[U-Boot] [PATCH] pci: Support parsing PCI controller DT subnodes

[Bin Meng]

Hi Marek,

On Tue, Aug 14, 2018 at 4:54 PM, Marek Vasut <marek.vasut at gmail.com> wrote:
> On 08/14/2018 04:34 AM, Bin Meng wrote:
>> Hi Tom,
>>
>> On Tue, Aug 14, 2018 at 1:16 AM, Tom Rini <trini at konsulko.com> wrote:
>>> On Mon, Aug 13, 2018 at 06:07:14PM +0200, Marek Vasut wrote:
>>>> On 08/13/2018 03:39 PM, Tom Rini wrote:
>>>> [...]
>>>>
>>>>>>>> Next step is to upstream the DT
>>>>>>>> changes to Linux kernel, then sync the changes to U-Boot to satisfy
>>>>>>>> this obsession - using exactly the same DT as Linux.
>>>>>>>
>>>>>>> This is not gonna happen.
>>>>>>>
>>>>>>> Sorry, you're really just wasting my time with this foolishness. If
>>>>>>> U-Boot cannot parse valid DT bindings while other OSes can, U-Boot is
>>>>>>> broken and must be fixed. So far I only see you attacking this patch and
>>>>>>> trying to pull in everything you can do avoid accepting this patch or
>>>>>>> providing a better alternative. This is not a constructive discussion,
>>>>>>> so I stop here.
>>>>>>
>>>>>> The fix in this patch is purely hack, period.
>>>>>
>>>>> Lets step back for a moment.
>>>>>
>>>>> First, U-Boot intends to be, in the case where a relevant DTS file
>>>>> exists, the Linux kernel one PLUS additions we require (u-boot,dm-spl,
>>>>> u-boot,dm-pre-reloc, but also sometimes stdout-path or properties that
>>>>> are omitted for various reasons).
>>>>
>>>> Right, which doesn't apply here. None of those u-boot,... props are
>>>> needed in this case.
>>>
>>> Which is why I also mentioned the non-u-boot specific props we also need
>>> sometimes.  My point is two-fold:
>>> 1: We can and will _add_ information to the dts files that come from
>>> Linux.
>>> 2: Not all information that we add is U-Boot prefixed.
>>>
>>
>> It would be better if we document such DT expectation somewhere.
>>
>>>>> Second, I've asked before (both in this thread and on IRC), and not
>>>>> gotten an answer yet, as to how Linux goes "Oh, _this_ PCI device and
>>>>> _this_ DT node need to be matched and populate some data
>>>>> structures".
>>>>
>>>> You did get an answer to that on irc from George. Looks like
>>>> of_pci_find_child_device() in drivers/pci/of.c
>>>
>>> Yeah, George said he thought that might be it but didn't have time to
>>> confirm.
>>>
>>>>> Marek's patch seems to be, in short "here's where U-Boot
>>>>> needs to wire things up".  Bin has said that no, the function in
>>>>> question is for other things.
>>>>
>>>> I disagree with this. It's a bind function and assigns other parameters
>>>> of the driver instance too.
>>>>
>>>>> I think knowing where Linux does this
>>>>> would be instructive to figure out where we need to have some additional
>>>>> logic added OR we can make some cost/benefit analysis to see if it makes
>>>>> more sense overall to add compatibles to some nodes rather than add to
>>>>> the binary size.
>>>>
>>>> Adding compatible does not make any sense, the PCI ID provides that
>>>> information. Adding compatible would only add redundancy which could
>>>> possibly be even harmful (ie. if the controller got replaced with
>>>> another one).
>>>
>>> To try and move things along rather than re-argue the same point, you're
>>> saying that our pci_find_and_bind_driver() is the rough equivalent of
>>> of_pci_find_child_device() or at least pci_set_of_node() (which calls
>>> of_pci_find_child_device()).
>>>
>>> So, Bin, if this isn't the right place to start down this path, where
>>> would be?  Given that Linux can take a DTB and PCI bus with devices and
>>> get things right, what would it look like for U-Boot to replicate the
>>> same behavior?  Instead of having to add explicit compatible nodes for
>>> each PCI device, as I understand (but correct me if I'm wrong) we're
>>> doing today.  Thanks!
>>
>> So is this a requirement for all U-Boot driver subsystems to replicate
>> the same Linux behavior? If yes, can we have it officially documented
>> somewhere?
>
> No, because we are not replicating Linux behavior.
>

But you kept mentioning you wanted U-Boot to use exactly the same DT
from Linux. And I pointed out that FreeBSD's DT files are not the same
as Linux. Here you are saying you don't want U-Boot to replicate Linux
behavior, if not the Linux behavior, what do you want to suggest then?

>> Since Marek refused to take the original U-Boot option 1 to support
>> his case, and asked U-Boot to follow Linux's practice on PCI device
>> binding, if we go that way, here is what we can do:
>
> You are inserting words into my mouth and I dislike that. I never said
> anything about Linux. I said DT is OS agnostic and U-Boot should be able
> to parse DT as fully as possible.

See above comment. I might have used some words that made you feel
uncomfortable, and for me I felt the same way. Part of the reason is
that I am not a native English speaker and I may mis-use/interpret
words. If that's the case I apologize. Anyway I won't quarrel against
this over and over again. This does not help to move things forward.

>
>> * Keep pci-uclass driver's post_bind() and child_post_bind() only for
>> Sandbox configuration
>> * Keep the call to pci_bus_find_devfn() in pci_bind_bus_devices() only
>> for Sandbox configuration
>> * Sandbox is special. We should limit the mechanism of matching PCI
>> emulation device via "compatible" to sandbox only
>
> I don't think this is limited to sandbox, although I don't see a
> real-world usecase right now.
>

You are welcome to send patches against sandbox PCI codes to remove
such limitation, to make it behave like a real-world bus device.

>> * Assign the DT node to the bound device in pci_find_and_bind_driver()
>> if there is a valid PCI "reg" encoding for a specific PCI device in
>> the device tree
>> * Create DM PCI test case against the DT node assignment
>> * Remove all compatible string in U-Boot's PCI device drivers: eg:
>> ehci_pci_ids[], xhci_pci_ids[], etc. IOW, all PCI device drivers
>> should only use U_BOOT_PCI_DEVICE(), aka the original U-Boot option 2
>> * Fork a "pci-ns16550" driver to support U_BOOT_PCI_DEVICE(), as
>> currently PCI ns16550 device driver uses "compatible" string to do the
>> matching, and update crownbay.dts and galileo.dts (so far I only know
>> two boards are using PCI ns16550 serial port)
>
> I cannot test such changes, but I believe there is
> PCI_CLASS_COMMUNICATION_SERIAL and matching on that would suffice ?
>

Maybe, but I need check the datasheet to confirm. This can be a good
start though. Note I can test the changes, and I can do the whole
bunch of such proposed design changes if you don't mind, but let's
wait for Simon's comments.

>> * Make sure all DM PCI test cases are not broken
>> * Document all of the above changes in doc/driver-model/pci-info.txt
>>
>> I am not sure if I missed anything. Simon, could you also comment on it?

Regards,
Bin