On Tue, Jun 30, 2009 at 12:33 PM, Jon Smirl[off-list ref] wrote:

On Tue, Jun 30, 2009 at 12:50 PM, Grant Likely[off-list ref]=

 wrote:

quoted

On Tue, Jun 30, 2009 at 12:18 AM, Wolfram Sang[off-list ref] wr=

ote:

quoted

quoted

quoted

+
+ =A0 =A0 /* Force clear the data valid bit */
+ =A0 =A0 in_be32(&psc_dma->psc_regs->ac97_data);
+

No mutex involved here. I think this is either a separate patch or it n=

eeds at

quoted

quoted

least to be mentioned in the patch description.

Oops, that was sloppy. =A0Yes, I'll put this into a separate patch. =A0T=

hanks.

Now that you have added the mutexes, do you ever need to force clear
the valid bit?
Maybe log an error if this happens so that we can track down why.

I know exactly why it happened.  spin_event_timeout() had a bug that
would cause it to report a false positive timeout.  The bug is fixed
and queued for merging.

However, I still think this explicit clear should be applied.  It is
just a cheap register read and it prevents the driver from getting
wedged after a timeout does event, for whatever reason.

g.

--=20
Grant Likely, B.Sc., P.Eng.
Secret Lab Technologies Ltd.