Re: [PATCH] Containment measures for slab objects on scatter gather lists

[Russell King <rmk+lkml@arm.linux.org.uk>]

On Sat, Jun 30, 2007 at 12:11:38AM +0100, Alan Cox wrote:
> > DMA to or from memory should be done via the DMA mapping API.  If we're
> > DMAing to/from a limited range within a page, either we should be using
> > dma_map_single(), or dma_map_page() with an appropriate offset and size.
> 
> If those ranges overlap a cache line then the dma mapping API will not
> save your backside.

There's nothing much that the DMA API can do though.  Consider DMA
to a result buffer which is, eg, only 16 bytes in size.  So you get
passed a size of '16' to the DMA API.  What should you do at this
point?  BUG()?

What if you have 64 or 128 byte cache lines?

> > sizes, but they do happen.  We handle this on ARM by writing back
> > the overlapped lines and invalidating the rest before the DMA operation
> > commences, and hope that the overlapped lines aren't touched for the
> > duration of the DMA.)
> 
> The combination of "hope" and "DMA" isn't a good one for stable system
> design. In this situation we should be waving large red flags

I agree.

However, I don't think this is an issue for the DMA API to handle; it's
something that driver authors need to be aware of.  If they wish to do
a DMA to a kmalloc'd buffer or even a page, we could require that
offsets and sizes are cache line aligned.

However, remember that turning on slab debugging turns off cache line
alignment, so imposing such a requirement implies that the slab
debugging will break DMA, or driver authors also have to be aware of
that and do their own alignment internally, *or* we provide an allocator
which does unconditionally align.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of: