Re: [PATCH v5 2/6] phy: cdns-dphy: Add Rx support

[Paul Kocialkowski <paul.kocialkowski@bootlin.com>]

[-- Attachment #1: Type: text/plain, Size: 3641 bytes --]

Hi,

On Fri 08 Oct 21, 13:27, Laurent Pinchart wrote:
> Hi Pratyush,
> 
> On Thu, Oct 07, 2021 at 05:44:38PM +0530, Pratyush Yadav wrote:
> > On 07/10/21 03:10AM, Laurent Pinchart wrote:
> > > On Fri, Oct 01, 2021 at 11:53:16AM +0530, Vinod Koul wrote:
> > > > On 17-09-21, 22:58, Pratyush Yadav wrote:
> > > > > On 16/09/21 12:22PM, Paul Kocialkowski wrote:
> > > > > > On Fri 03 Sep 21, 00:25, Pratyush Yadav wrote:
> > > > > > > The Cadence DPHY can be used to receive image data over the CSI-2
> > > > > > > protocol. Add support for Rx mode. The programming sequence differs from
> > > > > > > the Tx mode so it is added as a separate set of hooks to isolate the two
> > > > > > > paths. The mode in which the DPHY has to be used is selected based on
> > > > > > > the compatible.
> > > > > > 
> > > > > > I just realized that I didn't follow-up on a previous revision on the debate
> > > > > > about using the phy sub-mode to distinguish between rx/tx.
> > > > > > 
> > > > > > I see that you've been using a dedicated compatible, but I'm not sure this is a
> > > > > > good fit either. My understanding is that the compatible should describe a group
> > > > > > of register-compatible revisions of a hardware component, not how the hardware
> > > > > > is used specifically. I guess the distinction between rx/tx falls under
> > > > > > the latter rather than the former.
> > > > > 
> > > > > I am not sure if that is the case. For example, we use "ti,am654-ospi" 
> > > > > for Cadence Quadspi controller. The default compatible, "cdns,qspi-nor", 
> > > > > only supports Quad SPI (4 lines). The "ti,am654-ospi" compatible also 
> > > > > supports Octal SPI (8 lines).
> > > > 
> > > > Those are hardware defaults right?
> > > > 
> > > > > In addition, I feel like the Rx DPHY is almost a different type of 
> > > > > device from a Tx DPHY. The programming sequence is completely different, 
> > > > 
> > > > Is that due to direction or something else..?
> > > > 
> > > > > the clocks required are different, etc. So I think using a different 
> > > > > compatible for Rx mode makes sense.
> > > > 
> > > > Is the underlaying IP not capable of both TX and RX and in the specific
> > > > situations you are using it as TX and RX.
> > > > 
> > > > I am okay that default being TX but you can use Paul's approach of
> > > > direction with this to make it better proposal
> > > 
> > > 
> > > Given that the RX and TX implementations are very different (it's not a
> > > matter of selecting a mode at runtime), I'm actually tempted to
> > > recommend having two drivers, one for the RX PHY and one for the TX PHY.
> > > This can only be done with two different compatible strings, which I
> > > think would be a better approach.
> > 
> > FWIW, I think having different drivers would certainly make things 
> > easier to maintain.
> 
> I'm sorry for not having recommended this in the first place.
> 
> Any objection from anyone against going in this direction ?

So apparently there is not a single register that is shared between rx and tx
and clocks are not the same either so it feels to me like a separate driver
would be legit. This looks like two distinct IPs sharing the same base address.

Cheers,

Paul

> > > It's unfortunate that the original compatible string didn't contain
> > > "tx". We could rename it and keep the old one in the driver for backward
> > > compatibility, making things cleaner going forward.
> 
> -- 
> Regards,
> 
> Laurent Pinchart

-- 
Paul Kocialkowski, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]