Re: [Qemu-devel] [Question] dump memory when host pci device is used by guest

[Jan Kiszka]

On 2011-10-10 11:34, Richard W.M. Jones wrote:
> On Mon, Oct 10, 2011 at 10:10:21AM +0100, Daniel P. Berrange wrote:
>> On Mon, Oct 10, 2011 at 10:08:26AM +0100, Daniel P. Berrange wrote:
>>> On Mon, Oct 10, 2011 at 10:21:02AM +0800, Wen Congyang wrote:
>>>> At 10/09/2011 06:23 PM, Richard W.M. Jones Write:
>>>>> On Sun, Oct 09, 2011 at 10:49:57AM +0200, Jan Kiszka wrote:
>>>>>> As explained in the other replies: It is way more future-proof to use an
>>>>>> interface for this which was designed for it (remote gdb) instead of
>>>>>> artificially relaxing reasonable constraints of the migration mechanism
>>>>>> plus having to follow that format with the post-processing tool.
>>>>>
>>>>> Any interface that isn't "get this information off my production
>>>>> server *now*" so that I can get the server restarted, and send it to
>>>>> an expert to analyse -- is a poor interface, whether it was designed
>>>>> like that or not.  Perhaps we don't have the right interface at all,
>>>>> but remote gdb is not it.
>>>>
>>>> What about the following idea?
>>>>
>>>> Introduce a new monitor command named dump, and this command accepts a 
>>>> filename.
>>>> We can use almost all migration's code. We use this command to dump guest's
>>>> memory, so there is no need to check whether the guest has a unmigratable 
>>>> device.
>>>
>>> I think it would be a good idea of QEMU had a dedicated 'dump' command
>>> for this purpose, even if it was just an alias for 'migrate' initially.
>>> I have never really liked the fact that we abuse the 'migrate' command
>>> to generate a core dump. The resulting data file from this is more
>>> complex than it really needs to be, causing complexity for post-processing
>>> it. The needs of migration, are not entirely aligned with the needs of
>>> core dumping in the long term, so we should allow the possibility of
>>> their impls diverging without impacting apps using them.
>>>
>>> So adding a 'dump' command which wrote out data in a format that was
>>> optimized for offline processing by tools like 'crash' (or the windows
>>> equivalent) would be a good improvement, even if it just reuses the
>>> migrate code for now.
>>
>> The other reason why it would be good, is that we would then have a clearly
>> defined standard "QEMU dump format", instead of "libvirt dump format for 
>> QEMU"
> 
> Other reasons why this is a good idea:
> 
>  - so you don't need 3 processes doing a dance in order to capture a
>    core dump

If it made the overall implementation simpler, that would have been no
valid reason.

The key is only that banning of gdb on certain systems (though I wonder
if you cannot configure it to only support remote targets, thus making
it useless for potential attackers).

> 
>  - so that qemu can initiate a core dump itself (from watchdog events)

This is a management tool decision. A dump command would not have any
pre-condition or auto-trigger.

> 
>  - so that two qemu processes can do a core dump at the same time
>    without trying to use conflicting ports

That is a non-issue (just like you can have multiple monitors of
multiple VMs listening on different TCP ports).

Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux