[Pvfs2-users] Problems with pvfs-2.1.6 and suse-10.x

[Murali Vilayannur]

Hi,
Thanks for the test program and the report!
I am not able to trigger this on my machine with your program..
can you tell me the command line that causes it to reliably fail?
Also would it be possible to get an strace output of the failing program?
strace -f a.out
Sam and I had checked in a few fixes to straighten out some size_t and
ssize_t mixups
recently. However, those were fixes only for ppc64 and should not have
affected x86/32 bit archs..but something could have regressed..
Can you also do this prior to running the program and send the
relevant kernel logs?
Edit the src/kernel/linux-2.6/Makefile and uncomment the EXTRA_CFLAGS
+= -DPVFS2_KERNEL_DEBUG and recompile..
then after loading the kernel module.. run
sudo sh -c "echo 1028 > /proc/sys/pvfs2/debug"

then run your program and send me the kernel logs.. (if it is too
large, send it to me alone)
thanks again for the report!
Murali


On 1/4/07, Trach-Minh Tran <trach-minh.tran at epfl.ch> wrote:
> Dear all,
>
> I am testing pvfs-2.1.6 and suse-10.2 (smp kernel 2.6.18, gcc-4.1.2, glibc-2.5)
> on a single host as well as a 2 data servers configuration. The test
> platform is the Xeon 5160 bi-processor dual core node. All the build,
> config and some simple tests (with pvfs2-ping, pvfs2-cp, ...) work fine.
> Even a MPI program using MPI-IO (with mpich2-1.0.5) ran correctly!
>
> However, when I try to do simple read/write with the attached Fortran
> program on the pvfs2 mounted volume, I start to have problems. More
> specifically, when the program is compiled with the Intel Fortran-9.1,
> it crashes when an array *larger than 16 MB* is read (although a previous
> "write" has passed) with the following kernel message
> ...
> Jan  4 08:50:04 a1 kernel: pvfs2_bufmap_copy_to_user_iovec: failed allocating memory
> Jan  4 08:50:04 a1 kernel: /usr/local/src/pvfs2/pvfs-2.6.1/src/kernel/linux-2.6/file.c line 278: pvfs2_file_read: Failed to copy-out buffers.  Please make sure that the pvfs2-client is running (-12)
> ...
>
> While when compiled with gfortran, it crashes already on the "write"!
> ...
> Jan  4 08:50:51 a1 kernel: pvfs2_bufmap_copy_iovec_from_user: failed allocating memory
> Jan  4 08:50:51 a1 kernel: /usr/local/src/pvfs2/pvfs-2.6.1/src/kernel/linux-2.6/file.c line 220: Failed to copy-in buffers. Please make sure that the pvfs2-client is running. -12
> ...
>
> Additional infos:
>  - The same problems occur on suse-10.1 (smp kernel 2.6.16, gcc-4.1.0, glibc-2.4)
>  - pvfs2-1.5.1 has no such problems on suse-10.1; however this version has
>    build problem on suse-10.2, due to the kernel version :-(
>
> Thanks for any help!
>
> -Minh.
>
>
> program main
>   implicit none
>   real, dimension(:), allocatable :: array
>   real :: x0, x1
>   integer :: size, n, i, nits
> !
>   write(*,*) 'Enter size of record (MB) and nits'
>   read(*,*) size, nits
>   n=size*1024*1024/4
> !
>   allocate(array(n))
>   open(99, file='test.out', form='unformatted')
>   do i=1,nits
>      array(1:n)=i
>      write(99) array(1:n)
>   end do
>   close(99)
>   print*, 'writing ok!'
> !
>   open(99, file='test.out', form='unformatted')
>   do i=1,nits
>      read(99) array(1:n)
>      x0=minval(array(1:n))
>      x1=maxval(array(1:n))
>      if ( x0 .ne. real(i) .and. x1 .ne. real(i) ) then
>         print*, 'error in reading, x0, x1', x0, x1
>      else
>         print*, 'reading record', i, '   OK!'
>      end if
>   end do
>   close(99)
>
>
> end program main
>
>
> _______________________________________________
> Pvfs2-users mailing list
> Pvfs2-users at beowulf-underground.org
> http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
>
>
>