Re: [PATCH v1 2/2] virtio-net: virtio_net_flush_tx() check for per-queue reset

[Jason Wang]

On Mon, Jan 30, 2023 at 4:03 PM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote:
>
> On Mon, 30 Jan 2023 15:49:36 +0800, Jason Wang <jasowang@redhat.com> wrote:
> > On Mon, Jan 30, 2023 at 1:32 PM Michael S. Tsirkin <mst@redhat.com> wrote:
> > >
> > > On Mon, Jan 30, 2023 at 10:15:12AM +0800, Xuan Zhuo wrote:
> > > > On Sun, 29 Jan 2023 07:15:47 -0500, "Michael S. Tsirkin" 
> > > > <mst@redhat.com> wrote:
> > > > > On Sun, Jan 29, 2023 at 08:03:42PM +0800, Xuan Zhuo wrote:
> > > > > > On Sun, 29 Jan 2023 06:57:29 -0500, "Michael S. Tsirkin" 
> > > > > > <mst@redhat.com> wrote:
> > > > > > > On Sun, Jan 29, 2023 at 04:23:08PM +0800, Xuan Zhuo wrote:
> > > > > > > > On Sun, 29 Jan 2023 03:12:12 -0500, "Michael S. Tsirkin" 
> > > > > > > > <mst@redhat.com> wrote:
> > > > > > > > > On Sun, Jan 29, 2023 at 03:28:28PM +0800, Xuan Zhuo wrote:
> > > > > > > > > > On Sun, 29 Jan 2023 02:25:43 -0500, "Michael S. Tsirkin" 
> > > > > > > > > > <mst@redhat.com> wrote:
> > > > > > > > > > > On Sun, Jan 29, 2023 at 10:51:50AM +0800, Xuan Zhuo wrote:
> > > > > > > > > > > > Check whether it is per-queue reset state in 
> > > > > > > > > > > > virtio_net_flush_tx().
> > > > > > > > > > > >
> > > > > > > > > > > > Before per-queue reset, we need to recover async tx 
> > > > > > > > > > > > resources. At this
> > > > > > > > > > > > time, virtio_net_flush_tx() is called, but we should 
> > > > > > > > > > > > not try to send
> > > > > > > > > > > > new packets, so virtio_net_flush_tx() should check the 
> > > > > > > > > > > > current
> > > > > > > > > > > > per-queue reset state.
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > What does "at this time" mean here?
> > > > > > > > > > > Do you in fact mean it's called from 
> > > > > > > > > > > flush_or_purge_queued_packets?
> > > > > > > > > >
> > > > > > > > > > Yes
> > > > > > > > > >
> > > > > > > > > > virtio_queue_reset
> > > > > > > > > >   k->queue_reset
> > > > > > > > > >           virtio_net_queue_reset
> > > > > > > > > >                   flush_or_purge_queued_packets
> > > > > > > > > >                           qemu_flush_or_purge_queued_packets
> > > > > > > > > >                                   .....
> > > > > > > > > >                                   (callback) 
> > > > > > > > > > virtio_net_tx_complete
> > > > > > > > > >                                           
> > > > > > > > > > virtio_net_flush_tx <-- here send new packet. We need stop 
> > > > > > > > > > it.
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > Because it is inside the callback, I can't pass information 
> > > > > > > > > > through the stack. I
> > > > > > > > > > originally thought it was a general situation, so I wanted 
> > > > > > > > > > to put it in
> > > > > > > > > > struct VirtQueue.
> > > > > > > > > >
> > > > > > > > > > If it is not very suitable, it may be better to put it in 
> > > > > > > > > > VirtIONetQueue.
> > > > > > > > > >
> > > > > > > > > > Thanks.
> > > > > > > > >
> > > > > > > > > Hmm maybe. Another idea: isn't virtio_net_tx_complete called
> > > > > > > > > with length 0 here? Are there other cases where length is 0?
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > > > What does the call stack look like?
> > > > > > > > > > >
> > > > > > > > > > > If yes introducing a vq state just so virtio_net_flush_tx
> > > > > > > > > > > knows we are in the process of reset would be a bad idea.
> > > > > > > > > > > We want something much more local, ideally on stack even 
> > > > > > > > > > > ...
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > Fixes: 7dc6be52 ("virtio-net: support queue reset")
> > > > > > > > > > > > Fixes: 
> > > > > > > > > > > > https://gitlab.com/qemu-project/qemu/-/issues/1451
> > > > > > > > > > > > Reported-by: Alexander Bulekov <alxndr@bu.edu>
> > > > > > > > > > > > Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
> > > > > > > > > > > > ---
> > > > > > > > > > > >  hw/net/virtio-net.c | 3 ++-
> > > > > > > > > > > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > > > > > > > > > > >
> > > > > > > > > > > > diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
> > > > > > > > > > > > index 3ae909041a..fba6451a50 100644
> > > > > > > > > > > > --- a/hw/net/virtio-net.c
> > > > > > > > > > > > +++ b/hw/net/virtio-net.c
> > > > > > > > > > > > @@ -2627,7 +2627,8 @@ static int32_t 
> > > > > > > > > > > > virtio_net_flush_tx(VirtIONetQueue *q)
> > > > > > > > > > > >      VirtQueueElement *elem;
> > > > > > > > > > > >      int32_t num_packets = 0;
> > > > > > > > > > > >      int queue_index = 
> > > > > > > > > > > > vq2q(virtio_get_queue_index(q->tx_vq));
> > > > > > > > > > > > -    if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
> > > > > > > > > > > > +    if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) ||
> > > > > > > > > > > > +        virtio_queue_reset_state(q->tx_vq)) {
> > > > > > > > >
> > > > > > > > > btw this sounds like you are asking it to reset some state.
> > > > > > > > >
> > > > > > > > > > > >          return num_packets;
> > > > > > > > >
> > > > > > > > > and then
> > > > > > > > >
> > > > > > > > >     ret = virtio_net_flush_tx(q);
> > > > > > > > >     if (ret >= n->tx_burst)
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > will reschedule automatically won't it?
> > > > > > > > >
> > > > > > > > > also why check in virtio_net_flush_tx and not 
> > > > > > > > > virtio_net_tx_complete?
> > > > > > > >
> > > > > > > > virtio_net_flush_tx may been called by timer.
> >
> > We stop timer/bh during device reset, do we need to do the same with vq 
> > reset?
> >
> > > > > > > >
> > > > > > > > Thanks.
> > > > > > >
> > > > > > > timer won't run while flush_or_purge_queued_packets is in 
> > > > > > > progress.
> > > > > >
> > > > > > Is timer not executed during the VMEXIT process? Otherwise, we 
> > > > > > still have to
> > > > > > consider that after the flush_or_purge_queued_packets, this process 
> > > > > > before the
> > > > > > structure is cleared.
> > > > >
> > > > >
> > > > >
> > > > > void virtio_queue_reset(VirtIODevice *vdev, uint32_t queue_index)
> > > > > {
> > > > >     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
> > > > >
> > > > >     if (k->queue_reset) {
> > > > >         k->queue_reset(vdev, queue_index);
> > > > >     }
> > > > >
> > > > >     __virtio_queue_reset(vdev, queue_index);
> > > > > }
> > > > >
> > > > >
> > > > > No timers do not run between  k->queue_reset and __virtio_queue_reset.
> > > > >
> > > > >
> > > > > > Even if it can be processed in virtio_net_tx_complete, is there any 
> > > > > > good way?
> > > > > > This is a callback, it is not convenient to pass the parameters.
> > > > > >
> > > > > > Thanks
> > > > >
> > > > >
> > > > > How about checking that length is 0?
> > > >
> > > >
> > > > I think that check length is not a good way. This modifys the semantics 
> > > > of 0.
> >
> > 0 seems to mean "purge" and
> >
> > > It is
> > > > not friendly to the future maintenance. On the other hand, 
> > > > qemu_net_queue_purge()
> > > > will pass 0, and this function is called by many places.
> >
> > That's exactly what we want actually, when do purge we don't need a flush?
>
> Yes, but I'm not sure. If we stop flush, there will be any other effects.

So we did:

virtio_net_queue_reset():
    nc = qemu_get_subqueue(n->nic, vq2q(queue_index);
    flush_or_purge_queued_packets(nc);
        qemu_flush_or_purge_queued_packets(nc->peer, true); // [1]
            if (qemu_net_queue_flush(nc->incoming_queue)) {
            ....
            } else if (purge) {
                qemu_net_queue_purge(nc->incoming_queue, nc->peer);
                        packet->send_cb()
                            virtio_net_tx_complete()
                                virtio_net_flush_tx()
                                    qemu_sendv_packet_async() // [2]
            }

We try to flush the tap's incoming queue and if we fail we will purge
in [1]. But the sent_cb() tries to send more packets which could be
queued to the tap incoming queue [2]. This breaks the semantic of
qemu_flush_or_purge_queued_packets().

>
> On the other hand, if we use "0" as a judgment condition, do you mean only the
> implementation of the purge in the flush_or_purge_queued_packets()?

It should be all the users of qemu_net_queue_purge(). The rest users
seems all fine:

virtio_net_vhost_status(), if we do flush, it may end up with touching
vring during vhost is running.
filters: all do a flush before.

>
> >
> > > >
> > > > How about we add an api in queue.c to replace the sent_cb callback on 
> > > > queue?
> > > >
> > > > Thanks.
> > > > > OK I guess. Jason?
> >
> > Not sure, anything different from adding a check in
> > virtio_net_tx_complete()? (assuming bh and timer is cancelled or
> > deleted).
>
> We replaced the sent_cb with a function without flush.

I meant it won't be different from adding a

if (virtio_queue_is_rest())

somewhere in virtio_net_tx_complete()?

Thanks

>
> Thanks.
>
>
> >
> > Thanks
> >
> > >
> > > >
> > > > >
> > > > > >
> > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > > > >      }
> > > > > > > > > > > >
> > > > > > > > > > > > --
> > > > > > > > > > > > 2.32.0.3.g01195cf9f
> > > > > > > > > > >
> > > > > > > > >
> > > > > > >
> > > > >
> > >
> >
>