bitkeeper revision 1.835 (406d9481GqoZ_RrT3GukXhamv7rulA)

Delete shadow page tables when destroying domain

5 files changed, 161 insertions, 142 deletions

diff --git a/xen/common/domain.c b/xen/common/domain.c
index 2db0331d89..7b55ee89d4 100644
--- a/xen/common/domain.c
+++ b/xen/common/domain.c
@@ -19,6 +19,7 @@
 #include <xen/console.h>
 #include <xen/vbd.h>
 #include <asm/i387.h>
+#include <xen/shadow.h>
 
 #ifdef CONFIG_X86_64BITMODE
 #define ELFSIZE 64
@@ -382,6 +383,8 @@ void free_all_dom_mem(struct task_struct *p)
 
     INIT_LIST_HEAD(&zombies);
 
+    if ( p->mm.shadow_mode ) shadow_mode_disable(p);
+
     /* STEP 1. Drop the in-use reference to the page-table base. */
     put_page_and_type(&frame_table[pagetable_val(p->mm.pagetable) >>
                                   PAGE_SHIFT]);
diff --git a/xen/common/memory.c b/xen/common/memory.c
index edba5c02de..b7896e3cfb 100644
--- a/xen/common/memory.c
+++ b/xen/common/memory.c
@@ -770,7 +770,14 @@ void free_page_type(struct pfn_info *page, unsigned int type)
 	     (get_shadow_status(&current->mm, 
 				page-frame_table) & PSH_shadowed) )
 	{
-	    unshadow_table( page-frame_table, type );
+	    /* using 'current-mm' is safe because page type changes only
+	       occur within the context of the currently running domain as 
+	       pagetable pages can not be shared across domains. The one
+	       exception is when destroying a domain. However, we get away 
+	       with this as there's no way the current domain can have this
+	       mfn shadowed, so we won't get here... Phew! */
+
+ 	    unshadow_table( page-frame_table, type );
 	    put_shadow_status(&current->mm);
         }
 	return;
diff --git a/xen/common/schedule.c b/xen/common/schedule.c
index fcdf860351..6692bacaa3 100644
--- a/xen/common/schedule.c
+++ b/xen/common/schedule.c
@@ -283,7 +283,13 @@ long do_sched_op(unsigned long op)
 }
 
 
-/* sched_pause_sync - synchronously pause a domain's execution */
+/* sched_pause_sync - synchronously pause a domain's execution 
+
+XXXX This is horibly broken -- here just as a place holder at present,
+                               do not use.
+
+*/
+
 void sched_pause_sync(struct task_struct *p)
 {
     unsigned long flags;
@@ -293,7 +299,7 @@ void sched_pause_sync(struct task_struct *p)
 
     if ( schedule_data[cpu].curr != p )
         /* if not the current task, we can remove it from scheduling now */
-        SCHED_FN(pause, p);
+        SCHED_OP(pause, p);
 
     p->state = TASK_PAUSED;
     
diff --git a/xen/common/shadow.c b/xen/common/shadow.c
index 8e7b53db86..14d395cbf7 100644
--- a/xen/common/shadow.c
+++ b/xen/common/shadow.c
@@ -225,7 +225,7 @@ nomem:
     return -ENOMEM;
 }
 
-static void shadow_mode_disable( struct task_struct *p )
+void shadow_mode_disable( struct task_struct *p )
 {
     struct mm_struct *m = &p->mm;
     struct shadow_status *next;
@@ -353,144 +353,6 @@ static inline struct pfn_info *alloc_shadow_page( struct mm_struct *m )
     return alloc_domain_page( NULL );
 }
 
-/************************************************************************/
-
-static inline void mark_dirty( struct mm_struct *m, unsigned int mfn )
-{
-	unsigned int pfn = machine_to_phys_mapping[mfn];
-	ASSERT(m->shadow_dirty_bitmap);
-	if( likely(pfn<m->shadow_dirty_bitmap_size) )
-	{
-		// XXX use setbit
-		m->shadow_dirty_bitmap[pfn/(sizeof(int)*8)] |= 
-			(1<<(pfn%(sizeof(int)*8)));
-	}
-	else
-	{
-		printk("XXXX mark dirty overflow!");
-	}
-
-}
-
-/************************************************************************/
-
-static inline void l1pte_write_fault( struct mm_struct *m, 
-									  unsigned long *gpte_p, unsigned long *spte_p )
-{ 
-    unsigned long gpte = *gpte_p;
-    unsigned long spte = *spte_p;
-
-    switch( m->shadow_mode )
-    {
-    case SHM_test:
-		spte = gpte;
-		gpte |= _PAGE_DIRTY | _PAGE_ACCESSED;
-		spte |= _PAGE_RW | _PAGE_DIRTY | _PAGE_ACCESSED; 			
-		break;
-
-    case SHM_logdirty:
-		spte = gpte;
-		gpte |= _PAGE_DIRTY | _PAGE_ACCESSED;
-		spte |= _PAGE_RW | _PAGE_DIRTY | _PAGE_ACCESSED; 			
-		mark_dirty( m, gpte >> PAGE_SHIFT );
-		break;
-    }
-
-    *gpte_p = gpte;
-    *spte_p = spte;
-}
-
-static inline void l1pte_read_fault( struct mm_struct *m, 
-									 unsigned long *gpte_p, unsigned long *spte_p )
-{ 
-    unsigned long gpte = *gpte_p;
-    unsigned long spte = *spte_p;
-
-    switch( m->shadow_mode )
-    {
-    case SHM_test:
-		spte = gpte;
-		gpte |= _PAGE_ACCESSED;
-		spte |= _PAGE_ACCESSED; 			
-		if ( ! (gpte & _PAGE_DIRTY ) )
-			spte &= ~ _PAGE_RW;
-		break;
-
-    case SHM_logdirty:
-		spte = gpte;
-		gpte |= _PAGE_ACCESSED;
-		spte |= _PAGE_ACCESSED; 			
-		spte &= ~ _PAGE_RW;
-		break;
-    }
-
-    *gpte_p = gpte;
-    *spte_p = spte;
-}
-
-static inline void l1pte_no_fault( struct mm_struct *m, 
-								   unsigned long *gpte_p, unsigned long *spte_p )
-{ 
-    unsigned long gpte = *gpte_p;
-    unsigned long spte = *spte_p;
-
-    switch( m->shadow_mode )
-    {
-    case SHM_test:
-		spte = 0;
-		if ( (gpte & (_PAGE_PRESENT|_PAGE_ACCESSED) ) == 
-			 (_PAGE_PRESENT|_PAGE_ACCESSED) )
-		{
-			spte = gpte;
-			if ( ! (gpte & _PAGE_DIRTY ) )
-				spte &= ~ _PAGE_RW;
-		}
-		break;
-
-    case SHM_logdirty:
-		spte = 0;
-		if ( (gpte & (_PAGE_PRESENT|_PAGE_ACCESSED) ) == 
-			 (_PAGE_PRESENT|_PAGE_ACCESSED) )
-		{
-			spte = gpte;
-			spte &= ~ _PAGE_RW;
-		}
-
-		break;
-    }
-
-    *gpte_p = gpte;
-    *spte_p = spte;
-}
-
-static inline void l2pde_general( struct mm_struct *m, 
-			   unsigned long *gpde_p, unsigned long *spde_p,
-			   unsigned long sl1pfn)
-{
-    unsigned long gpde = *gpde_p;
-    unsigned long spde = *spde_p;
-
-	spde = 0;
-
-	if ( sl1pfn )
-	{
-		spde = (gpde & ~PAGE_MASK) | (sl1pfn<<PAGE_SHIFT) | 
-			_PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY;
-		gpde = gpde | _PAGE_ACCESSED | _PAGE_DIRTY;
-
-		if ( unlikely( (sl1pfn<<PAGE_SHIFT) == (gpde & PAGE_MASK)  ) )
-		{   
-			// detect linear map, and keep pointing at guest
-			SH_VLOG("4c: linear mapping ( %08lx )",sl1pfn);
-			spde = gpde & ~_PAGE_RW;
-		}
-	}
-
-    *gpde_p = gpde;
-    *spde_p = spde;
-}
-
-/*********************************************************************/
 
 void unshadow_table( unsigned long gpfn, unsigned int type )
 {
diff --git a/xen/include/xen/shadow.h b/xen/include/xen/shadow.h
index 0f4e2e19d4..f5c0d5327a 100644
--- a/xen/include/xen/shadow.h
+++ b/xen/include/xen/shadow.h
@@ -31,6 +31,7 @@ extern void shadow_l1_normal_pt_update( unsigned long pa, unsigned long gpte,
 extern void shadow_l2_normal_pt_update( unsigned long pa, unsigned long gpte );
 extern void unshadow_table( unsigned long gpfn, unsigned int type );
 extern int shadow_mode_enable( struct task_struct *p, unsigned int mode );
+extern void shadow_mode_disable( struct task_struct *p );
 extern unsigned long shadow_l2_table( 
                      struct mm_struct *m, unsigned long gpfn );
 
@@ -73,6 +74,146 @@ struct shadow_status {
 
 
 
+/************************************************************************/
+
+static inline void mark_dirty( struct mm_struct *m, unsigned int mfn )
+{
+	unsigned int pfn = machine_to_phys_mapping[mfn];
+	ASSERT(m->shadow_dirty_bitmap);
+	if( likely(pfn<m->shadow_dirty_bitmap_size) )
+	{
+		// use setbit to be smp guest safe
+		set_bit( pfn, m->shadow_dirty_bitmap );
+	}
+	else
+	{
+		SH_LOG("mark_dirty pfn out of range attempt!");
+	}
+
+}
+
+/************************************************************************/
+
+static inline void l1pte_write_fault( struct mm_struct *m, 
+									  unsigned long *gpte_p, unsigned long *spte_p )
+{ 
+    unsigned long gpte = *gpte_p;
+    unsigned long spte = *spte_p;
+
+    switch( m->shadow_mode )
+    {
+    case SHM_test:
+		spte = gpte;
+		gpte |= _PAGE_DIRTY | _PAGE_ACCESSED;
+		spte |= _PAGE_RW | _PAGE_DIRTY | _PAGE_ACCESSED; 			
+		break;
+
+    case SHM_logdirty:
+		spte = gpte;
+		gpte |= _PAGE_DIRTY | _PAGE_ACCESSED;
+		spte |= _PAGE_RW | _PAGE_DIRTY | _PAGE_ACCESSED; 			
+		mark_dirty( m, gpte >> PAGE_SHIFT );
+		break;
+    }
+
+    *gpte_p = gpte;
+    *spte_p = spte;
+}
+
+static inline void l1pte_read_fault( struct mm_struct *m, 
+									 unsigned long *gpte_p, unsigned long *spte_p )
+{ 
+    unsigned long gpte = *gpte_p;
+    unsigned long spte = *spte_p;
+
+    switch( m->shadow_mode )
+    {
+    case SHM_test:
+		spte = gpte;
+		gpte |= _PAGE_ACCESSED;
+		spte |= _PAGE_ACCESSED; 			
+		if ( ! (gpte & _PAGE_DIRTY ) )
+			spte &= ~ _PAGE_RW;
+		break;
+
+    case SHM_logdirty:
+		spte = gpte;
+		gpte |= _PAGE_ACCESSED;
+		spte |= _PAGE_ACCESSED; 			
+		spte &= ~ _PAGE_RW;
+		break;
+    }
+
+    *gpte_p = gpte;
+    *spte_p = spte;
+}
+
+static inline void l1pte_no_fault( struct mm_struct *m, 
+								   unsigned long *gpte_p, unsigned long *spte_p )
+{ 
+    unsigned long gpte = *gpte_p;
+    unsigned long spte = *spte_p;
+
+    switch( m->shadow_mode )
+    {
+    case SHM_test:
+		spte = 0;
+		if ( (gpte & (_PAGE_PRESENT|_PAGE_ACCESSED) ) == 
+			 (_PAGE_PRESENT|_PAGE_ACCESSED) )
+		{
+			spte = gpte;
+			if ( ! (gpte & _PAGE_DIRTY ) )
+				spte &= ~ _PAGE_RW;
+		}
+		break;
+
+    case SHM_logdirty:
+		spte = 0;
+		if ( (gpte & (_PAGE_PRESENT|_PAGE_ACCESSED) ) == 
+			 (_PAGE_PRESENT|_PAGE_ACCESSED) )
+		{
+			spte = gpte;
+			spte &= ~ _PAGE_RW;
+		}
+
+		break;
+    }
+
+    *gpte_p = gpte;
+    *spte_p = spte;
+}
+
+static inline void l2pde_general( struct mm_struct *m, 
+			   unsigned long *gpde_p, unsigned long *spde_p,
+			   unsigned long sl1pfn)
+{
+    unsigned long gpde = *gpde_p;
+    unsigned long spde = *spde_p;
+
+	spde = 0;
+
+	if ( sl1pfn )
+	{
+		spde = (gpde & ~PAGE_MASK) | (sl1pfn<<PAGE_SHIFT) | 
+			_PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY;
+		gpde = gpde | _PAGE_ACCESSED | _PAGE_DIRTY;
+
+		if ( unlikely( (sl1pfn<<PAGE_SHIFT) == (gpde & PAGE_MASK)  ) )
+		{   
+			// detect linear map, and keep pointing at guest
+			SH_VLOG("4c: linear mapping ( %08lx )",sl1pfn);
+			spde = gpde & ~_PAGE_RW;
+		}
+	}
+
+    *gpde_p = gpde;
+    *spde_p = spde;
+}
+
+/*********************************************************************/
+
+
+
 #if SHADOW_HASH_DEBUG
 static void shadow_audit(struct mm_struct *m, int print)
 {