[PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[FUJITA Tomonori]

This is the second version of the patchset to convert SPARC to
asm-generic/dma-mapping-common.h and pci-dma-compat.h:

http://marc.info/?l=linux-sparc&m=124686094931138&w=2

The changes are:

- I added Arnd's patch (1/8) to remove flush_write_buffers() in
  dma-mapping-common.h. so no need to define useless
  flush_write_buffers() on SPARC.

- I added a patch to remove useless flush_write_buffers() on IA64.

- I found and fixed a bug wrt dma_{alloc|free}_coherent that affects
  pci devices on SPARC32; the bug doesn't use get_dma_ops in
  dma_{alloc|free}_coherent (4/8).

- Robert asked me about the dma-debug feature so I added it (8/8).

- Robert tested [3-8/8] patches. I added 'Tested-by' tags.


This patchset touches multiple architectures so it would be appropriate
to merge it via -mm. However, I think that the changes to IA64 and X86
are not large so it would be easier to merge this via SPARC tree.

=
 arch/ia64/include/asm/dma-mapping.h      |    1 -
 arch/sparc/Kconfig                       |    2 +
 arch/sparc/include/asm/dma-mapping.h     |  145 ++++-------------------
 arch/sparc/include/asm/pci.h             |    3 +
 arch/sparc/include/asm/pci_32.h          |  105 ----------------
 arch/sparc/include/asm/pci_64.h          |   88 --------------
 arch/sparc/kernel/Makefile               |    2 +-
 arch/sparc/kernel/dma.c                  |  175 +--------------------------
 arch/sparc/kernel/dma.h                  |   14 --
 arch/sparc/kernel/iommu.c                |   20 ++-
 arch/sparc/kernel/ioport.c               |  190 ++++++++++++++++-------------
 arch/sparc/kernel/pci.c                  |    2 +-
 arch/sparc/kernel/pci_sun4v.c            |   30 ++----
 arch/x86/kernel/pci-nommu.c              |   27 ++++-
 include/asm-generic/dma-mapping-common.h |    6 -
 15 files changed, 183 insertions(+), 627 deletions(-)

[PATCH 1/8] remove flush_write_buffers() in dma-mapping-common.h

[FUJITA Tomonori]

From: Arnd Bergmann <arnd@arndb.de>

This moves flush_write_buffers() in asm-generic/dma-mapping-common.h to
arch/x86/kernel/pci-nommu.c.

The purpose of this patch is that, we can avoid defining NULL
flush_write_buffers() on IA64 and SPARC.

dma-mapping-common.h is used by X86 and IA64 (and SPARC soon) but only
X86 with CONFIG_X86_OOSTORE or CONFIG_X86_PPRO_FENCE actually uses
flush_write_buffers(). CONFIG_X86_OOSTORE or CONFIG_X86_PPRO_FENCE is
usable with only kernel/pci-nommu.c (that is, not usable with other
X86 IOMMU implementations such as SWIOTLB, VT-d, etc) so we can safely
move flush_write_buffers() in asm-generic/dma-mapping-common.h to
arch/x86/kernel/pci-nommu.c.

The further discussion is:

http://lkml.org/lkml/2009/6/28/104

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
---
 arch/x86/kernel/pci-nommu.c              |   27 ++++++++++++++++++++++-----
 include/asm-generic/dma-mapping-common.h |    6 ------
 2 files changed, 22 insertions(+), 11 deletions(-)

diff --git a/arch/x86/kernel/pci-nommu.c b/arch/x86/kernel/pci-nommu.c
index 71d412a..f7eb967 100644
--- a/arch/x86/kernel/pci-nommu.c
+++ b/arch/x86/kernel/pci-nommu.c
@@ -79,12 +79,29 @@ static void nommu_free_coherent(struct device *dev, size_t size, void *vaddr,
 	free_pages((unsigned long)vaddr, get_order(size));
 }
 
+static void nommu_sync_single_for_device(struct device *dev,
+			dma_addr_t addr, size_t size,
+			enum dma_data_direction dir)
+{
+	flush_write_buffers();
+}
+
+
+static void nommu_sync_sg_for_device(struct device *dev,
+			struct scatterlist *sg, int nelems,
+			enum dma_data_direction dir)
+{
+	flush_write_buffers();
+}
+
 struct dma_map_ops nommu_dma_ops = {
-	.alloc_coherent	= dma_generic_alloc_coherent,
-	.free_coherent	= nommu_free_coherent,
-	.map_sg		= nommu_map_sg,
-	.map_page	= nommu_map_page,
-	.is_phys	= 1,
+	.alloc_coherent		= dma_generic_alloc_coherent,
+	.free_coherent		= nommu_free_coherent,
+	.map_sg			= nommu_map_sg,
+	.map_page		= nommu_map_page,
+	.sync_single_for_device = nommu_sync_single_for_device,
+	.sync_sg_for_device	= nommu_sync_sg_for_device,
+	.is_phys		= 1,
 };
 
 void __init no_iommu_init(void)
diff --git a/include/asm-generic/dma-mapping-common.h b/include/asm-generic/dma-mapping-common.h
index 5406a60..e694263 100644
--- a/include/asm-generic/dma-mapping-common.h
+++ b/include/asm-generic/dma-mapping-common.h
@@ -103,7 +103,6 @@ static inline void dma_sync_single_for_cpu(struct device *dev, dma_addr_t addr,
 	if (ops->sync_single_for_cpu)
 		ops->sync_single_for_cpu(dev, addr, size, dir);
 	debug_dma_sync_single_for_cpu(dev, addr, size, dir);
-	flush_write_buffers();
 }
 
 static inline void dma_sync_single_for_device(struct device *dev,
@@ -116,7 +115,6 @@ static inline void dma_sync_single_for_device(struct device *dev,
 	if (ops->sync_single_for_device)
 		ops->sync_single_for_device(dev, addr, size, dir);
 	debug_dma_sync_single_for_device(dev, addr, size, dir);
-	flush_write_buffers();
 }
 
 static inline void dma_sync_single_range_for_cpu(struct device *dev,
@@ -132,7 +130,6 @@ static inline void dma_sync_single_range_for_cpu(struct device *dev,
 		ops->sync_single_range_for_cpu(dev, addr, offset, size, dir);
 		debug_dma_sync_single_range_for_cpu(dev, addr, offset, size, dir);
 
-		flush_write_buffers();
 	} else
 		dma_sync_single_for_cpu(dev, addr, size, dir);
 }
@@ -150,7 +147,6 @@ static inline void dma_sync_single_range_for_device(struct device *dev,
 		ops->sync_single_range_for_device(dev, addr, offset, size, dir);
 		debug_dma_sync_single_range_for_device(dev, addr, offset, size, dir);
 
-		flush_write_buffers();
 	} else
 		dma_sync_single_for_device(dev, addr, size, dir);
 }
@@ -165,7 +161,6 @@ dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg,
 	if (ops->sync_sg_for_cpu)
 		ops->sync_sg_for_cpu(dev, sg, nelems, dir);
 	debug_dma_sync_sg_for_cpu(dev, sg, nelems, dir);
-	flush_write_buffers();
 }
 
 static inline void
@@ -179,7 +174,6 @@ dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
 		ops->sync_sg_for_device(dev, sg, nelems, dir);
 	debug_dma_sync_sg_for_device(dev, sg, nelems, dir);
 
-	flush_write_buffers();
 }
 
 #define dma_map_single(d, a, s, r) dma_map_single_attrs(d, a, s, r, NULL)
-- 
1.6.0.6

Re: [PATCH 1/8] remove flush_write_buffers() in dma-mapping-common.h

[FUJITA Tomonori]

On Mon, 13 Jul 2009 15:25:53 +0900
FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp> wrote:

> From: Arnd Bergmann <arnd@arndb.de>
> 
> This moves flush_write_buffers() in asm-generic/dma-mapping-common.h to
> arch/x86/kernel/pci-nommu.c.
> 
> The purpose of this patch is that, we can avoid defining NULL
> flush_write_buffers() on IA64 and SPARC.
> 
> dma-mapping-common.h is used by X86 and IA64 (and SPARC soon) but only
> X86 with CONFIG_X86_OOSTORE or CONFIG_X86_PPRO_FENCE actually uses
> flush_write_buffers(). CONFIG_X86_OOSTORE or CONFIG_X86_PPRO_FENCE is
> usable with only kernel/pci-nommu.c (that is, not usable with other
> X86 IOMMU implementations such as SWIOTLB, VT-d, etc) so we can safely
> move flush_write_buffers() in asm-generic/dma-mapping-common.h to
> arch/x86/kernel/pci-nommu.c.
> 
> The further discussion is:
> 
> http://lkml.org/lkml/2009/6/28/104
> 
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> Acked-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>

Ping?

> ---
>  arch/x86/kernel/pci-nommu.c              |   27 ++++++++++++++++++++++-----
>  include/asm-generic/dma-mapping-common.h |    6 ------
>  2 files changed, 22 insertions(+), 11 deletions(-)
> 
> diff --git a/arch/x86/kernel/pci-nommu.c b/arch/x86/kernel/pci-nommu.c
> index 71d412a..f7eb967 100644
> --- a/arch/x86/kernel/pci-nommu.c
> +++ b/arch/x86/kernel/pci-nommu.c
> @@ -79,12 +79,29 @@ static void nommu_free_coherent(struct device *dev, size_t size, void *vaddr,
>  	free_pages((unsigned long)vaddr, get_order(size));
>  }
>  
> +static void nommu_sync_single_for_device(struct device *dev,
> +			dma_addr_t addr, size_t size,
> +			enum dma_data_direction dir)
> +{
> +	flush_write_buffers();
> +}
> +
> +
> +static void nommu_sync_sg_for_device(struct device *dev,
> +			struct scatterlist *sg, int nelems,
> +			enum dma_data_direction dir)
> +{
> +	flush_write_buffers();
> +}
> +
>  struct dma_map_ops nommu_dma_ops = {
> -	.alloc_coherent	= dma_generic_alloc_coherent,
> -	.free_coherent	= nommu_free_coherent,
> -	.map_sg		= nommu_map_sg,
> -	.map_page	= nommu_map_page,
> -	.is_phys	= 1,
> +	.alloc_coherent		= dma_generic_alloc_coherent,
> +	.free_coherent		= nommu_free_coherent,
> +	.map_sg			= nommu_map_sg,
> +	.map_page		= nommu_map_page,
> +	.sync_single_for_device = nommu_sync_single_for_device,
> +	.sync_sg_for_device	= nommu_sync_sg_for_device,
> +	.is_phys		= 1,
>  };
>  
>  void __init no_iommu_init(void)
> diff --git a/include/asm-generic/dma-mapping-common.h b/include/asm-generic/dma-mapping-common.h
> index 5406a60..e694263 100644
> --- a/include/asm-generic/dma-mapping-common.h
> +++ b/include/asm-generic/dma-mapping-common.h
> @@ -103,7 +103,6 @@ static inline void dma_sync_single_for_cpu(struct device *dev, dma_addr_t addr,
>  	if (ops->sync_single_for_cpu)
>  		ops->sync_single_for_cpu(dev, addr, size, dir);
>  	debug_dma_sync_single_for_cpu(dev, addr, size, dir);
> -	flush_write_buffers();
>  }
>  
>  static inline void dma_sync_single_for_device(struct device *dev,
> @@ -116,7 +115,6 @@ static inline void dma_sync_single_for_device(struct device *dev,
>  	if (ops->sync_single_for_device)
>  		ops->sync_single_for_device(dev, addr, size, dir);
>  	debug_dma_sync_single_for_device(dev, addr, size, dir);
> -	flush_write_buffers();
>  }
>  
>  static inline void dma_sync_single_range_for_cpu(struct device *dev,
> @@ -132,7 +130,6 @@ static inline void dma_sync_single_range_for_cpu(struct device *dev,
>  		ops->sync_single_range_for_cpu(dev, addr, offset, size, dir);
>  		debug_dma_sync_single_range_for_cpu(dev, addr, offset, size, dir);
>  
> -		flush_write_buffers();
>  	} else
>  		dma_sync_single_for_cpu(dev, addr, size, dir);
>  }
> @@ -150,7 +147,6 @@ static inline void dma_sync_single_range_for_device(struct device *dev,
>  		ops->sync_single_range_for_device(dev, addr, offset, size, dir);
>  		debug_dma_sync_single_range_for_device(dev, addr, offset, size, dir);
>  
> -		flush_write_buffers();
>  	} else
>  		dma_sync_single_for_device(dev, addr, size, dir);
>  }
> @@ -165,7 +161,6 @@ dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg,
>  	if (ops->sync_sg_for_cpu)
>  		ops->sync_sg_for_cpu(dev, sg, nelems, dir);
>  	debug_dma_sync_sg_for_cpu(dev, sg, nelems, dir);
> -	flush_write_buffers();
>  }
>  
>  static inline void
> @@ -179,7 +174,6 @@ dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
>  		ops->sync_sg_for_device(dev, sg, nelems, dir);
>  	debug_dma_sync_sg_for_device(dev, sg, nelems, dir);
>  
> -	flush_write_buffers();
>  }
>  
>  #define dma_map_single(d, a, s, r) dma_map_single_attrs(d, a, s, r, NULL)
> -- 
> 1.6.0.6
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

[PATCH 2/8] IA64: remove NULL flush_write_buffers

[FUJITA Tomonori]

flush_write_buffers() in dma-mapping-common.h was removed so we can
remove NULL flush_write_buffers() in IA64.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
---
 arch/ia64/include/asm/dma-mapping.h |    1 -
 1 files changed, 0 insertions(+), 1 deletions(-)

diff --git a/arch/ia64/include/asm/dma-mapping.h b/arch/ia64/include/asm/dma-mapping.h
index 5a61b5c..fe9c822 100644
--- a/arch/ia64/include/asm/dma-mapping.h
+++ b/arch/ia64/include/asm/dma-mapping.h
@@ -44,7 +44,6 @@ static inline void dma_free_coherent(struct device *dev, size_t size,
 #define dma_free_noncoherent(d, s, v, h) dma_free_coherent(d, s, v, h)
 
 #define get_dma_ops(dev) platform_dma_get_ops(dev)
-#define flush_write_buffers()
 
 #include <asm-generic/dma-mapping-common.h>
 
-- 
1.6.0.6

[PATCH 3/8] sparc: use dma_map_ops struct

[FUJITA Tomonori]

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Tested-by: Robert Reif <reif@earthlink.net>
---
 arch/sparc/include/asm/dma-mapping.h |   43 +++++----------------------------
 arch/sparc/kernel/dma.c              |   16 ++++++++----
 arch/sparc/kernel/iommu.c            |   16 ++++++++----
 arch/sparc/kernel/pci_sun4v.c        |   14 +++++++----
 4 files changed, 36 insertions(+), 53 deletions(-)

diff --git a/arch/sparc/include/asm/dma-mapping.h b/arch/sparc/include/asm/dma-mapping.h
index 204e4bf..893f3ec 100644
--- a/arch/sparc/include/asm/dma-mapping.h
+++ b/arch/sparc/include/asm/dma-mapping.h
@@ -13,36 +13,7 @@ extern int dma_set_mask(struct device *dev, u64 dma_mask);
 #define dma_free_noncoherent(d, s, v, h) dma_free_coherent(d, s, v, h)
 #define dma_is_consistent(d, h)	(1)
 
-struct dma_ops {
-	void *(*alloc_coherent)(struct device *dev, size_t size,
-				dma_addr_t *dma_handle, gfp_t flag);
-	void (*free_coherent)(struct device *dev, size_t size,
-			      void *cpu_addr, dma_addr_t dma_handle);
-	dma_addr_t (*map_page)(struct device *dev, struct page *page,
-			       unsigned long offset, size_t size,
-			       enum dma_data_direction direction);
-	void (*unmap_page)(struct device *dev, dma_addr_t dma_addr,
-			   size_t size,
-			   enum dma_data_direction direction);
-	int (*map_sg)(struct device *dev, struct scatterlist *sg, int nents,
-		      enum dma_data_direction direction);
-	void (*unmap_sg)(struct device *dev, struct scatterlist *sg,
-			 int nhwentries,
-			 enum dma_data_direction direction);
-	void (*sync_single_for_cpu)(struct device *dev,
-				    dma_addr_t dma_handle, size_t size,
-				    enum dma_data_direction direction);
-	void (*sync_single_for_device)(struct device *dev,
-				       dma_addr_t dma_handle, size_t size,
-				       enum dma_data_direction direction);
-	void (*sync_sg_for_cpu)(struct device *dev, struct scatterlist *sg,
-				int nelems,
-				enum dma_data_direction direction);
-	void (*sync_sg_for_device)(struct device *dev,
-				   struct scatterlist *sg, int nents,
-				   enum dma_data_direction dir);
-};
-extern const struct dma_ops *dma_ops;
+extern const struct dma_map_ops *dma_ops;
 
 static inline void *dma_alloc_coherent(struct device *dev, size_t size,
 				       dma_addr_t *dma_handle, gfp_t flag)
@@ -62,40 +33,40 @@ static inline dma_addr_t dma_map_single(struct device *dev, void *cpu_addr,
 {
 	return dma_ops->map_page(dev, virt_to_page(cpu_addr),
 				 (unsigned long)cpu_addr & ~PAGE_MASK, size,
-				 direction);
+				 direction, NULL);
 }
 
 static inline void dma_unmap_single(struct device *dev, dma_addr_t dma_addr,
 				    size_t size,
 				    enum dma_data_direction direction)
 {
-	dma_ops->unmap_page(dev, dma_addr, size, direction);
+	dma_ops->unmap_page(dev, dma_addr, size, direction, NULL);
 }
 
 static inline dma_addr_t dma_map_page(struct device *dev, struct page *page,
 				      unsigned long offset, size_t size,
 				      enum dma_data_direction direction)
 {
-	return dma_ops->map_page(dev, page, offset, size, direction);
+	return dma_ops->map_page(dev, page, offset, size, direction, NULL);
 }
 
 static inline void dma_unmap_page(struct device *dev, dma_addr_t dma_address,
 				  size_t size,
 				  enum dma_data_direction direction)
 {
-	dma_ops->unmap_page(dev, dma_address, size, direction);
+	dma_ops->unmap_page(dev, dma_address, size, direction, NULL);
 }
 
 static inline int dma_map_sg(struct device *dev, struct scatterlist *sg,
 			     int nents, enum dma_data_direction direction)
 {
-	return dma_ops->map_sg(dev, sg, nents, direction);
+	return dma_ops->map_sg(dev, sg, nents, direction, NULL);
 }
 
 static inline void dma_unmap_sg(struct device *dev, struct scatterlist *sg,
 				int nents, enum dma_data_direction direction)
 {
-	dma_ops->unmap_sg(dev, sg, nents, direction);
+	dma_ops->unmap_sg(dev, sg, nents, direction, NULL);
 }
 
 static inline void dma_sync_single_for_cpu(struct device *dev,
diff --git a/arch/sparc/kernel/dma.c b/arch/sparc/kernel/dma.c
index 524c32f..473a3fc 100644
--- a/arch/sparc/kernel/dma.c
+++ b/arch/sparc/kernel/dma.c
@@ -60,7 +60,8 @@ static void dma32_free_coherent(struct device *dev, size_t size,
 
 static dma_addr_t dma32_map_page(struct device *dev, struct page *page,
 				 unsigned long offset, size_t size,
-				 enum dma_data_direction direction)
+				 enum dma_data_direction direction,
+				 struct dma_attrs *attrs)
 {
 #ifdef CONFIG_PCI
 	if (dev->bus == &pci_bus_type)
@@ -72,7 +73,8 @@ static dma_addr_t dma32_map_page(struct device *dev, struct page *page,
 }
 
 static void dma32_unmap_page(struct device *dev, dma_addr_t dma_address,
-			     size_t size, enum dma_data_direction direction)
+			     size_t size, enum dma_data_direction direction,
+			     struct dma_attrs *attrs)
 {
 #ifdef CONFIG_PCI
 	if (dev->bus == &pci_bus_type) {
@@ -85,7 +87,8 @@ static void dma32_unmap_page(struct device *dev, dma_addr_t dma_address,
 }
 
 static int dma32_map_sg(struct device *dev, struct scatterlist *sg,
-			int nents, enum dma_data_direction direction)
+			int nents, enum dma_data_direction direction,
+			struct dma_attrs *attrs)
 {
 #ifdef CONFIG_PCI
 	if (dev->bus == &pci_bus_type)
@@ -95,7 +98,8 @@ static int dma32_map_sg(struct device *dev, struct scatterlist *sg,
 }
 
 void dma32_unmap_sg(struct device *dev, struct scatterlist *sg,
-		    int nents, enum dma_data_direction direction)
+		    int nents, enum dma_data_direction direction,
+		    struct dma_attrs *attrs)
 {
 #ifdef CONFIG_PCI
 	if (dev->bus == &pci_bus_type) {
@@ -161,7 +165,7 @@ static void dma32_sync_sg_for_device(struct device *dev,
 	BUG();
 }
 
-static const struct dma_ops dma32_dma_ops = {
+static const struct dma_map_ops dma32_dma_ops = {
 	.alloc_coherent		= dma32_alloc_coherent,
 	.free_coherent		= dma32_free_coherent,
 	.map_page		= dma32_map_page,
@@ -174,5 +178,5 @@ static const struct dma_ops dma32_dma_ops = {
 	.sync_sg_for_device	= dma32_sync_sg_for_device,
 };
 
-const struct dma_ops *dma_ops = &dma32_dma_ops;
+const struct dma_map_ops *dma_ops = &dma32_dma_ops;
 EXPORT_SYMBOL(dma_ops);
diff --git a/arch/sparc/kernel/iommu.c b/arch/sparc/kernel/iommu.c
index 0aeaefe..a9f0ad9 100644
--- a/arch/sparc/kernel/iommu.c
+++ b/arch/sparc/kernel/iommu.c
@@ -353,7 +353,8 @@ static void dma_4u_free_coherent(struct device *dev, size_t size,
 
 static dma_addr_t dma_4u_map_page(struct device *dev, struct page *page,
 				  unsigned long offset, size_t sz,
-				  enum dma_data_direction direction)
+				  enum dma_data_direction direction,
+				  struct dma_attrs *attrs)
 {
 	struct iommu *iommu;
 	struct strbuf *strbuf;
@@ -474,7 +475,8 @@ do_flush_sync:
 }
 
 static void dma_4u_unmap_page(struct device *dev, dma_addr_t bus_addr,
-			      size_t sz, enum dma_data_direction direction)
+			      size_t sz, enum dma_data_direction direction,
+			      struct dma_attrs *attrs)
 {
 	struct iommu *iommu;
 	struct strbuf *strbuf;
@@ -520,7 +522,8 @@ static void dma_4u_unmap_page(struct device *dev, dma_addr_t bus_addr,
 }
 
 static int dma_4u_map_sg(struct device *dev, struct scatterlist *sglist,
-			 int nelems, enum dma_data_direction direction)
+			 int nelems, enum dma_data_direction direction,
+			 struct dma_attrs *attrs)
 {
 	struct scatterlist *s, *outs, *segstart;
 	unsigned long flags, handle, prot, ctx;
@@ -691,7 +694,8 @@ static unsigned long fetch_sg_ctx(struct iommu *iommu, struct scatterlist *sg)
 }
 
 static void dma_4u_unmap_sg(struct device *dev, struct scatterlist *sglist,
-			    int nelems, enum dma_data_direction direction)
+			    int nelems, enum dma_data_direction direction,
+			    struct dma_attrs *attrs)
 {
 	unsigned long flags, ctx;
 	struct scatterlist *sg;
@@ -822,7 +826,7 @@ static void dma_4u_sync_sg_for_cpu(struct device *dev,
 	spin_unlock_irqrestore(&iommu->lock, flags);
 }
 
-static const struct dma_ops sun4u_dma_ops = {
+static const struct dma_map_ops sun4u_dma_ops = {
 	.alloc_coherent		= dma_4u_alloc_coherent,
 	.free_coherent		= dma_4u_free_coherent,
 	.map_page		= dma_4u_map_page,
@@ -833,7 +837,7 @@ static const struct dma_ops sun4u_dma_ops = {
 	.sync_sg_for_cpu	= dma_4u_sync_sg_for_cpu,
 };
 
-const struct dma_ops *dma_ops = &sun4u_dma_ops;
+const struct dma_map_ops *dma_ops = &sun4u_dma_ops;
 EXPORT_SYMBOL(dma_ops);
 
 int dma_supported(struct device *dev, u64 device_mask)
diff --git a/arch/sparc/kernel/pci_sun4v.c b/arch/sparc/kernel/pci_sun4v.c
index 2485eaa..c4f7dce 100644
--- a/arch/sparc/kernel/pci_sun4v.c
+++ b/arch/sparc/kernel/pci_sun4v.c
@@ -232,7 +232,8 @@ static void dma_4v_free_coherent(struct device *dev, size_t size, void *cpu,
 
 static dma_addr_t dma_4v_map_page(struct device *dev, struct page *page,
 				  unsigned long offset, size_t sz,
-				  enum dma_data_direction direction)
+				  enum dma_data_direction direction,
+				  struct dma_attrs *attrs)
 {
 	struct iommu *iommu;
 	unsigned long flags, npages, oaddr;
@@ -296,7 +297,8 @@ iommu_map_fail:
 }
 
 static void dma_4v_unmap_page(struct device *dev, dma_addr_t bus_addr,
-			      size_t sz, enum dma_data_direction direction)
+			      size_t sz, enum dma_data_direction direction,
+			      struct dma_attrs *attrs)
 {
 	struct pci_pbm_info *pbm;
 	struct iommu *iommu;
@@ -336,7 +338,8 @@ static void dma_4v_unmap_page(struct device *dev, dma_addr_t bus_addr,
 }
 
 static int dma_4v_map_sg(struct device *dev, struct scatterlist *sglist,
-			 int nelems, enum dma_data_direction direction)
+			 int nelems, enum dma_data_direction direction,
+			 struct dma_attrs *attrs)
 {
 	struct scatterlist *s, *outs, *segstart;
 	unsigned long flags, handle, prot;
@@ -478,7 +481,8 @@ iommu_map_failed:
 }
 
 static void dma_4v_unmap_sg(struct device *dev, struct scatterlist *sglist,
-			    int nelems, enum dma_data_direction direction)
+			    int nelems, enum dma_data_direction direction,
+			    struct dma_attrs *attrs)
 {
 	struct pci_pbm_info *pbm;
 	struct scatterlist *sg;
@@ -535,7 +539,7 @@ static void dma_4v_sync_sg_for_cpu(struct device *dev,
 	/* Nothing to do... */
 }
 
-static const struct dma_ops sun4v_dma_ops = {
+static const struct dma_map_ops sun4v_dma_ops = {
 	.alloc_coherent			= dma_4v_alloc_coherent,
 	.free_coherent			= dma_4v_free_coherent,
 	.map_page			= dma_4v_map_page,
-- 
1.6.0.6

Re: [PATCH 3/8] sparc: use dma_map_ops struct

[David Miller]

From: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Date: Mon, 13 Jul 2009 15:25:55 +0900

> Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
> Tested-by: Robert Reif <reif@earthlink.net>

Acked-by: David S. Miller <davem@davemloft.net>

[PATCH 4/8] sparc: use asm-generic/dma-mapping-common.h

[FUJITA Tomonori]

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Tested-by: Robert Reif <reif@earthlink.net>
---
 arch/sparc/Kconfig                   |    1 +
 arch/sparc/include/asm/dma-mapping.h |  107 ++++-----------------------------
 arch/sparc/kernel/dma.c              |    4 +-
 arch/sparc/kernel/iommu.c            |    4 +-
 arch/sparc/kernel/pci_sun4v.c        |    2 +-
 5 files changed, 19 insertions(+), 99 deletions(-)

diff --git a/arch/sparc/Kconfig b/arch/sparc/Kconfig
index 3f8b6a9..5f2df99 100644
--- a/arch/sparc/Kconfig
+++ b/arch/sparc/Kconfig
@@ -25,6 +25,7 @@ config SPARC
 	select ARCH_WANT_OPTIONAL_GPIOLIB
 	select RTC_CLASS
 	select RTC_DRV_M48T59
+	select HAVE_DMA_ATTRS
 
 config SPARC32
 	def_bool !64BIT
diff --git a/arch/sparc/include/asm/dma-mapping.h b/arch/sparc/include/asm/dma-mapping.h
index 893f3ec..34c9226 100644
--- a/arch/sparc/include/asm/dma-mapping.h
+++ b/arch/sparc/include/asm/dma-mapping.h
@@ -3,6 +3,7 @@
 
 #include <linux/scatterlist.h>
 #include <linux/mm.h>
+#include <linux/dma-debug.h>
 
 #define DMA_ERROR_CODE	(~(dma_addr_t)0x0)
 
@@ -13,113 +14,31 @@ extern int dma_set_mask(struct device *dev, u64 dma_mask);
 #define dma_free_noncoherent(d, s, v, h) dma_free_coherent(d, s, v, h)
 #define dma_is_consistent(d, h)	(1)
 
-extern const struct dma_map_ops *dma_ops;
+extern struct dma_map_ops *dma_ops;
 
-static inline void *dma_alloc_coherent(struct device *dev, size_t size,
-				       dma_addr_t *dma_handle, gfp_t flag)
-{
-	return dma_ops->alloc_coherent(dev, size, dma_handle, flag);
-}
-
-static inline void dma_free_coherent(struct device *dev, size_t size,
-				     void *cpu_addr, dma_addr_t dma_handle)
-{
-	dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
-}
-
-static inline dma_addr_t dma_map_single(struct device *dev, void *cpu_addr,
-					size_t size,
-					enum dma_data_direction direction)
+static inline struct dma_map_ops *get_dma_ops(struct device *dev)
 {
-	return dma_ops->map_page(dev, virt_to_page(cpu_addr),
-				 (unsigned long)cpu_addr & ~PAGE_MASK, size,
-				 direction, NULL);
+	return dma_ops;
 }
 
-static inline void dma_unmap_single(struct device *dev, dma_addr_t dma_addr,
-				    size_t size,
-				    enum dma_data_direction direction)
-{
-	dma_ops->unmap_page(dev, dma_addr, size, direction, NULL);
-}
-
-static inline dma_addr_t dma_map_page(struct device *dev, struct page *page,
-				      unsigned long offset, size_t size,
-				      enum dma_data_direction direction)
-{
-	return dma_ops->map_page(dev, page, offset, size, direction, NULL);
-}
+#include <asm-generic/dma-mapping-common.h>
 
-static inline void dma_unmap_page(struct device *dev, dma_addr_t dma_address,
-				  size_t size,
-				  enum dma_data_direction direction)
-{
-	dma_ops->unmap_page(dev, dma_address, size, direction, NULL);
-}
-
-static inline int dma_map_sg(struct device *dev, struct scatterlist *sg,
-			     int nents, enum dma_data_direction direction)
-{
-	return dma_ops->map_sg(dev, sg, nents, direction, NULL);
-}
-
-static inline void dma_unmap_sg(struct device *dev, struct scatterlist *sg,
-				int nents, enum dma_data_direction direction)
-{
-	dma_ops->unmap_sg(dev, sg, nents, direction, NULL);
-}
-
-static inline void dma_sync_single_for_cpu(struct device *dev,
-					   dma_addr_t dma_handle, size_t size,
-					   enum dma_data_direction direction)
-{
-	dma_ops->sync_single_for_cpu(dev, dma_handle, size, direction);
-}
-
-static inline void dma_sync_single_for_device(struct device *dev,
-					      dma_addr_t dma_handle,
-					      size_t size,
-					      enum dma_data_direction direction)
-{
-	if (dma_ops->sync_single_for_device)
-		dma_ops->sync_single_for_device(dev, dma_handle, size,
-						direction);
-}
-
-static inline void dma_sync_sg_for_cpu(struct device *dev,
-				       struct scatterlist *sg, int nelems,
-				       enum dma_data_direction direction)
+static inline void *dma_alloc_coherent(struct device *dev, size_t size,
+				       dma_addr_t *dma_handle, gfp_t flag)
 {
-	dma_ops->sync_sg_for_cpu(dev, sg, nelems, direction);
-}
+	struct dma_map_ops *ops = get_dma_ops(dev);
 
-static inline void dma_sync_sg_for_device(struct device *dev,
-					  struct scatterlist *sg, int nelems,
-					  enum dma_data_direction direction)
-{
-	if (dma_ops->sync_sg_for_device)
-		dma_ops->sync_sg_for_device(dev, sg, nelems, direction);
+	return ops->alloc_coherent(dev, size, dma_handle, flag);
 }
 
-static inline void dma_sync_single_range_for_cpu(struct device *dev,
-						 dma_addr_t dma_handle,
-						 unsigned long offset,
-						 size_t size,
-						 enum dma_data_direction dir)
+static inline void dma_free_coherent(struct device *dev, size_t size,
+				     void *cpu_addr, dma_addr_t dma_handle)
 {
-	dma_sync_single_for_cpu(dev, dma_handle+offset, size, dir);
-}
+	struct dma_map_ops *ops = get_dma_ops(dev);
 
-static inline void dma_sync_single_range_for_device(struct device *dev,
-						    dma_addr_t dma_handle,
-						    unsigned long offset,
-						    size_t size,
-						    enum dma_data_direction dir)
-{
-	dma_sync_single_for_device(dev, dma_handle+offset, size, dir);
+	ops->free_coherent(dev, size, cpu_addr, dma_handle);
 }
 
-
 static inline int dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
 	return (dma_addr == DMA_ERROR_CODE);
diff --git a/arch/sparc/kernel/dma.c b/arch/sparc/kernel/dma.c
index 473a3fc..15820a9 100644
--- a/arch/sparc/kernel/dma.c
+++ b/arch/sparc/kernel/dma.c
@@ -165,7 +165,7 @@ static void dma32_sync_sg_for_device(struct device *dev,
 	BUG();
 }
 
-static const struct dma_map_ops dma32_dma_ops = {
+static struct dma_map_ops dma32_dma_ops = {
 	.alloc_coherent		= dma32_alloc_coherent,
 	.free_coherent		= dma32_free_coherent,
 	.map_page		= dma32_map_page,
@@ -178,5 +178,5 @@ static const struct dma_map_ops dma32_dma_ops = {
 	.sync_sg_for_device	= dma32_sync_sg_for_device,
 };
 
-const struct dma_map_ops *dma_ops = &dma32_dma_ops;
+struct dma_map_ops *dma_ops = &dma32_dma_ops;
 EXPORT_SYMBOL(dma_ops);
diff --git a/arch/sparc/kernel/iommu.c b/arch/sparc/kernel/iommu.c
index a9f0ad9..74b289c 100644
--- a/arch/sparc/kernel/iommu.c
+++ b/arch/sparc/kernel/iommu.c
@@ -826,7 +826,7 @@ static void dma_4u_sync_sg_for_cpu(struct device *dev,
 	spin_unlock_irqrestore(&iommu->lock, flags);
 }
 
-static const struct dma_map_ops sun4u_dma_ops = {
+static struct dma_map_ops sun4u_dma_ops = {
 	.alloc_coherent		= dma_4u_alloc_coherent,
 	.free_coherent		= dma_4u_free_coherent,
 	.map_page		= dma_4u_map_page,
@@ -837,7 +837,7 @@ static const struct dma_map_ops sun4u_dma_ops = {
 	.sync_sg_for_cpu	= dma_4u_sync_sg_for_cpu,
 };
 
-const struct dma_map_ops *dma_ops = &sun4u_dma_ops;
+struct dma_map_ops *dma_ops = &sun4u_dma_ops;
 EXPORT_SYMBOL(dma_ops);
 
 int dma_supported(struct device *dev, u64 device_mask)
diff --git a/arch/sparc/kernel/pci_sun4v.c b/arch/sparc/kernel/pci_sun4v.c
index c4f7dce..ee800f9 100644
--- a/arch/sparc/kernel/pci_sun4v.c
+++ b/arch/sparc/kernel/pci_sun4v.c
@@ -539,7 +539,7 @@ static void dma_4v_sync_sg_for_cpu(struct device *dev,
 	/* Nothing to do... */
 }
 
-static const struct dma_map_ops sun4v_dma_ops = {
+static struct dma_map_ops sun4v_dma_ops = {
 	.alloc_coherent			= dma_4v_alloc_coherent,
 	.free_coherent			= dma_4v_free_coherent,
 	.map_page			= dma_4v_map_page,
-- 
1.6.0.6

Re: [PATCH 4/8] sparc: use asm-generic/dma-mapping-common.h

[David Miller]

From: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Date: Mon, 13 Jul 2009 15:25:56 +0900

> Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
> Tested-by: Robert Reif <reif@earthlink.net>

ACK

[PATCH 5/8] sparc: remove no-op dma_4v_sync_single_for_cpu and dma_4v_sync_sg_for_cpu

[FUJITA Tomonori]

Now sparc uses include/asm-generic/dma-mapping-common.h. pci_sun4v.c
doesn't need to have no-op dma_4v_sync_single_for_cpu and
dma_4v_sync_sg_for_cpu (dma-mapping-common.h does nothing if
sync_{single|sg}_for_cpu hook is not defined). So we can remove them
safely.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Tested-by: Robert Reif <reif@earthlink.net>
---
 arch/sparc/kernel/pci_sun4v.c |   16 ----------------
 1 files changed, 0 insertions(+), 16 deletions(-)

diff --git a/arch/sparc/kernel/pci_sun4v.c b/arch/sparc/kernel/pci_sun4v.c
index ee800f9..23c33ff 100644
--- a/arch/sparc/kernel/pci_sun4v.c
+++ b/arch/sparc/kernel/pci_sun4v.c
@@ -525,20 +525,6 @@ static void dma_4v_unmap_sg(struct device *dev, struct scatterlist *sglist,
 	spin_unlock_irqrestore(&iommu->lock, flags);
 }
 
-static void dma_4v_sync_single_for_cpu(struct device *dev,
-				       dma_addr_t bus_addr, size_t sz,
-				       enum dma_data_direction direction)
-{
-	/* Nothing to do... */
-}
-
-static void dma_4v_sync_sg_for_cpu(struct device *dev,
-				   struct scatterlist *sglist, int nelems,
-				   enum dma_data_direction direction)
-{
-	/* Nothing to do... */
-}
-
 static struct dma_map_ops sun4v_dma_ops = {
 	.alloc_coherent			= dma_4v_alloc_coherent,
 	.free_coherent			= dma_4v_free_coherent,
@@ -546,8 +532,6 @@ static struct dma_map_ops sun4v_dma_ops = {
 	.unmap_page			= dma_4v_unmap_page,
 	.map_sg				= dma_4v_map_sg,
 	.unmap_sg			= dma_4v_unmap_sg,
-	.sync_single_for_cpu		= dma_4v_sync_single_for_cpu,
-	.sync_sg_for_cpu		= dma_4v_sync_sg_for_cpu,
 };
 
 static void __devinit pci_sun4v_scan_bus(struct pci_pbm_info *pbm,
-- 
1.6.0.6

Re: [PATCH 5/8] sparc: remove no-op dma_4v_sync_single_for_cpu and dma_4v_sync_sg_for_cpu

[David Miller]

From: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Date: Mon, 13 Jul 2009 15:25:57 +0900

> Now sparc uses include/asm-generic/dma-mapping-common.h. pci_sun4v.c
> doesn't need to have no-op dma_4v_sync_single_for_cpu and
> dma_4v_sync_sg_for_cpu (dma-mapping-common.h does nothing if
> sync_{single|sg}_for_cpu hook is not defined). So we can remove them
> safely.
> 
> Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
> Tested-by: Robert Reif <reif@earthlink.net>

ACK

[PATCH 6/8] sparc: replace sbus_map_single and sbus_unmap_single with sbus_map_page and sbus_unmap_page

[FUJITA Tomonori]

This is a preparation for using asm-generic/pci-dma-compat.h; SPARC32
has two dma_map_ops structures for pci and sbus (removing
arch/sparc/kernel/dma.c, PCI and SBUS DMA accessor).

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Tested-by: Robert Reif <reif@earthlink.net>
---
 arch/sparc/kernel/dma.c    |    5 ++---
 arch/sparc/kernel/dma.h    |    6 +++---
 arch/sparc/kernel/ioport.c |    7 +++++--
 3 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/arch/sparc/kernel/dma.c b/arch/sparc/kernel/dma.c
index 15820a9..a5d50da 100644
--- a/arch/sparc/kernel/dma.c
+++ b/arch/sparc/kernel/dma.c
@@ -68,8 +68,7 @@ static dma_addr_t dma32_map_page(struct device *dev, struct page *page,
 		return pci_map_page(to_pci_dev(dev), page, offset,
 				    size, (int)direction);
 #endif
-	return sbus_map_single(dev, page_address(page) + offset,
-			       size, (int)direction);
+	return sbus_map_page(dev, page, offset, size, (int)direction);
 }
 
 static void dma32_unmap_page(struct device *dev, dma_addr_t dma_address,
@@ -83,7 +82,7 @@ static void dma32_unmap_page(struct device *dev, dma_addr_t dma_address,
 		return;
 	}
 #endif
-	sbus_unmap_single(dev, dma_address, size, (int)direction);
+	sbus_unmap_page(dev, dma_address, size, (int)direction);
 }
 
 static int dma32_map_sg(struct device *dev, struct scatterlist *sg,
diff --git a/arch/sparc/kernel/dma.h b/arch/sparc/kernel/dma.h
index f8d8951..680351e 100644
--- a/arch/sparc/kernel/dma.h
+++ b/arch/sparc/kernel/dma.h
@@ -1,8 +1,8 @@
 void *sbus_alloc_consistent(struct device *dev, long len, u32 *dma_addrp);
 void sbus_free_consistent(struct device *dev, long n, void *p, u32 ba);
-dma_addr_t sbus_map_single(struct device *dev, void *va,
-			   size_t len, int direction);
-void sbus_unmap_single(struct device *dev, dma_addr_t ba,
+dma_addr_t sbus_map_page(struct device *dev, struct page *page,
+			 unsigned long offset, size_t len, int direction);
+void sbus_unmap_page(struct device *dev, dma_addr_t ba,
 		       size_t n, int direction);
 int sbus_map_sg(struct device *dev, struct scatterlist *sg,
 		int n, int direction);
diff --git a/arch/sparc/kernel/ioport.c b/arch/sparc/kernel/ioport.c
index 87ea0d0..39ff1e0 100644
--- a/arch/sparc/kernel/ioport.c
+++ b/arch/sparc/kernel/ioport.c
@@ -337,8 +337,11 @@ void sbus_free_consistent(struct device *dev, long n, void *p, u32 ba)
  * CPU view of this memory may be inconsistent with
  * a device view and explicit flushing is necessary.
  */
-dma_addr_t sbus_map_single(struct device *dev, void *va, size_t len, int direction)
+dma_addr_t sbus_map_page(struct device *dev, struct page *page,
+			 unsigned long offset, size_t len, int direction)
 {
+	void *va = page_address(page) + offset;
+
 	/* XXX why are some lengths signed, others unsigned? */
 	if (len <= 0) {
 		return 0;
@@ -350,7 +353,7 @@ dma_addr_t sbus_map_single(struct device *dev, void *va, size_t len, int directi
 	return mmu_get_scsi_one(dev, va, len);
 }
 
-void sbus_unmap_single(struct device *dev, dma_addr_t ba, size_t n, int direction)
+void sbus_unmap_page(struct device *dev, dma_addr_t ba, size_t n, int direction)
 {
 	mmu_release_scsi_one(dev, ba, n);
 }
-- 
1.6.0.6

Re: [PATCH 6/8] sparc: replace sbus_map_single and sbus_unmap_single with sbus_map_page and sbus_unmap_page

[David Miller]

From: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Date: Mon, 13 Jul 2009 15:25:58 +0900

> This is a preparation for using asm-generic/pci-dma-compat.h; SPARC32
> has two dma_map_ops structures for pci and sbus (removing
> arch/sparc/kernel/dma.c, PCI and SBUS DMA accessor).
> 
> Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
> Tested-by: Robert Reif <reif@earthlink.net>

ACK

[PATCH 7/8] sparc: use asm-generic/pci-dma-compat

[FUJITA Tomonori]

This converts SPARC to use asm-generic/pci-dma-compat instead of the
homegrown mechnism.

SPARC32 has two dma_map_ops structures for pci and sbus (removing
arch/sparc/kernel/dma.c, PCI and SBUS DMA accessor). The global
'dma_ops' is set to sbus_dma_ops and get_dma_ops() returns
pci32_dma_ops for pci devices so we can use the appropriate dma
mapping operations.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Tested-by: Robert Reif <reif@earthlink.net>
---
 arch/sparc/include/asm/dma-mapping.h |    7 ++-
 arch/sparc/include/asm/pci.h         |    3 +
 arch/sparc/include/asm/pci_32.h      |  105 ----------------------
 arch/sparc/include/asm/pci_64.h      |   88 ------------------
 arch/sparc/kernel/dma.c              |  155 ++-------------------------------
 arch/sparc/kernel/dma.h              |   14 ---
 arch/sparc/kernel/iommu.c            |    4 +-
 arch/sparc/kernel/ioport.c           |  162 ++++++++++++++++------------------
 arch/sparc/kernel/pci.c              |    2 +-
 9 files changed, 96 insertions(+), 444 deletions(-)
 delete mode 100644 arch/sparc/kernel/dma.h

diff --git a/arch/sparc/include/asm/dma-mapping.h b/arch/sparc/include/asm/dma-mapping.h
index 34c9226..2677818 100644
--- a/arch/sparc/include/asm/dma-mapping.h
+++ b/arch/sparc/include/asm/dma-mapping.h
@@ -14,10 +14,15 @@ extern int dma_set_mask(struct device *dev, u64 dma_mask);
 #define dma_free_noncoherent(d, s, v, h) dma_free_coherent(d, s, v, h)
 #define dma_is_consistent(d, h)	(1)
 
-extern struct dma_map_ops *dma_ops;
+extern struct dma_map_ops *dma_ops, pci32_dma_ops;
+extern struct bus_type pci_bus_type;
 
 static inline struct dma_map_ops *get_dma_ops(struct device *dev)
 {
+#if defined(CONFIG_SPARC32) && defined(CONFIG_PCI)
+	if (dev->bus == &pci_bus_type)
+		return &pci32_dma_ops;
+#endif
 	return dma_ops;
 }
 
diff --git a/arch/sparc/include/asm/pci.h b/arch/sparc/include/asm/pci.h
index 6e14fd1..d9c031f 100644
--- a/arch/sparc/include/asm/pci.h
+++ b/arch/sparc/include/asm/pci.h
@@ -5,4 +5,7 @@
 #else
 #include <asm/pci_32.h>
 #endif
+
+#include <asm-generic/pci-dma-compat.h>
+
 #endif
diff --git a/arch/sparc/include/asm/pci_32.h b/arch/sparc/include/asm/pci_32.h
index b41c4c1..ac0e836 100644
--- a/arch/sparc/include/asm/pci_32.h
+++ b/arch/sparc/include/asm/pci_32.h
@@ -31,42 +31,8 @@ static inline void pcibios_penalize_isa_irq(int irq, int active)
  */
 #define PCI_DMA_BUS_IS_PHYS	(0)
 
-#include <asm/scatterlist.h>
-
 struct pci_dev;
 
-/* Allocate and map kernel buffer using consistent mode DMA for a device.
- * hwdev should be valid struct pci_dev pointer for PCI devices.
- */
-extern void *pci_alloc_consistent(struct pci_dev *hwdev, size_t size, dma_addr_t *dma_handle);
-
-/* Free and unmap a consistent DMA buffer.
- * cpu_addr is what was returned from pci_alloc_consistent,
- * size must be the same as what as passed into pci_alloc_consistent,
- * and likewise dma_addr must be the same as what *dma_addrp was set to.
- *
- * References to the memory and mappings assosciated with cpu_addr/dma_addr
- * past this call are illegal.
- */
-extern void pci_free_consistent(struct pci_dev *hwdev, size_t size, void *vaddr, dma_addr_t dma_handle);
-
-/* Map a single buffer of the indicated size for DMA in streaming mode.
- * The 32-bit bus address to use is returned.
- *
- * Once the device is given the dma address, the device owns this memory
- * until either pci_unmap_single or pci_dma_sync_single_for_cpu is performed.
- */
-extern dma_addr_t pci_map_single(struct pci_dev *hwdev, void *ptr, size_t size, int direction);
-
-/* Unmap a single streaming mode DMA translation.  The dma_addr and size
- * must match what was provided for in a previous pci_map_single call.  All
- * other usages are undefined.
- *
- * After this call, reads by the cpu to the buffer are guaranteed to see
- * whatever the device wrote there.
- */
-extern void pci_unmap_single(struct pci_dev *hwdev, dma_addr_t dma_addr, size_t size, int direction);
-
 /* pci_unmap_{single,page} is not a nop, thus... */
 #define DECLARE_PCI_UNMAP_ADDR(ADDR_NAME)	\
 	dma_addr_t ADDR_NAME;
@@ -81,69 +47,6 @@ extern void pci_unmap_single(struct pci_dev *hwdev, dma_addr_t dma_addr, size_t
 #define pci_unmap_len_set(PTR, LEN_NAME, VAL)		\
 	(((PTR)->LEN_NAME) = (VAL))
 
-/*
- * Same as above, only with pages instead of mapped addresses.
- */
-extern dma_addr_t pci_map_page(struct pci_dev *hwdev, struct page *page,
-			unsigned long offset, size_t size, int direction);
-extern void pci_unmap_page(struct pci_dev *hwdev,
-			dma_addr_t dma_address, size_t size, int direction);
-
-/* Map a set of buffers described by scatterlist in streaming
- * mode for DMA.  This is the scather-gather version of the
- * above pci_map_single interface.  Here the scatter gather list
- * elements are each tagged with the appropriate dma address
- * and length.  They are obtained via sg_dma_{address,length}(SG).
- *
- * NOTE: An implementation may be able to use a smaller number of
- *       DMA address/length pairs than there are SG table elements.
- *       (for example via virtual mapping capabilities)
- *       The routine returns the number of addr/length pairs actually
- *       used, at most nents.
- *
- * Device ownership issues as mentioned above for pci_map_single are
- * the same here.
- */
-extern int pci_map_sg(struct pci_dev *hwdev, struct scatterlist *sg, int nents, int direction);
-
-/* Unmap a set of streaming mode DMA translations.
- * Again, cpu read rules concerning calls here are the same as for
- * pci_unmap_single() above.
- */
-extern void pci_unmap_sg(struct pci_dev *hwdev, struct scatterlist *sg, int nhwents, int direction);
-
-/* Make physical memory consistent for a single
- * streaming mode DMA translation after a transfer.
- *
- * If you perform a pci_map_single() but wish to interrogate the
- * buffer using the cpu, yet do not wish to teardown the PCI dma
- * mapping, you must call this function before doing so.  At the
- * next point you give the PCI dma address back to the card, you
- * must first perform a pci_dma_sync_for_device, and then the device
- * again owns the buffer.
- */
-extern void pci_dma_sync_single_for_cpu(struct pci_dev *hwdev, dma_addr_t dma_handle, size_t size, int direction);
-extern void pci_dma_sync_single_for_device(struct pci_dev *hwdev, dma_addr_t dma_handle, size_t size, int direction);
-
-/* Make physical memory consistent for a set of streaming
- * mode DMA translations after a transfer.
- *
- * The same as pci_dma_sync_single_* but for a scatter-gather list,
- * same rules and usage.
- */
-extern void pci_dma_sync_sg_for_cpu(struct pci_dev *hwdev, struct scatterlist *sg, int nelems, int direction);
-extern void pci_dma_sync_sg_for_device(struct pci_dev *hwdev, struct scatterlist *sg, int nelems, int direction);
-
-/* Return whether the given PCI device DMA address mask can
- * be supported properly.  For example, if your device can
- * only drive the low 24-bits during PCI bus mastering, then
- * you would pass 0x00ffffff as the mask to this function.
- */
-static inline int pci_dma_supported(struct pci_dev *hwdev, u64 mask)
-{
-	return 1;
-}
-
 #ifdef CONFIG_PCI
 static inline void pci_dma_burst_advice(struct pci_dev *pdev,
 					enum pci_dma_burst_strategy *strat,
@@ -154,14 +57,6 @@ static inline void pci_dma_burst_advice(struct pci_dev *pdev,
 }
 #endif
 
-#define PCI_DMA_ERROR_CODE      (~(dma_addr_t)0x0)
-
-static inline int pci_dma_mapping_error(struct pci_dev *pdev,
-					dma_addr_t dma_addr)
-{
-        return (dma_addr == PCI_DMA_ERROR_CODE);
-}
-
 struct device_node;
 extern struct device_node *pci_device_to_OF_node(struct pci_dev *pdev);
 
diff --git a/arch/sparc/include/asm/pci_64.h b/arch/sparc/include/asm/pci_64.h
index 7a1e356..5cc9f6a 100644
--- a/arch/sparc/include/asm/pci_64.h
+++ b/arch/sparc/include/asm/pci_64.h
@@ -35,37 +35,6 @@ static inline void pcibios_penalize_isa_irq(int irq, int active)
  */
 #define PCI_DMA_BUS_IS_PHYS	(0)
 
-static inline void *pci_alloc_consistent(struct pci_dev *pdev, size_t size,
-					 dma_addr_t *dma_handle)
-{
-	return dma_alloc_coherent(&pdev->dev, size, dma_handle, GFP_ATOMIC);
-}
-
-static inline void pci_free_consistent(struct pci_dev *pdev, size_t size,
-				       void *vaddr, dma_addr_t dma_handle)
-{
-	return dma_free_coherent(&pdev->dev, size, vaddr, dma_handle);
-}
-
-static inline dma_addr_t pci_map_single(struct pci_dev *pdev, void *ptr,
-					size_t size, int direction)
-{
-	return dma_map_single(&pdev->dev, ptr, size,
-			      (enum dma_data_direction) direction);
-}
-
-static inline void pci_unmap_single(struct pci_dev *pdev, dma_addr_t dma_addr,
-				    size_t size, int direction)
-{
-	dma_unmap_single(&pdev->dev, dma_addr, size,
-			 (enum dma_data_direction) direction);
-}
-
-#define pci_map_page(dev, page, off, size, dir) \
-	pci_map_single(dev, (page_address(page) + (off)), size, dir)
-#define pci_unmap_page(dev,addr,sz,dir) \
-	pci_unmap_single(dev,addr,sz,dir)
-
 /* pci_unmap_{single,page} is not a nop, thus... */
 #define DECLARE_PCI_UNMAP_ADDR(ADDR_NAME)	\
 	dma_addr_t ADDR_NAME;
@@ -80,57 +49,6 @@ static inline void pci_unmap_single(struct pci_dev *pdev, dma_addr_t dma_addr,
 #define pci_unmap_len_set(PTR, LEN_NAME, VAL)		\
 	(((PTR)->LEN_NAME) = (VAL))
 
-static inline int pci_map_sg(struct pci_dev *pdev, struct scatterlist *sg,
-			     int nents, int direction)
-{
-	return dma_map_sg(&pdev->dev, sg, nents,
-			  (enum dma_data_direction) direction);
-}
-
-static inline void pci_unmap_sg(struct pci_dev *pdev, struct scatterlist *sg,
-				int nents, int direction)
-{
-	dma_unmap_sg(&pdev->dev, sg, nents,
-		     (enum dma_data_direction) direction);
-}
-
-static inline void pci_dma_sync_single_for_cpu(struct pci_dev *pdev,
-					       dma_addr_t dma_handle,
-					       size_t size, int direction)
-{
-	dma_sync_single_for_cpu(&pdev->dev, dma_handle, size,
-				(enum dma_data_direction) direction);
-}
-
-static inline void pci_dma_sync_single_for_device(struct pci_dev *pdev,
-						  dma_addr_t dma_handle,
-						  size_t size, int direction)
-{
-	/* No flushing needed to sync cpu writes to the device.  */
-}
-
-static inline void pci_dma_sync_sg_for_cpu(struct pci_dev *pdev,
-					   struct scatterlist *sg,
-					   int nents, int direction)
-{
-	dma_sync_sg_for_cpu(&pdev->dev, sg, nents,
-			    (enum dma_data_direction) direction);
-}
-
-static inline void pci_dma_sync_sg_for_device(struct pci_dev *pdev,
-					      struct scatterlist *sg,
-					      int nelems, int direction)
-{
-	/* No flushing needed to sync cpu writes to the device.  */
-}
-
-/* Return whether the given PCI device DMA address mask can
- * be supported properly.  For example, if your device can
- * only drive the low 24-bits during PCI bus mastering, then
- * you would pass 0x00ffffff as the mask to this function.
- */
-extern int pci_dma_supported(struct pci_dev *hwdev, u64 mask);
-
 /* PCI IOMMU mapping bypass support. */
 
 /* PCI 64-bit addressing works for all slots on all controller
@@ -140,12 +58,6 @@ extern int pci_dma_supported(struct pci_dev *hwdev, u64 mask);
 #define PCI64_REQUIRED_MASK	(~(dma64_addr_t)0)
 #define PCI64_ADDR_BASE		0xfffc000000000000UL
 
-static inline int pci_dma_mapping_error(struct pci_dev *pdev,
-					dma_addr_t dma_addr)
-{
-	return dma_mapping_error(&pdev->dev, dma_addr);
-}
-
 #ifdef CONFIG_PCI
 static inline void pci_dma_burst_advice(struct pci_dev *pdev,
 					enum pci_dma_burst_strategy *strat,
diff --git a/arch/sparc/kernel/dma.c b/arch/sparc/kernel/dma.c
index a5d50da..b2fa312 100644
--- a/arch/sparc/kernel/dma.c
+++ b/arch/sparc/kernel/dma.c
@@ -13,13 +13,17 @@
 #include <linux/pci.h>
 #endif
 
-#include "dma.h"
-
+/*
+ * Return whether the given PCI device DMA address mask can be
+ * supported properly.  For example, if your device can only drive the
+ * low 24-bits during PCI bus mastering, then you would pass
+ * 0x00ffffff as the mask to this function.
+ */
 int dma_supported(struct device *dev, u64 mask)
 {
 #ifdef CONFIG_PCI
 	if (dev->bus == &pci_bus_type)
-		return pci_dma_supported(to_pci_dev(dev), mask);
+		return 1;
 #endif
 	return 0;
 }
@@ -34,148 +38,3 @@ int dma_set_mask(struct device *dev, u64 dma_mask)
 	return -EOPNOTSUPP;
 }
 EXPORT_SYMBOL(dma_set_mask);
-
-static void *dma32_alloc_coherent(struct device *dev, size_t size,
-				  dma_addr_t *dma_handle, gfp_t flag)
-{
-#ifdef CONFIG_PCI
-	if (dev->bus == &pci_bus_type)
-		return pci_alloc_consistent(to_pci_dev(dev), size, dma_handle);
-#endif
-	return sbus_alloc_consistent(dev, size, dma_handle);
-}
-
-static void dma32_free_coherent(struct device *dev, size_t size,
-				void *cpu_addr, dma_addr_t dma_handle)
-{
-#ifdef CONFIG_PCI
-	if (dev->bus == &pci_bus_type) {
-		pci_free_consistent(to_pci_dev(dev), size,
-				    cpu_addr, dma_handle);
-		return;
-	}
-#endif
-	sbus_free_consistent(dev, size, cpu_addr, dma_handle);
-}
-
-static dma_addr_t dma32_map_page(struct device *dev, struct page *page,
-				 unsigned long offset, size_t size,
-				 enum dma_data_direction direction,
-				 struct dma_attrs *attrs)
-{
-#ifdef CONFIG_PCI
-	if (dev->bus == &pci_bus_type)
-		return pci_map_page(to_pci_dev(dev), page, offset,
-				    size, (int)direction);
-#endif
-	return sbus_map_page(dev, page, offset, size, (int)direction);
-}
-
-static void dma32_unmap_page(struct device *dev, dma_addr_t dma_address,
-			     size_t size, enum dma_data_direction direction,
-			     struct dma_attrs *attrs)
-{
-#ifdef CONFIG_PCI
-	if (dev->bus == &pci_bus_type) {
-		pci_unmap_page(to_pci_dev(dev), dma_address,
-			       size, (int)direction);
-		return;
-	}
-#endif
-	sbus_unmap_page(dev, dma_address, size, (int)direction);
-}
-
-static int dma32_map_sg(struct device *dev, struct scatterlist *sg,
-			int nents, enum dma_data_direction direction,
-			struct dma_attrs *attrs)
-{
-#ifdef CONFIG_PCI
-	if (dev->bus == &pci_bus_type)
-		return pci_map_sg(to_pci_dev(dev), sg, nents, (int)direction);
-#endif
-	return sbus_map_sg(dev, sg, nents, direction);
-}
-
-void dma32_unmap_sg(struct device *dev, struct scatterlist *sg,
-		    int nents, enum dma_data_direction direction,
-		    struct dma_attrs *attrs)
-{
-#ifdef CONFIG_PCI
-	if (dev->bus == &pci_bus_type) {
-		pci_unmap_sg(to_pci_dev(dev), sg, nents, (int)direction);
-		return;
-	}
-#endif
-	sbus_unmap_sg(dev, sg, nents, (int)direction);
-}
-
-static void dma32_sync_single_for_cpu(struct device *dev, dma_addr_t dma_handle,
-				      size_t size,
-				      enum dma_data_direction direction)
-{
-#ifdef CONFIG_PCI
-	if (dev->bus == &pci_bus_type) {
-		pci_dma_sync_single_for_cpu(to_pci_dev(dev), dma_handle,
-					    size, (int)direction);
-		return;
-	}
-#endif
-	sbus_dma_sync_single_for_cpu(dev, dma_handle, size, (int) direction);
-}
-
-static void dma32_sync_single_for_device(struct device *dev,
-					 dma_addr_t dma_handle, size_t size,
-					 enum dma_data_direction direction)
-{
-#ifdef CONFIG_PCI
-	if (dev->bus == &pci_bus_type) {
-		pci_dma_sync_single_for_device(to_pci_dev(dev), dma_handle,
-					       size, (int)direction);
-		return;
-	}
-#endif
-	sbus_dma_sync_single_for_device(dev, dma_handle, size, (int) direction);
-}
-
-static void dma32_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg,
-				  int nelems, enum dma_data_direction direction)
-{
-#ifdef CONFIG_PCI
-	if (dev->bus == &pci_bus_type) {
-		pci_dma_sync_sg_for_cpu(to_pci_dev(dev), sg,
-					nelems, (int)direction);
-		return;
-	}
-#endif
-	BUG();
-}
-
-static void dma32_sync_sg_for_device(struct device *dev,
-				     struct scatterlist *sg, int nelems,
-				     enum dma_data_direction direction)
-{
-#ifdef CONFIG_PCI
-	if (dev->bus == &pci_bus_type) {
-		pci_dma_sync_sg_for_device(to_pci_dev(dev), sg,
-					   nelems, (int)direction);
-		return;
-	}
-#endif
-	BUG();
-}
-
-static struct dma_map_ops dma32_dma_ops = {
-	.alloc_coherent		= dma32_alloc_coherent,
-	.free_coherent		= dma32_free_coherent,
-	.map_page		= dma32_map_page,
-	.unmap_page		= dma32_unmap_page,
-	.map_sg			= dma32_map_sg,
-	.unmap_sg		= dma32_unmap_sg,
-	.sync_single_for_cpu	= dma32_sync_single_for_cpu,
-	.sync_single_for_device	= dma32_sync_single_for_device,
-	.sync_sg_for_cpu	= dma32_sync_sg_for_cpu,
-	.sync_sg_for_device	= dma32_sync_sg_for_device,
-};
-
-struct dma_map_ops *dma_ops = &dma32_dma_ops;
-EXPORT_SYMBOL(dma_ops);
diff --git a/arch/sparc/kernel/dma.h b/arch/sparc/kernel/dma.h
deleted file mode 100644
index 680351e..0000000
--- a/arch/sparc/kernel/dma.h
+++ /dev/null
@@ -1,14 +0,0 @@
-void *sbus_alloc_consistent(struct device *dev, long len, u32 *dma_addrp);
-void sbus_free_consistent(struct device *dev, long n, void *p, u32 ba);
-dma_addr_t sbus_map_page(struct device *dev, struct page *page,
-			 unsigned long offset, size_t len, int direction);
-void sbus_unmap_page(struct device *dev, dma_addr_t ba,
-		       size_t n, int direction);
-int sbus_map_sg(struct device *dev, struct scatterlist *sg,
-		int n, int direction);
-void sbus_unmap_sg(struct device *dev, struct scatterlist *sg,
-		   int n, int direction);
-void sbus_dma_sync_single_for_cpu(struct device *dev, dma_addr_t ba,
-				  size_t size, int direction);
-void sbus_dma_sync_single_for_device(struct device *dev, dma_addr_t ba,
-				     size_t size, int direction);
diff --git a/arch/sparc/kernel/iommu.c b/arch/sparc/kernel/iommu.c
index 74b289c..7690cc2 100644
--- a/arch/sparc/kernel/iommu.c
+++ b/arch/sparc/kernel/iommu.c
@@ -840,6 +840,8 @@ static struct dma_map_ops sun4u_dma_ops = {
 struct dma_map_ops *dma_ops = &sun4u_dma_ops;
 EXPORT_SYMBOL(dma_ops);
 
+extern int pci64_dma_supported(struct pci_dev *pdev, u64 device_mask);
+
 int dma_supported(struct device *dev, u64 device_mask)
 {
 	struct iommu *iommu = dev->archdata.iommu;
@@ -853,7 +855,7 @@ int dma_supported(struct device *dev, u64 device_mask)
 
 #ifdef CONFIG_PCI
 	if (dev->bus == &pci_bus_type)
-		return pci_dma_supported(to_pci_dev(dev), device_mask);
+		return pci64_dma_supported(to_pci_dev(dev), device_mask);
 #endif
 
 	return 0;
diff --git a/arch/sparc/kernel/ioport.c b/arch/sparc/kernel/ioport.c
index 39ff1e0..1eb6043 100644
--- a/arch/sparc/kernel/ioport.c
+++ b/arch/sparc/kernel/ioport.c
@@ -48,8 +48,6 @@
 #include <asm/iommu.h>
 #include <asm/io-unit.h>
 
-#include "dma.h"
-
 #define mmu_inval_dma_area(p, l)	/* Anton pulled it out for 2.4.0-xx */
 
 static struct resource *_sparc_find_resource(struct resource *r,
@@ -246,7 +244,8 @@ EXPORT_SYMBOL(sbus_set_sbus64);
  * Typically devices use them for control blocks.
  * CPU may access them without any explicit flushing.
  */
-void *sbus_alloc_consistent(struct device *dev, long len, u32 *dma_addrp)
+static void *sbus_alloc_coherent(struct device *dev, size_t len,
+				 dma_addr_t *dma_addrp, gfp_t gfp)
 {
 	struct of_device *op = to_of_device(dev);
 	unsigned long len_total = (len + PAGE_SIZE-1) & PAGE_MASK;
@@ -299,7 +298,8 @@ err_nopages:
 	return NULL;
 }
 
-void sbus_free_consistent(struct device *dev, long n, void *p, u32 ba)
+static void sbus_free_coherent(struct device *dev, size_t n, void *p,
+			       dma_addr_t ba)
 {
 	struct resource *res;
 	struct page *pgv;
@@ -317,7 +317,7 @@ void sbus_free_consistent(struct device *dev, long n, void *p, u32 ba)
 
 	n = (n + PAGE_SIZE-1) & PAGE_MASK;
 	if ((res->end-res->start)+1 != n) {
-		printk("sbus_free_consistent: region 0x%lx asked 0x%lx\n",
+		printk("sbus_free_consistent: region 0x%lx asked 0x%zx\n",
 		    (long)((res->end-res->start)+1), n);
 		return;
 	}
@@ -337,8 +337,10 @@ void sbus_free_consistent(struct device *dev, long n, void *p, u32 ba)
  * CPU view of this memory may be inconsistent with
  * a device view and explicit flushing is necessary.
  */
-dma_addr_t sbus_map_page(struct device *dev, struct page *page,
-			 unsigned long offset, size_t len, int direction)
+static dma_addr_t sbus_map_page(struct device *dev, struct page *page,
+				unsigned long offset, size_t len,
+				enum dma_data_direction dir,
+				struct dma_attrs *attrs)
 {
 	void *va = page_address(page) + offset;
 
@@ -353,12 +355,14 @@ dma_addr_t sbus_map_page(struct device *dev, struct page *page,
 	return mmu_get_scsi_one(dev, va, len);
 }
 
-void sbus_unmap_page(struct device *dev, dma_addr_t ba, size_t n, int direction)
+static void sbus_unmap_page(struct device *dev, dma_addr_t ba, size_t n,
+			    enum dma_data_direction dir, struct dma_attrs *attrs)
 {
 	mmu_release_scsi_one(dev, ba, n);
 }
 
-int sbus_map_sg(struct device *dev, struct scatterlist *sg, int n, int direction)
+static int sbus_map_sg(struct device *dev, struct scatterlist *sg, int n,
+		       enum dma_data_direction dir, struct dma_attrs *attrs)
 {
 	mmu_get_scsi_sgl(dev, sg, n);
 
@@ -369,19 +373,38 @@ int sbus_map_sg(struct device *dev, struct scatterlist *sg, int n, int direction
 	return n;
 }
 
-void sbus_unmap_sg(struct device *dev, struct scatterlist *sg, int n, int direction)
+static void sbus_unmap_sg(struct device *dev, struct scatterlist *sg, int n,
+			  enum dma_data_direction dir, struct dma_attrs *attrs)
 {
 	mmu_release_scsi_sgl(dev, sg, n);
 }
 
-void sbus_dma_sync_single_for_cpu(struct device *dev, dma_addr_t ba, size_t size, int direction)
+static void sbus_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg,
+				 int n,	enum dma_data_direction dir)
 {
+	BUG();
 }
 
-void sbus_dma_sync_single_for_device(struct device *dev, dma_addr_t ba, size_t size, int direction)
+static void sbus_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
+				    int n, enum dma_data_direction dir)
 {
+	BUG();
 }
 
+struct dma_map_ops sbus_dma_ops = {
+	.alloc_coherent		= sbus_alloc_coherent,
+	.free_coherent		= sbus_free_coherent,
+	.map_page		= sbus_map_page,
+	.unmap_page		= sbus_unmap_page,
+	.map_sg			= sbus_map_sg,
+	.unmap_sg		= sbus_unmap_sg,
+	.sync_sg_for_cpu	= sbus_sync_sg_for_cpu,
+	.sync_sg_for_device	= sbus_sync_sg_for_device,
+};
+
+struct dma_map_ops *dma_ops = &sbus_dma_ops;
+EXPORT_SYMBOL(dma_ops);
+
 static int __init sparc_register_ioport(void)
 {
 	register_proc_sparc_ioport();
@@ -398,7 +421,8 @@ arch_initcall(sparc_register_ioport);
 /* Allocate and map kernel buffer using consistent mode DMA for a device.
  * hwdev should be valid struct pci_dev pointer for PCI devices.
  */
-void *pci_alloc_consistent(struct pci_dev *pdev, size_t len, dma_addr_t *pba)
+static void *pci32_alloc_coherent(struct device *dev, size_t len,
+				  dma_addr_t *pba, gfp_t gfp)
 {
 	unsigned long len_total = (len + PAGE_SIZE-1) & PAGE_MASK;
 	unsigned long va;
@@ -442,7 +466,6 @@ void *pci_alloc_consistent(struct pci_dev *pdev, size_t len, dma_addr_t *pba)
 	*pba = virt_to_phys(va); /* equals virt_to_bus (R.I.P.) for us. */
 	return (void *) res->start;
 }
-EXPORT_SYMBOL(pci_alloc_consistent);
 
 /* Free and unmap a consistent DMA buffer.
  * cpu_addr is what was returned from pci_alloc_consistent,
@@ -452,7 +475,8 @@ EXPORT_SYMBOL(pci_alloc_consistent);
  * References to the memory and mappings associated with cpu_addr/dma_addr
  * past this call are illegal.
  */
-void pci_free_consistent(struct pci_dev *pdev, size_t n, void *p, dma_addr_t ba)
+static void pci32_free_coherent(struct device *dev, size_t n, void *p,
+				dma_addr_t ba)
 {
 	struct resource *res;
 	unsigned long pgp;
@@ -484,60 +508,18 @@ void pci_free_consistent(struct pci_dev *pdev, size_t n, void *p, dma_addr_t ba)
 
 	free_pages(pgp, get_order(n));
 }
-EXPORT_SYMBOL(pci_free_consistent);
-
-/* Map a single buffer of the indicated size for DMA in streaming mode.
- * The 32-bit bus address to use is returned.
- *
- * Once the device is given the dma address, the device owns this memory
- * until either pci_unmap_single or pci_dma_sync_single_* is performed.
- */
-dma_addr_t pci_map_single(struct pci_dev *hwdev, void *ptr, size_t size,
-    int direction)
-{
-	BUG_ON(direction == PCI_DMA_NONE);
-	/* IIep is write-through, not flushing. */
-	return virt_to_phys(ptr);
-}
-EXPORT_SYMBOL(pci_map_single);
-
-/* Unmap a single streaming mode DMA translation.  The dma_addr and size
- * must match what was provided for in a previous pci_map_single call.  All
- * other usages are undefined.
- *
- * After this call, reads by the cpu to the buffer are guaranteed to see
- * whatever the device wrote there.
- */
-void pci_unmap_single(struct pci_dev *hwdev, dma_addr_t ba, size_t size,
-    int direction)
-{
-	BUG_ON(direction == PCI_DMA_NONE);
-	if (direction != PCI_DMA_TODEVICE) {
-		mmu_inval_dma_area((unsigned long)phys_to_virt(ba),
-		    (size + PAGE_SIZE-1) & PAGE_MASK);
-	}
-}
-EXPORT_SYMBOL(pci_unmap_single);
 
 /*
  * Same as pci_map_single, but with pages.
  */
-dma_addr_t pci_map_page(struct pci_dev *hwdev, struct page *page,
-			unsigned long offset, size_t size, int direction)
+static dma_addr_t pci32_map_page(struct device *dev, struct page *page,
+				 unsigned long offset, size_t size,
+				 enum dma_data_direction dir,
+				 struct dma_attrs *attrs)
 {
-	BUG_ON(direction == PCI_DMA_NONE);
 	/* IIep is write-through, not flushing. */
 	return page_to_phys(page) + offset;
 }
-EXPORT_SYMBOL(pci_map_page);
-
-void pci_unmap_page(struct pci_dev *hwdev,
-			dma_addr_t dma_address, size_t size, int direction)
-{
-	BUG_ON(direction == PCI_DMA_NONE);
-	/* mmu_inval_dma_area XXX */
-}
-EXPORT_SYMBOL(pci_unmap_page);
 
 /* Map a set of buffers described by scatterlist in streaming
  * mode for DMA.  This is the scather-gather version of the
@@ -554,13 +536,13 @@ EXPORT_SYMBOL(pci_unmap_page);
  * Device ownership issues as mentioned above for pci_map_single are
  * the same here.
  */
-int pci_map_sg(struct pci_dev *hwdev, struct scatterlist *sgl, int nents,
-    int direction)
+static int pci32_map_sg(struct device *device, struct scatterlist *sgl,
+			int nents, enum dma_data_direction dir,
+			struct dma_attrs *attrs)
 {
 	struct scatterlist *sg;
 	int n;
 
-	BUG_ON(direction == PCI_DMA_NONE);
 	/* IIep is write-through, not flushing. */
 	for_each_sg(sgl, sg, nents, n) {
 		BUG_ON(page_address(sg_page(sg)) == NULL);
@@ -569,20 +551,19 @@ int pci_map_sg(struct pci_dev *hwdev, struct scatterlist *sgl, int nents,
 	}
 	return nents;
 }
-EXPORT_SYMBOL(pci_map_sg);
 
 /* Unmap a set of streaming mode DMA translations.
  * Again, cpu read rules concerning calls here are the same as for
  * pci_unmap_single() above.
  */
-void pci_unmap_sg(struct pci_dev *hwdev, struct scatterlist *sgl, int nents,
-    int direction)
+static void pci32_unmap_sg(struct device *dev, struct scatterlist *sgl,
+			   int nents, enum dma_data_direction dir,
+			   struct dma_attrs *attrs)
 {
 	struct scatterlist *sg;
 	int n;
 
-	BUG_ON(direction == PCI_DMA_NONE);
-	if (direction != PCI_DMA_TODEVICE) {
+	if (dir != PCI_DMA_TODEVICE) {
 		for_each_sg(sgl, sg, nents, n) {
 			BUG_ON(page_address(sg_page(sg)) == NULL);
 			mmu_inval_dma_area(
@@ -591,7 +572,6 @@ void pci_unmap_sg(struct pci_dev *hwdev, struct scatterlist *sgl, int nents,
 		}
 	}
 }
-EXPORT_SYMBOL(pci_unmap_sg);
 
 /* Make physical memory consistent for a single
  * streaming mode DMA translation before or after a transfer.
@@ -603,25 +583,23 @@ EXPORT_SYMBOL(pci_unmap_sg);
  * must first perform a pci_dma_sync_for_device, and then the
  * device again owns the buffer.
  */
-void pci_dma_sync_single_for_cpu(struct pci_dev *hwdev, dma_addr_t ba, size_t size, int direction)
+static void pci32_sync_single_for_cpu(struct device *dev, dma_addr_t ba,
+				      size_t size, enum dma_data_direction dir)
 {
-	BUG_ON(direction == PCI_DMA_NONE);
-	if (direction != PCI_DMA_TODEVICE) {
+	if (dir != PCI_DMA_TODEVICE) {
 		mmu_inval_dma_area((unsigned long)phys_to_virt(ba),
 		    (size + PAGE_SIZE-1) & PAGE_MASK);
 	}
 }
-EXPORT_SYMBOL(pci_dma_sync_single_for_cpu);
 
-void pci_dma_sync_single_for_device(struct pci_dev *hwdev, dma_addr_t ba, size_t size, int direction)
+static void pci32_sync_single_for_device(struct device *dev, dma_addr_t ba,
+					 size_t size, enum dma_data_direction dir)
 {
-	BUG_ON(direction == PCI_DMA_NONE);
-	if (direction != PCI_DMA_TODEVICE) {
+	if (dir != PCI_DMA_TODEVICE) {
 		mmu_inval_dma_area((unsigned long)phys_to_virt(ba),
 		    (size + PAGE_SIZE-1) & PAGE_MASK);
 	}
 }
-EXPORT_SYMBOL(pci_dma_sync_single_for_device);
 
 /* Make physical memory consistent for a set of streaming
  * mode DMA translations after a transfer.
@@ -629,13 +607,13 @@ EXPORT_SYMBOL(pci_dma_sync_single_for_device);
  * The same as pci_dma_sync_single_* but for a scatter-gather list,
  * same rules and usage.
  */
-void pci_dma_sync_sg_for_cpu(struct pci_dev *hwdev, struct scatterlist *sgl, int nents, int direction)
+static void pci32_sync_sg_for_cpu(struct device *dev, struct scatterlist *sgl,
+				  int nents, enum dma_data_direction dir)
 {
 	struct scatterlist *sg;
 	int n;
 
-	BUG_ON(direction == PCI_DMA_NONE);
-	if (direction != PCI_DMA_TODEVICE) {
+	if (dir != PCI_DMA_TODEVICE) {
 		for_each_sg(sgl, sg, nents, n) {
 			BUG_ON(page_address(sg_page(sg)) == NULL);
 			mmu_inval_dma_area(
@@ -644,15 +622,14 @@ void pci_dma_sync_sg_for_cpu(struct pci_dev *hwdev, struct scatterlist *sgl, int
 		}
 	}
 }
-EXPORT_SYMBOL(pci_dma_sync_sg_for_cpu);
 
-void pci_dma_sync_sg_for_device(struct pci_dev *hwdev, struct scatterlist *sgl, int nents, int direction)
+static void pci32_sync_sg_for_device(struct device *device, struct scatterlist *sgl,
+				     int nents, enum dma_data_direction dir)
 {
 	struct scatterlist *sg;
 	int n;
 
-	BUG_ON(direction == PCI_DMA_NONE);
-	if (direction != PCI_DMA_TODEVICE) {
+	if (dir != PCI_DMA_TODEVICE) {
 		for_each_sg(sgl, sg, nents, n) {
 			BUG_ON(page_address(sg_page(sg)) == NULL);
 			mmu_inval_dma_area(
@@ -661,7 +638,20 @@ void pci_dma_sync_sg_for_device(struct pci_dev *hwdev, struct scatterlist *sgl,
 		}
 	}
 }
-EXPORT_SYMBOL(pci_dma_sync_sg_for_device);
+
+struct dma_map_ops pci32_dma_ops = {
+	.alloc_coherent		= pci32_alloc_coherent,
+	.free_coherent		= pci32_free_coherent,
+	.map_page		= pci32_map_page,
+	.map_sg			= pci32_map_sg,
+	.unmap_sg		= pci32_unmap_sg,
+	.sync_single_for_cpu	= pci32_sync_single_for_cpu,
+	.sync_single_for_device	= pci32_sync_single_for_device,
+	.sync_sg_for_cpu	= pci32_sync_sg_for_cpu,
+	.sync_sg_for_device	= pci32_sync_sg_for_device,
+};
+EXPORT_SYMBOL(pci32_dma_ops);
+
 #endif /* CONFIG_PCI */
 
 #ifdef CONFIG_PROC_FS
diff --git a/arch/sparc/kernel/pci.c b/arch/sparc/kernel/pci.c
index 57859ad..c686486 100644
--- a/arch/sparc/kernel/pci.c
+++ b/arch/sparc/kernel/pci.c
@@ -1039,7 +1039,7 @@ static void ali_sound_dma_hack(struct pci_dev *pdev, int set_bit)
 	pci_dev_put(ali_isa_bridge);
 }
 
-int pci_dma_supported(struct pci_dev *pdev, u64 device_mask)
+int pci64_dma_supported(struct pci_dev *pdev, u64 device_mask)
 {
 	u64 dma_addr_mask;
 
-- 
1.6.0.6

Re: [PATCH 7/8] sparc: use asm-generic/pci-dma-compat

[David Miller]

From: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Date: Mon, 13 Jul 2009 15:25:59 +0900

> This converts SPARC to use asm-generic/pci-dma-compat instead of the
> homegrown mechnism.
> 
> SPARC32 has two dma_map_ops structures for pci and sbus (removing
> arch/sparc/kernel/dma.c, PCI and SBUS DMA accessor). The global
> 'dma_ops' is set to sbus_dma_ops and get_dma_ops() returns
> pci32_dma_ops for pci devices so we can use the appropriate dma
> mapping operations.
> 
> Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
> Tested-by: Robert Reif <reif@earthlink.net>

ACK

[PATCH 8/8] sparc: add CONFIG_DMA_API_DEBUG support

[FUJITA Tomonori]

All we need to do for CONFIG_DMA_API_DEBUG support is call
dma_debug_init() in DMA code common for SPARC32 and SPARC64.

Now SPARC32 uses two dma_map_ops structures for pci and sbus so there
is not much dma stuff for SPARC32 in kernel/dma.c. kernel/ioport.c
also includes dma stuff for SPARC32. So let's put all the dma stuff
for SPARC32 in kernel/ioport.c and make kernel/dma.c common for
SPARC32 and SPARC64.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
---
 arch/sparc/Kconfig                   |    1 +
 arch/sparc/include/asm/dma-mapping.h |    6 ++++-
 arch/sparc/kernel/Makefile           |    2 +-
 arch/sparc/kernel/dma.c              |   37 ++++-----------------------------
 arch/sparc/kernel/ioport.c           |   27 ++++++++++++++++++++++++
 5 files changed, 39 insertions(+), 34 deletions(-)

diff --git a/arch/sparc/Kconfig b/arch/sparc/Kconfig
index 5f2df99..233cff5 100644
--- a/arch/sparc/Kconfig
+++ b/arch/sparc/Kconfig
@@ -26,6 +26,7 @@ config SPARC
 	select RTC_CLASS
 	select RTC_DRV_M48T59
 	select HAVE_DMA_ATTRS
+	select HAVE_DMA_API_DEBUG
 
 config SPARC32
 	def_bool !64BIT
diff --git a/arch/sparc/include/asm/dma-mapping.h b/arch/sparc/include/asm/dma-mapping.h
index 2677818..5a8c308 100644
--- a/arch/sparc/include/asm/dma-mapping.h
+++ b/arch/sparc/include/asm/dma-mapping.h
@@ -32,8 +32,11 @@ static inline void *dma_alloc_coherent(struct device *dev, size_t size,
 				       dma_addr_t *dma_handle, gfp_t flag)
 {
 	struct dma_map_ops *ops = get_dma_ops(dev);
+	void *cpu_addr;
 
-	return ops->alloc_coherent(dev, size, dma_handle, flag);
+	cpu_addr = ops->alloc_coherent(dev, size, dma_handle, flag);
+	debug_dma_alloc_coherent(dev, size, *dma_handle, cpu_addr);
+	return cpu_addr;
 }
 
 static inline void dma_free_coherent(struct device *dev, size_t size,
@@ -41,6 +44,7 @@ static inline void dma_free_coherent(struct device *dev, size_t size,
 {
 	struct dma_map_ops *ops = get_dma_ops(dev);
 
+	debug_dma_free_coherent(dev, size, cpu_addr, dma_handle);
 	ops->free_coherent(dev, size, cpu_addr, dma_handle);
 }
 
diff --git a/arch/sparc/kernel/Makefile b/arch/sparc/kernel/Makefile
index 475ce46..29b88a5 100644
--- a/arch/sparc/kernel/Makefile
+++ b/arch/sparc/kernel/Makefile
@@ -61,7 +61,7 @@ obj-$(CONFIG_SPARC64_SMP) += cpumap.o
 obj-$(CONFIG_SPARC32)     += devres.o
 devres-y                  := ../../../kernel/irq/devres.o
 
-obj-$(CONFIG_SPARC32)     += dma.o
+obj-y                     += dma.o
 
 obj-$(CONFIG_SPARC32_PCI) += pcic.o
 
diff --git a/arch/sparc/kernel/dma.c b/arch/sparc/kernel/dma.c
index b2fa312..e1ba8ee 100644
--- a/arch/sparc/kernel/dma.c
+++ b/arch/sparc/kernel/dma.c
@@ -1,40 +1,13 @@
-/* dma.c: PCI and SBUS DMA accessors for 32-bit sparc.
- *
- * Copyright (C) 2008 David S. Miller <davem@davemloft.net>
- */
-
 #include <linux/kernel.h>
 #include <linux/module.h>
 #include <linux/dma-mapping.h>
-#include <linux/scatterlist.h>
-#include <linux/mm.h>
+#include <linux/dma-debug.h>
 
-#ifdef CONFIG_PCI
-#include <linux/pci.h>
-#endif
+#define PREALLOC_DMA_DEBUG_ENTRIES       (1 << 15)
 
-/*
- * Return whether the given PCI device DMA address mask can be
- * supported properly.  For example, if your device can only drive the
- * low 24-bits during PCI bus mastering, then you would pass
- * 0x00ffffff as the mask to this function.
- */
-int dma_supported(struct device *dev, u64 mask)
+static int __init dma_init(void)
 {
-#ifdef CONFIG_PCI
-	if (dev->bus == &pci_bus_type)
-		return 1;
-#endif
+	dma_debug_init(PREALLOC_DMA_DEBUG_ENTRIES);
 	return 0;
 }
-EXPORT_SYMBOL(dma_supported);
-
-int dma_set_mask(struct device *dev, u64 dma_mask)
-{
-#ifdef CONFIG_PCI
-	if (dev->bus == &pci_bus_type)
-		return pci_set_dma_mask(to_pci_dev(dev), dma_mask);
-#endif
-	return -EOPNOTSUPP;
-}
-EXPORT_SYMBOL(dma_set_mask);
+fs_initcall(dma_init);
diff --git a/arch/sparc/kernel/ioport.c b/arch/sparc/kernel/ioport.c
index 1eb6043..edbea23 100644
--- a/arch/sparc/kernel/ioport.c
+++ b/arch/sparc/kernel/ioport.c
@@ -654,6 +654,33 @@ EXPORT_SYMBOL(pci32_dma_ops);
 
 #endif /* CONFIG_PCI */
 
+/*
+ * Return whether the given PCI device DMA address mask can be
+ * supported properly.  For example, if your device can only drive the
+ * low 24-bits during PCI bus mastering, then you would pass
+ * 0x00ffffff as the mask to this function.
+ */
+int dma_supported(struct device *dev, u64 mask)
+{
+#ifdef CONFIG_PCI
+	if (dev->bus == &pci_bus_type)
+		return 1;
+#endif
+	return 0;
+}
+EXPORT_SYMBOL(dma_supported);
+
+int dma_set_mask(struct device *dev, u64 dma_mask)
+{
+#ifdef CONFIG_PCI
+	if (dev->bus == &pci_bus_type)
+		return pci_set_dma_mask(to_pci_dev(dev), dma_mask);
+#endif
+	return -EOPNOTSUPP;
+}
+EXPORT_SYMBOL(dma_set_mask);
+
+
 #ifdef CONFIG_PROC_FS
 
 static int
-- 
1.6.0.6

Re: [PATCH 8/8] sparc: add CONFIG_DMA_API_DEBUG support

[David Miller]

From: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Date: Mon, 13 Jul 2009 15:26:00 +0900

> All we need to do for CONFIG_DMA_API_DEBUG support is call
> dma_debug_init() in DMA code common for SPARC32 and SPARC64.
> 
> Now SPARC32 uses two dma_map_ops structures for pci and sbus so there
> is not much dma stuff for SPARC32 in kernel/dma.c. kernel/ioport.c
> also includes dma stuff for SPARC32. So let's put all the dma stuff
> for SPARC32 in kernel/ioport.c and make kernel/dma.c common for
> SPARC32 and SPARC64.
> 
> Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>

ACK, we just need to tweak that illegal DMA address check so
that it doesn't erroneously trigger on sparc32.

Re: [PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[Robert Reif]

FUJITA Tomonori wrote:
> This is the second version of the patchset to convert SPARC to
> asm-generic/dma-mapping-common.h and pci-dma-compat.h:
>
>   
I still get the same runtime warning with this patch set at boot time.

The bad address is within the kernel so it looks like
it's catching a real bug.

cat kallsyms | grep f0007000
f0007000 T trapbase_cpu3

WARNING: at lib/dma-debug.c:873 check_for_illegal_area+0xc8/0x100()
esp ffd7ba30: DMA-API: device driver maps memory from kernel text or 
rodata [addr=f0007000] [len=4096]
Modules linked in: ext3 jbd sd_mod sun_esp esp_scsi scsi_transport_spi 
scsi_mod
[f011ffa0 : check_for_illegal_area+0xc8/0x100 ]
[f0120b1c : debug_dma_map_sg+0x158/0x1c4 ]
[fe6946d0 : sbus_esp_map_sg+0xec/0xfc [sun_esp] ]
[fe684454 : esp_maybe_execute_command+0x1f0/0x6f8 [esp_scsi] ]
[fe6849e4 : esp_queuecommand+0x88/0x98 [esp_scsi] ]
[fe638c00 : scsi_dispatch_cmd+0x200/0x2a8 [scsi_mod] ]
[fe63f038 : scsi_request_fn+0x3f4/0x52c [scsi_mod] ]
[f00f6d64 : __generic_unplug_device+0x40/0x50 ]
[f00f6d90 : generic_unplug_device+0x1c/0x38 ]
[f00f4be4 : blk_unplug+0x14/0x24 ]
[f00ade1c : sync_buffer+0x48/0x60 ]
[f01f6d58 : __wait_on_bit+0x4c/0xa4 ]
[f01f6e60 : out_of_line_wait_on_bit+0xb0/0xc0 ]
[fe709320 : __ext3_get_inode_loc+0x3bc/0x418 [ext3] ]
[fe7093b4 : ext3_iget+0x38/0x65c [ext3] ]
[fe712c70 : ext3_get_journal+0xc/0xd4 [ext3] ]

Re: [PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[FUJITA Tomonori]

On Mon, 13 Jul 2009 20:56:21 -0400
Robert Reif <reif@earthlink.net> wrote:

> FUJITA Tomonori wrote:
> > This is the second version of the patchset to convert SPARC to
> > asm-generic/dma-mapping-common.h and pci-dma-compat.h:
> >
> >   
> I still get the same runtime warning with this patch set at boot time.
> 
> The bad address is within the kernel so it looks like
> it's catching a real bug.
> 
> cat kallsyms | grep f0007000
> f0007000 T trapbase_cpu3
> 
> WARNING: at lib/dma-debug.c:873 check_for_illegal_area+0xc8/0x100()
> esp ffd7ba30: DMA-API: device driver maps memory from kernel text or 
> rodata [addr=f0007000] [len=4096]
> Modules linked in: ext3 jbd sd_mod sun_esp esp_scsi scsi_transport_spi 

Ok, I looked at check_for_illegal_area() in dma-debug.

What check_for_illegal_area() does looks bogus to me with some of I/O
remapping hardware.


> scsi_mod
> [f011ffa0 : check_for_illegal_area+0xc8/0x100 ]
> [f0120b1c : debug_dma_map_sg+0x158/0x1c4 ]
> [fe6946d0 : sbus_esp_map_sg+0xec/0xfc [sun_esp] ]
> [fe684454 : esp_maybe_execute_command+0x1f0/0x6f8 [esp_scsi] ]
> [fe6849e4 : esp_queuecommand+0x88/0x98 [esp_scsi] ]
> [fe638c00 : scsi_dispatch_cmd+0x200/0x2a8 [scsi_mod] ]
> [fe63f038 : scsi_request_fn+0x3f4/0x52c [scsi_mod] ]
> [f00f6d64 : __generic_unplug_device+0x40/0x50 ]
> [f00f6d90 : generic_unplug_device+0x1c/0x38 ]
> [f00f4be4 : blk_unplug+0x14/0x24 ]
> [f00ade1c : sync_buffer+0x48/0x60 ]
> [f01f6d58 : __wait_on_bit+0x4c/0xa4 ]
> [f01f6e60 : out_of_line_wait_on_bit+0xb0/0xc0 ]
> [fe709320 : __ext3_get_inode_loc+0x3bc/0x418 [ext3] ]
> [fe7093b4 : ext3_iget+0x38/0x65c [ext3] ]
> [fe712c70 : ext3_get_journal+0xc/0xd4 [ext3] ]
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Re: [PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[Joerg Roedel]

On Tue, Jul 14, 2009 at 10:40:16AM +0900, FUJITA Tomonori wrote:
> On Mon, 13 Jul 2009 20:56:21 -0400
> Robert Reif <reif@earthlink.net> wrote:
> 
> > FUJITA Tomonori wrote:
> > > This is the second version of the patchset to convert SPARC to
> > > asm-generic/dma-mapping-common.h and pci-dma-compat.h:
> > >
> > >   
> > I still get the same runtime warning with this patch set at boot time.
> > 
> > The bad address is within the kernel so it looks like
> > it's catching a real bug.
> > 
> > cat kallsyms | grep f0007000
> > f0007000 T trapbase_cpu3
> > 
> > WARNING: at lib/dma-debug.c:873 check_for_illegal_area+0xc8/0x100()
> > esp ffd7ba30: DMA-API: device driver maps memory from kernel text or 
> > rodata [addr=f0007000] [len=4096]
> > Modules linked in: ext3 jbd sd_mod sun_esp esp_scsi scsi_transport_spi 
> 
> Ok, I looked at check_for_illegal_area() in dma-debug.
> 
> What check_for_illegal_area() does looks bogus to me with some of I/O
> remapping hardware.

Can you be more specific about this one? check_for_illegal_area() should
not depend on any hardware because all it does is checking the machine
addresses to be mapped.

	Joerg

Re: [PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[David Miller]

From: Joerg Roedel <joerg.roedel@amd.com>
Date: Tue, 14 Jul 2009 11:23:55 +0200

> On Tue, Jul 14, 2009 at 10:40:16AM +0900, FUJITA Tomonori wrote:
>> On Mon, 13 Jul 2009 20:56:21 -0400
>> Robert Reif <reif@earthlink.net> wrote:
>> 
>> > The bad address is within the kernel so it looks like
>> > it's catching a real bug.
>> > 
>> > cat kallsyms | grep f0007000
>> > f0007000 T trapbase_cpu3
>> > 
>> > WARNING: at lib/dma-debug.c:873 check_for_illegal_area+0xc8/0x100()
>> > esp ffd7ba30: DMA-API: device driver maps memory from kernel text or 
>> > rodata [addr=f0007000] [len=4096]
>> > Modules linked in: ext3 jbd sd_mod sun_esp esp_scsi scsi_transport_spi 
>> 
>> Ok, I looked at check_for_illegal_area() in dma-debug.
>> 
>> What check_for_illegal_area() does looks bogus to me with some of I/O
>> remapping hardware.
> 
> Can you be more specific about this one? check_for_illegal_area() should
> not depend on any hardware because all it does is checking the machine
> addresses to be mapped.

The check can't work properly on sparc32.

Sparc32 always maps the kernel to a fixed physical location, and it
therefore can execute in the identity mapping area of physical memory
like where all the free pages and kmalloc areas live virtually.

So if we free up some pages within the kernel image (because the
memory is unused, for exmple that's what's happening here with the
extra trap table pages on Robert's machine) we have pages in the free
page pool that are located right inside of the kernel text, data, etc.

We'll thus need a way to turn off these checks somehow.  You could
also augment this check by seeing if there is a backing page, and if
so, whether it is PageReserved or not.  That's just one idea.

Re: [PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[Ingo Molnar]

* David Miller <davem@davemloft.net> wrote:

> From: Joerg Roedel <joerg.roedel@amd.com>
> Date: Tue, 14 Jul 2009 11:23:55 +0200
> 
> > On Tue, Jul 14, 2009 at 10:40:16AM +0900, FUJITA Tomonori wrote:
> >> On Mon, 13 Jul 2009 20:56:21 -0400
> >> Robert Reif <reif@earthlink.net> wrote:
> >> 
> >> > The bad address is within the kernel so it looks like
> >> > it's catching a real bug.
> >> > 
> >> > cat kallsyms | grep f0007000
> >> > f0007000 T trapbase_cpu3
> >> > 
> >> > WARNING: at lib/dma-debug.c:873 check_for_illegal_area+0xc8/0x100()
> >> > esp ffd7ba30: DMA-API: device driver maps memory from kernel text or 
> >> > rodata [addr=f0007000] [len=4096]
> >> > Modules linked in: ext3 jbd sd_mod sun_esp esp_scsi scsi_transport_spi 
> >> 
> >> Ok, I looked at check_for_illegal_area() in dma-debug.
> >> 
> >> What check_for_illegal_area() does looks bogus to me with some of I/O
> >> remapping hardware.
> > 
> > Can you be more specific about this one? 
> > check_for_illegal_area() should not depend on any hardware 
> > because all it does is checking the machine addresses to be 
> > mapped.
> 
> The check can't work properly on sparc32.
> 
> Sparc32 always maps the kernel to a fixed physical location, and 
> it therefore can execute in the identity mapping area of physical 
> memory like where all the free pages and kmalloc areas live 
> virtually.
> 
> So if we free up some pages within the kernel image (because the 
> memory is unused, for exmple that's what's happening here with the 
> extra trap table pages on Robert's machine) we have pages in the 
> free page pool that are located right inside of the kernel text, 
> data, etc.
> 
> We'll thus need a way to turn off these checks somehow.  You could 
> also augment this check by seeing if there is a backing page, and 
> if so, whether it is PageReserved or not.  That's just one idea.

Hm, note, this sparc32 behavior might break certain aspects of 
lockdep as well, see kernel/lockdep.c::static_obj(). You could get 
spurious non-printing of 'trying to register non-static key'.

I'm wondering why sparc32 frees from the middle of the kernel image. 
The way architectures generally do it is to put freeable pages into 
a separate section. That way it does not get mingled with the kernel 
core image area (which stays nicely continuous).

	Ingo

Re: [PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[David Miller]

From: Ingo Molnar <mingo@elte.hu>
Date: Sat, 18 Jul 2009 13:06:20 +0200

> I'm wondering why sparc32 frees from the middle of the kernel image. 
> The way architectures generally do it is to put freeable pages into 
> a separate section. That way it does not get mingled with the kernel 
> core image area (which stays nicely continuous).

It's conditionally free'able, we actually use the pages
if the system is SMP and has more than cpu 0 installed.

Re: [PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[Ingo Molnar]

* David Miller <davem@davemloft.net> wrote:

> From: Ingo Molnar <mingo@elte.hu>
> Date: Sat, 18 Jul 2009 13:06:20 +0200
> 
> > I'm wondering why sparc32 frees from the middle of the kernel 
> > image. The way architectures generally do it is to put freeable 
> > pages into a separate section. That way it does not get mingled 
> > with the kernel core image area (which stays nicely continuous).
> 
> It's conditionally free'able, we actually use the pages if the 
> system is SMP and has more than cpu 0 installed.

Sounds similar to the alternate instructions section on x86:

 arch/x86/kernel/alternative.c:

        if (smp_alt_once)
                free_init_pages("SMP alternatives",
                                (unsigned long)__smp_locks,
                                (unsigned long)__smp_locks_end);

It's in a separate section and linked + freed separately. The linker 
script page-aligns it:

 arch/x86/kernel/vmlinux.lds.S:

        /*
         * smp_locks might be freed after init
         * start/end must be page aligned
         */
        . = ALIGN(PAGE_SIZE);
        .smp_locks : AT(ADDR(.smp_locks) - LOAD_OFFSET) {
                __smp_locks = .;
                *(.smp_locks)
                __smp_locks_end = .;
                . = ALIGN(PAGE_SIZE);
        }

So that it can be freed. It's also placed next to the regular init 
section to not create too many holes. It also keeps the 'kernel text 
area' continuous and things like core_kernel_text() & co still work 
fine.

	Ingo

Re: [PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[David Miller]

From: Ingo Molnar <mingo@elte.hu>
Date: Mon, 3 Aug 2009 07:54:41 +0200

> So that it can be freed. It's also placed next to the regular init 
> section to not create too many holes. It also keeps the 'kernel text 
> area' continuous and things like core_kernel_text() & co still work 
> fine.

Ok, I'll look into doing similar on sparc.

Re: [PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[Joerg Roedel]

On Tue, Jul 14, 2009 at 01:12:24PM -0700, David Miller wrote:
> From: Joerg Roedel <joerg.roedel@amd.com>
> Date: Tue, 14 Jul 2009 11:23:55 +0200
> 
> > On Tue, Jul 14, 2009 at 10:40:16AM +0900, FUJITA Tomonori wrote:
> >> On Mon, 13 Jul 2009 20:56:21 -0400
> >> Robert Reif <reif@earthlink.net> wrote:
> >> 
> >> > The bad address is within the kernel so it looks like
> >> > it's catching a real bug.
> >> > 
> >> > cat kallsyms | grep f0007000
> >> > f0007000 T trapbase_cpu3
> >> > 
> >> > WARNING: at lib/dma-debug.c:873 check_for_illegal_area+0xc8/0x100()
> >> > esp ffd7ba30: DMA-API: device driver maps memory from kernel text or 
> >> > rodata [addr=f0007000] [len=4096]
> >> > Modules linked in: ext3 jbd sd_mod sun_esp esp_scsi scsi_transport_spi 
> >> 
> >> Ok, I looked at check_for_illegal_area() in dma-debug.
> >> 
> >> What check_for_illegal_area() does looks bogus to me with some of I/O
> >> remapping hardware.
> > 
> > Can you be more specific about this one? check_for_illegal_area() should
> > not depend on any hardware because all it does is checking the machine
> > addresses to be mapped.
> 
> The check can't work properly on sparc32.
> 
> Sparc32 always maps the kernel to a fixed physical location, and it
> therefore can execute in the identity mapping area of physical memory
> like where all the free pages and kmalloc areas live virtually.
> 
> So if we free up some pages within the kernel image (because the
> memory is unused, for exmple that's what's happening here with the
> extra trap table pages on Robert's machine) we have pages in the free
> page pool that are located right inside of the kernel text, data, etc.
> 
> We'll thus need a way to turn off these checks somehow.  You could
> also augment this check by seeing if there is a backing page, and if
> so, whether it is PageReserved or not.  That's just one idea.

Hmm these checks sound specific and hard to maintain. I think its best
to give architectures the option whether to enable this check or not.

	Joerg

Re: [PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[David Miller]

From: Joerg Roedel <joerg.roedel@amd.com>
Date: Tue, 21 Jul 2009 18:05:38 +0200

> Hmm these checks sound specific and hard to maintain. I think its best
> to give architectures the option whether to enable this check or not.

No matter how the arch works in this area, there should be some
kind of sanity check.

So, it probably makes sense to allow an arch to override the
validation function rather than turn the test completely off.

Re: [PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[Ingo Molnar]

* David Miller <davem@davemloft.net> wrote:

> From: Joerg Roedel <joerg.roedel@amd.com>
> Date: Tue, 21 Jul 2009 18:05:38 +0200
> 
> > Hmm these checks sound specific and hard to maintain. I think 
> > its best to give architectures the option whether to enable this 
> > check or not.
> 
> No matter how the arch works in this area, there should be some 
> kind of sanity check.
> 
> So, it probably makes sense to allow an arch to override the 
> validation function rather than turn the test completely off.

I'm wondering how sparc32 handles core_kernel_text() & friends? 

Those functions already assume that the bits between _stext and 
_etext are all core kernel text and not generic pages freed back to 
the buddy.

	Ingo

Re: [PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[David Miller]

From: Ingo Molnar <mingo@elte.hu>
Date: Mon, 3 Aug 2009 07:56:08 +0200

> I'm wondering how sparc32 handles core_kernel_text() & friends? 
> 
> Those functions already assume that the bits between _stext and 
> _etext are all core kernel text and not generic pages freed back to 
> the buddy.

They won't work for such pages obviously.  Shows how many sparc32
users who trigger this case of actually freeing those pages and
using such facilities there actually are :-)

Re: [PATCH v2 0/8] sparc: use asm-generic/dma-mapping-common.h and pci-dma-compat.h

[Ingo Molnar]

* David Miller <davem@davemloft.net> wrote:

> From: Ingo Molnar <mingo@elte.hu>
> Date: Mon, 3 Aug 2009 07:56:08 +0200
> 
> > I'm wondering how sparc32 handles core_kernel_text() & friends? 
> > 
> > Those functions already assume that the bits between _stext and 
> > _etext are all core kernel text and not generic pages freed back 
> > to the buddy.
> 
> They won't work for such pages obviously.  Shows how many sparc32 
> users who trigger this case of actually freeing those pages and 
> using such facilities there actually are :-)

:)

I suspect the other bit is probably that core_kernel_text() really 
didnt use to have any critical use at all - just debugging/lockdep 
mostly. So no functionality broke for people to report.

Such bugs often have years of latency on x86 too - we recently fixed 
one that was there for over a year and nobody noticed:

 4a44bac: symbols, stacktrace: look up init symbols after module symbols

	Ingo