From mboxrd@z Thu Jan 1 00:00:00 1970 From: Dave Chinner Subject: Re: [PATCH 3/4] fs: Lock the inode LRU list separately Date: Thu, 28 Oct 2010 09:24:44 +1100 Message-ID: <20101027222444.GC2715@dastard> References: <1288153384-8878-1-git-send-email-david@fromorbit.com> <1288153384-8878-4-git-send-email-david@fromorbit.com> <20101027090530.GB16443@infradead.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: viro@ZenIV.linux.org.uk, linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org To: Christoph Hellwig Return-path: Received: from bld-mail13.adl6.internode.on.net ([150.101.137.98]:38991 "EHLO mail.internode.on.net" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1754016Ab0J0WYs (ORCPT ); Wed, 27 Oct 2010 18:24:48 -0400 Content-Disposition: inline In-Reply-To: <20101027090530.GB16443@infradead.org> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Wed, Oct 27, 2010 at 05:05:30AM -0400, Christoph Hellwig wrote: > > @@ -537,15 +545,10 @@ void evict_inodes(struct super_block *sb) > > } > > > > inode->i_state |= I_FREEING; > > - if (!(inode->i_state & (I_DIRTY | I_SYNC))) > > - percpu_counter_dec(&nr_inodes_unused); > > + inode_lru_list_del(inode); > > spin_unlock(&inode->i_lock); > > > > - /* > > - * Move the inode off the IO lists and LRU once I_FREEING is > > - * set so that it won't get moved back on there if it is dirty. > > - */ > > - list_move(&inode->i_lru, &dispose); > > + list_add(&inode->i_lru, &dispose); > > } > > spin_unlock(&inode_lock); > > > > @@ -582,15 +585,10 @@ int invalidate_inodes(struct super_block *sb) > > } > > > > inode->i_state |= I_FREEING; > > - if (!(inode->i_state & (I_DIRTY | I_SYNC))) > > - percpu_counter_dec(&nr_inodes_unused); > > + inode_lru_list_del(inode); > > spin_unlock(&inode->i_lock); > > with this scheme we now decrement nr_inodes_unused twice - once in > invalidate_inodes/evict_inodes and once in dispose_one_inode. That doesn't happen because the counter is only modified when the inode is moved on/off the list and there are checks to avoid removing an inode that is not on the list. Also, the inode is not removed from the LRU in dispose_one_inode - it is always done when the inode is marked I_FREEING while the i_lock is held before calling dispose_one_inode(). Basically I wanted to remove the strange "inode is not on the LRU if it is dirty or under writeback" accounting checks and make the accounting symmetric with adding/removing the inodes from the LRU. These are protected by list_empty() checks, so should always end up with the correct accounting. hence the only special case now is prune_icache() which already holds the inode_lru_lock() so can't call the helper. Besides, we don't need any checks there because we know the inode is on the LRU already.... Cheers, Dave. -- Dave Chinner david@fromorbit.com