On Sun, 28 Jul 2019, Andy Lutomirski wrote:
> On Sun, Jul 28, 2019 at 6:20 AM Thomas Gleixner <[email protected]> wrote:
> >
> > The 32bit variants of vdso_clock_gettime()/getres() have a NULL pointer
> > check for the timespec pointer. That's inconsistent vs. 64bit.
> >
> > But the vdso implementation will never be consistent versus the syscall
> > because the only case which it can handle is NULL. Any other invalid
> > pointer will cause a segfault. So special casing NULL is not really useful.
> >
> > Remove it along with the superflouos syscall fallback invocation as that
> > will return -EFAULT anyway. That also gets rid of the dubious typecast
> > which only works because the pointer is NULL.
>
> Reviewed-by: Andy Lutomirski <[email protected]>
>
> FWIW, the equivalent change to gettimeofday would be an ABI break,
> since we historically have that check, and it even makes sense there.
Of course, because either of the two pointers can be NULL.
Thanks,
tglx
