Le dimanche 7 juin 2009 04:20:57, Bryan Kadzban a écrit :
> Bruce Dubbs wrote:
> > jp wrote:
> >>> What version of the kernel is you host system running?
> >>
> >> uname -r on my host returns: 2.6.29
> >
> > Oops. I made a change to the host requirements section to require
> > 2.6.18 or later, but never updated the glibc instruction. The option
> > should be
> >
> > --enable-kernel=2.6.18
> >
> > In both chapter 5 and 6. If you have time, I'd be interested if the
> > test runs OK for you when using that switch.
>
Re-using the tools dir that I had got backed up, I re-runned glibc chapter 6
with --enable-kernel=2.6.18 and tests results are even worse than with --
enable-kernel=2.6.0:
[/sources/glibc-build/stdio-common/bug22.out] Error 1
[/sources/glibc-build/posix/wordexp-test.out] Error 1
and the nptl/tst-mutex{5,9,5a}fail, as well as the nptl/tst-cond10.
The c++-types-check.out fails, too.
I can see a process still running after the tests, see below ( I used ps axww
from my host to get it ).
There is still a doubt in my mind, because in chapter 5, glibc had been
configured with --enable-kernel=2.6.0
Now it's a bit late for me. Tomorrow, will restart the build from the
beginning, to make sure.
> I don't really see a reason this flag would have any effect at all (barring
> bugs in the glibc kernel version compatibility code, which is one thing the
> test is there to find anyway... :-) ).
>
> As I said last time this came up, setting --enable-kernel newer only causes
> glibc to build without some compatibility code (needed for older kernels).
> I'm trying to envision how that would cause one of the testsuite programs
> to run forever, and can't come up with any valid reason.
>
> (jp: Which program was actually still running? pstree is helpful here.
> What was its full command line?)
>
Bruce, the command still running after the tests was: /sources/glibc-
build/elf/ld-linux.so.2 --library-path /sources/glibc-build:/sources/glibc-
build/math:/sources/glibc-build/elf:/sources/glibc-build/dlfcn:/sources/glibc-
build/nss:/sources/glibc-build/nis:/sources/glibc-build/rt:/sources/glibc-
build/resolv:/sources/glibc-build/crypt:/sources/glibc-build/nptl
/sources/glibc-build/nptl/tst-mutex9
> If I had to guess, I'd say that test sometimes loops forever and sometimes
> finishes (in this setup, whatever "this setup" actually is), and the
> --enable-kernel had no effect on this behavior other than re-running the
> whole testsuite. But that's just a guess...
>
> > Additionally, the behavior you describe did not occur in our test
> > build. Are you deviating from the book in any way?
>
> I would suspect this to be the root cause, but let's find out for sure.
>
> :-)
Brian, I don't think I deviated from the book. I start the whole build again,
and will double check that I don't deviate.
jp
--
http://linuxfromscratch.org/mailman/listinfo/lfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page
