Re: [LTP] expected errors on 64 bit zSeries

[Gerhard T. <TO...@de...>]

BTW, the pread02 testcase failure on 31 bit platforms has the same cause as
the pwrite02 failure described below.  We are also going to fix that in the
glibc.

Regards / Mit freundlichen Gruessen
Gerhard

Gerhard Tonn, Linux for eServer Development, +(49)-7031-16-4716, Lotus
Notes: ton@ibmde,
   Internet: to...@de...


Gerhard Tonn/Germany/IBM@IB...@li... on 10.07.2002 10:27:07

Please respond to Gerhard Tonn/Germany/IBM@IBMDE

Sent by:    ltp...@li...


To:    ltp...@li...
cc:    Susanne Wintenberger/Germany/IBM@IBMDE
Subject:    [LTP] expected errors on 64 bit zSeries



Hi,
I am currently fixing some of the expected errors of the ltp on the Linux
for zSeries kernel on 31 bit, 64 bit and the 31 bit emulation on 64 bit.
Below are some findings for the 64 bit case in advance. It refers to
http://ltp.sourceforge.net/errorstemp.php?file=20020507-Zseries64.html.

>Tests that fail all of the time:
>                    ioctl01 - investigating
>                    ioctl02 - investigating
>                    mmap01 - investigating
This is a known linux for zSeries kernel bug, will be fixed in 2.4.19
>                    mprotect03 - investigating
same as above
>                    personality01 - sf#511474
>                    personality02 - sf#511474
>                    pwrite02 - investigating
This is in my opinion a testcase bug and a glibc bug on 31 bit platforms.
http://www.opengroup.org/onlinepubs/007908799/xsh/write.html defines that
pwrite returns EINVAL, if the offset is negative. However the glibc
implementation on 31 bit doesn't check it but does a conversion from an
off_t (defined as long) to an loff_t (defined as lon long), but ignoring
the sign. The 64 bit implementation is correct. We are going to fix this in
the glibc function.
>                    read02 - investigating
It seems as if the buf address value is not checked as long as the accessed
file is empty. The testcase works fine if the file is non-empty. Could you
fix the testcase to make the file non-empty?
>                    setgroups03 - investigating
The testcase fails, because  on the 64 bit system the glibc implementation
issues directly a  system call without any additional checks. The kernel
checks first
 for permission which fails and returns EPERM. The attached file contains a
patch for the testcase that changes uid only for the second testcase.
>                    setrlimit03 - investigating
>                    sockioctl01 - investigating
>                    waitpid04 - investigating
The failure of the testcase seems to depend on the value of pid  that is
not explicitly set. Since it is 0 on the zSeries platforms for some reason,
the testcase fails. If I set pid explicitly to some invalid  value, the
testcase works. The attached file contains a patch for the testcase.
>                    writev02 - investigating
The writev call of the testcase passes as third parameter the value 3 to
indicate that iovec has  three elements, but it has only two. The attached
file contains a patch for the testcase.
>                    writev05 - investigating
same as above
>                    ar - investigating
>                    sem02 - investigating
>                    float_power - investigating

(See attached file: ltp.patch64)

Regards / Mit freundlichen Gruessen
Gerhard

Gerhard Tonn, Linux for eServer Development, +(49)-7031-16-4716, Lotus
Notes: ton@ibmde,
   Internet: to...@de...