LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
* common non-cache coherent direct dma mapping ops
@ 2018-05-11  7:59 Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 01/20] dma-mapping: simplify Kconfig dependencies Christoph Hellwig
                   ` (20 more replies)
  0 siblings, 21 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Hi all,

this series continues consolidating the dma-mapping code, with a focus
on architectures that do not (always) provide cache coherence for DMA.
Three architectures (arm, mips and powerpc) are still left to be
converted later due to complexity of their dma ops selection.

The dma-noncoherent ops calls the dma-direct ops for the actual
translation of streaming mappins and allow the architecture to provide
any cache flushing required for cpu to device and/or device to cpu
ownership transfers.  The dma coherent allocator is for now still left
entirely to architecture supplied implementations due the amount of
variations.  Hopefully we can do some consolidation for them later on
as well.

A lot of architectures are currently doing very questionable things
in their dma mapping routines, which are documented in the changelogs
for each patch.  Please review them very careful and correct me on
incorrect assumptions.

Because this series sits on top of two previously submitted series
a git tree might be useful to actually test it.  It is provided here:

    git://git.infradead.org/users/hch/misc.git generic-dma-noncoherent

Gitweb:

    http://git.infradead.org/users/hch/misc.git/shortlog/refs/heads/generic-dma-noncoherent

Changes since RFC:
 - fix a typo accidentally disabling the device to cpu transfer sync
 - fixed a few compile failures

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 01/20] dma-mapping: simplify Kconfig dependencies
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation Christoph Hellwig
                   ` (19 subsequent siblings)
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

ARCH_DMA_ADDR_T_64BIT is always true for 64-bit architectures now, so we
can skip the clause requiring it.  'n' is the default default, so no need
to explicitly state it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 lib/Kconfig | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/lib/Kconfig b/lib/Kconfig
index 1d84e61cccfe..6c4e9d0ce5d1 100644
--- a/lib/Kconfig
+++ b/lib/Kconfig
@@ -443,13 +443,11 @@ config IOMMU_HELPER
 
 config DMA_DIRECT_OPS
 	bool
-	depends on HAS_DMA && (!64BIT || ARCH_DMA_ADDR_T_64BIT)
-	default n
+	depends on HAS_DMA
 
 config DMA_VIRT_OPS
 	bool
-	depends on HAS_DMA && (!64BIT || ARCH_DMA_ADDR_T_64BIT)
-	default n
+	depends on HAS_DMA
 
 config SWIOTLB
 	bool
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 01/20] dma-mapping: simplify Kconfig dependencies Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-18 13:03   ` Alexey Brodkin
  2018-05-11  7:59 ` [PATCH 03/20] arc: use generic dma_noncoherent_ops Christoph Hellwig
                   ` (18 subsequent siblings)
  20 siblings, 1 reply; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Add a new dma_map_ops implementation that uses dma-direct for the
address mapping of streaming mappings, and which requires arch-specific
implemenations of coherent allocate/free.

Architectures have to provide flushing helpers to ownership trasnfers
to the device and/or CPU, and can provide optional implementations of
the coherent mmap functionality, and the cache_flush routines for
non-coherent long term allocations.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 MAINTAINERS                       |   2 +
 include/asm-generic/dma-mapping.h |   9 +++
 include/linux/dma-direct.h        |   7 ++-
 include/linux/dma-mapping.h       |   1 +
 include/linux/dma-noncoherent.h   |  47 ++++++++++++++
 lib/Kconfig                       |  20 ++++++
 lib/Makefile                      |   1 +
 lib/dma-direct.c                  |   8 +--
 lib/dma-noncoherent.c             | 101 ++++++++++++++++++++++++++++++
 9 files changed, 191 insertions(+), 5 deletions(-)
 create mode 100644 include/linux/dma-noncoherent.h
 create mode 100644 lib/dma-noncoherent.c

diff --git a/MAINTAINERS b/MAINTAINERS
index 79bb02ff812f..08d0d15d4958 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -4334,12 +4334,14 @@ W:	http://git.infradead.org/users/hch/dma-mapping.git
 S:	Supported
 F:	lib/dma-debug.c
 F:	lib/dma-direct.c
+F:	lib/dma-noncoherent.c
 F:	lib/dma-virt.c
 F:	drivers/base/dma-mapping.c
 F:	drivers/base/dma-coherent.c
 F:	include/asm-generic/dma-mapping.h
 F:	include/linux/dma-direct.h
 F:	include/linux/dma-mapping.h
+F:	include/linux/dma-noncoherent.h
 
 DME1737 HARDWARE MONITOR DRIVER
 M:	Juerg Haefliger <juergh@gmail.com>
diff --git a/include/asm-generic/dma-mapping.h b/include/asm-generic/dma-mapping.h
index 880a292d792f..ad2868263867 100644
--- a/include/asm-generic/dma-mapping.h
+++ b/include/asm-generic/dma-mapping.h
@@ -4,7 +4,16 @@
 
 static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
 {
+	/*
+	 * Use the non-coherent ops if available.  If an architecture wants a
+	 * more fine-grained selection of operations it will have to implement
+	 * get_arch_dma_ops itself or use the per-device dma_ops.
+	 */
+#ifdef CONFIG_DMA_NONCOHERENT_OPS
+	return &dma_noncoherent_ops;
+#else
 	return &dma_direct_ops;
+#endif
 }
 
 #endif /* _ASM_GENERIC_DMA_MAPPING_H */
diff --git a/include/linux/dma-direct.h b/include/linux/dma-direct.h
index 53ad6a47f513..8d9f33febde5 100644
--- a/include/linux/dma-direct.h
+++ b/include/linux/dma-direct.h
@@ -59,6 +59,11 @@ void *dma_direct_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
 		gfp_t gfp, unsigned long attrs);
 void dma_direct_free(struct device *dev, size_t size, void *cpu_addr,
 		dma_addr_t dma_addr, unsigned long attrs);
+dma_addr_t dma_direct_map_page(struct device *dev, struct page *page,
+		unsigned long offset, size_t size, enum dma_data_direction dir,
+		unsigned long attrs);
+int dma_direct_map_sg(struct device *dev, struct scatterlist *sgl, int nents,
+		enum dma_data_direction dir, unsigned long attrs);
 int dma_direct_supported(struct device *dev, u64 mask);
-
+int dma_direct_mapping_error(struct device *dev, dma_addr_t dma_addr);
 #endif /* _LINUX_DMA_DIRECT_H */
diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
index 25a9a2b04f78..4be070df5fc5 100644
--- a/include/linux/dma-mapping.h
+++ b/include/linux/dma-mapping.h
@@ -136,6 +136,7 @@ struct dma_map_ops {
 };
 
 extern const struct dma_map_ops dma_direct_ops;
+extern const struct dma_map_ops dma_noncoherent_ops;
 extern const struct dma_map_ops dma_virt_ops;
 
 #define DMA_BIT_MASK(n)	(((n) == 64) ? ~0ULL : ((1ULL<<(n))-1))
diff --git a/include/linux/dma-noncoherent.h b/include/linux/dma-noncoherent.h
new file mode 100644
index 000000000000..10b2654d549b
--- /dev/null
+++ b/include/linux/dma-noncoherent.h
@@ -0,0 +1,47 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _LINUX_DMA_NONCOHERENT_H
+#define _LINUX_DMA_NONCOHERENT_H 1
+
+#include <linux/dma-mapping.h>
+
+void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
+		gfp_t gfp, unsigned long attrs);
+void arch_dma_free(struct device *dev, size_t size, void *cpu_addr,
+		dma_addr_t dma_addr, unsigned long attrs);
+
+#ifdef CONFIG_DMA_NONCOHERENT_MMAP
+int arch_dma_mmap(struct device *dev, struct vm_area_struct *vma,
+		void *cpu_addr, dma_addr_t dma_addr, size_t size,
+		unsigned long attrs);
+#else
+#define arch_dma_mmap NULL
+#endif /* CONFIG_DMA_NONCOHERENT_MMAP */
+
+#ifdef CONFIG_DMA_NONCOHERENT_CACHE_SYNC
+void arch_dma_cache_sync(struct device *dev, void *vaddr, size_t size,
+		enum dma_data_direction direction);
+#else
+#define arch_dma_cache_sync NULL
+#endif /* CONFIG_DMA_NONCOHERENT_CACHE_SYNC */
+
+#ifdef CONFIG_ARCH_HAS_SYNC_DMA_FOR_DEVICE
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir);
+#else
+static inline void arch_sync_dma_for_device(struct device *dev,
+		phys_addr_t paddr, size_t size, enum dma_data_direction dir)
+{
+}
+#endif /* ARCH_HAS_SYNC_DMA_FOR_DEVICE */
+
+#ifdef CONFIG_ARCH_HAS_SYNC_DMA_FOR_CPU
+void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir);
+#else
+static inline void arch_sync_dma_for_cpu(struct device *dev,
+		phys_addr_t paddr, size_t size, enum dma_data_direction dir)
+{
+}
+#endif /* ARCH_HAS_SYNC_DMA_FOR_CPU */
+
+#endif /* _LINUX_DMA_NONCOHERENT_H */
diff --git a/lib/Kconfig b/lib/Kconfig
index 6c4e9d0ce5d1..7a913937888b 100644
--- a/lib/Kconfig
+++ b/lib/Kconfig
@@ -441,10 +441,30 @@ config ARCH_DMA_ADDR_T_64BIT
 config IOMMU_HELPER
 	bool
 
+config ARCH_HAS_SYNC_DMA_FOR_DEVICE
+	bool
+
+config ARCH_HAS_SYNC_DMA_FOR_CPU
+	bool
+	select NEED_DMA_MAP_STATE
+
 config DMA_DIRECT_OPS
 	bool
 	depends on HAS_DMA
 
+config DMA_NONCOHERENT_OPS
+	bool
+	depends on HAS_DMA
+	select DMA_DIRECT_OPS
+
+config DMA_NONCOHERENT_MMAP
+	bool
+	depends on DMA_NONCOHERENT_OPS
+
+config DMA_NONCOHERENT_CACHE_SYNC
+	bool
+	depends on DMA_NONCOHERENT_OPS
+
 config DMA_VIRT_OPS
 	bool
 	depends on HAS_DMA
diff --git a/lib/Makefile b/lib/Makefile
index 94203b5eecd4..9f18c8152281 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -30,6 +30,7 @@ lib-$(CONFIG_PRINTK) += dump_stack.o
 lib-$(CONFIG_MMU) += ioremap.o
 lib-$(CONFIG_SMP) += cpumask.o
 lib-$(CONFIG_DMA_DIRECT_OPS) += dma-direct.o
+lib-$(CONFIG_DMA_NONCOHERENT_OPS) += dma-noncoherent.o
 lib-$(CONFIG_DMA_VIRT_OPS) += dma-virt.o
 
 lib-y	+= kobject.o klist.o
diff --git a/lib/dma-direct.c b/lib/dma-direct.c
index df9e726e0712..b824eb218782 100644
--- a/lib/dma-direct.c
+++ b/lib/dma-direct.c
@@ -128,7 +128,7 @@ void dma_direct_free(struct device *dev, size_t size, void *cpu_addr,
 		free_pages((unsigned long)cpu_addr, page_order);
 }
 
-static dma_addr_t dma_direct_map_page(struct device *dev, struct page *page,
+dma_addr_t dma_direct_map_page(struct device *dev, struct page *page,
 		unsigned long offset, size_t size, enum dma_data_direction dir,
 		unsigned long attrs)
 {
@@ -139,8 +139,8 @@ static dma_addr_t dma_direct_map_page(struct device *dev, struct page *page,
 	return dma_addr;
 }
 
-static int dma_direct_map_sg(struct device *dev, struct scatterlist *sgl,
-		int nents, enum dma_data_direction dir, unsigned long attrs)
+int dma_direct_map_sg(struct device *dev, struct scatterlist *sgl, int nents,
+		enum dma_data_direction dir, unsigned long attrs)
 {
 	int i;
 	struct scatterlist *sg;
@@ -175,7 +175,7 @@ int dma_direct_supported(struct device *dev, u64 mask)
 	return 1;
 }
 
-static int dma_direct_mapping_error(struct device *dev, dma_addr_t dma_addr)
+int dma_direct_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
 	return dma_addr == DIRECT_MAPPING_ERROR;
 }
diff --git a/lib/dma-noncoherent.c b/lib/dma-noncoherent.c
new file mode 100644
index 000000000000..a2c192b3508d
--- /dev/null
+++ b/lib/dma-noncoherent.c
@@ -0,0 +1,101 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2018 Christoph Hellwig.
+ *
+ * DMA operations that map physical memory directly without providing cache
+ * coherence.
+ */
+#include <linux/export.h>
+#include <linux/mm.h>
+#include <linux/dma-direct.h>
+#include <linux/dma-noncoherent.h>
+#include <linux/scatterlist.h>
+
+static void dma_noncoherent_sync_single_for_device(struct device *dev,
+		dma_addr_t addr, size_t size, enum dma_data_direction dir)
+{
+	arch_sync_dma_for_device(dev, dma_to_phys(dev, addr), size, dir);
+}
+
+static void dma_noncoherent_sync_sg_for_device(struct device *dev,
+		struct scatterlist *sgl, int nents, enum dma_data_direction dir)
+{
+	struct scatterlist *sg;
+	int i;
+
+	for_each_sg(sgl, sg, nents, i)
+		arch_sync_dma_for_device(dev, sg_phys(sg), sg->length, dir);
+}
+
+static dma_addr_t dma_noncoherent_map_page(struct device *dev, struct page *page,
+		unsigned long offset, size_t size, enum dma_data_direction dir,
+		unsigned long attrs)
+{
+	dma_addr_t addr;
+
+	addr = dma_direct_map_page(dev, page, offset, size, dir, attrs);
+	if (!dma_mapping_error(dev, addr) && !(attrs & DMA_ATTR_SKIP_CPU_SYNC))
+		arch_sync_dma_for_device(dev, page_to_phys(page), size, dir);
+	return addr;
+}
+
+static int dma_noncoherent_map_sg(struct device *dev, struct scatterlist *sgl,
+		int nents, enum dma_data_direction dir, unsigned long attrs)
+{
+	nents = dma_direct_map_sg(dev, sgl, nents, dir, attrs);
+	if (nents > 0 && !(attrs & DMA_ATTR_SKIP_CPU_SYNC))
+		dma_noncoherent_sync_sg_for_device(dev, sgl, nents, dir);
+	return nents;
+}
+
+#ifdef CONFIG_ARCH_HAS_SYNC_DMA_FOR_CPU
+static void dma_noncoherent_sync_single_for_cpu(struct device *dev,
+		dma_addr_t addr, size_t size, enum dma_data_direction dir)
+{
+	arch_sync_dma_for_cpu(dev, dma_to_phys(dev, addr), size, dir);
+}
+
+static void dma_noncoherent_sync_sg_for_cpu(struct device *dev,
+		struct scatterlist *sgl, int nents, enum dma_data_direction dir)
+{
+	struct scatterlist *sg;
+	int i;
+
+	for_each_sg(sgl, sg, nents, i)
+		arch_sync_dma_for_cpu(dev, sg_phys(sg), sg->length, dir);
+}
+
+static void dma_noncoherent_unmap_page(struct device *dev, dma_addr_t addr,
+		size_t size, enum dma_data_direction dir, unsigned long attrs)
+{
+	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
+		dma_noncoherent_sync_single_for_cpu(dev, addr, size, dir);
+}
+
+static void dma_noncoherent_unmap_sg(struct device *dev, struct scatterlist *sgl,
+		int nents, enum dma_data_direction dir, unsigned long attrs)
+{
+	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
+		dma_noncoherent_sync_sg_for_cpu(dev, sgl, nents, dir);
+}
+#endif
+
+const struct dma_map_ops dma_noncoherent_ops = {
+	.alloc			= arch_dma_alloc,
+	.free			= arch_dma_free,
+	.mmap			= arch_dma_mmap,
+	.sync_single_for_device	= dma_noncoherent_sync_single_for_device,
+	.sync_sg_for_device	= dma_noncoherent_sync_sg_for_device,
+	.map_page		= dma_noncoherent_map_page,
+	.map_sg			= dma_noncoherent_map_sg,
+#ifdef CONFIG_ARCH_HAS_SYNC_DMA_FOR_CPU
+	.sync_single_for_cpu	= dma_noncoherent_sync_single_for_cpu,
+	.sync_sg_for_cpu	= dma_noncoherent_sync_sg_for_cpu,
+	.unmap_page		= dma_noncoherent_unmap_page,
+	.unmap_sg		= dma_noncoherent_unmap_sg,
+#endif
+	.dma_supported		= dma_direct_supported,
+	.mapping_error		= dma_direct_mapping_error,
+	.cache_sync		= arch_dma_cache_sync,
+};
+EXPORT_SYMBOL(dma_noncoherent_ops);
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 03/20] arc: use generic dma_noncoherent_ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 01/20] dma-mapping: simplify Kconfig dependencies Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11 12:44   ` Alexey Brodkin
  2018-05-11  7:59 ` [PATCH 04/20] arm-nommu: " Christoph Hellwig
                   ` (17 subsequent siblings)
  20 siblings, 1 reply; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Switch to the generic noncoherent direct mapping implementation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/arc/Kconfig                   |   4 +
 arch/arc/include/asm/Kbuild        |   1 +
 arch/arc/include/asm/dma-mapping.h |  21 -----
 arch/arc/mm/dma.c                  | 141 +++--------------------------
 4 files changed, 19 insertions(+), 148 deletions(-)
 delete mode 100644 arch/arc/include/asm/dma-mapping.h

diff --git a/arch/arc/Kconfig b/arch/arc/Kconfig
index 7498aca4b887..89d47eac18b2 100644
--- a/arch/arc/Kconfig
+++ b/arch/arc/Kconfig
@@ -9,11 +9,15 @@
 config ARC
 	def_bool y
 	select ARC_TIMERS
+	select ARCH_HAS_SYNC_DMA_FOR_CPU
+	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
 	select ARCH_HAS_SG_CHAIN
 	select ARCH_SUPPORTS_ATOMIC_RMW if ARC_HAS_LLSC
 	select BUILDTIME_EXTABLE_SORT
 	select CLONE_BACKWARDS
 	select COMMON_CLK
+	select DMA_NONCOHERENT_OPS
+	select DMA_NONCOHERENT_MMAP
 	select GENERIC_ATOMIC64 if !ISA_ARCV2 || !(ARC_HAS_LL64 && ARC_HAS_LLSC)
 	select GENERIC_CLOCKEVENTS
 	select GENERIC_FIND_FIRST_BIT
diff --git a/arch/arc/include/asm/Kbuild b/arch/arc/include/asm/Kbuild
index 4bd5d4369e05..bbdcb955e18f 100644
--- a/arch/arc/include/asm/Kbuild
+++ b/arch/arc/include/asm/Kbuild
@@ -2,6 +2,7 @@
 generic-y += bugs.h
 generic-y += device.h
 generic-y += div64.h
+generic-y += dma-mapping.h
 generic-y += emergency-restart.h
 generic-y += extable.h
 generic-y += fb.h
diff --git a/arch/arc/include/asm/dma-mapping.h b/arch/arc/include/asm/dma-mapping.h
deleted file mode 100644
index 7a16824bfe98..000000000000
--- a/arch/arc/include/asm/dma-mapping.h
+++ /dev/null
@@ -1,21 +0,0 @@
-/*
- * DMA Mapping glue for ARC
- *
- * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License version 2 as
- * published by the Free Software Foundation.
- */
-
-#ifndef ASM_ARC_DMA_MAPPING_H
-#define ASM_ARC_DMA_MAPPING_H
-
-extern const struct dma_map_ops arc_dma_ops;
-
-static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
-{
-	return &arc_dma_ops;
-}
-
-#endif
diff --git a/arch/arc/mm/dma.c b/arch/arc/mm/dma.c
index 1dcc404b5aec..e4cd485517b4 100644
--- a/arch/arc/mm/dma.c
+++ b/arch/arc/mm/dma.c
@@ -16,13 +16,12 @@
  * The default DMA address == Phy address which is 0x8000_0000 based.
  */
 
-#include <linux/dma-mapping.h>
+#include <linux/dma-noncoherent.h>
 #include <asm/cache.h>
 #include <asm/cacheflush.h>
 
-
-static void *arc_dma_alloc(struct device *dev, size_t size,
-		dma_addr_t *dma_handle, gfp_t gfp, unsigned long attrs)
+void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
+		gfp_t gfp, unsigned long attrs)
 {
 	unsigned long order = get_order(size);
 	struct page *page;
@@ -89,7 +88,7 @@ static void *arc_dma_alloc(struct device *dev, size_t size,
 	return kvaddr;
 }
 
-static void arc_dma_free(struct device *dev, size_t size, void *vaddr,
+void arch_dma_free(struct device *dev, size_t size, void *vaddr,
 		dma_addr_t dma_handle, unsigned long attrs)
 {
 	phys_addr_t paddr = dma_handle;
@@ -105,9 +104,9 @@ static void arc_dma_free(struct device *dev, size_t size, void *vaddr,
 	__free_pages(page, get_order(size));
 }
 
-static int arc_dma_mmap(struct device *dev, struct vm_area_struct *vma,
-			void *cpu_addr, dma_addr_t dma_addr, size_t size,
-			unsigned long attrs)
+int arch_dma_mmap(struct device *dev, struct vm_area_struct *vma,
+		void *cpu_addr, dma_addr_t dma_addr, size_t size,
+		unsigned long attrs)
 {
 	unsigned long user_count = vma_pages(vma);
 	unsigned long count = PAGE_ALIGN(size) >> PAGE_SHIFT;
@@ -135,7 +134,7 @@ static int arc_dma_mmap(struct device *dev, struct vm_area_struct *vma,
  * CPU accesses page via normal paddr, thus needs to explicitly made
  * consistent before each use
  */
-static void _dma_cache_sync(phys_addr_t paddr, size_t size,
+static void _dma_cache_sync(struct device *dev, phys_addr_t paddr, size_t size,
 		enum dma_data_direction dir)
 {
 	switch (dir) {
@@ -153,126 +152,14 @@ static void _dma_cache_sync(phys_addr_t paddr, size_t size,
 	}
 }
 
-/*
- * arc_dma_map_page - map a portion of a page for streaming DMA
- *
- * Ensure that any data held in the cache is appropriately discarded
- * or written back.
- *
- * The device owns this memory once this call has completed.  The CPU
- * can regain ownership by calling dma_unmap_page().
- *
- * Note: while it takes struct page as arg, caller can "abuse" it to pass
- * a region larger than PAGE_SIZE, provided it is physically contiguous
- * and this still works correctly
- */
-static dma_addr_t arc_dma_map_page(struct device *dev, struct page *page,
-		unsigned long offset, size_t size, enum dma_data_direction dir,
-		unsigned long attrs)
-{
-	phys_addr_t paddr = page_to_phys(page) + offset;
-
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		_dma_cache_sync(paddr, size, dir);
-
-	return paddr;
-}
-
-/*
- * arc_dma_unmap_page - unmap a buffer previously mapped through dma_map_page()
- *
- * After this call, reads by the CPU to the buffer are guaranteed to see
- * whatever the device wrote there.
- *
- * Note: historically this routine was not implemented for ARC
- */
-static void arc_dma_unmap_page(struct device *dev, dma_addr_t handle,
-			       size_t size, enum dma_data_direction dir,
-			       unsigned long attrs)
-{
-	phys_addr_t paddr = handle;
-
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		_dma_cache_sync(paddr, size, dir);
-}
-
-static int arc_dma_map_sg(struct device *dev, struct scatterlist *sg,
-	   int nents, enum dma_data_direction dir, unsigned long attrs)
-{
-	struct scatterlist *s;
-	int i;
-
-	for_each_sg(sg, s, nents, i)
-		s->dma_address = dma_map_page(dev, sg_page(s), s->offset,
-					       s->length, dir);
-
-	return nents;
-}
-
-static void arc_dma_unmap_sg(struct device *dev, struct scatterlist *sg,
-			     int nents, enum dma_data_direction dir,
-			     unsigned long attrs)
-{
-	struct scatterlist *s;
-	int i;
-
-	for_each_sg(sg, s, nents, i)
-		arc_dma_unmap_page(dev, sg_dma_address(s), sg_dma_len(s), dir,
-				   attrs);
-}
-
-static void arc_dma_sync_single_for_cpu(struct device *dev,
-		dma_addr_t dma_handle, size_t size, enum dma_data_direction dir)
-{
-	_dma_cache_sync(dma_handle, size, DMA_FROM_DEVICE);
-}
-
-static void arc_dma_sync_single_for_device(struct device *dev,
-		dma_addr_t dma_handle, size_t size, enum dma_data_direction dir)
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-	_dma_cache_sync(dma_handle, size, DMA_TO_DEVICE);
+	return _dma_cache_sync(dev, paddr, size, dir);
 }
 
-static void arc_dma_sync_sg_for_cpu(struct device *dev,
-		struct scatterlist *sglist, int nelems,
-		enum dma_data_direction dir)
+void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-	int i;
-	struct scatterlist *sg;
-
-	for_each_sg(sglist, sg, nelems, i)
-		_dma_cache_sync(sg_phys(sg), sg->length, dir);
+	return _dma_cache_sync(dev, paddr, size, dir);
 }
-
-static void arc_dma_sync_sg_for_device(struct device *dev,
-		struct scatterlist *sglist, int nelems,
-		enum dma_data_direction dir)
-{
-	int i;
-	struct scatterlist *sg;
-
-	for_each_sg(sglist, sg, nelems, i)
-		_dma_cache_sync(sg_phys(sg), sg->length, dir);
-}
-
-static int arc_dma_supported(struct device *dev, u64 dma_mask)
-{
-	/* Support 32 bit DMA mask exclusively */
-	return dma_mask == DMA_BIT_MASK(32);
-}
-
-const struct dma_map_ops arc_dma_ops = {
-	.alloc			= arc_dma_alloc,
-	.free			= arc_dma_free,
-	.mmap			= arc_dma_mmap,
-	.map_page		= arc_dma_map_page,
-	.unmap_page		= arc_dma_unmap_page,
-	.map_sg			= arc_dma_map_sg,
-	.unmap_sg		= arc_dma_unmap_sg,
-	.sync_single_for_device	= arc_dma_sync_single_for_device,
-	.sync_single_for_cpu	= arc_dma_sync_single_for_cpu,
-	.sync_sg_for_cpu	= arc_dma_sync_sg_for_cpu,
-	.sync_sg_for_device	= arc_dma_sync_sg_for_device,
-	.dma_supported		= arc_dma_supported,
-};
-EXPORT_SYMBOL(arc_dma_ops);
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 04/20] arm-nommu: use generic dma_noncoherent_ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (2 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 03/20] arc: use generic dma_noncoherent_ops Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  9:11   ` Russell King - ARM Linux
  2018-05-11 13:56   ` John Garry
  2018-05-11  7:59 ` [PATCH 05/20] c6x: " Christoph Hellwig
                   ` (16 subsequent siblings)
  20 siblings, 2 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Switch to the generic noncoherent direct mapping implementation for
the nommu dma map implementation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/arc/Kconfig                |   1 +
 arch/arm/Kconfig                |   4 +
 arch/arm/mm/dma-mapping-nommu.c | 139 +++++---------------------------
 3 files changed, 23 insertions(+), 121 deletions(-)

diff --git a/arch/arc/Kconfig b/arch/arc/Kconfig
index 89d47eac18b2..3a492a9aeaad 100644
--- a/arch/arc/Kconfig
+++ b/arch/arc/Kconfig
@@ -9,6 +9,7 @@
 config ARC
 	def_bool y
 	select ARC_TIMERS
+	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
 	select ARCH_HAS_SYNC_DMA_FOR_CPU
 	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
 	select ARCH_HAS_SG_CHAIN
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index c43f5bb55ac8..76ddd0064f87 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -12,6 +12,8 @@ config ARM
 	select ARCH_HAS_PHYS_TO_DMA
 	select ARCH_HAS_STRICT_KERNEL_RWX if MMU && !XIP_KERNEL
 	select ARCH_HAS_STRICT_MODULE_RWX if MMU
+	select ARCH_HAS_SYNC_DMA_FOR_CPU if !MMU
+	select ARCH_HAS_SYNC_DMA_FOR_DEVICE if !MMU
 	select ARCH_HAS_TICK_BROADCAST if GENERIC_CLOCKEVENTS_BROADCAST
 	select ARCH_HAVE_CUSTOM_GPIO_H
 	select ARCH_HAS_GCOV_PROFILE_ALL
@@ -27,6 +29,8 @@ config ARM
 	select CPU_PM if (SUSPEND || CPU_IDLE)
 	select DCACHE_WORD_ACCESS if HAVE_EFFICIENT_UNALIGNED_ACCESS
 	select DMA_DIRECT_OPS if !MMU
+	select DMA_NONCOHERENT_OPS if !MMU
+	select DMA_NONCOHERENT_MMAP if !MMU
 	select EDAC_SUPPORT
 	select EDAC_ATOMIC_SCRUB
 	select GENERIC_ALLOCATOR
diff --git a/arch/arm/mm/dma-mapping-nommu.c b/arch/arm/mm/dma-mapping-nommu.c
index f448a0663b10..a74ed6632982 100644
--- a/arch/arm/mm/dma-mapping-nommu.c
+++ b/arch/arm/mm/dma-mapping-nommu.c
@@ -12,6 +12,7 @@
 #include <linux/export.h>
 #include <linux/mm.h>
 #include <linux/dma-direct.h>
+#include <linux/dma-noncoherent.h>
 #include <linux/scatterlist.h>
 
 #include <asm/cachetype.h>
@@ -26,18 +27,16 @@
  *   - MMU/MPU is off
  *   - cpu is v7m w/o cache support
  *   - device is coherent
- *  otherwise arm_nommu_dma_ops is used.
+ *  otherwise dma_noncoherent_ops is used.
  *
- *  arm_nommu_dma_ops rely on consistent DMA memory (please, refer to
+ *  dma_noncoherent_ops rely on consistent DMA memory (please, refer to
  *  [1] on how to declare such memory).
  *
  *  [1] Documentation/devicetree/bindings/reserved-memory/reserved-memory.txt
  */
 
-static void *arm_nommu_dma_alloc(struct device *dev, size_t size,
-				 dma_addr_t *dma_handle, gfp_t gfp,
-				 unsigned long attrs)
-
+void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
+		gfp_t gfp, unsigned long attrs)
 {
 	void *ret;
 
@@ -65,9 +64,8 @@ static void *arm_nommu_dma_alloc(struct device *dev, size_t size,
 	return ret;
 }
 
-static void arm_nommu_dma_free(struct device *dev, size_t size,
-			       void *cpu_addr, dma_addr_t dma_addr,
-			       unsigned long attrs)
+void arch_dma_free(struct device *dev, size_t size, void *cpu_addr,
+		dma_addr_t dma_addr, unsigned long attrs)
 {
 	if (attrs & DMA_ATTR_NON_CONSISTENT) {
 		dma_direct_free(dev, size, cpu_addr, dma_addr, attrs);
@@ -81,9 +79,9 @@ static void arm_nommu_dma_free(struct device *dev, size_t size,
 	return;
 }
 
-static int arm_nommu_dma_mmap(struct device *dev, struct vm_area_struct *vma,
-			      void *cpu_addr, dma_addr_t dma_addr, size_t size,
-			      unsigned long attrs)
+int arch_dma_mmap(struct device *dev, struct vm_area_struct *vma,
+		void *cpu_addr, dma_addr_t dma_addr, size_t size,
+		unsigned long attrs)
 {
 	int ret;
 
@@ -93,9 +91,8 @@ static int arm_nommu_dma_mmap(struct device *dev, struct vm_area_struct *vma,
 	return dma_common_mmap(dev, vma, cpu_addr, dma_addr, size);
 }
 
-
-static void __dma_page_cpu_to_dev(phys_addr_t paddr, size_t size,
-				  enum dma_data_direction dir)
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
 	dmac_map_area(__va(paddr), size, dir);
 
@@ -105,8 +102,8 @@ static void __dma_page_cpu_to_dev(phys_addr_t paddr, size_t size,
 		outer_clean_range(paddr, paddr + size);
 }
 
-static void __dma_page_dev_to_cpu(phys_addr_t paddr, size_t size,
-				  enum dma_data_direction dir)
+void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
 	if (dir != DMA_TO_DEVICE) {
 		outer_inv_range(paddr, paddr + size);
@@ -114,110 +111,9 @@ static void __dma_page_dev_to_cpu(phys_addr_t paddr, size_t size,
 	}
 }
 
-static dma_addr_t arm_nommu_dma_map_page(struct device *dev, struct page *page,
-					 unsigned long offset, size_t size,
-					 enum dma_data_direction dir,
-					 unsigned long attrs)
-{
-	dma_addr_t handle = page_to_phys(page) + offset;
-
-	__dma_page_cpu_to_dev(handle, size, dir);
-
-	return handle;
-}
-
-static void arm_nommu_dma_unmap_page(struct device *dev, dma_addr_t handle,
-				     size_t size, enum dma_data_direction dir,
-				     unsigned long attrs)
-{
-	__dma_page_dev_to_cpu(handle, size, dir);
-}
-
-
-static int arm_nommu_dma_map_sg(struct device *dev, struct scatterlist *sgl,
-				int nents, enum dma_data_direction dir,
-				unsigned long attrs)
-{
-	int i;
-	struct scatterlist *sg;
-
-	for_each_sg(sgl, sg, nents, i) {
-		sg_dma_address(sg) = sg_phys(sg);
-		sg_dma_len(sg) = sg->length;
-		__dma_page_cpu_to_dev(sg_dma_address(sg), sg_dma_len(sg), dir);
-	}
-
-	return nents;
-}
-
-static void arm_nommu_dma_unmap_sg(struct device *dev, struct scatterlist *sgl,
-				   int nents, enum dma_data_direction dir,
-				   unsigned long attrs)
-{
-	struct scatterlist *sg;
-	int i;
-
-	for_each_sg(sgl, sg, nents, i)
-		__dma_page_dev_to_cpu(sg_dma_address(sg), sg_dma_len(sg), dir);
-}
-
-static void arm_nommu_dma_sync_single_for_device(struct device *dev,
-		dma_addr_t handle, size_t size, enum dma_data_direction dir)
-{
-	__dma_page_cpu_to_dev(handle, size, dir);
-}
-
-static void arm_nommu_dma_sync_single_for_cpu(struct device *dev,
-		dma_addr_t handle, size_t size, enum dma_data_direction dir)
-{
-	__dma_page_cpu_to_dev(handle, size, dir);
-}
-
-static void arm_nommu_dma_sync_sg_for_device(struct device *dev, struct scatterlist *sgl,
-					     int nents, enum dma_data_direction dir)
-{
-	struct scatterlist *sg;
-	int i;
-
-	for_each_sg(sgl, sg, nents, i)
-		__dma_page_cpu_to_dev(sg_dma_address(sg), sg_dma_len(sg), dir);
-}
-
-static void arm_nommu_dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sgl,
-					  int nents, enum dma_data_direction dir)
-{
-	struct scatterlist *sg;
-	int i;
-
-	for_each_sg(sgl, sg, nents, i)
-		__dma_page_dev_to_cpu(sg_dma_address(sg), sg_dma_len(sg), dir);
-}
-
-const struct dma_map_ops arm_nommu_dma_ops = {
-	.alloc			= arm_nommu_dma_alloc,
-	.free			= arm_nommu_dma_free,
-	.mmap			= arm_nommu_dma_mmap,
-	.map_page		= arm_nommu_dma_map_page,
-	.unmap_page		= arm_nommu_dma_unmap_page,
-	.map_sg			= arm_nommu_dma_map_sg,
-	.unmap_sg		= arm_nommu_dma_unmap_sg,
-	.sync_single_for_device	= arm_nommu_dma_sync_single_for_device,
-	.sync_single_for_cpu	= arm_nommu_dma_sync_single_for_cpu,
-	.sync_sg_for_device	= arm_nommu_dma_sync_sg_for_device,
-	.sync_sg_for_cpu	= arm_nommu_dma_sync_sg_for_cpu,
-};
-EXPORT_SYMBOL(arm_nommu_dma_ops);
-
-static const struct dma_map_ops *arm_nommu_get_dma_map_ops(bool coherent)
-{
-	return coherent ? &dma_direct_ops : &arm_nommu_dma_ops;
-}
-
 void arch_setup_dma_ops(struct device *dev, u64 dma_base, u64 size,
 			const struct iommu_ops *iommu, bool coherent)
 {
-	const struct dma_map_ops *dma_ops;
-
 	if (IS_ENABLED(CONFIG_CPU_V7M)) {
 		/*
 		 * Cache support for v7m is optional, so can be treated as
@@ -233,9 +129,10 @@ void arch_setup_dma_ops(struct device *dev, u64 dma_base, u64 size,
 		dev->archdata.dma_coherent = (get_cr() & CR_M) ? coherent : true;
 	}
 
-	dma_ops = arm_nommu_get_dma_map_ops(dev->archdata.dma_coherent);
-
-	set_dma_ops(dev, dma_ops);
+	if (dev->archdata.dma_coherent)
+		set_dma_ops(dev, &dma_direct_ops);
+	else
+		set_dma_ops(dev, &dma_noncoherent_ops);
 }
 
 void arch_teardown_dma_ops(struct device *dev)
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 05/20] c6x: use generic dma_noncoherent_ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (3 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 04/20] arm-nommu: " Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-15  0:25   ` [Linux-c6x-dev] " Mark Salter
  2018-05-11  7:59 ` [PATCH 06/20] hexagon: " Christoph Hellwig
                   ` (15 subsequent siblings)
  20 siblings, 1 reply; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Switch to the generic noncoherent direct mapping implementation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/c6x/Kconfig                   |   3 +
 arch/c6x/include/asm/Kbuild        |   1 +
 arch/c6x/include/asm/dma-mapping.h |  28 ------
 arch/c6x/include/asm/setup.h       |   2 +
 arch/c6x/kernel/Makefile           |   2 +-
 arch/c6x/kernel/dma.c              | 138 -----------------------------
 arch/c6x/mm/dma-coherent.c         |  40 ++++++++-
 7 files changed, 44 insertions(+), 170 deletions(-)
 delete mode 100644 arch/c6x/include/asm/dma-mapping.h
 delete mode 100644 arch/c6x/kernel/dma.c

diff --git a/arch/c6x/Kconfig b/arch/c6x/Kconfig
index 8c088b96e372..bf59855628ac 100644
--- a/arch/c6x/Kconfig
+++ b/arch/c6x/Kconfig
@@ -6,7 +6,10 @@
 
 config C6X
 	def_bool y
+	select ARCH_HAS_SYNC_DMA_FOR_CPU
+	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
 	select CLKDEV_LOOKUP
+	select DMA_NONCOHERENT_OPS
 	select GENERIC_ATOMIC64
 	select GENERIC_IRQ_SHOW
 	select HAVE_ARCH_TRACEHOOK
diff --git a/arch/c6x/include/asm/Kbuild b/arch/c6x/include/asm/Kbuild
index fd4c840de837..434600e47662 100644
--- a/arch/c6x/include/asm/Kbuild
+++ b/arch/c6x/include/asm/Kbuild
@@ -5,6 +5,7 @@ generic-y += current.h
 generic-y += device.h
 generic-y += div64.h
 generic-y += dma.h
+generic-y += dma-mapping.h
 generic-y += emergency-restart.h
 generic-y += exec.h
 generic-y += extable.h
diff --git a/arch/c6x/include/asm/dma-mapping.h b/arch/c6x/include/asm/dma-mapping.h
deleted file mode 100644
index 05daf1038111..000000000000
--- a/arch/c6x/include/asm/dma-mapping.h
+++ /dev/null
@@ -1,28 +0,0 @@
-/*
- *  Port on Texas Instruments TMS320C6x architecture
- *
- *  Copyright (C) 2004, 2009, 2010, 2011 Texas Instruments Incorporated
- *  Author: Aurelien Jacquiot <aurelien.jacquiot@ti.com>
- *
- *  This program is free software; you can redistribute it and/or modify
- *  it under the terms of the GNU General Public License version 2 as
- *  published by the Free Software Foundation.
- *
- */
-#ifndef _ASM_C6X_DMA_MAPPING_H
-#define _ASM_C6X_DMA_MAPPING_H
-
-extern const struct dma_map_ops c6x_dma_ops;
-
-static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
-{
-	return &c6x_dma_ops;
-}
-
-extern void coherent_mem_init(u32 start, u32 size);
-void *c6x_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
-		gfp_t gfp, unsigned long attrs);
-void c6x_dma_free(struct device *dev, size_t size, void *vaddr,
-		dma_addr_t dma_handle, unsigned long attrs);
-
-#endif	/* _ASM_C6X_DMA_MAPPING_H */
diff --git a/arch/c6x/include/asm/setup.h b/arch/c6x/include/asm/setup.h
index 852afb209afb..350f34debb19 100644
--- a/arch/c6x/include/asm/setup.h
+++ b/arch/c6x/include/asm/setup.h
@@ -28,5 +28,7 @@ extern unsigned char c6x_fuse_mac[6];
 extern void machine_init(unsigned long dt_ptr);
 extern void time_init(void);
 
+extern void coherent_mem_init(u32 start, u32 size);
+
 #endif /* !__ASSEMBLY__ */
 #endif /* _ASM_C6X_SETUP_H */
diff --git a/arch/c6x/kernel/Makefile b/arch/c6x/kernel/Makefile
index 02f340d7b8fe..fbe74174de87 100644
--- a/arch/c6x/kernel/Makefile
+++ b/arch/c6x/kernel/Makefile
@@ -8,6 +8,6 @@ extra-y := head.o vmlinux.lds
 obj-y := process.o traps.o irq.o signal.o ptrace.o
 obj-y += setup.o sys_c6x.o time.o devicetree.o
 obj-y += switch_to.o entry.o vectors.o c6x_ksyms.o
-obj-y += soc.o dma.o
+obj-y += soc.o
 
 obj-$(CONFIG_MODULES)           += module.o
diff --git a/arch/c6x/kernel/dma.c b/arch/c6x/kernel/dma.c
deleted file mode 100644
index 31e1a9ec3a9c..000000000000
--- a/arch/c6x/kernel/dma.c
+++ /dev/null
@@ -1,138 +0,0 @@
-/*
- *  Copyright (C) 2011 Texas Instruments Incorporated
- *  Author: Mark Salter <msalter@redhat.com>
- *
- *  This program is free software; you can redistribute it and/or modify
- *  it under the terms of the GNU General Public License version 2 as
- *  published by the Free Software Foundation.
- */
-#include <linux/module.h>
-#include <linux/dma-mapping.h>
-#include <linux/mm.h>
-#include <linux/mm_types.h>
-#include <linux/scatterlist.h>
-
-#include <asm/cacheflush.h>
-
-static void c6x_dma_sync(dma_addr_t handle, size_t size,
-			 enum dma_data_direction dir)
-{
-	unsigned long paddr = handle;
-
-	BUG_ON(!valid_dma_direction(dir));
-
-	switch (dir) {
-	case DMA_FROM_DEVICE:
-		L2_cache_block_invalidate(paddr, paddr + size);
-		break;
-	case DMA_TO_DEVICE:
-		L2_cache_block_writeback(paddr, paddr + size);
-		break;
-	case DMA_BIDIRECTIONAL:
-		L2_cache_block_writeback_invalidate(paddr, paddr + size);
-		break;
-	default:
-		break;
-	}
-}
-
-static dma_addr_t c6x_dma_map_page(struct device *dev, struct page *page,
-		unsigned long offset, size_t size, enum dma_data_direction dir,
-		unsigned long attrs)
-{
-	dma_addr_t handle = virt_to_phys(page_address(page) + offset);
-
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		c6x_dma_sync(handle, size, dir);
-
-	return handle;
-}
-
-static void c6x_dma_unmap_page(struct device *dev, dma_addr_t handle,
-		size_t size, enum dma_data_direction dir, unsigned long attrs)
-{
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		c6x_dma_sync(handle, size, dir);
-}
-
-static int c6x_dma_map_sg(struct device *dev, struct scatterlist *sglist,
-		int nents, enum dma_data_direction dir, unsigned long attrs)
-{
-	struct scatterlist *sg;
-	int i;
-
-	for_each_sg(sglist, sg, nents, i) {
-		sg->dma_address = sg_phys(sg);
-		if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-			c6x_dma_sync(sg->dma_address, sg->length, dir);
-	}
-
-	return nents;
-}
-
-static void c6x_dma_unmap_sg(struct device *dev, struct scatterlist *sglist,
-		  int nents, enum dma_data_direction dir, unsigned long attrs)
-{
-	struct scatterlist *sg;
-	int i;
-
-	if (attrs & DMA_ATTR_SKIP_CPU_SYNC)
-		return;
-
-	for_each_sg(sglist, sg, nents, i)
-		c6x_dma_sync(sg_dma_address(sg), sg->length, dir);
-}
-
-static void c6x_dma_sync_single_for_cpu(struct device *dev, dma_addr_t handle,
-		size_t size, enum dma_data_direction dir)
-{
-	c6x_dma_sync(handle, size, dir);
-
-}
-
-static void c6x_dma_sync_single_for_device(struct device *dev,
-		dma_addr_t handle, size_t size, enum dma_data_direction dir)
-{
-	c6x_dma_sync(handle, size, dir);
-
-}
-
-static void c6x_dma_sync_sg_for_cpu(struct device *dev,
-		struct scatterlist *sglist, int nents,
-		enum dma_data_direction dir)
-{
-	struct scatterlist *sg;
-	int i;
-
-	for_each_sg(sglist, sg, nents, i)
-		c6x_dma_sync_single_for_cpu(dev, sg_dma_address(sg),
-					sg->length, dir);
-
-}
-
-static void c6x_dma_sync_sg_for_device(struct device *dev,
-		struct scatterlist *sglist, int nents,
-		enum dma_data_direction dir)
-{
-	struct scatterlist *sg;
-	int i;
-
-	for_each_sg(sglist, sg, nents, i)
-		c6x_dma_sync_single_for_device(dev, sg_dma_address(sg),
-					   sg->length, dir);
-
-}
-
-const struct dma_map_ops c6x_dma_ops = {
-	.alloc			= c6x_dma_alloc,
-	.free			= c6x_dma_free,
-	.map_page		= c6x_dma_map_page,
-	.unmap_page		= c6x_dma_unmap_page,
-	.map_sg			= c6x_dma_map_sg,
-	.unmap_sg		= c6x_dma_unmap_sg,
-	.sync_single_for_device	= c6x_dma_sync_single_for_device,
-	.sync_single_for_cpu	= c6x_dma_sync_single_for_cpu,
-	.sync_sg_for_device	= c6x_dma_sync_sg_for_device,
-	.sync_sg_for_cpu	= c6x_dma_sync_sg_for_cpu,
-};
-EXPORT_SYMBOL(c6x_dma_ops);
diff --git a/arch/c6x/mm/dma-coherent.c b/arch/c6x/mm/dma-coherent.c
index 95e38ad27c69..d0a8e0c4b27e 100644
--- a/arch/c6x/mm/dma-coherent.c
+++ b/arch/c6x/mm/dma-coherent.c
@@ -19,10 +19,12 @@
 #include <linux/bitops.h>
 #include <linux/module.h>
 #include <linux/interrupt.h>
-#include <linux/dma-mapping.h>
+#include <linux/dma-noncoherent.h>
 #include <linux/memblock.h>
 
+#include <asm/cacheflush.h>
 #include <asm/page.h>
+#include <asm/setup.h>
 
 /*
  * DMA coherent memory management, can be redefined using the memdma=
@@ -73,7 +75,7 @@ static void __free_dma_pages(u32 addr, int order)
  * Allocate DMA coherent memory space and return both the kernel
  * virtual and DMA address for that space.
  */
-void *c6x_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
+void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
 		gfp_t gfp, unsigned long attrs)
 {
 	u32 paddr;
@@ -98,7 +100,7 @@ void *c6x_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
 /*
  * Free DMA coherent memory as defined by the above mapping.
  */
-void c6x_dma_free(struct device *dev, size_t size, void *vaddr,
+void arch_dma_free(struct device *dev, size_t size, void *vaddr,
 		dma_addr_t dma_handle, unsigned long attrs)
 {
 	int order;
@@ -139,3 +141,35 @@ void __init coherent_mem_init(phys_addr_t start, u32 size)
 	dma_bitmap = phys_to_virt(bitmap_phys);
 	memset(dma_bitmap, 0, dma_pages * PAGE_SIZE);
 }
+
+static void c6x_dma_sync(struct device *dev, phys_addr_t paddr, size_t size,
+		enum dma_data_direction dir)
+{
+	BUG_ON(!valid_dma_direction(dir));
+
+	switch (dir) {
+	case DMA_FROM_DEVICE:
+		L2_cache_block_invalidate(paddr, paddr + size);
+		break;
+	case DMA_TO_DEVICE:
+		L2_cache_block_writeback(paddr, paddr + size);
+		break;
+	case DMA_BIDIRECTIONAL:
+		L2_cache_block_writeback_invalidate(paddr, paddr + size);
+		break;
+	default:
+		break;
+	}
+}
+
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
+{
+	return c6x_dma_sync(dev, paddr, size, dir);
+}
+
+void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
+{
+	return c6x_dma_sync(dev, paddr, size, dir);
+}
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 06/20] hexagon: use generic dma_noncoherent_ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (4 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 05/20] c6x: " Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 07/20] m68k: " Christoph Hellwig
                   ` (14 subsequent siblings)
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Switch to the generic noncoherent direct mapping implementation.

This removes the previous sync_single_for_cpu implementation, which looks
bogus given that no syncing is happening in the similar but more
important unmap_single case.

This adds the previously missing sync_sg_for_device implementation that
matches the pre-existing sync_single_for_device implementation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/hexagon/Kconfig                   |   2 +
 arch/hexagon/include/asm/Kbuild        |   1 +
 arch/hexagon/include/asm/dma-mapping.h |  40 -------
 arch/hexagon/kernel/dma.c              | 143 ++-----------------------
 4 files changed, 11 insertions(+), 175 deletions(-)
 delete mode 100644 arch/hexagon/include/asm/dma-mapping.h

diff --git a/arch/hexagon/Kconfig b/arch/hexagon/Kconfig
index 37adb2003033..bcbdcb32935c 100644
--- a/arch/hexagon/Kconfig
+++ b/arch/hexagon/Kconfig
@@ -4,6 +4,7 @@ comment "Linux Kernel Configuration for Hexagon"
 
 config HEXAGON
 	def_bool y
+	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
 	select HAVE_OPROFILE
 	# Other pending projects/to-do items.
 	# select HAVE_REGS_AND_STACK_ACCESS_API
@@ -28,6 +29,7 @@ config HEXAGON
 	select GENERIC_CLOCKEVENTS_BROADCAST
 	select MODULES_USE_ELF_RELA
 	select GENERIC_CPU_DEVICES
+	select DMA_NONCOHERENT_OPS
 	---help---
 	  Qualcomm Hexagon is a processor architecture designed for high
 	  performance and low power across a wide variety of applications.
diff --git a/arch/hexagon/include/asm/Kbuild b/arch/hexagon/include/asm/Kbuild
index e9743f689fb8..843a8086e980 100644
--- a/arch/hexagon/include/asm/Kbuild
+++ b/arch/hexagon/include/asm/Kbuild
@@ -5,6 +5,7 @@ generic-y += bugs.h
 generic-y += current.h
 generic-y += device.h
 generic-y += div64.h
+generic-y += dma-mapping.h
 generic-y += emergency-restart.h
 generic-y += extable.h
 generic-y += fb.h
diff --git a/arch/hexagon/include/asm/dma-mapping.h b/arch/hexagon/include/asm/dma-mapping.h
deleted file mode 100644
index 263f6acbfb0f..000000000000
--- a/arch/hexagon/include/asm/dma-mapping.h
+++ /dev/null
@@ -1,40 +0,0 @@
-/*
- * DMA operations for the Hexagon architecture
- *
- * Copyright (c) 2010-2011, The Linux Foundation. All rights reserved.
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License version 2 and
- * only version 2 as published by the Free Software Foundation.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
- * 02110-1301, USA.
- */
-
-#ifndef _ASM_DMA_MAPPING_H
-#define _ASM_DMA_MAPPING_H
-
-#include <linux/types.h>
-#include <linux/cache.h>
-#include <linux/mm.h>
-#include <linux/scatterlist.h>
-#include <linux/dma-debug.h>
-#include <asm/io.h>
-
-struct device;
-
-extern const struct dma_map_ops *dma_ops;
-
-static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
-{
-	return dma_ops;
-}
-
-#endif
diff --git a/arch/hexagon/kernel/dma.c b/arch/hexagon/kernel/dma.c
index 77459df34e2e..ffc4ae8e126f 100644
--- a/arch/hexagon/kernel/dma.c
+++ b/arch/hexagon/kernel/dma.c
@@ -18,32 +18,19 @@
  * 02110-1301, USA.
  */
 
-#include <linux/dma-mapping.h>
-#include <linux/dma-direct.h>
+#include <linux/dma-noncoherent.h>
 #include <linux/bootmem.h>
 #include <linux/genalloc.h>
-#include <asm/dma-mapping.h>
 #include <linux/module.h>
 #include <asm/page.h>
 
-#define HEXAGON_MAPPING_ERROR	0
-
-const struct dma_map_ops *dma_ops;
-EXPORT_SYMBOL(dma_ops);
-
-static inline void *dma_addr_to_virt(dma_addr_t dma_addr)
-{
-	return phys_to_virt((unsigned long) dma_addr);
-}
-
 static struct gen_pool *coherent_pool;
 
 
 /* Allocates from a pool of uncached memory that was reserved at boot time */
 
-static void *hexagon_dma_alloc_coherent(struct device *dev, size_t size,
-				 dma_addr_t *dma_addr, gfp_t flag,
-				 unsigned long attrs)
+void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *dma_addr,
+		gfp_t flag, unsigned long attrs)
 {
 	void *ret;
 
@@ -75,58 +62,17 @@ static void *hexagon_dma_alloc_coherent(struct device *dev, size_t size,
 	return ret;
 }
 
-static void hexagon_free_coherent(struct device *dev, size_t size, void *vaddr,
-				  dma_addr_t dma_addr, unsigned long attrs)
+void arch_dma_free(struct device *dev, size_t size, void *vaddr,
+		dma_addr_t dma_addr, unsigned long attrs)
 {
 	gen_pool_free(coherent_pool, (unsigned long) vaddr, size);
 }
 
-static int check_addr(const char *name, struct device *hwdev,
-		      dma_addr_t bus, size_t size)
-{
-	if (hwdev && hwdev->dma_mask && !dma_capable(hwdev, bus, size)) {
-		if (*hwdev->dma_mask >= DMA_BIT_MASK(32))
-			printk(KERN_ERR
-				"%s: overflow %Lx+%zu of device mask %Lx\n",
-				name, (long long)bus, size,
-				(long long)*hwdev->dma_mask);
-		return 0;
-	}
-	return 1;
-}
-
-static int hexagon_map_sg(struct device *hwdev, struct scatterlist *sg,
-			  int nents, enum dma_data_direction dir,
-			  unsigned long attrs)
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-	struct scatterlist *s;
-	int i;
-
-	WARN_ON(nents == 0 || sg[0].length == 0);
-
-	for_each_sg(sg, s, nents, i) {
-		s->dma_address = sg_phys(s);
-		if (!check_addr("map_sg", hwdev, s->dma_address, s->length))
-			return 0;
-
-		s->dma_length = s->length;
+	void *addr = phys_to_virt(paddr);
 
-		if (attrs & DMA_ATTR_SKIP_CPU_SYNC)
-			continue;
-
-		flush_dcache_range(dma_addr_to_virt(s->dma_address),
-				   dma_addr_to_virt(s->dma_address + s->length));
-	}
-
-	return nents;
-}
-
-/*
- * address is virtual
- */
-static inline void dma_sync(void *addr, size_t size,
-			    enum dma_data_direction dir)
-{
 	switch (dir) {
 	case DMA_TO_DEVICE:
 		hexagon_clean_dcache_range((unsigned long) addr,
@@ -144,76 +90,3 @@ static inline void dma_sync(void *addr, size_t size,
 		BUG();
 	}
 }
-
-/**
- * hexagon_map_page() - maps an address for device DMA
- * @dev:	pointer to DMA device
- * @page:	pointer to page struct of DMA memory
- * @offset:	offset within page
- * @size:	size of memory to map
- * @dir:	transfer direction
- * @attrs:	pointer to DMA attrs (not used)
- *
- * Called to map a memory address to a DMA address prior
- * to accesses to/from device.
- *
- * We don't particularly have many hoops to jump through
- * so far.  Straight translation between phys and virtual.
- *
- * DMA is not cache coherent so sync is necessary; this
- * seems to be a convenient place to do it.
- *
- */
-static dma_addr_t hexagon_map_page(struct device *dev, struct page *page,
-				   unsigned long offset, size_t size,
-				   enum dma_data_direction dir,
-				   unsigned long attrs)
-{
-	dma_addr_t bus = page_to_phys(page) + offset;
-	WARN_ON(size == 0);
-
-	if (!check_addr("map_single", dev, bus, size))
-		return HEXAGON_MAPPING_ERROR;
-
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		dma_sync(dma_addr_to_virt(bus), size, dir);
-
-	return bus;
-}
-
-static void hexagon_sync_single_for_cpu(struct device *dev,
-					dma_addr_t dma_handle, size_t size,
-					enum dma_data_direction dir)
-{
-	dma_sync(dma_addr_to_virt(dma_handle), size, dir);
-}
-
-static void hexagon_sync_single_for_device(struct device *dev,
-					dma_addr_t dma_handle, size_t size,
-					enum dma_data_direction dir)
-{
-	dma_sync(dma_addr_to_virt(dma_handle), size, dir);
-}
-
-static int hexagon_mapping_error(struct device *dev, dma_addr_t dma_addr)
-{
-	return dma_addr == HEXAGON_MAPPING_ERROR;
-}
-
-const struct dma_map_ops hexagon_dma_ops = {
-	.alloc		= hexagon_dma_alloc_coherent,
-	.free		= hexagon_free_coherent,
-	.map_sg		= hexagon_map_sg,
-	.map_page	= hexagon_map_page,
-	.sync_single_for_cpu = hexagon_sync_single_for_cpu,
-	.sync_single_for_device = hexagon_sync_single_for_device,
-	.mapping_error	= hexagon_mapping_error,
-};
-
-void __init hexagon_dma_init(void)
-{
-	if (dma_ops)
-		return;
-
-	dma_ops = &hexagon_dma_ops;
-}
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 07/20] m68k: use generic dma_noncoherent_ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (5 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 06/20] hexagon: " Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 08/20] microblaze: " Christoph Hellwig
                   ` (13 subsequent siblings)
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Switch to the generic noncoherent direct mapping implementation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/m68k/Kconfig                   |  2 +
 arch/m68k/include/asm/Kbuild        |  1 +
 arch/m68k/include/asm/dma-mapping.h | 12 -----
 arch/m68k/kernel/dma.c              | 68 ++++-------------------------
 4 files changed, 11 insertions(+), 72 deletions(-)
 delete mode 100644 arch/m68k/include/asm/dma-mapping.h

diff --git a/arch/m68k/Kconfig b/arch/m68k/Kconfig
index 785612b576f7..3f61327da2d5 100644
--- a/arch/m68k/Kconfig
+++ b/arch/m68k/Kconfig
@@ -2,6 +2,7 @@
 config M68K
 	bool
 	default y
+	select ARCH_HAS_SYNC_DMA_FOR_DEVICE if HAS_DMA
 	select ARCH_MIGHT_HAVE_PC_PARPORT if ISA
 	select ARCH_NO_COHERENT_DMA_MMAP if !MMU
 	select HAVE_IDE
@@ -24,6 +25,7 @@ config M68K
 	select MODULES_USE_ELF_RELA
 	select OLD_SIGSUSPEND3
 	select OLD_SIGACTION
+	select DMA_NONCOHERENT_OPS if HAS_DMA
 
 config CPU_BIG_ENDIAN
 	def_bool y
diff --git a/arch/m68k/include/asm/Kbuild b/arch/m68k/include/asm/Kbuild
index 88a9d27df1ac..a853c00f1374 100644
--- a/arch/m68k/include/asm/Kbuild
+++ b/arch/m68k/include/asm/Kbuild
@@ -1,5 +1,6 @@
 generic-y += barrier.h
 generic-y += device.h
+generic-y += dma-mapping.h
 generic-y += emergency-restart.h
 generic-y += exec.h
 generic-y += extable.h
diff --git a/arch/m68k/include/asm/dma-mapping.h b/arch/m68k/include/asm/dma-mapping.h
deleted file mode 100644
index e3722ed04fbb..000000000000
--- a/arch/m68k/include/asm/dma-mapping.h
+++ /dev/null
@@ -1,12 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0 */
-#ifndef _M68K_DMA_MAPPING_H
-#define _M68K_DMA_MAPPING_H
-
-extern const struct dma_map_ops m68k_dma_ops;
-
-static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
-{
-        return &m68k_dma_ops;
-}
-
-#endif  /* _M68K_DMA_MAPPING_H */
diff --git a/arch/m68k/kernel/dma.c b/arch/m68k/kernel/dma.c
index c01b9b8f97bf..3d561c577d35 100644
--- a/arch/m68k/kernel/dma.c
+++ b/arch/m68k/kernel/dma.c
@@ -6,7 +6,7 @@
 
 #undef DEBUG
 
-#include <linux/dma-mapping.h>
+#include <linux/dma-noncoherent.h>
 #include <linux/device.h>
 #include <linux/kernel.h>
 #include <linux/scatterlist.h>
@@ -18,7 +18,7 @@
 
 #if defined(CONFIG_MMU) && !defined(CONFIG_COLDFIRE)
 
-static void *m68k_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
+void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
 		gfp_t flag, unsigned long attrs)
 {
 	struct page *page, **map;
@@ -61,7 +61,7 @@ static void *m68k_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
 	return addr;
 }
 
-static void m68k_dma_free(struct device *dev, size_t size, void *addr,
+void arch_dma_free(struct device *dev, size_t size, void *addr,
 		dma_addr_t handle, unsigned long attrs)
 {
 	pr_debug("dma_free_coherent: %p, %x\n", addr, handle);
@@ -72,8 +72,8 @@ static void m68k_dma_free(struct device *dev, size_t size, void *addr,
 
 #include <asm/cacheflush.h>
 
-static void *m68k_dma_alloc(struct device *dev, size_t size,
-		dma_addr_t *dma_handle, gfp_t gfp, unsigned long attrs)
+void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
+		gfp_t gfp, unsigned long attrs)
 {
 	void *ret;
 
@@ -88,7 +88,7 @@ static void *m68k_dma_alloc(struct device *dev, size_t size,
 	return ret;
 }
 
-static void m68k_dma_free(struct device *dev, size_t size, void *vaddr,
+void arch_dma_free(struct device *dev, size_t size, void *vaddr,
 		dma_addr_t dma_handle, unsigned long attrs)
 {
 	free_pages((unsigned long)vaddr, get_order(size));
@@ -96,8 +96,8 @@ static void m68k_dma_free(struct device *dev, size_t size, void *vaddr,
 
 #endif /* CONFIG_MMU && !CONFIG_COLDFIRE */
 
-static void m68k_dma_sync_single_for_device(struct device *dev,
-		dma_addr_t handle, size_t size, enum dma_data_direction dir)
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t handle,
+		size_t size, enum dma_data_direction dir)
 {
 	switch (dir) {
 	case DMA_BIDIRECTIONAL:
@@ -113,55 +113,3 @@ static void m68k_dma_sync_single_for_device(struct device *dev,
 		break;
 	}
 }
-
-static void m68k_dma_sync_sg_for_device(struct device *dev,
-		struct scatterlist *sglist, int nents, enum dma_data_direction dir)
-{
-	int i;
-	struct scatterlist *sg;
-
-	for_each_sg(sglist, sg, nents, i) {
-		dma_sync_single_for_device(dev, sg->dma_address, sg->length,
-					   dir);
-	}
-}
-
-static dma_addr_t m68k_dma_map_page(struct device *dev, struct page *page,
-		unsigned long offset, size_t size, enum dma_data_direction dir,
-		unsigned long attrs)
-{
-	dma_addr_t handle = page_to_phys(page) + offset;
-
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		dma_sync_single_for_device(dev, handle, size, dir);
-
-	return handle;
-}
-
-static int m68k_dma_map_sg(struct device *dev, struct scatterlist *sglist,
-		int nents, enum dma_data_direction dir, unsigned long attrs)
-{
-	int i;
-	struct scatterlist *sg;
-
-	for_each_sg(sglist, sg, nents, i) {
-		sg->dma_address = sg_phys(sg);
-
-		if (attrs & DMA_ATTR_SKIP_CPU_SYNC)
-			continue;
-
-		dma_sync_single_for_device(dev, sg->dma_address, sg->length,
-					   dir);
-	}
-	return nents;
-}
-
-const struct dma_map_ops m68k_dma_ops = {
-	.alloc			= m68k_dma_alloc,
-	.free			= m68k_dma_free,
-	.map_page		= m68k_dma_map_page,
-	.map_sg			= m68k_dma_map_sg,
-	.sync_single_for_device	= m68k_dma_sync_single_for_device,
-	.sync_sg_for_device	= m68k_dma_sync_sg_for_device,
-};
-EXPORT_SYMBOL(m68k_dma_ops);
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 08/20] microblaze: use generic dma_noncoherent_ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (6 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 07/20] m68k: " Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 09/20] microblaze: remove the consistent_sync and consistent_sync_page Christoph Hellwig
                   ` (12 subsequent siblings)
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Switch to the generic noncoherent direct mapping implementation.

This removes the direction-based optimizations in
sync_{single,sg}_for_{cpu,device} which were marked untestested and
do not match the usually very well tested {un,}map_{single,sg}
implementations.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/microblaze/Kconfig                   |   4 +
 arch/microblaze/include/asm/Kbuild        |   1 +
 arch/microblaze/include/asm/dma-mapping.h |  28 -----
 arch/microblaze/include/asm/pgtable.h     |   2 -
 arch/microblaze/kernel/dma.c              | 144 ++--------------------
 arch/microblaze/mm/consistent.c           |   9 +-
 6 files changed, 22 insertions(+), 166 deletions(-)
 delete mode 100644 arch/microblaze/include/asm/dma-mapping.h

diff --git a/arch/microblaze/Kconfig b/arch/microblaze/Kconfig
index d14782100088..848e31a86ba5 100644
--- a/arch/microblaze/Kconfig
+++ b/arch/microblaze/Kconfig
@@ -1,6 +1,8 @@
 config MICROBLAZE
 	def_bool y
 	select ARCH_HAS_GCOV_PROFILE_ALL
+	select ARCH_HAS_SYNC_DMA_FOR_CPU
+	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
 	select ARCH_MIGHT_HAVE_PC_PARPORT
 	select ARCH_NO_COHERENT_DMA_MMAP if !MMU
 	select ARCH_WANT_IPC_PARSE_VERSION
@@ -8,6 +10,8 @@ config MICROBLAZE
 	select TIMER_OF
 	select CLONE_BACKWARDS3
 	select COMMON_CLK
+	select DMA_NONCOHERENT_OPS
+	select DMA_NONCOHERENT_MMAP
 	select GENERIC_ATOMIC64
 	select GENERIC_CLOCKEVENTS
 	select GENERIC_CPU_DEVICES
diff --git a/arch/microblaze/include/asm/Kbuild b/arch/microblaze/include/asm/Kbuild
index 3c80a5a308ed..8d3e71f43a3e 100644
--- a/arch/microblaze/include/asm/Kbuild
+++ b/arch/microblaze/include/asm/Kbuild
@@ -4,6 +4,7 @@ generic-y += bug.h
 generic-y += bugs.h
 generic-y += device.h
 generic-y += div64.h
+generic-y += dma-mapping.h
 generic-y += emergency-restart.h
 generic-y += exec.h
 generic-y += extable.h
diff --git a/arch/microblaze/include/asm/dma-mapping.h b/arch/microblaze/include/asm/dma-mapping.h
deleted file mode 100644
index add50c1373bf..000000000000
--- a/arch/microblaze/include/asm/dma-mapping.h
+++ /dev/null
@@ -1,28 +0,0 @@
-/*
- * Implements the generic device dma API for microblaze and the pci
- *
- * Copyright (C) 2009-2010 Michal Simek <monstr@monstr.eu>
- * Copyright (C) 2009-2010 PetaLogix
- *
- * This file is subject to the terms and conditions of the GNU General
- * Public License. See the file COPYING in the main directory of this
- * archive for more details.
- *
- * This file is base on powerpc and x86 dma-mapping.h versions
- * Copyright (C) 2004 IBM
- */
-
-#ifndef _ASM_MICROBLAZE_DMA_MAPPING_H
-#define _ASM_MICROBLAZE_DMA_MAPPING_H
-
-/*
- * Available generic sets of operations
- */
-extern const struct dma_map_ops dma_nommu_ops;
-
-static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
-{
-	return &dma_nommu_ops;
-}
-
-#endif	/* _ASM_MICROBLAZE_DMA_MAPPING_H */
diff --git a/arch/microblaze/include/asm/pgtable.h b/arch/microblaze/include/asm/pgtable.h
index db8b1fa83452..8a2e654b709f 100644
--- a/arch/microblaze/include/asm/pgtable.h
+++ b/arch/microblaze/include/asm/pgtable.h
@@ -553,8 +553,6 @@ void __init *early_get_page(void);
 
 extern unsigned long ioremap_bot, ioremap_base;
 
-void *consistent_alloc(gfp_t gfp, size_t size, dma_addr_t *dma_handle);
-void consistent_free(size_t size, void *vaddr);
 void consistent_sync(void *vaddr, size_t size, int direction);
 void consistent_sync_page(struct page *page, unsigned long offset,
 	size_t size, int direction);
diff --git a/arch/microblaze/kernel/dma.c b/arch/microblaze/kernel/dma.c
index 3145e7dc8ab1..71032cf64669 100644
--- a/arch/microblaze/kernel/dma.c
+++ b/arch/microblaze/kernel/dma.c
@@ -8,29 +8,15 @@
  */
 
 #include <linux/device.h>
-#include <linux/dma-mapping.h>
+#include <linux/dma-noncoherent.h>
 #include <linux/gfp.h>
 #include <linux/dma-debug.h>
 #include <linux/export.h>
 #include <linux/bug.h>
 #include <asm/cacheflush.h>
 
-static void *dma_nommu_alloc_coherent(struct device *dev, size_t size,
-				       dma_addr_t *dma_handle, gfp_t flag,
-				       unsigned long attrs)
-{
-	return consistent_alloc(flag, size, dma_handle);
-}
-
-static void dma_nommu_free_coherent(struct device *dev, size_t size,
-				     void *vaddr, dma_addr_t dma_handle,
-				     unsigned long attrs)
-{
-	consistent_free(size, vaddr);
-}
-
-static inline void __dma_sync(unsigned long paddr,
-			      size_t size, enum dma_data_direction direction)
+static void __dma_sync(struct device *dev, phys_addr_t paddr, size_t size,
+		enum dma_data_direction direction)
 {
 	switch (direction) {
 	case DMA_TO_DEVICE:
@@ -45,113 +31,21 @@ static inline void __dma_sync(unsigned long paddr,
 	}
 }
 
-static int dma_nommu_map_sg(struct device *dev, struct scatterlist *sgl,
-			     int nents, enum dma_data_direction direction,
-			     unsigned long attrs)
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-	struct scatterlist *sg;
-	int i;
-
-	/* FIXME this part of code is untested */
-	for_each_sg(sgl, sg, nents, i) {
-		sg->dma_address = sg_phys(sg);
-
-		if (attrs & DMA_ATTR_SKIP_CPU_SYNC)
-			continue;
-
-		__dma_sync(sg_phys(sg), sg->length, direction);
-	}
-
-	return nents;
-}
-
-static inline dma_addr_t dma_nommu_map_page(struct device *dev,
-					     struct page *page,
-					     unsigned long offset,
-					     size_t size,
-					     enum dma_data_direction direction,
-					     unsigned long attrs)
-{
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		__dma_sync(page_to_phys(page) + offset, size, direction);
-	return page_to_phys(page) + offset;
+	__dma_sync(dev, paddr, size, dir);
 }
 
-static inline void dma_nommu_unmap_page(struct device *dev,
-					 dma_addr_t dma_address,
-					 size_t size,
-					 enum dma_data_direction direction,
-					 unsigned long attrs)
+void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-/* There is not necessary to do cache cleanup
- *
- * phys_to_virt is here because in __dma_sync_page is __virt_to_phys and
- * dma_address is physical address
- */
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		__dma_sync(dma_address, size, direction);
+	__dma_sync(dev, paddr, size, dir);
 }
 
-static inline void
-dma_nommu_sync_single_for_cpu(struct device *dev,
-			       dma_addr_t dma_handle, size_t size,
-			       enum dma_data_direction direction)
-{
-	/*
-	 * It's pointless to flush the cache as the memory segment
-	 * is given to the CPU
-	 */
-
-	if (direction == DMA_FROM_DEVICE)
-		__dma_sync(dma_handle, size, direction);
-}
-
-static inline void
-dma_nommu_sync_single_for_device(struct device *dev,
-				  dma_addr_t dma_handle, size_t size,
-				  enum dma_data_direction direction)
-{
-	/*
-	 * It's pointless to invalidate the cache if the device isn't
-	 * supposed to write to the relevant region
-	 */
-
-	if (direction == DMA_TO_DEVICE)
-		__dma_sync(dma_handle, size, direction);
-}
-
-static inline void
-dma_nommu_sync_sg_for_cpu(struct device *dev,
-			   struct scatterlist *sgl, int nents,
-			   enum dma_data_direction direction)
-{
-	struct scatterlist *sg;
-	int i;
-
-	/* FIXME this part of code is untested */
-	if (direction == DMA_FROM_DEVICE)
-		for_each_sg(sgl, sg, nents, i)
-			__dma_sync(sg->dma_address, sg->length, direction);
-}
-
-static inline void
-dma_nommu_sync_sg_for_device(struct device *dev,
-			      struct scatterlist *sgl, int nents,
-			      enum dma_data_direction direction)
-{
-	struct scatterlist *sg;
-	int i;
-
-	/* FIXME this part of code is untested */
-	if (direction == DMA_TO_DEVICE)
-		for_each_sg(sgl, sg, nents, i)
-			__dma_sync(sg->dma_address, sg->length, direction);
-}
-
-static
-int dma_nommu_mmap_coherent(struct device *dev, struct vm_area_struct *vma,
-			     void *cpu_addr, dma_addr_t handle, size_t size,
-			     unsigned long attrs)
+int arch_dma_mmap(struct device *dev, struct vm_area_struct *vma,
+		void *cpu_addr, dma_addr_t handle, size_t size,
+		unsigned long attrs)
 {
 #ifdef CONFIG_MMU
 	unsigned long user_count = vma_pages(vma);
@@ -170,17 +64,3 @@ int dma_nommu_mmap_coherent(struct device *dev, struct vm_area_struct *vma,
 	return -ENXIO;
 #endif
 }
-
-const struct dma_map_ops dma_nommu_ops = {
-	.alloc			= dma_nommu_alloc_coherent,
-	.free			= dma_nommu_free_coherent,
-	.mmap			= dma_nommu_mmap_coherent,
-	.map_sg			= dma_nommu_map_sg,
-	.map_page		= dma_nommu_map_page,
-	.unmap_page		= dma_nommu_unmap_page,
-	.sync_single_for_cpu	= dma_nommu_sync_single_for_cpu,
-	.sync_single_for_device	= dma_nommu_sync_single_for_device,
-	.sync_sg_for_cpu	= dma_nommu_sync_sg_for_cpu,
-	.sync_sg_for_device	= dma_nommu_sync_sg_for_device,
-};
-EXPORT_SYMBOL(dma_nommu_ops);
diff --git a/arch/microblaze/mm/consistent.c b/arch/microblaze/mm/consistent.c
index b06c3a7faf20..b9a9c8c3397b 100644
--- a/arch/microblaze/mm/consistent.c
+++ b/arch/microblaze/mm/consistent.c
@@ -33,6 +33,7 @@
 #include <linux/pci.h>
 #include <linux/interrupt.h>
 #include <linux/gfp.h>
+#include <linux/dma-noncoherent.h>
 
 #include <asm/pgalloc.h>
 #include <linux/io.h>
@@ -59,7 +60,8 @@
  * uncached region.  This will no doubt cause big problems if memory allocated
  * here is not also freed properly. -- JW
  */
-void *consistent_alloc(gfp_t gfp, size_t size, dma_addr_t *dma_handle)
+void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
+		gfp_t gfp, unsigned long attrs)
 {
 	unsigned long order, vaddr;
 	void *ret;
@@ -154,7 +156,6 @@ void *consistent_alloc(gfp_t gfp, size_t size, dma_addr_t *dma_handle)
 
 	return ret;
 }
-EXPORT_SYMBOL(consistent_alloc);
 
 #ifdef CONFIG_MMU
 static pte_t *consistent_virt_to_pte(void *vaddr)
@@ -178,7 +179,8 @@ unsigned long consistent_virt_to_pfn(void *vaddr)
 /*
  * free page(s) as defined by the above mapping.
  */
-void consistent_free(size_t size, void *vaddr)
+void arch_dma_free(struct device *dev, size_t size, void *vaddr,
+		dma_addr_t dma_addr, unsigned long attrs)
 {
 	struct page *page;
 
@@ -218,7 +220,6 @@ void consistent_free(size_t size, void *vaddr)
 	flush_tlb_all();
 #endif
 }
-EXPORT_SYMBOL(consistent_free);
 
 /*
  * make an area consistent.
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 09/20] microblaze: remove the consistent_sync and consistent_sync_page
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (7 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 08/20] microblaze: " Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 10/20] nds32: use generic dma_noncoherent_ops Christoph Hellwig
                   ` (11 subsequent siblings)
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Both unused.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/microblaze/include/asm/pgtable.h |  3 --
 arch/microblaze/mm/consistent.c       | 45 ---------------------------
 2 files changed, 48 deletions(-)

diff --git a/arch/microblaze/include/asm/pgtable.h b/arch/microblaze/include/asm/pgtable.h
index 8a2e654b709f..7b650ab14fa0 100644
--- a/arch/microblaze/include/asm/pgtable.h
+++ b/arch/microblaze/include/asm/pgtable.h
@@ -553,9 +553,6 @@ void __init *early_get_page(void);
 
 extern unsigned long ioremap_bot, ioremap_base;
 
-void consistent_sync(void *vaddr, size_t size, int direction);
-void consistent_sync_page(struct page *page, unsigned long offset,
-	size_t size, int direction);
 unsigned long consistent_virt_to_pfn(void *vaddr);
 
 void setup_memory(void);
diff --git a/arch/microblaze/mm/consistent.c b/arch/microblaze/mm/consistent.c
index b9a9c8c3397b..c9a278ac795a 100644
--- a/arch/microblaze/mm/consistent.c
+++ b/arch/microblaze/mm/consistent.c
@@ -220,48 +220,3 @@ void arch_dma_free(struct device *dev, size_t size, void *vaddr,
 	flush_tlb_all();
 #endif
 }
-
-/*
- * make an area consistent.
- */
-void consistent_sync(void *vaddr, size_t size, int direction)
-{
-	unsigned long start;
-	unsigned long end;
-
-	start = (unsigned long)vaddr;
-
-	/* Convert start address back down to unshadowed memory region */
-#ifdef CONFIG_XILINX_UNCACHED_SHADOW
-	start &= ~UNCACHED_SHADOW_MASK;
-#endif
-	end = start + size;
-
-	switch (direction) {
-	case PCI_DMA_NONE:
-		BUG();
-	case PCI_DMA_FROMDEVICE:	/* invalidate only */
-		invalidate_dcache_range(start, end);
-		break;
-	case PCI_DMA_TODEVICE:		/* writeback only */
-		flush_dcache_range(start, end);
-		break;
-	case PCI_DMA_BIDIRECTIONAL:	/* writeback and invalidate */
-		flush_dcache_range(start, end);
-		break;
-	}
-}
-EXPORT_SYMBOL(consistent_sync);
-
-/*
- * consistent_sync_page makes memory consistent. identical
- * to consistent_sync, but takes a struct page instead of a
- * virtual address
- */
-void consistent_sync_page(struct page *page, unsigned long offset,
-	size_t size, int direction)
-{
-	unsigned long start = (unsigned long)page_address(page) + offset;
-	consistent_sync((void *)start, size, direction);
-}
-EXPORT_SYMBOL(consistent_sync_page);
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 10/20] nds32: use generic dma_noncoherent_ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (8 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 09/20] microblaze: remove the consistent_sync and consistent_sync_page Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 11/20] nios2: " Christoph Hellwig
                   ` (10 subsequent siblings)
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Switch to the generic noncoherent direct mapping implementation.

This makes sure the cache_sync routines is called in the unmap_sg
case, to match the unmap_single and sync_{single,sg}_to_cpu cases.

Note that this now always uses page_address, as previously done
in the non-sg cache sync routines.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/nds32/Kconfig                   |   3 +
 arch/nds32/include/asm/Kbuild        |   1 +
 arch/nds32/include/asm/dma-mapping.h |  14 ---
 arch/nds32/kernel/dma.c              | 182 ++++++---------------------
 4 files changed, 39 insertions(+), 161 deletions(-)
 delete mode 100644 arch/nds32/include/asm/dma-mapping.h

diff --git a/arch/nds32/Kconfig b/arch/nds32/Kconfig
index 249f38d3388f..67d0ac0a989c 100644
--- a/arch/nds32/Kconfig
+++ b/arch/nds32/Kconfig
@@ -5,10 +5,13 @@
 
 config NDS32
         def_bool y
+	select ARCH_HAS_SYNC_DMA_FOR_CPU
+	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
 	select ARCH_WANT_FRAME_POINTERS if FTRACE
 	select CLKSRC_MMIO
 	select CLONE_BACKWARDS
 	select COMMON_CLK
+	select DMA_NONCOHERENT_OPS
 	select GENERIC_ATOMIC64
 	select GENERIC_CPU_DEVICES
 	select GENERIC_CLOCKEVENTS
diff --git a/arch/nds32/include/asm/Kbuild b/arch/nds32/include/asm/Kbuild
index 06bdf8167f5a..b3e951f805f8 100644
--- a/arch/nds32/include/asm/Kbuild
+++ b/arch/nds32/include/asm/Kbuild
@@ -13,6 +13,7 @@ generic-y += cputime.h
 generic-y += device.h
 generic-y += div64.h
 generic-y += dma.h
+generic-y += dma-mapping.h
 generic-y += emergency-restart.h
 generic-y += errno.h
 generic-y += exec.h
diff --git a/arch/nds32/include/asm/dma-mapping.h b/arch/nds32/include/asm/dma-mapping.h
deleted file mode 100644
index 2dd47d245c25..000000000000
--- a/arch/nds32/include/asm/dma-mapping.h
+++ /dev/null
@@ -1,14 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0
-// Copyright (C) 2005-2017 Andes Technology Corporation
-
-#ifndef ASMNDS32_DMA_MAPPING_H
-#define ASMNDS32_DMA_MAPPING_H
-
-extern struct dma_map_ops nds32_dma_ops;
-
-static inline struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
-{
-	return &nds32_dma_ops;
-}
-
-#endif
diff --git a/arch/nds32/kernel/dma.c b/arch/nds32/kernel/dma.c
index d291800fc621..48018275e7f4 100644
--- a/arch/nds32/kernel/dma.c
+++ b/arch/nds32/kernel/dma.c
@@ -3,17 +3,14 @@
 
 #include <linux/types.h>
 #include <linux/mm.h>
-#include <linux/export.h>
 #include <linux/string.h>
-#include <linux/scatterlist.h>
-#include <linux/dma-mapping.h>
+#include <linux/dma-noncoherent.h>
 #include <linux/io.h>
 #include <linux/cache.h>
 #include <linux/highmem.h>
 #include <linux/slab.h>
 #include <asm/cacheflush.h>
 #include <asm/tlbflush.h>
-#include <asm/dma-mapping.h>
 #include <asm/proc-fns.h>
 
 /*
@@ -22,11 +19,6 @@
 static pte_t *consistent_pte;
 static DEFINE_RAW_SPINLOCK(consistent_lock);
 
-enum master_type {
-	FOR_CPU = 0,
-	FOR_DEVICE = 1,
-};
-
 /*
  * VM region handling support.
  *
@@ -124,10 +116,8 @@ static struct arch_vm_region *vm_region_find(struct arch_vm_region *head,
 	return c;
 }
 
-/* FIXME: attrs is not used. */
-static void *nds32_dma_alloc_coherent(struct device *dev, size_t size,
-				      dma_addr_t * handle, gfp_t gfp,
-				      unsigned long attrs)
+void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
+		gfp_t gfp, unsigned long attrs)
 {
 	struct page *page;
 	struct arch_vm_region *c;
@@ -232,8 +222,8 @@ static void *nds32_dma_alloc_coherent(struct device *dev, size_t size,
 	return NULL;
 }
 
-static void nds32_dma_free(struct device *dev, size_t size, void *cpu_addr,
-			   dma_addr_t handle, unsigned long attrs)
+void arch_dma_free(struct device *dev, size_t size, void *cpu_addr,
+		dma_addr_t handle, unsigned long attrs)
 {
 	struct arch_vm_region *c;
 	unsigned long flags, addr;
@@ -333,145 +323,43 @@ static int __init consistent_init(void)
 }
 
 core_initcall(consistent_init);
-static void consistent_sync(void *vaddr, size_t size, int direction, int master_type);
-static dma_addr_t nds32_dma_map_page(struct device *dev, struct page *page,
-				     unsigned long offset, size_t size,
-				     enum dma_data_direction dir,
-				     unsigned long attrs)
-{
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		consistent_sync((void *)(page_address(page) + offset), size, dir, FOR_DEVICE);
-	return page_to_phys(page) + offset;
-}
-
-static void nds32_dma_unmap_page(struct device *dev, dma_addr_t handle,
-				 size_t size, enum dma_data_direction dir,
-				 unsigned long attrs)
-{
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		consistent_sync(phys_to_virt(handle), size, dir, FOR_CPU);
-}
 
-/*
- * Make an area consistent for devices.
- */
-static void consistent_sync(void *vaddr, size_t size, int direction, int master_type)
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-	unsigned long start = (unsigned long)vaddr;
-	unsigned long end = start + size;
-
-	if (master_type == FOR_CPU) {
-		switch (direction) {
-		case DMA_TO_DEVICE:
-			break;
-		case DMA_FROM_DEVICE:
-		case DMA_BIDIRECTIONAL:
-			cpu_dma_inval_range(start, end);
-			break;
-		default:
-			BUG();
-		}
-	} else {
-		/* FOR_DEVICE */
-		switch (direction) {
-		case DMA_FROM_DEVICE:
-			break;
-		case DMA_TO_DEVICE:
-		case DMA_BIDIRECTIONAL:
-			cpu_dma_wb_range(start, end);
-			break;
-		default:
-			BUG();
-		}
+	void *addr = kmap_atomic_pfn(PHYS_PFN(paddr));
+	unsigned long start = (unsigned long)addr;
+
+	switch (dir) {
+	case DMA_FROM_DEVICE:
+		break;
+	case DMA_TO_DEVICE:
+	case DMA_BIDIRECTIONAL:
+		cpu_dma_wb_range(start, start + size);
+		break;
+	default:
+		BUG();
 	}
-}
 
-static int nds32_dma_map_sg(struct device *dev, struct scatterlist *sg,
-			    int nents, enum dma_data_direction dir,
-			    unsigned long attrs)
-{
-	int i;
-
-	for (i = 0; i < nents; i++, sg++) {
-		void *virt;
-		unsigned long pfn;
-		struct page *page = sg_page(sg);
-
-		sg->dma_address = sg_phys(sg);
-		pfn = page_to_pfn(page) + sg->offset / PAGE_SIZE;
-		page = pfn_to_page(pfn);
-		if (PageHighMem(page)) {
-			virt = kmap_atomic(page);
-			consistent_sync(virt, sg->length, dir, FOR_CPU);
-			kunmap_atomic(virt);
-		} else {
-			if (sg->offset > PAGE_SIZE)
-				panic("sg->offset:%08x > PAGE_SIZE\n",
-				      sg->offset);
-			virt = page_address(page) + sg->offset;
-			consistent_sync(virt, sg->length, dir, FOR_CPU);
-		}
-	}
-	return nents;
-}
-
-static void nds32_dma_unmap_sg(struct device *dev, struct scatterlist *sg,
-			       int nhwentries, enum dma_data_direction dir,
-			       unsigned long attrs)
-{
+	kunmap_atomic(addr);
 }
 
-static void
-nds32_dma_sync_single_for_cpu(struct device *dev, dma_addr_t handle,
-			      size_t size, enum dma_data_direction dir)
+void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-	consistent_sync((void *)phys_to_virt(handle), size, dir, FOR_CPU);
-}
-
-static void
-nds32_dma_sync_single_for_device(struct device *dev, dma_addr_t handle,
-				 size_t size, enum dma_data_direction dir)
-{
-	consistent_sync((void *)phys_to_virt(handle), size, dir, FOR_DEVICE);
-}
-
-static void
-nds32_dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg, int nents,
-			  enum dma_data_direction dir)
-{
-	int i;
-
-	for (i = 0; i < nents; i++, sg++) {
-		char *virt =
-		    page_address((struct page *)sg->page_link) + sg->offset;
-		consistent_sync(virt, sg->length, dir, FOR_CPU);
+	void *addr = kmap_atomic_pfn(PHYS_PFN(paddr));
+	unsigned long start = (unsigned long)addr;
+
+	switch (dir) {
+	case DMA_TO_DEVICE:
+		break;
+	case DMA_FROM_DEVICE:
+	case DMA_BIDIRECTIONAL:
+		cpu_dma_inval_range(start, start + size);
+		break;
+	default:
+		BUG();
 	}
-}
-
-static void
-nds32_dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
-			     int nents, enum dma_data_direction dir)
-{
-	int i;
 
-	for (i = 0; i < nents; i++, sg++) {
-		char *virt =
-		    page_address((struct page *)sg->page_link) + sg->offset;
-		consistent_sync(virt, sg->length, dir, FOR_DEVICE);
-	}
+	kunmap_atomic(addr);
 }
-
-struct dma_map_ops nds32_dma_ops = {
-	.alloc = nds32_dma_alloc_coherent,
-	.free = nds32_dma_free,
-	.map_page = nds32_dma_map_page,
-	.unmap_page = nds32_dma_unmap_page,
-	.map_sg = nds32_dma_map_sg,
-	.unmap_sg = nds32_dma_unmap_sg,
-	.sync_single_for_device = nds32_dma_sync_single_for_device,
-	.sync_single_for_cpu = nds32_dma_sync_single_for_cpu,
-	.sync_sg_for_cpu = nds32_dma_sync_sg_for_cpu,
-	.sync_sg_for_device = nds32_dma_sync_sg_for_device,
-};
-
-EXPORT_SYMBOL(nds32_dma_ops);
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 11/20] nios2: use generic dma_noncoherent_ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (9 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 10/20] nds32: use generic dma_noncoherent_ops Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 12/20] openrisc: " Christoph Hellwig
                   ` (9 subsequent siblings)
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Switch to the generic noncoherent direct mapping implementation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/nios2/Kconfig                   |   3 +
 arch/nios2/include/asm/Kbuild        |   1 +
 arch/nios2/include/asm/dma-mapping.h |  20 ----
 arch/nios2/mm/dma-mapping.c          | 139 +++------------------------
 4 files changed, 17 insertions(+), 146 deletions(-)
 delete mode 100644 arch/nios2/include/asm/dma-mapping.h

diff --git a/arch/nios2/Kconfig b/arch/nios2/Kconfig
index 3d4ec88f1db1..92035042cf62 100644
--- a/arch/nios2/Kconfig
+++ b/arch/nios2/Kconfig
@@ -1,6 +1,9 @@
 # SPDX-License-Identifier: GPL-2.0
 config NIOS2
 	def_bool y
+	select ARCH_HAS_SYNC_DMA_FOR_CPU
+	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
+	select DMA_NONCOHERENT_OPS
 	select TIMER_OF
 	select GENERIC_ATOMIC64
 	select GENERIC_CLOCKEVENTS
diff --git a/arch/nios2/include/asm/Kbuild b/arch/nios2/include/asm/Kbuild
index d232da2cbb38..24f6ee1ee69b 100644
--- a/arch/nios2/include/asm/Kbuild
+++ b/arch/nios2/include/asm/Kbuild
@@ -8,6 +8,7 @@ generic-y += current.h
 generic-y += device.h
 generic-y += div64.h
 generic-y += dma.h
+generic-y += dma-mapping.h
 generic-y += emergency-restart.h
 generic-y += exec.h
 generic-y += extable.h
diff --git a/arch/nios2/include/asm/dma-mapping.h b/arch/nios2/include/asm/dma-mapping.h
deleted file mode 100644
index 6ceb92251da0..000000000000
--- a/arch/nios2/include/asm/dma-mapping.h
+++ /dev/null
@@ -1,20 +0,0 @@
-/*
- * Copyright (C) 2011 Tobias Klauser <tklauser@distanz.ch>
- * Copyright (C) 2009 Wind River Systems Inc
- *
- * This file is subject to the terms and conditions of the GNU General
- * Public License.  See the file COPYING in the main directory of this
- * archive for more details.
- */
-
-#ifndef _ASM_NIOS2_DMA_MAPPING_H
-#define _ASM_NIOS2_DMA_MAPPING_H
-
-extern const struct dma_map_ops nios2_dma_ops;
-
-static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
-{
-	return &nios2_dma_ops;
-}
-
-#endif /* _ASM_NIOS2_DMA_MAPPING_H */
diff --git a/arch/nios2/mm/dma-mapping.c b/arch/nios2/mm/dma-mapping.c
index 4be815519dd4..4af9e5b5ba1c 100644
--- a/arch/nios2/mm/dma-mapping.c
+++ b/arch/nios2/mm/dma-mapping.c
@@ -12,18 +12,18 @@
 
 #include <linux/types.h>
 #include <linux/mm.h>
-#include <linux/export.h>
 #include <linux/string.h>
-#include <linux/scatterlist.h>
 #include <linux/dma-mapping.h>
 #include <linux/io.h>
 #include <linux/cache.h>
 #include <asm/cacheflush.h>
 
-static inline void __dma_sync_for_device(void *vaddr, size_t size,
-			      enum dma_data_direction direction)
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-	switch (direction) {
+	void *vaddr = phys_to_virt(paddr);
+
+	switch (dir) {
 	case DMA_FROM_DEVICE:
 		invalidate_dcache_range((unsigned long)vaddr,
 			(unsigned long)(vaddr + size));
@@ -42,10 +42,12 @@ static inline void __dma_sync_for_device(void *vaddr, size_t size,
 	}
 }
 
-static inline void __dma_sync_for_cpu(void *vaddr, size_t size,
-			      enum dma_data_direction direction)
+void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-	switch (direction) {
+	void *vaddr = phys_to_virt(paddr);
+
+	switch (dir) {
 	case DMA_BIDIRECTIONAL:
 	case DMA_FROM_DEVICE:
 		invalidate_dcache_range((unsigned long)vaddr,
@@ -58,8 +60,8 @@ static inline void __dma_sync_for_cpu(void *vaddr, size_t size,
 	}
 }
 
-static void *nios2_dma_alloc(struct device *dev, size_t size,
-		dma_addr_t *dma_handle, gfp_t gfp, unsigned long attrs)
+void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
+		gfp_t gfp, unsigned long attrs)
 {
 	void *ret;
 
@@ -80,125 +82,10 @@ static void *nios2_dma_alloc(struct device *dev, size_t size,
 	return ret;
 }
 
-static void nios2_dma_free(struct device *dev, size_t size, void *vaddr,
+void arch_dma_free(struct device *dev, size_t size, void *vaddr,
 		dma_addr_t dma_handle, unsigned long attrs)
 {
 	unsigned long addr = (unsigned long) CAC_ADDR((unsigned long) vaddr);
 
 	free_pages(addr, get_order(size));
 }
-
-static int nios2_dma_map_sg(struct device *dev, struct scatterlist *sg,
-		int nents, enum dma_data_direction direction,
-		unsigned long attrs)
-{
-	int i;
-
-	for_each_sg(sg, sg, nents, i) {
-		void *addr = sg_virt(sg);
-
-		if (!addr)
-			continue;
-
-		sg->dma_address = sg_phys(sg);
-
-		if (attrs & DMA_ATTR_SKIP_CPU_SYNC)
-			continue;
-
-		__dma_sync_for_device(addr, sg->length, direction);
-	}
-
-	return nents;
-}
-
-static dma_addr_t nios2_dma_map_page(struct device *dev, struct page *page,
-			unsigned long offset, size_t size,
-			enum dma_data_direction direction,
-			unsigned long attrs)
-{
-	void *addr = page_address(page) + offset;
-
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		__dma_sync_for_device(addr, size, direction);
-
-	return page_to_phys(page) + offset;
-}
-
-static void nios2_dma_unmap_page(struct device *dev, dma_addr_t dma_address,
-		size_t size, enum dma_data_direction direction,
-		unsigned long attrs)
-{
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		__dma_sync_for_cpu(phys_to_virt(dma_address), size, direction);
-}
-
-static void nios2_dma_unmap_sg(struct device *dev, struct scatterlist *sg,
-		int nhwentries, enum dma_data_direction direction,
-		unsigned long attrs)
-{
-	void *addr;
-	int i;
-
-	if (direction == DMA_TO_DEVICE)
-		return;
-
-	if (attrs & DMA_ATTR_SKIP_CPU_SYNC)
-		return;
-
-	for_each_sg(sg, sg, nhwentries, i) {
-		addr = sg_virt(sg);
-		if (addr)
-			__dma_sync_for_cpu(addr, sg->length, direction);
-	}
-}
-
-static void nios2_dma_sync_single_for_cpu(struct device *dev,
-		dma_addr_t dma_handle, size_t size,
-		enum dma_data_direction direction)
-{
-	__dma_sync_for_cpu(phys_to_virt(dma_handle), size, direction);
-}
-
-static void nios2_dma_sync_single_for_device(struct device *dev,
-		dma_addr_t dma_handle, size_t size,
-		enum dma_data_direction direction)
-{
-	__dma_sync_for_device(phys_to_virt(dma_handle), size, direction);
-}
-
-static void nios2_dma_sync_sg_for_cpu(struct device *dev,
-		struct scatterlist *sg, int nelems,
-		enum dma_data_direction direction)
-{
-	int i;
-
-	/* Make sure that gcc doesn't leave the empty loop body.  */
-	for_each_sg(sg, sg, nelems, i)
-		__dma_sync_for_cpu(sg_virt(sg), sg->length, direction);
-}
-
-static void nios2_dma_sync_sg_for_device(struct device *dev,
-		struct scatterlist *sg, int nelems,
-		enum dma_data_direction direction)
-{
-	int i;
-
-	/* Make sure that gcc doesn't leave the empty loop body.  */
-	for_each_sg(sg, sg, nelems, i)
-		__dma_sync_for_device(sg_virt(sg), sg->length, direction);
-
-}
-
-const struct dma_map_ops nios2_dma_ops = {
-	.alloc			= nios2_dma_alloc,
-	.free			= nios2_dma_free,
-	.map_page		= nios2_dma_map_page,
-	.unmap_page		= nios2_dma_unmap_page,
-	.map_sg			= nios2_dma_map_sg,
-	.unmap_sg		= nios2_dma_unmap_sg,
-	.sync_single_for_device	= nios2_dma_sync_single_for_device,
-	.sync_single_for_cpu	= nios2_dma_sync_single_for_cpu,
-	.sync_sg_for_cpu	= nios2_dma_sync_sg_for_cpu,
-	.sync_sg_for_device	= nios2_dma_sync_sg_for_device,
-};
-EXPORT_SYMBOL(nios2_dma_ops);
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 12/20] openrisc: use generic dma_noncoherent_ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (10 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 11/20] nios2: " Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 13/20] sh: simplify get_arch_dma_ops Christoph Hellwig
                   ` (8 subsequent siblings)
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Switch to the generic noncoherent direct mapping implementation.

Fix sync_single_for_device to do the same cache coherency operations as
the more tested map_single path, as both should transfer ownership to
the device.

Remove the sync_single_for_cpu implementation as no cache coherency
operations are used in the more commonly used unmap_single case, both
of which transfer ownership to the CPU.

Implement the missing sync_sg_for_device operation, matching the cache
coherency operations in sync_single_for_device and map_{single,sg}.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/openrisc/Kconfig                   |   2 +
 arch/openrisc/include/asm/Kbuild        |   1 +
 arch/openrisc/include/asm/dma-mapping.h |  35 --------
 arch/openrisc/kernel/dma.c              | 109 +++---------------------
 4 files changed, 13 insertions(+), 134 deletions(-)
 delete mode 100644 arch/openrisc/include/asm/dma-mapping.h

diff --git a/arch/openrisc/Kconfig b/arch/openrisc/Kconfig
index 9ecad05bfc73..65e3c574c9d3 100644
--- a/arch/openrisc/Kconfig
+++ b/arch/openrisc/Kconfig
@@ -6,6 +6,8 @@
 
 config OPENRISC
 	def_bool y
+	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
+	select DMA_NONCOHERENT_OPS
 	select OF
 	select OF_EARLY_FLATTREE
 	select IRQ_DOMAIN
diff --git a/arch/openrisc/include/asm/Kbuild b/arch/openrisc/include/asm/Kbuild
index f05c722a21f8..e663a996b612 100644
--- a/arch/openrisc/include/asm/Kbuild
+++ b/arch/openrisc/include/asm/Kbuild
@@ -6,6 +6,7 @@ generic-y += current.h
 generic-y += device.h
 generic-y += div64.h
 generic-y += dma.h
+generic-y += dma-mapping.h
 generic-y += emergency-restart.h
 generic-y += exec.h
 generic-y += extable.h
diff --git a/arch/openrisc/include/asm/dma-mapping.h b/arch/openrisc/include/asm/dma-mapping.h
deleted file mode 100644
index e212a1f0b6d2..000000000000
--- a/arch/openrisc/include/asm/dma-mapping.h
+++ /dev/null
@@ -1,35 +0,0 @@
-/*
- * OpenRISC Linux
- *
- * Linux architectural port borrowing liberally from similar works of
- * others.  All original copyrights apply as per the original source
- * declaration.
- *
- * OpenRISC implementation:
- * Copyright (C) 2010-2011 Jonas Bonn <jonas@southpole.se>
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 of the License, or
- * (at your option) any later version.
- */
-
-#ifndef __ASM_OPENRISC_DMA_MAPPING_H
-#define __ASM_OPENRISC_DMA_MAPPING_H
-
-/*
- * See Documentation/DMA-API-HOWTO.txt and
- * Documentation/DMA-API.txt for documentation.
- */
-
-#include <linux/dma-debug.h>
-#include <linux/dma-mapping.h>
-
-extern const struct dma_map_ops or1k_dma_map_ops;
-
-static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
-{
-	return &or1k_dma_map_ops;
-}
-
-#endif	/* __ASM_OPENRISC_DMA_MAPPING_H */
diff --git a/arch/openrisc/kernel/dma.c b/arch/openrisc/kernel/dma.c
index ec7fd45704d2..cce99405edf4 100644
--- a/arch/openrisc/kernel/dma.c
+++ b/arch/openrisc/kernel/dma.c
@@ -19,9 +19,7 @@
  * the only thing implemented properly.  The rest need looking into...
  */
 
-#include <linux/dma-mapping.h>
-#include <linux/dma-debug.h>
-#include <linux/export.h>
+#include <linux/dma-noncoherent.h>
 
 #include <asm/cpuinfo.h>
 #include <asm/spr_defs.h>
@@ -80,10 +78,9 @@ page_clear_nocache(pte_t *pte, unsigned long addr,
  * is being ignored for now; uncached but write-combined memory is a
  * missing feature of the OR1K.
  */
-static void *
-or1k_dma_alloc(struct device *dev, size_t size,
-	       dma_addr_t *dma_handle, gfp_t gfp,
-	       unsigned long attrs)
+void *
+arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
+		gfp_t gfp, unsigned long attrs)
 {
 	unsigned long va;
 	void *page;
@@ -115,9 +112,9 @@ or1k_dma_alloc(struct device *dev, size_t size,
 	return (void *)va;
 }
 
-static void
-or1k_dma_free(struct device *dev, size_t size, void *vaddr,
-	      dma_addr_t dma_handle, unsigned long attrs)
+void
+arch_dma_free(struct device *dev, size_t size, void *vaddr,
+		dma_addr_t dma_handle, unsigned long attrs)
 {
 	unsigned long va = (unsigned long)vaddr;
 	struct mm_walk walk = {
@@ -133,18 +130,11 @@ or1k_dma_free(struct device *dev, size_t size, void *vaddr,
 	free_pages_exact(vaddr, size);
 }
 
-static dma_addr_t
-or1k_map_page(struct device *dev, struct page *page,
-	      unsigned long offset, size_t size,
-	      enum dma_data_direction dir,
-	      unsigned long attrs)
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t addr, size_t size,
+		enum dma_data_direction dir)
 {
-	unsigned long cl;
-	dma_addr_t addr = page_to_phys(page) + offset;
 	struct cpuinfo_or1k *cpuinfo = &cpuinfo_or1k[smp_processor_id()];
-
-	if (attrs & DMA_ATTR_SKIP_CPU_SYNC)
-		return addr;
+	unsigned long cl;
 
 	switch (dir) {
 	case DMA_TO_DEVICE:
@@ -167,83 +157,4 @@ or1k_map_page(struct device *dev, struct page *page,
 		 */
 		break;
 	}
-
-	return addr;
 }
-
-static void
-or1k_unmap_page(struct device *dev, dma_addr_t dma_handle,
-		size_t size, enum dma_data_direction dir,
-		unsigned long attrs)
-{
-	/* Nothing special to do here... */
-}
-
-static int
-or1k_map_sg(struct device *dev, struct scatterlist *sg,
-	    int nents, enum dma_data_direction dir,
-	    unsigned long attrs)
-{
-	struct scatterlist *s;
-	int i;
-
-	for_each_sg(sg, s, nents, i) {
-		s->dma_address = or1k_map_page(dev, sg_page(s), s->offset,
-					       s->length, dir, 0);
-	}
-
-	return nents;
-}
-
-static void
-or1k_unmap_sg(struct device *dev, struct scatterlist *sg,
-	      int nents, enum dma_data_direction dir,
-	      unsigned long attrs)
-{
-	struct scatterlist *s;
-	int i;
-
-	for_each_sg(sg, s, nents, i) {
-		or1k_unmap_page(dev, sg_dma_address(s), sg_dma_len(s), dir, 0);
-	}
-}
-
-static void
-or1k_sync_single_for_cpu(struct device *dev,
-			 dma_addr_t dma_handle, size_t size,
-			 enum dma_data_direction dir)
-{
-	unsigned long cl;
-	dma_addr_t addr = dma_handle;
-	struct cpuinfo_or1k *cpuinfo = &cpuinfo_or1k[smp_processor_id()];
-
-	/* Invalidate the dcache for the requested range */
-	for (cl = addr; cl < addr + size; cl += cpuinfo->dcache_block_size)
-		mtspr(SPR_DCBIR, cl);
-}
-
-static void
-or1k_sync_single_for_device(struct device *dev,
-			    dma_addr_t dma_handle, size_t size,
-			    enum dma_data_direction dir)
-{
-	unsigned long cl;
-	dma_addr_t addr = dma_handle;
-	struct cpuinfo_or1k *cpuinfo = &cpuinfo_or1k[smp_processor_id()];
-
-	/* Flush the dcache for the requested range */
-	for (cl = addr; cl < addr + size; cl += cpuinfo->dcache_block_size)
-		mtspr(SPR_DCBFR, cl);
-}
-
-const struct dma_map_ops or1k_dma_map_ops = {
-	.alloc = or1k_dma_alloc,
-	.free = or1k_dma_free,
-	.map_page = or1k_map_page,
-	.unmap_page = or1k_unmap_page,
-	.map_sg = or1k_map_sg,
-	.unmap_sg = or1k_unmap_sg,
-	.sync_single_for_cpu = or1k_sync_single_for_cpu,
-	.sync_single_for_device = or1k_sync_single_for_device,
-};
-EXPORT_SYMBOL(or1k_dma_map_ops);
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 13/20] sh: simplify get_arch_dma_ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (11 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 12/20] openrisc: " Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 14/20] sh: introduce a sh_cacheop_vaddr helper Christoph Hellwig
                   ` (7 subsequent siblings)
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Remove the indirection through the dma_ops variable, and just return
nommu_dma_ops directly from get_arch_dma_ops.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/sh/include/asm/dma-mapping.h |  5 ++---
 arch/sh/kernel/dma-nommu.c        |  8 +-------
 arch/sh/mm/consistent.c           |  3 ---
 arch/sh/mm/init.c                 | 10 ----------
 4 files changed, 3 insertions(+), 23 deletions(-)

diff --git a/arch/sh/include/asm/dma-mapping.h b/arch/sh/include/asm/dma-mapping.h
index 41167931e5d9..149e71f95be7 100644
--- a/arch/sh/include/asm/dma-mapping.h
+++ b/arch/sh/include/asm/dma-mapping.h
@@ -2,12 +2,11 @@
 #ifndef __ASM_SH_DMA_MAPPING_H
 #define __ASM_SH_DMA_MAPPING_H
 
-extern const struct dma_map_ops *dma_ops;
-extern void no_iommu_init(void);
+extern const struct dma_map_ops nommu_dma_ops;
 
 static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
 {
-	return dma_ops;
+	return &nommu_dma_ops;
 }
 
 extern void *dma_generic_alloc_coherent(struct device *dev, size_t size,
diff --git a/arch/sh/kernel/dma-nommu.c b/arch/sh/kernel/dma-nommu.c
index 3e3a32fc676e..79a9edafa5b0 100644
--- a/arch/sh/kernel/dma-nommu.c
+++ b/arch/sh/kernel/dma-nommu.c
@@ -79,10 +79,4 @@ const struct dma_map_ops nommu_dma_ops = {
 	.sync_sg_for_device	= nommu_sync_sg_for_device,
 #endif
 };
-
-void __init no_iommu_init(void)
-{
-	if (dma_ops)
-		return;
-	dma_ops = &nommu_dma_ops;
-}
+EXPORT_SYMBOL(nommu_dma_ops);
diff --git a/arch/sh/mm/consistent.c b/arch/sh/mm/consistent.c
index 35ea3099a3b6..221832eec33b 100644
--- a/arch/sh/mm/consistent.c
+++ b/arch/sh/mm/consistent.c
@@ -20,9 +20,6 @@
 #include <asm/cacheflush.h>
 #include <asm/addrspace.h>
 
-const struct dma_map_ops *dma_ops;
-EXPORT_SYMBOL(dma_ops);
-
 void *dma_generic_alloc_coherent(struct device *dev, size_t size,
 				 dma_addr_t *dma_handle, gfp_t gfp,
 				 unsigned long attrs)
diff --git a/arch/sh/mm/init.c b/arch/sh/mm/init.c
index ce0bbaa7e404..32e09f03e6bf 100644
--- a/arch/sh/mm/init.c
+++ b/arch/sh/mm/init.c
@@ -395,22 +395,12 @@ void __init paging_init(void)
 	free_area_init_nodes(max_zone_pfns);
 }
 
-/*
- * Early initialization for any I/O MMUs we might have.
- */
-static void __init iommu_init(void)
-{
-	no_iommu_init();
-}
-
 unsigned int mem_init_done = 0;
 
 void __init mem_init(void)
 {
 	pg_data_t *pgdat;
 
-	iommu_init();
-
 	high_memory = NULL;
 	for_each_online_pgdat(pgdat)
 		high_memory = max_t(void *, high_memory,
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 14/20] sh: introduce a sh_cacheop_vaddr helper
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (12 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 13/20] sh: simplify get_arch_dma_ops Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 15/20] sh: use dma_direct_ops for the CONFIG_DMA_COHERENT case Christoph Hellwig
                   ` (6 subsequent siblings)
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

And use it in the maple bus code to avoid a dma API dependency.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/sh/include/asm/cacheflush.h | 7 +++++++
 arch/sh/mm/consistent.c          | 5 +----
 drivers/sh/maple/maple.c         | 7 ++++---
 3 files changed, 12 insertions(+), 7 deletions(-)

diff --git a/arch/sh/include/asm/cacheflush.h b/arch/sh/include/asm/cacheflush.h
index d103ab5a4e4b..b932e42ef028 100644
--- a/arch/sh/include/asm/cacheflush.h
+++ b/arch/sh/include/asm/cacheflush.h
@@ -101,5 +101,12 @@ void kunmap_coherent(void *kvaddr);
 
 void cpu_cache_init(void);
 
+static inline void *sh_cacheop_vaddr(void *vaddr)
+{
+	if (__in_29bit_mode())
+		vaddr = (void *)CAC_ADDR((unsigned long)vaddr);
+	return vaddr;
+}
+
 #endif /* __KERNEL__ */
 #endif /* __ASM_SH_CACHEFLUSH_H */
diff --git a/arch/sh/mm/consistent.c b/arch/sh/mm/consistent.c
index 221832eec33b..f18483494c46 100644
--- a/arch/sh/mm/consistent.c
+++ b/arch/sh/mm/consistent.c
@@ -69,10 +69,7 @@ void dma_generic_free_coherent(struct device *dev, size_t size,
 void sh_sync_dma_for_device(void *vaddr, size_t size,
 		    enum dma_data_direction direction)
 {
-	void *addr;
-
-	addr = __in_29bit_mode() ?
-	       (void *)CAC_ADDR((unsigned long)vaddr) : vaddr;
+	void *addr = sh_cacheop_vaddr(vaddr);
 
 	switch (direction) {
 	case DMA_FROM_DEVICE:		/* invalidate only */
diff --git a/drivers/sh/maple/maple.c b/drivers/sh/maple/maple.c
index 7525039d812c..c9c354bd713a 100644
--- a/drivers/sh/maple/maple.c
+++ b/drivers/sh/maple/maple.c
@@ -300,8 +300,8 @@ static void maple_send(void)
 	mutex_unlock(&maple_wlist_lock);
 	if (maple_packets > 0) {
 		for (i = 0; i < (1 << MAPLE_DMA_PAGES); i++)
-			sh_sync_dma_for_device(maple_sendbuf + i * PAGE_SIZE,
-				       PAGE_SIZE, DMA_BIDIRECTIONAL);
+			__flush_purge_region(maple_sendbuf + i * PAGE_SIZE,
+					PAGE_SIZE);
 	}
 
 finish:
@@ -642,7 +642,8 @@ static void maple_dma_handler(struct work_struct *work)
 		list_for_each_entry_safe(mq, nmq, &maple_sentq, list) {
 			mdev = mq->dev;
 			recvbuf = mq->recvbuf->buf;
-			sh_sync_dma_for_device(recvbuf, 0x400, DMA_FROM_DEVICE);
+			__flush_invalidate_region(sh_cacheop_vaddr(recvbuf),
+					0x400);
 			code = recvbuf[0];
 			kfree(mq->sendbuf);
 			list_del_init(&mq->list);
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 15/20] sh: use dma_direct_ops for the CONFIG_DMA_COHERENT case
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (13 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 14/20] sh: introduce a sh_cacheop_vaddr helper Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 16/20] mm: split arch/sh/mm/consistent.c Christoph Hellwig
                   ` (5 subsequent siblings)
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

This is a slight change in behavior as we avoid the detour through the
virtual mapping for the coherent allocator, but if this CPU really is
coherent that should be the right thing to do.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/sh/Kconfig                   | 1 +
 arch/sh/include/asm/dma-mapping.h | 4 ++++
 arch/sh/kernel/Makefile           | 4 ++--
 arch/sh/kernel/dma-nommu.c        | 4 ----
 4 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/arch/sh/Kconfig b/arch/sh/Kconfig
index 7d521926041e..d0b095323d62 100644
--- a/arch/sh/Kconfig
+++ b/arch/sh/Kconfig
@@ -157,6 +157,7 @@ config SWAP_IO_SPACE
 	bool
 
 config DMA_COHERENT
+	select DMA_DIRECT_OPS
 	bool
 
 config DMA_NONCOHERENT
diff --git a/arch/sh/include/asm/dma-mapping.h b/arch/sh/include/asm/dma-mapping.h
index 149e71f95be7..1ebc6a4eb1c5 100644
--- a/arch/sh/include/asm/dma-mapping.h
+++ b/arch/sh/include/asm/dma-mapping.h
@@ -6,7 +6,11 @@ extern const struct dma_map_ops nommu_dma_ops;
 
 static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
 {
+#ifdef CONFIG_DMA_NONCOHERENT
 	return &nommu_dma_ops;
+#else
+	return &dma_direct_ops;
+#endif
 }
 
 extern void *dma_generic_alloc_coherent(struct device *dev, size_t size,
diff --git a/arch/sh/kernel/Makefile b/arch/sh/kernel/Makefile
index dc80041f7363..cb5f1bfb52de 100644
--- a/arch/sh/kernel/Makefile
+++ b/arch/sh/kernel/Makefile
@@ -12,7 +12,7 @@ endif
 
 CFLAGS_REMOVE_return_address.o = -pg
 
-obj-y	:= debugtraps.o dma-nommu.o dumpstack.o 		\
+obj-y	:= debugtraps.o dumpstack.o 		\
 	   idle.o io.o irq.o irq_$(BITS).o kdebugfs.o			\
 	   machvec.o nmi_debug.o process.o				\
 	   process_$(BITS).o ptrace.o ptrace_$(BITS).o			\
@@ -45,7 +45,7 @@ obj-$(CONFIG_DUMP_CODE)		+= disassemble.o
 obj-$(CONFIG_HIBERNATION)	+= swsusp.o
 obj-$(CONFIG_DWARF_UNWINDER)	+= dwarf.o
 obj-$(CONFIG_PERF_EVENTS)	+= perf_event.o perf_callchain.o
-
+obj-$(CONFIG_DMA_NONCOHERENT)	+= dma-nommu.o
 obj-$(CONFIG_HAVE_HW_BREAKPOINT)		+= hw_breakpoint.o
 
 ccflags-y := -Werror
diff --git a/arch/sh/kernel/dma-nommu.c b/arch/sh/kernel/dma-nommu.c
index 79a9edafa5b0..d8689b1cb743 100644
--- a/arch/sh/kernel/dma-nommu.c
+++ b/arch/sh/kernel/dma-nommu.c
@@ -51,7 +51,6 @@ static int nommu_map_sg(struct device *dev, struct scatterlist *sg,
 	return nents;
 }
 
-#ifdef CONFIG_DMA_NONCOHERENT
 static void nommu_sync_single_for_device(struct device *dev, dma_addr_t addr,
 			      size_t size, enum dma_data_direction dir)
 {
@@ -67,16 +66,13 @@ static void nommu_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
 	for_each_sg(sg, s, nelems, i)
 		sh_sync_dma_for_device(sg_virt(s), s->length, dir);
 }
-#endif
 
 const struct dma_map_ops nommu_dma_ops = {
 	.alloc			= dma_generic_alloc_coherent,
 	.free			= dma_generic_free_coherent,
 	.map_page		= nommu_map_page,
 	.map_sg			= nommu_map_sg,
-#ifdef CONFIG_DMA_NONCOHERENT
 	.sync_single_for_device	= nommu_sync_single_for_device,
 	.sync_sg_for_device	= nommu_sync_sg_for_device,
-#endif
 };
 EXPORT_SYMBOL(nommu_dma_ops);
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 16/20] mm: split arch/sh/mm/consistent.c
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (14 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 15/20] sh: use dma_direct_ops for the CONFIG_DMA_COHERENT case Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 17/20] sh: use generic dma_noncoherent_ops Christoph Hellwig
                   ` (4 subsequent siblings)
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Half of the file just contains platform device memory setup code which
is required for all builds, and half contains helpers for dma coherent
allocation, which is only needed if CONFIG_DMA_NONCOHERENT is enabled.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/sh/kernel/Makefile       |  2 +-
 arch/sh/kernel/dma-coherent.c | 80 +++++++++++++++++++++++++++++++++++
 arch/sh/mm/consistent.c       | 76 ---------------------------------
 3 files changed, 81 insertions(+), 77 deletions(-)
 create mode 100644 arch/sh/kernel/dma-coherent.c

diff --git a/arch/sh/kernel/Makefile b/arch/sh/kernel/Makefile
index cb5f1bfb52de..d5ddb64bfffe 100644
--- a/arch/sh/kernel/Makefile
+++ b/arch/sh/kernel/Makefile
@@ -45,7 +45,7 @@ obj-$(CONFIG_DUMP_CODE)		+= disassemble.o
 obj-$(CONFIG_HIBERNATION)	+= swsusp.o
 obj-$(CONFIG_DWARF_UNWINDER)	+= dwarf.o
 obj-$(CONFIG_PERF_EVENTS)	+= perf_event.o perf_callchain.o
-obj-$(CONFIG_DMA_NONCOHERENT)	+= dma-nommu.o
+obj-$(CONFIG_DMA_NONCOHERENT)	+= dma-nommu.o dma-coherent.o
 obj-$(CONFIG_HAVE_HW_BREAKPOINT)		+= hw_breakpoint.o
 
 ccflags-y := -Werror
diff --git a/arch/sh/kernel/dma-coherent.c b/arch/sh/kernel/dma-coherent.c
new file mode 100644
index 000000000000..4f41e5cd5207
--- /dev/null
+++ b/arch/sh/kernel/dma-coherent.c
@@ -0,0 +1,80 @@
+/*
+ * Copyright (C) 2004 - 2007  Paul Mundt
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ */
+#include <linux/mm.h>
+#include <linux/init.h>
+#include <linux/dma-mapping.h>
+#include <linux/module.h>
+#include <asm/cacheflush.h>
+#include <asm/addrspace.h>
+
+void *dma_generic_alloc_coherent(struct device *dev, size_t size,
+				 dma_addr_t *dma_handle, gfp_t gfp,
+				 unsigned long attrs)
+{
+	void *ret, *ret_nocache;
+	int order = get_order(size);
+
+	gfp |= __GFP_ZERO;
+
+	ret = (void *)__get_free_pages(gfp, order);
+	if (!ret)
+		return NULL;
+
+	/*
+	 * Pages from the page allocator may have data present in
+	 * cache. So flush the cache before using uncached memory.
+	 */
+	sh_sync_dma_for_device(ret, size, DMA_BIDIRECTIONAL);
+
+	ret_nocache = (void __force *)ioremap_nocache(virt_to_phys(ret), size);
+	if (!ret_nocache) {
+		free_pages((unsigned long)ret, order);
+		return NULL;
+	}
+
+	split_page(pfn_to_page(virt_to_phys(ret) >> PAGE_SHIFT), order);
+
+	*dma_handle = virt_to_phys(ret) - PFN_PHYS(dev->dma_pfn_offset);
+
+	return ret_nocache;
+}
+
+void dma_generic_free_coherent(struct device *dev, size_t size,
+			       void *vaddr, dma_addr_t dma_handle,
+			       unsigned long attrs)
+{
+	int order = get_order(size);
+	unsigned long pfn = (dma_handle >> PAGE_SHIFT) + dev->dma_pfn_offset;
+	int k;
+
+	for (k = 0; k < (1 << order); k++)
+		__free_pages(pfn_to_page(pfn + k), 0);
+
+	iounmap(vaddr);
+}
+
+void sh_sync_dma_for_device(void *vaddr, size_t size,
+		    enum dma_data_direction direction)
+{
+	void *addr = sh_cacheop_vaddr(vaddr);
+
+	switch (direction) {
+	case DMA_FROM_DEVICE:		/* invalidate only */
+		__flush_invalidate_region(addr, size);
+		break;
+	case DMA_TO_DEVICE:		/* writeback only */
+		__flush_wback_region(addr, size);
+		break;
+	case DMA_BIDIRECTIONAL:		/* writeback and invalidate */
+		__flush_purge_region(addr, size);
+		break;
+	default:
+		BUG();
+	}
+}
+EXPORT_SYMBOL(sh_sync_dma_for_device);
diff --git a/arch/sh/mm/consistent.c b/arch/sh/mm/consistent.c
index f18483494c46..5da5be74fb6b 100644
--- a/arch/sh/mm/consistent.c
+++ b/arch/sh/mm/consistent.c
@@ -1,10 +1,6 @@
 /*
- * arch/sh/mm/consistent.c
- *
  * Copyright (C) 2004 - 2007  Paul Mundt
  *
- * Declared coherent memory functions based on arch/x86/kernel/pci-dma_32.c
- *
  * This file is subject to the terms and conditions of the GNU General Public
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
@@ -13,79 +9,7 @@
 #include <linux/init.h>
 #include <linux/platform_device.h>
 #include <linux/dma-mapping.h>
-#include <linux/dma-debug.h>
 #include <linux/io.h>
-#include <linux/module.h>
-#include <linux/gfp.h>
-#include <asm/cacheflush.h>
-#include <asm/addrspace.h>
-
-void *dma_generic_alloc_coherent(struct device *dev, size_t size,
-				 dma_addr_t *dma_handle, gfp_t gfp,
-				 unsigned long attrs)
-{
-	void *ret, *ret_nocache;
-	int order = get_order(size);
-
-	gfp |= __GFP_ZERO;
-
-	ret = (void *)__get_free_pages(gfp, order);
-	if (!ret)
-		return NULL;
-
-	/*
-	 * Pages from the page allocator may have data present in
-	 * cache. So flush the cache before using uncached memory.
-	 */
-	sh_sync_dma_for_device(ret, size, DMA_BIDIRECTIONAL);
-
-	ret_nocache = (void __force *)ioremap_nocache(virt_to_phys(ret), size);
-	if (!ret_nocache) {
-		free_pages((unsigned long)ret, order);
-		return NULL;
-	}
-
-	split_page(pfn_to_page(virt_to_phys(ret) >> PAGE_SHIFT), order);
-
-	*dma_handle = virt_to_phys(ret) - PFN_PHYS(dev->dma_pfn_offset);
-
-	return ret_nocache;
-}
-
-void dma_generic_free_coherent(struct device *dev, size_t size,
-			       void *vaddr, dma_addr_t dma_handle,
-			       unsigned long attrs)
-{
-	int order = get_order(size);
-	unsigned long pfn = (dma_handle >> PAGE_SHIFT) + dev->dma_pfn_offset;
-	int k;
-
-	for (k = 0; k < (1 << order); k++)
-		__free_pages(pfn_to_page(pfn + k), 0);
-
-	iounmap(vaddr);
-}
-
-void sh_sync_dma_for_device(void *vaddr, size_t size,
-		    enum dma_data_direction direction)
-{
-	void *addr = sh_cacheop_vaddr(vaddr);
-
-	switch (direction) {
-	case DMA_FROM_DEVICE:		/* invalidate only */
-		__flush_invalidate_region(addr, size);
-		break;
-	case DMA_TO_DEVICE:		/* writeback only */
-		__flush_wback_region(addr, size);
-		break;
-	case DMA_BIDIRECTIONAL:		/* writeback and invalidate */
-		__flush_purge_region(addr, size);
-		break;
-	default:
-		BUG();
-	}
-}
-EXPORT_SYMBOL(sh_sync_dma_for_device);
 
 static int __init memchunk_setup(char *str)
 {
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 17/20] sh: use generic dma_noncoherent_ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (15 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 16/20] mm: split arch/sh/mm/consistent.c Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 18/20] xtensa: " Christoph Hellwig
                   ` (3 subsequent siblings)
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Switch to the generic noncoherent direct mapping implementation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/sh/Kconfig                   |  3 +-
 arch/sh/include/asm/Kbuild        |  1 +
 arch/sh/include/asm/dma-mapping.h | 26 -----------
 arch/sh/kernel/Makefile           |  2 +-
 arch/sh/kernel/dma-coherent.c     | 20 ++++----
 arch/sh/kernel/dma-nommu.c        | 78 -------------------------------
 6 files changed, 13 insertions(+), 117 deletions(-)
 delete mode 100644 arch/sh/include/asm/dma-mapping.h
 delete mode 100644 arch/sh/kernel/dma-nommu.c

diff --git a/arch/sh/Kconfig b/arch/sh/Kconfig
index d0b095323d62..9809e0604af3 100644
--- a/arch/sh/Kconfig
+++ b/arch/sh/Kconfig
@@ -49,7 +49,6 @@ config SUPERH
 	select HAVE_ARCH_AUDITSYSCALL
 	select HAVE_FUTEX_CMPXCHG if FUTEX
 	select HAVE_NMI
-	select NEED_DMA_MAP_STATE
 	select NEED_SG_DMA_LENGTH
 
 	help
@@ -162,6 +161,8 @@ config DMA_COHERENT
 
 config DMA_NONCOHERENT
 	def_bool !DMA_COHERENT
+	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
+	select DMA_NONCOHERENT_OPS
 
 config PGTABLE_LEVELS
 	default 3 if X2TLB
diff --git a/arch/sh/include/asm/Kbuild b/arch/sh/include/asm/Kbuild
index 1efcce74997b..50f7e878ea1b 100644
--- a/arch/sh/include/asm/Kbuild
+++ b/arch/sh/include/asm/Kbuild
@@ -1,6 +1,7 @@
 generic-y += current.h
 generic-y += delay.h
 generic-y += div64.h
+generic-y += dma-mapping.h
 generic-y += emergency-restart.h
 generic-y += exec.h
 generic-y += irq_regs.h
diff --git a/arch/sh/include/asm/dma-mapping.h b/arch/sh/include/asm/dma-mapping.h
deleted file mode 100644
index 1ebc6a4eb1c5..000000000000
--- a/arch/sh/include/asm/dma-mapping.h
+++ /dev/null
@@ -1,26 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0 */
-#ifndef __ASM_SH_DMA_MAPPING_H
-#define __ASM_SH_DMA_MAPPING_H
-
-extern const struct dma_map_ops nommu_dma_ops;
-
-static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
-{
-#ifdef CONFIG_DMA_NONCOHERENT
-	return &nommu_dma_ops;
-#else
-	return &dma_direct_ops;
-#endif
-}
-
-extern void *dma_generic_alloc_coherent(struct device *dev, size_t size,
-					dma_addr_t *dma_addr, gfp_t flag,
-					unsigned long attrs);
-extern void dma_generic_free_coherent(struct device *dev, size_t size,
-				      void *vaddr, dma_addr_t dma_handle,
-				      unsigned long attrs);
-
-void sh_sync_dma_for_device(void *vaddr, size_t size,
-	    enum dma_data_direction dir);
-
-#endif /* __ASM_SH_DMA_MAPPING_H */
diff --git a/arch/sh/kernel/Makefile b/arch/sh/kernel/Makefile
index d5ddb64bfffe..59673f8a3379 100644
--- a/arch/sh/kernel/Makefile
+++ b/arch/sh/kernel/Makefile
@@ -45,7 +45,7 @@ obj-$(CONFIG_DUMP_CODE)		+= disassemble.o
 obj-$(CONFIG_HIBERNATION)	+= swsusp.o
 obj-$(CONFIG_DWARF_UNWINDER)	+= dwarf.o
 obj-$(CONFIG_PERF_EVENTS)	+= perf_event.o perf_callchain.o
-obj-$(CONFIG_DMA_NONCOHERENT)	+= dma-nommu.o dma-coherent.o
+obj-$(CONFIG_DMA_NONCOHERENT)	+= dma-coherent.o
 obj-$(CONFIG_HAVE_HW_BREAKPOINT)		+= hw_breakpoint.o
 
 ccflags-y := -Werror
diff --git a/arch/sh/kernel/dma-coherent.c b/arch/sh/kernel/dma-coherent.c
index 4f41e5cd5207..11b8c4ac6adb 100644
--- a/arch/sh/kernel/dma-coherent.c
+++ b/arch/sh/kernel/dma-coherent.c
@@ -7,14 +7,13 @@
  */
 #include <linux/mm.h>
 #include <linux/init.h>
-#include <linux/dma-mapping.h>
+#include <linux/dma-noncoherent.h>
 #include <linux/module.h>
 #include <asm/cacheflush.h>
 #include <asm/addrspace.h>
 
-void *dma_generic_alloc_coherent(struct device *dev, size_t size,
-				 dma_addr_t *dma_handle, gfp_t gfp,
-				 unsigned long attrs)
+void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
+		gfp_t gfp, unsigned long attrs)
 {
 	void *ret, *ret_nocache;
 	int order = get_order(size);
@@ -44,9 +43,8 @@ void *dma_generic_alloc_coherent(struct device *dev, size_t size,
 	return ret_nocache;
 }
 
-void dma_generic_free_coherent(struct device *dev, size_t size,
-			       void *vaddr, dma_addr_t dma_handle,
-			       unsigned long attrs)
+void arch_dma_free(struct device *dev, size_t size, void *vaddr,
+		dma_addr_t dma_handle, unsigned long attrs)
 {
 	int order = get_order(size);
 	unsigned long pfn = (dma_handle >> PAGE_SHIFT) + dev->dma_pfn_offset;
@@ -58,12 +56,12 @@ void dma_generic_free_coherent(struct device *dev, size_t size,
 	iounmap(vaddr);
 }
 
-void sh_sync_dma_for_device(void *vaddr, size_t size,
-		    enum dma_data_direction direction)
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-	void *addr = sh_cacheop_vaddr(vaddr);
+	void *addr = sh_cacheop_vaddr(phys_to_virt(paddr));
 
-	switch (direction) {
+	switch (dir) {
 	case DMA_FROM_DEVICE:		/* invalidate only */
 		__flush_invalidate_region(addr, size);
 		break;
diff --git a/arch/sh/kernel/dma-nommu.c b/arch/sh/kernel/dma-nommu.c
deleted file mode 100644
index d8689b1cb743..000000000000
--- a/arch/sh/kernel/dma-nommu.c
+++ /dev/null
@@ -1,78 +0,0 @@
-/*
- * DMA mapping support for platforms lacking IOMMUs.
- *
- * Copyright (C) 2009  Paul Mundt
- *
- * This file is subject to the terms and conditions of the GNU General Public
- * License.  See the file "COPYING" in the main directory of this archive
- * for more details.
- */
-#include <linux/dma-mapping.h>
-#include <linux/io.h>
-#include <asm/cacheflush.h>
-
-static dma_addr_t nommu_map_page(struct device *dev, struct page *page,
-				 unsigned long offset, size_t size,
-				 enum dma_data_direction dir,
-				 unsigned long attrs)
-{
-	dma_addr_t addr = page_to_phys(page) + offset
-		- PFN_PHYS(dev->dma_pfn_offset);
-
-	WARN_ON(size == 0);
-
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		sh_sync_dma_for_device(page_address(page) + offset, size, dir);
-
-	return addr;
-}
-
-static int nommu_map_sg(struct device *dev, struct scatterlist *sg,
-			int nents, enum dma_data_direction dir,
-			unsigned long attrs)
-{
-	struct scatterlist *s;
-	int i;
-
-	WARN_ON(nents == 0 || sg[0].length == 0);
-
-	for_each_sg(sg, s, nents, i) {
-		dma_addr_t offset = PFN_PHYS(dev->dma_pfn_offset);
-
-		BUG_ON(!sg_page(s));
-
-		if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-			sh_sync_dma_for_device(sg_virt(s), s->length, dir);
-
-		s->dma_address = sg_phys(s) - offset;
-		s->dma_length = s->length;
-	}
-
-	return nents;
-}
-
-static void nommu_sync_single_for_device(struct device *dev, dma_addr_t addr,
-			      size_t size, enum dma_data_direction dir)
-{
-	sh_sync_dma_for_device(phys_to_virt(addr), size, dir);
-}
-
-static void nommu_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
-			  int nelems, enum dma_data_direction dir)
-{
-	struct scatterlist *s;
-	int i;
-
-	for_each_sg(sg, s, nelems, i)
-		sh_sync_dma_for_device(sg_virt(s), s->length, dir);
-}
-
-const struct dma_map_ops nommu_dma_ops = {
-	.alloc			= dma_generic_alloc_coherent,
-	.free			= dma_generic_free_coherent,
-	.map_page		= nommu_map_page,
-	.map_sg			= nommu_map_sg,
-	.sync_single_for_device	= nommu_sync_single_for_device,
-	.sync_sg_for_device	= nommu_sync_sg_for_device,
-};
-EXPORT_SYMBOL(nommu_dma_ops);
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 18/20] xtensa: use generic dma_noncoherent_ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (16 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 17/20] sh: use generic dma_noncoherent_ops Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 19/20] sparc: " Christoph Hellwig
                   ` (2 subsequent siblings)
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Switch to the generic noncoherent direct mapping implementation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/xtensa/Kconfig                   |   3 +
 arch/xtensa/include/asm/Kbuild        |   1 +
 arch/xtensa/include/asm/dma-mapping.h |  26 ------
 arch/xtensa/kernel/pci-dma.c          | 130 +++-----------------------
 4 files changed, 19 insertions(+), 141 deletions(-)
 delete mode 100644 arch/xtensa/include/asm/dma-mapping.h

diff --git a/arch/xtensa/Kconfig b/arch/xtensa/Kconfig
index 17df332269b2..ef114648e954 100644
--- a/arch/xtensa/Kconfig
+++ b/arch/xtensa/Kconfig
@@ -5,11 +5,14 @@ config ZONE_DMA
 config XTENSA
 	def_bool y
 	select ARCH_NO_COHERENT_DMA_MMAP if !MMU
+	select ARCH_HAS_SYNC_DMA_FOR_CPU
+	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
 	select ARCH_WANT_FRAME_POINTERS
 	select ARCH_WANT_IPC_PARSE_VERSION
 	select BUILDTIME_EXTABLE_SORT
 	select CLONE_BACKWARDS
 	select COMMON_CLK
+	select DMA_NONCOHERENT_OPS
 	select GENERIC_ATOMIC64
 	select GENERIC_CLOCKEVENTS
 	select GENERIC_IRQ_SHOW
diff --git a/arch/xtensa/include/asm/Kbuild b/arch/xtensa/include/asm/Kbuild
index 436b20337168..a8d6cd3bee4b 100644
--- a/arch/xtensa/include/asm/Kbuild
+++ b/arch/xtensa/include/asm/Kbuild
@@ -2,6 +2,7 @@ generic-y += bug.h
 generic-y += device.h
 generic-y += div64.h
 generic-y += dma-contiguous.h
+generic-y += dma-mapping.h
 generic-y += emergency-restart.h
 generic-y += exec.h
 generic-y += extable.h
diff --git a/arch/xtensa/include/asm/dma-mapping.h b/arch/xtensa/include/asm/dma-mapping.h
deleted file mode 100644
index 44098800dad7..000000000000
--- a/arch/xtensa/include/asm/dma-mapping.h
+++ /dev/null
@@ -1,26 +0,0 @@
-/*
- * This file is subject to the terms and conditions of the GNU General Public
- * License.  See the file "COPYING" in the main directory of this archive
- * for more details.
- *
- * Copyright (C) 2003 - 2005 Tensilica Inc.
- * Copyright (C) 2015 Cadence Design Systems Inc.
- */
-
-#ifndef _XTENSA_DMA_MAPPING_H
-#define _XTENSA_DMA_MAPPING_H
-
-#include <asm/cache.h>
-#include <asm/io.h>
-
-#include <linux/mm.h>
-#include <linux/scatterlist.h>
-
-extern const struct dma_map_ops xtensa_dma_map_ops;
-
-static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
-{
-	return &xtensa_dma_map_ops;
-}
-
-#endif	/* _XTENSA_DMA_MAPPING_H */
diff --git a/arch/xtensa/kernel/pci-dma.c b/arch/xtensa/kernel/pci-dma.c
index 392b4a80ebc2..a83d60e92908 100644
--- a/arch/xtensa/kernel/pci-dma.c
+++ b/arch/xtensa/kernel/pci-dma.c
@@ -16,26 +16,24 @@
  */
 
 #include <linux/dma-contiguous.h>
+#include <linux/dma-noncoherent.h>
 #include <linux/dma-direct.h>
 #include <linux/gfp.h>
 #include <linux/highmem.h>
 #include <linux/mm.h>
-#include <linux/module.h>
-#include <linux/pci.h>
-#include <linux/string.h>
 #include <linux/types.h>
 #include <asm/cacheflush.h>
 #include <asm/io.h>
 
-static void do_cache_op(dma_addr_t dma_handle, size_t size,
+static void do_cache_op(phys_addr_t paddr, size_t size,
 			void (*fn)(unsigned long, unsigned long))
 {
-	unsigned long off = dma_handle & (PAGE_SIZE - 1);
-	unsigned long pfn = PFN_DOWN(dma_handle);
+	unsigned long off = paddr & (PAGE_SIZE - 1);
+	unsigned long pfn = PFN_DOWN(paddr);
 	struct page *page = pfn_to_page(pfn);
 
 	if (!PageHighMem(page))
-		fn((unsigned long)bus_to_virt(dma_handle), size);
+		fn((unsigned long)phys_to_virt(paddr), size);
 	else
 		while (size > 0) {
 			size_t sz = min_t(size_t, size, PAGE_SIZE - off);
@@ -49,14 +47,13 @@ static void do_cache_op(dma_addr_t dma_handle, size_t size,
 		}
 }
 
-static void xtensa_sync_single_for_cpu(struct device *dev,
-				       dma_addr_t dma_handle, size_t size,
-				       enum dma_data_direction dir)
+void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
 	switch (dir) {
 	case DMA_BIDIRECTIONAL:
 	case DMA_FROM_DEVICE:
-		do_cache_op(dma_handle, size, __invalidate_dcache_range);
+		do_cache_op(paddr, size, __invalidate_dcache_range);
 		break;
 
 	case DMA_NONE:
@@ -68,15 +65,14 @@ static void xtensa_sync_single_for_cpu(struct device *dev,
 	}
 }
 
-static void xtensa_sync_single_for_device(struct device *dev,
-					  dma_addr_t dma_handle, size_t size,
-					  enum dma_data_direction dir)
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
 	switch (dir) {
 	case DMA_BIDIRECTIONAL:
 	case DMA_TO_DEVICE:
 		if (XCHAL_DCACHE_IS_WRITEBACK)
-			do_cache_op(dma_handle, size, __flush_dcache_range);
+			do_cache_op(paddr, size, __flush_dcache_range);
 		break;
 
 	case DMA_NONE:
@@ -88,40 +84,13 @@ static void xtensa_sync_single_for_device(struct device *dev,
 	}
 }
 
-static void xtensa_sync_sg_for_cpu(struct device *dev,
-				   struct scatterlist *sg, int nents,
-				   enum dma_data_direction dir)
-{
-	struct scatterlist *s;
-	int i;
-
-	for_each_sg(sg, s, nents, i) {
-		xtensa_sync_single_for_cpu(dev, sg_dma_address(s),
-					   sg_dma_len(s), dir);
-	}
-}
-
-static void xtensa_sync_sg_for_device(struct device *dev,
-				      struct scatterlist *sg, int nents,
-				      enum dma_data_direction dir)
-{
-	struct scatterlist *s;
-	int i;
-
-	for_each_sg(sg, s, nents, i) {
-		xtensa_sync_single_for_device(dev, sg_dma_address(s),
-					      sg_dma_len(s), dir);
-	}
-}
-
 /*
  * Note: We assume that the full memory space is always mapped to 'kseg'
  *	 Otherwise we have to use page attributes (not implemented).
  */
 
-static void *xtensa_dma_alloc(struct device *dev, size_t size,
-			      dma_addr_t *handle, gfp_t flag,
-			      unsigned long attrs)
+void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
+		gfp_t flag, unsigned long attrs)
 {
 	unsigned long ret;
 	unsigned long uncached;
@@ -171,8 +140,8 @@ static void *xtensa_dma_alloc(struct device *dev, size_t size,
 	return (void *)uncached;
 }
 
-static void xtensa_dma_free(struct device *dev, size_t size, void *vaddr,
-			    dma_addr_t dma_handle, unsigned long attrs)
+void arch_dma_free(struct device *dev, size_t size, void *vaddr,
+		dma_addr_t dma_handle, unsigned long attrs)
 {
 	unsigned long count = PAGE_ALIGN(size) >> PAGE_SHIFT;
 	unsigned long addr = (unsigned long)vaddr;
@@ -192,72 +161,3 @@ static void xtensa_dma_free(struct device *dev, size_t size, void *vaddr,
 	if (!dma_release_from_contiguous(dev, page, count))
 		__free_pages(page, get_order(size));
 }
-
-static dma_addr_t xtensa_map_page(struct device *dev, struct page *page,
-				  unsigned long offset, size_t size,
-				  enum dma_data_direction dir,
-				  unsigned long attrs)
-{
-	dma_addr_t dma_handle = page_to_phys(page) + offset;
-
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		xtensa_sync_single_for_device(dev, dma_handle, size, dir);
-
-	return dma_handle;
-}
-
-static void xtensa_unmap_page(struct device *dev, dma_addr_t dma_handle,
-			      size_t size, enum dma_data_direction dir,
-			      unsigned long attrs)
-{
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		xtensa_sync_single_for_cpu(dev, dma_handle, size, dir);
-}
-
-static int xtensa_map_sg(struct device *dev, struct scatterlist *sg,
-			 int nents, enum dma_data_direction dir,
-			 unsigned long attrs)
-{
-	struct scatterlist *s;
-	int i;
-
-	for_each_sg(sg, s, nents, i) {
-		s->dma_address = xtensa_map_page(dev, sg_page(s), s->offset,
-						 s->length, dir, attrs);
-	}
-	return nents;
-}
-
-static void xtensa_unmap_sg(struct device *dev,
-			    struct scatterlist *sg, int nents,
-			    enum dma_data_direction dir,
-			    unsigned long attrs)
-{
-	struct scatterlist *s;
-	int i;
-
-	for_each_sg(sg, s, nents, i) {
-		xtensa_unmap_page(dev, sg_dma_address(s),
-				  sg_dma_len(s), dir, attrs);
-	}
-}
-
-int xtensa_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
-{
-	return 0;
-}
-
-const struct dma_map_ops xtensa_dma_map_ops = {
-	.alloc = xtensa_dma_alloc,
-	.free = xtensa_dma_free,
-	.map_page = xtensa_map_page,
-	.unmap_page = xtensa_unmap_page,
-	.map_sg = xtensa_map_sg,
-	.unmap_sg = xtensa_unmap_sg,
-	.sync_single_for_cpu = xtensa_sync_single_for_cpu,
-	.sync_single_for_device = xtensa_sync_single_for_device,
-	.sync_sg_for_cpu = xtensa_sync_sg_for_cpu,
-	.sync_sg_for_device = xtensa_sync_sg_for_device,
-	.mapping_error = xtensa_dma_mapping_error,
-};
-EXPORT_SYMBOL(xtensa_dma_map_ops);
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 19/20] sparc: use generic dma_noncoherent_ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (17 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 18/20] xtensa: " Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-11  7:59 ` [PATCH 20/20] parisc: " Christoph Hellwig
  2018-05-13 13:26 ` common non-cache coherent direct dma mapping ops Helge Deller
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Switch to the generic noncoherent direct mapping implementation.

This removes the previous sync_single_for_device implementation, which
looks bogus given that no syncing is happening in the similar but more
important map_single case.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/sparc/Kconfig                   |   2 +
 arch/sparc/include/asm/dma-mapping.h |   5 +-
 arch/sparc/kernel/ioport.c           | 151 ++-------------------------
 3 files changed, 14 insertions(+), 144 deletions(-)

diff --git a/arch/sparc/Kconfig b/arch/sparc/Kconfig
index 435dbc033afe..0889b4eabf8b 100644
--- a/arch/sparc/Kconfig
+++ b/arch/sparc/Kconfig
@@ -48,6 +48,8 @@ config SPARC
 
 config SPARC32
 	def_bool !64BIT
+	select ARCH_HAS_SYNC_DMA_FOR_CPU
+	select DMA_NONCOHERENT_OPS
 	select GENERIC_ATOMIC64
 	select CLZ_TAB
 	select HAVE_UID16
diff --git a/arch/sparc/include/asm/dma-mapping.h b/arch/sparc/include/asm/dma-mapping.h
index 12ae33daf52f..e17566376934 100644
--- a/arch/sparc/include/asm/dma-mapping.h
+++ b/arch/sparc/include/asm/dma-mapping.h
@@ -7,7 +7,6 @@
 #include <linux/dma-debug.h>
 
 extern const struct dma_map_ops *dma_ops;
-extern const struct dma_map_ops pci32_dma_ops;
 
 extern struct bus_type pci_bus_type;
 
@@ -15,11 +14,11 @@ static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
 {
 #ifdef CONFIG_SPARC_LEON
 	if (sparc_cpu_model == sparc_leon)
-		return &pci32_dma_ops;
+		return &dma_noncoherent_ops;
 #endif
 #if defined(CONFIG_SPARC32) && defined(CONFIG_PCI)
 	if (bus == &pci_bus_type)
-		return &pci32_dma_ops;
+		return &dma_noncoherent_ops;
 #endif
 	return dma_ops;
 }
diff --git a/arch/sparc/kernel/ioport.c b/arch/sparc/kernel/ioport.c
index 3bcef9ce74df..7954512c42e7 100644
--- a/arch/sparc/kernel/ioport.c
+++ b/arch/sparc/kernel/ioport.c
@@ -38,6 +38,7 @@
 #include <linux/proc_fs.h>
 #include <linux/seq_file.h>
 #include <linux/scatterlist.h>
+#include <linux/dma-noncoherent.h>
 #include <linux/of_device.h>
 
 #include <asm/io.h>
@@ -434,9 +435,8 @@ 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.
  */
-static void *pci32_alloc_coherent(struct device *dev, size_t len,
-				  dma_addr_t *pba, gfp_t gfp,
-				  unsigned long attrs)
+void *arch_dma_alloc(struct device *dev, size_t len, dma_addr_t *pba, gfp_t gfp,
+		unsigned long attrs)
 {
 	unsigned long len_total = PAGE_ALIGN(len);
 	void *va;
@@ -488,8 +488,8 @@ static void *pci32_alloc_coherent(struct device *dev, size_t len,
  * References to the memory and mappings associated with cpu_addr/dma_addr
  * past this call are illegal.
  */
-static void pci32_free_coherent(struct device *dev, size_t n, void *p,
-				dma_addr_t ba, unsigned long attrs)
+void arch_dma_free(struct device *dev, size_t n, void *p, dma_addr_t ba,
+		unsigned long attrs)
 {
 	struct resource *res;
 
@@ -519,146 +519,15 @@ static void pci32_free_coherent(struct device *dev, size_t n, void *p,
 	free_pages((unsigned long)phys_to_virt(ba), get_order(n));
 }
 
-/*
- * Same as pci_map_single, but with pages.
- */
-static dma_addr_t pci32_map_page(struct device *dev, struct page *page,
-				 unsigned long offset, size_t size,
-				 enum dma_data_direction dir,
-				 unsigned long attrs)
-{
-	/* IIep is write-through, not flushing. */
-	return page_to_phys(page) + offset;
-}
-
-static void pci32_unmap_page(struct device *dev, dma_addr_t ba, size_t size,
-			     enum dma_data_direction dir, unsigned long attrs)
-{
-	if (dir != PCI_DMA_TODEVICE && !(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		dma_make_coherent(ba, PAGE_ALIGN(size));
-}
-
-/* Map a set of buffers described by scatterlist in streaming
- * mode for DMA.  This is the scatter-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.
- */
-static int pci32_map_sg(struct device *device, struct scatterlist *sgl,
-			int nents, enum dma_data_direction dir,
-			unsigned long attrs)
-{
-	struct scatterlist *sg;
-	int n;
-
-	/* IIep is write-through, not flushing. */
-	for_each_sg(sgl, sg, nents, n) {
-		sg->dma_address = sg_phys(sg);
-		sg->dma_length = sg->length;
-	}
-	return nents;
-}
-
-/* Unmap a set of streaming mode DMA translations.
- * Again, cpu read rules concerning calls here are the same as for
- * pci_unmap_single() above.
- */
-static void pci32_unmap_sg(struct device *dev, struct scatterlist *sgl,
-			   int nents, enum dma_data_direction dir,
-			   unsigned long attrs)
-{
-	struct scatterlist *sg;
-	int n;
-
-	if (dir != PCI_DMA_TODEVICE && !(attrs & DMA_ATTR_SKIP_CPU_SYNC)) {
-		for_each_sg(sgl, sg, nents, n) {
-			dma_make_coherent(sg_phys(sg), PAGE_ALIGN(sg->length));
-		}
-	}
-}
+/* IIep is write-through, not flushing on cpu to device transfer. */
 
-/* Make physical memory consistent for a single
- * streaming mode DMA translation before or 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.
- */
-static void pci32_sync_single_for_cpu(struct device *dev, dma_addr_t ba,
-				      size_t size, enum dma_data_direction dir)
+void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-	if (dir != PCI_DMA_TODEVICE) {
-		dma_make_coherent(ba, PAGE_ALIGN(size));
-	}
-}
-
-static void pci32_sync_single_for_device(struct device *dev, dma_addr_t ba,
-					 size_t size, enum dma_data_direction dir)
-{
-	if (dir != PCI_DMA_TODEVICE) {
-		dma_make_coherent(ba, PAGE_ALIGN(size));
-	}
+	if (dir != PCI_DMA_TODEVICE)
+		dma_make_coherent(paddr, PAGE_ALIGN(size));
 }
 
-/* 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.
- */
-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;
-
-	if (dir != PCI_DMA_TODEVICE) {
-		for_each_sg(sgl, sg, nents, n) {
-			dma_make_coherent(sg_phys(sg), PAGE_ALIGN(sg->length));
-		}
-	}
-}
-
-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;
-
-	if (dir != PCI_DMA_TODEVICE) {
-		for_each_sg(sgl, sg, nents, n) {
-			dma_make_coherent(sg_phys(sg), PAGE_ALIGN(sg->length));
-		}
-	}
-}
-
-/* note: leon re-uses pci32_dma_ops */
-const struct dma_map_ops pci32_dma_ops = {
-	.alloc			= pci32_alloc_coherent,
-	.free			= pci32_free_coherent,
-	.map_page		= pci32_map_page,
-	.unmap_page		= pci32_unmap_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);
-
 const struct dma_map_ops *dma_ops = &sbus_dma_ops;
 EXPORT_SYMBOL(dma_ops);
 
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* [PATCH 20/20] parisc: use generic dma_noncoherent_ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (18 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 19/20] sparc: " Christoph Hellwig
@ 2018-05-11  7:59 ` Christoph Hellwig
  2018-05-13 13:26 ` common non-cache coherent direct dma mapping ops Helge Deller
  20 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-11  7:59 UTC (permalink / raw)
  To: iommu
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-xtensa, linux-kernel

Switch to the generic noncoherent direct mapping implementation.

Parisc previously had two different non-coherent dma ops implementation
that just different in the way coherent allocations were handled or not
handled.  The different behavior is not selected at runtime in the
arch_dma_alloc and arch_dma_free routines.  The non-coherent allocation
in the pcx cases now uses the dma_direct helpers that are a little more
sophisticated and used by a lot of other architectures.

Fix sync_single_for_cpu to do skip the cache flush unless the transfer
is to the device to match the more tested unmap_single path which should
have the same cache coherency implications.

This also now consistenly uses flush_kernel_dcache_range for cache
flushing while previously some of the SG based operations used
flush_kernel_vmap_range instead.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/parisc/Kconfig                   |   4 +
 arch/parisc/include/asm/dma-mapping.h |   5 -
 arch/parisc/kernel/pci-dma.c          | 184 +++-----------------------
 arch/parisc/kernel/setup.c            |   8 +-
 arch/parisc/mm/init.c                 |  11 +-
 5 files changed, 32 insertions(+), 180 deletions(-)

diff --git a/arch/parisc/Kconfig b/arch/parisc/Kconfig
index 4d8f64d48597..4993c6dc8358 100644
--- a/arch/parisc/Kconfig
+++ b/arch/parisc/Kconfig
@@ -188,6 +188,10 @@ config PA20
 config PA11
 	def_bool y
 	depends on PA7000 || PA7100LC || PA7200 || PA7300LC
+	select ARCH_HAS_SYNC_DMA_FOR_CPU
+	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
+	select DMA_NONCOHERENT_OPS
+	select DMA_NONCOHERENT_CACHE_SYNC
 
 config PREFETCH
 	def_bool y
diff --git a/arch/parisc/include/asm/dma-mapping.h b/arch/parisc/include/asm/dma-mapping.h
index 01e1fc057c83..44a9f97194aa 100644
--- a/arch/parisc/include/asm/dma-mapping.h
+++ b/arch/parisc/include/asm/dma-mapping.h
@@ -21,11 +21,6 @@
 ** flush/purge and allocate "regular" cacheable pages for everything.
 */
 
-#ifdef CONFIG_PA11
-extern const struct dma_map_ops pcxl_dma_ops;
-extern const struct dma_map_ops pcx_dma_ops;
-#endif
-
 extern const struct dma_map_ops *hppa_dma_ops;
 
 static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
diff --git a/arch/parisc/kernel/pci-dma.c b/arch/parisc/kernel/pci-dma.c
index 91bc0cac03a1..02de2eb22d85 100644
--- a/arch/parisc/kernel/pci-dma.c
+++ b/arch/parisc/kernel/pci-dma.c
@@ -21,13 +21,12 @@
 #include <linux/init.h>
 #include <linux/gfp.h>
 #include <linux/mm.h>
-#include <linux/pci.h>
 #include <linux/proc_fs.h>
 #include <linux/seq_file.h>
 #include <linux/string.h>
 #include <linux/types.h>
-#include <linux/scatterlist.h>
-#include <linux/export.h>
+#include <linux/dma-direct.h>
+#include <linux/dma-noncoherent.h>
 
 #include <asm/cacheflush.h>
 #include <asm/dma.h>    /* for DMA_CHUNK_SIZE */
@@ -447,178 +446,39 @@ static void pa11_dma_free(struct device *dev, size_t size, void *vaddr,
 	free_pages((unsigned long)__va(dma_handle), order);
 }
 
-static dma_addr_t pa11_dma_map_page(struct device *dev, struct page *page,
-		unsigned long offset, size_t size,
-		enum dma_data_direction direction, unsigned long attrs)
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-	void *addr = page_address(page) + offset;
-	BUG_ON(direction == DMA_NONE);
-
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		flush_kernel_dcache_range((unsigned long) addr, size);
-
-	return virt_to_phys(addr);
-}
-
-static void pa11_dma_unmap_page(struct device *dev, dma_addr_t dma_handle,
-		size_t size, enum dma_data_direction direction,
-		unsigned long attrs)
-{
-	BUG_ON(direction == DMA_NONE);
-
-	if (attrs & DMA_ATTR_SKIP_CPU_SYNC)
-		return;
-
-	if (direction == DMA_TO_DEVICE)
-		return;
-
-	/*
-	 * For PCI_DMA_FROMDEVICE this flush is not necessary for the
-	 * simple map/unmap case. However, it IS necessary if if
-	 * pci_dma_sync_single_* has been called and the buffer reused.
-	 */
-
-	flush_kernel_dcache_range((unsigned long) phys_to_virt(dma_handle), size);
-}
-
-static int pa11_dma_map_sg(struct device *dev, struct scatterlist *sglist,
-		int nents, enum dma_data_direction direction,
-		unsigned long attrs)
-{
-	int i;
-	struct scatterlist *sg;
-
-	BUG_ON(direction == DMA_NONE);
-
-	for_each_sg(sglist, sg, nents, i) {
-		unsigned long vaddr = (unsigned long)sg_virt(sg);
-
-		sg_dma_address(sg) = (dma_addr_t) virt_to_phys(vaddr);
-		sg_dma_len(sg) = sg->length;
-
-		if (attrs & DMA_ATTR_SKIP_CPU_SYNC)
-			continue;
-
-		flush_kernel_dcache_range(vaddr, sg->length);
-	}
-	return nents;
-}
-
-static void pa11_dma_unmap_sg(struct device *dev, struct scatterlist *sglist,
-		int nents, enum dma_data_direction direction,
-		unsigned long attrs)
-{
-	int i;
-	struct scatterlist *sg;
-
-	BUG_ON(direction == DMA_NONE);
-
-	if (attrs & DMA_ATTR_SKIP_CPU_SYNC)
-		return;
-
-	if (direction == DMA_TO_DEVICE)
-		return;
-
-	/* once we do combining we'll need to use phys_to_virt(sg_dma_address(sglist)) */
-
-	for_each_sg(sglist, sg, nents, i)
-		flush_kernel_vmap_range(sg_virt(sg), sg->length);
-}
-
-static void pa11_dma_sync_single_for_cpu(struct device *dev,
-		dma_addr_t dma_handle, size_t size,
-		enum dma_data_direction direction)
-{
-	BUG_ON(direction == DMA_NONE);
-
-	flush_kernel_dcache_range((unsigned long) phys_to_virt(dma_handle),
-			size);
+	flush_kernel_dcache_range((unsigned long)phys_to_virt(paddr), size);
 }
 
-static void pa11_dma_sync_single_for_device(struct device *dev,
-		dma_addr_t dma_handle, size_t size,
-		enum dma_data_direction direction)
+void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-	BUG_ON(direction == DMA_NONE);
-
-	flush_kernel_dcache_range((unsigned long) phys_to_virt(dma_handle),
-			size);
-}
-
-static void pa11_dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sglist, int nents, enum dma_data_direction direction)
-{
-	int i;
-	struct scatterlist *sg;
-
-	/* once we do combining we'll need to use phys_to_virt(sg_dma_address(sglist)) */
-
-	for_each_sg(sglist, sg, nents, i)
-		flush_kernel_vmap_range(sg_virt(sg), sg->length);
+	flush_kernel_dcache_range((unsigned long)phys_to_virt(paddr), size);
 }
 
-static void pa11_dma_sync_sg_for_device(struct device *dev, struct scatterlist *sglist, int nents, enum dma_data_direction direction)
-{
-	int i;
-	struct scatterlist *sg;
-
-	/* once we do combining we'll need to use phys_to_virt(sg_dma_address(sglist)) */
-
-	for_each_sg(sglist, sg, nents, i)
-		flush_kernel_vmap_range(sg_virt(sg), sg->length);
-}
-
-static void pa11_dma_cache_sync(struct device *dev, void *vaddr, size_t size,
+void arch_dma_cache_sync(struct device *dev, void *vaddr, size_t size,
 	       enum dma_data_direction direction)
 {
 	flush_kernel_dcache_range((unsigned long)vaddr, size);
 }
 
-const struct dma_map_ops pcxl_dma_ops = {
-	.alloc =		pa11_dma_alloc,
-	.free =			pa11_dma_free,
-	.map_page =		pa11_dma_map_page,
-	.unmap_page =		pa11_dma_unmap_page,
-	.map_sg =		pa11_dma_map_sg,
-	.unmap_sg =		pa11_dma_unmap_sg,
-	.sync_single_for_cpu =	pa11_dma_sync_single_for_cpu,
-	.sync_single_for_device = pa11_dma_sync_single_for_device,
-	.sync_sg_for_cpu =	pa11_dma_sync_sg_for_cpu,
-	.sync_sg_for_device =	pa11_dma_sync_sg_for_device,
-	.cache_sync =		pa11_dma_cache_sync,
-};
-
-static void *pcx_dma_alloc(struct device *dev, size_t size,
-		dma_addr_t *dma_handle, gfp_t flag, unsigned long attrs)
+void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
+		gfp_t gfp, unsigned long attrs)
 {
-	void *addr;
-
-	if ((attrs & DMA_ATTR_NON_CONSISTENT) == 0)
-		return NULL;
-
-	addr = (void *)__get_free_pages(flag, get_order(size));
-	if (addr)
-		*dma_handle = (dma_addr_t)virt_to_phys(addr);
-
-	return addr;
+	if (boot_cpu_data.cpu_type == pcxl2 || boot_cpu_data.cpu_type == pcxl)
+		return pa11_dma_alloc(dev, size, dma_handle, gfp, attrs);
+	if (attrs & DMA_ATTR_NON_CONSISTENT)
+		return dma_direct_alloc(dev, size, dma_handle, gfp, attrs);
+	return NULL;
 }
 
-static void pcx_dma_free(struct device *dev, size_t size, void *vaddr,
-		dma_addr_t iova, unsigned long attrs)
+void arch_dma_free(struct device *dev, size_t size, void *cpu_addr,
+		dma_addr_t dma_addr, unsigned long attrs)
 {
-	free_pages((unsigned long)vaddr, get_order(size));
-	return;
+	if (boot_cpu_data.cpu_type == pcxl2 || boot_cpu_data.cpu_type == pcxl)
+		pa11_dma_free(dev, size, cpu_addr, dma_addr, attrs);
+	else
+		dma_direct_free(dev, size, cpu_addr, dma_addr, attrs);
 }
-
-const struct dma_map_ops pcx_dma_ops = {
-	.alloc =		pcx_dma_alloc,
-	.free =			pcx_dma_free,
-	.map_page =		pa11_dma_map_page,
-	.unmap_page =		pa11_dma_unmap_page,
-	.map_sg =		pa11_dma_map_sg,
-	.unmap_sg =		pa11_dma_unmap_sg,
-	.sync_single_for_cpu =	pa11_dma_sync_single_for_cpu,
-	.sync_single_for_device = pa11_dma_sync_single_for_device,
-	.sync_sg_for_cpu =	pa11_dma_sync_sg_for_cpu,
-	.sync_sg_for_device =	pa11_dma_sync_sg_for_device,
-	.cache_sync =		pa11_dma_cache_sync,
-};
diff --git a/arch/parisc/kernel/setup.c b/arch/parisc/kernel/setup.c
index 8d3a7b80ac42..4e87c35c22b7 100644
--- a/arch/parisc/kernel/setup.c
+++ b/arch/parisc/kernel/setup.c
@@ -97,14 +97,12 @@ void __init dma_ops_init(void)
 		panic(	"PA-RISC Linux currently only supports machines that conform to\n"
 			"the PA-RISC 1.1 or 2.0 architecture specification.\n");
 
-	case pcxs:
-	case pcxt:
-		hppa_dma_ops = &pcx_dma_ops;
-		break;
 	case pcxl2:
 		pa7300lc_init();
 	case pcxl: /* falls through */
-		hppa_dma_ops = &pcxl_dma_ops;
+	case pcxs:
+	case pcxt:
+		hppa_dma_ops = &dma_noncoherent_ops;
 		break;
 	default:
 		break;
diff --git a/arch/parisc/mm/init.c b/arch/parisc/mm/init.c
index cab32ee824d2..4ad91c28ecbe 100644
--- a/arch/parisc/mm/init.c
+++ b/arch/parisc/mm/init.c
@@ -19,7 +19,6 @@
 #include <linux/gfp.h>
 #include <linux/delay.h>
 #include <linux/init.h>
-#include <linux/pci.h>		/* for hppa_dma_ops and pcxl_dma_ops */
 #include <linux/initrd.h>
 #include <linux/swap.h>
 #include <linux/unistd.h>
@@ -616,17 +615,13 @@ void __init mem_init(void)
 	free_all_bootmem();
 
 #ifdef CONFIG_PA11
-	if (hppa_dma_ops == &pcxl_dma_ops) {
+	if (boot_cpu_data.cpu_type == pcxl2 || boot_cpu_data.cpu_type == pcxl) {
 		pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
 		parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
 						+ PCXL_DMA_MAP_SIZE);
-	} else {
-		pcxl_dma_start = 0;
-		parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
-	}
-#else
-	parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
+	} else
 #endif
+		parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
 
 	mem_init_print_info(NULL);
 
-- 
2.17.0

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: [PATCH 04/20] arm-nommu: use generic dma_noncoherent_ops
  2018-05-11  7:59 ` [PATCH 04/20] arm-nommu: " Christoph Hellwig
@ 2018-05-11  9:11   ` Russell King - ARM Linux
  2018-05-22 11:53     ` Christoph Hellwig
  2018-05-11 13:56   ` John Garry
  1 sibling, 1 reply; 39+ messages in thread
From: Russell King - ARM Linux @ 2018-05-11  9:11 UTC (permalink / raw)
  To: Christoph Hellwig
  Cc: iommu, linux-arch, linux-xtensa, Michal Simek, Vincent Chen,
	linux-c6x-dev, linux-parisc, linux-sh, linux-hexagon,
	linux-kernel, linux-m68k, openrisc, Greentime Hu, linux-alpha,
	sparclinux, nios2-dev, linux-snps-arc, linux-arm-kernel

On Fri, May 11, 2018 at 09:59:29AM +0200, Christoph Hellwig wrote:
> Switch to the generic noncoherent direct mapping implementation for
> the nommu dma map implementation.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  arch/arc/Kconfig                |   1 +
>  arch/arm/Kconfig                |   4 +
>  arch/arm/mm/dma-mapping-nommu.c | 139 +++++---------------------------
>  3 files changed, 23 insertions(+), 121 deletions(-)
> 
> diff --git a/arch/arc/Kconfig b/arch/arc/Kconfig
> index 89d47eac18b2..3a492a9aeaad 100644
> --- a/arch/arc/Kconfig
> +++ b/arch/arc/Kconfig
> @@ -9,6 +9,7 @@
>  config ARC
>  	def_bool y
>  	select ARC_TIMERS
> +	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
>  	select ARCH_HAS_SYNC_DMA_FOR_CPU
>  	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
>  	select ARCH_HAS_SG_CHAIN
> diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
> index c43f5bb55ac8..76ddd0064f87 100644
> --- a/arch/arm/Kconfig
> +++ b/arch/arm/Kconfig
> @@ -12,6 +12,8 @@ config ARM
>  	select ARCH_HAS_PHYS_TO_DMA
>  	select ARCH_HAS_STRICT_KERNEL_RWX if MMU && !XIP_KERNEL
>  	select ARCH_HAS_STRICT_MODULE_RWX if MMU
> +	select ARCH_HAS_SYNC_DMA_FOR_CPU if !MMU
> +	select ARCH_HAS_SYNC_DMA_FOR_DEVICE if !MMU
>  	select ARCH_HAS_TICK_BROADCAST if GENERIC_CLOCKEVENTS_BROADCAST
>  	select ARCH_HAVE_CUSTOM_GPIO_H
>  	select ARCH_HAS_GCOV_PROFILE_ALL
> @@ -27,6 +29,8 @@ config ARM
>  	select CPU_PM if (SUSPEND || CPU_IDLE)
>  	select DCACHE_WORD_ACCESS if HAVE_EFFICIENT_UNALIGNED_ACCESS
>  	select DMA_DIRECT_OPS if !MMU
> +	select DMA_NONCOHERENT_OPS if !MMU
> +	select DMA_NONCOHERENT_MMAP if !MMU
>  	select EDAC_SUPPORT
>  	select EDAC_ATOMIC_SCRUB
>  	select GENERIC_ALLOCATOR
> diff --git a/arch/arm/mm/dma-mapping-nommu.c b/arch/arm/mm/dma-mapping-nommu.c
> index f448a0663b10..a74ed6632982 100644
> --- a/arch/arm/mm/dma-mapping-nommu.c
> +++ b/arch/arm/mm/dma-mapping-nommu.c
> @@ -12,6 +12,7 @@
>  #include <linux/export.h>
>  #include <linux/mm.h>
>  #include <linux/dma-direct.h>
> +#include <linux/dma-noncoherent.h>
>  #include <linux/scatterlist.h>
>  
>  #include <asm/cachetype.h>
> @@ -26,18 +27,16 @@
>   *   - MMU/MPU is off
>   *   - cpu is v7m w/o cache support
>   *   - device is coherent
> - *  otherwise arm_nommu_dma_ops is used.
> + *  otherwise dma_noncoherent_ops is used.
>   *
> - *  arm_nommu_dma_ops rely on consistent DMA memory (please, refer to
> + *  dma_noncoherent_ops rely on consistent DMA memory (please, refer to
>   *  [1] on how to declare such memory).
>   *
>   *  [1] Documentation/devicetree/bindings/reserved-memory/reserved-memory.txt
>   */
>  
> -static void *arm_nommu_dma_alloc(struct device *dev, size_t size,
> -				 dma_addr_t *dma_handle, gfp_t gfp,
> -				 unsigned long attrs)
> -
> +void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
> +		gfp_t gfp, unsigned long attrs)
>  {
>  	void *ret;
>  
> @@ -65,9 +64,8 @@ static void *arm_nommu_dma_alloc(struct device *dev, size_t size,
>  	return ret;
>  }
>  
> -static void arm_nommu_dma_free(struct device *dev, size_t size,
> -			       void *cpu_addr, dma_addr_t dma_addr,
> -			       unsigned long attrs)
> +void arch_dma_free(struct device *dev, size_t size, void *cpu_addr,
> +		dma_addr_t dma_addr, unsigned long attrs)
>  {
>  	if (attrs & DMA_ATTR_NON_CONSISTENT) {
>  		dma_direct_free(dev, size, cpu_addr, dma_addr, attrs);
> @@ -81,9 +79,9 @@ static void arm_nommu_dma_free(struct device *dev, size_t size,
>  	return;
>  }
>  
> -static int arm_nommu_dma_mmap(struct device *dev, struct vm_area_struct *vma,
> -			      void *cpu_addr, dma_addr_t dma_addr, size_t size,
> -			      unsigned long attrs)
> +int arch_dma_mmap(struct device *dev, struct vm_area_struct *vma,
> +		void *cpu_addr, dma_addr_t dma_addr, size_t size,
> +		unsigned long attrs)
>  {
>  	int ret;
>  
> @@ -93,9 +91,8 @@ static int arm_nommu_dma_mmap(struct device *dev, struct vm_area_struct *vma,
>  	return dma_common_mmap(dev, vma, cpu_addr, dma_addr, size);
>  }
>  
> -
> -static void __dma_page_cpu_to_dev(phys_addr_t paddr, size_t size,
> -				  enum dma_data_direction dir)
> +void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
> +		size_t size, enum dma_data_direction dir)

Please no.  There is a lot of history of these (__dma_page_cpu_to_dev etc)
functions being abused by out of tree drivers, because they think they
know better.  This is stopped by making them static and ensuring that
drivers have no access to these functions.

Please do not re-expose these to the global kernel.

While it may make things easier for a cross-architecture point of view,
it makes it a lot easier for people to abuse these private APIs.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 8.8Mbps down 630kbps up
According to speedtest.net: 8.21Mbps down 510kbps up

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: [PATCH 03/20] arc: use generic dma_noncoherent_ops
  2018-05-11  7:59 ` [PATCH 03/20] arc: use generic dma_noncoherent_ops Christoph Hellwig
@ 2018-05-11 12:44   ` Alexey Brodkin
  0 siblings, 0 replies; 39+ messages in thread
From: Alexey Brodkin @ 2018-05-11 12:44 UTC (permalink / raw)
  To: hch
  Cc: deanbo422, linux-sh, linux-kernel, nios2-dev, linux-xtensa,
	linux-m68k, linux-alpha, linux-hexagon, linux-snps-arc, iommu,
	green.hu, openrisc, linux-arm-kernel, monstr, linux-parisc,
	linux-c6x-dev, linux-arch, sparclinux

Hi Christoph,

On Fri, 2018-05-11 at 09:59 +0200, Christoph Hellwig wrote:
> Switch to the generic noncoherent direct mapping implementation.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  arch/arc/Kconfig                   |   4 +
>  arch/arc/include/asm/Kbuild        |   1 +
>  arch/arc/include/asm/dma-mapping.h |  21 -----
>  arch/arc/mm/dma.c                  | 141 +++--------------------------
>  4 files changed, 19 insertions(+), 148 deletions(-)
>  delete mode 100644 arch/arc/include/asm/dma-mapping.h

Sorry I didn't try your proposed fix earlier - was away for a couple of days.

But that newer version with fixed lib/dma-noncoherent.c still doesn't work on ARC.

USB ECHI controller fails like that:
------------------------->8-------------------------
usb 1-1: new high-speed USB device number 2 using ehci-platform
usb 1-1: can't set config #1, error -32
------------------------->8-------------------------

Again with entire series reverted USB starts to work.
Let's wait for other arches previously affected to update on their status.

-Alexey

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: [PATCH 04/20] arm-nommu: use generic dma_noncoherent_ops
  2018-05-11  7:59 ` [PATCH 04/20] arm-nommu: " Christoph Hellwig
  2018-05-11  9:11   ` Russell King - ARM Linux
@ 2018-05-11 13:56   ` John Garry
  1 sibling, 0 replies; 39+ messages in thread
From: John Garry @ 2018-05-11 13:56 UTC (permalink / raw)
  To: Christoph Hellwig, iommu
  Cc: linux-arch, linux-xtensa, Michal Simek, Vincent Chen,
	linux-c6x-dev, linux-parisc, linux-sh, linux-hexagon,
	linux-kernel, linux-m68k, openrisc, Greentime Hu, linux-alpha,
	sparclinux, nios2-dev, linux-snps-arc, linux-arm-kernel

On 11/05/2018 08:59, Christoph Hellwig wrote:
> Switch to the generic noncoherent direct mapping implementation for
> the nommu dma map implementation.
>
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  arch/arc/Kconfig                |   1 +
>  arch/arm/Kconfig                |   4 +
>  arch/arm/mm/dma-mapping-nommu.c | 139 +++++---------------------------
>  3 files changed, 23 insertions(+), 121 deletions(-)
>
> diff --git a/arch/arc/Kconfig b/arch/arc/Kconfig
> index 89d47eac18b2..3a492a9aeaad 100644
> --- a/arch/arc/Kconfig
> +++ b/arch/arc/Kconfig
> @@ -9,6 +9,7 @@
>  config ARC
>  	def_bool y
>  	select ARC_TIMERS
> +	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
>  	select ARCH_HAS_SYNC_DMA_FOR_CPU
>  	select ARCH_HAS_SYNC_DMA_FOR_DEVICE

I guess that this arc change is here by accident, no? And isn't 
ARCH_HAS_SYNC_DMA_FOR_DEVICE already selected (by 3/20)?

>  	select ARCH_HAS_SG_CHAIN
> diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
> index c43f5bb55ac8..76ddd0064f87 100644
> --- a/arch/arm/Kconfig
> +++ b/arch/arm/Kconfig

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: common non-cache coherent direct dma mapping ops
  2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
                   ` (19 preceding siblings ...)
  2018-05-11  7:59 ` [PATCH 20/20] parisc: " Christoph Hellwig
@ 2018-05-13 13:26 ` Helge Deller
  20 siblings, 0 replies; 39+ messages in thread
From: Helge Deller @ 2018-05-13 13:26 UTC (permalink / raw)
  To: Christoph Hellwig, iommu, James Bottomley
  Cc: linux-arch, Michal Simek, Greentime Hu, Vincent Chen,
	linux-alpha, linux-snps-arc, linux-arm-kernel, linux-c6x-dev,
	linux-hexagon, linux-m68k, nios2-dev, openrisc, linux-parisc,
	linux-sh, sparclinux, linux-kernel

On 11.05.2018 09:59, Christoph Hellwig wrote:
> this series continues consolidating the dma-mapping code, with a focus
> on architectures that do not (always) provide cache coherence for DMA.
> Three architectures (arm, mips and powerpc) are still left to be
> converted later due to complexity of their dma ops selection.
> 
> The dma-noncoherent ops calls the dma-direct ops for the actual
> translation of streaming mappins and allow the architecture to provide
> any cache flushing required for cpu to device and/or device to cpu
> ownership transfers.  The dma coherent allocator is for now still left
> entirely to architecture supplied implementations due the amount of
> variations.  Hopefully we can do some consolidation for them later on
> as well.
> 
> A lot of architectures are currently doing very questionable things
> in their dma mapping routines, which are documented in the changelogs
> for each patch.  Please review them very careful and correct me on
> incorrect assumptions.
> 
> Because this series sits on top of two previously submitted series
> a git tree might be useful to actually test it.  It is provided here:
> 
>     git://git.infradead.org/users/hch/misc.git generic-dma-noncoherent
> 
> Gitweb:
> 
>     http://git.infradead.org/users/hch/misc.git/shortlog/refs/heads/generic-dma-noncoherent
> 
> Changes since RFC:
>  - fix a typo accidentally disabling the device to cpu transfer sync
>  - fixed a few compile failures


I tested it again on parisc (this time again on top of git head) and it still breaks
the same way as I reported in my mail on April 21st: the lasi82956 network driver works
unreliable. NIC gets IP, but ping doesn't work.
See drivers/net/ethernet/i825xx/lasi_82596.c, it uses dma*sync() functions.

See comment in James mail from April 21st too:
-> you just made every 32 bit parisc system unnecessarily use non-coherent.

Helge

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: [Linux-c6x-dev] [PATCH 05/20] c6x: use generic dma_noncoherent_ops
  2018-05-11  7:59 ` [PATCH 05/20] c6x: " Christoph Hellwig
@ 2018-05-15  0:25   ` Mark Salter
  0 siblings, 0 replies; 39+ messages in thread
From: Mark Salter @ 2018-05-15  0:25 UTC (permalink / raw)
  To: Christoph Hellwig, iommu
  Cc: linux-arch, linux-xtensa, Michal Simek, Vincent Chen,
	linux-c6x-dev, linux-parisc, linux-sh, linux-hexagon,
	linux-kernel, linux-m68k, openrisc, Greentime Hu, linux-alpha,
	sparclinux, nios2-dev, linux-snps-arc, linux-arm-kernel

On Fri, 2018-05-11 at 09:59 +0200, Christoph Hellwig wrote:
> Switch to the generic noncoherent direct mapping implementation.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  arch/c6x/Kconfig                   |   3 +
>  arch/c6x/include/asm/Kbuild        |   1 +
>  arch/c6x/include/asm/dma-mapping.h |  28 ------
>  arch/c6x/include/asm/setup.h       |   2 +
>  arch/c6x/kernel/Makefile           |   2 +-
>  arch/c6x/kernel/dma.c              | 138 -----------------------------
>  arch/c6x/mm/dma-coherent.c         |  40 ++++++++-
>  7 files changed, 44 insertions(+), 170 deletions(-)
>  delete mode 100644 arch/c6x/include/asm/dma-mapping.h
>  delete mode 100644 arch/c6x/kernel/dma.c
> 
> diff --git a/arch/c6x/Kconfig b/arch/c6x/Kconfig
> index 8c088b96e372..bf59855628ac 100644
> --- a/arch/c6x/Kconfig
> +++ b/arch/c6x/Kconfig
> @@ -6,7 +6,10 @@
>  
>  config C6X
>  	def_bool y
> +	select ARCH_HAS_SYNC_DMA_FOR_CPU
> +	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
>  	select CLKDEV_LOOKUP
> +	select DMA_NONCOHERENT_OPS
>  	select GENERIC_ATOMIC64
>  	select GENERIC_IRQ_SHOW
>  	select HAVE_ARCH_TRACEHOOK
> diff --git a/arch/c6x/include/asm/Kbuild b/arch/c6x/include/asm/Kbuild
> index fd4c840de837..434600e47662 100644
> --- a/arch/c6x/include/asm/Kbuild
> +++ b/arch/c6x/include/asm/Kbuild
> @@ -5,6 +5,7 @@ generic-y += current.h
>  generic-y += device.h
>  generic-y += div64.h
>  generic-y += dma.h
> +generic-y += dma-mapping.h
>  generic-y += emergency-restart.h
>  generic-y += exec.h
>  generic-y += extable.h
> diff --git a/arch/c6x/include/asm/dma-mapping.h b/arch/c6x/include/asm/dma-mapping.h
> deleted file mode 100644
> index 05daf1038111..000000000000
> --- a/arch/c6x/include/asm/dma-mapping.h
> +++ /dev/null
> @@ -1,28 +0,0 @@
> -/*
> - *  Port on Texas Instruments TMS320C6x architecture
> - *
> - *  Copyright (C) 2004, 2009, 2010, 2011 Texas Instruments Incorporated
> - *  Author: Aurelien Jacquiot <aurelien.jacquiot@ti.com>
> - *
> - *  This program is free software; you can redistribute it and/or modify
> - *  it under the terms of the GNU General Public License version 2 as
> - *  published by the Free Software Foundation.
> - *
> - */
> -#ifndef _ASM_C6X_DMA_MAPPING_H
> -#define _ASM_C6X_DMA_MAPPING_H
> -
> -extern const struct dma_map_ops c6x_dma_ops;
> -
> -static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
> -{
> -	return &c6x_dma_ops;
> -}
> -
> -extern void coherent_mem_init(u32 start, u32 size);
> -void *c6x_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
> -		gfp_t gfp, unsigned long attrs);
> -void c6x_dma_free(struct device *dev, size_t size, void *vaddr,
> -		dma_addr_t dma_handle, unsigned long attrs);
> -
> -#endif	/* _ASM_C6X_DMA_MAPPING_H */
> diff --git a/arch/c6x/include/asm/setup.h b/arch/c6x/include/asm/setup.h
> index 852afb209afb..350f34debb19 100644
> --- a/arch/c6x/include/asm/setup.h
> +++ b/arch/c6x/include/asm/setup.h
> @@ -28,5 +28,7 @@ extern unsigned char c6x_fuse_mac[6];
>  extern void machine_init(unsigned long dt_ptr);
>  extern void time_init(void);
>  
> +extern void coherent_mem_init(u32 start, u32 size);
> +
>  #endif /* !__ASSEMBLY__ */
>  #endif /* _ASM_C6X_SETUP_H */
> diff --git a/arch/c6x/kernel/Makefile b/arch/c6x/kernel/Makefile
> index 02f340d7b8fe..fbe74174de87 100644
> --- a/arch/c6x/kernel/Makefile
> +++ b/arch/c6x/kernel/Makefile
> @@ -8,6 +8,6 @@ extra-y := head.o vmlinux.lds
>  obj-y := process.o traps.o irq.o signal.o ptrace.o
>  obj-y += setup.o sys_c6x.o time.o devicetree.o
>  obj-y += switch_to.o entry.o vectors.o c6x_ksyms.o
> -obj-y += soc.o dma.o
> +obj-y += soc.o
>  
>  obj-$(CONFIG_MODULES)           += module.o
> diff --git a/arch/c6x/kernel/dma.c b/arch/c6x/kernel/dma.c
> deleted file mode 100644
> index 31e1a9ec3a9c..000000000000
> --- a/arch/c6x/kernel/dma.c
> +++ /dev/null
> @@ -1,138 +0,0 @@
> -/*
> - *  Copyright (C) 2011 Texas Instruments Incorporated
> - *  Author: Mark Salter <msalter@redhat.com>
> - *
> - *  This program is free software; you can redistribute it and/or modify
> - *  it under the terms of the GNU General Public License version 2 as
> - *  published by the Free Software Foundation.
> - */
> -#include <linux/module.h>
> -#include <linux/dma-mapping.h>
> -#include <linux/mm.h>
> -#include <linux/mm_types.h>
> -#include <linux/scatterlist.h>
> -
> -#include <asm/cacheflush.h>
> -
> -static void c6x_dma_sync(dma_addr_t handle, size_t size,
> -			 enum dma_data_direction dir)
> -{
> -	unsigned long paddr = handle;
> -
> -	BUG_ON(!valid_dma_direction(dir));
> -
> -	switch (dir) {
> -	case DMA_FROM_DEVICE:
> -		L2_cache_block_invalidate(paddr, paddr + size);
> -		break;
> -	case DMA_TO_DEVICE:
> -		L2_cache_block_writeback(paddr, paddr + size);
> -		break;
> -	case DMA_BIDIRECTIONAL:
> -		L2_cache_block_writeback_invalidate(paddr, paddr + size);
> -		break;
> -	default:
> -		break;
> -	}
> -}
> -
> -static dma_addr_t c6x_dma_map_page(struct device *dev, struct page *page,
> -		unsigned long offset, size_t size, enum dma_data_direction dir,
> -		unsigned long attrs)
> -{
> -	dma_addr_t handle = virt_to_phys(page_address(page) + offset);
> -
> -	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
> -		c6x_dma_sync(handle, size, dir);
> -
> -	return handle;
> -}
> -
> -static void c6x_dma_unmap_page(struct device *dev, dma_addr_t handle,
> -		size_t size, enum dma_data_direction dir, unsigned long attrs)
> -{
> -	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
> -		c6x_dma_sync(handle, size, dir);
> -}
> -
> -static int c6x_dma_map_sg(struct device *dev, struct scatterlist *sglist,
> -		int nents, enum dma_data_direction dir, unsigned long attrs)
> -{
> -	struct scatterlist *sg;
> -	int i;
> -
> -	for_each_sg(sglist, sg, nents, i) {
> -		sg->dma_address = sg_phys(sg);
> -		if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
> -			c6x_dma_sync(sg->dma_address, sg->length, dir);
> -	}
> -
> -	return nents;
> -}
> -
> -static void c6x_dma_unmap_sg(struct device *dev, struct scatterlist *sglist,
> -		  int nents, enum dma_data_direction dir, unsigned long attrs)
> -{
> -	struct scatterlist *sg;
> -	int i;
> -
> -	if (attrs & DMA_ATTR_SKIP_CPU_SYNC)
> -		return;
> -
> -	for_each_sg(sglist, sg, nents, i)
> -		c6x_dma_sync(sg_dma_address(sg), sg->length, dir);
> -}
> -
> -static void c6x_dma_sync_single_for_cpu(struct device *dev, dma_addr_t handle,
> -		size_t size, enum dma_data_direction dir)
> -{
> -	c6x_dma_sync(handle, size, dir);
> -
> -}
> -
> -static void c6x_dma_sync_single_for_device(struct device *dev,
> -		dma_addr_t handle, size_t size, enum dma_data_direction dir)
> -{
> -	c6x_dma_sync(handle, size, dir);
> -
> -}
> -
> -static void c6x_dma_sync_sg_for_cpu(struct device *dev,
> -		struct scatterlist *sglist, int nents,
> -		enum dma_data_direction dir)
> -{
> -	struct scatterlist *sg;
> -	int i;
> -
> -	for_each_sg(sglist, sg, nents, i)
> -		c6x_dma_sync_single_for_cpu(dev, sg_dma_address(sg),
> -					sg->length, dir);
> -
> -}
> -
> -static void c6x_dma_sync_sg_for_device(struct device *dev,
> -		struct scatterlist *sglist, int nents,
> -		enum dma_data_direction dir)
> -{
> -	struct scatterlist *sg;
> -	int i;
> -
> -	for_each_sg(sglist, sg, nents, i)
> -		c6x_dma_sync_single_for_device(dev, sg_dma_address(sg),
> -					   sg->length, dir);
> -
> -}
> -
> -const struct dma_map_ops c6x_dma_ops = {
> -	.alloc			= c6x_dma_alloc,
> -	.free			= c6x_dma_free,
> -	.map_page		= c6x_dma_map_page,
> -	.unmap_page		= c6x_dma_unmap_page,
> -	.map_sg			= c6x_dma_map_sg,
> -	.unmap_sg		= c6x_dma_unmap_sg,
> -	.sync_single_for_device	= c6x_dma_sync_single_for_device,
> -	.sync_single_for_cpu	= c6x_dma_sync_single_for_cpu,
> -	.sync_sg_for_device	= c6x_dma_sync_sg_for_device,
> -	.sync_sg_for_cpu	= c6x_dma_sync_sg_for_cpu,
> -};
> -EXPORT_SYMBOL(c6x_dma_ops);
> diff --git a/arch/c6x/mm/dma-coherent.c b/arch/c6x/mm/dma-coherent.c
> index 95e38ad27c69..d0a8e0c4b27e 100644
> --- a/arch/c6x/mm/dma-coherent.c
> +++ b/arch/c6x/mm/dma-coherent.c
> @@ -19,10 +19,12 @@
>  #include <linux/bitops.h>
>  #include <linux/module.h>
>  #include <linux/interrupt.h>
> -#include <linux/dma-mapping.h>
> +#include <linux/dma-noncoherent.h>
>  #include <linux/memblock.h>
>  
> +#include <asm/cacheflush.h>
>  #include <asm/page.h>
> +#include <asm/setup.h>
>  
>  /*
>   * DMA coherent memory management, can be redefined using the memdma=
> @@ -73,7 +75,7 @@ static void __free_dma_pages(u32 addr, int order)
>   * Allocate DMA coherent memory space and return both the kernel
>   * virtual and DMA address for that space.
>   */
> -void *c6x_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
> +void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
>  		gfp_t gfp, unsigned long attrs)
>  {
>  	u32 paddr;
> @@ -98,7 +100,7 @@ void *c6x_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
>  /*
>   * Free DMA coherent memory as defined by the above mapping.
>   */
> -void c6x_dma_free(struct device *dev, size_t size, void *vaddr,
> +void arch_dma_free(struct device *dev, size_t size, void *vaddr,
>  		dma_addr_t dma_handle, unsigned long attrs)
>  {
>  	int order;
> @@ -139,3 +141,35 @@ void __init coherent_mem_init(phys_addr_t start, u32 size)
>  	dma_bitmap = phys_to_virt(bitmap_phys);
>  	memset(dma_bitmap, 0, dma_pages * PAGE_SIZE);
>  }
> +
> +static void c6x_dma_sync(struct device *dev, phys_addr_t paddr, size_t size,
> +		enum dma_data_direction dir)
> +{
> +	BUG_ON(!valid_dma_direction(dir));
> +
> +	switch (dir) {
> +	case DMA_FROM_DEVICE:
> +		L2_cache_block_invalidate(paddr, paddr + size);
> +		break;
> +	case DMA_TO_DEVICE:
> +		L2_cache_block_writeback(paddr, paddr + size);
> +		break;
> +	case DMA_BIDIRECTIONAL:
> +		L2_cache_block_writeback_invalidate(paddr, paddr + size);
> +		break;
> +	default:
> +		break;
> +	}
> +}
> +
> +void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
> +		size_t size, enum dma_data_direction dir)
> +{
> +	return c6x_dma_sync(dev, paddr, size, dir);
> +}
> +
> +void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
> +		size_t size, enum dma_data_direction dir)
> +{
> +	return c6x_dma_sync(dev, paddr, size, dir);
> +}
Acked-by: Mark Salter <msalter@redhat.com>

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation
  2018-05-11  7:59 ` [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation Christoph Hellwig
@ 2018-05-18 13:03   ` Alexey Brodkin
  2018-05-18 13:27     ` hch
                       ` (2 more replies)
  0 siblings, 3 replies; 39+ messages in thread
From: Alexey Brodkin @ 2018-05-18 13:03 UTC (permalink / raw)
  To: hch
  Cc: deanbo422, linux-sh, linux-kernel, nios2-dev, linux-xtensa,
	linux-m68k@lists.linux-m68k.org, linux-alpha, linux-hexagon,
	linux-snps-arc, iommu, green.hu, openrisc, linux-arm-kernel,
	monstr, linux-parisc, linux-c6x-dev, linux-arch, sparclinux

Hi Christoph,

On Fri, 2018-05-11 at 09:59 +0200, Christoph Hellwig wrote:

[snip]

There seems to be one subtle issue with map/unmap code.
While investigating problems on ARC I added instrumentation as below:
---------------------------------------->8------------------------------------
--- a/arch/arc/mm/dma.c
+++ b/arch/arc/mm/dma.c
@@ -152,14 +152,37 @@ static void _dma_cache_sync(struct device *dev, phys_addr_t paddr, size_t size,
        }
 }
 
+static const char *dir_to_str(enum dma_data_direction dir)
+{
+       switch (dir) {
+       case DMA_BIDIRECTIONAL: return "DMA_BIDIRECTIONAL";
+       case DMA_TO_DEVICE: return "DMA_TO_DEVICE";
+       case DMA_FROM_DEVICE: return "DMA_FROM_DEVICE";
+       case DMA_NONE: return "DMA_NONE";
+       default: return "WRONG_VALUE!";
+       }
+}
+
 void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
                size_t size, enum dma_data_direction dir)
 {
+       if (dir != DMA_TO_DEVICE){
+               dump_stack();
+               printk(" *** %s@%d: DMA direction is %s instead of %s\n",
+                      __func__, __LINE__, dir_to_str(dir), dir_to_str(DMA_TO_DEVICE));
+       }
+
        return _dma_cache_sync(dev, paddr, size, dir);
 }
 
 void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
                size_t size, enum dma_data_direction dir)
 {
+       if (dir != DMA_FROM_DEVICE) {
+               dump_stack();
+               printk(" *** %s@%d: DMA direction is %s instead of %s\n",
+                      __func__, __LINE__, dir_to_str(dir), dir_to_str(DMA_FROM_DEVICE));
+       }
+
        return _dma_cache_sync(dev, paddr, size, dir);
 }
---------------------------------------->8------------------------------------

And with that I noticed a bit unexpected output, see below:
---------------------------------------->8------------------------------------
Stack Trace:
  arc_unwind_core.constprop.1+0xd4/0xf8
  dump_stack+0x68/0x80
  arch_sync_dma_for_device+0x34/0xc4
  dma_noncoherent_map_sg+0x80/0x94
  __dw_mci_start_request+0x1ee/0x868
  dw_mci_request+0x17e/0x1c8
  mmc_wait_for_req+0x106/0x1ac
  mmc_app_sd_status+0x108/0x130
  mmc_sd_setup_card+0xc6/0x2e8
  mmc_attach_sd+0x1b6/0x394
  mmc_rescan+0x2f4/0x3bc
  process_one_work+0x194/0x348
  worker_thread+0xf2/0x478
  kthread+0x120/0x13c
  ret_from_fork+0x18/0x1c
 *** arch_sync_dma_for_device@172: DMA direction is DMA_FROM_DEVICE instead of DMA_TO_DEVICE
...
Stack Trace:
  arc_unwind_core.constprop.1+0xd4/0xf8
  dump_stack+0x68/0x80
  arch_sync_dma_for_device+0x34/0xc4
  dma_noncoherent_map_page+0x86/0x8c
  usb_hcd_map_urb_for_dma+0x49e/0x53c
  usb_hcd_submit_urb+0x43c/0x8c4
  usb_control_msg+0xbe/0x16c
  hub_port_init+0x5e0/0xb0c
  hub_event+0x4e6/0x1164
  process_one_work+0x194/0x348
  worker_thread+0xf2/0x478
  kthread+0x120/0x13c
  ret_from_fork+0x18/0x1c
 mmcblk0: p1 p2
 *** arch_sync_dma_for_device@172: DMA direction is DMA_FROM_DEVICE instead of DMA_TO_DEVICE

...
and quite some more of the similar
...
---------------------------------------->8------------------------------------

In case of MMC/DW_MCI (AKA DesignWare MobileStorage controller) that's an execution flow:
1) __dw_mci_start_request()
2) dw_mci_pre_dma_transfer()
3) dma_map_sg(..., mmc_get_dma_dir(data))

Note mmc_get_dma_dir() is just "data->flags & MMC_DATA_WRITE ? DMA_TO_DEVICE : DMA_FROM_DEVICE".
I.e. if we're preparing for sending data dma_noncoherent_map_sg() will have DMA_TO_DEVICE which
is quite OK for passing to dma_noncoherent_sync_sg_for_device() but in case of reading we'll have
DMA_FROM_DEVICE which we'll pass to dma_noncoherent_sync_sg_for_device() in dma_noncoherent_map_sg().

I'd say this is not entirely correct because IMHO arch_sync_dma_for_cpu() is supposed to only be used
in case of DMA_FROM_DEVICE and arch_sync_dma_for_device() only in case of DMA_TO_DEVICE.


> +static dma_addr_t dma_noncoherent_map_page(struct device *dev, struct page *page,
> +		unsigned long offset, size_t size, enum dma_data_direction dir,
> +		unsigned long attrs)
> +{
> +	dma_addr_t addr;
> +
> +	addr = dma_direct_map_page(dev, page, offset, size, dir, attrs);
> +	if (!dma_mapping_error(dev, addr) && !(attrs & DMA_ATTR_SKIP_CPU_SYNC))
> +		arch_sync_dma_for_device(dev, page_to_phys(page), size, dir);
> +	return addr;
> +}
> +
> +static int dma_noncoherent_map_sg(struct device *dev, struct scatterlist *sgl,
> +		int nents, enum dma_data_direction dir, unsigned long attrs)
> +{
> +	nents = dma_direct_map_sg(dev, sgl, nents, dir, attrs);
> +	if (nents > 0 && !(attrs & DMA_ATTR_SKIP_CPU_SYNC))
> +		dma_noncoherent_sync_sg_for_device(dev, sgl, nents, dir);
> +	return nents;
> +}

The same is for unmap functions.
My guess is we need to respect direction in map/unmap functions and use
either dma_noncoherent_sync_single_for_cpu(..., DMA_FROM_DEVICE) or
dma_noncoherent_sync_single_for_device(...,DMA_TO_DEVICE).


> +static void dma_noncoherent_unmap_page(struct device *dev, dma_addr_t addr,
> +		size_t size, enum dma_data_direction dir, unsigned long attrs)
> +{
> +	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
> +		dma_noncoherent_sync_single_for_cpu(dev, addr, size, dir);
> +}
> +
> +static void dma_noncoherent_unmap_sg(struct device *dev, struct scatterlist *sgl,
> +		int nents, enum dma_data_direction dir, unsigned long attrs)
> +{
> +	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
> +		dma_noncoherent_sync_sg_for_cpu(dev, sgl, nents, dir);
> +}
> +#endif

But the real fix of my problem is:
---------------------------------------->8------------------------------------
--- a/lib/dma-noncoherent.c
+++ b/lib/dma-noncoherent.c
@@ -35,7 +35,7 @@ static dma_addr_t dma_noncoherent_map_page(struct device *dev, struct page *page
 
        addr = dma_direct_map_page(dev, page, offset, size, dir, attrs);
        if (!dma_mapping_error(dev, addr) && !(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-               arch_sync_dma_for_device(dev, page_to_phys(page), size, dir);
+               arch_sync_dma_for_device(dev, page_to_phys(page) + offset, size, dir);
        return addr;
 }
---------------------------------------->8------------------------------------

You seem to lost an offset in the page so if we happen to have a buffer not aligned to
a page boundary then we were obviously corrupting data outside our data :)

-Alexey

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation
  2018-05-18 13:03   ` Alexey Brodkin
@ 2018-05-18 13:27     ` hch
  2018-05-18 14:13       ` Alexey Brodkin
  2018-05-18 17:28       ` Vineet Gupta
  2018-05-18 17:20     ` dma_sync_*_for_cpu and direction=TO_DEVICE (was Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation) Vineet Gupta
  2018-05-18 20:05     ` [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation Helge Deller
  2 siblings, 2 replies; 39+ messages in thread
From: hch @ 2018-05-18 13:27 UTC (permalink / raw)
  To: Alexey Brodkin
  Cc: hch, deanbo422, linux-sh, linux-kernel, nios2-dev, linux-xtensa,
	linux-m68k, linux-alpha, linux-hexagon, linux-snps-arc, iommu,
	green.hu, openrisc, linux-arm-kernel, monstr, linux-parisc,
	linux-c6x-dev, linux-arch, sparclinux

On Fri, May 18, 2018 at 01:03:46PM +0000, Alexey Brodkin wrote:
> Note mmc_get_dma_dir() is just "data->flags & MMC_DATA_WRITE ? DMA_TO_DEVICE : DMA_FROM_DEVICE".
> I.e. if we're preparing for sending data dma_noncoherent_map_sg() will have DMA_TO_DEVICE which
> is quite OK for passing to dma_noncoherent_sync_sg_for_device() but in case of reading we'll have
> DMA_FROM_DEVICE which we'll pass to dma_noncoherent_sync_sg_for_device() in dma_noncoherent_map_sg().
> 
> I'd say this is not entirely correct because IMHO arch_sync_dma_for_cpu() is supposed to only be used
> in case of DMA_FROM_DEVICE and arch_sync_dma_for_device() only in case of DMA_TO_DEVICE.

arc overrides the dir paramter of the dma_sync_single_for_device/
dma_sync_single_for_cpu calls.  My patches dropped that, and I have
restored that, and audit for the other architectures is pending.

That being said the existing arc code still looks rather odd as it
didn't do the same thing for the scatterlist versions of the calls.
I've thrown in a few patches into my new tree to make the sg versions
make the normal calls, and to clean up the area a bit.

> You seem to lost an offset in the page so if we happen to have a buffer not aligned to
> a page boundary then we were obviously corrupting data outside our data :)

Oops!  Thank you for all the debugging!

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation
  2018-05-18 13:27     ` hch
@ 2018-05-18 14:13       ` Alexey Brodkin
  2018-05-18 17:28       ` Vineet Gupta
  1 sibling, 0 replies; 39+ messages in thread
From: Alexey Brodkin @ 2018-05-18 14:13 UTC (permalink / raw)
  To: hch
  Cc: deanbo422, linux-kernel, linux-sh, nios2-dev, linux-xtensa,
	linux-m68k@lists.linux-m68k.org, linux-alpha, linux-hexagon,
	linux-snps-arc, iommu, green.hu, openrisc, linux-arm-kernel,
	monstr, linux-parisc, linux-c6x-dev, linux-arch, sparclinux

Hi Christoph,

On Fri, 2018-05-18 at 15:27 +0200, hch@lst.de wrote:
> On Fri, May 18, 2018 at 01:03:46PM +0000, Alexey Brodkin wrote:
> > Note mmc_get_dma_dir() is just "data->flags & MMC_DATA_WRITE ? DMA_TO_DEVICE : DMA_FROM_DEVICE".
> > I.e. if we're preparing for sending data dma_noncoherent_map_sg() will have DMA_TO_DEVICE which
> > is quite OK for passing to dma_noncoherent_sync_sg_for_device() but in case of reading we'll have
> > DMA_FROM_DEVICE which we'll pass to dma_noncoherent_sync_sg_for_device() in dma_noncoherent_map_sg().
> > 
> > I'd say this is not entirely correct because IMHO arch_sync_dma_for_cpu() is supposed to only be used
> > in case of DMA_FROM_DEVICE and arch_sync_dma_for_device() only in case of DMA_TO_DEVICE.
> 
> arc overrides the dir paramter of the dma_sync_single_for_device/
> dma_sync_single_for_cpu calls.  My patches dropped that, and I have
> restored that, and audit for the other architectures is pending.

Well at least for me that's a confusion what is a reason to pass direction
to function which purpose is already known.

I'd say that XXX_sync_for_device() doesn't need _variable_ direction as an argument,
otherwise what does that mean if we pass DMA_FROM_DEVICE to that function?

> That being said the existing arc code still looks rather odd as it
> didn't do the same thing for the scatterlist versions of the calls.

That might easily be the case so good we caught that now and it will be fixed :)

> I've thrown in a few patches into my new tree to make the sg versions
> make the normal calls, and to clean up the area a bit.

I'll try your newer series now, thanks!

-Alexey

^ permalink raw reply	[flat|nested] 39+ messages in thread

* dma_sync_*_for_cpu and direction=TO_DEVICE (was Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation)
  2018-05-18 13:03   ` Alexey Brodkin
  2018-05-18 13:27     ` hch
@ 2018-05-18 17:20     ` Vineet Gupta
  2018-05-18 17:50       ` Russell King - ARM Linux
  2018-05-18 20:05     ` [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation Helge Deller
  2 siblings, 1 reply; 39+ messages in thread
From: Vineet Gupta @ 2018-05-18 17:20 UTC (permalink / raw)
  To: Alexey Brodkin, hch
  Cc: linux-arch, linux-xtensa, monstr, linux-snps-arc, linux-c6x-dev,
	linux-parisc, linux-sh, linux-hexagon, linux-kernel, iommu,
	linux-m68k, openrisc, green.hu, linux-alpha, sparclinux,
	nios2-dev, deanbo422, linux-arm-kernel, linux-mm, Andrew Morton

On 05/18/2018 06:11 AM, Alexey Brodkin wrote:
>   void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
>                  size_t size, enum dma_data_direction dir)
>   {
> +       if (dir != DMA_TO_DEVICE){
> +               dump_stack();
> +               printk(" *** %s@%d: DMA direction is %s instead of %s\n",
> +                      __func__, __LINE__, dir_to_str(dir), dir_to_str(DMA_TO_DEVICE));
> +       }
> +
>          return _dma_cache_sync(dev, paddr, size, dir);
>   }
>   
>   void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
>                  size_t size, enum dma_data_direction dir)
>   {
> +       if (dir != DMA_FROM_DEVICE) {
> +               dump_stack();
> +               printk(" *** %s@%d: DMA direction is %s instead of %s\n",
> +                      __func__, __LINE__, dir_to_str(dir), dir_to_str(DMA_FROM_DEVICE));
> +       }
> +
>          return _dma_cache_sync(dev, paddr, size, dir);
>   }

...
> In case of MMC/DW_MCI (AKA DesignWare MobileStorage controller) that's an execution flow:
> 1) __dw_mci_start_request()
> 2) dw_mci_pre_dma_transfer()
> 3) dma_map_sg(..., mmc_get_dma_dir(data))
>
> Note mmc_get_dma_dir() is just "data->flags & MMC_DATA_WRITE ? DMA_TO_DEVICE : DMA_FROM_DEVICE".
> I.e. if we're preparing for sending data dma_noncoherent_map_sg() will have DMA_TO_DEVICE which
> is quite OK for passing to dma_noncoherent_sync_sg_for_device() but in case of reading we'll have
> DMA_FROM_DEVICE which we'll pass to dma_noncoherent_sync_sg_for_device() in dma_noncoherent_map_sg().
>
> I'd say this is not entirely correct because IMHO arch_sync_dma_for_cpu() is supposed to only be used
> in case of DMA_FROM_DEVICE and arch_sync_dma_for_device() only in case of DMA_TO_DEVICE.

So roughly 10 years ago, some kernel rookie name Vineet Gupta, asked the exact 
same question :-)

http://kernelnewbies.kernelnewbies.narkive.com/aGW1QcDv/query-about-dma-sync-for-cpu-and-direction-to-device

I never understood the need for this direction. And if memory serves me right, at 
that time I was seeing twice the amount of cache flushing !

> But the real fix of my problem is:
> ---------------------------------------->8------------------------------------
> --- a/lib/dma-noncoherent.c
> +++ b/lib/dma-noncoherent.c
> @@ -35,7 +35,7 @@ static dma_addr_t dma_noncoherent_map_page(struct device *dev, struct page *page
>   
>          addr = dma_direct_map_page(dev, page, offset, size, dir, attrs);
>          if (!dma_mapping_error(dev, addr) && !(attrs & DMA_ATTR_SKIP_CPU_SYNC))
> -               arch_sync_dma_for_device(dev, page_to_phys(page), size, dir);
> +               arch_sync_dma_for_device(dev, page_to_phys(page) + offset, size, dir);
>          return addr;
>   }
> ---------------------------------------->8------------------------------------
>
> You seem to lost an offset in the page so if we happen to have a buffer not aligned to
> a page boundary then we were obviously corrupting data outside our data :)

Neat !

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation
  2018-05-18 13:27     ` hch
  2018-05-18 14:13       ` Alexey Brodkin
@ 2018-05-18 17:28       ` Vineet Gupta
  1 sibling, 0 replies; 39+ messages in thread
From: Vineet Gupta @ 2018-05-18 17:28 UTC (permalink / raw)
  To: hch, Alexey Brodkin
  Cc: linux-arch, linux-xtensa, monstr, linux-snps-arc, linux-c6x-dev,
	linux-parisc, linux-sh, linux-hexagon, linux-kernel, iommu,
	linux-m68k, openrisc, green.hu, linux-alpha, sparclinux,
	nios2-dev, deanbo422, linux-arm-kernel, linux-mm, Andrew Morton

On 05/18/2018 06:23 AM, hch@lst.de wrote:
>   Fri, May 18, 2018 at 01:03:46PM +0000, Alexey Brodkin wrote:
>> Note mmc_get_dma_dir() is just "data->flags & MMC_DATA_WRITE ? DMA_TO_DEVICE : DMA_FROM_DEVICE".
>> I.e. if we're preparing for sending data dma_noncoherent_map_sg() will have DMA_TO_DEVICE which
>> is quite OK for passing to dma_noncoherent_sync_sg_for_device() but in case of reading we'll have
>> DMA_FROM_DEVICE which we'll pass to dma_noncoherent_sync_sg_for_device() in dma_noncoherent_map_sg().
>>
>> I'd say this is not entirely correct because IMHO arch_sync_dma_for_cpu() is supposed to only be used
>> in case of DMA_FROM_DEVICE and arch_sync_dma_for_device() only in case of DMA_TO_DEVICE.
> arc overrides the dir paramter of the dma_sync_single_for_device/
> dma_sync_single_for_cpu calls.  My patches dropped that, and I have
> restored that, and audit for the other architectures is pending.

Right, for now lets retain that and do a sweeping audit of @direction - to me it 
seems extraneous (as it did 10 years ago), but I'm not an expert in this are so 
perhaps it is needed for some device / arches and it would be good to understand 
that finally.

> That being said the existing arc code still looks rather odd as it
> didn't do the same thing for the scatterlist versions of the calls.
> I've thrown in a few patches into my new tree to make the sg versions
> make the normal calls, and to clean up the area a bit.

Not calling names or anything here, but it doesn't exist for sg variants, because 
I didn't write that code :-)
It was introduced by your commi:

2016-01-20 052c96dbe33b arc: convert to dma_map_ops

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: dma_sync_*_for_cpu and direction=TO_DEVICE (was Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation)
  2018-05-18 17:20     ` dma_sync_*_for_cpu and direction=TO_DEVICE (was Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation) Vineet Gupta
@ 2018-05-18 17:50       ` Russell King - ARM Linux
  2018-05-18 19:57         ` Alexey Brodkin
  2018-05-18 20:35         ` Vineet Gupta
  0 siblings, 2 replies; 39+ messages in thread
From: Russell King - ARM Linux @ 2018-05-18 17:50 UTC (permalink / raw)
  To: Vineet Gupta
  Cc: Alexey Brodkin, hch, linux-arch, linux-xtensa, monstr, deanbo422,
	linux-c6x-dev, linux-parisc, linux-sh, linux-m68k, linux-hexagon,
	linux-kernel, linux-mm, iommu, openrisc, green.hu, linux-alpha,
	sparclinux, nios2-dev, Andrew Morton, linux-snps-arc,
	linux-arm-kernel

On Fri, May 18, 2018 at 10:20:02AM -0700, Vineet Gupta wrote:
> I never understood the need for this direction. And if memory serves me
> right, at that time I was seeing twice the amount of cache flushing !

It's necessary.  Take a moment to think carefully about this:

	dma_map_single(, dir)

	dma_sync_single_for_cpu(, dir)

	dma_sync_single_for_device(, dir)

	dma_unmap_single(, dir)

In the case of a DMA-incoherent architecture, the operations done at each
stage depend on the direction argument:

	map		for_cpu		for_device	unmap
TO_DEV	writeback	none		writeback	none
TO_CPU	invalidate	invalidate*	invalidate	invalidate*
BIDIR	writeback	invalidate	writeback	invalidate

* - only necessary if the CPU speculatively prefetches.

The multiple invalidations for the TO_CPU case handles different
conditions that can result in data corruption, and for some CPUs, all
four are necessary.

This is what is implemented for 32-bit ARM, depending on the CPU
capabilities, as we have DMA incoherent devices and we have CPUs that
speculatively prefetch data, and so may load data into the caches while
DMA is in operation.


Things get more interesting if the implementation behind the DMA API has
to copy data between the buffer supplied to the mapping and some DMA
accessible buffer:

	map		for_cpu		for_device	unmap
TO_DEV	copy to dma	none		copy to dma	none
TO_CPU	none		copy to cpu	none		copy to cpu
BIDIR	copy to dma	copy to cpu	copy to dma	copy to cpu

So, in both cases, the value of the direction argument defines what you
need to do in each call.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 8.8Mbps down 630kbps up
According to speedtest.net: 8.21Mbps down 510kbps up

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: dma_sync_*_for_cpu and direction=TO_DEVICE (was Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation)
  2018-05-18 17:50       ` Russell King - ARM Linux
@ 2018-05-18 19:57         ` Alexey Brodkin
  2018-05-18 21:33           ` Russell King - ARM Linux
  2018-05-18 20:35         ` Vineet Gupta
  1 sibling, 1 reply; 39+ messages in thread
From: Alexey Brodkin @ 2018-05-18 19:57 UTC (permalink / raw)
  To: linux
  Cc: deanbo422, linux-sh, linux-kernel, nios2-dev, linux-xtensa,
	linux-m68k, linux-mm, linux-hexagon, hch, linux-alpha,
	linux-snps-arc, iommu, green.hu, Vineet.Gupta1, akpm, openrisc,
	linux-arm-kernel, monstr, linux-parisc, linux-c6x-dev,
	linux-arch, sparclinux

Hi Russel,

On Fri, 2018-05-18 at 18:50 +0100, Russell King - ARM Linux wrote:
> It's necessary.  Take a moment to think carefully about this:
> 
>         dma_map_single(, dir)
> 
>         dma_sync_single_for_cpu(, dir)
> 
>         dma_sync_single_for_device(, dir)
> 
>         dma_unmap_single(, dir)
> 
> In the case of a DMA-incoherent architecture, the operations done at each
> stage depend on the direction argument:
> 
>         map             for_cpu         for_device      unmap
> TO_DEV  writeback       none            writeback       none
> TO_CPU  invalidate      invalidate*     invalidate      invalidate*
> BIDIR   writeback       invalidate      writeback       invalidate
> 
> * - only necessary if the CPU speculatively prefetches.

I think invalidation of DMA buffer is required on for_cpu(TO_CPU) even
if CPU doesn't preferch - what if we reuse the same buffer for multiple
reads from DMA device?

> The multiple invalidations for the TO_CPU case handles different
> conditions that can result in data corruption, and for some CPUs, all
> four are necessary.

I would agree that map()/unmap() a quite a special cases and so depending
on direction we need to execute in them either for_cpu() or for_device()
call-backs depending on direction.

As for invalidation in case of for_device(TO_CPU) I still don't see
a rationale behind it. Would be interesting to see a real example where
we benefit from this.

> This is what is implemented for 32-bit ARM, depending on the CPU
> capabilities, as we have DMA incoherent devices and we have CPUs that
> speculatively prefetch data, and so may load data into the caches while
> DMA is in operation.
> 
> 
> Things get more interesting if the implementation behind the DMA API has
> to copy data between the buffer supplied to the mapping and some DMA
> accessible buffer:
> 
>         map             for_cpu         for_device      unmap
> TO_DEV  copy to dma     none            copy to dma     none
> TO_CPU  none            copy to cpu     none            copy to cpu
> BIDIR   copy to dma     copy to cpu     copy to dma     copy to cpu
> 
> So, in both cases, the value of the direction argument defines what you
> need to do in each call.

Interesting enough in your seond table (which describes more complicated
case indeed) you set "none" for for_device(TO_CPU) which looks logical
to me.

So IMHO that's what make sense:
---------------------------->8-----------------------------
        map             for_cpu         for_device      unmap
TO_DEV  writeback       none            writeback       none
TO_CPU  none            invalidate      none            invalidate*
BIDIR   writeback       invalidate      writeback       invalidate*
---------------------------->8-----------------------------

* is the case for prefetching CPU.

-Alexey

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation
  2018-05-18 13:03   ` Alexey Brodkin
  2018-05-18 13:27     ` hch
  2018-05-18 17:20     ` dma_sync_*_for_cpu and direction=TO_DEVICE (was Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation) Vineet Gupta
@ 2018-05-18 20:05     ` Helge Deller
  2018-05-19  6:38       ` hch
  2 siblings, 1 reply; 39+ messages in thread
From: Helge Deller @ 2018-05-18 20:05 UTC (permalink / raw)
  To: Alexey Brodkin, hch
  Cc: deanbo422, linux-sh, linux-kernel, nios2-dev, linux-m68k,
	linux-alpha, linux-hexagon, linux-snps-arc, iommu, green.hu,
	openrisc, linux-arm-kernel, monstr, linux-parisc, linux-c6x-dev,
	linux-arch, sparclinux, James Bottomley

On 18.05.2018 15:03, Alexey Brodkin wrote:
> But the real fix of my problem is:
> ---------------------------------------->8------------------------------------
> --- a/lib/dma-noncoherent.c
> +++ b/lib/dma-noncoherent.c
> @@ -35,7 +35,7 @@ static dma_addr_t dma_noncoherent_map_page(struct device *dev, struct page *page
>  
>         addr = dma_direct_map_page(dev, page, offset, size, dir, attrs);
>         if (!dma_mapping_error(dev, addr) && !(attrs & DMA_ATTR_SKIP_CPU_SYNC))
> -               arch_sync_dma_for_device(dev, page_to_phys(page), size, dir);
> +               arch_sync_dma_for_device(dev, page_to_phys(page) + offset, size, dir);
>         return addr;
>  }
> ---------------------------------------->8------------------------------------
> 
> You seem to lost an offset in the page so if we happen to have a buffer not aligned to
> a page boundary then we were obviously corrupting data outside our data :)

Good.
This patch seems to fix the dma issues I faced on my 32bit B160L parisc box.

So it leaves only one open issue on parisc:
Now every 32 bit parisc system is unnecessarily non-coherent.

Helge

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: dma_sync_*_for_cpu and direction=TO_DEVICE (was Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation)
  2018-05-18 17:50       ` Russell King - ARM Linux
  2018-05-18 19:57         ` Alexey Brodkin
@ 2018-05-18 20:35         ` Vineet Gupta
  2018-05-18 21:55           ` Russell King - ARM Linux
  1 sibling, 1 reply; 39+ messages in thread
From: Vineet Gupta @ 2018-05-18 20:35 UTC (permalink / raw)
  To: Russell King - ARM Linux
  Cc: Alexey Brodkin, hch, linux-arch, linux-xtensa, monstr, deanbo422,
	linux-c6x-dev, linux-parisc, linux-sh,
	linux-m68k@lists.linux-m68k.org, linux-hexagon, linux-kernel,
	linux-mm, iommu, openrisc, green.hu, linux-alpha, sparclinux,
	nios2-dev, Andrew Morton, linux-snps-arc, linux-arm-kernel

On 05/18/2018 10:50 AM, Russell King - ARM Linux wrote:
> On Fri, May 18, 2018 at 10:20:02AM -0700, Vineet Gupta wrote:
>> I never understood the need for this direction. And if memory serves me
>> right, at that time I was seeing twice the amount of cache flushing !
> It's necessary.  Take a moment to think carefully about this:
>
> 	dma_map_single(, dir)
>
> 	dma_sync_single_for_cpu(, dir)
>
> 	dma_sync_single_for_device(, dir)
>
> 	dma_unmap_single(, dir)

As an aside, do these imply a state machine of sorts - does a driver needs to 
always call map_single first ?

My original point of contention/confusion is the specific combinations of API and 
direction, specifically for_cpu(TO_DEV) and for_device(TO_CPU)

Semantically what does dma_sync_single_for_cpu(TO_DEV) even imply for a non dma 
coherent arch.
Your tables below have "none" for both, implying it is unlikely to be a real 
combination (for ARM and ARC atleast).

The other case, actually @dir TO_CPU, independent of for_{cpu, device}  implies 
driver intends to touch it after the call, so it would invalidate any stray lines, 
unconditionally (and not just for speculative prefetch case).


> In the case of a DMA-incoherent architecture, the operations done at each
> stage depend on the direction argument:
>
> 	map		for_cpu		for_device	unmap
> TO_DEV	writeback	none		writeback	none
> TO_CPU	invalidate	invalidate*	invalidate	invalidate*
> BIDIR	writeback	invalidate	writeback	invalidate
>
> * - only necessary if the CPU speculatively prefetches.
>
> The multiple invalidations for the TO_CPU case handles different
> conditions that can result in data corruption, and for some CPUs, all
> four are necessary.

Can you please explain in some more detail, TO_CPU row, why invalidate is 
conditional sometimes.

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: dma_sync_*_for_cpu and direction=TO_DEVICE (was Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation)
  2018-05-18 19:57         ` Alexey Brodkin
@ 2018-05-18 21:33           ` Russell King - ARM Linux
  0 siblings, 0 replies; 39+ messages in thread
From: Russell King - ARM Linux @ 2018-05-18 21:33 UTC (permalink / raw)
  To: Alexey Brodkin
  Cc: deanbo422, linux-sh, linux-kernel, nios2-dev, linux-xtensa,
	linux-m68k, linux-mm, linux-hexagon, hch, linux-alpha,
	linux-snps-arc, iommu, green.hu, Vineet.Gupta1, akpm, openrisc,
	linux-arm-kernel, monstr, linux-parisc, linux-c6x-dev,
	linux-arch, sparclinux

On Fri, May 18, 2018 at 07:57:34PM +0000, Alexey Brodkin wrote:
> Hi Russel,

That's Russell.

> On Fri, 2018-05-18 at 18:50 +0100, Russell King - ARM Linux wrote:
> > It's necessary.  Take a moment to think carefully about this:
> > 
> >         dma_map_single(, dir)
> > 
> >         dma_sync_single_for_cpu(, dir)
> > 
> >         dma_sync_single_for_device(, dir)
> > 
> >         dma_unmap_single(, dir)
> > 
> > In the case of a DMA-incoherent architecture, the operations done at each
> > stage depend on the direction argument:
> > 
> >         map             for_cpu         for_device      unmap
> > TO_DEV  writeback       none            writeback       none
> > TO_CPU  invalidate      invalidate*     invalidate      invalidate*
> > BIDIR   writeback       invalidate      writeback       invalidate
> > 
> > * - only necessary if the CPU speculatively prefetches.
> 
> I think invalidation of DMA buffer is required on for_cpu(TO_CPU) even
> if CPU doesn't preferch - what if we reuse the same buffer for multiple
> reads from DMA device?

That's fine - for non-coherent DMA, the CPU caches will only end up
containing data for that memory if:
- the CPU speculatively fetches data from that memory, or
- the CPU explicitly touches that memory

> > The multiple invalidations for the TO_CPU case handles different
> > conditions that can result in data corruption, and for some CPUs, all
> > four are necessary.
> 
> I would agree that map()/unmap() a quite a special cases and so depending
> on direction we need to execute in them either for_cpu() or for_device()
> call-backs depending on direction.
> 
> As for invalidation in case of for_device(TO_CPU) I still don't see
> a rationale behind it. Would be interesting to see a real example where
> we benefit from this.

Yes, you could avoid that, but depending how you structure the
architecture implementation, it can turn out to be a corner case.
The above table is precisely how 32-bit ARM is implemented, because
the way we implement them is based on who owns the memory - the "map"
and "for_device" operations translate internally to a cpu-to-device
ownership transition of the buffer.  Similar for "unmap" and "to_cpu".
It basically avoids having to add additional functions at the lower
implementation levels.

> > Things get more interesting if the implementation behind the DMA API has
> > to copy data between the buffer supplied to the mapping and some DMA
> > accessible buffer:
> > 
> >         map             for_cpu         for_device      unmap
> > TO_DEV  copy to dma     none            copy to dma     none
> > TO_CPU  none            copy to cpu     none            copy to cpu
> > BIDIR   copy to dma     copy to cpu     copy to dma     copy to cpu
> > 
> > So, in both cases, the value of the direction argument defines what you
> > need to do in each call.
> 
> Interesting enough in your seond table (which describes more complicated
> case indeed) you set "none" for for_device(TO_CPU) which looks logical
> to me.
> 
> So IMHO that's what make sense:
> ---------------------------->8-----------------------------
>         map             for_cpu         for_device      unmap
> TO_DEV  writeback       none            writeback       none
> TO_CPU  none            invalidate      none            invalidate*
> BIDIR   writeback       invalidate      writeback       invalidate*
> ---------------------------->8-----------------------------

That doesn't make sense for the TO_CPU case.  If the caches contain
dirty cache lines, and you're DMAing from the device to the system
RAM, other system activity can cause the dirty cache lines to be
evicted (written back) to memory which the DMA has already overwritten.
The result is data corruption.  So, you really can't have "none" in
the "map" case there.

Given that, the "for_cpu" case becomes dependent on whether the CPU
speculatively prefetches.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 8.8Mbps down 630kbps up
According to speedtest.net: 8.21Mbps down 510kbps up

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: dma_sync_*_for_cpu and direction=TO_DEVICE (was Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation)
  2018-05-18 20:35         ` Vineet Gupta
@ 2018-05-18 21:55           ` Russell King - ARM Linux
  0 siblings, 0 replies; 39+ messages in thread
From: Russell King - ARM Linux @ 2018-05-18 21:55 UTC (permalink / raw)
  To: Vineet Gupta
  Cc: Alexey Brodkin, hch, linux-arch, linux-xtensa, monstr, deanbo422,
	linux-c6x-dev, linux-parisc, linux-sh, linux-m68k, linux-hexagon,
	linux-kernel, linux-mm, iommu, openrisc, green.hu, linux-alpha,
	sparclinux, nios2-dev, Andrew Morton, linux-snps-arc,
	linux-arm-kernel

On Fri, May 18, 2018 at 01:35:08PM -0700, Vineet Gupta wrote:
> On 05/18/2018 10:50 AM, Russell King - ARM Linux wrote:
> >On Fri, May 18, 2018 at 10:20:02AM -0700, Vineet Gupta wrote:
> >>I never understood the need for this direction. And if memory serves me
> >>right, at that time I was seeing twice the amount of cache flushing !
> >It's necessary.  Take a moment to think carefully about this:
> >
> >	dma_map_single(, dir)
> >
> >	dma_sync_single_for_cpu(, dir)
> >
> >	dma_sync_single_for_device(, dir)
> >
> >	dma_unmap_single(, dir)
> 
> As an aside, do these imply a state machine of sorts - does a driver needs
> to always call map_single first ?

Kind-of, but some drivers do omit some of the dma_sync_*() calls.
For example, if a buffer is written to, then mapped with TO_DEVICE,
and then the CPU wishes to write to it, it's fairly common that a
driver omits the dma_sync_single_for_cpu() call.  If you think about
the cases I gave and what cache operations happen, such a scenario
practically turns out to be safe.

> My original point of contention/confusion is the specific combinations of
> API and direction, specifically for_cpu(TO_DEV) and for_device(TO_CPU)

Remember that it is expected that all calls for a mapping use the
same direction argument while that mapping exists.  In other words,
if you call dma_map_single(TO_DEVICE) and then use any of the other
functions, the other functions will also use TO_DEVICE.  The DMA
direction argument describes the direction of the DMA operation
being performed on the buffer, not on the individual dma_* operation.

What isn't expected at arch level is for drivers to do:

	dma_map_single(TO_DEVICE)
	dma_sync_single_for_cpu(FROM_DEVICE)

or vice versa.

> Semantically what does dma_sync_single_for_cpu(TO_DEV) even imply for a non
> dma coherent arch.
> 
> Your tables below have "none" for both, implying it is unlikely to be a real
> combination (for ARM and ARC atleast).

Very little for the cases that I've stated (and as I mentioned
above, some drivers do omit the call in that case.)

> The other case, actually @dir TO_CPU, independent of for_{cpu, device} 
> implies driver intends to touch it after the call, so it would invalidate
> any stray lines, unconditionally (and not just for speculative prefetch
> case).

If you don't have a CPU that speculatively prefetches, and you've
already had to invalidate the cache lines (to avoid write-backs
corrupting DMA'd data) then there's no need for the architecture
to do any work at the for_cpu(TO_CPU) case - the CPU shouldn't
be touching cache lines that are part of the buffer while it is
mapped, which means a non-speculating CPU won't pull in any
cache lines without an explicit access.

Speculating CPUs are different.  The action of the speculation is
to try and guess what data the program wants to access ahead of
the program flow.  That causes the CPU to prefetch data into the
cache.  The point in the program flow that this happens is not
really determinant to the programmer.  This means that if you try
to read from the DMA buffer after the DMA operation has complete
without invalidating the cache between the DMA completing and the
CPU reading, you have no guarantee that you're reading the data
that the DMA operation has been written.  The cache may have
loaded itself with data before the DMA operation completed, and
the CPU may see that stale data.

The difference between non-speculating CPUs and speculating CPUs
is that for non-speculating CPUs, caches work according to explicit
accesses by the program, and the program is stalled while the data
is fetched from external memory.  Speculating CPUs try to predict
ahead of time what data the program will require in the future,
and attempt to load that data into the caches _before_ the program
requires it - which means that the program suffers fewer stalls.

> >In the case of a DMA-incoherent architecture, the operations done at each
> >stage depend on the direction argument:
> >
> >	map		for_cpu		for_device	unmap
> >TO_DEV	writeback	none		writeback	none
> >TO_CPU	invalidate	invalidate*	invalidate	invalidate*
> >BIDIR	writeback	invalidate	writeback	invalidate
> >
> >* - only necessary if the CPU speculatively prefetches.
> >
> >The multiple invalidations for the TO_CPU case handles different
> >conditions that can result in data corruption, and for some CPUs, all
> >four are necessary.
> 
> Can you please explain in some more detail, TO_CPU row, why invalidate is
> conditional sometimes.

See above - I hope my explanation above is sufficient.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 8.8Mbps down 630kbps up
According to speedtest.net: 8.21Mbps down 510kbps up

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation
  2018-05-18 20:05     ` [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation Helge Deller
@ 2018-05-19  6:38       ` hch
  0 siblings, 0 replies; 39+ messages in thread
From: hch @ 2018-05-19  6:38 UTC (permalink / raw)
  To: Helge Deller
  Cc: Alexey Brodkin, hch, deanbo422, linux-sh, linux-kernel,
	nios2-dev, linux-m68k, linux-alpha, linux-hexagon,
	linux-snps-arc, iommu, green.hu, openrisc, linux-arm-kernel,
	monstr, linux-parisc, linux-c6x-dev, linux-arch, sparclinux,
	James Bottomley

On Fri, May 18, 2018 at 10:05:51PM +0200, Helge Deller wrote:
> This patch seems to fix the dma issues I faced on my 32bit B160L parisc box.
> 
> So it leaves only one open issue on parisc:
> Now every 32 bit parisc system is unnecessarily non-coherent.

I diagree with those comments, let me resend the refactored patch
to make it more clear.

^ permalink raw reply	[flat|nested] 39+ messages in thread

* Re: [PATCH 04/20] arm-nommu: use generic dma_noncoherent_ops
  2018-05-11  9:11   ` Russell King - ARM Linux
@ 2018-05-22 11:53     ` Christoph Hellwig
  0 siblings, 0 replies; 39+ messages in thread
From: Christoph Hellwig @ 2018-05-22 11:53 UTC (permalink / raw)
  To: Russell King - ARM Linux
  Cc: Christoph Hellwig, iommu, linux-arch, linux-xtensa, Michal Simek,
	Vincent Chen, linux-c6x-dev, linux-parisc, linux-sh,
	linux-hexagon, linux-kernel, linux-m68k, openrisc, Greentime Hu,
	linux-alpha, sparclinux, nios2-dev, linux-snps-arc,
	linux-arm-kernel

On Fri, May 11, 2018 at 10:11:15AM +0100, Russell King - ARM Linux wrote:
> > +void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
> > +		size_t size, enum dma_data_direction dir)
> 
> Please no.  There is a lot of history of these (__dma_page_cpu_to_dev etc)
> functions being abused by out of tree drivers, because they think they
> know better.  This is stopped by making them static and ensuring that
> drivers have no access to these functions.
> 
> Please do not re-expose these to the global kernel.
> 
> While it may make things easier for a cross-architecture point of view,
> it makes it a lot easier for people to abuse these private APIs.

The point of this series, which isn't fully archived yet, is to
consolidate the direct mapping dma code, that is all dma_map_ops
instances except for iommus.  It is in fact in many ways modelled
after the ARM code.  For that we need the architectures to supply
the cache maintainance routines.

However, even if they now appear in dma-noncoherent.h they are NOT
and will NOT be exported, so using them directly from drivers won't
easily be possible.

I'll drop the arm-nommu patch for now, but I will pester your with this
again once arm is the last architecture not sharing the common code.

> 
> -- 
> RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
> FTTC broadband for 0.8mile line in suburbia: sync at 8.8Mbps down 630kbps up
> According to speedtest.net: 8.21Mbps down 510kbps up
---end quoted text---

^ permalink raw reply	[flat|nested] 39+ messages in thread

end of thread, other threads:[~2018-05-22 11:48 UTC | newest]

Thread overview: 39+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-05-11  7:59 common non-cache coherent direct dma mapping ops Christoph Hellwig
2018-05-11  7:59 ` [PATCH 01/20] dma-mapping: simplify Kconfig dependencies Christoph Hellwig
2018-05-11  7:59 ` [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation Christoph Hellwig
2018-05-18 13:03   ` Alexey Brodkin
2018-05-18 13:27     ` hch
2018-05-18 14:13       ` Alexey Brodkin
2018-05-18 17:28       ` Vineet Gupta
2018-05-18 17:20     ` dma_sync_*_for_cpu and direction=TO_DEVICE (was Re: [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation) Vineet Gupta
2018-05-18 17:50       ` Russell King - ARM Linux
2018-05-18 19:57         ` Alexey Brodkin
2018-05-18 21:33           ` Russell King - ARM Linux
2018-05-18 20:35         ` Vineet Gupta
2018-05-18 21:55           ` Russell King - ARM Linux
2018-05-18 20:05     ` [PATCH 02/20] dma-mapping: provide a generic dma-noncoherent implementation Helge Deller
2018-05-19  6:38       ` hch
2018-05-11  7:59 ` [PATCH 03/20] arc: use generic dma_noncoherent_ops Christoph Hellwig
2018-05-11 12:44   ` Alexey Brodkin
2018-05-11  7:59 ` [PATCH 04/20] arm-nommu: " Christoph Hellwig
2018-05-11  9:11   ` Russell King - ARM Linux
2018-05-22 11:53     ` Christoph Hellwig
2018-05-11 13:56   ` John Garry
2018-05-11  7:59 ` [PATCH 05/20] c6x: " Christoph Hellwig
2018-05-15  0:25   ` [Linux-c6x-dev] " Mark Salter
2018-05-11  7:59 ` [PATCH 06/20] hexagon: " Christoph Hellwig
2018-05-11  7:59 ` [PATCH 07/20] m68k: " Christoph Hellwig
2018-05-11  7:59 ` [PATCH 08/20] microblaze: " Christoph Hellwig
2018-05-11  7:59 ` [PATCH 09/20] microblaze: remove the consistent_sync and consistent_sync_page Christoph Hellwig
2018-05-11  7:59 ` [PATCH 10/20] nds32: use generic dma_noncoherent_ops Christoph Hellwig
2018-05-11  7:59 ` [PATCH 11/20] nios2: " Christoph Hellwig
2018-05-11  7:59 ` [PATCH 12/20] openrisc: " Christoph Hellwig
2018-05-11  7:59 ` [PATCH 13/20] sh: simplify get_arch_dma_ops Christoph Hellwig
2018-05-11  7:59 ` [PATCH 14/20] sh: introduce a sh_cacheop_vaddr helper Christoph Hellwig
2018-05-11  7:59 ` [PATCH 15/20] sh: use dma_direct_ops for the CONFIG_DMA_COHERENT case Christoph Hellwig
2018-05-11  7:59 ` [PATCH 16/20] mm: split arch/sh/mm/consistent.c Christoph Hellwig
2018-05-11  7:59 ` [PATCH 17/20] sh: use generic dma_noncoherent_ops Christoph Hellwig
2018-05-11  7:59 ` [PATCH 18/20] xtensa: " Christoph Hellwig
2018-05-11  7:59 ` [PATCH 19/20] sparc: " Christoph Hellwig
2018-05-11  7:59 ` [PATCH 20/20] parisc: " Christoph Hellwig
2018-05-13 13:26 ` common non-cache coherent direct dma mapping ops Helge Deller

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).