drivers/net/ethernet/mellanox/mlx5/core/health.c

   1 /*
   2  * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
   3  *
   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:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  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.
  22  *
  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
  30  * SOFTWARE.
  31  */
  32
  33 #include <linux/kernel.h>
  34 #include <linux/module.h>
  35 #include <linux/random.h>
  36 #include <linux/vmalloc.h>
  37 #include <linux/mlx5/driver.h>
  38 #include <linux/mlx5/cmd.h>
  39 #include "mlx5_core.h"
  40
  41 enum {
  42         MLX5_HEALTH_POLL_INTERVAL       = 2 * HZ,
  43         MAX_MISSES                      = 3,
  44 };
  45
  46 enum {
  47         MLX5_HEALTH_SYNDR_FW_ERR                = 0x1,
  48         MLX5_HEALTH_SYNDR_IRISC_ERR             = 0x7,
  49         MLX5_HEALTH_SYNDR_HW_UNRECOVERABLE_ERR  = 0x8,
  50         MLX5_HEALTH_SYNDR_CRC_ERR               = 0x9,
  51         MLX5_HEALTH_SYNDR_FETCH_PCI_ERR         = 0xa,
  52         MLX5_HEALTH_SYNDR_HW_FTL_ERR            = 0xb,
  53         MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR  = 0xc,
  54         MLX5_HEALTH_SYNDR_EQ_ERR                = 0xd,
  55         MLX5_HEALTH_SYNDR_EQ_INV                = 0xe,
  56         MLX5_HEALTH_SYNDR_FFSER_ERR             = 0xf,
  57         MLX5_HEALTH_SYNDR_HIGH_TEMP             = 0x10
  58 };
  59
  60 static void health_care(struct work_struct *work)
  61 {
  62         struct mlx5_core_health *health;
  63         struct mlx5_core_dev *dev;
  64         struct mlx5_priv *priv;
  65
  66         health = container_of(work, struct mlx5_core_health, work);
  67         priv = container_of(health, struct mlx5_priv, health);
  68         dev = container_of(priv, struct mlx5_core_dev, priv);
  69         mlx5_core_warn(dev, "handling bad device here\n");
  70 }
  71
  72 static const char *hsynd_str(u8 synd)
  73 {
  74         switch (synd) {
  75         case MLX5_HEALTH_SYNDR_FW_ERR:
  76                 return "firmware internal error";
  77         case MLX5_HEALTH_SYNDR_IRISC_ERR:
  78                 return "irisc not responding";
  79         case MLX5_HEALTH_SYNDR_HW_UNRECOVERABLE_ERR:
  80                 return "unrecoverable hardware error";
  81         case MLX5_HEALTH_SYNDR_CRC_ERR:
  82                 return "firmware CRC error";
  83         case MLX5_HEALTH_SYNDR_FETCH_PCI_ERR:
  84                 return "ICM fetch PCI error";
  85         case MLX5_HEALTH_SYNDR_HW_FTL_ERR:
  86                 return "HW fatal error\n";
  87         case MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR:
  88                 return "async EQ buffer overrun";
  89         case MLX5_HEALTH_SYNDR_EQ_ERR:
  90                 return "EQ error";
  91         case MLX5_HEALTH_SYNDR_EQ_INV:
  92                 return "Invalid EQ refrenced";
  93         case MLX5_HEALTH_SYNDR_FFSER_ERR:
  94                 return "FFSER error";
  95         case MLX5_HEALTH_SYNDR_HIGH_TEMP:
  96                 return "High temprature";
  97         default:
  98                 return "unrecognized error";
  99         }
 100 }
 101
 102 static u16 get_maj(u32 fw)
 103 {
 104         return fw >> 28;
 105 }
 106
 107 static u16 get_min(u32 fw)
 108 {
 109         return fw >> 16 & 0xfff;
 110 }
 111
 112 static u16 get_sub(u32 fw)
 113 {
 114         return fw & 0xffff;
 115 }
 116
 117 static void print_health_info(struct mlx5_core_dev *dev)
 118 {
 119         struct mlx5_core_health *health = &dev->priv.health;
 120         struct health_buffer __iomem *h = health->health;
 121         char fw_str[18];
 122         u32 fw;
 123         int i;
 124
 125         for (i = 0; i < ARRAY_SIZE(h->assert_var); i++)
 126                 dev_err(&dev->pdev->dev, "assert_var[%d] 0x%08x\n", i, ioread32be(h->assert_var + i));
 127
 128         dev_err(&dev->pdev->dev, "assert_exit_ptr 0x%08x\n", ioread32be(&h->assert_exit_ptr));
 129         dev_err(&dev->pdev->dev, "assert_callra 0x%08x\n", ioread32be(&h->assert_callra));
 130         fw = ioread32be(&h->fw_ver);
 131         sprintf(fw_str, "%d.%d.%d", get_maj(fw), get_min(fw), get_sub(fw));
 132         dev_err(&dev->pdev->dev, "fw_ver %s\n", fw_str);
 133         dev_err(&dev->pdev->dev, "hw_id 0x%08x\n", ioread32be(&h->hw_id));
 134         dev_err(&dev->pdev->dev, "irisc_index %d\n", ioread8(&h->irisc_index));
 135         dev_err(&dev->pdev->dev, "synd 0x%x: %s\n", ioread8(&h->synd), hsynd_str(ioread8(&h->synd)));
 136         dev_err(&dev->pdev->dev, "ext_synd 0x%04x\n", ioread16be(&h->ext_synd));
 137 }
 138
 139 static void poll_health(unsigned long data)
 140 {
 141         struct mlx5_core_dev *dev = (struct mlx5_core_dev *)data;
 142         struct mlx5_core_health *health = &dev->priv.health;
 143         unsigned long next;
 144         u32 count;
 145
 146         count = ioread32be(health->health_counter);
 147         if (count == health->prev)
 148                 ++health->miss_counter;
 149         else
 150                 health->miss_counter = 0;
 151
 152         health->prev = count;
 153         if (health->miss_counter == MAX_MISSES) {
 154                 mlx5_core_err(dev, "device's health compromised\n");
 155                 print_health_info(dev);
 156                 queue_work(health->wq, &health->work);
 157         } else {
 158                 get_random_bytes(&next, sizeof(next));
 159                 next %= HZ;
 160                 next += jiffies + MLX5_HEALTH_POLL_INTERVAL;
 161                 mod_timer(&health->timer, next);
 162         }
 163 }
 164
 165 void mlx5_start_health_poll(struct mlx5_core_dev *dev)
 166 {
 167         struct mlx5_core_health *health = &dev->priv.health;
 168
 169         init_timer(&health->timer);
 170         health->health = &dev->iseg->health;
 171         health->health_counter = &dev->iseg->health_counter;
 172
 173         health->timer.data = (unsigned long)dev;
 174         health->timer.function = poll_health;
 175         health->timer.expires = round_jiffies(jiffies + MLX5_HEALTH_POLL_INTERVAL);
 176         add_timer(&health->timer);
 177 }
 178
 179 void mlx5_stop_health_poll(struct mlx5_core_dev *dev)
 180 {
 181         struct mlx5_core_health *health = &dev->priv.health;
 182
 183         del_timer_sync(&health->timer);
 184 }
 185
 186 void mlx5_health_cleanup(struct mlx5_core_dev *dev)
 187 {
 188         struct mlx5_core_health *health = &dev->priv.health;
 189
 190         destroy_workqueue(health->wq);
 191 }
 192
 193 int mlx5_health_init(struct mlx5_core_dev *dev)
 194 {
 195         struct mlx5_core_health *health;
 196         char *name;
 197
 198         health = &dev->priv.health;
 199         name = kmalloc(64, GFP_KERNEL);
 200         if (!name)
 201                 return -ENOMEM;
 202
 203         strcpy(name, "mlx5_health");
 204         strcat(name, dev_name(&dev->pdev->dev));
 205         health->wq = create_singlethread_workqueue(name);
 206         kfree(name);
 207         if (!health->wq)
 208                 return -ENOMEM;
 209
 210         INIT_WORK(&health->work, health_care);
 211
 212         return 0;
 213 }