Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S933060Ab2EKUbX (ORCPT ); Fri, 11 May 2012 16:31:23 -0400 Received: from mga14.intel.com ([143.182.124.37]:15055 "EHLO mga14.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932157Ab2EKUbW convert rfc822-to-8bit (ORCPT ); Fri, 11 May 2012 16:31:22 -0400 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="4.71,315,1320652800"; d="scan'208";a="99104894" From: "Liu, Jinsong" To: Konrad Rzeszutek Wilk CC: "'xen-devel@lists.xensource.com'" , "'linux-kernel@vger.kernel.org'" Subject: RE: [Xen-devel] [PATCH 3/3] Xen physical cpus interface Thread-Topic: [Xen-devel] [PATCH 3/3] Xen physical cpus interface Thread-Index: AQHNLr1B1b8pDRiwSFiwkpXixs4X8JbDIYNQgAHTcp+AABcXMA== Date: Fri, 11 May 2012 20:31:15 +0000 Message-ID: References: <20120420192439.GA32170@phenom.dumpdata.com> <20120510145745.GO26152@phenom.dumpdata.com> <20120511142758.GA29677@phenom.dumpdata.com> <20120511190041.GA3785@phenom.dumpdata.com> In-Reply-To: <20120511190041.GA3785@phenom.dumpdata.com> Accept-Language: en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: x-originating-ip: [10.239.127.40] Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 8BIT MIME-Version: 1.0 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 3254 Lines: 88 Konrad Rzeszutek Wilk wrote: > On Fri, May 11, 2012 at 06:04:21PM +0000, Liu, Jinsong wrote: >> Konrad Rzeszutek Wilk wrote: >>> On Fri, May 11, 2012 at 01:12:13PM +0000, Liu, Jinsong wrote: >>>> Liu, Jinsong wrote: >>>>> Just notice your reply (so quick :) >>>>> >>>>> Agree and will update later, except 1 concern below. >>>>> >>>>> Konrad Rzeszutek Wilk wrote: >>>>>>> >>>>>>> Hmm, it's good if it's convenient to do it automatically via >>>>>>> dev->release. However, dev container (pcpu) would be free at >>>>>>> some other error cases, so I prefer do it 'manually'. >>>>>> >>>>>> You could also call pcpu_release(..) to do it manually. >>>>>> >>>>> >>>>> that means kfree(pcpu) would be done twice at some error cases, >>>>> do you think it really good? >>>>> >>>> >>>> Ping. >>>> >>>> I think error recovery should be kept inside error logic level >>>> itself, if try to recover upper level error would bring trouble. >>>> >>>> In our example, there are 2 logic levels: >>>> pcpu level (as container), and dev level (subfield used for sys) >>> >>> So you need to untangle free_pcpu from doing both. Meaning one does >>> the SysFS and the other deals with free-ing the structure and >>> removing itself from the list. >>> >> >> free_cpu is very samll, just consist of the 2 parts your said: >> * pcpu_sys_remove() deal with sysfs >> * list_del/kfree(pcpu) deal with pcpu >> >>> >>>> dev->release should only recover error occurred at dev/sys level, >>>> and the pcpu error should be recovered at pcpu level. >>>> >>>> If dev->release try to recover its container pcpu level error, like >>>> list_del/kfree(pcpu), it would make confusing. i.e., considering >>>> pcpu_sys_create(), 2 error cases: device_register fail, and >>>> device_create_file fail --> how can the caller decide kfree(pcpu) >>>> or not? >>> >>> Then you should free it manually. But you can do this by a wrapper >>> function: >>> >>> __pcpu_release(..) { >>> .. >>> /* Does the removing itself from the list and kfree the pcpu */ } >>> pcpu_release(..) { >>> struct pcpcu *p= container_of(..) >>> __pcpu_release(p); >>> } >>> >>> dev->release = &pcpu_release; >>> >> >> Too weird way. If we want to release dev itself it's good to use >> dev->release, but for pcpu I doubt it. (consider the example I gave >> --> why we create issue (it maybe solved in weird method I agree), >> just for using dev->release?) >> >> In kernel many dev->release keep NULL. >> An example of using dev->release is cpu/mcheck/mce.c --> >> mce_device_release(), it *just* deal dev itself. > > OK? I am not sure what are we arguing here anymore? > I think using 'kfree(pcpu)' on the error paths (as long as it is > done before device_register) is OK. I think that seperating > the SysFS deletion from the pcpu deletion should be done to > avoid races. Perhaps the SysFS deletion function should also > remove the pcpu from the list. How about static array pcpu[NR_CPUS]? It seems solve all issues we argued :) Thanks, Jinsong-- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/