xref: /freebsd-src/sys/dev/mlx5/mlx5_core/mlx5_eq.c (revision 2eb4d8dc723da3cf7d735a3226ae49da4c8c5dbc)
1 /*-
2  * Copyright (c) 2013-2017, Mellanox Technologies, Ltd.  All rights reserved.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  * 1. Redistributions of source code must retain the above copyright
8  *    notice, this list of conditions and the following disclaimer.
9  * 2. Redistributions in binary form must reproduce the above copyright
10  *    notice, this list of conditions and the following disclaimer in the
11  *    documentation and/or other materials provided with the distribution.
12  *
13  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND
14  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16  * ARE DISCLAIMED.  IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
17  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
23  * SUCH DAMAGE.
24  *
25  * $FreeBSD$
26  */
27 
28 #include <linux/interrupt.h>
29 #include <linux/module.h>
30 #include <dev/mlx5/port.h>
31 #include <dev/mlx5/mlx5_ifc.h>
32 #include <dev/mlx5/mlx5_fpga/core.h>
33 #include "mlx5_core.h"
34 #include "eswitch.h"
35 
36 #include "opt_rss.h"
37 
38 #ifdef  RSS
39 #include <net/rss_config.h>
40 #include <netinet/in_rss.h>
41 #endif
42 
43 enum {
44 	MLX5_EQE_SIZE		= sizeof(struct mlx5_eqe),
45 	MLX5_EQE_OWNER_INIT_VAL	= 0x1,
46 };
47 
48 enum {
49 	MLX5_NUM_SPARE_EQE	= 0x80,
50 	MLX5_NUM_ASYNC_EQE	= 0x100,
51 	MLX5_NUM_CMD_EQE	= 32,
52 };
53 
54 enum {
55 	MLX5_EQ_DOORBEL_OFFSET	= 0x40,
56 };
57 
58 #define MLX5_ASYNC_EVENT_MASK ((1ull << MLX5_EVENT_TYPE_PATH_MIG)	    | \
59 			       (1ull << MLX5_EVENT_TYPE_COMM_EST)	    | \
60 			       (1ull << MLX5_EVENT_TYPE_SQ_DRAINED)	    | \
61 			       (1ull << MLX5_EVENT_TYPE_CQ_ERROR)	    | \
62 			       (1ull << MLX5_EVENT_TYPE_WQ_CATAS_ERROR)	    | \
63 			       (1ull << MLX5_EVENT_TYPE_PATH_MIG_FAILED)    | \
64 			       (1ull << MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR) | \
65 			       (1ull << MLX5_EVENT_TYPE_WQ_ACCESS_ERROR)    | \
66 			       (1ull << MLX5_EVENT_TYPE_PORT_CHANGE)	    | \
67 			       (1ull << MLX5_EVENT_TYPE_SRQ_CATAS_ERROR)    | \
68 			       (1ull << MLX5_EVENT_TYPE_SRQ_LAST_WQE)	    | \
69 			       (1ull << MLX5_EVENT_TYPE_SRQ_RQ_LIMIT)	    | \
70 			       (1ull << MLX5_EVENT_TYPE_NIC_VPORT_CHANGE))
71 
72 struct map_eq_in {
73 	u64	mask;
74 	u32	reserved;
75 	u32	unmap_eqn;
76 };
77 
78 struct cre_des_eq {
79 	u8	reserved[15];
80 	u8	eqn;
81 };
82 
83 /*Function prototype*/
84 static void mlx5_port_module_event(struct mlx5_core_dev *dev,
85 				   struct mlx5_eqe *eqe);
86 static void mlx5_port_general_notification_event(struct mlx5_core_dev *dev,
87 						 struct mlx5_eqe *eqe);
88 
89 static int mlx5_cmd_destroy_eq(struct mlx5_core_dev *dev, u8 eqn)
90 {
91 	u32 in[MLX5_ST_SZ_DW(destroy_eq_in)] = {0};
92 	u32 out[MLX5_ST_SZ_DW(destroy_eq_out)] = {0};
93 
94 	MLX5_SET(destroy_eq_in, in, opcode, MLX5_CMD_OP_DESTROY_EQ);
95 	MLX5_SET(destroy_eq_in, in, eq_number, eqn);
96 
97 	return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
98 }
99 
100 static struct mlx5_eqe *get_eqe(struct mlx5_eq *eq, u32 entry)
101 {
102 	return mlx5_buf_offset(&eq->buf, entry * MLX5_EQE_SIZE);
103 }
104 
105 static struct mlx5_eqe *next_eqe_sw(struct mlx5_eq *eq)
106 {
107 	struct mlx5_eqe *eqe = get_eqe(eq, eq->cons_index & (eq->nent - 1));
108 
109 	return ((eqe->owner & 1) ^ !!(eq->cons_index & eq->nent)) ? NULL : eqe;
110 }
111 
112 static const char *eqe_type_str(u8 type)
113 {
114 	switch (type) {
115 	case MLX5_EVENT_TYPE_COMP:
116 		return "MLX5_EVENT_TYPE_COMP";
117 	case MLX5_EVENT_TYPE_PATH_MIG:
118 		return "MLX5_EVENT_TYPE_PATH_MIG";
119 	case MLX5_EVENT_TYPE_COMM_EST:
120 		return "MLX5_EVENT_TYPE_COMM_EST";
121 	case MLX5_EVENT_TYPE_SQ_DRAINED:
122 		return "MLX5_EVENT_TYPE_SQ_DRAINED";
123 	case MLX5_EVENT_TYPE_SRQ_LAST_WQE:
124 		return "MLX5_EVENT_TYPE_SRQ_LAST_WQE";
125 	case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
126 		return "MLX5_EVENT_TYPE_SRQ_RQ_LIMIT";
127 	case MLX5_EVENT_TYPE_CQ_ERROR:
128 		return "MLX5_EVENT_TYPE_CQ_ERROR";
129 	case MLX5_EVENT_TYPE_WQ_CATAS_ERROR:
130 		return "MLX5_EVENT_TYPE_WQ_CATAS_ERROR";
131 	case MLX5_EVENT_TYPE_PATH_MIG_FAILED:
132 		return "MLX5_EVENT_TYPE_PATH_MIG_FAILED";
133 	case MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
134 		return "MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR";
135 	case MLX5_EVENT_TYPE_WQ_ACCESS_ERROR:
136 		return "MLX5_EVENT_TYPE_WQ_ACCESS_ERROR";
137 	case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
138 		return "MLX5_EVENT_TYPE_SRQ_CATAS_ERROR";
139 	case MLX5_EVENT_TYPE_INTERNAL_ERROR:
140 		return "MLX5_EVENT_TYPE_INTERNAL_ERROR";
141 	case MLX5_EVENT_TYPE_PORT_CHANGE:
142 		return "MLX5_EVENT_TYPE_PORT_CHANGE";
143 	case MLX5_EVENT_TYPE_GPIO_EVENT:
144 		return "MLX5_EVENT_TYPE_GPIO_EVENT";
145 	case MLX5_EVENT_TYPE_CODING_PORT_MODULE_EVENT:
146 		return "MLX5_EVENT_TYPE_PORT_MODULE_EVENT";
147 	case MLX5_EVENT_TYPE_TEMP_WARN_EVENT:
148 		return "MLX5_EVENT_TYPE_TEMP_WARN_EVENT";
149 	case MLX5_EVENT_TYPE_REMOTE_CONFIG:
150 		return "MLX5_EVENT_TYPE_REMOTE_CONFIG";
151 	case MLX5_EVENT_TYPE_DB_BF_CONGESTION:
152 		return "MLX5_EVENT_TYPE_DB_BF_CONGESTION";
153 	case MLX5_EVENT_TYPE_STALL_EVENT:
154 		return "MLX5_EVENT_TYPE_STALL_EVENT";
155 	case MLX5_EVENT_TYPE_CMD:
156 		return "MLX5_EVENT_TYPE_CMD";
157 	case MLX5_EVENT_TYPE_PAGE_REQUEST:
158 		return "MLX5_EVENT_TYPE_PAGE_REQUEST";
159 	case MLX5_EVENT_TYPE_NIC_VPORT_CHANGE:
160 		return "MLX5_EVENT_TYPE_NIC_VPORT_CHANGE";
161 	case MLX5_EVENT_TYPE_FPGA_ERROR:
162 		return "MLX5_EVENT_TYPE_FPGA_ERROR";
163 	case MLX5_EVENT_TYPE_FPGA_QP_ERROR:
164 		return "MLX5_EVENT_TYPE_FPGA_QP_ERROR";
165 	case MLX5_EVENT_TYPE_CODING_DCBX_CHANGE_EVENT:
166 		return "MLX5_EVENT_TYPE_CODING_DCBX_CHANGE_EVENT";
167 	case MLX5_EVENT_TYPE_CODING_GENERAL_NOTIFICATION_EVENT:
168 		return "MLX5_EVENT_TYPE_CODING_GENERAL_NOTIFICATION_EVENT";
169 	default:
170 		return "Unrecognized event";
171 	}
172 }
173 
174 static enum mlx5_dev_event port_subtype_event(u8 subtype)
175 {
176 	switch (subtype) {
177 	case MLX5_PORT_CHANGE_SUBTYPE_DOWN:
178 		return MLX5_DEV_EVENT_PORT_DOWN;
179 	case MLX5_PORT_CHANGE_SUBTYPE_ACTIVE:
180 		return MLX5_DEV_EVENT_PORT_UP;
181 	case MLX5_PORT_CHANGE_SUBTYPE_INITIALIZED:
182 		return MLX5_DEV_EVENT_PORT_INITIALIZED;
183 	case MLX5_PORT_CHANGE_SUBTYPE_LID:
184 		return MLX5_DEV_EVENT_LID_CHANGE;
185 	case MLX5_PORT_CHANGE_SUBTYPE_PKEY:
186 		return MLX5_DEV_EVENT_PKEY_CHANGE;
187 	case MLX5_PORT_CHANGE_SUBTYPE_GUID:
188 		return MLX5_DEV_EVENT_GUID_CHANGE;
189 	case MLX5_PORT_CHANGE_SUBTYPE_CLIENT_REREG:
190 		return MLX5_DEV_EVENT_CLIENT_REREG;
191 	}
192 	return -1;
193 }
194 
195 static enum mlx5_dev_event dcbx_subevent(u8 subtype)
196 {
197 	switch (subtype) {
198 	case MLX5_DCBX_EVENT_SUBTYPE_ERROR_STATE_DCBX:
199 		return MLX5_DEV_EVENT_ERROR_STATE_DCBX;
200 	case MLX5_DCBX_EVENT_SUBTYPE_REMOTE_CONFIG_CHANGE:
201 		return MLX5_DEV_EVENT_REMOTE_CONFIG_CHANGE;
202 	case MLX5_DCBX_EVENT_SUBTYPE_LOCAL_OPER_CHANGE:
203 		return MLX5_DEV_EVENT_LOCAL_OPER_CHANGE;
204 	case MLX5_DCBX_EVENT_SUBTYPE_REMOTE_CONFIG_APP_PRIORITY_CHANGE:
205 		return MLX5_DEV_EVENT_REMOTE_CONFIG_APPLICATION_PRIORITY_CHANGE;
206 	}
207 	return -1;
208 }
209 
210 static void eq_update_ci(struct mlx5_eq *eq, int arm)
211 {
212 	__be32 __iomem *addr = eq->doorbell + (arm ? 0 : 2);
213 	u32 val = (eq->cons_index & 0xffffff) | (eq->eqn << 24);
214 	__raw_writel((__force u32) cpu_to_be32(val), addr);
215 	/* We still want ordering, just not swabbing, so add a barrier */
216 	mb();
217 }
218 
219 static void
220 mlx5_temp_warning_event(struct mlx5_core_dev *dev, struct mlx5_eqe *eqe)
221 {
222 
223 	mlx5_core_warn(dev,
224 	    "High temperature on sensors with bit set %#jx %#jx\n",
225 	    (uintmax_t)be64_to_cpu(eqe->data.temp_warning.sensor_warning_msb),
226 	    (uintmax_t)be64_to_cpu(eqe->data.temp_warning.sensor_warning_lsb));
227 }
228 
229 static int mlx5_eq_int(struct mlx5_core_dev *dev, struct mlx5_eq *eq)
230 {
231 	struct mlx5_eqe *eqe;
232 	int eqes_found = 0;
233 	int set_ci = 0;
234 	u32 cqn;
235 	u32 rsn;
236 	u8 port;
237 
238 	while ((eqe = next_eqe_sw(eq))) {
239 		/*
240 		 * Make sure we read EQ entry contents after we've
241 		 * checked the ownership bit.
242 		 */
243 		atomic_thread_fence_acq();
244 
245 		mlx5_core_dbg(eq->dev, "eqn %d, eqe type %s\n",
246 			      eq->eqn, eqe_type_str(eqe->type));
247 		switch (eqe->type) {
248 		case MLX5_EVENT_TYPE_COMP:
249 			mlx5_cq_completion(dev, eqe);
250 			break;
251 
252 		case MLX5_EVENT_TYPE_PATH_MIG:
253 		case MLX5_EVENT_TYPE_COMM_EST:
254 		case MLX5_EVENT_TYPE_SQ_DRAINED:
255 		case MLX5_EVENT_TYPE_SRQ_LAST_WQE:
256 		case MLX5_EVENT_TYPE_WQ_CATAS_ERROR:
257 		case MLX5_EVENT_TYPE_PATH_MIG_FAILED:
258 		case MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
259 		case MLX5_EVENT_TYPE_WQ_ACCESS_ERROR:
260 			rsn = be32_to_cpu(eqe->data.qp_srq.qp_srq_n) & 0xffffff;
261 			mlx5_core_dbg(dev, "event %s(%d) arrived on resource 0x%x\n",
262 				      eqe_type_str(eqe->type), eqe->type, rsn);
263 			mlx5_rsc_event(dev, rsn, eqe->type);
264 			break;
265 
266 		case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
267 		case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
268 			rsn = be32_to_cpu(eqe->data.qp_srq.qp_srq_n) & 0xffffff;
269 			mlx5_core_dbg(dev, "SRQ event %s(%d): srqn 0x%x\n",
270 				      eqe_type_str(eqe->type), eqe->type, rsn);
271 			mlx5_srq_event(dev, rsn, eqe->type);
272 			break;
273 
274 		case MLX5_EVENT_TYPE_CMD:
275 			if (dev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
276 				mlx5_cmd_comp_handler(dev, be32_to_cpu(eqe->data.cmd.vector),
277 				    MLX5_CMD_MODE_EVENTS);
278 			}
279 			break;
280 
281 		case MLX5_EVENT_TYPE_PORT_CHANGE:
282 			port = (eqe->data.port.port >> 4) & 0xf;
283 			switch (eqe->sub_type) {
284 			case MLX5_PORT_CHANGE_SUBTYPE_DOWN:
285 			case MLX5_PORT_CHANGE_SUBTYPE_ACTIVE:
286 			case MLX5_PORT_CHANGE_SUBTYPE_LID:
287 			case MLX5_PORT_CHANGE_SUBTYPE_PKEY:
288 			case MLX5_PORT_CHANGE_SUBTYPE_GUID:
289 			case MLX5_PORT_CHANGE_SUBTYPE_CLIENT_REREG:
290 			case MLX5_PORT_CHANGE_SUBTYPE_INITIALIZED:
291 				if (dev->event)
292 					dev->event(dev, port_subtype_event(eqe->sub_type),
293 						   (unsigned long)port);
294 				break;
295 			default:
296 				mlx5_core_warn(dev, "Port event with unrecognized subtype: port %d, sub_type %d\n",
297 					       port, eqe->sub_type);
298 			}
299 			break;
300 
301 		case MLX5_EVENT_TYPE_CODING_DCBX_CHANGE_EVENT:
302 			port = (eqe->data.port.port >> 4) & 0xf;
303 			switch (eqe->sub_type) {
304 			case MLX5_DCBX_EVENT_SUBTYPE_ERROR_STATE_DCBX:
305 			case MLX5_DCBX_EVENT_SUBTYPE_REMOTE_CONFIG_CHANGE:
306 			case MLX5_DCBX_EVENT_SUBTYPE_LOCAL_OPER_CHANGE:
307 			case MLX5_DCBX_EVENT_SUBTYPE_REMOTE_CONFIG_APP_PRIORITY_CHANGE:
308 				if (dev->event)
309 					dev->event(dev,
310 						   dcbx_subevent(eqe->sub_type),
311 						   0);
312 				break;
313 			default:
314 				mlx5_core_warn(dev,
315 					       "dcbx event with unrecognized subtype: port %d, sub_type %d\n",
316 					       port, eqe->sub_type);
317 			}
318 			break;
319 
320 		case MLX5_EVENT_TYPE_CODING_GENERAL_NOTIFICATION_EVENT:
321 			mlx5_port_general_notification_event(dev, eqe);
322 			break;
323 
324 		case MLX5_EVENT_TYPE_CQ_ERROR:
325 			cqn = be32_to_cpu(eqe->data.cq_err.cqn) & 0xffffff;
326 			mlx5_core_warn(dev, "CQ error on CQN 0x%x, syndrom 0x%x\n",
327 				       cqn, eqe->data.cq_err.syndrome);
328 			mlx5_cq_event(dev, cqn, eqe->type);
329 			break;
330 
331 		case MLX5_EVENT_TYPE_PAGE_REQUEST:
332 			{
333 				u16 func_id = be16_to_cpu(eqe->data.req_pages.func_id);
334 				s32 npages = be32_to_cpu(eqe->data.req_pages.num_pages);
335 
336 				mlx5_core_dbg(dev, "page request for func 0x%x, npages %d\n",
337 					      func_id, npages);
338 				mlx5_core_req_pages_handler(dev, func_id, npages);
339 			}
340 			break;
341 
342 		case MLX5_EVENT_TYPE_CODING_PORT_MODULE_EVENT:
343 			mlx5_port_module_event(dev, eqe);
344 			break;
345 
346 		case MLX5_EVENT_TYPE_NIC_VPORT_CHANGE:
347 			{
348 				struct mlx5_eqe_vport_change *vc_eqe =
349 						&eqe->data.vport_change;
350 				u16 vport_num = be16_to_cpu(vc_eqe->vport_num);
351 
352 				if (dev->event)
353 					dev->event(dev,
354 					     MLX5_DEV_EVENT_VPORT_CHANGE,
355 					     (unsigned long)vport_num);
356 			}
357 			if (dev->priv.eswitch != NULL)
358 				mlx5_eswitch_vport_event(dev->priv.eswitch,
359 				    eqe);
360 			break;
361 
362 		case MLX5_EVENT_TYPE_FPGA_ERROR:
363 		case MLX5_EVENT_TYPE_FPGA_QP_ERROR:
364 			mlx5_fpga_event(dev, eqe->type, &eqe->data.raw);
365 			break;
366 		case MLX5_EVENT_TYPE_TEMP_WARN_EVENT:
367 			mlx5_temp_warning_event(dev, eqe);
368 			break;
369 
370 		default:
371 			mlx5_core_warn(dev, "Unhandled event 0x%x on EQ 0x%x\n",
372 				       eqe->type, eq->eqn);
373 			break;
374 		}
375 
376 		++eq->cons_index;
377 		eqes_found = 1;
378 		++set_ci;
379 
380 		/* The HCA will think the queue has overflowed if we
381 		 * don't tell it we've been processing events.  We
382 		 * create our EQs with MLX5_NUM_SPARE_EQE extra
383 		 * entries, so we must update our consumer index at
384 		 * least that often.
385 		 */
386 		if (unlikely(set_ci >= MLX5_NUM_SPARE_EQE)) {
387 			eq_update_ci(eq, 0);
388 			set_ci = 0;
389 		}
390 	}
391 
392 	eq_update_ci(eq, 1);
393 
394 	return eqes_found;
395 }
396 
397 static irqreturn_t mlx5_msix_handler(int irq, void *eq_ptr)
398 {
399 	struct mlx5_eq *eq = eq_ptr;
400 	struct mlx5_core_dev *dev = eq->dev;
401 
402 	/* check if IRQs are not disabled */
403 	if (likely(dev->priv.disable_irqs == 0))
404 		mlx5_eq_int(dev, eq);
405 
406 	/* MSI-X vectors always belong to us */
407 	return IRQ_HANDLED;
408 }
409 
410 static void init_eq_buf(struct mlx5_eq *eq)
411 {
412 	struct mlx5_eqe *eqe;
413 	int i;
414 
415 	for (i = 0; i < eq->nent; i++) {
416 		eqe = get_eqe(eq, i);
417 		eqe->owner = MLX5_EQE_OWNER_INIT_VAL;
418 	}
419 }
420 
421 int mlx5_create_map_eq(struct mlx5_core_dev *dev, struct mlx5_eq *eq, u8 vecidx,
422 		       int nent, u64 mask)
423 {
424 	u32 out[MLX5_ST_SZ_DW(create_eq_out)] = {0};
425 	struct mlx5_priv *priv = &dev->priv;
426 	__be64 *pas;
427 	void *eqc;
428 	int inlen;
429 	u32 *in;
430 	int err;
431 
432 	eq->nent = roundup_pow_of_two(nent + MLX5_NUM_SPARE_EQE);
433 	eq->cons_index = 0;
434 	err = mlx5_buf_alloc(dev, eq->nent * MLX5_EQE_SIZE, 2 * PAGE_SIZE,
435 			     &eq->buf);
436 	if (err)
437 		return err;
438 
439 	init_eq_buf(eq);
440 
441 	inlen = MLX5_ST_SZ_BYTES(create_eq_in) +
442 		MLX5_FLD_SZ_BYTES(create_eq_in, pas[0]) * eq->buf.npages;
443 	in = mlx5_vzalloc(inlen);
444 	if (!in) {
445 		err = -ENOMEM;
446 		goto err_buf;
447 	}
448 
449 	pas = (__be64 *)MLX5_ADDR_OF(create_eq_in, in, pas);
450 	mlx5_fill_page_array(&eq->buf, pas);
451 
452 	MLX5_SET(create_eq_in, in, opcode, MLX5_CMD_OP_CREATE_EQ);
453 	MLX5_SET64(create_eq_in, in, event_bitmask, mask);
454 
455 	eqc = MLX5_ADDR_OF(create_eq_in, in, eq_context_entry);
456 	MLX5_SET(eqc, eqc, log_eq_size, ilog2(eq->nent));
457 	MLX5_SET(eqc, eqc, uar_page, priv->uar->index);
458 	MLX5_SET(eqc, eqc, intr, vecidx);
459 	MLX5_SET(eqc, eqc, log_page_size,
460 		 eq->buf.page_shift - MLX5_ADAPTER_PAGE_SHIFT);
461 
462 	err = mlx5_cmd_exec(dev, in, inlen, out, sizeof(out));
463 	if (err)
464 		goto err_in;
465 
466 	eq->eqn = MLX5_GET(create_eq_out, out, eq_number);
467 	eq->irqn = vecidx;
468 	eq->dev = dev;
469 	eq->doorbell = priv->uar->map + MLX5_EQ_DOORBEL_OFFSET;
470 	err = request_irq(priv->msix_arr[vecidx].vector, mlx5_msix_handler, 0,
471 			  "mlx5_core", eq);
472 	if (err)
473 		goto err_eq;
474 #ifdef RSS
475 	if (vecidx >= MLX5_EQ_VEC_COMP_BASE) {
476 		u8 bucket = vecidx - MLX5_EQ_VEC_COMP_BASE;
477 		err = bind_irq_to_cpu(priv->msix_arr[vecidx].vector,
478 				      rss_getcpu(bucket % rss_getnumbuckets()));
479 		if (err)
480 			goto err_irq;
481 	}
482 #else
483 	if (0)
484 		goto err_irq;
485 #endif
486 
487 
488 	/* EQs are created in ARMED state
489 	 */
490 	eq_update_ci(eq, 1);
491 
492 	kvfree(in);
493 	return 0;
494 
495 err_irq:
496 	free_irq(priv->msix_arr[vecidx].vector, eq);
497 
498 err_eq:
499 	mlx5_cmd_destroy_eq(dev, eq->eqn);
500 
501 err_in:
502 	kvfree(in);
503 
504 err_buf:
505 	mlx5_buf_free(dev, &eq->buf);
506 	return err;
507 }
508 EXPORT_SYMBOL_GPL(mlx5_create_map_eq);
509 
510 int mlx5_destroy_unmap_eq(struct mlx5_core_dev *dev, struct mlx5_eq *eq)
511 {
512 	int err;
513 
514 	free_irq(dev->priv.msix_arr[eq->irqn].vector, eq);
515 	err = mlx5_cmd_destroy_eq(dev, eq->eqn);
516 	if (err)
517 		mlx5_core_warn(dev, "failed to destroy a previously created eq: eqn %d\n",
518 			       eq->eqn);
519 	mlx5_buf_free(dev, &eq->buf);
520 
521 	return err;
522 }
523 EXPORT_SYMBOL_GPL(mlx5_destroy_unmap_eq);
524 
525 int mlx5_eq_init(struct mlx5_core_dev *dev)
526 {
527 	int err;
528 
529 	spin_lock_init(&dev->priv.eq_table.lock);
530 
531 	err = 0;
532 
533 	return err;
534 }
535 
536 
537 void mlx5_eq_cleanup(struct mlx5_core_dev *dev)
538 {
539 }
540 
541 int mlx5_start_eqs(struct mlx5_core_dev *dev)
542 {
543 	struct mlx5_eq_table *table = &dev->priv.eq_table;
544 	u64 async_event_mask = MLX5_ASYNC_EVENT_MASK;
545 	int err;
546 
547 	if (MLX5_CAP_GEN(dev, port_module_event))
548 		async_event_mask |= (1ull <<
549 				     MLX5_EVENT_TYPE_CODING_PORT_MODULE_EVENT);
550 
551 	if (MLX5_CAP_GEN(dev, nic_vport_change_event))
552 		async_event_mask |= (1ull <<
553 				     MLX5_EVENT_TYPE_NIC_VPORT_CHANGE);
554 
555 	if (MLX5_CAP_GEN(dev, dcbx))
556 		async_event_mask |= (1ull <<
557 				     MLX5_EVENT_TYPE_CODING_DCBX_CHANGE_EVENT);
558 
559 	if (MLX5_CAP_GEN(dev, fpga))
560 		async_event_mask |= (1ull << MLX5_EVENT_TYPE_FPGA_ERROR) |
561 				    (1ull << MLX5_EVENT_TYPE_FPGA_QP_ERROR);
562 
563 	if (MLX5_CAP_GEN(dev, temp_warn_event))
564 		async_event_mask |= (1ull << MLX5_EVENT_TYPE_TEMP_WARN_EVENT);
565 
566 	if (MLX5_CAP_GEN(dev, general_notification_event)) {
567 		async_event_mask |= (1ull <<
568 		    MLX5_EVENT_TYPE_CODING_GENERAL_NOTIFICATION_EVENT);
569 	}
570 
571 	err = mlx5_create_map_eq(dev, &table->cmd_eq, MLX5_EQ_VEC_CMD,
572 				 MLX5_NUM_CMD_EQE, 1ull << MLX5_EVENT_TYPE_CMD);
573 	if (err) {
574 		mlx5_core_warn(dev, "failed to create cmd EQ %d\n", err);
575 		return err;
576 	}
577 
578 	mlx5_cmd_use_events(dev);
579 
580 	err = mlx5_create_map_eq(dev, &table->async_eq, MLX5_EQ_VEC_ASYNC,
581 				 MLX5_NUM_ASYNC_EQE, async_event_mask);
582 	if (err) {
583 		mlx5_core_warn(dev, "failed to create async EQ %d\n", err);
584 		goto err1;
585 	}
586 
587 	err = mlx5_create_map_eq(dev, &table->pages_eq,
588 				 MLX5_EQ_VEC_PAGES,
589 				 /* TODO: sriov max_vf + */ 1,
590 				 1 << MLX5_EVENT_TYPE_PAGE_REQUEST);
591 	if (err) {
592 		mlx5_core_warn(dev, "failed to create pages EQ %d\n", err);
593 		goto err2;
594 	}
595 
596 	return err;
597 
598 err2:
599 	mlx5_destroy_unmap_eq(dev, &table->async_eq);
600 
601 err1:
602 	mlx5_cmd_use_polling(dev);
603 	mlx5_destroy_unmap_eq(dev, &table->cmd_eq);
604 	return err;
605 }
606 
607 int mlx5_stop_eqs(struct mlx5_core_dev *dev)
608 {
609 	struct mlx5_eq_table *table = &dev->priv.eq_table;
610 	int err;
611 
612 	err = mlx5_destroy_unmap_eq(dev, &table->pages_eq);
613 	if (err)
614 		return err;
615 
616 	mlx5_destroy_unmap_eq(dev, &table->async_eq);
617 	mlx5_cmd_use_polling(dev);
618 
619 	err = mlx5_destroy_unmap_eq(dev, &table->cmd_eq);
620 	if (err)
621 		mlx5_cmd_use_events(dev);
622 
623 	return err;
624 }
625 
626 int mlx5_core_eq_query(struct mlx5_core_dev *dev, struct mlx5_eq *eq,
627 		       u32 *out, int outlen)
628 {
629 	u32 in[MLX5_ST_SZ_DW(query_eq_in)] = {0};
630 
631 	memset(out, 0, outlen);
632 	MLX5_SET(query_eq_in, in, opcode, MLX5_CMD_OP_QUERY_EQ);
633 	MLX5_SET(query_eq_in, in, eq_number, eq->eqn);
634 
635 	return mlx5_cmd_exec(dev, in, sizeof(in), out, outlen);
636 }
637 EXPORT_SYMBOL_GPL(mlx5_core_eq_query);
638 
639 static const char *mlx5_port_module_event_error_type_to_string(u8 error_type)
640 {
641 	switch (error_type) {
642 	case MLX5_MODULE_EVENT_ERROR_POWER_BUDGET_EXCEEDED:
643 		return "Power budget exceeded";
644 	case MLX5_MODULE_EVENT_ERROR_LONG_RANGE_FOR_NON_MLNX_CABLE_MODULE:
645 		return "Long Range for non MLNX cable";
646 	case MLX5_MODULE_EVENT_ERROR_BUS_STUCK:
647 		return "Bus stuck(I2C or data shorted)";
648 	case MLX5_MODULE_EVENT_ERROR_NO_EEPROM_RETRY_TIMEOUT:
649 		return "No EEPROM/retry timeout";
650 	case MLX5_MODULE_EVENT_ERROR_ENFORCE_PART_NUMBER_LIST:
651 		return "Enforce part number list";
652 	case MLX5_MODULE_EVENT_ERROR_UNSUPPORTED_CABLE:
653 		return "Unknown identifier";
654 	case MLX5_MODULE_EVENT_ERROR_HIGH_TEMPERATURE:
655 		return "High Temperature";
656 	case MLX5_MODULE_EVENT_ERROR_CABLE_IS_SHORTED:
657 		return "Bad or shorted cable/module";
658 	case MLX5_MODULE_EVENT_ERROR_PMD_TYPE_NOT_ENABLED:
659 		return "PMD type is not enabled";
660 	case MLX5_MODULE_EVENT_ERROR_LASTER_TEC_FAILURE:
661 		return "Laster_TEC_failure";
662 	case MLX5_MODULE_EVENT_ERROR_HIGH_CURRENT:
663 		return "High_current";
664 	case MLX5_MODULE_EVENT_ERROR_HIGH_VOLTAGE:
665 		return "High_voltage";
666 	case MLX5_MODULE_EVENT_ERROR_PCIE_SYS_POWER_SLOT_EXCEEDED:
667 		return "pcie_system_power_slot_Exceeded";
668 	case MLX5_MODULE_EVENT_ERROR_HIGH_POWER:
669 		return "High_power";
670 	case MLX5_MODULE_EVENT_ERROR_MODULE_STATE_MACHINE_FAULT:
671 		return "Module_state_machine_fault";
672 	default:
673 		return "Unknown error type";
674 	}
675 }
676 
677 unsigned int mlx5_query_module_status(struct mlx5_core_dev *dev, int module_num)
678 {
679 	if (module_num < 0 || module_num >= MLX5_MAX_PORTS)
680 		return 0;		/* undefined */
681 	return dev->module_status[module_num];
682 }
683 
684 static void mlx5_port_module_event(struct mlx5_core_dev *dev,
685 				   struct mlx5_eqe *eqe)
686 {
687 	unsigned int module_num;
688 	unsigned int module_status;
689 	unsigned int error_type;
690 	struct mlx5_eqe_port_module_event *module_event_eqe;
691 
692 	module_event_eqe = &eqe->data.port_module_event;
693 
694 	module_num = (unsigned int)module_event_eqe->module;
695 	module_status = (unsigned int)module_event_eqe->module_status &
696 	    PORT_MODULE_EVENT_MODULE_STATUS_MASK;
697 	error_type = (unsigned int)module_event_eqe->error_type &
698 	    PORT_MODULE_EVENT_ERROR_TYPE_MASK;
699 
700 	if (module_status < MLX5_MODULE_STATUS_NUM)
701 		dev->priv.pme_stats.status_counters[module_status]++;
702 	switch (module_status) {
703 	case MLX5_MODULE_STATUS_PLUGGED_ENABLED:
704 		mlx5_core_info(dev,
705 		    "Module %u, status: plugged and enabled\n",
706 		    module_num);
707 		break;
708 
709 	case MLX5_MODULE_STATUS_UNPLUGGED:
710 		mlx5_core_info(dev,
711 		    "Module %u, status: unplugged\n", module_num);
712 		break;
713 
714 	case MLX5_MODULE_STATUS_ERROR:
715 		mlx5_core_err(dev,
716 		    "Module %u, status: error, %s (%d)\n",
717 		    module_num,
718 		    mlx5_port_module_event_error_type_to_string(error_type),
719 		    error_type);
720 		if (error_type < MLX5_MODULE_EVENT_ERROR_NUM)
721 			dev->priv.pme_stats.error_counters[error_type]++;
722 		break;
723 
724 	default:
725 		mlx5_core_info(dev,
726 		    "Module %u, unknown status %d\n", module_num, module_status);
727 	}
728 	/* store module status */
729 	if (module_num < MLX5_MAX_PORTS)
730 		dev->module_status[module_num] = module_status;
731 }
732 
733 static void mlx5_port_general_notification_event(struct mlx5_core_dev *dev,
734 						 struct mlx5_eqe *eqe)
735 {
736 	u8 port = (eqe->data.port.port >> 4) & 0xf;
737 	u32 rqn;
738 	struct mlx5_eqe_general_notification_event *general_event;
739 
740 	switch (eqe->sub_type) {
741 	case MLX5_GEN_EVENT_SUBTYPE_DELAY_DROP_TIMEOUT:
742 		general_event = &eqe->data.general_notifications;
743 		rqn = be32_to_cpu(general_event->rq_user_index_delay_drop) &
744 			  0xffffff;
745 		break;
746 	case MLX5_GEN_EVENT_SUBTYPE_PCI_POWER_CHANGE_EVENT:
747 		mlx5_trigger_health_watchdog(dev);
748 		break;
749 	default:
750 		mlx5_core_warn(dev,
751 			       "general event with unrecognized subtype: port %d, sub_type %d\n",
752 			       port, eqe->sub_type);
753 		break;
754 	}
755 }
756 
757 void
758 mlx5_disable_interrupts(struct mlx5_core_dev *dev)
759 {
760 	int nvec = dev->priv.eq_table.num_comp_vectors + MLX5_EQ_VEC_COMP_BASE;
761 	int x;
762 
763 	for (x = 0; x != nvec; x++)
764 		disable_irq(dev->priv.msix_arr[x].vector);
765 }
766 
767 void
768 mlx5_poll_interrupts(struct mlx5_core_dev *dev)
769 {
770 	struct mlx5_eq *eq;
771 
772 	if (unlikely(dev->priv.disable_irqs != 0))
773 		return;
774 
775 	mlx5_eq_int(dev, &dev->priv.eq_table.cmd_eq);
776 	mlx5_eq_int(dev, &dev->priv.eq_table.async_eq);
777 	mlx5_eq_int(dev, &dev->priv.eq_table.pages_eq);
778 
779 	list_for_each_entry(eq, &dev->priv.eq_table.comp_eqs_list, list)
780 		mlx5_eq_int(dev, eq);
781 }
782