From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S935086AbWKXVvI (ORCPT ); Fri, 24 Nov 2006 16:51:08 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S935089AbWKXVvI (ORCPT ); Fri, 24 Nov 2006 16:51:08 -0500 Received: from 74-93-104-97-Washington.hfc.comcastbusiness.net ([74.93.104.97]:36051 "EHLO sunset.davemloft.net") by vger.kernel.org with ESMTP id S935086AbWKXVvG (ORCPT ); Fri, 24 Nov 2006 16:51:06 -0500 Date: Fri, 24 Nov 2006 13:51:15 -0800 (PST) Message-Id: <20061124.135115.26533586.davem@davemloft.net> To: ralf@linux-mips.org Cc: linux-kernel@vger.kernel.org Subject: Re: RFC: Removal of pci_dma_* functions From: David Miller In-Reply-To: <20061124170359.GA9355@linux-mips.org> References: <20061124170359.GA9355@linux-mips.org> X-Mailer: Mew version 4.2 on Emacs 21.4 / Mule 5.0 (SAKAKI) Mime-Version: 1.0 Content-Type: Text/Plain; charset=us-ascii Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org From: Ralf Baechle Date: Fri, 24 Nov 2006 17:03:59 +0000 > A grep for > > pci_dac_dma_supported() > pci_dac_page_to_dma() > pci_dac_dma_to_page() > pci_dac_dma_to_offset() > pci_dac_dma_sync_single_for_cpu() > pci_dac_dma_sync_single_for_device() > > reveals that only a some platforms implement these functions and there seems > to be no single user. Any objections against removing? Or why did we ever > add them anyay ... They were meant for things like those Dolphin clustering cards that can map arbitrary amounts of remote memory on the local system. Unfortunately, none of the drivers for these things are upstream, but I can nearly guarentee you that those drivers use these interfaces because they map so much memory that they cannot tolerate the limitations imposed by IOMMUs, even the ones with 2GB mapping windows like on sparc64.