This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [RFC][PATCH] tunables: Add elision tunable
- From: Aurelien Jarno <aurelien at aurel32 dot net>
- To: Tulio Magno Quites Machado Filho <tuliom at linux dot vnet dot ibm dot com>
- Cc: Stefan Liebler <stli at linux dot vnet dot ibm dot com>, libc-alpha at sourceware dot org, Rogerio Alves Cardoso <rcardoso at linux dot vnet dot ibm dot com>
- Date: Tue, 9 Jan 2018 16:54:19 +0100
- Subject: Re: [RFC][PATCH] tunables: Add elision tunable
- Authentication-results: sourceware.org; auth=none
- References: <68de6583-4f5c-5fa4-9328-1682e614ef56@linux.vnet.ibm.com> <581813ef-53f8-ba33-4026-21e8970d0155@linux.vnet.ibm.com> <d0be5934-6926-35b7-ff30-8e975464e571@redhat.com> <82fe9665-1329-98c1-7a15-c1b953bda91d@linux.vnet.ibm.com> <605c6017-4e5d-553a-d1ff-36b6480e7313@linux.vnet.ibm.com> <87mv2x2brz.fsf@linux.vnet.ibm.com>
On 2017-12-05 17:52, Tulio Magno Quites Machado Filho wrote:
> Stefan Liebler <stli@linux.vnet.ibm.com> writes:
>
> > On 11/16/2017 07:44 PM, Rogerio Alves wrote:
> >> All documentation suggestions was made on patch v5. Thanks for the review.
> >>
> >> Regards
> >>
> >> Rogerio.
> >>
> >
> > I've applied your patch in order to test it on s390.
> > Unfortunately building glibc is now failing.
> >
> > Here are some notes:
> >
> > git apply warned:
> > 0001-Add-elision-tunables.patch:291: trailing whitespace.
> > # Disable elision for tst-mutex8 so it can verify error case for
> > warning: 1 line adds whitespace errors.
> >
> > Changing configure.ac is not sufficient.
> > Please regenerate configure in order to reflect the changes in configure.ac.
> >
> > There are also usages of enable_lock_elision and ENABLE_LOCK_ELISION in
> > config.make.in, sysdeps/unix/sysv/linux/s390/Makefile,
> > sysdeps/s390/nptl/bits/pthreadtypes-arch.h, sysdeps/s390/configure.ac.
> >
> > Please have a look into attached diff which applies on top of your patch
> > v5. With those changes I was able to build glibc on s390.
> >
> > Afterwards I've run a small test program which determines if
> > pthread_mutex_lock is using a transaction or not:
> >
> > ./prog
> > Lock was a normal lock!
> >
> > GLIBC_TUNABLES=glibc.elision.enable=1 ./prog
> > Lock was elided via a transaction! (nesting-depth=1)
> >
> > GLIBC_TUNABLES=glibc.elision.enable=0 ./prog
> > Lock was a normal lock!
> >
> > chmod +s prog_secure
> > GLIBC_TUNABLES=glibc.elision.enable=1 ./prog_secure
> > Lock was a normal lock!
>
> I applied this patch, updated the ChangeLog, run tests on ppc{|64|64le},
> s390x and x86_64, and pushed the patch as 07ed18d26a342.
This commit cause the following tests to fail with the default
configuration on ppc{|64|64le}:
FAIL: elf/tst-env-setuid
FAIL: elf/tst-env-setuid-tunables
FAIL: stdlib/tst-secure-getenv
The problem is not new and it was possible to observe it with glibc-2.25
with --enable-tunables --enable-lock-elision=yes. Now --enable-tunables
is the default and following this commit lock elision code is always
there.
Please see https://sourceware.org/bugzilla/show_bug.cgi?id=22685 for
more details.
--
Aurelien Jarno GPG: 4096R/1DDD8C9B
aurelien@aurel32.net http://www.aurel32.net