On 17.02.20 15:24, David Hildenbrand wrote: > On 14.02.20 23:26, Christian Borntraeger wrote: >> From: Janosch Frank <frankja@xxxxxxxxxxxxx> >> >> STHYI data has to go through the bounce buffer. >> >> Signed-off-by: Janosch Frank <frankja@xxxxxxxxxxxxx> >> Reviewed-by: Thomas Huth <thuth@xxxxxxxxxx> >> Reviewed-by: Cornelia Huck <cohuck@xxxxxxxxxx> >> [borntraeger@xxxxxxxxxx: patch merging, splitting, fixing] >> Signed-off-by: Christian Borntraeger <borntraeger@xxxxxxxxxx> >> --- >> arch/s390/kvm/intercept.c | 15 ++++++++++----- >> 1 file changed, 10 insertions(+), 5 deletions(-) >> >> diff --git a/arch/s390/kvm/intercept.c b/arch/s390/kvm/intercept.c >> index 1e231058e4b3..cfabeecbb777 100644 >> --- a/arch/s390/kvm/intercept.c >> +++ b/arch/s390/kvm/intercept.c >> @@ -392,7 +392,7 @@ int handle_sthyi(struct kvm_vcpu *vcpu) >> goto out; >> } >> >> - if (addr & ~PAGE_MASK) >> + if (!kvm_s390_pv_is_protected(vcpu->kvm) && (addr & ~PAGE_MASK)) >> return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); >> >> sctns = (void *)get_zeroed_page(GFP_KERNEL); >> @@ -403,10 +403,15 @@ int handle_sthyi(struct kvm_vcpu *vcpu) >> >> out: >> if (!cc) { >> - r = write_guest(vcpu, addr, reg2, sctns, PAGE_SIZE); >> - if (r) { >> - free_page((unsigned long)sctns); >> - return kvm_s390_inject_prog_cond(vcpu, r); >> + if (kvm_s390_pv_is_protected(vcpu->kvm)) { > > I have the feeling that we might have to think about proper locking for > kvm_s390_pv_is_protected(). We have to make sure there cannot be any > races with user space. Smells like a new r/w lock maybe. I think we can keep that within kvm->lock (for the global changes) and vcpu->mutex. All the intercepts only happen within the vcpu run ioctl and that takes the vcpu->mutex. See my other patch (with mutex_lock(&vcpu->mutex) and mutex_unlock(&vcpu->mutex) around the create/destroy functions). lockdep_assert_help is happy and as long as the pv_handle reflects the state of the control block of that CPU we are good. I pushed out the patch/rfcs to pv_worktree on kvms390.git.