Re: Unresolved simbols error when trying to install BLCR modules

From: Paul H. Hargrove (PHHargrove_at_lbl_dot_gov)
Date: Thu Sep 01 2005 - 11:16:33 PDT

  • Next message: Paul H. Hargrove: "Re: thaw_threads returned error"
    Adolfo,
    
       It looks like this is a variant on bug 466.  The lines from 
    System.map show that the symbol irq_stat is exported, but not versioned. 
      Somehow, the headers used to build blcr are corresponding to a 
    versioned symbol.
       I am not sure how this would happen, and am concerned that it may be 
    evidence that the kernel and headers might not match.  However, I'd like 
    to suggest a possible work-around.  Please try rebuilding blcr after 
    commenting out the following 6 lines near the top of vmadump/vmadump.c:
    
    #if defined(CONFIG_MODVERSIONS) && ! defined(MODVERSIONS)
       #define MODVERSIONS
    #endif
    #if defined(MODVERSIONS)
       #include <linux/modversions.h>
    #endif
    
    -Paul
    
    Adolfo J. Banchio wrote:
    > Paul,
    > 
    > thanks for the prompt answer. 
    > 
    > These are the outputs of the commands you asked:
    > 
    > $ uname -r
    > 2.4.21-20.EL
    > 
    > $ grep irq_stat /boot/System.map
    > c029ef6a R __kstrtab_irq_stat
    > c02ab190 R __ksymtab_irq_stat
    > c03b7fc0 B irq_stat
    > 
    > 
    > I do not know if this proves the presence of the 466 bug,
    > but to complete the information the kernel is from
    > ROCKS (a cluster distribution) 3.3.0, based on RH EL.
    > 
    > The installed kernel-source package is
    > 
    >  kernel-source-2.4.21-20.EL.i386.rpm
    > 
    > and could be downloaded from:
    > http://www.rocksclusters.org/ftp-site/rocks-3.3.0/rocks-dist/enterprise/3/en/os/i386/RedHat/RPMS/
    > 
    > 
    > But I'm not whure if the running kernel is exactly this
    > (although the configure script doesn't claim is a different one).
    > When istalling the distribution it also install the following
    > packages
    > 
    >    kernel-unsupported-2.4.21-20.EL.athlon.
    >    kernel-2.4.21-20.EL.athlon.
    >    kernel-pcmcia-cs-3.1.31-13.i386.
    >    kernel-source-2.4.21-20.EL.i386.
    >    kernel-utils-2.4-8.37.6.i386.
    >    rocks-kernel-3.3.0-13.i386.
    > 
    > All these (and the problem) is on an Athlon machine (the frontend).
    > The nodes are Pentium 4 machines, and there the installation 
    > of BLCR and loading of modules happens without problems. On those
    > machines is the same kernel (but the smp version) running, and
    > the installed packages are:
    > 
    >  kernel-smp-unsupported-2.4.21-20.EL.i686.
    >  kernel-unsupported-2.4.21-20.EL.i686.
    >  kernel-2.4.21-20.EL.i686.
    >  kernel-smp-2.4.21-20.EL.i686.
    >  kernel-pcmcia-cs-3.1.31-13.i386.
    >  kernel-source-2.4.21-20.EL.i386.
    >  kernel-utils-2.4-8.37.6.i386.
    >  rocks-kernel-3.3.0-13.i386.
    > 
    > 
    > I hope this helps to understand the problem. Since it work on the
    > nodes, it is not from high priority for me to solve the issue, but
    > it would be nice to, and specially if this could help to improve 
    > BLCR. Which, by the way, it's GREAT !!.  
    > 
    > thanks in advance,
    > 
    > adolfo
    > 
    > 
    > P.S.: separately I send another question from the working BLCR.
    > 
    > 
    > 
    > 
    > On Wed, 2005-08-31 at 19:46, Paul H. Hargrove wrote:
    > 
    >>Adolfo,
    >>
    >>  I am not certain about the cause of your problem.  The unresolved
    >>symbol is a versioned one (the _R94d0d943) and therefore must be
    >>exported by the kernel.  However, if the running kernel and the source
    >>(as configured) don't match this is just the kind of thing one could
    >>expect to happen.  Just for a sanity check, please ensure that "uname
    >>-r" says "2.4.21-20.EL", and tell me what "grep irq_stat
    >>/boot/System.map" yields.
    >>  If I can get a hold of the sources for the kernel in question (please
    >>do *NOT* e-mail them to me), I should be able to see if there is
    >>something wierd in BLCR.  Could you point me an an ftp or http download
    >>site for the kernel source (I assume EL is RedHat Enterprise Linux, but
    >>I  don't know which release).
    >>  It is also possible that this is a variation of bug 466 in which the
    >>kernel is not using versioned symbols but kernel modules are.  If so,
    >>the grep output requested above will prove it.
    >>
    >>-Paul
    >>
    >>Adolfo J. Banchio wrote:
    >>
    >>>Hi,
    >>>
    >>>I'm trying to install the BLCR modules, but after
    >>>compiling the I get the following error when 
    >>>loading the first one:
    >>>
    >>>%/sbin/insmod /usr/local/lib/blcr/2.4.21-20.EL/vmadump_blcr.o
    >>>
    >>>/usr/local/lib/blcr/2.4.21-20.EL/vmadump_blcr.o: unresolved symbol
    >>>irq_stat_R94d0d943
    >>>
    >>>
    >>>I've checked te mailing list and find a similar thread, and
    >>>followed the suggestions there:
    >>>
    >>>- copy /boot/config file to /usr/src/linux/.config
    >>>- make oldconfig
    >>>-  make dep
    >>>
    >>>and the ./configure --with-system-map=/boot/System.map
    >>>etc
    >>>
    >>>But still get the mentioned error.
    >>>
    >>>Any help will be apreciated.
    >>>
    >>>thanks in advance
    >>>
    >>>
    >>>adolfo
    >>>
    >>>
    >>> 
    >>
    > 
    
    
    -- 
    Paul H. Hargrove                          PHHargrove_at_lbl_dot_gov
    Future Technologies Group
    HPC Research Department                   Tel: +1-510-495-2352
    Lawrence Berkeley National Laboratory     Fax: +1-510-486-6900
    

  • Next message: Paul H. Hargrove: "Re: thaw_threads returned error"