xref: /linux/drivers/net/ethernet/mellanox/mlx4/eq.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
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/interrupt.h>
35 #include <linux/slab.h>
36 #include <linux/export.h>
37 #include <linux/mm.h>
38 #include <linux/dma-mapping.h>
39 
40 #include <linux/mlx4/cmd.h>
41 #include <linux/cpu_rmap.h>
42 
43 #include "mlx4.h"
44 #include "fw.h"
45 
46 enum {
47 	MLX4_IRQNAME_SIZE	= 32
48 };
49 
50 enum {
51 	MLX4_NUM_ASYNC_EQE	= 0x100,
52 	MLX4_NUM_SPARE_EQE	= 0x80,
53 	MLX4_EQ_ENTRY_SIZE	= 0x20
54 };
55 
56 #define MLX4_EQ_STATUS_OK	   ( 0 << 28)
57 #define MLX4_EQ_STATUS_WRITE_FAIL  (10 << 28)
58 #define MLX4_EQ_OWNER_SW	   ( 0 << 24)
59 #define MLX4_EQ_OWNER_HW	   ( 1 << 24)
60 #define MLX4_EQ_FLAG_EC		   ( 1 << 18)
61 #define MLX4_EQ_FLAG_OI		   ( 1 << 17)
62 #define MLX4_EQ_STATE_ARMED	   ( 9 <<  8)
63 #define MLX4_EQ_STATE_FIRED	   (10 <<  8)
64 #define MLX4_EQ_STATE_ALWAYS_ARMED (11 <<  8)
65 
66 #define MLX4_ASYNC_EVENT_MASK ((1ull << MLX4_EVENT_TYPE_PATH_MIG)	    | \
67 			       (1ull << MLX4_EVENT_TYPE_COMM_EST)	    | \
68 			       (1ull << MLX4_EVENT_TYPE_SQ_DRAINED)	    | \
69 			       (1ull << MLX4_EVENT_TYPE_CQ_ERROR)	    | \
70 			       (1ull << MLX4_EVENT_TYPE_WQ_CATAS_ERROR)	    | \
71 			       (1ull << MLX4_EVENT_TYPE_EEC_CATAS_ERROR)    | \
72 			       (1ull << MLX4_EVENT_TYPE_PATH_MIG_FAILED)    | \
73 			       (1ull << MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR) | \
74 			       (1ull << MLX4_EVENT_TYPE_WQ_ACCESS_ERROR)    | \
75 			       (1ull << MLX4_EVENT_TYPE_PORT_CHANGE)	    | \
76 			       (1ull << MLX4_EVENT_TYPE_ECC_DETECT)	    | \
77 			       (1ull << MLX4_EVENT_TYPE_SRQ_CATAS_ERROR)    | \
78 			       (1ull << MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE)    | \
79 			       (1ull << MLX4_EVENT_TYPE_SRQ_LIMIT)	    | \
80 			       (1ull << MLX4_EVENT_TYPE_CMD)		    | \
81 			       (1ull << MLX4_EVENT_TYPE_OP_REQUIRED)	    | \
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 generate event for slave %d\n",
156 							  i);
157 			}
158 		} else {
159 			if (mlx4_GEN_EQE(dev, slave, eqe))
160 				mlx4_warn(dev, "Failed to generate event for slave %d\n",
161 					  slave);
162 		}
163 		++slave_eq->cons;
164 	}
165 }
166 
167 
168 static void slave_event(struct mlx4_dev *dev, u8 slave, struct mlx4_eqe *eqe)
169 {
170 	struct mlx4_priv *priv = mlx4_priv(dev);
171 	struct mlx4_slave_event_eq *slave_eq = &priv->mfunc.master.slave_eq;
172 	struct mlx4_eqe *s_eqe;
173 	unsigned long flags;
174 
175 	spin_lock_irqsave(&slave_eq->event_lock, flags);
176 	s_eqe = &slave_eq->event_eqe[slave_eq->prod & (SLAVE_EVENT_EQ_SIZE - 1)];
177 	if ((!!(s_eqe->owner & 0x80)) ^
178 	    (!!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE))) {
179 		mlx4_warn(dev, "Master failed to generate an EQE for slave: %d. No free EQE on slave events queue\n",
180 			  slave);
181 		spin_unlock_irqrestore(&slave_eq->event_lock, flags);
182 		return;
183 	}
184 
185 	memcpy(s_eqe, eqe, dev->caps.eqe_size - 1);
186 	s_eqe->slave_id = slave;
187 	/* ensure all information is written before setting the ownersip bit */
188 	wmb();
189 	s_eqe->owner = !!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE) ? 0x0 : 0x80;
190 	++slave_eq->prod;
191 
192 	queue_work(priv->mfunc.master.comm_wq,
193 		   &priv->mfunc.master.slave_event_work);
194 	spin_unlock_irqrestore(&slave_eq->event_lock, flags);
195 }
196 
197 static void mlx4_slave_event(struct mlx4_dev *dev, int slave,
198 			     struct mlx4_eqe *eqe)
199 {
200 	struct mlx4_priv *priv = mlx4_priv(dev);
201 	struct mlx4_slave_state *s_slave =
202 		&priv->mfunc.master.slave_state[slave];
203 
204 	if (!s_slave->active) {
205 		/*mlx4_warn(dev, "Trying to pass event to inactive slave\n");*/
206 		return;
207 	}
208 
209 	slave_event(dev, slave, eqe);
210 }
211 
212 int mlx4_gen_pkey_eqe(struct mlx4_dev *dev, int slave, u8 port)
213 {
214 	struct mlx4_eqe eqe;
215 
216 	struct mlx4_priv *priv = mlx4_priv(dev);
217 	struct mlx4_slave_state *s_slave = &priv->mfunc.master.slave_state[slave];
218 
219 	if (!s_slave->active)
220 		return 0;
221 
222 	memset(&eqe, 0, sizeof eqe);
223 
224 	eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
225 	eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE;
226 	eqe.event.port_mgmt_change.port = port;
227 
228 	return mlx4_GEN_EQE(dev, slave, &eqe);
229 }
230 EXPORT_SYMBOL(mlx4_gen_pkey_eqe);
231 
232 int mlx4_gen_guid_change_eqe(struct mlx4_dev *dev, int slave, u8 port)
233 {
234 	struct mlx4_eqe eqe;
235 
236 	/*don't send if we don't have the that slave */
237 	if (dev->num_vfs < slave)
238 		return 0;
239 	memset(&eqe, 0, sizeof eqe);
240 
241 	eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
242 	eqe.subtype = MLX4_DEV_PMC_SUBTYPE_GUID_INFO;
243 	eqe.event.port_mgmt_change.port = port;
244 
245 	return mlx4_GEN_EQE(dev, slave, &eqe);
246 }
247 EXPORT_SYMBOL(mlx4_gen_guid_change_eqe);
248 
249 int mlx4_gen_port_state_change_eqe(struct mlx4_dev *dev, int slave, u8 port,
250 				   u8 port_subtype_change)
251 {
252 	struct mlx4_eqe eqe;
253 
254 	/*don't send if we don't have the that slave */
255 	if (dev->num_vfs < slave)
256 		return 0;
257 	memset(&eqe, 0, sizeof eqe);
258 
259 	eqe.type = MLX4_EVENT_TYPE_PORT_CHANGE;
260 	eqe.subtype = port_subtype_change;
261 	eqe.event.port_change.port = cpu_to_be32(port << 28);
262 
263 	mlx4_dbg(dev, "%s: sending: %d to slave: %d on port: %d\n", __func__,
264 		 port_subtype_change, slave, port);
265 	return mlx4_GEN_EQE(dev, slave, &eqe);
266 }
267 EXPORT_SYMBOL(mlx4_gen_port_state_change_eqe);
268 
269 enum slave_port_state mlx4_get_slave_port_state(struct mlx4_dev *dev, int slave, u8 port)
270 {
271 	struct mlx4_priv *priv = mlx4_priv(dev);
272 	struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state;
273 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
274 
275 	if (slave >= dev->num_slaves || port > dev->caps.num_ports ||
276 	    port <= 0 || !test_bit(port - 1, actv_ports.ports)) {
277 		pr_err("%s: Error: asking for slave:%d, port:%d\n",
278 		       __func__, slave, port);
279 		return SLAVE_PORT_DOWN;
280 	}
281 	return s_state[slave].port_state[port];
282 }
283 EXPORT_SYMBOL(mlx4_get_slave_port_state);
284 
285 static int mlx4_set_slave_port_state(struct mlx4_dev *dev, int slave, u8 port,
286 				     enum slave_port_state state)
287 {
288 	struct mlx4_priv *priv = mlx4_priv(dev);
289 	struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state;
290 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
291 
292 	if (slave >= dev->num_slaves || port > dev->caps.num_ports ||
293 	    port <= 0 || !test_bit(port - 1, actv_ports.ports)) {
294 		pr_err("%s: Error: asking for slave:%d, port:%d\n",
295 		       __func__, slave, port);
296 		return -1;
297 	}
298 	s_state[slave].port_state[port] = state;
299 
300 	return 0;
301 }
302 
303 static void set_all_slave_state(struct mlx4_dev *dev, u8 port, int event)
304 {
305 	int i;
306 	enum slave_port_gen_event gen_event;
307 	struct mlx4_slaves_pport slaves_pport = mlx4_phys_to_slaves_pport(dev,
308 									  port);
309 
310 	for (i = 0; i < dev->num_vfs + 1; i++)
311 		if (test_bit(i, slaves_pport.slaves))
312 			set_and_calc_slave_port_state(dev, i, port,
313 						      event, &gen_event);
314 }
315 /**************************************************************************
316 	The function get as input the new event to that port,
317 	and according to the prev state change the slave's port state.
318 	The events are:
319 		MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
320 		MLX4_PORT_STATE_DEV_EVENT_PORT_UP
321 		MLX4_PORT_STATE_IB_EVENT_GID_VALID
322 		MLX4_PORT_STATE_IB_EVENT_GID_INVALID
323 ***************************************************************************/
324 int set_and_calc_slave_port_state(struct mlx4_dev *dev, int slave,
325 				  u8 port, int event,
326 				  enum slave_port_gen_event *gen_event)
327 {
328 	struct mlx4_priv *priv = mlx4_priv(dev);
329 	struct mlx4_slave_state *ctx = NULL;
330 	unsigned long flags;
331 	int ret = -1;
332 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
333 	enum slave_port_state cur_state =
334 		mlx4_get_slave_port_state(dev, slave, port);
335 
336 	*gen_event = SLAVE_PORT_GEN_EVENT_NONE;
337 
338 	if (slave >= dev->num_slaves || port > dev->caps.num_ports ||
339 	    port <= 0 || !test_bit(port - 1, actv_ports.ports)) {
340 		pr_err("%s: Error: asking for slave:%d, port:%d\n",
341 		       __func__, slave, port);
342 		return ret;
343 	}
344 
345 	ctx = &priv->mfunc.master.slave_state[slave];
346 	spin_lock_irqsave(&ctx->lock, flags);
347 
348 	switch (cur_state) {
349 	case SLAVE_PORT_DOWN:
350 		if (MLX4_PORT_STATE_DEV_EVENT_PORT_UP == event)
351 			mlx4_set_slave_port_state(dev, slave, port,
352 						  SLAVE_PENDING_UP);
353 		break;
354 	case SLAVE_PENDING_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 		else if (MLX4_PORT_STATE_IB_PORT_STATE_EVENT_GID_VALID == event) {
359 			mlx4_set_slave_port_state(dev, slave, port,
360 						  SLAVE_PORT_UP);
361 			*gen_event = SLAVE_PORT_GEN_EVENT_UP;
362 		}
363 		break;
364 	case SLAVE_PORT_UP:
365 		if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event) {
366 			mlx4_set_slave_port_state(dev, slave, port,
367 						  SLAVE_PORT_DOWN);
368 			*gen_event = SLAVE_PORT_GEN_EVENT_DOWN;
369 		} else if (MLX4_PORT_STATE_IB_EVENT_GID_INVALID ==
370 				event) {
371 			mlx4_set_slave_port_state(dev, slave, port,
372 						  SLAVE_PENDING_UP);
373 			*gen_event = SLAVE_PORT_GEN_EVENT_DOWN;
374 		}
375 		break;
376 	default:
377 		pr_err("%s: BUG!!! UNKNOWN state: slave:%d, port:%d\n",
378 		       __func__, slave, port);
379 		goto out;
380 	}
381 	ret = mlx4_get_slave_port_state(dev, slave, port);
382 
383 out:
384 	spin_unlock_irqrestore(&ctx->lock, flags);
385 	return ret;
386 }
387 
388 EXPORT_SYMBOL(set_and_calc_slave_port_state);
389 
390 int mlx4_gen_slaves_port_mgt_ev(struct mlx4_dev *dev, u8 port, int attr)
391 {
392 	struct mlx4_eqe eqe;
393 
394 	memset(&eqe, 0, sizeof eqe);
395 
396 	eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
397 	eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PORT_INFO;
398 	eqe.event.port_mgmt_change.port = port;
399 	eqe.event.port_mgmt_change.params.port_info.changed_attr =
400 		cpu_to_be32((u32) attr);
401 
402 	slave_event(dev, ALL_SLAVES, &eqe);
403 	return 0;
404 }
405 EXPORT_SYMBOL(mlx4_gen_slaves_port_mgt_ev);
406 
407 void mlx4_master_handle_slave_flr(struct work_struct *work)
408 {
409 	struct mlx4_mfunc_master_ctx *master =
410 		container_of(work, struct mlx4_mfunc_master_ctx,
411 			     slave_flr_event_work);
412 	struct mlx4_mfunc *mfunc =
413 		container_of(master, struct mlx4_mfunc, master);
414 	struct mlx4_priv *priv =
415 		container_of(mfunc, struct mlx4_priv, mfunc);
416 	struct mlx4_dev *dev = &priv->dev;
417 	struct mlx4_slave_state *slave_state = priv->mfunc.master.slave_state;
418 	int i;
419 	int err;
420 	unsigned long flags;
421 
422 	mlx4_dbg(dev, "mlx4_handle_slave_flr\n");
423 
424 	for (i = 0 ; i < dev->num_slaves; i++) {
425 
426 		if (MLX4_COMM_CMD_FLR == slave_state[i].last_cmd) {
427 			mlx4_dbg(dev, "mlx4_handle_slave_flr: clean slave: %d\n",
428 				 i);
429 
430 			mlx4_delete_all_resources_for_slave(dev, i);
431 			/*return the slave to running mode*/
432 			spin_lock_irqsave(&priv->mfunc.master.slave_state_lock, flags);
433 			slave_state[i].last_cmd = MLX4_COMM_CMD_RESET;
434 			slave_state[i].is_slave_going_down = 0;
435 			spin_unlock_irqrestore(&priv->mfunc.master.slave_state_lock, flags);
436 			/*notify the FW:*/
437 			err = mlx4_cmd(dev, 0, i, 0, MLX4_CMD_INFORM_FLR_DONE,
438 				       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
439 			if (err)
440 				mlx4_warn(dev, "Failed to notify FW on FLR done (slave:%d)\n",
441 					  i);
442 		}
443 	}
444 }
445 
446 static int mlx4_eq_int(struct mlx4_dev *dev, struct mlx4_eq *eq)
447 {
448 	struct mlx4_priv *priv = mlx4_priv(dev);
449 	struct mlx4_eqe *eqe;
450 	int cqn;
451 	int eqes_found = 0;
452 	int set_ci = 0;
453 	int port;
454 	int slave = 0;
455 	int ret;
456 	u32 flr_slave;
457 	u8 update_slave_state;
458 	int i;
459 	enum slave_port_gen_event gen_event;
460 	unsigned long flags;
461 	struct mlx4_vport_state *s_info;
462 
463 	while ((eqe = next_eqe_sw(eq, dev->caps.eqe_factor))) {
464 		/*
465 		 * Make sure we read EQ entry contents after we've
466 		 * checked the ownership bit.
467 		 */
468 		rmb();
469 
470 		switch (eqe->type) {
471 		case MLX4_EVENT_TYPE_COMP:
472 			cqn = be32_to_cpu(eqe->event.comp.cqn) & 0xffffff;
473 			mlx4_cq_completion(dev, cqn);
474 			break;
475 
476 		case MLX4_EVENT_TYPE_PATH_MIG:
477 		case MLX4_EVENT_TYPE_COMM_EST:
478 		case MLX4_EVENT_TYPE_SQ_DRAINED:
479 		case MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE:
480 		case MLX4_EVENT_TYPE_WQ_CATAS_ERROR:
481 		case MLX4_EVENT_TYPE_PATH_MIG_FAILED:
482 		case MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
483 		case MLX4_EVENT_TYPE_WQ_ACCESS_ERROR:
484 			mlx4_dbg(dev, "event %d arrived\n", eqe->type);
485 			if (mlx4_is_master(dev)) {
486 				/* forward only to slave owning the QP */
487 				ret = mlx4_get_slave_from_resource_id(dev,
488 						RES_QP,
489 						be32_to_cpu(eqe->event.qp.qpn)
490 						& 0xffffff, &slave);
491 				if (ret && ret != -ENOENT) {
492 					mlx4_dbg(dev, "QP event %02x(%02x) on EQ %d at index %u: could not get slave id (%d)\n",
493 						 eqe->type, eqe->subtype,
494 						 eq->eqn, eq->cons_index, ret);
495 					break;
496 				}
497 
498 				if (!ret && slave != dev->caps.function) {
499 					mlx4_slave_event(dev, slave, eqe);
500 					break;
501 				}
502 
503 			}
504 			mlx4_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) &
505 				      0xffffff, eqe->type);
506 			break;
507 
508 		case MLX4_EVENT_TYPE_SRQ_LIMIT:
509 			mlx4_dbg(dev, "%s: MLX4_EVENT_TYPE_SRQ_LIMIT\n",
510 				 __func__);
511 		case MLX4_EVENT_TYPE_SRQ_CATAS_ERROR:
512 			if (mlx4_is_master(dev)) {
513 				/* forward only to slave owning the SRQ */
514 				ret = mlx4_get_slave_from_resource_id(dev,
515 						RES_SRQ,
516 						be32_to_cpu(eqe->event.srq.srqn)
517 						& 0xffffff,
518 						&slave);
519 				if (ret && ret != -ENOENT) {
520 					mlx4_warn(dev, "SRQ event %02x(%02x) on EQ %d at index %u: could not get slave id (%d)\n",
521 						  eqe->type, eqe->subtype,
522 						  eq->eqn, eq->cons_index, ret);
523 					break;
524 				}
525 				mlx4_warn(dev, "%s: slave:%d, srq_no:0x%x, event: %02x(%02x)\n",
526 					  __func__, slave,
527 					  be32_to_cpu(eqe->event.srq.srqn),
528 					  eqe->type, eqe->subtype);
529 
530 				if (!ret && slave != dev->caps.function) {
531 					mlx4_warn(dev, "%s: sending event %02x(%02x) to slave:%d\n",
532 						  __func__, eqe->type,
533 						  eqe->subtype, slave);
534 					mlx4_slave_event(dev, slave, eqe);
535 					break;
536 				}
537 			}
538 			mlx4_srq_event(dev, be32_to_cpu(eqe->event.srq.srqn) &
539 				       0xffffff, eqe->type);
540 			break;
541 
542 		case MLX4_EVENT_TYPE_CMD:
543 			mlx4_cmd_event(dev,
544 				       be16_to_cpu(eqe->event.cmd.token),
545 				       eqe->event.cmd.status,
546 				       be64_to_cpu(eqe->event.cmd.out_param));
547 			break;
548 
549 		case MLX4_EVENT_TYPE_PORT_CHANGE: {
550 			struct mlx4_slaves_pport slaves_port;
551 			port = be32_to_cpu(eqe->event.port_change.port) >> 28;
552 			slaves_port = mlx4_phys_to_slaves_pport(dev, port);
553 			if (eqe->subtype == MLX4_PORT_CHANGE_SUBTYPE_DOWN) {
554 				mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_DOWN,
555 						    port);
556 				mlx4_priv(dev)->sense.do_sense_port[port] = 1;
557 				if (!mlx4_is_master(dev))
558 					break;
559 				for (i = 0; i < dev->num_vfs + 1; i++) {
560 					if (!test_bit(i, slaves_port.slaves))
561 						continue;
562 					if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH) {
563 						if (i == mlx4_master_func_num(dev))
564 							continue;
565 						mlx4_dbg(dev, "%s: Sending MLX4_PORT_CHANGE_SUBTYPE_DOWN to slave: %d, port:%d\n",
566 							 __func__, i, port);
567 						s_info = &priv->mfunc.master.vf_oper[slave].vport[port].state;
568 						if (IFLA_VF_LINK_STATE_AUTO == s_info->link_state) {
569 							eqe->event.port_change.port =
570 								cpu_to_be32(
571 								(be32_to_cpu(eqe->event.port_change.port) & 0xFFFFFFF)
572 								| (mlx4_phys_to_slave_port(dev, i, port) << 28));
573 							mlx4_slave_event(dev, i, eqe);
574 						}
575 					} else {  /* IB port */
576 						set_and_calc_slave_port_state(dev, i, port,
577 									      MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
578 									      &gen_event);
579 						/*we can be in pending state, then do not send port_down event*/
580 						if (SLAVE_PORT_GEN_EVENT_DOWN ==  gen_event) {
581 							if (i == mlx4_master_func_num(dev))
582 								continue;
583 							mlx4_slave_event(dev, i, eqe);
584 						}
585 					}
586 				}
587 			} else {
588 				mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_UP, port);
589 
590 				mlx4_priv(dev)->sense.do_sense_port[port] = 0;
591 
592 				if (!mlx4_is_master(dev))
593 					break;
594 				if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
595 					for (i = 0; i < dev->num_vfs + 1; i++) {
596 						if (!test_bit(i, slaves_port.slaves))
597 							continue;
598 						if (i == mlx4_master_func_num(dev))
599 							continue;
600 						s_info = &priv->mfunc.master.vf_oper[slave].vport[port].state;
601 						if (IFLA_VF_LINK_STATE_AUTO == s_info->link_state) {
602 							eqe->event.port_change.port =
603 								cpu_to_be32(
604 								(be32_to_cpu(eqe->event.port_change.port) & 0xFFFFFFF)
605 								| (mlx4_phys_to_slave_port(dev, i, port) << 28));
606 							mlx4_slave_event(dev, i, eqe);
607 						}
608 					}
609 				else /* IB port */
610 					/* port-up event will be sent to a slave when the
611 					 * slave's alias-guid is set. This is done in alias_GUID.c
612 					 */
613 					set_all_slave_state(dev, port, MLX4_DEV_EVENT_PORT_UP);
614 			}
615 			break;
616 		}
617 
618 		case MLX4_EVENT_TYPE_CQ_ERROR:
619 			mlx4_warn(dev, "CQ %s on CQN %06x\n",
620 				  eqe->event.cq_err.syndrome == 1 ?
621 				  "overrun" : "access violation",
622 				  be32_to_cpu(eqe->event.cq_err.cqn) & 0xffffff);
623 			if (mlx4_is_master(dev)) {
624 				ret = mlx4_get_slave_from_resource_id(dev,
625 					RES_CQ,
626 					be32_to_cpu(eqe->event.cq_err.cqn)
627 					& 0xffffff, &slave);
628 				if (ret && ret != -ENOENT) {
629 					mlx4_dbg(dev, "CQ event %02x(%02x) on EQ %d at index %u: could not get slave id (%d)\n",
630 						 eqe->type, eqe->subtype,
631 						 eq->eqn, eq->cons_index, ret);
632 					break;
633 				}
634 
635 				if (!ret && slave != dev->caps.function) {
636 					mlx4_slave_event(dev, slave, eqe);
637 					break;
638 				}
639 			}
640 			mlx4_cq_event(dev,
641 				      be32_to_cpu(eqe->event.cq_err.cqn)
642 				      & 0xffffff,
643 				      eqe->type);
644 			break;
645 
646 		case MLX4_EVENT_TYPE_EQ_OVERFLOW:
647 			mlx4_warn(dev, "EQ overrun on EQN %d\n", eq->eqn);
648 			break;
649 
650 		case MLX4_EVENT_TYPE_OP_REQUIRED:
651 			atomic_inc(&priv->opreq_count);
652 			/* FW commands can't be executed from interrupt context
653 			 * working in deferred task
654 			 */
655 			queue_work(mlx4_wq, &priv->opreq_task);
656 			break;
657 
658 		case MLX4_EVENT_TYPE_COMM_CHANNEL:
659 			if (!mlx4_is_master(dev)) {
660 				mlx4_warn(dev, "Received comm channel event for non master device\n");
661 				break;
662 			}
663 			memcpy(&priv->mfunc.master.comm_arm_bit_vector,
664 			       eqe->event.comm_channel_arm.bit_vec,
665 			       sizeof eqe->event.comm_channel_arm.bit_vec);
666 			queue_work(priv->mfunc.master.comm_wq,
667 				   &priv->mfunc.master.comm_work);
668 			break;
669 
670 		case MLX4_EVENT_TYPE_FLR_EVENT:
671 			flr_slave = be32_to_cpu(eqe->event.flr_event.slave_id);
672 			if (!mlx4_is_master(dev)) {
673 				mlx4_warn(dev, "Non-master function received FLR event\n");
674 				break;
675 			}
676 
677 			mlx4_dbg(dev, "FLR event for slave: %d\n", flr_slave);
678 
679 			if (flr_slave >= dev->num_slaves) {
680 				mlx4_warn(dev,
681 					  "Got FLR for unknown function: %d\n",
682 					  flr_slave);
683 				update_slave_state = 0;
684 			} else
685 				update_slave_state = 1;
686 
687 			spin_lock_irqsave(&priv->mfunc.master.slave_state_lock, flags);
688 			if (update_slave_state) {
689 				priv->mfunc.master.slave_state[flr_slave].active = false;
690 				priv->mfunc.master.slave_state[flr_slave].last_cmd = MLX4_COMM_CMD_FLR;
691 				priv->mfunc.master.slave_state[flr_slave].is_slave_going_down = 1;
692 			}
693 			spin_unlock_irqrestore(&priv->mfunc.master.slave_state_lock, flags);
694 			queue_work(priv->mfunc.master.comm_wq,
695 				   &priv->mfunc.master.slave_flr_event_work);
696 			break;
697 
698 		case MLX4_EVENT_TYPE_FATAL_WARNING:
699 			if (eqe->subtype == MLX4_FATAL_WARNING_SUBTYPE_WARMING) {
700 				if (mlx4_is_master(dev))
701 					for (i = 0; i < dev->num_slaves; i++) {
702 						mlx4_dbg(dev, "%s: Sending MLX4_FATAL_WARNING_SUBTYPE_WARMING to slave: %d\n",
703 							 __func__, i);
704 						if (i == dev->caps.function)
705 							continue;
706 						mlx4_slave_event(dev, i, eqe);
707 					}
708 				mlx4_err(dev, "Temperature Threshold was reached! Threshold: %d celsius degrees; Current Temperature: %d\n",
709 					 be16_to_cpu(eqe->event.warming.warning_threshold),
710 					 be16_to_cpu(eqe->event.warming.current_temperature));
711 			} else
712 				mlx4_warn(dev, "Unhandled event FATAL WARNING (%02x), subtype %02x on EQ %d at index %u. owner=%x, nent=0x%x, slave=%x, 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 
719 			break;
720 
721 		case MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT:
722 			mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_MGMT_CHANGE,
723 					    (unsigned long) eqe);
724 			break;
725 
726 		case MLX4_EVENT_TYPE_EEC_CATAS_ERROR:
727 		case MLX4_EVENT_TYPE_ECC_DETECT:
728 		default:
729 			mlx4_warn(dev, "Unhandled event %02x(%02x) on EQ %d at index %u. owner=%x, nent=0x%x, slave=%x, ownership=%s\n",
730 				  eqe->type, eqe->subtype, eq->eqn,
731 				  eq->cons_index, eqe->owner, eq->nent,
732 				  eqe->slave_id,
733 				  !!(eqe->owner & 0x80) ^
734 				  !!(eq->cons_index & eq->nent) ? "HW" : "SW");
735 			break;
736 		};
737 
738 		++eq->cons_index;
739 		eqes_found = 1;
740 		++set_ci;
741 
742 		/*
743 		 * The HCA will think the queue has overflowed if we
744 		 * don't tell it we've been processing events.  We
745 		 * create our EQs with MLX4_NUM_SPARE_EQE extra
746 		 * entries, so we must update our consumer index at
747 		 * least that often.
748 		 */
749 		if (unlikely(set_ci >= MLX4_NUM_SPARE_EQE)) {
750 			eq_set_ci(eq, 0);
751 			set_ci = 0;
752 		}
753 	}
754 
755 	eq_set_ci(eq, 1);
756 
757 	return eqes_found;
758 }
759 
760 static irqreturn_t mlx4_interrupt(int irq, void *dev_ptr)
761 {
762 	struct mlx4_dev *dev = dev_ptr;
763 	struct mlx4_priv *priv = mlx4_priv(dev);
764 	int work = 0;
765 	int i;
766 
767 	writel(priv->eq_table.clr_mask, priv->eq_table.clr_int);
768 
769 	for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
770 		work |= mlx4_eq_int(dev, &priv->eq_table.eq[i]);
771 
772 	return IRQ_RETVAL(work);
773 }
774 
775 static irqreturn_t mlx4_msi_x_interrupt(int irq, void *eq_ptr)
776 {
777 	struct mlx4_eq  *eq  = eq_ptr;
778 	struct mlx4_dev *dev = eq->dev;
779 
780 	mlx4_eq_int(dev, eq);
781 
782 	/* MSI-X vectors always belong to us */
783 	return IRQ_HANDLED;
784 }
785 
786 int mlx4_MAP_EQ_wrapper(struct mlx4_dev *dev, int slave,
787 			struct mlx4_vhcr *vhcr,
788 			struct mlx4_cmd_mailbox *inbox,
789 			struct mlx4_cmd_mailbox *outbox,
790 			struct mlx4_cmd_info *cmd)
791 {
792 	struct mlx4_priv *priv = mlx4_priv(dev);
793 	struct mlx4_slave_event_eq_info *event_eq =
794 		priv->mfunc.master.slave_state[slave].event_eq;
795 	u32 in_modifier = vhcr->in_modifier;
796 	u32 eqn = in_modifier & 0x3FF;
797 	u64 in_param =  vhcr->in_param;
798 	int err = 0;
799 	int i;
800 
801 	if (slave == dev->caps.function)
802 		err = mlx4_cmd(dev, in_param, (in_modifier & 0x80000000) | eqn,
803 			       0, MLX4_CMD_MAP_EQ, MLX4_CMD_TIME_CLASS_B,
804 			       MLX4_CMD_NATIVE);
805 	if (!err)
806 		for (i = 0; i < MLX4_EVENT_TYPES_NUM; ++i)
807 			if (in_param & (1LL << i))
808 				event_eq[i].eqn = in_modifier >> 31 ? -1 : eqn;
809 
810 	return err;
811 }
812 
813 static int mlx4_MAP_EQ(struct mlx4_dev *dev, u64 event_mask, int unmap,
814 			int eq_num)
815 {
816 	return mlx4_cmd(dev, event_mask, (unmap << 31) | eq_num,
817 			0, MLX4_CMD_MAP_EQ, MLX4_CMD_TIME_CLASS_B,
818 			MLX4_CMD_WRAPPED);
819 }
820 
821 static int mlx4_SW2HW_EQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
822 			 int eq_num)
823 {
824 	return mlx4_cmd(dev, mailbox->dma, eq_num, 0,
825 			MLX4_CMD_SW2HW_EQ, MLX4_CMD_TIME_CLASS_A,
826 			MLX4_CMD_WRAPPED);
827 }
828 
829 static int mlx4_HW2SW_EQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
830 			 int eq_num)
831 {
832 	return mlx4_cmd_box(dev, 0, mailbox->dma, eq_num,
833 			    0, MLX4_CMD_HW2SW_EQ, MLX4_CMD_TIME_CLASS_A,
834 			    MLX4_CMD_WRAPPED);
835 }
836 
837 static int mlx4_num_eq_uar(struct mlx4_dev *dev)
838 {
839 	/*
840 	 * Each UAR holds 4 EQ doorbells.  To figure out how many UARs
841 	 * we need to map, take the difference of highest index and
842 	 * the lowest index we'll use and add 1.
843 	 */
844 	return (dev->caps.num_comp_vectors + 1 + dev->caps.reserved_eqs +
845 		 dev->caps.comp_pool)/4 - dev->caps.reserved_eqs/4 + 1;
846 }
847 
848 static void __iomem *mlx4_get_eq_uar(struct mlx4_dev *dev, struct mlx4_eq *eq)
849 {
850 	struct mlx4_priv *priv = mlx4_priv(dev);
851 	int index;
852 
853 	index = eq->eqn / 4 - dev->caps.reserved_eqs / 4;
854 
855 	if (!priv->eq_table.uar_map[index]) {
856 		priv->eq_table.uar_map[index] =
857 			ioremap(pci_resource_start(dev->pdev, 2) +
858 				((eq->eqn / 4) << PAGE_SHIFT),
859 				PAGE_SIZE);
860 		if (!priv->eq_table.uar_map[index]) {
861 			mlx4_err(dev, "Couldn't map EQ doorbell for EQN 0x%06x\n",
862 				 eq->eqn);
863 			return NULL;
864 		}
865 	}
866 
867 	return priv->eq_table.uar_map[index] + 0x800 + 8 * (eq->eqn % 4);
868 }
869 
870 static void mlx4_unmap_uar(struct mlx4_dev *dev)
871 {
872 	struct mlx4_priv *priv = mlx4_priv(dev);
873 	int i;
874 
875 	for (i = 0; i < mlx4_num_eq_uar(dev); ++i)
876 		if (priv->eq_table.uar_map[i]) {
877 			iounmap(priv->eq_table.uar_map[i]);
878 			priv->eq_table.uar_map[i] = NULL;
879 		}
880 }
881 
882 static int mlx4_create_eq(struct mlx4_dev *dev, int nent,
883 			  u8 intr, struct mlx4_eq *eq)
884 {
885 	struct mlx4_priv *priv = mlx4_priv(dev);
886 	struct mlx4_cmd_mailbox *mailbox;
887 	struct mlx4_eq_context *eq_context;
888 	int npages;
889 	u64 *dma_list = NULL;
890 	dma_addr_t t;
891 	u64 mtt_addr;
892 	int err = -ENOMEM;
893 	int i;
894 
895 	eq->dev   = dev;
896 	eq->nent  = roundup_pow_of_two(max(nent, 2));
897 	/* CX3 is capable of extending the CQE/EQE from 32 to 64 bytes */
898 	npages = PAGE_ALIGN(eq->nent * (MLX4_EQ_ENTRY_SIZE << dev->caps.eqe_factor)) / PAGE_SIZE;
899 
900 	eq->page_list = kmalloc(npages * sizeof *eq->page_list,
901 				GFP_KERNEL);
902 	if (!eq->page_list)
903 		goto err_out;
904 
905 	for (i = 0; i < npages; ++i)
906 		eq->page_list[i].buf = NULL;
907 
908 	dma_list = kmalloc(npages * sizeof *dma_list, GFP_KERNEL);
909 	if (!dma_list)
910 		goto err_out_free;
911 
912 	mailbox = mlx4_alloc_cmd_mailbox(dev);
913 	if (IS_ERR(mailbox))
914 		goto err_out_free;
915 	eq_context = mailbox->buf;
916 
917 	for (i = 0; i < npages; ++i) {
918 		eq->page_list[i].buf = dma_alloc_coherent(&dev->pdev->dev,
919 							  PAGE_SIZE, &t, GFP_KERNEL);
920 		if (!eq->page_list[i].buf)
921 			goto err_out_free_pages;
922 
923 		dma_list[i] = t;
924 		eq->page_list[i].map = t;
925 
926 		memset(eq->page_list[i].buf, 0, PAGE_SIZE);
927 	}
928 
929 	eq->eqn = mlx4_bitmap_alloc(&priv->eq_table.bitmap);
930 	if (eq->eqn == -1)
931 		goto err_out_free_pages;
932 
933 	eq->doorbell = mlx4_get_eq_uar(dev, eq);
934 	if (!eq->doorbell) {
935 		err = -ENOMEM;
936 		goto err_out_free_eq;
937 	}
938 
939 	err = mlx4_mtt_init(dev, npages, PAGE_SHIFT, &eq->mtt);
940 	if (err)
941 		goto err_out_free_eq;
942 
943 	err = mlx4_write_mtt(dev, &eq->mtt, 0, npages, dma_list);
944 	if (err)
945 		goto err_out_free_mtt;
946 
947 	eq_context->flags	  = cpu_to_be32(MLX4_EQ_STATUS_OK   |
948 						MLX4_EQ_STATE_ARMED);
949 	eq_context->log_eq_size	  = ilog2(eq->nent);
950 	eq_context->intr	  = intr;
951 	eq_context->log_page_size = PAGE_SHIFT - MLX4_ICM_PAGE_SHIFT;
952 
953 	mtt_addr = mlx4_mtt_addr(dev, &eq->mtt);
954 	eq_context->mtt_base_addr_h = mtt_addr >> 32;
955 	eq_context->mtt_base_addr_l = cpu_to_be32(mtt_addr & 0xffffffff);
956 
957 	err = mlx4_SW2HW_EQ(dev, mailbox, eq->eqn);
958 	if (err) {
959 		mlx4_warn(dev, "SW2HW_EQ failed (%d)\n", err);
960 		goto err_out_free_mtt;
961 	}
962 
963 	kfree(dma_list);
964 	mlx4_free_cmd_mailbox(dev, mailbox);
965 
966 	eq->cons_index = 0;
967 
968 	return err;
969 
970 err_out_free_mtt:
971 	mlx4_mtt_cleanup(dev, &eq->mtt);
972 
973 err_out_free_eq:
974 	mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn, MLX4_USE_RR);
975 
976 err_out_free_pages:
977 	for (i = 0; i < npages; ++i)
978 		if (eq->page_list[i].buf)
979 			dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
980 					  eq->page_list[i].buf,
981 					  eq->page_list[i].map);
982 
983 	mlx4_free_cmd_mailbox(dev, mailbox);
984 
985 err_out_free:
986 	kfree(eq->page_list);
987 	kfree(dma_list);
988 
989 err_out:
990 	return err;
991 }
992 
993 static void mlx4_free_eq(struct mlx4_dev *dev,
994 			 struct mlx4_eq *eq)
995 {
996 	struct mlx4_priv *priv = mlx4_priv(dev);
997 	struct mlx4_cmd_mailbox *mailbox;
998 	int err;
999 	int i;
1000 	/* CX3 is capable of extending the CQE/EQE from 32 to 64 bytes */
1001 	int npages = PAGE_ALIGN((MLX4_EQ_ENTRY_SIZE << dev->caps.eqe_factor) * eq->nent) / PAGE_SIZE;
1002 
1003 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1004 	if (IS_ERR(mailbox))
1005 		return;
1006 
1007 	err = mlx4_HW2SW_EQ(dev, mailbox, eq->eqn);
1008 	if (err)
1009 		mlx4_warn(dev, "HW2SW_EQ failed (%d)\n", err);
1010 
1011 	if (0) {
1012 		mlx4_dbg(dev, "Dumping EQ context %02x:\n", eq->eqn);
1013 		for (i = 0; i < sizeof (struct mlx4_eq_context) / 4; ++i) {
1014 			if (i % 4 == 0)
1015 				pr_cont("[%02x] ", i * 4);
1016 			pr_cont(" %08x", be32_to_cpup(mailbox->buf + i * 4));
1017 			if ((i + 1) % 4 == 0)
1018 				pr_cont("\n");
1019 		}
1020 	}
1021 
1022 	mlx4_mtt_cleanup(dev, &eq->mtt);
1023 	for (i = 0; i < npages; ++i)
1024 		dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
1025 				    eq->page_list[i].buf,
1026 				    eq->page_list[i].map);
1027 
1028 	kfree(eq->page_list);
1029 	mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn, MLX4_USE_RR);
1030 	mlx4_free_cmd_mailbox(dev, mailbox);
1031 }
1032 
1033 static void mlx4_free_irqs(struct mlx4_dev *dev)
1034 {
1035 	struct mlx4_eq_table *eq_table = &mlx4_priv(dev)->eq_table;
1036 	struct mlx4_priv *priv = mlx4_priv(dev);
1037 	int	i, vec;
1038 
1039 	if (eq_table->have_irq)
1040 		free_irq(dev->pdev->irq, dev);
1041 
1042 	for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
1043 		if (eq_table->eq[i].have_irq) {
1044 			free_irq(eq_table->eq[i].irq, eq_table->eq + i);
1045 			eq_table->eq[i].have_irq = 0;
1046 		}
1047 
1048 	for (i = 0; i < dev->caps.comp_pool; i++) {
1049 		/*
1050 		 * Freeing the assigned irq's
1051 		 * all bits should be 0, but we need to validate
1052 		 */
1053 		if (priv->msix_ctl.pool_bm & 1ULL << i) {
1054 			/* NO need protecting*/
1055 			vec = dev->caps.num_comp_vectors + 1 + i;
1056 			free_irq(priv->eq_table.eq[vec].irq,
1057 				 &priv->eq_table.eq[vec]);
1058 		}
1059 	}
1060 
1061 
1062 	kfree(eq_table->irq_names);
1063 }
1064 
1065 static int mlx4_map_clr_int(struct mlx4_dev *dev)
1066 {
1067 	struct mlx4_priv *priv = mlx4_priv(dev);
1068 
1069 	priv->clr_base = ioremap(pci_resource_start(dev->pdev, priv->fw.clr_int_bar) +
1070 				 priv->fw.clr_int_base, MLX4_CLR_INT_SIZE);
1071 	if (!priv->clr_base) {
1072 		mlx4_err(dev, "Couldn't map interrupt clear register, aborting\n");
1073 		return -ENOMEM;
1074 	}
1075 
1076 	return 0;
1077 }
1078 
1079 static void mlx4_unmap_clr_int(struct mlx4_dev *dev)
1080 {
1081 	struct mlx4_priv *priv = mlx4_priv(dev);
1082 
1083 	iounmap(priv->clr_base);
1084 }
1085 
1086 int mlx4_alloc_eq_table(struct mlx4_dev *dev)
1087 {
1088 	struct mlx4_priv *priv = mlx4_priv(dev);
1089 
1090 	priv->eq_table.eq = kcalloc(dev->caps.num_eqs - dev->caps.reserved_eqs,
1091 				    sizeof *priv->eq_table.eq, GFP_KERNEL);
1092 	if (!priv->eq_table.eq)
1093 		return -ENOMEM;
1094 
1095 	return 0;
1096 }
1097 
1098 void mlx4_free_eq_table(struct mlx4_dev *dev)
1099 {
1100 	kfree(mlx4_priv(dev)->eq_table.eq);
1101 }
1102 
1103 int mlx4_init_eq_table(struct mlx4_dev *dev)
1104 {
1105 	struct mlx4_priv *priv = mlx4_priv(dev);
1106 	int err;
1107 	int i;
1108 
1109 	priv->eq_table.uar_map = kcalloc(mlx4_num_eq_uar(dev),
1110 					 sizeof *priv->eq_table.uar_map,
1111 					 GFP_KERNEL);
1112 	if (!priv->eq_table.uar_map) {
1113 		err = -ENOMEM;
1114 		goto err_out_free;
1115 	}
1116 
1117 	err = mlx4_bitmap_init(&priv->eq_table.bitmap, dev->caps.num_eqs,
1118 			       dev->caps.num_eqs - 1, dev->caps.reserved_eqs, 0);
1119 	if (err)
1120 		goto err_out_free;
1121 
1122 	for (i = 0; i < mlx4_num_eq_uar(dev); ++i)
1123 		priv->eq_table.uar_map[i] = NULL;
1124 
1125 	if (!mlx4_is_slave(dev)) {
1126 		err = mlx4_map_clr_int(dev);
1127 		if (err)
1128 			goto err_out_bitmap;
1129 
1130 		priv->eq_table.clr_mask =
1131 			swab32(1 << (priv->eq_table.inta_pin & 31));
1132 		priv->eq_table.clr_int  = priv->clr_base +
1133 			(priv->eq_table.inta_pin < 32 ? 4 : 0);
1134 	}
1135 
1136 	priv->eq_table.irq_names =
1137 		kmalloc(MLX4_IRQNAME_SIZE * (dev->caps.num_comp_vectors + 1 +
1138 					     dev->caps.comp_pool),
1139 			GFP_KERNEL);
1140 	if (!priv->eq_table.irq_names) {
1141 		err = -ENOMEM;
1142 		goto err_out_bitmap;
1143 	}
1144 
1145 	for (i = 0; i < dev->caps.num_comp_vectors; ++i) {
1146 		err = mlx4_create_eq(dev, dev->caps.num_cqs -
1147 					  dev->caps.reserved_cqs +
1148 					  MLX4_NUM_SPARE_EQE,
1149 				     (dev->flags & MLX4_FLAG_MSI_X) ? i : 0,
1150 				     &priv->eq_table.eq[i]);
1151 		if (err) {
1152 			--i;
1153 			goto err_out_unmap;
1154 		}
1155 	}
1156 
1157 	err = mlx4_create_eq(dev, MLX4_NUM_ASYNC_EQE + MLX4_NUM_SPARE_EQE,
1158 			     (dev->flags & MLX4_FLAG_MSI_X) ? dev->caps.num_comp_vectors : 0,
1159 			     &priv->eq_table.eq[dev->caps.num_comp_vectors]);
1160 	if (err)
1161 		goto err_out_comp;
1162 
1163 	/*if additional completion vectors poolsize is 0 this loop will not run*/
1164 	for (i = dev->caps.num_comp_vectors + 1;
1165 	      i < dev->caps.num_comp_vectors + dev->caps.comp_pool + 1; ++i) {
1166 
1167 		err = mlx4_create_eq(dev, dev->caps.num_cqs -
1168 					  dev->caps.reserved_cqs +
1169 					  MLX4_NUM_SPARE_EQE,
1170 				     (dev->flags & MLX4_FLAG_MSI_X) ? i : 0,
1171 				     &priv->eq_table.eq[i]);
1172 		if (err) {
1173 			--i;
1174 			goto err_out_unmap;
1175 		}
1176 	}
1177 
1178 
1179 	if (dev->flags & MLX4_FLAG_MSI_X) {
1180 		const char *eq_name;
1181 
1182 		for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i) {
1183 			if (i < dev->caps.num_comp_vectors) {
1184 				snprintf(priv->eq_table.irq_names +
1185 					 i * MLX4_IRQNAME_SIZE,
1186 					 MLX4_IRQNAME_SIZE,
1187 					 "mlx4-comp-%d@pci:%s", i,
1188 					 pci_name(dev->pdev));
1189 			} else {
1190 				snprintf(priv->eq_table.irq_names +
1191 					 i * MLX4_IRQNAME_SIZE,
1192 					 MLX4_IRQNAME_SIZE,
1193 					 "mlx4-async@pci:%s",
1194 					 pci_name(dev->pdev));
1195 			}
1196 
1197 			eq_name = priv->eq_table.irq_names +
1198 				  i * MLX4_IRQNAME_SIZE;
1199 			err = request_irq(priv->eq_table.eq[i].irq,
1200 					  mlx4_msi_x_interrupt, 0, eq_name,
1201 					  priv->eq_table.eq + i);
1202 			if (err)
1203 				goto err_out_async;
1204 
1205 			priv->eq_table.eq[i].have_irq = 1;
1206 		}
1207 	} else {
1208 		snprintf(priv->eq_table.irq_names,
1209 			 MLX4_IRQNAME_SIZE,
1210 			 DRV_NAME "@pci:%s",
1211 			 pci_name(dev->pdev));
1212 		err = request_irq(dev->pdev->irq, mlx4_interrupt,
1213 				  IRQF_SHARED, priv->eq_table.irq_names, dev);
1214 		if (err)
1215 			goto err_out_async;
1216 
1217 		priv->eq_table.have_irq = 1;
1218 	}
1219 
1220 	err = mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1221 			  priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
1222 	if (err)
1223 		mlx4_warn(dev, "MAP_EQ for async EQ %d failed (%d)\n",
1224 			   priv->eq_table.eq[dev->caps.num_comp_vectors].eqn, err);
1225 
1226 	for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
1227 		eq_set_ci(&priv->eq_table.eq[i], 1);
1228 
1229 	return 0;
1230 
1231 err_out_async:
1232 	mlx4_free_eq(dev, &priv->eq_table.eq[dev->caps.num_comp_vectors]);
1233 
1234 err_out_comp:
1235 	i = dev->caps.num_comp_vectors - 1;
1236 
1237 err_out_unmap:
1238 	while (i >= 0) {
1239 		mlx4_free_eq(dev, &priv->eq_table.eq[i]);
1240 		--i;
1241 	}
1242 	if (!mlx4_is_slave(dev))
1243 		mlx4_unmap_clr_int(dev);
1244 	mlx4_free_irqs(dev);
1245 
1246 err_out_bitmap:
1247 	mlx4_unmap_uar(dev);
1248 	mlx4_bitmap_cleanup(&priv->eq_table.bitmap);
1249 
1250 err_out_free:
1251 	kfree(priv->eq_table.uar_map);
1252 
1253 	return err;
1254 }
1255 
1256 void mlx4_cleanup_eq_table(struct mlx4_dev *dev)
1257 {
1258 	struct mlx4_priv *priv = mlx4_priv(dev);
1259 	int i;
1260 
1261 	mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 1,
1262 		    priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
1263 
1264 	mlx4_free_irqs(dev);
1265 
1266 	for (i = 0; i < dev->caps.num_comp_vectors + dev->caps.comp_pool + 1; ++i)
1267 		mlx4_free_eq(dev, &priv->eq_table.eq[i]);
1268 
1269 	if (!mlx4_is_slave(dev))
1270 		mlx4_unmap_clr_int(dev);
1271 
1272 	mlx4_unmap_uar(dev);
1273 	mlx4_bitmap_cleanup(&priv->eq_table.bitmap);
1274 
1275 	kfree(priv->eq_table.uar_map);
1276 }
1277 
1278 /* A test that verifies that we can accept interrupts on all
1279  * the irq vectors of the device.
1280  * Interrupts are checked using the NOP command.
1281  */
1282 int mlx4_test_interrupts(struct mlx4_dev *dev)
1283 {
1284 	struct mlx4_priv *priv = mlx4_priv(dev);
1285 	int i;
1286 	int err;
1287 
1288 	err = mlx4_NOP(dev);
1289 	/* When not in MSI_X, there is only one irq to check */
1290 	if (!(dev->flags & MLX4_FLAG_MSI_X) || mlx4_is_slave(dev))
1291 		return err;
1292 
1293 	/* A loop over all completion vectors, for each vector we will check
1294 	 * whether it works by mapping command completions to that vector
1295 	 * and performing a NOP command
1296 	 */
1297 	for(i = 0; !err && (i < dev->caps.num_comp_vectors); ++i) {
1298 		/* Temporary use polling for command completions */
1299 		mlx4_cmd_use_polling(dev);
1300 
1301 		/* Map the new eq to handle all asynchronous events */
1302 		err = mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1303 				  priv->eq_table.eq[i].eqn);
1304 		if (err) {
1305 			mlx4_warn(dev, "Failed mapping eq for interrupt test\n");
1306 			mlx4_cmd_use_events(dev);
1307 			break;
1308 		}
1309 
1310 		/* Go back to using events */
1311 		mlx4_cmd_use_events(dev);
1312 		err = mlx4_NOP(dev);
1313 	}
1314 
1315 	/* Return to default */
1316 	mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1317 		    priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
1318 	return err;
1319 }
1320 EXPORT_SYMBOL(mlx4_test_interrupts);
1321 
1322 int mlx4_assign_eq(struct mlx4_dev *dev, char *name, struct cpu_rmap *rmap,
1323 		   int *vector)
1324 {
1325 
1326 	struct mlx4_priv *priv = mlx4_priv(dev);
1327 	int vec = 0, err = 0, i;
1328 
1329 	mutex_lock(&priv->msix_ctl.pool_lock);
1330 	for (i = 0; !vec && i < dev->caps.comp_pool; i++) {
1331 		if (~priv->msix_ctl.pool_bm & 1ULL << i) {
1332 			priv->msix_ctl.pool_bm |= 1ULL << i;
1333 			vec = dev->caps.num_comp_vectors + 1 + i;
1334 			snprintf(priv->eq_table.irq_names +
1335 					vec * MLX4_IRQNAME_SIZE,
1336 					MLX4_IRQNAME_SIZE, "%s", name);
1337 #ifdef CONFIG_RFS_ACCEL
1338 			if (rmap) {
1339 				err = irq_cpu_rmap_add(rmap,
1340 						       priv->eq_table.eq[vec].irq);
1341 				if (err)
1342 					mlx4_warn(dev, "Failed adding irq rmap\n");
1343 			}
1344 #endif
1345 			err = request_irq(priv->eq_table.eq[vec].irq,
1346 					  mlx4_msi_x_interrupt, 0,
1347 					  &priv->eq_table.irq_names[vec<<5],
1348 					  priv->eq_table.eq + vec);
1349 			if (err) {
1350 				/*zero out bit by fliping it*/
1351 				priv->msix_ctl.pool_bm ^= 1 << i;
1352 				vec = 0;
1353 				continue;
1354 				/*we dont want to break here*/
1355 			}
1356 
1357 			eq_set_ci(&priv->eq_table.eq[vec], 1);
1358 		}
1359 	}
1360 	mutex_unlock(&priv->msix_ctl.pool_lock);
1361 
1362 	if (vec) {
1363 		*vector = vec;
1364 	} else {
1365 		*vector = 0;
1366 		err = (i == dev->caps.comp_pool) ? -ENOSPC : err;
1367 	}
1368 	return err;
1369 }
1370 EXPORT_SYMBOL(mlx4_assign_eq);
1371 
1372 int mlx4_eq_get_irq(struct mlx4_dev *dev, int vec)
1373 {
1374 	struct mlx4_priv *priv = mlx4_priv(dev);
1375 
1376 	return priv->eq_table.eq[vec].irq;
1377 }
1378 EXPORT_SYMBOL(mlx4_eq_get_irq);
1379 
1380 void mlx4_release_eq(struct mlx4_dev *dev, int vec)
1381 {
1382 	struct mlx4_priv *priv = mlx4_priv(dev);
1383 	/*bm index*/
1384 	int i = vec - dev->caps.num_comp_vectors - 1;
1385 
1386 	if (likely(i >= 0)) {
1387 		/*sanity check , making sure were not trying to free irq's
1388 		  Belonging to a legacy EQ*/
1389 		mutex_lock(&priv->msix_ctl.pool_lock);
1390 		if (priv->msix_ctl.pool_bm & 1ULL << i) {
1391 			free_irq(priv->eq_table.eq[vec].irq,
1392 				 &priv->eq_table.eq[vec]);
1393 			priv->msix_ctl.pool_bm &= ~(1ULL << i);
1394 		}
1395 		mutex_unlock(&priv->msix_ctl.pool_lock);
1396 	}
1397 
1398 }
1399 EXPORT_SYMBOL(mlx4_release_eq);
1400 
1401