[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20240130105246-mutt-send-email-mst@kernel.org>
Date: Tue, 30 Jan 2024 10:53:29 -0500
From: "Michael S. Tsirkin" <mst@...hat.com>
To: Daniel Jurgens <danielj@...dia.com>
Cc: "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
"jasowang@...hat.com" <jasowang@...hat.com>,
"xuanzhuo@...ux.alibaba.com" <xuanzhuo@...ux.alibaba.com>,
"virtualization@...ts.linux.dev" <virtualization@...ts.linux.dev>,
"davem@...emloft.net" <davem@...emloft.net>,
"edumazet@...gle.com" <edumazet@...gle.com>,
"kuba@...nel.org" <kuba@...nel.org>,
"abeni@...hat.com" <abeni@...hat.com>,
Parav Pandit <parav@...dia.com>
Subject: Re: [PATCH net-next] virtio_net: Add TX stop and wake counters
On Tue, Jan 30, 2024 at 03:50:29PM +0000, Daniel Jurgens wrote:
> > From: Michael S. Tsirkin <mst@...hat.com>
> > Sent: Tuesday, January 30, 2024 9:42 AM
> > On Tue, Jan 30, 2024 at 03:40:21PM +0000, Daniel Jurgens wrote:
> > > > From: Michael S. Tsirkin <mst@...hat.com>
> > > > Sent: Tuesday, January 30, 2024 8:58 AM
> > > >
> > > > On Tue, Jan 30, 2024 at 08:25:21AM -0600, Daniel Jurgens wrote:
> > > > > Add a tx queue stop and wake counters, they are useful for debugging.
> > > > >
> > > > > $ ethtool -S ens5f2 | grep 'tx_stop\|tx_wake'
> > > > > ...
> > > > > tx_queue_1_tx_stop: 16726
> > > > > tx_queue_1_tx_wake: 16726
> > > > > ...
> > > > > tx_queue_8_tx_stop: 1500110
> > > > > tx_queue_8_tx_wake: 1500110
> > > > >
> > > > > Signed-off-by: Daniel Jurgens <danielj@...dia.com>
> > > > > Reviewed-by: Parav Pandit <parav@...dia.com>
> > > >
> > > > Hmm isn't one always same as the other, except when queue is stopped?
> > > > And when it is stopped you can see that in the status?
> > > > So how is having two useful?
> > >
> > > At idle the counters will be the same, unless a tx_timeout occurs. But
> > under load they can be monitored to see which queues are stopped and get
> > an idea of how long they are stopped.
> >
> > how does it give you the idea of how long they are stopped?
>
> By serially monitoring the counter you can see stops that persist long intervals that are less than the tx_timeout time.
Why don't you monitor queue status directly?
> >
> > > Other net drivers (not all), also have the wake counter.
> >
> > Examples?
>
> [danielj@...mtx-051 upstream]$ ethtool -i ens2f1np1
> driver: mlx5_core
> version: 6.7.0+
> ...
> [danielj@...mtx-051 upstream]$ ethtool -S ens2f1np1 | grep wake
> tx_queue_wake: 0
> tx0_wake: 0
Do they have a stop counter too?
> >
> > > In my opinion it makes the stop counter more useful, at little cost.
> > >
> > > >
> > > >
> > > > > ---
> > > > > drivers/net/virtio_net.c | 26 ++++++++++++++++++++++++--
> > > > > 1 file changed, 24 insertions(+), 2 deletions(-)
> > > > >
> > > > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > > > > index 3cb8aa193884..7e3c31ceaf7e 100644
> > > > > --- a/drivers/net/virtio_net.c
> > > > > +++ b/drivers/net/virtio_net.c
> > > > > @@ -88,6 +88,8 @@ struct virtnet_sq_stats {
> > > > > u64_stats_t xdp_tx_drops;
> > > > > u64_stats_t kicks;
> > > > > u64_stats_t tx_timeouts;
> > > > > + u64_stats_t tx_stop;
> > > > > + u64_stats_t tx_wake;
> > > > > };
> > > > >
> > > > > struct virtnet_rq_stats {
> > > > > @@ -112,6 +114,8 @@ static const struct virtnet_stat_desc
> > > > virtnet_sq_stats_desc[] = {
> > > > > { "xdp_tx_drops", VIRTNET_SQ_STAT(xdp_tx_drops) },
> > > > > { "kicks", VIRTNET_SQ_STAT(kicks) },
> > > > > { "tx_timeouts", VIRTNET_SQ_STAT(tx_timeouts) },
> > > > > + { "tx_stop", VIRTNET_SQ_STAT(tx_stop) },
> > > > > + { "tx_wake", VIRTNET_SQ_STAT(tx_wake) },
> > > > > };
> > > > >
> > > > > static const struct virtnet_stat_desc virtnet_rq_stats_desc[] = {
> > > > > @@
> > > > > -843,6 +847,9 @@ static void check_sq_full_and_disable(struct
> > > > > virtnet_info
> > > > *vi,
> > > > > */
> > > > > if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
> > > > > netif_stop_subqueue(dev, qnum);
> > > > > + u64_stats_update_begin(&sq->stats.syncp);
> > > > > + u64_stats_inc(&sq->stats.tx_stop);
> > > > > + u64_stats_update_end(&sq->stats.syncp);
> > > > > if (use_napi) {
> > > > > if (unlikely(!virtqueue_enable_cb_delayed(sq->vq)))
> > > > > virtqueue_napi_schedule(&sq->napi, sq- vq);
> > @@ -851,6 +858,9
> > > > >@@ static void check_sq_full_and_disable(struct virtnet_info *vi,
> > > > > free_old_xmit_skbs(sq, false);
> > > > > if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
> > > > > netif_start_subqueue(dev, qnum);
> > > > > + u64_stats_update_begin(&sq->stats.syncp);
> > > > > + u64_stats_inc(&sq->stats.tx_wake);
> > > > > + u64_stats_update_end(&sq->stats.syncp);
> > > > > virtqueue_disable_cb(sq->vq);
> > > > > }
> > > > > }
> > > > > @@ -2163,8 +2173,14 @@ static void virtnet_poll_cleantx(struct
> > > > receive_queue *rq)
> > > > > free_old_xmit_skbs(sq, true);
> > > > > } while (unlikely(!virtqueue_enable_cb_delayed(sq->vq)));
> > > > >
> > > > > - if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS)
> > > > > + if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) {
> > > > > + if (netif_tx_queue_stopped(txq)) {
> > > > > + u64_stats_update_begin(&sq->stats.syncp);
> > > > > + u64_stats_inc(&sq->stats.tx_wake);
> > > > > + u64_stats_update_end(&sq->stats.syncp);
> > > > > + }
> > > > > netif_tx_wake_queue(txq);
> > > > > + }
> > > > >
> > > > > __netif_tx_unlock(txq);
> > > > > }
> > > > > @@ -2310,8 +2326,14 @@ static int virtnet_poll_tx(struct
> > > > > napi_struct
> > > > *napi, int budget)
> > > > > virtqueue_disable_cb(sq->vq);
> > > > > free_old_xmit_skbs(sq, true);
> > > > >
> > > > > - if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS)
> > > > > + if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) {
> > > > > + if (netif_tx_queue_stopped(txq)) {
> > > > > + u64_stats_update_begin(&sq->stats.syncp);
> > > > > + u64_stats_inc(&sq->stats.tx_wake);
> > > > > + u64_stats_update_end(&sq->stats.syncp);
> > > > > + }
> > > > > netif_tx_wake_queue(txq);
> > > > > + }
> > > > >
> > > > > opaque = virtqueue_enable_cb_prepare(sq->vq);
> > > > >
> > > > > --
> > > > > 2.42.0
Powered by blists - more mailing lists