| Message ID | 1512620115-7569-1-git-send-email-makita.toshiaki@lab.ntt.co.jp |
|---|---|
| State | Accepted, archived |
| Delegated to: | David Miller |
| Headers | show |
| Series | [net-next] virtio_net: Disable interrupts if napi_complete_done rescheduled napi | expand |
On Thu, Dec 07, 2017 at 01:15:15PM +0900, Toshiaki Makita wrote: > Since commit 39e6c8208d7b ("net: solve a NAPI race") napi has been able > to be rescheduled within napi_complete_done() even in non-busypoll case, > but virtnet_poll() always enabled interrupts before complete, and when > napi was rescheduled within napi_complete_done() it did not disable > interrupts. > This caused more interrupts when event idx is disabled. > > According to commit cbdadbbf0c79 ("virtio_net: fix race in RX VQ > processing") we cannot place virtqueue_enable_cb_prepare() after > NAPI_STATE_SCHED is cleared, so disable interrupts again if > napi_complete_done() returned false. > > Tested with vhost-user of OVS 2.7 on host, which does not have the event > idx feature. > > * Before patch: > > $ netperf -t UDP_STREAM -H 192.168.150.253 -l 60 -- -m 1472 > MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.150.253 () port 0 AF_INET > Socket Message Elapsed Messages > Size Size Time Okay Errors Throughput > bytes bytes secs # # 10^6bits/sec > > 212992 1472 60.00 32763206 0 6430.32 > 212992 60.00 23384299 4589.56 > > Interrupts on guest: 9872369 > Packets/interrupt: 2.37 > > * After patch > > $ netperf -t UDP_STREAM -H 192.168.150.253 -l 60 -- -m 1472 > MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.150.253 () port 0 AF_INET > Socket Message Elapsed Messages > Size Size Time Okay Errors Throughput > bytes bytes secs # # 10^6bits/sec > > 212992 1472 60.00 32794646 0 6436.49 > 212992 60.00 32793501 6436.27 > > Interrupts on guest: 4941299 > Packets/interrupt: 6.64 > > Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp> Acked-by: Michael S. Tsirkin <mst@redhat.com> it might make sense in net and not -next since tx napi regressed performance in some configs, this might bring it back at least partially. Jason - what do you think? > --- > drivers/net/virtio_net.c | 9 ++++++--- > 1 file changed, 6 insertions(+), 3 deletions(-) > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c > index 19a985e..c0db48d 100644 > --- a/drivers/net/virtio_net.c > +++ b/drivers/net/virtio_net.c > @@ -261,9 +261,12 @@ static void virtqueue_napi_complete(struct napi_struct *napi, > int opaque; > > opaque = virtqueue_enable_cb_prepare(vq); > - if (napi_complete_done(napi, processed) && > - unlikely(virtqueue_poll(vq, opaque))) > - virtqueue_napi_schedule(napi, vq); > + if (napi_complete_done(napi, processed)) { > + if (unlikely(virtqueue_poll(vq, opaque))) > + virtqueue_napi_schedule(napi, vq); > + } else { > + virtqueue_disable_cb(vq); > + } > } > > static void skb_xmit_done(struct virtqueue *vq) > -- > 1.8.3.1 >
On 2017年12月07日 13:09, Michael S. Tsirkin wrote: > On Thu, Dec 07, 2017 at 01:15:15PM +0900, Toshiaki Makita wrote: >> Since commit 39e6c8208d7b ("net: solve a NAPI race") napi has been able >> to be rescheduled within napi_complete_done() even in non-busypoll case, >> but virtnet_poll() always enabled interrupts before complete, and when >> napi was rescheduled within napi_complete_done() it did not disable >> interrupts. >> This caused more interrupts when event idx is disabled. >> >> According to commit cbdadbbf0c79 ("virtio_net: fix race in RX VQ >> processing") we cannot place virtqueue_enable_cb_prepare() after >> NAPI_STATE_SCHED is cleared, so disable interrupts again if >> napi_complete_done() returned false. >> >> Tested with vhost-user of OVS 2.7 on host, which does not have the event >> idx feature. >> >> * Before patch: >> >> $ netperf -t UDP_STREAM -H 192.168.150.253 -l 60 -- -m 1472 >> MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.150.253 () port 0 AF_INET >> Socket Message Elapsed Messages >> Size Size Time Okay Errors Throughput >> bytes bytes secs # # 10^6bits/sec >> >> 212992 1472 60.00 32763206 0 6430.32 >> 212992 60.00 23384299 4589.56 >> >> Interrupts on guest: 9872369 >> Packets/interrupt: 2.37 >> >> * After patch >> >> $ netperf -t UDP_STREAM -H 192.168.150.253 -l 60 -- -m 1472 >> MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.150.253 () port 0 AF_INET >> Socket Message Elapsed Messages >> Size Size Time Okay Errors Throughput >> bytes bytes secs # # 10^6bits/sec >> >> 212992 1472 60.00 32794646 0 6436.49 >> 212992 60.00 32793501 6436.27 >> >> Interrupts on guest: 4941299 >> Packets/interrupt: 6.64 >> >> Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp> > Acked-by: Michael S. Tsirkin <mst@redhat.com> > > it might make sense in net and not -next since tx napi regressed performance > in some configs, this might bring it back at least partially. > Jason - what do you think? No sure, the regression I saw was tested with event idx on. And virtqueue_disable_cb() does almost nothing for event idx (or even a little bit slower). The patch looks good. Acked-by: Jason Wang <jasowang@redhat.com> > >> --- >> drivers/net/virtio_net.c | 9 ++++++--- >> 1 file changed, 6 insertions(+), 3 deletions(-) >> >> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c >> index 19a985e..c0db48d 100644 >> --- a/drivers/net/virtio_net.c >> +++ b/drivers/net/virtio_net.c >> @@ -261,9 +261,12 @@ static void virtqueue_napi_complete(struct napi_struct *napi, >> int opaque; >> >> opaque = virtqueue_enable_cb_prepare(vq); >> - if (napi_complete_done(napi, processed) && >> - unlikely(virtqueue_poll(vq, opaque))) >> - virtqueue_napi_schedule(napi, vq); >> + if (napi_complete_done(napi, processed)) { >> + if (unlikely(virtqueue_poll(vq, opaque))) >> + virtqueue_napi_schedule(napi, vq); >> + } else { >> + virtqueue_disable_cb(vq); >> + } >> } >> >> static void skb_xmit_done(struct virtqueue *vq) >> -- >> 1.8.3.1 >>
From: Jason Wang <jasowang@redhat.com> Date: Thu, 7 Dec 2017 15:08:45 +0800 > > > On 2017年12月07日 13:09, Michael S. Tsirkin wrote: >> On Thu, Dec 07, 2017 at 01:15:15PM +0900, Toshiaki Makita wrote: >>> Since commit 39e6c8208d7b ("net: solve a NAPI race") napi has been >>> able >>> to be rescheduled within napi_complete_done() even in non-busypoll >>> case, >>> but virtnet_poll() always enabled interrupts before complete, and when >>> napi was rescheduled within napi_complete_done() it did not disable >>> interrupts. >>> This caused more interrupts when event idx is disabled. >>> >>> According to commit cbdadbbf0c79 ("virtio_net: fix race in RX VQ >>> processing") we cannot place virtqueue_enable_cb_prepare() after >>> NAPI_STATE_SCHED is cleared, so disable interrupts again if >>> napi_complete_done() returned false. >>> >>> Tested with vhost-user of OVS 2.7 on host, which does not have the >>> event >>> idx feature. ... >>> Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp> >> Acked-by: Michael S. Tsirkin <mst@redhat.com> >> >> it might make sense in net and not -next since tx napi regressed >> performance >> in some configs, this might bring it back at least partially. >> Jason - what do you think? > > No sure, the regression I saw was tested with event idx on. And > virtqueue_disable_cb() does almost nothing for event idx (or even a > little bit slower). > > The patch looks good. > > Acked-by: Jason Wang <jasowang@redhat.com> I'm going to put this into net-next for now.
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c index 19a985e..c0db48d 100644 --- a/drivers/net/virtio_net.c +++ b/drivers/net/virtio_net.c @@ -261,9 +261,12 @@ static void virtqueue_napi_complete(struct napi_struct *napi, int opaque; opaque = virtqueue_enable_cb_prepare(vq); - if (napi_complete_done(napi, processed) && - unlikely(virtqueue_poll(vq, opaque))) - virtqueue_napi_schedule(napi, vq); + if (napi_complete_done(napi, processed)) { + if (unlikely(virtqueue_poll(vq, opaque))) + virtqueue_napi_schedule(napi, vq); + } else { + virtqueue_disable_cb(vq); + } } static void skb_xmit_done(struct virtqueue *vq)
Since commit 39e6c8208d7b ("net: solve a NAPI race") napi has been able to be rescheduled within napi_complete_done() even in non-busypoll case, but virtnet_poll() always enabled interrupts before complete, and when napi was rescheduled within napi_complete_done() it did not disable interrupts. This caused more interrupts when event idx is disabled. According to commit cbdadbbf0c79 ("virtio_net: fix race in RX VQ processing") we cannot place virtqueue_enable_cb_prepare() after NAPI_STATE_SCHED is cleared, so disable interrupts again if napi_complete_done() returned false. Tested with vhost-user of OVS 2.7 on host, which does not have the event idx feature. * Before patch: $ netperf -t UDP_STREAM -H 192.168.150.253 -l 60 -- -m 1472 MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.150.253 () port 0 AF_INET Socket Message Elapsed Messages Size Size Time Okay Errors Throughput bytes bytes secs # # 10^6bits/sec 212992 1472 60.00 32763206 0 6430.32 212992 60.00 23384299 4589.56 Interrupts on guest: 9872369 Packets/interrupt: 2.37 * After patch $ netperf -t UDP_STREAM -H 192.168.150.253 -l 60 -- -m 1472 MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.150.253 () port 0 AF_INET Socket Message Elapsed Messages Size Size Time Okay Errors Throughput bytes bytes secs # # 10^6bits/sec 212992 1472 60.00 32794646 0 6436.49 212992 60.00 32793501 6436.27 Interrupts on guest: 4941299 Packets/interrupt: 6.64 Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp> --- drivers/net/virtio_net.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-)