[PATCH 1/2] mmu: define MAP_WRITECOMBINE for all architectures

[PATCH 1/2] mmu: define MAP_WRITECOMBINE for all architectures

[Ahmad Fatoum]

Write Combine (WC) memory is an optimization over uncached memory,
especially suitable for framebuffers. We allocate WC memory using
dma_alloc_writecombine(), but it can be useful to remap existing memory
not allocated within barebox as WC.

To allow this, let's define MAP_WRITECOMBINE globally with a fallback to
MAP_UNCACHED if not available.

Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
---
 arch/Kconfig              |  3 +++
 arch/arm/Kconfig          |  1 +
 arch/arm/cpu/mmu-common.h |  1 -
 arch/arm/cpu/mmu_32.c     |  6 +++---
 arch/arm/cpu/mmu_64.c     |  4 ++--
 include/mmu.h             | 14 ++++++++++----
 6 files changed, 19 insertions(+), 10 deletions(-)

diff --git a/arch/Kconfig b/arch/Kconfig
index 919c8cfebab5..55618bf896c2 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -59,6 +59,9 @@ config ARCH_HAS_CTRLC
 config ARCH_DMA_DEFAULT_COHERENT
 	bool
 
+config ARCH_HAS_DMA_WRITE_COMBINE
+	bool
+
 config ARCH_HAS_ASAN_FIBER_API
 	bool
 
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 7a3952700aa8..9694cb5b7463 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -16,6 +16,7 @@ config ARM
 	select HAVE_ARCH_BOARD_GENERIC_DT if OFDEVICE
 	select HAVE_ARCH_BOOTM_OFTREE
 	select HW_HAS_PCI
+	select ARCH_HAS_DMA_WRITE_COMBINE
 	default y
 
 config ARCH_LINUX_NAME
diff --git a/arch/arm/cpu/mmu-common.h b/arch/arm/cpu/mmu-common.h
index ac11a87be416..3bca5cc3b821 100644
--- a/arch/arm/cpu/mmu-common.h
+++ b/arch/arm/cpu/mmu-common.h
@@ -10,7 +10,6 @@
 #include <linux/kernel.h>
 #include <linux/sizes.h>
 
-#define ARCH_MAP_WRITECOMBINE	((unsigned)-1)
 #define ARCH_MAP_CACHED_RWX	((unsigned)-2)
 #define ARCH_MAP_CACHED_RO	((unsigned)-3)
 
diff --git a/arch/arm/cpu/mmu_32.c b/arch/arm/cpu/mmu_32.c
index 67f1fe59886a..89a18d342b80 100644
--- a/arch/arm/cpu/mmu_32.c
+++ b/arch/arm/cpu/mmu_32.c
@@ -238,7 +238,7 @@ static uint32_t get_pte_flags(int map_type)
 			return PTE_FLAGS_UNCACHED_V7;
 		case MAP_CODE:
 			return PTE_FLAGS_CODE_V7;
-		case ARCH_MAP_WRITECOMBINE:
+		case MAP_WRITECOMBINE:
 			return PTE_FLAGS_WC_V7;
 		case MAP_FAULT:
 		default:
@@ -253,7 +253,7 @@ static uint32_t get_pte_flags(int map_type)
 		case MAP_CACHED:
 			return PTE_FLAGS_CACHED_V4;
 		case MAP_UNCACHED:
-		case ARCH_MAP_WRITECOMBINE:
+		case MAP_WRITECOMBINE:
 			return PTE_FLAGS_UNCACHED_V4;
 		case MAP_FAULT:
 		default:
@@ -651,7 +651,7 @@ void mmu_disable(void)
 
 void *dma_alloc_writecombine(struct device *dev, size_t size, dma_addr_t *dma_handle)
 {
-	return dma_alloc_map(dev, size, dma_handle, ARCH_MAP_WRITECOMBINE);
+	return dma_alloc_map(dev, size, dma_handle, MAP_WRITECOMBINE);
 }
 
 void mmu_early_enable(unsigned long membase, unsigned long memsize, unsigned long barebox_start)
diff --git a/arch/arm/cpu/mmu_64.c b/arch/arm/cpu/mmu_64.c
index abcc970f4bff..a229e4cb5526 100644
--- a/arch/arm/cpu/mmu_64.c
+++ b/arch/arm/cpu/mmu_64.c
@@ -295,7 +295,7 @@ static unsigned long get_pte_attrs(unsigned flags)
 		return attrs_xn() | UNCACHED_MEM;
 	case MAP_FAULT:
 		return 0x0;
-	case ARCH_MAP_WRITECOMBINE:
+	case MAP_WRITECOMBINE:
 		return attrs_xn() | MEM_ALLOC_WRITECOMBINE;
 	case MAP_CODE:
 		return CACHED_MEM | PTE_BLOCK_RO;
@@ -448,7 +448,7 @@ void dma_flush_range(void *ptr, size_t size)
 
 void *dma_alloc_writecombine(struct device *dev, size_t size, dma_addr_t *dma_handle)
 {
-	return dma_alloc_map(dev, size, dma_handle, ARCH_MAP_WRITECOMBINE);
+	return dma_alloc_map(dev, size, dma_handle, MAP_WRITECOMBINE);
 }
 
 static void init_range(size_t total_level0_tables)
diff --git a/include/mmu.h b/include/mmu.h
index 20855e89eda3..17c04d2fa05f 100644
--- a/include/mmu.h
+++ b/include/mmu.h
@@ -5,10 +5,16 @@
 #include <linux/types.h>
 #include <errno.h>
 
-#define MAP_UNCACHED	0
-#define MAP_CACHED	1
-#define MAP_FAULT	2
-#define MAP_CODE	3
+#define MAP_UNCACHED		0
+#define MAP_CACHED		1
+#define MAP_FAULT		2
+#define MAP_CODE		3
+
+#ifdef CONFIG_ARCH_HAS_DMA_WRITE_COMBINE
+#define MAP_WRITECOMBINE	4
+#else
+#define MAP_WRITECOMBINE	MAP_UNCACHED
+#endif
 
 /*
  * Depending on the architecture the default mapping can be
-- 
2.39.5

[PATCH 2/2] video: remap framebuffer as writecombine if possible

[Ahmad Fatoum]

We do not do cache maintenance for framebuffers and expect that memory
is mapped suitably, either writecombine or uncached (coherent).

The simplefb-client was first added to barebox for a cache-coherent
platform and did not observe this requirement.

To enable its use on ARM as well without graphic artifacts, let's remap
the range as writecombine, which will fall back to uncached if
unsupported.

Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
---
 drivers/video/simplefb-client.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/video/simplefb-client.c b/drivers/video/simplefb-client.c
index dafec6178fe6..1c0ee3e7d426 100644
--- a/drivers/video/simplefb-client.c
+++ b/drivers/video/simplefb-client.c
@@ -18,6 +18,7 @@
 #include <linux/platform_data/simplefb.h>
 #include <driver.h>
 #include <of.h>
+#include <mmu.h>
 
 static struct fb_ops simplefb_ops;
 
@@ -119,6 +120,11 @@ static int simplefb_probe(struct device *dev)
 	info->screen_base = (void *)mem->start;
 	info->screen_size = resource_size(mem);
 
+	/*
+	 * Best effort: Some platforms don't need this and those that do,
+	 * will at worst have some graphic artifacts on lack of remap_range.
+	 */
+	(void)remap_range(info->screen_base, info->screen_size, MAP_WRITECOMBINE);
 
 	info->fbops = &simplefb_ops;
 
-- 
2.39.5

Re: [PATCH 1/2] mmu: define MAP_WRITECOMBINE for all architectures

[Sascha Hauer]

On Wed, 30 Jul 2025 14:31:25 +0200, Ahmad Fatoum wrote:
> Write Combine (WC) memory is an optimization over uncached memory,
> especially suitable for framebuffers. We allocate WC memory using
> dma_alloc_writecombine(), but it can be useful to remap existing memory
> not allocated within barebox as WC.
> 
> To allow this, let's define MAP_WRITECOMBINE globally with a fallback to
> MAP_UNCACHED if not available.
> 
> [...]

Applied, thanks!

[1/2] mmu: define MAP_WRITECOMBINE for all architectures
      https://git.pengutronix.de/cgit/barebox/commit/?id=b87902baeb53 (link may not be stable)
[2/2] video: remap framebuffer as writecombine if possible
      https://git.pengutronix.de/cgit/barebox/commit/?id=16d0cfa203cb (link may not be stable)

Best regards,
-- 
Sascha Hauer <s.hauer@pengutronix.de>