Re: [Qemu-devel] [PATCH] Flush pending AIO on reboot and shutdown.

[Anthony Liguori <anthony@codemonkey.ws>]

Gleb Natapov wrote:
> On Wed, Aug 13, 2008 at 08:52:50AM -0500, Anthony Liguori wrote:
>   
>> Gleb Natapov wrote:
>>     
>>> If there is outstanding IDE IO when BIOS starts execution then IDE
>>> commands sent by BIOS will interfere with it and will leave IDE
>>> subsystem in unpredictable state. This can happen when system reboots
>>> unexpectedly without waiting for IO completion. Flushing IO before exit
>>> prevents data lose.
>>>
>>> Signed-off-by: Gleb Natapov <gleb@qumranet.com>
>>> ---
>>>
>>>  vl.c |    2 ++
>>>  1 files changed, 2 insertions(+), 0 deletions(-)
>>>
>>> diff --git a/vl.c b/vl.c
>>> index e42ae64..afa2a3a 100644
>>> --- a/vl.c
>>> +++ b/vl.c
>>> @@ -7553,6 +7553,7 @@ static int main_loop(void)
>>>              if (reset_requested) {
>>>                  reset_requested = 0;
>>>                  qemu_system_reset();
>>> +                qemu_aio_flush();
>>>   
>>>       
>> Perhaps the aio block layer should do qemu_register_reset() with a  
>> handler that does a flush.
>>
>>     
> I though about doing it this way, but then I saw that qemu_register_reset()
> is used for HW reset notification only. I don't think that hw/ide.c is the
> right place to register the notifier though, so I decided to call it
> explicitly. Do you think I should do qemu_register_reset() in block.c?
>   

But shouldn't the various disk types be the ones to flush their 
outstanding IO on reset?

Regards,

Anthony Liguori

> --
> 			Gleb.
>
>
>