Re: [PATCH] x86/irq: Precompute nr_legacy_irqs() for performance optimization

[Jiri Slaby]

Hi,

On 01. 11. 25, 12:26, ray wrote:
> Precompute the return value of nr_legacy_irqs() in init_ISA_irqs(),
> init_IRQ(), and native_init_IRQ() functions to avoid repeated function
> calls in loops, improving performance.
>
> Changes made:
> - Precompute nr_legacy_irqs() return value in three functions
> - Use local variables instead of function calls in loops
> - Maintain original logic unchanged
>
> Signed-off-by: ray <veidongray@xxxxxx>

Is this how you sign legal documents?

> index 6ab9eac64670..bb7cbf99f65c 100644
> --- a/arch/x86/kernel/irqinit.c
> +++ b/arch/x86/kernel/irqinit.c

...
> @@ -66,7 +66,9 @@ void __init init_ISA_irqs(void)
>   
>   	legacy_pic->init(0);
>   
> -	for (i = 0; i < nr_legacy_irqs(); i++) {
> +	nr_legacy = nr_legacy_irqs();
> +
> +	for (i = 0; i < nr_legacy; i++) {

nr_legacy_irqs() is a static inline, so this was not a call per se. 
Instead, this resolves to 2 memory reads. Is the change worth it at all, 
provided these are NOT a fast path?

> @@ -106,7 +112,9 @@ void __init native_init_IRQ(void)
>   
>   	lapic_assign_system_vectors();
>   
> -	if (!acpi_ioapic && !of_ioapic && nr_legacy_irqs()) {
> +	nr_legacy = nr_legacy_irqs();
> +
> +	if (!acpi_ioapic && !of_ioapic && nr_legacy) {

What's the added value in here at all?

thanks,
--
js
suse labs