[lttng-dev] [PATCH lttng-tools 2/2] Tests: select_poll_epoll: Add support for _time64

[Jérémie Galarneau]

On Thu, Oct 13, 2022 at 10:12:48AM +1000, Alistair Francis wrote:
> On Wed, Oct 12, 2022 at 11:19 PM Jérémie Galarneau
> <jeremie.galarneau at efficios.com> wrote:
> >
> > On Fri, Oct 07, 2022 at 10:39:18AM +1000, Alistair Francis via lttng-dev wrote:
> >
> > Hi Alistair,
> >
> > The first patch is good, I'll merge it in master.
> > Some comments on this patch follow.
> 
> Thanks!
> 
> >
> > > From: Alistair Francis <alistair.francis at wdc.com>
> > >
> > > Add support for the  64-bit time_t syscalls SYS_ppoll_time64
> > > and SYS_pselect6_time64.
> > >
> > > These are the syscalls that exist 32-bit platforms since the 5.1 kernel.
> > > 32-bit platforms with a 64-bit time_t only have these and don't have the
> > > original syscalls (such as 32-bit RISC-V).
> > >
> > > Fixes: https://github.com/lttng/lttng-tools/pull/162
> > > Signed-off-by: Alistair Francis <alistair.francis at wdc.com>
> > > ---
> > >  tests/regression/kernel/select_poll_epoll.cpp | 53 +++++++++++++++++++
> > >  1 file changed, 53 insertions(+)
> > >
> > > diff --git a/tests/regression/kernel/select_poll_epoll.cpp b/tests/regression/kernel/select_poll_epoll.cpp
> > > index c0b688217..4a6d394f4 100644
> > > --- a/tests/regression/kernel/select_poll_epoll.cpp
> > > +++ b/tests/regression/kernel/select_poll_epoll.cpp
> > > @@ -5,6 +5,7 @@
> > >   *
> > >   */
> > >
> > > +#include <errno.h>
> > >  #include <fcntl.h>
> > >  #include <limits.h>
> > >  #include <poll.h>
> > > @@ -456,8 +457,22 @@ void ppoll_fds_buffer_overflow(
> > >       ufds[0].fd = wait_fd;
> > >       ufds[0].events = POLLIN|POLLPRI;
> > >
> > > +#ifdef SYS_ppoll_time64
> > > +     /*
> > > +      * As there is no timeout value, we don't convert to/from
> > > +      * 64/32-bit time_t.
> > > +      */
> > > +     ret = syscall(SYS_ppoll_time64, ufds, 100, NULL, NULL);
> > > +     if (ret == 0 || errno != ENOSYS) {
> > > +             goto ppoll_fds_buffer_overflow_done;
> > > +     }
> > > +#endif
> > > +
> >
> > This results in the following warning when building for an architecture
> > that doesn't have SYS_ppoll_time64 defined:
> >
> > label ‘ppoll_fds_buffer_overflow_done’ defined but not used [-Wunused-label]
> > ppoll_fds_buffer_overflow_done:
> > ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > select_poll_epoll.cpp: In function ‘void ppoll_fds_ulong_max(FILE*)’:
> 
> Argh, I'll fix this
> 
> >
> >
> > Also, it is my understanding that both syscalls can be available on some
> > platforms. In that case, it would make sense to add them as separate
> > tests and skip tests that target non-existant syscalls.
> 
> So all 32-bit platforms since the 5.1 (or 5.4?) kernel have both syscalls.
> 
> From my understanding the original syscalls will be removed on 32-bit
> platforms at some point (before 2038) and there will only be *_time64
> variants.
> 
> If you want I can copy the tests to test both syscall types, but I
> don't think that's necessary.
>

My fear is that on those platforms the test will pass if the kernel tracer
succeeds in tracing any of the two syscalls.

I think it will be easier to simply have separate tests than validate the
two invocations independently in the trace.

Thanks!
Jérémie

> Alistair