linux-kernel-fault-codes

Linux 内核 6.6.0 故障错误码与系统影响 — 完整参考手册

范围：rasdaemon 关心的所有 trace event + 错误码定义 — MCE / AER / EDAC / CXL / extlog / non-standard / memory-failure / devlink / block / signal
基线：Linux kernel 6.6.0（/work/work/git_code/linux）
性质：内核侧错误码字典 + 严重性 + 系统影响 + 来源溯源（file:line）

---

0. 总览：rasdaemon 关心的 14 个 trace event 在内核的来源

Trace event	头文件 / 源文件	触发点
mce_record	include/trace/events/mce.h	arch/x86/kernel/cpu/mce/core.c
mc_event	include/ras/ras_event.h:98-163	drivers/edac/edac_mc.c:929 edac_raw_mc_handle_error()
aer_event	include/ras/ras_event.h:269-339	drivers/pci/pcie/aer.c
cxl_aer_uncorrectable_error	drivers/cxl/core/trace.h:51	drivers/cxl/core/pci.c
cxl_aer_correctable_error	drivers/cxl/core/trace.h:99	drivers/cxl/core/pci.c
cxl_overflow	drivers/cxl/core/trace.h:127	drivers/cxl/core/mbox.c
cxl_generic_event	drivers/cxl/core/trace.h:225	drivers/cxl/core/mbox.c:863 cxl_event_trace_record()
cxl_general_media	drivers/cxl/core/trace.h:315	同上
cxl_dram	drivers/cxl/core/trace.h:398	同上
cxl_memory_module	drivers/cxl/core/trace.h:547	同上
cxl_poison	drivers/cxl/core/trace.h:643	drivers/cxl/core/mbox.c
arm_event	include/ras/ras_event.h:171-208	drivers/ras/ras.c:24 log_arm_hw_error() ← ghes.c:507
extlog_mem_event	include/ras/ras_event.h:27-77	drivers/acpi/acpi_extlog.c:178
non_standard_event	include/ras/ras_event.h:219-253	drivers/ras/ras.c:17 ← ghes.c:676
memory_failure_event	include/ras/ras_event.h:399-423	mm/memory-failure.c:1323
signal_generate	include/trace/events/signal.h:50-80	kernel/signal.c
devlink_health_report	include/trace/events/devlink.h:81-107	net/devlink/health.c
block_rq_error	include/trace/events/block.h:165-170	block/blk-core.c

---

1. MCE — Machine Check Exception（x86）

1.1 mce_record trace event 字段

include/trace/events/mce.h:12-70 — 对应 struct mce (arch/x86/include/uapi/asm/mce.h:13-39)：

字段	类型	来源 MSR / 寄存器
cpu	u32	m->extcpu
mcgcap	u64	MCGCAP
mcgstatus	u64	MCG_STATUS
bank	u8	0…63（MCA bank index）
status	u64	MCi_STATUS
ipid	u64	MCA_IPID（SMCA 才有）
addr	u64	MCi_ADDR
misc	u64	MCi_MISC
synd	u64	MCA_SYND（SMCA 才有）
cs	u8	code segment
ip	u64	instruction pointer
tsc	u64	RDTSC
cpuvendor	u32	X86_VENDOR_*（INTEL=0, AMD=2, HYGON=9, ZHAOXIN=11…）
cpuid	u32	CPUID 1.EAX（family/model/stepping）
walltime	u64	m->time（秒）
socketid	u32	CPU socket id
apicid	u32	initial APIC ID

1.2 MCG_STATUS 位（全局机器检查状态）

arch/x86/include/asm/mce.h:23-27

位	名称	含义
0	MCG_STATUS_RIPV	Restart IP valid — 可在 m->ip 处恢复执行
1	MCG_STATUS_EIPV	Error IP valid — m->ip 指向出错指令
2	MCG_STATUS_MCIP	Machine Check In Progress — 软件处理完后必须清零
3	MCG_STATUS_LMCES	LMCE Signaled — MCE 作为 Local MCE 投递

MCG_CAP 特性位（mce.h:12-21）：

位	名称	含义
7:0	MCG_BANKCNT_MASK	MCA bank 数量
8	MCG_CTL_P	MCG_CTL 存在
9	MCG_EXT_P	扩展寄存器（MCG_EIP 等）
10	MCG_CMCI_P	CMCI 支持
23:16	MCG_EXT_CNT_MASK	扩展寄存器计数
24	MCG_SER_P	Software Error Recovery 支持（UCNA/SRAR）
26	MCG_ELOG_P	Extended error log 支持
27	MCG_LMCE_P	Local MCE 支持

1.3 MCI_STATUS 位（每 bank MCA 状态）

arch/x86/include/asm/mce.h:32-52

位	名称	含义	严重性
63	MCI_STATUS_VAL	寄存器有效（必先检查）	—
62	MCI_STATUS_OVER	之前的错误被丢弃（溢出）	—
61	MCI_STATUS_UC	Uncorrected error	Uncorrected
60	MCI_STATUS_EN	Error enabled（已配置为检测）	—
59	MCI_STATUS_MISCV	MCi_MISC 有效	—
58	MCI_STATUS_ADDRV	MCi_ADDR 有效	—
57	MCI_STATUS_PCC	Processor Context Corrupt — 内核无法继续	致命
56	MCI_STATUS_S	Signaled — MCE 已通过异常/CMCI 投递	—
55	MCI_STATUS_AR	Action Required — 软件必须动作（如下线页）	Recoverable w/ action
52:38	MCI_STATUS_CEC_MASK	Corrected Error Count（已纠正错误数）	—
31:16	MSCOD	Model-Specific Error Code	—
15:0	MCACOD	MCA Error Code（bit 12 过滤位忽略）	—

AMD 特有位（mce.h:48-52）：

位	名称	含义
55	MCI_STATUS_TCC	Task Context Corrupt（MCA 扩展）
53	MCI_STATUS_SYNDV	MCi_SYND 有效
44	MCI_STATUS_DEFERRED	未纠正但延迟（无异常；通过 APIC 记录）
43	MCI_STATUS_POISON	访问到已被下毒的数据（数据 poisoning / SUCCOR）
40	MCI_STATUS_SCRUB	巡检 scrub 期间发现

MCACOD 宏（mce.h:73-80）：

名称	值	含义
MCACOD	0xefff	MCACOD 字段掩码
MCACOD_SCRUB	0x00C0	Memory scrubbing error（bits 0xC0…0xCF）
MCACOD_SCRUBMSK	0xeff0	Scrub 掩码（跳过 bit 12）
MCACOD_L3WB	0x017A	L3 explicit writeback 错误
MCACOD_DATA	0x0134	Data Load（SRAR）
MCACOD_INSTR	0x0150	Instruction Fetch（SRAR）

1.4 MCI_MISC 位

mce.h:82-92

宏	位	含义
MCI_MISC_ADDR_LSB(m)	m & 0x3f	Address LSB（亚页位掩码）
MCI_MISC_ADDR_MODE(m)	(m>>6) & 7	Address mode
MCI_MISC_ADDR_SEGOFF	0	Segment offset
MCI_MISC_ADDR_LINEAR	1	Linear
MCI_MISC_ADDR_PHYS	2	Physical（mce_usable_address 使用）
MCI_MISC_ADDR_MEM	3	Memory
MCI_MISC_ADDR_GENERIC	7	Generic
MCI_ADDR_PHYSADDR	GENMASK_ULL(phys_bits-1, 0)	物理地址掩码

MCI_CTL2（mce.h:94-96）：

位	名称	含义
30	MCI_CTL2_CMCI_EN	CMCI 中断使能
14:0	MCI_CTL2_CMCI_THRESHOLD_MASK	CMCI 阈值

1.5 严重性等级 enum severity_level

arch/x86/kernel/cpu/mce/internal.h:11-21

值	常量	含义
0	MCE_NO_SEVERITY	无效 / 不启用（!VAL 或 !EN）— 丢弃
1	MCE_DEFERRED_SEVERITY (=MCE_UCNA_SEVERITY)	延迟错误或 UCNA — 保留
2	MCE_KEEP_SEVERITY	CE / 非致命 UC — 保留
3	MCE_SOME_SEVERITY	动作可选但未知
4	MCE_AO_SEVERITY	Action Optional（scrub / L3WB）
5	MCE_UC_SEVERITY	Uncorrected 无 AR
6	MCE_AR_SEVERITY	Action Required（页必须离线）
7	MCE_PANIC_SEVERITY	致命 — 触发 panic

1.6 Intel 严重性表（severity.c:38-218）

核心规则（按顺序，首匹配胜）：

严重性	触发条件	消息
NO	!MCI_STATUS_VAL	“Invalid”
NO	!MCI_STATUS_EN（异常上下文）	“Not enabled”
PANIC	MCI_STATUS_PCC set	“Processor context corrupt”
PANIC	MCG_STATUS_MCIP clear	“MCIP not set in MCA handler”
PANIC	RIPV & EIPV 都清零	“Neither restart nor error IP”
PANIC	RIPV clear in KERNEL/KERNEL_RECOV	“In kernel and no restart IP”
KEEP	UC clear	“Corrected error”
AO	`MCI_UC_AR	MCACOD_SCRUBMSK==UC
AO	`MCI_UC_AR	MCACOD==UC
AO	SKX step≥4, banks 13-18, `ADDR	0x001000c0`
UCNA	MCI_UC_SAR == UC（S, AR clear）	“Uncorrected no action required”
PANIC	OVER+UC+AR+!S	“Illegal combination (UCNA with AR=1)”
KEEP	S clear in SER mode	“Non signaled machine check”
PANIC	OVER+UC+S+AR	“Action required with lost events”
KEEP	UC+S+AR+ADDRV, RIPV	“Action required but unaffected thread is continuable”
AR	UC+S+AR+ADDRV+MCACOD_DATA, KERNEL_RECOV	“Data load in error recoverable area of kernel”
AR	UC+S+AR+ADDRV+MCACOD_DATA, USER	“Data load error in a user process”
AR	UC+S+AR+ADDRV+MCACOD_INSTR, USER	“Instruction fetch error in a user process”
PANIC	UC+S+AR+ADDRV+MCACOD_DATA, KERNEL	“Data load in unrecoverable area of kernel”
PANIC	UC+S+AR+ADDRV+MCACOD_INSTR, KERNEL	“Instruction fetch error in kernel”
PANIC	UC+S+AR, no MCACOD	“Action required: unknown MCACOD”
SOME	UC+S, no AR	“Action optional: unknown MCACOD”
SOME	OVER+UC+S	“Action optional with lost events”
PANIC	OVER+UC	“Overflowed uncorrected”
PANIC	UC, KERNEL	“Uncorrected in kernel”
UC	UC, no AR	“Uncorrected”
SOME	catch-all	“No match”

复合宏（severity.c:67-70）：

• MCI_UC_S = MCI_STATUS_UC | MCI_STATUS_S
• MCI_UC_AR = MCI_STATUS_UC | MCI_STATUS_AR
• MCI_UC_SAR = MCI_STATUS_UC | MCI_STATUS_S | MCI_STATUS_AR
• MCI_ADDR = MCI_STATUS_ADDRV | MCI_STATUS_MISCV

1.7 AMD 严重性（severity_amd()，severity.c:310-368）

检查	严重性	消息
MCI_STATUS_PCC	MCE_PANIC_SEVERITY	“Processor Context Corrupt”
MCI_STATUS_DEFERRED	MCE_DEFERRED_SEVERITY	—
!MCI_STATUS_UC	MCE_KEEP_SEVERITY	“Corrected or deferred-by-UC=0”
OVER && !overflow_recov	MCE_PANIC_SEVERITY	“Overflowed uncorrected without MCA Overflow Recovery”
!succor	MCE_PANIC_SEVERITY	“Uncorrected without MCA Recovery”
error_context == IN_KERNEL	MCE_PANIC_SEVERITY	“Uncorrected unrecoverable in kernel context”
默认	MCE_AR_SEVERITY	—

error_context()（severity.c:274-307）：

• IN_USER — cs & 3 == 3
• IN_KERNEL — 无 RIPV
• IN_KERNEL_RECOV — uaccess/copy with extable fixup，置 MCE_IN_KERNEL_RECOV / MCE_IN_KERNEL_COPYIN

1.8 公共函数

函数	位置	行为
mce_usable_address()	core.c:462-482	UC 无 DEFERRED
mce_is_correctable()	core.c:525-537	满足 ADDRV && MISCV && LSB≤PAGE_SHIFT && ADDR_MODE==PHYS
mce_is_memory_error()	core.c:485-514	AMD：SMCA_UMC/UMC_V2 且 XEC=0，或 K8 bank 4 XEC=0x8。Intel：`(status & 0xef80)==BIT(7)

1.9 AMD SMCA 26 个 bank type

arch/x86/include/asm/mce.h:299-334（enum smca_bank_types）：

索引	名称	简名
0	SMCA_LS	load_store
1	SMCA_LS_V2	load_store
2	SMCA_IF	insn_fetch
3	SMCA_L2_CACHE	l2_cache
4	SMCA_DE	decode_unit
5	SMCA_RESERVED	reserved
6	SMCA_EX	execution_unit
7	SMCA_FP	floating_point
8	SMCA_L3_CACHE	l3_cache
9	SMCA_CS	coherent_slave
10	SMCA_CS_V2	coherent_slave
11	SMCA_PIE	pie
12	SMCA_UMC	umc
13	SMCA_UMC_V2	umc_v2
14	SMCA_PB	param_block
15	SMCA_PSP	psp
16	SMCA_PSP_V2	psp
17	SMCA_SMU	smu
18	SMCA_SMU_V2	smu
19	SMCA_MP5	mp5
20	SMCA_MPDMA	mpdma
21	SMCA_NBIO	nbio
22	SMCA_PCIE	pcie
23	SMCA_PCIE_V2	pcie
24	SMCA_XGMI_PCS	xgmi_pcs
25	SMCA_NBIF	nbif
26	SMCA_SHUB	shub
27	SMCA_SATA	sata
28	SMCA_USB	usb
29	SMCA_GMI_PCS	gmi_pcs
30	SMCA_XGMI_PHY	xgmi_phy
31	SMCA_WAFL_PHY	wafl_phy
32	SMCA_GMI_PHY	gmi_phy
—	N_SMCA_BANK_TYPES	33（sentinel）

HWID/MCATYPE 映射（amd.c:160-219）：

Bank	HWID	MCATYPE
SMCA_LS	0xB0	0x0
SMCA_LS_V2	0xB0	0x10
SMCA_IF	0xB0	0x1
SMCA_L2_CACHE	0xB0	0x2
SMCA_DE	0xB0	0x3
SMCA_EX	0xB0	0x5
SMCA_FP	0xB0	0x6
SMCA_CS	0x2E	0x0
SMCA_PIE	0x2E	0x1
SMCA_CS_V2	0x2E	0x2
SMCA_UMC	0x96	0x0
SMCA_UMC_V2	0x96	0x1
SMCA_PB	0x05	0x0
SMCA_PSP	0xFF	0x0
SMCA_PSP_V2	0xFF	0x1
SMCA_SMU	0x01	0x0
SMCA_SMU_V2	0x01	0x1
SMCA_MP5	0x01	0x2
SMCA_MPDMA	0x01	0x3
SMCA_NBIO	0x18	0x0
SMCA_PCIE	0x46	0x0
SMCA_PCIE_V2	0x46	0x1
SMCA_XGMI_PCS	0x50	0x0
SMCA_NBIF	0x6C	0x0
SMCA_SHUB	0x80	0x0
SMCA_SATA	0xA8	0x0
SMCA_USB	0xAA	0x0
SMCA_GMI_PCS	0x241	0x0
SMCA_XGMI_PHY	0x259	0x0
SMCA_WAFL_PHY	0x267	0x0
SMCA_GMI_PHY	0x269	0x0

宏：HWID_MCATYPE(hwid, mcatype) = ((hwid) << 16) | (mcatype)（amd.c:74）

• MCI_IPID_MCATYPE = 0xFFFF0000
• MCI_IPID_HWID = 0xFFF
• MCI_CONFIG_MCAX = 0x1（指示 MCA 扩展）

1.10 AMD 阈值 / 延迟错误位

amd.c:32-56

常量	值	含义
THRESHOLD_MAX	0xFFF	12-bit 最大计数
NR_BLOCKS	5	每 bank 最多 MISC block
INT_TYPE_APIC	0x00020000	中断类型 = APIC LVT
MASK_VALID_HI	0x80000000	MCi_MISC.high block 有效
MASK_CNTP_HI	0x40000000	计数器存在
MASK_LOCKED_HI	0x20000000	阈值已锁
MASK_LVTOFF_HI	0x00F00000	LVT offset（bits 23:20）
MASK_COUNT_EN_HI	0x00080000	计数使能
MASK_INT_TYPE_HI	0x00060000	中断类型字段
MASK_OVERFLOW_HI	0x00010000	溢出位
MASK_ERR_COUNT_HI	0x00000FFF	错误计数（bits 11:0）
MASK_BLKPTR_LO	0xFF000000	low word block 指针
MCG_XBLK_ADDR	0xC0000400	Extended block 基址
MSR_CU_DEF_ERR	0xC0000410	延迟错误配置 MSR
MASK_DEF_LVTOFF	0x000000F0	延迟 LVT offset
MASK_DEF_INT_TYPE	0x00000006	延迟中断类型
DEF_LVT_OFF	0x2	默认 LVT offset
DEF_INT_TYPE_APIC	0x2	默认 APIC
SMCA_THR_LVT_OFF	0xF000	SMCA 阈值 LVT offset

K8/Family 0x15 bank 4 子块（amd.c:357-374）：

• 0x00000413 → “dram”
• 0xc0000408 → “ht_links”
• 0xc0000409 → “l3_cache”

1.11 AMD SMCA MSR 地址

arch/x86/include/asm/mce.h:113-132

MSR	值
MSR_AMD64_SMCA_MC0_CTL	0xc0002000
MSR_AMD64_SMCA_MC0_STATUS	0xc0002001
MSR_AMD64_SMCA_MC0_ADDR	0xc0002002
MSR_AMD64_SMCA_MC0_MISC0	0xc0002003
MSR_AMD64_SMCA_MC0_CONFIG	0xc0002004
MSR_AMD64_SMCA_MC0_IPID	0xc0002005
MSR_AMD64_SMCA_MC0_SYND	0xc0002006
MSR_AMD64_SMCA_MC0_DESTAT	0xc0002008（延迟错误状态）
MSR_AMD64_SMCA_MC0_DEADDR	0xc0002009（延迟错误地址）
MSR_AMD64_SMCA_MC0_MISC1	0xc000200a
MSR_AMD64_SMCA_MCx_*	基址 + 0x10 * x

MCA_CONFIG 用到的位（amd.c:274-334）：

• bit 32 (high & 0x1) = McaX — 启用 SMCA 寄存器布局
• bit 37 (low & 0x20) = DeferredIntTypeSupported
• bits 38:37 = DeferredIntType（0x1 = APIC）
• bit 40 (low & 0x100) = McaLsbInStatusSupported

1.12 CMCI（Corrected Machine Check Interrupt）

intel.c:66-79 + mce.h:30, 95-96

常量	值	含义
CMCI_THRESHOLD	1	默认阈值
CMCI_POLL_INTERVAL	30*HZ	正常轮询间隔
CMCI_STORM_INTERVAL	HZ	风暴间隔
CMCI_STORM_THRESHOLD	15	风暴阈值（每秒事件数）

CMCI 风暴状态（intel.c:75-79）：

值	名称	含义
0	CMCI_STORM_NONE	无风暴
1	CMCI_STORM_ACTIVE	风暴中；切到轮询
2	CMCI_STORM_SUBSIDED	风暴结束；即将恢复中断模式

MCG_CMCI_P = BIT_ULL(10) — CMCI 支持位

1.13 mce-inject 操作类型

inject.c:47-61（enum injection_type）：

值	常量	触发
0	SW_INJ	mce_log(&i_mce) — 解码 only，安全
1	HW_INJ	int $18 → #MC 异常 — 真实 MCE，PCC 时 panic
2	DFR_INT_INJ	int $DEFERRED_ERROR_VECTOR — AMD 延迟错误 APIC
3	THR_INT_INJ	int $THRESHOLD_APIC_VECTOR — AMD 阈值 APIC
4	N_INJ_TYPES	sentinel

注入上下文标志（mce.h:98-105）：

• MCJ_CTX_MASK = 3, MCJ_CTX(flags) = (flags) & 3
• MCJ_CTX_RANDOM = 0（默认）
• MCJ_CTX_PROCESS = 1（进程上下文）
• MCJ_CTX_IRQ = 2（IRQ 上下文）
• MCJ_NMI_BROADCAST = 0x4
• MCJ_EXCEPTION = 0x8
• MCJ_IRQ_BROADCAST = 0x10

Debugfs 文件（inject.c:696-710）：status、misc、addr、synd、ipid、bank、flags、cpu、README

1.14 mce.kflags（kernel-only）

mce.h:137-158

位	名称	含义
0	MCE_HANDLED_CEC	由 CEC 处理
1	MCE_HANDLED_UC	由 UC handler 处理
2	MCE_HANDLED_EXTLOG	由 Extended Log 处理
3	MCE_HANDLED_NFIT	由 NFIT 处理
4	MCE_HANDLED_EDAC	由 EDAC 处理
5	MCE_HANDLED_MCELOG	由 /dev/mcelog 处理
6	MCE_IN_KERNEL_RECOV	内核可恢复
7	MCE_IN_KERNEL_COPYIN	copy_from_user 期间

1.15 mce_notifier_prios

mce.h:176-186（notifier chain 优先级）

优先级	常量	用途
0	MCE_PRIO_LOWEST	默认 print-only
1	MCE_PRIO_MCELOG	/dev/mcelog 旧设备
2	MCE_PRIO_EDAC	EDAC 子系统
3	MCE_PRIO_NFIT	NVDIMM Firmware Interface
4	MCE_PRIO_EXTLOG	Extended error log
5	MCE_PRIO_UC	UC handler（页离线）
6	MCE_PRIO_EARLY	最早消费方（emit trace_mce_record）
7	MCE_PRIO_CEC = MCE_PRIO_HIGHEST	Corrected error collector

1.16 mcp_flags（轮询标志）

mce.h:256-262

标志	含义
MCP_TIMESTAMP	读 bank 时盖 TSC
MCP_UC	记录 UC 错误
MCP_DONTLOG	清但不记录
MCP_QUEUE_LOG	仅入 genpool（boot 时用）

1.17 Vendor flags / quirks

internal.h:139-179

位	名称	含义
0	overflow_recov	MCA overflow recovery（F15h 00h-0fh 总是有）
1	succor	AMD S/W UnCorrectable COntainment & Recovery
2	smca	AMD Scalable MCA
3	zen_ifu_quirk	Zen IFU poison-consumption EIPV/RIPV bug
4	amd_threshold	AMD-style 阈值 bank
5	p5	Pentium family-5
6	winchip	Centaur Winchip C6
7	snb_ifu_quirk	Sandy Bridge IFU EIPV/RIPV bug
8	skx_repmov_quirk	Skylake/Cascade/Cooper Lake REP MOVS bug

1.18 CPU vendor 枚举

arch/x86/include/asm/processor.h：X86_VENDOR_INTEL=0, X86_VENDOR_AMD=2, X86_VENDOR_HYGON=9, X86_VENDOR_CENTAUR=5/8, X86_VENDOR_ZHAOXIN=11, X86_VENDOR_UNKNOWN

1.19 Kernel cmdline MCE 选项

core.c:2238-2290

选项	效果
mce	启用 Pentium P5 MCE
mce=off	禁用全部 MCE
mce=no_cmci	禁用 CMCI
mce=no_lmce	禁用 LMCE
mce=dont_log_ce	不记录 CE
mce=print_all	全部 MCE 打到控制台
mce=ignore_ce	禁用 CE 轮询和 CMCI
mce=bootlog / nobootlog	启动前日志切换
mce=bios_cmci_threshold	不编程 CMCI 阈值
mce=recovery	强制启用 copy_mc_fragile()
mce=TOLERANCELEVEL	monarch 超时（微秒）
nomce	mce=off 的别名

1.20 其他常量

名称	值	位置
MAX_NR_BANKS	64	mce.h:227
MCE_LOG_MIN_LEN	32	mce.h:107
MCE_LOG_SIGNATURE	“MACHINECHECK”	mce.h:108
MCE_POOLSZ	2*PAGE_SIZE (8KiB)	genpool.c:22
MCE_OVERFLOW	0	mce.h:107
NBCFG	0x44	inject.c:45

/dev/mcelog ioctls（uapi/asm/mce.h:41-43）：

• MCE_GET_RECORD_LEN = _IOR('M', 1, int)
• MCE_GET_LOG_LEN = _IOR('M', 2, int)
• MCE_GETCLEAR_FLAGS = _IOR('M', 3, int)

CPUHP 状态（core.c:2767-2778）：

• CPUHP_X86_MCE_DEAD = “x86/mce:dead” → mce_cpu_dead()
• CPUHP_AP_ONLINE_DYN = “x86/mce:online” → mce_cpu_online()

---

2. PCIe AER

2.1 aer_event trace event 字段

include/ras/ras_event.h:269-339

TP_PROTO(const char *dev_name,
         const u32 status,
         const u8 severity,
         const u8 tlp_header_valid,
         struct aer_header_log_regs *tlp)

字段	类型	描述
dev_name	string	设备 slot 名（[domain:]bus:device.function）
status	u32	Correctable 或 Uncorrectable 寄存器值
severity	u8	AER_* 枚举值
tlp_header_valid	u8	是否捕获到 TLP header
tlp_header	u32[4]	TLP header log 4 个 DWORD

2.2 AER 严重性枚举

include/linux/aer.h:14-17

值	名称	含义
0	AER_NONFATAL	Uncorrected, non-fatal
1	AER_FATAL	Uncorrected, fatal
2	AER_CORRECTABLE	Corrected
3	DPC_FATAL	Fatal via Downstream Port Containment（驱动内部）

2.3 Correctable Error Status（PCI_ERR_COR_STATUS = 0x10）

include/uapi/linux/pci_regs.h:748-802

位	掩码	名称	含义
0	0x00000001	PCI_ERR_COR_RCVR	Receiver Error（PHY）
6	0x00000040	PCI_ERR_COR_BAD_TLP	Bad TLP
7	0x00000080	PCI_ERR_COR_BAD_DLLP	Bad DLLP
8	0x00000100	PCI_ERR_COR_REP_ROLL	REPLAY_NUM Rollover
12	0x00001000	PCI_ERR_COR_REP_TIMER	Replay Timer Timeout
13	0x00002000	PCI_ERR_COR_ADV_NFAT	Advisory Non-Fatal
14	0x00004000	PCI_ERR_COR_INTERNAL	Corrected Internal
15	0x00008000	PCI_ERR_COR_LOG_OVER	Header Log Overflow

AER_MAX_TYPEOF_COR_ERRS = 16

2.4 Uncorrectable Error Status（PCI_ERR_UNCOR_STATUS = 0x04）

位	掩码	名称	含义
0	0x00000001	PCI_ERR_UNC_UND	Undefined
4	0x00000010	PCI_ERR_UNC_DLP	Data Link Protocol Error
5	0x00000020	PCI_ERR_UNC_SURPDN	Surprise Down
12	0x00001000	PCI_ERR_UNC_POISON_TLP	Poisoned TLP
13	0x00002000	PCI_ERR_UNC_FCP	Flow Control Protocol
14	0x00004000	PCI_ERR_UNC_COMP_TIME	Completion Timeout
15	0x00008000	PCI_ERR_UNC_COMP_ABORT	Completer Abort
16	0x00010000	PCI_ERR_UNC_UNX_COMP	Unexpected Completion
17	0x00020000	PCI_ERR_UNC_RX_OVER	Receiver Overflow
18	0x00040000	PCI_ERR_UNC_MALF_TLP	Malformed TLP
19	0x00080000	PCI_ERR_UNC_ECRC	ECRC Error
20	0x00100000	PCI_ERR_UNC_UNSUP	Unsupported Request
21	0x00200000	PCI_ERR_UNC_ACSV	ACS Violation
22	0x00400000	PCI_ERR_UNC_INTN	Uncorrectable Internal
23	0x00800000	PCI_ERR_UNC_MCBTLP	MC Blocked TLP
24	0x01000000	PCI_ERR_UNC_ATOMEG	AtomicOp Egress Blocked
25	0x02000000	PCI_ERR_UNC_TLPPRE	TLP Prefix Blocked

AER_MAX_TYPEOF_UNCOR_ERRS = 27

2.5 AER Capability 寄存器布局

偏移	寄存器
0x04	PCI_ERR_UNCOR_STATUS
0x08	PCI_ERR_UNCOR_MASK
0x0c	PCI_ERR_UNCOR_SEVER（bit set = Fatal）
0x10	PCI_ERR_COR_STATUS
0x14	PCI_ERR_COR_MASK
0x18	PCI_ERR_CAP
0x1c	PCI_ERR_HEADER_LOG（16 字节）
0x2c	PCI_ERR_ROOT_COMMAND
0x30	PCI_ERR_ROOT_STATUS
0x34	PCI_ERR_ROOT_ERR_SRC

PCI_ERR_CAP 字段：

• 0x1f PCI_ERR_CAP_FEP(x) — First Error Pointer（5-bit）
• 0x20 PCI_ERR_CAP_ECRC_GENC — ECRC Generation Capable
• 0x40 PCI_ERR_CAP_ECRC_GENE — ECRC Generation Enable
• 0x80 PCI_ERR_CAP_ECRC_CHKC — ECRC Check Capable
• 0x100 PCI_ERR_CAP_ECRC_CHKE — ECRC Check Enable

PCI_ERR_ROOT_COMMAND：

• 0x01 PCI_ERR_ROOT_CMD_COR_EN
• 0x02 PCI_ERR_ROOT_CMD_NONFATAL_EN
• 0x04 PCI_ERR_ROOT_CMD_FATAL_EN

PCI_ERR_ROOT_STATUS：

• 0x01 PCI_ERR_ROOT_COR_RCV
• 0x02 PCI_ERR_ROOT_MULTI_COR_RCV
• 0x04 PCI_ERR_ROOT_UNCOR_RCV
• 0x08 PCI_ERR_ROOT_MULTI_UNCOR_RCV
• 0x10 PCI_ERR_ROOT_FIRST_FATAL
• 0x20 PCI_ERR_ROOT_NONFATAL_RCV
• 0x40 PCI_ERR_ROOT_FATAL_RCV
• 0xf8000000 PCI_ERR_ROOT_AER_IRQ

AER_ERR_STATUS_MASK = PCI_ERR_ROOT_UNCOR_RCV | PCI_ERR_ROOT_COR_RCV | PCI_ERR_ROOT_MULTI_COR_RCV | PCI_ERR_ROOT_MULTI_UNCOR_RCV（aer.c:104-107）

PCI_ERR_ROOT_ERR_SRC 解码（aer.c:101-102）：

• ERR_COR_ID(d) = d & 0xffff
• ERR_UNCOR_ID(d) = d >> 16

2.6 AER recovery

状态	触发
pci_channel_io_normal	Non-Fatal → pcie_do_recovery()
pci_channel_io_frozen	Fatal → pcie_do_recovery()

aer_recover_ring = 16 entries（AER_RECOVER_RING_SIZE=16，aer.c:972），由 aer_recover_work_func 处理。aer_root_reset() 选择 FLR（PCI_EXP_TYPE_RC_EC）或 pci_bus_error_reset()（RP / Downstream）。

2.7 Agent / Layer 分类

aer.c:399-432

Agents（4）：AER_AGENT_RECEIVER(0), _REQUESTER(1), _COMPLETER(2), _TRANSMITTER(3)
Layers（3）：AER_PHYSICAL_LAYER_ERROR(0), _DATA_LINK_LAYER_ERROR(1), _TRANSACTION_LAYER_ERROR(2)

宏：AER_GET_AGENT()、AER_GET_LAYER_ERROR()

2.8 AER_LOG_TLP_MASKS

aer.c:88-93：

#define AER_LOG_TLP_MASKS (PCI_ERR_UNC_POISON_TLP | PCI_ERR_UNC_ECRC | \
                           PCI_ERR_UNC_UNSUP | PCI_ERR_UNC_COMP_ABORT | \
                           PCI_ERR_UNC_UNX_COMP | PCI_ERR_UNC_MALF_TLP)

2.9 AER_INJECT

drivers/pci/pcie/aer_inject.c

struct aer_error_inj {
    u8 bus, dev, fn;
    u32 uncor_status, cor_status;
    u32 header_log0..3;
    u32 domain;
};

/dev/aer_inject 字符设备（aer_inject.c:512-516），拦截 pci_ops 写 root port AER 寄存器，触发 root IRQ。

2.10 CPER AER section

include/linux/cper.h:246-253, 509-542

struct cper_sec_pcie {
    u64 validation_bits;   // CPER_PCIE_VALID_AER_INFO = 0x80
    u8 aer_info[96];       // uncor_status, uncor_mask, uncor_severity,
                           // cor_status, cor_mask, cap_control,
                           // header_log[4], root_command, root_status
    // ... other fields
};

cper_severity_to_aer() 映射（aer.c:750-761）：

• CPER_SEV_RECOVERABLE → AER_NONFATAL
• CPER_SEV_FATAL → AER_FATAL
• 其他 → AER_CORRECTABLE

---

3. EDAC / Memory Controller

3.1 enum hw_event_mc_err_type

include/linux/edac.h:113-119

值	名称	含义	系统影响
0	HW_EVENT_ERR_CORRECTED	ECC 在硬件中纠正	单/多位已纠正；可能触发 page scrub
1	HW_EVENT_ERR_UNCORRECTED	未被 ECC 纠正但不致命	页被下毒；可能杀进程；数据不可用
2	HW_EVENT_ERR_DEFERRED	数据下毒，非紧急处理	预防性下线页
3	HW_EVENT_ERR_FATAL	未恢复的不可纠正错误	panic / machine check
4	HW_EVENT_ERR_INFO	信息性（按 CPER 规范）	诊断用

3.2 enum dev_type（DRAM 设备宽度）

include/linux/edac.h:72-81

值	名称	含义
0	DEV_UNKNOWN	未知
1	DEV_X1	1-bit 数据宽度
2	DEV_X2	2-bit
3	DEV_X4	4-bit（典型）
4	DEV_X8	8-bit（典型）
5	DEV_X16	16-bit
6	DEV_X32	32-bit
7	DEV_X64	64-bit

DEV_FLAG_X1…DEV_FLAG_X64 派生位标志（edac.h:83-90）。x4/x8 通常用 EDAC_S4ECD4ED / EDAC_S8ECD8ED 启用 chipkill。

3.3 enum edac_mc_layer_type（层级）

include/linux/edac.h:346-352

值	名称	含义
0	EDAC_MC_LAYER_BRANCH	FB-DIMM branch
1	EDAC_MC_LAYER_CHANNEL	内存 channel
2	EDAC_MC_LAYER_SLOT	DIMM slot
3	EDAC_MC_LAYER_CHIP_SELECT	Rank（csrow）
4	EDAC_MC_LAYER_ALL_MEM	全部内存（firmware-driven）

显示名（edac_mc.c:794-800）：“branch” / “channel” / “slot” / “csrow” / “memory”

3.4 enum mem_type（DRAM 类型，参考）

include/linux/edac.h:191-221

值	名称	类型
0	MEM_EMPTY	—
1	MEM_RESERVED	—
2	MEM_UNKNOWN	—
3	MEM_FPM	~1995
4	MEM_EDO	~1998
5	MEM_BEDO	Burst EDO
6	MEM_SDR	SDR SDRAM
7	MEM_RDR	Registered SDR
8	MEM_DDR	—
9	MEM_RDDR	—
10	MEM_RMBS	Rambus
11	MEM_DDR2	—
12	MEM_FB_DDR2	Fully-Buffered
13	MEM_RDDR2	—
14	MEM_XDR	Rambus XDR
15	MEM_DDR3	—
16	MEM_RDDR3	—
17	MEM_LRDDR3	—
18	MEM_LPDDR3	—
19	MEM_DDR4	—
20	MEM_RDDR4	—
21	MEM_LRDDR4	—
22	MEM_LPDDR4	—
23	MEM_DDR5	—
24	MEM_RDDR5	—
25	MEM_LRDDR5	—
26	MEM_NVDIMM	—
27	MEM_WIO2	—
28	MEM_HBM2	—

3.5 enum edac_type（ECC 能力）

include/linux/edac.h:265-276

值	名称	含义
0	EDAC_UNKNOWN	—
1	EDAC_NONE	—
2	EDAC_RESERVED	—
3	EDAC_PARITY	—
4	EDAC_EC	仅检测
5	EDAC_SECDED	单错纠正 + 双错检测
6	EDAC_S2ECD2ED	—
7	EDAC_S4ECD4ED	chipkill x4
8	EDAC_S8ECD8ED	chipkill x8
9	EDAC_S16ECD16ED	chipkill x16

3.6 enum scrub_type（scrub 模式）

include/linux/edac.h:301-312：SCRUB_UNKNOWN, SCRUB_NONE, SCRUB_SW_PROG, SCRUB_SW_SRC, SCRUB_SW_PROG_SRC, SCRUB_SW_TUNABLE, SCRUB_HW_PROG, SCRUB_HW_SRC, SCRUB_HW_PROG_SRC, SCRUB_HW_TUNABLE

3.7 EDAC_OPSTATE 状态

include/linux/edac.h:26-29, 326-330

值	名称	含义
-1	EDAC_OPSTATE_INVAL	无效/未初始化
0	EDAC_OPSTATE_POLL	轮询
1	EDAC_OPSTATE_NMI	NMI
2	EDAC_OPSTATE_INT	中断

mci->op_state 内部（edac.h:603）：0x100 OP_ALLOC, 0x201 OP_RUNNING_POLL, 0x202 OP_RUNNING_INTERRUPT, 0x203 OP_RUNNING_POLL_INTR, 0x300 OP_OFFLINE

3.8 各平台驱动常量

平台	文件	关键常量
Skylake	drivers/edac/skx_common.c	NUM_CHANNELS, NUM_DIMMS；MTR[12:13]=rank, MTR[2:4]=row+12, MTR[0:1]=col+10；标签 "CPU_SrcID#%u_MC#%u_Chan#%u_DIMM#%u"
Icelake/SPR	drivers/edac/i10nm_base.c	I10NM_REVISION "v0.0.6", I10NM_HBM_IMC_MMIO_SIZE=0x9000, I10NM_IS_HBM_PRESENT(reg)=bits[27:30], I10NM_IS_HBM_IMC(reg)=bit 29
SB/IVB/HSW/BDW	drivers/edac/sb_edac.c	NUM_CHANNELS=6, MAX_DIMMS=3, KNL_MAX_CHAS=38, KNL_MAX_CHANNELS=6, KNL_MAX_EDCS=8；chipset enum：SANDY_BRIDGE, IVY_BRIDGE, HASWELL, BROADWELL, KNIGHTS_LANDING
Apollo Lake / Denverton	drivers/edac/pnd2_edac.c	APL_NUM_CHANNELS=4, DNV_NUM_CHANNELS=2, DNV_MAX_DIMMS=2；PMI ports apl_dports[4]={0x18,0x10,0x11,0x19}, dnv_dports[2]={0x10,0x12}
E7520/i3100	drivers/edac/e752x_edac.c	E752X_NR_CSROWS=8；chip enum：E7520=0, E7525=1, E7320=2, I3100=3；PCI device IDs: 0x3590/0x3591/0x359E/0x3593/0x3592/0x35B0/0x35B1；err regs: 0x40-0x84
AMD64	drivers/edac/amd64_edac.c	DCTs: K8=1, F10h=2, F16h=1, F15h=dynamic；DCT_CFG_SEL F1x10C；DCSB_CS_ENABLE；F10_NB_ARRAY_DRAM（注入）
X-Gene	drivers/edac/xgene_edac.c	MCU_MAX_RANK=8, MCU_RANK_STRIDE=0x40；MCUGECR=0x0110（MCU_GECR_DEMANDUCINTREN_MASK=BIT(0) 等）；MCUGESR=0x0114（MCU_GESR_ADDRNOMATCH_ERR_MASK=BIT(7) 等）；MCUESRR0=0x0314（MCU_ESRR_MULTUCERR_MASK=BIT(3), BACKUCERR_MASK=BIT(2), DEMANDUCERR_MASK=BIT(1), CERR_MASK=BIT(0)）

3.9 mc_event trace event 字段

include/ras/ras_event.h:98-163

字段	类型	含义
error_type	u32	HW_EVENT_ERR_* 严重性
msg	string	人类可读错误消息
label	string	DIMM 标签
error_count	u16	错误数
mc_index	u8	MC 索引
top_layer	s8	顶层层级索引
middle_layer	s8	中层
lower_layer	s8	底层
address	long	物理地址（page << PAGE_SHIFT | offset）
grain_bits	u8	log2(grain)
syndrome	long	ECC 综合征
driver_detail	string	驱动额外信息

---

4. CXL（Compute Express Link）

4.1 8 个 CXL trace events（v6.6.0，无 cxl_memory_sparing）

#	Event	行号
1	cxl_aer_uncorrectable_error	drivers/cxl/core/trace.h:51
2	cxl_aer_correctable_error	drivers/cxl/core/trace.h:99
3	cxl_overflow	drivers/cxl/core/trace.h:127
4	cxl_generic_event	drivers/cxl/core/trace.h:225
5	cxl_general_media	drivers/cxl/core/trace.h:315
6	cxl_dram	drivers/cxl/core/trace.h:398
7	cxl_memory_module	drivers/cxl/core/trace.h:547
8	cxl_poison	drivers/cxl/core/trace.h:643

4.2 CXL AER Uncorrectable 位

drivers/cxl/core/trace.h:17-31（注意是 CXL_RAS_UC_* 而非 CXL_AER_UE_*）

位	符号	含义
0	CXL_RAS_UC_CACHE_DATA_PARITY	Cache data parity
1	CXL_RAS_UC_CACHE_ADDR_PARITY	Cache address parity
2	CXL_RAS_UC_CACHE_BE_PARITY	Cache byte-enable parity
3	CXL_RAS_UC_CACHE_DATA_ECC	Cache data ECC UC
4	CXL_RAS_UC_MEM_DATA_PARITY	Memory data parity
5	CXL_RAS_UC_MEM_ADDR_PARITY	Memory address parity
6	CXL_RAS_UC_MEM_BE_PARITY	Memory byte-enable parity
7	CXL_RAS_UC_MEM_DATA_ECC	Memory data ECC UC
8	CXL_RAS_UC_REINIT_THRESH	REINIT 阈值
9	CXL_RAS_UC_RSVD_ENCODE	收到未识别编码
10	CXL_RAS_UC_POISON	收到对端 poison
11	CXL_RAS_UC_RECV_OVERFLOW	接收方溢出
14	CXL_RAS_UC_INTERNAL_ERR	设备内部错
15	CXL_RAS_UC_IDE_TX_ERR	IDE 发送错
16	CXL_RAS_UC_IDE_RX_ERR	IDE 接收错

掩码：cxl.h:130 = GENMASK(16,14) | GENMASK(11,0)。show_uc_errs() 在 trace.h:33-49。

4.3 CXL AER Correctable 位

trace.h:81-87（CXL_RAS_CE_*）

位	符号	含义
0	CXL_RAS_CE_CACHE_DATA_ECC	Cache ECC 纠正
1	CXL_RAS_CE_MEM_DATA_ECC	Memory ECC 纠正
2	CXL_RAS_CE_CRC_THRESH	CRC 阈值
3	CLX_RAS_CE_RETRY_THRESH (内核拼写错：缺 ‘X’)	Retry 阈值
4	CXL_RAS_CE_CACHE_POISON	收到 cache poison
5	CXL_RAS_CE_MEM_POISON	收到 memory poison
6	CXL_RAS_CE_PHYS_LAYER_ERR	PHY 错误

掩码：cxl.h:137 = GENMASK(6,0)。show_ce_errs() 在 trace.h:89-97。

4.4 CXL Event Record Type UUIDs（mbox.c:843-861）

UUID	Type	Trace event	CXL Spec
fbcd0a77-c260-417f-85a9-088b1621eba6	General Media (GMER)	cxl_general_media	3.0 §8.2.9.2.1.1
601dcbb3-9c06-4eab-b8af-4e9bfb5c9624	DRAM (DER)	cxl_dram	3.0 §8.2.9.2.1.2
fe927475-dd59-4339-a586-79bab113b774	Memory Module (MMER)	cxl_memory_module	3.0 §8.2.9.2.1.3
其他	Generic	cxl_generic_event	—

4.5 memory_event_type (GMER/DER type)

trace.h:278-285

值	符号	含义
0x00	CXL_GMER_MEM_EVT_TYPE_ECC_ERROR	ECC error
0x01	CXL_GMER_MEM_EVT_TYPE_INV_ADDR	Invalid address
0x02	CXL_GMER_MEM_EVT_TYPE_DATA_PATH_ERROR	Data path error

（注意：v6.6.0 没有 memory_event_sub_type 枚举）

4.6 transaction_type

trace.h:287-302

值	符号	含义
0x00	CXL_GMER_TRANS_UNKNOWN	Unknown
0x01	CXL_GMER_TRANS_HOST_READ	Host read
0x02	CXL_GMER_TRANS_HOST_WRITE	Host write
0x03	CXL_GMER_TRANS_HOST_SCAN_MEDIA	Host scan-media
0x04	CXL_GMER_TRANS_HOST_INJECT_POISON	Host inject-poison
0x05	CXL_GMER_TRANS_INTERNAL_MEDIA_SCRUB	Internal media scrub
0x06	CXL_GMER_TRANS_INTERNAL_MEDIA_MANAGEMENT	Internal media mgmt

4.7 cxl_event_log_type（严重性）

cxlmem.h:620-626 和 cxl.h:161-170

值	名称	状态寄存器位
0x00	CXL_EVENT_TYPE_INFO	CXLDEV_EVENT_STATUS_INFO = BIT(0)
0x01	CXL_EVENT_TYPE_WARN	CXLDEV_EVENT_STATUS_WARN = BIT(1)
0x02	CXL_EVENT_TYPE_FAIL	CXLDEV_EVENT_STATUS_FAIL = BIT(2)
0x03	CXL_EVENT_TYPE_FATAL	CXLDEV_EVENT_STATUS_FATAL = BIT(3)
—	CXLDEV_EVENT_STATUS_ALL	0x0F（OR）
0x04	CXL_EVENT_TYPE_MAX	sentinel

cxl_event_thread() ISR（pci.c:619）读 32-bit 状态，按设置位触发 Get Event Records mailbox（mbox.c:1023，FATAL→INFO 顺序）。

4.8 DPA Flags

trace.h:255-263

符号	值	含义
CXL_DPA_FLAGS_MASK	0x3F	低 6 位掩码
CXL_DPA_MASK	~CXL_DPA_FLAGS_MASK	高位
CXL_DPA_VOLATILE	BIT(0)	易失性
CXL_DPA_NOT_REPAIRABLE	BIT(1)	不可修复

4.9 Event Descriptor Flags

trace.h:269-276

符号	值	含义
CXL_GMER_EVT_DESC_UNCORECTABLE_EVENT	BIT(0)	(内核拼写：UNCORECTABLE)
CXL_GMER_EVT_DESC_THRESHOLD_EVENT	BIT(1)	阈值触发
CXL_GMER_EVT_DESC_POISON_LIST_OVERFLOW	BIT(2)	poison list 溢出

4.10 Health Status Flags (DHI)

trace.h:487-494

符号	值	含义
CXL_DHI_HS_MAINTENANCE_NEEDED	BIT(0)	维护需要
CXL_DHI_HS_PERFORMANCE_DEGRADED	BIT(1)	性能降级
CXL_DHI_HS_HW_REPLACEMENT_NEEDED	BIT(2)	硬件需更换

（v6.6.0 无 MEM_CAPACITY_DEGRADED）

4.11 Media Status Enum (DHI)

trace.h:496-527

值	符号	含义
0x00	CXL_DHI_MS_NORMAL	Normal
0x01	CXL_DHI_MS_NOT_READY	Not ready
0x02	CXL_DHI_MS_WRITE_PERSISTENCY_LOST	写持久性丢失
0x03	CXL_DHI_MS_ALL_DATA_LOST	全部数据丢失
0x04	CXL_DHI_MS_WRITE_PERSISTENCY_LOSS_EVENT_POWER_LOSS	掉电时持久性丢失
0x05	CXL_DHI_MS_WRITE_PERSISTENCY_LOSS_EVENT_SHUTDOWN	关机时持久性丢失
0x06	CXL_DHI_MS_WRITE_PERSISTENCY_LOSS_IMMINENT	即将丢失
0x07	CXL_DHI_MS_WRITE_ALL_DATA_LOSS_EVENT_POWER_LOSS	掉电时全数据丢失
0x08	CXL_DHI_MS_WRITE_ALL_DATA_LOSS_EVENT_SHUTDOWN	关机时全数据丢失
0x09	CXL_DHI_MS_WRITE_ALL_DATA_LOSS_IMMINENT	即将全数据丢失

4.12 DHI Add-Status 位字段

trace.h:529-545

符号	值	含义
CXL_DHI_AS_NORMAL	0x0	Normal
CXL_DHI_AS_WARNING	0x1	Warning
CXL_DHI_AS_CRITICAL	0x2	Critical
CXL_DHI_AS_LIFE_USED(as)	as & 0x3	寿命 2-bit
CXL_DHI_AS_DEV_TEMP(as)	(as & 0xC) >> 2	温度 2-bit
CXL_DHI_AS_COR_VOL_ERR_CNT(as)	(as & 0x10) >> 4	易失性错误 1-bit
CXL_DHI_AS_COR_PER_ERR_CNT(as)	(as & 0x20) >> 5	持久性错误 1-bit

4.13 Memory Module Event Types

trace.h:467-480

值	符号	含义
0x00	CXL_MMER_HEALTH_STATUS_CHANGE	Health changed
0x01	CXL_MMER_MEDIA_STATUS_CHANGE	Media status changed
0x02	CXL_MMER_LIFE_USED_CHANGE	寿命阈值跨越
0x03	CXL_MMER_TEMP_CHANGE	温度阈值跨越
0x04	CXL_MMER_DATA_PATH_ERROR	Data path 错误
0x05	CXL_MMER_LSA_ERROR	Label Storage Area 错误

Sparing 标志（v6.6.0 不存在，后加）：

• HARD_SPARING、QUERY_RESOURCES、DEVICE_INITIATED — 在更新内核

4.14 Poison List Source Enum

cxlmem.h:778-782

值	符号	含义
0	CXL_POISON_SOURCE_UNKNOWN	Unknown
1	CXL_POISON_SOURCE_EXTERNAL	External（peer）
2	CXL_POISON_SOURCE_INTERNAL	Internal
3	CXL_POISON_SOURCE_INJECTED	Injected（host）
7	CXL_POISON_SOURCE_VENDOR	Vendor（4-6 保留）

存储：64-bit poison 地址低 3 位。cxlmem.h:763-764：

• CXL_POISON_START_MASK = GENMASK_ULL(63, 6)
• CXL_POISON_SOURCE_MASK = GENMASK(2, 0)

长度单位：CXL_POISON_LEN_MULT = 64 字节。cxlmem.h:773-775 flags：

• CXL_POISON_FLAG_MORE = BIT(0)
• CXL_POISON_FLAG_OVERFLOW = BIT(1)
• CXL_POISON_FLAG_SCANNING = BIT(2)
• CXL_POISON_LIST_MAX = 1024

4.15 Common Event Record Flags（header）

trace.h:165-174

符号	值	含义
CXL_EVENT_RECORD_FLAG_PERMANENT	BIT(2)	永久
CXL_EVENT_RECORD_FLAG_MAINT_NEEDED	BIT(3)	维护需要
CXL_EVENT_RECORD_FLAG_PERF_DEGRADED	BIT(4)	性能降级
CXL_EVENT_RECORD_FLAG_HW_REPLACE	BIT(5)	硬件更换

4.16 CXLMDEV Memory Device Status

cxlmem.h:11-32

符号	值	含义
CXLMDEV_STATUS_OFFSET	0x0	—
CXLMDEV_DEV_FATAL	BIT(0)	设备致命
CXLMDEV_FW_HALT	BIT(1)	固件停机
CXLMDEV_STATUS_MEDIA_STATUS_MASK	GENMASK(3,2)	—
CXLMDEV_MS_NOT_READY	0	—
CXLMDEV_MS_READY	1	—
CXLMDEV_MS_ERROR	2	—
CXLMDEV_MS_DISABLED	3	—
CXLMDEV_MBOX_IF_READY	BIT(4)	—
CXLMDEV_RESET_NEEDED_MASK	GENMASK(7,5)	—
CXLMDEV_RESET_NEEDED_NOT	0	—
CXLMDEV_RESET_NEEDED_COLD	1	—
CXLMDEV_RESET_NEEDED_WARM	2	—
CXLMDEV_RESET_NEEDED_HOT	3	—
CXLMDEV_RESET_NEEDED_CXL	4	—

CXLMDEV_DEV_FATAL 和 CXLMDEV_FW_HALT 触发 cxl_err()（pci.c:77-80）→ dev_err_ratelimited。

4.17 Mailbox Return Codes（cxlmem.h:142-174）

CMD_CMD_RC_TABLE 30 个 CXL_MBOX_CMD_RC_*：

SUCCESS, BACKGROUND, INPUT, UNSUPPORTED, INTERNAL, RETRY, BUSY, MEDIADISABLED, FWINPROGRESS, FWOOO, FWAUTH, FWSLOT, FWROLLBACK, FWRESET, HANDLE, PADDR(-EFAULT), POISONLMT, MEDIAFAILURE, ABORT, SECURITY, PASSPHRASE, MBUNSUPPORTED, PAYLOADLEN, LOG, INTERRUPTED, FEATUREVERSION, FEATURESELVALUE, FEATURETRANSFERIP, FEATURETRANSFEROOO, RESOURCEEXHAUSTED, EXTLIST

4.18 Mailbox Opcodes for Event（cxlmem.h:493-531）

Opcode	符号	规范
0x0100	CXL_MBOX_OP_GET_EVENT_RECORD	3.0 §8.2.9.2.2
0x0101	CXL_MBOX_OP_CLEAR_EVENT_RECORD	3.0 §8.2.9.2.3
0x0102	CXL_MBOX_OP_GET_EVT_INT_POLICY	3.0 §8.2.9.2.4
0x0103	CXL_MBOX_OP_SET_EVT_INT_POLICY	3.0 §8.2.9.2.4
0x4300	CXL_MBOX_OP_GET_POISON	3.0 §8.2.9.8.4.1
0x4301	CXL_MBOX_OP_INJECT_POISON	3.0 §8.2.9.8.4.2
0x4302	CXL_MBOX_OP_CLEAR_POISON	3.0 §8.2.9.8.4.3
0x4303	CXL_MBOX_OP_GET_SCAN_MEDIA_CAPS	3.0
0x4304	CXL_MBOX_OP_SCAN_MEDIA	3.0（UAPI 标记 deprecated）
0x4305	CXL_MBOX_OP_GET_SCAN_MEDIA	3.0（UAPI 标记 deprecated）
0x4400	CXL_MBOX_OP_SANITIZE	3.0 §8.2.9.8.5.1
0x4401	CXL_MBOX_OP_SECURE_ERASE	3.0 §8.2.9.8.5.2
0x4500-0x4505	Security commands	3.0
0x10000	CXL_MBOX_OP_MAX	sentinel

4.19 Get Event Payload Flags（cxlmem.h:604-615）

符号	值
CXL_GET_EVENT_FLAG_OVERFLOW	BIT(0)
CXL_GET_EVENT_FLAG_MORE_RECORDS	BIT(1)

struct cxl_get_event_payload {
    u8 flags;
    u8 reserved1;
    __le16 overflow_err_count;
    __le64 first_overflow_timestamp;
    __le64 last_overflow_timestamp;
    __le16 record_count;
    u8 reserved2[10];
    struct cxl_event_record_raw records[];
};

cxl_overflow 事件在 flags & CXL_GET_EVENT_FLAG_OVERFLOW 时发射。

4.20 Event Interrupt Modes（cxlmem.h:213-217）

enum cxl_event_int_mode {
    CXL_INT_NONE     = 0x00,
    CXL_INT_MSI_MSIX = 0x01,
    CXL_INT_FW       = 0x02
};

cxl_event_config_msgnums() (pci.c:677) 硬设 4 个严重性 → CXL_INT_MSI_MSIX。

4.21 PMEM Security State Flags（cxlmem.h:819-824）

符号	值	含义
CXL_PMEM_SEC_STATE_USER_PASS_SET	0x01	—
CXL_PMEM_SEC_STATE_MASTER_PASS_SET	0x02	—
CXL_PMEM_SEC_STATE_LOCKED	0x04	—
CXL_PMEM_SEC_STATE_FROZEN	0x08	—
CXL_PMEM_SEC_STATE_USER_PLIMIT	0x10	—
CXL_PMEM_SEC_STATE_MASTER_PLIMIT	0x20	—

4.22 DVSEC 标识符（cxlpci.h:15-56）

ID	符号	规范
0x1E98	PCI_DVSEC_VENDOR_ID_CXL	CXL 2.0 §8.1
0	CXL_DVSEC_PCIE_DEVICE	2.0 §8.1.3
2	CXL_DVSEC_FUNCTION_MAP	2.0 §8.1.4
3	CXL_DVSEC_PORT_EXTENSIONS	2.0 §8.1.5
4	CXL_DVSEC_PORT_GPF	2.0 §8.1.6
5	CXL_DVSEC_DEVICE_GPF	2.0 §8.1.7
7	CXL_DVSEC_PCIE_FLEXBUS_PORT	2.0 §8.1.8
8	CXL_DVSEC_REG_LOCATOR	2.0 §8.1.9

Device DVSEC 控制位：CXL_DVSEC_MEM_CAPABLE=BIT(2), CXL_DVSEC_HDM_COUNT_MASK=GENMASK(5,4), CXL_DVSEC_MEM_ENABLE=BIT(2), CXL_DVSEC_MEM_INFO_VALID=BIT(0), CXL_DVSEC_MEM_ACTIVE=BIT(1)

4.23 RAS 寄存器偏移和掩码（cxl.h:129-145）

偏移	寄存器	掩码
0x0	CXL_RAS_UNCORRECTABLE_STATUS_OFFSET	`GENMASK(16,14)
0x4	CXL_RAS_UNCORRECTABLE_MASK_OFFSET	`GENMASK(16,14)
0x8	CXL_RAS_UNCORRECTABLE_SEVERITY_OFFSET	—
0xC	CXL_RAS_CORRECTABLE_STATUS_OFFSET	GENMASK(6,0)
0x10	CXL_RAS_CORRECTABLE_MASK_OFFSET	—
0x14	CXL_RAS_CAP_CONTROL_OFFSET	GENMASK(5,0) (FE pointer)
0x18	CXL_RAS_HEADER_LOG_OFFSET	—
—	CXL_RAS_CAPABILITY_LENGTH	0x58
—	CXL_HEADERLOG_SIZE	SZ_512
—	CXL_HEADERLOG_SIZE_U32	SZ_512 / sizeof(u32)

4.24 Misc

名称	值	位置
CXL_EVENT_RECORD_DATA_LENGTH	0x50	cxlmem.h:594
CXL_EVENT_GEN_MED_COMP_ID_SIZE	0x10	cxlmem.h:645
CXL_EVENT_DER_CORRECTION_MASK_SIZE	0x20	cxlmem.h:664
CXL_CLEAR_EVENT_MAX_HANDLES	U8_MAX	cxlmem.h:639
CXL_FW_TRANSFER_ALIGNMENT	128	cxlmem.h:323

重要：v6.6.0 内核没有 CXL_1_1 / CXL_2_0 / CXL_3_0 版本宏 — 硬编码 3.0。

---

5. extlog（Extended Log）

5.1 extlog_mem_event trace event 字段

include/ras/ras_event.h:27-77，触发：drivers/acpi/acpi_extlog.c:178

字段	类型	含义
err_seq	u32	MCE extlog 错误序号
etype	u8	CPER 内存 error type，或 ~0（无效）
sev	u8	CPER 严重性
pa	u64	错误物理地址，或 ~0ull（无效）
pa_mask_lsb	u8	__ffs64(pa_mask)，或 ~0
fru_id	guid_t (16 字节)	FRU GUID
fru_text	string (≤20 字符)	FRU 文本
data	cper_mem_err_compact	压缩的 CPER 内存错误字段

---

6. CPER（Common Platform Error Record）

6.1 cper_severity 枚举

include/linux/cper.h:38-43

值	名称	含义
0	CPER_SEV_RECOVERABLE	可恢复
1	CPER_SEV_FATAL	致命
2	CPER_SEV_CORRECTED	已纠正
3	CPER_SEV_INFORMATIONAL	信息性

cper_severity_str() 映射（drivers/firmware/efi/cper.c:58-69）：

• RECOVERABLE → “recoverable”
• FATAL → “fatal”
• CORRECTED → “corrected”
• INFORMATIONAL → “info”
• 其他 → “unknown”

6.2 CPER 内存 error type 字符串

drivers/firmware/efi/cper.c:189-213（mem_err_type_strs[]）

Index	字符串	含义	影响
0	“unknown”	Unknown	不定
1	“no error”	Spurious	无
2	“single-bit ECC”	CE 单 bit	已纠正
3	“multi-bit ECC”	UE 多 bit	页下毒
4	“single-symbol chipkill ECC”	单符号 chipkill	已纠正
5	“multi-symbol chipkill ECC”	多符号 chipkill 失败	UE
6	“master abort”	Bus master abort	事务中止
7	“target abort”	Bus target abort	事务中止
8	“parity error”	Bus parity	完整性破坏
9	“watchdog timeout”	Device watchdog	事务停滞
10	“invalid address”	非法地址	UE
11	“mirror Broken”	镜像损坏	镜像禁用
12	“memory sparing”	备用 rank 激活	已纠正
13	“scrub corrected error”	巡检纠错	已纠正
14	“scrub uncorrected error”	巡检发现 UE	页下毒
15	“physical memory map-out event”	DIMM map-out	设备移除

6.3 CPER 内存 validation bits

include/linux/cper.h:215-244

位	名称	字段
0x0001	CPER_MEM_VALID_ERROR_STATUS	error_status
0x0002	CPER_MEM_VALID_PA	physical_addr
0x0004	CPER_MEM_VALID_PA_MASK	physical_addr_mask
0x0008	CPER_MEM_VALID_NODE	node
0x0010	CPER_MEM_VALID_CARD	card
0x0020	CPER_MEM_VALID_MODULE	module
0x0040	CPER_MEM_VALID_BANK	bank
0x0080	CPER_MEM_VALID_DEVICE	device
0x0100	CPER_MEM_VALID_ROW	row
0x0200	CPER_MEM_VALID_COLUMN	column
0x0400	CPER_MEM_VALID_BIT_POSITION	bit_pos
0x0800	CPER_MEM_VALID_REQUESTOR_ID	requestor_id
0x1000	CPER_MEM_VALID_RESPONDER_ID	responder_id
0x2000	CPER_MEM_VALID_TARGET_ID	target_id
0x4000	CPER_MEM_VALID_ERROR_TYPE	error_type
0x8000	CPER_MEM_VALID_RANK_NUMBER	rank
0x10000	CPER_MEM_VALID_CARD_HANDLE	mem_array_handle
0x20000	CPER_MEM_VALID_MODULE_HANDLE	mem_dev_handle
0x40000	CPER_MEM_VALID_ROW_EXT	Extended row
0x80000	CPER_MEM_VALID_BANK_GROUP	Bank group
0x100000	CPER_MEM_VALID_BANK_ADDRESS	Bank address
0x200000	CPER_MEM_VALID_CHIP_ID	Chip ID

掩码/位移：

• CPER_MEM_EXT_ROW_MASK = 0x3，CPER_MEM_EXT_ROW_SHIFT = 16
• CPER_MEM_BANK_ADDRESS_MASK = 0xff
• CPER_MEM_BANK_GROUP_SHIFT = 8
• CPER_MEM_CHIP_ID_SHIFT = 5

6.4 CPER Section Type GUIDs

include/linux/cper.h:154-200

GUID	Section Type	厂商
9876CCAD-47B4-4bdb-B65E-16F193C4F3DB	CPER_SEC_PROC_GENERIC	通用处理器
DC3EA0B0-A144-4797-B95B-53FA242B6E1D	CPER_SEC_PROC_IA	Intel x86/x64
E429FAF1-3CB7-11D4-0BCA-070080C73C88	CPER_SEC_PROC_IPF	Itanium
E19E3D16-BC11-11E4-9CAAC2051D5D46B0	CPER_SEC_PROC_ARM	ARM
A5BC1114-6F64-4EDE-B863-3E83ED7C83B1	CPER_SEC_PLATFORM_MEM	Platform memory
D995E954-BBC1-430F-AD91-B44DCB3C6F35	CPER_SEC_PCIE	PCIe
81212A96-09ED-4996-94718D729C8E69ED	CPER_SEC_FW_ERR_REC_REF	Firmware error record ref
C5753963-3B84-4095-BF78-EDDAD3F9C9DD	CPER_SEC_PCI_X_BUS	PCI/PCI-X bus
EB5E4685-CA66-4769-B6A2-26068B001326	CPER_SEC_PCI_DEV	PCI component/device
5B51FEF7-C79D-4434-8F1B-AA62DE3E2C64	CPER_SEC_DMAR_GENERIC	Generic DMAr
71761D37-32B2-45cd-A7D0-B0FEDD93E8CF	CPER_SEC_DMAR_VT	Intel VT-d DMAr
036F84E1-7F37-428c-A79E-575FDFAA84EC	CPER_SEC_DMAR_IOMMU	IOMMU DMAr

6.5 CPER Section Flags

include/linux/cper.h:123-146

标志	值	含义
CPER_SEC_VALID_FRU_ID	0x1	—
CPER_SEC_VALID_FRU_TEXT	0x2	—
CPER_SEC_PRIMARY	0x0001	直接关联错误
CPER_SEC_CONTAINMENT_WARNING	0x0002	可能已传播
CPER_SEC_RESET	0x0004	组件需重新初始化
CPER_SEC_ERROR_THRESHOLD_EXCEEDED	0x0008	错误阈值
CPER_SEC_RESOURCE_NOT_ACCESSIBLE	0x0010	查询冲突
CPER_SEC_LATENT_ERROR	0x0020	错误已包含但未纠正
CPER_SEC_REV	0x0100	—

6.6 CPER Record Flags

include/linux/cper.h:97-101

标志	值	含义
CPER_HW_ERROR_FLAGS_RECOVERED	0x1	错误已恢复
CPER_HW_ERROR_FLAGS_PREVERR	0x2	来自上 boot
CPER_HW_ERROR_FLAGS_SIMULATED	0x4	测试注入

6.7 CPER Processor Validation Bits

include/linux/cper.h:201-213

位	名称	字段
0x0001	CPER_PROC_VALID_TYPE	proc_type
0x0002	CPER_PROC_VALID_ISA	proc_isa
0x0004	CPER_PROC_VALID_ERROR_TYPE	proc_error_type
0x0008	CPER_PROC_VALID_OPERATION	—
0x0010	CPER_PROC_VALID_FLAGS	—
0x0020	CPER_PROC_VALID_LEVEL	—
0x0040	CPER_PROC_VALID_VERSION	—
0x0080	CPER_PROC_VALID_BRAND_INFO	—
0x0100	CPER_PROC_VALID_ID	—
0x0200	CPER_PROC_VALID_TARGET_ADDRESS	—
0x0400	CPER_PROC_VALID_REQUESTOR_ID	—
0x0800	CPER_PROC_VALID_RESPONDER_ID	—
0x1000	CPER_PROC_VALID_IP	—

6.8 CPER PCIe Validation Bits

include/linux/cper.h:246-255

位	名称	字段
0x1	CPER_PCIE_VALID_PORT_TYPE	—
0x2	CPER_PCIE_VALID_VERSION	—
0x4	CPER_PCIE_VALID_COMMAND_STATUS	—
0x8	CPER_PCIE_VALID_DEVICE_ID	—
0x10	CPER_PCIE_VALID_SERIAL_NUMBER	—
0x20	CPER_PCIE_VALID_BRIDGE_CONTROL_STATUS	—
0x40	CPER_PCIE_VALID_CAPABILITY	—
0x80	CPER_PCIE_VALID_AER_INFO	aer_info[] populated

CPER_PCIE_SLOT_SHIFT = 3

6.9 CPER ARM Error Types

include/linux/cper.h:273-315

值	名称
0	CPER_ARM_CACHE_ERROR
1	CPER_ARM_TLB_ERROR
2	CPER_ARM_BUS_ERROR
3	CPER_ARM_VENDOR_ERROR (= CPER_ARM_MAX_TYPE)

ARM validation bits 和 err_info 字段位移定义在 cper.h:257-315（CPER_ARM_ERR_TRANSACTION_SHIFT=16, OPERATION_SHIFT=18 等）。

6.10 CPER MCE creator / section UUIDs

arch/x86/kernel/cpu/mce/apei.c:127-132

名称	UUID
CPER_CREATOR_MCE	75a574e3-5052-4b29-8a8e-be2c6490b89d
CPER_SECTION_TYPE_MCE	fe08ffbe-95e4-4be7-bc73-4096044a38fc

apei_write_mce() (apei.c:144-174) 写 ERST 用 CPER_HW_ERROR_FLAGS_PREVERR | CPER_SEC_PRIMARY | CPER_SEV_FATAL。

6.11 CPER Notification Type GUIDs

include/linux/cper.h:60-91

名称	含义
CPER_NOTIFY_CMC	Corrected Machine Check
CPER_NOTIFY_CPE	Corrected Platform Error
CPER_NOTIFY_MCE	Machine Check Exception
CPER_NOTIFY_PCIE	PCI Express Error
CPER_NOTIFY_INIT	INIT Record（IPF）
CPER_NOTIFY_NMI	Non-Maskable Interrupt
CPER_NOTIFY_BOOT	BOOT Error Record
CPER_NOTIFY_DMAR	DMA Remapping Error

---

7. APEI / GHES

7.1 ghes_severity 枚举

include/acpi/ghes.h:51-56

值	名称	含义	Mapped to MCE.status
0x0	GHES_SEV_NO	无 / 未知	—
0x1	GHES_SEV_CORRECTED	已纠正	无 MCI_STATUS_UC
0x2	GHES_SEV_RECOVERABLE	可恢复	设 MCI_STATUS_UC（apei.c:53-54）
0x3	GHES_SEV_PANIC	致命 / panic	设 `MCI_STATUS_UC

GHES_EXITING = 0x0002（ghes.h:16）— ghes->flags 标记系统退出

CPER → GHES 映射（ghes.c:291-302）：

• CPER_SEV_INFORMATIONAL → GHES_SEV_NO
• CPER_SEV_CORRECTED → GHES_SEV_CORRECTED
• CPER_SEV_RECOVERABLE → GHES_SEV_RECOVERABLE
• CPER_SEV_FATAL → GHES_SEV_PANIC

7.2 APEI → MCE 桥

apei_mce_report_mem_error() (apei.c:29-64)：

status = MCI_STATUS_VAL | MCI_STATUS_EN | MCI_STATUS_ADDRV | MCI_STATUS_MISCV | 0x9f
misc = (MCI_MISC_ADDR_PHYS << 6) | lsb
bank = -1  // synthetic

SMCA：apei_smca_report_x86_error() (apei.c:66-125) 提取 6 个 SMCA 寄存器（MCA_STATUS, MCA_ADDR, MCA_MISC, MCA_CONFIG, MCA_IPID, MCA_SYND）

7.3 APEI HEST Source Types

include/acpi/actbl1.h:1406-1419

值	名称	含义
0	ACPI_HEST_TYPE_IA32_CHECK	IA-32 MCE 源
1	ACPI_HEST_TYPE_IA32_CORRECTED_CHECK	IA-32 CMCI
2	ACPI_HEST_TYPE_IA32_NMI	IA-32 NMI
6	ACPI_HEST_TYPE_AER_ROOT_PORT	PCIe AER root port
7	ACPI_HEST_TYPE_AER_ENDPOINT	PCIe AER endpoint
8	ACPI_HEST_TYPE_AER_BRIDGE	PCIe AER bridge
9	ACPI_HEST_TYPE_GENERIC_ERROR	GHES
10	ACPI_HEST_TYPE_GENERIC_ERROR_V2	GHES v2（read_ack 寄存器）
11	ACPI_HEST_TYPE_IA32_DEFERRED_CHECK	IA-32 延迟 MCE

7.4 APEI HEST Notification Types

include/acpi/actbl1.h:1491-1505

值	名称
0	ACPI_HEST_NOTIFY_POLLED
1	ACPI_HEST_NOTIFY_EXTERNAL
2	ACPI_HEST_NOTIFY_LOCAL
3	ACPI_HEST_NOTIFY_SCI
4	ACPI_HEST_NOTIFY_NMI
5	ACPI_HEST_NOTIFY_CMCI（ACPI 5.0）
6	ACPI_HEST_NOTIFY_MCE（ACPI 5.0）
7	ACPI_HEST_NOTIFY_GPIO（ACPI 6.0）
8	ACPI_HEST_NOTIFY_SEA（ACPI 6.1, ARM）
9	ACPI_HEST_NOTIFY_SEI（ACPI 6.1, ARM）
10	ACPI_HEST_NOTIFY_GSIV（ACPI 6.1）
11	ACPI_HEST_NOTIFY_SOFTWARE_DELEGATED（ACPI 6.2）

---

8. arm_event trace event

include/ras/ras_event.h:171-208，触发：drivers/ras/ras.c:24 log_arm_hw_error() ← ghes.c:507

字段	类型	含义
mpidr	u64	Multiprocessor Affinity Register
midr	u64	Main ID Register
running_state	u32	PSCI state (bit 0 set)
psci_state	u32	PSCI state
affinity	u8	亲和性 level

Validation bits（include/linux/cper.h:257-260）：CPER_ARM_VALID_AFFINITY_LEVEL, CPER_ARM_VALID_MPIDR, CPER_ARM_VALID_RUNNING_STATE

---

9. non_standard_event trace event

include/ras/ras_event.h:219-253，触发：drivers/ras/ras.c:17 ← ghes.c:676

字段	类型	含义
sec_type[16]	char 数组	section type GUID raw bytes
fru_id[16]	char 数组	FRU ID GUID
fru_text	string	FRU text
sev	u8	CPER severity
len	u32	raw data 长度
buf	dynamic u8	raw error data hex dump

---

10. signal_generate trace event

include/trace/events/signal.h:50-80

字段	类型	含义
sig	int	信号号
errno	int	siginfo si_errno
code	int	siginfo si_code
comm[]	char[16]	目标 task comm
pid	pid_t	目标 PID
group	int	1 = 进程组, 0 = 单进程
result	int	TRACE_SIGNAL_* 结果

10.1 TRACE_SIGNAL_* 结果码

include/trace/events/signal.h:27-33

值	名称	含义
0	TRACE_SIGNAL_DELIVERED	已投递
1	TRACE_SIGNAL_IGNORED	被忽略
2	TRACE_SIGNAL_ALREADY_PENDING	已 pending
3	TRACE_SIGNAL_OVERFLOW_FAIL	sigqueue 溢出
4	TRACE_SIGNAL_LOSE_INFO	siginfo 丢失

10.2 SIGBUS codes (si_code 值)

include/uapi/asm-generic/siginfo.h:251-258

值	名称	含义
1	BUS_ADRALN	地址对齐无效
2	BUS_ADRERR	物理地址不存在
3	BUS_OBJERR	对象特定硬件错误
4	BUS_MCEERR_AR	MCE 硬件错误已消费（action required）
5	BUS_MCEERR_AO	MCE 硬件错误已发现但未消费（action optional）

---

11. devlink_health_report trace event

include/trace/events/devlink.h:81-107

字段	类型	含义
bus_name	string	devlink 设备 bus name
dev_name	string	devlink 设备 name
driver_name	string	驱动名
reporter_name	string	health reporter 名
msg	string	消息字符串

11.1 devlink_health_reporter_state 枚举

include/net/devlink.h:714-717

值	名称	含义
0	DEVLINK_HEALTH_REPORTER_STATE_HEALTHY	健康
1	DEVLINK_HEALTH_REPORTER_STATE_ERROR	错误状态

（v6.6 没有 REPLAYING — 后加）

---

12. block_rq_error trace event

include/trace/events/block.h:165-170（继承自 block_rq_completion class 行 105-135）

字段	类型	含义
dev	dev_t	块设备 (MAJOR,MINOR)
sector	sector_t	起始 sector
nr_sector	u32	512-byte sectors 数
error	int	blk_status_to_errno() 转换的 errno
rwbs[]	char	R/W/B/S 标志
cmd	dynamic char	命令字符串

12.1 blk_status_t 枚举（BLK_STS_*）

include/linux/blk_types.h:99-179

值	名称	errno	含义
0	BLK_STS_OK	0	Success
1	BLK_STS_NOTSUPP	-EOPNOTSUPP	op not supported
2	BLK_STS_TIMEOUT	-ETIMEDOUT	timeout
3	BLK_STS_NOSPC	-ENOSPC	critical space allocation
4	BLK_STS_TRANSPORT	-ENOLINK	recoverable transport
5	BLK_STS_TARGET	-EREMOTEIO	critical target
6	BLK_STS_RESV_CONFLICT	-EBADE	reservation conflict
7	BLK_STS_MEDIUM	-ENODATA	critical medium
8	BLK_STS_PROTECTION	-EILSEQ	DIX/PIX protection
9	BLK_STS_RESOURCE	-ENOMEM	kernel resource
10	BLK_STS_IOERR	-EIO	generic I/O
11	BLK_STS_DM_REQUEUE	-EREMCHG	DM internal retry
12	BLK_STS_AGAIN	-EAGAIN	nonblocking retry
13	BLK_STS_DEV_RESOURCE	-EBUSY	device-specific resource
14	BLK_STS_ZONE_RESOURCE	-	zone resource
15	BLK_STS_ZONE_OPEN_RESOURCE	-ETOOMANYREFS	too many open zones
16	BLK_STS_ZONE_ACTIVE_RESOURCE	-EOVERFLOW	too many active zones
17	BLK_STS_OFFLINE	-ENODEV	device offline
18	BLK_STS_DURATION_LIMIT	-ETIME	command duration limit exceeded

---

13. memory_failure_event trace event

include/ras/ras_event.h:399-423，触发：mm/memory-failure.c:1323

字段	类型	含义
pfn	unsigned long	Page Frame Number
type	int	enum mf_action_page_type
result	int	enum mf_result

13.1 MF_MSG_* 页面类型

include/linux/mm.h:3917-3938 + include/ras/ras_event.h:356-376

值	名称	字符串
0	MF_MSG_KERNEL	“reserved kernel page”
1	MF_MSG_KERNEL_HIGH_ORDER	“high-order kernel page”
2	MF_MSG_SLAB	“kernel slab page”
3	MF_MSG_DIFFERENT_COMPOUND	“different compound page after locking”
4	MF_MSG_HUGE	“huge page”
5	MF_MSG_FREE_HUGE	“free huge page”
6	MF_MSG_UNMAP_FAILED	“unmapping failed page”
7	MF_MSG_DIRTY_SWAPCACHE	“dirty swapcache page”
8	MF_MSG_CLEAN_SWAPCACHE	“clean swapcache page”
9	MF_MSG_DIRTY_MLOCKED_LRU	“dirty mlocked LRU page”
10	MF_MSG_CLEAN_MLOCKED_LRU	“clean mlocked LRU page”
11	MF_MSG_DIRTY_UNEVICTABLE_LRU	“dirty unevictable LRU page”
12	MF_MSG_CLEAN_UNEVICTABLE_LRU	“clean unevictable LRU page”
13	MF_MSG_DIRTY_LRU	“dirty LRU page”
14	MF_MSG_CLEAN_LRU	“clean LRU page”
15	MF_MSG_TRUNCATED_LRU	“already truncated LRU page”
16	MF_MSG_BUDDY	“free buddy page”
17	MF_MSG_DAX	“dax page”
18	MF_MSG_UNSPLIT_THP	“unsplit thp”
19	MF_MSG_UNKNOWN	“unknown page”

13.2 mf_action_result / enum mf_result

include/linux/mm.h:3910-3915

值	名称	字符串	含义
0	MF_IGNORED	“Ignored”	无法处理
1	MF_FAILED	“Failed”	处理失败
2	MF_DELAYED	“Delayed”	稍后处理
3	MF_RECOVERED	“Recovered”	已恢复

13.3 enum mf_flags

include/linux/mm.h:3826-3834

位	名称	含义
1<<0	MF_COUNT_INCREASED	refcount 已增加
1<<1	MF_ACTION_REQUIRED	需要动作
1<<2	MF_MUST_KILL	必须杀进程
1<<3	MF_SOFT_OFFLINE	软离线
1<<4	MF_UNPOISON	unpoison 请求
1<<5	MF_SW_SIMULATED	软件模拟注入
1<<6	MF_NO_RETRY	不重试

13.4 HWPOISON page flag

include/linux/page-flags.h：

• PG_hwpoison (bit 129) — “hardware poisoned page. Don’t touch”
• __PG_HWPOISON 掩码
• SetPageHWPoison(page) / ClearPageHWPoison(page) / PageHWPoison(page) — flag ops
• SetPageHWPoisonTakenOff(page) / ClearPageHWPoisonTakenOff(page) — 子页跟踪
• PageHasHWPoisoned(page) — 复合页检测

13.5 hwpoison-inject 调试接口

mm/hwpoison-inject.c：

• 路径：/sys/kernel/debug/hwpoison/
• 文件：corrupt-pfn（调 hwpoison_inject() with MF_SW_SIMULATED），unpoison-pfn
• 过滤：corrupt-filter-enable, corrupt-filter-dev-major/-minor, corrupt-filter-flags-mask/-value, corrupt-filter-memcg
• 需要 CAP_SYS_ADMIN 否则 -EPERM；!pfn_valid → -ENXIO
• 返回 -EOPNOTSUPP → 0

---

14. ERST（APEI Error Record Serialization Table）

include/acpi/apei.h:13-51

ERST 存储原始 cper_record_header blobs（signature = “CPER”）。ioctl 接口：

名称	值	含义
APEI_ERST_INVALID_RECORD_ID	0xffffffffffffffffULL	“无 record” sentinel
APEI_ERST_CLEAR_RECORD	_IOW('E', 1, u64)	按 id 清记录
APEI_ERST_GET_RECORD_COUNT	_IOR('E', 2, u32)	总记录数

ERST 没有 “record type” 枚举 — 格式由 cper_record_header.signature == "CPER" 决定。

---

15. pstore

include/linux/pstore.h:28-44 + fs/pstore/platform.c:44-54

15.1 enum pstore_type_id

值	名称	字符串	含义
0	PSTORE_TYPE_DMESG	“dmesg”	内核 log 快照
1	PSTORE_TYPE_MCE	“mce”	MCE 记录
2	PSTORE_TYPE_CONSOLE	“console”	控制台输出
3	PSTORE_TYPE_FTRACE	“ftrace”	ftrace 环形缓冲
4	PSTORE_TYPE_PPC_RTAS	“rtas”	PowerPC RTAS
5	PSTORE_TYPE_PPC_OF	“powerpc-ofw”	PowerPC Open Firmware
6	PSTORE_TYPE_PPC_COMMON	“powerpc-common”	PowerPC common
7	PSTORE_TYPE_PMSG	“pmsg”	用户态 pmsg
8	PSTORE_TYPE_PPC_OPAL	“powerpc-opal”	PowerPC OPAL
9	PSTORE_TYPE_MAX	sentinel	列表结束

15.2 PSTORE_FLAGS_*（frontends，pstore.h:205-208）

标志	值
PSTORE_FLAGS_DMESG	BIT(0)
PSTORE_FLAGS_CONSOLE	BIT(1)
PSTORE_FLAGS_FTRACE	BIT(2)
PSTORE_FLAGS_PMSG	BIT(3)

---

16. 内核侧 → rasdaemon 侧 错误码对照

域	内核源	错误码数	在内核的根
MCE	arch/x86/include/asm/mce.h	~50 (bits + helpers)	MSR 0xc0002000+
MCE Severity	arch/x86/kernel/cpu/mce/severity.c	22 rules	表驱动
AER	include/uapi/linux/pci_regs.h	8 CE + 16 UE bits	PCI config 0x04/0x10
EDAC	include/linux/edac.h	5 err_type + 4 层级 + 29 mem_type + 9 edac_type + 9 scrub + 3 op_state	驱动实现
CXL	drivers/cxl/core/trace.h + cxlmem.h	15 UC + 7 CE + 16 mailbox + 30 RC + 3 mem_evt + 7 trans + 5 health + 10 media + 8 dev_status	设备寄存器
extlog	include/ras/ras_event.h	16 err_type + 4 sev	ACPI 6456 接口
CPER	include/linux/cper.h	4 sev + 16 mem err_type + 22 mem valid + 12 section type	UEFI spec
APEI	include/acpi/ghes.h + actbl1.h	4 sev + 12 HEST type + 12 HEST notify	ACPI 表
ARM	include/ras/ras_event.h	4 err type (CPER)	CPER PEI section
memory-failure	include/linux/mm.h	20 MF_MSG + 4 MF result + 7 mf_flags	内核
hwpoison	include/linux/page-flags.h	PG_hwpoison 1 bit	内核 page flag
signal	include/trace/events/signal.h + uapi/asm-generic/siginfo.h	5 result + 5 BUS code	内核
devlink	include/trace/events/devlink.h + include/net/devlink.h	2 reporter_state	内核
block	include/linux/blk_types.h	19 BLK_STS_*	内核
ERST	include/acpi/apei.h	ioctl 3	ACPI 表
pstore	include/linux/pstore.h	9 type + 4 flags	内核

总计：约 400+ 错误码 来自内核源码。

---

17. 关键文件索引

文件	行数	作用
arch/x86/include/asm/mce.h	334	MCE 状态位 / SMCA enum / MSR 地址
arch/x86/include/uapi/asm/mce.h	43	struct mce / ioctls
arch/x86/kernel/cpu/mce/internal.h	250+	severity_level / vendor flags / mca_config
arch/x86/kernel/cpu/mce/core.c	2800+	主分发、severity 决策、CPUHP、cmdline
arch/x86/kernel/cpu/mce/intel.c	1000+	CMCI 实现、Intel filter quirks
arch/x86/kernel/cpu/mce/amd.c	800+	smca_hwid_mcatypes、threshold 处理
arch/x86/kernel/cpu/mce/severity.c	400+	Intel/AMD 严重性表
arch/x86/kernel/cpu/mce/inject.c	800+	mce-inject / injection_type
arch/x86/kernel/cpu/mce/genpool.c	150+	MCE_POOLSZ / genpool
arch/x86/kernel/cpu/mce/apei.c	200+	APEI/MCE 桥
arch/x86/kernel/cpu/mce/dev-mcelog.c	380+	/dev/mcelog
arch/x86/kernel/cpu/mce/p5.c winchip.c	100+	旧 MCE
include/trace/events/mce.h	70	mce_record trace event
include/linux/cper.h	560+	CPER 全部定义
include/acpi/ghes.h	80+	GHES_SEV_* / 桥接口
include/acpi/apei.h	80+	ERST ioctls
include/acpi/actbl1.h	1500+	HEST 表
include/uapi/linux/pci_regs.h	800+	PCI_ERR_* 位
include/linux/aer.h	50+	AER severity / 头结构
drivers/pci/pcie/aer.c	1400+	AER ISR / 恢复 / 分类
drivers/pci/pcie/aer_inject.c	600+	AER 软件注入
drivers/pci/pcie/err.c	280+	pcie_do_recovery()
include/linux/edac.h	600+	EDAC 全部 enum / 结构
drivers/edac/edac_mc.c	1300+	主分发
drivers/edac/skx_common.c i10nm_base.c sb_edac.c	各 1000-3500	Intel 各代解码
drivers/edac/amd64_edac.c	3500+	AMD64 解码
drivers/cxl/core/trace.h	700+	8 个 cxl trace events
drivers/cxl/cxlmem.h	830+	mailbox / 事件 / poison
drivers/cxl/cxl.h	470+	RAS 寄存器 / event log type
drivers/cxl/cxlpci.h	100+	DVSEC / regloc type
drivers/cxl/core/mbox.c	1200+	mailbox 分发
drivers/cxl/core/memdev.c	800+	memdev 驱动
drivers/cxl/pci.c	800+	CXL.io AER / event ISR
include/uapi/linux/cxl_mem.h	50+	UAPI CXL 命令
include/ras/ras_event.h	430+	全部 RAS trace events
include/linux/mm.h	4000+	mf_result / mf_action_page_type / mf_flags
mm/memory-failure.c	1500+	memory_failure / page 处理
mm/hwpoison-inject.c	110+	debugfs hwpoison 注入
include/linux/blk_types.h	300+	blk_status_t / REQ_*
include/trace/events/signal.h	80+	signal_generate / TRACE_SIGNAL_*
include/trace/events/devlink.h	170+	devlink_health_* trace events
include/trace/events/block.h	200+	block_rq_error
include/net/devlink.h	1000+	devlink 全部
include/linux/pstore.h	300+	pstore 全套

---

18. 调试与快速定位

问题	在哪查
MCE 错误位定义	arch/x86/include/asm/mce.h
MCE 严重性决策	arch/x86/kernel/cpu/mce/severity.c
SMCA bank → HWID 映射	arch/x86/kernel/cpu/mce/amd.c:160-219
AER 错误位	include/uapi/linux/pci_regs.h:748-802
AER 严重性映射	drivers/pci/pcie/aer.c:750-761 (cper_severity_to_aer)
EDAC 错误类型字符串	edac_mc.c:121-136 (mc_event_error_type)
CXL 错误位	drivers/cxl/core/trace.h:17-31 (UC) 81-87 (CE)
CXL 事件分发	drivers/cxl/core/mbox.c:843-861
CPER 内存 err_type 字符串	drivers/firmware/efi/cper.c:189-213
GHES 桥	drivers/acpi/apei/ghes.c
signal si_code	include/uapi/asm-generic/siginfo.h:251-258
blk_status errno	block/blk-core.c:151-178 (blk_status_to_errno)
memory_failure 字符串	mm/memory-failure.c:870-897
hwpoison 注入	mm/hwpoison-inject.c
ERST ioctl	include/acpi/apei.h:13-51
测点	/sys/kernel/debug/tracing/events/<system>/

---

19. 关键事实总结

1. rasdaemon 的所有错误码都源于内核 — 它是消费者，不是定义者
2. CPER 规范是通用传输格式 — 几乎所有硬件平台错误（PEI 段、mem 段、PCIe 段）都用 CPER
3. APEI 是固件 → 内核的桥 — 错误从固件通过 GHES / AER 通知进内核
4. 错误严重性在内核决策 — rasdaemon 看到的 error_type 是内核已经分类过的
5. 操作恢复由内核完成 — rasdaemon 只观察；少量 EDAC PFA 写 sysfs 是例外
6. SMCA 26 类 bank — 通过 (HWID, MCATYPE) 二元组路由
7. CXL v6.6.0 没有 sparing 事件 — 后加在更新内核
8. CXL 内核宏用 CXL_RAS_UC_*/CXL_RAS_CE_* 命名 — 不是 CXL_AER_UE_*
9. arm_event 仅解 PEI 段 — 不解 SMMU/GIC/CCI/CCN 专用段
10. hwpoison 是 page flag 而非 enum — 配合 mf_action_page_type 字符串一起用