KVM: MMU: large page support

Create large pages mappings if the guest PTE's are marked as such and
the underlying memory is hugetlbfs backed.  If the largepage contains
write-protected pages, a large pte is not used.

Gives a consistent 2% improvement for data copies on ram mounted
filesystem, without NPT/EPT.

Anthony measures a 4% improvement on 4-way kernbench, with NPT.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>

arch/x86/kvm/paging_tmpl.h

@@ -248,6 +248,7 @@ static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page,
 	pt_element_t gpte;
 	unsigned pte_access;
 	struct page *npage;
+	int largepage = vcpu->arch.update_pte.largepage;
 
 	gpte = *(const pt_element_t *)pte;
 	if (~gpte & (PT_PRESENT_MASK | PT_ACCESSED_MASK)) {
@@ -264,7 +265,8 @@ static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page,
 		return;
 	get_page(npage);
 	mmu_set_spte(vcpu, spte, page->role.access, pte_access, 0, 0,
-		     gpte & PT_DIRTY_MASK, NULL, gpte_to_gfn(gpte), npage);
+		     gpte & PT_DIRTY_MASK, NULL, largepage, gpte_to_gfn(gpte),
+		     npage);
 }
 
 /*
@@ -272,8 +274,8 @@ static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page,
  */
 static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
 			 struct guest_walker *walker,
-			 int user_fault, int write_fault, int *ptwrite,
-			 struct page *page)
+			 int user_fault, int write_fault, int largepage,
+			 int *ptwrite, struct page *page)
 {
 	hpa_t shadow_addr;
 	int level;
@@ -301,11 +303,19 @@ static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
 		shadow_ent = ((u64 *)__va(shadow_addr)) + index;
 		if (level == PT_PAGE_TABLE_LEVEL)
 			break;
-		if (is_shadow_present_pte(*shadow_ent)) {
+
+		if (largepage && level == PT_DIRECTORY_LEVEL)
+			break;
+
+		if (is_shadow_present_pte(*shadow_ent)
+		    && !is_large_pte(*shadow_ent)) {
 			shadow_addr = *shadow_ent & PT64_BASE_ADDR_MASK;
 			continue;
 		}
 
+		if (is_large_pte(*shadow_ent))
+			rmap_remove(vcpu->kvm, shadow_ent);
+
 		if (level - 1 == PT_PAGE_TABLE_LEVEL
 		    && walker->level == PT_DIRECTORY_LEVEL) {
 			metaphysical = 1;
@@ -339,7 +349,7 @@ static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
 	mmu_set_spte(vcpu, shadow_ent, access, walker->pte_access & access,
 		     user_fault, write_fault,
 		     walker->ptes[walker->level-1] & PT_DIRTY_MASK,
-		     ptwrite, walker->gfn, page);
+		     ptwrite, largepage, walker->gfn, page);
 
 	return shadow_ent;
 }
@@ -369,6 +379,7 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
 	int write_pt = 0;
 	int r;
 	struct page *page;
+	int largepage = 0;
 
 	pgprintk("%s: addr %lx err %x\n", __FUNCTION__, addr, error_code);
 	kvm_mmu_audit(vcpu, "pre page fault");
@@ -396,6 +407,14 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
 	}
 
 	down_read(&current->mm->mmap_sem);
+	if (walker.level == PT_DIRECTORY_LEVEL) {
+		gfn_t large_gfn;
+		large_gfn = walker.gfn & ~(KVM_PAGES_PER_HPAGE-1);
+		if (is_largepage_backed(vcpu, large_gfn)) {
+			walker.gfn = large_gfn;
+			largepage = 1;
+		}
+	}
 	page = gfn_to_page(vcpu->kvm, walker.gfn);
 	up_read(&current->mm->mmap_sem);
 
@@ -410,7 +429,8 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
 	spin_lock(&vcpu->kvm->mmu_lock);
 	kvm_mmu_free_some_pages(vcpu);
 	shadow_pte = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault,
-				  &write_pt, page);
+				  largepage, &write_pt, page);
+
 	pgprintk("%s: shadow pte %p %llx ptwrite %d\n", __FUNCTION__,
 		 shadow_pte, *shadow_pte, write_pt);
 

arch/x86/kvm/x86.c

@@ -88,6 +88,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
+	{ "largepages", VM_STAT(lpages) },
 	{ NULL }
 };
 

include/asm-x86/kvm_host.h

@@ -39,6 +39,13 @@
 #define INVALID_PAGE (~(hpa_t)0)
 #define UNMAPPED_GVA (~(gpa_t)0)
 
+/* shadow tables are PAE even on non-PAE hosts */
+#define KVM_HPAGE_SHIFT 21
+#define KVM_HPAGE_SIZE (1UL << KVM_HPAGE_SHIFT)
+#define KVM_HPAGE_MASK (~(KVM_HPAGE_SIZE - 1))
+
+#define KVM_PAGES_PER_HPAGE (KVM_HPAGE_SIZE / PAGE_SIZE)
+
 #define DE_VECTOR 0
 #define UD_VECTOR 6
 #define NM_VECTOR 7
@@ -230,6 +237,7 @@ struct kvm_vcpu_arch {
 	struct {
 		gfn_t gfn;          /* presumed gfn during guest pte update */
 		struct page *page;  /* page corresponding to that gfn */
+		int largepage;
 	} update_pte;
 
 	struct i387_fxsave_struct host_fx_image;
@@ -307,6 +315,7 @@ struct kvm_vm_stat {
 	u32 mmu_recycled;
 	u32 mmu_cache_miss;
 	u32 remote_tlb_flush;
+	u32 lpages;
 };
 
 struct kvm_vcpu_stat {

include/linux/kvm_host.h

@@ -103,6 +103,10 @@ struct kvm_memory_slot {
 	unsigned long flags;
 	unsigned long *rmap;
 	unsigned long *dirty_bitmap;
+	struct {
+		unsigned long rmap_pde;
+		int write_count;
+	} *lpage_info;
 	unsigned long userspace_addr;
 	int user_alloc;
 };
@@ -169,6 +173,7 @@ int kvm_arch_set_memory_region(struct kvm *kvm,
 				int user_alloc);
 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn);
 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
+unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
 void kvm_release_page_clean(struct page *page);
 void kvm_release_page_dirty(struct page *page);
 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,

virt/kvm/kvm_main.c

@@ -212,9 +212,13 @@ static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 	if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 		vfree(free->dirty_bitmap);
 
+	if (!dont || free->lpage_info != dont->lpage_info)
+		vfree(free->lpage_info);
+
 	free->npages = 0;
 	free->dirty_bitmap = NULL;
 	free->rmap = NULL;
+	free->lpage_info = NULL;
 }
 
 void kvm_free_physmem(struct kvm *kvm)
@@ -324,6 +328,25 @@ int __kvm_set_memory_region(struct kvm *kvm,
 		new.user_alloc = user_alloc;
 		new.userspace_addr = mem->userspace_addr;
 	}
+	if (npages && !new.lpage_info) {
+		int largepages = npages / KVM_PAGES_PER_HPAGE;
+		if (npages % KVM_PAGES_PER_HPAGE)
+			largepages++;
+		if (base_gfn % KVM_PAGES_PER_HPAGE)
+			largepages++;
+
+		new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
+
+		if (!new.lpage_info)
+			goto out_free;
+
+		memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
+
+		if (base_gfn % KVM_PAGES_PER_HPAGE)
+			new.lpage_info[0].write_count = 1;
+		if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
+			new.lpage_info[largepages-1].write_count = 1;
+	}
 
 	/* Allocate page dirty bitmap if needed */
 	if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
@@ -467,7 +490,7 @@ int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 }
 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 
-static unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
+unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 {
 	struct kvm_memory_slot *slot;