You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
regions: rename REGION_MODULE_REGION to REGION_MODULE_BAND
The pair read backwards: "module_region" says region twice and looks like
the more specific of the two, when it is the weaker one — the band itself,
or an address merely assumed to be a module because it fell inside it. The
codebase already calls that concept a band everywhere in prose, and the
MODULES_BAND_EXACT arch axis uses the same word, so the enum was the odd
one out. The wire name follows: module_region -> module_band.
kasld_addr_is_module_region() becomes kasld_addr_is_module_band(), and its
comment gains the distinction it was missing: a source that already knows
it read a module's address uses the predicate as a sanity filter and still
emits REGION_MODULE, while a source holding a bare pointer uses it to
classify and must emit REGION_MODULE_BAND.
Two comments were stale rather than merely misnamed. The region note said
module_text_bound "reads both" and that its arches have a narrow
text-relative band; neither is true — it reads REGION_MODULE only, and the
band contains the whole kernel-text range on both riscv64 and s390. And the
wire-token parse was documented against "longest module_region = 13" when
virt_page_offset and efi_loader_image are 16.
0 commit comments