Add MT_MEMORY_SO, mark L3 and L4 to use it
authorNathan Monson <nmonson@gmail.com>
Thu, 16 Oct 2008 20:10:47 +0000 (13:10 -0700)
committerTony Lindgren <tony@atomide.com>
Fri, 17 Oct 2008 00:16:15 +0000 (17:16 -0700)
commitacb7f883f1ec61fd4dcb840a66ddca051ad8f2ef
tree465ac55e912abe950882cbb2f6304c9dbfb6e915
parentc61516af59f3fdfc63d6c0b9e25d9e942929db93
Add MT_MEMORY_SO, mark L3 and L4 to use it

On Wed, Oct 8, 2008 at 11:36 AM, Nathan Monson <nmonson@gmail.com> wrote:
> "Felipe Contreras" <felipe.contreras@gmail.com> writes:
>> irq -33, desc: c0335cf8, depth: 0, count: 0, unhandled: 0
>
> On the BeagleBoard list, Pratheesh Gangadhar said that mapping I/O
> regions as Strongly Ordered suppresses this problem:
> http://groups.google.com/group/beagleboard/browse_thread/thread/23e1c95b4bfb09b5/70d12dca569ca503?show_docid=70d12dca569ca503

Pratheesh helped me make a patch against the latest linux-omap git to
try this.

With this patch, my IRQ -33 problems with the DSP have disappeared.
Before, I would end up in IRQ -33 loop after 10 invocations of the DSP
Bridge 'ping.out' utility.  I just finished running it 50,000 times
without error.

As stated before, this patch is just a workaround for testing
purposes, not a fix.  Who knows what performance side effects it
has...

Signed-off-by: Nathan Monson <nmonson@gmail.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
arch/arm/include/asm/mach/map.h
arch/arm/mach-omap2/io.c
arch/arm/mm/mmu.c