[alsa-devel] [PATCH] ASoC: AM3517: Fix AIC23 suspend/resume hang

[Philby John]

On Thu, 2009-11-26 at 20:52 +0530, Aggarwal, Anuj wrote:
> > -----Original Message-----
> > From: Mark Brown [mailto:broonie at opensource.wolfsonmicro.com]
> > Sent: Thursday, November 26, 2009 8:41 PM
> > To: Aggarwal, Anuj
> > Cc: 'Troy Kisky'; alsa-devel at alsa-project.org; linux-omap at vger.kernel.org;
> > Arun KS
> > Subject: Re: [alsa-devel] [PATCH] ASoC: AM3517: Fix AIC23 suspend/resume
> > hang
> > 
> > On Thu, Nov 26, 2009 at 08:26:44PM +0530, Aggarwal, Anuj wrote:
> > 
> > > driver, although I am not sure if the problem is in the audio driver.
> > > When tried to capture, using NFS as storage, it gives overrun error and
> > comes out with:
> > > arecord: pcm_read:1617: read error: Input/output error
> > 
> > > It happen always after ~20 sec, file size ~5MB. Tried with multiple
> > > configurations in arecord but no use.
> > > When tried: arecord -f cd /dev/null, it works fine. Same issue doesn't
> > > come too when I try to store the captured audio file on a MMC card.
> > 
> > > Any idea what could be the problem? Why arecord goes for a toss after a
> > > single overrun error and why it is happening always after ~20 sec? Is
> > > there something which can be tried to narrow down the problem?
> > 
> > Sounds like you've narrowed the problem down to a performance issue with
> > NFS writeout - it's probably having trouble keeping up with your I/O
> > rate.  This isn't 100% surprising with smaller systems, sometimes tuning
> > the NFS configuration can resolve the issue but sometimes the hardware
> > is just underspecified.
> [Aggarwal, Anuj] I am still surprised how this could be a NFS writeout issue
> as we are seeing a consistent read/write rate of 2Mbps over tftp. When dd
> command is used for read/write to further check NFS performance, 2Mbps for write and 4Mbps for read is observed.
> Does that still mean nfs is the culprit? What could be tweaked in audio/network driver to avoid this problem, any suggestions?

My two cents... could also try connecting the target directly to your
host machine (using nfs), through a cross-cable (one with the Tx and Rx
crossed). If no errors are seen, then it surely must be an attribution
of network congestion.

Regards,
Philby