[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <yw1xd1oshgvq.fsf@unicorn.mansr.com>
Date: Wed, 11 May 2016 14:19:05 +0100
From: Måns Rullgård <mans@...sr.com>
To: Arnd Bergmann <arnd@...db.de>
Cc: Tejun Heo <tj@...nel.org>,
Andy Shevchenko <andriy.shevchenko@...ux.intel.com>,
linux-ide@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] ata: dwc: add DMADEVICES dependency
Arnd Bergmann <arnd@...db.de> writes:
> On Wednesday 11 May 2016 13:57:08 Måns Rullgård wrote:
>> > diff --git a/drivers/ata/Kconfig b/drivers/ata/Kconfig
>> > index 41b0725e58ad..8f7a4a4d2566 100644
>> > --- a/drivers/ata/Kconfig
>> > +++ b/drivers/ata/Kconfig
>> > @@ -322,7 +322,7 @@ config SATA_DWC
>> >
>> > config SATA_DWC_OLD_DMA
>> > bool "Support old device trees"
>> > - depends on SATA_DWC
>> > + depends on SATA_DWC && DMADEVICES
>> > select DW_DMAC_CORE
>> > default y if 460EX
>> > help
>> > --
>>
>> Isn't the proper fix here to have DW_DMAC_CORE select DMADEVICES?
>
> No, all dmaengine drivers depend on DMADEVICES, so that would
> cause a circular dependency.
>
> The normal behavior for DMA slave drivers is to use the generic
> DMA API and have a dependency on DMADEVICES but not on a particular
> DMA engine implementation. In this case, the driver uses an exported
> interface of the dw_dma driver directly, so we also need to select
> that one, but it doesn't mean we can skip the normal dependency.
OK, but then shouldn't the main SATA_DWC have this dependency since it
uses the DMA API?
--
Måns Rullgård
Powered by blists - more mailing lists