Re: [PATCH v3 -tip 1/5] x86, MSI: Support multiple MSIs in presense of IRQ remapping

[Ingo Molnar <mingo@kernel.org>]

* Alexander Gordeev <agordeev@redhat.com> wrote:

> On Tue, Oct 02, 2012 at 06:55:18AM +0200, Ingo Molnar wrote:
> 
> Thanks for the review, Ingo.
> 
> > > @@ -584,8 +586,12 @@ int __irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node,
> > >  #define irq_alloc_desc_from(from, node)		\
> > >  	irq_alloc_descs(-1, from, 1, node)
> > >  
> > > +#define irq_alloc_descs_from(from, cnt, node)	\
> > > +	irq_alloc_descs(-1, from, cnt, node)
> > > +
> > 
> > Please use inlines instead of macros. Might transform the one 
> > above it as well in the process.
> 
> You mean here do not introduce irq_alloc_descs_from, but rather use
> irq_alloc_descs() directly?

My suggestion is to add irq_alloc_descs_from() as a (very 
simple) inline function and change irq_alloc_desc_from() to be 
an inline function as well.

> > > +int irq_can_alloc_irqs(unsigned int from, unsigned int cnt)
> > > +{
> > > +	unsigned int start;
> > > +	int ret = 0;
> > > +
> > > +	if (!cnt)
> > > +		return -EINVAL;
> > > +
> > > +	mutex_lock(&sparse_irq_lock);
> > > +	start = bitmap_find_next_zero_area(allocated_irqs, IRQ_BITMAP_BITS,
> > > +					   from, cnt, 0);
> > > +	mutex_unlock(&sparse_irq_lock);
> > > +	if (start + cnt > nr_irqs)
> > > +		ret = irq_can_expand_nr_irqs(start + cnt);
> > > +	return ret;
> > 
> > How is this supposed to work wrt. races?
> 
> It is not supposed. Just a quick check if there are enough bits before an
> attempt to allocate memory in __create_irqs(). Otherwise __create_irqs()
> might allocate irq_cfg's, then realize there are no bits, then deallocate
> and fail.
>
> But strictly speaking, irq_can_alloc_irqs() is unnecessary.

Why complicate it if it's unnecessary? The function is inviting 
wrong logic: it *cannot* tell whether there are enough bits, 
because the check is racy.

So I'd suggest to keep this out - this will further simplify the 
patches.

Thanks,

	Ingo