GPU kernel and userspace border

Session information has not yet been published for this event.


One Line Summary

Building an efficient and maintenable API for allowing userspace to program GPU is one of the most important piece of the graphic stacks


The API provided by the kernel GPU driver is crucial for the graphic stack performance and maintenability. It could be a bottleneck or prove very hard to be adapted/improved over time.

Unlike the closed driver, open source graphic stack have to ensure computer security by preventing use of the GPU to compromise the computer security (over writting system memory, crashing the computer, …). Hence the open source driver can’t allow userspace to directly feed GPU command, unless GPU have capacity to enforce security (memory protection, context, privilegied context, …).

This talk will discuss alternative way to build the graphic stack in order to work around those issue. The aim being to have the most efficience and fastest graphic stack while also having strong security and maintainability.


memory management, GPU, drm, kms, security

Presentation Materials



  • Jerome Glisse

    Red Hat


    I have been working on the open source radeon driver for several years now. I am one of the author of radeon KMS and major contributor to TTM code too.

    I have also participated to the mesa OpenGL driver (from r3xx to lastest HD6xxx serie support).