On Mon, 7 Dec 2020 at 20:56, Peter Maydell <peter.mayd...@linaro.org> wrote: > These are fairly obscure, and it wouldn't surprise me if > other target archs had picked a different separation of > concerns between userspace and the kernel such that userspace > didn't need to manually deliver an exception.
...looking at the target/i386 code, it looks like the approach taken on that architecture is that userspace just asks the kernel to take the exception using the KVM_SET_VCPU_EVENTS ioctl. (For Arm we only use that ioctl to handle SError, because the information that an SError is pending is a piece of state that's external to the CPU.) thanks -- PMM
