On Tue, 07/07 12:19, Michael S. Tsirkin wrote: > On Tue, Jul 07, 2015 at 05:09:09PM +0800, Fam Zheng wrote: > > On Tue, 07/07 11:13, Michael S. Tsirkin wrote: > > > On Tue, Jul 07, 2015 at 09:21:07AM +0800, Fam Zheng wrote: > > > > Since commit 6e99c63 "net/socket: Drop net_socket_can_send" and friends, > > > > net queues need to be explicitly flushed after qemu_can_send_packet() > > > > returns false, because the netdev side will disable the polling of fd. > > > > > > > > This fixes the case of "cont" after "stop" (or migration). > > > > > > > > Signed-off-by: Fam Zheng <f...@redhat.com> > > > > > > Note virtio has its own handler which must be used to > > > flush packets - this one might run too early or too late. > > > > Which handler do you mean? I don't think virtio-net handles resume now. (If > > it > > does, we probably should drop it together with this change, since it's > > needed > > by as all NICs.) > > > > Fam > > virtio_vmstate_change > > It's all far from trivial. I suspect these whack-a-mole approach > spreading purge here and there will only create more bugs. > > Why would we ever need to process network packets when > VM is not running? I don't see any point to it. > How about we simply stop the job processing network on > vm stop and restart on vm start?
I suppose it is too much for 2.4. I think this approach, adding qemu_flush_queued_packets(), is consistent with its existing usage (when a device is becoming active from inactive), like in e1000_write_config. How about applying this and let's work on "stopping tap when VM not running" for 2.5? Fam
