Re: [PATCH v3 0/4] SysFS driver for QEMU fw_cfg device

[Laszlo Ersek <lersek-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>]

On 10/05/15 15:05, Mark Rutland wrote:
>>> I'm not sure I follow what the difficulty with supporting DT in addition
>>> to ACPI is? It looks like all you need is a compatible string and a reg
>>> entry.
>>
>> Bearing in mind that I have almost no experience with arm:
>>
>> I started out by probing all possible port-io and mmio locations where
>> fw_cfg registers might have been found, from a "classic" module_init
>> method.
>>
>> Arm has DT, which as far as I understand will answer the following two
>> questions: 1. Do I have fw_cfg ? 2. If yes, what address range does it use ?
>> So that I could continue using a classic module_init, but won't need
>> to probe for the device.
>>
>> PC (my primary architecture, the one I actually care about) does not
>> have DT. If I want to share the same code, I can't probe, so if I try
>> DT and don't find fw_cfg there (or somehow DT is no-op-ed out because
>> I'm on a PC guest), I could somehow look it up in ACPI the same way
>> (i.e., use ACPI as sort of a stand-in for DT).
> 
> I'd imagine that it's simple to have something in your probe path like:
> 
> if (pdev->dev.of_node)
> 	parse_dt(pdev);
> else
> 	parse_acpi(pdev);
> 
>> But all ACPI-enabled drivers I could find use dedicated macros (i.e.
>> no more classic module_init() and module_exit(), but rather
>> module_acpi_driver() with .add and .remove methods on an acpi_driver
>> object, etc.) Not sure how I'd glue DT back into something like that.
> 
> You don't have to use those macros, and can simply use the classic
> module_{init,exit} functions, calling the requisite acpi driver
> registration functions at module {init,exit} time.
> 
>> In addition, Michael's comment earlier in the thread suggests that
>> even my current acpi version isn't sufficiently "orthodox" w.r.t.
>> ACPI, and I should be providing the hardware access routine as
>> an ACPI/AML routine, to avoid race conditions with the rest of ACPI,
>> and for encapsulation. I.e. it's even rude to use the fw_cfg node's
>> ACPI _CRS method (the part where I'd be treating it like a DT stand-in
>> only to query fw_cfg's hardware specifics).
> 
> As Peter stated, this sounds very much like it rules out sharing the
> interface with FW generally (and is certainly scary).
> 
>> So far, all the information I've been able to pull together points
>> away from a dual DT + ACPI all-in-one solution for fw_cfg. If you know
>> of an example where that's done in an acceptable way, please let
>> me know so I can use it for inspiration...
> 
> I'm not immediately aware, but I would imagine you could search for
> files that had both an of_match_table and a acpi_bus_register_driver
> call.

One file that I think is an example for this (and I have looked at
before) is: "drivers/virtio/virtio_mmio.c".

Virtio-mmio is supposed to be enumerable in both ACPI and DT virtual
machines. For the QEMU side, grep QEMU for "LNRO0005" vs. "virtio,mmio".

Thanks
Laszlo