Re: [PATCH v2 3/3] dt-bindings: designware: add binding for Designware PCIe in ECAM mode

[Rob Herring <robh@kernel.org>]

On Thu, Aug 24, 2017 at 3:12 PM, Ard Biesheuvel
<ard.biesheuvel@linaro.org> wrote:
> On 24 August 2017 at 21:02, Rob Herring <robh@kernel.org> wrote:
>> Cc the DT list for bindings please.
>>
>> On Thu, Aug 24, 2017 at 1:43 PM, Ard Biesheuvel
>> <ard.biesheuvel@linaro.org> wrote:
>>> Describe the binding for firmware-configured instances of the Synopsys
>>> Designware PCIe controller in RC mode.
>>>
>>> Cc: Rob Herring <robh@kernel.org>
>>> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
>>> ---
>>>  Documentation/devicetree/bindings/pci/designware-pcie-ecam.txt | 56 ++++++++++++++++++++
>>>  1 file changed, 56 insertions(+)
>>>
>>> diff --git a/Documentation/devicetree/bindings/pci/designware-pcie-ecam.txt b/Documentation/devicetree/bindings/pci/designware-pcie-ecam.txt
>>> new file mode 100644
>>> index 000000000000..b8127b19c220
>>> --- /dev/null
>>> +++ b/Documentation/devicetree/bindings/pci/designware-pcie-ecam.txt
>>> @@ -0,0 +1,56 @@
>>> +* Synopsys Designware PCIe root complex in ECAM mode
>>> +
>>> +In some cases, firmware may already have configured the Synopsys Designware
>>> +PCIe controller in RC mode with static ATU window mappings that cover all
>>> +config, MMIO and I/O spaces in a [mostly] ECAM compatible fashion.
>>> +In this case, there is no need for the OS to perform any low level setup
>>> +of clocks or device registers, nor is there any reason for the driver to
>>> +reconfigure ATU windows for config and/or IO space accesses at runtime.
>>> +
>>> +Such hardware configurations should be described as "pci-host-ecam-generic"
>>> +if they are truly ECAM compatible. Configurations that require no low-level
>>> +setup by the OS nor any ATU window reconfiguration at runtime, but do
>>> +require special handling for type 0 config TLPs may instead be described as
>>> +"snps,dw-pcie-ecam".
>>
>> Humm, what happens when we have the next exception that's SoC specific
>> or another vendor?
>
> This is not SoC specific, but IP specific. We are working with two
> different SoCs from completely different vendors that both synthesized
> this IP with a 64 KB ATU window size, not expecting this to break ECAM
> compatibility.

This case is not SoC specific, but quirks/bugs show up in both places.
Plus "dw-pcie" is fairly meaningless because the IP is both
configurable and has multiple releases.

>> Seems like perhaps "firmware initialized" should
>> have been a separate property flag for bootloaders to add rather than
>> a compatible string.
>>
>
> Yes, but then you still have 10 different drivers that all retain the
> low-level bits that are all different between SoCs. That is exactly
> what I want to get rid of, and usually we can do that with existing
> bindings, because we simply expose it as pci-host-ecam-generic. Only
> in this particular case, that doesn't fly due to the quirk.

You can expose it as "vendor,soc-pcie", "pci-host-ecam-generic" and
then the kernel can decide. Then you can use: the default ecam driver,
the ecam driver but match on "vendor,soc-pcie" for quirks, or a vendor
specific driver. The only thing that wouldn't work would be having
both quirks in the ecam driver and a vendor driver which IMO we just
shouldn't support anyway.

Now if you have 10 SoCs all needing this same quirk, then maybe adding
"snps,dw-pcie-ecam" addtionally to the above makes sense. But that
only saves you match strings.

>> I'd rather see this done in a way that does not require DT updates if
>> quirks have to be handled/added later.
>>
>
> Do you see a way that still allows us to keep the abstraction? I don't
> want a flag, I simply don't want to expose any low-level specifics
> about the device to the OS, beyond what it needs to use it in its
> configured state.

That's not how DT works. Detailed information is exposed to the OS and
the OS decides if it can handle things generically or not because that
decision can change over time. Whenever we don't make things specific
enough, we've gotten burned.

Rob