KVM: arm/arm64: vgic-new: Add userland access to VGIC dist registers
authorAndre Przywara <andre.przywara@arm.com>
Tue, 1 Dec 2015 22:36:37 +0000 (22:36 +0000)
committerChristoffer Dall <christoffer.dall@linaro.org>
Fri, 20 May 2016 13:40:03 +0000 (15:40 +0200)
Userland may want to save and restore the state of the in-kernel VGIC,
so we provide the code which takes a userland request and translate
that into calls to our MMIO framework.

From Christoffer:
When accessing the VGIC state from userspace we really don't want a VCPU
to be messing with the state at the same time, and the API specifies
that we should return -EBUSY if any VCPUs are running.
Check and prevent VCPUs from running by grabbing their mutexes, one by
one, and error out if we fail.
(Note: This could potentially be simplified to just do a simple check
and see if any VCPUs are running, and return -EBUSY then, without
enforcing the locking throughout the duration of the uaccess, if we
think that taking/releasing all these mutexes for every single GIC
register access is too heavyweight.)

Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
virt/kvm/arm/vgic/vgic-kvm-device.c

index 7862128..9ee27cb 100644 (file)
@@ -238,7 +238,60 @@ static int vgic_attr_regs_access(struct kvm_device *dev,
                                 struct kvm_device_attr *attr,
                                 u32 *reg, bool is_write)
 {
-       return -ENXIO;
+       gpa_t addr;
+       int cpuid, ret, c;
+       struct kvm_vcpu *vcpu, *tmp_vcpu;
+       int vcpu_lock_idx = -1;
+
+       cpuid = (attr->attr & KVM_DEV_ARM_VGIC_CPUID_MASK) >>
+                KVM_DEV_ARM_VGIC_CPUID_SHIFT;
+       vcpu = kvm_get_vcpu(dev->kvm, cpuid);
+       addr = attr->attr & KVM_DEV_ARM_VGIC_OFFSET_MASK;
+
+       mutex_lock(&dev->kvm->lock);
+
+       ret = vgic_init(dev->kvm);
+       if (ret)
+               goto out;
+
+       if (cpuid >= atomic_read(&dev->kvm->online_vcpus)) {
+               ret = -EINVAL;
+               goto out;
+       }
+
+       /*
+        * Any time a vcpu is run, vcpu_load is called which tries to grab the
+        * vcpu->mutex.  By grabbing the vcpu->mutex of all VCPUs we ensure
+        * that no other VCPUs are run and fiddle with the vgic state while we
+        * access it.
+        */
+       ret = -EBUSY;
+       kvm_for_each_vcpu(c, tmp_vcpu, dev->kvm) {
+               if (!mutex_trylock(&tmp_vcpu->mutex))
+                       goto out;
+               vcpu_lock_idx = c;
+       }
+
+       switch (attr->group) {
+       case KVM_DEV_ARM_VGIC_GRP_CPU_REGS:
+               ret = -EINVAL;
+               break;
+       case KVM_DEV_ARM_VGIC_GRP_DIST_REGS:
+               ret = vgic_v2_dist_uaccess(vcpu, is_write, addr, reg);
+               break;
+       default:
+               ret = -EINVAL;
+               break;
+       }
+
+out:
+       for (; vcpu_lock_idx >= 0; vcpu_lock_idx--) {
+               tmp_vcpu = kvm_get_vcpu(dev->kvm, vcpu_lock_idx);
+               mutex_unlock(&tmp_vcpu->mutex);
+       }
+
+       mutex_unlock(&dev->kvm->lock);
+       return ret;
 }
 
 /* V2 ops */