Re: Intel IOMMU (and IOMMU for Virtualization) performances

[Andi Kleen]

On Mon, Jun 09, 2008 at 06:36:29PM +0900, FUJITA Tomonori wrote:
> On Mon, 09 Jun 2008 10:17:11 +0200
> Andi Kleen <andi@xxxxxxxxxxxxxx> wrote:
> 
> > "Grant Grundler" <grundler@xxxxxxxxxx> writes:
> > >
> > > The historical DMA mapping "failure mode" is a kernel panic.  Resizing or
> > 
> > Hasn't been for a long time, except in some extreme cases.  All drivers
> > are expected to check return values for a long time now.
> 
> Agreed, but I think that lots of network drivers still assume that DMA
> mapping always succeeds (they don't check return values).

Those should be just fixed. 

You're right. I reviewed some new drivers and was surprised that they
even got that wrong (code reviewers were supposed to flag this in the
first place)

I would guess in practice the main offender would be tg3.c/bnx2.c.
On the other hand Intel IOMMU systems should usually use e1000e which
checks.

-Andi
--
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/