From: Prarit Bhargava <prarit@redhat.com>
Date: Wed, 7 Apr 2010 17:41:19 -0400
Subject: [acpi] warn on hot-add of memory exceeding 4G boundary
Message-id: <20100407174119.28938.83703.sendpatchset@prarit.bos.redhat.com>
Patchwork-id: 24039
O-Subject: [RHEL5 PATCH]: Warn on hot-add of memory exceeding 4G boundary [v2]
Bugzilla: 571544
RH-Acked-by: Larry Woodman <lwoodman@redhat.com>
RH-Acked-by: Dean Nelson <dnelson@redhat.com>
RH-Acked-by: Stefan Assmann <sassmann@redhat.com>

If you hot-add memory to a system, and it has lt 4G of memory, 32-bit DMA
devices will have problems because swiotlb is not enabled.  Default enabling
of swiotlb is not an option because of the performance hit systems take when
doing so.

1.  Warn the user that they are adding memory which would exceed 4G of memory.
2.  Stop adding the memory.
3.  ia64 cleanup
	- fix the headers to properly include swiotlb.h
	- remove redefinition of dma_mapping_error in benet driver
	- remove stale swiotlb extern definition.  int swiotlb is not used
	in ia64.

Successfully brew built by me, and a similar patch was tested by the customer.

Resolves BZ 571544.

Signed-off-by: Jarod Wilson <jarod@redhat.com>

diff --git a/drivers/acpi/acpi_memhotplug.c b/drivers/acpi/acpi_memhotplug.c
index ce6329b..c440623 100644
--- a/drivers/acpi/acpi_memhotplug.c
+++ b/drivers/acpi/acpi_memhotplug.c
@@ -31,6 +31,7 @@
 #include <linux/types.h>
 #include <linux/memory_hotplug.h>
 #include <acpi/acpi_drivers.h>
+#include <linux/dma-mapping.h>
 
 #define ACPI_MEMORY_DEVICE_COMPONENT		0x08000000UL
 #define ACPI_MEMORY_DEVICE_CLASS		"memory"
@@ -240,6 +241,18 @@ static int acpi_memory_enable_device(struct acpi_memory_device *mem_device)
 			num_enabled++;
 			continue;
 		}
+#ifdef CONFIG_X86
+		/* Systems that boot with lt 4G of memory cannot add memory
+		 * to exceed 4G because bounce-buffering is disabled by
+		 * default. */
+		if (!swiotlb &&
+		    (info->start_addr + info->length >= 0x100000000)) {
+			printk(KERN_ERR "Adding memory would exceed the 4G "
+			       "boundary.  If you need to do this, please "
+			       "reboot with 'iommu=soft'.\n");
+			continue;
+		}
+#endif
 		if (node < 0)
 			node =  memory_add_physaddr_to_nid(info->start_addr);
 		result = add_memory(node, info->start_addr, info->length);
diff --git a/drivers/net/benet/be_compat.h b/drivers/net/benet/be_compat.h
index b668632..697c5e7 100644
--- a/drivers/net/benet/be_compat.h
+++ b/drivers/net/benet/be_compat.h
@@ -77,8 +77,6 @@ extern void be_netdev_ops_init(struct net_device *netdev,
 			struct net_device_ops *ops);
 extern int eth_validate_addr(struct net_device *);
 
-#define dma_mapping_error(dev, map) dma_mapping_error(map)
-
 /*
  * Back port of new NAPI: simulate polling on multiple napi instances
  * using tasklets
diff --git a/include/asm-ia64/dma-mapping.h b/include/asm-ia64/dma-mapping.h
index a85a10c..d7f98af 100644
--- a/include/asm-ia64/dma-mapping.h
+++ b/include/asm-ia64/dma-mapping.h
@@ -6,6 +6,7 @@
  *	David Mosberger-Tang <davidm@hpl.hp.com>
  */
 #include <asm/machvec.h>
+#include <asm/swiotlb.h>
 
 #define dma_alloc_coherent      platform_dma_alloc_coherent
 #define dma_free_coherent       platform_dma_free_coherent
diff --git a/include/asm-ia64/swiotlb.h b/include/asm-ia64/swiotlb.h
index cf4ba38..67e602f 100644
--- a/include/asm-ia64/swiotlb.h
+++ b/include/asm-ia64/swiotlb.h
@@ -35,10 +35,4 @@ extern void swiotlb_init(void);
 /* Width of DMA addresses in the IO TLB. 30 bits is a b44 limitation. */
 #define IO_TLB_DMA_BITS 30
 
-#ifdef CONFIG_SWIOTLB
-extern int swiotlb;
-#else
-#define swiotlb 0
-#endif
-
 #endif