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 lock_sem_sw_reset(struct mlx5_core_dev *dev)
67 ret = -mlx5_vsc_lock(dev);
69 mlx5_core_warn(dev, "Timed out locking gateway %d\n", ret);
73 ret = -mlx5_vsc_lock_addr_space(dev, MLX5_SEMAPHORE_SW_RESET);
76 mlx5_core_dbg(dev, "SW reset FW semaphore already locked, another function will handle the reset\n");
78 mlx5_core_warn(dev, "SW reset semaphore lock return %d\n", ret);
81 /* Unlock GW access */
87 static int unlock_sem_sw_reset(struct mlx5_core_dev *dev)
92 ret = -mlx5_vsc_lock(dev);
94 mlx5_core_warn(dev, "Timed out locking gateway %d\n", ret);
98 ret = -mlx5_vsc_unlock_addr_space(dev, MLX5_SEMAPHORE_SW_RESET);
100 /* Unlock GW access */
101 mlx5_vsc_unlock(dev);
106 static u8 get_nic_mode(struct mlx5_core_dev *dev)
108 return (ioread32be(&dev->iseg->cmdq_addr_l_sz) >> 8) & 7;
111 static bool sensor_fw_synd_rfr(struct mlx5_core_dev *dev)
113 struct mlx5_core_health *health = &dev->priv.health;
114 struct mlx5_health_buffer __iomem *h = health->health;
115 u32 rfr = ioread32be(&h->rfr) >> MLX5_RFR_OFFSET;
116 u8 synd = ioread8(&h->synd);
119 mlx5_core_dbg(dev, "FW requests reset, synd: %d\n", synd);
123 static void mlx5_trigger_cmd_completions(struct mlx5_core_dev *dev)
128 /* wait for pending handlers to complete */
129 synchronize_irq(dev->priv.msix_arr[MLX5_EQ_VEC_CMD].vector);
130 spin_lock_irqsave(&dev->cmd.alloc_lock, flags);
131 vector = ~dev->cmd.bitmask & ((1ul << (1 << dev->cmd.log_sz)) - 1);
135 vector |= MLX5_TRIGGERED_CMD_COMP;
136 spin_unlock_irqrestore(&dev->cmd.alloc_lock, flags);
138 mlx5_core_dbg(dev, "vector 0x%jx\n", (uintmax_t)vector);
139 mlx5_cmd_comp_handler(dev, vector);
143 spin_unlock_irqrestore(&dev->cmd.alloc_lock, flags);
146 static bool sensor_pci_no_comm(struct mlx5_core_dev *dev)
148 struct mlx5_core_health *health = &dev->priv.health;
149 struct mlx5_health_buffer __iomem *h = health->health;
150 bool err = ioread32be(&h->fw_ver) == 0xffffffff;
155 static bool sensor_nic_disabled(struct mlx5_core_dev *dev)
157 return get_nic_mode(dev) == MLX5_NIC_IFC_DISABLED;
160 static bool sensor_nic_sw_reset(struct mlx5_core_dev *dev)
162 return get_nic_mode(dev) == MLX5_NIC_IFC_SW_RESET;
165 static u32 check_fatal_sensors(struct mlx5_core_dev *dev)
167 if (sensor_pci_no_comm(dev))
168 return MLX5_SENSOR_PCI_COMM_ERR;
169 if (pci_channel_offline(dev->pdev))
170 return MLX5_SENSOR_PCI_ERR;
171 if (sensor_nic_disabled(dev))
172 return MLX5_SENSOR_NIC_DISABLED;
173 if (sensor_nic_sw_reset(dev))
174 return MLX5_SENSOR_NIC_SW_RESET;
175 if (sensor_fw_synd_rfr(dev))
176 return MLX5_SENSOR_FW_SYND_RFR;
178 return MLX5_SENSOR_NO_ERR;
181 static void reset_fw_if_needed(struct mlx5_core_dev *dev)
183 bool supported = (ioread32be(&dev->iseg->initializing) >>
184 MLX5_FW_RESET_SUPPORTED_OFFSET) & 1;
185 u32 cmdq_addr, fatal_error;
190 /* The reset only needs to be issued by one PF. The health buffer is
191 * shared between all functions, and will be cleared during a reset.
192 * Check again to avoid a redundant 2nd reset. If the fatal erros was
193 * PCI related a reset won't help.
195 fatal_error = check_fatal_sensors(dev);
196 if (fatal_error == MLX5_SENSOR_PCI_COMM_ERR ||
197 fatal_error == MLX5_SENSOR_NIC_DISABLED ||
198 fatal_error == MLX5_SENSOR_NIC_SW_RESET) {
199 mlx5_core_warn(dev, "Not issuing FW reset. Either it's already done or won't help.\n");
203 mlx5_core_warn(dev, "Issuing FW Reset\n");
204 /* Write the NIC interface field to initiate the reset, the command
205 * interface address also resides here, don't overwrite it.
207 cmdq_addr = ioread32be(&dev->iseg->cmdq_addr_l_sz);
208 iowrite32be((cmdq_addr & 0xFFFFF000) |
209 MLX5_NIC_IFC_SW_RESET << MLX5_NIC_IFC_OFFSET,
210 &dev->iseg->cmdq_addr_l_sz);
213 #define MLX5_CRDUMP_WAIT_MS 60000
214 #define MLX5_FW_RESET_WAIT_MS 1000
215 #define MLX5_NIC_STATE_POLL_MS 5
216 void mlx5_enter_error_state(struct mlx5_core_dev *dev, bool force)
218 unsigned long end, delay_ms = MLX5_CRDUMP_WAIT_MS;
222 fatal_error = check_fatal_sensors(dev);
224 if (fatal_error || force) {
225 if (xchg(&dev->state, MLX5_DEVICE_STATE_INTERNAL_ERROR) ==
226 MLX5_DEVICE_STATE_INTERNAL_ERROR)
229 mlx5_core_err(dev, "internal state error detected\n");
230 mlx5_trigger_cmd_completions(dev);
233 mutex_lock(&dev->intf_state_mutex);
238 if (fatal_error == MLX5_SENSOR_FW_SYND_RFR) {
239 /* Get cr-dump and reset FW semaphore */
240 if (mlx5_core_is_pf(dev))
241 lock = lock_sem_sw_reset(dev);
243 /* Execute cr-dump and SW reset */
244 if (lock != -EBUSY) {
246 reset_fw_if_needed(dev);
247 delay_ms = MLX5_FW_RESET_WAIT_MS;
251 /* Recover from SW reset */
252 end = jiffies + msecs_to_jiffies(delay_ms);
254 if (sensor_nic_disabled(dev))
257 msleep(MLX5_NIC_STATE_POLL_MS);
258 } while (!time_after(jiffies, end));
260 if (!sensor_nic_disabled(dev)) {
261 dev_err(&dev->pdev->dev, "NIC IFC still %d after %lums.\n",
262 get_nic_mode(dev), delay_ms);
265 /* Release FW semaphore if you are the lock owner */
267 unlock_sem_sw_reset(dev);
269 mlx5_core_err(dev, "system error event triggered\n");
272 mlx5_core_event(dev, MLX5_DEV_EVENT_SYS_ERROR, 0);
273 mutex_unlock(&dev->intf_state_mutex);
276 static void mlx5_handle_bad_state(struct mlx5_core_dev *dev)
278 u8 nic_mode = get_nic_mode(dev);
280 if (nic_mode == MLX5_NIC_IFC_SW_RESET) {
281 /* The IFC mode field is 3 bits, so it will read 0x7 in two cases:
282 * 1. PCI has been disabled (ie. PCI-AER, PF driver unloaded
283 * and this is a VF), this is not recoverable by SW reset.
284 * Logging of this is handled elsewhere.
285 * 2. FW reset has been issued by another function, driver can
286 * be reloaded to recover after the mode switches to
287 * MLX5_NIC_IFC_DISABLED.
289 if (dev->priv.health.fatal_error != MLX5_SENSOR_PCI_COMM_ERR)
290 mlx5_core_warn(dev, "NIC SW reset is already progress\n");
292 mlx5_core_warn(dev, "Communication with FW over the PCI link is down\n");
294 mlx5_core_warn(dev, "NIC mode %d\n", nic_mode);
297 mlx5_disable_device(dev);
300 #define MLX5_FW_RESET_WAIT_MS 1000
301 #define MLX5_NIC_STATE_POLL_MS 5
302 static void health_recover(struct work_struct *work)
304 unsigned long end = jiffies + msecs_to_jiffies(MLX5_FW_RESET_WAIT_MS);
305 struct mlx5_core_health *health;
306 struct delayed_work *dwork;
307 struct mlx5_core_dev *dev;
308 struct mlx5_priv *priv;
312 dwork = container_of(work, struct delayed_work, work);
313 health = container_of(dwork, struct mlx5_core_health, recover_work);
314 priv = container_of(health, struct mlx5_priv, health);
315 dev = container_of(priv, struct mlx5_core_dev, priv);
317 mtx_lock(&Giant); /* XXX newbus needs this */
319 if (sensor_pci_no_comm(dev)) {
320 dev_err(&dev->pdev->dev, "health recovery flow aborted, PCI reads still not working\n");
324 nic_mode = get_nic_mode(dev);
325 while (nic_mode != MLX5_NIC_IFC_DISABLED &&
326 !time_after(jiffies, end)) {
327 msleep(MLX5_NIC_STATE_POLL_MS);
328 nic_mode = get_nic_mode(dev);
331 if (nic_mode != MLX5_NIC_IFC_DISABLED) {
332 dev_err(&dev->pdev->dev, "health recovery flow aborted, unexpected NIC IFC mode %d.\n",
338 dev_err(&dev->pdev->dev, "starting health recovery flow\n");
339 mlx5_recover_device(dev);
345 /* How much time to wait until health resetting the driver (in msecs) */
346 #define MLX5_RECOVERY_DELAY_MSECS 60000
347 #define MLX5_RECOVERY_NO_DELAY 0
348 static unsigned long get_recovery_delay(struct mlx5_core_dev *dev)
350 return dev->priv.health.fatal_error == MLX5_SENSOR_PCI_ERR ||
351 dev->priv.health.fatal_error == MLX5_SENSOR_PCI_COMM_ERR ?
352 MLX5_RECOVERY_DELAY_MSECS : MLX5_RECOVERY_NO_DELAY;
355 static void health_care(struct work_struct *work)
357 struct mlx5_core_health *health;
358 unsigned long recover_delay;
359 struct mlx5_core_dev *dev;
360 struct mlx5_priv *priv;
363 health = container_of(work, struct mlx5_core_health, work);
364 priv = container_of(health, struct mlx5_priv, health);
365 dev = container_of(priv, struct mlx5_core_dev, priv);
367 mlx5_core_warn(dev, "handling bad device here\n");
368 mlx5_handle_bad_state(dev);
369 recover_delay = msecs_to_jiffies(get_recovery_delay(dev));
371 spin_lock_irqsave(&health->wq_lock, flags);
372 if (!test_bit(MLX5_DROP_NEW_RECOVERY_WORK, &health->flags)) {
373 mlx5_core_warn(dev, "Scheduling recovery work with %lums delay\n",
375 schedule_delayed_work(&health->recover_work, recover_delay);
377 dev_err(&dev->pdev->dev,
378 "new health works are not permitted at this stage\n");
380 spin_unlock_irqrestore(&health->wq_lock, flags);
383 static int get_next_poll_jiffies(void)
387 get_random_bytes(&next, sizeof(next));
389 next += jiffies + MLX5_HEALTH_POLL_INTERVAL;
394 void mlx5_trigger_health_work(struct mlx5_core_dev *dev)
396 struct mlx5_core_health *health = &dev->priv.health;
399 spin_lock_irqsave(&health->wq_lock, flags);
400 if (!test_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags))
401 queue_work(health->wq, &health->work);
403 dev_err(&dev->pdev->dev,
404 "new health works are not permitted at this stage\n");
405 spin_unlock_irqrestore(&health->wq_lock, flags);
408 static const char *hsynd_str(u8 synd)
411 case MLX5_HEALTH_SYNDR_FW_ERR:
412 return "firmware internal error";
413 case MLX5_HEALTH_SYNDR_IRISC_ERR:
414 return "irisc not responding";
415 case MLX5_HEALTH_SYNDR_HW_UNRECOVERABLE_ERR:
416 return "unrecoverable hardware error";
417 case MLX5_HEALTH_SYNDR_CRC_ERR:
418 return "firmware CRC error";
419 case MLX5_HEALTH_SYNDR_FETCH_PCI_ERR:
420 return "ICM fetch PCI error";
421 case MLX5_HEALTH_SYNDR_HW_FTL_ERR:
422 return "HW fatal error\n";
423 case MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR:
424 return "async EQ buffer overrun";
425 case MLX5_HEALTH_SYNDR_EQ_ERR:
427 case MLX5_HEALTH_SYNDR_EQ_INV:
428 return "Invalid EQ referenced";
429 case MLX5_HEALTH_SYNDR_FFSER_ERR:
430 return "FFSER error";
431 case MLX5_HEALTH_SYNDR_HIGH_TEMP:
432 return "High temprature";
434 return "unrecognized error";
438 static void print_health_info(struct mlx5_core_dev *dev)
440 struct mlx5_core_health *health = &dev->priv.health;
441 struct mlx5_health_buffer __iomem *h = health->health;
446 /* If the syndrom is 0, the device is OK and no need to print buffer */
447 if (!ioread8(&h->synd))
450 for (i = 0; i < ARRAY_SIZE(h->assert_var); i++)
451 printf("mlx5_core: INFO: ""assert_var[%d] 0x%08x\n", i, ioread32be(h->assert_var + i));
453 printf("mlx5_core: INFO: ""assert_exit_ptr 0x%08x\n", ioread32be(&h->assert_exit_ptr));
454 printf("mlx5_core: INFO: ""assert_callra 0x%08x\n", ioread32be(&h->assert_callra));
455 snprintf(fw_str, sizeof(fw_str), "%d.%d.%d", fw_rev_maj(dev), fw_rev_min(dev), fw_rev_sub(dev));
456 printf("mlx5_core: INFO: ""fw_ver %s\n", fw_str);
457 printf("mlx5_core: INFO: ""hw_id 0x%08x\n", ioread32be(&h->hw_id));
458 printf("mlx5_core: INFO: ""irisc_index %d\n", ioread8(&h->irisc_index));
459 printf("mlx5_core: INFO: ""synd 0x%x: %s\n", ioread8(&h->synd), hsynd_str(ioread8(&h->synd)));
460 printf("mlx5_core: INFO: ""ext_synd 0x%04x\n", ioread16be(&h->ext_synd));
461 fw = ioread32be(&h->fw_ver);
462 printf("mlx5_core: INFO: ""raw fw_ver 0x%08x\n", fw);
465 static void poll_health(unsigned long data)
467 struct mlx5_core_dev *dev = (struct mlx5_core_dev *)data;
468 struct mlx5_core_health *health = &dev->priv.health;
472 if (dev->state != MLX5_DEVICE_STATE_UP)
475 if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)
478 count = ioread32be(health->health_counter);
479 if (count == health->prev)
480 ++health->miss_counter;
482 health->miss_counter = 0;
484 health->prev = count;
485 if (health->miss_counter == MAX_MISSES) {
486 mlx5_core_err(dev, "device's health compromised - reached miss count\n");
487 print_health_info(dev);
490 fatal_error = check_fatal_sensors(dev);
492 if (fatal_error && !health->fatal_error) {
493 mlx5_core_err(dev, "Fatal error %u detected\n", fatal_error);
494 dev->priv.health.fatal_error = fatal_error;
495 print_health_info(dev);
496 mlx5_trigger_health_work(dev);
500 mod_timer(&health->timer, get_next_poll_jiffies());
503 void mlx5_start_health_poll(struct mlx5_core_dev *dev)
505 struct mlx5_core_health *health = &dev->priv.health;
507 init_timer(&health->timer);
508 health->fatal_error = MLX5_SENSOR_NO_ERR;
509 clear_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
510 clear_bit(MLX5_DROP_NEW_RECOVERY_WORK, &health->flags);
511 health->health = &dev->iseg->health;
512 health->health_counter = &dev->iseg->health_counter;
514 setup_timer(&health->timer, poll_health, (unsigned long)dev);
515 mod_timer(&health->timer,
516 round_jiffies(jiffies + MLX5_HEALTH_POLL_INTERVAL));
519 void mlx5_stop_health_poll(struct mlx5_core_dev *dev)
521 struct mlx5_core_health *health = &dev->priv.health;
523 del_timer_sync(&health->timer);
526 void mlx5_drain_health_wq(struct mlx5_core_dev *dev)
528 struct mlx5_core_health *health = &dev->priv.health;
531 spin_lock_irqsave(&health->wq_lock, flags);
532 set_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
533 set_bit(MLX5_DROP_NEW_RECOVERY_WORK, &health->flags);
534 spin_unlock_irqrestore(&health->wq_lock, flags);
535 cancel_delayed_work_sync(&health->recover_work);
536 cancel_work_sync(&health->work);
539 void mlx5_drain_health_recovery(struct mlx5_core_dev *dev)
541 struct mlx5_core_health *health = &dev->priv.health;
544 spin_lock_irqsave(&health->wq_lock, flags);
545 set_bit(MLX5_DROP_NEW_RECOVERY_WORK, &health->flags);
546 spin_unlock_irqrestore(&health->wq_lock, flags);
547 cancel_delayed_work_sync(&dev->priv.health.recover_work);
550 void mlx5_health_cleanup(struct mlx5_core_dev *dev)
552 struct mlx5_core_health *health = &dev->priv.health;
554 destroy_workqueue(health->wq);
557 #define HEALTH_NAME "mlx5_health"
558 int mlx5_health_init(struct mlx5_core_dev *dev)
560 struct mlx5_core_health *health;
564 health = &dev->priv.health;
565 len = strlen(HEALTH_NAME) + strlen(dev_name(&dev->pdev->dev));
566 name = kmalloc(len + 1, GFP_KERNEL);
570 snprintf(name, len, "%s:%s", HEALTH_NAME, dev_name(&dev->pdev->dev));
571 health->wq = create_singlethread_workqueue(name);
576 spin_lock_init(&health->wq_lock);
577 INIT_WORK(&health->work, health_care);
578 INIT_DELAYED_WORK(&health->recover_work, health_recover);