Re: [RFC PATCH v2 3/7] objtool/klp: allow special section entry size overrides

[Josh Poimboeuf <jpoimboe@kernel.org>]

On Wed, Aug 26, 2026 at 03:49:56PM -0400, Joe Lawrence wrote:
> @@ -1781,6 +1799,27 @@ static int create_fake_symbols(struct elf *elf)
>  		}
>  
>  		entry_size = sec->sh.sh_entsize;
> +
> +		/*
> +		 * Some special sections have multiple relocs per entry,
> +		 * so the reloc-based heuristic below doesn't work.  Use
> +		 * the arch-defined entry sizes for known special sections.
> +		 */

This comment doesn't seem correct: for the special sections with
multiple relocs per entry, the .o file should have annotations for that
(either entsize or ANNOTATE_DATA_SPECIAL).

Also the comment doesn't seem to match the commit log (that this is
needed for unit testing?)

> +		if (!entry_size) {
> +			if (!strcmp(sec->name, ".altinstructions"))
> +				entry_size = entry_size_from_env("ALT_ENTRY_SIZE", ALT_ENTRY_SIZE);
> +			else if (!strcmp(sec->name, "__jump_table"))
> +				entry_size = entry_size_from_env("JUMP_ENTRY_SIZE", JUMP_ENTRY_SIZE);
> +			else if (!strcmp(sec->name, "__ex_table"))
> +				entry_size = entry_size_from_env("EX_ENTRY_SIZE", EX_ENTRY_SIZE);

I really don't like this because it removes klp-diff's hard-earned
agnosticism about these entry sizes.  And they should already be encoded
in the binary anyway.

-- 
Josh