On Sat, 2014-02-08 at 13:06 -0800, Andy Lutomirski wrote:
This toggles TIF_SYSCALL_AUDIT as needed when rules change instead
of leaving it set whenever rules might be set in the future. This
reduces syscall latency from >60ns to closer to 40ns on my laptop.
Al also politely reminded me it might be wise to get some perf data
about where exactly we are spending out time. I don't know squat about
perf, but Linus always tells me to do:
perf record -g -e cycles:pp -F 25000 $YOURTEST
perf report -s symbol
(the "-s symbol" is so that you don't get separate data for the
different processes that are part of the kernel build - you'll just
want "general kernel data"), and on one of the kernel symbols just
select it and do "Zoom into kernel DSO". You should see something like
this: