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

   1 /*
   2  * Copyright (c) 2013, Mellanox Technologies inc.  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_CRC_ERR               = 0x9,
  50         MLX5_HEALTH_SYNDR_FETCH_PCI_ERR         = 0xa,
  51         MLX5_HEALTH_SYNDR_HW_FTL_ERR            = 0xb,
  52         MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR  = 0xc,
  53         MLX5_HEALTH_SYNDR_EQ_ERR                = 0xd,
  54         MLX5_HEALTH_SYNDR_FFSER_ERR             = 0xf,
  55 };
  56
  57 static DEFINE_SPINLOCK(health_lock);
  58
  59 static LIST_HEAD(health_list);
  60 static struct work_struct health_work;
  61
  62 static health_handler_t reg_handler;
  63 int mlx5_register_health_report_handler(health_handler_t handler)
  64 {
  65         spin_lock_irq(&health_lock);
  66         if (reg_handler) {
  67                 spin_unlock_irq(&health_lock);
  68                 return -EEXIST;
  69         }
  70         reg_handler = handler;
  71         spin_unlock_irq(&health_lock);
  72
  73         return 0;
  74 }
  75 EXPORT_SYMBOL(mlx5_register_health_report_handler);
  76
  77 void mlx5_unregister_health_report_handler(void)
  78 {
  79         spin_lock_irq(&health_lock);
  80         reg_handler = NULL;
  81         spin_unlock_irq(&health_lock);
  82 }
  83 EXPORT_SYMBOL(mlx5_unregister_health_report_handler);
  84
  85 static void health_care(struct work_struct *work)
  86 {
  87         struct mlx5_core_health *health, *n;
  88         struct mlx5_core_dev *dev;
  89         struct mlx5_priv *priv;
  90         LIST_HEAD(tlist);
  91
  92         spin_lock_irq(&health_lock);
  93         list_splice_init(&health_list, &tlist);
  94
  95         spin_unlock_irq(&health_lock);
  96
  97         list_for_each_entry_safe(health, n, &tlist, list) {
  98                 priv = container_of(health, struct mlx5_priv, health);
  99                 dev = container_of(priv, struct mlx5_core_dev, priv);
 100                 mlx5_core_warn(dev, "handling bad device here\n");
 101                 spin_lock_irq(&health_lock);
 102                 if (reg_handler)
 103                         reg_handler(dev->pdev, health->health,
 104                                     sizeof(health->health));
 105
 106                 list_del_init(&health->list);
 107                 spin_unlock_irq(&health_lock);
 108         }
 109 }
 110
 111 static const char *hsynd_str(u8 synd)
 112 {
 113         switch (synd) {
 114         case MLX5_HEALTH_SYNDR_FW_ERR:
 115                 return "firmware internal error";
 116         case MLX5_HEALTH_SYNDR_IRISC_ERR:
 117                 return "irisc not responding";
 118         case MLX5_HEALTH_SYNDR_CRC_ERR:
 119                 return "firmware CRC error";
 120         case MLX5_HEALTH_SYNDR_FETCH_PCI_ERR:
 121                 return "ICM fetch PCI error";
 122         case MLX5_HEALTH_SYNDR_HW_FTL_ERR:
 123                 return "HW fatal error\n";
 124         case MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR:
 125                 return "async EQ buffer overrun";
 126         case MLX5_HEALTH_SYNDR_EQ_ERR:
 127                 return "EQ error";
 128         case MLX5_HEALTH_SYNDR_FFSER_ERR:
 129                 return "FFSER error";
 130         default:
 131                 return "unrecognized error";
 132         }
 133 }
 134
 135 static void print_health_info(struct mlx5_core_dev *dev)
 136 {
 137         struct mlx5_core_health *health = &dev->priv.health;
 138         struct health_buffer __iomem *h = health->health;
 139         int i;
 140
 141         for (i = 0; i < ARRAY_SIZE(h->assert_var); i++)
 142                 pr_info("assert_var[%d] 0x%08x\n", i, be32_to_cpu(h->assert_var[i]));
 143
 144         pr_info("assert_exit_ptr 0x%08x\n", be32_to_cpu(h->assert_exit_ptr));
 145         pr_info("assert_callra 0x%08x\n", be32_to_cpu(h->assert_callra));
 146         pr_info("fw_ver 0x%08x\n", be32_to_cpu(h->fw_ver));
 147         pr_info("hw_id 0x%08x\n", be32_to_cpu(h->hw_id));
 148         pr_info("irisc_index %d\n", h->irisc_index);
 149         pr_info("synd 0x%x: %s\n", h->synd, hsynd_str(h->synd));
 150         pr_info("ext_sync 0x%04x\n", be16_to_cpu(h->ext_sync));
 151 }
 152
 153 static void poll_health(unsigned long data)
 154 {
 155         struct mlx5_core_dev *dev = (struct mlx5_core_dev *)data;
 156         struct mlx5_core_health *health = &dev->priv.health;
 157         unsigned long next;
 158         u32 count;
 159
 160         count = ioread32be(health->health_counter);
 161         if (count == health->prev)
 162                 ++health->miss_counter;
 163         else
 164                 health->miss_counter = 0;
 165
 166         health->prev = count;
 167         if (health->miss_counter == MAX_MISSES) {
 168                 mlx5_core_err(dev, "device's health compromised\n");
 169                 print_health_info(dev);
 170                 spin_lock_irq(&health_lock);
 171                 list_add_tail(&health->list, &health_list);
 172                 spin_unlock_irq(&health_lock);
 173
 174                 queue_work(mlx5_core_wq, &health_work);
 175         } else {
 176                 get_random_bytes(&next, sizeof(next));
 177                 next %= HZ;
 178                 next += jiffies + MLX5_HEALTH_POLL_INTERVAL;
 179                 mod_timer(&health->timer, next);
 180         }
 181 }
 182
 183 void mlx5_start_health_poll(struct mlx5_core_dev *dev)
 184 {
 185         struct mlx5_core_health *health = &dev->priv.health;
 186
 187         INIT_LIST_HEAD(&health->list);
 188         init_timer(&health->timer);
 189         health->health = &dev->iseg->health;
 190         health->health_counter = &dev->iseg->health_counter;
 191
 192         health->timer.data = (unsigned long)dev;
 193         health->timer.function = poll_health;
 194         health->timer.expires = round_jiffies(jiffies + MLX5_HEALTH_POLL_INTERVAL);
 195         add_timer(&health->timer);
 196 }
 197
 198 void mlx5_stop_health_poll(struct mlx5_core_dev *dev)
 199 {
 200         struct mlx5_core_health *health = &dev->priv.health;
 201
 202         del_timer_sync(&health->timer);
 203
 204         spin_lock_irq(&health_lock);
 205         if (!list_empty(&health->list))
 206                 list_del_init(&health->list);
 207         spin_unlock_irq(&health_lock);
 208 }
 209
 210 void mlx5_health_cleanup(void)
 211 {
 212 }
 213
 214 void  __init mlx5_health_init(void)
 215 {
 216         INIT_WORK(&health_work, health_care);
 217 }