x86/MCE: Make correctable error detection look at the Deferred bit
[sfrench/cifs-2.6.git] / arch / x86 / kernel / check.c
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/init.h>
3 #include <linux/sched.h>
4 #include <linux/kthread.h>
5 #include <linux/workqueue.h>
6 #include <linux/memblock.h>
7
8 #include <asm/proto.h>
9
10 /*
11  * Some BIOSes seem to corrupt the low 64k of memory during events
12  * like suspend/resume and unplugging an HDMI cable.  Reserve all
13  * remaining free memory in that area and fill it with a distinct
14  * pattern.
15  */
16 #define MAX_SCAN_AREAS  8
17
18 static int __read_mostly memory_corruption_check = -1;
19
20 static unsigned __read_mostly corruption_check_size = 64*1024;
21 static unsigned __read_mostly corruption_check_period = 60; /* seconds */
22
23 static struct scan_area {
24         u64 addr;
25         u64 size;
26 } scan_areas[MAX_SCAN_AREAS];
27 static int num_scan_areas;
28
29 static __init int set_corruption_check(char *arg)
30 {
31         ssize_t ret;
32         unsigned long val;
33
34         ret = kstrtoul(arg, 10, &val);
35         if (ret)
36                 return ret;
37
38         memory_corruption_check = val;
39         return 0;
40 }
41 early_param("memory_corruption_check", set_corruption_check);
42
43 static __init int set_corruption_check_period(char *arg)
44 {
45         ssize_t ret;
46         unsigned long val;
47
48         ret = kstrtoul(arg, 10, &val);
49         if (ret)
50                 return ret;
51
52         corruption_check_period = val;
53         return 0;
54 }
55 early_param("memory_corruption_check_period", set_corruption_check_period);
56
57 static __init int set_corruption_check_size(char *arg)
58 {
59         char *end;
60         unsigned size;
61
62         size = memparse(arg, &end);
63
64         if (*end == '\0')
65                 corruption_check_size = size;
66
67         return (size == corruption_check_size) ? 0 : -EINVAL;
68 }
69 early_param("memory_corruption_check_size", set_corruption_check_size);
70
71
72 void __init setup_bios_corruption_check(void)
73 {
74         phys_addr_t start, end;
75         u64 i;
76
77         if (memory_corruption_check == -1) {
78                 memory_corruption_check =
79 #ifdef CONFIG_X86_BOOTPARAM_MEMORY_CORRUPTION_CHECK
80                         1
81 #else
82                         0
83 #endif
84                         ;
85         }
86
87         if (corruption_check_size == 0)
88                 memory_corruption_check = 0;
89
90         if (!memory_corruption_check)
91                 return;
92
93         corruption_check_size = round_up(corruption_check_size, PAGE_SIZE);
94
95         for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
96                                 NULL) {
97                 start = clamp_t(phys_addr_t, round_up(start, PAGE_SIZE),
98                                 PAGE_SIZE, corruption_check_size);
99                 end = clamp_t(phys_addr_t, round_down(end, PAGE_SIZE),
100                               PAGE_SIZE, corruption_check_size);
101                 if (start >= end)
102                         continue;
103
104                 memblock_reserve(start, end - start);
105                 scan_areas[num_scan_areas].addr = start;
106                 scan_areas[num_scan_areas].size = end - start;
107
108                 /* Assume we've already mapped this early memory */
109                 memset(__va(start), 0, end - start);
110
111                 if (++num_scan_areas >= MAX_SCAN_AREAS)
112                         break;
113         }
114
115         if (num_scan_areas)
116                 printk(KERN_INFO "Scanning %d areas for low memory corruption\n", num_scan_areas);
117 }
118
119
120 void check_for_bios_corruption(void)
121 {
122         int i;
123         int corruption = 0;
124
125         if (!memory_corruption_check)
126                 return;
127
128         for (i = 0; i < num_scan_areas; i++) {
129                 unsigned long *addr = __va(scan_areas[i].addr);
130                 unsigned long size = scan_areas[i].size;
131
132                 for (; size; addr++, size -= sizeof(unsigned long)) {
133                         if (!*addr)
134                                 continue;
135                         printk(KERN_ERR "Corrupted low memory at %p (%lx phys) = %08lx\n",
136                                addr, __pa(addr), *addr);
137                         corruption = 1;
138                         *addr = 0;
139                 }
140         }
141
142         WARN_ONCE(corruption, KERN_ERR "Memory corruption detected in low memory\n");
143 }
144
145 static void check_corruption(struct work_struct *dummy);
146 static DECLARE_DELAYED_WORK(bios_check_work, check_corruption);
147
148 static void check_corruption(struct work_struct *dummy)
149 {
150         check_for_bios_corruption();
151         schedule_delayed_work(&bios_check_work,
152                 round_jiffies_relative(corruption_check_period*HZ));
153 }
154
155 static int start_periodic_check_for_corruption(void)
156 {
157         if (!num_scan_areas || !memory_corruption_check || corruption_check_period == 0)
158                 return 0;
159
160         printk(KERN_INFO "Scanning for low memory corruption every %d seconds\n",
161                corruption_check_period);
162
163         /* First time we run the checks right away */
164         schedule_delayed_work(&bios_check_work, 0);
165         return 0;
166 }
167 device_initcall(start_periodic_check_for_corruption);
168