model-writtenrun 727 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Daria Shanina reported that setting huge_page_size to 1 TB on a 32-bit system caused overflow and SIGFPE, attaching a patch that rejected a zero huge-page size during shared-memory allocation #1.
- Kyotaro Horiguchi agreed the guard would prevent SIGFPE but recommended rejecting unusable sizes earlier in check_huge_page_size(), rather than widening the later arithmetic #2.
- Michael Paquier identified the existing MAX_KILOBYTES limit as the appropriate solution and proposed replacing huge_page_size's INT_MAX upper bound; Kyotaro endorsed that approach #3 #4.
- Michael chose to apply the fix only to the development branch because the problematic configuration had attracted no complaints for years, despite considering stable-branch changes technically acceptable #5.
- The discussion-derived upper-bound change landed as dbaa4dc3c8, replacing the attached runtime-guard approach, and Michael confirmed the development-branch-only application #6.
- Daria subsequently confirmed that the MAX_KILOBYTES limit resolved the problem, leaving the fix landed with no further work requested #7.