migration: Support periodic RAMBlock dirty bitmap sync

When VM is configured with huge memory, the current throttle logic
doesn't look like to scale, because migration_trigger_throttle()
is only called for each iteration, so it won't be invoked for a long
time if one iteration can take a long time.

The periodic dirty sync aims to fix the above issue by synchronizing
the ramblock from remote dirty bitmap and, when necessary, triggering
the CPU throttle multiple times during a long iteration.

This is a trade-off between synchronization overhead and CPU throttle
impact.

Signed-off-by: Hyman Huang <yong.huang@smartx.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/f61f1b3653f2acf026901103e1c73d157d38b08f.1729146786.git.yong.huang@smartx.com
[peterx: make prev_cnt global, and reset for each migration]
Signed-off-by: Peter Xu <peterx@redhat.com>

5 files changed, 84 insertions, 4 deletions

diff --git a/migration/cpu-throttle.c b/migration/cpu-throttle.c
index 7632dc6143..5179019e33 100644
--- a/migration/cpu-throttle.c
+++ b/migration/cpu-throttle.c
@@ -28,16 +28,23 @@
 #include "qemu/main-loop.h"
 #include "sysemu/cpus.h"
 #include "sysemu/cpu-throttle.h"
+#include "migration.h"
+#include "migration-stats.h"
 #include "trace.h"
 
 /* vcpu throttling controls */
-static QEMUTimer *throttle_timer;
+static QEMUTimer *throttle_timer, *throttle_dirty_sync_timer;
 static unsigned int throttle_percentage;
+static bool throttle_dirty_sync_timer_active;
+static uint64_t throttle_dirty_sync_count_prev;
 
 #define CPU_THROTTLE_PCT_MIN 1
 #define CPU_THROTTLE_PCT_MAX 99
 #define CPU_THROTTLE_TIMESLICE_NS 10000000
 
+/* Making sure RAMBlock dirty bitmap is synchronized every five seconds */
+#define CPU_THROTTLE_DIRTY_SYNC_TIMESLICE_MS 5000
+
 static void cpu_throttle_thread(CPUState *cpu, run_on_cpu_data opaque)
 {
     double pct;
@@ -112,6 +119,7 @@ void cpu_throttle_set(int new_throttle_pct)
 void cpu_throttle_stop(void)
 {
     qatomic_set(&throttle_percentage, 0);
+    cpu_throttle_dirty_sync_timer(false);
 }
 
 bool cpu_throttle_active(void)
@@ -124,8 +132,68 @@ int cpu_throttle_get_percentage(void)
     return qatomic_read(&throttle_percentage);
 }
 
+void cpu_throttle_dirty_sync_timer_tick(void *opaque)
+{
+    uint64_t sync_cnt = stat64_get(&mig_stats.dirty_sync_count);
+
+    /*
+     * The first iteration copies all memory anyhow and has no
+     * effect on guest performance, therefore omit it to avoid
+     * paying extra for the sync penalty.
+     */
+    if (sync_cnt <= 1) {
+        goto end;
+    }
+
+    if (sync_cnt == throttle_dirty_sync_count_prev) {
+        trace_cpu_throttle_dirty_sync();
+        WITH_RCU_READ_LOCK_GUARD() {
+            migration_bitmap_sync_precopy(false);
+        }
+    }
+
+end:
+    throttle_dirty_sync_count_prev = stat64_get(&mig_stats.dirty_sync_count);
+
+    timer_mod(throttle_dirty_sync_timer,
+        qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL_RT) +
+            CPU_THROTTLE_DIRTY_SYNC_TIMESLICE_MS);
+}
+
+static bool cpu_throttle_dirty_sync_active(void)
+{
+    return qatomic_read(&throttle_dirty_sync_timer_active);
+}
+
+void cpu_throttle_dirty_sync_timer(bool enable)
+{
+    assert(throttle_dirty_sync_timer);
+
+    if (enable) {
+        if (!cpu_throttle_dirty_sync_active()) {
+            /*
+             * Always reset the dirty sync count cache, in case migration
+             * was cancelled once.
+             */
+            throttle_dirty_sync_count_prev = 0;
+            timer_mod(throttle_dirty_sync_timer,
+                qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL_RT) +
+                    CPU_THROTTLE_DIRTY_SYNC_TIMESLICE_MS);
+            qatomic_set(&throttle_dirty_sync_timer_active, 1);
+        }
+    } else {
+        if (cpu_throttle_dirty_sync_active()) {
+            timer_del(throttle_dirty_sync_timer);
+            qatomic_set(&throttle_dirty_sync_timer_active, 0);
+        }
+    }
+}
+
 void cpu_throttle_init(void)
 {
     throttle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL_RT,
                                   cpu_throttle_timer_tick, NULL);
+    throttle_dirty_sync_timer =
+        timer_new_ms(QEMU_CLOCK_VIRTUAL_RT,
+                     cpu_throttle_dirty_sync_timer_tick, NULL);
 }
diff --git a/migration/migration.c b/migration/migration.c
index 64c1d0c1db..9bcc9db4fb 100644
--- a/migration/migration.c
+++ b/migration/migration.c
@@ -3297,12 +3297,17 @@ static MigIterateState migration_iteration_run(MigrationState *s)
 
 static void migration_iteration_finish(MigrationState *s)
 {
-    /* If we enabled cpu throttling for auto-converge, turn it off. */
+    bql_lock();
+
+    /*
+     * If we enabled cpu throttling for auto-converge, turn it off.
+     * Stopping CPU throttle should be serialized by BQL to avoid
+     * racing for the throttle_dirty_sync_timer.
+     */
     if (migrate_auto_converge()) {
         cpu_throttle_stop();
     }
 
-    bql_lock();
     switch (s->state) {
     case MIGRATION_STATUS_COMPLETED:
         runstate_set(RUN_STATE_POSTMIGRATE);
@@ -3520,6 +3525,11 @@ static void *migration_thread(void *opaque)
         qemu_savevm_send_colo_enable(s->to_dst_file);
     }
 
+    if (migrate_auto_converge()) {
+        /* Start RAMBlock dirty bitmap sync timer */
+        cpu_throttle_dirty_sync_timer(true);
+    }
+
     bql_lock();
     ret = qemu_savevm_state_setup(s->to_dst_file, &local_err);
     bql_unlock();
diff --git a/migration/migration.h b/migration/migration.h
index b9ce5aa4ff..7dc59c5e8d 100644
--- a/migration/migration.h
+++ b/migration/migration.h
@@ -551,4 +551,5 @@ int migration_rp_wait(MigrationState *s);
  */
 void migration_rp_kick(MigrationState *s);
 
+void migration_bitmap_sync_precopy(bool last_stage);
 #endif
diff --git a/migration/ram.c b/migration/ram.c
index 04fd805901..b24e45442c 100644
--- a/migration/ram.c
+++ b/migration/ram.c
@@ -1088,7 +1088,7 @@ static void migration_bitmap_sync(RAMState *rs, bool last_stage)
     }
 }
 
-static void migration_bitmap_sync_precopy(bool last_stage)
+void migration_bitmap_sync_precopy(bool last_stage)
 {
     Error *local_err = NULL;
     assert(ram_state);
diff --git a/migration/trace-events b/migration/trace-events
index 9a19599804..0638183056 100644
--- a/migration/trace-events
+++ b/migration/trace-events
@@ -381,3 +381,4 @@ migration_pagecache_insert(void) "Error allocating page"
 
 # cpu-throttle.c
 cpu_throttle_set(int new_throttle_pct)  "set guest CPU throttled by %d%%"
+cpu_throttle_dirty_sync(void) ""