[PATCH v3 0/4] Use dma_default_coherent for devicetree default coherency

Hi all,

This series split out second half of my previous series
"[PATCH 0/4] MIPS DMA coherence fixes".

It intends to use dma_default_coherent to determine the default coherency of
devicetree probed devices instead of hardcoding it with Kconfig options.

For some MIPS systems, dma_default_coherent is determined with either
bootloader or hardware registers in platform initilization code, and devicetree
does not explicility specify the coherency of the device, so we need the ability
to change the default coherency of devicetree probed devices.

For other platforms that supports noncoherent, dma_default_coherent is a fixed
value set by arch code. It's defaulted to false for most archs except RISC-V.

Thanks
- Jiaxun
---
v2:
  - Add PATCH 1 to help with backporting
  - Use Kconfig option to set dma_default_coherent 

v3:
  - Style fixes
  - Squash setting ARCH_DMA_DEFAULT_COHERENT into PATCH 4
  - Setting ARCH_DMA_DEFAULT_COHERENT for PowerPC

Jiaxun Yang (4):
  of: address: Fix default coherency for MIPS
  dma-mapping: Provide a fallback dma_default_coherent
  dma-mapping: Provide CONFIG_ARCH_DMA_DEFAULT_COHERENT
  of: address: Always use dma_default_coherent for default coherency

 arch/powerpc/Kconfig        | 2 +-
 arch/riscv/Kconfig          | 2 +-
 drivers/of/Kconfig          | 4 ----
 drivers/of/address.c        | 2 +-
 include/linux/dma-map-ops.h | 2 ++
 kernel/dma/Kconfig          | 7 +++++++
 kernel/dma/mapping.c        | 6 +++++-
 7 files changed, 17 insertions(+), 8 deletions(-)

-- 
2.37.1 (Apple Git-137.1)

DT-based MIPS doesn't use OF_DMA_DEFAULT_COHERENT, but
might override the system-wide default at runtime.

Use dma_default_coherent to override default coherence for
MIPS.

Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
---
 drivers/of/address.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/drivers/of/address.c b/drivers/of/address.c
index 4c0b169ef9bf..c105d66a1fa4 100644
--- a/drivers/of/address.c
+++ b/drivers/of/address.c

@@ -1105,6 +1105,14 @@ bool of_dma_is_coherent(struct device_node *np)
 	struct device_node *node;
 	bool is_coherent = IS_ENABLED(CONFIG_OF_DMA_DEFAULT_COHERENT);
 
+	/*
+	 * DT-based MIPS doesn't use OF_DMA_DEFAULT_COHERENT, but
+	 * might override the system-wide default at runtime.
+	 */
+#if defined(CONFIG_MIPS) && defined(CONFIG_DMA_NONCOHERENT)
+	is_coherent = dma_default_coherent;
+#endif
+
 	node = of_node_get(np);
 
 	while (node) {

-- 
2.37.1 (Apple Git-137.1)

dma_default_coherent was decleared unconditionally at kernel/dma/mapping.c
but only decleared when any of non-coherent options is enabled in
dma-map-ops.h.

Guard the declaration in mapping.c with non-coherent options and provide
a fallback definition.

Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
---
v3: Style fix
---
 include/linux/dma-map-ops.h | 2 ++
 kernel/dma/mapping.c        | 4 ++++
 2 files changed, 6 insertions(+)

diff --git a/include/linux/dma-map-ops.h b/include/linux/dma-map-ops.h
index 41bf4bdb117a..31f114f486c4 100644
--- a/include/linux/dma-map-ops.h
+++ b/include/linux/dma-map-ops.h

@@ -269,6 +269,8 @@ static inline bool dev_is_dma_coherent(struct device *dev)
 	return dev->dma_coherent;
 }
 #else
+#define dma_default_coherent true
+
 static inline bool dev_is_dma_coherent(struct device *dev)
 {
 	return true;

diff --git a/kernel/dma/mapping.c b/kernel/dma/mapping.c
index 68106e3791f6..80f9663ffe26 100644
--- a/kernel/dma/mapping.c
+++ b/kernel/dma/mapping.c

@@ -17,7 +17,11 @@
 #include "debug.h"
 #include "direct.h"
 
+#if defined(CONFIG_ARCH_HAS_SYNC_DMA_FOR_DEVICE) || \
+	defined(CONFIG_ARCH_HAS_SYNC_DMA_FOR_CPU) || \
+	defined(CONFIG_ARCH_HAS_SYNC_DMA_FOR_CPU_ALL)
 bool dma_default_coherent;
+#endif
 
 /*
  * Managed DMA API

-- 
2.37.1 (Apple Git-137.1)

As for now all arches have dma_default_coherent reflecting default
DMA coherency for of devices, so there is no need to have a standalone
config option.

Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
---
v3: Squash setting ARCH_DMA_DEFAULT_COHERENT into this patch.
---
 arch/powerpc/Kconfig |  2 +-
 arch/riscv/Kconfig   |  2 +-
 drivers/of/Kconfig   |  4 ----
 drivers/of/address.c | 10 +---------
 4 files changed, 3 insertions(+), 15 deletions(-)

diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
index 57f5d2f53d06..824e00a1277b 100644
--- a/arch/powerpc/Kconfig
+++ b/arch/powerpc/Kconfig

@@ -113,6 +113,7 @@ config PPC
 	#
 	select ARCH_32BIT_OFF_T if PPC32
 	select ARCH_DISABLE_KASAN_INLINE	if PPC_RADIX_MMU
+	select ARCH_DMA_DEFAULT_COHERENT	if !NOT_COHERENT_CACHE
 	select ARCH_ENABLE_MEMORY_HOTPLUG
 	select ARCH_ENABLE_MEMORY_HOTREMOVE
 	select ARCH_HAS_COPY_MC			if PPC64

@@ -273,7 +274,6 @@ config PPC
 	select NEED_PER_CPU_PAGE_FIRST_CHUNK	if PPC64
 	select NEED_SG_DMA_LENGTH
 	select OF
-	select OF_DMA_DEFAULT_COHERENT		if !NOT_COHERENT_CACHE
 	select OF_EARLY_FLATTREE
 	select OLD_SIGACTION			if PPC32
 	select OLD_SIGSUSPEND

diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
index 36a5b6fed0d3..6425b5c5d6d4 100644
--- a/arch/riscv/Kconfig
+++ b/arch/riscv/Kconfig

@@ -12,6 +12,7 @@ config 32BIT
 
 config RISCV
 	def_bool y
+	select ARCH_DMA_DEFAULT_COHERENT
 	select ARCH_ENABLE_HUGEPAGE_MIGRATION if HUGETLB_PAGE && MIGRATION
 	select ARCH_ENABLE_SPLIT_PMD_PTLOCK if PGTABLE_LEVELS > 2
 	select ARCH_ENABLE_THP_MIGRATION if TRANSPARENT_HUGEPAGE

@@ -121,7 +122,6 @@ config RISCV
 	select MODULES_USE_ELF_RELA if MODULES
 	select MODULE_SECTIONS if MODULES
 	select OF
-	select OF_DMA_DEFAULT_COHERENT
 	select OF_EARLY_FLATTREE
 	select OF_IRQ
 	select PCI_DOMAINS_GENERIC if PCI

diff --git a/drivers/of/Kconfig b/drivers/of/Kconfig
index 644386833a7b..e40f10bf2ba4 100644
--- a/drivers/of/Kconfig
+++ b/drivers/of/Kconfig

@@ -102,8 +102,4 @@ config OF_OVERLAY
 config OF_NUMA
 	bool
 
-config OF_DMA_DEFAULT_COHERENT
-	# arches should select this if DMA is coherent by default for OF devices
-	bool
-
 endif # OF

diff --git a/drivers/of/address.c b/drivers/of/address.c
index c105d66a1fa4..23ade4919853 100644
--- a/drivers/of/address.c
+++ b/drivers/of/address.c

@@ -1103,15 +1103,7 @@ phys_addr_t __init of_dma_get_max_cpu_address(struct device_node *np)
 bool of_dma_is_coherent(struct device_node *np)
 {
 	struct device_node *node;
-	bool is_coherent = IS_ENABLED(CONFIG_OF_DMA_DEFAULT_COHERENT);
-
-	/*
-	 * DT-based MIPS doesn't use OF_DMA_DEFAULT_COHERENT, but
-	 * might override the system-wide default at runtime.
-	 */
-#if defined(CONFIG_MIPS) && defined(CONFIG_DMA_NONCOHERENT)
-	is_coherent = dma_default_coherent;
-#endif
+	bool is_coherent = dma_default_coherent;
 
 	node = of_node_get(np);
 

-- 
2.37.1 (Apple Git-137.1)

Provide a kconfig option to allow arches to manipulate default
value of dma_default_coherent in Kconfig.

Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
---
v3: Add comments
---
 kernel/dma/Kconfig   | 7 +++++++
 kernel/dma/mapping.c | 2 +-
 2 files changed, 8 insertions(+), 1 deletion(-)

diff --git a/kernel/dma/Kconfig b/kernel/dma/Kconfig
index 56866aaa2ae1..6677d0e64d27 100644
--- a/kernel/dma/Kconfig
+++ b/kernel/dma/Kconfig

@@ -76,6 +76,13 @@ config ARCH_HAS_DMA_PREP_COHERENT
 config ARCH_HAS_FORCE_DMA_UNENCRYPTED
 	bool
 
+#
+# Select this option if the architecture assumes DMA devices are coherent
+# by default.
+#
+config ARCH_DMA_DEFAULT_COHERENT
+	bool
+
 config SWIOTLB
 	bool
 	select NEED_DMA_MAP_STATE

diff --git a/kernel/dma/mapping.c b/kernel/dma/mapping.c
index 80f9663ffe26..9a4db5cce600 100644
--- a/kernel/dma/mapping.c
+++ b/kernel/dma/mapping.c

@@ -20,7 +20,7 @@
 #if defined(CONFIG_ARCH_HAS_SYNC_DMA_FOR_DEVICE) || \
 	defined(CONFIG_ARCH_HAS_SYNC_DMA_FOR_CPU) || \
 	defined(CONFIG_ARCH_HAS_SYNC_DMA_FOR_CPU_ALL)
-bool dma_default_coherent;
+bool dma_default_coherent = IS_ENABLED(CONFIG_ARCH_DMA_DEFAULT_COHERENT);
 #endif
 
 /*

-- 
2.37.1 (Apple Git-137.1)

The series looks fine to me.  How should we merge it?

2023年3月23日 07:29，Christoph Hellwig [off-list ref] 写道：

The series looks fine to me.  How should we merge it?

Perhaps go through dma-mapping tree?

Thanks
- Jiaxun

On Thu, Mar 23, 2023 at 09:07:31PM +0000, Jiaxun Yang wrote:

quoted

2023年3月23日 07:29，Christoph Hellwig [off-list ref] 写道：

The series looks fine to me.  How should we merge it?

Perhaps go through dma-mapping tree?

Is patch a 6.3 candidate or should all of it go into 6.4?

2023年3月23日 21:39，Christoph Hellwig [off-list ref] 写道：

On Thu, Mar 23, 2023 at 09:07:31PM +0000, Jiaxun Yang wrote:

quoted

quoted

2023年3月23日 07:29，Christoph Hellwig [off-list ref] 写道：

The series looks fine to me.  How should we merge it?

Perhaps go through dma-mapping tree?

Is patch a 6.3 candidate or should all of it go into 6.4?

Please leave it for 6.4, as corresponding MIPS arch part will be a part of 6.4.

Thanks
Jiaxun

On Fri, Mar 24, 2023 at 09:17:38AM +0000, Jiaxun Yang wrote:

quoted

Is patch a 6.3 candidate or should all of it go into 6.4?

Please leave it for 6.4, as corresponding MIPS arch part will be a part of 6.4.

Ok.  I'll really need review from the MIPS and drivers/of/ maintainers,
through.

On Tue, Mar 28, 2023 at 03:18:12AM +0200, Christoph Hellwig wrote:

On Fri, Mar 24, 2023 at 09:17:38AM +0000, Jiaxun Yang wrote:

quoted

quoted

Is patch a 6.3 candidate or should all of it go into 6.4?

Please leave it for 6.4, as corresponding MIPS arch part will be a part of 6.4.

Ok.  I'll really need review from the MIPS and drivers/of/ maintainers,
through.

I don't see any MIPS changes in the series besides the ifdef CONFIG_MIPS
part in patch 1, which gets removed again in patch 4 (chance to drop
that completely ?).

I've merged the corresponding MIPS patches into mips-next last week.

Thomas.

-- 
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea.                                                [ RFC1925, 2.3 ]

2023年3月28日 08:45，Thomas Bogendoerfer [off-list ref] 写道：

On Tue, Mar 28, 2023 at 03:18:12AM +0200, Christoph Hellwig wrote:

quoted

On Fri, Mar 24, 2023 at 09:17:38AM +0000, Jiaxun Yang wrote:

quoted

quoted

Is patch a 6.3 candidate or should all of it go into 6.4?

Please leave it for 6.4, as corresponding MIPS arch part will be a part of 6.4.

Ok.  I'll really need review from the MIPS and drivers/of/ maintainers,
through.

+cc devicetree foks.

I don't see any MIPS changes in the series besides the ifdef CONFIG_MIPS
part in patch 1, which gets removed again in patch 4 (chance to drop
that completely ?).

It was suggested by DMA folks to have that patch.

I've merged the corresponding MIPS patches into mips-next last week.

Thanks
- Jiaxun

Thomas.

-- 
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea.                                                [ RFC1925, 2.3 ]

On Tue, Mar 21, 2023 at 6:08 AM Jiaxun Yang [off-list ref] wrote:

DT-based MIPS doesn't use OF_DMA_DEFAULT_COHERENT, but
might override the system-wide default at runtime.

Use dma_default_coherent to override default coherence for
MIPS.

I assume you want this tagged for stable? Otherwise, I don't
understand why you add this here and then remove in patch 4.

On Tue, Mar 21, 2023 at 6:08 AM Jiaxun Yang [off-list ref] wrote:

As for now all arches have dma_default_coherent reflecting default
DMA coherency for of devices, so there is no need to have a standalone
config option.

Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
---
v3: Squash setting ARCH_DMA_DEFAULT_COHERENT into this patch.
---
 arch/powerpc/Kconfig |  2 +-
 arch/riscv/Kconfig   |  2 +-
 drivers/of/Kconfig   |  4 ----
 drivers/of/address.c | 10 +---------
 4 files changed, 3 insertions(+), 15 deletions(-)

Reviewed-by: Rob Herring <robh@kernel.org>

Jiaxun Yang [off-list ref] writes:

Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc)

cheers