It is important to handle the exit code on the same CPU as the guest,
specially as we're accessing resources that are per-CPU (caches,
for exemple).
To achieve this, make the section that encompassing both
__kvm_vcpu_run() and handle_exit().
user_mem_abort() can sleep though (as it calls gfp_to_pfn()), so
preemption has to be reenabled at that stage.
Reported-by: Will Deacon <will.deacon at arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier at arm.com>
---
arch/arm/kvm/arm.c | 2 ++
arch/arm/kvm/mmu.c | 2 ++
2 files changed, 4 insertions(+), 0 deletions(-)
diff --git a/arch/arm/kvm/arm.c b/arch/arm/kvm/arm.c
index 13681a1..b96462b 100644
--- a/arch/arm/kvm/arm.c
+++ b/arch/arm/kvm/arm.c
@@ -522,6 +522,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
update_vttbr(vcpu->kvm);
+ preempt_disable();
local_irq_disable();
kvm_guest_enter();
vcpu->mode = IN_GUEST_MODE;
@@ -540,6 +541,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
trace_kvm_exit(vcpu->arch.regs.pc);
ret = handle_exit(vcpu, run, ret);
+ preempt_enable();
if (ret) {
kvm_err("Error in handle_exit\n");
break;
diff --git a/arch/arm/kvm/mmu.c b/arch/arm/kvm/mmu.c
index 992d39a..ae38c21 100644
--- a/arch/arm/kvm/mmu.c
+++ b/arch/arm/kvm/mmu.c
@@ -419,7 +419,9 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
pfn_t pfn;
int ret;
+ preempt_enable();
pfn = gfn_to_pfn(vcpu->kvm, gfn);
+ preempt_disable();
if (is_error_pfn(pfn)) {
put_page(pfn_to_page(pfn));
--
1.7.3.4
