Re: [PATCH 4/4] omap: Make get_irqnr_and_base common for mach-omap2 multiboot

[Russell King - ARM Linux <linux@arm.linux.org.uk>]

On Fri, Jan 15, 2010 at 05:35:20PM -0800, Tony Lindgren wrote:
> -#ifndef CONFIG_ARCH_OMAP4
> +#if defined(CONFIG_ARCH_OMAP2420) || defined(CONFIG_ARCH_OMAP2430) || \
> +	defined(CONFIG_ARCH_OMAP34XX)
> +
> +#define INTCPS_SIR_IRQ_OFFSET	0x0040		/* Active interrupt offset */
> +#define	ACTIVEIRQ_MASK		0x7f		/* Active interrupt bits */
> +
> +omap_irq_base:	.word	0x0

You have no idea where this word will be placed - it could be in the middle
of a read only section, which might really be read only.  It might also be
out of reach of the "ldr \base, [pc, #offset]" instructions you're using
below - which is what an "ldr \base, address" instruction really is.

> +
>  		.macro	get_irqnr_and_base, irqnr, irqstat, base, tmp
> -		ldr	\base, =OMAP2_VA_IC_BASE
> -		ldr	\irqnr, [\base, #0x98] /* IRQ pending reg 1 */
> +		ldr	\base, omap_irq_base
> +		cmp	\base, #0		@ is irq base configured?
> +		bne	9998f			@ already configured

First thing, why not use get_irqnr_preamble to load the base address?
That means you only do all this computation once per group of IRQs
processed.

Secondly, is this really worth the overhead, or can't you move
omap_irq_base into the data section, and have your IRQ initialization
function set the pointer?

You then don't have to make these two tests every time you check for
an interrupt/enter to process interrupts.