Skip to contentSkip to conversation

generic noncoherent dma ops for sparc32, 2nd resend

PATCH In review Christoph Hellwig 2 participants 4 messages Jul 27, 2018 – Jul 30, 2018 3 messages not in the archive SPARC + UltraSPARC (sparc/sparc64)

lore
Share & tools
lore
#0 lore raw

Dave, can you take a review this and apply if ok? I've been sending this twice already but never got an answer. Out of my batch of conversion sparc is one of only two that haven't been merged yet.

A message that is not in the archive (sent off-list or not yet received): [email protected]

#2 lore raw

PATCH 1/1 sparc: use generic dma_noncoherent_ops

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.

---
 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(-)
arch/sparc/Kconfig +2 -0
index 0f535debf802..79f29c67291a 100644
@@ -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
arch/sparc/include/asm/dma-mapping.h +2 -3
index 12ae33daf52f..e17566376934 100644
@@ -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;
 }
arch/sparc/kernel/ioport.c +10 -141

179 lines not on this page · raw message

Signature · 1 line
2.18.0

A message that is not in the archive (sent off-list or not yet received): [email protected]

Acked-by
#4 lore raw

Re: [PATCH] sparc: use generic dma_noncoherent_ops

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:

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

Sam

Quoted 18 lines · 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.

---
 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(-)
arch/sparc/Kconfig +0 -0
index 0f535debf802..79f29c67291a 100644

arch/sparc/Kconfig config SPARC · line 52

Quoted 5 lines
arch/sparc/Kconfig +2 -0
@@ -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.

Quoted 37 lines
arch/sparc/include/asm/dma-mapping.h +2 -3
@@ -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;
 }
arch/sparc/kernel/ioport.c +1 -0
index cca9134cfa7d..960c1bc22c2e 100644
@@ -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>

arch/sparc/kernel/ioport.c arch_initcall(sparc_register_ioport); · line 439

Quoted 8 lines
arch/sparc/kernel/ioport.c +2 -3
@@ -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.

arch/sparc/kernel/ioport.c static void *pci32_alloc_coherent(struct device *dev, size_t len, · line 492

Quoted 11 lines
arch/sparc/kernel/ioport.c +2 -2
 {
 	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

A message that is not in the archive (sent off-list or not yet received): [email protected]

#6 lore raw

Re: [PATCH] sparc: use generic dma_noncoherent_ops

Quoted 9 lines · 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'll change it.

Quoted 2 lines

You can add my:

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

Quoted 5 lines

arch/sparc/kernel/ioport.c arch_initcall(sparc_register_ioport); · line 439

Quoted 2 lines
arch/sparc/kernel/ioport.c +2 -0
+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.