| Authors: | Jesper Dangaard Brouer <[email protected]> |
|---|
Here is a quick starting guide, that ends showing the output from the time_bench_sample kernel module. I would like to know the micro-benchmark performance of different CPUs, thus feel free to email me the output after completing this exercise ;-)
Download source:
$ git clone https://github.com/netoptimizer/prototype-kernel.git
Try to compile modules (you need RPM kernel-devel or a local kernel source tree avail):
$ cd prototype-kernel/kernel $ make
After first make, run script to symlink kernel modules into your local kernel module tree:
$ dirs ~/git/prototype-kernel/kernel $ sudo ./scripts/setup_prototype_devel_env.sh [output shortened:] Symlink all kernel modules './time_bench_sample.ko' -> 'prototype-kernel/kernel/lib/time_bench_sample.ko' [...] Depmod - update kernel module symbols (on hostname)
Now you can load the kernel modules, and load dependencies also work. Run the setup script again if you have added a new module.
Lets look at module time_bench_sample:
$ modinfo time_bench_sample filename: /lib/modules/3.16.4-200.fc20.x86_64/kernel/extra/time_bench_sample.ko license: GPL author: Jesper Dangaard Brouer <[email protected]> description: Sample: Benchmarking code execution time in kernel depends: time_bench vermagic: 3.16.4-200.fc20.x86_64 SMP mod_unload
The time_bench_sample module simply prints its results to the kernel log. I usually "run-it" like this:
$ make && \ (sudo rmmod time_bench_sample; sudo modprobe time_bench_sample) \ && sudo dmesg -c
On my laptop CPU i7-2620M the output looks like this. I've cleaned it up by adding line-breaks before "Type" and "-"
[1523779.968385] time_bench_sample: Unloaded [1523779.975410] time_bench_sample: Loaded [1523780.298424] time_bench: Type:for_loop Per elem: 0 cycles(tsc) 0.322 ns (step:0) - (measurement period time:0.322574148 sec time_interval:322574148) - (invoke count:1000000000 tsc_interval:868128306) [1523781.558964] time_bench: Type:spin_lock_unlock Per elem: 33 cycles(tsc) 12.588 ns (step:0) - (measurement period time:1.258850730 sec time_interval:1258850730) - (invoke count:100000000 tsc_interval:3387886371) [1523782.240786] time_bench: Type:local_BH_disable_enable Per elem: 18 cycles(tsc) 6.809 ns (step:0) - (measurement period time:0.680904887 sec time_interval:680904887) - (invoke count:100000000 tsc_interval:1832487465) [1523782.523465] time_bench: Type:local_IRQ_disable_enable Per elem: 7 cycles(tsc) 2.822 ns (step:0) - (measurement period time:0.282294042 sec time_interval:282294042) - (invoke count:100000000 tsc_interval:759724455) [1523783.885408] time_bench: Type:local_irq_save_restore Per elem: 36 cycles(tsc) 13.601 ns (step:0) - (measurement period time:1.360118333 sec time_interval:1360118333) - (invoke count:100000000 tsc_interval:3660422754) [1523783.918188] time_bench: Type:preempt_disable_enable Per elem: 0 cycles(tsc) 0.327 ns (step:0) - (measurement period time:0.032728667 sec time_interval:32728667) - (invoke count:100000000 tsc_interval:88080549) [1523784.139055] time_bench: Type:funcion_call_cost Per elem: 5 cycles(tsc) 2.205 ns (step:0) - (measurement period time:0.220564571 sec time_interval:220564571) - (invoke count:100000000 tsc_interval:593594688) [1523784.325287] time_bench: Type:func_ptr_call_cost Per elem: 5 cycles(tsc) 1.859 ns (step:0) - (measurement period time:0.185976490 sec time_interval:185976490) - (invoke count:100000000 tsc_interval:500509440) [1523784.483675] time_bench: Type:page_alloc_put Per elem: 425 cycles(tsc) 158.170 ns (step:0) - (measurement period time:0.158170395 sec time_interval:158170395) - (invoke count:1000000 tsc_interval:425676012)