2 * Copyright (c) 2013-2017, Mellanox Technologies, Ltd. All rights reserved.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
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.
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
28 #include <linux/kernel.h>
29 #include <linux/module.h>
30 #include <linux/random.h>
31 #include <linux/vmalloc.h>
32 #include <linux/hardirq.h>
33 #include <linux/delay.h>
34 #include <dev/mlx5/driver.h>
35 #include <dev/mlx5/mlx5_ifc.h>
36 #include "mlx5_core.h"
38 #define MLX5_HEALTH_POLL_INTERVAL (2 * HZ)
42 MLX5_NIC_IFC_FULL = 0,
43 MLX5_NIC_IFC_DISABLED = 1,
44 MLX5_NIC_IFC_NO_DRAM_NIC = 2,
45 MLX5_NIC_IFC_SW_RESET = 7,
49 MLX5_DROP_NEW_HEALTH_WORK,
50 MLX5_DROP_NEW_RECOVERY_WORK,
54 MLX5_SENSOR_NO_ERR = 0,
55 MLX5_SENSOR_PCI_COMM_ERR = 1,
56 MLX5_SENSOR_PCI_ERR = 2,
57 MLX5_SENSOR_NIC_DISABLED = 3,
58 MLX5_SENSOR_NIC_SW_RESET = 4,
59 MLX5_SENSOR_FW_SYND_RFR = 5,
62 static int mlx5_fw_reset_enable = 1;
63 SYSCTL_INT(_hw_mlx5, OID_AUTO, fw_reset_enable, CTLFLAG_RWTUN,
64 &mlx5_fw_reset_enable, 0,
65 "Enable firmware reset");
67 static int lock_sem_sw_reset(struct mlx5_core_dev *dev)
72 ret = -mlx5_vsc_lock(dev);
74 mlx5_core_warn(dev, "Timed out locking gateway %d\n", ret);
78 ret = -mlx5_vsc_lock_addr_space(dev, MLX5_SEMAPHORE_SW_RESET);
81 mlx5_core_dbg(dev, "SW reset FW semaphore already locked, another function will handle the reset\n");
83 mlx5_core_warn(dev, "SW reset semaphore lock return %d\n", ret);
86 /* Unlock GW access */
92 static int unlock_sem_sw_reset(struct mlx5_core_dev *dev)
97 ret = -mlx5_vsc_lock(dev);
99 mlx5_core_warn(dev, "Timed out locking gateway %d\n", ret);
103 ret = -mlx5_vsc_unlock_addr_space(dev, MLX5_SEMAPHORE_SW_RESET);
105 /* Unlock GW access */
106 mlx5_vsc_unlock(dev);
111 static u8 get_nic_mode(struct mlx5_core_dev *dev)
113 return (ioread32be(&dev->iseg->cmdq_addr_l_sz) >> 8) & 7;
116 static bool sensor_fw_synd_rfr(struct mlx5_core_dev *dev)
118 struct mlx5_core_health *health = &dev->priv.health;
119 struct mlx5_health_buffer __iomem *h = health->health;
120 u32 rfr = ioread32be(&h->rfr) >> MLX5_RFR_OFFSET;
121 u8 synd = ioread8(&h->synd);
124 mlx5_core_dbg(dev, "FW requests reset, synd: %d\n", synd);
128 static void mlx5_trigger_cmd_completions(struct mlx5_core_dev *dev)
133 /* wait for pending handlers to complete */
134 synchronize_irq(dev->priv.msix_arr[MLX5_EQ_VEC_CMD].vector);
135 spin_lock_irqsave(&dev->cmd.alloc_lock, flags);
136 vector = ~dev->cmd.bitmask & ((1ul << (1 << dev->cmd.log_sz)) - 1);
140 vector |= MLX5_TRIGGERED_CMD_COMP;
141 spin_unlock_irqrestore(&dev->cmd.alloc_lock, flags);
143 mlx5_core_dbg(dev, "vector 0x%jx\n", (uintmax_t)vector);
144 mlx5_cmd_comp_handler(dev, vector, MLX5_CMD_MODE_EVENTS);
148 spin_unlock_irqrestore(&dev->cmd.alloc_lock, flags);
151 static bool sensor_pci_no_comm(struct mlx5_core_dev *dev)
153 struct mlx5_core_health *health = &dev->priv.health;
154 struct mlx5_health_buffer __iomem *h = health->health;
155 bool err = ioread32be(&h->fw_ver) == 0xffffffff;
160 static bool sensor_nic_disabled(struct mlx5_core_dev *dev)
162 return get_nic_mode(dev) == MLX5_NIC_IFC_DISABLED;
165 static bool sensor_nic_sw_reset(struct mlx5_core_dev *dev)
167 return get_nic_mode(dev) == MLX5_NIC_IFC_SW_RESET;
170 static u32 check_fatal_sensors(struct mlx5_core_dev *dev)
172 if (sensor_pci_no_comm(dev))
173 return MLX5_SENSOR_PCI_COMM_ERR;
174 if (pci_channel_offline(dev->pdev))
175 return MLX5_SENSOR_PCI_ERR;
176 if (sensor_nic_disabled(dev))
177 return MLX5_SENSOR_NIC_DISABLED;
178 if (sensor_nic_sw_reset(dev))
179 return MLX5_SENSOR_NIC_SW_RESET;
180 if (sensor_fw_synd_rfr(dev))
181 return MLX5_SENSOR_FW_SYND_RFR;
183 return MLX5_SENSOR_NO_ERR;
186 static void reset_fw_if_needed(struct mlx5_core_dev *dev)
189 u32 cmdq_addr, fatal_error;
191 if (!mlx5_fw_reset_enable)
193 supported = (ioread32be(&dev->iseg->initializing) >>
194 MLX5_FW_RESET_SUPPORTED_OFFSET) & 1;
198 /* The reset only needs to be issued by one PF. The health buffer is
199 * shared between all functions, and will be cleared during a reset.
200 * Check again to avoid a redundant 2nd reset. If the fatal erros was
201 * PCI related a reset won't help.
203 fatal_error = check_fatal_sensors(dev);
204 if (fatal_error == MLX5_SENSOR_PCI_COMM_ERR ||
205 fatal_error == MLX5_SENSOR_NIC_DISABLED ||
206 fatal_error == MLX5_SENSOR_NIC_SW_RESET) {
207 mlx5_core_warn(dev, "Not issuing FW reset. Either it's already done or won't help.\n");
211 mlx5_core_warn(dev, "Issuing FW Reset\n");
212 /* Write the NIC interface field to initiate the reset, the command
213 * interface address also resides here, don't overwrite it.
215 cmdq_addr = ioread32be(&dev->iseg->cmdq_addr_l_sz);
216 iowrite32be((cmdq_addr & 0xFFFFF000) |
217 MLX5_NIC_IFC_SW_RESET << MLX5_NIC_IFC_OFFSET,
218 &dev->iseg->cmdq_addr_l_sz);
221 #define MLX5_CRDUMP_WAIT_MS 60000
222 #define MLX5_FW_RESET_WAIT_MS 1000
223 #define MLX5_NIC_STATE_POLL_MS 5
224 void mlx5_enter_error_state(struct mlx5_core_dev *dev, bool force)
226 unsigned long end, delay_ms = MLX5_CRDUMP_WAIT_MS;
230 fatal_error = check_fatal_sensors(dev);
232 if (fatal_error || force) {
233 if (xchg(&dev->state, MLX5_DEVICE_STATE_INTERNAL_ERROR) ==
234 MLX5_DEVICE_STATE_INTERNAL_ERROR)
237 mlx5_core_err(dev, "internal state error detected\n");
238 mlx5_trigger_cmd_completions(dev);
241 mutex_lock(&dev->intf_state_mutex);
246 if (fatal_error == MLX5_SENSOR_FW_SYND_RFR) {
247 /* Get cr-dump and reset FW semaphore */
248 if (mlx5_core_is_pf(dev))
249 lock = lock_sem_sw_reset(dev);
251 /* Execute cr-dump and SW reset */
252 if (lock != -EBUSY) {
254 reset_fw_if_needed(dev);
255 delay_ms = MLX5_FW_RESET_WAIT_MS;
259 /* Recover from SW reset */
260 end = jiffies + msecs_to_jiffies(delay_ms);
262 if (sensor_nic_disabled(dev))
265 msleep(MLX5_NIC_STATE_POLL_MS);
266 } while (!time_after(jiffies, end));
268 if (!sensor_nic_disabled(dev)) {
269 dev_err(&dev->pdev->dev, "NIC IFC still %d after %lums.\n",
270 get_nic_mode(dev), delay_ms);
273 /* Release FW semaphore if you are the lock owner */
275 unlock_sem_sw_reset(dev);
277 mlx5_core_err(dev, "system error event triggered\n");
280 mlx5_core_event(dev, MLX5_DEV_EVENT_SYS_ERROR, 1);
281 mutex_unlock(&dev->intf_state_mutex);
284 static void mlx5_handle_bad_state(struct mlx5_core_dev *dev)
286 u8 nic_mode = get_nic_mode(dev);
288 if (nic_mode == MLX5_NIC_IFC_SW_RESET) {
289 /* The IFC mode field is 3 bits, so it will read 0x7 in two cases:
290 * 1. PCI has been disabled (ie. PCI-AER, PF driver unloaded
291 * and this is a VF), this is not recoverable by SW reset.
292 * Logging of this is handled elsewhere.
293 * 2. FW reset has been issued by another function, driver can
294 * be reloaded to recover after the mode switches to
295 * MLX5_NIC_IFC_DISABLED.
297 if (dev->priv.health.fatal_error != MLX5_SENSOR_PCI_COMM_ERR)
298 mlx5_core_warn(dev, "NIC SW reset is already progress\n");
300 mlx5_core_warn(dev, "Communication with FW over the PCI link is down\n");
302 mlx5_core_warn(dev, "NIC mode %d\n", nic_mode);
305 mlx5_disable_device(dev);
308 #define MLX5_FW_RESET_WAIT_MS 1000
309 #define MLX5_NIC_STATE_POLL_MS 5
310 static void health_recover(struct work_struct *work)
312 unsigned long end = jiffies + msecs_to_jiffies(MLX5_FW_RESET_WAIT_MS);
313 struct mlx5_core_health *health;
314 struct delayed_work *dwork;
315 struct mlx5_core_dev *dev;
316 struct mlx5_priv *priv;
320 dwork = container_of(work, struct delayed_work, work);
321 health = container_of(dwork, struct mlx5_core_health, recover_work);
322 priv = container_of(health, struct mlx5_priv, health);
323 dev = container_of(priv, struct mlx5_core_dev, priv);
325 mtx_lock(&Giant); /* XXX newbus needs this */
327 if (sensor_pci_no_comm(dev)) {
328 dev_err(&dev->pdev->dev, "health recovery flow aborted, PCI reads still not working\n");
332 nic_mode = get_nic_mode(dev);
333 while (nic_mode != MLX5_NIC_IFC_DISABLED &&
334 !time_after(jiffies, end)) {
335 msleep(MLX5_NIC_STATE_POLL_MS);
336 nic_mode = get_nic_mode(dev);
339 if (nic_mode != MLX5_NIC_IFC_DISABLED) {
340 dev_err(&dev->pdev->dev, "health recovery flow aborted, unexpected NIC IFC mode %d.\n",
346 dev_err(&dev->pdev->dev, "starting health recovery flow\n");
347 mlx5_recover_device(dev);
353 /* How much time to wait until health resetting the driver (in msecs) */
354 #define MLX5_RECOVERY_DELAY_MSECS 60000
355 #define MLX5_RECOVERY_NO_DELAY 0
356 static unsigned long get_recovery_delay(struct mlx5_core_dev *dev)
358 return dev->priv.health.fatal_error == MLX5_SENSOR_PCI_ERR ||
359 dev->priv.health.fatal_error == MLX5_SENSOR_PCI_COMM_ERR ?
360 MLX5_RECOVERY_DELAY_MSECS : MLX5_RECOVERY_NO_DELAY;
363 static void health_care(struct work_struct *work)
365 struct mlx5_core_health *health;
366 unsigned long recover_delay;
367 struct mlx5_core_dev *dev;
368 struct mlx5_priv *priv;
371 health = container_of(work, struct mlx5_core_health, work);
372 priv = container_of(health, struct mlx5_priv, health);
373 dev = container_of(priv, struct mlx5_core_dev, priv);
375 mlx5_core_warn(dev, "handling bad device here\n");
376 mlx5_handle_bad_state(dev);
377 recover_delay = msecs_to_jiffies(get_recovery_delay(dev));
379 spin_lock_irqsave(&health->wq_lock, flags);
380 if (!test_bit(MLX5_DROP_NEW_RECOVERY_WORK, &health->flags)) {
381 mlx5_core_warn(dev, "Scheduling recovery work with %lums delay\n",
383 schedule_delayed_work(&health->recover_work, recover_delay);
385 dev_err(&dev->pdev->dev,
386 "new health works are not permitted at this stage\n");
388 spin_unlock_irqrestore(&health->wq_lock, flags);
391 static int get_next_poll_jiffies(void)
395 get_random_bytes(&next, sizeof(next));
397 next += jiffies + MLX5_HEALTH_POLL_INTERVAL;
402 void mlx5_trigger_health_work(struct mlx5_core_dev *dev)
404 struct mlx5_core_health *health = &dev->priv.health;
407 spin_lock_irqsave(&health->wq_lock, flags);
408 if (!test_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags))
409 queue_work(health->wq, &health->work);
411 dev_err(&dev->pdev->dev,
412 "new health works are not permitted at this stage\n");
413 spin_unlock_irqrestore(&health->wq_lock, flags);
416 static const char *hsynd_str(u8 synd)
419 case MLX5_HEALTH_SYNDR_FW_ERR:
420 return "firmware internal error";
421 case MLX5_HEALTH_SYNDR_IRISC_ERR:
422 return "irisc not responding";
423 case MLX5_HEALTH_SYNDR_HW_UNRECOVERABLE_ERR:
424 return "unrecoverable hardware error";
425 case MLX5_HEALTH_SYNDR_CRC_ERR:
426 return "firmware CRC error";
427 case MLX5_HEALTH_SYNDR_FETCH_PCI_ERR:
428 return "ICM fetch PCI error";
429 case MLX5_HEALTH_SYNDR_HW_FTL_ERR:
430 return "HW fatal error\n";
431 case MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR:
432 return "async EQ buffer overrun";
433 case MLX5_HEALTH_SYNDR_EQ_ERR:
435 case MLX5_HEALTH_SYNDR_EQ_INV:
436 return "Invalid EQ referenced";
437 case MLX5_HEALTH_SYNDR_FFSER_ERR:
438 return "FFSER error";
439 case MLX5_HEALTH_SYNDR_HIGH_TEMP:
440 return "High temprature";
442 return "unrecognized error";
446 static void print_health_info(struct mlx5_core_dev *dev)
448 struct mlx5_core_health *health = &dev->priv.health;
449 struct mlx5_health_buffer __iomem *h = health->health;
454 /* If the syndrom is 0, the device is OK and no need to print buffer */
455 if (!ioread8(&h->synd))
458 for (i = 0; i < ARRAY_SIZE(h->assert_var); i++)
459 printf("mlx5_core: INFO: ""assert_var[%d] 0x%08x\n", i, ioread32be(h->assert_var + i));
461 printf("mlx5_core: INFO: ""assert_exit_ptr 0x%08x\n", ioread32be(&h->assert_exit_ptr));
462 printf("mlx5_core: INFO: ""assert_callra 0x%08x\n", ioread32be(&h->assert_callra));
463 snprintf(fw_str, sizeof(fw_str), "%d.%d.%d", fw_rev_maj(dev), fw_rev_min(dev), fw_rev_sub(dev));
464 printf("mlx5_core: INFO: ""fw_ver %s\n", fw_str);
465 printf("mlx5_core: INFO: ""hw_id 0x%08x\n", ioread32be(&h->hw_id));
466 printf("mlx5_core: INFO: ""irisc_index %d\n", ioread8(&h->irisc_index));
467 printf("mlx5_core: INFO: ""synd 0x%x: %s\n", ioread8(&h->synd), hsynd_str(ioread8(&h->synd)));
468 printf("mlx5_core: INFO: ""ext_synd 0x%04x\n", ioread16be(&h->ext_synd));
469 fw = ioread32be(&h->fw_ver);
470 printf("mlx5_core: INFO: ""raw fw_ver 0x%08x\n", fw);
473 static void poll_health(unsigned long data)
475 struct mlx5_core_dev *dev = (struct mlx5_core_dev *)data;
476 struct mlx5_core_health *health = &dev->priv.health;
480 if (dev->state != MLX5_DEVICE_STATE_UP)
483 if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)
486 count = ioread32be(health->health_counter);
487 if (count == health->prev)
488 ++health->miss_counter;
490 health->miss_counter = 0;
492 health->prev = count;
493 if (health->miss_counter == MAX_MISSES) {
494 mlx5_core_err(dev, "device's health compromised - reached miss count\n");
495 print_health_info(dev);
498 fatal_error = check_fatal_sensors(dev);
500 if (fatal_error && !health->fatal_error) {
501 mlx5_core_err(dev, "Fatal error %u detected\n", fatal_error);
502 dev->priv.health.fatal_error = fatal_error;
503 print_health_info(dev);
504 mlx5_trigger_health_work(dev);
508 mod_timer(&health->timer, get_next_poll_jiffies());
511 void mlx5_start_health_poll(struct mlx5_core_dev *dev)
513 struct mlx5_core_health *health = &dev->priv.health;
515 init_timer(&health->timer);
516 health->fatal_error = MLX5_SENSOR_NO_ERR;
517 clear_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
518 clear_bit(MLX5_DROP_NEW_RECOVERY_WORK, &health->flags);
519 health->health = &dev->iseg->health;
520 health->health_counter = &dev->iseg->health_counter;
522 setup_timer(&health->timer, poll_health, (unsigned long)dev);
523 mod_timer(&health->timer,
524 round_jiffies(jiffies + MLX5_HEALTH_POLL_INTERVAL));
527 void mlx5_stop_health_poll(struct mlx5_core_dev *dev, bool disable_health)
529 struct mlx5_core_health *health = &dev->priv.health;
532 if (disable_health) {
533 spin_lock_irqsave(&health->wq_lock, flags);
534 set_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
535 set_bit(MLX5_DROP_NEW_RECOVERY_WORK, &health->flags);
536 spin_unlock_irqrestore(&health->wq_lock, flags);
539 del_timer_sync(&health->timer);
542 void mlx5_drain_health_wq(struct mlx5_core_dev *dev)
544 struct mlx5_core_health *health = &dev->priv.health;
547 spin_lock_irqsave(&health->wq_lock, flags);
548 set_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
549 set_bit(MLX5_DROP_NEW_RECOVERY_WORK, &health->flags);
550 spin_unlock_irqrestore(&health->wq_lock, flags);
551 cancel_delayed_work_sync(&health->recover_work);
552 cancel_work_sync(&health->work);
555 void mlx5_drain_health_recovery(struct mlx5_core_dev *dev)
557 struct mlx5_core_health *health = &dev->priv.health;
560 spin_lock_irqsave(&health->wq_lock, flags);
561 set_bit(MLX5_DROP_NEW_RECOVERY_WORK, &health->flags);
562 spin_unlock_irqrestore(&health->wq_lock, flags);
563 cancel_delayed_work_sync(&dev->priv.health.recover_work);
566 void mlx5_health_cleanup(struct mlx5_core_dev *dev)
568 struct mlx5_core_health *health = &dev->priv.health;
570 destroy_workqueue(health->wq);
573 #define HEALTH_NAME "mlx5_health"
574 int mlx5_health_init(struct mlx5_core_dev *dev)
576 struct mlx5_core_health *health;
580 health = &dev->priv.health;
581 len = strlen(HEALTH_NAME) + strlen(dev_name(&dev->pdev->dev));
582 name = kmalloc(len + 1, GFP_KERNEL);
586 snprintf(name, len, "%s:%s", HEALTH_NAME, dev_name(&dev->pdev->dev));
587 health->wq = create_singlethread_workqueue(name);
592 spin_lock_init(&health->wq_lock);
593 INIT_WORK(&health->work, health_care);
594 INIT_DELAYED_WORK(&health->recover_work, health_recover);