[LTP] [PATCH 3/5] syscalls: Don't use tst_syscall() unnecessarily

[Arnd Bergmann]

On Tue, May 19, 2020 at 2:23 PM Cyril Hrubis <chrubis@suse.cz> wrote:
> > These syscall are old enough and must have support in libc for everyone.
> > Don't use tst_syscall() for them unnecessarily.
> >
> > Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
> > ---
> >  lib/parse_opts.c | 3 +--
> >  1 file changed, 1 insertion(+), 2 deletions(-)
> >
> > diff --git a/lib/parse_opts.c b/lib/parse_opts.c
> > index a9d50589a3f9..b3ab69c0a539 100644
> > --- a/lib/parse_opts.c
> > +++ b/lib/parse_opts.c
> > @@ -45,7 +45,6 @@
> >  #include "test.h"
> >  #include "ltp_priv.h"
> >  #include "usctest.h"
> > -#include "tst_clocks.h"
> >
> >  #ifndef UNIT_TEST
> >  #define UNIT_TEST    0
> > @@ -472,7 +471,7 @@ static uint64_t get_current_time(void)
> >  {
> >       struct timespec ts;
> >
> > -     tst_clock_gettime(CLOCK_MONOTONIC, &ts);
> > +     clock_gettime(CLOCK_MONOTONIC, &ts);
>
> I guess that this will reintroduce LTP compilation failures on older
> glibc, which was the primary reason we used the tst_clock_gettime()
> instead of clock_gettime().

I see that clock_gettime was first added in glibc-2.1.3 back in 1999.
Can that actually run LTP any more? If it can and this is considered
important, I fear the tst_clock_gettime() call needs to be extended
to call the clock_gettime()/clock_gettime64()/gettimeofday() syscalls,
whichever is the first to work, and convert the formats from the
native kernel format to the glibc format.

         Arnd