sys/dev/mlx5/mlx5_core/mlx5_health.c

   1 /*-
   2  * Copyright (c) 2013-2015, Mellanox Technologies, Ltd.  All rights reserved.
   3  *
   4  * Redistribution and use in source and binary forms, with or without
   5  * modification, are permitted provided that the following conditions
   6  * are met:
   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.
  12  *
  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
  23  * SUCH DAMAGE.
  24  *
  25  * $FreeBSD$
  26  */
  27
  28 #include <linux/kernel.h>
  29 #include <linux/module.h>
  30 #include <linux/random.h>
  31 #include <linux/vmalloc.h>
  32 #include <dev/mlx5/driver.h>
  33 #include <dev/mlx5/mlx5_ifc.h>
  34 #include "mlx5_core.h"
  35
  36 #define MLX5_HEALTH_POLL_INTERVAL       (2 * HZ)
  37 #define MAX_MISSES                      3
  38
  39 static DEFINE_SPINLOCK(health_lock);
  40 static LIST_HEAD(health_list);
  41 static struct work_struct health_work;
  42
  43 static void health_care(struct work_struct *work)
  44 {
  45         struct mlx5_core_health *health, *n;
  46         struct mlx5_core_dev *dev;
  47         struct mlx5_priv *priv;
  48         LIST_HEAD(tlist);
  49
  50         spin_lock_irq(&health_lock);
  51         list_splice_init(&health_list, &tlist);
  52
  53         spin_unlock_irq(&health_lock);
  54
  55         list_for_each_entry_safe(health, n, &tlist, list) {
  56                 priv = container_of(health, struct mlx5_priv, health);
  57                 dev = container_of(priv, struct mlx5_core_dev, priv);
  58                 mlx5_core_warn(dev, "handling bad device here\n");
  59
  60                 spin_lock_irq(&health_lock);
  61                 list_del_init(&health->list);
  62                 spin_unlock_irq(&health_lock);
  63
  64                 /* enter error state */
  65                 mlx5_enter_error_state(dev);
  66         }
  67 }
  68
  69 static const char *hsynd_str(u8 synd)
  70 {
  71         switch (synd) {
  72         case MLX5_HEALTH_SYNDR_FW_ERR:
  73                 return "firmware internal error";
  74         case MLX5_HEALTH_SYNDR_IRISC_ERR:
  75                 return "irisc not responding";
  76         case MLX5_HEALTH_SYNDR_CRC_ERR:
  77                 return "firmware CRC error";
  78         case MLX5_HEALTH_SYNDR_FETCH_PCI_ERR:
  79                 return "ICM fetch PCI error";
  80         case MLX5_HEALTH_SYNDR_HW_FTL_ERR:
  81                 return "HW fatal error\n";
  82         case MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR:
  83                 return "async EQ buffer overrun";
  84         case MLX5_HEALTH_SYNDR_EQ_ERR:
  85                 return "EQ error";
  86         case MLX5_HEALTH_SYNDR_FFSER_ERR:
  87                 return "FFSER error";
  88         default:
  89                 return "unrecognized error";
  90         }
  91 }
  92
  93 static u16 read_be16(__be16 __iomem *p)
  94 {
  95         return swab16(readl((__force u16 __iomem *) p));
  96 }
  97
  98 static u32 read_be32(__be32 __iomem *p)
  99 {
 100         return swab32(readl((__force u32 __iomem *) p));
 101 }
 102
 103 static void print_health_info(struct mlx5_core_dev *dev)
 104 {
 105         struct mlx5_core_health *health = &dev->priv.health;
 106         struct mlx5_health_buffer __iomem *h = health->health;
 107         int i;
 108
 109         for (i = 0; i < ARRAY_SIZE(h->assert_var); i++)
 110                 printf("mlx5_core: INFO: ""assert_var[%d] 0x%08x\n", i, read_be32(h->assert_var + i));
 111
 112         printf("mlx5_core: INFO: ""assert_exit_ptr 0x%08x\n", read_be32(&h->assert_exit_ptr));
 113         printf("mlx5_core: INFO: ""assert_callra 0x%08x\n", read_be32(&h->assert_callra));
 114         printf("mlx5_core: INFO: ""fw_ver 0x%08x\n", read_be32(&h->fw_ver));
 115         printf("mlx5_core: INFO: ""hw_id 0x%08x\n", read_be32(&h->hw_id));
 116         printf("mlx5_core: INFO: ""irisc_index %d\n", readb(&h->irisc_index));
 117         printf("mlx5_core: INFO: ""synd 0x%x: %s\n", readb(&h->synd), hsynd_str(readb(&h->synd)));
 118         printf("mlx5_core: INFO: ""ext_sync 0x%04x\n", read_be16(&h->ext_sync));
 119 }
 120
 121 static void poll_health(unsigned long data)
 122 {
 123         struct mlx5_core_dev *dev = (struct mlx5_core_dev *)data;
 124         struct mlx5_core_health *health = &dev->priv.health;
 125         int next;
 126         u32 count;
 127
 128         if (dev->state != MLX5_DEVICE_STATE_UP)
 129                 return;
 130
 131         count = ioread32be(health->health_counter);
 132         if (count == health->prev)
 133                 ++health->miss_counter;
 134         else
 135                 health->miss_counter = 0;
 136
 137         health->prev = count;
 138         if (health->miss_counter == MAX_MISSES) {
 139                 mlx5_core_err(dev, "device's health compromised\n");
 140                 print_health_info(dev);
 141                 spin_lock_irq(&health_lock);
 142                 list_add_tail(&health->list, &health_list);
 143                 spin_unlock_irq(&health_lock);
 144
 145                 if (!queue_work(mlx5_core_wq, &health_work))
 146                         mlx5_core_warn(dev, "failed to queue health work\n");
 147         } else {
 148                 get_random_bytes(&next, sizeof(next));
 149                 next %= HZ;
 150                 next += jiffies + MLX5_HEALTH_POLL_INTERVAL;
 151                 mod_timer(&health->timer, next);
 152         }
 153 }
 154
 155 void mlx5_start_health_poll(struct mlx5_core_dev *dev)
 156 {
 157         struct mlx5_core_health *health = &dev->priv.health;
 158
 159         INIT_LIST_HEAD(&health->list);
 160         init_timer(&health->timer);
 161         health->health = &dev->iseg->health;
 162         health->health_counter = &dev->iseg->health_counter;
 163
 164         setup_timer(&health->timer, poll_health, (unsigned long)dev);
 165         mod_timer(&health->timer,
 166                   round_jiffies(jiffies + MLX5_HEALTH_POLL_INTERVAL));
 167 }
 168
 169 void mlx5_stop_health_poll(struct mlx5_core_dev *dev)
 170 {
 171         struct mlx5_core_health *health = &dev->priv.health;
 172
 173         del_timer_sync(&health->timer);
 174
 175         spin_lock_irq(&health_lock);
 176         if (!list_empty(&health->list))
 177                 list_del_init(&health->list);
 178         spin_unlock_irq(&health_lock);
 179 }
 180
 181 void mlx5_health_cleanup(void)
 182 {
 183 }
 184
 185 void  __init mlx5_health_init(void)
 186 {
 187
 188         INIT_WORK(&health_work, health_care);
 189 }