Re: [PATCH 2/4] x86-64: Calgary IOMMU - move valid_dma_direction into the callers

[Jeff Garzik <jeff@garzik.org>]

Muli Ben-Yehuda wrote:
> On Thu, May 25, 2006 at 12:35:07AM -0400, Jeff Garzik wrote:
>> Jon Mason wrote:
>>> >From Andi Kleen's comments on the original Calgary patch, move
>>> valid_dma_direction into the calling functions.
>>>
>>> Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com>
>>> Signed-off-by: Jon Mason <jdmason@us.ibm.com>
>> Even though BUG_ON() includes unlikely(), this introduces additional 
>> tests in very hot paths.
> 
> Are they really very hot? I mean if you're calling the DMA API, you're
> about to frob the hardware or have already frobbed it - does this
> check really matter?

When you are adding a check that will _never_ be hit in production, to 
the _hottest_ paths in the kernel, you can be assured it matters...


>> _Why_ do we need this at all?
> 
> It was helpful for us during the dma-ops work and Calgary bringup and
> Andi requested that we move it from Calgary to common code. I think
> we're fine with dropping it if that's the consensus, but it did catch
> a few bugs early on and the cost is tiny.

Key phrase:  "early on"

These checks will only be useful during _early_ development of a new DMA 
platform.  For _100%_ of the real world users, these checks are useless. 
  Not 99%, 100%.

	Jeff