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