Re: [Qemu-devel] ahci drive: how to make it non-bootable?

[Gleb Natapov]

On Wed, May 09, 2012 at 08:10:07PM +0400, Michael Tokarev wrote:
> On 09.05.2012 12:02, Gleb Natapov wrote:
> > On Tue, May 08, 2012 at 09:56:10PM +0400, Michael Tokarev wrote:
> []
> >>> It's two only:
> >>>   -drive if=none,id=<name>,...
> >>>   -device virtio-blk-pci,drive=<name>
> >>
> >> Ok, at least it is not entirely insane :)
> 
> > That's the way you suppose to do that. Does it work if you specify
> > bootindex for virtio this way? If it does then use it. Restoring
> > deprecated hacks is not the way to deal with it. If it does not, then
> > it is a bug that should be fixed.
> 
> Using the "too verbose" version is, well, difficult to type, when
> you have to start many guests - eg, testing etc.  That's why I think
> that 3 options for ahci drives is somewhat insane.
> 
You are complaining to later. Verbose syntax is the only syntax going
forward AFAIK. All other syntaxes are legacy and, as you can see, new
devices will not support them.

But you didn't answer my question :) Does specifying bootindex correctly
solves your problem?

> Technically there is no problem at all at combining, eg, -netdev and
> -device rtl8139, in the same way as current -drive file=foo,if=bar
> works, and it is MUCH easier to type and hence is MUCH user-friendly.
> And much less error-prone, too.
> 
For the record, I do not like this verbosity too.

> I think it'd be a good solution to allow bootindex and some other
> options to be recognizable in context of, eg -drive with if!=none,
> by combining options recognizable from if= with the ones recognizable
> by -drive.  Ditto for -netdev..-device: let -netdev to recognize
> model=, and allow model-specific (e1000|8130|virtio-net-pci|whatever)
> to be recognized as well.  This way there will be no need to specify
> id=foo and netdev=foo, the command line will be shorter and also more
> understandable.
> 
But I do not thing that correct solution is to support multiple command
line syntaxes forever. Lets have one that works even if it is too
verbose.

> Please don't say that the only sane option to start qemu guests is
> to use libvirt.  Qemu has its great value exactly because it allows
> starting guests from command line, which is well-scriptable.
I do not use libvirt. Verbose command line is still scriptable.

> 
> The old hacks which were prematurely removed from qemu-kvm makes the
> life easier for the _user_, which is the main target of the software.
What makes you think it was removed prematurely? If user refuses to
learn proper command line syntax it will not be able to use more things
than bootindex. AHCI for one and IIRC you need verbose syntax to enable
vhost. Changing most of device properties require verbose syntax.

> I'd love to stop using them but sometimes it is not possible.  With
> this extboot thing, qemu-kvm dropped ability to boot from SCSI for
> example, and it turns out there are quite some users of this interface
That's the only example and extboot was never written for booting from
SCSI, that was merely unintended side effect. LSI was always too broken
to be used. Debian would be much better compiling it out from the start
(IMHO).

> exists - despite the fact SCSI is broken, there is a proprietary
virtio-scsi is not broken.

> bootrom exists etc -- we just broke users setups without providing
> viable alternative, which, to my view, is unacceptable.
> 
Use proprietary bootrom.

--
                        Gleb.