xref: /linux/drivers/net/ethernet/mellanox/mlx4/fw.c (revision 69bfec7548f4c1595bac0e3ddfc0458a5af31f4c)
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 #include <linux/kernel.h>
40 #include <uapi/rdma/mlx4-abi.h>
41 
42 #include "fw.h"
43 #include "icm.h"
44 
45 enum {
46 	MLX4_COMMAND_INTERFACE_MIN_REV		= 2,
47 	MLX4_COMMAND_INTERFACE_MAX_REV		= 3,
48 	MLX4_COMMAND_INTERFACE_NEW_PORT_CMDS	= 3,
49 };
50 
51 extern void __buggy_use_of_MLX4_GET(void);
52 extern void __buggy_use_of_MLX4_PUT(void);
53 
54 static bool enable_qos;
55 module_param(enable_qos, bool, 0444);
56 MODULE_PARM_DESC(enable_qos, "Enable Enhanced QoS support (default: off)");
57 
58 #define MLX4_GET(dest, source, offset)				      \
59 	do {							      \
60 		void *__p = (char *) (source) + (offset);	      \
61 		__be64 val;                                           \
62 		switch (sizeof(dest)) {				      \
63 		case 1: (dest) = *(u8 *) __p;	    break;	      \
64 		case 2: (dest) = be16_to_cpup(__p); break;	      \
65 		case 4: (dest) = be32_to_cpup(__p); break;	      \
66 		case 8: val = get_unaligned((__be64 *)__p);           \
67 			(dest) = be64_to_cpu(val);  break;            \
68 		default: __buggy_use_of_MLX4_GET();		      \
69 		}						      \
70 	} while (0)
71 
72 #define MLX4_PUT(dest, source, offset)				      \
73 	do {							      \
74 		void *__d = ((char *) (dest) + (offset));	      \
75 		switch (sizeof(source)) {			      \
76 		case 1: *(u8 *) __d = (source);		       break; \
77 		case 2:	*(__be16 *) __d = cpu_to_be16(source); break; \
78 		case 4:	*(__be32 *) __d = cpu_to_be32(source); break; \
79 		case 8:	*(__be64 *) __d = cpu_to_be64(source); break; \
80 		default: __buggy_use_of_MLX4_PUT();		      \
81 		}						      \
82 	} while (0)
83 
84 static void dump_dev_cap_flags(struct mlx4_dev *dev, u64 flags)
85 {
86 	static const char *fname[] = {
87 		[ 0] = "RC transport",
88 		[ 1] = "UC transport",
89 		[ 2] = "UD transport",
90 		[ 3] = "XRC transport",
91 		[ 6] = "SRQ support",
92 		[ 7] = "IPoIB checksum offload",
93 		[ 8] = "P_Key violation counter",
94 		[ 9] = "Q_Key violation counter",
95 		[12] = "Dual Port Different Protocol (DPDP) support",
96 		[15] = "Big LSO headers",
97 		[16] = "MW support",
98 		[17] = "APM support",
99 		[18] = "Atomic ops support",
100 		[19] = "Raw multicast support",
101 		[20] = "Address vector port checking support",
102 		[21] = "UD multicast support",
103 		[30] = "IBoE support",
104 		[32] = "Unicast loopback support",
105 		[34] = "FCS header control",
106 		[37] = "Wake On LAN (port1) support",
107 		[38] = "Wake On LAN (port2) support",
108 		[40] = "UDP RSS support",
109 		[41] = "Unicast VEP steering support",
110 		[42] = "Multicast VEP steering support",
111 		[48] = "Counters support",
112 		[52] = "RSS IP fragments support",
113 		[53] = "Port ETS Scheduler support",
114 		[55] = "Port link type sensing support",
115 		[59] = "Port management change event support",
116 		[61] = "64 byte EQE support",
117 		[62] = "64 byte CQE support",
118 	};
119 	int i;
120 
121 	mlx4_dbg(dev, "DEV_CAP flags:\n");
122 	for (i = 0; i < ARRAY_SIZE(fname); ++i)
123 		if (fname[i] && (flags & (1LL << i)))
124 			mlx4_dbg(dev, "    %s\n", fname[i]);
125 }
126 
127 static void dump_dev_cap_flags2(struct mlx4_dev *dev, u64 flags)
128 {
129 	static const char * const fname[] = {
130 		[0] = "RSS support",
131 		[1] = "RSS Toeplitz Hash Function support",
132 		[2] = "RSS XOR Hash Function support",
133 		[3] = "Device managed flow steering support",
134 		[4] = "Automatic MAC reassignment support",
135 		[5] = "Time stamping support",
136 		[6] = "VST (control vlan insertion/stripping) support",
137 		[7] = "FSM (MAC anti-spoofing) support",
138 		[8] = "Dynamic QP updates support",
139 		[9] = "Device managed flow steering IPoIB support",
140 		[10] = "TCP/IP offloads/flow-steering for VXLAN support",
141 		[11] = "MAD DEMUX (Secure-Host) support",
142 		[12] = "Large cache line (>64B) CQE stride support",
143 		[13] = "Large cache line (>64B) EQE stride support",
144 		[14] = "Ethernet protocol control support",
145 		[15] = "Ethernet Backplane autoneg support",
146 		[16] = "CONFIG DEV support",
147 		[17] = "Asymmetric EQs support",
148 		[18] = "More than 80 VFs support",
149 		[19] = "Performance optimized for limited rule configuration flow steering support",
150 		[20] = "Recoverable error events support",
151 		[21] = "Port Remap support",
152 		[22] = "QCN support",
153 		[23] = "QP rate limiting support",
154 		[24] = "Ethernet Flow control statistics support",
155 		[25] = "Granular QoS per VF support",
156 		[26] = "Port ETS Scheduler support",
157 		[27] = "Port beacon support",
158 		[28] = "RX-ALL support",
159 		[29] = "802.1ad offload support",
160 		[31] = "Modifying loopback source checks using UPDATE_QP support",
161 		[32] = "Loopback source checks support",
162 		[33] = "RoCEv2 support",
163 		[34] = "DMFS Sniffer support (UC & MC)",
164 		[35] = "Diag counters per port",
165 		[36] = "QinQ VST mode support",
166 		[37] = "sl to vl mapping table change event support",
167 		[38] = "user MAC support",
168 		[39] = "Report driver version to FW support",
169 		[40] = "SW CQ initialization support",
170 	};
171 	int i;
172 
173 	for (i = 0; i < ARRAY_SIZE(fname); ++i)
174 		if (fname[i] && (flags & (1LL << i)))
175 			mlx4_dbg(dev, "    %s\n", fname[i]);
176 }
177 
178 int mlx4_MOD_STAT_CFG(struct mlx4_dev *dev, struct mlx4_mod_stat_cfg *cfg)
179 {
180 	struct mlx4_cmd_mailbox *mailbox;
181 	u32 *inbox;
182 	int err = 0;
183 
184 #define MOD_STAT_CFG_IN_SIZE		0x100
185 
186 #define MOD_STAT_CFG_PG_SZ_M_OFFSET	0x002
187 #define MOD_STAT_CFG_PG_SZ_OFFSET	0x003
188 
189 	mailbox = mlx4_alloc_cmd_mailbox(dev);
190 	if (IS_ERR(mailbox))
191 		return PTR_ERR(mailbox);
192 	inbox = mailbox->buf;
193 
194 	MLX4_PUT(inbox, cfg->log_pg_sz, MOD_STAT_CFG_PG_SZ_OFFSET);
195 	MLX4_PUT(inbox, cfg->log_pg_sz_m, MOD_STAT_CFG_PG_SZ_M_OFFSET);
196 
197 	err = mlx4_cmd(dev, mailbox->dma, 0, 0, MLX4_CMD_MOD_STAT_CFG,
198 			MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
199 
200 	mlx4_free_cmd_mailbox(dev, mailbox);
201 	return err;
202 }
203 
204 int mlx4_QUERY_FUNC(struct mlx4_dev *dev, struct mlx4_func *func, int slave)
205 {
206 	struct mlx4_cmd_mailbox *mailbox;
207 	u32 *outbox;
208 	u8 in_modifier;
209 	u8 field;
210 	u16 field16;
211 	int err;
212 
213 #define QUERY_FUNC_BUS_OFFSET			0x00
214 #define QUERY_FUNC_DEVICE_OFFSET		0x01
215 #define QUERY_FUNC_FUNCTION_OFFSET		0x01
216 #define QUERY_FUNC_PHYSICAL_FUNCTION_OFFSET	0x03
217 #define QUERY_FUNC_RSVD_EQS_OFFSET		0x04
218 #define QUERY_FUNC_MAX_EQ_OFFSET		0x06
219 #define QUERY_FUNC_RSVD_UARS_OFFSET		0x0b
220 
221 	mailbox = mlx4_alloc_cmd_mailbox(dev);
222 	if (IS_ERR(mailbox))
223 		return PTR_ERR(mailbox);
224 	outbox = mailbox->buf;
225 
226 	in_modifier = slave;
227 
228 	err = mlx4_cmd_box(dev, 0, mailbox->dma, in_modifier, 0,
229 			   MLX4_CMD_QUERY_FUNC,
230 			   MLX4_CMD_TIME_CLASS_A,
231 			   MLX4_CMD_NATIVE);
232 	if (err)
233 		goto out;
234 
235 	MLX4_GET(field, outbox, QUERY_FUNC_BUS_OFFSET);
236 	func->bus = field & 0xf;
237 	MLX4_GET(field, outbox, QUERY_FUNC_DEVICE_OFFSET);
238 	func->device = field & 0xf1;
239 	MLX4_GET(field, outbox, QUERY_FUNC_FUNCTION_OFFSET);
240 	func->function = field & 0x7;
241 	MLX4_GET(field, outbox, QUERY_FUNC_PHYSICAL_FUNCTION_OFFSET);
242 	func->physical_function = field & 0xf;
243 	MLX4_GET(field16, outbox, QUERY_FUNC_RSVD_EQS_OFFSET);
244 	func->rsvd_eqs = field16 & 0xffff;
245 	MLX4_GET(field16, outbox, QUERY_FUNC_MAX_EQ_OFFSET);
246 	func->max_eq = field16 & 0xffff;
247 	MLX4_GET(field, outbox, QUERY_FUNC_RSVD_UARS_OFFSET);
248 	func->rsvd_uars = field & 0x0f;
249 
250 	mlx4_dbg(dev, "Bus: %d, Device: %d, Function: %d, Physical function: %d, Max EQs: %d, Reserved EQs: %d, Reserved UARs: %d\n",
251 		 func->bus, func->device, func->function, func->physical_function,
252 		 func->max_eq, func->rsvd_eqs, func->rsvd_uars);
253 
254 out:
255 	mlx4_free_cmd_mailbox(dev, mailbox);
256 	return err;
257 }
258 
259 static int mlx4_activate_vst_qinq(struct mlx4_priv *priv, int slave, int port)
260 {
261 	struct mlx4_vport_oper_state *vp_oper;
262 	struct mlx4_vport_state *vp_admin;
263 	int err;
264 
265 	vp_oper = &priv->mfunc.master.vf_oper[slave].vport[port];
266 	vp_admin = &priv->mfunc.master.vf_admin[slave].vport[port];
267 
268 	if (vp_admin->default_vlan != vp_oper->state.default_vlan) {
269 		err = __mlx4_register_vlan(&priv->dev, port,
270 					   vp_admin->default_vlan,
271 					   &vp_oper->vlan_idx);
272 		if (err) {
273 			vp_oper->vlan_idx = NO_INDX;
274 			mlx4_warn(&priv->dev,
275 				  "No vlan resources slave %d, port %d\n",
276 				  slave, port);
277 			return err;
278 		}
279 		mlx4_dbg(&priv->dev, "alloc vlan %d idx  %d slave %d port %d\n",
280 			 (int)(vp_oper->state.default_vlan),
281 			 vp_oper->vlan_idx, slave, port);
282 	}
283 	vp_oper->state.vlan_proto   = vp_admin->vlan_proto;
284 	vp_oper->state.default_vlan = vp_admin->default_vlan;
285 	vp_oper->state.default_qos  = vp_admin->default_qos;
286 
287 	return 0;
288 }
289 
290 static int mlx4_handle_vst_qinq(struct mlx4_priv *priv, int slave, int port)
291 {
292 	struct mlx4_vport_oper_state *vp_oper;
293 	struct mlx4_slave_state *slave_state;
294 	struct mlx4_vport_state *vp_admin;
295 	int err;
296 
297 	vp_oper = &priv->mfunc.master.vf_oper[slave].vport[port];
298 	vp_admin = &priv->mfunc.master.vf_admin[slave].vport[port];
299 	slave_state = &priv->mfunc.master.slave_state[slave];
300 
301 	if ((vp_admin->vlan_proto != htons(ETH_P_8021AD)) ||
302 	    (!slave_state->active))
303 		return 0;
304 
305 	if (vp_oper->state.vlan_proto == vp_admin->vlan_proto &&
306 	    vp_oper->state.default_vlan == vp_admin->default_vlan &&
307 	    vp_oper->state.default_qos == vp_admin->default_qos)
308 		return 0;
309 
310 	if (!slave_state->vst_qinq_supported) {
311 		/* Warn and revert the request to set vst QinQ mode */
312 		vp_admin->vlan_proto   = vp_oper->state.vlan_proto;
313 		vp_admin->default_vlan = vp_oper->state.default_vlan;
314 		vp_admin->default_qos  = vp_oper->state.default_qos;
315 
316 		mlx4_warn(&priv->dev,
317 			  "Slave %d does not support VST QinQ mode\n", slave);
318 		return 0;
319 	}
320 
321 	err = mlx4_activate_vst_qinq(priv, slave, port);
322 	return err;
323 }
324 
325 int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
326 				struct mlx4_vhcr *vhcr,
327 				struct mlx4_cmd_mailbox *inbox,
328 				struct mlx4_cmd_mailbox *outbox,
329 				struct mlx4_cmd_info *cmd)
330 {
331 	struct mlx4_priv *priv = mlx4_priv(dev);
332 	u8	field, port;
333 	u32	size, proxy_qp, qkey;
334 	int	err = 0;
335 	struct mlx4_func func;
336 
337 #define QUERY_FUNC_CAP_FLAGS_OFFSET		0x0
338 #define QUERY_FUNC_CAP_NUM_PORTS_OFFSET		0x1
339 #define QUERY_FUNC_CAP_PF_BHVR_OFFSET		0x4
340 #define QUERY_FUNC_CAP_FMR_OFFSET		0x8
341 #define QUERY_FUNC_CAP_QP_QUOTA_OFFSET_DEP	0x10
342 #define QUERY_FUNC_CAP_CQ_QUOTA_OFFSET_DEP	0x14
343 #define QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET_DEP	0x18
344 #define QUERY_FUNC_CAP_MPT_QUOTA_OFFSET_DEP	0x20
345 #define QUERY_FUNC_CAP_MTT_QUOTA_OFFSET_DEP	0x24
346 #define QUERY_FUNC_CAP_MCG_QUOTA_OFFSET_DEP	0x28
347 #define QUERY_FUNC_CAP_MAX_EQ_OFFSET		0x2c
348 #define QUERY_FUNC_CAP_RESERVED_EQ_OFFSET	0x30
349 #define QUERY_FUNC_CAP_QP_RESD_LKEY_OFFSET	0x48
350 
351 #define QUERY_FUNC_CAP_QP_QUOTA_OFFSET		0x50
352 #define QUERY_FUNC_CAP_CQ_QUOTA_OFFSET		0x54
353 #define QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET		0x58
354 #define QUERY_FUNC_CAP_MPT_QUOTA_OFFSET		0x60
355 #define QUERY_FUNC_CAP_MTT_QUOTA_OFFSET		0x64
356 #define QUERY_FUNC_CAP_MCG_QUOTA_OFFSET		0x68
357 
358 #define QUERY_FUNC_CAP_EXTRA_FLAGS_OFFSET	0x6c
359 
360 #define QUERY_FUNC_CAP_FMR_FLAG			0x80
361 #define QUERY_FUNC_CAP_FLAG_RDMA		0x40
362 #define QUERY_FUNC_CAP_FLAG_ETH			0x80
363 #define QUERY_FUNC_CAP_FLAG_QUOTAS		0x10
364 #define QUERY_FUNC_CAP_FLAG_RESD_LKEY		0x08
365 #define QUERY_FUNC_CAP_FLAG_VALID_MAILBOX	0x04
366 
367 #define QUERY_FUNC_CAP_EXTRA_FLAGS_BF_QP_ALLOC_FLAG	(1UL << 31)
368 #define QUERY_FUNC_CAP_EXTRA_FLAGS_A0_QP_ALLOC_FLAG	(1UL << 30)
369 
370 /* when opcode modifier = 1 */
371 #define QUERY_FUNC_CAP_PHYS_PORT_OFFSET		0x3
372 #define QUERY_FUNC_CAP_PRIV_VF_QKEY_OFFSET	0x4
373 #define QUERY_FUNC_CAP_FLAGS0_OFFSET		0x8
374 #define QUERY_FUNC_CAP_FLAGS1_OFFSET		0xc
375 
376 #define QUERY_FUNC_CAP_QP0_TUNNEL		0x10
377 #define QUERY_FUNC_CAP_QP0_PROXY		0x14
378 #define QUERY_FUNC_CAP_QP1_TUNNEL		0x18
379 #define QUERY_FUNC_CAP_QP1_PROXY		0x1c
380 #define QUERY_FUNC_CAP_PHYS_PORT_ID		0x28
381 
382 #define QUERY_FUNC_CAP_FLAGS1_FORCE_MAC		0x40
383 #define QUERY_FUNC_CAP_FLAGS1_FORCE_VLAN	0x80
384 #define QUERY_FUNC_CAP_FLAGS1_NIC_INFO			0x10
385 #define QUERY_FUNC_CAP_VF_ENABLE_QP0		0x08
386 
387 #define QUERY_FUNC_CAP_FLAGS0_FORCE_PHY_WQE_GID 0x80
388 #define QUERY_FUNC_CAP_PHV_BIT			0x40
389 #define QUERY_FUNC_CAP_VLAN_OFFLOAD_DISABLE	0x20
390 
391 #define QUERY_FUNC_CAP_SUPPORTS_VST_QINQ	BIT(30)
392 #define QUERY_FUNC_CAP_SUPPORTS_NON_POWER_OF_2_NUM_EQS BIT(31)
393 
394 	if (vhcr->op_modifier == 1) {
395 		struct mlx4_active_ports actv_ports =
396 			mlx4_get_active_ports(dev, slave);
397 		int converted_port = mlx4_slave_convert_port(
398 				dev, slave, vhcr->in_modifier);
399 		struct mlx4_vport_oper_state *vp_oper;
400 
401 		if (converted_port < 0)
402 			return -EINVAL;
403 
404 		vhcr->in_modifier = converted_port;
405 		/* phys-port = logical-port */
406 		field = vhcr->in_modifier -
407 			find_first_bit(actv_ports.ports, dev->caps.num_ports);
408 		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_PHYS_PORT_OFFSET);
409 
410 		port = vhcr->in_modifier;
411 		proxy_qp = dev->phys_caps.base_proxy_sqpn + 8 * slave + port - 1;
412 
413 		/* Set nic_info bit to mark new fields support */
414 		field  = QUERY_FUNC_CAP_FLAGS1_NIC_INFO;
415 
416 		if (mlx4_vf_smi_enabled(dev, slave, port) &&
417 		    !mlx4_get_parav_qkey(dev, proxy_qp, &qkey)) {
418 			field |= QUERY_FUNC_CAP_VF_ENABLE_QP0;
419 			MLX4_PUT(outbox->buf, qkey,
420 				 QUERY_FUNC_CAP_PRIV_VF_QKEY_OFFSET);
421 		}
422 		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_FLAGS1_OFFSET);
423 
424 		/* size is now the QP number */
425 		size = dev->phys_caps.base_tunnel_sqpn + 8 * slave + port - 1;
426 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP0_TUNNEL);
427 
428 		size += 2;
429 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP1_TUNNEL);
430 
431 		MLX4_PUT(outbox->buf, proxy_qp, QUERY_FUNC_CAP_QP0_PROXY);
432 		proxy_qp += 2;
433 		MLX4_PUT(outbox->buf, proxy_qp, QUERY_FUNC_CAP_QP1_PROXY);
434 
435 		MLX4_PUT(outbox->buf, dev->caps.phys_port_id[vhcr->in_modifier],
436 			 QUERY_FUNC_CAP_PHYS_PORT_ID);
437 
438 		vp_oper = &priv->mfunc.master.vf_oper[slave].vport[port];
439 		err = mlx4_handle_vst_qinq(priv, slave, port);
440 		if (err)
441 			return err;
442 
443 		field = 0;
444 		if (dev->caps.phv_bit[port])
445 			field |= QUERY_FUNC_CAP_PHV_BIT;
446 		if (vp_oper->state.vlan_proto == htons(ETH_P_8021AD))
447 			field |= QUERY_FUNC_CAP_VLAN_OFFLOAD_DISABLE;
448 		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_FLAGS0_OFFSET);
449 
450 	} else if (vhcr->op_modifier == 0) {
451 		struct mlx4_active_ports actv_ports =
452 			mlx4_get_active_ports(dev, slave);
453 		struct mlx4_slave_state *slave_state =
454 			&priv->mfunc.master.slave_state[slave];
455 
456 		/* enable rdma and ethernet interfaces, new quota locations,
457 		 * and reserved lkey
458 		 */
459 		field = (QUERY_FUNC_CAP_FLAG_ETH | QUERY_FUNC_CAP_FLAG_RDMA |
460 			 QUERY_FUNC_CAP_FLAG_QUOTAS | QUERY_FUNC_CAP_FLAG_VALID_MAILBOX |
461 			 QUERY_FUNC_CAP_FLAG_RESD_LKEY);
462 		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_FLAGS_OFFSET);
463 
464 		field = min(
465 			bitmap_weight(actv_ports.ports, dev->caps.num_ports),
466 			(unsigned int) dev->caps.num_ports);
467 		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_NUM_PORTS_OFFSET);
468 
469 		size = dev->caps.function_caps; /* set PF behaviours */
470 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_PF_BHVR_OFFSET);
471 
472 		field = 0; /* protected FMR support not available as yet */
473 		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_FMR_OFFSET);
474 
475 		size = priv->mfunc.master.res_tracker.res_alloc[RES_QP].quota[slave];
476 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP_QUOTA_OFFSET);
477 		size = dev->caps.num_qps;
478 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP_QUOTA_OFFSET_DEP);
479 
480 		size = priv->mfunc.master.res_tracker.res_alloc[RES_SRQ].quota[slave];
481 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET);
482 		size = dev->caps.num_srqs;
483 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET_DEP);
484 
485 		size = priv->mfunc.master.res_tracker.res_alloc[RES_CQ].quota[slave];
486 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET);
487 		size = dev->caps.num_cqs;
488 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET_DEP);
489 
490 		if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) ||
491 		    mlx4_QUERY_FUNC(dev, &func, slave)) {
492 			size = vhcr->in_modifier &
493 				QUERY_FUNC_CAP_SUPPORTS_NON_POWER_OF_2_NUM_EQS ?
494 				dev->caps.num_eqs :
495 				rounddown_pow_of_two(dev->caps.num_eqs);
496 			MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MAX_EQ_OFFSET);
497 			size = dev->caps.reserved_eqs;
498 			MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET);
499 		} else {
500 			size = vhcr->in_modifier &
501 				QUERY_FUNC_CAP_SUPPORTS_NON_POWER_OF_2_NUM_EQS ?
502 				func.max_eq :
503 				rounddown_pow_of_two(func.max_eq);
504 			MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MAX_EQ_OFFSET);
505 			size = func.rsvd_eqs;
506 			MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET);
507 		}
508 
509 		size = priv->mfunc.master.res_tracker.res_alloc[RES_MPT].quota[slave];
510 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET);
511 		size = dev->caps.num_mpts;
512 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET_DEP);
513 
514 		size = priv->mfunc.master.res_tracker.res_alloc[RES_MTT].quota[slave];
515 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET);
516 		size = dev->caps.num_mtts;
517 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET_DEP);
518 
519 		size = dev->caps.num_mgms + dev->caps.num_amgms;
520 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET);
521 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET_DEP);
522 
523 		size = QUERY_FUNC_CAP_EXTRA_FLAGS_BF_QP_ALLOC_FLAG |
524 			QUERY_FUNC_CAP_EXTRA_FLAGS_A0_QP_ALLOC_FLAG;
525 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_EXTRA_FLAGS_OFFSET);
526 
527 		size = dev->caps.reserved_lkey + ((slave << 8) & 0xFF00);
528 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP_RESD_LKEY_OFFSET);
529 
530 		if (vhcr->in_modifier & QUERY_FUNC_CAP_SUPPORTS_VST_QINQ)
531 			slave_state->vst_qinq_supported = true;
532 
533 	} else
534 		err = -EINVAL;
535 
536 	return err;
537 }
538 
539 int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, u8 gen_or_port,
540 			struct mlx4_func_cap *func_cap)
541 {
542 	struct mlx4_cmd_mailbox *mailbox;
543 	u32			*outbox;
544 	u8			field, op_modifier;
545 	u32			size, qkey;
546 	int			err = 0, quotas = 0;
547 	u32                     in_modifier;
548 	u32			slave_caps;
549 
550 	op_modifier = !!gen_or_port; /* 0 = general, 1 = logical port */
551 	slave_caps = QUERY_FUNC_CAP_SUPPORTS_VST_QINQ |
552 		QUERY_FUNC_CAP_SUPPORTS_NON_POWER_OF_2_NUM_EQS;
553 	in_modifier = op_modifier ? gen_or_port : slave_caps;
554 
555 	mailbox = mlx4_alloc_cmd_mailbox(dev);
556 	if (IS_ERR(mailbox))
557 		return PTR_ERR(mailbox);
558 
559 	err = mlx4_cmd_box(dev, 0, mailbox->dma, in_modifier, op_modifier,
560 			   MLX4_CMD_QUERY_FUNC_CAP,
561 			   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
562 	if (err)
563 		goto out;
564 
565 	outbox = mailbox->buf;
566 
567 	if (!op_modifier) {
568 		MLX4_GET(field, outbox, QUERY_FUNC_CAP_FLAGS_OFFSET);
569 		if (!(field & (QUERY_FUNC_CAP_FLAG_ETH | QUERY_FUNC_CAP_FLAG_RDMA))) {
570 			mlx4_err(dev, "The host supports neither eth nor rdma interfaces\n");
571 			err = -EPROTONOSUPPORT;
572 			goto out;
573 		}
574 		func_cap->flags = field;
575 		quotas = !!(func_cap->flags & QUERY_FUNC_CAP_FLAG_QUOTAS);
576 
577 		MLX4_GET(field, outbox, QUERY_FUNC_CAP_NUM_PORTS_OFFSET);
578 		func_cap->num_ports = field;
579 
580 		MLX4_GET(size, outbox, QUERY_FUNC_CAP_PF_BHVR_OFFSET);
581 		func_cap->pf_context_behaviour = size;
582 
583 		if (quotas) {
584 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP_QUOTA_OFFSET);
585 			func_cap->qp_quota = size & 0xFFFFFF;
586 
587 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET);
588 			func_cap->srq_quota = size & 0xFFFFFF;
589 
590 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET);
591 			func_cap->cq_quota = size & 0xFFFFFF;
592 
593 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET);
594 			func_cap->mpt_quota = size & 0xFFFFFF;
595 
596 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET);
597 			func_cap->mtt_quota = size & 0xFFFFFF;
598 
599 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET);
600 			func_cap->mcg_quota = size & 0xFFFFFF;
601 
602 		} else {
603 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP_QUOTA_OFFSET_DEP);
604 			func_cap->qp_quota = size & 0xFFFFFF;
605 
606 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET_DEP);
607 			func_cap->srq_quota = size & 0xFFFFFF;
608 
609 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET_DEP);
610 			func_cap->cq_quota = size & 0xFFFFFF;
611 
612 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET_DEP);
613 			func_cap->mpt_quota = size & 0xFFFFFF;
614 
615 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET_DEP);
616 			func_cap->mtt_quota = size & 0xFFFFFF;
617 
618 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET_DEP);
619 			func_cap->mcg_quota = size & 0xFFFFFF;
620 		}
621 		MLX4_GET(size, outbox, QUERY_FUNC_CAP_MAX_EQ_OFFSET);
622 		func_cap->max_eq = size & 0xFFFFFF;
623 
624 		MLX4_GET(size, outbox, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET);
625 		func_cap->reserved_eq = size & 0xFFFFFF;
626 
627 		if (func_cap->flags & QUERY_FUNC_CAP_FLAG_RESD_LKEY) {
628 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP_RESD_LKEY_OFFSET);
629 			func_cap->reserved_lkey = size;
630 		} else {
631 			func_cap->reserved_lkey = 0;
632 		}
633 
634 		func_cap->extra_flags = 0;
635 
636 		/* Mailbox data from 0x6c and onward should only be treated if
637 		 * QUERY_FUNC_CAP_FLAG_VALID_MAILBOX is set in func_cap->flags
638 		 */
639 		if (func_cap->flags & QUERY_FUNC_CAP_FLAG_VALID_MAILBOX) {
640 			MLX4_GET(size, outbox, QUERY_FUNC_CAP_EXTRA_FLAGS_OFFSET);
641 			if (size & QUERY_FUNC_CAP_EXTRA_FLAGS_BF_QP_ALLOC_FLAG)
642 				func_cap->extra_flags |= MLX4_QUERY_FUNC_FLAGS_BF_RES_QP;
643 			if (size & QUERY_FUNC_CAP_EXTRA_FLAGS_A0_QP_ALLOC_FLAG)
644 				func_cap->extra_flags |= MLX4_QUERY_FUNC_FLAGS_A0_RES_QP;
645 		}
646 
647 		goto out;
648 	}
649 
650 	/* logical port query */
651 	if (gen_or_port > dev->caps.num_ports) {
652 		err = -EINVAL;
653 		goto out;
654 	}
655 
656 	MLX4_GET(func_cap->flags1, outbox, QUERY_FUNC_CAP_FLAGS1_OFFSET);
657 	if (dev->caps.port_type[gen_or_port] == MLX4_PORT_TYPE_ETH) {
658 		if (func_cap->flags1 & QUERY_FUNC_CAP_FLAGS1_FORCE_VLAN) {
659 			mlx4_err(dev, "VLAN is enforced on this port\n");
660 			err = -EPROTONOSUPPORT;
661 			goto out;
662 		}
663 
664 		if (func_cap->flags1 & QUERY_FUNC_CAP_FLAGS1_FORCE_MAC) {
665 			mlx4_err(dev, "Force mac is enabled on this port\n");
666 			err = -EPROTONOSUPPORT;
667 			goto out;
668 		}
669 	} else if (dev->caps.port_type[gen_or_port] == MLX4_PORT_TYPE_IB) {
670 		MLX4_GET(field, outbox, QUERY_FUNC_CAP_FLAGS0_OFFSET);
671 		if (field & QUERY_FUNC_CAP_FLAGS0_FORCE_PHY_WQE_GID) {
672 			mlx4_err(dev, "phy_wqe_gid is enforced on this ib port\n");
673 			err = -EPROTONOSUPPORT;
674 			goto out;
675 		}
676 	}
677 
678 	MLX4_GET(field, outbox, QUERY_FUNC_CAP_PHYS_PORT_OFFSET);
679 	func_cap->physical_port = field;
680 	if (func_cap->physical_port != gen_or_port) {
681 		err = -EINVAL;
682 		goto out;
683 	}
684 
685 	if (func_cap->flags1 & QUERY_FUNC_CAP_VF_ENABLE_QP0) {
686 		MLX4_GET(qkey, outbox, QUERY_FUNC_CAP_PRIV_VF_QKEY_OFFSET);
687 		func_cap->spec_qps.qp0_qkey = qkey;
688 	} else {
689 		func_cap->spec_qps.qp0_qkey = 0;
690 	}
691 
692 	MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP0_TUNNEL);
693 	func_cap->spec_qps.qp0_tunnel = size & 0xFFFFFF;
694 
695 	MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP0_PROXY);
696 	func_cap->spec_qps.qp0_proxy = size & 0xFFFFFF;
697 
698 	MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP1_TUNNEL);
699 	func_cap->spec_qps.qp1_tunnel = size & 0xFFFFFF;
700 
701 	MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP1_PROXY);
702 	func_cap->spec_qps.qp1_proxy = size & 0xFFFFFF;
703 
704 	if (func_cap->flags1 & QUERY_FUNC_CAP_FLAGS1_NIC_INFO)
705 		MLX4_GET(func_cap->phys_port_id, outbox,
706 			 QUERY_FUNC_CAP_PHYS_PORT_ID);
707 
708 	MLX4_GET(func_cap->flags0, outbox, QUERY_FUNC_CAP_FLAGS0_OFFSET);
709 
710 	/* All other resources are allocated by the master, but we still report
711 	 * 'num' and 'reserved' capabilities as follows:
712 	 * - num remains the maximum resource index
713 	 * - 'num - reserved' is the total available objects of a resource, but
714 	 *   resource indices may be less than 'reserved'
715 	 * TODO: set per-resource quotas */
716 
717 out:
718 	mlx4_free_cmd_mailbox(dev, mailbox);
719 
720 	return err;
721 }
722 
723 static void disable_unsupported_roce_caps(void *buf);
724 
725 int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
726 {
727 	struct mlx4_cmd_mailbox *mailbox;
728 	u32 *outbox;
729 	u8 field;
730 	u32 field32, flags, ext_flags;
731 	u16 size;
732 	u16 stat_rate;
733 	int err;
734 	int i;
735 
736 #define QUERY_DEV_CAP_OUT_SIZE		       0x100
737 #define QUERY_DEV_CAP_MAX_SRQ_SZ_OFFSET		0x10
738 #define QUERY_DEV_CAP_MAX_QP_SZ_OFFSET		0x11
739 #define QUERY_DEV_CAP_RSVD_QP_OFFSET		0x12
740 #define QUERY_DEV_CAP_MAX_QP_OFFSET		0x13
741 #define QUERY_DEV_CAP_RSVD_SRQ_OFFSET		0x14
742 #define QUERY_DEV_CAP_MAX_SRQ_OFFSET		0x15
743 #define QUERY_DEV_CAP_RSVD_EEC_OFFSET		0x16
744 #define QUERY_DEV_CAP_MAX_EEC_OFFSET		0x17
745 #define QUERY_DEV_CAP_MAX_CQ_SZ_OFFSET		0x19
746 #define QUERY_DEV_CAP_RSVD_CQ_OFFSET		0x1a
747 #define QUERY_DEV_CAP_MAX_CQ_OFFSET		0x1b
748 #define QUERY_DEV_CAP_MAX_MPT_OFFSET		0x1d
749 #define QUERY_DEV_CAP_RSVD_EQ_OFFSET		0x1e
750 #define QUERY_DEV_CAP_MAX_EQ_OFFSET		0x1f
751 #define QUERY_DEV_CAP_RSVD_MTT_OFFSET		0x20
752 #define QUERY_DEV_CAP_MAX_MRW_SZ_OFFSET		0x21
753 #define QUERY_DEV_CAP_RSVD_MRW_OFFSET		0x22
754 #define QUERY_DEV_CAP_MAX_MTT_SEG_OFFSET	0x23
755 #define QUERY_DEV_CAP_NUM_SYS_EQ_OFFSET		0x26
756 #define QUERY_DEV_CAP_MAX_AV_OFFSET		0x27
757 #define QUERY_DEV_CAP_MAX_REQ_QP_OFFSET		0x29
758 #define QUERY_DEV_CAP_MAX_RES_QP_OFFSET		0x2b
759 #define QUERY_DEV_CAP_MAX_GSO_OFFSET		0x2d
760 #define QUERY_DEV_CAP_RSS_OFFSET		0x2e
761 #define QUERY_DEV_CAP_MAX_RDMA_OFFSET		0x2f
762 #define QUERY_DEV_CAP_RSZ_SRQ_OFFSET		0x33
763 #define QUERY_DEV_CAP_PORT_BEACON_OFFSET	0x34
764 #define QUERY_DEV_CAP_ACK_DELAY_OFFSET		0x35
765 #define QUERY_DEV_CAP_MTU_WIDTH_OFFSET		0x36
766 #define QUERY_DEV_CAP_VL_PORT_OFFSET		0x37
767 #define QUERY_DEV_CAP_MAX_MSG_SZ_OFFSET		0x38
768 #define QUERY_DEV_CAP_MAX_GID_OFFSET		0x3b
769 #define QUERY_DEV_CAP_RATE_SUPPORT_OFFSET	0x3c
770 #define QUERY_DEV_CAP_CQ_TS_SUPPORT_OFFSET	0x3e
771 #define QUERY_DEV_CAP_MAX_PKEY_OFFSET		0x3f
772 #define QUERY_DEV_CAP_EXT_FLAGS_OFFSET		0x40
773 #define QUERY_DEV_CAP_WOL_OFFSET		0x43
774 #define QUERY_DEV_CAP_FLAGS_OFFSET		0x44
775 #define QUERY_DEV_CAP_RSVD_UAR_OFFSET		0x48
776 #define QUERY_DEV_CAP_UAR_SZ_OFFSET		0x49
777 #define QUERY_DEV_CAP_PAGE_SZ_OFFSET		0x4b
778 #define QUERY_DEV_CAP_BF_OFFSET			0x4c
779 #define QUERY_DEV_CAP_LOG_BF_REG_SZ_OFFSET	0x4d
780 #define QUERY_DEV_CAP_LOG_MAX_BF_REGS_PER_PAGE_OFFSET	0x4e
781 #define QUERY_DEV_CAP_LOG_MAX_BF_PAGES_OFFSET	0x4f
782 #define QUERY_DEV_CAP_MAX_SG_SQ_OFFSET		0x51
783 #define QUERY_DEV_CAP_MAX_DESC_SZ_SQ_OFFSET	0x52
784 #define QUERY_DEV_CAP_MAX_SG_RQ_OFFSET		0x55
785 #define QUERY_DEV_CAP_MAX_DESC_SZ_RQ_OFFSET	0x56
786 #define QUERY_DEV_CAP_USER_MAC_EN_OFFSET	0x5C
787 #define QUERY_DEV_CAP_SVLAN_BY_QP_OFFSET	0x5D
788 #define QUERY_DEV_CAP_MAX_QP_MCG_OFFSET		0x61
789 #define QUERY_DEV_CAP_RSVD_MCG_OFFSET		0x62
790 #define QUERY_DEV_CAP_MAX_MCG_OFFSET		0x63
791 #define QUERY_DEV_CAP_RSVD_PD_OFFSET		0x64
792 #define QUERY_DEV_CAP_MAX_PD_OFFSET		0x65
793 #define QUERY_DEV_CAP_RSVD_XRC_OFFSET		0x66
794 #define QUERY_DEV_CAP_MAX_XRC_OFFSET		0x67
795 #define QUERY_DEV_CAP_MAX_COUNTERS_OFFSET	0x68
796 #define QUERY_DEV_CAP_PORT_FLOWSTATS_COUNTERS_OFFSET	0x70
797 #define QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET	0x70
798 #define QUERY_DEV_CAP_FLOW_STEERING_IPOIB_OFFSET	0x74
799 #define QUERY_DEV_CAP_FLOW_STEERING_RANGE_EN_OFFSET	0x76
800 #define QUERY_DEV_CAP_FLOW_STEERING_MAX_QP_OFFSET	0x77
801 #define QUERY_DEV_CAP_SL2VL_EVENT_OFFSET	0x78
802 #define QUERY_DEV_CAP_CQ_EQ_CACHE_LINE_STRIDE	0x7a
803 #define QUERY_DEV_CAP_ECN_QCN_VER_OFFSET	0x7b
804 #define QUERY_DEV_CAP_RDMARC_ENTRY_SZ_OFFSET	0x80
805 #define QUERY_DEV_CAP_QPC_ENTRY_SZ_OFFSET	0x82
806 #define QUERY_DEV_CAP_AUX_ENTRY_SZ_OFFSET	0x84
807 #define QUERY_DEV_CAP_ALTC_ENTRY_SZ_OFFSET	0x86
808 #define QUERY_DEV_CAP_EQC_ENTRY_SZ_OFFSET	0x88
809 #define QUERY_DEV_CAP_CQC_ENTRY_SZ_OFFSET	0x8a
810 #define QUERY_DEV_CAP_SRQ_ENTRY_SZ_OFFSET	0x8c
811 #define QUERY_DEV_CAP_C_MPT_ENTRY_SZ_OFFSET	0x8e
812 #define QUERY_DEV_CAP_MTT_ENTRY_SZ_OFFSET	0x90
813 #define QUERY_DEV_CAP_D_MPT_ENTRY_SZ_OFFSET	0x92
814 #define QUERY_DEV_CAP_BMME_FLAGS_OFFSET		0x94
815 #define QUERY_DEV_CAP_CONFIG_DEV_OFFSET		0x94
816 #define QUERY_DEV_CAP_PHV_EN_OFFSET		0x96
817 #define QUERY_DEV_CAP_RSVD_LKEY_OFFSET		0x98
818 #define QUERY_DEV_CAP_MAX_ICM_SZ_OFFSET		0xa0
819 #define QUERY_DEV_CAP_ETH_BACKPL_OFFSET		0x9c
820 #define QUERY_DEV_CAP_DIAG_RPRT_PER_PORT	0x9c
821 #define QUERY_DEV_CAP_FW_REASSIGN_MAC		0x9d
822 #define QUERY_DEV_CAP_VXLAN			0x9e
823 #define QUERY_DEV_CAP_MAD_DEMUX_OFFSET		0xb0
824 #define QUERY_DEV_CAP_DMFS_HIGH_RATE_QPN_BASE_OFFSET	0xa8
825 #define QUERY_DEV_CAP_DMFS_HIGH_RATE_QPN_RANGE_OFFSET	0xac
826 #define QUERY_DEV_CAP_MAP_CLOCK_TO_USER 0xc1
827 #define QUERY_DEV_CAP_QP_RATE_LIMIT_NUM_OFFSET	0xcc
828 #define QUERY_DEV_CAP_QP_RATE_LIMIT_MAX_OFFSET	0xd0
829 #define QUERY_DEV_CAP_QP_RATE_LIMIT_MIN_OFFSET	0xd2
830 #define QUERY_DEV_CAP_HEALTH_BUFFER_ADDRESS_OFFSET	0xe4
831 
832 	dev_cap->flags2 = 0;
833 	mailbox = mlx4_alloc_cmd_mailbox(dev);
834 	if (IS_ERR(mailbox))
835 		return PTR_ERR(mailbox);
836 	outbox = mailbox->buf;
837 
838 	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0, MLX4_CMD_QUERY_DEV_CAP,
839 			   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
840 	if (err)
841 		goto out;
842 
843 	if (mlx4_is_mfunc(dev))
844 		disable_unsupported_roce_caps(outbox);
845 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAP_CLOCK_TO_USER);
846 	dev_cap->map_clock_to_user = field & 0x80;
847 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_QP_OFFSET);
848 	dev_cap->reserved_qps = 1 << (field & 0xf);
849 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_QP_OFFSET);
850 	dev_cap->max_qps = 1 << (field & 0x1f);
851 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_SRQ_OFFSET);
852 	dev_cap->reserved_srqs = 1 << (field >> 4);
853 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_SRQ_OFFSET);
854 	dev_cap->max_srqs = 1 << (field & 0x1f);
855 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_CQ_SZ_OFFSET);
856 	dev_cap->max_cq_sz = 1 << field;
857 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_CQ_OFFSET);
858 	dev_cap->reserved_cqs = 1 << (field & 0xf);
859 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_CQ_OFFSET);
860 	dev_cap->max_cqs = 1 << (field & 0x1f);
861 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_MPT_OFFSET);
862 	dev_cap->max_mpts = 1 << (field & 0x3f);
863 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_EQ_OFFSET);
864 	dev_cap->reserved_eqs = 1 << (field & 0xf);
865 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_EQ_OFFSET);
866 	dev_cap->max_eqs = 1 << (field & 0xf);
867 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_MTT_OFFSET);
868 	dev_cap->reserved_mtts = 1 << (field >> 4);
869 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_MRW_OFFSET);
870 	dev_cap->reserved_mrws = 1 << (field & 0xf);
871 	MLX4_GET(size, outbox, QUERY_DEV_CAP_NUM_SYS_EQ_OFFSET);
872 	dev_cap->num_sys_eqs = size & 0xfff;
873 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_REQ_QP_OFFSET);
874 	dev_cap->max_requester_per_qp = 1 << (field & 0x3f);
875 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_RES_QP_OFFSET);
876 	dev_cap->max_responder_per_qp = 1 << (field & 0x3f);
877 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_GSO_OFFSET);
878 	field &= 0x1f;
879 	if (!field)
880 		dev_cap->max_gso_sz = 0;
881 	else
882 		dev_cap->max_gso_sz = 1 << field;
883 
884 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSS_OFFSET);
885 	if (field & 0x20)
886 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_RSS_XOR;
887 	if (field & 0x10)
888 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_RSS_TOP;
889 	field &= 0xf;
890 	if (field) {
891 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_RSS;
892 		dev_cap->max_rss_tbl_sz = 1 << field;
893 	} else
894 		dev_cap->max_rss_tbl_sz = 0;
895 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_RDMA_OFFSET);
896 	dev_cap->max_rdma_global = 1 << (field & 0x3f);
897 	MLX4_GET(field, outbox, QUERY_DEV_CAP_ACK_DELAY_OFFSET);
898 	dev_cap->local_ca_ack_delay = field & 0x1f;
899 	MLX4_GET(field, outbox, QUERY_DEV_CAP_VL_PORT_OFFSET);
900 	dev_cap->num_ports = field & 0xf;
901 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_MSG_SZ_OFFSET);
902 	dev_cap->max_msg_sz = 1 << (field & 0x1f);
903 	MLX4_GET(field, outbox, QUERY_DEV_CAP_PORT_FLOWSTATS_COUNTERS_OFFSET);
904 	if (field & 0x10)
905 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN;
906 	MLX4_GET(field, outbox, QUERY_DEV_CAP_FLOW_STEERING_RANGE_EN_OFFSET);
907 	if (field & 0x80)
908 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_FS_EN;
909 	dev_cap->fs_log_max_ucast_qp_range_size = field & 0x1f;
910 	if (field & 0x20)
911 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER;
912 	MLX4_GET(field, outbox, QUERY_DEV_CAP_PORT_BEACON_OFFSET);
913 	if (field & 0x80)
914 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_PORT_BEACON;
915 	MLX4_GET(field, outbox, QUERY_DEV_CAP_FLOW_STEERING_IPOIB_OFFSET);
916 	if (field & 0x80)
917 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_DMFS_IPOIB;
918 	MLX4_GET(field, outbox, QUERY_DEV_CAP_FLOW_STEERING_MAX_QP_OFFSET);
919 	dev_cap->fs_max_num_qp_per_entry = field;
920 	MLX4_GET(field, outbox, QUERY_DEV_CAP_SL2VL_EVENT_OFFSET);
921 	if (field & (1 << 5))
922 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT;
923 	MLX4_GET(field, outbox, QUERY_DEV_CAP_ECN_QCN_VER_OFFSET);
924 	if (field & 0x1)
925 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_QCN;
926 	MLX4_GET(stat_rate, outbox, QUERY_DEV_CAP_RATE_SUPPORT_OFFSET);
927 	dev_cap->stat_rate_support = stat_rate;
928 	MLX4_GET(field, outbox, QUERY_DEV_CAP_CQ_TS_SUPPORT_OFFSET);
929 	if (field & 0x80)
930 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_TS;
931 	MLX4_GET(ext_flags, outbox, QUERY_DEV_CAP_EXT_FLAGS_OFFSET);
932 	MLX4_GET(flags, outbox, QUERY_DEV_CAP_FLAGS_OFFSET);
933 	dev_cap->flags = flags | (u64)ext_flags << 32;
934 	MLX4_GET(field, outbox, QUERY_DEV_CAP_WOL_OFFSET);
935 	dev_cap->wol_port[1] = !!(field & 0x20);
936 	dev_cap->wol_port[2] = !!(field & 0x40);
937 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_UAR_OFFSET);
938 	dev_cap->reserved_uars = field >> 4;
939 	MLX4_GET(field, outbox, QUERY_DEV_CAP_UAR_SZ_OFFSET);
940 	dev_cap->uar_size = 1 << ((field & 0x3f) + 20);
941 	MLX4_GET(field, outbox, QUERY_DEV_CAP_PAGE_SZ_OFFSET);
942 	dev_cap->min_page_sz = 1 << field;
943 
944 	MLX4_GET(field, outbox, QUERY_DEV_CAP_BF_OFFSET);
945 	if (field & 0x80) {
946 		MLX4_GET(field, outbox, QUERY_DEV_CAP_LOG_BF_REG_SZ_OFFSET);
947 		dev_cap->bf_reg_size = 1 << (field & 0x1f);
948 		MLX4_GET(field, outbox, QUERY_DEV_CAP_LOG_MAX_BF_REGS_PER_PAGE_OFFSET);
949 		if ((1 << (field & 0x3f)) > (PAGE_SIZE / dev_cap->bf_reg_size))
950 			field = 3;
951 		dev_cap->bf_regs_per_page = 1 << (field & 0x3f);
952 	} else {
953 		dev_cap->bf_reg_size = 0;
954 	}
955 
956 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_SG_SQ_OFFSET);
957 	dev_cap->max_sq_sg = field;
958 	MLX4_GET(size, outbox, QUERY_DEV_CAP_MAX_DESC_SZ_SQ_OFFSET);
959 	dev_cap->max_sq_desc_sz = size;
960 
961 	MLX4_GET(field, outbox, QUERY_DEV_CAP_USER_MAC_EN_OFFSET);
962 	if (field & (1 << 2))
963 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_USER_MAC_EN;
964 	MLX4_GET(field, outbox, QUERY_DEV_CAP_SVLAN_BY_QP_OFFSET);
965 	if (field & 0x1)
966 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_SVLAN_BY_QP;
967 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_QP_MCG_OFFSET);
968 	dev_cap->max_qp_per_mcg = 1 << field;
969 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_MCG_OFFSET);
970 	dev_cap->reserved_mgms = field & 0xf;
971 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_MCG_OFFSET);
972 	dev_cap->max_mcgs = 1 << field;
973 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_PD_OFFSET);
974 	dev_cap->reserved_pds = field >> 4;
975 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_PD_OFFSET);
976 	dev_cap->max_pds = 1 << (field & 0x3f);
977 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_XRC_OFFSET);
978 	dev_cap->reserved_xrcds = field >> 4;
979 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_XRC_OFFSET);
980 	dev_cap->max_xrcds = 1 << (field & 0x1f);
981 
982 	MLX4_GET(size, outbox, QUERY_DEV_CAP_RDMARC_ENTRY_SZ_OFFSET);
983 	dev_cap->rdmarc_entry_sz = size;
984 	MLX4_GET(size, outbox, QUERY_DEV_CAP_QPC_ENTRY_SZ_OFFSET);
985 	dev_cap->qpc_entry_sz = size;
986 	MLX4_GET(size, outbox, QUERY_DEV_CAP_AUX_ENTRY_SZ_OFFSET);
987 	dev_cap->aux_entry_sz = size;
988 	MLX4_GET(size, outbox, QUERY_DEV_CAP_ALTC_ENTRY_SZ_OFFSET);
989 	dev_cap->altc_entry_sz = size;
990 	MLX4_GET(size, outbox, QUERY_DEV_CAP_EQC_ENTRY_SZ_OFFSET);
991 	dev_cap->eqc_entry_sz = size;
992 	MLX4_GET(size, outbox, QUERY_DEV_CAP_CQC_ENTRY_SZ_OFFSET);
993 	dev_cap->cqc_entry_sz = size;
994 	MLX4_GET(size, outbox, QUERY_DEV_CAP_SRQ_ENTRY_SZ_OFFSET);
995 	dev_cap->srq_entry_sz = size;
996 	MLX4_GET(size, outbox, QUERY_DEV_CAP_C_MPT_ENTRY_SZ_OFFSET);
997 	dev_cap->cmpt_entry_sz = size;
998 	MLX4_GET(size, outbox, QUERY_DEV_CAP_MTT_ENTRY_SZ_OFFSET);
999 	dev_cap->mtt_entry_sz = size;
1000 	MLX4_GET(size, outbox, QUERY_DEV_CAP_D_MPT_ENTRY_SZ_OFFSET);
1001 	dev_cap->dmpt_entry_sz = size;
1002 
1003 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_SRQ_SZ_OFFSET);
1004 	dev_cap->max_srq_sz = 1 << field;
1005 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_QP_SZ_OFFSET);
1006 	dev_cap->max_qp_sz = 1 << field;
1007 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSZ_SRQ_OFFSET);
1008 	dev_cap->resize_srq = field & 1;
1009 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_SG_RQ_OFFSET);
1010 	dev_cap->max_rq_sg = field;
1011 	MLX4_GET(size, outbox, QUERY_DEV_CAP_MAX_DESC_SZ_RQ_OFFSET);
1012 	dev_cap->max_rq_desc_sz = size;
1013 	MLX4_GET(field, outbox, QUERY_DEV_CAP_CQ_EQ_CACHE_LINE_STRIDE);
1014 	if (field & (1 << 4))
1015 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_QOS_VPP;
1016 	if (field & (1 << 5))
1017 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_ETH_PROT_CTRL;
1018 	if (field & (1 << 6))
1019 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
1020 	if (field & (1 << 7))
1021 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
1022 	MLX4_GET(dev_cap->bmme_flags, outbox,
1023 		 QUERY_DEV_CAP_BMME_FLAGS_OFFSET);
1024 	if (dev_cap->bmme_flags & MLX4_FLAG_ROCE_V1_V2)
1025 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_ROCE_V1_V2;
1026 	if (dev_cap->bmme_flags & MLX4_FLAG_PORT_REMAP)
1027 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_PORT_REMAP;
1028 	MLX4_GET(field, outbox, QUERY_DEV_CAP_CONFIG_DEV_OFFSET);
1029 	if (field & 0x20)
1030 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_CONFIG_DEV;
1031 	if (field & (1 << 2))
1032 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
1033 	MLX4_GET(field, outbox, QUERY_DEV_CAP_PHV_EN_OFFSET);
1034 	if (field & 0x80)
1035 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_PHV_EN;
1036 	if (field & 0x40)
1037 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN;
1038 
1039 	MLX4_GET(dev_cap->reserved_lkey, outbox,
1040 		 QUERY_DEV_CAP_RSVD_LKEY_OFFSET);
1041 	MLX4_GET(field32, outbox, QUERY_DEV_CAP_ETH_BACKPL_OFFSET);
1042 	if (field32 & (1 << 0))
1043 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_ETH_BACKPL_AN_REP;
1044 	if (field32 & (1 << 7))
1045 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_RECOVERABLE_ERROR_EVENT;
1046 	if (field32 & (1 << 8))
1047 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_DRIVER_VERSION_TO_FW;
1048 	MLX4_GET(field32, outbox, QUERY_DEV_CAP_DIAG_RPRT_PER_PORT);
1049 	if (field32 & (1 << 17))
1050 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT;
1051 	MLX4_GET(field, outbox, QUERY_DEV_CAP_FW_REASSIGN_MAC);
1052 	if (field & 1<<6)
1053 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN;
1054 	MLX4_GET(field, outbox, QUERY_DEV_CAP_VXLAN);
1055 	if (field & 1<<3)
1056 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS;
1057 	if (field & (1 << 5))
1058 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
1059 	MLX4_GET(dev_cap->max_icm_sz, outbox,
1060 		 QUERY_DEV_CAP_MAX_ICM_SZ_OFFSET);
1061 	if (dev_cap->flags & MLX4_DEV_CAP_FLAG_COUNTERS)
1062 		MLX4_GET(dev_cap->max_counters, outbox,
1063 			 QUERY_DEV_CAP_MAX_COUNTERS_OFFSET);
1064 
1065 	MLX4_GET(field32, outbox,
1066 		 QUERY_DEV_CAP_MAD_DEMUX_OFFSET);
1067 	if (field32 & (1 << 0))
1068 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_MAD_DEMUX;
1069 
1070 	MLX4_GET(dev_cap->dmfs_high_rate_qpn_base, outbox,
1071 		 QUERY_DEV_CAP_DMFS_HIGH_RATE_QPN_BASE_OFFSET);
1072 	dev_cap->dmfs_high_rate_qpn_base &= MGM_QPN_MASK;
1073 	MLX4_GET(dev_cap->dmfs_high_rate_qpn_range, outbox,
1074 		 QUERY_DEV_CAP_DMFS_HIGH_RATE_QPN_RANGE_OFFSET);
1075 	dev_cap->dmfs_high_rate_qpn_range &= MGM_QPN_MASK;
1076 
1077 	MLX4_GET(size, outbox, QUERY_DEV_CAP_QP_RATE_LIMIT_NUM_OFFSET);
1078 	dev_cap->rl_caps.num_rates = size;
1079 	if (dev_cap->rl_caps.num_rates) {
1080 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT;
1081 		MLX4_GET(size, outbox, QUERY_DEV_CAP_QP_RATE_LIMIT_MAX_OFFSET);
1082 		dev_cap->rl_caps.max_val  = size & 0xfff;
1083 		dev_cap->rl_caps.max_unit = size >> 14;
1084 		MLX4_GET(size, outbox, QUERY_DEV_CAP_QP_RATE_LIMIT_MIN_OFFSET);
1085 		dev_cap->rl_caps.min_val  = size & 0xfff;
1086 		dev_cap->rl_caps.min_unit = size >> 14;
1087 	}
1088 
1089 	MLX4_GET(dev_cap->health_buffer_addrs, outbox,
1090 		 QUERY_DEV_CAP_HEALTH_BUFFER_ADDRESS_OFFSET);
1091 
1092 	MLX4_GET(field32, outbox, QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET);
1093 	if (field32 & (1 << 16))
1094 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_UPDATE_QP;
1095 	if (field32 & (1 << 18))
1096 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_UPDATE_QP_SRC_CHECK_LB;
1097 	if (field32 & (1 << 19))
1098 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_LB_SRC_CHK;
1099 	if (field32 & (1 << 26))
1100 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_VLAN_CONTROL;
1101 	if (field32 & (1 << 20))
1102 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_FSM;
1103 	if (field32 & (1 << 21))
1104 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_80_VFS;
1105 	if (field32 & (1 << 23))
1106 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_SW_CQ_INIT;
1107 
1108 	for (i = 1; i <= dev_cap->num_ports; i++) {
1109 		err = mlx4_QUERY_PORT(dev, i, dev_cap->port_cap + i);
1110 		if (err)
1111 			goto out;
1112 	}
1113 
1114 	/*
1115 	 * Each UAR has 4 EQ doorbells; so if a UAR is reserved, then
1116 	 * we can't use any EQs whose doorbell falls on that page,
1117 	 * even if the EQ itself isn't reserved.
1118 	 */
1119 	if (dev_cap->num_sys_eqs == 0)
1120 		dev_cap->reserved_eqs = max(dev_cap->reserved_uars * 4,
1121 					    dev_cap->reserved_eqs);
1122 	else
1123 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_SYS_EQS;
1124 
1125 out:
1126 	mlx4_free_cmd_mailbox(dev, mailbox);
1127 	return err;
1128 }
1129 
1130 void mlx4_dev_cap_dump(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
1131 {
1132 	if (dev_cap->bf_reg_size > 0)
1133 		mlx4_dbg(dev, "BlueFlame available (reg size %d, regs/page %d)\n",
1134 			 dev_cap->bf_reg_size, dev_cap->bf_regs_per_page);
1135 	else
1136 		mlx4_dbg(dev, "BlueFlame not available\n");
1137 
1138 	mlx4_dbg(dev, "Base MM extensions: flags %08x, rsvd L_Key %08x\n",
1139 		 dev_cap->bmme_flags, dev_cap->reserved_lkey);
1140 	mlx4_dbg(dev, "Max ICM size %lld MB\n",
1141 		 (unsigned long long) dev_cap->max_icm_sz >> 20);
1142 	mlx4_dbg(dev, "Max QPs: %d, reserved QPs: %d, entry size: %d\n",
1143 		 dev_cap->max_qps, dev_cap->reserved_qps, dev_cap->qpc_entry_sz);
1144 	mlx4_dbg(dev, "Max SRQs: %d, reserved SRQs: %d, entry size: %d\n",
1145 		 dev_cap->max_srqs, dev_cap->reserved_srqs, dev_cap->srq_entry_sz);
1146 	mlx4_dbg(dev, "Max CQs: %d, reserved CQs: %d, entry size: %d\n",
1147 		 dev_cap->max_cqs, dev_cap->reserved_cqs, dev_cap->cqc_entry_sz);
1148 	mlx4_dbg(dev, "Num sys EQs: %d, max EQs: %d, reserved EQs: %d, entry size: %d\n",
1149 		 dev_cap->num_sys_eqs, dev_cap->max_eqs, dev_cap->reserved_eqs,
1150 		 dev_cap->eqc_entry_sz);
1151 	mlx4_dbg(dev, "reserved MPTs: %d, reserved MTTs: %d\n",
1152 		 dev_cap->reserved_mrws, dev_cap->reserved_mtts);
1153 	mlx4_dbg(dev, "Max PDs: %d, reserved PDs: %d, reserved UARs: %d\n",
1154 		 dev_cap->max_pds, dev_cap->reserved_pds, dev_cap->reserved_uars);
1155 	mlx4_dbg(dev, "Max QP/MCG: %d, reserved MGMs: %d\n",
1156 		 dev_cap->max_pds, dev_cap->reserved_mgms);
1157 	mlx4_dbg(dev, "Max CQEs: %d, max WQEs: %d, max SRQ WQEs: %d\n",
1158 		 dev_cap->max_cq_sz, dev_cap->max_qp_sz, dev_cap->max_srq_sz);
1159 	mlx4_dbg(dev, "Local CA ACK delay: %d, max MTU: %d, port width cap: %d\n",
1160 		 dev_cap->local_ca_ack_delay, 128 << dev_cap->port_cap[1].ib_mtu,
1161 		 dev_cap->port_cap[1].max_port_width);
1162 	mlx4_dbg(dev, "Max SQ desc size: %d, max SQ S/G: %d\n",
1163 		 dev_cap->max_sq_desc_sz, dev_cap->max_sq_sg);
1164 	mlx4_dbg(dev, "Max RQ desc size: %d, max RQ S/G: %d\n",
1165 		 dev_cap->max_rq_desc_sz, dev_cap->max_rq_sg);
1166 	mlx4_dbg(dev, "Max GSO size: %d\n", dev_cap->max_gso_sz);
1167 	mlx4_dbg(dev, "Max counters: %d\n", dev_cap->max_counters);
1168 	mlx4_dbg(dev, "Max RSS Table size: %d\n", dev_cap->max_rss_tbl_sz);
1169 	mlx4_dbg(dev, "DMFS high rate steer QPn base: %d\n",
1170 		 dev_cap->dmfs_high_rate_qpn_base);
1171 	mlx4_dbg(dev, "DMFS high rate steer QPn range: %d\n",
1172 		 dev_cap->dmfs_high_rate_qpn_range);
1173 
1174 	if (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT) {
1175 		struct mlx4_rate_limit_caps *rl_caps = &dev_cap->rl_caps;
1176 
1177 		mlx4_dbg(dev, "QP Rate-Limit: #rates %d, unit/val max %d/%d, min %d/%d\n",
1178 			 rl_caps->num_rates, rl_caps->max_unit, rl_caps->max_val,
1179 			 rl_caps->min_unit, rl_caps->min_val);
1180 	}
1181 
1182 	dump_dev_cap_flags(dev, dev_cap->flags);
1183 	dump_dev_cap_flags2(dev, dev_cap->flags2);
1184 }
1185 
1186 int mlx4_QUERY_PORT(struct mlx4_dev *dev, int port, struct mlx4_port_cap *port_cap)
1187 {
1188 	struct mlx4_cmd_mailbox *mailbox;
1189 	u32 *outbox;
1190 	u8 field;
1191 	u32 field32;
1192 	int err;
1193 
1194 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1195 	if (IS_ERR(mailbox))
1196 		return PTR_ERR(mailbox);
1197 	outbox = mailbox->buf;
1198 
1199 	if (dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1200 		err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0, MLX4_CMD_QUERY_DEV_CAP,
1201 				   MLX4_CMD_TIME_CLASS_A,
1202 				   MLX4_CMD_NATIVE);
1203 
1204 		if (err)
1205 			goto out;
1206 
1207 		MLX4_GET(field, outbox, QUERY_DEV_CAP_VL_PORT_OFFSET);
1208 		port_cap->max_vl	   = field >> 4;
1209 		MLX4_GET(field, outbox, QUERY_DEV_CAP_MTU_WIDTH_OFFSET);
1210 		port_cap->ib_mtu	   = field >> 4;
1211 		port_cap->max_port_width = field & 0xf;
1212 		MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_GID_OFFSET);
1213 		port_cap->max_gids	   = 1 << (field & 0xf);
1214 		MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_PKEY_OFFSET);
1215 		port_cap->max_pkeys	   = 1 << (field & 0xf);
1216 	} else {
1217 #define QUERY_PORT_SUPPORTED_TYPE_OFFSET	0x00
1218 #define QUERY_PORT_MTU_OFFSET			0x01
1219 #define QUERY_PORT_ETH_MTU_OFFSET		0x02
1220 #define QUERY_PORT_WIDTH_OFFSET			0x06
1221 #define QUERY_PORT_MAX_GID_PKEY_OFFSET		0x07
1222 #define QUERY_PORT_MAX_MACVLAN_OFFSET		0x0a
1223 #define QUERY_PORT_MAX_VL_OFFSET		0x0b
1224 #define QUERY_PORT_MAC_OFFSET			0x10
1225 #define QUERY_PORT_TRANS_VENDOR_OFFSET		0x18
1226 #define QUERY_PORT_WAVELENGTH_OFFSET		0x1c
1227 #define QUERY_PORT_TRANS_CODE_OFFSET		0x20
1228 
1229 		err = mlx4_cmd_box(dev, 0, mailbox->dma, port, 0, MLX4_CMD_QUERY_PORT,
1230 				   MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
1231 		if (err)
1232 			goto out;
1233 
1234 		MLX4_GET(field, outbox, QUERY_PORT_SUPPORTED_TYPE_OFFSET);
1235 		port_cap->link_state = (field & 0x80) >> 7;
1236 		port_cap->supported_port_types = field & 3;
1237 		port_cap->suggested_type = (field >> 3) & 1;
1238 		port_cap->default_sense = (field >> 4) & 1;
1239 		port_cap->dmfs_optimized_state = (field >> 5) & 1;
1240 		MLX4_GET(field, outbox, QUERY_PORT_MTU_OFFSET);
1241 		port_cap->ib_mtu	   = field & 0xf;
1242 		MLX4_GET(field, outbox, QUERY_PORT_WIDTH_OFFSET);
1243 		port_cap->max_port_width = field & 0xf;
1244 		MLX4_GET(field, outbox, QUERY_PORT_MAX_GID_PKEY_OFFSET);
1245 		port_cap->max_gids	   = 1 << (field >> 4);
1246 		port_cap->max_pkeys	   = 1 << (field & 0xf);
1247 		MLX4_GET(field, outbox, QUERY_PORT_MAX_VL_OFFSET);
1248 		port_cap->max_vl	   = field & 0xf;
1249 		port_cap->max_tc_eth	   = field >> 4;
1250 		MLX4_GET(field, outbox, QUERY_PORT_MAX_MACVLAN_OFFSET);
1251 		port_cap->log_max_macs  = field & 0xf;
1252 		port_cap->log_max_vlans = field >> 4;
1253 		MLX4_GET(port_cap->eth_mtu, outbox, QUERY_PORT_ETH_MTU_OFFSET);
1254 		MLX4_GET(port_cap->def_mac, outbox, QUERY_PORT_MAC_OFFSET);
1255 		MLX4_GET(field32, outbox, QUERY_PORT_TRANS_VENDOR_OFFSET);
1256 		port_cap->trans_type = field32 >> 24;
1257 		port_cap->vendor_oui = field32 & 0xffffff;
1258 		MLX4_GET(port_cap->wavelength, outbox, QUERY_PORT_WAVELENGTH_OFFSET);
1259 		MLX4_GET(port_cap->trans_code, outbox, QUERY_PORT_TRANS_CODE_OFFSET);
1260 	}
1261 
1262 out:
1263 	mlx4_free_cmd_mailbox(dev, mailbox);
1264 	return err;
1265 }
1266 
1267 #define DEV_CAP_EXT_2_FLAG_PFC_COUNTERS	(1 << 28)
1268 #define DEV_CAP_EXT_2_FLAG_VLAN_CONTROL (1 << 26)
1269 #define DEV_CAP_EXT_2_FLAG_80_VFS	(1 << 21)
1270 #define DEV_CAP_EXT_2_FLAG_FSM		(1 << 20)
1271 
1272 int mlx4_QUERY_DEV_CAP_wrapper(struct mlx4_dev *dev, int slave,
1273 			       struct mlx4_vhcr *vhcr,
1274 			       struct mlx4_cmd_mailbox *inbox,
1275 			       struct mlx4_cmd_mailbox *outbox,
1276 			       struct mlx4_cmd_info *cmd)
1277 {
1278 	u64	flags;
1279 	int	err = 0;
1280 	u8	field;
1281 	u16	field16;
1282 	u32	bmme_flags, field32;
1283 	int	real_port;
1284 	int	slave_port;
1285 	int	first_port;
1286 	struct mlx4_active_ports actv_ports;
1287 
1288 	err = mlx4_cmd_box(dev, 0, outbox->dma, 0, 0, MLX4_CMD_QUERY_DEV_CAP,
1289 			   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1290 	if (err)
1291 		return err;
1292 
1293 	disable_unsupported_roce_caps(outbox->buf);
1294 	/* add port mng change event capability and disable mw type 1
1295 	 * unconditionally to slaves
1296 	 */
1297 	MLX4_GET(flags, outbox->buf, QUERY_DEV_CAP_EXT_FLAGS_OFFSET);
1298 	flags |= MLX4_DEV_CAP_FLAG_PORT_MNG_CHG_EV;
1299 	flags &= ~MLX4_DEV_CAP_FLAG_MEM_WINDOW;
1300 	actv_ports = mlx4_get_active_ports(dev, slave);
1301 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
1302 	for (slave_port = 0, real_port = first_port;
1303 	     real_port < first_port +
1304 	     bitmap_weight(actv_ports.ports, dev->caps.num_ports);
1305 	     ++real_port, ++slave_port) {
1306 		if (flags & (MLX4_DEV_CAP_FLAG_WOL_PORT1 << real_port))
1307 			flags |= MLX4_DEV_CAP_FLAG_WOL_PORT1 << slave_port;
1308 		else
1309 			flags &= ~(MLX4_DEV_CAP_FLAG_WOL_PORT1 << slave_port);
1310 	}
1311 	for (; slave_port < dev->caps.num_ports; ++slave_port)
1312 		flags &= ~(MLX4_DEV_CAP_FLAG_WOL_PORT1 << slave_port);
1313 
1314 	/* Not exposing RSS IP fragments to guests */
1315 	flags &= ~MLX4_DEV_CAP_FLAG_RSS_IP_FRAG;
1316 	MLX4_PUT(outbox->buf, flags, QUERY_DEV_CAP_EXT_FLAGS_OFFSET);
1317 
1318 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_VL_PORT_OFFSET);
1319 	field &= ~0x0F;
1320 	field |= bitmap_weight(actv_ports.ports, dev->caps.num_ports) & 0x0F;
1321 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_VL_PORT_OFFSET);
1322 
1323 	/* For guests, disable timestamp */
1324 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_CQ_TS_SUPPORT_OFFSET);
1325 	field &= 0x7f;
1326 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_CQ_TS_SUPPORT_OFFSET);
1327 
1328 	/* For guests, disable vxlan tunneling and QoS support */
1329 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_VXLAN);
1330 	field &= 0xd7;
1331 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_VXLAN);
1332 
1333 	/* For guests, disable port BEACON */
1334 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_PORT_BEACON_OFFSET);
1335 	field &= 0x7f;
1336 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_PORT_BEACON_OFFSET);
1337 
1338 	/* For guests, report Blueflame disabled */
1339 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_BF_OFFSET);
1340 	field &= 0x7f;
1341 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_BF_OFFSET);
1342 
1343 	/* For guests, disable mw type 2 and port remap*/
1344 	MLX4_GET(bmme_flags, outbox->buf, QUERY_DEV_CAP_BMME_FLAGS_OFFSET);
1345 	bmme_flags &= ~MLX4_BMME_FLAG_TYPE_2_WIN;
1346 	bmme_flags &= ~MLX4_FLAG_PORT_REMAP;
1347 	MLX4_PUT(outbox->buf, bmme_flags, QUERY_DEV_CAP_BMME_FLAGS_OFFSET);
1348 
1349 	/* turn off device-managed steering capability if not enabled */
1350 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) {
1351 		MLX4_GET(field, outbox->buf,
1352 			 QUERY_DEV_CAP_FLOW_STEERING_RANGE_EN_OFFSET);
1353 		field &= 0x7f;
1354 		MLX4_PUT(outbox->buf, field,
1355 			 QUERY_DEV_CAP_FLOW_STEERING_RANGE_EN_OFFSET);
1356 	}
1357 
1358 	/* turn off ipoib managed steering for guests */
1359 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_FLOW_STEERING_IPOIB_OFFSET);
1360 	field &= ~0x80;
1361 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_FLOW_STEERING_IPOIB_OFFSET);
1362 
1363 	/* turn off host side virt features (VST, FSM, etc) for guests */
1364 	MLX4_GET(field32, outbox->buf, QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET);
1365 	field32 &= ~(DEV_CAP_EXT_2_FLAG_VLAN_CONTROL | DEV_CAP_EXT_2_FLAG_80_VFS |
1366 		     DEV_CAP_EXT_2_FLAG_FSM | DEV_CAP_EXT_2_FLAG_PFC_COUNTERS);
1367 	MLX4_PUT(outbox->buf, field32, QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET);
1368 
1369 	/* turn off QCN for guests */
1370 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_ECN_QCN_VER_OFFSET);
1371 	field &= 0xfe;
1372 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_ECN_QCN_VER_OFFSET);
1373 
1374 	/* turn off QP max-rate limiting for guests */
1375 	field16 = 0;
1376 	MLX4_PUT(outbox->buf, field16, QUERY_DEV_CAP_QP_RATE_LIMIT_NUM_OFFSET);
1377 
1378 	/* turn off QoS per VF support for guests */
1379 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_CQ_EQ_CACHE_LINE_STRIDE);
1380 	field &= 0xef;
1381 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_CQ_EQ_CACHE_LINE_STRIDE);
1382 
1383 	/* turn off ignore FCS feature for guests */
1384 	MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_CONFIG_DEV_OFFSET);
1385 	field &= 0xfb;
1386 	MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_CONFIG_DEV_OFFSET);
1387 
1388 	return 0;
1389 }
1390 
1391 static void disable_unsupported_roce_caps(void *buf)
1392 {
1393 	u32 flags;
1394 
1395 	MLX4_GET(flags, buf, QUERY_DEV_CAP_EXT_FLAGS_OFFSET);
1396 	flags &= ~(1UL << 31);
1397 	MLX4_PUT(buf, flags, QUERY_DEV_CAP_EXT_FLAGS_OFFSET);
1398 	MLX4_GET(flags, buf, QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET);
1399 	flags &= ~(1UL << 24);
1400 	MLX4_PUT(buf, flags, QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET);
1401 	MLX4_GET(flags, buf, QUERY_DEV_CAP_BMME_FLAGS_OFFSET);
1402 	flags &= ~(MLX4_FLAG_ROCE_V1_V2);
1403 	MLX4_PUT(buf, flags, QUERY_DEV_CAP_BMME_FLAGS_OFFSET);
1404 }
1405 
1406 int mlx4_QUERY_PORT_wrapper(struct mlx4_dev *dev, int slave,
1407 			    struct mlx4_vhcr *vhcr,
1408 			    struct mlx4_cmd_mailbox *inbox,
1409 			    struct mlx4_cmd_mailbox *outbox,
1410 			    struct mlx4_cmd_info *cmd)
1411 {
1412 	struct mlx4_priv *priv = mlx4_priv(dev);
1413 	u64 def_mac;
1414 	u8 port_type;
1415 	u16 short_field;
1416 	int err;
1417 	int admin_link_state;
1418 	int port = mlx4_slave_convert_port(dev, slave,
1419 					   vhcr->in_modifier & 0xFF);
1420 
1421 #define MLX4_VF_PORT_NO_LINK_SENSE_MASK	0xE0
1422 #define MLX4_PORT_LINK_UP_MASK		0x80
1423 #define QUERY_PORT_CUR_MAX_PKEY_OFFSET	0x0c
1424 #define QUERY_PORT_CUR_MAX_GID_OFFSET	0x0e
1425 
1426 	if (port < 0)
1427 		return -EINVAL;
1428 
1429 	/* Protect against untrusted guests: enforce that this is the
1430 	 * QUERY_PORT general query.
1431 	 */
1432 	if (vhcr->op_modifier || vhcr->in_modifier & ~0xFF)
1433 		return -EINVAL;
1434 
1435 	vhcr->in_modifier = port;
1436 
1437 	err = mlx4_cmd_box(dev, 0, outbox->dma, vhcr->in_modifier, 0,
1438 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
1439 			   MLX4_CMD_NATIVE);
1440 
1441 	if (!err && dev->caps.function != slave) {
1442 		def_mac = priv->mfunc.master.vf_oper[slave].vport[vhcr->in_modifier].state.mac;
1443 		MLX4_PUT(outbox->buf, def_mac, QUERY_PORT_MAC_OFFSET);
1444 
1445 		/* get port type - currently only eth is enabled */
1446 		MLX4_GET(port_type, outbox->buf,
1447 			 QUERY_PORT_SUPPORTED_TYPE_OFFSET);
1448 
1449 		/* No link sensing allowed */
1450 		port_type &= MLX4_VF_PORT_NO_LINK_SENSE_MASK;
1451 		/* set port type to currently operating port type */
1452 		port_type |= (dev->caps.port_type[vhcr->in_modifier] & 0x3);
1453 
1454 		admin_link_state = priv->mfunc.master.vf_oper[slave].vport[vhcr->in_modifier].state.link_state;
1455 		if (IFLA_VF_LINK_STATE_ENABLE == admin_link_state)
1456 			port_type |= MLX4_PORT_LINK_UP_MASK;
1457 		else if (IFLA_VF_LINK_STATE_DISABLE == admin_link_state)
1458 			port_type &= ~MLX4_PORT_LINK_UP_MASK;
1459 		else if (IFLA_VF_LINK_STATE_AUTO == admin_link_state && mlx4_is_bonded(dev)) {
1460 			int other_port = (port == 1) ? 2 : 1;
1461 			struct mlx4_port_cap port_cap;
1462 
1463 			err = mlx4_QUERY_PORT(dev, other_port, &port_cap);
1464 			if (err)
1465 				goto out;
1466 			port_type |= (port_cap.link_state << 7);
1467 		}
1468 
1469 		MLX4_PUT(outbox->buf, port_type,
1470 			 QUERY_PORT_SUPPORTED_TYPE_OFFSET);
1471 
1472 		if (dev->caps.port_type[vhcr->in_modifier] == MLX4_PORT_TYPE_ETH)
1473 			short_field = mlx4_get_slave_num_gids(dev, slave, port);
1474 		else
1475 			short_field = 1; /* slave max gids */
1476 		MLX4_PUT(outbox->buf, short_field,
1477 			 QUERY_PORT_CUR_MAX_GID_OFFSET);
1478 
1479 		short_field = dev->caps.pkey_table_len[vhcr->in_modifier];
1480 		MLX4_PUT(outbox->buf, short_field,
1481 			 QUERY_PORT_CUR_MAX_PKEY_OFFSET);
1482 	}
1483 out:
1484 	return err;
1485 }
1486 
1487 int mlx4_get_slave_pkey_gid_tbl_len(struct mlx4_dev *dev, u8 port,
1488 				    int *gid_tbl_len, int *pkey_tbl_len)
1489 {
1490 	struct mlx4_cmd_mailbox *mailbox;
1491 	u32			*outbox;
1492 	u16			field;
1493 	int			err;
1494 
1495 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1496 	if (IS_ERR(mailbox))
1497 		return PTR_ERR(mailbox);
1498 
1499 	err =  mlx4_cmd_box(dev, 0, mailbox->dma, port, 0,
1500 			    MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
1501 			    MLX4_CMD_WRAPPED);
1502 	if (err)
1503 		goto out;
1504 
1505 	outbox = mailbox->buf;
1506 
1507 	MLX4_GET(field, outbox, QUERY_PORT_CUR_MAX_GID_OFFSET);
1508 	*gid_tbl_len = field;
1509 
1510 	MLX4_GET(field, outbox, QUERY_PORT_CUR_MAX_PKEY_OFFSET);
1511 	*pkey_tbl_len = field;
1512 
1513 out:
1514 	mlx4_free_cmd_mailbox(dev, mailbox);
1515 	return err;
1516 }
1517 EXPORT_SYMBOL(mlx4_get_slave_pkey_gid_tbl_len);
1518 
1519 int mlx4_map_cmd(struct mlx4_dev *dev, u16 op, struct mlx4_icm *icm, u64 virt)
1520 {
1521 	struct mlx4_cmd_mailbox *mailbox;
1522 	struct mlx4_icm_iter iter;
1523 	__be64 *pages;
1524 	int lg;
1525 	int nent = 0;
1526 	int i;
1527 	int err = 0;
1528 	int ts = 0, tc = 0;
1529 
1530 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1531 	if (IS_ERR(mailbox))
1532 		return PTR_ERR(mailbox);
1533 	pages = mailbox->buf;
1534 
1535 	for (mlx4_icm_first(icm, &iter);
1536 	     !mlx4_icm_last(&iter);
1537 	     mlx4_icm_next(&iter)) {
1538 		/*
1539 		 * We have to pass pages that are aligned to their
1540 		 * size, so find the least significant 1 in the
1541 		 * address or size and use that as our log2 size.
1542 		 */
1543 		lg = ffs(mlx4_icm_addr(&iter) | mlx4_icm_size(&iter)) - 1;
1544 		if (lg < MLX4_ICM_PAGE_SHIFT) {
1545 			mlx4_warn(dev, "Got FW area not aligned to %d (%llx/%lx)\n",
1546 				  MLX4_ICM_PAGE_SIZE,
1547 				  (unsigned long long) mlx4_icm_addr(&iter),
1548 				  mlx4_icm_size(&iter));
1549 			err = -EINVAL;
1550 			goto out;
1551 		}
1552 
1553 		for (i = 0; i < mlx4_icm_size(&iter) >> lg; ++i) {
1554 			if (virt != -1) {
1555 				pages[nent * 2] = cpu_to_be64(virt);
1556 				virt += 1ULL << lg;
1557 			}
1558 
1559 			pages[nent * 2 + 1] =
1560 				cpu_to_be64((mlx4_icm_addr(&iter) + (i << lg)) |
1561 					    (lg - MLX4_ICM_PAGE_SHIFT));
1562 			ts += 1 << (lg - 10);
1563 			++tc;
1564 
1565 			if (++nent == MLX4_MAILBOX_SIZE / 16) {
1566 				err = mlx4_cmd(dev, mailbox->dma, nent, 0, op,
1567 						MLX4_CMD_TIME_CLASS_B,
1568 						MLX4_CMD_NATIVE);
1569 				if (err)
1570 					goto out;
1571 				nent = 0;
1572 			}
1573 		}
1574 	}
1575 
1576 	if (nent)
1577 		err = mlx4_cmd(dev, mailbox->dma, nent, 0, op,
1578 			       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
1579 	if (err)
1580 		goto out;
1581 
1582 	switch (op) {
1583 	case MLX4_CMD_MAP_FA:
1584 		mlx4_dbg(dev, "Mapped %d chunks/%d KB for FW\n", tc, ts);
1585 		break;
1586 	case MLX4_CMD_MAP_ICM_AUX:
1587 		mlx4_dbg(dev, "Mapped %d chunks/%d KB for ICM aux\n", tc, ts);
1588 		break;
1589 	case MLX4_CMD_MAP_ICM:
1590 		mlx4_dbg(dev, "Mapped %d chunks/%d KB at %llx for ICM\n",
1591 			 tc, ts, (unsigned long long) virt - (ts << 10));
1592 		break;
1593 	}
1594 
1595 out:
1596 	mlx4_free_cmd_mailbox(dev, mailbox);
1597 	return err;
1598 }
1599 
1600 int mlx4_MAP_FA(struct mlx4_dev *dev, struct mlx4_icm *icm)
1601 {
1602 	return mlx4_map_cmd(dev, MLX4_CMD_MAP_FA, icm, -1);
1603 }
1604 
1605 int mlx4_UNMAP_FA(struct mlx4_dev *dev)
1606 {
1607 	return mlx4_cmd(dev, 0, 0, 0, MLX4_CMD_UNMAP_FA,
1608 			MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
1609 }
1610 
1611 
1612 int mlx4_RUN_FW(struct mlx4_dev *dev)
1613 {
1614 	return mlx4_cmd(dev, 0, 0, 0, MLX4_CMD_RUN_FW,
1615 			MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1616 }
1617 
1618 int mlx4_QUERY_FW(struct mlx4_dev *dev)
1619 {
1620 	struct mlx4_fw  *fw  = &mlx4_priv(dev)->fw;
1621 	struct mlx4_cmd *cmd = &mlx4_priv(dev)->cmd;
1622 	struct mlx4_cmd_mailbox *mailbox;
1623 	u32 *outbox;
1624 	int err = 0;
1625 	u64 fw_ver;
1626 	u16 cmd_if_rev;
1627 	u8 lg;
1628 
1629 #define QUERY_FW_OUT_SIZE             0x100
1630 #define QUERY_FW_VER_OFFSET            0x00
1631 #define QUERY_FW_PPF_ID		       0x09
1632 #define QUERY_FW_CMD_IF_REV_OFFSET     0x0a
1633 #define QUERY_FW_MAX_CMD_OFFSET        0x0f
1634 #define QUERY_FW_ERR_START_OFFSET      0x30
1635 #define QUERY_FW_ERR_SIZE_OFFSET       0x38
1636 #define QUERY_FW_ERR_BAR_OFFSET        0x3c
1637 
1638 #define QUERY_FW_SIZE_OFFSET           0x00
1639 #define QUERY_FW_CLR_INT_BASE_OFFSET   0x20
1640 #define QUERY_FW_CLR_INT_BAR_OFFSET    0x28
1641 
1642 #define QUERY_FW_COMM_BASE_OFFSET      0x40
1643 #define QUERY_FW_COMM_BAR_OFFSET       0x48
1644 
1645 #define QUERY_FW_CLOCK_OFFSET	       0x50
1646 #define QUERY_FW_CLOCK_BAR	       0x58
1647 
1648 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1649 	if (IS_ERR(mailbox))
1650 		return PTR_ERR(mailbox);
1651 	outbox = mailbox->buf;
1652 
1653 	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0, MLX4_CMD_QUERY_FW,
1654 			    MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1655 	if (err)
1656 		goto out;
1657 
1658 	MLX4_GET(fw_ver, outbox, QUERY_FW_VER_OFFSET);
1659 	/*
1660 	 * FW subminor version is at more significant bits than minor
1661 	 * version, so swap here.
1662 	 */
1663 	dev->caps.fw_ver = (fw_ver & 0xffff00000000ull) |
1664 		((fw_ver & 0xffff0000ull) >> 16) |
1665 		((fw_ver & 0x0000ffffull) << 16);
1666 
1667 	MLX4_GET(lg, outbox, QUERY_FW_PPF_ID);
1668 	dev->caps.function = lg;
1669 
1670 	if (mlx4_is_slave(dev))
1671 		goto out;
1672 
1673 
1674 	MLX4_GET(cmd_if_rev, outbox, QUERY_FW_CMD_IF_REV_OFFSET);
1675 	if (cmd_if_rev < MLX4_COMMAND_INTERFACE_MIN_REV ||
1676 	    cmd_if_rev > MLX4_COMMAND_INTERFACE_MAX_REV) {
1677 		mlx4_err(dev, "Installed FW has unsupported command interface revision %d\n",
1678 			 cmd_if_rev);
1679 		mlx4_err(dev, "(Installed FW version is %d.%d.%03d)\n",
1680 			 (int) (dev->caps.fw_ver >> 32),
1681 			 (int) (dev->caps.fw_ver >> 16) & 0xffff,
1682 			 (int) dev->caps.fw_ver & 0xffff);
1683 		mlx4_err(dev, "This driver version supports only revisions %d to %d\n",
1684 			 MLX4_COMMAND_INTERFACE_MIN_REV, MLX4_COMMAND_INTERFACE_MAX_REV);
1685 		err = -ENODEV;
1686 		goto out;
1687 	}
1688 
1689 	if (cmd_if_rev < MLX4_COMMAND_INTERFACE_NEW_PORT_CMDS)
1690 		dev->flags |= MLX4_FLAG_OLD_PORT_CMDS;
1691 
1692 	MLX4_GET(lg, outbox, QUERY_FW_MAX_CMD_OFFSET);
1693 	cmd->max_cmds = 1 << lg;
1694 
1695 	mlx4_dbg(dev, "FW version %d.%d.%03d (cmd intf rev %d), max commands %d\n",
1696 		 (int) (dev->caps.fw_ver >> 32),
1697 		 (int) (dev->caps.fw_ver >> 16) & 0xffff,
1698 		 (int) dev->caps.fw_ver & 0xffff,
1699 		 cmd_if_rev, cmd->max_cmds);
1700 
1701 	MLX4_GET(fw->catas_offset, outbox, QUERY_FW_ERR_START_OFFSET);
1702 	MLX4_GET(fw->catas_size,   outbox, QUERY_FW_ERR_SIZE_OFFSET);
1703 	MLX4_GET(fw->catas_bar,    outbox, QUERY_FW_ERR_BAR_OFFSET);
1704 	fw->catas_bar = (fw->catas_bar >> 6) * 2;
1705 
1706 	mlx4_dbg(dev, "Catastrophic error buffer at 0x%llx, size 0x%x, BAR %d\n",
1707 		 (unsigned long long) fw->catas_offset, fw->catas_size, fw->catas_bar);
1708 
1709 	MLX4_GET(fw->fw_pages,     outbox, QUERY_FW_SIZE_OFFSET);
1710 	MLX4_GET(fw->clr_int_base, outbox, QUERY_FW_CLR_INT_BASE_OFFSET);
1711 	MLX4_GET(fw->clr_int_bar,  outbox, QUERY_FW_CLR_INT_BAR_OFFSET);
1712 	fw->clr_int_bar = (fw->clr_int_bar >> 6) * 2;
1713 
1714 	MLX4_GET(fw->comm_base, outbox, QUERY_FW_COMM_BASE_OFFSET);
1715 	MLX4_GET(fw->comm_bar,  outbox, QUERY_FW_COMM_BAR_OFFSET);
1716 	fw->comm_bar = (fw->comm_bar >> 6) * 2;
1717 	mlx4_dbg(dev, "Communication vector bar:%d offset:0x%llx\n",
1718 		 fw->comm_bar, fw->comm_base);
1719 	mlx4_dbg(dev, "FW size %d KB\n", fw->fw_pages >> 2);
1720 
1721 	MLX4_GET(fw->clock_offset, outbox, QUERY_FW_CLOCK_OFFSET);
1722 	MLX4_GET(fw->clock_bar,    outbox, QUERY_FW_CLOCK_BAR);
1723 	fw->clock_bar = (fw->clock_bar >> 6) * 2;
1724 	mlx4_dbg(dev, "Internal clock bar:%d offset:0x%llx\n",
1725 		 fw->clock_bar, fw->clock_offset);
1726 
1727 	/*
1728 	 * Round up number of system pages needed in case
1729 	 * MLX4_ICM_PAGE_SIZE < PAGE_SIZE.
1730 	 */
1731 	fw->fw_pages =
1732 		ALIGN(fw->fw_pages, PAGE_SIZE / MLX4_ICM_PAGE_SIZE) >>
1733 		(PAGE_SHIFT - MLX4_ICM_PAGE_SHIFT);
1734 
1735 	mlx4_dbg(dev, "Clear int @ %llx, BAR %d\n",
1736 		 (unsigned long long) fw->clr_int_base, fw->clr_int_bar);
1737 
1738 out:
1739 	mlx4_free_cmd_mailbox(dev, mailbox);
1740 	return err;
1741 }
1742 
1743 int mlx4_QUERY_FW_wrapper(struct mlx4_dev *dev, int slave,
1744 			  struct mlx4_vhcr *vhcr,
1745 			  struct mlx4_cmd_mailbox *inbox,
1746 			  struct mlx4_cmd_mailbox *outbox,
1747 			  struct mlx4_cmd_info *cmd)
1748 {
1749 	u8 *outbuf;
1750 	int err;
1751 
1752 	outbuf = outbox->buf;
1753 	err = mlx4_cmd_box(dev, 0, outbox->dma, 0, 0, MLX4_CMD_QUERY_FW,
1754 			    MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1755 	if (err)
1756 		return err;
1757 
1758 	/* for slaves, set pci PPF ID to invalid and zero out everything
1759 	 * else except FW version */
1760 	outbuf[0] = outbuf[1] = 0;
1761 	memset(&outbuf[8], 0, QUERY_FW_OUT_SIZE - 8);
1762 	outbuf[QUERY_FW_PPF_ID] = MLX4_INVALID_SLAVE_ID;
1763 
1764 	return 0;
1765 }
1766 
1767 static void get_board_id(void *vsd, char *board_id)
1768 {
1769 	int i;
1770 
1771 #define VSD_OFFSET_SIG1		0x00
1772 #define VSD_OFFSET_SIG2		0xde
1773 #define VSD_OFFSET_MLX_BOARD_ID	0xd0
1774 #define VSD_OFFSET_TS_BOARD_ID	0x20
1775 
1776 #define VSD_SIGNATURE_TOPSPIN	0x5ad
1777 
1778 	memset(board_id, 0, MLX4_BOARD_ID_LEN);
1779 
1780 	if (be16_to_cpup(vsd + VSD_OFFSET_SIG1) == VSD_SIGNATURE_TOPSPIN &&
1781 	    be16_to_cpup(vsd + VSD_OFFSET_SIG2) == VSD_SIGNATURE_TOPSPIN) {
1782 		strscpy(board_id, vsd + VSD_OFFSET_TS_BOARD_ID, MLX4_BOARD_ID_LEN);
1783 	} else {
1784 		/*
1785 		 * The board ID is a string but the firmware byte
1786 		 * swaps each 4-byte word before passing it back to
1787 		 * us.  Therefore we need to swab it before printing.
1788 		 */
1789 		u32 *bid_u32 = (u32 *)board_id;
1790 
1791 		for (i = 0; i < 4; ++i) {
1792 			u32 *addr;
1793 			u32 val;
1794 
1795 			addr = (u32 *) (vsd + VSD_OFFSET_MLX_BOARD_ID + i * 4);
1796 			val = get_unaligned(addr);
1797 			val = swab32(val);
1798 			put_unaligned(val, &bid_u32[i]);
1799 		}
1800 	}
1801 }
1802 
1803 int mlx4_QUERY_ADAPTER(struct mlx4_dev *dev, struct mlx4_adapter *adapter)
1804 {
1805 	struct mlx4_cmd_mailbox *mailbox;
1806 	u32 *outbox;
1807 	int err;
1808 
1809 #define QUERY_ADAPTER_OUT_SIZE             0x100
1810 #define QUERY_ADAPTER_INTA_PIN_OFFSET      0x10
1811 #define QUERY_ADAPTER_VSD_OFFSET           0x20
1812 
1813 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1814 	if (IS_ERR(mailbox))
1815 		return PTR_ERR(mailbox);
1816 	outbox = mailbox->buf;
1817 
1818 	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0, MLX4_CMD_QUERY_ADAPTER,
1819 			   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1820 	if (err)
1821 		goto out;
1822 
1823 	MLX4_GET(adapter->inta_pin, outbox,    QUERY_ADAPTER_INTA_PIN_OFFSET);
1824 
1825 	get_board_id(outbox + QUERY_ADAPTER_VSD_OFFSET / 4,
1826 		     adapter->board_id);
1827 
1828 out:
1829 	mlx4_free_cmd_mailbox(dev, mailbox);
1830 	return err;
1831 }
1832 
1833 int mlx4_INIT_HCA(struct mlx4_dev *dev, struct mlx4_init_hca_param *param)
1834 {
1835 	struct mlx4_cmd_mailbox *mailbox;
1836 	__be32 *inbox;
1837 	int err;
1838 	static const u8 a0_dmfs_hw_steering[] =  {
1839 		[MLX4_STEERING_DMFS_A0_DEFAULT]		= 0,
1840 		[MLX4_STEERING_DMFS_A0_DYNAMIC]		= 1,
1841 		[MLX4_STEERING_DMFS_A0_STATIC]		= 2,
1842 		[MLX4_STEERING_DMFS_A0_DISABLE]		= 3
1843 	};
1844 
1845 #define INIT_HCA_IN_SIZE		 0x200
1846 #define INIT_HCA_VERSION_OFFSET		 0x000
1847 #define	 INIT_HCA_VERSION		 2
1848 #define INIT_HCA_VXLAN_OFFSET		 0x0c
1849 #define INIT_HCA_CACHELINE_SZ_OFFSET	 0x0e
1850 #define INIT_HCA_FLAGS_OFFSET		 0x014
1851 #define INIT_HCA_RECOVERABLE_ERROR_EVENT_OFFSET 0x018
1852 #define INIT_HCA_QPC_OFFSET		 0x020
1853 #define	 INIT_HCA_QPC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x10)
1854 #define	 INIT_HCA_LOG_QP_OFFSET		 (INIT_HCA_QPC_OFFSET + 0x17)
1855 #define	 INIT_HCA_SRQC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x28)
1856 #define	 INIT_HCA_LOG_SRQ_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x2f)
1857 #define	 INIT_HCA_CQC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x30)
1858 #define	 INIT_HCA_LOG_CQ_OFFSET		 (INIT_HCA_QPC_OFFSET + 0x37)
1859 #define	 INIT_HCA_EQE_CQE_OFFSETS	 (INIT_HCA_QPC_OFFSET + 0x38)
1860 #define	 INIT_HCA_EQE_CQE_STRIDE_OFFSET  (INIT_HCA_QPC_OFFSET + 0x3b)
1861 #define	 INIT_HCA_ALTC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x40)
1862 #define	 INIT_HCA_AUXC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x50)
1863 #define	 INIT_HCA_EQC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x60)
1864 #define	 INIT_HCA_LOG_EQ_OFFSET		 (INIT_HCA_QPC_OFFSET + 0x67)
1865 #define	INIT_HCA_NUM_SYS_EQS_OFFSET	(INIT_HCA_QPC_OFFSET + 0x6a)
1866 #define	 INIT_HCA_RDMARC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x70)
1867 #define	 INIT_HCA_LOG_RD_OFFSET		 (INIT_HCA_QPC_OFFSET + 0x77)
1868 #define INIT_HCA_MCAST_OFFSET		 0x0c0
1869 #define	 INIT_HCA_MC_BASE_OFFSET	 (INIT_HCA_MCAST_OFFSET + 0x00)
1870 #define	 INIT_HCA_LOG_MC_ENTRY_SZ_OFFSET (INIT_HCA_MCAST_OFFSET + 0x13)
1871 #define	 INIT_HCA_LOG_MC_HASH_SZ_OFFSET	 (INIT_HCA_MCAST_OFFSET + 0x17)
1872 #define  INIT_HCA_UC_STEERING_OFFSET	 (INIT_HCA_MCAST_OFFSET + 0x18)
1873 #define	 INIT_HCA_LOG_MC_TABLE_SZ_OFFSET (INIT_HCA_MCAST_OFFSET + 0x1b)
1874 #define  INIT_HCA_DEVICE_MANAGED_FLOW_STEERING_EN	0x6
1875 #define  INIT_HCA_DRIVER_VERSION_OFFSET   0x140
1876 #define  INIT_HCA_DRIVER_VERSION_SZ       0x40
1877 #define  INIT_HCA_FS_PARAM_OFFSET         0x1d0
1878 #define  INIT_HCA_FS_BASE_OFFSET          (INIT_HCA_FS_PARAM_OFFSET + 0x00)
1879 #define  INIT_HCA_FS_LOG_ENTRY_SZ_OFFSET  (INIT_HCA_FS_PARAM_OFFSET + 0x13)
1880 #define  INIT_HCA_FS_A0_OFFSET		  (INIT_HCA_FS_PARAM_OFFSET + 0x18)
1881 #define  INIT_HCA_FS_LOG_TABLE_SZ_OFFSET  (INIT_HCA_FS_PARAM_OFFSET + 0x1b)
1882 #define  INIT_HCA_FS_ETH_BITS_OFFSET      (INIT_HCA_FS_PARAM_OFFSET + 0x21)
1883 #define  INIT_HCA_FS_ETH_NUM_ADDRS_OFFSET (INIT_HCA_FS_PARAM_OFFSET + 0x22)
1884 #define  INIT_HCA_FS_IB_BITS_OFFSET       (INIT_HCA_FS_PARAM_OFFSET + 0x25)
1885 #define  INIT_HCA_FS_IB_NUM_ADDRS_OFFSET  (INIT_HCA_FS_PARAM_OFFSET + 0x26)
1886 #define INIT_HCA_TPT_OFFSET		 0x0f0
1887 #define	 INIT_HCA_DMPT_BASE_OFFSET	 (INIT_HCA_TPT_OFFSET + 0x00)
1888 #define  INIT_HCA_TPT_MW_OFFSET		 (INIT_HCA_TPT_OFFSET + 0x08)
1889 #define	 INIT_HCA_LOG_MPT_SZ_OFFSET	 (INIT_HCA_TPT_OFFSET + 0x0b)
1890 #define	 INIT_HCA_MTT_BASE_OFFSET	 (INIT_HCA_TPT_OFFSET + 0x10)
1891 #define	 INIT_HCA_CMPT_BASE_OFFSET	 (INIT_HCA_TPT_OFFSET + 0x18)
1892 #define INIT_HCA_UAR_OFFSET		 0x120
1893 #define	 INIT_HCA_LOG_UAR_SZ_OFFSET	 (INIT_HCA_UAR_OFFSET + 0x0a)
1894 #define  INIT_HCA_UAR_PAGE_SZ_OFFSET     (INIT_HCA_UAR_OFFSET + 0x0b)
1895 
1896 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1897 	if (IS_ERR(mailbox))
1898 		return PTR_ERR(mailbox);
1899 	inbox = mailbox->buf;
1900 
1901 	*((u8 *) mailbox->buf + INIT_HCA_VERSION_OFFSET) = INIT_HCA_VERSION;
1902 
1903 	*((u8 *) mailbox->buf + INIT_HCA_CACHELINE_SZ_OFFSET) =
1904 		((ilog2(cache_line_size()) - 4) << 5) | (1 << 4);
1905 
1906 #if defined(__LITTLE_ENDIAN)
1907 	*(inbox + INIT_HCA_FLAGS_OFFSET / 4) &= ~cpu_to_be32(1 << 1);
1908 #elif defined(__BIG_ENDIAN)
1909 	*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1 << 1);
1910 #else
1911 #error Host endianness not defined
1912 #endif
1913 	/* Check port for UD address vector: */
1914 	*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1);
1915 
1916 	/* Enable IPoIB checksumming if we can: */
1917 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
1918 		*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1 << 3);
1919 
1920 	/* Enable QoS support if module parameter set */
1921 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG && enable_qos)
1922 		*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1 << 2);
1923 
1924 	/* enable counters */
1925 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS)
1926 		*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1 << 4);
1927 
1928 	/* Enable RSS spread to fragmented IP packets when supported */
1929 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_RSS_IP_FRAG)
1930 		*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1 << 13);
1931 
1932 	/* CX3 is capable of extending CQEs/EQEs from 32 to 64 bytes */
1933 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_64B_EQE) {
1934 		*(inbox + INIT_HCA_EQE_CQE_OFFSETS / 4) |= cpu_to_be32(1 << 29);
1935 		dev->caps.eqe_size   = 64;
1936 		dev->caps.eqe_factor = 1;
1937 	} else {
1938 		dev->caps.eqe_size   = 32;
1939 		dev->caps.eqe_factor = 0;
1940 	}
1941 
1942 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_64B_CQE) {
1943 		*(inbox + INIT_HCA_EQE_CQE_OFFSETS / 4) |= cpu_to_be32(1 << 30);
1944 		dev->caps.cqe_size   = 64;
1945 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1946 	} else {
1947 		dev->caps.cqe_size   = 32;
1948 	}
1949 
1950 	/* CX3 is capable of extending CQEs\EQEs to strides larger than 64B */
1951 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) &&
1952 	    (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE)) {
1953 		dev->caps.eqe_size = cache_line_size();
1954 		dev->caps.cqe_size = cache_line_size();
1955 		dev->caps.eqe_factor = 0;
1956 		MLX4_PUT(inbox, (u8)((ilog2(dev->caps.eqe_size) - 5) << 4 |
1957 				      (ilog2(dev->caps.eqe_size) - 5)),
1958 			 INIT_HCA_EQE_CQE_STRIDE_OFFSET);
1959 
1960 		/* User still need to know to support CQE > 32B */
1961 		dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1962 	}
1963 
1964 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RECOVERABLE_ERROR_EVENT)
1965 		*(inbox + INIT_HCA_RECOVERABLE_ERROR_EVENT_OFFSET / 4) |= cpu_to_be32(1 << 31);
1966 
1967 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DRIVER_VERSION_TO_FW) {
1968 		u8 *dst = (u8 *)(inbox + INIT_HCA_DRIVER_VERSION_OFFSET / 4);
1969 
1970 		strncpy(dst, DRV_NAME_FOR_FW, INIT_HCA_DRIVER_VERSION_SZ - 1);
1971 		mlx4_dbg(dev, "Reporting Driver Version to FW: %s\n", dst);
1972 	}
1973 
1974 	/* QPC/EEC/CQC/EQC/RDMARC attributes */
1975 
1976 	MLX4_PUT(inbox, param->qpc_base,      INIT_HCA_QPC_BASE_OFFSET);
1977 	MLX4_PUT(inbox, param->log_num_qps,   INIT_HCA_LOG_QP_OFFSET);
1978 	MLX4_PUT(inbox, param->srqc_base,     INIT_HCA_SRQC_BASE_OFFSET);
1979 	MLX4_PUT(inbox, param->log_num_srqs,  INIT_HCA_LOG_SRQ_OFFSET);
1980 	MLX4_PUT(inbox, param->cqc_base,      INIT_HCA_CQC_BASE_OFFSET);
1981 	MLX4_PUT(inbox, param->log_num_cqs,   INIT_HCA_LOG_CQ_OFFSET);
1982 	MLX4_PUT(inbox, param->altc_base,     INIT_HCA_ALTC_BASE_OFFSET);
1983 	MLX4_PUT(inbox, param->auxc_base,     INIT_HCA_AUXC_BASE_OFFSET);
1984 	MLX4_PUT(inbox, param->eqc_base,      INIT_HCA_EQC_BASE_OFFSET);
1985 	MLX4_PUT(inbox, param->log_num_eqs,   INIT_HCA_LOG_EQ_OFFSET);
1986 	MLX4_PUT(inbox, param->num_sys_eqs,   INIT_HCA_NUM_SYS_EQS_OFFSET);
1987 	MLX4_PUT(inbox, param->rdmarc_base,   INIT_HCA_RDMARC_BASE_OFFSET);
1988 	MLX4_PUT(inbox, param->log_rd_per_qp, INIT_HCA_LOG_RD_OFFSET);
1989 
1990 	/* steering attributes */
1991 	if (dev->caps.steering_mode ==
1992 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1993 		*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |=
1994 			cpu_to_be32(1 <<
1995 				    INIT_HCA_DEVICE_MANAGED_FLOW_STEERING_EN);
1996 
1997 		MLX4_PUT(inbox, param->mc_base, INIT_HCA_FS_BASE_OFFSET);
1998 		MLX4_PUT(inbox, param->log_mc_entry_sz,
1999 			 INIT_HCA_FS_LOG_ENTRY_SZ_OFFSET);
2000 		MLX4_PUT(inbox, param->log_mc_table_sz,
2001 			 INIT_HCA_FS_LOG_TABLE_SZ_OFFSET);
2002 		/* Enable Ethernet flow steering
2003 		 * with udp unicast and tcp unicast
2004 		 */
2005 		if (dev->caps.dmfs_high_steer_mode !=
2006 		    MLX4_STEERING_DMFS_A0_STATIC)
2007 			MLX4_PUT(inbox,
2008 				 (u8)(MLX4_FS_UDP_UC_EN | MLX4_FS_TCP_UC_EN),
2009 				 INIT_HCA_FS_ETH_BITS_OFFSET);
2010 		MLX4_PUT(inbox, (u16) MLX4_FS_NUM_OF_L2_ADDR,
2011 			 INIT_HCA_FS_ETH_NUM_ADDRS_OFFSET);
2012 		/* Enable IPoIB flow steering
2013 		 * with udp unicast and tcp unicast
2014 		 */
2015 		MLX4_PUT(inbox, (u8) (MLX4_FS_UDP_UC_EN | MLX4_FS_TCP_UC_EN),
2016 			 INIT_HCA_FS_IB_BITS_OFFSET);
2017 		MLX4_PUT(inbox, (u16) MLX4_FS_NUM_OF_L2_ADDR,
2018 			 INIT_HCA_FS_IB_NUM_ADDRS_OFFSET);
2019 
2020 		if (dev->caps.dmfs_high_steer_mode !=
2021 		    MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2022 			MLX4_PUT(inbox,
2023 				 ((u8)(a0_dmfs_hw_steering[dev->caps.dmfs_high_steer_mode]
2024 				       << 6)),
2025 				 INIT_HCA_FS_A0_OFFSET);
2026 	} else {
2027 		MLX4_PUT(inbox, param->mc_base,	INIT_HCA_MC_BASE_OFFSET);
2028 		MLX4_PUT(inbox, param->log_mc_entry_sz,
2029 			 INIT_HCA_LOG_MC_ENTRY_SZ_OFFSET);
2030 		MLX4_PUT(inbox, param->log_mc_hash_sz,
2031 			 INIT_HCA_LOG_MC_HASH_SZ_OFFSET);
2032 		MLX4_PUT(inbox, param->log_mc_table_sz,
2033 			 INIT_HCA_LOG_MC_TABLE_SZ_OFFSET);
2034 		if (dev->caps.steering_mode == MLX4_STEERING_MODE_B0)
2035 			MLX4_PUT(inbox, (u8) (1 << 3),
2036 				 INIT_HCA_UC_STEERING_OFFSET);
2037 	}
2038 
2039 	/* TPT attributes */
2040 
2041 	MLX4_PUT(inbox, param->dmpt_base,  INIT_HCA_DMPT_BASE_OFFSET);
2042 	MLX4_PUT(inbox, param->mw_enabled, INIT_HCA_TPT_MW_OFFSET);
2043 	MLX4_PUT(inbox, param->log_mpt_sz, INIT_HCA_LOG_MPT_SZ_OFFSET);
2044 	MLX4_PUT(inbox, param->mtt_base,   INIT_HCA_MTT_BASE_OFFSET);
2045 	MLX4_PUT(inbox, param->cmpt_base,  INIT_HCA_CMPT_BASE_OFFSET);
2046 
2047 	/* UAR attributes */
2048 
2049 	MLX4_PUT(inbox, param->uar_page_sz,	INIT_HCA_UAR_PAGE_SZ_OFFSET);
2050 	MLX4_PUT(inbox, param->log_uar_sz,      INIT_HCA_LOG_UAR_SZ_OFFSET);
2051 
2052 	/* set parser VXLAN attributes */
2053 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS) {
2054 		u8 parser_params = 0;
2055 		MLX4_PUT(inbox, parser_params,	INIT_HCA_VXLAN_OFFSET);
2056 	}
2057 
2058 	err = mlx4_cmd(dev, mailbox->dma, 0, 0, MLX4_CMD_INIT_HCA,
2059 		       MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
2060 
2061 	if (err)
2062 		mlx4_err(dev, "INIT_HCA returns %d\n", err);
2063 
2064 	mlx4_free_cmd_mailbox(dev, mailbox);
2065 	return err;
2066 }
2067 
2068 int mlx4_QUERY_HCA(struct mlx4_dev *dev,
2069 		   struct mlx4_init_hca_param *param)
2070 {
2071 	struct mlx4_cmd_mailbox *mailbox;
2072 	__be32 *outbox;
2073 	u64 qword_field;
2074 	u32 dword_field;
2075 	u16 word_field;
2076 	u8 byte_field;
2077 	int err;
2078 	static const u8 a0_dmfs_query_hw_steering[] =  {
2079 		[0] = MLX4_STEERING_DMFS_A0_DEFAULT,
2080 		[1] = MLX4_STEERING_DMFS_A0_DYNAMIC,
2081 		[2] = MLX4_STEERING_DMFS_A0_STATIC,
2082 		[3] = MLX4_STEERING_DMFS_A0_DISABLE
2083 	};
2084 
2085 #define QUERY_HCA_GLOBAL_CAPS_OFFSET	0x04
2086 #define QUERY_HCA_CORE_CLOCK_OFFSET	0x0c
2087 
2088 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2089 	if (IS_ERR(mailbox))
2090 		return PTR_ERR(mailbox);
2091 	outbox = mailbox->buf;
2092 
2093 	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0,
2094 			   MLX4_CMD_QUERY_HCA,
2095 			   MLX4_CMD_TIME_CLASS_B,
2096 			   !mlx4_is_slave(dev));
2097 	if (err)
2098 		goto out;
2099 
2100 	MLX4_GET(param->global_caps, outbox, QUERY_HCA_GLOBAL_CAPS_OFFSET);
2101 	MLX4_GET(param->hca_core_clock, outbox, QUERY_HCA_CORE_CLOCK_OFFSET);
2102 
2103 	/* QPC/EEC/CQC/EQC/RDMARC attributes */
2104 
2105 	MLX4_GET(qword_field, outbox, INIT_HCA_QPC_BASE_OFFSET);
2106 	param->qpc_base = qword_field & ~((u64)0x1f);
2107 	MLX4_GET(byte_field, outbox, INIT_HCA_LOG_QP_OFFSET);
2108 	param->log_num_qps = byte_field & 0x1f;
2109 	MLX4_GET(qword_field, outbox, INIT_HCA_SRQC_BASE_OFFSET);
2110 	param->srqc_base = qword_field & ~((u64)0x1f);
2111 	MLX4_GET(byte_field, outbox, INIT_HCA_LOG_SRQ_OFFSET);
2112 	param->log_num_srqs = byte_field & 0x1f;
2113 	MLX4_GET(qword_field, outbox, INIT_HCA_CQC_BASE_OFFSET);
2114 	param->cqc_base = qword_field & ~((u64)0x1f);
2115 	MLX4_GET(byte_field, outbox, INIT_HCA_LOG_CQ_OFFSET);
2116 	param->log_num_cqs = byte_field & 0x1f;
2117 	MLX4_GET(qword_field, outbox, INIT_HCA_ALTC_BASE_OFFSET);
2118 	param->altc_base = qword_field;
2119 	MLX4_GET(qword_field, outbox, INIT_HCA_AUXC_BASE_OFFSET);
2120 	param->auxc_base = qword_field;
2121 	MLX4_GET(qword_field, outbox, INIT_HCA_EQC_BASE_OFFSET);
2122 	param->eqc_base = qword_field & ~((u64)0x1f);
2123 	MLX4_GET(byte_field, outbox, INIT_HCA_LOG_EQ_OFFSET);
2124 	param->log_num_eqs = byte_field & 0x1f;
2125 	MLX4_GET(word_field, outbox, INIT_HCA_NUM_SYS_EQS_OFFSET);
2126 	param->num_sys_eqs = word_field & 0xfff;
2127 	MLX4_GET(qword_field, outbox, INIT_HCA_RDMARC_BASE_OFFSET);
2128 	param->rdmarc_base = qword_field & ~((u64)0x1f);
2129 	MLX4_GET(byte_field, outbox, INIT_HCA_LOG_RD_OFFSET);
2130 	param->log_rd_per_qp = byte_field & 0x7;
2131 
2132 	MLX4_GET(dword_field, outbox, INIT_HCA_FLAGS_OFFSET);
2133 	if (dword_field & (1 << INIT_HCA_DEVICE_MANAGED_FLOW_STEERING_EN)) {
2134 		param->steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
2135 	} else {
2136 		MLX4_GET(byte_field, outbox, INIT_HCA_UC_STEERING_OFFSET);
2137 		if (byte_field & 0x8)
2138 			param->steering_mode = MLX4_STEERING_MODE_B0;
2139 		else
2140 			param->steering_mode = MLX4_STEERING_MODE_A0;
2141 	}
2142 
2143 	if (dword_field & (1 << 13))
2144 		param->rss_ip_frags = 1;
2145 
2146 	/* steering attributes */
2147 	if (param->steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2148 		MLX4_GET(param->mc_base, outbox, INIT_HCA_FS_BASE_OFFSET);
2149 		MLX4_GET(byte_field, outbox, INIT_HCA_FS_LOG_ENTRY_SZ_OFFSET);
2150 		param->log_mc_entry_sz = byte_field & 0x1f;
2151 		MLX4_GET(byte_field, outbox, INIT_HCA_FS_LOG_TABLE_SZ_OFFSET);
2152 		param->log_mc_table_sz = byte_field & 0x1f;
2153 		MLX4_GET(byte_field, outbox, INIT_HCA_FS_A0_OFFSET);
2154 		param->dmfs_high_steer_mode =
2155 			a0_dmfs_query_hw_steering[(byte_field >> 6) & 3];
2156 	} else {
2157 		MLX4_GET(param->mc_base, outbox, INIT_HCA_MC_BASE_OFFSET);
2158 		MLX4_GET(byte_field, outbox, INIT_HCA_LOG_MC_ENTRY_SZ_OFFSET);
2159 		param->log_mc_entry_sz = byte_field & 0x1f;
2160 		MLX4_GET(byte_field,  outbox, INIT_HCA_LOG_MC_HASH_SZ_OFFSET);
2161 		param->log_mc_hash_sz = byte_field & 0x1f;
2162 		MLX4_GET(byte_field, outbox, INIT_HCA_LOG_MC_TABLE_SZ_OFFSET);
2163 		param->log_mc_table_sz = byte_field & 0x1f;
2164 	}
2165 
2166 	/* CX3 is capable of extending CQEs/EQEs from 32 to 64 bytes */
2167 	MLX4_GET(byte_field, outbox, INIT_HCA_EQE_CQE_OFFSETS);
2168 	if (byte_field & 0x20) /* 64-bytes eqe enabled */
2169 		param->dev_cap_enabled |= MLX4_DEV_CAP_64B_EQE_ENABLED;
2170 	if (byte_field & 0x40) /* 64-bytes cqe enabled */
2171 		param->dev_cap_enabled |= MLX4_DEV_CAP_64B_CQE_ENABLED;
2172 
2173 	/* CX3 is capable of extending CQEs\EQEs to strides larger than 64B */
2174 	MLX4_GET(byte_field, outbox, INIT_HCA_EQE_CQE_STRIDE_OFFSET);
2175 	if (byte_field) {
2176 		param->dev_cap_enabled |= MLX4_DEV_CAP_EQE_STRIDE_ENABLED;
2177 		param->dev_cap_enabled |= MLX4_DEV_CAP_CQE_STRIDE_ENABLED;
2178 		param->cqe_size = 1 << ((byte_field &
2179 					 MLX4_CQE_SIZE_MASK_STRIDE) + 5);
2180 		param->eqe_size = 1 << (((byte_field &
2181 					  MLX4_EQE_SIZE_MASK_STRIDE) >> 4) + 5);
2182 	}
2183 
2184 	/* TPT attributes */
2185 
2186 	MLX4_GET(param->dmpt_base,  outbox, INIT_HCA_DMPT_BASE_OFFSET);
2187 	MLX4_GET(byte_field, outbox, INIT_HCA_TPT_MW_OFFSET);
2188 	param->mw_enabled = byte_field >> 7;
2189 	MLX4_GET(byte_field, outbox, INIT_HCA_LOG_MPT_SZ_OFFSET);
2190 	param->log_mpt_sz = byte_field & 0x3f;
2191 	MLX4_GET(param->mtt_base,   outbox, INIT_HCA_MTT_BASE_OFFSET);
2192 	MLX4_GET(param->cmpt_base,  outbox, INIT_HCA_CMPT_BASE_OFFSET);
2193 
2194 	/* UAR attributes */
2195 
2196 	MLX4_GET(param->uar_page_sz, outbox, INIT_HCA_UAR_PAGE_SZ_OFFSET);
2197 	MLX4_GET(byte_field, outbox, INIT_HCA_LOG_UAR_SZ_OFFSET);
2198 	param->log_uar_sz = byte_field & 0xf;
2199 
2200 	/* phv_check enable */
2201 	MLX4_GET(byte_field, outbox, INIT_HCA_CACHELINE_SZ_OFFSET);
2202 	if (byte_field & 0x2)
2203 		param->phv_check_en = 1;
2204 out:
2205 	mlx4_free_cmd_mailbox(dev, mailbox);
2206 
2207 	return err;
2208 }
2209 
2210 static int mlx4_hca_core_clock_update(struct mlx4_dev *dev)
2211 {
2212 	struct mlx4_cmd_mailbox *mailbox;
2213 	__be32 *outbox;
2214 	int err;
2215 
2216 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2217 	if (IS_ERR(mailbox)) {
2218 		mlx4_warn(dev, "hca_core_clock mailbox allocation failed\n");
2219 		return PTR_ERR(mailbox);
2220 	}
2221 	outbox = mailbox->buf;
2222 
2223 	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0,
2224 			   MLX4_CMD_QUERY_HCA,
2225 			   MLX4_CMD_TIME_CLASS_B,
2226 			   !mlx4_is_slave(dev));
2227 	if (err) {
2228 		mlx4_warn(dev, "hca_core_clock update failed\n");
2229 		goto out;
2230 	}
2231 
2232 	MLX4_GET(dev->caps.hca_core_clock, outbox, QUERY_HCA_CORE_CLOCK_OFFSET);
2233 
2234 out:
2235 	mlx4_free_cmd_mailbox(dev, mailbox);
2236 
2237 	return err;
2238 }
2239 
2240 /* for IB-type ports only in SRIOV mode. Checks that both proxy QP0
2241  * and real QP0 are active, so that the paravirtualized QP0 is ready
2242  * to operate */
2243 static int check_qp0_state(struct mlx4_dev *dev, int function, int port)
2244 {
2245 	struct mlx4_priv *priv = mlx4_priv(dev);
2246 	/* irrelevant if not infiniband */
2247 	if (priv->mfunc.master.qp0_state[port].proxy_qp0_active &&
2248 	    priv->mfunc.master.qp0_state[port].qp0_active)
2249 		return 1;
2250 	return 0;
2251 }
2252 
2253 int mlx4_INIT_PORT_wrapper(struct mlx4_dev *dev, int slave,
2254 			   struct mlx4_vhcr *vhcr,
2255 			   struct mlx4_cmd_mailbox *inbox,
2256 			   struct mlx4_cmd_mailbox *outbox,
2257 			   struct mlx4_cmd_info *cmd)
2258 {
2259 	struct mlx4_priv *priv = mlx4_priv(dev);
2260 	int port = mlx4_slave_convert_port(dev, slave, vhcr->in_modifier);
2261 	int err;
2262 
2263 	if (port < 0)
2264 		return -EINVAL;
2265 
2266 	if (priv->mfunc.master.slave_state[slave].init_port_mask & (1 << port))
2267 		return 0;
2268 
2269 	if (dev->caps.port_mask[port] != MLX4_PORT_TYPE_IB) {
2270 		/* Enable port only if it was previously disabled */
2271 		if (!priv->mfunc.master.init_port_ref[port]) {
2272 			err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_INIT_PORT,
2273 				       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2274 			if (err)
2275 				return err;
2276 		}
2277 		priv->mfunc.master.slave_state[slave].init_port_mask |= (1 << port);
2278 	} else {
2279 		if (slave == mlx4_master_func_num(dev)) {
2280 			if (check_qp0_state(dev, slave, port) &&
2281 			    !priv->mfunc.master.qp0_state[port].port_active) {
2282 				err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_INIT_PORT,
2283 					       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2284 				if (err)
2285 					return err;
2286 				priv->mfunc.master.qp0_state[port].port_active = 1;
2287 				priv->mfunc.master.slave_state[slave].init_port_mask |= (1 << port);
2288 			}
2289 		} else
2290 			priv->mfunc.master.slave_state[slave].init_port_mask |= (1 << port);
2291 	}
2292 	++priv->mfunc.master.init_port_ref[port];
2293 	return 0;
2294 }
2295 
2296 int mlx4_INIT_PORT(struct mlx4_dev *dev, int port)
2297 {
2298 	struct mlx4_cmd_mailbox *mailbox;
2299 	u32 *inbox;
2300 	int err;
2301 	u32 flags;
2302 	u16 field;
2303 
2304 	if (dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
2305 #define INIT_PORT_IN_SIZE          256
2306 #define INIT_PORT_FLAGS_OFFSET     0x00
2307 #define INIT_PORT_FLAG_SIG         (1 << 18)
2308 #define INIT_PORT_FLAG_NG          (1 << 17)
2309 #define INIT_PORT_FLAG_G0          (1 << 16)
2310 #define INIT_PORT_VL_SHIFT         4
2311 #define INIT_PORT_PORT_WIDTH_SHIFT 8
2312 #define INIT_PORT_MTU_OFFSET       0x04
2313 #define INIT_PORT_MAX_GID_OFFSET   0x06
2314 #define INIT_PORT_MAX_PKEY_OFFSET  0x0a
2315 #define INIT_PORT_GUID0_OFFSET     0x10
2316 #define INIT_PORT_NODE_GUID_OFFSET 0x18
2317 #define INIT_PORT_SI_GUID_OFFSET   0x20
2318 
2319 		mailbox = mlx4_alloc_cmd_mailbox(dev);
2320 		if (IS_ERR(mailbox))
2321 			return PTR_ERR(mailbox);
2322 		inbox = mailbox->buf;
2323 
2324 		flags = 0;
2325 		flags |= (dev->caps.vl_cap[port] & 0xf) << INIT_PORT_VL_SHIFT;
2326 		flags |= (dev->caps.port_width_cap[port] & 0xf) << INIT_PORT_PORT_WIDTH_SHIFT;
2327 		MLX4_PUT(inbox, flags,		  INIT_PORT_FLAGS_OFFSET);
2328 
2329 		field = 128 << dev->caps.ib_mtu_cap[port];
2330 		MLX4_PUT(inbox, field, INIT_PORT_MTU_OFFSET);
2331 		field = dev->caps.gid_table_len[port];
2332 		MLX4_PUT(inbox, field, INIT_PORT_MAX_GID_OFFSET);
2333 		field = dev->caps.pkey_table_len[port];
2334 		MLX4_PUT(inbox, field, INIT_PORT_MAX_PKEY_OFFSET);
2335 
2336 		err = mlx4_cmd(dev, mailbox->dma, port, 0, MLX4_CMD_INIT_PORT,
2337 			       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2338 
2339 		mlx4_free_cmd_mailbox(dev, mailbox);
2340 	} else
2341 		err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_INIT_PORT,
2342 			       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2343 
2344 	if (!err)
2345 		mlx4_hca_core_clock_update(dev);
2346 
2347 	return err;
2348 }
2349 EXPORT_SYMBOL_GPL(mlx4_INIT_PORT);
2350 
2351 int mlx4_CLOSE_PORT_wrapper(struct mlx4_dev *dev, int slave,
2352 			    struct mlx4_vhcr *vhcr,
2353 			    struct mlx4_cmd_mailbox *inbox,
2354 			    struct mlx4_cmd_mailbox *outbox,
2355 			    struct mlx4_cmd_info *cmd)
2356 {
2357 	struct mlx4_priv *priv = mlx4_priv(dev);
2358 	int port = mlx4_slave_convert_port(dev, slave, vhcr->in_modifier);
2359 	int err;
2360 
2361 	if (port < 0)
2362 		return -EINVAL;
2363 
2364 	if (!(priv->mfunc.master.slave_state[slave].init_port_mask &
2365 	    (1 << port)))
2366 		return 0;
2367 
2368 	if (dev->caps.port_mask[port] != MLX4_PORT_TYPE_IB) {
2369 		if (priv->mfunc.master.init_port_ref[port] == 1) {
2370 			err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_CLOSE_PORT,
2371 				       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2372 			if (err)
2373 				return err;
2374 		}
2375 		priv->mfunc.master.slave_state[slave].init_port_mask &= ~(1 << port);
2376 	} else {
2377 		/* infiniband port */
2378 		if (slave == mlx4_master_func_num(dev)) {
2379 			if (!priv->mfunc.master.qp0_state[port].qp0_active &&
2380 			    priv->mfunc.master.qp0_state[port].port_active) {
2381 				err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_CLOSE_PORT,
2382 					       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2383 				if (err)
2384 					return err;
2385 				priv->mfunc.master.slave_state[slave].init_port_mask &= ~(1 << port);
2386 				priv->mfunc.master.qp0_state[port].port_active = 0;
2387 			}
2388 		} else
2389 			priv->mfunc.master.slave_state[slave].init_port_mask &= ~(1 << port);
2390 	}
2391 	--priv->mfunc.master.init_port_ref[port];
2392 	return 0;
2393 }
2394 
2395 int mlx4_CLOSE_PORT(struct mlx4_dev *dev, int port)
2396 {
2397 	return mlx4_cmd(dev, 0, port, 0, MLX4_CMD_CLOSE_PORT,
2398 			MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2399 }
2400 EXPORT_SYMBOL_GPL(mlx4_CLOSE_PORT);
2401 
2402 int mlx4_CLOSE_HCA(struct mlx4_dev *dev, int panic)
2403 {
2404 	return mlx4_cmd(dev, 0, 0, panic, MLX4_CMD_CLOSE_HCA,
2405 			MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
2406 }
2407 
2408 struct mlx4_config_dev {
2409 	__be32	update_flags;
2410 	__be32	rsvd1[3];
2411 	__be16	vxlan_udp_dport;
2412 	__be16	rsvd2;
2413 	__be16  roce_v2_entropy;
2414 	__be16  roce_v2_udp_dport;
2415 	__be32	roce_flags;
2416 	__be32	rsvd4[25];
2417 	__be16	rsvd5;
2418 	u8	rsvd6;
2419 	u8	rx_checksum_val;
2420 };
2421 
2422 #define MLX4_VXLAN_UDP_DPORT (1 << 0)
2423 #define MLX4_ROCE_V2_UDP_DPORT BIT(3)
2424 #define MLX4_DISABLE_RX_PORT BIT(18)
2425 
2426 static int mlx4_CONFIG_DEV_set(struct mlx4_dev *dev, struct mlx4_config_dev *config_dev)
2427 {
2428 	int err;
2429 	struct mlx4_cmd_mailbox *mailbox;
2430 
2431 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2432 	if (IS_ERR(mailbox))
2433 		return PTR_ERR(mailbox);
2434 
2435 	memcpy(mailbox->buf, config_dev, sizeof(*config_dev));
2436 
2437 	err = mlx4_cmd(dev, mailbox->dma, 0, 0, MLX4_CMD_CONFIG_DEV,
2438 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
2439 
2440 	mlx4_free_cmd_mailbox(dev, mailbox);
2441 	return err;
2442 }
2443 
2444 static int mlx4_CONFIG_DEV_get(struct mlx4_dev *dev, struct mlx4_config_dev *config_dev)
2445 {
2446 	int err;
2447 	struct mlx4_cmd_mailbox *mailbox;
2448 
2449 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2450 	if (IS_ERR(mailbox))
2451 		return PTR_ERR(mailbox);
2452 
2453 	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 1, MLX4_CMD_CONFIG_DEV,
2454 			   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2455 
2456 	if (!err)
2457 		memcpy(config_dev, mailbox->buf, sizeof(*config_dev));
2458 
2459 	mlx4_free_cmd_mailbox(dev, mailbox);
2460 	return err;
2461 }
2462 
2463 /* Conversion between the HW values and the actual functionality.
2464  * The value represented by the array index,
2465  * and the functionality determined by the flags.
2466  */
2467 static const u8 config_dev_csum_flags[] = {
2468 	[0] =	0,
2469 	[1] =	MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP,
2470 	[2] =	MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP	|
2471 		MLX4_RX_CSUM_MODE_L4,
2472 	[3] =	MLX4_RX_CSUM_MODE_L4			|
2473 		MLX4_RX_CSUM_MODE_IP_OK_IP_NON_TCP_UDP	|
2474 		MLX4_RX_CSUM_MODE_MULTI_VLAN
2475 };
2476 
2477 int mlx4_config_dev_retrieval(struct mlx4_dev *dev,
2478 			      struct mlx4_config_dev_params *params)
2479 {
2480 	struct mlx4_config_dev config_dev = {0};
2481 	int err;
2482 	u8 csum_mask;
2483 
2484 #define CONFIG_DEV_RX_CSUM_MODE_MASK			0x7
2485 #define CONFIG_DEV_RX_CSUM_MODE_PORT1_BIT_OFFSET	0
2486 #define CONFIG_DEV_RX_CSUM_MODE_PORT2_BIT_OFFSET	4
2487 
2488 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_CONFIG_DEV))
2489 		return -EOPNOTSUPP;
2490 
2491 	err = mlx4_CONFIG_DEV_get(dev, &config_dev);
2492 	if (err)
2493 		return err;
2494 
2495 	csum_mask = (config_dev.rx_checksum_val >> CONFIG_DEV_RX_CSUM_MODE_PORT1_BIT_OFFSET) &
2496 			CONFIG_DEV_RX_CSUM_MODE_MASK;
2497 
2498 	if (csum_mask >= ARRAY_SIZE(config_dev_csum_flags))
2499 		return -EINVAL;
2500 	params->rx_csum_flags_port_1 = config_dev_csum_flags[csum_mask];
2501 
2502 	csum_mask = (config_dev.rx_checksum_val >> CONFIG_DEV_RX_CSUM_MODE_PORT2_BIT_OFFSET) &
2503 			CONFIG_DEV_RX_CSUM_MODE_MASK;
2504 
2505 	if (csum_mask >= ARRAY_SIZE(config_dev_csum_flags))
2506 		return -EINVAL;
2507 	params->rx_csum_flags_port_2 = config_dev_csum_flags[csum_mask];
2508 
2509 	params->vxlan_udp_dport = be16_to_cpu(config_dev.vxlan_udp_dport);
2510 
2511 	return 0;
2512 }
2513 EXPORT_SYMBOL_GPL(mlx4_config_dev_retrieval);
2514 
2515 int mlx4_config_vxlan_port(struct mlx4_dev *dev, __be16 udp_port)
2516 {
2517 	struct mlx4_config_dev config_dev;
2518 
2519 	memset(&config_dev, 0, sizeof(config_dev));
2520 	config_dev.update_flags    = cpu_to_be32(MLX4_VXLAN_UDP_DPORT);
2521 	config_dev.vxlan_udp_dport = udp_port;
2522 
2523 	return mlx4_CONFIG_DEV_set(dev, &config_dev);
2524 }
2525 EXPORT_SYMBOL_GPL(mlx4_config_vxlan_port);
2526 
2527 #define CONFIG_DISABLE_RX_PORT BIT(15)
2528 int mlx4_disable_rx_port_check(struct mlx4_dev *dev, bool dis)
2529 {
2530 	struct mlx4_config_dev config_dev;
2531 
2532 	memset(&config_dev, 0, sizeof(config_dev));
2533 	config_dev.update_flags = cpu_to_be32(MLX4_DISABLE_RX_PORT);
2534 	if (dis)
2535 		config_dev.roce_flags =
2536 			cpu_to_be32(CONFIG_DISABLE_RX_PORT);
2537 
2538 	return mlx4_CONFIG_DEV_set(dev, &config_dev);
2539 }
2540 
2541 int mlx4_config_roce_v2_port(struct mlx4_dev *dev, u16 udp_port)
2542 {
2543 	struct mlx4_config_dev config_dev;
2544 
2545 	memset(&config_dev, 0, sizeof(config_dev));
2546 	config_dev.update_flags    = cpu_to_be32(MLX4_ROCE_V2_UDP_DPORT);
2547 	config_dev.roce_v2_udp_dport = cpu_to_be16(udp_port);
2548 
2549 	return mlx4_CONFIG_DEV_set(dev, &config_dev);
2550 }
2551 EXPORT_SYMBOL_GPL(mlx4_config_roce_v2_port);
2552 
2553 int mlx4_virt2phy_port_map(struct mlx4_dev *dev, u32 port1, u32 port2)
2554 {
2555 	struct mlx4_cmd_mailbox *mailbox;
2556 	struct {
2557 		__be32 v_port1;
2558 		__be32 v_port2;
2559 	} *v2p;
2560 	int err;
2561 
2562 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2563 	if (IS_ERR(mailbox))
2564 		return -ENOMEM;
2565 
2566 	v2p = mailbox->buf;
2567 	v2p->v_port1 = cpu_to_be32(port1);
2568 	v2p->v_port2 = cpu_to_be32(port2);
2569 
2570 	err = mlx4_cmd(dev, mailbox->dma, 0,
2571 		       MLX4_SET_PORT_VIRT2PHY, MLX4_CMD_VIRT_PORT_MAP,
2572 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
2573 
2574 	mlx4_free_cmd_mailbox(dev, mailbox);
2575 	return err;
2576 }
2577 
2578 
2579 int mlx4_SET_ICM_SIZE(struct mlx4_dev *dev, u64 icm_size, u64 *aux_pages)
2580 {
2581 	int ret = mlx4_cmd_imm(dev, icm_size, aux_pages, 0, 0,
2582 			       MLX4_CMD_SET_ICM_SIZE,
2583 			       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2584 	if (ret)
2585 		return ret;
2586 
2587 	/*
2588 	 * Round up number of system pages needed in case
2589 	 * MLX4_ICM_PAGE_SIZE < PAGE_SIZE.
2590 	 */
2591 	*aux_pages = ALIGN(*aux_pages, PAGE_SIZE / MLX4_ICM_PAGE_SIZE) >>
2592 		(PAGE_SHIFT - MLX4_ICM_PAGE_SHIFT);
2593 
2594 	return 0;
2595 }
2596 
2597 int mlx4_NOP(struct mlx4_dev *dev)
2598 {
2599 	/* Input modifier of 0x1f means "finish as soon as possible." */
2600 	return mlx4_cmd(dev, 0, 0x1f, 0, MLX4_CMD_NOP, MLX4_CMD_TIME_CLASS_A,
2601 			MLX4_CMD_NATIVE);
2602 }
2603 
2604 int mlx4_query_diag_counters(struct mlx4_dev *dev, u8 op_modifier,
2605 			     const u32 offset[],
2606 			     u32 value[], size_t array_len, u8 port)
2607 {
2608 	struct mlx4_cmd_mailbox *mailbox;
2609 	u32 *outbox;
2610 	size_t i;
2611 	int ret;
2612 
2613 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2614 	if (IS_ERR(mailbox))
2615 		return PTR_ERR(mailbox);
2616 
2617 	outbox = mailbox->buf;
2618 
2619 	ret = mlx4_cmd_box(dev, 0, mailbox->dma, port, op_modifier,
2620 			   MLX4_CMD_DIAG_RPRT, MLX4_CMD_TIME_CLASS_A,
2621 			   MLX4_CMD_NATIVE);
2622 	if (ret)
2623 		goto out;
2624 
2625 	for (i = 0; i < array_len; i++) {
2626 		if (offset[i] > MLX4_MAILBOX_SIZE) {
2627 			ret = -EINVAL;
2628 			goto out;
2629 		}
2630 
2631 		MLX4_GET(value[i], outbox, offset[i]);
2632 	}
2633 
2634 out:
2635 	mlx4_free_cmd_mailbox(dev, mailbox);
2636 	return ret;
2637 }
2638 EXPORT_SYMBOL(mlx4_query_diag_counters);
2639 
2640 int mlx4_get_phys_port_id(struct mlx4_dev *dev)
2641 {
2642 	u8 port;
2643 	u32 *outbox;
2644 	struct mlx4_cmd_mailbox *mailbox;
2645 	u32 in_mod;
2646 	u32 guid_hi, guid_lo;
2647 	int err, ret = 0;
2648 #define MOD_STAT_CFG_PORT_OFFSET 8
2649 #define MOD_STAT_CFG_GUID_H	 0X14
2650 #define MOD_STAT_CFG_GUID_L	 0X1c
2651 
2652 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2653 	if (IS_ERR(mailbox))
2654 		return PTR_ERR(mailbox);
2655 	outbox = mailbox->buf;
2656 
2657 	for (port = 1; port <= dev->caps.num_ports; port++) {
2658 		in_mod = port << MOD_STAT_CFG_PORT_OFFSET;
2659 		err = mlx4_cmd_box(dev, 0, mailbox->dma, in_mod, 0x2,
2660 				   MLX4_CMD_MOD_STAT_CFG, MLX4_CMD_TIME_CLASS_A,
2661 				   MLX4_CMD_NATIVE);
2662 		if (err) {
2663 			mlx4_err(dev, "Fail to get port %d uplink guid\n",
2664 				 port);
2665 			ret = err;
2666 		} else {
2667 			MLX4_GET(guid_hi, outbox, MOD_STAT_CFG_GUID_H);
2668 			MLX4_GET(guid_lo, outbox, MOD_STAT_CFG_GUID_L);
2669 			dev->caps.phys_port_id[port] = (u64)guid_lo |
2670 						       (u64)guid_hi << 32;
2671 		}
2672 	}
2673 	mlx4_free_cmd_mailbox(dev, mailbox);
2674 	return ret;
2675 }
2676 
2677 #define MLX4_WOL_SETUP_MODE (5 << 28)
2678 int mlx4_wol_read(struct mlx4_dev *dev, u64 *config, int port)
2679 {
2680 	u32 in_mod = MLX4_WOL_SETUP_MODE | port << 8;
2681 
2682 	return mlx4_cmd_imm(dev, 0, config, in_mod, 0x3,
2683 			    MLX4_CMD_MOD_STAT_CFG, MLX4_CMD_TIME_CLASS_A,
2684 			    MLX4_CMD_NATIVE);
2685 }
2686 EXPORT_SYMBOL_GPL(mlx4_wol_read);
2687 
2688 int mlx4_wol_write(struct mlx4_dev *dev, u64 config, int port)
2689 {
2690 	u32 in_mod = MLX4_WOL_SETUP_MODE | port << 8;
2691 
2692 	return mlx4_cmd(dev, config, in_mod, 0x1, MLX4_CMD_MOD_STAT_CFG,
2693 			MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
2694 }
2695 EXPORT_SYMBOL_GPL(mlx4_wol_write);
2696 
2697 enum {
2698 	ADD_TO_MCG = 0x26,
2699 };
2700 
2701 
2702 void mlx4_opreq_action(struct work_struct *work)
2703 {
2704 	struct mlx4_priv *priv = container_of(work, struct mlx4_priv,
2705 					      opreq_task);
2706 	struct mlx4_dev *dev = &priv->dev;
2707 	int num_tasks = atomic_read(&priv->opreq_count);
2708 	struct mlx4_cmd_mailbox *mailbox;
2709 	struct mlx4_mgm *mgm;
2710 	u32 *outbox;
2711 	u32 modifier;
2712 	u16 token;
2713 	u16 type;
2714 	int err;
2715 	u32 num_qps;
2716 	struct mlx4_qp qp;
2717 	int i;
2718 	u8 rem_mcg;
2719 	u8 prot;
2720 
2721 #define GET_OP_REQ_MODIFIER_OFFSET	0x08
2722 #define GET_OP_REQ_TOKEN_OFFSET		0x14
2723 #define GET_OP_REQ_TYPE_OFFSET		0x1a
2724 #define GET_OP_REQ_DATA_OFFSET		0x20
2725 
2726 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2727 	if (IS_ERR(mailbox)) {
2728 		mlx4_err(dev, "Failed to allocate mailbox for GET_OP_REQ\n");
2729 		return;
2730 	}
2731 	outbox = mailbox->buf;
2732 
2733 	while (num_tasks) {
2734 		err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0,
2735 				   MLX4_CMD_GET_OP_REQ, MLX4_CMD_TIME_CLASS_A,
2736 				   MLX4_CMD_NATIVE);
2737 		if (err) {
2738 			mlx4_err(dev, "Failed to retrieve required operation: %d\n",
2739 				 err);
2740 			goto out;
2741 		}
2742 		MLX4_GET(modifier, outbox, GET_OP_REQ_MODIFIER_OFFSET);
2743 		MLX4_GET(token, outbox, GET_OP_REQ_TOKEN_OFFSET);
2744 		MLX4_GET(type, outbox, GET_OP_REQ_TYPE_OFFSET);
2745 		type &= 0xfff;
2746 
2747 		switch (type) {
2748 		case ADD_TO_MCG:
2749 			if (dev->caps.steering_mode ==
2750 			    MLX4_STEERING_MODE_DEVICE_MANAGED) {
2751 				mlx4_warn(dev, "ADD MCG operation is not supported in DEVICE_MANAGED steering mode\n");
2752 				err = EPERM;
2753 				break;
2754 			}
2755 			mgm = (struct mlx4_mgm *)((u8 *)(outbox) +
2756 						  GET_OP_REQ_DATA_OFFSET);
2757 			num_qps = be32_to_cpu(mgm->members_count) &
2758 				  MGM_QPN_MASK;
2759 			rem_mcg = ((u8 *)(&mgm->members_count))[0] & 1;
2760 			prot = ((u8 *)(&mgm->members_count))[0] >> 6;
2761 
2762 			for (i = 0; i < num_qps; i++) {
2763 				qp.qpn = be32_to_cpu(mgm->qp[i]);
2764 				if (rem_mcg)
2765 					err = mlx4_multicast_detach(dev, &qp,
2766 								    mgm->gid,
2767 								    prot, 0);
2768 				else
2769 					err = mlx4_multicast_attach(dev, &qp,
2770 								    mgm->gid,
2771 								    mgm->gid[5]
2772 								    , 0, prot,
2773 								    NULL);
2774 				if (err)
2775 					break;
2776 			}
2777 			break;
2778 		default:
2779 			mlx4_warn(dev, "Bad type for required operation\n");
2780 			err = EINVAL;
2781 			break;
2782 		}
2783 		err = mlx4_cmd(dev, 0, ((u32) err |
2784 					(__force u32)cpu_to_be32(token) << 16),
2785 			       1, MLX4_CMD_GET_OP_REQ, MLX4_CMD_TIME_CLASS_A,
2786 			       MLX4_CMD_NATIVE);
2787 		if (err) {
2788 			mlx4_err(dev, "Failed to acknowledge required request: %d\n",
2789 				 err);
2790 			goto out;
2791 		}
2792 		memset(outbox, 0, 0xffc);
2793 		num_tasks = atomic_dec_return(&priv->opreq_count);
2794 	}
2795 
2796 out:
2797 	mlx4_free_cmd_mailbox(dev, mailbox);
2798 }
2799 
2800 static int mlx4_check_smp_firewall_active(struct mlx4_dev *dev,
2801 					  struct mlx4_cmd_mailbox *mailbox)
2802 {
2803 #define MLX4_CMD_MAD_DEMUX_SET_ATTR_OFFSET		0x10
2804 #define MLX4_CMD_MAD_DEMUX_GETRESP_ATTR_OFFSET		0x20
2805 #define MLX4_CMD_MAD_DEMUX_TRAP_ATTR_OFFSET		0x40
2806 #define MLX4_CMD_MAD_DEMUX_TRAP_REPRESS_ATTR_OFFSET	0x70
2807 
2808 	u32 set_attr_mask, getresp_attr_mask;
2809 	u32 trap_attr_mask, traprepress_attr_mask;
2810 
2811 	MLX4_GET(set_attr_mask, mailbox->buf,
2812 		 MLX4_CMD_MAD_DEMUX_SET_ATTR_OFFSET);
2813 	mlx4_dbg(dev, "SMP firewall set_attribute_mask = 0x%x\n",
2814 		 set_attr_mask);
2815 
2816 	MLX4_GET(getresp_attr_mask, mailbox->buf,
2817 		 MLX4_CMD_MAD_DEMUX_GETRESP_ATTR_OFFSET);
2818 	mlx4_dbg(dev, "SMP firewall getresp_attribute_mask = 0x%x\n",
2819 		 getresp_attr_mask);
2820 
2821 	MLX4_GET(trap_attr_mask, mailbox->buf,
2822 		 MLX4_CMD_MAD_DEMUX_TRAP_ATTR_OFFSET);
2823 	mlx4_dbg(dev, "SMP firewall trap_attribute_mask = 0x%x\n",
2824 		 trap_attr_mask);
2825 
2826 	MLX4_GET(traprepress_attr_mask, mailbox->buf,
2827 		 MLX4_CMD_MAD_DEMUX_TRAP_REPRESS_ATTR_OFFSET);
2828 	mlx4_dbg(dev, "SMP firewall traprepress_attribute_mask = 0x%x\n",
2829 		 traprepress_attr_mask);
2830 
2831 	if (set_attr_mask && getresp_attr_mask && trap_attr_mask &&
2832 	    traprepress_attr_mask)
2833 		return 1;
2834 
2835 	return 0;
2836 }
2837 
2838 int mlx4_config_mad_demux(struct mlx4_dev *dev)
2839 {
2840 	struct mlx4_cmd_mailbox *mailbox;
2841 	int err;
2842 
2843 	/* Check if mad_demux is supported */
2844 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_MAD_DEMUX))
2845 		return 0;
2846 
2847 	mailbox = mlx4_alloc_cmd_mailbox(dev);
2848 	if (IS_ERR(mailbox)) {
2849 		mlx4_warn(dev, "Failed to allocate mailbox for cmd MAD_DEMUX");
2850 		return -ENOMEM;
2851 	}
2852 
2853 	/* Query mad_demux to find out which MADs are handled by internal sma */
2854 	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0x01 /* subn mgmt class */,
2855 			   MLX4_CMD_MAD_DEMUX_QUERY_RESTR, MLX4_CMD_MAD_DEMUX,
2856 			   MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
2857 	if (err) {
2858 		mlx4_warn(dev, "MLX4_CMD_MAD_DEMUX: query restrictions failed (%d)\n",
2859 			  err);
2860 		goto out;
2861 	}
2862 
2863 	if (mlx4_check_smp_firewall_active(dev, mailbox))
2864 		dev->flags |= MLX4_FLAG_SECURE_HOST;
2865 
2866 	/* Config mad_demux to handle all MADs returned by the query above */
2867 	err = mlx4_cmd(dev, mailbox->dma, 0x01 /* subn mgmt class */,
2868 		       MLX4_CMD_MAD_DEMUX_CONFIG, MLX4_CMD_MAD_DEMUX,
2869 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
2870 	if (err) {
2871 		mlx4_warn(dev, "MLX4_CMD_MAD_DEMUX: configure failed (%d)\n", err);
2872 		goto out;
2873 	}
2874 
2875 	if (dev->flags & MLX4_FLAG_SECURE_HOST)
2876 		mlx4_warn(dev, "HCA operating in secure-host mode. SMP firewall activated.\n");
2877 out:
2878 	mlx4_free_cmd_mailbox(dev, mailbox);
2879 	return err;
2880 }
2881 
2882 /* Access Reg commands */
2883 enum mlx4_access_reg_masks {
2884 	MLX4_ACCESS_REG_STATUS_MASK = 0x7f,
2885 	MLX4_ACCESS_REG_METHOD_MASK = 0x7f,
2886 	MLX4_ACCESS_REG_LEN_MASK = 0x7ff
2887 };
2888 
2889 struct mlx4_access_reg {
2890 	__be16 constant1;
2891 	u8 status;
2892 	u8 resrvd1;
2893 	__be16 reg_id;
2894 	u8 method;
2895 	u8 constant2;
2896 	__be32 resrvd2[2];
2897 	__be16 len_const;
2898 	__be16 resrvd3;
2899 #define MLX4_ACCESS_REG_HEADER_SIZE (20)
2900 	u8 reg_data[MLX4_MAILBOX_SIZE-MLX4_ACCESS_REG_HEADER_SIZE];
2901 } __attribute__((__packed__));
2902 
2903 /**
2904  * mlx4_ACCESS_REG - Generic access reg command.
2905  * @dev: mlx4_dev.
2906  * @reg_id: register ID to access.
2907  * @method: Access method Read/Write.
2908  * @reg_len: register length to Read/Write in bytes.
2909  * @reg_data: reg_data pointer to Read/Write From/To.
2910  *
2911  * Access ConnectX registers FW command.
2912  * Returns 0 on success and copies outbox mlx4_access_reg data
2913  * field into reg_data or a negative error code.
2914  */
2915 static int mlx4_ACCESS_REG(struct mlx4_dev *dev, u16 reg_id,
2916 			   enum mlx4_access_reg_method method,
2917 			   u16 reg_len, void *reg_data)
2918 {
2919 	struct mlx4_cmd_mailbox *inbox, *outbox;
2920 	struct mlx4_access_reg *inbuf, *outbuf;
2921 	int err;
2922 
2923 	inbox = mlx4_alloc_cmd_mailbox(dev);
2924 	if (IS_ERR(inbox))
2925 		return PTR_ERR(inbox);
2926 
2927 	outbox = mlx4_alloc_cmd_mailbox(dev);
2928 	if (IS_ERR(outbox)) {
2929 		mlx4_free_cmd_mailbox(dev, inbox);
2930 		return PTR_ERR(outbox);
2931 	}
2932 
2933 	inbuf = inbox->buf;
2934 	outbuf = outbox->buf;
2935 
2936 	inbuf->constant1 = cpu_to_be16(0x1<<11 | 0x4);
2937 	inbuf->constant2 = 0x1;
2938 	inbuf->reg_id = cpu_to_be16(reg_id);
2939 	inbuf->method = method & MLX4_ACCESS_REG_METHOD_MASK;
2940 
2941 	reg_len = min(reg_len, (u16)(sizeof(inbuf->reg_data)));
2942 	inbuf->len_const =
2943 		cpu_to_be16(((reg_len/4 + 1) & MLX4_ACCESS_REG_LEN_MASK) |
2944 			    ((0x3) << 12));
2945 
2946 	memcpy(inbuf->reg_data, reg_data, reg_len);
2947 	err = mlx4_cmd_box(dev, inbox->dma, outbox->dma, 0, 0,
2948 			   MLX4_CMD_ACCESS_REG, MLX4_CMD_TIME_CLASS_C,
2949 			   MLX4_CMD_WRAPPED);
2950 	if (err)
2951 		goto out;
2952 
2953 	if (outbuf->status & MLX4_ACCESS_REG_STATUS_MASK) {
2954 		err = outbuf->status & MLX4_ACCESS_REG_STATUS_MASK;
2955 		mlx4_err(dev,
2956 			 "MLX4_CMD_ACCESS_REG(%x) returned REG status (%x)\n",
2957 			 reg_id, err);
2958 		goto out;
2959 	}
2960 
2961 	memcpy(reg_data, outbuf->reg_data, reg_len);
2962 out:
2963 	mlx4_free_cmd_mailbox(dev, inbox);
2964 	mlx4_free_cmd_mailbox(dev, outbox);
2965 	return err;
2966 }
2967 
2968 /* ConnectX registers IDs */
2969 enum mlx4_reg_id {
2970 	MLX4_REG_ID_PTYS = 0x5004,
2971 };
2972 
2973 /**
2974  * mlx4_ACCESS_PTYS_REG - Access PTYs (Port Type and Speed)
2975  * register
2976  * @dev: mlx4_dev.
2977  * @method: Access method Read/Write.
2978  * @ptys_reg: PTYS register data pointer.
2979  *
2980  * Access ConnectX PTYS register, to Read/Write Port Type/Speed
2981  * configuration
2982  * Returns 0 on success or a negative error code.
2983  */
2984 int mlx4_ACCESS_PTYS_REG(struct mlx4_dev *dev,
2985 			 enum mlx4_access_reg_method method,
2986 			 struct mlx4_ptys_reg *ptys_reg)
2987 {
2988 	return mlx4_ACCESS_REG(dev, MLX4_REG_ID_PTYS,
2989 			       method, sizeof(*ptys_reg), ptys_reg);
2990 }
2991 EXPORT_SYMBOL_GPL(mlx4_ACCESS_PTYS_REG);
2992 
2993 int mlx4_ACCESS_REG_wrapper(struct mlx4_dev *dev, int slave,
2994 			    struct mlx4_vhcr *vhcr,
2995 			    struct mlx4_cmd_mailbox *inbox,
2996 			    struct mlx4_cmd_mailbox *outbox,
2997 			    struct mlx4_cmd_info *cmd)
2998 {
2999 	struct mlx4_access_reg *inbuf = inbox->buf;
3000 	u8 method = inbuf->method & MLX4_ACCESS_REG_METHOD_MASK;
3001 	u16 reg_id = be16_to_cpu(inbuf->reg_id);
3002 
3003 	if (slave != mlx4_master_func_num(dev) &&
3004 	    method == MLX4_ACCESS_REG_WRITE)
3005 		return -EPERM;
3006 
3007 	if (reg_id == MLX4_REG_ID_PTYS) {
3008 		struct mlx4_ptys_reg *ptys_reg =
3009 			(struct mlx4_ptys_reg *)inbuf->reg_data;
3010 
3011 		ptys_reg->local_port =
3012 			mlx4_slave_convert_port(dev, slave,
3013 						ptys_reg->local_port);
3014 	}
3015 
3016 	return mlx4_cmd_box(dev, inbox->dma, outbox->dma, vhcr->in_modifier,
3017 			    0, MLX4_CMD_ACCESS_REG, MLX4_CMD_TIME_CLASS_C,
3018 			    MLX4_CMD_NATIVE);
3019 }
3020 
3021 static int mlx4_SET_PORT_phv_bit(struct mlx4_dev *dev, u8 port, u8 phv_bit)
3022 {
3023 #define SET_PORT_GEN_PHV_VALID	0x10
3024 #define SET_PORT_GEN_PHV_EN	0x80
3025 
3026 	struct mlx4_cmd_mailbox *mailbox;
3027 	struct mlx4_set_port_general_context *context;
3028 	u32 in_mod;
3029 	int err;
3030 
3031 	mailbox = mlx4_alloc_cmd_mailbox(dev);
3032 	if (IS_ERR(mailbox))
3033 		return PTR_ERR(mailbox);
3034 	context = mailbox->buf;
3035 
3036 	context->flags2 |=  SET_PORT_GEN_PHV_VALID;
3037 	if (phv_bit)
3038 		context->phv_en |=  SET_PORT_GEN_PHV_EN;
3039 
3040 	in_mod = MLX4_SET_PORT_GENERAL << 8 | port;
3041 	err = mlx4_cmd(dev, mailbox->dma, in_mod, MLX4_SET_PORT_ETH_OPCODE,
3042 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
3043 		       MLX4_CMD_NATIVE);
3044 
3045 	mlx4_free_cmd_mailbox(dev, mailbox);
3046 	return err;
3047 }
3048 
3049 int get_phv_bit(struct mlx4_dev *dev, u8 port, int *phv)
3050 {
3051 	int err;
3052 	struct mlx4_func_cap func_cap;
3053 
3054 	memset(&func_cap, 0, sizeof(func_cap));
3055 	err = mlx4_QUERY_FUNC_CAP(dev, port, &func_cap);
3056 	if (!err)
3057 		*phv = func_cap.flags0 & QUERY_FUNC_CAP_PHV_BIT;
3058 	return err;
3059 }
3060 EXPORT_SYMBOL(get_phv_bit);
3061 
3062 int set_phv_bit(struct mlx4_dev *dev, u8 port, int new_val)
3063 {
3064 	int ret;
3065 
3066 	if (mlx4_is_slave(dev))
3067 		return -EPERM;
3068 
3069 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
3070 	    !(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
3071 		ret = mlx4_SET_PORT_phv_bit(dev, port, new_val);
3072 		if (!ret)
3073 			dev->caps.phv_bit[port] = new_val;
3074 		return ret;
3075 	}
3076 
3077 	return -EOPNOTSUPP;
3078 }
3079 EXPORT_SYMBOL(set_phv_bit);
3080 
3081 int mlx4_get_is_vlan_offload_disabled(struct mlx4_dev *dev, u8 port,
3082 				      bool *vlan_offload_disabled)
3083 {
3084 	struct mlx4_func_cap func_cap;
3085 	int err;
3086 
3087 	memset(&func_cap, 0, sizeof(func_cap));
3088 	err = mlx4_QUERY_FUNC_CAP(dev, port, &func_cap);
3089 	if (!err)
3090 		*vlan_offload_disabled =
3091 			!!(func_cap.flags0 &
3092 			   QUERY_FUNC_CAP_VLAN_OFFLOAD_DISABLE);
3093 	return err;
3094 }
3095 EXPORT_SYMBOL(mlx4_get_is_vlan_offload_disabled);
3096 
3097 void mlx4_replace_zero_macs(struct mlx4_dev *dev)
3098 {
3099 	int i;
3100 	u8 mac_addr[ETH_ALEN];
3101 
3102 	dev->port_random_macs = 0;
3103 	for (i = 1; i <= dev->caps.num_ports; ++i)
3104 		if (!dev->caps.def_mac[i] &&
3105 		    dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH) {
3106 			eth_random_addr(mac_addr);
3107 			dev->port_random_macs |= 1 << i;
3108 			dev->caps.def_mac[i] = ether_addr_to_u64(mac_addr);
3109 		}
3110 }
3111 EXPORT_SYMBOL_GPL(mlx4_replace_zero_macs);
3112