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 = 3;
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 = 3;
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 
109 static char mlx4_version[] =
110 	DRV_NAME ": Mellanox ConnectX core driver v"
111 	DRV_VERSION " (" DRV_RELDATE ")\n";
112 
113 static struct mlx4_profile default_profile = {
114 	.num_qp		= 1 << 18,
115 	.num_srq	= 1 << 16,
116 	.rdmarc_per_qp	= 1 << 4,
117 	.num_cq		= 1 << 16,
118 	.num_mcg	= 1 << 13,
119 	.num_mpt	= 1 << 19,
120 	.num_mtt	= 1 << 20, /* It is really num mtt segements */
121 };
122 
123 static int log_num_mac = 7;
124 module_param_named(log_num_mac, log_num_mac, int, 0444);
125 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
126 
127 static int log_num_vlan;
128 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
129 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
130 /* Log2 max number of VLANs per ETH port (0-7) */
131 #define MLX4_LOG_NUM_VLANS 7
132 
133 static bool use_prio;
134 module_param_named(use_prio, use_prio, bool, 0444);
135 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
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 static atomic_t pf_loading = ATOMIC_INIT(0);
154 
155 int mlx4_check_port_params(struct mlx4_dev *dev,
156 			   enum mlx4_port_type *port_type)
157 {
158 	int i;
159 
160 	for (i = 0; i < dev->caps.num_ports - 1; i++) {
161 		if (port_type[i] != port_type[i + 1]) {
162 			if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
163 				mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
164 				return -EINVAL;
165 			}
166 		}
167 	}
168 
169 	for (i = 0; i < dev->caps.num_ports; i++) {
170 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
171 			mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
172 				 i + 1);
173 			return -EINVAL;
174 		}
175 	}
176 	return 0;
177 }
178 
179 static void mlx4_set_port_mask(struct mlx4_dev *dev)
180 {
181 	int i;
182 
183 	for (i = 1; i <= dev->caps.num_ports; ++i)
184 		dev->caps.port_mask[i] = dev->caps.port_type[i];
185 }
186 
187 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
188 {
189 	int err;
190 	int i;
191 
192 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
193 	if (err) {
194 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
195 		return err;
196 	}
197 
198 	if (dev_cap->min_page_sz > PAGE_SIZE) {
199 		mlx4_err(dev, "HCA minimum page size of %d bigger than 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, aborting\n",
205 			 dev_cap->num_ports, MLX4_MAX_PORTS);
206 		return -ENODEV;
207 	}
208 
209 	if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
210 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
211 			 dev_cap->uar_size,
212 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
213 		return -ENODEV;
214 	}
215 
216 	dev->caps.num_ports	     = dev_cap->num_ports;
217 	dev->phys_caps.num_phys_eqs  = MLX4_MAX_EQ_NUM;
218 	for (i = 1; i <= dev->caps.num_ports; ++i) {
219 		dev->caps.vl_cap[i]	    = dev_cap->max_vl[i];
220 		dev->caps.ib_mtu_cap[i]	    = dev_cap->ib_mtu[i];
221 		dev->phys_caps.gid_phys_table_len[i]  = dev_cap->max_gids[i];
222 		dev->phys_caps.pkey_phys_table_len[i] = dev_cap->max_pkeys[i];
223 		/* set gid and pkey table operating lengths by default
224 		 * to non-sriov values */
225 		dev->caps.gid_table_len[i]  = dev_cap->max_gids[i];
226 		dev->caps.pkey_table_len[i] = dev_cap->max_pkeys[i];
227 		dev->caps.port_width_cap[i] = dev_cap->max_port_width[i];
228 		dev->caps.eth_mtu_cap[i]    = dev_cap->eth_mtu[i];
229 		dev->caps.def_mac[i]        = dev_cap->def_mac[i];
230 		dev->caps.supported_type[i] = dev_cap->supported_port_types[i];
231 		dev->caps.suggested_type[i] = dev_cap->suggested_type[i];
232 		dev->caps.default_sense[i] = dev_cap->default_sense[i];
233 		dev->caps.trans_type[i]	    = dev_cap->trans_type[i];
234 		dev->caps.vendor_oui[i]     = dev_cap->vendor_oui[i];
235 		dev->caps.wavelength[i]     = dev_cap->wavelength[i];
236 		dev->caps.trans_code[i]     = dev_cap->trans_code[i];
237 	}
238 
239 	dev->caps.uar_page_size	     = PAGE_SIZE;
240 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
241 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
242 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
243 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
244 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
245 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
246 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
247 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
248 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
249 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
250 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
251 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
252 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
253 	/*
254 	 * Subtract 1 from the limit because we need to allocate a
255 	 * spare CQE so the HCA HW can tell the difference between an
256 	 * empty CQ and a full CQ.
257 	 */
258 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
259 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
260 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
261 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
262 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
263 
264 	/* The first 128 UARs are used for EQ doorbells */
265 	dev->caps.reserved_uars	     = max_t(int, 128, dev_cap->reserved_uars);
266 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
267 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
268 					dev_cap->reserved_xrcds : 0;
269 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
270 					dev_cap->max_xrcds : 0;
271 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
272 
273 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
274 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
275 	dev->caps.flags		     = dev_cap->flags;
276 	dev->caps.flags2	     = dev_cap->flags2;
277 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
278 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
279 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
280 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
281 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
282 
283 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
284 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
285 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
286 	/* Don't do sense port on multifunction devices (for now at least) */
287 	if (mlx4_is_mfunc(dev))
288 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
289 
290 	dev->caps.log_num_macs  = log_num_mac;
291 	dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
292 
293 	for (i = 1; i <= dev->caps.num_ports; ++i) {
294 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
295 		if (dev->caps.supported_type[i]) {
296 			/* if only ETH is supported - assign ETH */
297 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
298 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
299 			/* if only IB is supported, assign IB */
300 			else if (dev->caps.supported_type[i] ==
301 				 MLX4_PORT_TYPE_IB)
302 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
303 			else {
304 				/* if IB and ETH are supported, we set the port
305 				 * type according to user selection of port type;
306 				 * if user selected none, take the FW hint */
307 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
308 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
309 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
310 				else
311 					dev->caps.port_type[i] = port_type_array[i - 1];
312 			}
313 		}
314 		/*
315 		 * Link sensing is allowed on the port if 3 conditions are true:
316 		 * 1. Both protocols are supported on the port.
317 		 * 2. Different types are supported on the port
318 		 * 3. FW declared that it supports link sensing
319 		 */
320 		mlx4_priv(dev)->sense.sense_allowed[i] =
321 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
322 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
323 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
324 
325 		/*
326 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
327 		 * and perform sense_port FW command to try and set the correct
328 		 * port type from beginning
329 		 */
330 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
331 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
332 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
333 			mlx4_SENSE_PORT(dev, i, &sensed_port);
334 			if (sensed_port != MLX4_PORT_TYPE_NONE)
335 				dev->caps.port_type[i] = sensed_port;
336 		} else {
337 			dev->caps.possible_type[i] = dev->caps.port_type[i];
338 		}
339 
340 		if (dev->caps.log_num_macs > dev_cap->log_max_macs[i]) {
341 			dev->caps.log_num_macs = dev_cap->log_max_macs[i];
342 			mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
343 				  i, 1 << dev->caps.log_num_macs);
344 		}
345 		if (dev->caps.log_num_vlans > dev_cap->log_max_vlans[i]) {
346 			dev->caps.log_num_vlans = dev_cap->log_max_vlans[i];
347 			mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
348 				  i, 1 << dev->caps.log_num_vlans);
349 		}
350 	}
351 
352 	dev->caps.max_counters = 1 << ilog2(dev_cap->max_counters);
353 
354 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
355 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
356 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
357 		(1 << dev->caps.log_num_macs) *
358 		(1 << dev->caps.log_num_vlans) *
359 		dev->caps.num_ports;
360 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
361 
362 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
363 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
364 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
365 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
366 
367 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
368 
369 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
370 		if (dev_cap->flags &
371 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
372 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
373 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
374 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
375 		}
376 	}
377 
378 	if ((dev->caps.flags &
379 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
380 	    mlx4_is_master(dev))
381 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
382 
383 	return 0;
384 }
385 
386 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
387 				       enum pci_bus_speed *speed,
388 				       enum pcie_link_width *width)
389 {
390 	u32 lnkcap1, lnkcap2;
391 	int err1, err2;
392 
393 #define  PCIE_MLW_CAP_SHIFT 4	/* start of MLW mask in link capabilities */
394 
395 	*speed = PCI_SPEED_UNKNOWN;
396 	*width = PCIE_LNK_WIDTH_UNKNOWN;
397 
398 	err1 = pcie_capability_read_dword(dev->pdev, PCI_EXP_LNKCAP, &lnkcap1);
399 	err2 = pcie_capability_read_dword(dev->pdev, PCI_EXP_LNKCAP2, &lnkcap2);
400 	if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
401 		if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
402 			*speed = PCIE_SPEED_8_0GT;
403 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
404 			*speed = PCIE_SPEED_5_0GT;
405 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
406 			*speed = PCIE_SPEED_2_5GT;
407 	}
408 	if (!err1) {
409 		*width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
410 		if (!lnkcap2) { /* pre-r3.0 */
411 			if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
412 				*speed = PCIE_SPEED_5_0GT;
413 			else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
414 				*speed = PCIE_SPEED_2_5GT;
415 		}
416 	}
417 
418 	if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
419 		return err1 ? err1 :
420 			err2 ? err2 : -EINVAL;
421 	}
422 	return 0;
423 }
424 
425 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
426 {
427 	enum pcie_link_width width, width_cap;
428 	enum pci_bus_speed speed, speed_cap;
429 	int err;
430 
431 #define PCIE_SPEED_STR(speed) \
432 	(speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
433 	 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
434 	 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
435 	 "Unknown")
436 
437 	err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
438 	if (err) {
439 		mlx4_warn(dev,
440 			  "Unable to determine PCIe device BW capabilities\n");
441 		return;
442 	}
443 
444 	err = pcie_get_minimum_link(dev->pdev, &speed, &width);
445 	if (err || speed == PCI_SPEED_UNKNOWN ||
446 	    width == PCIE_LNK_WIDTH_UNKNOWN) {
447 		mlx4_warn(dev,
448 			  "Unable to determine PCI device chain minimum BW\n");
449 		return;
450 	}
451 
452 	if (width != width_cap || speed != speed_cap)
453 		mlx4_warn(dev,
454 			  "PCIe BW is different than device's capability\n");
455 
456 	mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
457 		  PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
458 	mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
459 		  width, width_cap);
460 	return;
461 }
462 
463 /*The function checks if there are live vf, return the num of them*/
464 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
465 {
466 	struct mlx4_priv *priv = mlx4_priv(dev);
467 	struct mlx4_slave_state *s_state;
468 	int i;
469 	int ret = 0;
470 
471 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
472 		s_state = &priv->mfunc.master.slave_state[i];
473 		if (s_state->active && s_state->last_cmd !=
474 		    MLX4_COMM_CMD_RESET) {
475 			mlx4_warn(dev, "%s: slave: %d is still active\n",
476 				  __func__, i);
477 			ret++;
478 		}
479 	}
480 	return ret;
481 }
482 
483 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
484 {
485 	u32 qk = MLX4_RESERVED_QKEY_BASE;
486 
487 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
488 	    qpn < dev->phys_caps.base_proxy_sqpn)
489 		return -EINVAL;
490 
491 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
492 		/* tunnel qp */
493 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
494 	else
495 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
496 	*qkey = qk;
497 	return 0;
498 }
499 EXPORT_SYMBOL(mlx4_get_parav_qkey);
500 
501 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
502 {
503 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
504 
505 	if (!mlx4_is_master(dev))
506 		return;
507 
508 	priv->virt2phys_pkey[slave][port - 1][i] = val;
509 }
510 EXPORT_SYMBOL(mlx4_sync_pkey_table);
511 
512 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
513 {
514 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
515 
516 	if (!mlx4_is_master(dev))
517 		return;
518 
519 	priv->slave_node_guids[slave] = guid;
520 }
521 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
522 
523 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
524 {
525 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
526 
527 	if (!mlx4_is_master(dev))
528 		return 0;
529 
530 	return priv->slave_node_guids[slave];
531 }
532 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
533 
534 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
535 {
536 	struct mlx4_priv *priv = mlx4_priv(dev);
537 	struct mlx4_slave_state *s_slave;
538 
539 	if (!mlx4_is_master(dev))
540 		return 0;
541 
542 	s_slave = &priv->mfunc.master.slave_state[slave];
543 	return !!s_slave->active;
544 }
545 EXPORT_SYMBOL(mlx4_is_slave_active);
546 
547 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
548 				       struct mlx4_dev_cap *dev_cap,
549 				       struct mlx4_init_hca_param *hca_param)
550 {
551 	dev->caps.steering_mode = hca_param->steering_mode;
552 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
553 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
554 		dev->caps.fs_log_max_ucast_qp_range_size =
555 			dev_cap->fs_log_max_ucast_qp_range_size;
556 	} else
557 		dev->caps.num_qp_per_mgm =
558 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
559 
560 	mlx4_dbg(dev, "Steering mode is: %s\n",
561 		 mlx4_steering_mode_str(dev->caps.steering_mode));
562 }
563 
564 static int mlx4_slave_cap(struct mlx4_dev *dev)
565 {
566 	int			   err;
567 	u32			   page_size;
568 	struct mlx4_dev_cap	   dev_cap;
569 	struct mlx4_func_cap	   func_cap;
570 	struct mlx4_init_hca_param hca_param;
571 	int			   i;
572 
573 	memset(&hca_param, 0, sizeof(hca_param));
574 	err = mlx4_QUERY_HCA(dev, &hca_param);
575 	if (err) {
576 		mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
577 		return err;
578 	}
579 
580 	/* fail if the hca has an unknown global capability
581 	 * at this time global_caps should be always zeroed
582 	 */
583 	if (hca_param.global_caps) {
584 		mlx4_err(dev, "Unknown hca global capabilities\n");
585 		return -ENOSYS;
586 	}
587 
588 	mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
589 
590 	dev->caps.hca_core_clock = hca_param.hca_core_clock;
591 
592 	memset(&dev_cap, 0, sizeof(dev_cap));
593 	dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
594 	err = mlx4_dev_cap(dev, &dev_cap);
595 	if (err) {
596 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
597 		return err;
598 	}
599 
600 	err = mlx4_QUERY_FW(dev);
601 	if (err)
602 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
603 
604 	page_size = ~dev->caps.page_size_cap + 1;
605 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
606 	if (page_size > PAGE_SIZE) {
607 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
608 			 page_size, PAGE_SIZE);
609 		return -ENODEV;
610 	}
611 
612 	/* slave gets uar page size from QUERY_HCA fw command */
613 	dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
614 
615 	/* TODO: relax this assumption */
616 	if (dev->caps.uar_page_size != PAGE_SIZE) {
617 		mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
618 			 dev->caps.uar_page_size, PAGE_SIZE);
619 		return -ENODEV;
620 	}
621 
622 	memset(&func_cap, 0, sizeof(func_cap));
623 	err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
624 	if (err) {
625 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
626 			 err);
627 		return err;
628 	}
629 
630 	if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
631 	    PF_CONTEXT_BEHAVIOUR_MASK) {
632 		mlx4_err(dev, "Unknown pf context behaviour\n");
633 		return -ENOSYS;
634 	}
635 
636 	dev->caps.num_ports		= func_cap.num_ports;
637 	dev->quotas.qp			= func_cap.qp_quota;
638 	dev->quotas.srq			= func_cap.srq_quota;
639 	dev->quotas.cq			= func_cap.cq_quota;
640 	dev->quotas.mpt			= func_cap.mpt_quota;
641 	dev->quotas.mtt			= func_cap.mtt_quota;
642 	dev->caps.num_qps		= 1 << hca_param.log_num_qps;
643 	dev->caps.num_srqs		= 1 << hca_param.log_num_srqs;
644 	dev->caps.num_cqs		= 1 << hca_param.log_num_cqs;
645 	dev->caps.num_mpts		= 1 << hca_param.log_mpt_sz;
646 	dev->caps.num_eqs		= func_cap.max_eq;
647 	dev->caps.reserved_eqs		= func_cap.reserved_eq;
648 	dev->caps.num_pds               = MLX4_NUM_PDS;
649 	dev->caps.num_mgms              = 0;
650 	dev->caps.num_amgms             = 0;
651 
652 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
653 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
654 			 dev->caps.num_ports, MLX4_MAX_PORTS);
655 		return -ENODEV;
656 	}
657 
658 	dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
659 	dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
660 	dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
661 	dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
662 	dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
663 
664 	if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
665 	    !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
666 	    !dev->caps.qp0_qkey) {
667 		err = -ENOMEM;
668 		goto err_mem;
669 	}
670 
671 	for (i = 1; i <= dev->caps.num_ports; ++i) {
672 		err = mlx4_QUERY_FUNC_CAP(dev, (u32) i, &func_cap);
673 		if (err) {
674 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
675 				 i, err);
676 			goto err_mem;
677 		}
678 		dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
679 		dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
680 		dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
681 		dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
682 		dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
683 		dev->caps.port_mask[i] = dev->caps.port_type[i];
684 		dev->caps.phys_port_id[i] = func_cap.phys_port_id;
685 		if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
686 						    &dev->caps.gid_table_len[i],
687 						    &dev->caps.pkey_table_len[i]))
688 			goto err_mem;
689 	}
690 
691 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
692 				       dev->caps.reserved_uars) >
693 				       pci_resource_len(dev->pdev, 2)) {
694 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
695 			 dev->caps.uar_page_size * dev->caps.num_uars,
696 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
697 		goto err_mem;
698 	}
699 
700 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
701 		dev->caps.eqe_size   = 64;
702 		dev->caps.eqe_factor = 1;
703 	} else {
704 		dev->caps.eqe_size   = 32;
705 		dev->caps.eqe_factor = 0;
706 	}
707 
708 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
709 		dev->caps.cqe_size   = 64;
710 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_64B_CQE;
711 	} else {
712 		dev->caps.cqe_size   = 32;
713 	}
714 
715 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
716 	mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
717 
718 	slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
719 
720 	return 0;
721 
722 err_mem:
723 	kfree(dev->caps.qp0_qkey);
724 	kfree(dev->caps.qp0_tunnel);
725 	kfree(dev->caps.qp0_proxy);
726 	kfree(dev->caps.qp1_tunnel);
727 	kfree(dev->caps.qp1_proxy);
728 	dev->caps.qp0_qkey = NULL;
729 	dev->caps.qp0_tunnel = NULL;
730 	dev->caps.qp0_proxy = NULL;
731 	dev->caps.qp1_tunnel = NULL;
732 	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_eth_port)
753 		request_module_nowait(EN_DRV_NAME);
754 	if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
755 		request_module_nowait(IB_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, aborting\n",
783 					 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. Set only 'eth' or 'ib' for both ports (should be the same)\n");
867 		goto out;
868 	}
869 
870 	mlx4_do_sense_ports(mdev, new_types, types);
871 
872 	err = mlx4_check_port_params(mdev, new_types);
873 	if (err)
874 		goto out;
875 
876 	/* We are about to apply the changes after the configuration
877 	 * was verified, no need to remember the temporary types
878 	 * any more */
879 	for (i = 0; i < mdev->caps.num_ports; i++)
880 		priv->port[i + 1].tmp_type = 0;
881 
882 	err = mlx4_change_port_types(mdev, new_types);
883 
884 out:
885 	mlx4_start_sense(mdev);
886 	mutex_unlock(&priv->port_mutex);
887 	return err ? err : count;
888 }
889 
890 enum ibta_mtu {
891 	IB_MTU_256  = 1,
892 	IB_MTU_512  = 2,
893 	IB_MTU_1024 = 3,
894 	IB_MTU_2048 = 4,
895 	IB_MTU_4096 = 5
896 };
897 
898 static inline int int_to_ibta_mtu(int mtu)
899 {
900 	switch (mtu) {
901 	case 256:  return IB_MTU_256;
902 	case 512:  return IB_MTU_512;
903 	case 1024: return IB_MTU_1024;
904 	case 2048: return IB_MTU_2048;
905 	case 4096: return IB_MTU_4096;
906 	default: return -1;
907 	}
908 }
909 
910 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
911 {
912 	switch (mtu) {
913 	case IB_MTU_256:  return  256;
914 	case IB_MTU_512:  return  512;
915 	case IB_MTU_1024: return 1024;
916 	case IB_MTU_2048: return 2048;
917 	case IB_MTU_4096: return 4096;
918 	default: return -1;
919 	}
920 }
921 
922 static ssize_t show_port_ib_mtu(struct device *dev,
923 			     struct device_attribute *attr,
924 			     char *buf)
925 {
926 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
927 						   port_mtu_attr);
928 	struct mlx4_dev *mdev = info->dev;
929 
930 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
931 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
932 
933 	sprintf(buf, "%d\n",
934 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
935 	return strlen(buf);
936 }
937 
938 static ssize_t set_port_ib_mtu(struct device *dev,
939 			     struct device_attribute *attr,
940 			     const char *buf, size_t count)
941 {
942 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
943 						   port_mtu_attr);
944 	struct mlx4_dev *mdev = info->dev;
945 	struct mlx4_priv *priv = mlx4_priv(mdev);
946 	int err, port, mtu, ibta_mtu = -1;
947 
948 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
949 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
950 		return -EINVAL;
951 	}
952 
953 	err = kstrtoint(buf, 0, &mtu);
954 	if (!err)
955 		ibta_mtu = int_to_ibta_mtu(mtu);
956 
957 	if (err || ibta_mtu < 0) {
958 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
959 		return -EINVAL;
960 	}
961 
962 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
963 
964 	mlx4_stop_sense(mdev);
965 	mutex_lock(&priv->port_mutex);
966 	mlx4_unregister_device(mdev);
967 	for (port = 1; port <= mdev->caps.num_ports; port++) {
968 		mlx4_CLOSE_PORT(mdev, port);
969 		err = mlx4_SET_PORT(mdev, port, -1);
970 		if (err) {
971 			mlx4_err(mdev, "Failed to set port %d, aborting\n",
972 				 port);
973 			goto err_set_port;
974 		}
975 	}
976 	err = mlx4_register_device(mdev);
977 err_set_port:
978 	mutex_unlock(&priv->port_mutex);
979 	mlx4_start_sense(mdev);
980 	return err ? err : count;
981 }
982 
983 static int mlx4_load_fw(struct mlx4_dev *dev)
984 {
985 	struct mlx4_priv *priv = mlx4_priv(dev);
986 	int err;
987 
988 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
989 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
990 	if (!priv->fw.fw_icm) {
991 		mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
992 		return -ENOMEM;
993 	}
994 
995 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
996 	if (err) {
997 		mlx4_err(dev, "MAP_FA command failed, aborting\n");
998 		goto err_free;
999 	}
1000 
1001 	err = mlx4_RUN_FW(dev);
1002 	if (err) {
1003 		mlx4_err(dev, "RUN_FW command failed, aborting\n");
1004 		goto err_unmap_fa;
1005 	}
1006 
1007 	return 0;
1008 
1009 err_unmap_fa:
1010 	mlx4_UNMAP_FA(dev);
1011 
1012 err_free:
1013 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1014 	return err;
1015 }
1016 
1017 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1018 				int cmpt_entry_sz)
1019 {
1020 	struct mlx4_priv *priv = mlx4_priv(dev);
1021 	int err;
1022 	int num_eqs;
1023 
1024 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1025 				  cmpt_base +
1026 				  ((u64) (MLX4_CMPT_TYPE_QP *
1027 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1028 				  cmpt_entry_sz, dev->caps.num_qps,
1029 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1030 				  0, 0);
1031 	if (err)
1032 		goto err;
1033 
1034 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1035 				  cmpt_base +
1036 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1037 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1038 				  cmpt_entry_sz, dev->caps.num_srqs,
1039 				  dev->caps.reserved_srqs, 0, 0);
1040 	if (err)
1041 		goto err_qp;
1042 
1043 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1044 				  cmpt_base +
1045 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1046 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1047 				  cmpt_entry_sz, dev->caps.num_cqs,
1048 				  dev->caps.reserved_cqs, 0, 0);
1049 	if (err)
1050 		goto err_srq;
1051 
1052 	num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
1053 		  dev->caps.num_eqs;
1054 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1055 				  cmpt_base +
1056 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1057 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1058 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1059 	if (err)
1060 		goto err_cq;
1061 
1062 	return 0;
1063 
1064 err_cq:
1065 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1066 
1067 err_srq:
1068 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1069 
1070 err_qp:
1071 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1072 
1073 err:
1074 	return err;
1075 }
1076 
1077 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1078 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1079 {
1080 	struct mlx4_priv *priv = mlx4_priv(dev);
1081 	u64 aux_pages;
1082 	int num_eqs;
1083 	int err;
1084 
1085 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1086 	if (err) {
1087 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1088 		return err;
1089 	}
1090 
1091 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1092 		 (unsigned long long) icm_size >> 10,
1093 		 (unsigned long long) aux_pages << 2);
1094 
1095 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1096 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1097 	if (!priv->fw.aux_icm) {
1098 		mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1099 		return -ENOMEM;
1100 	}
1101 
1102 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1103 	if (err) {
1104 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1105 		goto err_free_aux;
1106 	}
1107 
1108 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1109 	if (err) {
1110 		mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1111 		goto err_unmap_aux;
1112 	}
1113 
1114 
1115 	num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
1116 		   dev->caps.num_eqs;
1117 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1118 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1119 				  num_eqs, num_eqs, 0, 0);
1120 	if (err) {
1121 		mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1122 		goto err_unmap_cmpt;
1123 	}
1124 
1125 	/*
1126 	 * Reserved MTT entries must be aligned up to a cacheline
1127 	 * boundary, since the FW will write to them, while the driver
1128 	 * writes to all other MTT entries. (The variable
1129 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1130 	 * size, not the raw entry size)
1131 	 */
1132 	dev->caps.reserved_mtts =
1133 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1134 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1135 
1136 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1137 				  init_hca->mtt_base,
1138 				  dev->caps.mtt_entry_sz,
1139 				  dev->caps.num_mtts,
1140 				  dev->caps.reserved_mtts, 1, 0);
1141 	if (err) {
1142 		mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1143 		goto err_unmap_eq;
1144 	}
1145 
1146 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1147 				  init_hca->dmpt_base,
1148 				  dev_cap->dmpt_entry_sz,
1149 				  dev->caps.num_mpts,
1150 				  dev->caps.reserved_mrws, 1, 1);
1151 	if (err) {
1152 		mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1153 		goto err_unmap_mtt;
1154 	}
1155 
1156 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1157 				  init_hca->qpc_base,
1158 				  dev_cap->qpc_entry_sz,
1159 				  dev->caps.num_qps,
1160 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1161 				  0, 0);
1162 	if (err) {
1163 		mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1164 		goto err_unmap_dmpt;
1165 	}
1166 
1167 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1168 				  init_hca->auxc_base,
1169 				  dev_cap->aux_entry_sz,
1170 				  dev->caps.num_qps,
1171 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1172 				  0, 0);
1173 	if (err) {
1174 		mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1175 		goto err_unmap_qp;
1176 	}
1177 
1178 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1179 				  init_hca->altc_base,
1180 				  dev_cap->altc_entry_sz,
1181 				  dev->caps.num_qps,
1182 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1183 				  0, 0);
1184 	if (err) {
1185 		mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1186 		goto err_unmap_auxc;
1187 	}
1188 
1189 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1190 				  init_hca->rdmarc_base,
1191 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1192 				  dev->caps.num_qps,
1193 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1194 				  0, 0);
1195 	if (err) {
1196 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1197 		goto err_unmap_altc;
1198 	}
1199 
1200 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1201 				  init_hca->cqc_base,
1202 				  dev_cap->cqc_entry_sz,
1203 				  dev->caps.num_cqs,
1204 				  dev->caps.reserved_cqs, 0, 0);
1205 	if (err) {
1206 		mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1207 		goto err_unmap_rdmarc;
1208 	}
1209 
1210 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1211 				  init_hca->srqc_base,
1212 				  dev_cap->srq_entry_sz,
1213 				  dev->caps.num_srqs,
1214 				  dev->caps.reserved_srqs, 0, 0);
1215 	if (err) {
1216 		mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1217 		goto err_unmap_cq;
1218 	}
1219 
1220 	/*
1221 	 * For flow steering device managed mode it is required to use
1222 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1223 	 * required, but for simplicity just map the whole multicast
1224 	 * group table now.  The table isn't very big and it's a lot
1225 	 * easier than trying to track ref counts.
1226 	 */
1227 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1228 				  init_hca->mc_base,
1229 				  mlx4_get_mgm_entry_size(dev),
1230 				  dev->caps.num_mgms + dev->caps.num_amgms,
1231 				  dev->caps.num_mgms + dev->caps.num_amgms,
1232 				  0, 0);
1233 	if (err) {
1234 		mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1235 		goto err_unmap_srq;
1236 	}
1237 
1238 	return 0;
1239 
1240 err_unmap_srq:
1241 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1242 
1243 err_unmap_cq:
1244 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1245 
1246 err_unmap_rdmarc:
1247 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1248 
1249 err_unmap_altc:
1250 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1251 
1252 err_unmap_auxc:
1253 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1254 
1255 err_unmap_qp:
1256 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1257 
1258 err_unmap_dmpt:
1259 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1260 
1261 err_unmap_mtt:
1262 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1263 
1264 err_unmap_eq:
1265 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1266 
1267 err_unmap_cmpt:
1268 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1269 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1270 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1271 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1272 
1273 err_unmap_aux:
1274 	mlx4_UNMAP_ICM_AUX(dev);
1275 
1276 err_free_aux:
1277 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1278 
1279 	return err;
1280 }
1281 
1282 static void mlx4_free_icms(struct mlx4_dev *dev)
1283 {
1284 	struct mlx4_priv *priv = mlx4_priv(dev);
1285 
1286 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1287 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1288 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1289 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1290 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1291 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1292 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1293 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1294 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1295 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1296 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1297 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1298 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1299 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1300 
1301 	mlx4_UNMAP_ICM_AUX(dev);
1302 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1303 }
1304 
1305 static void mlx4_slave_exit(struct mlx4_dev *dev)
1306 {
1307 	struct mlx4_priv *priv = mlx4_priv(dev);
1308 
1309 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1310 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_TIME))
1311 		mlx4_warn(dev, "Failed to close slave function\n");
1312 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1313 }
1314 
1315 static int map_bf_area(struct mlx4_dev *dev)
1316 {
1317 	struct mlx4_priv *priv = mlx4_priv(dev);
1318 	resource_size_t bf_start;
1319 	resource_size_t bf_len;
1320 	int err = 0;
1321 
1322 	if (!dev->caps.bf_reg_size)
1323 		return -ENXIO;
1324 
1325 	bf_start = pci_resource_start(dev->pdev, 2) +
1326 			(dev->caps.num_uars << PAGE_SHIFT);
1327 	bf_len = pci_resource_len(dev->pdev, 2) -
1328 			(dev->caps.num_uars << PAGE_SHIFT);
1329 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1330 	if (!priv->bf_mapping)
1331 		err = -ENOMEM;
1332 
1333 	return err;
1334 }
1335 
1336 static void unmap_bf_area(struct mlx4_dev *dev)
1337 {
1338 	if (mlx4_priv(dev)->bf_mapping)
1339 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1340 }
1341 
1342 cycle_t mlx4_read_clock(struct mlx4_dev *dev)
1343 {
1344 	u32 clockhi, clocklo, clockhi1;
1345 	cycle_t cycles;
1346 	int i;
1347 	struct mlx4_priv *priv = mlx4_priv(dev);
1348 
1349 	for (i = 0; i < 10; i++) {
1350 		clockhi = swab32(readl(priv->clock_mapping));
1351 		clocklo = swab32(readl(priv->clock_mapping + 4));
1352 		clockhi1 = swab32(readl(priv->clock_mapping));
1353 		if (clockhi == clockhi1)
1354 			break;
1355 	}
1356 
1357 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1358 
1359 	return cycles;
1360 }
1361 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1362 
1363 
1364 static int map_internal_clock(struct mlx4_dev *dev)
1365 {
1366 	struct mlx4_priv *priv = mlx4_priv(dev);
1367 
1368 	priv->clock_mapping =
1369 		ioremap(pci_resource_start(dev->pdev, priv->fw.clock_bar) +
1370 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1371 
1372 	if (!priv->clock_mapping)
1373 		return -ENOMEM;
1374 
1375 	return 0;
1376 }
1377 
1378 static void unmap_internal_clock(struct mlx4_dev *dev)
1379 {
1380 	struct mlx4_priv *priv = mlx4_priv(dev);
1381 
1382 	if (priv->clock_mapping)
1383 		iounmap(priv->clock_mapping);
1384 }
1385 
1386 static void mlx4_close_hca(struct mlx4_dev *dev)
1387 {
1388 	unmap_internal_clock(dev);
1389 	unmap_bf_area(dev);
1390 	if (mlx4_is_slave(dev))
1391 		mlx4_slave_exit(dev);
1392 	else {
1393 		mlx4_CLOSE_HCA(dev, 0);
1394 		mlx4_free_icms(dev);
1395 		mlx4_UNMAP_FA(dev);
1396 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1397 	}
1398 }
1399 
1400 static int mlx4_init_slave(struct mlx4_dev *dev)
1401 {
1402 	struct mlx4_priv *priv = mlx4_priv(dev);
1403 	u64 dma = (u64) priv->mfunc.vhcr_dma;
1404 	int ret_from_reset = 0;
1405 	u32 slave_read;
1406 	u32 cmd_channel_ver;
1407 
1408 	if (atomic_read(&pf_loading)) {
1409 		mlx4_warn(dev, "PF is not ready - Deferring probe\n");
1410 		return -EPROBE_DEFER;
1411 	}
1412 
1413 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1414 	priv->cmd.max_cmds = 1;
1415 	mlx4_warn(dev, "Sending reset\n");
1416 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1417 				       MLX4_COMM_TIME);
1418 	/* if we are in the middle of flr the slave will try
1419 	 * NUM_OF_RESET_RETRIES times before leaving.*/
1420 	if (ret_from_reset) {
1421 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1422 			mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
1423 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
1424 			return -EPROBE_DEFER;
1425 		} else
1426 			goto err;
1427 	}
1428 
1429 	/* check the driver version - the slave I/F revision
1430 	 * must match the master's */
1431 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1432 	cmd_channel_ver = mlx4_comm_get_version();
1433 
1434 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1435 		MLX4_COMM_GET_IF_REV(slave_read)) {
1436 		mlx4_err(dev, "slave driver version is not supported by the master\n");
1437 		goto err;
1438 	}
1439 
1440 	mlx4_warn(dev, "Sending vhcr0\n");
1441 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1442 						    MLX4_COMM_TIME))
1443 		goto err;
1444 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1445 						    MLX4_COMM_TIME))
1446 		goto err;
1447 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1448 						    MLX4_COMM_TIME))
1449 		goto err;
1450 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma, MLX4_COMM_TIME))
1451 		goto err;
1452 
1453 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1454 	return 0;
1455 
1456 err:
1457 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, 0);
1458 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1459 	return -EIO;
1460 }
1461 
1462 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1463 {
1464 	int i;
1465 
1466 	for (i = 1; i <= dev->caps.num_ports; i++) {
1467 		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
1468 			dev->caps.gid_table_len[i] =
1469 				mlx4_get_slave_num_gids(dev, 0, i);
1470 		else
1471 			dev->caps.gid_table_len[i] = 1;
1472 		dev->caps.pkey_table_len[i] =
1473 			dev->phys_caps.pkey_phys_table_len[i] - 1;
1474 	}
1475 }
1476 
1477 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
1478 {
1479 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
1480 
1481 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
1482 	      i++) {
1483 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
1484 			break;
1485 	}
1486 
1487 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
1488 }
1489 
1490 static void choose_steering_mode(struct mlx4_dev *dev,
1491 				 struct mlx4_dev_cap *dev_cap)
1492 {
1493 	if (mlx4_log_num_mgm_entry_size == -1 &&
1494 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
1495 	    (!mlx4_is_mfunc(dev) ||
1496 	     (dev_cap->fs_max_num_qp_per_entry >= (dev->num_vfs + 1))) &&
1497 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
1498 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
1499 		dev->oper_log_mgm_entry_size =
1500 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
1501 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1502 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
1503 		dev->caps.fs_log_max_ucast_qp_range_size =
1504 			dev_cap->fs_log_max_ucast_qp_range_size;
1505 	} else {
1506 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
1507 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1508 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
1509 		else {
1510 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
1511 
1512 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
1513 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1514 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
1515 		}
1516 		dev->oper_log_mgm_entry_size =
1517 			mlx4_log_num_mgm_entry_size > 0 ?
1518 			mlx4_log_num_mgm_entry_size :
1519 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
1520 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
1521 	}
1522 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
1523 		 mlx4_steering_mode_str(dev->caps.steering_mode),
1524 		 dev->oper_log_mgm_entry_size,
1525 		 mlx4_log_num_mgm_entry_size);
1526 }
1527 
1528 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
1529 				       struct mlx4_dev_cap *dev_cap)
1530 {
1531 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
1532 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
1533 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
1534 	else
1535 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
1536 
1537 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
1538 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
1539 }
1540 
1541 static int mlx4_init_hca(struct mlx4_dev *dev)
1542 {
1543 	struct mlx4_priv	  *priv = mlx4_priv(dev);
1544 	struct mlx4_adapter	   adapter;
1545 	struct mlx4_dev_cap	   dev_cap;
1546 	struct mlx4_mod_stat_cfg   mlx4_cfg;
1547 	struct mlx4_profile	   profile;
1548 	struct mlx4_init_hca_param init_hca;
1549 	u64 icm_size;
1550 	int err;
1551 
1552 	if (!mlx4_is_slave(dev)) {
1553 		err = mlx4_QUERY_FW(dev);
1554 		if (err) {
1555 			if (err == -EACCES)
1556 				mlx4_info(dev, "non-primary physical function, skipping\n");
1557 			else
1558 				mlx4_err(dev, "QUERY_FW command failed, aborting\n");
1559 			return err;
1560 		}
1561 
1562 		err = mlx4_load_fw(dev);
1563 		if (err) {
1564 			mlx4_err(dev, "Failed to start FW, aborting\n");
1565 			return err;
1566 		}
1567 
1568 		mlx4_cfg.log_pg_sz_m = 1;
1569 		mlx4_cfg.log_pg_sz = 0;
1570 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
1571 		if (err)
1572 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
1573 
1574 		err = mlx4_dev_cap(dev, &dev_cap);
1575 		if (err) {
1576 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
1577 			goto err_stop_fw;
1578 		}
1579 
1580 		choose_steering_mode(dev, &dev_cap);
1581 		choose_tunnel_offload_mode(dev, &dev_cap);
1582 
1583 		err = mlx4_get_phys_port_id(dev);
1584 		if (err)
1585 			mlx4_err(dev, "Fail to get physical port id\n");
1586 
1587 		if (mlx4_is_master(dev))
1588 			mlx4_parav_master_pf_caps(dev);
1589 
1590 		profile = default_profile;
1591 		if (dev->caps.steering_mode ==
1592 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
1593 			profile.num_mcg = MLX4_FS_NUM_MCG;
1594 
1595 		icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
1596 					     &init_hca);
1597 		if ((long long) icm_size < 0) {
1598 			err = icm_size;
1599 			goto err_stop_fw;
1600 		}
1601 
1602 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
1603 
1604 		init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
1605 		init_hca.uar_page_sz = PAGE_SHIFT - 12;
1606 		init_hca.mw_enabled = 0;
1607 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1608 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
1609 			init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
1610 
1611 		err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
1612 		if (err)
1613 			goto err_stop_fw;
1614 
1615 		err = mlx4_INIT_HCA(dev, &init_hca);
1616 		if (err) {
1617 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
1618 			goto err_free_icm;
1619 		}
1620 		/*
1621 		 * If TS is supported by FW
1622 		 * read HCA frequency by QUERY_HCA command
1623 		 */
1624 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
1625 			memset(&init_hca, 0, sizeof(init_hca));
1626 			err = mlx4_QUERY_HCA(dev, &init_hca);
1627 			if (err) {
1628 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
1629 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1630 			} else {
1631 				dev->caps.hca_core_clock =
1632 					init_hca.hca_core_clock;
1633 			}
1634 
1635 			/* In case we got HCA frequency 0 - disable timestamping
1636 			 * to avoid dividing by zero
1637 			 */
1638 			if (!dev->caps.hca_core_clock) {
1639 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1640 				mlx4_err(dev,
1641 					 "HCA frequency is 0 - timestamping is not supported\n");
1642 			} else if (map_internal_clock(dev)) {
1643 				/*
1644 				 * Map internal clock,
1645 				 * in case of failure disable timestamping
1646 				 */
1647 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1648 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
1649 			}
1650 		}
1651 	} else {
1652 		err = mlx4_init_slave(dev);
1653 		if (err) {
1654 			if (err != -EPROBE_DEFER)
1655 				mlx4_err(dev, "Failed to initialize slave\n");
1656 			return err;
1657 		}
1658 
1659 		err = mlx4_slave_cap(dev);
1660 		if (err) {
1661 			mlx4_err(dev, "Failed to obtain slave caps\n");
1662 			goto err_close;
1663 		}
1664 	}
1665 
1666 	if (map_bf_area(dev))
1667 		mlx4_dbg(dev, "Failed to map blue flame area\n");
1668 
1669 	/*Only the master set the ports, all the rest got it from it.*/
1670 	if (!mlx4_is_slave(dev))
1671 		mlx4_set_port_mask(dev);
1672 
1673 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
1674 	if (err) {
1675 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
1676 		goto unmap_bf;
1677 	}
1678 
1679 	priv->eq_table.inta_pin = adapter.inta_pin;
1680 	memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
1681 
1682 	return 0;
1683 
1684 unmap_bf:
1685 	unmap_internal_clock(dev);
1686 	unmap_bf_area(dev);
1687 
1688 	if (mlx4_is_slave(dev)) {
1689 		kfree(dev->caps.qp0_qkey);
1690 		kfree(dev->caps.qp0_tunnel);
1691 		kfree(dev->caps.qp0_proxy);
1692 		kfree(dev->caps.qp1_tunnel);
1693 		kfree(dev->caps.qp1_proxy);
1694 	}
1695 
1696 err_close:
1697 	if (mlx4_is_slave(dev))
1698 		mlx4_slave_exit(dev);
1699 	else
1700 		mlx4_CLOSE_HCA(dev, 0);
1701 
1702 err_free_icm:
1703 	if (!mlx4_is_slave(dev))
1704 		mlx4_free_icms(dev);
1705 
1706 err_stop_fw:
1707 	if (!mlx4_is_slave(dev)) {
1708 		mlx4_UNMAP_FA(dev);
1709 		mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1710 	}
1711 	return err;
1712 }
1713 
1714 static int mlx4_init_counters_table(struct mlx4_dev *dev)
1715 {
1716 	struct mlx4_priv *priv = mlx4_priv(dev);
1717 	int nent;
1718 
1719 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1720 		return -ENOENT;
1721 
1722 	nent = dev->caps.max_counters;
1723 	return mlx4_bitmap_init(&priv->counters_bitmap, nent, nent - 1, 0, 0);
1724 }
1725 
1726 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
1727 {
1728 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
1729 }
1730 
1731 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1732 {
1733 	struct mlx4_priv *priv = mlx4_priv(dev);
1734 
1735 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1736 		return -ENOENT;
1737 
1738 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
1739 	if (*idx == -1)
1740 		return -ENOMEM;
1741 
1742 	return 0;
1743 }
1744 
1745 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1746 {
1747 	u64 out_param;
1748 	int err;
1749 
1750 	if (mlx4_is_mfunc(dev)) {
1751 		err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
1752 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
1753 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
1754 		if (!err)
1755 			*idx = get_param_l(&out_param);
1756 
1757 		return err;
1758 	}
1759 	return __mlx4_counter_alloc(dev, idx);
1760 }
1761 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
1762 
1763 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1764 {
1765 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
1766 	return;
1767 }
1768 
1769 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1770 {
1771 	u64 in_param = 0;
1772 
1773 	if (mlx4_is_mfunc(dev)) {
1774 		set_param_l(&in_param, idx);
1775 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
1776 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
1777 			 MLX4_CMD_WRAPPED);
1778 		return;
1779 	}
1780 	__mlx4_counter_free(dev, idx);
1781 }
1782 EXPORT_SYMBOL_GPL(mlx4_counter_free);
1783 
1784 static int mlx4_setup_hca(struct mlx4_dev *dev)
1785 {
1786 	struct mlx4_priv *priv = mlx4_priv(dev);
1787 	int err;
1788 	int port;
1789 	__be32 ib_port_default_caps;
1790 
1791 	err = mlx4_init_uar_table(dev);
1792 	if (err) {
1793 		mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
1794 		 return err;
1795 	}
1796 
1797 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
1798 	if (err) {
1799 		mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
1800 		goto err_uar_table_free;
1801 	}
1802 
1803 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
1804 	if (!priv->kar) {
1805 		mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
1806 		err = -ENOMEM;
1807 		goto err_uar_free;
1808 	}
1809 
1810 	err = mlx4_init_pd_table(dev);
1811 	if (err) {
1812 		mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
1813 		goto err_kar_unmap;
1814 	}
1815 
1816 	err = mlx4_init_xrcd_table(dev);
1817 	if (err) {
1818 		mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
1819 		goto err_pd_table_free;
1820 	}
1821 
1822 	err = mlx4_init_mr_table(dev);
1823 	if (err) {
1824 		mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
1825 		goto err_xrcd_table_free;
1826 	}
1827 
1828 	if (!mlx4_is_slave(dev)) {
1829 		err = mlx4_init_mcg_table(dev);
1830 		if (err) {
1831 			mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
1832 			goto err_mr_table_free;
1833 		}
1834 	}
1835 
1836 	err = mlx4_init_eq_table(dev);
1837 	if (err) {
1838 		mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
1839 		goto err_mcg_table_free;
1840 	}
1841 
1842 	err = mlx4_cmd_use_events(dev);
1843 	if (err) {
1844 		mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
1845 		goto err_eq_table_free;
1846 	}
1847 
1848 	err = mlx4_NOP(dev);
1849 	if (err) {
1850 		if (dev->flags & MLX4_FLAG_MSI_X) {
1851 			mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
1852 				  priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1853 			mlx4_warn(dev, "Trying again without MSI-X\n");
1854 		} else {
1855 			mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
1856 				 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1857 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
1858 		}
1859 
1860 		goto err_cmd_poll;
1861 	}
1862 
1863 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
1864 
1865 	err = mlx4_init_cq_table(dev);
1866 	if (err) {
1867 		mlx4_err(dev, "Failed to initialize 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 shared receive queue table, aborting\n");
1874 		goto err_cq_table_free;
1875 	}
1876 
1877 	err = mlx4_init_qp_table(dev);
1878 	if (err) {
1879 		mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
1880 		goto err_srq_table_free;
1881 	}
1882 
1883 	err = mlx4_init_counters_table(dev);
1884 	if (err && err != -ENOENT) {
1885 		mlx4_err(dev, "Failed to initialize counters table, aborting\n");
1886 		goto err_qp_table_free;
1887 	}
1888 
1889 	if (!mlx4_is_slave(dev)) {
1890 		for (port = 1; port <= dev->caps.num_ports; port++) {
1891 			ib_port_default_caps = 0;
1892 			err = mlx4_get_port_ib_caps(dev, port,
1893 						    &ib_port_default_caps);
1894 			if (err)
1895 				mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
1896 					  port, err);
1897 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
1898 
1899 			/* initialize per-slave default ib port capabilities */
1900 			if (mlx4_is_master(dev)) {
1901 				int i;
1902 				for (i = 0; i < dev->num_slaves; i++) {
1903 					if (i == mlx4_master_func_num(dev))
1904 						continue;
1905 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
1906 						ib_port_default_caps;
1907 				}
1908 			}
1909 
1910 			if (mlx4_is_mfunc(dev))
1911 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
1912 			else
1913 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
1914 
1915 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
1916 					    dev->caps.pkey_table_len[port] : -1);
1917 			if (err) {
1918 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1919 					 port);
1920 				goto err_counters_table_free;
1921 			}
1922 		}
1923 	}
1924 
1925 	return 0;
1926 
1927 err_counters_table_free:
1928 	mlx4_cleanup_counters_table(dev);
1929 
1930 err_qp_table_free:
1931 	mlx4_cleanup_qp_table(dev);
1932 
1933 err_srq_table_free:
1934 	mlx4_cleanup_srq_table(dev);
1935 
1936 err_cq_table_free:
1937 	mlx4_cleanup_cq_table(dev);
1938 
1939 err_cmd_poll:
1940 	mlx4_cmd_use_polling(dev);
1941 
1942 err_eq_table_free:
1943 	mlx4_cleanup_eq_table(dev);
1944 
1945 err_mcg_table_free:
1946 	if (!mlx4_is_slave(dev))
1947 		mlx4_cleanup_mcg_table(dev);
1948 
1949 err_mr_table_free:
1950 	mlx4_cleanup_mr_table(dev);
1951 
1952 err_xrcd_table_free:
1953 	mlx4_cleanup_xrcd_table(dev);
1954 
1955 err_pd_table_free:
1956 	mlx4_cleanup_pd_table(dev);
1957 
1958 err_kar_unmap:
1959 	iounmap(priv->kar);
1960 
1961 err_uar_free:
1962 	mlx4_uar_free(dev, &priv->driver_uar);
1963 
1964 err_uar_table_free:
1965 	mlx4_cleanup_uar_table(dev);
1966 	return err;
1967 }
1968 
1969 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
1970 {
1971 	struct mlx4_priv *priv = mlx4_priv(dev);
1972 	struct msix_entry *entries;
1973 	int nreq = min_t(int, dev->caps.num_ports *
1974 			 min_t(int, num_online_cpus() + 1,
1975 			       MAX_MSIX_P_PORT) + MSIX_LEGACY_SZ, MAX_MSIX);
1976 	int i;
1977 
1978 	if (msi_x) {
1979 		nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
1980 			     nreq);
1981 
1982 		entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
1983 		if (!entries)
1984 			goto no_msi;
1985 
1986 		for (i = 0; i < nreq; ++i)
1987 			entries[i].entry = i;
1988 
1989 		nreq = pci_enable_msix_range(dev->pdev, entries, 2, nreq);
1990 
1991 		if (nreq < 0) {
1992 			kfree(entries);
1993 			goto no_msi;
1994 		} else if (nreq < MSIX_LEGACY_SZ +
1995 			   dev->caps.num_ports * MIN_MSIX_P_PORT) {
1996 			/*Working in legacy mode , all EQ's shared*/
1997 			dev->caps.comp_pool           = 0;
1998 			dev->caps.num_comp_vectors = nreq - 1;
1999 		} else {
2000 			dev->caps.comp_pool           = nreq - MSIX_LEGACY_SZ;
2001 			dev->caps.num_comp_vectors = MSIX_LEGACY_SZ - 1;
2002 		}
2003 		for (i = 0; i < nreq; ++i)
2004 			priv->eq_table.eq[i].irq = entries[i].vector;
2005 
2006 		dev->flags |= MLX4_FLAG_MSI_X;
2007 
2008 		kfree(entries);
2009 		return;
2010 	}
2011 
2012 no_msi:
2013 	dev->caps.num_comp_vectors = 1;
2014 	dev->caps.comp_pool	   = 0;
2015 
2016 	for (i = 0; i < 2; ++i)
2017 		priv->eq_table.eq[i].irq = dev->pdev->irq;
2018 }
2019 
2020 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2021 {
2022 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2023 	int err = 0;
2024 
2025 	info->dev = dev;
2026 	info->port = port;
2027 	if (!mlx4_is_slave(dev)) {
2028 		mlx4_init_mac_table(dev, &info->mac_table);
2029 		mlx4_init_vlan_table(dev, &info->vlan_table);
2030 		mlx4_init_roce_gid_table(dev, &info->gid_table);
2031 		info->base_qpn = mlx4_get_base_qpn(dev, port);
2032 	}
2033 
2034 	sprintf(info->dev_name, "mlx4_port%d", port);
2035 	info->port_attr.attr.name = info->dev_name;
2036 	if (mlx4_is_mfunc(dev))
2037 		info->port_attr.attr.mode = S_IRUGO;
2038 	else {
2039 		info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2040 		info->port_attr.store     = set_port_type;
2041 	}
2042 	info->port_attr.show      = show_port_type;
2043 	sysfs_attr_init(&info->port_attr.attr);
2044 
2045 	err = device_create_file(&dev->pdev->dev, &info->port_attr);
2046 	if (err) {
2047 		mlx4_err(dev, "Failed to create file for port %d\n", port);
2048 		info->port = -1;
2049 	}
2050 
2051 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2052 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
2053 	if (mlx4_is_mfunc(dev))
2054 		info->port_mtu_attr.attr.mode = S_IRUGO;
2055 	else {
2056 		info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2057 		info->port_mtu_attr.store     = set_port_ib_mtu;
2058 	}
2059 	info->port_mtu_attr.show      = show_port_ib_mtu;
2060 	sysfs_attr_init(&info->port_mtu_attr.attr);
2061 
2062 	err = device_create_file(&dev->pdev->dev, &info->port_mtu_attr);
2063 	if (err) {
2064 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
2065 		device_remove_file(&info->dev->pdev->dev, &info->port_attr);
2066 		info->port = -1;
2067 	}
2068 
2069 	return err;
2070 }
2071 
2072 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
2073 {
2074 	if (info->port < 0)
2075 		return;
2076 
2077 	device_remove_file(&info->dev->pdev->dev, &info->port_attr);
2078 	device_remove_file(&info->dev->pdev->dev, &info->port_mtu_attr);
2079 }
2080 
2081 static int mlx4_init_steering(struct mlx4_dev *dev)
2082 {
2083 	struct mlx4_priv *priv = mlx4_priv(dev);
2084 	int num_entries = dev->caps.num_ports;
2085 	int i, j;
2086 
2087 	priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
2088 	if (!priv->steer)
2089 		return -ENOMEM;
2090 
2091 	for (i = 0; i < num_entries; i++)
2092 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
2093 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
2094 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
2095 		}
2096 	return 0;
2097 }
2098 
2099 static void mlx4_clear_steering(struct mlx4_dev *dev)
2100 {
2101 	struct mlx4_priv *priv = mlx4_priv(dev);
2102 	struct mlx4_steer_index *entry, *tmp_entry;
2103 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
2104 	int num_entries = dev->caps.num_ports;
2105 	int i, j;
2106 
2107 	for (i = 0; i < num_entries; i++) {
2108 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
2109 			list_for_each_entry_safe(pqp, tmp_pqp,
2110 						 &priv->steer[i].promisc_qps[j],
2111 						 list) {
2112 				list_del(&pqp->list);
2113 				kfree(pqp);
2114 			}
2115 			list_for_each_entry_safe(entry, tmp_entry,
2116 						 &priv->steer[i].steer_entries[j],
2117 						 list) {
2118 				list_del(&entry->list);
2119 				list_for_each_entry_safe(pqp, tmp_pqp,
2120 							 &entry->duplicates,
2121 							 list) {
2122 					list_del(&pqp->list);
2123 					kfree(pqp);
2124 				}
2125 				kfree(entry);
2126 			}
2127 		}
2128 	}
2129 	kfree(priv->steer);
2130 }
2131 
2132 static int extended_func_num(struct pci_dev *pdev)
2133 {
2134 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
2135 }
2136 
2137 #define MLX4_OWNER_BASE	0x8069c
2138 #define MLX4_OWNER_SIZE	4
2139 
2140 static int mlx4_get_ownership(struct mlx4_dev *dev)
2141 {
2142 	void __iomem *owner;
2143 	u32 ret;
2144 
2145 	if (pci_channel_offline(dev->pdev))
2146 		return -EIO;
2147 
2148 	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
2149 			MLX4_OWNER_SIZE);
2150 	if (!owner) {
2151 		mlx4_err(dev, "Failed to obtain ownership bit\n");
2152 		return -ENOMEM;
2153 	}
2154 
2155 	ret = readl(owner);
2156 	iounmap(owner);
2157 	return (int) !!ret;
2158 }
2159 
2160 static void mlx4_free_ownership(struct mlx4_dev *dev)
2161 {
2162 	void __iomem *owner;
2163 
2164 	if (pci_channel_offline(dev->pdev))
2165 		return;
2166 
2167 	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
2168 			MLX4_OWNER_SIZE);
2169 	if (!owner) {
2170 		mlx4_err(dev, "Failed to obtain ownership bit\n");
2171 		return;
2172 	}
2173 	writel(0, owner);
2174 	msleep(1000);
2175 	iounmap(owner);
2176 }
2177 
2178 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data)
2179 {
2180 	struct mlx4_priv *priv;
2181 	struct mlx4_dev *dev;
2182 	int err;
2183 	int port;
2184 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
2185 	int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
2186 	const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
2187 		{2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
2188 	unsigned total_vfs = 0;
2189 	int sriov_initialized = 0;
2190 	unsigned int i;
2191 
2192 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
2193 
2194 	err = pci_enable_device(pdev);
2195 	if (err) {
2196 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
2197 		return err;
2198 	}
2199 
2200 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
2201 	 * per port, we must limit the number of VFs to 63 (since their are
2202 	 * 128 MACs)
2203 	 */
2204 	for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
2205 	     total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
2206 		nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
2207 		if (nvfs[i] < 0) {
2208 			dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
2209 			return -EINVAL;
2210 		}
2211 	}
2212 	for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
2213 	     i++) {
2214 		prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
2215 		if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
2216 			dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
2217 			return -EINVAL;
2218 		}
2219 	}
2220 	if (total_vfs >= MLX4_MAX_NUM_VF) {
2221 		dev_err(&pdev->dev,
2222 			"Requested more VF's (%d) than allowed (%d)\n",
2223 			total_vfs, MLX4_MAX_NUM_VF - 1);
2224 		return -EINVAL;
2225 	}
2226 
2227 	for (i = 0; i < MLX4_MAX_PORTS; i++) {
2228 		if (nvfs[i] + nvfs[2] >= MLX4_MAX_NUM_VF_P_PORT) {
2229 			dev_err(&pdev->dev,
2230 				"Requested more VF's (%d) for port (%d) than allowed (%d)\n",
2231 				nvfs[i] + nvfs[2], i + 1,
2232 				MLX4_MAX_NUM_VF_P_PORT - 1);
2233 			return -EINVAL;
2234 		}
2235 	}
2236 
2237 
2238 	/*
2239 	 * Check for BARs.
2240 	 */
2241 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
2242 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
2243 		dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
2244 			pci_dev_data, pci_resource_flags(pdev, 0));
2245 		err = -ENODEV;
2246 		goto err_disable_pdev;
2247 	}
2248 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
2249 		dev_err(&pdev->dev, "Missing UAR, aborting\n");
2250 		err = -ENODEV;
2251 		goto err_disable_pdev;
2252 	}
2253 
2254 	err = pci_request_regions(pdev, DRV_NAME);
2255 	if (err) {
2256 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
2257 		goto err_disable_pdev;
2258 	}
2259 
2260 	pci_set_master(pdev);
2261 
2262 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2263 	if (err) {
2264 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
2265 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2266 		if (err) {
2267 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
2268 			goto err_release_regions;
2269 		}
2270 	}
2271 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2272 	if (err) {
2273 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
2274 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2275 		if (err) {
2276 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
2277 			goto err_release_regions;
2278 		}
2279 	}
2280 
2281 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
2282 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
2283 
2284 	dev       = pci_get_drvdata(pdev);
2285 	priv      = mlx4_priv(dev);
2286 	dev->pdev = pdev;
2287 	INIT_LIST_HEAD(&priv->ctx_list);
2288 	spin_lock_init(&priv->ctx_lock);
2289 
2290 	mutex_init(&priv->port_mutex);
2291 
2292 	INIT_LIST_HEAD(&priv->pgdir_list);
2293 	mutex_init(&priv->pgdir_mutex);
2294 
2295 	INIT_LIST_HEAD(&priv->bf_list);
2296 	mutex_init(&priv->bf_mutex);
2297 
2298 	dev->rev_id = pdev->revision;
2299 	dev->numa_node = dev_to_node(&pdev->dev);
2300 	/* Detect if this device is a virtual function */
2301 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
2302 		/* When acting as pf, we normally skip vfs unless explicitly
2303 		 * requested to probe them. */
2304 		if (total_vfs) {
2305 			unsigned vfs_offset = 0;
2306 			for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
2307 				     vfs_offset + nvfs[i] < extended_func_num(pdev);
2308 			     vfs_offset += nvfs[i], i++)
2309 				;
2310 			if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
2311 				err = -ENODEV;
2312 				goto err_free_dev;
2313 			}
2314 			if ((extended_func_num(pdev) - vfs_offset)
2315 			    > prb_vf[i]) {
2316 				mlx4_warn(dev, "Skipping virtual function:%d\n",
2317 					  extended_func_num(pdev));
2318 				err = -ENODEV;
2319 				goto err_free_dev;
2320 			}
2321 		}
2322 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
2323 		dev->flags |= MLX4_FLAG_SLAVE;
2324 	} else {
2325 		/* We reset the device and enable SRIOV only for physical
2326 		 * devices.  Try to claim ownership on the device;
2327 		 * if already taken, skip -- do not allow multiple PFs */
2328 		err = mlx4_get_ownership(dev);
2329 		if (err) {
2330 			if (err < 0)
2331 				goto err_free_dev;
2332 			else {
2333 				mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
2334 				err = -EINVAL;
2335 				goto err_free_dev;
2336 			}
2337 		}
2338 
2339 		if (total_vfs) {
2340 			mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n",
2341 				  total_vfs);
2342 			dev->dev_vfs = kzalloc(
2343 				total_vfs * sizeof(*dev->dev_vfs),
2344 				GFP_KERNEL);
2345 			if (NULL == dev->dev_vfs) {
2346 				mlx4_err(dev, "Failed to allocate memory for VFs\n");
2347 				err = 0;
2348 			} else {
2349 				atomic_inc(&pf_loading);
2350 				err = pci_enable_sriov(pdev, total_vfs);
2351 				if (err) {
2352 					mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
2353 						 err);
2354 					atomic_dec(&pf_loading);
2355 					err = 0;
2356 				} else {
2357 					mlx4_warn(dev, "Running in master mode\n");
2358 					dev->flags |= MLX4_FLAG_SRIOV |
2359 						MLX4_FLAG_MASTER;
2360 					dev->num_vfs = total_vfs;
2361 					sriov_initialized = 1;
2362 				}
2363 			}
2364 		}
2365 
2366 		atomic_set(&priv->opreq_count, 0);
2367 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
2368 
2369 		/*
2370 		 * Now reset the HCA before we touch the PCI capabilities or
2371 		 * attempt a firmware command, since a boot ROM may have left
2372 		 * the HCA in an undefined state.
2373 		 */
2374 		err = mlx4_reset(dev);
2375 		if (err) {
2376 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
2377 			goto err_rel_own;
2378 		}
2379 	}
2380 
2381 slave_start:
2382 	err = mlx4_cmd_init(dev);
2383 	if (err) {
2384 		mlx4_err(dev, "Failed to init command interface, aborting\n");
2385 		goto err_sriov;
2386 	}
2387 
2388 	/* In slave functions, the communication channel must be initialized
2389 	 * before posting commands. Also, init num_slaves before calling
2390 	 * mlx4_init_hca */
2391 	if (mlx4_is_mfunc(dev)) {
2392 		if (mlx4_is_master(dev))
2393 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
2394 		else {
2395 			dev->num_slaves = 0;
2396 			err = mlx4_multi_func_init(dev);
2397 			if (err) {
2398 				mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
2399 				goto err_cmd;
2400 			}
2401 		}
2402 	}
2403 
2404 	err = mlx4_init_hca(dev);
2405 	if (err) {
2406 		if (err == -EACCES) {
2407 			/* Not primary Physical function
2408 			 * Running in slave mode */
2409 			mlx4_cmd_cleanup(dev);
2410 			dev->flags |= MLX4_FLAG_SLAVE;
2411 			dev->flags &= ~MLX4_FLAG_MASTER;
2412 			goto slave_start;
2413 		} else
2414 			goto err_mfunc;
2415 	}
2416 
2417 	/* check if the device is functioning at its maximum possible speed.
2418 	 * No return code for this call, just warn the user in case of PCI
2419 	 * express device capabilities are under-satisfied by the bus.
2420 	 */
2421 	if (!mlx4_is_slave(dev))
2422 		mlx4_check_pcie_caps(dev);
2423 
2424 	/* In master functions, the communication channel must be initialized
2425 	 * after obtaining its address from fw */
2426 	if (mlx4_is_master(dev)) {
2427 		unsigned sum = 0;
2428 		err = mlx4_multi_func_init(dev);
2429 		if (err) {
2430 			mlx4_err(dev, "Failed to init master mfunc interface, aborting\n");
2431 			goto err_close;
2432 		}
2433 		if (sriov_initialized) {
2434 			int ib_ports = 0;
2435 			mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2436 				ib_ports++;
2437 
2438 			if (ib_ports &&
2439 			    (num_vfs_argc > 1 || probe_vfs_argc > 1)) {
2440 				mlx4_err(dev,
2441 					 "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");
2442 				goto err_close;
2443 			}
2444 			for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]); i++) {
2445 				unsigned j;
2446 				for (j = 0; j < nvfs[i]; ++sum, ++j) {
2447 					dev->dev_vfs[sum].min_port =
2448 						i < 2 ? i + 1 : 1;
2449 					dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
2450 						dev->caps.num_ports;
2451 				}
2452 			}
2453 		}
2454 	}
2455 
2456 	err = mlx4_alloc_eq_table(dev);
2457 	if (err)
2458 		goto err_master_mfunc;
2459 
2460 	priv->msix_ctl.pool_bm = 0;
2461 	mutex_init(&priv->msix_ctl.pool_lock);
2462 
2463 	mlx4_enable_msi_x(dev);
2464 	if ((mlx4_is_mfunc(dev)) &&
2465 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
2466 		err = -ENOSYS;
2467 		mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
2468 		goto err_free_eq;
2469 	}
2470 
2471 	if (!mlx4_is_slave(dev)) {
2472 		err = mlx4_init_steering(dev);
2473 		if (err)
2474 			goto err_free_eq;
2475 	}
2476 
2477 	err = mlx4_setup_hca(dev);
2478 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
2479 	    !mlx4_is_mfunc(dev)) {
2480 		dev->flags &= ~MLX4_FLAG_MSI_X;
2481 		dev->caps.num_comp_vectors = 1;
2482 		dev->caps.comp_pool	   = 0;
2483 		pci_disable_msix(pdev);
2484 		err = mlx4_setup_hca(dev);
2485 	}
2486 
2487 	if (err)
2488 		goto err_steer;
2489 
2490 	mlx4_init_quotas(dev);
2491 
2492 	for (port = 1; port <= dev->caps.num_ports; port++) {
2493 		err = mlx4_init_port_info(dev, port);
2494 		if (err)
2495 			goto err_port;
2496 	}
2497 
2498 	err = mlx4_register_device(dev);
2499 	if (err)
2500 		goto err_port;
2501 
2502 	mlx4_request_modules(dev);
2503 
2504 	mlx4_sense_init(dev);
2505 	mlx4_start_sense(dev);
2506 
2507 	priv->removed = 0;
2508 
2509 	if (mlx4_is_master(dev) && dev->num_vfs)
2510 		atomic_dec(&pf_loading);
2511 
2512 	return 0;
2513 
2514 err_port:
2515 	for (--port; port >= 1; --port)
2516 		mlx4_cleanup_port_info(&priv->port[port]);
2517 
2518 	mlx4_cleanup_counters_table(dev);
2519 	mlx4_cleanup_qp_table(dev);
2520 	mlx4_cleanup_srq_table(dev);
2521 	mlx4_cleanup_cq_table(dev);
2522 	mlx4_cmd_use_polling(dev);
2523 	mlx4_cleanup_eq_table(dev);
2524 	mlx4_cleanup_mcg_table(dev);
2525 	mlx4_cleanup_mr_table(dev);
2526 	mlx4_cleanup_xrcd_table(dev);
2527 	mlx4_cleanup_pd_table(dev);
2528 	mlx4_cleanup_uar_table(dev);
2529 
2530 err_steer:
2531 	if (!mlx4_is_slave(dev))
2532 		mlx4_clear_steering(dev);
2533 
2534 err_free_eq:
2535 	mlx4_free_eq_table(dev);
2536 
2537 err_master_mfunc:
2538 	if (mlx4_is_master(dev))
2539 		mlx4_multi_func_cleanup(dev);
2540 
2541 	if (mlx4_is_slave(dev)) {
2542 		kfree(dev->caps.qp0_qkey);
2543 		kfree(dev->caps.qp0_tunnel);
2544 		kfree(dev->caps.qp0_proxy);
2545 		kfree(dev->caps.qp1_tunnel);
2546 		kfree(dev->caps.qp1_proxy);
2547 	}
2548 
2549 err_close:
2550 	if (dev->flags & MLX4_FLAG_MSI_X)
2551 		pci_disable_msix(pdev);
2552 
2553 	mlx4_close_hca(dev);
2554 
2555 err_mfunc:
2556 	if (mlx4_is_slave(dev))
2557 		mlx4_multi_func_cleanup(dev);
2558 
2559 err_cmd:
2560 	mlx4_cmd_cleanup(dev);
2561 
2562 err_sriov:
2563 	if (dev->flags & MLX4_FLAG_SRIOV)
2564 		pci_disable_sriov(pdev);
2565 
2566 err_rel_own:
2567 	if (!mlx4_is_slave(dev))
2568 		mlx4_free_ownership(dev);
2569 
2570 	if (mlx4_is_master(dev) && dev->num_vfs)
2571 		atomic_dec(&pf_loading);
2572 
2573 	kfree(priv->dev.dev_vfs);
2574 
2575 err_free_dev:
2576 	kfree(priv);
2577 
2578 err_release_regions:
2579 	pci_release_regions(pdev);
2580 
2581 err_disable_pdev:
2582 	pci_disable_device(pdev);
2583 	pci_set_drvdata(pdev, NULL);
2584 	return err;
2585 }
2586 
2587 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
2588 {
2589 	struct mlx4_priv *priv;
2590 	struct mlx4_dev *dev;
2591 
2592 	printk_once(KERN_INFO "%s", mlx4_version);
2593 
2594 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
2595 	if (!priv)
2596 		return -ENOMEM;
2597 
2598 	dev       = &priv->dev;
2599 	pci_set_drvdata(pdev, dev);
2600 	priv->pci_dev_data = id->driver_data;
2601 
2602 	return __mlx4_init_one(pdev, id->driver_data);
2603 }
2604 
2605 static void __mlx4_remove_one(struct pci_dev *pdev)
2606 {
2607 	struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
2608 	struct mlx4_priv *priv = mlx4_priv(dev);
2609 	int               pci_dev_data;
2610 	int p;
2611 
2612 	if (priv->removed)
2613 		return;
2614 
2615 	pci_dev_data = priv->pci_dev_data;
2616 
2617 	/* in SRIOV it is not allowed to unload the pf's
2618 	 * driver while there are alive vf's */
2619 	if (mlx4_is_master(dev) && mlx4_how_many_lives_vf(dev))
2620 		pr_warn("Removing PF when there are assigned VF's !!!\n");
2621 	mlx4_stop_sense(dev);
2622 	mlx4_unregister_device(dev);
2623 
2624 	for (p = 1; p <= dev->caps.num_ports; p++) {
2625 		mlx4_cleanup_port_info(&priv->port[p]);
2626 		mlx4_CLOSE_PORT(dev, p);
2627 	}
2628 
2629 	if (mlx4_is_master(dev))
2630 		mlx4_free_resource_tracker(dev,
2631 					   RES_TR_FREE_SLAVES_ONLY);
2632 
2633 	mlx4_cleanup_counters_table(dev);
2634 	mlx4_cleanup_qp_table(dev);
2635 	mlx4_cleanup_srq_table(dev);
2636 	mlx4_cleanup_cq_table(dev);
2637 	mlx4_cmd_use_polling(dev);
2638 	mlx4_cleanup_eq_table(dev);
2639 	mlx4_cleanup_mcg_table(dev);
2640 	mlx4_cleanup_mr_table(dev);
2641 	mlx4_cleanup_xrcd_table(dev);
2642 	mlx4_cleanup_pd_table(dev);
2643 
2644 	if (mlx4_is_master(dev))
2645 		mlx4_free_resource_tracker(dev,
2646 					   RES_TR_FREE_STRUCTS_ONLY);
2647 
2648 	iounmap(priv->kar);
2649 	mlx4_uar_free(dev, &priv->driver_uar);
2650 	mlx4_cleanup_uar_table(dev);
2651 	if (!mlx4_is_slave(dev))
2652 		mlx4_clear_steering(dev);
2653 	mlx4_free_eq_table(dev);
2654 	if (mlx4_is_master(dev))
2655 		mlx4_multi_func_cleanup(dev);
2656 	mlx4_close_hca(dev);
2657 	if (mlx4_is_slave(dev))
2658 		mlx4_multi_func_cleanup(dev);
2659 	mlx4_cmd_cleanup(dev);
2660 
2661 	if (dev->flags & MLX4_FLAG_MSI_X)
2662 		pci_disable_msix(pdev);
2663 	if (dev->flags & MLX4_FLAG_SRIOV) {
2664 		mlx4_warn(dev, "Disabling SR-IOV\n");
2665 		pci_disable_sriov(pdev);
2666 		dev->num_vfs = 0;
2667 	}
2668 
2669 	if (!mlx4_is_slave(dev))
2670 		mlx4_free_ownership(dev);
2671 
2672 	kfree(dev->caps.qp0_qkey);
2673 	kfree(dev->caps.qp0_tunnel);
2674 	kfree(dev->caps.qp0_proxy);
2675 	kfree(dev->caps.qp1_tunnel);
2676 	kfree(dev->caps.qp1_proxy);
2677 	kfree(dev->dev_vfs);
2678 
2679 	pci_release_regions(pdev);
2680 	pci_disable_device(pdev);
2681 	memset(priv, 0, sizeof(*priv));
2682 	priv->pci_dev_data = pci_dev_data;
2683 	priv->removed = 1;
2684 }
2685 
2686 static void mlx4_remove_one(struct pci_dev *pdev)
2687 {
2688 	struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
2689 	struct mlx4_priv *priv = mlx4_priv(dev);
2690 
2691 	__mlx4_remove_one(pdev);
2692 	kfree(priv);
2693 	pci_set_drvdata(pdev, NULL);
2694 }
2695 
2696 int mlx4_restart_one(struct pci_dev *pdev)
2697 {
2698 	struct mlx4_dev	 *dev  = pci_get_drvdata(pdev);
2699 	struct mlx4_priv *priv = mlx4_priv(dev);
2700 	int		  pci_dev_data;
2701 
2702 	pci_dev_data = priv->pci_dev_data;
2703 	__mlx4_remove_one(pdev);
2704 	return __mlx4_init_one(pdev, pci_dev_data);
2705 }
2706 
2707 static DEFINE_PCI_DEVICE_TABLE(mlx4_pci_table) = {
2708 	/* MT25408 "Hermon" SDR */
2709 	{ PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2710 	/* MT25408 "Hermon" DDR */
2711 	{ PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2712 	/* MT25408 "Hermon" QDR */
2713 	{ PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2714 	/* MT25408 "Hermon" DDR PCIe gen2 */
2715 	{ PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2716 	/* MT25408 "Hermon" QDR PCIe gen2 */
2717 	{ PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2718 	/* MT25408 "Hermon" EN 10GigE */
2719 	{ PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2720 	/* MT25408 "Hermon" EN 10GigE PCIe gen2 */
2721 	{ PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2722 	/* MT25458 ConnectX EN 10GBASE-T 10GigE */
2723 	{ PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2724 	/* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
2725 	{ PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2726 	/* MT26468 ConnectX EN 10GigE PCIe gen2*/
2727 	{ PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2728 	/* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
2729 	{ PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2730 	/* MT26478 ConnectX2 40GigE PCIe gen2 */
2731 	{ PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2732 	/* MT25400 Family [ConnectX-2 Virtual Function] */
2733 	{ PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
2734 	/* MT27500 Family [ConnectX-3] */
2735 	{ PCI_VDEVICE(MELLANOX, 0x1003), 0 },
2736 	/* MT27500 Family [ConnectX-3 Virtual Function] */
2737 	{ PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
2738 	{ PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
2739 	{ PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
2740 	{ PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
2741 	{ PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
2742 	{ PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
2743 	{ PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
2744 	{ PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
2745 	{ PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
2746 	{ PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
2747 	{ PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
2748 	{ PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
2749 	{ PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
2750 	{ 0, }
2751 };
2752 
2753 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
2754 
2755 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
2756 					      pci_channel_state_t state)
2757 {
2758 	__mlx4_remove_one(pdev);
2759 
2760 	return state == pci_channel_io_perm_failure ?
2761 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
2762 }
2763 
2764 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
2765 {
2766 	struct mlx4_dev	 *dev  = pci_get_drvdata(pdev);
2767 	struct mlx4_priv *priv = mlx4_priv(dev);
2768 	int               ret;
2769 
2770 	ret = __mlx4_init_one(pdev, priv->pci_dev_data);
2771 
2772 	return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
2773 }
2774 
2775 static const struct pci_error_handlers mlx4_err_handler = {
2776 	.error_detected = mlx4_pci_err_detected,
2777 	.slot_reset     = mlx4_pci_slot_reset,
2778 };
2779 
2780 static struct pci_driver mlx4_driver = {
2781 	.name		= DRV_NAME,
2782 	.id_table	= mlx4_pci_table,
2783 	.probe		= mlx4_init_one,
2784 	.shutdown	= __mlx4_remove_one,
2785 	.remove		= mlx4_remove_one,
2786 	.err_handler    = &mlx4_err_handler,
2787 };
2788 
2789 static int __init mlx4_verify_params(void)
2790 {
2791 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
2792 		pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
2793 		return -1;
2794 	}
2795 
2796 	if (log_num_vlan != 0)
2797 		pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
2798 			MLX4_LOG_NUM_VLANS);
2799 
2800 	if (use_prio != 0)
2801 		pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
2802 
2803 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
2804 		pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
2805 			log_mtts_per_seg);
2806 		return -1;
2807 	}
2808 
2809 	/* Check if module param for ports type has legal combination */
2810 	if (port_type_array[0] == false && port_type_array[1] == true) {
2811 		pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
2812 		port_type_array[0] = true;
2813 	}
2814 
2815 	if (mlx4_log_num_mgm_entry_size != -1 &&
2816 	    (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
2817 	     mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE)) {
2818 		pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-1 or %d..%d)\n",
2819 			mlx4_log_num_mgm_entry_size,
2820 			MLX4_MIN_MGM_LOG_ENTRY_SIZE,
2821 			MLX4_MAX_MGM_LOG_ENTRY_SIZE);
2822 		return -1;
2823 	}
2824 
2825 	return 0;
2826 }
2827 
2828 static int __init mlx4_init(void)
2829 {
2830 	int ret;
2831 
2832 	if (mlx4_verify_params())
2833 		return -EINVAL;
2834 
2835 	mlx4_catas_init();
2836 
2837 	mlx4_wq = create_singlethread_workqueue("mlx4");
2838 	if (!mlx4_wq)
2839 		return -ENOMEM;
2840 
2841 	ret = pci_register_driver(&mlx4_driver);
2842 	if (ret < 0)
2843 		destroy_workqueue(mlx4_wq);
2844 	return ret < 0 ? ret : 0;
2845 }
2846 
2847 static void __exit mlx4_cleanup(void)
2848 {
2849 	pci_unregister_driver(&mlx4_driver);
2850 	destroy_workqueue(mlx4_wq);
2851 }
2852 
2853 module_init(mlx4_init);
2854 module_exit(mlx4_cleanup);
2855