qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH 1/2] cpu: Provide vcpu throttling interface


From: Jason J. Herne
Subject: Re: [Qemu-devel] [PATCH 1/2] cpu: Provide vcpu throttling interface
Date: Mon, 01 Jun 2015 13:04:56 -0400
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.5.0

On 06/01/2015 11:23 AM, Andrey Korolyov wrote:
On Mon, Jun 1, 2015 at 6:17 PM, Jason J. Herne
<address@hidden> wrote:
Provide a method to throttle guest cpu execution. CPUState is augmented with
timeout controls and throttle start/stop functions. To throttle the guest cpu
the caller simply has to call the throttle start function and provide a ratio of
sleep time to normal execution time.

Signed-off-by: Jason J. Herne <address@hidden>
Reviewed-by: Matthew Rosato <address@hidden>
---
  cpus.c            | 62 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
  include/qom/cpu.h | 46 +++++++++++++++++++++++++++++++++++++++++
  2 files changed, 108 insertions(+)

diff --git a/cpus.c b/cpus.c
index de6469f..7568357 100644
--- a/cpus.c
+++ b/cpus.c
@@ -64,6 +64,9 @@

  #endif /* CONFIG_LINUX */

+/* Number of ms between cpu throttle operations */
+#define CPU_THROTTLE_TIMESLICE 10
+
  static CPUState *next_cpu;
  int64_t max_delay;
  int64_t max_advance;
@@ -919,6 +922,65 @@ static void qemu_kvm_wait_io_event(CPUState *cpu)
      qemu_wait_io_event_common(cpu);
  }

+static void cpu_throttle_thread(void *opq)
+{
+    CPUState *cpu = (CPUState *)opq;
+    long sleeptime_ms = (long)(cpu->throttle_ratio * CPU_THROTTLE_TIMESLICE);
+
+    /* Stop the timer if needed */
+    if (cpu->throttle_timer_stop) {
+        timer_del(cpu->throttle_timer);
+        timer_free(cpu->throttle_timer);
+        cpu->throttle_timer = NULL;
+        return;
+    }
+
+    qemu_mutex_unlock_iothread();
+    g_usleep(sleeptime_ms * 1000); /* Convert ms to us for usleep call */
+    qemu_mutex_lock_iothread();
+
+    timer_mod(cpu->throttle_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) +
+                                   CPU_THROTTLE_TIMESLICE);
+}
+
+static void cpu_throttle_timer_pop(void *opq)
+{
+    CPUState *cpu = (CPUState *)opq;
+
+    async_run_on_cpu(cpu, cpu_throttle_thread, cpu);
+}
+
+void cpu_throttle_start(CPUState *cpu, float throttle_ratio)
+{
+    assert(throttle_ratio > 0);
+    cpu->throttle_ratio = throttle_ratio;
+
+    if (!cpu_throttle_active(cpu)) {
+        cpu->throttle_timer = timer_new_ms(QEMU_CLOCK_REALTIME,
+                                           cpu_throttle_timer_pop, cpu);
+        timer_mod(cpu->throttle_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) +
+                                       CPU_THROTTLE_TIMESLICE);
+        cpu->throttle_timer_stop = false;
+    }
+}
+
+void cpu_throttle_stop(CPUState *cpu)
+{
+    assert(cpu_throttle_active(cpu));
+    cpu->throttle_timer_stop = true;
+}
+
+bool cpu_throttle_active(CPUState *cpu)
+{
+    return (cpu->throttle_timer != NULL);
+}
+
+float cpu_throttle_get_ratio(CPUState *cpu)
+{
+    assert(cpu_throttle_active(cpu));
+    return cpu->throttle_ratio;
+}
+
  static void *qemu_kvm_cpu_thread_fn(void *arg)
  {
      CPUState *cpu = arg;
diff --git a/include/qom/cpu.h b/include/qom/cpu.h
index 39f0f19..9d16e6a 100644
--- a/include/qom/cpu.h
+++ b/include/qom/cpu.h
@@ -310,6 +310,11 @@ struct CPUState {
      uint32_t can_do_io;
      int32_t exception_index; /* used by m68k TCG */

+    /* vcpu throttling controls */
+    QEMUTimer *throttle_timer;
+    bool throttle_timer_stop;
+    float throttle_ratio;
+
      /* Note that this is accessed at the start of every TB via a negative
         offset from AREG0.  Leave this field at the end so as to make the
         (absolute value) offset as small as possible.  This reduces code
@@ -553,6 +558,47 @@ CPUState *qemu_get_cpu(int index);
   */
  bool cpu_exists(int64_t id);

+/**
+ * cpu_throttle_start:
+ * @cpu: The vcpu to throttle
+ *
+ * Throttles a vcpu by forcing it to sleep. The duration of the sleep is a
+ * ratio of sleep time to running time. A ratio of 1.0 corresponds to a 50%
+ * duty cycle (example: 10ms sleep for every 10ms awake).
+ *
+ * cpu_throttle_start can be called as needed to adjust the throttle ratio.
+ * Once the throttling starts, it will remain in effect until cpu_throttle_stop
+ * is called.
+ */
+void cpu_throttle_start(CPUState *cpu, float throttle_ratio);
+
+/**
+ * cpu_throttle_stop:
+ * @cpu: The vcpu to stop throttling
+ *
+ * Stops the vcpu throttling started by cpu_throttle_start.
+ */
+void cpu_throttle_stop(CPUState *cpu);
+
+/**
+ * cpu_throttle_active:
+ * @cpu: The vcpu to check
+ *
+ * Returns %true if this vcpu is currently being throttled, %false otherwise.
+ */
+bool cpu_throttle_active(CPUState *cpu);
+
+/**
+ * cpu_throttle_get_ratio:
+ * @cpu: The vcpu whose throttle ratio to return.
+ *
+ * Returns the ratio being used to throttle this vcpu. See cpu_throttle_start
+ * for details.
+ *
+ * Returns The ratio being used to throttle this vcpu.
+ */
+float cpu_throttle_get_ratio(CPUState *cpu);
+
  #ifndef CONFIG_USER_ONLY

  typedef void (*CPUInterruptHandler)(CPUState *, int);
--
1.9.1



Thanks Jason, this patch would be quite interesting as it eliminates
slight overhead from scheduler when cgroups are actively used for same
task (~5% for per-vcpu cgroup layout simular to libvirt`s one for
guest perf numa bench). Are you planning to add wakeup frequency
throttler as well to same interface?


No, I was not planning on adding anything. I recognise that the controls
this patch provides are not an ideal throttling mechanism. But it seems
better than what we have today for auto-converge. That said, if people
like the interface and the mechanism it certainly can be enhanced.


--
-- Jason J. Herne (address@hidden)




reply via email to

[Prev in Thread] Current Thread [Next in Thread]