VFIO and Management of Errors

This proposal has been accepted as a session.

*
Accepted Session
60 Minute BoF
Scheduled: Thursday, October 16, 2014 from 5:30 – 6:30pm in Room 28

One Line Summary

Devices assigned to user space through vfio can trigger a number of error conditions. This discussion looks at approaches to managing errors.

Abstract

This discussion focuses on approaches to managing errors triggered by devices bound to vfio. In-band device errors are fairly straightforward and in most cases would be sent to user space by the standard eventfd mechnaism.

Out-of-band errors would include things like IOMMU faults (e.g. an access violation) are more difficult. The IOMMU layer in Linux supports a fault handler, but VFIO does not use this currently. How should an access violation by a vfio-bound device be handled? SIGSEGV to the app? Do KVM virtual machines have any different requirements than other user space apps?

How would IOMMU faults work in an architecture supporting two stage IOMMU address translations where a fault was caused by a vfio-assigned device in guest user space?

This discussion focuses on these types of questions.

Tags

VFIO, IOMMU

Speakers

Leave a private comment to organizers about this proposal