Add qemu 2.4.0
[kvmfornfv.git] / qemu / qemu-timer.c
diff --git a/qemu/qemu-timer.c b/qemu/qemu-timer.c
new file mode 100644 (file)
index 0000000..2463fe6
--- /dev/null
@@ -0,0 +1,625 @@
+/*
+ * QEMU System Emulator
+ *
+ * Copyright (c) 2003-2008 Fabrice Bellard
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+
+#include "qemu/main-loop.h"
+#include "qemu/timer.h"
+
+#ifdef CONFIG_POSIX
+#include <pthread.h>
+#endif
+
+#ifdef CONFIG_PPOLL
+#include <poll.h>
+#endif
+
+#ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
+#include <sys/prctl.h>
+#endif
+
+/***********************************************************/
+/* timers */
+
+typedef struct QEMUClock {
+    /* We rely on BQL to protect the timerlists */
+    QLIST_HEAD(, QEMUTimerList) timerlists;
+
+    NotifierList reset_notifiers;
+    int64_t last;
+
+    QEMUClockType type;
+    bool enabled;
+} QEMUClock;
+
+QEMUTimerListGroup main_loop_tlg;
+static QEMUClock qemu_clocks[QEMU_CLOCK_MAX];
+
+/* A QEMUTimerList is a list of timers attached to a clock. More
+ * than one QEMUTimerList can be attached to each clock, for instance
+ * used by different AioContexts / threads. Each clock also has
+ * a list of the QEMUTimerLists associated with it, in order that
+ * reenabling the clock can call all the notifiers.
+ */
+
+struct QEMUTimerList {
+    QEMUClock *clock;
+    QemuMutex active_timers_lock;
+    QEMUTimer *active_timers;
+    QLIST_ENTRY(QEMUTimerList) list;
+    QEMUTimerListNotifyCB *notify_cb;
+    void *notify_opaque;
+
+    /* lightweight method to mark the end of timerlist's running */
+    QemuEvent timers_done_ev;
+};
+
+/**
+ * qemu_clock_ptr:
+ * @type: type of clock
+ *
+ * Translate a clock type into a pointer to QEMUClock object.
+ *
+ * Returns: a pointer to the QEMUClock object
+ */
+static inline QEMUClock *qemu_clock_ptr(QEMUClockType type)
+{
+    return &qemu_clocks[type];
+}
+
+static bool timer_expired_ns(QEMUTimer *timer_head, int64_t current_time)
+{
+    return timer_head && (timer_head->expire_time <= current_time);
+}
+
+QEMUTimerList *timerlist_new(QEMUClockType type,
+                             QEMUTimerListNotifyCB *cb,
+                             void *opaque)
+{
+    QEMUTimerList *timer_list;
+    QEMUClock *clock = qemu_clock_ptr(type);
+
+    timer_list = g_malloc0(sizeof(QEMUTimerList));
+    qemu_event_init(&timer_list->timers_done_ev, true);
+    timer_list->clock = clock;
+    timer_list->notify_cb = cb;
+    timer_list->notify_opaque = opaque;
+    qemu_mutex_init(&timer_list->active_timers_lock);
+    QLIST_INSERT_HEAD(&clock->timerlists, timer_list, list);
+    return timer_list;
+}
+
+void timerlist_free(QEMUTimerList *timer_list)
+{
+    assert(!timerlist_has_timers(timer_list));
+    if (timer_list->clock) {
+        QLIST_REMOVE(timer_list, list);
+    }
+    qemu_mutex_destroy(&timer_list->active_timers_lock);
+    g_free(timer_list);
+}
+
+static void qemu_clock_init(QEMUClockType type)
+{
+    QEMUClock *clock = qemu_clock_ptr(type);
+
+    /* Assert that the clock of type TYPE has not been initialized yet. */
+    assert(main_loop_tlg.tl[type] == NULL);
+
+    clock->type = type;
+    clock->enabled = true;
+    clock->last = INT64_MIN;
+    QLIST_INIT(&clock->timerlists);
+    notifier_list_init(&clock->reset_notifiers);
+    main_loop_tlg.tl[type] = timerlist_new(type, NULL, NULL);
+}
+
+bool qemu_clock_use_for_deadline(QEMUClockType type)
+{
+    return !(use_icount && (type == QEMU_CLOCK_VIRTUAL));
+}
+
+void qemu_clock_notify(QEMUClockType type)
+{
+    QEMUTimerList *timer_list;
+    QEMUClock *clock = qemu_clock_ptr(type);
+    QLIST_FOREACH(timer_list, &clock->timerlists, list) {
+        timerlist_notify(timer_list);
+    }
+}
+
+/* Disabling the clock will wait for related timerlists to stop
+ * executing qemu_run_timers.  Thus, this functions should not
+ * be used from the callback of a timer that is based on @clock.
+ * Doing so would cause a deadlock.
+ *
+ * Caller should hold BQL.
+ */
+void qemu_clock_enable(QEMUClockType type, bool enabled)
+{
+    QEMUClock *clock = qemu_clock_ptr(type);
+    QEMUTimerList *tl;
+    bool old = clock->enabled;
+    clock->enabled = enabled;
+    if (enabled && !old) {
+        qemu_clock_notify(type);
+    } else if (!enabled && old) {
+        QLIST_FOREACH(tl, &clock->timerlists, list) {
+            qemu_event_wait(&tl->timers_done_ev);
+        }
+    }
+}
+
+bool timerlist_has_timers(QEMUTimerList *timer_list)
+{
+    return !!timer_list->active_timers;
+}
+
+bool qemu_clock_has_timers(QEMUClockType type)
+{
+    return timerlist_has_timers(
+        main_loop_tlg.tl[type]);
+}
+
+bool timerlist_expired(QEMUTimerList *timer_list)
+{
+    int64_t expire_time;
+
+    qemu_mutex_lock(&timer_list->active_timers_lock);
+    if (!timer_list->active_timers) {
+        qemu_mutex_unlock(&timer_list->active_timers_lock);
+        return false;
+    }
+    expire_time = timer_list->active_timers->expire_time;
+    qemu_mutex_unlock(&timer_list->active_timers_lock);
+
+    return expire_time < qemu_clock_get_ns(timer_list->clock->type);
+}
+
+bool qemu_clock_expired(QEMUClockType type)
+{
+    return timerlist_expired(
+        main_loop_tlg.tl[type]);
+}
+
+/*
+ * As above, but return -1 for no deadline, and do not cap to 2^32
+ * as we know the result is always positive.
+ */
+
+int64_t timerlist_deadline_ns(QEMUTimerList *timer_list)
+{
+    int64_t delta;
+    int64_t expire_time;
+
+    if (!timer_list->clock->enabled) {
+        return -1;
+    }
+
+    /* The active timers list may be modified before the caller uses our return
+     * value but ->notify_cb() is called when the deadline changes.  Therefore
+     * the caller should notice the change and there is no race condition.
+     */
+    qemu_mutex_lock(&timer_list->active_timers_lock);
+    if (!timer_list->active_timers) {
+        qemu_mutex_unlock(&timer_list->active_timers_lock);
+        return -1;
+    }
+    expire_time = timer_list->active_timers->expire_time;
+    qemu_mutex_unlock(&timer_list->active_timers_lock);
+
+    delta = expire_time - qemu_clock_get_ns(timer_list->clock->type);
+
+    if (delta <= 0) {
+        return 0;
+    }
+
+    return delta;
+}
+
+/* Calculate the soonest deadline across all timerlists attached
+ * to the clock. This is used for the icount timeout so we
+ * ignore whether or not the clock should be used in deadline
+ * calculations.
+ */
+int64_t qemu_clock_deadline_ns_all(QEMUClockType type)
+{
+    int64_t deadline = -1;
+    QEMUTimerList *timer_list;
+    QEMUClock *clock = qemu_clock_ptr(type);
+    QLIST_FOREACH(timer_list, &clock->timerlists, list) {
+        deadline = qemu_soonest_timeout(deadline,
+                                        timerlist_deadline_ns(timer_list));
+    }
+    return deadline;
+}
+
+QEMUClockType timerlist_get_clock(QEMUTimerList *timer_list)
+{
+    return timer_list->clock->type;
+}
+
+QEMUTimerList *qemu_clock_get_main_loop_timerlist(QEMUClockType type)
+{
+    return main_loop_tlg.tl[type];
+}
+
+void timerlist_notify(QEMUTimerList *timer_list)
+{
+    if (timer_list->notify_cb) {
+        timer_list->notify_cb(timer_list->notify_opaque);
+    } else {
+        qemu_notify_event();
+    }
+}
+
+/* Transition function to convert a nanosecond timeout to ms
+ * This is used where a system does not support ppoll
+ */
+int qemu_timeout_ns_to_ms(int64_t ns)
+{
+    int64_t ms;
+    if (ns < 0) {
+        return -1;
+    }
+
+    if (!ns) {
+        return 0;
+    }
+
+    /* Always round up, because it's better to wait too long than to wait too
+     * little and effectively busy-wait
+     */
+    ms = (ns + SCALE_MS - 1) / SCALE_MS;
+
+    /* To avoid overflow problems, limit this to 2^31, i.e. approx 25 days */
+    if (ms > (int64_t) INT32_MAX) {
+        ms = INT32_MAX;
+    }
+
+    return (int) ms;
+}
+
+
+/* qemu implementation of g_poll which uses a nanosecond timeout but is
+ * otherwise identical to g_poll
+ */
+int qemu_poll_ns(GPollFD *fds, guint nfds, int64_t timeout)
+{
+#ifdef CONFIG_PPOLL
+    if (timeout < 0) {
+        return ppoll((struct pollfd *)fds, nfds, NULL, NULL);
+    } else {
+        struct timespec ts;
+        int64_t tvsec = timeout / 1000000000LL;
+        /* Avoid possibly overflowing and specifying a negative number of
+         * seconds, which would turn a very long timeout into a busy-wait.
+         */
+        if (tvsec > (int64_t)INT32_MAX) {
+            tvsec = INT32_MAX;
+        }
+        ts.tv_sec = tvsec;
+        ts.tv_nsec = timeout % 1000000000LL;
+        return ppoll((struct pollfd *)fds, nfds, &ts, NULL);
+    }
+#else
+    return g_poll(fds, nfds, qemu_timeout_ns_to_ms(timeout));
+#endif
+}
+
+
+void timer_init_tl(QEMUTimer *ts,
+                   QEMUTimerList *timer_list, int scale,
+                   QEMUTimerCB *cb, void *opaque)
+{
+    ts->timer_list = timer_list;
+    ts->cb = cb;
+    ts->opaque = opaque;
+    ts->scale = scale;
+    ts->expire_time = -1;
+}
+
+void timer_deinit(QEMUTimer *ts)
+{
+    assert(ts->expire_time == -1);
+    ts->timer_list = NULL;
+}
+
+void timer_free(QEMUTimer *ts)
+{
+    g_free(ts);
+}
+
+static void timer_del_locked(QEMUTimerList *timer_list, QEMUTimer *ts)
+{
+    QEMUTimer **pt, *t;
+
+    ts->expire_time = -1;
+    pt = &timer_list->active_timers;
+    for(;;) {
+        t = *pt;
+        if (!t)
+            break;
+        if (t == ts) {
+            *pt = t->next;
+            break;
+        }
+        pt = &t->next;
+    }
+}
+
+static bool timer_mod_ns_locked(QEMUTimerList *timer_list,
+                                QEMUTimer *ts, int64_t expire_time)
+{
+    QEMUTimer **pt, *t;
+
+    /* add the timer in the sorted list */
+    pt = &timer_list->active_timers;
+    for (;;) {
+        t = *pt;
+        if (!timer_expired_ns(t, expire_time)) {
+            break;
+        }
+        pt = &t->next;
+    }
+    ts->expire_time = MAX(expire_time, 0);
+    ts->next = *pt;
+    *pt = ts;
+
+    return pt == &timer_list->active_timers;
+}
+
+static void timerlist_rearm(QEMUTimerList *timer_list)
+{
+    /* Interrupt execution to force deadline recalculation.  */
+    qemu_clock_warp(timer_list->clock->type);
+    timerlist_notify(timer_list);
+}
+
+/* stop a timer, but do not dealloc it */
+void timer_del(QEMUTimer *ts)
+{
+    QEMUTimerList *timer_list = ts->timer_list;
+
+    if (timer_list) {
+        qemu_mutex_lock(&timer_list->active_timers_lock);
+        timer_del_locked(timer_list, ts);
+        qemu_mutex_unlock(&timer_list->active_timers_lock);
+    }
+}
+
+/* modify the current timer so that it will be fired when current_time
+   >= expire_time. The corresponding callback will be called. */
+void timer_mod_ns(QEMUTimer *ts, int64_t expire_time)
+{
+    QEMUTimerList *timer_list = ts->timer_list;
+    bool rearm;
+
+    qemu_mutex_lock(&timer_list->active_timers_lock);
+    timer_del_locked(timer_list, ts);
+    rearm = timer_mod_ns_locked(timer_list, ts, expire_time);
+    qemu_mutex_unlock(&timer_list->active_timers_lock);
+
+    if (rearm) {
+        timerlist_rearm(timer_list);
+    }
+}
+
+/* modify the current timer so that it will be fired when current_time
+   >= expire_time or the current deadline, whichever comes earlier.
+   The corresponding callback will be called. */
+void timer_mod_anticipate_ns(QEMUTimer *ts, int64_t expire_time)
+{
+    QEMUTimerList *timer_list = ts->timer_list;
+    bool rearm;
+
+    qemu_mutex_lock(&timer_list->active_timers_lock);
+    if (ts->expire_time == -1 || ts->expire_time > expire_time) {
+        if (ts->expire_time != -1) {
+            timer_del_locked(timer_list, ts);
+        }
+        rearm = timer_mod_ns_locked(timer_list, ts, expire_time);
+    } else {
+        rearm = false;
+    }
+    qemu_mutex_unlock(&timer_list->active_timers_lock);
+
+    if (rearm) {
+        timerlist_rearm(timer_list);
+    }
+}
+
+void timer_mod(QEMUTimer *ts, int64_t expire_time)
+{
+    timer_mod_ns(ts, expire_time * ts->scale);
+}
+
+void timer_mod_anticipate(QEMUTimer *ts, int64_t expire_time)
+{
+    timer_mod_anticipate_ns(ts, expire_time * ts->scale);
+}
+
+bool timer_pending(QEMUTimer *ts)
+{
+    return ts->expire_time >= 0;
+}
+
+bool timer_expired(QEMUTimer *timer_head, int64_t current_time)
+{
+    return timer_expired_ns(timer_head, current_time * timer_head->scale);
+}
+
+bool timerlist_run_timers(QEMUTimerList *timer_list)
+{
+    QEMUTimer *ts;
+    int64_t current_time;
+    bool progress = false;
+    QEMUTimerCB *cb;
+    void *opaque;
+
+    qemu_event_reset(&timer_list->timers_done_ev);
+    if (!timer_list->clock->enabled) {
+        goto out;
+    }
+
+    current_time = qemu_clock_get_ns(timer_list->clock->type);
+    for(;;) {
+        qemu_mutex_lock(&timer_list->active_timers_lock);
+        ts = timer_list->active_timers;
+        if (!timer_expired_ns(ts, current_time)) {
+            qemu_mutex_unlock(&timer_list->active_timers_lock);
+            break;
+        }
+
+        /* remove timer from the list before calling the callback */
+        timer_list->active_timers = ts->next;
+        ts->next = NULL;
+        ts->expire_time = -1;
+        cb = ts->cb;
+        opaque = ts->opaque;
+        qemu_mutex_unlock(&timer_list->active_timers_lock);
+
+        /* run the callback (the timer list can be modified) */
+        cb(opaque);
+        progress = true;
+    }
+
+out:
+    qemu_event_set(&timer_list->timers_done_ev);
+    return progress;
+}
+
+bool qemu_clock_run_timers(QEMUClockType type)
+{
+    return timerlist_run_timers(main_loop_tlg.tl[type]);
+}
+
+void timerlistgroup_init(QEMUTimerListGroup *tlg,
+                         QEMUTimerListNotifyCB *cb, void *opaque)
+{
+    QEMUClockType type;
+    for (type = 0; type < QEMU_CLOCK_MAX; type++) {
+        tlg->tl[type] = timerlist_new(type, cb, opaque);
+    }
+}
+
+void timerlistgroup_deinit(QEMUTimerListGroup *tlg)
+{
+    QEMUClockType type;
+    for (type = 0; type < QEMU_CLOCK_MAX; type++) {
+        timerlist_free(tlg->tl[type]);
+    }
+}
+
+bool timerlistgroup_run_timers(QEMUTimerListGroup *tlg)
+{
+    QEMUClockType type;
+    bool progress = false;
+    for (type = 0; type < QEMU_CLOCK_MAX; type++) {
+        progress |= timerlist_run_timers(tlg->tl[type]);
+    }
+    return progress;
+}
+
+int64_t timerlistgroup_deadline_ns(QEMUTimerListGroup *tlg)
+{
+    int64_t deadline = -1;
+    QEMUClockType type;
+    for (type = 0; type < QEMU_CLOCK_MAX; type++) {
+        if (qemu_clock_use_for_deadline(tlg->tl[type]->clock->type)) {
+            deadline = qemu_soonest_timeout(deadline,
+                                            timerlist_deadline_ns(
+                                                tlg->tl[type]));
+        }
+    }
+    return deadline;
+}
+
+int64_t qemu_clock_get_ns(QEMUClockType type)
+{
+    int64_t now, last;
+    QEMUClock *clock = qemu_clock_ptr(type);
+
+    switch (type) {
+    case QEMU_CLOCK_REALTIME:
+        return get_clock();
+    default:
+    case QEMU_CLOCK_VIRTUAL:
+        if (use_icount) {
+            return cpu_get_icount();
+        } else {
+            return cpu_get_clock();
+        }
+    case QEMU_CLOCK_HOST:
+        now = get_clock_realtime();
+        last = clock->last;
+        clock->last = now;
+        if (now < last || now > (last + get_max_clock_jump())) {
+            notifier_list_notify(&clock->reset_notifiers, &now);
+        }
+        return now;
+    case QEMU_CLOCK_VIRTUAL_RT:
+        return cpu_get_clock();
+    }
+}
+
+void qemu_clock_register_reset_notifier(QEMUClockType type,
+                                        Notifier *notifier)
+{
+    QEMUClock *clock = qemu_clock_ptr(type);
+    notifier_list_add(&clock->reset_notifiers, notifier);
+}
+
+void qemu_clock_unregister_reset_notifier(QEMUClockType type,
+                                          Notifier *notifier)
+{
+    notifier_remove(notifier);
+}
+
+void init_clocks(void)
+{
+    QEMUClockType type;
+    for (type = 0; type < QEMU_CLOCK_MAX; type++) {
+        qemu_clock_init(type);
+    }
+
+#ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
+    prctl(PR_SET_TIMERSLACK, 1, 0, 0, 0);
+#endif
+}
+
+uint64_t timer_expire_time_ns(QEMUTimer *ts)
+{
+    return timer_pending(ts) ? ts->expire_time : -1;
+}
+
+bool qemu_clock_run_all_timers(void)
+{
+    bool progress = false;
+    QEMUClockType type;
+
+    for (type = 0; type < QEMU_CLOCK_MAX; type++) {
+        progress |= qemu_clock_run_timers(type);
+    }
+
+    return progress;
+}