Offset calculation bug in copy_forward_misaligned_words

[uris-genlabs]

There's a bug in the function copy_forward_misaligned_words under library/compiler-builtins/compiler-builtins/src/mem/impls.rs (line 153), with the mem-unaligned feature not enabled.

The code calculates src_aligned incorrectly - instead of adding the missing alignment to src, it subtracts. This causes the wrong bytes to be read by load_aligned_end_partial, and may potentially access invalid memory. Compare this with copy_backward_misaligned_words, where the alignment is correctly calculated by subtraction.

This bug is probably the cause of this issue:
#145769

[RalfJung]

You closed this as completed -- why that? Did it get fixed or is there not a bug after all?

[uris-genlabs]

Hi, yesterday I thought there was a bug, but after further investigation I realized the root cause is actually different, though I do believe I've encountered the same alignment issue as described here <#145769>. This only happens in the misaligned implementation of memcpy, which was not the one used in 1.87.0.

…

On Tue, Sep 16, 2025 at 11:46 PM Ralf Jung ***@***.***> wrote: *RalfJung* left a comment (rust-lang/rust#146643) <#146643 (comment)> You closed this as completed -- why that? Did it get fixed or is there not a bug after all? — Reply to this email directly, view it on GitHub <#146643 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BKAK6X2TJUXVDXCN2U3AT3T3TBZK7AVCNFSM6AAAAACGVMI4G2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGMBQGMYDCNRTGQ> . You are receiving this because you modified the open/close state.Message ID: ***@***.***>