Re: [RFC v2 6/8] gpu: drm: tegra: Remove redundant host1x

[Stephen Warren <swarren@wwwdotorg.org>]

On 12/13/2012 01:57 AM, Thierry Reding wrote:
> On Thu, Dec 13, 2012 at 10:48:55AM +0200, Terje Bergström wrote:
>> On 12.12.2012 18:08, Thierry Reding wrote:
>>> I've briefly discussed this with Stephen on IRC because I
>>> thought I had remembered him objecting to the idea of adding a
>>> dummy device just for this purpose. It turns out, however, that
>>> what he didn't like was to add a dummy node to the DT just to
>>> make this happen, but he has no (strong) objections to a dummy
>>> platform device.
>>> 
>>> While I'm not very happy about that solution, I've been going
>>> over it for a week now and haven't come up with any better
>>> alternative that doesn't have its own disadvantages. So perhaps
>>> we should go ahead and implement that. For the host1x driver
>>> this really just means creating a platform device and adding it
>>> to the system, with some of the fields tweaked to make things
>>> work.
>> 
>> Even the virtual device is not too beautiful. The problem is that
>> the virtual device is not physical parent for DC, HDMI, etc, so 
>> dev_get_drvdata(pdev->dev.parent) returns the data from host1x
>> device, not the virtual device.
>> 
>> We'll post with something that goes around this, but it's not
>> going to be too pretty. Let's try to find the solution once we
>> get the code out.
> 
> After some more discussion with Stephen on IRC we came to the
> conclusion that the easiest might be to have tegra-drm call into
> host1x with something like:
> 
> void host1x_set_drm_device(struct host1x *host1x, struct device
> *dev);

If host1x is registering the dummy device that causes tegradrm to be
instantiated, then presumably there's no need for the API above, since
host1x will already have the struct device * for tegradrm, since it
created it?

> Once the dummy device has been properly set up and have each client
> use
> 
> struct device *host1x_get_drm_device(struct host1x *host1x);
> 
> to obtain a pointer to the dummy device, such that it can receive
> the driver-private data using dev_get_drvdata() on the returned
> device. As long as tegra-drm hasn't registered with host1x yet, the
> above function could return ERR_PTR(-EPROBE_DEFER), so that
> dependencies are automatically handled. This is required because,
> tegra-drm not being the parent of the child devices, it can be
> guaranteed that it is probed before any of the children.
> 
> That way we should be able to get around any circular
> dependencies, since we only call into host1x from tegra-drm but not
> the other way around.