EN FR
EN FR


Section: Research Program

Research direction: developing drivers using Genes

We believe that weaknesses of previous methods for easing device driver development arise from an insufficient understanding of the range and scope of driver functionality, as required by real devices and OSes. We propose a new methodology for understanding device drivers, inspired by the biological field of genomics. Rather than focusing on the input/output behavior of a device, we take the radically new methodology of studying existing device driver code itself. On the one hand, this methodology makes it possible to identify the behaviors performed by real device drivers, whether to support the features of the device and the OS, or to improve properties such as safety or performance. On the other hand, this methodology makes it possible to capture the actual patterns of code used to implement these behaviors, raising the level of abstraction from individual operations to collections of operations implementing a single functionality, which we refer to as genes. Because the requirements of the device remain fixed, regardless of the OS, we expect to find genes with common behaviors across different OSes, even when those genes have a different internal structure. This leads to a view of a device driver as being constructed as a composition of genes, thus opening the door to new methodologies to address the problems faced by real driver developers. Among these, we have so far identified the problems of developing drivers, porting existing drivers to other OSes, backporting existing drivers to older OS versions, and long-term maintenance of the driver code.

Our short term goal is to “sequence” the complete set of genes for a set of related drivers. In the longer term, we plan to develop methodologies based on genes for aiding in driver development and maintenance. This work is currently financed by a grant from the Direction Générale de l'Armement (DGA) that supports the PhD of Peter Senna Tschudin. Valentin Rothberg's PhD is supported by an Inria Cordi-S grant.