minutes: IO Visor TSC and Dev Members Call


Brenden Blanco <bblanco@...>
 

Thank you all for joining today. Please find my notes from the meeting below.

=== Works in progress ===

Tracing:
* add hw cache events reading support in PERF_EVENT_ARRAY (brenden)
* working on nmi support for bpf tracing (alexei)
 - to attach programs when counters are firing (overflow wakeup_xxx)

XDP:
* add improved rcu protection to mlx4 rx path (brenden)
* prefetch tuneable in xdp (or elsewhere) netlink
* mlx4 refactoring of TX rings
  (mlx5 is TBD - prototyped but not concrete yet)
* ixgbe support starting (intel)
* e1000 would be very nice to have (william tu?)
 - as template for others to follow
 - needs recycling logic
 - needs to take lock - only 1 tx ring assumed currently

TC:
* cls_bpf needs test harness (daniel)
* defrag (ipv4) support in cls_bpf (daniel)
 - redirects to upper stack, and last packet in frag goes to bpf again

LSM:
* Next drop possible near end of month
* use case: want to be able to write to kernel memory

current_in_cgroup (sargun):
* should be able to land with a few more alterations
* todo: bcc tools support for filtering by cgroup once landed (brenden)

ceth (huawei):
 - trying to integrate xdp hook into i40e driver
 - mlx4 integration with ceth+xdp

=== Discussion of upcoming events ===

LinuxCon Aug 22-24
* Thomas Graf to present Cilium

Aug 23 Brendan Gregg @fb
* email alexei for invites if interested

Facebook @Scale Aug 31
* email alexei for invites if interested

Tracing iovisor meetup in Sept @fb, details TBD
 - Similar proposed format to XDP meetup

Netdev in Oct, Tokyo (see netdev mailer for details)
Topics being proposed:
* xdp talk from Tom, Yunsong
* cls_bpf from Daniel
* ktls from facebook
* ILA from prem @ barefoot
* netronome: offloading

Plumbers in Nov @ Santa Fe
* Needs tracing topics!

USENIX LISA16 Dec 4-9
* sasha + brendan = perf+bpf tutorial

Linuxcon Australia jan
Brendan = bpf?

Brendan reminds us all to attend more conferences, write more blogs, etc.!
Spread the word :)

Idea for blog post on power of BPF in networking:
tc+bpf vs iptables filtering for ingress filtering of a source address

Late update from William Tu on OVS->BPF:
* goal: how to offload parsing+lookup+actions with bpf
  - starting with parsing
  - lookup
  - ovs installs entries into bpf map
  - ovs actions:
   - redirect - ok
   - tunnel - set tunnel key - ok
   - upcall - difficult, using event output
    - how to push packet down to program
     - daniel: AF_PACKET?
   - conntrack helper - doable
   - daniel says was able to implement connection tracking directly with bpf maps


=== Attendees ===
Jakub Kicinski
Alban Crequy
Brenden Blanco
Daniel Borkmann
Kyle Laracey
Nicolaas Viljoen
Sargun Dhillon
Prem Jonnalagadda
William Tu
Alexei Starovoitov
Brendan Gregg
Yan Chen
Wenzhe