From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754786AbbHJQgp (ORCPT ); Mon, 10 Aug 2015 12:36:45 -0400 Received: from mail-yk0-f178.google.com ([209.85.160.178]:35783 "EHLO mail-yk0-f178.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754494AbbHJQgk (ORCPT ); Mon, 10 Aug 2015 12:36:40 -0400 Date: Mon, 10 Aug 2015 12:36:38 -0400 From: Tejun Heo To: Guenter Roeck Cc: Christoph Lameter , linux-mm@kvack.org, linux-kernel@vger.kernel.org Subject: Re: [RFC PATCH] percpu: Prevent endless loop if there is no unallocated region Message-ID: <20150810163638.GC23408@mtj.duckdns.org> References: <1439122659-31442-1-git-send-email-linux@roeck-us.net> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <1439122659-31442-1-git-send-email-linux@roeck-us.net> User-Agent: Mutt/1.5.23 (2014-03-12) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hello, On Sun, Aug 09, 2015 at 05:17:39AM -0700, Guenter Roeck wrote: > Qemu tests with unicore32 show memory management code entering an endless > loop in pcpu_alloc(). Bisect points to commit a93ace487a33 ("percpu: move > region iterations out of pcpu_[de]populate_chunk()"). Code analysis > identifies the following relevant changes. > > - rs = page_start; > - pcpu_next_pop(chunk, &rs, &re, page_end); > - > - if (rs != page_start || re != page_end) { > + pcpu_for_each_unpop_region(chunk, rs, re, page_start, page_end) { > > For unicore32, values were page_start==0, page_end==1, rs==0, re==1. > This worked fine with the old code. With the new code, however, the loop > is always entered. Debugging information added into the loop shows > an endless repetition of > > in loop chunk c5c53100 populated 0xff rs 1 re 2 page start 0 page end 1 > in loop chunk c5c53100 populated 0xff rs 1 re 2 page start 0 page end 1 > in loop chunk c5c53100 populated 0xff rs 1 re 2 page start 0 page end 1 > in loop chunk c5c53100 populated 0xff rs 1 re 2 page start 0 page end 1 That's a bug in the find bit functions in unicore32. If @offset >= @end, it should return @end, not @offset. Thanks. -- tejun