From mboxrd@z Thu Jan 1 00:00:00 1970 From: Dave Chinner Date: Thu, 02 Jul 2015 00:19:51 +0000 Subject: Re: XFS: Fine-tuning for checks before the function call "xfs_qm_dqrele"? Message-Id: <20150702001951.GX22807@dastard> List-Id: References: <530CF8FF.8080600@users.sourceforge.net> <530DD06F.4090703@users.sourceforge.net> <5317A59D.4@users.sourceforge.net> <5479F823.60900@users.sourceforge.net> <20141130230904.GF16151@dastard> <558D1833.3080009@users.sourceforge.net> <20150629214355.GE7943@dastard> <55939BA8.1090400@users.sourceforge.net> In-Reply-To: <55939BA8.1090400@users.sourceforge.net> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: SF Markus Elfring Cc: Julia Lawall , kernel-janitors@vger.kernel.org, LKML , xfs@oss.sgi.com On Wed, Jul 01, 2015 at 09:50:00AM +0200, SF Markus Elfring wrote: > > xfs_dqrele_inode() gets called on every inode in the inode cache, > > and this change results in a cacheline in every inode being dirtied > > even if they don't have dquots attached. Given the inode cache can > > hold tens to hundreds of millions of inodes on large machines, we > > don't want to dirty any cachelines we don't need to while walking > > the inode cache and releasing dquots... > > Would it make sense to annotate checks before such function calls > as "LIKELY"? No - it will be random as to whether the inodes have dquots attached or not and so a static hint is always going to be wrong for someone.... Cheers, Dave. -- Dave Chinner david@fromorbit.com