Re: [Qemu-devel] [PATCH v4 0/3] High downtime with 95+ throttle pct

[Yury Kotov]

Hi,

19.08.2019, 20:11, "Paolo Bonzini" <address@hidden>:
> On 19/08/19 18:39, Paolo Bonzini wrote:
>>  On 15/08/19 11:13, Yury Kotov wrote:
>>>  Ping ping
>>
>>  Hi,
>>
>>  sorry for the delay, I was waiting for the 4.1 release.
>>
>>  I would like to make a small change so that preemption of QEMU does not
>>  result in overly long sleeps. The following patch on top of yours computes
>>  the throttle-end time just once. Of course you can still be unlucky if
>>  you are preempted at the wrong time, but the window becomes much smaller.

Thanks for your suggestion! I'll use it in the v5.

>
> The new unit test is hanging for me on aarch64-softmmu.
>

Ok, I'll try to fix it in v5.

> Paolo
>
>>  diff --git a/cpus.c b/cpus.c
>>  index d091dbd..d7e2145 100644
>>  --- a/cpus.c
>>  +++ b/cpus.c
>>  @@ -781,7 +781,7 @@ static void cpu_throttle_thread(CPUState *cpu, 
>> run_on_cpu_data opaque)
>>   {
>>       double pct;
>>       double throttle_ratio;
>>  - int64_t sleeptime_ns;
>>  + int64_t sleeptime_ns, end;
>>
>>       if (!cpu_throttle_get_percentage()) {
>>           return;
>>  @@ -792,18 +792,17 @@ static void cpu_throttle_thread(CPUState *cpu, 
>> run_on_cpu_data opaque)
>>       /* Add 1ns to fix double's rounding error (like 0.9999999...) */
>>       sleeptime_ns = (int64_t)(throttle_ratio * CPU_THROTTLE_TIMESLICE_NS + 
>> 1);
>>
>>  - while (sleeptime_ns >= SCALE_MS && !cpu->stop) {
>>  - int64_t start, end;
>>  - start = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
>>  - qemu_cond_timedwait(cpu->halt_cond, &qemu_global_mutex,
>>  - sleeptime_ns / SCALE_MS);
>>  - end = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
>>  - sleeptime_ns -= end - start;
>>  - }
>>  - if (sleeptime_ns >= SCALE_US && !cpu->stop) {
>>  - qemu_mutex_unlock_iothread();
>>  - g_usleep(sleeptime_ns / SCALE_US);
>>  - qemu_mutex_lock_iothread();
>>  + end = 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 = end - qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
>>       }
>>       atomic_set(&cpu->throttle_thread_scheduled, 0);
>>   }

Regards,
Yury