blob: 8945be9ad2b10018fc91862af8977e12dcad1014 [file] [log] [blame]
H. Peter Anvin1965aae2008-10-22 22:26:29 -07001#ifndef _ASM_X86_MCE_H
2#define _ASM_X86_MCE_H
Thomas Gleixnere2f43022007-10-17 18:04:40 +02003
Jaswinder Singh Rajput999b6972009-01-30 22:47:27 +05304#include <linux/types.h>
Thomas Gleixnere2f43022007-10-17 18:04:40 +02005#include <asm/ioctls.h>
Thomas Gleixnere2f43022007-10-17 18:04:40 +02006
7/*
8 * Machine Check support for x86
9 */
10
Thomas Gleixner01c66802009-04-08 12:31:24 +020011#define MCG_BANKCNT_MASK 0xff /* Number of Banks */
Borislav Petkove4876832009-06-20 23:27:16 -070012#define MCG_CTL_P (1ULL<<8) /* MCG_CTL register available */
Thomas Gleixner01c66802009-04-08 12:31:24 +020013#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)
Andi Kleened7290d2009-05-27 21:56:57 +020018#define MCG_SER_P (1ULL<<24) /* MCA recovery/new status bits */
Thomas Gleixnere2f43022007-10-17 18:04:40 +020019
Ingo Molnar06b851d2009-04-08 12:31:25 +020020#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 */
Thomas Gleixnere2f43022007-10-17 18:04:40 +020023
Ingo Molnar06b851d2009-04-08 12:31:25 +020024#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 */
Andi Kleened7290d2009-05-27 21:56:57 +020031#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 */
Thomas Gleixnere2f43022007-10-17 18:04:40 +020040
Huang Ying5b7e88e2009-07-31 09:41:40 +080041#define MCJ_CTX_MASK 3
42#define MCJ_CTX(flags) ((flags) & MCJ_CTX_MASK)
43#define MCJ_CTX_RANDOM 0 /* inject context: random */
44#define MCJ_CTX_PROCESS 1 /* inject context: process */
45#define MCJ_CTX_IRQ 2 /* inject context: IRQ */
46#define MCJ_NMI_BROADCAST 4 /* do NMI broadcasting */
47
Thomas Gleixnere2f43022007-10-17 18:04:40 +020048/* Fields are zero when not available */
49struct mce {
50 __u64 status;
51 __u64 misc;
52 __u64 addr;
53 __u64 mcgstatus;
H. Peter Anvin65ea5b02008-01-30 13:30:56 +010054 __u64 ip;
Thomas Gleixnere2f43022007-10-17 18:04:40 +020055 __u64 tsc; /* cpu time stamp counter */
Andi Kleen8ee08342009-05-27 21:56:56 +020056 __u64 time; /* wall time_t when error was detected */
57 __u8 cpuvendor; /* cpu vendor as encoded in system.h */
Huang Ying5b7e88e2009-07-31 09:41:40 +080058 __u8 inject_flags; /* software inject flags */
59 __u16 pad;
Andi Kleen8ee08342009-05-27 21:56:56 +020060 __u32 cpuid; /* CPUID 1 EAX */
Thomas Gleixnere2f43022007-10-17 18:04:40 +020061 __u8 cs; /* code segment */
62 __u8 bank; /* machine check bank */
Andi Kleend620c672009-05-27 21:56:56 +020063 __u8 cpu; /* cpu number; obsolete; use extcpu now */
Thomas Gleixnere2f43022007-10-17 18:04:40 +020064 __u8 finished; /* entry is valid */
Andi Kleend620c672009-05-27 21:56:56 +020065 __u32 extcpu; /* linux cpu number that detected the error */
Andi Kleen8ee08342009-05-27 21:56:56 +020066 __u32 socketid; /* CPU socket ID */
67 __u32 apicid; /* CPU initial apic ID */
68 __u64 mcgcap; /* MCGCAP MSR: machine check capabilities of CPU */
Thomas Gleixnere2f43022007-10-17 18:04:40 +020069};
70
71/*
72 * This structure contains all data related to the MCE log. Also
73 * carries a signature to make it easier to find from external
74 * debugging tools. Each entry is only valid when its finished flag
75 * is set.
76 */
77
78#define MCE_LOG_LEN 32
79
80struct mce_log {
81 char signature[12]; /* "MACHINECHECK" */
82 unsigned len; /* = MCE_LOG_LEN */
83 unsigned next;
84 unsigned flags;
Andi Kleenf6fb0ac2009-05-27 21:56:55 +020085 unsigned recordlen; /* length of struct mce */
Thomas Gleixnere2f43022007-10-17 18:04:40 +020086 struct mce entry[MCE_LOG_LEN];
87};
88
89#define MCE_OVERFLOW 0 /* bit 0 in flags means overflow */
90
91#define MCE_LOG_SIGNATURE "MACHINECHECK"
92
93#define MCE_GET_RECORD_LEN _IOR('M', 1, int)
94#define MCE_GET_LOG_LEN _IOR('M', 2, int)
95#define MCE_GETCLEAR_FLAGS _IOR('M', 3, int)
96
97/* Software defined banks */
98#define MCE_EXTENDED_BANK 128
99#define MCE_THERMAL_BANK MCE_EXTENDED_BANK + 0
100
101#define K8_MCE_THRESHOLD_BASE (MCE_EXTENDED_BANK + 1) /* MCE_AMD */
102#define K8_MCE_THRESHOLD_BANK_0 (MCE_THRESHOLD_BASE + 0 * 9)
103#define K8_MCE_THRESHOLD_BANK_1 (MCE_THRESHOLD_BASE + 1 * 9)
104#define K8_MCE_THRESHOLD_BANK_2 (MCE_THRESHOLD_BASE + 2 * 9)
105#define K8_MCE_THRESHOLD_BANK_3 (MCE_THRESHOLD_BASE + 3 * 9)
106#define K8_MCE_THRESHOLD_BANK_4 (MCE_THRESHOLD_BASE + 4 * 9)
107#define K8_MCE_THRESHOLD_BANK_5 (MCE_THRESHOLD_BASE + 5 * 9)
108#define K8_MCE_THRESHOLD_DRAM_ECC (MCE_THRESHOLD_BANK_4 + 0)
109
Thomas Gleixnere2f43022007-10-17 18:04:40 +0200110#ifdef __KERNEL__
111
Hidetoshi Seto9e55e442009-06-15 17:22:15 +0900112#include <linux/percpu.h>
113#include <linux/init.h>
114#include <asm/atomic.h>
115
Thomas Gleixnere2f43022007-10-17 18:04:40 +0200116extern int mce_disabled;
Hidetoshi Setoc6978362009-06-15 17:22:49 +0900117extern int mce_p5_enabled;
Thomas Gleixnere2f43022007-10-17 18:04:40 +0200118
Hidetoshi Seto58995d22009-06-15 17:27:47 +0900119#ifdef CONFIG_X86_MCE
120void mcheck_init(struct cpuinfo_x86 *c);
121#else
122static inline void mcheck_init(struct cpuinfo_x86 *c) {}
123#endif
124
Hidetoshi Seto9e55e442009-06-15 17:22:15 +0900125#ifdef CONFIG_X86_ANCIENT_MCE
126void intel_p5_mcheck_init(struct cpuinfo_x86 *c);
127void winchip_mcheck_init(struct cpuinfo_x86 *c);
Hidetoshi Setoc6978362009-06-15 17:22:49 +0900128static inline void enable_p5_mce(void) { mce_p5_enabled = 1; }
Hidetoshi Seto9e55e442009-06-15 17:22:15 +0900129#else
130static inline void intel_p5_mcheck_init(struct cpuinfo_x86 *c) {}
131static inline void winchip_mcheck_init(struct cpuinfo_x86 *c) {}
Hidetoshi Setoc6978362009-06-15 17:22:49 +0900132static inline void enable_p5_mce(void) {}
Hidetoshi Seto9e55e442009-06-15 17:22:15 +0900133#endif
134
Andi Kleenb5f2fa42009-02-12 13:43:22 +0100135void mce_setup(struct mce *m);
Thomas Gleixnere2f43022007-10-17 18:04:40 +0200136void mce_log(struct mce *m);
Ingo Molnarcb491fc2009-04-08 12:31:17 +0200137DECLARE_PER_CPU(struct sys_device, mce_dev);
Thomas Gleixnere2f43022007-10-17 18:04:40 +0200138
Andi Kleen41fdff32009-02-12 13:49:30 +0100139/*
Andi Kleen3ccdccf2009-07-09 00:31:45 +0200140 * Maximum banks number.
141 * This is the limit of the current register layout on
142 * Intel CPUs.
Andi Kleen41fdff32009-02-12 13:49:30 +0100143 */
Andi Kleen3ccdccf2009-07-09 00:31:45 +0200144#define MAX_NR_BANKS 32
Andi Kleen41fdff32009-02-12 13:49:30 +0100145
Thomas Gleixnere2f43022007-10-17 18:04:40 +0200146#ifdef CONFIG_X86_MCE_INTEL
Hidetoshi Seto62fdac52009-06-11 16:06:07 +0900147extern int mce_cmci_disabled;
148extern int mce_ignore_ce;
Thomas Gleixnere2f43022007-10-17 18:04:40 +0200149void mce_intel_feature_init(struct cpuinfo_x86 *c);
Andi Kleen88ccbed2009-02-12 13:49:36 +0100150void cmci_clear(void);
151void cmci_reenable(void);
152void cmci_rediscover(int dying);
153void cmci_recheck(void);
Thomas Gleixnere2f43022007-10-17 18:04:40 +0200154#else
155static inline void mce_intel_feature_init(struct cpuinfo_x86 *c) { }
Andi Kleen88ccbed2009-02-12 13:49:36 +0100156static inline void cmci_clear(void) {}
157static inline void cmci_reenable(void) {}
158static inline void cmci_rediscover(int dying) {}
159static inline void cmci_recheck(void) {}
Thomas Gleixnere2f43022007-10-17 18:04:40 +0200160#endif
161
162#ifdef CONFIG_X86_MCE_AMD
163void mce_amd_feature_init(struct cpuinfo_x86 *c);
164#else
165static inline void mce_amd_feature_init(struct cpuinfo_x86 *c) { }
166#endif
167
H. Peter Anvin38736072009-05-28 10:05:33 -0700168int mce_available(struct cpuinfo_x86 *c);
Andi Kleen88ccbed2009-02-12 13:49:36 +0100169
Andi Kleen01ca79f2009-05-27 21:56:52 +0200170DECLARE_PER_CPU(unsigned, mce_exception_count);
Andi Kleenca84f692009-05-27 21:56:57 +0200171DECLARE_PER_CPU(unsigned, mce_poll_count);
Andi Kleen01ca79f2009-05-27 21:56:52 +0200172
Thomas Gleixnere2f43022007-10-17 18:04:40 +0200173extern atomic_t mce_entry;
174
Andi Kleenee031c32009-02-12 13:49:34 +0100175typedef DECLARE_BITMAP(mce_banks_t, MAX_NR_BANKS);
176DECLARE_PER_CPU(mce_banks_t, mce_poll_banks);
177
Andi Kleenb79109c2009-02-12 13:43:23 +0100178enum mcp_flags {
179 MCP_TIMESTAMP = (1 << 0), /* log time stamp */
180 MCP_UC = (1 << 1), /* log uncorrected errors */
Andi Kleen5679af42009-04-07 17:06:55 +0200181 MCP_DONTLOG = (1 << 2), /* only clear, don't log */
Andi Kleenb79109c2009-02-12 13:43:23 +0100182};
H. Peter Anvin38736072009-05-28 10:05:33 -0700183void machine_check_poll(enum mcp_flags flags, mce_banks_t *b);
Andi Kleenb79109c2009-02-12 13:43:23 +0100184
Andi Kleen9ff36ee2009-05-27 21:56:58 +0200185int mce_notify_irq(void);
Andi Kleen9b1beaf2009-05-27 21:56:59 +0200186void mce_notify_process(void);
Thomas Gleixnere2f43022007-10-17 18:04:40 +0200187
Andi Kleenea149b32009-04-29 19:31:00 +0200188DECLARE_PER_CPU(struct mce, injectm);
189extern struct file_operations mce_chrdev_ops;
190
Hidetoshi Seto58995d22009-06-15 17:27:47 +0900191/*
192 * Exception handler
193 */
194
195/* Call the installed machine check handler for this CPU setup. */
196extern void (*machine_check_vector)(struct pt_regs *, long error_code);
197void do_machine_check(struct pt_regs *, long);
198
199/*
200 * Threshold handler
201 */
Thomas Gleixnere2f43022007-10-17 18:04:40 +0200202
Andi Kleenb2762682009-02-12 13:49:31 +0100203extern void (*mce_threshold_vector)(void);
Hidetoshi Seto58995d22009-06-15 17:27:47 +0900204extern void (*threshold_cpu_callback)(unsigned long action, unsigned int cpu);
Andi Kleenb2762682009-02-12 13:49:31 +0100205
Hidetoshi Setoe8ce2c52009-06-15 17:24:40 +0900206/*
207 * Thermal handler
208 */
209
Hidetoshi Setoe8ce2c52009-06-15 17:24:40 +0900210void intel_init_thermal(struct cpuinfo_x86 *c);
211
Hidetoshi Setoe8ce2c52009-06-15 17:24:40 +0900212void mce_log_therm_throt_event(__u64 status);
Hidetoshi Setoe8ce2c52009-06-15 17:24:40 +0900213
Thomas Gleixnere2f43022007-10-17 18:04:40 +0200214#endif /* __KERNEL__ */
H. Peter Anvin1965aae2008-10-22 22:26:29 -0700215#endif /* _ASM_X86_MCE_H */