ELFs with specific alignment can fail rewriting

Discussed separately, but just placing this as a tracking issue for reference:

cargo run -p litebox_syscall_rewriter ./python3 -o /dev/null
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.08s
     Running `target/debug/litebox_syscall_rewriter ./python3 -o /dev/null`
Error: failed to generate object file: Unaligned sh_offset value 0x201000 for section '.text', expected 0x400000

python312.zip

@wdcui said

You’re hitting a layout quirk from BOLT/huge-page alignment:
In your ELF, .text has sh_addralign = 0x200000 (2 MiB).
But its sh_offset = 0x201000, which is not a multiple of 0x200000 (it’s 0x1000 past it).
Some writers (likely via the object crate under the hood) validate that sh_offset % sh_addralign == 0 when (re)emitting an object, and they error out. > Their “expected 0x400000” is the next 2 MiB multiple after 0x201000.

@CvvT said

Looks like a bug from object crate. Should we submit an issue to upstream?