Re: [PATCH 1/3] ARM: tegra: Add AHB driver

[Arnd Bergmann <arnd-r2nGTMty4D4@public.gmane.org>]

On Monday 23 April 2012, Hiroshi DOYU wrote:
> The AHB Bus conforms to the AMBA Specification (Rev 2.0) Advanced
> High-performance Bus (AHB) architecture.
> 
> Both Tegra20/30 have this.
> 
> Signed-off-by: Hiroshi DOYU <hdoyu-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>

Please explain in the text above why it's a good idea to have this
driver.


> @@ -122,6 +123,7 @@ void __init tegra20_init_early(void)
>  	tegra_init_cache(0x331, 0x441);
>  	tegra_pmc_init();
>  	tegra_powergate_init();
> +	tegra_ahb_gizmo_init();
>  }
>  #endif
>  #ifdef CONFIG_ARCH_TEGRA_3x_SOC
> @@ -132,5 +134,6 @@ void __init tegra30_init_early(void)
>  	tegra_init_cache(0x441, 0x551);
>  	tegra_pmc_init();
>  	tegra_powergate_init();
> +	tegra_ahb_gizmo_init();
>  }
>  #endif

Does it really have to be "early", rather than an initcall? Why?

> +
> +static inline unsigned long gizmo_readl(unsigned long offset)
> +{
> +	return readl(IO_TO_VIRT(TEGRA_AHB_GIZMO_BASE + offset));
> +}
> +
> +static inline void gizmo_writel(unsigned long value, unsigned long offset)
> +{
> +	writel(value, IO_TO_VIRT(TEGRA_AHB_GIZMO_BASE + offset));
> +}

Please change this to no longer use hardcoded addresses. A good
implementation would scan the device tree for the physical address
and then ioremap the registers in the init function, in order to
save a local __iomem pointer.

	Arnd