view README @ 171:a8efe285e7fa

dti.lib compiles
author Mychaela Falconia <falcon@freecalypso.org>
date Tue, 11 Oct 2016 07:44:31 +0000
parents 9fb9f896bd77
children 75ea63a3fce5
line wrap: on
line source

FreeCalypso Magnetite firmware project
======================================

This source tree contains yet another firmware offering created under the
FreeCalypso umbrella.  The key qualities of this firmware offering are:

Negatives:

  * Builds with TI's proprietary TMS470 compiler and thus requires Wine;
  * Some of the fw components are still in the form of binary blobs -
    but see below on the deblobbing progress.

Positives:

  * Can be built to run on Motorola C139, Openmoko GTA02 and Pirelli DP-L10
    targets, as well as the to-be-built FCDEV3B, all from the same source tree;

  * Works as solidly as the TCS211 "golden" reference from TI, on all of the
    supported targets - deep sleep works, voice calls work in all codec modes
    including AMR, the DSP dynamic download mechanism does its magic, the call
    audio passes reliably in both directions.

The present FC Magnetite firmware is built on the principle of starting with
the known working TCS211 code base, without any major restructuring, and making
small incremental evolutionary changes, testing at every step to ensure that
nothing breaks.  It is the direct opposite of the "rebuild from the ground up"
approach taken with our previous Citrine firmware (aka "gsm-fw"), the approach
that produced very disappointing results.

Functionality
=============

TI's GSM mobile station firmware architecture supports two ways in which the
GSM device may be controlled: via AT commands from an external host and/or via
a local UI on devices with LCD & keypad hardware.  (I said "and/or" because the
two mechanisms can coexist.)  At the present time, however, only the AT command
mode of operation is supported in FreeCalypso Magnetite.  The effect of this
limitation is that if you run this fw on a Motorola C139 or a Pirelli DP-L10,
the phone's LCD will stay dark, the buttons won't do anything, and the
phone-turned-modem can only be operated via AT commands sent via FreeCalypso
host utility fc-shell.

The demo/prototype phone UI code in TI's reference fw delivery is written for
a 176x220 pixel 16-bit color LCD (such an LCD was used on TI's official
development platform called D-Sample), whereas our available Mot C139 and
Pirelli targets have significantly smaller LCDs: 96x64 and 128x128 pixels,
respectively.  Prior to the D-Sample, TI had used 84x48 pixel black&white
(1 bit per pixel) LCDs, and this old C-Sample LCD code is still there, albeit
bitrotten.  In late 2015 this author made a very dirty hack to resurrect TI's
old C-Sample UI code and get it to display on the C139 phone LCD (84x48 is a
proper subset of 96x64) - you can find this hack in the tcs211-c139 Hg tree.

The upshot of this LCD situation is that porting TI's phone UI code to run on
Mot C139 or Pirelli DP-L10 will require major rework of the affected parts of
the firmware.  While I would like to do it eventually, I am not willing to do
it right now - I would like to get this code running in its pristine state on
its native 176x220 pix LCD *before* I hack the holy **** out of it for the
C139/Pirelli port.  I do have a real TI-made D-Sample kit with the right LCD,
but unfortunately the main board has an older version of the core chipset for
which we lack the necessary fw support, hence it doesn't help.  Therefore, we
will need to build our own Calypso board with a 176x220 pix 16-bit color LCD,
get the UI-enabled GSM firmware running on that board, and only then proceed
with C139 and/or Pirelli ports if such are still desired.

For now, modem or pseudo-modem operation with control via AT commands is all
we have.

Build system
============

Even though FC Magnetite is essentially unchanged TCS211 code base and builds
using TI's proprietary TMS470 compiler under Wine, the build system is entirely
new.  TI's TCS211 build system, called BuSyB, works by way of a Java tool
generating a customized makefile for each desired build configuration, based on
lots of magic contained in a big repository of XML files.  There are a bunch of
Perl scripts involved as well.  The Java tool that does the heavy lifting exists
only as compiled Java bytecode sans source, and the surrounding Perl scripts
aren't very understandable either.  And the whole thing thoroughly assumes a
Windows environment (drive letters, backslashes, case-insensitive file system)
throughout.  As a result, when working with TCS211 fw with its original build
system, we had to treat these BuSyB-generated makefiles almost as being blobs in
themselves: regenerating a makefile from XML magic required major effort, there
were some bugs in the makefile generation which we couldn't fix and thus we had
to edit the makefiles manually after each regeneration - it was an utter mess,
and absolutely not an acceptable way to build a forward-looking, community-
serving project.

In FC Magnetite I have recreated the relevant parts of the TCS211 build system,
using Bourne shell magic instead of Java and XML.  Just like TI's BuSyB, ours
is a makefile generation system: in order to compile the firmware in a
particular desired configuration, you run a shell script to select the config
you would like.  This shell script will create a dedicated build directory tree
to fully contain this build, and populate it with generated Makefiles and some
other bits - then you go into the just-created build directory and run make
there.  The source and build trees are thus cleanly separated.  See
doc/Compiling for detailed instructions.

Another key difference from our previous TCS211-based firmware offerings is that
even though we still have to run TI's compiler binaries under Wine, the Wine
invokation has been moved from the top (root) of the build process to the
bottom leaves.  With our previous TCS211-based works you would run Wine at the
top, and then the entire build process would proceed in the Windows environment,
using Windows versions of make and other nonsense.  Not so in FC Magnetite:
in this firmware project all shell scripts, Makefiles, Perl scripts and other
build system accessories run at the native Unix level, and Wine is only invoked
at the lowest level by individual tool wrappers: for example, TI's compiler
binary cl470.exe is encapsulated in a Unix shell script called cl470 that
invokes Wine to run the Windows binary, presenting the illusion of a native
Unix tool to all upper levels.

As yet another defenestration measure, all source files are checked into this
tree with Unix line endings.

Blob status
===========

A long-term FreeCalypso goal is to have our phone/modem firmware rebuild fully
from source without any blobs, but this goal has not been achieved yet.  While
we do have what *seems* to be a suitable replacement source (or feasible ability
to reconstruct such) for every piece of TCS211 fw that came in binary-only form,
actually making this replacement without breaking functionality is a very
non-trivial endeavor.  Our previous attempt to rebuild the firmware from the
ground up, using source pieces lifted from different available leaks and
building with gcc so that no TMS470 COFF blobs could be used, produced very
disappointing results.

Instead the new FreeCalypso firmware approach implemented in FC Magnetite is to
approach the blob-free goal incrementally.  The new Magnetite build system is
specifically designed to enable the transition from the use of blobs to
recompilation from source to be made with very fine granularity, down to the
level of individual object modules within libs if necessary.  We tackle one
binary-only component at a time, either reconstructing the missing source from
disassembly or adapting the source from a different version as works best in
each individual case, and we make a test build of the firmware using the
reconstructed or fitted component instead of the original blob.  If the firmware
still works (doesn't break), we make this deblobbing transition permanent and
move on to the next component.

As of this writing, most of Layer 1 and a few housekeeping parts of the fw have
already been deblobbed, i.e., are now recompiled from source.  The G23M protocol
stack is our next deblobbing target - we have a newer version of it in full
source form, and we are hoping to be able to retrofit this newer G23M version
into the TCS211 environment in order to replace the binary blob version we are
using currently.

Further reading
===============

For various instructions and notes specific to this FreeCalypso Magnetite
firmware, look in the doc directory.  For more information about the overall
FreeCalypso project and our hardware building aspirations, go to our website:

https://www.freecalypso.org/