bitkeeper revision 1.1299 (425eef16s6fX6jXC3hHMyjTl5h4U4Q)

Forward port of I/O-page fix in 2.0 series.
Signed-off-by: Keir Fraser <keir@xensource.com>

diff --git a/xen/arch/x86/mm.c b/xen/arch/x86/mm.c
index 05a5a3649cf7a1dc520fac18e018445e4c0daa35..1a0a9f95120887bd439cd1d432e38b89d349e829 100644 (file)
--- a/xen/arch/x86/mm.c
+++ b/xen/arch/x86/mm.c
@@ -171,6 +171,9 @@ void arch_init_memory(void)
 {
     extern void subarch_init_memory(struct domain *);
 
+    unsigned long i, j, pfn, nr_pfns;
+    struct pfn_info *page;
+
     memset(percpu_info, 0, sizeof(percpu_info));
 
     /*
@@ -184,13 +187,42 @@ void arch_init_memory(void)
 
     /*
      * Initialise our DOMID_IO domain.
-     * This domain owns no pages but is considered a special case when
-     * mapping I/O pages, as the mappings occur at the priv of the caller.
+     * This domain owns I/O pages that are within the range of the pfn_info
+     * array. Mappings occur at the priv of the caller.
      */
     dom_io = alloc_domain_struct();
     atomic_set(&dom_io->refcnt, 1);
     dom_io->id = DOMID_IO;
 
+    /* First 1MB of RAM is historically marked as I/O. */
+    for ( i = 0; i < 0x100; i++ )
+    {
+        page = &frame_table[i];
+        page->count_info        = PGC_allocated | 1;
+        page->u.inuse.type_info = PGT_writable_page | PGT_validated | 1;
+        page_set_owner(page, dom_io);
+    }
+ 
+    /* Any non-RAM areas in the e820 map are considered to be for I/O. */
+    for ( i = 0; i < e820.nr_map; i++ )
+    {
+        if ( e820.map[i].type == E820_RAM )
+            continue;
+        pfn = e820.map[i].addr >> PAGE_SHIFT;
+        nr_pfns = (e820.map[i].size +
+                   (e820.map[i].addr & ~PAGE_MASK) +
+                   ~PAGE_MASK) >> PAGE_SHIFT;
+        for ( j = 0; j < nr_pfns; j++ )
+        {
+            if ( !pfn_valid(pfn+j) )
+                continue;
+            page = &frame_table[pfn+j];
+            page->count_info        = PGC_allocated | 1;
+            page->u.inuse.type_info = PGT_writable_page | PGT_validated | 1;
+            page_set_owner(page, dom_io);
+        }
+    }
+
     subarch_init_memory(dom_xen);
 }
 
@@ -306,13 +338,7 @@ static int get_page_from_pagenr(unsigned long page_nr, struct domain *d)
 {
     struct pfn_info *page = &frame_table[page_nr];
 
-    if ( unlikely(!pfn_is_ram(page_nr)) )
-    {
-        MEM_LOG("Pfn %p is not RAM", page_nr);
-        return 0;
-    }
-
-    if ( unlikely(!get_page(page, d)) )
+    if ( unlikely(!pfn_valid(page_nr)) || unlikely(!get_page(page, d)) )
     {
         MEM_LOG("Could not get page ref for pfn %p", page_nr);
         return 0;
@@ -419,20 +445,25 @@ get_page_from_l1e(
         return 0;
     }
 
-    if ( unlikely(!pfn_is_ram(mfn)) )
+    if ( unlikely(!pfn_valid(mfn)) ||
+         unlikely(page_get_owner(page) == dom_io) )
     {
-        /* Revert to caller privileges if FD == DOMID_IO. */
+        /* DOMID_IO reverts to caller for privilege checks. */
         if ( d == dom_io )
             d = current->domain;
 
-        if ( IS_PRIV(d) )
-            return 1;
+        if ( (!IS_PRIV(d)) &&
+             (!IS_CAPABLE_PHYSDEV(d) || !domain_iomem_in_pfn(d, mfn)) )
+        {
+            MEM_LOG("Non-privileged attempt to map I/O space %08lx", mfn);
+            return 0;
+        }
 
-        if ( IS_CAPABLE_PHYSDEV(d) )
-            return domain_iomem_in_pfn(d, mfn);
+        /* No reference counting for out-of-range I/O pages. */
+        if ( !pfn_valid(mfn) )
+            return 1;
 
-        MEM_LOG("Non-privileged attempt to map I/O space %p", mfn);
-        return 0;
+        d = dom_io;
     }
 
     return ((l1v & _PAGE_RW) ?
@@ -529,7 +560,7 @@ void put_page_from_l1e(l1_pgentry_t l1e, struct domain *d)
     struct pfn_info *page = &frame_table[pfn];
     struct domain   *e;
 
-    if ( !(l1v & _PAGE_PRESENT) || !pfn_is_ram(pfn) )
+    if ( !(l1v & _PAGE_PRESENT) || !pfn_valid(pfn) )
         return;
 
     e = page_get_owner(page);
@@ -2851,7 +2882,7 @@ void ptwr_destroy(struct domain *d)
         gntref = (grant_ref_t)((val & 0xFF00) | ((ptr >> 2) & 0x00FF));
         
         if ( unlikely(IS_XEN_HEAP_FRAME(page)) ||
-             unlikely(!pfn_is_ram(pfn)) ||
+             unlikely(!pfn_valid(pfn)) ||
              unlikely((e = find_domain_by_id(domid)) == NULL) )
         {
             MEM_LOG("Bad frame (%p) or bad domid (%d).\n", pfn, domid);

diff --git a/xen/arch/x86/shadow.c b/xen/arch/x86/shadow.c
index 94c460faa94d56c67f7d440dbf1c80d70a1cd70e..71867b7649faf2b467a4698e2ca816492603742f 100644 (file)
--- a/xen/arch/x86/shadow.c
+++ b/xen/arch/x86/shadow.c
@@ -1727,7 +1727,7 @@ shadow_mark_mfn_out_of_sync(struct exec_domain *ed, unsigned long gpfn,
     struct out_of_sync_entry *entry = shadow_alloc_oos_entry(d);
 
     ASSERT(spin_is_locked(&d->arch.shadow_lock));
-    ASSERT(pfn_is_ram(mfn));
+    ASSERT(pfn_valid(mfn));
     ASSERT((page->u.inuse.type_info & PGT_type_mask) == PGT_writable_page);
 
     FSH_LOG("%s(gpfn=%p, mfn=%p) c=%p t=%p", __func__,

diff --git a/xen/common/grant_table.c b/xen/common/grant_table.c
index 11258a488855b56aa70ee531381e3b853ed8fe96..76ea9819348998aad2699f3214bacdb2e5f11143 100644 (file)
--- a/xen/common/grant_table.c
+++ b/xen/common/grant_table.c
@@ -161,7 +161,7 @@ __gnttab_activate_grant_ref(
 
         frame = __gpfn_to_mfn_foreign(granting_d, sha->frame);
 
-        if ( unlikely(!pfn_is_ram(frame)) ||
+        if ( unlikely(!pfn_valid(frame)) ||
              unlikely(!((dev_hst_ro_flags & GNTMAP_readonly) ?
                         get_page(&frame_table[frame], granting_d) :
                         get_page_and_type(&frame_table[frame], granting_d,

diff --git a/xen/include/asm-x86/page.h b/xen/include/asm-x86/page.h
index d3410ba76ee90891f3308152e936788b9d23de90..cc222a545ec3e4d96c936368aa41b74038ad1222 100644 (file)
--- a/xen/include/asm-x86/page.h
+++ b/xen/include/asm-x86/page.h
@@ -35,14 +35,7 @@ typedef struct { unsigned long pt_lo; } pagetable_t;
 #define pfn_to_page(_pfn)   (frame_table + (_pfn))
 #define phys_to_page(kaddr) (frame_table + ((kaddr) >> PAGE_SHIFT))
 #define virt_to_page(kaddr) (frame_table + (__pa(kaddr) >> PAGE_SHIFT))
-#define VALID_PAGE(page)    ((page - frame_table) < max_mapnr)
-
-/*
- * NB. We don't currently track I/O holes in the physical RAM space.
- * For now we guess that I/O devices will be mapped in the first 1MB
- * (e.g., VGA buffers) or beyond the end of physical RAM.
- */
-#define pfn_is_ram(_pfn)    (((_pfn) > 0x100) && ((_pfn) < max_page))
+#define pfn_valid(_pfn)     ((_pfn) < max_page)
 
 /* High table entries are reserved by the hypervisor. */
 #define DOMAIN_ENTRIES_PER_L2_PAGETABLE     \

diff --git a/xen/include/asm-x86/shadow.h b/xen/include/asm-x86/shadow.h
index 0874bfc6403cf9213f71edf3da6307f3dd72482e..f9a1b8b988845e4f5e91752aa3f486e0c1ba22b2 100644 (file)
--- a/xen/include/asm-x86/shadow.h
+++ b/xen/include/asm-x86/shadow.h
@@ -85,7 +85,7 @@ static inline int page_is_page_table(struct pfn_info *page)
 
 static inline int mfn_is_page_table(unsigned long mfn)
 {
-    if ( !pfn_is_ram(mfn) )
+    if ( !pfn_valid(mfn) )
         return 0;
 
     return frame_table[mfn].count_info & PGC_page_table;
@@ -98,7 +98,7 @@ static inline int page_out_of_sync(struct pfn_info *page)
 
 static inline int mfn_out_of_sync(unsigned long mfn)
 {
-    if ( !pfn_is_ram(mfn) )
+    if ( !pfn_valid(mfn) )
         return 0;
 
     return frame_table[mfn].count_info & PGC_out_of_sync;
@@ -280,7 +280,7 @@ shadow_get_page_from_l1e(l1_pgentry_t l1e, struct domain *d)
     if ( unlikely(!res) && IS_PRIV(d) && !shadow_mode_translate(d) &&
          !(l1_pgentry_val(nl1e) & L1_DISALLOW_MASK) &&
          (mfn = l1_pgentry_to_pfn(nl1e)) &&
-         pfn_is_ram(mfn) &&
+         pfn_valid(mfn) &&
          (owner = page_get_owner(pfn_to_page(l1_pgentry_to_pfn(nl1e)))) &&
          (d != owner) )
     {
@@ -426,7 +426,7 @@ get_shadow_ref(unsigned long smfn)
 {
     u32 x, nx;
 
-    ASSERT(pfn_is_ram(smfn));
+    ASSERT(pfn_valid(smfn));
 
     x = frame_table[smfn].count_info;
     nx = x + 1;
@@ -455,7 +455,7 @@ put_shadow_ref(unsigned long smfn)
 {
     u32 x, nx;
 
-    ASSERT(pfn_is_ram(smfn));
+    ASSERT(pfn_valid(smfn));
 
     x = frame_table[smfn].count_info;
     nx = x - 1;