On Mon, Oct 26, 2020 at 1:22 PM Arnaldo Carvalho de Melo
<acme@xxxxxxxxxx> wrote:
>
> Em Mon, Oct 26, 2020 at 05:10:06PM -0300, Arnaldo Carvalho de Melo escreveu:
> > Now with a correct subject :-)
> >
> > Em Mon, Oct 26, 2020 at 04:58:30PM -0300, Arnaldo Carvalho de Melo escreveu:
> > > Hi guys,
> > >
> > > I just stumbled on this, investigating... This is with what is
> > > in the tmp branch at
> > > git://git.kernel.org/pub/scm/devel/pahole/pahole.git.
> > >
> > > error: found variable in CU '/home/acme/git/linux/security/selinux/hooks.c' that has void type
> > > Encountered error while encoding BTF.
> > > LD .tmp_vmlinux.kallsyms1
> > > KSYMS .tmp_vmlinux.kallsyms1.S
> > > AS .tmp_vmlinux.kallsyms1.S
> > > LD .tmp_vmlinux.kallsyms2
> > > KSYMS .tmp_vmlinux.kallsyms2.S
> > > AS .tmp_vmlinux.kallsyms2.S
> > > LD vmlinux
> > > BTFIDS vmlinux
> > > FAILED: load BTF from vmlinux: Unknown error -2make[1]: *** [/home/acme/git/linux/Makefile:1164: vmlinux] Error 255
> > > make[1]: Leaving directory '/home/acme/git/build/v5.10.0-rc1+'
> > > make: *** [Makefile:185: __sub-make] Error 2
> >
> > It is something recent:
> >
> > c815d26689313d8d7 (Hao Luo 2020-08-24 17:45:23 -0700 440) if (var->ip.tag.type == 0) {
> > 2e719cca667212840 (Andrii Nakryiko 2020-10-08 16:39:57 -0700 441) fprintf(stderr, "error: found variable in CU '%s' that has void type\n",
> > 2e719cca667212840 (Andrii Nakryiko 2020-10-08 16:39:57 -0700 442) cu->name);
> > f3d9054ba8ff1df0f (Hao Luo 2020-07-08 13:44:10 -0700 443) if (force)
> > f3d9054ba8ff1df0f (Hao Luo 2020-07-08 13:44:10 -0700 444) continue;
> > f3d9054ba8ff1df0f (Hao Luo 2020-07-08 13:44:10 -0700 445) err = -1;
> > f3d9054ba8ff1df0f (Hao Luo 2020-07-08 13:44:10 -0700 446) break;
>
> Printing the variable name and address gives:
>
> LD .tmp_vmlinux.btf
> BTF .btf.vmlinux.bin.o
> error: found variable "(null)" (addr: 0xffffffff82f4e260) in CU '/home/acme/git/linux/security/selinux/hooks.c' that has void type
> Encountered error while encoding BTF.
>
> Unsure if this is related:
>
> [root@five ~]# gcc --version | head -1
> gcc (GCC) 10.2.1 20200723 (Red Hat 10.2.1-1)
> [root@five ~]#
>
> [acme@five linux]$ readelf -sw ../build/v5.9.0+/vmlinux | grep ffffffff82f4e260
> readelf: ../build/v5.9.0+/vmlinux: Error: LEB value too large
> readelf: ../build/v5.9.0+/vmlinux: Error: LEB value too large
> readelf: ../build/v5.9.0+/vmlinux: Error: LEB value too large
> readelf: ../build/v5.9.0+/vmlinux: Error: LEB value too large
> readelf: ../build/v5.9.0+/vmlinux: Error: LEB value too large
> readelf: ../build/v5.9.0+/vmlinux: Error: LEB value too large
> readelf: ../build/v5.9.0+/vmlinux: Error: LEB value too large
> readelf: ../build/v5.9.0+/vmlinux: Error: LEB value too large
> readelf: ../build/v5.9.0+/vmlinux: Error: LEB value too large
> readelf: ../build/v5.9.0+/vmlinux: Error: LEB value too large
> readelf: ../build/v5.9.0+/vmlinux: Error: LEB value too large
> readelf: ../build/v5.9.0+/vmlinux: Error: LEB value too large
>
> <SNIP lots more such lines>
>
Don't know what this is, but what Hao proposed (to check everything
only after we make sure that the variable has a proper per-CPU symbol
in ELF) would eliminate all that, so I think we should do just that.
But I'm surprised I didn't see any of those issues in my kernel
configurations and environment (though I use GCC 8, still, so that
might explain).
> - Arnaldo
