[PATCH 2/2] KVM: Convert read-only users of vm_list to RCU

[Jan Kiszka <jan.kiszka@siemens.com>]

Only for walking the list of VMs, we do not need to hold the preemption
disabling kvm_lock. Convert stat services, the cpufreq callback and
mmu_shrink to RCU. For the latter, special care is required to
synchronize its list_move_tail with kvm_destroy_vm.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
---
 arch/x86/kvm/mmu.c       |   14 +++++++++-----
 arch/x86/kvm/x86.c       |    4 ++--
 include/linux/kvm_host.h |    1 +
 virt/kvm/kvm_main.c      |   17 ++++++++++-------
 4 files changed, 22 insertions(+), 14 deletions(-)

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index b6a9963..e9d0ed8 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -3587,9 +3587,9 @@ static int mmu_shrink(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask)
 	if (nr_to_scan == 0)
 		goto out;
 
-	raw_spin_lock(&kvm_lock);
+	rcu_read_lock();
 
-	list_for_each_entry(kvm, &vm_list, vm_list) {
+	list_for_each_entry_rcu(kvm, &vm_list, vm_list) {
 		int idx, freed_pages;
 		LIST_HEAD(invalid_list);
 
@@ -3607,10 +3607,14 @@ static int mmu_shrink(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask)
 		spin_unlock(&kvm->mmu_lock);
 		srcu_read_unlock(&kvm->srcu, idx);
 	}
-	if (kvm_freed)
-		list_move_tail(&kvm_freed->vm_list, &vm_list);
+	if (kvm_freed) {
+		raw_spin_lock(&kvm_lock);
+		if (!kvm->deleted)
+			list_move_tail(&kvm_freed->vm_list, &vm_list);
+		raw_spin_unlock(&kvm_lock);
+	}
 
-	raw_spin_unlock(&kvm_lock);
+	rcu_read_unlock();
 
 out:
 	return percpu_counter_read_positive(&kvm_total_used_mmu_pages);
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index ae6e75b..01fa1ea 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -4578,7 +4578,7 @@ static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long va
 
 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
 
-	raw_spin_lock(&kvm_lock);
+	rcu_read_lock();
 	list_for_each_entry(kvm, &vm_list, vm_list) {
 		kvm_for_each_vcpu(i, vcpu, kvm) {
 			if (vcpu->cpu != freq->cpu)
@@ -4588,7 +4588,7 @@ static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long va
 				send_ipi = 1;
 		}
 	}
-	raw_spin_unlock(&kvm_lock);
+	rcu_read_unlock();
 
 	if (freq->old < freq->new && send_ipi) {
 		/*
diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
index c8dee22..9074cac 100644
--- a/include/linux/kvm_host.h
+++ b/include/linux/kvm_host.h
@@ -234,6 +234,7 @@ struct kvm {
 #endif
 	struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
 	atomic_t online_vcpus;
+	bool deleted;
 	struct list_head vm_list;
 	struct mutex lock;
 	struct kvm_io_bus *buses[KVM_NR_BUSES];
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 7b70c67..b5a05b5 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -546,9 +546,13 @@ static void kvm_destroy_vm(struct kvm *kvm)
 	struct mm_struct *mm = kvm->mm;
 
 	kvm_arch_sync_events(kvm);
+
 	raw_spin_lock(&kvm_lock);
 	list_del(&kvm->vm_list);
+	kvm->deleted = true;
 	raw_spin_unlock(&kvm_lock);
+	synchronize_rcu();
+
 	kvm_free_irq_routing(kvm);
 	for (i = 0; i < KVM_NR_BUSES; i++)
 		kvm_io_bus_destroy(kvm->buses[i]);
@@ -2347,10 +2351,10 @@ static int vm_stat_get(void *_offset, u64 *val)
 	struct kvm *kvm;
 
 	*val = 0;
-	raw_spin_lock(&kvm_lock);
-	list_for_each_entry(kvm, &vm_list, vm_list)
+	rcu_read_lock();
+	list_for_each_entry_rcu(kvm, &vm_list, vm_list)
 		*val += *(u32 *)((void *)kvm + offset);
-	raw_spin_unlock(&kvm_lock);
+	rcu_read_unlock();
 	return 0;
 }
 
@@ -2364,12 +2368,11 @@ static int vcpu_stat_get(void *_offset, u64 *val)
 	int i;
 
 	*val = 0;
-	raw_spin_lock(&kvm_lock);
-	list_for_each_entry(kvm, &vm_list, vm_list)
+	rcu_read_lock();
+	list_for_each_entry_rcu(kvm, &vm_list, vm_list)
 		kvm_for_each_vcpu(i, vcpu, kvm)
 			*val += *(u32 *)((void *)vcpu + offset);
-
-	raw_spin_unlock(&kvm_lock);
+	rcu_read_unlock();
 	return 0;
 }
 
-- 
1.7.1