Re: [LTP] [PATCH] clone10: add support archs

[Petr Vorel <pvorel@suse.cz>]

> Petr Vorel <pvorel@suse.cz> wrote:

> > > Acked-by: Jan Stancek <jstancek@redhat.com>

> > LGTM. Indeed only these 3 archs don't TCONF.
> > Acked-by: Petr Vorel <pvorel@suse.cz>

> > Out of curiosity, where is the support defined in kernel?
> > "if (clone_flags & CLONE_SETTLS)" is in many archs:

> Not based on this, the clone10.c test was written and gets tested
> only on the known archs (x86_64, s390x, aarch64) by now.

> And in case that other archs (not tested) have different behavior like i386,
> so we are limited to the know/tested archs.

> If we can get another arch to verified we can add it to the
> .supported_archs as well.

> $ cat -n ltp/include/lapi/tls.h
>     ...
>     53 static inline void init_tls(void)
>     54 {
>     55     #if defined(__x86_64__) || defined(__aarch64__) || defined(__s390x__)
>     56     tls_ptr = allocate_tls_area();
>     57 #else
>     ...

include/lapi/tls.h

static inline void init_tls(void)
{
#if defined(__x86_64__) || defined(__aarch64__) || defined(__s390x__)
	tls_ptr = allocate_tls_area();
#else
	tst_brk(TCONF, "Unsupported architecture for TLS");
#endif

I see. First IMHO the message in include/lapi/tls.h is pretty misleading. It
does not look to me as a test limitation, but as a missing arch support in
kernel.  IMHO the message should have been something like:

tst_brk(TCONF, "Test not supported only architecture");

(i.e. to mention "test")

Also having arch listed in the test and also in tls.h is redundant.
IMHO it should be only on a single place.

Kind regards,
Petr

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp