Merge branch kvm-arm64/mmu/reduce-vmemmap-overhead into kvmarm-master/next

Host stage-2 optimisations from Quentin Perret

* kvm-arm64/mmu/reduce-vmemmap-overhead:
  KVM: arm64: Use less bits for hyp_page refcount
  KVM: arm64: Use less bits for hyp_page order
  KVM: arm64: Remove hyp_pool pointer from struct hyp_page
  KVM: arm64: Unify MMIO and mem host stage-2 pools
  KVM: arm64: Remove list_head from hyp_page
  KVM: arm64: Use refcount at hyp to check page availability
  KVM: arm64: Move hyp_pool locking out of refcount helpers

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

@@ -7,7 +7,7 @@
 #include <nvhe/memory.h>
 #include <nvhe/spinlock.h>
 
-#define HYP_NO_ORDER	UINT_MAX
+#define HYP_NO_ORDER	USHRT_MAX
 
 struct hyp_pool {
 	/*
@@ -19,48 +19,13 @@ struct hyp_pool {
 	struct list_head free_area[MAX_ORDER];
 	phys_addr_t range_start;
 	phys_addr_t range_end;
-	unsigned int max_order;
+	unsigned short max_order;
 };
 
-static inline void hyp_page_ref_inc(struct hyp_page *p)
-{
-	struct hyp_pool *pool = hyp_page_to_pool(p);
-
-	hyp_spin_lock(&pool->lock);
-	p->refcount++;
-	hyp_spin_unlock(&pool->lock);
-}
-
-static inline int hyp_page_ref_dec_and_test(struct hyp_page *p)
-{
-	struct hyp_pool *pool = hyp_page_to_pool(p);
-	int ret;
-
-	hyp_spin_lock(&pool->lock);
-	p->refcount--;
-	ret = (p->refcount == 0);
-	hyp_spin_unlock(&pool->lock);
-
-	return ret;
-}
-
-static inline void hyp_set_page_refcounted(struct hyp_page *p)
-{
-	struct hyp_pool *pool = hyp_page_to_pool(p);
-
-	hyp_spin_lock(&pool->lock);
-	if (p->refcount) {
-		hyp_spin_unlock(&pool->lock);
-		BUG();
-	}
-	p->refcount = 1;
-	hyp_spin_unlock(&pool->lock);
-}
-
 /* Allocation */
-void *hyp_alloc_pages(struct hyp_pool *pool, unsigned int order);
-void hyp_get_page(void *addr);
-void hyp_put_page(void *addr);
+void *hyp_alloc_pages(struct hyp_pool *pool, unsigned short order);
+void hyp_get_page(struct hyp_pool *pool, void *addr);
+void hyp_put_page(struct hyp_pool *pool, void *addr);
 
 /* Used pages cannot be freed */
 int hyp_pool_init(struct hyp_pool *pool, u64 pfn, unsigned int nr_pages,

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

@@ -23,7 +23,7 @@ extern struct host_kvm host_kvm;
 int __pkvm_prot_finalize(void);
 int __pkvm_mark_hyp(phys_addr_t start, phys_addr_t end);
 
-int kvm_host_prepare_stage2(void *mem_pgt_pool, void *dev_pgt_pool);
+int kvm_host_prepare_stage2(void *pgt_pool_base);
 void handle_host_mem_abort(struct kvm_cpu_context *host_ctxt);
 
 static __always_inline void __load_host_stage2(void)

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

@@ -7,12 +7,9 @@
 
 #include <linux/types.h>
 
-struct hyp_pool;
 struct hyp_page {
-	unsigned int refcount;
-	unsigned int order;
-	struct hyp_pool *pool;
-	struct list_head node;
+	unsigned short refcount;
+	unsigned short order;
 };
 
 extern u64 __hyp_vmemmap;

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

@@ -78,19 +78,20 @@ static inline unsigned long hyp_s1_pgtable_pages(void)
 	return res;
 }
 
-static inline unsigned long host_s2_mem_pgtable_pages(void)
+static inline unsigned long host_s2_pgtable_pages(void)
 {
+	unsigned long res;
+
 	/*
 	 * Include an extra 16 pages to safely upper-bound the worst case of
 	 * concatenated pgds.
 	 */
-	return __hyp_pgtable_total_pages() + 16;
-}
+	res = __hyp_pgtable_total_pages() + 16;
 
-static inline unsigned long host_s2_dev_pgtable_pages(void)
-{
 	/* Allow 1 GiB for MMIO mappings */
-	return __hyp_pgtable_max_pages(SZ_1G >> PAGE_SHIFT);
+	res += __hyp_pgtable_max_pages(SZ_1G >> PAGE_SHIFT);
+
+	return res;
 }
 
 #endif /* __KVM_HYP_MM_H */

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

@@ -23,8 +23,7 @@
 extern unsigned long hyp_nr_cpus;
 struct host_kvm host_kvm;
 
-static struct hyp_pool host_s2_mem;
-static struct hyp_pool host_s2_dev;
+static struct hyp_pool host_s2_pool;
 
 /*
  * Copies of the host's CPU features registers holding sanitized values.
@@ -36,7 +35,7 @@ static const u8 pkvm_hyp_id = 1;
 
 static void *host_s2_zalloc_pages_exact(size_t size)
 {
-	return hyp_alloc_pages(&host_s2_mem, get_order(size));
+	return hyp_alloc_pages(&host_s2_pool, get_order(size));
 }
 
 static void *host_s2_zalloc_page(void *pool)
@@ -44,20 +43,24 @@ static void *host_s2_zalloc_page(void *pool)
 	return hyp_alloc_pages(pool, 0);
 }
 
-static int prepare_s2_pools(void *mem_pgt_pool, void *dev_pgt_pool)
+static void host_s2_get_page(void *addr)
+{
+	hyp_get_page(&host_s2_pool, addr);
+}
+
+static void host_s2_put_page(void *addr)
+{
+	hyp_put_page(&host_s2_pool, addr);
+}
+
+static int prepare_s2_pool(void *pgt_pool_base)
 {
 	unsigned long nr_pages, pfn;
 	int ret;
 
-	pfn = hyp_virt_to_pfn(mem_pgt_pool);
-	nr_pages = host_s2_mem_pgtable_pages();
-	ret = hyp_pool_init(&host_s2_mem, pfn, nr_pages, 0);
-	if (ret)
-		return ret;
-
-	pfn = hyp_virt_to_pfn(dev_pgt_pool);
-	nr_pages = host_s2_dev_pgtable_pages();
-	ret = hyp_pool_init(&host_s2_dev, pfn, nr_pages, 0);
+	pfn = hyp_virt_to_pfn(pgt_pool_base);
+	nr_pages = host_s2_pgtable_pages();
+	ret = hyp_pool_init(&host_s2_pool, pfn, nr_pages, 0);
 	if (ret)
 		return ret;
 
@@ -67,8 +70,8 @@ static int prepare_s2_pools(void *mem_pgt_pool, void *dev_pgt_pool)
 		.phys_to_virt = hyp_phys_to_virt,
 		.virt_to_phys = hyp_virt_to_phys,
 		.page_count = hyp_page_count,
-		.get_page = hyp_get_page,
-		.put_page = hyp_put_page,
+		.get_page = host_s2_get_page,
+		.put_page = host_s2_put_page,
 	};
 
 	return 0;
@@ -86,7 +89,7 @@ static void prepare_host_vtcr(void)
 					  id_aa64mmfr1_el1_sys_val, phys_shift);
 }
 
-int kvm_host_prepare_stage2(void *mem_pgt_pool, void *dev_pgt_pool)
+int kvm_host_prepare_stage2(void *pgt_pool_base)
 {
 	struct kvm_s2_mmu *mmu = &host_kvm.arch.mmu;
 	int ret;
@@ -94,7 +97,7 @@ int kvm_host_prepare_stage2(void *mem_pgt_pool, void *dev_pgt_pool)
 	prepare_host_vtcr();
 	hyp_spin_lock_init(&host_kvm.lock);
 
-	ret = prepare_s2_pools(mem_pgt_pool, dev_pgt_pool);
+	ret = prepare_s2_pool(pgt_pool_base);
 	if (ret)
 		return ret;
 
@@ -199,11 +202,10 @@ static bool range_is_memory(u64 start, u64 end)
 }
 
 static inline int __host_stage2_idmap(u64 start, u64 end,
-				      enum kvm_pgtable_prot prot,
-				      struct hyp_pool *pool)
+				      enum kvm_pgtable_prot prot)
 {
 	return kvm_pgtable_stage2_map(&host_kvm.pgt, start, end - start, start,
-				      prot, pool);
+				      prot, &host_s2_pool);
 }
 
 static int host_stage2_idmap(u64 addr)
@@ -211,7 +213,6 @@ static int host_stage2_idmap(u64 addr)
 	enum kvm_pgtable_prot prot = KVM_PGTABLE_PROT_R | KVM_PGTABLE_PROT_W;
 	struct kvm_mem_range range;
 	bool is_memory = find_mem_range(addr, &range);
-	struct hyp_pool *pool = is_memory ? &host_s2_mem : &host_s2_dev;
 	int ret;
 
 	if (is_memory)
@@ -222,22 +223,21 @@ static int host_stage2_idmap(u64 addr)
 	if (ret)
 		goto unlock;
 
-	ret = __host_stage2_idmap(range.start, range.end, prot, pool);
-	if (is_memory || ret != -ENOMEM)
+	ret = __host_stage2_idmap(range.start, range.end, prot);
+	if (ret != -ENOMEM)
 		goto unlock;
 
 	/*
-	 * host_s2_mem has been provided with enough pages to cover all of
-	 * memory with page granularity, so we should never hit the ENOMEM case.
-	 * However, it is difficult to know how much of the MMIO range we will
-	 * need to cover upfront, so we may need to 'recycle' the pages if we
-	 * run out.
+	 * The pool has been provided with enough pages to cover all of memory
+	 * with page granularity, but it is difficult to know how much of the
+	 * MMIO range we will need to cover upfront, so we may need to 'recycle'
+	 * the pages if we run out.
 	 */
 	ret = host_stage2_unmap_dev_all();
 	if (ret)
 		goto unlock;
 
-	ret = __host_stage2_idmap(range.start, range.end, prot, pool);
+	ret = __host_stage2_idmap(range.start, range.end, prot);
 
 unlock:
 	hyp_spin_unlock(&host_kvm.lock);
@@ -258,7 +258,7 @@ int __pkvm_mark_hyp(phys_addr_t start, phys_addr_t end)
 
 	hyp_spin_lock(&host_kvm.lock);
 	ret = kvm_pgtable_stage2_set_owner(&host_kvm.pgt, start, end - start,
-					   &host_s2_mem, pkvm_hyp_id);
+					   &host_s2_pool, pkvm_hyp_id);
 	hyp_spin_unlock(&host_kvm.lock);
 
 	return ret != -EAGAIN ? ret : 0;

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

@@ -32,7 +32,7 @@ u64 __hyp_vmemmap;
  */
 static struct hyp_page *__find_buddy_nocheck(struct hyp_pool *pool,
 					     struct hyp_page *p,
-					     unsigned int order)
+					     unsigned short order)
 {
 	phys_addr_t addr = hyp_page_to_phys(p);
 
@@ -51,21 +51,49 @@ static struct hyp_page *__find_buddy_nocheck(struct hyp_pool *pool,
 /* Find a buddy page currently available for allocation */
 static struct hyp_page *__find_buddy_avail(struct hyp_pool *pool,
 					   struct hyp_page *p,
-					   unsigned int order)
+					   unsigned short order)
 {
 	struct hyp_page *buddy = __find_buddy_nocheck(pool, p, order);
 
-	if (!buddy || buddy->order != order || list_empty(&buddy->node))
+	if (!buddy || buddy->order != order || buddy->refcount)
 		return NULL;
 
 	return buddy;
 
 }
 
+/*
+ * Pages that are available for allocation are tracked in free-lists, so we use
+ * the pages themselves to store the list nodes to avoid wasting space. As the
+ * allocator always returns zeroed pages (which are zeroed on the hyp_put_page()
+ * path to optimize allocation speed), we also need to clean-up the list node in
+ * each page when we take it out of the list.
+ */
+static inline void page_remove_from_list(struct hyp_page *p)
+{
+	struct list_head *node = hyp_page_to_virt(p);
+
+	__list_del_entry(node);
+	memset(node, 0, sizeof(*node));
+}
+
+static inline void page_add_to_list(struct hyp_page *p, struct list_head *head)
+{
+	struct list_head *node = hyp_page_to_virt(p);
+
+	INIT_LIST_HEAD(node);
+	list_add_tail(node, head);
+}
+
+static inline struct hyp_page *node_to_page(struct list_head *node)
+{
+	return hyp_virt_to_page(node);
+}
+
 static void __hyp_attach_page(struct hyp_pool *pool,
 			      struct hyp_page *p)
 {
-	unsigned int order = p->order;
+	unsigned short order = p->order;
 	struct hyp_page *buddy;
 
 	memset(hyp_page_to_virt(p), 0, PAGE_SIZE << p->order);
@@ -83,32 +111,23 @@ static void __hyp_attach_page(struct hyp_pool *pool,
 			break;
 
 		/* Take the buddy out of its list, and coallesce with @p */
-		list_del_init(&buddy->node);
+		page_remove_from_list(buddy);
 		buddy->order = HYP_NO_ORDER;
 		p = min(p, buddy);
 	}
 
 	/* Mark the new head, and insert it */
 	p->order = order;
-	list_add_tail(&p->node, &pool->free_area[order]);
-}
-
-static void hyp_attach_page(struct hyp_page *p)
-{
-	struct hyp_pool *pool = hyp_page_to_pool(p);
-
-	hyp_spin_lock(&pool->lock);
-	__hyp_attach_page(pool, p);
-	hyp_spin_unlock(&pool->lock);
+	page_add_to_list(p, &pool->free_area[order]);
 }
 
 static struct hyp_page *__hyp_extract_page(struct hyp_pool *pool,
 					   struct hyp_page *p,
-					   unsigned int order)
+					   unsigned short order)
 {
 	struct hyp_page *buddy;
 
-	list_del_init(&p->node);
+	page_remove_from_list(p);
 	while (p->order > order) {
 		/*
 		 * The buddy of order n - 1 currently has HYP_NO_ORDER as it
@@ -119,30 +138,64 @@ static struct hyp_page *__hyp_extract_page(struct hyp_pool *pool,
 		p->order--;
 		buddy = __find_buddy_nocheck(pool, p, p->order);
 		buddy->order = p->order;
-		list_add_tail(&buddy->node, &pool->free_area[buddy->order]);
+		page_add_to_list(buddy, &pool->free_area[buddy->order]);
 	}
 
 	return p;
 }
 
-void hyp_put_page(void *addr)
+static inline void hyp_page_ref_inc(struct hyp_page *p)
 {
-	struct hyp_page *p = hyp_virt_to_page(addr);
+	BUG_ON(p->refcount == USHRT_MAX);
+	p->refcount++;
+}
 
+static inline int hyp_page_ref_dec_and_test(struct hyp_page *p)
+{
+	p->refcount--;
+	return (p->refcount == 0);
+}
+
+static inline void hyp_set_page_refcounted(struct hyp_page *p)
+{
+	BUG_ON(p->refcount);
+	p->refcount = 1;
+}
+
+static void __hyp_put_page(struct hyp_pool *pool, struct hyp_page *p)
+{
 	if (hyp_page_ref_dec_and_test(p))
-		hyp_attach_page(p);
+		__hyp_attach_page(pool, p);
+}
+
+/*
+ * Changes to the buddy tree and page refcounts must be done with the hyp_pool
+ * lock held. If a refcount change requires an update to the buddy tree (e.g.
+ * hyp_put_page()), both operations must be done within the same critical
+ * section to guarantee transient states (e.g. a page with null refcount but
+ * not yet attached to a free list) can't be observed by well-behaved readers.
+ */
+void hyp_put_page(struct hyp_pool *pool, void *addr)
+{
+	struct hyp_page *p = hyp_virt_to_page(addr);
+
+	hyp_spin_lock(&pool->lock);
+	__hyp_put_page(pool, p);
+	hyp_spin_unlock(&pool->lock);
 }
 
-void hyp_get_page(void *addr)
+void hyp_get_page(struct hyp_pool *pool, void *addr)
 {
 	struct hyp_page *p = hyp_virt_to_page(addr);
 
+	hyp_spin_lock(&pool->lock);
 	hyp_page_ref_inc(p);
+	hyp_spin_unlock(&pool->lock);
 }
 
-void *hyp_alloc_pages(struct hyp_pool *pool, unsigned int order)
+void *hyp_alloc_pages(struct hyp_pool *pool, unsigned short order)
 {
-	unsigned int i = order;
+	unsigned short i = order;
 	struct hyp_page *p;
 
 	hyp_spin_lock(&pool->lock);
@@ -156,11 +209,11 @@ void *hyp_alloc_pages(struct hyp_pool *pool, unsigned int order)
 	}
 
 	/* Extract it from the tree at the right order */
-	p = list_first_entry(&pool->free_area[i], struct hyp_page, node);
+	p = node_to_page(pool->free_area[i].next);
 	p = __hyp_extract_page(pool, p, order);
 
-	hyp_spin_unlock(&pool->lock);
 	hyp_set_page_refcounted(p);
+	hyp_spin_unlock(&pool->lock);
 
 	return hyp_page_to_virt(p);
 }
@@ -181,15 +234,14 @@ int hyp_pool_init(struct hyp_pool *pool, u64 pfn, unsigned int nr_pages,
 
 	/* Init the vmemmap portion */
 	p = hyp_phys_to_page(phys);
-	memset(p, 0, sizeof(*p) * nr_pages);
 	for (i = 0; i < nr_pages; i++) {
-		p[i].pool = pool;
-		INIT_LIST_HEAD(&p[i].node);
+		p[i].order = 0;
+		hyp_set_page_refcounted(&p[i]);
 	}
 
 	/* Attach the unused pages to the buddy tree */
 	for (i = reserved_pages; i < nr_pages; i++)
-		__hyp_attach_page(pool, &p[i]);
+		__hyp_put_page(pool, &p[i]);
 
 	return 0;
 }

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

@@ -24,8 +24,7 @@ unsigned long hyp_nr_cpus;
 
 static void *vmemmap_base;
 static void *hyp_pgt_base;
-static void *host_s2_mem_pgt_base;
-static void *host_s2_dev_pgt_base;
+static void *host_s2_pgt_base;
 static struct kvm_pgtable_mm_ops pkvm_pgtable_mm_ops;
 
 static int divide_memory_pool(void *virt, unsigned long size)
@@ -45,14 +44,9 @@ static int divide_memory_pool(void *virt, unsigned long size)
 	if (!hyp_pgt_base)
 		return -ENOMEM;
 
-	nr_pages = host_s2_mem_pgtable_pages();
-	host_s2_mem_pgt_base = hyp_early_alloc_contig(nr_pages);
-	if (!host_s2_mem_pgt_base)
-		return -ENOMEM;
-
-	nr_pages = host_s2_dev_pgtable_pages();
-	host_s2_dev_pgt_base = hyp_early_alloc_contig(nr_pages);
-	if (!host_s2_dev_pgt_base)
+	nr_pages = host_s2_pgtable_pages();
+	host_s2_pgt_base = hyp_early_alloc_contig(nr_pages);
+	if (!host_s2_pgt_base)
 		return -ENOMEM;
 
 	return 0;
@@ -143,6 +137,16 @@ static void *hyp_zalloc_hyp_page(void *arg)
 	return hyp_alloc_pages(&hpool, 0);
 }
 
+static void hpool_get_page(void *addr)
+{
+	hyp_get_page(&hpool, addr);
+}
+
+static void hpool_put_page(void *addr)
+{
+	hyp_put_page(&hpool, addr);
+}
+
 void __noreturn __pkvm_init_finalise(void)
 {
 	struct kvm_host_data *host_data = this_cpu_ptr(&kvm_host_data);
@@ -158,7 +162,7 @@ void __noreturn __pkvm_init_finalise(void)
 	if (ret)
 		goto out;
 
-	ret = kvm_host_prepare_stage2(host_s2_mem_pgt_base, host_s2_dev_pgt_base);
+	ret = kvm_host_prepare_stage2(host_s2_pgt_base);
 	if (ret)
 		goto out;
 
@@ -166,8 +170,8 @@ void __noreturn __pkvm_init_finalise(void)
 		.zalloc_page = hyp_zalloc_hyp_page,
 		.phys_to_virt = hyp_phys_to_virt,
 		.virt_to_phys = hyp_virt_to_phys,
-		.get_page = hyp_get_page,
-		.put_page = hyp_put_page,
+		.get_page = hpool_get_page,
+		.put_page = hpool_put_page,
 	};
 	pkvm_pgtable.mm_ops = &pkvm_pgtable_mm_ops;
 

arch/arm64/kvm/hyp/reserved_mem.c

@@ -71,8 +71,7 @@ void __init kvm_hyp_reserve(void)
 	}
 
 	hyp_mem_pages += hyp_s1_pgtable_pages();
-	hyp_mem_pages += host_s2_mem_pgtable_pages();
-	hyp_mem_pages += host_s2_dev_pgtable_pages();
+	hyp_mem_pages += host_s2_pgtable_pages();
 
 	/*
 	 * The hyp_vmemmap needs to be backed by pages, but these pages