[XEN] Snapshot PAE l3es when they are shadowed.

We don't update the shadows so we mustn't look at the guest l3es
or we'll be confused by them if they change.
Signed-off-by: Tim Deegan <Tim.Deegan@xensource.com>

diff --git a/xen/arch/x86/hvm/svm/svm.c b/xen/arch/x86/hvm/svm/svm.c
index cc9043e4133e5fa80f28713806118bfee084caa3..d3df4a5c3e4d6dc6b448c60e72511fcbcf7f69cd 100644 (file)
--- a/xen/arch/x86/hvm/svm/svm.c
+++ b/xen/arch/x86/hvm/svm/svm.c
@@ -2679,7 +2679,7 @@ void walk_shadow_and_guest_pt(unsigned long gva)
     shadow_sync_va(v, gva);
 
     gpte.l1 = 0;
-    __copy_from_user(&gpte, &linear_pg_table[ l1_linear_offset(gva) ],
+    __copy_from_user(&gpte, &__linear_l1_table[ l1_linear_offset(gva) ],
                      sizeof(gpte) );
     printk( "G-PTE = %x, flags=%x\n", gpte.l1, l1e_get_flags(gpte) );
 

diff --git a/xen/arch/x86/mm.c b/xen/arch/x86/mm.c
index 2d0afd735fe91b4980254db967a9b5ac4bdf1969..5d043ad598c0f93f5051f7e630969e24a15582cd 100644 (file)
--- a/xen/arch/x86/mm.c
+++ b/xen/arch/x86/mm.c
@@ -374,9 +374,6 @@ void write_ptbase(struct vcpu *v)
 /* Should be called after CR3 is updated.
  * Updates vcpu->arch.cr3 and, for HVM guests, vcpu->arch.hvm_vcpu.cpu_cr3.
  * 
- * Also updates other state derived from CR3 (vcpu->arch.guest_vtable,
- * shadow_vtable, etc).
- *
  * Uses values found in vcpu->arch.(guest_table and guest_table_user), and
  * for HVM guests, arch.monitor_table and hvm's guest CR3.
  *

diff --git a/xen/arch/x86/mm/shadow/multi.c b/xen/arch/x86/mm/shadow/multi.c
index 5a1638df19c3cbd5a6d6ee65f746ae83ae3f1e86..fec6551d8c534e4d3f65c34c82da5591d269d7ec 100644 (file)
--- a/xen/arch/x86/mm/shadow/multi.c
+++ b/xen/arch/x86/mm/shadow/multi.c
@@ -237,7 +237,8 @@ guest_walk_tables(struct vcpu *v, unsigned long va, walk_t *gw, int guest_op)
 #if GUEST_PAGING_LEVELS >= 4 /* 64-bit only... */
     /* Get l4e from the top level table */
     gw->l4mfn = pagetable_get_mfn(v->arch.guest_table);
-    gw->l4e = (guest_l4e_t *)v->arch.guest_vtable + guest_l4_table_offset(va);
+    gw->l4e = (guest_l4e_t *)v->arch.paging.shadow.guest_vtable 
+        + guest_l4_table_offset(va);
     /* Walk down to the l3e */
     if ( !(guest_l4e_get_flags(*gw->l4e) & _PAGE_PRESENT) ) return 0;
     gw->l3mfn = vcpu_gfn_to_mfn(v, guest_l4e_get_gfn(*gw->l4e));
@@ -248,9 +249,8 @@ guest_walk_tables(struct vcpu *v, unsigned long va, walk_t *gw, int guest_op)
     gw->l3e = ((guest_l3e_t *)sh_map_domain_page(gw->l3mfn))
         + guest_l3_table_offset(va);
 #else /* PAE only... */
-    /* Get l3e from the top level table */
-    gw->l3mfn = pagetable_get_mfn(v->arch.guest_table);
-    gw->l3e = (guest_l3e_t *)v->arch.guest_vtable + guest_l3_table_offset(va);
+    /* Get l3e from the cache of the guest's top level table */
+    gw->l3e = (guest_l3e_t *)&v->arch.paging.shadow.gl3e[guest_l3_table_offset(va)];
 #endif /* PAE or 64... */
     /* Walk down to the l2e */
     if ( !(guest_l3e_get_flags(*gw->l3e) & _PAGE_PRESENT) ) return 0;
@@ -264,7 +264,8 @@ guest_walk_tables(struct vcpu *v, unsigned long va, walk_t *gw, int guest_op)
 #else /* 32-bit only... */
     /* Get l2e from the top level table */
     gw->l2mfn = pagetable_get_mfn(v->arch.guest_table);
-    gw->l2e = (guest_l2e_t *)v->arch.guest_vtable + guest_l2_table_offset(va);
+    gw->l2e = (guest_l2e_t *)v->arch.paging.shadow.guest_vtable 
+        + guest_l2_table_offset(va);
 #endif /* All levels... */
     
     if ( !(guest_l2e_get_flags(*gw->l2e) & _PAGE_PRESENT) ) return 0;
@@ -357,8 +358,8 @@ static inline void print_gw(walk_t *gw)
     SHADOW_PRINTK("   l4e=%p\n", gw->l4e);
     if ( gw->l4e )
         SHADOW_PRINTK("   *l4e=%" SH_PRI_gpte "\n", gw->l4e->l4);
-#endif /* PAE or 64... */
     SHADOW_PRINTK("   l3mfn=%" PRI_mfn "\n", mfn_x(gw->l3mfn));
+#endif /* PAE or 64... */
     SHADOW_PRINTK("   l3e=%p\n", gw->l3e);
     if ( gw->l3e )
         SHADOW_PRINTK("   *l3e=%" SH_PRI_gpte "\n", gw->l3e->l3);
@@ -3169,8 +3170,7 @@ sh_update_linear_entries(struct vcpu *v)
 #else /* GUEST_PAGING_LEVELS == 3 */
         
         shadow_l3e = (shadow_l3e_t *)&v->arch.paging.shadow.l3table;
-        /* Always safe to use guest_vtable, because it's globally mapped */
-        guest_l3e = v->arch.guest_vtable;
+        guest_l3e = (guest_l3e_t *)&v->arch.paging.shadow.gl3e;
 
 #endif /* GUEST_PAGING_LEVELS */
         
@@ -3268,39 +3268,37 @@ sh_update_linear_entries(struct vcpu *v)
 }
 
 
-/* Removes vcpu->arch.guest_vtable and vcpu->arch.shadow_table[].
+/* Removes vcpu->arch.paging.shadow.guest_vtable and vcpu->arch.shadow_table[].
  * Does all appropriate management/bookkeeping/refcounting/etc...
  */
 static void
 sh_detach_old_tables(struct vcpu *v)
 {
-    struct domain *d = v->domain;
     mfn_t smfn;
     int i = 0;
 
     ////
-    //// vcpu->arch.guest_vtable
+    //// vcpu->arch.paging.shadow.guest_vtable
     ////
-    if ( v->arch.guest_vtable )
+
+#if GUEST_PAGING_LEVELS == 3
+    /* PAE guests don't have a mapping of the guest top-level table */
+    ASSERT(v->arch.paging.shadow.guest_vtable == NULL);
+#else
+    if ( v->arch.paging.shadow.guest_vtable )
     {
-#if GUEST_PAGING_LEVELS == 4
-        if ( shadow_mode_external(d) || shadow_mode_translate(d) )
-            sh_unmap_domain_page_global(v->arch.guest_vtable);
-#elif GUEST_PAGING_LEVELS == 3
-        if ( 1 || shadow_mode_external(d) || shadow_mode_translate(d) )
-            sh_unmap_domain_page_global(v->arch.guest_vtable);
-#elif GUEST_PAGING_LEVELS == 2
+        struct domain *d = v->domain;
         if ( shadow_mode_external(d) || shadow_mode_translate(d) )
-            sh_unmap_domain_page_global(v->arch.guest_vtable);
-#endif
-        v->arch.guest_vtable = NULL;
+            sh_unmap_domain_page_global(v->arch.paging.shadow.guest_vtable);
+        v->arch.paging.shadow.guest_vtable = NULL;
     }
+#endif
+
 
     ////
     //// vcpu->arch.shadow_table[]
     ////
 
-
 #if GUEST_PAGING_LEVELS == 3
     /* PAE guests have four shadow_table entries */
     for ( i = 0 ; i < 4 ; i++ )
@@ -3398,7 +3396,9 @@ sh_update_cr3(struct vcpu *v, int do_locking)
     struct domain *d = v->domain;
     mfn_t gmfn;
 #if GUEST_PAGING_LEVELS == 3
+    guest_l3e_t *gl3e;
     u32 guest_idx=0;
+    int i;
 #endif
 
     /* Don't do anything on an uninitialised vcpu */
@@ -3457,55 +3457,54 @@ sh_update_cr3(struct vcpu *v, int do_locking)
 
 
     ////
-    //// vcpu->arch.guest_vtable
+    //// vcpu->arch.paging.shadow.guest_vtable
     ////
 #if GUEST_PAGING_LEVELS == 4
     if ( shadow_mode_external(d) || shadow_mode_translate(d) )
     {
-        if ( v->arch.guest_vtable )
-            sh_unmap_domain_page_global(v->arch.guest_vtable);
-        v->arch.guest_vtable = sh_map_domain_page_global(gmfn);
+        if ( v->arch.paging.shadow.guest_vtable )
+            sh_unmap_domain_page_global(v->arch.paging.shadow.guest_vtable);
+        v->arch.paging.shadow.guest_vtable = sh_map_domain_page_global(gmfn);
     }
     else
-        v->arch.guest_vtable = __linear_l4_table;
+        v->arch.paging.shadow.guest_vtable = __linear_l4_table;
 #elif GUEST_PAGING_LEVELS == 3
-    if ( v->arch.guest_vtable )
-        sh_unmap_domain_page_global(v->arch.guest_vtable);
-    if ( shadow_mode_external(d) )
-    {
-        if ( paging_vcpu_mode_translate(v) ) 
-            /* Paging enabled: find where in the page the l3 table is */
-            guest_idx = guest_index((void *)hvm_get_guest_ctrl_reg(v, 3));
-        else
-            /* Paging disabled: l3 is at the start of a page (in the p2m) */ 
-            guest_idx = 0; 
-
-        // Ignore the low 2 bits of guest_idx -- they are really just
-        // cache control.
-        guest_idx &= ~3;
-
-        // XXX - why does this need a global map?
-        v->arch.guest_vtable =
-            (guest_l3e_t *)sh_map_domain_page_global(gmfn) + guest_idx;
-    }
+     /* On PAE guests we don't use a mapping of the guest's own top-level
+      * table.  We cache the current state of that table and shadow that,
+      * until the next CR3 write makes us refresh our cache. */
+     ASSERT(v->arch.paging.shadow.guest_vtable == NULL);
+ 
+     if ( shadow_mode_external(d) && paging_vcpu_mode_translate(v) ) 
+         /* Paging enabled: find where in the page the l3 table is */
+         guest_idx = guest_index((void *)hvm_get_guest_ctrl_reg(v, 3));
     else
-        v->arch.guest_vtable = sh_map_domain_page_global(gmfn);
+        /* Paging disabled or PV: l3 is at the start of a page */ 
+        guest_idx = 0; 
+     
+     // Ignore the low 2 bits of guest_idx -- they are really just
+     // cache control.
+     guest_idx &= ~3;
+     
+     gl3e = ((guest_l3e_t *)sh_map_domain_page(gmfn)) + guest_idx;
+     for ( i = 0; i < 4 ; i++ )
+         v->arch.paging.shadow.gl3e[i] = gl3e[i];
+     sh_unmap_domain_page(gl3e);
 #elif GUEST_PAGING_LEVELS == 2
     if ( shadow_mode_external(d) || shadow_mode_translate(d) )
     {
-        if ( v->arch.guest_vtable )
-            sh_unmap_domain_page_global(v->arch.guest_vtable);
-        v->arch.guest_vtable = sh_map_domain_page_global(gmfn);
+        if ( v->arch.paging.shadow.guest_vtable )
+            sh_unmap_domain_page_global(v->arch.paging.shadow.guest_vtable);
+        v->arch.paging.shadow.guest_vtable = sh_map_domain_page_global(gmfn);
     }
     else
-        v->arch.guest_vtable = __linear_l2_table;
+        v->arch.paging.shadow.guest_vtable = __linear_l2_table;
 #else
 #error this should never happen
 #endif
 
 #if 0
-    printk("%s %s %d gmfn=%05lx guest_vtable=%p\n",
-           __func__, __FILE__, __LINE__, gmfn, v->arch.guest_vtable);
+    printk("%s %s %d gmfn=%05lx shadow.guest_vtable=%p\n",
+           __func__, __FILE__, __LINE__, gmfn, v->arch.paging.shadow.guest_vtable);
 #endif
 
     ////
@@ -3523,10 +3522,10 @@ sh_update_cr3(struct vcpu *v, int do_locking)
     /* PAE guests have four shadow_table entries, based on the 
      * current values of the guest's four l3es. */
     {
-        int i, flush = 0;
+        int flush = 0;
         gfn_t gl2gfn;
         mfn_t gl2mfn;
-        guest_l3e_t *gl3e = (guest_l3e_t*)v->arch.guest_vtable;
+        guest_l3e_t *gl3e = (guest_l3e_t*)&v->arch.paging.shadow.gl3e;
         /* First, make all four entries read-only. */
         for ( i = 0; i < 4; i++ )
         {

diff --git a/xen/arch/x86/mm/shadow/page-guest32.h b/xen/arch/x86/mm/shadow/page-guest32.h
index e93206169ae9a0a922af70592054c3d195e063f9..5d333bd91bb74594ece4c509cfda954605b83f47 100644 (file)
--- a/xen/arch/x86/mm/shadow/page-guest32.h
+++ b/xen/arch/x86/mm/shadow/page-guest32.h
@@ -87,11 +87,6 @@ static inline l2_pgentry_32_t l2e_from_paddr_32(paddr_t pa, unsigned int flags)
 #define l2_table_offset_32(a)         \
     (((a) >> L2_PAGETABLE_SHIFT_32) & (L2_PAGETABLE_ENTRIES_32 - 1))
 
-#define linear_l1_table_32                                                 \
-    ((l1_pgentry_32_t *)(LINEAR_PT_VIRT_START))
-
-#define linear_pg_table_32 linear_l1_table_32
-
 #endif /* __X86_PAGE_GUEST_H__ */
 
 /*

diff --git a/xen/arch/x86/mm/shadow/types.h b/xen/arch/x86/mm/shadow/types.h
index b94289e6747b0bcd3a4a3afa78a78a556120ccec..48670c31c1f8a3ff835305e2f42622e43e4da878 100644 (file)
--- a/xen/arch/x86/mm/shadow/types.h
+++ b/xen/arch/x86/mm/shadow/types.h
@@ -447,10 +447,8 @@ struct shadow_walk_t
     guest_l2e_t *l2e;           /* Pointer to guest's level 2 entry */
     guest_l1e_t *l1e;           /* Pointer to guest's level 1 entry */
     guest_l1e_t eff_l1e;        /* Effective level 1 entry */
-#if GUEST_PAGING_LEVELS >= 3
 #if GUEST_PAGING_LEVELS >= 4
     mfn_t l4mfn;                /* MFN that the level 4 entry is in */
-#endif
     mfn_t l3mfn;                /* MFN that the level 3 entry is in */
 #endif
     mfn_t l2mfn;                /* MFN that the level 2 entry is in */

diff --git a/xen/include/asm-x86/domain.h b/xen/include/asm-x86/domain.h
index b148d5682e532e3a2c661fed2ee45402c84d3252..3237f53e8e311451c0ac762085ffcccb91ae4ebf 100644 (file)
--- a/xen/include/asm-x86/domain.h
+++ b/xen/include/asm-x86/domain.h
@@ -92,7 +92,11 @@ struct shadow_vcpu {
 #if CONFIG_PAGING_LEVELS >= 3
     /* PAE guests: per-vcpu shadow top-level table */
     l3_pgentry_t l3table[4] __attribute__((__aligned__(32)));
+    /* PAE guests: per-vcpu cache of the top-level *guest* entries */
+    l3_pgentry_t gl3e[4] __attribute__((__aligned__(32)));
 #endif
+    /* Non-PAE guests: pointer to guest top-level pagetable */
+    void *guest_vtable;
     /* Last MFN that we emulated a write to. */
     unsigned long last_emulated_mfn;
     /* MFN of the last shadow that we shot a writeable mapping in */
@@ -242,8 +246,6 @@ struct arch_vcpu
     pagetable_t monitor_table;          /* (MFN) hypervisor PT (for HVM) */
     unsigned long cr3;                     /* (MA) value to install in HW CR3 */
 
-    void *guest_vtable;                 /* virtual addr of pagetable */
-
     /* Current LDT details. */
     unsigned long shadow_ldt_mapcnt;
 

diff --git a/xen/include/asm-x86/page.h b/xen/include/asm-x86/page.h
index 1adc4b481dc877700a7c0ba7fe69dfad2c348f2e..31ec281920662c6dda6a947030b2e1da7c316dd7 100644 (file)
--- a/xen/include/asm-x86/page.h
+++ b/xen/include/asm-x86/page.h
@@ -278,9 +278,6 @@ typedef struct { u64 pfn; } pagetable_t;
 #define __linear_l4_table \
  ((l4_pgentry_t *)(__linear_l3_table + l3_linear_offset(LINEAR_PT_VIRT_START)))
 
-#define linear_l1_table __linear_l1_table
-#define linear_pg_table linear_l1_table
-#define linear_l2_table(v) ((l2_pgentry_t *)(v)->arch.guest_vtable)
 
 #ifndef __ASSEMBLY__
 #if CONFIG_PAGING_LEVELS == 3