[PATCH] KVM: MMU: Implement simple reverse mapping
authorAvi Kivity <avi@qumranet.com>
Sat, 6 Jan 2007 00:36:38 +0000 (16:36 -0800)
committerLinus Torvalds <torvalds@woody.osdl.org>
Sat, 6 Jan 2007 07:55:24 +0000 (23:55 -0800)
Keep in each host page frame's page->private a pointer to the shadow pte which
maps it.  If there are multiple shadow ptes mapping the page, set bit 0 of
page->private, and use the rest as a pointer to a linked list of all such
mappings.

Reverse mappings are needed because we when we cache shadow page tables, we
must protect the guest page tables from being modified by the guest, as that
would invalidate the cached ptes.

Signed-off-by: Avi Kivity <avi@qumranet.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
drivers/kvm/kvm.h
drivers/kvm/kvm_main.c
drivers/kvm/mmu.c
drivers/kvm/paging_tmpl.h

index e8fe1039e3b52a3f304f2c89da25180a4b39cdac..b65511ed4388af3ca63655e0b4e79d108ee19d15 100644 (file)
@@ -236,6 +236,7 @@ struct kvm {
        struct kvm_vcpu vcpus[KVM_MAX_VCPUS];
        int memory_config_version;
        int busy;
+       unsigned long rmap_overflow;
 };
 
 struct kvm_stat {
index bc88c334664b67c56843b07183afc626d4776de1..f2a6b6f0e929994f3b246e475c57033dbaa75507 100644 (file)
@@ -638,6 +638,7 @@ raced:
                                                     | __GFP_ZERO);
                        if (!new.phys_mem[i])
                                goto out_free;
+                       new.phys_mem[i]->private = 0;
                }
        }
 
index 790423c5f23d2dab18dcf98eba55330c93177935..0f27beb6c5df80f334107a9413d1a74309c85a64 100644 (file)
@@ -27,6 +27,7 @@
 #include "kvm.h"
 
 #define pgprintk(x...) do { } while (0)
+#define rmap_printk(x...) do { } while (0)
 
 #define ASSERT(x)                                                      \
        if (!(x)) {                                                     \
 #define PT_DIRECTORY_LEVEL 2
 #define PT_PAGE_TABLE_LEVEL 1
 
+#define RMAP_EXT 4
+
+struct kvm_rmap_desc {
+       u64 *shadow_ptes[RMAP_EXT];
+       struct kvm_rmap_desc *more;
+};
+
 static int is_write_protection(struct kvm_vcpu *vcpu)
 {
        return vcpu->cr0 & CR0_WP_MASK;
@@ -150,6 +158,120 @@ static int is_io_pte(unsigned long pte)
        return pte & PT_SHADOW_IO_MARK;
 }
 
+static int is_rmap_pte(u64 pte)
+{
+       return (pte & (PT_WRITABLE_MASK | PT_PRESENT_MASK))
+               == (PT_WRITABLE_MASK | PT_PRESENT_MASK);
+}
+
+/*
+ * Reverse mapping data structures:
+ *
+ * If page->private bit zero is zero, then page->private points to the
+ * shadow page table entry that points to page_address(page).
+ *
+ * If page->private bit zero is one, (then page->private & ~1) points
+ * to a struct kvm_rmap_desc containing more mappings.
+ */
+static void rmap_add(struct kvm *kvm, u64 *spte)
+{
+       struct page *page;
+       struct kvm_rmap_desc *desc;
+       int i;
+
+       if (!is_rmap_pte(*spte))
+               return;
+       page = pfn_to_page((*spte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT);
+       if (!page->private) {
+               rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte);
+               page->private = (unsigned long)spte;
+       } else if (!(page->private & 1)) {
+               rmap_printk("rmap_add: %p %llx 1->many\n", spte, *spte);
+               desc = kzalloc(sizeof *desc, GFP_NOWAIT);
+               if (!desc)
+                       BUG(); /* FIXME: return error */
+               desc->shadow_ptes[0] = (u64 *)page->private;
+               desc->shadow_ptes[1] = spte;
+               page->private = (unsigned long)desc | 1;
+       } else {
+               rmap_printk("rmap_add: %p %llx many->many\n", spte, *spte);
+               desc = (struct kvm_rmap_desc *)(page->private & ~1ul);
+               while (desc->shadow_ptes[RMAP_EXT-1] && desc->more)
+                       desc = desc->more;
+               if (desc->shadow_ptes[RMAP_EXT-1]) {
+                       desc->more = kzalloc(sizeof *desc->more, GFP_NOWAIT);
+                       if (!desc->more)
+                               BUG(); /* FIXME: return error */
+                       desc = desc->more;
+               }
+               for (i = 0; desc->shadow_ptes[i]; ++i)
+                       ;
+               desc->shadow_ptes[i] = spte;
+       }
+}
+
+static void rmap_desc_remove_entry(struct page *page,
+                                  struct kvm_rmap_desc *desc,
+                                  int i,
+                                  struct kvm_rmap_desc *prev_desc)
+{
+       int j;
+
+       for (j = RMAP_EXT - 1; !desc->shadow_ptes[j] && j > i; --j)
+               ;
+       desc->shadow_ptes[i] = desc->shadow_ptes[j];
+       desc->shadow_ptes[j] = 0;
+       if (j != 0)
+               return;
+       if (!prev_desc && !desc->more)
+               page->private = (unsigned long)desc->shadow_ptes[0];
+       else
+               if (prev_desc)
+                       prev_desc->more = desc->more;
+               else
+                       page->private = (unsigned long)desc->more | 1;
+       kfree(desc);
+}
+
+static void rmap_remove(struct kvm *kvm, u64 *spte)
+{
+       struct page *page;
+       struct kvm_rmap_desc *desc;
+       struct kvm_rmap_desc *prev_desc;
+       int i;
+
+       if (!is_rmap_pte(*spte))
+               return;
+       page = pfn_to_page((*spte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT);
+       if (!page->private) {
+               printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte);
+               BUG();
+       } else if (!(page->private & 1)) {
+               rmap_printk("rmap_remove:  %p %llx 1->0\n", spte, *spte);
+               if ((u64 *)page->private != spte) {
+                       printk(KERN_ERR "rmap_remove:  %p %llx 1->BUG\n",
+                              spte, *spte);
+                       BUG();
+               }
+               page->private = 0;
+       } else {
+               rmap_printk("rmap_remove:  %p %llx many->many\n", spte, *spte);
+               desc = (struct kvm_rmap_desc *)(page->private & ~1ul);
+               prev_desc = NULL;
+               while (desc) {
+                       for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i)
+                               if (desc->shadow_ptes[i] == spte) {
+                                       rmap_desc_remove_entry(page, desc, i,
+                                                              prev_desc);
+                                       return;
+                               }
+                       prev_desc = desc;
+                       desc = desc->more;
+               }
+               BUG();
+       }
+}
+
 static void kvm_mmu_free_page(struct kvm_vcpu *vcpu, hpa_t page_hpa)
 {
        struct kvm_mmu_page *page_head = page_header(page_hpa);
@@ -229,27 +351,27 @@ hpa_t gva_to_hpa(struct kvm_vcpu *vcpu, gva_t gva)
 static void release_pt_page_64(struct kvm_vcpu *vcpu, hpa_t page_hpa,
                               int level)
 {
+       u64 *pos;
+       u64 *end;
+
        ASSERT(vcpu);
        ASSERT(VALID_PAGE(page_hpa));
        ASSERT(level <= PT64_ROOT_LEVEL && level > 0);
 
-       if (level == 1)
-               memset(__va(page_hpa), 0, PAGE_SIZE);
-       else {
-               u64 *pos;
-               u64 *end;
+       for (pos = __va(page_hpa), end = pos + PT64_ENT_PER_PAGE;
+            pos != end; pos++) {
+               u64 current_ent = *pos;
 
-               for (pos = __va(page_hpa), end = pos + PT64_ENT_PER_PAGE;
-                    pos != end; pos++) {
-                       u64 current_ent = *pos;
-
-                       *pos = 0;
-                       if (is_present_pte(current_ent))
+               if (is_present_pte(current_ent)) {
+                       if (level != 1)
                                release_pt_page_64(vcpu,
                                                  current_ent &
                                                  PT64_BASE_ADDR_MASK,
                                                  level - 1);
+                       else
+                               rmap_remove(vcpu->kvm, pos);
                }
+               *pos = 0;
        }
        kvm_mmu_free_page(vcpu, page_hpa);
 }
@@ -275,6 +397,7 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, hpa_t p)
                        page_header_update_slot(vcpu->kvm, table, v);
                        table[index] = p | PT_PRESENT_MASK | PT_WRITABLE_MASK |
                                                                PT_USER_MASK;
+                       rmap_add(vcpu->kvm, &table[index]);
                        return 0;
                }
 
@@ -437,6 +560,7 @@ static inline void set_pte_common(struct kvm_vcpu *vcpu,
        } else {
                *shadow_pte |= paddr;
                page_header_update_slot(vcpu->kvm, shadow_pte, gaddr);
+               rmap_add(vcpu->kvm, shadow_pte);
        }
 }
 
@@ -489,6 +613,7 @@ static void paging_inval_page(struct kvm_vcpu *vcpu, gva_t addr)
                u64 *table = __va(page_addr);
 
                if (level == PT_PAGE_TABLE_LEVEL ) {
+                       rmap_remove(vcpu->kvm, &table[index]);
                        table[index] = 0;
                        return;
                }
@@ -679,8 +804,9 @@ void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot)
                pt = __va(page->page_hpa);
                for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
                        /* avoid RMW */
-                       if (pt[i] & PT_WRITABLE_MASK)
+                       if (pt[i] & PT_WRITABLE_MASK) {
+                               rmap_remove(kvm, &pt[i]);
                                pt[i] &= ~PT_WRITABLE_MASK;
-
+                       }
        }
 }
index 09bb9b4ed12d1918d5843dfae6a53328a98500a7..8c48528a6e890ebcb82973292278d794eb876808 100644 (file)
@@ -261,6 +261,7 @@ static int FNAME(fix_write_pf)(struct kvm_vcpu *vcpu,
        mark_page_dirty(vcpu->kvm, gfn);
        *shadow_ent |= PT_WRITABLE_MASK;
        *guest_ent |= PT_DIRTY_MASK;
+       rmap_add(vcpu->kvm, shadow_ent);
 
        return 1;
 }