Re: [RFC 2/2] dt-bindings: firmware: tegra186-bpmp: Document interconnects property

From: Dmitry Osipenko <digetx@gmail.com>
To: Thierry Reding <thierry.reding@gmail.com>
Cc: Georgi Djakov <georgi.djakov@linaro.org>,
	Rob Herring <robh+dt@kernel.org>,
	Jon Hunter <jonathanh@nvidia.com>,
	linux-tegra@vger.kernel.org, devicetree@vger.kernel.org
Subject: Re: [RFC 2/2] dt-bindings: firmware: tegra186-bpmp: Document interconnects property
Date: Wed, 29 Jan 2020 19:02:40 +0300	[thread overview]
Message-ID: <0b8692ab-4e06-b277-bbe2-93922e47c2f6@gmail.com> (raw)
In-Reply-To: <20200129093602.GC2479935@ulmo>

29.01.2020 12:36, Thierry Reding пишет:
> On Tue, Jan 28, 2020 at 10:27:00PM +0300, Dmitry Osipenko wrote:
>> 27.01.2020 15:21, Thierry Reding пишет:
>>> On Tue, Jan 21, 2020 at 11:12:11PM +0300, Dmitry Osipenko wrote:
>>>> 21.01.2020 18:54, Thierry Reding пишет:
>>>>> On Tue, Jan 21, 2020 at 05:18:43PM +0200, Georgi Djakov wrote:
>>>>>> On 1/21/20 16:10, Thierry Reding wrote:
>>> [...]
>>>>>>> I'm not sure if that TEGRA_ICC_EMEM makes a lot of sense. It's always
>>>>>>> going to be the same and it's arbitrarily defined, so it's effectively
>>>>>>> useless. But other than that it looks good.
>>>>>>
>>>>>> Well, in most cases the target would be the EMEM, so that's fine. I have seen
>>>>>> that other vendors that may have an additional internal memory, especially
>>>>>> dedicated to some DSPs and in such cases the bandwidth needs are different for
>>>>>> the two paths (to internal memory and DDR).
>>>>>
>>>>> Most chips have a small internal memory that can be used, though it
>>>>> seldomly is. However, in that case I would expect the target to be a
>>>>> completely different device, so it'd look more like this:
>>>>>
>>>>> 	interconnects = <&mc TEGRA186_MEMORY_CLIENT_BPMPR &iram>,
>>>>> 			...;
>>>>>
>>>>> I don't think EMEM has any "downstream" other than external memory.
>>>>
>>>> The node ID should be mandatory in terms of interconnect, even if it's a
>>>> single node. EMC (provider) != EMEM (endpoint).
>>>
>>> I don't understand why. An ID only makes sense if you've got multiple
>>> endpoints. For example, a regulator is a provider with a single endpoint
>>> so we don't specify an ID.
>>
>> Because this is how ICC binding is defined, unless I'm missing something.
> 
> I don't think so. It's defined as "pairs of phandles and interconnect
> provider specifiers", which is equivalent to what pretty much all of the
> resource bindings define. The #interconnect-cells property defines the
> number of cells used for the specifier. In the normal case this would be
> 1, and the value of the one cell would be the ID of the endpoint. But if
> there's only a single endpoint, it's customary to set the number of
> cells to 0, in which case only the phandle is required.

Right, setting interconnect-cells=0 should work. I'll give it a try,
thank you!