undefined symbol in modpost "zero"

[Conor Dooley]

On Fri, Dec 23, 2022 at 04:59:51PM +0000, Ben Dooks wrote:
> On 23/12/2022 16:38, Ben Dooks wrote:
> > I'm getting the following error from modpost:
> > 
> > ERROR: modpost: "zero" [arch/riscv/kvm/kvm.ko] undefined!
> > ERROR: modpost: "zero" [drivers/gpu/drm/drm.ko] undefined!
> > ERROR: modpost: "zero" [drivers/nvme/host/nvme.ko] undefined!
> > ERROR: modpost: "zero" [drivers/i2c/algos/i2c-algo-bit.ko] undefined!
> > 
> > This seems to be coming from arch/riscv/include/asm/jump_label.h
> > with the arch_static_branch_jump() assembling some code:
> > 
> > >     40  {
> > >     41          asm_volatile_goto(
> > >     42                  "       .option
> > > push                            \n\t"
> > >     43                  "       .option
> > > norelax                         \n\t"
> > >     44                  "       .option
> > > norvc                           \n\t"
> > >     45                  "1:     jal             zero,
> > > %l[label]         \n\t"
> > >     46                  "       .option
> > > pop                             \n\t"
> > >     47                  "       .pushsection    __jump_table,
> > > \"aw\"    \n\t"
> > >     48                  "       .align          " RISCV_LGPTR
> > > "         \n\t"
> > >     49                  "       .long           1b - ., %l[label] -
> > > .   \n\t"
> > >     50                  "       " RISCV_PTR "   %0 -
> > > .                  \n\t"
> > >     51                  "
> > > .popsection                             \n\t"
> > >     52                  :  :  "i"(&((char *)key)[branch]) :  : label);
> > >     53
> > >     54          return false;
> > >     55  label:
> > >     56          return true;
> > >     57  }
> > > 
> > 
> > Changing the jal zero to jal x0 doesn't fix it, it just comes up with
> > "x0" being undefined.
> > 
> > gcc version 12.2.0 (Debian 12.2.0-9)
> > GNU assembler version 2.39.50 (riscv64-linux-gnu) using BFD version (GNU
> > Binutils for Debian) 2.39.50.20221208
> > 
> > I suspect this is a gas bug where it means to assemble a J instruction
> > to the label without outputting the unused symbol "zero" into the symbol
> > table. Is the right thing to fix gas, or can the kernel be changed?
> > 
> > FYI:
> > 
> > tmp.s:
> >       1          .text
> >       2
> >       3  1:
> >       4          jal zero, 1b
> > 
> > 
> > produces:
> > 
> > 0000000000000000 <.L1^B1>:
> >     0:   0000006f                j       0 <.L1^B1>
> >                          0: R_RISCV_JAL  .L1^B1
> > 
> > and
> > 
> > $ riscv64-linux-gnu-nm -a tmp.o
> > 0000000000000000 b .bss
> > 0000000000000000 d .data
> > 0000000000000000 n .riscv.attributes
> > 0000000000000000 t .text
> >                   U zero
> > 
> > Current fix is to:
> > 
> > 
> > diff --git a/arch/riscv/include/asm/jump_label.h
> > b/arch/riscv/include/asm/jump_label.h
> > index 6d58bbb5da46..a4abbacd0b62 100644
> > --- a/arch/riscv/include/asm/jump_label.h
> > +++ b/arch/riscv/include/asm/jump_label.h
> > @@ -42,7 +42,7 @@ static __always_inline bool
> > arch_static_branch_jump(struct static_key * const ke
> >                  "       .option push                            \n\t"
> >                  "       .option norelax                         \n\t"
> >                  "       .option norvc                           \n\t"
> > -               "1:     jal             zero, %l[label]         \n\t"
> > +               "1:     j               %l[label]               \n\t"
> >                  "       .option pop                             \n\t"
> >                  "       .pushsection    __jump_table, \"aw\"    \n\t"
> >                  "       .align          " RISCV_LGPTR "         \n\t"
> > 
> > 
> > This fixes modpost, not sure if this should be applied or not.
> > 
> 
> I just checked riscv64-linux-gnu-nm vmlinux.o and it has
> 
>                  U zero
> 
> 
> So I guess the linker just ignores this undefined symbol as it is never
> actually used for a relocation.

For anyone not paying attention to the RISC-V IRC, it is an upstream
binutils issue that does not affect released versions (yet) although it
does affect what is in debian-testing.

The upstream report for the issue is here:
https://sourceware.org/bugzilla/show_bug.cgi?id=29940

and for debian here:
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1026979

Thanks,
Conor.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-riscv/attachments/20221227/b1825ee1/attachment.sig>