arch/powerpc/platforms/powernv/opal-memory-errors.c

   1 /*
   2  * OPAL asynchronus Memory error handling support in PowreNV.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  17  *
  18  * Copyright 2013 IBM Corporation
  19  * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
  20  */
  21
  22 #undef DEBUG
  23
  24 #include <linux/kernel.h>
  25 #include <linux/init.h>
  26 #include <linux/of.h>
  27 #include <linux/mm.h>
  28 #include <linux/slab.h>
  29
  30 #include <asm/opal.h>
  31 #include <asm/cputable.h>
  32
  33 static int opal_mem_err_nb_init;
  34 static LIST_HEAD(opal_memory_err_list);
  35 static DEFINE_SPINLOCK(opal_mem_err_lock);
  36
  37 struct OpalMsgNode {
  38         struct list_head list;
  39         struct opal_msg msg;
  40 };
  41
  42 static void handle_memory_error_event(struct OpalMemoryErrorData *merr_evt)
  43 {
  44         uint64_t paddr_start, paddr_end;
  45
  46         pr_debug("%s: Retrived memory error event, type: 0x%x\n",
  47                   __func__, merr_evt->type);
  48         switch (merr_evt->type) {
  49         case OPAL_MEM_ERR_TYPE_RESILIENCE:
  50                 paddr_start = merr_evt->u.resilience.physical_address_start;
  51                 paddr_end = merr_evt->u.resilience.physical_address_end;
  52                 break;
  53         case OPAL_MEM_ERR_TYPE_DYN_DALLOC:
  54                 paddr_start = merr_evt->u.dyn_dealloc.physical_address_start;
  55                 paddr_end = merr_evt->u.dyn_dealloc.physical_address_end;
  56                 break;
  57         default:
  58                 return;
  59         }
  60
  61         for (; paddr_start < paddr_end; paddr_start += PAGE_SIZE) {
  62                 memory_failure(paddr_start >> PAGE_SHIFT, 0, 0);
  63         }
  64 }
  65
  66 static void handle_memory_error(void)
  67 {
  68         unsigned long flags;
  69         struct OpalMemoryErrorData *merr_evt;
  70         struct OpalMsgNode *msg_node;
  71
  72         spin_lock_irqsave(&opal_mem_err_lock, flags);
  73         while (!list_empty(&opal_memory_err_list)) {
  74                  msg_node = list_entry(opal_memory_err_list.next,
  75                                            struct OpalMsgNode, list);
  76                 list_del(&msg_node->list);
  77                 spin_unlock_irqrestore(&opal_mem_err_lock, flags);
  78
  79                 merr_evt = (struct OpalMemoryErrorData *)
  80                                         &msg_node->msg.params[0];
  81                 handle_memory_error_event(merr_evt);
  82                 kfree(msg_node);
  83                 spin_lock_irqsave(&opal_mem_err_lock, flags);
  84         }
  85         spin_unlock_irqrestore(&opal_mem_err_lock, flags);
  86 }
  87
  88 static void mem_error_handler(struct work_struct *work)
  89 {
  90         handle_memory_error();
  91 }
  92
  93 static DECLARE_WORK(mem_error_work, mem_error_handler);
  94
  95 /*
  96  * opal_memory_err_event - notifier handler that queues up the opal message
  97  * to be preocessed later.
  98  */
  99 static int opal_memory_err_event(struct notifier_block *nb,
 100                           unsigned long msg_type, void *msg)
 101 {
 102         unsigned long flags;
 103         struct OpalMsgNode *msg_node;
 104
 105         if (msg_type != OPAL_MSG_MEM_ERR)
 106                 return 0;
 107
 108         msg_node = kzalloc(sizeof(*msg_node), GFP_ATOMIC);
 109         if (!msg_node) {
 110                 pr_err("MEMORY_ERROR: out of memory, Opal message event not"
 111                        "handled\n");
 112                 return -ENOMEM;
 113         }
 114         memcpy(&msg_node->msg, msg, sizeof(struct opal_msg));
 115
 116         spin_lock_irqsave(&opal_mem_err_lock, flags);
 117         list_add(&msg_node->list, &opal_memory_err_list);
 118         spin_unlock_irqrestore(&opal_mem_err_lock, flags);
 119
 120         schedule_work(&mem_error_work);
 121         return 0;
 122 }
 123
 124 static struct notifier_block opal_mem_err_nb = {
 125         .notifier_call  = opal_memory_err_event,
 126         .next           = NULL,
 127         .priority       = 0,
 128 };
 129
 130 static int __init opal_mem_err_init(void)
 131 {
 132         int ret;
 133
 134         if (!opal_mem_err_nb_init) {
 135                 ret = opal_message_notifier_register(
 136                                         OPAL_MSG_MEM_ERR, &opal_mem_err_nb);
 137                 if (ret) {
 138                         pr_err("%s: Can't register OPAL event notifier (%d)\n",
 139                                __func__, ret);
 140                         return ret;
 141                 }
 142                 opal_mem_err_nb_init = 1;
 143         }
 144         return 0;
 145 }
 146 subsys_initcall(opal_mem_err_init);