xref: /linux/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c (revision 071bf69a0220253a44acb8b2a27f7a262b9a46bf)
1 /*
2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies.
4  * All rights reserved.
5  * Copyright (c) 2005, 2006, 2007 Cisco Systems, Inc.  All rights reserved.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * OpenIB.org BSD license below:
12  *
13  *     Redistribution and use in source and binary forms, with or
14  *     without modification, are permitted provided that the following
15  *     conditions are met:
16  *
17  *      - Redistributions of source code must retain the above
18  *        copyright notice, this list of conditions and the following
19  *        disclaimer.
20  *
21  *      - Redistributions in binary form must reproduce the above
22  *        copyright notice, this list of conditions and the following
23  *        disclaimer in the documentation and/or other materials
24  *        provided with the distribution.
25  *
26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33  * SOFTWARE.
34  */
35 
36 #include <linux/sched.h>
37 #include <linux/pci.h>
38 #include <linux/errno.h>
39 #include <linux/kernel.h>
40 #include <linux/io.h>
41 #include <linux/slab.h>
42 #include <linux/mlx4/cmd.h>
43 #include <linux/mlx4/qp.h>
44 #include <linux/if_ether.h>
45 #include <linux/etherdevice.h>
46 
47 #include "mlx4.h"
48 #include "fw.h"
49 #include "mlx4_stats.h"
50 
51 #define MLX4_MAC_VALID		(1ull << 63)
52 #define MLX4_PF_COUNTERS_PER_PORT	2
53 #define MLX4_VF_COUNTERS_PER_PORT	1
54 
55 struct mac_res {
56 	struct list_head list;
57 	u64 mac;
58 	int ref_count;
59 	u8 smac_index;
60 	u8 port;
61 };
62 
63 struct vlan_res {
64 	struct list_head list;
65 	u16 vlan;
66 	int ref_count;
67 	int vlan_index;
68 	u8 port;
69 };
70 
71 struct res_common {
72 	struct list_head	list;
73 	struct rb_node		node;
74 	u64		        res_id;
75 	int			owner;
76 	int			state;
77 	int			from_state;
78 	int			to_state;
79 	int			removing;
80 };
81 
82 enum {
83 	RES_ANY_BUSY = 1
84 };
85 
86 struct res_gid {
87 	struct list_head	list;
88 	u8			gid[16];
89 	enum mlx4_protocol	prot;
90 	enum mlx4_steer_type	steer;
91 	u64			reg_id;
92 };
93 
94 enum res_qp_states {
95 	RES_QP_BUSY = RES_ANY_BUSY,
96 
97 	/* QP number was allocated */
98 	RES_QP_RESERVED,
99 
100 	/* ICM memory for QP context was mapped */
101 	RES_QP_MAPPED,
102 
103 	/* QP is in hw ownership */
104 	RES_QP_HW
105 };
106 
107 struct res_qp {
108 	struct res_common	com;
109 	struct res_mtt	       *mtt;
110 	struct res_cq	       *rcq;
111 	struct res_cq	       *scq;
112 	struct res_srq	       *srq;
113 	struct list_head	mcg_list;
114 	spinlock_t		mcg_spl;
115 	int			local_qpn;
116 	atomic_t		ref_count;
117 	u32			qpc_flags;
118 	/* saved qp params before VST enforcement in order to restore on VGT */
119 	u8			sched_queue;
120 	__be32			param3;
121 	u8			vlan_control;
122 	u8			fvl_rx;
123 	u8			pri_path_fl;
124 	u8			vlan_index;
125 	u8			feup;
126 };
127 
128 enum res_mtt_states {
129 	RES_MTT_BUSY = RES_ANY_BUSY,
130 	RES_MTT_ALLOCATED,
131 };
132 
133 static inline const char *mtt_states_str(enum res_mtt_states state)
134 {
135 	switch (state) {
136 	case RES_MTT_BUSY: return "RES_MTT_BUSY";
137 	case RES_MTT_ALLOCATED: return "RES_MTT_ALLOCATED";
138 	default: return "Unknown";
139 	}
140 }
141 
142 struct res_mtt {
143 	struct res_common	com;
144 	int			order;
145 	atomic_t		ref_count;
146 };
147 
148 enum res_mpt_states {
149 	RES_MPT_BUSY = RES_ANY_BUSY,
150 	RES_MPT_RESERVED,
151 	RES_MPT_MAPPED,
152 	RES_MPT_HW,
153 };
154 
155 struct res_mpt {
156 	struct res_common	com;
157 	struct res_mtt	       *mtt;
158 	int			key;
159 };
160 
161 enum res_eq_states {
162 	RES_EQ_BUSY = RES_ANY_BUSY,
163 	RES_EQ_RESERVED,
164 	RES_EQ_HW,
165 };
166 
167 struct res_eq {
168 	struct res_common	com;
169 	struct res_mtt	       *mtt;
170 };
171 
172 enum res_cq_states {
173 	RES_CQ_BUSY = RES_ANY_BUSY,
174 	RES_CQ_ALLOCATED,
175 	RES_CQ_HW,
176 };
177 
178 struct res_cq {
179 	struct res_common	com;
180 	struct res_mtt	       *mtt;
181 	atomic_t		ref_count;
182 };
183 
184 enum res_srq_states {
185 	RES_SRQ_BUSY = RES_ANY_BUSY,
186 	RES_SRQ_ALLOCATED,
187 	RES_SRQ_HW,
188 };
189 
190 struct res_srq {
191 	struct res_common	com;
192 	struct res_mtt	       *mtt;
193 	struct res_cq	       *cq;
194 	atomic_t		ref_count;
195 };
196 
197 enum res_counter_states {
198 	RES_COUNTER_BUSY = RES_ANY_BUSY,
199 	RES_COUNTER_ALLOCATED,
200 };
201 
202 struct res_counter {
203 	struct res_common	com;
204 	int			port;
205 };
206 
207 enum res_xrcdn_states {
208 	RES_XRCD_BUSY = RES_ANY_BUSY,
209 	RES_XRCD_ALLOCATED,
210 };
211 
212 struct res_xrcdn {
213 	struct res_common	com;
214 	int			port;
215 };
216 
217 enum res_fs_rule_states {
218 	RES_FS_RULE_BUSY = RES_ANY_BUSY,
219 	RES_FS_RULE_ALLOCATED,
220 };
221 
222 struct res_fs_rule {
223 	struct res_common	com;
224 	int			qpn;
225 	/* VF DMFS mbox with port flipped */
226 	void			*mirr_mbox;
227 	/* > 0 --> apply mirror when getting into HA mode      */
228 	/* = 0 --> un-apply mirror when getting out of HA mode */
229 	u32			mirr_mbox_size;
230 	struct list_head	mirr_list;
231 	u64			mirr_rule_id;
232 };
233 
234 static void *res_tracker_lookup(struct rb_root *root, u64 res_id)
235 {
236 	struct rb_node *node = root->rb_node;
237 
238 	while (node) {
239 		struct res_common *res = container_of(node, struct res_common,
240 						      node);
241 
242 		if (res_id < res->res_id)
243 			node = node->rb_left;
244 		else if (res_id > res->res_id)
245 			node = node->rb_right;
246 		else
247 			return res;
248 	}
249 	return NULL;
250 }
251 
252 static int res_tracker_insert(struct rb_root *root, struct res_common *res)
253 {
254 	struct rb_node **new = &(root->rb_node), *parent = NULL;
255 
256 	/* Figure out where to put new node */
257 	while (*new) {
258 		struct res_common *this = container_of(*new, struct res_common,
259 						       node);
260 
261 		parent = *new;
262 		if (res->res_id < this->res_id)
263 			new = &((*new)->rb_left);
264 		else if (res->res_id > this->res_id)
265 			new = &((*new)->rb_right);
266 		else
267 			return -EEXIST;
268 	}
269 
270 	/* Add new node and rebalance tree. */
271 	rb_link_node(&res->node, parent, new);
272 	rb_insert_color(&res->node, root);
273 
274 	return 0;
275 }
276 
277 enum qp_transition {
278 	QP_TRANS_INIT2RTR,
279 	QP_TRANS_RTR2RTS,
280 	QP_TRANS_RTS2RTS,
281 	QP_TRANS_SQERR2RTS,
282 	QP_TRANS_SQD2SQD,
283 	QP_TRANS_SQD2RTS
284 };
285 
286 /* For Debug uses */
287 static const char *resource_str(enum mlx4_resource rt)
288 {
289 	switch (rt) {
290 	case RES_QP: return "RES_QP";
291 	case RES_CQ: return "RES_CQ";
292 	case RES_SRQ: return "RES_SRQ";
293 	case RES_MPT: return "RES_MPT";
294 	case RES_MTT: return "RES_MTT";
295 	case RES_MAC: return  "RES_MAC";
296 	case RES_VLAN: return  "RES_VLAN";
297 	case RES_EQ: return "RES_EQ";
298 	case RES_COUNTER: return "RES_COUNTER";
299 	case RES_FS_RULE: return "RES_FS_RULE";
300 	case RES_XRCD: return "RES_XRCD";
301 	default: return "Unknown resource type !!!";
302 	};
303 }
304 
305 static void rem_slave_vlans(struct mlx4_dev *dev, int slave);
306 static inline int mlx4_grant_resource(struct mlx4_dev *dev, int slave,
307 				      enum mlx4_resource res_type, int count,
308 				      int port)
309 {
310 	struct mlx4_priv *priv = mlx4_priv(dev);
311 	struct resource_allocator *res_alloc =
312 		&priv->mfunc.master.res_tracker.res_alloc[res_type];
313 	int err = -EINVAL;
314 	int allocated, free, reserved, guaranteed, from_free;
315 	int from_rsvd;
316 
317 	if (slave > dev->persist->num_vfs)
318 		return -EINVAL;
319 
320 	spin_lock(&res_alloc->alloc_lock);
321 	allocated = (port > 0) ?
322 		res_alloc->allocated[(port - 1) *
323 		(dev->persist->num_vfs + 1) + slave] :
324 		res_alloc->allocated[slave];
325 	free = (port > 0) ? res_alloc->res_port_free[port - 1] :
326 		res_alloc->res_free;
327 	reserved = (port > 0) ? res_alloc->res_port_rsvd[port - 1] :
328 		res_alloc->res_reserved;
329 	guaranteed = res_alloc->guaranteed[slave];
330 
331 	if (allocated + count > res_alloc->quota[slave]) {
332 		mlx4_warn(dev, "VF %d port %d res %s: quota exceeded, count %d alloc %d quota %d\n",
333 			  slave, port, resource_str(res_type), count,
334 			  allocated, res_alloc->quota[slave]);
335 		goto out;
336 	}
337 
338 	if (allocated + count <= guaranteed) {
339 		err = 0;
340 		from_rsvd = count;
341 	} else {
342 		/* portion may need to be obtained from free area */
343 		if (guaranteed - allocated > 0)
344 			from_free = count - (guaranteed - allocated);
345 		else
346 			from_free = count;
347 
348 		from_rsvd = count - from_free;
349 
350 		if (free - from_free >= reserved)
351 			err = 0;
352 		else
353 			mlx4_warn(dev, "VF %d port %d res %s: free pool empty, free %d from_free %d rsvd %d\n",
354 				  slave, port, resource_str(res_type), free,
355 				  from_free, reserved);
356 	}
357 
358 	if (!err) {
359 		/* grant the request */
360 		if (port > 0) {
361 			res_alloc->allocated[(port - 1) *
362 			(dev->persist->num_vfs + 1) + slave] += count;
363 			res_alloc->res_port_free[port - 1] -= count;
364 			res_alloc->res_port_rsvd[port - 1] -= from_rsvd;
365 		} else {
366 			res_alloc->allocated[slave] += count;
367 			res_alloc->res_free -= count;
368 			res_alloc->res_reserved -= from_rsvd;
369 		}
370 	}
371 
372 out:
373 	spin_unlock(&res_alloc->alloc_lock);
374 	return err;
375 }
376 
377 static inline void mlx4_release_resource(struct mlx4_dev *dev, int slave,
378 				    enum mlx4_resource res_type, int count,
379 				    int port)
380 {
381 	struct mlx4_priv *priv = mlx4_priv(dev);
382 	struct resource_allocator *res_alloc =
383 		&priv->mfunc.master.res_tracker.res_alloc[res_type];
384 	int allocated, guaranteed, from_rsvd;
385 
386 	if (slave > dev->persist->num_vfs)
387 		return;
388 
389 	spin_lock(&res_alloc->alloc_lock);
390 
391 	allocated = (port > 0) ?
392 		res_alloc->allocated[(port - 1) *
393 		(dev->persist->num_vfs + 1) + slave] :
394 		res_alloc->allocated[slave];
395 	guaranteed = res_alloc->guaranteed[slave];
396 
397 	if (allocated - count >= guaranteed) {
398 		from_rsvd = 0;
399 	} else {
400 		/* portion may need to be returned to reserved area */
401 		if (allocated - guaranteed > 0)
402 			from_rsvd = count - (allocated - guaranteed);
403 		else
404 			from_rsvd = count;
405 	}
406 
407 	if (port > 0) {
408 		res_alloc->allocated[(port - 1) *
409 		(dev->persist->num_vfs + 1) + slave] -= count;
410 		res_alloc->res_port_free[port - 1] += count;
411 		res_alloc->res_port_rsvd[port - 1] += from_rsvd;
412 	} else {
413 		res_alloc->allocated[slave] -= count;
414 		res_alloc->res_free += count;
415 		res_alloc->res_reserved += from_rsvd;
416 	}
417 
418 	spin_unlock(&res_alloc->alloc_lock);
419 	return;
420 }
421 
422 static inline void initialize_res_quotas(struct mlx4_dev *dev,
423 					 struct resource_allocator *res_alloc,
424 					 enum mlx4_resource res_type,
425 					 int vf, int num_instances)
426 {
427 	res_alloc->guaranteed[vf] = num_instances /
428 				    (2 * (dev->persist->num_vfs + 1));
429 	res_alloc->quota[vf] = (num_instances / 2) + res_alloc->guaranteed[vf];
430 	if (vf == mlx4_master_func_num(dev)) {
431 		res_alloc->res_free = num_instances;
432 		if (res_type == RES_MTT) {
433 			/* reserved mtts will be taken out of the PF allocation */
434 			res_alloc->res_free += dev->caps.reserved_mtts;
435 			res_alloc->guaranteed[vf] += dev->caps.reserved_mtts;
436 			res_alloc->quota[vf] += dev->caps.reserved_mtts;
437 		}
438 	}
439 }
440 
441 void mlx4_init_quotas(struct mlx4_dev *dev)
442 {
443 	struct mlx4_priv *priv = mlx4_priv(dev);
444 	int pf;
445 
446 	/* quotas for VFs are initialized in mlx4_slave_cap */
447 	if (mlx4_is_slave(dev))
448 		return;
449 
450 	if (!mlx4_is_mfunc(dev)) {
451 		dev->quotas.qp = dev->caps.num_qps - dev->caps.reserved_qps -
452 			mlx4_num_reserved_sqps(dev);
453 		dev->quotas.cq = dev->caps.num_cqs - dev->caps.reserved_cqs;
454 		dev->quotas.srq = dev->caps.num_srqs - dev->caps.reserved_srqs;
455 		dev->quotas.mtt = dev->caps.num_mtts - dev->caps.reserved_mtts;
456 		dev->quotas.mpt = dev->caps.num_mpts - dev->caps.reserved_mrws;
457 		return;
458 	}
459 
460 	pf = mlx4_master_func_num(dev);
461 	dev->quotas.qp =
462 		priv->mfunc.master.res_tracker.res_alloc[RES_QP].quota[pf];
463 	dev->quotas.cq =
464 		priv->mfunc.master.res_tracker.res_alloc[RES_CQ].quota[pf];
465 	dev->quotas.srq =
466 		priv->mfunc.master.res_tracker.res_alloc[RES_SRQ].quota[pf];
467 	dev->quotas.mtt =
468 		priv->mfunc.master.res_tracker.res_alloc[RES_MTT].quota[pf];
469 	dev->quotas.mpt =
470 		priv->mfunc.master.res_tracker.res_alloc[RES_MPT].quota[pf];
471 }
472 
473 static int get_max_gauranteed_vfs_counter(struct mlx4_dev *dev)
474 {
475 	/* reduce the sink counter */
476 	return (dev->caps.max_counters - 1 -
477 		(MLX4_PF_COUNTERS_PER_PORT * MLX4_MAX_PORTS))
478 		/ MLX4_MAX_PORTS;
479 }
480 
481 int mlx4_init_resource_tracker(struct mlx4_dev *dev)
482 {
483 	struct mlx4_priv *priv = mlx4_priv(dev);
484 	int i, j;
485 	int t;
486 	int max_vfs_guarantee_counter = get_max_gauranteed_vfs_counter(dev);
487 
488 	priv->mfunc.master.res_tracker.slave_list =
489 		kzalloc(dev->num_slaves * sizeof(struct slave_list),
490 			GFP_KERNEL);
491 	if (!priv->mfunc.master.res_tracker.slave_list)
492 		return -ENOMEM;
493 
494 	for (i = 0 ; i < dev->num_slaves; i++) {
495 		for (t = 0; t < MLX4_NUM_OF_RESOURCE_TYPE; ++t)
496 			INIT_LIST_HEAD(&priv->mfunc.master.res_tracker.
497 				       slave_list[i].res_list[t]);
498 		mutex_init(&priv->mfunc.master.res_tracker.slave_list[i].mutex);
499 	}
500 
501 	mlx4_dbg(dev, "Started init_resource_tracker: %ld slaves\n",
502 		 dev->num_slaves);
503 	for (i = 0 ; i < MLX4_NUM_OF_RESOURCE_TYPE; i++)
504 		priv->mfunc.master.res_tracker.res_tree[i] = RB_ROOT;
505 
506 	for (i = 0; i < MLX4_NUM_OF_RESOURCE_TYPE; i++) {
507 		struct resource_allocator *res_alloc =
508 			&priv->mfunc.master.res_tracker.res_alloc[i];
509 		res_alloc->quota = kmalloc((dev->persist->num_vfs + 1) *
510 					   sizeof(int), GFP_KERNEL);
511 		res_alloc->guaranteed = kmalloc((dev->persist->num_vfs + 1) *
512 						sizeof(int), GFP_KERNEL);
513 		if (i == RES_MAC || i == RES_VLAN)
514 			res_alloc->allocated = kzalloc(MLX4_MAX_PORTS *
515 						       (dev->persist->num_vfs
516 						       + 1) *
517 						       sizeof(int), GFP_KERNEL);
518 		else
519 			res_alloc->allocated = kzalloc((dev->persist->
520 							num_vfs + 1) *
521 						       sizeof(int), GFP_KERNEL);
522 		/* Reduce the sink counter */
523 		if (i == RES_COUNTER)
524 			res_alloc->res_free = dev->caps.max_counters - 1;
525 
526 		if (!res_alloc->quota || !res_alloc->guaranteed ||
527 		    !res_alloc->allocated)
528 			goto no_mem_err;
529 
530 		spin_lock_init(&res_alloc->alloc_lock);
531 		for (t = 0; t < dev->persist->num_vfs + 1; t++) {
532 			struct mlx4_active_ports actv_ports =
533 				mlx4_get_active_ports(dev, t);
534 			switch (i) {
535 			case RES_QP:
536 				initialize_res_quotas(dev, res_alloc, RES_QP,
537 						      t, dev->caps.num_qps -
538 						      dev->caps.reserved_qps -
539 						      mlx4_num_reserved_sqps(dev));
540 				break;
541 			case RES_CQ:
542 				initialize_res_quotas(dev, res_alloc, RES_CQ,
543 						      t, dev->caps.num_cqs -
544 						      dev->caps.reserved_cqs);
545 				break;
546 			case RES_SRQ:
547 				initialize_res_quotas(dev, res_alloc, RES_SRQ,
548 						      t, dev->caps.num_srqs -
549 						      dev->caps.reserved_srqs);
550 				break;
551 			case RES_MPT:
552 				initialize_res_quotas(dev, res_alloc, RES_MPT,
553 						      t, dev->caps.num_mpts -
554 						      dev->caps.reserved_mrws);
555 				break;
556 			case RES_MTT:
557 				initialize_res_quotas(dev, res_alloc, RES_MTT,
558 						      t, dev->caps.num_mtts -
559 						      dev->caps.reserved_mtts);
560 				break;
561 			case RES_MAC:
562 				if (t == mlx4_master_func_num(dev)) {
563 					int max_vfs_pport = 0;
564 					/* Calculate the max vfs per port for */
565 					/* both ports.			      */
566 					for (j = 0; j < dev->caps.num_ports;
567 					     j++) {
568 						struct mlx4_slaves_pport slaves_pport =
569 							mlx4_phys_to_slaves_pport(dev, j + 1);
570 						unsigned current_slaves =
571 							bitmap_weight(slaves_pport.slaves,
572 								      dev->caps.num_ports) - 1;
573 						if (max_vfs_pport < current_slaves)
574 							max_vfs_pport =
575 								current_slaves;
576 					}
577 					res_alloc->quota[t] =
578 						MLX4_MAX_MAC_NUM -
579 						2 * max_vfs_pport;
580 					res_alloc->guaranteed[t] = 2;
581 					for (j = 0; j < MLX4_MAX_PORTS; j++)
582 						res_alloc->res_port_free[j] =
583 							MLX4_MAX_MAC_NUM;
584 				} else {
585 					res_alloc->quota[t] = MLX4_MAX_MAC_NUM;
586 					res_alloc->guaranteed[t] = 2;
587 				}
588 				break;
589 			case RES_VLAN:
590 				if (t == mlx4_master_func_num(dev)) {
591 					res_alloc->quota[t] = MLX4_MAX_VLAN_NUM;
592 					res_alloc->guaranteed[t] = MLX4_MAX_VLAN_NUM / 2;
593 					for (j = 0; j < MLX4_MAX_PORTS; j++)
594 						res_alloc->res_port_free[j] =
595 							res_alloc->quota[t];
596 				} else {
597 					res_alloc->quota[t] = MLX4_MAX_VLAN_NUM / 2;
598 					res_alloc->guaranteed[t] = 0;
599 				}
600 				break;
601 			case RES_COUNTER:
602 				res_alloc->quota[t] = dev->caps.max_counters;
603 				if (t == mlx4_master_func_num(dev))
604 					res_alloc->guaranteed[t] =
605 						MLX4_PF_COUNTERS_PER_PORT *
606 						MLX4_MAX_PORTS;
607 				else if (t <= max_vfs_guarantee_counter)
608 					res_alloc->guaranteed[t] =
609 						MLX4_VF_COUNTERS_PER_PORT *
610 						MLX4_MAX_PORTS;
611 				else
612 					res_alloc->guaranteed[t] = 0;
613 				res_alloc->res_free -= res_alloc->guaranteed[t];
614 				break;
615 			default:
616 				break;
617 			}
618 			if (i == RES_MAC || i == RES_VLAN) {
619 				for (j = 0; j < dev->caps.num_ports; j++)
620 					if (test_bit(j, actv_ports.ports))
621 						res_alloc->res_port_rsvd[j] +=
622 							res_alloc->guaranteed[t];
623 			} else {
624 				res_alloc->res_reserved += res_alloc->guaranteed[t];
625 			}
626 		}
627 	}
628 	spin_lock_init(&priv->mfunc.master.res_tracker.lock);
629 	return 0;
630 
631 no_mem_err:
632 	for (i = 0; i < MLX4_NUM_OF_RESOURCE_TYPE; i++) {
633 		kfree(priv->mfunc.master.res_tracker.res_alloc[i].allocated);
634 		priv->mfunc.master.res_tracker.res_alloc[i].allocated = NULL;
635 		kfree(priv->mfunc.master.res_tracker.res_alloc[i].guaranteed);
636 		priv->mfunc.master.res_tracker.res_alloc[i].guaranteed = NULL;
637 		kfree(priv->mfunc.master.res_tracker.res_alloc[i].quota);
638 		priv->mfunc.master.res_tracker.res_alloc[i].quota = NULL;
639 	}
640 	return -ENOMEM;
641 }
642 
643 void mlx4_free_resource_tracker(struct mlx4_dev *dev,
644 				enum mlx4_res_tracker_free_type type)
645 {
646 	struct mlx4_priv *priv = mlx4_priv(dev);
647 	int i;
648 
649 	if (priv->mfunc.master.res_tracker.slave_list) {
650 		if (type != RES_TR_FREE_STRUCTS_ONLY) {
651 			for (i = 0; i < dev->num_slaves; i++) {
652 				if (type == RES_TR_FREE_ALL ||
653 				    dev->caps.function != i)
654 					mlx4_delete_all_resources_for_slave(dev, i);
655 			}
656 			/* free master's vlans */
657 			i = dev->caps.function;
658 			mlx4_reset_roce_gids(dev, i);
659 			mutex_lock(&priv->mfunc.master.res_tracker.slave_list[i].mutex);
660 			rem_slave_vlans(dev, i);
661 			mutex_unlock(&priv->mfunc.master.res_tracker.slave_list[i].mutex);
662 		}
663 
664 		if (type != RES_TR_FREE_SLAVES_ONLY) {
665 			for (i = 0; i < MLX4_NUM_OF_RESOURCE_TYPE; i++) {
666 				kfree(priv->mfunc.master.res_tracker.res_alloc[i].allocated);
667 				priv->mfunc.master.res_tracker.res_alloc[i].allocated = NULL;
668 				kfree(priv->mfunc.master.res_tracker.res_alloc[i].guaranteed);
669 				priv->mfunc.master.res_tracker.res_alloc[i].guaranteed = NULL;
670 				kfree(priv->mfunc.master.res_tracker.res_alloc[i].quota);
671 				priv->mfunc.master.res_tracker.res_alloc[i].quota = NULL;
672 			}
673 			kfree(priv->mfunc.master.res_tracker.slave_list);
674 			priv->mfunc.master.res_tracker.slave_list = NULL;
675 		}
676 	}
677 }
678 
679 static void update_pkey_index(struct mlx4_dev *dev, int slave,
680 			      struct mlx4_cmd_mailbox *inbox)
681 {
682 	u8 sched = *(u8 *)(inbox->buf + 64);
683 	u8 orig_index = *(u8 *)(inbox->buf + 35);
684 	u8 new_index;
685 	struct mlx4_priv *priv = mlx4_priv(dev);
686 	int port;
687 
688 	port = (sched >> 6 & 1) + 1;
689 
690 	new_index = priv->virt2phys_pkey[slave][port - 1][orig_index];
691 	*(u8 *)(inbox->buf + 35) = new_index;
692 }
693 
694 static void update_gid(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *inbox,
695 		       u8 slave)
696 {
697 	struct mlx4_qp_context	*qp_ctx = inbox->buf + 8;
698 	enum mlx4_qp_optpar	optpar = be32_to_cpu(*(__be32 *) inbox->buf);
699 	u32			ts = (be32_to_cpu(qp_ctx->flags) >> 16) & 0xff;
700 	int port;
701 
702 	if (MLX4_QP_ST_UD == ts) {
703 		port = (qp_ctx->pri_path.sched_queue >> 6 & 1) + 1;
704 		if (mlx4_is_eth(dev, port))
705 			qp_ctx->pri_path.mgid_index =
706 				mlx4_get_base_gid_ix(dev, slave, port) | 0x80;
707 		else
708 			qp_ctx->pri_path.mgid_index = slave | 0x80;
709 
710 	} else if (MLX4_QP_ST_RC == ts || MLX4_QP_ST_XRC == ts || MLX4_QP_ST_UC == ts) {
711 		if (optpar & MLX4_QP_OPTPAR_PRIMARY_ADDR_PATH) {
712 			port = (qp_ctx->pri_path.sched_queue >> 6 & 1) + 1;
713 			if (mlx4_is_eth(dev, port)) {
714 				qp_ctx->pri_path.mgid_index +=
715 					mlx4_get_base_gid_ix(dev, slave, port);
716 				qp_ctx->pri_path.mgid_index &= 0x7f;
717 			} else {
718 				qp_ctx->pri_path.mgid_index = slave & 0x7F;
719 			}
720 		}
721 		if (optpar & MLX4_QP_OPTPAR_ALT_ADDR_PATH) {
722 			port = (qp_ctx->alt_path.sched_queue >> 6 & 1) + 1;
723 			if (mlx4_is_eth(dev, port)) {
724 				qp_ctx->alt_path.mgid_index +=
725 					mlx4_get_base_gid_ix(dev, slave, port);
726 				qp_ctx->alt_path.mgid_index &= 0x7f;
727 			} else {
728 				qp_ctx->alt_path.mgid_index = slave & 0x7F;
729 			}
730 		}
731 	}
732 }
733 
734 static int handle_counter(struct mlx4_dev *dev, struct mlx4_qp_context *qpc,
735 			  u8 slave, int port);
736 
737 static int update_vport_qp_param(struct mlx4_dev *dev,
738 				 struct mlx4_cmd_mailbox *inbox,
739 				 u8 slave, u32 qpn)
740 {
741 	struct mlx4_qp_context	*qpc = inbox->buf + 8;
742 	struct mlx4_vport_oper_state *vp_oper;
743 	struct mlx4_priv *priv;
744 	u32 qp_type;
745 	int port, err = 0;
746 
747 	port = (qpc->pri_path.sched_queue & 0x40) ? 2 : 1;
748 	priv = mlx4_priv(dev);
749 	vp_oper = &priv->mfunc.master.vf_oper[slave].vport[port];
750 	qp_type	= (be32_to_cpu(qpc->flags) >> 16) & 0xff;
751 
752 	err = handle_counter(dev, qpc, slave, port);
753 	if (err)
754 		goto out;
755 
756 	if (MLX4_VGT != vp_oper->state.default_vlan) {
757 		/* the reserved QPs (special, proxy, tunnel)
758 		 * do not operate over vlans
759 		 */
760 		if (mlx4_is_qp_reserved(dev, qpn))
761 			return 0;
762 
763 		/* force strip vlan by clear vsd, MLX QP refers to Raw Ethernet */
764 		if (qp_type == MLX4_QP_ST_UD ||
765 		    (qp_type == MLX4_QP_ST_MLX && mlx4_is_eth(dev, port))) {
766 			if (dev->caps.bmme_flags & MLX4_BMME_FLAG_VSD_INIT2RTR) {
767 				*(__be32 *)inbox->buf =
768 					cpu_to_be32(be32_to_cpu(*(__be32 *)inbox->buf) |
769 					MLX4_QP_OPTPAR_VLAN_STRIPPING);
770 				qpc->param3 &= ~cpu_to_be32(MLX4_STRIP_VLAN);
771 			} else {
772 				struct mlx4_update_qp_params params = {.flags = 0};
773 
774 				err = mlx4_update_qp(dev, qpn, MLX4_UPDATE_QP_VSD, &params);
775 				if (err)
776 					goto out;
777 			}
778 		}
779 
780 		/* preserve IF_COUNTER flag */
781 		qpc->pri_path.vlan_control &=
782 			MLX4_CTRL_ETH_SRC_CHECK_IF_COUNTER;
783 		if (vp_oper->state.link_state == IFLA_VF_LINK_STATE_DISABLE &&
784 		    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_UPDATE_QP) {
785 			qpc->pri_path.vlan_control |=
786 				MLX4_VLAN_CTRL_ETH_TX_BLOCK_TAGGED |
787 				MLX4_VLAN_CTRL_ETH_TX_BLOCK_PRIO_TAGGED |
788 				MLX4_VLAN_CTRL_ETH_TX_BLOCK_UNTAGGED |
789 				MLX4_VLAN_CTRL_ETH_RX_BLOCK_PRIO_TAGGED |
790 				MLX4_VLAN_CTRL_ETH_RX_BLOCK_UNTAGGED |
791 				MLX4_VLAN_CTRL_ETH_RX_BLOCK_TAGGED;
792 		} else if (0 != vp_oper->state.default_vlan) {
793 			qpc->pri_path.vlan_control |=
794 				MLX4_VLAN_CTRL_ETH_TX_BLOCK_TAGGED |
795 				MLX4_VLAN_CTRL_ETH_RX_BLOCK_PRIO_TAGGED |
796 				MLX4_VLAN_CTRL_ETH_RX_BLOCK_UNTAGGED;
797 		} else { /* priority tagged */
798 			qpc->pri_path.vlan_control |=
799 				MLX4_VLAN_CTRL_ETH_TX_BLOCK_TAGGED |
800 				MLX4_VLAN_CTRL_ETH_RX_BLOCK_TAGGED;
801 		}
802 
803 		qpc->pri_path.fvl_rx |= MLX4_FVL_RX_FORCE_ETH_VLAN;
804 		qpc->pri_path.vlan_index = vp_oper->vlan_idx;
805 		qpc->pri_path.fl |= MLX4_FL_CV | MLX4_FL_ETH_HIDE_CQE_VLAN;
806 		qpc->pri_path.feup |= MLX4_FEUP_FORCE_ETH_UP | MLX4_FVL_FORCE_ETH_VLAN;
807 		qpc->pri_path.sched_queue &= 0xC7;
808 		qpc->pri_path.sched_queue |= (vp_oper->state.default_qos) << 3;
809 		qpc->qos_vport = vp_oper->state.qos_vport;
810 	}
811 	if (vp_oper->state.spoofchk) {
812 		qpc->pri_path.feup |= MLX4_FSM_FORCE_ETH_SRC_MAC;
813 		qpc->pri_path.grh_mylmc = (0x80 & qpc->pri_path.grh_mylmc) + vp_oper->mac_idx;
814 	}
815 out:
816 	return err;
817 }
818 
819 static int mpt_mask(struct mlx4_dev *dev)
820 {
821 	return dev->caps.num_mpts - 1;
822 }
823 
824 static void *find_res(struct mlx4_dev *dev, u64 res_id,
825 		      enum mlx4_resource type)
826 {
827 	struct mlx4_priv *priv = mlx4_priv(dev);
828 
829 	return res_tracker_lookup(&priv->mfunc.master.res_tracker.res_tree[type],
830 				  res_id);
831 }
832 
833 static int get_res(struct mlx4_dev *dev, int slave, u64 res_id,
834 		   enum mlx4_resource type,
835 		   void *res)
836 {
837 	struct res_common *r;
838 	int err = 0;
839 
840 	spin_lock_irq(mlx4_tlock(dev));
841 	r = find_res(dev, res_id, type);
842 	if (!r) {
843 		err = -ENONET;
844 		goto exit;
845 	}
846 
847 	if (r->state == RES_ANY_BUSY) {
848 		err = -EBUSY;
849 		goto exit;
850 	}
851 
852 	if (r->owner != slave) {
853 		err = -EPERM;
854 		goto exit;
855 	}
856 
857 	r->from_state = r->state;
858 	r->state = RES_ANY_BUSY;
859 
860 	if (res)
861 		*((struct res_common **)res) = r;
862 
863 exit:
864 	spin_unlock_irq(mlx4_tlock(dev));
865 	return err;
866 }
867 
868 int mlx4_get_slave_from_resource_id(struct mlx4_dev *dev,
869 				    enum mlx4_resource type,
870 				    u64 res_id, int *slave)
871 {
872 
873 	struct res_common *r;
874 	int err = -ENOENT;
875 	int id = res_id;
876 
877 	if (type == RES_QP)
878 		id &= 0x7fffff;
879 	spin_lock(mlx4_tlock(dev));
880 
881 	r = find_res(dev, id, type);
882 	if (r) {
883 		*slave = r->owner;
884 		err = 0;
885 	}
886 	spin_unlock(mlx4_tlock(dev));
887 
888 	return err;
889 }
890 
891 static void put_res(struct mlx4_dev *dev, int slave, u64 res_id,
892 		    enum mlx4_resource type)
893 {
894 	struct res_common *r;
895 
896 	spin_lock_irq(mlx4_tlock(dev));
897 	r = find_res(dev, res_id, type);
898 	if (r)
899 		r->state = r->from_state;
900 	spin_unlock_irq(mlx4_tlock(dev));
901 }
902 
903 static int counter_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
904 			     u64 in_param, u64 *out_param, int port);
905 
906 static int handle_existing_counter(struct mlx4_dev *dev, u8 slave, int port,
907 				   int counter_index)
908 {
909 	struct res_common *r;
910 	struct res_counter *counter;
911 	int ret = 0;
912 
913 	if (counter_index == MLX4_SINK_COUNTER_INDEX(dev))
914 		return ret;
915 
916 	spin_lock_irq(mlx4_tlock(dev));
917 	r = find_res(dev, counter_index, RES_COUNTER);
918 	if (!r || r->owner != slave) {
919 		ret = -EINVAL;
920 	} else {
921 		counter = container_of(r, struct res_counter, com);
922 		if (!counter->port)
923 			counter->port = port;
924 	}
925 
926 	spin_unlock_irq(mlx4_tlock(dev));
927 	return ret;
928 }
929 
930 static int handle_unexisting_counter(struct mlx4_dev *dev,
931 				     struct mlx4_qp_context *qpc, u8 slave,
932 				     int port)
933 {
934 	struct mlx4_priv *priv = mlx4_priv(dev);
935 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
936 	struct res_common *tmp;
937 	struct res_counter *counter;
938 	u64 counter_idx = MLX4_SINK_COUNTER_INDEX(dev);
939 	int err = 0;
940 
941 	spin_lock_irq(mlx4_tlock(dev));
942 	list_for_each_entry(tmp,
943 			    &tracker->slave_list[slave].res_list[RES_COUNTER],
944 			    list) {
945 		counter = container_of(tmp, struct res_counter, com);
946 		if (port == counter->port) {
947 			qpc->pri_path.counter_index  = counter->com.res_id;
948 			spin_unlock_irq(mlx4_tlock(dev));
949 			return 0;
950 		}
951 	}
952 	spin_unlock_irq(mlx4_tlock(dev));
953 
954 	/* No existing counter, need to allocate a new counter */
955 	err = counter_alloc_res(dev, slave, RES_OP_RESERVE, 0, 0, &counter_idx,
956 				port);
957 	if (err == -ENOENT) {
958 		err = 0;
959 	} else if (err && err != -ENOSPC) {
960 		mlx4_err(dev, "%s: failed to create new counter for slave %d err %d\n",
961 			 __func__, slave, err);
962 	} else {
963 		qpc->pri_path.counter_index = counter_idx;
964 		mlx4_dbg(dev, "%s: alloc new counter for slave %d index %d\n",
965 			 __func__, slave, qpc->pri_path.counter_index);
966 		err = 0;
967 	}
968 
969 	return err;
970 }
971 
972 static int handle_counter(struct mlx4_dev *dev, struct mlx4_qp_context *qpc,
973 			  u8 slave, int port)
974 {
975 	if (qpc->pri_path.counter_index != MLX4_SINK_COUNTER_INDEX(dev))
976 		return handle_existing_counter(dev, slave, port,
977 					       qpc->pri_path.counter_index);
978 
979 	return handle_unexisting_counter(dev, qpc, slave, port);
980 }
981 
982 static struct res_common *alloc_qp_tr(int id)
983 {
984 	struct res_qp *ret;
985 
986 	ret = kzalloc(sizeof *ret, GFP_KERNEL);
987 	if (!ret)
988 		return NULL;
989 
990 	ret->com.res_id = id;
991 	ret->com.state = RES_QP_RESERVED;
992 	ret->local_qpn = id;
993 	INIT_LIST_HEAD(&ret->mcg_list);
994 	spin_lock_init(&ret->mcg_spl);
995 	atomic_set(&ret->ref_count, 0);
996 
997 	return &ret->com;
998 }
999 
1000 static struct res_common *alloc_mtt_tr(int id, int order)
1001 {
1002 	struct res_mtt *ret;
1003 
1004 	ret = kzalloc(sizeof *ret, GFP_KERNEL);
1005 	if (!ret)
1006 		return NULL;
1007 
1008 	ret->com.res_id = id;
1009 	ret->order = order;
1010 	ret->com.state = RES_MTT_ALLOCATED;
1011 	atomic_set(&ret->ref_count, 0);
1012 
1013 	return &ret->com;
1014 }
1015 
1016 static struct res_common *alloc_mpt_tr(int id, int key)
1017 {
1018 	struct res_mpt *ret;
1019 
1020 	ret = kzalloc(sizeof *ret, GFP_KERNEL);
1021 	if (!ret)
1022 		return NULL;
1023 
1024 	ret->com.res_id = id;
1025 	ret->com.state = RES_MPT_RESERVED;
1026 	ret->key = key;
1027 
1028 	return &ret->com;
1029 }
1030 
1031 static struct res_common *alloc_eq_tr(int id)
1032 {
1033 	struct res_eq *ret;
1034 
1035 	ret = kzalloc(sizeof *ret, GFP_KERNEL);
1036 	if (!ret)
1037 		return NULL;
1038 
1039 	ret->com.res_id = id;
1040 	ret->com.state = RES_EQ_RESERVED;
1041 
1042 	return &ret->com;
1043 }
1044 
1045 static struct res_common *alloc_cq_tr(int id)
1046 {
1047 	struct res_cq *ret;
1048 
1049 	ret = kzalloc(sizeof *ret, GFP_KERNEL);
1050 	if (!ret)
1051 		return NULL;
1052 
1053 	ret->com.res_id = id;
1054 	ret->com.state = RES_CQ_ALLOCATED;
1055 	atomic_set(&ret->ref_count, 0);
1056 
1057 	return &ret->com;
1058 }
1059 
1060 static struct res_common *alloc_srq_tr(int id)
1061 {
1062 	struct res_srq *ret;
1063 
1064 	ret = kzalloc(sizeof *ret, GFP_KERNEL);
1065 	if (!ret)
1066 		return NULL;
1067 
1068 	ret->com.res_id = id;
1069 	ret->com.state = RES_SRQ_ALLOCATED;
1070 	atomic_set(&ret->ref_count, 0);
1071 
1072 	return &ret->com;
1073 }
1074 
1075 static struct res_common *alloc_counter_tr(int id, int port)
1076 {
1077 	struct res_counter *ret;
1078 
1079 	ret = kzalloc(sizeof *ret, GFP_KERNEL);
1080 	if (!ret)
1081 		return NULL;
1082 
1083 	ret->com.res_id = id;
1084 	ret->com.state = RES_COUNTER_ALLOCATED;
1085 	ret->port = port;
1086 
1087 	return &ret->com;
1088 }
1089 
1090 static struct res_common *alloc_xrcdn_tr(int id)
1091 {
1092 	struct res_xrcdn *ret;
1093 
1094 	ret = kzalloc(sizeof *ret, GFP_KERNEL);
1095 	if (!ret)
1096 		return NULL;
1097 
1098 	ret->com.res_id = id;
1099 	ret->com.state = RES_XRCD_ALLOCATED;
1100 
1101 	return &ret->com;
1102 }
1103 
1104 static struct res_common *alloc_fs_rule_tr(u64 id, int qpn)
1105 {
1106 	struct res_fs_rule *ret;
1107 
1108 	ret = kzalloc(sizeof *ret, GFP_KERNEL);
1109 	if (!ret)
1110 		return NULL;
1111 
1112 	ret->com.res_id = id;
1113 	ret->com.state = RES_FS_RULE_ALLOCATED;
1114 	ret->qpn = qpn;
1115 	return &ret->com;
1116 }
1117 
1118 static struct res_common *alloc_tr(u64 id, enum mlx4_resource type, int slave,
1119 				   int extra)
1120 {
1121 	struct res_common *ret;
1122 
1123 	switch (type) {
1124 	case RES_QP:
1125 		ret = alloc_qp_tr(id);
1126 		break;
1127 	case RES_MPT:
1128 		ret = alloc_mpt_tr(id, extra);
1129 		break;
1130 	case RES_MTT:
1131 		ret = alloc_mtt_tr(id, extra);
1132 		break;
1133 	case RES_EQ:
1134 		ret = alloc_eq_tr(id);
1135 		break;
1136 	case RES_CQ:
1137 		ret = alloc_cq_tr(id);
1138 		break;
1139 	case RES_SRQ:
1140 		ret = alloc_srq_tr(id);
1141 		break;
1142 	case RES_MAC:
1143 		pr_err("implementation missing\n");
1144 		return NULL;
1145 	case RES_COUNTER:
1146 		ret = alloc_counter_tr(id, extra);
1147 		break;
1148 	case RES_XRCD:
1149 		ret = alloc_xrcdn_tr(id);
1150 		break;
1151 	case RES_FS_RULE:
1152 		ret = alloc_fs_rule_tr(id, extra);
1153 		break;
1154 	default:
1155 		return NULL;
1156 	}
1157 	if (ret)
1158 		ret->owner = slave;
1159 
1160 	return ret;
1161 }
1162 
1163 int mlx4_calc_vf_counters(struct mlx4_dev *dev, int slave, int port,
1164 			  struct mlx4_counter *data)
1165 {
1166 	struct mlx4_priv *priv = mlx4_priv(dev);
1167 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
1168 	struct res_common *tmp;
1169 	struct res_counter *counter;
1170 	int *counters_arr;
1171 	int i = 0, err = 0;
1172 
1173 	memset(data, 0, sizeof(*data));
1174 
1175 	counters_arr = kmalloc_array(dev->caps.max_counters,
1176 				     sizeof(*counters_arr), GFP_KERNEL);
1177 	if (!counters_arr)
1178 		return -ENOMEM;
1179 
1180 	spin_lock_irq(mlx4_tlock(dev));
1181 	list_for_each_entry(tmp,
1182 			    &tracker->slave_list[slave].res_list[RES_COUNTER],
1183 			    list) {
1184 		counter = container_of(tmp, struct res_counter, com);
1185 		if (counter->port == port) {
1186 			counters_arr[i] = (int)tmp->res_id;
1187 			i++;
1188 		}
1189 	}
1190 	spin_unlock_irq(mlx4_tlock(dev));
1191 	counters_arr[i] = -1;
1192 
1193 	i = 0;
1194 
1195 	while (counters_arr[i] != -1) {
1196 		err = mlx4_get_counter_stats(dev, counters_arr[i], data,
1197 					     0);
1198 		if (err) {
1199 			memset(data, 0, sizeof(*data));
1200 			goto table_changed;
1201 		}
1202 		i++;
1203 	}
1204 
1205 table_changed:
1206 	kfree(counters_arr);
1207 	return 0;
1208 }
1209 
1210 static int add_res_range(struct mlx4_dev *dev, int slave, u64 base, int count,
1211 			 enum mlx4_resource type, int extra)
1212 {
1213 	int i;
1214 	int err;
1215 	struct mlx4_priv *priv = mlx4_priv(dev);
1216 	struct res_common **res_arr;
1217 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
1218 	struct rb_root *root = &tracker->res_tree[type];
1219 
1220 	res_arr = kzalloc(count * sizeof *res_arr, GFP_KERNEL);
1221 	if (!res_arr)
1222 		return -ENOMEM;
1223 
1224 	for (i = 0; i < count; ++i) {
1225 		res_arr[i] = alloc_tr(base + i, type, slave, extra);
1226 		if (!res_arr[i]) {
1227 			for (--i; i >= 0; --i)
1228 				kfree(res_arr[i]);
1229 
1230 			kfree(res_arr);
1231 			return -ENOMEM;
1232 		}
1233 	}
1234 
1235 	spin_lock_irq(mlx4_tlock(dev));
1236 	for (i = 0; i < count; ++i) {
1237 		if (find_res(dev, base + i, type)) {
1238 			err = -EEXIST;
1239 			goto undo;
1240 		}
1241 		err = res_tracker_insert(root, res_arr[i]);
1242 		if (err)
1243 			goto undo;
1244 		list_add_tail(&res_arr[i]->list,
1245 			      &tracker->slave_list[slave].res_list[type]);
1246 	}
1247 	spin_unlock_irq(mlx4_tlock(dev));
1248 	kfree(res_arr);
1249 
1250 	return 0;
1251 
1252 undo:
1253 	for (--i; i >= 0; --i) {
1254 		rb_erase(&res_arr[i]->node, root);
1255 		list_del_init(&res_arr[i]->list);
1256 	}
1257 
1258 	spin_unlock_irq(mlx4_tlock(dev));
1259 
1260 	for (i = 0; i < count; ++i)
1261 		kfree(res_arr[i]);
1262 
1263 	kfree(res_arr);
1264 
1265 	return err;
1266 }
1267 
1268 static int remove_qp_ok(struct res_qp *res)
1269 {
1270 	if (res->com.state == RES_QP_BUSY || atomic_read(&res->ref_count) ||
1271 	    !list_empty(&res->mcg_list)) {
1272 		pr_err("resource tracker: fail to remove qp, state %d, ref_count %d\n",
1273 		       res->com.state, atomic_read(&res->ref_count));
1274 		return -EBUSY;
1275 	} else if (res->com.state != RES_QP_RESERVED) {
1276 		return -EPERM;
1277 	}
1278 
1279 	return 0;
1280 }
1281 
1282 static int remove_mtt_ok(struct res_mtt *res, int order)
1283 {
1284 	if (res->com.state == RES_MTT_BUSY ||
1285 	    atomic_read(&res->ref_count)) {
1286 		pr_devel("%s-%d: state %s, ref_count %d\n",
1287 			 __func__, __LINE__,
1288 			 mtt_states_str(res->com.state),
1289 			 atomic_read(&res->ref_count));
1290 		return -EBUSY;
1291 	} else if (res->com.state != RES_MTT_ALLOCATED)
1292 		return -EPERM;
1293 	else if (res->order != order)
1294 		return -EINVAL;
1295 
1296 	return 0;
1297 }
1298 
1299 static int remove_mpt_ok(struct res_mpt *res)
1300 {
1301 	if (res->com.state == RES_MPT_BUSY)
1302 		return -EBUSY;
1303 	else if (res->com.state != RES_MPT_RESERVED)
1304 		return -EPERM;
1305 
1306 	return 0;
1307 }
1308 
1309 static int remove_eq_ok(struct res_eq *res)
1310 {
1311 	if (res->com.state == RES_MPT_BUSY)
1312 		return -EBUSY;
1313 	else if (res->com.state != RES_MPT_RESERVED)
1314 		return -EPERM;
1315 
1316 	return 0;
1317 }
1318 
1319 static int remove_counter_ok(struct res_counter *res)
1320 {
1321 	if (res->com.state == RES_COUNTER_BUSY)
1322 		return -EBUSY;
1323 	else if (res->com.state != RES_COUNTER_ALLOCATED)
1324 		return -EPERM;
1325 
1326 	return 0;
1327 }
1328 
1329 static int remove_xrcdn_ok(struct res_xrcdn *res)
1330 {
1331 	if (res->com.state == RES_XRCD_BUSY)
1332 		return -EBUSY;
1333 	else if (res->com.state != RES_XRCD_ALLOCATED)
1334 		return -EPERM;
1335 
1336 	return 0;
1337 }
1338 
1339 static int remove_fs_rule_ok(struct res_fs_rule *res)
1340 {
1341 	if (res->com.state == RES_FS_RULE_BUSY)
1342 		return -EBUSY;
1343 	else if (res->com.state != RES_FS_RULE_ALLOCATED)
1344 		return -EPERM;
1345 
1346 	return 0;
1347 }
1348 
1349 static int remove_cq_ok(struct res_cq *res)
1350 {
1351 	if (res->com.state == RES_CQ_BUSY)
1352 		return -EBUSY;
1353 	else if (res->com.state != RES_CQ_ALLOCATED)
1354 		return -EPERM;
1355 
1356 	return 0;
1357 }
1358 
1359 static int remove_srq_ok(struct res_srq *res)
1360 {
1361 	if (res->com.state == RES_SRQ_BUSY)
1362 		return -EBUSY;
1363 	else if (res->com.state != RES_SRQ_ALLOCATED)
1364 		return -EPERM;
1365 
1366 	return 0;
1367 }
1368 
1369 static int remove_ok(struct res_common *res, enum mlx4_resource type, int extra)
1370 {
1371 	switch (type) {
1372 	case RES_QP:
1373 		return remove_qp_ok((struct res_qp *)res);
1374 	case RES_CQ:
1375 		return remove_cq_ok((struct res_cq *)res);
1376 	case RES_SRQ:
1377 		return remove_srq_ok((struct res_srq *)res);
1378 	case RES_MPT:
1379 		return remove_mpt_ok((struct res_mpt *)res);
1380 	case RES_MTT:
1381 		return remove_mtt_ok((struct res_mtt *)res, extra);
1382 	case RES_MAC:
1383 		return -ENOSYS;
1384 	case RES_EQ:
1385 		return remove_eq_ok((struct res_eq *)res);
1386 	case RES_COUNTER:
1387 		return remove_counter_ok((struct res_counter *)res);
1388 	case RES_XRCD:
1389 		return remove_xrcdn_ok((struct res_xrcdn *)res);
1390 	case RES_FS_RULE:
1391 		return remove_fs_rule_ok((struct res_fs_rule *)res);
1392 	default:
1393 		return -EINVAL;
1394 	}
1395 }
1396 
1397 static int rem_res_range(struct mlx4_dev *dev, int slave, u64 base, int count,
1398 			 enum mlx4_resource type, int extra)
1399 {
1400 	u64 i;
1401 	int err;
1402 	struct mlx4_priv *priv = mlx4_priv(dev);
1403 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
1404 	struct res_common *r;
1405 
1406 	spin_lock_irq(mlx4_tlock(dev));
1407 	for (i = base; i < base + count; ++i) {
1408 		r = res_tracker_lookup(&tracker->res_tree[type], i);
1409 		if (!r) {
1410 			err = -ENOENT;
1411 			goto out;
1412 		}
1413 		if (r->owner != slave) {
1414 			err = -EPERM;
1415 			goto out;
1416 		}
1417 		err = remove_ok(r, type, extra);
1418 		if (err)
1419 			goto out;
1420 	}
1421 
1422 	for (i = base; i < base + count; ++i) {
1423 		r = res_tracker_lookup(&tracker->res_tree[type], i);
1424 		rb_erase(&r->node, &tracker->res_tree[type]);
1425 		list_del(&r->list);
1426 		kfree(r);
1427 	}
1428 	err = 0;
1429 
1430 out:
1431 	spin_unlock_irq(mlx4_tlock(dev));
1432 
1433 	return err;
1434 }
1435 
1436 static int qp_res_start_move_to(struct mlx4_dev *dev, int slave, int qpn,
1437 				enum res_qp_states state, struct res_qp **qp,
1438 				int alloc)
1439 {
1440 	struct mlx4_priv *priv = mlx4_priv(dev);
1441 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
1442 	struct res_qp *r;
1443 	int err = 0;
1444 
1445 	spin_lock_irq(mlx4_tlock(dev));
1446 	r = res_tracker_lookup(&tracker->res_tree[RES_QP], qpn);
1447 	if (!r)
1448 		err = -ENOENT;
1449 	else if (r->com.owner != slave)
1450 		err = -EPERM;
1451 	else {
1452 		switch (state) {
1453 		case RES_QP_BUSY:
1454 			mlx4_dbg(dev, "%s: failed RES_QP, 0x%llx\n",
1455 				 __func__, r->com.res_id);
1456 			err = -EBUSY;
1457 			break;
1458 
1459 		case RES_QP_RESERVED:
1460 			if (r->com.state == RES_QP_MAPPED && !alloc)
1461 				break;
1462 
1463 			mlx4_dbg(dev, "failed RES_QP, 0x%llx\n", r->com.res_id);
1464 			err = -EINVAL;
1465 			break;
1466 
1467 		case RES_QP_MAPPED:
1468 			if ((r->com.state == RES_QP_RESERVED && alloc) ||
1469 			    r->com.state == RES_QP_HW)
1470 				break;
1471 			else {
1472 				mlx4_dbg(dev, "failed RES_QP, 0x%llx\n",
1473 					  r->com.res_id);
1474 				err = -EINVAL;
1475 			}
1476 
1477 			break;
1478 
1479 		case RES_QP_HW:
1480 			if (r->com.state != RES_QP_MAPPED)
1481 				err = -EINVAL;
1482 			break;
1483 		default:
1484 			err = -EINVAL;
1485 		}
1486 
1487 		if (!err) {
1488 			r->com.from_state = r->com.state;
1489 			r->com.to_state = state;
1490 			r->com.state = RES_QP_BUSY;
1491 			if (qp)
1492 				*qp = r;
1493 		}
1494 	}
1495 
1496 	spin_unlock_irq(mlx4_tlock(dev));
1497 
1498 	return err;
1499 }
1500 
1501 static int mr_res_start_move_to(struct mlx4_dev *dev, int slave, int index,
1502 				enum res_mpt_states state, struct res_mpt **mpt)
1503 {
1504 	struct mlx4_priv *priv = mlx4_priv(dev);
1505 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
1506 	struct res_mpt *r;
1507 	int err = 0;
1508 
1509 	spin_lock_irq(mlx4_tlock(dev));
1510 	r = res_tracker_lookup(&tracker->res_tree[RES_MPT], index);
1511 	if (!r)
1512 		err = -ENOENT;
1513 	else if (r->com.owner != slave)
1514 		err = -EPERM;
1515 	else {
1516 		switch (state) {
1517 		case RES_MPT_BUSY:
1518 			err = -EINVAL;
1519 			break;
1520 
1521 		case RES_MPT_RESERVED:
1522 			if (r->com.state != RES_MPT_MAPPED)
1523 				err = -EINVAL;
1524 			break;
1525 
1526 		case RES_MPT_MAPPED:
1527 			if (r->com.state != RES_MPT_RESERVED &&
1528 			    r->com.state != RES_MPT_HW)
1529 				err = -EINVAL;
1530 			break;
1531 
1532 		case RES_MPT_HW:
1533 			if (r->com.state != RES_MPT_MAPPED)
1534 				err = -EINVAL;
1535 			break;
1536 		default:
1537 			err = -EINVAL;
1538 		}
1539 
1540 		if (!err) {
1541 			r->com.from_state = r->com.state;
1542 			r->com.to_state = state;
1543 			r->com.state = RES_MPT_BUSY;
1544 			if (mpt)
1545 				*mpt = r;
1546 		}
1547 	}
1548 
1549 	spin_unlock_irq(mlx4_tlock(dev));
1550 
1551 	return err;
1552 }
1553 
1554 static int eq_res_start_move_to(struct mlx4_dev *dev, int slave, int index,
1555 				enum res_eq_states state, struct res_eq **eq)
1556 {
1557 	struct mlx4_priv *priv = mlx4_priv(dev);
1558 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
1559 	struct res_eq *r;
1560 	int err = 0;
1561 
1562 	spin_lock_irq(mlx4_tlock(dev));
1563 	r = res_tracker_lookup(&tracker->res_tree[RES_EQ], index);
1564 	if (!r)
1565 		err = -ENOENT;
1566 	else if (r->com.owner != slave)
1567 		err = -EPERM;
1568 	else {
1569 		switch (state) {
1570 		case RES_EQ_BUSY:
1571 			err = -EINVAL;
1572 			break;
1573 
1574 		case RES_EQ_RESERVED:
1575 			if (r->com.state != RES_EQ_HW)
1576 				err = -EINVAL;
1577 			break;
1578 
1579 		case RES_EQ_HW:
1580 			if (r->com.state != RES_EQ_RESERVED)
1581 				err = -EINVAL;
1582 			break;
1583 
1584 		default:
1585 			err = -EINVAL;
1586 		}
1587 
1588 		if (!err) {
1589 			r->com.from_state = r->com.state;
1590 			r->com.to_state = state;
1591 			r->com.state = RES_EQ_BUSY;
1592 			if (eq)
1593 				*eq = r;
1594 		}
1595 	}
1596 
1597 	spin_unlock_irq(mlx4_tlock(dev));
1598 
1599 	return err;
1600 }
1601 
1602 static int cq_res_start_move_to(struct mlx4_dev *dev, int slave, int cqn,
1603 				enum res_cq_states state, struct res_cq **cq)
1604 {
1605 	struct mlx4_priv *priv = mlx4_priv(dev);
1606 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
1607 	struct res_cq *r;
1608 	int err;
1609 
1610 	spin_lock_irq(mlx4_tlock(dev));
1611 	r = res_tracker_lookup(&tracker->res_tree[RES_CQ], cqn);
1612 	if (!r) {
1613 		err = -ENOENT;
1614 	} else if (r->com.owner != slave) {
1615 		err = -EPERM;
1616 	} else if (state == RES_CQ_ALLOCATED) {
1617 		if (r->com.state != RES_CQ_HW)
1618 			err = -EINVAL;
1619 		else if (atomic_read(&r->ref_count))
1620 			err = -EBUSY;
1621 		else
1622 			err = 0;
1623 	} else if (state != RES_CQ_HW || r->com.state != RES_CQ_ALLOCATED) {
1624 		err = -EINVAL;
1625 	} else {
1626 		err = 0;
1627 	}
1628 
1629 	if (!err) {
1630 		r->com.from_state = r->com.state;
1631 		r->com.to_state = state;
1632 		r->com.state = RES_CQ_BUSY;
1633 		if (cq)
1634 			*cq = r;
1635 	}
1636 
1637 	spin_unlock_irq(mlx4_tlock(dev));
1638 
1639 	return err;
1640 }
1641 
1642 static int srq_res_start_move_to(struct mlx4_dev *dev, int slave, int index,
1643 				 enum res_srq_states state, struct res_srq **srq)
1644 {
1645 	struct mlx4_priv *priv = mlx4_priv(dev);
1646 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
1647 	struct res_srq *r;
1648 	int err = 0;
1649 
1650 	spin_lock_irq(mlx4_tlock(dev));
1651 	r = res_tracker_lookup(&tracker->res_tree[RES_SRQ], index);
1652 	if (!r) {
1653 		err = -ENOENT;
1654 	} else if (r->com.owner != slave) {
1655 		err = -EPERM;
1656 	} else if (state == RES_SRQ_ALLOCATED) {
1657 		if (r->com.state != RES_SRQ_HW)
1658 			err = -EINVAL;
1659 		else if (atomic_read(&r->ref_count))
1660 			err = -EBUSY;
1661 	} else if (state != RES_SRQ_HW || r->com.state != RES_SRQ_ALLOCATED) {
1662 		err = -EINVAL;
1663 	}
1664 
1665 	if (!err) {
1666 		r->com.from_state = r->com.state;
1667 		r->com.to_state = state;
1668 		r->com.state = RES_SRQ_BUSY;
1669 		if (srq)
1670 			*srq = r;
1671 	}
1672 
1673 	spin_unlock_irq(mlx4_tlock(dev));
1674 
1675 	return err;
1676 }
1677 
1678 static void res_abort_move(struct mlx4_dev *dev, int slave,
1679 			   enum mlx4_resource type, int id)
1680 {
1681 	struct mlx4_priv *priv = mlx4_priv(dev);
1682 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
1683 	struct res_common *r;
1684 
1685 	spin_lock_irq(mlx4_tlock(dev));
1686 	r = res_tracker_lookup(&tracker->res_tree[type], id);
1687 	if (r && (r->owner == slave))
1688 		r->state = r->from_state;
1689 	spin_unlock_irq(mlx4_tlock(dev));
1690 }
1691 
1692 static void res_end_move(struct mlx4_dev *dev, int slave,
1693 			 enum mlx4_resource type, int id)
1694 {
1695 	struct mlx4_priv *priv = mlx4_priv(dev);
1696 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
1697 	struct res_common *r;
1698 
1699 	spin_lock_irq(mlx4_tlock(dev));
1700 	r = res_tracker_lookup(&tracker->res_tree[type], id);
1701 	if (r && (r->owner == slave))
1702 		r->state = r->to_state;
1703 	spin_unlock_irq(mlx4_tlock(dev));
1704 }
1705 
1706 static int valid_reserved(struct mlx4_dev *dev, int slave, int qpn)
1707 {
1708 	return mlx4_is_qp_reserved(dev, qpn) &&
1709 		(mlx4_is_master(dev) || mlx4_is_guest_proxy(dev, slave, qpn));
1710 }
1711 
1712 static int fw_reserved(struct mlx4_dev *dev, int qpn)
1713 {
1714 	return qpn < dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
1715 }
1716 
1717 static int qp_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
1718 			u64 in_param, u64 *out_param)
1719 {
1720 	int err;
1721 	int count;
1722 	int align;
1723 	int base;
1724 	int qpn;
1725 	u8 flags;
1726 
1727 	switch (op) {
1728 	case RES_OP_RESERVE:
1729 		count = get_param_l(&in_param) & 0xffffff;
1730 		/* Turn off all unsupported QP allocation flags that the
1731 		 * slave tries to set.
1732 		 */
1733 		flags = (get_param_l(&in_param) >> 24) & dev->caps.alloc_res_qp_mask;
1734 		align = get_param_h(&in_param);
1735 		err = mlx4_grant_resource(dev, slave, RES_QP, count, 0);
1736 		if (err)
1737 			return err;
1738 
1739 		err = __mlx4_qp_reserve_range(dev, count, align, &base, flags);
1740 		if (err) {
1741 			mlx4_release_resource(dev, slave, RES_QP, count, 0);
1742 			return err;
1743 		}
1744 
1745 		err = add_res_range(dev, slave, base, count, RES_QP, 0);
1746 		if (err) {
1747 			mlx4_release_resource(dev, slave, RES_QP, count, 0);
1748 			__mlx4_qp_release_range(dev, base, count);
1749 			return err;
1750 		}
1751 		set_param_l(out_param, base);
1752 		break;
1753 	case RES_OP_MAP_ICM:
1754 		qpn = get_param_l(&in_param) & 0x7fffff;
1755 		if (valid_reserved(dev, slave, qpn)) {
1756 			err = add_res_range(dev, slave, qpn, 1, RES_QP, 0);
1757 			if (err)
1758 				return err;
1759 		}
1760 
1761 		err = qp_res_start_move_to(dev, slave, qpn, RES_QP_MAPPED,
1762 					   NULL, 1);
1763 		if (err)
1764 			return err;
1765 
1766 		if (!fw_reserved(dev, qpn)) {
1767 			err = __mlx4_qp_alloc_icm(dev, qpn, GFP_KERNEL);
1768 			if (err) {
1769 				res_abort_move(dev, slave, RES_QP, qpn);
1770 				return err;
1771 			}
1772 		}
1773 
1774 		res_end_move(dev, slave, RES_QP, qpn);
1775 		break;
1776 
1777 	default:
1778 		err = -EINVAL;
1779 		break;
1780 	}
1781 	return err;
1782 }
1783 
1784 static int mtt_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
1785 			 u64 in_param, u64 *out_param)
1786 {
1787 	int err = -EINVAL;
1788 	int base;
1789 	int order;
1790 
1791 	if (op != RES_OP_RESERVE_AND_MAP)
1792 		return err;
1793 
1794 	order = get_param_l(&in_param);
1795 
1796 	err = mlx4_grant_resource(dev, slave, RES_MTT, 1 << order, 0);
1797 	if (err)
1798 		return err;
1799 
1800 	base = __mlx4_alloc_mtt_range(dev, order);
1801 	if (base == -1) {
1802 		mlx4_release_resource(dev, slave, RES_MTT, 1 << order, 0);
1803 		return -ENOMEM;
1804 	}
1805 
1806 	err = add_res_range(dev, slave, base, 1, RES_MTT, order);
1807 	if (err) {
1808 		mlx4_release_resource(dev, slave, RES_MTT, 1 << order, 0);
1809 		__mlx4_free_mtt_range(dev, base, order);
1810 	} else {
1811 		set_param_l(out_param, base);
1812 	}
1813 
1814 	return err;
1815 }
1816 
1817 static int mpt_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
1818 			 u64 in_param, u64 *out_param)
1819 {
1820 	int err = -EINVAL;
1821 	int index;
1822 	int id;
1823 	struct res_mpt *mpt;
1824 
1825 	switch (op) {
1826 	case RES_OP_RESERVE:
1827 		err = mlx4_grant_resource(dev, slave, RES_MPT, 1, 0);
1828 		if (err)
1829 			break;
1830 
1831 		index = __mlx4_mpt_reserve(dev);
1832 		if (index == -1) {
1833 			mlx4_release_resource(dev, slave, RES_MPT, 1, 0);
1834 			break;
1835 		}
1836 		id = index & mpt_mask(dev);
1837 
1838 		err = add_res_range(dev, slave, id, 1, RES_MPT, index);
1839 		if (err) {
1840 			mlx4_release_resource(dev, slave, RES_MPT, 1, 0);
1841 			__mlx4_mpt_release(dev, index);
1842 			break;
1843 		}
1844 		set_param_l(out_param, index);
1845 		break;
1846 	case RES_OP_MAP_ICM:
1847 		index = get_param_l(&in_param);
1848 		id = index & mpt_mask(dev);
1849 		err = mr_res_start_move_to(dev, slave, id,
1850 					   RES_MPT_MAPPED, &mpt);
1851 		if (err)
1852 			return err;
1853 
1854 		err = __mlx4_mpt_alloc_icm(dev, mpt->key, GFP_KERNEL);
1855 		if (err) {
1856 			res_abort_move(dev, slave, RES_MPT, id);
1857 			return err;
1858 		}
1859 
1860 		res_end_move(dev, slave, RES_MPT, id);
1861 		break;
1862 	}
1863 	return err;
1864 }
1865 
1866 static int cq_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
1867 			u64 in_param, u64 *out_param)
1868 {
1869 	int cqn;
1870 	int err;
1871 
1872 	switch (op) {
1873 	case RES_OP_RESERVE_AND_MAP:
1874 		err = mlx4_grant_resource(dev, slave, RES_CQ, 1, 0);
1875 		if (err)
1876 			break;
1877 
1878 		err = __mlx4_cq_alloc_icm(dev, &cqn);
1879 		if (err) {
1880 			mlx4_release_resource(dev, slave, RES_CQ, 1, 0);
1881 			break;
1882 		}
1883 
1884 		err = add_res_range(dev, slave, cqn, 1, RES_CQ, 0);
1885 		if (err) {
1886 			mlx4_release_resource(dev, slave, RES_CQ, 1, 0);
1887 			__mlx4_cq_free_icm(dev, cqn);
1888 			break;
1889 		}
1890 
1891 		set_param_l(out_param, cqn);
1892 		break;
1893 
1894 	default:
1895 		err = -EINVAL;
1896 	}
1897 
1898 	return err;
1899 }
1900 
1901 static int srq_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
1902 			 u64 in_param, u64 *out_param)
1903 {
1904 	int srqn;
1905 	int err;
1906 
1907 	switch (op) {
1908 	case RES_OP_RESERVE_AND_MAP:
1909 		err = mlx4_grant_resource(dev, slave, RES_SRQ, 1, 0);
1910 		if (err)
1911 			break;
1912 
1913 		err = __mlx4_srq_alloc_icm(dev, &srqn);
1914 		if (err) {
1915 			mlx4_release_resource(dev, slave, RES_SRQ, 1, 0);
1916 			break;
1917 		}
1918 
1919 		err = add_res_range(dev, slave, srqn, 1, RES_SRQ, 0);
1920 		if (err) {
1921 			mlx4_release_resource(dev, slave, RES_SRQ, 1, 0);
1922 			__mlx4_srq_free_icm(dev, srqn);
1923 			break;
1924 		}
1925 
1926 		set_param_l(out_param, srqn);
1927 		break;
1928 
1929 	default:
1930 		err = -EINVAL;
1931 	}
1932 
1933 	return err;
1934 }
1935 
1936 static int mac_find_smac_ix_in_slave(struct mlx4_dev *dev, int slave, int port,
1937 				     u8 smac_index, u64 *mac)
1938 {
1939 	struct mlx4_priv *priv = mlx4_priv(dev);
1940 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
1941 	struct list_head *mac_list =
1942 		&tracker->slave_list[slave].res_list[RES_MAC];
1943 	struct mac_res *res, *tmp;
1944 
1945 	list_for_each_entry_safe(res, tmp, mac_list, list) {
1946 		if (res->smac_index == smac_index && res->port == (u8) port) {
1947 			*mac = res->mac;
1948 			return 0;
1949 		}
1950 	}
1951 	return -ENOENT;
1952 }
1953 
1954 static int mac_add_to_slave(struct mlx4_dev *dev, int slave, u64 mac, int port, u8 smac_index)
1955 {
1956 	struct mlx4_priv *priv = mlx4_priv(dev);
1957 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
1958 	struct list_head *mac_list =
1959 		&tracker->slave_list[slave].res_list[RES_MAC];
1960 	struct mac_res *res, *tmp;
1961 
1962 	list_for_each_entry_safe(res, tmp, mac_list, list) {
1963 		if (res->mac == mac && res->port == (u8) port) {
1964 			/* mac found. update ref count */
1965 			++res->ref_count;
1966 			return 0;
1967 		}
1968 	}
1969 
1970 	if (mlx4_grant_resource(dev, slave, RES_MAC, 1, port))
1971 		return -EINVAL;
1972 	res = kzalloc(sizeof *res, GFP_KERNEL);
1973 	if (!res) {
1974 		mlx4_release_resource(dev, slave, RES_MAC, 1, port);
1975 		return -ENOMEM;
1976 	}
1977 	res->mac = mac;
1978 	res->port = (u8) port;
1979 	res->smac_index = smac_index;
1980 	res->ref_count = 1;
1981 	list_add_tail(&res->list,
1982 		      &tracker->slave_list[slave].res_list[RES_MAC]);
1983 	return 0;
1984 }
1985 
1986 static void mac_del_from_slave(struct mlx4_dev *dev, int slave, u64 mac,
1987 			       int port)
1988 {
1989 	struct mlx4_priv *priv = mlx4_priv(dev);
1990 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
1991 	struct list_head *mac_list =
1992 		&tracker->slave_list[slave].res_list[RES_MAC];
1993 	struct mac_res *res, *tmp;
1994 
1995 	list_for_each_entry_safe(res, tmp, mac_list, list) {
1996 		if (res->mac == mac && res->port == (u8) port) {
1997 			if (!--res->ref_count) {
1998 				list_del(&res->list);
1999 				mlx4_release_resource(dev, slave, RES_MAC, 1, port);
2000 				kfree(res);
2001 			}
2002 			break;
2003 		}
2004 	}
2005 }
2006 
2007 static void rem_slave_macs(struct mlx4_dev *dev, int slave)
2008 {
2009 	struct mlx4_priv *priv = mlx4_priv(dev);
2010 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
2011 	struct list_head *mac_list =
2012 		&tracker->slave_list[slave].res_list[RES_MAC];
2013 	struct mac_res *res, *tmp;
2014 	int i;
2015 
2016 	list_for_each_entry_safe(res, tmp, mac_list, list) {
2017 		list_del(&res->list);
2018 		/* dereference the mac the num times the slave referenced it */
2019 		for (i = 0; i < res->ref_count; i++)
2020 			__mlx4_unregister_mac(dev, res->port, res->mac);
2021 		mlx4_release_resource(dev, slave, RES_MAC, 1, res->port);
2022 		kfree(res);
2023 	}
2024 }
2025 
2026 static int mac_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
2027 			 u64 in_param, u64 *out_param, int in_port)
2028 {
2029 	int err = -EINVAL;
2030 	int port;
2031 	u64 mac;
2032 	u8 smac_index;
2033 
2034 	if (op != RES_OP_RESERVE_AND_MAP)
2035 		return err;
2036 
2037 	port = !in_port ? get_param_l(out_param) : in_port;
2038 	port = mlx4_slave_convert_port(
2039 			dev, slave, port);
2040 
2041 	if (port < 0)
2042 		return -EINVAL;
2043 	mac = in_param;
2044 
2045 	err = __mlx4_register_mac(dev, port, mac);
2046 	if (err >= 0) {
2047 		smac_index = err;
2048 		set_param_l(out_param, err);
2049 		err = 0;
2050 	}
2051 
2052 	if (!err) {
2053 		err = mac_add_to_slave(dev, slave, mac, port, smac_index);
2054 		if (err)
2055 			__mlx4_unregister_mac(dev, port, mac);
2056 	}
2057 	return err;
2058 }
2059 
2060 static int vlan_add_to_slave(struct mlx4_dev *dev, int slave, u16 vlan,
2061 			     int port, int vlan_index)
2062 {
2063 	struct mlx4_priv *priv = mlx4_priv(dev);
2064 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
2065 	struct list_head *vlan_list =
2066 		&tracker->slave_list[slave].res_list[RES_VLAN];
2067 	struct vlan_res *res, *tmp;
2068 
2069 	list_for_each_entry_safe(res, tmp, vlan_list, list) {
2070 		if (res->vlan == vlan && res->port == (u8) port) {
2071 			/* vlan found. update ref count */
2072 			++res->ref_count;
2073 			return 0;
2074 		}
2075 	}
2076 
2077 	if (mlx4_grant_resource(dev, slave, RES_VLAN, 1, port))
2078 		return -EINVAL;
2079 	res = kzalloc(sizeof(*res), GFP_KERNEL);
2080 	if (!res) {
2081 		mlx4_release_resource(dev, slave, RES_VLAN, 1, port);
2082 		return -ENOMEM;
2083 	}
2084 	res->vlan = vlan;
2085 	res->port = (u8) port;
2086 	res->vlan_index = vlan_index;
2087 	res->ref_count = 1;
2088 	list_add_tail(&res->list,
2089 		      &tracker->slave_list[slave].res_list[RES_VLAN]);
2090 	return 0;
2091 }
2092 
2093 
2094 static void vlan_del_from_slave(struct mlx4_dev *dev, int slave, u16 vlan,
2095 				int port)
2096 {
2097 	struct mlx4_priv *priv = mlx4_priv(dev);
2098 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
2099 	struct list_head *vlan_list =
2100 		&tracker->slave_list[slave].res_list[RES_VLAN];
2101 	struct vlan_res *res, *tmp;
2102 
2103 	list_for_each_entry_safe(res, tmp, vlan_list, list) {
2104 		if (res->vlan == vlan && res->port == (u8) port) {
2105 			if (!--res->ref_count) {
2106 				list_del(&res->list);
2107 				mlx4_release_resource(dev, slave, RES_VLAN,
2108 						      1, port);
2109 				kfree(res);
2110 			}
2111 			break;
2112 		}
2113 	}
2114 }
2115 
2116 static void rem_slave_vlans(struct mlx4_dev *dev, int slave)
2117 {
2118 	struct mlx4_priv *priv = mlx4_priv(dev);
2119 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
2120 	struct list_head *vlan_list =
2121 		&tracker->slave_list[slave].res_list[RES_VLAN];
2122 	struct vlan_res *res, *tmp;
2123 	int i;
2124 
2125 	list_for_each_entry_safe(res, tmp, vlan_list, list) {
2126 		list_del(&res->list);
2127 		/* dereference the vlan the num times the slave referenced it */
2128 		for (i = 0; i < res->ref_count; i++)
2129 			__mlx4_unregister_vlan(dev, res->port, res->vlan);
2130 		mlx4_release_resource(dev, slave, RES_VLAN, 1, res->port);
2131 		kfree(res);
2132 	}
2133 }
2134 
2135 static int vlan_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
2136 			  u64 in_param, u64 *out_param, int in_port)
2137 {
2138 	struct mlx4_priv *priv = mlx4_priv(dev);
2139 	struct mlx4_slave_state *slave_state = priv->mfunc.master.slave_state;
2140 	int err;
2141 	u16 vlan;
2142 	int vlan_index;
2143 	int port;
2144 
2145 	port = !in_port ? get_param_l(out_param) : in_port;
2146 
2147 	if (!port || op != RES_OP_RESERVE_AND_MAP)
2148 		return -EINVAL;
2149 
2150 	port = mlx4_slave_convert_port(
2151 			dev, slave, port);
2152 
2153 	if (port < 0)
2154 		return -EINVAL;
2155 	/* upstream kernels had NOP for reg/unreg vlan. Continue this. */
2156 	if (!in_port && port > 0 && port <= dev->caps.num_ports) {
2157 		slave_state[slave].old_vlan_api = true;
2158 		return 0;
2159 	}
2160 
2161 	vlan = (u16) in_param;
2162 
2163 	err = __mlx4_register_vlan(dev, port, vlan, &vlan_index);
2164 	if (!err) {
2165 		set_param_l(out_param, (u32) vlan_index);
2166 		err = vlan_add_to_slave(dev, slave, vlan, port, vlan_index);
2167 		if (err)
2168 			__mlx4_unregister_vlan(dev, port, vlan);
2169 	}
2170 	return err;
2171 }
2172 
2173 static int counter_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
2174 			     u64 in_param, u64 *out_param, int port)
2175 {
2176 	u32 index;
2177 	int err;
2178 
2179 	if (op != RES_OP_RESERVE)
2180 		return -EINVAL;
2181 
2182 	err = mlx4_grant_resource(dev, slave, RES_COUNTER, 1, 0);
2183 	if (err)
2184 		return err;
2185 
2186 	err = __mlx4_counter_alloc(dev, &index);
2187 	if (err) {
2188 		mlx4_release_resource(dev, slave, RES_COUNTER, 1, 0);
2189 		return err;
2190 	}
2191 
2192 	err = add_res_range(dev, slave, index, 1, RES_COUNTER, port);
2193 	if (err) {
2194 		__mlx4_counter_free(dev, index);
2195 		mlx4_release_resource(dev, slave, RES_COUNTER, 1, 0);
2196 	} else {
2197 		set_param_l(out_param, index);
2198 	}
2199 
2200 	return err;
2201 }
2202 
2203 static int xrcdn_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
2204 			   u64 in_param, u64 *out_param)
2205 {
2206 	u32 xrcdn;
2207 	int err;
2208 
2209 	if (op != RES_OP_RESERVE)
2210 		return -EINVAL;
2211 
2212 	err = __mlx4_xrcd_alloc(dev, &xrcdn);
2213 	if (err)
2214 		return err;
2215 
2216 	err = add_res_range(dev, slave, xrcdn, 1, RES_XRCD, 0);
2217 	if (err)
2218 		__mlx4_xrcd_free(dev, xrcdn);
2219 	else
2220 		set_param_l(out_param, xrcdn);
2221 
2222 	return err;
2223 }
2224 
2225 int mlx4_ALLOC_RES_wrapper(struct mlx4_dev *dev, int slave,
2226 			   struct mlx4_vhcr *vhcr,
2227 			   struct mlx4_cmd_mailbox *inbox,
2228 			   struct mlx4_cmd_mailbox *outbox,
2229 			   struct mlx4_cmd_info *cmd)
2230 {
2231 	int err;
2232 	int alop = vhcr->op_modifier;
2233 
2234 	switch (vhcr->in_modifier & 0xFF) {
2235 	case RES_QP:
2236 		err = qp_alloc_res(dev, slave, vhcr->op_modifier, alop,
2237 				   vhcr->in_param, &vhcr->out_param);
2238 		break;
2239 
2240 	case RES_MTT:
2241 		err = mtt_alloc_res(dev, slave, vhcr->op_modifier, alop,
2242 				    vhcr->in_param, &vhcr->out_param);
2243 		break;
2244 
2245 	case RES_MPT:
2246 		err = mpt_alloc_res(dev, slave, vhcr->op_modifier, alop,
2247 				    vhcr->in_param, &vhcr->out_param);
2248 		break;
2249 
2250 	case RES_CQ:
2251 		err = cq_alloc_res(dev, slave, vhcr->op_modifier, alop,
2252 				   vhcr->in_param, &vhcr->out_param);
2253 		break;
2254 
2255 	case RES_SRQ:
2256 		err = srq_alloc_res(dev, slave, vhcr->op_modifier, alop,
2257 				    vhcr->in_param, &vhcr->out_param);
2258 		break;
2259 
2260 	case RES_MAC:
2261 		err = mac_alloc_res(dev, slave, vhcr->op_modifier, alop,
2262 				    vhcr->in_param, &vhcr->out_param,
2263 				    (vhcr->in_modifier >> 8) & 0xFF);
2264 		break;
2265 
2266 	case RES_VLAN:
2267 		err = vlan_alloc_res(dev, slave, vhcr->op_modifier, alop,
2268 				     vhcr->in_param, &vhcr->out_param,
2269 				     (vhcr->in_modifier >> 8) & 0xFF);
2270 		break;
2271 
2272 	case RES_COUNTER:
2273 		err = counter_alloc_res(dev, slave, vhcr->op_modifier, alop,
2274 					vhcr->in_param, &vhcr->out_param, 0);
2275 		break;
2276 
2277 	case RES_XRCD:
2278 		err = xrcdn_alloc_res(dev, slave, vhcr->op_modifier, alop,
2279 				      vhcr->in_param, &vhcr->out_param);
2280 		break;
2281 
2282 	default:
2283 		err = -EINVAL;
2284 		break;
2285 	}
2286 
2287 	return err;
2288 }
2289 
2290 static int qp_free_res(struct mlx4_dev *dev, int slave, int op, int cmd,
2291 		       u64 in_param)
2292 {
2293 	int err;
2294 	int count;
2295 	int base;
2296 	int qpn;
2297 
2298 	switch (op) {
2299 	case RES_OP_RESERVE:
2300 		base = get_param_l(&in_param) & 0x7fffff;
2301 		count = get_param_h(&in_param);
2302 		err = rem_res_range(dev, slave, base, count, RES_QP, 0);
2303 		if (err)
2304 			break;
2305 		mlx4_release_resource(dev, slave, RES_QP, count, 0);
2306 		__mlx4_qp_release_range(dev, base, count);
2307 		break;
2308 	case RES_OP_MAP_ICM:
2309 		qpn = get_param_l(&in_param) & 0x7fffff;
2310 		err = qp_res_start_move_to(dev, slave, qpn, RES_QP_RESERVED,
2311 					   NULL, 0);
2312 		if (err)
2313 			return err;
2314 
2315 		if (!fw_reserved(dev, qpn))
2316 			__mlx4_qp_free_icm(dev, qpn);
2317 
2318 		res_end_move(dev, slave, RES_QP, qpn);
2319 
2320 		if (valid_reserved(dev, slave, qpn))
2321 			err = rem_res_range(dev, slave, qpn, 1, RES_QP, 0);
2322 		break;
2323 	default:
2324 		err = -EINVAL;
2325 		break;
2326 	}
2327 	return err;
2328 }
2329 
2330 static int mtt_free_res(struct mlx4_dev *dev, int slave, int op, int cmd,
2331 			u64 in_param, u64 *out_param)
2332 {
2333 	int err = -EINVAL;
2334 	int base;
2335 	int order;
2336 
2337 	if (op != RES_OP_RESERVE_AND_MAP)
2338 		return err;
2339 
2340 	base = get_param_l(&in_param);
2341 	order = get_param_h(&in_param);
2342 	err = rem_res_range(dev, slave, base, 1, RES_MTT, order);
2343 	if (!err) {
2344 		mlx4_release_resource(dev, slave, RES_MTT, 1 << order, 0);
2345 		__mlx4_free_mtt_range(dev, base, order);
2346 	}
2347 	return err;
2348 }
2349 
2350 static int mpt_free_res(struct mlx4_dev *dev, int slave, int op, int cmd,
2351 			u64 in_param)
2352 {
2353 	int err = -EINVAL;
2354 	int index;
2355 	int id;
2356 	struct res_mpt *mpt;
2357 
2358 	switch (op) {
2359 	case RES_OP_RESERVE:
2360 		index = get_param_l(&in_param);
2361 		id = index & mpt_mask(dev);
2362 		err = get_res(dev, slave, id, RES_MPT, &mpt);
2363 		if (err)
2364 			break;
2365 		index = mpt->key;
2366 		put_res(dev, slave, id, RES_MPT);
2367 
2368 		err = rem_res_range(dev, slave, id, 1, RES_MPT, 0);
2369 		if (err)
2370 			break;
2371 		mlx4_release_resource(dev, slave, RES_MPT, 1, 0);
2372 		__mlx4_mpt_release(dev, index);
2373 		break;
2374 	case RES_OP_MAP_ICM:
2375 		index = get_param_l(&in_param);
2376 		id = index & mpt_mask(dev);
2377 		err = mr_res_start_move_to(dev, slave, id,
2378 					   RES_MPT_RESERVED, &mpt);
2379 		if (err)
2380 			return err;
2381 
2382 		__mlx4_mpt_free_icm(dev, mpt->key);
2383 		res_end_move(dev, slave, RES_MPT, id);
2384 		break;
2385 	default:
2386 		err = -EINVAL;
2387 		break;
2388 	}
2389 	return err;
2390 }
2391 
2392 static int cq_free_res(struct mlx4_dev *dev, int slave, int op, int cmd,
2393 		       u64 in_param, u64 *out_param)
2394 {
2395 	int cqn;
2396 	int err;
2397 
2398 	switch (op) {
2399 	case RES_OP_RESERVE_AND_MAP:
2400 		cqn = get_param_l(&in_param);
2401 		err = rem_res_range(dev, slave, cqn, 1, RES_CQ, 0);
2402 		if (err)
2403 			break;
2404 
2405 		mlx4_release_resource(dev, slave, RES_CQ, 1, 0);
2406 		__mlx4_cq_free_icm(dev, cqn);
2407 		break;
2408 
2409 	default:
2410 		err = -EINVAL;
2411 		break;
2412 	}
2413 
2414 	return err;
2415 }
2416 
2417 static int srq_free_res(struct mlx4_dev *dev, int slave, int op, int cmd,
2418 			u64 in_param, u64 *out_param)
2419 {
2420 	int srqn;
2421 	int err;
2422 
2423 	switch (op) {
2424 	case RES_OP_RESERVE_AND_MAP:
2425 		srqn = get_param_l(&in_param);
2426 		err = rem_res_range(dev, slave, srqn, 1, RES_SRQ, 0);
2427 		if (err)
2428 			break;
2429 
2430 		mlx4_release_resource(dev, slave, RES_SRQ, 1, 0);
2431 		__mlx4_srq_free_icm(dev, srqn);
2432 		break;
2433 
2434 	default:
2435 		err = -EINVAL;
2436 		break;
2437 	}
2438 
2439 	return err;
2440 }
2441 
2442 static int mac_free_res(struct mlx4_dev *dev, int slave, int op, int cmd,
2443 			    u64 in_param, u64 *out_param, int in_port)
2444 {
2445 	int port;
2446 	int err = 0;
2447 
2448 	switch (op) {
2449 	case RES_OP_RESERVE_AND_MAP:
2450 		port = !in_port ? get_param_l(out_param) : in_port;
2451 		port = mlx4_slave_convert_port(
2452 				dev, slave, port);
2453 
2454 		if (port < 0)
2455 			return -EINVAL;
2456 		mac_del_from_slave(dev, slave, in_param, port);
2457 		__mlx4_unregister_mac(dev, port, in_param);
2458 		break;
2459 	default:
2460 		err = -EINVAL;
2461 		break;
2462 	}
2463 
2464 	return err;
2465 
2466 }
2467 
2468 static int vlan_free_res(struct mlx4_dev *dev, int slave, int op, int cmd,
2469 			    u64 in_param, u64 *out_param, int port)
2470 {
2471 	struct mlx4_priv *priv = mlx4_priv(dev);
2472 	struct mlx4_slave_state *slave_state = priv->mfunc.master.slave_state;
2473 	int err = 0;
2474 
2475 	port = mlx4_slave_convert_port(
2476 			dev, slave, port);
2477 
2478 	if (port < 0)
2479 		return -EINVAL;
2480 	switch (op) {
2481 	case RES_OP_RESERVE_AND_MAP:
2482 		if (slave_state[slave].old_vlan_api)
2483 			return 0;
2484 		if (!port)
2485 			return -EINVAL;
2486 		vlan_del_from_slave(dev, slave, in_param, port);
2487 		__mlx4_unregister_vlan(dev, port, in_param);
2488 		break;
2489 	default:
2490 		err = -EINVAL;
2491 		break;
2492 	}
2493 
2494 	return err;
2495 }
2496 
2497 static int counter_free_res(struct mlx4_dev *dev, int slave, int op, int cmd,
2498 			    u64 in_param, u64 *out_param)
2499 {
2500 	int index;
2501 	int err;
2502 
2503 	if (op != RES_OP_RESERVE)
2504 		return -EINVAL;
2505 
2506 	index = get_param_l(&in_param);
2507 	if (index == MLX4_SINK_COUNTER_INDEX(dev))
2508 		return 0;
2509 
2510 	err = rem_res_range(dev, slave, index, 1, RES_COUNTER, 0);
2511 	if (err)
2512 		return err;
2513 
2514 	__mlx4_counter_free(dev, index);
2515 	mlx4_release_resource(dev, slave, RES_COUNTER, 1, 0);
2516 
2517 	return err;
2518 }
2519 
2520 static int xrcdn_free_res(struct mlx4_dev *dev, int slave, int op, int cmd,
2521 			  u64 in_param, u64 *out_param)
2522 {
2523 	int xrcdn;
2524 	int err;
2525 
2526 	if (op != RES_OP_RESERVE)
2527 		return -EINVAL;
2528 
2529 	xrcdn = get_param_l(&in_param);
2530 	err = rem_res_range(dev, slave, xrcdn, 1, RES_XRCD, 0);
2531 	if (err)
2532 		return err;
2533 
2534 	__mlx4_xrcd_free(dev, xrcdn);
2535 
2536 	return err;
2537 }
2538 
2539 int mlx4_FREE_RES_wrapper(struct mlx4_dev *dev, int slave,
2540 			  struct mlx4_vhcr *vhcr,
2541 			  struct mlx4_cmd_mailbox *inbox,
2542 			  struct mlx4_cmd_mailbox *outbox,
2543 			  struct mlx4_cmd_info *cmd)
2544 {
2545 	int err = -EINVAL;
2546 	int alop = vhcr->op_modifier;
2547 
2548 	switch (vhcr->in_modifier & 0xFF) {
2549 	case RES_QP:
2550 		err = qp_free_res(dev, slave, vhcr->op_modifier, alop,
2551 				  vhcr->in_param);
2552 		break;
2553 
2554 	case RES_MTT:
2555 		err = mtt_free_res(dev, slave, vhcr->op_modifier, alop,
2556 				   vhcr->in_param, &vhcr->out_param);
2557 		break;
2558 
2559 	case RES_MPT:
2560 		err = mpt_free_res(dev, slave, vhcr->op_modifier, alop,
2561 				   vhcr->in_param);
2562 		break;
2563 
2564 	case RES_CQ:
2565 		err = cq_free_res(dev, slave, vhcr->op_modifier, alop,
2566 				  vhcr->in_param, &vhcr->out_param);
2567 		break;
2568 
2569 	case RES_SRQ:
2570 		err = srq_free_res(dev, slave, vhcr->op_modifier, alop,
2571 				   vhcr->in_param, &vhcr->out_param);
2572 		break;
2573 
2574 	case RES_MAC:
2575 		err = mac_free_res(dev, slave, vhcr->op_modifier, alop,
2576 				   vhcr->in_param, &vhcr->out_param,
2577 				   (vhcr->in_modifier >> 8) & 0xFF);
2578 		break;
2579 
2580 	case RES_VLAN:
2581 		err = vlan_free_res(dev, slave, vhcr->op_modifier, alop,
2582 				    vhcr->in_param, &vhcr->out_param,
2583 				    (vhcr->in_modifier >> 8) & 0xFF);
2584 		break;
2585 
2586 	case RES_COUNTER:
2587 		err = counter_free_res(dev, slave, vhcr->op_modifier, alop,
2588 				       vhcr->in_param, &vhcr->out_param);
2589 		break;
2590 
2591 	case RES_XRCD:
2592 		err = xrcdn_free_res(dev, slave, vhcr->op_modifier, alop,
2593 				     vhcr->in_param, &vhcr->out_param);
2594 
2595 	default:
2596 		break;
2597 	}
2598 	return err;
2599 }
2600 
2601 /* ugly but other choices are uglier */
2602 static int mr_phys_mpt(struct mlx4_mpt_entry *mpt)
2603 {
2604 	return (be32_to_cpu(mpt->flags) >> 9) & 1;
2605 }
2606 
2607 static int mr_get_mtt_addr(struct mlx4_mpt_entry *mpt)
2608 {
2609 	return (int)be64_to_cpu(mpt->mtt_addr) & 0xfffffff8;
2610 }
2611 
2612 static int mr_get_mtt_size(struct mlx4_mpt_entry *mpt)
2613 {
2614 	return be32_to_cpu(mpt->mtt_sz);
2615 }
2616 
2617 static u32 mr_get_pd(struct mlx4_mpt_entry *mpt)
2618 {
2619 	return be32_to_cpu(mpt->pd_flags) & 0x00ffffff;
2620 }
2621 
2622 static int mr_is_fmr(struct mlx4_mpt_entry *mpt)
2623 {
2624 	return be32_to_cpu(mpt->pd_flags) & MLX4_MPT_PD_FLAG_FAST_REG;
2625 }
2626 
2627 static int mr_is_bind_enabled(struct mlx4_mpt_entry *mpt)
2628 {
2629 	return be32_to_cpu(mpt->flags) & MLX4_MPT_FLAG_BIND_ENABLE;
2630 }
2631 
2632 static int mr_is_region(struct mlx4_mpt_entry *mpt)
2633 {
2634 	return be32_to_cpu(mpt->flags) & MLX4_MPT_FLAG_REGION;
2635 }
2636 
2637 static int qp_get_mtt_addr(struct mlx4_qp_context *qpc)
2638 {
2639 	return be32_to_cpu(qpc->mtt_base_addr_l) & 0xfffffff8;
2640 }
2641 
2642 static int srq_get_mtt_addr(struct mlx4_srq_context *srqc)
2643 {
2644 	return be32_to_cpu(srqc->mtt_base_addr_l) & 0xfffffff8;
2645 }
2646 
2647 static int qp_get_mtt_size(struct mlx4_qp_context *qpc)
2648 {
2649 	int page_shift = (qpc->log_page_size & 0x3f) + 12;
2650 	int log_sq_size = (qpc->sq_size_stride >> 3) & 0xf;
2651 	int log_sq_sride = qpc->sq_size_stride & 7;
2652 	int log_rq_size = (qpc->rq_size_stride >> 3) & 0xf;
2653 	int log_rq_stride = qpc->rq_size_stride & 7;
2654 	int srq = (be32_to_cpu(qpc->srqn) >> 24) & 1;
2655 	int rss = (be32_to_cpu(qpc->flags) >> 13) & 1;
2656 	u32 ts = (be32_to_cpu(qpc->flags) >> 16) & 0xff;
2657 	int xrc = (ts == MLX4_QP_ST_XRC) ? 1 : 0;
2658 	int sq_size;
2659 	int rq_size;
2660 	int total_pages;
2661 	int total_mem;
2662 	int page_offset = (be32_to_cpu(qpc->params2) >> 6) & 0x3f;
2663 
2664 	sq_size = 1 << (log_sq_size + log_sq_sride + 4);
2665 	rq_size = (srq|rss|xrc) ? 0 : (1 << (log_rq_size + log_rq_stride + 4));
2666 	total_mem = sq_size + rq_size;
2667 	total_pages =
2668 		roundup_pow_of_two((total_mem + (page_offset << 6)) >>
2669 				   page_shift);
2670 
2671 	return total_pages;
2672 }
2673 
2674 static int check_mtt_range(struct mlx4_dev *dev, int slave, int start,
2675 			   int size, struct res_mtt *mtt)
2676 {
2677 	int res_start = mtt->com.res_id;
2678 	int res_size = (1 << mtt->order);
2679 
2680 	if (start < res_start || start + size > res_start + res_size)
2681 		return -EPERM;
2682 	return 0;
2683 }
2684 
2685 int mlx4_SW2HW_MPT_wrapper(struct mlx4_dev *dev, int slave,
2686 			   struct mlx4_vhcr *vhcr,
2687 			   struct mlx4_cmd_mailbox *inbox,
2688 			   struct mlx4_cmd_mailbox *outbox,
2689 			   struct mlx4_cmd_info *cmd)
2690 {
2691 	int err;
2692 	int index = vhcr->in_modifier;
2693 	struct res_mtt *mtt;
2694 	struct res_mpt *mpt;
2695 	int mtt_base = mr_get_mtt_addr(inbox->buf) / dev->caps.mtt_entry_sz;
2696 	int phys;
2697 	int id;
2698 	u32 pd;
2699 	int pd_slave;
2700 
2701 	id = index & mpt_mask(dev);
2702 	err = mr_res_start_move_to(dev, slave, id, RES_MPT_HW, &mpt);
2703 	if (err)
2704 		return err;
2705 
2706 	/* Disable memory windows for VFs. */
2707 	if (!mr_is_region(inbox->buf)) {
2708 		err = -EPERM;
2709 		goto ex_abort;
2710 	}
2711 
2712 	/* Make sure that the PD bits related to the slave id are zeros. */
2713 	pd = mr_get_pd(inbox->buf);
2714 	pd_slave = (pd >> 17) & 0x7f;
2715 	if (pd_slave != 0 && --pd_slave != slave) {
2716 		err = -EPERM;
2717 		goto ex_abort;
2718 	}
2719 
2720 	if (mr_is_fmr(inbox->buf)) {
2721 		/* FMR and Bind Enable are forbidden in slave devices. */
2722 		if (mr_is_bind_enabled(inbox->buf)) {
2723 			err = -EPERM;
2724 			goto ex_abort;
2725 		}
2726 		/* FMR and Memory Windows are also forbidden. */
2727 		if (!mr_is_region(inbox->buf)) {
2728 			err = -EPERM;
2729 			goto ex_abort;
2730 		}
2731 	}
2732 
2733 	phys = mr_phys_mpt(inbox->buf);
2734 	if (!phys) {
2735 		err = get_res(dev, slave, mtt_base, RES_MTT, &mtt);
2736 		if (err)
2737 			goto ex_abort;
2738 
2739 		err = check_mtt_range(dev, slave, mtt_base,
2740 				      mr_get_mtt_size(inbox->buf), mtt);
2741 		if (err)
2742 			goto ex_put;
2743 
2744 		mpt->mtt = mtt;
2745 	}
2746 
2747 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
2748 	if (err)
2749 		goto ex_put;
2750 
2751 	if (!phys) {
2752 		atomic_inc(&mtt->ref_count);
2753 		put_res(dev, slave, mtt->com.res_id, RES_MTT);
2754 	}
2755 
2756 	res_end_move(dev, slave, RES_MPT, id);
2757 	return 0;
2758 
2759 ex_put:
2760 	if (!phys)
2761 		put_res(dev, slave, mtt->com.res_id, RES_MTT);
2762 ex_abort:
2763 	res_abort_move(dev, slave, RES_MPT, id);
2764 
2765 	return err;
2766 }
2767 
2768 int mlx4_HW2SW_MPT_wrapper(struct mlx4_dev *dev, int slave,
2769 			   struct mlx4_vhcr *vhcr,
2770 			   struct mlx4_cmd_mailbox *inbox,
2771 			   struct mlx4_cmd_mailbox *outbox,
2772 			   struct mlx4_cmd_info *cmd)
2773 {
2774 	int err;
2775 	int index = vhcr->in_modifier;
2776 	struct res_mpt *mpt;
2777 	int id;
2778 
2779 	id = index & mpt_mask(dev);
2780 	err = mr_res_start_move_to(dev, slave, id, RES_MPT_MAPPED, &mpt);
2781 	if (err)
2782 		return err;
2783 
2784 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
2785 	if (err)
2786 		goto ex_abort;
2787 
2788 	if (mpt->mtt)
2789 		atomic_dec(&mpt->mtt->ref_count);
2790 
2791 	res_end_move(dev, slave, RES_MPT, id);
2792 	return 0;
2793 
2794 ex_abort:
2795 	res_abort_move(dev, slave, RES_MPT, id);
2796 
2797 	return err;
2798 }
2799 
2800 int mlx4_QUERY_MPT_wrapper(struct mlx4_dev *dev, int slave,
2801 			   struct mlx4_vhcr *vhcr,
2802 			   struct mlx4_cmd_mailbox *inbox,
2803 			   struct mlx4_cmd_mailbox *outbox,
2804 			   struct mlx4_cmd_info *cmd)
2805 {
2806 	int err;
2807 	int index = vhcr->in_modifier;
2808 	struct res_mpt *mpt;
2809 	int id;
2810 
2811 	id = index & mpt_mask(dev);
2812 	err = get_res(dev, slave, id, RES_MPT, &mpt);
2813 	if (err)
2814 		return err;
2815 
2816 	if (mpt->com.from_state == RES_MPT_MAPPED) {
2817 		/* In order to allow rereg in SRIOV, we need to alter the MPT entry. To do
2818 		 * that, the VF must read the MPT. But since the MPT entry memory is not
2819 		 * in the VF's virtual memory space, it must use QUERY_MPT to obtain the
2820 		 * entry contents. To guarantee that the MPT cannot be changed, the driver
2821 		 * must perform HW2SW_MPT before this query and return the MPT entry to HW
2822 		 * ownership fofollowing the change. The change here allows the VF to
2823 		 * perform QUERY_MPT also when the entry is in SW ownership.
2824 		 */
2825 		struct mlx4_mpt_entry *mpt_entry = mlx4_table_find(
2826 					&mlx4_priv(dev)->mr_table.dmpt_table,
2827 					mpt->key, NULL);
2828 
2829 		if (NULL == mpt_entry || NULL == outbox->buf) {
2830 			err = -EINVAL;
2831 			goto out;
2832 		}
2833 
2834 		memcpy(outbox->buf, mpt_entry, sizeof(*mpt_entry));
2835 
2836 		err = 0;
2837 	} else if (mpt->com.from_state == RES_MPT_HW) {
2838 		err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
2839 	} else {
2840 		err = -EBUSY;
2841 		goto out;
2842 	}
2843 
2844 
2845 out:
2846 	put_res(dev, slave, id, RES_MPT);
2847 	return err;
2848 }
2849 
2850 static int qp_get_rcqn(struct mlx4_qp_context *qpc)
2851 {
2852 	return be32_to_cpu(qpc->cqn_recv) & 0xffffff;
2853 }
2854 
2855 static int qp_get_scqn(struct mlx4_qp_context *qpc)
2856 {
2857 	return be32_to_cpu(qpc->cqn_send) & 0xffffff;
2858 }
2859 
2860 static u32 qp_get_srqn(struct mlx4_qp_context *qpc)
2861 {
2862 	return be32_to_cpu(qpc->srqn) & 0x1ffffff;
2863 }
2864 
2865 static void adjust_proxy_tun_qkey(struct mlx4_dev *dev, struct mlx4_vhcr *vhcr,
2866 				  struct mlx4_qp_context *context)
2867 {
2868 	u32 qpn = vhcr->in_modifier & 0xffffff;
2869 	u32 qkey = 0;
2870 
2871 	if (mlx4_get_parav_qkey(dev, qpn, &qkey))
2872 		return;
2873 
2874 	/* adjust qkey in qp context */
2875 	context->qkey = cpu_to_be32(qkey);
2876 }
2877 
2878 static int adjust_qp_sched_queue(struct mlx4_dev *dev, int slave,
2879 				 struct mlx4_qp_context *qpc,
2880 				 struct mlx4_cmd_mailbox *inbox);
2881 
2882 int mlx4_RST2INIT_QP_wrapper(struct mlx4_dev *dev, int slave,
2883 			     struct mlx4_vhcr *vhcr,
2884 			     struct mlx4_cmd_mailbox *inbox,
2885 			     struct mlx4_cmd_mailbox *outbox,
2886 			     struct mlx4_cmd_info *cmd)
2887 {
2888 	int err;
2889 	int qpn = vhcr->in_modifier & 0x7fffff;
2890 	struct res_mtt *mtt;
2891 	struct res_qp *qp;
2892 	struct mlx4_qp_context *qpc = inbox->buf + 8;
2893 	int mtt_base = qp_get_mtt_addr(qpc) / dev->caps.mtt_entry_sz;
2894 	int mtt_size = qp_get_mtt_size(qpc);
2895 	struct res_cq *rcq;
2896 	struct res_cq *scq;
2897 	int rcqn = qp_get_rcqn(qpc);
2898 	int scqn = qp_get_scqn(qpc);
2899 	u32 srqn = qp_get_srqn(qpc) & 0xffffff;
2900 	int use_srq = (qp_get_srqn(qpc) >> 24) & 1;
2901 	struct res_srq *srq;
2902 	int local_qpn = be32_to_cpu(qpc->local_qpn) & 0xffffff;
2903 
2904 	err = adjust_qp_sched_queue(dev, slave, qpc, inbox);
2905 	if (err)
2906 		return err;
2907 
2908 	err = qp_res_start_move_to(dev, slave, qpn, RES_QP_HW, &qp, 0);
2909 	if (err)
2910 		return err;
2911 	qp->local_qpn = local_qpn;
2912 	qp->sched_queue = 0;
2913 	qp->param3 = 0;
2914 	qp->vlan_control = 0;
2915 	qp->fvl_rx = 0;
2916 	qp->pri_path_fl = 0;
2917 	qp->vlan_index = 0;
2918 	qp->feup = 0;
2919 	qp->qpc_flags = be32_to_cpu(qpc->flags);
2920 
2921 	err = get_res(dev, slave, mtt_base, RES_MTT, &mtt);
2922 	if (err)
2923 		goto ex_abort;
2924 
2925 	err = check_mtt_range(dev, slave, mtt_base, mtt_size, mtt);
2926 	if (err)
2927 		goto ex_put_mtt;
2928 
2929 	err = get_res(dev, slave, rcqn, RES_CQ, &rcq);
2930 	if (err)
2931 		goto ex_put_mtt;
2932 
2933 	if (scqn != rcqn) {
2934 		err = get_res(dev, slave, scqn, RES_CQ, &scq);
2935 		if (err)
2936 			goto ex_put_rcq;
2937 	} else
2938 		scq = rcq;
2939 
2940 	if (use_srq) {
2941 		err = get_res(dev, slave, srqn, RES_SRQ, &srq);
2942 		if (err)
2943 			goto ex_put_scq;
2944 	}
2945 
2946 	adjust_proxy_tun_qkey(dev, vhcr, qpc);
2947 	update_pkey_index(dev, slave, inbox);
2948 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
2949 	if (err)
2950 		goto ex_put_srq;
2951 	atomic_inc(&mtt->ref_count);
2952 	qp->mtt = mtt;
2953 	atomic_inc(&rcq->ref_count);
2954 	qp->rcq = rcq;
2955 	atomic_inc(&scq->ref_count);
2956 	qp->scq = scq;
2957 
2958 	if (scqn != rcqn)
2959 		put_res(dev, slave, scqn, RES_CQ);
2960 
2961 	if (use_srq) {
2962 		atomic_inc(&srq->ref_count);
2963 		put_res(dev, slave, srqn, RES_SRQ);
2964 		qp->srq = srq;
2965 	}
2966 	put_res(dev, slave, rcqn, RES_CQ);
2967 	put_res(dev, slave, mtt_base, RES_MTT);
2968 	res_end_move(dev, slave, RES_QP, qpn);
2969 
2970 	return 0;
2971 
2972 ex_put_srq:
2973 	if (use_srq)
2974 		put_res(dev, slave, srqn, RES_SRQ);
2975 ex_put_scq:
2976 	if (scqn != rcqn)
2977 		put_res(dev, slave, scqn, RES_CQ);
2978 ex_put_rcq:
2979 	put_res(dev, slave, rcqn, RES_CQ);
2980 ex_put_mtt:
2981 	put_res(dev, slave, mtt_base, RES_MTT);
2982 ex_abort:
2983 	res_abort_move(dev, slave, RES_QP, qpn);
2984 
2985 	return err;
2986 }
2987 
2988 static int eq_get_mtt_addr(struct mlx4_eq_context *eqc)
2989 {
2990 	return be32_to_cpu(eqc->mtt_base_addr_l) & 0xfffffff8;
2991 }
2992 
2993 static int eq_get_mtt_size(struct mlx4_eq_context *eqc)
2994 {
2995 	int log_eq_size = eqc->log_eq_size & 0x1f;
2996 	int page_shift = (eqc->log_page_size & 0x3f) + 12;
2997 
2998 	if (log_eq_size + 5 < page_shift)
2999 		return 1;
3000 
3001 	return 1 << (log_eq_size + 5 - page_shift);
3002 }
3003 
3004 static int cq_get_mtt_addr(struct mlx4_cq_context *cqc)
3005 {
3006 	return be32_to_cpu(cqc->mtt_base_addr_l) & 0xfffffff8;
3007 }
3008 
3009 static int cq_get_mtt_size(struct mlx4_cq_context *cqc)
3010 {
3011 	int log_cq_size = (be32_to_cpu(cqc->logsize_usrpage) >> 24) & 0x1f;
3012 	int page_shift = (cqc->log_page_size & 0x3f) + 12;
3013 
3014 	if (log_cq_size + 5 < page_shift)
3015 		return 1;
3016 
3017 	return 1 << (log_cq_size + 5 - page_shift);
3018 }
3019 
3020 int mlx4_SW2HW_EQ_wrapper(struct mlx4_dev *dev, int slave,
3021 			  struct mlx4_vhcr *vhcr,
3022 			  struct mlx4_cmd_mailbox *inbox,
3023 			  struct mlx4_cmd_mailbox *outbox,
3024 			  struct mlx4_cmd_info *cmd)
3025 {
3026 	int err;
3027 	int eqn = vhcr->in_modifier;
3028 	int res_id = (slave << 10) | eqn;
3029 	struct mlx4_eq_context *eqc = inbox->buf;
3030 	int mtt_base = eq_get_mtt_addr(eqc) / dev->caps.mtt_entry_sz;
3031 	int mtt_size = eq_get_mtt_size(eqc);
3032 	struct res_eq *eq;
3033 	struct res_mtt *mtt;
3034 
3035 	err = add_res_range(dev, slave, res_id, 1, RES_EQ, 0);
3036 	if (err)
3037 		return err;
3038 	err = eq_res_start_move_to(dev, slave, res_id, RES_EQ_HW, &eq);
3039 	if (err)
3040 		goto out_add;
3041 
3042 	err = get_res(dev, slave, mtt_base, RES_MTT, &mtt);
3043 	if (err)
3044 		goto out_move;
3045 
3046 	err = check_mtt_range(dev, slave, mtt_base, mtt_size, mtt);
3047 	if (err)
3048 		goto out_put;
3049 
3050 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3051 	if (err)
3052 		goto out_put;
3053 
3054 	atomic_inc(&mtt->ref_count);
3055 	eq->mtt = mtt;
3056 	put_res(dev, slave, mtt->com.res_id, RES_MTT);
3057 	res_end_move(dev, slave, RES_EQ, res_id);
3058 	return 0;
3059 
3060 out_put:
3061 	put_res(dev, slave, mtt->com.res_id, RES_MTT);
3062 out_move:
3063 	res_abort_move(dev, slave, RES_EQ, res_id);
3064 out_add:
3065 	rem_res_range(dev, slave, res_id, 1, RES_EQ, 0);
3066 	return err;
3067 }
3068 
3069 int mlx4_CONFIG_DEV_wrapper(struct mlx4_dev *dev, int slave,
3070 			    struct mlx4_vhcr *vhcr,
3071 			    struct mlx4_cmd_mailbox *inbox,
3072 			    struct mlx4_cmd_mailbox *outbox,
3073 			    struct mlx4_cmd_info *cmd)
3074 {
3075 	int err;
3076 	u8 get = vhcr->op_modifier;
3077 
3078 	if (get != 1)
3079 		return -EPERM;
3080 
3081 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3082 
3083 	return err;
3084 }
3085 
3086 static int get_containing_mtt(struct mlx4_dev *dev, int slave, int start,
3087 			      int len, struct res_mtt **res)
3088 {
3089 	struct mlx4_priv *priv = mlx4_priv(dev);
3090 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
3091 	struct res_mtt *mtt;
3092 	int err = -EINVAL;
3093 
3094 	spin_lock_irq(mlx4_tlock(dev));
3095 	list_for_each_entry(mtt, &tracker->slave_list[slave].res_list[RES_MTT],
3096 			    com.list) {
3097 		if (!check_mtt_range(dev, slave, start, len, mtt)) {
3098 			*res = mtt;
3099 			mtt->com.from_state = mtt->com.state;
3100 			mtt->com.state = RES_MTT_BUSY;
3101 			err = 0;
3102 			break;
3103 		}
3104 	}
3105 	spin_unlock_irq(mlx4_tlock(dev));
3106 
3107 	return err;
3108 }
3109 
3110 static int verify_qp_parameters(struct mlx4_dev *dev,
3111 				struct mlx4_vhcr *vhcr,
3112 				struct mlx4_cmd_mailbox *inbox,
3113 				enum qp_transition transition, u8 slave)
3114 {
3115 	u32			qp_type;
3116 	u32			qpn;
3117 	struct mlx4_qp_context	*qp_ctx;
3118 	enum mlx4_qp_optpar	optpar;
3119 	int port;
3120 	int num_gids;
3121 
3122 	qp_ctx  = inbox->buf + 8;
3123 	qp_type	= (be32_to_cpu(qp_ctx->flags) >> 16) & 0xff;
3124 	optpar	= be32_to_cpu(*(__be32 *) inbox->buf);
3125 
3126 	if (slave != mlx4_master_func_num(dev)) {
3127 		qp_ctx->params2 &= ~MLX4_QP_BIT_FPP;
3128 		/* setting QP rate-limit is disallowed for VFs */
3129 		if (qp_ctx->rate_limit_params)
3130 			return -EPERM;
3131 	}
3132 
3133 	switch (qp_type) {
3134 	case MLX4_QP_ST_RC:
3135 	case MLX4_QP_ST_XRC:
3136 	case MLX4_QP_ST_UC:
3137 		switch (transition) {
3138 		case QP_TRANS_INIT2RTR:
3139 		case QP_TRANS_RTR2RTS:
3140 		case QP_TRANS_RTS2RTS:
3141 		case QP_TRANS_SQD2SQD:
3142 		case QP_TRANS_SQD2RTS:
3143 			if (slave != mlx4_master_func_num(dev)) {
3144 				if (optpar & MLX4_QP_OPTPAR_PRIMARY_ADDR_PATH) {
3145 					port = (qp_ctx->pri_path.sched_queue >> 6 & 1) + 1;
3146 					if (dev->caps.port_mask[port] != MLX4_PORT_TYPE_IB)
3147 						num_gids = mlx4_get_slave_num_gids(dev, slave, port);
3148 					else
3149 						num_gids = 1;
3150 					if (qp_ctx->pri_path.mgid_index >= num_gids)
3151 						return -EINVAL;
3152 				}
3153 				if (optpar & MLX4_QP_OPTPAR_ALT_ADDR_PATH) {
3154 					port = (qp_ctx->alt_path.sched_queue >> 6 & 1) + 1;
3155 					if (dev->caps.port_mask[port] != MLX4_PORT_TYPE_IB)
3156 						num_gids = mlx4_get_slave_num_gids(dev, slave, port);
3157 					else
3158 						num_gids = 1;
3159 					if (qp_ctx->alt_path.mgid_index >= num_gids)
3160 						return -EINVAL;
3161 				}
3162 			}
3163 			break;
3164 		default:
3165 			break;
3166 		}
3167 		break;
3168 
3169 	case MLX4_QP_ST_MLX:
3170 		qpn = vhcr->in_modifier & 0x7fffff;
3171 		port = (qp_ctx->pri_path.sched_queue >> 6 & 1) + 1;
3172 		if (transition == QP_TRANS_INIT2RTR &&
3173 		    slave != mlx4_master_func_num(dev) &&
3174 		    mlx4_is_qp_reserved(dev, qpn) &&
3175 		    !mlx4_vf_smi_enabled(dev, slave, port)) {
3176 			/* only enabled VFs may create MLX proxy QPs */
3177 			mlx4_err(dev, "%s: unprivileged slave %d attempting to create an MLX proxy special QP on port %d\n",
3178 				 __func__, slave, port);
3179 			return -EPERM;
3180 		}
3181 		break;
3182 
3183 	default:
3184 		break;
3185 	}
3186 
3187 	return 0;
3188 }
3189 
3190 int mlx4_WRITE_MTT_wrapper(struct mlx4_dev *dev, int slave,
3191 			   struct mlx4_vhcr *vhcr,
3192 			   struct mlx4_cmd_mailbox *inbox,
3193 			   struct mlx4_cmd_mailbox *outbox,
3194 			   struct mlx4_cmd_info *cmd)
3195 {
3196 	struct mlx4_mtt mtt;
3197 	__be64 *page_list = inbox->buf;
3198 	u64 *pg_list = (u64 *)page_list;
3199 	int i;
3200 	struct res_mtt *rmtt = NULL;
3201 	int start = be64_to_cpu(page_list[0]);
3202 	int npages = vhcr->in_modifier;
3203 	int err;
3204 
3205 	err = get_containing_mtt(dev, slave, start, npages, &rmtt);
3206 	if (err)
3207 		return err;
3208 
3209 	/* Call the SW implementation of write_mtt:
3210 	 * - Prepare a dummy mtt struct
3211 	 * - Translate inbox contents to simple addresses in host endianness */
3212 	mtt.offset = 0;  /* TBD this is broken but I don't handle it since
3213 			    we don't really use it */
3214 	mtt.order = 0;
3215 	mtt.page_shift = 0;
3216 	for (i = 0; i < npages; ++i)
3217 		pg_list[i + 2] = (be64_to_cpu(page_list[i + 2]) & ~1ULL);
3218 
3219 	err = __mlx4_write_mtt(dev, &mtt, be64_to_cpu(page_list[0]), npages,
3220 			       ((u64 *)page_list + 2));
3221 
3222 	if (rmtt)
3223 		put_res(dev, slave, rmtt->com.res_id, RES_MTT);
3224 
3225 	return err;
3226 }
3227 
3228 int mlx4_HW2SW_EQ_wrapper(struct mlx4_dev *dev, int slave,
3229 			  struct mlx4_vhcr *vhcr,
3230 			  struct mlx4_cmd_mailbox *inbox,
3231 			  struct mlx4_cmd_mailbox *outbox,
3232 			  struct mlx4_cmd_info *cmd)
3233 {
3234 	int eqn = vhcr->in_modifier;
3235 	int res_id = eqn | (slave << 10);
3236 	struct res_eq *eq;
3237 	int err;
3238 
3239 	err = eq_res_start_move_to(dev, slave, res_id, RES_EQ_RESERVED, &eq);
3240 	if (err)
3241 		return err;
3242 
3243 	err = get_res(dev, slave, eq->mtt->com.res_id, RES_MTT, NULL);
3244 	if (err)
3245 		goto ex_abort;
3246 
3247 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3248 	if (err)
3249 		goto ex_put;
3250 
3251 	atomic_dec(&eq->mtt->ref_count);
3252 	put_res(dev, slave, eq->mtt->com.res_id, RES_MTT);
3253 	res_end_move(dev, slave, RES_EQ, res_id);
3254 	rem_res_range(dev, slave, res_id, 1, RES_EQ, 0);
3255 
3256 	return 0;
3257 
3258 ex_put:
3259 	put_res(dev, slave, eq->mtt->com.res_id, RES_MTT);
3260 ex_abort:
3261 	res_abort_move(dev, slave, RES_EQ, res_id);
3262 
3263 	return err;
3264 }
3265 
3266 int mlx4_GEN_EQE(struct mlx4_dev *dev, int slave, struct mlx4_eqe *eqe)
3267 {
3268 	struct mlx4_priv *priv = mlx4_priv(dev);
3269 	struct mlx4_slave_event_eq_info *event_eq;
3270 	struct mlx4_cmd_mailbox *mailbox;
3271 	u32 in_modifier = 0;
3272 	int err;
3273 	int res_id;
3274 	struct res_eq *req;
3275 
3276 	if (!priv->mfunc.master.slave_state)
3277 		return -EINVAL;
3278 
3279 	/* check for slave valid, slave not PF, and slave active */
3280 	if (slave < 0 || slave > dev->persist->num_vfs ||
3281 	    slave == dev->caps.function ||
3282 	    !priv->mfunc.master.slave_state[slave].active)
3283 		return 0;
3284 
3285 	event_eq = &priv->mfunc.master.slave_state[slave].event_eq[eqe->type];
3286 
3287 	/* Create the event only if the slave is registered */
3288 	if (event_eq->eqn < 0)
3289 		return 0;
3290 
3291 	mutex_lock(&priv->mfunc.master.gen_eqe_mutex[slave]);
3292 	res_id = (slave << 10) | event_eq->eqn;
3293 	err = get_res(dev, slave, res_id, RES_EQ, &req);
3294 	if (err)
3295 		goto unlock;
3296 
3297 	if (req->com.from_state != RES_EQ_HW) {
3298 		err = -EINVAL;
3299 		goto put;
3300 	}
3301 
3302 	mailbox = mlx4_alloc_cmd_mailbox(dev);
3303 	if (IS_ERR(mailbox)) {
3304 		err = PTR_ERR(mailbox);
3305 		goto put;
3306 	}
3307 
3308 	if (eqe->type == MLX4_EVENT_TYPE_CMD) {
3309 		++event_eq->token;
3310 		eqe->event.cmd.token = cpu_to_be16(event_eq->token);
3311 	}
3312 
3313 	memcpy(mailbox->buf, (u8 *) eqe, 28);
3314 
3315 	in_modifier = (slave & 0xff) | ((event_eq->eqn & 0x3ff) << 16);
3316 
3317 	err = mlx4_cmd(dev, mailbox->dma, in_modifier, 0,
3318 		       MLX4_CMD_GEN_EQE, MLX4_CMD_TIME_CLASS_B,
3319 		       MLX4_CMD_NATIVE);
3320 
3321 	put_res(dev, slave, res_id, RES_EQ);
3322 	mutex_unlock(&priv->mfunc.master.gen_eqe_mutex[slave]);
3323 	mlx4_free_cmd_mailbox(dev, mailbox);
3324 	return err;
3325 
3326 put:
3327 	put_res(dev, slave, res_id, RES_EQ);
3328 
3329 unlock:
3330 	mutex_unlock(&priv->mfunc.master.gen_eqe_mutex[slave]);
3331 	return err;
3332 }
3333 
3334 int mlx4_QUERY_EQ_wrapper(struct mlx4_dev *dev, int slave,
3335 			  struct mlx4_vhcr *vhcr,
3336 			  struct mlx4_cmd_mailbox *inbox,
3337 			  struct mlx4_cmd_mailbox *outbox,
3338 			  struct mlx4_cmd_info *cmd)
3339 {
3340 	int eqn = vhcr->in_modifier;
3341 	int res_id = eqn | (slave << 10);
3342 	struct res_eq *eq;
3343 	int err;
3344 
3345 	err = get_res(dev, slave, res_id, RES_EQ, &eq);
3346 	if (err)
3347 		return err;
3348 
3349 	if (eq->com.from_state != RES_EQ_HW) {
3350 		err = -EINVAL;
3351 		goto ex_put;
3352 	}
3353 
3354 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3355 
3356 ex_put:
3357 	put_res(dev, slave, res_id, RES_EQ);
3358 	return err;
3359 }
3360 
3361 int mlx4_SW2HW_CQ_wrapper(struct mlx4_dev *dev, int slave,
3362 			  struct mlx4_vhcr *vhcr,
3363 			  struct mlx4_cmd_mailbox *inbox,
3364 			  struct mlx4_cmd_mailbox *outbox,
3365 			  struct mlx4_cmd_info *cmd)
3366 {
3367 	int err;
3368 	int cqn = vhcr->in_modifier;
3369 	struct mlx4_cq_context *cqc = inbox->buf;
3370 	int mtt_base = cq_get_mtt_addr(cqc) / dev->caps.mtt_entry_sz;
3371 	struct res_cq *cq = NULL;
3372 	struct res_mtt *mtt;
3373 
3374 	err = cq_res_start_move_to(dev, slave, cqn, RES_CQ_HW, &cq);
3375 	if (err)
3376 		return err;
3377 	err = get_res(dev, slave, mtt_base, RES_MTT, &mtt);
3378 	if (err)
3379 		goto out_move;
3380 	err = check_mtt_range(dev, slave, mtt_base, cq_get_mtt_size(cqc), mtt);
3381 	if (err)
3382 		goto out_put;
3383 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3384 	if (err)
3385 		goto out_put;
3386 	atomic_inc(&mtt->ref_count);
3387 	cq->mtt = mtt;
3388 	put_res(dev, slave, mtt->com.res_id, RES_MTT);
3389 	res_end_move(dev, slave, RES_CQ, cqn);
3390 	return 0;
3391 
3392 out_put:
3393 	put_res(dev, slave, mtt->com.res_id, RES_MTT);
3394 out_move:
3395 	res_abort_move(dev, slave, RES_CQ, cqn);
3396 	return err;
3397 }
3398 
3399 int mlx4_HW2SW_CQ_wrapper(struct mlx4_dev *dev, int slave,
3400 			  struct mlx4_vhcr *vhcr,
3401 			  struct mlx4_cmd_mailbox *inbox,
3402 			  struct mlx4_cmd_mailbox *outbox,
3403 			  struct mlx4_cmd_info *cmd)
3404 {
3405 	int err;
3406 	int cqn = vhcr->in_modifier;
3407 	struct res_cq *cq = NULL;
3408 
3409 	err = cq_res_start_move_to(dev, slave, cqn, RES_CQ_ALLOCATED, &cq);
3410 	if (err)
3411 		return err;
3412 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3413 	if (err)
3414 		goto out_move;
3415 	atomic_dec(&cq->mtt->ref_count);
3416 	res_end_move(dev, slave, RES_CQ, cqn);
3417 	return 0;
3418 
3419 out_move:
3420 	res_abort_move(dev, slave, RES_CQ, cqn);
3421 	return err;
3422 }
3423 
3424 int mlx4_QUERY_CQ_wrapper(struct mlx4_dev *dev, int slave,
3425 			  struct mlx4_vhcr *vhcr,
3426 			  struct mlx4_cmd_mailbox *inbox,
3427 			  struct mlx4_cmd_mailbox *outbox,
3428 			  struct mlx4_cmd_info *cmd)
3429 {
3430 	int cqn = vhcr->in_modifier;
3431 	struct res_cq *cq;
3432 	int err;
3433 
3434 	err = get_res(dev, slave, cqn, RES_CQ, &cq);
3435 	if (err)
3436 		return err;
3437 
3438 	if (cq->com.from_state != RES_CQ_HW)
3439 		goto ex_put;
3440 
3441 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3442 ex_put:
3443 	put_res(dev, slave, cqn, RES_CQ);
3444 
3445 	return err;
3446 }
3447 
3448 static int handle_resize(struct mlx4_dev *dev, int slave,
3449 			 struct mlx4_vhcr *vhcr,
3450 			 struct mlx4_cmd_mailbox *inbox,
3451 			 struct mlx4_cmd_mailbox *outbox,
3452 			 struct mlx4_cmd_info *cmd,
3453 			 struct res_cq *cq)
3454 {
3455 	int err;
3456 	struct res_mtt *orig_mtt;
3457 	struct res_mtt *mtt;
3458 	struct mlx4_cq_context *cqc = inbox->buf;
3459 	int mtt_base = cq_get_mtt_addr(cqc) / dev->caps.mtt_entry_sz;
3460 
3461 	err = get_res(dev, slave, cq->mtt->com.res_id, RES_MTT, &orig_mtt);
3462 	if (err)
3463 		return err;
3464 
3465 	if (orig_mtt != cq->mtt) {
3466 		err = -EINVAL;
3467 		goto ex_put;
3468 	}
3469 
3470 	err = get_res(dev, slave, mtt_base, RES_MTT, &mtt);
3471 	if (err)
3472 		goto ex_put;
3473 
3474 	err = check_mtt_range(dev, slave, mtt_base, cq_get_mtt_size(cqc), mtt);
3475 	if (err)
3476 		goto ex_put1;
3477 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3478 	if (err)
3479 		goto ex_put1;
3480 	atomic_dec(&orig_mtt->ref_count);
3481 	put_res(dev, slave, orig_mtt->com.res_id, RES_MTT);
3482 	atomic_inc(&mtt->ref_count);
3483 	cq->mtt = mtt;
3484 	put_res(dev, slave, mtt->com.res_id, RES_MTT);
3485 	return 0;
3486 
3487 ex_put1:
3488 	put_res(dev, slave, mtt->com.res_id, RES_MTT);
3489 ex_put:
3490 	put_res(dev, slave, orig_mtt->com.res_id, RES_MTT);
3491 
3492 	return err;
3493 
3494 }
3495 
3496 int mlx4_MODIFY_CQ_wrapper(struct mlx4_dev *dev, int slave,
3497 			   struct mlx4_vhcr *vhcr,
3498 			   struct mlx4_cmd_mailbox *inbox,
3499 			   struct mlx4_cmd_mailbox *outbox,
3500 			   struct mlx4_cmd_info *cmd)
3501 {
3502 	int cqn = vhcr->in_modifier;
3503 	struct res_cq *cq;
3504 	int err;
3505 
3506 	err = get_res(dev, slave, cqn, RES_CQ, &cq);
3507 	if (err)
3508 		return err;
3509 
3510 	if (cq->com.from_state != RES_CQ_HW)
3511 		goto ex_put;
3512 
3513 	if (vhcr->op_modifier == 0) {
3514 		err = handle_resize(dev, slave, vhcr, inbox, outbox, cmd, cq);
3515 		goto ex_put;
3516 	}
3517 
3518 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3519 ex_put:
3520 	put_res(dev, slave, cqn, RES_CQ);
3521 
3522 	return err;
3523 }
3524 
3525 static int srq_get_mtt_size(struct mlx4_srq_context *srqc)
3526 {
3527 	int log_srq_size = (be32_to_cpu(srqc->state_logsize_srqn) >> 24) & 0xf;
3528 	int log_rq_stride = srqc->logstride & 7;
3529 	int page_shift = (srqc->log_page_size & 0x3f) + 12;
3530 
3531 	if (log_srq_size + log_rq_stride + 4 < page_shift)
3532 		return 1;
3533 
3534 	return 1 << (log_srq_size + log_rq_stride + 4 - page_shift);
3535 }
3536 
3537 int mlx4_SW2HW_SRQ_wrapper(struct mlx4_dev *dev, int slave,
3538 			   struct mlx4_vhcr *vhcr,
3539 			   struct mlx4_cmd_mailbox *inbox,
3540 			   struct mlx4_cmd_mailbox *outbox,
3541 			   struct mlx4_cmd_info *cmd)
3542 {
3543 	int err;
3544 	int srqn = vhcr->in_modifier;
3545 	struct res_mtt *mtt;
3546 	struct res_srq *srq = NULL;
3547 	struct mlx4_srq_context *srqc = inbox->buf;
3548 	int mtt_base = srq_get_mtt_addr(srqc) / dev->caps.mtt_entry_sz;
3549 
3550 	if (srqn != (be32_to_cpu(srqc->state_logsize_srqn) & 0xffffff))
3551 		return -EINVAL;
3552 
3553 	err = srq_res_start_move_to(dev, slave, srqn, RES_SRQ_HW, &srq);
3554 	if (err)
3555 		return err;
3556 	err = get_res(dev, slave, mtt_base, RES_MTT, &mtt);
3557 	if (err)
3558 		goto ex_abort;
3559 	err = check_mtt_range(dev, slave, mtt_base, srq_get_mtt_size(srqc),
3560 			      mtt);
3561 	if (err)
3562 		goto ex_put_mtt;
3563 
3564 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3565 	if (err)
3566 		goto ex_put_mtt;
3567 
3568 	atomic_inc(&mtt->ref_count);
3569 	srq->mtt = mtt;
3570 	put_res(dev, slave, mtt->com.res_id, RES_MTT);
3571 	res_end_move(dev, slave, RES_SRQ, srqn);
3572 	return 0;
3573 
3574 ex_put_mtt:
3575 	put_res(dev, slave, mtt->com.res_id, RES_MTT);
3576 ex_abort:
3577 	res_abort_move(dev, slave, RES_SRQ, srqn);
3578 
3579 	return err;
3580 }
3581 
3582 int mlx4_HW2SW_SRQ_wrapper(struct mlx4_dev *dev, int slave,
3583 			   struct mlx4_vhcr *vhcr,
3584 			   struct mlx4_cmd_mailbox *inbox,
3585 			   struct mlx4_cmd_mailbox *outbox,
3586 			   struct mlx4_cmd_info *cmd)
3587 {
3588 	int err;
3589 	int srqn = vhcr->in_modifier;
3590 	struct res_srq *srq = NULL;
3591 
3592 	err = srq_res_start_move_to(dev, slave, srqn, RES_SRQ_ALLOCATED, &srq);
3593 	if (err)
3594 		return err;
3595 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3596 	if (err)
3597 		goto ex_abort;
3598 	atomic_dec(&srq->mtt->ref_count);
3599 	if (srq->cq)
3600 		atomic_dec(&srq->cq->ref_count);
3601 	res_end_move(dev, slave, RES_SRQ, srqn);
3602 
3603 	return 0;
3604 
3605 ex_abort:
3606 	res_abort_move(dev, slave, RES_SRQ, srqn);
3607 
3608 	return err;
3609 }
3610 
3611 int mlx4_QUERY_SRQ_wrapper(struct mlx4_dev *dev, int slave,
3612 			   struct mlx4_vhcr *vhcr,
3613 			   struct mlx4_cmd_mailbox *inbox,
3614 			   struct mlx4_cmd_mailbox *outbox,
3615 			   struct mlx4_cmd_info *cmd)
3616 {
3617 	int err;
3618 	int srqn = vhcr->in_modifier;
3619 	struct res_srq *srq;
3620 
3621 	err = get_res(dev, slave, srqn, RES_SRQ, &srq);
3622 	if (err)
3623 		return err;
3624 	if (srq->com.from_state != RES_SRQ_HW) {
3625 		err = -EBUSY;
3626 		goto out;
3627 	}
3628 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3629 out:
3630 	put_res(dev, slave, srqn, RES_SRQ);
3631 	return err;
3632 }
3633 
3634 int mlx4_ARM_SRQ_wrapper(struct mlx4_dev *dev, int slave,
3635 			 struct mlx4_vhcr *vhcr,
3636 			 struct mlx4_cmd_mailbox *inbox,
3637 			 struct mlx4_cmd_mailbox *outbox,
3638 			 struct mlx4_cmd_info *cmd)
3639 {
3640 	int err;
3641 	int srqn = vhcr->in_modifier;
3642 	struct res_srq *srq;
3643 
3644 	err = get_res(dev, slave, srqn, RES_SRQ, &srq);
3645 	if (err)
3646 		return err;
3647 
3648 	if (srq->com.from_state != RES_SRQ_HW) {
3649 		err = -EBUSY;
3650 		goto out;
3651 	}
3652 
3653 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3654 out:
3655 	put_res(dev, slave, srqn, RES_SRQ);
3656 	return err;
3657 }
3658 
3659 int mlx4_GEN_QP_wrapper(struct mlx4_dev *dev, int slave,
3660 			struct mlx4_vhcr *vhcr,
3661 			struct mlx4_cmd_mailbox *inbox,
3662 			struct mlx4_cmd_mailbox *outbox,
3663 			struct mlx4_cmd_info *cmd)
3664 {
3665 	int err;
3666 	int qpn = vhcr->in_modifier & 0x7fffff;
3667 	struct res_qp *qp;
3668 
3669 	err = get_res(dev, slave, qpn, RES_QP, &qp);
3670 	if (err)
3671 		return err;
3672 	if (qp->com.from_state != RES_QP_HW) {
3673 		err = -EBUSY;
3674 		goto out;
3675 	}
3676 
3677 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3678 out:
3679 	put_res(dev, slave, qpn, RES_QP);
3680 	return err;
3681 }
3682 
3683 int mlx4_INIT2INIT_QP_wrapper(struct mlx4_dev *dev, int slave,
3684 			      struct mlx4_vhcr *vhcr,
3685 			      struct mlx4_cmd_mailbox *inbox,
3686 			      struct mlx4_cmd_mailbox *outbox,
3687 			      struct mlx4_cmd_info *cmd)
3688 {
3689 	struct mlx4_qp_context *context = inbox->buf + 8;
3690 	adjust_proxy_tun_qkey(dev, vhcr, context);
3691 	update_pkey_index(dev, slave, inbox);
3692 	return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3693 }
3694 
3695 static int adjust_qp_sched_queue(struct mlx4_dev *dev, int slave,
3696 				  struct mlx4_qp_context *qpc,
3697 				  struct mlx4_cmd_mailbox *inbox)
3698 {
3699 	enum mlx4_qp_optpar optpar = be32_to_cpu(*(__be32 *)inbox->buf);
3700 	u8 pri_sched_queue;
3701 	int port = mlx4_slave_convert_port(
3702 		   dev, slave, (qpc->pri_path.sched_queue >> 6 & 1) + 1) - 1;
3703 
3704 	if (port < 0)
3705 		return -EINVAL;
3706 
3707 	pri_sched_queue = (qpc->pri_path.sched_queue & ~(1 << 6)) |
3708 			  ((port & 1) << 6);
3709 
3710 	if (optpar & (MLX4_QP_OPTPAR_PRIMARY_ADDR_PATH | MLX4_QP_OPTPAR_SCHED_QUEUE) ||
3711 	    qpc->pri_path.sched_queue || mlx4_is_eth(dev, port + 1)) {
3712 		qpc->pri_path.sched_queue = pri_sched_queue;
3713 	}
3714 
3715 	if (optpar & MLX4_QP_OPTPAR_ALT_ADDR_PATH) {
3716 		port = mlx4_slave_convert_port(
3717 				dev, slave, (qpc->alt_path.sched_queue >> 6 & 1)
3718 				+ 1) - 1;
3719 		if (port < 0)
3720 			return -EINVAL;
3721 		qpc->alt_path.sched_queue =
3722 			(qpc->alt_path.sched_queue & ~(1 << 6)) |
3723 			(port & 1) << 6;
3724 	}
3725 	return 0;
3726 }
3727 
3728 static int roce_verify_mac(struct mlx4_dev *dev, int slave,
3729 				struct mlx4_qp_context *qpc,
3730 				struct mlx4_cmd_mailbox *inbox)
3731 {
3732 	u64 mac;
3733 	int port;
3734 	u32 ts = (be32_to_cpu(qpc->flags) >> 16) & 0xff;
3735 	u8 sched = *(u8 *)(inbox->buf + 64);
3736 	u8 smac_ix;
3737 
3738 	port = (sched >> 6 & 1) + 1;
3739 	if (mlx4_is_eth(dev, port) && (ts != MLX4_QP_ST_MLX)) {
3740 		smac_ix = qpc->pri_path.grh_mylmc & 0x7f;
3741 		if (mac_find_smac_ix_in_slave(dev, slave, port, smac_ix, &mac))
3742 			return -ENOENT;
3743 	}
3744 	return 0;
3745 }
3746 
3747 int mlx4_INIT2RTR_QP_wrapper(struct mlx4_dev *dev, int slave,
3748 			     struct mlx4_vhcr *vhcr,
3749 			     struct mlx4_cmd_mailbox *inbox,
3750 			     struct mlx4_cmd_mailbox *outbox,
3751 			     struct mlx4_cmd_info *cmd)
3752 {
3753 	int err;
3754 	struct mlx4_qp_context *qpc = inbox->buf + 8;
3755 	int qpn = vhcr->in_modifier & 0x7fffff;
3756 	struct res_qp *qp;
3757 	u8 orig_sched_queue;
3758 	__be32	orig_param3 = qpc->param3;
3759 	u8 orig_vlan_control = qpc->pri_path.vlan_control;
3760 	u8 orig_fvl_rx = qpc->pri_path.fvl_rx;
3761 	u8 orig_pri_path_fl = qpc->pri_path.fl;
3762 	u8 orig_vlan_index = qpc->pri_path.vlan_index;
3763 	u8 orig_feup = qpc->pri_path.feup;
3764 
3765 	err = adjust_qp_sched_queue(dev, slave, qpc, inbox);
3766 	if (err)
3767 		return err;
3768 	err = verify_qp_parameters(dev, vhcr, inbox, QP_TRANS_INIT2RTR, slave);
3769 	if (err)
3770 		return err;
3771 
3772 	if (roce_verify_mac(dev, slave, qpc, inbox))
3773 		return -EINVAL;
3774 
3775 	update_pkey_index(dev, slave, inbox);
3776 	update_gid(dev, inbox, (u8)slave);
3777 	adjust_proxy_tun_qkey(dev, vhcr, qpc);
3778 	orig_sched_queue = qpc->pri_path.sched_queue;
3779 
3780 	err = get_res(dev, slave, qpn, RES_QP, &qp);
3781 	if (err)
3782 		return err;
3783 	if (qp->com.from_state != RES_QP_HW) {
3784 		err = -EBUSY;
3785 		goto out;
3786 	}
3787 
3788 	err = update_vport_qp_param(dev, inbox, slave, qpn);
3789 	if (err)
3790 		goto out;
3791 
3792 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3793 out:
3794 	/* if no error, save sched queue value passed in by VF. This is
3795 	 * essentially the QOS value provided by the VF. This will be useful
3796 	 * if we allow dynamic changes from VST back to VGT
3797 	 */
3798 	if (!err) {
3799 		qp->sched_queue = orig_sched_queue;
3800 		qp->param3	= orig_param3;
3801 		qp->vlan_control = orig_vlan_control;
3802 		qp->fvl_rx	=  orig_fvl_rx;
3803 		qp->pri_path_fl = orig_pri_path_fl;
3804 		qp->vlan_index  = orig_vlan_index;
3805 		qp->feup	= orig_feup;
3806 	}
3807 	put_res(dev, slave, qpn, RES_QP);
3808 	return err;
3809 }
3810 
3811 int mlx4_RTR2RTS_QP_wrapper(struct mlx4_dev *dev, int slave,
3812 			    struct mlx4_vhcr *vhcr,
3813 			    struct mlx4_cmd_mailbox *inbox,
3814 			    struct mlx4_cmd_mailbox *outbox,
3815 			    struct mlx4_cmd_info *cmd)
3816 {
3817 	int err;
3818 	struct mlx4_qp_context *context = inbox->buf + 8;
3819 
3820 	err = adjust_qp_sched_queue(dev, slave, context, inbox);
3821 	if (err)
3822 		return err;
3823 	err = verify_qp_parameters(dev, vhcr, inbox, QP_TRANS_RTR2RTS, slave);
3824 	if (err)
3825 		return err;
3826 
3827 	update_pkey_index(dev, slave, inbox);
3828 	update_gid(dev, inbox, (u8)slave);
3829 	adjust_proxy_tun_qkey(dev, vhcr, context);
3830 	return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3831 }
3832 
3833 int mlx4_RTS2RTS_QP_wrapper(struct mlx4_dev *dev, int slave,
3834 			    struct mlx4_vhcr *vhcr,
3835 			    struct mlx4_cmd_mailbox *inbox,
3836 			    struct mlx4_cmd_mailbox *outbox,
3837 			    struct mlx4_cmd_info *cmd)
3838 {
3839 	int err;
3840 	struct mlx4_qp_context *context = inbox->buf + 8;
3841 
3842 	err = adjust_qp_sched_queue(dev, slave, context, inbox);
3843 	if (err)
3844 		return err;
3845 	err = verify_qp_parameters(dev, vhcr, inbox, QP_TRANS_RTS2RTS, slave);
3846 	if (err)
3847 		return err;
3848 
3849 	update_pkey_index(dev, slave, inbox);
3850 	update_gid(dev, inbox, (u8)slave);
3851 	adjust_proxy_tun_qkey(dev, vhcr, context);
3852 	return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3853 }
3854 
3855 
3856 int mlx4_SQERR2RTS_QP_wrapper(struct mlx4_dev *dev, int slave,
3857 			      struct mlx4_vhcr *vhcr,
3858 			      struct mlx4_cmd_mailbox *inbox,
3859 			      struct mlx4_cmd_mailbox *outbox,
3860 			      struct mlx4_cmd_info *cmd)
3861 {
3862 	struct mlx4_qp_context *context = inbox->buf + 8;
3863 	int err = adjust_qp_sched_queue(dev, slave, context, inbox);
3864 	if (err)
3865 		return err;
3866 	adjust_proxy_tun_qkey(dev, vhcr, context);
3867 	return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3868 }
3869 
3870 int mlx4_SQD2SQD_QP_wrapper(struct mlx4_dev *dev, int slave,
3871 			    struct mlx4_vhcr *vhcr,
3872 			    struct mlx4_cmd_mailbox *inbox,
3873 			    struct mlx4_cmd_mailbox *outbox,
3874 			    struct mlx4_cmd_info *cmd)
3875 {
3876 	int err;
3877 	struct mlx4_qp_context *context = inbox->buf + 8;
3878 
3879 	err = adjust_qp_sched_queue(dev, slave, context, inbox);
3880 	if (err)
3881 		return err;
3882 	err = verify_qp_parameters(dev, vhcr, inbox, QP_TRANS_SQD2SQD, slave);
3883 	if (err)
3884 		return err;
3885 
3886 	adjust_proxy_tun_qkey(dev, vhcr, context);
3887 	update_gid(dev, inbox, (u8)slave);
3888 	update_pkey_index(dev, slave, inbox);
3889 	return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3890 }
3891 
3892 int mlx4_SQD2RTS_QP_wrapper(struct mlx4_dev *dev, int slave,
3893 			    struct mlx4_vhcr *vhcr,
3894 			    struct mlx4_cmd_mailbox *inbox,
3895 			    struct mlx4_cmd_mailbox *outbox,
3896 			    struct mlx4_cmd_info *cmd)
3897 {
3898 	int err;
3899 	struct mlx4_qp_context *context = inbox->buf + 8;
3900 
3901 	err = adjust_qp_sched_queue(dev, slave, context, inbox);
3902 	if (err)
3903 		return err;
3904 	err = verify_qp_parameters(dev, vhcr, inbox, QP_TRANS_SQD2RTS, slave);
3905 	if (err)
3906 		return err;
3907 
3908 	adjust_proxy_tun_qkey(dev, vhcr, context);
3909 	update_gid(dev, inbox, (u8)slave);
3910 	update_pkey_index(dev, slave, inbox);
3911 	return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3912 }
3913 
3914 int mlx4_2RST_QP_wrapper(struct mlx4_dev *dev, int slave,
3915 			 struct mlx4_vhcr *vhcr,
3916 			 struct mlx4_cmd_mailbox *inbox,
3917 			 struct mlx4_cmd_mailbox *outbox,
3918 			 struct mlx4_cmd_info *cmd)
3919 {
3920 	int err;
3921 	int qpn = vhcr->in_modifier & 0x7fffff;
3922 	struct res_qp *qp;
3923 
3924 	err = qp_res_start_move_to(dev, slave, qpn, RES_QP_MAPPED, &qp, 0);
3925 	if (err)
3926 		return err;
3927 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
3928 	if (err)
3929 		goto ex_abort;
3930 
3931 	atomic_dec(&qp->mtt->ref_count);
3932 	atomic_dec(&qp->rcq->ref_count);
3933 	atomic_dec(&qp->scq->ref_count);
3934 	if (qp->srq)
3935 		atomic_dec(&qp->srq->ref_count);
3936 	res_end_move(dev, slave, RES_QP, qpn);
3937 	return 0;
3938 
3939 ex_abort:
3940 	res_abort_move(dev, slave, RES_QP, qpn);
3941 
3942 	return err;
3943 }
3944 
3945 static struct res_gid *find_gid(struct mlx4_dev *dev, int slave,
3946 				struct res_qp *rqp, u8 *gid)
3947 {
3948 	struct res_gid *res;
3949 
3950 	list_for_each_entry(res, &rqp->mcg_list, list) {
3951 		if (!memcmp(res->gid, gid, 16))
3952 			return res;
3953 	}
3954 	return NULL;
3955 }
3956 
3957 static int add_mcg_res(struct mlx4_dev *dev, int slave, struct res_qp *rqp,
3958 		       u8 *gid, enum mlx4_protocol prot,
3959 		       enum mlx4_steer_type steer, u64 reg_id)
3960 {
3961 	struct res_gid *res;
3962 	int err;
3963 
3964 	res = kzalloc(sizeof *res, GFP_KERNEL);
3965 	if (!res)
3966 		return -ENOMEM;
3967 
3968 	spin_lock_irq(&rqp->mcg_spl);
3969 	if (find_gid(dev, slave, rqp, gid)) {
3970 		kfree(res);
3971 		err = -EEXIST;
3972 	} else {
3973 		memcpy(res->gid, gid, 16);
3974 		res->prot = prot;
3975 		res->steer = steer;
3976 		res->reg_id = reg_id;
3977 		list_add_tail(&res->list, &rqp->mcg_list);
3978 		err = 0;
3979 	}
3980 	spin_unlock_irq(&rqp->mcg_spl);
3981 
3982 	return err;
3983 }
3984 
3985 static int rem_mcg_res(struct mlx4_dev *dev, int slave, struct res_qp *rqp,
3986 		       u8 *gid, enum mlx4_protocol prot,
3987 		       enum mlx4_steer_type steer, u64 *reg_id)
3988 {
3989 	struct res_gid *res;
3990 	int err;
3991 
3992 	spin_lock_irq(&rqp->mcg_spl);
3993 	res = find_gid(dev, slave, rqp, gid);
3994 	if (!res || res->prot != prot || res->steer != steer)
3995 		err = -EINVAL;
3996 	else {
3997 		*reg_id = res->reg_id;
3998 		list_del(&res->list);
3999 		kfree(res);
4000 		err = 0;
4001 	}
4002 	spin_unlock_irq(&rqp->mcg_spl);
4003 
4004 	return err;
4005 }
4006 
4007 static int qp_attach(struct mlx4_dev *dev, int slave, struct mlx4_qp *qp,
4008 		     u8 gid[16], int block_loopback, enum mlx4_protocol prot,
4009 		     enum mlx4_steer_type type, u64 *reg_id)
4010 {
4011 	switch (dev->caps.steering_mode) {
4012 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
4013 		int port = mlx4_slave_convert_port(dev, slave, gid[5]);
4014 		if (port < 0)
4015 			return port;
4016 		return mlx4_trans_to_dmfs_attach(dev, qp, gid, port,
4017 						block_loopback, prot,
4018 						reg_id);
4019 	}
4020 	case MLX4_STEERING_MODE_B0:
4021 		if (prot == MLX4_PROT_ETH) {
4022 			int port = mlx4_slave_convert_port(dev, slave, gid[5]);
4023 			if (port < 0)
4024 				return port;
4025 			gid[5] = port;
4026 		}
4027 		return mlx4_qp_attach_common(dev, qp, gid,
4028 					    block_loopback, prot, type);
4029 	default:
4030 		return -EINVAL;
4031 	}
4032 }
4033 
4034 static int qp_detach(struct mlx4_dev *dev, struct mlx4_qp *qp,
4035 		     u8 gid[16], enum mlx4_protocol prot,
4036 		     enum mlx4_steer_type type, u64 reg_id)
4037 {
4038 	switch (dev->caps.steering_mode) {
4039 	case MLX4_STEERING_MODE_DEVICE_MANAGED:
4040 		return mlx4_flow_detach(dev, reg_id);
4041 	case MLX4_STEERING_MODE_B0:
4042 		return mlx4_qp_detach_common(dev, qp, gid, prot, type);
4043 	default:
4044 		return -EINVAL;
4045 	}
4046 }
4047 
4048 static int mlx4_adjust_port(struct mlx4_dev *dev, int slave,
4049 			    u8 *gid, enum mlx4_protocol prot)
4050 {
4051 	int real_port;
4052 
4053 	if (prot != MLX4_PROT_ETH)
4054 		return 0;
4055 
4056 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_B0 ||
4057 	    dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
4058 		real_port = mlx4_slave_convert_port(dev, slave, gid[5]);
4059 		if (real_port < 0)
4060 			return -EINVAL;
4061 		gid[5] = real_port;
4062 	}
4063 
4064 	return 0;
4065 }
4066 
4067 int mlx4_QP_ATTACH_wrapper(struct mlx4_dev *dev, int slave,
4068 			       struct mlx4_vhcr *vhcr,
4069 			       struct mlx4_cmd_mailbox *inbox,
4070 			       struct mlx4_cmd_mailbox *outbox,
4071 			       struct mlx4_cmd_info *cmd)
4072 {
4073 	struct mlx4_qp qp; /* dummy for calling attach/detach */
4074 	u8 *gid = inbox->buf;
4075 	enum mlx4_protocol prot = (vhcr->in_modifier >> 28) & 0x7;
4076 	int err;
4077 	int qpn;
4078 	struct res_qp *rqp;
4079 	u64 reg_id = 0;
4080 	int attach = vhcr->op_modifier;
4081 	int block_loopback = vhcr->in_modifier >> 31;
4082 	u8 steer_type_mask = 2;
4083 	enum mlx4_steer_type type = (gid[7] & steer_type_mask) >> 1;
4084 
4085 	qpn = vhcr->in_modifier & 0xffffff;
4086 	err = get_res(dev, slave, qpn, RES_QP, &rqp);
4087 	if (err)
4088 		return err;
4089 
4090 	qp.qpn = qpn;
4091 	if (attach) {
4092 		err = qp_attach(dev, slave, &qp, gid, block_loopback, prot,
4093 				type, &reg_id);
4094 		if (err) {
4095 			pr_err("Fail to attach rule to qp 0x%x\n", qpn);
4096 			goto ex_put;
4097 		}
4098 		err = add_mcg_res(dev, slave, rqp, gid, prot, type, reg_id);
4099 		if (err)
4100 			goto ex_detach;
4101 	} else {
4102 		err = mlx4_adjust_port(dev, slave, gid, prot);
4103 		if (err)
4104 			goto ex_put;
4105 
4106 		err = rem_mcg_res(dev, slave, rqp, gid, prot, type, &reg_id);
4107 		if (err)
4108 			goto ex_put;
4109 
4110 		err = qp_detach(dev, &qp, gid, prot, type, reg_id);
4111 		if (err)
4112 			pr_err("Fail to detach rule from qp 0x%x reg_id = 0x%llx\n",
4113 			       qpn, reg_id);
4114 	}
4115 	put_res(dev, slave, qpn, RES_QP);
4116 	return err;
4117 
4118 ex_detach:
4119 	qp_detach(dev, &qp, gid, prot, type, reg_id);
4120 ex_put:
4121 	put_res(dev, slave, qpn, RES_QP);
4122 	return err;
4123 }
4124 
4125 /*
4126  * MAC validation for Flow Steering rules.
4127  * VF can attach rules only with a mac address which is assigned to it.
4128  */
4129 static int validate_eth_header_mac(int slave, struct _rule_hw *eth_header,
4130 				   struct list_head *rlist)
4131 {
4132 	struct mac_res *res, *tmp;
4133 	__be64 be_mac;
4134 
4135 	/* make sure it isn't multicast or broadcast mac*/
4136 	if (!is_multicast_ether_addr(eth_header->eth.dst_mac) &&
4137 	    !is_broadcast_ether_addr(eth_header->eth.dst_mac)) {
4138 		list_for_each_entry_safe(res, tmp, rlist, list) {
4139 			be_mac = cpu_to_be64(res->mac << 16);
4140 			if (ether_addr_equal((u8 *)&be_mac, eth_header->eth.dst_mac))
4141 				return 0;
4142 		}
4143 		pr_err("MAC %pM doesn't belong to VF %d, Steering rule rejected\n",
4144 		       eth_header->eth.dst_mac, slave);
4145 		return -EINVAL;
4146 	}
4147 	return 0;
4148 }
4149 
4150 static void handle_eth_header_mcast_prio(struct mlx4_net_trans_rule_hw_ctrl *ctrl,
4151 					 struct _rule_hw *eth_header)
4152 {
4153 	if (is_multicast_ether_addr(eth_header->eth.dst_mac) ||
4154 	    is_broadcast_ether_addr(eth_header->eth.dst_mac)) {
4155 		struct mlx4_net_trans_rule_hw_eth *eth =
4156 			(struct mlx4_net_trans_rule_hw_eth *)eth_header;
4157 		struct _rule_hw *next_rule = (struct _rule_hw *)(eth + 1);
4158 		bool last_rule = next_rule->size == 0 && next_rule->id == 0 &&
4159 			next_rule->rsvd == 0;
4160 
4161 		if (last_rule)
4162 			ctrl->prio = cpu_to_be16(MLX4_DOMAIN_NIC);
4163 	}
4164 }
4165 
4166 /*
4167  * In case of missing eth header, append eth header with a MAC address
4168  * assigned to the VF.
4169  */
4170 static int add_eth_header(struct mlx4_dev *dev, int slave,
4171 			  struct mlx4_cmd_mailbox *inbox,
4172 			  struct list_head *rlist, int header_id)
4173 {
4174 	struct mac_res *res, *tmp;
4175 	u8 port;
4176 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
4177 	struct mlx4_net_trans_rule_hw_eth *eth_header;
4178 	struct mlx4_net_trans_rule_hw_ipv4 *ip_header;
4179 	struct mlx4_net_trans_rule_hw_tcp_udp *l4_header;
4180 	__be64 be_mac = 0;
4181 	__be64 mac_msk = cpu_to_be64(MLX4_MAC_MASK << 16);
4182 
4183 	ctrl = (struct mlx4_net_trans_rule_hw_ctrl *)inbox->buf;
4184 	port = ctrl->port;
4185 	eth_header = (struct mlx4_net_trans_rule_hw_eth *)(ctrl + 1);
4186 
4187 	/* Clear a space in the inbox for eth header */
4188 	switch (header_id) {
4189 	case MLX4_NET_TRANS_RULE_ID_IPV4:
4190 		ip_header =
4191 			(struct mlx4_net_trans_rule_hw_ipv4 *)(eth_header + 1);
4192 		memmove(ip_header, eth_header,
4193 			sizeof(*ip_header) + sizeof(*l4_header));
4194 		break;
4195 	case MLX4_NET_TRANS_RULE_ID_TCP:
4196 	case MLX4_NET_TRANS_RULE_ID_UDP:
4197 		l4_header = (struct mlx4_net_trans_rule_hw_tcp_udp *)
4198 			    (eth_header + 1);
4199 		memmove(l4_header, eth_header, sizeof(*l4_header));
4200 		break;
4201 	default:
4202 		return -EINVAL;
4203 	}
4204 	list_for_each_entry_safe(res, tmp, rlist, list) {
4205 		if (port == res->port) {
4206 			be_mac = cpu_to_be64(res->mac << 16);
4207 			break;
4208 		}
4209 	}
4210 	if (!be_mac) {
4211 		pr_err("Failed adding eth header to FS rule, Can't find matching MAC for port %d\n",
4212 		       port);
4213 		return -EINVAL;
4214 	}
4215 
4216 	memset(eth_header, 0, sizeof(*eth_header));
4217 	eth_header->size = sizeof(*eth_header) >> 2;
4218 	eth_header->id = cpu_to_be16(__sw_id_hw[MLX4_NET_TRANS_RULE_ID_ETH]);
4219 	memcpy(eth_header->dst_mac, &be_mac, ETH_ALEN);
4220 	memcpy(eth_header->dst_mac_msk, &mac_msk, ETH_ALEN);
4221 
4222 	return 0;
4223 
4224 }
4225 
4226 #define MLX4_UPD_QP_PATH_MASK_SUPPORTED      (                                \
4227 	1ULL << MLX4_UPD_QP_PATH_MASK_MAC_INDEX                     |\
4228 	1ULL << MLX4_UPD_QP_PATH_MASK_ETH_SRC_CHECK_MC_LB)
4229 int mlx4_UPDATE_QP_wrapper(struct mlx4_dev *dev, int slave,
4230 			   struct mlx4_vhcr *vhcr,
4231 			   struct mlx4_cmd_mailbox *inbox,
4232 			   struct mlx4_cmd_mailbox *outbox,
4233 			   struct mlx4_cmd_info *cmd_info)
4234 {
4235 	int err;
4236 	u32 qpn = vhcr->in_modifier & 0xffffff;
4237 	struct res_qp *rqp;
4238 	u64 mac;
4239 	unsigned port;
4240 	u64 pri_addr_path_mask;
4241 	struct mlx4_update_qp_context *cmd;
4242 	int smac_index;
4243 
4244 	cmd = (struct mlx4_update_qp_context *)inbox->buf;
4245 
4246 	pri_addr_path_mask = be64_to_cpu(cmd->primary_addr_path_mask);
4247 	if (cmd->qp_mask || cmd->secondary_addr_path_mask ||
4248 	    (pri_addr_path_mask & ~MLX4_UPD_QP_PATH_MASK_SUPPORTED))
4249 		return -EPERM;
4250 
4251 	if ((pri_addr_path_mask &
4252 	     (1ULL << MLX4_UPD_QP_PATH_MASK_ETH_SRC_CHECK_MC_LB)) &&
4253 		!(dev->caps.flags2 &
4254 		  MLX4_DEV_CAP_FLAG2_UPDATE_QP_SRC_CHECK_LB)) {
4255 		mlx4_warn(dev, "Src check LB for slave %d isn't supported\n",
4256 			  slave);
4257 		return -ENOTSUPP;
4258 	}
4259 
4260 	/* Just change the smac for the QP */
4261 	err = get_res(dev, slave, qpn, RES_QP, &rqp);
4262 	if (err) {
4263 		mlx4_err(dev, "Updating qpn 0x%x for slave %d rejected\n", qpn, slave);
4264 		return err;
4265 	}
4266 
4267 	port = (rqp->sched_queue >> 6 & 1) + 1;
4268 
4269 	if (pri_addr_path_mask & (1ULL << MLX4_UPD_QP_PATH_MASK_MAC_INDEX)) {
4270 		smac_index = cmd->qp_context.pri_path.grh_mylmc;
4271 		err = mac_find_smac_ix_in_slave(dev, slave, port,
4272 						smac_index, &mac);
4273 
4274 		if (err) {
4275 			mlx4_err(dev, "Failed to update qpn 0x%x, MAC is invalid. smac_ix: %d\n",
4276 				 qpn, smac_index);
4277 			goto err_mac;
4278 		}
4279 	}
4280 
4281 	err = mlx4_cmd(dev, inbox->dma,
4282 		       vhcr->in_modifier, 0,
4283 		       MLX4_CMD_UPDATE_QP, MLX4_CMD_TIME_CLASS_A,
4284 		       MLX4_CMD_NATIVE);
4285 	if (err) {
4286 		mlx4_err(dev, "Failed to update qpn on qpn 0x%x, command failed\n", qpn);
4287 		goto err_mac;
4288 	}
4289 
4290 err_mac:
4291 	put_res(dev, slave, qpn, RES_QP);
4292 	return err;
4293 }
4294 
4295 static u32 qp_attach_mbox_size(void *mbox)
4296 {
4297 	u32 size = sizeof(struct mlx4_net_trans_rule_hw_ctrl);
4298 	struct _rule_hw  *rule_header;
4299 
4300 	rule_header = (struct _rule_hw *)(mbox + size);
4301 
4302 	while (rule_header->size) {
4303 		size += rule_header->size * sizeof(u32);
4304 		rule_header += 1;
4305 	}
4306 	return size;
4307 }
4308 
4309 static int mlx4_do_mirror_rule(struct mlx4_dev *dev, struct res_fs_rule *fs_rule);
4310 
4311 int mlx4_QP_FLOW_STEERING_ATTACH_wrapper(struct mlx4_dev *dev, int slave,
4312 					 struct mlx4_vhcr *vhcr,
4313 					 struct mlx4_cmd_mailbox *inbox,
4314 					 struct mlx4_cmd_mailbox *outbox,
4315 					 struct mlx4_cmd_info *cmd)
4316 {
4317 
4318 	struct mlx4_priv *priv = mlx4_priv(dev);
4319 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
4320 	struct list_head *rlist = &tracker->slave_list[slave].res_list[RES_MAC];
4321 	int err;
4322 	int qpn;
4323 	struct res_qp *rqp;
4324 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
4325 	struct _rule_hw  *rule_header;
4326 	int header_id;
4327 	struct res_fs_rule *rrule;
4328 	u32 mbox_size;
4329 
4330 	if (dev->caps.steering_mode !=
4331 	    MLX4_STEERING_MODE_DEVICE_MANAGED)
4332 		return -EOPNOTSUPP;
4333 
4334 	ctrl = (struct mlx4_net_trans_rule_hw_ctrl *)inbox->buf;
4335 	err = mlx4_slave_convert_port(dev, slave, ctrl->port);
4336 	if (err <= 0)
4337 		return -EINVAL;
4338 	ctrl->port = err;
4339 	qpn = be32_to_cpu(ctrl->qpn) & 0xffffff;
4340 	err = get_res(dev, slave, qpn, RES_QP, &rqp);
4341 	if (err) {
4342 		pr_err("Steering rule with qpn 0x%x rejected\n", qpn);
4343 		return err;
4344 	}
4345 	rule_header = (struct _rule_hw *)(ctrl + 1);
4346 	header_id = map_hw_to_sw_id(be16_to_cpu(rule_header->id));
4347 
4348 	if (header_id == MLX4_NET_TRANS_RULE_ID_ETH)
4349 		handle_eth_header_mcast_prio(ctrl, rule_header);
4350 
4351 	if (slave == dev->caps.function)
4352 		goto execute;
4353 
4354 	switch (header_id) {
4355 	case MLX4_NET_TRANS_RULE_ID_ETH:
4356 		if (validate_eth_header_mac(slave, rule_header, rlist)) {
4357 			err = -EINVAL;
4358 			goto err_put_qp;
4359 		}
4360 		break;
4361 	case MLX4_NET_TRANS_RULE_ID_IB:
4362 		break;
4363 	case MLX4_NET_TRANS_RULE_ID_IPV4:
4364 	case MLX4_NET_TRANS_RULE_ID_TCP:
4365 	case MLX4_NET_TRANS_RULE_ID_UDP:
4366 		pr_warn("Can't attach FS rule without L2 headers, adding L2 header\n");
4367 		if (add_eth_header(dev, slave, inbox, rlist, header_id)) {
4368 			err = -EINVAL;
4369 			goto err_put_qp;
4370 		}
4371 		vhcr->in_modifier +=
4372 			sizeof(struct mlx4_net_trans_rule_hw_eth) >> 2;
4373 		break;
4374 	default:
4375 		pr_err("Corrupted mailbox\n");
4376 		err = -EINVAL;
4377 		goto err_put_qp;
4378 	}
4379 
4380 execute:
4381 	err = mlx4_cmd_imm(dev, inbox->dma, &vhcr->out_param,
4382 			   vhcr->in_modifier, 0,
4383 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
4384 			   MLX4_CMD_NATIVE);
4385 	if (err)
4386 		goto err_put_qp;
4387 
4388 
4389 	err = add_res_range(dev, slave, vhcr->out_param, 1, RES_FS_RULE, qpn);
4390 	if (err) {
4391 		mlx4_err(dev, "Fail to add flow steering resources\n");
4392 		goto err_detach;
4393 	}
4394 
4395 	err = get_res(dev, slave, vhcr->out_param, RES_FS_RULE, &rrule);
4396 	if (err)
4397 		goto err_detach;
4398 
4399 	mbox_size = qp_attach_mbox_size(inbox->buf);
4400 	rrule->mirr_mbox = kmalloc(mbox_size, GFP_KERNEL);
4401 	if (!rrule->mirr_mbox) {
4402 		err = -ENOMEM;
4403 		goto err_put_rule;
4404 	}
4405 	rrule->mirr_mbox_size = mbox_size;
4406 	rrule->mirr_rule_id = 0;
4407 	memcpy(rrule->mirr_mbox, inbox->buf, mbox_size);
4408 
4409 	/* set different port */
4410 	ctrl = (struct mlx4_net_trans_rule_hw_ctrl *)rrule->mirr_mbox;
4411 	if (ctrl->port == 1)
4412 		ctrl->port = 2;
4413 	else
4414 		ctrl->port = 1;
4415 
4416 	if (mlx4_is_bonded(dev))
4417 		mlx4_do_mirror_rule(dev, rrule);
4418 
4419 	atomic_inc(&rqp->ref_count);
4420 
4421 err_put_rule:
4422 	put_res(dev, slave, vhcr->out_param, RES_FS_RULE);
4423 err_detach:
4424 	/* detach rule on error */
4425 	if (err)
4426 		mlx4_cmd(dev, vhcr->out_param, 0, 0,
4427 			 MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
4428 			 MLX4_CMD_NATIVE);
4429 err_put_qp:
4430 	put_res(dev, slave, qpn, RES_QP);
4431 	return err;
4432 }
4433 
4434 static int mlx4_undo_mirror_rule(struct mlx4_dev *dev, struct res_fs_rule *fs_rule)
4435 {
4436 	int err;
4437 
4438 	err = rem_res_range(dev, fs_rule->com.owner, fs_rule->com.res_id, 1, RES_FS_RULE, 0);
4439 	if (err) {
4440 		mlx4_err(dev, "Fail to remove flow steering resources\n");
4441 		return err;
4442 	}
4443 
4444 	mlx4_cmd(dev, fs_rule->com.res_id, 0, 0, MLX4_QP_FLOW_STEERING_DETACH,
4445 		 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
4446 	return 0;
4447 }
4448 
4449 int mlx4_QP_FLOW_STEERING_DETACH_wrapper(struct mlx4_dev *dev, int slave,
4450 					 struct mlx4_vhcr *vhcr,
4451 					 struct mlx4_cmd_mailbox *inbox,
4452 					 struct mlx4_cmd_mailbox *outbox,
4453 					 struct mlx4_cmd_info *cmd)
4454 {
4455 	int err;
4456 	struct res_qp *rqp;
4457 	struct res_fs_rule *rrule;
4458 	u64 mirr_reg_id;
4459 
4460 	if (dev->caps.steering_mode !=
4461 	    MLX4_STEERING_MODE_DEVICE_MANAGED)
4462 		return -EOPNOTSUPP;
4463 
4464 	err = get_res(dev, slave, vhcr->in_param, RES_FS_RULE, &rrule);
4465 	if (err)
4466 		return err;
4467 
4468 	if (!rrule->mirr_mbox) {
4469 		mlx4_err(dev, "Mirror rules cannot be removed explicitly\n");
4470 		put_res(dev, slave, vhcr->in_param, RES_FS_RULE);
4471 		return -EINVAL;
4472 	}
4473 	mirr_reg_id = rrule->mirr_rule_id;
4474 	kfree(rrule->mirr_mbox);
4475 
4476 	/* Release the rule form busy state before removal */
4477 	put_res(dev, slave, vhcr->in_param, RES_FS_RULE);
4478 	err = get_res(dev, slave, rrule->qpn, RES_QP, &rqp);
4479 	if (err)
4480 		return err;
4481 
4482 	if (mirr_reg_id && mlx4_is_bonded(dev)) {
4483 		err = get_res(dev, slave, mirr_reg_id, RES_FS_RULE, &rrule);
4484 		if (err) {
4485 			mlx4_err(dev, "Fail to get resource of mirror rule\n");
4486 		} else {
4487 			put_res(dev, slave, mirr_reg_id, RES_FS_RULE);
4488 			mlx4_undo_mirror_rule(dev, rrule);
4489 		}
4490 	}
4491 	err = rem_res_range(dev, slave, vhcr->in_param, 1, RES_FS_RULE, 0);
4492 	if (err) {
4493 		mlx4_err(dev, "Fail to remove flow steering resources\n");
4494 		goto out;
4495 	}
4496 
4497 	err = mlx4_cmd(dev, vhcr->in_param, 0, 0,
4498 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
4499 		       MLX4_CMD_NATIVE);
4500 	if (!err)
4501 		atomic_dec(&rqp->ref_count);
4502 out:
4503 	put_res(dev, slave, rrule->qpn, RES_QP);
4504 	return err;
4505 }
4506 
4507 enum {
4508 	BUSY_MAX_RETRIES = 10
4509 };
4510 
4511 int mlx4_QUERY_IF_STAT_wrapper(struct mlx4_dev *dev, int slave,
4512 			       struct mlx4_vhcr *vhcr,
4513 			       struct mlx4_cmd_mailbox *inbox,
4514 			       struct mlx4_cmd_mailbox *outbox,
4515 			       struct mlx4_cmd_info *cmd)
4516 {
4517 	int err;
4518 	int index = vhcr->in_modifier & 0xffff;
4519 
4520 	err = get_res(dev, slave, index, RES_COUNTER, NULL);
4521 	if (err)
4522 		return err;
4523 
4524 	err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
4525 	put_res(dev, slave, index, RES_COUNTER);
4526 	return err;
4527 }
4528 
4529 static void detach_qp(struct mlx4_dev *dev, int slave, struct res_qp *rqp)
4530 {
4531 	struct res_gid *rgid;
4532 	struct res_gid *tmp;
4533 	struct mlx4_qp qp; /* dummy for calling attach/detach */
4534 
4535 	list_for_each_entry_safe(rgid, tmp, &rqp->mcg_list, list) {
4536 		switch (dev->caps.steering_mode) {
4537 		case MLX4_STEERING_MODE_DEVICE_MANAGED:
4538 			mlx4_flow_detach(dev, rgid->reg_id);
4539 			break;
4540 		case MLX4_STEERING_MODE_B0:
4541 			qp.qpn = rqp->local_qpn;
4542 			(void) mlx4_qp_detach_common(dev, &qp, rgid->gid,
4543 						     rgid->prot, rgid->steer);
4544 			break;
4545 		}
4546 		list_del(&rgid->list);
4547 		kfree(rgid);
4548 	}
4549 }
4550 
4551 static int _move_all_busy(struct mlx4_dev *dev, int slave,
4552 			  enum mlx4_resource type, int print)
4553 {
4554 	struct mlx4_priv *priv = mlx4_priv(dev);
4555 	struct mlx4_resource_tracker *tracker =
4556 		&priv->mfunc.master.res_tracker;
4557 	struct list_head *rlist = &tracker->slave_list[slave].res_list[type];
4558 	struct res_common *r;
4559 	struct res_common *tmp;
4560 	int busy;
4561 
4562 	busy = 0;
4563 	spin_lock_irq(mlx4_tlock(dev));
4564 	list_for_each_entry_safe(r, tmp, rlist, list) {
4565 		if (r->owner == slave) {
4566 			if (!r->removing) {
4567 				if (r->state == RES_ANY_BUSY) {
4568 					if (print)
4569 						mlx4_dbg(dev,
4570 							 "%s id 0x%llx is busy\n",
4571 							  resource_str(type),
4572 							  r->res_id);
4573 					++busy;
4574 				} else {
4575 					r->from_state = r->state;
4576 					r->state = RES_ANY_BUSY;
4577 					r->removing = 1;
4578 				}
4579 			}
4580 		}
4581 	}
4582 	spin_unlock_irq(mlx4_tlock(dev));
4583 
4584 	return busy;
4585 }
4586 
4587 static int move_all_busy(struct mlx4_dev *dev, int slave,
4588 			 enum mlx4_resource type)
4589 {
4590 	unsigned long begin;
4591 	int busy;
4592 
4593 	begin = jiffies;
4594 	do {
4595 		busy = _move_all_busy(dev, slave, type, 0);
4596 		if (time_after(jiffies, begin + 5 * HZ))
4597 			break;
4598 		if (busy)
4599 			cond_resched();
4600 	} while (busy);
4601 
4602 	if (busy)
4603 		busy = _move_all_busy(dev, slave, type, 1);
4604 
4605 	return busy;
4606 }
4607 static void rem_slave_qps(struct mlx4_dev *dev, int slave)
4608 {
4609 	struct mlx4_priv *priv = mlx4_priv(dev);
4610 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
4611 	struct list_head *qp_list =
4612 		&tracker->slave_list[slave].res_list[RES_QP];
4613 	struct res_qp *qp;
4614 	struct res_qp *tmp;
4615 	int state;
4616 	u64 in_param;
4617 	int qpn;
4618 	int err;
4619 
4620 	err = move_all_busy(dev, slave, RES_QP);
4621 	if (err)
4622 		mlx4_warn(dev, "rem_slave_qps: Could not move all qps to busy for slave %d\n",
4623 			  slave);
4624 
4625 	spin_lock_irq(mlx4_tlock(dev));
4626 	list_for_each_entry_safe(qp, tmp, qp_list, com.list) {
4627 		spin_unlock_irq(mlx4_tlock(dev));
4628 		if (qp->com.owner == slave) {
4629 			qpn = qp->com.res_id;
4630 			detach_qp(dev, slave, qp);
4631 			state = qp->com.from_state;
4632 			while (state != 0) {
4633 				switch (state) {
4634 				case RES_QP_RESERVED:
4635 					spin_lock_irq(mlx4_tlock(dev));
4636 					rb_erase(&qp->com.node,
4637 						 &tracker->res_tree[RES_QP]);
4638 					list_del(&qp->com.list);
4639 					spin_unlock_irq(mlx4_tlock(dev));
4640 					if (!valid_reserved(dev, slave, qpn)) {
4641 						__mlx4_qp_release_range(dev, qpn, 1);
4642 						mlx4_release_resource(dev, slave,
4643 								      RES_QP, 1, 0);
4644 					}
4645 					kfree(qp);
4646 					state = 0;
4647 					break;
4648 				case RES_QP_MAPPED:
4649 					if (!valid_reserved(dev, slave, qpn))
4650 						__mlx4_qp_free_icm(dev, qpn);
4651 					state = RES_QP_RESERVED;
4652 					break;
4653 				case RES_QP_HW:
4654 					in_param = slave;
4655 					err = mlx4_cmd(dev, in_param,
4656 						       qp->local_qpn, 2,
4657 						       MLX4_CMD_2RST_QP,
4658 						       MLX4_CMD_TIME_CLASS_A,
4659 						       MLX4_CMD_NATIVE);
4660 					if (err)
4661 						mlx4_dbg(dev, "rem_slave_qps: failed to move slave %d qpn %d to reset\n",
4662 							 slave, qp->local_qpn);
4663 					atomic_dec(&qp->rcq->ref_count);
4664 					atomic_dec(&qp->scq->ref_count);
4665 					atomic_dec(&qp->mtt->ref_count);
4666 					if (qp->srq)
4667 						atomic_dec(&qp->srq->ref_count);
4668 					state = RES_QP_MAPPED;
4669 					break;
4670 				default:
4671 					state = 0;
4672 				}
4673 			}
4674 		}
4675 		spin_lock_irq(mlx4_tlock(dev));
4676 	}
4677 	spin_unlock_irq(mlx4_tlock(dev));
4678 }
4679 
4680 static void rem_slave_srqs(struct mlx4_dev *dev, int slave)
4681 {
4682 	struct mlx4_priv *priv = mlx4_priv(dev);
4683 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
4684 	struct list_head *srq_list =
4685 		&tracker->slave_list[slave].res_list[RES_SRQ];
4686 	struct res_srq *srq;
4687 	struct res_srq *tmp;
4688 	int state;
4689 	u64 in_param;
4690 	LIST_HEAD(tlist);
4691 	int srqn;
4692 	int err;
4693 
4694 	err = move_all_busy(dev, slave, RES_SRQ);
4695 	if (err)
4696 		mlx4_warn(dev, "rem_slave_srqs: Could not move all srqs - too busy for slave %d\n",
4697 			  slave);
4698 
4699 	spin_lock_irq(mlx4_tlock(dev));
4700 	list_for_each_entry_safe(srq, tmp, srq_list, com.list) {
4701 		spin_unlock_irq(mlx4_tlock(dev));
4702 		if (srq->com.owner == slave) {
4703 			srqn = srq->com.res_id;
4704 			state = srq->com.from_state;
4705 			while (state != 0) {
4706 				switch (state) {
4707 				case RES_SRQ_ALLOCATED:
4708 					__mlx4_srq_free_icm(dev, srqn);
4709 					spin_lock_irq(mlx4_tlock(dev));
4710 					rb_erase(&srq->com.node,
4711 						 &tracker->res_tree[RES_SRQ]);
4712 					list_del(&srq->com.list);
4713 					spin_unlock_irq(mlx4_tlock(dev));
4714 					mlx4_release_resource(dev, slave,
4715 							      RES_SRQ, 1, 0);
4716 					kfree(srq);
4717 					state = 0;
4718 					break;
4719 
4720 				case RES_SRQ_HW:
4721 					in_param = slave;
4722 					err = mlx4_cmd(dev, in_param, srqn, 1,
4723 						       MLX4_CMD_HW2SW_SRQ,
4724 						       MLX4_CMD_TIME_CLASS_A,
4725 						       MLX4_CMD_NATIVE);
4726 					if (err)
4727 						mlx4_dbg(dev, "rem_slave_srqs: failed to move slave %d srq %d to SW ownership\n",
4728 							 slave, srqn);
4729 
4730 					atomic_dec(&srq->mtt->ref_count);
4731 					if (srq->cq)
4732 						atomic_dec(&srq->cq->ref_count);
4733 					state = RES_SRQ_ALLOCATED;
4734 					break;
4735 
4736 				default:
4737 					state = 0;
4738 				}
4739 			}
4740 		}
4741 		spin_lock_irq(mlx4_tlock(dev));
4742 	}
4743 	spin_unlock_irq(mlx4_tlock(dev));
4744 }
4745 
4746 static void rem_slave_cqs(struct mlx4_dev *dev, int slave)
4747 {
4748 	struct mlx4_priv *priv = mlx4_priv(dev);
4749 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
4750 	struct list_head *cq_list =
4751 		&tracker->slave_list[slave].res_list[RES_CQ];
4752 	struct res_cq *cq;
4753 	struct res_cq *tmp;
4754 	int state;
4755 	u64 in_param;
4756 	LIST_HEAD(tlist);
4757 	int cqn;
4758 	int err;
4759 
4760 	err = move_all_busy(dev, slave, RES_CQ);
4761 	if (err)
4762 		mlx4_warn(dev, "rem_slave_cqs: Could not move all cqs - too busy for slave %d\n",
4763 			  slave);
4764 
4765 	spin_lock_irq(mlx4_tlock(dev));
4766 	list_for_each_entry_safe(cq, tmp, cq_list, com.list) {
4767 		spin_unlock_irq(mlx4_tlock(dev));
4768 		if (cq->com.owner == slave && !atomic_read(&cq->ref_count)) {
4769 			cqn = cq->com.res_id;
4770 			state = cq->com.from_state;
4771 			while (state != 0) {
4772 				switch (state) {
4773 				case RES_CQ_ALLOCATED:
4774 					__mlx4_cq_free_icm(dev, cqn);
4775 					spin_lock_irq(mlx4_tlock(dev));
4776 					rb_erase(&cq->com.node,
4777 						 &tracker->res_tree[RES_CQ]);
4778 					list_del(&cq->com.list);
4779 					spin_unlock_irq(mlx4_tlock(dev));
4780 					mlx4_release_resource(dev, slave,
4781 							      RES_CQ, 1, 0);
4782 					kfree(cq);
4783 					state = 0;
4784 					break;
4785 
4786 				case RES_CQ_HW:
4787 					in_param = slave;
4788 					err = mlx4_cmd(dev, in_param, cqn, 1,
4789 						       MLX4_CMD_HW2SW_CQ,
4790 						       MLX4_CMD_TIME_CLASS_A,
4791 						       MLX4_CMD_NATIVE);
4792 					if (err)
4793 						mlx4_dbg(dev, "rem_slave_cqs: failed to move slave %d cq %d to SW ownership\n",
4794 							 slave, cqn);
4795 					atomic_dec(&cq->mtt->ref_count);
4796 					state = RES_CQ_ALLOCATED;
4797 					break;
4798 
4799 				default:
4800 					state = 0;
4801 				}
4802 			}
4803 		}
4804 		spin_lock_irq(mlx4_tlock(dev));
4805 	}
4806 	spin_unlock_irq(mlx4_tlock(dev));
4807 }
4808 
4809 static void rem_slave_mrs(struct mlx4_dev *dev, int slave)
4810 {
4811 	struct mlx4_priv *priv = mlx4_priv(dev);
4812 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
4813 	struct list_head *mpt_list =
4814 		&tracker->slave_list[slave].res_list[RES_MPT];
4815 	struct res_mpt *mpt;
4816 	struct res_mpt *tmp;
4817 	int state;
4818 	u64 in_param;
4819 	LIST_HEAD(tlist);
4820 	int mptn;
4821 	int err;
4822 
4823 	err = move_all_busy(dev, slave, RES_MPT);
4824 	if (err)
4825 		mlx4_warn(dev, "rem_slave_mrs: Could not move all mpts - too busy for slave %d\n",
4826 			  slave);
4827 
4828 	spin_lock_irq(mlx4_tlock(dev));
4829 	list_for_each_entry_safe(mpt, tmp, mpt_list, com.list) {
4830 		spin_unlock_irq(mlx4_tlock(dev));
4831 		if (mpt->com.owner == slave) {
4832 			mptn = mpt->com.res_id;
4833 			state = mpt->com.from_state;
4834 			while (state != 0) {
4835 				switch (state) {
4836 				case RES_MPT_RESERVED:
4837 					__mlx4_mpt_release(dev, mpt->key);
4838 					spin_lock_irq(mlx4_tlock(dev));
4839 					rb_erase(&mpt->com.node,
4840 						 &tracker->res_tree[RES_MPT]);
4841 					list_del(&mpt->com.list);
4842 					spin_unlock_irq(mlx4_tlock(dev));
4843 					mlx4_release_resource(dev, slave,
4844 							      RES_MPT, 1, 0);
4845 					kfree(mpt);
4846 					state = 0;
4847 					break;
4848 
4849 				case RES_MPT_MAPPED:
4850 					__mlx4_mpt_free_icm(dev, mpt->key);
4851 					state = RES_MPT_RESERVED;
4852 					break;
4853 
4854 				case RES_MPT_HW:
4855 					in_param = slave;
4856 					err = mlx4_cmd(dev, in_param, mptn, 0,
4857 						     MLX4_CMD_HW2SW_MPT,
4858 						     MLX4_CMD_TIME_CLASS_A,
4859 						     MLX4_CMD_NATIVE);
4860 					if (err)
4861 						mlx4_dbg(dev, "rem_slave_mrs: failed to move slave %d mpt %d to SW ownership\n",
4862 							 slave, mptn);
4863 					if (mpt->mtt)
4864 						atomic_dec(&mpt->mtt->ref_count);
4865 					state = RES_MPT_MAPPED;
4866 					break;
4867 				default:
4868 					state = 0;
4869 				}
4870 			}
4871 		}
4872 		spin_lock_irq(mlx4_tlock(dev));
4873 	}
4874 	spin_unlock_irq(mlx4_tlock(dev));
4875 }
4876 
4877 static void rem_slave_mtts(struct mlx4_dev *dev, int slave)
4878 {
4879 	struct mlx4_priv *priv = mlx4_priv(dev);
4880 	struct mlx4_resource_tracker *tracker =
4881 		&priv->mfunc.master.res_tracker;
4882 	struct list_head *mtt_list =
4883 		&tracker->slave_list[slave].res_list[RES_MTT];
4884 	struct res_mtt *mtt;
4885 	struct res_mtt *tmp;
4886 	int state;
4887 	LIST_HEAD(tlist);
4888 	int base;
4889 	int err;
4890 
4891 	err = move_all_busy(dev, slave, RES_MTT);
4892 	if (err)
4893 		mlx4_warn(dev, "rem_slave_mtts: Could not move all mtts  - too busy for slave %d\n",
4894 			  slave);
4895 
4896 	spin_lock_irq(mlx4_tlock(dev));
4897 	list_for_each_entry_safe(mtt, tmp, mtt_list, com.list) {
4898 		spin_unlock_irq(mlx4_tlock(dev));
4899 		if (mtt->com.owner == slave) {
4900 			base = mtt->com.res_id;
4901 			state = mtt->com.from_state;
4902 			while (state != 0) {
4903 				switch (state) {
4904 				case RES_MTT_ALLOCATED:
4905 					__mlx4_free_mtt_range(dev, base,
4906 							      mtt->order);
4907 					spin_lock_irq(mlx4_tlock(dev));
4908 					rb_erase(&mtt->com.node,
4909 						 &tracker->res_tree[RES_MTT]);
4910 					list_del(&mtt->com.list);
4911 					spin_unlock_irq(mlx4_tlock(dev));
4912 					mlx4_release_resource(dev, slave, RES_MTT,
4913 							      1 << mtt->order, 0);
4914 					kfree(mtt);
4915 					state = 0;
4916 					break;
4917 
4918 				default:
4919 					state = 0;
4920 				}
4921 			}
4922 		}
4923 		spin_lock_irq(mlx4_tlock(dev));
4924 	}
4925 	spin_unlock_irq(mlx4_tlock(dev));
4926 }
4927 
4928 static int mlx4_do_mirror_rule(struct mlx4_dev *dev, struct res_fs_rule *fs_rule)
4929 {
4930 	struct mlx4_cmd_mailbox *mailbox;
4931 	int err;
4932 	struct res_fs_rule *mirr_rule;
4933 	u64 reg_id;
4934 
4935 	mailbox = mlx4_alloc_cmd_mailbox(dev);
4936 	if (IS_ERR(mailbox))
4937 		return PTR_ERR(mailbox);
4938 
4939 	if (!fs_rule->mirr_mbox) {
4940 		mlx4_err(dev, "rule mirroring mailbox is null\n");
4941 		return -EINVAL;
4942 	}
4943 	memcpy(mailbox->buf, fs_rule->mirr_mbox, fs_rule->mirr_mbox_size);
4944 	err = mlx4_cmd_imm(dev, mailbox->dma, &reg_id, fs_rule->mirr_mbox_size >> 2, 0,
4945 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
4946 			   MLX4_CMD_NATIVE);
4947 	mlx4_free_cmd_mailbox(dev, mailbox);
4948 
4949 	if (err)
4950 		goto err;
4951 
4952 	err = add_res_range(dev, fs_rule->com.owner, reg_id, 1, RES_FS_RULE, fs_rule->qpn);
4953 	if (err)
4954 		goto err_detach;
4955 
4956 	err = get_res(dev, fs_rule->com.owner, reg_id, RES_FS_RULE, &mirr_rule);
4957 	if (err)
4958 		goto err_rem;
4959 
4960 	fs_rule->mirr_rule_id = reg_id;
4961 	mirr_rule->mirr_rule_id = 0;
4962 	mirr_rule->mirr_mbox_size = 0;
4963 	mirr_rule->mirr_mbox = NULL;
4964 	put_res(dev, fs_rule->com.owner, reg_id, RES_FS_RULE);
4965 
4966 	return 0;
4967 err_rem:
4968 	rem_res_range(dev, fs_rule->com.owner, reg_id, 1, RES_FS_RULE, 0);
4969 err_detach:
4970 	mlx4_cmd(dev, reg_id, 0, 0, MLX4_QP_FLOW_STEERING_DETACH,
4971 		 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
4972 err:
4973 	return err;
4974 }
4975 
4976 static int mlx4_mirror_fs_rules(struct mlx4_dev *dev, bool bond)
4977 {
4978 	struct mlx4_priv *priv = mlx4_priv(dev);
4979 	struct mlx4_resource_tracker *tracker =
4980 		&priv->mfunc.master.res_tracker;
4981 	struct rb_root *root = &tracker->res_tree[RES_FS_RULE];
4982 	struct rb_node *p;
4983 	struct res_fs_rule *fs_rule;
4984 	int err = 0;
4985 	LIST_HEAD(mirr_list);
4986 
4987 	for (p = rb_first(root); p; p = rb_next(p)) {
4988 		fs_rule = rb_entry(p, struct res_fs_rule, com.node);
4989 		if ((bond && fs_rule->mirr_mbox_size) ||
4990 		    (!bond && !fs_rule->mirr_mbox_size))
4991 			list_add_tail(&fs_rule->mirr_list, &mirr_list);
4992 	}
4993 
4994 	list_for_each_entry(fs_rule, &mirr_list, mirr_list) {
4995 		if (bond)
4996 			err += mlx4_do_mirror_rule(dev, fs_rule);
4997 		else
4998 			err += mlx4_undo_mirror_rule(dev, fs_rule);
4999 	}
5000 	return err;
5001 }
5002 
5003 int mlx4_bond_fs_rules(struct mlx4_dev *dev)
5004 {
5005 	return mlx4_mirror_fs_rules(dev, true);
5006 }
5007 
5008 int mlx4_unbond_fs_rules(struct mlx4_dev *dev)
5009 {
5010 	return mlx4_mirror_fs_rules(dev, false);
5011 }
5012 
5013 static void rem_slave_fs_rule(struct mlx4_dev *dev, int slave)
5014 {
5015 	struct mlx4_priv *priv = mlx4_priv(dev);
5016 	struct mlx4_resource_tracker *tracker =
5017 		&priv->mfunc.master.res_tracker;
5018 	struct list_head *fs_rule_list =
5019 		&tracker->slave_list[slave].res_list[RES_FS_RULE];
5020 	struct res_fs_rule *fs_rule;
5021 	struct res_fs_rule *tmp;
5022 	int state;
5023 	u64 base;
5024 	int err;
5025 
5026 	err = move_all_busy(dev, slave, RES_FS_RULE);
5027 	if (err)
5028 		mlx4_warn(dev, "rem_slave_fs_rule: Could not move all mtts to busy for slave %d\n",
5029 			  slave);
5030 
5031 	spin_lock_irq(mlx4_tlock(dev));
5032 	list_for_each_entry_safe(fs_rule, tmp, fs_rule_list, com.list) {
5033 		spin_unlock_irq(mlx4_tlock(dev));
5034 		if (fs_rule->com.owner == slave) {
5035 			base = fs_rule->com.res_id;
5036 			state = fs_rule->com.from_state;
5037 			while (state != 0) {
5038 				switch (state) {
5039 				case RES_FS_RULE_ALLOCATED:
5040 					/* detach rule */
5041 					err = mlx4_cmd(dev, base, 0, 0,
5042 						       MLX4_QP_FLOW_STEERING_DETACH,
5043 						       MLX4_CMD_TIME_CLASS_A,
5044 						       MLX4_CMD_NATIVE);
5045 
5046 					spin_lock_irq(mlx4_tlock(dev));
5047 					rb_erase(&fs_rule->com.node,
5048 						 &tracker->res_tree[RES_FS_RULE]);
5049 					list_del(&fs_rule->com.list);
5050 					spin_unlock_irq(mlx4_tlock(dev));
5051 					kfree(fs_rule);
5052 					state = 0;
5053 					break;
5054 
5055 				default:
5056 					state = 0;
5057 				}
5058 			}
5059 		}
5060 		spin_lock_irq(mlx4_tlock(dev));
5061 	}
5062 	spin_unlock_irq(mlx4_tlock(dev));
5063 }
5064 
5065 static void rem_slave_eqs(struct mlx4_dev *dev, int slave)
5066 {
5067 	struct mlx4_priv *priv = mlx4_priv(dev);
5068 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
5069 	struct list_head *eq_list =
5070 		&tracker->slave_list[slave].res_list[RES_EQ];
5071 	struct res_eq *eq;
5072 	struct res_eq *tmp;
5073 	int err;
5074 	int state;
5075 	LIST_HEAD(tlist);
5076 	int eqn;
5077 
5078 	err = move_all_busy(dev, slave, RES_EQ);
5079 	if (err)
5080 		mlx4_warn(dev, "rem_slave_eqs: Could not move all eqs - too busy for slave %d\n",
5081 			  slave);
5082 
5083 	spin_lock_irq(mlx4_tlock(dev));
5084 	list_for_each_entry_safe(eq, tmp, eq_list, com.list) {
5085 		spin_unlock_irq(mlx4_tlock(dev));
5086 		if (eq->com.owner == slave) {
5087 			eqn = eq->com.res_id;
5088 			state = eq->com.from_state;
5089 			while (state != 0) {
5090 				switch (state) {
5091 				case RES_EQ_RESERVED:
5092 					spin_lock_irq(mlx4_tlock(dev));
5093 					rb_erase(&eq->com.node,
5094 						 &tracker->res_tree[RES_EQ]);
5095 					list_del(&eq->com.list);
5096 					spin_unlock_irq(mlx4_tlock(dev));
5097 					kfree(eq);
5098 					state = 0;
5099 					break;
5100 
5101 				case RES_EQ_HW:
5102 					err = mlx4_cmd(dev, slave, eqn & 0x3ff,
5103 						       1, MLX4_CMD_HW2SW_EQ,
5104 						       MLX4_CMD_TIME_CLASS_A,
5105 						       MLX4_CMD_NATIVE);
5106 					if (err)
5107 						mlx4_dbg(dev, "rem_slave_eqs: failed to move slave %d eqs %d to SW ownership\n",
5108 							 slave, eqn & 0x3ff);
5109 					atomic_dec(&eq->mtt->ref_count);
5110 					state = RES_EQ_RESERVED;
5111 					break;
5112 
5113 				default:
5114 					state = 0;
5115 				}
5116 			}
5117 		}
5118 		spin_lock_irq(mlx4_tlock(dev));
5119 	}
5120 	spin_unlock_irq(mlx4_tlock(dev));
5121 }
5122 
5123 static void rem_slave_counters(struct mlx4_dev *dev, int slave)
5124 {
5125 	struct mlx4_priv *priv = mlx4_priv(dev);
5126 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
5127 	struct list_head *counter_list =
5128 		&tracker->slave_list[slave].res_list[RES_COUNTER];
5129 	struct res_counter *counter;
5130 	struct res_counter *tmp;
5131 	int err;
5132 	int *counters_arr = NULL;
5133 	int i, j;
5134 
5135 	err = move_all_busy(dev, slave, RES_COUNTER);
5136 	if (err)
5137 		mlx4_warn(dev, "rem_slave_counters: Could not move all counters - too busy for slave %d\n",
5138 			  slave);
5139 
5140 	counters_arr = kmalloc_array(dev->caps.max_counters,
5141 				     sizeof(*counters_arr), GFP_KERNEL);
5142 	if (!counters_arr)
5143 		return;
5144 
5145 	do {
5146 		i = 0;
5147 		j = 0;
5148 		spin_lock_irq(mlx4_tlock(dev));
5149 		list_for_each_entry_safe(counter, tmp, counter_list, com.list) {
5150 			if (counter->com.owner == slave) {
5151 				counters_arr[i++] = counter->com.res_id;
5152 				rb_erase(&counter->com.node,
5153 					 &tracker->res_tree[RES_COUNTER]);
5154 				list_del(&counter->com.list);
5155 				kfree(counter);
5156 			}
5157 		}
5158 		spin_unlock_irq(mlx4_tlock(dev));
5159 
5160 		while (j < i) {
5161 			__mlx4_counter_free(dev, counters_arr[j++]);
5162 			mlx4_release_resource(dev, slave, RES_COUNTER, 1, 0);
5163 		}
5164 	} while (i);
5165 
5166 	kfree(counters_arr);
5167 }
5168 
5169 static void rem_slave_xrcdns(struct mlx4_dev *dev, int slave)
5170 {
5171 	struct mlx4_priv *priv = mlx4_priv(dev);
5172 	struct mlx4_resource_tracker *tracker = &priv->mfunc.master.res_tracker;
5173 	struct list_head *xrcdn_list =
5174 		&tracker->slave_list[slave].res_list[RES_XRCD];
5175 	struct res_xrcdn *xrcd;
5176 	struct res_xrcdn *tmp;
5177 	int err;
5178 	int xrcdn;
5179 
5180 	err = move_all_busy(dev, slave, RES_XRCD);
5181 	if (err)
5182 		mlx4_warn(dev, "rem_slave_xrcdns: Could not move all xrcdns - too busy for slave %d\n",
5183 			  slave);
5184 
5185 	spin_lock_irq(mlx4_tlock(dev));
5186 	list_for_each_entry_safe(xrcd, tmp, xrcdn_list, com.list) {
5187 		if (xrcd->com.owner == slave) {
5188 			xrcdn = xrcd->com.res_id;
5189 			rb_erase(&xrcd->com.node, &tracker->res_tree[RES_XRCD]);
5190 			list_del(&xrcd->com.list);
5191 			kfree(xrcd);
5192 			__mlx4_xrcd_free(dev, xrcdn);
5193 		}
5194 	}
5195 	spin_unlock_irq(mlx4_tlock(dev));
5196 }
5197 
5198 void mlx4_delete_all_resources_for_slave(struct mlx4_dev *dev, int slave)
5199 {
5200 	struct mlx4_priv *priv = mlx4_priv(dev);
5201 	mlx4_reset_roce_gids(dev, slave);
5202 	mutex_lock(&priv->mfunc.master.res_tracker.slave_list[slave].mutex);
5203 	rem_slave_vlans(dev, slave);
5204 	rem_slave_macs(dev, slave);
5205 	rem_slave_fs_rule(dev, slave);
5206 	rem_slave_qps(dev, slave);
5207 	rem_slave_srqs(dev, slave);
5208 	rem_slave_cqs(dev, slave);
5209 	rem_slave_mrs(dev, slave);
5210 	rem_slave_eqs(dev, slave);
5211 	rem_slave_mtts(dev, slave);
5212 	rem_slave_counters(dev, slave);
5213 	rem_slave_xrcdns(dev, slave);
5214 	mutex_unlock(&priv->mfunc.master.res_tracker.slave_list[slave].mutex);
5215 }
5216 
5217 void mlx4_vf_immed_vlan_work_handler(struct work_struct *_work)
5218 {
5219 	struct mlx4_vf_immed_vlan_work *work =
5220 		container_of(_work, struct mlx4_vf_immed_vlan_work, work);
5221 	struct mlx4_cmd_mailbox *mailbox;
5222 	struct mlx4_update_qp_context *upd_context;
5223 	struct mlx4_dev *dev = &work->priv->dev;
5224 	struct mlx4_resource_tracker *tracker =
5225 		&work->priv->mfunc.master.res_tracker;
5226 	struct list_head *qp_list =
5227 		&tracker->slave_list[work->slave].res_list[RES_QP];
5228 	struct res_qp *qp;
5229 	struct res_qp *tmp;
5230 	u64 qp_path_mask_vlan_ctrl =
5231 		       ((1ULL << MLX4_UPD_QP_PATH_MASK_ETH_TX_BLOCK_UNTAGGED) |
5232 		       (1ULL << MLX4_UPD_QP_PATH_MASK_ETH_TX_BLOCK_1P) |
5233 		       (1ULL << MLX4_UPD_QP_PATH_MASK_ETH_TX_BLOCK_TAGGED) |
5234 		       (1ULL << MLX4_UPD_QP_PATH_MASK_ETH_RX_BLOCK_UNTAGGED) |
5235 		       (1ULL << MLX4_UPD_QP_PATH_MASK_ETH_RX_BLOCK_1P) |
5236 		       (1ULL << MLX4_UPD_QP_PATH_MASK_ETH_RX_BLOCK_TAGGED));
5237 
5238 	u64 qp_path_mask = ((1ULL << MLX4_UPD_QP_PATH_MASK_VLAN_INDEX) |
5239 		       (1ULL << MLX4_UPD_QP_PATH_MASK_FVL) |
5240 		       (1ULL << MLX4_UPD_QP_PATH_MASK_CV) |
5241 		       (1ULL << MLX4_UPD_QP_PATH_MASK_ETH_HIDE_CQE_VLAN) |
5242 		       (1ULL << MLX4_UPD_QP_PATH_MASK_FEUP) |
5243 		       (1ULL << MLX4_UPD_QP_PATH_MASK_FVL_RX) |
5244 		       (1ULL << MLX4_UPD_QP_PATH_MASK_SCHED_QUEUE));
5245 
5246 	int err;
5247 	int port, errors = 0;
5248 	u8 vlan_control;
5249 
5250 	if (mlx4_is_slave(dev)) {
5251 		mlx4_warn(dev, "Trying to update-qp in slave %d\n",
5252 			  work->slave);
5253 		goto out;
5254 	}
5255 
5256 	mailbox = mlx4_alloc_cmd_mailbox(dev);
5257 	if (IS_ERR(mailbox))
5258 		goto out;
5259 	if (work->flags & MLX4_VF_IMMED_VLAN_FLAG_LINK_DISABLE) /* block all */
5260 		vlan_control = MLX4_VLAN_CTRL_ETH_TX_BLOCK_TAGGED |
5261 			MLX4_VLAN_CTRL_ETH_TX_BLOCK_PRIO_TAGGED |
5262 			MLX4_VLAN_CTRL_ETH_TX_BLOCK_UNTAGGED |
5263 			MLX4_VLAN_CTRL_ETH_RX_BLOCK_PRIO_TAGGED |
5264 			MLX4_VLAN_CTRL_ETH_RX_BLOCK_UNTAGGED |
5265 			MLX4_VLAN_CTRL_ETH_RX_BLOCK_TAGGED;
5266 	else if (!work->vlan_id)
5267 		vlan_control = MLX4_VLAN_CTRL_ETH_TX_BLOCK_TAGGED |
5268 			MLX4_VLAN_CTRL_ETH_RX_BLOCK_TAGGED;
5269 	else
5270 		vlan_control = MLX4_VLAN_CTRL_ETH_TX_BLOCK_TAGGED |
5271 			MLX4_VLAN_CTRL_ETH_RX_BLOCK_PRIO_TAGGED |
5272 			MLX4_VLAN_CTRL_ETH_RX_BLOCK_UNTAGGED;
5273 
5274 	upd_context = mailbox->buf;
5275 	upd_context->qp_mask = cpu_to_be64(1ULL << MLX4_UPD_QP_MASK_VSD);
5276 
5277 	spin_lock_irq(mlx4_tlock(dev));
5278 	list_for_each_entry_safe(qp, tmp, qp_list, com.list) {
5279 		spin_unlock_irq(mlx4_tlock(dev));
5280 		if (qp->com.owner == work->slave) {
5281 			if (qp->com.from_state != RES_QP_HW ||
5282 			    !qp->sched_queue ||  /* no INIT2RTR trans yet */
5283 			    mlx4_is_qp_reserved(dev, qp->local_qpn) ||
5284 			    qp->qpc_flags & (1 << MLX4_RSS_QPC_FLAG_OFFSET)) {
5285 				spin_lock_irq(mlx4_tlock(dev));
5286 				continue;
5287 			}
5288 			port = (qp->sched_queue >> 6 & 1) + 1;
5289 			if (port != work->port) {
5290 				spin_lock_irq(mlx4_tlock(dev));
5291 				continue;
5292 			}
5293 			if (MLX4_QP_ST_RC == ((qp->qpc_flags >> 16) & 0xff))
5294 				upd_context->primary_addr_path_mask = cpu_to_be64(qp_path_mask);
5295 			else
5296 				upd_context->primary_addr_path_mask =
5297 					cpu_to_be64(qp_path_mask | qp_path_mask_vlan_ctrl);
5298 			if (work->vlan_id == MLX4_VGT) {
5299 				upd_context->qp_context.param3 = qp->param3;
5300 				upd_context->qp_context.pri_path.vlan_control = qp->vlan_control;
5301 				upd_context->qp_context.pri_path.fvl_rx = qp->fvl_rx;
5302 				upd_context->qp_context.pri_path.vlan_index = qp->vlan_index;
5303 				upd_context->qp_context.pri_path.fl = qp->pri_path_fl;
5304 				upd_context->qp_context.pri_path.feup = qp->feup;
5305 				upd_context->qp_context.pri_path.sched_queue =
5306 					qp->sched_queue;
5307 			} else {
5308 				upd_context->qp_context.param3 = qp->param3 & ~cpu_to_be32(MLX4_STRIP_VLAN);
5309 				upd_context->qp_context.pri_path.vlan_control = vlan_control;
5310 				upd_context->qp_context.pri_path.vlan_index = work->vlan_ix;
5311 				upd_context->qp_context.pri_path.fvl_rx =
5312 					qp->fvl_rx | MLX4_FVL_RX_FORCE_ETH_VLAN;
5313 				upd_context->qp_context.pri_path.fl =
5314 					qp->pri_path_fl | MLX4_FL_CV | MLX4_FL_ETH_HIDE_CQE_VLAN;
5315 				upd_context->qp_context.pri_path.feup =
5316 					qp->feup | MLX4_FEUP_FORCE_ETH_UP | MLX4_FVL_FORCE_ETH_VLAN;
5317 				upd_context->qp_context.pri_path.sched_queue =
5318 					qp->sched_queue & 0xC7;
5319 				upd_context->qp_context.pri_path.sched_queue |=
5320 					((work->qos & 0x7) << 3);
5321 				upd_context->qp_mask |=
5322 					cpu_to_be64(1ULL <<
5323 						    MLX4_UPD_QP_MASK_QOS_VPP);
5324 				upd_context->qp_context.qos_vport =
5325 					work->qos_vport;
5326 			}
5327 
5328 			err = mlx4_cmd(dev, mailbox->dma,
5329 				       qp->local_qpn & 0xffffff,
5330 				       0, MLX4_CMD_UPDATE_QP,
5331 				       MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
5332 			if (err) {
5333 				mlx4_info(dev, "UPDATE_QP failed for slave %d, port %d, qpn %d (%d)\n",
5334 					  work->slave, port, qp->local_qpn, err);
5335 				errors++;
5336 			}
5337 		}
5338 		spin_lock_irq(mlx4_tlock(dev));
5339 	}
5340 	spin_unlock_irq(mlx4_tlock(dev));
5341 	mlx4_free_cmd_mailbox(dev, mailbox);
5342 
5343 	if (errors)
5344 		mlx4_err(dev, "%d UPDATE_QP failures for slave %d, port %d\n",
5345 			 errors, work->slave, work->port);
5346 
5347 	/* unregister previous vlan_id if needed and we had no errors
5348 	 * while updating the QPs
5349 	 */
5350 	if (work->flags & MLX4_VF_IMMED_VLAN_FLAG_VLAN && !errors &&
5351 	    NO_INDX != work->orig_vlan_ix)
5352 		__mlx4_unregister_vlan(&work->priv->dev, work->port,
5353 				       work->orig_vlan_id);
5354 out:
5355 	kfree(work);
5356 	return;
5357 }
5358