Re: [Buildroot] [PATCH v5] Config.in: ban textrels on musl toolchains

["J. Neuschäfer via buildroot" <buildroot@buildroot.org>]

On Tue, Aug 13, 2024 at 11:09:02AM +0200, yann.morin@orange.com wrote:
> J., All,
>
> +Bernd and +David for x264:
>
> On 2024-07-19 17:27 +0200, J. Neuschäfer via buildroot spake thusly:
> > This patch forbids the use of TEXTRELs on musl toolchains with dynamic
> > linking.
>
> Another build failure due to this change:
>
>     [...]/buildroot/output/per-package/x264/host/bin/../lib/gcc/i586-buildroot-linux-musl/12.4.0/../../../../i586-buildroot-linux-musl/bin/ld: common/x86/dct-a-8.o: warning: relocation in read-only section `.text'
>     [...]/buildroot/output/per-package/x264/host/bin/i586-buildroot-linux-musl-gcc-ranlib libx264.a
>     [...]/buildroot/output/per-package/x264/host/bin/../lib/gcc/i586-buildroot-linux-musl/12.4.0/../../../../i586-buildroot-linux-musl/bin/ld: read-only segment has dynamic relocations
>     collect2: error: ld returned 1 exit status
>     make[2]: *** [Makefile:251: libx264.so.164] Error 1
>     make[2]: *** Waiting for unfinished jobs....
>     make[1]: *** [package/pkg-generic.mk:289: [...]/buildroot/output/build/x264-baee400fa9ced6f5481a728138fed6e867b0ff7f/.stamp_built] Error 2
>     make: *** [Makefile:83: _all] Error 2
>
> There is still no failure in the autobuilders, as x264 is quite late
> in the alphabetical order, so it probably is just hidden by earlier
> failures...
>
> I have no idea what to look for to avoid those relocations in the first
> place, so could you have a look, please?
>
> Regards,
> Yann E. MORIN.

I took a look, the issue is that libx264's x86 assembly is position
dependent (patch coming in a separate email). For example:


SECTION .text

cextern pw_32
[...]

INIT_MMX mmx
cglobal add4x4_idct, 2,2
    pxor m7, m7
.skip_prologue:
    movq  m1, [r1+ 8]
    movq  m3, [r1+24]
    movq  m2, [r1+16]
    movq  m0, [r1+ 0]
    IDCT4_1D w,0,1,2,3,4,5
    TRANSPOSE4x4W 0,1,2,3,4
    paddw m0, [pw_32]           <--- here
    IDCT4_1D w,0,1,2,3,4,5
    STORE_DIFF  m0, m4, m7, [r0+0*FDEC_STRIDE]
    STORE_DIFF  m1, m4, m7, [r0+1*FDEC_STRIDE]
    STORE_DIFF  m2, m4, m7, [r0+2*FDEC_STRIDE]
    STORE_DIFF  m3, m4, m7, [r0+3*FDEC_STRIDE]
    RET


In my experience, hand-written assembly code is usually the culprit, but
of course not all hand-written assembly code has these issues.


-- jn
_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot