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 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RECOVERABLE_ERROR_EVENT)
92 		async_ev_mask |= (1ull << MLX4_EVENT_TYPE_RECOVERABLE_ERROR_EVENT);
93 
94 	return async_ev_mask;
95 }
96 
97 static void eq_set_ci(struct mlx4_eq *eq, int req_not)
98 {
99 	__raw_writel((__force u32) cpu_to_be32((eq->cons_index & 0xffffff) |
100 					       req_not << 31),
101 		     eq->doorbell);
102 	/* We still want ordering, just not swabbing, so add a barrier */
103 	wmb();
104 }
105 
106 static struct mlx4_eqe *get_eqe(struct mlx4_eq *eq, u32 entry, u8 eqe_factor,
107 				u8 eqe_size)
108 {
109 	/* (entry & (eq->nent - 1)) gives us a cyclic array */
110 	unsigned long offset = (entry & (eq->nent - 1)) * eqe_size;
111 	/* CX3 is capable of extending the EQE from 32 to 64 bytes with
112 	 * strides of 64B,128B and 256B.
113 	 * When 64B EQE is used, the first (in the lower addresses)
114 	 * 32 bytes in the 64 byte EQE are reserved and the next 32 bytes
115 	 * contain the legacy EQE information.
116 	 * In all other cases, the first 32B contains the legacy EQE info.
117 	 */
118 	return eq->page_list[offset / PAGE_SIZE].buf + (offset + (eqe_factor ? MLX4_EQ_ENTRY_SIZE : 0)) % PAGE_SIZE;
119 }
120 
121 static struct mlx4_eqe *next_eqe_sw(struct mlx4_eq *eq, u8 eqe_factor, u8 size)
122 {
123 	struct mlx4_eqe *eqe = get_eqe(eq, eq->cons_index, eqe_factor, size);
124 	return !!(eqe->owner & 0x80) ^ !!(eq->cons_index & eq->nent) ? NULL : eqe;
125 }
126 
127 static struct mlx4_eqe *next_slave_event_eqe(struct mlx4_slave_event_eq *slave_eq)
128 {
129 	struct mlx4_eqe *eqe =
130 		&slave_eq->event_eqe[slave_eq->cons & (SLAVE_EVENT_EQ_SIZE - 1)];
131 	return (!!(eqe->owner & 0x80) ^
132 		!!(slave_eq->cons & SLAVE_EVENT_EQ_SIZE)) ?
133 		eqe : NULL;
134 }
135 
136 void mlx4_gen_slave_eqe(struct work_struct *work)
137 {
138 	struct mlx4_mfunc_master_ctx *master =
139 		container_of(work, struct mlx4_mfunc_master_ctx,
140 			     slave_event_work);
141 	struct mlx4_mfunc *mfunc =
142 		container_of(master, struct mlx4_mfunc, master);
143 	struct mlx4_priv *priv = container_of(mfunc, struct mlx4_priv, mfunc);
144 	struct mlx4_dev *dev = &priv->dev;
145 	struct mlx4_slave_event_eq *slave_eq = &mfunc->master.slave_eq;
146 	struct mlx4_eqe *eqe;
147 	u8 slave;
148 	int i, phys_port, slave_port;
149 
150 	for (eqe = next_slave_event_eqe(slave_eq); eqe;
151 	      eqe = next_slave_event_eqe(slave_eq)) {
152 		slave = eqe->slave_id;
153 
154 		if (eqe->type == MLX4_EVENT_TYPE_PORT_CHANGE &&
155 		    eqe->subtype == MLX4_PORT_CHANGE_SUBTYPE_DOWN &&
156 		    mlx4_is_bonded(dev)) {
157 			struct mlx4_port_cap port_cap;
158 
159 			if (!mlx4_QUERY_PORT(dev, 1, &port_cap) && port_cap.link_state)
160 				goto consume;
161 
162 			if (!mlx4_QUERY_PORT(dev, 2, &port_cap) && port_cap.link_state)
163 				goto consume;
164 		}
165 		/* All active slaves need to receive the event */
166 		if (slave == ALL_SLAVES) {
167 			for (i = 0; i <= dev->persist->num_vfs; i++) {
168 				phys_port = 0;
169 				if (eqe->type == MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT &&
170 				    eqe->subtype == MLX4_DEV_PMC_SUBTYPE_PORT_INFO) {
171 					phys_port  = eqe->event.port_mgmt_change.port;
172 					slave_port = mlx4_phys_to_slave_port(dev, i, phys_port);
173 					if (slave_port < 0) /* VF doesn't have this port */
174 						continue;
175 					eqe->event.port_mgmt_change.port = slave_port;
176 				}
177 				if (mlx4_GEN_EQE(dev, i, eqe))
178 					mlx4_warn(dev, "Failed to generate event for slave %d\n",
179 						  i);
180 				if (phys_port)
181 					eqe->event.port_mgmt_change.port = phys_port;
182 			}
183 		} else {
184 			if (mlx4_GEN_EQE(dev, slave, eqe))
185 				mlx4_warn(dev, "Failed to generate event for slave %d\n",
186 					  slave);
187 		}
188 consume:
189 		++slave_eq->cons;
190 	}
191 }
192 
193 
194 static void slave_event(struct mlx4_dev *dev, u8 slave, struct mlx4_eqe *eqe)
195 {
196 	struct mlx4_priv *priv = mlx4_priv(dev);
197 	struct mlx4_slave_event_eq *slave_eq = &priv->mfunc.master.slave_eq;
198 	struct mlx4_eqe *s_eqe;
199 	unsigned long flags;
200 
201 	spin_lock_irqsave(&slave_eq->event_lock, flags);
202 	s_eqe = &slave_eq->event_eqe[slave_eq->prod & (SLAVE_EVENT_EQ_SIZE - 1)];
203 	if ((!!(s_eqe->owner & 0x80)) ^
204 	    (!!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE))) {
205 		mlx4_warn(dev, "Master failed to generate an EQE for slave: %d. No free EQE on slave events queue\n",
206 			  slave);
207 		spin_unlock_irqrestore(&slave_eq->event_lock, flags);
208 		return;
209 	}
210 
211 	memcpy(s_eqe, eqe, sizeof(struct mlx4_eqe) - 1);
212 	s_eqe->slave_id = slave;
213 	/* ensure all information is written before setting the ownersip bit */
214 	dma_wmb();
215 	s_eqe->owner = !!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE) ? 0x0 : 0x80;
216 	++slave_eq->prod;
217 
218 	queue_work(priv->mfunc.master.comm_wq,
219 		   &priv->mfunc.master.slave_event_work);
220 	spin_unlock_irqrestore(&slave_eq->event_lock, flags);
221 }
222 
223 static void mlx4_slave_event(struct mlx4_dev *dev, int slave,
224 			     struct mlx4_eqe *eqe)
225 {
226 	struct mlx4_priv *priv = mlx4_priv(dev);
227 
228 	if (slave < 0 || slave > dev->persist->num_vfs ||
229 	    slave == dev->caps.function ||
230 	    !priv->mfunc.master.slave_state[slave].active)
231 		return;
232 
233 	slave_event(dev, slave, eqe);
234 }
235 
236 #if defined(CONFIG_SMP)
237 static void mlx4_set_eq_affinity_hint(struct mlx4_priv *priv, int vec)
238 {
239 	int hint_err;
240 	struct mlx4_dev *dev = &priv->dev;
241 	struct mlx4_eq *eq = &priv->eq_table.eq[vec];
242 
243 	if (!cpumask_available(eq->affinity_mask) ||
244 	    cpumask_empty(eq->affinity_mask))
245 		return;
246 
247 	hint_err = irq_update_affinity_hint(eq->irq, eq->affinity_mask);
248 	if (hint_err)
249 		mlx4_warn(dev, "irq_update_affinity_hint failed, err %d\n", hint_err);
250 }
251 #endif
252 
253 int mlx4_gen_pkey_eqe(struct mlx4_dev *dev, int slave, u8 port)
254 {
255 	struct mlx4_eqe eqe;
256 
257 	struct mlx4_priv *priv = mlx4_priv(dev);
258 	struct mlx4_slave_state *s_slave = &priv->mfunc.master.slave_state[slave];
259 
260 	if (!s_slave->active)
261 		return 0;
262 
263 	memset(&eqe, 0, sizeof(eqe));
264 
265 	eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
266 	eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE;
267 	eqe.event.port_mgmt_change.port = mlx4_phys_to_slave_port(dev, slave, port);
268 
269 	return mlx4_GEN_EQE(dev, slave, &eqe);
270 }
271 EXPORT_SYMBOL(mlx4_gen_pkey_eqe);
272 
273 int mlx4_gen_guid_change_eqe(struct mlx4_dev *dev, int slave, u8 port)
274 {
275 	struct mlx4_eqe eqe;
276 
277 	/*don't send if we don't have the that slave */
278 	if (dev->persist->num_vfs < slave)
279 		return 0;
280 	memset(&eqe, 0, sizeof(eqe));
281 
282 	eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
283 	eqe.subtype = MLX4_DEV_PMC_SUBTYPE_GUID_INFO;
284 	eqe.event.port_mgmt_change.port = mlx4_phys_to_slave_port(dev, slave, port);
285 
286 	return mlx4_GEN_EQE(dev, slave, &eqe);
287 }
288 EXPORT_SYMBOL(mlx4_gen_guid_change_eqe);
289 
290 int mlx4_gen_port_state_change_eqe(struct mlx4_dev *dev, int slave, u8 port,
291 				   u8 port_subtype_change)
292 {
293 	struct mlx4_eqe eqe;
294 	u8 slave_port = mlx4_phys_to_slave_port(dev, slave, port);
295 
296 	/*don't send if we don't have the that slave */
297 	if (dev->persist->num_vfs < slave)
298 		return 0;
299 	memset(&eqe, 0, sizeof(eqe));
300 
301 	eqe.type = MLX4_EVENT_TYPE_PORT_CHANGE;
302 	eqe.subtype = port_subtype_change;
303 	eqe.event.port_change.port = cpu_to_be32(slave_port << 28);
304 
305 	mlx4_dbg(dev, "%s: sending: %d to slave: %d on port: %d\n", __func__,
306 		 port_subtype_change, slave, port);
307 	return mlx4_GEN_EQE(dev, slave, &eqe);
308 }
309 EXPORT_SYMBOL(mlx4_gen_port_state_change_eqe);
310 
311 enum slave_port_state mlx4_get_slave_port_state(struct mlx4_dev *dev, int slave, u8 port)
312 {
313 	struct mlx4_priv *priv = mlx4_priv(dev);
314 	struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state;
315 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
316 
317 	if (slave >= dev->num_slaves || port > dev->caps.num_ports ||
318 	    port <= 0 || !test_bit(port - 1, actv_ports.ports)) {
319 		pr_err("%s: Error: asking for slave:%d, port:%d\n",
320 		       __func__, slave, port);
321 		return SLAVE_PORT_DOWN;
322 	}
323 	return s_state[slave].port_state[port];
324 }
325 EXPORT_SYMBOL(mlx4_get_slave_port_state);
326 
327 static int mlx4_set_slave_port_state(struct mlx4_dev *dev, int slave, u8 port,
328 				     enum slave_port_state state)
329 {
330 	struct mlx4_priv *priv = mlx4_priv(dev);
331 	struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state;
332 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
333 
334 	if (slave >= dev->num_slaves || port > dev->caps.num_ports ||
335 	    port <= 0 || !test_bit(port - 1, actv_ports.ports)) {
336 		pr_err("%s: Error: asking for slave:%d, port:%d\n",
337 		       __func__, slave, port);
338 		return -1;
339 	}
340 	s_state[slave].port_state[port] = state;
341 
342 	return 0;
343 }
344 
345 static void set_all_slave_state(struct mlx4_dev *dev, u8 port, int event)
346 {
347 	int i;
348 	enum slave_port_gen_event gen_event;
349 	struct mlx4_slaves_pport slaves_pport = mlx4_phys_to_slaves_pport(dev,
350 									  port);
351 
352 	for (i = 0; i < dev->persist->num_vfs + 1; i++)
353 		if (test_bit(i, slaves_pport.slaves))
354 			set_and_calc_slave_port_state(dev, i, port,
355 						      event, &gen_event);
356 }
357 /**************************************************************************
358 	The function get as input the new event to that port,
359 	and according to the prev state change the slave's port state.
360 	The events are:
361 		MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
362 		MLX4_PORT_STATE_DEV_EVENT_PORT_UP
363 		MLX4_PORT_STATE_IB_EVENT_GID_VALID
364 		MLX4_PORT_STATE_IB_EVENT_GID_INVALID
365 ***************************************************************************/
366 int set_and_calc_slave_port_state(struct mlx4_dev *dev, int slave,
367 				  u8 port, int event,
368 				  enum slave_port_gen_event *gen_event)
369 {
370 	struct mlx4_priv *priv = mlx4_priv(dev);
371 	struct mlx4_slave_state *ctx = NULL;
372 	unsigned long flags;
373 	int ret = -1;
374 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
375 	enum slave_port_state cur_state =
376 		mlx4_get_slave_port_state(dev, slave, port);
377 
378 	*gen_event = SLAVE_PORT_GEN_EVENT_NONE;
379 
380 	if (slave >= dev->num_slaves || port > dev->caps.num_ports ||
381 	    port <= 0 || !test_bit(port - 1, actv_ports.ports)) {
382 		pr_err("%s: Error: asking for slave:%d, port:%d\n",
383 		       __func__, slave, port);
384 		return ret;
385 	}
386 
387 	ctx = &priv->mfunc.master.slave_state[slave];
388 	spin_lock_irqsave(&ctx->lock, flags);
389 
390 	switch (cur_state) {
391 	case SLAVE_PORT_DOWN:
392 		if (MLX4_PORT_STATE_DEV_EVENT_PORT_UP == event)
393 			mlx4_set_slave_port_state(dev, slave, port,
394 						  SLAVE_PENDING_UP);
395 		break;
396 	case SLAVE_PENDING_UP:
397 		if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event)
398 			mlx4_set_slave_port_state(dev, slave, port,
399 						  SLAVE_PORT_DOWN);
400 		else if (MLX4_PORT_STATE_IB_PORT_STATE_EVENT_GID_VALID == event) {
401 			mlx4_set_slave_port_state(dev, slave, port,
402 						  SLAVE_PORT_UP);
403 			*gen_event = SLAVE_PORT_GEN_EVENT_UP;
404 		}
405 		break;
406 	case SLAVE_PORT_UP:
407 		if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event) {
408 			mlx4_set_slave_port_state(dev, slave, port,
409 						  SLAVE_PORT_DOWN);
410 			*gen_event = SLAVE_PORT_GEN_EVENT_DOWN;
411 		} else if (MLX4_PORT_STATE_IB_EVENT_GID_INVALID ==
412 				event) {
413 			mlx4_set_slave_port_state(dev, slave, port,
414 						  SLAVE_PENDING_UP);
415 			*gen_event = SLAVE_PORT_GEN_EVENT_DOWN;
416 		}
417 		break;
418 	default:
419 		pr_err("%s: BUG!!! UNKNOWN state: slave:%d, port:%d\n",
420 		       __func__, slave, port);
421 		goto out;
422 	}
423 	ret = mlx4_get_slave_port_state(dev, slave, port);
424 
425 out:
426 	spin_unlock_irqrestore(&ctx->lock, flags);
427 	return ret;
428 }
429 
430 EXPORT_SYMBOL(set_and_calc_slave_port_state);
431 
432 int mlx4_gen_slaves_port_mgt_ev(struct mlx4_dev *dev, u8 port, int attr)
433 {
434 	struct mlx4_eqe eqe;
435 
436 	memset(&eqe, 0, sizeof(eqe));
437 
438 	eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
439 	eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PORT_INFO;
440 	eqe.event.port_mgmt_change.port = port;
441 	eqe.event.port_mgmt_change.params.port_info.changed_attr =
442 		cpu_to_be32((u32) attr);
443 
444 	slave_event(dev, ALL_SLAVES, &eqe);
445 	return 0;
446 }
447 EXPORT_SYMBOL(mlx4_gen_slaves_port_mgt_ev);
448 
449 void mlx4_master_handle_slave_flr(struct work_struct *work)
450 {
451 	struct mlx4_mfunc_master_ctx *master =
452 		container_of(work, struct mlx4_mfunc_master_ctx,
453 			     slave_flr_event_work);
454 	struct mlx4_mfunc *mfunc =
455 		container_of(master, struct mlx4_mfunc, master);
456 	struct mlx4_priv *priv =
457 		container_of(mfunc, struct mlx4_priv, mfunc);
458 	struct mlx4_dev *dev = &priv->dev;
459 	struct mlx4_slave_state *slave_state = priv->mfunc.master.slave_state;
460 	int i;
461 	int err;
462 	unsigned long flags;
463 
464 	mlx4_dbg(dev, "mlx4_handle_slave_flr\n");
465 
466 	for (i = 0 ; i < dev->num_slaves; i++) {
467 
468 		if (MLX4_COMM_CMD_FLR == slave_state[i].last_cmd) {
469 			mlx4_dbg(dev, "mlx4_handle_slave_flr: clean slave: %d\n",
470 				 i);
471 			/* In case of 'Reset flow' FLR can be generated for
472 			 * a slave before mlx4_load_one is done.
473 			 * make sure interface is up before trying to delete
474 			 * slave resources which weren't allocated yet.
475 			 */
476 			if (dev->persist->interface_state &
477 			    MLX4_INTERFACE_STATE_UP)
478 				mlx4_delete_all_resources_for_slave(dev, i);
479 			/*return the slave to running mode*/
480 			spin_lock_irqsave(&priv->mfunc.master.slave_state_lock, flags);
481 			slave_state[i].last_cmd = MLX4_COMM_CMD_RESET;
482 			slave_state[i].is_slave_going_down = 0;
483 			spin_unlock_irqrestore(&priv->mfunc.master.slave_state_lock, flags);
484 			/*notify the FW:*/
485 			err = mlx4_cmd(dev, 0, i, 0, MLX4_CMD_INFORM_FLR_DONE,
486 				       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
487 			if (err)
488 				mlx4_warn(dev, "Failed to notify FW on FLR done (slave:%d)\n",
489 					  i);
490 		}
491 	}
492 }
493 
494 static int mlx4_eq_int(struct mlx4_dev *dev, struct mlx4_eq *eq)
495 {
496 	struct mlx4_priv *priv = mlx4_priv(dev);
497 	struct mlx4_eqe *eqe;
498 	int cqn;
499 	int eqes_found = 0;
500 	int set_ci = 0;
501 	int port;
502 	int slave = 0;
503 	int ret;
504 	u32 flr_slave;
505 	u8 update_slave_state;
506 	int i;
507 	enum slave_port_gen_event gen_event;
508 	unsigned long flags;
509 	struct mlx4_vport_state *s_info;
510 	int eqe_size = dev->caps.eqe_size;
511 
512 	while ((eqe = next_eqe_sw(eq, dev->caps.eqe_factor, eqe_size))) {
513 		/*
514 		 * Make sure we read EQ entry contents after we've
515 		 * checked the ownership bit.
516 		 */
517 		dma_rmb();
518 
519 		switch (eqe->type) {
520 		case MLX4_EVENT_TYPE_COMP:
521 			cqn = be32_to_cpu(eqe->event.comp.cqn) & 0xffffff;
522 			mlx4_cq_completion(dev, cqn);
523 			break;
524 
525 		case MLX4_EVENT_TYPE_PATH_MIG:
526 		case MLX4_EVENT_TYPE_COMM_EST:
527 		case MLX4_EVENT_TYPE_SQ_DRAINED:
528 		case MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE:
529 		case MLX4_EVENT_TYPE_WQ_CATAS_ERROR:
530 		case MLX4_EVENT_TYPE_PATH_MIG_FAILED:
531 		case MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
532 		case MLX4_EVENT_TYPE_WQ_ACCESS_ERROR:
533 			mlx4_dbg(dev, "event %d arrived\n", eqe->type);
534 			if (mlx4_is_master(dev)) {
535 				/* forward only to slave owning the QP */
536 				ret = mlx4_get_slave_from_resource_id(dev,
537 						RES_QP,
538 						be32_to_cpu(eqe->event.qp.qpn)
539 						& 0xffffff, &slave);
540 				if (ret && ret != -ENOENT) {
541 					mlx4_dbg(dev, "QP event %02x(%02x) on EQ %d at index %u: could not get slave id (%d)\n",
542 						 eqe->type, eqe->subtype,
543 						 eq->eqn, eq->cons_index, ret);
544 					break;
545 				}
546 
547 				if (!ret && slave != dev->caps.function) {
548 					mlx4_slave_event(dev, slave, eqe);
549 					break;
550 				}
551 
552 			}
553 			mlx4_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) &
554 				      0xffffff, eqe->type);
555 			break;
556 
557 		case MLX4_EVENT_TYPE_SRQ_LIMIT:
558 			mlx4_dbg(dev, "%s: MLX4_EVENT_TYPE_SRQ_LIMIT. srq_no=0x%x, eq 0x%x\n",
559 				 __func__, be32_to_cpu(eqe->event.srq.srqn),
560 				 eq->eqn);
561 			fallthrough;
562 		case MLX4_EVENT_TYPE_SRQ_CATAS_ERROR:
563 			if (mlx4_is_master(dev)) {
564 				/* forward only to slave owning the SRQ */
565 				ret = mlx4_get_slave_from_resource_id(dev,
566 						RES_SRQ,
567 						be32_to_cpu(eqe->event.srq.srqn)
568 						& 0xffffff,
569 						&slave);
570 				if (ret && ret != -ENOENT) {
571 					mlx4_warn(dev, "SRQ event %02x(%02x) on EQ %d at index %u: could not get slave id (%d)\n",
572 						  eqe->type, eqe->subtype,
573 						  eq->eqn, eq->cons_index, ret);
574 					break;
575 				}
576 				if (eqe->type ==
577 				    MLX4_EVENT_TYPE_SRQ_CATAS_ERROR)
578 					mlx4_warn(dev, "%s: slave:%d, srq_no:0x%x, event: %02x(%02x)\n",
579 						  __func__, slave,
580 						  be32_to_cpu(eqe->event.srq.srqn),
581 						  eqe->type, eqe->subtype);
582 
583 				if (!ret && slave != dev->caps.function) {
584 					if (eqe->type ==
585 					    MLX4_EVENT_TYPE_SRQ_CATAS_ERROR)
586 						mlx4_warn(dev, "%s: sending event %02x(%02x) to slave:%d\n",
587 							  __func__, eqe->type,
588 							  eqe->subtype, slave);
589 					mlx4_slave_event(dev, slave, eqe);
590 					break;
591 				}
592 			}
593 			mlx4_srq_event(dev, be32_to_cpu(eqe->event.srq.srqn) &
594 				       0xffffff, eqe->type);
595 			break;
596 
597 		case MLX4_EVENT_TYPE_CMD:
598 			mlx4_cmd_event(dev,
599 				       be16_to_cpu(eqe->event.cmd.token),
600 				       eqe->event.cmd.status,
601 				       be64_to_cpu(eqe->event.cmd.out_param));
602 			break;
603 
604 		case MLX4_EVENT_TYPE_PORT_CHANGE: {
605 			struct mlx4_slaves_pport slaves_port;
606 			port = be32_to_cpu(eqe->event.port_change.port) >> 28;
607 			slaves_port = mlx4_phys_to_slaves_pport(dev, port);
608 			if (eqe->subtype == MLX4_PORT_CHANGE_SUBTYPE_DOWN) {
609 				mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_DOWN,
610 						    port);
611 				mlx4_priv(dev)->sense.do_sense_port[port] = 1;
612 				if (!mlx4_is_master(dev))
613 					break;
614 				for (i = 0; i < dev->persist->num_vfs + 1;
615 				     i++) {
616 					int reported_port = mlx4_is_bonded(dev) ? 1 : mlx4_phys_to_slave_port(dev, i, port);
617 
618 					if (!test_bit(i, slaves_port.slaves) && !mlx4_is_bonded(dev))
619 						continue;
620 					if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH) {
621 						if (i == mlx4_master_func_num(dev))
622 							continue;
623 						mlx4_dbg(dev, "%s: Sending MLX4_PORT_CHANGE_SUBTYPE_DOWN to slave: %d, port:%d\n",
624 							 __func__, i, port);
625 						s_info = &priv->mfunc.master.vf_oper[i].vport[port].state;
626 						if (IFLA_VF_LINK_STATE_AUTO == s_info->link_state) {
627 							eqe->event.port_change.port =
628 								cpu_to_be32(
629 								(be32_to_cpu(eqe->event.port_change.port) & 0xFFFFFFF)
630 								| (reported_port << 28));
631 							mlx4_slave_event(dev, i, eqe);
632 						}
633 					} else {  /* IB port */
634 						set_and_calc_slave_port_state(dev, i, port,
635 									      MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
636 									      &gen_event);
637 						/*we can be in pending state, then do not send port_down event*/
638 						if (SLAVE_PORT_GEN_EVENT_DOWN ==  gen_event) {
639 							if (i == mlx4_master_func_num(dev))
640 								continue;
641 							eqe->event.port_change.port =
642 								cpu_to_be32(
643 								(be32_to_cpu(eqe->event.port_change.port) & 0xFFFFFFF)
644 								| (mlx4_phys_to_slave_port(dev, i, port) << 28));
645 							mlx4_slave_event(dev, i, eqe);
646 						}
647 					}
648 				}
649 			} else {
650 				mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_UP, port);
651 
652 				mlx4_priv(dev)->sense.do_sense_port[port] = 0;
653 
654 				if (!mlx4_is_master(dev))
655 					break;
656 				if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
657 					for (i = 0;
658 					     i < dev->persist->num_vfs + 1;
659 					     i++) {
660 						int reported_port = mlx4_is_bonded(dev) ? 1 : mlx4_phys_to_slave_port(dev, i, port);
661 
662 						if (!test_bit(i, slaves_port.slaves) && !mlx4_is_bonded(dev))
663 							continue;
664 						if (i == mlx4_master_func_num(dev))
665 							continue;
666 						s_info = &priv->mfunc.master.vf_oper[i].vport[port].state;
667 						if (IFLA_VF_LINK_STATE_AUTO == s_info->link_state) {
668 							eqe->event.port_change.port =
669 								cpu_to_be32(
670 								(be32_to_cpu(eqe->event.port_change.port) & 0xFFFFFFF)
671 								| (reported_port << 28));
672 							mlx4_slave_event(dev, i, eqe);
673 						}
674 					}
675 				else /* IB port */
676 					/* port-up event will be sent to a slave when the
677 					 * slave's alias-guid is set. This is done in alias_GUID.c
678 					 */
679 					set_all_slave_state(dev, port, MLX4_DEV_EVENT_PORT_UP);
680 			}
681 			break;
682 		}
683 
684 		case MLX4_EVENT_TYPE_CQ_ERROR:
685 			mlx4_warn(dev, "CQ %s on CQN %06x\n",
686 				  eqe->event.cq_err.syndrome == 1 ?
687 				  "overrun" : "access violation",
688 				  be32_to_cpu(eqe->event.cq_err.cqn) & 0xffffff);
689 			if (mlx4_is_master(dev)) {
690 				ret = mlx4_get_slave_from_resource_id(dev,
691 					RES_CQ,
692 					be32_to_cpu(eqe->event.cq_err.cqn)
693 					& 0xffffff, &slave);
694 				if (ret && ret != -ENOENT) {
695 					mlx4_dbg(dev, "CQ event %02x(%02x) on EQ %d at index %u: could not get slave id (%d)\n",
696 						 eqe->type, eqe->subtype,
697 						 eq->eqn, eq->cons_index, ret);
698 					break;
699 				}
700 
701 				if (!ret && slave != dev->caps.function) {
702 					mlx4_slave_event(dev, slave, eqe);
703 					break;
704 				}
705 			}
706 			mlx4_cq_event(dev,
707 				      be32_to_cpu(eqe->event.cq_err.cqn)
708 				      & 0xffffff,
709 				      eqe->type);
710 			break;
711 
712 		case MLX4_EVENT_TYPE_EQ_OVERFLOW:
713 			mlx4_warn(dev, "EQ overrun on EQN %d\n", eq->eqn);
714 			break;
715 
716 		case MLX4_EVENT_TYPE_OP_REQUIRED:
717 			atomic_inc(&priv->opreq_count);
718 			/* FW commands can't be executed from interrupt context
719 			 * working in deferred task
720 			 */
721 			queue_work(mlx4_wq, &priv->opreq_task);
722 			break;
723 
724 		case MLX4_EVENT_TYPE_COMM_CHANNEL:
725 			if (!mlx4_is_master(dev)) {
726 				mlx4_warn(dev, "Received comm channel event for non master device\n");
727 				break;
728 			}
729 			memcpy(&priv->mfunc.master.comm_arm_bit_vector,
730 			       eqe->event.comm_channel_arm.bit_vec,
731 			       sizeof(eqe->event.comm_channel_arm.bit_vec));
732 			queue_work(priv->mfunc.master.comm_wq,
733 				   &priv->mfunc.master.comm_work);
734 			break;
735 
736 		case MLX4_EVENT_TYPE_FLR_EVENT:
737 			flr_slave = be32_to_cpu(eqe->event.flr_event.slave_id);
738 			if (!mlx4_is_master(dev)) {
739 				mlx4_warn(dev, "Non-master function received FLR event\n");
740 				break;
741 			}
742 
743 			mlx4_dbg(dev, "FLR event for slave: %d\n", flr_slave);
744 
745 			if (flr_slave >= dev->num_slaves) {
746 				mlx4_warn(dev,
747 					  "Got FLR for unknown function: %d\n",
748 					  flr_slave);
749 				update_slave_state = 0;
750 			} else
751 				update_slave_state = 1;
752 
753 			spin_lock_irqsave(&priv->mfunc.master.slave_state_lock, flags);
754 			if (update_slave_state) {
755 				priv->mfunc.master.slave_state[flr_slave].active = false;
756 				priv->mfunc.master.slave_state[flr_slave].last_cmd = MLX4_COMM_CMD_FLR;
757 				priv->mfunc.master.slave_state[flr_slave].is_slave_going_down = 1;
758 			}
759 			spin_unlock_irqrestore(&priv->mfunc.master.slave_state_lock, flags);
760 			mlx4_dispatch_event(dev, MLX4_DEV_EVENT_SLAVE_SHUTDOWN,
761 					    flr_slave);
762 			queue_work(priv->mfunc.master.comm_wq,
763 				   &priv->mfunc.master.slave_flr_event_work);
764 			break;
765 
766 		case MLX4_EVENT_TYPE_FATAL_WARNING:
767 			if (eqe->subtype == MLX4_FATAL_WARNING_SUBTYPE_WARMING) {
768 				if (mlx4_is_master(dev))
769 					for (i = 0; i < dev->num_slaves; i++) {
770 						mlx4_dbg(dev, "%s: Sending MLX4_FATAL_WARNING_SUBTYPE_WARMING to slave: %d\n",
771 							 __func__, i);
772 						if (i == dev->caps.function)
773 							continue;
774 						mlx4_slave_event(dev, i, eqe);
775 					}
776 				mlx4_err(dev, "Temperature Threshold was reached! Threshold: %d celsius degrees; Current Temperature: %d\n",
777 					 be16_to_cpu(eqe->event.warming.warning_threshold),
778 					 be16_to_cpu(eqe->event.warming.current_temperature));
779 			} else
780 				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",
781 					  eqe->type, eqe->subtype, eq->eqn,
782 					  eq->cons_index, eqe->owner, eq->nent,
783 					  eqe->slave_id,
784 					  !!(eqe->owner & 0x80) ^
785 					  !!(eq->cons_index & eq->nent) ? "HW" : "SW");
786 
787 			break;
788 
789 		case MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT:
790 			mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_MGMT_CHANGE,
791 					    (unsigned long) eqe);
792 			break;
793 
794 		case MLX4_EVENT_TYPE_RECOVERABLE_ERROR_EVENT:
795 			switch (eqe->subtype) {
796 			case MLX4_RECOVERABLE_ERROR_EVENT_SUBTYPE_BAD_CABLE:
797 				mlx4_warn(dev, "Bad cable detected on port %u\n",
798 					  eqe->event.bad_cable.port);
799 				break;
800 			case MLX4_RECOVERABLE_ERROR_EVENT_SUBTYPE_UNSUPPORTED_CABLE:
801 				mlx4_warn(dev, "Unsupported cable detected\n");
802 				break;
803 			default:
804 				mlx4_dbg(dev,
805 					 "Unhandled recoverable error event detected: %02x(%02x) on EQ %d at index %u. owner=%x, nent=0x%x, ownership=%s\n",
806 					 eqe->type, eqe->subtype, eq->eqn,
807 					 eq->cons_index, eqe->owner, eq->nent,
808 					 !!(eqe->owner & 0x80) ^
809 					 !!(eq->cons_index & eq->nent) ? "HW" : "SW");
810 				break;
811 			}
812 			break;
813 
814 		case MLX4_EVENT_TYPE_EEC_CATAS_ERROR:
815 		case MLX4_EVENT_TYPE_ECC_DETECT:
816 		default:
817 			mlx4_warn(dev, "Unhandled event %02x(%02x) on EQ %d at index %u. owner=%x, nent=0x%x, slave=%x, ownership=%s\n",
818 				  eqe->type, eqe->subtype, eq->eqn,
819 				  eq->cons_index, eqe->owner, eq->nent,
820 				  eqe->slave_id,
821 				  !!(eqe->owner & 0x80) ^
822 				  !!(eq->cons_index & eq->nent) ? "HW" : "SW");
823 			break;
824 		}
825 
826 		++eq->cons_index;
827 		eqes_found = 1;
828 		++set_ci;
829 
830 		/*
831 		 * The HCA will think the queue has overflowed if we
832 		 * don't tell it we've been processing events.  We
833 		 * create our EQs with MLX4_NUM_SPARE_EQE extra
834 		 * entries, so we must update our consumer index at
835 		 * least that often.
836 		 */
837 		if (unlikely(set_ci >= MLX4_NUM_SPARE_EQE)) {
838 			eq_set_ci(eq, 0);
839 			set_ci = 0;
840 		}
841 	}
842 
843 	eq_set_ci(eq, 1);
844 
845 	return eqes_found;
846 }
847 
848 static irqreturn_t mlx4_interrupt(int irq, void *dev_ptr)
849 {
850 	struct mlx4_dev *dev = dev_ptr;
851 	struct mlx4_priv *priv = mlx4_priv(dev);
852 	int work = 0;
853 	int i;
854 
855 	writel(priv->eq_table.clr_mask, priv->eq_table.clr_int);
856 
857 	for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
858 		work |= mlx4_eq_int(dev, &priv->eq_table.eq[i]);
859 
860 	return IRQ_RETVAL(work);
861 }
862 
863 static irqreturn_t mlx4_msi_x_interrupt(int irq, void *eq_ptr)
864 {
865 	struct mlx4_eq  *eq  = eq_ptr;
866 	struct mlx4_dev *dev = eq->dev;
867 
868 	mlx4_eq_int(dev, eq);
869 
870 	/* MSI-X vectors always belong to us */
871 	return IRQ_HANDLED;
872 }
873 
874 int mlx4_MAP_EQ_wrapper(struct mlx4_dev *dev, int slave,
875 			struct mlx4_vhcr *vhcr,
876 			struct mlx4_cmd_mailbox *inbox,
877 			struct mlx4_cmd_mailbox *outbox,
878 			struct mlx4_cmd_info *cmd)
879 {
880 	struct mlx4_priv *priv = mlx4_priv(dev);
881 	struct mlx4_slave_event_eq_info *event_eq =
882 		priv->mfunc.master.slave_state[slave].event_eq;
883 	u32 in_modifier = vhcr->in_modifier;
884 	u32 eqn = in_modifier & 0x3FF;
885 	u64 in_param =  vhcr->in_param;
886 	int err = 0;
887 	int i;
888 
889 	if (slave == dev->caps.function)
890 		err = mlx4_cmd(dev, in_param, (in_modifier & 0x80000000) | eqn,
891 			       0, MLX4_CMD_MAP_EQ, MLX4_CMD_TIME_CLASS_B,
892 			       MLX4_CMD_NATIVE);
893 	if (!err)
894 		for (i = 0; i < MLX4_EVENT_TYPES_NUM; ++i)
895 			if (in_param & (1LL << i))
896 				event_eq[i].eqn = in_modifier >> 31 ? -1 : eqn;
897 
898 	return err;
899 }
900 
901 static int mlx4_MAP_EQ(struct mlx4_dev *dev, u64 event_mask, int unmap,
902 			int eq_num)
903 {
904 	return mlx4_cmd(dev, event_mask, (unmap << 31) | eq_num,
905 			0, MLX4_CMD_MAP_EQ, MLX4_CMD_TIME_CLASS_B,
906 			MLX4_CMD_WRAPPED);
907 }
908 
909 static int mlx4_SW2HW_EQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
910 			 int eq_num)
911 {
912 	return mlx4_cmd(dev, mailbox->dma, eq_num, 0,
913 			MLX4_CMD_SW2HW_EQ, MLX4_CMD_TIME_CLASS_A,
914 			MLX4_CMD_WRAPPED);
915 }
916 
917 static int mlx4_HW2SW_EQ(struct mlx4_dev *dev,  int eq_num)
918 {
919 	return mlx4_cmd(dev, 0, eq_num, 1, MLX4_CMD_HW2SW_EQ,
920 			MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
921 }
922 
923 static int mlx4_num_eq_uar(struct mlx4_dev *dev)
924 {
925 	/*
926 	 * Each UAR holds 4 EQ doorbells.  To figure out how many UARs
927 	 * we need to map, take the difference of highest index and
928 	 * the lowest index we'll use and add 1.
929 	 */
930 	return (dev->caps.num_comp_vectors + 1 + dev->caps.reserved_eqs) / 4 -
931 		dev->caps.reserved_eqs / 4 + 1;
932 }
933 
934 static void __iomem *mlx4_get_eq_uar(struct mlx4_dev *dev, struct mlx4_eq *eq)
935 {
936 	struct mlx4_priv *priv = mlx4_priv(dev);
937 	int index;
938 
939 	index = eq->eqn / 4 - dev->caps.reserved_eqs / 4;
940 
941 	if (!priv->eq_table.uar_map[index]) {
942 		priv->eq_table.uar_map[index] =
943 			ioremap(
944 				pci_resource_start(dev->persist->pdev, 2) +
945 				((eq->eqn / 4) << (dev->uar_page_shift)),
946 				(1 << (dev->uar_page_shift)));
947 		if (!priv->eq_table.uar_map[index]) {
948 			mlx4_err(dev, "Couldn't map EQ doorbell for EQN 0x%06x\n",
949 				 eq->eqn);
950 			return NULL;
951 		}
952 	}
953 
954 	return priv->eq_table.uar_map[index] + 0x800 + 8 * (eq->eqn % 4);
955 }
956 
957 static void mlx4_unmap_uar(struct mlx4_dev *dev)
958 {
959 	struct mlx4_priv *priv = mlx4_priv(dev);
960 	int i;
961 
962 	for (i = 0; i < mlx4_num_eq_uar(dev); ++i)
963 		if (priv->eq_table.uar_map[i]) {
964 			iounmap(priv->eq_table.uar_map[i]);
965 			priv->eq_table.uar_map[i] = NULL;
966 		}
967 }
968 
969 static int mlx4_create_eq(struct mlx4_dev *dev, int nent,
970 			  u8 intr, struct mlx4_eq *eq)
971 {
972 	struct mlx4_priv *priv = mlx4_priv(dev);
973 	struct mlx4_cmd_mailbox *mailbox;
974 	struct mlx4_eq_context *eq_context;
975 	int npages;
976 	u64 *dma_list = NULL;
977 	dma_addr_t t;
978 	u64 mtt_addr;
979 	int err = -ENOMEM;
980 	int i;
981 
982 	eq->dev   = dev;
983 	eq->nent  = roundup_pow_of_two(max(nent, 2));
984 	/* CX3 is capable of extending the CQE/EQE from 32 to 64 bytes, with
985 	 * strides of 64B,128B and 256B.
986 	 */
987 	npages = PAGE_ALIGN(eq->nent * dev->caps.eqe_size) / PAGE_SIZE;
988 
989 	eq->page_list = kmalloc_array(npages, sizeof(*eq->page_list),
990 				      GFP_KERNEL);
991 	if (!eq->page_list)
992 		goto err_out;
993 
994 	for (i = 0; i < npages; ++i)
995 		eq->page_list[i].buf = NULL;
996 
997 	dma_list = kmalloc_array(npages, sizeof(*dma_list), GFP_KERNEL);
998 	if (!dma_list)
999 		goto err_out_free;
1000 
1001 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1002 	if (IS_ERR(mailbox))
1003 		goto err_out_free;
1004 	eq_context = mailbox->buf;
1005 
1006 	for (i = 0; i < npages; ++i) {
1007 		eq->page_list[i].buf = dma_alloc_coherent(&dev->persist->
1008 							  pdev->dev,
1009 							  PAGE_SIZE, &t,
1010 							  GFP_KERNEL);
1011 		if (!eq->page_list[i].buf)
1012 			goto err_out_free_pages;
1013 
1014 		dma_list[i] = t;
1015 		eq->page_list[i].map = t;
1016 	}
1017 
1018 	eq->eqn = mlx4_bitmap_alloc(&priv->eq_table.bitmap);
1019 	if (eq->eqn == -1)
1020 		goto err_out_free_pages;
1021 
1022 	eq->doorbell = mlx4_get_eq_uar(dev, eq);
1023 	if (!eq->doorbell) {
1024 		err = -ENOMEM;
1025 		goto err_out_free_eq;
1026 	}
1027 
1028 	err = mlx4_mtt_init(dev, npages, PAGE_SHIFT, &eq->mtt);
1029 	if (err)
1030 		goto err_out_free_eq;
1031 
1032 	err = mlx4_write_mtt(dev, &eq->mtt, 0, npages, dma_list);
1033 	if (err)
1034 		goto err_out_free_mtt;
1035 
1036 	eq_context->flags	  = cpu_to_be32(MLX4_EQ_STATUS_OK   |
1037 						MLX4_EQ_STATE_ARMED);
1038 	eq_context->log_eq_size	  = ilog2(eq->nent);
1039 	eq_context->intr	  = intr;
1040 	eq_context->log_page_size = PAGE_SHIFT - MLX4_ICM_PAGE_SHIFT;
1041 
1042 	mtt_addr = mlx4_mtt_addr(dev, &eq->mtt);
1043 	eq_context->mtt_base_addr_h = mtt_addr >> 32;
1044 	eq_context->mtt_base_addr_l = cpu_to_be32(mtt_addr & 0xffffffff);
1045 
1046 	err = mlx4_SW2HW_EQ(dev, mailbox, eq->eqn);
1047 	if (err) {
1048 		mlx4_warn(dev, "SW2HW_EQ failed (%d)\n", err);
1049 		goto err_out_free_mtt;
1050 	}
1051 
1052 	kfree(dma_list);
1053 	mlx4_free_cmd_mailbox(dev, mailbox);
1054 
1055 	eq->cons_index = 0;
1056 
1057 	INIT_LIST_HEAD(&eq->tasklet_ctx.list);
1058 	INIT_LIST_HEAD(&eq->tasklet_ctx.process_list);
1059 	spin_lock_init(&eq->tasklet_ctx.lock);
1060 	tasklet_setup(&eq->tasklet_ctx.task, mlx4_cq_tasklet_cb);
1061 
1062 	return err;
1063 
1064 err_out_free_mtt:
1065 	mlx4_mtt_cleanup(dev, &eq->mtt);
1066 
1067 err_out_free_eq:
1068 	mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn, MLX4_USE_RR);
1069 
1070 err_out_free_pages:
1071 	for (i = 0; i < npages; ++i)
1072 		if (eq->page_list[i].buf)
1073 			dma_free_coherent(&dev->persist->pdev->dev, PAGE_SIZE,
1074 					  eq->page_list[i].buf,
1075 					  eq->page_list[i].map);
1076 
1077 	mlx4_free_cmd_mailbox(dev, mailbox);
1078 
1079 err_out_free:
1080 	kfree(eq->page_list);
1081 	kfree(dma_list);
1082 
1083 err_out:
1084 	return err;
1085 }
1086 
1087 static void mlx4_free_eq(struct mlx4_dev *dev,
1088 			 struct mlx4_eq *eq)
1089 {
1090 	struct mlx4_priv *priv = mlx4_priv(dev);
1091 	int err;
1092 	int i;
1093 	/* CX3 is capable of extending the CQE/EQE from 32 to 64 bytes, with
1094 	 * strides of 64B,128B and 256B
1095 	 */
1096 	int npages = PAGE_ALIGN(dev->caps.eqe_size  * eq->nent) / PAGE_SIZE;
1097 
1098 	err = mlx4_HW2SW_EQ(dev, eq->eqn);
1099 	if (err)
1100 		mlx4_warn(dev, "HW2SW_EQ failed (%d)\n", err);
1101 
1102 	synchronize_irq(eq->irq);
1103 	tasklet_disable(&eq->tasklet_ctx.task);
1104 
1105 	mlx4_mtt_cleanup(dev, &eq->mtt);
1106 	for (i = 0; i < npages; ++i)
1107 		dma_free_coherent(&dev->persist->pdev->dev, PAGE_SIZE,
1108 				  eq->page_list[i].buf,
1109 				  eq->page_list[i].map);
1110 
1111 	kfree(eq->page_list);
1112 	mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn, MLX4_USE_RR);
1113 }
1114 
1115 static void mlx4_free_irqs(struct mlx4_dev *dev)
1116 {
1117 	struct mlx4_eq_table *eq_table = &mlx4_priv(dev)->eq_table;
1118 	int	i;
1119 
1120 	if (eq_table->have_irq)
1121 		free_irq(dev->persist->pdev->irq, dev);
1122 
1123 	for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
1124 		if (eq_table->eq[i].have_irq) {
1125 			free_cpumask_var(eq_table->eq[i].affinity_mask);
1126 			irq_update_affinity_hint(eq_table->eq[i].irq, NULL);
1127 			free_irq(eq_table->eq[i].irq, eq_table->eq + i);
1128 			eq_table->eq[i].have_irq = 0;
1129 		}
1130 
1131 	kfree(eq_table->irq_names);
1132 }
1133 
1134 static int mlx4_map_clr_int(struct mlx4_dev *dev)
1135 {
1136 	struct mlx4_priv *priv = mlx4_priv(dev);
1137 
1138 	priv->clr_base = ioremap(pci_resource_start(dev->persist->pdev,
1139 				 priv->fw.clr_int_bar) +
1140 				 priv->fw.clr_int_base, MLX4_CLR_INT_SIZE);
1141 	if (!priv->clr_base) {
1142 		mlx4_err(dev, "Couldn't map interrupt clear register, aborting\n");
1143 		return -ENOMEM;
1144 	}
1145 
1146 	return 0;
1147 }
1148 
1149 static void mlx4_unmap_clr_int(struct mlx4_dev *dev)
1150 {
1151 	struct mlx4_priv *priv = mlx4_priv(dev);
1152 
1153 	iounmap(priv->clr_base);
1154 }
1155 
1156 int mlx4_alloc_eq_table(struct mlx4_dev *dev)
1157 {
1158 	struct mlx4_priv *priv = mlx4_priv(dev);
1159 
1160 	priv->eq_table.eq = kcalloc(dev->caps.num_eqs - dev->caps.reserved_eqs,
1161 				    sizeof(*priv->eq_table.eq), GFP_KERNEL);
1162 	if (!priv->eq_table.eq)
1163 		return -ENOMEM;
1164 
1165 	return 0;
1166 }
1167 
1168 void mlx4_free_eq_table(struct mlx4_dev *dev)
1169 {
1170 	kfree(mlx4_priv(dev)->eq_table.eq);
1171 }
1172 
1173 int mlx4_init_eq_table(struct mlx4_dev *dev)
1174 {
1175 	struct mlx4_priv *priv = mlx4_priv(dev);
1176 	int err;
1177 	int i;
1178 
1179 	priv->eq_table.uar_map = kcalloc(mlx4_num_eq_uar(dev),
1180 					 sizeof(*priv->eq_table.uar_map),
1181 					 GFP_KERNEL);
1182 	if (!priv->eq_table.uar_map) {
1183 		err = -ENOMEM;
1184 		goto err_out_free;
1185 	}
1186 
1187 	err = mlx4_bitmap_init(&priv->eq_table.bitmap,
1188 			       roundup_pow_of_two(dev->caps.num_eqs),
1189 			       dev->caps.num_eqs - 1,
1190 			       dev->caps.reserved_eqs,
1191 			       roundup_pow_of_two(dev->caps.num_eqs) -
1192 			       dev->caps.num_eqs);
1193 	if (err)
1194 		goto err_out_free;
1195 
1196 	for (i = 0; i < mlx4_num_eq_uar(dev); ++i)
1197 		priv->eq_table.uar_map[i] = NULL;
1198 
1199 	if (!mlx4_is_slave(dev)) {
1200 		err = mlx4_map_clr_int(dev);
1201 		if (err)
1202 			goto err_out_bitmap;
1203 
1204 		priv->eq_table.clr_mask =
1205 			swab32(1 << (priv->eq_table.inta_pin & 31));
1206 		priv->eq_table.clr_int  = priv->clr_base +
1207 			(priv->eq_table.inta_pin < 32 ? 4 : 0);
1208 	}
1209 
1210 	priv->eq_table.irq_names =
1211 		kmalloc_array(MLX4_IRQNAME_SIZE,
1212 			      (dev->caps.num_comp_vectors + 1),
1213 			      GFP_KERNEL);
1214 	if (!priv->eq_table.irq_names) {
1215 		err = -ENOMEM;
1216 		goto err_out_clr_int;
1217 	}
1218 
1219 	for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i) {
1220 		if (i == MLX4_EQ_ASYNC) {
1221 			err = mlx4_create_eq(dev,
1222 					     MLX4_NUM_ASYNC_EQE + MLX4_NUM_SPARE_EQE,
1223 					     0, &priv->eq_table.eq[MLX4_EQ_ASYNC]);
1224 		} else {
1225 			struct mlx4_eq	*eq = &priv->eq_table.eq[i];
1226 #ifdef CONFIG_RFS_ACCEL
1227 			int port = find_first_bit(eq->actv_ports.ports,
1228 						  dev->caps.num_ports) + 1;
1229 
1230 			if (port <= dev->caps.num_ports) {
1231 				struct mlx4_port_info *info =
1232 					&mlx4_priv(dev)->port[port];
1233 
1234 				if (!info->rmap) {
1235 					info->rmap = alloc_irq_cpu_rmap(
1236 						mlx4_get_eqs_per_port(dev, port));
1237 					if (!info->rmap) {
1238 						mlx4_warn(dev, "Failed to allocate cpu rmap\n");
1239 						err = -ENOMEM;
1240 						goto err_out_unmap;
1241 					}
1242 				}
1243 
1244 				err = irq_cpu_rmap_add(
1245 					info->rmap, eq->irq);
1246 				if (err)
1247 					mlx4_warn(dev, "Failed adding irq rmap\n");
1248 			}
1249 #endif
1250 			err = mlx4_create_eq(dev, dev->quotas.cq +
1251 					     MLX4_NUM_SPARE_EQE,
1252 					     (dev->flags & MLX4_FLAG_MSI_X) ?
1253 					     i + 1 - !!(i > MLX4_EQ_ASYNC) : 0,
1254 					     eq);
1255 		}
1256 		if (err)
1257 			goto err_out_unmap;
1258 	}
1259 
1260 	if (dev->flags & MLX4_FLAG_MSI_X) {
1261 		const char *eq_name;
1262 
1263 		snprintf(priv->eq_table.irq_names +
1264 			 MLX4_EQ_ASYNC * MLX4_IRQNAME_SIZE,
1265 			 MLX4_IRQNAME_SIZE,
1266 			 "mlx4-async@pci:%s",
1267 			 pci_name(dev->persist->pdev));
1268 		eq_name = priv->eq_table.irq_names +
1269 			MLX4_EQ_ASYNC * MLX4_IRQNAME_SIZE;
1270 
1271 		err = request_irq(priv->eq_table.eq[MLX4_EQ_ASYNC].irq,
1272 				  mlx4_msi_x_interrupt, 0, eq_name,
1273 				  priv->eq_table.eq + MLX4_EQ_ASYNC);
1274 		if (err)
1275 			goto err_out_unmap;
1276 
1277 		priv->eq_table.eq[MLX4_EQ_ASYNC].have_irq = 1;
1278 	} else {
1279 		snprintf(priv->eq_table.irq_names,
1280 			 MLX4_IRQNAME_SIZE,
1281 			 DRV_NAME "@pci:%s",
1282 			 pci_name(dev->persist->pdev));
1283 		err = request_irq(dev->persist->pdev->irq, mlx4_interrupt,
1284 				  IRQF_SHARED, priv->eq_table.irq_names, dev);
1285 		if (err)
1286 			goto err_out_unmap;
1287 
1288 		priv->eq_table.have_irq = 1;
1289 	}
1290 
1291 	err = mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1292 			  priv->eq_table.eq[MLX4_EQ_ASYNC].eqn);
1293 	if (err)
1294 		mlx4_warn(dev, "MAP_EQ for async EQ %d failed (%d)\n",
1295 			   priv->eq_table.eq[MLX4_EQ_ASYNC].eqn, err);
1296 
1297 	/* arm ASYNC eq */
1298 	eq_set_ci(&priv->eq_table.eq[MLX4_EQ_ASYNC], 1);
1299 
1300 	return 0;
1301 
1302 err_out_unmap:
1303 	while (i > 0)
1304 		mlx4_free_eq(dev, &priv->eq_table.eq[--i]);
1305 #ifdef CONFIG_RFS_ACCEL
1306 	for (i = 1; i <= dev->caps.num_ports; i++) {
1307 		if (mlx4_priv(dev)->port[i].rmap) {
1308 			free_irq_cpu_rmap(mlx4_priv(dev)->port[i].rmap);
1309 			mlx4_priv(dev)->port[i].rmap = NULL;
1310 		}
1311 	}
1312 #endif
1313 	mlx4_free_irqs(dev);
1314 
1315 err_out_clr_int:
1316 	if (!mlx4_is_slave(dev))
1317 		mlx4_unmap_clr_int(dev);
1318 
1319 err_out_bitmap:
1320 	mlx4_unmap_uar(dev);
1321 	mlx4_bitmap_cleanup(&priv->eq_table.bitmap);
1322 
1323 err_out_free:
1324 	kfree(priv->eq_table.uar_map);
1325 
1326 	return err;
1327 }
1328 
1329 void mlx4_cleanup_eq_table(struct mlx4_dev *dev)
1330 {
1331 	struct mlx4_priv *priv = mlx4_priv(dev);
1332 	int i;
1333 
1334 	mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 1,
1335 		    priv->eq_table.eq[MLX4_EQ_ASYNC].eqn);
1336 
1337 #ifdef CONFIG_RFS_ACCEL
1338 	for (i = 1; i <= dev->caps.num_ports; i++) {
1339 		if (mlx4_priv(dev)->port[i].rmap) {
1340 			free_irq_cpu_rmap(mlx4_priv(dev)->port[i].rmap);
1341 			mlx4_priv(dev)->port[i].rmap = NULL;
1342 		}
1343 	}
1344 #endif
1345 	mlx4_free_irqs(dev);
1346 
1347 	for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
1348 		mlx4_free_eq(dev, &priv->eq_table.eq[i]);
1349 
1350 	if (!mlx4_is_slave(dev))
1351 		mlx4_unmap_clr_int(dev);
1352 
1353 	mlx4_unmap_uar(dev);
1354 	mlx4_bitmap_cleanup(&priv->eq_table.bitmap);
1355 
1356 	kfree(priv->eq_table.uar_map);
1357 }
1358 
1359 /* A test that verifies that we can accept interrupts
1360  * on the vector allocated for asynchronous events
1361  */
1362 int mlx4_test_async(struct mlx4_dev *dev)
1363 {
1364 	return mlx4_NOP(dev);
1365 }
1366 EXPORT_SYMBOL(mlx4_test_async);
1367 
1368 /* A test that verifies that we can accept interrupts
1369  * on the given irq vector of the tested port.
1370  * Interrupts are checked using the NOP command.
1371  */
1372 int mlx4_test_interrupt(struct mlx4_dev *dev, int vector)
1373 {
1374 	struct mlx4_priv *priv = mlx4_priv(dev);
1375 	int err;
1376 
1377 	/* Temporary use polling for command completions */
1378 	mlx4_cmd_use_polling(dev);
1379 
1380 	/* Map the new eq to handle all asynchronous events */
1381 	err = mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1382 			  priv->eq_table.eq[MLX4_CQ_TO_EQ_VECTOR(vector)].eqn);
1383 	if (err) {
1384 		mlx4_warn(dev, "Failed mapping eq for interrupt test\n");
1385 		goto out;
1386 	}
1387 
1388 	/* Go back to using events */
1389 	mlx4_cmd_use_events(dev);
1390 	err = mlx4_NOP(dev);
1391 
1392 	/* Return to default */
1393 	mlx4_cmd_use_polling(dev);
1394 out:
1395 	mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1396 		    priv->eq_table.eq[MLX4_EQ_ASYNC].eqn);
1397 	mlx4_cmd_use_events(dev);
1398 
1399 	return err;
1400 }
1401 EXPORT_SYMBOL(mlx4_test_interrupt);
1402 
1403 bool mlx4_is_eq_vector_valid(struct mlx4_dev *dev, u8 port, int vector)
1404 {
1405 	struct mlx4_priv *priv = mlx4_priv(dev);
1406 
1407 	vector = MLX4_CQ_TO_EQ_VECTOR(vector);
1408 	if (vector < 0 || (vector >= dev->caps.num_comp_vectors + 1) ||
1409 	    (vector == MLX4_EQ_ASYNC))
1410 		return false;
1411 
1412 	return test_bit(port - 1, priv->eq_table.eq[vector].actv_ports.ports);
1413 }
1414 EXPORT_SYMBOL(mlx4_is_eq_vector_valid);
1415 
1416 u32 mlx4_get_eqs_per_port(struct mlx4_dev *dev, u8 port)
1417 {
1418 	struct mlx4_priv *priv = mlx4_priv(dev);
1419 	unsigned int i;
1420 	unsigned int sum = 0;
1421 
1422 	for (i = 0; i < dev->caps.num_comp_vectors + 1; i++)
1423 		sum += !!test_bit(port - 1,
1424 				  priv->eq_table.eq[i].actv_ports.ports);
1425 
1426 	return sum;
1427 }
1428 EXPORT_SYMBOL(mlx4_get_eqs_per_port);
1429 
1430 int mlx4_is_eq_shared(struct mlx4_dev *dev, int vector)
1431 {
1432 	struct mlx4_priv *priv = mlx4_priv(dev);
1433 
1434 	vector = MLX4_CQ_TO_EQ_VECTOR(vector);
1435 	if (vector <= 0 || (vector >= dev->caps.num_comp_vectors + 1))
1436 		return -EINVAL;
1437 
1438 	return !!(bitmap_weight(priv->eq_table.eq[vector].actv_ports.ports,
1439 				dev->caps.num_ports) > 1);
1440 }
1441 EXPORT_SYMBOL(mlx4_is_eq_shared);
1442 
1443 struct cpu_rmap *mlx4_get_cpu_rmap(struct mlx4_dev *dev, int port)
1444 {
1445 	return mlx4_priv(dev)->port[port].rmap;
1446 }
1447 EXPORT_SYMBOL(mlx4_get_cpu_rmap);
1448 
1449 int mlx4_assign_eq(struct mlx4_dev *dev, u8 port, int *vector)
1450 {
1451 	struct mlx4_priv *priv = mlx4_priv(dev);
1452 	int err = 0, i = 0;
1453 	u32 min_ref_count_val = (u32)-1;
1454 	int requested_vector = MLX4_CQ_TO_EQ_VECTOR(*vector);
1455 	int *prequested_vector = NULL;
1456 
1457 
1458 	mutex_lock(&priv->msix_ctl.pool_lock);
1459 	if (requested_vector < (dev->caps.num_comp_vectors + 1) &&
1460 	    (requested_vector >= 0) &&
1461 	    (requested_vector != MLX4_EQ_ASYNC)) {
1462 		if (test_bit(port - 1,
1463 			     priv->eq_table.eq[requested_vector].actv_ports.ports)) {
1464 			prequested_vector = &requested_vector;
1465 		} else {
1466 			struct mlx4_eq *eq;
1467 
1468 			for (i = 1; i < port;
1469 			     requested_vector += mlx4_get_eqs_per_port(dev, i++))
1470 				;
1471 
1472 			eq = &priv->eq_table.eq[requested_vector];
1473 			if (requested_vector < dev->caps.num_comp_vectors + 1 &&
1474 			    test_bit(port - 1, eq->actv_ports.ports)) {
1475 				prequested_vector = &requested_vector;
1476 			}
1477 		}
1478 	}
1479 
1480 	if  (!prequested_vector) {
1481 		requested_vector = -1;
1482 		for (i = 0; min_ref_count_val && i < dev->caps.num_comp_vectors + 1;
1483 		     i++) {
1484 			struct mlx4_eq *eq = &priv->eq_table.eq[i];
1485 
1486 			if (min_ref_count_val > eq->ref_count &&
1487 			    test_bit(port - 1, eq->actv_ports.ports)) {
1488 				min_ref_count_val = eq->ref_count;
1489 				requested_vector = i;
1490 			}
1491 		}
1492 
1493 		if (requested_vector < 0) {
1494 			err = -ENOSPC;
1495 			goto err_unlock;
1496 		}
1497 
1498 		prequested_vector = &requested_vector;
1499 	}
1500 
1501 	if (!test_bit(*prequested_vector, priv->msix_ctl.pool_bm) &&
1502 	    dev->flags & MLX4_FLAG_MSI_X) {
1503 		set_bit(*prequested_vector, priv->msix_ctl.pool_bm);
1504 		snprintf(priv->eq_table.irq_names +
1505 			 *prequested_vector * MLX4_IRQNAME_SIZE,
1506 			 MLX4_IRQNAME_SIZE, "mlx4-%d@%s",
1507 			 *prequested_vector, dev_name(&dev->persist->pdev->dev));
1508 
1509 		err = request_irq(priv->eq_table.eq[*prequested_vector].irq,
1510 				  mlx4_msi_x_interrupt, 0,
1511 				  &priv->eq_table.irq_names[*prequested_vector << 5],
1512 				  priv->eq_table.eq + *prequested_vector);
1513 
1514 		if (err) {
1515 			clear_bit(*prequested_vector, priv->msix_ctl.pool_bm);
1516 			*prequested_vector = -1;
1517 		} else {
1518 #if defined(CONFIG_SMP)
1519 			mlx4_set_eq_affinity_hint(priv, *prequested_vector);
1520 #endif
1521 			eq_set_ci(&priv->eq_table.eq[*prequested_vector], 1);
1522 			priv->eq_table.eq[*prequested_vector].have_irq = 1;
1523 		}
1524 	}
1525 
1526 	if (!err && *prequested_vector >= 0)
1527 		priv->eq_table.eq[*prequested_vector].ref_count++;
1528 
1529 err_unlock:
1530 	mutex_unlock(&priv->msix_ctl.pool_lock);
1531 
1532 	if (!err && *prequested_vector >= 0)
1533 		*vector = MLX4_EQ_TO_CQ_VECTOR(*prequested_vector);
1534 	else
1535 		*vector = 0;
1536 
1537 	return err;
1538 }
1539 EXPORT_SYMBOL(mlx4_assign_eq);
1540 
1541 int mlx4_eq_get_irq(struct mlx4_dev *dev, int cq_vec)
1542 {
1543 	struct mlx4_priv *priv = mlx4_priv(dev);
1544 
1545 	return priv->eq_table.eq[MLX4_CQ_TO_EQ_VECTOR(cq_vec)].irq;
1546 }
1547 EXPORT_SYMBOL(mlx4_eq_get_irq);
1548 
1549 void mlx4_release_eq(struct mlx4_dev *dev, int vec)
1550 {
1551 	struct mlx4_priv *priv = mlx4_priv(dev);
1552 	int eq_vec = MLX4_CQ_TO_EQ_VECTOR(vec);
1553 
1554 	mutex_lock(&priv->msix_ctl.pool_lock);
1555 	priv->eq_table.eq[eq_vec].ref_count--;
1556 
1557 	/* once we allocated EQ, we don't release it because it might be binded
1558 	 * to cpu_rmap.
1559 	 */
1560 	mutex_unlock(&priv->msix_ctl.pool_lock);
1561 }
1562 EXPORT_SYMBOL(mlx4_release_eq);
1563 
1564