hvm: Fix up guest_table handling after p2m changes.

Fixes a host crash on HVM guest restore.
Signed-off-by: Keir Fraser <keir@xensource.com>

4 files changed, 76 insertions, 100 deletions

diff --git a/xen/arch/x86/hvm/hvm.c b/xen/arch/x86/hvm/hvm.c
index fa9752cd78..8560076cec 100644
--- a/xen/arch/x86/hvm/hvm.c
+++ b/xen/arch/x86/hvm/hvm.c
@@ -586,8 +586,7 @@ int hvm_set_cr0(unsigned long value)
 
         if ( !paging_mode_hap(v->domain) )
         {
-            put_page(mfn_to_page(get_mfn_from_gpfn(
-                v->arch.hvm_vcpu.guest_cr[3] >> PAGE_SHIFT)));
+            put_page(pagetable_get_page(v->arch.guest_table));
             v->arch.guest_table = pagetable_null();
         }
     }
@@ -603,21 +602,11 @@ int hvm_set_cr0(unsigned long value)
 
 int hvm_set_cr3(unsigned long value)
 {
-    unsigned long old_base_mfn, mfn;
+    unsigned long mfn;
     struct vcpu *v = current;
 
-    if ( paging_mode_hap(v->domain) || !hvm_paging_enabled(v) )
-    {
-        /* Nothing to do. */
-    }
-    else if ( value == v->arch.hvm_vcpu.guest_cr[3] )
-    {
-        /* Shadow-mode TLB flush. Invalidate the shadow. */
-        mfn = get_mfn_from_gpfn(value >> PAGE_SHIFT);
-        if ( mfn != pagetable_get_pfn(v->arch.guest_table) )
-            goto bad_cr3;
-    }
-    else 
+    if ( hvm_paging_enabled(v) && !paging_mode_hap(v->domain) &&
+         (value != v->arch.hvm_vcpu.guest_cr[3]) )
     {
         /* Shadow-mode CR3 change. Check PDBR and then make a new shadow. */
         HVM_DBG_LOG(DBG_LEVEL_VMMU, "CR3 value = %lx", value);
@@ -625,12 +614,9 @@ int hvm_set_cr3(unsigned long value)
         if ( !mfn_valid(mfn) || !get_page(mfn_to_page(mfn), v->domain) )
             goto bad_cr3;
 
-        old_base_mfn = pagetable_get_pfn(v->arch.guest_table);
+        put_page(pagetable_get_page(v->arch.guest_table));
         v->arch.guest_table = pagetable_from_pfn(mfn);
 
-        if ( old_base_mfn )
-            put_page(mfn_to_page(old_base_mfn));
-
         HVM_DBG_LOG(DBG_LEVEL_VMMU, "Update CR3 value = %lx", value);
     }
 
diff --git a/xen/arch/x86/hvm/svm/svm.c b/xen/arch/x86/hvm/svm/svm.c
index 46cefe2591..9e56ab6a29 100644
--- a/xen/arch/x86/hvm/svm/svm.c
+++ b/xen/arch/x86/hvm/svm/svm.c
@@ -337,9 +337,37 @@ int svm_vmcb_save(struct vcpu *v, struct hvm_hw_cpu *c)
 
 int svm_vmcb_restore(struct vcpu *v, struct hvm_hw_cpu *c)
 {
-    unsigned long mfn, old_base_mfn;
+    unsigned long mfn = 0;
     struct vmcb_struct *vmcb = v->arch.hvm_svm.vmcb;
 
+    if ( c->pending_valid &&
+         ((c->pending_type == 1) || (c->pending_type > 6) ||
+          (c->pending_reserved != 0)) )
+    {
+        gdprintk(XENLOG_ERR, "Invalid pending event 0x%"PRIx32".\n",
+                 c->pending_event);
+        return -EINVAL;
+    }
+
+    if ( !paging_mode_hap(v->domain) )
+    {
+        if ( c->cr0 & X86_CR0_PG )
+        {
+            mfn = gmfn_to_mfn(v->domain, c->cr3 >> PAGE_SHIFT);
+            if ( !mfn_valid(mfn) || !get_page(mfn_to_page(mfn), v->domain) )
+            {
+                gdprintk(XENLOG_ERR, "Invalid CR3 value=0x%"PRIx64"\n",
+                         c->cr3);
+                return -EINVAL;
+            }
+        }
+
+        if ( v->arch.hvm_vcpu.guest_cr[0] & X86_CR0_PG )
+            put_page(pagetable_get_page(v->arch.guest_table));
+
+        v->arch.guest_table = pagetable_from_pfn(mfn);
+    }
+
     vmcb->rip    = c->rip;
     vmcb->rsp    = c->rsp;
     vmcb->rflags = c->rflags;
@@ -357,18 +385,6 @@ int svm_vmcb_restore(struct vcpu *v, struct hvm_hw_cpu *c)
            __func__, c->cr3, c->cr0, c->cr4);
 #endif
 
-    if ( hvm_paging_enabled(v) && !paging_mode_hap(v->domain) )
-    {
-        HVM_DBG_LOG(DBG_LEVEL_VMMU, "CR3 = %"PRIx64, c->cr3);
-        mfn = gmfn_to_mfn(v->domain, c->cr3 >> PAGE_SHIFT);
-        if( !mfn_valid(mfn) || !get_page(mfn_to_page(mfn), v->domain) ) 
-            goto bad_cr3;
-        old_base_mfn = pagetable_get_pfn(v->arch.guest_table);
-        v->arch.guest_table = pagetable_from_pfn(mfn);
-        if ( old_base_mfn )
-             put_page(mfn_to_page(old_base_mfn));
-    }
-
     vmcb->idtr.limit = c->idtr_limit;
     vmcb->idtr.base  = c->idtr_base;
 
@@ -435,14 +451,6 @@ int svm_vmcb_restore(struct vcpu *v, struct hvm_hw_cpu *c)
         gdprintk(XENLOG_INFO, "Re-injecting 0x%"PRIx32", 0x%"PRIx32"\n",
                  c->pending_event, c->error_code);
 
-        if ( (c->pending_type == 1) || (c->pending_type > 6) ||
-             (c->pending_reserved != 0) )
-        {
-            gdprintk(XENLOG_ERR, "Invalid pending event 0x%"PRIx32"\n", 
-                     c->pending_event);
-            return -EINVAL;
-        }
-
         if ( hvm_event_needs_reinjection(c->pending_type, c->pending_vector) )
         {
             vmcb->eventinj.bytes = c->pending_event;
@@ -453,10 +461,6 @@ int svm_vmcb_restore(struct vcpu *v, struct hvm_hw_cpu *c)
     paging_update_paging_modes(v);
 
     return 0;
- 
- bad_cr3:
-    gdprintk(XENLOG_ERR, "Invalid CR3 value=0x%"PRIx64"\n", c->cr3);
-    return -EINVAL;
 }
 
         
diff --git a/xen/arch/x86/hvm/vmx/vmx.c b/xen/arch/x86/hvm/vmx/vmx.c
index 0dd069ab2e..79d95e07c8 100644
--- a/xen/arch/x86/hvm/vmx/vmx.c
+++ b/xen/arch/x86/hvm/vmx/vmx.c
@@ -565,7 +565,31 @@ void vmx_vmcs_save(struct vcpu *v, struct hvm_hw_cpu *c)
 
 int vmx_vmcs_restore(struct vcpu *v, struct hvm_hw_cpu *c)
 {
-    unsigned long mfn, old_base_mfn;
+    unsigned long mfn = 0;
+
+    if ( c->pending_valid &&
+         ((c->pending_type == 1) || (c->pending_type > 6) ||
+          (c->pending_reserved != 0)) )
+    {
+        gdprintk(XENLOG_ERR, "Invalid pending event 0x%"PRIx32".\n",
+                 c->pending_event);
+        return -EINVAL;
+    }
+
+    if ( c->cr0 & X86_CR0_PG )
+    {
+        mfn = gmfn_to_mfn(v->domain, c->cr3 >> PAGE_SHIFT);
+        if ( !mfn_valid(mfn) || !get_page(mfn_to_page(mfn), v->domain) )
+        {
+            gdprintk(XENLOG_ERR, "Invalid CR3 value=0x%"PRIx64"\n", c->cr3);
+            return -EINVAL;
+        }
+    }
+
+    if ( v->arch.hvm_vcpu.guest_cr[0] & X86_CR0_PG )
+        put_page(pagetable_get_page(v->arch.guest_table));
+
+    v->arch.guest_table = pagetable_from_pfn(mfn);
 
     vmx_vmcs_enter(v);
 
@@ -586,18 +610,6 @@ int vmx_vmcs_restore(struct vcpu *v, struct hvm_hw_cpu *c)
            __func__, c->cr3, c->cr0, c->cr4);
 #endif
 
-    if ( hvm_paging_enabled(v) )
-    {
-        HVM_DBG_LOG(DBG_LEVEL_VMMU, "CR3 = %"PRIx64, c->cr3);
-        mfn = gmfn_to_mfn(v->domain, c->cr3 >> PAGE_SHIFT);
-        if ( !mfn_valid(mfn) || !get_page(mfn_to_page(mfn), v->domain) )
-            goto bad_cr3;
-        old_base_mfn = pagetable_get_pfn(v->arch.guest_table);
-        v->arch.guest_table = pagetable_from_pfn(mfn);
-        if ( old_base_mfn )
-            put_page(mfn_to_page(old_base_mfn));
-    }
-
     v->arch.hvm_vcpu.guest_efer = c->msr_efer;
     vmx_update_guest_efer(v);
 
@@ -662,14 +674,6 @@ int vmx_vmcs_restore(struct vcpu *v, struct hvm_hw_cpu *c)
         gdprintk(XENLOG_INFO, "Re-injecting 0x%"PRIx32", 0x%"PRIx32"\n",
                  c->pending_event, c->error_code);
 
-        if ( (c->pending_type == 1) || (c->pending_type > 6) ||
-             (c->pending_reserved != 0) )
-        {
-            gdprintk(XENLOG_ERR, "Invalid pending event 0x%"PRIx32".\n",
-                     c->pending_event);
-            return -EINVAL;
-        }
-
         if ( hvm_event_needs_reinjection(c->pending_type, c->pending_vector) )
         {
             vmx_vmcs_enter(v);
@@ -680,11 +684,6 @@ int vmx_vmcs_restore(struct vcpu *v, struct hvm_hw_cpu *c)
     }
 
     return 0;
-
- bad_cr3:
-    gdprintk(XENLOG_ERR, "Invalid CR3 value=0x%"PRIx64"\n", c->cr3);
-    vmx_vmcs_exit(v);
-    return -EINVAL;
 }
 
 #if defined(__x86_64__) && defined(HVM_DEBUG_SUSPEND)
@@ -1905,7 +1904,22 @@ static void vmx_world_save(struct vcpu *v, struct vmx_assist_context *c)
 
 static int vmx_world_restore(struct vcpu *v, struct vmx_assist_context *c)
 {
-    unsigned long mfn, old_base_mfn;
+    unsigned long mfn = 0;
+
+    if ( c->cr0 & X86_CR0_PG )
+    {
+        mfn = gmfn_to_mfn(v->domain, c->cr3 >> PAGE_SHIFT);
+        if ( !mfn_valid(mfn) || !get_page(mfn_to_page(mfn), v->domain) )
+        {
+            gdprintk(XENLOG_ERR, "Invalid CR3 value=%x", c->cr3);
+            return -EINVAL;
+        }
+    }
+
+    if ( v->arch.hvm_vcpu.guest_cr[0] & X86_CR0_PG )
+        put_page(pagetable_get_page(v->arch.guest_table));
+
+    v->arch.guest_table = pagetable_from_pfn(mfn);
 
     __vmwrite(GUEST_RIP, c->eip);
     __vmwrite(GUEST_RSP, c->esp);
@@ -1917,18 +1931,6 @@ static int vmx_world_restore(struct vcpu *v, struct vmx_assist_context *c)
     vmx_update_guest_cr(v, 0);
     vmx_update_guest_cr(v, 4);
 
-    if ( hvm_paging_enabled(v) )
-    {
-        HVM_DBG_LOG(DBG_LEVEL_VMMU, "CR3 = %x", c->cr3);
-        mfn = get_mfn_from_gpfn(c->cr3 >> PAGE_SHIFT);
-        if ( !mfn_valid(mfn) || !get_page(mfn_to_page(mfn), v->domain) )
-            goto bad_cr3;
-        old_base_mfn = pagetable_get_pfn(v->arch.guest_table);
-        v->arch.guest_table = pagetable_from_pfn(mfn);
-        if ( old_base_mfn )
-             put_page(mfn_to_page(old_base_mfn));
-    }
-
     __vmwrite(GUEST_IDTR_LIMIT, c->idtr_limit);
     __vmwrite(GUEST_IDTR_BASE, c->idtr_base);
 
@@ -1977,10 +1979,6 @@ static int vmx_world_restore(struct vcpu *v, struct vmx_assist_context *c)
 
     paging_update_paging_modes(v);
     return 0;
-
- bad_cr3:
-    gdprintk(XENLOG_ERR, "Invalid CR3 value=%x", c->cr3);
-    return -EINVAL;
 }
 
 enum { VMX_ASSIST_INVOKE = 0, VMX_ASSIST_RESTORE };
diff --git a/xen/arch/x86/mm/shadow/multi.c b/xen/arch/x86/mm/shadow/multi.c
index 34a2d0d074..e0a20bfac8 100644
--- a/xen/arch/x86/mm/shadow/multi.c
+++ b/xen/arch/x86/mm/shadow/multi.c
@@ -3502,24 +3502,12 @@ sh_update_cr3(struct vcpu *v, int do_locking)
     /* Double-check that the HVM code has sent us a sane guest_table */
     if ( is_hvm_domain(d) )
     {
-        gfn_t gfn;
-
         ASSERT(shadow_mode_external(d));
-
-        // Is paging enabled on this vcpu?
         if ( hvm_paging_enabled(v) )
-        {
-            gfn = _gfn(paddr_to_pfn(v->arch.hvm_vcpu.guest_cr[3]));
-            gmfn = gfn_to_mfn(d, gfn);
-            ASSERT(mfn_valid(gmfn));
-            ASSERT(pagetable_get_pfn(v->arch.guest_table) == mfn_x(gmfn));
-        } 
+            ASSERT(pagetable_get_pfn(v->arch.guest_table));
         else 
-        {
-            /* Paging disabled: guest_table points at a 32-bit 1-to-1 map */
             ASSERT(v->arch.guest_table.pfn
                    == d->arch.paging.shadow.unpaged_pagetable.pfn);
-        }
     }
 #endif