x86/mm: Provide support for multiple frame buffers in HVM guests.

Support is provided for both shadow and hardware assisted paging (HAP)
modes. This code bookkeeps the set of video frame buffers (vram),
detects when the guest has modified any of those buffers and, upon request,
returns a bitmap of the modified pages.
This lets other software components re-paint the portions of the monitor
(or monitors) that have changed.
Each monitor has a frame buffer of some size at some position
in guest physical memory.
The set of frame buffers being tracked can change over time as monitors
are plugged and unplugged.

Signed-off-by: Robert Phillips <robert.phillips@citrix.com>
Acked-by: Tim Deegan <tim@xen.org>

Removed a stray #include and a few hard tabs.

Signed-off-by: Tim Deegan <tim@xen.org>

Committed-by: Tim Deegan <tim@xen.org>

9 files changed, 1193 insertions, 368 deletions

diff --git a/xen/arch/x86/mm/Makefile b/xen/arch/x86/mm/Makefile
index 73dcdf4cb1..becd0c9bea 100644
--- a/xen/arch/x86/mm/Makefile
+++ b/xen/arch/x86/mm/Makefile
@@ -5,6 +5,7 @@ obj-y += paging.o
 obj-y += p2m.o p2m-pt.o p2m-ept.o p2m-pod.o
 obj-y += guest_walk_2.o
 obj-y += guest_walk_3.o
+obj-y += dirty_vram.o
 obj-$(x86_64) += guest_walk_4.o
 obj-$(x86_64) += mem_event.o
 obj-$(x86_64) += mem_paging.o
diff --git a/xen/arch/x86/mm/dirty_vram.c b/xen/arch/x86/mm/dirty_vram.c
new file mode 100644
index 0000000000..68e39731ba
--- /dev/null
+++ b/xen/arch/x86/mm/dirty_vram.c
@@ -0,0 +1,864 @@
+/*
+ * arch/x86/mm/dirty_vram.c: Bookkeep/query dirty VRAM pages
+ * with support for multiple frame buffers.
+ *
+ * Copyright (c) 2012, Citrix Systems, Inc. (Robert Phillips)
+ * Parts of this code are Copyright (c) 2007 Advanced Micro Devices (Wei Huang)
+ * Parts of this code are Copyright (c) 2007 XenSource Inc.
+ * Parts of this code are Copyright (c) 2006 by Michael A Fetterman
+ * Parts based on earlier work by Michael A Fetterman, Ian Pratt et al.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+ * Place - Suite 330, Boston, MA 02111-1307 USA.
+ */
+
+
+#include <xen/types.h>
+#include <xen/sched.h>
+#include <xen/guest_access.h>
+#include <asm/shadow.h>
+#include <asm/dirty_vram.h>
+#include <asm/hap.h>
+#include <asm/config.h>
+#include "mm-locks.h"
+
+#define DEBUG_stop_tracking_all_vram          0
+#define DEBUG_allocating_dirty_vram_range     0
+#define DEBUG_high_water_mark_for_vram_ranges 0
+#define DEBUG_freeing_dirty_vram_range        0
+#define DEBUG_allocate_paddr_links_page       0
+#define DEBUG_update_vram_mapping             0
+#define DEBUG_alloc_paddr_inject_fault        0
+#define DEBUG_link_limit_exceeded             0
+
+/* Allocates domain's dirty_vram structure */
+dv_dirty_vram_t *
+dirty_vram_alloc(struct domain *d)
+{
+    dv_dirty_vram_t *dirty_vram;
+    ASSERT( paging_locked_by_me(d) );
+    dirty_vram = d->arch.hvm_domain.dirty_vram = xzalloc(dv_dirty_vram_t);
+    if ( dirty_vram )
+    {
+        INIT_LIST_HEAD(&dirty_vram->range_head);
+        INIT_LIST_HEAD(&dirty_vram->ext_head);
+    }
+    return dirty_vram;
+}
+
+/*
+ * Returns domain's dirty_vram structure,
+ * allocating it if necessary
+ */
+dv_dirty_vram_t *
+dirty_vram_find_or_alloc(struct domain *d)
+{
+    dv_dirty_vram_t *dirty_vram = d->arch.hvm_domain.dirty_vram;
+    ASSERT( paging_locked_by_me(d) );
+    if ( !dirty_vram )
+        dirty_vram = dirty_vram_alloc(d);
+    return dirty_vram;
+}
+
+
+/* Free domain's dirty_vram structure */
+void dirty_vram_free(struct domain *d)
+{
+    dv_dirty_vram_t *dirty_vram = d->arch.hvm_domain.dirty_vram;
+    ASSERT( paging_locked_by_me(d) );
+    if ( dirty_vram )
+    {
+        struct list_head *curr, *next;
+        /* Free all the ranges */
+        list_for_each_safe(curr, next, &dirty_vram->range_head)
+        {
+            dv_range_t *range = list_entry(curr, dv_range_t, range_link);
+#if DEBUG_stop_tracking_all_vram
+            gdprintk(XENLOG_DEBUG, "[%05lx:%05lx] stop tracking all vram\n",
+                     range->begin_pfn, range->end_pfn);
+#endif
+            xfree(range->pl_tab);
+            xfree(range);
+        }
+        /* Free all the extension pages */
+        list_for_each_safe(curr, next, &dirty_vram->ext_head)
+        {
+            struct dv_paddr_link_ext *ext =
+                container_of(
+                    curr, struct dv_paddr_link_ext, ext_link);
+            struct page_info *pg = __virt_to_page(ext);
+            d->arch.paging.free_page(d, pg);
+        }
+
+        xfree(dirty_vram);
+        d->arch.hvm_domain.dirty_vram = NULL;
+    }
+}
+
+/* Returns dirty vram range containing gfn, NULL if none */
+struct dv_range *
+dirty_vram_range_find_gfn(struct domain *d,
+                          unsigned long gfn)
+{
+    struct dv_dirty_vram *dirty_vram = d->arch.hvm_domain.dirty_vram;
+    ASSERT( paging_locked_by_me(d) );
+    if ( dirty_vram )
+    {
+        struct list_head *curr;
+        list_for_each(curr, &dirty_vram->range_head)
+        {
+            dv_range_t *range = list_entry(curr, dv_range_t, range_link);
+            if ( gfn >= range->begin_pfn &&
+                 gfn <  range->end_pfn )
+                return range;
+        }
+    }
+    return NULL;
+}
+
+/*
+ * Returns pointer to dirty vram range matching [begin_pfn .. end_pfn ),
+ * NULL if none.
+ */
+dv_range_t *
+dirty_vram_range_find(struct domain *d,
+                      unsigned long begin_pfn,
+                      unsigned long nr)
+{
+    unsigned long end_pfn = begin_pfn + nr;
+    dv_dirty_vram_t *dirty_vram = d->arch.hvm_domain.dirty_vram;
+    ASSERT( paging_locked_by_me(d) );
+    if ( dirty_vram )
+    {
+        struct list_head *curr;
+        list_for_each(curr, &dirty_vram->range_head)
+        {
+            dv_range_t *range = list_entry(curr, dv_range_t, range_link);
+            if ( begin_pfn == range->begin_pfn &&
+                 end_pfn   == range->end_pfn )
+                return range;
+        }
+    }
+    return NULL;
+}
+
+/* Allocate specified dirty_vram range */
+static dv_range_t *
+_dirty_vram_range_alloc(struct domain *d,
+                        unsigned long begin_pfn,
+                        unsigned long nr)
+{
+    dv_dirty_vram_t *dirty_vram = d->arch.hvm_domain.dirty_vram;
+    dv_range_t *range = NULL;
+    unsigned long end_pfn = begin_pfn + nr;
+    dv_pl_entry_t *pl_tab = NULL;
+    int i;
+
+    ASSERT( paging_locked_by_me(d) );
+    ASSERT( dirty_vram != NULL );
+
+#if DEBUG_allocating_dirty_vram_range
+    gdprintk(XENLOG_DEBUG,
+             "[%05lx:%05lx] Allocating dirty vram range hap:%d\n",
+             begin_pfn, end_pfn,
+             d->arch.hvm_domain.hap_enabled);
+#endif
+
+    range = xzalloc(dv_range_t);
+    if ( range == NULL )
+        goto err_out;
+
+    INIT_LIST_HEAD(&range->range_link);
+
+    range->begin_pfn = begin_pfn;
+    range->end_pfn = end_pfn;
+
+    if ( !hap_enabled(d) )
+    {
+        if ( (pl_tab = xzalloc_array(dv_pl_entry_t, nr)) == NULL )
+            goto err_out;
+
+        for ( i = 0; i != nr; i++ )
+        {
+            pl_tab[i].mapping.sl1ma = INVALID_PADDR;
+        }
+    }
+
+    range->pl_tab = pl_tab;
+    range->mappings_hwm = 1;
+
+    list_add(&range->range_link, &dirty_vram->range_head);
+    if ( ++dirty_vram->nr_ranges > dirty_vram->ranges_hwm )
+    {
+        dirty_vram->ranges_hwm = dirty_vram->nr_ranges;
+#if DEBUG_high_water_mark_for_vram_ranges
+        gdprintk(XENLOG_DEBUG,
+                 "High water mark for number of vram ranges is now:%d\n",
+                 dirty_vram->ranges_hwm);
+#endif
+    }
+    return range;
+
+ err_out:
+    xfree(pl_tab);
+    xfree(range);
+    return NULL;
+}
+
+
+/* Frees specified dirty_vram range */
+void dirty_vram_range_free(struct domain *d,
+                           dv_range_t *range)
+{
+    dv_dirty_vram_t *dirty_vram = d->arch.hvm_domain.dirty_vram;
+    ASSERT( paging_locked_by_me(d) );
+    if ( dirty_vram )
+    {
+        int i, nr = range->end_pfn - range->begin_pfn;
+
+#if DEBUG_freeing_dirty_vram_range
+        gdprintk(XENLOG_DEBUG,
+                 "[%05lx:%05lx] Freeing dirty vram range\n",
+                 range->begin_pfn, range->end_pfn);
+#endif
+
+        if ( range->pl_tab )
+        {
+            for ( i = 0; i != nr; i++ )
+            {
+                dv_paddr_link_t *plx;
+                plx = range->pl_tab[i].mapping.pl_next;
+                /* Does current FB page have multiple mappings? */
+                if ( plx ) /* yes */
+                {
+                    /* Find the last element in singly-linked list */
+                    while ( plx->pl_next != NULL )
+                        plx = plx->pl_next;
+                    
+                    /* Prepend whole list to the free list */
+                    plx->pl_next = dirty_vram->pl_free;
+                    dirty_vram->pl_free = range->pl_tab[i].mapping.pl_next;
+                }
+            }
+            xfree(range->pl_tab);
+            range->pl_tab = NULL;
+        }
+
+        /* Remove range from the linked list, free it, and adjust count*/
+        list_del(&range->range_link);
+        xfree(range);
+        dirty_vram->nr_ranges--;
+    }
+}
+
+/*
+ * dirty_vram_range_alloc()
+ * This function ensures that the new range does not overlap any existing
+ * ranges -- deleting them if necessary -- and then calls
+ * _dirty_vram_range_alloc to actually allocate the new range.
+ */
+dv_range_t *
+dirty_vram_range_alloc(struct domain *d,
+                        unsigned long begin_pfn,
+                        unsigned long nr)
+{
+    unsigned long end_pfn = begin_pfn + nr;
+    dv_dirty_vram_t *dirty_vram = d->arch.hvm_domain.dirty_vram;
+    dv_range_t *range;
+    struct list_head *curr, *next;
+
+    ASSERT( paging_locked_by_me(d) );
+    ASSERT( dirty_vram != NULL );
+
+    /*
+     * Ranges cannot overlap so
+     * free any range that overlaps [ begin_pfn .. end_pfn )
+     */
+    list_for_each_safe(curr, next, &dirty_vram->range_head)
+    {
+        dv_range_t *rng = list_entry(curr, dv_range_t, range_link);
+        if ( ( ( rng->begin_pfn <= begin_pfn ) &&
+               ( begin_pfn <  rng->end_pfn   )
+                 ) ||
+             ( ( begin_pfn <= rng->begin_pfn ) &&
+               ( rng->begin_pfn < end_pfn    )
+                 ) )
+        {
+            /* Different tracking, tear the previous down. */
+            dirty_vram_range_free(d, rng);
+        }
+    }
+
+    range = _dirty_vram_range_alloc(d, begin_pfn, nr);
+    if ( !range )
+        goto out;
+
+ out:
+    return range;
+}
+
+/*
+ * dirty_vram_range_find_or_alloc()
+ * Find the range for [begin_pfn:begin_pfn+nr).
+ * If it doesn't exists, create it.
+ */
+dv_range_t *
+dirty_vram_range_find_or_alloc(struct domain *d,
+                                unsigned long begin_pfn,
+                                unsigned long nr)
+{
+    dv_range_t *range;
+    ASSERT( paging_locked_by_me(d) );
+    range = dirty_vram_range_find(d, begin_pfn, nr);
+    if ( !range )
+        range = dirty_vram_range_alloc(d, begin_pfn, nr);
+    
+    return range;
+}
+
+
+
+/* Allocate a dv_paddr_link struct */
+static dv_paddr_link_t *
+alloc_paddr_link(struct domain *d)
+{
+    dv_paddr_link_t * pl = NULL;
+    dv_dirty_vram_t *dirty_vram = d->arch.hvm_domain.dirty_vram;
+    dv_paddr_link_ext_t *ext = NULL;
+    
+
+    ASSERT( paging_locked_by_me(d) );
+    BUILD_BUG_ON(sizeof(dv_paddr_link_ext_t) > PAGE_SIZE);
+    /* Is the list of free pl's empty? */
+    if ( dirty_vram->pl_free == NULL ) /* yes */
+    {
+        /*
+         * Allocate another page of pl's.
+         * Link them all together and point the free list head at them
+         */
+        int i;
+        struct page_info *pg = d->arch.paging.alloc_page(d);
+
+        ext = map_domain_page(pg);
+        if ( ext == NULL )
+            goto out;
+
+#if DEBUG_allocate_paddr_links_page
+        gdprintk(XENLOG_DEBUG, "Allocated another page of paddr_links\n");
+#endif
+        list_add(&ext->ext_link, &dirty_vram->ext_head);
+
+        /* initialize and link together the new pl entries */
+        for ( i = 0; i != ARRAY_SIZE(ext->entries); i++ )
+        {
+            ext->entries[i].sl1ma = INVALID_PADDR;
+            ext->entries[i].pl_next = &ext->entries[i+1];
+        }
+        ext->entries[ARRAY_SIZE(ext->entries) - 1].pl_next = NULL;
+        dirty_vram->pl_free = &ext->entries[0];
+    }
+    pl = dirty_vram->pl_free;
+    dirty_vram->pl_free = pl->pl_next;
+
+    pl->sl1ma = INVALID_PADDR;
+    pl->pl_next = NULL;
+ out:
+    if ( ext )
+        unmap_domain_page(ext);
+    
+    return pl;
+}
+
+
+/*
+ * Free a paddr_link struct.
+ *
+ * The caller has walked the singly-linked list of elements
+ * that have, as their head, an element in a pl_tab cell.
+ * The list walks has reached the element to be freed.
+ * (Each element is a dv_paddr_link_t struct.)
+ *
+ * @pl points to the element to be freed.
+ * @ppl points to its predecessor element's next member.
+ *
+ * After linking the precessor to the element's successor,
+ * we can free @pl by prepending it to the list of free
+ * elements.
+ *
+ * As a boundary case (which happens to be the common case),
+ * @pl points to a cell in the pl_tab rather than to some
+ * extension element danging from that cell.
+ * We recognize this case because @ppl is NULL.
+ * In that case we promote the first extension element by
+ * copying it into the pl_tab cell and free it.
+ */
+
+dv_paddr_link_t *
+free_paddr_link(struct domain *d,
+                dv_paddr_link_t **ppl,
+                dv_paddr_link_t *pl)
+{
+    dv_dirty_vram_t *dirty_vram = d->arch.hvm_domain.dirty_vram;
+    dv_paddr_link_t *npl; /* next pl */
+
+    ASSERT( paging_locked_by_me(d) );
+    /* extension mapping? */
+    if ( ppl ) /* yes. free it */
+    {
+        ASSERT(pl == (*ppl));
+        (*ppl) = npl = pl->pl_next;
+    }
+    else  /* main table */
+    {
+        /*
+         * move 2nd mapping to main table.
+         * and free 2nd mapping
+         */
+        dv_paddr_link_t * spl;
+        spl = pl->pl_next;
+        if ( spl == NULL )
+        {
+            pl->sl1ma = INVALID_PADDR;
+            return pl;
+        }
+        pl->sl1ma = spl->sl1ma;
+        pl->pl_next = spl->pl_next;
+        npl = pl; /* reprocess main table entry again */
+        pl = spl;
+    }
+    pl->sl1ma = INVALID_PADDR;
+    pl->pl_next = dirty_vram->pl_free;
+    dirty_vram->pl_free = pl;
+    return npl;
+}
+
+
+/*
+ * dirty_vram_range_update()
+ *
+ * This is called whenever a level 1 page table entry is modified.
+ * If the L1PTE is being cleared, the function removes any paddr_links
+ * that refer to it.
+ * If the L1PTE is being set to a frame buffer page, a paddr_link is
+ * created for that page's entry in pl_tab.
+ * Returns 1 iff entry found and set or cleared.
+ */
+int dirty_vram_range_update(struct domain *d,
+                            unsigned long gfn,
+                            paddr_t sl1ma,
+                            int set)
+{
+    int effective = 0;
+    dv_range_t *range;
+    unsigned long i;
+    dv_paddr_link_t *pl;
+    dv_paddr_link_t **ppl;
+    int len = 0;
+
+    ASSERT(paging_locked_by_me(d));
+    range = dirty_vram_range_find_gfn(d, gfn);
+    if ( !range )
+        return effective;
+
+    
+    i = gfn - range->begin_pfn;
+    pl = &range->pl_tab[ i ].mapping;
+    ppl = NULL;
+
+    /*
+     * find matching entry (pl), if any, and its predecessor
+     * in linked list (ppl)
+     */
+    while ( pl != NULL )
+    {
+        if ( pl->sl1ma == sl1ma || pl->sl1ma == INVALID_PADDR )
+            break;
+            
+        ppl = &pl->pl_next;
+        pl = *ppl;
+        len++;
+    }
+
+    if ( set )
+    {
+        /* Did we find sl1ma in either the main table or the linked list? */
+        if ( pl == NULL ) /* no, so we'll need to alloc a link */
+        {
+            ASSERT(ppl != NULL);
+            
+#if DEBUG_alloc_paddr_inject_fault
+            {
+                static int counter;
+                
+                /* Test stuck_dirty logic for some cases */
+                if ( (++counter) % 4 == 0 )
+                {
+                    /* Simply mark the frame buffer page as always dirty */
+                    range->pl_tab[ i ].stuck_dirty = 1;
+                    gdprintk(XENLOG_DEBUG,
+                             "[%lx] inject stuck dirty fault\n",
+                             gfn );
+                    goto out;
+                }
+            }
+#endif
+            /*
+             * Have we reached the limit of mappings we're willing
+             * to bookkeep?
+             */
+            if ( len > DV_ADDR_LINK_LIST_LIMIT ) /* yes */
+            {
+#if DEBUG_link_limit_exceeded
+                if ( !range->pl_tab[ i ].stuck_dirty )
+                    gdprintk(XENLOG_DEBUG,
+                             "[%lx] link limit exceeded\n",
+                             gfn );
+#endif            
+                /* Simply mark the frame buffer page as always dirty */
+                range->pl_tab[ i ].stuck_dirty = 1;
+                goto out;
+            }
+
+            /* alloc link and append it to list */
+            (*ppl) = pl = alloc_paddr_link(d);
+            /* Were we able to allocate a link? */
+            if ( pl == NULL ) /* no */
+            {
+                /* Simply mark the frame buffer page as always dirty */
+                range->pl_tab[ i ].stuck_dirty = 1;
+                
+                gdprintk(XENLOG_DEBUG,
+                         "[%lx] alloc failure\n",
+                         gfn );
+                
+                goto out;
+            }
+        }
+        if ( pl->sl1ma != sl1ma )
+        {
+            ASSERT(pl->sl1ma == INVALID_PADDR);
+            pl->sl1ma = sl1ma;
+            range->nr_mappings++;
+        }
+        effective = 1;
+        if ( len > range->mappings_hwm )
+        {
+            range->mappings_hwm = len;
+#if DEBUG_update_vram_mapping
+            gdprintk(XENLOG_DEBUG,
+                     "[%lx] set      sl1ma:%lx hwm:%d mappings:%d "
+                     "freepages:%d\n",
+                     gfn, sl1ma,
+                     range->mappings_hwm,
+                     range->nr_mappings,
+                     d->arch.paging.shadow.free_pages);
+#endif
+        }
+    }
+    else /* clear */
+    {
+        if ( pl && pl->sl1ma == sl1ma )
+        {
+#if DEBUG_update_vram_mapping
+            gdprintk(XENLOG_DEBUG,
+                     "[%lx] clear    sl1ma:%lx mappings:%d\n",
+                     gfn, sl1ma,
+                     range->nr_mappings - 1);
+#endif
+            free_paddr_link(d, ppl, pl);
+            --range->nr_mappings;
+            effective = 1;
+        }
+    }
+ out:
+    return effective;
+}
+
+
+/*
+ * shadow_scan_dirty_flags()
+ * This produces a dirty bitmap for the range by examining every
+ * L1PTE referenced by some dv_paddr_link in the range's pl_tab table.
+ * It tests and clears each such L1PTE's dirty flag.
+ */
+static int shadow_scan_dirty_flags(struct domain *d,
+                                   dv_range_t *range,
+                                   uint8_t *dirty_bitmap)
+{
+    int flush_tlb = 0;
+    unsigned long i;
+    unsigned long nr = range->end_pfn - range->begin_pfn;
+    l1_pgentry_t *sl1e = NULL;
+
+    ASSERT( paging_locked_by_me(d) );
+    /* Iterate over VRAM to track dirty bits. */
+    for ( i = 0; i < nr; i++ )
+    {
+        int dirty = 0, len = 1;
+        dv_paddr_link_t *pl;
+        /* Does the frame buffer have an incomplete set of mappings? */
+        if ( unlikely(range->pl_tab[i].stuck_dirty) ) /* yes */
+            dirty = 1;
+        else /* The frame buffer's set of mappings is complete.  Scan it. */
+            for ( pl = &range->pl_tab[i].mapping;
+                  pl;
+                  pl = pl->pl_next, len++ )
+            {
+                paddr_t sl1ma = pl->sl1ma;
+                if ( sl1ma == INVALID_PADDR ) /* FB page is unmapped */
+                    continue;
+
+                if ( sl1e ) /* cleanup from previous iteration */
+                    unmap_domain_page(sl1e);
+
+                sl1e = map_domain_page(sl1ma >> PAGE_SHIFT);
+                if ( l1e_get_flags(*sl1e) & _PAGE_DIRTY )
+                {
+                    dirty = 1;
+                    /* Clear dirty so we can detect if page gets re-dirtied.
+                     * Note: this is atomic, so we may clear a
+                     * _PAGE_ACCESSED set by another processor.
+                     */
+                    l1e_remove_flags(*sl1e, _PAGE_DIRTY);
+                    flush_tlb = 1;
+                }
+            } /* for */
+        
+        if ( dirty )
+            dirty_bitmap[i >> 3] |= (1 << (i & 7));
+
+    }
+    
+    if ( sl1e )
+        unmap_domain_page(sl1e);
+
+    return flush_tlb;
+}
+
+
+/*
+ * shadow_track_dirty_vram()
+ * This is the API called by the guest to determine which pages in the range
+ * from [begin_pfn:begin_pfn+nr) have been dirtied since the last call.
+ * It creates the domain's dv_dirty_vram on demand.
+ * It creates ranges on demand when some [begin_pfn:nr) is first encountered.
+ * To collect the dirty bitmask it calls shadow_scan_dirty_flags().
+ * It copies the dirty bitmask into guest storage.
+ */
+int shadow_track_dirty_vram(struct domain *d,
+                            unsigned long begin_pfn,
+                            unsigned long nr,
+                            XEN_GUEST_HANDLE_64(uint8) guest_dirty_bitmap)
+{
+    int rc = 0;
+    unsigned long end_pfn = begin_pfn + nr;
+    int flush_tlb = 0;
+    dv_range_t *range;
+    struct p2m_domain *p2m = p2m_get_hostp2m(d);
+    uint8_t *dirty_bitmap = NULL;
+
+    /*
+     * This range test is tricky.
+     *
+     * The range [begin_pfn..end_pfn) is an open interval, so end_pfn
+     * is a pfn beyond the end of the range.
+     *
+     * p2m->max_mapped_pfn is a valid PFN so p2m->max_mapped_pfn + 1 is an
+     * invalid PFN.
+     *
+     * If end_pfn is beyond *that* then the range is invalid.
+     */
+    if ( end_pfn < begin_pfn
+         || begin_pfn > p2m->max_mapped_pfn
+         || end_pfn > p2m->max_mapped_pfn + 1 )
+        return -EINVAL;
+
+    paging_lock(d);
+
+    if ( !nr )
+    {
+        dirty_vram_free(d);
+        goto out;
+    }
+
+    if ( guest_handle_is_null(guest_dirty_bitmap) )
+        goto out;
+
+    if ( !dirty_vram_find_or_alloc(d) )
+    {
+        rc = -ENOMEM;
+        goto out;
+    }
+
+    range = dirty_vram_range_find(d, begin_pfn, nr);
+    if ( !range )
+    {
+        range = dirty_vram_range_alloc(d, begin_pfn, nr);
+        if ( range )
+            sh_find_all_vram_mappings(d->vcpu[0], range);
+    }
+    if ( range )
+    {
+        int size = ( nr + BITS_PER_BYTE - 1 ) / BITS_PER_BYTE;
+        
+        rc = -ENOMEM;
+        dirty_bitmap = xzalloc_bytes( size );
+        if ( !dirty_bitmap )
+            goto out;
+
+        flush_tlb |= shadow_scan_dirty_flags(d, range, dirty_bitmap);
+
+        rc = -EFAULT;
+        if ( copy_to_guest(guest_dirty_bitmap,
+                           dirty_bitmap,
+                           size) == 0 )
+            rc = 0;
+    }
+    
+    if ( flush_tlb )
+        flush_tlb_mask(d->domain_dirty_cpumask);
+
+out:
+    paging_unlock(d);
+    
+    if ( dirty_bitmap )
+        xfree(dirty_bitmap);
+    return rc;
+}
+
+
+/************************************************/
+/*          HAP VRAM TRACKING SUPPORT           */
+/************************************************/
+
+/*
+ * hap_track_dirty_vram()
+ * Create the domain's dv_dirty_vram struct on demand.
+ * Create a dirty vram range on demand when some [begin_pfn:begin_pfn+nr] is
+ * first encountered.
+ * Collect the guest_dirty bitmask, a bit mask of the dirty vram pages, by
+ * calling paging_log_dirty_range(), which interrogates each vram
+ * page's p2m type looking for pages that have been made writable.
+ */
+int hap_track_dirty_vram(struct domain *d,
+                         unsigned long begin_pfn,
+                         unsigned long nr,
+                         XEN_GUEST_HANDLE_64(uint8) guest_dirty_bitmap)
+{
+    long rc = 0;
+    dv_dirty_vram_t *dirty_vram;
+    uint8_t *dirty_bitmap = NULL;
+
+    if ( nr )
+    {
+        dv_range_t *range = NULL;
+        int size = ( nr + BITS_PER_BYTE - 1 ) / BITS_PER_BYTE;
+        
+        if ( !paging_mode_log_dirty(d) )
+        {
+            hap_logdirty_init(d);
+            rc = paging_log_dirty_enable(d);
+            if ( rc )
+                goto out;
+        }
+
+        rc = -ENOMEM;
+        dirty_bitmap = xzalloc_bytes( size );
+        if ( !dirty_bitmap )
+            goto out;
+        
+        paging_lock(d);
+        
+        dirty_vram = d->arch.hvm_domain.dirty_vram;
+        if ( !dirty_vram ) 
+        {
+            rc = -ENOMEM;
+            if ( !(dirty_vram = dirty_vram_alloc(d)) )
+            {
+                paging_unlock(d);
+                goto out;
+            }
+        }
+        
+        range = dirty_vram_range_find(d, begin_pfn, nr);
+        if ( !range )
+        {
+            rc = -ENOMEM;
+            if ( !(range = dirty_vram_range_alloc(d, begin_pfn, nr)) )
+            {
+                paging_unlock(d);
+                goto out;
+            }
+            
+            paging_unlock(d);
+            
+            /* set l1e entries of range within P2M table to be read-only. */
+            p2m_change_type_range(d, begin_pfn, begin_pfn + nr,
+                                  p2m_ram_rw, p2m_ram_logdirty);
+            
+            flush_tlb_mask(d->domain_dirty_cpumask);
+            
+            memset(dirty_bitmap, 0xff, size); /* consider all pages dirty */
+        }
+        else
+        {
+            paging_unlock(d);
+            
+            domain_pause(d);
+            
+            /* get the bitmap */
+            paging_log_dirty_range(d, begin_pfn, nr, dirty_bitmap);
+            
+            domain_unpause(d);
+        }
+        
+        
+        rc = -EFAULT;
+        if ( copy_to_guest(guest_dirty_bitmap,
+                           dirty_bitmap,
+                           size) == 0 )
+        {
+            rc = 0;
+        }
+    }
+    else {
+        paging_lock(d);
+        
+        dirty_vram = d->arch.hvm_domain.dirty_vram;
+        if ( dirty_vram )
+        {
+            /*
+             * If zero pages specified while tracking dirty vram
+             * then stop tracking
+             */
+            dirty_vram_free(d);
+        
+        }
+        
+        paging_unlock(d);
+    }
+out:
+    if ( dirty_bitmap )
+        xfree(dirty_bitmap);
+    
+    return rc;
+}
+
+/*
+ * Local variables:
+ * mode: C
+ * c-file-style: "BSD"
+ * c-basic-offset: 4
+ * indent-tabs-mode: nil
+ * End:
+ */
diff --git a/xen/arch/x86/mm/hap/hap.c b/xen/arch/x86/mm/hap/hap.c
index a95ccbf62c..f7d979bcdc 100644
--- a/xen/arch/x86/mm/hap/hap.c
+++ b/xen/arch/x86/mm/hap/hap.c
@@ -53,117 +53,6 @@
 #define page_to_mfn(_pg) _mfn(__page_to_mfn(_pg))
 
 /************************************************/
-/*          HAP VRAM TRACKING SUPPORT           */
-/************************************************/
-
-/*
- * hap_track_dirty_vram()
- * Create the domain's dv_dirty_vram struct on demand.
- * Create a dirty vram range on demand when some [begin_pfn:begin_pfn+nr] is
- * first encountered.
- * Collect the guest_dirty bitmask, a bit mask of the dirty vram pages, by
- * calling paging_log_dirty_range(), which interrogates each vram
- * page's p2m type looking for pages that have been made writable.
- */
-
-int hap_track_dirty_vram(struct domain *d,
-                         unsigned long begin_pfn,
-                         unsigned long nr,
-                         XEN_GUEST_HANDLE_64(uint8) guest_dirty_bitmap)
-{
-    long rc = 0;
-    struct sh_dirty_vram *dirty_vram;
-    uint8_t *dirty_bitmap = NULL;
-
-    if ( nr )
-    {
-        int size = (nr + BITS_PER_BYTE - 1) / BITS_PER_BYTE;
-
-        if ( !paging_mode_log_dirty(d) )
-        {
-            hap_logdirty_init(d);
-            rc = paging_log_dirty_enable(d);
-            if ( rc )
-                goto out;
-        }
-
-        rc = -ENOMEM;
-        dirty_bitmap = xzalloc_bytes(size);
-        if ( !dirty_bitmap )
-            goto out;
-
-        paging_lock(d);
-
-        dirty_vram = d->arch.hvm_domain.dirty_vram;
-        if ( !dirty_vram )
-        {
-            rc = -ENOMEM;
-            if ( (dirty_vram = xzalloc(struct sh_dirty_vram)) == NULL )
-            {
-                paging_unlock(d);
-                goto out;
-            }
-
-            d->arch.hvm_domain.dirty_vram = dirty_vram;
-        }
-
-        if ( begin_pfn != dirty_vram->begin_pfn ||
-             begin_pfn + nr != dirty_vram->end_pfn )
-        {
-            dirty_vram->begin_pfn = begin_pfn;
-            dirty_vram->end_pfn = begin_pfn + nr;
-
-            paging_unlock(d);
-
-            /* set l1e entries of range within P2M table to be read-only. */
-            p2m_change_type_range(d, begin_pfn, begin_pfn + nr,
-                                  p2m_ram_rw, p2m_ram_logdirty);
-
-            flush_tlb_mask(d->domain_dirty_cpumask);
-
-            memset(dirty_bitmap, 0xff, size); /* consider all pages dirty */
-        }
-        else
-        {
-            paging_unlock(d);
-
-            domain_pause(d);
-
-            /* get the bitmap */
-            paging_log_dirty_range(d, begin_pfn, nr, dirty_bitmap);
-
-            domain_unpause(d);
-        }
-
-        rc = -EFAULT;
-        if ( copy_to_guest(guest_dirty_bitmap, dirty_bitmap, size) == 0 )
-            rc = 0;
-    }
-    else
-    {
-        paging_lock(d);
-
-        dirty_vram = d->arch.hvm_domain.dirty_vram;
-        if ( dirty_vram )
-        {
-            /*
-             * If zero pages specified while tracking dirty vram
-             * then stop tracking
-             */
-            xfree(dirty_vram);
-            d->arch.hvm_domain.dirty_vram = NULL;
-        }
-
-        paging_unlock(d);
-    }
-out:
-    if ( dirty_bitmap )
-        xfree(dirty_bitmap);
-
-    return rc;
-}
-
-/************************************************/
 /*            HAP LOG DIRTY SUPPORT             */
 /************************************************/
 
diff --git a/xen/arch/x86/mm/p2m.c b/xen/arch/x86/mm/p2m.c
index de1dd822ea..6f638a27d1 100644
--- a/xen/arch/x86/mm/p2m.c
+++ b/xen/arch/x86/mm/p2m.c
@@ -741,20 +741,23 @@ void p2m_change_type_range(struct domain *d,
     struct p2m_domain *p2m = p2m_get_hostp2m(d);
 
     BUG_ON(p2m_is_grant(ot) || p2m_is_grant(nt));
-
     p2m_lock(p2m);
-    p2m->defer_nested_flush = 1;
 
+    p2m->defer_nested_flush = 1;
+    
     for ( gfn = start; gfn < end; gfn++ )
     {
         mfn = p2m->get_entry(p2m, gfn, &pt, &a, 0, NULL);
         if ( pt == ot )
-            set_p2m_entry(p2m, gfn, mfn, PAGE_ORDER_4K, nt, p2m->default_access);
+            set_p2m_entry(p2m, gfn, mfn, PAGE_ORDER_4K, nt,
+                          p2m->default_access);
     }
-
+    
     p2m->defer_nested_flush = 0;
+
     if ( nestedhvm_enabled(d) )
         p2m_flush_nestedp2m(d);
+
     p2m_unlock(p2m);
 }
 
diff --git a/xen/arch/x86/mm/paging.c b/xen/arch/x86/mm/paging.c
index ac9bb1a0ed..d59e8d6523 100644
--- a/xen/arch/x86/mm/paging.c
+++ b/xen/arch/x86/mm/paging.c
@@ -27,6 +27,7 @@
 #include <asm/p2m.h>
 #include <asm/hap.h>
 #include <asm/hvm/nestedhvm.h>
+#include <asm/dirty_vram.h>
 #include <xen/numa.h>
 #include <xsm/xsm.h>
 
@@ -192,15 +193,11 @@ int paging_log_dirty_disable(struct domain *d)
     return ret;
 }
 
-/* Mark a page as dirty */
+/* Given a guest mfn, mark a page as dirty */
 void paging_mark_dirty(struct domain *d, unsigned long guest_mfn)
 {
     unsigned long pfn;
     mfn_t gmfn;
-    int changed;
-    mfn_t mfn, *l4, *l3, *l2;
-    unsigned long *l1;
-    int i1, i2, i3, i4;
 
     gmfn = _mfn(guest_mfn);
 
@@ -210,6 +207,19 @@ void paging_mark_dirty(struct domain *d, unsigned long guest_mfn)
 
     /* We /really/ mean PFN here, even for non-translated guests. */
     pfn = get_gpfn_from_mfn(mfn_x(gmfn));
+    paging_mark_dirty_gpfn(d, pfn);
+}
+
+
+/* Given a guest pfn, mark a page as dirty */
+void paging_mark_dirty_gpfn(struct domain *d, unsigned long pfn)
+{
+    int changed;
+    mfn_t mfn, *l4, *l3, *l2;
+    unsigned long *l1;
+    int i1, i2, i3, i4;
+    dv_range_t *range;
+    
     /* Shared MFNs should NEVER be marked dirty */
     BUG_ON(SHARED_M2P(pfn));
 
@@ -229,6 +239,11 @@ void paging_mark_dirty(struct domain *d, unsigned long guest_mfn)
     /* Recursive: this is called from inside the shadow code */
     paging_lock_recursive(d);
 
+    d->arch.paging.log_dirty.dirty_count++;
+    range = dirty_vram_range_find_gfn(d, pfn);
+    if ( range )
+        range->dirty_count++;
+
     if ( unlikely(!mfn_valid(d->arch.paging.log_dirty.top)) ) 
     {
          d->arch.paging.log_dirty.top = paging_new_log_dirty_node(d);
@@ -445,7 +460,31 @@ void paging_log_dirty_range(struct domain *d,
     struct p2m_domain *p2m = p2m_get_hostp2m(d);
     int i;
     unsigned long pfn;
+    dv_range_t *range;
+    unsigned int range_dirty_count;
 
+    paging_lock(d);
+    range = dirty_vram_range_find_gfn(d, begin_pfn);
+    if ( !range )
+    {
+        paging_unlock(d);
+        goto out;
+    }
+    
+    range_dirty_count = range->dirty_count;
+    range->dirty_count = 0;
+
+    paging_unlock(d);
+    
+    if ( !range_dirty_count)
+        goto out;
+
+    PAGING_DEBUG(LOGDIRTY,
+                 "log-dirty-range: dom %u [%05lx:%05lx] range_dirty=%u\n",
+                 d->domain_id,
+                 begin_pfn,
+                 begin_pfn + nr,
+                 range_dirty_count);
     /*
      * Set l1e entries of P2M table to be read-only.
      *
@@ -460,15 +499,17 @@ void paging_log_dirty_range(struct domain *d,
 
     for ( i = 0, pfn = begin_pfn; pfn < begin_pfn + nr; i++, pfn++ )
     {
-        p2m_type_t pt;
-        pt = p2m_change_type(d, pfn, p2m_ram_rw, p2m_ram_logdirty);
-        if ( pt == p2m_ram_rw )
+        if ( p2m_change_type(d, pfn, p2m_ram_rw, p2m_ram_logdirty) ==
+             p2m_ram_rw )
             dirty_bitmap[i >> 3] |= (1 << (i & 7));
     }
 
     p2m_unlock(p2m);
 
     flush_tlb_mask(d->domain_dirty_cpumask);
+
+ out:
+    return;
 }
 
 /* Note that this function takes three function pointers. Callers must supply
diff --git a/xen/arch/x86/mm/shadow/common.c b/xen/arch/x86/mm/shadow/common.c
index ce79131e1f..1e4b880187 100644
--- a/xen/arch/x86/mm/shadow/common.c
+++ b/xen/arch/x86/mm/shadow/common.c
@@ -36,6 +36,7 @@
 #include <asm/current.h>
 #include <asm/flushtlb.h>
 #include <asm/shadow.h>
+#include <asm/dirty_vram.h>
 #include <xen/numa.h>
 #include "private.h"
 
@@ -3121,12 +3122,7 @@ void shadow_teardown(struct domain *d)
      * calls now that we've torn down the bitmap */
     d->arch.paging.mode &= ~PG_log_dirty;
 
-    if (d->arch.hvm_domain.dirty_vram) {
-        xfree(d->arch.hvm_domain.dirty_vram->sl1ma);
-        xfree(d->arch.hvm_domain.dirty_vram->dirty_bitmap);
-        xfree(d->arch.hvm_domain.dirty_vram);
-        d->arch.hvm_domain.dirty_vram = NULL;
-    }
+    dirty_vram_free(d);
 
     paging_unlock(d);
 
@@ -3464,178 +3460,217 @@ void shadow_clean_dirty_bitmap(struct domain *d)
 
 /**************************************************************************/
 /* VRAM dirty tracking support */
-int shadow_track_dirty_vram(struct domain *d,
-                            unsigned long begin_pfn,
-                            unsigned long nr,
-                            XEN_GUEST_HANDLE_64(uint8) dirty_bitmap)
-{
-    int rc;
-    unsigned long end_pfn = begin_pfn + nr;
-    unsigned long dirty_size = (nr + 7) / 8;
-    int flush_tlb = 0;
-    unsigned long i;
-    p2m_type_t t;
-    struct sh_dirty_vram *dirty_vram = d->arch.hvm_domain.dirty_vram;
-    struct p2m_domain *p2m = p2m_get_hostp2m(d);
 
-    if (end_pfn < begin_pfn
-            || begin_pfn > p2m->max_mapped_pfn
-            || end_pfn >= p2m->max_mapped_pfn)
-        return -EINVAL;
 
-    /* We perform p2m lookups, so lock the p2m upfront to avoid deadlock */
-    p2m_lock(p2m_get_hostp2m(d));
-    paging_lock(d);
+/* Support functions for shadow-based dirty VRAM code */
 
-    if ( dirty_vram && (!nr ||
-             ( begin_pfn != dirty_vram->begin_pfn
-            || end_pfn   != dirty_vram->end_pfn )) )
-    {
-        /* Different tracking, tear the previous down. */
-        gdprintk(XENLOG_INFO, "stopping tracking VRAM %lx - %lx\n", dirty_vram->begin_pfn, dirty_vram->end_pfn);
-        xfree(dirty_vram->sl1ma);
-        xfree(dirty_vram->dirty_bitmap);
-        xfree(dirty_vram);
-        dirty_vram = d->arch.hvm_domain.dirty_vram = NULL;
-    }
+#define DEBUG_unshadow_sl1ma                  0          
+#define DEBUG_unshadow_sl1ma_detail           0
+#define DEBUG_count_initial_mappings          0
 
-    if ( !nr )
+/* smfn is no longer a shadow page.  Remove it from any
+ * dirty vram range mapping. */
+void
+dirty_vram_delete_shadow(struct vcpu *v,
+                         unsigned long gfn,
+                         unsigned int shadow_type, 
+                         mfn_t smfn)
+{
+    static unsigned int l1_shadow_mask = 
+          1 << SH_type_l1_32_shadow
+        | 1 << SH_type_fl1_32_shadow
+        | 1 << SH_type_l1_pae_shadow
+        | 1 << SH_type_fl1_pae_shadow
+        | 1 << SH_type_l1_64_shadow
+        | 1 << SH_type_fl1_64_shadow
+        ;
+    struct domain *d = v->domain;
+    dv_dirty_vram_t *dirty_vram;
+    struct list_head *curr, *next;
+    
+    ASSERT(paging_locked_by_me(d));
+    /* Ignore all but level 1 shadows */
+    
+    if ( (l1_shadow_mask & (1 << shadow_type)) == 0 )
     {
-        rc = 0;
         goto out;
     }
 
-    /* This should happen seldomly (Video mode change),
-     * no need to be careful. */
+    dirty_vram = d->arch.hvm_domain.dirty_vram;
     if ( !dirty_vram )
     {
-        /* Throw away all the shadows rather than walking through them 
-         * up to nr times getting rid of mappings of each pfn */
-        shadow_blow_tables(d);
-
-        gdprintk(XENLOG_INFO, "tracking VRAM %lx - %lx\n", begin_pfn, end_pfn);
-
-        rc = -ENOMEM;
-        if ( (dirty_vram = xmalloc(struct sh_dirty_vram)) == NULL )
-            goto out;
-        dirty_vram->begin_pfn = begin_pfn;
-        dirty_vram->end_pfn = end_pfn;
-        d->arch.hvm_domain.dirty_vram = dirty_vram;
-
-        if ( (dirty_vram->sl1ma = xmalloc_array(paddr_t, nr)) == NULL )
-            goto out_dirty_vram;
-        memset(dirty_vram->sl1ma, ~0, sizeof(paddr_t) * nr);
-
-        if ( (dirty_vram->dirty_bitmap = xzalloc_array(uint8_t, dirty_size)) == NULL )
-            goto out_sl1ma;
-
-        dirty_vram->last_dirty = NOW();
-
-        /* Tell the caller that this time we could not track dirty bits. */
-        rc = -ENODATA;
-    }
-    else if (dirty_vram->last_dirty == -1)
-    {
-        /* still completely clean, just copy our empty bitmap */
-        rc = -EFAULT;
-        if ( copy_to_guest(dirty_bitmap, dirty_vram->dirty_bitmap, dirty_size) == 0 )
-            rc = 0;
+        goto out;
     }
-    else
+        
+    list_for_each_safe(curr, next, &dirty_vram->range_head)
     {
-        /* Iterate over VRAM to track dirty bits. */
-        for ( i = 0; i < nr; i++ ) {
-            mfn_t mfn = get_gfn_query_unlocked(d, begin_pfn + i, &t);
-            struct page_info *page;
-            int dirty = 0;
-            paddr_t sl1ma = dirty_vram->sl1ma[i];
-
-            if (mfn_x(mfn) == INVALID_MFN)
-            {
-                dirty = 1;
-            }
-            else
+        dv_range_t *range = list_entry(curr, dv_range_t, range_link);
+        unsigned long i;
+        int max_mappings = 1, mappings = 0;
+        int unshadowed = 0;
+        for ( i = 0; i != range->end_pfn - range->begin_pfn; i++ )
+        {
+            dv_paddr_link_t *pl = &range->pl_tab[ i ].mapping;
+            dv_paddr_link_t **ppl = NULL;
+            mappings = 0;
+            
+            while ( pl != NULL )
             {
-                page = mfn_to_page(mfn);
-                switch (page->u.inuse.type_info & PGT_count_mask)
-                {
-                case 0:
-                    /* No guest reference, nothing to track. */
-                    break;
-                case 1:
-                    /* One guest reference. */
-                    if ( sl1ma == INVALID_PADDR )
-                    {
-                        /* We don't know which sl1e points to this, too bad. */
-                        dirty = 1;
-                        /* TODO: Heuristics for finding the single mapping of
-                         * this gmfn */
-                        flush_tlb |= sh_remove_all_mappings(d->vcpu[0], mfn);
-                    }
-                    else
-                    {
-                        /* Hopefully the most common case: only one mapping,
-                         * whose dirty bit we can use. */
-                        l1_pgentry_t *sl1e = maddr_to_virt(sl1ma);
-
-                        if ( l1e_get_flags(*sl1e) & _PAGE_DIRTY )
-                        {
-                            dirty = 1;
-                            /* Note: this is atomic, so we may clear a
-                             * _PAGE_ACCESSED set by another processor. */
-                            l1e_remove_flags(*sl1e, _PAGE_DIRTY);
-                            flush_tlb = 1;
-                        }
-                    }
-                    break;
-                default:
-                    /* More than one guest reference,
-                     * we don't afford tracking that. */
-                    dirty = 1;
+                paddr_t sl1ma = pl->sl1ma;
+                unsigned long sl1mn;
+                
+                if ( sl1ma == INVALID_PADDR )
                     break;
+                
+                sl1mn = sl1ma >> PAGE_SHIFT;
+                if ( sl1mn == mfn_x(smfn) ) {
+#if DEBUG_unshadow_sl1ma_detail
+                    gdprintk(XENLOG_DEBUG,
+                             "[%lx] gfn[%lx] unshadow sl1ma:%lx\n",
+                             mfn_x(smfn),
+                             range->begin_pfn + i,
+                             sl1ma);
+#endif
+                    unshadowed++;
+                    pl = free_paddr_link(d, ppl, pl);
+                    --range->nr_mappings;
+                }
+                else
+                {
+                    ppl = &pl->pl_next;
+                    pl = *ppl;
+                    mappings++;
                 }
             }
-
-            if ( dirty )
+        }
+        if ( mappings > max_mappings )
+            max_mappings = mappings;
+        
+        if ( unshadowed ) {
+#if DEBUG_unshadow_sl1ma
+            gdprintk(XENLOG_DEBUG,
+                     "[%lx] gfn[%05lx:%05lx] unshadowed:%d mappings:0x%x "
+                     "max_mappings:%d\n",
+                     mfn_x(smfn),
+                     range->begin_pfn, range->end_pfn,
+                     unshadowed, range->nr_mappings, max_mappings);
+#endif
+            if ( range->nr_mappings == 0 )
             {
-                dirty_vram->dirty_bitmap[i / 8] |= 1 << (i % 8);
-                dirty_vram->last_dirty = NOW();
+                dirty_vram_range_free(d, range);                    
             }
         }
+    }
+ out:
+    return;
+}
+
 
-        rc = -EFAULT;
-        if ( copy_to_guest(dirty_bitmap, dirty_vram->dirty_bitmap, dirty_size) == 0 ) {
-            memset(dirty_vram->dirty_bitmap, 0, dirty_size);
-            if (dirty_vram->last_dirty + SECONDS(2) < NOW())
+typedef int (*hash_pfn_callback_t)(struct vcpu *v,
+                                   mfn_t smfn,
+                                   unsigned long begin_pfn,
+                                   unsigned long end_pfn,
+                                   int *removed);
+
+static int hash_pfn_foreach(struct vcpu *v, 
+                            unsigned int callback_mask, 
+                            hash_pfn_callback_t callbacks[], 
+                            unsigned long begin_pfn,
+                            unsigned long end_pfn)
+/* Walk the hash table looking at the types of the entries and 
+ * calling the appropriate callback function for each entry. 
+ * The mask determines which shadow types we call back for, and the array
+ * of callbacks tells us which function to call.
+ * Any callback may return non-zero to let us skip the rest of the scan. 
+ *
+ * WARNING: Callbacks MUST NOT add or remove hash entries unless they 
+ * then return non-zero to terminate the scan. */
+{
+    int i, done = 0, removed = 0;
+    struct domain *d = v->domain;
+    struct page_info *x;
+
+    /* Say we're here, to stop hash-lookups reordering the chains */
+    ASSERT(paging_locked_by_me(d));
+    ASSERT(d->arch.paging.shadow.hash_walking == 0);
+    d->arch.paging.shadow.hash_walking = 1;
+
+    for ( i = 0; i < SHADOW_HASH_BUCKETS; i++ ) 
+    {
+        /* WARNING: This is not safe against changes to the hash table.
+         * The callback *must* return non-zero if it has inserted or
+         * deleted anything from the hash (lookups are OK, though). */
+        for ( x = d->arch.paging.shadow.hash_table[i];
+              x;
+              x = next_shadow(x) )
+        {
+            if ( callback_mask & (1 << x->u.sh.type) )
             {
-                /* was clean for more than two seconds, try to disable guest
-                 * write access */
-                for ( i = begin_pfn; i < end_pfn; i++ ) {
-                    mfn_t mfn = get_gfn_query_unlocked(d, i, &t);
-                    if (mfn_x(mfn) != INVALID_MFN)
-                        flush_tlb |= sh_remove_write_access(d->vcpu[0], mfn, 1, 0);
-                }
-                dirty_vram->last_dirty = -1;
+                ASSERT(x->u.sh.type <= 15);
+                ASSERT(callbacks[x->u.sh.type] != NULL);
+                done = callbacks[x->u.sh.type](v, page_to_mfn(x), 
+                                               begin_pfn, end_pfn,
+                                               &removed);
+                if ( done ) break;
             }
-            rc = 0;
         }
+        if ( done ) break; 
     }
-    if ( flush_tlb )
-        flush_tlb_mask(d->domain_dirty_cpumask);
-    goto out;
+    d->arch.paging.shadow.hash_walking = 0;
+    return removed;
+}
 
-out_sl1ma:
-    xfree(dirty_vram->sl1ma);
-out_dirty_vram:
-    xfree(dirty_vram);
-    dirty_vram = d->arch.hvm_domain.dirty_vram = NULL;
+void sh_find_all_vram_mappings(struct vcpu *v,
+                               dv_range_t *range)
+{
+    /* Dispatch table for getting per-type functions */
+    static hash_pfn_callback_t callbacks[SH_type_unused] = {
+        NULL, /* none    */
+        SHADOW_INTERNAL_NAME(sh_find_vram_mappings_in_l1, 2), /* l1_32   */
+        SHADOW_INTERNAL_NAME(sh_find_vram_mappings_in_l1, 2), /* fl1_32  */
+        NULL, /* l2_32   */
+        SHADOW_INTERNAL_NAME(sh_find_vram_mappings_in_l1, 3), /* l1_pae  */
+        SHADOW_INTERNAL_NAME(sh_find_vram_mappings_in_l1, 3), /* fl1_pae */
+        NULL, /* l2_pae  */
+        NULL, /* l2h_pae */
+#if CONFIG_PAGING_LEVELS >= 4
+        SHADOW_INTERNAL_NAME(sh_find_vram_mappings_in_l1, 4), /* l1_64   */
+        SHADOW_INTERNAL_NAME(sh_find_vram_mappings_in_l1, 4), /* fl1_64  */
+#else
+        NULL, /* l1_64   */
+        NULL, /* fl1_64  */
+#endif
+        NULL, /* l2_64   */
+        NULL, /* l2h_64  */
+        NULL, /* l3_64   */
+        NULL, /* l4_64   */
+        NULL, /* p2m     */
+        NULL  /* unused  */
+    };
 
-out:
-    paging_unlock(d);
-    p2m_unlock(p2m_get_hostp2m(d));
-    return rc;
+    static unsigned int callback_mask = 
+          1 << SH_type_l1_32_shadow
+        | 1 << SH_type_fl1_32_shadow
+        | 1 << SH_type_l1_pae_shadow
+        | 1 << SH_type_fl1_pae_shadow
+        | 1 << SH_type_l1_64_shadow
+        | 1 << SH_type_fl1_64_shadow
+        ;
+
+    perfc_incr(shadow_mappings);
+
+    hash_pfn_foreach(v, callback_mask, callbacks,
+                     range->begin_pfn,
+                     range->end_pfn);
+
+#if DEBUG_count_initial_mappings
+    gdprintk(XENLOG_DEBUG, "[%05lx:%05lx] count of initial mappings:%d\n",
+             range->begin_pfn, range->end_pfn,
+             range->nr_mappings);
+#endif
 }
 
+
 /**************************************************************************/
 /* Shadow-control XEN_DOMCTL dispatcher */
 
diff --git a/xen/arch/x86/mm/shadow/multi.c b/xen/arch/x86/mm/shadow/multi.c
index 4967da1d35..bb983bcecc 100644
--- a/xen/arch/x86/mm/shadow/multi.c
+++ b/xen/arch/x86/mm/shadow/multi.c
@@ -35,6 +35,7 @@
 #include <asm/flushtlb.h>
 #include <asm/hvm/hvm.h>
 #include <asm/hvm/cacheattr.h>
+#include <asm/dirty_vram.h>
 #include <asm/mtrr.h>
 #include <asm/guest_pt.h>
 #include <public/sched.h>
@@ -149,6 +150,10 @@ delete_fl1_shadow_status(struct vcpu *v, gfn_t gfn, mfn_t smfn)
     SHADOW_PRINTK("gfn=%"SH_PRI_gfn", type=%08x, smfn=%05lx\n",
                    gfn_x(gfn), SH_type_fl1_shadow, mfn_x(smfn));
     ASSERT(mfn_to_page(smfn)->u.sh.head);
+
+    /* Removing any dv_paddr_links to the erstwhile shadow page */
+    dirty_vram_delete_shadow(v, gfn_x(gfn), SH_type_fl1_shadow, smfn);
+    
     shadow_hash_delete(v, gfn_x(gfn), SH_type_fl1_shadow, smfn);
 }
 
@@ -160,6 +165,10 @@ delete_shadow_status(struct vcpu *v, mfn_t gmfn, u32 shadow_type, mfn_t smfn)
                    v->domain->domain_id, v->vcpu_id,
                    mfn_x(gmfn), shadow_type, mfn_x(smfn));
     ASSERT(mfn_to_page(smfn)->u.sh.head);
+    
+    /* Removing any dv_paddr_links to the erstwhile shadow page */
+    dirty_vram_delete_shadow(v, mfn_x(gmfn), shadow_type, smfn);
+    
     shadow_hash_delete(v, mfn_x(gmfn), shadow_type, smfn);
     /* 32-on-64 PV guests don't own their l4 pages; see set_shadow_status */
     if ( !is_pv_32on64_vcpu(v) || shadow_type != SH_type_l4_64_shadow )
@@ -516,7 +525,6 @@ _sh_propagate(struct vcpu *v,
     guest_l1e_t guest_entry = { guest_intpte };
     shadow_l1e_t *sp = shadow_entry_ptr;
     struct domain *d = v->domain;
-    struct sh_dirty_vram *dirty_vram = d->arch.hvm_domain.dirty_vram;
     gfn_t target_gfn = guest_l1e_get_gfn(guest_entry);
     u32 pass_thru_flags;
     u32 gflags, sflags;
@@ -663,17 +671,6 @@ _sh_propagate(struct vcpu *v,
         }
     }
 
-    if ( unlikely((level == 1) && dirty_vram
-            && dirty_vram->last_dirty == -1
-            && gfn_x(target_gfn) >= dirty_vram->begin_pfn
-            && gfn_x(target_gfn) < dirty_vram->end_pfn) )
-    {
-        if ( ft & FETCH_TYPE_WRITE )
-            dirty_vram->last_dirty = NOW();
-        else
-            sflags &= ~_PAGE_RW;
-    }
-
     /* Read-only memory */
     if ( p2m_is_readonly(p2mt) ||
          (p2mt == p2m_mmio_direct &&
@@ -1072,101 +1069,60 @@ static int shadow_set_l2e(struct vcpu *v,
     return flags;
 }
 
-static inline void shadow_vram_get_l1e(shadow_l1e_t new_sl1e,
+/* shadow_vram_fix_l1e()
+ *
+ * Tests L1PTEs as they are modified, looking for when they start to (or
+ * cease to) point to frame buffer pages.  If the old and new gfns differ,
+ * calls dirty_vram_range_update() to updates the dirty_vram structures.
+ */
+static inline void shadow_vram_fix_l1e(shadow_l1e_t old_sl1e,
+                                       shadow_l1e_t new_sl1e,
                                        shadow_l1e_t *sl1e,
                                        mfn_t sl1mfn,
                                        struct domain *d)
 { 
-    mfn_t mfn = shadow_l1e_get_mfn(new_sl1e);
-    int flags = shadow_l1e_get_flags(new_sl1e);
-    unsigned long gfn;
-    struct sh_dirty_vram *dirty_vram = d->arch.hvm_domain.dirty_vram;
+    mfn_t new_mfn, old_mfn;
+    unsigned long new_gfn = INVALID_M2P_ENTRY, old_gfn = INVALID_M2P_ENTRY;
+    paddr_t sl1ma;
+    dv_dirty_vram_t *dirty_vram = d->arch.hvm_domain.dirty_vram;
 
-    if ( !dirty_vram         /* tracking disabled? */
-         || !(flags & _PAGE_RW) /* read-only mapping? */
-         || !mfn_valid(mfn) )   /* mfn can be invalid in mmio_direct */
+    if ( !dirty_vram )
         return;
 
-    gfn = mfn_to_gfn(d, mfn);
-    /* Page sharing not supported on shadow PTs */
-    BUG_ON(SHARED_M2P(gfn));
+    sl1ma = pfn_to_paddr(mfn_x(sl1mfn)) | ((unsigned long)sl1e & ~PAGE_MASK);
 
-    if ( (gfn >= dirty_vram->begin_pfn) && (gfn < dirty_vram->end_pfn) )
+    old_mfn = shadow_l1e_get_mfn(old_sl1e);
+
+    if ( !sh_l1e_is_magic(old_sl1e) &&
+         (l1e_get_flags(old_sl1e) & _PAGE_PRESENT) &&
+         mfn_valid(old_mfn))
     {
-        unsigned long i = gfn - dirty_vram->begin_pfn;
-        struct page_info *page = mfn_to_page(mfn);
-        
-        if ( (page->u.inuse.type_info & PGT_count_mask) == 1 )
-            /* Initial guest reference, record it */
-            dirty_vram->sl1ma[i] = pfn_to_paddr(mfn_x(sl1mfn))
-                | ((unsigned long)sl1e & ~PAGE_MASK);
+        old_gfn = mfn_to_gfn(d, old_mfn);
     }
-}
-
-static inline void shadow_vram_put_l1e(shadow_l1e_t old_sl1e,
-                                       shadow_l1e_t *sl1e,
-                                       mfn_t sl1mfn,
-                                       struct domain *d)
-{
-    mfn_t mfn = shadow_l1e_get_mfn(old_sl1e);
-    int flags = shadow_l1e_get_flags(old_sl1e);
-    unsigned long gfn;
-    struct sh_dirty_vram *dirty_vram = d->arch.hvm_domain.dirty_vram;
-
-    if ( !dirty_vram         /* tracking disabled? */
-         || !(flags & _PAGE_RW) /* read-only mapping? */
-         || !mfn_valid(mfn) )   /* mfn can be invalid in mmio_direct */
-        return;
-
-    gfn = mfn_to_gfn(d, mfn);
-    /* Page sharing not supported on shadow PTs */
-    BUG_ON(SHARED_M2P(gfn));
-
-    if ( (gfn >= dirty_vram->begin_pfn) && (gfn < dirty_vram->end_pfn) )
+    
+    new_mfn = shadow_l1e_get_mfn(new_sl1e);
+    if ( !sh_l1e_is_magic(new_sl1e) &&
+         (l1e_get_flags(new_sl1e) & _PAGE_PRESENT) &&
+         mfn_valid(new_mfn))
     {
-        unsigned long i = gfn - dirty_vram->begin_pfn;
-        struct page_info *page = mfn_to_page(mfn);
-        int dirty = 0;
-        paddr_t sl1ma = pfn_to_paddr(mfn_x(sl1mfn))
-            | ((unsigned long)sl1e & ~PAGE_MASK);
+        new_gfn = mfn_to_gfn(d, new_mfn);
+    }
 
-        if ( (page->u.inuse.type_info & PGT_count_mask) == 1 )
-        {
-            /* Last reference */
-            if ( dirty_vram->sl1ma[i] == INVALID_PADDR ) {
-                /* We didn't know it was that one, let's say it is dirty */
-                dirty = 1;
-            }
-            else
-            {
-                ASSERT(dirty_vram->sl1ma[i] == sl1ma);
-                dirty_vram->sl1ma[i] = INVALID_PADDR;
-                if ( flags & _PAGE_DIRTY )
-                    dirty = 1;
-            }
-        }
-        else
+    if ( old_gfn == new_gfn ) return;
+
+    if ( VALID_M2P(old_gfn) )
+        if ( dirty_vram_range_update(d, old_gfn, sl1ma, 0/*clear*/) )
         {
-            /* We had more than one reference, just consider the page dirty. */
-            dirty = 1;
-            /* Check that it's not the one we recorded. */
-            if ( dirty_vram->sl1ma[i] == sl1ma )
-            {
-                /* Too bad, we remembered the wrong one... */
-                dirty_vram->sl1ma[i] = INVALID_PADDR;
-            }
-            else
-            {
-                /* Ok, our recorded sl1e is still pointing to this page, let's
-                 * just hope it will remain. */
-            }
+            SHADOW_PRINTK("gfn %lx (mfn %lx) cleared vram pte\n",
+                          old_gfn, mfn_x(old_mfn));
         }
-        if ( dirty )
+
+    if ( VALID_M2P(new_gfn) )
+        if ( dirty_vram_range_update(d, new_gfn, sl1ma, 1/*set*/) )
         {
-            dirty_vram->dirty_bitmap[i / 8] |= 1 << (i % 8);
-            dirty_vram->last_dirty = NOW();
+            SHADOW_PRINTK("gfn %lx (mfn %lx) set vram pte\n",
+                          new_gfn, mfn_x(new_mfn));
         }
-    }
 }
 
 static int shadow_set_l1e(struct vcpu *v, 
@@ -1211,12 +1167,13 @@ static int shadow_set_l1e(struct vcpu *v,
                 shadow_l1e_remove_flags(new_sl1e, _PAGE_RW);
                 /* fall through */
             case 0:
-                shadow_vram_get_l1e(new_sl1e, sl1e, sl1mfn, d);
                 break;
             }
         }
     } 
 
+    shadow_vram_fix_l1e(old_sl1e, new_sl1e, sl1e, sl1mfn, d);
+
     /* Write the new entry */
     shadow_write_entries(sl1e, &new_sl1e, 1, sl1mfn);
     flags |= SHADOW_SET_CHANGED;
@@ -1231,7 +1188,6 @@ static int shadow_set_l1e(struct vcpu *v,
          * trigger a flush later. */
         if ( shadow_mode_refcounts(d) ) 
         {
-            shadow_vram_put_l1e(old_sl1e, sl1e, sl1mfn, d);
             shadow_put_page_from_l1e(old_sl1e, d);
             TRACE_SHADOW_PATH_FLAG(TRCE_SFLAG_SHADOW_L1_PUT_REF);
         } 
@@ -2018,7 +1974,6 @@ void sh_destroy_l1_shadow(struct vcpu *v, mfn_t smfn)
         SHADOW_FOREACH_L1E(sl1mfn, sl1e, 0, 0, {
             if ( (shadow_l1e_get_flags(*sl1e) & _PAGE_PRESENT)
                  && !sh_l1e_is_magic(*sl1e) ) {
-                shadow_vram_put_l1e(*sl1e, sl1e, sl1mfn, d);
                 shadow_put_page_from_l1e(*sl1e, d);
             }
         });
@@ -4336,6 +4291,37 @@ int sh_rm_mappings_from_l1(struct vcpu *v, mfn_t sl1mfn, mfn_t target_mfn)
     return done;
 }
 
+
+int sh_find_vram_mappings_in_l1(struct vcpu *v,
+                                mfn_t sl1mfn,
+                                unsigned long begin_pfn,
+                                unsigned long end_pfn,
+                                int *removed)
+/* Find all VRAM mappings in this shadow l1 table */
+{
+    struct domain *d = v->domain;
+    shadow_l1e_t *sl1e;
+    int done = 0;
+
+    /* only returns _PAGE_PRESENT entries */
+    SHADOW_FOREACH_L1E(sl1mfn, sl1e, 0, done, 
+    {
+        unsigned long gfn;
+        mfn_t gmfn = shadow_l1e_get_mfn(*sl1e);
+        if ( !mfn_valid(gmfn) )
+            continue;
+        gfn = mfn_to_gfn(d, gmfn);
+        if ( VALID_M2P(gfn) && (begin_pfn <= gfn) && (gfn < end_pfn) ) 
+        {
+            paddr_t sl1ma =
+                pfn_to_paddr(mfn_x(sl1mfn)) |
+                ( (unsigned long)sl1e & ~PAGE_MASK );
+            dirty_vram_range_update(v->domain, gfn, sl1ma, 1/*set*/);
+        }
+    });
+    return 0;
+}
+
 /**************************************************************************/
 /* Functions to excise all pointers to shadows from higher-level shadows. */
 
diff --git a/xen/arch/x86/mm/shadow/multi.h b/xen/arch/x86/mm/shadow/multi.h
index 835121e494..436a4ace8a 100644
--- a/xen/arch/x86/mm/shadow/multi.h
+++ b/xen/arch/x86/mm/shadow/multi.h
@@ -66,7 +66,12 @@ SHADOW_INTERNAL_NAME(sh_rm_write_access_from_l1, GUEST_LEVELS)
 extern int
 SHADOW_INTERNAL_NAME(sh_rm_mappings_from_l1, GUEST_LEVELS)
     (struct vcpu *v, mfn_t sl1mfn, mfn_t target_mfn);
-
+extern int
+SHADOW_INTERNAL_NAME(sh_find_vram_mappings_in_l1, GUEST_LEVELS)
+     (struct vcpu *v, mfn_t sl1mfn, 
+      unsigned long begin_pfn,
+      unsigned long end_pfn,
+      int *removed);
 extern void
 SHADOW_INTERNAL_NAME(sh_clear_shadow_entry, GUEST_LEVELS)
     (struct vcpu *v, void *ep, mfn_t smfn);
diff --git a/xen/arch/x86/mm/shadow/types.h b/xen/arch/x86/mm/shadow/types.h
index 43ce1dbb0a..5b0f9f7644 100644
--- a/xen/arch/x86/mm/shadow/types.h
+++ b/xen/arch/x86/mm/shadow/types.h
@@ -229,6 +229,7 @@ static inline shadow_l4e_t shadow_l4e_from_mfn(mfn_t mfn, u32 flags)
 #define sh_update_cr3              INTERNAL_NAME(sh_update_cr3)
 #define sh_rm_write_access_from_l1 INTERNAL_NAME(sh_rm_write_access_from_l1)
 #define sh_rm_mappings_from_l1     INTERNAL_NAME(sh_rm_mappings_from_l1)
+#define sh_find_vram_mappings_in_l1 INTERNAL_NAME(sh_find_vram_mappings_in_l1)
 #define sh_remove_l1_shadow        INTERNAL_NAME(sh_remove_l1_shadow)
 #define sh_remove_l2_shadow        INTERNAL_NAME(sh_remove_l2_shadow)
 #define sh_remove_l3_shadow        INTERNAL_NAME(sh_remove_l3_shadow)