Re: [PATCH v3 4/6] irqchip/renesas-rzv2h: Make IRQ type handling range-aware

[Thomas Gleixner <tglx@kernel.org>]

On Mon, Feb 09 2026 at 10:41, Prabhakar wrote:
> From: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com>
>
> Refine IRQ type handling to explicitly bound IRQ and TINT ranges and
> dispatch based on the hardware IRQ number.

Changelog should not start with an explanation of the what. See

https://www.kernel.org/doc/html/latest/process/maintainer-tip.html#changelog

> This restructures the logic to clearly separate NMI, IRQ, and TINT

Restructure the ...


> @@ -175,18 +177,27 @@ static void rzv2h_icu_eoi(struct irq_data *d)
>  	u32 bit;
>  
>  	scoped_guard(raw_spinlock, &priv->lock) {
> -		if (hw_irq >= ICU_TINT_START) {
> -			tintirq_nr = hw_irq - ICU_TINT_START;
> -			bit = BIT(tintirq_nr);
> -			if (!irqd_is_level_type(d))
> -				writel_relaxed(bit, priv->base + priv->info->t_offs + ICU_TSCLR);
> -		} else if (hw_irq >= ICU_IRQ_START) {
> +		switch (hw_irq) {
> +		case 0:
> +			/* Clear NMI */
> +			writel_relaxed(ICU_NSCLR_NCLR, priv->base + ICU_NSCLR);
> +			break;
> +		case ICU_IRQ_START ... ICU_IRQ_LAST:
> +			/* Clear IRQ */
>  			tintirq_nr = hw_irq - ICU_IRQ_START;
>  			bit = BIT(tintirq_nr);
>  			if (!irqd_is_level_type(d))
>  				writel_relaxed(bit, priv->base + ICU_ISCLR);
> -		} else {
> -			writel_relaxed(ICU_NSCLR_NCLR, priv->base + ICU_NSCLR);
> +			break;
> +		case ICU_TINT_START ... ICU_TINT_LAST:
> +			/* Clear TINT */
> +			tintirq_nr = hw_irq - ICU_TINT_START;
> +			bit = BIT(tintirq_nr);
> +			if (!irqd_is_level_type(d))
> +				writel_relaxed(bit, priv->base + priv->info->t_offs + ICU_TSCLR);
> +			break;
> +		default:
> +			break;
>  		}
>  	}

TBH, I personally do not care about the performance of your platform at
all, but are you really serious about having a switch case like that in
a hotpath function?

Instead of sprinkling this switch case gunk all over the place you can
simply have separate interrupt chips for each region and install the
proper one at setup time. Then the functions are clearly separated and
just handling the type they are written for and nothing else. No?

Thanks,

        tglx