1 /*
2  * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 
33 #include <linux/kernel.h>
34 #include <linux/random.h>
35 #include <linux/vmalloc.h>
36 #include <linux/hardirq.h>
37 #include <linux/mlx5/driver.h>
38 #include <linux/kern_levels.h>
39 #include "mlx5_core.h"
40 #include "lib/eq.h"
41 #include "lib/mlx5.h"
42 #include "lib/events.h"
43 #include "lib/pci_vsc.h"
44 #include "lib/tout.h"
45 #include "diag/fw_tracer.h"
46 #include "diag/reporter_vnic.h"
47 
48 enum {
49 	MAX_MISSES			= 3,
50 };
51 
52 enum {
53 	MLX5_DROP_HEALTH_WORK,
54 };
55 
56 enum  {
57 	MLX5_SENSOR_NO_ERR		= 0,
58 	MLX5_SENSOR_PCI_COMM_ERR	= 1,
59 	MLX5_SENSOR_PCI_ERR		= 2,
60 	MLX5_SENSOR_NIC_DISABLED	= 3,
61 	MLX5_SENSOR_NIC_SW_RESET	= 4,
62 	MLX5_SENSOR_FW_SYND_RFR		= 5,
63 };
64 
65 enum {
66 	MLX5_SEVERITY_MASK		= 0x7,
67 	MLX5_SEVERITY_VALID_MASK	= 0x8,
68 };
69 
mlx5_get_nic_state(struct mlx5_core_dev * dev)70 u8 mlx5_get_nic_state(struct mlx5_core_dev *dev)
71 {
72 	return (ioread32be(&dev->iseg->cmdq_addr_l_sz) >> 8) & 7;
73 }
74 
mlx5_set_nic_state(struct mlx5_core_dev * dev,u8 state)75 void mlx5_set_nic_state(struct mlx5_core_dev *dev, u8 state)
76 {
77 	u32 cur_cmdq_addr_l_sz;
78 
79 	cur_cmdq_addr_l_sz = ioread32be(&dev->iseg->cmdq_addr_l_sz);
80 	iowrite32be((cur_cmdq_addr_l_sz & 0xFFFFF000) |
81 		    state << MLX5_NIC_IFC_OFFSET,
82 		    &dev->iseg->cmdq_addr_l_sz);
83 }
84 
sensor_pci_not_working(struct mlx5_core_dev * dev)85 static bool sensor_pci_not_working(struct mlx5_core_dev *dev)
86 {
87 	struct mlx5_core_health *health = &dev->priv.health;
88 	struct health_buffer __iomem *h = health->health;
89 
90 	/* Offline PCI reads return 0xffffffff */
91 	return (ioread32be(&h->fw_ver) == 0xffffffff);
92 }
93 
mlx5_health_get_rfr(u8 rfr_severity)94 static int mlx5_health_get_rfr(u8 rfr_severity)
95 {
96 	return rfr_severity >> MLX5_RFR_BIT_OFFSET;
97 }
98 
sensor_fw_synd_rfr(struct mlx5_core_dev * dev)99 static bool sensor_fw_synd_rfr(struct mlx5_core_dev *dev)
100 {
101 	struct mlx5_core_health *health = &dev->priv.health;
102 	struct health_buffer __iomem *h = health->health;
103 	u8 synd = ioread8(&h->synd);
104 	u8 rfr;
105 
106 	rfr = mlx5_health_get_rfr(ioread8(&h->rfr_severity));
107 
108 	if (rfr && synd)
109 		mlx5_core_dbg(dev, "FW requests reset, synd: %d\n", synd);
110 	return rfr && synd;
111 }
112 
mlx5_health_check_fatal_sensors(struct mlx5_core_dev * dev)113 u32 mlx5_health_check_fatal_sensors(struct mlx5_core_dev *dev)
114 {
115 	if (sensor_pci_not_working(dev))
116 		return MLX5_SENSOR_PCI_COMM_ERR;
117 	if (pci_channel_offline(dev->pdev))
118 		return MLX5_SENSOR_PCI_ERR;
119 	if (mlx5_get_nic_state(dev) == MLX5_NIC_IFC_DISABLED)
120 		return MLX5_SENSOR_NIC_DISABLED;
121 	if (mlx5_get_nic_state(dev) == MLX5_NIC_IFC_SW_RESET)
122 		return MLX5_SENSOR_NIC_SW_RESET;
123 	if (sensor_fw_synd_rfr(dev))
124 		return MLX5_SENSOR_FW_SYND_RFR;
125 
126 	return MLX5_SENSOR_NO_ERR;
127 }
128 
lock_sem_sw_reset(struct mlx5_core_dev * dev,bool lock)129 static int lock_sem_sw_reset(struct mlx5_core_dev *dev, bool lock)
130 {
131 	enum mlx5_vsc_state state;
132 	int ret;
133 
134 	if (!mlx5_core_is_pf(dev))
135 		return -EBUSY;
136 
137 	/* Try to lock GW access, this stage doesn't return
138 	 * EBUSY because locked GW does not mean that other PF
139 	 * already started the reset.
140 	 */
141 	ret = mlx5_vsc_gw_lock(dev);
142 	if (ret == -EBUSY)
143 		return -EINVAL;
144 	if (ret)
145 		return ret;
146 
147 	state = lock ? MLX5_VSC_LOCK : MLX5_VSC_UNLOCK;
148 	/* At this stage, if the return status == EBUSY, then we know
149 	 * for sure that another PF started the reset, so don't allow
150 	 * another reset.
151 	 */
152 	ret = mlx5_vsc_sem_set_space(dev, MLX5_SEMAPHORE_SW_RESET, state);
153 	if (ret)
154 		mlx5_core_warn(dev, "Failed to lock SW reset semaphore\n");
155 
156 	/* Unlock GW access */
157 	mlx5_vsc_gw_unlock(dev);
158 
159 	return ret;
160 }
161 
reset_fw_if_needed(struct mlx5_core_dev * dev)162 static bool reset_fw_if_needed(struct mlx5_core_dev *dev)
163 {
164 	bool supported = (ioread32be(&dev->iseg->initializing) >>
165 			  MLX5_FW_RESET_SUPPORTED_OFFSET) & 1;
166 	u32 fatal_error;
167 
168 	if (!supported)
169 		return false;
170 
171 	/* The reset only needs to be issued by one PF. The health buffer is
172 	 * shared between all functions, and will be cleared during a reset.
173 	 * Check again to avoid a redundant 2nd reset. If the fatal errors was
174 	 * PCI related a reset won't help.
175 	 */
176 	fatal_error = mlx5_health_check_fatal_sensors(dev);
177 	if (fatal_error == MLX5_SENSOR_PCI_COMM_ERR ||
178 	    fatal_error == MLX5_SENSOR_NIC_DISABLED ||
179 	    fatal_error == MLX5_SENSOR_NIC_SW_RESET) {
180 		mlx5_core_warn(dev, "Not issuing FW reset. Either it's already done or won't help.");
181 		return false;
182 	}
183 
184 	mlx5_core_warn(dev, "Issuing FW Reset\n");
185 	/* Write the NIC interface field to initiate the reset, the command
186 	 * interface address also resides here, don't overwrite it.
187 	 */
188 	mlx5_set_nic_state(dev, MLX5_NIC_IFC_SW_RESET);
189 
190 	return true;
191 }
192 
enter_error_state(struct mlx5_core_dev * dev,bool force)193 static void enter_error_state(struct mlx5_core_dev *dev, bool force)
194 {
195 	if (mlx5_health_check_fatal_sensors(dev) || force) { /* protected state setting */
196 		dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
197 		mlx5_cmd_flush(dev);
198 	}
199 
200 	mlx5_notifier_call_chain(dev->priv.events, MLX5_DEV_EVENT_SYS_ERROR, (void *)1);
201 }
202 
mlx5_enter_error_state(struct mlx5_core_dev * dev,bool force)203 void mlx5_enter_error_state(struct mlx5_core_dev *dev, bool force)
204 {
205 	bool err_detected = false;
206 
207 	/* Mark the device as fatal in order to abort FW commands */
208 	if ((mlx5_health_check_fatal_sensors(dev) || force) &&
209 	    dev->state == MLX5_DEVICE_STATE_UP) {
210 		dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
211 		err_detected = true;
212 	}
213 	mutex_lock(&dev->intf_state_mutex);
214 	if (!err_detected && dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)
215 		goto unlock;/* a previous error is still being handled */
216 
217 	enter_error_state(dev, force);
218 unlock:
219 	mutex_unlock(&dev->intf_state_mutex);
220 }
221 
mlx5_error_sw_reset(struct mlx5_core_dev * dev)222 void mlx5_error_sw_reset(struct mlx5_core_dev *dev)
223 {
224 	unsigned long end, delay_ms = mlx5_tout_ms(dev, PCI_TOGGLE);
225 	int lock = -EBUSY;
226 
227 	mutex_lock(&dev->intf_state_mutex);
228 	if (dev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR)
229 		goto unlock;
230 
231 	mlx5_core_err(dev, "start\n");
232 
233 	if (mlx5_health_check_fatal_sensors(dev) == MLX5_SENSOR_FW_SYND_RFR) {
234 		/* Get cr-dump and reset FW semaphore */
235 		lock = lock_sem_sw_reset(dev, true);
236 
237 		if (lock == -EBUSY) {
238 			delay_ms = mlx5_tout_ms(dev, FULL_CRDUMP);
239 			goto recover_from_sw_reset;
240 		}
241 		/* Execute SW reset */
242 		reset_fw_if_needed(dev);
243 	}
244 
245 recover_from_sw_reset:
246 	/* Recover from SW reset */
247 	end = jiffies + msecs_to_jiffies(delay_ms);
248 	do {
249 		if (mlx5_get_nic_state(dev) == MLX5_NIC_IFC_DISABLED)
250 			break;
251 		if (pci_channel_offline(dev->pdev)) {
252 			mlx5_core_err(dev, "PCI channel offline, stop waiting for NIC IFC\n");
253 			goto unlock;
254 		}
255 
256 		msleep(20);
257 	} while (!time_after(jiffies, end));
258 
259 	if (mlx5_get_nic_state(dev) != MLX5_NIC_IFC_DISABLED) {
260 		dev_err(&dev->pdev->dev, "NIC IFC still %d after %lums.\n",
261 			mlx5_get_nic_state(dev), delay_ms);
262 	}
263 
264 	/* Release FW semaphore if you are the lock owner */
265 	if (!lock)
266 		lock_sem_sw_reset(dev, false);
267 
268 	mlx5_core_err(dev, "end\n");
269 
270 unlock:
271 	mutex_unlock(&dev->intf_state_mutex);
272 }
273 
mlx5_handle_bad_state(struct mlx5_core_dev * dev)274 static void mlx5_handle_bad_state(struct mlx5_core_dev *dev)
275 {
276 	u8 nic_interface = mlx5_get_nic_state(dev);
277 
278 	switch (nic_interface) {
279 	case MLX5_NIC_IFC_FULL:
280 		mlx5_core_warn(dev, "Expected to see disabled NIC but it is full driver\n");
281 		break;
282 
283 	case MLX5_NIC_IFC_DISABLED:
284 		mlx5_core_warn(dev, "starting teardown\n");
285 		break;
286 
287 	case MLX5_NIC_IFC_NO_DRAM_NIC:
288 		mlx5_core_warn(dev, "Expected to see disabled NIC but it is no dram nic\n");
289 		break;
290 
291 	case MLX5_NIC_IFC_SW_RESET:
292 		/* The IFC mode field is 3 bits, so it will read 0x7 in 2 cases:
293 		 * 1. PCI has been disabled (ie. PCI-AER, PF driver unloaded
294 		 *    and this is a VF), this is not recoverable by SW reset.
295 		 *    Logging of this is handled elsewhere.
296 		 * 2. FW reset has been issued by another function, driver can
297 		 *    be reloaded to recover after the mode switches to
298 		 *    MLX5_NIC_IFC_DISABLED.
299 		 */
300 		if (dev->priv.health.fatal_error != MLX5_SENSOR_PCI_COMM_ERR)
301 			mlx5_core_warn(dev, "NIC SW reset in progress\n");
302 		break;
303 
304 	default:
305 		mlx5_core_warn(dev, "Expected to see disabled NIC but it is has invalid value %d\n",
306 			       nic_interface);
307 	}
308 
309 	mlx5_disable_device(dev);
310 }
311 
mlx5_health_wait_pci_up(struct mlx5_core_dev * dev)312 int mlx5_health_wait_pci_up(struct mlx5_core_dev *dev)
313 {
314 	unsigned long end;
315 
316 	end = jiffies + msecs_to_jiffies(mlx5_tout_ms(dev, FW_RESET));
317 	while (sensor_pci_not_working(dev)) {
318 		if (time_after(jiffies, end))
319 			return -ETIMEDOUT;
320 		if (test_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state)) {
321 			mlx5_core_warn(dev, "device is being removed, stop waiting for PCI\n");
322 			return -ENODEV;
323 		}
324 		if (pci_channel_offline(dev->pdev)) {
325 			mlx5_core_err(dev, "PCI channel offline, stop waiting for PCI\n");
326 			return -EACCES;
327 		}
328 		msleep(100);
329 	}
330 	return 0;
331 }
332 
mlx5_health_try_recover(struct mlx5_core_dev * dev)333 static int mlx5_health_try_recover(struct mlx5_core_dev *dev)
334 {
335 	mlx5_core_warn(dev, "handling bad device here\n");
336 	mlx5_handle_bad_state(dev);
337 	if (mlx5_health_wait_pci_up(dev)) {
338 		mlx5_core_err(dev, "health recovery flow aborted, PCI reads still not working\n");
339 		return -EIO;
340 	}
341 	mlx5_core_err(dev, "starting health recovery flow\n");
342 	if (mlx5_recover_device(dev) || mlx5_health_check_fatal_sensors(dev)) {
343 		mlx5_core_err(dev, "health recovery failed\n");
344 		return -EIO;
345 	}
346 
347 	mlx5_core_info(dev, "health recovery succeeded\n");
348 	return 0;
349 }
350 
hsynd_str(u8 synd)351 static const char *hsynd_str(u8 synd)
352 {
353 	switch (synd) {
354 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_FW_INTERNAL_ERR:
355 		return "firmware internal error";
356 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_DEAD_IRISC:
357 		return "irisc not responding";
358 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_HW_FATAL_ERR:
359 		return "unrecoverable hardware error";
360 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_FW_CRC_ERR:
361 		return "firmware CRC error";
362 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_ICM_FETCH_PCI_ERR:
363 		return "ICM fetch PCI error";
364 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_ICM_PAGE_ERR:
365 		return "HW fatal error\n";
366 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_ASYNCHRONOUS_EQ_BUF_OVERRUN:
367 		return "async EQ buffer overrun";
368 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_EQ_IN_ERR:
369 		return "EQ error";
370 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_EQ_INV:
371 		return "Invalid EQ referenced";
372 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_FFSER_ERR:
373 		return "FFSER error";
374 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_HIGH_TEMP_ERR:
375 		return "High temperature";
376 	default:
377 		return "unrecognized error";
378 	}
379 }
380 
mlx5_loglevel_str(int level)381 static const char *mlx5_loglevel_str(int level)
382 {
383 	switch (level) {
384 	case LOGLEVEL_EMERG:
385 		return "EMERGENCY";
386 	case LOGLEVEL_ALERT:
387 		return "ALERT";
388 	case LOGLEVEL_CRIT:
389 		return "CRITICAL";
390 	case LOGLEVEL_ERR:
391 		return "ERROR";
392 	case LOGLEVEL_WARNING:
393 		return "WARNING";
394 	case LOGLEVEL_NOTICE:
395 		return "NOTICE";
396 	case LOGLEVEL_INFO:
397 		return "INFO";
398 	case LOGLEVEL_DEBUG:
399 		return "DEBUG";
400 	}
401 	return "Unknown log level";
402 }
403 
mlx5_health_get_severity(u8 rfr_severity)404 static int mlx5_health_get_severity(u8 rfr_severity)
405 {
406 	return rfr_severity & MLX5_SEVERITY_VALID_MASK ?
407 	       rfr_severity & MLX5_SEVERITY_MASK : LOGLEVEL_ERR;
408 }
409 
print_health_info(struct mlx5_core_dev * dev)410 static void print_health_info(struct mlx5_core_dev *dev)
411 {
412 	struct mlx5_core_health *health = &dev->priv.health;
413 	struct health_buffer __iomem *h = health->health;
414 	u8 rfr_severity;
415 	int severity;
416 	int i;
417 
418 	/* If the syndrome is 0, the device is OK and no need to print buffer */
419 	if (!ioread8(&h->synd))
420 		return;
421 
422 	if (ioread32be(&h->fw_ver) == 0xFFFFFFFF) {
423 		mlx5_log(dev, LOGLEVEL_ERR, "PCI slot is unavailable\n");
424 		return;
425 	}
426 
427 	rfr_severity = ioread8(&h->rfr_severity);
428 	severity  = mlx5_health_get_severity(rfr_severity);
429 	mlx5_log(dev, severity, "Health issue observed, %s, severity(%d) %s:\n",
430 		 hsynd_str(ioread8(&h->synd)), severity, mlx5_loglevel_str(severity));
431 
432 	for (i = 0; i < ARRAY_SIZE(h->assert_var); i++)
433 		mlx5_log(dev, severity, "assert_var[%d] 0x%08x\n", i,
434 			 ioread32be(h->assert_var + i));
435 
436 	mlx5_log(dev, severity, "assert_exit_ptr 0x%08x\n", ioread32be(&h->assert_exit_ptr));
437 	mlx5_log(dev, severity, "assert_callra 0x%08x\n", ioread32be(&h->assert_callra));
438 	mlx5_log(dev, severity, "fw_ver %d.%d.%d", fw_rev_maj(dev), fw_rev_min(dev),
439 		 fw_rev_sub(dev));
440 	mlx5_log(dev, severity, "time %u\n", ioread32be(&h->time));
441 	mlx5_log(dev, severity, "hw_id 0x%08x\n", ioread32be(&h->hw_id));
442 	mlx5_log(dev, severity, "rfr %d\n", mlx5_health_get_rfr(rfr_severity));
443 	mlx5_log(dev, severity, "severity %d (%s)\n", severity, mlx5_loglevel_str(severity));
444 	mlx5_log(dev, severity, "irisc_index %d\n", ioread8(&h->irisc_index));
445 	mlx5_log(dev, severity, "synd 0x%x: %s\n", ioread8(&h->synd),
446 		 hsynd_str(ioread8(&h->synd)));
447 	mlx5_log(dev, severity, "ext_synd 0x%04x\n", ioread16be(&h->ext_synd));
448 	mlx5_log(dev, severity, "raw fw_ver 0x%08x\n", ioread32be(&h->fw_ver));
449 }
450 
451 static int
mlx5_fw_reporter_diagnose(struct devlink_health_reporter * reporter,struct devlink_fmsg * fmsg,struct netlink_ext_ack * extack)452 mlx5_fw_reporter_diagnose(struct devlink_health_reporter *reporter,
453 			  struct devlink_fmsg *fmsg,
454 			  struct netlink_ext_ack *extack)
455 {
456 	struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
457 	struct mlx5_core_health *health = &dev->priv.health;
458 	struct health_buffer __iomem *h = health->health;
459 	u8 synd;
460 	int err;
461 
462 	synd = ioread8(&h->synd);
463 	err = devlink_fmsg_u8_pair_put(fmsg, "Syndrome", synd);
464 	if (err || !synd)
465 		return err;
466 	return devlink_fmsg_string_pair_put(fmsg, "Description", hsynd_str(synd));
467 }
468 
469 struct mlx5_fw_reporter_ctx {
470 	u8 err_synd;
471 	int miss_counter;
472 };
473 
474 static int
mlx5_fw_reporter_ctx_pairs_put(struct devlink_fmsg * fmsg,struct mlx5_fw_reporter_ctx * fw_reporter_ctx)475 mlx5_fw_reporter_ctx_pairs_put(struct devlink_fmsg *fmsg,
476 			       struct mlx5_fw_reporter_ctx *fw_reporter_ctx)
477 {
478 	int err;
479 
480 	err = devlink_fmsg_u8_pair_put(fmsg, "syndrome",
481 				       fw_reporter_ctx->err_synd);
482 	if (err)
483 		return err;
484 	err = devlink_fmsg_u32_pair_put(fmsg, "fw_miss_counter",
485 					fw_reporter_ctx->miss_counter);
486 	if (err)
487 		return err;
488 	return 0;
489 }
490 
491 static int
mlx5_fw_reporter_heath_buffer_data_put(struct mlx5_core_dev * dev,struct devlink_fmsg * fmsg)492 mlx5_fw_reporter_heath_buffer_data_put(struct mlx5_core_dev *dev,
493 				       struct devlink_fmsg *fmsg)
494 {
495 	struct mlx5_core_health *health = &dev->priv.health;
496 	struct health_buffer __iomem *h = health->health;
497 	u8 rfr_severity;
498 	int err;
499 	int i;
500 
501 	if (!ioread8(&h->synd))
502 		return 0;
503 
504 	err = devlink_fmsg_pair_nest_start(fmsg, "health buffer");
505 	if (err)
506 		return err;
507 	err = devlink_fmsg_obj_nest_start(fmsg);
508 	if (err)
509 		return err;
510 	err = devlink_fmsg_arr_pair_nest_start(fmsg, "assert_var");
511 	if (err)
512 		return err;
513 
514 	for (i = 0; i < ARRAY_SIZE(h->assert_var); i++) {
515 		err = devlink_fmsg_u32_put(fmsg, ioread32be(h->assert_var + i));
516 		if (err)
517 			return err;
518 	}
519 	err = devlink_fmsg_arr_pair_nest_end(fmsg);
520 	if (err)
521 		return err;
522 	err = devlink_fmsg_u32_pair_put(fmsg, "assert_exit_ptr",
523 					ioread32be(&h->assert_exit_ptr));
524 	if (err)
525 		return err;
526 	err = devlink_fmsg_u32_pair_put(fmsg, "assert_callra",
527 					ioread32be(&h->assert_callra));
528 	if (err)
529 		return err;
530 	err = devlink_fmsg_u32_pair_put(fmsg, "time", ioread32be(&h->time));
531 	if (err)
532 		return err;
533 	err = devlink_fmsg_u32_pair_put(fmsg, "hw_id", ioread32be(&h->hw_id));
534 	if (err)
535 		return err;
536 	rfr_severity = ioread8(&h->rfr_severity);
537 	err = devlink_fmsg_u8_pair_put(fmsg, "rfr", mlx5_health_get_rfr(rfr_severity));
538 	if (err)
539 		return err;
540 	err = devlink_fmsg_u8_pair_put(fmsg, "severity", mlx5_health_get_severity(rfr_severity));
541 	if (err)
542 		return err;
543 	err = devlink_fmsg_u8_pair_put(fmsg, "irisc_index",
544 				       ioread8(&h->irisc_index));
545 	if (err)
546 		return err;
547 	err = devlink_fmsg_u8_pair_put(fmsg, "synd", ioread8(&h->synd));
548 	if (err)
549 		return err;
550 	err = devlink_fmsg_u32_pair_put(fmsg, "ext_synd",
551 					ioread16be(&h->ext_synd));
552 	if (err)
553 		return err;
554 	err = devlink_fmsg_u32_pair_put(fmsg, "raw_fw_ver",
555 					ioread32be(&h->fw_ver));
556 	if (err)
557 		return err;
558 	err = devlink_fmsg_obj_nest_end(fmsg);
559 	if (err)
560 		return err;
561 	return devlink_fmsg_pair_nest_end(fmsg);
562 }
563 
564 static int
mlx5_fw_reporter_dump(struct devlink_health_reporter * reporter,struct devlink_fmsg * fmsg,void * priv_ctx,struct netlink_ext_ack * extack)565 mlx5_fw_reporter_dump(struct devlink_health_reporter *reporter,
566 		      struct devlink_fmsg *fmsg, void *priv_ctx,
567 		      struct netlink_ext_ack *extack)
568 {
569 	struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
570 	int err;
571 
572 	err = mlx5_fw_tracer_trigger_core_dump_general(dev);
573 	if (err)
574 		return err;
575 
576 	if (priv_ctx) {
577 		struct mlx5_fw_reporter_ctx *fw_reporter_ctx = priv_ctx;
578 
579 		err = mlx5_fw_reporter_ctx_pairs_put(fmsg, fw_reporter_ctx);
580 		if (err)
581 			return err;
582 	}
583 
584 	err = mlx5_fw_reporter_heath_buffer_data_put(dev, fmsg);
585 	if (err)
586 		return err;
587 	return mlx5_fw_tracer_get_saved_traces_objects(dev->tracer, fmsg);
588 }
589 
mlx5_fw_reporter_err_work(struct work_struct * work)590 static void mlx5_fw_reporter_err_work(struct work_struct *work)
591 {
592 	struct mlx5_fw_reporter_ctx fw_reporter_ctx;
593 	struct mlx5_core_health *health;
594 
595 	health = container_of(work, struct mlx5_core_health, report_work);
596 
597 	if (IS_ERR_OR_NULL(health->fw_reporter))
598 		return;
599 
600 	fw_reporter_ctx.err_synd = health->synd;
601 	fw_reporter_ctx.miss_counter = health->miss_counter;
602 	if (fw_reporter_ctx.err_synd) {
603 		devlink_health_report(health->fw_reporter,
604 				      "FW syndrome reported", &fw_reporter_ctx);
605 		return;
606 	}
607 	if (fw_reporter_ctx.miss_counter)
608 		devlink_health_report(health->fw_reporter,
609 				      "FW miss counter reported",
610 				      &fw_reporter_ctx);
611 }
612 
613 static const struct devlink_health_reporter_ops mlx5_fw_reporter_ops = {
614 		.name = "fw",
615 		.diagnose = mlx5_fw_reporter_diagnose,
616 		.dump = mlx5_fw_reporter_dump,
617 };
618 
619 static int
mlx5_fw_fatal_reporter_recover(struct devlink_health_reporter * reporter,void * priv_ctx,struct netlink_ext_ack * extack)620 mlx5_fw_fatal_reporter_recover(struct devlink_health_reporter *reporter,
621 			       void *priv_ctx,
622 			       struct netlink_ext_ack *extack)
623 {
624 	struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
625 
626 	return mlx5_health_try_recover(dev);
627 }
628 
629 static int
mlx5_fw_fatal_reporter_dump(struct devlink_health_reporter * reporter,struct devlink_fmsg * fmsg,void * priv_ctx,struct netlink_ext_ack * extack)630 mlx5_fw_fatal_reporter_dump(struct devlink_health_reporter *reporter,
631 			    struct devlink_fmsg *fmsg, void *priv_ctx,
632 			    struct netlink_ext_ack *extack)
633 {
634 	struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
635 	u32 crdump_size = dev->priv.health.crdump_size;
636 	u32 *cr_data;
637 	int err;
638 
639 	if (!mlx5_core_is_pf(dev))
640 		return -EPERM;
641 
642 	cr_data = kvmalloc(crdump_size, GFP_KERNEL);
643 	if (!cr_data)
644 		return -ENOMEM;
645 	err = mlx5_crdump_collect(dev, cr_data);
646 	if (err)
647 		goto free_data;
648 
649 	if (priv_ctx) {
650 		struct mlx5_fw_reporter_ctx *fw_reporter_ctx = priv_ctx;
651 
652 		err = mlx5_fw_reporter_ctx_pairs_put(fmsg, fw_reporter_ctx);
653 		if (err)
654 			goto free_data;
655 	}
656 
657 	err = devlink_fmsg_binary_pair_put(fmsg, "crdump_data", cr_data, crdump_size);
658 
659 free_data:
660 	kvfree(cr_data);
661 	return err;
662 }
663 
mlx5_fw_fatal_reporter_err_work(struct work_struct * work)664 static void mlx5_fw_fatal_reporter_err_work(struct work_struct *work)
665 {
666 	struct mlx5_fw_reporter_ctx fw_reporter_ctx;
667 	struct mlx5_core_health *health;
668 	struct mlx5_core_dev *dev;
669 	struct devlink *devlink;
670 	struct mlx5_priv *priv;
671 
672 	health = container_of(work, struct mlx5_core_health, fatal_report_work);
673 	priv = container_of(health, struct mlx5_priv, health);
674 	dev = container_of(priv, struct mlx5_core_dev, priv);
675 	devlink = priv_to_devlink(dev);
676 
677 	mutex_lock(&dev->intf_state_mutex);
678 	if (test_bit(MLX5_DROP_HEALTH_WORK, &health->flags)) {
679 		mlx5_core_err(dev, "health works are not permitted at this stage\n");
680 		mutex_unlock(&dev->intf_state_mutex);
681 		return;
682 	}
683 	mutex_unlock(&dev->intf_state_mutex);
684 	enter_error_state(dev, false);
685 	if (IS_ERR_OR_NULL(health->fw_fatal_reporter)) {
686 		devl_lock(devlink);
687 		if (mlx5_health_try_recover(dev))
688 			mlx5_core_err(dev, "health recovery failed\n");
689 		devl_unlock(devlink);
690 		return;
691 	}
692 	fw_reporter_ctx.err_synd = health->synd;
693 	fw_reporter_ctx.miss_counter = health->miss_counter;
694 	if (devlink_health_report(health->fw_fatal_reporter,
695 				  "FW fatal error reported", &fw_reporter_ctx) == -ECANCELED) {
696 		/* If recovery wasn't performed, due to grace period,
697 		 * unload the driver. This ensures that the driver
698 		 * closes all its resources and it is not subjected to
699 		 * requests from the kernel.
700 		 */
701 		mlx5_core_err(dev, "Driver is in error state. Unloading\n");
702 		mlx5_unload_one(dev, false);
703 	}
704 }
705 
706 static const struct devlink_health_reporter_ops mlx5_fw_fatal_reporter_ops = {
707 		.name = "fw_fatal",
708 		.recover = mlx5_fw_fatal_reporter_recover,
709 		.dump = mlx5_fw_fatal_reporter_dump,
710 };
711 
712 #define MLX5_FW_REPORTER_ECPF_GRACEFUL_PERIOD 180000
713 #define MLX5_FW_REPORTER_PF_GRACEFUL_PERIOD 60000
714 #define MLX5_FW_REPORTER_VF_GRACEFUL_PERIOD 30000
715 #define MLX5_FW_REPORTER_DEFAULT_GRACEFUL_PERIOD MLX5_FW_REPORTER_VF_GRACEFUL_PERIOD
716 
mlx5_fw_reporters_create(struct mlx5_core_dev * dev)717 void mlx5_fw_reporters_create(struct mlx5_core_dev *dev)
718 {
719 	struct mlx5_core_health *health = &dev->priv.health;
720 	struct devlink *devlink = priv_to_devlink(dev);
721 	u64 grace_period;
722 
723 	if (mlx5_core_is_ecpf(dev)) {
724 		grace_period = MLX5_FW_REPORTER_ECPF_GRACEFUL_PERIOD;
725 	} else if (mlx5_core_is_pf(dev)) {
726 		grace_period = MLX5_FW_REPORTER_PF_GRACEFUL_PERIOD;
727 	} else {
728 		/* VF or SF */
729 		grace_period = MLX5_FW_REPORTER_DEFAULT_GRACEFUL_PERIOD;
730 	}
731 
732 	health->fw_reporter =
733 		devl_health_reporter_create(devlink, &mlx5_fw_reporter_ops,
734 					    0, dev);
735 	if (IS_ERR(health->fw_reporter))
736 		mlx5_core_warn(dev, "Failed to create fw reporter, err = %ld\n",
737 			       PTR_ERR(health->fw_reporter));
738 
739 	health->fw_fatal_reporter =
740 		devl_health_reporter_create(devlink,
741 					    &mlx5_fw_fatal_reporter_ops,
742 					    grace_period,
743 					    dev);
744 	if (IS_ERR(health->fw_fatal_reporter))
745 		mlx5_core_warn(dev, "Failed to create fw fatal reporter, err = %ld\n",
746 			       PTR_ERR(health->fw_fatal_reporter));
747 }
748 
mlx5_fw_reporters_destroy(struct mlx5_core_dev * dev)749 static void mlx5_fw_reporters_destroy(struct mlx5_core_dev *dev)
750 {
751 	struct mlx5_core_health *health = &dev->priv.health;
752 
753 	if (!IS_ERR_OR_NULL(health->fw_reporter))
754 		devlink_health_reporter_destroy(health->fw_reporter);
755 
756 	if (!IS_ERR_OR_NULL(health->fw_fatal_reporter))
757 		devlink_health_reporter_destroy(health->fw_fatal_reporter);
758 }
759 
get_next_poll_jiffies(struct mlx5_core_dev * dev)760 static unsigned long get_next_poll_jiffies(struct mlx5_core_dev *dev)
761 {
762 	unsigned long next;
763 
764 	get_random_bytes(&next, sizeof(next));
765 	next %= HZ;
766 	next += jiffies + msecs_to_jiffies(mlx5_tout_ms(dev, HEALTH_POLL_INTERVAL));
767 
768 	return next;
769 }
770 
mlx5_trigger_health_work(struct mlx5_core_dev * dev)771 void mlx5_trigger_health_work(struct mlx5_core_dev *dev)
772 {
773 	struct mlx5_core_health *health = &dev->priv.health;
774 
775 	if (!mlx5_dev_is_lightweight(dev))
776 		queue_work(health->wq, &health->fatal_report_work);
777 }
778 
779 #define MLX5_MSEC_PER_HOUR (MSEC_PER_SEC * 60 * 60)
mlx5_health_log_ts_update(struct work_struct * work)780 static void mlx5_health_log_ts_update(struct work_struct *work)
781 {
782 	struct delayed_work *dwork = to_delayed_work(work);
783 	u32 out[MLX5_ST_SZ_DW(mrtc_reg)] = {};
784 	u32 in[MLX5_ST_SZ_DW(mrtc_reg)] = {};
785 	struct mlx5_core_health *health;
786 	struct mlx5_core_dev *dev;
787 	struct mlx5_priv *priv;
788 	u64 now_us;
789 
790 	health = container_of(dwork, struct mlx5_core_health, update_fw_log_ts_work);
791 	priv = container_of(health, struct mlx5_priv, health);
792 	dev = container_of(priv, struct mlx5_core_dev, priv);
793 
794 	now_us =  ktime_to_us(ktime_get_real());
795 
796 	MLX5_SET(mrtc_reg, in, time_h, now_us >> 32);
797 	MLX5_SET(mrtc_reg, in, time_l, now_us & 0xFFFFFFFF);
798 	mlx5_core_access_reg(dev, in, sizeof(in), out, sizeof(out), MLX5_REG_MRTC, 0, 1);
799 
800 	queue_delayed_work(health->wq, &health->update_fw_log_ts_work,
801 			   msecs_to_jiffies(MLX5_MSEC_PER_HOUR));
802 }
803 
poll_health(struct timer_list * t)804 static void poll_health(struct timer_list *t)
805 {
806 	struct mlx5_core_dev *dev = from_timer(dev, t, priv.health.timer);
807 	struct mlx5_core_health *health = &dev->priv.health;
808 	struct health_buffer __iomem *h = health->health;
809 	u32 fatal_error;
810 	u8 prev_synd;
811 	u32 count;
812 
813 	if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)
814 		goto out;
815 
816 	fatal_error = mlx5_health_check_fatal_sensors(dev);
817 
818 	if (fatal_error && !health->fatal_error) {
819 		mlx5_core_err(dev, "Fatal error %u detected\n", fatal_error);
820 		dev->priv.health.fatal_error = fatal_error;
821 		print_health_info(dev);
822 		dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
823 		mlx5_trigger_health_work(dev);
824 		return;
825 	}
826 
827 	count = ioread32be(health->health_counter);
828 	if (count == health->prev)
829 		++health->miss_counter;
830 	else
831 		health->miss_counter = 0;
832 
833 	health->prev = count;
834 	if (health->miss_counter == MAX_MISSES) {
835 		mlx5_core_err(dev, "device's health compromised - reached miss count\n");
836 		print_health_info(dev);
837 		queue_work(health->wq, &health->report_work);
838 	}
839 
840 	prev_synd = health->synd;
841 	health->synd = ioread8(&h->synd);
842 	if (health->synd && health->synd != prev_synd)
843 		queue_work(health->wq, &health->report_work);
844 
845 out:
846 	mod_timer(&health->timer, get_next_poll_jiffies(dev));
847 }
848 
mlx5_start_health_poll(struct mlx5_core_dev * dev)849 void mlx5_start_health_poll(struct mlx5_core_dev *dev)
850 {
851 	u64 poll_interval_ms =  mlx5_tout_ms(dev, HEALTH_POLL_INTERVAL);
852 	struct mlx5_core_health *health = &dev->priv.health;
853 
854 	timer_setup(&health->timer, poll_health, 0);
855 	health->fatal_error = MLX5_SENSOR_NO_ERR;
856 	clear_bit(MLX5_DROP_HEALTH_WORK, &health->flags);
857 	health->health = &dev->iseg->health;
858 	health->health_counter = &dev->iseg->health_counter;
859 
860 	health->timer.expires = jiffies + msecs_to_jiffies(poll_interval_ms);
861 	add_timer(&health->timer);
862 }
863 
mlx5_stop_health_poll(struct mlx5_core_dev * dev,bool disable_health)864 void mlx5_stop_health_poll(struct mlx5_core_dev *dev, bool disable_health)
865 {
866 	struct mlx5_core_health *health = &dev->priv.health;
867 
868 	if (disable_health)
869 		set_bit(MLX5_DROP_HEALTH_WORK, &health->flags);
870 
871 	del_timer_sync(&health->timer);
872 }
873 
mlx5_start_health_fw_log_up(struct mlx5_core_dev * dev)874 void mlx5_start_health_fw_log_up(struct mlx5_core_dev *dev)
875 {
876 	struct mlx5_core_health *health = &dev->priv.health;
877 
878 	if (mlx5_core_is_pf(dev) && MLX5_CAP_MCAM_REG(dev, mrtc))
879 		queue_delayed_work(health->wq, &health->update_fw_log_ts_work, 0);
880 }
881 
mlx5_drain_health_wq(struct mlx5_core_dev * dev)882 void mlx5_drain_health_wq(struct mlx5_core_dev *dev)
883 {
884 	struct mlx5_core_health *health = &dev->priv.health;
885 
886 	set_bit(MLX5_DROP_HEALTH_WORK, &health->flags);
887 	cancel_delayed_work_sync(&health->update_fw_log_ts_work);
888 	cancel_work_sync(&health->report_work);
889 	cancel_work_sync(&health->fatal_report_work);
890 }
891 
mlx5_health_cleanup(struct mlx5_core_dev * dev)892 void mlx5_health_cleanup(struct mlx5_core_dev *dev)
893 {
894 	struct mlx5_core_health *health = &dev->priv.health;
895 
896 	cancel_delayed_work_sync(&health->update_fw_log_ts_work);
897 	destroy_workqueue(health->wq);
898 	mlx5_reporter_vnic_destroy(dev);
899 	mlx5_fw_reporters_destroy(dev);
900 }
901 
mlx5_health_init(struct mlx5_core_dev * dev)902 int mlx5_health_init(struct mlx5_core_dev *dev)
903 {
904 	struct devlink *devlink = priv_to_devlink(dev);
905 	struct mlx5_core_health *health;
906 	char *name;
907 
908 	if (!mlx5_dev_is_lightweight(dev)) {
909 		devl_lock(devlink);
910 		mlx5_fw_reporters_create(dev);
911 		devl_unlock(devlink);
912 	}
913 	mlx5_reporter_vnic_create(dev);
914 
915 	health = &dev->priv.health;
916 	name = kmalloc(64, GFP_KERNEL);
917 	if (!name)
918 		goto out_err;
919 
920 	strcpy(name, "mlx5_health");
921 	strcat(name, dev_name(dev->device));
922 	health->wq = create_singlethread_workqueue(name);
923 	kfree(name);
924 	if (!health->wq)
925 		goto out_err;
926 	INIT_WORK(&health->fatal_report_work, mlx5_fw_fatal_reporter_err_work);
927 	INIT_WORK(&health->report_work, mlx5_fw_reporter_err_work);
928 	INIT_DELAYED_WORK(&health->update_fw_log_ts_work, mlx5_health_log_ts_update);
929 
930 	return 0;
931 
932 out_err:
933 	mlx5_reporter_vnic_destroy(dev);
934 	mlx5_fw_reporters_destroy(dev);
935 	return -ENOMEM;
936 }
937