netdev - Re: [PATCH v2 net-next] ravb: RX checksum offload

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

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <58512067-ed8a-d9d1-1f29-0351c87c02b7@cogentembedded.com>
Date:   Wed, 4 Oct 2017 21:53:09 +0300
From:   Sergei Shtylyov <sergei.shtylyov@...entembedded.com>
To:     Simon Horman <horms+renesas@...ge.net.au>,
        David Miller <davem@...emloft.net>
Cc:     Magnus Damm <magnus.damm@...il.com>, netdev@...r.kernel.org,
        linux-renesas-soc@...r.kernel.org
Subject: Re: [PATCH v2 net-next] ravb: RX checksum offload

On 10/04/2017 10:54 AM, Simon Horman wrote:

> Add support for 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
> 
> The RAVB provides 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.
> 
> In terms of performance throughput is close to gigabit line-rate both with
> and without RX checksum offload enabled. 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:
>   # /usr/bin/perf_3.16 record -o /run/perf.data -a netperf -t TCP_MAERTS -H 10.4.3.162
>   MIGRATED TCP MAERTS TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.4.3.162 () port 0 AF_INET : demo
>   enable_enobufs failed: getprotobyname
>   Recv   Send    Send
>   Socket Socket  Message  Elapsed
>   Size   Size    Size     Time     Throughput
>   bytes  bytes   bytes    secs.    10^6bits/sec
> 
>    87380  16384  16384    10.00     937.54
> 
>   Summary of output of perf report:
>      18.28%      ksoftirqd/0  [kernel.kallsyms]  [k] _raw_spin_unlock_irqrestore
>      10.34%      ksoftirqd/0  [kernel.kallsyms]  [k] __pi_memcpy
>       9.83%      ksoftirqd/0  [kernel.kallsyms]  [k] ravb_poll
>       7.89%      ksoftirqd/0  [kernel.kallsyms]  [k] skb_put
>       4.01%      ksoftirqd/0  [kernel.kallsyms]  [k] dev_gro_receive
>       3.37%          netperf  [kernel.kallsyms]  [k] __arch_copy_to_user
>       3.17%          swapper  [kernel.kallsyms]  [k] arch_cpu_idle
>       2.55%          swapper  [kernel.kallsyms]  [k] tick_nohz_idle_enter
>       2.04%      ksoftirqd/0  [kernel.kallsyms]  [k] __pi___inval_dcache_area
>       2.03%          swapper  [kernel.kallsyms]  [k] _raw_spin_unlock_irq
>       1.96%      ksoftirqd/0  [kernel.kallsyms]  [k] __netdev_alloc_skb
>       1.59%      ksoftirqd/0  [kernel.kallsyms]  [k] __slab_alloc.isra.83
> 
> Test results without RX checksum offload enabled:
>   # /usr/bin/perf_3.16 record -o /run/perf.data -a netperf -t TCP_MAERTS -H 10.4.3.162
>   MIGRATED TCP MAERTS TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.4.3.162 () port 0 AF_INET : demo
>   enable_enobufs failed: getprotobyname
>   Recv   Send    Send
>   Socket Socket  Message  Elapsed
>   Size   Size    Size     Time     Throughput
>   bytes  bytes   bytes    secs.    10^6bits/sec
> 
>    87380  16384  16384    10.00     940.20
> 
>   Summary of output of perf report:
>      17.10%    ksoftirqd/0  [kernel.kallsyms]  [k] _raw_spin_unlock_irqrestore
>      10.99%    ksoftirqd/0  [kernel.kallsyms]  [k] __pi_memcpy
>       8.87%    ksoftirqd/0  [kernel.kallsyms]  [k] ravb_poll
>       8.16%    ksoftirqd/0  [kernel.kallsyms]  [k] skb_put
>       7.42%    ksoftirqd/0  [kernel.kallsyms]  [k] do_csum
>       3.91%    ksoftirqd/0  [kernel.kallsyms]  [k] dev_gro_receive
>       2.31%        swapper  [kernel.kallsyms]  [k] arch_cpu_idle
>       2.16%    ksoftirqd/0  [kernel.kallsyms]  [k] __pi___inval_dcache_area
>       2.14%    ksoftirqd/0  [kernel.kallsyms]  [k] __netdev_alloc_skb
>       1.93%        netperf  [kernel.kallsyms]  [k] __arch_copy_to_user
>       1.79%        swapper  [kernel.kallsyms]  [k] tick_nohz_idle_enter
>       1.63%    ksoftirqd/0  [kernel.kallsyms]  [k] __slab_alloc.isra.83
> 
> Above results collected on an R-Car Gen 3 Salvator-X/r8a7796 ES1.0.
> Also tested on a R-Car Gen 3 Salvator-X/r8a7795 ES1.0.
> 
> By inspection this also appears to be compatible with the ravb found
> on R-Car Gen 2 SoCs, however, this patch is currently untested on such
> hardware.
> 
> Signed-off-by: Simon Horman <horms+renesas@...ge.net.au>
> 
> ---
> v2
> Address review of v1 by Sergei Shtylyov
> * set features rather than oring them with (zero) existing values:
> * Set/unset using a single call to ravb_modify()

   Already belated but still:

Reviewed-by: Sergei Shtylyov <sergei.shtylyov@...entembedded.com>

MBR, Sergei