[PATCH] mips: Fix argument passing for inlined syscalls on Linux [BZ #25523]

[Maciej W. Rozycki]

On Thu, 27 Feb 2020, Matt Turner wrote:

> > > diff --git a/sysdeps/unix/sysv/linux/mips/mips32/sysdep.h b/sysdeps/unix/sysv/linux/mips/mips32/sysdep.h
> > > index beefcf284b..c275d63f67 100644
> > > --- a/sysdeps/unix/sysv/linux/mips/mips32/sysdep.h
> > > +++ b/sysdeps/unix/sysv/linux/mips/mips32/sysdep.h
> > [...]
> > > @@ -202,11 +203,13 @@ union __mips_syscall_return
> > >       long _sys_result;                                               \
> > >                                                                       \
> > >       {                                                               \
> > > +     long _arg1 = (long) (arg1);                                     \
> > > +     long _arg2 = (long) (arg2);                                     \
> > >       register long __s0 asm ("$16") __attribute__ ((unused))         \
> > >         = (number);                                                   \
> > >       register long __v0 asm ("$2");                                  \
> > > -     register long __a0 asm ("$4") = (long) (arg1);                  \
> > > -     register long __a1 asm ("$5") = (long) (arg2);                  \
> > > +     register long __a0 asm ("$4") = _arg1;                          \
> > > +     register long __a1 asm ("$5") = _arg2;                          \
> > >       register long __a3 asm ("$7");                                  \
> > >       __asm__ volatile (                                              \
> > >       ".set\tnoreorder\n\t"                                           \
> >
> >  Can e.g. `(long) (arg1)' end up as a library call?  Do you have a case,
> > even potential, where such clobbering actually happens?
> 
> We found that GNU make 4.3 fails to work on MIPS without this patch to
> glibc. See https://bugs.gentoo.org/708758

 Ah, indeed it can then, when whatever is passed as `arg1' gets inlined. 
Thanks for the pointer and sorry about the confusion.

  Maciej