Skip to content

Fix resolve_address to account for multiple mappings of the same file… - #4035

Merged
rocallahan merged 1 commit into
rr-debugger:masterfrom
pcc:lld-fix
Dec 19, 2025
Merged

rocallahan merged 1 commit into
rr-debugger:masterfrom
pcc:lld-fix

Conversation

@pcc

@pcc pcc commented Dec 19, 2025

Copy link
Copy Markdown
Contributor

… offset

Some linkers such as lld will create program headers with multiple mappings of the same file offset. This can lead to problems when a symbol of interest to rr, such as __aarch64_ldadd4_relax, is covered by more than one mapping, as that will lead to us finding the function in multiple mappings. For that symbol in particular, we can end up misinterpreting the instructions in the wrong mapping and incorrectly computing an address to write to, which can lead to an assertion failure, or worse, silent memory corruption. Fix it by changing resolve_address to check whether the mapping is the correct one (fully covers the appropriate program header and has the same memory permissions) before returning the address.

… offset

Some linkers such as lld will create program headers with multiple
mappings of the same file offset. This can lead to problems when a symbol
of interest to rr, such as __aarch64_ldadd4_relax, is covered by more
than one mapping, as that will lead to us finding the function in multiple
mappings. For that symbol in particular, we can end up misinterpreting the
instructions in the wrong mapping and incorrectly computing an address
to write to, which can lead to an assertion failure, or worse, silent
memory corruption. Fix it by changing resolve_address to check whether
the mapping is the correct one (fully covers the appropriate program
header and has the same memory permissions) before returning the address.
@rocallahan
rocallahan merged commit 5dbba89 into rr-debugger:master Dec 19, 2025
5 checks passed
@rocallahan

Copy link
Copy Markdown
Collaborator

Thanks!

rocallahan pushed a commit that referenced this pull request Jun 24, 2026
When using a non-glibc loader (e.g. musl), __aarch64_ldadd4_relax will
not necessarily be present, but the loader could still be using other
outlined atomic functions, so the search for __aarch64_ldadd4_relax
would not necessarily succeed. Therefore, remove the logic for
disassembling __aarch64_ldadd4_relax, and replace it with code that sets
__aarch64_have_lse_atomics directly.

This logic was considered necessary because __aarch64_have_lse_atomics
isn't exported. However, __aarch64_ldadd4_relax isn't exported either. So
either way, the symbol should be findable via .symtab if it is present.

It's possible that this wouldn't have worked before because of the bug
fixed by #4035.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants