Re: [Gumstix-users] Compiling JamVM + Classpath to avoid segfault

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

Robert,
	I just went thru your patch and found that on 1410, the only thing that
is needed is the patch to classpath.mk and to install jikes on my build
machine (free-java-sdk on ubuntu). dosnt seam to require the uClib patch
Jason


On Wed, 2007-05-16 at 14:44 -0600, ro...@in... wrote:
> Hello,
> 
> I believe I have found a combination of JamVM + Classpath + other stuff
> which does not trigger a segfault (a bug which has been reported by
> various people). Here are my notes. I hope this is useful to someone.
> 
> There is some stuff about wifi in there. You can ignore that.
> 
> Robert Dodier
> 
> PS. The caps-lock stuff is to help me remember. 8^)
> 
> ------------------------- cut here -------------------------
> YOU NEED TO EXECUTE MAKE AT LEAST THREE TIMES. FIRST TIME TO BUILD A VANILLA
> BASE SYSTEM, SECOND TIME TO BUILD SPECIAL-PURPOSE PACKAGES. MY EXPERIENCE HAS
> BEEN THAT IT WILL FAIL IF YOU TRY TO BUILD THE SPECIAL-PURPOSE STUFF AND BASE
> SYSTEM ALL IN ONE GO.
> 
> SUMMARY: Gumstix svn -r1089 (includes Linux kernel 2.6.17) + uClibc 0.9.28.3 +
> Classpath 0.90 + JamVM 1.4.2. (I tried MANY other combinations and this is the
> only one which works reliably in my experience.)
> 
> Before executing make for the first time:
> 
>    1. svn co -r1089 http://svn.gumstix.com/gumstix-buildroot/trunk
> gumstix-buildroot-r1089
>           * ... and then cd to the new sandbox directory
>    2. Toolchain options via make menuconfig
>           * Verify that uClibc daily snapshot is unselected
>           * Verify that gcc 3.4.5 (not any gcc 4.x version) is selected
>    3. uClibc version
>           * Edit toolchain/uClibc/uclibc.mk and change 0.9.28 to 0.9.28.3
>    4. Copy toolchain/uClibc/uClibc-revert-ptfork-malloc-lock.patch (shown below)
>           * Subversion log for ptfork.c is unclear as to purpose of the malloc
> lock stuff. Comments in code suggest it is just a hack. I don't know of any bad
> effects of reverting the malloc lock stuff; I haven't observed any and I haven't
> read of any.
>    5. Remove toolchain/uClibc/uClibc-gcc-41-raise-error.patch
>           * The patched code has changed, so the patch fails. I tried to adjust
> the patch accordingly but could not make it succeed. Gcc 3.4.5 is selected (see
> above) so any gcc 4.x patch is irrelevant, so I just erased it.
>    6. Busybox (multi-purpose executable) customization
>           * Edit target/device/Gumstix/basix-connex/busybox.config to customize
> (NOT package/busybox/busybox.config, that file is ignored!)
>                 o Enable nohup via CONFIG_NOHUP=y
>    7. Linux kernel configuration
>           * Linux kernel version is chosen by svn revision number (r1089 =>
> Linux 2.6.17)
> 
> 
> Now run make (for the first time, you'll need to run make again later).
> 
> After first make succeeds and before second make:
> 
>    1. Package selection via make menuconfig
>           * Select jamvm, classpath, and zlib headers
>           * Select ntp, strace, and tcpdump
>           * Select wifistix and wireless tools
>    2. Apply patch to classpath.mk to cause Classpath to be compiled with Jikes
> (Java compiler)
>           * classpath.mk patch is shown below
>           * In addition to enabling compile by Jikes, patch also removes
> dependency of Classpath on GUI libraries
>           * Requires Jikes installation.
>                1. Download jikes via ftp from rpmfind.net in
> linux/fedora/extras/5/SRPMS/jikes-1.22-5.fc5.src.rpm
>                2. Unpack & rebuild srpm, install new jikes rpm
> 
> 
> Now run make again.
> 
> After second make succeeds:
> 
>    1. Modify build_arm_nofpu/root/etc/network/interfaces with your Wifi key and
> essid.
>           * Final line needs to be something like: pre-up /sbin/iwconfig $IFACE
> key <MY-KEY> key open essid <MY-ESSID> txpower 100mW where <MY-KEY> and
> <MY-ESSID> represent the key and essid for the device, respectively. Note that
> $IFACE is a literal dollar-sign followed by IFACE.
>    2. Need to generate shared object (.so) libraries from .a libraries for
> Classpath (Java implementation).
>          1. CC=`pwd`/build_arm_nofpu/staging_dir/bin-ccache/arm-linux-uclibc-gcc
>          2. pushd ./build_arm_nofpu/root/usr/lib/classpath
>          3. mkdir /tmp/FOO
>          4. d=`pwd`
>          5. for f in io lang langreflect net nio util; do pushd /tmp/FOO; ar x
> $d/libjava$f.a; $CC --shared *.o; mv a.out $d/libjava$f.so; rm *.o; popd; done
>          6. rm libjava*.a libjava*.la
>          7. popd
> 
> 
> Now run make yet again after these changes (so they are incorporated into the
> filesystem image)
> 
> You're done!! Now burn the filesystem image onto your Gumstix. Have fun!!
> 
> Patch for classpath.mk
> 
> Save the following as /tmp/classpath.mk-patch and replace all <TAB> with literal
> tab characters. The resulting should have md5sum =
> 88b0338fe4e365592c35e5b3614c7cf1. Then apply the patch by cat
> /tmp/classpath.mk-patch | patch -p0.
> 
> --- package/classpath/classpath.mk<TAB>(revision 1089)
> +++ package/classpath/classpath.mk<TAB>(working copy)
> @@ -24,10 +24,10 @@
>  $(CLASSPATH_DIR)/.configured: $(CLASSPATH_DIR)/.unpacked
>  <TAB>(cd $(CLASSPATH_DIR); rm -rf config.cache; \
>  <TAB><TAB>$(TARGET_CONFIGURE_OPTS) CC_FOR_BUILD=$(HOSTCC) \
> -<TAB><TAB>CFLAGS="$(TARGET_CFLAGS)" \
> +<TAB><TAB>CFLAGS="$(TARGET_CFLAGS) -msoft-float -fPIC" \
>  <TAB><TAB>./configure \
>  <TAB><TAB>--target=$(GNU_TARGET_NAME) \
> -<TAB><TAB>--host=$(GNU_TARGET_NAME) \
> +<TAB><TAB>--host=arm \
>  <TAB><TAB>--build=$(GNU_HOST_NAME) \
>  <TAB><TAB>--prefix=/usr \
>  <TAB><TAB>--exec-prefix=/usr \
> @@ -39,6 +39,12 @@
>  <TAB><TAB>--localstatedir=/var \
>  <TAB><TAB>--mandir=/usr/man \
>  <TAB><TAB>--infodir=/usr/info \
> +<TAB><TAB>--with-jikes \
> +<TAB><TAB>--without-gcj \
> +<TAB><TAB>--without-x \
> +<TAB><TAB>--disable-qt-peer \
> +<TAB><TAB>--disable-gconf-peer \
> +<TAB><TAB>--disable-plugin \
>  <TAB><TAB>--disable-gtk-peer \
>  <TAB>);
>  <TAB>touch  $(CLASSPATH_DIR)/.configured
> 
> 
> 
> New patch for uClibc, toolchain/uClibc/uClibc-revert-ptfork-malloc-lock.patch
> 
> Save the following and replace all <TAB> with literal tab characters, and
> <SPACE> with a literal space character. The resulting should have md5sum =
> 11d2de2cde8f3378635f2838189e933c.
> 
> --- uClibc/libpthread/linuxthreads/ptfork.c<TAB>2007/01/26 00:33:53<TAB>17533
> +++ uClibc/libpthread/linuxthreads/ptfork.c<TAB>2007/01/26 00:47:48<TAB>17534
> @@ -26,15 +26,6 @@
>  #include "pthread.h"
>  #include "internals.h"
> <SPACE>
> -#warning hack alert... should be sufficent for system(), but what about other
> libc mutexes?
> -#include <bits/uClibc_mutex.h>
> -
> -__UCLIBC_MUTEX_EXTERN(__malloc_lock);
> -
> -#define __MALLOC_LOCK<TAB><TAB>__UCLIBC_MUTEX_LOCK(__malloc_lock)
> -#define __MALLOC_UNLOCK<TAB><TAB>__UCLIBC_MUTEX_UNLOCK(__malloc_lock)
> -#warning hack alert block end
> -
>  struct handler_list {
>    void (*handler)(void);
>    struct handler_list * next;
> @@ -100,18 +91,9 @@
>    parent = pthread_atfork_parent;
>    pthread_mutex_unlock(&pthread_atfork_lock);
>    pthread_call_handlers(prepare);
> -
> -#warning hack alert
> -  __MALLOC_LOCK;
> -
>    pid = __libc_fork();
> -
> -#warning hack alert
> -  __MALLOC_UNLOCK;
> -
>    if (pid == 0) {
>      __pthread_reset_main_thread();
> -#warning need to reconsider __fresetlockfiles!
>      __fresetlockfiles();
>      pthread_call_handlers(child);
>    } else {
> 
> -------------------------------------------------------------------------
> This SF.net email is sponsored by DB2 Express
> Download DB2 Express C - the FREE version of DB2 express and take
> control of your XML. No limits. Just data. Click to get it now.
> http://sourceforge.net/powerbar/db2/
> _______________________________________________
> gumstix-users mailing list
> gum...@li...
> https://lists.sourceforge.net/lists/listinfo/gumstix-users
>