2 * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
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"
42 #include "lib/pci_vsc.h"
44 #include "diag/fw_tracer.h"
45 #include "diag/reporter_vnic.h"
52 MLX5_HEALTH_SYNDR_FW_ERR = 0x1,
53 MLX5_HEALTH_SYNDR_IRISC_ERR = 0x7,
54 MLX5_HEALTH_SYNDR_HW_UNRECOVERABLE_ERR = 0x8,
55 MLX5_HEALTH_SYNDR_CRC_ERR = 0x9,
56 MLX5_HEALTH_SYNDR_FETCH_PCI_ERR = 0xa,
57 MLX5_HEALTH_SYNDR_HW_FTL_ERR = 0xb,
58 MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR = 0xc,
59 MLX5_HEALTH_SYNDR_EQ_ERR = 0xd,
60 MLX5_HEALTH_SYNDR_EQ_INV = 0xe,
61 MLX5_HEALTH_SYNDR_FFSER_ERR = 0xf,
62 MLX5_HEALTH_SYNDR_HIGH_TEMP = 0x10
66 MLX5_DROP_HEALTH_WORK,
70 MLX5_SENSOR_NO_ERR = 0,
71 MLX5_SENSOR_PCI_COMM_ERR = 1,
72 MLX5_SENSOR_PCI_ERR = 2,
73 MLX5_SENSOR_NIC_DISABLED = 3,
74 MLX5_SENSOR_NIC_SW_RESET = 4,
75 MLX5_SENSOR_FW_SYND_RFR = 5,
79 MLX5_SEVERITY_MASK = 0x7,
80 MLX5_SEVERITY_VALID_MASK = 0x8,
83 u8 mlx5_get_nic_state(struct mlx5_core_dev *dev)
85 return (ioread32be(&dev->iseg->cmdq_addr_l_sz) >> 8) & 7;
88 void mlx5_set_nic_state(struct mlx5_core_dev *dev, u8 state)
90 u32 cur_cmdq_addr_l_sz;
92 cur_cmdq_addr_l_sz = ioread32be(&dev->iseg->cmdq_addr_l_sz);
93 iowrite32be((cur_cmdq_addr_l_sz & 0xFFFFF000) |
94 state << MLX5_NIC_IFC_OFFSET,
95 &dev->iseg->cmdq_addr_l_sz);
98 static bool sensor_pci_not_working(struct mlx5_core_dev *dev)
100 struct mlx5_core_health *health = &dev->priv.health;
101 struct health_buffer __iomem *h = health->health;
103 /* Offline PCI reads return 0xffffffff */
104 return (ioread32be(&h->fw_ver) == 0xffffffff);
107 static int mlx5_health_get_rfr(u8 rfr_severity)
109 return rfr_severity >> MLX5_RFR_BIT_OFFSET;
112 static bool sensor_fw_synd_rfr(struct mlx5_core_dev *dev)
114 struct mlx5_core_health *health = &dev->priv.health;
115 struct health_buffer __iomem *h = health->health;
116 u8 synd = ioread8(&h->synd);
119 rfr = mlx5_health_get_rfr(ioread8(&h->rfr_severity));
122 mlx5_core_dbg(dev, "FW requests reset, synd: %d\n", synd);
126 u32 mlx5_health_check_fatal_sensors(struct mlx5_core_dev *dev)
128 if (sensor_pci_not_working(dev))
129 return MLX5_SENSOR_PCI_COMM_ERR;
130 if (pci_channel_offline(dev->pdev))
131 return MLX5_SENSOR_PCI_ERR;
132 if (mlx5_get_nic_state(dev) == MLX5_NIC_IFC_DISABLED)
133 return MLX5_SENSOR_NIC_DISABLED;
134 if (mlx5_get_nic_state(dev) == MLX5_NIC_IFC_SW_RESET)
135 return MLX5_SENSOR_NIC_SW_RESET;
136 if (sensor_fw_synd_rfr(dev))
137 return MLX5_SENSOR_FW_SYND_RFR;
139 return MLX5_SENSOR_NO_ERR;
142 static int lock_sem_sw_reset(struct mlx5_core_dev *dev, bool lock)
144 enum mlx5_vsc_state state;
147 if (!mlx5_core_is_pf(dev))
150 /* Try to lock GW access, this stage doesn't return
151 * EBUSY because locked GW does not mean that other PF
152 * already started the reset.
154 ret = mlx5_vsc_gw_lock(dev);
160 state = lock ? MLX5_VSC_LOCK : MLX5_VSC_UNLOCK;
161 /* At this stage, if the return status == EBUSY, then we know
162 * for sure that another PF started the reset, so don't allow
165 ret = mlx5_vsc_sem_set_space(dev, MLX5_SEMAPHORE_SW_RESET, state);
167 mlx5_core_warn(dev, "Failed to lock SW reset semaphore\n");
169 /* Unlock GW access */
170 mlx5_vsc_gw_unlock(dev);
175 static bool reset_fw_if_needed(struct mlx5_core_dev *dev)
177 bool supported = (ioread32be(&dev->iseg->initializing) >>
178 MLX5_FW_RESET_SUPPORTED_OFFSET) & 1;
184 /* The reset only needs to be issued by one PF. The health buffer is
185 * shared between all functions, and will be cleared during a reset.
186 * Check again to avoid a redundant 2nd reset. If the fatal errors was
187 * PCI related a reset won't help.
189 fatal_error = mlx5_health_check_fatal_sensors(dev);
190 if (fatal_error == MLX5_SENSOR_PCI_COMM_ERR ||
191 fatal_error == MLX5_SENSOR_NIC_DISABLED ||
192 fatal_error == MLX5_SENSOR_NIC_SW_RESET) {
193 mlx5_core_warn(dev, "Not issuing FW reset. Either it's already done or won't help.");
197 mlx5_core_warn(dev, "Issuing FW Reset\n");
198 /* Write the NIC interface field to initiate the reset, the command
199 * interface address also resides here, don't overwrite it.
201 mlx5_set_nic_state(dev, MLX5_NIC_IFC_SW_RESET);
206 static void enter_error_state(struct mlx5_core_dev *dev, bool force)
208 if (mlx5_health_check_fatal_sensors(dev) || force) { /* protected state setting */
209 dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
213 mlx5_notifier_call_chain(dev->priv.events, MLX5_DEV_EVENT_SYS_ERROR, (void *)1);
216 void mlx5_enter_error_state(struct mlx5_core_dev *dev, bool force)
218 bool err_detected = false;
220 /* Mark the device as fatal in order to abort FW commands */
221 if ((mlx5_health_check_fatal_sensors(dev) || force) &&
222 dev->state == MLX5_DEVICE_STATE_UP) {
223 dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
226 mutex_lock(&dev->intf_state_mutex);
227 if (!err_detected && dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)
228 goto unlock;/* a previous error is still being handled */
230 enter_error_state(dev, force);
232 mutex_unlock(&dev->intf_state_mutex);
235 void mlx5_error_sw_reset(struct mlx5_core_dev *dev)
237 unsigned long end, delay_ms = mlx5_tout_ms(dev, PCI_TOGGLE);
240 mutex_lock(&dev->intf_state_mutex);
241 if (dev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR)
244 mlx5_core_err(dev, "start\n");
246 if (mlx5_health_check_fatal_sensors(dev) == MLX5_SENSOR_FW_SYND_RFR) {
247 /* Get cr-dump and reset FW semaphore */
248 lock = lock_sem_sw_reset(dev, true);
250 if (lock == -EBUSY) {
251 delay_ms = mlx5_tout_ms(dev, FULL_CRDUMP);
252 goto recover_from_sw_reset;
254 /* Execute SW reset */
255 reset_fw_if_needed(dev);
258 recover_from_sw_reset:
259 /* Recover from SW reset */
260 end = jiffies + msecs_to_jiffies(delay_ms);
262 if (mlx5_get_nic_state(dev) == MLX5_NIC_IFC_DISABLED)
266 } while (!time_after(jiffies, end));
268 if (mlx5_get_nic_state(dev) != MLX5_NIC_IFC_DISABLED) {
269 dev_err(&dev->pdev->dev, "NIC IFC still %d after %lums.\n",
270 mlx5_get_nic_state(dev), delay_ms);
273 /* Release FW semaphore if you are the lock owner */
275 lock_sem_sw_reset(dev, false);
277 mlx5_core_err(dev, "end\n");
280 mutex_unlock(&dev->intf_state_mutex);
283 static void mlx5_handle_bad_state(struct mlx5_core_dev *dev)
285 u8 nic_interface = mlx5_get_nic_state(dev);
287 switch (nic_interface) {
288 case MLX5_NIC_IFC_FULL:
289 mlx5_core_warn(dev, "Expected to see disabled NIC but it is full driver\n");
292 case MLX5_NIC_IFC_DISABLED:
293 mlx5_core_warn(dev, "starting teardown\n");
296 case MLX5_NIC_IFC_NO_DRAM_NIC:
297 mlx5_core_warn(dev, "Expected to see disabled NIC but it is no dram nic\n");
300 case MLX5_NIC_IFC_SW_RESET:
301 /* The IFC mode field is 3 bits, so it will read 0x7 in 2 cases:
302 * 1. PCI has been disabled (ie. PCI-AER, PF driver unloaded
303 * and this is a VF), this is not recoverable by SW reset.
304 * Logging of this is handled elsewhere.
305 * 2. FW reset has been issued by another function, driver can
306 * be reloaded to recover after the mode switches to
307 * MLX5_NIC_IFC_DISABLED.
309 if (dev->priv.health.fatal_error != MLX5_SENSOR_PCI_COMM_ERR)
310 mlx5_core_warn(dev, "NIC SW reset in progress\n");
314 mlx5_core_warn(dev, "Expected to see disabled NIC but it is has invalid value %d\n",
318 mlx5_disable_device(dev);
321 int mlx5_health_wait_pci_up(struct mlx5_core_dev *dev)
325 end = jiffies + msecs_to_jiffies(mlx5_tout_ms(dev, FW_RESET));
326 while (sensor_pci_not_working(dev)) {
327 if (time_after(jiffies, end))
329 if (test_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state)) {
330 mlx5_core_warn(dev, "device is being removed, stop waiting for PCI\n");
338 static int mlx5_health_try_recover(struct mlx5_core_dev *dev)
340 mlx5_core_warn(dev, "handling bad device here\n");
341 mlx5_handle_bad_state(dev);
342 if (mlx5_health_wait_pci_up(dev)) {
343 mlx5_core_err(dev, "health recovery flow aborted, PCI reads still not working\n");
346 mlx5_core_err(dev, "starting health recovery flow\n");
347 if (mlx5_recover_device(dev) || mlx5_health_check_fatal_sensors(dev)) {
348 mlx5_core_err(dev, "health recovery failed\n");
352 mlx5_core_info(dev, "health recovery succeeded\n");
356 static const char *hsynd_str(u8 synd)
359 case MLX5_HEALTH_SYNDR_FW_ERR:
360 return "firmware internal error";
361 case MLX5_HEALTH_SYNDR_IRISC_ERR:
362 return "irisc not responding";
363 case MLX5_HEALTH_SYNDR_HW_UNRECOVERABLE_ERR:
364 return "unrecoverable hardware error";
365 case MLX5_HEALTH_SYNDR_CRC_ERR:
366 return "firmware CRC error";
367 case MLX5_HEALTH_SYNDR_FETCH_PCI_ERR:
368 return "ICM fetch PCI error";
369 case MLX5_HEALTH_SYNDR_HW_FTL_ERR:
370 return "HW fatal error\n";
371 case MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR:
372 return "async EQ buffer overrun";
373 case MLX5_HEALTH_SYNDR_EQ_ERR:
375 case MLX5_HEALTH_SYNDR_EQ_INV:
376 return "Invalid EQ referenced";
377 case MLX5_HEALTH_SYNDR_FFSER_ERR:
378 return "FFSER error";
379 case MLX5_HEALTH_SYNDR_HIGH_TEMP:
380 return "High temperature";
382 return "unrecognized error";
386 static const char *mlx5_loglevel_str(int level)
397 case LOGLEVEL_WARNING:
399 case LOGLEVEL_NOTICE:
406 return "Unknown log level";
409 static int mlx5_health_get_severity(u8 rfr_severity)
411 return rfr_severity & MLX5_SEVERITY_VALID_MASK ?
412 rfr_severity & MLX5_SEVERITY_MASK : LOGLEVEL_ERR;
415 static void print_health_info(struct mlx5_core_dev *dev)
417 struct mlx5_core_health *health = &dev->priv.health;
418 struct health_buffer __iomem *h = health->health;
423 /* If the syndrome is 0, the device is OK and no need to print buffer */
424 if (!ioread8(&h->synd))
427 if (ioread32be(&h->fw_ver) == 0xFFFFFFFF) {
428 mlx5_log(dev, LOGLEVEL_ERR, "PCI slot is unavailable\n");
432 rfr_severity = ioread8(&h->rfr_severity);
433 severity = mlx5_health_get_severity(rfr_severity);
434 mlx5_log(dev, severity, "Health issue observed, %s, severity(%d) %s:\n",
435 hsynd_str(ioread8(&h->synd)), severity, mlx5_loglevel_str(severity));
437 for (i = 0; i < ARRAY_SIZE(h->assert_var); i++)
438 mlx5_log(dev, severity, "assert_var[%d] 0x%08x\n", i,
439 ioread32be(h->assert_var + i));
441 mlx5_log(dev, severity, "assert_exit_ptr 0x%08x\n", ioread32be(&h->assert_exit_ptr));
442 mlx5_log(dev, severity, "assert_callra 0x%08x\n", ioread32be(&h->assert_callra));
443 mlx5_log(dev, severity, "fw_ver %d.%d.%d", fw_rev_maj(dev), fw_rev_min(dev),
445 mlx5_log(dev, severity, "time %u\n", ioread32be(&h->time));
446 mlx5_log(dev, severity, "hw_id 0x%08x\n", ioread32be(&h->hw_id));
447 mlx5_log(dev, severity, "rfr %d\n", mlx5_health_get_rfr(rfr_severity));
448 mlx5_log(dev, severity, "severity %d (%s)\n", severity, mlx5_loglevel_str(severity));
449 mlx5_log(dev, severity, "irisc_index %d\n", ioread8(&h->irisc_index));
450 mlx5_log(dev, severity, "synd 0x%x: %s\n", ioread8(&h->synd),
451 hsynd_str(ioread8(&h->synd)));
452 mlx5_log(dev, severity, "ext_synd 0x%04x\n", ioread16be(&h->ext_synd));
453 mlx5_log(dev, severity, "raw fw_ver 0x%08x\n", ioread32be(&h->fw_ver));
457 mlx5_fw_reporter_diagnose(struct devlink_health_reporter *reporter,
458 struct devlink_fmsg *fmsg,
459 struct netlink_ext_ack *extack)
461 struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
462 struct mlx5_core_health *health = &dev->priv.health;
463 struct health_buffer __iomem *h = health->health;
467 synd = ioread8(&h->synd);
468 err = devlink_fmsg_u8_pair_put(fmsg, "Syndrome", synd);
471 return devlink_fmsg_string_pair_put(fmsg, "Description", hsynd_str(synd));
474 struct mlx5_fw_reporter_ctx {
480 mlx5_fw_reporter_ctx_pairs_put(struct devlink_fmsg *fmsg,
481 struct mlx5_fw_reporter_ctx *fw_reporter_ctx)
485 err = devlink_fmsg_u8_pair_put(fmsg, "syndrome",
486 fw_reporter_ctx->err_synd);
489 err = devlink_fmsg_u32_pair_put(fmsg, "fw_miss_counter",
490 fw_reporter_ctx->miss_counter);
497 mlx5_fw_reporter_heath_buffer_data_put(struct mlx5_core_dev *dev,
498 struct devlink_fmsg *fmsg)
500 struct mlx5_core_health *health = &dev->priv.health;
501 struct health_buffer __iomem *h = health->health;
506 if (!ioread8(&h->synd))
509 err = devlink_fmsg_pair_nest_start(fmsg, "health buffer");
512 err = devlink_fmsg_obj_nest_start(fmsg);
515 err = devlink_fmsg_arr_pair_nest_start(fmsg, "assert_var");
519 for (i = 0; i < ARRAY_SIZE(h->assert_var); i++) {
520 err = devlink_fmsg_u32_put(fmsg, ioread32be(h->assert_var + i));
524 err = devlink_fmsg_arr_pair_nest_end(fmsg);
527 err = devlink_fmsg_u32_pair_put(fmsg, "assert_exit_ptr",
528 ioread32be(&h->assert_exit_ptr));
531 err = devlink_fmsg_u32_pair_put(fmsg, "assert_callra",
532 ioread32be(&h->assert_callra));
535 err = devlink_fmsg_u32_pair_put(fmsg, "time", ioread32be(&h->time));
538 err = devlink_fmsg_u32_pair_put(fmsg, "hw_id", ioread32be(&h->hw_id));
541 rfr_severity = ioread8(&h->rfr_severity);
542 err = devlink_fmsg_u8_pair_put(fmsg, "rfr", mlx5_health_get_rfr(rfr_severity));
545 err = devlink_fmsg_u8_pair_put(fmsg, "severity", mlx5_health_get_severity(rfr_severity));
548 err = devlink_fmsg_u8_pair_put(fmsg, "irisc_index",
549 ioread8(&h->irisc_index));
552 err = devlink_fmsg_u8_pair_put(fmsg, "synd", ioread8(&h->synd));
555 err = devlink_fmsg_u32_pair_put(fmsg, "ext_synd",
556 ioread16be(&h->ext_synd));
559 err = devlink_fmsg_u32_pair_put(fmsg, "raw_fw_ver",
560 ioread32be(&h->fw_ver));
563 err = devlink_fmsg_obj_nest_end(fmsg);
566 return devlink_fmsg_pair_nest_end(fmsg);
570 mlx5_fw_reporter_dump(struct devlink_health_reporter *reporter,
571 struct devlink_fmsg *fmsg, void *priv_ctx,
572 struct netlink_ext_ack *extack)
574 struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
577 err = mlx5_fw_tracer_trigger_core_dump_general(dev);
582 struct mlx5_fw_reporter_ctx *fw_reporter_ctx = priv_ctx;
584 err = mlx5_fw_reporter_ctx_pairs_put(fmsg, fw_reporter_ctx);
589 err = mlx5_fw_reporter_heath_buffer_data_put(dev, fmsg);
592 return mlx5_fw_tracer_get_saved_traces_objects(dev->tracer, fmsg);
595 static void mlx5_fw_reporter_err_work(struct work_struct *work)
597 struct mlx5_fw_reporter_ctx fw_reporter_ctx;
598 struct mlx5_core_health *health;
600 health = container_of(work, struct mlx5_core_health, report_work);
602 if (IS_ERR_OR_NULL(health->fw_reporter))
605 fw_reporter_ctx.err_synd = health->synd;
606 fw_reporter_ctx.miss_counter = health->miss_counter;
607 if (fw_reporter_ctx.err_synd) {
608 devlink_health_report(health->fw_reporter,
609 "FW syndrome reported", &fw_reporter_ctx);
612 if (fw_reporter_ctx.miss_counter)
613 devlink_health_report(health->fw_reporter,
614 "FW miss counter reported",
618 static const struct devlink_health_reporter_ops mlx5_fw_reporter_ops = {
620 .diagnose = mlx5_fw_reporter_diagnose,
621 .dump = mlx5_fw_reporter_dump,
625 mlx5_fw_fatal_reporter_recover(struct devlink_health_reporter *reporter,
627 struct netlink_ext_ack *extack)
629 struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
631 return mlx5_health_try_recover(dev);
635 mlx5_fw_fatal_reporter_dump(struct devlink_health_reporter *reporter,
636 struct devlink_fmsg *fmsg, void *priv_ctx,
637 struct netlink_ext_ack *extack)
639 struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
640 u32 crdump_size = dev->priv.health.crdump_size;
644 if (!mlx5_core_is_pf(dev))
647 cr_data = kvmalloc(crdump_size, GFP_KERNEL);
650 err = mlx5_crdump_collect(dev, cr_data);
655 struct mlx5_fw_reporter_ctx *fw_reporter_ctx = priv_ctx;
657 err = mlx5_fw_reporter_ctx_pairs_put(fmsg, fw_reporter_ctx);
662 err = devlink_fmsg_binary_pair_put(fmsg, "crdump_data", cr_data, crdump_size);
669 static void mlx5_fw_fatal_reporter_err_work(struct work_struct *work)
671 struct mlx5_fw_reporter_ctx fw_reporter_ctx;
672 struct mlx5_core_health *health;
673 struct mlx5_core_dev *dev;
674 struct devlink *devlink;
675 struct mlx5_priv *priv;
677 health = container_of(work, struct mlx5_core_health, fatal_report_work);
678 priv = container_of(health, struct mlx5_priv, health);
679 dev = container_of(priv, struct mlx5_core_dev, priv);
680 devlink = priv_to_devlink(dev);
682 mutex_lock(&dev->intf_state_mutex);
683 if (test_bit(MLX5_DROP_HEALTH_WORK, &health->flags)) {
684 mlx5_core_err(dev, "health works are not permitted at this stage\n");
685 mutex_unlock(&dev->intf_state_mutex);
688 mutex_unlock(&dev->intf_state_mutex);
689 enter_error_state(dev, false);
690 if (IS_ERR_OR_NULL(health->fw_fatal_reporter)) {
692 if (mlx5_health_try_recover(dev))
693 mlx5_core_err(dev, "health recovery failed\n");
694 devl_unlock(devlink);
697 fw_reporter_ctx.err_synd = health->synd;
698 fw_reporter_ctx.miss_counter = health->miss_counter;
699 if (devlink_health_report(health->fw_fatal_reporter,
700 "FW fatal error reported", &fw_reporter_ctx) == -ECANCELED) {
701 /* If recovery wasn't performed, due to grace period,
702 * unload the driver. This ensures that the driver
703 * closes all its resources and it is not subjected to
704 * requests from the kernel.
706 mlx5_core_err(dev, "Driver is in error state. Unloading\n");
707 mlx5_unload_one(dev, false);
711 static const struct devlink_health_reporter_ops mlx5_fw_fatal_reporter_ops = {
713 .recover = mlx5_fw_fatal_reporter_recover,
714 .dump = mlx5_fw_fatal_reporter_dump,
717 #define MLX5_FW_REPORTER_ECPF_GRACEFUL_PERIOD 180000
718 #define MLX5_FW_REPORTER_PF_GRACEFUL_PERIOD 60000
719 #define MLX5_FW_REPORTER_VF_GRACEFUL_PERIOD 30000
720 #define MLX5_FW_REPORTER_DEFAULT_GRACEFUL_PERIOD MLX5_FW_REPORTER_VF_GRACEFUL_PERIOD
722 static void mlx5_fw_reporters_create(struct mlx5_core_dev *dev)
724 struct mlx5_core_health *health = &dev->priv.health;
725 struct devlink *devlink = priv_to_devlink(dev);
728 if (mlx5_core_is_ecpf(dev)) {
729 grace_period = MLX5_FW_REPORTER_ECPF_GRACEFUL_PERIOD;
730 } else if (mlx5_core_is_pf(dev)) {
731 grace_period = MLX5_FW_REPORTER_PF_GRACEFUL_PERIOD;
734 grace_period = MLX5_FW_REPORTER_DEFAULT_GRACEFUL_PERIOD;
737 health->fw_reporter =
738 devlink_health_reporter_create(devlink, &mlx5_fw_reporter_ops,
740 if (IS_ERR(health->fw_reporter))
741 mlx5_core_warn(dev, "Failed to create fw reporter, err = %ld\n",
742 PTR_ERR(health->fw_reporter));
744 health->fw_fatal_reporter =
745 devlink_health_reporter_create(devlink,
746 &mlx5_fw_fatal_reporter_ops,
749 if (IS_ERR(health->fw_fatal_reporter))
750 mlx5_core_warn(dev, "Failed to create fw fatal reporter, err = %ld\n",
751 PTR_ERR(health->fw_fatal_reporter));
754 static void mlx5_fw_reporters_destroy(struct mlx5_core_dev *dev)
756 struct mlx5_core_health *health = &dev->priv.health;
758 if (!IS_ERR_OR_NULL(health->fw_reporter))
759 devlink_health_reporter_destroy(health->fw_reporter);
761 if (!IS_ERR_OR_NULL(health->fw_fatal_reporter))
762 devlink_health_reporter_destroy(health->fw_fatal_reporter);
765 static unsigned long get_next_poll_jiffies(struct mlx5_core_dev *dev)
769 get_random_bytes(&next, sizeof(next));
771 next += jiffies + msecs_to_jiffies(mlx5_tout_ms(dev, HEALTH_POLL_INTERVAL));
776 void mlx5_trigger_health_work(struct mlx5_core_dev *dev)
778 struct mlx5_core_health *health = &dev->priv.health;
780 queue_work(health->wq, &health->fatal_report_work);
783 #define MLX5_MSEC_PER_HOUR (MSEC_PER_SEC * 60 * 60)
784 static void mlx5_health_log_ts_update(struct work_struct *work)
786 struct delayed_work *dwork = to_delayed_work(work);
787 u32 out[MLX5_ST_SZ_DW(mrtc_reg)] = {};
788 u32 in[MLX5_ST_SZ_DW(mrtc_reg)] = {};
789 struct mlx5_core_health *health;
790 struct mlx5_core_dev *dev;
791 struct mlx5_priv *priv;
794 health = container_of(dwork, struct mlx5_core_health, update_fw_log_ts_work);
795 priv = container_of(health, struct mlx5_priv, health);
796 dev = container_of(priv, struct mlx5_core_dev, priv);
798 now_us = ktime_to_us(ktime_get_real());
800 MLX5_SET(mrtc_reg, in, time_h, now_us >> 32);
801 MLX5_SET(mrtc_reg, in, time_l, now_us & 0xFFFFFFFF);
802 mlx5_core_access_reg(dev, in, sizeof(in), out, sizeof(out), MLX5_REG_MRTC, 0, 1);
804 queue_delayed_work(health->wq, &health->update_fw_log_ts_work,
805 msecs_to_jiffies(MLX5_MSEC_PER_HOUR));
808 static void poll_health(struct timer_list *t)
810 struct mlx5_core_dev *dev = from_timer(dev, t, priv.health.timer);
811 struct mlx5_core_health *health = &dev->priv.health;
812 struct health_buffer __iomem *h = health->health;
817 if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)
820 fatal_error = mlx5_health_check_fatal_sensors(dev);
822 if (fatal_error && !health->fatal_error) {
823 mlx5_core_err(dev, "Fatal error %u detected\n", fatal_error);
824 dev->priv.health.fatal_error = fatal_error;
825 print_health_info(dev);
826 dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
827 mlx5_trigger_health_work(dev);
831 count = ioread32be(health->health_counter);
832 if (count == health->prev)
833 ++health->miss_counter;
835 health->miss_counter = 0;
837 health->prev = count;
838 if (health->miss_counter == MAX_MISSES) {
839 mlx5_core_err(dev, "device's health compromised - reached miss count\n");
840 print_health_info(dev);
841 queue_work(health->wq, &health->report_work);
844 prev_synd = health->synd;
845 health->synd = ioread8(&h->synd);
846 if (health->synd && health->synd != prev_synd)
847 queue_work(health->wq, &health->report_work);
850 mod_timer(&health->timer, get_next_poll_jiffies(dev));
853 void mlx5_start_health_poll(struct mlx5_core_dev *dev)
855 u64 poll_interval_ms = mlx5_tout_ms(dev, HEALTH_POLL_INTERVAL);
856 struct mlx5_core_health *health = &dev->priv.health;
858 timer_setup(&health->timer, poll_health, 0);
859 health->fatal_error = MLX5_SENSOR_NO_ERR;
860 clear_bit(MLX5_DROP_HEALTH_WORK, &health->flags);
861 health->health = &dev->iseg->health;
862 health->health_counter = &dev->iseg->health_counter;
864 health->timer.expires = jiffies + msecs_to_jiffies(poll_interval_ms);
865 add_timer(&health->timer);
868 void mlx5_stop_health_poll(struct mlx5_core_dev *dev, bool disable_health)
870 struct mlx5_core_health *health = &dev->priv.health;
873 set_bit(MLX5_DROP_HEALTH_WORK, &health->flags);
875 del_timer_sync(&health->timer);
878 void mlx5_start_health_fw_log_up(struct mlx5_core_dev *dev)
880 struct mlx5_core_health *health = &dev->priv.health;
882 if (mlx5_core_is_pf(dev) && MLX5_CAP_MCAM_REG(dev, mrtc))
883 queue_delayed_work(health->wq, &health->update_fw_log_ts_work, 0);
886 void mlx5_drain_health_wq(struct mlx5_core_dev *dev)
888 struct mlx5_core_health *health = &dev->priv.health;
890 set_bit(MLX5_DROP_HEALTH_WORK, &health->flags);
891 cancel_delayed_work_sync(&health->update_fw_log_ts_work);
892 cancel_work_sync(&health->report_work);
893 cancel_work_sync(&health->fatal_report_work);
896 void mlx5_health_cleanup(struct mlx5_core_dev *dev)
898 struct mlx5_core_health *health = &dev->priv.health;
900 cancel_delayed_work_sync(&health->update_fw_log_ts_work);
901 destroy_workqueue(health->wq);
902 mlx5_reporter_vnic_destroy(dev);
903 mlx5_fw_reporters_destroy(dev);
906 int mlx5_health_init(struct mlx5_core_dev *dev)
908 struct mlx5_core_health *health;
911 mlx5_fw_reporters_create(dev);
912 mlx5_reporter_vnic_create(dev);
914 health = &dev->priv.health;
915 name = kmalloc(64, GFP_KERNEL);
919 strcpy(name, "mlx5_health");
920 strcat(name, dev_name(dev->device));
921 health->wq = create_singlethread_workqueue(name);
925 INIT_WORK(&health->fatal_report_work, mlx5_fw_fatal_reporter_err_work);
926 INIT_WORK(&health->report_work, mlx5_fw_reporter_err_work);
927 INIT_DELAYED_WORK(&health->update_fw_log_ts_work, mlx5_health_log_ts_update);
932 mlx5_reporter_vnic_destroy(dev);
933 mlx5_fw_reporters_destroy(dev);