Re: [patch v3 21/36] Hexagon: Add SMP support

[Benjamin Herrenschmidt]

On Thu, 2011-09-08 at 20:09 -0500, Richard Kuo wrote:

> +irqreturn_t handle_ipi(int irq, void *desc)
> +{
> +	int cpu = smp_processor_id();
> +	struct ipi_data *ipi = &per_cpu(ipi_data, cpu);
> +	unsigned long ops;
> +
> +	while ((ops = xchg(&ipi->bits, 0)) != 0)
> +		__handle_ipi(&ops, ipi, cpu);
> +	return IRQ_HANDLED;
> +}

So on the consumer side you are using xchg() ...

> +void send_ipi(const struct cpumask *cpumask, enum ipi_message_type msg)
> +{
> +	unsigned long flags;
> +	unsigned long cpu;
> +	unsigned long retval;
> +
> +	local_irq_save(flags);
> +
> +	for_each_cpu(cpu, cpumask) {
> +		struct ipi_data *ipi = &per_cpu(ipi_data, cpu);
> +
> +		spin_lock(&ipi->lock);
> +		ipi->bits |= 1 << msg;  /*  one bit per message  */
> +		spin_unlock(&ipi->lock);
> +
> +		retval = __vmintop_post(BASE_IPI_IRQ+cpu);
> +		if (retval != 0) {
> +			printk(KERN_ERR "interrupt %ld not configured?\n",
> +				BASE_IPI_IRQ+cpu);
> +		}
> +	}
> +
> +	local_irq_restore(flags);
> +}

And on the producer side, a spinlock for one or... that looks like wheel
re-inventing to me :-)

In fact your bit iteration loop also re-invents find_*_bit interfaces.

Any reason why you don't simply use bitops or atomics here ?

Also what is your memory model ? ordered or weakly ordered ? Do you need
some kind of memory barrier between setting ipi->bits and
__vmintop_post() ?

Cheers,
Ben.


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/