|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [PATCH 2/4] hw/xen: update Xen PV NIC to XenDevice model
On Tue, 2023-10-24 at 15:47 +0100, Paul Durrant wrote:
> On 17/10/2023 19:25, David Woodhouse wrote:
> > +
> > +#define xen_pv_printf(a, n, ...) qemu_printf(__VA_ARGS__)
>
> Why define this...
In the first place, just to make it build in the short term. Then I
forgot to clean it up before posting. In my tree this is all tracing
now.
>
> > @@ -232,7 +258,7 @@ static ssize_t net_rx_packet(NetClientState *nc, const
> > uint8_t *buf, size_t size
> > RING_IDX rc, rp;
> > void *page;
> >
> > - if (netdev->xendev.be_state != XenbusStateConnected) {
> > + if (netdev->rx_ring.sring == NULL) {
>
> Why not a straight swap for xen_device_backend_get_state()? Hard to see
> whether there any hidden side effects of this change otherwise.
Could do, but what I *actually* cared about when looking at that check,
was whether the ring pointer was NULL. So I checked that explicitly.
It should be identical.
> > +static void xen_netdev_frontend_changed(XenDevice *xendev,
> > + enum xenbus_state frontend_state,
> > + Error **errp)
> > {
> > - struct XenNetDev *netdev = container_of(xendev, struct XenNetDev,
> > xendev);
> > - net_tx_packets(netdev);
> > - qemu_flush_queued_packets(qemu_get_queue(netdev->nic));
> > + ERRP_GUARD();
> > + enum xenbus_state backend_state = xen_device_backend_get_state(xendev);
> > +
> > + trace_xen_netdev_frontend_changed(xendev->name, frontend_state);
> > +
> > + switch (frontend_state) {
> > + case XenbusStateInitialised:
>
> I don't think that's really a valid state for a network frontend. Linux
> netback just ignores it.
Must we? I was thinking of making the ->frontend_changed() methods
optional and allowing backends to just provide ->connect() and
->disconnect() methods instead if they wanted to. Because we have three
identical ->frontend_changed() methods now...
Attachment:
smime.p7s
|
![]() |
Lists.xenproject.org is hosted with RackSpace, monitoring our |