xref: /openbmc/u-boot/doc/README.generic-board (revision 94db5120)
1#
2# (C) Copyright 2014 Google, Inc
3# Simon Glass <sjg@chromium.org>
4#
5# SPDX-License-Identifier:	GPL-2.0+
6#
7
8Background
9----------
10
11U-Boot traditionally had a board.c file for each architecture. This introduced
12quite a lot of duplication, with each architecture tending to do
13initialisation slightly differently. To address this, a new 'generic board
14init' feature was introduced in March 2013 (further motivation is
15provided in the cover letter below).
16
17All boards and architectures have moved to this as of mid 2016.
18
19
20What has changed?
21-----------------
22
23The main change is that the arch/<arch>/lib/board.c file is removed in
24favour of common/board_f.c (for pre-relocation init) and common/board_r.c
25(for post-relocation init).
26
27Related to this, the global_data and bd_t structures now have a core set of
28fields which are common to all architectures. Architecture-specific fields
29have been moved to separate structures.
30
31
32Further Background
33------------------
34
35The full text of the original generic board series is reproduced below.
36
37--8<-------------
38
39This series creates a generic board.c implementation which contains
40the essential functions of the major arch/xxx/lib/board.c files.
41
42What is the motivation for this change?
43
441. There is a lot of repeated code in the board.c files. Any change to
45things like setting up the baud rate requires a change in 10 separate
46places.
47
482. Since there are 10 separate files, adding a new feature which requires
49initialisation is painful since it must be independently added in 10
50places.
51
523. As time goes by the architectures naturally diverge since there is limited
53pressure to compare features or even CONFIG options against similar things
54in other board.c files.
55
564. New architectures must implement all the features all over again, and
57sometimes in subtle different ways. This places an unfair burden on getting
58a new architecture fully functional and running with U-Boot.
59
605. While it is a bit of a tricky change, I believe it is worthwhile and
61achievable. There is no requirement that all code be common, only that
62the code that is common should be located in common/board.c rather than
63arch/xxx/lib/board.c.
64
65All the functions of board_init_f() and board_init_r() are broken into
66separate function calls so that they can easily be included or excluded
67for a particular architecture. It also makes it easier to adopt Graeme's
68initcall proposal when it is ready.
69
70http://lists.denx.de/pipermail/u-boot/2012-January/114499.html
71
72This series removes the dependency on generic relocation. So relocation
73happens as one big chunk and is still completely arch-specific. See the
74relocation series for a proposed solution to this for ARM:
75
76http://lists.denx.de/pipermail/u-boot/2011-December/112928.html
77
78or Graeme's recent x86 series v2:
79
80http://lists.denx.de/pipermail/u-boot/2012-January/114467.html
81
82Instead of moving over a whole architecture, this series takes the approach
83of simply enabling generic board support for an architecture. It is then up
84to each board to opt in by defining CONFIG_SYS_GENERIC_BOARD in the board
85config file. If this is not done, then the code will be generated as
86before. This allows both sets of code to co-exist until we are comfortable
87with the generic approach, and enough boards run.
88
89ARM is a relatively large board.c file and one which I can test, therefore
90I think it is a good target for this series. On the other hand, x86 is
91relatively small and simple, but different enough that it introduces a
92few issues to be solved. So I have chosen both ARM and x86 for this series.
93After a suggestion from Wolfgang I have added PPC also. This is the
94largest and most feature-full board, so hopefully we have all bases
95covered in this RFC.
96
97A generic global_data structure is also required. This might upset a few
98people. Here is my basic reasoning: most fields are the same, all
99architectures include and need it, most global_data.h files already have
100#ifdefs to select fields for a particular SOC, so it is hard to
101see why architecures are different in this area. We can perhaps add a
102way to put architecture-specific fields into a separate header file, but
103for now I have judged that to be counter-productive.
104
105Similarly we need a generic bd_info structure, since generic code will
106be accessing it. I have done this in the same way as global_data and the
107same comments apply.
108
109There was dicussion on the list about passing gd_t around as a parameter
110to pre-relocation init functions. I think this makes sense, but it can
111be done as a separate change, and this series does not require it.
112
113While this series needs to stand on its own (as with the link script
114cleanup series and the generic relocation series) the goal is the
115unification of the board init code. So I hope we can address issues with
116this in mind, rather than focusing too narrowly on particular ARM, x86 or
117PPC issues.
118
119I have run-tested ARM on Tegra Seaboard only. To try it out, define
120CONFIG_SYS_GENERIC_BOARD in your board file and rebuild. Most likely on
121x86 and PPC at least it will hang, but if you are lucky it will print
122something first :-)
123
124I have run this though MAKEALL with CONFIG_SYS_GENERIC_BOARD on for all
125ARM, PPC and x86 boards. There are a few failures due to errors in
126the board config, which I have sent patches for. The main issue is
127just the difference between __bss_end and __bss_end__.
128
129Note: the first group of commits are required for this series to build,
130but could be separated out if required. I have included them here for
131convenience.
132
133------------->8--
134
135Simon Glass, sjg@chromium.org
136March 2014
137Updated after final removal, May 2016
138