DSM-G600, DNS-3xx and NSA-220 Hack Forum

HaydnH wrote:

Hi Bzhou,

I was just working on something that requires ffmpeg but I erased my custom compiled version a while ago so thought I'd use the optware version - however, one of the libraries required by it wasn't compiled with -fPIC so it won't work... oh well think I still have my log of the compilation somewhere...

Hi Haydn,

http://trac.nslu2-linux.org/optware/changeset/8201 should force -fPIC into CFLAGS on dns-323 now. Please test ffmpeg_0.svn20080409-2_arm.ipk when it gets in the feed.

Please edit http://www.nslu2-linux.org/wiki/Optware/Dns323 if -fPIC problem is resolved.

Last edited by bzhou (2008-04-16 00:25:11)

The DNS323 with the two most recent firmwares provides a kernel numbered 2.6.12.6-arm1

/ # uname -a
Linux nas1 2.6.12.6-arm1 #29 Thu Dec 27 09:59:48 CST 2007 armv5tejl unknown

The readme for the DLink tool chain states

vi kernel-headers.mk
        change ==> LINUX_SOURCE:=kernel-headers-2.4.21.tar.bz2 --> LINUX_SOURCE:=kernel-headers-2.6.9.1.tar.bz2

It looks like that's what I have set up in my optware setup (I'm sure I've seen "kernel-headers 2.6.9" scroll past at some point, but I can't remember where!)

Should we "fix" this to match the actual kernel version, or is the minor version irrelevant?

To try this, I assume I need to change sources/toolchain-dns323/kernel-headers.mk.patch:

--- make/kernel-headers.mk.orig    2006-05-17 02:10:48.000000000 +0000
+++ make/kernel-headers.mk    2007-01-10 20:21:09.000000000 +0000
@@ -43,8 +43,8 @@
         (cd $(LINUX_DIR)/include; ln -fs asm-$(ARCH)$(NOMMU) asm;) \
     fi
     touch $(LINUX_DIR)/include/linux/autoconf.h;
-    echo "#define UTS_RELEASE \"2.4.21\"" > $(LINUX_DIR)/include/linux/version.h;
-    echo "#define LINUX_VERSION_CODE 132117" >> $(LINUX_DIR)/include/linux/version.h;
+    echo "#define UTS_RELEASE \"2.6.6\"" > $(LINUX_DIR)/include/linux/version.h;
+    echo "#define LINUX_VERSION_CODE 132617" >> $(LINUX_DIR)/include/linux/version.h;
     echo "#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))" >> \
         $(LINUX_DIR)/include/linux/version.h;
     touch $(LINUX_DIR)/.configured

Something that I've noticed about that patch is that the new UTS_RELEASE does not match the KERNEL_VERSION
(in the above case 2^16 * 2 + 2^8 * 4 + 2^0 * 21 =  132117, but  2^16 * 2 + 2^8 * 6 + 2^0 * 6 = 132614, not 132617 as coded).

I think I should have  UTS_RELEASE 2.6.12 and LINUX_VERSION_CODE 132620. Do I need to make those changes anywhere else?

Or a waste of time?

Edit:  KERNEL_VERSION => LINUX_VERSION_CODE

Last edited by sjmac (2008-04-16 13:07:35)

sjmac, you can also take a look at http://trac.nslu2-linux.org/optware/bro … yno-x07.mk
In this case the toolchain was missing kernel header in the first place, and I hacked on a matching one.
But this issue probably is not very urgent in the short term. With older kernel header, there might be some feature missing (things like inotify), but mostly it will be ok.

sjmac, I got that patch from http://dns323.kood.org/howto:crosscompile build_dns323.tgz patches/kernel-headers.mk.
Could it be that at one time kernel version of older firmware was 2.6.6 or 2.6.9?
But I agree, if the UTS_RELEASE and LINUX_VERSION_CODE is inconsistent, that's certainly a bug.

No difference (I don't think disable-assembler is an option, but I changed the enable-assembler to disable-assembler, which will work as we expect I think)

Actually, looking at the build output from the svn build it looks like the build process spots that assembly isn't an option  anyway ...

...
nm: init-posix.o: no symbols
nm: lseek64.o: no symbols
nm: oldgetrlimit64.o: no symbols
nm: getutmpx.o: no symbols
checking if we should use assembler functions... no
checking if we should use RAID... no
checking If we should should enable LOAD DATA LOCAL by default... no
checking for arm-linux-getconf... no
checking for getconf... getconf
...

I also used -O0 -g, but forgot to turn off stripping. I'll try again this evening.

Edit/notes:
I'll take a look at the options used in this script too ...

http://216.239.59.104/search?q=cache:S8 … =firefox-a

linked from
http://dns323.kood.org/forum/t1737-MySQL-DNS-323.html (mysql builds and runs for fonz' fun_plug 0.4 and 0.5.)

Last edited by sjmac (2008-04-17 13:22:38)

STACK_DIRECTION , try sed and change it to 1 in config.h after configure

Not so simple :-(

I'm taking parts of Petter Larsson's build script in to mysql.mk one at a time, but at almost 13 minutes a go to compile mysql it's a bit tedious. I noticed that he used -static, and also seemed to edit configure.h in a place that it doesn't exist in the optware downloaded source.

BTW, I just typed 'make dirclean' instead of 'make mysql-dirclean' by mistake. What's the difference?

Edit: Patches to build this so there is no SIGILL on startup. (Some of the makefile changes will need isolating from the rest of the targets).

I used -Os (optimize for small size and performance) instead of -O2. Seems sensible for DNS323?

Last edited by sjmac (2008-04-18 13:26:52)

---

Attachments:
mysql-mk.patch, Size: 1,473 bytes, Downloads: 635

sjmac, glad you got it working!
In the mean time, I've updated mysql.mk to 4.1.22, and mysql5.mk to 5.0.45 (still needs some finishing touch, but basically working).
Can you re-base your patch to these?
We should give you svn write access when rwhitby is back, before that I can commit changes for you.

bzhou wrote:

Could it be that at one time kernel version of older firmware was 2.6.6 or 2.6.9?
But I agree, if the UTS_RELEASE and LINUX_VERSION_CODE is inconsistent, that's certainly a bug.

Yes, earlier firmwares used a 2.6.6 kernel according to the front page of the wiki http://dns323.kood.org/

I don't think its making much difference, but I've fixed the UTS_RELEASE to 2.6.9 in my tree, so that it matches the LINUX_VERSION_CODE

132617 decimal is 0x00020609 (hex, for version 2.6.9), which doesn't match 2.6.6.

See if zcat is compiled in busybox by

# /bin/busybox --help

If zcat is available in the PATH, try

# cd /mnt/HD_a2/ipkg/
# zcat ipkg-opt_0.99.163-9.arm.ipk | tar -xf - ./data.tar.gz
# zcat ./data.tar.gz | tar -xf -

instead of the two "tar -xzf" commands.

Last edited by bzhou (2008-04-24 02:02:06)

This is probably because fun_plug has gnu tar, while busybox tar is missing some feature. Once bootstrapped, you can also install optware tar, which is a very recent full version gnu tar.

For the make error, it is missing gettext as a dependent ipk. You can manually install that as a workaround. I will fix this when I have time.

The ocaml package in the feed has the known problem of hardcoding gcc path as the cross compiler. I don't know if it will work by  symlinking a native compiler to the expected place. The pure ocaml code will be compiled fine.

bzhou wrote:

The ocaml package in the feed has the known problem of hardcoding gcc path as the cross compiler. I don't know if it will work by  symlinking a native compiler to the expected place. The pure ocaml code will be compiled fine.

ocaml 3.10.2-2 should have this problem fixed.

bzhou wrote:

pulz wrote:

Anyone that would happend to have old xmlrpc package laying around? xmlrpc-c v1.11.00-2

I looked at the change between 1.11.00-2 and 1.11.00-3, it adapts the code to the libcurl 7.18.0 -> 7.18.1 change.
So you might want both xmlrpc-c 1.11.00-2 and libcurl 7.18.0. Unless someone saves backup of the feed, this needs subversion r8100 and rebuild.

Im trying to use it xmlrpc package with rtorrent and it isnt working.
I been reading over at the synologi forums and they say it crashed with the latest build -3 and some smart souls saved the -2 package, but sadly isn´t  working with the dns-323.
To bad, it would have been fun getting this to work.