This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Correct minimal test suite changes for elision enabled kernels.
- From: Torvald Riegel <triegel at redhat dot com>
- To: Stefan Liebler <stli at linux dot vnet dot ibm dot com>
- Cc: libc-alpha at sourceware dot org
- Date: Mon, 15 Dec 2014 22:20:12 +0100
- Subject: Re: [PATCH] Correct minimal test suite changes for elision enabled kernels.
- Authentication-results: sourceware.org; auth=none
- References: <533E9CF2 dot 8030306 at linux dot vnet dot ibm dot com> <lhub7k$7jj$1 at ger dot gmane dot org>
On Mon, 2014-04-07 at 16:03 +0200, Stefan Liebler wrote:
> Hi,
> according to the comment of Andi Kleen (see below),
> iÂve dropped the ifdef.
> Please test it on an intel machine with lock elision support!
I have committed this patch for you as
7f786dc12bd60f0a134e538429fef98350e4c814. The part of the test that was
previously disabled if elision was enabled was black box testing
according of semantics required by POSIX. Using lock elision must not
change the outcome.
This also fixes a -Werror with a recent GCC build, in that in a build
that enables elision, the timespec values ts, tv, and tv2 had been
declared but not used (the only use was in the disabled part of the
tests).
Thanks.
> Bye
>
> ---
> 2014-04-07 Stefan Liebler <stli@linux.vnet.ibm.com>
>
> * nptl/tst-mutex5 (do_test):
> Remove #ifdef ENABLE_LOCK_ELISION.
> ---
>
> On Fri, Apr 04, 2014 at 01:52:18PM +0200, Stefan Liebler wrote:
> > Hi,
> >
> > according to the comment in
> > "Re: [PATCH 03/14] Add minimal test suite changes for elision
> > enabled kernels"
> > (https://www.sourceware.org/ml/libc-alpha/2013-06/msg01145.html),
> > the tests tst-mutex5/8 are changed in order to not check some
> > behaviour of POSIX if lock elision is enabled at configure time.
> > Currently, the timeout of pthread_mutex_timedlock is only tested
> > with elision enabled at configure time and not for the default case.
>
> If it works with elision we can actually drop the ifdef,
> as it should work in both cases.
>
> May have been a left over from before trylock aborted.
>
> -Andi
>
> >
> > Bye
> >
> > ---
> > 2014-04-04 Stefan Liebler <stli@linux.vnet.ibm.com>
> >
> > * nptl/tst-mutex5 (do_test):
> > Use #ifndef ENABLE_LOCK_ELISION instead of #ifdef.
> > ---
>
> > diff --git a/nptl/tst-mutex5.c b/nptl/tst-mutex5.c
> > index 14d3025..a829272 100644
> > --- a/nptl/tst-mutex5.c
> > +++ b/nptl/tst-mutex5.c
> > @@ -87,7 +87,7 @@ do_test (void)
> > }
> >
> > /* Elided locks do not time out. */
> > -#ifdef ENABLE_LOCK_ELISION
> > +#ifndef ENABLE_LOCK_ELISION
> > if (pthread_mutex_trylock (&m) == 0)
> > {
> > puts ("mutex_trylock succeeded");
>
>
> -- ak@linux.intel.com -- Speaking for myself only