On Tue, 5 Dec 2017 10:39:26 +0100 Christian Borntraeger <borntraeger@xxxxxxxxxx> wrote: > On 12/05/2017 10:13 AM, Cornelia Huck wrote: > > This reminds me of something I stumbled upon the other day: > > > > handle_ri() and handle_gs() (both implemented in priv.c) don't seem to > > have a check for PSTATE, yet they enable ri/gs before retrying the > > instruction. Is that correct? > > The guarded storage ops (e3 49 and e3 4d) are problem state. > Most of the ri instruction are as well, so problem state can enable RI > interpretion. > > We could do some optimization to only enable RI if the > instruction would enable in for the guest (e.g. an inspection of the > control block could leave RI disabled). On the other hand that would > require to implement these instruction in KVM, which I would like > to avoid. Right now we enable RI and re-drive the instruction. It's probably not worth it, I think.
