From: linux@arm.linux.org.uk (Russell King - ARM Linux) Date: Mon, 25 Nov 2013 19:09:36 +0000 Subject: [PATCH 11/31] dma: add channel request API that supports deferred probe In-Reply-To: References: <528E4F55.9070204@wwwdotorg.org> <528F95A9.2050305@wwwdotorg.org> <528F9DF9.6080706@wwwdotorg.org> <20131123003442.GH16735@n2100.arm.linux.org.uk> <5293883A.8050808@wwwdotorg.org> <20131125180000.GR16735@n2100.arm.linux.org.uk> Message-ID: <20131125190935.GS16735@n2100.arm.linux.org.uk> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org On Mon, Nov 25, 2013 at 10:42:52AM -0800, Dan Williams wrote: > On Mon, Nov 25, 2013 at 10:00 AM, Russell King - ARM Linux > wrote: > > On Mon, Nov 25, 2013 at 09:45:18AM -0800, Dan Williams wrote: > >> Regardless of whether the driver was dma_request_channel as a > >> fallback, it will currently use NULL to indicate an allocation > >> failure. > > > > At the moment, dma_request_slave_channel()'s return values are valid > > pointer or NULL. I'd suggest as that's how it's been established, > > that function is left alone - changing the return values in this kind > > of invisible way is Really Bad(tm) because you can't check that it's > > right in any future users - unless you're prepared to review every > > single new user of this function for the next few years or more. > > > > Instead, leaving it as is and introducing a new function name with > > the different return error method is the right way to do this. > > I was attempting to tap the brakes on the number of request_channel > apis in circulation. > dma_request_channel > dma_request_slave_channel > dma_request_slave_channel_compat > ... and now dma_request_slave_channel_reason > > ...but given we already have things like samsung_dmadev_request() in > the wild, tracking correct usage going forward will be an uphill > battle. We can do this the other way round. Introduce the new api > and delete the old one (after some time). Yes, new and deprecate the old is the right way to do this. > Can we make the new api not pass the 'defer' flag. You mention it is > for compat reasons, but since there are no users I'm not seeing how > there can be compatibility concerns. Can you elaborate? Yes, I agree. I don't think there's a compatibility concern here - with the old interface, a non-present resource would be indicated as "does not exist". If we convert any error pointer including the one representing -EPROBE_DEFER to the "does not exist" representation, we're not changing the behaviour.