From mboxrd@z Thu Jan 1 00:00:00 1970 From: John Stultz Subject: Re: [RFC][PATCH] Input: Add infrastrucutre for selecting clockid for event time stamps. Date: Fri, 06 Jan 2012 11:39:16 -0800 Message-ID: <1325878756.2290.101.camel@work-vm> References: <1325818212-24752-1-git-send-email-john.stultz@linaro.org> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Return-path: Received: from e34.co.us.ibm.com ([32.97.110.152]:34714 "EHLO e34.co.us.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755282Ab2AFTjW (ORCPT ); Fri, 6 Jan 2012 14:39:22 -0500 Received: from /spool/local by e34.co.us.ibm.com with IBM ESMTP SMTP Gateway: Authorized Use Only! Violators will be prosecuted for from ; Fri, 6 Jan 2012 12:39:21 -0700 Received: from d03av05.boulder.ibm.com (d03av05.boulder.ibm.com [9.17.195.85]) by d03relay04.boulder.ibm.com (8.13.8/8.13.8/NCO v10.0) with ESMTP id q06JdI6d030126 for ; Fri, 6 Jan 2012 12:39:18 -0700 Received: from d03av05.boulder.ibm.com (loopback [127.0.0.1]) by d03av05.boulder.ibm.com (8.14.4/8.13.1/NCO v10.0 AVout) with ESMTP id q06JdHYO001932 for ; Fri, 6 Jan 2012 12:39:18 -0700 In-Reply-To: Sender: linux-input-owner@vger.kernel.org List-Id: linux-input@vger.kernel.org To: Daniel Kurtz Cc: linux-input@vger.kernel.org, Dmitry Torokhov , Arve =?ISO-8859-1?Q?Hj=F8nnev=E5g?= On Fri, 2012-01-06 at 17:20 +0800, Daniel Kurtz wrote: > On Fri, Jan 6, 2012 at 10:50 AM, John Stultz wrote: > > Here's another revision, incorperating Dmitry's suggestion. > > > > As noted by Arve and others, since wall time can jump backwards, it > > is difficult to use for input because one cannot determine if one > > event occured before another or for how long a key was pressed. > > > > However, the timestamp field is part of the kernel ABI, and cannot > > be changed without possibly breaking existing users. > > > > This patch adds a new IOCTL that allows a clockid to be set in > > the evdev_client struct that will specify which time base to > > use for event timestamps (ie: CLOCK_MONOTONIC instead > > of CLOCK_REALTIME). > > > > For now we only support CLOCK_MONOTONIC and CLOCK_REALTIME, but > > in the future we could support other clockids if appropriate. > > What about CLOCK_MONOTONIC_RAW? It could be used here, although I'm still not clear on the benefit of using monotonic raw over just monotonic. > Last time we discussed, I thought this clock was the most useful for > use with input devices. But, you wrote this: > > So rawmonotonic isn't frequency corrected via NTP, while the monotonic > > clock is. So if you're calculating intervals, you will get more accurate > > times (where a second is a second) w/ ktime_get_ts(). > > Does this frequency correction involve timestamp jumps? No. clock monotonic isn't jumped. But its rate may be slightly adjusted (max of +/- 500ppm), so that it accurately matches the passing of time. > If so, of what magnitude? At worse, assuming some sort of terribly mis-configured or malicious ntp daemon, if you had two one second intervals that you had measured, they could actually differ by up to a millisecond. > In my experience, input event timestamp intervals are usually on the > order of a few milliseconds (5-25 ms). If CLOCK_MONOTONIC experiences > frequent adjustments near this order of magnitude, I still think > CLOCK_MONOTONIC_RAW might be a better choice for event timestamps. So for 5-25 ms intervals, you're looking at a worse case difference of 5-25 us. And again, this isn't likely, as the ntp freq adjustment would have to go from -500ppm to +500ppm mid-interval. So I really suspect there isn't much practical difference between CLOCK_MONOTONIC and CLOCK_MONOTONIC_RAW for input events. But I'd be interested to hear if anyone has actually run into such complications. CLOCK_MONOTONIC_RAW can be useful (especially for providing a stable baseline when doing time adjustment calculations), but the downside is that measurements (especially longer measurements) with CLOCK_MONOTONIC_RAW may not be accurate. And as it always is with time, everything is relative: even CLOCK_MONOTONIC_RAW can vary over time, as the crystal driving the clocksource may slightly fluctuate with temperature. But again, nothing is keeping CLOCK_MONOTONIC_RAW from being supported via the proposed ioctl. > > + bool timestamp_clkid; > > This isn't bool anymore. Good catch! thanks! -john