tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1#ifndef _ASM_X86_MCE_H
2#define _ASM_X86_MCE_H
3
4#include <linux/types.h>
5#include <asm/ioctls.h>
6
7/*
8 * Machine Check support for x86
9 */
10
11#define MCG_BANKCNT_MASK 0xff /* Number of Banks */
12#define MCG_CTL_P (1ULL<<8) /* MCG_CAP register available */
13#define MCG_EXT_P (1ULL<<9) /* Extended registers available */
14#define MCG_CMCI_P (1ULL<<10) /* CMCI supported */
15#define MCG_EXT_CNT_MASK 0xff0000 /* Number of Extended registers */
16#define MCG_EXT_CNT_SHIFT 16
17#define MCG_EXT_CNT(c) (((c) & MCG_EXT_CNT_MASK) >> MCG_EXT_CNT_SHIFT)
18#define MCG_SER_P (1ULL<<24) /* MCA recovery/new status bits */
19
20#define MCG_STATUS_RIPV (1ULL<<0) /* restart ip valid */
21#define MCG_STATUS_EIPV (1ULL<<1) /* ip points to correct instruction */
22#define MCG_STATUS_MCIP (1ULL<<2) /* machine check in progress */
23
24#define MCI_STATUS_VAL (1ULL<<63) /* valid error */
25#define MCI_STATUS_OVER (1ULL<<62) /* previous errors lost */
26#define MCI_STATUS_UC (1ULL<<61) /* uncorrected error */
27#define MCI_STATUS_EN (1ULL<<60) /* error enabled */
28#define MCI_STATUS_MISCV (1ULL<<59) /* misc error reg. valid */
29#define MCI_STATUS_ADDRV (1ULL<<58) /* addr reg. valid */
30#define MCI_STATUS_PCC (1ULL<<57) /* processor context corrupt */
31#define MCI_STATUS_S (1ULL<<56) /* Signaled machine check */
32#define MCI_STATUS_AR (1ULL<<55) /* Action required */
33
34/* MISC register defines */
35#define MCM_ADDR_SEGOFF 0 /* segment offset */
36#define MCM_ADDR_LINEAR 1 /* linear address */
37#define MCM_ADDR_PHYS 2 /* physical address */
38#define MCM_ADDR_MEM 3 /* memory address */
39#define MCM_ADDR_GENERIC 7 /* generic */
40
41/* Fields are zero when not available */
42struct mce {
43 __u64 status;
44 __u64 misc;
45 __u64 addr;
46 __u64 mcgstatus;
47 __u64 ip;
48 __u64 tsc; /* cpu time stamp counter */
49 __u64 time; /* wall time_t when error was detected */
50 __u8 cpuvendor; /* cpu vendor as encoded in system.h */
51 __u8 pad1;
52 __u16 pad2;
53 __u32 cpuid; /* CPUID 1 EAX */
54 __u8 cs; /* code segment */
55 __u8 bank; /* machine check bank */
56 __u8 cpu; /* cpu number; obsolete; use extcpu now */
57 __u8 finished; /* entry is valid */
58 __u32 extcpu; /* linux cpu number that detected the error */
59 __u32 socketid; /* CPU socket ID */
60 __u32 apicid; /* CPU initial apic ID */
61 __u64 mcgcap; /* MCGCAP MSR: machine check capabilities of CPU */
62};
63
64/*
65 * This structure contains all data related to the MCE log. Also
66 * carries a signature to make it easier to find from external
67 * debugging tools. Each entry is only valid when its finished flag
68 * is set.
69 */
70
71#define MCE_LOG_LEN 32
72
73struct mce_log {
74 char signature[12]; /* "MACHINECHECK" */
75 unsigned len; /* = MCE_LOG_LEN */
76 unsigned next;
77 unsigned flags;
78 unsigned recordlen; /* length of struct mce */
79 struct mce entry[MCE_LOG_LEN];
80};
81
82#define MCE_OVERFLOW 0 /* bit 0 in flags means overflow */
83
84#define MCE_LOG_SIGNATURE "MACHINECHECK"
85
86#define MCE_GET_RECORD_LEN _IOR('M', 1, int)
87#define MCE_GET_LOG_LEN _IOR('M', 2, int)
88#define MCE_GETCLEAR_FLAGS _IOR('M', 3, int)
89
90/* Software defined banks */
91#define MCE_EXTENDED_BANK 128
92#define MCE_THERMAL_BANK MCE_EXTENDED_BANK + 0
93
94#define K8_MCE_THRESHOLD_BASE (MCE_EXTENDED_BANK + 1) /* MCE_AMD */
95#define K8_MCE_THRESHOLD_BANK_0 (MCE_THRESHOLD_BASE + 0 * 9)
96#define K8_MCE_THRESHOLD_BANK_1 (MCE_THRESHOLD_BASE + 1 * 9)
97#define K8_MCE_THRESHOLD_BANK_2 (MCE_THRESHOLD_BASE + 2 * 9)
98#define K8_MCE_THRESHOLD_BANK_3 (MCE_THRESHOLD_BASE + 3 * 9)
99#define K8_MCE_THRESHOLD_BANK_4 (MCE_THRESHOLD_BASE + 4 * 9)
100#define K8_MCE_THRESHOLD_BANK_5 (MCE_THRESHOLD_BASE + 5 * 9)
101#define K8_MCE_THRESHOLD_DRAM_ECC (MCE_THRESHOLD_BANK_4 + 0)
102
103#ifdef __KERNEL__
104
105#include <linux/percpu.h>
106#include <linux/init.h>
107#include <asm/atomic.h>
108
109extern int mce_disabled;
110extern int mce_p5_enabled;
111
112#ifdef CONFIG_X86_MCE
113void mcheck_init(struct cpuinfo_x86 *c);
114#else
115static inline void mcheck_init(struct cpuinfo_x86 *c) {}
116#endif
117
118#ifdef CONFIG_X86_OLD_MCE
119extern int nr_mce_banks;
120void amd_mcheck_init(struct cpuinfo_x86 *c);
121void intel_p4_mcheck_init(struct cpuinfo_x86 *c);
122void intel_p6_mcheck_init(struct cpuinfo_x86 *c);
123#endif
124
125#ifdef CONFIG_X86_ANCIENT_MCE
126void intel_p5_mcheck_init(struct cpuinfo_x86 *c);
127void winchip_mcheck_init(struct cpuinfo_x86 *c);
128static inline void enable_p5_mce(void) { mce_p5_enabled = 1; }
129#else
130static inline void intel_p5_mcheck_init(struct cpuinfo_x86 *c) {}
131static inline void winchip_mcheck_init(struct cpuinfo_x86 *c) {}
132static inline void enable_p5_mce(void) {}
133#endif
134
135void mce_setup(struct mce *m);
136void mce_log(struct mce *m);
137DECLARE_PER_CPU(struct sys_device, mce_dev);
138
139/*
140 * To support more than 128 would need to escape the predefined
141 * Linux defined extended banks first.
142 */
143#define MAX_NR_BANKS (MCE_EXTENDED_BANK - 1)
144
145#ifdef CONFIG_X86_MCE_INTEL
146extern int mce_cmci_disabled;
147extern int mce_ignore_ce;
148void mce_intel_feature_init(struct cpuinfo_x86 *c);
149void cmci_clear(void);
150void cmci_reenable(void);
151void cmci_rediscover(int dying);
152void cmci_recheck(void);
153#else
154static inline void mce_intel_feature_init(struct cpuinfo_x86 *c) { }
155static inline void cmci_clear(void) {}
156static inline void cmci_reenable(void) {}
157static inline void cmci_rediscover(int dying) {}
158static inline void cmci_recheck(void) {}
159#endif
160
161#ifdef CONFIG_X86_MCE_AMD
162void mce_amd_feature_init(struct cpuinfo_x86 *c);
163#else
164static inline void mce_amd_feature_init(struct cpuinfo_x86 *c) { }
165#endif
166
167int mce_available(struct cpuinfo_x86 *c);
168
169DECLARE_PER_CPU(unsigned, mce_exception_count);
170DECLARE_PER_CPU(unsigned, mce_poll_count);
171
172extern atomic_t mce_entry;
173
174typedef DECLARE_BITMAP(mce_banks_t, MAX_NR_BANKS);
175DECLARE_PER_CPU(mce_banks_t, mce_poll_banks);
176
177enum mcp_flags {
178 MCP_TIMESTAMP = (1 << 0), /* log time stamp */
179 MCP_UC = (1 << 1), /* log uncorrected errors */
180 MCP_DONTLOG = (1 << 2), /* only clear, don't log */
181};
182void machine_check_poll(enum mcp_flags flags, mce_banks_t *b);
183
184int mce_notify_irq(void);
185void mce_notify_process(void);
186
187DECLARE_PER_CPU(struct mce, injectm);
188extern struct file_operations mce_chrdev_ops;
189
190/*
191 * Exception handler
192 */
193
194/* Call the installed machine check handler for this CPU setup. */
195extern void (*machine_check_vector)(struct pt_regs *, long error_code);
196void do_machine_check(struct pt_regs *, long);
197
198/*
199 * Threshold handler
200 */
201
202extern void (*mce_threshold_vector)(void);
203extern void (*threshold_cpu_callback)(unsigned long action, unsigned int cpu);
204
205/*
206 * Thermal handler
207 */
208
209void intel_init_thermal(struct cpuinfo_x86 *c);
210
211#ifdef CONFIG_X86_NEW_MCE
212void mce_log_therm_throt_event(__u64 status);
213#else
214static inline void mce_log_therm_throt_event(__u64 status) {}
215#endif
216
217#endif /* __KERNEL__ */
218#endif /* _ASM_X86_MCE_H */