From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752938AbbCSWtB (ORCPT ); Thu, 19 Mar 2015 18:49:01 -0400 Received: from lists.s-osg.org ([54.187.51.154]:43779 "EHLO lists.s-osg.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752317AbbCSWsq (ORCPT ); Thu, 19 Mar 2015 18:48:46 -0400 Message-ID: <550B524C.5030104@osg.samsung.com> Date: Thu, 19 Mar 2015 16:48:44 -0600 From: Shuah Khan Organization: Samsung Open Source Group User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.5.0 MIME-Version: 1.0 To: John Stultz CC: lkml , Prarit Bhargava , Thomas Gleixner , Richard Cochran Subject: Re: [PATCH] kselftest/timers: Set default threadtest values to simplify execution scripts References: <1426693913-5738-1-git-send-email-john.stultz@linaro.org> <550B4756.2040208@osg.samsung.com> In-Reply-To: Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 03/19/2015 04:34 PM, John Stultz wrote: > On Thu, Mar 19, 2015 at 3:01 PM, Shuah Khan wrote: >> On 03/18/2015 09:51 AM, John Stultz wrote: >>> In order to keep the kselftest Makefiles simpler, set the threadtest >>> default values to the ones used in standard run_tests >>> >>> Cc: Shuah Khan >>> Cc: Prarit Bhargava >>> Cc: Thomas Gleixner >>> Cc: Richard Cochran >>> Signed-off-by: John Stultz >>> --- >>> tools/testing/selftests/timers/threadtest.c | 8 ++++++-- >>> 1 file changed, 6 insertions(+), 2 deletions(-) >>> >> >> Applied to next for 4.1 >> >> Some numbers for you with timer tests included: >> >> make kselftest target takes: >> >> real 11m50.499s >> user 3m25.979s >> sys 5m45.433s >> >> It is creeping up, previous timing was >> >> real 9.41 >> user 3.55 >> system 0:24.86 >> >> Not concerned yet. Might be getting closer to >> needing to defining quick vs long test categories. > > Yea, the timekeeping tests are particularly rough about how long the > run. In some cases we're having to watch for behavior that could be > somewhat rare, so we need to watch for a fair amount of time. In some > cases we're doing our own calibrations which require a larger amount > of time to ensure accuracy. And in other cases, we want to have timers > that fire far enough out that any scheduler variance/noise is easy to > filter out. > > With the destructive tests, which re-run the validation tests > repeatedly under different conditions, it ends up being about an hour! > So I feel this pain. > > But there's also probably some spots where 3 seconds seemed like a > good value, but could be shorter. So I'll have to take another look > to see if we could reasonably compress some of the intervals we use > down. There may also be some spots where we could parallelize the > tests across the various clockids. > If you can take a look to see which tests can be included in a quick test run vs. longer test run in addition to destructive vs. normal. I will work on adding categories soon. The logic can be isolated in selftest/Makefile and timers/Makefile. thanks, -- Shuah -- Shuah Khan Sr. Linux Kernel Developer Open Source Innovation Group Samsung Research America (Silicon Valley) shuahkh@osg.samsung.com | (970) 217-8978