Re: [Patch] LoongArch: Restore pointer_equality_needed for GOT data relocations
Project / Subsystem
binutils / loongarch
Date
2026-08-03
Proposer
Yanjun Yang <yangyj.ee@gmail.com>
Source type
public_inbox
Consensus
Proposed
Sentiment
—/10
Technical tradeoffs
- • Reintroducing `pointer_equality_needed` for data symbols to fix static executable crashes, potentially reintroducing issues for function pointers if not handled carefully.
- • Distinguishing between function and data symbols for correct GOT relocation handling.
All attributes
- project
- binutils
- subsystem
- loongarch
- patch_id
- —
- discussion_id
- anBEsrlkwMTUZ1Kl@void-bee.tail05c47.ts.net
- source_type
- public_inbox
- title
- Re: [Patch] LoongArch: Restore pointer_equality_needed for GOT data relocations
- headline
- LoongArch: Restore pointer_equality_needed for GOT data relocations
- tldr
- Discusses a regression in LoongArch GOT relocations; the proposer acknowledges Xi Ruoyao's critique and plans to bisect.
- proposer
- Yanjun Yang <yangyj.ee@gmail.com>
- consensus
- Proposed
- outcome
- proposed
- sentiment_score
- —
- technical_tradeoffs
-
- • Reintroducing `pointer_equality_needed` for data symbols to fix static executable crashes, potentially reintroducing issues for function pointers if not handled carefully.
- • Distinguishing between function and data symbols for correct GOT relocation handling.
- series_id
- —
- series_role
- reply
- series_parts
- []
- tags
-
- • binutils
- • linker
- • loongarch
- • relocation
- • GOT
- bugzilla_url
- —
- date
- 2026-08-03T00:00:00.000Z
Re: [Patch] LoongArch: Restore pointer_equality_needed for GOT data relocations
This email is a reply in a discussion about a regression in LoongArch GOT (Global Offset Table) relocations. The original patch aimed to fix PR 29655 by removing pointer_equality_needed for GOT relocations, but this caused runtime segfaults in static executables due to altered GOT layout for data symbols. Xi Ruoyao questioned the logic, particularly concerning data symbols. The author acknowledges the critique, states that commit 7f9d6d6ead5 is the regression source, and plans to bisect further.