A call to memcpy() in a single binary that uses glibc may behave in 12 different ways depending on the features of the specific x86-64 CPU you run it on.

Here is a list of those impls in glibc:

https://github.com/bminor/glibc/blob/12a497c716f0a06be5946cabb8c3ec22a079771e/sysdeps/x86_64/multiarch/ifunc-impl-list.c#L1174-L1218

Fwiw this may matter a lot during binary exploitation. This was important in a challenge from PlaidCTF 2025. E.g. passing a negative (or: very huge) length allowed you to write past a buffer without a crash (the given implementation was not doing a wild copy).

#linux #binaryexploitation #ctf

@disconnect3d https://github.com/v-p-b/reversing-memcpy-again

@disconnect3d The chosen implementation also depends on the copy size, right?