机器型号:ASUS ROG Strix SCAR 18 G834JZR,i9-14900HX,两条 16 GB DDR5-5600,RTX 4080 Laptop,BIOS 322(2025-10-01),微码 0x143。
系统是 Windows 11 专业工作站版 26H2,当前 Build 26300.9711(2026-10 累积更新)。两块 NVMe 都健康,C 盘还有 51 GB。
蓝屏从 2026-10-31 开始,没有停过。最近五天(11-13 到 11-17)崩了 5 次。绝大多数时候 PROCESS_NAME 是 Registry,崩溃点在加载 C:\Windows\AppCompat\Programs\Amcache.hve 的过程中。
我把能找到的 minidump 全部过了一遍 WinDbg,一共 22 份:
| 家族 |
次数 |
占比 |
代表符号 |
| hive 校验(Amcache) |
17 |
77% |
nt!CmpCheckRegistry2+0x2a9、+0x2ad,少数 HvpGetCellPaged、HvpMapEntryGetBinAddress |
| 内存管理(VAD/引用计数) |
2 |
9% |
nt!KiFastFailDispatch+b2(MiUnlockAndDereferenceVadShared) |
| PPM 电源/频率 |
2 |
9% |
nt!PpmPerfSnapProcessorCyclesWorkloadClass、PpmParkComputeUnparkMaskEx |
| Hypervisor 空闲 |
1 |
5% |
intelppm!HvRequestIdle+0x2d |
完整时间线(按日期排序,BugCheck 取 kd 的 BUGCHECK_CODE):
| # |
日期 |
BugCheck |
家族 |
进程 |
| 1 |
2026-10-31 |
0x50 |
hive 校验(Amcache) |
Registry |
| 2 |
2026-11-02 |
0x50 |
hive 校验(Amcache) |
Registry |
| 3 |
2026-11-02 |
0x20001 |
Hypervisor 空闲 |
System |
| 4 |
2026-11-03 |
0x1e |
hive 校验(Amcache) |
Registry |
| 5 |
2026-11-03 |
0x50 |
hive 校验(Amcache) |
Registry |
| 6 |
2026-11-03 |
0x50 |
hive 校验(Amcache) |
Registry |
| 7 |
2026-11-04 |
0x139 |
内存管理(VAD/引用计数) |
dotnet.exe |
| 8 |
2026-11-05 |
0x1e |
hive 校验(Amcache) |
Registry |
| 9 |
2026-11-05 |
0x3b |
hive 校验(Amcache) |
Registry |
| 10 |
2026-11-07 |
0x3b |
hive 校验(Amcache) |
Registry |
| 11 |
2026-11-07 |
0x3b |
hive 校验(Amcache) |
Registry |
| 12 |
2026-11-07 |
0x3b |
hive 校验(Amcache) |
Registry |
| 13 |
2026-11-09 |
0x50 |
hive 校验(Amcache) |
Registry |
| 14 |
2026-11-09 |
0x50 |
hive 校验(Amcache) |
Registry |
| 15 |
2026-11-09 |
0x50 |
hive 校验(Amcache) |
Registry |
| 16 |
2026-11-10 |
0xa |
PPM 电源/频率 |
Code.exe |
| 17 |
2026-11-10 |
0x50 |
hive 校验(Amcache) |
Registry |
| 18 |
2026-11-15 |
0x139 |
内存管理(VAD/引用计数) |
sampler-batch-probe.exe |
| 19 |
2026-11-16 |
0x1e |
hive 校验(Amcache) |
Registry |
| 20 |
2026-11-16 |
0xa |
PPM 电源/频率 |
HipsDaemon.exe |
| 21 |
2026-11-17 |
0x50 |
hive 校验(Amcache) |
Registry |
| 22 |
2026-11-17 |
0x50 |
hive 校验(Amcache) |
Registry |
0x50 样本的调用链是这样:
代码块
nt!KeBugCheckEx
nt!MiSystemFault
nt!MmAccessFault
nt!KiPageFault
nt!CmpCheckRegistry2+0x2a9 <- 崩溃点
nt!CmCheckRegistry+0x19f
nt!CmpCreateHive+0x3d3
nt!CmpInitHiveFromFile+0x387
nt!CmpCmdHiveOpen+0xe6
nt!CmLoadAppKey+0x716
nt!CmLoadDifferencingKey+0x5c8
nt!NtLoadKeyEx+0x64
nt!KiSystemServiceCopyEnd
崩溃指令地址的低 20 位每次都是 0x6EC39,跨重启、跨 KASLR 不变。FAILURE_ID_HASH 是 {1bd2a583-35dd-23f1-c0af-bbcbe0774988},FAILURE_BUCKET_ID 出现过 AV_nt!CmpCheckRegistry2 和 0x1E_C0000096_nt!CmpCheckRegistry2。
我另外写了个 REGF 解析器,把 Amcache.hve 逐 cell 验了一遍:66,617 个 cell、11,708 个键节点,没有越界或尺寸异常,头部主/次序列号一致。文件本身没坏。
试过的东西都没有效果,列在这里免得重复建议:
-
DISM /Online /Cleanup-Image /RestoreHealth 用 ISO 作源修过组件存储,结果干净
- 关掉 HVCI 和 VBS,关掉 Hyper-V、Windows 沙盒、IIS、WCF、MSMQ
- 暂停 Store 自动更新,禁用 CompatTelRunner、Appraiser 一组计划任务,停掉 InventorySvc
- 给 Amcache.hve 和它的 LOG 加 ACL 禁止写入,内核用备份/恢复特权绕过去了,照崩
- 卸载 Intel XTU,排除降压和超频
- 内存诊断跑过(事件 1101/1201 报告无错误),MemTest86 也过了几轮;CPU 压力测试 180分钟没有新增 WHEA
- 电源方案里最大处理器状态 95% 和 10% 都试过
WHEA 的情况单独说一句 :11-10 10:57 那次崩溃之前一条新增 WHEA 都没有,所以它至少不是必要条件。
外面已经有两个同类报告。6054557 那位是 MECHREVO 的 14900HX 加 RTX 5070Ti,同样 26300.9711,18 次 0x50,函数和 hash 都一样。6069097 那位是 MSI Z890 加 Ultra 9 285K,装完 KB5132061(26200.9974)之后一天崩 8 次 0x3B,进程也是 Registry,故障也在 hive cell map 查找里,卸载这个 LCU 后恢复正常。
想问四个问题:
- 这算不算微软已知的内核缺陷,有没有对应的修复计划或 KB?
- 11 月的 KB5132061(26300.9974)到底能不能装?6069097 那位装完崩得更频繁。
- 要推动定位,还需要什么数据?完整 MEMORY.DMP、WPR 的 ETL、Amcache.hve 副本,我都可以提供。
- 修复出来之前,有没有官方认可的缓解办法?
English summary
Windows 11 26H2, Build 26300.9711. 22 minidumps analyzed with WinDbg against public symbols.
17 of the 22 (77%) fault in nt!CmpCheckRegistry2 (+0x2a9 / +0x2ad, a few in HvpGetCellPaged / HvpMapEntryGetBinAddress), with PROCESS_NAME = Registry, while loading C:\Windows\AppCompat\Programs\Amcache.hve. Bugchecks are 0x50 and 0x1E (C0000096). Stack: NtLoadKeyEx → CmLoadDifferencingKey → CmLoadAppKey → CmpCmdHiveOpen → CmpInitHiveFromFile → CmpCreateHive → CmCheckRegistry → CmpCheckRegistry2 → KiPageFault. The low 20 bits of the faulting instruction are always 0x6EC39. FAILURE_ID_HASH is {1bd2a583-35dd-23f1-c0af-bbcbe0774988}.
The remaining dumps: VAD / reference count 0, PPM 0, hypervisor idle 1, unparsed 0.
Amcache.hve was validated with a hand-written REGF parser: 66,617 cells, 11,708 key nodes, no bounds or size errors, primary and secondary sequence numbers match. The file is structurally sound.
Things already tried without effect: DISM RestoreHealth from ISO media, HVCI and VBS off, Hyper-V / Sandbox / IIS / WCF / MSMQ disabled, Store auto-update paused, CompatTelRunner and Appraiser tasks disabled, InventorySvc stopped, ACL write-deny on the hive (the kernel bypasses it via backup/restore privilege), Intel XTU uninstalled, memory diagnostics and several MemTest86 passes, 30 minutes of CPU stress with no new WHEA, maximum processor state at 95% and at 10%.
WHEA: 13 corrected events since 11-15, 12 of them on APIC 33 (P-core #4), both parity and TLB. One of the crashes had no preceding WHEA event at all.
Related reports: 6054557 (same function, same hash, another 14900HX on 26300.9711) and 6069097 (KB5132061 / 26200.9974 causing repeated 0x3B in Registry during hive cell map lookups, resolved by uninstalling the LCU).
Questions: is this a known defect with a fix planned? Should KB5132061 be installed? What additional data would help? Is there an officially supported mitigation?