Re: [BUG, bisected] atmel_spi: hard lockups in transfer

[Marc Pignat <marc.pignat@hevs.ch>]

On Monday 07 April 2008, Haavard Skinnemoen wrote:
> On Sun, 06 Apr 2008 09:28:38 +1000
> Ben Nizette <bn@niasdigital.com> wrote:
> 
> > Hi,
> > 
> > For some weeks now there have been reports flying across the main AVR32
> > support forum, avrfreaks [1], that atmel_spi now hard-hangs on most
> > transfers.  The behaviour is not apparent in .23 and is in .24+.  One
> > tester was good enough to bisect the problem and turned up
> 
> Thanks for reporting this.
> 
> > 154443c72f47169ebcb3a7befbff0e934c49bff3
> > atmel_spi: chain DMA transfers
> > 
> > as the culprit.  The tester also had to revert 
> > 
> > commit 8bacb219018a52e6f02a3cff6a7badf102ddfc44
> > atmel_spi: fix dmachain oops with DEBUG enabled
> > 
> > when they reverted 154443 as they were not confident enough to resolve
> > the conflicts that resulted from just reverting the first one.
> 
> That is indeed the correct way to resolve the conflict since the latter
> fixes a bug introduced by the former.
> 
> > The discussion thread involving this can be seen at [2].
> > 
> > As I say, this is present in .24+ so any fix will be a stable candidate
> > as well as .25 material (if we haven't missed the boat by then).
mainline .24+ is not affected (linux-2.6.24.3.atmel.3.tar.bz2 is).

> 
> I'm afraid I don't have much time to look at this right now...but I'll
> see what I can do. It's strange that I never noticed this problem
> myself -- I did see some overrun issues, but I blamed it on the
> throughput improvements and just reduced the speed slightly.
This patch seems to have problems at least on avr32, since this is only
a speed-up patch, we should revert them.

Regards

Marc