Re: lib/crypto/mips/poly1305-core.S:95: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $8,0+3($5)'

Re: lib/crypto/mips/poly1305-core.S:95: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $8,0+3($5)'

[Eric Biggers]

[+Cc linux-mips and linux-crypto]

On Sat, Mar 21, 2026 at 01:47:15AM +0800, kernel test robot wrote:
> Hi Eric,
> 
> First bad commit (maybe != root cause):
> 
> tree:   https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
> head:   0e4f8f1a3d081e834be5fd0a62bdb2554fadd307
> commit: 7e54e993ab8c98d912f54ad6f46bfcc9dcd65368 lib/crypto: mips: Move arch/mips/lib/crypto/ into lib/crypto/
> date:   9 months ago
> config: mips-randconfig-r131-20260320 (https://download.01.org/0day-ci/archive/20260321/202603210105.sdD4rsTq-lkp@intel.com/config)
> compiler: mips-linux-gcc (GCC) 11.5.0
> sparse: v0.6.5-rc1
> reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20260321/202603210105.sdD4rsTq-lkp@intel.com/reproduce)
> 
> If you fix the issue in a separate patch/commit (i.e. not just a new version of
> the same patch/commit), kindly add following tags
> | Reported-by: kernel test robot <lkp@intel.com>
> | Closes: https://lore.kernel.org/oe-kbuild-all/202603210105.sdD4rsTq-lkp@intel.com/
> 
> All errors (new ones prefixed by >>):
> 
>    lib/crypto/mips/poly1305-core.S: Assembler messages:
> >> lib/crypto/mips/poly1305-core.S:95: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $8,0+3($5)'
>    lib/crypto/mips/poly1305-core.S:96: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $9,4+3($5)'
>    lib/crypto/mips/poly1305-core.S:97: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $10,8+3($5)'
>    lib/crypto/mips/poly1305-core.S:98: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $11,12+3($5)'

This isn't new.  It was first reported in 2021:
https://lore.kernel.org/all/202108040636.P6t1LvPP-lkp@intel.com/

It's caused by:

CONFIG_64BIT=n causes the 32-bit Poly1305 assembly code to be generated.
That code checks the _MIPS_ARCH_MIPS32R6 macro.  But, the code is
actually built with -march=mips64r6, due to CONFIG_CPU_MIPS64_R6=y.
Thus _MIPS_ARCH_MIPS32R6 is not defined; _MIPS_ARCH_MIPS64R6 is defined
instead.  So the wrong code gets built.

Maybe someone from linux-mips@vger.kernel.org can confirm whether
CONFIG_64BIT=n && CONFIG_CPU_MIPS64_R6=y actually makes sense?

- Eric

Re: lib/crypto/mips/poly1305-core.S:95: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $8,0+3($5)'

[Maciej W. Rozycki]

On Mon, 30 Mar 2026, Eric Biggers wrote:

> > All errors (new ones prefixed by >>):
> > 
> >    lib/crypto/mips/poly1305-core.S: Assembler messages:
> > >> lib/crypto/mips/poly1305-core.S:95: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $8,0+3($5)'
> >    lib/crypto/mips/poly1305-core.S:96: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $9,4+3($5)'
> >    lib/crypto/mips/poly1305-core.S:97: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $10,8+3($5)'
> >    lib/crypto/mips/poly1305-core.S:98: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $11,12+3($5)'
> 
> This isn't new.  It was first reported in 2021:
> https://lore.kernel.org/all/202108040636.P6t1LvPP-lkp@intel.com/
> 
> It's caused by:
> 
> CONFIG_64BIT=n causes the 32-bit Poly1305 assembly code to be generated.
> That code checks the _MIPS_ARCH_MIPS32R6 macro.  But, the code is
> actually built with -march=mips64r6, due to CONFIG_CPU_MIPS64_R6=y.
> Thus _MIPS_ARCH_MIPS32R6 is not defined; _MIPS_ARCH_MIPS64R6 is defined
> instead.  So the wrong code gets built.
> 
> Maybe someone from linux-mips@vger.kernel.org can confirm whether
> CONFIG_64BIT=n && CONFIG_CPU_MIPS64_R6=y actually makes sense?

 Absolutely, you can build a 32-bit kernel for and run on 64-bit hardware 
and we have supported it since forever across all the MIPS architecture 
revisions.

 You probably want to check for CONFIG_CPU_MIPSR6 instead (and similarly 
for CONFIG_CPU_MIPSR2, etc.).

  Maciej

Re: lib/crypto/mips/poly1305-core.S:95: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $8,0+3($5)'

[Eric Biggers]

On Tue, Mar 31, 2026 at 02:01:44AM +0100, Maciej W. Rozycki wrote:
> On Mon, 30 Mar 2026, Eric Biggers wrote:
> 
> > > All errors (new ones prefixed by >>):
> > > 
> > >    lib/crypto/mips/poly1305-core.S: Assembler messages:
> > > >> lib/crypto/mips/poly1305-core.S:95: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $8,0+3($5)'
> > >    lib/crypto/mips/poly1305-core.S:96: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $9,4+3($5)'
> > >    lib/crypto/mips/poly1305-core.S:97: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $10,8+3($5)'
> > >    lib/crypto/mips/poly1305-core.S:98: Error: opcode not supported on this processor: mips64r6 (mips64r6) `lwl $11,12+3($5)'
> > 
> > This isn't new.  It was first reported in 2021:
> > https://lore.kernel.org/all/202108040636.P6t1LvPP-lkp@intel.com/
> > 
> > It's caused by:
> > 
> > CONFIG_64BIT=n causes the 32-bit Poly1305 assembly code to be generated.
> > That code checks the _MIPS_ARCH_MIPS32R6 macro.  But, the code is
> > actually built with -march=mips64r6, due to CONFIG_CPU_MIPS64_R6=y.
> > Thus _MIPS_ARCH_MIPS32R6 is not defined; _MIPS_ARCH_MIPS64R6 is defined
> > instead.  So the wrong code gets built.
> > 
> > Maybe someone from linux-mips@vger.kernel.org can confirm whether
> > CONFIG_64BIT=n && CONFIG_CPU_MIPS64_R6=y actually makes sense?
> 
>  Absolutely, you can build a 32-bit kernel for and run on 64-bit hardware 
> and we have supported it since forever across all the MIPS architecture 
> revisions.
> 
>  You probably want to check for CONFIG_CPU_MIPSR6 instead (and similarly 
> for CONFIG_CPU_MIPSR2, etc.).

Okay, but does it make sense to use -march=mips64r6 in that case,
instead of -march=mips32r6?  I guess you get something that uses the
32-bit ABI but is free to use 64-bit instructions?  It looks like
poly1305-mips.pl doesn't really support that combination: it just has
32-bit ABI + 32-bit instructions, and 64-bit ABI + 64-bit instructions.

Maybe it would be easiest to always compile poly1305-mips.pl with the
corresponding mips32 flag when CONFIG_64BIT is unset?  Note: since
poly1305-mips.pl was taken from an external source
(https://github.com/dot-asm/cryptogams/blob/master/mips/poly1305-mips.pl),
modifying the build system to pass the appropriate flag might be
preferable to extensive local patching of that file.

- Eric