[LTP] [PATCH V7 14/19] syscalls/select6: Add support for time64 tests

[Viresh Kumar]

On 27-07-20, 11:40, Cyril Hrubis wrote:
> > @@ -38,7 +39,7 @@ static int do_select(int nfds, fd_set *readfds, fd_set *writefds, fd_set *except
> >  	}
> >  	case 2: {
> >  		int ret;
> > -		struct timespec ts = {
> > +		struct __kernel_old_timespec ts = {
> >  			.tv_sec = timeout->tv_sec,
> >  			.tv_nsec = timeout->tv_usec * 1000,
> >  		};
> 
> I'm a bit lost here, should we actually pass the __kernel_old_timespec
> to all the tst_syscall() fuctions here?

select, pselect6, pselect6_time64, and newselect, all have different
requirements, some take timespec and others take timeval.

Though after looking again at kernel sources I feel pselect6 may need
__kernel_timespec instead of __kernel_old_timespec, which is different than what
we did with other syscalls.

Arnd, can you confirm this please ?

> I guess that the only function that would take the argument as struct
> timeval is the select() glibc function, or do I miss something?

select in kernel also takes old timeval.

> > @@ -47,7 +48,22 @@ static int do_select(int nfds, fd_set *readfds, fd_set *writefds, fd_set *except
> >  		timeout->tv_usec = ts.tv_nsec / 1000;
> >  		return ret;
> >  	}
> > -	case 3:
> > +	case 3: {
> > +		int ret = 0;
> > +#if (__NR_clock_settime64 != __LTP__NR_INVALID_SYSCALL)
>             ^
> 	    __NR_pselect6_time64 ?
> 
> > +		struct __kernel_timespec ts = {
> > +			.tv_sec = timeout->tv_sec,
> > +			.tv_nsec = timeout->tv_usec * 1000,
> > +		};
> > +		ret = tst_syscall(__NR_pselect6_time64, nfds, readfds, writefds, exceptfds, &ts, NULL);
> > +		timeout->tv_sec = ts.tv_sec;
> > +		timeout->tv_usec = ts.tv_nsec / 1000;
> > +#else
> > +		tst_brk(TCONF, "__NR_pselect6 time64 variant not supported");
> > +#endif
> > +		return ret;
> > +	}
> > +	case 4:
> >  #ifdef __NR__newselect
> >  		return tst_syscall(__NR__newselect, nfds, readfds, writefds, exceptfds, timeout);
> >  #else
> > @@ -72,11 +88,14 @@ static void select_info(void)
> >  		tst_res(TINFO, "Testing SYS_pselect6 syscall");
> >  	break;
> >  	case 3:
> > +		tst_res(TINFO, "Testing SYS_pselect6 time64 syscall");
> > +	break;
> > +	case 4:
> >  		tst_res(TINFO, "Testing SYS__newselect syscall");
> >  	break;
> >  	}
> >  }
> >  
> > -#define TEST_VARIANTS 4
> > +#define TEST_VARIANTS 5
> 
> Also lastly but not least we should clean up the rest of the select
> tests and add support for the different variants there as well.

Maybe not. IIUC only pselect6 got changed to adapt to different timespec
structures and the other ones aren't.

Arnd: Can you confirm this as well ?

> Looking at them these are just copy&paste of the same test with a
> different fds, we can easily merge them into a single test.
> 
> And the coverate in these tests is a bit lacking, we do not have a
> single tests that would send a data over a pipe to a fd select is
> watching and check that select was woken up by that. There is no such
> test in the pselect/ directory either.

Hmm, I will look at that separately then.

-- 
viresh