sysutils/cpu-microcode-intel: Handle extended signature tables

[jrm]

Summary

Intel microcode files can contain an extended signature table after the
main payload, listing additional processor signatures and platform flag
combinations that are covered by the same microcode blob.

ucode-split was only reading the primary header signature when naming
output files and ignored the extended table entirely. As a result,
processors whose signature appeared only in an extended table entry had
no matching split file in ${DATADIR} and therefore no microcode via
cpucontrol(8).

This affected a number of CPU families in the current release, including
Raptor Lake (06-bf-02/05/06/07, Core Gen13/Gen14), Sapphire Rapids
steppings E0-E3 (06-8f-04 through 06-8f-07), Arrow Lake-H (06-c5-02),
Panther Lake (06-cc-02/03), and others.

Extend ucode-split to read the extended signature table and write an
additional output file for each entry that differs from the primary
signature.

PR: 295351

[markj]

sysutils/cpu-microcode-intel/files/ucode-split.c

271

Wouldn't it be simpler to make copy_entry() use pread(2)?

[jrm]

• Use pread(2) as suggested by markj, which means we can remove extsig_off save/restore around copy_entry()
• Reference the 'Optional Extended Signature Table' in vol 3A section 12.11.2 of the March 2026 SDM

[jrm]

Fix a typo (s/feee/free/)

[jrm]

sysutils/cpu-microcode-intel/files/ucode-split.c

271

Indeed. Thanks.

[markj]

My comments are just nits, I think the change is fine.

sysutils/cpu-microcode-intel/files/ucode-split.c

246

You might handle this error case in the same way as in copy_entry(), i.e., check for rv < 0 separately so you can print errno.

251

Same here.

259

I'm not sure there's much point in skipping identical entries? copy_entry() will just overwrite the file again.