xref: /freebsd/sys/dev/mlx5/mlx5_core/mlx5_eq.c (revision e23731db48ef9c6568d4768b1f87d48514339faa)
1 /*-
2  * Copyright (c) 2013-2021, 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 
26 #include "opt_rss.h"
27 #include "opt_ratelimit.h"
28 
29 #include <linux/interrupt.h>
30 #include <linux/module.h>
31 #include <dev/mlx5/port.h>
32 #include <dev/mlx5/mlx5_ifc.h>
33 #include <dev/mlx5/mlx5_fpga/core.h>
34 #include <dev/mlx5/mlx5_core/mlx5_core.h>
35 #include <dev/mlx5/mlx5_core/eswitch.h>
36 #include <dev/mlx5/mlx5_accel/ipsec.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 	case MLX5_EVENT_TYPE_OBJECT_CHANGE:
170 		return "MLX5_EVENT_TYPE_OBJECT_CHANGE";
171 	default:
172 		return "Unrecognized event";
173 	}
174 }
175 
176 static enum mlx5_dev_event port_subtype_event(u8 subtype)
177 {
178 	switch (subtype) {
179 	case MLX5_PORT_CHANGE_SUBTYPE_DOWN:
180 		return MLX5_DEV_EVENT_PORT_DOWN;
181 	case MLX5_PORT_CHANGE_SUBTYPE_ACTIVE:
182 		return MLX5_DEV_EVENT_PORT_UP;
183 	case MLX5_PORT_CHANGE_SUBTYPE_INITIALIZED:
184 		return MLX5_DEV_EVENT_PORT_INITIALIZED;
185 	case MLX5_PORT_CHANGE_SUBTYPE_LID:
186 		return MLX5_DEV_EVENT_LID_CHANGE;
187 	case MLX5_PORT_CHANGE_SUBTYPE_PKEY:
188 		return MLX5_DEV_EVENT_PKEY_CHANGE;
189 	case MLX5_PORT_CHANGE_SUBTYPE_GUID:
190 		return MLX5_DEV_EVENT_GUID_CHANGE;
191 	case MLX5_PORT_CHANGE_SUBTYPE_CLIENT_REREG:
192 		return MLX5_DEV_EVENT_CLIENT_REREG;
193 	}
194 	return -1;
195 }
196 
197 static enum mlx5_dev_event dcbx_subevent(u8 subtype)
198 {
199 	switch (subtype) {
200 	case MLX5_DCBX_EVENT_SUBTYPE_ERROR_STATE_DCBX:
201 		return MLX5_DEV_EVENT_ERROR_STATE_DCBX;
202 	case MLX5_DCBX_EVENT_SUBTYPE_REMOTE_CONFIG_CHANGE:
203 		return MLX5_DEV_EVENT_REMOTE_CONFIG_CHANGE;
204 	case MLX5_DCBX_EVENT_SUBTYPE_LOCAL_OPER_CHANGE:
205 		return MLX5_DEV_EVENT_LOCAL_OPER_CHANGE;
206 	case MLX5_DCBX_EVENT_SUBTYPE_REMOTE_CONFIG_APP_PRIORITY_CHANGE:
207 		return MLX5_DEV_EVENT_REMOTE_CONFIG_APPLICATION_PRIORITY_CHANGE;
208 	}
209 	return -1;
210 }
211 
212 static void eq_update_ci(struct mlx5_eq *eq, int arm)
213 {
214 	__be32 __iomem *addr = eq->doorbell + (arm ? 0 : 2);
215 	u32 val = (eq->cons_index & 0xffffff) | (eq->eqn << 24);
216 	__raw_writel((__force u32) cpu_to_be32(val), addr);
217 	/* We still want ordering, just not swabbing, so add a barrier */
218 	mb();
219 }
220 
221 static void
222 mlx5_temp_warning_event(struct mlx5_core_dev *dev, struct mlx5_eqe *eqe)
223 {
224 
225 	mlx5_core_warn(dev,
226 	    "High temperature on sensors with bit set %#jx %#jx\n",
227 	    (uintmax_t)be64_to_cpu(eqe->data.temp_warning.sensor_warning_msb),
228 	    (uintmax_t)be64_to_cpu(eqe->data.temp_warning.sensor_warning_lsb));
229 }
230 
231 static int mlx5_eq_int(struct mlx5_core_dev *dev, struct mlx5_eq *eq)
232 {
233 	struct mlx5_eqe *eqe;
234 	int eqes_found = 0;
235 	int set_ci = 0;
236 	u32 cqn;
237 	u32 rsn;
238 	u8 port;
239 
240 	while ((eqe = next_eqe_sw(eq))) {
241 		/*
242 		 * Make sure we read EQ entry contents after we've
243 		 * checked the ownership bit.
244 		 */
245 		atomic_thread_fence_acq();
246 
247 		mlx5_core_dbg(eq->dev, "eqn %d, eqe type %s\n",
248 			      eq->eqn, eqe_type_str(eqe->type));
249 
250 		if (dev->priv.eq_table.cb != NULL &&
251 		    dev->priv.eq_table.cb(dev, eqe->type, &eqe->data)) {
252 			/* FALLTHROUGH */
253 		} else switch (eqe->type) {
254 		case MLX5_EVENT_TYPE_COMP:
255 			mlx5_cq_completion(dev, eqe);
256 			break;
257 
258 		case MLX5_EVENT_TYPE_PATH_MIG:
259 		case MLX5_EVENT_TYPE_COMM_EST:
260 		case MLX5_EVENT_TYPE_SQ_DRAINED:
261 		case MLX5_EVENT_TYPE_SRQ_LAST_WQE:
262 		case MLX5_EVENT_TYPE_WQ_CATAS_ERROR:
263 		case MLX5_EVENT_TYPE_PATH_MIG_FAILED:
264 		case MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
265 		case MLX5_EVENT_TYPE_WQ_ACCESS_ERROR:
266 			rsn = be32_to_cpu(eqe->data.qp_srq.qp_srq_n) & 0xffffff;
267 			mlx5_core_dbg(dev, "event %s(%d) arrived on resource 0x%x\n",
268 				      eqe_type_str(eqe->type), eqe->type, rsn);
269 			mlx5_rsc_event(dev, rsn, eqe->type);
270 			break;
271 
272 		case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
273 		case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
274 			rsn = be32_to_cpu(eqe->data.qp_srq.qp_srq_n) & 0xffffff;
275 			mlx5_core_dbg(dev, "SRQ event %s(%d): srqn 0x%x\n",
276 				      eqe_type_str(eqe->type), eqe->type, rsn);
277 			mlx5_srq_event(dev, rsn, eqe->type);
278 			break;
279 
280 		case MLX5_EVENT_TYPE_CMD:
281 			if (dev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR) {
282 				mlx5_cmd_comp_handler(dev, be32_to_cpu(eqe->data.cmd.vector),
283 				    MLX5_CMD_MODE_EVENTS);
284 			}
285 			break;
286 
287 		case MLX5_EVENT_TYPE_PORT_CHANGE:
288 			port = (eqe->data.port.port >> 4) & 0xf;
289 			switch (eqe->sub_type) {
290 			case MLX5_PORT_CHANGE_SUBTYPE_DOWN:
291 			case MLX5_PORT_CHANGE_SUBTYPE_ACTIVE:
292 			case MLX5_PORT_CHANGE_SUBTYPE_LID:
293 			case MLX5_PORT_CHANGE_SUBTYPE_PKEY:
294 			case MLX5_PORT_CHANGE_SUBTYPE_GUID:
295 			case MLX5_PORT_CHANGE_SUBTYPE_CLIENT_REREG:
296 			case MLX5_PORT_CHANGE_SUBTYPE_INITIALIZED:
297 				if (dev->event)
298 					dev->event(dev, port_subtype_event(eqe->sub_type),
299 						   (unsigned long)port);
300 				break;
301 			default:
302 				mlx5_core_warn(dev, "Port event with unrecognized subtype: port %d, sub_type %d\n",
303 					       port, eqe->sub_type);
304 			}
305 			break;
306 
307 		case MLX5_EVENT_TYPE_CODING_DCBX_CHANGE_EVENT:
308 			port = (eqe->data.port.port >> 4) & 0xf;
309 			switch (eqe->sub_type) {
310 			case MLX5_DCBX_EVENT_SUBTYPE_ERROR_STATE_DCBX:
311 			case MLX5_DCBX_EVENT_SUBTYPE_REMOTE_CONFIG_CHANGE:
312 			case MLX5_DCBX_EVENT_SUBTYPE_LOCAL_OPER_CHANGE:
313 			case MLX5_DCBX_EVENT_SUBTYPE_REMOTE_CONFIG_APP_PRIORITY_CHANGE:
314 				if (dev->event)
315 					dev->event(dev,
316 						   dcbx_subevent(eqe->sub_type),
317 						   0);
318 				break;
319 			default:
320 				mlx5_core_warn(dev,
321 					       "dcbx event with unrecognized subtype: port %d, sub_type %d\n",
322 					       port, eqe->sub_type);
323 			}
324 			break;
325 
326 		case MLX5_EVENT_TYPE_CODING_GENERAL_NOTIFICATION_EVENT:
327 			mlx5_port_general_notification_event(dev, eqe);
328 			break;
329 
330 		case MLX5_EVENT_TYPE_CQ_ERROR:
331 			cqn = be32_to_cpu(eqe->data.cq_err.cqn) & 0xffffff;
332 			mlx5_core_warn(dev, "CQ error on CQN 0x%x, syndrom 0x%x\n",
333 				       cqn, eqe->data.cq_err.syndrome);
334 			mlx5_cq_event(dev, cqn, eqe->type);
335 			break;
336 
337 		case MLX5_EVENT_TYPE_PAGE_REQUEST:
338 			{
339 				u16 func_id = be16_to_cpu(eqe->data.req_pages.func_id);
340 				s32 npages = be32_to_cpu(eqe->data.req_pages.num_pages);
341 
342 				mlx5_core_dbg(dev, "page request for func 0x%x, npages %d\n",
343 					      func_id, npages);
344 				mlx5_core_req_pages_handler(dev, func_id, npages);
345 			}
346 			break;
347 
348 		case MLX5_EVENT_TYPE_CODING_PORT_MODULE_EVENT:
349 			mlx5_port_module_event(dev, eqe);
350 			break;
351 
352 		case MLX5_EVENT_TYPE_NIC_VPORT_CHANGE:
353 			{
354 				struct mlx5_eqe_vport_change *vc_eqe =
355 						&eqe->data.vport_change;
356 				u16 vport_num = be16_to_cpu(vc_eqe->vport_num);
357 
358 				if (dev->event)
359 					dev->event(dev,
360 					     MLX5_DEV_EVENT_VPORT_CHANGE,
361 					     (unsigned long)vport_num);
362 			}
363 			if (dev->priv.eswitch != NULL)
364 				mlx5_eswitch_vport_event(dev->priv.eswitch,
365 				    eqe);
366 			break;
367 
368 		case MLX5_EVENT_TYPE_FPGA_ERROR:
369 		case MLX5_EVENT_TYPE_FPGA_QP_ERROR:
370 			mlx5_fpga_event(dev, eqe->type, &eqe->data.raw);
371 			break;
372 		case MLX5_EVENT_TYPE_TEMP_WARN_EVENT:
373 			mlx5_temp_warning_event(dev, eqe);
374 			break;
375 
376 		case MLX5_EVENT_TYPE_OBJECT_CHANGE:
377 			mlx5_object_change_event(dev, eqe);
378 			break;
379 
380 		default:
381 			mlx5_core_warn(dev, "Unhandled event 0x%x on EQ 0x%x\n",
382 				       eqe->type, eq->eqn);
383 			break;
384 		}
385 
386 		++eq->cons_index;
387 		eqes_found = 1;
388 		++set_ci;
389 
390 		/* The HCA will think the queue has overflowed if we
391 		 * don't tell it we've been processing events.  We
392 		 * create our EQs with MLX5_NUM_SPARE_EQE extra
393 		 * entries, so we must update our consumer index at
394 		 * least that often.
395 		 */
396 		if (unlikely(set_ci >= MLX5_NUM_SPARE_EQE)) {
397 			eq_update_ci(eq, 0);
398 			set_ci = 0;
399 		}
400 	}
401 
402 	eq_update_ci(eq, 1);
403 
404 	return eqes_found;
405 }
406 
407 static irqreturn_t mlx5_msix_handler(int irq, void *eq_ptr)
408 {
409 	struct mlx5_eq *eq = eq_ptr;
410 	struct mlx5_core_dev *dev = eq->dev;
411 
412 	/* check if IRQs are not disabled */
413 	if (likely(dev->priv.disable_irqs == 0))
414 		mlx5_eq_int(dev, eq);
415 
416 	/* MSI-X vectors always belong to us */
417 	return IRQ_HANDLED;
418 }
419 
420 static void init_eq_buf(struct mlx5_eq *eq)
421 {
422 	struct mlx5_eqe *eqe;
423 	int i;
424 
425 	for (i = 0; i < eq->nent; i++) {
426 		eqe = get_eqe(eq, i);
427 		eqe->owner = MLX5_EQE_OWNER_INIT_VAL;
428 	}
429 }
430 
431 int mlx5_create_map_eq(struct mlx5_core_dev *dev, struct mlx5_eq *eq, u8 vecidx,
432 		       int nent, u64 mask)
433 {
434 	u32 out[MLX5_ST_SZ_DW(create_eq_out)] = {0};
435 	struct mlx5_priv *priv = &dev->priv;
436 	__be64 *pas;
437 	void *eqc;
438 	int inlen;
439 	u32 *in;
440 	int err;
441 
442 	eq->nent = roundup_pow_of_two(nent + MLX5_NUM_SPARE_EQE);
443 	eq->cons_index = 0;
444 	err = mlx5_buf_alloc(dev, eq->nent * MLX5_EQE_SIZE, 2 * PAGE_SIZE,
445 			     &eq->buf);
446 	if (err)
447 		return err;
448 
449 	init_eq_buf(eq);
450 
451 	inlen = MLX5_ST_SZ_BYTES(create_eq_in) +
452 		MLX5_FLD_SZ_BYTES(create_eq_in, pas[0]) * eq->buf.npages;
453 	in = mlx5_vzalloc(inlen);
454 	if (!in) {
455 		err = -ENOMEM;
456 		goto err_buf;
457 	}
458 
459 	pas = (__be64 *)MLX5_ADDR_OF(create_eq_in, in, pas);
460 	mlx5_fill_page_array(&eq->buf, pas);
461 
462 	MLX5_SET(create_eq_in, in, opcode, MLX5_CMD_OP_CREATE_EQ);
463 	MLX5_SET64(create_eq_in, in, event_bitmask, mask);
464 
465 	eqc = MLX5_ADDR_OF(create_eq_in, in, eq_context_entry);
466 	MLX5_SET(eqc, eqc, log_eq_size, ilog2(eq->nent));
467 	MLX5_SET(eqc, eqc, uar_page, priv->uar->index);
468 	MLX5_SET(eqc, eqc, intr, vecidx);
469 	MLX5_SET(eqc, eqc, log_page_size,
470 		 eq->buf.page_shift - MLX5_ADAPTER_PAGE_SHIFT);
471 
472 	err = mlx5_cmd_exec(dev, in, inlen, out, sizeof(out));
473 	if (err)
474 		goto err_in;
475 
476 	eq->eqn = MLX5_GET(create_eq_out, out, eq_number);
477 	eq->irqn = vecidx;
478 	eq->dev = dev;
479 	eq->doorbell = priv->uar->map + MLX5_EQ_DOORBEL_OFFSET;
480 	err = request_irq(priv->msix_arr[vecidx].vector, mlx5_msix_handler, 0,
481 			  "mlx5_core", eq);
482 	if (err)
483 		goto err_eq;
484 #ifdef RSS
485 	if (vecidx >= MLX5_EQ_VEC_COMP_BASE) {
486 		u8 bucket = vecidx - MLX5_EQ_VEC_COMP_BASE;
487 		err = bind_irq_to_cpu(priv->msix_arr[vecidx].vector,
488 				      rss_getcpu(bucket % rss_getnumbuckets()));
489 		if (err)
490 			goto err_irq;
491 	}
492 #else
493 	if (0)
494 		goto err_irq;
495 #endif
496 
497 
498 	/* EQs are created in ARMED state
499 	 */
500 	eq_update_ci(eq, 1);
501 
502 	kvfree(in);
503 	return 0;
504 
505 err_irq:
506 	free_irq(priv->msix_arr[vecidx].vector, eq);
507 
508 err_eq:
509 	mlx5_cmd_destroy_eq(dev, eq->eqn);
510 
511 err_in:
512 	kvfree(in);
513 
514 err_buf:
515 	mlx5_buf_free(dev, &eq->buf);
516 	return err;
517 }
518 EXPORT_SYMBOL_GPL(mlx5_create_map_eq);
519 
520 int mlx5_destroy_unmap_eq(struct mlx5_core_dev *dev, struct mlx5_eq *eq)
521 {
522 	int err;
523 
524 	free_irq(dev->priv.msix_arr[eq->irqn].vector, eq);
525 	err = mlx5_cmd_destroy_eq(dev, eq->eqn);
526 	if (err)
527 		mlx5_core_warn(dev, "failed to destroy a previously created eq: eqn %d\n",
528 			       eq->eqn);
529 	mlx5_buf_free(dev, &eq->buf);
530 
531 	return err;
532 }
533 EXPORT_SYMBOL_GPL(mlx5_destroy_unmap_eq);
534 
535 int mlx5_eq_init(struct mlx5_core_dev *dev)
536 {
537 	int err;
538 
539 	spin_lock_init(&dev->priv.eq_table.lock);
540 
541 	err = 0;
542 
543 	return err;
544 }
545 
546 
547 void mlx5_eq_cleanup(struct mlx5_core_dev *dev)
548 {
549 }
550 
551 int mlx5_start_eqs(struct mlx5_core_dev *dev)
552 {
553 	struct mlx5_eq_table *table = &dev->priv.eq_table;
554 	u64 async_event_mask = MLX5_ASYNC_EVENT_MASK;
555 	int err;
556 
557 	if (MLX5_CAP_GEN(dev, port_module_event))
558 		async_event_mask |= (1ull <<
559 				     MLX5_EVENT_TYPE_CODING_PORT_MODULE_EVENT);
560 
561 	if (MLX5_CAP_GEN(dev, nic_vport_change_event))
562 		async_event_mask |= (1ull <<
563 				     MLX5_EVENT_TYPE_NIC_VPORT_CHANGE);
564 
565 	if (MLX5_CAP_GEN(dev, dcbx))
566 		async_event_mask |= (1ull <<
567 				     MLX5_EVENT_TYPE_CODING_DCBX_CHANGE_EVENT);
568 
569 	if (MLX5_CAP_GEN(dev, fpga))
570 		async_event_mask |= (1ull << MLX5_EVENT_TYPE_FPGA_ERROR) |
571 				    (1ull << MLX5_EVENT_TYPE_FPGA_QP_ERROR);
572 
573 	if (MLX5_CAP_GEN(dev, temp_warn_event))
574 		async_event_mask |= (1ull << MLX5_EVENT_TYPE_TEMP_WARN_EVENT);
575 
576 	if (MLX5_CAP_GEN(dev, general_notification_event)) {
577 		async_event_mask |= (1ull <<
578 		    MLX5_EVENT_TYPE_CODING_GENERAL_NOTIFICATION_EVENT);
579 	}
580 
581 	if (mlx5_ipsec_device_caps(dev) & MLX5_IPSEC_CAP_PACKET_OFFLOAD)
582 		async_event_mask |=
583 			(1ull << MLX5_EVENT_TYPE_OBJECT_CHANGE);
584 
585 	err = mlx5_create_map_eq(dev, &table->cmd_eq, MLX5_EQ_VEC_CMD,
586 				 MLX5_NUM_CMD_EQE, 1ull << MLX5_EVENT_TYPE_CMD);
587 	if (err) {
588 		mlx5_core_warn(dev, "failed to create cmd EQ %d\n", err);
589 		return err;
590 	}
591 
592 	mlx5_cmd_use_events(dev);
593 
594 	err = mlx5_create_map_eq(dev, &table->async_eq, MLX5_EQ_VEC_ASYNC,
595 				 MLX5_NUM_ASYNC_EQE, async_event_mask);
596 	if (err) {
597 		mlx5_core_warn(dev, "failed to create async EQ %d\n", err);
598 		goto err1;
599 	}
600 
601 	err = mlx5_create_map_eq(dev, &table->pages_eq,
602 				 MLX5_EQ_VEC_PAGES,
603 				 /* TODO: sriov max_vf + */ 1,
604 				 1 << MLX5_EVENT_TYPE_PAGE_REQUEST);
605 	if (err) {
606 		mlx5_core_warn(dev, "failed to create pages EQ %d\n", err);
607 		goto err2;
608 	}
609 
610 	return err;
611 
612 err2:
613 	mlx5_destroy_unmap_eq(dev, &table->async_eq);
614 
615 err1:
616 	mlx5_cmd_use_polling(dev);
617 	mlx5_destroy_unmap_eq(dev, &table->cmd_eq);
618 	return err;
619 }
620 
621 int mlx5_stop_eqs(struct mlx5_core_dev *dev)
622 {
623 	struct mlx5_eq_table *table = &dev->priv.eq_table;
624 	int err;
625 
626 	err = mlx5_destroy_unmap_eq(dev, &table->pages_eq);
627 	if (err)
628 		return err;
629 
630 	mlx5_destroy_unmap_eq(dev, &table->async_eq);
631 	mlx5_cmd_use_polling(dev);
632 
633 	err = mlx5_destroy_unmap_eq(dev, &table->cmd_eq);
634 	if (err)
635 		mlx5_cmd_use_events(dev);
636 
637 	return err;
638 }
639 
640 int mlx5_core_eq_query(struct mlx5_core_dev *dev, struct mlx5_eq *eq,
641 		       u32 *out, int outlen)
642 {
643 	u32 in[MLX5_ST_SZ_DW(query_eq_in)] = {0};
644 
645 	memset(out, 0, outlen);
646 	MLX5_SET(query_eq_in, in, opcode, MLX5_CMD_OP_QUERY_EQ);
647 	MLX5_SET(query_eq_in, in, eq_number, eq->eqn);
648 
649 	return mlx5_cmd_exec(dev, in, sizeof(in), out, outlen);
650 }
651 EXPORT_SYMBOL_GPL(mlx5_core_eq_query);
652 
653 static const char *mlx5_port_module_event_error_type_to_string(u8 error_type)
654 {
655 	switch (error_type) {
656 	case MLX5_MODULE_EVENT_ERROR_POWER_BUDGET_EXCEEDED:
657 		return "Power budget exceeded";
658 	case MLX5_MODULE_EVENT_ERROR_LONG_RANGE_FOR_NON_MLNX_CABLE_MODULE:
659 		return "Long Range for non MLNX cable";
660 	case MLX5_MODULE_EVENT_ERROR_BUS_STUCK:
661 		return "Bus stuck(I2C or data shorted)";
662 	case MLX5_MODULE_EVENT_ERROR_NO_EEPROM_RETRY_TIMEOUT:
663 		return "No EEPROM/retry timeout";
664 	case MLX5_MODULE_EVENT_ERROR_ENFORCE_PART_NUMBER_LIST:
665 		return "Enforce part number list";
666 	case MLX5_MODULE_EVENT_ERROR_UNSUPPORTED_CABLE:
667 		return "Unknown identifier";
668 	case MLX5_MODULE_EVENT_ERROR_HIGH_TEMPERATURE:
669 		return "High Temperature";
670 	case MLX5_MODULE_EVENT_ERROR_CABLE_IS_SHORTED:
671 		return "Bad or shorted cable/module";
672 	case MLX5_MODULE_EVENT_ERROR_PMD_TYPE_NOT_ENABLED:
673 		return "PMD type is not enabled";
674 	case MLX5_MODULE_EVENT_ERROR_LASTER_TEC_FAILURE:
675 		return "Laster_TEC_failure";
676 	case MLX5_MODULE_EVENT_ERROR_HIGH_CURRENT:
677 		return "High_current";
678 	case MLX5_MODULE_EVENT_ERROR_HIGH_VOLTAGE:
679 		return "High_voltage";
680 	case MLX5_MODULE_EVENT_ERROR_PCIE_SYS_POWER_SLOT_EXCEEDED:
681 		return "pcie_system_power_slot_Exceeded";
682 	case MLX5_MODULE_EVENT_ERROR_HIGH_POWER:
683 		return "High_power";
684 	case MLX5_MODULE_EVENT_ERROR_MODULE_STATE_MACHINE_FAULT:
685 		return "Module_state_machine_fault";
686 	default:
687 		return "Unknown error type";
688 	}
689 }
690 
691 unsigned int mlx5_query_module_status(struct mlx5_core_dev *dev, int module_num)
692 {
693 	if (module_num < 0 || module_num >= MLX5_MAX_PORTS)
694 		return 0;		/* undefined */
695 	return dev->module_status[module_num];
696 }
697 
698 static void mlx5_port_module_event(struct mlx5_core_dev *dev,
699 				   struct mlx5_eqe *eqe)
700 {
701 	unsigned int module_num;
702 	unsigned int module_status;
703 	unsigned int error_type;
704 	struct mlx5_eqe_port_module_event *module_event_eqe;
705 
706 	module_event_eqe = &eqe->data.port_module_event;
707 
708 	module_num = (unsigned int)module_event_eqe->module;
709 	module_status = (unsigned int)module_event_eqe->module_status &
710 	    PORT_MODULE_EVENT_MODULE_STATUS_MASK;
711 	error_type = (unsigned int)module_event_eqe->error_type &
712 	    PORT_MODULE_EVENT_ERROR_TYPE_MASK;
713 
714 	if (module_status < MLX5_MODULE_STATUS_NUM)
715 		dev->priv.pme_stats.status_counters[module_status]++;
716 	switch (module_status) {
717 	case MLX5_MODULE_STATUS_PLUGGED_ENABLED:
718 		mlx5_core_info(dev,
719 		    "Module %u, status: plugged and enabled\n",
720 		    module_num);
721 		break;
722 
723 	case MLX5_MODULE_STATUS_UNPLUGGED:
724 		mlx5_core_info(dev,
725 		    "Module %u, status: unplugged\n", module_num);
726 		break;
727 
728 	case MLX5_MODULE_STATUS_ERROR:
729 		mlx5_core_err(dev,
730 		    "Module %u, status: error, %s (%d)\n",
731 		    module_num,
732 		    mlx5_port_module_event_error_type_to_string(error_type),
733 		    error_type);
734 		if (error_type < MLX5_MODULE_EVENT_ERROR_NUM)
735 			dev->priv.pme_stats.error_counters[error_type]++;
736 		break;
737 
738 	default:
739 		mlx5_core_info(dev,
740 		    "Module %u, unknown status %d\n", module_num, module_status);
741 	}
742 	/* store module status */
743 	if (module_num < MLX5_MAX_PORTS)
744 		dev->module_status[module_num] = module_status;
745 }
746 
747 static void mlx5_port_general_notification_event(struct mlx5_core_dev *dev,
748 						 struct mlx5_eqe *eqe)
749 {
750 	u8 port = (eqe->data.port.port >> 4) & 0xf;
751 
752 	switch (eqe->sub_type) {
753 	case MLX5_GEN_EVENT_SUBTYPE_DELAY_DROP_TIMEOUT:
754 		break;
755 	case MLX5_GEN_EVENT_SUBTYPE_PCI_POWER_CHANGE_EVENT:
756 		mlx5_trigger_health_watchdog(dev);
757 		break;
758 	default:
759 		mlx5_core_warn(dev,
760 			       "general event with unrecognized subtype: port %d, sub_type %d\n",
761 			       port, eqe->sub_type);
762 		break;
763 	}
764 }
765 
766 void
767 mlx5_disable_interrupts(struct mlx5_core_dev *dev)
768 {
769 	int nvec = dev->priv.eq_table.num_comp_vectors + MLX5_EQ_VEC_COMP_BASE;
770 	int x;
771 
772 	for (x = 0; x != nvec; x++)
773 		disable_irq(dev->priv.msix_arr[x].vector);
774 }
775 
776 void
777 mlx5_poll_interrupts(struct mlx5_core_dev *dev)
778 {
779 	struct mlx5_eq *eq;
780 
781 	if (unlikely(dev->priv.disable_irqs != 0))
782 		return;
783 
784 	mlx5_eq_int(dev, &dev->priv.eq_table.cmd_eq);
785 	mlx5_eq_int(dev, &dev->priv.eq_table.async_eq);
786 	mlx5_eq_int(dev, &dev->priv.eq_table.pages_eq);
787 
788 	list_for_each_entry(eq, &dev->priv.eq_table.comp_eqs_list, list)
789 		mlx5_eq_int(dev, eq);
790 }
791