Add support for the RX checksum offload. This is enabled by default and
may be disabled and re-enabled using 'ethtool':
# ethtool -K eth0 rx off
# ethtool -K eth0 rx on
Some Ether MACs provide a simple checksumming scheme which appears to be
completely compatible with CHECKSUM_COMPLETE: sum of all packet data after
the L2 header is appended to packet data; this may be trivially read by
the driver and used to update the skb accordingly. The same checksumming
scheme is implemented in the EtherAVB MACs and now supported by the 'ravb'
driver.
In terms of performance, throughput is close to gigabit line rate with the
RX checksum offload both enabled and disabled. The 'perf' output, however,
appears to indicate that significantly less time is spent in do_csum() --
this is as expected.
Test results with RX checksum offload enabled:
~/netperf-2.2pl4# perf record -a ./netperf -t TCP_MAERTS -H 192.168.2.4
TCP MAERTS TEST to 192.168.2.4
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
131072 16384 16384 10.01 933.93
[ perf record: Woken up 8 times to write data ]
[ perf record: Captured and wrote 1.955 MB perf.data (41940 samples) ]
~/netperf-2.2pl4# perf report
Samples: 41K of event 'cycles:ppp', Event count (approx.): 9915302763
Overhead Command Shared Object Symbol
9.44% netperf [kernel.kallsyms] [k] __arch_copy_to_user
7.75% swapper [kernel.kallsyms] [k] _raw_spin_unlock_irq
6.31% swapper [kernel.kallsyms] [k] default_idle_call
5.89% swapper [kernel.kallsyms] [k] arch_cpu_idle
4.37% swapper [kernel.kallsyms] [k] tick_nohz_idle_exit
4.02% netperf [kernel.kallsyms] [k] _raw_spin_unlock_irq
2.52% netperf [kernel.kallsyms] [k] preempt_count_sub
1.81% netperf [kernel.kallsyms] [k] tcp_recvmsg
1.80% netperf [kernel.kallsyms] [k] _raw_spin_unlock_irqres
1.78% netperf [kernel.kallsyms] [k] preempt_count_add
1.36% netperf [kernel.kallsyms] [k] __tcp_transmit_skb
1.20% netperf [kernel.kallsyms] [k] __local_bh_enable_ip
1.10% netperf [kernel.kallsyms] [k] sh_eth_start_xmit
Test results with RX checksum offload disabled:
~/netperf-2.2pl4# perf record -a ./netperf -t TCP_MAERTS -H 192.168.2.4
TCP MAERTS TEST to 192.168.2.4
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
131072 16384 16384 10.01 932.04
[ perf record: Woken up 14 times to write data ]
[ perf record: Captured and wrote 3.642 MB perf.data (78817 samples) ]
~/netperf-2.2pl4# perf report
Samples: 78K of event 'cycles:ppp', Event count (approx.): 18091442796
Overhead Command Shared Object Symbol
7.00% swapper [kernel.kallsyms] [k] do_csum
3.94% swapper [kernel.kallsyms] [k] sh_eth_poll
3.83% ksoftirqd/0 [kernel.kallsyms] [k] do_csum
3.23% swapper [kernel.kallsyms] [k] _raw_spin_unlock_irq
2.87% netperf [kernel.kallsyms] [k] __arch_copy_to_user
2.86% swapper [kernel.kallsyms] [k] arch_cpu_idle
2.13% swapper [kernel.kallsyms] [k] default_idle_call
2.12% ksoftirqd/0 [kernel.kallsyms] [k] sh_eth_poll
2.02% swapper [kernel.kallsyms] [k] _raw_spin_unlock_irqrestore
1.84% swapper [kernel.kallsyms] [k] __softirqentry_text_start
1.64% swapper [kernel.kallsyms] [k] tick_nohz_idle_exit
1.53% netperf [kernel.kallsyms] [k] _raw_spin_unlock_irq
1.32% netperf [kernel.kallsyms] [k] preempt_count_sub
1.27% swapper [kernel.kallsyms] [k] __pi___inval_dcache_area
1.22% swapper [kernel.kallsyms] [k] check_preemption_disabled
1.01% ksoftirqd/0 [kernel.kallsyms] [k] _raw_spin_unlock_irqrestore
The above results collected on the R-Car V3H Starter Kit board.
Based on the commit 4d86d38186 ("ravb: RX checksum offload")...
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
f8e022db50