Re: [Qemu-devel] [PATCH 4/5] Add generic drive hotplugging

[Alexander Graf]

On 27.08.2010, at 11:53, Markus Armbruster wrote:

> Alexander Graf <address@hidden> writes:
> 
>> The monitor command for hotplugging is in i386 specific code. This is just
>> plain wrong, as S390 just learned how to do hotplugging too and needs to
>> get drives for that.
>> 
>> So let's add a generic copy to generic code that handles drive_add in a
>> way that doesn't have pci dependencies.
>> 
>> I'm not fully happy with the patch as is. IMHO there should only be a
>> single target agnostic drive_hot_add function available. How we could
>> potentially fit IF_SCSI in there I don't know though.
>> 
>> Signed-off-by: Alexander Graf <address@hidden>
> 
> Yes, we need a target-agnostic command to add host block devices.  All
> we have now is x86's drive_add, which isn't target-agnostic, because its
> normal function is to add both host and guest part, and the latter is
> entangled with PCI.
> 
> Your patch creates a copy of x86's drive_add with the guest bits
> stripped from the code, so it's usable for non-PCI targets.  Guest bits
> remain in the syntax (first argument, useless baggage with your
> command).  Ugly.
> 
> I'm working on a clean solution for this problem.  If you must have a
> solution right away, and ugly is fine, then your patch should do for
> now.

Yes. IMHO it'd be useful to just broaden the scope of drive_add for now and 
remove it later in favor of your improved blockdev_add framework. So I guess 
the code duplication doesn't really hurt either, since this will all go away 
anyways.

> In my opinion, pci_add, pci_del and drive_add all need to go.  They're
> insufficiently general, and they fail to separate host and guest parts.

Sure :). Please make sure to always keep non-PCI capably systems in mind on 
your redesign!


Alex