xref: /linux/drivers/net/ethernet/mellanox/mlx4/eq.c (revision f49f4ab95c301dbccad0efe85296d908b8ae7ad4)
1 /*
2  * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
3  * Copyright (c) 2005, 2006, 2007 Cisco Systems, Inc. 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/init.h>
35 #include <linux/interrupt.h>
36 #include <linux/slab.h>
37 #include <linux/export.h>
38 #include <linux/mm.h>
39 #include <linux/dma-mapping.h>
40 
41 #include <linux/mlx4/cmd.h>
42 #include <linux/cpu_rmap.h>
43 
44 #include "mlx4.h"
45 #include "fw.h"
46 
47 enum {
48 	MLX4_IRQNAME_SIZE	= 32
49 };
50 
51 enum {
52 	MLX4_NUM_ASYNC_EQE	= 0x100,
53 	MLX4_NUM_SPARE_EQE	= 0x80,
54 	MLX4_EQ_ENTRY_SIZE	= 0x20
55 };
56 
57 #define MLX4_EQ_STATUS_OK	   ( 0 << 28)
58 #define MLX4_EQ_STATUS_WRITE_FAIL  (10 << 28)
59 #define MLX4_EQ_OWNER_SW	   ( 0 << 24)
60 #define MLX4_EQ_OWNER_HW	   ( 1 << 24)
61 #define MLX4_EQ_FLAG_EC		   ( 1 << 18)
62 #define MLX4_EQ_FLAG_OI		   ( 1 << 17)
63 #define MLX4_EQ_STATE_ARMED	   ( 9 <<  8)
64 #define MLX4_EQ_STATE_FIRED	   (10 <<  8)
65 #define MLX4_EQ_STATE_ALWAYS_ARMED (11 <<  8)
66 
67 #define MLX4_ASYNC_EVENT_MASK ((1ull << MLX4_EVENT_TYPE_PATH_MIG)	    | \
68 			       (1ull << MLX4_EVENT_TYPE_COMM_EST)	    | \
69 			       (1ull << MLX4_EVENT_TYPE_SQ_DRAINED)	    | \
70 			       (1ull << MLX4_EVENT_TYPE_CQ_ERROR)	    | \
71 			       (1ull << MLX4_EVENT_TYPE_WQ_CATAS_ERROR)	    | \
72 			       (1ull << MLX4_EVENT_TYPE_EEC_CATAS_ERROR)    | \
73 			       (1ull << MLX4_EVENT_TYPE_PATH_MIG_FAILED)    | \
74 			       (1ull << MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR) | \
75 			       (1ull << MLX4_EVENT_TYPE_WQ_ACCESS_ERROR)    | \
76 			       (1ull << MLX4_EVENT_TYPE_PORT_CHANGE)	    | \
77 			       (1ull << MLX4_EVENT_TYPE_ECC_DETECT)	    | \
78 			       (1ull << MLX4_EVENT_TYPE_SRQ_CATAS_ERROR)    | \
79 			       (1ull << MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE)    | \
80 			       (1ull << MLX4_EVENT_TYPE_SRQ_LIMIT)	    | \
81 			       (1ull << MLX4_EVENT_TYPE_CMD)		    | \
82 			       (1ull << MLX4_EVENT_TYPE_COMM_CHANNEL)       | \
83 			       (1ull << MLX4_EVENT_TYPE_FLR_EVENT)	    | \
84 			       (1ull << MLX4_EVENT_TYPE_FATAL_WARNING))
85 
86 static u64 get_async_ev_mask(struct mlx4_dev *dev)
87 {
88 	u64 async_ev_mask = MLX4_ASYNC_EVENT_MASK;
89 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_PORT_MNG_CHG_EV)
90 		async_ev_mask |= (1ull << MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT);
91 
92 	return async_ev_mask;
93 }
94 
95 static void eq_set_ci(struct mlx4_eq *eq, int req_not)
96 {
97 	__raw_writel((__force u32) cpu_to_be32((eq->cons_index & 0xffffff) |
98 					       req_not << 31),
99 		     eq->doorbell);
100 	/* We still want ordering, just not swabbing, so add a barrier */
101 	mb();
102 }
103 
104 static struct mlx4_eqe *get_eqe(struct mlx4_eq *eq, u32 entry)
105 {
106 	unsigned long off = (entry & (eq->nent - 1)) * MLX4_EQ_ENTRY_SIZE;
107 	return eq->page_list[off / PAGE_SIZE].buf + off % PAGE_SIZE;
108 }
109 
110 static struct mlx4_eqe *next_eqe_sw(struct mlx4_eq *eq)
111 {
112 	struct mlx4_eqe *eqe = get_eqe(eq, eq->cons_index);
113 	return !!(eqe->owner & 0x80) ^ !!(eq->cons_index & eq->nent) ? NULL : eqe;
114 }
115 
116 static struct mlx4_eqe *next_slave_event_eqe(struct mlx4_slave_event_eq *slave_eq)
117 {
118 	struct mlx4_eqe *eqe =
119 		&slave_eq->event_eqe[slave_eq->cons & (SLAVE_EVENT_EQ_SIZE - 1)];
120 	return (!!(eqe->owner & 0x80) ^
121 		!!(slave_eq->cons & SLAVE_EVENT_EQ_SIZE)) ?
122 		eqe : NULL;
123 }
124 
125 void mlx4_gen_slave_eqe(struct work_struct *work)
126 {
127 	struct mlx4_mfunc_master_ctx *master =
128 		container_of(work, struct mlx4_mfunc_master_ctx,
129 			     slave_event_work);
130 	struct mlx4_mfunc *mfunc =
131 		container_of(master, struct mlx4_mfunc, master);
132 	struct mlx4_priv *priv = container_of(mfunc, struct mlx4_priv, mfunc);
133 	struct mlx4_dev *dev = &priv->dev;
134 	struct mlx4_slave_event_eq *slave_eq = &mfunc->master.slave_eq;
135 	struct mlx4_eqe *eqe;
136 	u8 slave;
137 	int i;
138 
139 	for (eqe = next_slave_event_eqe(slave_eq); eqe;
140 	      eqe = next_slave_event_eqe(slave_eq)) {
141 		slave = eqe->slave_id;
142 
143 		/* All active slaves need to receive the event */
144 		if (slave == ALL_SLAVES) {
145 			for (i = 0; i < dev->num_slaves; i++) {
146 				if (i != dev->caps.function &&
147 				    master->slave_state[i].active)
148 					if (mlx4_GEN_EQE(dev, i, eqe))
149 						mlx4_warn(dev, "Failed to "
150 							  " generate event "
151 							  "for slave %d\n", i);
152 			}
153 		} else {
154 			if (mlx4_GEN_EQE(dev, slave, eqe))
155 				mlx4_warn(dev, "Failed to generate event "
156 					       "for slave %d\n", slave);
157 		}
158 		++slave_eq->cons;
159 	}
160 }
161 
162 
163 static void slave_event(struct mlx4_dev *dev, u8 slave, struct mlx4_eqe *eqe)
164 {
165 	struct mlx4_priv *priv = mlx4_priv(dev);
166 	struct mlx4_slave_event_eq *slave_eq = &priv->mfunc.master.slave_eq;
167 	struct mlx4_eqe *s_eqe;
168 	unsigned long flags;
169 
170 	spin_lock_irqsave(&slave_eq->event_lock, flags);
171 	s_eqe = &slave_eq->event_eqe[slave_eq->prod & (SLAVE_EVENT_EQ_SIZE - 1)];
172 	if ((!!(s_eqe->owner & 0x80)) ^
173 	    (!!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE))) {
174 		mlx4_warn(dev, "Master failed to generate an EQE for slave: %d. "
175 			  "No free EQE on slave events queue\n", slave);
176 		spin_unlock_irqrestore(&slave_eq->event_lock, flags);
177 		return;
178 	}
179 
180 	memcpy(s_eqe, eqe, sizeof(struct mlx4_eqe) - 1);
181 	s_eqe->slave_id = slave;
182 	/* ensure all information is written before setting the ownersip bit */
183 	wmb();
184 	s_eqe->owner = !!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE) ? 0x0 : 0x80;
185 	++slave_eq->prod;
186 
187 	queue_work(priv->mfunc.master.comm_wq,
188 		   &priv->mfunc.master.slave_event_work);
189 	spin_unlock_irqrestore(&slave_eq->event_lock, flags);
190 }
191 
192 static void mlx4_slave_event(struct mlx4_dev *dev, int slave,
193 			     struct mlx4_eqe *eqe)
194 {
195 	struct mlx4_priv *priv = mlx4_priv(dev);
196 	struct mlx4_slave_state *s_slave =
197 		&priv->mfunc.master.slave_state[slave];
198 
199 	if (!s_slave->active) {
200 		/*mlx4_warn(dev, "Trying to pass event to inactive slave\n");*/
201 		return;
202 	}
203 
204 	slave_event(dev, slave, eqe);
205 }
206 
207 int mlx4_gen_pkey_eqe(struct mlx4_dev *dev, int slave, u8 port)
208 {
209 	struct mlx4_eqe eqe;
210 
211 	struct mlx4_priv *priv = mlx4_priv(dev);
212 	struct mlx4_slave_state *s_slave = &priv->mfunc.master.slave_state[slave];
213 
214 	if (!s_slave->active)
215 		return 0;
216 
217 	memset(&eqe, 0, sizeof eqe);
218 
219 	eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
220 	eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE;
221 	eqe.event.port_mgmt_change.port = port;
222 
223 	return mlx4_GEN_EQE(dev, slave, &eqe);
224 }
225 EXPORT_SYMBOL(mlx4_gen_pkey_eqe);
226 
227 int mlx4_gen_guid_change_eqe(struct mlx4_dev *dev, int slave, u8 port)
228 {
229 	struct mlx4_eqe eqe;
230 
231 	/*don't send if we don't have the that slave */
232 	if (dev->num_vfs < slave)
233 		return 0;
234 	memset(&eqe, 0, sizeof eqe);
235 
236 	eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
237 	eqe.subtype = MLX4_DEV_PMC_SUBTYPE_GUID_INFO;
238 	eqe.event.port_mgmt_change.port = port;
239 
240 	return mlx4_GEN_EQE(dev, slave, &eqe);
241 }
242 EXPORT_SYMBOL(mlx4_gen_guid_change_eqe);
243 
244 int mlx4_gen_port_state_change_eqe(struct mlx4_dev *dev, int slave, u8 port,
245 				   u8 port_subtype_change)
246 {
247 	struct mlx4_eqe eqe;
248 
249 	/*don't send if we don't have the that slave */
250 	if (dev->num_vfs < slave)
251 		return 0;
252 	memset(&eqe, 0, sizeof eqe);
253 
254 	eqe.type = MLX4_EVENT_TYPE_PORT_CHANGE;
255 	eqe.subtype = port_subtype_change;
256 	eqe.event.port_change.port = cpu_to_be32(port << 28);
257 
258 	mlx4_dbg(dev, "%s: sending: %d to slave: %d on port: %d\n", __func__,
259 		 port_subtype_change, slave, port);
260 	return mlx4_GEN_EQE(dev, slave, &eqe);
261 }
262 EXPORT_SYMBOL(mlx4_gen_port_state_change_eqe);
263 
264 enum slave_port_state mlx4_get_slave_port_state(struct mlx4_dev *dev, int slave, u8 port)
265 {
266 	struct mlx4_priv *priv = mlx4_priv(dev);
267 	struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state;
268 	if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS) {
269 		pr_err("%s: Error: asking for slave:%d, port:%d\n",
270 		       __func__, slave, port);
271 		return SLAVE_PORT_DOWN;
272 	}
273 	return s_state[slave].port_state[port];
274 }
275 EXPORT_SYMBOL(mlx4_get_slave_port_state);
276 
277 static int mlx4_set_slave_port_state(struct mlx4_dev *dev, int slave, u8 port,
278 				     enum slave_port_state state)
279 {
280 	struct mlx4_priv *priv = mlx4_priv(dev);
281 	struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state;
282 
283 	if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS || port == 0) {
284 		pr_err("%s: Error: asking for slave:%d, port:%d\n",
285 		       __func__, slave, port);
286 		return -1;
287 	}
288 	s_state[slave].port_state[port] = state;
289 
290 	return 0;
291 }
292 
293 static void set_all_slave_state(struct mlx4_dev *dev, u8 port, int event)
294 {
295 	int i;
296 	enum slave_port_gen_event gen_event;
297 
298 	for (i = 0; i < dev->num_slaves; i++)
299 		set_and_calc_slave_port_state(dev, i, port, event, &gen_event);
300 }
301 /**************************************************************************
302 	The function get as input the new event to that port,
303 	and according to the prev state change the slave's port state.
304 	The events are:
305 		MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
306 		MLX4_PORT_STATE_DEV_EVENT_PORT_UP
307 		MLX4_PORT_STATE_IB_EVENT_GID_VALID
308 		MLX4_PORT_STATE_IB_EVENT_GID_INVALID
309 ***************************************************************************/
310 int set_and_calc_slave_port_state(struct mlx4_dev *dev, int slave,
311 				  u8 port, int event,
312 				  enum slave_port_gen_event *gen_event)
313 {
314 	struct mlx4_priv *priv = mlx4_priv(dev);
315 	struct mlx4_slave_state *ctx = NULL;
316 	unsigned long flags;
317 	int ret = -1;
318 	enum slave_port_state cur_state =
319 		mlx4_get_slave_port_state(dev, slave, port);
320 
321 	*gen_event = SLAVE_PORT_GEN_EVENT_NONE;
322 
323 	if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS || port == 0) {
324 		pr_err("%s: Error: asking for slave:%d, port:%d\n",
325 		       __func__, slave, port);
326 		return ret;
327 	}
328 
329 	ctx = &priv->mfunc.master.slave_state[slave];
330 	spin_lock_irqsave(&ctx->lock, flags);
331 
332 	switch (cur_state) {
333 	case SLAVE_PORT_DOWN:
334 		if (MLX4_PORT_STATE_DEV_EVENT_PORT_UP == event)
335 			mlx4_set_slave_port_state(dev, slave, port,
336 						  SLAVE_PENDING_UP);
337 		break;
338 	case SLAVE_PENDING_UP:
339 		if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event)
340 			mlx4_set_slave_port_state(dev, slave, port,
341 						  SLAVE_PORT_DOWN);
342 		else if (MLX4_PORT_STATE_IB_PORT_STATE_EVENT_GID_VALID == event) {
343 			mlx4_set_slave_port_state(dev, slave, port,
344 						  SLAVE_PORT_UP);
345 			*gen_event = SLAVE_PORT_GEN_EVENT_UP;
346 		}
347 		break;
348 	case SLAVE_PORT_UP:
349 		if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event) {
350 			mlx4_set_slave_port_state(dev, slave, port,
351 						  SLAVE_PORT_DOWN);
352 			*gen_event = SLAVE_PORT_GEN_EVENT_DOWN;
353 		} else if (MLX4_PORT_STATE_IB_EVENT_GID_INVALID ==
354 				event) {
355 			mlx4_set_slave_port_state(dev, slave, port,
356 						  SLAVE_PENDING_UP);
357 			*gen_event = SLAVE_PORT_GEN_EVENT_DOWN;
358 		}
359 		break;
360 	default:
361 		pr_err("%s: BUG!!! UNKNOWN state: "
362 		       "slave:%d, port:%d\n", __func__, slave, port);
363 			goto out;
364 	}
365 	ret = mlx4_get_slave_port_state(dev, slave, port);
366 
367 out:
368 	spin_unlock_irqrestore(&ctx->lock, flags);
369 	return ret;
370 }
371 
372 EXPORT_SYMBOL(set_and_calc_slave_port_state);
373 
374 int mlx4_gen_slaves_port_mgt_ev(struct mlx4_dev *dev, u8 port, int attr)
375 {
376 	struct mlx4_eqe eqe;
377 
378 	memset(&eqe, 0, sizeof eqe);
379 
380 	eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
381 	eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PORT_INFO;
382 	eqe.event.port_mgmt_change.port = port;
383 	eqe.event.port_mgmt_change.params.port_info.changed_attr =
384 		cpu_to_be32((u32) attr);
385 
386 	slave_event(dev, ALL_SLAVES, &eqe);
387 	return 0;
388 }
389 EXPORT_SYMBOL(mlx4_gen_slaves_port_mgt_ev);
390 
391 void mlx4_master_handle_slave_flr(struct work_struct *work)
392 {
393 	struct mlx4_mfunc_master_ctx *master =
394 		container_of(work, struct mlx4_mfunc_master_ctx,
395 			     slave_flr_event_work);
396 	struct mlx4_mfunc *mfunc =
397 		container_of(master, struct mlx4_mfunc, master);
398 	struct mlx4_priv *priv =
399 		container_of(mfunc, struct mlx4_priv, mfunc);
400 	struct mlx4_dev *dev = &priv->dev;
401 	struct mlx4_slave_state *slave_state = priv->mfunc.master.slave_state;
402 	int i;
403 	int err;
404 
405 	mlx4_dbg(dev, "mlx4_handle_slave_flr\n");
406 
407 	for (i = 0 ; i < dev->num_slaves; i++) {
408 
409 		if (MLX4_COMM_CMD_FLR == slave_state[i].last_cmd) {
410 			mlx4_dbg(dev, "mlx4_handle_slave_flr: "
411 				 "clean slave: %d\n", i);
412 
413 			mlx4_delete_all_resources_for_slave(dev, i);
414 			/*return the slave to running mode*/
415 			spin_lock(&priv->mfunc.master.slave_state_lock);
416 			slave_state[i].last_cmd = MLX4_COMM_CMD_RESET;
417 			slave_state[i].is_slave_going_down = 0;
418 			spin_unlock(&priv->mfunc.master.slave_state_lock);
419 			/*notify the FW:*/
420 			err = mlx4_cmd(dev, 0, i, 0, MLX4_CMD_INFORM_FLR_DONE,
421 				       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
422 			if (err)
423 				mlx4_warn(dev, "Failed to notify FW on "
424 					  "FLR done (slave:%d)\n", i);
425 		}
426 	}
427 }
428 
429 static int mlx4_eq_int(struct mlx4_dev *dev, struct mlx4_eq *eq)
430 {
431 	struct mlx4_priv *priv = mlx4_priv(dev);
432 	struct mlx4_eqe *eqe;
433 	int cqn;
434 	int eqes_found = 0;
435 	int set_ci = 0;
436 	int port;
437 	int slave = 0;
438 	int ret;
439 	u32 flr_slave;
440 	u8 update_slave_state;
441 	int i;
442 	enum slave_port_gen_event gen_event;
443 
444 	while ((eqe = next_eqe_sw(eq))) {
445 		/*
446 		 * Make sure we read EQ entry contents after we've
447 		 * checked the ownership bit.
448 		 */
449 		rmb();
450 
451 		switch (eqe->type) {
452 		case MLX4_EVENT_TYPE_COMP:
453 			cqn = be32_to_cpu(eqe->event.comp.cqn) & 0xffffff;
454 			mlx4_cq_completion(dev, cqn);
455 			break;
456 
457 		case MLX4_EVENT_TYPE_PATH_MIG:
458 		case MLX4_EVENT_TYPE_COMM_EST:
459 		case MLX4_EVENT_TYPE_SQ_DRAINED:
460 		case MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE:
461 		case MLX4_EVENT_TYPE_WQ_CATAS_ERROR:
462 		case MLX4_EVENT_TYPE_PATH_MIG_FAILED:
463 		case MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
464 		case MLX4_EVENT_TYPE_WQ_ACCESS_ERROR:
465 			mlx4_dbg(dev, "event %d arrived\n", eqe->type);
466 			if (mlx4_is_master(dev)) {
467 				/* forward only to slave owning the QP */
468 				ret = mlx4_get_slave_from_resource_id(dev,
469 						RES_QP,
470 						be32_to_cpu(eqe->event.qp.qpn)
471 						& 0xffffff, &slave);
472 				if (ret && ret != -ENOENT) {
473 					mlx4_dbg(dev, "QP event %02x(%02x) on "
474 						 "EQ %d at index %u: could "
475 						 "not get slave id (%d)\n",
476 						 eqe->type, eqe->subtype,
477 						 eq->eqn, eq->cons_index, ret);
478 					break;
479 				}
480 
481 				if (!ret && slave != dev->caps.function) {
482 					mlx4_slave_event(dev, slave, eqe);
483 					break;
484 				}
485 
486 			}
487 			mlx4_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) &
488 				      0xffffff, eqe->type);
489 			break;
490 
491 		case MLX4_EVENT_TYPE_SRQ_LIMIT:
492 			mlx4_warn(dev, "%s: MLX4_EVENT_TYPE_SRQ_LIMIT\n",
493 				  __func__);
494 		case MLX4_EVENT_TYPE_SRQ_CATAS_ERROR:
495 			if (mlx4_is_master(dev)) {
496 				/* forward only to slave owning the SRQ */
497 				ret = mlx4_get_slave_from_resource_id(dev,
498 						RES_SRQ,
499 						be32_to_cpu(eqe->event.srq.srqn)
500 						& 0xffffff,
501 						&slave);
502 				if (ret && ret != -ENOENT) {
503 					mlx4_warn(dev, "SRQ event %02x(%02x) "
504 						  "on EQ %d at index %u: could"
505 						  " not get slave id (%d)\n",
506 						  eqe->type, eqe->subtype,
507 						  eq->eqn, eq->cons_index, ret);
508 					break;
509 				}
510 				mlx4_warn(dev, "%s: slave:%d, srq_no:0x%x,"
511 					  " event: %02x(%02x)\n", __func__,
512 					  slave,
513 					  be32_to_cpu(eqe->event.srq.srqn),
514 					  eqe->type, eqe->subtype);
515 
516 				if (!ret && slave != dev->caps.function) {
517 					mlx4_warn(dev, "%s: sending event "
518 						  "%02x(%02x) to slave:%d\n",
519 						   __func__, eqe->type,
520 						  eqe->subtype, slave);
521 					mlx4_slave_event(dev, slave, eqe);
522 					break;
523 				}
524 			}
525 			mlx4_srq_event(dev, be32_to_cpu(eqe->event.srq.srqn) &
526 				       0xffffff, eqe->type);
527 			break;
528 
529 		case MLX4_EVENT_TYPE_CMD:
530 			mlx4_cmd_event(dev,
531 				       be16_to_cpu(eqe->event.cmd.token),
532 				       eqe->event.cmd.status,
533 				       be64_to_cpu(eqe->event.cmd.out_param));
534 			break;
535 
536 		case MLX4_EVENT_TYPE_PORT_CHANGE:
537 			port = be32_to_cpu(eqe->event.port_change.port) >> 28;
538 			if (eqe->subtype == MLX4_PORT_CHANGE_SUBTYPE_DOWN) {
539 				mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_DOWN,
540 						    port);
541 				mlx4_priv(dev)->sense.do_sense_port[port] = 1;
542 				if (!mlx4_is_master(dev))
543 					break;
544 				for (i = 0; i < dev->num_slaves; i++) {
545 					if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH) {
546 						if (i == mlx4_master_func_num(dev))
547 							continue;
548 						mlx4_dbg(dev, "%s: Sending MLX4_PORT_CHANGE_SUBTYPE_DOWN"
549 							 " to slave: %d, port:%d\n",
550 							 __func__, i, port);
551 						mlx4_slave_event(dev, i, eqe);
552 					} else {  /* IB port */
553 						set_and_calc_slave_port_state(dev, i, port,
554 									      MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
555 									      &gen_event);
556 						/*we can be in pending state, then do not send port_down event*/
557 						if (SLAVE_PORT_GEN_EVENT_DOWN ==  gen_event) {
558 							if (i == mlx4_master_func_num(dev))
559 								continue;
560 							mlx4_slave_event(dev, i, eqe);
561 						}
562 					}
563 				}
564 			} else {
565 				mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_UP, port);
566 
567 				mlx4_priv(dev)->sense.do_sense_port[port] = 0;
568 
569 				if (!mlx4_is_master(dev))
570 					break;
571 				if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
572 					for (i = 0; i < dev->num_slaves; i++) {
573 						if (i == mlx4_master_func_num(dev))
574 							continue;
575 						mlx4_slave_event(dev, i, eqe);
576 					}
577 				else /* IB port */
578 					/* port-up event will be sent to a slave when the
579 					 * slave's alias-guid is set. This is done in alias_GUID.c
580 					 */
581 					set_all_slave_state(dev, port, MLX4_DEV_EVENT_PORT_UP);
582 			}
583 			break;
584 
585 		case MLX4_EVENT_TYPE_CQ_ERROR:
586 			mlx4_warn(dev, "CQ %s on CQN %06x\n",
587 				  eqe->event.cq_err.syndrome == 1 ?
588 				  "overrun" : "access violation",
589 				  be32_to_cpu(eqe->event.cq_err.cqn) & 0xffffff);
590 			if (mlx4_is_master(dev)) {
591 				ret = mlx4_get_slave_from_resource_id(dev,
592 					RES_CQ,
593 					be32_to_cpu(eqe->event.cq_err.cqn)
594 					& 0xffffff, &slave);
595 				if (ret && ret != -ENOENT) {
596 					mlx4_dbg(dev, "CQ event %02x(%02x) on "
597 						 "EQ %d at index %u: could "
598 						  "not get slave id (%d)\n",
599 						  eqe->type, eqe->subtype,
600 						  eq->eqn, eq->cons_index, ret);
601 					break;
602 				}
603 
604 				if (!ret && slave != dev->caps.function) {
605 					mlx4_slave_event(dev, slave, eqe);
606 					break;
607 				}
608 			}
609 			mlx4_cq_event(dev,
610 				      be32_to_cpu(eqe->event.cq_err.cqn)
611 				      & 0xffffff,
612 				      eqe->type);
613 			break;
614 
615 		case MLX4_EVENT_TYPE_EQ_OVERFLOW:
616 			mlx4_warn(dev, "EQ overrun on EQN %d\n", eq->eqn);
617 			break;
618 
619 		case MLX4_EVENT_TYPE_COMM_CHANNEL:
620 			if (!mlx4_is_master(dev)) {
621 				mlx4_warn(dev, "Received comm channel event "
622 					       "for non master device\n");
623 				break;
624 			}
625 			memcpy(&priv->mfunc.master.comm_arm_bit_vector,
626 			       eqe->event.comm_channel_arm.bit_vec,
627 			       sizeof eqe->event.comm_channel_arm.bit_vec);
628 			queue_work(priv->mfunc.master.comm_wq,
629 				   &priv->mfunc.master.comm_work);
630 			break;
631 
632 		case MLX4_EVENT_TYPE_FLR_EVENT:
633 			flr_slave = be32_to_cpu(eqe->event.flr_event.slave_id);
634 			if (!mlx4_is_master(dev)) {
635 				mlx4_warn(dev, "Non-master function received"
636 					       "FLR event\n");
637 				break;
638 			}
639 
640 			mlx4_dbg(dev, "FLR event for slave: %d\n", flr_slave);
641 
642 			if (flr_slave >= dev->num_slaves) {
643 				mlx4_warn(dev,
644 					  "Got FLR for unknown function: %d\n",
645 					  flr_slave);
646 				update_slave_state = 0;
647 			} else
648 				update_slave_state = 1;
649 
650 			spin_lock(&priv->mfunc.master.slave_state_lock);
651 			if (update_slave_state) {
652 				priv->mfunc.master.slave_state[flr_slave].active = false;
653 				priv->mfunc.master.slave_state[flr_slave].last_cmd = MLX4_COMM_CMD_FLR;
654 				priv->mfunc.master.slave_state[flr_slave].is_slave_going_down = 1;
655 			}
656 			spin_unlock(&priv->mfunc.master.slave_state_lock);
657 			queue_work(priv->mfunc.master.comm_wq,
658 				   &priv->mfunc.master.slave_flr_event_work);
659 			break;
660 
661 		case MLX4_EVENT_TYPE_FATAL_WARNING:
662 			if (eqe->subtype == MLX4_FATAL_WARNING_SUBTYPE_WARMING) {
663 				if (mlx4_is_master(dev))
664 					for (i = 0; i < dev->num_slaves; i++) {
665 						mlx4_dbg(dev, "%s: Sending "
666 							"MLX4_FATAL_WARNING_SUBTYPE_WARMING"
667 							" to slave: %d\n", __func__, i);
668 						if (i == dev->caps.function)
669 							continue;
670 						mlx4_slave_event(dev, i, eqe);
671 					}
672 				mlx4_err(dev, "Temperature Threshold was reached! "
673 					"Threshold: %d celsius degrees; "
674 					"Current Temperature: %d\n",
675 					be16_to_cpu(eqe->event.warming.warning_threshold),
676 					be16_to_cpu(eqe->event.warming.current_temperature));
677 			} else
678 				mlx4_warn(dev, "Unhandled event FATAL WARNING (%02x), "
679 					  "subtype %02x on EQ %d at index %u. owner=%x, "
680 					  "nent=0x%x, slave=%x, ownership=%s\n",
681 					  eqe->type, eqe->subtype, eq->eqn,
682 					  eq->cons_index, eqe->owner, eq->nent,
683 					  eqe->slave_id,
684 					  !!(eqe->owner & 0x80) ^
685 					  !!(eq->cons_index & eq->nent) ? "HW" : "SW");
686 
687 			break;
688 
689 		case MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT:
690 			mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_MGMT_CHANGE,
691 					    (unsigned long) eqe);
692 			break;
693 
694 		case MLX4_EVENT_TYPE_EEC_CATAS_ERROR:
695 		case MLX4_EVENT_TYPE_ECC_DETECT:
696 		default:
697 			mlx4_warn(dev, "Unhandled event %02x(%02x) on EQ %d at "
698 				  "index %u. owner=%x, nent=0x%x, slave=%x, "
699 				  "ownership=%s\n",
700 				  eqe->type, eqe->subtype, eq->eqn,
701 				  eq->cons_index, eqe->owner, eq->nent,
702 				  eqe->slave_id,
703 				  !!(eqe->owner & 0x80) ^
704 				  !!(eq->cons_index & eq->nent) ? "HW" : "SW");
705 			break;
706 		};
707 
708 		++eq->cons_index;
709 		eqes_found = 1;
710 		++set_ci;
711 
712 		/*
713 		 * The HCA will think the queue has overflowed if we
714 		 * don't tell it we've been processing events.  We
715 		 * create our EQs with MLX4_NUM_SPARE_EQE extra
716 		 * entries, so we must update our consumer index at
717 		 * least that often.
718 		 */
719 		if (unlikely(set_ci >= MLX4_NUM_SPARE_EQE)) {
720 			eq_set_ci(eq, 0);
721 			set_ci = 0;
722 		}
723 	}
724 
725 	eq_set_ci(eq, 1);
726 
727 	return eqes_found;
728 }
729 
730 static irqreturn_t mlx4_interrupt(int irq, void *dev_ptr)
731 {
732 	struct mlx4_dev *dev = dev_ptr;
733 	struct mlx4_priv *priv = mlx4_priv(dev);
734 	int work = 0;
735 	int i;
736 
737 	writel(priv->eq_table.clr_mask, priv->eq_table.clr_int);
738 
739 	for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
740 		work |= mlx4_eq_int(dev, &priv->eq_table.eq[i]);
741 
742 	return IRQ_RETVAL(work);
743 }
744 
745 static irqreturn_t mlx4_msi_x_interrupt(int irq, void *eq_ptr)
746 {
747 	struct mlx4_eq  *eq  = eq_ptr;
748 	struct mlx4_dev *dev = eq->dev;
749 
750 	mlx4_eq_int(dev, eq);
751 
752 	/* MSI-X vectors always belong to us */
753 	return IRQ_HANDLED;
754 }
755 
756 int mlx4_MAP_EQ_wrapper(struct mlx4_dev *dev, int slave,
757 			struct mlx4_vhcr *vhcr,
758 			struct mlx4_cmd_mailbox *inbox,
759 			struct mlx4_cmd_mailbox *outbox,
760 			struct mlx4_cmd_info *cmd)
761 {
762 	struct mlx4_priv *priv = mlx4_priv(dev);
763 	struct mlx4_slave_event_eq_info *event_eq =
764 		priv->mfunc.master.slave_state[slave].event_eq;
765 	u32 in_modifier = vhcr->in_modifier;
766 	u32 eqn = in_modifier & 0x1FF;
767 	u64 in_param =  vhcr->in_param;
768 	int err = 0;
769 	int i;
770 
771 	if (slave == dev->caps.function)
772 		err = mlx4_cmd(dev, in_param, (in_modifier & 0x80000000) | eqn,
773 			       0, MLX4_CMD_MAP_EQ, MLX4_CMD_TIME_CLASS_B,
774 			       MLX4_CMD_NATIVE);
775 	if (!err)
776 		for (i = 0; i < MLX4_EVENT_TYPES_NUM; ++i)
777 			if (in_param & (1LL << i))
778 				event_eq[i].eqn = in_modifier >> 31 ? -1 : eqn;
779 
780 	return err;
781 }
782 
783 static int mlx4_MAP_EQ(struct mlx4_dev *dev, u64 event_mask, int unmap,
784 			int eq_num)
785 {
786 	return mlx4_cmd(dev, event_mask, (unmap << 31) | eq_num,
787 			0, MLX4_CMD_MAP_EQ, MLX4_CMD_TIME_CLASS_B,
788 			MLX4_CMD_WRAPPED);
789 }
790 
791 static int mlx4_SW2HW_EQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
792 			 int eq_num)
793 {
794 	return mlx4_cmd(dev, mailbox->dma, eq_num, 0,
795 			MLX4_CMD_SW2HW_EQ, MLX4_CMD_TIME_CLASS_A,
796 			MLX4_CMD_WRAPPED);
797 }
798 
799 static int mlx4_HW2SW_EQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
800 			 int eq_num)
801 {
802 	return mlx4_cmd_box(dev, 0, mailbox->dma, eq_num,
803 			    0, MLX4_CMD_HW2SW_EQ, MLX4_CMD_TIME_CLASS_A,
804 			    MLX4_CMD_WRAPPED);
805 }
806 
807 static int mlx4_num_eq_uar(struct mlx4_dev *dev)
808 {
809 	/*
810 	 * Each UAR holds 4 EQ doorbells.  To figure out how many UARs
811 	 * we need to map, take the difference of highest index and
812 	 * the lowest index we'll use and add 1.
813 	 */
814 	return (dev->caps.num_comp_vectors + 1 + dev->caps.reserved_eqs +
815 		 dev->caps.comp_pool)/4 - dev->caps.reserved_eqs/4 + 1;
816 }
817 
818 static void __iomem *mlx4_get_eq_uar(struct mlx4_dev *dev, struct mlx4_eq *eq)
819 {
820 	struct mlx4_priv *priv = mlx4_priv(dev);
821 	int index;
822 
823 	index = eq->eqn / 4 - dev->caps.reserved_eqs / 4;
824 
825 	if (!priv->eq_table.uar_map[index]) {
826 		priv->eq_table.uar_map[index] =
827 			ioremap(pci_resource_start(dev->pdev, 2) +
828 				((eq->eqn / 4) << PAGE_SHIFT),
829 				PAGE_SIZE);
830 		if (!priv->eq_table.uar_map[index]) {
831 			mlx4_err(dev, "Couldn't map EQ doorbell for EQN 0x%06x\n",
832 				 eq->eqn);
833 			return NULL;
834 		}
835 	}
836 
837 	return priv->eq_table.uar_map[index] + 0x800 + 8 * (eq->eqn % 4);
838 }
839 
840 static void mlx4_unmap_uar(struct mlx4_dev *dev)
841 {
842 	struct mlx4_priv *priv = mlx4_priv(dev);
843 	int i;
844 
845 	for (i = 0; i < mlx4_num_eq_uar(dev); ++i)
846 		if (priv->eq_table.uar_map[i]) {
847 			iounmap(priv->eq_table.uar_map[i]);
848 			priv->eq_table.uar_map[i] = NULL;
849 		}
850 }
851 
852 static int mlx4_create_eq(struct mlx4_dev *dev, int nent,
853 			  u8 intr, struct mlx4_eq *eq)
854 {
855 	struct mlx4_priv *priv = mlx4_priv(dev);
856 	struct mlx4_cmd_mailbox *mailbox;
857 	struct mlx4_eq_context *eq_context;
858 	int npages;
859 	u64 *dma_list = NULL;
860 	dma_addr_t t;
861 	u64 mtt_addr;
862 	int err = -ENOMEM;
863 	int i;
864 
865 	eq->dev   = dev;
866 	eq->nent  = roundup_pow_of_two(max(nent, 2));
867 	npages = PAGE_ALIGN(eq->nent * MLX4_EQ_ENTRY_SIZE) / PAGE_SIZE;
868 
869 	eq->page_list = kmalloc(npages * sizeof *eq->page_list,
870 				GFP_KERNEL);
871 	if (!eq->page_list)
872 		goto err_out;
873 
874 	for (i = 0; i < npages; ++i)
875 		eq->page_list[i].buf = NULL;
876 
877 	dma_list = kmalloc(npages * sizeof *dma_list, GFP_KERNEL);
878 	if (!dma_list)
879 		goto err_out_free;
880 
881 	mailbox = mlx4_alloc_cmd_mailbox(dev);
882 	if (IS_ERR(mailbox))
883 		goto err_out_free;
884 	eq_context = mailbox->buf;
885 
886 	for (i = 0; i < npages; ++i) {
887 		eq->page_list[i].buf = dma_alloc_coherent(&dev->pdev->dev,
888 							  PAGE_SIZE, &t, GFP_KERNEL);
889 		if (!eq->page_list[i].buf)
890 			goto err_out_free_pages;
891 
892 		dma_list[i] = t;
893 		eq->page_list[i].map = t;
894 
895 		memset(eq->page_list[i].buf, 0, PAGE_SIZE);
896 	}
897 
898 	eq->eqn = mlx4_bitmap_alloc(&priv->eq_table.bitmap);
899 	if (eq->eqn == -1)
900 		goto err_out_free_pages;
901 
902 	eq->doorbell = mlx4_get_eq_uar(dev, eq);
903 	if (!eq->doorbell) {
904 		err = -ENOMEM;
905 		goto err_out_free_eq;
906 	}
907 
908 	err = mlx4_mtt_init(dev, npages, PAGE_SHIFT, &eq->mtt);
909 	if (err)
910 		goto err_out_free_eq;
911 
912 	err = mlx4_write_mtt(dev, &eq->mtt, 0, npages, dma_list);
913 	if (err)
914 		goto err_out_free_mtt;
915 
916 	memset(eq_context, 0, sizeof *eq_context);
917 	eq_context->flags	  = cpu_to_be32(MLX4_EQ_STATUS_OK   |
918 						MLX4_EQ_STATE_ARMED);
919 	eq_context->log_eq_size	  = ilog2(eq->nent);
920 	eq_context->intr	  = intr;
921 	eq_context->log_page_size = PAGE_SHIFT - MLX4_ICM_PAGE_SHIFT;
922 
923 	mtt_addr = mlx4_mtt_addr(dev, &eq->mtt);
924 	eq_context->mtt_base_addr_h = mtt_addr >> 32;
925 	eq_context->mtt_base_addr_l = cpu_to_be32(mtt_addr & 0xffffffff);
926 
927 	err = mlx4_SW2HW_EQ(dev, mailbox, eq->eqn);
928 	if (err) {
929 		mlx4_warn(dev, "SW2HW_EQ failed (%d)\n", err);
930 		goto err_out_free_mtt;
931 	}
932 
933 	kfree(dma_list);
934 	mlx4_free_cmd_mailbox(dev, mailbox);
935 
936 	eq->cons_index = 0;
937 
938 	return err;
939 
940 err_out_free_mtt:
941 	mlx4_mtt_cleanup(dev, &eq->mtt);
942 
943 err_out_free_eq:
944 	mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn);
945 
946 err_out_free_pages:
947 	for (i = 0; i < npages; ++i)
948 		if (eq->page_list[i].buf)
949 			dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
950 					  eq->page_list[i].buf,
951 					  eq->page_list[i].map);
952 
953 	mlx4_free_cmd_mailbox(dev, mailbox);
954 
955 err_out_free:
956 	kfree(eq->page_list);
957 	kfree(dma_list);
958 
959 err_out:
960 	return err;
961 }
962 
963 static void mlx4_free_eq(struct mlx4_dev *dev,
964 			 struct mlx4_eq *eq)
965 {
966 	struct mlx4_priv *priv = mlx4_priv(dev);
967 	struct mlx4_cmd_mailbox *mailbox;
968 	int err;
969 	int npages = PAGE_ALIGN(MLX4_EQ_ENTRY_SIZE * eq->nent) / PAGE_SIZE;
970 	int i;
971 
972 	mailbox = mlx4_alloc_cmd_mailbox(dev);
973 	if (IS_ERR(mailbox))
974 		return;
975 
976 	err = mlx4_HW2SW_EQ(dev, mailbox, eq->eqn);
977 	if (err)
978 		mlx4_warn(dev, "HW2SW_EQ failed (%d)\n", err);
979 
980 	if (0) {
981 		mlx4_dbg(dev, "Dumping EQ context %02x:\n", eq->eqn);
982 		for (i = 0; i < sizeof (struct mlx4_eq_context) / 4; ++i) {
983 			if (i % 4 == 0)
984 				pr_cont("[%02x] ", i * 4);
985 			pr_cont(" %08x", be32_to_cpup(mailbox->buf + i * 4));
986 			if ((i + 1) % 4 == 0)
987 				pr_cont("\n");
988 		}
989 	}
990 
991 	mlx4_mtt_cleanup(dev, &eq->mtt);
992 	for (i = 0; i < npages; ++i)
993 		dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
994 				    eq->page_list[i].buf,
995 				    eq->page_list[i].map);
996 
997 	kfree(eq->page_list);
998 	mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn);
999 	mlx4_free_cmd_mailbox(dev, mailbox);
1000 }
1001 
1002 static void mlx4_free_irqs(struct mlx4_dev *dev)
1003 {
1004 	struct mlx4_eq_table *eq_table = &mlx4_priv(dev)->eq_table;
1005 	struct mlx4_priv *priv = mlx4_priv(dev);
1006 	int	i, vec;
1007 
1008 	if (eq_table->have_irq)
1009 		free_irq(dev->pdev->irq, dev);
1010 
1011 	for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
1012 		if (eq_table->eq[i].have_irq) {
1013 			free_irq(eq_table->eq[i].irq, eq_table->eq + i);
1014 			eq_table->eq[i].have_irq = 0;
1015 		}
1016 
1017 	for (i = 0; i < dev->caps.comp_pool; i++) {
1018 		/*
1019 		 * Freeing the assigned irq's
1020 		 * all bits should be 0, but we need to validate
1021 		 */
1022 		if (priv->msix_ctl.pool_bm & 1ULL << i) {
1023 			/* NO need protecting*/
1024 			vec = dev->caps.num_comp_vectors + 1 + i;
1025 			free_irq(priv->eq_table.eq[vec].irq,
1026 				 &priv->eq_table.eq[vec]);
1027 		}
1028 	}
1029 
1030 
1031 	kfree(eq_table->irq_names);
1032 }
1033 
1034 static int mlx4_map_clr_int(struct mlx4_dev *dev)
1035 {
1036 	struct mlx4_priv *priv = mlx4_priv(dev);
1037 
1038 	priv->clr_base = ioremap(pci_resource_start(dev->pdev, priv->fw.clr_int_bar) +
1039 				 priv->fw.clr_int_base, MLX4_CLR_INT_SIZE);
1040 	if (!priv->clr_base) {
1041 		mlx4_err(dev, "Couldn't map interrupt clear register, aborting.\n");
1042 		return -ENOMEM;
1043 	}
1044 
1045 	return 0;
1046 }
1047 
1048 static void mlx4_unmap_clr_int(struct mlx4_dev *dev)
1049 {
1050 	struct mlx4_priv *priv = mlx4_priv(dev);
1051 
1052 	iounmap(priv->clr_base);
1053 }
1054 
1055 int mlx4_alloc_eq_table(struct mlx4_dev *dev)
1056 {
1057 	struct mlx4_priv *priv = mlx4_priv(dev);
1058 
1059 	priv->eq_table.eq = kcalloc(dev->caps.num_eqs - dev->caps.reserved_eqs,
1060 				    sizeof *priv->eq_table.eq, GFP_KERNEL);
1061 	if (!priv->eq_table.eq)
1062 		return -ENOMEM;
1063 
1064 	return 0;
1065 }
1066 
1067 void mlx4_free_eq_table(struct mlx4_dev *dev)
1068 {
1069 	kfree(mlx4_priv(dev)->eq_table.eq);
1070 }
1071 
1072 int mlx4_init_eq_table(struct mlx4_dev *dev)
1073 {
1074 	struct mlx4_priv *priv = mlx4_priv(dev);
1075 	int err;
1076 	int i;
1077 
1078 	priv->eq_table.uar_map = kcalloc(mlx4_num_eq_uar(dev),
1079 					 sizeof *priv->eq_table.uar_map,
1080 					 GFP_KERNEL);
1081 	if (!priv->eq_table.uar_map) {
1082 		err = -ENOMEM;
1083 		goto err_out_free;
1084 	}
1085 
1086 	err = mlx4_bitmap_init(&priv->eq_table.bitmap, dev->caps.num_eqs,
1087 			       dev->caps.num_eqs - 1, dev->caps.reserved_eqs, 0);
1088 	if (err)
1089 		goto err_out_free;
1090 
1091 	for (i = 0; i < mlx4_num_eq_uar(dev); ++i)
1092 		priv->eq_table.uar_map[i] = NULL;
1093 
1094 	if (!mlx4_is_slave(dev)) {
1095 		err = mlx4_map_clr_int(dev);
1096 		if (err)
1097 			goto err_out_bitmap;
1098 
1099 		priv->eq_table.clr_mask =
1100 			swab32(1 << (priv->eq_table.inta_pin & 31));
1101 		priv->eq_table.clr_int  = priv->clr_base +
1102 			(priv->eq_table.inta_pin < 32 ? 4 : 0);
1103 	}
1104 
1105 	priv->eq_table.irq_names =
1106 		kmalloc(MLX4_IRQNAME_SIZE * (dev->caps.num_comp_vectors + 1 +
1107 					     dev->caps.comp_pool),
1108 			GFP_KERNEL);
1109 	if (!priv->eq_table.irq_names) {
1110 		err = -ENOMEM;
1111 		goto err_out_bitmap;
1112 	}
1113 
1114 	for (i = 0; i < dev->caps.num_comp_vectors; ++i) {
1115 		err = mlx4_create_eq(dev, dev->caps.num_cqs -
1116 					  dev->caps.reserved_cqs +
1117 					  MLX4_NUM_SPARE_EQE,
1118 				     (dev->flags & MLX4_FLAG_MSI_X) ? i : 0,
1119 				     &priv->eq_table.eq[i]);
1120 		if (err) {
1121 			--i;
1122 			goto err_out_unmap;
1123 		}
1124 	}
1125 
1126 	err = mlx4_create_eq(dev, MLX4_NUM_ASYNC_EQE + MLX4_NUM_SPARE_EQE,
1127 			     (dev->flags & MLX4_FLAG_MSI_X) ? dev->caps.num_comp_vectors : 0,
1128 			     &priv->eq_table.eq[dev->caps.num_comp_vectors]);
1129 	if (err)
1130 		goto err_out_comp;
1131 
1132 	/*if additional completion vectors poolsize is 0 this loop will not run*/
1133 	for (i = dev->caps.num_comp_vectors + 1;
1134 	      i < dev->caps.num_comp_vectors + dev->caps.comp_pool + 1; ++i) {
1135 
1136 		err = mlx4_create_eq(dev, dev->caps.num_cqs -
1137 					  dev->caps.reserved_cqs +
1138 					  MLX4_NUM_SPARE_EQE,
1139 				     (dev->flags & MLX4_FLAG_MSI_X) ? i : 0,
1140 				     &priv->eq_table.eq[i]);
1141 		if (err) {
1142 			--i;
1143 			goto err_out_unmap;
1144 		}
1145 	}
1146 
1147 
1148 	if (dev->flags & MLX4_FLAG_MSI_X) {
1149 		const char *eq_name;
1150 
1151 		for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i) {
1152 			if (i < dev->caps.num_comp_vectors) {
1153 				snprintf(priv->eq_table.irq_names +
1154 					 i * MLX4_IRQNAME_SIZE,
1155 					 MLX4_IRQNAME_SIZE,
1156 					 "mlx4-comp-%d@pci:%s", i,
1157 					 pci_name(dev->pdev));
1158 			} else {
1159 				snprintf(priv->eq_table.irq_names +
1160 					 i * MLX4_IRQNAME_SIZE,
1161 					 MLX4_IRQNAME_SIZE,
1162 					 "mlx4-async@pci:%s",
1163 					 pci_name(dev->pdev));
1164 			}
1165 
1166 			eq_name = priv->eq_table.irq_names +
1167 				  i * MLX4_IRQNAME_SIZE;
1168 			err = request_irq(priv->eq_table.eq[i].irq,
1169 					  mlx4_msi_x_interrupt, 0, eq_name,
1170 					  priv->eq_table.eq + i);
1171 			if (err)
1172 				goto err_out_async;
1173 
1174 			priv->eq_table.eq[i].have_irq = 1;
1175 		}
1176 	} else {
1177 		snprintf(priv->eq_table.irq_names,
1178 			 MLX4_IRQNAME_SIZE,
1179 			 DRV_NAME "@pci:%s",
1180 			 pci_name(dev->pdev));
1181 		err = request_irq(dev->pdev->irq, mlx4_interrupt,
1182 				  IRQF_SHARED, priv->eq_table.irq_names, dev);
1183 		if (err)
1184 			goto err_out_async;
1185 
1186 		priv->eq_table.have_irq = 1;
1187 	}
1188 
1189 	err = mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1190 			  priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
1191 	if (err)
1192 		mlx4_warn(dev, "MAP_EQ for async EQ %d failed (%d)\n",
1193 			   priv->eq_table.eq[dev->caps.num_comp_vectors].eqn, err);
1194 
1195 	for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
1196 		eq_set_ci(&priv->eq_table.eq[i], 1);
1197 
1198 	return 0;
1199 
1200 err_out_async:
1201 	mlx4_free_eq(dev, &priv->eq_table.eq[dev->caps.num_comp_vectors]);
1202 
1203 err_out_comp:
1204 	i = dev->caps.num_comp_vectors - 1;
1205 
1206 err_out_unmap:
1207 	while (i >= 0) {
1208 		mlx4_free_eq(dev, &priv->eq_table.eq[i]);
1209 		--i;
1210 	}
1211 	if (!mlx4_is_slave(dev))
1212 		mlx4_unmap_clr_int(dev);
1213 	mlx4_free_irqs(dev);
1214 
1215 err_out_bitmap:
1216 	mlx4_unmap_uar(dev);
1217 	mlx4_bitmap_cleanup(&priv->eq_table.bitmap);
1218 
1219 err_out_free:
1220 	kfree(priv->eq_table.uar_map);
1221 
1222 	return err;
1223 }
1224 
1225 void mlx4_cleanup_eq_table(struct mlx4_dev *dev)
1226 {
1227 	struct mlx4_priv *priv = mlx4_priv(dev);
1228 	int i;
1229 
1230 	mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 1,
1231 		    priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
1232 
1233 	mlx4_free_irqs(dev);
1234 
1235 	for (i = 0; i < dev->caps.num_comp_vectors + dev->caps.comp_pool + 1; ++i)
1236 		mlx4_free_eq(dev, &priv->eq_table.eq[i]);
1237 
1238 	if (!mlx4_is_slave(dev))
1239 		mlx4_unmap_clr_int(dev);
1240 
1241 	mlx4_unmap_uar(dev);
1242 	mlx4_bitmap_cleanup(&priv->eq_table.bitmap);
1243 
1244 	kfree(priv->eq_table.uar_map);
1245 }
1246 
1247 /* A test that verifies that we can accept interrupts on all
1248  * the irq vectors of the device.
1249  * Interrupts are checked using the NOP command.
1250  */
1251 int mlx4_test_interrupts(struct mlx4_dev *dev)
1252 {
1253 	struct mlx4_priv *priv = mlx4_priv(dev);
1254 	int i;
1255 	int err;
1256 
1257 	err = mlx4_NOP(dev);
1258 	/* When not in MSI_X, there is only one irq to check */
1259 	if (!(dev->flags & MLX4_FLAG_MSI_X) || mlx4_is_slave(dev))
1260 		return err;
1261 
1262 	/* A loop over all completion vectors, for each vector we will check
1263 	 * whether it works by mapping command completions to that vector
1264 	 * and performing a NOP command
1265 	 */
1266 	for(i = 0; !err && (i < dev->caps.num_comp_vectors); ++i) {
1267 		/* Temporary use polling for command completions */
1268 		mlx4_cmd_use_polling(dev);
1269 
1270 		/* Map the new eq to handle all asyncronous events */
1271 		err = mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1272 				  priv->eq_table.eq[i].eqn);
1273 		if (err) {
1274 			mlx4_warn(dev, "Failed mapping eq for interrupt test\n");
1275 			mlx4_cmd_use_events(dev);
1276 			break;
1277 		}
1278 
1279 		/* Go back to using events */
1280 		mlx4_cmd_use_events(dev);
1281 		err = mlx4_NOP(dev);
1282 	}
1283 
1284 	/* Return to default */
1285 	mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1286 		    priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
1287 	return err;
1288 }
1289 EXPORT_SYMBOL(mlx4_test_interrupts);
1290 
1291 int mlx4_assign_eq(struct mlx4_dev *dev, char *name, struct cpu_rmap *rmap,
1292 		   int *vector)
1293 {
1294 
1295 	struct mlx4_priv *priv = mlx4_priv(dev);
1296 	int vec = 0, err = 0, i;
1297 
1298 	mutex_lock(&priv->msix_ctl.pool_lock);
1299 	for (i = 0; !vec && i < dev->caps.comp_pool; i++) {
1300 		if (~priv->msix_ctl.pool_bm & 1ULL << i) {
1301 			priv->msix_ctl.pool_bm |= 1ULL << i;
1302 			vec = dev->caps.num_comp_vectors + 1 + i;
1303 			snprintf(priv->eq_table.irq_names +
1304 					vec * MLX4_IRQNAME_SIZE,
1305 					MLX4_IRQNAME_SIZE, "%s", name);
1306 #ifdef CONFIG_RFS_ACCEL
1307 			if (rmap) {
1308 				err = irq_cpu_rmap_add(rmap,
1309 						       priv->eq_table.eq[vec].irq);
1310 				if (err)
1311 					mlx4_warn(dev, "Failed adding irq rmap\n");
1312 			}
1313 #endif
1314 			err = request_irq(priv->eq_table.eq[vec].irq,
1315 					  mlx4_msi_x_interrupt, 0,
1316 					  &priv->eq_table.irq_names[vec<<5],
1317 					  priv->eq_table.eq + vec);
1318 			if (err) {
1319 				/*zero out bit by fliping it*/
1320 				priv->msix_ctl.pool_bm ^= 1 << i;
1321 				vec = 0;
1322 				continue;
1323 				/*we dont want to break here*/
1324 			}
1325 			eq_set_ci(&priv->eq_table.eq[vec], 1);
1326 		}
1327 	}
1328 	mutex_unlock(&priv->msix_ctl.pool_lock);
1329 
1330 	if (vec) {
1331 		*vector = vec;
1332 	} else {
1333 		*vector = 0;
1334 		err = (i == dev->caps.comp_pool) ? -ENOSPC : err;
1335 	}
1336 	return err;
1337 }
1338 EXPORT_SYMBOL(mlx4_assign_eq);
1339 
1340 void mlx4_release_eq(struct mlx4_dev *dev, int vec)
1341 {
1342 	struct mlx4_priv *priv = mlx4_priv(dev);
1343 	/*bm index*/
1344 	int i = vec - dev->caps.num_comp_vectors - 1;
1345 
1346 	if (likely(i >= 0)) {
1347 		/*sanity check , making sure were not trying to free irq's
1348 		  Belonging to a legacy EQ*/
1349 		mutex_lock(&priv->msix_ctl.pool_lock);
1350 		if (priv->msix_ctl.pool_bm & 1ULL << i) {
1351 			free_irq(priv->eq_table.eq[vec].irq,
1352 				 &priv->eq_table.eq[vec]);
1353 			priv->msix_ctl.pool_bm &= ~(1ULL << i);
1354 		}
1355 		mutex_unlock(&priv->msix_ctl.pool_lock);
1356 	}
1357 
1358 }
1359 EXPORT_SYMBOL(mlx4_release_eq);
1360 
1361