On Mon, 02 Nov 2020, Daniel Vetter wrote:
> On Mon, Nov 2, 2020 at 12:09 PM Lee Jones <lee.jones@xxxxxxxxxx> wrote:
> >
> > On Mon, 02 Nov 2020, Daniel Vetter wrote:
> >
> > > On Fri, Oct 30, 2020 at 7:18 PM Lee Jones <lee.jones@xxxxxxxxxx> wrote:
> > > >
> > > > Commit 09e5b3fd5672 ("Fonts: Support FONT_EXTRA_WORDS macros for
> > > > built-in fonts") introduced the following error when building
> > > > rpc_defconfig (only this build appears to be affected):
> > > >
> > > > `acorndata_8x8' referenced in section `.text' of arch/arm/boot/compressed/ll_char_wr.o:
> > > > defined in discarded section `.data' of arch/arm/boot/compressed/font.o
> > > > `acorndata_8x8' referenced in section `.data.rel.ro' of arch/arm/boot/compressed/font.o:
> > > > defined in discarded section `.data' of arch/arm/boot/compressed/font.o
> > > > make[3]: *** [/scratch/linux/arch/arm/boot/compressed/Makefile:191: arch/arm/boot/compressed/vmlinux] Error 1
> > > > make[2]: *** [/scratch/linux/arch/arm/boot/Makefile:61: arch/arm/boot/compressed/vmlinux] Error 2
> > > > make[1]: *** [/scratch/linux/arch/arm/Makefile:317: zImage] Error 2
> > > >
> > > > The .data section is discarded at link time. Reinstating
> > > > acorndata_8x8 as const ensures it is still available after linking.
> > > >
> > > > Cc: <stable@xxxxxxxxxxxxxxx>
> > > > Cc: Russell King <linux@xxxxxxxxxxxxxxx>
> > > > Signed-off-by: Lee Jones <lee.jones@xxxxxxxxxx>
> > >
> > > Shouldn't we add the const to all of them, for consistency?
> >
> > The thought did cross my mind. However, I do not see any further
> > issues which need addressing. Nor do I have any visibility into what
> > issues may be caused by doing so. The only thing I know for sure is
> > that this patch fixes the compile error pertained to in the commit
> > message, and I'd like for this fix to be as atomic as possible, as
> > it's designed to be routed through the Stable/LTS trees.
>
> The trouble is that if we only make one of them const, then it'll take
> so much longer to hit any issues due to code not handling this
> correctly. Being consistent with all fonts sounds like the best
> approach.
>
> And the original patch that lost the const for the additional data
> also went through cc: stable for all fonts together. So that shouldn't
> be the hold-up.
My plan was to keep the fix as simple as possible.
This is only an issue due to the odd handling of the compressed Arm
image which exclusively references 'acorndata_8x8' and discards it's
.data section.
I am happy to go with the majority on this though.
Does anyone else have an opinion?
--
Lee Jones [李琼斯]
Senior Technical Lead - Developer Services
Linaro.org │ Open source software for Arm SoCs
Follow Linaro: Facebook | Twitter | Blog
