xref: /linux/drivers/net/ethernet/mellanox/mlx4/mcg.c (revision b85d45947951d23cb22d90caecf4c1eb81342c96)
1 /*
2  * Copyright (c) 2006, 2007 Cisco Systems, Inc.  All rights reserved.
3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #include <linux/string.h>
35 #include <linux/etherdevice.h>
36 
37 #include <linux/mlx4/cmd.h>
38 #include <linux/export.h>
39 
40 #include "mlx4.h"
41 
42 static const u8 zero_gid[16];	/* automatically initialized to 0 */
43 
44 int mlx4_get_mgm_entry_size(struct mlx4_dev *dev)
45 {
46 	return 1 << dev->oper_log_mgm_entry_size;
47 }
48 
49 int mlx4_get_qp_per_mgm(struct mlx4_dev *dev)
50 {
51 	return 4 * (mlx4_get_mgm_entry_size(dev) / 16 - 2);
52 }
53 
54 static int mlx4_QP_FLOW_STEERING_ATTACH(struct mlx4_dev *dev,
55 					struct mlx4_cmd_mailbox *mailbox,
56 					u32 size,
57 					u64 *reg_id)
58 {
59 	u64 imm;
60 	int err = 0;
61 
62 	err = mlx4_cmd_imm(dev, mailbox->dma, &imm, size, 0,
63 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
64 			   MLX4_CMD_NATIVE);
65 	if (err)
66 		return err;
67 	*reg_id = imm;
68 
69 	return err;
70 }
71 
72 static int mlx4_QP_FLOW_STEERING_DETACH(struct mlx4_dev *dev, u64 regid)
73 {
74 	int err = 0;
75 
76 	err = mlx4_cmd(dev, regid, 0, 0,
77 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
78 		       MLX4_CMD_NATIVE);
79 
80 	return err;
81 }
82 
83 static int mlx4_READ_ENTRY(struct mlx4_dev *dev, int index,
84 			   struct mlx4_cmd_mailbox *mailbox)
85 {
86 	return mlx4_cmd_box(dev, 0, mailbox->dma, index, 0, MLX4_CMD_READ_MCG,
87 			    MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
88 }
89 
90 static int mlx4_WRITE_ENTRY(struct mlx4_dev *dev, int index,
91 			    struct mlx4_cmd_mailbox *mailbox)
92 {
93 	return mlx4_cmd(dev, mailbox->dma, index, 0, MLX4_CMD_WRITE_MCG,
94 			MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
95 }
96 
97 static int mlx4_WRITE_PROMISC(struct mlx4_dev *dev, u8 port, u8 steer,
98 			      struct mlx4_cmd_mailbox *mailbox)
99 {
100 	u32 in_mod;
101 
102 	in_mod = (u32) port << 16 | steer << 1;
103 	return mlx4_cmd(dev, mailbox->dma, in_mod, 0x1,
104 			MLX4_CMD_WRITE_MCG, MLX4_CMD_TIME_CLASS_A,
105 			MLX4_CMD_NATIVE);
106 }
107 
108 static int mlx4_GID_HASH(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
109 			 u16 *hash, u8 op_mod)
110 {
111 	u64 imm;
112 	int err;
113 
114 	err = mlx4_cmd_imm(dev, mailbox->dma, &imm, 0, op_mod,
115 			   MLX4_CMD_MGID_HASH, MLX4_CMD_TIME_CLASS_A,
116 			   MLX4_CMD_NATIVE);
117 
118 	if (!err)
119 		*hash = imm;
120 
121 	return err;
122 }
123 
124 static struct mlx4_promisc_qp *get_promisc_qp(struct mlx4_dev *dev, u8 port,
125 					      enum mlx4_steer_type steer,
126 					      u32 qpn)
127 {
128 	struct mlx4_steer *s_steer;
129 	struct mlx4_promisc_qp *pqp;
130 
131 	if (port < 1 || port > dev->caps.num_ports)
132 		return NULL;
133 
134 	s_steer = &mlx4_priv(dev)->steer[port - 1];
135 
136 	list_for_each_entry(pqp, &s_steer->promisc_qps[steer], list) {
137 		if (pqp->qpn == qpn)
138 			return pqp;
139 	}
140 	/* not found */
141 	return NULL;
142 }
143 
144 /*
145  * Add new entry to steering data structure.
146  * All promisc QPs should be added as well
147  */
148 static int new_steering_entry(struct mlx4_dev *dev, u8 port,
149 			      enum mlx4_steer_type steer,
150 			      unsigned int index, u32 qpn)
151 {
152 	struct mlx4_steer *s_steer;
153 	struct mlx4_cmd_mailbox *mailbox;
154 	struct mlx4_mgm *mgm;
155 	u32 members_count;
156 	struct mlx4_steer_index *new_entry;
157 	struct mlx4_promisc_qp *pqp;
158 	struct mlx4_promisc_qp *dqp = NULL;
159 	u32 prot;
160 	int err;
161 
162 	if (port < 1 || port > dev->caps.num_ports)
163 		return -EINVAL;
164 
165 	s_steer = &mlx4_priv(dev)->steer[port - 1];
166 	new_entry = kzalloc(sizeof *new_entry, GFP_KERNEL);
167 	if (!new_entry)
168 		return -ENOMEM;
169 
170 	INIT_LIST_HEAD(&new_entry->duplicates);
171 	new_entry->index = index;
172 	list_add_tail(&new_entry->list, &s_steer->steer_entries[steer]);
173 
174 	/* If the given qpn is also a promisc qp,
175 	 * it should be inserted to duplicates list
176 	 */
177 	pqp = get_promisc_qp(dev, port, steer, qpn);
178 	if (pqp) {
179 		dqp = kmalloc(sizeof *dqp, GFP_KERNEL);
180 		if (!dqp) {
181 			err = -ENOMEM;
182 			goto out_alloc;
183 		}
184 		dqp->qpn = qpn;
185 		list_add_tail(&dqp->list, &new_entry->duplicates);
186 	}
187 
188 	/* if no promisc qps for this vep, we are done */
189 	if (list_empty(&s_steer->promisc_qps[steer]))
190 		return 0;
191 
192 	/* now need to add all the promisc qps to the new
193 	 * steering entry, as they should also receive the packets
194 	 * destined to this address */
195 	mailbox = mlx4_alloc_cmd_mailbox(dev);
196 	if (IS_ERR(mailbox)) {
197 		err = -ENOMEM;
198 		goto out_alloc;
199 	}
200 	mgm = mailbox->buf;
201 
202 	err = mlx4_READ_ENTRY(dev, index, mailbox);
203 	if (err)
204 		goto out_mailbox;
205 
206 	members_count = be32_to_cpu(mgm->members_count) & 0xffffff;
207 	prot = be32_to_cpu(mgm->members_count) >> 30;
208 	list_for_each_entry(pqp, &s_steer->promisc_qps[steer], list) {
209 		/* don't add already existing qpn */
210 		if (pqp->qpn == qpn)
211 			continue;
212 		if (members_count == dev->caps.num_qp_per_mgm) {
213 			/* out of space */
214 			err = -ENOMEM;
215 			goto out_mailbox;
216 		}
217 
218 		/* add the qpn */
219 		mgm->qp[members_count++] = cpu_to_be32(pqp->qpn & MGM_QPN_MASK);
220 	}
221 	/* update the qps count and update the entry with all the promisc qps*/
222 	mgm->members_count = cpu_to_be32(members_count | (prot << 30));
223 	err = mlx4_WRITE_ENTRY(dev, index, mailbox);
224 
225 out_mailbox:
226 	mlx4_free_cmd_mailbox(dev, mailbox);
227 	if (!err)
228 		return 0;
229 out_alloc:
230 	if (dqp) {
231 		list_del(&dqp->list);
232 		kfree(dqp);
233 	}
234 	list_del(&new_entry->list);
235 	kfree(new_entry);
236 	return err;
237 }
238 
239 /* update the data structures with existing steering entry */
240 static int existing_steering_entry(struct mlx4_dev *dev, u8 port,
241 				   enum mlx4_steer_type steer,
242 				   unsigned int index, u32 qpn)
243 {
244 	struct mlx4_steer *s_steer;
245 	struct mlx4_steer_index *tmp_entry, *entry = NULL;
246 	struct mlx4_promisc_qp *pqp;
247 	struct mlx4_promisc_qp *dqp;
248 
249 	if (port < 1 || port > dev->caps.num_ports)
250 		return -EINVAL;
251 
252 	s_steer = &mlx4_priv(dev)->steer[port - 1];
253 
254 	pqp = get_promisc_qp(dev, port, steer, qpn);
255 	if (!pqp)
256 		return 0; /* nothing to do */
257 
258 	list_for_each_entry(tmp_entry, &s_steer->steer_entries[steer], list) {
259 		if (tmp_entry->index == index) {
260 			entry = tmp_entry;
261 			break;
262 		}
263 	}
264 	if (unlikely(!entry)) {
265 		mlx4_warn(dev, "Steering entry at index %x is not registered\n", index);
266 		return -EINVAL;
267 	}
268 
269 	/* the given qpn is listed as a promisc qpn
270 	 * we need to add it as a duplicate to this entry
271 	 * for future references */
272 	list_for_each_entry(dqp, &entry->duplicates, list) {
273 		if (qpn == dqp->qpn)
274 			return 0; /* qp is already duplicated */
275 	}
276 
277 	/* add the qp as a duplicate on this index */
278 	dqp = kmalloc(sizeof *dqp, GFP_KERNEL);
279 	if (!dqp)
280 		return -ENOMEM;
281 	dqp->qpn = qpn;
282 	list_add_tail(&dqp->list, &entry->duplicates);
283 
284 	return 0;
285 }
286 
287 /* Check whether a qpn is a duplicate on steering entry
288  * If so, it should not be removed from mgm */
289 static bool check_duplicate_entry(struct mlx4_dev *dev, u8 port,
290 				  enum mlx4_steer_type steer,
291 				  unsigned int index, u32 qpn)
292 {
293 	struct mlx4_steer *s_steer;
294 	struct mlx4_steer_index *tmp_entry, *entry = NULL;
295 	struct mlx4_promisc_qp *dqp, *tmp_dqp;
296 
297 	if (port < 1 || port > dev->caps.num_ports)
298 		return NULL;
299 
300 	s_steer = &mlx4_priv(dev)->steer[port - 1];
301 
302 	/* if qp is not promisc, it cannot be duplicated */
303 	if (!get_promisc_qp(dev, port, steer, qpn))
304 		return false;
305 
306 	/* The qp is promisc qp so it is a duplicate on this index
307 	 * Find the index entry, and remove the duplicate */
308 	list_for_each_entry(tmp_entry, &s_steer->steer_entries[steer], list) {
309 		if (tmp_entry->index == index) {
310 			entry = tmp_entry;
311 			break;
312 		}
313 	}
314 	if (unlikely(!entry)) {
315 		mlx4_warn(dev, "Steering entry for index %x is not registered\n", index);
316 		return false;
317 	}
318 	list_for_each_entry_safe(dqp, tmp_dqp, &entry->duplicates, list) {
319 		if (dqp->qpn == qpn) {
320 			list_del(&dqp->list);
321 			kfree(dqp);
322 		}
323 	}
324 	return true;
325 }
326 
327 /* Returns true if all the QPs != tqpn contained in this entry
328  * are Promisc QPs. Returns false otherwise.
329  */
330 static bool promisc_steering_entry(struct mlx4_dev *dev, u8 port,
331 				   enum mlx4_steer_type steer,
332 				   unsigned int index, u32 tqpn,
333 				   u32 *members_count)
334 {
335 	struct mlx4_cmd_mailbox *mailbox;
336 	struct mlx4_mgm *mgm;
337 	u32 m_count;
338 	bool ret = false;
339 	int i;
340 
341 	if (port < 1 || port > dev->caps.num_ports)
342 		return false;
343 
344 	mailbox = mlx4_alloc_cmd_mailbox(dev);
345 	if (IS_ERR(mailbox))
346 		return false;
347 	mgm = mailbox->buf;
348 
349 	if (mlx4_READ_ENTRY(dev, index, mailbox))
350 		goto out;
351 	m_count = be32_to_cpu(mgm->members_count) & 0xffffff;
352 	if (members_count)
353 		*members_count = m_count;
354 
355 	for (i = 0;  i < m_count; i++) {
356 		u32 qpn = be32_to_cpu(mgm->qp[i]) & MGM_QPN_MASK;
357 		if (!get_promisc_qp(dev, port, steer, qpn) && qpn != tqpn) {
358 			/* the qp is not promisc, the entry can't be removed */
359 			goto out;
360 		}
361 	}
362 	ret = true;
363 out:
364 	mlx4_free_cmd_mailbox(dev, mailbox);
365 	return ret;
366 }
367 
368 /* IF a steering entry contains only promisc QPs, it can be removed. */
369 static bool can_remove_steering_entry(struct mlx4_dev *dev, u8 port,
370 				      enum mlx4_steer_type steer,
371 				      unsigned int index, u32 tqpn)
372 {
373 	struct mlx4_steer *s_steer;
374 	struct mlx4_steer_index *entry = NULL, *tmp_entry;
375 	u32 members_count;
376 	bool ret = false;
377 
378 	if (port < 1 || port > dev->caps.num_ports)
379 		return NULL;
380 
381 	s_steer = &mlx4_priv(dev)->steer[port - 1];
382 
383 	if (!promisc_steering_entry(dev, port, steer, index,
384 				    tqpn, &members_count))
385 		goto out;
386 
387 	/* All the qps currently registered for this entry are promiscuous,
388 	  * Checking for duplicates */
389 	ret = true;
390 	list_for_each_entry_safe(entry, tmp_entry, &s_steer->steer_entries[steer], list) {
391 		if (entry->index == index) {
392 			if (list_empty(&entry->duplicates) ||
393 			    members_count == 1) {
394 				struct mlx4_promisc_qp *pqp, *tmp_pqp;
395 				/* If there is only 1 entry in duplicates then
396 				 * this is the QP we want to delete, going over
397 				 * the list and deleting the entry.
398 				 */
399 				list_del(&entry->list);
400 				list_for_each_entry_safe(pqp, tmp_pqp,
401 							 &entry->duplicates,
402 							 list) {
403 					list_del(&pqp->list);
404 					kfree(pqp);
405 				}
406 				kfree(entry);
407 			} else {
408 				/* This entry contains duplicates so it shouldn't be removed */
409 				ret = false;
410 				goto out;
411 			}
412 		}
413 	}
414 
415 out:
416 	return ret;
417 }
418 
419 static int add_promisc_qp(struct mlx4_dev *dev, u8 port,
420 			  enum mlx4_steer_type steer, u32 qpn)
421 {
422 	struct mlx4_steer *s_steer;
423 	struct mlx4_cmd_mailbox *mailbox;
424 	struct mlx4_mgm *mgm;
425 	struct mlx4_steer_index *entry;
426 	struct mlx4_promisc_qp *pqp;
427 	struct mlx4_promisc_qp *dqp;
428 	u32 members_count;
429 	u32 prot;
430 	int i;
431 	bool found;
432 	int err;
433 	struct mlx4_priv *priv = mlx4_priv(dev);
434 
435 	if (port < 1 || port > dev->caps.num_ports)
436 		return -EINVAL;
437 
438 	s_steer = &mlx4_priv(dev)->steer[port - 1];
439 
440 	mutex_lock(&priv->mcg_table.mutex);
441 
442 	if (get_promisc_qp(dev, port, steer, qpn)) {
443 		err = 0;  /* Noting to do, already exists */
444 		goto out_mutex;
445 	}
446 
447 	pqp = kmalloc(sizeof *pqp, GFP_KERNEL);
448 	if (!pqp) {
449 		err = -ENOMEM;
450 		goto out_mutex;
451 	}
452 	pqp->qpn = qpn;
453 
454 	mailbox = mlx4_alloc_cmd_mailbox(dev);
455 	if (IS_ERR(mailbox)) {
456 		err = -ENOMEM;
457 		goto out_alloc;
458 	}
459 	mgm = mailbox->buf;
460 
461 	if (!(mlx4_is_mfunc(dev) && steer == MLX4_UC_STEER)) {
462 		/* The promisc QP needs to be added for each one of the steering
463 		 * entries. If it already exists, needs to be added as
464 		 * a duplicate for this entry.
465 		 */
466 		list_for_each_entry(entry,
467 				    &s_steer->steer_entries[steer],
468 				    list) {
469 			err = mlx4_READ_ENTRY(dev, entry->index, mailbox);
470 			if (err)
471 				goto out_mailbox;
472 
473 			members_count = be32_to_cpu(mgm->members_count) &
474 					0xffffff;
475 			prot = be32_to_cpu(mgm->members_count) >> 30;
476 			found = false;
477 			for (i = 0; i < members_count; i++) {
478 				if ((be32_to_cpu(mgm->qp[i]) &
479 				     MGM_QPN_MASK) == qpn) {
480 					/* Entry already exists.
481 					 * Add to duplicates.
482 					 */
483 					dqp = kmalloc(sizeof(*dqp), GFP_KERNEL);
484 					if (!dqp) {
485 						err = -ENOMEM;
486 						goto out_mailbox;
487 					}
488 					dqp->qpn = qpn;
489 					list_add_tail(&dqp->list,
490 						      &entry->duplicates);
491 					found = true;
492 				}
493 			}
494 			if (!found) {
495 				/* Need to add the qpn to mgm */
496 				if (members_count ==
497 				    dev->caps.num_qp_per_mgm) {
498 					/* entry is full */
499 					err = -ENOMEM;
500 					goto out_mailbox;
501 				}
502 				mgm->qp[members_count++] =
503 					cpu_to_be32(qpn & MGM_QPN_MASK);
504 				mgm->members_count =
505 					cpu_to_be32(members_count |
506 						    (prot << 30));
507 				err = mlx4_WRITE_ENTRY(dev, entry->index,
508 						       mailbox);
509 				if (err)
510 					goto out_mailbox;
511 			}
512 		}
513 	}
514 
515 	/* add the new qpn to list of promisc qps */
516 	list_add_tail(&pqp->list, &s_steer->promisc_qps[steer]);
517 	/* now need to add all the promisc qps to default entry */
518 	memset(mgm, 0, sizeof *mgm);
519 	members_count = 0;
520 	list_for_each_entry(dqp, &s_steer->promisc_qps[steer], list) {
521 		if (members_count == dev->caps.num_qp_per_mgm) {
522 			/* entry is full */
523 			err = -ENOMEM;
524 			goto out_list;
525 		}
526 		mgm->qp[members_count++] = cpu_to_be32(dqp->qpn & MGM_QPN_MASK);
527 	}
528 	mgm->members_count = cpu_to_be32(members_count | MLX4_PROT_ETH << 30);
529 
530 	err = mlx4_WRITE_PROMISC(dev, port, steer, mailbox);
531 	if (err)
532 		goto out_list;
533 
534 	mlx4_free_cmd_mailbox(dev, mailbox);
535 	mutex_unlock(&priv->mcg_table.mutex);
536 	return 0;
537 
538 out_list:
539 	list_del(&pqp->list);
540 out_mailbox:
541 	mlx4_free_cmd_mailbox(dev, mailbox);
542 out_alloc:
543 	kfree(pqp);
544 out_mutex:
545 	mutex_unlock(&priv->mcg_table.mutex);
546 	return err;
547 }
548 
549 static int remove_promisc_qp(struct mlx4_dev *dev, u8 port,
550 			     enum mlx4_steer_type steer, u32 qpn)
551 {
552 	struct mlx4_priv *priv = mlx4_priv(dev);
553 	struct mlx4_steer *s_steer;
554 	struct mlx4_cmd_mailbox *mailbox;
555 	struct mlx4_mgm *mgm;
556 	struct mlx4_steer_index *entry, *tmp_entry;
557 	struct mlx4_promisc_qp *pqp;
558 	struct mlx4_promisc_qp *dqp;
559 	u32 members_count;
560 	bool found;
561 	bool back_to_list = false;
562 	int i;
563 	int err;
564 
565 	if (port < 1 || port > dev->caps.num_ports)
566 		return -EINVAL;
567 
568 	s_steer = &mlx4_priv(dev)->steer[port - 1];
569 	mutex_lock(&priv->mcg_table.mutex);
570 
571 	pqp = get_promisc_qp(dev, port, steer, qpn);
572 	if (unlikely(!pqp)) {
573 		mlx4_warn(dev, "QP %x is not promiscuous QP\n", qpn);
574 		/* nothing to do */
575 		err = 0;
576 		goto out_mutex;
577 	}
578 
579 	/*remove from list of promisc qps */
580 	list_del(&pqp->list);
581 
582 	/* set the default entry not to include the removed one */
583 	mailbox = mlx4_alloc_cmd_mailbox(dev);
584 	if (IS_ERR(mailbox)) {
585 		err = -ENOMEM;
586 		back_to_list = true;
587 		goto out_list;
588 	}
589 	mgm = mailbox->buf;
590 	members_count = 0;
591 	list_for_each_entry(dqp, &s_steer->promisc_qps[steer], list)
592 		mgm->qp[members_count++] = cpu_to_be32(dqp->qpn & MGM_QPN_MASK);
593 	mgm->members_count = cpu_to_be32(members_count | MLX4_PROT_ETH << 30);
594 
595 	err = mlx4_WRITE_PROMISC(dev, port, steer, mailbox);
596 	if (err)
597 		goto out_mailbox;
598 
599 	if (!(mlx4_is_mfunc(dev) && steer == MLX4_UC_STEER)) {
600 		/* Remove the QP from all the steering entries */
601 		list_for_each_entry_safe(entry, tmp_entry,
602 					 &s_steer->steer_entries[steer],
603 					 list) {
604 			found = false;
605 			list_for_each_entry(dqp, &entry->duplicates, list) {
606 				if (dqp->qpn == qpn) {
607 					found = true;
608 					break;
609 				}
610 			}
611 			if (found) {
612 				/* A duplicate, no need to change the MGM,
613 				 * only update the duplicates list
614 				 */
615 				list_del(&dqp->list);
616 				kfree(dqp);
617 			} else {
618 				int loc = -1;
619 
620 				err = mlx4_READ_ENTRY(dev,
621 						      entry->index,
622 						      mailbox);
623 					if (err)
624 						goto out_mailbox;
625 				members_count =
626 					be32_to_cpu(mgm->members_count) &
627 					0xffffff;
628 				if (!members_count) {
629 					mlx4_warn(dev, "QP %06x wasn't found in entry %x mcount=0. deleting entry...\n",
630 						  qpn, entry->index);
631 					list_del(&entry->list);
632 					kfree(entry);
633 					continue;
634 				}
635 
636 				for (i = 0; i < members_count; ++i)
637 					if ((be32_to_cpu(mgm->qp[i]) &
638 					     MGM_QPN_MASK) == qpn) {
639 						loc = i;
640 						break;
641 					}
642 
643 				if (loc < 0) {
644 					mlx4_err(dev, "QP %06x wasn't found in entry %d\n",
645 						 qpn, entry->index);
646 					err = -EINVAL;
647 					goto out_mailbox;
648 				}
649 
650 				/* Copy the last QP in this MGM
651 				 * over removed QP
652 				 */
653 				mgm->qp[loc] = mgm->qp[members_count - 1];
654 				mgm->qp[members_count - 1] = 0;
655 				mgm->members_count =
656 					cpu_to_be32(--members_count |
657 						    (MLX4_PROT_ETH << 30));
658 
659 				err = mlx4_WRITE_ENTRY(dev,
660 						       entry->index,
661 						       mailbox);
662 					if (err)
663 						goto out_mailbox;
664 			}
665 		}
666 	}
667 
668 out_mailbox:
669 	mlx4_free_cmd_mailbox(dev, mailbox);
670 out_list:
671 	if (back_to_list)
672 		list_add_tail(&pqp->list, &s_steer->promisc_qps[steer]);
673 	else
674 		kfree(pqp);
675 out_mutex:
676 	mutex_unlock(&priv->mcg_table.mutex);
677 	return err;
678 }
679 
680 /*
681  * Caller must hold MCG table semaphore.  gid and mgm parameters must
682  * be properly aligned for command interface.
683  *
684  *  Returns 0 unless a firmware command error occurs.
685  *
686  * If GID is found in MGM or MGM is empty, *index = *hash, *prev = -1
687  * and *mgm holds MGM entry.
688  *
689  * if GID is found in AMGM, *index = index in AMGM, *prev = index of
690  * previous entry in hash chain and *mgm holds AMGM entry.
691  *
692  * If no AMGM exists for given gid, *index = -1, *prev = index of last
693  * entry in hash chain and *mgm holds end of hash chain.
694  */
695 static int find_entry(struct mlx4_dev *dev, u8 port,
696 		      u8 *gid, enum mlx4_protocol prot,
697 		      struct mlx4_cmd_mailbox *mgm_mailbox,
698 		      int *prev, int *index)
699 {
700 	struct mlx4_cmd_mailbox *mailbox;
701 	struct mlx4_mgm *mgm = mgm_mailbox->buf;
702 	u8 *mgid;
703 	int err;
704 	u16 hash;
705 	u8 op_mod = (prot == MLX4_PROT_ETH) ?
706 		!!(dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER) : 0;
707 
708 	mailbox = mlx4_alloc_cmd_mailbox(dev);
709 	if (IS_ERR(mailbox))
710 		return -ENOMEM;
711 	mgid = mailbox->buf;
712 
713 	memcpy(mgid, gid, 16);
714 
715 	err = mlx4_GID_HASH(dev, mailbox, &hash, op_mod);
716 	mlx4_free_cmd_mailbox(dev, mailbox);
717 	if (err)
718 		return err;
719 
720 	if (0)
721 		mlx4_dbg(dev, "Hash for %pI6 is %04x\n", gid, hash);
722 
723 	*index = hash;
724 	*prev  = -1;
725 
726 	do {
727 		err = mlx4_READ_ENTRY(dev, *index, mgm_mailbox);
728 		if (err)
729 			return err;
730 
731 		if (!(be32_to_cpu(mgm->members_count) & 0xffffff)) {
732 			if (*index != hash) {
733 				mlx4_err(dev, "Found zero MGID in AMGM\n");
734 				err = -EINVAL;
735 			}
736 			return err;
737 		}
738 
739 		if (!memcmp(mgm->gid, gid, 16) &&
740 		    be32_to_cpu(mgm->members_count) >> 30 == prot)
741 			return err;
742 
743 		*prev = *index;
744 		*index = be32_to_cpu(mgm->next_gid_index) >> 6;
745 	} while (*index);
746 
747 	*index = -1;
748 	return err;
749 }
750 
751 static const u8 __promisc_mode[] = {
752 	[MLX4_FS_REGULAR]   = 0x0,
753 	[MLX4_FS_ALL_DEFAULT] = 0x1,
754 	[MLX4_FS_MC_DEFAULT] = 0x3,
755 	[MLX4_FS_UC_SNIFFER] = 0x4,
756 	[MLX4_FS_MC_SNIFFER] = 0x5,
757 };
758 
759 int mlx4_map_sw_to_hw_steering_mode(struct mlx4_dev *dev,
760 				    enum mlx4_net_trans_promisc_mode flow_type)
761 {
762 	if (flow_type >= MLX4_FS_MODE_NUM) {
763 		mlx4_err(dev, "Invalid flow type. type = %d\n", flow_type);
764 		return -EINVAL;
765 	}
766 	return __promisc_mode[flow_type];
767 }
768 EXPORT_SYMBOL_GPL(mlx4_map_sw_to_hw_steering_mode);
769 
770 static void trans_rule_ctrl_to_hw(struct mlx4_net_trans_rule *ctrl,
771 				  struct mlx4_net_trans_rule_hw_ctrl *hw)
772 {
773 	u8 flags = 0;
774 
775 	flags = ctrl->queue_mode == MLX4_NET_TRANS_Q_LIFO ? 1 : 0;
776 	flags |= ctrl->exclusive ? (1 << 2) : 0;
777 	flags |= ctrl->allow_loopback ? (1 << 3) : 0;
778 
779 	hw->flags = flags;
780 	hw->type = __promisc_mode[ctrl->promisc_mode];
781 	hw->prio = cpu_to_be16(ctrl->priority);
782 	hw->port = ctrl->port;
783 	hw->qpn = cpu_to_be32(ctrl->qpn);
784 }
785 
786 const u16 __sw_id_hw[] = {
787 	[MLX4_NET_TRANS_RULE_ID_ETH]     = 0xE001,
788 	[MLX4_NET_TRANS_RULE_ID_IB]      = 0xE005,
789 	[MLX4_NET_TRANS_RULE_ID_IPV6]    = 0xE003,
790 	[MLX4_NET_TRANS_RULE_ID_IPV4]    = 0xE002,
791 	[MLX4_NET_TRANS_RULE_ID_TCP]     = 0xE004,
792 	[MLX4_NET_TRANS_RULE_ID_UDP]     = 0xE006,
793 	[MLX4_NET_TRANS_RULE_ID_VXLAN]	 = 0xE008
794 };
795 
796 int mlx4_map_sw_to_hw_steering_id(struct mlx4_dev *dev,
797 				  enum mlx4_net_trans_rule_id id)
798 {
799 	if (id >= MLX4_NET_TRANS_RULE_NUM) {
800 		mlx4_err(dev, "Invalid network rule id. id = %d\n", id);
801 		return -EINVAL;
802 	}
803 	return __sw_id_hw[id];
804 }
805 EXPORT_SYMBOL_GPL(mlx4_map_sw_to_hw_steering_id);
806 
807 static const int __rule_hw_sz[] = {
808 	[MLX4_NET_TRANS_RULE_ID_ETH] =
809 		sizeof(struct mlx4_net_trans_rule_hw_eth),
810 	[MLX4_NET_TRANS_RULE_ID_IB] =
811 		sizeof(struct mlx4_net_trans_rule_hw_ib),
812 	[MLX4_NET_TRANS_RULE_ID_IPV6] = 0,
813 	[MLX4_NET_TRANS_RULE_ID_IPV4] =
814 		sizeof(struct mlx4_net_trans_rule_hw_ipv4),
815 	[MLX4_NET_TRANS_RULE_ID_TCP] =
816 		sizeof(struct mlx4_net_trans_rule_hw_tcp_udp),
817 	[MLX4_NET_TRANS_RULE_ID_UDP] =
818 		sizeof(struct mlx4_net_trans_rule_hw_tcp_udp),
819 	[MLX4_NET_TRANS_RULE_ID_VXLAN] =
820 		sizeof(struct mlx4_net_trans_rule_hw_vxlan)
821 };
822 
823 int mlx4_hw_rule_sz(struct mlx4_dev *dev,
824 	       enum mlx4_net_trans_rule_id id)
825 {
826 	if (id >= MLX4_NET_TRANS_RULE_NUM) {
827 		mlx4_err(dev, "Invalid network rule id. id = %d\n", id);
828 		return -EINVAL;
829 	}
830 
831 	return __rule_hw_sz[id];
832 }
833 EXPORT_SYMBOL_GPL(mlx4_hw_rule_sz);
834 
835 static int parse_trans_rule(struct mlx4_dev *dev, struct mlx4_spec_list *spec,
836 			    struct _rule_hw *rule_hw)
837 {
838 	if (mlx4_hw_rule_sz(dev, spec->id) < 0)
839 		return -EINVAL;
840 	memset(rule_hw, 0, mlx4_hw_rule_sz(dev, spec->id));
841 	rule_hw->id = cpu_to_be16(__sw_id_hw[spec->id]);
842 	rule_hw->size = mlx4_hw_rule_sz(dev, spec->id) >> 2;
843 
844 	switch (spec->id) {
845 	case MLX4_NET_TRANS_RULE_ID_ETH:
846 		memcpy(rule_hw->eth.dst_mac, spec->eth.dst_mac, ETH_ALEN);
847 		memcpy(rule_hw->eth.dst_mac_msk, spec->eth.dst_mac_msk,
848 		       ETH_ALEN);
849 		memcpy(rule_hw->eth.src_mac, spec->eth.src_mac, ETH_ALEN);
850 		memcpy(rule_hw->eth.src_mac_msk, spec->eth.src_mac_msk,
851 		       ETH_ALEN);
852 		if (spec->eth.ether_type_enable) {
853 			rule_hw->eth.ether_type_enable = 1;
854 			rule_hw->eth.ether_type = spec->eth.ether_type;
855 		}
856 		rule_hw->eth.vlan_tag = spec->eth.vlan_id;
857 		rule_hw->eth.vlan_tag_msk = spec->eth.vlan_id_msk;
858 		break;
859 
860 	case MLX4_NET_TRANS_RULE_ID_IB:
861 		rule_hw->ib.l3_qpn = spec->ib.l3_qpn;
862 		rule_hw->ib.qpn_mask = spec->ib.qpn_msk;
863 		memcpy(&rule_hw->ib.dst_gid, &spec->ib.dst_gid, 16);
864 		memcpy(&rule_hw->ib.dst_gid_msk, &spec->ib.dst_gid_msk, 16);
865 		break;
866 
867 	case MLX4_NET_TRANS_RULE_ID_IPV6:
868 		return -EOPNOTSUPP;
869 
870 	case MLX4_NET_TRANS_RULE_ID_IPV4:
871 		rule_hw->ipv4.src_ip = spec->ipv4.src_ip;
872 		rule_hw->ipv4.src_ip_msk = spec->ipv4.src_ip_msk;
873 		rule_hw->ipv4.dst_ip = spec->ipv4.dst_ip;
874 		rule_hw->ipv4.dst_ip_msk = spec->ipv4.dst_ip_msk;
875 		break;
876 
877 	case MLX4_NET_TRANS_RULE_ID_TCP:
878 	case MLX4_NET_TRANS_RULE_ID_UDP:
879 		rule_hw->tcp_udp.dst_port = spec->tcp_udp.dst_port;
880 		rule_hw->tcp_udp.dst_port_msk = spec->tcp_udp.dst_port_msk;
881 		rule_hw->tcp_udp.src_port = spec->tcp_udp.src_port;
882 		rule_hw->tcp_udp.src_port_msk = spec->tcp_udp.src_port_msk;
883 		break;
884 
885 	case MLX4_NET_TRANS_RULE_ID_VXLAN:
886 		rule_hw->vxlan.vni =
887 			cpu_to_be32(be32_to_cpu(spec->vxlan.vni) << 8);
888 		rule_hw->vxlan.vni_mask =
889 			cpu_to_be32(be32_to_cpu(spec->vxlan.vni_mask) << 8);
890 		break;
891 
892 	default:
893 		return -EINVAL;
894 	}
895 
896 	return __rule_hw_sz[spec->id];
897 }
898 
899 static void mlx4_err_rule(struct mlx4_dev *dev, char *str,
900 			  struct mlx4_net_trans_rule *rule)
901 {
902 #define BUF_SIZE 256
903 	struct mlx4_spec_list *cur;
904 	char buf[BUF_SIZE];
905 	int len = 0;
906 
907 	mlx4_err(dev, "%s", str);
908 	len += snprintf(buf + len, BUF_SIZE - len,
909 			"port = %d prio = 0x%x qp = 0x%x ",
910 			rule->port, rule->priority, rule->qpn);
911 
912 	list_for_each_entry(cur, &rule->list, list) {
913 		switch (cur->id) {
914 		case MLX4_NET_TRANS_RULE_ID_ETH:
915 			len += snprintf(buf + len, BUF_SIZE - len,
916 					"dmac = %pM ", &cur->eth.dst_mac);
917 			if (cur->eth.ether_type)
918 				len += snprintf(buf + len, BUF_SIZE - len,
919 						"ethertype = 0x%x ",
920 						be16_to_cpu(cur->eth.ether_type));
921 			if (cur->eth.vlan_id)
922 				len += snprintf(buf + len, BUF_SIZE - len,
923 						"vlan-id = %d ",
924 						be16_to_cpu(cur->eth.vlan_id));
925 			break;
926 
927 		case MLX4_NET_TRANS_RULE_ID_IPV4:
928 			if (cur->ipv4.src_ip)
929 				len += snprintf(buf + len, BUF_SIZE - len,
930 						"src-ip = %pI4 ",
931 						&cur->ipv4.src_ip);
932 			if (cur->ipv4.dst_ip)
933 				len += snprintf(buf + len, BUF_SIZE - len,
934 						"dst-ip = %pI4 ",
935 						&cur->ipv4.dst_ip);
936 			break;
937 
938 		case MLX4_NET_TRANS_RULE_ID_TCP:
939 		case MLX4_NET_TRANS_RULE_ID_UDP:
940 			if (cur->tcp_udp.src_port)
941 				len += snprintf(buf + len, BUF_SIZE - len,
942 						"src-port = %d ",
943 						be16_to_cpu(cur->tcp_udp.src_port));
944 			if (cur->tcp_udp.dst_port)
945 				len += snprintf(buf + len, BUF_SIZE - len,
946 						"dst-port = %d ",
947 						be16_to_cpu(cur->tcp_udp.dst_port));
948 			break;
949 
950 		case MLX4_NET_TRANS_RULE_ID_IB:
951 			len += snprintf(buf + len, BUF_SIZE - len,
952 					"dst-gid = %pI6\n", cur->ib.dst_gid);
953 			len += snprintf(buf + len, BUF_SIZE - len,
954 					"dst-gid-mask = %pI6\n",
955 					cur->ib.dst_gid_msk);
956 			break;
957 
958 		case MLX4_NET_TRANS_RULE_ID_VXLAN:
959 			len += snprintf(buf + len, BUF_SIZE - len,
960 					"VNID = %d ", be32_to_cpu(cur->vxlan.vni));
961 			break;
962 		case MLX4_NET_TRANS_RULE_ID_IPV6:
963 			break;
964 
965 		default:
966 			break;
967 		}
968 	}
969 	len += snprintf(buf + len, BUF_SIZE - len, "\n");
970 	mlx4_err(dev, "%s", buf);
971 
972 	if (len >= BUF_SIZE)
973 		mlx4_err(dev, "Network rule error message was truncated, print buffer is too small\n");
974 }
975 
976 int mlx4_flow_attach(struct mlx4_dev *dev,
977 		     struct mlx4_net_trans_rule *rule, u64 *reg_id)
978 {
979 	struct mlx4_cmd_mailbox *mailbox;
980 	struct mlx4_spec_list *cur;
981 	u32 size = 0;
982 	int ret;
983 
984 	mailbox = mlx4_alloc_cmd_mailbox(dev);
985 	if (IS_ERR(mailbox))
986 		return PTR_ERR(mailbox);
987 
988 	trans_rule_ctrl_to_hw(rule, mailbox->buf);
989 
990 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
991 
992 	list_for_each_entry(cur, &rule->list, list) {
993 		ret = parse_trans_rule(dev, cur, mailbox->buf + size);
994 		if (ret < 0) {
995 			mlx4_free_cmd_mailbox(dev, mailbox);
996 			return ret;
997 		}
998 		size += ret;
999 	}
1000 
1001 	ret = mlx4_QP_FLOW_STEERING_ATTACH(dev, mailbox, size >> 2, reg_id);
1002 	if (ret == -ENOMEM) {
1003 		mlx4_err_rule(dev,
1004 			      "mcg table is full. Fail to register network rule\n",
1005 			      rule);
1006 	} else if (ret) {
1007 		if (ret == -ENXIO) {
1008 			if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED)
1009 				mlx4_err_rule(dev,
1010 					      "DMFS is not enabled, "
1011 					      "failed to register network rule.\n",
1012 					      rule);
1013 			else
1014 				mlx4_err_rule(dev,
1015 					      "Rule exceeds the dmfs_high_rate_mode limitations, "
1016 					      "failed to register network rule.\n",
1017 					      rule);
1018 
1019 		} else {
1020 			mlx4_err_rule(dev, "Fail to register network rule.\n", rule);
1021 		}
1022 	}
1023 
1024 	mlx4_free_cmd_mailbox(dev, mailbox);
1025 
1026 	return ret;
1027 }
1028 EXPORT_SYMBOL_GPL(mlx4_flow_attach);
1029 
1030 int mlx4_flow_detach(struct mlx4_dev *dev, u64 reg_id)
1031 {
1032 	int err;
1033 
1034 	err = mlx4_QP_FLOW_STEERING_DETACH(dev, reg_id);
1035 	if (err)
1036 		mlx4_err(dev, "Fail to detach network rule. registration id = 0x%llx\n",
1037 			 reg_id);
1038 	return err;
1039 }
1040 EXPORT_SYMBOL_GPL(mlx4_flow_detach);
1041 
1042 int mlx4_tunnel_steer_add(struct mlx4_dev *dev, unsigned char *addr,
1043 			  int port, int qpn, u16 prio, u64 *reg_id)
1044 {
1045 	int err;
1046 	struct mlx4_spec_list spec_eth_outer = { {NULL} };
1047 	struct mlx4_spec_list spec_vxlan     = { {NULL} };
1048 	struct mlx4_spec_list spec_eth_inner = { {NULL} };
1049 
1050 	struct mlx4_net_trans_rule rule = {
1051 		.queue_mode = MLX4_NET_TRANS_Q_FIFO,
1052 		.exclusive = 0,
1053 		.allow_loopback = 1,
1054 		.promisc_mode = MLX4_FS_REGULAR,
1055 	};
1056 
1057 	__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
1058 
1059 	rule.port = port;
1060 	rule.qpn = qpn;
1061 	rule.priority = prio;
1062 	INIT_LIST_HEAD(&rule.list);
1063 
1064 	spec_eth_outer.id = MLX4_NET_TRANS_RULE_ID_ETH;
1065 	memcpy(spec_eth_outer.eth.dst_mac, addr, ETH_ALEN);
1066 	memcpy(spec_eth_outer.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
1067 
1068 	spec_vxlan.id = MLX4_NET_TRANS_RULE_ID_VXLAN;    /* any vxlan header */
1069 	spec_eth_inner.id = MLX4_NET_TRANS_RULE_ID_ETH;	 /* any inner eth header */
1070 
1071 	list_add_tail(&spec_eth_outer.list, &rule.list);
1072 	list_add_tail(&spec_vxlan.list,     &rule.list);
1073 	list_add_tail(&spec_eth_inner.list, &rule.list);
1074 
1075 	err = mlx4_flow_attach(dev, &rule, reg_id);
1076 	return err;
1077 }
1078 EXPORT_SYMBOL(mlx4_tunnel_steer_add);
1079 
1080 int mlx4_FLOW_STEERING_IB_UC_QP_RANGE(struct mlx4_dev *dev, u32 min_range_qpn,
1081 				      u32 max_range_qpn)
1082 {
1083 	int err;
1084 	u64 in_param;
1085 
1086 	in_param = ((u64) min_range_qpn) << 32;
1087 	in_param |= ((u64) max_range_qpn) & 0xFFFFFFFF;
1088 
1089 	err = mlx4_cmd(dev, in_param, 0, 0,
1090 			MLX4_FLOW_STEERING_IB_UC_QP_RANGE,
1091 			MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1092 
1093 	return err;
1094 }
1095 EXPORT_SYMBOL_GPL(mlx4_FLOW_STEERING_IB_UC_QP_RANGE);
1096 
1097 int mlx4_qp_attach_common(struct mlx4_dev *dev, struct mlx4_qp *qp, u8 gid[16],
1098 			  int block_mcast_loopback, enum mlx4_protocol prot,
1099 			  enum mlx4_steer_type steer)
1100 {
1101 	struct mlx4_priv *priv = mlx4_priv(dev);
1102 	struct mlx4_cmd_mailbox *mailbox;
1103 	struct mlx4_mgm *mgm;
1104 	u32 members_count;
1105 	int index, prev;
1106 	int link = 0;
1107 	int i;
1108 	int err;
1109 	u8 port = gid[5];
1110 	u8 new_entry = 0;
1111 
1112 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1113 	if (IS_ERR(mailbox))
1114 		return PTR_ERR(mailbox);
1115 	mgm = mailbox->buf;
1116 
1117 	mutex_lock(&priv->mcg_table.mutex);
1118 	err = find_entry(dev, port, gid, prot,
1119 			 mailbox, &prev, &index);
1120 	if (err)
1121 		goto out;
1122 
1123 	if (index != -1) {
1124 		if (!(be32_to_cpu(mgm->members_count) & 0xffffff)) {
1125 			new_entry = 1;
1126 			memcpy(mgm->gid, gid, 16);
1127 		}
1128 	} else {
1129 		link = 1;
1130 
1131 		index = mlx4_bitmap_alloc(&priv->mcg_table.bitmap);
1132 		if (index == -1) {
1133 			mlx4_err(dev, "No AMGM entries left\n");
1134 			err = -ENOMEM;
1135 			goto out;
1136 		}
1137 		index += dev->caps.num_mgms;
1138 
1139 		new_entry = 1;
1140 		memset(mgm, 0, sizeof *mgm);
1141 		memcpy(mgm->gid, gid, 16);
1142 	}
1143 
1144 	members_count = be32_to_cpu(mgm->members_count) & 0xffffff;
1145 	if (members_count == dev->caps.num_qp_per_mgm) {
1146 		mlx4_err(dev, "MGM at index %x is full\n", index);
1147 		err = -ENOMEM;
1148 		goto out;
1149 	}
1150 
1151 	for (i = 0; i < members_count; ++i)
1152 		if ((be32_to_cpu(mgm->qp[i]) & MGM_QPN_MASK) == qp->qpn) {
1153 			mlx4_dbg(dev, "QP %06x already a member of MGM\n", qp->qpn);
1154 			err = 0;
1155 			goto out;
1156 		}
1157 
1158 	if (block_mcast_loopback)
1159 		mgm->qp[members_count++] = cpu_to_be32((qp->qpn & MGM_QPN_MASK) |
1160 						       (1U << MGM_BLCK_LB_BIT));
1161 	else
1162 		mgm->qp[members_count++] = cpu_to_be32(qp->qpn & MGM_QPN_MASK);
1163 
1164 	mgm->members_count = cpu_to_be32(members_count | (u32) prot << 30);
1165 
1166 	err = mlx4_WRITE_ENTRY(dev, index, mailbox);
1167 	if (err)
1168 		goto out;
1169 
1170 	if (!link)
1171 		goto out;
1172 
1173 	err = mlx4_READ_ENTRY(dev, prev, mailbox);
1174 	if (err)
1175 		goto out;
1176 
1177 	mgm->next_gid_index = cpu_to_be32(index << 6);
1178 
1179 	err = mlx4_WRITE_ENTRY(dev, prev, mailbox);
1180 	if (err)
1181 		goto out;
1182 
1183 out:
1184 	if (prot == MLX4_PROT_ETH) {
1185 		/* manage the steering entry for promisc mode */
1186 		if (new_entry)
1187 			err = new_steering_entry(dev, port, steer,
1188 						 index, qp->qpn);
1189 		else
1190 			err = existing_steering_entry(dev, port, steer,
1191 						      index, qp->qpn);
1192 	}
1193 	if (err && link && index != -1) {
1194 		if (index < dev->caps.num_mgms)
1195 			mlx4_warn(dev, "Got AMGM index %d < %d\n",
1196 				  index, dev->caps.num_mgms);
1197 		else
1198 			mlx4_bitmap_free(&priv->mcg_table.bitmap,
1199 					 index - dev->caps.num_mgms, MLX4_USE_RR);
1200 	}
1201 	mutex_unlock(&priv->mcg_table.mutex);
1202 
1203 	mlx4_free_cmd_mailbox(dev, mailbox);
1204 	return err;
1205 }
1206 
1207 int mlx4_qp_detach_common(struct mlx4_dev *dev, struct mlx4_qp *qp, u8 gid[16],
1208 			  enum mlx4_protocol prot, enum mlx4_steer_type steer)
1209 {
1210 	struct mlx4_priv *priv = mlx4_priv(dev);
1211 	struct mlx4_cmd_mailbox *mailbox;
1212 	struct mlx4_mgm *mgm;
1213 	u32 members_count;
1214 	int prev, index;
1215 	int i, loc = -1;
1216 	int err;
1217 	u8 port = gid[5];
1218 	bool removed_entry = false;
1219 
1220 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1221 	if (IS_ERR(mailbox))
1222 		return PTR_ERR(mailbox);
1223 	mgm = mailbox->buf;
1224 
1225 	mutex_lock(&priv->mcg_table.mutex);
1226 
1227 	err = find_entry(dev, port, gid, prot,
1228 			 mailbox, &prev, &index);
1229 	if (err)
1230 		goto out;
1231 
1232 	if (index == -1) {
1233 		mlx4_err(dev, "MGID %pI6 not found\n", gid);
1234 		err = -EINVAL;
1235 		goto out;
1236 	}
1237 
1238 	/* If this QP is also a promisc QP, it shouldn't be removed only if
1239 	 * at least one none promisc QP is also attached to this MCG
1240 	 */
1241 	if (prot == MLX4_PROT_ETH &&
1242 	    check_duplicate_entry(dev, port, steer, index, qp->qpn) &&
1243 	    !promisc_steering_entry(dev, port, steer, index, qp->qpn, NULL))
1244 			goto out;
1245 
1246 	members_count = be32_to_cpu(mgm->members_count) & 0xffffff;
1247 	for (i = 0; i < members_count; ++i)
1248 		if ((be32_to_cpu(mgm->qp[i]) & MGM_QPN_MASK) == qp->qpn) {
1249 			loc = i;
1250 			break;
1251 		}
1252 
1253 	if (loc == -1) {
1254 		mlx4_err(dev, "QP %06x not found in MGM\n", qp->qpn);
1255 		err = -EINVAL;
1256 		goto out;
1257 	}
1258 
1259 	/* copy the last QP in this MGM over removed QP */
1260 	mgm->qp[loc] = mgm->qp[members_count - 1];
1261 	mgm->qp[members_count - 1] = 0;
1262 	mgm->members_count = cpu_to_be32(--members_count | (u32) prot << 30);
1263 
1264 	if (prot == MLX4_PROT_ETH)
1265 		removed_entry = can_remove_steering_entry(dev, port, steer,
1266 								index, qp->qpn);
1267 	if (members_count && (prot != MLX4_PROT_ETH || !removed_entry)) {
1268 		err = mlx4_WRITE_ENTRY(dev, index, mailbox);
1269 		goto out;
1270 	}
1271 
1272 	/* We are going to delete the entry, members count should be 0 */
1273 	mgm->members_count = cpu_to_be32((u32) prot << 30);
1274 
1275 	if (prev == -1) {
1276 		/* Remove entry from MGM */
1277 		int amgm_index = be32_to_cpu(mgm->next_gid_index) >> 6;
1278 		if (amgm_index) {
1279 			err = mlx4_READ_ENTRY(dev, amgm_index, mailbox);
1280 			if (err)
1281 				goto out;
1282 		} else
1283 			memset(mgm->gid, 0, 16);
1284 
1285 		err = mlx4_WRITE_ENTRY(dev, index, mailbox);
1286 		if (err)
1287 			goto out;
1288 
1289 		if (amgm_index) {
1290 			if (amgm_index < dev->caps.num_mgms)
1291 				mlx4_warn(dev, "MGM entry %d had AMGM index %d < %d\n",
1292 					  index, amgm_index, dev->caps.num_mgms);
1293 			else
1294 				mlx4_bitmap_free(&priv->mcg_table.bitmap,
1295 						 amgm_index - dev->caps.num_mgms, MLX4_USE_RR);
1296 		}
1297 	} else {
1298 		/* Remove entry from AMGM */
1299 		int cur_next_index = be32_to_cpu(mgm->next_gid_index) >> 6;
1300 		err = mlx4_READ_ENTRY(dev, prev, mailbox);
1301 		if (err)
1302 			goto out;
1303 
1304 		mgm->next_gid_index = cpu_to_be32(cur_next_index << 6);
1305 
1306 		err = mlx4_WRITE_ENTRY(dev, prev, mailbox);
1307 		if (err)
1308 			goto out;
1309 
1310 		if (index < dev->caps.num_mgms)
1311 			mlx4_warn(dev, "entry %d had next AMGM index %d < %d\n",
1312 				  prev, index, dev->caps.num_mgms);
1313 		else
1314 			mlx4_bitmap_free(&priv->mcg_table.bitmap,
1315 					 index - dev->caps.num_mgms, MLX4_USE_RR);
1316 	}
1317 
1318 out:
1319 	mutex_unlock(&priv->mcg_table.mutex);
1320 
1321 	mlx4_free_cmd_mailbox(dev, mailbox);
1322 	if (err && dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
1323 		/* In case device is under an error, return success as a closing command */
1324 		err = 0;
1325 	return err;
1326 }
1327 
1328 static int mlx4_QP_ATTACH(struct mlx4_dev *dev, struct mlx4_qp *qp,
1329 			  u8 gid[16], u8 attach, u8 block_loopback,
1330 			  enum mlx4_protocol prot)
1331 {
1332 	struct mlx4_cmd_mailbox *mailbox;
1333 	int err = 0;
1334 	int qpn;
1335 
1336 	if (!mlx4_is_mfunc(dev))
1337 		return -EBADF;
1338 
1339 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1340 	if (IS_ERR(mailbox))
1341 		return PTR_ERR(mailbox);
1342 
1343 	memcpy(mailbox->buf, gid, 16);
1344 	qpn = qp->qpn;
1345 	qpn |= (prot << 28);
1346 	if (attach && block_loopback)
1347 		qpn |= (1 << 31);
1348 
1349 	err = mlx4_cmd(dev, mailbox->dma, qpn, attach,
1350 		       MLX4_CMD_QP_ATTACH, MLX4_CMD_TIME_CLASS_A,
1351 		       MLX4_CMD_WRAPPED);
1352 
1353 	mlx4_free_cmd_mailbox(dev, mailbox);
1354 	if (err && !attach &&
1355 	    dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
1356 		err = 0;
1357 	return err;
1358 }
1359 
1360 int mlx4_trans_to_dmfs_attach(struct mlx4_dev *dev, struct mlx4_qp *qp,
1361 			      u8 gid[16], u8 port,
1362 			      int block_mcast_loopback,
1363 			      enum mlx4_protocol prot, u64 *reg_id)
1364 {
1365 		struct mlx4_spec_list spec = { {NULL} };
1366 		__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
1367 
1368 		struct mlx4_net_trans_rule rule = {
1369 			.queue_mode = MLX4_NET_TRANS_Q_FIFO,
1370 			.exclusive = 0,
1371 			.promisc_mode = MLX4_FS_REGULAR,
1372 			.priority = MLX4_DOMAIN_NIC,
1373 		};
1374 
1375 		rule.allow_loopback = !block_mcast_loopback;
1376 		rule.port = port;
1377 		rule.qpn = qp->qpn;
1378 		INIT_LIST_HEAD(&rule.list);
1379 
1380 		switch (prot) {
1381 		case MLX4_PROT_ETH:
1382 			spec.id = MLX4_NET_TRANS_RULE_ID_ETH;
1383 			memcpy(spec.eth.dst_mac, &gid[10], ETH_ALEN);
1384 			memcpy(spec.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
1385 			break;
1386 
1387 		case MLX4_PROT_IB_IPV6:
1388 			spec.id = MLX4_NET_TRANS_RULE_ID_IB;
1389 			memcpy(spec.ib.dst_gid, gid, 16);
1390 			memset(&spec.ib.dst_gid_msk, 0xff, 16);
1391 			break;
1392 		default:
1393 			return -EINVAL;
1394 		}
1395 		list_add_tail(&spec.list, &rule.list);
1396 
1397 		return mlx4_flow_attach(dev, &rule, reg_id);
1398 }
1399 
1400 int mlx4_multicast_attach(struct mlx4_dev *dev, struct mlx4_qp *qp, u8 gid[16],
1401 			  u8 port, int block_mcast_loopback,
1402 			  enum mlx4_protocol prot, u64 *reg_id)
1403 {
1404 	switch (dev->caps.steering_mode) {
1405 	case MLX4_STEERING_MODE_A0:
1406 		if (prot == MLX4_PROT_ETH)
1407 			return 0;
1408 
1409 	case MLX4_STEERING_MODE_B0:
1410 		if (prot == MLX4_PROT_ETH)
1411 			gid[7] |= (MLX4_MC_STEER << 1);
1412 
1413 		if (mlx4_is_mfunc(dev))
1414 			return mlx4_QP_ATTACH(dev, qp, gid, 1,
1415 					      block_mcast_loopback, prot);
1416 		return mlx4_qp_attach_common(dev, qp, gid,
1417 					     block_mcast_loopback, prot,
1418 					     MLX4_MC_STEER);
1419 
1420 	case MLX4_STEERING_MODE_DEVICE_MANAGED:
1421 		return mlx4_trans_to_dmfs_attach(dev, qp, gid, port,
1422 						 block_mcast_loopback,
1423 						 prot, reg_id);
1424 	default:
1425 		return -EINVAL;
1426 	}
1427 }
1428 EXPORT_SYMBOL_GPL(mlx4_multicast_attach);
1429 
1430 int mlx4_multicast_detach(struct mlx4_dev *dev, struct mlx4_qp *qp, u8 gid[16],
1431 			  enum mlx4_protocol prot, u64 reg_id)
1432 {
1433 	switch (dev->caps.steering_mode) {
1434 	case MLX4_STEERING_MODE_A0:
1435 		if (prot == MLX4_PROT_ETH)
1436 			return 0;
1437 
1438 	case MLX4_STEERING_MODE_B0:
1439 		if (prot == MLX4_PROT_ETH)
1440 			gid[7] |= (MLX4_MC_STEER << 1);
1441 
1442 		if (mlx4_is_mfunc(dev))
1443 			return mlx4_QP_ATTACH(dev, qp, gid, 0, 0, prot);
1444 
1445 		return mlx4_qp_detach_common(dev, qp, gid, prot,
1446 					     MLX4_MC_STEER);
1447 
1448 	case MLX4_STEERING_MODE_DEVICE_MANAGED:
1449 		return mlx4_flow_detach(dev, reg_id);
1450 
1451 	default:
1452 		return -EINVAL;
1453 	}
1454 }
1455 EXPORT_SYMBOL_GPL(mlx4_multicast_detach);
1456 
1457 int mlx4_flow_steer_promisc_add(struct mlx4_dev *dev, u8 port,
1458 				u32 qpn, enum mlx4_net_trans_promisc_mode mode)
1459 {
1460 	struct mlx4_net_trans_rule rule;
1461 	u64 *regid_p;
1462 
1463 	switch (mode) {
1464 	case MLX4_FS_ALL_DEFAULT:
1465 		regid_p = &dev->regid_promisc_array[port];
1466 		break;
1467 	case MLX4_FS_MC_DEFAULT:
1468 		regid_p = &dev->regid_allmulti_array[port];
1469 		break;
1470 	default:
1471 		return -1;
1472 	}
1473 
1474 	if (*regid_p != 0)
1475 		return -1;
1476 
1477 	rule.promisc_mode = mode;
1478 	rule.port = port;
1479 	rule.qpn = qpn;
1480 	INIT_LIST_HEAD(&rule.list);
1481 	mlx4_err(dev, "going promisc on %x\n", port);
1482 
1483 	return  mlx4_flow_attach(dev, &rule, regid_p);
1484 }
1485 EXPORT_SYMBOL_GPL(mlx4_flow_steer_promisc_add);
1486 
1487 int mlx4_flow_steer_promisc_remove(struct mlx4_dev *dev, u8 port,
1488 				   enum mlx4_net_trans_promisc_mode mode)
1489 {
1490 	int ret;
1491 	u64 *regid_p;
1492 
1493 	switch (mode) {
1494 	case MLX4_FS_ALL_DEFAULT:
1495 		regid_p = &dev->regid_promisc_array[port];
1496 		break;
1497 	case MLX4_FS_MC_DEFAULT:
1498 		regid_p = &dev->regid_allmulti_array[port];
1499 		break;
1500 	default:
1501 		return -1;
1502 	}
1503 
1504 	if (*regid_p == 0)
1505 		return -1;
1506 
1507 	ret =  mlx4_flow_detach(dev, *regid_p);
1508 	if (ret == 0)
1509 		*regid_p = 0;
1510 
1511 	return ret;
1512 }
1513 EXPORT_SYMBOL_GPL(mlx4_flow_steer_promisc_remove);
1514 
1515 int mlx4_unicast_attach(struct mlx4_dev *dev,
1516 			struct mlx4_qp *qp, u8 gid[16],
1517 			int block_mcast_loopback, enum mlx4_protocol prot)
1518 {
1519 	if (prot == MLX4_PROT_ETH)
1520 		gid[7] |= (MLX4_UC_STEER << 1);
1521 
1522 	if (mlx4_is_mfunc(dev))
1523 		return mlx4_QP_ATTACH(dev, qp, gid, 1,
1524 					block_mcast_loopback, prot);
1525 
1526 	return mlx4_qp_attach_common(dev, qp, gid, block_mcast_loopback,
1527 					prot, MLX4_UC_STEER);
1528 }
1529 EXPORT_SYMBOL_GPL(mlx4_unicast_attach);
1530 
1531 int mlx4_unicast_detach(struct mlx4_dev *dev, struct mlx4_qp *qp,
1532 			       u8 gid[16], enum mlx4_protocol prot)
1533 {
1534 	if (prot == MLX4_PROT_ETH)
1535 		gid[7] |= (MLX4_UC_STEER << 1);
1536 
1537 	if (mlx4_is_mfunc(dev))
1538 		return mlx4_QP_ATTACH(dev, qp, gid, 0, 0, prot);
1539 
1540 	return mlx4_qp_detach_common(dev, qp, gid, prot, MLX4_UC_STEER);
1541 }
1542 EXPORT_SYMBOL_GPL(mlx4_unicast_detach);
1543 
1544 int mlx4_PROMISC_wrapper(struct mlx4_dev *dev, int slave,
1545 			 struct mlx4_vhcr *vhcr,
1546 			 struct mlx4_cmd_mailbox *inbox,
1547 			 struct mlx4_cmd_mailbox *outbox,
1548 			 struct mlx4_cmd_info *cmd)
1549 {
1550 	u32 qpn = (u32) vhcr->in_param & 0xffffffff;
1551 	int port = mlx4_slave_convert_port(dev, slave, vhcr->in_param >> 62);
1552 	enum mlx4_steer_type steer = vhcr->in_modifier;
1553 
1554 	if (port < 0)
1555 		return -EINVAL;
1556 
1557 	/* Promiscuous unicast is not allowed in mfunc */
1558 	if (mlx4_is_mfunc(dev) && steer == MLX4_UC_STEER)
1559 		return 0;
1560 
1561 	if (vhcr->op_modifier)
1562 		return add_promisc_qp(dev, port, steer, qpn);
1563 	else
1564 		return remove_promisc_qp(dev, port, steer, qpn);
1565 }
1566 
1567 static int mlx4_PROMISC(struct mlx4_dev *dev, u32 qpn,
1568 			enum mlx4_steer_type steer, u8 add, u8 port)
1569 {
1570 	return mlx4_cmd(dev, (u64) qpn | (u64) port << 62, (u32) steer, add,
1571 			MLX4_CMD_PROMISC, MLX4_CMD_TIME_CLASS_A,
1572 			MLX4_CMD_WRAPPED);
1573 }
1574 
1575 int mlx4_multicast_promisc_add(struct mlx4_dev *dev, u32 qpn, u8 port)
1576 {
1577 	if (mlx4_is_mfunc(dev))
1578 		return mlx4_PROMISC(dev, qpn, MLX4_MC_STEER, 1, port);
1579 
1580 	return add_promisc_qp(dev, port, MLX4_MC_STEER, qpn);
1581 }
1582 EXPORT_SYMBOL_GPL(mlx4_multicast_promisc_add);
1583 
1584 int mlx4_multicast_promisc_remove(struct mlx4_dev *dev, u32 qpn, u8 port)
1585 {
1586 	if (mlx4_is_mfunc(dev))
1587 		return mlx4_PROMISC(dev, qpn, MLX4_MC_STEER, 0, port);
1588 
1589 	return remove_promisc_qp(dev, port, MLX4_MC_STEER, qpn);
1590 }
1591 EXPORT_SYMBOL_GPL(mlx4_multicast_promisc_remove);
1592 
1593 int mlx4_unicast_promisc_add(struct mlx4_dev *dev, u32 qpn, u8 port)
1594 {
1595 	if (mlx4_is_mfunc(dev))
1596 		return mlx4_PROMISC(dev, qpn, MLX4_UC_STEER, 1, port);
1597 
1598 	return add_promisc_qp(dev, port, MLX4_UC_STEER, qpn);
1599 }
1600 EXPORT_SYMBOL_GPL(mlx4_unicast_promisc_add);
1601 
1602 int mlx4_unicast_promisc_remove(struct mlx4_dev *dev, u32 qpn, u8 port)
1603 {
1604 	if (mlx4_is_mfunc(dev))
1605 		return mlx4_PROMISC(dev, qpn, MLX4_UC_STEER, 0, port);
1606 
1607 	return remove_promisc_qp(dev, port, MLX4_UC_STEER, qpn);
1608 }
1609 EXPORT_SYMBOL_GPL(mlx4_unicast_promisc_remove);
1610 
1611 int mlx4_init_mcg_table(struct mlx4_dev *dev)
1612 {
1613 	struct mlx4_priv *priv = mlx4_priv(dev);
1614 	int err;
1615 
1616 	/* No need for mcg_table when fw managed the mcg table*/
1617 	if (dev->caps.steering_mode ==
1618 	    MLX4_STEERING_MODE_DEVICE_MANAGED)
1619 		return 0;
1620 	err = mlx4_bitmap_init(&priv->mcg_table.bitmap, dev->caps.num_amgms,
1621 			       dev->caps.num_amgms - 1, 0, 0);
1622 	if (err)
1623 		return err;
1624 
1625 	mutex_init(&priv->mcg_table.mutex);
1626 
1627 	return 0;
1628 }
1629 
1630 void mlx4_cleanup_mcg_table(struct mlx4_dev *dev)
1631 {
1632 	if (dev->caps.steering_mode !=
1633 	    MLX4_STEERING_MODE_DEVICE_MANAGED)
1634 		mlx4_bitmap_cleanup(&mlx4_priv(dev)->mcg_table.bitmap);
1635 }
1636