Description of problem: rpm --rebuild perl-5.6.1-26.72.3.src.rpm on Linux390 causes: (...) op/study.............FAILED at test 0 (...) running ./perl harness in /usr/src/redhat/BUILD/perl-5.6.1/t results in: (...) op/study............dubious Test returned status 0 (wstat 11, 0xb) (...) How reproducible: The problem occurs by testing op/study.t we run ./perl op/study.t and get 1..24 ok 1 ok 2 Segmentation fault Everytime we run perl 5.6.1 using study.t or start our application we can see following in /var/log/messages: Jan 14 12:53:37 vsysad01 kernel: User process fault: interruption code 0x10 Jan 14 12:53:37 vsysad01 kernel: failing address: 0 Jan 14 12:53:37 vsysad01 kernel: CPU: 0 Tainted: P Jan 14 12:53:37 vsysad01 kernel: Process perl (pid: 15530, task: 01070000, ksp: 01071e78) Jan 14 12:53:37 vsysad01 kernel: User PSW : 070de000 804ab796 Jan 14 12:53:37 vsysad01 kernel: User GPRS: 00300001 ffffffff 00000000 00000000 Jan 14 12:53:37 vsysad01 kernel: 00000000 00000006 7ffff718 004e1278 Jan 14 12:53:37 vsysad01 kernel: 004dc8f0 004d9160 004d9160 004e1278 Jan 14 12:53:37 vsysad01 kernel: 00000003 804ab3e0 80461f60 7ffff6b0 Jan 14 12:53:37 vsysad01 kernel: User ACRS: 00000000 00000000 00000000 00000000 Jan 14 12:53:37 vsysad01 kernel: 00000000 00000000 00000000 00000000 Jan 14 12:53:37 vsysad01 last message repeated 2 times Jan 14 12:53:37 vsysad01 kernel: User Code: 58 60 20 04 58 30 f0 84 58 20 d0 24 58 40 30 00 58 30 20 00 Additional info: perl-5.6.1-26.72.3a.src.rpm does not solve our problem! If we run our application using perl 5.6.1 this fault occurs too. So just bypassing the test like in .72.3a release doesn't make us lucky! :-( This problem does not occur on x86 plattform. diff of perl.spec in perl-5.6.1-26.72.3.src.rpm and perl-5.6.1-26.72.3a.src.rpm: --- perl.spec Wed Feb 20 20:54:45 2002 +++ perl.spec.a Sun Mar 10 21:01:51 2002 @@ -9,7 +9,7 @@ %define threading 0 %define largefiles 0 %define ndbm 1 -%define rhrelease .72.3 +%define rhrelease .72.3a %endif %define perlver 5.6.1 @@ -265,7 +265,7 @@ make -f Makefile -%ifnarch ia64 +%ifnarch ia64 s390 s390x make -f Makefile test %endif Please help! Thanks in advance and best regards Jan As
Some tests _want_ to have a seg-fault, so you have to look into the tests to see what is going on. Also the current perl seems to run these tests without excluding them. greetings, Florian La Roche
Sorry! I don't agree to. On x86 Plattform the test pass. 1..24 ok 1 ok 2 (...) ok 23 ok 24 By editing of .rpmrc: optflags: s390 -O0 I turn off the optimizing compilation. After rebuild of perl rpm the test pass on s390, too. But this is a bypass only! I want use the default optimizing -O2 for better performace. Customer needs perl-5.6.1 exactly. So we can not use perl-5.6.0 or perl-5.8.0. Thanks and best regards Jan As
I suggest building this then with -O0 if no perl update is allowed. Current perl rpm builds with all tests passing AFAIK. Thanks for this bug report, Florian La Roche