->caps.num_ports;
366 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
367 
368 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
369 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
370 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
371 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
372 
373 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
374 
375 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
376 		if (dev_cap->flags &
377 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
378 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
379 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
380 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
381 		}
382 	}
383 
384 	if ((dev->caps.flags &
385 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
386 	    mlx4_is_master(dev))
387 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
388 
389 	return 0;
390 }
391 
392 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
393 				       enum pci_bus_speed *speed,
394 				       enum pcie_link_width *width)
395 {
396 	u32 lnkcap1, lnkcap2;
397 	int err1, err2;
398 
399 #define  PCIE_MLW_CAP_SHIFT 4	/* start of MLW mask in link capabilities */
400 
401 	*speed = PCI_SPEED_UNKNOWN;
402 	*width = PCIE_LNK_WIDTH_UNKNOWN;
403 
404 	err1 = pcie_capability_read_dword(dev->pdev, PCI_EXP_LNKCAP, &lnkcap1);
405 	err2 = pcie_capability_read_dword(dev->pdev, PCI_EXP_LNKCAP2, &lnkcap2);
406 	if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
407 		if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
408 			*speed = PCIE_SPEED_8_0GT;
409 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
410 			*speed = PCIE_SPEED_5_0GT;
411 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
412 			*speed = PCIE_SPEED_2_5GT;
413 	}
414 	if (!err1) {
415 		*width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
416 		if (!lnkcap2) { /* pre-r3.0 */
417 			if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
418 				*speed = PCIE_SPEED_5_0GT;
419 			else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
420 				*speed = PCIE_SPEED_2_5GT;
421 		}
422 	}
423 
424 	if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
425 		return err1 ? err1 :
426 			err2 ? err2 : -EINVAL;
427 	}
428 	return 0;
429 }
430 
431 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
432 {
433 	enum pcie_link_width width, width_cap;
434 	enum pci_bus_speed speed, speed_cap;
435 	int err;
436 
437 #define PCIE_SPEED_STR(speed) \
438 	(speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
439 	 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
440 	 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
441 	 "Unknown")
442 
443 	err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
444 	if (err) {
445 		mlx4_warn(dev,
446 			  "Unable to determine PCIe device BW capabilities\n");
447 		return;
448 	}
449 
450 	err = pcie_get_minimum_link(dev->pdev, &speed, &width);
451 	if (err || speed == PCI_SPEED_UNKNOWN ||
452 	    width == PCIE_LNK_WIDTH_UNKNOWN) {
453 		mlx4_warn(dev,
454 			  "Unable to determine PCI device chain minimum BW\n");
455 		return;
456 	}
457 
458 	if (width != width_cap || speed != speed_cap)
459 		mlx4_warn(dev,
460 			  "PCIe BW is different than device's capability\n");
461 
462 	mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
463 		  PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
464 	mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
465 		  width, width_cap);
466 	return;
467 }
468 
469 /*The function checks if there are live vf, return the num of them*/
470 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
471 {
472 	struct mlx4_priv *priv = mlx4_priv(dev);
473 	struct mlx4_slave_state *s_state;
474 	int i;
475 	int ret = 0;
476 
477 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
478 		s_state = &priv->mfunc.master.slave_state[i];
479 		if (s_state->active && s_state->last_cmd !=
480 		    MLX4_COMM_CMD_RESET) {
481 			mlx4_warn(dev, "%s: slave: %d is still active\n",
482 				  __func__, i);
483 			ret++;
484 		}
485 	}
486 	return ret;
487 }
488 
489 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
490 {
491 	u32 qk = MLX4_RESERVED_QKEY_BASE;
492 
493 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
494 	    qpn < dev->phys_caps.base_proxy_sqpn)
495 		return -EINVAL;
496 
497 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
498 		/* tunnel qp */
499 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
500 	else
501 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
502 	*qkey = qk;
503 	return 0;
504 }
505 EXPORT_SYMBOL(mlx4_get_parav_qkey);
506 
507 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
508 {
509 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
510 
511 	if (!mlx4_is_master(dev))
512 		return;
513 
514 	priv->virt2phys_pkey[slave][port - 1][i] = val;
515 }
516 EXPORT_SYMBOL(mlx4_sync_pkey_table);
517 
518 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
519 {
520 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
521 
522 	if (!mlx4_is_master(dev))
523 		return;
524 
525 	priv->slave_node_guids[slave] = guid;
526 }
527 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
528 
529 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
530 {
531 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
532 
533 	if (!mlx4_is_master(dev))
534 		return 0;
535 
536 	return priv->slave_node_guids[slave];
537 }
538 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
539 
540 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
541 {
542 	struct mlx4_priv *priv = mlx4_priv(dev);
543 	struct mlx4_slave_state *s_slave;
544 
545 	if (!mlx4_is_master(dev))
546 		return 0;
547 
548 	s_slave = &priv->mfunc.master.slave_state[slave];
549 	return !!s_slave->active;
550 }
551 EXPORT_SYMBOL(mlx4_is_slave_active);
552 
553 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
554 				       struct mlx4_dev_cap *dev_cap,
555 				       struct mlx4_init_hca_param *hca_param)
556 {
557 	dev->caps.steering_mode = hca_param->steering_mode;
558 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
559 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
560 		dev->caps.fs_log_max_ucast_qp_range_size =
561 			dev_cap->fs_log_max_ucast_qp_range_size;
562 	} else
563 		dev->caps.num_qp_per_mgm =
564 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
565 
566 	mlx4_dbg(dev, "Steering mode is: %s\n",
567 		 mlx4_steering_mode_str(dev->caps.steering_mode));
568 }
569 
570 static int mlx4_slave_cap(struct mlx4_dev *dev)
571 {
572 	int			   err;
573 	u32			   page_size;
574 	struct mlx4_dev_cap	   dev_cap;
575 	struct mlx4_func_cap	   func_cap;
576 	struct mlx4_init_hca_param hca_param;
577 	int			   i;
578 
579 	memset(&hca_param, 0, sizeof(hca_param));
580 	err = mlx4_QUERY_HCA(dev, &hca_param);
581 	if (err) {
582 		mlx4_err(dev, "QUERY_HCA command failed, aborting.\n");
583 		return err;
584 	}
585 
586 	/*fail if the hca has an unknown capability */
587 	if ((hca_param.global_caps | HCA_GLOBAL_CAP_MASK) !=
588 	    HCA_GLOBAL_CAP_MASK) {
589 		mlx4_err(dev, "Unknown hca global capabilities\n");
590 		return -ENOSYS;
591 	}
592 
593 	mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
594 
595 	dev->caps.hca_core_clock = hca_param.hca_core_clock;
596 
597 	memset(&dev_cap, 0, sizeof(dev_cap));
598 	dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
599 	err = mlx4_dev_cap(dev, &dev_cap);
600 	if (err) {
601 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
602 		return err;
603 	}
604 
605 	err = mlx4_QUERY_FW(dev);
606 	if (err)
607 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version.\n");
608 
609 	page_size = ~dev->caps.page_size_cap + 1;
610 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
611 	if (page_size > PAGE_SIZE) {
612 		mlx4_err(dev, "HCA minimum page size of %d bigger than "
613 			 "kernel PAGE_SIZE of %ld, aborting.\n",
614 			 page_size, PAGE_SIZE);
615 		return -ENODEV;
616 	}
617 
618 	/* slave gets uar page size from QUERY_HCA fw command */
619 	dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
620 
621 	/* TODO: relax this assumption */
622 	if (dev->caps.uar_page_size != PAGE_SIZE) {
623 		mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
624 			 dev->caps.uar_page_size, PAGE_SIZE);
625 		return -ENODEV;
626 	}
627 
628 	memset(&func_cap, 0, sizeof(func_cap));
629 	err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
630 	if (err) {
631 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d).\n",
632 			  err);
633 		return err;
634 	}
635 
636 	if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
637 	    PF_CONTEXT_BEHAVIOUR_MASK) {
638 		mlx4_err(dev, "Unknown pf context behaviour\n");
639 		return -ENOSYS;
640 	}
641 
642 	dev->caps.num_ports		= func_cap.num_ports;
643 	dev->quotas.qp			= func_cap.qp_quota;
644 	dev->quotas.srq			= func_cap.srq_quota;
645 	dev->quotas.cq			= func_cap.cq_quota;
646 	dev->quotas.mpt			= func_cap.mpt_quota;
647 	dev->quotas.mtt			= func_cap.mtt_quota;
648 	dev->caps.num_qps		= 1 << hca_param.log_num_qps;
649 	dev->caps.num_srqs		= 1 << hca_param.log_num_srqs;
650 	dev->caps.num_cqs		= 1 << hca_param.log_num_cqs;
651 	dev->caps.num_mpts		= 1 << hca_param.log_mpt_sz;
652 	dev->caps.num_eqs		= func_cap.max_eq;
653 	dev->caps.reserved_eqs		= func_cap.reserved_eq;
654 	dev->caps.num_pds               = MLX4_NUM_PDS;
655 	dev->caps.num_mgms              = 0;
656 	dev->caps.num_amgms             = 0;
657 
658 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
659 		mlx4_err(dev, "HCA has %d ports, but we only support %d, "
660 			 "aborting.\n", dev->caps.num_ports, MLX4_MAX_PORTS);
661 		return -ENODEV;
662 	}
663 
664 	dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
665 	dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
666 	dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
667 	dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
668 
669 	if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
670 	    !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy) {
671 		err = -ENOMEM;
672 		goto err_mem;
673 	}
674 
675 	for (i = 1; i <= dev->caps.num_ports; ++i) {
676 		err = mlx4_QUERY_FUNC_CAP(dev, (u32) i, &func_cap);
677 		if (err) {
678 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for"
679 				 " port %d, aborting (%d).\n", i, err);
680 			goto err_mem;
681 		}
682 		dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
683 		dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
684 		dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
685 		dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
686 		dev->caps.port_mask[i] = dev->caps.port_type[i];
687 		dev->caps.phys_port_id[i] = func_cap.phys_port_id;
688 		if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
689 						    &dev->caps.gid_table_len[i],
690 						    &dev->caps.pkey_table_len[i]))
691 			goto err_mem;
692 	}
693 
694 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
695 				       dev->caps.reserved_uars) >
696 				       pci_resource_len(dev->pdev, 2)) {
697 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than "
698 			 "PCI resource 2 size of 0x%llx, aborting.\n",
699 			 dev->caps.uar_page_size * dev->caps.num_uars,
700 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
701 		goto err_mem;
702 	}
703 
704 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
705 		dev->caps.eqe_size   = 64;
706 		dev->caps.eqe_factor = 1;
707 	} else {
708 		dev->caps.eqe_size   = 32;
709 		dev->caps.eqe_factor = 0;
710 	}
711 
712 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
713 		dev->caps.cqe_size   = 64;
714 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_64B_CQE;
715 	} else {
716 		dev->caps.cqe_size   = 32;
717 	}
718 
719 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
720 	mlx4_warn(dev, "Timestamping is not supported in slave mode.\n");
721 
722 	slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
723 
724 	return 0;
725 
726 err_mem:
727 	kfree(dev->caps.qp0_tunnel);
728 	kfree(dev->caps.qp0_proxy);
729 	kfree(dev->caps.qp1_tunnel);
730 	kfree(dev->caps.qp1_proxy);
731 	dev->caps.qp0_tunnel = dev->caps.qp0_proxy =
732 		dev->caps.qp1_tunnel = dev->caps.qp1_proxy = NULL;
733 
734 	return err;
735 }
736 
737 static void mlx4_request_modules(struct mlx4_dev *dev)
738 {
739 	int port;
740 	int has_ib_port = false;
741 	int has_eth_port = false;
742 #define EN_DRV_NAME	"mlx4_en"
743 #define IB_DRV_NAME	"mlx4_ib"
744 
745 	for (port = 1; port <= dev->caps.num_ports; port++) {
746 		if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
747 			has_ib_port = true;
748 		else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
749 			has_eth_port = true;
750 	}
751 
752 	if (has_ib_port)
753 		request_module_nowait(IB_DRV_NAME);
754 	if (has_eth_port)
755 		request_module_nowait(EN_DRV_NAME);
756 }
757 
758 /*
759  * Change the port configuration of the device.
760  * Every user of this function must hold the port mutex.
761  */
762 int mlx4_change_port_types(struct mlx4_dev *dev,
763 			   enum mlx4_port_type *port_types)
764 {
765 	int err = 0;
766 	int change = 0;
767 	int port;
768 
769 	for (port = 0; port <  dev->caps.num_ports; port++) {
770 		/* Change the port type only if the new type is different
771 		 * from the current, and not set to Auto */
772 		if (port_types[port] != dev->caps.port_type[port + 1])
773 			change = 1;
774 	}
775 	if (change) {
776 		mlx4_unregister_device(dev);
777 		for (port = 1; port <= dev->caps.num_ports; port++) {
778 			mlx4_CLOSE_PORT(dev, port);
779 			dev->caps.port_type[port] = port_types[port - 1];
780 			err = mlx4_SET_PORT(dev, port, -1);
781 			if (err) {
782 				mlx4_err(dev, "Failed to set port %d, "
783 					      "aborting\n", port);
784 				goto out;
785 			}
786 		}
787 		mlx4_set_port_mask(dev);
788 		err = mlx4_register_device(dev);
789 		if (err) {
790 			mlx4_err(dev, "Failed to register device\n");
791 			goto out;
792 		}
793 		mlx4_request_modules(dev);
794 	}
795 
796 out:
797 	return err;
798 }
799 
800 static ssize_t show_port_type(struct device *dev,
801 			      struct device_attribute *attr,
802 			      char *buf)
803 {
804 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
805 						   port_attr);
806 	struct mlx4_dev *mdev = info->dev;
807 	char type[8];
808 
809 	sprintf(type, "%s",
810 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
811 		"ib" : "eth");
812 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
813 		sprintf(buf, "auto (%s)\n", type);
814 	else
815 		sprintf(buf, "%s\n", type);
816 
817 	return strlen(buf);
818 }
819 
820 static ssize_t set_port_type(struct device *dev,
821 			     struct device_attribute *attr,
822 			     const char *buf, size_t count)
823 {
824 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
825 						   port_attr);
826 	struct mlx4_dev *mdev = info->dev;
827 	struct mlx4_priv *priv = mlx4_priv(mdev);
828 	enum mlx4_port_type types[MLX4_MAX_PORTS];
829 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
830 	int i;
831 	int err = 0;
832 
833 	if (!strcmp(buf, "ib\n"))
834 		info->tmp_type = MLX4_PORT_TYPE_IB;
835 	else if (!strcmp(buf, "eth\n"))
836 		info->tmp_type = MLX4_PORT_TYPE_ETH;
837 	else if (!strcmp(buf, "auto\n"))
838 		info->tmp_type = MLX4_PORT_TYPE_AUTO;
839 	else {
840 		mlx4_err(mdev, "%s is not supported port type\n", buf);
841 		return -EINVAL;
842 	}
843 
844 	mlx4_stop_sense(mdev);
845 	mutex_lock(&priv->port_mutex);
846 	/* Possible type is always the one that was delivered */
847 	mdev->caps.possible_type[info->port] = info->tmp_type;
848 
849 	for (i = 0; i < mdev->caps.num_ports; i++) {
850 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
851 					mdev->caps.possible_type[i+1];
852 		if (types[i] == MLX4_PORT_TYPE_AUTO)
853 			types[i] = mdev->caps.port_type[i+1];
854 	}
855 
856 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
857 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
858 		for (i = 1; i <= mdev->caps.num_ports; i++) {
859 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
860 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
861 				err = -EINVAL;
862 			}
863 		}
864 	}
865 	if (err) {
866 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. "
867 			       "Set only 'eth' or 'ib' for both ports "
868 			       "(should be the same)\n");
869 		goto out;
870 	}
871 
872 	mlx4_do_sense_ports(mdev, new_types, types);
873 
874 	err = mlx4_check_port_params(mdev, new_types);
875 	if (err)
876 		goto out;
877 
878 	/* We are about to apply the changes after the configuration
879 	 * was verified, no need to remember the temporary types
880 	 * any more */
881 	for (i = 0; i < mdev->caps.num_ports; i++)
882 		priv->port[i + 1].tmp_type = 0;
883 
884 	err = mlx4_change_port_types(mdev, new_types);
885 
886 out:
887 	mlx4_start_sense(mdev);
888 	mutex_unlock(&priv->port_mutex);
889 	return err ? err : count;
890 }
891 
892 enum ibta_mtu {
893 	IB_MTU_256  = 1,
894 	IB_MTU_512  = 2,
895 	IB_MTU_1024 = 3,
896 	IB_MTU_2048 = 4,
897 	IB_MTU_4096 = 5
898 };
899 
900 static inline int int_to_ibta_mtu(int mtu)
901 {
902 	switch (mtu) {
903 	case 256:  return IB_MTU_256;
904 	case 512:  return IB_MTU_512;
905 	case 1024: return IB_MTU_1024;
906 	case 2048: return IB_MTU_2048;
907 	case 4096: return IB_MTU_4096;
908 	default: return -1;
909 	}
910 }
911 
912 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
913 {
914 	switch (mtu) {
915 	case IB_MTU_256:  return  256;
916 	case IB_MTU_512:  return  512;
917 	case IB_MTU_1024: return 1024;
918 	case IB_MTU_2048: return 2048;
919 	case IB_MTU_4096: return 4096;
920 	default: return -1;
921 	}
922 }
923 
924 static ssize_t show_port_ib_mtu(struct device *dev,
925 			     struct device_attribute *attr,
926 			     char *buf)
927 {
928 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
929 						   port_mtu_attr);
930 	struct mlx4_dev *mdev = info->dev;
931 
932 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
933 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
934 
935 	sprintf(buf, "%d\n",
936 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
937 	return strlen(buf);
938 }
939 
940 static ssize_t set_port_ib_mtu(struct device *dev,
941 			     struct device_attribute *attr,
942 			     const char *buf, size_t count)
943 {
944 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
945 						   port_mtu_attr);
946 	struct mlx4_dev *mdev = info->dev;
947 	struct mlx4_priv *priv = mlx4_priv(mdev);
948 	int err, port, mtu, ibta_mtu = -1;
949 
950 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
951 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
952 		return -EINVAL;
953 	}
954 
955 	err = kstrtoint(buf, 0, &mtu);
956 	if (!err)
957 		ibta_mtu = int_to_ibta_mtu(mtu);
958 
959 	if (err || ibta_mtu < 0) {
960 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
961 		return -EINVAL;
962 	}
963 
964 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
965 
966 	mlx4_stop_sense(mdev);
967 	mutex_lock(&priv->port_mutex);
968 	mlx4_unregister_device(mdev);
969 	for (port = 1; port <= mdev->caps.num_ports; port++) {
970 		mlx4_CLOSE_PORT(mdev, port);
971 		err = mlx4_SET_PORT(mdev, port, -1);
972 		if (err) {
973 			mlx4_err(mdev, "Failed to set port %d, "
974 				      "aborting\n", port);
975 			goto err_set_port;
976 		}
977 	}
978 	err = mlx4_register_device(mdev);
979 err_set_port:
980 	mutex_unlock(&priv->port_mutex);
981 	mlx4_start_sense(mdev);
982 	return err ? err : count;
983 }
984 
985 static int mlx4_load_fw(struct mlx4_dev *dev)
986 {
987 	struct mlx4_priv *priv = mlx4_priv(dev);
988 	int err;
989 
990 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
991 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
992 	if (!priv->fw.fw_icm) {
993 		mlx4_err(dev, "Couldn't allocate FW area, aborting.\n");
994 		return -ENOMEM;
995 	}
996 
997 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
998 	if (err) {
999 		mlx4_err(dev, "MAP_FA command failed, aborting.\n");
1000 		goto err_free;
1001 	}
1002 
1003 	err = mlx4_RUN_FW(dev);
1004 	if (err) {
1005 		mlx4_err(dev, "RUN_FW command failed, aborting.\n");
1006 		goto err_unmap_fa;
1007 	}
1008 
1009 	return 0;
1010 
1011 err_unmap_fa:
1012 	mlx4_UNMAP_FA(dev);
1013 
1014 err_free:
1015 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1016 	return err;
1017 }
1018 
1019 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1020 				int cmpt_entry_sz)
1021 {
1022 	struct mlx4_priv *priv = mlx4_priv(dev);
1023 	int err;
1024 	int num_eqs;
1025 
1026 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1027 				  cmpt_base +
1028 				  ((u64) (MLX4_CMPT_TYPE_QP *
1029 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1030 				  cmpt_entry_sz, dev->caps.num_qps,
1031 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1032 				  0, 0);
1033 	if (err)
1034 		goto err;
1035 
1036 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1037 				  cmpt_base +
1038 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1039 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1040 				  cmpt_entry_sz, dev->caps.num_srqs,
1041 				  dev->caps.reserved_srqs, 0, 0);
1042 	if (err)
1043 		goto err_qp;
1044 
1045 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1046 				  cmpt_base +
1047 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1048 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1049 				  cmpt_entry_sz, dev->caps.num_cqs,
1050 				  dev->caps.reserved_cqs, 0, 0);
1051 	if (err)
1052 		goto err_srq;
1053 
1054 	num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
1055 		  dev->caps.num_eqs;
1056 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1057 				  cmpt_base +
1058 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1059 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1060 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1061 	if (err)
1062 		goto err_cq;
1063 
1064 	return 0;
1065 
1066 err_cq:
1067 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1068 
1069 err_srq:
1070 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1071 
1072 err_qp:
1073 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1074 
1075 err:
1076 	return err;
1077 }
1078 
1079 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1080 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1081 {
1082 	struct mlx4_priv *priv = mlx4_priv(dev);
1083 	u64 aux_pages;
1084 	int num_eqs;
1085 	int err;
1086 
1087 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1088 	if (err) {
1089 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting.\n");
1090 		return err;
1091 	}
1092 
1093 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory.\n",
1094 		 (unsigned long long) icm_size >> 10,
1095 		 (unsigned long long) aux_pages << 2);
1096 
1097 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1098 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1099 	if (!priv->fw.aux_icm) {
1100 		mlx4_err(dev, "Couldn't allocate aux memory, aborting.\n");
1101 		return -ENOMEM;
1102 	}
1103 
1104 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1105 	if (err) {
1106 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting.\n");
1107 		goto err_free_aux;
1108 	}
1109 
1110 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1111 	if (err) {
1112 		mlx4_err(dev, "Failed to map cMPT context memory, aborting.\n");
1113 		goto err_unmap_aux;
1114 	}
1115 
1116 
1117 	num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
1118 		   dev->caps.num_eqs;
1119 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1120 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1121 				  num_eqs, num_eqs, 0, 0);
1122 	if (err) {
1123 		mlx4_err(dev, "Failed to map EQ context memory, aborting.\n");
1124 		goto err_unmap_cmpt;
1125 	}
1126 
1127 	/*
1128 	 * Reserved MTT entries must be aligned up to a cacheline
1129 	 * boundary, since the FW will write to them, while the driver
1130 	 * writes to all other MTT entries. (The variable
1131 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1132 	 * size, not the raw entry size)
1133 	 */
1134 	dev->caps.reserved_mtts =
1135 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1136 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1137 
1138 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1139 				  init_hca->mtt_base,
1140 				  dev->caps.mtt_entry_sz,
1141 				  dev->caps.num_mtts,
1142 				  dev->caps.reserved_mtts, 1, 0);
1143 	if (err) {
1144 		mlx4_err(dev, "Failed to map MTT context memory, aborting.\n");
1145 		goto err_unmap_eq;
1146 	}
1147 
1148 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1149 				  init_hca->dmpt_base,
1150 				  dev_cap->dmpt_entry_sz,
1151 				  dev->caps.num_mpts,
1152 				  dev->caps.reserved_mrws, 1, 1);
1153 	if (err) {
1154 		mlx4_err(dev, "Failed to map dMPT context memory, aborting.\n");
1155 		goto err_unmap_mtt;
1156 	}
1157 
1158 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1159 				  init_hca->qpc_base,
1160 				  dev_cap->qpc_entry_sz,
1161 				  dev->caps.num_qps,
1162 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1163 				  0, 0);
1164 	if (err) {
1165 		mlx4_err(dev, "Failed to map QP context memory, aborting.\n");
1166 		goto err_unmap_dmpt;
1167 	}
1168 
1169 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1170 				  init_hca->auxc_base,
1171 				  dev_cap->aux_entry_sz,
1172 				  dev->caps.num_qps,
1173 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1174 				  0, 0);
1175 	if (err) {
1176 		mlx4_err(dev, "Failed to map AUXC context memory, aborting.\n");
1177 		goto err_unmap_qp;
1178 	}
1179 
1180 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1181 				  init_hca->altc_base,
1182 				  dev_cap->altc_entry_sz,
1183 				  dev->caps.num_qps,
1184 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1185 				  0, 0);
1186 	if (err) {
1187 		mlx4_err(dev, "Failed to map ALTC context memory, aborting.\n");
1188 		goto err_unmap_auxc;
1189 	}
1190 
1191 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1192 				  init_hca->rdmarc_base,
1193 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1194 				  dev->caps.num_qps,
1195 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1196 				  0, 0);
1197 	if (err) {
1198 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1199 		goto err_unmap_altc;
1200 	}
1201 
1202 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1203 				  init_hca->cqc_base,
1204 				  dev_cap->cqc_entry_sz,
1205 				  dev->caps.num_cqs,
1206 				  dev->caps.reserved_cqs, 0, 0);
1207 	if (err) {
1208 		mlx4_err(dev, "Failed to map CQ context memory, aborting.\n");
1209 		goto err_unmap_rdmarc;
1210 	}
1211 
1212 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1213 				  init_hca->srqc_base,
1214 				  dev_cap->srq_entry_sz,
1215 				  dev->caps.num_srqs,
1216 				  dev->caps.reserved_srqs, 0, 0);
1217 	if (err) {
1218 		mlx4_err(dev, "Failed to map SRQ context memory, aborting.\n");
1219 		goto err_unmap_cq;
1220 	}
1221 
1222 	/*
1223 	 * For flow steering device managed mode it is required to use
1224 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1225 	 * required, but for simplicity just map the whole multicast
1226 	 * group table now.  The table isn't very big and it's a lot
1227 	 * easier than trying to track ref counts.
1228 	 */
1229 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1230 				  init_hca->mc_base,
1231 				  mlx4_get_mgm_entry_size(dev),
1232 				  dev->caps.num_mgms + dev->caps.num_amgms,
1233 				  dev->caps.num_mgms + dev->caps.num_amgms,
1234 				  0, 0);
1235 	if (err) {
1236 		mlx4_err(dev, "Failed to map MCG context memory, aborting.\n");
1237 		goto err_unmap_srq;
1238 	}
1239 
1240 	return 0;
1241 
1242 err_unmap_srq:
1243 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1244 
1245 err_unmap_cq:
1246 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1247 
1248 err_unmap_rdmarc:
1249 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1250 
1251 err_unmap_altc:
1252 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1253 
1254 err_unmap_auxc:
1255 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1256 
1257 err_unmap_qp:
1258 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1259 
1260 err_unmap_dmpt:
1261 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1262 
1263 err_unmap_mtt:
1264 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1265 
1266 err_unmap_eq:
1267 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1268 
1269 err_unmap_cmpt:
1270 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1271 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1272 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1273 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1274 
1275 err_unmap_aux:
1276 	mlx4_UNMAP_ICM_AUX(dev);
1277 
1278 err_free_aux:
1279 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1280 
1281 	return err;
1282 }
1283 
1284 static void mlx4_free_icms(struct mlx4_dev *dev)
1285 {
1286 	struct mlx4_priv *priv = mlx4_priv(dev);
1287 
1288 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1289 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1290 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1291 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1292 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1293 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1294 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1295 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1296 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1297 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1298 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1299 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1300 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1301 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1302 
1303 	mlx4_UNMAP_ICM_AUX(dev);
1304 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1305 }
1306 
1307 static void mlx4_slave_exit(struct mlx4_dev *dev)
1308 {
1309 	struct mlx4_priv *priv = mlx4_priv(dev);
1310 
1311 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1312 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_TIME))
1313 		mlx4_warn(dev, "Failed to close slave function.\n");
1314 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1315 }
1316 
1317 static int map_bf_area(struct mlx4_dev *dev)
1318 {
1319 	struct mlx4_priv *priv = mlx4_priv(dev);
1320 	resource_size_t bf_start;
1321 	resource_size_t bf_len;
1322 	int err = 0;
1323 
1324 	if (!dev->caps.bf_reg_size)
1325 		return -ENXIO;
1326 
1327 	bf_start = pci_resource_start(dev->pdev, 2) +
1328 			(dev->caps.num_uars << PAGE_SHIFT);
1329 	bf_len = pci_resource_len(dev->pdev, 2) -
1330 			(dev->caps.num_uars << PAGE_SHIFT);
1331 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1332 	if (!priv->bf_mapping)
1333 		err = -ENOMEM;
1334 
1335 	return err;
1336 }
1337 
1338 static void unmap_bf_area(struct mlx4_dev *dev)
1339 {
1340 	if (mlx4_priv(dev)->bf_mapping)
1341 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1342 }
1343 
1344 cycle_t mlx4_read_clock(struct mlx4_dev *dev)
1345 {
1346 	u32 clockhi, clocklo, clockhi1;
1347 	cycle_t cycles;
1348 	int i;
1349 	struct mlx4_priv *priv = mlx4_priv(dev);
1350 
1351 	for (i = 0; i < 10; i++) {
1352 		clockhi = swab32(readl(priv->clock_mapping));
1353 		clocklo = swab32(readl(priv->clock_mapping + 4));
1354 		clockhi1 = swab32(readl(priv->clock_mapping));
1355 		if (clockhi == clockhi1)
1356 			break;
1357 	}
1358 
1359 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1360 
1361 	return cycles;
1362 }
1363 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1364 
1365 
1366 static int map_internal_clock(struct mlx4_dev *dev)
1367 {
1368 	struct mlx4_priv *priv = mlx4_priv(dev);
1369 
1370 	priv->clock_mapping =
1371 		ioremap(pci_resource_start(dev->pdev, priv->fw.clock_bar) +
1372 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1373 
1374 	if (!priv->clock_mapping)
1375 		return -ENOMEM;
1376 
1377 	return 0;
1378 }
1379 
1380 static void unmap_internal_clock(struct mlx4_dev *dev)
1381 {
1382 	struct mlx4_priv *priv = mlx4_priv(dev);
1383 
1384 	if (priv->clock_mapping)
1385 		iounmap(priv->clock_mapping);
1386 }
1387 
1388 static void mlx4_close_hca(struct mlx4_dev *dev)
1389 {
1390 	unmap_internal_clock(dev);
1391 	unmap_bf_area(dev);
1392 	if (mlx4_is_slave(dev))
1393 		mlx4_slave_exit(dev);
1394 	else {
1395 		mlx4_CLOSE_HCA(dev, 0);
1396 		mlx4_free_icms(dev);
1397 		mlx4_UNMAP_FA(dev);
1398 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1399 	}
1400 }
1401 
1402 static int mlx4_init_slave(struct mlx4_dev *dev)
1403 {
1404 	struct mlx4_priv *priv = mlx4_priv(dev);
1405 	u64 dma = (u64) priv->mfunc.vhcr_dma;
1406 	int ret_from_reset = 0;
1407 	u32 slave_read;
1408 	u32 cmd_channel_ver;
1409 
1410 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1411 	priv->cmd.max_cmds = 1;
1412 	mlx4_warn(dev, "Sending reset\n");
1413 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1414 				       MLX4_COMM_TIME);
1415 	/* if we are in the middle of flr the slave will try
1416 	 * NUM_OF_RESET_RETRIES times before leaving.*/
1417 	if (ret_from_reset) {
1418 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1419 			mlx4_warn(dev, "slave is currently in the "
1420 				  "middle of FLR. Deferring probe.\n");
1421 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
1422 			return -EPROBE_DEFER;
1423 		} else
1424 			goto err;
1425 	}
1426 
1427 	/* check the driver version - the slave I/F revision
1428 	 * must match the master's */
1429 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1430 	cmd_channel_ver = mlx4_comm_get_version();
1431 
1432 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1433 		MLX4_COMM_GET_IF_REV(slave_read)) {
1434 		mlx4_err(dev, "slave driver version is not supported"
1435 			 " by the master\n");
1436 		goto err;
1437 	}
1438 
1439 	mlx4_warn(dev, "Sending vhcr0\n");
1440 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1441 						    MLX4_COMM_TIME))
1442 		goto err;
1443 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1444 						    MLX4_COMM_TIME))
1445 		goto err;
1446 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1447 						    MLX4_COMM_TIME))
1448 		goto err;
1449 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma, MLX4_COMM_TIME))
1450 		goto err;
1451 
1452 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1453 	return 0;
1454 
1455 err:
1456 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, 0);
1457 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1458 	return -EIO;
1459 }
1460 
1461 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1462 {
1463 	int i;
1464 
1465 	for (i = 1; i <= dev->caps.num_ports; i++) {
1466 		dev->caps.gid_table_len[i] = 1;
1467 		dev->caps.pkey_table_len[i] =
1468 			dev->phys_caps.pkey_phys_table_len[i] - 1;
1469 	}
1470 }
1471 
1472 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
1473 {
1474 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
1475 
1476 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
1477 	      i++) {
1478 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
1479 			break;
1480 	}
1481 
1482 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
1483 }
1484 
1485 static void choose_steering_mode(struct mlx4_dev *dev,
1486 				 struct mlx4_dev_cap *dev_cap)
1487 {
1488 	if (mlx4_log_num_mgm_entry_size == -1 &&
1489 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
1490 	    (!mlx4_is_mfunc(dev) ||
1491 	     (dev_cap->fs_max_num_qp_per_entry >= (num_vfs + 1))) &&
1492 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
1493 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
1494 		dev->oper_log_mgm_entry_size =
1495 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
1496 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1497 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
1498 		dev->caps.fs_log_max_ucast_qp_range_size =
1499 			dev_cap->fs_log_max_ucast_qp_range_size;
1500 	} else {
1501 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
1502 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1503 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
1504 		else {
1505 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
1506 
1507 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
1508 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1509 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags "
1510 					  "set to use B0 steering. Falling back to A0 steering mode.\n");
1511 		}
1512 		dev->oper_log_mgm_entry_size =
1513 			mlx4_log_num_mgm_entry_size > 0 ?
1514 			mlx4_log_num_mgm_entry_size :
1515 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
1516 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
1517 	}
1518 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, "
1519 		 "modparam log_num_mgm_entry_size = %d\n",
1520 		 mlx4_steering_mode_str(dev->caps.steering_mode),
1521 		 dev->oper_log_mgm_entry_size,
1522 		 mlx4_log_num_mgm_entry_size);
1523 }
1524 
1525 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
1526 				       struct mlx4_dev_cap *dev_cap)
1527 {
1528 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
1529 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
1530 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
1531 	else
1532 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
1533 
1534 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
1535 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
1536 }
1537 
1538 static int mlx4_init_hca(struct mlx4_dev *dev)
1539 {
1540 	struct mlx4_priv	  *priv = mlx4_priv(dev);
1541 	struct mlx4_adapter	   adapter;
1542 	struct mlx4_dev_cap	   dev_cap;
1543 	struct mlx4_mod_stat_cfg   mlx4_cfg;
1544 	struct mlx4_profile	   profile;
1545 	struct mlx4_init_hca_param init_hca;
1546 	u64 icm_size;
1547 	int err;
1548 
1549 	if (!mlx4_is_slave(dev)) {
1550 		err = mlx4_QUERY_FW(dev);
1551 		if (err) {
1552 			if (err == -EACCES)
1553 				mlx4_info(dev, "non-primary physical function, skipping.\n");
1554 			else
1555 				mlx4_err(dev, "QUERY_FW command failed, aborting.\n");
1556 			return err;
1557 		}
1558 
1559 		err = mlx4_load_fw(dev);
1560 		if (err) {
1561 			mlx4_err(dev, "Failed to start FW, aborting.\n");
1562 			return err;
1563 		}
1564 
1565 		mlx4_cfg.log_pg_sz_m = 1;
1566 		mlx4_cfg.log_pg_sz = 0;
1567 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
1568 		if (err)
1569 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
1570 
1571 		err = mlx4_dev_cap(dev, &dev_cap);
1572 		if (err) {
1573 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
1574 			goto err_stop_fw;
1575 		}
1576 
1577 		choose_steering_mode(dev, &dev_cap);
1578 		choose_tunnel_offload_mode(dev, &dev_cap);
1579 
1580 		err = mlx4_get_phys_port_id(dev);
1581 		if (err)
1582 			mlx4_err(dev, "Fail to get physical port id\n");
1583 
1584 		if (mlx4_is_master(dev))
1585 			mlx4_parav_master_pf_caps(dev);
1586 
1587 		profile = default_profile;
1588 		if (dev->caps.steering_mode ==
1589 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
1590 			profile.num_mcg = MLX4_FS_NUM_MCG;
1591 
1592 		icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
1593 					     &init_hca);
1594 		if ((long long) icm_size < 0) {
1595 			err = icm_size;
1596 			goto err_stop_fw;
1597 		}
1598 
1599 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
1600 
1601 		init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
1602 		init_hca.uar_page_sz = PAGE_SHIFT - 12;
1603 		init_hca.mw_enabled = 0;
1604 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1605 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
1606 			init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
1607 
1608 		err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
1609 		if (err)
1610 			goto err_stop_fw;
1611 
1612 		err = mlx4_INIT_HCA(dev, &init_hca);
1613 		if (err) {
1614 			mlx4_err(dev, "INIT_HCA command failed, aborting.\n");
1615 			goto err_free_icm;
1616 		}
1617 		/*
1618 		 * If TS is supported by FW
1619 		 * read HCA frequency by QUERY_HCA command
1620 		 */
1621 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
1622 			memset(&init_hca, 0, sizeof(init_hca));
1623 			err = mlx4_QUERY_HCA(dev, &init_hca);
1624 			if (err) {
1625 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp.\n");
1626 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1627 			} else {
1628 				dev->caps.hca_core_clock =
1629 					init_hca.hca_core_clock;
1630 			}
1631 
1632 			/* In case we got HCA frequency 0 - disable timestamping
1633 			 * to avoid dividing by zero
1634 			 */
1635 			if (!dev->caps.hca_core_clock) {
1636 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1637 				mlx4_err(dev,
1638 					 "HCA frequency is 0. Timestamping is not supported.");
1639 			} else if (map_internal_clock(dev)) {
1640 				/*
1641 				 * Map internal clock,
1642 				 * in case of failure disable timestamping
1643 				 */
1644 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1645 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported.\n");
1646 			}
1647 		}
1648 	} else {
1649 		err = mlx4_init_slave(dev);
1650 		if (err) {
1651 			if (err != -EPROBE_DEFER)
1652 				mlx4_err(dev, "Failed to initialize slave\n");
1653 			return err;
1654 		}
1655 
1656 		err = mlx4_slave_cap(dev);
1657 		if (err) {
1658 			mlx4_err(dev, "Failed to obtain slave caps\n");
1659 			goto err_close;
1660 		}
1661 	}
1662 
1663 	if (map_bf_area(dev))
1664 		mlx4_dbg(dev, "Failed to map blue flame area\n");
1665 
1666 	/*Only the master set the ports, all the rest got it from it.*/
1667 	if (!mlx4_is_slave(dev))
1668 		mlx4_set_port_mask(dev);
1669 
1670 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
1671 	if (err) {
1672 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting.\n");
1673 		goto unmap_bf;
1674 	}
1675 
1676 	priv->eq_table.inta_pin = adapter.inta_pin;
1677 	memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
1678 
1679 	return 0;
1680 
1681 unmap_bf:
1682 	unmap_internal_clock(dev);
1683 	unmap_bf_area(dev);
1684 
1685 err_close:
1686 	if (mlx4_is_slave(dev))
1687 		mlx4_slave_exit(dev);
1688 	else
1689 		mlx4_CLOSE_HCA(dev, 0);
1690 
1691 err_free_icm:
1692 	if (!mlx4_is_slave(dev))
1693 		mlx4_free_icms(dev);
1694 
1695 err_stop_fw:
1696 	if (!mlx4_is_slave(dev)) {
1697 		mlx4_UNMAP_FA(dev);
1698 		mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1699 	}
1700 	return err;
1701 }
1702 
1703 static int mlx4_init_counters_table(struct mlx4_dev *dev)
1704 {
1705 	struct mlx4_priv *priv = mlx4_priv(dev);
1706 	int nent;
1707 
1708 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1709 		return -ENOENT;
1710 
1711 	nent = dev->caps.max_counters;
1712 	return mlx4_bitmap_init(&priv->counters_bitmap, nent, nent - 1, 0, 0);
1713 }
1714 
1715 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
1716 {
1717 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
1718 }
1719 
1720 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1721 {
1722 	struct mlx4_priv *priv = mlx4_priv(dev);
1723 
1724 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1725 		return -ENOENT;
1726 
1727 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
1728 	if (*idx == -1)
1729 		return -ENOMEM;
1730 
1731 	return 0;
1732 }
1733 
1734 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1735 {
1736 	u64 out_param;
1737 	int err;
1738 
1739 	if (mlx4_is_mfunc(dev)) {
1740 		err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
1741 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
1742 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
1743 		if (!err)
1744 			*idx = get_param_l(&out_param);
1745 
1746 		return err;
1747 	}
1748 	return __mlx4_counter_alloc(dev, idx);
1749 }
1750 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
1751 
1752 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1753 {
1754 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
1755 	return;
1756 }
1757 
1758 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1759 {
1760 	u64 in_param = 0;
1761 
1762 	if (mlx4_is_mfunc(dev)) {
1763 		set_param_l(&in_param, idx);
1764 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
1765 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
1766 			 MLX4_CMD_WRAPPED);
1767 		return;
1768 	}
1769 	__mlx4_counter_free(dev, idx);
1770 }
1771 EXPORT_SYMBOL_GPL(mlx4_counter_free);
1772 
1773 static int mlx4_setup_hca(struct mlx4_dev *dev)
1774 {
1775 	struct mlx4_priv *priv = mlx4_priv(dev);
1776 	int err;
1777 	int port;
1778 	__be32 ib_port_default_caps;
1779 
1780 	err = mlx4_init_uar_table(dev);
1781 	if (err) {
1782 		mlx4_err(dev, "Failed to initialize "
1783 			 "user access region table, aborting.\n");
1784 		return err;
1785 	}
1786 
1787 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
1788 	if (err) {
1789 		mlx4_err(dev, "Failed to allocate driver access region, "
1790 			 "aborting.\n");
1791 		goto err_uar_table_free;
1792 	}
1793 
1794 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
1795 	if (!priv->kar) {
1796 		mlx4_err(dev, "Couldn't map kernel access region, "
1797 			 "aborting.\n");
1798 		err = -ENOMEM;
1799 		goto err_uar_free;
1800 	}
1801 
1802 	err = mlx4_init_pd_table(dev);
1803 	if (err) {
1804 		mlx4_err(dev, "Failed to initialize "
1805 			 "protection domain table, aborting.\n");
1806 		goto err_kar_unmap;
1807 	}
1808 
1809 	err = mlx4_init_xrcd_table(dev);
1810 	if (err) {
1811 		mlx4_err(dev, "Failed to initialize "
1812 			 "reliable connection domain table, aborting.\n");
1813 		goto err_pd_table_free;
1814 	}
1815 
1816 	err = mlx4_init_mr_table(dev);
1817 	if (err) {
1818 		mlx4_err(dev, "Failed to initialize "
1819 			 "memory region table, aborting.\n");
1820 		goto err_xrcd_table_free;
1821 	}
1822 
1823 	if (!mlx4_is_slave(dev)) {
1824 		err = mlx4_init_mcg_table(dev);
1825 		if (err) {
1826 			mlx4_err(dev, "Failed to initialize multicast group table, aborting.\n");
1827 			goto err_mr_table_free;
1828 		}
1829 	}
1830 
1831 	err = mlx4_init_eq_table(dev);
1832 	if (err) {
1833 		mlx4_err(dev, "Failed to initialize "
1834 			 "event queue table, aborting.\n");
1835 		goto err_mcg_table_free;
1836 	}
1837 
1838 	err = mlx4_cmd_use_events(dev);
1839 	if (err) {
1840 		mlx4_err(dev, "Failed to switch to event-driven "
1841 			 "firmware commands, aborting.\n");
1842 		goto err_eq_table_free;
1843 	}
1844 
1845 	err = mlx4_NOP(dev);
1846 	if (err) {
1847 		if (dev->flags & MLX4_FLAG_MSI_X) {
1848 			mlx4_warn(dev, "NOP command failed to generate MSI-X "
1849 				  "interrupt IRQ %d).\n",
1850 				  priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1851 			mlx4_warn(dev, "Trying again without MSI-X.\n");
1852 		} else {
1853 			mlx4_err(dev, "NOP command failed to generate interrupt "
1854 				 "(IRQ %d), aborting.\n",
1855 				 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1856 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
1857 		}
1858 
1859 		goto err_cmd_poll;
1860 	}
1861 
1862 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
1863 
1864 	err = mlx4_init_cq_table(dev);
1865 	if (err) {
1866 		mlx4_err(dev, "Failed to initialize "
1867 			 "completion queue table, aborting.\n");
1868 		goto err_cmd_poll;
1869 	}
1870 
1871 	err = mlx4_init_srq_table(dev);
1872 	if (err) {
1873 		mlx4_err(dev, "Failed to initialize "
1874 			 "shared receive queue table, aborting.\n");
1875 		goto err_cq_table_free;
1876 	}
1877 
1878 	err = mlx4_init_qp_table(dev);
1879 	if (err) {
1880 		mlx4_err(dev, "Failed to initialize "
1881 			 "queue pair table, aborting.\n");
1882 		goto err_srq_table_free;
1883 	}
1884 
1885 	err = mlx4_init_counters_table(dev);
1886 	if (err && err != -ENOENT) {
1887 		mlx4_err(dev, "Failed to initialize counters table, aborting.\n");
1888 		goto err_qp_table_free;
1889 	}
1890 
1891 	if (!mlx4_is_slave(dev)) {
1892 		for (port = 1; port <= dev->caps.num_ports; port++) {
1893 			ib_port_default_caps = 0;
1894 			err = mlx4_get_port_ib_caps(dev, port,
1895 						    &ib_port_default_caps);
1896 			if (err)
1897 				mlx4_warn(dev, "failed to get port %d default "
1898 					  "ib capabilities (%d). Continuing "
1899 					  "with caps = 0\n", port, err);
1900 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
1901 
1902 			/* initialize per-slave default ib port capabilities */
1903 			if (mlx4_is_master(dev)) {
1904 				int i;
1905 				for (i = 0; i < dev->num_slaves; i++) {
1906 					if (i == mlx4_master_func_num(dev))
1907 						continue;
1908 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
1909 							ib_port_default_caps;
1910 				}
1911 			}
1912 
1913 			if (mlx4_is_mfunc(dev))
1914 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
1915 			else
1916 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
1917 
1918 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
1919 					    dev->caps.pkey_table_len[port] : -1);
1920 			if (err) {
1921 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1922 					port);
1923 				goto err_counters_table_free;
1924 			}
1925 		}
1926 	}
1927 
1928 	return 0;
1929 
1930 err_counters_table_free:
1931 	mlx4_cleanup_counters_table(dev);
1932 
1933 err_qp_table_free:
1934 	mlx4_cleanup_qp_table(dev);
1935 
1936 err_srq_table_free:
1937 	mlx4_cleanup_srq_table(dev);
1938 
1939 err_cq_table_free:
1940 	mlx4_cleanup_cq_table(dev);
1941 
1942 err_cmd_poll:
1943 	mlx4_cmd_use_polling(dev);
1944 
1945 err_eq_table_free:
1946 	mlx4_cleanup_eq_table(dev);
1947 
1948 err_mcg_table_free:
1949 	if (!mlx4_is_slave(dev))
1950 		mlx4_cleanup_mcg_table(dev);
1951 
1952 err_mr_table_free:
1953 	mlx4_cleanup_mr_table(dev);
1954 
1955 err_xrcd_table_free:
1956 	mlx4_cleanup_xrcd_table(dev);
1957 
1958 err_pd_table_free:
1959 	mlx4_cleanup_pd_table(dev);
1960 
1961 err_kar_unmap:
1962 	iounmap(priv->kar);
1963 
1964 err_uar_free:
1965 	mlx4_uar_free(dev, &priv->driver_uar);
1966 
1967 err_uar_table_free:
1968 	mlx4_cleanup_uar_table(dev);
1969 	return err;
1970 }
1971 
1972 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
1973 {
1974 	struct mlx4_priv *priv = mlx4_priv(dev);
1975 	struct msix_entry *entries;
1976 	int nreq = min_t(int, dev->caps.num_ports *
1977 			 min_t(int, netif_get_num_default_rss_queues() + 1,
1978 			       MAX_MSIX_P_PORT) + MSIX_LEGACY_SZ, MAX_MSIX);
1979 	int err;
1980 	int i;
1981 
1982 	if (msi_x) {
1983 		nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
1984 			     nreq);
1985 
1986 		entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
1987 		if (!entries)
1988 			goto no_msi;
1989 
1990 		for (i = 0; i < nreq; ++i)
1991 			entries[i].entry = i;
1992 
1993 	retry:
1994 		err = pci_enable_msix(dev->pdev, entries, nreq);
1995 		if (err) {
1996 			/* Try again if at least 2 vectors are available */
1997 			if (err > 1) {
1998 				mlx4_info(dev, "Requested %d vectors, "
1999 					  "but only %d MSI-X vectors available, "
2000 					  "trying again\n", nreq, err);
2001 				nreq = err;
2002 				goto retry;
2003 			}
2004 			kfree(entries);
2005 			goto no_msi;
2006 		}
2007 
2008 		if (nreq <
2009 		    MSIX_LEGACY_SZ + dev->caps.num_ports * MIN_MSIX_P_PORT) {
2010 			/*Working in legacy mode , all EQ's shared*/
2011 			dev->caps.comp_pool           = 0;
2012 			dev->caps.num_comp_vectors = nreq - 1;
2013 		} else {
2014 			dev->caps.comp_pool           = nreq - MSIX_LEGACY_SZ;
2015 			dev->caps.num_comp_vectors = MSIX_LEGACY_SZ - 1;
2016 		}
2017 		for (i = 0; i < nreq; ++i)
2018 			priv->eq_table.eq[i].irq = entries[i].vector;
2019 
2020 		dev->flags |= MLX4_FLAG_MSI_X;
2021 
2022 		kfree(entries);
2023 		return;
2024 	}
2025 
2026 no_msi:
2027 	dev->caps.num_comp_vectors = 1;
2028 	dev->caps.comp_pool	   = 0;
2029 
2030 	for (i = 0; i < 2; ++i)
2031 		priv->eq_table.eq[i].irq = dev->pdev->irq;
2032 }
2033 
2034 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2035 {
2036 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2037 	int err = 0;
2038 
2039 	info->dev = dev;
2040 	info->port = port;
2041 	if (!mlx4_is_slave(dev)) {
2042 		mlx4_init_mac_table(dev, &info->mac_table);
2043 		mlx4_init_vlan_table(dev, &info->vlan_table);
2044 		info->base_qpn = mlx4_get_base_qpn(dev, port);
2045 	}
2046 
2047 	sprintf(info->dev_name, "mlx4_port%d", port);
2048 	info->port_attr.attr.name = info->dev_name;
2049 	if (mlx4_is_mfunc(dev))
2050 		info->port_attr.attr.mode = S_IRUGO;
2051 	else {
2052 		info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2053 		info->port_attr.store     = set_port_type;
2054 	}
2055 	info->port_attr.show      = show_port_type;
2056 	sysfs_attr_init(&info->port_attr.attr);
2057 
2058 	err = device_create_file(&dev->pdev->dev, &info->port_attr);
2059 	if (err) {
2060 		mlx4_err(dev, "Failed to create file for port %d\n", port);
2061 		info->port = -1;
2062 	}
2063 
2064 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2065 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
2066 	if (mlx4_is_mfunc(dev))
2067 		info->port_mtu_attr.attr.mode = S_IRUGO;
2068 	else {
2069 		info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2070 		info->port_mtu_attr.store     = set_port_ib_mtu;
2071 	}
2072 	info->port_mtu_attr.show      = show_port_ib_mtu;
2073 	sysfs_attr_init(&info->port_mtu_attr.attr);
2074 
2075 	err = device_create_file(&dev->pdev->dev, &info->port_mtu_attr);
2076 	if (err) {
2077 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
2078 		device_remove_file(&info->dev->pdev->dev, &info->port_attr);
2079 		info->port = -1;
2080 	}
2081 
2082 	return err;
2083 }
2084 
2085 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
2086 {
2087 	if (info->port < 0)
2088 		return;
2089 
2090 	device_remove_file(&info->dev->pdev->dev, &info->port_attr);
2091 	device_remove_file(&info->dev->pdev->dev, &info->port_mtu_attr);
2092 }
2093 
2094 static int mlx4_init_steering(struct mlx4_dev *dev)
2095 {
2096 	struct mlx4_priv *priv = mlx4_priv(dev);
2097 	int num_entries = dev->caps.num_ports;
2098 	int i, j;
2099 
2100 	priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
2101 	if (!priv->steer)
2102 		return -ENOMEM;
2103 
2104 	for (i = 0; i < num_entries; i++)
2105 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
2106 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
2107 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
2108 		}
2109 	return 0;
2110 }
2111 
2112 static void mlx4_clear_steering(struct mlx4_dev *dev)
2113 {
2114 	struct mlx4_priv *priv = mlx4_priv(dev);
2115 	struct mlx4_steer_index *entry, *tmp_entry;
2116 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
2117 	int num_entries = dev->caps.num_ports;
2118 	int i, j;
2119 
2120 	for (i = 0; i < num_entries; i++) {
2121 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
2122 			list_for_each_entry_safe(pqp, tmp_pqp,
2123 						 &priv->steer[i].promisc_qps[j],
2124 						 list) {
2125 				list_del(&pqp->list);
2126 				kfree(pqp);
2127 			}
2128 			list_for_each_entry_safe(entry, tmp_entry,
2129 						 &priv->steer[i].steer_entries[j],
2130 						 list) {
2131 				list_del(&entry->list);
2132 				list_for_each_entry_safe(pqp, tmp_pqp,
2133 							 &entry->duplicates,
2134 							 list) {
2135 					list_del(&pqp->list);
2136 					kfree(pqp);
2137 				}
2138 				kfree(entry);
2139 			}
2140 		}
2141 	}
2142 	kfree(priv->steer);
2143 }
2144 
2145 static int extended_func_num(struct pci_dev *pdev)
2146 {
2147 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
2148 }
2149 
2150 #define MLX4_OWNER_BASE	0x8069c
2151 #define MLX4_OWNER_SIZE	4
2152 
2153 static int mlx4_get_ownership(struct mlx4_dev *dev)
2154 {
2155 	void __iomem *owner;
2156 	u32 ret;
2157 
2158 	if (pci_channel_offline(dev->pdev))
2159 		return -EIO;
2160 
2161 	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
2162 			MLX4_OWNER_SIZE);
2163 	if (!owner) {
2164 		mlx4_err(dev, "Failed to obtain ownership bit\n");
2165 		return -ENOMEM;
2166 	}
2167 
2168 	ret = readl(owner);
2169 	iounmap(owner);
2170 	return (int) !!ret;
2171 }
2172 
2173 static void mlx4_free_ownership(struct mlx4_dev *dev)
2174 {
2175 	void __iomem *owner;
2176 
2177 	if (pci_channel_offline(dev->pdev))
2178 		return;
2179 
2180 	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
2181 			MLX4_OWNER_SIZE);
2182 	if (!owner) {
2183 		mlx4_err(dev, "Failed to obtain ownership bit\n");
2184 		return;
2185 	}
2186 	writel(0, owner);
2187 	msleep(1000);
2188 	iounmap(owner);
2189 }
2190 
2191 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data)
2192 {
2193 	struct mlx4_priv *priv;
2194 	struct mlx4_dev *dev;
2195 	int err;
2196 	int port;
2197 
2198 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
2199 
2200 	err = pci_enable_device(pdev);
2201 	if (err) {
2202 		dev_err(&pdev->dev, "Cannot enable PCI device, "
2203 			"aborting.\n");
2204 		return err;
2205 	}
2206 
2207 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
2208 	 * per port, we must limit the number of VFs to 63 (since their are
2209 	 * 128 MACs)
2210 	 */
2211 	if (num_vfs >= MLX4_MAX_NUM_VF) {
2212 		dev_err(&pdev->dev,
2213 			"Requested more VF's (%d) than allowed (%d)\n",
2214 			num_vfs, MLX4_MAX_NUM_VF - 1);
2215 		return -EINVAL;
2216 	}
2217 
2218 	if (num_vfs < 0) {
2219 		pr_err("num_vfs module parameter cannot be negative\n");
2220 		return -EINVAL;
2221 	}
2222 	/*
2223 	 * Check for BARs.
2224 	 */
2225 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
2226 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
2227 		dev_err(&pdev->dev, "Missing DCS, aborting."
2228 			"(driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
2229 			pci_dev_data, pci_resource_flags(pdev, 0));
2230 		err = -ENODEV;
2231 		goto err_disable_pdev;
2232 	}
2233 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
2234 		dev_err(&pdev->dev, "Missing UAR, aborting.\n");
2235 		err = -ENODEV;
2236 		goto err_disable_pdev;
2237 	}
2238 
2239 	err = pci_request_regions(pdev, DRV_NAME);
2240 	if (err) {
2241 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
2242 		goto err_disable_pdev;
2243 	}
2244 
2245 	pci_set_master(pdev);
2246 
2247 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2248 	if (err) {
2249 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
2250 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2251 		if (err) {
2252 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
2253 			goto err_release_regions;
2254 		}
2255 	}
2256 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2257 	if (err) {
2258 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
2259 			 "consistent PCI DMA mask.\n");
2260 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2261 		if (err) {
2262 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
2263 				"aborting.\n");
2264 			goto err_release_regions;
2265 		}
2266 	}
2267 
2268 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
2269 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
2270 
2271 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
2272 	if (!priv) {
2273 		err = -ENOMEM;
2274 		goto err_release_regions;
2275 	}
2276 
2277 	dev       = &priv->dev;
2278 	dev->pdev = pdev;
2279 	INIT_LIST_HEAD(&priv->ctx_list);
2280 	spin_lock_init(&priv->ctx_lock);
2281 
2282 	mutex_init(&priv->port_mutex);
2283 
2284 	INIT_LIST_HEAD(&priv->pgdir_list);
2285 	mutex_init(&priv->pgdir_mutex);
2286 
2287 	INIT_LIST_HEAD(&priv->bf_list);
2288 	mutex_init(&priv->bf_mutex);
2289 
2290 	dev->rev_id = pdev->revision;
2291 	dev->numa_node = dev_to_node(&pdev->dev);
2292 	/* Detect if this device is a virtual function */
2293 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
2294 		/* When acting as pf, we normally skip vfs unless explicitly
2295 		 * requested to probe them. */
2296 		if (num_vfs && extended_func_num(pdev) > probe_vf) {
2297 			mlx4_warn(dev, "Skipping virtual function:%d\n",
2298 						extended_func_num(pdev));
2299 			err = -ENODEV;
2300 			goto err_free_dev;
2301 		}
2302 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
2303 		dev->flags |= MLX4_FLAG_SLAVE;
2304 	} else {
2305 		/* We reset the device and enable SRIOV only for physical
2306 		 * devices.  Try to claim ownership on the device;
2307 		 * if already taken, skip -- do not allow multiple PFs */
2308 		err = mlx4_get_ownership(dev);
2309 		if (err) {
2310 			if (err < 0)
2311 				goto err_free_dev;
2312 			else {
2313 				mlx4_warn(dev, "Multiple PFs not yet supported."
2314 					  " Skipping PF.\n");
2315 				err = -EINVAL;
2316 				goto err_free_dev;
2317 			}
2318 		}
2319 
2320 		if (num_vfs) {
2321 			mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", num_vfs);
2322 			err = pci_enable_sriov(pdev, num_vfs);
2323 			if (err) {
2324 				mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d).\n",
2325 					 err);
2326 				err = 0;
2327 			} else {
2328 				mlx4_warn(dev, "Running in master mode\n");
2329 				dev->flags |= MLX4_FLAG_SRIOV |
2330 					      MLX4_FLAG_MASTER;
2331 				dev->num_vfs = num_vfs;
2332 			}
2333 		}
2334 
2335 		atomic_set(&priv->opreq_count, 0);
2336 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
2337 
2338 		/*
2339 		 * Now reset the HCA before we touch the PCI capabilities or
2340 		 * attempt a firmware command, since a boot ROM may have left
2341 		 * the HCA in an undefined state.
2342 		 */
2343 		err = mlx4_reset(dev);
2344 		if (err) {
2345 			mlx4_err(dev, "Failed to reset HCA, aborting.\n");
2346 			goto err_rel_own;
2347 		}
2348 	}
2349 
2350 slave_start:
2351 	err = mlx4_cmd_init(dev);
2352 	if (err) {
2353 		mlx4_err(dev, "Failed to init command interface, aborting.\n");
2354 		goto err_sriov;
2355 	}
2356 
2357 	/* In slave functions, the communication channel must be initialized
2358 	 * before posting commands. Also, init num_slaves before calling
2359 	 * mlx4_init_hca */
2360 	if (mlx4_is_mfunc(dev)) {
2361 		if (mlx4_is_master(dev))
2362 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
2363 		else {
2364 			dev->num_slaves = 0;
2365 			err = mlx4_multi_func_init(dev);
2366 			if (err) {
2367 				mlx4_err(dev, "Failed to init slave mfunc"
2368 					 " interface, aborting.\n");
2369 				goto err_cmd;
2370 			}
2371 		}
2372 	}
2373 
2374 	err = mlx4_init_hca(dev);
2375 	if (err) {
2376 		if (err == -EACCES) {
2377 			/* Not primary Physical function
2378 			 * Running in slave mode */
2379 			mlx4_cmd_cleanup(dev);
2380 			dev->flags |= MLX4_FLAG_SLAVE;
2381 			dev->flags &= ~MLX4_FLAG_MASTER;
2382 			goto slave_start;
2383 		} else
2384 			goto err_mfunc;
2385 	}
2386 
2387 	/* check if the device is functioning at its maximum possible speed.
2388 	 * No return code for this call, just warn the user in case of PCI
2389 	 * express device capabilities are under-satisfied by the bus.
2390 	 */
2391 	mlx4_check_pcie_caps(dev);
2392 
2393 	/* In master functions, the communication channel must be initialized
2394 	 * after obtaining its address from fw */
2395 	if (mlx4_is_master(dev)) {
2396 		err = mlx4_multi_func_init(dev);
2397 		if (err) {
2398 			mlx4_err(dev, "Failed to init master mfunc"
2399 				 "interface, aborting.\n");
2400 			goto err_close;
2401 		}
2402 	}
2403 
2404 	err = mlx4_alloc_eq_table(dev);
2405 	if (err)
2406 		goto err_master_mfunc;
2407 
2408 	priv->msix_ctl.pool_bm = 0;
2409 	mutex_init(&priv->msix_ctl.pool_lock);
2410 
2411 	mlx4_enable_msi_x(dev);
2412 	if ((mlx4_is_mfunc(dev)) &&
2413 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
2414 		err = -ENOSYS;
2415 		mlx4_err(dev, "INTx is not supported in multi-function mode."
2416 			 " aborting.\n");
2417 		goto err_free_eq;
2418 	}
2419 
2420 	if (!mlx4_is_slave(dev)) {
2421 		err = mlx4_init_steering(dev);
2422 		if (err)
2423 			goto err_free_eq;
2424 	}
2425 
2426 	err = mlx4_setup_hca(dev);
2427 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
2428 	    !mlx4_is_mfunc(dev)) {
2429 		dev->flags &= ~MLX4_FLAG_MSI_X;
2430 		dev->caps.num_comp_vectors = 1;
2431 		dev->caps.comp_pool	   = 0;
2432 		pci_disable_msix(pdev);
2433 		err = mlx4_setup_hca(dev);
2434 	}
2435 
2436 	if (err)
2437 		goto err_steer;
2438 
2439 	mlx4_init_quotas(dev);
2440 
2441 	for (port = 1; port <= dev->caps.num_ports; port++) {
2442 		err = mlx4_init_port_info(dev, port);
2443 		if (err)
2444 			goto err_port;
2445 	}
2446 
2447 	err = mlx4_register_device(dev);
2448 	if (err)
2449 		goto err_port;
2450 
2451 	mlx4_request_modules(dev);
2452 
2453 	mlx4_sense_init(dev);
2454 	mlx4_start_sense(dev);
2455 
2456 	priv->pci_dev_data = pci_dev_data;
2457 	pci_set_drvdata(pdev, dev);
2458 
2459 	return 0;
2460 
2461 err_port:
2462 	for (--port; port >= 1; --port)
2463 		mlx4_cleanup_port_info(&priv->port[port]);
2464 
2465 	mlx4_cleanup_counters_table(dev);
2466 	mlx4_cleanup_qp_table(dev);
2467 	mlx4_cleanup_srq_table(dev);
2468 	mlx4_cleanup_cq_table(dev);
2469 	mlx4_cmd_use_polling(dev);
2470 	mlx4_cleanup_eq_table(dev);
2471 	mlx4_cleanup_mcg_table(dev);
2472 	mlx4_cleanup_mr_table(dev);
2473 	mlx4_cleanup_xrcd_table(dev);
2474 	mlx4_cleanup_pd_table(dev);
2475 	mlx4_cleanup_uar_table(dev);
2476 
2477 err_steer:
2478 	if (!mlx4_is_slave(dev))
2479 		mlx4_clear_steering(dev);
2480 
2481 err_free_eq:
2482 	mlx4_free_eq_table(dev);
2483 
2484 err_master_mfunc:
2485 	if (mlx4_is_master(dev))
2486 		mlx4_multi_func_cleanup(dev);
2487 
2488 err_close:
2489 	if (dev->flags & MLX4_FLAG_MSI_X)
2490 		pci_disable_msix(pdev);
2491 
2492 	mlx4_close_hca(dev);
2493 
2494 err_mfunc:
2495 	if (mlx4_is_slave(dev))
2496 		mlx4_multi_func_cleanup(dev);
2497 
2498 err_cmd:
2499 	mlx4_cmd_cleanup(dev);
2500 
2501 err_sriov:
2502 	if (dev->flags & MLX4_FLAG_SRIOV)
2503 		pci_disable_sriov(pdev);
2504 
2505 err_rel_own:
2506 	if (!mlx4_is_slave(dev))
2507 		mlx4_free_ownership(dev);
2508 
2509 err_free_dev:
2510 	kfree(priv);
2511 
2512 err_release_regions:
2513 	pci_release_regions(pdev);
2514 
2515 err_disable_pdev:
2516 	pci_disable_device(pdev);
2517 	pci_set_drvdata(pdev, NULL);
2518 	return err;
2519 }
2520 
2521 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
2522 {
2523 	printk_once(KERN_INFO "%s", mlx4_version);
2524 
2525 	return __mlx4_init_one(pdev, id->driver_data);
2526 }
2527 
2528 static void mlx4_remove_one(struct pci_dev *pdev)
2529 {
2530 	struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
2531 	struct mlx4_priv *priv = mlx4_priv(dev);
2532 	int p;
2533 
2534 	if (dev) {
2535 		/* in SRIOV it is not allowed to unload the pf's
2536 		 * driver while there are alive vf's */
2537 		if (mlx4_is_master(dev)) {
2538 			if (mlx4_how_many_lives_vf(dev))
2539 				printk(KERN_ERR "Removing PF when there are assigned VF's !!!\n");
2540 		}
2541 		mlx4_stop_sense(dev);
2542 		mlx4_unregister_device(dev);
2543 
2544 		for (p = 1; p <= dev->caps.num_ports; p++) {
2545 			mlx4_cleanup_port_info(&priv->port[p]);
2546 			mlx4_CLOSE_PORT(dev, p);
2547 		}
2548 
2549 		if (mlx4_is_master(dev))
2550 			mlx4_free_resource_tracker(dev,
2551 						   RES_TR_FREE_SLAVES_ONLY);
2552 
2553 		mlx4_cleanup_counters_table(dev);
2554 		mlx4_cleanup_qp_table(dev);
2555 		mlx4_cleanup_srq_table(dev);
2556 		mlx4_cleanup_cq_table(dev);
2557 		mlx4_cmd_use_polling(dev);
2558 		mlx4_cleanup_eq_table(dev);
2559 		mlx4_cleanup_mcg_table(dev);
2560 		mlx4_cleanup_mr_table(dev);
2561 		mlx4_cleanup_xrcd_table(dev);
2562 		mlx4_cleanup_pd_table(dev);
2563 
2564 		if (mlx4_is_master(dev))
2565 			mlx4_free_resource_tracker(dev,
2566 						   RES_TR_FREE_STRUCTS_ONLY);
2567 
2568 		iounmap(priv->kar);
2569 		mlx4_uar_free(dev, &priv->driver_uar);
2570 		mlx4_cleanup_uar_table(dev);
2571 		if (!mlx4_is_slave(dev))
2572 			mlx4_clear_steering(dev);
2573 		mlx4_free_eq_table(dev);
2574 		if (mlx4_is_master(dev))
2575 			mlx4_multi_func_cleanup(dev);
2576 		mlx4_close_hca(dev);
2577 		if (mlx4_is_slave(dev))
2578 			mlx4_multi_func_cleanup(dev);
2579 		mlx4_cmd_cleanup(dev);
2580 
2581 		if (dev->flags & MLX4_FLAG_MSI_X)
2582 			pci_disable_msix(pdev);
2583 		if (dev->flags & MLX4_FLAG_SRIOV) {
2584 			mlx4_warn(dev, "Disabling SR-IOV\n");
2585 			pci_disable_sriov(pdev);
2586 		}
2587 
2588 		if (!mlx4_is_slave(dev))
2589 			mlx4_free_ownership(dev);
2590 
2591 		kfree(dev->caps.qp0_tunnel);
2592 		kfree(dev->caps.qp0_proxy);
2593 		kfree(dev->caps.qp1_tunnel);
2594 		kfree(dev->caps.qp1_proxy);
2595 
2596 		kfree(priv);
2597 		pci_release_regions(pdev);
2598 		pci_disable_device(pdev);
2599 		pci_set_drvdata(pdev, NULL);
2600 	}
2601 }
2602 
2603 int mlx4_restart_one(struct pci_dev *pdev)
2604 {
2605 	struct mlx4_dev	 *dev  = pci_get_drvdata(pdev);
2606 	struct mlx4_priv *priv = mlx4_priv(dev);
2607 	int		  pci_dev_data;
2608 
2609 	pci_dev_data = priv->pci_dev_data;
2610 	mlx4_remove_one(pdev);
2611 	return __mlx4_init_one(pdev, pci_dev_data);
2612 }
2613 
2614 static DEFINE_PCI_DEVICE_TABLE(mlx4_pci_table) = {
2615 	/* MT25408 "Hermon" SDR */
2616 	{ PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2617 	/* MT25408 "Hermon" DDR */
2618 	{ PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2619 	/* MT25408 "Hermon" QDR */
2620 	{ PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2621 	/* MT25408 "Hermon" DDR PCIe gen2 */
2622 	{ PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2623 	/* MT25408 "Hermon" QDR PCIe gen2 */
2624 	{ PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2625 	/* MT25408 "Hermon" EN 10GigE */
2626 	{ PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2627 	/* MT25408 "Hermon" EN 10GigE PCIe gen2 */
2628 	{ PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2629 	/* MT25458 ConnectX EN 10GBASE-T 10GigE */
2630 	{ PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2631 	/* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
2632 	{ PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2633 	/* MT26468 ConnectX EN 10GigE PCIe gen2*/
2634 	{ PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2635 	/* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
2636 	{ PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2637 	/* MT26478 ConnectX2 40GigE PCIe gen2 */
2638 	{ PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2639 	/* MT25400 Family [ConnectX-2 Virtual Function] */
2640 	{ PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
2641 	/* MT27500 Family [ConnectX-3] */
2642 	{ PCI_VDEVICE(MELLANOX, 0x1003), 0 },
2643 	/* MT27500 Family [ConnectX-3 Virtual Function] */
2644 	{ PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
2645 	{ PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
2646 	{ PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
2647 	{ PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
2648 	{ PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
2649 	{ PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
2650 	{ PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
2651 	{ PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
2652 	{ PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
2653 	{ PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
2654 	{ PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
2655 	{ PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
2656 	{ PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
2657 	{ 0, }
2658 };
2659 
2660 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
2661 
2662 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
2663 					      pci_channel_state_t state)
2664 {
2665 	mlx4_remove_one(pdev);
2666 
2667 	return state == pci_channel_io_perm_failure ?
2668 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
2669 }
2670 
2671 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
2672 {
2673 	int ret = __mlx4_init_one(pdev, 0);
2674 
2675 	return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
2676 }
2677 
2678 static const struct pci_error_handlers mlx4_err_handler = {
2679 	.error_detected = mlx4_pci_err_detected,
2680 	.slot_reset     = mlx4_pci_slot_reset,
2681 };
2682 
2683 static struct pci_driver mlx4_driver = {
2684 	.name		= DRV_NAME,
2685 	.id_table	= mlx4_pci_table,
2686 	.probe		= mlx4_init_one,
2687 	.remove		= mlx4_remove_one,
2688 	.err_handler    = &mlx4_err_handler,
2689 };
2690 
2691 static int __init mlx4_verify_params(void)
2692 {
2693 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
2694 		pr_warning("mlx4_core: bad num_mac: %d\n", log_num_mac);
2695 		return -1;
2696 	}
2697 
2698 	if (log_num_vlan != 0)
2699 		pr_warning("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
2700 			   MLX4_LOG_NUM_VLANS);
2701 
2702 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
2703 		pr_warning("mlx4_core: bad log_mtts_per_seg: %d\n", log_mtts_per_seg);
2704 		return -1;
2705 	}
2706 
2707 	/* Check if module param for ports type has legal combination */
2708 	if (port_type_array[0] == false && port_type_array[1] == true) {
2709 		printk(KERN_WARNING "Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
2710 		port_type_array[0] = true;
2711 	}
2712 
2713 	if (mlx4_log_num_mgm_entry_size != -1 &&
2714 	    (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
2715 	     mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE)) {
2716 		pr_warning("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not "
2717 			   "in legal range (-1 or %d..%d)\n",
2718 			   mlx4_log_num_mgm_entry_size,
2719 			   MLX4_MIN_MGM_LOG_ENTRY_SIZE,
2720 			   MLX4_MAX_MGM_LOG_ENTRY_SIZE);
2721 		return -1;
2722 	}
2723 
2724 	return 0;
2725 }
2726 
2727 static int __init mlx4_init(void)
2728 {
2729 	int ret;
2730 
2731 	if (mlx4_verify_params())
2732 		return -EINVAL;
2733 
2734 	mlx4_catas_init();
2735 
2736 	mlx4_wq = create_singlethread_workqueue("mlx4");
2737 	if (!mlx4_wq)
2738 		return -ENOMEM;
2739 
2740 	ret = pci_register_driver(&mlx4_driver);
2741 	if (ret < 0)
2742 		destroy_workqueue(mlx4_wq);
2743 	return ret < 0 ? ret : 0;
2744 }
2745 
2746 static void __exit mlx4_cleanup(void)
2747 {
2748 	pci_unregister_driver(&mlx4_driver);
2749 	destroy_workqueue(mlx4_wq);
2750 }
2751 
2752 module_init(mlx4_init);
2753 module_exit(mlx4_cleanup);
2754