Re: [PATCH v2 3/4] ASoC: dt-bindings: fsl-sai: allow only one dma-names

[Conor Dooley <conor@xxxxxxxxxx>]

On Fri, Mar 01, 2024 at 11:46:20AM -0500, Frank Li wrote:
> On Fri, Mar 01, 2024 at 04:05:25PM +0000, Conor Dooley wrote:
> > On Thu, Feb 29, 2024 at 02:38:24PM -0500, Frank Li wrote:
> > > On Thu, Feb 29, 2024 at 06:57:29PM +0000, Conor Dooley wrote:
> > > > On Thu, Feb 29, 2024 at 06:55:58PM +0000, Conor Dooley wrote:
> > > > > On Tue, Feb 27, 2024 at 03:54:11PM -0500, Frank Li wrote:
> > > > > > Some sai only connect one direction. So allow only "rx" or "tx" for
> > > > > > dma-names.
> > > > > 
> > > > > Which sai? Can you restrict this per compatible please, so that someone
> > > > > cannot add 2 dmas for ones where only the tx is supported.
> > > > > 
> > > > > |  dmas:
> > > > > |    minItems: 1
> > > > > |    items:
> > > > > |      - description: DMA controller phandle and request line for RX
> > > > > |      - description: DMA controller phandle and request line for TX
> > > > > 
> > > > > The binding already allows only one, but it documents that the first dma
> > > > > is always the RX dma, and that doesn't change with this patch..
> > > > 
> > > > I said "doesn't change" - but I don't think you can change this
> > > > trivially, as something could rely on the first dma being the rx one.
> > > > You'd have to check that there is nothing using these using indices
> > > > rather than names before making any changes here.
> > > 
> > > Linux driver and dts with tx only work well. Only issue is dtb_check will
> > > report error. I want to eliminate these DTB_CHECK warning.
> > 
> > Linux is not the only user of these bindings, citing linux as your
> > evidence here is only sufficient if no other users exist. Do they?
> 
> But, 'dmas' should be common property for all these bindings? I don't think
> they use 'descriptions:' property, which should guide dts writer to write
> dts file. actually words 'DMA controller phandle and request line' just
> nonsense words. let 'regs', it'd better descript at 'reg-names' instead
> of 'regs' if reg-names exist. Only meansful words is "RX" and "TX", which
> already show at "dma-names".

None of this matters. If there's a documented order for these, which
there is, software is not obligated to use the names and can rely on the
order alone. You need to check that there are no other users which will
be broken by your proposed change.

> > > And it also reasonable, only rx or tx for a special SAI.
> > > 
> > > Can we remove 'description'? dmas should already descripted at common place
> > > and 'RX' and 'TX' are listed at 'dma-names'
> > 
> > Removing the description has the same problem. The existing binding has
> > set a fixed order that you now want to make flexible.
> 
> Actually original set minItems: is 1, which means allow 1 channel. but
> set items to force two channel. 
> 
> Does it work
> 
> oneOf:
>   items: 
>   	- description: TX
> 	- description: RX
>   items:
> 	- description: TX
>   items:
> 	- description: RX
> 
> > 
> > Thanks,
> > Conor.
> 
>

Attachment: signature.asc
Description: PGP signature