[PATCH] sparc: use generic dma_noncoherent_ops

[PATCH] sparc: use generic dma_noncoherent_ops

[Christoph Hellwig]

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 0f535debf802..79f29c67291a 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 cca9134cfa7d..960c1bc22c2e 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);
 
-- 
2.17.1

[PATCH] sparc: use generic dma_noncoherent_ops

[Christoph Hellwig]

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>
Acked-by: Sam Ravnborg <sam@ravnborg.org>
---
 Makefile                             |   2 +-
 arch/sparc/Kconfig                   |   2 +
 arch/sparc/include/asm/dma-mapping.h |   5 +-
 arch/sparc/kernel/ioport.c           | 193 +++++----------------------
 4 files changed, 36 insertions(+), 166 deletions(-)

diff --git a/Makefile b/Makefile
index 85f3481a56d6..8a3fd0c4a76e 100644
--- a/Makefile
+++ b/Makefile
@@ -368,7 +368,7 @@ HOST_LOADLIBES := $(HOST_LFS_LIBS)
 # Make variables (CC, etc...)
 AS		= $(CROSS_COMPILE)as
 LD		= $(CROSS_COMPILE)ld
-CC		= $(CROSS_COMPILE)gcc
+CC		= sparc64-linux-gnu-gcc-8
 CPP		= $(CC) -E
 AR		= $(CROSS_COMPILE)ar
 NM		= $(CROSS_COMPILE)nm
diff --git a/arch/sparc/Kconfig b/arch/sparc/Kconfig
index 0f535debf802..79f29c67291a 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 cca9134cfa7d..6799c93c9f27 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>
@@ -434,42 +435,41 @@ 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 size, dma_addr_t *dma_handle,
+		gfp_t gfp, unsigned long attrs)
 {
-	unsigned long len_total = PAGE_ALIGN(len);
+	unsigned long len_total = PAGE_ALIGN(size);
 	void *va;
 	struct resource *res;
 	int order;
 
-	if (len = 0) {
+	if (size = 0) {
 		return NULL;
 	}
-	if (len > 256*1024) {			/* __get_free_pages() limit */
+	if (size > 256*1024) {			/* __get_free_pages() limit */
 		return NULL;
 	}
 
 	order = get_order(len_total);
 	va = (void *) __get_free_pages(gfp, order);
 	if (va = NULL) {
-		printk("pci_alloc_consistent: no %ld pages\n", len_total>>PAGE_SHIFT);
+		printk("%s: no %ld pages\n", __func__, len_total>>PAGE_SHIFT);
 		goto err_nopages;
 	}
 
 	if ((res = kzalloc(sizeof(struct resource), GFP_KERNEL)) = NULL) {
-		printk("pci_alloc_consistent: no core\n");
+		printk("%s: no core\n", __func__);
 		goto err_nomem;
 	}
 
 	if (allocate_resource(&_sparc_dvma, res, len_total,
 	    _sparc_dvma.start, _sparc_dvma.end, PAGE_SIZE, NULL, NULL) != 0) {
-		printk("pci_alloc_consistent: cannot occupy 0x%lx", len_total);
+		printk("%s: cannot occupy 0x%lx", __func__, len_total);
 		goto err_nova;
 	}
 	srmmu_mapiorange(0, virt_to_phys(va), res->start, len_total);
 
-	*pba = virt_to_phys(va); /* equals virt_to_bus (R.I.P.) for us. */
+	*dma_handle = virt_to_phys(va);
 	return (void *) res->start;
 
 err_nova:
@@ -481,184 +481,53 @@ static void *pci32_alloc_coherent(struct device *dev, size_t len,
 }
 
 /* Free and unmap a consistent DMA buffer.
- * cpu_addr is what was returned from pci_alloc_consistent,
- * size must be the same as what as passed into pci_alloc_consistent,
- * and likewise dma_addr must be the same as what *dma_addrp was set to.
+ * cpu_addr is what was returned arch_dma_alloc, size must be the same as what
+ * was passed into arch_dma_alloc, and likewise dma_addr must be the same as
+ * what *dma_ndler was set to.
  *
  * 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 size, void *cpu_addr,
+		dma_addr_t dma_addr, unsigned long attrs)
 {
 	struct resource *res;
 
 	if ((res = lookup_resource(&_sparc_dvma,
-	    (unsigned long)p)) = NULL) {
-		printk("pci_free_consistent: cannot free %p\n", p);
+	    (unsigned long)cpu_addr)) = NULL) {
+		printk("%s: cannot free %p\n", __func__, cpu_addr);
 		return;
 	}
 
-	if (((unsigned long)p & (PAGE_SIZE-1)) != 0) {
-		printk("pci_free_consistent: unaligned va %p\n", p);
+	if (((unsigned long)cpu_addr & (PAGE_SIZE-1)) != 0) {
+		printk("%s: unaligned va %p\n", __func__, cpu_addr);
 		return;
 	}
 
-	n = PAGE_ALIGN(n);
-	if (resource_size(res) != n) {
-		printk("pci_free_consistent: region 0x%lx asked 0x%lx\n",
-		    (long)resource_size(res), (long)n);
+	size = PAGE_ALIGN(size);
+	if (resource_size(res) != size) {
+		printk("%s: region 0x%lx asked 0x%zx\n", __func__,
+		    (long)resource_size(res), size);
 		return;
 	}
 
-	dma_make_coherent(ba, n);
-	srmmu_unmapiorange((unsigned long)p, n);
+	dma_make_coherent(dma_addr, size);
+	srmmu_unmapiorange((unsigned long)cpu_addr, size);
 
 	release_resource(res);
 	kfree(res);
-	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));
-		}
-	}
-}
-
-/* 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)
-{
-	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));
-	}
+	free_pages((unsigned long)phys_to_virt(dma_addr), get_order(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));
-		}
-	}
-}
+/* IIep is write-through, not flushing on cpu to device transfer. */
 
-static void pci32_sync_sg_for_device(struct device *device, struct scatterlist *sgl,
-				     int nents, 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)
 {
-	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));
-		}
-	}
+	if (dir != PCI_DMA_TODEVICE)
+		dma_make_coherent(paddr, PAGE_ALIGN(size));
 }
 
-/* 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);
 
-- 
2.18.0

[PATCH] sparc: use generic dma_noncoherent_ops

[Christoph Hellwig]

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-jcswGhMUV9g@public.gmane.org>
Acked-by: Sam Ravnborg <sam-uyr5N9Q2VtJg9hUCZPvPmw@public.gmane.org>
---
 Makefile                             |   2 +-
 arch/sparc/Kconfig                   |   2 +
 arch/sparc/include/asm/dma-mapping.h |   5 +-
 arch/sparc/kernel/ioport.c           | 193 +++++----------------------
 4 files changed, 36 insertions(+), 166 deletions(-)

diff --git a/Makefile b/Makefile
index 85f3481a56d6..8a3fd0c4a76e 100644
--- a/Makefile
+++ b/Makefile
@@ -368,7 +368,7 @@ HOST_LOADLIBES := $(HOST_LFS_LIBS)
 # Make variables (CC, etc...)
 AS		= $(CROSS_COMPILE)as
 LD		= $(CROSS_COMPILE)ld
-CC		= $(CROSS_COMPILE)gcc
+CC		= sparc64-linux-gnu-gcc-8
 CPP		= $(CC) -E
 AR		= $(CROSS_COMPILE)ar
 NM		= $(CROSS_COMPILE)nm
diff --git a/arch/sparc/Kconfig b/arch/sparc/Kconfig
index 0f535debf802..79f29c67291a 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 cca9134cfa7d..6799c93c9f27 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>
@@ -434,42 +435,41 @@ 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 size, dma_addr_t *dma_handle,
+		gfp_t gfp, unsigned long attrs)
 {
-	unsigned long len_total = PAGE_ALIGN(len);
+	unsigned long len_total = PAGE_ALIGN(size);
 	void *va;
 	struct resource *res;
 	int order;
 
-	if (len == 0) {
+	if (size == 0) {
 		return NULL;
 	}
-	if (len > 256*1024) {			/* __get_free_pages() limit */
+	if (size > 256*1024) {			/* __get_free_pages() limit */
 		return NULL;
 	}
 
 	order = get_order(len_total);
 	va = (void *) __get_free_pages(gfp, order);
 	if (va == NULL) {
-		printk("pci_alloc_consistent: no %ld pages\n", len_total>>PAGE_SHIFT);
+		printk("%s: no %ld pages\n", __func__, len_total>>PAGE_SHIFT);
 		goto err_nopages;
 	}
 
 	if ((res = kzalloc(sizeof(struct resource), GFP_KERNEL)) == NULL) {
-		printk("pci_alloc_consistent: no core\n");
+		printk("%s: no core\n", __func__);
 		goto err_nomem;
 	}
 
 	if (allocate_resource(&_sparc_dvma, res, len_total,
 	    _sparc_dvma.start, _sparc_dvma.end, PAGE_SIZE, NULL, NULL) != 0) {
-		printk("pci_alloc_consistent: cannot occupy 0x%lx", len_total);
+		printk("%s: cannot occupy 0x%lx", __func__, len_total);
 		goto err_nova;
 	}
 	srmmu_mapiorange(0, virt_to_phys(va), res->start, len_total);
 
-	*pba = virt_to_phys(va); /* equals virt_to_bus (R.I.P.) for us. */
+	*dma_handle = virt_to_phys(va);
 	return (void *) res->start;
 
 err_nova:
@@ -481,184 +481,53 @@ static void *pci32_alloc_coherent(struct device *dev, size_t len,
 }
 
 /* Free and unmap a consistent DMA buffer.
- * cpu_addr is what was returned from pci_alloc_consistent,
- * size must be the same as what as passed into pci_alloc_consistent,
- * and likewise dma_addr must be the same as what *dma_addrp was set to.
+ * cpu_addr is what was returned arch_dma_alloc, size must be the same as what
+ * was passed into arch_dma_alloc, and likewise dma_addr must be the same as
+ * what *dma_ndler was set to.
  *
  * 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 size, void *cpu_addr,
+		dma_addr_t dma_addr, unsigned long attrs)
 {
 	struct resource *res;
 
 	if ((res = lookup_resource(&_sparc_dvma,
-	    (unsigned long)p)) == NULL) {
-		printk("pci_free_consistent: cannot free %p\n", p);
+	    (unsigned long)cpu_addr)) == NULL) {
+		printk("%s: cannot free %p\n", __func__, cpu_addr);
 		return;
 	}
 
-	if (((unsigned long)p & (PAGE_SIZE-1)) != 0) {
-		printk("pci_free_consistent: unaligned va %p\n", p);
+	if (((unsigned long)cpu_addr & (PAGE_SIZE-1)) != 0) {
+		printk("%s: unaligned va %p\n", __func__, cpu_addr);
 		return;
 	}
 
-	n = PAGE_ALIGN(n);
-	if (resource_size(res) != n) {
-		printk("pci_free_consistent: region 0x%lx asked 0x%lx\n",
-		    (long)resource_size(res), (long)n);
+	size = PAGE_ALIGN(size);
+	if (resource_size(res) != size) {
+		printk("%s: region 0x%lx asked 0x%zx\n", __func__,
+		    (long)resource_size(res), size);
 		return;
 	}
 
-	dma_make_coherent(ba, n);
-	srmmu_unmapiorange((unsigned long)p, n);
+	dma_make_coherent(dma_addr, size);
+	srmmu_unmapiorange((unsigned long)cpu_addr, size);
 
 	release_resource(res);
 	kfree(res);
-	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));
-		}
-	}
-}
-
-/* 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)
-{
-	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));
-	}
+	free_pages((unsigned long)phys_to_virt(dma_addr), get_order(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));
-		}
-	}
-}
+/* IIep is write-through, not flushing on cpu to device transfer. */
 
-static void pci32_sync_sg_for_device(struct device *device, struct scatterlist *sgl,
-				     int nents, 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)
 {
-	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));
-		}
-	}
+	if (dir != PCI_DMA_TODEVICE)
+		dma_make_coherent(paddr, PAGE_ALIGN(size));
 }
 
-/* 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);
 
-- 
2.18.0

Re: [PATCH] sparc: use generic dma_noncoherent_ops

[Sam Ravnborg]

Hi Christoph.

On Mon, Jul 30, 2018 at 06:17:23PM +0200, Christoph Hellwig wrote:
> 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>
> Acked-by: Sam Ravnborg <sam@ravnborg.org>
> ---
>  Makefile                             |   2 +-
>  arch/sparc/Kconfig                   |   2 +
>  arch/sparc/include/asm/dma-mapping.h |   5 +-
>  arch/sparc/kernel/ioport.c           | 193 +++++----------------------
>  4 files changed, 36 insertions(+), 166 deletions(-)
> 
> diff --git a/Makefile b/Makefile
> index 85f3481a56d6..8a3fd0c4a76e 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -368,7 +368,7 @@ HOST_LOADLIBES := $(HOST_LFS_LIBS)
>  # Make variables (CC, etc...)
>  AS		= $(CROSS_COMPILE)as
>  LD		= $(CROSS_COMPILE)ld
> -CC		= $(CROSS_COMPILE)gcc
> +CC		= sparc64-linux-gnu-gcc-8
>  CPP		= $(CC) -E
>  AR		= $(CROSS_COMPILE)ar
>  NM		= $(CROSS_COMPILE)nm
This sneaked in by accident...

	Sam

Re: [PATCH] sparc: use generic dma_noncoherent_ops

[Sam Ravnborg]

Hi Christoph.

On Mon, Jul 30, 2018 at 06:17:23PM +0200, Christoph Hellwig wrote:
> 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-jcswGhMUV9g@public.gmane.org>
> Acked-by: Sam Ravnborg <sam-uyr5N9Q2VtJg9hUCZPvPmw@public.gmane.org>
> ---
>  Makefile                             |   2 +-
>  arch/sparc/Kconfig                   |   2 +
>  arch/sparc/include/asm/dma-mapping.h |   5 +-
>  arch/sparc/kernel/ioport.c           | 193 +++++----------------------
>  4 files changed, 36 insertions(+), 166 deletions(-)
> 
> diff --git a/Makefile b/Makefile
> index 85f3481a56d6..8a3fd0c4a76e 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -368,7 +368,7 @@ HOST_LOADLIBES := $(HOST_LFS_LIBS)
>  # Make variables (CC, etc...)
>  AS		= $(CROSS_COMPILE)as
>  LD		= $(CROSS_COMPILE)ld
> -CC		= $(CROSS_COMPILE)gcc
> +CC		= sparc64-linux-gnu-gcc-8
>  CPP		= $(CC) -E
>  AR		= $(CROSS_COMPILE)ar
>  NM		= $(CROSS_COMPILE)nm
This sneaked in by accident...

	Sam

Re: [PATCH] sparc: use generic dma_noncoherent_ops

[Christoph Hellwig]

On Fri, Jul 27, 2018 at 11:05:48PM +0200, Sam Ravnborg wrote:
> Hi Christoph.
> 
> Some observations below - nitpick and bikeshedding only.
> 
> The parameter of phys_addr_t is sometimes renamed
> to use the same name as in the original prototype (good),
> and sometimes uses the old name (bad).
> This makes it inconsistent as the local name changes in the
> different functions, but they represents the same.

I'll change it.

> You can add my:
> Acked-by: Sam Ravnborg <sam-uyr5N9Q2VtJg9hUCZPvPmw@public.gmane.org>

I will resend it with your nitpicks addressed and your ack added.

> > +void *arch_dma_alloc(struct device *dev, size_t len, dma_addr_t *pba, gfp_t gfp,
> > +		unsigned long attrs)
> 
> This function was renamed in ee664a9252d24 and is now renamed again.
> The printk statements should be updated to use arch_dma_alloc.

I've switched the printk statements to use __func__ instead, not that
I plan for another rename any time soon.

I've also update the comments referring to this function to use the
right name.

Re: [PATCH] sparc: use generic dma_noncoherent_ops

[Christoph Hellwig]

On Fri, Jul 27, 2018 at 11:05:48PM +0200, Sam Ravnborg wrote:
> Hi Christoph.
> 
> Some observations below - nitpick and bikeshedding only.
> 
> The parameter of phys_addr_t is sometimes renamed
> to use the same name as in the original prototype (good),
> and sometimes uses the old name (bad).
> This makes it inconsistent as the local name changes in the
> different functions, but they represents the same.

I'll change it.

> You can add my:
> Acked-by: Sam Ravnborg <sam@ravnborg.org>

I will resend it with your nitpicks addressed and your ack added.

> > +void *arch_dma_alloc(struct device *dev, size_t len, dma_addr_t *pba, gfp_t gfp,
> > +		unsigned long attrs)
> 
> This function was renamed in ee664a9252d24 and is now renamed again.
> The printk statements should be updated to use arch_dma_alloc.

I've switched the printk statements to use __func__ instead, not that
I plan for another rename any time soon.

I've also update the comments referring to this function to use the
right name.

Re: [PATCH] sparc: use generic dma_noncoherent_ops

[Sam Ravnborg]

Hi Christoph.

Some observations below - nitpick and bikeshedding only.

The parameter of phys_addr_t is sometimes renamed
to use the same name as in the original prototype (good),
and sometimes uses the old name (bad).
This makes it inconsistent as the local name changes in the
different functions, but they represents the same.

I really like how you managed to kill a lot of code
with this patch.

You can add my:
Acked-by: Sam Ravnborg <sam@ravnborg.org>

But that does imply that this is OK, you
must wait for davem.

	Sam

On Fri, Jul 27, 2018 at 06:44:09PM +0200, Christoph Hellwig wrote:
> 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 0f535debf802..79f29c67291a 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

The current implementation has both:
pci32_sync_single_for_cpu() AND
pci32_sync_single_for_device

Yet, the new implementation only include arch_sync_dma_for_cpu()
You addressed this in the changelog - so I guess it is OK.


> --- 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 cca9134cfa7d..960c1bc22c2e 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)

This function was renamed in ee664a9252d24 and is now renamed again.
The printk statements should be updated to use arch_dma_alloc.


>  {
>  	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)
Likewise, use arch_dma_free in printk statements

Re: [PATCH] sparc: use generic dma_noncoherent_ops

[Sam Ravnborg]

Hi Christoph.

Some observations below - nitpick and bikeshedding only.

The parameter of phys_addr_t is sometimes renamed
to use the same name as in the original prototype (good),
and sometimes uses the old name (bad).
This makes it inconsistent as the local name changes in the
different functions, but they represents the same.

I really like how you managed to kill a lot of code
with this patch.

You can add my:
Acked-by: Sam Ravnborg <sam-uyr5N9Q2VtJg9hUCZPvPmw@public.gmane.org>

But that does imply that this is OK, you
must wait for davem.

	Sam

On Fri, Jul 27, 2018 at 06:44:09PM +0200, Christoph Hellwig wrote:
> 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-jcswGhMUV9g@public.gmane.org>
> ---
>  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 0f535debf802..79f29c67291a 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

The current implementation has both:
pci32_sync_single_for_cpu() AND
pci32_sync_single_for_device

Yet, the new implementation only include arch_sync_dma_for_cpu()
You addressed this in the changelog - so I guess it is OK.


> --- 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 cca9134cfa7d..960c1bc22c2e 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)

This function was renamed in ee664a9252d24 and is now renamed again.
The printk statements should be updated to use arch_dma_alloc.


>  {
>  	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)
Likewise, use arch_dma_free in printk statements

[PATCH] sparc: use generic dma_noncoherent_ops

[Christoph Hellwig]

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 0f535debf802..79f29c67291a 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 cca9134cfa7d..960c1bc22c2e 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);
 
-- 
2.18.0

[PATCH] sparc: use generic dma_noncoherent_ops

[Christoph Hellwig]

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-jcswGhMUV9g@public.gmane.org>
---
 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 0f535debf802..79f29c67291a 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 cca9134cfa7d..960c1bc22c2e 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);
 
-- 
2.18.0