Date:   Thu, 29 Nov 2018 10:18:58 -0600
From:   David Lechner <david@...hnology.com>
To:     Roger Quadros <rogerq@...com>, tony@...mide.com
Cc:     robh+dt@...nel.org, bcousson@...libre.com, ssantosh@...nel.org,
        ohad@...ery.com, bjorn.andersson@...aro.org, s-anna@...com,
        nsekhar@...com, t-kristo@...com, nsaulnier@...com, jreeder@...com,
        m-karicheri2@...com, woods.technical@...il.com,
        linux-omap@...r.kernel.org, linux-remoteproc@...r.kernel.org,
        linux-kernel@...r.kernel.org, devicetree@...r.kernel.org
Subject: Re: [PATCH 01/17] dt-bindings: remoteproc: Add TI PRUSS bindings

On 11/29/18 2:49 AM, Roger Quadros wrote:
> David,
> 
> On 28/11/18 17:42, David Lechner wrote:
>> On 11/27/18 9:15 AM, Roger Quadros wrote:
>>>
>>> On 26/11/18 23:14, David Lechner wrote:
>>>> On 11/22/18 5:38 AM, Roger Quadros wrote:
>>>>> From: Suman Anna <s-anna@...com>
>>>>>
>>>>> This patch adds the bindings for the Programmable Real-Time Unit
>>>>> and Industrial Communication Subsystem (PRU-ICSS) present on various
>>>>> TI SoCs. The IP is present on multiple TI SoC architecture families
>>>>> including the OMAP architecture SoCs such as AM33xx, AM437x and
>>>>> AM57xx; and on a Keystone 2 architecture based 66AK2G SoC. It is
>>>>> also present on the Davinci based OMAPL138 SoCs and K3 architecture
>>>>> based AM65x SoCs as well (not covered for now). Details have been
>>>>> added to include bindings for various core sub-modules like the PRU
>>>>> Cores, the PRUSS Interrupt Controller, and other sub-modules used
>>>>> for Industrial Communication purposes, covering the MDIO, MII_RT
>>>>> and the IEP sub-modules. The binding mostly uses standard DT
>>>>> properties.
>>>>>
>>>>> Signed-off-by: Suman Anna <s-anna@...com>
>>>>> Signed-off-by: Roger Quadros <rogerq@...com>
>>>>> ---
>>>>>     .../devicetree/bindings/soc/ti/ti,pruss.txt        | 360 +++++++++++++++++++++
>>>>>     1 file changed, 360 insertions(+)
>>>>>     create mode 100644 Documentation/devicetree/bindings/soc/ti/ti,pruss.txt
>>>>>
>>>>> diff --git a/Documentation/devicetree/bindings/soc/ti/ti,pruss.txt b/Documentation/devicetree/bindings/soc/ti/ti,pruss.txt
>>>>> new file mode 100644
>>>>> index 0000000..24fedad
>>>>> --- /dev/null
>>>>> +++ b/Documentation/devicetree/bindings/soc/ti/ti,pruss.txt
>>>>
>>>> ...
>>>>
>>>>> +
>>>>> +PRU-ICSS SoC Bus Parent Node
>>>>> +=============================
>>>>> +This node represents the integration of the PRU-ICSS IP into a SoC, and is
>>>>> +required for all SoCs. The PRU-ICSS parent nodes need to be defined as child
>>>>> +nodes of this node.
>>>>> +
>>>>> +Required Properties:
>>>>> +--------------------
>>>>> +- compatible     : should be one of,
>>>>> +                       "ti,am3356-pruss-soc-bus" for AM335x family of SoCs
>>>>> +                       "ti,am4376-pruss-soc-bus" for AM437x family of SoCs
>>>>> +                       "ti,am5728-pruss-soc-bus" for AM57xx family of SoCs
>>>>> +                       "ti,k2g-pruss-soc-bus" for 66AK2G family of SoCs
>>>>> +- reg            : address and size of the PRUSS CFG sub-module registers
>>>>> +                   dictating the interconnect configuration
>>>>
>>>> I haven't looked into Tony's suggestion of using ti-sysc yet, so this may be a
>>>> moot point, but how will this work with AM18xx that does not have a PRUSS CFG
>>>> register? It seems to me that reg here should be the address and size of the
>>>> entire PRUSS IP block and the CFG register should be a syscon node or something
>>>> like that.
>>>
>>> The reg property description is incorrect in the patch. It should have been
>>>
>>> reg        : address of SYSCFG register.
>>>
>>> The SYSCFG register is used to enable and reset the module.
>>>
>>> But based on Tony's suggestion this wrapper driver will change to ti,sysc for
>>> OMAP like SoCs.
>>>
>>> For AM18xx it could be a simple wrapper driver that just populates the children?
>>
>> I suppose that could work. I will look into it (perhaps after seeing what you
>> come up with in v2).
>>
>>>
>>>>
>>>>> +- #address-cells : should be 1
>>>>> +- #size-cells    : should be 1
>>>>> +- ranges         : standard ranges definition
>>>>> +
>>>>
>>>> ...
>>>>
>>>>> +
>>>>> +PRUSS INTC Child Node
>>>>> +======================
>>>>> +Each PRUSS has a single interrupt controller instance that is common to both
>>>>> +the PRU cores. Each interrupt controller can detect 64 input events which are
>>>>> +then mapped to 10 possible output interrupts through two levels of mapping. The
>>>>> +input events can be triggered by either the PRUs and/or various other PRUSS
>>>>> +internal and external peripherals. The first 2 output interrupts are fed
>>>>> +exclusively to the internal PRU cores, with the remaining 8 connected to
>>>>> +external interrupt controllers including the MPU.
>>>>
>>>> FYI, on AM18xx, there is a PRUSSEVTSEL bit in CFGCHIP3[3] (already a syscon node
>>>> in the device tree) that allows selecting one of two groups of 32 input events
>>>> out of this group of 64. This is perhaps getting out of the scope of this patch
>>>> series, but I just want to make sure we end up with something that can be easily
>>>> extended for this case. For example, I was thinking that this binding could be
>>>> modified so that #interrupt-cells could be 1 or 2. If it is 2, then the first
>>>> cell specifies the PRUSSEVTSEL value and the second value is the event number.
>>>>
>>>
>>> this is da850.dtsi correct?
>>
>> Yes.
>>
>>>
>>> As PRUSSEVTSEL is not SYSEVENT specific but applies to all the SYSEVENTs at a time.
>>> I don't think interrupt-cells is the right place to specify this.
>>>
>>> Can it be set in DT in the board file? But this can't change once booted so maybe restrictive.
>>
>> I guess the way I see this is that it is like specifying the bank and index for
>> a GPIO. If you only specify the system event number, then it is not clear which
>> event you mean - it could be one of two events. You have to also specify the
>> PRUSSEVTSEL value (one could call this the bank or group, I suppose) to fully
>> describe the system event.
> 
> But this PRUSSEVTSEL is not present on most SoCs. I think it is only on the AM18xx and OMAP-L13x SoCs.
> 
> How about modelling this as a linear map of 64 events and decode this internally in INTC driver.
> i.e. 0 to 31 set PRUSSEVTSEL to 0 and 32 to 63 set PRUSSEVTSEL to 1.
> 
> If any interrupt map provides sysevents from both groups simultaneously then we complain and error out.
> 
> This should keep the dt-binding and resource table format identical across all SoCs.


This sounds reasonable to me.


> 
>>
>>>
>>> If runtime change is required it can only be done before a PRU boots.
>>>
>>> How about providing this info in the resource table and/or application DT node?
>>
>> This seems like this would make it easy to end up with a broken interrupts
>> if you accidentally try to use interrupts from both groups. Instead this
>> could be implemented in the irqchip driver such that the first interrupt
>> (system event) requested gets to select the group. If any interrupts from
>> the other group are requested later, they can just return an error. Any
>> interrupts in the same group as the first can be requested successfully.
> 
> cheers,
> -roger
> 

Powered by blists - more mailing lists