perf tools build broken for RISCV 32 bit

[Arnd Bergmann]

On Mon, Jan 18, 2021 at 3:25 PM Emiliano Ingrassia
<ingrassia at epigenesys.com> wrote:
> On Mon, Jan 18, 2021 at 12:56:11PM +0100, Arnd Bergmann wrote:
> > On Mon, Jan 18, 2021 at 12:35 PM Emiliano Ingrassia
> > <ingrassia at epigenesys.com> wrote:
> > > |#if __WORDSIZE == 32
> > > |/* Workarounds for generic code needing to handle 64-bit time_t.
> > > |...
> > > |/* Fix sysdeps/nptl/lowlevellock-futex.h.  */
> > > |#define __NR_futex              __NR_futex_time64
> > >
> > > a possible solution to fix the build of perf tool on RISCV 32 bit could
> > > be the following patch to file tools/arch/riscv/include/uapi/asm/unistd.h:
> > > |+
> > > |+#ifndef __NR_futex
> > > |+#define __NR_futex __NR_futex_time64
> > > |+#endif
> > >
> > > What do you think about it? Could it be a correct solution?
> >
> > No, that would be much worse: __NR_futex must only ever point
> > to the system call that takes the old timespec argument. If you define
> > something like this, it breaks applications that try to do the right thing.
>
> OK, but what should be the right solution for RISCV 32 bit in this case?
> Is it a problem in the libc implementation or in RISCV/perf implementation?

The problem is in perf, along with as any other application that wants to
use futex on a 32-bit architecture with 64-bit time_t.

I already said what the application needs to do in order to get the
right system call, in the absence of a wrapper that is part of every
C library.

       Arnd