KVM: arm/arm64: Introduce kvm_arch_vcpu_run_pid_change
authorChristoffer Dall <christoffer.dall@linaro.org>
Fri, 23 Feb 2018 16:23:57 +0000 (17:23 +0100)
committerMarc Zyngier <marc.zyngier@arm.com>
Fri, 25 May 2018 11:27:54 +0000 (12:27 +0100)
KVM/ARM differs from other architectures in having to maintain an
additional virtual address space from that of the host and the
guest, because we split the execution of KVM across both EL1 and
EL2.

This results in a need to explicitly map data structures into EL2
(hyp) which are accessed from the hyp code.  As we are about to be
more clever with our FPSIMD handling on arm64, which stores data in
the task struct and uses thread_info flags, we will have to map
parts of the currently executing task struct into the EL2 virtual
address space.

However, we don't want to do this on every KVM_RUN, because it is a
fairly expensive operation to walk the page tables, and the common
execution mode is to map a single thread to a VCPU.  By introducing
a hook that architectures can select with
HAVE_KVM_VCPU_RUN_PID_CHANGE, we do not introduce overhead for
other architectures, but have a simple way to only map the data we
need when required for arm64.

This patch introduces the framework only, and wires it up in the
arm/arm64 KVM common code.

No functional change.

Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
include/linux/kvm_host.h
virt/kvm/Kconfig
virt/kvm/kvm_main.c

index 6930c63126c78a9ef665b5b5653a60a8773b4d4c..4268ace60bf12a23782aab8b71479e6529e73ddd 100644 (file)
@@ -1276,4 +1276,13 @@ static inline long kvm_arch_vcpu_async_ioctl(struct file *filp,
 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
                unsigned long start, unsigned long end);
 
+#ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE
+int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu);
+#else
+static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
+{
+       return 0;
+}
+#endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */
+
 #endif
index cca7e065a075d8eebb33097ec4a46f3e5bd2c3a3..72143cfaf6ec39404dad5f72a8cf08c5e5fefc7e 100644 (file)
@@ -54,3 +54,6 @@ config HAVE_KVM_IRQ_BYPASS
 
 config HAVE_KVM_VCPU_ASYNC_IOCTL
        bool
+
+config HAVE_KVM_VCPU_RUN_PID_CHANGE
+       bool
index c7b2e927f69903c4eb8bb571282d7a5b7ca5ca27..c32e2407713d83fcbcb282daa40ffce764b65cf8 100644 (file)
@@ -2550,8 +2550,13 @@ static long kvm_vcpu_ioctl(struct file *filp,
                oldpid = rcu_access_pointer(vcpu->pid);
                if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) {
                        /* The thread running this VCPU changed. */
-                       struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
+                       struct pid *newpid;
 
+                       r = kvm_arch_vcpu_run_pid_change(vcpu);
+                       if (r)
+                               break;
+
+                       newpid = get_task_pid(current, PIDTYPE_PID);
                        rcu_assign_pointer(vcpu->pid, newpid);
                        if (oldpid)
                                synchronize_rcu();