Hello,
On 20 June 2014 16:49, Arnd Bergmann arnd@arndb.de wrote:
On Friday 20 June 2014 15:29:18 Ashwin Chaugule wrote:
On 20 June 2014 15:08, Arnd Bergmann arnd@arndb.de wrote:
On Friday 20 June 2014 14:55:16 Ashwin Chaugule wrote:
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.
It's mostly a matter of consistency: We can have multiple interrupt controllers, pin controllers, clock controllers, dma engines, etc, and in the DT case we use references to the nodes wherever we have other devices referring to a mailbox name.
I believe Intel's embedded chips are moving in the same direction with their ACPI support. If the ACPI spec gains support for mailbox devices, locking them into having only a single device may be a problem later for them.
Note that "device" here doesn't have to mean a platform device that is instantiated from DSDT, it can be any mailbox provider that is registered in an arbitrary way, as long as you have a method to map back from the (consumer-device, name-string) tuple back to the (provider, channel) tuple. I have read your patch again now and noticed that you actually tried to do this, but unfortunately you got it wrong by requiring the consumer to fill out the name of the provider in the request. You can't do that, because it's not generic enough to support devices that can be reused, and it means that drivers using the API are never portable between DT and ACPI. You have to get rid of the "ctrl_name" field in the mbox_client structure and change the lookup to be based only on cd->dev and cl->chan_name, using whatever tables you have available in ACPI.
I think you looked at the previous version of the patch. I'm attaching the latest version here again FWIW. In this version, I removed the "ctrl_name" field and rely on the cl->chan_name to provide the info as described in Jassi' original patch.
linux/mailbox_client.h
18 * struct mbox_client - User of a mailbox 19 * @dev: The client device 20 * @chan_name: The "controller:channel" this client wants
Instead of dev, I added a name string to the mbox controller structure. So now the client gets its channel by requesting "controller:channel" where controller should match with mbox->name and channel becomes an index into mbox->chans[].
Cheers, Ashwin