-next fails to boot as of today on S3C6410

[Nicolas Pitre]

On Wed, 23 Nov 2011, Mark Brown wrote:

> On Wed, Nov 23, 2011 at 11:52:10AM -0600, Rob Herring wrote:
> > On 11/23/2011 08:55 AM, Mark Brown wrote:
> 
> > > My bisect turned up a second candidate - Nicholas' commit 549158d (ARM:
> > > move iotable mappings within the vmalloc region).  I reverted that and
> > > subsequent commits depending on it and my system now appears to boot
> > > fine with -next (with OF turned on to get the ops for the VIC).  This
> > > code is all a bit deep for me so I've no real idea what might be wrong.
> 
> > The first suspect should be overlapping static mappings in the iotable.
> > This was allowed before and now is not. Looking at some of the 6410
> > platforms, I don't see any though. But you didn't mention which platform
> > you are on. There is a 1KB mapping which is a bit unusual.
> 
> I'm on Cragganmore 6410 which just calls a bunch of arch-generic stuff
> (though now I look I do see a comment saying that we're relying on some
> setup from the bootloader which I get may have been trashed...).

I'm working on a way to get sensible debug messages out.

The problem is that, as soon as we execute this loop in 
devicemaps_init():

        for (addr = VMALLOC_START; addr; addr += PMD_SIZE)
                pmd_clear(pmd_off_k(addr));

we lose all means of communication with the outside world until 
mdesc->map_io() and the following TLB/cache flushes are done.  It is 
most likely that you are having a problem in the middle of that.


Nicolas