Posts Tagged ‘toolchain’

In the last couple of weeks, I decided it was time to port my Erlang framework, ELIoT, that I develop as part of my Ph.D. thesis, to several other devices than the ones on which I already ported it (the Raspberry Pi and the Carambola). So last week I ported it to Android, and while I still have to refine it so that it can be executed on non-rooted devices (i.e., you don’t need to be root to do it), I decided to give the Kindle a try. This first post describes my findings in exploring the device itself, in a later post I will talk about ELIoT itself.

I have a Kindle 4th generation non-touch, and this “non-touch” adjective is quite important, since it changes the operating system version (and probably the toolchains/hacks). The first step is to jailbreak it (beware of the consequences! this may at best break your warranty, at worse brick the device!), using the main guides it is quite easy and (seems) with little risks, from the point of view of losing all your settings and books; at the end of the procedure, you can access it through WiFi and SSH.

Given the fact that my main objective was to cross-compile stuff, I started trying to figure out how to obtain the toolchain, and the answer is, as usual, crosstool-ng. Given the fact that the Kindle runs Linux, the non-proprietary part of the OS (i.e., everything except their ebook applications) is open source and they allow us to download the source code; in this way, we know that the 4th gen runs the 2.6.31 kernel and 2.12 libc, so I started to specify those in the ct-ng configuration; for binutils I decided for the latest stable release available, and in my first try I compiled gcc 4.7 Linaro edition.

Some guides on the Web seemed to say that the floating point type was softfp, but since there was so little clue about the toolchains for this particular Kindle, I decided to give the hardfp a try. It didn’t work out, as expected, but it led to an interesting side note: if you want to know what kind of fp a distribution is, you can execute

readelf -A

on an existing binary (I tried on the wpa_supplicant binary): if it contains the line

Tag_ABI_VFP_args: VFP registers

then it means it is hardfp; otherwise it is softfp (I checked against a binary for the Raspberry Pi, which I know is the former).

So, I recompiled the toolchain with softfp and tried the usual hello world: it worked! But the joy was limited: on a more complex application some assembly errors emerged, and at this point I thought the reasons could be

  1. I was using the vanilla as kernel, but I didn’t know the subsubsubversion of the Kindle kernel and it may have been different
  2. maybe 4.7 was a too much advanced version of gcc to use.

So I changed the ct-ng configuration again and used the existing kernel sources downloaded from Amazon for my specific Kindle version, and changed the gcc version to 4.5.4, since /proc/version says:

Linux version 2.6.31-rt11-lab126 (jenkins-official@lucid-build02) (gcc version 4.5.3 20110406 (prerelease) (Linaro GCC 4.5-2011.04-0) ) #5 Sat Jan 12 20:39:09 PST 2013

this worked and I was able to compile and execute Erlang on the Kindle! Unfortunately, no screenshots here: access was from SSH, so all I can show you is:

[root@kindle erlangarm]# ./bin/erl -name erlang@ -setcookie abc
Eshell V5.9.3.1 (abort with ^G)
(erlang@> net_adm:ping(‘erlang@’).

copied directly from the terminal.

The final version of the ct-ng configuration is here, if anyone wants it; you have to adjust paths for the output directory and the kernel sources path (that you have to download from Amazon).

Read Full Post »