FreeCalypso > hg > tcs211-l1-reconst

comparison README @ 290:95f308ba55c7

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
README for the almost fully reconstructed state
author Mychaela Falconia <falcon@freecalypso.org>
date Mon, 20 Mar 2017 00:37:25 +0000
parents 53d8cb470dc7
children ed3e588600c1
comparison
equal deleted inserted replaced
289:f2f7f4dff6d7 290:95f308ba55c7
1 The purpose of this Hg tree (derived from leo2moko-debug) is to facilitate an 1 This Hg repository contains a version of TI's TCS211 firmware based on our sole
2 attempt at reconstructing an L1 source that would match the binary object 2 available TCS211 semi-src with the L1 component almost fully deblobbed.
3 version we got with our TCS211 semi-src delivery from Sotovik. 3 Specifically, we have deblobbed all of L1 with the exception of GPRS-specific
4 l1p_*, macs and p_tpudr12 modules - all other L1 modules are recompiled from
5 reconstructed C sources, producing objects that work in the place of the
6 original blobs - sometimes matching bit for bit, in other cases differing in
7 the bits, but matching in logic. See STATUS for a detailed analysis of the
8 reconstruction matches and differences.
4 9
5 The idea is to deblob TCS211 L1 one module or small group of modules at a time. 10 Building the firmware in this tree requires a trick because of the not-yet-
6 Take individual C modules from the LoCosto L1 source and get them to compile 11 deblobbed L1_GPRS modules - see README.old for the details. This tree only
7 in the TCS211 environment with TCS211 versions of all include files, producing 12 supports the original gtamodem target; see FreeCalypso Magnetite for other
8 new from-source objects that would serve as candidate replacements for the ones 13 target support.
9 we got as blobs. An absolutely essential aspect of this approach is
10 incrementality: we should be able to deblob just a few L1 objects (replace them
11 with build-from-source versions) while keeping the original blobs for all other
12 parts of L1 we haven't reached yet, and regression-test at each step to ensure
13 that the firmware still works.
14 14
15 TI's build system for this fw does not provide very fine granularity as far as 15 Choice of version for reconstruction
16 compiling each individual module from source or taking it as a delivered blob. 16 ====================================
17 Instead it operates on the granularity of libs, or sometimes (as in the case of
18 L1) groups of libs. In the case of L1, this tree has the build system set up
19 to rebuild l1_ext.lib and l1_int.lib from source, which necessarily includes
20 _everything_ that goes into these libs. We are nowhere close to having a
21 working reconstructed source for every module in these two libs, thus we have
22 dummies in the place of most of them to appease BuSyB. Therefore, if you
23 simply run ./winebuild.sh to build this fw version, the link step will fail.
24 17
25 Instead this fw has to be built in a somewhat convoluted way: 18 For some of the L1 objects (those packaged in l1_ext.lib and l1_int.lib) we
19 have two versions: the original from 20070608 (the same date as the rest of our
20 TCS211 semi-src) and an updated version of the l1_{ext,int}.lib set from
21 20080421. To the best of our knowledge, this 20080421 update was provided by
22 TI-Taiwan while engaged in a wild goose chase after the infamous bug #1024,
23 which later turned out to be a pure hardware problem, not involving the firmware
24 at all.
26 25
27 1. Run ./winebuild.sh initially to compile everything leading up to the link 26 The differences between the two versions are limited to just 3 objects:
28 step, and let it fail. 27 l1_cmplx.obj and l1p_cmpl.obj in l1_ext.lib, and l1_ctl.obj in l1_int.lib.
29 28
30 2. Go into g23m/__out__/gsm*/obj/l1_ext, and copy in original blob objects for 29 When we were reconstructing l1_cmplx.c and l1_ctl.c from the available LoCosto
31 all those modules which have not been source-reconstructed yet. Do likewise 30 source, the reconstruction of the original 20070608 version came more easily
32 for l1_int. 31 than the 20080421 update, and our reconstruction of these two modules matches
32 the 20070608 objects bit for bit. It thus appears that whatever changes TI's
33 customer support group made in that 20080421 version were not on TI's internal
34 mainline.
33 35
34 [Update: the copyblobs.sh script does this job now.] 36 Because the 20080421 L1 libs update was made for no other purpose except an
37 attempt to debug or fix what was erroneously believed to be a firmware problem
38 but turned out to be hw instead, we have every reason to believe that it does
39 not improve anything and is ultimately unnecessary. And because the original
40 20070608 version was easier to source-reconstruct than the other, we are
41 adopting the original 20070608 version and our reconstruction of its C source
42 form as the canonical version of TCS211 L1 for FreeCalypso.
35 43
36 3. Run ./winebuild.sh again. As long as the dummy C files have not been 44 Note that because we have not source-reconstructed the GPRS-specific l1p_*
37 touched and the copied-in objects have newer timestamps, the dummies won't 45 modules yet, we haven't looked at the l1p_cmpl.obj differences yet. However,
38 be recompiled and the copied-in objects won't get blown away. TI's make 46 when building a complete fw image with GPRS included using our partially
39 will rerun the ar packaging step to produce l1_ext.lib and l1_int.lib from 47 reconstructed L1, one should use the 20070608 version of the l1p_cmpl.obj blob
40 the mix of objects (some compiled from source, others copied-in blobs) in 48 for consistency, like we do in FreeCalypso Magnetite.
41 the obj directories, and then the final link should succeed.
42
43 The development process is as follows:
44
45 1. Choose an individual L1 C module to work on.
46
47 2. Copy the source for the chosen module from the LoCosto source in the place
48 of the dummy in this tree.
49
50 3. Run ./winebuild.sh in the g23m directory and examine the output in
51 report.txt. If the imported C module fails to compile, massage it until it
52 passes compilation.
53
54 4. Once compilation passes, look at the generated COFF object in
55 g23m/__out__/gsm*/obj/l1_{ext,int}. Disassemble it with the tiobjd tool
56 from freecalypso-reveng, and compare against the disassembly of the
57 original ("official") binary blob version we got. Massage the reconstructed
58 source until it compiles into object code that matches the original.
59
60 [Update: use the scripts under g23m/objdiff to facilitate the disassembly
61 and diffing.]