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