Re: Raspberry Pi 4 DMA range fix

[Jared McNeill <jmcneill%invisible.ca@localhost>]

On May 3, 2022, at 7:19 PM, Jason Thorpe <thorpej%me.com@localhost> wrote:
> 
> 
>> On May 3, 2022, at 2:58 PM, Jared McNeill <jmcneill%invisible.ca@localhost> wrote:
>> 
>> On May 3, 2022, at 6:49 PM, Jason Thorpe <thorpej%me.com@localhost> wrote:
>>> 
>>> 
>>> 
>>>> On May 3, 2022, at 2:46 PM, Jared McNeill <jmcneill%invisible.ca@localhost> wrote:
>>>> 
>>>> That patch isn’t quite right because _DMA restrictions apply only to children of the bus that it appears on (ACPI 6.4 section 6.2.4). So yeah, we want _DMA to apply to PCI devices that are a child of the PCI0 device here, but this isn’t the right way to go about it.
>>>> 
>>>> I think we need to special case this for acpipchb because we don’t really have a clean way of expressing the relationship between a PCI device and firmware tables.
>>> 
>>> Do you mean a PCI bus instance or a child device of the PCI bus?
>> 
>> In this case it’s the bus that needs to know about the restrictions that it has to apply to child devices. I got a bit ahead of myself here thinking about device properties and lumping those two problems together :)
> 
> The code in the tree should already associate ACPI handles for the bus nodes with “pci" bus instances (at least I’m pretty sure I checked that in).  Is that sufficient to be able to find the correct entries in the tables?

Yeah we’ve had the ability to lookup a node by SBDF for at least a decade now (acpi_pcidev_find). Always seemed sort of clunky to me. But that’s separate from the topic of this thread.