1 /*
2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
4  * Copyright (c) 2005, 2006, 2007 Cisco Systems, Inc.  All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer in the documentation and/or other materials
23  *        provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 
35 #include <linux/etherdevice.h>
36 #include <linux/mlx4/cmd.h>
37 #include <linux/module.h>
38 #include <linux/cache.h>
39 
40 #include "fw.h"
41 #include "icm.h"
42 
43 enum {
44 	MLX4_COMMAND_INTERFACE_MIN_REV		= 2,
45 	MLX4_COMMAND_INTERFACE_MAX_REV		= 3,
46 	MLX4_COMMAND_INTERFACE_NEW_PORT_CMDS	= 3,
47 };
48 
49 extern void __buggy_use_of_MLX4_GET(void);
50 extern void __buggy_use_of_MLX4_PUT(void);
51 
52 static bool enable_qos = true;
53 module_param(enable_qos, bool, 0444);
54 MODULE_PARM_DESC(enable_qos, "Enable Enhanced QoS support (default: on)");
55 
56 #define MLX4_GET(dest, source, offset)				      \
57 	do {							      \
58 		void *__p = (char *) (source) + (offset);	      \
59 		u64 val;                                              \
60 		switch (sizeof (dest)) {			      \
61 		case 1: (dest) = *(u8 *) __p;	    break;	      \
62 		case 2: (dest) = be16_to_cpup(__p); break;	      \
63 		case 4: (dest) = be32_to_cpup(__p); break;	      \
64 		case 8: val = get_unaligned((u64 *)__p);              \
65 			(dest) = be64_to_cpu(val);  break;            \
66 		default: __buggy_use_of_MLX4_GET();		      \
67 		}						      \
68 	} while (0)
69 
70 #define MLX4_PUT(dest, source, offset)				      \
71 	do {							      \
72 		void *__d = ((char *) (dest) + (offset));	      \
73 		switch (sizeof(source)) {			      \
74 		case 1: *(u8 *) __d = (source);		       break; \
75 		case 2:	*(__be16 *) __d = cpu_to_be16(source); break; \
76 		case 4:	*(__be32 *) __d = cpu_to_be32(source); break; \
77 		case 8:	*(__be64 *) __d = cpu_to_be64(source); break; \
78 		default: __buggy_use_of_MLX4_PUT();		      \
79 		}						      \
80 	} while (0)
81 
82 static void dump_dev_cap_flags(struct mlx4_dev *dev, u64 flags)
83 {
84 	static const char *fname[] = {
85 		[ 0] = "RC transport",
86 		[ 1] = "UC transport",
87 		[ 2] = "UD transport",
88 		[ 3] = "XRC transport",
89 		[ 6] = "SRQ support",
90 		[ 7] = "IPoIB checksum offload",
91 		[ 8] = "P_Key violation counter",
92 		[ 9] = "Q_Key violation counter",
93 		[12] = "Dual Port Different Protocol (DPDP) support",
94 		[15] = "Big LSO headers",
95 		[16] = "MW support",
96 		[17] = "APM support",
97 		[18] = "Atomic ops support",
98 		[19] = "Raw multicast support",
99 		[20] = "Address vector port checking support",
100 		[21] = "UD multicast support",
101 		[30] = "IBoE support",
102 		[32] = "Unicast loopback support",
103 		[34] = "FCS header control",
104 		[37] = "Wake On LAN (port1) support",
105 		[38] = "Wake On LAN (port2) support",
106 		[40] = "UDP RSS support",
107 		[41] = "Unicast VEP steering support",
108 		[42] = "Multicast VEP steering support",
109 		[48] = "Counters support",
110 		[52] = "RSS IP fragments support",
111 		[53] = "Port ETS Scheduler support",
112 		[55] = "Port link type sensing support",
113 		[59] = "Port management change event support",
114 		[61] = "64 byte EQE support",
115 		[62] = "64 byte CQE support",
116 	};
117 	int i;
118 
119 	mlx4_dbg(dev, "DEV_CAP flags:\n");
120 	for (i = 0; i < ARRAY_SIZE(fname); ++i)
121 		if (fname[i] && (flags & (1LL << i)))
122 			mlx4_dbg(dev, "    %s\n", fname[i]);
123 }
124 
125 static void dump_dev_cap_flags2(struct mlx4_dev *dev, u64 flags)
126 {
127 	static const char * const fname[] = {
128 		[0] = "RSS support",
129 		[1] = "RSS Toeplitz Hash Function support",
130 		[2] = "RSS XOR Hash Function support",
131 		[3] = "Device managed flow steering support",
132 		[4] = "Automatic MAC reassignment support",
133 		[5] = "Time stamping support",
134 		[6] = "VST (control vlan insertion/stripping) support",
135 		[7] = "FSM (MAC anti-spoofing) support",
136 		[8] = "Dynamic QP updates support",
137 		[9] = "Device managed flow steering IPoIB support",
138 		[10] = "TCP/IP offloads/flow-steering for VXLAN support",
139 		[11] = "MAD DEMUX (Secure-Host) support",
140 		[12] = "Large cache line (>64B) CQE stride support",
141 		[13] = "Large cache line (>64B) EQE stride support",
142 		[14] = "Ethernet protocol control support",
143 		[15] = "Ethernet Backplane autoneg support",
144 		[16] = "CONFIG DEV support",
145 		[17] = "Asymmetric EQs support",
146 		[18] = "More than 80 VFs support",
147 		[19] = "Performance optimized for limited rule configuration flow steering support",
148 		[20] = "Recoverable error events support",
149 		[21] = "Port Remap support",
150 		[22] = "QCN support",
151 		[23] = "QP rate limiting support",
152 		[24] = "Ethernet Flow control statistics support",
153 		[25] = "Granular QoS per VF support",
154 		[26] = "Port ETS Scheduler support",
155 		[27] = "Port beacon support",
156 		[28] = "RX-ALL support",
157 		[29] = "802.1ad offload support",
158 		[31] = "Modifying loopback source checks using UPDATE_QP support",
159 		[32] = "Loopback source checks support",
160 		[33] = "RoCEv2 support"
161 	};
162 	int i;
163 
164 	for (i = 0; i < ARRAY_SIZE(fname); ++i)
165 		if (fname[i] && (flags & (1LL << i)))
166 			mlx4_dbg(dev, "    %s\n", fname[i]);
167 }
168 
169 int mlx4_MOD_STAT_CFG(struct mlx4_dev *dev, struct mlx4_mod_stat_cfg *cfg)
170 {
171 	struct mlx4_cmd_mailbox *mailbox;
172 	u32 *inbox;
173 	int err = 0;
174 
175 #define MOD_STAT_CFG_IN_SIZE		0x100
176 
177 #define MOD_STAT_CFG_PG_SZ_M_OFFSET	0x002
178 #define MOD_STAT_CFG_PG_SZ_OFFSET	0x003
179 
180 	mailbox = mlx4_alloc_cmd_mailbox(dev);
181 	if (IS_ERR(mailbox))
182 		return PTR_ERR(mailbox);
183 	inbox = mailbox->buf;
184 
185 	MLX4_PUT(inbox, cfg->log_pg_sz, MOD_STAT_CFG_PG_SZ_OFFSET);
186 	MLX4_PUT(inbox, cfg->log_pg_sz_m, MOD_STAT_CFG_PG_SZ_M_OFFSET);
187 
188 	err = mlx4_cmd(dev, mailbox->dma, 0, 0, MLX4_CMD_MOD_STAT_CFG,
189 			MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
190 
191 	mlx4_free_cmd_mailbox(dev, mailbox);
192 	return err;
193 }
194 
195 int mlx4_QUERY_FUNC(struct mlx4_dev *dev, struct mlx4_func *func, int slave)
196 {
197 	struct mlx4_cmd_mailbox *mailbox;
198 	u32 *outbox;
199 	u8 in_modifier;
200 	u8 field;
201 	u16 field16;
202 	int err;
203 
204 #define QUERY_FUNC_BUS_OFFSET			0x00
205 #define QUERY_FUNC_DEVICE_OFFSET		0x01
206 #define QUERY_FUNC_FUNCTION_OFFSET		0x01
207 #define QUERY_FUNC_PHYSICAL_FUNCTION_OFFSET	0x03
208 #define QUERY_FUNC_RSVD_EQS_OFFSET		0x04
209 #define QUERY_FUNC_MAX_EQ_OFFSET		0x06
210 #define QUERY_FUNC_RSVD_UARS_OFFSET		0x0b
211 
212 	mailbox = mlx4_alloc_cmd_mailbox(dev);
213 	if (IS_ERR(mailbox))
214 		return PTR_ERR(mailbox);
215 	outbox = mailbox->buf;
216 
217 	in_modifier = slave;
218 
219 	err = mlx4_cmd_box(dev, 0, mailbox->dma, in_modifier, 0,
220 			   MLX4_CMD_QUERY_FUNC,
221 			   MLX4_CMD_TIME_CLASS_A,
222 			   MLX4_CMD_NATIVE);
223 	if (err)
224 		goto out;
225 
226 	MLX4_GET(field, outbox, QUERY_FUNC_BUS_OFFSET);
227 	func->bus = field & 0xf;
228 	MLX4_GET(field, outbox, QUERY_FUNC_DEVICE_OFFSET);
229 	func->device = field & 0xf1;
230 	MLX4_GET(field, outbox, QUERY_FUNC_FUNCTION_OFFSET);
231 	func->function = field & 0x7;
232 	MLX4_GET(field, outbox, QUERY_FUNC_PHYSICAL_FUNCTION_OFFSET);
233 	func->physical_function = field & 0xf;
234 	MLX4_GET(field16, outbox, QUERY_FUNC_RSVD_EQS_OFFSET);
235 	func->rsvd_eqs = field16 & 0xffff;
236 	MLX4_GET(field16, outbox, QUERY_FUNC_MAX_EQ_OFFSET);
237 	func->max_eq = field16 & 0xffff;
238 	MLX4_GET(field, outbox, QUERY_FUNC_RSVD_UARS_OFFSET);
239 	func->rsvd_uars = field & 0x0f;
240 
241 	mlx4_dbg(dev, "Bus: %d, Device: %d, Function: %d, Physical function: %d, Max EQs: %d, Reserved EQs: %d, Reserved UARs: %d\n",
242 		 func->bus, func->device, func->function, func->physical_function,
243 		 func->max_eq, func->rsvd_eqs, func->rsvd_uars);
244 
245 out:
246 	mlx4_free_cmd_mailbox(dev, mailbox);
247 	return err;
248 }
249 
250 int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
251 				struct mlx4_vhcr *vhcr,
252 				struct mlx4_cmd_mailbox *inbox,
253 				struct mlx4_cmd_mailbox *outbox,
254 				struct mlx4_cmd_info *cmd)
255 {
256 	struct mlx4_priv *priv = mlx4_priv(dev);
257 	u8	field, port;
258 	u32	size, proxy_qp, qkey;
259 	int	err = 0;
260 	struct mlx4_func func;
261 
262 #define QUERY_FUNC_CAP_FLAGS_OFFSET		0x0
263 #define QUERY_FUNC_CAP_NUM_PORTS_OFFSET		0x1
264 #define QUERY_FUNC_CAP_PF_BHVR_OFFSET		0x4
265 #define QUERY_FUNC_CAP_FMR_OFFSET		0x8
266 #define QUERY_FUNC_CAP_QP_QUOTA_OFFSET_DEP	0x10
267 #define QUERY_FUNC_CAP_CQ_QUOTA_OFFSET_DEP	0x14
268 #define QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET_DEP	0x18
269 #define QUERY_FUNC_CAP_MPT_QUOTA_OFFSET_DEP	0x20
270 #define QUERY_FUNC_CAP_MTT_QUOTA_OFFSET_DEP	0x24
271 #define QUERY_FUNC_CAP_MCG_QUOTA_OFFSET_DEP	0x28
272 #define QUERY_FUNC_CAP_MAX_EQ_OFFSET		0x2c
273 #define QUERY_FUNC_CAP_RESERVED_EQ_OFFSET	0x30
274 #define QUERY_FUNC_CAP_QP_RESD_LKEY_OFFSET	0x48
275 
276 #define QUERY_FUNC_CAP_QP_QUOTA_OFFSET		0x50
277 #define QUERY_FUNC_CAP_CQ_QUOTA_OFFSET		0x54
278 #define QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET		0x58
279 #define QUERY_FUNC_CAP_MPT_QUOTA_OFFSET		0x60
280 #define QUERY_FUNC_CAP_MTT_QUOTA_OFFSET		0x64
281 #define QUERY_FUNC_CAP_MCG_QUOTA_OFFSET		0x68
282 
283 #define QUERY_FUNC_CAP_EXTRA_FLAGS_OFFSET	0x6c
284 
285 #define QUERY_FUNC_CAP_FMR_FLAG			0x80
286 #define QUERY_FUNC_CAP_FLAG_RDMA		0x40
287 #define QUERY_FUNC_CAP_FLAG_ETH			0x80
288 #define QUERY_FUNC_CAP_FLAG_QUOTAS		0x10
289 #define QUERY_FUNC_CAP_FLAG_RESD_LKEY		0x08
290 #define QUERY_FUNC_CAP_FLAG_VALID_MAILBOX	0x04
291 
292 #define QUERY_FUNC_CAP_EXTRA_FLAGS_BF_QP_ALLOC_FLAG	(1UL << 31)
293 #define QUERY_FUNC_CAP_EXTRA_FLAGS_A0_QP_ALLOC_FLAG	(1UL << 30)
294 
295 /* when opcode modifier = 1 */
296 #define QUERY_FUNC_CAP_PHYS_PORT_OFFSET		0x3
297 #define QUERY_FUNC_CAP_PRIV_VF_QKEY_OFFSET	0x4
298 #define QUERY_FUNC_CAP_FLAGS0_OFFSET		0x8
299 #define QUERY_FUNC_CAP_FLAGS1_OFFSET		0xc
300 
301 #define QUERY_FUNC_CAP_QP0_TUNNEL		0x10
302 #define QUERY_FUNC_CAP_QP0_PROXY		0x14
303 #define QUERY_FUNC_CAP_QP1_TUNNEL		0x18
304 #define QUERY_FUNC_CAP_QP1_PROXY		0x1c
305 #define QUERY_FUNC_CAP_PHYS_PORT_ID		0x28
306 
307 #define QUERY_FUNC_CAP_FLAGS1_FORCE_MAC		0x40
308 #define QUERY_FUNC_CAP_FLAGS1_FORCE_VLAN	0x80
309 #define QUERY_FUNC_CAP_FLAGS1_NIC_INFO			0x10
310 #define QUERY_FUNC_CAP_VF_ENABLE_QP0		0x08
311 
312 #define QUERY_FUNC_CAP_FLAGS0_FORCE_PHY_WQE_GID 0x80
313 #define QUERY_FUNC_CAP_SUPPORTS_NON_POWER_OF_2_NUM_EQS (1 << 31)
314 #define QUERY_FUNC_CAP_PHV_BIT			0x40
315 
316 	if (vhcr->op_modifier == 1) {
317 		struct mlx4_active_ports actv_ports =
318 			mlx4_get_active_ports(dev, slave);
319 		int converted_port = mlx4_slave_convert_port(
320 				dev, slave, vhcr->in_modifier);
321 
322 		if (converted_port < 0)
323 			return -EINVAL;
324 
325 		vhcr->in_modifier = converted_port;
326 		/* phys-port = logical-port */
327 		field = vhcr->in_modifier -
328 			find_first_bit(actv_ports.ports, dev->caps.num_ports);
329 		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_PHYS_PORT_OFFSET);
330 
331 		port = vhcr->in_modifier;
332 		proxy_qp = dev->phys_caps.base_proxy_sqpn + 8 * slave + port - 1;
333 
334 		/* Set nic_info bit to mark new fields support */
335 		field  = QUERY_FUNC_CAP_FLAGS1_NIC_INFO;
336 
337 		if (mlx4_vf_smi_enabled(dev, slave, port) &&
338 		    !mlx4_get_parav_qkey(dev, proxy_qp, &qkey)) {
339 			field |= QUERY_FUNC_CAP_VF_ENABLE_QP0;
340 			MLX4_PUT(outbox->buf, qkey,
341 				 QUERY_FUNC_CAP_PRIV_VF_QKEY_OFFSET);
342 		}
343 		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_FLAGS1_OFFSET);
344 
345 		/* size is now the QP number */
346 		size = dev->phys_caps.base_tunnel_sqpn + 8 * slave + port - 1;
347 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP0_TUNNEL);
348 
349 		size += 2;
350 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP1_TUNNEL);
351 
352 		MLX4_PUT(outbox->buf, proxy_qp, QUERY_FUNC_CAP_QP0_PROXY);
353 		proxy_qp += 2;
354 		MLX4_PUT(outbox->buf, proxy_qp, QUERY_FUNC_CAP_QP1_PROXY);
355 
356 		MLX4_PUT(outbox->buf, dev->caps.phys_port_id[vhcr->in_modifier],
357 			 QUERY_FUNC_CAP_PHYS_PORT_ID);
358 
359 		if (dev->caps.phv_bit[port]) {
360 			field = QUERY_FUNC_CAP_PHV_BIT;
361 			MLX4_PUT(outbox->buf, field,
362 				 QUERY_FUNC_CAP_FLAGS0_OFFSET);
363 		}
364 
365 	} else if (vhcr->op_modifier == 0) {
366 		struct mlx4_active_ports actv_ports =
367 			mlx4_get_active_ports(dev, slave);
368 		/* enable rdma and ethernet interfaces, new quota locations,
369 		 * and reserved lkey
370 		 */
371 		field = (QUERY_FUNC_CAP_FLAG_ETH | QUERY_FUNC_CAP_FLAG_RDMA |
372 			 QUERY_FUNC_CAP_FLAG_QUOTAS | QUERY_FUNC_CAP_FLAG_VALID_MAILBOX |
373 			 QUERY_FUNC_CAP_FLAG_RESD_LKEY);
374 		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_FLAGS_OFFSET);
375 
376 		field = min(
377 			bitmap_weight(actv_ports.ports, dev->caps.num_ports),
378 			dev->caps.num_ports);
379 		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_NUM_PORTS_OFFSET);
380 
381 		size = dev->caps.function_caps; /* set PF behaviours */
382 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_PF_BHVR_OFFSET);
383 
384 		field = 0; /* protected FMR support not available as yet */
385 		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_FMR_OFFSET);
386 
387 		size = priv->mfunc.master.res_tracker.res_alloc[RES_QP].quota[slave];
388 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP_QUOTA_OFFSET);
389 		size = dev->caps.num_qps;
390 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP_QUOTA_OFFSET_DEP);
391 
392 		size = priv->mfunc.master.res_tracker.res_alloc[RES_SRQ].quota[slave];
393 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET);
394 		size = dev->caps.num_srqs;
395 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET_DEP);
396 
397 		size = priv->mfunc.master.res_tracker.res_alloc[RES_CQ].quota[slave];
398 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET);
399 		size = dev->caps.num_cqs;
400 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET_DEP);
401 
402 		if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) ||
403 		    mlx4_QUERY_FUNC(dev, &func, slave)) {
404 			size = vhcr->in_modifier &
405 				QUERY_FUNC_CAP_SUPPORTS_NON_POWER_OF_2_NUM_EQS ?
406 				dev->caps.num_eqs :
407 				rounddown_pow_of_two(dev->caps.num_eqs);
408 			MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MAX_EQ_OFFSET);
409 			size = dev->caps.reserved_eqs;
410 			MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET);
411 		} else {
412 			size = vhcr->in_modifier &
413 				QUERY_FUNC_CAP_SUPPORTS_NON_POWER_OF_2_NUM_EQS ?
414 				func.max_eq :
415 				rounddown_pow_of_two(func.max_eq);
416 			MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MAX_EQ_OFFSET);
417 			size = func.rsvd_eqs;
418 			MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET);
419 		}
420 
421 		size = priv->mfunc.master.res_tracker.res_alloc[RES_MPT].quota[slave];
422 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET);
423 		size = dev->caps.num_mpts;
424 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET_DEP);
425 
426 		size = priv->mfunc.master.res_tracker.res_alloc[RES_MTT].quota[slave];
427 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET);
428 		size = dev->caps.num_mtts;
429 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET_DEP);
430 
431 		size = dev->caps.num_mgms + dev->caps.num_amgms;
432 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET);
433 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET_DEP);
434 
435 		size = QUERY_FUNC_CAP_EXTRA_FLAGS_BF_QP_ALLOC_FLAG |
436 			QUERY_FUNC_CAP_EXTRA_FLAGS_A0_QP_ALLOC_FLAG;
437 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_EXTRA_FLAGS_OFFSET);
438 
439 		size = dev->caps.reserved_lkey + ((slave << 8) & 0xFF00);
440 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP_RESD_LKEY_OFFSET);
441 	} else
442 		err = -EINVAL;
443 
444 	return err;
445 }
446 
447 int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, u8 gen_or_port,
448 			struct mlx4_func_cap *func_cap)
449 {
450 	struct mlx4_cmd_mailbox *mailbox;
451 	u32			*outbox;
452 	u8			field, op_modifier;
453 	u32			size, qkey;
454 	int			err = 0, quotas = 0;
455 	u32                     in_modifier;
456 
457 	op_modifier = !!gen_or_port; /* 0 = general, 1 = logical port */
458 	in_modifier = op_modifier ? gen_or_port :
459 		QUERY_FUNC_CAP_SUPPORTS_NON_POWER_OF_2_NUM_EQS;
460 
461 	mailbox = mlx4_alloc_cmd_mailbox(dev);
462 	if (IS_ERR(mailbox))
463 		return PTR_ERR(mailbox);
464 
465 	err = mlx4_cmd_box(dev, 0, mailbox->dma, in_modifier, op_modifier,
466 			   MLX4_CMD_QUERY_FUNC_CAP,
467 			   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
468 	if (err)
469 		goto out;
470 
471 	outbox = mailbox->buf;
472 
473 	if (!op_modifier) {
474 		MLX4_GET(field, outbox, QUERY_FUNC_CAP_FLAGS_OFFSET);
475 		if (!(field & (QUERY_FUNC_CAP_FLAG_ETH | QUERY_FUNC_CAP_FLAG_RDMA))) {
476 			mlx4_err(dev, "The host supports neither eth nor rdma interfaces\n");
477 			err = -EPROTONOSUPPORT;
478 			goto out;
479 		}
480 		func_cap->flags = field;
481 		quotas = !!(func_cap->flags & QUERY_FUNC_CAP_FLAG_QUOTAS);
482 
483 		MLX4_GET(field, outbox, QUERY_FUNC_CAP_NUM_PORTS_OFFSET);
484 		func_cap->num_ports = field;
485 
486 		MLX4_GET(size, outbox, QUERY_FUNC_CAP_PF_BHVR_OFFSET);
487 		func_cap->pf_context_behaviour = size;
488 
489 		if (quotas) {
490 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP_QUOTA_OFFSET);
491 			func_cap->qp_quota = size & 0xFFFFFF;
492 
493 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET);
494 			func_cap->srq_quota = size & 0xFFFFFF;
495 
496 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET);
497 			func_cap->cq_quota = size & 0xFFFFFF;
498 
499 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET);
500 			func_cap->mpt_quota = size & 0xFFFFFF;
501 
502 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET);
503 			func_cap->mtt_quota = size & 0xFFFFFF;
504 
505 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET);
506 			func_cap->mcg_quota = size & 0xFFFFFF;
507 
508 		} else {
509 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP_QUOTA_OFFSET_DEP);
510 			func_cap->qp_quota = size & 0xFFFFFF;
511 
512 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET_DEP);
513 			func_cap->srq_quota = size & 0xFFFFFF;
514 
515 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET_DEP);
516 			func_cap->cq_quota = size & 0xFFFFFF;
517 
518 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET_DEP);
519 			func_cap->mpt_quota = size & 0xFFFFFF;
520 
521 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET_DEP);
522 			func_cap->mtt_quota = size & 0xFFFFFF;
523 
524 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET_DEP);
525 			func_cap->mcg_quota = size & 0xFFFFFF;
526 		}
527 		MLX4_GET(size, outbox, QUERY_FUNC_CAP_MAX_EQ_OFFSET);
528 		func_cap->max_eq = size & 0xFFFFFF;
529 
530 		MLX4_GET(size, outbox, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET);
531 		func_cap->reserved_eq = size & 0xFFFFFF;
532 
533 		if (func_cap->flags & QUERY_FUNC_CAP_FLAG_RESD_LKEY) {
534 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP_RESD_LKEY_OFFSET);
535 			func_cap->reserved_lkey = size;
536 		} else {
537 			func_cap->reserved_lkey = 0;
538 		}
539 
540 		func_cap->extra_flags = 0;
541 
542 		/* Mailbox data from 0x6c and onward should only be treated if
543 		 * QUERY_FUNC_CAP_FLAG_VALID_MAILBOX is set in func_cap->flags
544 		 */
545 		if (func_cap->flags & QUERY_FUNC_CAP_FLAG_VALID_MAILBOX) {
546 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_EXTRA_FLAGS_OFFSET);
547 			if (size & QUERY_FUNC_CAP_EXTRA_FLAGS_BF_QP_ALLOC_FLAG)
548 				func_cap->extra_flags |= MLX4_QUERY_FUNC_FLAGS_BF_RES_QP;
549 			if (size & QUERY_FUNC_CAP_EXTRA_FLAGS_A0_QP_ALLOC_FLAG)
550 				func_cap->extra_flags |= MLX4_QUERY_FUNC_FLAGS_A0_RES_QP;
551 		}
552 
553 		goto out;
554 	}
555 
556 	/* logical port query */
557 	if (gen_or_port > dev->caps.num_ports) {
558 		err = -EINVAL;
559 		goto out;
560 	}
561 
562 	MLX4_GET(func_cap->flags1, outbox, QUERY_FUNC_CAP_FLAGS1_OFFSET);
563 	if (dev->caps.port_type[gen_or_port] == MLX4_PORT_TYPE_ETH) {
564 		if (func_cap->flags1 & QUERY_FUNC_CAP_FLAGS1_FORCE_VLAN) {
565 			mlx4_err(dev, "VLAN is enforced on this port\n");
566 			err = -EPROTONOSUPPORT;
567 			goto out;
568 		}
569 
570 		if (func_cap->flags1 & QUERY_FUNC_CAP_FLAGS1_FORCE_MAC) {
571 			mlx4_err(dev, "Force mac is enabled on this port\n");
572 			err = -EPROTONOSUPPORT;
573 			goto out;
574 		}
575 	} else if (dev->caps.port_type[gen_or_port] == MLX4_PORT_TYPE_IB) {
576 		MLX4_GET(field, outbox, QUERY_FUNC_CAP_FLAGS0_OFFSET);
577 		if (field & QUERY_FUNC_CAP_FLAGS0_FORCE_PHY_WQE_GID) {
578 			mlx4_err(dev, "phy_wqe_gid is enforced on this ib port\n");
579 			err = -EPROTONOSUPPORT;
580 			goto out;
581 		}
582 	}
583 
584 	MLX4_GET(field, outbox, QUERY_FUNC_CAP_PHYS_PORT_OFFSET);
585 	func_cap->physical_port = field;
586 	if (func_cap->physical_port != gen_or_port) {
587 		err = -ENOSYS;
588 		goto out;
589 	}
590 
591 	if (func_cap->flags1 & QUERY_FUNC_CAP_VF_ENABLE_QP0) {
592 		MLX4_GET(qkey, outbox, QUERY_FUNC_CAP_PRIV_VF_QKEY_OFFSET);
593 		func_cap->qp0_qkey = qkey;
594 	} else {
595 		func_cap->qp0_qkey = 0;
596 	}
597 
598 	MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP0_TUNNEL);
599 	func_cap->qp0_tunnel_qpn = size & 0xFFFFFF;
600 
601 	MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP0_PROXY);
602 	func_cap->qp0_proxy_qpn = size & 0xFFFFFF;
603 
604 	MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP1_TUNNEL);
605 	func_cap->qp1_tunnel_qpn = size & 0xFFFFFF;
606 
607 	MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP1_PROXY);
608 	func_cap->qp1_proxy_qpn = size & 0xFFFFFF;
609 
610 	if (func_cap->flags1 & QUERY_FUNC_CAP_FLAGS1_NIC_INFO)
611 		MLX4_GET(func_cap->phys_port_id, outbox,
612 			 QUERY_FUNC_CAP_PHYS_PORT_ID);
613 
614 	MLX4_GET(field, outbox, QUERY_FUNC_CAP_FLAGS0_OFFSET);
615 	func_cap->flags |= (field & QUERY_FUNC_CAP_PHV_BIT);
616 
617 	/* All other resources are allocated by the master, but we still report
618 	 * 'num' and 'reserved' capabilities as follows:
619 	 * - num remains the maximum resource index
620 	 * - 'num - reserved' is the total available objects of a resource, but
621 	 *   resource indices may be less than 'reserved'
622 	 * TODO: set per-resource quotas */
623 
624 out:
625 	mlx4_free_cmd_mailbox(dev, mailbox);
626 
627 	return err;
628 }
629 
630 static void disable_unsupported_roce_caps(void *buf);
631 
632 int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
633 {
634 	struct mlx4_cmd_mailbox *mailbox;
635 	u32 *outbox;
636 	u8 field;
637 	u32 field32, flags, ext_flags;
638 	u16 size;
639 	u16 stat_rate;
640 	int err;
641 	int i;
642 
643 #define QUERY_DEV_CAP_OUT_SIZE		       0x100
644 #define QUERY_DEV_CAP_MAX_SRQ_SZ_OFFSET		0x10
645 #define QUERY_DEV_CAP_MAX_QP_SZ_OFFSET		0x11
646 #define QUERY_DEV_CAP_RSVD_QP_OFFSET		0x12
647 #define QUERY_DEV_CAP_MAX_QP_OFFSET		0x13
648 #define QUERY_DEV_CAP_RSVD_SRQ_OFFSET		0x14
649 #define QUERY_DEV_CAP_MAX_SRQ_OFFSET		0x15
650 #define QUERY_DEV_CAP_RSVD_EEC_OFFSET		0x16
651 #define QUERY_DEV_CAP_MAX_EEC_OFFSET		0x17
652 #define QUERY_DEV_CAP_MAX_CQ_SZ_OFFSET		0x19
653 #define QUERY_DEV_CAP_RSVD_CQ_OFFSET		0x1a
654 #define QUERY_DEV_CAP_MAX_CQ_OFFSET		0x1b
655 #define QUERY_DEV_CAP_MAX_MPT_OFFSET		0x1d
656 #define QUERY_DEV_CAP_RSVD_EQ_OFFSET		0x1e
657 #define QUERY_DEV_CAP_MAX_EQ_OFFSET		0x1f
658 #define QUERY_DEV_CAP_RSVD_MTT_OFFSET		0x20
659 #define QUERY_DEV_CAP_MAX_MRW_SZ_OFFSET		0x21
660 #define QUERY_DEV_CAP_RSVD_MRW_OFFSET		0x22
661 #define QUERY_DEV_CAP_MAX_MTT_SEG_OFFSET	0x23
662 #define QUERY_DEV_CAP_NUM_SYS_EQ_OFFSET		0x26
663 #define QUERY_DEV_CAP_MAX_AV_OFFSET		0x27
664 #define QUERY_DEV_CAP_MAX_REQ_QP_OFFSET		0x29
665 #define QUERY_DEV_CAP_MAX_RES_QP_OFFSET		0x2b
666 #define QUERY_DEV_CAP_MAX_GSO_OFFSET		0x2d
667 #define QUERY_DEV_CAP_RSS_OFFSET		0x2e
668 #define QUERY_DEV_CAP_MAX_RDMA_OFFSET		0x2f
669 #define QUERY_DEV_CAP_RSZ_SRQ_OFFSET		0x33
670 #define QUERY_DEV_CAP_PORT_BEACON_OFFSET	0x34
671 #define QUERY_DEV_CAP_ACK_DELAY_OFFSET		0x35
672 #define QUERY_DEV_CAP_MTU_WIDTH_OFFSET		0x36
673 #define QUERY_DEV_CAP_VL_PORT_OFFSET		0x37
674 #define QUERY_DEV_CAP_MAX_MSG_SZ_OFFSET		0x38
675 #define QUERY_DEV_CAP_MAX_GID_OFFSET		0x3b
676 #define QUERY_DEV_CAP_RATE_SUPPORT_OFFSET	0x3c
677 #define QUERY_DEV_CAP_CQ_TS_SUPPORT_OFFSET	0x3e
678 #define QUERY_DEV_CAP_MAX_PKEY_OFFSET		0x3f
679 #define QUERY_DEV_CAP_EXT_FLAGS_OFFSET		0x40
680 #define QUERY_DEV_CAP_FLAGS_OFFSET		0x44
681 #define QUERY_DEV_CAP_RSVD_UAR_OFFSET		0x48
682 #define QUERY_DEV_CAP_UAR_SZ_OFFSET		0x49
683 #define QUERY_DEV_CAP_PAGE_SZ_OFFSET		0x4b
684 #define QUERY_DEV_CAP_BF_OFFSET			0x4c
685 #define QUERY_DEV_CAP_LOG_BF_REG_SZ_OFFSET	0x4d
686 #define QUERY_DEV_CAP_LOG_MAX_BF_REGS_PER_PAGE_OFFSET	0x4e
687 #define QUERY_DEV_CAP_LOG_MAX_BF_PAGES_OFFSET	0x4f
688 #define QUERY_DEV_CAP_MAX_SG_SQ_OFFSET		0x51
689 #define QUERY_DEV_CAP_MAX_DESC_SZ_SQ_OFFSET	0x52
690 #define QUERY_DEV_CAP_MAX_SG_RQ_OFFSET		0x55
691 #define QUERY_DEV_CAP_MAX_DESC_SZ_RQ_OFFSET	0x56
692 #define QUERY_DEV_CAP_MAX_QP_MCG_OFFSET		0x61
693 #define QUERY_DEV_CAP_RSVD_MCG_OFFSET		0x62
694 #define QUERY_DEV_CAP_MAX_MCG_OFFSET		0x63
695 #define QUERY_DEV_CAP_RSVD_PD_OFFSET		0x64
696 #define QUERY_DEV_CAP_MAX_PD_OFFSET		0x65
697 #define QUERY_DEV_CAP_RSVD_XRC_OFFSET		0x66
698 #define QUERY_DEV_CAP_MAX_XRC_OFFSET		0x67
699 #define QUERY_DEV_CAP_MAX_COUNTERS_OFFSET	0x68
700 #define QUERY_DEV_CAP_PORT_FLOWSTATS_COUNTERS_OFFSET	0x70
701 #define QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET	0x70
702 #define QUERY_DEV_CAP_FLOW_STEERING_IPOIB_OFFSET	0x74
703 #define QUERY_DEV_CAP_FLOW_STEERING_RANGE_EN_OFFSET	0x76
704 #define QUERY_DEV_CAP_FLOW_STEERING_MAX_QP_OFFSET	0x77
705 #define QUERY_DEV_CAP_CQ_EQ_CACHE_LINE_STRIDE	0x7a
706 #define QUERY_DEV_CAP_ECN_QCN_VER_OFFSET	0x7b
707 #define QUERY_DEV_CAP_RDMARC_ENTRY_SZ_OFFSET	0x80
708 #define QUERY_DEV_CAP_QPC_ENTRY_SZ_OFFSET	0x82
709 #define QUERY_DEV_CAP_AUX_ENTRY_SZ_OFFSET	0x84
710 #define QUERY_DEV_CAP_ALTC_ENTRY_SZ_OFFSET	0x86
711 #define QUERY_DEV_CAP_EQC_ENTRY_SZ_OFFSET	0x88
712 #define QUERY_DEV_CAP_CQC_ENTRY_SZ_OFFSET	0x8a
713 #define QUERY_DEV_CAP_SRQ_ENTRY_SZ_OFFSET	0x8c
714 #define QUERY_DEV_CAP_C_MPT_ENTRY_SZ_OFFSET	0x8e
715 #define QUERY_DEV_CAP_MTT_ENTRY_SZ_OFFSET	0x90
716 #define QUERY_DEV_CAP_D_MPT_ENTRY_SZ_OFFSET	0x92
717 #define QUERY_DEV_CAP_BMME_FLAGS_OFFSET		0x94
718 #define QUERY_DEV_CAP_CONFIG_DEV_OFFSET		0x94
719 #define QUERY_DEV_CAP_PHV_EN_OFFSET		0x96
720 #define QUERY_DEV_CAP_RSVD_LKEY_OFFSET		0x98
721 #define QUERY_DEV_CAP_MAX_ICM_SZ_OFFSET		0xa0
722 #define QUERY_DEV_CAP_ETH_BACKPL_OFFSET		0x9c
723 #define QUERY_DEV_CAP_FW_REASSIGN_MAC		0x9d
724 #define QUERY_DEV_CAP_VXLAN			0x9e
725 #define QUERY_DEV_CAP_MAD_DEMUX_OFFSET		0xb0
726 #define QUERY_DEV_CAP_DMFS_HIGH_RATE_QPN_BASE_OFFSET	0xa8
727 #define QUERY_DEV_CAP_DMFS_HIGH_RATE_QPN_RANGE_OFFSET	0xac
728 #define QUERY_DEV_CAP_QP_RATE_LIMIT_NUM_OFFSET	0xcc
729 #define QUERY_DEV_CAP_QP_RATE_LIMIT_MAX_OFFSET	0xd0
730 #define QUERY_DEV_CAP_QP_RATE_LIMIT_MIN_OFFSET	0xd2
731 
732 
733 	dev_cap->flags2 = 0;
734 	mailbox = mlx4_alloc_cmd_mailbox(dev);
735 	if (IS_ERR(mailbox))
736 		return PTR_ERR(mailbox);
737 	outbox = mailbox->buf;
738 
739 	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0, MLX4_CMD_QUERY_DEV_CAP,
740 			   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
741 	if (err)
742 		goto out;
743 
744 	if (mlx4_is_mfunc(dev))
745 		disable_unsupported_roce_caps(outbox);
746 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_QP_OFFSET);
747 	dev_cap->reserved_qps = 1 << (field & 0xf);
748 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_QP_OFFSET);
749 	dev_cap->max_qps = 1 << (field & 0x1f);
750 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_SRQ_OFFSET);
751 	dev_cap->reserved_srqs = 1 << (field >> 4);
752 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_SRQ_OFFSET);
753 	dev_cap->max_srqs = 1 << (field & 0x1f);
754 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_CQ_SZ_OFFSET);
755 	dev_cap->max_cq_sz = 1 << field;
756 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_CQ_OFFSET);
757 	dev_cap->reserved_cqs = 1 << (field & 0xf);
758 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_CQ_OFFSET);
759 	dev_cap->max_cqs = 1 << (field & 0x1f);
760 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_MPT_OFFSET);
761 	dev_cap->max_mpts = 1 << (field & 0x3f);
762 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_EQ_OFFSET);
763 	dev_cap->reserved_eqs = 1 << (field & 0xf);
764 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_EQ_OFFSET);
765 	dev_cap->max_eqs = 1 << (field & 0xf);
766 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_MTT_OFFSET);
767 	dev_cap->reserved_mtts = 1 << (field >> 4);
768 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_MRW_SZ_OFFSET);
769 	dev_cap->max_mrw_sz = 1 << field;
770 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_MRW_OFFSET);
771 	dev_cap->reserved_mrws = 1 << (field & 0xf);
772 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_MTT_SEG_OFFSET);
773 	dev_cap->max_mtt_seg = 1 << (field & 0x3f);
774 	MLX4_GET(size, outbox, QUERY_DEV_CAP_NUM_SYS_EQ_OFFSET);
775 	dev_cap->num_sys_eqs = size & 0xfff;
776 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_REQ_QP_OFFSET);
777 	dev_cap->max_requester_per_qp = 1 << (field & 0x3f);
778 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_RES_QP_OFFSET);
779 	dev_cap->max_responder_per_qp = 1 << (field & 0x3f);
780 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_GSO_OFFSET);
781 	field &= 0x1f;
782 	if (!field)
783 		dev_cap->max_gso_sz = 0;
784 	else
785 		dev_cap->max_gso_sz = 1 << field;
786 
787 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSS_OFFSET);
788 	if (field & 0x20)
789 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_RSS_XOR;
790 	if (field & 0x10)
791 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_RSS_TOP;
792 	field &= 0xf;
793 	if (field) {
794 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_RSS;
795 		dev_cap->max_rss_tbl_sz = 1 << field;
796 	} else
797 		dev_cap->max_rss_tbl_sz = 0;
798 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_RDMA_OFFSET);
799 	dev_cap->max_rdma_global = 1 << (field & 0x3f);
800 	MLX4_GET(field, outbox, QUERY_DEV_CAP_ACK_DELAY_OFFSET);
801 	dev_cap->local_ca_ack_delay = field & 0x1f;
802 	MLX4_GET(field, outbox, QUERY_DEV_CAP_VL_PORT_OFFSET);
803 	dev_cap->num_ports = field & 0xf;
804 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_MSG_SZ_OFFSET);
805 	dev_cap->max_msg_sz = 1 << (field & 0x1f);
806 	MLX4_GET(field, outbox, QUERY_DEV_CAP_PORT_FLOWSTATS_COUNTERS_OFFSET);
807 	if (field & 0x10)
808 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN;
809 	MLX4_GET(field, outbox, QUERY_DEV_CAP_FLOW_STEERING_RANGE_EN_OFFSET);
810 	if (field & 0x80)
811 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_FS_EN;
812 	dev_cap->fs_log_max_ucast_qp_range_size = field & 0x1f;
813 	MLX4_GET(field, outbox, QUERY_DEV_CAP_PORT_BEACON_OFFSET);
814 	if (field & 0x80)
815 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_PORT_BEACON;
816 	MLX4_GET(field, outbox, QUERY_DEV_CAP_FLOW_STEERING_IPOIB_OFFSET);
817 	if (field & 0x80)
818 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_DMFS_IPOIB;
819 	MLX4_GET(field, outbox, QUERY_DEV_CAP_FLOW_STEERING_MAX_QP_OFFSET);
820 	dev_cap->fs_max_num_qp_per_entry = field;
821 	MLX4_GET(field, outbox, QUERY_DEV_CAP_ECN_QCN_VER_OFFSET);
822 	if (field & 0x1)
823 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_QCN;
824 	MLX4_GET(stat_rate, outbox, QUERY_DEV_CAP_RATE_SUPPORT_OFFSET);
825 	dev_cap->stat_rate_support = stat_rate;
826 	MLX4_GET(field, outbox, QUERY_DEV_CAP_CQ_TS_SUPPORT_OFFSET);
827 	if (field & 0x80)
828 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_TS;
829 	MLX4_GET(ext_flags, outbox, QUERY_DEV_CAP_EXT_FLAGS_OFFSET);
830 	MLX4_GET(flags, outbox, QUERY_DEV_CAP_FLAGS_OFFSET);
831 	dev_cap->flags = flags | (u64)ext_flags << 32;
832 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_UAR_OFFSET);
833 	dev_cap->reserved_uars = field >> 4;
834 	MLX4_GET(field, outbox, QUERY_DEV_CAP_UAR_SZ_OFFSET);
835 	dev_cap->uar_size = 1 << ((field & 0x3f) + 20);
836 	MLX4_GET(field, outbox, QUERY_DEV_CAP_PAGE_SZ_OFFSET);
837 	dev_cap->min_page_sz = 1 << field;
838 
839 	MLX4_GET(field, outbox, QUERY_DEV_CAP_BF_OFFSET);
840 	if (field & 0x80) {
841 		MLX4_GET(field, outbox, QUERY_DEV_CAP_LOG_BF_REG_SZ_OFFSET);
842 		dev_cap->bf_reg_size = 1 << (field & 0x1f);
843 		MLX4_GET(field, outbox, QUERY_DEV_CAP_LOG_MAX_BF_REGS_PER_PAGE_OFFSET);
844 		if ((1 << (field & 0x3f)) > (PAGE_SIZE / dev_cap->bf_reg_size))
845 			field = 3;
846 		dev_cap->bf_regs_per_page = 1 << (field & 0x3f);
847 	} else {
848 		dev_cap->bf_reg_size = 0;
849 	}
850 
851 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_SG_SQ_OFFSET);
852 	dev_cap->max_sq_sg = field;
853 	MLX4_GET(size, outbox, QUERY_DEV_CAP_MAX_DESC_SZ_SQ_OFFSET);
854 	dev_cap->max_sq_desc_sz = size;
855 
856 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_QP_MCG_OFFSET);
857 	dev_cap->max_qp_per_mcg = 1 << field;
858 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_MCG_OFFSET);
859 	dev_cap->reserved_mgms = field & 0xf;
860 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_MCG_OFFSET);
861 	dev_cap->max_mcgs = 1 << field;
862 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_PD_OFFSET);
863 	dev_cap->reserved_pds = field >> 4;
864 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_PD_OFFSET);
865 	dev_cap->max_pds = 1 << (field & 0x3f);
866 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_XRC_OFFSET);
867 	dev_cap->reserved_xrcds = field >> 4;
868 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_XRC_OFFSET);
869 	dev_cap->max_xrcds = 1 << (field & 0x1f);
870 
871 	MLX4_GET(size, outbox, QUERY_DEV_CAP_RDMARC_ENTRY_SZ_OFFSET);
872 	dev_cap->rdmarc_entry_sz = size;
873 	MLX4_GET(size, outbox, QUERY_DEV_CAP_QPC_ENTRY_SZ_OFFSET);
874 	dev_cap->qpc_entry_sz = size;
875 	MLX4_GET(size, outbox, QUERY_DEV_CAP_AUX_ENTRY_SZ_OFFSET);
876 	dev_cap->aux_entry_sz = size;
877 	MLX4_GET(size, outbox, QUERY_DEV_CAP_ALTC_ENTRY_SZ_OFFSET);
878 	dev_cap->altc_entry_sz = size;
879 	MLX4_GET(size, outbox, QUERY_DEV_CAP_EQC_ENTRY_SZ_OFFSET);
880 	dev_cap->eqc_entry_sz = size;
881 	MLX4_GET(size, outbox, QUERY_DEV_CAP_CQC_ENTRY_SZ_OFFSET);
882 	dev_cap->cqc_entry_sz = size;
883 	MLX4_GET(size, outbox, QUERY_DEV_CAP_SRQ_ENTRY_SZ_OFFSET);
884 	dev_cap->srq_entry_sz = size;
885 	MLX4_GET(size, outbox, QUERY_DEV_CAP_C_MPT_ENTRY_SZ_OFFSET);
886 	dev_cap->cmpt_entry_sz = size;
887 	MLX4_GET(size, outbox, QUERY_DEV_CAP_MTT_ENTRY_SZ_OFFSET);
888 	dev_cap->mtt_entry_sz = size;
889 	MLX4_GET(size, outbox, QUERY_DEV_CAP_D_MPT_ENTRY_SZ_OFFSET);
890 	dev_cap->dmpt_entry_sz = size;
891 
892 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_SRQ_SZ_OFFSET);
893 	dev_cap->max_srq_sz = 1 << field;
894 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_QP_SZ_OFFSET);
895 	dev_cap->max_qp_sz = 1 << field;
896 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSZ_SRQ_OFFSET);
897 	dev_cap->resize_srq = field & 1;
898 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_SG_RQ_OFFSET);
899 	dev_cap->max_rq_sg = field;
900 	MLX4_GET(size, outbox, QUERY_DEV_CAP_MAX_DESC_SZ_RQ_OFFSET);
901 	dev_cap->max_rq_desc_sz = size;
902 	MLX4_GET(field, outbox, QUERY_DEV_CAP_CQ_EQ_CACHE_LINE_STRIDE);
903 	if (field & (1 << 4))
904 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_QOS_VPP;
905 	if (field & (1 << 5))
906 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_ETH_PROT_CTRL;
907 	if (field & (1 << 6))
908 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
909 	if (field & (1 << 7))
910 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
911 	MLX4_GET(dev_cap->bmme_flags, outbox,
912 		 QUERY_DEV_CAP_BMME_FLAGS_OFFSET);
913 	if (dev_cap->bmme_flags & MLX4_FLAG_ROCE_V1_V2)
914 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_ROCE_V1_V2;
915 	if (dev_cap->bmme_flags & MLX4_FLAG_PORT_REMAP)
916 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_PORT_REMAP;
917 	MLX4_GET(field, outbox, QUERY_DEV_CAP_CONFIG_DEV_OFFSET);
918 	if (field & 0x20)
919 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_CONFIG_DEV;
920 	if (field & (1 << 2))
921 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
922 	MLX4_GET(field, outbox, QUERY_DEV_CAP_PHV_EN_OFFSET);
923 	if (field & 0x80)
924 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_PHV_EN;
925 	if (field & 0x40)
926 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN;
927 
928 	MLX4_GET(dev_cap->reserved_lkey, outbox,
929 		 QUERY_DEV_CAP_RSVD_LKEY_OFFSET);
930 	MLX4_GET(field32, outbox, QUERY_DEV_CAP_ETH_BACKPL_OFFSET);
931 	if (field32 & (1 << 0))
932 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_ETH_BACKPL_AN_REP;
933 	if (field32 & (1 << 7))
934 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_RECOVERABLE_ERROR_EVENT;
935 	MLX4_GET(field, outbox, QUERY_DEV_CAP_FW_REASSIGN_MAC);
936 	if (field & 1<<6)
937 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN;
938 	MLX4_GET(field, outbox, QUERY_DEV_CAP_VXLAN);
939 	if (field & 1<<3)
940 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS;
941 	if (field & (1 << 5))
942 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
943 	MLX4_GET(dev_cap->max_icm_sz, outbox,
944 		 QUERY_DEV_CAP_MAX_ICM_SZ_OFFSET);
945 	if (dev_cap->flags & MLX4_DEV_CAP_FLAG_COUNTERS)
946 		MLX4_GET(dev_cap->max_counters, outbox,
947 			 QUERY_DEV_CAP_MAX_COUNTERS_OFFSET);
948 
949 	MLX4_GET(field32, outbox,
950 		 QUERY_DEV_CAP_MAD_DEMUX_OFFSET);
951 	if (field32 & (1 << 0))
952 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_MAD_DEMUX;
953 
954 	MLX4_GET(dev_cap->dmfs_high_rate_qpn_base, outbox,
955 		 QUERY_DEV_CAP_DMFS_HIGH_RATE_QPN_BASE_OFFSET);
956 	dev_cap->dmfs_high_rate_qpn_base &= MGM_QPN_MASK;
957 	MLX4_GET(dev_cap->dmfs_high_rate_qpn_range, outbox,
958 		 QUERY_DEV_CAP_DMFS_HIGH_RATE_QPN_RANGE_OFFSET);
959 	dev_cap->dmfs_high_rate_qpn_range &= MGM_QPN_MASK;
960 
961 	MLX4_GET(size, outbox, QUERY_DEV_CAP_QP_RATE_LIMIT_NUM_OFFSET);
962 	dev_cap->rl_caps.num_rates = size;
963 	if (dev_cap->rl_caps.num_rates) {
964 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT;
965 		MLX4_GET(size, outbox, QUERY_DEV_CAP_QP_RATE_LIMIT_MAX_OFFSET);
966 		dev_cap->rl_caps.max_val  = size & 0xfff;
967 		dev_cap->rl_caps.max_unit = size >> 14;
968 		MLX4_GET(size, outbox, QUERY_DEV_CAP_QP_RATE_LIMIT_MIN_OFFSET);
969 		dev_cap->rl_caps.min_val  = size & 0xfff;
970 		dev_cap->rl_caps.min_unit = size >> 14;
971 	}
972 
973 	MLX4_GET(field32, outbox, QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET);
974 	if (field32 & (1 << 16))
975 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_UPDATE_QP;
976 	if (field32 & (1 << 18))
977 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_UPDATE_QP_SRC_CHECK_LB;
978 	if (field32 & (1 << 19))
979 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_LB_SRC_CHK;
980 	if (field32 & (1 << 26))
981 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_VLAN_CONTROL;
982 	if (field32 & (1 << 20))
983 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_FSM;
984 	if (field32 & (1 << 21))
985 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_80_VFS;
986 
987 	for (i = 1; i <= dev_cap->num_ports; i++) {
988 		err = mlx4_QUERY_PORT(dev, i, dev_cap->port_cap + i);
989 		if (err)
990 			goto out;
991 	}
992 
993 	/*
994 	 * Each UAR has 4 EQ doorbells; so if a UAR is reserved, then
995 	 * we can't use any EQs whose doorbell falls on that page,
996 	 * even if the EQ itself isn't reserved.
997 	 */
998 	if (dev_cap->num_sys_eqs == 0)
999 		dev_cap->reserved_eqs = max(dev_cap->reserved_uars * 4,
1000 					    dev_cap->reserved_eqs);
1001 	else
1002 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_SYS_EQS;
1003 
1004 out:
1005 	mlx4_free_cmd_mailbox(dev, mailbox);
1006 	return err;
1007 }
1008 
1009 void mlx4_dev_cap_dump(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
1010 {
1011 	if (dev_cap->bf_reg_size > 0)
1012 		mlx4_dbg(dev, "BlueFlame available (reg size %d, regs/page %d)\n",
1013 			 dev_cap->bf_reg_size, dev_cap->bf_regs_per_page);
1014 	else
1015 		mlx4_dbg(dev, "BlueFlame not available\n");
1016 
1017 	mlx4_dbg(dev, "Base MM extensions: flags %08x, rsvd L_Key %08x\n",
1018 		 dev_cap->bmme_flags, dev_cap->reserved_lkey);
1019 	mlx4_dbg(dev, "Max ICM size %lld MB\n",
1020 		 (unsigned long long) dev_cap->max_icm_sz >> 20);
1021 	mlx4_dbg(dev, "Max QPs: %d, reserved QPs: %d, entry size: %d\n",
1022 		 dev_cap->max_qps, dev_cap->reserved_qps, dev_cap->qpc_entry_sz);
1023 	mlx4_dbg(dev, "Max SRQs: %d, reserved SRQs: %d, entry size: %d\n",
1024 		 dev_cap->max_srqs, dev_cap->reserved_srqs, dev_cap->srq_entry_sz);
1025 	mlx4_dbg(dev, "Max CQs: %d, reserved CQs: %d, entry size: %d\n",
1026 		 dev_cap->max_cqs, dev_cap->reserved_cqs, dev_cap->cqc_entry_sz);
1027 	mlx4_dbg(dev, "Num sys EQs: %d, max EQs: %d, reserved EQs: %d, entry size: %d\n",
1028 		 dev_cap->num_sys_eqs, dev_cap->max_eqs, dev_cap->reserved_eqs,
1029 		 dev_cap->eqc_entry_sz);
1030 	mlx4_dbg(dev, "reserved MPTs: %d, reserved MTTs: %d\n",
1031 		 dev_cap->reserved_mrws, dev_cap->reserved_mtts);
1032 	mlx4_dbg(dev, "Max PDs: %d, reserved PDs: %d, reserved UARs: %d\n",
1033 		 dev_cap->max_pds, dev_cap->reserved_pds, dev_cap->reserved_uars);
1034 	mlx4_dbg(dev, "Max QP/MCG: %d, reserved MGMs: %d\n",
1035 		 dev_cap->max_pds, dev_cap->reserved_mgms);
1036 	mlx4_dbg(dev, "Max CQEs: %d, max WQEs: %d, max SRQ WQEs: %d\n",
1037 		 dev_cap->max_cq_sz, dev_cap->max_qp_sz, dev_cap->max_srq_sz);
1038 	mlx4_dbg(dev, "Local CA ACK delay: %d, max MTU: %d, port width cap: %d\n",
1039 		 dev_cap->local_ca_ack_delay, 128 << dev_cap->port_cap[1].ib_mtu,
1040 		 dev_cap->port_cap[1].max_port_width);
1041 	mlx4_dbg(dev, "Max SQ desc size: %d, max SQ S/G: %d\n",
1042 		 dev_cap->max_sq_desc_sz, dev_cap->max_sq_sg);
1043 	mlx4_dbg(dev, "Max RQ desc size: %d, max RQ S/G: %d\n",
1044 		 dev_cap->max_rq_desc_sz, dev_cap->max_rq_sg);
1045 	mlx4_dbg(dev, "Max GSO size: %d\n", dev_cap->max_gso_sz);
1046 	mlx4_dbg(dev, "Max counters: %d\n", dev_cap->max_counters);
1047 	mlx4_dbg(dev, "Max RSS Table size: %d\n", dev_cap->max_rss_tbl_sz);
1048 	mlx4_dbg(dev, "DMFS high rate steer QPn base: %d\n",
1049 		 dev_cap->dmfs_high_rate_qpn_base);
1050 	mlx4_dbg(dev, "DMFS high rate steer QPn range: %d\n",
1051 		 dev_cap->dmfs_high_rate_qpn_range);
1052 
1053 	if (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT) {
1054 		struct mlx4_rate_limit_caps *rl_caps = &dev_cap->rl_caps;
1055 
1056 		mlx4_dbg(dev, "QP Rate-Limit: #rates %d, unit/val max %d/%d, min %d/%d\n",
1057 			 rl_caps->num_rates, rl_caps->max_unit, rl_caps->max_val,
1058 			 rl_caps->min_unit, rl_caps->min_val);
1059 	}
1060 
1061 	dump_dev_cap_flags(dev, dev_cap->flags);
1062 	dump_dev_cap_flags2(dev, dev_cap->flags2);
1063 }
1064 
1065 int mlx4_QUERY_PORT(struct mlx4_dev *dev, int port, struct mlx4_port_cap *port_cap)
1066 {
1067 	struct mlx4_cmd_mailbox *mailbox;
1068 	u32 *outbox;
1069 	u8 field;
1070 	u32 field32;
1071 	int err;
1072 
1073 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1074 	if (IS_ERR(mailbox))
1075 		return PTR_ERR(mailbox);
1076 	outbox = mailbox->buf;
1077 
1078 	if (dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1079 		err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0, MLX4_CMD_QUERY_DEV_CAP,
1080 				   MLX4_CMD_TIME_CLASS_A,
1081 				   MLX4_CMD_NATIVE);
1082 
1083 		if (err)
1084 			goto out;
1085 
1086 		MLX4_GET(field, outbox, QUERY_DEV_CAP_VL_PORT_OFFSET);
1087 		port_cap->max_vl	   = field >> 4;
1088 		MLX4_GET(field, outbox, QUERY_DEV_CAP_MTU_WIDTH_OFFSET);
1089 		port_cap->ib_mtu	   = field >> 4;
1090 		port_cap->max_port_width = field & 0xf;
1091 		MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_GID_OFFSET);
1092 		port_cap->max_gids	   = 1 << (field & 0xf);
1093 		MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_PKEY_OFFSET);
1094 		port_cap->max_pkeys	   = 1 << (field & 0xf);
1095 	} else {
1096 #define QUERY_PORT_SUPPORTED_TYPE_OFFSET	0x00
1097 #define QUERY_PORT_MTU_OFFSET			0x01
1098 #define QUERY_PORT_ETH_MTU_OFFSET		0x02
1099 #define QUERY_PORT_WIDTH_OFFSET			0x06
1100 #define QUERY_PORT_MAX_GID_PKEY_OFFSET		0x07
1101 #define QUERY_PORT_MAX_MACVLAN_OFFSET		0x0a
1102 #define QUERY_PORT_MAX_VL_OFFSET		0x0b
1103 #define QUERY_PORT_MAC_OFFSET			0x10
1104 #define QUERY_PORT_TRANS_VENDOR_OFFSET		0x18
1105 #define QUERY_PORT_WAVELENGTH_OFFSET		0x1c
1106 #define QUERY_PORT_TRANS_CODE_OFFSET		0x20
1107 
1108 		err = mlx4_cmd_box(dev, 0, mailbox->dma, port, 0, MLX4_CMD_QUERY_PORT,
1109 				   MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
1110 		if (err)
1111 			goto out;
1112 
1113 		MLX4_GET(field, outbox, QUERY_PORT_SUPPORTED_TYPE_OFFSET);
1114 		port_cap->link_state = (field & 0x80) >> 7;
1115 		port_cap->supported_port_types = field & 3;
1116 		port_cap->suggested_type = (field >> 3) & 1;
1117 		port_cap->default_sense = (field >> 4) & 1;
1118 		port_cap->dmfs_optimized_state = (field >> 5) & 1;
1119 		MLX4_GET(field, outbox, QUERY_PORT_MTU_OFFSET);
1120 		port_cap->ib_mtu	   = field & 0xf;
1121 		MLX4_GET(field, outbox, QUERY_PORT_WIDTH_OFFSET);
1122 		port_cap->max_port_width = field & 0xf;
1123 		MLX4_GET(field, outbox, QUERY_PORT_MAX_GID_PKEY_OFFSET);
1124 		port_cap->max_gids	   = 1 << (field >> 4);
1125 		port_cap->max_pkeys	   = 1 << (field & 0xf);
1126 		MLX4_GET(field, outbox, QUERY_PORT_MAX_VL_OFFSET);
1127 		port_cap->max_vl	   = field & 0xf;
1128 		MLX4_GET(field, outbox, QUERY_PORT_MAX_MACVLAN_OFFSET);
1129 		port_cap->log_max_macs  = field & 0xf;
1130 		port_cap->log_max_vlans = field >> 4;
1131 		MLX4_GET(port_cap->eth_mtu, outbox, QUERY_PORT_ETH_MTU_OFFSET);
1132 		MLX4_GET(port_cap->def_mac, outbox, QUERY_PORT_MAC_OFFSET);
1133 		MLX4_GET(field32, outbox, QUERY_PORT_TRANS_VENDOR_OFFSET);
1134 		port_cap->trans_type = field32 >> 24;
1135 		port_cap->vendor_oui = field32 & 0xffffff;
1136 		MLX4_GET(port_cap->wavelength, outbox, QUERY_PORT_WAVELENGTH_OFFSET);
1137 		MLX4_GET(port_cap->trans_code, outbox, QUERY_PORT_TRANS_CODE_OFFSET);
1138 	}
1139 
1140 out:
1141 	mlx4_free_cmd_mailbox(dev, mailbox);
1142 	return err;
1143 }
1144 
1145 #define DEV_CAP_EXT_2_FLAG_PFC_COUNTERS	(1 << 28)
1146 #define DEV_CAP_EXT_2_FLAG_VLAN_CONTROL (1 << 26)
1147 #define DEV_CAP_EXT_2_FLAG_80_VFS	(1 << 21)
1148 #define DEV_CAP_EXT_2_FLAG_FSM		(1 << 20)
1149 
1150 int mlx4_QUERY_DEV_CAP_wrapper(struct mlx4_dev *dev, int slave,
1151 			       struct mlx4_vhcr *vhcr,
1152 			       struct mlx4_cmd_mailbox *inbox,
1153 			       struct mlx4_cmd_mailbox *outbox,
1154 			       struct mlx4_cmd_info *cmd)
1155 {
1156 	u64	flags;
1157 	int	err = 0;
1158 	u8	field;
1159 	u16	field16;
1160 	u32	bmme_flags, field32;
1161 	int	real_port;
1162 	int	slave_port;
1163 	int	first_port;
1164 	struct mlx4_active_ports actv_ports;
1165 
1166 	err = mlx4_cmd_box(dev, 0, outbox->dma, 0, 0, MLX4_CMD_QUERY_DEV_CAP,
1167 			   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1168 	if (err)
1169 		return err;
1170 
1171 	disable_unsupported_roce_caps(outbox->buf);
1172 	/* add port mng change event capability and disable mw type 1
1173 	 * unconditionally to slaves
1174 	 */
1175 	MLX4_GET(flags, outbox->buf, QUERY_DEV_CAP_EXT_FLAGS_OFFSET);
1176 	flags |= MLX4_DEV_CAP_FLAG_PORT_MNG_CHG_EV;
1177 	flags &= ~MLX4_DEV_CAP_FLAG_MEM_WINDOW;
1178 	actv_ports = mlx4_get_active_ports(dev, slave);
1179 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
1180 	for (slave_port = 0, real_port = first_port;
1181 	     real_port < first_port +
1182 	     bitmap_weight(actv_ports.ports, dev->caps.num_ports);
1183 	     ++real_port, ++slave_port) {
1184 		if (flags & (MLX4_DEV_CAP_FLAG_WOL_PORT1 << real_port))
1185 			flags |= MLX4_DEV_CAP_FLAG_WOL_PORT1 << slave_port;
1186 		else
1187 			flags &= ~(MLX4_DEV_CAP_FLAG_WOL_PORT1 << slave_port);
1188 	}
1189 	for (; slave_port < dev->caps.num_ports; ++slave_port)
1190 		flags &= ~(MLX4_DEV_CAP_FLAG_WOL_PORT1 << slave_port);
1191 
1192 	/* Not exposing RSS IP fragments to guests */
1193 	flags &= ~MLX4_DEV_CAP_FLAG_RSS_IP_FRAG;
1194 	MLX4_PUT(outbox->buf, flags, QUERY_DEV_CAP_EXT_FLAGS_OFFSET);
1195 
1196 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_VL_PORT_OFFSET);
1197 	field &= ~0x0F;
1198 	field |= bitmap_weight(actv_ports.ports, dev->caps.num_ports) & 0x0F;
1199 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_VL_PORT_OFFSET);
1200 
1201 	/* For guests, disable timestamp */
1202 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_CQ_TS_SUPPORT_OFFSET);
1203 	field &= 0x7f;
1204 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_CQ_TS_SUPPORT_OFFSET);
1205 
1206 	/* For guests, disable vxlan tunneling and QoS support */
1207 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_VXLAN);
1208 	field &= 0xd7;
1209 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_VXLAN);
1210 
1211 	/* For guests, disable port BEACON */
1212 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_PORT_BEACON_OFFSET);
1213 	field &= 0x7f;
1214 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_PORT_BEACON_OFFSET);
1215 
1216 	/* For guests, report Blueflame disabled */
1217 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_BF_OFFSET);
1218 	field &= 0x7f;
1219 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_BF_OFFSET);
1220 
1221 	/* For guests, disable mw type 2 and port remap*/
1222 	MLX4_GET(bmme_flags, outbox->buf, QUERY_DEV_CAP_BMME_FLAGS_OFFSET);
1223 	bmme_flags &= ~MLX4_BMME_FLAG_TYPE_2_WIN;
1224 	bmme_flags &= ~MLX4_FLAG_PORT_REMAP;
1225 	MLX4_PUT(outbox->buf, bmme_flags, QUERY_DEV_CAP_BMME_FLAGS_OFFSET);
1226 
1227 	/* turn off device-managed steering capability if not enabled */
1228 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) {
1229 		MLX4_GET(field, outbox->buf,
1230 			 QUERY_DEV_CAP_FLOW_STEERING_RANGE_EN_OFFSET);
1231 		field &= 0x7f;
1232 		MLX4_PUT(outbox->buf, field,
1233 			 QUERY_DEV_CAP_FLOW_STEERING_RANGE_EN_OFFSET);
1234 	}
1235 
1236 	/* turn off ipoib managed steering for guests */
1237 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_FLOW_STEERING_IPOIB_OFFSET);
1238 	field &= ~0x80;
1239 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_FLOW_STEERING_IPOIB_OFFSET);
1240 
1241 	/* turn off host side virt features (VST, FSM, etc) for guests */
1242 	MLX4_GET(field32, outbox->buf, QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET);
1243 	field32 &= ~(DEV_CAP_EXT_2_FLAG_VLAN_CONTROL | DEV_CAP_EXT_2_FLAG_80_VFS |
1244 		     DEV_CAP_EXT_2_FLAG_FSM | DEV_CAP_EXT_2_FLAG_PFC_COUNTERS);
1245 	MLX4_PUT(outbox->buf, field32, QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET);
1246 
1247 	/* turn off QCN for guests */
1248 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_ECN_QCN_VER_OFFSET);
1249 	field &= 0xfe;
1250 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_ECN_QCN_VER_OFFSET);
1251 
1252 	/* turn off QP max-rate limiting for guests */
1253 	field16 = 0;
1254 	MLX4_PUT(outbox->buf, field16, QUERY_DEV_CAP_QP_RATE_LIMIT_NUM_OFFSET);
1255 
1256 	/* turn off QoS per VF support for guests */
1257 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_CQ_EQ_CACHE_LINE_STRIDE);
1258 	field &= 0xef;
1259 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_CQ_EQ_CACHE_LINE_STRIDE);
1260 
1261 	/* turn off ignore FCS feature for guests */
1262 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_CONFIG_DEV_OFFSET);
1263 	field &= 0xfb;
1264 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_CONFIG_DEV_OFFSET);
1265 
1266 	return 0;
1267 }
1268 
1269 static void disable_unsupported_roce_caps(void *buf)
1270 {
1271 	u32 flags;
1272 
1273 	MLX4_GET(flags, buf, QUERY_DEV_CAP_EXT_FLAGS_OFFSET);
1274 	flags &= ~(1UL << 31);
1275 	MLX4_PUT(buf, flags, QUERY_DEV_CAP_EXT_FLAGS_OFFSET);
1276 	MLX4_GET(flags, buf, QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET);
1277 	flags &= ~(1UL << 24);
1278 	MLX4_PUT(buf, flags, QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET);
1279 	MLX4_GET(flags, buf, QUERY_DEV_CAP_BMME_FLAGS_OFFSET);
1280 	flags &= ~(MLX4_FLAG_ROCE_V1_V2);
1281 	MLX4_PUT(buf, flags, QUERY_DEV_CAP_BMME_FLAGS_OFFSET);
1282 }
1283 
1284 int mlx4_QUERY_PORT_wrapper(struct mlx4_dev *dev, int slave,
1285 			    struct mlx4_vhcr *vhcr,
1286 			    struct mlx4_cmd_mailbox *inbox,
1287 			    struct mlx4_cmd_mailbox *outbox,
1288 			    struct mlx4_cmd_info *cmd)
1289 {
1290 	struct mlx4_priv *priv = mlx4_priv(dev);
1291 	u64 def_mac;
1292 	u8 port_type;
1293 	u16 short_field;
1294 	int err;
1295 	int admin_link_state;
1296 	int port = mlx4_slave_convert_port(dev, slave,
1297 					   vhcr->in_modifier & 0xFF);
1298 
1299 #define MLX4_VF_PORT_NO_LINK_SENSE_MASK	0xE0
1300 #define MLX4_PORT_LINK_UP_MASK		0x80
1301 #define QUERY_PORT_CUR_MAX_PKEY_OFFSET	0x0c
1302 #define QUERY_PORT_CUR_MAX_GID_OFFSET	0x0e
1303 
1304 	if (port < 0)
1305 		return -EINVAL;
1306 
1307 	/* Protect against untrusted guests: enforce that this is the
1308 	 * QUERY_PORT general query.
1309 	 */
1310 	if (vhcr->op_modifier || vhcr->in_modifier & ~0xFF)
1311 		return -EINVAL;
1312 
1313 	vhcr->in_modifier = port;
1314 
1315 	err = mlx4_cmd_box(dev, 0, outbox->dma, vhcr->in_modifier, 0,
1316 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
1317 			   MLX4_CMD_NATIVE);
1318 
1319 	if (!err && dev->caps.function != slave) {
1320 		def_mac = priv->mfunc.master.vf_oper[slave].vport[vhcr->in_modifier].state.mac;
1321 		MLX4_PUT(outbox->buf, def_mac, QUERY_PORT_MAC_OFFSET);
1322 
1323 		/* get port type - currently only eth is enabled */
1324 		MLX4_GET(port_type, outbox->buf,
1325 			 QUERY_PORT_SUPPORTED_TYPE_OFFSET);
1326 
1327 		/* No link sensing allowed */
1328 		port_type &= MLX4_VF_PORT_NO_LINK_SENSE_MASK;
1329 		/* set port type to currently operating port type */
1330 		port_type |= (dev->caps.port_type[vhcr->in_modifier] & 0x3);
1331 
1332 		admin_link_state = priv->mfunc.master.vf_oper[slave].vport[vhcr->in_modifier].state.link_state;
1333 		if (IFLA_VF_LINK_STATE_ENABLE == admin_link_state)
1334 			port_type |= MLX4_PORT_LINK_UP_MASK;
1335 		else if (IFLA_VF_LINK_STATE_DISABLE == admin_link_state)
1336 			port_type &= ~MLX4_PORT_LINK_UP_MASK;
1337 		else if (IFLA_VF_LINK_STATE_AUTO == admin_link_state && mlx4_is_bonded(dev)) {
1338 			int other_port = (port == 1) ? 2 : 1;
1339 			struct mlx4_port_cap port_cap;
1340 
1341 			err = mlx4_QUERY_PORT(dev, other_port, &port_cap);
1342 			if (err)
1343 				goto out;
1344 			port_type |= (port_cap.link_state << 7);
1345 		}
1346 
1347 		MLX4_PUT(outbox->buf, port_type,
1348 			 QUERY_PORT_SUPPORTED_TYPE_OFFSET);
1349 
1350 		if (dev->caps.port_type[vhcr->in_modifier] == MLX4_PORT_TYPE_ETH)
1351 			short_field = mlx4_get_slave_num_gids(dev, slave, port);
1352 		else
1353 			short_field = 1; /* slave max gids */
1354 		MLX4_PUT(outbox->buf, short_field,
1355 			 QUERY_PORT_CUR_MAX_GID_OFFSET);
1356 
1357 		short_field = dev->caps.pkey_table_len[vhcr->in_modifier];
1358 		MLX4_PUT(outbox->buf, short_field,
1359 			 QUERY_PORT_CUR_MAX_PKEY_OFFSET);
1360 	}
1361 out:
1362 	return err;
1363 }
1364 
1365 int mlx4_get_slave_pkey_gid_tbl_len(struct mlx4_dev *dev, u8 port,
1366 				    int *gid_tbl_len, int *pkey_tbl_len)
1367 {
1368 	struct mlx4_cmd_mailbox *mailbox;
1369 	u32			*outbox;
1370 	u16			field;
1371 	int			err;
1372 
1373 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1374 	if (IS_ERR(mailbox))
1375 		return PTR_ERR(mailbox);
1376 
1377 	err =  mlx4_cmd_box(dev, 0, mailbox->dma, port, 0,
1378 			    MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
1379 			    MLX4_CMD_WRAPPED);
1380 	if (err)
1381 		goto out;
1382 
1383 	outbox = mailbox->buf;
1384 
1385 	MLX4_GET(field, outbox, QUERY_PORT_CUR_MAX_GID_OFFSET);
1386 	*gid_tbl_len = field;
1387 
1388 	MLX4_GET(field, outbox, QUERY_PORT_CUR_MAX_PKEY_OFFSET);
1389 	*pkey_tbl_len = field;
1390 
1391 out:
1392 	mlx4_free_cmd_mailbox(dev, mailbox);
1393 	return err;
1394 }
1395 EXPORT_SYMBOL(mlx4_get_slave_pkey_gid_tbl_len);
1396 
1397 int mlx4_map_cmd(struct mlx4_dev *dev, u16 op, struct mlx4_icm *icm, u64 virt)
1398 {
1399 	struct mlx4_cmd_mailbox *mailbox;
1400 	struct mlx4_icm_iter iter;
1401 	__be64 *pages;
1402 	int lg;
1403 	int nent = 0;
1404 	int i;
1405 	int err = 0;
1406 	int ts = 0, tc = 0;
1407 
1408 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1409 	if (IS_ERR(mailbox))
1410 		return PTR_ERR(mailbox);
1411 	pages = mailbox->buf;
1412 
1413 	for (mlx4_icm_first(icm, &iter);
1414 	     !mlx4_icm_last(&iter);
1415 	     mlx4_icm_next(&iter)) {
1416 		/*
1417 		 * We have to pass pages that are aligned to their
1418 		 * size, so find the least significant 1 in the
1419 		 * address or size and use that as our log2 size.
1420 		 */
1421 		lg = ffs(mlx4_icm_addr(&iter) | mlx4_icm_size(&iter)) - 1;
1422 		if (lg < MLX4_ICM_PAGE_SHIFT) {
1423 			mlx4_warn(dev, "Got FW area not aligned to %d (%llx/%lx)\n",
1424 				  MLX4_ICM_PAGE_SIZE,
1425 				  (unsigned long long) mlx4_icm_addr(&iter),
1426 				  mlx4_icm_size(&iter));
1427 			err = -EINVAL;
1428 			goto out;
1429 		}
1430 
1431 		for (i = 0; i < mlx4_icm_size(&iter) >> lg; ++i) {
1432 			if (virt != -1) {
1433 				pages[nent * 2] = cpu_to_be64(virt);
1434 				virt += 1 << lg;
1435 			}
1436 
1437 			pages[nent * 2 + 1] =
1438 				cpu_to_be64((mlx4_icm_addr(&iter) + (i << lg)) |
1439 					    (lg - MLX4_ICM_PAGE_SHIFT));
1440 			ts += 1 << (lg - 10);
1441 			++tc;
1442 
1443 			if (++nent == MLX4_MAILBOX_SIZE / 16) {
1444 				err = mlx4_cmd(dev, mailbox->dma, nent, 0, op,
1445 						MLX4_CMD_TIME_CLASS_B,
1446 						MLX4_CMD_NATIVE);
1447 				if (err)
1448 					goto out;
1449 				nent = 0;
1450 			}
1451 		}
1452 	}
1453 
1454 	if (nent)
1455 		err = mlx4_cmd(dev, mailbox->dma, nent, 0, op,
1456 			       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
1457 	if (err)
1458 		goto out;
1459 
1460 	switch (op) {
1461 	case MLX4_CMD_MAP_FA:
1462 		mlx4_dbg(dev, "Mapped %d chunks/%d KB for FW\n", tc, ts);
1463 		break;
1464 	case MLX4_CMD_MAP_ICM_AUX:
1465 		mlx4_dbg(dev, "Mapped %d chunks/%d KB for ICM aux\n", tc, ts);
1466 		break;
1467 	case MLX4_CMD_MAP_ICM:
1468 		mlx4_dbg(dev, "Mapped %d chunks/%d KB at %llx for ICM\n",
1469 			 tc, ts, (unsigned long long) virt - (ts << 10));
1470 		break;
1471 	}
1472 
1473 out:
1474 	mlx4_free_cmd_mailbox(dev, mailbox);
1475 	return err;
1476 }
1477 
1478 int mlx4_MAP_FA(struct mlx4_dev *dev, struct mlx4_icm *icm)
1479 {
1480 	return mlx4_map_cmd(dev, MLX4_CMD_MAP_FA, icm, -1);
1481 }
1482 
1483 int mlx4_UNMAP_FA(struct mlx4_dev *dev)
1484 {
1485 	return mlx4_cmd(dev, 0, 0, 0, MLX4_CMD_UNMAP_FA,
1486 			MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
1487 }
1488 
1489 
1490 int mlx4_RUN_FW(struct mlx4_dev *dev)
1491 {
1492 	return mlx4_cmd(dev, 0, 0, 0, MLX4_CMD_RUN_FW,
1493 			MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1494 }
1495 
1496 int mlx4_QUERY_FW(struct mlx4_dev *dev)
1497 {
1498 	struct mlx4_fw  *fw  = &mlx4_priv(dev)->fw;
1499 	struct mlx4_cmd *cmd = &mlx4_priv(dev)->cmd;
1500 	struct mlx4_cmd_mailbox *mailbox;
1501 	u32 *outbox;
1502 	int err = 0;
1503 	u64 fw_ver;
1504 	u16 cmd_if_rev;
1505 	u8 lg;
1506 
1507 #define QUERY_FW_OUT_SIZE             0x100
1508 #define QUERY_FW_VER_OFFSET            0x00
1509 #define QUERY_FW_PPF_ID		       0x09
1510 #define QUERY_FW_CMD_IF_REV_OFFSET     0x0a
1511 #define QUERY_FW_MAX_CMD_OFFSET        0x0f
1512 #define QUERY_FW_ERR_START_OFFSET      0x30
1513 #define QUERY_FW_ERR_SIZE_OFFSET       0x38
1514 #define QUERY_FW_ERR_BAR_OFFSET        0x3c
1515 
1516 #define QUERY_FW_SIZE_OFFSET           0x00
1517 #define QUERY_FW_CLR_INT_BASE_OFFSET   0x20
1518 #define QUERY_FW_CLR_INT_BAR_OFFSET    0x28
1519 
1520 #define QUERY_FW_COMM_BASE_OFFSET      0x40
1521 #define QUERY_FW_COMM_BAR_OFFSET       0x48
1522 
1523 #define QUERY_FW_CLOCK_OFFSET	       0x50
1524 #define QUERY_FW_CLOCK_BAR	       0x58
1525 
1526 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1527 	if (IS_ERR(mailbox))
1528 		return PTR_ERR(mailbox);
1529 	outbox = mailbox->buf;
1530 
1531 	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0, MLX4_CMD_QUERY_FW,
1532 			    MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1533 	if (err)
1534 		goto out;
1535 
1536 	MLX4_GET(fw_ver, outbox, QUERY_FW_VER_OFFSET);
1537 	/*
1538 	 * FW subminor version is at more significant bits than minor
1539 	 * version, so swap here.
1540 	 */
1541 	dev->caps.fw_ver = (fw_ver & 0xffff00000000ull) |
1542 		((fw_ver & 0xffff0000ull) >> 16) |
1543 		((fw_ver & 0x0000ffffull) << 16);
1544 
1545 	MLX4_GET(lg, outbox, QUERY_FW_PPF_ID);
1546 	dev->caps.function = lg;
1547 
1548 	if (mlx4_is_slave(dev))
1549 		goto out;
1550 
1551 
1552 	MLX4_GET(cmd_if_rev, outbox, QUERY_FW_CMD_IF_REV_OFFSET);
1553 	if (cmd_if_rev < MLX4_COMMAND_INTERFACE_MIN_REV ||
1554 	    cmd_if_rev > MLX4_COMMAND_INTERFACE_MAX_REV) {
1555 		mlx4_err(dev, "Installed FW has unsupported command interface revision %d\n",
1556 			 cmd_if_rev);
1557 		mlx4_err(dev, "(Installed FW version is %d.%d.%03d)\n",
1558 			 (int) (dev->caps.fw_ver >> 32),
1559 			 (int) (dev->caps.fw_ver >> 16) & 0xffff,
1560 			 (int) dev->caps.fw_ver & 0xffff);
1561 		mlx4_err(dev, "This driver version supports only revisions %d to %d\n",
1562 			 MLX4_COMMAND_INTERFACE_MIN_REV, MLX4_COMMAND_INTERFACE_MAX_REV);
1563 		err = -ENODEV;
1564 		goto out;
1565 	}
1566 
1567 	if (cmd_if_rev < MLX4_COMMAND_INTERFACE_NEW_PORT_CMDS)
1568 		dev->flags |= MLX4_FLAG_OLD_PORT_CMDS;
1569 
1570 	MLX4_GET(lg, outbox, QUERY_FW_MAX_CMD_OFFSET);
1571 	cmd->max_cmds = 1 << lg;
1572 
1573 	mlx4_dbg(dev, "FW version %d.%d.%03d (cmd intf rev %d), max commands %d\n",
1574 		 (int) (dev->caps.fw_ver >> 32),
1575 		 (int) (dev->caps.fw_ver >> 16) & 0xffff,
1576 		 (int) dev->caps.fw_ver & 0xffff,
1577 		 cmd_if_rev, cmd->max_cmds);
1578 
1579 	MLX4_GET(fw->catas_offset, outbox, QUERY_FW_ERR_START_OFFSET);
1580 	MLX4_GET(fw->catas_size,   outbox, QUERY_FW_ERR_SIZE_OFFSET);
1581 	MLX4_GET(fw->catas_bar,    outbox, QUERY_FW_ERR_BAR_OFFSET);
1582 	fw->catas_bar = (fw->catas_bar >> 6) * 2;
1583 
1584 	mlx4_dbg(dev, "Catastrophic error buffer at 0x%llx, size 0x%x, BAR %d\n",
1585 		 (unsigned long long) fw->catas_offset, fw->catas_size, fw->catas_bar);
1586 
1587 	MLX4_GET(fw->fw_pages,     outbox, QUERY_FW_SIZE_OFFSET);
1588 	MLX4_GET(fw->clr_int_base, outbox, QUERY_FW_CLR_INT_BASE_OFFSET);
1589 	MLX4_GET(fw->clr_int_bar,  outbox, QUERY_FW_CLR_INT_BAR_OFFSET);
1590 	fw->clr_int_bar = (fw->clr_int_bar >> 6) * 2;
1591 
1592 	MLX4_GET(fw->comm_base, outbox, QUERY_FW_COMM_BASE_OFFSET);
1593 	MLX4_GET(fw->comm_bar,  outbox, QUERY_FW_COMM_BAR_OFFSET);
1594 	fw->comm_bar = (fw->comm_bar >> 6) * 2;
1595 	mlx4_dbg(dev, "Communication vector bar:%d offset:0x%llx\n",
1596 		 fw->comm_bar, fw->comm_base);
1597 	mlx4_dbg(dev, "FW size %d KB\n", fw->fw_pages >> 2);
1598 
1599 	MLX4_GET(fw->clock_offset, outbox, QUERY_FW_CLOCK_OFFSET);
1600 	MLX4_GET(fw->clock_bar,    outbox, QUERY_FW_CLOCK_BAR);
1601 	fw->clock_bar = (fw->clock_bar >> 6) * 2;
1602 	mlx4_dbg(dev, "Internal clock bar:%d offset:0x%llx\n",
1603 		 fw->clock_bar, fw->clock_offset);
1604 
1605 	/*
1606 	 * Round up number of system pages needed in case
1607 	 * MLX4_ICM_PAGE_SIZE < PAGE_SIZE.
1608 	 */
1609 	fw->fw_pages =
1610 		ALIGN(fw->fw_pages, PAGE_SIZE / MLX4_ICM_PAGE_SIZE) >>
1611 		(PAGE_SHIFT - MLX4_ICM_PAGE_SHIFT);
1612 
1613 	mlx4_dbg(dev, "Clear int @ %llx, BAR %d\n",
1614 		 (unsigned long long) fw->clr_int_base, fw->clr_int_bar);
1615 
1616 out:
1617 	mlx4_free_cmd_mailbox(dev, mailbox);
1618 	return err;
1619 }
1620 
1621 int mlx4_QUERY_FW_wrapper(struct mlx4_dev *dev, int slave,
1622 			  struct mlx4_vhcr *vhcr,
1623 			  struct mlx4_cmd_mailbox *inbox,
1624 			  struct mlx4_cmd_mailbox *outbox,
1625 			  struct mlx4_cmd_info *cmd)
1626 {
1627 	u8 *outbuf;
1628 	int err;
1629 
1630 	outbuf = outbox->buf;
1631 	err = mlx4_cmd_box(dev, 0, outbox->dma, 0, 0, MLX4_CMD_QUERY_FW,
1632 			    MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1633 	if (err)
1634 		return err;
1635 
1636 	/* for slaves, set pci PPF ID to invalid and zero out everything
1637 	 * else except FW version */
1638 	outbuf[0] = outbuf[1] = 0;
1639 	memset(&outbuf[8], 0, QUERY_FW_OUT_SIZE - 8);
1640 	outbuf[QUERY_FW_PPF_ID] = MLX4_INVALID_SLAVE_ID;
1641 
1642 	return 0;
1643 }
1644 
1645 static void get_board_id(void *vsd, char *board_id)
1646 {
1647 	int i;
1648 
1649 #define VSD_OFFSET_SIG1		0x00
1650 #define VSD_OFFSET_SIG2		0xde
1651 #define VSD_OFFSET_MLX_BOARD_ID	0xd0
1652 #define VSD_OFFSET_TS_BOARD_ID	0x20
1653 
1654 #define VSD_SIGNATURE_TOPSPIN	0x5ad
1655 
1656 	memset(board_id, 0, MLX4_BOARD_ID_LEN);
1657 
1658 	if (be16_to_cpup(vsd + VSD_OFFSET_SIG1) == VSD_SIGNATURE_TOPSPIN &&
1659 	    be16_to_cpup(vsd + VSD_OFFSET_SIG2) == VSD_SIGNATURE_TOPSPIN) {
1660 		strlcpy(board_id, vsd + VSD_OFFSET_TS_BOARD_ID, MLX4_BOARD_ID_LEN);
1661 	} else {
1662 		/*
1663 		 * The board ID is a string but the firmware byte
1664 		 * swaps each 4-byte word before passing it back to
1665 		 * us.  Therefore we need to swab it before printing.
1666 		 */
1667 		u32 *bid_u32 = (u32 *)board_id;
1668 
1669 		for (i = 0; i < 4; ++i) {
1670 			u32 *addr;
1671 			u32 val;
1672 
1673 			addr = (u32 *) (vsd + VSD_OFFSET_MLX_BOARD_ID + i * 4);
1674 			val = get_unaligned(addr);
1675 			val = swab32(val);
1676 			put_unaligned(val, &bid_u32[i]);
1677 		}
1678 	}
1679 }
1680 
1681 int mlx4_QUERY_ADAPTER(struct mlx4_dev *dev, struct mlx4_adapter *adapter)
1682 {
1683 	struct mlx4_cmd_mailbox *mailbox;
1684 	u32 *outbox;
1685 	int err;
1686 
1687 #define QUERY_ADAPTER_OUT_SIZE             0x100
1688 #define QUERY_ADAPTER_INTA_PIN_OFFSET      0x10
1689 #define QUERY_ADAPTER_VSD_OFFSET           0x20
1690 
1691 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1692 	if (IS_ERR(mailbox))
1693 		return PTR_ERR(mailbox);
1694 	outbox = mailbox->buf;
1695 
1696 	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0, MLX4_CMD_QUERY_ADAPTER,
1697 			   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1698 	if (err)
1699 		goto out;
1700 
1701 	MLX4_GET(adapter->inta_pin, outbox,    QUERY_ADAPTER_INTA_PIN_OFFSET);
1702 
1703 	get_board_id(outbox + QUERY_ADAPTER_VSD_OFFSET / 4,
1704 		     adapter->board_id);
1705 
1706 out:
1707 	mlx4_free_cmd_mailbox(dev, mailbox);
1708 	return err;
1709 }
1710 
1711 int mlx4_INIT_HCA(struct mlx4_dev *dev, struct mlx4_init_hca_param *param)
1712 {
1713 	struct mlx4_cmd_mailbox *mailbox;
1714 	__be32 *inbox;
1715 	int err;
1716 	static const u8 a0_dmfs_hw_steering[] =  {
1717 		[MLX4_STEERING_DMFS_A0_DEFAULT]		= 0,
1718 		[MLX4_STEERING_DMFS_A0_DYNAMIC]		= 1,
1719 		[MLX4_STEERING_DMFS_A0_STATIC]		= 2,
1720 		[MLX4_STEERING_DMFS_A0_DISABLE]		= 3
1721 	};
1722 
1723 #define INIT_HCA_IN_SIZE		 0x200
1724 #define INIT_HCA_VERSION_OFFSET		 0x000
1725 #define	 INIT_HCA_VERSION		 2
1726 #define INIT_HCA_VXLAN_OFFSET		 0x0c
1727 #define INIT_HCA_CACHELINE_SZ_OFFSET	 0x0e
1728 #define INIT_HCA_FLAGS_OFFSET		 0x014
1729 #define INIT_HCA_RECOVERABLE_ERROR_EVENT_OFFSET 0x018
1730 #define INIT_HCA_QPC_OFFSET		 0x020
1731 #define	 INIT_HCA_QPC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x10)
1732 #define	 INIT_HCA_LOG_QP_OFFSET		 (INIT_HCA_QPC_OFFSET + 0x17)
1733 #define	 INIT_HCA_SRQC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x28)
1734 #define	 INIT_HCA_LOG_SRQ_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x2f)
1735 #define	 INIT_HCA_CQC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x30)
1736 #define	 INIT_HCA_LOG_CQ_OFFSET		 (INIT_HCA_QPC_OFFSET + 0x37)
1737 #define	 INIT_HCA_EQE_CQE_OFFSETS	 (INIT_HCA_QPC_OFFSET + 0x38)
1738 #define	 INIT_HCA_EQE_CQE_STRIDE_OFFSET  (INIT_HCA_QPC_OFFSET + 0x3b)
1739 #define	 INIT_HCA_ALTC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x40)
1740 #define	 INIT_HCA_AUXC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x50)
1741 #define	 INIT_HCA_EQC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x60)
1742 #define	 INIT_HCA_LOG_EQ_OFFSET		 (INIT_HCA_QPC_OFFSET + 0x67)
1743 #define	INIT_HCA_NUM_SYS_EQS_OFFSET	(INIT_HCA_QPC_OFFSET + 0x6a)
1744 #define	 INIT_HCA_RDMARC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x70)
1745 #define	 INIT_HCA_LOG_RD_OFFSET		 (INIT_HCA_QPC_OFFSET + 0x77)
1746 #define INIT_HCA_MCAST_OFFSET		 0x0c0
1747 #define	 INIT_HCA_MC_BASE_OFFSET	 (INIT_HCA_MCAST_OFFSET + 0x00)
1748 #define	 INIT_HCA_LOG_MC_ENTRY_SZ_OFFSET (INIT_HCA_MCAST_OFFSET + 0x12)
1749 #define	 INIT_HCA_LOG_MC_HASH_SZ_OFFSET	 (INIT_HCA_MCAST_OFFSET + 0x16)
1750 #define  INIT_HCA_UC_STEERING_OFFSET	 (INIT_HCA_MCAST_OFFSET + 0x18)
1751 #define	 INIT_HCA_LOG_MC_TABLE_SZ_OFFSET (INIT_HCA_MCAST_OFFSET + 0x1b)
1752 #define  INIT_HCA_DEVICE_MANAGED_FLOW_STEERING_EN	0x6
1753 #define  INIT_HCA_FS_PARAM_OFFSET         0x1d0
1754 #define  INIT_HCA_FS_BASE_OFFSET          (INIT_HCA_FS_PARAM_OFFSET + 0x00)
1755 #define  INIT_HCA_FS_LOG_ENTRY_SZ_OFFSET  (INIT_HCA_FS_PARAM_OFFSET + 0x12)
1756 #define  INIT_HCA_FS_A0_OFFSET		  (INIT_HCA_FS_PARAM_OFFSET + 0x18)
1757 #define  INIT_HCA_FS_LOG_TABLE_SZ_OFFSET  (INIT_HCA_FS_PARAM_OFFSET + 0x1b)
1758 #define  INIT_HCA_FS_ETH_BITS_OFFSET      (INIT_HCA_FS_PARAM_OFFSET + 0x21)
1759 #define  INIT_HCA_FS_ETH_NUM_ADDRS_OFFSET (INIT_HCA_FS_PARAM_OFFSET + 0x22)
1760 #define  INIT_HCA_FS_IB_BITS_OFFSET       (INIT_HCA_FS_PARAM_OFFSET + 0x25)
1761 #define  INIT_HCA_FS_IB_NUM_ADDRS_OFFSET  (INIT_HCA_FS_PARAM_OFFSET + 0x26)
1762 #define INIT_HCA_TPT_OFFSET		 0x0f0
1763 #define	 INIT_HCA_DMPT_BASE_OFFSET	 (INIT_HCA_TPT_OFFSET + 0x00)
1764 #define  INIT_HCA_TPT_MW_OFFSET		 (INIT_HCA_TPT_OFFSET + 0x08)
1765 #define	 INIT_HCA_LOG_MPT_SZ_OFFSET	 (INIT_HCA_TPT_OFFSET + 0x0b)
1766 #define	 INIT_HCA_MTT_BASE_OFFSET	 (INIT_HCA_TPT_OFFSET + 0x10)
1767 #define	 INIT_HCA_CMPT_BASE_OFFSET	 (INIT_HCA_TPT_OFFSET + 0x18)
1768 #define INIT_HCA_UAR_OFFSET		 0x120
1769 #define	 INIT_HCA_LOG_UAR_SZ_OFFSET	 (INIT_HCA_UAR_OFFSET + 0x0a)
1770 #define  INIT_HCA_UAR_PAGE_SZ_OFFSET     (INIT_HCA_UAR_OFFSET + 0x0b)
1771 
1772 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1773 	if (IS_ERR(mailbox))
1774 		return PTR_ERR(mailbox);
1775 	inbox = mailbox->buf;
1776 
1777 	*((u8 *) mailbox->buf + INIT_HCA_VERSION_OFFSET) = INIT_HCA_VERSION;
1778 
1779 	*((u8 *) mailbox->buf + INIT_HCA_CACHELINE_SZ_OFFSET) =
1780 		(ilog2(cache_line_size()) - 4) << 5;
1781 
1782 #if defined(__LITTLE_ENDIAN)
1783 	*(inbox + INIT_HCA_FLAGS_OFFSET / 4) &= ~cpu_to_be32(1 << 1);
1784 #elif defined(__BIG_ENDIAN)
1785 	*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1 << 1);
1786 #else
1787 #error Host endianness not defined
1788 #endif
1789 	/* Check port for UD address vector: */
1790 	*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1);
1791 
1792 	/* Enable IPoIB checksumming if we can: */
1793 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
1794 		*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1 << 3);
1795 
1796 	/* Enable QoS support if module parameter set */
1797 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG && enable_qos)
1798 		*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1 << 2);
1799 
1800 	/* enable counters */
1801 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS)
1802 		*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1 << 4);
1803 
1804 	/* Enable RSS spread to fragmented IP packets when supported */
1805 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_RSS_IP_FRAG)
1806 		*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1 << 13);
1807 
1808 	/* CX3 is capable of extending CQEs/EQEs from 32 to 64 bytes */
1809 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_64B_EQE) {
1810 		*(inbox + INIT_HCA_EQE_CQE_OFFSETS / 4) |= cpu_to_be32(1 << 29);
1811 		dev->caps.eqe_size   = 64;
1812 		dev->caps.eqe_factor = 1;
1813 	} else {
1814 		dev->caps.eqe_size   = 32;
1815 		dev->caps.eqe_factor = 0;
1816 	}
1817 
1818 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_64B_CQE) {
1819 		*(inbox + INIT_HCA_EQE_CQE_OFFSETS / 4) |= cpu_to_be32(1 << 30);
1820 		dev->caps.cqe_size   = 64;
1821 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1822 	} else {
1823 		dev->caps.cqe_size   = 32;
1824 	}
1825 
1826 	/* CX3 is capable of extending CQEs\EQEs to strides larger than 64B */
1827 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) &&
1828 	    (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE)) {
1829 		dev->caps.eqe_size = cache_line_size();
1830 		dev->caps.cqe_size = cache_line_size();
1831 		dev->caps.eqe_factor = 0;
1832 		MLX4_PUT(inbox, (u8)((ilog2(dev->caps.eqe_size) - 5) << 4 |
1833 				      (ilog2(dev->caps.eqe_size) - 5)),
1834 			 INIT_HCA_EQE_CQE_STRIDE_OFFSET);
1835 
1836 		/* User still need to know to support CQE > 32B */
1837 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1838 	}
1839 
1840 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RECOVERABLE_ERROR_EVENT)
1841 		*(inbox + INIT_HCA_RECOVERABLE_ERROR_EVENT_OFFSET / 4) |= cpu_to_be32(1 << 31);
1842 
1843 	/* QPC/EEC/CQC/EQC/RDMARC attributes */
1844 
1845 	MLX4_PUT(inbox, param->qpc_base,      INIT_HCA_QPC_BASE_OFFSET);
1846 	MLX4_PUT(inbox, param->log_num_qps,   INIT_HCA_LOG_QP_OFFSET);
1847 	MLX4_PUT(inbox, param->srqc_base,     INIT_HCA_SRQC_BASE_OFFSET);
1848 	MLX4_PUT(inbox, param->log_num_srqs,  INIT_HCA_LOG_SRQ_OFFSET);
1849 	MLX4_PUT(inbox, param->cqc_base,      INIT_HCA_CQC_BASE_OFFSET);
1850 	MLX4_PUT(inbox, param->log_num_cqs,   INIT_HCA_LOG_CQ_OFFSET);
1851 	MLX4_PUT(inbox, param->altc_base,     INIT_HCA_ALTC_BASE_OFFSET);
1852 	MLX4_PUT(inbox, param->auxc_base,     INIT_HCA_AUXC_BASE_OFFSET);
1853 	MLX4_PUT(inbox, param->eqc_base,      INIT_HCA_EQC_BASE_OFFSET);
1854 	MLX4_PUT(inbox, param->log_num_eqs,   INIT_HCA_LOG_EQ_OFFSET);
1855 	MLX4_PUT(inbox, param->num_sys_eqs,   INIT_HCA_NUM_SYS_EQS_OFFSET);
1856 	MLX4_PUT(inbox, param->rdmarc_base,   INIT_HCA_RDMARC_BASE_OFFSET);
1857 	MLX4_PUT(inbox, param->log_rd_per_qp, INIT_HCA_LOG_RD_OFFSET);
1858 
1859 	/* steering attributes */
1860 	if (dev->caps.steering_mode ==
1861 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1862 		*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |=
1863 			cpu_to_be32(1 <<
1864 				    INIT_HCA_DEVICE_MANAGED_FLOW_STEERING_EN);
1865 
1866 		MLX4_PUT(inbox, param->mc_base, INIT_HCA_FS_BASE_OFFSET);
1867 		MLX4_PUT(inbox, param->log_mc_entry_sz,
1868 			 INIT_HCA_FS_LOG_ENTRY_SZ_OFFSET);
1869 		MLX4_PUT(inbox, param->log_mc_table_sz,
1870 			 INIT_HCA_FS_LOG_TABLE_SZ_OFFSET);
1871 		/* Enable Ethernet flow steering
1872 		 * with udp unicast and tcp unicast
1873 		 */
1874 		if (dev->caps.dmfs_high_steer_mode !=
1875 		    MLX4_STEERING_DMFS_A0_STATIC)
1876 			MLX4_PUT(inbox,
1877 				 (u8)(MLX4_FS_UDP_UC_EN | MLX4_FS_TCP_UC_EN),
1878 				 INIT_HCA_FS_ETH_BITS_OFFSET);
1879 		MLX4_PUT(inbox, (u16) MLX4_FS_NUM_OF_L2_ADDR,
1880 			 INIT_HCA_FS_ETH_NUM_ADDRS_OFFSET);
1881 		/* Enable IPoIB flow steering
1882 		 * with udp unicast and tcp unicast
1883 		 */
1884 		MLX4_PUT(inbox, (u8) (MLX4_FS_UDP_UC_EN | MLX4_FS_TCP_UC_EN),
1885 			 INIT_HCA_FS_IB_BITS_OFFSET);
1886 		MLX4_PUT(inbox, (u16) MLX4_FS_NUM_OF_L2_ADDR,
1887 			 INIT_HCA_FS_IB_NUM_ADDRS_OFFSET);
1888 
1889 		if (dev->caps.dmfs_high_steer_mode !=
1890 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1891 			MLX4_PUT(inbox,
1892 				 ((u8)(a0_dmfs_hw_steering[dev->caps.dmfs_high_steer_mode]
1893 				       << 6)),
1894 				 INIT_HCA_FS_A0_OFFSET);
1895 	} else {
1896 		MLX4_PUT(inbox, param->mc_base,	INIT_HCA_MC_BASE_OFFSET);
1897 		MLX4_PUT(inbox, param->log_mc_entry_sz,
1898 			 INIT_HCA_LOG_MC_ENTRY_SZ_OFFSET);
1899 		MLX4_PUT(inbox, param->log_mc_hash_sz,
1900 			 INIT_HCA_LOG_MC_HASH_SZ_OFFSET);
1901 		MLX4_PUT(inbox, param->log_mc_table_sz,
1902 			 INIT_HCA_LOG_MC_TABLE_SZ_OFFSET);
1903 		if (dev->caps.steering_mode == MLX4_STEERING_MODE_B0)
1904 			MLX4_PUT(inbox, (u8) (1 << 3),
1905 				 INIT_HCA_UC_STEERING_OFFSET);
1906 	}
1907 
1908 	/* TPT attributes */
1909 
1910 	MLX4_PUT(inbox, param->dmpt_base,  INIT_HCA_DMPT_BASE_OFFSET);
1911 	MLX4_PUT(inbox, param->mw_enabled, INIT_HCA_TPT_MW_OFFSET);
1912 	MLX4_PUT(inbox, param->log_mpt_sz, INIT_HCA_LOG_MPT_SZ_OFFSET);
1913 	MLX4_PUT(inbox, param->mtt_base,   INIT_HCA_MTT_BASE_OFFSET);
1914 	MLX4_PUT(inbox, param->cmpt_base,  INIT_HCA_CMPT_BASE_OFFSET);
1915 
1916 	/* UAR attributes */
1917 
1918 	MLX4_PUT(inbox, param->uar_page_sz,	INIT_HCA_UAR_PAGE_SZ_OFFSET);
1919 	MLX4_PUT(inbox, param->log_uar_sz,      INIT_HCA_LOG_UAR_SZ_OFFSET);
1920 
1921 	/* set parser VXLAN attributes */
1922 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS) {
1923 		u8 parser_params = 0;
1924 		MLX4_PUT(inbox, parser_params,	INIT_HCA_VXLAN_OFFSET);
1925 	}
1926 
1927 	err = mlx4_cmd(dev, mailbox->dma, 0, 0, MLX4_CMD_INIT_HCA,
1928 		       MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
1929 
1930 	if (err)
1931 		mlx4_err(dev, "INIT_HCA returns %d\n", err);
1932 
1933 	mlx4_free_cmd_mailbox(dev, mailbox);
1934 	return err;
1935 }
1936 
1937 int mlx4_QUERY_HCA(struct mlx4_dev *dev,
1938 		   struct mlx4_init_hca_param *param)
1939 {
1940 	struct mlx4_cmd_mailbox *mailbox;
1941 	__be32 *outbox;
1942 	u32 dword_field;
1943 	int err;
1944 	u8 byte_field;
1945 	static const u8 a0_dmfs_query_hw_steering[] =  {
1946 		[0] = MLX4_STEERING_DMFS_A0_DEFAULT,
1947 		[1] = MLX4_STEERING_DMFS_A0_DYNAMIC,
1948 		[2] = MLX4_STEERING_DMFS_A0_STATIC,
1949 		[3] = MLX4_STEERING_DMFS_A0_DISABLE
1950 	};
1951 
1952 #define QUERY_HCA_GLOBAL_CAPS_OFFSET	0x04
1953 #define QUERY_HCA_CORE_CLOCK_OFFSET	0x0c
1954 
1955 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1956 	if (IS_ERR(mailbox))
1957 		return PTR_ERR(mailbox);
1958 	outbox = mailbox->buf;
1959 
1960 	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0,
1961 			   MLX4_CMD_QUERY_HCA,
1962 			   MLX4_CMD_TIME_CLASS_B,
1963 			   !mlx4_is_slave(dev));
1964 	if (err)
1965 		goto out;
1966 
1967 	MLX4_GET(param->global_caps, outbox, QUERY_HCA_GLOBAL_CAPS_OFFSET);
1968 	MLX4_GET(param->hca_core_clock, outbox, QUERY_HCA_CORE_CLOCK_OFFSET);
1969 
1970 	/* QPC/EEC/CQC/EQC/RDMARC attributes */
1971 
1972 	MLX4_GET(param->qpc_base,      outbox, INIT_HCA_QPC_BASE_OFFSET);
1973 	MLX4_GET(param->log_num_qps,   outbox, INIT_HCA_LOG_QP_OFFSET);
1974 	MLX4_GET(param->srqc_base,     outbox, INIT_HCA_SRQC_BASE_OFFSET);
1975 	MLX4_GET(param->log_num_srqs,  outbox, INIT_HCA_LOG_SRQ_OFFSET);
1976 	MLX4_GET(param->cqc_base,      outbox, INIT_HCA_CQC_BASE_OFFSET);
1977 	MLX4_GET(param->log_num_cqs,   outbox, INIT_HCA_LOG_CQ_OFFSET);
1978 	MLX4_GET(param->altc_base,     outbox, INIT_HCA_ALTC_BASE_OFFSET);
1979 	MLX4_GET(param->auxc_base,     outbox, INIT_HCA_AUXC_BASE_OFFSET);
1980 	MLX4_GET(param->eqc_base,      outbox, INIT_HCA_EQC_BASE_OFFSET);
1981 	MLX4_GET(param->log_num_eqs,   outbox, INIT_HCA_LOG_EQ_OFFSET);
1982 	MLX4_GET(param->num_sys_eqs,   outbox, INIT_HCA_NUM_SYS_EQS_OFFSET);
1983 	MLX4_GET(param->rdmarc_base,   outbox, INIT_HCA_RDMARC_BASE_OFFSET);
1984 	MLX4_GET(param->log_rd_per_qp, outbox, INIT_HCA_LOG_RD_OFFSET);
1985 
1986 	MLX4_GET(dword_field, outbox, INIT_HCA_FLAGS_OFFSET);
1987 	if (dword_field & (1 << INIT_HCA_DEVICE_MANAGED_FLOW_STEERING_EN)) {
1988 		param->steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1989 	} else {
1990 		MLX4_GET(byte_field, outbox, INIT_HCA_UC_STEERING_OFFSET);
1991 		if (byte_field & 0x8)
1992 			param->steering_mode = MLX4_STEERING_MODE_B0;
1993 		else
1994 			param->steering_mode = MLX4_STEERING_MODE_A0;
1995 	}
1996 
1997 	if (dword_field & (1 << 13))
1998 		param->rss_ip_frags = 1;
1999 
2000 	/* steering attributes */
2001 	if (param->steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2002 		MLX4_GET(param->mc_base, outbox, INIT_HCA_FS_BASE_OFFSET);
2003 		MLX4_GET(param->log_mc_entry_sz, outbox,
2004 			 INIT_HCA_FS_LOG_ENTRY_SZ_OFFSET);
2005 		MLX4_GET(param->log_mc_table_sz, outbox,
2006 			 INIT_HCA_FS_LOG_TABLE_SZ_OFFSET);
2007 		MLX4_GET(byte_field, outbox,
2008 			 INIT_HCA_FS_A0_OFFSET);
2009 		param->dmfs_high_steer_mode =
2010 			a0_dmfs_query_hw_steering[(byte_field >> 6) & 3];
2011 	} else {
2012 		MLX4_GET(param->mc_base, outbox, INIT_HCA_MC_BASE_OFFSET);
2013 		MLX4_GET(param->log_mc_entry_sz, outbox,
2014 			 INIT_HCA_LOG_MC_ENTRY_SZ_OFFSET);
2015 		MLX4_GET(param->log_mc_hash_sz,  outbox,
2016 			 INIT_HCA_LOG_MC_HASH_SZ_OFFSET);
2017 		MLX4_GET(param->log_mc_table_sz, outbox,
2018 			 INIT_HCA_LOG_MC_TABLE_SZ_OFFSET);
2019 	}
2020 
2021 	/* CX3 is capable of extending CQEs/EQEs from 32 to 64 bytes */
2022 	MLX4_GET(byte_field, outbox, INIT_HCA_EQE_CQE_OFFSETS);
2023 	if (byte_field & 0x20) /* 64-bytes eqe enabled */
2024 		param->dev_cap_enabled |= MLX4_DEV_CAP_64B_EQE_ENABLED;
2025 	if (byte_field & 0x40) /* 64-bytes cqe enabled */
2026 		param->dev_cap_enabled |= MLX4_DEV_CAP_64B_CQE_ENABLED;
2027 
2028 	/* CX3 is capable of extending CQEs\EQEs to strides larger than 64B */
2029 	MLX4_GET(byte_field, outbox, INIT_HCA_EQE_CQE_STRIDE_OFFSET);
2030 	if (byte_field) {
2031 		param->dev_cap_enabled |= MLX4_DEV_CAP_EQE_STRIDE_ENABLED;
2032 		param->dev_cap_enabled |= MLX4_DEV_CAP_CQE_STRIDE_ENABLED;
2033 		param->cqe_size = 1 << ((byte_field &
2034 					 MLX4_CQE_SIZE_MASK_STRIDE) + 5);
2035 		param->eqe_size = 1 << (((byte_field &
2036 					  MLX4_EQE_SIZE_MASK_STRIDE) >> 4) + 5);
2037 	}
2038 
2039 	/* TPT attributes */
2040 
2041 	MLX4_GET(param->dmpt_base,  outbox, INIT_HCA_DMPT_BASE_OFFSET);
2042 	MLX4_GET(param->mw_enabled, outbox, INIT_HCA_TPT_MW_OFFSET);
2043 	MLX4_GET(param->log_mpt_sz, outbox, INIT_HCA_LOG_MPT_SZ_OFFSET);
2044 	MLX4_GET(param->mtt_base,   outbox, INIT_HCA_MTT_BASE_OFFSET);
2045 	MLX4_GET(param->cmpt_base,  outbox, INIT_HCA_CMPT_BASE_OFFSET);
2046 
2047 	/* UAR attributes */
2048 
2049 	MLX4_GET(param->uar_page_sz, outbox, INIT_HCA_UAR_PAGE_SZ_OFFSET);
2050 	MLX4_GET(param->log_uar_sz, outbox, INIT_HCA_LOG_UAR_SZ_OFFSET);
2051 
2052 	/* phv_check enable */
2053 	MLX4_GET(byte_field, outbox, INIT_HCA_CACHELINE_SZ_OFFSET);
2054 	if (byte_field & 0x2)
2055 		param->phv_check_en = 1;
2056 out:
2057 	mlx4_free_cmd_mailbox(dev, mailbox);
2058 
2059 	return err;
2060 }
2061 
2062 static int mlx4_hca_core_clock_update(struct mlx4_dev *dev)
2063 {
2064 	struct mlx4_cmd_mailbox *mailbox;
2065 	__be32 *outbox;
2066 	int err;
2067 
2068 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2069 	if (IS_ERR(mailbox)) {
2070 		mlx4_warn(dev, "hca_core_clock mailbox allocation failed\n");
2071 		return PTR_ERR(mailbox);
2072 	}
2073 	outbox = mailbox->buf;
2074 
2075 	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0,
2076 			   MLX4_CMD_QUERY_HCA,
2077 			   MLX4_CMD_TIME_CLASS_B,
2078 			   !mlx4_is_slave(dev));
2079 	if (err) {
2080 		mlx4_warn(dev, "hca_core_clock update failed\n");
2081 		goto out;
2082 	}
2083 
2084 	MLX4_GET(dev->caps.hca_core_clock, outbox, QUERY_HCA_CORE_CLOCK_OFFSET);
2085 
2086 out:
2087 	mlx4_free_cmd_mailbox(dev, mailbox);
2088 
2089 	return err;
2090 }
2091 
2092 /* for IB-type ports only in SRIOV mode. Checks that both proxy QP0
2093  * and real QP0 are active, so that the paravirtualized QP0 is ready
2094  * to operate */
2095 static int check_qp0_state(struct mlx4_dev *dev, int function, int port)
2096 {
2097 	struct mlx4_priv *priv = mlx4_priv(dev);
2098 	/* irrelevant if not infiniband */
2099 	if (priv->mfunc.master.qp0_state[port].proxy_qp0_active &&
2100 	    priv->mfunc.master.qp0_state[port].qp0_active)
2101 		return 1;
2102 	return 0;
2103 }
2104 
2105 int mlx4_INIT_PORT_wrapper(struct mlx4_dev *dev, int slave,
2106 			   struct mlx4_vhcr *vhcr,
2107 			   struct mlx4_cmd_mailbox *inbox,
2108 			   struct mlx4_cmd_mailbox *outbox,
2109 			   struct mlx4_cmd_info *cmd)
2110 {
2111 	struct mlx4_priv *priv = mlx4_priv(dev);
2112 	int port = mlx4_slave_convert_port(dev, slave, vhcr->in_modifier);
2113 	int err;
2114 
2115 	if (port < 0)
2116 		return -EINVAL;
2117 
2118 	if (priv->mfunc.master.slave_state[slave].init_port_mask & (1 << port))
2119 		return 0;
2120 
2121 	if (dev->caps.port_mask[port] != MLX4_PORT_TYPE_IB) {
2122 		/* Enable port only if it was previously disabled */
2123 		if (!priv->mfunc.master.init_port_ref[port]) {
2124 			err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_INIT_PORT,
2125 				       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2126 			if (err)
2127 				return err;
2128 		}
2129 		priv->mfunc.master.slave_state[slave].init_port_mask |= (1 << port);
2130 	} else {
2131 		if (slave == mlx4_master_func_num(dev)) {
2132 			if (check_qp0_state(dev, slave, port) &&
2133 			    !priv->mfunc.master.qp0_state[port].port_active) {
2134 				err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_INIT_PORT,
2135 					       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2136 				if (err)
2137 					return err;
2138 				priv->mfunc.master.qp0_state[port].port_active = 1;
2139 				priv->mfunc.master.slave_state[slave].init_port_mask |= (1 << port);
2140 			}
2141 		} else
2142 			priv->mfunc.master.slave_state[slave].init_port_mask |= (1 << port);
2143 	}
2144 	++priv->mfunc.master.init_port_ref[port];
2145 	return 0;
2146 }
2147 
2148 int mlx4_INIT_PORT(struct mlx4_dev *dev, int port)
2149 {
2150 	struct mlx4_cmd_mailbox *mailbox;
2151 	u32 *inbox;
2152 	int err;
2153 	u32 flags;
2154 	u16 field;
2155 
2156 	if (dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
2157 #define INIT_PORT_IN_SIZE          256
2158 #define INIT_PORT_FLAGS_OFFSET     0x00
2159 #define INIT_PORT_FLAG_SIG         (1 << 18)
2160 #define INIT_PORT_FLAG_NG          (1 << 17)
2161 #define INIT_PORT_FLAG_G0          (1 << 16)
2162 #define INIT_PORT_VL_SHIFT         4
2163 #define INIT_PORT_PORT_WIDTH_SHIFT 8
2164 #define INIT_PORT_MTU_OFFSET       0x04
2165 #define INIT_PORT_MAX_GID_OFFSET   0x06
2166 #define INIT_PORT_MAX_PKEY_OFFSET  0x0a
2167 #define INIT_PORT_GUID0_OFFSET     0x10
2168 #define INIT_PORT_NODE_GUID_OFFSET 0x18
2169 #define INIT_PORT_SI_GUID_OFFSET   0x20
2170 
2171 		mailbox = mlx4_alloc_cmd_mailbox(dev);
2172 		if (IS_ERR(mailbox))
2173 			return PTR_ERR(mailbox);
2174 		inbox = mailbox->buf;
2175 
2176 		flags = 0;
2177 		flags |= (dev->caps.vl_cap[port] & 0xf) << INIT_PORT_VL_SHIFT;
2178 		flags |= (dev->caps.port_width_cap[port] & 0xf) << INIT_PORT_PORT_WIDTH_SHIFT;
2179 		MLX4_PUT(inbox, flags,		  INIT_PORT_FLAGS_OFFSET);
2180 
2181 		field = 128 << dev->caps.ib_mtu_cap[port];
2182 		MLX4_PUT(inbox, field, INIT_PORT_MTU_OFFSET);
2183 		field = dev->caps.gid_table_len[port];
2184 		MLX4_PUT(inbox, field, INIT_PORT_MAX_GID_OFFSET);
2185 		field = dev->caps.pkey_table_len[port];
2186 		MLX4_PUT(inbox, field, INIT_PORT_MAX_PKEY_OFFSET);
2187 
2188 		err = mlx4_cmd(dev, mailbox->dma, port, 0, MLX4_CMD_INIT_PORT,
2189 			       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2190 
2191 		mlx4_free_cmd_mailbox(dev, mailbox);
2192 	} else
2193 		err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_INIT_PORT,
2194 			       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2195 
2196 	if (!err)
2197 		mlx4_hca_core_clock_update(dev);
2198 
2199 	return err;
2200 }
2201 EXPORT_SYMBOL_GPL(mlx4_INIT_PORT);
2202 
2203 int mlx4_CLOSE_PORT_wrapper(struct mlx4_dev *dev, int slave,
2204 			    struct mlx4_vhcr *vhcr,
2205 			    struct mlx4_cmd_mailbox *inbox,
2206 			    struct mlx4_cmd_mailbox *outbox,
2207 			    struct mlx4_cmd_info *cmd)
2208 {
2209 	struct mlx4_priv *priv = mlx4_priv(dev);
2210 	int port = mlx4_slave_convert_port(dev, slave, vhcr->in_modifier);
2211 	int err;
2212 
2213 	if (port < 0)
2214 		return -EINVAL;
2215 
2216 	if (!(priv->mfunc.master.slave_state[slave].init_port_mask &
2217 	    (1 << port)))
2218 		return 0;
2219 
2220 	if (dev->caps.port_mask[port] != MLX4_PORT_TYPE_IB) {
2221 		if (priv->mfunc.master.init_port_ref[port] == 1) {
2222 			err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_CLOSE_PORT,
2223 				       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2224 			if (err)
2225 				return err;
2226 		}
2227 		priv->mfunc.master.slave_state[slave].init_port_mask &= ~(1 << port);
2228 	} else {
2229 		/* infiniband port */
2230 		if (slave == mlx4_master_func_num(dev)) {
2231 			if (!priv->mfunc.master.qp0_state[port].qp0_active &&
2232 			    priv->mfunc.master.qp0_state[port].port_active) {
2233 				err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_CLOSE_PORT,
2234 					       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2235 				if (err)
2236 					return err;
2237 				priv->mfunc.master.slave_state[slave].init_port_mask &= ~(1 << port);
2238 				priv->mfunc.master.qp0_state[port].port_active = 0;
2239 			}
2240 		} else
2241 			priv->mfunc.master.slave_state[slave].init_port_mask &= ~(1 << port);
2242 	}
2243 	--priv->mfunc.master.init_port_ref[port];
2244 	return 0;
2245 }
2246 
2247 int mlx4_CLOSE_PORT(struct mlx4_dev *dev, int port)
2248 {
2249 	return mlx4_cmd(dev, 0, port, 0, MLX4_CMD_CLOSE_PORT,
2250 			MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2251 }
2252 EXPORT_SYMBOL_GPL(mlx4_CLOSE_PORT);
2253 
2254 int mlx4_CLOSE_HCA(struct mlx4_dev *dev, int panic)
2255 {
2256 	return mlx4_cmd(dev, 0, 0, panic, MLX4_CMD_CLOSE_HCA,
2257 			MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
2258 }
2259 
2260 struct mlx4_config_dev {
2261 	__be32	update_flags;
2262 	__be32	rsvd1[3];
2263 	__be16	vxlan_udp_dport;
2264 	__be16	rsvd2;
2265 	__be16  roce_v2_entropy;
2266 	__be16  roce_v2_udp_dport;
2267 	__be32	roce_flags;
2268 	__be32	rsvd4[25];
2269 	__be16	rsvd5;
2270 	u8	rsvd6;
2271 	u8	rx_checksum_val;
2272 };
2273 
2274 #define MLX4_VXLAN_UDP_DPORT (1 << 0)
2275 #define MLX4_ROCE_V2_UDP_DPORT BIT(3)
2276 #define MLX4_DISABLE_RX_PORT BIT(18)
2277 
2278 static int mlx4_CONFIG_DEV_set(struct mlx4_dev *dev, struct mlx4_config_dev *config_dev)
2279 {
2280 	int err;
2281 	struct mlx4_cmd_mailbox *mailbox;
2282 
2283 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2284 	if (IS_ERR(mailbox))
2285 		return PTR_ERR(mailbox);
2286 
2287 	memcpy(mailbox->buf, config_dev, sizeof(*config_dev));
2288 
2289 	err = mlx4_cmd(dev, mailbox->dma, 0, 0, MLX4_CMD_CONFIG_DEV,
2290 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
2291 
2292 	mlx4_free_cmd_mailbox(dev, mailbox);
2293 	return err;
2294 }
2295 
2296 static int mlx4_CONFIG_DEV_get(struct mlx4_dev *dev, struct mlx4_config_dev *config_dev)
2297 {
2298 	int err;
2299 	struct mlx4_cmd_mailbox *mailbox;
2300 
2301 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2302 	if (IS_ERR(mailbox))
2303 		return PTR_ERR(mailbox);
2304 
2305 	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 1, MLX4_CMD_CONFIG_DEV,
2306 			   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2307 
2308 	if (!err)
2309 		memcpy(config_dev, mailbox->buf, sizeof(*config_dev));
2310 
2311 	mlx4_free_cmd_mailbox(dev, mailbox);
2312 	return err;
2313 }
2314 
2315 /* Conversion between the HW values and the actual functionality.
2316  * The value represented by the array index,
2317  * and the functionality determined by the flags.
2318  */
2319 static const u8 config_dev_csum_flags[] = {
2320 	[0] =	0,
2321 	[1] =	MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP,
2322 	[2] =	MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP	|
2323 		MLX4_RX_CSUM_MODE_L4,
2324 	[3] =	MLX4_RX_CSUM_MODE_L4			|
2325 		MLX4_RX_CSUM_MODE_IP_OK_IP_NON_TCP_UDP	|
2326 		MLX4_RX_CSUM_MODE_MULTI_VLAN
2327 };
2328 
2329 int mlx4_config_dev_retrieval(struct mlx4_dev *dev,
2330 			      struct mlx4_config_dev_params *params)
2331 {
2332 	struct mlx4_config_dev config_dev = {0};
2333 	int err;
2334 	u8 csum_mask;
2335 
2336 #define CONFIG_DEV_RX_CSUM_MODE_MASK			0x7
2337 #define CONFIG_DEV_RX_CSUM_MODE_PORT1_BIT_OFFSET	0
2338 #define CONFIG_DEV_RX_CSUM_MODE_PORT2_BIT_OFFSET	4
2339 
2340 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_CONFIG_DEV))
2341 		return -ENOTSUPP;
2342 
2343 	err = mlx4_CONFIG_DEV_get(dev, &config_dev);
2344 	if (err)
2345 		return err;
2346 
2347 	csum_mask = (config_dev.rx_checksum_val >> CONFIG_DEV_RX_CSUM_MODE_PORT1_BIT_OFFSET) &
2348 			CONFIG_DEV_RX_CSUM_MODE_MASK;
2349 
2350 	if (csum_mask >= sizeof(config_dev_csum_flags)/sizeof(config_dev_csum_flags[0]))
2351 		return -EINVAL;
2352 	params->rx_csum_flags_port_1 = config_dev_csum_flags[csum_mask];
2353 
2354 	csum_mask = (config_dev.rx_checksum_val >> CONFIG_DEV_RX_CSUM_MODE_PORT2_BIT_OFFSET) &
2355 			CONFIG_DEV_RX_CSUM_MODE_MASK;
2356 
2357 	if (csum_mask >= sizeof(config_dev_csum_flags)/sizeof(config_dev_csum_flags[0]))
2358 		return -EINVAL;
2359 	params->rx_csum_flags_port_2 = config_dev_csum_flags[csum_mask];
2360 
2361 	params->vxlan_udp_dport = be16_to_cpu(config_dev.vxlan_udp_dport);
2362 
2363 	return 0;
2364 }
2365 EXPORT_SYMBOL_GPL(mlx4_config_dev_retrieval);
2366 
2367 int mlx4_config_vxlan_port(struct mlx4_dev *dev, __be16 udp_port)
2368 {
2369 	struct mlx4_config_dev config_dev;
2370 
2371 	memset(&config_dev, 0, sizeof(config_dev));
2372 	config_dev.update_flags    = cpu_to_be32(MLX4_VXLAN_UDP_DPORT);
2373 	config_dev.vxlan_udp_dport = udp_port;
2374 
2375 	return mlx4_CONFIG_DEV_set(dev, &config_dev);
2376 }
2377 EXPORT_SYMBOL_GPL(mlx4_config_vxlan_port);
2378 
2379 #define CONFIG_DISABLE_RX_PORT BIT(15)
2380 int mlx4_disable_rx_port_check(struct mlx4_dev *dev, bool dis)
2381 {
2382 	struct mlx4_config_dev config_dev;
2383 
2384 	memset(&config_dev, 0, sizeof(config_dev));
2385 	config_dev.update_flags = cpu_to_be32(MLX4_DISABLE_RX_PORT);
2386 	if (dis)
2387 		config_dev.roce_flags =
2388 			cpu_to_be32(CONFIG_DISABLE_RX_PORT);
2389 
2390 	return mlx4_CONFIG_DEV_set(dev, &config_dev);
2391 }
2392 
2393 int mlx4_config_roce_v2_port(struct mlx4_dev *dev, u16 udp_port)
2394 {
2395 	struct mlx4_config_dev config_dev;
2396 
2397 	memset(&config_dev, 0, sizeof(config_dev));
2398 	config_dev.update_flags    = cpu_to_be32(MLX4_ROCE_V2_UDP_DPORT);
2399 	config_dev.roce_v2_udp_dport = cpu_to_be16(udp_port);
2400 
2401 	return mlx4_CONFIG_DEV_set(dev, &config_dev);
2402 }
2403 EXPORT_SYMBOL_GPL(mlx4_config_roce_v2_port);
2404 
2405 int mlx4_virt2phy_port_map(struct mlx4_dev *dev, u32 port1, u32 port2)
2406 {
2407 	struct mlx4_cmd_mailbox *mailbox;
2408 	struct {
2409 		__be32 v_port1;
2410 		__be32 v_port2;
2411 	} *v2p;
2412 	int err;
2413 
2414 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2415 	if (IS_ERR(mailbox))
2416 		return -ENOMEM;
2417 
2418 	v2p = mailbox->buf;
2419 	v2p->v_port1 = cpu_to_be32(port1);
2420 	v2p->v_port2 = cpu_to_be32(port2);
2421 
2422 	err = mlx4_cmd(dev, mailbox->dma, 0,
2423 		       MLX4_SET_PORT_VIRT2PHY, MLX4_CMD_VIRT_PORT_MAP,
2424 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
2425 
2426 	mlx4_free_cmd_mailbox(dev, mailbox);
2427 	return err;
2428 }
2429 
2430 
2431 int mlx4_SET_ICM_SIZE(struct mlx4_dev *dev, u64 icm_size, u64 *aux_pages)
2432 {
2433 	int ret = mlx4_cmd_imm(dev, icm_size, aux_pages, 0, 0,
2434 			       MLX4_CMD_SET_ICM_SIZE,
2435 			       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2436 	if (ret)
2437 		return ret;
2438 
2439 	/*
2440 	 * Round up number of system pages needed in case
2441 	 * MLX4_ICM_PAGE_SIZE < PAGE_SIZE.
2442 	 */
2443 	*aux_pages = ALIGN(*aux_pages, PAGE_SIZE / MLX4_ICM_PAGE_SIZE) >>
2444 		(PAGE_SHIFT - MLX4_ICM_PAGE_SHIFT);
2445 
2446 	return 0;
2447 }
2448 
2449 int mlx4_NOP(struct mlx4_dev *dev)
2450 {
2451 	/* Input modifier of 0x1f means "finish as soon as possible." */
2452 	return mlx4_cmd(dev, 0, 0x1f, 0, MLX4_CMD_NOP, MLX4_CMD_TIME_CLASS_A,
2453 			MLX4_CMD_NATIVE);
2454 }
2455 
2456 int mlx4_get_phys_port_id(struct mlx4_dev *dev)
2457 {
2458 	u8 port;
2459 	u32 *outbox;
2460 	struct mlx4_cmd_mailbox *mailbox;
2461 	u32 in_mod;
2462 	u32 guid_hi, guid_lo;
2463 	int err, ret = 0;
2464 #define MOD_STAT_CFG_PORT_OFFSET 8
2465 #define MOD_STAT_CFG_GUID_H	 0X14
2466 #define MOD_STAT_CFG_GUID_L	 0X1c
2467 
2468 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2469 	if (IS_ERR(mailbox))
2470 		return PTR_ERR(mailbox);
2471 	outbox = mailbox->buf;
2472 
2473 	for (port = 1; port <= dev->caps.num_ports; port++) {
2474 		in_mod = port << MOD_STAT_CFG_PORT_OFFSET;
2475 		err = mlx4_cmd_box(dev, 0, mailbox->dma, in_mod, 0x2,
2476 				   MLX4_CMD_MOD_STAT_CFG, MLX4_CMD_TIME_CLASS_A,
2477 				   MLX4_CMD_NATIVE);
2478 		if (err) {
2479 			mlx4_err(dev, "Fail to get port %d uplink guid\n",
2480 				 port);
2481 			ret = err;
2482 		} else {
2483 			MLX4_GET(guid_hi, outbox, MOD_STAT_CFG_GUID_H);
2484 			MLX4_GET(guid_lo, outbox, MOD_STAT_CFG_GUID_L);
2485 			dev->caps.phys_port_id[port] = (u64)guid_lo |
2486 						       (u64)guid_hi << 32;
2487 		}
2488 	}
2489 	mlx4_free_cmd_mailbox(dev, mailbox);
2490 	return ret;
2491 }
2492 
2493 #define MLX4_WOL_SETUP_MODE (5 << 28)
2494 int mlx4_wol_read(struct mlx4_dev *dev, u64 *config, int port)
2495 {
2496 	u32 in_mod = MLX4_WOL_SETUP_MODE | port << 8;
2497 
2498 	return mlx4_cmd_imm(dev, 0, config, in_mod, 0x3,
2499 			    MLX4_CMD_MOD_STAT_CFG, MLX4_CMD_TIME_CLASS_A,
2500 			    MLX4_CMD_NATIVE);
2501 }
2502 EXPORT_SYMBOL_GPL(mlx4_wol_read);
2503 
2504 int mlx4_wol_write(struct mlx4_dev *dev, u64 config, int port)
2505 {
2506 	u32 in_mod = MLX4_WOL_SETUP_MODE | port << 8;
2507 
2508 	return mlx4_cmd(dev, config, in_mod, 0x1, MLX4_CMD_MOD_STAT_CFG,
2509 			MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2510 }
2511 EXPORT_SYMBOL_GPL(mlx4_wol_write);
2512 
2513 enum {
2514 	ADD_TO_MCG = 0x26,
2515 };
2516 
2517 
2518 void mlx4_opreq_action(struct work_struct *work)
2519 {
2520 	struct mlx4_priv *priv = container_of(work, struct mlx4_priv,
2521 					      opreq_task);
2522 	struct mlx4_dev *dev = &priv->dev;
2523 	int num_tasks = atomic_read(&priv->opreq_count);
2524 	struct mlx4_cmd_mailbox *mailbox;
2525 	struct mlx4_mgm *mgm;
2526 	u32 *outbox;
2527 	u32 modifier;
2528 	u16 token;
2529 	u16 type;
2530 	int err;
2531 	u32 num_qps;
2532 	struct mlx4_qp qp;
2533 	int i;
2534 	u8 rem_mcg;
2535 	u8 prot;
2536 
2537 #define GET_OP_REQ_MODIFIER_OFFSET	0x08
2538 #define GET_OP_REQ_TOKEN_OFFSET		0x14
2539 #define GET_OP_REQ_TYPE_OFFSET		0x1a
2540 #define GET_OP_REQ_DATA_OFFSET		0x20
2541 
2542 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2543 	if (IS_ERR(mailbox)) {
2544 		mlx4_err(dev, "Failed to allocate mailbox for GET_OP_REQ\n");
2545 		return;
2546 	}
2547 	outbox = mailbox->buf;
2548 
2549 	while (num_tasks) {
2550 		err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0,
2551 				   MLX4_CMD_GET_OP_REQ, MLX4_CMD_TIME_CLASS_A,
2552 				   MLX4_CMD_NATIVE);
2553 		if (err) {
2554 			mlx4_err(dev, "Failed to retrieve required operation: %d\n",
2555 				 err);
2556 			return;
2557 		}
2558 		MLX4_GET(modifier, outbox, GET_OP_REQ_MODIFIER_OFFSET);
2559 		MLX4_GET(token, outbox, GET_OP_REQ_TOKEN_OFFSET);
2560 		MLX4_GET(type, outbox, GET_OP_REQ_TYPE_OFFSET);
2561 		type &= 0xfff;
2562 
2563 		switch (type) {
2564 		case ADD_TO_MCG:
2565 			if (dev->caps.steering_mode ==
2566 			    MLX4_STEERING_MODE_DEVICE_MANAGED) {
2567 				mlx4_warn(dev, "ADD MCG operation is not supported in DEVICE_MANAGED steering mode\n");
2568 				err = EPERM;
2569 				break;
2570 			}
2571 			mgm = (struct mlx4_mgm *)((u8 *)(outbox) +
2572 						  GET_OP_REQ_DATA_OFFSET);
2573 			num_qps = be32_to_cpu(mgm->members_count) &
2574 				  MGM_QPN_MASK;
2575 			rem_mcg = ((u8 *)(&mgm->members_count))[0] & 1;
2576 			prot = ((u8 *)(&mgm->members_count))[0] >> 6;
2577 
2578 			for (i = 0; i < num_qps; i++) {
2579 				qp.qpn = be32_to_cpu(mgm->qp[i]);
2580 				if (rem_mcg)
2581 					err = mlx4_multicast_detach(dev, &qp,
2582 								    mgm->gid,
2583 								    prot, 0);
2584 				else
2585 					err = mlx4_multicast_attach(dev, &qp,
2586 								    mgm->gid,
2587 								    mgm->gid[5]
2588 								    , 0, prot,
2589 								    NULL);
2590 				if (err)
2591 					break;
2592 			}
2593 			break;
2594 		default:
2595 			mlx4_warn(dev, "Bad type for required operation\n");
2596 			err = EINVAL;
2597 			break;
2598 		}
2599 		err = mlx4_cmd(dev, 0, ((u32) err |
2600 					(__force u32)cpu_to_be32(token) << 16),
2601 			       1, MLX4_CMD_GET_OP_REQ, MLX4_CMD_TIME_CLASS_A,
2602 			       MLX4_CMD_NATIVE);
2603 		if (err) {
2604 			mlx4_err(dev, "Failed to acknowledge required request: %d\n",
2605 				 err);
2606 			goto out;
2607 		}
2608 		memset(outbox, 0, 0xffc);
2609 		num_tasks = atomic_dec_return(&priv->opreq_count);
2610 	}
2611 
2612 out:
2613 	mlx4_free_cmd_mailbox(dev, mailbox);
2614 }
2615 
2616 static int mlx4_check_smp_firewall_active(struct mlx4_dev *dev,
2617 					  struct mlx4_cmd_mailbox *mailbox)
2618 {
2619 #define MLX4_CMD_MAD_DEMUX_SET_ATTR_OFFSET		0x10
2620 #define MLX4_CMD_MAD_DEMUX_GETRESP_ATTR_OFFSET		0x20
2621 #define MLX4_CMD_MAD_DEMUX_TRAP_ATTR_OFFSET		0x40
2622 #define MLX4_CMD_MAD_DEMUX_TRAP_REPRESS_ATTR_OFFSET	0x70
2623 
2624 	u32 set_attr_mask, getresp_attr_mask;
2625 	u32 trap_attr_mask, traprepress_attr_mask;
2626 
2627 	MLX4_GET(set_attr_mask, mailbox->buf,
2628 		 MLX4_CMD_MAD_DEMUX_SET_ATTR_OFFSET);
2629 	mlx4_dbg(dev, "SMP firewall set_attribute_mask = 0x%x\n",
2630 		 set_attr_mask);
2631 
2632 	MLX4_GET(getresp_attr_mask, mailbox->buf,
2633 		 MLX4_CMD_MAD_DEMUX_GETRESP_ATTR_OFFSET);
2634 	mlx4_dbg(dev, "SMP firewall getresp_attribute_mask = 0x%x\n",
2635 		 getresp_attr_mask);
2636 
2637 	MLX4_GET(trap_attr_mask, mailbox->buf,
2638 		 MLX4_CMD_MAD_DEMUX_TRAP_ATTR_OFFSET);
2639 	mlx4_dbg(dev, "SMP firewall trap_attribute_mask = 0x%x\n",
2640 		 trap_attr_mask);
2641 
2642 	MLX4_GET(traprepress_attr_mask, mailbox->buf,
2643 		 MLX4_CMD_MAD_DEMUX_TRAP_REPRESS_ATTR_OFFSET);
2644 	mlx4_dbg(dev, "SMP firewall traprepress_attribute_mask = 0x%x\n",
2645 		 traprepress_attr_mask);
2646 
2647 	if (set_attr_mask && getresp_attr_mask && trap_attr_mask &&
2648 	    traprepress_attr_mask)
2649 		return 1;
2650 
2651 	return 0;
2652 }
2653 
2654 int mlx4_config_mad_demux(struct mlx4_dev *dev)
2655 {
2656 	struct mlx4_cmd_mailbox *mailbox;
2657 	int secure_host_active;
2658 	int err;
2659 
2660 	/* Check if mad_demux is supported */
2661 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_MAD_DEMUX))
2662 		return 0;
2663 
2664 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2665 	if (IS_ERR(mailbox)) {
2666 		mlx4_warn(dev, "Failed to allocate mailbox for cmd MAD_DEMUX");
2667 		return -ENOMEM;
2668 	}
2669 
2670 	/* Query mad_demux to find out which MADs are handled by internal sma */
2671 	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0x01 /* subn mgmt class */,
2672 			   MLX4_CMD_MAD_DEMUX_QUERY_RESTR, MLX4_CMD_MAD_DEMUX,
2673 			   MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
2674 	if (err) {
2675 		mlx4_warn(dev, "MLX4_CMD_MAD_DEMUX: query restrictions failed (%d)\n",
2676 			  err);
2677 		goto out;
2678 	}
2679 
2680 	secure_host_active = mlx4_check_smp_firewall_active(dev, mailbox);
2681 
2682 	/* Config mad_demux to handle all MADs returned by the query above */
2683 	err = mlx4_cmd(dev, mailbox->dma, 0x01 /* subn mgmt class */,
2684 		       MLX4_CMD_MAD_DEMUX_CONFIG, MLX4_CMD_MAD_DEMUX,
2685 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
2686 	if (err) {
2687 		mlx4_warn(dev, "MLX4_CMD_MAD_DEMUX: configure failed (%d)\n", err);
2688 		goto out;
2689 	}
2690 
2691 	if (secure_host_active)
2692 		mlx4_warn(dev, "HCA operating in secure-host mode. SMP firewall activated.\n");
2693 out:
2694 	mlx4_free_cmd_mailbox(dev, mailbox);
2695 	return err;
2696 }
2697 
2698 /* Access Reg commands */
2699 enum mlx4_access_reg_masks {
2700 	MLX4_ACCESS_REG_STATUS_MASK = 0x7f,
2701 	MLX4_ACCESS_REG_METHOD_MASK = 0x7f,
2702 	MLX4_ACCESS_REG_LEN_MASK = 0x7ff
2703 };
2704 
2705 struct mlx4_access_reg {
2706 	__be16 constant1;
2707 	u8 status;
2708 	u8 resrvd1;
2709 	__be16 reg_id;
2710 	u8 method;
2711 	u8 constant2;
2712 	__be32 resrvd2[2];
2713 	__be16 len_const;
2714 	__be16 resrvd3;
2715 #define MLX4_ACCESS_REG_HEADER_SIZE (20)
2716 	u8 reg_data[MLX4_MAILBOX_SIZE-MLX4_ACCESS_REG_HEADER_SIZE];
2717 } __attribute__((__packed__));
2718 
2719 /**
2720  * mlx4_ACCESS_REG - Generic access reg command.
2721  * @dev: mlx4_dev.
2722  * @reg_id: register ID to access.
2723  * @method: Access method Read/Write.
2724  * @reg_len: register length to Read/Write in bytes.
2725  * @reg_data: reg_data pointer to Read/Write From/To.
2726  *
2727  * Access ConnectX registers FW command.
2728  * Returns 0 on success and copies outbox mlx4_access_reg data
2729  * field into reg_data or a negative error code.
2730  */
2731 static int mlx4_ACCESS_REG(struct mlx4_dev *dev, u16 reg_id,
2732 			   enum mlx4_access_reg_method method,
2733 			   u16 reg_len, void *reg_data)
2734 {
2735 	struct mlx4_cmd_mailbox *inbox, *outbox;
2736 	struct mlx4_access_reg *inbuf, *outbuf;
2737 	int err;
2738 
2739 	inbox = mlx4_alloc_cmd_mailbox(dev);
2740 	if (IS_ERR(inbox))
2741 		return PTR_ERR(inbox);
2742 
2743 	outbox = mlx4_alloc_cmd_mailbox(dev);
2744 	if (IS_ERR(outbox)) {
2745 		mlx4_free_cmd_mailbox(dev, inbox);
2746 		return PTR_ERR(outbox);
2747 	}
2748 
2749 	inbuf = inbox->buf;
2750 	outbuf = outbox->buf;
2751 
2752 	inbuf->constant1 = cpu_to_be16(0x1<<11 | 0x4);
2753 	inbuf->constant2 = 0x1;
2754 	inbuf->reg_id = cpu_to_be16(reg_id);
2755 	inbuf->method = method & MLX4_ACCESS_REG_METHOD_MASK;
2756 
2757 	reg_len = min(reg_len, (u16)(sizeof(inbuf->reg_data)));
2758 	inbuf->len_const =
2759 		cpu_to_be16(((reg_len/4 + 1) & MLX4_ACCESS_REG_LEN_MASK) |
2760 			    ((0x3) << 12));
2761 
2762 	memcpy(inbuf->reg_data, reg_data, reg_len);
2763 	err = mlx4_cmd_box(dev, inbox->dma, outbox->dma, 0, 0,
2764 			   MLX4_CMD_ACCESS_REG, MLX4_CMD_TIME_CLASS_C,
2765 			   MLX4_CMD_WRAPPED);
2766 	if (err)
2767 		goto out;
2768 
2769 	if (outbuf->status & MLX4_ACCESS_REG_STATUS_MASK) {
2770 		err = outbuf->status & MLX4_ACCESS_REG_STATUS_MASK;
2771 		mlx4_err(dev,
2772 			 "MLX4_CMD_ACCESS_REG(%x) returned REG status (%x)\n",
2773 			 reg_id, err);
2774 		goto out;
2775 	}
2776 
2777 	memcpy(reg_data, outbuf->reg_data, reg_len);
2778 out:
2779 	mlx4_free_cmd_mailbox(dev, inbox);
2780 	mlx4_free_cmd_mailbox(dev, outbox);
2781 	return err;
2782 }
2783 
2784 /* ConnectX registers IDs */
2785 enum mlx4_reg_id {
2786 	MLX4_REG_ID_PTYS = 0x5004,
2787 };
2788 
2789 /**
2790  * mlx4_ACCESS_PTYS_REG - Access PTYs (Port Type and Speed)
2791  * register
2792  * @dev: mlx4_dev.
2793  * @method: Access method Read/Write.
2794  * @ptys_reg: PTYS register data pointer.
2795  *
2796  * Access ConnectX PTYS register, to Read/Write Port Type/Speed
2797  * configuration
2798  * Returns 0 on success or a negative error code.
2799  */
2800 int mlx4_ACCESS_PTYS_REG(struct mlx4_dev *dev,
2801 			 enum mlx4_access_reg_method method,
2802 			 struct mlx4_ptys_reg *ptys_reg)
2803 {
2804 	return mlx4_ACCESS_REG(dev, MLX4_REG_ID_PTYS,
2805 			       method, sizeof(*ptys_reg), ptys_reg);
2806 }
2807 EXPORT_SYMBOL_GPL(mlx4_ACCESS_PTYS_REG);
2808 
2809 int mlx4_ACCESS_REG_wrapper(struct mlx4_dev *dev, int slave,
2810 			    struct mlx4_vhcr *vhcr,
2811 			    struct mlx4_cmd_mailbox *inbox,
2812 			    struct mlx4_cmd_mailbox *outbox,
2813 			    struct mlx4_cmd_info *cmd)
2814 {
2815 	struct mlx4_access_reg *inbuf = inbox->buf;
2816 	u8 method = inbuf->method & MLX4_ACCESS_REG_METHOD_MASK;
2817 	u16 reg_id = be16_to_cpu(inbuf->reg_id);
2818 
2819 	if (slave != mlx4_master_func_num(dev) &&
2820 	    method == MLX4_ACCESS_REG_WRITE)
2821 		return -EPERM;
2822 
2823 	if (reg_id == MLX4_REG_ID_PTYS) {
2824 		struct mlx4_ptys_reg *ptys_reg =
2825 			(struct mlx4_ptys_reg *)inbuf->reg_data;
2826 
2827 		ptys_reg->local_port =
2828 			mlx4_slave_convert_port(dev, slave,
2829 						ptys_reg->local_port);
2830 	}
2831 
2832 	return mlx4_cmd_box(dev, inbox->dma, outbox->dma, vhcr->in_modifier,
2833 			    0, MLX4_CMD_ACCESS_REG, MLX4_CMD_TIME_CLASS_C,
2834 			    MLX4_CMD_NATIVE);
2835 }
2836 
2837 static int mlx4_SET_PORT_phv_bit(struct mlx4_dev *dev, u8 port, u8 phv_bit)
2838 {
2839 #define SET_PORT_GEN_PHV_VALID	0x10
2840 #define SET_PORT_GEN_PHV_EN	0x80
2841 
2842 	struct mlx4_cmd_mailbox *mailbox;
2843 	struct mlx4_set_port_general_context *context;
2844 	u32 in_mod;
2845 	int err;
2846 
2847 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2848 	if (IS_ERR(mailbox))
2849 		return PTR_ERR(mailbox);
2850 	context = mailbox->buf;
2851 
2852 	context->v_ignore_fcs |=  SET_PORT_GEN_PHV_VALID;
2853 	if (phv_bit)
2854 		context->phv_en |=  SET_PORT_GEN_PHV_EN;
2855 
2856 	in_mod = MLX4_SET_PORT_GENERAL << 8 | port;
2857 	err = mlx4_cmd(dev, mailbox->dma, in_mod, MLX4_SET_PORT_ETH_OPCODE,
2858 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
2859 		       MLX4_CMD_NATIVE);
2860 
2861 	mlx4_free_cmd_mailbox(dev, mailbox);
2862 	return err;
2863 }
2864 
2865 int get_phv_bit(struct mlx4_dev *dev, u8 port, int *phv)
2866 {
2867 	int err;
2868 	struct mlx4_func_cap func_cap;
2869 
2870 	memset(&func_cap, 0, sizeof(func_cap));
2871 	err = mlx4_QUERY_FUNC_CAP(dev, port, &func_cap);
2872 	if (!err)
2873 		*phv = func_cap.flags & QUERY_FUNC_CAP_PHV_BIT;
2874 	return err;
2875 }
2876 EXPORT_SYMBOL(get_phv_bit);
2877 
2878 int set_phv_bit(struct mlx4_dev *dev, u8 port, int new_val)
2879 {
2880 	int ret;
2881 
2882 	if (mlx4_is_slave(dev))
2883 		return -EPERM;
2884 
2885 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
2886 	    !(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
2887 		ret = mlx4_SET_PORT_phv_bit(dev, port, new_val);
2888 		if (!ret)
2889 			dev->caps.phv_bit[port] = new_val;
2890 		return ret;
2891 	}
2892 
2893 	return -EOPNOTSUPP;
2894 }
2895 EXPORT_SYMBOL(set_phv_bit);
2896 
2897 void mlx4_replace_zero_macs(struct mlx4_dev *dev)
2898 {
2899 	int i;
2900 	u8 mac_addr[ETH_ALEN];
2901 
2902 	dev->port_random_macs = 0;
2903 	for (i = 1; i <= dev->caps.num_ports; ++i)
2904 		if (!dev->caps.def_mac[i] &&
2905 		    dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH) {
2906 			eth_random_addr(mac_addr);
2907 			dev->port_random_macs |= 1 << i;
2908 			dev->caps.def_mac[i] = mlx4_mac_to_u64(mac_addr);
2909 		}
2910 }
2911 EXPORT_SYMBOL_GPL(mlx4_replace_zero_macs);
2912