Re: [RFC PATCH 07/10] arm: efi_loader: discard hash, unwind information

[Heinrich Schuchardt]

Am 22. Mai 2023 21:25:50 MESZ schrieb Sam Edwards <cfsworks at gmail.com>:
>Hi Ilias,
>
>On 5/22/23 01:00, Ilias Apalodimas wrote:
>> The reason we end up with both hash and gnu.hash is because the hash
>> style is set to 'both'.  Should we perhaps use (and strip) only one of
>> them?
>
>If we do keep one, it should probably be .hash -- see commit b02bfc4dfc.
>
>I admit I'm completely mystified as to why we need the hash tables at all. The ELF spec says those are just for the dynamic linker, but neither the EFI code nor the self-relocating thunk require it, and I don't know of any target where the u-boot ELF itself is the shipped binary. For all I know, there never was a need to include .hash and Albert's commit fixed whatever problem he was facing only accidentally. Do you have any insights?
>

Ubuntu's and Debian's u-boot-qemu package ships uboot.elf for multiple architectures. Cf. https://packages.debian.org/bullseye/all/u-boot-qemu/filelist

You can pass uboot.elf as -kernel parameter to QEMU.

Best regards

Heinrich 


>LLD's --hash-style option doesn't appear to have a "none" option or I'd just be making use of that here.
>
>Cheers,
>Sam