Re: DMA sync question

[David Dillow]

On Mon, 2010-10-18 at 06:11 -0400, Bart Van Assche wrote:
> On Sun, Oct 17, 2010 at 5:25 AM, David Dillow <dillowda@xxxxxxxx> wrote:
> > [ ... ]
> > +static int srp_response_common(struct srp_target_port *target, s32 req_delta,
> > +                              void *rsp, int len)
[ ... ]
> > +
> > +       ib_dma_sync_single_for_cpu(dev, iu->dma, len, DMA_TO_DEVICE);
> > +       memcpy(iu->buf, rsp, len);
> > +       ib_dma_sync_single_for_device(dev, iu->dma, len, DMA_TO_DEVICE);
> > [ ... ]

> Hi Dave,
> 
> As far as I can see the above sync call applies to a buffer in the
> tx_ring[]. Data in that buffer is only modified by the CPU and never
> by the HCA. So why is the above sync call present ? Is that call
> necessary ?

In this instance, I copied existing practice from srp_queuecommand().

Documentation/DMA-API.txt seems to indicate we don't need the initial
*_sync_for_cpu() since the device never writes to the buffer, but I seem
to have an old memory that says we should do it anyways.

Perhaps LKML can clear it up -- do we need to call *_sync_for_cpu()
after a *sync_for_device(), even if the device never writes to the
buffer?
-- 
Dave Dillow
National Center for Computational Science
Oak Ridge National Laboratory
(865) 241-6602 office


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/