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_OP_REQUIRED)	    | \
83 			       (1ull << MLX4_EVENT_TYPE_COMM_CHANNEL)       | \
84 			       (1ull << MLX4_EVENT_TYPE_FLR_EVENT)	    | \
85 			       (1ull << MLX4_EVENT_TYPE_FATAL_WARNING))
86 
87 static u64 get_async_ev_mask(struct mlx4_dev *dev)
88 {
89 	u64 async_ev_mask = MLX4_ASYNC_EVENT_MASK;
90 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_PORT_MNG_CHG_EV)
91 		async_ev_mask |= (1ull << MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT);
92 
93 	return async_ev_mask;
94 }
95 
96 static void eq_set_ci(struct mlx4_eq *eq, int req_not)
97 {
98 	__raw_writel((__force u32) cpu_to_be32((eq->cons_index & 0xffffff) |
99 					       req_not << 31),
100 		     eq->doorbell);
101 	/* We still want ordering, just not swabbing, so add a barrier */
102 	mb();
103 }
104 
105 static struct mlx4_eqe *get_eqe(struct mlx4_eq *eq, u32 entry, u8 eqe_factor)
106 {
107 	/* (entry & (eq->nent - 1)) gives us a cyclic array */
108 	unsigned long offset = (entry & (eq->nent - 1)) * (MLX4_EQ_ENTRY_SIZE << eqe_factor);
109 	/* CX3 is capable of extending the EQE from 32 to 64 bytes.
110 	 * When this feature is enabled, the first (in the lower addresses)
111 	 * 32 bytes in the 64 byte EQE are reserved and the next 32 bytes
112 	 * contain the legacy EQE information.
113 	 */
114 	return eq->page_list[offset / PAGE_SIZE].buf + (offset + (eqe_factor ? MLX4_EQ_ENTRY_SIZE : 0)) % PAGE_SIZE;
115 }
116 
117 static struct mlx4_eqe *next_eqe_sw(struct mlx4_eq *eq, u8 eqe_factor)
118 {
119 	struct mlx4_eqe *eqe = get_eqe(eq, eq->cons_index, eqe_factor);
120 	return !!(eqe->owner & 0x80) ^ !!(eq->cons_index & eq->nent) ? NULL : eqe;
121 }
122 
123 static struct mlx4_eqe *next_slave_event_eqe(struct mlx4_slave_event_eq *slave_eq)
124 {
125 	struct mlx4_eqe *eqe =
126 		&slave_eq->event_eqe[slave_eq->cons & (SLAVE_EVENT_EQ_SIZE - 1)];
127 	return (!!(eqe->owner & 0x80) ^
128 		!!(slave_eq->cons & SLAVE_EVENT_EQ_SIZE)) ?
129 		eqe : NULL;
130 }
131 
132 void mlx4_gen_slave_eqe(struct work_struct *work)
133 {
134 	struct mlx4_mfunc_master_ctx *master =
135 		container_of(work, struct mlx4_mfunc_master_ctx,
136 			     slave_event_work);
137 	struct mlx4_mfunc *mfunc =
138 		container_of(master, struct mlx4_mfunc, master);
139 	struct mlx4_priv *priv = container_of(mfunc, struct mlx4_priv, mfunc);
140 	struct mlx4_dev *dev = &priv->dev;
141 	struct mlx4_slave_event_eq *slave_eq = &mfunc->master.slave_eq;
142 	struct mlx4_eqe *eqe;
143 	u8 slave;
144 	int i;
145 
146 	for (eqe = next_slave_event_eqe(slave_eq); eqe;
147 	      eqe = next_slave_event_eqe(slave_eq)) {
148 		slave = eqe->slave_id;
149 
150 		/* All active slaves need to receive the event */
151 		if (slave == ALL_SLAVES) {
152 			for (i = 0; i < dev->num_slaves; i++) {
153 				if (i != dev->caps.function &&
154 				    master->slave_state[i].active)
155 					if (mlx4_GEN_EQE(dev, i, eqe))
156 						mlx4_warn(dev, "Failed to "
157 							  " generate event "
158 							  "for slave %d\n", i);
159 			}
160 		} else {
161 			if (mlx4_GEN_EQE(dev, slave, eqe))
162 				mlx4_warn(dev, "Failed to generate event "
163 					       "for slave %d\n", slave);
164 		}
165 		++slave_eq->cons;
166 	}
167 }
168 
169 
170 static void slave_event(struct mlx4_dev *dev, u8 slave, struct mlx4_eqe *eqe)
171 {
172 	struct mlx4_priv *priv = mlx4_priv(dev);
173 	struct mlx4_slave_event_eq *slave_eq = &priv->mfunc.master.slave_eq;
174 	struct mlx4_eqe *s_eqe;
175 	unsigned long flags;
176 
177 	spin_lock_irqsave(&slave_eq->event_lock, flags);
178 	s_eqe = &slave_eq->event_eqe[slave_eq->prod & (SLAVE_EVENT_EQ_SIZE - 1)];
179 	if ((!!(s_eqe->owner & 0x80)) ^
180 	    (!!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE))) {
181 		mlx4_warn(dev, "Master failed to generate an EQE for slave: %d. "
182 			  "No free EQE on slave events queue\n", slave);
183 		spin_unlock_irqrestore(&slave_eq->event_lock, flags);
184 		return;
185 	}
186 
187 	memcpy(s_eqe, eqe, dev->caps.eqe_size - 1);
188 	s_eqe->slave_id = slave;
189 	/* ensure all information is written before setting the ownersip bit */
190 	wmb();
191 	s_eqe->owner = !!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE) ? 0x0 : 0x80;
192 	++slave_eq->prod;
193 
194 	queue_work(priv->mfunc.master.comm_wq,
195 		   &priv->mfunc.master.slave_event_work);
196 	spin_unlock_irqrestore(&slave_eq->event_lock, flags);
197 }
198 
199 static void mlx4_slave_event(struct mlx4_dev *dev, int slave,
200 			     struct mlx4_eqe *eqe)
201 {
202 	struct mlx4_priv *priv = mlx4_priv(dev);
203 	struct mlx4_slave_state *s_slave =
204 		&priv->mfunc.master.slave_state[slave];
205 
206 	if (!s_slave->active) {
207 		/*mlx4_warn(dev, "Trying to pass event to inactive slave\n");*/
208 		return;
209 	}
210 
211 	slave_event(dev, slave, eqe);
212 }
213 
214 int mlx4_gen_pkey_eqe(struct mlx4_dev *dev, int slave, u8 port)
215 {
216 	struct mlx4_eqe eqe;
217 
218 	struct mlx4_priv *priv = mlx4_priv(dev);
219 	struct mlx4_slave_state *s_slave = &priv->mfunc.master.slave_state[slave];
220 
221 	if (!s_slave->active)
222 		return 0;
223 
224 	memset(&eqe, 0, sizeof eqe);
225 
226 	eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
227 	eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE;
228 	eqe.event.port_mgmt_change.port = port;
229 
230 	return mlx4_GEN_EQE(dev, slave, &eqe);
231 }
232 EXPORT_SYMBOL(mlx4_gen_pkey_eqe);
233 
234 int mlx4_gen_guid_change_eqe(struct mlx4_dev *dev, int slave, u8 port)
235 {
236 	struct mlx4_eqe eqe;
237 
238 	/*don't send if we don't have the that slave */
239 	if (dev->num_vfs < slave)
240 		return 0;
241 	memset(&eqe, 0, sizeof eqe);
242 
243 	eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
244 	eqe.subtype = MLX4_DEV_PMC_SUBTYPE_GUID_INFO;
245 	eqe.event.port_mgmt_change.port = port;
246 
247 	return mlx4_GEN_EQE(dev, slave, &eqe);
248 }
249 EXPORT_SYMBOL(mlx4_gen_guid_change_eqe);
250 
251 int mlx4_gen_port_state_change_eqe(struct mlx4_dev *dev, int slave, u8 port,
252 				   u8 port_subtype_change)
253 {
254 	struct mlx4_eqe eqe;
255 
256 	/*don't send if we don't have the that slave */
257 	if (dev->num_vfs < slave)
258 		return 0;
259 	memset(&eqe, 0, sizeof eqe);
260 
261 	eqe.type = MLX4_EVENT_TYPE_PORT_CHANGE;
262 	eqe.subtype = port_subtype_change;
263 	eqe.event.port_change.port = cpu_to_be32(port << 28);
264 
265 	mlx4_dbg(dev, "%s: sending: %d to slave: %d on port: %d\n", __func__,
266 		 port_subtype_change, slave, port);
267 	return mlx4_GEN_EQE(dev, slave, &eqe);
268 }
269 EXPORT_SYMBOL(mlx4_gen_port_state_change_eqe);
270 
271 enum slave_port_state mlx4_get_slave_port_state(struct mlx4_dev *dev, int slave, u8 port)
272 {
273 	struct mlx4_priv *priv = mlx4_priv(dev);
274 	struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state;
275 	if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS) {
276 		pr_err("%s: Error: asking for slave:%d, port:%d\n",
277 		       __func__, slave, port);
278 		return SLAVE_PORT_DOWN;
279 	}
280 	return s_state[slave].port_state[port];
281 }
282 EXPORT_SYMBOL(mlx4_get_slave_port_state);
283 
284 static int mlx4_set_slave_port_state(struct mlx4_dev *dev, int slave, u8 port,
285 				     enum slave_port_state state)
286 {
287 	struct mlx4_priv *priv = mlx4_priv(dev);
288 	struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state;
289 
290 	if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS || port == 0) {
291 		pr_err("%s: Error: asking for slave:%d, port:%d\n",
292 		       __func__, slave, port);
293 		return -1;
294 	}
295 	s_state[slave].port_state[port] = state;
296 
297 	return 0;
298 }
299 
300 static void set_all_slave_state(struct mlx4_dev *dev, u8 port, int event)
301 {
302 	int i;
303 	enum slave_port_gen_event gen_event;
304 
305 	for (i = 0; i < dev->num_slaves; i++)
306 		set_and_calc_slave_port_state(dev, i, port, event, &gen_event);
307 }
308 /**************************************************************************
309 	The function get as input the new event to that port,
310 	and according to the prev state change the slave's port state.
311 	The events are:
312 		MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
313 		MLX4_PORT_STATE_DEV_EVENT_PORT_UP
314 		MLX4_PORT_STATE_IB_EVENT_GID_VALID
315 		MLX4_PORT_STATE_IB_EVENT_GID_INVALID
316 ***************************************************************************/
317 int set_and_calc_slave_port_state(struct mlx4_dev *dev, int slave,
318 				  u8 port, int event,
319 				  enum slave_port_gen_event *gen_event)
320 {
321 	struct mlx4_priv *priv = mlx4_priv(dev);
322 	struct mlx4_slave_state *ctx = NULL;
323 	unsigned long flags;
324 	int ret = -1;
325 	enum slave_port_state cur_state =
326 		mlx4_get_slave_port_state(dev, slave, port);
327 
328 	*gen_event = SLAVE_PORT_GEN_EVENT_NONE;
329 
330 	if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS || port == 0) {
331 		pr_err("%s: Error: asking for slave:%d, port:%d\n",
332 		       __func__, slave, port);
333 		return ret;
334 	}
335 
336 	ctx = &priv->mfunc.master.slave_state[slave];
337 	spin_lock_irqsave(&ctx->lock, flags);
338 
339 	switch (cur_state) {
340 	case SLAVE_PORT_DOWN:
341 		if (MLX4_PORT_STATE_DEV_EVENT_PORT_UP == event)
342 			mlx4_set_slave_port_state(dev, slave, port,
343 						  SLAVE_PENDING_UP);
344 		break;
345 	case SLAVE_PENDING_UP:
346 		if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event)
347 			mlx4_set_slave_port_state(dev, slave, port,
348 						  SLAVE_PORT_DOWN);
349 		else if (MLX4_PORT_STATE_IB_PORT_STATE_EVENT_GID_VALID == event) {
350 			mlx4_set_slave_port_state(dev, slave, port,
351 						  SLAVE_PORT_UP);
352 			*gen_event = SLAVE_PORT_GEN_EVENT_UP;
353 		}
354 		break;
355 	case SLAVE_PORT_UP:
356 		if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event) {
357 			mlx4_set_slave_port_state(dev, slave, port,
358 						  SLAVE_PORT_DOWN);
359 			*gen_event = SLAVE_PORT_GEN_EVENT_DOWN;
360 		} else if (MLX4_PORT_STATE_IB_EVENT_GID_INVALID ==
361 				event) {
362 			mlx4_set_slave_port_state(dev, slave, port,
363 						  SLAVE_PENDING_UP);
364 			*gen_event = SLAVE_PORT_GEN_EVENT_DOWN;
365 		}
366 		break;
367 	default:
368 		pr_err("%s: BUG!!! UNKNOWN state: "
369 		       "slave:%d, port:%d\n", __func__, slave, port);
370 			goto out;
371 	}
372 	ret = mlx4_get_slave_port_state(dev, slave, port);
373 
374 out:
375 	spin_unlock_irqrestore(&ctx->lock, flags);
376 	return ret;
377 }
378 
379 EXPORT_SYMBOL(set_and_calc_slave_port_state);
380 
381 int mlx4_gen_slaves_port_mgt_ev(struct mlx4_dev *dev, u8 port, int attr)
382 {
383 	struct mlx4_eqe eqe;
384 
385 	memset(&eqe, 0, sizeof eqe);
386 
387 	eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
388 	eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PORT_INFO;
389 	eqe.event.port_mgmt_change.port = port;
390 	eqe.event.port_mgmt_change.params.port_info.changed_attr =
391 		cpu_to_be32((u32) attr);
392 
393 	slave_event(dev, ALL_SLAVES, &eqe);
394 	return 0;
395 }
396 EXPORT_SYMBOL(mlx4_gen_slaves_port_mgt_ev);
397 
398 void mlx4_master_handle_slave_flr(struct work_struct *work)
399 {
400 	struct mlx4_mfunc_master_ctx *master =
401 		container_of(work, struct mlx4_mfunc_master_ctx,
402 			     slave_flr_event_work);
403 	struct mlx4_mfunc *mfunc =
404 		container_of(master, struct mlx4_mfunc, master);
405 	struct mlx4_priv *priv =
406 		container_of(mfunc, struct mlx4_priv, mfunc);
407 	struct mlx4_dev *dev = &priv->dev;
408 	struct mlx4_slave_state *slave_state = priv->mfunc.master.slave_state;
409 	int i;
410 	int err;
411 	unsigned long flags;
412 
413 	mlx4_dbg(dev, "mlx4_handle_slave_flr\n");
414 
415 	for (i = 0 ; i < dev->num_slaves; i++) {
416 
417 		if (MLX4_COMM_CMD_FLR == slave_state[i].last_cmd) {
418 			mlx4_dbg(dev, "mlx4_handle_slave_flr: "
419 				 "clean slave: %d\n", i);
420 
421 			mlx4_delete_all_resources_for_slave(dev, i);
422 			/*return the slave to running mode*/
423 			spin_lock_irqsave(&priv->mfunc.master.slave_state_lock, flags);
424 			slave_state[i].last_cmd = MLX4_COMM_CMD_RESET;
425 			slave_state[i].is_slave_going_down = 0;
426 			spin_unlock_irqrestore(&priv->mfunc.master.slave_state_lock, flags);
427 			/*notify the FW:*/
428 			err = mlx4_cmd(dev, 0, i, 0, MLX4_CMD_INFORM_FLR_DONE,
429 				       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
430 			if (err)
431 				mlx4_warn(dev, "Failed to notify FW on "
432 					  "FLR done (slave:%d)\n", i);
433 		}
434 	}
435 }
436 
437 static int mlx4_eq_int(struct mlx4_dev *dev, struct mlx4_eq *eq)
438 {
439 	struct mlx4_priv *priv = mlx4_priv(dev);
440 	struct mlx4_eqe *eqe;
441 	int cqn;
442 	int eqes_found = 0;
443 	int set_ci = 0;
444 	int port;
445 	int slave = 0;
446 	int ret;
447 	u32 flr_slave;
448 	u8 update_slave_state;
449 	int i;
450 	enum slave_port_gen_event gen_event;
451 	unsigned long flags;
452 	struct mlx4_vport_state *s_info;
453 
454 	while ((eqe = next_eqe_sw(eq, dev->caps.eqe_factor))) {
455 		/*
456 		 * Make sure we read EQ entry contents after we've
457 		 * checked the ownership bit.
458 		 */
459 		rmb();
460 
461 		switch (eqe->type) {
462 		case MLX4_EVENT_TYPE_COMP:
463 			cqn = be32_to_cpu(eqe->event.comp.cqn) & 0xffffff;
464 			mlx4_cq_completion(dev, cqn);
465 			break;
466 
467 		case MLX4_EVENT_TYPE_PATH_MIG:
468 		case MLX4_EVENT_TYPE_COMM_EST:
469 		case MLX4_EVENT_TYPE_SQ_DRAINED:
470 		case MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE:
471 		case MLX4_EVENT_TYPE_WQ_CATAS_ERROR:
472 		case MLX4_EVENT_TYPE_PATH_MIG_FAILED:
473 		case MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
474 		case MLX4_EVENT_TYPE_WQ_ACCESS_ERROR:
475 			mlx4_dbg(dev, "event %d arrived\n", eqe->type);
476 			if (mlx4_is_master(dev)) {
477 				/* forward only to slave owning the QP */
478 				ret = mlx4_get_slave_from_resource_id(dev,
479 						RES_QP,
480 						be32_to_cpu(eqe->event.qp.qpn)
481 						& 0xffffff, &slave);
482 				if (ret && ret != -ENOENT) {
483 					mlx4_dbg(dev, "QP event %02x(%02x) on "
484 						 "EQ %d at index %u: could "
485 						 "not get slave id (%d)\n",
486 						 eqe->type, eqe->subtype,
487 						 eq->eqn, eq->cons_index, ret);
488 					break;
489 				}
490 
491 				if (!ret && slave != dev->caps.function) {
492 					mlx4_slave_event(dev, slave, eqe);
493 					break;
494 				}
495 
496 			}
497 			mlx4_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) &
498 				      0xffffff, eqe->type);
499 			break;
500 
501 		case MLX4_EVENT_TYPE_SRQ_LIMIT:
502 			mlx4_dbg(dev, "%s: MLX4_EVENT_TYPE_SRQ_LIMIT\n",
503 				 __func__);
504 		case MLX4_EVENT_TYPE_SRQ_CATAS_ERROR:
505 			if (mlx4_is_master(dev)) {
506 				/* forward only to slave owning the SRQ */
507 				ret = mlx4_get_slave_from_resource_id(dev,
508 						RES_SRQ,
509 						be32_to_cpu(eqe->event.srq.srqn)
510 						& 0xffffff,
511 						&slave);
512 				if (ret && ret != -ENOENT) {
513 					mlx4_warn(dev, "SRQ event %02x(%02x) "
514 						  "on EQ %d at index %u: could"
515 						  " not get slave id (%d)\n",
516 						  eqe->type, eqe->subtype,
517 						  eq->eqn, eq->cons_index, ret);
518 					break;
519 				}
520 				mlx4_warn(dev, "%s: slave:%d, srq_no:0x%x,"
521 					  " event: %02x(%02x)\n", __func__,
522 					  slave,
523 					  be32_to_cpu(eqe->event.srq.srqn),
524 					  eqe->type, eqe->subtype);
525 
526 				if (!ret && slave != dev->caps.function) {
527 					mlx4_warn(dev, "%s: sending event "
528 						  "%02x(%02x) to slave:%d\n",
529 						   __func__, eqe->type,
530 						  eqe->subtype, slave);
531 					mlx4_slave_event(dev, slave, eqe);
532 					break;
533 				}
534 			}
535 			mlx4_srq_event(dev, be32_to_cpu(eqe->event.srq.srqn) &
536 				       0xffffff, eqe->type);
537 			break;
538 
539 		case MLX4_EVENT_TYPE_CMD:
540 			mlx4_cmd_event(dev,
541 				       be16_to_cpu(eqe->event.cmd.token),
542 				       eqe->event.cmd.status,
543 				       be64_to_cpu(eqe->event.cmd.out_param));
544 			break;
545 
546 		case MLX4_EVENT_TYPE_PORT_CHANGE:
547 			port = be32_to_cpu(eqe->event.port_change.port) >> 28;
548 			if (eqe->subtype == MLX4_PORT_CHANGE_SUBTYPE_DOWN) {
549 				mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_DOWN,
550 						    port);
551 				mlx4_priv(dev)->sense.do_sense_port[port] = 1;
552 				if (!mlx4_is_master(dev))
553 					break;
554 				for (i = 0; i < dev->num_slaves; i++) {
555 					if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH) {
556 						if (i == mlx4_master_func_num(dev))
557 							continue;
558 						mlx4_dbg(dev, "%s: Sending MLX4_PORT_CHANGE_SUBTYPE_DOWN"
559 							 " to slave: %d, port:%d\n",
560 							 __func__, i, port);
561 						s_info = &priv->mfunc.master.vf_oper[slave].vport[port].state;
562 						if (IFLA_VF_LINK_STATE_AUTO == s_info->link_state)
563 							mlx4_slave_event(dev, i, eqe);
564 					} else {  /* IB port */
565 						set_and_calc_slave_port_state(dev, i, port,
566 									      MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
567 									      &gen_event);
568 						/*we can be in pending state, then do not send port_down event*/
569 						if (SLAVE_PORT_GEN_EVENT_DOWN ==  gen_event) {
570 							if (i == mlx4_master_func_num(dev))
571 								continue;
572 							mlx4_slave_event(dev, i, eqe);
573 						}
574 					}
575 				}
576 			} else {
577 				mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_UP, port);
578 
579 				mlx4_priv(dev)->sense.do_sense_port[port] = 0;
580 
581 				if (!mlx4_is_master(dev))
582 					break;
583 				if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
584 					for (i = 0; i < dev->num_slaves; i++) {
585 						if (i == mlx4_master_func_num(dev))
586 							continue;
587 						s_info = &priv->mfunc.master.vf_oper[slave].vport[port].state;
588 						if (IFLA_VF_LINK_STATE_AUTO == s_info->link_state)
589 							mlx4_slave_event(dev, i, eqe);
590 					}
591 				else /* IB port */
592 					/* port-up event will be sent to a slave when the
593 					 * slave's alias-guid is set. This is done in alias_GUID.c
594 					 */
595 					set_all_slave_state(dev, port, MLX4_DEV_EVENT_PORT_UP);
596 			}
597 			break;
598 
599 		case MLX4_EVENT_TYPE_CQ_ERROR:
600 			mlx4_warn(dev, "CQ %s on CQN %06x\n",
601 				  eqe->event.cq_err.syndrome == 1 ?
602 				  "overrun" : "access violation",
603 				  be32_to_cpu(eqe->event.cq_err.cqn) & 0xffffff);
604 			if (mlx4_is_master(dev)) {
605 				ret = mlx4_get_slave_from_resource_id(dev,
606 					RES_CQ,
607 					be32_to_cpu(eqe->event.cq_err.cqn)
608 					& 0xffffff, &slave);
609 				if (ret && ret != -ENOENT) {
610 					mlx4_dbg(dev, "CQ event %02x(%02x) on "
611 						 "EQ %d at index %u: could "
612 						  "not get slave id (%d)\n",
613 						  eqe->type, eqe->subtype,
614 						  eq->eqn, eq->cons_index, ret);
615 					break;
616 				}
617 
618 				if (!ret && slave != dev->caps.function) {
619 					mlx4_slave_event(dev, slave, eqe);
620 					break;
621 				}
622 			}
623 			mlx4_cq_event(dev,
624 				      be32_to_cpu(eqe->event.cq_err.cqn)
625 				      & 0xffffff,
626 				      eqe->type);
627 			break;
628 
629 		case MLX4_EVENT_TYPE_EQ_OVERFLOW:
630 			mlx4_warn(dev, "EQ overrun on EQN %d\n", eq->eqn);
631 			break;
632 
633 		case MLX4_EVENT_TYPE_OP_REQUIRED:
634 			atomic_inc(&priv->opreq_count);
635 			/* FW commands can't be executed from interrupt context
636 			 * working in deferred task
637 			 */
638 			queue_work(mlx4_wq, &priv->opreq_task);
639 			break;
640 
641 		case MLX4_EVENT_TYPE_COMM_CHANNEL:
642 			if (!mlx4_is_master(dev)) {
643 				mlx4_warn(dev, "Received comm channel event "
644 					       "for non master device\n");
645 				break;
646 			}
647 			memcpy(&priv->mfunc.master.comm_arm_bit_vector,
648 			       eqe->event.comm_channel_arm.bit_vec,
649 			       sizeof eqe->event.comm_channel_arm.bit_vec);
650 			queue_work(priv->mfunc.master.comm_wq,
651 				   &priv->mfunc.master.comm_work);
652 			break;
653 
654 		case MLX4_EVENT_TYPE_FLR_EVENT:
655 			flr_slave = be32_to_cpu(eqe->event.flr_event.slave_id);
656 			if (!mlx4_is_master(dev)) {
657 				mlx4_warn(dev, "Non-master function received"
658 					       "FLR event\n");
659 				break;
660 			}
661 
662 			mlx4_dbg(dev, "FLR event for slave: %d\n", flr_slave);
663 
664 			if (flr_slave >= dev->num_slaves) {
665 				mlx4_warn(dev,
666 					  "Got FLR for unknown function: %d\n",
667 					  flr_slave);
668 				update_slave_state = 0;
669 			} else
670 				update_slave_state = 1;
671 
672 			spin_lock_irqsave(&priv->mfunc.master.slave_state_lock, flags);
673 			if (update_slave_state) {
674 				priv->mfunc.master.slave_state[flr_slave].active = false;
675 				priv->mfunc.master.slave_state[flr_slave].last_cmd = MLX4_COMM_CMD_FLR;
676 				priv->mfunc.master.slave_state[flr_slave].is_slave_going_down = 1;
677 			}
678 			spin_unlock_irqrestore(&priv->mfunc.master.slave_state_lock, flags);
679 			queue_work(priv->mfunc.master.comm_wq,
680 				   &priv->mfunc.master.slave_flr_event_work);
681 			break;
682 
683 		case MLX4_EVENT_TYPE_FATAL_WARNING:
684 			if (eqe->subtype == MLX4_FATAL_WARNING_SUBTYPE_WARMING) {
685 				if (mlx4_is_master(dev))
686 					for (i = 0; i < dev->num_slaves; i++) {
687 						mlx4_dbg(dev, "%s: Sending "
688 							"MLX4_FATAL_WARNING_SUBTYPE_WARMING"
689 							" to slave: %d\n", __func__, i);
690 						if (i == dev->caps.function)
691 							continue;
692 						mlx4_slave_event(dev, i, eqe);
693 					}
694 				mlx4_err(dev, "Temperature Threshold was reached! "
695 					"Threshold: %d celsius degrees; "
696 					"Current Temperature: %d\n",
697 					be16_to_cpu(eqe->event.warming.warning_threshold),
698 					be16_to_cpu(eqe->event.warming.current_temperature));
699 			} else
700 				mlx4_warn(dev, "Unhandled event FATAL WARNING (%02x), "
701 					  "subtype %02x on EQ %d at index %u. owner=%x, "
702 					  "nent=0x%x, slave=%x, ownership=%s\n",
703 					  eqe->type, eqe->subtype, eq->eqn,
704 					  eq->cons_index, eqe->owner, eq->nent,
705 					  eqe->slave_id,
706 					  !!(eqe->owner & 0x80) ^
707 					  !!(eq->cons_index & eq->nent) ? "HW" : "SW");
708 
709 			break;
710 
711 		case MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT:
712 			mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_MGMT_CHANGE,
713 					    (unsigned long) eqe);
714 			break;
715 
716 		case MLX4_EVENT_TYPE_EEC_CATAS_ERROR:
717 		case MLX4_EVENT_TYPE_ECC_DETECT:
718 		default:
719 			mlx4_warn(dev, "Unhandled event %02x(%02x) on EQ %d at "
720 				  "index %u. owner=%x, nent=0x%x, slave=%x, "
721 				  "ownership=%s\n",
722 				  eqe->type, eqe->subtype, eq->eqn,
723 				  eq->cons_index, eqe->owner, eq->nent,
724 				  eqe->slave_id,
725 				  !!(eqe->owner & 0x80) ^
726 				  !!(eq->cons_index & eq->nent) ? "HW" : "SW");
727 			break;
728 		};
729 
730 		++eq->cons_index;
731 		eqes_found = 1;
732 		++set_ci;
733 
734 		/*
735 		 * The HCA will think the queue has overflowed if we
736 		 * don't tell it we've been processing events.  We
737 		 * create our EQs with MLX4_NUM_SPARE_EQE extra
738 		 * entries, so we must update our consumer index at
739 		 * least that often.
740 		 */
741 		if (unlikely(set_ci >= MLX4_NUM_SPARE_EQE)) {
742 			eq_set_ci(eq, 0);
743 			set_ci = 0;
744 		}
745 	}
746 
747 	eq_set_ci(eq, 1);
748 
749 	return eqes_found;
750 }
751 
752 static irqreturn_t mlx4_interrupt(int irq, void *dev_ptr)
753 {
754 	struct mlx4_dev *dev = dev_ptr;
755 	struct mlx4_priv *priv = mlx4_priv(dev);
756 	int work = 0;
757 	int i;
758 
759 	writel(priv->eq_table.clr_mask, priv->eq_table.clr_int);
760 
761 	for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
762 		work |= mlx4_eq_int(dev, &priv->eq_table.eq[i]);
763 
764 	return IRQ_RETVAL(work);
765 }
766 
767 static irqreturn_t mlx4_msi_x_interrupt(int irq, void *eq_ptr)
768 {
769 	struct mlx4_eq  *eq  = eq_ptr;
770 	struct mlx4_dev *dev = eq->dev;
771 
772 	mlx4_eq_int(dev, eq);
773 
774 	/* MSI-X vectors always belong to us */
775 	return IRQ_HANDLED;
776 }
777 
778 int mlx4_MAP_EQ_wrapper(struct mlx4_dev *dev, int slave,
779 			struct mlx4_vhcr *vhcr,
780 			struct mlx4_cmd_mailbox *inbox,
781 			struct mlx4_cmd_mailbox *outbox,
782 			struct mlx4_cmd_info *cmd)
783 {
784 	struct mlx4_priv *priv = mlx4_priv(dev);
785 	struct mlx4_slave_event_eq_info *event_eq =
786 		priv->mfunc.master.slave_state[slave].event_eq;
787 	u32 in_modifier = vhcr->in_modifier;
788 	u32 eqn = in_modifier & 0x3FF;
789 	u64 in_param =  vhcr->in_param;
790 	int err = 0;
791 	int i;
792 
793 	if (slave == dev->caps.function)
794 		err = mlx4_cmd(dev, in_param, (in_modifier & 0x80000000) | eqn,
795 			       0, MLX4_CMD_MAP_EQ, MLX4_CMD_TIME_CLASS_B,
796 			       MLX4_CMD_NATIVE);
797 	if (!err)
798 		for (i = 0; i < MLX4_EVENT_TYPES_NUM; ++i)
799 			if (in_param & (1LL << i))
800 				event_eq[i].eqn = in_modifier >> 31 ? -1 : eqn;
801 
802 	return err;
803 }
804 
805 static int mlx4_MAP_EQ(struct mlx4_dev *dev, u64 event_mask, int unmap,
806 			int eq_num)
807 {
808 	return mlx4_cmd(dev, event_mask, (unmap << 31) | eq_num,
809 			0, MLX4_CMD_MAP_EQ, MLX4_CMD_TIME_CLASS_B,
810 			MLX4_CMD_WRAPPED);
811 }
812 
813 static int mlx4_SW2HW_EQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
814 			 int eq_num)
815 {
816 	return mlx4_cmd(dev, mailbox->dma, eq_num, 0,
817 			MLX4_CMD_SW2HW_EQ, MLX4_CMD_TIME_CLASS_A,
818 			MLX4_CMD_WRAPPED);
819 }
820 
821 static int mlx4_HW2SW_EQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
822 			 int eq_num)
823 {
824 	return mlx4_cmd_box(dev, 0, mailbox->dma, eq_num,
825 			    0, MLX4_CMD_HW2SW_EQ, MLX4_CMD_TIME_CLASS_A,
826 			    MLX4_CMD_WRAPPED);
827 }
828 
829 static int mlx4_num_eq_uar(struct mlx4_dev *dev)
830 {
831 	/*
832 	 * Each UAR holds 4 EQ doorbells.  To figure out how many UARs
833 	 * we need to map, take the difference of highest index and
834 	 * the lowest index we'll use and add 1.
835 	 */
836 	return (dev->caps.num_comp_vectors + 1 + dev->caps.reserved_eqs +
837 		 dev->caps.comp_pool)/4 - dev->caps.reserved_eqs/4 + 1;
838 }
839 
840 static void __iomem *mlx4_get_eq_uar(struct mlx4_dev *dev, struct mlx4_eq *eq)
841 {
842 	struct mlx4_priv *priv = mlx4_priv(dev);
843 	int index;
844 
845 	index = eq->eqn / 4 - dev->caps.reserved_eqs / 4;
846 
847 	if (!priv->eq_table.uar_map[index]) {
848 		priv->eq_table.uar_map[index] =
849 			ioremap(pci_resource_start(dev->pdev, 2) +
850 				((eq->eqn / 4) << PAGE_SHIFT),
851 				PAGE_SIZE);
852 		if (!priv->eq_table.uar_map[index]) {
853 			mlx4_err(dev, "Couldn't map EQ doorbell for EQN 0x%06x\n",
854 				 eq->eqn);
855 			return NULL;
856 		}
857 	}
858 
859 	return priv->eq_table.uar_map[index] + 0x800 + 8 * (eq->eqn % 4);
860 }
861 
862 static void mlx4_unmap_uar(struct mlx4_dev *dev)
863 {
864 	struct mlx4_priv *priv = mlx4_priv(dev);
865 	int i;
866 
867 	for (i = 0; i < mlx4_num_eq_uar(dev); ++i)
868 		if (priv->eq_table.uar_map[i]) {
869 			iounmap(priv->eq_table.uar_map[i]);
870 			priv->eq_table.uar_map[i] = NULL;
871 		}
872 }
873 
874 static int mlx4_create_eq(struct mlx4_dev *dev, int nent,
875 			  u8 intr, struct mlx4_eq *eq)
876 {
877 	struct mlx4_priv *priv = mlx4_priv(dev);
878 	struct mlx4_cmd_mailbox *mailbox;
879 	struct mlx4_eq_context *eq_context;
880 	int npages;
881 	u64 *dma_list = NULL;
882 	dma_addr_t t;
883 	u64 mtt_addr;
884 	int err = -ENOMEM;
885 	int i;
886 
887 	eq->dev   = dev;
888 	eq->nent  = roundup_pow_of_two(max(nent, 2));
889 	/* CX3 is capable of extending the CQE/EQE from 32 to 64 bytes */
890 	npages = PAGE_ALIGN(eq->nent * (MLX4_EQ_ENTRY_SIZE << dev->caps.eqe_factor)) / PAGE_SIZE;
891 
892 	eq->page_list = kmalloc(npages * sizeof *eq->page_list,
893 				GFP_KERNEL);
894 	if (!eq->page_list)
895 		goto err_out;
896 
897 	for (i = 0; i < npages; ++i)
898 		eq->page_list[i].buf = NULL;
899 
900 	dma_list = kmalloc(npages * sizeof *dma_list, GFP_KERNEL);
901 	if (!dma_list)
902 		goto err_out_free;
903 
904 	mailbox = mlx4_alloc_cmd_mailbox(dev);
905 	if (IS_ERR(mailbox))
906 		goto err_out_free;
907 	eq_context = mailbox->buf;
908 
909 	for (i = 0; i < npages; ++i) {
910 		eq->page_list[i].buf = dma_alloc_coherent(&dev->pdev->dev,
911 							  PAGE_SIZE, &t, GFP_KERNEL);
912 		if (!eq->page_list[i].buf)
913 			goto err_out_free_pages;
914 
915 		dma_list[i] = t;
916 		eq->page_list[i].map = t;
917 
918 		memset(eq->page_list[i].buf, 0, PAGE_SIZE);
919 	}
920 
921 	eq->eqn = mlx4_bitmap_alloc(&priv->eq_table.bitmap);
922 	if (eq->eqn == -1)
923 		goto err_out_free_pages;
924 
925 	eq->doorbell = mlx4_get_eq_uar(dev, eq);
926 	if (!eq->doorbell) {
927 		err = -ENOMEM;
928 		goto err_out_free_eq;
929 	}
930 
931 	err = mlx4_mtt_init(dev, npages, PAGE_SHIFT, &eq->mtt);
932 	if (err)
933 		goto err_out_free_eq;
934 
935 	err = mlx4_write_mtt(dev, &eq->mtt, 0, npages, dma_list);
936 	if (err)
937 		goto err_out_free_mtt;
938 
939 	eq_context->flags	  = cpu_to_be32(MLX4_EQ_STATUS_OK   |
940 						MLX4_EQ_STATE_ARMED);
941 	eq_context->log_eq_size	  = ilog2(eq->nent);
942 	eq_context->intr	  = intr;
943 	eq_context->log_page_size = PAGE_SHIFT - MLX4_ICM_PAGE_SHIFT;
944 
945 	mtt_addr = mlx4_mtt_addr(dev, &eq->mtt);
946 	eq_context->mtt_base_addr_h = mtt_addr >> 32;
947 	eq_context->mtt_base_addr_l = cpu_to_be32(mtt_addr & 0xffffffff);
948 
949 	err = mlx4_SW2HW_EQ(dev, mailbox, eq->eqn);
950 	if (err) {
951 		mlx4_warn(dev, "SW2HW_EQ failed (%d)\n", err);
952 		goto err_out_free_mtt;
953 	}
954 
955 	kfree(dma_list);
956 	mlx4_free_cmd_mailbox(dev, mailbox);
957 
958 	eq->cons_index = 0;
959 
960 	return err;
961 
962 err_out_free_mtt:
963 	mlx4_mtt_cleanup(dev, &eq->mtt);
964 
965 err_out_free_eq:
966 	mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn);
967 
968 err_out_free_pages:
969 	for (i = 0; i < npages; ++i)
970 		if (eq->page_list[i].buf)
971 			dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
972 					  eq->page_list[i].buf,
973 					  eq->page_list[i].map);
974 
975 	mlx4_free_cmd_mailbox(dev, mailbox);
976 
977 err_out_free:
978 	kfree(eq->page_list);
979 	kfree(dma_list);
980 
981 err_out:
982 	return err;
983 }
984 
985 static void mlx4_free_eq(struct mlx4_dev *dev,
986 			 struct mlx4_eq *eq)
987 {
988 	struct mlx4_priv *priv = mlx4_priv(dev);
989 	struct mlx4_cmd_mailbox *mailbox;
990 	int err;
991 	int i;
992 	/* CX3 is capable of extending the CQE/EQE from 32 to 64 bytes */
993 	int npages = PAGE_ALIGN((MLX4_EQ_ENTRY_SIZE << dev->caps.eqe_factor) * eq->nent) / PAGE_SIZE;
994 
995 	mailbox = mlx4_alloc_cmd_mailbox(dev);
996 	if (IS_ERR(mailbox))
997 		return;
998 
999 	err = mlx4_HW2SW_EQ(dev, mailbox, eq->eqn);
1000 	if (err)
1001 		mlx4_warn(dev, "HW2SW_EQ failed (%d)\n", err);
1002 
1003 	if (0) {
1004 		mlx4_dbg(dev, "Dumping EQ context %02x:\n", eq->eqn);
1005 		for (i = 0; i < sizeof (struct mlx4_eq_context) / 4; ++i) {
1006 			if (i % 4 == 0)
1007 				pr_cont("[%02x] ", i * 4);
1008 			pr_cont(" %08x", be32_to_cpup(mailbox->buf + i * 4));
1009 			if ((i + 1) % 4 == 0)
1010 				pr_cont("\n");
1011 		}
1012 	}
1013 
1014 	mlx4_mtt_cleanup(dev, &eq->mtt);
1015 	for (i = 0; i < npages; ++i)
1016 		dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
1017 				    eq->page_list[i].buf,
1018 				    eq->page_list[i].map);
1019 
1020 	kfree(eq->page_list);
1021 	mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn);
1022 	mlx4_free_cmd_mailbox(dev, mailbox);
1023 }
1024 
1025 static void mlx4_free_irqs(struct mlx4_dev *dev)
1026 {
1027 	struct mlx4_eq_table *eq_table = &mlx4_priv(dev)->eq_table;
1028 	struct mlx4_priv *priv = mlx4_priv(dev);
1029 	int	i, vec;
1030 
1031 	if (eq_table->have_irq)
1032 		free_irq(dev->pdev->irq, dev);
1033 
1034 	for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
1035 		if (eq_table->eq[i].have_irq) {
1036 			free_irq(eq_table->eq[i].irq, eq_table->eq + i);
1037 			eq_table->eq[i].have_irq = 0;
1038 		}
1039 
1040 	for (i = 0; i < dev->caps.comp_pool; i++) {
1041 		/*
1042 		 * Freeing the assigned irq's
1043 		 * all bits should be 0, but we need to validate
1044 		 */
1045 		if (priv->msix_ctl.pool_bm & 1ULL << i) {
1046 			/* NO need protecting*/
1047 			vec = dev->caps.num_comp_vectors + 1 + i;
1048 			free_irq(priv->eq_table.eq[vec].irq,
1049 				 &priv->eq_table.eq[vec]);
1050 		}
1051 	}
1052 
1053 
1054 	kfree(eq_table->irq_names);
1055 }
1056 
1057 static int mlx4_map_clr_int(struct mlx4_dev *dev)
1058 {
1059 	struct mlx4_priv *priv = mlx4_priv(dev);
1060 
1061 	priv->clr_base = ioremap(pci_resource_start(dev->pdev, priv->fw.clr_int_bar) +
1062 				 priv->fw.clr_int_base, MLX4_CLR_INT_SIZE);
1063 	if (!priv->clr_base) {
1064 		mlx4_err(dev, "Couldn't map interrupt clear register, aborting.\n");
1065 		return -ENOMEM;
1066 	}
1067 
1068 	return 0;
1069 }
1070 
1071 static void mlx4_unmap_clr_int(struct mlx4_dev *dev)
1072 {
1073 	struct mlx4_priv *priv = mlx4_priv(dev);
1074 
1075 	iounmap(priv->clr_base);
1076 }
1077 
1078 int mlx4_alloc_eq_table(struct mlx4_dev *dev)
1079 {
1080 	struct mlx4_priv *priv = mlx4_priv(dev);
1081 
1082 	priv->eq_table.eq = kcalloc(dev->caps.num_eqs - dev->caps.reserved_eqs,
1083 				    sizeof *priv->eq_table.eq, GFP_KERNEL);
1084 	if (!priv->eq_table.eq)
1085 		return -ENOMEM;
1086 
1087 	return 0;
1088 }
1089 
1090 void mlx4_free_eq_table(struct mlx4_dev *dev)
1091 {
1092 	kfree(mlx4_priv(dev)->eq_table.eq);
1093 }
1094 
1095 int mlx4_init_eq_table(struct mlx4_dev *dev)
1096 {
1097 	struct mlx4_priv *priv = mlx4_priv(dev);
1098 	int err;
1099 	int i;
1100 
1101 	priv->eq_table.uar_map = kcalloc(mlx4_num_eq_uar(dev),
1102 					 sizeof *priv->eq_table.uar_map,
1103 					 GFP_KERNEL);
1104 	if (!priv->eq_table.uar_map) {
1105 		err = -ENOMEM;
1106 		goto err_out_free;
1107 	}
1108 
1109 	err = mlx4_bitmap_init(&priv->eq_table.bitmap, dev->caps.num_eqs,
1110 			       dev->caps.num_eqs - 1, dev->caps.reserved_eqs, 0);
1111 	if (err)
1112 		goto err_out_free;
1113 
1114 	for (i = 0; i < mlx4_num_eq_uar(dev); ++i)
1115 		priv->eq_table.uar_map[i] = NULL;
1116 
1117 	if (!mlx4_is_slave(dev)) {
1118 		err = mlx4_map_clr_int(dev);
1119 		if (err)
1120 			goto err_out_bitmap;
1121 
1122 		priv->eq_table.clr_mask =
1123 			swab32(1 << (priv->eq_table.inta_pin & 31));
1124 		priv->eq_table.clr_int  = priv->clr_base +
1125 			(priv->eq_table.inta_pin < 32 ? 4 : 0);
1126 	}
1127 
1128 	priv->eq_table.irq_names =
1129 		kmalloc(MLX4_IRQNAME_SIZE * (dev->caps.num_comp_vectors + 1 +
1130 					     dev->caps.comp_pool),
1131 			GFP_KERNEL);
1132 	if (!priv->eq_table.irq_names) {
1133 		err = -ENOMEM;
1134 		goto err_out_bitmap;
1135 	}
1136 
1137 	for (i = 0; i < dev->caps.num_comp_vectors; ++i) {
1138 		err = mlx4_create_eq(dev, dev->caps.num_cqs -
1139 					  dev->caps.reserved_cqs +
1140 					  MLX4_NUM_SPARE_EQE,
1141 				     (dev->flags & MLX4_FLAG_MSI_X) ? i : 0,
1142 				     &priv->eq_table.eq[i]);
1143 		if (err) {
1144 			--i;
1145 			goto err_out_unmap;
1146 		}
1147 	}
1148 
1149 	err = mlx4_create_eq(dev, MLX4_NUM_ASYNC_EQE + MLX4_NUM_SPARE_EQE,
1150 			     (dev->flags & MLX4_FLAG_MSI_X) ? dev->caps.num_comp_vectors : 0,
1151 			     &priv->eq_table.eq[dev->caps.num_comp_vectors]);
1152 	if (err)
1153 		goto err_out_comp;
1154 
1155 	/*if additional completion vectors poolsize is 0 this loop will not run*/
1156 	for (i = dev->caps.num_comp_vectors + 1;
1157 	      i < dev->caps.num_comp_vectors + dev->caps.comp_pool + 1; ++i) {
1158 
1159 		err = mlx4_create_eq(dev, dev->caps.num_cqs -
1160 					  dev->caps.reserved_cqs +
1161 					  MLX4_NUM_SPARE_EQE,
1162 				     (dev->flags & MLX4_FLAG_MSI_X) ? i : 0,
1163 				     &priv->eq_table.eq[i]);
1164 		if (err) {
1165 			--i;
1166 			goto err_out_unmap;
1167 		}
1168 	}
1169 
1170 
1171 	if (dev->flags & MLX4_FLAG_MSI_X) {
1172 		const char *eq_name;
1173 
1174 		for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i) {
1175 			if (i < dev->caps.num_comp_vectors) {
1176 				snprintf(priv->eq_table.irq_names +
1177 					 i * MLX4_IRQNAME_SIZE,
1178 					 MLX4_IRQNAME_SIZE,
1179 					 "mlx4-comp-%d@pci:%s", i,
1180 					 pci_name(dev->pdev));
1181 			} else {
1182 				snprintf(priv->eq_table.irq_names +
1183 					 i * MLX4_IRQNAME_SIZE,
1184 					 MLX4_IRQNAME_SIZE,
1185 					 "mlx4-async@pci:%s",
1186 					 pci_name(dev->pdev));
1187 			}
1188 
1189 			eq_name = priv->eq_table.irq_names +
1190 				  i * MLX4_IRQNAME_SIZE;
1191 			err = request_irq(priv->eq_table.eq[i].irq,
1192 					  mlx4_msi_x_interrupt, 0, eq_name,
1193 					  priv->eq_table.eq + i);
1194 			if (err)
1195 				goto err_out_async;
1196 
1197 			priv->eq_table.eq[i].have_irq = 1;
1198 		}
1199 	} else {
1200 		snprintf(priv->eq_table.irq_names,
1201 			 MLX4_IRQNAME_SIZE,
1202 			 DRV_NAME "@pci:%s",
1203 			 pci_name(dev->pdev));
1204 		err = request_irq(dev->pdev->irq, mlx4_interrupt,
1205 				  IRQF_SHARED, priv->eq_table.irq_names, dev);
1206 		if (err)
1207 			goto err_out_async;
1208 
1209 		priv->eq_table.have_irq = 1;
1210 	}
1211 
1212 	err = mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1213 			  priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
1214 	if (err)
1215 		mlx4_warn(dev, "MAP_EQ for async EQ %d failed (%d)\n",
1216 			   priv->eq_table.eq[dev->caps.num_comp_vectors].eqn, err);
1217 
1218 	for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
1219 		eq_set_ci(&priv->eq_table.eq[i], 1);
1220 
1221 	return 0;
1222 
1223 err_out_async:
1224 	mlx4_free_eq(dev, &priv->eq_table.eq[dev->caps.num_comp_vectors]);
1225 
1226 err_out_comp:
1227 	i = dev->caps.num_comp_vectors - 1;
1228 
1229 err_out_unmap:
1230 	while (i >= 0) {
1231 		mlx4_free_eq(dev, &priv->eq_table.eq[i]);
1232 		--i;
1233 	}
1234 	if (!mlx4_is_slave(dev))
1235 		mlx4_unmap_clr_int(dev);
1236 	mlx4_free_irqs(dev);
1237 
1238 err_out_bitmap:
1239 	mlx4_unmap_uar(dev);
1240 	mlx4_bitmap_cleanup(&priv->eq_table.bitmap);
1241 
1242 err_out_free:
1243 	kfree(priv->eq_table.uar_map);
1244 
1245 	return err;
1246 }
1247 
1248 void mlx4_cleanup_eq_table(struct mlx4_dev *dev)
1249 {
1250 	struct mlx4_priv *priv = mlx4_priv(dev);
1251 	int i;
1252 
1253 	mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 1,
1254 		    priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
1255 
1256 	mlx4_free_irqs(dev);
1257 
1258 	for (i = 0; i < dev->caps.num_comp_vectors + dev->caps.comp_pool + 1; ++i)
1259 		mlx4_free_eq(dev, &priv->eq_table.eq[i]);
1260 
1261 	if (!mlx4_is_slave(dev))
1262 		mlx4_unmap_clr_int(dev);
1263 
1264 	mlx4_unmap_uar(dev);
1265 	mlx4_bitmap_cleanup(&priv->eq_table.bitmap);
1266 
1267 	kfree(priv->eq_table.uar_map);
1268 }
1269 
1270 /* A test that verifies that we can accept interrupts on all
1271  * the irq vectors of the device.
1272  * Interrupts are checked using the NOP command.
1273  */
1274 int mlx4_test_interrupts(struct mlx4_dev *dev)
1275 {
1276 	struct mlx4_priv *priv = mlx4_priv(dev);
1277 	int i;
1278 	int err;
1279 
1280 	err = mlx4_NOP(dev);
1281 	/* When not in MSI_X, there is only one irq to check */
1282 	if (!(dev->flags & MLX4_FLAG_MSI_X) || mlx4_is_slave(dev))
1283 		return err;
1284 
1285 	/* A loop over all completion vectors, for each vector we will check
1286 	 * whether it works by mapping command completions to that vector
1287 	 * and performing a NOP command
1288 	 */
1289 	for(i = 0; !err && (i < dev->caps.num_comp_vectors); ++i) {
1290 		/* Temporary use polling for command completions */
1291 		mlx4_cmd_use_polling(dev);
1292 
1293 		/* Map the new eq to handle all asynchronous events */
1294 		err = mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1295 				  priv->eq_table.eq[i].eqn);
1296 		if (err) {
1297 			mlx4_warn(dev, "Failed mapping eq for interrupt test\n");
1298 			mlx4_cmd_use_events(dev);
1299 			break;
1300 		}
1301 
1302 		/* Go back to using events */
1303 		mlx4_cmd_use_events(dev);
1304 		err = mlx4_NOP(dev);
1305 	}
1306 
1307 	/* Return to default */
1308 	mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1309 		    priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
1310 	return err;
1311 }
1312 EXPORT_SYMBOL(mlx4_test_interrupts);
1313 
1314 int mlx4_assign_eq(struct mlx4_dev *dev, char *name, struct cpu_rmap *rmap,
1315 		   int *vector)
1316 {
1317 
1318 	struct mlx4_priv *priv = mlx4_priv(dev);
1319 	int vec = 0, err = 0, i;
1320 
1321 	mutex_lock(&priv->msix_ctl.pool_lock);
1322 	for (i = 0; !vec && i < dev->caps.comp_pool; i++) {
1323 		if (~priv->msix_ctl.pool_bm & 1ULL << i) {
1324 			priv->msix_ctl.pool_bm |= 1ULL << i;
1325 			vec = dev->caps.num_comp_vectors + 1 + i;
1326 			snprintf(priv->eq_table.irq_names +
1327 					vec * MLX4_IRQNAME_SIZE,
1328 					MLX4_IRQNAME_SIZE, "%s", name);
1329 #ifdef CONFIG_RFS_ACCEL
1330 			if (rmap) {
1331 				err = irq_cpu_rmap_add(rmap,
1332 						       priv->eq_table.eq[vec].irq);
1333 				if (err)
1334 					mlx4_warn(dev, "Failed adding irq rmap\n");
1335 			}
1336 #endif
1337 			err = request_irq(priv->eq_table.eq[vec].irq,
1338 					  mlx4_msi_x_interrupt, 0,
1339 					  &priv->eq_table.irq_names[vec<<5],
1340 					  priv->eq_table.eq + vec);
1341 			if (err) {
1342 				/*zero out bit by fliping it*/
1343 				priv->msix_ctl.pool_bm ^= 1 << i;
1344 				vec = 0;
1345 				continue;
1346 				/*we dont want to break here*/
1347 			}
1348 			eq_set_ci(&priv->eq_table.eq[vec], 1);
1349 		}
1350 	}
1351 	mutex_unlock(&priv->msix_ctl.pool_lock);
1352 
1353 	if (vec) {
1354 		*vector = vec;
1355 	} else {
1356 		*vector = 0;
1357 		err = (i == dev->caps.comp_pool) ? -ENOSPC : err;
1358 	}
1359 	return err;
1360 }
1361 EXPORT_SYMBOL(mlx4_assign_eq);
1362 
1363 void mlx4_release_eq(struct mlx4_dev *dev, int vec)
1364 {
1365 	struct mlx4_priv *priv = mlx4_priv(dev);
1366 	/*bm index*/
1367 	int i = vec - dev->caps.num_comp_vectors - 1;
1368 
1369 	if (likely(i >= 0)) {
1370 		/*sanity check , making sure were not trying to free irq's
1371 		  Belonging to a legacy EQ*/
1372 		mutex_lock(&priv->msix_ctl.pool_lock);
1373 		if (priv->msix_ctl.pool_bm & 1ULL << i) {
1374 			free_irq(priv->eq_table.eq[vec].irq,
1375 				 &priv->eq_table.eq[vec]);
1376 			priv->msix_ctl.pool_bm &= ~(1ULL << i);
1377 		}
1378 		mutex_unlock(&priv->msix_ctl.pool_lock);
1379 	}
1380 
1381 }
1382 EXPORT_SYMBOL(mlx4_release_eq);
1383 
1384