[LTP] [PATCH v2] getrlimit/getrlimit03: Skip test if __NR_getrlimit_ulong not implemented

[Li Wang]

On Wed, Jul 9, 2025 at 9:43 AM Tiezhu Yang <yangtiezhu@loongson.cn> wrote:

> On 2025/7/8 下午9:10, Li Wang wrote:
> >
> >
> > On Tue, Jul 8, 2025 at 7:28 PM Tiezhu Yang <yangtiezhu@loongson.cn
> > <mailto:yangtiezhu@loongson.cn>> wrote:
> >
> >     In the LTP code, __NR_getrlimit_ulong is defined as __NR_ugetrlimit
> >     or __NR_getrlimit.
> >
> >     In the Linux kernel, LoongArch uses the generic syscall table
> >     which is defined in include/uapi/asm-generic/unistd.h. We can
> >     see that __NR_ugetrlimit is not defined in the generic header,
> >     and also there is no __ARCH_WANT_SET_GET_RLIMIT definition in
> >     the arch specific header arch/loongarch/include/asm/unistd.h,
> >     so both __NR_ugetrlimit and __NR_getrlimit are not implemented
> >     on LoongArch.
> >
> >     That is to say, no need to compare the return value and errno
> >     about the syscall numbers __NR_prlimit64 and __NR_{u}getrlimit,
> >     just check this case and then skip the test.
> >
> >     While at it, modify the value of test.tcnt as 1 to skip only
> >     once if the errno of getrlimit_ulong() or getrlimit_long() is
> >     ENOSYS.
>
> ...
>
> >     @@ -181,6 +187,7 @@ static void run(unsigned int resource)
> >              if (errno_l == ENOSYS) {
> >                      tst_res(TCONF | TERRNO,
> >                              "__NR_getrlimit(%d) not implemented",
> >     __NR_getrlimit);
> >
> >     +               test.tcnt = 1;
> >
> >
> > This second test.tcnt resetting is redundant, if the SIGNED_GETRLIMIT
> > get defined, that means both __NR_getrlimit and __NR_ugetrlimit are
> > detected,
> > and the only negative situation is getrlimit_ulong(resource, &rlim_ul);
> > works well, but getrlimit_long(resource, &rlim_l); return ENOSYS, to
> > make the
> > test can be interaction the second should no test.tcnt = 1.
> >
> > See:
> >
> https://github.com/linux-test-project/ltp/commit/fac783b5d6631efa70d2ea3257142763d944ba32
> >
> > Anyway, I modified the patch and merged!
>
> Thank you. If so, it is better to remove "or getrlimit_long()" in the
> commit message to reflect the reality, like this:
>
> "While at it, modify the value of test.tcnt as 1 to skip only
> once if the errno of getrlimit_ulong() is ENOSYS."
>

Indeed, fixed it.


-- 
Regards,
Li Wang