[gmx-users] pdb2gmx segmentation fault

Matthew Zwier mczwier at gmail.com
Fri Apr 22 20:09:20 CEST 2011


Nice catch on the readdir_r().  I wonder if the developers would
appreciate a bug report and patch for your fix.

On Fri, Apr 22, 2011 at 1:05 PM, Ragothaman Yennamalli
<ragothaman at gmail.com> wrote:
> Hi all,
> With the help of my colleague Nathan Weeks I am able to run pdb2gmx and all
> other commands successfully . Here are his responses which he wanted me to
> share with you in case someone is trying to install Gromacs in a similar
> system.
>
> We set CFLAGS that way due to the following message when running configure:
>
> ########################################
> checking whether gcc-4.3.2 accepts -native -fast -xO5 -fsimple=2
> -fnonstd -dalign -O3 -fomit-frame-pointer -finline-functions -Wall
> -Wno-unused -msse2 -funroll-
> all-loops -std=gnu99... no
> *******************************************************************
> * Sorry, these optimization settings don't seem to work for       *
> * your C compiler. Use make CFLAGS=..., or edit the top Makefile. *
> *******************************************************************
> ########################################
>
> The "-native -fast -xO5 -fsimple=2 -fnonstd -dalign" options that get
> generated are for Solaris Studio, but we're using GCC.  Also,
> "-D_POSIX_PTHREAD_SEMANTICS" is required on Solaris to make use of the POSIX
> versions of reentrant functions like ctime_r() and readdir_r() (if the
> standard _POSIX_C_SOURCE or _XOPEN_SOURCE feature test macros
> are set, _POSIX_PTHREAD_SEMANTICS will automatically be set as well).
>
> The memory fault that occurs when running pdb2gmx is due to a non-portable
> use
> of readdir_r() in futil.c; specifically, on Linux, the dirent structure
> has a "char d_name[256]" member, and on Solaris, it is "char d_name[1]", and
> the user is required to allocate memory to store its contents. The Linux
> readdir(3) man page gives an example of how to do this portably:
>
> http://www.kernel.org/doc/man-pages/online/pages/man3/readdir.3.html
>
> I applied this method to futil.c, and it fixed the memory fault problem:
>
> ########################################
> --- src/gmxlib/futil.c.orig     2011-03-15 07:44:30.000000000 -0500
> +++ src/gmxlib/futil.c  2011-04-20 13:04:58.388912208 -0500
> @@ -37,6 +37,7 @@
>  #include <config.h>
>  #endif
>
> +#include <stddef.h>
>  #include <stdio.h>
>  #include <stdlib.h>
>  #include <string.h>
> @@ -595,22 +596,28 @@
>
>  #ifdef HAVE_DIRENT_H
>
> -    struct dirent           tmp_dirent;
> +    struct dirent *         tmp_dirent;
>     struct dirent *         p;
> +    size_t len;
>
>
>     if(gmxdir!=NULL && gmxdir->dirent_handle!=NULL)
>     {
> -        rc = readdir_r(gmxdir->dirent_handle,&tmp_dirent,&p);
> +        len  = offsetof(struct dirent, d_name)
> +               + fpathconf(dirfd(gmxdir->dirent_handle), _PC_NAME_MAX) + 1;
> +        smalloc(tmp_dirent, len);
> +        rc = readdir_r(gmxdir->dirent_handle,tmp_dirent,&p);
> +
>         if(p!=NULL && rc==0)
>         {
> -            strncpy(name,tmp_dirent.d_name,maxlength_name);
> +            strncpy(name,tmp_dirent->d_name,maxlength_name);
>         }
>         else
>         {
>             name[0] = '\0';
>             rc      = ENOENT;
>         }
> +        sfree(tmp_dirent);
>     }
>     else
>     {
> ########################################
>
>> Also why the make -j 48, when you only have 8 cores?
>
> We have 8 6-core CPUs (48 cores total).
>
>> I'm not even sure if the linker is ok with make -jN for building gromacs,
>> it
>> is not listed as compatible for *BSD ports. Also, does your make (I
>> imagine
>> this is Solaris make) have all the same semantics as gnu-make (gmake)?
>
> We are using GNU make.
>
>> And why did you switch your CC between fftw and gromacs configures?
>> Solaris 'cc' may or may not be gcc43. Also, do we know if gcc43 is stable
>> especially with -O3...Try removing -O3 or move it to -O2 at most...
>
> The Sun Studio 12.1 C compiler worked for fftw, but not for gromacs.
>
>> To rule out a bad fftw library, you can set --with-fft to use the builtin
>> (slower) libfft instead of fftw (but fftw should not be affecting pdb2gmx
>> I don't think...).
>>
>> I too have the questions of your CFLAGS, do you not require -m64 either?
>
> With gcc 4.3, there was a compilation error when -m64 was added to CFLAGS:
>
> ###################################################
> make[5]: Entering directory
> `/tmp//gromacs-4.5.4/src/gmxlib/nonbonded/nb_kernel_ia32_sse'
> /bin/sh ../../../../libtool   --mode=compile gcc-4.3.2   -m64 -g
> -msse2 -std=gnu99 -D_POSIX_PTHREAD_SEMANTICS -D_REENTRANT -pthread
> -I./include -c -o nb_kernel010_ia32_sse.lo nb_kernel010_ia32_sse.s
>  gcc-4.3.2 -m64 -g -msse2 -std=gnu99 -D_POSIX_PTHREAD_SEMANTICS
> -D_REENTRANT -pthread -I./include -c nb_kernel010_ia32_sse.s  -fPIC
> -DPIC -o .libs/nb_kernel010_ia32_sse.o
> nb_kernel010_ia32_sse.s: Assembler messages:
> nb_kernel010_ia32_sse.s:87: Error: suffix or operands invalid for `push'
> nb_kernel010_ia32_sse.s:89: Error: suffix or operands invalid for `push'
> nb_kernel010_ia32_sse.s:90: Error: suffix or operands invalid for `push'
> nb_kernel010_ia32_sse.s:91: Error: suffix or operands invalid for `push'
> nb_kernel010_ia32_sse.s:92: Error: suffix or operands invalid for `push'
> nb_kernel010_ia32_sse.s:93: Error: suffix or operands invalid for `push'
> nb_kernel010_ia32_sse.s:94: Error: suffix or operands invalid for `push'
> nb_kernel010_ia32_sse.s:795: Error: suffix or operands invalid for `pop'
> nb_kernel010_ia32_sse.s:796: Error: suffix or operands invalid for `pop'
> nb_kernel010_ia32_sse.s:797: Error: suffix or operands invalid for `pop'
> nb_kernel010_ia32_sse.s:798: Error: suffix or operands invalid for `pop'
> nb_kernel010_ia32_sse.s:799: Error: suffix or operands invalid for `pop'
> nb_kernel010_ia32_sse.s:800: Error: suffix or operands invalid for `pop'
> nb_kernel010_ia32_sse.s:865: Error: suffix or operands invalid for `push'
> nb_kernel010_ia32_sse.s:867: Error: suffix or operands invalid for `push'
> nb_kernel010_ia32_sse.s:868: Error: suffix or operands invalid for `push'
> nb_kernel010_ia32_sse.s:869: Error: suffix or operands invalid for `push'
> nb_kernel010_ia32_sse.s:870: Error: suffix or operands invalid for `push'
> nb_kernel010_ia32_sse.s:871: Error: suffix or operands invalid for `push'
> nb_kernel010_ia32_sse.s:872: Error: suffix or operands invalid for `push'
> nb_kernel010_ia32_sse.s:1344: Error: suffix or operands invalid for `pop'
> nb_kernel010_ia32_sse.s:1345: Error: suffix or operands invalid for `pop'
> nb_kernel010_ia32_sse.s:1346: Error: suffix or operands invalid for `pop'
> nb_kernel010_ia32_sse.s:1347: Error: suffix or operands invalid for `pop'
> nb_kernel010_ia32_sse.s:1348: Error: suffix or operands invalid for `pop'
> nb_kernel010_ia32_sse.s:1349: Error: suffix or operands invalid for `pop'
> make[5]: *** [nb_kernel010_ia32_sse.lo] Error 1
> ##############################################
>
>
> --
> ****************************************
> Ragothaman M Yennamalli, Ph.D.
> Postdoctoral Research Associate
> 1012 Crop Genome Informatics Laboratory
> Department of Genetics, Development and Cell Biology
> Iowa State University
> Ames, Iowa 50011-3260 USA
>
> +1 515-294-8971 (Office)
> +1 515-294-8280 (Fax)
> +1 515-851-1016 (Mobile)
>
> "When you can do the common things of life in an uncommon way, you will
> command the attention of the world." -George Washington Carver
>
> http://www.public.iastate.edu/~raghu/
> http://www.artistrkrishnarao.com/
>
> ***************************************
>
>
> --
> gmx-users mailing list    gmx-users at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-users
> Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
> Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>



More information about the gromacs.org_gmx-users mailing list