Re: arch/mips/boot/compressed/decompress.c:undefined reference to `__ashldi3'

[David Laight <david.laight.linux@gmail.com>]

On Thu, 1 Oct 2026 11:30:09 +0100
Pedro Falcato <pfalcato@suse.de> wrote:

> On Wed, Sep 30, 2026 at 12:01:59PM -0700, Andrew Morton wrote:
> > On Wed, 30 Sep 2026 18:36:50 +0200 kernel test robot <lkp@intel.com> wrote:
> >   
> > > tree:   https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
> > > head:   551c722f40809618230001baccf219193e22fc5a
> > > commit: d4dba3b9c03a326cfa73833d6b166aeb442f82b5 lib: decompress_bunzip2: fix 32-bit shift undefined behavior
> > > date:   6 months ago
> > > config: mips-randconfig-r2300-20260930 (https://download.01.org/0day-ci/archive/20260930/202609301826.MFaFMsmd-lkp@intel.com/config)
> > > compiler: mips-linux-gcc (GCC) 15.2.0
> > > sparse: v0.6.5-rc1
> > > reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20260930/202609301826.MFaFMsmd-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
> > > | Fixes: d4dba3b9c03a ("lib: decompress_bunzip2: fix 32-bit shift undefined behavior")
> > > | Reported-by: kernel test robot <lkp@intel.com>
> > > | Closes: https://lore.kernel.org/oe-kbuild-all/202609301826.MFaFMsmd-lkp@intel.com/
> > > 
> > > All errors (new ones prefixed by >>):
> > > 
> > >    mips-linux-ld: arch/mips/boot/compressed/decompress.o: in function `get_bits':  
> > > >> arch/mips/boot/compressed/decompress.c:(.text+0xf4): undefined reference to `__ashldi3'
> > > >> mips-linux-ld: arch/mips/boot/compressed/decompress.c:(.text+0x168): undefined reference to `__ashldi3'  
> > 
> > I dunno, I'd be suspecting a toolchain issue here?
> > 
> > --- a/lib/decompress_bunzip2.c
> > +++ b/lib/decompress_bunzip2.c
> > @@ -135,7 +135,7 @@ static unsigned int INIT get_bits(struct bunzip_data *bd, char bits_wanted)
> >  		}
> >  		/* Avoid 32-bit overflow (dump bit buffer to top of output) */
> >  		if (bd->inbufBitCount >= 24) {
> > -			bits = bd->inbufBits&((1 << bd->inbufBitCount)-1);
> > +			bits = bd->inbufBits & ((1ULL << bd->inbufBitCount) - 1);
> >  			bits_wanted -= bd->inbufBitCount;
> >  			bits <<= bits_wanted;
> >  			bd->inbufBitCount = 0;
> > @@ -146,7 +146,7 @@ static unsigned int INIT get_bits(struct bunzip_data *bd, char bits_wanted)
> >  	}
> >  	/* Calculate result */
> >  	bd->inbufBitCount -= bits_wanted;
> > -	bits |= (bd->inbufBits >> bd->inbufBitCount)&((1 << bits_wanted)-1);
> > +	bits |= (bd->inbufBits >> bd->inbufBitCount) & ((1ULL << bits_wanted) - 1);
> >  
> >  	return bits;
> >  }
> > 
> > 32-bit MIPS should be able to do this?  
> 
> FWIW the patch seems entirely bogus. I don't think you can have inbufBitCount >=
> 32. If you could, then the patch would still not fix anything, because
> bd->inbufBits is an (32-bit) unsigned int.

The 1 should probably be 1U (or possibly 1UL is 64bit values can happen on
64bit) rather than 1ULL.

Wasn't the problem being solved '1 << 31' being undefined (or UB) and one of
the static checkers deciding that should never shift a signed value left?
That is entirely bogus anyway.
It is only non-portable because C allows 1's compliment and sign overpunch
for negative integers, gcc supports neither (and I think the next C standard
mandates 2's compliment).

That is separate from shift right of negative values being UB to allow for
cpu that don't have a sign replicating shift (all modern ones do).
Think about it, for a shift/rotate left you need one of 0/carry/high-bit and
for shift right 0/carry/low-bit so two instruction bits can select one of
0/carry/low-bit/high-bit (giving the often implemented but undocumented shift
left that replicates the low bit).

David

> 
> As per MIPS itself, MIPS does not link ashldi3.o into the decompressor for BZ2.
> It only does so for XZ and ZSTD.
> 
> Changing arch/mips/boot/compressed/Makefile would work, but since the change is
> entirely wrong I would suggest perhaps reverting?
> 
>