Re: [PATCH net-next] loopback: sctp: add NETIF_F_SCTP_CSUM to device features

[Daniel Borkmann <dborkman@redhat.com>]

On 02/24/2014 11:42 AM, David Laight wrote:
...
> I'm sure it shouldn't be that expensive, you are implying that it spent
> about 70% of the time doing crc32.

In this scenario, the following perf log I get that shows where cycles
are being spent on my machine:

  65.95%       netperf  [kernel.kallsyms]  [k] __crc32c_le
   3.79%       netperf  [kernel.kallsyms]  [k] memcpy
   2.38%       netperf  [kernel.kallsyms]  [k] copy_user_enhanced_fast_string
   0.62%       netperf  [sctp]             [k] sctp_datamsg_from_user
   0.62%       netperf  [sctp]             [k] sctp_sendmsg
   0.55%       netperf  [kernel.kallsyms]  [k] __slab_free
   0.52%       netperf  [sctp]             [k] sctp_outq_flush
   0.50%       netperf  [kernel.kallsyms]  [k] kfree
   0.49%       netperf  [kernel.kallsyms]  [k] cmpxchg_double_slab.isra.52
   0.48%       netperf  [kernel.kallsyms]  [k] kmem_cache_alloc
   0.43%       netperf  [kernel.kallsyms]  [k] __slab_alloc
   0.42%       netperf  [kernel.kallsyms]  [k] __copy_skb_header
   0.41%       netperf  [kernel.kallsyms]  [k] __alloc_skb

> The loop should be dominated by the per-byte lookup in a 256 word table.
> With 4k data the table will soon be in the data cache.
> Unless it is (stupidly) generating the table on each call, or trying
> to use a crc32 instruction, faulting, and emulating it, I wouldn't
> really have expected more than a few % improvement.