KVM: arm64: Move hyp state to hyp_vmemmap

Tracking the hypervisor's ownership state into struct hyp_page has
several benefits, including allowing far more efficient lookups (no
page-table walk needed) and de-corelating the state from the presence
of a mapping. This will later allow to map pages into EL2 stage-1 less
proactively which is generally a good thing for security. And in the
future this will help with tracking the state of pages mapped into the
hypervisor's private range without requiring an alias into the 'linear
map' range.

Reviewed-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: Quentin Perret <qperret@google.com>
Link: https://lore.kernel.org/r/20250416152648.2982950-6-qperret@google.com
Signed-off-by: Marc Zyngier <maz@kernel.org>

Author: Quentin Perret

arch/arm64/kvm/hyp/include/nvhe/memory.h

@@ -16,22 +16,22 @@
  *       accessible by another component;
  *   10: The page is accessible but not owned by the component;
  * The storage of this state depends on the component: either in the
- * hyp_vmemmap for the host state or in PTE software bits for the hypervisor
- * and guests.
+ * hyp_vmemmap for the host and hyp states or in PTE software bits for guests.
  */
 enum pkvm_page_state {
 	PKVM_PAGE_OWNED			= 0ULL,
 	PKVM_PAGE_SHARED_OWNED		= BIT(0),
 	PKVM_PAGE_SHARED_BORROWED	= BIT(1),
 
 	/*
-	 * 'Meta-states' are not stored directly in PTE SW bits for hyp and
-	 * guest states, but inferred from the context (e.g. invalid PTE
-	 * entries). For the host, meta-states are stored directly in the
+	 * 'Meta-states' are not stored directly in PTE SW bits for guest
+	 * states, but inferred from the context (e.g. invalid PTE entries).
+	 * For the host and hyp, meta-states are stored directly in the
 	 * struct hyp_page.
 	 */
 	PKVM_NOPAGE			= BIT(0) | BIT(1),
 };
+#define PKVM_PAGE_STATE_MASK		(BIT(0) | BIT(1))
 
 #define PKVM_PAGE_STATE_PROT_MASK	(KVM_PGTABLE_PROT_SW0 | KVM_PGTABLE_PROT_SW1)
 static inline enum kvm_pgtable_prot pkvm_mkstate(enum kvm_pgtable_prot prot,
@@ -52,7 +52,14 @@ struct hyp_page {
 	u8 order;
 
 	/* Host state. Guarded by the host stage-2 lock. */
-	unsigned __host_state : 8;
+	unsigned __host_state : 4;
+
+	/*
+	 * Complement of the hyp state. Guarded by the hyp stage-1 lock. We use
+	 * the complement so that the initial 0 in __hyp_state_comp (due to the
+	 * entire vmemmap starting off zeroed) encodes PKVM_NOPAGE.
+	 */
+	unsigned __hyp_state_comp : 4;
 
 	u32 host_share_guest_count;
 };
@@ -99,6 +106,16 @@ static inline void set_host_state(phys_addr_t phys, enum pkvm_page_state state)
 	hyp_phys_to_page(phys)->__host_state = state;
 }
 
+static inline enum pkvm_page_state get_hyp_state(phys_addr_t phys)
+{
+	return hyp_phys_to_page(phys)->__hyp_state_comp ^ PKVM_PAGE_STATE_MASK;
+}
+
+static inline void set_hyp_state(phys_addr_t phys, enum pkvm_page_state state)
+{
+	hyp_phys_to_page(phys)->__hyp_state_comp = state ^ PKVM_PAGE_STATE_MASK;
+}
+
 /*
  * Refcounting for 'struct hyp_page'.
  * hyp_pool::lock must be held if atomic access to the refcount is required.

arch/arm64/kvm/hyp/nvhe/mem_protect.c

@@ -649,24 +649,24 @@ static int __host_set_page_state_range(u64 addr, u64 size,
 	return 0;
 }
 
-static enum pkvm_page_state hyp_get_page_state(kvm_pte_t pte, u64 addr)
+static void __hyp_set_page_state_range(phys_addr_t phys, u64 size, enum pkvm_page_state state)
 {
-	if (!kvm_pte_valid(pte))
-		return PKVM_NOPAGE;
+	phys_addr_t end = phys + size;
 
-	return pkvm_getstate(kvm_pgtable_hyp_pte_prot(pte));
+	for (; phys < end; phys += PAGE_SIZE)
+		set_hyp_state(phys, state);
 }
 
-static int __hyp_check_page_state_range(u64 addr, u64 size,
-					enum pkvm_page_state state)
+static int __hyp_check_page_state_range(phys_addr_t phys, u64 size, enum pkvm_page_state state)
 {
-	struct check_walk_data d = {
-		.desired	= state,
-		.get_page_state	= hyp_get_page_state,
-	};
+	phys_addr_t end = phys + size;
+
+	for (; phys < end; phys += PAGE_SIZE) {
+		if (get_hyp_state(phys) != state)
+			return -EPERM;
+	}
 
-	hyp_assert_lock_held(&pkvm_pgd_lock);
-	return check_page_state_range(&pkvm_pgtable, addr, size, &d);
+	return 0;
 }
 
 static enum pkvm_page_state guest_get_page_state(kvm_pte_t pte, u64 addr)
@@ -694,7 +694,6 @@ int __pkvm_host_share_hyp(u64 pfn)
 {
 	u64 phys = hyp_pfn_to_phys(pfn);
 	void *virt = __hyp_va(phys);
-	enum kvm_pgtable_prot prot;
 	u64 size = PAGE_SIZE;
 	int ret;
 
@@ -705,13 +704,13 @@ int __pkvm_host_share_hyp(u64 pfn)
 	if (ret)
 		goto unlock;
 	if (IS_ENABLED(CONFIG_NVHE_EL2_DEBUG)) {
-		ret = __hyp_check_page_state_range((u64)virt, size, PKVM_NOPAGE);
+		ret = __hyp_check_page_state_range(phys, size, PKVM_NOPAGE);
 		if (ret)
 			goto unlock;
 	}
 
-	prot = pkvm_mkstate(PAGE_HYP, PKVM_PAGE_SHARED_BORROWED);
-	WARN_ON(pkvm_create_mappings_locked(virt, virt + size, prot));
+	__hyp_set_page_state_range(phys, size, PKVM_PAGE_SHARED_BORROWED);
+	WARN_ON(pkvm_create_mappings_locked(virt, virt + size, PAGE_HYP));
 	WARN_ON(__host_set_page_state_range(phys, size, PKVM_PAGE_SHARED_OWNED));
 
 unlock:
@@ -734,14 +733,15 @@ int __pkvm_host_unshare_hyp(u64 pfn)
 	ret = __host_check_page_state_range(phys, size, PKVM_PAGE_SHARED_OWNED);
 	if (ret)
 		goto unlock;
-	ret = __hyp_check_page_state_range(virt, size, PKVM_PAGE_SHARED_BORROWED);
+	ret = __hyp_check_page_state_range(phys, size, PKVM_PAGE_SHARED_BORROWED);
 	if (ret)
 		goto unlock;
 	if (hyp_page_count((void *)virt)) {
 		ret = -EBUSY;
 		goto unlock;
 	}
 
+	__hyp_set_page_state_range(phys, size, PKVM_NOPAGE);
 	WARN_ON(kvm_pgtable_hyp_unmap(&pkvm_pgtable, virt, size) != size);
 	WARN_ON(__host_set_page_state_range(phys, size, PKVM_PAGE_OWNED));
 
@@ -757,7 +757,6 @@ int __pkvm_host_donate_hyp(u64 pfn, u64 nr_pages)
 	u64 phys = hyp_pfn_to_phys(pfn);
 	u64 size = PAGE_SIZE * nr_pages;
 	void *virt = __hyp_va(phys);
-	enum kvm_pgtable_prot prot;
 	int ret;
 
 	host_lock_component();
@@ -767,13 +766,13 @@ int __pkvm_host_donate_hyp(u64 pfn, u64 nr_pages)
 	if (ret)
 		goto unlock;
 	if (IS_ENABLED(CONFIG_NVHE_EL2_DEBUG)) {
-		ret = __hyp_check_page_state_range((u64)virt, size, PKVM_NOPAGE);
+		ret = __hyp_check_page_state_range(phys, size, PKVM_NOPAGE);
 		if (ret)
 			goto unlock;
 	}
 
-	prot = pkvm_mkstate(PAGE_HYP, PKVM_PAGE_OWNED);
-	WARN_ON(pkvm_create_mappings_locked(virt, virt + size, prot));
+	__hyp_set_page_state_range(phys, size, PKVM_PAGE_OWNED);
+	WARN_ON(pkvm_create_mappings_locked(virt, virt + size, PAGE_HYP));
 	WARN_ON(host_stage2_set_owner_locked(phys, size, PKVM_ID_HYP));
 
 unlock:
@@ -793,7 +792,7 @@ int __pkvm_hyp_donate_host(u64 pfn, u64 nr_pages)
 	host_lock_component();
 	hyp_lock_component();
 
-	ret = __hyp_check_page_state_range(virt, size, PKVM_PAGE_OWNED);
+	ret = __hyp_check_page_state_range(phys, size, PKVM_PAGE_OWNED);
 	if (ret)
 		goto unlock;
 	if (IS_ENABLED(CONFIG_NVHE_EL2_DEBUG)) {
@@ -802,6 +801,7 @@ int __pkvm_hyp_donate_host(u64 pfn, u64 nr_pages)
 			goto unlock;
 	}
 
+	__hyp_set_page_state_range(phys, size, PKVM_NOPAGE);
 	WARN_ON(kvm_pgtable_hyp_unmap(&pkvm_pgtable, virt, size) != size);
 	WARN_ON(host_stage2_set_owner_locked(phys, size, PKVM_ID_HOST));
 
@@ -816,19 +816,18 @@ int hyp_pin_shared_mem(void *from, void *to)
 {
 	u64 cur, start = ALIGN_DOWN((u64)from, PAGE_SIZE);
 	u64 end = PAGE_ALIGN((u64)to);
+	u64 phys = __hyp_pa(start);
 	u64 size = end - start;
 	int ret;
 
 	host_lock_component();
 	hyp_lock_component();
 
-	ret = __host_check_page_state_range(__hyp_pa(start), size,
-					    PKVM_PAGE_SHARED_OWNED);
+	ret = __host_check_page_state_range(phys, size, PKVM_PAGE_SHARED_OWNED);
 	if (ret)
 		goto unlock;
 
-	ret = __hyp_check_page_state_range(start, size,
-					   PKVM_PAGE_SHARED_BORROWED);
+	ret = __hyp_check_page_state_range(phys, size, PKVM_PAGE_SHARED_BORROWED);
 	if (ret)
 		goto unlock;
 

arch/arm64/kvm/hyp/nvhe/setup.c

@@ -194,16 +194,20 @@ static int fix_host_ownership_walker(const struct kvm_pgtable_visit_ctx *ctx,
 
 	/*
 	 * Adjust the host stage-2 mappings to match the ownership attributes
-	 * configured in the hypervisor stage-1.
+	 * configured in the hypervisor stage-1, and make sure to propagate them
+	 * to the hyp_vmemmap state.
 	 */
 	state = pkvm_getstate(kvm_pgtable_hyp_pte_prot(ctx->old));
 	switch (state) {
 	case PKVM_PAGE_OWNED:
+		set_hyp_state(phys, PKVM_PAGE_OWNED);
 		return host_stage2_set_owner_locked(phys, PAGE_SIZE, PKVM_ID_HYP);
 	case PKVM_PAGE_SHARED_OWNED:
+		set_hyp_state(phys, PKVM_PAGE_SHARED_OWNED);
 		set_host_state(phys, PKVM_PAGE_SHARED_BORROWED);
 		break;
 	case PKVM_PAGE_SHARED_BORROWED:
+		set_hyp_state(phys, PKVM_PAGE_SHARED_BORROWED);
 		set_host_state(phys, PKVM_PAGE_SHARED_OWNED);
 		break;
 	default: