Merge branch 'master' of /home/nchip/public_html/qemu into garage-push
[qemu] / vl.c
diff --git a/vl.c b/vl.c
index 1342665..fcf8532 100644 (file)
--- a/vl.c
+++ b/vl.c
@@ -33,6 +33,7 @@
 #include "config-host.h"
 
 #ifndef _WIN32
+#include <libgen.h>
 #include <pwd.h>
 #include <sys/times.h>
 #include <sys/wait.h>
@@ -138,6 +139,7 @@ int main(int argc, char **argv)
 #include "hw/isa.h"
 #include "hw/baum.h"
 #include "hw/bt.h"
+#include "hw/watchdog.h"
 #include "hw/smbios.h"
 #include "hw/xen.h"
 #include "bt-host.h"
@@ -155,6 +157,7 @@ int main(int argc, char **argv)
 #include "migration.h"
 #include "kvm.h"
 #include "balloon.h"
+#include "qemu-option.h"
 
 #include "disas.h"
 
@@ -189,7 +192,7 @@ int main(int argc, char **argv)
 /* XXX: use a two level table to limit memory usage */
 #define MAX_IOPORTS 65536
 
-const char *bios_dir = CONFIG_QEMU_SHAREDIR;
+static const char *data_dir;
 const char *bios_name = NULL;
 static void *ioport_opaque[MAX_IOPORTS];
 static IOPortReadFunc *ioport_read_table[3][MAX_IOPORTS];
@@ -198,12 +201,9 @@ static IOPortWriteFunc *ioport_write_table[3][MAX_IOPORTS];
    to store the VM snapshots */
 DriveInfo drives_table[MAX_DRIVES+1];
 int nb_drives;
-static int vga_ram_size;
 enum vga_retrace_method vga_retrace_method = VGA_RETRACE_DUMB;
 static DisplayState *display_state;
-int nographic;
-static int curses;
-static int sdl;
+DisplayType display_type = DT_DEFAULT;
 const char* keyboard_layout = NULL;
 int64_t ticks_per_sec;
 ram_addr_t ram_size;
@@ -216,6 +216,7 @@ static int rtc_date_offset = -1; /* -1 means no change */
 int cirrus_vga_enabled = 1;
 int std_vga_enabled = 0;
 int vmsvga_enabled = 0;
+int xenfb_enabled = 0;
 #ifdef TARGET_SPARC
 int graphic_width = 1024;
 int graphic_height = 768;
@@ -251,6 +252,8 @@ int graphic_rotate = 0;
 #ifndef _WIN32
 int daemonize = 0;
 #endif
+WatchdogTimerModel *watchdog = NULL;
+int watchdog_action = WDT_RESET;
 const char *option_rom[MAX_OPTION_ROMS];
 int nb_option_roms;
 int semihosting_enabled = 0;
@@ -272,7 +275,7 @@ uint64_t node_cpumask[MAX_NODES];
 
 static CPUState *cur_cpu;
 static CPUState *next_cpu;
-static int event_pending = 1;
+static int timer_alarm_pending = 1;
 /* Conversion factor from emulated instructions to virtual clock ticks.  */
 static int icount_time_shift;
 /* Arbitrarily pick 1MIPS as the minimum allowable speed.  */
@@ -912,7 +915,7 @@ struct qemu_alarm_timer {
 
 static inline int alarm_has_dynticks(struct qemu_alarm_timer *t)
 {
-    return t->flags & ALARM_FLAG_DYNTICKS;
+    return t && (t->flags & ALARM_FLAG_DYNTICKS);
 }
 
 static void qemu_rearm_alarm_timer(struct qemu_alarm_timer *t)
@@ -927,17 +930,13 @@ static void qemu_rearm_alarm_timer(struct qemu_alarm_timer *t)
 #define MIN_TIMER_REARM_US 250
 
 static struct qemu_alarm_timer *alarm_timer;
-#ifndef _WIN32
-static int alarm_timer_rfd, alarm_timer_wfd;
-#endif
 
 #ifdef _WIN32
 
 struct qemu_alarm_win32 {
     MMRESULT timerId;
-    HANDLE host_alarm;
     unsigned int period;
-} alarm_win32_data = {0, NULL, -1};
+} alarm_win32_data = {0, -1};
 
 static int win32_start_timer(struct qemu_alarm_timer *t);
 static void win32_stop_timer(struct qemu_alarm_timer *t);
@@ -1192,9 +1191,8 @@ void qemu_mod_timer(QEMUTimer *ts, int64_t expire_time)
             qemu_rearm_alarm_timer(alarm_timer);
         }
         /* Interrupt execution to force deadline recalculation.  */
-        if (use_icount && cpu_single_env) {
-            cpu_exit(cpu_single_env);
-        }
+        if (use_icount)
+            qemu_notify_event();
     }
 }
 
@@ -1305,6 +1303,8 @@ static int timer_load(QEMUFile *f, void *opaque, int version_id)
     return 0;
 }
 
+static void qemu_event_increment(void);
+
 #ifdef _WIN32
 static void CALLBACK host_alarm_handler(UINT uTimerID, UINT uMsg,
                                         DWORD_PTR dwUser, DWORD_PTR dw1,
@@ -1348,27 +1348,22 @@ static void host_alarm_handler(int host_signum)
                                qemu_get_clock(vm_clock))) ||
         qemu_timer_expired(active_timers[QEMU_TIMER_REALTIME],
                            qemu_get_clock(rt_clock))) {
-        CPUState *env = next_cpu;
-
-#ifdef _WIN32
-        struct qemu_alarm_win32 *data = ((struct qemu_alarm_timer*)dwUser)->priv;
-        SetEvent(data->host_alarm);
-#else
-        static const char byte = 0;
-        write(alarm_timer_wfd, &byte, sizeof(byte));
-#endif
-        alarm_timer->flags |= ALARM_FLAG_EXPIRED;
+        qemu_event_increment();
+        if (alarm_timer) alarm_timer->flags |= ALARM_FLAG_EXPIRED;
 
-        if (env) {
+#ifndef CONFIG_IOTHREAD
+        if (next_cpu) {
             /* stop the currently executing cpu because a timer occured */
-            cpu_exit(env);
+            cpu_exit(next_cpu);
 #ifdef CONFIG_KQEMU
-            if (env->kqemu_enabled) {
-                kqemu_cpu_interrupt(env);
+            if (next_cpu->kqemu_enabled) {
+                kqemu_cpu_interrupt(next_cpu);
             }
 #endif
         }
-        event_pending = 1;
+#endif
+        timer_alarm_pending = 1;
+        qemu_notify_event();
     }
 }
 
@@ -1547,6 +1542,11 @@ static int dynticks_start_timer(struct qemu_alarm_timer *t)
 
     sigaction(SIGALRM, &act, NULL);
 
+    /* 
+     * Initialize ev struct to 0 to avoid valgrind complaining
+     * about uninitialized data in timer_create call
+     */
+    memset(&ev, 0, sizeof(ev));
     ev.sigev_value.sival_int = 0;
     ev.sigev_notify = SIGEV_SIGNAL;
     ev.sigev_signo = SIGALRM;
@@ -1644,24 +1644,6 @@ static void unix_stop_timer(struct qemu_alarm_timer *t)
 
 #endif /* !defined(_WIN32) */
 
-static void try_to_rearm_timer(void *opaque)
-{
-    struct qemu_alarm_timer *t = opaque;
-#ifndef _WIN32
-    ssize_t len;
-
-    /* Drain the notify pipe */
-    do {
-        char buffer[512];
-        len = read(alarm_timer_rfd, buffer, sizeof(buffer));
-    } while ((len == -1 && errno == EINTR) || len > 0);
-#endif
-
-    if (t->flags & ALARM_FLAG_EXPIRED) {
-        alarm_timer->flags &= ~ALARM_FLAG_EXPIRED;
-        qemu_rearm_alarm_timer(alarm_timer);
-    }
-}
 
 #ifdef _WIN32
 
@@ -1671,12 +1653,6 @@ static int win32_start_timer(struct qemu_alarm_timer *t)
     struct qemu_alarm_win32 *data = t->priv;
     UINT flags;
 
-    data->host_alarm = CreateEvent(NULL, FALSE, FALSE, NULL);
-    if (!data->host_alarm) {
-        perror("Failed CreateEvent");
-        return -1;
-    }
-
     memset(&tc, 0, sizeof(tc));
     timeGetDevCaps(&tc, sizeof(tc));
 
@@ -1699,14 +1675,10 @@ static int win32_start_timer(struct qemu_alarm_timer *t)
 
     if (!data->timerId) {
         perror("Failed to initialize win32 alarm timer");
-
         timeEndPeriod(data->period);
-        CloseHandle(data->host_alarm);
         return -1;
     }
 
-    qemu_add_wait_object(data->host_alarm, try_to_rearm_timer, t);
-
     return 0;
 }
 
@@ -1716,8 +1688,6 @@ static void win32_stop_timer(struct qemu_alarm_timer *t)
 
     timeKillEvent(data->timerId);
     timeEndPeriod(data->period);
-
-    CloseHandle(data->host_alarm);
 }
 
 static void win32_rearm_timer(struct qemu_alarm_timer *t)
@@ -1744,7 +1714,6 @@ static void win32_rearm_timer(struct qemu_alarm_timer *t)
         perror("Failed to re-arm win32 alarm timer");
 
         timeEndPeriod(data->period);
-        CloseHandle(data->host_alarm);
         exit(1);
     }
 }
@@ -1756,25 +1725,6 @@ static int init_timer_alarm(void)
     struct qemu_alarm_timer *t = NULL;
     int i, err = -1;
 
-#ifndef _WIN32
-    int fds[2];
-
-    err = pipe(fds);
-    if (err == -1)
-        return -errno;
-
-    err = fcntl_setfl(fds[0], O_NONBLOCK);
-    if (err < 0)
-        goto fail;
-
-    err = fcntl_setfl(fds[1], O_NONBLOCK);
-    if (err < 0)
-        goto fail;
-
-    alarm_timer_rfd = fds[0];
-    alarm_timer_wfd = fds[1];
-#endif
-
     for (i = 0; alarm_timers[i].name; i++) {
         t = &alarm_timers[i];
 
@@ -1788,20 +1738,11 @@ static int init_timer_alarm(void)
         goto fail;
     }
 
-#ifndef _WIN32
-    qemu_set_fd_handler2(alarm_timer_rfd, NULL,
-                         try_to_rearm_timer, NULL, t);
-#endif
-
     alarm_timer = t;
 
     return 0;
 
 fail:
-#ifndef _WIN32
-    close(fds[0]);
-    close(fds[1]);
-#endif
     return err;
 }
 
@@ -1870,43 +1811,6 @@ static int socket_init(void)
 }
 #endif
 
-const char *get_opt_name(char *buf, int buf_size, const char *p, char delim)
-{
-    char *q;
-
-    q = buf;
-    while (*p != '\0' && *p != delim) {
-        if (q && (q - buf) < buf_size - 1)
-            *q++ = *p;
-        p++;
-    }
-    if (q)
-        *q = '\0';
-
-    return p;
-}
-
-const char *get_opt_value(char *buf, int buf_size, const char *p)
-{
-    char *q;
-
-    q = buf;
-    while (*p != '\0') {
-        if (*p == ',') {
-            if (*(p + 1) != ',')
-                break;
-            p++;
-        }
-        if (q && (q - buf) < buf_size - 1)
-            *q++ = *p;
-        p++;
-    }
-    if (q)
-        *q = '\0';
-
-    return p;
-}
-
 int get_param_value(char *buf, int buf_size,
                     const char *tag, const char *str)
 {
@@ -1932,29 +1836,45 @@ int get_param_value(char *buf, int buf_size,
     return 0;
 }
 
-int check_params(char *buf, int buf_size,
-                 const char * const *params, const char *str)
+int check_params(const char * const *params, const char *str)
 {
+    int name_buf_size = 1;
     const char *p;
-    int i;
+    char *name_buf;
+    int i, len;
+    int ret = 0;
+
+    for (i = 0; params[i] != NULL; i++) {
+        len = strlen(params[i]) + 1;
+        if (len > name_buf_size) {
+            name_buf_size = len;
+        }
+    }
+    name_buf = qemu_malloc(name_buf_size);
 
     p = str;
     while (*p != '\0') {
-        p = get_opt_name(buf, buf_size, p, '=');
-        if (*p != '=')
-            return -1;
+        p = get_opt_name(name_buf, name_buf_size, p, '=');
+        if (*p != '=') {
+            ret = -1;
+            break;
+        }
         p++;
         for(i = 0; params[i] != NULL; i++)
-            if (!strcmp(params[i], buf))
+            if (!strcmp(params[i], name_buf))
                 break;
-        if (params[i] == NULL)
-            return -1;
+        if (params[i] == NULL) {
+            ret = -1;
+            break;
+        }
         p = get_opt_value(NULL, 0, p);
         if (*p != ',')
             break;
         p++;
     }
-    return 0;
+
+    qemu_free(name_buf);
+    return ret;
 }
 
 /***********************************************************/
@@ -2307,9 +2227,8 @@ int drive_init(struct drive_opt *arg, int snapshot, void *opaque)
                                            "cache", "format", "serial", "werror",
                                            NULL };
 
-    if (check_params(buf, sizeof(buf), params, str) < 0) {
-         fprintf(stderr, "qemu: unknown parameter '%s' in '%s'\n",
-                         buf, str);
+    if (check_params(params, str) < 0) {
+         fprintf(stderr, "qemu: unknown parameter in '%s'\n", str);
          return -1;
     }
 
@@ -2584,7 +2503,7 @@ int drive_init(struct drive_opt *arg, int snapshot, void *opaque)
     drives_table[drives_table_idx].unit = unit_id;
     drives_table[drives_table_idx].onerror = onerror;
     drives_table[drives_table_idx].drive_opt_idx = arg - drives_opt;
-    strncpy(drives_table[nb_drives].serial, serial, sizeof(serial));
+    strncpy(drives_table[drives_table_idx].serial, serial, sizeof(serial));
     nb_drives++;
 
     switch(type) {
@@ -2613,6 +2532,8 @@ int drive_init(struct drive_opt *arg, int snapshot, void *opaque)
     case IF_MTD:
     case IF_VIRTIO:
         break;
+    case IF_COUNT:
+        abort();
     }
     if (!file[0])
         return -2;
@@ -2903,11 +2824,11 @@ void usb_info(Monitor *mon)
 /* PCMCIA/Cardbus */
 
 static struct pcmcia_socket_entry_s {
-    struct pcmcia_socket_s *socket;
+    PCMCIASocket *socket;
     struct pcmcia_socket_entry_s *next;
 } *pcmcia_sockets = 0;
 
-void pcmcia_socket_register(struct pcmcia_socket_s *socket)
+void pcmcia_socket_register(PCMCIASocket *socket)
 {
     struct pcmcia_socket_entry_s *entry;
 
@@ -2917,7 +2838,7 @@ void pcmcia_socket_register(struct pcmcia_socket_s *socket)
     pcmcia_sockets = entry;
 }
 
-void pcmcia_socket_unregister(struct pcmcia_socket_s *socket)
+void pcmcia_socket_unregister(PCMCIASocket *socket)
 {
     struct pcmcia_socket_entry_s *entry, **ptr;
 
@@ -3277,6 +3198,7 @@ static int ram_save_block(QEMUFile *f)
 }
 
 static ram_addr_t ram_save_threshold = 10;
+static uint64_t bytes_transferred = 0;
 
 static ram_addr_t ram_save_remaining(void)
 {
@@ -3291,17 +3213,37 @@ static ram_addr_t ram_save_remaining(void)
     return count;
 }
 
+uint64_t ram_bytes_remaining(void)
+{
+    return ram_save_remaining() * TARGET_PAGE_SIZE;
+}
+
+uint64_t ram_bytes_transferred(void)
+{
+    return bytes_transferred;
+}
+
+uint64_t ram_bytes_total(void)
+{
+    return last_ram_offset;
+}
+
 static int ram_save_live(QEMUFile *f, int stage, void *opaque)
 {
     ram_addr_t addr;
 
+    if (cpu_physical_sync_dirty_bitmap(0, TARGET_PHYS_ADDR_MAX) != 0) {
+        qemu_file_set_error(f);
+        return 0;
+    }
+
     if (stage == 1) {
         /* Make sure all dirty bits are set */
         for (addr = 0; addr < last_ram_offset; addr += TARGET_PAGE_SIZE) {
             if (!cpu_physical_memory_get_dirty(addr, MIGRATION_DIRTY_FLAG))
                 cpu_physical_memory_set_dirty(addr);
         }
-        
+
         /* Enable dirty memory tracking */
         cpu_physical_memory_set_dirty_tracking(1);
 
@@ -3312,6 +3254,7 @@ static int ram_save_live(QEMUFile *f, int stage, void *opaque)
         int ret;
 
         ret = ram_save_block(f);
+        bytes_transferred += ret * TARGET_PAGE_SIZE;
         if (ret == 0) /* no more blocks */
             break;
     }
@@ -3321,7 +3264,9 @@ static int ram_save_live(QEMUFile *f, int stage, void *opaque)
     if (stage == 3) {
 
         /* flush all remaining blocks regardless of rate limiting */
-        while (ram_save_block(f) != 0);
+        while (ram_save_block(f) != 0) {
+            bytes_transferred += TARGET_PAGE_SIZE;
+        }
         cpu_physical_memory_set_dirty_tracking(0);
     }
 
@@ -3405,15 +3350,7 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
 
 void qemu_service_io(void)
 {
-    CPUState *env = cpu_single_env;
-    if (env) {
-        cpu_exit(env);
-#ifdef CONFIG_KQEMU
-        if (env->kqemu_enabled) {
-            kqemu_cpu_interrupt(env);
-        }
-#endif
-    }
+    qemu_notify_event();
 }
 
 /***********************************************************/
@@ -3481,15 +3418,12 @@ void qemu_bh_schedule_idle(QEMUBH *bh)
 
 void qemu_bh_schedule(QEMUBH *bh)
 {
-    CPUState *env = cpu_single_env;
     if (bh->scheduled)
         return;
     bh->scheduled = 1;
     bh->idle = 0;
     /* stop the currently executing CPU to execute the BH ASAP */
-    if (env) {
-        cpu_exit(env);
-    }
+    qemu_notify_event();
 }
 
 void qemu_bh_cancel(QEMUBH *bh)
@@ -3551,6 +3485,18 @@ static QEMUMachine *find_machine(const char *name)
     return NULL;
 }
 
+static QEMUMachine *find_default_machine(void)
+{
+    QEMUMachine *m;
+
+    for(m = first_machine; m != NULL; m = m->next) {
+        if (m->is_default) {
+            return m;
+        }
+    }
+    return NULL;
+}
+
 /***********************************************************/
 /* main execution loop */
 
@@ -3614,6 +3560,9 @@ static void vm_state_notify(int running, int reason)
     }
 }
 
+static void resume_all_vcpus(void);
+static void pause_all_vcpus(void);
+
 void vm_start(void)
 {
     if (!vm_running) {
@@ -3621,15 +3570,7 @@ void vm_start(void)
         vm_running = 1;
         vm_state_notify(1, 0);
         qemu_rearm_alarm_timer(alarm_timer);
-    }
-}
-
-void vm_stop(int reason)
-{
-    if (vm_running) {
-        cpu_disable_ticks();
-        vm_running = 0;
-        vm_state_notify(0, reason);
+        resume_all_vcpus();
     }
 }
 
@@ -3638,6 +3579,7 @@ void vm_stop(int reason)
 typedef struct QEMUResetEntry {
     QEMUResetHandler *func;
     void *opaque;
+    int order;
     struct QEMUResetEntry *next;
 } QEMUResetEntry;
 
@@ -3645,6 +3587,8 @@ static QEMUResetEntry *first_reset_entry;
 static int reset_requested;
 static int shutdown_requested;
 static int powerdown_requested;
+static int debug_requested;
+static int vmstop_requested;
 
 int qemu_shutdown_requested(void)
 {
@@ -3667,16 +3611,42 @@ int qemu_powerdown_requested(void)
     return r;
 }
 
-void qemu_register_reset(QEMUResetHandler *func, void *opaque)
+static int qemu_debug_requested(void)
+{
+    int r = debug_requested;
+    debug_requested = 0;
+    return r;
+}
+
+static int qemu_vmstop_requested(void)
+{
+    int r = vmstop_requested;
+    vmstop_requested = 0;
+    return r;
+}
+
+static void do_vm_stop(int reason)
+{
+    if (vm_running) {
+        cpu_disable_ticks();
+        vm_running = 0;
+        pause_all_vcpus();
+        vm_state_notify(0, reason);
+    }
+}
+
+void qemu_register_reset(QEMUResetHandler *func, int order, void *opaque)
 {
     QEMUResetEntry **pre, *re;
 
     pre = &first_reset_entry;
-    while (*pre != NULL)
+    while (*pre != NULL && (*pre)->order >= order) {
         pre = &(*pre)->next;
+    }
     re = qemu_mallocz(sizeof(QEMUResetEntry));
     re->func = func;
     re->opaque = opaque;
+    re->order = order;
     re->next = NULL;
     *pre = re;
 }
@@ -3689,8 +3659,6 @@ void qemu_system_reset(void)
     for(re = first_reset_entry; re != NULL; re = re->next) {
         re->func(re->opaque);
     }
-    if (kvm_enabled())
-        kvm_sync_vcpus();
 }
 
 void qemu_system_reset_request(void)
@@ -3700,84 +3668,544 @@ void qemu_system_reset_request(void)
     } else {
         reset_requested = 1;
     }
-    if (cpu_single_env)
-        cpu_exit(cpu_single_env);
+    qemu_notify_event();
 }
 
 void qemu_system_shutdown_request(void)
 {
     shutdown_requested = 1;
-    if (cpu_single_env)
-        cpu_exit(cpu_single_env);
+    qemu_notify_event();
 }
 
 void qemu_system_powerdown_request(void)
 {
     powerdown_requested = 1;
-    if (cpu_single_env)
-        cpu_exit(cpu_single_env);
+    qemu_notify_event();
 }
 
-#ifdef _WIN32
-static void host_main_loop_wait(int *timeout)
+#ifdef CONFIG_IOTHREAD
+static void qemu_system_vmstop_request(int reason)
 {
-    int ret, ret2, i;
-    PollingEntry *pe;
+    vmstop_requested = reason;
+    qemu_notify_event();
+}
+#endif
 
+#ifndef _WIN32
+static int io_thread_fd = -1;
 
-    /* XXX: need to suppress polling by better using win32 events */
-    ret = 0;
-    for(pe = first_polling_entry; pe != NULL; pe = pe->next) {
-        ret |= pe->func(pe->opaque);
-    }
-    if (ret == 0) {
-        int err;
-        WaitObjects *w = &wait_objects;
+static void qemu_event_increment(void)
+{
+    static const char byte = 0;
 
-        ret = WaitForMultipleObjects(w->num, w->events, FALSE, *timeout);
-        if (WAIT_OBJECT_0 + 0 <= ret && ret <= WAIT_OBJECT_0 + w->num - 1) {
-            if (w->func[ret - WAIT_OBJECT_0])
-                w->func[ret - WAIT_OBJECT_0](w->opaque[ret - WAIT_OBJECT_0]);
+    if (io_thread_fd == -1)
+        return;
 
-            /* Check for additional signaled events */
-            for(i = (ret - WAIT_OBJECT_0 + 1); i < w->num; i++) {
+    write(io_thread_fd, &byte, sizeof(byte));
+}
 
-                /* Check if event is signaled */
-                ret2 = WaitForSingleObject(w->events[i], 0);
-                if(ret2 == WAIT_OBJECT_0) {
-                    if (w->func[i])
-                        w->func[i](w->opaque[i]);
-                } else if (ret2 == WAIT_TIMEOUT) {
-                } else {
-                    err = GetLastError();
-                    fprintf(stderr, "WaitForSingleObject error %d %d\n", i, err);
-                }
-            }
-        } else if (ret == WAIT_TIMEOUT) {
-        } else {
-            err = GetLastError();
-            fprintf(stderr, "WaitForMultipleObjects error %d %d\n", ret, err);
-        }
-    }
+static void qemu_event_read(void *opaque)
+{
+    int fd = (unsigned long)opaque;
+    ssize_t len;
 
-    *timeout = 0;
+    /* Drain the notify pipe */
+    do {
+        char buffer[512];
+        len = read(fd, buffer, sizeof(buffer));
+    } while ((len == -1 && errno == EINTR) || len > 0);
+}
+
+static int qemu_event_init(void)
+{
+    int err;
+    int fds[2];
+
+    err = pipe(fds);
+    if (err == -1)
+        return -errno;
+
+    err = fcntl_setfl(fds[0], O_NONBLOCK);
+    if (err < 0)
+        goto fail;
+
+    err = fcntl_setfl(fds[1], O_NONBLOCK);
+    if (err < 0)
+        goto fail;
+
+    qemu_set_fd_handler2(fds[0], NULL, qemu_event_read, NULL,
+                         (void *)(unsigned long)fds[0]);
+
+    io_thread_fd = fds[1];
+    return 0;
+
+fail:
+    close(fds[0]);
+    close(fds[1]);
+    return err;
 }
 #else
-static void host_main_loop_wait(int *timeout)
+HANDLE qemu_event_handle;
+
+static void dummy_event_handler(void *opaque)
+{
+}
+
+static int qemu_event_init(void)
+{
+    qemu_event_handle = CreateEvent(NULL, FALSE, FALSE, NULL);
+    if (!qemu_event_handle) {
+        perror("Failed CreateEvent");
+        return -1;
+    }
+    qemu_add_wait_object(qemu_event_handle, dummy_event_handler, NULL);
+    return 0;
+}
+
+static void qemu_event_increment(void)
 {
+    SetEvent(qemu_event_handle);
 }
 #endif
 
-void main_loop_wait(int timeout)
+static int cpu_can_run(CPUState *env)
 {
-    IOHandlerRecord *ioh;
-    fd_set rfds, wfds, xfds;
-    int ret, nfds;
-    struct timeval tv;
+    if (env->stop)
+        return 0;
+    if (env->stopped)
+        return 0;
+    return 1;
+}
 
-    qemu_bh_update_timeout(&timeout);
+#ifndef CONFIG_IOTHREAD
+static int qemu_init_main_loop(void)
+{
+    return qemu_event_init();
+}
 
-    host_main_loop_wait(&timeout);
+void qemu_init_vcpu(void *_env)
+{
+    CPUState *env = _env;
+
+    if (kvm_enabled())
+        kvm_init_vcpu(env);
+    return;
+}
+
+int qemu_cpu_self(void *env)
+{
+    return 1;
+}
+
+static void resume_all_vcpus(void)
+{
+}
+
+static void pause_all_vcpus(void)
+{
+}
+
+void qemu_cpu_kick(void *env)
+{
+    return;
+}
+
+void qemu_notify_event(void)
+{
+    CPUState *env = cpu_single_env;
+
+    if (env) {
+        cpu_exit(env);
+#ifdef USE_KQEMU
+        if (env->kqemu_enabled)
+            kqemu_cpu_interrupt(env);
+#endif
+     }
+}
+
+#define qemu_mutex_lock_iothread() do { } while (0)
+#define qemu_mutex_unlock_iothread() do { } while (0)
+
+void vm_stop(int reason)
+{
+    do_vm_stop(reason);
+}
+
+#else /* CONFIG_IOTHREAD */
+
+#include "qemu-thread.h"
+
+QemuMutex qemu_global_mutex;
+static QemuMutex qemu_fair_mutex;
+
+static QemuThread io_thread;
+
+static QemuThread *tcg_cpu_thread;
+static QemuCond *tcg_halt_cond;
+
+static int qemu_system_ready;
+/* cpu creation */
+static QemuCond qemu_cpu_cond;
+/* system init */
+static QemuCond qemu_system_cond;
+static QemuCond qemu_pause_cond;
+
+static void block_io_signals(void);
+static void unblock_io_signals(void);
+static int tcg_has_work(void);
+
+static int qemu_init_main_loop(void)
+{
+    int ret;
+
+    ret = qemu_event_init();
+    if (ret)
+        return ret;
+
+    qemu_cond_init(&qemu_pause_cond);
+    qemu_mutex_init(&qemu_fair_mutex);
+    qemu_mutex_init(&qemu_global_mutex);
+    qemu_mutex_lock(&qemu_global_mutex);
+
+    unblock_io_signals();
+    qemu_thread_self(&io_thread);
+
+    return 0;
+}
+
+static void qemu_wait_io_event(CPUState *env)
+{
+    while (!tcg_has_work())
+        qemu_cond_timedwait(env->halt_cond, &qemu_global_mutex, 1000);
+
+    qemu_mutex_unlock(&qemu_global_mutex);
+
+    /*
+     * Users of qemu_global_mutex can be starved, having no chance
+     * to acquire it since this path will get to it first.
+     * So use another lock to provide fairness.
+     */
+    qemu_mutex_lock(&qemu_fair_mutex);
+    qemu_mutex_unlock(&qemu_fair_mutex);
+
+    qemu_mutex_lock(&qemu_global_mutex);
+    if (env->stop) {
+        env->stop = 0;
+        env->stopped = 1;
+        qemu_cond_signal(&qemu_pause_cond);
+    }
+}
+
+static int qemu_cpu_exec(CPUState *env);
+
+static void *kvm_cpu_thread_fn(void *arg)
+{
+    CPUState *env = arg;
+
+    block_io_signals();
+    qemu_thread_self(env->thread);
+
+    /* signal CPU creation */
+    qemu_mutex_lock(&qemu_global_mutex);
+    env->created = 1;
+    qemu_cond_signal(&qemu_cpu_cond);
+
+    /* and wait for machine initialization */
+    while (!qemu_system_ready)
+        qemu_cond_timedwait(&qemu_system_cond, &qemu_global_mutex, 100);
+
+    while (1) {
+        if (cpu_can_run(env))
+            qemu_cpu_exec(env);
+        qemu_wait_io_event(env);
+    }
+
+    return NULL;
+}
+
+static void tcg_cpu_exec(void);
+
+static void *tcg_cpu_thread_fn(void *arg)
+{
+    CPUState *env = arg;
+
+    block_io_signals();
+    qemu_thread_self(env->thread);
+
+    /* signal CPU creation */
+    qemu_mutex_lock(&qemu_global_mutex);
+    for (env = first_cpu; env != NULL; env = env->next_cpu)
+        env->created = 1;
+    qemu_cond_signal(&qemu_cpu_cond);
+
+    /* and wait for machine initialization */
+    while (!qemu_system_ready)
+        qemu_cond_timedwait(&qemu_system_cond, &qemu_global_mutex, 100);
+
+    while (1) {
+        tcg_cpu_exec();
+        qemu_wait_io_event(cur_cpu);
+    }
+
+    return NULL;
+}
+
+void qemu_cpu_kick(void *_env)
+{
+    CPUState *env = _env;
+    qemu_cond_broadcast(env->halt_cond);
+    if (kvm_enabled())
+        qemu_thread_signal(env->thread, SIGUSR1);
+}
+
+int qemu_cpu_self(void *env)
+{
+    return (cpu_single_env != NULL);
+}
+
+static void cpu_signal(int sig)
+{
+    if (cpu_single_env)
+        cpu_exit(cpu_single_env);
+}
+
+static void block_io_signals(void)
+{
+    sigset_t set;
+    struct sigaction sigact;
+
+    sigemptyset(&set);
+    sigaddset(&set, SIGUSR2);
+    sigaddset(&set, SIGIO);
+    sigaddset(&set, SIGALRM);
+    pthread_sigmask(SIG_BLOCK, &set, NULL);
+
+    sigemptyset(&set);
+    sigaddset(&set, SIGUSR1);
+    pthread_sigmask(SIG_UNBLOCK, &set, NULL);
+
+    memset(&sigact, 0, sizeof(sigact));
+    sigact.sa_handler = cpu_signal;
+    sigaction(SIGUSR1, &sigact, NULL);
+}
+
+static void unblock_io_signals(void)
+{
+    sigset_t set;
+
+    sigemptyset(&set);
+    sigaddset(&set, SIGUSR2);
+    sigaddset(&set, SIGIO);
+    sigaddset(&set, SIGALRM);
+    pthread_sigmask(SIG_UNBLOCK, &set, NULL);
+
+    sigemptyset(&set);
+    sigaddset(&set, SIGUSR1);
+    pthread_sigmask(SIG_BLOCK, &set, NULL);
+}
+
+static void qemu_signal_lock(unsigned int msecs)
+{
+    qemu_mutex_lock(&qemu_fair_mutex);
+
+    while (qemu_mutex_trylock(&qemu_global_mutex)) {
+        qemu_thread_signal(tcg_cpu_thread, SIGUSR1);
+        if (!qemu_mutex_timedlock(&qemu_global_mutex, msecs))
+            break;
+    }
+    qemu_mutex_unlock(&qemu_fair_mutex);
+}
+
+static void qemu_mutex_lock_iothread(void)
+{
+    if (kvm_enabled()) {
+        qemu_mutex_lock(&qemu_fair_mutex);
+        qemu_mutex_lock(&qemu_global_mutex);
+        qemu_mutex_unlock(&qemu_fair_mutex);
+    } else
+        qemu_signal_lock(100);
+}
+
+static void qemu_mutex_unlock_iothread(void)
+{
+    qemu_mutex_unlock(&qemu_global_mutex);
+}
+
+static int all_vcpus_paused(void)
+{
+    CPUState *penv = first_cpu;
+
+    while (penv) {
+        if (!penv->stopped)
+            return 0;
+        penv = (CPUState *)penv->next_cpu;
+    }
+
+    return 1;
+}
+
+static void pause_all_vcpus(void)
+{
+    CPUState *penv = first_cpu;
+
+    while (penv) {
+        penv->stop = 1;
+        qemu_thread_signal(penv->thread, SIGUSR1);
+        qemu_cpu_kick(penv);
+        penv = (CPUState *)penv->next_cpu;
+    }
+
+    while (!all_vcpus_paused()) {
+        qemu_cond_timedwait(&qemu_pause_cond, &qemu_global_mutex, 100);
+        penv = first_cpu;
+        while (penv) {
+            qemu_thread_signal(penv->thread, SIGUSR1);
+            penv = (CPUState *)penv->next_cpu;
+        }
+    }
+}
+
+static void resume_all_vcpus(void)
+{
+    CPUState *penv = first_cpu;
+
+    while (penv) {
+        penv->stop = 0;
+        penv->stopped = 0;
+        qemu_thread_signal(penv->thread, SIGUSR1);
+        qemu_cpu_kick(penv);
+        penv = (CPUState *)penv->next_cpu;
+    }
+}
+
+static void tcg_init_vcpu(void *_env)
+{
+    CPUState *env = _env;
+    /* share a single thread for all cpus with TCG */
+    if (!tcg_cpu_thread) {
+        env->thread = qemu_mallocz(sizeof(QemuThread));
+        env->halt_cond = qemu_mallocz(sizeof(QemuCond));
+        qemu_cond_init(env->halt_cond);
+        qemu_thread_create(env->thread, tcg_cpu_thread_fn, env);
+        while (env->created == 0)
+            qemu_cond_timedwait(&qemu_cpu_cond, &qemu_global_mutex, 100);
+        tcg_cpu_thread = env->thread;
+        tcg_halt_cond = env->halt_cond;
+    } else {
+        env->thread = tcg_cpu_thread;
+        env->halt_cond = tcg_halt_cond;
+    }
+}
+
+static void kvm_start_vcpu(CPUState *env)
+{
+    kvm_init_vcpu(env);
+    env->thread = qemu_mallocz(sizeof(QemuThread));
+    env->halt_cond = qemu_mallocz(sizeof(QemuCond));
+    qemu_cond_init(env->halt_cond);
+    qemu_thread_create(env->thread, kvm_cpu_thread_fn, env);
+    while (env->created == 0)
+        qemu_cond_timedwait(&qemu_cpu_cond, &qemu_global_mutex, 100);
+}
+
+void qemu_init_vcpu(void *_env)
+{
+    CPUState *env = _env;
+
+    if (kvm_enabled())
+        kvm_start_vcpu(env);
+    else
+        tcg_init_vcpu(env);
+}
+
+void qemu_notify_event(void)
+{
+    qemu_event_increment();
+}
+
+void vm_stop(int reason)
+{
+    QemuThread me;
+    qemu_thread_self(&me);
+
+    if (!qemu_thread_equal(&me, &io_thread)) {
+        qemu_system_vmstop_request(reason);
+        /*
+         * FIXME: should not return to device code in case
+         * vm_stop() has been requested.
+         */
+        if (cpu_single_env) {
+            cpu_exit(cpu_single_env);
+            cpu_single_env->stop = 1;
+        }
+        return;
+    }
+    do_vm_stop(reason);
+}
+
+#endif
+
+
+#ifdef _WIN32
+static void host_main_loop_wait(int *timeout)
+{
+    int ret, ret2, i;
+    PollingEntry *pe;
+
+
+    /* XXX: need to suppress polling by better using win32 events */
+    ret = 0;
+    for(pe = first_polling_entry; pe != NULL; pe = pe->next) {
+        ret |= pe->func(pe->opaque);
+    }
+    if (ret == 0) {
+        int err;
+        WaitObjects *w = &wait_objects;
+
+        ret = WaitForMultipleObjects(w->num, w->events, FALSE, *timeout);
+        if (WAIT_OBJECT_0 + 0 <= ret && ret <= WAIT_OBJECT_0 + w->num - 1) {
+            if (w->func[ret - WAIT_OBJECT_0])
+                w->func[ret - WAIT_OBJECT_0](w->opaque[ret - WAIT_OBJECT_0]);
+
+            /* Check for additional signaled events */
+            for(i = (ret - WAIT_OBJECT_0 + 1); i < w->num; i++) {
+
+                /* Check if event is signaled */
+                ret2 = WaitForSingleObject(w->events[i], 0);
+                if(ret2 == WAIT_OBJECT_0) {
+                    if (w->func[i])
+                        w->func[i](w->opaque[i]);
+                } else if (ret2 == WAIT_TIMEOUT) {
+                } else {
+                    err = GetLastError();
+                    fprintf(stderr, "WaitForSingleObject error %d %d\n", i, err);
+                }
+            }
+        } else if (ret == WAIT_TIMEOUT) {
+        } else {
+            err = GetLastError();
+            fprintf(stderr, "WaitForMultipleObjects error %d %d\n", ret, err);
+        }
+    }
+
+    *timeout = 0;
+}
+#else
+static void host_main_loop_wait(int *timeout)
+{
+}
+#endif
+
+void main_loop_wait(int timeout)
+{
+    IOHandlerRecord *ioh;
+    fd_set rfds, wfds, xfds;
+    int ret, nfds;
+    struct timeval tv;
+
+    qemu_bh_update_timeout(&timeout);
+
+    host_main_loop_wait(&timeout);
 
     /* poll any events */
     /* XXX: separate device handlers from system ones */
@@ -3810,7 +4238,9 @@ void main_loop_wait(int timeout)
         slirp_select_fill(&nfds, &rfds, &wfds, &xfds);
     }
 #endif
+    qemu_mutex_unlock_iothread();
     ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv);
+    qemu_mutex_lock_iothread();
     if (ret > 0) {
         IOHandlerRecord **pioh;
 
@@ -3845,10 +4275,18 @@ void main_loop_wait(int timeout)
     }
 #endif
 
+    /* rearm timer, if not periodic */
+    if (alarm_timer->flags & ALARM_FLAG_EXPIRED) {
+        alarm_timer->flags &= ~ALARM_FLAG_EXPIRED;
+        qemu_rearm_alarm_timer(alarm_timer);
+    }
+
     /* vm time timers */
-    if (vm_running && likely(!(cur_cpu->singlestep_enabled & SSTEP_NOTIMER)))
-        qemu_run_timers(&active_timers[QEMU_TIMER_VIRTUAL],
-                        qemu_get_clock(vm_clock));
+    if (vm_running) {
+        if (!cur_cpu || likely(!(cur_cpu->singlestep_enabled & SSTEP_NOTIMER)))
+            qemu_run_timers(&active_timers[QEMU_TIMER_VIRTUAL],
+                qemu_get_clock(vm_clock));
+    }
 
     /* real time timers */
     qemu_run_timers(&active_timers[QEMU_TIMER_REALTIME],
@@ -3860,153 +4298,206 @@ void main_loop_wait(int timeout)
 
 }
 
-static int main_loop(void)
+static int qemu_cpu_exec(CPUState *env)
 {
-    int ret, timeout;
+    int ret;
 #ifdef CONFIG_PROFILER
     int64_t ti;
 #endif
-    CPUState *env;
-
-    cur_cpu = first_cpu;
-    next_cpu = cur_cpu->next_cpu ?: first_cpu;
-    for(;;) {
-        if (vm_running) {
 
-            for(;;) {
-                /* get next cpu */
-                env = next_cpu;
 #ifdef CONFIG_PROFILER
-                ti = profile_getclock();
+    ti = profile_getclock();
 #endif
-                if (use_icount) {
-                    int64_t count;
-                    int decr;
-                    qemu_icount -= (env->icount_decr.u16.low + env->icount_extra);
-                    env->icount_decr.u16.low = 0;
-                    env->icount_extra = 0;
-                    count = qemu_next_deadline();
-                    count = (count + (1 << icount_time_shift) - 1)
-                            >> icount_time_shift;
-                    qemu_icount += count;
-                    decr = (count > 0xffff) ? 0xffff : count;
-                    count -= decr;
-                    env->icount_decr.u16.low = decr;
-                    env->icount_extra = count;
-                }
-                ret = cpu_exec(env);
+    if (use_icount) {
+        int64_t count;
+        int decr;
+        qemu_icount -= (env->icount_decr.u16.low + env->icount_extra);
+        env->icount_decr.u16.low = 0;
+        env->icount_extra = 0;
+        count = qemu_next_deadline();
+        count = (count + (1 << icount_time_shift) - 1)
+                >> icount_time_shift;
+        qemu_icount += count;
+        decr = (count > 0xffff) ? 0xffff : count;
+        count -= decr;
+        env->icount_decr.u16.low = decr;
+        env->icount_extra = count;
+    }
+    ret = cpu_exec(env);
 #ifdef CONFIG_PROFILER
-                qemu_time += profile_getclock() - ti;
+    qemu_time += profile_getclock() - ti;
 #endif
-                if (use_icount) {
-                    /* Fold pending instructions back into the
-                       instruction counter, and clear the interrupt flag.  */
-                    qemu_icount -= (env->icount_decr.u16.low
-                                    + env->icount_extra);
-                    env->icount_decr.u32 = 0;
-                    env->icount_extra = 0;
-                }
-                next_cpu = env->next_cpu ?: first_cpu;
-                if (event_pending && likely(ret != EXCP_DEBUG)) {
-                    ret = EXCP_INTERRUPT;
-                    event_pending = 0;
-                    break;
-                }
-                if (ret == EXCP_HLT) {
-                    /* Give the next CPU a chance to run.  */
-                    cur_cpu = env;
-                    continue;
-                }
-                if (ret != EXCP_HALTED)
-                    break;
-                /* all CPUs are halted ? */
-                if (env == cur_cpu)
-                    break;
-            }
-            cur_cpu = env;
+    if (use_icount) {
+        /* Fold pending instructions back into the
+           instruction counter, and clear the interrupt flag.  */
+        qemu_icount -= (env->icount_decr.u16.low
+                        + env->icount_extra);
+        env->icount_decr.u32 = 0;
+        env->icount_extra = 0;
+    }
+    return ret;
+}
 
-            if (shutdown_requested) {
-                ret = EXCP_INTERRUPT;
-                if (no_shutdown) {
-                    vm_stop(0);
-                    no_shutdown = 0;
-                }
-                else
-                    break;
-            }
-            if (reset_requested) {
-                reset_requested = 0;
-                qemu_system_reset();
-                ret = EXCP_INTERRUPT;
-            }
-            if (powerdown_requested) {
-                powerdown_requested = 0;
-               qemu_system_powerdown();
-                ret = EXCP_INTERRUPT;
-            }
-            if (unlikely(ret == EXCP_DEBUG)) {
-                gdb_set_stop_cpu(cur_cpu);
-                vm_stop(EXCP_DEBUG);
-            }
-            /* If all cpus are halted then wait until the next IRQ */
-            /* XXX: use timeout computed from timers */
-            if (ret == EXCP_HALTED) {
-                if (use_icount) {
-                    int64_t add;
-                    int64_t delta;
-                    /* Advance virtual time to the next event.  */
-                    if (use_icount == 1) {
-                        /* When not using an adaptive execution frequency
-                           we tend to get badly out of sync with real time,
-                           so just delay for a reasonable amount of time.  */
-                        delta = 0;
-                    } else {
-                        delta = cpu_get_icount() - cpu_get_clock();
-                    }
-                    if (delta > 0) {
-                        /* If virtual time is ahead of real time then just
-                           wait for IO.  */
-                        timeout = (delta / 1000000) + 1;
-                    } else {
-                        /* Wait for either IO to occur or the next
-                           timer event.  */
-                        add = qemu_next_deadline();
-                        /* We advance the timer before checking for IO.
-                           Limit the amount we advance so that early IO
-                           activity won't get the guest too far ahead.  */
-                        if (add > 10000000)
-                            add = 10000000;
-                        delta += add;
-                        add = (add + (1 << icount_time_shift) - 1)
-                              >> icount_time_shift;
-                        qemu_icount += add;
-                        timeout = delta / 1000000;
-                        if (timeout < 0)
-                            timeout = 0;
-                    }
-                } else {
-                    timeout = 5000;
-                }
-            } else {
-                timeout = 0;
-            }
+static void tcg_cpu_exec(void)
+{
+    int ret = 0;
+
+    if (next_cpu == NULL)
+        next_cpu = first_cpu;
+    for (; next_cpu != NULL; next_cpu = next_cpu->next_cpu) {
+        CPUState *env = cur_cpu = next_cpu;
+
+        if (!vm_running)
+            break;
+        if (timer_alarm_pending) {
+            timer_alarm_pending = 0;
+            break;
+        }
+        if (cpu_can_run(env))
+            ret = qemu_cpu_exec(env);
+        if (ret == EXCP_DEBUG) {
+            gdb_set_stop_cpu(env);
+            debug_requested = 1;
+            break;
+        }
+    }
+}
+
+static int cpu_has_work(CPUState *env)
+{
+    if (env->stop)
+        return 1;
+    if (env->stopped)
+        return 0;
+    if (!env->halted)
+        return 1;
+    if (qemu_cpu_has_work(env))
+        return 1;
+    return 0;
+}
+
+static int tcg_has_work(void)
+{
+    CPUState *env;
+
+    for (env = first_cpu; env != NULL; env = env->next_cpu)
+        if (cpu_has_work(env))
+            return 1;
+    return 0;
+}
+
+static int qemu_calculate_timeout(void)
+{
+    int timeout;
+
+    if (!vm_running)
+        timeout = 5000;
+    else if (tcg_has_work())
+        timeout = 0;
+    else if (!use_icount)
+        timeout = 5000;
+    else {
+     /* XXX: use timeout computed from timers */
+        int64_t add;
+        int64_t delta;
+        /* Advance virtual time to the next event.  */
+        if (use_icount == 1) {
+            /* When not using an adaptive execution frequency
+               we tend to get badly out of sync with real time,
+               so just delay for a reasonable amount of time.  */
+            delta = 0;
         } else {
-            if (shutdown_requested) {
-                ret = EXCP_INTERRUPT;
-                break;
-            }
-            timeout = 5000;
+            delta = cpu_get_icount() - cpu_get_clock();
         }
+        if (delta > 0) {
+            /* If virtual time is ahead of real time then just
+               wait for IO.  */
+            timeout = (delta / 1000000) + 1;
+        } else {
+            /* Wait for either IO to occur or the next
+               timer event.  */
+            add = qemu_next_deadline();
+            /* We advance the timer before checking for IO.
+               Limit the amount we advance so that early IO
+               activity won't get the guest too far ahead.  */
+            if (add > 10000000)
+                add = 10000000;
+            delta += add;
+            add = (add + (1 << icount_time_shift) - 1)
+                  >> icount_time_shift;
+            qemu_icount += add;
+            timeout = delta / 1000000;
+            if (timeout < 0)
+                timeout = 0;
+        }
+    }
+
+    return timeout;
+}
+
+static int vm_can_run(void)
+{
+    if (powerdown_requested)
+        return 0;
+    if (reset_requested)
+        return 0;
+    if (shutdown_requested)
+        return 0;
+    if (debug_requested)
+        return 0;
+    return 1;
+}
+
+static void main_loop(void)
+{
+    int r;
+
+#ifdef CONFIG_IOTHREAD
+    qemu_system_ready = 1;
+    qemu_cond_broadcast(&qemu_system_cond);
+#endif
+
+    for (;;) {
+        do {
 #ifdef CONFIG_PROFILER
-        ti = profile_getclock();
+            int64_t ti;
+#endif
+#ifndef CONFIG_IOTHREAD
+            tcg_cpu_exec();
+#endif
+#ifdef CONFIG_PROFILER
+            ti = profile_getclock();
+#endif
+#ifdef CONFIG_IOTHREAD
+            main_loop_wait(1000);
+#else
+            main_loop_wait(qemu_calculate_timeout());
 #endif
-        main_loop_wait(timeout);
 #ifdef CONFIG_PROFILER
-        dev_time += profile_getclock() - ti;
+            dev_time += profile_getclock() - ti;
 #endif
+        } while (vm_can_run());
+
+        if (qemu_debug_requested())
+            vm_stop(EXCP_DEBUG);
+        if (qemu_shutdown_requested()) {
+            if (no_shutdown) {
+                vm_stop(0);
+                no_shutdown = 0;
+            } else
+                break;
+        }
+        if (qemu_reset_requested()) {
+            pause_all_vcpus();
+            qemu_system_reset();
+            resume_all_vcpus();
+        }
+        if (qemu_powerdown_requested())
+            qemu_system_powerdown();
+        if ((r = qemu_vmstop_requested()))
+            vm_stop(r);
     }
-    cpu_disable_ticks();
-    return ret;
+    pause_all_vcpus();
 }
 
 static void version(void)
@@ -4222,23 +4713,19 @@ static void select_vgahw (const char *p)
 {
     const char *opts;
 
+    cirrus_vga_enabled = 0;
+    std_vga_enabled = 0;
+    vmsvga_enabled = 0;
+    xenfb_enabled = 0;
     if (strstart(p, "std", &opts)) {
         std_vga_enabled = 1;
-        cirrus_vga_enabled = 0;
-        vmsvga_enabled = 0;
     } else if (strstart(p, "cirrus", &opts)) {
         cirrus_vga_enabled = 1;
-        std_vga_enabled = 0;
-        vmsvga_enabled = 0;
     } else if (strstart(p, "vmware", &opts)) {
-        cirrus_vga_enabled = 0;
-        std_vga_enabled = 0;
         vmsvga_enabled = 1;
-    } else if (strstart(p, "none", &opts)) {
-        cirrus_vga_enabled = 0;
-        std_vga_enabled = 0;
-        vmsvga_enabled = 0;
-    } else {
+    } else if (strstart(p, "xenfb", &opts)) {
+        xenfb_enabled = 1;
+    } else if (!strstart(p, "none", &opts)) {
     invalid_vga:
         fprintf(stderr, "Unknown vga type: %s\n", p);
         exit(1);
@@ -4309,11 +4796,131 @@ static void termsig_setup(void)
 
 #endif
 
+#ifdef _WIN32
+/* Look for support files in the same directory as the executable.  */
+static char *find_datadir(const char *argv0)
+{
+    char *p;
+    char buf[MAX_PATH];
+    DWORD len;
+
+    len = GetModuleFileName(NULL, buf, sizeof(buf) - 1);
+    if (len == 0) {
+        return len;
+    }
+
+    buf[len] = 0;
+    p = buf + len - 1;
+    while (p != buf && *p != '\\')
+        p--;
+    *p = 0;
+    if (access(buf, R_OK) == 0) {
+        return qemu_strdup(buf);
+    }
+    return NULL;
+}
+#else /* !_WIN32 */
+
+/* Find a likely location for support files using the location of the binary.
+   For installed binaries this will be "$bindir/../share/qemu".  When
+   running from the build tree this will be "$bindir/../pc-bios".  */
+#define SHARE_SUFFIX "/share/qemu"
+#define BUILD_SUFFIX "/pc-bios"
+static char *find_datadir(const char *argv0)
+{
+    char *dir;
+    char *p = NULL;
+    char *res;
+#ifdef PATH_MAX
+    char buf[PATH_MAX];
+#endif
+
+#if defined(__linux__)
+    {
+        int len;
+        len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
+        if (len > 0) {
+            buf[len] = 0;
+            p = buf;
+        }
+    }
+#elif defined(__FreeBSD__)
+    {
+        int len;
+        len = readlink("/proc/curproc/file", buf, sizeof(buf) - 1);
+        if (len > 0) {
+            buf[len] = 0;
+            p = buf;
+        }
+    }
+#endif
+    /* If we don't have any way of figuring out the actual executable
+       location then try argv[0].  */
+    if (!p) {
+#ifdef PATH_MAX
+        p = buf;
+#endif
+        p = realpath(argv0, p);
+        if (!p) {
+            return NULL;
+        }
+    }
+    dir = dirname(p);
+    dir = dirname(dir);
+
+    res = qemu_mallocz(strlen(dir) + 
+        MAX(strlen(SHARE_SUFFIX), strlen(BUILD_SUFFIX)) + 1);
+    sprintf(res, "%s%s", dir, SHARE_SUFFIX);
+    if (access(res, R_OK)) {
+        sprintf(res, "%s%s", dir, BUILD_SUFFIX);
+        if (access(res, R_OK)) {
+            qemu_free(res);
+            res = NULL;
+        }
+    }
+#ifndef PATH_MAX
+    free(p);
+#endif
+    return res;
+}
+#undef SHARE_SUFFIX
+#undef BUILD_SUFFIX
+#endif
+
+char *qemu_find_file(int type, const char *name)
+{
+    int len;
+    const char *subdir;
+    char *buf;
+
+    /* If name contains path separators then try it as a straight path.  */
+    if ((strchr(name, '/') || strchr(name, '\\'))
+        && access(name, R_OK) == 0) {
+        return strdup(name);
+    }
+    switch (type) {
+    case QEMU_FILE_TYPE_BIOS:
+        subdir = "";
+        break;
+    case QEMU_FILE_TYPE_KEYMAP:
+        subdir = "keymaps/";
+        break;
+    default:
+        abort();
+    }
+    len = strlen(data_dir) + strlen(name) + strlen(subdir) + 2;
+    buf = qemu_mallocz(len);
+    sprintf(buf, "%s/%s%s", data_dir, subdir, name);
+    if (access(buf, R_OK)) {
+        qemu_free(buf);
+        return NULL;
+    }
+    return buf;
+}
+
 int main(int argc, char **argv, char **envp)
 {
-#ifdef CONFIG_GDBSTUB
     const char *gdbstub_dev = NULL;
-#endif
     uint32_t boot_devices_bitmap = 0;
     int i;
     int snapshot, linux_boot, net_boot;
@@ -4356,6 +4963,7 @@ int main(int argc, char **argv, char **envp)
     const char *run_as = NULL;
 #endif
     CPUState *env;
+    int show_vnc_port = 0;
 
     qemu_cache_utils_init(envp);
 
@@ -4390,15 +4998,12 @@ int main(int argc, char **argv, char **envp)
     }
 #endif
 
-    register_machines();
-    machine = first_machine;
+    module_call_init(MODULE_INIT_MACHINE);
+    machine = find_default_machine();
     cpu_model = NULL;
     initrd_filename = NULL;
     ram_size = 0;
-    vga_ram_size = VGA_RAM_SIZE;
     snapshot = 0;
-    nographic = 0;
-    curses = 0;
     kernel_filename = NULL;
     kernel_cmdline = "";
     cyls = heads = secs = 0;
@@ -4438,6 +5043,8 @@ int main(int argc, char **argv, char **envp)
     tb_size = 0;
     autostart= 1;
 
+    register_watchdogs();
+
     optind = 1;
     for(;;) {
         if (optind >= argc)
@@ -4483,7 +5090,7 @@ int main(int argc, char **argv, char **envp)
                     for(m = first_machine; m != NULL; m = m->next) {
                         printf("%-10s %s%s\n",
                                m->name, m->desc,
-                               m == first_machine ? " (default)" : "");
+                               m->is_default ? " (default)" : "");
                     }
                     exit(*optarg != '?');
                 }
@@ -4588,11 +5195,11 @@ int main(int argc, char **argv, char **envp)
                 numa_add(optarg);
                 break;
             case QEMU_OPTION_nographic:
-                nographic = 1;
+                display_type = DT_NOGRAPHIC;
                 break;
 #ifdef CONFIG_CURSES
             case QEMU_OPTION_curses:
-                curses = 1;
+                display_type = DT_CURSES;
                 break;
 #endif
             case QEMU_OPTION_portrait:
@@ -4668,7 +5275,7 @@ int main(int argc, char **argv, char **envp)
                 break;
 #endif
             case QEMU_OPTION_redir:
-                net_slirp_redir(NULL, optarg);
+                net_slirp_redir(NULL, optarg, NULL);
                 break;
 #endif
             case QEMU_OPTION_bt:
@@ -4743,16 +5350,14 @@ int main(int argc, char **argv, char **envp)
                     cpu_set_log(mask);
                 }
                 break;
-#ifdef CONFIG_GDBSTUB
             case QEMU_OPTION_s:
                 gdbstub_dev = "tcp::" DEFAULT_GDBSTUB_PORT;
                 break;
             case QEMU_OPTION_gdb:
                 gdbstub_dev = optarg;
                 break;
-#endif
             case QEMU_OPTION_L:
-                bios_dir = optarg;
+                data_dir = optarg;
                 break;
             case QEMU_OPTION_bios:
                 bios_name = optarg;
@@ -4829,6 +5434,17 @@ int main(int argc, char **argv, char **envp)
                 serial_devices[serial_device_index] = optarg;
                 serial_device_index++;
                 break;
+            case QEMU_OPTION_watchdog:
+                i = select_watchdog(optarg);
+                if (i > 0)
+                    exit (i == 1 ? 1 : 0);
+                break;
+            case QEMU_OPTION_watchdog_action:
+                if (select_watchdog_action(optarg) == -1) {
+                    fprintf(stderr, "Unknown -watchdog-action parameter\n");
+                    exit(1);
+                }
+                break;
             case QEMU_OPTION_virtiocon:
                 if (virtio_console_index >= MAX_VIRTIO_CONSOLES) {
                     fprintf(stderr, "qemu: too many virtio consoles\n");
@@ -4862,7 +5478,7 @@ int main(int argc, char **argv, char **envp)
                 no_quit = 1;
                 break;
             case QEMU_OPTION_sdl:
-                sdl = 1;
+                display_type = DT_SDL;
                 break;
 #endif
             case QEMU_OPTION_pidfile:
@@ -4924,6 +5540,7 @@ int main(int argc, char **argv, char **envp)
                 }
                 break;
            case QEMU_OPTION_vnc:
+                display_type = DT_VNC;
                vnc_display = optarg;
                break;
 #ifdef TARGET_I386
@@ -5066,6 +5683,16 @@ int main(int argc, char **argv, char **envp)
         }
     }
 
+    /* If no data_dir is specified then try to find it relative to the
+       executable path.  */
+    if (!data_dir) {
+        data_dir = find_datadir(argv[0]);
+    }
+    /* If all else fails use the install patch specified when building.  */
+    if (!data_dir) {
+        data_dir = CONFIG_QEMU_SHAREDIR;
+    }
+
 #if defined(CONFIG_KVM) && defined(CONFIG_KQEMU)
     if (kvm_allowed && kqemu_allowed) {
         fprintf(stderr,
@@ -5082,7 +5709,7 @@ int main(int argc, char **argv, char **envp)
         exit(1);
     }
 
-    if (nographic) {
+    if (display_type == DT_NOGRAPHIC) {
        if (serial_device_index == 0)
            serial_devices[0] = "stdio";
        if (parallel_device_index == 0)
@@ -5149,6 +5776,10 @@ int main(int argc, char **argv, char **envp)
     if (smp_cpus > 1)
         kqemu_allowed = 0;
 #endif
+    if (qemu_init_main_loop()) {
+        fprintf(stderr, "qemu_init_main_loop failed\n");
+        exit(1);
+    }
     linux_boot = (kernel_filename != NULL);
     net_boot = (boot_devices_bitmap >> ('n' - 'a')) & 0xF;
 
@@ -5207,19 +5838,24 @@ int main(int argc, char **argv, char **envp)
        for (i = 0; i < nb_nics && i < 4; i++) {
            const char *model = nd_table[i].model;
            char buf[1024];
+            char *filename;
             if (net_boot & (1 << i)) {
                 if (model == NULL)
                     model = "ne2k_pci";
-                snprintf(buf, sizeof(buf), "%s/pxe-%s.bin", bios_dir, model);
-                if (get_image_size(buf) > 0) {
+                snprintf(buf, sizeof(buf), "pxe-%s.bin", model);
+                filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, buf);
+                if (filename && get_image_size(filename) > 0) {
                     if (nb_option_roms >= MAX_OPTION_ROMS) {
                         fprintf(stderr, "Too many option ROMs\n");
                         exit(1);
                     }
-                    option_rom[nb_option_roms] = strdup(buf);
+                    option_rom[nb_option_roms] = qemu_strdup(buf);
                     nb_option_roms++;
                     netroms++;
                 }
+                if (filename) {
+                    qemu_free(filename);
+                }
             }
        }
        if (netroms == 0) {
@@ -5242,7 +5878,7 @@ int main(int argc, char **argv, char **envp)
     /* FIXME: This is a nasty hack because kqemu can't cope with dynamic
        guest ram allocation.  It needs to go away.  */
     if (kqemu_allowed) {
-        kqemu_phys_ram_size = ram_size + VGA_RAM_SIZE + 4 * 1024 * 1024;
+        kqemu_phys_ram_size = ram_size + 8 * 1024 * 1024 + 4 * 1024 * 1024;
         kqemu_phys_ram_base = qemu_vmalloc(kqemu_phys_ram_size);
         if (!kqemu_phys_ram_base) {
             fprintf(stderr, "Could not allocate physical memory\n");
@@ -5255,7 +5891,6 @@ int main(int argc, char **argv, char **envp)
     cpu_exec_init_all(tb_size * 1024 * 1024);
 
     bdrv_init();
-    dma_helper_init();
 
     /* we always create the cdrom drive, even if no disk is there */
 
@@ -5403,7 +6038,9 @@ int main(int argc, char **argv, char **envp)
         }
     }
 
-    machine->init(ram_size, vga_ram_size, boot_devices,
+    module_call_init(MODULE_INIT_DEVICE);
+
+    machine->init(ram_size, boot_devices,
                   kernel_filename, kernel_cmdline, initrd_filename, cpu_model);
 
 
@@ -5442,33 +6079,46 @@ int main(int argc, char **argv, char **envp)
         dumb_display_init();
     /* just use the first displaystate for the moment */
     ds = display_state;
-    /* terminal init */
-    if (nographic) {
-        if (curses) {
-            fprintf(stderr, "fatal: -nographic can't be used with -curses\n");
-            exit(1);
-        }
-    } else { 
+
+    if (display_type == DT_DEFAULT) {
+#if defined(CONFIG_SDL) || defined(CONFIG_COCOA)
+        display_type = DT_SDL;
+#else
+        display_type = DT_VNC;
+        vnc_display = "localhost:0,to=99";
+        show_vnc_port = 1;
+#endif
+    }
+        
+
+    switch (display_type) {
+    case DT_NOGRAPHIC:
+        break;
 #if defined(CONFIG_CURSES)
-            if (curses) {
-                /* At the moment curses cannot be used with other displays */
-                curses_display_init(ds, full_screen);
-            } else
+    case DT_CURSES:
+        curses_display_init(ds, full_screen);
+        break;
 #endif
-            {
-                if (vnc_display != NULL) {
-                    vnc_display_init(ds);
-                    if (vnc_display_open(ds, vnc_display) < 0)
-                        exit(1);
-                }
 #if defined(CONFIG_SDL)
-                if (sdl || !vnc_display)
-                    sdl_display_init(ds, full_screen, no_frame);
+    case DT_SDL:
+        sdl_display_init(ds, full_screen, no_frame);
+        break;
 #elif defined(CONFIG_COCOA)
-                if (sdl || !vnc_display)
-                    cocoa_display_init(ds, full_screen);
+    case DT_SDL:
+        cocoa_display_init(ds, full_screen);
+        break;
 #endif
-            }
+    case DT_VNC:
+        vnc_display_init(ds);
+        if (vnc_display_open(ds, vnc_display) < 0)
+            exit(1);
+
+        if (show_vnc_port) {
+            printf("VNC server running on `%s'\n", vnc_display_local_addr(ds));
+        }
+        break;
+    default:
+        break;
     }
     dpy_resize(ds);
 
@@ -5481,7 +6131,7 @@ int main(int argc, char **argv, char **envp)
         dcl = dcl->next;
     }
 
-    if (nographic || (vnc_display && !sdl)) {
+    if (display_type == DT_NOGRAPHIC || display_type == DT_VNC) {
         nographic_timer = qemu_new_timer(rt_clock, nographic_update, NULL);
         qemu_mod_timer(nographic_timer, qemu_get_clock(rt_clock));
     }
@@ -5522,13 +6172,11 @@ int main(int argc, char **argv, char **envp)
         }
     }
 
-#ifdef CONFIG_GDBSTUB
     if (gdbstub_dev && gdbserver_start(gdbstub_dev) < 0) {
         fprintf(stderr, "qemu: could not open gdbserver on device '%s'\n",
                 gdbstub_dev);
         exit(1);
     }
-#endif
 
     if (loadvm)
         do_loadvm(cur_mon, loadvm);