Re: [PATCH v3 18/19] iommu: exynos: init from dt-specific callback instead of initcall

[Will Deacon <will.deacon-5wv7dgnIgG8@public.gmane.org>]

On Mon, Dec 15, 2014 at 05:27:30PM +0000, Laurent Pinchart wrote:
> On Monday 15 December 2014 17:17:00 Will Deacon wrote:
> > On Sun, Dec 14, 2014 at 12:45:36PM +0000, Laurent Pinchart wrote:
> > > On Wednesday 19 November 2014 12:15:47 Marek Szyprowski wrote:
> > > > +static int __init exynos_iommu_of_setup(struct device_node *np)
> > > > +{
> > > > +	struct platform_device *pdev;
> > > > +
> > > > +	if (!init_done)
> > > > +		exynos_iommu_init();
> > > > +
> > > > +	pdev = of_platform_device_create(np, NULL,
> > > > platform_bus_type.dev_root);
> > > > +	if (IS_ERR(pdev))
> > > > +		return PTR_ERR(pdev);
> > > 
> > > If we end up having to create the IOMMU platform devices from within the
> > > drivers, the introduction of IOMMU_OF_DECLARE starts to feel like a
> > > workaround to me. I wonder whether it wouldn't then be better to let the
> > > driver core instantiate the IOMMU platform device from DT as for all
> > > other devices, and use device notifiers to defer probe of the bus masters
> > > until the required IOMMU(s) are registered.
> > > 
> > > Will, what's your opinion on that ?
> > 
> > Creating the platform device manually for the IOMMU is indeed grotty, but I
> > don't really understand why it's needed. Interrupt controllers, for example,
> > seem to get by without one.
> 
> There's several reasons, one of the most compelling ones I can think of at the 
> moment is runtime PM. IRQ controllers close to the CPU use CPU PM notifiers 
> instead. Note that IRQ controllers that are further away from the CPU (such as 
> GPIO-based IRQ controllers) are real platform devices and use runtime PM.

Ok, that's a good point, but the IOMMU will still probe later anyway, right?

> IOMMUs are not as low-level as system interrupt controllers or system clocks. 
> I'm beginning to agree with Thierry that they should be treated as normal 
> platform devices as they're not required earlier than probe time of their bus 
> master devices.

Well, I think you'd have to propose patches for discussion since I'm
certainly not wed to the current approach; I just want something that
allows of_{dma,iommu}_configure to run with the information it needs.

The usual answer is `we should model buses properly', but that's really
not practical afaict.

Will