On Tue Aug 19, 2014 at 17:43:29 +0200, Martin Schröder wrote:
Am 19.08.2014 um 14:25 schrieb Martin Schröder:
Hi Adam,
I tracked the problem down to l4linux/arch/l4/kernel/arch-arm/setup.c: In static customize_machine(void) either machine_desc->init_machine() is called and the static/generic platform devices are added (and the DT disregarded). If init_machine is not set the DT is parsed instead by of_platform_populate. So I changed in l4.c the .init_machine property to NULL and got all the devices generated from the DT.
While trying to get fec_main.c probed, I recognized that skipping .init_machine is not a good idea, since dma_alloc_coherent fails while allocating the memory for the network buffer descriptors in fec_main.c. This is due to the missing call of l4io_request_iomem in the "dmamem" device L4X_DEVICE_CB. So it seems that both init_machine() and of_platform_populate() is needed. If I remove the "else" line from the if statement in customize_machine the dma_alloc_coherent succeeds again.
Ok, I should probably rip this all out and do DT only.
Now at least I got the pinctrl-mxs.c working (utilizes only memory). But gpoi-mxs.c does not get it's irq and fails probing. What is the correct interrupt-parent definition if the interrupt controller is outside of l4linux?
Maybe the interrupt-parent adjustment could also be done "in memory" after loading the dtb?
I have some still hacky code on my disk which I think handles this. I have to check and clean this up to commit it.
Adam