Increasing Automation in the Backporting of Linux Drivers Using Coccinelle Luis R. Rodriguez Julia Lawall Rutgers University/SUSE Labs Sorbonne Universites/Inria/UPMC/LIP6´
[email protected] [email protected] [email protected],
[email protected] Abstract—Software is continually evolving, to fix bugs and to a kernel upgrade. Upgrading a kernel may also require add new features. Industry users, however, often value stability, experience to understand what features to enable, disable, or and thus may not be able to update their code base to the tune to meet existing deployment criteria. In the worst case, latest versions. This raises the need to selectively backport new some systems may rely on components that have not yet been features to older software versions. Traditionally, backporting has merged into the mainline Linux kernel, potentially making it been done by cluttering the backported code with preprocessor impossible to upgrade the kernel without cooperation from the directives, to replace behaviors that are unsupported in an earlier version by appropriate workarounds. This approach however component vendor or a slew of partners that need to collaborate involves writing a lot of error-prone backporting code, and results on developing a new productized image for a system. As an in implementations that are hard to read and maintain. We example, development for 802.11n AR9003 chipset support consider this issue in the context of the Linux kernel, for which on the mainline ath9k device driver started on March 20, older versions are in wide use. We present a new backporting 2010 with an early version of the silicon, at which point the strategy that relies on the use of a backporting compatability most recent major release of the Linux kernel was v2.6.32.