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/module.h>
35 #include <linux/random.h>
36 #include <linux/vmalloc.h>
37 #include <linux/hardirq.h>
38 #include <linux/mlx5/driver.h>
39 #include <linux/mlx5/cmd.h>
40 #include "mlx5_core.h"
43 #include "lib/pci_vsc.h"
46 MLX5_HEALTH_POLL_INTERVAL = 2 * HZ,
51 MLX5_HEALTH_SYNDR_FW_ERR = 0x1,
52 MLX5_HEALTH_SYNDR_IRISC_ERR = 0x7,
53 MLX5_HEALTH_SYNDR_HW_UNRECOVERABLE_ERR = 0x8,
54 MLX5_HEALTH_SYNDR_CRC_ERR = 0x9,
55 MLX5_HEALTH_SYNDR_FETCH_PCI_ERR = 0xa,
56 MLX5_HEALTH_SYNDR_HW_FTL_ERR = 0xb,
57 MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR = 0xc,
58 MLX5_HEALTH_SYNDR_EQ_ERR = 0xd,
59 MLX5_HEALTH_SYNDR_EQ_INV = 0xe,
60 MLX5_HEALTH_SYNDR_FFSER_ERR = 0xf,
61 MLX5_HEALTH_SYNDR_HIGH_TEMP = 0x10
65 MLX5_DROP_NEW_HEALTH_WORK,
69 MLX5_SENSOR_NO_ERR = 0,
70 MLX5_SENSOR_PCI_COMM_ERR = 1,
71 MLX5_SENSOR_PCI_ERR = 2,
72 MLX5_SENSOR_NIC_DISABLED = 3,
73 MLX5_SENSOR_NIC_SW_RESET = 4,
74 MLX5_SENSOR_FW_SYND_RFR = 5,
77 u8 mlx5_get_nic_state(struct mlx5_core_dev *dev)
79 return (ioread32be(&dev->iseg->cmdq_addr_l_sz) >> 8) & 7;
82 void mlx5_set_nic_state(struct mlx5_core_dev *dev, u8 state)
84 u32 cur_cmdq_addr_l_sz;
86 cur_cmdq_addr_l_sz = ioread32be(&dev->iseg->cmdq_addr_l_sz);
87 iowrite32be((cur_cmdq_addr_l_sz & 0xFFFFF000) |
88 state << MLX5_NIC_IFC_OFFSET,
89 &dev->iseg->cmdq_addr_l_sz);
92 static bool sensor_pci_not_working(struct mlx5_core_dev *dev)
94 struct mlx5_core_health *health = &dev->priv.health;
95 struct health_buffer __iomem *h = health->health;
97 /* Offline PCI reads return 0xffffffff */
98 return (ioread32be(&h->fw_ver) == 0xffffffff);
101 static bool sensor_fw_synd_rfr(struct mlx5_core_dev *dev)
103 struct mlx5_core_health *health = &dev->priv.health;
104 struct health_buffer __iomem *h = health->health;
105 u32 rfr = ioread32be(&h->rfr) >> MLX5_RFR_OFFSET;
106 u8 synd = ioread8(&h->synd);
109 mlx5_core_dbg(dev, "FW requests reset, synd: %d\n", synd);
113 static u32 check_fatal_sensors(struct mlx5_core_dev *dev)
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;
126 return MLX5_SENSOR_NO_ERR;
129 static int lock_sem_sw_reset(struct mlx5_core_dev *dev, bool lock)
131 enum mlx5_vsc_state state;
134 if (!mlx5_core_is_pf(dev))
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.
141 ret = mlx5_vsc_gw_lock(dev);
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
152 ret = mlx5_vsc_sem_set_space(dev, MLX5_SEMAPHORE_SW_RESET, state);
154 mlx5_core_warn(dev, "Failed to lock SW reset semaphore\n");
156 /* Unlock GW access */
157 mlx5_vsc_gw_unlock(dev);
162 static bool reset_fw_if_needed(struct mlx5_core_dev *dev)
164 bool supported = (ioread32be(&dev->iseg->initializing) >>
165 MLX5_FW_RESET_SUPPORTED_OFFSET) & 1;
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 erros was
174 * PCI related a reset won't help.
176 fatal_error = 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.");
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.
188 mlx5_set_nic_state(dev, MLX5_NIC_IFC_SW_RESET);
193 void mlx5_enter_error_state(struct mlx5_core_dev *dev, bool force)
195 mutex_lock(&dev->intf_state_mutex);
196 if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)
198 if (dev->state == MLX5_DEVICE_STATE_UNINITIALIZED) {
199 dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
203 if (check_fatal_sensors(dev) || force) {
204 dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
208 mlx5_notifier_call_chain(dev->priv.events, MLX5_DEV_EVENT_SYS_ERROR, (void *)1);
210 mutex_unlock(&dev->intf_state_mutex);
213 #define MLX5_CRDUMP_WAIT_MS 60000
214 #define MLX5_FW_RESET_WAIT_MS 1000
215 void mlx5_error_sw_reset(struct mlx5_core_dev *dev)
217 unsigned long end, delay_ms = MLX5_FW_RESET_WAIT_MS;
220 mutex_lock(&dev->intf_state_mutex);
221 if (dev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR)
224 mlx5_core_err(dev, "start\n");
226 if (check_fatal_sensors(dev) == MLX5_SENSOR_FW_SYND_RFR) {
227 /* Get cr-dump and reset FW semaphore */
228 lock = lock_sem_sw_reset(dev, true);
230 if (lock == -EBUSY) {
231 delay_ms = MLX5_CRDUMP_WAIT_MS;
232 goto recover_from_sw_reset;
234 /* Execute SW reset */
235 reset_fw_if_needed(dev);
238 recover_from_sw_reset:
239 /* Recover from SW reset */
240 end = jiffies + msecs_to_jiffies(delay_ms);
242 if (mlx5_get_nic_state(dev) == MLX5_NIC_IFC_DISABLED)
246 } while (!time_after(jiffies, end));
248 if (mlx5_get_nic_state(dev) != MLX5_NIC_IFC_DISABLED) {
249 dev_err(&dev->pdev->dev, "NIC IFC still %d after %lums.\n",
250 mlx5_get_nic_state(dev), delay_ms);
253 /* Release FW semaphore if you are the lock owner */
255 lock_sem_sw_reset(dev, false);
257 mlx5_core_err(dev, "end\n");
260 mutex_unlock(&dev->intf_state_mutex);
263 static void mlx5_handle_bad_state(struct mlx5_core_dev *dev)
265 u8 nic_interface = mlx5_get_nic_state(dev);
267 switch (nic_interface) {
268 case MLX5_NIC_IFC_FULL:
269 mlx5_core_warn(dev, "Expected to see disabled NIC but it is full driver\n");
272 case MLX5_NIC_IFC_DISABLED:
273 mlx5_core_warn(dev, "starting teardown\n");
276 case MLX5_NIC_IFC_NO_DRAM_NIC:
277 mlx5_core_warn(dev, "Expected to see disabled NIC but it is no dram nic\n");
280 case MLX5_NIC_IFC_SW_RESET:
281 /* The IFC mode field is 3 bits, so it will read 0x7 in 2 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 in progress\n");
294 mlx5_core_warn(dev, "Expected to see disabled NIC but it is has invalid value %d\n",
298 mlx5_disable_device(dev);
301 /* How much time to wait until health resetting the driver (in msecs) */
302 #define MLX5_RECOVERY_WAIT_MSECS 60000
303 static void health_care(struct work_struct *work)
305 struct mlx5_core_health *health;
306 struct mlx5_core_dev *dev;
307 struct mlx5_priv *priv;
310 health = container_of(work, struct mlx5_core_health, work);
311 priv = container_of(health, struct mlx5_priv, health);
312 dev = container_of(priv, struct mlx5_core_dev, priv);
313 mlx5_core_warn(dev, "handling bad device here\n");
314 mlx5_handle_bad_state(dev);
316 end = jiffies + msecs_to_jiffies(MLX5_RECOVERY_WAIT_MSECS);
317 while (sensor_pci_not_working(dev)) {
318 if (time_after(jiffies, end)) {
320 "health recovery flow aborted, PCI reads still not working\n");
326 mlx5_core_err(dev, "starting health recovery flow\n");
327 mlx5_recover_device(dev);
330 static const char *hsynd_str(u8 synd)
333 case MLX5_HEALTH_SYNDR_FW_ERR:
334 return "firmware internal error";
335 case MLX5_HEALTH_SYNDR_IRISC_ERR:
336 return "irisc not responding";
337 case MLX5_HEALTH_SYNDR_HW_UNRECOVERABLE_ERR:
338 return "unrecoverable hardware error";
339 case MLX5_HEALTH_SYNDR_CRC_ERR:
340 return "firmware CRC error";
341 case MLX5_HEALTH_SYNDR_FETCH_PCI_ERR:
342 return "ICM fetch PCI error";
343 case MLX5_HEALTH_SYNDR_HW_FTL_ERR:
344 return "HW fatal error\n";
345 case MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR:
346 return "async EQ buffer overrun";
347 case MLX5_HEALTH_SYNDR_EQ_ERR:
349 case MLX5_HEALTH_SYNDR_EQ_INV:
350 return "Invalid EQ referenced";
351 case MLX5_HEALTH_SYNDR_FFSER_ERR:
352 return "FFSER error";
353 case MLX5_HEALTH_SYNDR_HIGH_TEMP:
354 return "High temperature";
356 return "unrecognized error";
360 static void print_health_info(struct mlx5_core_dev *dev)
362 struct mlx5_core_health *health = &dev->priv.health;
363 struct health_buffer __iomem *h = health->health;
368 /* If the syndrome is 0, the device is OK and no need to print buffer */
369 if (!ioread8(&h->synd))
372 for (i = 0; i < ARRAY_SIZE(h->assert_var); i++)
373 mlx5_core_err(dev, "assert_var[%d] 0x%08x\n", i,
374 ioread32be(h->assert_var + i));
376 mlx5_core_err(dev, "assert_exit_ptr 0x%08x\n",
377 ioread32be(&h->assert_exit_ptr));
378 mlx5_core_err(dev, "assert_callra 0x%08x\n",
379 ioread32be(&h->assert_callra));
380 sprintf(fw_str, "%d.%d.%d", fw_rev_maj(dev), fw_rev_min(dev), fw_rev_sub(dev));
381 mlx5_core_err(dev, "fw_ver %s\n", fw_str);
382 mlx5_core_err(dev, "hw_id 0x%08x\n", ioread32be(&h->hw_id));
383 mlx5_core_err(dev, "irisc_index %d\n", ioread8(&h->irisc_index));
384 mlx5_core_err(dev, "synd 0x%x: %s\n", ioread8(&h->synd),
385 hsynd_str(ioread8(&h->synd)));
386 mlx5_core_err(dev, "ext_synd 0x%04x\n", ioread16be(&h->ext_synd));
387 fw = ioread32be(&h->fw_ver);
388 mlx5_core_err(dev, "raw fw_ver 0x%08x\n", fw);
391 static unsigned long 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 mlx5_core_err(dev, "new health works are not permitted at this stage\n");
412 spin_unlock_irqrestore(&health->wq_lock, flags);
415 static void poll_health(struct timer_list *t)
417 struct mlx5_core_dev *dev = from_timer(dev, t, priv.health.timer);
418 struct mlx5_core_health *health = &dev->priv.health;
422 if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)
425 count = ioread32be(health->health_counter);
426 if (count == health->prev)
427 ++health->miss_counter;
429 health->miss_counter = 0;
431 health->prev = count;
432 if (health->miss_counter == MAX_MISSES) {
433 mlx5_core_err(dev, "device's health compromised - reached miss count\n");
434 print_health_info(dev);
437 fatal_error = check_fatal_sensors(dev);
439 if (fatal_error && !health->fatal_error) {
440 mlx5_core_err(dev, "Fatal error %u detected\n", fatal_error);
441 dev->priv.health.fatal_error = fatal_error;
442 print_health_info(dev);
443 mlx5_trigger_health_work(dev);
447 mod_timer(&health->timer, get_next_poll_jiffies());
450 void mlx5_start_health_poll(struct mlx5_core_dev *dev)
452 struct mlx5_core_health *health = &dev->priv.health;
454 timer_setup(&health->timer, poll_health, 0);
455 health->fatal_error = MLX5_SENSOR_NO_ERR;
456 clear_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
457 health->health = &dev->iseg->health;
458 health->health_counter = &dev->iseg->health_counter;
460 health->timer.expires = round_jiffies(jiffies + MLX5_HEALTH_POLL_INTERVAL);
461 add_timer(&health->timer);
464 void mlx5_stop_health_poll(struct mlx5_core_dev *dev, bool disable_health)
466 struct mlx5_core_health *health = &dev->priv.health;
469 if (disable_health) {
470 spin_lock_irqsave(&health->wq_lock, flags);
471 set_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
472 spin_unlock_irqrestore(&health->wq_lock, flags);
475 del_timer_sync(&health->timer);
478 void mlx5_drain_health_wq(struct mlx5_core_dev *dev)
480 struct mlx5_core_health *health = &dev->priv.health;
483 spin_lock_irqsave(&health->wq_lock, flags);
484 set_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
485 spin_unlock_irqrestore(&health->wq_lock, flags);
486 cancel_work_sync(&health->work);
489 void mlx5_health_flush(struct mlx5_core_dev *dev)
491 struct mlx5_core_health *health = &dev->priv.health;
493 flush_workqueue(health->wq);
496 void mlx5_health_cleanup(struct mlx5_core_dev *dev)
498 struct mlx5_core_health *health = &dev->priv.health;
500 destroy_workqueue(health->wq);
503 int mlx5_health_init(struct mlx5_core_dev *dev)
505 struct mlx5_core_health *health;
508 health = &dev->priv.health;
509 name = kmalloc(64, GFP_KERNEL);
513 strcpy(name, "mlx5_health");
514 strcat(name, dev_name(dev->device));
515 health->wq = create_singlethread_workqueue(name);
519 spin_lock_init(&health->wq_lock);
520 INIT_WORK(&health->work, health_care);