| Message ID | cover.1609837120.git.xuanzhuo@linux.alibaba.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | virtio-net support xdp socket zero copy xmit | expand |
On 2021/1/5 下午5:11, Xuan Zhuo wrote: > The first patch made some adjustments to xsk. Thanks a lot for the work. It's rather interesting. > > The second patch itself can be used as an independent patch to solve the problem > that XDP may fail to load when the number of queues is insufficient. It would be better to send this as a separated patch. Several people asked for this before. > > The third to last patch implements support for xsk in virtio-net. > > A practical problem with virtio is that tx interrupts are not very reliable. > There will always be some missing or delayed tx interrupts. So I specially added > a point timer to solve this problem. Of course, considering performance issues, > The timer only triggers when the ring of the network card is full. This is sub-optimal. We need figure out the root cause. We don't meet such issue before. Several questions: - is tx interrupt enabled? - can you still see the issue if you disable event index? - what's backend did you use? qemu or vhost(user)? > > Regarding the issue of virtio-net supporting xsk's zero copy rx, I am also > developing it, but I found that the modification may be relatively large, so I > consider this patch set to be separated from the code related to xsk zero copy > rx. That's fine, but a question here. How is the multieuque being handled here. I'm asking since there's no programmable filters/directors support in virtio spec now. Thanks > > Xuan Zhuo (5): > xsk: support get page for drv > virtio-net: support XDP_TX when not more queues > virtio-net, xsk: distinguish XDP_TX and XSK XMIT ctx > xsk, virtio-net: prepare for support xsk > virtio-net, xsk: virtio-net support xsk zero copy tx > > drivers/net/virtio_net.c | 643 +++++++++++++++++++++++++++++++++++++++----- > include/linux/netdevice.h | 1 + > include/net/xdp_sock_drv.h | 10 + > include/net/xsk_buff_pool.h | 1 + > net/xdp/xsk_buff_pool.c | 10 +- > 5 files changed, 597 insertions(+), 68 deletions(-) > > -- > 1.8.3.1 >
On Tue, Jan 05, 2021 at 05:11:38PM +0800, Xuan Zhuo wrote: > The first patch made some adjustments to xsk. > > The second patch itself can be used as an independent patch to solve the problem > that XDP may fail to load when the number of queues is insufficient. > > The third to last patch implements support for xsk in virtio-net. > > A practical problem with virtio is that tx interrupts are not very reliable. > There will always be some missing or delayed tx interrupts. Would appreciate a bit more data on this one. Is this a host bug? Device bug? Can we limit the work around somehow? > So I specially added > a point timer to solve this problem. Of course, considering performance issues, > The timer only triggers when the ring of the network card is full. > > Regarding the issue of virtio-net supporting xsk's zero copy rx, I am also > developing it, but I found that the modification may be relatively large, so I > consider this patch set to be separated from the code related to xsk zero copy > rx. > > Xuan Zhuo (5): > xsk: support get page for drv > virtio-net: support XDP_TX when not more queues > virtio-net, xsk: distinguish XDP_TX and XSK XMIT ctx > xsk, virtio-net: prepare for support xsk > virtio-net, xsk: virtio-net support xsk zero copy tx > > drivers/net/virtio_net.c | 643 +++++++++++++++++++++++++++++++++++++++----- > include/linux/netdevice.h | 1 + > include/net/xdp_sock_drv.h | 10 + > include/net/xsk_buff_pool.h | 1 + > net/xdp/xsk_buff_pool.c | 10 +- > 5 files changed, 597 insertions(+), 68 deletions(-) > > -- > 1.8.3.1
XDP socket is an excellent by pass kernel network transmission framework. The zero copy feature of xsk (XDP socket) needs to be supported by the driver. The performance of zero copy is very good. mlx5 and intel ixgbe already support this feature, This patch set allows virtio-net to support xsk's zerocopy xmit feature. And xsk's zerocopy rx has made major changes to virtio-net, and I hope to submit it after this patch set are received. Compared with other drivers, virtio-net does not directly obtain the dma address, so I first obtain the xsk page, and then pass the page to virtio. When recycling the sent packets, we have to distinguish between skb and xdp. Now we have to distinguish between skb, xdp, xsk. So the second patch solves this problem first. The last four patches are used to support xsk zerocopy in virtio-net: 1. support xsk enable/disable 2. realize the function of xsk packet sending 3. implement xsk wakeup callback 4. set xsk completed when packet sent done ---------------- Performance Testing ------------ The udp package tool implemented by the interface of xsk vs sockperf(kernel udp) for performance testing: xsk zero copy in virtio-net: CPU PPS MSGSIZE 28.7% 3833857 64 38.5% 3689491 512 38.9% 2787096 1456 xsk without zero copy in virtio-net: CPU PPS MSGSIZE 100% 1916747 64 100% 1775988 512 100% 1440054 1456 sockperf: CPU PPS MSGSIZE 100% 713274 64 100% 701024 512 100% 695832 1456 Xuan Zhuo (7): xsk: support get page for drv virtio-net, xsk: distinguish XDP_TX and XSK XMIT ctx xsk, virtio-net: prepare for support xsk zerocopy xmit virtio-net, xsk: support xsk enable/disable virtio-net, xsk: realize the function of xsk packet sending virtio-net, xsk: implement xsk wakeup callback virtio-net, xsk: set xsk completed when packet sent done drivers/net/virtio_net.c | 559 +++++++++++++++++++++++++++++++++++++++----- include/linux/netdevice.h | 1 + include/net/xdp_sock_drv.h | 10 + include/net/xsk_buff_pool.h | 1 + net/xdp/xsk_buff_pool.c | 10 +- 5 files changed, 523 insertions(+), 58 deletions(-) -- 1.8.3.1