On Thu, 2008-07-17 at 12:55 +1000, David Gibson wrote:
> On Tue, Jul 15, 2008 at 03:14:54PM +0000, Adam Litke wrote:
> > Depending on the version of GNU binutils in use, the linkhuge or linkhuge_rw
> > tests may be known to fail and in some cases hang. Allow the test suite to
> > complete and produce clean output again by skipping tests that are known to
> > be
> > bad and suppressing redundant warning messages.
> >
> > This patch handles the following cases:
> >
> > 1) If an xBDT linker script is used on a system that requires the use of the
> > SPECIAL() linker script command for placement of the .plt and .got sections,
> > the resulting binary will behave erratically (function calls will call the
> > wrong functions) due to misplacement of the plt and got. The solution is to
> > not run linkhuge tests if the use of SPECIAL is discovered in the system
> > linker
> > scripts.
> >
> > 2) GNU binutils < 2.17 does not support -zcommon-page-size and
> > -zmax-page-size.
> > These older versions will silently ignore the unrecognized options and
> > produce
> > binaries that are not suitably aligned for linkhuge_rw to be remapped into
> > huge
> q> pages. The solution is to use linkhuge_rw in a test mode to check its
> > alignment. If that test fails, suppress the redundant warning messages for
> > the
> > rest of linkhuge_rw tests.
>
> [snip]
> > + # When the support is missing, libhugetlbfs is very verbose to aid in
> > + # problem diagnosis. To keep the test suite output clean in this case,
> > + # only allow the verbose messages to be printed once.
> > + bits=($WORDSIZES)
> > + # Just run the test once for the first word size
> > + run_test_bits ${bits[0]} linkhuge_rw --test-alignment
> > + if [ $? -ne 0 ]; then
> > + LINKHUGERW_QUIET=1
>
> This LINKHUGERW_QUIET stuff seems a bit overly complicated. Since the
> testcases will still get a FAIL, can't we just always suppress
> messages from the library, when QUIET_TEST is set. If everything
> PASSes then we're all dandy and the output is clean. If there are
> failures then the tester might want to look at the detailed library
> messages, in which case they can do a "make checkv" and get them from
> all the runs.
Ok that's fine. I will add a patch to the series that converts
run_tests.sh to using HUGETLB_VERBOSE=0 by default and remove the
LINKHUGERW_QUIET stuff. This should help with some of the library
warnings that multiple page size support will add as well.
--
Adam Litke - (agl at us.ibm.com)
IBM Linux Technology Center
-------------------------------------------------------------------------
This SF.Net email is sponsored by the Moblin Your Move Developer's challenge
Build the coolest Linux based applications with Moblin SDK & win great prizes
Grand prize is a trip for two to an Open Source event anywhere in the world
http://moblin-contest.org/redirect.php?banner_id=100&url=/
_______________________________________________
Libhugetlbfs-devel mailing list
Libhugetlbfs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libhugetlbfs-devel
