On 26/07/12 18:31, Peter Maydell wrote: > On 26 July 2012 18:11, Marc Zyngier <marc.zyngier at arm.com> wrote: >> Instead of relying on SMC calls to setup HVBAR, introduce a new >> HVC-based API. This of course requires a hypervisor stub to be >> installed already. At the moment, this is implemented as part of the >> boot wrapper, but will be moved to the kernel itself when it actually >> understands being booted in HYP mode. >> >> The kernel now uses "hvc #0xff", with r0 containing the pointer to >> the new vectors. > > This will still go bang if you try to boot a KVM-enabled kernel > on a non-virtualization capable system (eg as the qemu guest), > right? Indeed. Once we can boot the kernel in HYP mode (working on it at the moment), there will be a flag indicating if it is available or not. At that point, it becomes fairly trivial not to initialize KVM if we cannot reach HYP mode. This patch is just a way to decouple the two issues, and to get comments on the API itself. M. -- Jazz is not dead. It just smells funny...
