[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <1473879623-15382-1-git-send-email-jakub.kicinski@netronome.com>
Date: Wed, 14 Sep 2016 20:00:08 +0100
From: Jakub Kicinski <jakub.kicinski@...ronome.com>
To: netdev@...r.kernel.org
Cc: ast@...nel.org, daniel@...earbox.net, jiri@...nulli.us,
john.fastabend@...il.com, kubakici@...pl,
Jakub Kicinski <jakub.kicinski@...ronome.com>
Subject: [PATCHv3 net-next 00/15] BPF hardware offload (cls_bpf for now)
Hi!
In the last year a lot of progress have been made on offloading
simpler TC classifiers. There is also growing interest in using
BPF for generic high-speed packet processing in the kernel.
It seems beneficial to tie those two trends together and think
about hardware offloads of BPF programs. This patch set presents
such offload to Netronome smart NICs. cls_bpf is extended with
hardware offload capabilities and NFP driver gets a JIT translator
which in presence of capable firmware can be used to offload
the BPF program onto the card.
BPF JIT implementation is not 100% complete (e.g. missing instructions)
but it is functional. Encouragingly it should be possible to
offload most (if not all) advanced BPF features onto the NIC -
including packet modification, maps, tunnel encap/decap etc.
Example of basic tests I used:
__section_cls_entry
int cls_entry(struct __sk_buff *skb)
{
if (load_byte(skb, 0) != 0x0)
return 0;
if (load_byte(skb, 4) != 0x1)
return 0;
skb->mark = 0xcafe;
if (load_byte(skb, 50) != 0xff)
return 0;
return ~0U;
}
Above code can be compiled with Clang and loaded like this:
# ethtool -K p1p1 hw-tc-offload on
# tc qdisc add dev p1p1 ingress
# tc filter add dev p1p1 parent ffff: bpf obj prog.o action drop
This set implements the basic transparent offload, the skip_{sw,hw}
flags and reporting statistics for cls_bpf.
---
I'm posting the BPF offload as non-RFC with changes requested.
I did not carry John's and Jiri's acks on patches 2 and 3
because unlike other classifiers in v3 I'm rejecting unsupported
flags instead of ignoring them.
As suggested by Daniel I prefixed the now-exported verfier
structures with bpf_. I kept it as a separate patch for
clarity.
I removed the dependency on bitfield.h for now by copying
the macros into the driver's main header, I'll drop that copy
once bitfield.h trickles into net-next (it'll be a 12 LOC patch).
I've also revised the metadata format into something I'm more
happy with (patch 12).
Other noteworthy changes:
- dropped the reg ^ reg -> CONST_IMM(0) interpretation;
- made the 64bit load only interpreted in non-verifier mode;
- create&use per-insn storage instead of reserved fields.
Please see individual patches for detailed changelog.
Jakub Kicinski (15):
net: cls_bpf: add hardware offload
net: cls_bpf: limit hardware offload by software-only flag
net: cls_bpf: add support for marking filters as hardware-only
bpf: don't (ab)use instructions to store state
bpf: enable non-core use of the verfier
bpf: prefix structures in bpf_parser.h with bpf_
bpf: recognize 64bit immediate loads as consts
nfp: add BPF to NFP code translator
nfp: bpf: add hardware bpf offload
net: cls_bpf: allow offloaded filters to update stats
net: bpf: allow offloaded filters to update stats
nfp: bpf: add packet marking support
net: act_mirred: allow statistic updates from offloaded actions
nfp: bpf: add support for legacy redirect action
nfp: bpf: add offload of TC direct action mode
drivers/net/ethernet/netronome/nfp/Makefile | 7 +
drivers/net/ethernet/netronome/nfp/nfp_asm.h | 233 +++
drivers/net/ethernet/netronome/nfp/nfp_bpf.h | 223 +++
drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c | 1816 ++++++++++++++++++++
.../net/ethernet/netronome/nfp/nfp_bpf_verifier.c | 160 ++
drivers/net/ethernet/netronome/nfp/nfp_net.h | 47 +-
.../net/ethernet/netronome/nfp/nfp_net_common.c | 134 +-
drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h | 51 +-
.../net/ethernet/netronome/nfp/nfp_net_ethtool.c | 12 +
.../net/ethernet/netronome/nfp/nfp_net_offload.c | 291 ++++
.../net/ethernet/netronome/nfp/nfp_netvf_main.c | 2 +-
include/linux/bpf_parser.h | 89 +
include/linux/netdevice.h | 2 +
include/net/pkt_cls.h | 16 +
include/uapi/linux/pkt_cls.h | 1 +
kernel/bpf/verifier.c | 384 +++--
net/sched/act_mirred.c | 8 +
net/sched/cls_bpf.c | 117 +-
18 files changed, 3387 insertions(+), 206 deletions(-)
create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_asm.h
create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_bpf.h
create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_bpf_jit.c
create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_bpf_verifier.c
create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_net_offload.c
create mode 100644 include/linux/bpf_parser.h
--
1.9.1
Powered by blists - more mailing lists