The way it is written, CONFIG_AGP is always y for x86_64 - not so nice 
when you want to use nvagp instead which requries agpgart to be 
unloadable.

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 6c70fed..1b2ce6c 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -429,12 +429,17 @@ config HPET_EMULATE_RTC
 	def_bool y
 	depends on HPET_TIMER && (RTC=y || RTC=m || RTC_DRV_CMOS=m || RTC_DRV_CMOS=y)
 
+config GART_IOMMU
+	bool
+
+# Hack to make AGP selectable as module
+config GART_IOMMU_SELECTOR
 # Mark as embedded because too many people got it wrong.
 # The code disables itself when not needed.
-config GART_IOMMU
-	bool "GART IOMMU support" if EMBEDDED
-	default y
+	tristate "GART IOMMU support" if EMBEDDED
+	default m
 	select SWIOTLB
+	select GART_IOMMU_BASE
 	select AGP
 	depends on X86_64 && PCI
 	help

--

[ message continues ]

Subject typo.

Please fix so people don't think this is a networking patch :-)
--

[ message continues ]

On Fri, 18 Apr 2008 23:21:44 +0200 (CEST)

where's the source code for nvagp? what does it do that normal agp does not do?
Also.. running without agp on x86-64 (heck on x86) tends to be rather unpleasant wrt resource allocations..

So I don't think this patch is a good idea at all..

-- 
If you want to reach me at my work email, use arjan@linux.intel.com
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org
--

[ message continues ]

The patch is wrong. GART_IOMMU cannot be modular. It has to be 
initialized early.

-Andi
--

[ message continues ]