From mboxrd@z Thu Jan 1 00:00:00 1970 From: Peter Zijlstra Subject: Re: [RFC 0/2] drivers: spi/i2c: account completions as iowait Date: Sun, 2 Nov 2014 17:59:43 +0100 Message-ID: <20141102165943.GT10501@worktop.programming.kicks-ass.net> References: <1414936689-2707-1-git-send-email-wsa@the-dreams.de> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-i2c-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-spi-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, Mark Brown , Ingo Molnar , Balbir Singh To: Wolfram Sang Return-path: Content-Disposition: inline In-Reply-To: <1414936689-2707-1-git-send-email-wsa-z923LK4zBo2bacvFa/9K2g@public.gmane.org> Sender: linux-i2c-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org List-Id: linux-spi.vger.kernel.org On Sun, Nov 02, 2014 at 02:58:07PM +0100, Wolfram Sang wrote: > However, researching the net, users currently interpret iowait entirely as > blkio wait. Furthermore, io_schedule() calls delayacct_blkio_{start|end}() which > worked fine for my tests with I2C but might show that iowait was really meant as > blkiowait? So, should other subsystems use it? I would tend to agree with that; historically this has always been about blkio, not device io. > To make it more confusing, some people (like Peter Zijlstra [1]) seem to like > iowait gone, so maybe it is all not worth it? Yeah, iowait accounting is terminally broken :-) Mostly because the iowait is accounted per-cpu but that is a very tenuous relation because the IO devices are not per IO and blocking tasks are not associated with any particular CPU -- after all they're not consuming CPU time. If people really think its worth; we could invent new IO-wait measure that do make sense -- maybe, but the current thing is complete bollocks.