On Thu, 17 Apr 2014, Joel Fernandes wrote:
> On 04/17/2014 09:40 AM, Thomas Gleixner wrote:
> > + /*
> > + * If the channel is active, we need to double read as we
> > + * might see half updated data. We limit this to 5
> > + * attempts. If that fails we return -EBUSY and let the caller
> > + * deal with it.
> > + */
> > + dat = edma_read(ctlr, offs);
> > + for (i = 0; i < 5; i++) {
> > + u32 tmp = edma_read(ctlr, offs);
> > +
> > + if (tmp == dat) {
> > + *pos = dat;
> > + return 0;
> > + }
> > + dat = tmp;
> > + }
> > + return -EBUSY;
> > }
> > -EXPORT_SYMBOL(edma_get_position);
> >
> > /**
>
> The access is synchronized though and you shouldn't see unreliable
> results, unless you _are_ seeing unreliable results in which case it
> could be a silicon issue.
>
> The original code also doesn't have the double read so I would just drop
> that, unless you are seeing reliability issues.
>
> Here's a thread that discusses it and the end conclusion is that it
> should be fine (Kyle Kastile is one of the EDMA hardware designers).
> http://e2e.ti.com/support/dsp/davinci_digital_media_processors/f/99/t/66011.aspx
Gah, right I saw the issues with the original code, which does a
memcpy_fromio which copies byte by byte. Did not think about retrying
with the edma_read() based one.
Thanks,
tglx
--
To unsubscribe from this list: send the line "unsubscribe dmaengine" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
