Linux – Why modify the linux kernel instead of creating a kernel module

driverskernelkernel-moduleslinux

I have a development board which has an older version of Linux installed on it. The vendor supplies an image for the device with a heavily modified linux kernel, some loadable kernel modules, and some example software.

I would like to install a newer version of the linux kernel on the device, but the vendor has no support for this, as their modified linux kernel is based off of an older kernel version.

What I don't understand, is why start hacking away at the linux kernel, when you can make the kernel compatible with the device it is running on by writing drivers as kernel modules. It could be easily recompiled for any kernel version without problems. This way, if the vendor only supports a certain kernel version, you are "stuck" 🙁

But there must be some reason I am missing, because I see many projects use this approach of grabbing some version of the kernel, and heavily modifying it to fit their board. What I would be interested in, is:

Why modify the linux kernel instead of creating a kernel module?
What can be done if I need to run a newer kernel, but I get no support from the vendor (Device drivers should work on newer versions of the kernel…)

Best Answer

This question has a lot of assumptions in it.

Here are some reasons.

The kernel interface is not stable so a module for one version may not compile for a different version.

The kernel may not expose a required facility.

The kernel may expose a required facility but not in a way that is acceptable, for example requiring the module to have a particular license.

The people writing the code found it quicker to write the code this way.

As to your options if you need a newer kernel.

find someone else who has already ported the code
port it yourself
pay someone else to port it (may not need money, beer, flattery and curiosity may work).

Related Solutions

Detect If Custom Kernel Was Built with Module Support

If you have a /proc filesystem, the file /proc/modules exists if and only if the kernel if compiled with module support. If the file exists but is empty, your kernel supports modules but none are loaded at the moment. If the file doesn't exist, your kernel cannot load any module.

It's technically possible to have loadable module support without /proc. You can check for the presence of the init_module and delete_module system calls in the kernel binary. This may not be easy if you only have a compressed binary (e.g. vmlinuz or uImage). See How do I uncompress vmlinuz to vmlinux? for vmlinuz. Once you've managed to decompress the bulk of the kernel, search for the string sys_init_module.

Note that if modules are supported, you'll need additional files to compile your own modules anyway: kernel headers. These are C header files (*.h), some of which are generated when the kernel is compiled (so you can't just take them from the kernel source). See What does a kernel source tree contain? Is this related to Linux kernel headers?

Linux – Why do user space apps need kernel headers

Because those programs are build to use things defined in the kernel headers:

busybox-1.22.1]$ egrep -RHn '^#include <linux'
modutils/modutils-24.c:194:#include <linux/elf-em.h>
include/fix_u32.h:17:#include <linux/types.h>
libbb/loop.c:11:#include <linux/version.h>
console-tools/openvt.c:23:#include <linux/vt.h>
console-tools/kbd_mode.c:23:#include <linux/kd.h>
console-tools/showkey.c:19:#include <linux/kd.h>
util-linux/blockdev.c:36:#include <linux/fs.h>
util-linux/mkfs_ext2.c:50:#include <linux/fs.h>
util-linux/mkfs_vfat.c:28:#include <linux/hdreg.h> /* HDIO_GETGEO */
util-linux/mkfs_vfat.c:29:#include <linux/fd.h>    /* FDGETPRM */
....

For each specific tool, you'd need to read the source of the tool and the relevant kernel header to figure out exactly what.

You can see a few things are commented to make it easy.

For example, mkfs_vfat includes linux/fd.h to get FDGETPRM:

$ egrep -RHn FDGETPRM util-linux/mkfs_vfat.c
util-linux/mkfs_vfat.c:29:#include <linux/fd.h>    /* FDGETPRM */
util-linux/mkfs_vfat.c:351:         int not_floppy = ioctl(dev, FDGETPRM, &param);

You could probably remove the relevant #include and watch for compiler errors to make it easier, you'll get warnings that some things are not defined. Those things likely come from the kernel headers.

Best Answer

Related Solutions

Detect If Custom Kernel Was Built with Module Support

Linux – Why do user space apps need kernel headers

Related Question