Message-ID: <c263b843-50bc-4c2c-b15e-9b87dfb201ab@linaro.org>
Date: Mon, 23 Sep 2024 17:29:46 +0200
From: Krzysztof Kozlowski <krzysztof.kozlowski@...aro.org>
To: Abdellatif El Khlifi <abdellatif.elkhlifi@....com>,
 Krzysztof Kozlowski <krzk@...nel.org>
Cc: mathieu.poirier@...aro.org, Adam.Johnston@....com,
 Hugues.KambaMpiana@....com, Drew.Reed@....com, andersson@...nel.org,
 conor+dt@...nel.org, devicetree@...r.kernel.org,
 krzysztof.kozlowski+dt@...aro.org, linux-arm-kernel@...ts.infradead.org,
 linux-kernel@...r.kernel.org, linux-remoteproc@...r.kernel.org,
 liviu.dudau@....com, lpieralisi@...nel.org, robh@...nel.org,
 sudeep.holla@....com, robin.murphy@....com
Subject: Re: [PATCH v2 1/5] dt-bindings: remoteproc: sse710: Add the External
 Systems remote processors

On 23/09/2024 13:49, Abdellatif El Khlifi wrote:
> Hi Krzysztof,
> 
>>>>>>>>>>> +  '#extsys-id':
>>>>>>>>>>
>>>>>>>>>> '#' is not correct for sure, that's not a cell specifier.
>>>>>>>>>>
>>>>>>>>>> But anyway, we do not accept in general instance IDs.
>>>>>>>>>
>>>>>>>>> I'm happy to replace the instance ID with  another solution.
>>>>>>>>> In our case the remoteproc instance does not have a base address
>>>>>>>>> to use. So, we can't put remoteproc@...ress
>>>>>>>>>
>>>>>>>>> What do you recommend in this case please ?
>>>>>>>>
>>>>>>>> Waiting one month to respond is a great way to drop all context from my
>>>>>>>> memory. The emails are not even available for me - gone from inbox.
>>>>>>>>
>>>>>>>> Bus addressing could note it. Or you have different devices, so
>>>>>>>> different compatibles. Tricky to say, because you did not describe the
>>>>>>>> hardware really and it's one month later...
>>>>>>>>
>>>>>>>
>>>>>>> Sorry for waiting. I was in holidays.
>>>>>>>
>>>>>>> I'll add more documentation about the external system for more clarity [1].
>>>>>>>
>>>>>>> Basically, Linux runs on the Cortex-A35. The External system is a
>>>>>>> Cortex-M core. The Cortex-A35 can not access the memory of the Cortex-M.
>>>>>>> It can only control Cortex-M core using the reset control and status registers mapped
>>>>>>> in the memory space of the Cortex-A35.
>>>>>>
>>>>>> That's pretty standard.
>>>>>>
>>>>>> It does not explain me why bus addressing or different compatible are
>>>>>> not sufficient here.
>>>>>
>>>>> Using an instance ID was a design choice.
>>>>> I'm happy to replace it with the use of compatible and match data (WIP).
>>>>>
>>>>> The match data will be pointing to a data structure containing the right offsets
>>>>> to be used with regmap APIs.
>>>>>
>>>>> syscon node is used to represent the Host Base System Control register area [1]
>>>>> where the external system reset registers are mapped (EXT_SYS*).
>>>>>
>>>>> The nodes will look like this:
>>>>>
>>>>> syscon@...10000 {
>>>>>         compatible = "arm,sse710-host-base-sysctrl", "simple-mfd", "syscon";
>>>>>         reg = <0x1a010000 0x1000>;
>>>>>
>>>>>         #address-cells = <1>;
>>>>>         #size-cells = <1>;
>>>>>
>>>>>         remoteproc@310 {
>>>>>             compatible = "arm,sse710-extsys0";
>>>>>             reg = <0x310 4>;
>>>>
>>>> Uh, why do you create device nodes for one word? This really suggests it
>>>> is part of parent device and your split is artificial.
>>>
>>> The external system registers (described by the remoteproc node) are part
>>> of the parent device (the Host Base System Control register area) described
>>> by syscon.
>>>
>>> In case of the external system 0 , its registers are located at offset 0x310
>>> (physical address: 0x1a010310)
>>>
>>> When instantiating the devices without @address, the DTC compiler
>>> detects 2 nodes with the same name (remoteproc).
>>
>> There should be no children at all. DT is not for instantiating your
>> drivers. I claim you have only one device and that's
>> arm,sse710-host-base-sysctrl. If you create child node for one word,
>> that's not a device.
> 
> The Host Base System Control [3] is the big block containing various functionalities (MMIO registers).
> Among the functionalities, the two remote cores registers (aka External system 0 and 1).
> The remote cores have two registers each.
> 
> 1/ In the v1 patchset, a valid point was made by the community:
> 
>    Right now it seems somewhat tenuous to describe two consecutive
>    32-bit registers as separate "reg" entries, but *maybe* it's OK if that's

ARM is not special, neither this hardware is. Therefore:
1. Each register as reg: nope, for obvious reasons.
2. One device for entire syscon: quite common, why do you think it is
somehow odd?
3. If you quote other person, please provide the lore link, so I won't
spend useless 5 minutes to find who said that or if it was even said...

>    all there ever is. However if it's actually going to end up needing several
>    more additional MMIO and/or memory regions for other functionality, then
>    describing each register and location individually is liable to get
>    unmanageable really fast, and a higher-level functional grouping (e.g. these
>    reset-related registers together as a single 8-byte region) would likely be
>    a better design.
> 
>    The Exernal system registers are part of a bigger block with other functionality in place.
>    MFD/syscon might be better way to use these registers. You never know in
>    future you might want to use another set of 2-4 registers with a different
>    functionality in another driver.
> 
>    I would see if it makes sense to put together a single binding for
>    this "Host Base System Control" register (not sure what exactly that means).
>    Use MFD/regmap you access parts of this block. The remoteproc driver can
>    then be semi-generic (meaning applicable to group of similar platforms)
>    based on the platform compatible and use this regmap to provide the
>    functionality needed.

I don't understand how this lengthy semi-quote answers my concerns.
Please write concise points as arguments to my questions.

For example, I don't care what your remote proc driver does and it
should not matter in the terms of this binding.

> 
> 2/ There are many examples in the kernel that use syscon as a parent node of
>    child nodes for devices located at an offset from the syscon base address.
>    Please see these two examples [1][2]. I'm trying to follow a similar design if that
>    makes sense.

Yeah, for separate devices. If you have two words without any resources,
I claim you might not have here any separate devices or "not separate
enough", because all this is somehow fluid. Remote core sounds like
separate device, but all your arguments about need of extid which cannot
be used in reg does not support this case.

The example in the binding is also not complete - missing rest of
devices - which does not help.

> 
> 3/ Since there are two registers for each remote core. I'm suggesting to set the
>    size in the reg property to 8. 

How is this related?

> The driver will read the match data to get the right
>    offset to be used with regmap APIs.

Sorry, no talks about driver. Don't care, at least in this context.

You can completely omit address space from children in DT and everything
will work fine and look fine from bindings point of view.

> 
> Suggested nodes:
> 
> 
>     syscon@...10000 {
>         compatible = "arm,sse710-host-base-sysctrl", "simple-mfd", "syscon";
>         reg = <0x1a010000 0x1000>;
> 
>         #address-cells = <1>;
>         #size-cells = <1>;
> 
>         remoteproc@310 {
>             compatible = "arm,sse710-extsys0";
>             reg = <0x310 8>;
>             firmware-name = "es_flashfw.elf";
>             mboxes = <&mhu0_hes0 0 1>, <&mhu0_es0h 0 1>;
>             mbox-names = "txes0", "rxes0";
>             memory-region = <&extsys0_vring0>, <&extsys0_vring1>;
>         };
> 
>         remoteproc@318 {
>             compatible = "arm,sse710-extsys1";
>             reg = <0x318 8>;
>             firmware-name = "es_flashfw.elf";

Same firmware? Always or only depends?

>             mboxes = <&mhu0_hes1 0 1>, <&mhu0_es1h 0 1>;
>             mbox-names = "txes0", "rxes0";
>             memory-region = <&extsys1_vring0>, <&extsys1_vring1>;

The rest of resources support the idea of two children but I still have
doubts about need of identifying remote instances.

Looking at your driver it is totally not needed. Your driver just
duplicates the regs here, so it's a proof that you are not using DT
correctly.

>         };
> };
> 
> 
> [1]: Documentation/devicetree/bindings/clock/sprd,sc9863a-clk.yaml
> 
>     syscon@...00000 {
>       compatible = "sprd,sc9863a-glbregs", "syscon", "simple-mfd";
>       reg = <0x20e00000 0x4000>;
>       #address-cells = <1>;
>       #size-cells = <1>;
>       ranges = <0 0x20e00000 0x4000>;
> 
>       apahb_gate: apahb-gate@0 {
>         compatible = "sprd,sc9863a-apahb-gate";
>         reg = <0x0 0x1020>;

Well, size 1020, but please never use sprd as an example. You can as
well point to a buggy code and say that "I can implement bugs as well,
because there are bugs already".

Same for few other almost abandoned, poorly maintained platforms.

>         #clock-cells = <1>;
>       };
>     };
> 
> 
> [2]: Documentation/devicetree/bindings/arm/arm,juno-fpga-apb-regs.yaml:
> 
>     syscon@...00 {
>         compatible = "arm,juno-fpga-apb-regs", "syscon", "simple-mfd";
>         reg = <0x010000 0x1000>;
>         ranges = <0x0 0x10000 0x1000>;
>         #address-cells = <1>;
>         #size-cells = <1>;
> 
>         led@8,0 {
>             compatible = "register-bit-led";

register-bit-led... what do you want to prove? You will find
clocks-per-reg and try to implement them? That's known no-go.

Best regards,
Krzysztof

Powered by blists - more mailing lists