Re: [PATCH] x86, irq: Check if irq is remapped before freeing irte

[Thomas Gleixner <tglx@linutronix.de>]

On Mon, 18 Oct 2010, Yinghai Lu wrote:

> On 10/18/2010 02:17 PM, Thomas Gleixner wrote:
> > 
> > 
> > On Mon, 18 Oct 2010, Thomas Gleixner wrote:
> > 
> >> On Mon, 18 Oct 2010, Yinghai Lu wrote:
> >>>
> >>> Index: linux-2.6/drivers/pci/intr_remapping.c
> >>> ===================================================================
> >>> --- linux-2.6.orig/drivers/pci/intr_remapping.c
> >>> +++ linux-2.6/drivers/pci/intr_remapping.c
> >>> @@ -60,7 +60,7 @@ int get_irte(int irq, struct irte *entry
> >>>  	unsigned long flags;
> >>>  	int index;
> >>>  
> >>> -	if (!entry || !irq_iommu)
> >>> +	if (!entry || !irq_iommu || !irq_iommu->iommu)
> >>>  		return -1;
> >>
> >> Hmm, why do we need this? This is only called from
> >> ir_ioapic_set_affinity() and ir_msi_set_affinity().

That does not answer that question !

> >> We should never end up there when intr_remapping=off, right ?
> > 
> > Thinking more about it, this check is actively bogus. The call sites do:
> > 
> >         struct irte irte;
> > 
> >         if (get_irte(irq, &irte))
> >                 return -1;
> > 
> > So entry _CANNOT_ be NULL.
> > 
> > And in fact we should change get_irte() to
> > 
> > get_irte(struct irq_2_iommu *irq_iommu, struct irte *entry)
> > 
> > The call site already knows about it. No need to lookup irq_iommu
> > based on the irq number.
> 
> looks like all irq-irte related API could replace "int irq" to "struct irq_2_iommu *irq_iommu"
> 
> extern int get_irte(int irq, struct irte *entry);
> extern int modify_irte(int irq, struct irte *irte_modified);
> extern int alloc_irte(struct intel_iommu *iommu, int irq, u16 count);
> extern int set_irte_irq(int irq, struct intel_iommu *iommu, u16 index,
>                         u16 sub_handle);
> extern int map_irq_to_irte_handle(int irq, u16 *sub_handle);
> extern int free_irte(int irq);

Probably, but we need to figure out which functions need which checks
instead of having either redundant or superflous ones there.

Thanks,

	tglx