[PATCH v3 8/8] KVM: x86/mmu: Lazily allocate memslot rmaps

From: Ben Gardon
Date: Thu May 06 2021 - 14:43:41 EST


If the TDP MMU is in use, wait to allocate the rmaps until the shadow
MMU is actually used. (i.e. a nested VM is launched.) This saves memory
equal to 0.2% of guest memory in cases where the TDP MMU is used and
there are no nested guests involved.

Signed-off-by: Ben Gardon <bgardon@xxxxxxxxxx>
---
arch/x86/include/asm/kvm_host.h | 1 +
arch/x86/kvm/mmu/mmu.c | 14 ++++++++++---
arch/x86/kvm/mmu/tdp_mmu.c | 6 ++++--
arch/x86/kvm/mmu/tdp_mmu.h | 4 ++--
arch/x86/kvm/x86.c | 37 ++++++++++++++++++++++++++++++++-
5 files changed, 54 insertions(+), 8 deletions(-)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 00065f9bbc5e..7b8e1532fb55 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -1860,5 +1860,6 @@ static inline int kvm_cpu_get_apicid(int mps_cpu)
int kvm_cpu_dirty_log_size(void);

inline bool kvm_memslots_have_rmaps(struct kvm *kvm);
+int alloc_all_memslots_rmaps(struct kvm *kvm);

#endif /* _ASM_X86_KVM_HOST_H */
diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
index 48067c572c02..e3a3b65829c5 100644
--- a/arch/x86/kvm/mmu/mmu.c
+++ b/arch/x86/kvm/mmu/mmu.c
@@ -3306,6 +3306,10 @@ static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu)
}
}

+ r = alloc_all_memslots_rmaps(vcpu->kvm);
+ if (r)
+ return r;
+
write_lock(&vcpu->kvm->mmu_lock);
r = make_mmu_pages_available(vcpu);
if (r < 0)
@@ -5494,9 +5498,13 @@ void kvm_mmu_init_vm(struct kvm *kvm)
{
struct kvm_page_track_notifier_node *node = &kvm->arch.mmu_sp_tracker;

- kvm_mmu_init_tdp_mmu(kvm);
-
- kvm->arch.memslots_have_rmaps = true;
+ if (!kvm_mmu_init_tdp_mmu(kvm))
+ /*
+ * No smp_load/store wrappers needed here as we are in
+ * VM init and there cannot be any memslots / other threads
+ * accessing this struct kvm yet.
+ */
+ kvm->arch.memslots_have_rmaps = true;

node->track_write = kvm_mmu_pte_write;
node->track_flush_slot = kvm_mmu_invalidate_zap_pages_in_memslot;
diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
index 83cbdbe5de5a..5342aca2c8e0 100644
--- a/arch/x86/kvm/mmu/tdp_mmu.c
+++ b/arch/x86/kvm/mmu/tdp_mmu.c
@@ -14,10 +14,10 @@ static bool __read_mostly tdp_mmu_enabled = false;
module_param_named(tdp_mmu, tdp_mmu_enabled, bool, 0644);

/* Initializes the TDP MMU for the VM, if enabled. */
-void kvm_mmu_init_tdp_mmu(struct kvm *kvm)
+bool kvm_mmu_init_tdp_mmu(struct kvm *kvm)
{
if (!tdp_enabled || !READ_ONCE(tdp_mmu_enabled))
- return;
+ return false;

/* This should not be changed for the lifetime of the VM. */
kvm->arch.tdp_mmu_enabled = true;
@@ -25,6 +25,8 @@ void kvm_mmu_init_tdp_mmu(struct kvm *kvm)
INIT_LIST_HEAD(&kvm->arch.tdp_mmu_roots);
spin_lock_init(&kvm->arch.tdp_mmu_pages_lock);
INIT_LIST_HEAD(&kvm->arch.tdp_mmu_pages);
+
+ return true;
}

static __always_inline void kvm_lockdep_assert_mmu_lock_held(struct kvm *kvm,
diff --git a/arch/x86/kvm/mmu/tdp_mmu.h b/arch/x86/kvm/mmu/tdp_mmu.h
index 5fdf63090451..b046ab5137a1 100644
--- a/arch/x86/kvm/mmu/tdp_mmu.h
+++ b/arch/x86/kvm/mmu/tdp_mmu.h
@@ -80,12 +80,12 @@ int kvm_tdp_mmu_get_walk(struct kvm_vcpu *vcpu, u64 addr, u64 *sptes,
int *root_level);

#ifdef CONFIG_X86_64
-void kvm_mmu_init_tdp_mmu(struct kvm *kvm);
+bool kvm_mmu_init_tdp_mmu(struct kvm *kvm);
void kvm_mmu_uninit_tdp_mmu(struct kvm *kvm);
static inline bool is_tdp_mmu_enabled(struct kvm *kvm) { return kvm->arch.tdp_mmu_enabled; }
static inline bool is_tdp_mmu_page(struct kvm_mmu_page *sp) { return sp->tdp_mmu_page; }
#else
-static inline void kvm_mmu_init_tdp_mmu(struct kvm *kvm) {}
+static inline bool kvm_mmu_init_tdp_mmu(struct kvm *kvm) { return false; }
static inline void kvm_mmu_uninit_tdp_mmu(struct kvm *kvm) {}
static inline bool is_tdp_mmu_enabled(struct kvm *kvm) { return false; }
static inline bool is_tdp_mmu_page(struct kvm_mmu_page *sp) { return false; }
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 1098ab73a704..95e74fb9fc20 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -10868,9 +10868,44 @@ static int alloc_memslot_rmap(struct kvm_memory_slot *slot,
return -ENOMEM;
}

+int alloc_all_memslots_rmaps(struct kvm *kvm)
+{
+ struct kvm_memslots *slots;
+ struct kvm_memory_slot *slot;
+ int r = 0;
+ int i;
+
+ if (kvm_memslots_have_rmaps(kvm))
+ return 0;
+
+ mutex_lock(&kvm->slots_arch_lock);
+ for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
+ slots = __kvm_memslots(kvm, i);
+ kvm_for_each_memslot(slot, slots) {
+ r = alloc_memslot_rmap(slot, slot->npages);
+ if (r) {
+ mutex_unlock(&kvm->slots_arch_lock);
+ return r;
+ }
+ }
+ }
+
+ /*
+ * memslots_have_rmaps is set and read in different lock contexts,
+ * so protect it with smp_load/store.
+ */
+ smp_store_release(&kvm->arch.memslots_have_rmaps, true);
+ mutex_unlock(&kvm->slots_arch_lock);
+ return 0;
+}
+
bool kvm_memslots_have_rmaps(struct kvm *kvm)
{
- return kvm->arch.memslots_have_rmaps;
+ /*
+ * memslots_have_rmaps is set and read in different lock contexts,
+ * so protect it with smp_load/store.
+ */
+ return smp_load_acquire(&kvm->arch.memslots_have_rmaps);
}

static int kvm_alloc_memslot_metadata(struct kvm *kvm,
--
2.31.1.607.g51e8a6a459-goog