Re: [V9 PATCH 2/2] irqchip: gicv2m: Add supports for ARM GICv2m MSI(-X)

[Suravee Suthikulanit <suravee.suthikulpanit@amd.com>]

On 10/31/2014 4:40 AM, Thomas Gleixner wrote:
> On Fri, 31 Oct 2014, suravee.suthikulpanit@amd.com wrote:
>> +/*
>> + * alloc_msi_irq - Allocate MSIs from available MSI bitmap.
>> + * @data: Pointer to v2m_data
>> + * @nvec: Number of interrupts to allocate
>> + * @irq: Pointer to the allocated irq
>> + *
>> + * Allocates interrupts only if the contiguous range of MSIs
>> + * with specified nvec are available. Otherwise return the number
>> + * of available interrupts. If none are available, then returns -ENOENT.
>
> And the exact purpose of returning the number of available interrupts
> is?

Initially, I intended to use this function to allocate irqs for both MSI 
and multi-MSI case. But I'll simplify this and revisit it again when 
adding the multi-MSI.

>
>> +	virq = __irq_domain_alloc_irqs(v2m->domain, hwirq,
>> +				       1, NUMA_NO_NODE, v2m, true);
>
> And surely the ability of alloc_msi_irq() to allocate a contiguous
> vector space is required to satisfy an hardcoded allocation of ONE
> interrupt.
>
> What is guaranteeing that the caller only requests a single interrupt?

Since this is calling from gicv2m_setup_msi_irq(), it should only setup 
1 MSI interrupt.

 >[...]
>> +err_out:
>
> Single error exit which undoes the stuff in the same order it got
> initialized is just plain wrong. Ever looked at proper error exits in
> other kernel files?
>

I'll clean this up in V10. Thanks for pointing this out.

Suravee