From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ian Kent Subject: Re: [PATCH 0/5] RCU-walk support for autofs Date: Wed, 20 Aug 2014 17:52:47 +0800 Message-ID: <1408528367.11028.44.camel@perseus.themaw.net> References: <20140818062844.9860.43007.stgit@notabene.brown> <1408350340.2533.3.camel@perseus.fritz.box> <1408442547.11314.9.camel@perseus.themaw.net> <20140819211654.5ec3853a@notabene.brown> <1408451815.11314.15.camel@perseus.themaw.net> <20140820131352.2aad3ca5@notabene.brown> <1408506167.11028.32.camel@perseus.themaw.net> Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Return-path: DKIM-Signature: v=1; a=rsa-sha1; c=relaxed/relaxed; d=themaw.net; h= message-id:subject:from:to:cc:date:in-reply-to:references :content-type:mime-version:content-transfer-encoding; s=mesmtp; bh=KzggeGkN4B5JGRcQcgULJNZefHw=; b=Mec5Zvs5OVx191Hr6Ic/9qp9ehD6 Ze/55fNzg5xPGUmqv+mOAxQta7rDDAI/uB0dlO5bALsQQ5MxbA7MqY7OxrNfiFRc 4S/lwKsugapjRr6ONH3JD/zOqzTm8WBktkeBd3eAQkxzqWniqLIHz/MjMtYtaHEd zl7SxcftxNT1GNs= DKIM-Signature: v=1; a=rsa-sha1; c=relaxed/relaxed; d= messagingengine.com; h=message-id:subject:from:to:cc:date :in-reply-to:references:content-type:mime-version :content-transfer-encoding; s=smtpout; bh=KzggeGkN4B5JGRcQcgULJN ZefHw=; b=lxJDG3gc9YEClPN+aX1SXS8ajoKRHLqE9PQN6HhHohJ4wFdE0mjqZ5 FqMib8PBtWOwXljz2XY3dYo0MAPofqUtCo3L/1CgJdsiTiFwj14eT1PgcAdvLKMT fkU63G5J+1ih5ESH4vRpASXoxb7YZy3Iz95v1zMB9ym/d7AyMxoX4= In-Reply-To: <1408506167.11028.32.camel@perseus.themaw.net> Sender: autofs-owner@vger.kernel.org List-ID: Content-Type: text/plain; charset="us-ascii" To: NeilBrown Cc: autofs@vger.kernel.org, linux-kernel@vger.kernel.org On Wed, 2014-08-20 at 11:42 +0800, Ian Kent wrote: > Great minds think alike. > I have basically the exact same patch and I'm testing now. > It probably should be folded into "autofs4: avoid taking fs_lock during > rcu-walk". > > It takes ages though, the test has two configurations to test and each > one takes about 70 minutes. > > fyi, my criteria for the test is if it runs through to completion once > then that's essentially a pass. If not then there is a race that needs > to be fixed before merge. If it fails on runs two or three then there's > a hard to find race that needs attention but is not serious enough to > block merging. I usually stop after three runs but if fails are seen > later then, they too need to be resolved in the fullness of time. > > The reason for this is that in heavy use sites we rarely see more than > about three processes concurrently accessing mount points and the test > run uses 10 client instances (currently on a 6 CPU machine) that all > access the mount tree at the same time so that should be somewhat more > pressure than is seen under heavy use. > > And indeed this rule of has proven reliable so far. I've run my test three times now. The first time I had debug logging enabled. I disabled it for the second and third runs as that sometimes makes a difference. All three runs went through fine. The only thing I noticed was that the test took a little longer (two and a half to three minutes longer, of 70) to complete than without the patch series but I've seen that sort of variation in kernels at different times in the past. So I think the series is fine and should be merged. Please feel free to add any of my acked, reviewed and tested by to them. Ian