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 struct mlx4_profile low_mem_profile = {
124 	.num_qp		= 1 << 17,
125 	.num_srq	= 1 << 6,
126 	.rdmarc_per_qp	= 1 << 4,
127 	.num_cq		= 1 << 8,
128 	.num_mcg	= 1 << 8,
129 	.num_mpt	= 1 << 9,
130 	.num_mtt	= 1 << 7,
131 };
132 
133 static int log_num_mac = 7;
134 module_param_named(log_num_mac, log_num_mac, int, 0444);
135 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
136 
137 static int log_num_vlan;
138 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
139 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
140 /* Log2 max number of VLANs per ETH port (0-7) */
141 #define MLX4_LOG_NUM_VLANS 7
142 #define MLX4_MIN_LOG_NUM_VLANS 0
143 #define MLX4_MIN_LOG_NUM_MAC 1
144 
145 static bool use_prio;
146 module_param_named(use_prio, use_prio, bool, 0444);
147 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
148 
149 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
150 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
151 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
152 
153 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
154 static int arr_argc = 2;
155 module_param_array(port_type_array, int, &arr_argc, 0444);
156 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
157 				"1 for IB, 2 for Ethernet");
158 
159 struct mlx4_port_config {
160 	struct list_head list;
161 	enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
162 	struct pci_dev *pdev;
163 };
164 
165 static atomic_t pf_loading = ATOMIC_INIT(0);
166 
167 int mlx4_check_port_params(struct mlx4_dev *dev,
168 			   enum mlx4_port_type *port_type)
169 {
170 	int i;
171 
172 	for (i = 0; i < dev->caps.num_ports - 1; i++) {
173 		if (port_type[i] != port_type[i + 1]) {
174 			if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
175 				mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
176 				return -EINVAL;
177 			}
178 		}
179 	}
180 
181 	for (i = 0; i < dev->caps.num_ports; i++) {
182 		if (!(port_type[i] & dev->caps.supported_type[i+1])) {
183 			mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
184 				 i + 1);
185 			return -EINVAL;
186 		}
187 	}
188 	return 0;
189 }
190 
191 static void mlx4_set_port_mask(struct mlx4_dev *dev)
192 {
193 	int i;
194 
195 	for (i = 1; i <= dev->caps.num_ports; ++i)
196 		dev->caps.port_mask[i] = dev->caps.port_type[i];
197 }
198 
199 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
200 {
201 	int err;
202 	int i;
203 
204 	err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
205 	if (err) {
206 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
207 		return err;
208 	}
209 
210 	if (dev_cap->min_page_sz > PAGE_SIZE) {
211 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
212 			 dev_cap->min_page_sz, PAGE_SIZE);
213 		return -ENODEV;
214 	}
215 	if (dev_cap->num_ports > MLX4_MAX_PORTS) {
216 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
217 			 dev_cap->num_ports, MLX4_MAX_PORTS);
218 		return -ENODEV;
219 	}
220 
221 	if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
222 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
223 			 dev_cap->uar_size,
224 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
225 		return -ENODEV;
226 	}
227 
228 	dev->caps.num_ports	     = dev_cap->num_ports;
229 	dev->phys_caps.num_phys_eqs  = MLX4_MAX_EQ_NUM;
230 	for (i = 1; i <= dev->caps.num_ports; ++i) {
231 		dev->caps.vl_cap[i]	    = dev_cap->max_vl[i];
232 		dev->caps.ib_mtu_cap[i]	    = dev_cap->ib_mtu[i];
233 		dev->phys_caps.gid_phys_table_len[i]  = dev_cap->max_gids[i];
234 		dev->phys_caps.pkey_phys_table_len[i] = dev_cap->max_pkeys[i];
235 		/* set gid and pkey table operating lengths by default
236 		 * to non-sriov values */
237 		dev->caps.gid_table_len[i]  = dev_cap->max_gids[i];
238 		dev->caps.pkey_table_len[i] = dev_cap->max_pkeys[i];
239 		dev->caps.port_width_cap[i] = dev_cap->max_port_width[i];
240 		dev->caps.eth_mtu_cap[i]    = dev_cap->eth_mtu[i];
241 		dev->caps.def_mac[i]        = dev_cap->def_mac[i];
242 		dev->caps.supported_type[i] = dev_cap->supported_port_types[i];
243 		dev->caps.suggested_type[i] = dev_cap->suggested_type[i];
244 		dev->caps.default_sense[i] = dev_cap->default_sense[i];
245 		dev->caps.trans_type[i]	    = dev_cap->trans_type[i];
246 		dev->caps.vendor_oui[i]     = dev_cap->vendor_oui[i];
247 		dev->caps.wavelength[i]     = dev_cap->wavelength[i];
248 		dev->caps.trans_code[i]     = dev_cap->trans_code[i];
249 	}
250 
251 	dev->caps.uar_page_size	     = PAGE_SIZE;
252 	dev->caps.num_uars	     = dev_cap->uar_size / PAGE_SIZE;
253 	dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
254 	dev->caps.bf_reg_size	     = dev_cap->bf_reg_size;
255 	dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
256 	dev->caps.max_sq_sg	     = dev_cap->max_sq_sg;
257 	dev->caps.max_rq_sg	     = dev_cap->max_rq_sg;
258 	dev->caps.max_wqes	     = dev_cap->max_qp_sz;
259 	dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
260 	dev->caps.max_srq_wqes	     = dev_cap->max_srq_sz;
261 	dev->caps.max_srq_sge	     = dev_cap->max_rq_sg - 1;
262 	dev->caps.reserved_srqs	     = dev_cap->reserved_srqs;
263 	dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
264 	dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
265 	/*
266 	 * Subtract 1 from the limit because we need to allocate a
267 	 * spare CQE so the HCA HW can tell the difference between an
268 	 * empty CQ and a full CQ.
269 	 */
270 	dev->caps.max_cqes	     = dev_cap->max_cq_sz - 1;
271 	dev->caps.reserved_cqs	     = dev_cap->reserved_cqs;
272 	dev->caps.reserved_eqs	     = dev_cap->reserved_eqs;
273 	dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
274 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
275 
276 	/* The first 128 UARs are used for EQ doorbells */
277 	dev->caps.reserved_uars	     = max_t(int, 128, dev_cap->reserved_uars);
278 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
279 	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
280 					dev_cap->reserved_xrcds : 0;
281 	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
282 					dev_cap->max_xrcds : 0;
283 	dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
284 
285 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
286 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
287 	dev->caps.flags		     = dev_cap->flags;
288 	dev->caps.flags2	     = dev_cap->flags2;
289 	dev->caps.bmme_flags	     = dev_cap->bmme_flags;
290 	dev->caps.reserved_lkey	     = dev_cap->reserved_lkey;
291 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
292 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
293 	dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
294 
295 	/* Sense port always allowed on supported devices for ConnectX-1 and -2 */
296 	if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
297 		dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
298 	/* Don't do sense port on multifunction devices (for now at least) */
299 	if (mlx4_is_mfunc(dev))
300 		dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
301 
302 	if (mlx4_low_memory_profile()) {
303 		dev->caps.log_num_macs  = MLX4_MIN_LOG_NUM_MAC;
304 		dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
305 	} else {
306 		dev->caps.log_num_macs  = log_num_mac;
307 		dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
308 	}
309 
310 	for (i = 1; i <= dev->caps.num_ports; ++i) {
311 		dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
312 		if (dev->caps.supported_type[i]) {
313 			/* if only ETH is supported - assign ETH */
314 			if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
315 				dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
316 			/* if only IB is supported, assign IB */
317 			else if (dev->caps.supported_type[i] ==
318 				 MLX4_PORT_TYPE_IB)
319 				dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
320 			else {
321 				/* if IB and ETH are supported, we set the port
322 				 * type according to user selection of port type;
323 				 * if user selected none, take the FW hint */
324 				if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
325 					dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
326 						MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
327 				else
328 					dev->caps.port_type[i] = port_type_array[i - 1];
329 			}
330 		}
331 		/*
332 		 * Link sensing is allowed on the port if 3 conditions are true:
333 		 * 1. Both protocols are supported on the port.
334 		 * 2. Different types are supported on the port
335 		 * 3. FW declared that it supports link sensing
336 		 */
337 		mlx4_priv(dev)->sense.sense_allowed[i] =
338 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
339 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
340 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
341 
342 		/*
343 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
344 		 * and perform sense_port FW command to try and set the correct
345 		 * port type from beginning
346 		 */
347 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
348 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
349 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
350 			mlx4_SENSE_PORT(dev, i, &sensed_port);
351 			if (sensed_port != MLX4_PORT_TYPE_NONE)
352 				dev->caps.port_type[i] = sensed_port;
353 		} else {
354 			dev->caps.possible_type[i] = dev->caps.port_type[i];
355 		}
356 
357 		if (dev->caps.log_num_macs > dev_cap->log_max_macs[i]) {
358 			dev->caps.log_num_macs = dev_cap->log_max_macs[i];
359 			mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
360 				  i, 1 << dev->caps.log_num_macs);
361 		}
362 		if (dev->caps.log_num_vlans > dev_cap->log_max_vlans[i]) {
363 			dev->caps.log_num_vlans = dev_cap->log_max_vlans[i];
364 			mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
365 				  i, 1 << dev->caps.log_num_vlans);
366 		}
367 	}
368 
369 	dev->caps.max_counters = 1 << ilog2(dev_cap->max_counters);
370 
371 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
372 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
373 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
374 		(1 << dev->caps.log_num_macs) *
375 		(1 << dev->caps.log_num_vlans) *
376 		dev->caps.num_ports;
377 	dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
378 
379 	dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
380 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
381 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
382 		dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
383 
384 	dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
385 
386 	if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
387 		if (dev_cap->flags &
388 		    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
389 			mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
390 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
391 			dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
392 		}
393 	}
394 
395 	if ((dev->caps.flags &
396 	    (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
397 	    mlx4_is_master(dev))
398 		dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
399 
400 	return 0;
401 }
402 
403 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
404 				       enum pci_bus_speed *speed,
405 				       enum pcie_link_width *width)
406 {
407 	u32 lnkcap1, lnkcap2;
408 	int err1, err2;
409 
410 #define  PCIE_MLW_CAP_SHIFT 4	/* start of MLW mask in link capabilities */
411 
412 	*speed = PCI_SPEED_UNKNOWN;
413 	*width = PCIE_LNK_WIDTH_UNKNOWN;
414 
415 	err1 = pcie_capability_read_dword(dev->pdev, PCI_EXP_LNKCAP, &lnkcap1);
416 	err2 = pcie_capability_read_dword(dev->pdev, PCI_EXP_LNKCAP2, &lnkcap2);
417 	if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
418 		if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
419 			*speed = PCIE_SPEED_8_0GT;
420 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
421 			*speed = PCIE_SPEED_5_0GT;
422 		else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
423 			*speed = PCIE_SPEED_2_5GT;
424 	}
425 	if (!err1) {
426 		*width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
427 		if (!lnkcap2) { /* pre-r3.0 */
428 			if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
429 				*speed = PCIE_SPEED_5_0GT;
430 			else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
431 				*speed = PCIE_SPEED_2_5GT;
432 		}
433 	}
434 
435 	if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
436 		return err1 ? err1 :
437 			err2 ? err2 : -EINVAL;
438 	}
439 	return 0;
440 }
441 
442 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
443 {
444 	enum pcie_link_width width, width_cap;
445 	enum pci_bus_speed speed, speed_cap;
446 	int err;
447 
448 #define PCIE_SPEED_STR(speed) \
449 	(speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
450 	 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
451 	 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
452 	 "Unknown")
453 
454 	err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
455 	if (err) {
456 		mlx4_warn(dev,
457 			  "Unable to determine PCIe device BW capabilities\n");
458 		return;
459 	}
460 
461 	err = pcie_get_minimum_link(dev->pdev, &speed, &width);
462 	if (err || speed == PCI_SPEED_UNKNOWN ||
463 	    width == PCIE_LNK_WIDTH_UNKNOWN) {
464 		mlx4_warn(dev,
465 			  "Unable to determine PCI device chain minimum BW\n");
466 		return;
467 	}
468 
469 	if (width != width_cap || speed != speed_cap)
470 		mlx4_warn(dev,
471 			  "PCIe BW is different than device's capability\n");
472 
473 	mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
474 		  PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
475 	mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
476 		  width, width_cap);
477 	return;
478 }
479 
480 /*The function checks if there are live vf, return the num of them*/
481 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
482 {
483 	struct mlx4_priv *priv = mlx4_priv(dev);
484 	struct mlx4_slave_state *s_state;
485 	int i;
486 	int ret = 0;
487 
488 	for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
489 		s_state = &priv->mfunc.master.slave_state[i];
490 		if (s_state->active && s_state->last_cmd !=
491 		    MLX4_COMM_CMD_RESET) {
492 			mlx4_warn(dev, "%s: slave: %d is still active\n",
493 				  __func__, i);
494 			ret++;
495 		}
496 	}
497 	return ret;
498 }
499 
500 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
501 {
502 	u32 qk = MLX4_RESERVED_QKEY_BASE;
503 
504 	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
505 	    qpn < dev->phys_caps.base_proxy_sqpn)
506 		return -EINVAL;
507 
508 	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
509 		/* tunnel qp */
510 		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
511 	else
512 		qk += qpn - dev->phys_caps.base_proxy_sqpn;
513 	*qkey = qk;
514 	return 0;
515 }
516 EXPORT_SYMBOL(mlx4_get_parav_qkey);
517 
518 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
519 {
520 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
521 
522 	if (!mlx4_is_master(dev))
523 		return;
524 
525 	priv->virt2phys_pkey[slave][port - 1][i] = val;
526 }
527 EXPORT_SYMBOL(mlx4_sync_pkey_table);
528 
529 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
530 {
531 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
532 
533 	if (!mlx4_is_master(dev))
534 		return;
535 
536 	priv->slave_node_guids[slave] = guid;
537 }
538 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
539 
540 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
541 {
542 	struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
543 
544 	if (!mlx4_is_master(dev))
545 		return 0;
546 
547 	return priv->slave_node_guids[slave];
548 }
549 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
550 
551 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
552 {
553 	struct mlx4_priv *priv = mlx4_priv(dev);
554 	struct mlx4_slave_state *s_slave;
555 
556 	if (!mlx4_is_master(dev))
557 		return 0;
558 
559 	s_slave = &priv->mfunc.master.slave_state[slave];
560 	return !!s_slave->active;
561 }
562 EXPORT_SYMBOL(mlx4_is_slave_active);
563 
564 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
565 				       struct mlx4_dev_cap *dev_cap,
566 				       struct mlx4_init_hca_param *hca_param)
567 {
568 	dev->caps.steering_mode = hca_param->steering_mode;
569 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
570 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
571 		dev->caps.fs_log_max_ucast_qp_range_size =
572 			dev_cap->fs_log_max_ucast_qp_range_size;
573 	} else
574 		dev->caps.num_qp_per_mgm =
575 			4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
576 
577 	mlx4_dbg(dev, "Steering mode is: %s\n",
578 		 mlx4_steering_mode_str(dev->caps.steering_mode));
579 }
580 
581 static int mlx4_slave_cap(struct mlx4_dev *dev)
582 {
583 	int			   err;
584 	u32			   page_size;
585 	struct mlx4_dev_cap	   dev_cap;
586 	struct mlx4_func_cap	   func_cap;
587 	struct mlx4_init_hca_param hca_param;
588 	int			   i;
589 
590 	memset(&hca_param, 0, sizeof(hca_param));
591 	err = mlx4_QUERY_HCA(dev, &hca_param);
592 	if (err) {
593 		mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
594 		return err;
595 	}
596 
597 	/* fail if the hca has an unknown global capability
598 	 * at this time global_caps should be always zeroed
599 	 */
600 	if (hca_param.global_caps) {
601 		mlx4_err(dev, "Unknown hca global capabilities\n");
602 		return -ENOSYS;
603 	}
604 
605 	mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
606 
607 	dev->caps.hca_core_clock = hca_param.hca_core_clock;
608 
609 	memset(&dev_cap, 0, sizeof(dev_cap));
610 	dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
611 	err = mlx4_dev_cap(dev, &dev_cap);
612 	if (err) {
613 		mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
614 		return err;
615 	}
616 
617 	err = mlx4_QUERY_FW(dev);
618 	if (err)
619 		mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
620 
621 	page_size = ~dev->caps.page_size_cap + 1;
622 	mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
623 	if (page_size > PAGE_SIZE) {
624 		mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
625 			 page_size, PAGE_SIZE);
626 		return -ENODEV;
627 	}
628 
629 	/* slave gets uar page size from QUERY_HCA fw command */
630 	dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
631 
632 	/* TODO: relax this assumption */
633 	if (dev->caps.uar_page_size != PAGE_SIZE) {
634 		mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
635 			 dev->caps.uar_page_size, PAGE_SIZE);
636 		return -ENODEV;
637 	}
638 
639 	memset(&func_cap, 0, sizeof(func_cap));
640 	err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
641 	if (err) {
642 		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
643 			 err);
644 		return err;
645 	}
646 
647 	if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
648 	    PF_CONTEXT_BEHAVIOUR_MASK) {
649 		mlx4_err(dev, "Unknown pf context behaviour\n");
650 		return -ENOSYS;
651 	}
652 
653 	dev->caps.num_ports		= func_cap.num_ports;
654 	dev->quotas.qp			= func_cap.qp_quota;
655 	dev->quotas.srq			= func_cap.srq_quota;
656 	dev->quotas.cq			= func_cap.cq_quota;
657 	dev->quotas.mpt			= func_cap.mpt_quota;
658 	dev->quotas.mtt			= func_cap.mtt_quota;
659 	dev->caps.num_qps		= 1 << hca_param.log_num_qps;
660 	dev->caps.num_srqs		= 1 << hca_param.log_num_srqs;
661 	dev->caps.num_cqs		= 1 << hca_param.log_num_cqs;
662 	dev->caps.num_mpts		= 1 << hca_param.log_mpt_sz;
663 	dev->caps.num_eqs		= func_cap.max_eq;
664 	dev->caps.reserved_eqs		= func_cap.reserved_eq;
665 	dev->caps.num_pds               = MLX4_NUM_PDS;
666 	dev->caps.num_mgms              = 0;
667 	dev->caps.num_amgms             = 0;
668 
669 	if (dev->caps.num_ports > MLX4_MAX_PORTS) {
670 		mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
671 			 dev->caps.num_ports, MLX4_MAX_PORTS);
672 		return -ENODEV;
673 	}
674 
675 	dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
676 	dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
677 	dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
678 	dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
679 	dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
680 
681 	if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
682 	    !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
683 	    !dev->caps.qp0_qkey) {
684 		err = -ENOMEM;
685 		goto err_mem;
686 	}
687 
688 	for (i = 1; i <= dev->caps.num_ports; ++i) {
689 		err = mlx4_QUERY_FUNC_CAP(dev, (u32) i, &func_cap);
690 		if (err) {
691 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
692 				 i, err);
693 			goto err_mem;
694 		}
695 		dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
696 		dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
697 		dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
698 		dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
699 		dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
700 		dev->caps.port_mask[i] = dev->caps.port_type[i];
701 		dev->caps.phys_port_id[i] = func_cap.phys_port_id;
702 		if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
703 						    &dev->caps.gid_table_len[i],
704 						    &dev->caps.pkey_table_len[i]))
705 			goto err_mem;
706 	}
707 
708 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
709 				       dev->caps.reserved_uars) >
710 				       pci_resource_len(dev->pdev, 2)) {
711 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
712 			 dev->caps.uar_page_size * dev->caps.num_uars,
713 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
714 		goto err_mem;
715 	}
716 
717 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
718 		dev->caps.eqe_size   = 64;
719 		dev->caps.eqe_factor = 1;
720 	} else {
721 		dev->caps.eqe_size   = 32;
722 		dev->caps.eqe_factor = 0;
723 	}
724 
725 	if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
726 		dev->caps.cqe_size   = 64;
727 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_64B_CQE;
728 	} else {
729 		dev->caps.cqe_size   = 32;
730 	}
731 
732 	dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
733 	mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
734 
735 	slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
736 
737 	return 0;
738 
739 err_mem:
740 	kfree(dev->caps.qp0_qkey);
741 	kfree(dev->caps.qp0_tunnel);
742 	kfree(dev->caps.qp0_proxy);
743 	kfree(dev->caps.qp1_tunnel);
744 	kfree(dev->caps.qp1_proxy);
745 	dev->caps.qp0_qkey = NULL;
746 	dev->caps.qp0_tunnel = NULL;
747 	dev->caps.qp0_proxy = NULL;
748 	dev->caps.qp1_tunnel = NULL;
749 	dev->caps.qp1_proxy = NULL;
750 
751 	return err;
752 }
753 
754 static void mlx4_request_modules(struct mlx4_dev *dev)
755 {
756 	int port;
757 	int has_ib_port = false;
758 	int has_eth_port = false;
759 #define EN_DRV_NAME	"mlx4_en"
760 #define IB_DRV_NAME	"mlx4_ib"
761 
762 	for (port = 1; port <= dev->caps.num_ports; port++) {
763 		if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
764 			has_ib_port = true;
765 		else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
766 			has_eth_port = true;
767 	}
768 
769 	if (has_eth_port)
770 		request_module_nowait(EN_DRV_NAME);
771 	if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
772 		request_module_nowait(IB_DRV_NAME);
773 }
774 
775 /*
776  * Change the port configuration of the device.
777  * Every user of this function must hold the port mutex.
778  */
779 int mlx4_change_port_types(struct mlx4_dev *dev,
780 			   enum mlx4_port_type *port_types)
781 {
782 	int err = 0;
783 	int change = 0;
784 	int port;
785 
786 	for (port = 0; port <  dev->caps.num_ports; port++) {
787 		/* Change the port type only if the new type is different
788 		 * from the current, and not set to Auto */
789 		if (port_types[port] != dev->caps.port_type[port + 1])
790 			change = 1;
791 	}
792 	if (change) {
793 		mlx4_unregister_device(dev);
794 		for (port = 1; port <= dev->caps.num_ports; port++) {
795 			mlx4_CLOSE_PORT(dev, port);
796 			dev->caps.port_type[port] = port_types[port - 1];
797 			err = mlx4_SET_PORT(dev, port, -1);
798 			if (err) {
799 				mlx4_err(dev, "Failed to set port %d, aborting\n",
800 					 port);
801 				goto out;
802 			}
803 		}
804 		mlx4_set_port_mask(dev);
805 		err = mlx4_register_device(dev);
806 		if (err) {
807 			mlx4_err(dev, "Failed to register device\n");
808 			goto out;
809 		}
810 		mlx4_request_modules(dev);
811 	}
812 
813 out:
814 	return err;
815 }
816 
817 static ssize_t show_port_type(struct device *dev,
818 			      struct device_attribute *attr,
819 			      char *buf)
820 {
821 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
822 						   port_attr);
823 	struct mlx4_dev *mdev = info->dev;
824 	char type[8];
825 
826 	sprintf(type, "%s",
827 		(mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
828 		"ib" : "eth");
829 	if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
830 		sprintf(buf, "auto (%s)\n", type);
831 	else
832 		sprintf(buf, "%s\n", type);
833 
834 	return strlen(buf);
835 }
836 
837 static ssize_t set_port_type(struct device *dev,
838 			     struct device_attribute *attr,
839 			     const char *buf, size_t count)
840 {
841 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
842 						   port_attr);
843 	struct mlx4_dev *mdev = info->dev;
844 	struct mlx4_priv *priv = mlx4_priv(mdev);
845 	enum mlx4_port_type types[MLX4_MAX_PORTS];
846 	enum mlx4_port_type new_types[MLX4_MAX_PORTS];
847 	int i;
848 	int err = 0;
849 
850 	if (!strcmp(buf, "ib\n"))
851 		info->tmp_type = MLX4_PORT_TYPE_IB;
852 	else if (!strcmp(buf, "eth\n"))
853 		info->tmp_type = MLX4_PORT_TYPE_ETH;
854 	else if (!strcmp(buf, "auto\n"))
855 		info->tmp_type = MLX4_PORT_TYPE_AUTO;
856 	else {
857 		mlx4_err(mdev, "%s is not supported port type\n", buf);
858 		return -EINVAL;
859 	}
860 
861 	mlx4_stop_sense(mdev);
862 	mutex_lock(&priv->port_mutex);
863 	/* Possible type is always the one that was delivered */
864 	mdev->caps.possible_type[info->port] = info->tmp_type;
865 
866 	for (i = 0; i < mdev->caps.num_ports; i++) {
867 		types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
868 					mdev->caps.possible_type[i+1];
869 		if (types[i] == MLX4_PORT_TYPE_AUTO)
870 			types[i] = mdev->caps.port_type[i+1];
871 	}
872 
873 	if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
874 	    !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
875 		for (i = 1; i <= mdev->caps.num_ports; i++) {
876 			if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
877 				mdev->caps.possible_type[i] = mdev->caps.port_type[i];
878 				err = -EINVAL;
879 			}
880 		}
881 	}
882 	if (err) {
883 		mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
884 		goto out;
885 	}
886 
887 	mlx4_do_sense_ports(mdev, new_types, types);
888 
889 	err = mlx4_check_port_params(mdev, new_types);
890 	if (err)
891 		goto out;
892 
893 	/* We are about to apply the changes after the configuration
894 	 * was verified, no need to remember the temporary types
895 	 * any more */
896 	for (i = 0; i < mdev->caps.num_ports; i++)
897 		priv->port[i + 1].tmp_type = 0;
898 
899 	err = mlx4_change_port_types(mdev, new_types);
900 
901 out:
902 	mlx4_start_sense(mdev);
903 	mutex_unlock(&priv->port_mutex);
904 	return err ? err : count;
905 }
906 
907 enum ibta_mtu {
908 	IB_MTU_256  = 1,
909 	IB_MTU_512  = 2,
910 	IB_MTU_1024 = 3,
911 	IB_MTU_2048 = 4,
912 	IB_MTU_4096 = 5
913 };
914 
915 static inline int int_to_ibta_mtu(int mtu)
916 {
917 	switch (mtu) {
918 	case 256:  return IB_MTU_256;
919 	case 512:  return IB_MTU_512;
920 	case 1024: return IB_MTU_1024;
921 	case 2048: return IB_MTU_2048;
922 	case 4096: return IB_MTU_4096;
923 	default: return -1;
924 	}
925 }
926 
927 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
928 {
929 	switch (mtu) {
930 	case IB_MTU_256:  return  256;
931 	case IB_MTU_512:  return  512;
932 	case IB_MTU_1024: return 1024;
933 	case IB_MTU_2048: return 2048;
934 	case IB_MTU_4096: return 4096;
935 	default: return -1;
936 	}
937 }
938 
939 static ssize_t show_port_ib_mtu(struct device *dev,
940 			     struct device_attribute *attr,
941 			     char *buf)
942 {
943 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
944 						   port_mtu_attr);
945 	struct mlx4_dev *mdev = info->dev;
946 
947 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
948 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
949 
950 	sprintf(buf, "%d\n",
951 			ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
952 	return strlen(buf);
953 }
954 
955 static ssize_t set_port_ib_mtu(struct device *dev,
956 			     struct device_attribute *attr,
957 			     const char *buf, size_t count)
958 {
959 	struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
960 						   port_mtu_attr);
961 	struct mlx4_dev *mdev = info->dev;
962 	struct mlx4_priv *priv = mlx4_priv(mdev);
963 	int err, port, mtu, ibta_mtu = -1;
964 
965 	if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
966 		mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
967 		return -EINVAL;
968 	}
969 
970 	err = kstrtoint(buf, 0, &mtu);
971 	if (!err)
972 		ibta_mtu = int_to_ibta_mtu(mtu);
973 
974 	if (err || ibta_mtu < 0) {
975 		mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
976 		return -EINVAL;
977 	}
978 
979 	mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
980 
981 	mlx4_stop_sense(mdev);
982 	mutex_lock(&priv->port_mutex);
983 	mlx4_unregister_device(mdev);
984 	for (port = 1; port <= mdev->caps.num_ports; port++) {
985 		mlx4_CLOSE_PORT(mdev, port);
986 		err = mlx4_SET_PORT(mdev, port, -1);
987 		if (err) {
988 			mlx4_err(mdev, "Failed to set port %d, aborting\n",
989 				 port);
990 			goto err_set_port;
991 		}
992 	}
993 	err = mlx4_register_device(mdev);
994 err_set_port:
995 	mutex_unlock(&priv->port_mutex);
996 	mlx4_start_sense(mdev);
997 	return err ? err : count;
998 }
999 
1000 static int mlx4_load_fw(struct mlx4_dev *dev)
1001 {
1002 	struct mlx4_priv *priv = mlx4_priv(dev);
1003 	int err;
1004 
1005 	priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1006 					 GFP_HIGHUSER | __GFP_NOWARN, 0);
1007 	if (!priv->fw.fw_icm) {
1008 		mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1009 		return -ENOMEM;
1010 	}
1011 
1012 	err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1013 	if (err) {
1014 		mlx4_err(dev, "MAP_FA command failed, aborting\n");
1015 		goto err_free;
1016 	}
1017 
1018 	err = mlx4_RUN_FW(dev);
1019 	if (err) {
1020 		mlx4_err(dev, "RUN_FW command failed, aborting\n");
1021 		goto err_unmap_fa;
1022 	}
1023 
1024 	return 0;
1025 
1026 err_unmap_fa:
1027 	mlx4_UNMAP_FA(dev);
1028 
1029 err_free:
1030 	mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1031 	return err;
1032 }
1033 
1034 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1035 				int cmpt_entry_sz)
1036 {
1037 	struct mlx4_priv *priv = mlx4_priv(dev);
1038 	int err;
1039 	int num_eqs;
1040 
1041 	err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1042 				  cmpt_base +
1043 				  ((u64) (MLX4_CMPT_TYPE_QP *
1044 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1045 				  cmpt_entry_sz, dev->caps.num_qps,
1046 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1047 				  0, 0);
1048 	if (err)
1049 		goto err;
1050 
1051 	err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1052 				  cmpt_base +
1053 				  ((u64) (MLX4_CMPT_TYPE_SRQ *
1054 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1055 				  cmpt_entry_sz, dev->caps.num_srqs,
1056 				  dev->caps.reserved_srqs, 0, 0);
1057 	if (err)
1058 		goto err_qp;
1059 
1060 	err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1061 				  cmpt_base +
1062 				  ((u64) (MLX4_CMPT_TYPE_CQ *
1063 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1064 				  cmpt_entry_sz, dev->caps.num_cqs,
1065 				  dev->caps.reserved_cqs, 0, 0);
1066 	if (err)
1067 		goto err_srq;
1068 
1069 	num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
1070 		  dev->caps.num_eqs;
1071 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1072 				  cmpt_base +
1073 				  ((u64) (MLX4_CMPT_TYPE_EQ *
1074 					  cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1075 				  cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1076 	if (err)
1077 		goto err_cq;
1078 
1079 	return 0;
1080 
1081 err_cq:
1082 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1083 
1084 err_srq:
1085 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1086 
1087 err_qp:
1088 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1089 
1090 err:
1091 	return err;
1092 }
1093 
1094 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1095 			 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1096 {
1097 	struct mlx4_priv *priv = mlx4_priv(dev);
1098 	u64 aux_pages;
1099 	int num_eqs;
1100 	int err;
1101 
1102 	err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1103 	if (err) {
1104 		mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1105 		return err;
1106 	}
1107 
1108 	mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1109 		 (unsigned long long) icm_size >> 10,
1110 		 (unsigned long long) aux_pages << 2);
1111 
1112 	priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1113 					  GFP_HIGHUSER | __GFP_NOWARN, 0);
1114 	if (!priv->fw.aux_icm) {
1115 		mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1116 		return -ENOMEM;
1117 	}
1118 
1119 	err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1120 	if (err) {
1121 		mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1122 		goto err_free_aux;
1123 	}
1124 
1125 	err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1126 	if (err) {
1127 		mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1128 		goto err_unmap_aux;
1129 	}
1130 
1131 
1132 	num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
1133 		   dev->caps.num_eqs;
1134 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1135 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
1136 				  num_eqs, num_eqs, 0, 0);
1137 	if (err) {
1138 		mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1139 		goto err_unmap_cmpt;
1140 	}
1141 
1142 	/*
1143 	 * Reserved MTT entries must be aligned up to a cacheline
1144 	 * boundary, since the FW will write to them, while the driver
1145 	 * writes to all other MTT entries. (The variable
1146 	 * dev->caps.mtt_entry_sz below is really the MTT segment
1147 	 * size, not the raw entry size)
1148 	 */
1149 	dev->caps.reserved_mtts =
1150 		ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1151 		      dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1152 
1153 	err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1154 				  init_hca->mtt_base,
1155 				  dev->caps.mtt_entry_sz,
1156 				  dev->caps.num_mtts,
1157 				  dev->caps.reserved_mtts, 1, 0);
1158 	if (err) {
1159 		mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1160 		goto err_unmap_eq;
1161 	}
1162 
1163 	err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1164 				  init_hca->dmpt_base,
1165 				  dev_cap->dmpt_entry_sz,
1166 				  dev->caps.num_mpts,
1167 				  dev->caps.reserved_mrws, 1, 1);
1168 	if (err) {
1169 		mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1170 		goto err_unmap_mtt;
1171 	}
1172 
1173 	err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1174 				  init_hca->qpc_base,
1175 				  dev_cap->qpc_entry_sz,
1176 				  dev->caps.num_qps,
1177 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1178 				  0, 0);
1179 	if (err) {
1180 		mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1181 		goto err_unmap_dmpt;
1182 	}
1183 
1184 	err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1185 				  init_hca->auxc_base,
1186 				  dev_cap->aux_entry_sz,
1187 				  dev->caps.num_qps,
1188 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1189 				  0, 0);
1190 	if (err) {
1191 		mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1192 		goto err_unmap_qp;
1193 	}
1194 
1195 	err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1196 				  init_hca->altc_base,
1197 				  dev_cap->altc_entry_sz,
1198 				  dev->caps.num_qps,
1199 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1200 				  0, 0);
1201 	if (err) {
1202 		mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1203 		goto err_unmap_auxc;
1204 	}
1205 
1206 	err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1207 				  init_hca->rdmarc_base,
1208 				  dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1209 				  dev->caps.num_qps,
1210 				  dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1211 				  0, 0);
1212 	if (err) {
1213 		mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1214 		goto err_unmap_altc;
1215 	}
1216 
1217 	err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1218 				  init_hca->cqc_base,
1219 				  dev_cap->cqc_entry_sz,
1220 				  dev->caps.num_cqs,
1221 				  dev->caps.reserved_cqs, 0, 0);
1222 	if (err) {
1223 		mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1224 		goto err_unmap_rdmarc;
1225 	}
1226 
1227 	err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1228 				  init_hca->srqc_base,
1229 				  dev_cap->srq_entry_sz,
1230 				  dev->caps.num_srqs,
1231 				  dev->caps.reserved_srqs, 0, 0);
1232 	if (err) {
1233 		mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1234 		goto err_unmap_cq;
1235 	}
1236 
1237 	/*
1238 	 * For flow steering device managed mode it is required to use
1239 	 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1240 	 * required, but for simplicity just map the whole multicast
1241 	 * group table now.  The table isn't very big and it's a lot
1242 	 * easier than trying to track ref counts.
1243 	 */
1244 	err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1245 				  init_hca->mc_base,
1246 				  mlx4_get_mgm_entry_size(dev),
1247 				  dev->caps.num_mgms + dev->caps.num_amgms,
1248 				  dev->caps.num_mgms + dev->caps.num_amgms,
1249 				  0, 0);
1250 	if (err) {
1251 		mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1252 		goto err_unmap_srq;
1253 	}
1254 
1255 	return 0;
1256 
1257 err_unmap_srq:
1258 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1259 
1260 err_unmap_cq:
1261 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1262 
1263 err_unmap_rdmarc:
1264 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1265 
1266 err_unmap_altc:
1267 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1268 
1269 err_unmap_auxc:
1270 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1271 
1272 err_unmap_qp:
1273 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1274 
1275 err_unmap_dmpt:
1276 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1277 
1278 err_unmap_mtt:
1279 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1280 
1281 err_unmap_eq:
1282 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1283 
1284 err_unmap_cmpt:
1285 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1286 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1287 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1288 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1289 
1290 err_unmap_aux:
1291 	mlx4_UNMAP_ICM_AUX(dev);
1292 
1293 err_free_aux:
1294 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1295 
1296 	return err;
1297 }
1298 
1299 static void mlx4_free_icms(struct mlx4_dev *dev)
1300 {
1301 	struct mlx4_priv *priv = mlx4_priv(dev);
1302 
1303 	mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1304 	mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1305 	mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1306 	mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1307 	mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1308 	mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1309 	mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1310 	mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1311 	mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1312 	mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1313 	mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1314 	mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1315 	mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1316 	mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1317 
1318 	mlx4_UNMAP_ICM_AUX(dev);
1319 	mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1320 }
1321 
1322 static void mlx4_slave_exit(struct mlx4_dev *dev)
1323 {
1324 	struct mlx4_priv *priv = mlx4_priv(dev);
1325 
1326 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1327 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_TIME))
1328 		mlx4_warn(dev, "Failed to close slave function\n");
1329 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1330 }
1331 
1332 static int map_bf_area(struct mlx4_dev *dev)
1333 {
1334 	struct mlx4_priv *priv = mlx4_priv(dev);
1335 	resource_size_t bf_start;
1336 	resource_size_t bf_len;
1337 	int err = 0;
1338 
1339 	if (!dev->caps.bf_reg_size)
1340 		return -ENXIO;
1341 
1342 	bf_start = pci_resource_start(dev->pdev, 2) +
1343 			(dev->caps.num_uars << PAGE_SHIFT);
1344 	bf_len = pci_resource_len(dev->pdev, 2) -
1345 			(dev->caps.num_uars << PAGE_SHIFT);
1346 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1347 	if (!priv->bf_mapping)
1348 		err = -ENOMEM;
1349 
1350 	return err;
1351 }
1352 
1353 static void unmap_bf_area(struct mlx4_dev *dev)
1354 {
1355 	if (mlx4_priv(dev)->bf_mapping)
1356 		io_mapping_free(mlx4_priv(dev)->bf_mapping);
1357 }
1358 
1359 cycle_t mlx4_read_clock(struct mlx4_dev *dev)
1360 {
1361 	u32 clockhi, clocklo, clockhi1;
1362 	cycle_t cycles;
1363 	int i;
1364 	struct mlx4_priv *priv = mlx4_priv(dev);
1365 
1366 	for (i = 0; i < 10; i++) {
1367 		clockhi = swab32(readl(priv->clock_mapping));
1368 		clocklo = swab32(readl(priv->clock_mapping + 4));
1369 		clockhi1 = swab32(readl(priv->clock_mapping));
1370 		if (clockhi == clockhi1)
1371 			break;
1372 	}
1373 
1374 	cycles = (u64) clockhi << 32 | (u64) clocklo;
1375 
1376 	return cycles;
1377 }
1378 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1379 
1380 
1381 static int map_internal_clock(struct mlx4_dev *dev)
1382 {
1383 	struct mlx4_priv *priv = mlx4_priv(dev);
1384 
1385 	priv->clock_mapping =
1386 		ioremap(pci_resource_start(dev->pdev, priv->fw.clock_bar) +
1387 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1388 
1389 	if (!priv->clock_mapping)
1390 		return -ENOMEM;
1391 
1392 	return 0;
1393 }
1394 
1395 static void unmap_internal_clock(struct mlx4_dev *dev)
1396 {
1397 	struct mlx4_priv *priv = mlx4_priv(dev);
1398 
1399 	if (priv->clock_mapping)
1400 		iounmap(priv->clock_mapping);
1401 }
1402 
1403 static void mlx4_close_hca(struct mlx4_dev *dev)
1404 {
1405 	unmap_internal_clock(dev);
1406 	unmap_bf_area(dev);
1407 	if (mlx4_is_slave(dev))
1408 		mlx4_slave_exit(dev);
1409 	else {
1410 		mlx4_CLOSE_HCA(dev, 0);
1411 		mlx4_free_icms(dev);
1412 		mlx4_UNMAP_FA(dev);
1413 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1414 	}
1415 }
1416 
1417 static int mlx4_init_slave(struct mlx4_dev *dev)
1418 {
1419 	struct mlx4_priv *priv = mlx4_priv(dev);
1420 	u64 dma = (u64) priv->mfunc.vhcr_dma;
1421 	int ret_from_reset = 0;
1422 	u32 slave_read;
1423 	u32 cmd_channel_ver;
1424 
1425 	if (atomic_read(&pf_loading)) {
1426 		mlx4_warn(dev, "PF is not ready - Deferring probe\n");
1427 		return -EPROBE_DEFER;
1428 	}
1429 
1430 	mutex_lock(&priv->cmd.slave_cmd_mutex);
1431 	priv->cmd.max_cmds = 1;
1432 	mlx4_warn(dev, "Sending reset\n");
1433 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1434 				       MLX4_COMM_TIME);
1435 	/* if we are in the middle of flr the slave will try
1436 	 * NUM_OF_RESET_RETRIES times before leaving.*/
1437 	if (ret_from_reset) {
1438 		if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1439 			mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
1440 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
1441 			return -EPROBE_DEFER;
1442 		} else
1443 			goto err;
1444 	}
1445 
1446 	/* check the driver version - the slave I/F revision
1447 	 * must match the master's */
1448 	slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1449 	cmd_channel_ver = mlx4_comm_get_version();
1450 
1451 	if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1452 		MLX4_COMM_GET_IF_REV(slave_read)) {
1453 		mlx4_err(dev, "slave driver version is not supported by the master\n");
1454 		goto err;
1455 	}
1456 
1457 	mlx4_warn(dev, "Sending vhcr0\n");
1458 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1459 						    MLX4_COMM_TIME))
1460 		goto err;
1461 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1462 						    MLX4_COMM_TIME))
1463 		goto err;
1464 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1465 						    MLX4_COMM_TIME))
1466 		goto err;
1467 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma, MLX4_COMM_TIME))
1468 		goto err;
1469 
1470 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1471 	return 0;
1472 
1473 err:
1474 	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, 0);
1475 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
1476 	return -EIO;
1477 }
1478 
1479 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1480 {
1481 	int i;
1482 
1483 	for (i = 1; i <= dev->caps.num_ports; i++) {
1484 		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
1485 			dev->caps.gid_table_len[i] =
1486 				mlx4_get_slave_num_gids(dev, 0, i);
1487 		else
1488 			dev->caps.gid_table_len[i] = 1;
1489 		dev->caps.pkey_table_len[i] =
1490 			dev->phys_caps.pkey_phys_table_len[i] - 1;
1491 	}
1492 }
1493 
1494 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
1495 {
1496 	int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
1497 
1498 	for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
1499 	      i++) {
1500 		if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
1501 			break;
1502 	}
1503 
1504 	return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
1505 }
1506 
1507 static void choose_steering_mode(struct mlx4_dev *dev,
1508 				 struct mlx4_dev_cap *dev_cap)
1509 {
1510 	if (mlx4_log_num_mgm_entry_size == -1 &&
1511 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
1512 	    (!mlx4_is_mfunc(dev) ||
1513 	     (dev_cap->fs_max_num_qp_per_entry >= (dev->num_vfs + 1))) &&
1514 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
1515 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
1516 		dev->oper_log_mgm_entry_size =
1517 			choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
1518 		dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1519 		dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
1520 		dev->caps.fs_log_max_ucast_qp_range_size =
1521 			dev_cap->fs_log_max_ucast_qp_range_size;
1522 	} else {
1523 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
1524 		    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1525 			dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
1526 		else {
1527 			dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
1528 
1529 			if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
1530 			    dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1531 				mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
1532 		}
1533 		dev->oper_log_mgm_entry_size =
1534 			mlx4_log_num_mgm_entry_size > 0 ?
1535 			mlx4_log_num_mgm_entry_size :
1536 			MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
1537 		dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
1538 	}
1539 	mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
1540 		 mlx4_steering_mode_str(dev->caps.steering_mode),
1541 		 dev->oper_log_mgm_entry_size,
1542 		 mlx4_log_num_mgm_entry_size);
1543 }
1544 
1545 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
1546 				       struct mlx4_dev_cap *dev_cap)
1547 {
1548 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
1549 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
1550 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
1551 	else
1552 		dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
1553 
1554 	mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
1555 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
1556 }
1557 
1558 static int mlx4_init_hca(struct mlx4_dev *dev)
1559 {
1560 	struct mlx4_priv	  *priv = mlx4_priv(dev);
1561 	struct mlx4_adapter	   adapter;
1562 	struct mlx4_dev_cap	   dev_cap;
1563 	struct mlx4_mod_stat_cfg   mlx4_cfg;
1564 	struct mlx4_profile	   profile;
1565 	struct mlx4_init_hca_param init_hca;
1566 	u64 icm_size;
1567 	int err;
1568 
1569 	if (!mlx4_is_slave(dev)) {
1570 		err = mlx4_QUERY_FW(dev);
1571 		if (err) {
1572 			if (err == -EACCES)
1573 				mlx4_info(dev, "non-primary physical function, skipping\n");
1574 			else
1575 				mlx4_err(dev, "QUERY_FW command failed, aborting\n");
1576 			return err;
1577 		}
1578 
1579 		err = mlx4_load_fw(dev);
1580 		if (err) {
1581 			mlx4_err(dev, "Failed to start FW, aborting\n");
1582 			return err;
1583 		}
1584 
1585 		mlx4_cfg.log_pg_sz_m = 1;
1586 		mlx4_cfg.log_pg_sz = 0;
1587 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
1588 		if (err)
1589 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
1590 
1591 		err = mlx4_dev_cap(dev, &dev_cap);
1592 		if (err) {
1593 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
1594 			goto err_stop_fw;
1595 		}
1596 
1597 		choose_steering_mode(dev, &dev_cap);
1598 		choose_tunnel_offload_mode(dev, &dev_cap);
1599 
1600 		err = mlx4_get_phys_port_id(dev);
1601 		if (err)
1602 			mlx4_err(dev, "Fail to get physical port id\n");
1603 
1604 		if (mlx4_is_master(dev))
1605 			mlx4_parav_master_pf_caps(dev);
1606 
1607 		if (mlx4_low_memory_profile()) {
1608 			mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
1609 			profile = low_mem_profile;
1610 		} else {
1611 			profile = default_profile;
1612 		}
1613 		if (dev->caps.steering_mode ==
1614 		    MLX4_STEERING_MODE_DEVICE_MANAGED)
1615 			profile.num_mcg = MLX4_FS_NUM_MCG;
1616 
1617 		icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
1618 					     &init_hca);
1619 		if ((long long) icm_size < 0) {
1620 			err = icm_size;
1621 			goto err_stop_fw;
1622 		}
1623 
1624 		dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
1625 
1626 		init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
1627 		init_hca.uar_page_sz = PAGE_SHIFT - 12;
1628 		init_hca.mw_enabled = 0;
1629 		if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1630 		    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
1631 			init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
1632 
1633 		err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
1634 		if (err)
1635 			goto err_stop_fw;
1636 
1637 		err = mlx4_INIT_HCA(dev, &init_hca);
1638 		if (err) {
1639 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
1640 			goto err_free_icm;
1641 		}
1642 		/*
1643 		 * If TS is supported by FW
1644 		 * read HCA frequency by QUERY_HCA command
1645 		 */
1646 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
1647 			memset(&init_hca, 0, sizeof(init_hca));
1648 			err = mlx4_QUERY_HCA(dev, &init_hca);
1649 			if (err) {
1650 				mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
1651 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1652 			} else {
1653 				dev->caps.hca_core_clock =
1654 					init_hca.hca_core_clock;
1655 			}
1656 
1657 			/* In case we got HCA frequency 0 - disable timestamping
1658 			 * to avoid dividing by zero
1659 			 */
1660 			if (!dev->caps.hca_core_clock) {
1661 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1662 				mlx4_err(dev,
1663 					 "HCA frequency is 0 - timestamping is not supported\n");
1664 			} else if (map_internal_clock(dev)) {
1665 				/*
1666 				 * Map internal clock,
1667 				 * in case of failure disable timestamping
1668 				 */
1669 				dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1670 				mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
1671 			}
1672 		}
1673 	} else {
1674 		err = mlx4_init_slave(dev);
1675 		if (err) {
1676 			if (err != -EPROBE_DEFER)
1677 				mlx4_err(dev, "Failed to initialize slave\n");
1678 			return err;
1679 		}
1680 
1681 		err = mlx4_slave_cap(dev);
1682 		if (err) {
1683 			mlx4_err(dev, "Failed to obtain slave caps\n");
1684 			goto err_close;
1685 		}
1686 	}
1687 
1688 	if (map_bf_area(dev))
1689 		mlx4_dbg(dev, "Failed to map blue flame area\n");
1690 
1691 	/*Only the master set the ports, all the rest got it from it.*/
1692 	if (!mlx4_is_slave(dev))
1693 		mlx4_set_port_mask(dev);
1694 
1695 	err = mlx4_QUERY_ADAPTER(dev, &adapter);
1696 	if (err) {
1697 		mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
1698 		goto unmap_bf;
1699 	}
1700 
1701 	priv->eq_table.inta_pin = adapter.inta_pin;
1702 	memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
1703 
1704 	return 0;
1705 
1706 unmap_bf:
1707 	unmap_internal_clock(dev);
1708 	unmap_bf_area(dev);
1709 
1710 	if (mlx4_is_slave(dev)) {
1711 		kfree(dev->caps.qp0_qkey);
1712 		kfree(dev->caps.qp0_tunnel);
1713 		kfree(dev->caps.qp0_proxy);
1714 		kfree(dev->caps.qp1_tunnel);
1715 		kfree(dev->caps.qp1_proxy);
1716 	}
1717 
1718 err_close:
1719 	if (mlx4_is_slave(dev))
1720 		mlx4_slave_exit(dev);
1721 	else
1722 		mlx4_CLOSE_HCA(dev, 0);
1723 
1724 err_free_icm:
1725 	if (!mlx4_is_slave(dev))
1726 		mlx4_free_icms(dev);
1727 
1728 err_stop_fw:
1729 	if (!mlx4_is_slave(dev)) {
1730 		mlx4_UNMAP_FA(dev);
1731 		mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1732 	}
1733 	return err;
1734 }
1735 
1736 static int mlx4_init_counters_table(struct mlx4_dev *dev)
1737 {
1738 	struct mlx4_priv *priv = mlx4_priv(dev);
1739 	int nent;
1740 
1741 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1742 		return -ENOENT;
1743 
1744 	nent = dev->caps.max_counters;
1745 	return mlx4_bitmap_init(&priv->counters_bitmap, nent, nent - 1, 0, 0);
1746 }
1747 
1748 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
1749 {
1750 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
1751 }
1752 
1753 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1754 {
1755 	struct mlx4_priv *priv = mlx4_priv(dev);
1756 
1757 	if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1758 		return -ENOENT;
1759 
1760 	*idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
1761 	if (*idx == -1)
1762 		return -ENOMEM;
1763 
1764 	return 0;
1765 }
1766 
1767 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1768 {
1769 	u64 out_param;
1770 	int err;
1771 
1772 	if (mlx4_is_mfunc(dev)) {
1773 		err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
1774 				   RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
1775 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
1776 		if (!err)
1777 			*idx = get_param_l(&out_param);
1778 
1779 		return err;
1780 	}
1781 	return __mlx4_counter_alloc(dev, idx);
1782 }
1783 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
1784 
1785 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1786 {
1787 	mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
1788 	return;
1789 }
1790 
1791 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1792 {
1793 	u64 in_param = 0;
1794 
1795 	if (mlx4_is_mfunc(dev)) {
1796 		set_param_l(&in_param, idx);
1797 		mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
1798 			 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
1799 			 MLX4_CMD_WRAPPED);
1800 		return;
1801 	}
1802 	__mlx4_counter_free(dev, idx);
1803 }
1804 EXPORT_SYMBOL_GPL(mlx4_counter_free);
1805 
1806 static int mlx4_setup_hca(struct mlx4_dev *dev)
1807 {
1808 	struct mlx4_priv *priv = mlx4_priv(dev);
1809 	int err;
1810 	int port;
1811 	__be32 ib_port_default_caps;
1812 
1813 	err = mlx4_init_uar_table(dev);
1814 	if (err) {
1815 		mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
1816 		 return err;
1817 	}
1818 
1819 	err = mlx4_uar_alloc(dev, &priv->driver_uar);
1820 	if (err) {
1821 		mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
1822 		goto err_uar_table_free;
1823 	}
1824 
1825 	priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
1826 	if (!priv->kar) {
1827 		mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
1828 		err = -ENOMEM;
1829 		goto err_uar_free;
1830 	}
1831 
1832 	err = mlx4_init_pd_table(dev);
1833 	if (err) {
1834 		mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
1835 		goto err_kar_unmap;
1836 	}
1837 
1838 	err = mlx4_init_xrcd_table(dev);
1839 	if (err) {
1840 		mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
1841 		goto err_pd_table_free;
1842 	}
1843 
1844 	err = mlx4_init_mr_table(dev);
1845 	if (err) {
1846 		mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
1847 		goto err_xrcd_table_free;
1848 	}
1849 
1850 	if (!mlx4_is_slave(dev)) {
1851 		err = mlx4_init_mcg_table(dev);
1852 		if (err) {
1853 			mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
1854 			goto err_mr_table_free;
1855 		}
1856 		err = mlx4_config_mad_demux(dev);
1857 		if (err) {
1858 			mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
1859 			goto err_mcg_table_free;
1860 		}
1861 	}
1862 
1863 	err = mlx4_init_eq_table(dev);
1864 	if (err) {
1865 		mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
1866 		goto err_mcg_table_free;
1867 	}
1868 
1869 	err = mlx4_cmd_use_events(dev);
1870 	if (err) {
1871 		mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
1872 		goto err_eq_table_free;
1873 	}
1874 
1875 	err = mlx4_NOP(dev);
1876 	if (err) {
1877 		if (dev->flags & MLX4_FLAG_MSI_X) {
1878 			mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
1879 				  priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1880 			mlx4_warn(dev, "Trying again without MSI-X\n");
1881 		} else {
1882 			mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
1883 				 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1884 			mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
1885 		}
1886 
1887 		goto err_cmd_poll;
1888 	}
1889 
1890 	mlx4_dbg(dev, "NOP command IRQ test passed\n");
1891 
1892 	err = mlx4_init_cq_table(dev);
1893 	if (err) {
1894 		mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
1895 		goto err_cmd_poll;
1896 	}
1897 
1898 	err = mlx4_init_srq_table(dev);
1899 	if (err) {
1900 		mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
1901 		goto err_cq_table_free;
1902 	}
1903 
1904 	err = mlx4_init_qp_table(dev);
1905 	if (err) {
1906 		mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
1907 		goto err_srq_table_free;
1908 	}
1909 
1910 	err = mlx4_init_counters_table(dev);
1911 	if (err && err != -ENOENT) {
1912 		mlx4_err(dev, "Failed to initialize counters table, aborting\n");
1913 		goto err_qp_table_free;
1914 	}
1915 
1916 	if (!mlx4_is_slave(dev)) {
1917 		for (port = 1; port <= dev->caps.num_ports; port++) {
1918 			ib_port_default_caps = 0;
1919 			err = mlx4_get_port_ib_caps(dev, port,
1920 						    &ib_port_default_caps);
1921 			if (err)
1922 				mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
1923 					  port, err);
1924 			dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
1925 
1926 			/* initialize per-slave default ib port capabilities */
1927 			if (mlx4_is_master(dev)) {
1928 				int i;
1929 				for (i = 0; i < dev->num_slaves; i++) {
1930 					if (i == mlx4_master_func_num(dev))
1931 						continue;
1932 					priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
1933 						ib_port_default_caps;
1934 				}
1935 			}
1936 
1937 			if (mlx4_is_mfunc(dev))
1938 				dev->caps.port_ib_mtu[port] = IB_MTU_2048;
1939 			else
1940 				dev->caps.port_ib_mtu[port] = IB_MTU_4096;
1941 
1942 			err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
1943 					    dev->caps.pkey_table_len[port] : -1);
1944 			if (err) {
1945 				mlx4_err(dev, "Failed to set port %d, aborting\n",
1946 					 port);
1947 				goto err_counters_table_free;
1948 			}
1949 		}
1950 	}
1951 
1952 	return 0;
1953 
1954 err_counters_table_free:
1955 	mlx4_cleanup_counters_table(dev);
1956 
1957 err_qp_table_free:
1958 	mlx4_cleanup_qp_table(dev);
1959 
1960 err_srq_table_free:
1961 	mlx4_cleanup_srq_table(dev);
1962 
1963 err_cq_table_free:
1964 	mlx4_cleanup_cq_table(dev);
1965 
1966 err_cmd_poll:
1967 	mlx4_cmd_use_polling(dev);
1968 
1969 err_eq_table_free:
1970 	mlx4_cleanup_eq_table(dev);
1971 
1972 err_mcg_table_free:
1973 	if (!mlx4_is_slave(dev))
1974 		mlx4_cleanup_mcg_table(dev);
1975 
1976 err_mr_table_free:
1977 	mlx4_cleanup_mr_table(dev);
1978 
1979 err_xrcd_table_free:
1980 	mlx4_cleanup_xrcd_table(dev);
1981 
1982 err_pd_table_free:
1983 	mlx4_cleanup_pd_table(dev);
1984 
1985 err_kar_unmap:
1986 	iounmap(priv->kar);
1987 
1988 err_uar_free:
1989 	mlx4_uar_free(dev, &priv->driver_uar);
1990 
1991 err_uar_table_free:
1992 	mlx4_cleanup_uar_table(dev);
1993 	return err;
1994 }
1995 
1996 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
1997 {
1998 	struct mlx4_priv *priv = mlx4_priv(dev);
1999 	struct msix_entry *entries;
2000 	int nreq = min_t(int, dev->caps.num_ports *
2001 			 min_t(int, num_online_cpus() + 1,
2002 			       MAX_MSIX_P_PORT) + MSIX_LEGACY_SZ, MAX_MSIX);
2003 	int i;
2004 
2005 	if (msi_x) {
2006 		nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
2007 			     nreq);
2008 
2009 		entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
2010 		if (!entries)
2011 			goto no_msi;
2012 
2013 		for (i = 0; i < nreq; ++i)
2014 			entries[i].entry = i;
2015 
2016 		nreq = pci_enable_msix_range(dev->pdev, entries, 2, nreq);
2017 
2018 		if (nreq < 0) {
2019 			kfree(entries);
2020 			goto no_msi;
2021 		} else if (nreq < MSIX_LEGACY_SZ +
2022 			   dev->caps.num_ports * MIN_MSIX_P_PORT) {
2023 			/*Working in legacy mode , all EQ's shared*/
2024 			dev->caps.comp_pool           = 0;
2025 			dev->caps.num_comp_vectors = nreq - 1;
2026 		} else {
2027 			dev->caps.comp_pool           = nreq - MSIX_LEGACY_SZ;
2028 			dev->caps.num_comp_vectors = MSIX_LEGACY_SZ - 1;
2029 		}
2030 		for (i = 0; i < nreq; ++i)
2031 			priv->eq_table.eq[i].irq = entries[i].vector;
2032 
2033 		dev->flags |= MLX4_FLAG_MSI_X;
2034 
2035 		kfree(entries);
2036 		return;
2037 	}
2038 
2039 no_msi:
2040 	dev->caps.num_comp_vectors = 1;
2041 	dev->caps.comp_pool	   = 0;
2042 
2043 	for (i = 0; i < 2; ++i)
2044 		priv->eq_table.eq[i].irq = dev->pdev->irq;
2045 }
2046 
2047 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2048 {
2049 	struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2050 	int err = 0;
2051 
2052 	info->dev = dev;
2053 	info->port = port;
2054 	if (!mlx4_is_slave(dev)) {
2055 		mlx4_init_mac_table(dev, &info->mac_table);
2056 		mlx4_init_vlan_table(dev, &info->vlan_table);
2057 		mlx4_init_roce_gid_table(dev, &info->gid_table);
2058 		info->base_qpn = mlx4_get_base_qpn(dev, port);
2059 	}
2060 
2061 	sprintf(info->dev_name, "mlx4_port%d", port);
2062 	info->port_attr.attr.name = info->dev_name;
2063 	if (mlx4_is_mfunc(dev))
2064 		info->port_attr.attr.mode = S_IRUGO;
2065 	else {
2066 		info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2067 		info->port_attr.store     = set_port_type;
2068 	}
2069 	info->port_attr.show      = show_port_type;
2070 	sysfs_attr_init(&info->port_attr.attr);
2071 
2072 	err = device_create_file(&dev->pdev->dev, &info->port_attr);
2073 	if (err) {
2074 		mlx4_err(dev, "Failed to create file for port %d\n", port);
2075 		info->port = -1;
2076 	}
2077 
2078 	sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2079 	info->port_mtu_attr.attr.name = info->dev_mtu_name;
2080 	if (mlx4_is_mfunc(dev))
2081 		info->port_mtu_attr.attr.mode = S_IRUGO;
2082 	else {
2083 		info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2084 		info->port_mtu_attr.store     = set_port_ib_mtu;
2085 	}
2086 	info->port_mtu_attr.show      = show_port_ib_mtu;
2087 	sysfs_attr_init(&info->port_mtu_attr.attr);
2088 
2089 	err = device_create_file(&dev->pdev->dev, &info->port_mtu_attr);
2090 	if (err) {
2091 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
2092 		device_remove_file(&info->dev->pdev->dev, &info->port_attr);
2093 		info->port = -1;
2094 	}
2095 
2096 	return err;
2097 }
2098 
2099 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
2100 {
2101 	if (info->port < 0)
2102 		return;
2103 
2104 	device_remove_file(&info->dev->pdev->dev, &info->port_attr);
2105 	device_remove_file(&info->dev->pdev->dev, &info->port_mtu_attr);
2106 }
2107 
2108 static int mlx4_init_steering(struct mlx4_dev *dev)
2109 {
2110 	struct mlx4_priv *priv = mlx4_priv(dev);
2111 	int num_entries = dev->caps.num_ports;
2112 	int i, j;
2113 
2114 	priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
2115 	if (!priv->steer)
2116 		return -ENOMEM;
2117 
2118 	for (i = 0; i < num_entries; i++)
2119 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
2120 			INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
2121 			INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
2122 		}
2123 	return 0;
2124 }
2125 
2126 static void mlx4_clear_steering(struct mlx4_dev *dev)
2127 {
2128 	struct mlx4_priv *priv = mlx4_priv(dev);
2129 	struct mlx4_steer_index *entry, *tmp_entry;
2130 	struct mlx4_promisc_qp *pqp, *tmp_pqp;
2131 	int num_entries = dev->caps.num_ports;
2132 	int i, j;
2133 
2134 	for (i = 0; i < num_entries; i++) {
2135 		for (j = 0; j < MLX4_NUM_STEERS; j++) {
2136 			list_for_each_entry_safe(pqp, tmp_pqp,
2137 						 &priv->steer[i].promisc_qps[j],
2138 						 list) {
2139 				list_del(&pqp->list);
2140 				kfree(pqp);
2141 			}
2142 			list_for_each_entry_safe(entry, tmp_entry,
2143 						 &priv->steer[i].steer_entries[j],
2144 						 list) {
2145 				list_del(&entry->list);
2146 				list_for_each_entry_safe(pqp, tmp_pqp,
2147 							 &entry->duplicates,
2148 							 list) {
2149 					list_del(&pqp->list);
2150 					kfree(pqp);
2151 				}
2152 				kfree(entry);
2153 			}
2154 		}
2155 	}
2156 	kfree(priv->steer);
2157 }
2158 
2159 static int extended_func_num(struct pci_dev *pdev)
2160 {
2161 	return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
2162 }
2163 
2164 #define MLX4_OWNER_BASE	0x8069c
2165 #define MLX4_OWNER_SIZE	4
2166 
2167 static int mlx4_get_ownership(struct mlx4_dev *dev)
2168 {
2169 	void __iomem *owner;
2170 	u32 ret;
2171 
2172 	if (pci_channel_offline(dev->pdev))
2173 		return -EIO;
2174 
2175 	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
2176 			MLX4_OWNER_SIZE);
2177 	if (!owner) {
2178 		mlx4_err(dev, "Failed to obtain ownership bit\n");
2179 		return -ENOMEM;
2180 	}
2181 
2182 	ret = readl(owner);
2183 	iounmap(owner);
2184 	return (int) !!ret;
2185 }
2186 
2187 static void mlx4_free_ownership(struct mlx4_dev *dev)
2188 {
2189 	void __iomem *owner;
2190 
2191 	if (pci_channel_offline(dev->pdev))
2192 		return;
2193 
2194 	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
2195 			MLX4_OWNER_SIZE);
2196 	if (!owner) {
2197 		mlx4_err(dev, "Failed to obtain ownership bit\n");
2198 		return;
2199 	}
2200 	writel(0, owner);
2201 	msleep(1000);
2202 	iounmap(owner);
2203 }
2204 
2205 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data)
2206 {
2207 	struct mlx4_priv *priv;
2208 	struct mlx4_dev *dev;
2209 	int err;
2210 	int port;
2211 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
2212 	int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
2213 	const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
2214 		{2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
2215 	unsigned total_vfs = 0;
2216 	int sriov_initialized = 0;
2217 	unsigned int i;
2218 
2219 	pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
2220 
2221 	err = pci_enable_device(pdev);
2222 	if (err) {
2223 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
2224 		return err;
2225 	}
2226 
2227 	/* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
2228 	 * per port, we must limit the number of VFs to 63 (since their are
2229 	 * 128 MACs)
2230 	 */
2231 	for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
2232 	     total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
2233 		nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
2234 		if (nvfs[i] < 0) {
2235 			dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
2236 			return -EINVAL;
2237 		}
2238 	}
2239 	for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
2240 	     i++) {
2241 		prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
2242 		if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
2243 			dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
2244 			return -EINVAL;
2245 		}
2246 	}
2247 	if (total_vfs >= MLX4_MAX_NUM_VF) {
2248 		dev_err(&pdev->dev,
2249 			"Requested more VF's (%d) than allowed (%d)\n",
2250 			total_vfs, MLX4_MAX_NUM_VF - 1);
2251 		return -EINVAL;
2252 	}
2253 
2254 	for (i = 0; i < MLX4_MAX_PORTS; i++) {
2255 		if (nvfs[i] + nvfs[2] >= MLX4_MAX_NUM_VF_P_PORT) {
2256 			dev_err(&pdev->dev,
2257 				"Requested more VF's (%d) for port (%d) than allowed (%d)\n",
2258 				nvfs[i] + nvfs[2], i + 1,
2259 				MLX4_MAX_NUM_VF_P_PORT - 1);
2260 			return -EINVAL;
2261 		}
2262 	}
2263 
2264 
2265 	/*
2266 	 * Check for BARs.
2267 	 */
2268 	if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
2269 	    !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
2270 		dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
2271 			pci_dev_data, pci_resource_flags(pdev, 0));
2272 		err = -ENODEV;
2273 		goto err_disable_pdev;
2274 	}
2275 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
2276 		dev_err(&pdev->dev, "Missing UAR, aborting\n");
2277 		err = -ENODEV;
2278 		goto err_disable_pdev;
2279 	}
2280 
2281 	err = pci_request_regions(pdev, DRV_NAME);
2282 	if (err) {
2283 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
2284 		goto err_disable_pdev;
2285 	}
2286 
2287 	pci_set_master(pdev);
2288 
2289 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2290 	if (err) {
2291 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
2292 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2293 		if (err) {
2294 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
2295 			goto err_release_regions;
2296 		}
2297 	}
2298 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2299 	if (err) {
2300 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
2301 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2302 		if (err) {
2303 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
2304 			goto err_release_regions;
2305 		}
2306 	}
2307 
2308 	/* Allow large DMA segments, up to the firmware limit of 1 GB */
2309 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
2310 
2311 	dev       = pci_get_drvdata(pdev);
2312 	priv      = mlx4_priv(dev);
2313 	dev->pdev = pdev;
2314 	INIT_LIST_HEAD(&priv->ctx_list);
2315 	spin_lock_init(&priv->ctx_lock);
2316 
2317 	mutex_init(&priv->port_mutex);
2318 
2319 	INIT_LIST_HEAD(&priv->pgdir_list);
2320 	mutex_init(&priv->pgdir_mutex);
2321 
2322 	INIT_LIST_HEAD(&priv->bf_list);
2323 	mutex_init(&priv->bf_mutex);
2324 
2325 	dev->rev_id = pdev->revision;
2326 	dev->numa_node = dev_to_node(&pdev->dev);
2327 	/* Detect if this device is a virtual function */
2328 	if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
2329 		/* When acting as pf, we normally skip vfs unless explicitly
2330 		 * requested to probe them. */
2331 		if (total_vfs) {
2332 			unsigned vfs_offset = 0;
2333 			for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
2334 				     vfs_offset + nvfs[i] < extended_func_num(pdev);
2335 			     vfs_offset += nvfs[i], i++)
2336 				;
2337 			if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
2338 				err = -ENODEV;
2339 				goto err_free_dev;
2340 			}
2341 			if ((extended_func_num(pdev) - vfs_offset)
2342 			    > prb_vf[i]) {
2343 				mlx4_warn(dev, "Skipping virtual function:%d\n",
2344 					  extended_func_num(pdev));
2345 				err = -ENODEV;
2346 				goto err_free_dev;
2347 			}
2348 		}
2349 		mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
2350 		dev->flags |= MLX4_FLAG_SLAVE;
2351 	} else {
2352 		/* We reset the device and enable SRIOV only for physical
2353 		 * devices.  Try to claim ownership on the device;
2354 		 * if already taken, skip -- do not allow multiple PFs */
2355 		err = mlx4_get_ownership(dev);
2356 		if (err) {
2357 			if (err < 0)
2358 				goto err_free_dev;
2359 			else {
2360 				mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
2361 				err = -EINVAL;
2362 				goto err_free_dev;
2363 			}
2364 		}
2365 
2366 		if (total_vfs) {
2367 			mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n",
2368 				  total_vfs);
2369 			dev->dev_vfs = kzalloc(
2370 				total_vfs * sizeof(*dev->dev_vfs),
2371 				GFP_KERNEL);
2372 			if (NULL == dev->dev_vfs) {
2373 				mlx4_err(dev, "Failed to allocate memory for VFs\n");
2374 				err = 0;
2375 			} else {
2376 				atomic_inc(&pf_loading);
2377 				err = pci_enable_sriov(pdev, total_vfs);
2378 				if (err) {
2379 					mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
2380 						 err);
2381 					atomic_dec(&pf_loading);
2382 					err = 0;
2383 				} else {
2384 					mlx4_warn(dev, "Running in master mode\n");
2385 					dev->flags |= MLX4_FLAG_SRIOV |
2386 						MLX4_FLAG_MASTER;
2387 					dev->num_vfs = total_vfs;
2388 					sriov_initialized = 1;
2389 				}
2390 			}
2391 		}
2392 
2393 		atomic_set(&priv->opreq_count, 0);
2394 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
2395 
2396 		/*
2397 		 * Now reset the HCA before we touch the PCI capabilities or
2398 		 * attempt a firmware command, since a boot ROM may have left
2399 		 * the HCA in an undefined state.
2400 		 */
2401 		err = mlx4_reset(dev);
2402 		if (err) {
2403 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
2404 			goto err_rel_own;
2405 		}
2406 	}
2407 
2408 slave_start:
2409 	err = mlx4_cmd_init(dev);
2410 	if (err) {
2411 		mlx4_err(dev, "Failed to init command interface, aborting\n");
2412 		goto err_sriov;
2413 	}
2414 
2415 	/* In slave functions, the communication channel must be initialized
2416 	 * before posting commands. Also, init num_slaves before calling
2417 	 * mlx4_init_hca */
2418 	if (mlx4_is_mfunc(dev)) {
2419 		if (mlx4_is_master(dev))
2420 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
2421 		else {
2422 			dev->num_slaves = 0;
2423 			err = mlx4_multi_func_init(dev);
2424 			if (err) {
2425 				mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
2426 				goto err_cmd;
2427 			}
2428 		}
2429 	}
2430 
2431 	err = mlx4_init_hca(dev);
2432 	if (err) {
2433 		if (err == -EACCES) {
2434 			/* Not primary Physical function
2435 			 * Running in slave mode */
2436 			mlx4_cmd_cleanup(dev);
2437 			dev->flags |= MLX4_FLAG_SLAVE;
2438 			dev->flags &= ~MLX4_FLAG_MASTER;
2439 			goto slave_start;
2440 		} else
2441 			goto err_mfunc;
2442 	}
2443 
2444 	/* check if the device is functioning at its maximum possible speed.
2445 	 * No return code for this call, just warn the user in case of PCI
2446 	 * express device capabilities are under-satisfied by the bus.
2447 	 */
2448 	if (!mlx4_is_slave(dev))
2449 		mlx4_check_pcie_caps(dev);
2450 
2451 	/* In master functions, the communication channel must be initialized
2452 	 * after obtaining its address from fw */
2453 	if (mlx4_is_master(dev)) {
2454 		unsigned sum = 0;
2455 		err = mlx4_multi_func_init(dev);
2456 		if (err) {
2457 			mlx4_err(dev, "Failed to init master mfunc interface, aborting\n");
2458 			goto err_close;
2459 		}
2460 		if (sriov_initialized) {
2461 			int ib_ports = 0;
2462 			mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2463 				ib_ports++;
2464 
2465 			if (ib_ports &&
2466 			    (num_vfs_argc > 1 || probe_vfs_argc > 1)) {
2467 				mlx4_err(dev,
2468 					 "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");
2469 				err = -EINVAL;
2470 				goto err_master_mfunc;
2471 			}
2472 			for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]); i++) {
2473 				unsigned j;
2474 				for (j = 0; j < nvfs[i]; ++sum, ++j) {
2475 					dev->dev_vfs[sum].min_port =
2476 						i < 2 ? i + 1 : 1;
2477 					dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
2478 						dev->caps.num_ports;
2479 				}
2480 			}
2481 		}
2482 	}
2483 
2484 	err = mlx4_alloc_eq_table(dev);
2485 	if (err)
2486 		goto err_master_mfunc;
2487 
2488 	priv->msix_ctl.pool_bm = 0;
2489 	mutex_init(&priv->msix_ctl.pool_lock);
2490 
2491 	mlx4_enable_msi_x(dev);
2492 	if ((mlx4_is_mfunc(dev)) &&
2493 	    !(dev->flags & MLX4_FLAG_MSI_X)) {
2494 		err = -ENOSYS;
2495 		mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
2496 		goto err_free_eq;
2497 	}
2498 
2499 	if (!mlx4_is_slave(dev)) {
2500 		err = mlx4_init_steering(dev);
2501 		if (err)
2502 			goto err_free_eq;
2503 	}
2504 
2505 	err = mlx4_setup_hca(dev);
2506 	if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
2507 	    !mlx4_is_mfunc(dev)) {
2508 		dev->flags &= ~MLX4_FLAG_MSI_X;
2509 		dev->caps.num_comp_vectors = 1;
2510 		dev->caps.comp_pool	   = 0;
2511 		pci_disable_msix(pdev);
2512 		err = mlx4_setup_hca(dev);
2513 	}
2514 
2515 	if (err)
2516 		goto err_steer;
2517 
2518 	mlx4_init_quotas(dev);
2519 
2520 	for (port = 1; port <= dev->caps.num_ports; port++) {
2521 		err = mlx4_init_port_info(dev, port);
2522 		if (err)
2523 			goto err_port;
2524 	}
2525 
2526 	err = mlx4_register_device(dev);
2527 	if (err)
2528 		goto err_port;
2529 
2530 	mlx4_request_modules(dev);
2531 
2532 	mlx4_sense_init(dev);
2533 	mlx4_start_sense(dev);
2534 
2535 	priv->removed = 0;
2536 
2537 	if (mlx4_is_master(dev) && dev->num_vfs)
2538 		atomic_dec(&pf_loading);
2539 
2540 	return 0;
2541 
2542 err_port:
2543 	for (--port; port >= 1; --port)
2544 		mlx4_cleanup_port_info(&priv->port[port]);
2545 
2546 	mlx4_cleanup_counters_table(dev);
2547 	mlx4_cleanup_qp_table(dev);
2548 	mlx4_cleanup_srq_table(dev);
2549 	mlx4_cleanup_cq_table(dev);
2550 	mlx4_cmd_use_polling(dev);
2551 	mlx4_cleanup_eq_table(dev);
2552 	mlx4_cleanup_mcg_table(dev);
2553 	mlx4_cleanup_mr_table(dev);
2554 	mlx4_cleanup_xrcd_table(dev);
2555 	mlx4_cleanup_pd_table(dev);
2556 	mlx4_cleanup_uar_table(dev);
2557 
2558 err_steer:
2559 	if (!mlx4_is_slave(dev))
2560 		mlx4_clear_steering(dev);
2561 
2562 err_free_eq:
2563 	mlx4_free_eq_table(dev);
2564 
2565 err_master_mfunc:
2566 	if (mlx4_is_master(dev))
2567 		mlx4_multi_func_cleanup(dev);
2568 
2569 	if (mlx4_is_slave(dev)) {
2570 		kfree(dev->caps.qp0_qkey);
2571 		kfree(dev->caps.qp0_tunnel);
2572 		kfree(dev->caps.qp0_proxy);
2573 		kfree(dev->caps.qp1_tunnel);
2574 		kfree(dev->caps.qp1_proxy);
2575 	}
2576 
2577 err_close:
2578 	if (dev->flags & MLX4_FLAG_MSI_X)
2579 		pci_disable_msix(pdev);
2580 
2581 	mlx4_close_hca(dev);
2582 
2583 err_mfunc:
2584 	if (mlx4_is_slave(dev))
2585 		mlx4_multi_func_cleanup(dev);
2586 
2587 err_cmd:
2588 	mlx4_cmd_cleanup(dev);
2589 
2590 err_sriov:
2591 	if (dev->flags & MLX4_FLAG_SRIOV)
2592 		pci_disable_sriov(pdev);
2593 
2594 err_rel_own:
2595 	if (!mlx4_is_slave(dev))
2596 		mlx4_free_ownership(dev);
2597 
2598 	if (mlx4_is_master(dev) && dev->num_vfs)
2599 		atomic_dec(&pf_loading);
2600 
2601 	kfree(priv->dev.dev_vfs);
2602 
2603 err_free_dev:
2604 	kfree(priv);
2605 
2606 err_release_regions:
2607 	pci_release_regions(pdev);
2608 
2609 err_disable_pdev:
2610 	pci_disable_device(pdev);
2611 	pci_set_drvdata(pdev, NULL);
2612 	return err;
2613 }
2614 
2615 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
2616 {
2617 	struct mlx4_priv *priv;
2618 	struct mlx4_dev *dev;
2619 
2620 	printk_once(KERN_INFO "%s", mlx4_version);
2621 
2622 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
2623 	if (!priv)
2624 		return -ENOMEM;
2625 
2626 	dev       = &priv->dev;
2627 	pci_set_drvdata(pdev, dev);
2628 	priv->pci_dev_data = id->driver_data;
2629 
2630 	return __mlx4_init_one(pdev, id->driver_data);
2631 }
2632 
2633 static void __mlx4_remove_one(struct pci_dev *pdev)
2634 {
2635 	struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
2636 	struct mlx4_priv *priv = mlx4_priv(dev);
2637 	int               pci_dev_data;
2638 	int p;
2639 
2640 	if (priv->removed)
2641 		return;
2642 
2643 	pci_dev_data = priv->pci_dev_data;
2644 
2645 	/* in SRIOV it is not allowed to unload the pf's
2646 	 * driver while there are alive vf's */
2647 	if (mlx4_is_master(dev) && mlx4_how_many_lives_vf(dev))
2648 		pr_warn("Removing PF when there are assigned VF's !!!\n");
2649 	mlx4_stop_sense(dev);
2650 	mlx4_unregister_device(dev);
2651 
2652 	for (p = 1; p <= dev->caps.num_ports; p++) {
2653 		mlx4_cleanup_port_info(&priv->port[p]);
2654 		mlx4_CLOSE_PORT(dev, p);
2655 	}
2656 
2657 	if (mlx4_is_master(dev))
2658 		mlx4_free_resource_tracker(dev,
2659 					   RES_TR_FREE_SLAVES_ONLY);
2660 
2661 	mlx4_cleanup_counters_table(dev);
2662 	mlx4_cleanup_qp_table(dev);
2663 	mlx4_cleanup_srq_table(dev);
2664 	mlx4_cleanup_cq_table(dev);
2665 	mlx4_cmd_use_polling(dev);
2666 	mlx4_cleanup_eq_table(dev);
2667 	mlx4_cleanup_mcg_table(dev);
2668 	mlx4_cleanup_mr_table(dev);
2669 	mlx4_cleanup_xrcd_table(dev);
2670 	mlx4_cleanup_pd_table(dev);
2671 
2672 	if (mlx4_is_master(dev))
2673 		mlx4_free_resource_tracker(dev,
2674 					   RES_TR_FREE_STRUCTS_ONLY);
2675 
2676 	iounmap(priv->kar);
2677 	mlx4_uar_free(dev, &priv->driver_uar);
2678 	mlx4_cleanup_uar_table(dev);
2679 	if (!mlx4_is_slave(dev))
2680 		mlx4_clear_steering(dev);
2681 	mlx4_free_eq_table(dev);
2682 	if (mlx4_is_master(dev))
2683 		mlx4_multi_func_cleanup(dev);
2684 	mlx4_close_hca(dev);
2685 	if (mlx4_is_slave(dev))
2686 		mlx4_multi_func_cleanup(dev);
2687 	mlx4_cmd_cleanup(dev);
2688 
2689 	if (dev->flags & MLX4_FLAG_MSI_X)
2690 		pci_disable_msix(pdev);
2691 	if (dev->flags & MLX4_FLAG_SRIOV) {
2692 		mlx4_warn(dev, "Disabling SR-IOV\n");
2693 		pci_disable_sriov(pdev);
2694 		dev->num_vfs = 0;
2695 	}
2696 
2697 	if (!mlx4_is_slave(dev))
2698 		mlx4_free_ownership(dev);
2699 
2700 	kfree(dev->caps.qp0_qkey);
2701 	kfree(dev->caps.qp0_tunnel);
2702 	kfree(dev->caps.qp0_proxy);
2703 	kfree(dev->caps.qp1_tunnel);
2704 	kfree(dev->caps.qp1_proxy);
2705 	kfree(dev->dev_vfs);
2706 
2707 	pci_release_regions(pdev);
2708 	pci_disable_device(pdev);
2709 	memset(priv, 0, sizeof(*priv));
2710 	priv->pci_dev_data = pci_dev_data;
2711 	priv->removed = 1;
2712 }
2713 
2714 static void mlx4_remove_one(struct pci_dev *pdev)
2715 {
2716 	struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
2717 	struct mlx4_priv *priv = mlx4_priv(dev);
2718 
2719 	__mlx4_remove_one(pdev);
2720 	kfree(priv);
2721 	pci_set_drvdata(pdev, NULL);
2722 }
2723 
2724 int mlx4_restart_one(struct pci_dev *pdev)
2725 {
2726 	struct mlx4_dev	 *dev  = pci_get_drvdata(pdev);
2727 	struct mlx4_priv *priv = mlx4_priv(dev);
2728 	int		  pci_dev_data;
2729 
2730 	pci_dev_data = priv->pci_dev_data;
2731 	__mlx4_remove_one(pdev);
2732 	return __mlx4_init_one(pdev, pci_dev_data);
2733 }
2734 
2735 static const struct pci_device_id mlx4_pci_table[] = {
2736 	/* MT25408 "Hermon" SDR */
2737 	{ PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2738 	/* MT25408 "Hermon" DDR */
2739 	{ PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2740 	/* MT25408 "Hermon" QDR */
2741 	{ PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2742 	/* MT25408 "Hermon" DDR PCIe gen2 */
2743 	{ PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2744 	/* MT25408 "Hermon" QDR PCIe gen2 */
2745 	{ PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2746 	/* MT25408 "Hermon" EN 10GigE */
2747 	{ PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2748 	/* MT25408 "Hermon" EN 10GigE PCIe gen2 */
2749 	{ PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2750 	/* MT25458 ConnectX EN 10GBASE-T 10GigE */
2751 	{ PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2752 	/* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
2753 	{ PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2754 	/* MT26468 ConnectX EN 10GigE PCIe gen2*/
2755 	{ PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2756 	/* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
2757 	{ PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2758 	/* MT26478 ConnectX2 40GigE PCIe gen2 */
2759 	{ PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2760 	/* MT25400 Family [ConnectX-2 Virtual Function] */
2761 	{ PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
2762 	/* MT27500 Family [ConnectX-3] */
2763 	{ PCI_VDEVICE(MELLANOX, 0x1003), 0 },
2764 	/* MT27500 Family [ConnectX-3 Virtual Function] */
2765 	{ PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
2766 	{ PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
2767 	{ PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
2768 	{ PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
2769 	{ PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
2770 	{ PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
2771 	{ PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
2772 	{ PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
2773 	{ PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
2774 	{ PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
2775 	{ PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
2776 	{ PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
2777 	{ PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
2778 	{ 0, }
2779 };
2780 
2781 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
2782 
2783 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
2784 					      pci_channel_state_t state)
2785 {
2786 	__mlx4_remove_one(pdev);
2787 
2788 	return state == pci_channel_io_perm_failure ?
2789 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
2790 }
2791 
2792 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
2793 {
2794 	struct mlx4_dev	 *dev  = pci_get_drvdata(pdev);
2795 	struct mlx4_priv *priv = mlx4_priv(dev);
2796 	int               ret;
2797 
2798 	ret = __mlx4_init_one(pdev, priv->pci_dev_data);
2799 
2800 	return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
2801 }
2802 
2803 static const struct pci_error_handlers mlx4_err_handler = {
2804 	.error_detected = mlx4_pci_err_detected,
2805 	.slot_reset     = mlx4_pci_slot_reset,
2806 };
2807 
2808 static struct pci_driver mlx4_driver = {
2809 	.name		= DRV_NAME,
2810 	.id_table	= mlx4_pci_table,
2811 	.probe		= mlx4_init_one,
2812 	.shutdown	= __mlx4_remove_one,
2813 	.remove		= mlx4_remove_one,
2814 	.err_handler    = &mlx4_err_handler,
2815 };
2816 
2817 static int __init mlx4_verify_params(void)
2818 {
2819 	if ((log_num_mac < 0) || (log_num_mac > 7)) {
2820 		pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
2821 		return -1;
2822 	}
2823 
2824 	if (log_num_vlan != 0)
2825 		pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
2826 			MLX4_LOG_NUM_VLANS);
2827 
2828 	if (use_prio != 0)
2829 		pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
2830 
2831 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
2832 		pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
2833 			log_mtts_per_seg);
2834 		return -1;
2835 	}
2836 
2837 	/* Check if module param for ports type has legal combination */
2838 	if (port_type_array[0] == false && port_type_array[1] == true) {
2839 		pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
2840 		port_type_array[0] = true;
2841 	}
2842 
2843 	if (mlx4_log_num_mgm_entry_size != -1 &&
2844 	    (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
2845 	     mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE)) {
2846 		pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-1 or %d..%d)\n",
2847 			mlx4_log_num_mgm_entry_size,
2848 			MLX4_MIN_MGM_LOG_ENTRY_SIZE,
2849 			MLX4_MAX_MGM_LOG_ENTRY_SIZE);
2850 		return -1;
2851 	}
2852 
2853 	return 0;
2854 }
2855 
2856 static int __init mlx4_init(void)
2857 {
2858 	int ret;
2859 
2860 	if (mlx4_verify_params())
2861 		return -EINVAL;
2862 
2863 	mlx4_catas_init();
2864 
2865 	mlx4_wq = create_singlethread_workqueue("mlx4");
2866 	if (!mlx4_wq)
2867 		return -ENOMEM;
2868 
2869 	ret = pci_register_driver(&mlx4_driver);
2870 	if (ret < 0)
2871 		destroy_workqueue(mlx4_wq);
2872 	return ret < 0 ? ret : 0;
2873 }
2874 
2875 static void __exit mlx4_cleanup(void)
2876 {
2877 	pci_unregister_driver(&mlx4_driver);
2878 	destroy_workqueue(mlx4_wq);
2879 }
2880 
2881 module_init(mlx4_init);
2882 module_exit(mlx4_cleanup);
2883