Hi Arnd,

On 20 June 2014 15:08, Arnd Bergmann [off-list ref] wrote:

On Friday 20 June 2014 14:55:16 Ashwin Chaugule wrote:

quoted

So, in order to get an mbox->dev for ACPI platforms, we'd need an
entry in the DSDT table. That seems rather pointless, since the DSDT
is reserved for devices and is supposed to be OS agnostic. Since the
mailbox controller itself is not really a "device" with a resource
descriptor, I dont see the point in adding a dummy DSDT entry for the
sake of getting this `struct device`. Also, I'm told adding new
entries to this table requires registering a unique 4 character
identifier and approval from some committees. If there are other ways
to get this structure I'd like to hear about it.

The other alternative would be to piggy back on the ACPI CPU detection
code, which looks for the ACPI0007 device node in the DSDT and use
that as the mbox controller device. This node is already registered
and is an established method to detect CPUs. But I'm not sure what
happens when CPUs are hotplugged off, we surely dont want mailbox
clients such as PCC to break.

The main question here is whether you expect having to support multiple
mailbox devices in an ACPI system. If you think there is never more than
one, you wouldn't need a DSDT entry, but if you can end up in a situation
where another device needs to specify which mailbox it is using, then
you need that entry anyway.

At this point, I dont see the need for multiple mailbox devices. But
I'm not seeing why we'd need a DSDT entry only if there are more than
one mailbox devices? I'd obviously prefer not having a DSDT entry for
this, and the patch I posted is the only way I could see to keep DT
and ACPI mbox supported at runtime without DSDT involved. Please let
me know if there are better ways.

Cheers,
Ashwin