On 10/15/13 7:52 PM, Namhyung Kim wrote:
Aha, okay. So it mostly matters to syscall tracing, right? For a normal record session, it seems that the effect is not that large:
Yes, that's in the description "When recording raw_syscalls for the entire system"
There is a small benefit to all record sessions -- mmap+memcpy has less overhead than write(). I still need to look at Ingo's suggestion to use non-temporal stores which might reduce the overhead of the memcpy.
Try a workload that generates a HUGE data file -- say a full kernel build (e.g., perf record -g -- make O=/tmp/kbuild -j 16). You should see a much larger benefit from the mmap route. Be sure to use your callchain enhancements to look at that 1+G file. ;-)
David -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
