PCI DAC DMA APIs

PCI DAC DMA APIs

[Jan Beulich]

While the kernel headers provide for this, there don't appear to be any
in-tree users (which seems contrary to general Linux policies). Would there
be objections to remove all of these?

Jan

Re: PCI DAC DMA APIs

[Andi Kleen]

On Thursday 15 March 2007 13:38, Jan Beulich wrote:
> While the kernel headers provide for this, there don't appear to be any
> in-tree users (which seems contrary to general Linux policies). Would there
> be objections to remove all of these?

Would be fine for me.

I think the original idea was to optimize for some SPARC systems, but they
were never really used for that.

x86 never needed them.

-Andi

Re: PCI DAC DMA APIs

[Jesse Barnes]

On Thursday, March 15, 2007 5:38 am Jan Beulich wrote:
> While the kernel headers provide for this, there don't appear to be
> any in-tree users (which seems contrary to general Linux policies).
> Would there be objections to remove all of these?

It should be safe to kill them, but I remember arguing with davem about 
this stuff in the past...

Jesse

Re: PCI DAC DMA APIs

[Christoph Hellwig]

On Thu, Mar 15, 2007 at 12:38:13PM +0000, Jan Beulich wrote:
> While the kernel headers provide for this, there don't appear to be any
> in-tree users (which seems contrary to general Linux policies). Would there
> be objections to remove all of these?

They should go away.  Having them in for more than five years without
any users is almost a guarantee for bitrot.

Re: PCI DAC DMA APIs

[David Miller]

From: Christoph Hellwig <hch@infradead.org>
Date: Thu, 15 Mar 2007 19:18:34 +0000

> On Thu, Mar 15, 2007 at 12:38:13PM +0000, Jan Beulich wrote:
> > While the kernel headers provide for this, there don't appear to be any
> > in-tree users (which seems contrary to general Linux policies). Would there
> > be objections to remove all of these?
> 
> They should go away.  Having them in for more than five years without
> any users is almost a guarantee for bitrot.

Yes, probably we should get rid of them.

The idea wasn't sparc optimizations, it was for things like those
Dolphin clustering cards that essentially want to get at all of
physical memory from the PCI card.

The IOMMU is a limited resource, so at the expense of lack of
prefetching and write caching we provide a way to do unlimited DMA
mapping with 64-bit DAC addresses.

None of these drivers ever got integrated, so it's a total loss.

Someone will complain when we pull it out, but fsck them, they
had years to do something about this. :)