gdb support for user mode (Paul Brook)
[qemu] / exec.c
diff --git a/exec.c b/exec.c
index e7f5081..f923043 100644 (file)
--- a/exec.c
+++ b/exec.c
  * License along with this library; if not, write to the Free Software
  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  */
+#include "config.h"
+#ifdef _WIN32
+#include <windows.h>
+#else
+#include <sys/types.h>
+#include <sys/mman.h>
+#endif
 #include <stdlib.h>
 #include <stdio.h>
 #include <stdarg.h>
 #include <errno.h>
 #include <unistd.h>
 #include <inttypes.h>
-#include <sys/mman.h>
 
-#include "config.h"
-#ifdef TARGET_I386
-#include "cpu-i386.h"
-#endif
-#ifdef TARGET_ARM
-#include "cpu-arm.h"
-#endif
-#include "exec.h"
+#include "cpu.h"
+#include "exec-all.h"
 
 //#define DEBUG_TB_INVALIDATE
 //#define DEBUG_FLUSH
+//#define DEBUG_TLB
 
 /* make various TB consistency checks */
 //#define DEBUG_TB_CHECK 
+//#define DEBUG_TLB_CHECK 
 
 /* threshold to flush the translated code buffer */
 #define CODE_GEN_BUFFER_MAX_SIZE (CODE_GEN_BUFFER_SIZE - CODE_GEN_MAX_SIZE)
 
-#define CODE_GEN_MAX_BLOCKS    (CODE_GEN_BUFFER_SIZE / 64)
+#define SMC_BITMAP_USE_THRESHOLD 10
+
+#define MMAP_AREA_START        0x00000000
+#define MMAP_AREA_END          0xa8000000
 
 TranslationBlock tbs[CODE_GEN_MAX_BLOCKS];
 TranslationBlock *tb_hash[CODE_GEN_HASH_SIZE];
+TranslationBlock *tb_phys_hash[CODE_GEN_PHYS_HASH_SIZE];
 int nb_tbs;
 /* any access to the tbs or the page table must use this lock */
 spinlock_t tb_lock = SPIN_LOCK_UNLOCKED;
 
-uint8_t code_gen_buffer[CODE_GEN_BUFFER_SIZE];
+uint8_t code_gen_buffer[CODE_GEN_BUFFER_SIZE] __attribute__((aligned (32)));
 uint8_t *code_gen_ptr;
 
-/* XXX: pack the flags in the low bits of the pointer ? */
+int phys_ram_size;
+int phys_ram_fd;
+uint8_t *phys_ram_base;
+uint8_t *phys_ram_dirty;
+
 typedef struct PageDesc {
-    unsigned long flags;
+    /* list of TBs intersecting this ram page */
     TranslationBlock *first_tb;
+    /* in order to optimize self modifying code, we count the number
+       of lookups we do to a given page to use a bitmap */
+    unsigned int code_write_count;
+    uint8_t *code_bitmap;
+#if defined(CONFIG_USER_ONLY)
+    unsigned long flags;
+#endif
 } PageDesc;
 
+typedef struct PhysPageDesc {
+    /* offset in host memory of the page + io_index in the low 12 bits */
+    unsigned long phys_offset;
+} PhysPageDesc;
+
+typedef struct VirtPageDesc {
+    /* physical address of code page. It is valid only if 'valid_tag'
+       matches 'virt_valid_tag' */ 
+    target_ulong phys_addr; 
+    unsigned int valid_tag;
+#if !defined(CONFIG_SOFTMMU)
+    /* original page access rights. It is valid only if 'valid_tag'
+       matches 'virt_valid_tag' */
+    unsigned int prot;
+#endif
+} VirtPageDesc;
+
 #define L2_BITS 10
 #define L1_BITS (32 - L2_BITS - TARGET_PAGE_BITS)
 
 #define L1_SIZE (1 << L1_BITS)
 #define L2_SIZE (1 << L2_BITS)
 
-static void tb_invalidate_page(unsigned long address);
+static void io_mem_init(void);
 
-unsigned long real_host_page_size;
-unsigned long host_page_bits;
-unsigned long host_page_size;
-unsigned long host_page_mask;
+unsigned long qemu_real_host_page_size;
+unsigned long qemu_host_page_bits;
+unsigned long qemu_host_page_size;
+unsigned long qemu_host_page_mask;
 
+/* XXX: for system emulation, it could just be an array */
 static PageDesc *l1_map[L1_SIZE];
+PhysPageDesc **l1_phys_map;
+
+#if !defined(CONFIG_USER_ONLY)
+static VirtPageDesc *l1_virt_map[L1_SIZE];
+static unsigned int virt_valid_tag;
+#endif
+
+/* io memory support */
+CPUWriteMemoryFunc *io_mem_write[IO_MEM_NB_ENTRIES][4];
+CPUReadMemoryFunc *io_mem_read[IO_MEM_NB_ENTRIES][4];
+void *io_mem_opaque[IO_MEM_NB_ENTRIES];
+static int io_mem_nb;
+
+/* log support */
+char *logfilename = "/tmp/qemu.log";
+FILE *logfile;
+int loglevel;
+
+/* statistics */
+static int tlb_flush_count;
+static int tb_flush_count;
+static int tb_phys_invalidate_count;
 
 static void page_init(void)
 {
-    /* NOTE: we can always suppose that host_page_size >=
+    /* NOTE: we can always suppose that qemu_host_page_size >=
        TARGET_PAGE_SIZE */
-    real_host_page_size = getpagesize();
-    if (host_page_size == 0)
-        host_page_size = real_host_page_size;
-    if (host_page_size < TARGET_PAGE_SIZE)
-        host_page_size = TARGET_PAGE_SIZE;
-    host_page_bits = 0;
-    while ((1 << host_page_bits) < host_page_size)
-        host_page_bits++;
-    host_page_mask = ~(host_page_size - 1);
-}
-
-/* dump memory mappings */
-void page_dump(FILE *f)
-{
-    unsigned long start, end;
-    int i, j, prot, prot1;
-    PageDesc *p;
+#ifdef _WIN32
+    {
+        SYSTEM_INFO system_info;
+        DWORD old_protect;
+        
+        GetSystemInfo(&system_info);
+        qemu_real_host_page_size = system_info.dwPageSize;
+        
+        VirtualProtect(code_gen_buffer, sizeof(code_gen_buffer),
+                       PAGE_EXECUTE_READWRITE, &old_protect);
+    }
+#else
+    qemu_real_host_page_size = getpagesize();
+    {
+        unsigned long start, end;
 
-    fprintf(f, "%-8s %-8s %-8s %s\n",
-            "start", "end", "size", "prot");
-    start = -1;
-    end = -1;
-    prot = 0;
-    for(i = 0; i <= L1_SIZE; i++) {
-        if (i < L1_SIZE)
-            p = l1_map[i];
-        else
-            p = NULL;
-        for(j = 0;j < L2_SIZE; j++) {
-            if (!p)
-                prot1 = 0;
-            else
-                prot1 = p[j].flags;
-            if (prot1 != prot) {
-                end = (i << (32 - L1_BITS)) | (j << TARGET_PAGE_BITS);
-                if (start != -1) {
-                    fprintf(f, "%08lx-%08lx %08lx %c%c%c\n",
-                            start, end, end - start, 
-                            prot & PAGE_READ ? 'r' : '-',
-                            prot & PAGE_WRITE ? 'w' : '-',
-                            prot & PAGE_EXEC ? 'x' : '-');
-                }
-                if (prot1 != 0)
-                    start = end;
-                else
-                    start = -1;
-                prot = prot1;
-            }
-            if (!p)
-                break;
-        }
+        start = (unsigned long)code_gen_buffer;
+        start &= ~(qemu_real_host_page_size - 1);
+        
+        end = (unsigned long)code_gen_buffer + sizeof(code_gen_buffer);
+        end += qemu_real_host_page_size - 1;
+        end &= ~(qemu_real_host_page_size - 1);
+        
+        mprotect((void *)start, end - start, 
+                 PROT_READ | PROT_WRITE | PROT_EXEC);
     }
+#endif
+
+    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_bits = 0;
+    while ((1 << qemu_host_page_bits) < qemu_host_page_size)
+        qemu_host_page_bits++;
+    qemu_host_page_mask = ~(qemu_host_page_size - 1);
+#if !defined(CONFIG_USER_ONLY)
+    virt_valid_tag = 1;
+#endif
+    l1_phys_map = qemu_vmalloc(L1_SIZE * sizeof(PhysPageDesc *));
+    memset(l1_phys_map, 0, L1_SIZE * sizeof(PhysPageDesc *));
 }
 
 static inline PageDesc *page_find_alloc(unsigned int index)
@@ -142,7 +188,7 @@ static inline PageDesc *page_find_alloc(unsigned int index)
     p = *lp;
     if (!p) {
         /* allocate if not found */
-        p = malloc(sizeof(PageDesc) * L2_SIZE);
+        p = qemu_malloc(sizeof(PageDesc) * L2_SIZE);
         memset(p, 0, sizeof(PageDesc) * L2_SIZE);
         *lp = p;
     }
@@ -159,51 +205,106 @@ static inline PageDesc *page_find(unsigned int index)
     return p + (index & (L2_SIZE - 1));
 }
 
-int page_get_flags(unsigned long address)
+static inline PhysPageDesc *phys_page_find_alloc(unsigned int index)
 {
-    PageDesc *p;
+    PhysPageDesc **lp, *p;
 
-    p = page_find(address >> TARGET_PAGE_BITS);
+    lp = &l1_phys_map[index >> L2_BITS];
+    p = *lp;
+    if (!p) {
+        /* allocate if not found */
+        p = qemu_vmalloc(sizeof(PhysPageDesc) * L2_SIZE);
+        memset(p, 0, sizeof(PhysPageDesc) * L2_SIZE);
+        *lp = p;
+    }
+    return p + (index & (L2_SIZE - 1));
+}
+
+static inline PhysPageDesc *phys_page_find(unsigned int index)
+{
+    PhysPageDesc *p;
+
+    p = l1_phys_map[index >> L2_BITS];
     if (!p)
         return 0;
-    return p->flags;
+    return p + (index & (L2_SIZE - 1));
 }
 
-/* modify the flags of a page and invalidate the code if
-   necessary. The flag PAGE_WRITE_ORG is positionned automatically
-   depending on PAGE_WRITE */
-void page_set_flags(unsigned long start, unsigned long end, int flags)
+#if !defined(CONFIG_USER_ONLY)
+static void tlb_protect_code(CPUState *env, target_ulong addr);
+static void tlb_unprotect_code_phys(CPUState *env, unsigned long phys_addr, target_ulong vaddr);
+
+static inline VirtPageDesc *virt_page_find_alloc(unsigned int index)
 {
-    PageDesc *p;
-    unsigned long addr;
+    VirtPageDesc **lp, *p;
 
-    start = start & TARGET_PAGE_MASK;
-    end = TARGET_PAGE_ALIGN(end);
-    if (flags & PAGE_WRITE)
-        flags |= PAGE_WRITE_ORG;
-    spin_lock(&tb_lock);
-    for(addr = start; addr < end; addr += TARGET_PAGE_SIZE) {
-        p = page_find_alloc(addr >> TARGET_PAGE_BITS);
-        /* if the write protection is set, then we invalidate the code
-           inside */
-        if (!(p->flags & PAGE_WRITE) && 
-            (flags & PAGE_WRITE) &&
-            p->first_tb) {
-            tb_invalidate_page(addr);
+    /* XXX: should not truncate for 64 bit addresses */
+#if TARGET_LONG_BITS > 32
+    index &= (L1_SIZE - 1);
+#endif
+    lp = &l1_virt_map[index >> L2_BITS];
+    p = *lp;
+    if (!p) {
+        /* allocate if not found */
+        p = qemu_malloc(sizeof(VirtPageDesc) * L2_SIZE);
+        memset(p, 0, sizeof(VirtPageDesc) * L2_SIZE);
+        *lp = p;
+    }
+    return p + (index & (L2_SIZE - 1));
+}
+
+static inline VirtPageDesc *virt_page_find(unsigned int index)
+{
+    VirtPageDesc *p;
+
+    p = l1_virt_map[index >> L2_BITS];
+    if (!p)
+        return 0;
+    return p + (index & (L2_SIZE - 1));
+}
+
+static void virt_page_flush(void)
+{
+    int i, j;
+    VirtPageDesc *p;
+    
+    virt_valid_tag++;
+
+    if (virt_valid_tag == 0) {
+        virt_valid_tag = 1;
+        for(i = 0; i < L1_SIZE; i++) {
+            p = l1_virt_map[i];
+            if (p) {
+                for(j = 0; j < L2_SIZE; j++)
+                    p[j].valid_tag = 0;
+            }
         }
-        p->flags = flags;
     }
-    spin_unlock(&tb_lock);
 }
+#else
+static void virt_page_flush(void)
+{
+}
+#endif
 
 void cpu_exec_init(void)
 {
     if (!code_gen_ptr) {
         code_gen_ptr = code_gen_buffer;
         page_init();
+        io_mem_init();
     }
 }
 
+static inline void invalidate_page_bitmap(PageDesc *p)
+{
+    if (p->code_bitmap) {
+        qemu_free(p->code_bitmap);
+        p->code_bitmap = NULL;
+    }
+    p->code_write_count = 0;
+}
+
 /* set to NULL all the 'first_tb' fields in all PageDescs */
 static void page_flush_tb(void)
 {
@@ -213,30 +314,36 @@ static void page_flush_tb(void)
     for(i = 0; i < L1_SIZE; i++) {
         p = l1_map[i];
         if (p) {
-            for(j = 0; j < L2_SIZE; j++)
-                p[j].first_tb = NULL;
+            for(j = 0; j < L2_SIZE; j++) {
+                p->first_tb = NULL;
+                invalidate_page_bitmap(p);
+                p++;
+            }
         }
     }
 }
 
 /* flush all the translation blocks */
 /* XXX: tb_flush is currently not thread safe */
-void tb_flush(void)
+void tb_flush(CPUState *env)
 {
-    int i;
-#ifdef DEBUG_FLUSH
+#if defined(DEBUG_FLUSH)
     printf("qemu: flush code_size=%d nb_tbs=%d avg_tb_size=%d\n", 
            code_gen_ptr - code_gen_buffer, 
            nb_tbs, 
-           (code_gen_ptr - code_gen_buffer) / nb_tbs);
+           nb_tbs > 0 ? (code_gen_ptr - code_gen_buffer) / nb_tbs : 0);
 #endif
     nb_tbs = 0;
-    for(i = 0;i < CODE_GEN_HASH_SIZE; i++)
-        tb_hash[i] = NULL;
+    memset (tb_hash, 0, CODE_GEN_HASH_SIZE * sizeof (void *));
+    virt_page_flush();
+
+    memset (tb_phys_hash, 0, CODE_GEN_PHYS_HASH_SIZE * sizeof (void *));
     page_flush_tb();
+
     code_gen_ptr = code_gen_buffer;
     /* XXX: flush processor icache at this point if cache flush is
        expensive */
+    tb_flush_count++;
 }
 
 #ifdef DEBUG_TB_CHECK
@@ -312,6 +419,23 @@ static inline void tb_remove(TranslationBlock **ptb, TranslationBlock *tb,
     }
 }
 
+static inline void tb_page_remove(TranslationBlock **ptb, TranslationBlock *tb)
+{
+    TranslationBlock *tb1;
+    unsigned int n1;
+
+    for(;;) {
+        tb1 = *ptb;
+        n1 = (long)tb1 & 3;
+        tb1 = (TranslationBlock *)((long)tb1 & ~3);
+        if (tb1 == tb) {
+            *ptb = tb1->page_next[n1];
+            break;
+        }
+        ptb = &tb1->page_next[n1];
+    }
+}
+
 static inline void tb_jmp_remove(TranslationBlock *tb, int n)
 {
     TranslationBlock *tb1, **ptb;
@@ -347,29 +471,27 @@ static inline void tb_reset_jump(TranslationBlock *tb, int n)
     tb_set_jmp_target(tb, n, (unsigned long)(tb->tc_ptr + tb->tb_next_offset[n]));
 }
 
-static inline void tb_invalidate(TranslationBlock *tb, int parity)
+static inline void tb_invalidate(TranslationBlock *tb)
 {
-    PageDesc *p;
-    unsigned int page_index1, page_index2;
     unsigned int h, n1;
-    TranslationBlock *tb1, *tb2;
+    TranslationBlock *tb1, *tb2, **ptb;
     
+    tb_invalidated_flag = 1;
+
     /* remove the TB from the hash list */
     h = tb_hash_func(tb->pc);
-    tb_remove(&tb_hash[h], tb, 
-              offsetof(TranslationBlock, hash_next));
-    /* remove the TB from the page list */
-    page_index1 = tb->pc >> TARGET_PAGE_BITS;
-    if ((page_index1 & 1) == parity) {
-        p = page_find(page_index1);
-        tb_remove(&p->first_tb, tb, 
-                  offsetof(TranslationBlock, page_next[page_index1 & 1]));
-    }
-    page_index2 = (tb->pc + tb->size - 1) >> TARGET_PAGE_BITS;
-    if ((page_index2 & 1) == parity) {
-        p = page_find(page_index2);
-        tb_remove(&p->first_tb, tb, 
-                  offsetof(TranslationBlock, page_next[page_index2 & 1]));
+    ptb = &tb_hash[h];
+    for(;;) {
+        tb1 = *ptb;
+        /* NOTE: the TB is not necessarily linked in the hash. It
+           indicates that it is not currently used */
+        if (tb1 == NULL)
+            return;
+        if (tb1 == tb) {
+            *ptb = tb1->hash_next;
+            break;
+        }
+        ptb = &tb1->hash_next;
     }
 
     /* suppress this TB from the two jump lists */
@@ -391,67 +513,391 @@ static inline void tb_invalidate(TranslationBlock *tb, int parity)
     tb->jmp_first = (TranslationBlock *)((long)tb | 2); /* fail safe */
 }
 
-/* invalidate all TBs which intersect with the target page starting at addr */
-static void tb_invalidate_page(unsigned long address)
+static inline void tb_phys_invalidate(TranslationBlock *tb, unsigned int page_addr)
 {
-    TranslationBlock *tb_next, *tb;
-    unsigned int page_index;
-    int parity1, parity2;
     PageDesc *p;
-#ifdef DEBUG_TB_INVALIDATE
-    printf("tb_invalidate_page: %lx\n", address);
+    unsigned int h;
+    target_ulong phys_pc;
+    
+    /* remove the TB from the hash list */
+    phys_pc = tb->page_addr[0] + (tb->pc & ~TARGET_PAGE_MASK);
+    h = tb_phys_hash_func(phys_pc);
+    tb_remove(&tb_phys_hash[h], tb, 
+              offsetof(TranslationBlock, phys_hash_next));
+
+    /* remove the TB from the page list */
+    if (tb->page_addr[0] != page_addr) {
+        p = page_find(tb->page_addr[0] >> TARGET_PAGE_BITS);
+        tb_page_remove(&p->first_tb, tb);
+        invalidate_page_bitmap(p);
+    }
+    if (tb->page_addr[1] != -1 && tb->page_addr[1] != page_addr) {
+        p = page_find(tb->page_addr[1] >> TARGET_PAGE_BITS);
+        tb_page_remove(&p->first_tb, tb);
+        invalidate_page_bitmap(p);
+    }
+
+    tb_invalidate(tb);
+    tb_phys_invalidate_count++;
+}
+
+static inline void set_bits(uint8_t *tab, int start, int len)
+{
+    int end, mask, end1;
+
+    end = start + len;
+    tab += start >> 3;
+    mask = 0xff << (start & 7);
+    if ((start & ~7) == (end & ~7)) {
+        if (start < end) {
+            mask &= ~(0xff << (end & 7));
+            *tab |= mask;
+        }
+    } else {
+        *tab++ |= mask;
+        start = (start + 8) & ~7;
+        end1 = end & ~7;
+        while (start < end1) {
+            *tab++ = 0xff;
+            start += 8;
+        }
+        if (start < end) {
+            mask = ~(0xff << (end & 7));
+            *tab |= mask;
+        }
+    }
+}
+
+static void build_page_bitmap(PageDesc *p)
+{
+    int n, tb_start, tb_end;
+    TranslationBlock *tb;
+    
+    p->code_bitmap = qemu_malloc(TARGET_PAGE_SIZE / 8);
+    if (!p->code_bitmap)
+        return;
+    memset(p->code_bitmap, 0, TARGET_PAGE_SIZE / 8);
+
+    tb = p->first_tb;
+    while (tb != NULL) {
+        n = (long)tb & 3;
+        tb = (TranslationBlock *)((long)tb & ~3);
+        /* NOTE: this is subtle as a TB may span two physical pages */
+        if (n == 0) {
+            /* NOTE: tb_end may be after the end of the page, but
+               it is not a problem */
+            tb_start = tb->pc & ~TARGET_PAGE_MASK;
+            tb_end = tb_start + tb->size;
+            if (tb_end > TARGET_PAGE_SIZE)
+                tb_end = TARGET_PAGE_SIZE;
+        } else {
+            tb_start = 0;
+            tb_end = ((tb->pc + tb->size) & ~TARGET_PAGE_MASK);
+        }
+        set_bits(p->code_bitmap, tb_start, tb_end - tb_start);
+        tb = tb->page_next[n];
+    }
+}
+
+#ifdef TARGET_HAS_PRECISE_SMC
+
+static void tb_gen_code(CPUState *env, 
+                        target_ulong pc, target_ulong cs_base, int flags,
+                        int cflags)
+{
+    TranslationBlock *tb;
+    uint8_t *tc_ptr;
+    target_ulong phys_pc, phys_page2, virt_page2;
+    int code_gen_size;
+
+    phys_pc = get_phys_addr_code(env, pc);
+    tb = tb_alloc(pc);
+    if (!tb) {
+        /* flush must be done */
+        tb_flush(env);
+        /* cannot fail at this point */
+        tb = tb_alloc(pc);
+    }
+    tc_ptr = code_gen_ptr;
+    tb->tc_ptr = tc_ptr;
+    tb->cs_base = cs_base;
+    tb->flags = flags;
+    tb->cflags = cflags;
+    cpu_gen_code(env, tb, CODE_GEN_MAX_SIZE, &code_gen_size);
+    code_gen_ptr = (void *)(((unsigned long)code_gen_ptr + code_gen_size + CODE_GEN_ALIGN - 1) & ~(CODE_GEN_ALIGN - 1));
+    
+    /* check next page if needed */
+    virt_page2 = (pc + tb->size - 1) & TARGET_PAGE_MASK;
+    phys_page2 = -1;
+    if ((pc & TARGET_PAGE_MASK) != virt_page2) {
+        phys_page2 = get_phys_addr_code(env, virt_page2);
+    }
+    tb_link_phys(tb, phys_pc, phys_page2);
+}
 #endif
+    
+/* invalidate all TBs which intersect with the target physical page
+   starting in range [start;end[. NOTE: start and end must refer to
+   the same physical page. 'is_cpu_write_access' should be true if called
+   from a real cpu write access: the virtual CPU will exit the current
+   TB if code is modified inside this TB. */
+void tb_invalidate_phys_page_range(target_ulong start, target_ulong end, 
+                                   int is_cpu_write_access)
+{
+    int n, current_tb_modified, current_tb_not_found, current_flags;
+    CPUState *env = cpu_single_env;
+    PageDesc *p;
+    TranslationBlock *tb, *tb_next, *current_tb, *saved_tb;
+    target_ulong tb_start, tb_end;
+    target_ulong current_pc, current_cs_base;
 
-    page_index = address >> TARGET_PAGE_BITS;
-    p = page_find(page_index);
-    if (!p)
+    p = page_find(start >> TARGET_PAGE_BITS);
+    if (!p) 
         return;
+    if (!p->code_bitmap && 
+        ++p->code_write_count >= SMC_BITMAP_USE_THRESHOLD &&
+        is_cpu_write_access) {
+        /* build code bitmap */
+        build_page_bitmap(p);
+    }
+
+    /* we remove all the TBs in the range [start, end[ */
+    /* XXX: see if in some cases it could be faster to invalidate all the code */
+    current_tb_not_found = is_cpu_write_access;
+    current_tb_modified = 0;
+    current_tb = NULL; /* avoid warning */
+    current_pc = 0; /* avoid warning */
+    current_cs_base = 0; /* avoid warning */
+    current_flags = 0; /* avoid warning */
     tb = p->first_tb;
-    parity1 = page_index & 1;
-    parity2 = parity1 ^ 1;
     while (tb != NULL) {
-        tb_next = tb->page_next[parity1];
-        tb_invalidate(tb, parity2);
+        n = (long)tb & 3;
+        tb = (TranslationBlock *)((long)tb & ~3);
+        tb_next = tb->page_next[n];
+        /* NOTE: this is subtle as a TB may span two physical pages */
+        if (n == 0) {
+            /* NOTE: tb_end may be after the end of the page, but
+               it is not a problem */
+            tb_start = tb->page_addr[0] + (tb->pc & ~TARGET_PAGE_MASK);
+            tb_end = tb_start + tb->size;
+        } else {
+            tb_start = tb->page_addr[1];
+            tb_end = tb_start + ((tb->pc + tb->size) & ~TARGET_PAGE_MASK);
+        }
+        if (!(tb_end <= start || tb_start >= end)) {
+#ifdef TARGET_HAS_PRECISE_SMC
+            if (current_tb_not_found) {
+                current_tb_not_found = 0;
+                current_tb = NULL;
+                if (env->mem_write_pc) {
+                    /* now we have a real cpu fault */
+                    current_tb = tb_find_pc(env->mem_write_pc);
+                }
+            }
+            if (current_tb == tb &&
+                !(current_tb->cflags & CF_SINGLE_INSN)) {
+                /* If we are modifying the current TB, we must stop
+                its execution. We could be more precise by checking
+                that the modification is after the current PC, but it
+                would require a specialized function to partially
+                restore the CPU state */
+                
+                current_tb_modified = 1;
+                cpu_restore_state(current_tb, env, 
+                                  env->mem_write_pc, NULL);
+#if defined(TARGET_I386)
+                current_flags = env->hflags;
+                current_flags |= (env->eflags & (IOPL_MASK | TF_MASK | VM_MASK));
+                current_cs_base = (target_ulong)env->segs[R_CS].base;
+                current_pc = current_cs_base + env->eip;
+#else
+#error unsupported CPU
+#endif
+            }
+#endif /* TARGET_HAS_PRECISE_SMC */
+            saved_tb = env->current_tb;
+            env->current_tb = NULL;
+            tb_phys_invalidate(tb, -1);
+            env->current_tb = saved_tb;
+            if (env->interrupt_request && env->current_tb)
+                cpu_interrupt(env, env->interrupt_request);
+        }
         tb = tb_next;
     }
+#if !defined(CONFIG_USER_ONLY)
+    /* if no code remaining, no need to continue to use slow writes */
+    if (!p->first_tb) {
+        invalidate_page_bitmap(p);
+        if (is_cpu_write_access) {
+            tlb_unprotect_code_phys(env, start, env->mem_write_vaddr);
+        }
+    }
+#endif
+#ifdef TARGET_HAS_PRECISE_SMC
+    if (current_tb_modified) {
+        /* we generate a block containing just the instruction
+           modifying the memory. It will ensure that it cannot modify
+           itself */
+        env->current_tb = NULL;
+        tb_gen_code(env, current_pc, current_cs_base, current_flags, 
+                    CF_SINGLE_INSN);
+        cpu_resume_from_signal(env, NULL);
+    }
+#endif
+}
+
+/* len must be <= 8 and start must be a multiple of len */
+static inline void tb_invalidate_phys_page_fast(target_ulong start, int len)
+{
+    PageDesc *p;
+    int offset, b;
+#if 0
+    if (1) {
+        if (loglevel) {
+            fprintf(logfile, "modifying code at 0x%x size=%d EIP=%x PC=%08x\n", 
+                   cpu_single_env->mem_write_vaddr, len, 
+                   cpu_single_env->eip, 
+                   cpu_single_env->eip + (long)cpu_single_env->segs[R_CS].base);
+        }
+    }
+#endif
+    p = page_find(start >> TARGET_PAGE_BITS);
+    if (!p) 
+        return;
+    if (p->code_bitmap) {
+        offset = start & ~TARGET_PAGE_MASK;
+        b = p->code_bitmap[offset >> 3] >> (offset & 7);
+        if (b & ((1 << len) - 1))
+            goto do_invalidate;
+    } else {
+    do_invalidate:
+        tb_invalidate_phys_page_range(start, start + len, 1);
+    }
+}
+
+#if !defined(CONFIG_SOFTMMU)
+static void tb_invalidate_phys_page(target_ulong addr, 
+                                    unsigned long pc, void *puc)
+{
+    int n, current_flags, current_tb_modified;
+    target_ulong current_pc, current_cs_base;
+    PageDesc *p;
+    TranslationBlock *tb, *current_tb;
+#ifdef TARGET_HAS_PRECISE_SMC
+    CPUState *env = cpu_single_env;
+#endif
+
+    addr &= TARGET_PAGE_MASK;
+    p = page_find(addr >> TARGET_PAGE_BITS);
+    if (!p) 
+        return;
+    tb = p->first_tb;
+    current_tb_modified = 0;
+    current_tb = NULL;
+    current_pc = 0; /* avoid warning */
+    current_cs_base = 0; /* avoid warning */
+    current_flags = 0; /* avoid warning */
+#ifdef TARGET_HAS_PRECISE_SMC
+    if (tb && pc != 0) {
+        current_tb = tb_find_pc(pc);
+    }
+#endif
+    while (tb != NULL) {
+        n = (long)tb & 3;
+        tb = (TranslationBlock *)((long)tb & ~3);
+#ifdef TARGET_HAS_PRECISE_SMC
+        if (current_tb == tb &&
+            !(current_tb->cflags & CF_SINGLE_INSN)) {
+                /* If we are modifying the current TB, we must stop
+                   its execution. We could be more precise by checking
+                   that the modification is after the current PC, but it
+                   would require a specialized function to partially
+                   restore the CPU state */
+            
+            current_tb_modified = 1;
+            cpu_restore_state(current_tb, env, pc, puc);
+#if defined(TARGET_I386)
+            current_flags = env->hflags;
+            current_flags |= (env->eflags & (IOPL_MASK | TF_MASK | VM_MASK));
+            current_cs_base = (target_ulong)env->segs[R_CS].base;
+            current_pc = current_cs_base + env->eip;
+#else
+#error unsupported CPU
+#endif
+        }
+#endif /* TARGET_HAS_PRECISE_SMC */
+        tb_phys_invalidate(tb, addr);
+        tb = tb->page_next[n];
+    }
     p->first_tb = NULL;
+#ifdef TARGET_HAS_PRECISE_SMC
+    if (current_tb_modified) {
+        /* we generate a block containing just the instruction
+           modifying the memory. It will ensure that it cannot modify
+           itself */
+        env->current_tb = NULL;
+        tb_gen_code(env, current_pc, current_cs_base, current_flags, 
+                    CF_SINGLE_INSN);
+        cpu_resume_from_signal(env, puc);
+    }
+#endif
 }
+#endif
 
 /* add the tb in the target page and protect it if necessary */
-static inline void tb_alloc_page(TranslationBlock *tb, unsigned int page_index)
+static inline void tb_alloc_page(TranslationBlock *tb, 
+                                 unsigned int n, unsigned int page_addr)
 {
     PageDesc *p;
-    unsigned long host_start, host_end, addr, page_addr;
-    int prot;
+    TranslationBlock *last_first_tb;
+
+    tb->page_addr[n] = page_addr;
+    p = page_find(page_addr >> TARGET_PAGE_BITS);
+    tb->page_next[n] = p->first_tb;
+    last_first_tb = p->first_tb;
+    p->first_tb = (TranslationBlock *)((long)tb | n);
+    invalidate_page_bitmap(p);
 
-    p = page_find_alloc(page_index);
-    tb->page_next[page_index & 1] = p->first_tb;
-    p->first_tb = tb;
+#if defined(TARGET_HAS_SMC) || 1
+
+#if defined(CONFIG_USER_ONLY)
     if (p->flags & PAGE_WRITE) {
+        unsigned long host_start, host_end, addr;
+        int prot;
+
         /* force the host page as non writable (writes will have a
            page fault + mprotect overhead) */
-        page_addr = (page_index << TARGET_PAGE_BITS);
-        host_start = page_addr & host_page_mask;
-        host_end = host_start + host_page_size;
+        host_start = page_addr & qemu_host_page_mask;
+        host_end = host_start + qemu_host_page_size;
         prot = 0;
         for(addr = host_start; addr < host_end; addr += TARGET_PAGE_SIZE)
             prot |= page_get_flags(addr);
-        mprotect((void *)host_start, host_page_size, 
+        mprotect((void *)host_start, qemu_host_page_size, 
                  (prot & PAGE_BITS) & ~PAGE_WRITE);
 #ifdef DEBUG_TB_INVALIDATE
         printf("protecting code page: 0x%08lx\n", 
                host_start);
 #endif
         p->flags &= ~PAGE_WRITE;
-#ifdef DEBUG_TB_CHECK
-        tb_page_check();
-#endif
     }
+#else
+    /* if some code is already present, then the pages are already
+       protected. So we handle the case where only the first TB is
+       allocated in a physical page */
+    if (!last_first_tb) {
+        target_ulong virt_addr;
+
+        virt_addr = (tb->pc & TARGET_PAGE_MASK) + (n << TARGET_PAGE_BITS);
+        tlb_protect_code(cpu_single_env, virt_addr);        
+    }
+#endif
+
+#endif /* TARGET_HAS_SMC */
 }
 
 /* Allocate a new translation block. Flush the translation buffer if
    too many translation blocks or too much generated code. */
-TranslationBlock *tb_alloc(unsigned long pc)
+TranslationBlock *tb_alloc(target_ulong pc)
 {
     TranslationBlock *tb;
 
@@ -460,83 +906,96 @@ TranslationBlock *tb_alloc(unsigned long pc)
         return NULL;
     tb = &tbs[nb_tbs++];
     tb->pc = pc;
+    tb->cflags = 0;
     return tb;
 }
 
-/* link the tb with the other TBs */
-void tb_link(TranslationBlock *tb)
+/* add a new TB and link it to the physical page tables. phys_page2 is
+   (-1) to indicate that only one page contains the TB. */
+void tb_link_phys(TranslationBlock *tb, 
+                  target_ulong phys_pc, target_ulong phys_page2)
 {
-    unsigned int page_index1, page_index2;
+    unsigned int h;
+    TranslationBlock **ptb;
 
-    /* add in the page list */
-    page_index1 = tb->pc >> TARGET_PAGE_BITS;
-    tb_alloc_page(tb, page_index1);
-    page_index2 = (tb->pc + tb->size - 1) >> TARGET_PAGE_BITS;
-    if (page_index2 != page_index1) {
-        tb_alloc_page(tb, page_index2);
-    }
-    tb->jmp_first = (TranslationBlock *)((long)tb | 2);
-    tb->jmp_next[0] = NULL;
-    tb->jmp_next[1] = NULL;
+    /* add in the physical hash table */
+    h = tb_phys_hash_func(phys_pc);
+    ptb = &tb_phys_hash[h];
+    tb->phys_hash_next = *ptb;
+    *ptb = tb;
 
-    /* init original jump addresses */
-    if (tb->tb_next_offset[0] != 0xffff)
-        tb_reset_jump(tb, 0);
-    if (tb->tb_next_offset[1] != 0xffff)
-        tb_reset_jump(tb, 1);
+    /* add in the page list */
+    tb_alloc_page(tb, 0, phys_pc & TARGET_PAGE_MASK);
+    if (phys_page2 != -1)
+        tb_alloc_page(tb, 1, phys_page2);
+    else
+        tb->page_addr[1] = -1;
+#ifdef DEBUG_TB_CHECK
+    tb_page_check();
+#endif
 }
 
-/* called from signal handler: invalidate the code and unprotect the
-   page. Return TRUE if the fault was succesfully handled. */
-int page_unprotect(unsigned long address)
+/* link the tb with the other TBs */
+void tb_link(TranslationBlock *tb)
 {
-    unsigned int page_index, prot, pindex;
-    PageDesc *p, *p1;
-    unsigned long host_start, host_end, addr;
-
-    host_start = address & host_page_mask;
-    page_index = host_start >> TARGET_PAGE_BITS;
-    p1 = page_find(page_index);
-    if (!p1)
-        return 0;
-    host_end = host_start + host_page_size;
-    p = p1;
-    prot = 0;
-    for(addr = host_start;addr < host_end; addr += TARGET_PAGE_SIZE) {
-        prot |= p->flags;
-        p++;
-    }
-    /* if the page was really writable, then we change its
-       protection back to writable */
-    if (prot & PAGE_WRITE_ORG) {
-        mprotect((void *)host_start, host_page_size, 
-                 (prot & PAGE_BITS) | PAGE_WRITE);
-        pindex = (address - host_start) >> TARGET_PAGE_BITS;
-        p1[pindex].flags |= PAGE_WRITE;
-        /* and since the content will be modified, we must invalidate
-           the corresponding translated code. */
-        tb_invalidate_page(address);
-#ifdef DEBUG_TB_CHECK
-        tb_invalidate_check(address);
+#if !defined(CONFIG_USER_ONLY)
+    {
+        VirtPageDesc *vp;
+        target_ulong addr;
+        
+        /* save the code memory mappings (needed to invalidate the code) */
+        addr = tb->pc & TARGET_PAGE_MASK;
+        vp = virt_page_find_alloc(addr >> TARGET_PAGE_BITS);
+#ifdef DEBUG_TLB_CHECK 
+        if (vp->valid_tag == virt_valid_tag &&
+            vp->phys_addr != tb->page_addr[0]) {
+            printf("Error tb addr=0x%x phys=0x%x vp->phys_addr=0x%x\n",
+                   addr, tb->page_addr[0], vp->phys_addr);
+        }
 #endif
-        return 1;
-    } else {
-        return 0;
+        vp->phys_addr = tb->page_addr[0];
+        if (vp->valid_tag != virt_valid_tag) {
+            vp->valid_tag = virt_valid_tag;
+#if !defined(CONFIG_SOFTMMU)
+            vp->prot = 0;
+#endif
+        }
+        
+        if (tb->page_addr[1] != -1) {
+            addr += TARGET_PAGE_SIZE;
+            vp = virt_page_find_alloc(addr >> TARGET_PAGE_BITS);
+#ifdef DEBUG_TLB_CHECK 
+            if (vp->valid_tag == virt_valid_tag &&
+                vp->phys_addr != tb->page_addr[1]) { 
+                printf("Error tb addr=0x%x phys=0x%x vp->phys_addr=0x%x\n",
+                       addr, tb->page_addr[1], vp->phys_addr);
+            }
+#endif
+            vp->phys_addr = tb->page_addr[1];
+            if (vp->valid_tag != virt_valid_tag) {
+                vp->valid_tag = virt_valid_tag;
+#if !defined(CONFIG_SOFTMMU)
+                vp->prot = 0;
+#endif
+            }
+        }
     }
-}
+#endif
 
-/* call this function when system calls directly modify a memory area */
-void page_unprotect_range(uint8_t *data, unsigned long data_size)
-{
-    unsigned long start, end, addr;
+    tb->jmp_first = (TranslationBlock *)((long)tb | 2);
+    tb->jmp_next[0] = NULL;
+    tb->jmp_next[1] = NULL;
+#ifdef USE_CODE_COPY
+    tb->cflags &= ~CF_FP_USED;
+    if (tb->cflags & CF_TB_FP_USED)
+        tb->cflags |= CF_FP_USED;
+#endif
 
-    start = (unsigned long)data;
-    end = start + data_size;
-    start &= TARGET_PAGE_MASK;
-    end = TARGET_PAGE_ALIGN(end);
-    for(addr = start; addr < end; addr += TARGET_PAGE_SIZE) {
-        page_unprotect(addr);
-    }
+    /* init original jump addresses */
+    if (tb->tb_next_offset[0] != 0xffff)
+        tb_reset_jump(tb, 0);
+    if (tb->tb_next_offset[1] != 0xffff)
+        tb_reset_jump(tb, 1);
 }
 
 /* find the TB 'tb' such that tb[0].tc_ptr <= tc_ptr <
@@ -606,7 +1065,7 @@ static inline void tb_reset_jump_recursive2(TranslationBlock *tb, int n)
         /* suppress the jump to next tb in generated code */
         tb_reset_jump(tb, n);
 
-        /* suppress jumps in the tb on which we could have jump */
+        /* suppress jumps in the tb on which we could have jumped */
         tb_reset_jump_recursive(tb_next);
     }
 }
@@ -617,20 +1076,191 @@ static void tb_reset_jump_recursive(TranslationBlock *tb)
     tb_reset_jump_recursive2(tb, 1);
 }
 
-/* mask must never be zero */
+#if defined(TARGET_HAS_ICE)
+static void breakpoint_invalidate(CPUState *env, target_ulong pc)
+{
+    target_ulong phys_addr;
+
+    phys_addr = cpu_get_phys_page_debug(env, pc);
+    tb_invalidate_phys_page_range(phys_addr, phys_addr + 1, 0);
+}
+#endif
+
+/* add a breakpoint. EXCP_DEBUG is returned by the CPU loop if a
+   breakpoint is reached */
+int cpu_breakpoint_insert(CPUState *env, target_ulong pc)
+{
+#if defined(TARGET_HAS_ICE)
+    int i;
+    
+    for(i = 0; i < env->nb_breakpoints; i++) {
+        if (env->breakpoints[i] == pc)
+            return 0;
+    }
+
+    if (env->nb_breakpoints >= MAX_BREAKPOINTS)
+        return -1;
+    env->breakpoints[env->nb_breakpoints++] = pc;
+    
+    breakpoint_invalidate(env, pc);
+    return 0;
+#else
+    return -1;
+#endif
+}
+
+/* remove a breakpoint */
+int cpu_breakpoint_remove(CPUState *env, target_ulong pc)
+{
+#if defined(TARGET_HAS_ICE)
+    int i;
+    for(i = 0; i < env->nb_breakpoints; i++) {
+        if (env->breakpoints[i] == pc)
+            goto found;
+    }
+    return -1;
+ found:
+    env->nb_breakpoints--;
+    if (i < env->nb_breakpoints)
+      env->breakpoints[i] = env->breakpoints[env->nb_breakpoints];
+
+    breakpoint_invalidate(env, pc);
+    return 0;
+#else
+    return -1;
+#endif
+}
+
+/* enable or disable single step mode. EXCP_DEBUG is returned by the
+   CPU loop after each instruction */
+void cpu_single_step(CPUState *env, int enabled)
+{
+#if defined(TARGET_HAS_ICE)
+    if (env->singlestep_enabled != enabled) {
+        env->singlestep_enabled = enabled;
+        /* must flush all the translated code to avoid inconsistancies */
+        /* XXX: only flush what is necessary */
+        tb_flush(env);
+    }
+#endif
+}
+
+/* enable or disable low levels log */
+void cpu_set_log(int log_flags)
+{
+    loglevel = log_flags;
+    if (loglevel && !logfile) {
+        logfile = fopen(logfilename, "w");
+        if (!logfile) {
+            perror(logfilename);
+            _exit(1);
+        }
+#if !defined(CONFIG_SOFTMMU)
+        /* must avoid mmap() usage of glibc by setting a buffer "by hand" */
+        {
+            static uint8_t logfile_buf[4096];
+            setvbuf(logfile, logfile_buf, _IOLBF, sizeof(logfile_buf));
+        }
+#else
+        setvbuf(logfile, NULL, _IOLBF, 0);
+#endif
+    }
+}
+
+void cpu_set_log_filename(const char *filename)
+{
+    logfilename = strdup(filename);
+}
+
+/* mask must never be zero, except for A20 change call */
 void cpu_interrupt(CPUState *env, int mask)
 {
     TranslationBlock *tb;
-    
+    static int interrupt_lock;
+
     env->interrupt_request |= mask;
     /* if the cpu is currently executing code, we must unlink it and
        all the potentially executing TB */
     tb = env->current_tb;
-    if (tb) {
+    if (tb && !testandset(&interrupt_lock)) {
+        env->current_tb = NULL;
         tb_reset_jump_recursive(tb);
+        interrupt_lock = 0;
     }
 }
 
+void cpu_reset_interrupt(CPUState *env, int mask)
+{
+    env->interrupt_request &= ~mask;
+}
+
+CPULogItem cpu_log_items[] = {
+    { CPU_LOG_TB_OUT_ASM, "out_asm", 
+      "show generated host assembly code for each compiled TB" },
+    { CPU_LOG_TB_IN_ASM, "in_asm",
+      "show target assembly code for each compiled TB" },
+    { CPU_LOG_TB_OP, "op", 
+      "show micro ops for each compiled TB (only usable if 'in_asm' used)" },
+#ifdef TARGET_I386
+    { CPU_LOG_TB_OP_OPT, "op_opt",
+      "show micro ops after optimization for each compiled TB" },
+#endif
+    { CPU_LOG_INT, "int",
+      "show interrupts/exceptions in short format" },
+    { CPU_LOG_EXEC, "exec",
+      "show trace before each executed TB (lots of logs)" },
+    { CPU_LOG_TB_CPU, "cpu",
+      "show CPU state before bloc translation" },
+#ifdef TARGET_I386
+    { CPU_LOG_PCALL, "pcall",
+      "show protected mode far calls/returns/exceptions" },
+#endif
+#ifdef DEBUG_IOPORT
+    { CPU_LOG_IOPORT, "ioport",
+      "show all i/o ports accesses" },
+#endif
+    { 0, NULL, NULL },
+};
+
+static int cmp1(const char *s1, int n, const char *s2)
+{
+    if (strlen(s2) != n)
+        return 0;
+    return memcmp(s1, s2, n) == 0;
+}
+      
+/* takes a comma separated list of log masks. Return 0 if error. */
+int cpu_str_to_log_mask(const char *str)
+{
+    CPULogItem *item;
+    int mask;
+    const char *p, *p1;
+
+    p = str;
+    mask = 0;
+    for(;;) {
+        p1 = strchr(p, ',');
+        if (!p1)
+            p1 = p + strlen(p);
+       if(cmp1(p,p1-p,"all")) {
+               for(item = cpu_log_items; item->mask != 0; item++) {
+                       mask |= item->mask;
+               }
+       } else {
+        for(item = cpu_log_items; item->mask != 0; item++) {
+            if (cmp1(p, p1 - p, item->name))
+                goto found;
+        }
+        return 0;
+       }
+    found:
+        mask |= item->mask;
+        if (*p1 != ',')
+            break;
+        p = p1 + 1;
+    }
+    return mask;
+}
 
 void cpu_abort(CPUState *env, const char *fmt, ...)
 {
@@ -641,49 +1271,1069 @@ void cpu_abort(CPUState *env, const char *fmt, ...)
     vfprintf(stderr, fmt, ap);
     fprintf(stderr, "\n");
 #ifdef TARGET_I386
-    cpu_x86_dump_state(env, stderr, X86_DUMP_FPU | X86_DUMP_CCOP);
+    cpu_dump_state(env, stderr, fprintf, X86_DUMP_FPU | X86_DUMP_CCOP);
+#else
+    cpu_dump_state(env, stderr, fprintf, 0);
 #endif
     va_end(ap);
     abort();
 }
 
-#ifdef TARGET_I386
-/* unmap all maped pages and flush all associated code */
-void page_unmap(void)
+#if !defined(CONFIG_USER_ONLY)
+
+/* NOTE: if flush_global is true, also flush global entries (not
+   implemented yet) */
+void tlb_flush(CPUState *env, int flush_global)
+{
+    int i;
+
+#if defined(DEBUG_TLB)
+    printf("tlb_flush:\n");
+#endif
+    /* must reset current TB so that interrupts cannot modify the
+       links while we are modifying them */
+    env->current_tb = NULL;
+
+    for(i = 0; i < CPU_TLB_SIZE; i++) {
+        env->tlb_read[0][i].address = -1;
+        env->tlb_write[0][i].address = -1;
+        env->tlb_read[1][i].address = -1;
+        env->tlb_write[1][i].address = -1;
+    }
+
+    virt_page_flush();
+    memset (tb_hash, 0, CODE_GEN_HASH_SIZE * sizeof (void *));
+
+#if !defined(CONFIG_SOFTMMU)
+    munmap((void *)MMAP_AREA_START, MMAP_AREA_END - MMAP_AREA_START);
+#endif
+#ifdef USE_KQEMU
+    if (env->kqemu_enabled) {
+        kqemu_flush(env, flush_global);
+    }
+#endif
+    tlb_flush_count++;
+}
+
+static inline void tlb_flush_entry(CPUTLBEntry *tlb_entry, target_ulong addr)
+{
+    if (addr == (tlb_entry->address & 
+                 (TARGET_PAGE_MASK | TLB_INVALID_MASK)))
+        tlb_entry->address = -1;
+}
+
+void tlb_flush_page(CPUState *env, target_ulong addr)
+{
+    int i, n;
+    VirtPageDesc *vp;
+    PageDesc *p;
+    TranslationBlock *tb;
+
+#if defined(DEBUG_TLB)
+    printf("tlb_flush_page: 0x%08x\n", addr);
+#endif
+    /* must reset current TB so that interrupts cannot modify the
+       links while we are modifying them */
+    env->current_tb = NULL;
+
+    addr &= TARGET_PAGE_MASK;
+    i = (addr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
+    tlb_flush_entry(&env->tlb_read[0][i], addr);
+    tlb_flush_entry(&env->tlb_write[0][i], addr);
+    tlb_flush_entry(&env->tlb_read[1][i], addr);
+    tlb_flush_entry(&env->tlb_write[1][i], addr);
+
+    /* remove from the virtual pc hash table all the TB at this
+       virtual address */
+    
+    vp = virt_page_find(addr >> TARGET_PAGE_BITS);
+    if (vp && vp->valid_tag == virt_valid_tag) {
+        p = page_find(vp->phys_addr >> TARGET_PAGE_BITS);
+        if (p) {
+            /* we remove all the links to the TBs in this virtual page */
+            tb = p->first_tb;
+            while (tb != NULL) {
+                n = (long)tb & 3;
+                tb = (TranslationBlock *)((long)tb & ~3);
+                if ((tb->pc & TARGET_PAGE_MASK) == addr ||
+                    ((tb->pc + tb->size - 1) & TARGET_PAGE_MASK) == addr) {
+                    tb_invalidate(tb);
+                }
+                tb = tb->page_next[n];
+            }
+        }
+        vp->valid_tag = 0;
+    }
+
+#if !defined(CONFIG_SOFTMMU)
+    if (addr < MMAP_AREA_END)
+        munmap((void *)addr, TARGET_PAGE_SIZE);
+#endif
+#ifdef USE_KQEMU
+    if (env->kqemu_enabled) {
+        kqemu_flush_page(env, addr);
+    }
+#endif
+}
+
+static inline void tlb_protect_code1(CPUTLBEntry *tlb_entry, target_ulong addr)
+{
+    if (addr == (tlb_entry->address & 
+                 (TARGET_PAGE_MASK | TLB_INVALID_MASK)) &&
+        (tlb_entry->address & ~TARGET_PAGE_MASK) != IO_MEM_CODE &&
+        (tlb_entry->address & ~TARGET_PAGE_MASK) != IO_MEM_ROM) {
+        tlb_entry->address = (tlb_entry->address & TARGET_PAGE_MASK) | IO_MEM_CODE;
+    }
+}
+
+/* update the TLBs so that writes to code in the virtual page 'addr'
+   can be detected */
+static void tlb_protect_code(CPUState *env, target_ulong addr)
+{
+    int i;
+
+    addr &= TARGET_PAGE_MASK;
+    i = (addr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
+    tlb_protect_code1(&env->tlb_write[0][i], addr);
+    tlb_protect_code1(&env->tlb_write[1][i], addr);
+#if !defined(CONFIG_SOFTMMU)
+    /* NOTE: as we generated the code for this page, it is already at
+       least readable */
+    if (addr < MMAP_AREA_END)
+        mprotect((void *)addr, TARGET_PAGE_SIZE, PROT_READ);
+#endif
+}
+
+static inline void tlb_unprotect_code2(CPUTLBEntry *tlb_entry, 
+                                       unsigned long phys_addr)
+{
+    if ((tlb_entry->address & ~TARGET_PAGE_MASK) == IO_MEM_CODE &&
+        ((tlb_entry->address & TARGET_PAGE_MASK) + tlb_entry->addend) == phys_addr) {
+        tlb_entry->address = (tlb_entry->address & TARGET_PAGE_MASK) | IO_MEM_NOTDIRTY;
+    }
+}
+
+/* update the TLB so that writes in physical page 'phys_addr' are no longer
+   tested self modifying code */
+static void tlb_unprotect_code_phys(CPUState *env, unsigned long phys_addr, target_ulong vaddr)
+{
+    int i;
+
+    phys_addr &= TARGET_PAGE_MASK;
+    phys_addr += (long)phys_ram_base;
+    i = (vaddr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
+    tlb_unprotect_code2(&env->tlb_write[0][i], phys_addr);
+    tlb_unprotect_code2(&env->tlb_write[1][i], phys_addr);
+}
+
+static inline void tlb_reset_dirty_range(CPUTLBEntry *tlb_entry, 
+                                         unsigned long start, unsigned long length)
 {
-    PageDesc *p, *pmap;
     unsigned long addr;
-    int i, j, ret, j1;
+    if ((tlb_entry->address & ~TARGET_PAGE_MASK) == IO_MEM_RAM) {
+        addr = (tlb_entry->address & TARGET_PAGE_MASK) + tlb_entry->addend;
+        if ((addr - start) < length) {
+            tlb_entry->address = (tlb_entry->address & TARGET_PAGE_MASK) | IO_MEM_NOTDIRTY;
+        }
+    }
+}
 
-    for(i = 0; i < L1_SIZE; i++) {
-        pmap = l1_map[i];
-        if (pmap) {
-            p = pmap;
-            for(j = 0;j < L2_SIZE;) {
-                if (p->flags & PAGE_VALID) {
-                    addr = (i << (32 - L1_BITS)) | (j << TARGET_PAGE_BITS);
-                    /* we try to find a range to make less syscalls */
-                    j1 = j;
-                    p++;
-                    j++;
-                    while (j < L2_SIZE && (p->flags & PAGE_VALID)) {
-                        p++;
-                        j++;
-                    }
-                    ret = munmap((void *)addr, (j - j1) << TARGET_PAGE_BITS);
-                    if (ret != 0) {
-                        fprintf(stderr, "Could not unmap page 0x%08lx\n", addr);
-                        exit(1);
+void cpu_physical_memory_reset_dirty(target_ulong start, target_ulong end,
+                                     int dirty_flags)
+{
+    CPUState *env;
+    unsigned long length, start1;
+    int i, mask, len;
+    uint8_t *p;
+
+    start &= TARGET_PAGE_MASK;
+    end = TARGET_PAGE_ALIGN(end);
+
+    length = end - start;
+    if (length == 0)
+        return;
+    mask = ~dirty_flags;
+    p = phys_ram_dirty + (start >> TARGET_PAGE_BITS);
+    len = length >> TARGET_PAGE_BITS;
+    for(i = 0; i < len; i++)
+        p[i] &= mask;
+
+    env = cpu_single_env;
+    /* we modify the TLB cache so that the dirty bit will be set again
+       when accessing the range */
+    start1 = start + (unsigned long)phys_ram_base;
+    for(i = 0; i < CPU_TLB_SIZE; i++)
+        tlb_reset_dirty_range(&env->tlb_write[0][i], start1, length);
+    for(i = 0; i < CPU_TLB_SIZE; i++)
+        tlb_reset_dirty_range(&env->tlb_write[1][i], start1, length);
+
+#if !defined(CONFIG_SOFTMMU)
+    /* XXX: this is expensive */
+    {
+        VirtPageDesc *p;
+        int j;
+        target_ulong addr;
+
+        for(i = 0; i < L1_SIZE; i++) {
+            p = l1_virt_map[i];
+            if (p) {
+                addr = i << (TARGET_PAGE_BITS + L2_BITS);
+                for(j = 0; j < L2_SIZE; j++) {
+                    if (p->valid_tag == virt_valid_tag &&
+                        p->phys_addr >= start && p->phys_addr < end &&
+                        (p->prot & PROT_WRITE)) {
+                        if (addr < MMAP_AREA_END) {
+                            mprotect((void *)addr, TARGET_PAGE_SIZE, 
+                                     p->prot & ~PROT_WRITE);
+                        }
                     }
-                } else {
+                    addr += TARGET_PAGE_SIZE;
                     p++;
-                    j++;
                 }
             }
-            free(pmap);
-            l1_map[i] = NULL;
         }
     }
-    tb_flush();
+#endif
+}
+
+static inline void tlb_set_dirty1(CPUTLBEntry *tlb_entry, 
+                                    unsigned long start)
+{
+    unsigned long addr;
+    if ((tlb_entry->address & ~TARGET_PAGE_MASK) == IO_MEM_NOTDIRTY) {
+        addr = (tlb_entry->address & TARGET_PAGE_MASK) + tlb_entry->addend;
+        if (addr == start) {
+            tlb_entry->address = (tlb_entry->address & TARGET_PAGE_MASK) | IO_MEM_RAM;
+        }
+    }
+}
+
+/* update the TLB corresponding to virtual page vaddr and phys addr
+   addr so that it is no longer dirty */
+static inline void tlb_set_dirty(unsigned long addr, target_ulong vaddr)
+{
+    CPUState *env = cpu_single_env;
+    int i;
+
+    phys_ram_dirty[(addr - (unsigned long)phys_ram_base) >> TARGET_PAGE_BITS] = 0xff;
+
+    addr &= TARGET_PAGE_MASK;
+    i = (vaddr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
+    tlb_set_dirty1(&env->tlb_write[0][i], addr);
+    tlb_set_dirty1(&env->tlb_write[1][i], addr);
+}
+
+/* add a new TLB entry. At most one entry for a given virtual address
+   is permitted. Return 0 if OK or 2 if the page could not be mapped
+   (can only happen in non SOFTMMU mode for I/O pages or pages
+   conflicting with the host address space). */
+int tlb_set_page(CPUState *env, target_ulong vaddr, 
+                 target_phys_addr_t paddr, int prot, 
+                 int is_user, int is_softmmu)
+{
+    PhysPageDesc *p;
+    unsigned long pd;
+    TranslationBlock *first_tb;
+    unsigned int index;
+    target_ulong address;
+    unsigned long addend;
+    int ret;
+
+    p = phys_page_find(paddr >> TARGET_PAGE_BITS);
+    first_tb = NULL;
+    if (!p) {
+        pd = IO_MEM_UNASSIGNED;
+    } else {
+        PageDesc *p1;
+        pd = p->phys_offset;
+        if ((pd & ~TARGET_PAGE_MASK) <= IO_MEM_ROM) {
+            /* NOTE: we also allocate the page at this stage */
+            p1 = page_find_alloc(pd >> TARGET_PAGE_BITS);
+            first_tb = p1->first_tb;
+        }
+    }
+#if defined(DEBUG_TLB)
+    printf("tlb_set_page: vaddr=0x%08x paddr=0x%08x prot=%x u=%d c=%d smmu=%d pd=0x%08x\n",
+           vaddr, paddr, prot, is_user, (first_tb != NULL), is_softmmu, pd);
+#endif
+
+    ret = 0;
+#if !defined(CONFIG_SOFTMMU)
+    if (is_softmmu) 
+#endif
+    {
+        if ((pd & ~TARGET_PAGE_MASK) > IO_MEM_ROM) {
+            /* IO memory case */
+            address = vaddr | pd;
+            addend = paddr;
+        } else {
+            /* standard memory */
+            address = vaddr;
+            addend = (unsigned long)phys_ram_base + (pd & TARGET_PAGE_MASK);
+        }
+        
+        index = (vaddr >> 12) & (CPU_TLB_SIZE - 1);
+        addend -= vaddr;
+        if (prot & PAGE_READ) {
+            env->tlb_read[is_user][index].address = address;
+            env->tlb_read[is_user][index].addend = addend;
+        } else {
+            env->tlb_read[is_user][index].address = -1;
+            env->tlb_read[is_user][index].addend = -1;
+        }
+        if (prot & PAGE_WRITE) {
+            if ((pd & ~TARGET_PAGE_MASK) == IO_MEM_ROM) {
+                /* ROM: access is ignored (same as unassigned) */
+                env->tlb_write[is_user][index].address = vaddr | IO_MEM_ROM;
+                env->tlb_write[is_user][index].addend = addend;
+            } else 
+                /* XXX: the PowerPC code seems not ready to handle
+                   self modifying code with DCBI */
+#if defined(TARGET_HAS_SMC) || 1
+            if (first_tb) {
+                /* if code is present, we use a specific memory
+                   handler. It works only for physical memory access */
+                env->tlb_write[is_user][index].address = vaddr | IO_MEM_CODE;
+                env->tlb_write[is_user][index].addend = addend;
+            } else 
+#endif
+            if ((pd & ~TARGET_PAGE_MASK) == IO_MEM_RAM && 
+                       !cpu_physical_memory_is_dirty(pd)) {
+                env->tlb_write[is_user][index].address = vaddr | IO_MEM_NOTDIRTY;
+                env->tlb_write[is_user][index].addend = addend;
+            } else {
+                env->tlb_write[is_user][index].address = address;
+                env->tlb_write[is_user][index].addend = addend;
+            }
+        } else {
+            env->tlb_write[is_user][index].address = -1;
+            env->tlb_write[is_user][index].addend = -1;
+        }
+    }
+#if !defined(CONFIG_SOFTMMU)
+    else {
+        if ((pd & ~TARGET_PAGE_MASK) > IO_MEM_ROM) {
+            /* IO access: no mapping is done as it will be handled by the
+               soft MMU */
+            if (!(env->hflags & HF_SOFTMMU_MASK))
+                ret = 2;
+        } else {
+            void *map_addr;
+
+            if (vaddr >= MMAP_AREA_END) {
+                ret = 2;
+            } else {
+                if (prot & PROT_WRITE) {
+                    if ((pd & ~TARGET_PAGE_MASK) == IO_MEM_ROM || 
+#if defined(TARGET_HAS_SMC) || 1
+                        first_tb ||
+#endif
+                        ((pd & ~TARGET_PAGE_MASK) == IO_MEM_RAM && 
+                         !cpu_physical_memory_is_dirty(pd))) {
+                        /* ROM: we do as if code was inside */
+                        /* if code is present, we only map as read only and save the
+                           original mapping */
+                        VirtPageDesc *vp;
+                        
+                        vp = virt_page_find_alloc(vaddr >> TARGET_PAGE_BITS);
+                        vp->phys_addr = pd;
+                        vp->prot = prot;
+                        vp->valid_tag = virt_valid_tag;
+                        prot &= ~PAGE_WRITE;
+                    }
+                }
+                map_addr = mmap((void *)vaddr, TARGET_PAGE_SIZE, prot, 
+                                MAP_SHARED | MAP_FIXED, phys_ram_fd, (pd & TARGET_PAGE_MASK));
+                if (map_addr == MAP_FAILED) {
+                    cpu_abort(env, "mmap failed when mapped physical address 0x%08x to virtual address 0x%08x\n",
+                              paddr, vaddr);
+                }
+            }
+        }
+    }
+#endif
+    return ret;
+}
+
+/* called from signal handler: invalidate the code and unprotect the
+   page. Return TRUE if the fault was succesfully handled. */
+int page_unprotect(unsigned long addr, unsigned long pc, void *puc)
+{
+#if !defined(CONFIG_SOFTMMU)
+    VirtPageDesc *vp;
+
+#if defined(DEBUG_TLB)
+    printf("page_unprotect: addr=0x%08x\n", addr);
+#endif
+    addr &= TARGET_PAGE_MASK;
+
+    /* if it is not mapped, no need to worry here */
+    if (addr >= MMAP_AREA_END)
+        return 0;
+    vp = virt_page_find(addr >> TARGET_PAGE_BITS);
+    if (!vp)
+        return 0;
+    /* NOTE: in this case, validate_tag is _not_ tested as it
+       validates only the code TLB */
+    if (vp->valid_tag != virt_valid_tag)
+        return 0;
+    if (!(vp->prot & PAGE_WRITE))
+        return 0;
+#if defined(DEBUG_TLB)
+    printf("page_unprotect: addr=0x%08x phys_addr=0x%08x prot=%x\n", 
+           addr, vp->phys_addr, vp->prot);
+#endif
+    if (mprotect((void *)addr, TARGET_PAGE_SIZE, vp->prot) < 0)
+        cpu_abort(cpu_single_env, "error mprotect addr=0x%lx prot=%d\n",
+                  (unsigned long)addr, vp->prot);
+    /* set the dirty bit */
+    phys_ram_dirty[vp->phys_addr >> TARGET_PAGE_BITS] = 0xff;
+    /* flush the code inside */
+    tb_invalidate_phys_page(vp->phys_addr, pc, puc);
+    return 1;
+#else
+    return 0;
+#endif
 }
+
+#else
+
+void tlb_flush(CPUState *env, int flush_global)
+{
+}
+
+void tlb_flush_page(CPUState *env, target_ulong addr)
+{
+}
+
+int tlb_set_page(CPUState *env, target_ulong vaddr, 
+                 target_phys_addr_t paddr, int prot, 
+                 int is_user, int is_softmmu)
+{
+    return 0;
+}
+
+/* dump memory mappings */
+void page_dump(FILE *f)
+{
+    unsigned long start, end;
+    int i, j, prot, prot1;
+    PageDesc *p;
+
+    fprintf(f, "%-8s %-8s %-8s %s\n",
+            "start", "end", "size", "prot");
+    start = -1;
+    end = -1;
+    prot = 0;
+    for(i = 0; i <= L1_SIZE; i++) {
+        if (i < L1_SIZE)
+            p = l1_map[i];
+        else
+            p = NULL;
+        for(j = 0;j < L2_SIZE; j++) {
+            if (!p)
+                prot1 = 0;
+            else
+                prot1 = p[j].flags;
+            if (prot1 != prot) {
+                end = (i << (32 - L1_BITS)) | (j << TARGET_PAGE_BITS);
+                if (start != -1) {
+                    fprintf(f, "%08lx-%08lx %08lx %c%c%c\n",
+                            start, end, end - start, 
+                            prot & PAGE_READ ? 'r' : '-',
+                            prot & PAGE_WRITE ? 'w' : '-',
+                            prot & PAGE_EXEC ? 'x' : '-');
+                }
+                if (prot1 != 0)
+                    start = end;
+                else
+                    start = -1;
+                prot = prot1;
+            }
+            if (!p)
+                break;
+        }
+    }
+}
+
+int page_get_flags(unsigned long address)
+{
+    PageDesc *p;
+
+    p = page_find(address >> TARGET_PAGE_BITS);
+    if (!p)
+        return 0;
+    return p->flags;
+}
+
+/* modify the flags of a page and invalidate the code if
+   necessary. The flag PAGE_WRITE_ORG is positionned automatically
+   depending on PAGE_WRITE */
+void page_set_flags(unsigned long start, unsigned long end, int flags)
+{
+    PageDesc *p;
+    unsigned long addr;
+
+    start = start & TARGET_PAGE_MASK;
+    end = TARGET_PAGE_ALIGN(end);
+    if (flags & PAGE_WRITE)
+        flags |= PAGE_WRITE_ORG;
+    spin_lock(&tb_lock);
+    for(addr = start; addr < end; addr += TARGET_PAGE_SIZE) {
+        p = page_find_alloc(addr >> TARGET_PAGE_BITS);
+        /* if the write protection is set, then we invalidate the code
+           inside */
+        if (!(p->flags & PAGE_WRITE) && 
+            (flags & PAGE_WRITE) &&
+            p->first_tb) {
+            tb_invalidate_phys_page(addr, 0, NULL);
+        }
+        p->flags = flags;
+    }
+    spin_unlock(&tb_lock);
+}
+
+/* called from signal handler: invalidate the code and unprotect the
+   page. Return TRUE if the fault was succesfully handled. */
+int page_unprotect(unsigned long address, unsigned long pc, void *puc)
+{
+    unsigned int page_index, prot, pindex;
+    PageDesc *p, *p1;
+    unsigned long host_start, host_end, addr;
+
+    host_start = address & qemu_host_page_mask;
+    page_index = host_start >> TARGET_PAGE_BITS;
+    p1 = page_find(page_index);
+    if (!p1)
+        return 0;
+    host_end = host_start + qemu_host_page_size;
+    p = p1;
+    prot = 0;
+    for(addr = host_start;addr < host_end; addr += TARGET_PAGE_SIZE) {
+        prot |= p->flags;
+        p++;
+    }
+    /* if the page was really writable, then we change its
+       protection back to writable */
+    if (prot & PAGE_WRITE_ORG) {
+        pindex = (address - host_start) >> TARGET_PAGE_BITS;
+        if (!(p1[pindex].flags & PAGE_WRITE)) {
+            mprotect((void *)host_start, qemu_host_page_size, 
+                     (prot & PAGE_BITS) | PAGE_WRITE);
+            p1[pindex].flags |= PAGE_WRITE;
+            /* and since the content will be modified, we must invalidate
+               the corresponding translated code. */
+            tb_invalidate_phys_page(address, pc, puc);
+#ifdef DEBUG_TB_CHECK
+            tb_invalidate_check(address);
+#endif
+            return 1;
+        }
+    }
+    return 0;
+}
+
+/* call this function when system calls directly modify a memory area */
+void page_unprotect_range(uint8_t *data, unsigned long data_size)
+{
+    unsigned long start, end, addr;
+
+    start = (unsigned long)data;
+    end = start + data_size;
+    start &= TARGET_PAGE_MASK;
+    end = TARGET_PAGE_ALIGN(end);
+    for(addr = start; addr < end; addr += TARGET_PAGE_SIZE) {
+        page_unprotect(addr, 0, NULL);
+    }
+}
+
+static inline void tlb_set_dirty(unsigned long addr, target_ulong vaddr)
+{
+}
+#endif /* defined(CONFIG_USER_ONLY) */
+
+/* register physical memory. 'size' must be a multiple of the target
+   page size. If (phys_offset & ~TARGET_PAGE_MASK) != 0, then it is an
+   io memory page */
+void cpu_register_physical_memory(target_phys_addr_t start_addr, 
+                                  unsigned long size,
+                                  unsigned long phys_offset)
+{
+    unsigned long addr, end_addr;
+    PhysPageDesc *p;
+
+    size = (size + TARGET_PAGE_SIZE - 1) & TARGET_PAGE_MASK;
+    end_addr = start_addr + size;
+    for(addr = start_addr; addr != end_addr; addr += TARGET_PAGE_SIZE) {
+        p = phys_page_find_alloc(addr >> TARGET_PAGE_BITS);
+        p->phys_offset = phys_offset;
+        if ((phys_offset & ~TARGET_PAGE_MASK) <= IO_MEM_ROM)
+            phys_offset += TARGET_PAGE_SIZE;
+    }
+}
+
+static uint32_t unassigned_mem_readb(void *opaque, target_phys_addr_t addr)
+{
+    return 0;
+}
+
+static void unassigned_mem_writeb(void *opaque, target_phys_addr_t addr, uint32_t val)
+{
+}
+
+static CPUReadMemoryFunc *unassigned_mem_read[3] = {
+    unassigned_mem_readb,
+    unassigned_mem_readb,
+    unassigned_mem_readb,
+};
+
+static CPUWriteMemoryFunc *unassigned_mem_write[3] = {
+    unassigned_mem_writeb,
+    unassigned_mem_writeb,
+    unassigned_mem_writeb,
+};
+
+/* self modifying code support in soft mmu mode : writing to a page
+   containing code comes to these functions */
+
+static void code_mem_writeb(void *opaque, target_phys_addr_t addr, uint32_t val)
+{
+    unsigned long phys_addr;
+
+    phys_addr = addr - (unsigned long)phys_ram_base;
+#if !defined(CONFIG_USER_ONLY)
+    tb_invalidate_phys_page_fast(phys_addr, 1);
+#endif
+    stb_p((uint8_t *)(long)addr, val);
+    phys_ram_dirty[phys_addr >> TARGET_PAGE_BITS] = 0xff;
+}
+
+static void code_mem_writew(void *opaque, target_phys_addr_t addr, uint32_t val)
+{
+    unsigned long phys_addr;
+
+    phys_addr = addr - (unsigned long)phys_ram_base;
+#if !defined(CONFIG_USER_ONLY)
+    tb_invalidate_phys_page_fast(phys_addr, 2);
+#endif
+    stw_p((uint8_t *)(long)addr, val);
+    phys_ram_dirty[phys_addr >> TARGET_PAGE_BITS] = 0xff;
+}
+
+static void code_mem_writel(void *opaque, target_phys_addr_t addr, uint32_t val)
+{
+    unsigned long phys_addr;
+
+    phys_addr = addr - (unsigned long)phys_ram_base;
+#if !defined(CONFIG_USER_ONLY)
+    tb_invalidate_phys_page_fast(phys_addr, 4);
+#endif
+    stl_p((uint8_t *)(long)addr, val);
+    phys_ram_dirty[phys_addr >> TARGET_PAGE_BITS] = 0xff;
+}
+
+static CPUReadMemoryFunc *code_mem_read[3] = {
+    NULL, /* never used */
+    NULL, /* never used */
+    NULL, /* never used */
+};
+
+static CPUWriteMemoryFunc *code_mem_write[3] = {
+    code_mem_writeb,
+    code_mem_writew,
+    code_mem_writel,
+};
+
+static void notdirty_mem_writeb(void *opaque, target_phys_addr_t addr, uint32_t val)
+{
+    stb_p((uint8_t *)(long)addr, val);
+    tlb_set_dirty(addr, cpu_single_env->mem_write_vaddr);
+}
+
+static void notdirty_mem_writew(void *opaque, target_phys_addr_t addr, uint32_t val)
+{
+    stw_p((uint8_t *)(long)addr, val);
+    tlb_set_dirty(addr, cpu_single_env->mem_write_vaddr);
+}
+
+static void notdirty_mem_writel(void *opaque, target_phys_addr_t addr, uint32_t val)
+{
+    stl_p((uint8_t *)(long)addr, val);
+    tlb_set_dirty(addr, cpu_single_env->mem_write_vaddr);
+}
+
+static CPUWriteMemoryFunc *notdirty_mem_write[3] = {
+    notdirty_mem_writeb,
+    notdirty_mem_writew,
+    notdirty_mem_writel,
+};
+
+static void io_mem_init(void)
+{
+    cpu_register_io_memory(IO_MEM_ROM >> IO_MEM_SHIFT, code_mem_read, unassigned_mem_write, NULL);
+    cpu_register_io_memory(IO_MEM_UNASSIGNED >> IO_MEM_SHIFT, unassigned_mem_read, unassigned_mem_write, NULL);
+    cpu_register_io_memory(IO_MEM_CODE >> IO_MEM_SHIFT, code_mem_read, code_mem_write, NULL);
+    cpu_register_io_memory(IO_MEM_NOTDIRTY >> IO_MEM_SHIFT, code_mem_read, notdirty_mem_write, NULL);
+    io_mem_nb = 5;
+
+    /* alloc dirty bits array */
+    phys_ram_dirty = qemu_vmalloc(phys_ram_size >> TARGET_PAGE_BITS);
+}
+
+/* mem_read and mem_write are arrays of functions containing the
+   function to access byte (index 0), word (index 1) and dword (index
+   2). All functions must be supplied. If io_index is non zero, the
+   corresponding io zone is modified. If it is zero, a new io zone is
+   allocated. The return value can be used with
+   cpu_register_physical_memory(). (-1) is returned if error. */
+int cpu_register_io_memory(int io_index,
+                           CPUReadMemoryFunc **mem_read,
+                           CPUWriteMemoryFunc **mem_write,
+                           void *opaque)
+{
+    int i;
+
+    if (io_index <= 0) {
+        if (io_index >= IO_MEM_NB_ENTRIES)
+            return -1;
+        io_index = io_mem_nb++;
+    } else {
+        if (io_index >= IO_MEM_NB_ENTRIES)
+            return -1;
+    }
+    
+    for(i = 0;i < 3; i++) {
+        io_mem_read[io_index][i] = mem_read[i];
+        io_mem_write[io_index][i] = mem_write[i];
+    }
+    io_mem_opaque[io_index] = opaque;
+    return io_index << IO_MEM_SHIFT;
+}
+
+CPUWriteMemoryFunc **cpu_get_io_memory_write(int io_index)
+{
+    return io_mem_write[io_index >> IO_MEM_SHIFT];
+}
+
+CPUReadMemoryFunc **cpu_get_io_memory_read(int io_index)
+{
+    return io_mem_read[io_index >> IO_MEM_SHIFT];
+}
+
+/* physical memory access (slow version, mainly for debug) */
+#if defined(CONFIG_USER_ONLY)
+void cpu_physical_memory_rw(target_phys_addr_t addr, uint8_t *buf, 
+                            int len, int is_write)
+{
+    int l, flags;
+    target_ulong page;
+
+    while (len > 0) {
+        page = addr & TARGET_PAGE_MASK;
+        l = (page + TARGET_PAGE_SIZE) - addr;
+        if (l > len)
+            l = len;
+        flags = page_get_flags(page);
+        if (!(flags & PAGE_VALID))
+            return;
+        if (is_write) {
+            if (!(flags & PAGE_WRITE))
+                return;
+            memcpy((uint8_t *)addr, buf, len);
+        } else {
+            if (!(flags & PAGE_READ))
+                return;
+            memcpy(buf, (uint8_t *)addr, len);
+        }
+        len -= l;
+        buf += l;
+        addr += l;
+    }
+}
+
+/* never used */
+uint32_t ldl_phys(target_phys_addr_t addr)
+{
+    return 0;
+}
+
+void stl_phys_notdirty(target_phys_addr_t addr, uint32_t val)
+{
+}
+
+void stl_phys(target_phys_addr_t addr, uint32_t val)
+{
+}
+
+#else
+void cpu_physical_memory_rw(target_phys_addr_t addr, uint8_t *buf, 
+                            int len, int is_write)
+{
+    int l, io_index;
+    uint8_t *ptr;
+    uint32_t val;
+    target_phys_addr_t page;
+    unsigned long pd;
+    PhysPageDesc *p;
+    
+    while (len > 0) {
+        page = addr & TARGET_PAGE_MASK;
+        l = (page + TARGET_PAGE_SIZE) - addr;
+        if (l > len)
+            l = len;
+        p = phys_page_find(page >> TARGET_PAGE_BITS);
+        if (!p) {
+            pd = IO_MEM_UNASSIGNED;
+        } else {
+            pd = p->phys_offset;
+        }
+        
+        if (is_write) {
+            if ((pd & ~TARGET_PAGE_MASK) != 0) {
+                io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
+                if (l >= 4 && ((addr & 3) == 0)) {
+                    /* 32 bit read access */
+                    val = ldl_p(buf);
+                    io_mem_write[io_index][2](io_mem_opaque[io_index], addr, val);
+                    l = 4;
+                } else if (l >= 2 && ((addr & 1) == 0)) {
+                    /* 16 bit read access */
+                    val = lduw_p(buf);
+                    io_mem_write[io_index][1](io_mem_opaque[io_index], addr, val);
+                    l = 2;
+                } else {
+                    /* 8 bit access */
+                    val = ldub_p(buf);
+                    io_mem_write[io_index][0](io_mem_opaque[io_index], addr, val);
+                    l = 1;
+                }
+            } else {
+                unsigned long addr1;
+                addr1 = (pd & TARGET_PAGE_MASK) + (addr & ~TARGET_PAGE_MASK);
+                /* RAM case */
+                ptr = phys_ram_base + addr1;
+                memcpy(ptr, buf, l);
+                /* invalidate code */
+                tb_invalidate_phys_page_range(addr1, addr1 + l, 0);
+                /* set dirty bit */
+                phys_ram_dirty[addr1 >> TARGET_PAGE_BITS] = 0xff;
+            }
+        } else {
+            if ((pd & ~TARGET_PAGE_MASK) > IO_MEM_ROM &&
+                (pd & ~TARGET_PAGE_MASK) != IO_MEM_CODE) {
+                /* I/O case */
+                io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
+                if (l >= 4 && ((addr & 3) == 0)) {
+                    /* 32 bit read access */
+                    val = io_mem_read[io_index][2](io_mem_opaque[io_index], addr);
+                    stl_p(buf, val);
+                    l = 4;
+                } else if (l >= 2 && ((addr & 1) == 0)) {
+                    /* 16 bit read access */
+                    val = io_mem_read[io_index][1](io_mem_opaque[io_index], addr);
+                    stw_p(buf, val);
+                    l = 2;
+                } else {
+                    /* 8 bit access */
+                    val = io_mem_read[io_index][0](io_mem_opaque[io_index], addr);
+                    stb_p(buf, val);
+                    l = 1;
+                }
+            } else {
+                /* RAM case */
+                ptr = phys_ram_base + (pd & TARGET_PAGE_MASK) + 
+                    (addr & ~TARGET_PAGE_MASK);
+                memcpy(buf, ptr, l);
+            }
+        }
+        len -= l;
+        buf += l;
+        addr += l;
+    }
+}
+
+/* warning: addr must be aligned */
+uint32_t ldl_phys(target_phys_addr_t addr)
+{
+    int io_index;
+    uint8_t *ptr;
+    uint32_t val;
+    unsigned long pd;
+    PhysPageDesc *p;
+
+    p = phys_page_find(addr >> TARGET_PAGE_BITS);
+    if (!p) {
+        pd = IO_MEM_UNASSIGNED;
+    } else {
+        pd = p->phys_offset;
+    }
+        
+    if ((pd & ~TARGET_PAGE_MASK) > IO_MEM_ROM &&
+        (pd & ~TARGET_PAGE_MASK) != IO_MEM_CODE) {
+        /* I/O case */
+        io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
+        val = io_mem_read[io_index][2](io_mem_opaque[io_index], addr);
+    } else {
+        /* RAM case */
+        ptr = phys_ram_base + (pd & TARGET_PAGE_MASK) + 
+            (addr & ~TARGET_PAGE_MASK);
+        val = ldl_p(ptr);
+    }
+    return val;
+}
+
+/* warning: addr must be aligned. The ram page is not masked as dirty
+   and the code inside is not invalidated. It is useful if the dirty
+   bits are used to track modified PTEs */
+void stl_phys_notdirty(target_phys_addr_t addr, uint32_t val)
+{
+    int io_index;
+    uint8_t *ptr;
+    unsigned long pd;
+    PhysPageDesc *p;
+
+    p = phys_page_find(addr >> TARGET_PAGE_BITS);
+    if (!p) {
+        pd = IO_MEM_UNASSIGNED;
+    } else {
+        pd = p->phys_offset;
+    }
+        
+    if ((pd & ~TARGET_PAGE_MASK) != 0) {
+        io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
+        io_mem_write[io_index][2](io_mem_opaque[io_index], addr, val);
+    } else {
+        ptr = phys_ram_base + (pd & TARGET_PAGE_MASK) + 
+            (addr & ~TARGET_PAGE_MASK);
+        stl_p(ptr, val);
+    }
+}
+
+/* warning: addr must be aligned */
+/* XXX: optimize code invalidation test */
+void stl_phys(target_phys_addr_t addr, uint32_t val)
+{
+    int io_index;
+    uint8_t *ptr;
+    unsigned long pd;
+    PhysPageDesc *p;
+
+    p = phys_page_find(addr >> TARGET_PAGE_BITS);
+    if (!p) {
+        pd = IO_MEM_UNASSIGNED;
+    } else {
+        pd = p->phys_offset;
+    }
+        
+    if ((pd & ~TARGET_PAGE_MASK) != 0) {
+        io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
+        io_mem_write[io_index][2](io_mem_opaque[io_index], addr, val);
+    } else {
+        unsigned long addr1;
+        addr1 = (pd & TARGET_PAGE_MASK) + (addr & ~TARGET_PAGE_MASK);
+        /* RAM case */
+        ptr = phys_ram_base + addr1;
+        stl_p(ptr, val);
+        /* invalidate code */
+        tb_invalidate_phys_page_range(addr1, addr1 + 4, 0);
+        /* set dirty bit */
+        phys_ram_dirty[addr1 >> TARGET_PAGE_BITS] = 0xff;
+    }
+}
+
+#endif
+
+/* virtual memory access for debug */
+int cpu_memory_rw_debug(CPUState *env, target_ulong addr, 
+                        uint8_t *buf, int len, int is_write)
+{
+    int l;
+    target_ulong page, phys_addr;
+
+    while (len > 0) {
+        page = addr & TARGET_PAGE_MASK;
+        phys_addr = cpu_get_phys_page_debug(env, page);
+        /* if no physical page mapped, return an error */
+        if (phys_addr == -1)
+            return -1;
+        l = (page + TARGET_PAGE_SIZE) - addr;
+        if (l > len)
+            l = len;
+        cpu_physical_memory_rw(phys_addr + (addr & ~TARGET_PAGE_MASK), 
+                               buf, l, is_write);
+        len -= l;
+        buf += l;
+        addr += l;
+    }
+    return 0;
+}
+
+void dump_exec_info(FILE *f,
+                    int (*cpu_fprintf)(FILE *f, const char *fmt, ...))
+{
+    int i, target_code_size, max_target_code_size;
+    int direct_jmp_count, direct_jmp2_count, cross_page;
+    TranslationBlock *tb;
+    
+    target_code_size = 0;
+    max_target_code_size = 0;
+    cross_page = 0;
+    direct_jmp_count = 0;
+    direct_jmp2_count = 0;
+    for(i = 0; i < nb_tbs; i++) {
+        tb = &tbs[i];
+        target_code_size += tb->size;
+        if (tb->size > max_target_code_size)
+            max_target_code_size = tb->size;
+        if (tb->page_addr[1] != -1)
+            cross_page++;
+        if (tb->tb_next_offset[0] != 0xffff) {
+            direct_jmp_count++;
+            if (tb->tb_next_offset[1] != 0xffff) {
+                direct_jmp2_count++;
+            }
+        }
+    }
+    /* XXX: avoid using doubles ? */
+    cpu_fprintf(f, "TB count            %d\n", nb_tbs);
+    cpu_fprintf(f, "TB avg target size  %d max=%d bytes\n", 
+                nb_tbs ? target_code_size / nb_tbs : 0,
+                max_target_code_size);
+    cpu_fprintf(f, "TB avg host size    %d bytes (expansion ratio: %0.1f)\n", 
+                nb_tbs ? (code_gen_ptr - code_gen_buffer) / nb_tbs : 0,
+                target_code_size ? (double) (code_gen_ptr - code_gen_buffer) / target_code_size : 0);
+    cpu_fprintf(f, "cross page TB count %d (%d%%)\n", 
+            cross_page, 
+            nb_tbs ? (cross_page * 100) / nb_tbs : 0);
+    cpu_fprintf(f, "direct jump count   %d (%d%%) (2 jumps=%d %d%%)\n",
+                direct_jmp_count, 
+                nb_tbs ? (direct_jmp_count * 100) / nb_tbs : 0,
+                direct_jmp2_count,
+                nb_tbs ? (direct_jmp2_count * 100) / nb_tbs : 0);
+    cpu_fprintf(f, "TB flush count      %d\n", tb_flush_count);
+    cpu_fprintf(f, "TB invalidate count %d\n", tb_phys_invalidate_count);
+    cpu_fprintf(f, "TLB flush count     %d\n", tlb_flush_count);
+}
+
+#if !defined(CONFIG_USER_ONLY) 
+
+#define MMUSUFFIX _cmmu
+#define GETPC() NULL
+#define env cpu_single_env
+#define SOFTMMU_CODE_ACCESS
+
+#define SHIFT 0
+#include "softmmu_template.h"
+
+#define SHIFT 1
+#include "softmmu_template.h"
+
+#define SHIFT 2
+#include "softmmu_template.h"
+
+#define SHIFT 3
+#include "softmmu_template.h"
+
+#undef env
+
 #endif