]> git.proxmox.com Git - mirror_qemu.git/blobdiff - exec.c
egl-helpers: add dmabuf import support
[mirror_qemu.git] / exec.c
diff --git a/exec.c b/exec.c
index 821bef34767195dd2474afa0bc09f0062cd58b63..6378714a2b0a2342237edfbd829349fb8fff5319 100644 (file)
--- a/exec.c
+++ b/exec.c
 #include "qemu/cutils.h"
 #include "cpu.h"
 #include "exec/exec-all.h"
+#include "exec/target_page.h"
 #include "tcg.h"
 #include "hw/qdev-core.h"
+#include "hw/qdev-properties.h"
 #if !defined(CONFIG_USER_ONLY)
 #include "hw/boards.h"
 #include "hw/xen/xen.h"
@@ -54,7 +56,6 @@
 #endif
 
 #endif
-#include "exec/cpu-all.h"
 #include "qemu/rcu_queue.h"
 #include "qemu/main-loop.h"
 #include "translate-all.h"
@@ -117,6 +118,9 @@ __thread CPUState *current_cpu;
    2 = Adaptive rate instruction counting.  */
 int use_icount;
 
+uintptr_t qemu_host_page_size;
+intptr_t qemu_host_page_mask;
+
 bool set_preferred_target_page_bits(int bits)
 {
     /* The target page size is the lowest common denominator for all
@@ -181,21 +185,18 @@ typedef struct PhysPageMap {
 } PhysPageMap;
 
 struct AddressSpaceDispatch {
-    struct rcu_head rcu;
-
     MemoryRegionSection *mru_section;
     /* This is a multi-level map on the physical address space.
      * The bottom level has pointers to MemoryRegionSections.
      */
     PhysPageEntry phys_map;
     PhysPageMap map;
-    AddressSpace *as;
 };
 
 #define SUBPAGE_IDX(addr) ((addr) & ~TARGET_PAGE_MASK)
 typedef struct subpage_t {
     MemoryRegion iomem;
-    AddressSpace *as;
+    FlatView *fv;
     hwaddr base;
     uint16_t sub_section[];
 } subpage_t;
@@ -355,7 +356,7 @@ static void phys_page_compact(PhysPageEntry *lp, Node *nodes)
     }
 }
 
-static void phys_page_compact_all(AddressSpaceDispatch *d, int nodes_nb)
+void address_space_dispatch_compact(AddressSpaceDispatch *d)
 {
     if (d->phys_map.skip) {
         phys_page_compact(&d->phys_map, d->map.nodes);
@@ -373,10 +374,11 @@ static inline bool section_covers_addr(const MemoryRegionSection *section,
                              int128_getlo(section->size), addr);
 }
 
-static MemoryRegionSection *phys_page_find(PhysPageEntry lp, hwaddr addr,
-                                           Node *nodes, MemoryRegionSection *sections)
+static MemoryRegionSection *phys_page_find(AddressSpaceDispatch *d, hwaddr addr)
 {
-    PhysPageEntry *p;
+    PhysPageEntry lp = d->phys_map, *p;
+    Node *nodes = d->map.nodes;
+    MemoryRegionSection *sections = d->map.sections;
     hwaddr index = addr >> TARGET_PAGE_BITS;
     int i;
 
@@ -414,8 +416,7 @@ static MemoryRegionSection *address_space_lookup_region(AddressSpaceDispatch *d,
         section_covers_addr(section, addr)) {
         update = false;
     } else {
-        section = phys_page_find(d->phys_map, addr, d->map.nodes,
-                                 d->map.sections);
+        section = phys_page_find(d, addr);
         update = true;
     }
     if (resolve_subpage && section->mr->subpage) {
@@ -465,74 +466,115 @@ address_space_translate_internal(AddressSpaceDispatch *d, hwaddr addr, hwaddr *x
 }
 
 /* Called from RCU critical section */
-IOMMUTLBEntry address_space_get_iotlb_entry(AddressSpace *as, hwaddr addr,
-                                            bool is_write)
+static MemoryRegionSection flatview_do_translate(FlatView *fv,
+                                                 hwaddr addr,
+                                                 hwaddr *xlat,
+                                                 hwaddr *plen,
+                                                 bool is_write,
+                                                 bool is_mmio,
+                                                 AddressSpace **target_as)
 {
-    IOMMUTLBEntry iotlb = {0};
+    IOMMUTLBEntry iotlb;
     MemoryRegionSection *section;
-    MemoryRegion *mr;
+    IOMMUMemoryRegion *iommu_mr;
+    IOMMUMemoryRegionClass *imrc;
 
     for (;;) {
-        AddressSpaceDispatch *d = atomic_rcu_read(&as->dispatch);
-        section = address_space_lookup_region(d, addr, false);
-        addr = addr - section->offset_within_address_space
-               + section->offset_within_region;
-        mr = section->mr;
+        section = address_space_translate_internal(
+                flatview_to_dispatch(fv), addr, &addr,
+                plen, is_mmio);
 
-        if (!mr->iommu_ops) {
+        iommu_mr = memory_region_get_iommu(section->mr);
+        if (!iommu_mr) {
             break;
         }
+        imrc = memory_region_get_iommu_class_nocheck(iommu_mr);
 
-        iotlb = mr->iommu_ops->translate(mr, addr, is_write);
+        iotlb = imrc->translate(iommu_mr, addr, is_write ?
+                                IOMMU_WO : IOMMU_RO);
+        addr = ((iotlb.translated_addr & ~iotlb.addr_mask)
+                | (addr & iotlb.addr_mask));
+        *plen = MIN(*plen, (addr | iotlb.addr_mask) - addr + 1);
         if (!(iotlb.perm & (1 << is_write))) {
-            iotlb.target_as = NULL;
-            break;
+            goto translate_fail;
         }
 
-        addr = ((iotlb.translated_addr & ~iotlb.addr_mask)
-                | (addr & iotlb.addr_mask));
-        as = iotlb.target_as;
+        fv = address_space_to_flatview(iotlb.target_as);
+        *target_as = iotlb.target_as;
     }
 
-    return iotlb;
+    *xlat = addr;
+
+    return *section;
+
+translate_fail:
+    return (MemoryRegionSection) { .mr = &io_mem_unassigned };
 }
 
 /* Called from RCU critical section */
-MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
-                                      hwaddr *xlat, hwaddr *plen,
-                                      bool is_write)
+IOMMUTLBEntry address_space_get_iotlb_entry(AddressSpace *as, hwaddr addr,
+                                            bool is_write)
 {
-    IOMMUTLBEntry iotlb;
-    MemoryRegionSection *section;
-    MemoryRegion *mr;
+    MemoryRegionSection section;
+    hwaddr xlat, plen;
 
-    for (;;) {
-        AddressSpaceDispatch *d = atomic_rcu_read(&as->dispatch);
-        section = address_space_translate_internal(d, addr, &addr, plen, true);
-        mr = section->mr;
+    /* Try to get maximum page mask during translation. */
+    plen = (hwaddr)-1;
 
-        if (!mr->iommu_ops) {
-            break;
-        }
+    /* This can never be MMIO. */
+    section = flatview_do_translate(address_space_to_flatview(as), addr,
+                                    &xlat, &plen, is_write, false, &as);
 
-        iotlb = mr->iommu_ops->translate(mr, addr, is_write);
-        addr = ((iotlb.translated_addr & ~iotlb.addr_mask)
-                | (addr & iotlb.addr_mask));
-        *plen = MIN(*plen, (addr | iotlb.addr_mask) - addr + 1);
-        if (!(iotlb.perm & (1 << is_write))) {
-            mr = &io_mem_unassigned;
-            break;
-        }
+    /* Illegal translation */
+    if (section.mr == &io_mem_unassigned) {
+        goto iotlb_fail;
+    }
+
+    /* Convert memory region offset into address space offset */
+    xlat += section.offset_within_address_space -
+        section.offset_within_region;
 
-        as = iotlb.target_as;
+    if (plen == (hwaddr)-1) {
+        /*
+         * We use default page size here. Logically it only happens
+         * for identity mappings.
+         */
+        plen = TARGET_PAGE_SIZE;
     }
 
+    /* Convert to address mask */
+    plen -= 1;
+
+    return (IOMMUTLBEntry) {
+        .target_as = as,
+        .iova = addr & ~plen,
+        .translated_addr = xlat & ~plen,
+        .addr_mask = plen,
+        /* IOTLBs are for DMAs, and DMA only allows on RAMs. */
+        .perm = IOMMU_RW,
+    };
+
+iotlb_fail:
+    return (IOMMUTLBEntry) {0};
+}
+
+/* Called from RCU critical section */
+MemoryRegion *flatview_translate(FlatView *fv, hwaddr addr, hwaddr *xlat,
+                                 hwaddr *plen, bool is_write)
+{
+    MemoryRegion *mr;
+    MemoryRegionSection section;
+    AddressSpace *as = NULL;
+
+    /* This can be MMIO, so setup MMIO bit. */
+    section = flatview_do_translate(fv, addr, xlat, plen, is_write, true, &as);
+    mr = section.mr;
+
     if (xen_enabled() && memory_access_is_direct(mr, is_write)) {
         hwaddr page = ((addr & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE) - addr;
         *plen = MIN(page, *plen);
     }
 
-    *xlat = addr;
     return mr;
 }
 
@@ -546,7 +588,7 @@ address_space_translate_for_iotlb(CPUState *cpu, int asidx, hwaddr addr,
 
     section = address_space_translate_internal(d, addr, xlat, plen, false);
 
-    assert(!section->mr->iommu_ops);
+    assert(!memory_region_is_iommu(section->mr));
     return section;
 }
 #endif
@@ -693,6 +735,20 @@ void cpu_exec_unrealizefn(CPUState *cpu)
     }
 }
 
+Property cpu_common_props[] = {
+#ifndef CONFIG_USER_ONLY
+    /* Create a memory property for softmmu CPU object,
+     * so users can wire up its memory. (This can't go in qom/cpu.c
+     * because that file is compiled only once for both user-mode
+     * and system builds.) The default if no link is set up is to use
+     * the system address space.
+     */
+    DEFINE_PROP_LINK("memory", CPUState, memory, TYPE_MEMORY_REGION,
+                     MemoryRegion *),
+#endif
+    DEFINE_PROP_END_OF_LIST(),
+};
+
 void cpu_exec_initfn(CPUState *cpu)
 {
     cpu->as = NULL;
@@ -700,18 +756,6 @@ void cpu_exec_initfn(CPUState *cpu)
 
 #ifndef CONFIG_USER_ONLY
     cpu->thread_id = qemu_get_thread_id();
-
-    /* This is a softmmu CPU object, so create a property for it
-     * so users can wire up its memory. (This can't go in qom/cpu.c
-     * because that file is compiled only once for both user-mode
-     * and system builds.) The default if no link is set up is to use
-     * the system address space.
-     */
-    object_property_add_link(OBJECT(cpu), "memory", TYPE_MEMORY_REGION,
-                             (Object **)&cpu->memory,
-                             qdev_prop_allow_set_link_before_realize,
-                             OBJ_PROP_LINK_UNREF_ON_RELEASE,
-                             &error_abort);
     cpu->memory = system_memory;
     object_ref(OBJECT(cpu->memory));
 #endif
@@ -733,15 +777,28 @@ void cpu_exec_realizefn(CPUState *cpu, Error **errp)
 #endif
 }
 
+#if defined(CONFIG_USER_ONLY)
 static void breakpoint_invalidate(CPUState *cpu, target_ulong pc)
 {
-    /* Flush the whole TB as this will not have race conditions
-     * even if we don't have proper locking yet.
-     * Ideally we would just invalidate the TBs for the
-     * specified PC.
-     */
-    tb_flush(cpu);
+    mmap_lock();
+    tb_lock();
+    tb_invalidate_phys_page_range(pc, pc + 1, 0);
+    tb_unlock();
+    mmap_unlock();
+}
+#else
+static void breakpoint_invalidate(CPUState *cpu, target_ulong pc)
+{
+    MemTxAttrs attrs;
+    hwaddr phys = cpu_get_phys_page_attrs_debug(cpu, pc, &attrs);
+    int asidx = cpu_asidx_from_attrs(cpu, attrs);
+    if (phys != -1) {
+        /* Locks grabbed by tb_invalidate_phys_addr */
+        tb_invalidate_phys_addr(cpu->cpu_ases[asidx].as,
+                                phys | (pc & ~TARGET_PAGE_MASK));
+    }
 }
+#endif
 
 #if defined(CONFIG_USER_ONLY)
 void cpu_watchpoint_remove_all(CPUState *cpu, int mask)
@@ -1160,7 +1217,7 @@ hwaddr memory_region_section_get_iotlb(CPUState *cpu,
     } else {
         AddressSpaceDispatch *d;
 
-        d = atomic_rcu_read(&section->address_space->dispatch);
+        d = flatview_to_dispatch(section->fv);
         iotlb = section - d->map.sections;
         iotlb += xlat;
     }
@@ -1186,7 +1243,7 @@ hwaddr memory_region_section_get_iotlb(CPUState *cpu,
 
 static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
                              uint16_t section);
-static subpage_t *subpage_init(AddressSpace *as, hwaddr base);
+static subpage_t *subpage_init(FlatView *fv, hwaddr base);
 
 static void *(*phys_mem_alloc)(size_t size, uint64_t *align) =
                                qemu_anon_ram_alloc;
@@ -1243,13 +1300,13 @@ static void phys_sections_free(PhysPageMap *map)
     g_free(map->nodes);
 }
 
-static void register_subpage(AddressSpaceDispatch *d, MemoryRegionSection *section)
+static void register_subpage(FlatView *fv, MemoryRegionSection *section)
 {
+    AddressSpaceDispatch *d = flatview_to_dispatch(fv);
     subpage_t *subpage;
     hwaddr base = section->offset_within_address_space
         & TARGET_PAGE_MASK;
-    MemoryRegionSection *existing = phys_page_find(d->phys_map, base,
-                                                   d->map.nodes, d->map.sections);
+    MemoryRegionSection *existing = phys_page_find(d, base);
     MemoryRegionSection subsection = {
         .offset_within_address_space = base,
         .size = int128_make64(TARGET_PAGE_SIZE),
@@ -1259,8 +1316,8 @@ static void register_subpage(AddressSpaceDispatch *d, MemoryRegionSection *secti
     assert(existing->mr->subpage || existing->mr == &io_mem_unassigned);
 
     if (!(existing->mr->subpage)) {
-        subpage = subpage_init(d->as, base);
-        subsection.address_space = d->as;
+        subpage = subpage_init(fv, base);
+        subsection.fv = fv;
         subsection.mr = &subpage->iomem;
         phys_page_set(d, base >> TARGET_PAGE_BITS, 1,
                       phys_section_add(&d->map, &subsection));
@@ -1274,9 +1331,10 @@ static void register_subpage(AddressSpaceDispatch *d, MemoryRegionSection *secti
 }
 
 
-static void register_multipage(AddressSpaceDispatch *d,
+static void register_multipage(FlatView *fv,
                                MemoryRegionSection *section)
 {
+    AddressSpaceDispatch *d = flatview_to_dispatch(fv);
     hwaddr start_addr = section->offset_within_address_space;
     uint16_t section_index = phys_section_add(&d->map, section);
     uint64_t num_pages = int128_get64(int128_rshift(section->size,
@@ -1286,10 +1344,8 @@ static void register_multipage(AddressSpaceDispatch *d,
     phys_page_set(d, start_addr >> TARGET_PAGE_BITS, num_pages, section_index);
 }
 
-static void mem_add(MemoryListener *listener, MemoryRegionSection *section)
+void flatview_add_to_dispatch(FlatView *fv, MemoryRegionSection *section)
 {
-    AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
-    AddressSpaceDispatch *d = as->next_dispatch;
     MemoryRegionSection now = *section, remain = *section;
     Int128 page_size = int128_make64(TARGET_PAGE_SIZE);
 
@@ -1298,7 +1354,7 @@ static void mem_add(MemoryListener *listener, MemoryRegionSection *section)
                        - now.offset_within_address_space;
 
         now.size = int128_min(int128_make64(left), now.size);
-        register_subpage(d, &now);
+        register_subpage(fv, &now);
     } else {
         now.size = int128_zero();
     }
@@ -1308,13 +1364,13 @@ static void mem_add(MemoryListener *listener, MemoryRegionSection *section)
         remain.offset_within_region += int128_get64(now.size);
         now = remain;
         if (int128_lt(remain.size, page_size)) {
-            register_subpage(d, &now);
+            register_subpage(fv, &now);
         } else if (remain.offset_within_address_space & ~TARGET_PAGE_MASK) {
             now.size = page_size;
-            register_subpage(d, &now);
+            register_subpage(fv, &now);
         } else {
             now.size = int128_and(now.size, int128_neg(page_size));
-            register_multipage(d, &now);
+            register_multipage(fv, &now);
         }
     }
 }
@@ -1446,25 +1502,17 @@ static int64_t get_file_size(int fd)
     return size;
 }
 
-static void *file_ram_alloc(RAMBlock *block,
-                            ram_addr_t memory,
-                            const char *path,
-                            Error **errp)
+static int file_ram_open(const char *path,
+                         const char *region_name,
+                         bool *created,
+                         Error **errp)
 {
-    bool unlink_on_error = false;
     char *filename;
     char *sanitized_name;
     char *c;
-    void *area = MAP_FAILED;
     int fd = -1;
-    int64_t file_size;
-
-    if (kvm_enabled() && !kvm_has_sync_mmu()) {
-        error_setg(errp,
-                   "host lacks kvm mmu notifiers, -mem-path unsupported");
-        return NULL;
-    }
 
+    *created = false;
     for (;;) {
         fd = open(path, O_RDWR);
         if (fd >= 0) {
@@ -1475,13 +1523,13 @@ static void *file_ram_alloc(RAMBlock *block,
             /* @path names a file that doesn't exist, create it */
             fd = open(path, O_RDWR | O_CREAT | O_EXCL, 0644);
             if (fd >= 0) {
-                unlink_on_error = true;
+                *created = true;
                 break;
             }
         } else if (errno == EISDIR) {
             /* @path names a directory, create a file there */
             /* Make name safe to use with mkstemp by replacing '/' with '_'. */
-            sanitized_name = g_strdup(memory_region_name(block->mr));
+            sanitized_name = g_strdup(region_name);
             for (c = sanitized_name; *c != '\0'; c++) {
                 if (*c == '/') {
                     *c = '_';
@@ -1504,7 +1552,7 @@ static void *file_ram_alloc(RAMBlock *block,
             error_setg_errno(errp, errno,
                              "can't open backing store %s for guest RAM",
                              path);
-            goto error;
+            return -1;
         }
         /*
          * Try again on EINTR and EEXIST.  The latter happens when
@@ -1512,6 +1560,17 @@ static void *file_ram_alloc(RAMBlock *block,
          */
     }
 
+    return fd;
+}
+
+static void *file_ram_alloc(RAMBlock *block,
+                            ram_addr_t memory,
+                            int fd,
+                            bool truncate,
+                            Error **errp)
+{
+    void *area;
+
     block->page_size = qemu_fd_getpagesize(fd);
     block->mr->align = block->page_size;
 #if defined(__s390x__)
@@ -1520,20 +1579,11 @@ static void *file_ram_alloc(RAMBlock *block,
     }
 #endif
 
-    file_size = get_file_size(fd);
-
     if (memory < block->page_size) {
         error_setg(errp, "memory size 0x" RAM_ADDR_FMT " must be equal to "
                    "or larger than page size 0x%zx",
                    memory, block->page_size);
-        goto error;
-    }
-
-    if (file_size > 0 && file_size < memory) {
-        error_setg(errp, "backing store %s size 0x%" PRIx64
-                   " does not match 'size' option 0x" RAM_ADDR_FMT,
-                   path, file_size, memory);
-        goto error;
+        return NULL;
     }
 
     memory = ROUND_UP(memory, block->page_size);
@@ -1552,7 +1602,7 @@ static void *file_ram_alloc(RAMBlock *block,
      * those labels. Therefore, extending the non-empty backend file
      * is disabled as well.
      */
-    if (!file_size && ftruncate(fd, memory)) {
+    if (truncate && ftruncate(fd, memory)) {
         perror("ftruncate");
     }
 
@@ -1561,30 +1611,19 @@ static void *file_ram_alloc(RAMBlock *block,
     if (area == MAP_FAILED) {
         error_setg_errno(errp, errno,
                          "unable to map backing store for guest RAM");
-        goto error;
+        return NULL;
     }
 
     if (mem_prealloc) {
         os_mem_prealloc(fd, area, memory, smp_cpus, errp);
         if (errp && *errp) {
-            goto error;
+            qemu_ram_munmap(area, memory);
+            return NULL;
         }
     }
 
     block->fd = fd;
     return area;
-
-error:
-    if (area != MAP_FAILED) {
-        qemu_ram_munmap(area, memory);
-    }
-    if (unlink_on_error) {
-        unlink(path);
-    }
-    if (fd != -1) {
-        close(fd);
-    }
-    return NULL;
 }
 #endif
 
@@ -1895,18 +1934,25 @@ static void ram_block_add(RAMBlock *new_block, Error **errp)
 }
 
 #ifdef __linux__
-RAMBlock *qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr,
-                                   bool share, const char *mem_path,
-                                   Error **errp)
+RAMBlock *qemu_ram_alloc_from_fd(ram_addr_t size, MemoryRegion *mr,
+                                 bool share, int fd,
+                                 Error **errp)
 {
     RAMBlock *new_block;
     Error *local_err = NULL;
+    int64_t file_size;
 
     if (xen_enabled()) {
         error_setg(errp, "-mem-path not supported with Xen");
         return NULL;
     }
 
+    if (kvm_enabled() && !kvm_has_sync_mmu()) {
+        error_setg(errp,
+                   "host lacks kvm mmu notifiers, -mem-path unsupported");
+        return NULL;
+    }
+
     if (phys_mem_alloc != qemu_anon_ram_alloc) {
         /*
          * file_ram_alloc() needs to allocate just like
@@ -1919,13 +1965,20 @@ RAMBlock *qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr,
     }
 
     size = HOST_PAGE_ALIGN(size);
+    file_size = get_file_size(fd);
+    if (file_size > 0 && file_size < size) {
+        error_setg(errp, "backing store %s size 0x%" PRIx64
+                   " does not match 'size' option 0x" RAM_ADDR_FMT,
+                   mem_path, file_size, size);
+        return NULL;
+    }
+
     new_block = g_malloc0(sizeof(*new_block));
     new_block->mr = mr;
     new_block->used_length = size;
     new_block->max_length = size;
     new_block->flags = share ? RAM_SHARED : 0;
-    new_block->host = file_ram_alloc(new_block, size,
-                                     mem_path, errp);
+    new_block->host = file_ram_alloc(new_block, size, fd, !file_size, errp);
     if (!new_block->host) {
         g_free(new_block);
         return NULL;
@@ -1938,6 +1991,33 @@ RAMBlock *qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr,
         return NULL;
     }
     return new_block;
+
+}
+
+
+RAMBlock *qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr,
+                                   bool share, const char *mem_path,
+                                   Error **errp)
+{
+    int fd;
+    bool created;
+    RAMBlock *block;
+
+    fd = file_ram_open(mem_path, memory_region_name(mr), &created, errp);
+    if (fd < 0) {
+        return NULL;
+    }
+
+    block = qemu_ram_alloc_from_fd(size, mr, share, fd, errp);
+    if (!block) {
+        if (created) {
+            unlink(mem_path);
+        }
+        close(fd);
+        return NULL;
+    }
+
+    return block;
 }
 #endif
 
@@ -2106,10 +2186,10 @@ void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr)
          * In that case just map until the end of the page.
          */
         if (block->offset == 0) {
-            return xen_map_cache(addr, 0, 0);
+            return xen_map_cache(addr, 0, 0, false);
         }
 
-        block->host = xen_map_cache(block->offset, block->max_length, 1);
+        block->host = xen_map_cache(block->offset, block->max_length, 1, false);
     }
     return ramblock_ptr(block, addr);
 }
@@ -2120,7 +2200,7 @@ void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr)
  * Called within RCU critical section.
  */
 static void *qemu_ram_ptr_length(RAMBlock *ram_block, ram_addr_t addr,
-                                 hwaddr *size)
+                                 hwaddr *size, bool lock)
 {
     RAMBlock *block = ram_block;
     if (*size == 0) {
@@ -2139,10 +2219,10 @@ static void *qemu_ram_ptr_length(RAMBlock *ram_block, ram_addr_t addr,
          * In that case just map the requested area.
          */
         if (block->offset == 0) {
-            return xen_map_cache(addr, *size, 1);
+            return xen_map_cache(addr, *size, lock, lock);
         }
 
-        block->host = xen_map_cache(block->offset, block->max_length, 1);
+        block->host = xen_map_cache(block->offset, block->max_length, 1, lock);
     }
 
     return ramblock_ptr(block, addr);
@@ -2252,6 +2332,7 @@ static void notdirty_mem_write(void *opaque, hwaddr ram_addr,
 {
     bool locked = false;
 
+    assert(tcg_enabled());
     if (!cpu_physical_memory_get_dirty_flag(ram_addr, DIRTY_MEMORY_CODE)) {
         locked = true;
         tb_lock();
@@ -2310,6 +2391,7 @@ static void check_watchpoint(int offset, int len, MemTxAttrs attrs, int flags)
     CPUWatchpoint *wp;
     uint32_t cpu_flags;
 
+    assert(tcg_enabled());
     if (cpu->watchpoint_hit) {
         /* We re-entered the check after replacing the TB. Now raise
          * the debug interrupt so that is will trigger after the
@@ -2416,6 +2498,11 @@ static const MemoryRegionOps watch_mem_ops = {
     .endianness = DEVICE_NATIVE_ENDIAN,
 };
 
+static MemTxResult flatview_write(FlatView *fv, hwaddr addr, MemTxAttrs attrs,
+                                  const uint8_t *buf, int len);
+static bool flatview_access_valid(FlatView *fv, hwaddr addr, int len,
+                                  bool is_write);
+
 static MemTxResult subpage_read(void *opaque, hwaddr addr, uint64_t *data,
                                 unsigned len, MemTxAttrs attrs)
 {
@@ -2427,8 +2514,7 @@ static MemTxResult subpage_read(void *opaque, hwaddr addr, uint64_t *data,
     printf("%s: subpage %p len %u addr " TARGET_FMT_plx "\n", __func__,
            subpage, len, addr);
 #endif
-    res = address_space_read(subpage->as, addr + subpage->base,
-                             attrs, buf, len);
+    res = flatview_read(subpage->fv, addr + subpage->base, attrs, buf, len);
     if (res) {
         return res;
     }
@@ -2477,8 +2563,7 @@ static MemTxResult subpage_write(void *opaque, hwaddr addr,
     default:
         abort();
     }
-    return address_space_write(subpage->as, addr + subpage->base,
-                               attrs, buf, len);
+    return flatview_write(subpage->fv, addr + subpage->base, attrs, buf, len);
 }
 
 static bool subpage_accepts(void *opaque, hwaddr addr,
@@ -2490,8 +2575,8 @@ static bool subpage_accepts(void *opaque, hwaddr addr,
            __func__, subpage, is_write ? 'w' : 'r', len, addr);
 #endif
 
-    return address_space_access_valid(subpage->as, addr + subpage->base,
-                                      len, is_write);
+    return flatview_access_valid(subpage->fv, addr + subpage->base,
+                                 len, is_write);
 }
 
 static const MemoryRegionOps subpage_ops = {
@@ -2525,12 +2610,12 @@ static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
     return 0;
 }
 
-static subpage_t *subpage_init(AddressSpace *as, hwaddr base)
+static subpage_t *subpage_init(FlatView *fv, hwaddr base)
 {
     subpage_t *mmio;
 
     mmio = g_malloc0(sizeof(subpage_t) + TARGET_PAGE_SIZE * sizeof(uint16_t));
-    mmio->as = as;
+    mmio->fv = fv;
     mmio->base = base;
     memory_region_init_io(&mmio->iomem, NULL, &subpage_ops, mmio,
                           NULL, TARGET_PAGE_SIZE);
@@ -2544,12 +2629,11 @@ static subpage_t *subpage_init(AddressSpace *as, hwaddr base)
     return mmio;
 }
 
-static uint16_t dummy_section(PhysPageMap *map, AddressSpace *as,
-                              MemoryRegion *mr)
+static uint16_t dummy_section(PhysPageMap *map, FlatView *fv, MemoryRegion *mr)
 {
-    assert(as);
+    assert(fv);
     MemoryRegionSection section = {
-        .address_space = as,
+        .fv = fv,
         .mr = mr,
         .offset_within_address_space = 0,
         .offset_within_region = 0,
@@ -2586,46 +2670,31 @@ static void io_mem_init(void)
                           NULL, UINT64_MAX);
 }
 
-static void mem_begin(MemoryListener *listener)
+AddressSpaceDispatch *address_space_dispatch_new(FlatView *fv)
 {
-    AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
     AddressSpaceDispatch *d = g_new0(AddressSpaceDispatch, 1);
     uint16_t n;
 
-    n = dummy_section(&d->map, as, &io_mem_unassigned);
+    n = dummy_section(&d->map, fv, &io_mem_unassigned);
     assert(n == PHYS_SECTION_UNASSIGNED);
-    n = dummy_section(&d->map, as, &io_mem_notdirty);
+    n = dummy_section(&d->map, fv, &io_mem_notdirty);
     assert(n == PHYS_SECTION_NOTDIRTY);
-    n = dummy_section(&d->map, as, &io_mem_rom);
+    n = dummy_section(&d->map, fv, &io_mem_rom);
     assert(n == PHYS_SECTION_ROM);
-    n = dummy_section(&d->map, as, &io_mem_watch);
+    n = dummy_section(&d->map, fv, &io_mem_watch);
     assert(n == PHYS_SECTION_WATCH);
 
     d->phys_map  = (PhysPageEntry) { .ptr = PHYS_MAP_NODE_NIL, .skip = 1 };
-    d->as = as;
-    as->next_dispatch = d;
+
+    return d;
 }
 
-static void address_space_dispatch_free(AddressSpaceDispatch *d)
+void address_space_dispatch_free(AddressSpaceDispatch *d)
 {
     phys_sections_free(&d->map);
     g_free(d);
 }
 
-static void mem_commit(MemoryListener *listener)
-{
-    AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
-    AddressSpaceDispatch *cur = as->dispatch;
-    AddressSpaceDispatch *next = as->next_dispatch;
-
-    phys_page_compact_all(next, next->map.nodes_nb);
-
-    atomic_rcu_set(&as->dispatch, next);
-    if (cur) {
-        call_rcu(cur, address_space_dispatch_free, rcu);
-    }
-}
-
 static void tcg_commit(MemoryListener *listener)
 {
     CPUAddressSpace *cpuas;
@@ -2639,39 +2708,11 @@ static void tcg_commit(MemoryListener *listener)
      * We reload the dispatch pointer now because cpu_reloading_memory_map()
      * may have split the RCU critical section.
      */
-    d = atomic_rcu_read(&cpuas->as->dispatch);
+    d = address_space_to_dispatch(cpuas->as);
     atomic_rcu_set(&cpuas->memory_dispatch, d);
     tlb_flush(cpuas->cpu);
 }
 
-void address_space_init_dispatch(AddressSpace *as)
-{
-    as->dispatch = NULL;
-    as->dispatch_listener = (MemoryListener) {
-        .begin = mem_begin,
-        .commit = mem_commit,
-        .region_add = mem_add,
-        .region_nop = mem_add,
-        .priority = 0,
-    };
-    memory_listener_register(&as->dispatch_listener, as);
-}
-
-void address_space_unregister(AddressSpace *as)
-{
-    memory_listener_unregister(&as->dispatch_listener);
-}
-
-void address_space_destroy_dispatch(AddressSpace *as)
-{
-    AddressSpaceDispatch *d = as->dispatch;
-
-    atomic_rcu_set(&as->dispatch, NULL);
-    if (d) {
-        call_rcu(d, address_space_dispatch_free, rcu);
-    }
-}
-
 static void memory_map_init(void)
 {
     system_memory = g_malloc(sizeof(*system_memory));
@@ -2755,6 +2796,7 @@ static void invalidate_and_set_dirty(MemoryRegion *mr, hwaddr addr,
             cpu_physical_memory_range_includes_clean(addr, length, dirty_log_mask);
     }
     if (dirty_log_mask & (1 << DIRTY_MEMORY_CODE)) {
+        assert(tcg_enabled());
         tb_lock();
         tb_invalidate_phys_range(addr, addr + length);
         tb_unlock();
@@ -2814,11 +2856,11 @@ static bool prepare_mmio_access(MemoryRegion *mr)
 }
 
 /* Called within RCU critical section.  */
-static MemTxResult address_space_write_continue(AddressSpace *as, hwaddr addr,
-                                                MemTxAttrs attrs,
-                                                const uint8_t *buf,
-                                                int len, hwaddr addr1,
-                                                hwaddr l, MemoryRegion *mr)
+static MemTxResult flatview_write_continue(FlatView *fv, hwaddr addr,
+                                           MemTxAttrs attrs,
+                                           const uint8_t *buf,
+                                           int len, hwaddr addr1,
+                                           hwaddr l, MemoryRegion *mr)
 {
     uint8_t *ptr;
     uint64_t val;
@@ -2861,7 +2903,7 @@ static MemTxResult address_space_write_continue(AddressSpace *as, hwaddr addr,
             }
         } else {
             /* RAM case */
-            ptr = qemu_map_ram_ptr(mr->ram_block, addr1);
+            ptr = qemu_ram_ptr_length(mr->ram_block, addr1, &l, false);
             memcpy(ptr, buf, l);
             invalidate_and_set_dirty(mr, addr1, l);
         }
@@ -2880,14 +2922,14 @@ static MemTxResult address_space_write_continue(AddressSpace *as, hwaddr addr,
         }
 
         l = len;
-        mr = address_space_translate(as, addr, &addr1, &l, true);
+        mr = flatview_translate(fv, addr, &addr1, &l, true);
     }
 
     return result;
 }
 
-MemTxResult address_space_write(AddressSpace *as, hwaddr addr, MemTxAttrs attrs,
-                                const uint8_t *buf, int len)
+static MemTxResult flatview_write(FlatView *fv, hwaddr addr, MemTxAttrs attrs,
+                                  const uint8_t *buf, int len)
 {
     hwaddr l;
     hwaddr addr1;
@@ -2897,20 +2939,27 @@ MemTxResult address_space_write(AddressSpace *as, hwaddr addr, MemTxAttrs attrs,
     if (len > 0) {
         rcu_read_lock();
         l = len;
-        mr = address_space_translate(as, addr, &addr1, &l, true);
-        result = address_space_write_continue(as, addr, attrs, buf, len,
-                                              addr1, l, mr);
+        mr = flatview_translate(fv, addr, &addr1, &l, true);
+        result = flatview_write_continue(fv, addr, attrs, buf, len,
+                                         addr1, l, mr);
         rcu_read_unlock();
     }
 
     return result;
 }
 
+MemTxResult address_space_write(AddressSpace *as, hwaddr addr,
+                                              MemTxAttrs attrs,
+                                              const uint8_t *buf, int len)
+{
+    return flatview_write(address_space_to_flatview(as), addr, attrs, buf, len);
+}
+
 /* Called within RCU critical section.  */
-MemTxResult address_space_read_continue(AddressSpace *as, hwaddr addr,
-                                        MemTxAttrs attrs, uint8_t *buf,
-                                        int len, hwaddr addr1, hwaddr l,
-                                        MemoryRegion *mr)
+MemTxResult flatview_read_continue(FlatView *fv, hwaddr addr,
+                                   MemTxAttrs attrs, uint8_t *buf,
+                                   int len, hwaddr addr1, hwaddr l,
+                                   MemoryRegion *mr)
 {
     uint8_t *ptr;
     uint64_t val;
@@ -2952,7 +3001,7 @@ MemTxResult address_space_read_continue(AddressSpace *as, hwaddr addr,
             }
         } else {
             /* RAM case */
-            ptr = qemu_map_ram_ptr(mr->ram_block, addr1);
+            ptr = qemu_ram_ptr_length(mr->ram_block, addr1, &l, false);
             memcpy(buf, ptr, l);
         }
 
@@ -2970,14 +3019,14 @@ MemTxResult address_space_read_continue(AddressSpace *as, hwaddr addr,
         }
 
         l = len;
-        mr = address_space_translate(as, addr, &addr1, &l, false);
+        mr = flatview_translate(fv, addr, &addr1, &l, false);
     }
 
     return result;
 }
 
-MemTxResult address_space_read_full(AddressSpace *as, hwaddr addr,
-                                    MemTxAttrs attrs, uint8_t *buf, int len)
+MemTxResult flatview_read_full(FlatView *fv, hwaddr addr,
+                               MemTxAttrs attrs, uint8_t *buf, int len)
 {
     hwaddr l;
     hwaddr addr1;
@@ -2987,25 +3036,33 @@ MemTxResult address_space_read_full(AddressSpace *as, hwaddr addr,
     if (len > 0) {
         rcu_read_lock();
         l = len;
-        mr = address_space_translate(as, addr, &addr1, &l, false);
-        result = address_space_read_continue(as, addr, attrs, buf, len,
-                                             addr1, l, mr);
+        mr = flatview_translate(fv, addr, &addr1, &l, false);
+        result = flatview_read_continue(fv, addr, attrs, buf, len,
+                                        addr1, l, mr);
         rcu_read_unlock();
     }
 
     return result;
 }
 
-MemTxResult address_space_rw(AddressSpace *as, hwaddr addr, MemTxAttrs attrs,
-                             uint8_t *buf, int len, bool is_write)
+static MemTxResult flatview_rw(FlatView *fv, hwaddr addr, MemTxAttrs attrs,
+                               uint8_t *buf, int len, bool is_write)
 {
     if (is_write) {
-        return address_space_write(as, addr, attrs, (uint8_t *)buf, len);
+        return flatview_write(fv, addr, attrs, (uint8_t *)buf, len);
     } else {
-        return address_space_read(as, addr, attrs, (uint8_t *)buf, len);
+        return flatview_read(fv, addr, attrs, (uint8_t *)buf, len);
     }
 }
 
+MemTxResult address_space_rw(AddressSpace *as, hwaddr addr,
+                             MemTxAttrs attrs, uint8_t *buf,
+                             int len, bool is_write)
+{
+    return flatview_rw(address_space_to_flatview(as),
+                       addr, attrs, buf, len, is_write);
+}
+
 void cpu_physical_memory_rw(hwaddr addr, uint8_t *buf,
                             int len, int is_write)
 {
@@ -3163,7 +3220,8 @@ static void cpu_notify_map_clients(void)
     qemu_mutex_unlock(&map_client_list_lock);
 }
 
-bool address_space_access_valid(AddressSpace *as, hwaddr addr, int len, bool is_write)
+static bool flatview_access_valid(FlatView *fv, hwaddr addr, int len,
+                                  bool is_write)
 {
     MemoryRegion *mr;
     hwaddr l, xlat;
@@ -3171,7 +3229,7 @@ bool address_space_access_valid(AddressSpace *as, hwaddr addr, int len, bool is_
     rcu_read_lock();
     while (len > 0) {
         l = len;
-        mr = address_space_translate(as, addr, &xlat, &l, is_write);
+        mr = flatview_translate(fv, addr, &xlat, &l, is_write);
         if (!memory_access_is_direct(mr, is_write)) {
             l = memory_access_size(mr, l, addr);
             if (!memory_region_access_valid(mr, xlat, l, is_write)) {
@@ -3187,8 +3245,16 @@ bool address_space_access_valid(AddressSpace *as, hwaddr addr, int len, bool is_
     return true;
 }
 
+bool address_space_access_valid(AddressSpace *as, hwaddr addr,
+                                int len, bool is_write)
+{
+    return flatview_access_valid(address_space_to_flatview(as),
+                                 addr, len, is_write);
+}
+
 static hwaddr
-address_space_extend_translation(AddressSpace *as, hwaddr addr, hwaddr target_len,
+flatview_extend_translation(FlatView *fv, hwaddr addr,
+                                 hwaddr target_len,
                                  MemoryRegion *mr, hwaddr base, hwaddr len,
                                  bool is_write)
 {
@@ -3205,7 +3271,8 @@ address_space_extend_translation(AddressSpace *as, hwaddr addr, hwaddr target_le
         }
 
         len = target_len;
-        this_mr = address_space_translate(as, addr, &xlat, &len, is_write);
+        this_mr = flatview_translate(fv, addr, &xlat,
+                                                   &len, is_write);
         if (this_mr != mr || xlat != base + done) {
             return done;
         }
@@ -3228,6 +3295,7 @@ void *address_space_map(AddressSpace *as,
     hwaddr l, xlat;
     MemoryRegion *mr;
     void *ptr;
+    FlatView *fv = address_space_to_flatview(as);
 
     if (len == 0) {
         return NULL;
@@ -3235,7 +3303,7 @@ void *address_space_map(AddressSpace *as,
 
     l = len;
     rcu_read_lock();
-    mr = address_space_translate(as, addr, &xlat, &l, is_write);
+    mr = flatview_translate(fv, addr, &xlat, &l, is_write);
 
     if (!memory_access_is_direct(mr, is_write)) {
         if (atomic_xchg(&bounce.in_use, true)) {
@@ -3251,7 +3319,7 @@ void *address_space_map(AddressSpace *as,
         memory_region_ref(mr);
         bounce.mr = mr;
         if (!is_write) {
-            address_space_read(as, addr, MEMTXATTRS_UNSPECIFIED,
+            flatview_read(fv, addr, MEMTXATTRS_UNSPECIFIED,
                                bounce.buffer, l);
         }
 
@@ -3262,8 +3330,9 @@ void *address_space_map(AddressSpace *as,
 
 
     memory_region_ref(mr);
-    *plen = address_space_extend_translation(as, addr, len, mr, xlat, l, is_write);
-    ptr = qemu_ram_ptr_length(mr->ram_block, xlat, plen);
+    *plen = flatview_extend_translation(fv, addr, len, mr, xlat,
+                                             l, is_write);
+    ptr = qemu_ram_ptr_length(mr->ram_block, xlat, plen, true);
     rcu_read_unlock();
 
     return ptr;
@@ -3408,6 +3477,15 @@ size_t qemu_target_page_size(void)
     return TARGET_PAGE_SIZE;
 }
 
+int qemu_target_page_bits(void)
+{
+    return TARGET_PAGE_BITS;
+}
+
+int qemu_target_page_bits_min(void)
+{
+    return TARGET_PAGE_BITS_MIN;
+}
 #endif
 
 /*
@@ -3521,3 +3599,100 @@ err:
 }
 
 #endif
+
+void page_size_init(void)
+{
+    /* NOTE: we can always suppose that qemu_host_page_size >=
+       TARGET_PAGE_SIZE */
+    if (qemu_host_page_size == 0) {
+        qemu_host_page_size = qemu_real_host_page_size;
+    }
+    if (qemu_host_page_size < TARGET_PAGE_SIZE) {
+        qemu_host_page_size = TARGET_PAGE_SIZE;
+    }
+    qemu_host_page_mask = -(intptr_t)qemu_host_page_size;
+}
+
+#if !defined(CONFIG_USER_ONLY)
+
+static void mtree_print_phys_entries(fprintf_function mon, void *f,
+                                     int start, int end, int skip, int ptr)
+{
+    if (start == end - 1) {
+        mon(f, "\t%3d      ", start);
+    } else {
+        mon(f, "\t%3d..%-3d ", start, end - 1);
+    }
+    mon(f, " skip=%d ", skip);
+    if (ptr == PHYS_MAP_NODE_NIL) {
+        mon(f, " ptr=NIL");
+    } else if (!skip) {
+        mon(f, " ptr=#%d", ptr);
+    } else {
+        mon(f, " ptr=[%d]", ptr);
+    }
+    mon(f, "\n");
+}
+
+#define MR_SIZE(size) (int128_nz(size) ? (hwaddr)int128_get64( \
+                           int128_sub((size), int128_one())) : 0)
+
+void mtree_print_dispatch(fprintf_function mon, void *f,
+                          AddressSpaceDispatch *d, MemoryRegion *root)
+{
+    int i;
+
+    mon(f, "  Dispatch\n");
+    mon(f, "    Physical sections\n");
+
+    for (i = 0; i < d->map.sections_nb; ++i) {
+        MemoryRegionSection *s = d->map.sections + i;
+        const char *names[] = { " [unassigned]", " [not dirty]",
+                                " [ROM]", " [watch]" };
+
+        mon(f, "      #%d @" TARGET_FMT_plx ".." TARGET_FMT_plx " %s%s%s%s%s",
+            i,
+            s->offset_within_address_space,
+            s->offset_within_address_space + MR_SIZE(s->mr->size),
+            s->mr->name ? s->mr->name : "(noname)",
+            i < ARRAY_SIZE(names) ? names[i] : "",
+            s->mr == root ? " [ROOT]" : "",
+            s == d->mru_section ? " [MRU]" : "",
+            s->mr->is_iommu ? " [iommu]" : "");
+
+        if (s->mr->alias) {
+            mon(f, " alias=%s", s->mr->alias->name ?
+                    s->mr->alias->name : "noname");
+        }
+        mon(f, "\n");
+    }
+
+    mon(f, "    Nodes (%d bits per level, %d levels) ptr=[%d] skip=%d\n",
+               P_L2_BITS, P_L2_LEVELS, d->phys_map.ptr, d->phys_map.skip);
+    for (i = 0; i < d->map.nodes_nb; ++i) {
+        int j, jprev;
+        PhysPageEntry prev;
+        Node *n = d->map.nodes + i;
+
+        mon(f, "      [%d]\n", i);
+
+        for (j = 0, jprev = 0, prev = *n[0]; j < ARRAY_SIZE(*n); ++j) {
+            PhysPageEntry *pe = *n + j;
+
+            if (pe->ptr == prev.ptr && pe->skip == prev.skip) {
+                continue;
+            }
+
+            mtree_print_phys_entries(mon, f, jprev, j, prev.skip, prev.ptr);
+
+            jprev = j;
+            prev = *pe;
+        }
+
+        if (jprev != ARRAY_SIZE(*n)) {
+            mtree_print_phys_entries(mon, f, jprev, j, prev.skip, prev.ptr);
+        }
+    }
+}
+
+#endif