dmatest regression in 3.10-rc1

[will.deacon@arm.com (Will Deacon)]

Hi Andy,

On Mon, May 20, 2013 at 08:52:34AM +0100, Andy Shevchenko wrote:
> On Fri, 2013-05-17 at 15:18 +0100, Will Deacon wrote: 
> > On Fri, May 17, 2013 at 01:34:23PM +0100, Vinod Koul wrote:
> > > On Thu, May 16, 2013 at 04:35:53PM +0100, Will Deacon wrote:
> > > > Right, so I think I understand what's causing this, but I'll leave it to
> > > > Andriy to suggest a fix.
> 
> Thanks for the report.
> 
> Unfortunately neither me nor other guys in the team could not reproduce
> mentioned issue on our hardware.
> Nevertheless I will try to allocate time and suggest possible fix if any
> comes to my mind. So, your testing and analysis is appreciated.

Sure. They way I trigger it is to unload the module whilst the DMA is
ongoing (I'm using a software model, so I can make the DMA nice and slow --
I guess you could try using some large buffers).

My script (I just run from busybox) is:

  #!/bin/sh

  mount proc -t proc /proc
  mount sysfs -t sysfs /sys
  mount debugfs -t debugfs /sys/kernel/debug

  echo 9 > /proc/sysrq-trigger

  modprobe dmatest
  echo 8 > /sys/kernel/debug/dmatest/iterations
  echo 8 > /sys/kernel/debug/dmatest/threads_per_chan
  echo 131072 > /sys/kernel/debug/dmatest/test_buf_size
  echo 1 > /sys/kernel/debug/dmatest/run
  modprobe -r dmatest

Will