Skip to content

Explain the 6.1 sections differences and leave objtool sections out - #47

Merged
tamnd merged 1 commit into
mainfrom
sections-61-clean
Oct 4, 2026
Merged

tamnd merged 1 commit into
mainfrom
sections-61-clean

Conversation

@tamnd

@tamnd tamnd commented Oct 4, 2026

Copy link
Copy Markdown
Owner

This gets rk sections-diff on 6.1.188 to 382 differences, all explained.

Twelve rows were .retpoline_sites or .return_sites found in one build's object and not the other's. On 6.1 objtool writes those into every object, and whether one is there just follows from whether the code has an indirect call or a ret. #46 already stopped counting their entries, so the sections an object is compared on now leave them out too, along with the rest of what objtool writes. There's a test for it.

The other five rows get rules:

  • .init.rodata in rtc.o: gcc unrolls the loop over the three ids in add_rtc_cmos, folds the loads and drops the array. rucc keeps it. This is the same cause as the .rodata rule.
  • The two crng BUG_ONs in random.c, the ext4_add_complete_io WARN and the check_copy_size WARN are the ones already explained on 6.12 and 7.2.8, at 6.1's line numbers.
  • The BUG_ON in ext4_split_extent_at for both DATA_VALID flags. None of its four callers can pass both, so rucc drops the check.

Part of #3.

@tamnd tamnd added kind/bug Something in the harness is wrong area/rk The rk command line labels Oct 4, 2026
@tamnd
tamnd merged commit 232359b into main Oct 4, 2026
3 of 5 checks passed
@tamnd
tamnd deleted the sections-61-clean branch October 4, 2026 09:31
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

area/rk The rk command line kind/bug Something in the harness is wrong

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant