[Bug ld/25665] aarch64 veneers not inserted with large .text object between caller and target

[tnfchris at sourceware dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=25665

--- Comment #11 from Tamar Christina <tnfchris at sourceware dot org> ---
(In reply to rrorden from comment #8)
> So the proposal is to add an aarch64 linker option that places stubs after
> input sections that need them, just like aarch32, and I would need to
> implement the -ffunction-sections behavior in our aot compiler so each
> function (and they are relatively small) gets a unique section name. The A53
> erratum would not force every stub section to align to 4k, but that is a
> worst-case scenario. If this is correct, I am okay with it.
> 

Yup, that correct, On the long run -ffunction-section would give you other
benefits like letting the linker remove unused functions so you would likely
end up with a size gain here if you don't use all functions.  They would be
merged in the end anyway so there's no downside after linking.

> How does the linker know it is targeting an A53 core? I see ld options
> --fix-cortex-a53-835769, --fix-cortex-a53-843419. Do these need to be set
> explicitly?

Yup, you need to explicit give it the option, though certain toolchains like
the GNU A released from Arm come with them on by default.

-- 
You are receiving this mail because:
You are on the CC list for the bug.