events-arch-x86: fix an uninitialized bank_type and an mce_priv leak

[agenticode]

Two x86 error-path fixes, one commit each.

mce-amd-smca: bank_type used uninitialized

decode_smca_error() only sets bank_type when the IPID matches an entry in smca_hwid_mcatypes[]. The no-match check is i >= MAX_NR_BANKS (64), but the loop stops at ARRAY_SIZE(smca_hwid_mcatypes) (37), so it is never true and stack garbage indexes smca_names[] and smca_mce_descs[].

One mce_record with family 0x17 and IPID 0xffff0fff00000000 (in no entry, not a non-CPU node), under valgrind:

Use of uninitialised value of size 8
   at decode_smca_error (mce-amd-smca.c:964)    <- smca_names[bank_type].name
Use of uninitialised value of size 8
   at decode_smca_error (mce-amd-smca.c:969)    <- smca_mce_descs[bank_type].num_descs

Outside valgrind the same record printed "Don't know how to decode this bank", so the result depends on what is on the stack. With the patch bank_type starts at N_SMCA_BANK_TYPES and that message is what you always get.

I left the i >= MAX_NR_BANKS test alone: aecf33a moved it off ARRAY_SIZE() so that non-CPU nodes with no table match keep SMCA_UMC_V2.

ras-erst: mce_priv leak

ras_erst_init() calls init_mce_priv(), but x86-mce-erst has no .cleanup. The only free_mce_priv() caller is the x86-mce-event cleanup, and modules_cleanup_type() skips it when that module failed to init.

Failing the allocation x86-mce-event needs:

rasdaemon: module x86-mce-event init failed: -12
==28371==ERROR: LeakSanitizer: detected memory leaks

Direct leak of 96 byte(s) in 1 object(s) allocated from:
    #2 init_mce_priv ../events-arch-x86/ras-mce-handler.c:401
    #3 ras_erst_init ../events-arch-x86/ras-erst.c:230

Indirect leak of 560 byte(s) in 1 object(s) allocated from:
    #1 __GI___getdelim

free_mce_priv() clears ras->mce_priv, so both modules calling it is fine.

Testing

Fedora 44, on top of d06e494:

• gcc 16.2 and clang 22.1: no build warnings
• meson test: 9/9 with both compilers
• python3 tests/run.py: 140 tests, rc=0
• scripts/checkpatch.pl --no-tree --strict: no issues