Re: [Patch v4 02/12] irqchip: axi-intc: Clean up irqdomain argument and read/write

[Zubair Lutfullah Kakakhel]

Hi,

Thanks for the review.
Comments inline.

On 09/01/2016 06:15 PM, Marc Zyngier wrote:
> On 01/09/16 17:50, Zubair Lutfullah Kakakhel wrote:
> > The drivers read/write function handling is a bit quirky.
> > And the irqmask is passed directly to the handler.
> >
> > Add a new irqchip struct to pass to the handler and
> > cleanup read/write handling.
> >
> > Signed-off-by: Zubair Lutfullah Kakakhel <Zubair.Kakakhel@xxxxxxxxxx>
> >
> > ---
> > V3 -> V4
> > Better error handling for kzalloc
> > Erroring out if the axi intc is probed twice as that isn't
> > supported

...

> >  static int __init xilinx_intc_of_init(struct device_node *intc,
> >  					     struct device_node *parent)
> >  {
> > -	u32 nr_irq, intr_mask;
> > +	u32 nr_irq;
> >  	int ret;
> > +	struct xintc_irq_chip *irqc;
> > +
> > +	irqc = kzalloc(sizeof(*irqc), GFP_KERNEL);
> > +	if (!irqc)
> > +		return -ENOMEM;
> >
> > -	intc_baseaddr = of_iomap(intc, 0);
> > -	BUG_ON(!intc_baseaddr);
> > +	if (xintc_irqc) {
> > +		pr_err("%s: Multiple instances of axi_intc aren't supported\n");
> > +		ret = -EINVAL;
> > +		goto err_alloc;
>
> How about testing the variable first and error-ing early, rather than
> performing the allocation and undoing it later?

Sure. Thanks

> > +	} else {
> > +		xintc_irqc = irqc;
> > +	}
> > +
> > +	irqc->base = of_iomap(intc, 0);
> > +	BUG_ON(!irqc->base);
> >
> >  	ret = of_property_read_u32(intc, "xlnx,num-intr-inputs", &nr_irq);
> >  	if (ret < 0) {
> >  		pr_err("%s: unable to read xlnx,num-intr-inputs\n", __func__);
> > -		return ret;
> > +		goto err_alloc;
> >  	}
> >
> > -	ret = of_property_read_u32(intc, "xlnx,kind-of-intr", &intr_mask);
> > +	ret = of_property_read_u32(intc, "xlnx,kind-of-intr", &irqc->intr_mask);
> >  	if (ret < 0) {
> >  		pr_err("%s: unable to read xlnx,kind-of-intr\n", __func__);
> > -		return ret;
> > +		goto err_alloc;
> >  	}
> >
> > -	if (intr_mask >> nr_irq)
> > +	if (irqc->intr_mask >> nr_irq)
> >  		pr_warn("%s: mismatch in kind-of-intr param\n", __func__);
> >
> >  	pr_info("%s: num_irq=%d, edge=0x%x\n",
> > -		intc->full_name, nr_irq, intr_mask);
> > +		intc->full_name, nr_irq, irqc->intr_mask);
> >
> > -	write_fn = intc_write32;
> > -	read_fn = intc_read32;
> > +	irqc->read = intc_read32;
> > +	irqc->write = intc_write32;
> >
> >  	/*
> >  	 * Disable all external interrupts until they are
> >  	 * explicity requested.
> >  	 */
> > -	write_fn(0, intc_baseaddr + IER);
> > +	xintc_write(irqc, IER, 0);
> >
> >  	/* Acknowledge any pending interrupts just in case. */
> > -	write_fn(0xffffffff, intc_baseaddr + IAR);
> > +	xintc_write(irqc, IAR, 0xffffffff);
> >
> >  	/* Turn on the Master Enable. */
> > -	write_fn(MER_HIE | MER_ME, intc_baseaddr + MER);
> > -	if (!(read_fn(intc_baseaddr + MER) & (MER_HIE | MER_ME))) {
> > -		write_fn = intc_write32_be;
> > -		read_fn = intc_read32_be;
> > -		write_fn(MER_HIE | MER_ME, intc_baseaddr + MER);
> > +	xintc_write(irqc, MER, MER_HIE | MER_ME);
> > +	if (!(xintc_read(irqc, MER) & (MER_HIE | MER_ME))) {
> > +		irqc->read = intc_read32_be;
> > +		irqc->write = intc_write32_be;
> > +		xintc_write(irqc, MER, MER_HIE | MER_ME);
> >  	}
> >
> > -	/* Yeah, okay, casting the intr_mask to a void* is butt-ugly, but I'm
> > -	 * lazy and Michal can clean it up to something nicer when he tests
> > -	 * and commits this patch.  ~~gcl */
> >  	root_domain = irq_domain_add_linear(intc, nr_irq, &xintc_irq_domain_ops,
> > -							(void *)intr_mask);
> > +					    irqc);
>
> What if the domain allocation fails? You've now configured the HW for
> something you can't use. What are the side effects? Hint: handle
> everything that can fail first, and only then poke the HW.

Thanks for pointing it out. I'll add an error check.

> >  	irq_set_default_host(root_domain);
> >
> >  	return 0;
> > +
> > +err_alloc:
> > +	xintc_irqc = NULL;
> > +	kfree(irqc);
> > +	return ret;
> > +
> >  }
> >
> >  IRQCHIP_DECLARE(xilinx_intc, "xlnx,xps-intc-1.00.a", xilinx_intc_of_init);
>
> Instead of posting 3 versions in 3 days, please take the time to
> correctly address the comments, and review your own code before
> re-posting it. Rushing to get it merged for 4.9 is really not the best
> approach.

Apologies for the spam. Combination of some free time this week + window of opportunity.
To be fair, AFAIK, this driver has lived in arch/microblaze without receiving a full thorough
review by any irqchip maintainer.

Hence the various missing bits. e.g. the root_domain error check didn't exist before.
And I didn't see it.

Regards,
ZubairLK

> Thanks,
>
> 	M.