From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from ipmail07.adl2.internode.on.net ([150.101.137.131]:22826 "EHLO ipmail07.adl2.internode.on.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726321AbeIZHKl (ORCPT ); Wed, 26 Sep 2018 03:10:41 -0400 Date: Wed, 26 Sep 2018 10:59:35 +1000 From: Dave Chinner Subject: Re: Strange behavior with log IO fake-completions Message-ID: <20180926005935.GC31060@dastard> References: <20180910235827.GD5631@dastard> <20180912004259.GJ5631@dastard> <20180917025601.GE16550@dastard> <20180921002336.GO27618@dastard> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: Sender: linux-xfs-owner@vger.kernel.org List-ID: List-Id: xfs To: Joshi Cc: linux-xfs@vger.kernel.org On Mon, Sep 24, 2018 at 07:55:02PM +0530, Joshi wrote: > > I changed all log wait queues (ic_force_wait, ic_write_wait, and > xc_commit_wait) into completion variables (despite knowing that it is > also implemented using wait_queue). > > Sorry, ignore this and subsequent part of the text. I think replacing > with completion requires more work than I did.Calling "complete_all" > is tricky, given the need of calling "reinit_completion" which may get > executed by multiple threads in this case. Yup. Perhaps you might like to look at this recent patch set aimed at reducing spurious wake-ups and lock contention in the log reservation code: https://marc.info/?l=linux-kernel&m=153531683127094&w=2 Not sure it applies to this case, but it still may be useful to you. BTW, when posting benchmark results, can you also include a baseline result from a completely unmodified kernel and the patches you are testing? Nobody can evaluate the correctness or effectiveness of the changes you are testing without these things.... Cheers, Dave. -- Dave Chinner david@fromorbit.com