From mboxrd@z Thu Jan 1 00:00:00 1970 From: Robert Hancock Subject: Re: Kernel Summit request for Discussion of future of ATA (libata) and IDE Date: Tue, 05 Aug 2008 20:30:01 -0600 Message-ID: <48990CA9.2060107@shaw.ca> References: <48976168.3020804@shaw.ca> <20080804205508.20a3f917@lxorguk.ukuu.org.uk> <48977200.3050307@shaw.ca> <20080804220619.76b94ceb@lxorguk.ukuu.org.uk> <4898EE70.1030604@shaw.ca> <4898F3EA.7000507@gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: Received: from idcmail-mo2no.shaw.ca ([64.59.134.9]:34969 "EHLO pd5mo1no-dmz.prod.shaw.ca" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1752817AbYHFCaI (ORCPT ); Tue, 5 Aug 2008 22:30:08 -0400 In-Reply-To: <4898F3EA.7000507@gmail.com> Sender: linux-ide-owner@vger.kernel.org List-Id: linux-ide@vger.kernel.org To: Tejun Heo Cc: Alan Cox , Bartlomiej Zolnierkiewicz , James Bottomley , linux-kernel , linux-ide , Jeff Garzik Tejun Heo wrote: > Robert Hancock wrote: >> Here's my first cut at it. Compile tested only. This sets most controllers >> to use 32-bit PIO except for those which could potentially be on a real ISA >> or other 16-bit bus. It's a bit non-obvious what to do with some of the >> drivers, so input is welcome. >> >> This implementation doesn't check the ata_id_has_dword_io at all, since it >> would only make a difference on controllers where we don't really want to >> use it anyway. >> >> It seems like regardless of whether we do 32-bit by default or not the 32-bit >> data_xfer function should be added to libata core as we have several drivers >> which duplicate the same code currently.. > > Great, just some minor nitpicks as I don't have much idea about 16 bit ones. > >> +unsigned int ata_sff_data_xfer(struct ata_device *dev, unsigned char *buf, >> + unsigned int buflen, int rw) >> +{ >> + struct ata_port *ap = dev->link->ap; >> + void __iomem *data_addr = ap->ioaddr.data_addr; >> + unsigned int words = buflen >> 2; > > dwords maybe? > >> + unsigned int slop = buflen & 3; >> + >> + /* Transfer multiple of 4 bytes */ >> + if (rw == READ) >> + ioread32_rep(data_addr, buf, words); >> + else >> + iowrite32_rep(data_addr, buf, words); >> + >> + /* Transfer trailing 1 byte, if any. */ > > 1byte? Yeah, those are both leftovers from the 16-bit code. I'll fix them up in the next version, if the approach looks good.. > > Thanks. >