| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <20170711204405.2f6b033c@redhat.com>
Date: Tue, 11 Jul 2017 20:44:05 +0200
From: Jesper Dangaard Brouer <brouer@...hat.com>
To: John Fastabend <john.fastabend@...il.com>
Cc: David Miller <davem@...emloft.net>, netdev@...r.kernel.org,
andy@...yhouse.net, daniel@...earbox.net, ast@...com,
alexander.duyck@...il.com, bjorn.topel@...el.com,
jakub.kicinski@...ronome.com, ecree@...arflare.com,
sgoutham@...ium.com, Yuval.Mintz@...ium.com, saeedm@...lanox.com,
brouer@...hat.com
Subject: Re: [RFC PATCH 00/12] Implement XDP bpf_redirect vairants
On Tue, 11 Jul 2017 20:01:36 +0200
Jesper Dangaard Brouer <brouer@...hat.com> wrote:
> On Tue, 11 Jul 2017 10:48:29 -0700
> John Fastabend <john.fastabend@...il.com> wrote:
>
> > On 07/11/2017 08:36 AM, Jesper Dangaard Brouer wrote:
> > > On Sat, 8 Jul 2017 21:06:17 +0200
> > > Jesper Dangaard Brouer <brouer@...hat.com> wrote:
> > >
> > >> My plan is to test this latest patchset again, Monday and Tuesday.
> > >> I'll try to assess stability and provide some performance numbers.
> > >
> > > Performance numbers:
> > >
> > > 14378479 pkt/s = XDP_DROP without touching memory
> > > 9222401 pkt/s = xdp1: XDP_DROP with reading packet data
> > > 6344472 pkt/s = xdp2: XDP_TX with swap mac (writes into pkt)
> > > 4595574 pkt/s = xdp_redirect: XDP_REDIRECT with swap mac (simulate XDP_TX)
> > > 5066243 pkt/s = xdp_redirect_map: XDP_REDIRECT with swap mac + devmap
> > >
> > > The performance drop between xdp2 and xdp_redirect, was expected due
> > > to the HW-tailptr flush per packet, which is costly.
> > >
> > > (1/6344472-1/4595574)*10^9 = -59.98 ns
> > >
> > > The performance drop between xdp2 and xdp_redirect_map, is higher than
> > > I expected, which is not good! The avoidance of the tailptr flush per
> > > packet was expected to give a higher boost. The cost increased with
> > > 40 ns, which is too high compared to the code added (on a 4GHz machine
> > > approx 160 cycles).
> > >
> > > (1/6344472-1/5066243)*10^9 = -39.77 ns
> > >
> > > This system doesn't have DDIO, thus we are stalling on cache-misses,
> > > but I was actually expecting that the added code could "hide" behind
> > > these cache-misses.
> > >
> > > I'm somewhat surprised to see this large a performance drop.
> > >
> >
> > Yep, although there is room for optimizations in the code path for sure. And
> > 5mpps is not horrible my preference is to get this series in plus any
> > small optimization we come up with while the merge window is closed. Then
> > follow up patches can do optimizations.
>
> IMHO 5Mpps is a very bad number for XDP.
>
> > One easy optimization is to get rid of the atomic bitops. They are not needed
> > here we have a per cpu unsigned long. Another easy one would be to move
> > some of the checks out of the hotpath. For example checking for ndo_xdp_xmit
> > and flush ops on the net device in the hotpath really should be done in the
> > slow path.
>
> I'm already running with a similar patch as below, but it
> (surprisingly) only gave my 3 ns improvement. I also tried a
> prefetchw() on xdp.data that gave me 10 ns (which is quite good).
>
> I'm booting up another system with a CPU E5-1650 v4 @ 3.60GHz, which
> have DDIO ... I have high hopes for this, as the major bottleneck on
> this CPU i7-4790K CPU @ 4.00GHz is clearly cache-misses.
>
> Something is definitely wrong on this CPU, as perf stats shows, a very
> bad utilization of the CPU pipeline with 0.89 insn per cycle.
Wow, getting DDIO working and avoiding the cache-miss, was really
_the_ issue. On this CPU E5-1650 v4 @ 3.60GHz things look really
really good for XDP_REDIRECT with maps. (p.s. with __set_bit()
optimization)
13,939,674 pkt/s = XDP_DROP without touching memory
14,290,650 pkt/s = xdp1: XDP_DROP with reading packet data
13,221,812 pkt/s = xdp2: XDP_TX with swap mac (writes into pkt)
7,596,576 pkt/s = xdp_redirect: XDP_REDIRECT with swap mac (like XDP_TX)
13,058,435 pkt/s = xdp_redirect_map:XDP_REDIRECT with swap mac + devmap
Surprisingly, on this DDIO capable CPU it is slightly slower NOT to
read packet memory.
The large performance gap to xdp_redirect is due to the tailptr flush,
which really show up on this system. The CPU efficiency is 1.36 insn
per cycle, which for map variant is 2.15 insn per cycle.
Gap (1/13221812-1/7596576)*10^9 = -56 ns
The xdp_redirect_map performance is really really good, almost 10G
wirespeed on a single CPU!!! This is amazing, and we know that this
code is not even optimal yet. The performance difference to xdp2 is
only around 1 ns.
--
Best regards,
Jesper Dangaard Brouer
MSc.CS, Principal Kernel Engineer at Red Hat
LinkedIn: http://www.linkedin.com/in/brouer
[jbrouer@...1650v4 bpf]$ sudo ./xdp1 6
proto 17: 11919302 pkt/s
proto 17: 14281659 pkt/s
proto 17: 14290650 pkt/s
proto 17: 14283120 pkt/s
proto 17: 14303023 pkt/s
proto 17: 14299496 pkt/s
[jbrouer@...1650v4 bpf]$ sudo ./xdp2 6
proto 0: 1 pkt/s
proto 17: 3225455 pkt/s
proto 17: 13266772 pkt/s
proto 17: 13221812 pkt/s
proto 17: 13222200 pkt/s
proto 17: 13225508 pkt/s
proto 17: 13226274 pkt/s
[jbrouer@...1650v4 bpf]$ sudo ./xdp_redirect 6 6
ifindex 6: 66040 pkt/s
ifindex 6: 7029143 pkt/s
ifindex 6: 7596576 pkt/s
ifindex 6: 7598499 pkt/s
ifindex 6: 7597025 pkt/s
ifindex 6: 7598462 pkt/s
[jbrouer@...1650v4 bpf]$ sudo ./xdp_redirect_map 6 6
map[0] (vports) = 4, map[1] (map) = 5, map[2] (count) = 0
ifindex 6: 95429 pkt/s
ifindex 6: 12156600 pkt/s
ifindex 6: 13058435 pkt/s
ifindex 6: 13058515 pkt/s
ifindex 6: 13059213 pkt/s
ifindex 6: 13058322 pkt/s
ifindex 6: 13059342 pkt/s
[jbrouer@...1650v4 prototype-kernel]$
sudo ./xdp_bench01_mem_access_cost --dev ixgbe2 --action XDP_DROP
XDP_action pps pps-human-readable mem
XDP_DROP 0 0 no_touch
XDP_DROP 1 1 no_touch
XDP_DROP 11959667 11,959,667 no_touch
XDP_DROP 13939674 13,939,674 no_touch
XDP_DROP 13954549 13,954,549 no_touch
XDP_DROP 13953897 13,953,897 no_touch
XDP_DROP 13963531 13,963,531 no_touch
[jbrouer@...1650v4 prototype-kernel]$
sudo ./xdp_bench01_mem_access_cost --dev ixgbe2 --action XDP_DROP --read
XDP_action pps pps-human-readable mem
XDP_DROP 0 0 read
XDP_DROP 0 0 read
XDP_DROP 0 0 read
XDP_DROP 8611099 8,611,099 read
XDP_DROP 14300230 14,300,230 read
XDP_DROP 14293416 14,293,416 read
XDP_DROP 14297247 14,297,247 read
XDP_DROP 14300563 14,300,563 read
XDP_DROP 14299873 14,299,873 read
^CInterrupted: Removing XDP program on ifindex:6 device:ixgbe2
[jbrouer@...1650v4 prototype-kernel]$
sudo ./xdp_bench01_mem_access_cost --dev ixgbe2 --action XDP_TX --swap
XDP_action pps pps-human-readable mem
XDP_TX 1 1 swap_mac
XDP_TX 3007657 3,007,657 swap_mac
XDP_TX 13322885 13,322,885 swap_mac
XDP_TX 13200845 13,200,845 swap_mac
XDP_TX 13189829 13,189,829 swap_mac
XDP_TX 13197952 13,197,952 swap_mac
XDP_TX 13198856 13,198,856 swap_mac
^CInterrupted: Removing XDP program on ifindex:6 device:ixgbe2
Normal xdp_redirect:
[jbrouer@...alhost ~]$ sudo perf stat -C10 -e L1-icache-load-misses -e cycles:k -e instructions:k -e cache-misses:k -e cache-references:k -e LLC-store-misses:k -e LLC-store -e LLC-load-misses:k -e LLC-load -e L1-dcache-load-misses -e L1-dcache-loads -e L1-dcache-stores -e ref-cycles -e bus-cycles -r 3 sleep 1
Performance counter stats for 'CPU(s) 10' (3 runs):
105,645 L1-icache-load-misses ( +- 0.90% ) (35.48%)
3,790,481,191 cycles:k ( +- 0.00% ) (42.67%)
5,159,580,049 instructions:k # 1.36 insn per cycle ( +- 0.02% ) (49.87%)
931 cache-misses:k # 0.004 % of all cache refs ( +- 31.80% ) (49.97%)
21,394,789 cache-references:k ( +- 0.03% ) (50.07%)
0 LLC-store-misses (43.07%)
840,689 LLC-store ( +- 0.13% ) (43.16%)
0 LLC-load-misses (14.37%)
8,031,535 LLC-load ( +- 0.02% ) (14.27%)
42,211,992 L1-dcache-load-misses # 2.49% of all L1-dcache hits ( +- 0.01% ) (21.36%)
1,692,701,894 L1-dcache-loads ( +- 0.02% ) (28.46%)
922,985,760 L1-dcache-stores ( +- 0.02% ) (28.37%)
3,591,805,402 ref-cycles ( +- 0.00% ) (35.47%)
99,762,378 bus-cycles ( +- 0.00% ) (35.47%)
1.000964284 seconds time elapsed ( +- 0.01% )
xdp_redirect_map::
[jbrouer@...alhost ~]$ sudo perf stat -C10 -e L1-icache-load-misses -e cycles:k -e instructions:k -e cache-misses:k -e cache-references:k -e LLC-store-misses:k -e LLC-store -e LLC-load-misses:k -e LLC-load -e L1-dcache-load-misses -e L1-dcache-loads -e L1-dcache-stores -e ref-cycles -e bus-cycles -r 3 sleep 1
Performance counter stats for 'CPU(s) 10' (3 runs):
103,113 L1-icache-load-misses ( +- 11.60% ) (35.48%)
3,789,133,467 cycles:k ( +- 0.03% ) (42.66%)
8,152,033,594 instructions:k # 2.15 insn per cycle ( +- 9.00% ) (49.85%)
1,414 cache-misses:k # 0.004 % of all cache refs ( +- 21.42% ) (49.95%)
32,480,603 cache-references:k ( +- 8.63% ) (50.05%)
0 LLC-store-misses (43.06%)
786,799 LLC-store ( +- 1.57% ) (43.14%)
67 LLC-load-misses ( +-100.00% ) (14.37%)
12,445,529 LLC-load ( +- 9.05% ) (14.28%)
77,013,768 L1-dcache-load-misses # 2.83% of all L1-dcache hits ( +- 9.07% ) (21.38%)
2,725,676,877 L1-dcache-loads ( +- 8.98% ) (28.47%)
1,566,087,361 L1-dcache-stores ( +- 8.95% ) (28.39%)
3,590,481,746 ref-cycles ( +- 0.03% ) (35.48%)
99,725,522 bus-cycles ( +- 0.03% ) (35.47%)
1.000920909 seconds time elapsed ( +- 0.00% )
Powered by blists - more mailing lists