KVM: s390: consider epoch index on TOD clock syncs
authorDavid Hildenbrand <david@redhat.com>
Wed, 7 Feb 2018 11:46:45 +0000 (12:46 +0100)
committerChristian Borntraeger <borntraeger@de.ibm.com>
Tue, 20 Feb 2018 20:51:21 +0000 (20:51 +0000)
For now, we don't take care of over/underflows. Especially underflows
are critical:

Assume the epoch is currently 0 and we get a sync request for delta=1,
meaning the TOD is moved forward by 1 and we have to fix it up by
subtracting 1 from the epoch. Right now, this will leave the epoch
index untouched, resulting in epoch=-1, epoch_idx=0, which is wrong.

We have to take care of over and underflows, also for the VSIE case. So
let's factor out calculation into a separate function.

Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20180207114647.6220-5-david@redhat.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Fixes: 8fa1696ea781 ("KVM: s390: Multiple Epoch Facility support")
Cc: stable@vger.kernel.org
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
[use u8 for idx]

arch/s390/kvm/kvm-s390.c

index 5b7fe80cda56f8a47ae0b7d989ca9f3240f6c835..b07aa16dcf0659386e838787eea051705a433ff6 100644 (file)
@@ -179,6 +179,28 @@ int kvm_arch_hardware_enable(void)
 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
                              unsigned long end);
 
+static void kvm_clock_sync_scb(struct kvm_s390_sie_block *scb, u64 delta)
+{
+       u8 delta_idx = 0;
+
+       /*
+        * The TOD jumps by delta, we have to compensate this by adding
+        * -delta to the epoch.
+        */
+       delta = -delta;
+
+       /* sign-extension - we're adding to signed values below */
+       if ((s64)delta < 0)
+               delta_idx = -1;
+
+       scb->epoch += delta;
+       if (scb->ecd & ECD_MEF) {
+               scb->epdx += delta_idx;
+               if (scb->epoch < delta)
+                       scb->epdx += 1;
+       }
+}
+
 /*
  * This callback is executed during stop_machine(). All CPUs are therefore
  * temporarily stopped. In order not to change guest behavior, we have to
@@ -194,13 +216,17 @@ static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
        unsigned long long *delta = v;
 
        list_for_each_entry(kvm, &vm_list, vm_list) {
-               kvm->arch.epoch -= *delta;
                kvm_for_each_vcpu(i, vcpu, kvm) {
-                       vcpu->arch.sie_block->epoch -= *delta;
+                       kvm_clock_sync_scb(vcpu->arch.sie_block, *delta);
+                       if (i == 0) {
+                               kvm->arch.epoch = vcpu->arch.sie_block->epoch;
+                               kvm->arch.epdx = vcpu->arch.sie_block->epdx;
+                       }
                        if (vcpu->arch.cputm_enabled)
                                vcpu->arch.cputm_start += *delta;
                        if (vcpu->arch.vsie_block)
-                               vcpu->arch.vsie_block->epoch -= *delta;
+                               kvm_clock_sync_scb(vcpu->arch.vsie_block,
+                                                  *delta);
                }
        }
        return NOTIFY_OK;