acpi, pci, spibus -- tying it all together

[Konstantin Belousov]

On Sun, Sep 09, 2018 at 10:25:44AM +0300, Yuri Pankov wrote:
> I have modified intelspi to attach to SPI master (Sunrise Point-H Serial 
> IO SPI) located on pci bus, and on attach I have the spibus0 and spibus1 
> buses added.
> 
> As spibus is not self-enumerating, the only way to find the slave device 
> I need is via acpi bus probe, that works, but I don't see a way to make 
> it a child of spibus1 (where it's located).
> 
> In ACPI terms it looks like the below (from DSDT):
> 
> ...
> \_SB_.PCI0.SPI0 "Device (SPI0)"
> \_SB_.PCI0.SPI1 "Device (SPI1)"
> ...
> Scope (SPI1) {
>      Device (SPIT) {
>          ...here comes all the data we need,
>             including the ACPI ID we can probe...
>      }
> }
> ...
> 
> I hope that made at least some sense.  And the question is if there any 
> existing way of tying it all together, and adding that slave device to 
> the spibus1 (which only knows about hinted children at the moment)?

I had very similar situation where I wrote NVDIMM driver.

The suggestion I got was to implement DEVICE_IDENTIFY() method.  It
gets the driver_t and the parent bus device_t arguments.  In the method,
you create children' device_t, iterating over the ACPI enumerated entries.
Practically, you would use acpica AcpiWalkNamespace(ACPI_TYPE_DEVICE, ...)
function and call BUS_ADD_CHILD() when needed.  Set the ACPI handle for
the created child.  [Get the Intel' ACPICA reference manual to understand
the KPI].

Then probe does nothing, and attach verifies against the handle.

You can see that in https://kib.kiev.ua/kib/nvdimm.7.patch
nvdimm_identify(), nvdimm_foreach_acpi(), and nvdimm_create_dev().