From: Hyman Huang <yong.huang@smartx.com>
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>
---
migration/cpu-throttle.c | 65 +++++++++++++++++++++++++++++++++++++++-
migration/cpu-throttle.h | 14 +++++++++
migration/migration.c | 14 +++++++--
migration/migration.h | 1 +
migration/ram.c | 2 +-
migration/trace-events | 1 +
6 files changed, 93 insertions(+), 4 deletions(-)
diff --git a/migration/cpu-throttle.c b/migration/cpu-throttle.c
index fa47ee2e21..342681cdd4 100644
--- a/migration/cpu-throttle.c
+++ b/migration/cpu-throttle.c
@@ -28,16 +28,22 @@
#include "qemu/main-loop.h"
#include "sysemu/cpus.h"
#include "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;
#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 +118,7 @@ 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 +131,64 @@ int cpu_throttle_get_percentage(void)
return qatomic_read(&throttle_percentage);
}
+void cpu_throttle_dirty_sync_timer_tick(void *opaque)
+{
+ static uint64_t prev_sync_cnt;
+ 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 == prev_sync_cnt) {
+ trace_cpu_throttle_dirty_sync();
+ WITH_RCU_READ_LOCK_GUARD() {
+ migration_bitmap_sync_precopy(false);
+ }
+ }
+
+end:
+ prev_sync_cnt = 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()) {
+ 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/cpu-throttle.h b/migration/cpu-throttle.h
index d65bdef6d0..420702b8d3 100644
--- a/migration/cpu-throttle.h
+++ b/migration/cpu-throttle.h
@@ -65,4 +65,18 @@ bool cpu_throttle_active(void);
*/
int cpu_throttle_get_percentage(void);
+/**
+ * cpu_throttle_dirty_sync_timer_tick:
+ *
+ * Dirty sync timer hook.
+ */
+void cpu_throttle_dirty_sync_timer_tick(void *opaque);
+
+/**
+ * cpu_throttle_dirty_sync_timer:
+ *
+ * Start or stop the dirty sync timer.
+ */
+void cpu_throttle_dirty_sync_timer(bool enable);
+
#endif /* SYSEMU_CPU_THROTTLE_H */
diff --git a/migration/migration.c b/migration/migration.c
index 2e10ca77af..f673e30069 100644
--- a/migration/migration.c
+++ b/migration/migration.c
@@ -3291,12 +3291,17 @@ 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);
@@ -3513,6 +3518,11 @@ 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 38aa1402d5..fbd0d19092 100644
--- a/migration/migration.h
+++ b/migration/migration.h
@@ -537,4 +537,5 @@ 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 9b5b350405..d284f63854 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) ""
--
2.27.0
On Thu, Oct 17, 2024 at 02:42:54PM +0800, yong.huang@smartx.com wrote: > +void cpu_throttle_dirty_sync_timer_tick(void *opaque) > +{ > + static uint64_t prev_sync_cnt; We may need to reset this in case migration got cancelled and invoked again, to make sure it keeps working in the 2nd run. > + 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 == prev_sync_cnt) { > + trace_cpu_throttle_dirty_sync(); > + WITH_RCU_READ_LOCK_GUARD() { > + migration_bitmap_sync_precopy(false); > + } > + } > + > +end: > + prev_sync_cnt = 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); > +} Please both of you have a look on whether you agree I squash below into this patch when merge: ===8<=== From 84a2544eab73e35dbd35fed3b1440169915f9aa4 Mon Sep 17 00:00:00 2001 From: Peter Xu <peterx@redhat.com> Date: Thu, 17 Oct 2024 15:27:19 -0400 Subject: [PATCH] fixup! migration: Support periodic RAMBlock dirty bitmap sync Signed-off-by: Peter Xu <peterx@redhat.com> --- migration/cpu-throttle.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) diff --git a/migration/cpu-throttle.c b/migration/cpu-throttle.c index 342681cdd4..3df287d8d3 100644 --- a/migration/cpu-throttle.c +++ b/migration/cpu-throttle.c @@ -36,6 +36,7 @@ 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 @@ -133,7 +134,6 @@ int cpu_throttle_get_percentage(void) void cpu_throttle_dirty_sync_timer_tick(void *opaque) { - static uint64_t prev_sync_cnt; uint64_t sync_cnt = stat64_get(&mig_stats.dirty_sync_count); /* @@ -145,7 +145,7 @@ void cpu_throttle_dirty_sync_timer_tick(void *opaque) goto end; } - if (sync_cnt == prev_sync_cnt) { + if (sync_cnt == throttle_dirty_sync_count_prev) { trace_cpu_throttle_dirty_sync(); WITH_RCU_READ_LOCK_GUARD() { migration_bitmap_sync_precopy(false); @@ -153,7 +153,7 @@ void cpu_throttle_dirty_sync_timer_tick(void *opaque) } end: - prev_sync_cnt = stat64_get(&mig_stats.dirty_sync_count); + 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) + @@ -171,6 +171,11 @@ void cpu_throttle_dirty_sync_timer(bool enable) 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); -- 2.45.0 -- Peter Xu
On Fri, Oct 18, 2024 at 3:33 AM Peter Xu <peterx@redhat.com> wrote: > On Thu, Oct 17, 2024 at 02:42:54PM +0800, yong.huang@smartx.com wrote: > > +void cpu_throttle_dirty_sync_timer_tick(void *opaque) > > +{ > > + static uint64_t prev_sync_cnt; > > We may need to reset this in case migration got cancelled and invoked > again, to make sure it keeps working in the 2nd run. > > > + 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 == prev_sync_cnt) { > > + trace_cpu_throttle_dirty_sync(); > > + WITH_RCU_READ_LOCK_GUARD() { > > + migration_bitmap_sync_precopy(false); > > + } > > + } > > + > > +end: > > + prev_sync_cnt = 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); > > +} > > Please both of you have a look on whether you agree I squash below into > this patch when merge: > Thanks for the fixup, it looks good to me. > > ===8<=== > From 84a2544eab73e35dbd35fed3b1440169915f9aa4 Mon Sep 17 00:00:00 2001 > From: Peter Xu <peterx@redhat.com> > Date: Thu, 17 Oct 2024 15:27:19 -0400 > Subject: [PATCH] fixup! migration: Support periodic RAMBlock dirty bitmap > sync > > Signed-off-by: Peter Xu <peterx@redhat.com> > --- > migration/cpu-throttle.c | 11 ++++++++--- > 1 file changed, 8 insertions(+), 3 deletions(-) > > diff --git a/migration/cpu-throttle.c b/migration/cpu-throttle.c > index 342681cdd4..3df287d8d3 100644 > --- a/migration/cpu-throttle.c > +++ b/migration/cpu-throttle.c > @@ -36,6 +36,7 @@ > 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 > @@ -133,7 +134,6 @@ int cpu_throttle_get_percentage(void) > > void cpu_throttle_dirty_sync_timer_tick(void *opaque) > { > - static uint64_t prev_sync_cnt; > uint64_t sync_cnt = stat64_get(&mig_stats.dirty_sync_count); > > /* > @@ -145,7 +145,7 @@ void cpu_throttle_dirty_sync_timer_tick(void *opaque) > goto end; > } > > - if (sync_cnt == prev_sync_cnt) { > + if (sync_cnt == throttle_dirty_sync_count_prev) { > trace_cpu_throttle_dirty_sync(); > WITH_RCU_READ_LOCK_GUARD() { > migration_bitmap_sync_precopy(false); > @@ -153,7 +153,7 @@ void cpu_throttle_dirty_sync_timer_tick(void *opaque) > } > > end: > - prev_sync_cnt = stat64_get(&mig_stats.dirty_sync_count); > + 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) + > @@ -171,6 +171,11 @@ void cpu_throttle_dirty_sync_timer(bool enable) > > 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); > -- > 2.45.0 > > -- > Peter Xu > > Acked-by: Hyman Huang <yong.huang@smartx.com> -- Best regards
Peter Xu <peterx@redhat.com> writes: > On Thu, Oct 17, 2024 at 02:42:54PM +0800, yong.huang@smartx.com wrote: >> +void cpu_throttle_dirty_sync_timer_tick(void *opaque) >> +{ >> + static uint64_t prev_sync_cnt; > > We may need to reset this in case migration got cancelled and invoked > again, to make sure it keeps working in the 2nd run. > >> + 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 == prev_sync_cnt) { >> + trace_cpu_throttle_dirty_sync(); >> + WITH_RCU_READ_LOCK_GUARD() { >> + migration_bitmap_sync_precopy(false); >> + } >> + } >> + >> +end: >> + prev_sync_cnt = 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); >> +} > > Please both of you have a look on whether you agree I squash below into > this patch when merge: > > ===8<=== > From 84a2544eab73e35dbd35fed3b1440169915f9aa4 Mon Sep 17 00:00:00 2001 > From: Peter Xu <peterx@redhat.com> > Date: Thu, 17 Oct 2024 15:27:19 -0400 > Subject: [PATCH] fixup! migration: Support periodic RAMBlock dirty bitmap sync > > Signed-off-by: Peter Xu <peterx@redhat.com> > --- > migration/cpu-throttle.c | 11 ++++++++--- > 1 file changed, 8 insertions(+), 3 deletions(-) > > diff --git a/migration/cpu-throttle.c b/migration/cpu-throttle.c > index 342681cdd4..3df287d8d3 100644 > --- a/migration/cpu-throttle.c > +++ b/migration/cpu-throttle.c > @@ -36,6 +36,7 @@ > 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 > @@ -133,7 +134,6 @@ int cpu_throttle_get_percentage(void) > > void cpu_throttle_dirty_sync_timer_tick(void *opaque) > { > - static uint64_t prev_sync_cnt; > uint64_t sync_cnt = stat64_get(&mig_stats.dirty_sync_count); > > /* > @@ -145,7 +145,7 @@ void cpu_throttle_dirty_sync_timer_tick(void *opaque) > goto end; > } > > - if (sync_cnt == prev_sync_cnt) { > + if (sync_cnt == throttle_dirty_sync_count_prev) { > trace_cpu_throttle_dirty_sync(); > WITH_RCU_READ_LOCK_GUARD() { > migration_bitmap_sync_precopy(false); > @@ -153,7 +153,7 @@ void cpu_throttle_dirty_sync_timer_tick(void *opaque) > } > > end: > - prev_sync_cnt = stat64_get(&mig_stats.dirty_sync_count); > + 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) + > @@ -171,6 +171,11 @@ void cpu_throttle_dirty_sync_timer(bool enable) > > 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); > -- > 2.45.0 LGTM
yong.huang@smartx.com writes: > From: Hyman Huang <yong.huang@smartx.com> > > 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>
© 2016 - 2024 Red Hat, Inc.