From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-oi0-x230.google.com (mail-oi0-x230.google.com [IPv6:2607:f8b0:4003:c06::230]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by lists.ozlabs.org (Postfix) with ESMTPS id 0B16A1A04B8 for ; Fri, 26 Feb 2016 21:05:03 +1100 (AEDT) Received: by mail-oi0-x230.google.com with SMTP id m82so58739666oif.1 for ; Fri, 26 Feb 2016 02:05:02 -0800 (PST) Date: Fri, 26 Feb 2016 02:04:49 -0800 (PST) From: Hugh Dickins To: "Aneesh Kumar K.V" cc: Michael Ellerman , Paul Mackerras , linuxppc-dev@lists.ozlabs.org, linux-mm@kvack.org Subject: Re: Problems with swapping in v4.5-rc on POWER In-Reply-To: Message-ID: References: <877fhttmr1.fsf@linux.vnet.ibm.com> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , On Thu, 25 Feb 2016, Hugh Dickins wrote: > On Wed, 24 Feb 2016, Hugh Dickins wrote: > > On Thu, 25 Feb 2016, Aneesh Kumar K.V wrote: > > > > > > Can you test the impact of the merge listed below ?(ie, revert the merge and see if > > > we can reproduce and also verify with merge applied). This will give us a > > > set of commits to look closer. We had quiet a lot of page table > > > related changes going in this merge window. > > > > > > f689b742f217b2ffe7 ("Pull powerpc updates from Michael Ellerman:") > > > > > > That is the merge commit that added _PAGE_PTE. > > > > Another experiment running on it at the moment, I'd like to give that > > a few more hours, and then will try the revert you suggest. But does > > that merge revert cleanly, did you try? I'm afraid of interactions, > > whether obvious or subtle, with the THP refcounting rework. Oh, since > > I don't have THP configured on, maybe I can ignore any issues from that. > > That revert worked painlessly, only a very few and simple conflicts, > I ran that under load for 12 hours, no problem seen. > > I've now checked out an f689b742 tree and started on that, just to > confirm that it fails fairly quickly I hope; and will then proceed > to git bisect, giving that as bad and 37cea93b as good. > > Given the uncertainty of whether 12 hours is really long enough to be > sure, and perhaps difficulties along the way, I don't rate my chances > of a reliable bisection higher than 60%, but we'll see. I'm sure you won't want a breathless report from me on each bisection step, but I ought to report that: contrary to our expectations, the f689b742 survived without error for 12 hours, so appears to be good. I'll bisect between there and v4.5-rc1. Hugh