Re: [LTP] Patch for zSeries testcase failures

[Robert W. <ro...@us...>]

Thanks for the patch...I will try in get this into CVS by close of business
on Monday (July 15th).

- Robbie

Robert V. Williamson <ro...@us...>
Linux Test Project
IBM Linux Technology Center
Phone: (512) 838-9295   T/L: 638-9295
http://ltp.sourceforge.net


                                                                                                                             
                      Gerhard                                                                                                
                      Tonn/Germany/IBM@IBMDE           To:       ltp...@li...                              
                      Sent by:                         cc:       Susanne Wintenberger/Germany/IBM@IBMDE                      
                      ltp...@li...ur        Subject:  [LTP] Patch for zSeries testcase failures                   
                      ceforge.net                                                                                            
                                                                                                                             
                                                                                                                             
                      07/12/2002 05:02 AM                                                                                    
                                                                                                                             
                                                                                                                             




Hi,
attached is a file that has testcase fixes for several failing testcases on
the zSeries platforms. It includes also the fixes for 64 bit I have posted
some days ago, but adapted to the latest ltp version.

We have recently run the ltp testsuite on the IBM zSeries platforms. It has
been executed on a 31 bit and a 64 bit kernel and a 31 bit system call
emulation on a 64 bit kernel.

Most of the failing testcases don't succeed since the value -1 casted to an
address is not an invalid address on the 31 bit zSeries platform.
On most platforms kernel and user space share the same address space. An
address passed from user space to kernel space is invalid on these
platforms
if the memory at that address is not mapped in or protected or if it's a
kernel address. Passing -1 casted to an address doesn't fail on 31 bit
zSeries
platforms, since it addresses the bottom of the stack, and is not a kernel
address as on most other platforms.

In order to generate invalid addresses portable we should use the mmap call
with protection PROT_NONE. See the attached file for a possible solution.

Another common cause for failures is the fact that some testcases trigger
two errors and it is undefined which one has to be reported first. Below
are
the detailed findings.

The following testcases fail on the emulation layer, since -1 casted to an
address is valid in this environment.
   - access03    1
   - access03    2
   - access03    3
   - access03    4
   - access05    6
   - chdir04     3
   - chmod06     4
   - chown04     4
   - chroot03    4
   - creat06     5
   - execve03    4
   - lchown02    4
   - link04      7
   - link04     13
   - lstat02     3
   - mkdir01     1
   - mkdir03     1
   - mknod06     3
   - open08      6
   - rename08    1
   - rename08    2
   - rmdir02     5
   - rmdir05     4
   - stat03      3
   - stat06      7
   - statfs02    4
   - symlink03    4
   - truncate03    4
   - unlink07    8
   - readv02     2
   - read02      3

fcntl13 fails on the emulation layer
  The fourth run passes two invalid parameters, the first one is an
invalid file pointer, the third one an invalid address. Since the
  emulation wrapper does some additional parameter checking, the third one
is checked before the first one which results in a different error code.
  See patch file for fix.

statfs03 fails on the emulation layer
  The testcase passes an invalid buf parameter. Since the emulation wrapper
does some additional parameter checking, it is checked
  before permission is checked which results in a different error code.
  See patch file for fix.

pread02 fails on the 31 bit kernel
  This is a glibc bug on 31 bit platforms.
http://www.opengroup.org/onlinepubs/007908799/xsh/read.html defines that
pread
   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 long long), but ignoring the sign. The 64 bit
implementation is correct. We are going to fix this in the glibc function.

recv01 - 3 , recvfrom01 - 5 and recvmsg01 - 5 fail on all zSeries kernels
  These testcases fail even when passing an invalid address since the
address is never dereferenced. It succeeds on most platforms
  using -1 as address, since checking for kernel addresses is always done
in advance, but as said above not applicable to the zSeries platform.

(See attached file: ltp-20020709-s390.patch)

Regards / Mit freundlichen Gruessen
Gerhard

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


#### ltp-20020709-s390.patch has been removed from this note on July 12
2002 by Robert Williamson