> On Sep 19, 2018, at 10:00 AM, Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote: > >> On 19/09/2018 18:57, Sebastian Andrzej Siewior wrote: >>> On 2018-09-19 07:55:51 [+0200], Paolo Bonzini wrote: >>> A kthread can do use_mm/unuse_mm. >> >> indeed. The FPU struct for the kernel thread isn't valid / does not >> contain the expected PKRU value. So loading the pkru value from the >> struct FPU does not work as expected. We could set it to 0 for a kernel >> thread so we don't end up with a random value. >> If we want to get this usecase working then we would have to move pkru >> value from FPU to mm_struct and consider it in use_mm(). Do we want >> this? > > As a start, I think keeping it in the FPU struct but loading it > unconditionally will work. kthreads will not obey PKU but it will be > better already. > > I honestly don't know if PKRU should be per-mm, I don't know mm very > well despite my brilliant observation above. :) > > It must be per thread. I don’t think it’s possible to have sane semantics per mm. I also think that use_mm should set PKRU to the same value that signal handlers use. If we use 0, it’s a recipe for accidental PK bypass.
