> From: Steve Calfee
> On Wed, Dec 18, 2013 at 3:24 AM, David Laight <david.lai...@aculab.com> wrote:
> >
> > This saves a kzalloc() call on every transfer and some memory
> > indirections.
> >
> > The only possible downside is for isochronous tranfers with 64 td
> > when the allocate is 8+4096 bytes (on 64bit systems) so requires
> > an additional page.
...
> Are you sure this will not cause weird bugs on non-cache coherent
> systems? You are smashing two memory allocations into one. This does
> not insure that the memory in each allocation is cache line aligned
> does it? Arm and Mips systems don't like having two i/o actions or cpu
> and i/o activity on data sharing a cache line.
These are both internal data structures and not visible to the hardware.
David
