From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1762174AbXGDN7t (ORCPT ); Wed, 4 Jul 2007 09:59:49 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1760547AbXGDN7k (ORCPT ); Wed, 4 Jul 2007 09:59:40 -0400 Received: from mtagate7.de.ibm.com ([195.212.29.156]:56470 "EHLO mtagate7.de.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1760047AbXGDN7j (ORCPT ); Wed, 4 Jul 2007 09:59:39 -0400 From: Hoang-Nam Nguyen To: jim.houston@ccur.com Subject: Re: idr_get_new_above() limitation? Date: Wed, 4 Jul 2007 16:11:29 +0200 User-Agent: KMail/1.8.2 Cc: linux-kernel@vger.kernel.org, linuxppc-dev@ozlabs.org, openib-general@openib.org, Stefan Roscher , raisch@de.ibm.com, Andrew Morton References: <200707021919.27251.hnguyen@linux.vnet.ibm.com> <1183422700.3130.27.camel@localhost.localdomain> In-Reply-To: <1183422700.3130.27.camel@localhost.localdomain> MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 7bit Content-Disposition: inline Message-Id: <200707041611.30056.hnguyen@linux.vnet.ibm.com> Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Tuesday 03 July 2007 02:31, Jim Houston wrote: > The problem is in idr_get_new_above_int() in the loop which > adds new layers to the top of the radix tree. It is failing > the "layers < (MAX_LEVEL - 1)" test. It doesn't allocate the > new layer but still calls sub_alloc() which relies on having > the new layer properly constructed. I believe that it is > allocating the slot which corresponds to id = 0. Hi Jim, Thanks for your quick reply. Yes, I realized that while condition too and have tried with a tiny change like (layers < MAX_LEVEL), but without success with idr_find(), even though 6 layers were created and the object was added at proper location. After several debug cycles I think to find the root cause in the if-condition in idr_find(): void *idr_find(struct idr *idp, int id) { int n; struct idr_layer *p; n = idp->layers * IDR_BITS; p = idp->top; /* Mask off upper bits we don't use for the search. */ id &= MAX_ID_MASK; if (id >= (1 << n)) return NULL; ... } Since idp->layers is now 6, n is equal 36, ie out of 32-bit-range, and therefore (1 << n) = (1 << 36) = 0 causing that if-cond to be true ie idr_find() fails. Replacing that if-line by if ((long)id >= (1L << n)) makes idr_find() working properly until MAX_ID_MASK. Since there are other places to be changed like above as well eg. idr_replace() and because you're creating a patch too, I'm waiting first for your comment. Let me know if you prefer me to send a patch. Regards Nam