[Qemu-devel] future of the virtio-blk serial number support

[Qemu-devel] future of the virtio-blk serial number support

[Christoph Hellwig]

Back iSeptember 2007 Michael made the serial number support in qemu
optional and off by default, and in October 2009 Rusty reverted the
Linux virtio-blk support for it.  Given that I can't find support in
any other virtio implementation that makes the feature look essentially
dead.

How should we proceed with adding more fields to struct virtio_blk_config?
I would suggest removing the identity field, declaring
VIRTIO_BLK_F_IDENTIFY officially deprecated and adding the new fields
directly after blk_size again, maybe with a comment that these new
features can't be advertized together with VIRTIO_BLK_F_IDENTIFY.

I need to add a new optiomal_io_size field soon to support the block
topology information when using virtio which is quite important when
using RAID arrays as backend, and I'd prefer to do it in a way that's
compatibly with the PCI spec.

[Qemu-devel] Re: future of the virtio-blk serial number support

[Michael S. Tsirkin]

On Thu, Jan 28, 2010 at 01:43:27PM +0100, Christoph Hellwig wrote:
> Back iSeptember 2007 Michael made the serial number support in qemu
> optional and off by default, and in October 2009 Rusty reverted the
> Linux virtio-blk support for it.  Given that I can't find support in
> any other virtio implementation that makes the feature look essentially
> dead.
> 
> How should we proceed with adding more fields to struct virtio_blk_config?
> I would suggest removing the identity field, declaring
> VIRTIO_BLK_F_IDENTIFY officially deprecated and adding the new fields
> directly after blk_size again, maybe with a comment that these new
> features can't be advertized together with VIRTIO_BLK_F_IDENTIFY.
> 
> I need to add a new optiomal_io_size field soon to support the block
> topology information when using virtio which is quite important when
> using RAID arrays as backend, and I'd prefer to do it in a way that's
> compatibly with the PCI spec.


Makes sense to me.

[Qemu-devel] Re: future of the virtio-blk serial number support

[Anthony Liguori]

On 01/28/2010 06:43 AM, Christoph Hellwig wrote:
> Back iSeptember 2007 Michael made the serial number support in qemu
> optional and off by default, and in October 2009 Rusty reverted the
> Linux virtio-blk support for it.  Given that I can't find support in
> any other virtio implementation that makes the feature look essentially
> dead.
>
> How should we proceed with adding more fields to struct virtio_blk_config?
> I would suggest removing the identity field, declaring
> VIRTIO_BLK_F_IDENTIFY officially deprecated and adding the new fields
> directly after blk_size again, maybe with a comment that these new
> features can't be advertized together with VIRTIO_BLK_F_IDENTIFY.
>
> I need to add a new optiomal_io_size field soon to support the block
> topology information when using virtio which is quite important when
> using RAID arrays as backend, and I'd prefer to do it in a way that's
> compatibly with the PCI spec.
>    

Sounds good to me.  IDENTIFY is basically impossible to implement in 
it's current form with virtio-pci.

We should put something in the spec mentioning that config space is 
limited and large values should not be placed in it.  Instead, 
implementations should use a control ring queue to pass large data sets.

Regards,

Anthony Liguori

[Qemu-devel] Re: future of the virtio-blk serial number support

[john cooper]

Christoph Hellwig wrote:
> Back iSeptember 2007 Michael made the serial number support in qemu
> optional and off by default, and in October 2009 Rusty reverted the
> Linux virtio-blk support for it.  Given that I can't find support in
> any other virtio implementation that makes the feature look essentially
> dead.
> 
> How should we proceed with adding more fields to struct virtio_blk_config?
> I would suggest removing the identity field, declaring
> VIRTIO_BLK_F_IDENTIFY officially deprecated and adding the new fields
> directly after blk_size again, maybe with a comment that these new
> features can't be advertized together with VIRTIO_BLK_F_IDENTIFY.

Agreed.  Feel free to remove it from the PCI
config space.  Its an artifact at this point
and I wouldn't even worry about deprecating
it. 

-john

-- 
john.cooper@redhat.com

[Qemu-devel] Re: future of the virtio-blk serial number support

[Rusty Russell]

On Thu, 28 Jan 2010 11:13:27 pm Christoph Hellwig wrote:
> Back iSeptember 2007 Michael made the serial number support in qemu
> optional and off by default, and in October 2009 Rusty reverted the
> Linux virtio-blk support for it.  Given that I can't find support in
> any other virtio implementation that makes the feature look essentially
> dead.
> 
> How should we proceed with adding more fields to struct virtio_blk_config?
> I would suggest removing the identity field, declaring
> VIRTIO_BLK_F_IDENTIFY officially deprecated and adding the new fields
> directly after blk_size again, maybe with a comment that these new
> features can't be advertized together with VIRTIO_BLK_F_IDENTIFY.

Yes, this is the correct approach.

Agreed,
Rusty.