Hello,
Here there is a series of 20 patches that lays the foundations for

using dma_ops in i386 in the very same way x86_64, as well as many other

architectures already do.
The functions themselves for i386 are placed in a pci-base_32.c, but just

a few among them are actually implemented. Most were no-ops anyway.
Also, as I said, this is by no means a complete coverage of dma_ops.

there are still some call sites to be patches in pci-dma_32.c (although I don't

really plan to change them, but to integrate them in a single pci-dma.c).

I intend to have it done progressively.
The granularity is per-operation, meaning each patch moves one specific function

to the common header. This is compiled-tested in both i386 and x86_64 in

~5 randconfigs each, and boot-tested in my hardware with my default configs
The motivation for that is the ongoing work for pci-passthrough in KVM.

So ingo, avi, what do you think it's the best way to handle these patches through?
Thanks
--

Thanks; Please also see Stephen Tweedie's tree at
http://git.et.redhat.com/?p=linux-2.6-dom0-pvops.git;a=summary
I guess your work would be overlapped anyway, but just to make sure there's no

double work.

--

I see the headers are unified, but the .c files are duplicated.  I 
x86.git.
--

Do not meddle in the internals of kernels, for they are subtle and quick to panic.
--

presume correctly.
--

looks very nice to me! I've applied it to x86.git, lets see what

happens.
	Ingo

--

Houston, we've got a problem! :-/
randconfig testing found that this patchset broke sendfile on certain

.config's - DMA started returning all 0xfffffffff data, corrupting

files. (config attached)
After some bisection fun it turns out that the conversions from struct

page are wrong:
+static inline dma_addr_t dma_map_page(struct device *dev, struct page *page,

+                                     size_t offset, size_t size,

+                                     int direction)

+{

+       return dma_map_single(dev, page_address(page)+offset, size, direction);
because page_address() is not defined for highmem pages in general. (and

even if it's defined, it will corrupt data)
changing it to page_to_phys() is not good because it goes via a 32-bit

bottleneck that trims things on PAE:
       dma_addr_t      (*map_single)(struct device *hwdev, void *ptr,
the 'ptr' is 32-bit albeit it's a DMA target.
what i came up is the prototype 32-bit fix below - this works on 32-bit

but breaks 64-bit because we pass in physical addresses instead of

virtual direct addresses.
i'll fix the 64-bit side but that means materially touching all the

dma_mapping_ops instantiations materially on the 64-bit side - not

really something we wanted to do :-/
	Ingo
---------------->

Subject: x86: dma-ops on highmem fix

From: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

---

 arch/x86/kernel/pci-base_32.c |    4 ++--

 include/asm-x86/dma-mapping.h |   10 ++++++----

 2 files changed, 8 insertions(+), 6 deletions(-)
Index: linux-x86.q/arch/x86/kernel/pci-base_32.c

===================================================================

--- linux-x86.q.orig/arch/x86/kernel/pci-base_32.c

+++ linux-x86.q/arch/x86/kernel/pci-base_32.c

@@ -4,12 +4,12 @@

 #include <linux/dma-mapping.h>

 #include <asm/dma-mapping.h>
-static dma_addr_t pci32_map_single(struct device *dev, void *ptr,

+static dma_addr_t pci32_map_single(struct device *de...
[ message continues ]

the full fix ended up being the one below. It's not that bad - and

gart_64.c looks even a bit cleaner. Still, it needs careful review.
	Ingo
--------------->

Subject: x86: dma-ops on highmem fix

From: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

---

 arch/x86/kernel/pci-base_32.c    |    4 ++--

 arch/x86/kernel/pci-calgary_64.c |    3 ++-

 arch/x86/kernel/pci-dma_64.c     |    2 +-

 arch/x86/kernel/pci-gart_64.c    |   15 +++++++--------

 arch/x86/kernel/pci-nommu_64.c   |    4 ++--

 arch/x86/kernel/pci-swiotlb_64.c |    9 ++++++++-

 include/asm-x86/dma-mapping.h    |   10 ++++++----

 7 files changed, 28 insertions(+), 19 deletions(-)
Index: linux-x86.q/arch/x86/kernel/pci-base_32.c

===================================================================

--- linux-x86.q.orig/arch/x86/kernel/pci-base_32.c

+++ linux-x86.q/arch/x86/kernel/pci-base_32.c

@@ -4,12 +4,12 @@

 #include <linux/dma-mapping.h>

 #include <asm/dma-mapping.h>
-static dma_addr_t pci32_map_single(struct device *dev, void *ptr,

+static dma_addr_t pci32_map_single(struct device *dev, phys_addr_t ptr,

 				   size_t size, int direction)

 {

 	WARN_ON(size == 0);

 	flush_write_buffers();

-	return virt_to_phys(ptr);

+	return ptr;

 }
 static int pci32_dma_map_sg(struct device *dev, struct scatterlist *sglist,

Index: linux-x86.q/arch/x86/kernel/pci-calgary_64.c

===================================================================

--- linux-x86.q.orig/arch/x86/kernel/pci-calgary_64.c

+++ linux-x86.q/arch/x86/kernel/pci-calgary_64.c

@@ -470,10 +470,11 @@ error:

 	return 0;

 }
-static dma_addr_t calgary_map_single(struct device *dev, void *vaddr,

+static dma_addr_t calgary_map_single(struct device *dev, phys_addr_t paddr,

 	size_t size, int direction)

 {

 	dma_addr_t dma_handle = bad_dma_address;

+	void *vaddr = phys_to_virt(paddr);

 	unsigned long uaddr;

 	unsigned int npages;

 	struct iommu_table *tbl = find_iommu_table(dev);

Index: linux-x...
[ message continues ]

It looks all good to me.

I'll give a shot in my systems to see if it goes okay.

--

take it off the x86_64 specific header
Signed-off-by: Glauber Costa <gcosta@redhat.com>

---

 include/asm-x86/dma-mapping.h    |   54 ++++++++++++++++++++++++++++++++++++++

 include/asm-x86/dma-mapping_64.h |   49 ----------------------------------

 2 files changed, 54 insertions(+), 49 deletions(-)
diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h

index 58f790f..aebd178 100644

--- a/include/asm-x86/dma-mapping.h

+++ b/include/asm-x86/dma-mapping.h

@@ -1,5 +1,59 @@

+#ifndef _ASM_DMA_MAPPING_H_

+#define _ASM_DMA_MAPPING_H_

+

+/*

+ * IOMMU interface. See Documentation/DMA-mapping.txt and DMA-API.txt for

+ * documentation.

+ */

+

+#include <linux/scatterlist.h>

+#include <asm/io.h>

+#include <asm/swiotlb.h>

+

+struct dma_mapping_ops {

+	int             (*mapping_error)(dma_addr_t dma_addr);

+	void*           (*alloc_coherent)(struct device *dev, size_t size,

+				dma_addr_t *dma_handle, gfp_t gfp);

+	void            (*free_coherent)(struct device *dev, size_t size,

+				void *vaddr, dma_addr_t dma_handle);

+	dma_addr_t      (*map_single)(struct device *hwdev, void *ptr,

+				size_t size, int direction);

+	/* like map_single, but doesn't check the device mask */

+	dma_addr_t      (*map_simple)(struct device *hwdev, char *ptr,

+				size_t size, int direction);

+	void            (*unmap_single)(struct device *dev, dma_addr_t addr,

+				size_t size, int direction);

+	void            (*sync_single_for_cpu)(struct device *hwdev,

+				dma_addr_t dma_handle, size_t size,

+				int direction);

+	void            (*sync_single_for_device)(struct device *hwdev,

+				dma_addr_t dma_handle, size_t size,

+				int direction);

+	void            (*sync_single_range_for_cpu)(struct device *hwdev,

+				dma_addr_t dma_handle, unsigned long offset,

+				size_t size, int direction);

+	void            (*sync_single_range_for_device)(struct device *hwdev,

+				dma_addr_t dma_handle, unsigned long offset,

+				size_t size, int direction);

+	voi...
[ message continues ]

That's already the name of the game for x86_64. For i386,

we add a pci-base_32.c, that will hold the default operations.

The function call itself goes through dma-mapping.h , the common

header
Signed-off-by: Glauber Costa <gcosta@redhat.com>

---

 arch/x86/kernel/Makefile         |    1 +

 arch/x86/kernel/pci-base_32.c    |   20 ++++++++++++++++++++

 include/asm-x86/dma-mapping.h    |   10 ++++++++++

 include/asm-x86/dma-mapping_32.h |   10 ----------

 include/asm-x86/dma-mapping_64.h |    9 ---------

 5 files changed, 31 insertions(+), 19 deletions(-)

 create mode 100644 arch/x86/kernel/pci-base_32.c
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile

index 340221f..95fbf57 100644

--- a/arch/x86/kernel/Makefile

+++ b/arch/x86/kernel/Makefile

@@ -27,6 +27,7 @@ obj-y			+= pci-dma_$(BITS).o  bootflag.o e820_$(BITS).o

 obj-y			+= quirks.o i8237.o topology.o kdebugfs.o

 obj-y			+= alternative.o i8253.o

 obj-$(CONFIG_X86_64)	+= pci-nommu_64.o bugs_64.o

+obj-$(CONFIG_X86_32)	+= pci-base_32.o

 obj-y			+= tsc_$(BITS).o io_delay.o rtc.o
 obj-y				+= process.o

diff --git a/arch/x86/kernel/pci-base_32.c b/arch/x86/kernel/pci-base_32.c

new file mode 100644

index 0000000..b613d73

--- /dev/null

+++ b/arch/x86/kernel/pci-base_32.c

@@ -0,0 +1,20 @@

+#include <linux/mm.h>

+#include <linux/kernel.h>

+#include <linux/module.h>

+#include <linux/dma-mapping.h>

+#include <asm/dma-mapping.h>

+

+static dma_addr_t pci32_map_single(struct device *dev, void *ptr,

+				   size_t size, int direction)

+{

+	WARN_ON(size == 0);

+	flush_write_buffers();

+	return virt_to_phys(ptr);

+}

+

+static const struct dma_mapping_ops pci32_dma_ops = {

+	.map_single = pci32_map_single,

+};

+

+const struct dma_mapping_ops *dma_ops = &pci32_dma_ops;

+EXPORT_SYMBOL(dma_ops);

diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h

index aebd178..d320244 100644

--- a/include/asm-x86/dma-mapping.h

+++ b/include/asm-x86/dma-mapping.h

@...
[ message continues ]

i386 base does not need it, so it gets an empty function
Signed-off-by: Glauber Costa <gcosta@redhat.com>

---

 arch/x86/kernel/pci-base_32.c    |    1 +

 include/asm-x86/dma-mapping.h    |   10 ++++++++++

 include/asm-x86/dma-mapping_32.h |    7 -------

 include/asm-x86/dma-mapping_64.h |    8 --------

 4 files changed, 11 insertions(+), 15 deletions(-)
diff --git a/arch/x86/kernel/pci-base_32.c b/arch/x86/kernel/pci-base_32.c

index b613d73..a8a7c7f 100644

--- a/arch/x86/kernel/pci-base_32.c

+++ b/arch/x86/kernel/pci-base_32.c

@@ -14,6 +14,7 @@ static dma_addr_t pci32_map_single(struct device *dev, void *ptr,
 static const struct dma_mapping_ops pci32_dma_ops = {

 	.map_single = pci32_map_single,

+	.unmap_single = NULL,

 };
 const struct dma_mapping_ops *dma_ops = &pci32_dma_ops;

diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h

index d320244..bb0378f 100644

--- a/include/asm-x86/dma-mapping.h

+++ b/include/asm-x86/dma-mapping.h

@@ -66,4 +66,14 @@ dma_map_single(struct device *hwdev, void *ptr, size_t size,

 	return dma_ops->map_single(hwdev, ptr, size, direction);

 }
+static inline void

+dma_unmap_single(struct device *dev, dma_addr_t addr, size_t size,

+		 int direction)

+{

+	BUG_ON(!valid_dma_direction(direction));

+	if (dma_ops->unmap_single)

+		dma_ops->unmap_single(dev, addr, size, direction);

+}

+

+

 #endif

diff --git a/include/asm-x86/dma-mapping_32.h b/include/asm-x86/dma-mapping_32.h

index b496306..0b27cb0 100644

--- a/include/asm-x86/dma-mapping_32.h

+++ b/include/asm-x86/dma-mapping_32.h

@@ -17,13 +17,6 @@ void *dma_alloc_coherent(struct device *dev, size_t size,

 void dma_free_coherent(struct device *dev, size_t size,

 			 void *vaddr, dma_addr_t dma_handle);
-static inline void

-dma_unmap_single(struct device *dev, dma_addr_t dma_addr, size_t size,

-		 enum dma_data_direction direction)

-{

-	BUG_ON(!valid_dma_direction(direction));

-}

-

 static inline int

 dma_map_sg(struct device *dev, str...
[ message continues ]

the old i386 implementation is moved to pci-base_32.c
Signed-off-by: Glauber Costa <gcosta@redhat.com>

---

 arch/x86/kernel/pci-base_32.c    |   19 +++++++++++++++++++

 include/asm-x86/dma-mapping.h    |    8 +++++++-

 include/asm-x86/dma-mapping_32.h |   20 --------------------

 include/asm-x86/dma-mapping_64.h |    7 -------

 4 files changed, 26 insertions(+), 28 deletions(-)
diff --git a/arch/x86/kernel/pci-base_32.c b/arch/x86/kernel/pci-base_32.c

index a8a7c7f..2474152 100644

--- a/arch/x86/kernel/pci-base_32.c

+++ b/arch/x86/kernel/pci-base_32.c

@@ -12,9 +12,28 @@ static dma_addr_t pci32_map_single(struct device *dev, void *ptr,

 	return virt_to_phys(ptr);

 }
+static int pci32_dma_map_sg(struct device *dev, struct scatterlist *sglist,

+			    int nents, int direction)

+{

+	struct scatterlist *sg;

+	int i;

+

+	WARN_ON(nents == 0 || sglist[0].length == 0);

+

+	for_each_sg(sglist, sg, nents, i) {

+		BUG_ON(!sg_page(sg));

+

+		sg->dma_address = sg_phys(sg);

+	}

+

+	flush_write_buffers();

+	return nents;

+}

+

 static const struct dma_mapping_ops pci32_dma_ops = {

 	.map_single = pci32_map_single,

 	.unmap_single = NULL,

+	.map_sg = pci32_dma_map_sg,

 };
 const struct dma_mapping_ops *dma_ops = &pci32_dma_ops;

diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h

index bb0378f..0901154 100644

--- a/include/asm-x86/dma-mapping.h

+++ b/include/asm-x86/dma-mapping.h

@@ -75,5 +75,11 @@ dma_unmap_single(struct device *dev, dma_addr_t addr, size_t size,

 		dma_ops->unmap_single(dev, addr, size, direction);

 }
-

+static inline int

+dma_map_sg(struct device *hwdev, struct scatterlist *sg,

+	   int nents, int direction)

+{

+	BUG_ON(!valid_dma_direction(direction));

+	return dma_ops->map_sg(hwdev, sg, nents, direction);

+}

 #endif

diff --git a/include/asm-x86/dma-mapping_32.h b/include/asm-x86/dma-mapping_32.h

index 0b27cb0..cdcdeff 100644

--- a/include/asm-x86/dma-mapping_32.h

+++ b/include/asm-x86/dma-mapping_32...
[ message continues ]

i386 gets an empty function
Signed-off-by: Glauber Costa <gcosta@redhat.com>

---

 arch/x86/kernel/pci-base_32.c    |    1 +

 include/asm-x86/dma-mapping.h    |    9 +++++++++

 include/asm-x86/dma-mapping_32.h |    8 --------

 include/asm-x86/dma-mapping_64.h |    8 --------

 4 files changed, 10 insertions(+), 16 deletions(-)
diff --git a/arch/x86/kernel/pci-base_32.c b/arch/x86/kernel/pci-base_32.c

index 2474152..9205304 100644

--- a/arch/x86/kernel/pci-base_32.c

+++ b/arch/x86/kernel/pci-base_32.c

@@ -34,6 +34,7 @@ static const struct dma_mapping_ops pci32_dma_ops = {

 	.map_single = pci32_map_single,

 	.unmap_single = NULL,

 	.map_sg = pci32_dma_map_sg,

+	.unmap_sg = NULL,

 };
 const struct dma_mapping_ops *dma_ops = &pci32_dma_ops;

diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h

index 0901154..6e7747a 100644

--- a/include/asm-x86/dma-mapping.h

+++ b/include/asm-x86/dma-mapping.h

@@ -82,4 +82,13 @@ dma_map_sg(struct device *hwdev, struct scatterlist *sg,

 	BUG_ON(!valid_dma_direction(direction));

 	return dma_ops->map_sg(hwdev, sg, nents, direction);

 }

+

+static inline void

+dma_unmap_sg(struct device *hwdev, struct scatterlist *sg, int nents,

+	     int direction)

+{

+	BUG_ON(!valid_dma_direction(direction));

+	if (dma_ops->unmap_sg)

+		dma_ops->unmap_sg(hwdev, sg, nents, direction);

+}

 #endif

diff --git a/include/asm-x86/dma-mapping_32.h b/include/asm-x86/dma-mapping_32.h

index cdcdeff..55445e3 100644

--- a/include/asm-x86/dma-mapping_32.h

+++ b/include/asm-x86/dma-mapping_32.h

@@ -32,14 +32,6 @@ dma_unmap_page(struct device *dev, dma_addr_t dma_address, size_t size,

 	BUG_ON(!valid_dma_direction(direction));

 }
-

-static inline void

-dma_unmap_sg(struct device *dev, struct scatterlist *sg, int nhwentries,

-	     enum dma_data_direction direction)

-{

-	BUG_ON(!valid_dma_direction(direction));

-}

-

 static inline void

 dma_sync_single_for_cpu(struct device *dev, dma_addr_t dma_handle, size_t size,

 	...
[ message continues ]