[PATCH 2/2] ARM: use ARM_DMA_ZONE_SIZE to adjust the zone sizes

On Wed, May 11, 2011 at 06:21:38PM +0100, Catalin Marinas wrote:

On Wed, 2011-05-11 at 17:25 +0100, Russell King - ARM Linux wrote:

quoted

Rather than each platform providing its own function to adjust the
zone sizes, use the new ARM_DMA_ZONE_SIZE definition to perform this
adjustment.  This ensures that the actual DMA zone size and the
ISA_DMA_THRESHOLD/MAX_DMA_ADDRESS definitions are consistent with
each other, and moves this complexity out of the platform code.

...

quoted

diff --git a/arch/arm/mach-realview/include/mach/memory.h b/arch/arm/mach-realview/include/mach/memory.h
index 973428d..1759fa6 100644
--- a/arch/arm/mach-realview/include/mach/memory.h
+++ b/arch/arm/mach-realview/include/mach/memory.h

@@ -29,11 +29,7 @@
 #define PLAT_PHYS_OFFSET               UL(0x00000000)
 #endif

-#if !defined(__ASSEMBLY__) && defined(CONFIG_ZONE_DMA)
-extern void realview_adjust_zones(unsigned long *size, unsigned long *hole);
-#define arch_adjust_zones(size, hole) \
-       realview_adjust_zones(size, hole)
-
+#ifdef CONFIG_ZONE_DMA
 #define ARM_DMA_ZONE_SIZE      SZ_256M
 #endif

We only use ZONE_DMA for PBX and realview_adjust_zones() does this
check. We end up with limiting the DMA zone to 256MB on all RealView
platforms if we compile a single kernel image that includes PBX. Not a
big problem though:

You've already implicitly limited it by the ISA_DMA_THRESHOLD/
MAX_DMA_ADDRESS definition - what's left is an inconsistent relative
size of those definitions and the actual DMA zone.  That inconsistency
can cause the DMA allocator to fail unexpectedly as it assumes that
GFP_DMA will return a page which will satisfy ISA_DMA_THRESHOLD.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>

Thanks.