Message-ID: <80eb0949508d31a55f2b8ab999210a7c7cd5cbe2.camel@gmail.com>
Date: Tue, 01 Oct 2024 10:23:45 +0200
From: Nuno Sá <noname.nuno@...il.com>
To: Jonathan Cameron <Jonathan.Cameron@...wei.com>, Angelo Dureghello
	 <adureghello@...libre.com>
Cc: Jonathan Cameron <jic23@...nel.org>, Krzysztof Kozlowski
 <krzk@...nel.org>,  Lars-Peter Clausen	 <lars@...afoo.de>, Michael
 Hennerich <Michael.Hennerich@...log.com>, Nuno Sa	 <nuno.sa@...log.com>,
 Rob Herring <robh@...nel.org>, Krzysztof Kozlowski	 <krzk+dt@...nel.org>,
 Conor Dooley <conor+dt@...nel.org>, Olivier Moysan	
 <olivier.moysan@...s.st.com>, linux-iio@...r.kernel.org, 
	linux-kernel@...r.kernel.org, devicetree@...r.kernel.org,
 dlechner@...libre.com
Subject: Re: [PATCH v3 04/10] dt-bindings: iio: dac: ad3552r: add io-backend
 support

On Mon, 2024-09-30 at 16:09 +0100, Jonathan Cameron wrote:
> On Mon, 30 Sep 2024 15:22:01 +0200
> Angelo Dureghello <adureghello@...libre.com> wrote:
> 
> > On 30.09.2024 09:20, Nuno Sá wrote:
> > > On Sun, 2024-09-29 at 11:59 +0100, Jonathan Cameron wrote:  
> > > > On Sat, 28 Sep 2024 14:20:29 +0200
> > > > Krzysztof Kozlowski <krzk@...nel.org> wrote:
> > > >   
> > > > > On 25/09/2024 13:55, Nuno Sá wrote:  
> > > > > > On Wed, 2024-09-25 at 09:22 +0200, Krzysztof Kozlowski wrote:    
> > > > > > > On 24/09/2024 14:27, Nuno Sá wrote:    
> > > > > > > > On Tue, 2024-09-24 at 10:02 +0200, Krzysztof Kozlowski wrote:   
> > > > > > > > > On 23/09/2024 17:50, Angelo Dureghello wrote:    
> > > > > > > > > > Hi Krzysztof,
> > > > > > > > > > 
> > > > > > > > > > On 22/09/24 23:02, Krzysztof Kozlowski wrote:    
> > > > > > > > > > > On Thu, Sep 19, 2024 at 11:20:00AM +0200, Angelo
> > > > > > > > > > > Dureghello
> > > > > > > > > > > wrote:    
> > > > > > > > > > > > From: Angelo Dureghello <adureghello@...libre.com>
> > > > > > > > > > > > 
> > > > > > > > > > > > There is a version AXI DAC IP block (for FPGAs) that
> > > > > > > > > > > > provides
> > > > > > > > > > > > a physical bus for AD3552R and similar chips, and acts
> > > > > > > > > > > > as
> > > > > > > > > > > > an SPI controller.
> > > > > > > > > > > > 
> > > > > > > > > > > > For this case, the binding is modified to include some
> > > > > > > > > > > > additional properties.
> > > > > > > > > > > > 
> > > > > > > > > > > > Signed-off-by: Angelo Dureghello
> > > > > > > > > > > > <adureghello@...libre.com>
> > > > > > > > > > > > ---
> > > > > > > > > > > >   .../devicetree/bindings/iio/dac/adi,ad3552r.yaml   |
> > > > > > > > > > > > 42
> > > > > > > > > > > > ++++++++++++++++++++++
> > > > > > > > > > > >   1 file changed, 42 insertions(+)
> > > > > > > > > > > > 
> > > > > > > > > > > > diff --git
> > > > > > > > > > > > a/Documentation/devicetree/bindings/iio/dac/adi,ad3552r.
> > > > > > > > > > > > yaml
> > > > > > > > > > > > b/Documentation/devicetree/bindings/iio/dac/adi,ad3552r.
> > > > > > > > > > > > yaml
> > > > > > > > > > > > index 41fe00034742..aca4a41c2633 100644
> > > > > > > > > > > > ---
> > > > > > > > > > > > a/Documentation/devicetree/bindings/iio/dac/adi,ad3552r.
> > > > > > > > > > > > yaml
> > > > > > > > > > > > +++
> > > > > > > > > > > > b/Documentation/devicetree/bindings/iio/dac/adi,ad3552r.
> > > > > > > > > > > > yaml
> > > > > > > > > > > > @@ -60,6 +60,18 @@ properties:
> > > > > > > > > > > >       $ref: /schemas/types.yaml#/definitions/uint32
> > > > > > > > > > > >       enum: [0, 1, 2, 3]
> > > > > > > > > > > >   
> > > > > > > > > > > > +  io-backends:
> > > > > > > > > > > > +    description: The iio backend reference.
> > > > > > > > > > > > +      An example backend can be found at
> > > > > > > > > > > > +       
> > > > > > > > > > > > https://analogdevicesinc.github.io/hdl/library/axi_ad3552r/index.html
> > > > > > > > > > > > +    maxItems: 1
> > > > > > > > > > > > +
> > > > > > > > > > > > +  adi,synchronous-mode:
> > > > > > > > > > > > +    description: Enable waiting for external
> > > > > > > > > > > > synchronization
> > > > > > > > > > > > signal.
> > > > > > > > > > > > +      Some AXI IP configuration can implement a dual-IP
> > > > > > > > > > > > layout,
> > > > > > > > > > > > with
> > > > > > > > > > > > internal
> > > > > > > > > > > > +      wirings for streaming synchronization.
> > > > > > > > > > > > +    type: boolean
> > > > > > > > > > > > +
> > > > > > > > > > > >     '#address-cells':
> > > > > > > > > > > >       const: 1
> > > > > > > > > > > >   
> > > > > > > > > > > > @@ -128,6 +140,7 @@ patternProperties:
> > > > > > > > > > > >             - custom-output-range-config
> > > > > > > > > > > >   
> > > > > > > > > > > >   allOf:
> > > > > > > > > > > > +  - $ref: /schemas/spi/spi-peripheral-props.yaml#
> > > > > > > > > > > >     - if:
> > > > > > > > > > > >         properties:
> > > > > > > > > > > >           compatible:
> > > > > > > > > > > > @@ -238,4 +251,33 @@ examples:
> > > > > > > > > > > >               };
> > > > > > > > > > > >           };
> > > > > > > > > > > >       };
> > > > > > > > > > > > +
> > > > > > > > > > > > +  - |
> > > > > > > > > > > > +    axi_dac: spi@...70000 {
> > > > > > > > > > > > +        compatible = "adi,axi-ad3552r";    
> > > > > > > > > > > That is either redundant or entire example should go to
> > > > > > > > > > > the
> > > > > > > > > > > parent
> > > > > > > > > > > node,
> > > > > > > > > > > if this device is fixed child of complex device (IOW,
> > > > > > > > > > > adi,ad3552r
> > > > > > > > > > > cannot
> > > > > > > > > > > be used outside of adi,axi-ad3552r).    
> > > > > > > > > > 
> > > > > > > > > > ad3552r can still be used by a generic "classic" spi
> > > > > > > > > > controller (SCLK/CS/MISO) but at a slower samplerate, fpga
> > > > > > > > > > controller only (axi-ad3552r) can reach 33MUPS.    
> > > > > > > > > 
> > > > > > > > > OK, then this is just redundant. Drop the node. Parent example
> > > > > > > > > should
> > > > > > > > > contain the children, though.    
> > > > > > > > > >    
> > > > > > > > > > >    
> > > > > > > > > > > > +        reg = <0x44a70000 0x1000>;
> > > > > > > > > > > > +        dmas = <&dac_tx_dma 0>;
> > > > > > > > > > > > +        dma-names = "tx";
> > > > > > > > > > > > +        #io-backend-cells = <0>;
> > > > > > > > > > > > +        clocks = <&ref_clk>;
> > > > > > > > > > > > +
> > > > > > > > > > > > +        #address-cells = <1>;
> > > > > > > > > > > > +        #size-cells = <0>;
> > > > > > > > > > > > +
> > > > > > > > > > > > +        dac@0 {
> > > > > > > > > > > > +            compatible = "adi,ad3552r";
> > > > > > > > > > > > +            reg = <0>;
> > > > > > > > > > > > +            reset-gpios = <&gpio0 92 0>;    
> > > > > > > > > > > Use standard defines for GPIO flags.    
> > > > > > > > > > 
> > > > > > > > > > fixed, thanks
> > > > > > > > > >    
> > > > > > > > > > > > +            io-backends = <&axi_dac>;    
> > > > > > > > > > > Why do you need to point to the parent? How much coupled
> > > > > > > > > > > are
> > > > > > > > > > > these
> > > > > > > > > > > devices? Child pointing to parent is not usually expected,
> > > > > > > > > > > because
> > > > > > > > > > > that's obvious.    
> > > > > > > > > > 
> > > > > > > > > > 
> > > > > > > > > > "io-backends" is actually the way to refer to the backend
> > > > > > > > > > module,
> > > > > > > > > > (used already for i.e. ad9739a),
> > > > > > > > > > it is needed because the backend is not only acting as spi-
> > > > > > > > > > controller,
> > > > > > > > > > but is also providing some APIs for synchronization and bus
> > > > > > > > > > setup
> > > > > > > > > > support.    
> > > > > > > > > 
> > > > > > > > > 
> > > > > > > > > But if backend is the parent, then this is redundant. You can
> > > > > > > > > take
> > > > > > > > > it
> > > > > > > > > from the child-parent relationship. Is this pointing to other
> > > > > > > > > devices
> > > > > > > > > (non-parent) in other ad3552r configurations?
> > > > > > > > >    
> > > > > > > > 
> > > > > > > > The backend is a provider-consumer type of API. On the consumer
> > > > > > > > side
> > > > > > > > (which
> > > > > > > > is the
> > > > > > > > driver the child node will probe on), we need to call
> > > > > > > > devm_iio_backend_get()
> > > > > > > > to get
> > > > > > > > the backend object (which obviously is the parent). For that,
> > > > > > > > 'io-
> > > > > > > > backends'
> > > > > > > > is being    
> > > > > > > 
> > > > > > > You described the driver, so how does it matter? Driver can call
> > > > > > > get_backend_from_parent(), right? Or
> > > > > > > get_backend_from_fwnode(parent)?    
> > > > > > 
> > > > > > Well yes, just stating what the framework (also in terms of
> > > > > > bindings) is
> > > > > > expecting. Of course that on the driver side we can paper around it
> > > > > > the
> > > > > > way we
> > > > > > want. But my main point was that we can only paper around it if we
> > > > > > use
> > > > > > code that
> > > > > > is meant not to be used.
> > > > > > 
> > > > > > And, FWIW, I was (trying) replying to your comment
> > > > > > 
> > > > > > "You can take it from the child-parent relationship"
> > > > > > 
> > > > > > Again, we can only do that by introducing new code or use code
> > > > > > that's not
> > > > > > meant
> > > > > > to be used. The way we're supposed to reference backends is by
> > > > > > explicitly
> > > > > > using
> > > > > > the proper FW property.
> > > > > > 
> > > > > > Put it in another way and a completely hypothetical case. If we have
> > > > > > a spi
> > > > > > controller which happens to export some clock and one of it's
> > > > > > peripherals
> > > > > > ends
> > > > > > up using that clock, wouldn't we still use 'clocks' to reference
> > > > > > that
> > > > > > clock?    
> > > > > 
> > > > > I asked how coupled are these devices. Never got the answer and you
> > > > > are
> > > > > reflecting with question. Depends. Please do not create hypothetical,
> > > > > generic scenarios and then apply them to your one particular opposite
> > > > > case.  
> > > > 
> > > > I'll throw a possible clarifying question in here.  Could we use this
> > > > device with a multimaster SPI setup such that the control is on a
> > > > conventional
> > > > SPI controller (maybe a qspi capable one), and the data plane only goes
> > > > through
> > > > a specific purpose backend?  If so, then they are not tightly coupled
> > > > and
> > > > the reference makes sense.  Putting it another way, the difference
> > > > between
> > > > this case and all the prior iio-backend bindings is the control and
> > > > dataplanes
> > > > use the same pins.  Does that have to be the case at the host end?  If
> > > > it
> > > > does,
> > > > then the reference isn't strictly needed and this becomes a bit like
> > > > registering a single device on an spi bus or an i2c bus depending on who
> > > > does the registering (which is down to the parent in DT).
> > > >   
> > > 
> > > So, we currently have two drivers (with a new one being added in this
> > > series)
> > > for the same device:
> > > 
> > > 1) A SPI one tied to a typical spi controller. This is the "low speed"
> > > implementation and does not use backends;
> > > 2) The new platform device that is connected like this to the backend.
> > > 
> > > So yes, my understanding (but Angelo should know better :)) is that they
> > > are
> > > tightly coupled. Putting it in another way, the new platform device is
> > > very much
> > > specific to this parent (and yeah, this is a very special usecase where
> > > control
> > > and data planes are controlled by the IIO backend) and should not exist
> > > with it.  
> > 
> > ad3552r device can be coupled to the axi-ad3552r controller or to a generic 
> > spi controler.
> > 
> > We have actually 2 drivers, SPI and platform (for AXI controller, in this
> > patch).
> > 
> > Scenario 1 (SPI):
> > ad3522r can hypotetically work with whatever simple spi controller that can
> > read/write registers in raw mode. On simple SPI (CS, SCLK, MOSI), due to
> > ad3552r
> > chip limitation of 66Mhz clock, the maximum 33MUPS (16 bit samples) cannot
> > be
> > reached. Some QSPI DDR controller seems to be around, in that case, ad3552r
> > may work extending the SPI driver. 
> > 
> > Scenario 2 (AXI):
> > From an hardware-only point ov view axi-ad3552r IP acts as QSPI+DDR
> > controller
> > plus some additional features for stream synchronization.
> > From a sowftware point of view, really different from a spi controller
> > driver.
> > It's just a backend with APIes that can be called from the child driver.
> 
> Potential? scenario 3 is the one that interested me.
> 
> ad3552 double wired to a normal SPI controller (so like option 1) and
> to a an offload engine (so like option 2).  With a few pull up resistors
> (cs and clk?) and some care it should electrically work I think.
> In that case we'd need the io-backend reference because the parent
> would be the option 1 like SPI bus and the io-backend would not be
> the parent.
> 
> _______________________
> Host       SPI MOSI    |-------------------\
> hard       SPI MISO 0-3|----------------\  |
> QSPI       SPI CLK     |--------------\  | |
>            SPI CS      |----------\    | | |
>                        |           |   | | |
> FPGA                   |           |   | | |   |
> Soft       SPI MOSI    |-----------|---|-|-x---|
> QSPI       SPI MISO 0-3|-----------|---|-x-----|  DAC
> Offload    SPI CLK     |-----------|---x-------|
> with DDR   SPI CS      |-----------x-----------|
> _______________________|
> 
> Makes all sorts of assumptions about the SPI controllers being designed
> for multi controllers on the same SPI buses but I'm not aware of a reason
> you can't do that.
> 
> As the only control message that would need to go over the offload engine
> would be the exit DDR (I think) that might be hard coded into a slightly
> simpler soft IP along with the bulk data transfer stuff.
> 

Not even sure if DDR would be a problem. Right now, I __think__ we need to
enable DDR both the peripheral and on the backend. On the peripheral we could
use the control link on the non offload controller. On the offload controller,
we would set it through IIO backend and that would be a backend config and not
go over the bus.

To make a correction on my previous reply to Krzysztof, the HW folks made some
experiments with the SPI ENIGINE IP (with the offload engine) but without the
AXI DAC IP. So, effectively only one controller in place. That said, I'm also
not seeing any reason why something like the above would not be possible.

- Nuno Sá

Powered by blists - more mailing lists