Re: [Qemu-devel] [RFC v3 04/24] linux-user: riscv: Fix compile failure on riscv32 hosts

[Alistair Francis]

On Mon, Dec 10, 2018 at 9:04 AM Richard Henderson
<address@hidden> wrote:
>
> On 12/7/18 6:46 PM, Alistair Francis wrote:
> > When cross compilling for riscv32 hosts using GCC 8.2 this error is seen:
> >     error: '__NR__llseek' undeclared (first use in this function); did you 
> > mean '_llseek'?
> >
> > To avoid the error let's ensure that __NR__llseek is defined.
> >
> > Signed-off-by: Alistair Francis <address@hidden>
> > ---
> >  linux-user/riscv/target_syscall.h | 5 +++++
> >  1 file changed, 5 insertions(+)
> >
> > diff --git a/linux-user/riscv/target_syscall.h 
> > b/linux-user/riscv/target_syscall.h
> > index ee81d8bc88..af73f71839 100644
> > --- a/linux-user/riscv/target_syscall.h
> > +++ b/linux-user/riscv/target_syscall.h
> > @@ -47,6 +47,11 @@ struct target_pt_regs {
> >  #endif
> >  #define UNAME_MINIMUM_RELEASE "4.15.0"
> >
> > +/* This is sometimes needed to compile riscv32 Linux user mode */
> > +#if !defined(__NR__llseek) && !defined(__NR_lseek)
> > +#define __NR__llseek __NR3264_lseek
> > +#endif
>
> I'm not quite sure how this could be.  Upstream I see
>
> #if __BITS_PER_LONG == 64 && !defined(__SYSCALL_COMPAT)
> ...
> #define __NR_lseek __NR3264_lseek
> ...
> #else
> ...
> #define __NR_llseek __NR3264_lseek
>
> so this define should already exist.
>
> Do you have the wrong headers installed for the cross-build?

I'm using Yocto for the cross build, so I don't see how I can get the
wrong headers. I agree though that I probably shouldn't need this
patch. Maybe it's something to do with the glibc fork I'm using. I'll
drop this patch and see if I can figure out what's going on.

Alistair

>
>
> r~