Re: [Qemu-devel] migration: broken ram_save_pending

[Alexey Kardashevskiy]

On 02/07/2014 10:49 AM, Paolo Bonzini wrote:
> Il 06/02/2014 04:10, Alexey Kardashevskiy ha scritto:
>>>> Ok, I thought Alexey was saying we are not redirtying that handful of 
>>>> pages.
>>
>> Every iteration we read the dirty map from KVM and send all dirty pages
>> across the stream.
> 
> But we never finish because qemu_savevm_state_pending is only called _after_
> the g_usleep?  And thus there's time for the guest to redirty those pages.
> Does something like this fix it (of course for a proper pages the goto
> should be eliminated)?
> 
> diff --git a/migration.c b/migration.c
> index 7235c23..804c3bd 100644
> --- a/migration.c
> +++ b/migration.c
> @@ -589,6 +589,7 @@ static void *migration_thread(void *opaque)
>              } else {
>                  int ret;
>  
> +final_phase:
>                  DPRINTF("done iterating\n");
>                  qemu_mutex_lock_iothread();
>                  start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
> @@ -640,10 +641,16 @@ static void *migration_thread(void *opaque)
>              qemu_file_reset_rate_limit(s->file);
>              initial_time = current_time;
>              initial_bytes = qemu_ftell(s->file);
> -        }
> -        if (qemu_file_rate_limit(s->file)) {
> -            /* usleep expects microseconds */
> -            g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
> +        } else if (qemu_file_rate_limit(s->file)) {
> +            pending_size = qemu_savevm_state_pending(s->file, max_size);
> +            DPRINTF("pending size %" PRIu64 " max %" PRIu64 "\n",
> +                    pending_size, max_size);
> +            if (pending_size >= max_size) {
> +                /* usleep expects microseconds */
> +                g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
> +         } else {
> +                goto final_phase;
> +         }
>          }
>      }


It does not make any difference, will have a closer look on Monday.


-- 
Alexey