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 
252 		msleep(20);
253 	} while (!time_after(jiffies, end));
254 
255 	if (mlx5_get_nic_state(dev) != MLX5_NIC_IFC_DISABLED) {
256 		dev_err(&dev->pdev->dev, "NIC IFC still %d after %lums.\n",
257 			mlx5_get_nic_state(dev), delay_ms);
258 	}
259 
260 	/* Release FW semaphore if you are the lock owner */
261 	if (!lock)
262 		lock_sem_sw_reset(dev, false);
263 
264 	mlx5_core_err(dev, "end\n");
265 
266 unlock:
267 	mutex_unlock(&dev->intf_state_mutex);
268 }
269 
mlx5_handle_bad_state(struct mlx5_core_dev * dev)270 static void mlx5_handle_bad_state(struct mlx5_core_dev *dev)
271 {
272 	u8 nic_interface = mlx5_get_nic_state(dev);
273 
274 	switch (nic_interface) {
275 	case MLX5_NIC_IFC_FULL:
276 		mlx5_core_warn(dev, "Expected to see disabled NIC but it is full driver\n");
277 		break;
278 
279 	case MLX5_NIC_IFC_DISABLED:
280 		mlx5_core_warn(dev, "starting teardown\n");
281 		break;
282 
283 	case MLX5_NIC_IFC_NO_DRAM_NIC:
284 		mlx5_core_warn(dev, "Expected to see disabled NIC but it is no dram nic\n");
285 		break;
286 
287 	case MLX5_NIC_IFC_SW_RESET:
288 		/* The IFC mode field is 3 bits, so it will read 0x7 in 2 cases:
289 		 * 1. PCI has been disabled (ie. PCI-AER, PF driver unloaded
290 		 *    and this is a VF), this is not recoverable by SW reset.
291 		 *    Logging of this is handled elsewhere.
292 		 * 2. FW reset has been issued by another function, driver can
293 		 *    be reloaded to recover after the mode switches to
294 		 *    MLX5_NIC_IFC_DISABLED.
295 		 */
296 		if (dev->priv.health.fatal_error != MLX5_SENSOR_PCI_COMM_ERR)
297 			mlx5_core_warn(dev, "NIC SW reset in progress\n");
298 		break;
299 
300 	default:
301 		mlx5_core_warn(dev, "Expected to see disabled NIC but it is has invalid value %d\n",
302 			       nic_interface);
303 	}
304 
305 	mlx5_disable_device(dev);
306 }
307 
mlx5_health_wait_pci_up(struct mlx5_core_dev * dev)308 int mlx5_health_wait_pci_up(struct mlx5_core_dev *dev)
309 {
310 	unsigned long end;
311 
312 	end = jiffies + msecs_to_jiffies(mlx5_tout_ms(dev, FW_RESET));
313 	while (sensor_pci_not_working(dev)) {
314 		if (time_after(jiffies, end))
315 			return -ETIMEDOUT;
316 		if (test_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state)) {
317 			mlx5_core_warn(dev, "device is being removed, stop waiting for PCI\n");
318 			return -ENODEV;
319 		}
320 		msleep(100);
321 	}
322 	return 0;
323 }
324 
mlx5_health_try_recover(struct mlx5_core_dev * dev)325 static int mlx5_health_try_recover(struct mlx5_core_dev *dev)
326 {
327 	mlx5_core_warn(dev, "handling bad device here\n");
328 	mlx5_handle_bad_state(dev);
329 	if (mlx5_health_wait_pci_up(dev)) {
330 		mlx5_core_err(dev, "health recovery flow aborted, PCI reads still not working\n");
331 		return -EIO;
332 	}
333 	mlx5_core_err(dev, "starting health recovery flow\n");
334 	if (mlx5_recover_device(dev) || mlx5_health_check_fatal_sensors(dev)) {
335 		mlx5_core_err(dev, "health recovery failed\n");
336 		return -EIO;
337 	}
338 
339 	mlx5_core_info(dev, "health recovery succeeded\n");
340 	return 0;
341 }
342 
hsynd_str(u8 synd)343 static const char *hsynd_str(u8 synd)
344 {
345 	switch (synd) {
346 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_FW_INTERNAL_ERR:
347 		return "firmware internal error";
348 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_DEAD_IRISC:
349 		return "irisc not responding";
350 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_HW_FATAL_ERR:
351 		return "unrecoverable hardware error";
352 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_FW_CRC_ERR:
353 		return "firmware CRC error";
354 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_ICM_FETCH_PCI_ERR:
355 		return "ICM fetch PCI error";
356 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_ICM_PAGE_ERR:
357 		return "HW fatal error\n";
358 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_ASYNCHRONOUS_EQ_BUF_OVERRUN:
359 		return "async EQ buffer overrun";
360 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_EQ_IN_ERR:
361 		return "EQ error";
362 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_EQ_INV:
363 		return "Invalid EQ referenced";
364 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_FFSER_ERR:
365 		return "FFSER error";
366 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_HIGH_TEMP_ERR:
367 		return "High temperature";
368 	case MLX5_INITIAL_SEG_HEALTH_SYNDROME_ICM_PCI_POISONED_ERR:
369 		return "ICM fetch PCI data poisoned error";
370 	default:
371 		return "unrecognized error";
372 	}
373 }
374 
mlx5_loglevel_str(int level)375 static const char *mlx5_loglevel_str(int level)
376 {
377 	switch (level) {
378 	case LOGLEVEL_EMERG:
379 		return "EMERGENCY";
380 	case LOGLEVEL_ALERT:
381 		return "ALERT";
382 	case LOGLEVEL_CRIT:
383 		return "CRITICAL";
384 	case LOGLEVEL_ERR:
385 		return "ERROR";
386 	case LOGLEVEL_WARNING:
387 		return "WARNING";
388 	case LOGLEVEL_NOTICE:
389 		return "NOTICE";
390 	case LOGLEVEL_INFO:
391 		return "INFO";
392 	case LOGLEVEL_DEBUG:
393 		return "DEBUG";
394 	}
395 	return "Unknown log level";
396 }
397 
mlx5_health_get_severity(u8 rfr_severity)398 static int mlx5_health_get_severity(u8 rfr_severity)
399 {
400 	return rfr_severity & MLX5_SEVERITY_VALID_MASK ?
401 	       rfr_severity & MLX5_SEVERITY_MASK : LOGLEVEL_ERR;
402 }
403 
print_health_info(struct mlx5_core_dev * dev)404 static void print_health_info(struct mlx5_core_dev *dev)
405 {
406 	struct mlx5_core_health *health = &dev->priv.health;
407 	struct health_buffer __iomem *h = health->health;
408 	u8 rfr_severity;
409 	int severity;
410 	int i;
411 
412 	/* If the syndrome is 0, the device is OK and no need to print buffer */
413 	if (!ioread8(&h->synd))
414 		return;
415 
416 	if (ioread32be(&h->fw_ver) == 0xFFFFFFFF) {
417 		mlx5_log(dev, LOGLEVEL_ERR, "PCI slot is unavailable\n");
418 		return;
419 	}
420 
421 	rfr_severity = ioread8(&h->rfr_severity);
422 	severity  = mlx5_health_get_severity(rfr_severity);
423 	mlx5_log(dev, severity, "Health issue observed, %s, severity(%d) %s:\n",
424 		 hsynd_str(ioread8(&h->synd)), severity, mlx5_loglevel_str(severity));
425 
426 	for (i = 0; i < ARRAY_SIZE(h->assert_var); i++)
427 		mlx5_log(dev, severity, "assert_var[%d] 0x%08x\n", i,
428 			 ioread32be(h->assert_var + i));
429 
430 	mlx5_log(dev, severity, "assert_exit_ptr 0x%08x\n", ioread32be(&h->assert_exit_ptr));
431 	mlx5_log(dev, severity, "assert_callra 0x%08x\n", ioread32be(&h->assert_callra));
432 	mlx5_log(dev, severity, "fw_ver %d.%d.%d", fw_rev_maj(dev), fw_rev_min(dev),
433 		 fw_rev_sub(dev));
434 	mlx5_log(dev, severity, "time %u\n", ioread32be(&h->time));
435 	mlx5_log(dev, severity, "hw_id 0x%08x\n", ioread32be(&h->hw_id));
436 	mlx5_log(dev, severity, "rfr %d\n", mlx5_health_get_rfr(rfr_severity));
437 	mlx5_log(dev, severity, "severity %d (%s)\n", severity, mlx5_loglevel_str(severity));
438 	mlx5_log(dev, severity, "irisc_index %d\n", ioread8(&h->irisc_index));
439 	mlx5_log(dev, severity, "synd 0x%x: %s\n", ioread8(&h->synd),
440 		 hsynd_str(ioread8(&h->synd)));
441 	mlx5_log(dev, severity, "ext_synd 0x%04x\n", ioread16be(&h->ext_synd));
442 	mlx5_log(dev, severity, "raw fw_ver 0x%08x\n", ioread32be(&h->fw_ver));
443 }
444 
445 static int
mlx5_fw_reporter_diagnose(struct devlink_health_reporter * reporter,struct devlink_fmsg * fmsg,struct netlink_ext_ack * extack)446 mlx5_fw_reporter_diagnose(struct devlink_health_reporter *reporter,
447 			  struct devlink_fmsg *fmsg,
448 			  struct netlink_ext_ack *extack)
449 {
450 	struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
451 	struct mlx5_core_health *health = &dev->priv.health;
452 	struct health_buffer __iomem *h = health->health;
453 	u8 synd = ioread8(&h->synd);
454 
455 	devlink_fmsg_u8_pair_put(fmsg, "Syndrome", synd);
456 	if (!synd)
457 		return 0;
458 
459 	devlink_fmsg_string_pair_put(fmsg, "Description", hsynd_str(synd));
460 
461 	return 0;
462 }
463 
464 struct mlx5_fw_reporter_ctx {
465 	u8 err_synd;
466 	int miss_counter;
467 };
468 
469 static void
mlx5_fw_reporter_ctx_pairs_put(struct devlink_fmsg * fmsg,struct mlx5_fw_reporter_ctx * fw_reporter_ctx)470 mlx5_fw_reporter_ctx_pairs_put(struct devlink_fmsg *fmsg,
471 			       struct mlx5_fw_reporter_ctx *fw_reporter_ctx)
472 {
473 	devlink_fmsg_u8_pair_put(fmsg, "syndrome", fw_reporter_ctx->err_synd);
474 	devlink_fmsg_u32_pair_put(fmsg, "fw_miss_counter", fw_reporter_ctx->miss_counter);
475 }
476 
477 static void
mlx5_fw_reporter_heath_buffer_data_put(struct mlx5_core_dev * dev,struct devlink_fmsg * fmsg)478 mlx5_fw_reporter_heath_buffer_data_put(struct mlx5_core_dev *dev,
479 				       struct devlink_fmsg *fmsg)
480 {
481 	struct mlx5_core_health *health = &dev->priv.health;
482 	struct health_buffer __iomem *h = health->health;
483 	u8 rfr_severity;
484 	int i;
485 
486 	if (!ioread8(&h->synd))
487 		return;
488 
489 	devlink_fmsg_pair_nest_start(fmsg, "health buffer");
490 	devlink_fmsg_obj_nest_start(fmsg);
491 	devlink_fmsg_arr_pair_nest_start(fmsg, "assert_var");
492 	for (i = 0; i < ARRAY_SIZE(h->assert_var); i++)
493 		devlink_fmsg_u32_put(fmsg, ioread32be(h->assert_var + i));
494 	devlink_fmsg_arr_pair_nest_end(fmsg);
495 	devlink_fmsg_u32_pair_put(fmsg, "assert_exit_ptr",
496 				  ioread32be(&h->assert_exit_ptr));
497 	devlink_fmsg_u32_pair_put(fmsg, "assert_callra",
498 				  ioread32be(&h->assert_callra));
499 	devlink_fmsg_u32_pair_put(fmsg, "time", ioread32be(&h->time));
500 	devlink_fmsg_u32_pair_put(fmsg, "hw_id", ioread32be(&h->hw_id));
501 	rfr_severity = ioread8(&h->rfr_severity);
502 	devlink_fmsg_u8_pair_put(fmsg, "rfr", mlx5_health_get_rfr(rfr_severity));
503 	devlink_fmsg_u8_pair_put(fmsg, "severity", mlx5_health_get_severity(rfr_severity));
504 	devlink_fmsg_u8_pair_put(fmsg, "irisc_index", ioread8(&h->irisc_index));
505 	devlink_fmsg_u8_pair_put(fmsg, "synd", ioread8(&h->synd));
506 	devlink_fmsg_u32_pair_put(fmsg, "ext_synd", ioread16be(&h->ext_synd));
507 	devlink_fmsg_u32_pair_put(fmsg, "raw_fw_ver", ioread32be(&h->fw_ver));
508 	devlink_fmsg_obj_nest_end(fmsg);
509 	devlink_fmsg_pair_nest_end(fmsg);
510 }
511 
512 static int
mlx5_fw_reporter_dump(struct devlink_health_reporter * reporter,struct devlink_fmsg * fmsg,void * priv_ctx,struct netlink_ext_ack * extack)513 mlx5_fw_reporter_dump(struct devlink_health_reporter *reporter,
514 		      struct devlink_fmsg *fmsg, void *priv_ctx,
515 		      struct netlink_ext_ack *extack)
516 {
517 	struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
518 	int err;
519 
520 	err = mlx5_fw_tracer_trigger_core_dump_general(dev);
521 	if (err)
522 		return err;
523 
524 	if (priv_ctx) {
525 		struct mlx5_fw_reporter_ctx *fw_reporter_ctx = priv_ctx;
526 
527 		mlx5_fw_reporter_ctx_pairs_put(fmsg, fw_reporter_ctx);
528 	}
529 
530 	mlx5_fw_reporter_heath_buffer_data_put(dev, fmsg);
531 
532 	return mlx5_fw_tracer_get_saved_traces_objects(dev->tracer, fmsg);
533 }
534 
mlx5_fw_reporter_err_work(struct work_struct * work)535 static void mlx5_fw_reporter_err_work(struct work_struct *work)
536 {
537 	struct mlx5_fw_reporter_ctx fw_reporter_ctx;
538 	struct mlx5_core_health *health;
539 
540 	health = container_of(work, struct mlx5_core_health, report_work);
541 
542 	if (IS_ERR_OR_NULL(health->fw_reporter))
543 		return;
544 
545 	fw_reporter_ctx.err_synd = health->synd;
546 	fw_reporter_ctx.miss_counter = health->miss_counter;
547 	if (fw_reporter_ctx.err_synd) {
548 		devlink_health_report(health->fw_reporter,
549 				      "FW syndrome reported", &fw_reporter_ctx);
550 		return;
551 	}
552 	if (fw_reporter_ctx.miss_counter)
553 		devlink_health_report(health->fw_reporter,
554 				      "FW miss counter reported",
555 				      &fw_reporter_ctx);
556 }
557 
558 static const struct devlink_health_reporter_ops mlx5_fw_reporter_ops = {
559 		.name = "fw",
560 		.diagnose = mlx5_fw_reporter_diagnose,
561 		.dump = mlx5_fw_reporter_dump,
562 };
563 
564 static int
mlx5_fw_fatal_reporter_recover(struct devlink_health_reporter * reporter,void * priv_ctx,struct netlink_ext_ack * extack)565 mlx5_fw_fatal_reporter_recover(struct devlink_health_reporter *reporter,
566 			       void *priv_ctx,
567 			       struct netlink_ext_ack *extack)
568 {
569 	struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
570 
571 	return mlx5_health_try_recover(dev);
572 }
573 
574 static int
mlx5_fw_fatal_reporter_dump(struct devlink_health_reporter * reporter,struct devlink_fmsg * fmsg,void * priv_ctx,struct netlink_ext_ack * extack)575 mlx5_fw_fatal_reporter_dump(struct devlink_health_reporter *reporter,
576 			    struct devlink_fmsg *fmsg, void *priv_ctx,
577 			    struct netlink_ext_ack *extack)
578 {
579 	struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
580 	u32 crdump_size = dev->priv.health.crdump_size;
581 	u32 *cr_data;
582 	int err;
583 
584 	if (!mlx5_core_is_pf(dev))
585 		return -EPERM;
586 
587 	cr_data = kvmalloc(crdump_size, GFP_KERNEL);
588 	if (!cr_data)
589 		return -ENOMEM;
590 	err = mlx5_crdump_collect(dev, cr_data);
591 	if (err)
592 		goto free_data;
593 
594 	if (priv_ctx) {
595 		struct mlx5_fw_reporter_ctx *fw_reporter_ctx = priv_ctx;
596 
597 		mlx5_fw_reporter_ctx_pairs_put(fmsg, fw_reporter_ctx);
598 	}
599 
600 	devlink_fmsg_binary_pair_put(fmsg, "crdump_data", cr_data, crdump_size);
601 
602 free_data:
603 	kvfree(cr_data);
604 	return err;
605 }
606 
mlx5_fw_fatal_reporter_err_work(struct work_struct * work)607 static void mlx5_fw_fatal_reporter_err_work(struct work_struct *work)
608 {
609 	struct mlx5_fw_reporter_ctx fw_reporter_ctx;
610 	struct mlx5_core_health *health;
611 	struct mlx5_core_dev *dev;
612 	struct devlink *devlink;
613 	struct mlx5_priv *priv;
614 
615 	health = container_of(work, struct mlx5_core_health, fatal_report_work);
616 	priv = container_of(health, struct mlx5_priv, health);
617 	dev = container_of(priv, struct mlx5_core_dev, priv);
618 	devlink = priv_to_devlink(dev);
619 
620 	mutex_lock(&dev->intf_state_mutex);
621 	if (test_bit(MLX5_DROP_HEALTH_WORK, &health->flags)) {
622 		mlx5_core_err(dev, "health works are not permitted at this stage\n");
623 		mutex_unlock(&dev->intf_state_mutex);
624 		return;
625 	}
626 	mutex_unlock(&dev->intf_state_mutex);
627 	enter_error_state(dev, false);
628 	if (IS_ERR_OR_NULL(health->fw_fatal_reporter)) {
629 		devl_lock(devlink);
630 		if (mlx5_health_try_recover(dev))
631 			mlx5_core_err(dev, "health recovery failed\n");
632 		devl_unlock(devlink);
633 		return;
634 	}
635 	fw_reporter_ctx.err_synd = health->synd;
636 	fw_reporter_ctx.miss_counter = health->miss_counter;
637 	if (devlink_health_report(health->fw_fatal_reporter,
638 				  "FW fatal error reported", &fw_reporter_ctx) == -ECANCELED) {
639 		/* If recovery wasn't performed, due to grace period,
640 		 * unload the driver. This ensures that the driver
641 		 * closes all its resources and it is not subjected to
642 		 * requests from the kernel.
643 		 */
644 		mlx5_core_err(dev, "Driver is in error state. Unloading\n");
645 		mlx5_unload_one(dev, false);
646 	}
647 }
648 
649 static const struct devlink_health_reporter_ops mlx5_fw_fatal_reporter_ops = {
650 		.name = "fw_fatal",
651 		.recover = mlx5_fw_fatal_reporter_recover,
652 		.dump = mlx5_fw_fatal_reporter_dump,
653 };
654 
655 #define MLX5_FW_REPORTER_ECPF_GRACEFUL_PERIOD 180000
656 #define MLX5_FW_REPORTER_PF_GRACEFUL_PERIOD 60000
657 #define MLX5_FW_REPORTER_VF_GRACEFUL_PERIOD 30000
658 #define MLX5_FW_REPORTER_DEFAULT_GRACEFUL_PERIOD MLX5_FW_REPORTER_VF_GRACEFUL_PERIOD
659 
mlx5_fw_reporters_create(struct mlx5_core_dev * dev)660 void mlx5_fw_reporters_create(struct mlx5_core_dev *dev)
661 {
662 	struct mlx5_core_health *health = &dev->priv.health;
663 	struct devlink *devlink = priv_to_devlink(dev);
664 	u64 grace_period;
665 
666 	if (mlx5_core_is_ecpf(dev)) {
667 		grace_period = MLX5_FW_REPORTER_ECPF_GRACEFUL_PERIOD;
668 	} else if (mlx5_core_is_pf(dev)) {
669 		grace_period = MLX5_FW_REPORTER_PF_GRACEFUL_PERIOD;
670 	} else {
671 		/* VF or SF */
672 		grace_period = MLX5_FW_REPORTER_DEFAULT_GRACEFUL_PERIOD;
673 	}
674 
675 	health->fw_reporter =
676 		devl_health_reporter_create(devlink, &mlx5_fw_reporter_ops,
677 					    0, dev);
678 	if (IS_ERR(health->fw_reporter))
679 		mlx5_core_warn(dev, "Failed to create fw reporter, err = %ld\n",
680 			       PTR_ERR(health->fw_reporter));
681 
682 	health->fw_fatal_reporter =
683 		devl_health_reporter_create(devlink,
684 					    &mlx5_fw_fatal_reporter_ops,
685 					    grace_period,
686 					    dev);
687 	if (IS_ERR(health->fw_fatal_reporter))
688 		mlx5_core_warn(dev, "Failed to create fw fatal reporter, err = %ld\n",
689 			       PTR_ERR(health->fw_fatal_reporter));
690 }
691 
mlx5_fw_reporters_destroy(struct mlx5_core_dev * dev)692 static void mlx5_fw_reporters_destroy(struct mlx5_core_dev *dev)
693 {
694 	struct mlx5_core_health *health = &dev->priv.health;
695 
696 	if (!IS_ERR_OR_NULL(health->fw_reporter))
697 		devlink_health_reporter_destroy(health->fw_reporter);
698 
699 	if (!IS_ERR_OR_NULL(health->fw_fatal_reporter))
700 		devlink_health_reporter_destroy(health->fw_fatal_reporter);
701 }
702 
get_next_poll_jiffies(struct mlx5_core_dev * dev)703 static unsigned long get_next_poll_jiffies(struct mlx5_core_dev *dev)
704 {
705 	unsigned long next;
706 
707 	get_random_bytes(&next, sizeof(next));
708 	next %= HZ;
709 	next += jiffies + msecs_to_jiffies(mlx5_tout_ms(dev, HEALTH_POLL_INTERVAL));
710 
711 	return next;
712 }
713 
mlx5_trigger_health_work(struct mlx5_core_dev * dev)714 void mlx5_trigger_health_work(struct mlx5_core_dev *dev)
715 {
716 	struct mlx5_core_health *health = &dev->priv.health;
717 
718 	if (!mlx5_dev_is_lightweight(dev))
719 		queue_work(health->wq, &health->fatal_report_work);
720 }
721 
722 #define MLX5_MSEC_PER_HOUR (MSEC_PER_SEC * 60 * 60)
mlx5_health_log_ts_update(struct work_struct * work)723 static void mlx5_health_log_ts_update(struct work_struct *work)
724 {
725 	struct delayed_work *dwork = to_delayed_work(work);
726 	u32 out[MLX5_ST_SZ_DW(mrtc_reg)] = {};
727 	u32 in[MLX5_ST_SZ_DW(mrtc_reg)] = {};
728 	struct mlx5_core_health *health;
729 	struct mlx5_core_dev *dev;
730 	struct mlx5_priv *priv;
731 	u64 now_us;
732 
733 	health = container_of(dwork, struct mlx5_core_health, update_fw_log_ts_work);
734 	priv = container_of(health, struct mlx5_priv, health);
735 	dev = container_of(priv, struct mlx5_core_dev, priv);
736 
737 	now_us =  ktime_to_us(ktime_get_real());
738 
739 	MLX5_SET(mrtc_reg, in, time_h, now_us >> 32);
740 	MLX5_SET(mrtc_reg, in, time_l, now_us & 0xFFFFFFFF);
741 	mlx5_core_access_reg(dev, in, sizeof(in), out, sizeof(out), MLX5_REG_MRTC, 0, 1);
742 
743 	queue_delayed_work(health->wq, &health->update_fw_log_ts_work,
744 			   msecs_to_jiffies(MLX5_MSEC_PER_HOUR));
745 }
746 
poll_health(struct timer_list * t)747 static void poll_health(struct timer_list *t)
748 {
749 	struct mlx5_core_dev *dev = from_timer(dev, t, priv.health.timer);
750 	struct mlx5_core_health *health = &dev->priv.health;
751 	struct health_buffer __iomem *h = health->health;
752 	u32 fatal_error;
753 	u8 prev_synd;
754 	u32 count;
755 
756 	if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)
757 		goto out;
758 
759 	fatal_error = mlx5_health_check_fatal_sensors(dev);
760 
761 	if (fatal_error && !health->fatal_error) {
762 		mlx5_core_err(dev, "Fatal error %u detected\n", fatal_error);
763 		dev->priv.health.fatal_error = fatal_error;
764 		print_health_info(dev);
765 		dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
766 		mlx5_trigger_health_work(dev);
767 		return;
768 	}
769 
770 	count = ioread32be(health->health_counter);
771 	if (count == health->prev)
772 		++health->miss_counter;
773 	else
774 		health->miss_counter = 0;
775 
776 	health->prev = count;
777 	if (health->miss_counter == MAX_MISSES) {
778 		mlx5_core_err(dev, "device's health compromised - reached miss count\n");
779 		print_health_info(dev);
780 		queue_work(health->wq, &health->report_work);
781 	}
782 
783 	prev_synd = health->synd;
784 	health->synd = ioread8(&h->synd);
785 	if (health->synd && health->synd != prev_synd)
786 		queue_work(health->wq, &health->report_work);
787 
788 out:
789 	mod_timer(&health->timer, get_next_poll_jiffies(dev));
790 }
791 
mlx5_start_health_poll(struct mlx5_core_dev * dev)792 void mlx5_start_health_poll(struct mlx5_core_dev *dev)
793 {
794 	u64 poll_interval_ms =  mlx5_tout_ms(dev, HEALTH_POLL_INTERVAL);
795 	struct mlx5_core_health *health = &dev->priv.health;
796 
797 	timer_setup(&health->timer, poll_health, 0);
798 	health->fatal_error = MLX5_SENSOR_NO_ERR;
799 	clear_bit(MLX5_DROP_HEALTH_WORK, &health->flags);
800 	health->health = &dev->iseg->health;
801 	health->health_counter = &dev->iseg->health_counter;
802 
803 	health->timer.expires = jiffies + msecs_to_jiffies(poll_interval_ms);
804 	add_timer(&health->timer);
805 }
806 
mlx5_stop_health_poll(struct mlx5_core_dev * dev,bool disable_health)807 void mlx5_stop_health_poll(struct mlx5_core_dev *dev, bool disable_health)
808 {
809 	struct mlx5_core_health *health = &dev->priv.health;
810 
811 	if (disable_health)
812 		set_bit(MLX5_DROP_HEALTH_WORK, &health->flags);
813 
814 	del_timer_sync(&health->timer);
815 }
816 
mlx5_start_health_fw_log_up(struct mlx5_core_dev * dev)817 void mlx5_start_health_fw_log_up(struct mlx5_core_dev *dev)
818 {
819 	struct mlx5_core_health *health = &dev->priv.health;
820 
821 	if (mlx5_core_is_pf(dev) && MLX5_CAP_MCAM_REG(dev, mrtc))
822 		queue_delayed_work(health->wq, &health->update_fw_log_ts_work, 0);
823 }
824 
mlx5_drain_health_wq(struct mlx5_core_dev * dev)825 void mlx5_drain_health_wq(struct mlx5_core_dev *dev)
826 {
827 	struct mlx5_core_health *health = &dev->priv.health;
828 
829 	set_bit(MLX5_DROP_HEALTH_WORK, &health->flags);
830 	cancel_delayed_work_sync(&health->update_fw_log_ts_work);
831 	cancel_work_sync(&health->report_work);
832 	cancel_work_sync(&health->fatal_report_work);
833 }
834 
mlx5_health_cleanup(struct mlx5_core_dev * dev)835 void mlx5_health_cleanup(struct mlx5_core_dev *dev)
836 {
837 	struct mlx5_core_health *health = &dev->priv.health;
838 
839 	cancel_delayed_work_sync(&health->update_fw_log_ts_work);
840 	destroy_workqueue(health->wq);
841 	mlx5_reporter_vnic_destroy(dev);
842 	mlx5_fw_reporters_destroy(dev);
843 }
844 
mlx5_health_init(struct mlx5_core_dev * dev)845 int mlx5_health_init(struct mlx5_core_dev *dev)
846 {
847 	struct devlink *devlink = priv_to_devlink(dev);
848 	struct mlx5_core_health *health;
849 	char *name;
850 
851 	if (!mlx5_dev_is_lightweight(dev)) {
852 		devl_lock(devlink);
853 		mlx5_fw_reporters_create(dev);
854 		devl_unlock(devlink);
855 	}
856 	mlx5_reporter_vnic_create(dev);
857 
858 	health = &dev->priv.health;
859 	name = kmalloc(64, GFP_KERNEL);
860 	if (!name)
861 		goto out_err;
862 
863 	strcpy(name, "mlx5_health");
864 	strcat(name, dev_name(dev->device));
865 	health->wq = create_singlethread_workqueue(name);
866 	kfree(name);
867 	if (!health->wq)
868 		goto out_err;
869 	INIT_WORK(&health->fatal_report_work, mlx5_fw_fatal_reporter_err_work);
870 	INIT_WORK(&health->report_work, mlx5_fw_reporter_err_work);
871 	INIT_DELAYED_WORK(&health->update_fw_log_ts_work, mlx5_health_log_ts_update);
872 
873 	return 0;
874 
875 out_err:
876 	mlx5_reporter_vnic_destroy(dev);
877 	mlx5_fw_reporters_destroy(dev);
878 	return -ENOMEM;
879 }
880