On Wed, 09 Oct 2024 21:49:03 +0100,
Nathan Chancellor <nathan@xxxxxxxxxx> wrote:
>
> Hi Mark,
>
> On Tue, Oct 01, 2024 at 11:58:50PM +0100, Mark Brown wrote:
> > There is a control HCRX_EL2.GCSEn which must be set to allow GCS
> > features to take effect at lower ELs and also fine grained traps for GCS
> > usage at EL0 and EL1. Configure all these to allow GCS usage by EL0 and
> > EL1.
> >
> > We also initialise GCSCR_EL1 and GCSCRE0_EL1 to ensure that we can
> > execute function call instructions without faulting regardless of the
> > state when the kernel is started.
> >
> > Reviewed-by: Thiago Jung Bauermann <thiago.bauermann@xxxxxxxxxx>
> > Reviewed-by: Catalin Marinas <catalin.marinas@xxxxxxx>
> > Signed-off-by: Mark Brown <broonie@xxxxxxxxxx>
>
> I just bisected a build failure from a failed linker script assertion
> that I see with allmodconfig to this change in -next as commit
> ff5181d8a2a8 ("arm64/gcs: Provide basic EL2 setup to allow GCS usage at
> EL0 and EL1"):
>
> $ make -skj"$(nproc)" ARCH=arm64 CROSS_COMPILE=aarch64-linux- mrproper allmodconfig vmlinux
> aarch64-linux-ld: HYP init code too big
> make[4]: *** [scripts/Makefile.vmlinux:34: vmlinux] Error 1
> ...
>
> I see this with both GCC 14 and clang 19, in case toolchain version
> matters. Bisect log included as well.
Grmbl... 16 bytes too big. The hack below buys us about ~600 bytes by
removing some duplication, but we're losing half of the space to the
vectors.
Anyway, this is very lightly tested and it may eat your box.
Thanks,
M.
