[Qemu-devel] [PATCH v5 0/3] High downtime with 95+ throttle pct

[Qemu-devel] [PATCH v5 0/3] High downtime with 95+ throttle pct

[Yury Kotov]

Hi,

V5:
* Updated sleep loop in throttle_thread at the suggestion of Paolo Bonzini
* Fixed hanging of test

V4:
* The test was simplified to prevent false fails.

V3:
* Rebase fixes (migrate_set_parameter -> migrate_set_parameter_int)

V2:
* Added a test
* Fixed qemu_cond_timedwait for qsp

I wrote a test for migration auto converge and found out a strange thing:
1. Enable auto converge
2. Set max-bandwidth 1Gb/s
3. Set downtime-limit 1ms
4. Run standard test (just writes a byte per page)
5. Wait for converge
6. It's converged with 99% throttle percentage
7. The result downtime was about 300-600ms   <<<<

It's much higher than expected 1ms. I figured out that cpu_throttle_thread()
function sleeps for 100ms+ for high throttle percentage (>=95%) in VCPU thread.
And it sleeps even after a cpu kick.

Fixed it by using timedwait for ms part of sleep.
E.g timedwait(halt_cond, 1ms) + usleep(500).

Regards,
Yury

Yury Kotov (3):
  qemu-thread: Add qemu_cond_timedwait
  cpus: Fix throttling during vm_stop
  tests/migration: Add a test for auto converge

 cpus.c                   |  25 +++++---
 include/qemu/thread.h    |  18 ++++++
 tests/migration-test.c   | 120 +++++++++++++++++++++++++++++++++++----
 util/qemu-thread-posix.c |  40 +++++++++----
 util/qemu-thread-win32.c |  16 ++++++
 util/qsp.c               |  18 ++++++
 6 files changed, 206 insertions(+), 31 deletions(-)

-- 
2.22.0

[Qemu-devel] [PATCH v5 1/3] qemu-thread: Add qemu_cond_timedwait

[Yury Kotov]

Signed-off-by: Yury Kotov <yury-kotov@yandex-team.ru>
---
 include/qemu/thread.h    | 18 ++++++++++++++++++
 util/qemu-thread-posix.c | 40 ++++++++++++++++++++++++++++------------
 util/qemu-thread-win32.c | 16 ++++++++++++++++
 util/qsp.c               | 18 ++++++++++++++++++
 4 files changed, 80 insertions(+), 12 deletions(-)

diff --git a/include/qemu/thread.h b/include/qemu/thread.h
index 55d83a907c..d0cd7b9ae0 100644
--- a/include/qemu/thread.h
+++ b/include/qemu/thread.h
@@ -34,6 +34,8 @@ typedef void (*QemuRecMutexLockFunc)(QemuRecMutex *m, const char *f, int l);
 typedef int (*QemuRecMutexTrylockFunc)(QemuRecMutex *m, const char *f, int l);
 typedef void (*QemuCondWaitFunc)(QemuCond *c, QemuMutex *m, const char *f,
                                  int l);
+typedef void (*QemuCondTimedWaitFunc)(QemuCond *c, QemuMutex *m, int ms,
+                                      const char *f, int l);
 
 extern QemuMutexLockFunc qemu_bql_mutex_lock_func;
 extern QemuMutexLockFunc qemu_mutex_lock_func;
@@ -41,6 +43,7 @@ extern QemuMutexTrylockFunc qemu_mutex_trylock_func;
 extern QemuRecMutexLockFunc qemu_rec_mutex_lock_func;
 extern QemuRecMutexTrylockFunc qemu_rec_mutex_trylock_func;
 extern QemuCondWaitFunc qemu_cond_wait_func;
+extern QemuCondTimedWaitFunc qemu_cond_timedwait_func;
 
 /* convenience macros to bypass the profiler */
 #define qemu_mutex_lock__raw(m)                         \
@@ -63,6 +66,8 @@ extern QemuCondWaitFunc qemu_cond_wait_func;
             qemu_rec_mutex_trylock_impl(m, __FILE__, __LINE__);
 #define qemu_cond_wait(c, m)                                            \
             qemu_cond_wait_impl(c, m, __FILE__, __LINE__);
+#define qemu_cond_timedwait(c, m, ms)                                   \
+            qemu_cond_wait_impl(c, m, ms, __FILE__, __LINE__);
 #else
 #define qemu_mutex_lock(m) ({                                           \
             QemuMutexLockFunc _f = atomic_read(&qemu_mutex_lock_func);  \
@@ -89,6 +94,11 @@ extern QemuCondWaitFunc qemu_cond_wait_func;
             QemuCondWaitFunc _f = atomic_read(&qemu_cond_wait_func);    \
             _f(c, m, __FILE__, __LINE__);                               \
         })
+
+#define qemu_cond_timedwait(c, m, ms) ({                                       \
+            QemuCondTimedWaitFunc _f = atomic_read(&qemu_cond_timedwait_func); \
+            _f(c, m, ms, __FILE__, __LINE__);                                  \
+        })
 #endif
 
 #define qemu_mutex_unlock(mutex) \
@@ -134,12 +144,20 @@ void qemu_cond_signal(QemuCond *cond);
 void qemu_cond_broadcast(QemuCond *cond);
 void qemu_cond_wait_impl(QemuCond *cond, QemuMutex *mutex,
                          const char *file, const int line);
+void qemu_cond_timedwait_impl(QemuCond *cond, QemuMutex *mutex, int ms,
+                              const char *file, const int line);
 
 static inline void (qemu_cond_wait)(QemuCond *cond, QemuMutex *mutex)
 {
     qemu_cond_wait(cond, mutex);
 }
 
+static inline void (qemu_cond_timedwait)(QemuCond *cond, QemuMutex *mutex,
+                                         int ms)
+{
+    qemu_cond_timedwait(cond, mutex, ms);
+}
+
 void qemu_sem_init(QemuSemaphore *sem, int init);
 void qemu_sem_post(QemuSemaphore *sem);
 void qemu_sem_wait(QemuSemaphore *sem);
diff --git a/util/qemu-thread-posix.c b/util/qemu-thread-posix.c
index 1bf5e65dea..eed777d9ec 100644
--- a/util/qemu-thread-posix.c
+++ b/util/qemu-thread-posix.c
@@ -36,6 +36,18 @@ static void error_exit(int err, const char *msg)
     abort();
 }
 
+static void compute_abs_deadline(struct timespec *ts, int ms)
+{
+    struct timeval tv;
+    gettimeofday(&tv, NULL);
+    ts->tv_nsec = tv.tv_usec * 1000 + (ms % 1000) * 1000000;
+    ts->tv_sec = tv.tv_sec + ms / 1000;
+    if (ts->tv_nsec >= 1000000000) {
+        ts->tv_sec++;
+        ts->tv_nsec -= 1000000000;
+    }
+}
+
 void qemu_mutex_init(QemuMutex *mutex)
 {
     int err;
@@ -164,6 +176,22 @@ void qemu_cond_wait_impl(QemuCond *cond, QemuMutex *mutex, const char *file, con
         error_exit(err, __func__);
 }
 
+void qemu_cond_timedwait_impl(QemuCond *cond, QemuMutex *mutex, int ms,
+                              const char *file, const int line)
+{
+    int err;
+    struct timespec ts;
+
+    assert(cond->initialized);
+    trace_qemu_mutex_unlock(mutex, file, line);
+    compute_abs_deadline(&ts, ms);
+    err = pthread_cond_timedwait(&cond->cond, &mutex->lock, &ts);
+    trace_qemu_mutex_locked(mutex, file, line);
+    if (err && err != ETIMEDOUT) {
+        error_exit(err, __func__);
+    }
+}
+
 void qemu_sem_init(QemuSemaphore *sem, int init)
 {
     int rc;
@@ -238,18 +266,6 @@ void qemu_sem_post(QemuSemaphore *sem)
 #endif
 }
 
-static void compute_abs_deadline(struct timespec *ts, int ms)
-{
-    struct timeval tv;
-    gettimeofday(&tv, NULL);
-    ts->tv_nsec = tv.tv_usec * 1000 + (ms % 1000) * 1000000;
-    ts->tv_sec = tv.tv_sec + ms / 1000;
-    if (ts->tv_nsec >= 1000000000) {
-        ts->tv_sec++;
-        ts->tv_nsec -= 1000000000;
-    }
-}
-
 int qemu_sem_timedwait(QemuSemaphore *sem, int ms)
 {
     int rc;
diff --git a/util/qemu-thread-win32.c b/util/qemu-thread-win32.c
index 572f88535d..5faa01cb61 100644
--- a/util/qemu-thread-win32.c
+++ b/util/qemu-thread-win32.c
@@ -145,6 +145,22 @@ void qemu_cond_wait_impl(QemuCond *cond, QemuMutex *mutex, const char *file, con
     qemu_mutex_post_lock(mutex, file, line);
 }
 
+void qemu_cond_timedwait_impl(QemuCond *cond, QemuMutex *mutex, int ms,
+                              const char *file, const int line)
+{
+    int rc = 0;
+
+    assert(cond->initialized);
+    trace_qemu_mutex_unlock(mutex, file, line);
+    if (!SleepConditionVariableSRW(&cond->var, &mutex->lock, ms, 0)) {
+        rc = GetLastError();
+    }
+    trace_qemu_mutex_locked(mutex, file, line);
+    if (rc && rc != ERROR_TIMEOUT) {
+        error_exit(rc, __func__);
+    }
+}
+
 void qemu_sem_init(QemuSemaphore *sem, int init)
 {
     /* Manual reset.  */
diff --git a/util/qsp.c b/util/qsp.c
index 5264c97342..904dcb7436 100644
--- a/util/qsp.c
+++ b/util/qsp.c
@@ -131,6 +131,7 @@ QemuRecMutexLockFunc qemu_rec_mutex_lock_func = qemu_rec_mutex_lock_impl;
 QemuRecMutexTrylockFunc qemu_rec_mutex_trylock_func =
     qemu_rec_mutex_trylock_impl;
 QemuCondWaitFunc qemu_cond_wait_func = qemu_cond_wait_impl;
+QemuCondTimedWaitFunc qemu_cond_timedwait_func = qemu_cond_timedwait_impl;
 
 /*
  * It pays off to _not_ hash callsite->file; hashing a string is slow, and
@@ -412,6 +413,21 @@ qsp_cond_wait(QemuCond *cond, QemuMutex *mutex, const char *file, int line)
     qsp_entry_record(e, t1 - t0);
 }
 
+static void
+qsp_cond_timedwait(QemuCond *cond, QemuMutex *mutex, int ms,
+                   const char *file, int line)
+{
+    QSPEntry *e;
+    int64_t t0, t1;
+
+    t0 = get_clock();
+    qemu_cond_timedwait_impl(cond, mutex, ms, file, line);
+    t1 = get_clock();
+
+    e = qsp_entry_get(cond, file, line, QSP_CONDVAR);
+    qsp_entry_record(e, t1 - t0);
+}
+
 bool qsp_is_enabled(void)
 {
     return atomic_read(&qemu_mutex_lock_func) == qsp_mutex_lock;
@@ -425,6 +441,7 @@ void qsp_enable(void)
     atomic_set(&qemu_rec_mutex_lock_func, qsp_rec_mutex_lock);
     atomic_set(&qemu_rec_mutex_trylock_func, qsp_rec_mutex_trylock);
     atomic_set(&qemu_cond_wait_func, qsp_cond_wait);
+    atomic_set(&qemu_cond_timedwait_func, qsp_cond_timedwait);
 }
 
 void qsp_disable(void)
@@ -435,6 +452,7 @@ void qsp_disable(void)
     atomic_set(&qemu_rec_mutex_lock_func, qemu_rec_mutex_lock_impl);
     atomic_set(&qemu_rec_mutex_trylock_func, qemu_rec_mutex_trylock_impl);
     atomic_set(&qemu_cond_wait_func, qemu_cond_wait_impl);
+    atomic_set(&qemu_cond_timedwait_func, qemu_cond_timedwait_impl);
 }
 
 static gint qsp_tree_cmp(gconstpointer ap, gconstpointer bp, gpointer up)
-- 
2.22.0

[Qemu-devel] [PATCH v5 2/3] cpus: Fix throttling during vm_stop

[Yury Kotov]

Throttling thread sleeps in VCPU thread. For high throttle percentage
this sleep is more than 10ms. E.g. for 60% - 15ms, for 99% - 990ms.
vm_stop() kicks all VCPUs and waits for them. It's called at the end of
migration and because of the long sleep the migration downtime might be
more than 100ms even for downtime-limit 1ms.
Use qemu_cond_timedwait for high percentage to wake up during vm_stop.

Signed-off-by: Yury Kotov <yury-kotov@yandex-team.ru>
---
 cpus.c | 25 +++++++++++++++++--------
 1 file changed, 17 insertions(+), 8 deletions(-)

diff --git a/cpus.c b/cpus.c
index 85cd451a86..d2c61ff155 100644
--- a/cpus.c
+++ b/cpus.c
@@ -77,6 +77,8 @@
 
 #endif /* CONFIG_LINUX */
 
+static QemuMutex qemu_global_mutex;
+
 int64_t max_delay;
 int64_t max_advance;
 
@@ -782,7 +784,7 @@ static void cpu_throttle_thread(CPUState *cpu, run_on_cpu_data opaque)
 {
     double pct;
     double throttle_ratio;
-    long sleeptime_ns;
+    int64_t sleeptime_ns, endtime_ns;
 
     if (!cpu_throttle_get_percentage()) {
         return;
@@ -790,11 +792,20 @@ static void cpu_throttle_thread(CPUState *cpu, run_on_cpu_data opaque)
 
     pct = (double)cpu_throttle_get_percentage()/100;
     throttle_ratio = pct / (1 - pct);
-    sleeptime_ns = (long)(throttle_ratio * CPU_THROTTLE_TIMESLICE_NS);
-
-    qemu_mutex_unlock_iothread();
-    g_usleep(sleeptime_ns / 1000); /* Convert ns to us for usleep call */
-    qemu_mutex_lock_iothread();
+    /* Add 1ns to fix double's rounding error (like 0.9999999...) */
+    sleeptime_ns = (int64_t)(throttle_ratio * CPU_THROTTLE_TIMESLICE_NS + 1);
+    endtime_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + sleeptime_ns;
+    while (sleeptime_ns > 0 && !cpu->stop) {
+        if (sleeptime_ns > SCALE_MS) {
+            qemu_cond_timedwait(cpu->halt_cond, &qemu_global_mutex,
+                                sleeptime_ns / SCALE_MS);
+        } else {
+            qemu_mutex_unlock_iothread();
+            g_usleep(sleeptime_ns / SCALE_US);
+            qemu_mutex_lock_iothread();
+        }
+        sleeptime_ns = endtime_ns - qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
+    }
     atomic_set(&cpu->throttle_thread_scheduled, 0);
 }
 
@@ -1172,8 +1183,6 @@ static void qemu_init_sigbus(void)
 }
 #endif /* !CONFIG_LINUX */
 
-static QemuMutex qemu_global_mutex;
-
 static QemuThread io_thread;
 
 /* cpu creation */
-- 
2.22.0

[Qemu-devel] [PATCH v5 3/3] tests/migration: Add a test for auto converge

[Yury Kotov]

Signed-off-by: Yury Kotov <yury-kotov@yandex-team.ru>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
---
 tests/migration-test.c | 120 +++++++++++++++++++++++++++++++++++++----
 1 file changed, 109 insertions(+), 11 deletions(-)

diff --git a/tests/migration-test.c b/tests/migration-test.c
index b87ba99a9e..e5caf93dfa 100644
--- a/tests/migration-test.c
+++ b/tests/migration-test.c
@@ -240,6 +240,17 @@ static int64_t read_ram_property_int(QTestState *who, const char *property)
     return result;
 }
 
+static int64_t read_migrate_property_int(QTestState *who, const char *property)
+{
+    QDict *rsp_return;
+    int64_t result;
+
+    rsp_return = migrate_query(who);
+    result = qdict_get_try_int(rsp_return, property, 0);
+    qobject_unref(rsp_return);
+    return result;
+}
+
 static uint64_t get_migration_pass(QTestState *who)
 {
     return read_ram_property_int(who, "dirty-sync-count");
@@ -254,20 +265,22 @@ static void read_blocktime(QTestState *who)
     qobject_unref(rsp_return);
 }
 
+static bool check_migration_status(QTestState *who, const char *status)
+{
+    bool completed;
+    char *current_status;
+
+    current_status = migrate_query_status(who);
+    completed = strcmp(current_status, status) == 0;
+    g_assert_cmpstr(current_status, !=, "failed");
+    g_free(current_status);
+    return completed;
+}
+
 static void wait_for_migration_status(QTestState *who,
                                       const char *goal)
 {
-    while (true) {
-        bool completed;
-        char *status;
-
-        status = migrate_query_status(who);
-        completed = strcmp(status, goal) == 0;
-        g_assert_cmpstr(status, !=,  "failed");
-        g_free(status);
-        if (completed) {
-            return;
-        }
+    while (!check_migration_status(who, goal)) {
         usleep(1000);
     }
 }
@@ -1125,6 +1138,90 @@ static void test_migrate_fd_proto(void)
     test_migrate_end(from, to, true);
 }
 
+static void test_migrate_auto_converge(void)
+{
+    char *uri = g_strdup_printf("unix:%s/migsocket", tmpfs);
+    QTestState *from, *to;
+    int64_t remaining, percentage;
+
+    /*
+     * We want the test to be stable and as fast as possible.
+     * E.g., with 1Gb/s bandwith migration may pass without throttling,
+     * so we need to decrease a bandwidth.
+     */
+    const int64_t init_pct = 5, inc_pct = 50, max_pct = 95;
+    const int64_t max_bandwidth = 400000000; /* ~400Mb/s */
+    const int64_t downtime_limit = 250; /* 250ms */
+    /*
+     * We migrate through unix-socket (> 500Mb/s).
+     * Thus, expected migration speed ~= bandwidth limit (< 500Mb/s).
+     * So, we can predict expected_threshold
+     */
+    const int64_t expected_threshold = max_bandwidth * downtime_limit / 1000;
+
+    if (test_migrate_start(&from, &to, uri, false, false)) {
+        return;
+    }
+
+    migrate_set_capability(from, "auto-converge", true);
+    migrate_set_parameter_int(from, "cpu-throttle-initial", init_pct);
+    migrate_set_parameter_int(from, "cpu-throttle-increment", inc_pct);
+    migrate_set_parameter_int(from, "max-cpu-throttle", max_pct);
+
+    /*
+     * Set the initial parameters so that the migration could not converge
+     * without throttling.
+     */
+    migrate_set_parameter_int(from, "downtime-limit", 1);
+    migrate_set_parameter_int(from, "max-bandwidth", 100000000); /* ~100Mb/s */
+
+    /* To check remaining size after precopy */
+    migrate_set_capability(from, "pause-before-switchover", true);
+
+    /* Wait for the first serial output from the source */
+    wait_for_serial("src_serial");
+
+    migrate(from, uri, "{}");
+
+    /* Wait for throttling begins */
+    percentage = 0;
+    while (percentage == 0) {
+        percentage = read_migrate_property_int(from, "cpu-throttle-percentage");
+        usleep(100);
+        g_assert_false(got_stop);
+    }
+    /* The first percentage of throttling should be equal to init_pct */
+    g_assert_cmpint(percentage, ==, init_pct);
+    /* Now, when we tested that throttling works, let it converge */
+    migrate_set_parameter_int(from, "downtime-limit", downtime_limit);
+    migrate_set_parameter_int(from, "max-bandwidth", max_bandwidth);
+
+    /*
+     * Wait for pre-switchover status to check last throttle percentage
+     * and remaining. These values will be zeroed later
+     */
+    wait_for_migration_status(from, "pre-switchover");
+
+    /* The final percentage of throttling shouldn't be greater than max_pct */
+    percentage = read_migrate_property_int(from, "cpu-throttle-percentage");
+    g_assert_cmpint(percentage, <=, max_pct);
+
+    remaining = read_ram_property_int(from, "remaining");
+    g_assert_cmpint(remaining, <, expected_threshold);
+
+    wait_command(from, "{ 'execute': 'migrate-continue' , 'arguments':"
+                       "  { 'state': 'pre-switchover' } }");
+
+    qtest_qmp_eventwait(to, "RESUME");
+
+    wait_for_serial("dest_serial");
+    wait_for_migration_complete(from);
+
+    g_free(uri);
+
+    test_migrate_end(from, to, true);
+}
+
 int main(int argc, char **argv)
 {
     char template[] = "/tmp/migration-test-XXXXXX";
@@ -1180,6 +1277,7 @@ int main(int argc, char **argv)
     /* qtest_add_func("/migration/ignore_shared", test_ignore_shared); */
     qtest_add_func("/migration/xbzrle/unix", test_xbzrle_unix);
     qtest_add_func("/migration/fd_proto", test_migrate_fd_proto);
+    qtest_add_func("/migration/auto_converge", test_migrate_auto_converge);
 
     ret = g_test_run();
 
-- 
2.22.0

Re: [Qemu-devel] [PATCH v5 3/3] tests/migration: Add a test for auto converge

[Yury Kotov]

David,

I kept your Reviewed-by, but could you take another look at the test?
I've made many changes in it.

Thanks!

26.08.2019, 13:39, "Yury Kotov" <yury-kotov@yandex-team.ru>:
> Signed-off-by: Yury Kotov <yury-kotov@yandex-team.ru>
> Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
> ---
>  tests/migration-test.c | 120 +++++++++++++++++++++++++++++++++++++----
>  1 file changed, 109 insertions(+), 11 deletions(-)
>
> diff --git a/tests/migration-test.c b/tests/migration-test.c
> index b87ba99a9e..e5caf93dfa 100644
> --- a/tests/migration-test.c
> +++ b/tests/migration-test.c
> @@ -240,6 +240,17 @@ static int64_t read_ram_property_int(QTestState *who, const char *property)
>      return result;
>  }
>
> +static int64_t read_migrate_property_int(QTestState *who, const char *property)
> +{
> + QDict *rsp_return;
> + int64_t result;
> +
> + rsp_return = migrate_query(who);
> + result = qdict_get_try_int(rsp_return, property, 0);
> + qobject_unref(rsp_return);
> + return result;
> +}
> +
>  static uint64_t get_migration_pass(QTestState *who)
>  {
>      return read_ram_property_int(who, "dirty-sync-count");
> @@ -254,20 +265,22 @@ static void read_blocktime(QTestState *who)
>      qobject_unref(rsp_return);
>  }
>
> +static bool check_migration_status(QTestState *who, const char *status)
> +{
> + bool completed;
> + char *current_status;
> +
> + current_status = migrate_query_status(who);
> + completed = strcmp(current_status, status) == 0;
> + g_assert_cmpstr(current_status, !=, "failed");
> + g_free(current_status);
> + return completed;
> +}
> +
>  static void wait_for_migration_status(QTestState *who,
>                                        const char *goal)
>  {
> - while (true) {
> - bool completed;
> - char *status;
> -
> - status = migrate_query_status(who);
> - completed = strcmp(status, goal) == 0;
> - g_assert_cmpstr(status, !=, "failed");
> - g_free(status);
> - if (completed) {
> - return;
> - }
> + while (!check_migration_status(who, goal)) {
>          usleep(1000);
>      }
>  }
> @@ -1125,6 +1138,90 @@ static void test_migrate_fd_proto(void)
>      test_migrate_end(from, to, true);
>  }
>
> +static void test_migrate_auto_converge(void)
> +{
> + char *uri = g_strdup_printf("unix:%s/migsocket", tmpfs);
> + QTestState *from, *to;
> + int64_t remaining, percentage;
> +
> + /*
> + * We want the test to be stable and as fast as possible.
> + * E.g., with 1Gb/s bandwith migration may pass without throttling,
> + * so we need to decrease a bandwidth.
> + */
> + const int64_t init_pct = 5, inc_pct = 50, max_pct = 95;
> + const int64_t max_bandwidth = 400000000; /* ~400Mb/s */
> + const int64_t downtime_limit = 250; /* 250ms */
> + /*
> + * We migrate through unix-socket (> 500Mb/s).
> + * Thus, expected migration speed ~= bandwidth limit (< 500Mb/s).
> + * So, we can predict expected_threshold
> + */
> + const int64_t expected_threshold = max_bandwidth * downtime_limit / 1000;
> +
> + if (test_migrate_start(&from, &to, uri, false, false)) {
> + return;
> + }
> +
> + migrate_set_capability(from, "auto-converge", true);
> + migrate_set_parameter_int(from, "cpu-throttle-initial", init_pct);
> + migrate_set_parameter_int(from, "cpu-throttle-increment", inc_pct);
> + migrate_set_parameter_int(from, "max-cpu-throttle", max_pct);
> +
> + /*
> + * Set the initial parameters so that the migration could not converge
> + * without throttling.
> + */
> + migrate_set_parameter_int(from, "downtime-limit", 1);
> + migrate_set_parameter_int(from, "max-bandwidth", 100000000); /* ~100Mb/s */
> +
> + /* To check remaining size after precopy */
> + migrate_set_capability(from, "pause-before-switchover", true);
> +
> + /* Wait for the first serial output from the source */
> + wait_for_serial("src_serial");
> +
> + migrate(from, uri, "{}");
> +
> + /* Wait for throttling begins */
> + percentage = 0;
> + while (percentage == 0) {
> + percentage = read_migrate_property_int(from, "cpu-throttle-percentage");
> + usleep(100);
> + g_assert_false(got_stop);
> + }
> + /* The first percentage of throttling should be equal to init_pct */
> + g_assert_cmpint(percentage, ==, init_pct);
> + /* Now, when we tested that throttling works, let it converge */
> + migrate_set_parameter_int(from, "downtime-limit", downtime_limit);
> + migrate_set_parameter_int(from, "max-bandwidth", max_bandwidth);
> +
> + /*
> + * Wait for pre-switchover status to check last throttle percentage
> + * and remaining. These values will be zeroed later
> + */
> + wait_for_migration_status(from, "pre-switchover");
> +
> + /* The final percentage of throttling shouldn't be greater than max_pct */
> + percentage = read_migrate_property_int(from, "cpu-throttle-percentage");
> + g_assert_cmpint(percentage, <=, max_pct);
> +
> + remaining = read_ram_property_int(from, "remaining");
> + g_assert_cmpint(remaining, <, expected_threshold);
> +
> + wait_command(from, "{ 'execute': 'migrate-continue' , 'arguments':"
> + " { 'state': 'pre-switchover' } }");
> +
> + qtest_qmp_eventwait(to, "RESUME");
> +
> + wait_for_serial("dest_serial");
> + wait_for_migration_complete(from);
> +
> + g_free(uri);
> +
> + test_migrate_end(from, to, true);
> +}
> +
>  int main(int argc, char **argv)
>  {
>      char template[] = "/tmp/migration-test-XXXXXX";
> @@ -1180,6 +1277,7 @@ int main(int argc, char **argv)
>      /* qtest_add_func("/migration/ignore_shared", test_ignore_shared); */
>      qtest_add_func("/migration/xbzrle/unix", test_xbzrle_unix);
>      qtest_add_func("/migration/fd_proto", test_migrate_fd_proto);
> + qtest_add_func("/migration/auto_converge", test_migrate_auto_converge);
>
>      ret = g_test_run();
>
> --
> 2.22.0

Regards,
Yury

Re: [Qemu-devel] [PATCH v5 1/3] qemu-thread: Add qemu_cond_timedwait

[Eric Blake]

On 8/26/19 5:37 AM, Yury Kotov wrote:
> Signed-off-by: Yury Kotov <yury-kotov@yandex-team.ru>
> ---

Rather sparse on the commit message details.

>  include/qemu/thread.h    | 18 ++++++++++++++++++
>  util/qemu-thread-posix.c | 40 ++++++++++++++++++++++++++++------------
>  util/qemu-thread-win32.c | 16 ++++++++++++++++
>  util/qsp.c               | 18 ++++++++++++++++++
>  4 files changed, 80 insertions(+), 12 deletions(-)
> 

> +++ b/util/qemu-thread-posix.c
> @@ -36,6 +36,18 @@ static void error_exit(int err, const char *msg)
>      abort();
>  }
>  
> +static void compute_abs_deadline(struct timespec *ts, int ms)
> +{
> +    struct timeval tv;
> +    gettimeofday(&tv, NULL);
> +    ts->tv_nsec = tv.tv_usec * 1000 + (ms % 1000) * 1000000;
> +    ts->tv_sec = tv.tv_sec + ms / 1000;
> +    if (ts->tv_nsec >= 1000000000) {
> +        ts->tv_sec++;
> +        ts->tv_nsec -= 1000000000;
> +    }

I don't know if any named constants would make this easier or harder to
read (such as USEC_PER_SEC 1000000 or NSEC_PER_SEC 1000000000), but the
conversion from relative ms to absolute timespec looks correct. [1]

> +void qemu_cond_timedwait_impl(QemuCond *cond, QemuMutex *mutex, int ms,
> +                              const char *file, const int line)
> +{
> +    int err;
> +    struct timespec ts;
> +
> +    assert(cond->initialized);
> +    trace_qemu_mutex_unlock(mutex, file, line);
> +    compute_abs_deadline(&ts, ms);
> +    err = pthread_cond_timedwait(&cond->cond, &mutex->lock, &ts);
> +    trace_qemu_mutex_locked(mutex, file, line);
> +    if (err && err != ETIMEDOUT) {
> +        error_exit(err, __func__);
> +    }
> +}

However, this function returning void looks odd.  Although ETIMEDOUT is
the one error that guarantees that mutex is reobtained (all other errors
occur before the mutex is given up in the first place), and even though
the man page warns that you MUST recheck the condition variable in a
while loop regardless of success or failure (it might be a spurious
successful wake-up due to a broadcast where neither the condition nor
the timeout has actually been reached yet; or it might be a race where
the function reports a timeout immediately before the condition variable
became available after all), it still seems like callers might like to
know if a timeout happened, without having to calculate an ending
absolute time themselves.


>  
> -static void compute_abs_deadline(struct timespec *ts, int ms)
> -{
> -    struct timeval tv;

[1] Oh, you mixed code motion with new code, but the commit message
didn't mention that.  It's not necessarily worth splitting the patch,
but at least mentioning it would be worthwhile.

> +++ b/util/qemu-thread-win32.c
> @@ -145,6 +145,22 @@ void qemu_cond_wait_impl(QemuCond *cond, QemuMutex *mutex, const char *file, con
>      qemu_mutex_post_lock(mutex, file, line);
>  }
>  
> +void qemu_cond_timedwait_impl(QemuCond *cond, QemuMutex *mutex, int ms,
> +                              const char *file, const int line)
> +{
> +    int rc = 0;
> +
> +    assert(cond->initialized);
> +    trace_qemu_mutex_unlock(mutex, file, line);
> +    if (!SleepConditionVariableSRW(&cond->var, &mutex->lock, ms, 0)) {
> +        rc = GetLastError();
> +    }
> +    trace_qemu_mutex_locked(mutex, file, line);
> +    if (rc && rc != ERROR_TIMEOUT) {
> +        error_exit(rc, __func__);
> +    }
> +}

I am less certain that this implementation is correct, but on the
surface it seems okay.


>  
> +static void
> +qsp_cond_timedwait(QemuCond *cond, QemuMutex *mutex, int ms,
> +                   const char *file, int line)
> +{
> +    QSPEntry *e;
> +    int64_t t0, t1;
> +
> +    t0 = get_clock();
> +    qemu_cond_timedwait_impl(cond, mutex, ms, file, line);
> +    t1 = get_clock();
> +
> +    e = qsp_entry_get(cond, file, line, QSP_CONDVAR);
> +    qsp_entry_record(e, t1 - t0);
> +}

Another function where a bool or int return (to distinguish success from
timeout) might be worthwhile to some callers.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org

Re: [Qemu-devel] [PATCH v5 2/3] cpus: Fix throttling during vm_stop

[Eric Blake]

On 8/26/19 5:37 AM, Yury Kotov wrote:
> Throttling thread sleeps in VCPU thread. For high throttle percentage
> this sleep is more than 10ms. E.g. for 60% - 15ms, for 99% - 990ms.
> vm_stop() kicks all VCPUs and waits for them. It's called at the end of
> migration and because of the long sleep the migration downtime might be
> more than 100ms even for downtime-limit 1ms.
> Use qemu_cond_timedwait for high percentage to wake up during vm_stop.
> 
> Signed-off-by: Yury Kotov <yury-kotov@yandex-team.ru>
> ---
>  cpus.c | 25 +++++++++++++++++--------
>  1 file changed, 17 insertions(+), 8 deletions(-)
> 

> @@ -790,11 +792,20 @@ static void cpu_throttle_thread(CPUState *cpu, run_on_cpu_data opaque)
>  
>      pct = (double)cpu_throttle_get_percentage()/100;
>      throttle_ratio = pct / (1 - pct);
> -    sleeptime_ns = (long)(throttle_ratio * CPU_THROTTLE_TIMESLICE_NS);
> -
> -    qemu_mutex_unlock_iothread();
> -    g_usleep(sleeptime_ns / 1000); /* Convert ns to us for usleep call */
> -    qemu_mutex_lock_iothread();
> +    /* Add 1ns to fix double's rounding error (like 0.9999999...) */
> +    sleeptime_ns = (int64_t)(throttle_ratio * CPU_THROTTLE_TIMESLICE_NS + 1);

The cast to int64_t is not strictly necessary here, but doesn't hurt
(since it shows you DO know you are going from double to 64-bit int).

> +    endtime_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + sleeptime_ns;
> +    while (sleeptime_ns > 0 && !cpu->stop) {
> +        if (sleeptime_ns > SCALE_MS) {
> +            qemu_cond_timedwait(cpu->halt_cond, &qemu_global_mutex,
> +                                sleeptime_ns / SCALE_MS);
> +        } else {
> +            qemu_mutex_unlock_iothread();
> +            g_usleep(sleeptime_ns / SCALE_US);
> +            qemu_mutex_lock_iothread();
> +        }
> +        sleeptime_ns = endtime_ns - qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
> +    }

Looks reasonable.

(I wonder if an alternative approach, of doing a poll() or similar
instead of g_usleep, and using a pipe-to-self where we write to the pipe
in the same scenarios where cpu->halt_cond would be broadcast, in order
to wake up the sleeping poll in a responsive manner, would be any easier
or more efficient - but don't rewrite the patch just because of my question)

Reviewed-by: Eric Blake <eblake@redhat.com>

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org