1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * OPAL asynchronus Memory error handling support in PowerNV.
4 *
5 * Copyright 2013 IBM Corporation
6 * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
7 */
8
9 #undef DEBUG
10
11 #include <linux/kernel.h>
12 #include <linux/init.h>
13 #include <linux/of.h>
14 #include <linux/mm.h>
15 #include <linux/slab.h>
16
17 #include <asm/machdep.h>
18 #include <asm/opal.h>
19 #include <asm/cputable.h>
20
21 static int opal_mem_err_nb_init;
22 static LIST_HEAD(opal_memory_err_list);
23 static DEFINE_SPINLOCK(opal_mem_err_lock);
24
25 struct OpalMsgNode {
26 struct list_head list;
27 struct opal_msg msg;
28 };
29
handle_memory_error_event(struct OpalMemoryErrorData * merr_evt)30 static void handle_memory_error_event(struct OpalMemoryErrorData *merr_evt)
31 {
32 uint64_t paddr_start, paddr_end;
33
34 pr_debug("%s: Retrieved memory error event, type: 0x%x\n",
35 __func__, merr_evt->type);
36 switch (merr_evt->type) {
37 case OPAL_MEM_ERR_TYPE_RESILIENCE:
38 paddr_start = be64_to_cpu(merr_evt->u.resilience.physical_address_start);
39 paddr_end = be64_to_cpu(merr_evt->u.resilience.physical_address_end);
40 break;
41 case OPAL_MEM_ERR_TYPE_DYN_DALLOC:
42 paddr_start = be64_to_cpu(merr_evt->u.dyn_dealloc.physical_address_start);
43 paddr_end = be64_to_cpu(merr_evt->u.dyn_dealloc.physical_address_end);
44 break;
45 default:
46 return;
47 }
48
49 for (; paddr_start < paddr_end; paddr_start += PAGE_SIZE) {
50 memory_failure(paddr_start >> PAGE_SHIFT, 0);
51 }
52 }
53
handle_memory_error(void)54 static void handle_memory_error(void)
55 {
56 unsigned long flags;
57 struct OpalMemoryErrorData *merr_evt;
58 struct OpalMsgNode *msg_node;
59
60 spin_lock_irqsave(&opal_mem_err_lock, flags);
61 while (!list_empty(&opal_memory_err_list)) {
62 msg_node = list_entry(opal_memory_err_list.next,
63 struct OpalMsgNode, list);
64 list_del(&msg_node->list);
65 spin_unlock_irqrestore(&opal_mem_err_lock, flags);
66
67 merr_evt = (struct OpalMemoryErrorData *)
68 &msg_node->msg.params[0];
69 handle_memory_error_event(merr_evt);
70 kfree(msg_node);
71 spin_lock_irqsave(&opal_mem_err_lock, flags);
72 }
73 spin_unlock_irqrestore(&opal_mem_err_lock, flags);
74 }
75
mem_error_handler(struct work_struct * work)76 static void mem_error_handler(struct work_struct *work)
77 {
78 handle_memory_error();
79 }
80
81 static DECLARE_WORK(mem_error_work, mem_error_handler);
82
83 /*
84 * opal_memory_err_event - notifier handler that queues up the opal message
85 * to be processed later.
86 */
opal_memory_err_event(struct notifier_block * nb,unsigned long msg_type,void * msg)87 static int opal_memory_err_event(struct notifier_block *nb,
88 unsigned long msg_type, void *msg)
89 {
90 unsigned long flags;
91 struct OpalMsgNode *msg_node;
92
93 if (msg_type != OPAL_MSG_MEM_ERR)
94 return 0;
95
96 msg_node = kzalloc(sizeof(*msg_node), GFP_ATOMIC);
97 if (!msg_node) {
98 pr_err("MEMORY_ERROR: out of memory, Opal message event not"
99 "handled\n");
100 return -ENOMEM;
101 }
102 memcpy(&msg_node->msg, msg, sizeof(msg_node->msg));
103
104 spin_lock_irqsave(&opal_mem_err_lock, flags);
105 list_add(&msg_node->list, &opal_memory_err_list);
106 spin_unlock_irqrestore(&opal_mem_err_lock, flags);
107
108 schedule_work(&mem_error_work);
109 return 0;
110 }
111
112 static struct notifier_block opal_mem_err_nb = {
113 .notifier_call = opal_memory_err_event,
114 .next = NULL,
115 .priority = 0,
116 };
117
opal_mem_err_init(void)118 static int __init opal_mem_err_init(void)
119 {
120 int ret;
121
122 if (!opal_mem_err_nb_init) {
123 ret = opal_message_notifier_register(
124 OPAL_MSG_MEM_ERR, &opal_mem_err_nb);
125 if (ret) {
126 pr_err("%s: Can't register OPAL event notifier (%d)\n",
127 __func__, ret);
128 return ret;
129 }
130 opal_mem_err_nb_init = 1;
131 }
132 return 0;
133 }
134 machine_device_initcall(powernv, opal_mem_err_init);
135