Re: [PATCH] openvswitch: Orphan frags before sending to userspace via Netlink to avoid guest stall
From: Pravin Shelar <hidden>
Date: 2014-03-07 04:52:20
Also in:
kvm, lkml
From: Pravin Shelar <hidden>
Date: 2014-03-07 04:52:20
Also in:
kvm, lkml
On Thu, Mar 6, 2014 at 9:09 AM, Zoltan Kiss [off-list ref] wrote:
Do you have any feedback on this? I'm also adding KVM list as they might be interested in this. Zoli On 28/02/14 19:16, Zoltan Kiss wrote:quoted
The kernel datapath now switched to zerocopy Netlink messages, but that also means that the pages on frags array are sent straight to userspace. If those pages came outside the kernel, we have to swap them out with local copies. Signed-off-by: Zoltan Kiss <redacted>
I do not think this is required, netlink zero copy only maps pre-allocated buffers to user-space. But I found bug in datapath user-space queue code. I am not sure how this can work with skb fragments and MMAP-netlink socket. Here is what happens, OVS allocates netlink skb and adds fragments to skb using skb_zero_copy(), then calls genlmsg_unicast(). But if netlink sock is mmped then netlink-send queues netlink allocated skb->head (linear data of skb) and ignore skb frags. Currently this is not problem with OVS vswitchd since it does not use netlink MMAP sockets. But if vswitchd stats using MMAP-netlink socket, it can break it.