[PATCH v2 2/9] mailbox: Add NVIDIA Tegra XUSB mailbox driver

Stephen Warren swarren at wwwdotorg.org
Wed Aug 27 11:19:21 PDT 2014

On 08/27/2014 12:13 PM, Andrew Bresticker wrote:
> On Wed, Aug 27, 2014 at 10:50 AM, Stephen Warren <swarren at wwwdotorg.org> wrote:
>> On 08/27/2014 11:38 AM, Andrew Bresticker wrote:
>>> On Mon, Aug 25, 2014 at 12:01 PM, Stephen Warren <swarren at wwwdotorg.org>
>>> wrote:
>>>> On 08/18/2014 11:08 AM, Andrew Bresticker wrote:
>>>>> +static int tegra_xusb_mbox_probe(struct platform_device *pdev)
>>>>> +       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
>>>>> +       if (!res)
>>>>> +               return -ENODEV;
>>>> Should devm_request_mem_region() be called here to claim the region?
>>> No, the xHCI host driver also needs to map these registers, so they
>>> cannot be mapped exclusively here.
>> That's unfortunate. Having multiple drivers with overlapping register
>> regions is not a good idea. Can we instead have a top-level driver map all
>> the IO regions, then instantiate the various different sub-components
>> internally, and divide up the address space. Probably via MFD or similar.
>> That would prevent multiple drivers from touching the same register region.
> Perhaps I'm misunderstanding, but I don't see how MFD would prevent us
> from having to map this register space in two different locations -
> the XUSB FPCI address space cannot be divided cleanly between host and
> mailbox registers.  Or are you saying that there should be a separate
> device driver that exposes an API for accessing this register space,
> like the Tegra fuse or PMC drivers?

With MFD, there's typically a top-level driver for the HW module (or 
register space) that gets instantiated by the DT node. This driver then 
instantiates all the different sub-drivers that use that register space, 
and provides APIs for the sub-drivers to access the registers (either 
custom APIs or more recently by passing a regmap object down to the 

This top-level driver is the only driver that maps the space, and can 
manage sharing the space between the various sub-drivers.

That said, I haven't noticed many MFD drivers for MMIO devices. I 
certainly have seen multiple different drivers just re-mapping shared 
registers for themselves. It is simpler and does work. However, people 
usually mutter about it when it happens, since it's not clear which 
drivers are using what from the IO mapping registry. Using MFD or 
similar would allow the sharing to work in a clean fashion.

More information about the linux-arm-kernel mailing list