Re: Virtio-disk updates for latest Xen ?

[Juergen Gross <jgross@xxxxxxxx>]

On 07.12.22 01:21, Viresh Kumar wrote:
> +list
>
> On 06-12-22, 13:40, Oleksandr Tyshchenko wrote:
> > On Tue, Dec 6, 2022 at 1:15 PM Viresh Kumar <viresh.kumar@xxxxxxxxxx> wrote:
> > > Hi Oleksandr,
> > Hello Viresh
> > > I found that my rust counterpart [1] of virtio-disk repository broke
> > > with this commit:
> > >
> > > commit 3a96013a3e17 ("tools/xenstore: reduce number of watch events")
> > >
> > > And so was wondering if you have already fixed that for
> > > virtio-disk or not. And I can just copy what you have changed :)
> > Interesting, I am not aware of that issue. I haven't tried virtio-disk with
> > recent (latest) Xen, I was mostly playing with Qemu backends. Thanks for
> > letting me know, I will re-check once I have access to my hardware again.
> Hmm, moved a little closer :)
>
> We look at the watch events here [2] one by one and wait for the
> directory one to come ("backend/disk" in your case I think, it is
> "backend/virtio" with my latest patches). This is the second directory
> event which used to come after the directory is finalized.
>
> Earlier the events were read for following paths:
>
> "backend/virtio"
> "backend/virtio/1/0"
> "backend/virtio/1"
> "backend/virtio"
>
> and on the last event, we used to go ahead and look at the directory
> to find the FE domid.
>
> With the new changes, the events I am getting are:
>
> [src/xs.rs:164] &name = "backend/virtio"
> [src/xs.rs:164] &name = "backend/virtio/1/0"
> [src/xs.rs:164] &name = "backend/virtio/1/0/frontend"
> [src/xs.rs:164] &name = "backend/virtio/1/0/frontend-id"
> [src/xs.rs:164] &name = "backend/virtio/1/0/online"
> [src/xs.rs:164] &name = "backend/virtio/1/0/state"
> [src/xs.rs:164] &name = "backend/virtio/1/0/irq"
> [src/xs.rs:164] &name = "backend/virtio/1/0/base"
> [src/xs.rs:164] &name = "backend/virtio/1/0/type"
> [src/xs.rs:164] &name = "backend/virtio/1/0/transport"
>
> Here the second event for "backend/virtio" never comes and so we fail.
Yes, this is by design.

Note that all Xenstore entries for a device are written via a transaction,
so there should be no way you can see any of the backend nodes missing when
receiving a watch for "backend/virtio/1/0" due to creation of the device.

In future we are even planning to support setting up Xenstore watches for
a limited path depth, resulting in only the "backend/virtio", "backend/virtio/1"
and "backend/virtio/1/0" watch events being sent (this example is for the
kernel, which will use the last event to extract all the needed information
from the reported path).


Juergen

Attachment: OpenPGP_0xB0DE9DD628BF132F.asc
Description: OpenPGP public key

Attachment: OpenPGP_signature
Description: OpenPGP digital signature