Hi Henrique,
sid is still at 3.20260227.1 (last upload 2026-03-07).
microcode-20260925 carries the published fix for Meteor Lake (06-aa-04) and
ships the published fix for the revision current OEM firmware has been
distributing (20260811 also carries security updates (INTEL-SA-01379)):
https://github.com/intel/Intel-Linux-Processor-Microcode-Data-
Files/releases/tag/microcode-20260925
| MTL | C0 | 06-aa-04/e6 | old 00000028 | new 0000002b | Core Ultra |
Purpose: "Update for functional issues."
Why this matters: upstream shipped 06-aa-04 revision 0x2a in 20260811
and withdrew the file one day later (20260812: "functional issues
observed when loading the MCU in some platforms"). OEMs had already
baked 0x2a into firmware by then, so for machines whose firmware now
applies 0x2a, the currently packaged 0x28 is a guaranteed no-op (the
early loader does not apply equal-or-lower revisions) - those machines
permanently run the revision Intel retracted. Only 0x2b would take
over. (OS early-loading of 0x2a on pre-August firmware also caused
instant-reboot regressions, ~30 reports:
https://github.com/intel/Intel-Linux-Processor-Microcode-Data-
Files/issues/112
My machine is in that camp: HP ProBook 460 16 G11, Core Ultra 7 155H
(sig 0xa06a4), BIOS W72 Ver. 01.10.02. Verified locally:
/proc/cpuinfo reports microcode 0x2a, dmesg reports
"microcode: Current revision: 0x0000002a", and Debian's bundled file
for 06-aa-04 is revision 0x28.
Repeated fatal machine checks at runtime, signature stable across
7.2.12 and 7.2.14+deb14-amd64:
kern.log.1:2026-10-29T16:54:38 kernel: mce: [Hardware Error]: CPU 12: Machine
Check: 0 Bank 0: b200000000000400
kern.log.1:2026-10-29T16:54:38 kernel: mce: [Hardware Error]: PROCESSOR
0:a06a4 TIME 1793289190 SOCKET 0 APIC 0 microcode 2a
Identical signature again on 2026-10-31 (kernel panic both times);
on 2026-11-03, panic about 2 min after boot. Ruled out: RAM (EDAC
igen6, no CE/UE), NVMe (SMART clean), thermals.
Please upload 3.20260925.1 - for machines with firmware-0x2a it is
the first Debian-packaged revision the early loader would actually
accept. Thanks :-)
Best regards,
Dennis