From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932942AbcFJUo1 (ORCPT ); Fri, 10 Jun 2016 16:44:27 -0400 Received: from mail-lf0-f65.google.com ([209.85.215.65]:35754 "EHLO mail-lf0-f65.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932252AbcFJUoZ (ORCPT ); Fri, 10 Jun 2016 16:44:25 -0400 Date: Fri, 10 Jun 2016 22:44:19 +0200 From: Krzysztof Kozlowski To: "Luis R. Rodriguez" Cc: Krzysztof Kozlowski , Andrew Morton , hch@infradead.org, Bartlomiej Zolnierkiewicz , Jonathan Corbet , Russell King , Robin Murphy , Marek Szyprowski , Doug Anderson , Will Deacon , Joerg Roedel , Christian Borntraeger , Zhen Lei , "Michael S. Tsirkin" , Andy Lutomirski , linux-doc@vger.kernel.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH v4 01/44] dma-mapping: Use unsigned long for dma_attrs Message-ID: <20160610204419.GA4239@kozik-lap> References: <1465553521-27303-1-git-send-email-k.kozlowski@samsung.com> <1465553521-27303-2-git-send-email-k.kozlowski@samsung.com> <20160610144947.GD11948@wotan.suse.de> <20160610201600.GB2235@kozik-lap> <20160610202347.GE11948@wotan.suse.de> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline In-Reply-To: <20160610202347.GE11948@wotan.suse.de> 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 On Fri, Jun 10, 2016 at 10:23:47PM +0200, Luis R. Rodriguez wrote: > On Fri, Jun 10, 2016 at 10:16:00PM +0200, Krzysztof Kozlowski wrote: > > On Fri, Jun 10, 2016 at 04:49:47PM +0200, Luis R. Rodriguez wrote: > > > On Fri, Jun 10, 2016 at 12:11:18PM +0200, Krzysztof Kozlowski wrote: > > > > The dma-mapping core and the implementations do not change the > > > > DMA attributes passed by pointer. Thus the pointer can point to const > > > > data. However the attributes do not have to be a bitfield. Instead > > > > unsigned long will do fine: > > > > > > > > 1. This is just simpler. Both in terms of reading the code and setting > > > > attributes. Instead of initializing local attributes on the stack > > > > and passing pointer to it to dma_set_attr(), just set the bits. > > > > > > > > 2. It brings safeness and checking for const correctness because the > > > > attributes are passed by value. > > > > > > Do we not expect the number of argument to grow ? This "cleanup" would > > > do away with such possibilities, and then require adding the API later, > > > and this requiring a full set of collateral evolutions again when this > > > is needed. What was the original motivation for using this instead of > > > the approach you are suggesting ? > > > > What do you mean by "possibilities of argument to grow"? Something like > > adding new members to "struct dma_attrs" and changing its meaning? > > Yup that. > > > I think such growth is still constrained - you cannot put there anything > > without changing the meaning of the argument. > > Obviously, however it would mean no needed collateral evolutions, > just an extension to the struct and drivers that use the new member > can make use of it. For parts of the API, there is still possibility of adding new layer of wrapping, just like it was done with dma_map_single_attrs(): #define dma_map_single(d, a, s, r) dma_map_single_attrs(d, a, s, r, NULL) For the dma_map_ops not... > > The dma-attrs in current form were added around 2008 in 74bc7ceebfa1 > > ("dma: add dma_*map*_attrs() interfaces"), I think. Since that time, for > > example, the dma_map_*_attrs() did not change. > > So we don't expect this to change either? I do not know, I am not aware of planned changes to that. > > > > If the concern is the const data, why not require const struct dma_attr > > > for the APIs that we know can and should use const ? > > > > The const is one concern. Complicated (more than expected) usage of dma > > attributes by the caller is second. > > > > Switching it to const would also reduce the possibilities of API > > extension. > > My point was that const can be used for only APIs that we are sure of > that need it. As of now, dma_attrs should be const everywhere. That would be almost the same patchset as current one. If you consider extending the dma_attrs to something new and not yet known, then how will differentiate between cases when 'const' is needed for sure? I understand your concern. Sticking to current API for that reason might be a good defensive API programming... or might be way of keeping this function prototype for long... Best regards, Krzysztof