Re: [PATCH v2 04/18] xen/pvcalls: xenbus state handling

From: Stefano Stabellini
Date: Thu Jun 01 2017 - 16:54:34 EST


On Thu, 25 May 2017, Boris Ostrovsky wrote:
> On 05/19/2017 07:22 PM, Stefano Stabellini wrote:
> > Introduce the code to handle xenbus state changes.
> >
> > Implement the probe function for the pvcalls backend. Write the
> > supported versions, max-page-order and function-calls nodes to xenstore,
> > as required by the protocol.
> >
> > Introduce stub functions for disconnecting/connecting to a frontend.
> >
> > Signed-off-by: Stefano Stabellini <stefano@xxxxxxxxxxx>
> > CC: boris.ostrovsky@xxxxxxxxxx
> > CC: jgross@xxxxxxxx
> > ---
> > drivers/xen/pvcalls-back.c | 135 +++++++++++++++++++++++++++++++++++++++++++++
> > 1 file changed, 135 insertions(+)
> >
> > diff --git a/drivers/xen/pvcalls-back.c b/drivers/xen/pvcalls-back.c
> > index 9044cf2..b4da138 100644
> > --- a/drivers/xen/pvcalls-back.c
> > +++ b/drivers/xen/pvcalls-back.c
> > @@ -25,20 +25,155 @@
> > #include <xen/xenbus.h>
> > #include <xen/interface/io/pvcalls.h>
> >
> > +#define PVCALLS_VERSIONS "1"
>
> Shouldn't this be in a header file that will be shared with frontends?

I don't think it should: different backends could implement a different
pvcalls version. There is an handshake mechanism to agree on a version
of the protocol to use. It is probably not a good idea to expose this
specific version in the generic pvcalls header.


> > +#define MAX_RING_ORDER XENBUS_MAX_RING_GRANT_ORDER
> > +
> > struct pvcalls_back_global {
> > struct list_head frontends;
> > struct semaphore frontends_lock;
> > } pvcalls_back_global;
> >
> > +static int backend_connect(struct xenbus_device *dev)
> > +{
> > + return 0;
> > +}
> > +
> > +static int backend_disconnect(struct xenbus_device *dev)
> > +{
> > + return 0;
> > +}
> > +
> > static int pvcalls_back_probe(struct xenbus_device *dev,
> > const struct xenbus_device_id *id)
> > {
> > + int err;
> > +
> > + err = xenbus_printf(XBT_NIL, dev->nodename, "versions", "%s",
> > + PVCALLS_VERSIONS);
> > + if (err) {
> > + pr_warn("%s write out 'version' failed\n", __func__);
> > + return -EINVAL;
>
> err?

OK


> > + }
> > +
> > + err = xenbus_printf(XBT_NIL, dev->nodename, "max-page-order", "%u",
> > + MAX_RING_ORDER);
> > + if (err) {
> > + pr_warn("%s write out 'max-page-order' failed\n", __func__);
> > + return err;
> > + }
> > +
> > + /* "1" means socket, connect, release, bind, listen, accept and poll*/
>
> This comment is obsolete, or at least should refer to
> XENBUS_FUNCTIONS_CALLS (and is missing a space at the end).

OK


> > + err = xenbus_printf(XBT_NIL, dev->nodename, "function-calls",
> > + XENBUS_FUNCTIONS_CALLS);
> > + if (err) {
> > + pr_warn("%s write out 'function-calls' failed\n", __func__);
> > + return err;
> > + }
>
>
> In case of errors we will end up with all previous entries. I think this
> should be done as a transaction which would be aborted in case of an error.

OK


> > +
> > + err = xenbus_switch_state(dev, XenbusStateInitWait);
> > + if (err)
> > + return err;
>
>
> Not sure what to do on an error here (wrt xenstore entries).
> xenbus_switch_state() itself uses transactions. Are we allowed to have
> nested transactions?
>
> OTOH, xenbus_switch_state() never returns an error, at least now. In
> fact, in most cases we ignore return value.

I don't think we can have nested transactions. For simplicity, I'll
ignore errors from xenbus_switch_state.