[Qemu-devel] Re: [PATCH] Add monitor command for system_reboot

[Anthony Liguori <aliguori@us.ibm.com>]

Ryan Harper wrote:
> * Anthony Liguori <aliguori@us.ibm.com> [2009-05-08 12:00]:
>   
>> Ryan Harper wrote:
>>     
>>> Add a new monitor command (system_reboot) for a soft reboot which uses
>>> system_powerdown to trigger ACPI shutdown in the guest and once shutdown is
>>> complete, trigger a reset instead of exiting qemu.
>>>
>>> Depends on commit a6d6552426dcbf726e5549f08b70c9318d6be14b which enabled 
>>> ACPI
>>> power button support.
>>>
>>> Tested with Ubuntu 9.04 64-bit guest.
>>>
>>> Signed-off-by: Ryan Harper <ryanh@us.ibm.com>
>>>
>>> diff --git a/hw/acpi.c b/hw/acpi.c
>>> index dbaf18a..346d100 100644
>>> --- a/hw/acpi.c
>>> +++ b/hw/acpi.c
>>> @@ -151,7 +151,13 @@ static void pm_ioport_writew(void *opaque, uint32_t 
>>> addr, uint32_t val)
>>>                 sus_typ = (val >> 10) & 7;
>>>                 switch(sus_typ) {
>>>                 case 0: /* soft power off */
>>> -                    qemu_system_shutdown_request();
>>> +                    /* after powerdown, if on system_reboot path, call 
>>> reset +                       instead of shutdown */
>>> +                    if (qemu_reboot_requested()) {
>>> +                        qemu_system_reset_request();
>>> +                    } else {
>>> +                        qemu_system_shutdown_request();
>>> +                    }
>>>                     break;
>>>  
>>>       
>> If qemu_shutdown_requested(), then we'll immediately shutdown the system.
>>     
>
> Right, but this request happens after we've sent the ACPI powerdown
> event.  After we've done the powerdown (acpi aware guests can do a
> shutdown), instead of then calling shutdown, which exits, we call
> reset.
>   

I'm saying, semantically, if you call 'qemu_shutdown_requested()', if it 
returns 1, it means, immediately shutdown the VM--regardless of where 
it's called.

The semantics of qemu_reboot_requested() are, if it returns 1, then only 
when you see an ACPI soft power off, reset the VM.  It's that difference 
in semantics that I think could lead to confusion.

>> Also, the soft reset flag ought to get reset whenever a VM changes it's 
>> state.  That is, if you do a system_reboot, then a system_reset (imagine 
>> that the reboot fails), then you do a normal powerdown in the guest, 
>> instead of shutting off like the user would expect, we'll reboot because 
>> the qemu_reboot_requested() is still true.
>>
>> A good way to do that would be by registering a reset handler in 
>> hw/acpi.c.  In fact, I think that the whole functionality probably 
>> should live in hw/acpi.c.
>>     
>
> OK
>
>   
>> And I think this also needs to be stored as part of the savevm state for 
>> hw/acpi.c.  If you do a system_reboot followed by an immediate live 
>> migration, without the savevm handler, the VM will shutdown completely 
>> after the migration instead of rebooting as expected.
>>     
>
> Would it? I don't see that we are saving powerdown|shutdown|reset
> request flags?  Sounds like all of those flags need to be in the save
> state, and separate patch IMHO.
>   

No, they don't.

A qemu_powerdown_request() call happens from the monitor.  This is to 
allow a graceful shutdown (as opposed to exiting from the monitor).  
This will trigger the TCG loop to immediately exit.  The state doesn't 
need to be saved because you cannot do a migration in between when 
qemu_powerdown_request() is called and when the shutdown actually happens.

-- 
Regards,

Anthony Liguori