test

test2

Building toolchain 2.6 in a CRUX 2.7 host (using make 3.82) we can find a problem in glibc build:
*** mixed implicit and normal rules. Stop.

We can try to add a patch and fix this problem to build CRUX-ARM 2.6 (eabi/noeabi) in a CRUX 2.7 host (with make 3.82).
This should be tested in a 2.6 host and verify it builds right too. I’ll try to provide this last check while waiting for comments or opinions in this bug.

glibc-2.10.1-make382.patch
— glibc-2.10.1_orig/manual/Makefile 2012-02-08 08:05:11.777529411 +0000
+++ glibc-2.10.1/manual/Makefile 2012-02-08 08:06:46.189531779 +0000
@@ -232,7 +232,9 @@
.PHONY: stubs
stubs: $(objpfx)stubs
endif
-$(objpfx)stubs ../po/manual.pot $(objpfx)stamp%:
+$(objpfx)stubs ../po/manual.pot:
+ touch $@
+$(objpfx)stamp%:

      $(make-target-directory)
      touch $@

Makefile
diff –git a/Makefile b/Makefile
index aea2b53..5af5d60 100644
— a/Makefile
+++ b/Makefile
@@ -174,9 +174,9 @@ $(WORK)/glibc-$(GLIBC_VERSION).tar.bz2:
$(WORK)/glibc-ports-$(GLIBC_VERSION).tar.bz2:

      wget -P $(WORK) -c ftp://ftp.gnu.org/gnu/glibc/glibc-ports-$(GLIBC_VERSION).tar.bz2

$(WORK)/glibc-$(GLIBC_VERSION): $(WORK)/glibc-$(GLIBC_VERSION).tar.bz2 $(WORK)/glibc-ports-$(GLIBC_VERSION).tar.bz2
-

      tar -C $(WORK) -xvjf $(WORK)/glibc-$(GLIBC_VERSION).tar.bz2
      cd $(WORK)/glibc-$(GLIBC_VERSION) && \

+ patch -p1 -i $(WORK)/glibc-$(GLIBC_VERSION)-make382.patch && \

              tar xvjf $(WORK)/glibc-ports-$(GLIBC_VERSION).tar.bz2 && \
              mv glibc-ports-$(GLIBC_VERSION) ports && \
              sed -e 's/-lgcc_eh//g' -i Makeconfig

Error in acl references:

undefined reference to `acl_init’ undefined reference to `acl_create_entry’ ...

The underflow detection code of mpfr_gamma is buggy: some results may incorrectly be regarded as underflow.
This bug is fixed by the gamma-underflow patch, which also provides a testcase. Warning! While this patch avoids incorrect results on values close to underflow, it makes another problem appear: a possible infinite loop for some underflow cases. In particular, the testsuite will freeze on 32-bit machines.

We need to update ruby’s .footprint to fit with current arch and ABI.
NEW drwxr-xr-x root/root usr/lib/ruby/1.9/linux-eabi/

======⇒ ERROR: Footprint mismatch found:
MISSING -rwxr-xr-x root/root usr/lib/python2.7/lib-dynload/dbm.so
NEW -rwxr-xr-x root/root usr/lib/python2.7/lib-dynload/dbm_failed.so

We must to fix issues with archives

Check if we can add this option directly in toolchains.
This will allow to build initrd without building core-cross' glibc.

The script is useful to obtain ports from core-arm, opt-arm, xorg-arm overlays which are updated in upstream core, opt and xorg.
It would be fine to check ports for 'per device' overlays since gcc, pkgutils, ports and prt-get are ports that we overlayed for each device. In this case, the script should check each port against the upstream repo where the port is located.
For example,

case $repo in

efikamx|cubieboard|cubieboard2|raspberrypi)
# foreach port in repo
for port in $PORTS; do
  # get upstream repo where the port is located
  upstream_repo=""
  for up_repo in core opt xorg; do
    upstream_repo="$(basename $(dirname $(prt-get --no-std-config --config-prepend="prtdir /usr/ports/$up_repo" path $port 2>/dev/null) 2>/dev/null) 2>/dev/null)"
  done
  [ "$upstream_repo" == "" ] && exit
  # continue
  ...

Since we are adding all needed files to get dns resolved, probably we can think about this and remove it or fix it adding libc and ld-linux-armhf files.

# ping localhost
ping: bad address

Final initramfs size will grow 1M. Removing libnss and libresolv files won't make it decrease too much. Could this be interesting for a debugging initramfs and not for a media installer?

Small patch to add those files:
diff –git a/Makefile b/Makefile
index 5338441..7d88f5f 100644
— a/Makefile
+++ b/Makefile
@@ -152,6 +152,7 @@ $(WORK)/initramfs.cpio: check-root busybox e2fsprogs dialog $(WORK)/mnt gen_init

      done
      install -v -m 0755 $(WORK)/dialog-$(DIALOG_VERSION)/_install/usr/bin/dialog $(WORK)/mnt/usr/bin
      cp -dRv $(CLFS)/lib/libnss_{files*,dns*} $(CLFS)/lib/libresolv* $(WORK)/mnt/lib

+ cp -dRv $(CLFS)/lib/{ld-linux-armhf.so.3,libc-2.16.so,libc.so.6} $(WORK)/mnt/lib

      install -d  $(WORK)/mnt/usr/share/terminfo
      cp -dRv $(CLFS)/usr/share/terminfo/v $(WORK)/mnt/usr/share/terminfo
      install -v -m 0644 $(TOPDIR)/filesystem/{fstab,inittab,profile,protocols,*.conf} $(WORK)/mnt/etc

We need to be sure we build python (generic native build) with shm mounted.

======⇒ ERROR: Footprint mismatch found:
MISSING -rwxr-xr-x root/root usr/bin/nmap-update
MISSING -rw-r–r– root/root usr/man/man1/nmap-update.1.gz
======⇒ ERROR: Building '/pub/pkg/contrib/nmap#6.00-1.pkg.tar.xz' failed.

checking svn_client.h usability... no
checking svn_client.h presence... no
checking for svn_client.h... no
checking subversion-1/svn_client.h usability... no
checking subversion-1/svn_client.h presence... no
checking for subversion-1/svn_client.h... no
configure: WARNING: Not building nmap-update because libsvn1 was not found

IMHO, we should wait until upstream decission

There are two typos in Flyspray. The first is in the page titles. Note that there is a separator missing, and duplication of “CRUX-ARM” - here is the page title I have right now:

CRUX-ARM | bugsCRUX-ARM: New Task

I suggest this be fixed to look more like Crux’s, which reads:

Flyspray:: CRUX: New Task

So we should just replace “CRUX” with “CRUX-ARM”

The second typo is when a user enters an incorrect password. When a user enters a bad password, an error appears at the bottom of the webpage that says something about the “paassword” being incorrect or something. The word “password” is spelled incorrectly.

Current libgmp version 5.0.5 fails with assembler error.

The problem is located on upstream Pkgfile which forces the build for i686.
[...]
tmp-add_n.s: Assembler messages:
tmp-add_n.s:79: Error: bad instruction `xor %edx,%edx’ [...]

From the original Pkgfile, it specifies:
–build=i686-pc-linux-gnu

Solution:
- Use current build vars for both ABI’s:
- softfp: –build=arm-unknown-linux-gnueabi

It’s nice to be able to gzip modules to save space. Can –with-zlib please be added to kmod? I have it working well with my setup but it would be nice if this was mainlined.

Thanks!

Looks like there is a compile bug that’s been open since 2005 for arm platforms: http://sourceforge.net/p/libcgi/bugs/6/

Patching the string.c file and compiling fixes the package. We might want to fix it since upstream is unresponsive.

gcc-static is unnecessarily rebuilt each time make is invoked due to an incorrect filename extension in the Makefile target definition. Patch follows:

diff –git a/Makefile b/Makefile
index 4ab3882..568739d 100644
— a/Makefile
+++ b/Makefile
@@ -173,10 +173,10 @@ binutils-distclean: binutils-clean


# GCC-STATIC
-$(WORK)/gcc-$(GCC_VERSION).tar.bz2:
+$(WORK)/gcc-$(GCC_VERSION).tar.xz:

      wget -P $(WORK) -c ftp://gcc.gnu.org/pub/gcc/releases/gcc-$(GCC_VERSION)/gcc-$(GCC_VERSION).tar.xz

-$(WORK)/gcc-$(GCC_VERSION): $(WORK)/gcc-$(GCC_VERSION).tar.bz2
+$(WORK)/gcc-$(GCC_VERSION): $(WORK)/gcc-$(GCC_VERSION).tar.xz

      tar -C $(WORK) -xf $(WORK)/gcc-$(GCC_VERSION).tar.xz
      touch $(WORK)/gcc-$(GCC_VERSION)

There are some problems trying to cross-build perl to the new version.
Here there are some notes about the work done (but not working):
- The Pkgfile tries to build the needed hosts tools and then build perl.
- The config provided got from versatile native configure.
May be a bad way to try to cross-build it. Waiting comments and suggestions to cross-build perl.

Report:

/bin/sh ../../libtool –tag=CC –mode=compile gcc -DHAVE_CONFIG_H -I. -I../../src -I../../src -I../../src/libmpg123 -DOPT_ARM -DREAL_IS_FIXED -O2 -fomit-
frame-pointer -funroll-all-loops -finline-functions -ffast-math -O2 -march=armv7-a -mfpu=vfpv3-d16 -mthumb -MT layer3.lo -MD -MP -MF .deps/layer3.Tpo -c -o l
ayer3.lo layer3.c
libtool: compile: gcc -DHAVE_CONFIG_H -I. -I../../src -I../../src -I../../src/libmpg123 -DOPT_ARM -DREAL_IS_FIXED -O2 -fomit-frame-pointer -funroll-all-loops
-finline-functions -ffast-math -O2 -march=armv7-a -mfpu=vfpv3-d16 -mthumb -MT layer3.lo -MD -MP -MF .deps/layer3.Tpo -c layer3.c -fPIC -DPIC -o .libs/layer3
.o
/tmp/cckBsykf.s: Assembler messages:
/tmp/cckBsykf.s:2839: Error: shift must be constant – `orr fp,fp,r7,lsl sl’ /tmp/cckBsykf.s:2868: Error: shift must be constant – `orr fp,fp,r7,lsl sl’ /tmp/cckBsykf.s:3071: Error: shift must be constant – `orr r5,r5,r7,lsl sl’ /tmp/cckBsykf.s:3268: Error: shift must be constant – `orr r6,r6,r7,lsl sl’ /tmp/cckBsykf.s:3290: Error: shift must be constant – `orr r6,r6,r7,lsl sl’ /tmp/cckBsykf.s:3313: Error: shift must be constant – `orr r6,r6,r7,lsl sl’ /tmp/cckBsykf.s:3336: Error: shift must be constant – `orr fp,fp,r4,lsl r5’ /tmp/cckBsykf.s:3358: Error: shift must be constant – `orr fp,fp,r3,lsl r5’ /tmp/cckBsykf.s:4653: Error: shift must be constant – `orr sl,sl,r7,lsl r8’ /tmp/cckBsykf.s:4864: Error: shift must be constant – `orr sl,sl,r7,lsl r8’ /tmp/cckBsykf.s:4893: Error: shift must be constant – `orr r7,r7,r8,lsl sl’ /tmp/cckBsykf.s:5105: Error: shift must be constant – `orr sl,sl,r5,lsl r7’ /tmp/cckBsykf.s:5155: Error: shift must be constant – `orr sl,sl,r5,lsl r7’ /tmp/cckBsykf.s:5179: Error: shift must be constant – `orr sl,sl,r7,lsl r8’ /tmp/cckBsykf.s:5326: Error: shift must be constant – `orr sl,sl,r3,lsl r5’ /tmp/cckBsykf.s:5350: Error: shift must be constant – `orr sl,sl,r3,lsl r4’ make[3]: * [layer3.lo] Error 1
make[3]: Leaving directory `/nfs/crux/efikamx/work/src/mpg123-1.13.3/src/libmpg123’ make[2]: * [all-recursive] Error 1
make[2]: Leaving directory `/nfs/crux/efikamx/work/src/mpg123-1.13.3/src’ make[1]: * [all] Error 2
make[1]: Leaving directory `/nfs/crux/efikamx/work/src/mpg123-1.13.3/src’ make: * [all-recursive] Error 1
======⇒ ERROR: Building ‘/nfs/crux/efikamx/pkg/mpg123#1.13.3-1.pkg.tar.gz’ failed.

======⇒ ERROR: Footprint mismatch found:
NEW -rw-r–r– root/root usr/lib/perl5/5.12/linux-thread-multi/ODBM_File.pm
NEW drwxr-xr-x root/root usr/lib/perl5/5.12/linux-thread-multi/auto/ODBM_File/
NEW -rwxr-xr-x root/root usr/lib/perl5/5.12/linux-thread-multi/auto/ODBM_File/ODBM_File
.so
NEW -rw-r–r– root/root usr/man/man3/ODBM_File.3pm.gz
======⇒ ERROR: Building ‘/usr/ports/pkg/perl#5.12.4-1.pkg.tar.xz’ failed.

Xorg input devices would not work for me using core-arm/udev (version 167) on a cubox platform.

I upgraded to version 182 and the input devices worked fine. I had to patch udev to replace calls to accept4() with accept(). Pkgfile (based on core/udev) and patch appear below:

######## Pkgfile ############
# Description: Userspace device management daemon
# URL: http://www.kernel.org/pub/linux/utils/kernel/hotplug/udev.html # Maintainer: CRUX System Team, core-ports at crux dot nu
# Depends on: kmod

name=udev version=182
release=1
source=(ftp://ftp.kernel.org/pub/linux/utils/kernel/hotplug/$name-$version.tar.x
z \

      pre-accept4-kernel.patch 81-crux.rules start_udev)

build() {

  cd udev-$version

  patch -p1 < $SRC/pre-accept4-kernel.patch
  ./configure --prefix=/usr \
              --sbindir=/sbin --bindir=/sbin \
              --sysconfdir=/etc \
              --with-rootlibdir=/lib \
              --with-rootprefix= \
              --libexecdir=/lib \
              --mandir=/usr/man \
              --disable-introspection \
              --disable-gudev \
              --disable-udev_acl \                                                                                                                                                                                                                         --disable-keymap \
              --enable-static \
              --enable-rule_generator                                                                                                                                                                                                      
  make
  make install DESTDIR=$PKG

  mkdir -p $PKG/lib/{firmware,udev/devices/{pts,shm}}
  mkdir -p $PKG/{lib,sbin,run}
  
  # Symlink to udevd
  ln -s ../lib/udev/udevd $PKG/sbin/udevd

  # Add CRUX items
  install -m 0755 $SRC/start_udev $PKG/sbin
  install -m 0644 $SRC/81-crux.rules $PKG/lib/udev/rules.d
                                                                                                                                                                                                                                               # Remove junk
  rm -r $PKG/usr/share/{gtk-,}doc

}

######### pre-accept4-kernel.patch ###########
diff -urN a/src/udev-ctrl.c b/src/udev-ctrl.c
— a/src/udev-ctrl.c 2011-10-09 17:10:32.000000000 -0600
+++ b/src/udev-ctrl.c 2011-10-25 15:11:09.000000000 -0600
@@ -15,6 +15,7 @@
#include <stddef.h>
#include <string.h>
#include <unistd.h>
+#include <fcntl.h>
#include <sys/types.h>
#include <sys/poll.h>
#include <sys/socket.h> @@ -182,6 +183,7 @@

       struct ucred ucred;
       socklen_t slen;
       const int on = 1;

+ int flgs;

       conn = calloc(1, sizeof(struct udev_ctrl_connection));
       if (conn == NULL)

@@ -189,13 +191,19 @@

       conn->refcount = 1;
       conn->uctrl = uctrl;

- conn→sock = accept4(uctrl→sock, NULL, NULL, SOCK_CLOEXEC|SOCK_NONBLOCK);
+ conn→sock = accept4(uctrl→sock, NULL, NULL, SOCK_CLOEXEC|SOCK_NONBLOCK);
+ conn→sock = accept(uctrl→sock, NULL, NULL);
if (conn→sock < 0) {
if (errno != EINTR)
err(uctrl→udev, "unable to receive ctrl connection: %m\n");
goto err;
}
+ Since we don't have accept4
+ flgs = fcntl(conn→sock, F_GETFL, NULL); + if(flgs >= 0) fcntl(conn→sock, F_SETFL, flgs | O_NONBLOCK);
+ fcntl(conn→sock, F_SETFD, FD_CLOEXEC); +

       /* check peer credential of connection */                                                                                                                                                                                                    slen = sizeof(ucred);
       if (getsockopt(conn->sock, SOL_SOCKET, SO_PEERCRED, &ucred, &slen) < 0) {

Here is a build of opt/gst-plugins-base (on the cubox) during which the assembler reports:

{standard input}: Assembler messages:
{standard input}:586: Error: immediate value out of range – `movt r2,-32640’ {standard input}:656: Error: immediate value out of range – `movt r2,-32640’

http://sprunge.us/CdeD

The cubox may have some pre-patched or pre-build gstreamer packages for hardware acceleration, but I have not yet investigated them.

libtool tries to relink during installation but failed due to host libraries

libtool: relink: arm-crux-linux-gnueabihf-g++ -fPIC -DPIC -shared -nostdlib /home/sepen/devel/crux-arm/toolchain/clfs/usr/lib/crti.o /home/sepen/devel/crux-arm/toolchain/crosstools/lib/gcc/arm-crux-linux-gnueabihf/4.8.3/crtbeginS.o .libs/dummy.o cxx/.libs/isfuns.o cxx/.libs/ismpf.o cxx/.libs/ismpq.o cxx/.libs/ismpz.o cxx/.libs/ismpznw.o cxx/.libs/limits.o cxx/.libs/osdoprnti.o cxx/.libs/osfuns.o cxx/.libs/osmpf.o cxx/.libs/osmpq.o cxx/.libs/osmpz.o -Wl,-rpath -Wl,/home/sepen/devel/crux-arm/toolchain/crosstools/lib/gcc/arm-crux-linux-gnueabihf/4.8.3/../../../../arm-crux-linux-gnueabihf/lib -L/home/sepen/devel/crux-arm/ports/crossrootfs/libgmp/work/pkg/usr/lib -L/usr/lib -lgmp -L/home/sepen/devel/crux-arm/toolchain/crosstools/lib/gcc/arm-crux-linux-gnueabihf/4.8.3 -L/home/sepen/devel/crux-arm/toolchain/crosstools/lib/gcc/arm-crux-linux-gnueabihf/4.8.3/../../../../arm-crux-linux-gnueabihf/lib -L/home/sepen/devel/crux-arm/toolchain/clfs/lib -L/home/sepen/devel/crux-arm/toolchain/clfs/usr/lib -L/home/sepen/devel/crux-arm/ports/crossrootfs/libgmp/work/pkg/home/sepen/devel/crux-arm/toolchain/crosstools/lib/gcc/arm-crux-linux-gnueabihf/4.8.3/../../../../arm-crux-linux-gnueabihf/lib -lstdc++ -lm -lc -lgcc_s /home/sepen/devel/crux-arm/toolchain/crosstools/lib/gcc/arm-crux-linux-gnueabihf/4.8.3/crtendS.o /home/sepen/devel/crux-arm/toolchain/clfs/usr/lib/crtn.o -O2 -mfloat-abi=hard -Wl,-soname -Wl,libgmpxx.so.4 -o .libs/libgmpxx.so.4.4.0
/usr/lib/libstdc++.so: file not recognized: File format not recognized
collect2: error: ld returned 1 exit status
libtool: install: error: relink `libgmpxx.la’ with the above command before installing it

There is a good explanation of the problem in this thread:

https://lists.debian.org/debian-devel/2011/02/msg00196.html (Fun with libtool and cross-builds)

We can reproduce that problem when building crossrootfs on a x86_64 host. Our final target is arm 32bits so we can’t link against non-32bits objects, and that is for what we didn’t detect issues when building crossrootfs on a x86 host.
I’m not happy with this situation.

We can create a new toolchain with hardfp and thumb support.

It can be created with the same triplet used in the toolchain with a minor change (arm-cruxhf-linux-gnueabi) but the point could be to use another one: arm-crux-linux-gnueabihf

There are problems with this triplet which should be fixed before we start adopting it.

There are some checks which can be done to verify hardfp/thumb support.

readelf -e test:
The elf header should change (not implemented atm)

readelf -A test:
Tag_THUMB_ISA_use: Thumb-2
Tag_ABI_HardFP_use: SP and DP
These told us that this toolchain has thumb support and it’s built with hardfp

size test:
we can compare file’s sizes and verify that the thumb one it’s smaller.

We need to take a look to vexpress config and make a cleanup.
It'd interesting to add forgotten ext4 support.

Try to cut down not directly related options to this device

This isn't a real task. You should close it and start opening some real tasks.