lkml.org 
[lkml]   [2020]   [Jun]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] virtio_net: Unregister and re-register xdp_rxq across freeze/restore
On Sun, Jun 07, 2020 at 09:23:03AM -0400, Michael S. Tsirkin wrote:
> On Fri, Jun 05, 2020 at 02:46:24PM -0700, Sean Christopherson wrote:
> > @@ -1480,17 +1495,10 @@ static int virtnet_open(struct net_device *dev)
> > if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
> > schedule_delayed_work(&vi->refill, 0);
> >
> > - err = xdp_rxq_info_reg(&vi->rq[i].xdp_rxq, dev, i);
> > + err = virtnet_reg_xdp(&vi->rq[i].xdp_rxq, dev, i);
> > if (err < 0)
> > return err;
> >
> > - err = xdp_rxq_info_reg_mem_model(&vi->rq[i].xdp_rxq,
> > - MEM_TYPE_PAGE_SHARED, NULL);
> > - if (err < 0) {
> > - xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq);
> > - return err;
> > - }
> > -
> > virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
> > virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi);
> > }
> > @@ -2306,6 +2314,7 @@ static void virtnet_freeze_down(struct virtio_device *vdev)
> >
> > if (netif_running(vi->dev)) {
> > for (i = 0; i < vi->max_queue_pairs; i++) {
> > + xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq);
> > napi_disable(&vi->rq[i].napi);
> > virtnet_napi_tx_disable(&vi->sq[i].napi);
>
> I suspect the right thing to do is to first disable all NAPI,
> then play with XDP. Generally cleanup in the reverse order
> of init is a good idea.

Hmm, I was simply following virtnet_close(). Actually, the entire loop
could be factored out into a separate helper. Perhaps do that as part of
the fix, and then invert the ordering in a separate patch?

> > }
> > @@ -2313,6 +2322,8 @@ static void virtnet_freeze_down(struct virtio_device *vdev)
> > }
> >
> > static int init_vqs(struct virtnet_info *vi);
> > +static void virtnet_del_vqs(struct virtnet_info *vi);
> > +static void free_receive_page_frags(struct virtnet_info *vi);
>
> I'd really rather we reordered code so forward decls are not necessary.

Yeah, no argument from me. Would you prefer the reordering in a separate
patch on top, e.g. to simplify potential backporting?

> > static int virtnet_restore_up(struct virtio_device *vdev)
> > {
> > @@ -2331,6 +2342,10 @@ static int virtnet_restore_up(struct virtio_device *vdev)
> > schedule_delayed_work(&vi->refill, 0);
> >
> > for (i = 0; i < vi->max_queue_pairs; i++) {
> > + err = virtnet_reg_xdp(&vi->rq[i].xdp_rxq, vi->dev, i);
> > + if (err)
> > + goto free_vqs;
> > +
> > virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
> > virtnet_napi_tx_enable(vi, vi->sq[i].vq,
> > &vi->sq[i].napi);
> > @@ -2340,6 +2355,12 @@ static int virtnet_restore_up(struct virtio_device *vdev)
> > netif_tx_lock_bh(vi->dev);
> > netif_device_attach(vi->dev);
> > netif_tx_unlock_bh(vi->dev);
> > + return 0;
> > +
> > +free_vqs:
> > + cancel_delayed_work_sync(&vi->refill);
> > + free_receive_page_frags(vi);
> > + virtnet_del_vqs(vi);
>
>
> I am not sure this is safe to do after device-ready.
>
> Can reg xdp happen before device ready?

From a code perspective, I don't see anything that will explode, but I have
no idea if that's correct/sane behavior.

FWIW, the xdp error handling in virtnet_open() also looks bizarre to me,
e.g. bails in the middle of a loop without doing any cleanup. I assume
virtnet_close() wouldn't called if open failed? But I can't determine
whether or not that holds true based on code inspection, there are too many
call sites that lead to open and close.

\
 
 \ /
  Last update: 2020-06-08 23:10    [W:0.071 / U:1.364 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site