Problem with "ARM: tegra: Move tegra_common_init to tegra_init_early"

[Colin Cross]

On Tue, Feb 22, 2011 at 7:12 PM, Stephen Warren <swarren at nvidia.com> wrote:
> The following commit makes the Tegra APB DMA engine fail to initialize correctly:
>
> commit 0cf6230af909a86f81907455eca2a5c9b8f68fe6
>
>    ARM: tegra: Move tegra_common_init to tegra_init_early
>
>    Move tegra_common_init to tegra_init_early, and set it
>    as the init_early entry in the machine struct.
>    Initializes the clocks earlier so that timers can enable
>    their clocks.
>
>    Also reorders the members in the Harmony and Trimslice
>    boards' machine structs to match the order they are
>    called in.
>
> The reason is that tegra_init_early_ calls tegra_dma_init which calls
> request_threaded_irq, which fails since the IRQ hasn't yet been marked valid;
> that only happens in tegra_init_irq, which gets called after tegra_init_early.
>
> This used to work OK, since tegra_init_early was tegra_common_init, which got
> called after tegra_init_irq, basically from the beginning of tegra_harmony_init.
>
> I tried moving the call to tegra_dma_init back to each of the very end of
> tegra_init_irq and the very beginning of tegra_harmony_init. Both of these
> resulted in a kernel that wouldn't boot.
>
> Simply removing the call to tegra_dma_init completely does allow the kernel
> to boot, but obviously, DMA isn't available.
>
> I'm a little stumped why moving tegra_dma_init causes a boot failure. Does
> anyone have any helpful ideas, or any alternative ways to fix the underlying
> problem?

I don't know why it still fails if you put it back at the beginning of
the machine init, and it boots without errors on my board if I move
tegra_dma_init to a postcore_initcall, but I don't have any drivers on
my board that use DMA.

Can you provide kernel logs?