From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751542AbcAXJli (ORCPT ); Sun, 24 Jan 2016 04:41:38 -0500 Received: from mx2.suse.de ([195.135.220.15]:34297 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751233AbcAXJlc (ORCPT ); Sun, 24 Jan 2016 04:41:32 -0500 Date: Sun, 24 Jan 2016 10:40:56 +0100 From: Borislav Petkov To: Dan Williams Cc: Sudip Mukherjee , Christoph Hellwig , "H. Peter Anvin" , Ingo Molnar , Dave Hansen , Rik van Riel , Mel Gorman , Jerome Glisse , markk@clara.co.uk, Joerg Roedel , lkml , linux-mm@kvack.org Subject: Re: [BUG] Devices breaking due to CONFIG_ZONE_DEVICE Message-ID: <20160124094056.GA27266@pd.tnic> References: <20160123044643.GA3709@sudip-pc> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline Content-Transfer-Encoding: 8bit In-Reply-To: User-Agent: Mutt/1.5.24 (2015-08-30) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org + linux-mm On Fri, Jan 22, 2016 at 10:15:17PM -0800, Dan Williams wrote: > On Fri, Jan 22, 2016 at 9:47 PM, Dan Williams wrote: > > On Fri, Jan 22, 2016 at 8:46 PM, Sudip Mukherjee > > wrote: > >> Hi All, > >> Commit 033fbae988fc ("mm: ZONE_DEVICE for "device memory"") has > >> introduced CONFIG_ZONE_DEVICE while sacrificing CONFIG_ZONE_DMA. > >> Distributions like Ubuntu has started enabling CONFIG_ZONE_DEVICE and > >> thus breaking parallel port. Please have a look at > >> https://bugzilla.kernel.org/show_bug.cgi?id=110931 for the bug report. > >> > >> Apart from parallel port I can see some sound drivers will also break. > >> > >> Now what is the possible solution for this? > > > > The tradeoff here is enabling direct-I/O for persistent memory vs > > support for legacy devices. > > > > One possible solution is to alias ZONE_DMA and ZONE_DEVICE. At early > > boot if pmem is detected disable these legacy devices, or the reverse > > disable DMA to persistent memory if a legacy device is detected. The > > latter is a bit harder to do as I think we would want to make the > > decision early during memory init before we would know if any parallel > > ports or ISA sound cards are present. > > ...another option that might be cleaner is to teach GFP_DMA to get > memory from a different mechanism. I.e. don't use the mm-zone > infrastructure to organize that small 16MB pool of memory. -- Regards/Gruss, Boris. SUSE Linux GmbH, GF: Felix Imendörffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nürnberg) --