xref: /openbmc/docs/kernel-development.md (revision f4febd00)
1# OpenBMC kernel development
2
3The OpenBMC project maintains a kernel tree for use by the project. The tree's
4general development policy is that code must be upstream first. This is a strong
5requirement, not a hard requirement, and exceptions will be made on a
6case-by-case basis. If in doubt, start a discussion on the mailing list.
7
8The OpenBMC kernel tree is hosted at https://github.com/openbmc/linux and
9contains the set of patches that we carry. Ideally there would be no patches
10carried, as everything should be upstream.
11
12Your code will make it into the OpenBMC tree in these ways, from most to least
13desirable:
14
151. When the OpenBMC kernel moves to a new upstream release
162. By backporting upstream commits from a newer kernel version to the OpenBMC
17   kernel
183. Patches included in the OpenBMC tree temporarily
19
20## TL;DR
21
22If you require a patch added to the tree, follow these steps:
23
241. Submit your patch upstream. It doesn't need to be upstream, but it should be
25   on it's way
262. Use
27   `git format-patch --subject-prefix="PATCH linux dev-4.7" --to=openbmc@lists.ozlabs.org --to=joel@jms.id.au`
28   to create a formatted patch
29
30## Developing a new driver
31
32When developing a new driver, your goal is to have the code accepted upstream.
33The first step should be to check that there is no existing driver for the
34hardware you wish to support. Check the OpenBMC `-dev` tree, check upstream, and
35if you do not find support there ask on the mailing list.
36
37Once you are sure a driver needs to be written, you should develop and test the
38driver, before sending it upstream to the relevant maintainers. You should feel
39welcome to cc the OpenBMC list when sending upstream, so other kernel developers
40can provide input where appropriate. Be sure to follow the (upstream development
41process)[https://www.kernel.org/doc/Documentation/process/submitting-patches.rst].
42
43In the past patches underwent 'pre-review' on the OpenBMC mailing list. While
44this is useful for developers who are not familiar with writing kernel code, it
45has lead to confusion about the upstreaming process, so now we do all of our
46development in the community.
47
48Once the driver has been accepted upstream, send the good news to the OpenBMC
49list with a reference to the upstream tree. This may be Linus' tree, or it might
50be one of the subsystem maintainers. From there the OpenBMC kernel team can
51decide how best to include your code in the OpenBMC tree.
52
53### Exceptions
54
55There are cases where waiting for upstream acceptance will delay the bring-up of
56a new system. This should be avoided through careful planning and early
57development of the features upstream, but where this has not happened we can
58chose to carry the patches in the OpenBMC tree while the upstream development is
59ongoing.
60
61Another exception to the upstream first rule is where patches are modifying
62files that are not upstream. This currently includes the aspeed board file
63`arch/arm/mach-aspeed/aspeed.c`, and the device tree source files `dts`. The
64board file should go away when we get drivers written for all of the
65functionality; for now it contains some hacks relating to LPC and early init.
66
67If you find yourself adding to `arch/arm/mach-aspeed/aspeed.c`, first send an
68email to the OpenBMC list to get the opinion of the kernel developers. Patches
69to `aspeed.c` will be treated with some prejudice as the file will be removed
70once we have drivers for all of the Aspeed peripherals.
71
72## Getting existing code in the tree
73
74The OpenBMC kernel is currently based on the 4.7 series. If there is upstream
75code you would like backported, send it to the list. Be sure to include the
76upstream commit SHA in the commit message.
77
78## Testing
79
80When modifying the tree we currently test on the following platforms:
81
82- Palmetto, an OpenPower Power8 box containing an ast2400 with NCSI networking
83- ast2500-evb, the Aspeed dev board with two PHYs
84- Witherspoon, an OpenPower Power9 box containing an ast2500 with NCSI
85  networking
86- qemu-plametto and qemu-romulus
87
88Before submitting patches it is recommended you boot test on at least the Qemu
89platforms, and whatever hardware you have available.
90
91# Tips and Tricks
92
93Some general hints for kernel development
94
95## Out-of-tree builds
96
97You can build a kernel out of the yocto environment, by using the initramfs
98(from a pre-existing yocto build) directly:
99
100```
101make ARCH=arm \
102    O=obj \
103    CROSS_COMPILE=arm-linux-gnueabihf- \
104    CONFIG_INITRAMFS_SOURCE=/path/tp/obmc-phosphor-image-palmetto.cpio.gz
105```
106
107(adjust `O` and `CROSS_COMPILE` parameters as appropriate).
108
109You'll need to use `aspeed_g4_defconfig` or `aspeed_g5_defconfig` as your base
110kernel configuration.
111
112The cpio can be found in the following yocto output directory:
113
114```
115 build/tmp/deploy/images/palmetto/
116```
117
118## Building a uImage
119
120To build a uImage (for example, to netboot):
121
122```
123# build a zImage using the obmc rootfs
124make ARCH=arm \
125    O=obj \
126    CROSS_COMPILE=arm-linux-gnueabihf- \
127    CONFIG_INITRAMFS_SOURCE=/path/tp/obmc-phosphor-image-palmetto.cpio.gz
128
129# create a combined zImage + DTB image
130cat obj/arch/arm/boot/zImage \
131    obj/arch/arm/boot/dts/aspeed-bmc-opp-palmetto.dtb \
132        > obj/aspeed-zimage
133
134# create a uImage
135./scripts/mkuboot.sh -A arm -O linux -C none -T kernel \
136    -a 0x40008000 -e 0x40008000 -n $USER-`date +%Y%m%d%H%M` \
137    -d obj/aspeed-zimage obj/uImage
138```
139
140Note that some systems may have upgraded to a FIT-based u-boot, where the old
141uImage format is no longer accepted.
142