Re: [Xen-devel] [PATCH v2 RFC 2/8] arm: introduce a global dma_ops pointer

[Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>]

On Wed, 31 Jul 2013, Russell King - ARM Linux wrote:
> On Wed, Jul 31, 2013 at 08:54:06PM +0100, Stefano Stabellini wrote:
> > Initially set dma_ops to arm_dma_ops.
> > Use dma_ops instead of arm_dma_ops in dmabounce.
> > 
> > Signed-off-by: Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>
> > CC: will.deacon@xxxxxxx
> > CC: linux@xxxxxxxxxxxxxxxx
> 
> If you're using swiotlb, there's little point in converting dmabounce
> to be able to use it, because it's basically providing the same
> functionality - dmabounce is there to do software buffer bouncing within
> the DMA to move streaming buffers from DMA-inaccessible memory into
> DMA-accessible memory.
> 
> It's original use is with older SoCs with buggy DMA (eg, those which
> can only address alternate 1MB chunks of memory for example) but also
> got used in situations where alternative solutions would've been better
> (like using swiotlb.)  I've been discouraging its use as it's suffered
> from memory exhaustion problems (there's a number of threads and bug
> reports which were never solved about IXP4xx(?) platforms suffering
> this due to this bouncing.)

OK, I'll let dmabounce keep using arm_dma_ops directly (instead of
dma_ops).

Should I add "depends on !DMABOUNCE" to the SWIOTLB Kconfig entry too?

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel