Re: [PATCH V10 1/3] libxl: Add support for Virtio disk configuration

[Anthony PERARD <anthony.perard@xxxxxxxxxx>]

On Wed, Jul 13, 2022 at 08:03:17AM +0000, Oleksandr Tyshchenko wrote:
> 
> On 13.07.22 03:01, George Dunlap wrote:
> 
> Hello George, Anthony
> 
> >
> >
> >> On 30 Jun 2022, at 22:18, Oleksandr <olekstysh@xxxxxxxxx> wrote:
> >>
> >>
> >> Dear all.
> >>
> >>
> >> On 25.06.22 17:32, Oleksandr wrote:
> >>>
> >>> On 24.06.22 15:59, George Dunlap wrote:
> >>>
> >>> Hello George
> >>>
> >>>>
> >>>>> On 17 Jun 2022, at 17:14, Oleksandr Tyshchenko 
> >>>>> <olekstysh@xxxxxxxxx> wrote:
> >>>>>
> >>>>> From: Oleksandr Tyshchenko <oleksandr_tyshchenko@xxxxxxxx>
> >>>>>
> >>>>> This patch adds basic support for configuring and assisting 
> >>>>> virtio-mmio
> >>>>> based virtio-disk backend (emulator) which is intended to run out of
> >>>>> Qemu and could be run in any domain.
> >>>>> Although the Virtio block device is quite different from traditional
> >>>>> Xen PV block device (vbd) from the toolstack's point of view:
> >>>>> - as the frontend is virtio-blk which is not a Xenbus driver, nothing
> >>>>> written to Xenstore are fetched by the frontend currently ("vdev"
> >>>>> is not passed to the frontend). But this might need to be revised
> >>>>> in future, so frontend data might be written to Xenstore in order to
> >>>>> support hotplugging virtio devices or passing the backend domain id
> >>>>> on arch where the device-tree is not available.
> >>>>> - the ring-ref/event-channel are not used for the backend<->frontend
> >>>>> communication, the proposed IPC for Virtio is IOREQ/DM
> >>>>> it is still a "block device" and ought to be integrated in existing
> >>>>> "disk" handling. So, re-use (and adapt) "disk" parsing/configuration
> >>>>> logic to deal with Virtio devices as well.
> >>>>>
> >>>>> For the immediate purpose and an ability to extend that support for
> >>>>> other use-cases in future (Qemu, virtio-pci, etc) perform the 
> >>>>> following
> >>>>> actions:
> >>>>> - Add new disk backend type (LIBXL_DISK_BACKEND_OTHER) and reflect
> >>>>> that in the configuration
> >>>>> - Introduce new disk "specification" and "transport" fields to struct
> >>>>> libxl_device_disk. Both are written to the Xenstore. The transport
> >>>>> field is only used for the specification "virtio" and it assumes
> >>>>> only "mmio" value for now.
> >>>>> - Introduce new "specification" option with "xen" communication
> >>>>> protocol being default value.
> >>>>> - Add new device kind (LIBXL__DEVICE_KIND_VIRTIO_DISK) as current
> >>>>> one (LIBXL__DEVICE_KIND_VBD) doesn't fit into Virtio disk model
> >>>>>
> >>>>> An example of domain configuration for Virtio disk:
> >>>>> disk = [ 'phy:/dev/mmcblk0p3, xvda1, backendtype=other, 
> >>>>> specification=virtio']
> >>>>>
> >>>>> Nothing has changed for default Xen disk configuration.
> >>>>>
> >>>>> Please note, this patch is not enough for virtio-disk to work
> >>>>> on Xen (Arm), as for every Virtio device (including disk) we need
> >>>>> to allocate Virtio MMIO params (IRQ and memory region) and pass
> >>>>> them to the backend, also update Guest device-tree. The subsequent
> >>>>> patch will add these missing bits. For the current patch,
> >>>>> the default "irq" and "base" are just written to the Xenstore.
> >>>>> This is not an ideal splitting, but this way we avoid breaking
> >>>>> the bisectability.
> >>>>>
> >>>>> Signed-off-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@xxxxxxxx>
> >>>> OK, I am *really* sorry for coming in here at the last minute and 
> >>>> quibbling about things.
> >>>
> >>>
> >>> no problem
> >>>
> >>>
> >>>> But this introduces a public interface which looks really wrong to 
> >>>> me.  I’ve replied to the mail from December where Juergen proposed 
> >>>> the “Other” protocol.
> >>>>
> >>>> Hopefully this will be a simple matter of finding a better name 
> >>>> than “other”. (Or you guys convincing me that “other” is really the 
> >>>> best name for this value; or even Anthony asserting his right as a 
> >>>> maintainer to overrule my objection if he thinks I’m out of line.)
> >>>
> >>>
> >>> I saw your reply to V6 and Juergen's answer. I share Juergen's 
> >>> opinion as well as I understand your concern. I think, this is 
> >>> exactly the situation when finding a perfect name (obvious, short, 
> >>> etc) for the backendtype (in our particular case) is really difficult.
> >>>
> >>> Personally I tend to leave "other", because there is no better 
> >>> alternative from my PoV. Also I might be completely wrong here, but 
> >>> I don't think we will have to extend the "backendtype" for 
> >>> supporting other possible virtio backend implementations in the 
> >>> foreseeable future:
> >>>
> >>> - when Qemu gains the required support we will choose qdisk: 
> >>> backendtype qdisk specification virtio
> >>> - for the possible virtio alternative of the blkback we will choose 
> >>> phy: backendtype phy specification virtio
> >>>
> >>> If there will be a need to keep various implementation, we will be 
> >>> able to describe that by using "transport" (mmio, pci, xenbus, argo, 
> >>> whatever).
> >>> Actually this is why we also introduced "specification" and "transport".
> >>>
> >>> IIRC, there were other (suggested?) names except "other" which are 
> >>> "external" and "daemon". If you think that one of them is better 
> >>> than "other", I will happy to use it.
> >>
> >>
> >> Could we please make a decision on this?
> >>
> >> If "other" is not unambiguous, then maybe we could choose "daemon" to 
> >> describe arbitrary user-level backends, any thought?
> >
> > Unfortunately I didn’t have time to dig into this; I’m just going to 
> > have to withdraw my objection, and let you & Juergen decide what to 
> > call it.
> 
> George, thanks for letting me know. Juergen proposed to use "standalone" 
> for the new backendtype name which is far more specific. I agree with that.
> 
> 
> Anthony, would you be happy with that renaming?

I tried to figure out what backendtype is supposed to mean, how it's
used. I feel it's quite messy at the moment.

Man page xl-disk-configuration says it's a backend implementation to
use. Beside 'phy', which I guess is the kernel or blkback, the two other
point to QEMU ('qdisk') and tapdisk ('tap').

The, backendtype is used throughout libxl to deal with the different
backend implementation, and the value is stored in the xenstore key
"type". From "blkif.h", "type" should be 'file' or 'phy' or 'tap', but
we store 'qdisk' for 'qdisk'... so the "type" note in xenstore is
probably useless for qdisk, but maybe useful for 'phy'? (This "type"
node is only for the backend, so probably useless for a front end.)

Anyway, it seems to me that backendtype should be the name of the
implementation of the backend we want to use. It is just a parameter
to tell libxl how to communicate with the backend. At the moment libxl
uses xenstore to communicates with all backends even if that's not
required, because libxl works this way and it's hard to change. (We
could communicate with QEMU via QMP rather than xenstore for example.)

So I guess either you have a name for your implementation, or something
generic will do. So "standalone" is fine.

(We probably want to document somewhere that this new type would
simply mean "only-relying-on-xenstore-data" like Juergen is putting
it, and isn't blkback or QEMU.)

Thanks,

-- 
Anthony PERARD