Skip to contentSkip to conversation

dma_alloc_writecombine() and MAX_ORDER

DISCUSSION Michael Hunold 2 participants 4 messages Apr 29, 2011 – May 3, 2011 Arm (32-bit)

lore
Share & tools
lore
#0 lore raw

Hello,

I'm using dma_alloc_writecombine() to allocate memory in my framebuffer device driver just like many other drivers do.

Now I have a new panel with a higher resolution and more bits-per-pixel. I want to use double-buffering with DirectFB and so my framebuffer size exceeds 4Mb.

Even though my embedded system has plenty of memory, now I receive the following warning and the allocation fails:

WARNING: at /home/teemhu/mcu/has/topas/mm/page_alloc.c:2012
__alloc_pages_nodemask+0x444/0x650()

This is due to the following check in __alloc_pages_slowpath():

	if (order >= MAX_ORDER) {
		WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
		return NULL;
	}

Because MAX_ORDER is 11, only memory chunks of size (PAGE_SIZE << MAX_ORDER_NR_PAGES) = 4Mb can be allocated.

Apparently this topic has come up when using the OMAP DSS already:
http://www.spinics.net/lists/linux-omap/msg08124.html

Another user has asked this question on the linuxfb-dev mailing list in October 2010, but did not receive an answer:
http://www.spinics.net/lists/linux-fbdev/msg01745.html

So, would increasing MAX_ORDER in order to allow bigger memory chunks to be allocated via dma_alloc_writecombine() be acceptable?

If not, can anybody give me some advice how to do the memory allocation for the framebuffer memory instead in order to circumvent the problem?

CU
Michael.

#1 lore raw
Quoted 3 lines · Michael Hunold

Now I have a new panel with a higher resolution and more bits-per-pixel. I want to use double-buffering with DirectFB and so my framebuffer size exceeds 4Mb.

You could try this patch. It steals the memory for the coherent and writecombine memory at boot time, so MAX_ORDER no longer comes into the problem. You may need to define CONSISTENT_WC_SIZE and CONSISTENT_DMA_SIZE appropriately to gain sufficient memory. Note that the sum of those two must be a multiple of 2MB, and individually must be a multiple of 1MB.

arch/arm/include/asm/dma-mapping.h +1 -2
index c568da7..489c197 100644
@@ -230,8 +230,7 @@ int dma_mmap_coherent(struct device *, struct vm_area_struct *,
 extern void *dma_alloc_writecombine(struct device *, size_t, dma_addr_t *,
 		gfp_t);
 
-#define dma_free_writecombine(dev,size,cpu_addr,handle) \
-	dma_free_coherent(dev,size,cpu_addr,handle)
+extern void dma_free_writecombine(struct device *, size_t, void *, dma_addr_t);
 
 int dma_mmap_writecombine(struct device *, struct vm_area_struct *,
 		void *, dma_addr_t, size_t);
arch/arm/include/asm/mach/map.h +2 -0
index d2fedb5..3845215 100644
@@ -29,6 +29,8 @@ struct map_desc {
 #define MT_MEMORY_NONCACHED	11
 #define MT_MEMORY_DTCM		12
 #define MT_MEMORY_ITCM		13
+#define MT_DMA_COHERENT		14
+#define MT_WC_COHERENT		15
 
 #ifdef CONFIG_MMU
 extern void iotable_init(struct map_desc *, int);
arch/arm/include/asm/memory.h +7 -0
index 23c2e8e..fa54ecf 100644
@@ -88,6 +88,13 @@
 #define CONSISTENT_END		(0xffe00000UL)
 #define CONSISTENT_BASE		(CONSISTENT_END - CONSISTENT_DMA_SIZE)
 
+#ifndef CONSISTENT_WC_SIZE
+#define CONSISTENT_WC_SIZE	SZ_2M
+#endif
+
+#define CONSISTENT_WC_END	CONSISTENT_BASE
+#define CONSISTENT_WC_BASE	(CONSISTENT_WC_END - CONSISTENT_WC_SIZE)
+
 #else /* CONFIG_MMU */
 
 /*
arch/arm/mm/dma-mapping.c +164 -146

431 lines not on this page · raw message

arch/arm/mm/init.c +1 -0
index 5164069..95c826c 100644
@@ -313,6 +313,7 @@ void __init arm_memblock_init(struct meminfo *mi, struct machine_desc *mdesc)
 #endif
 
 	arm_mm_memblock_reserve();
+	dma_coherent_reserve();
 
 	/* reserve any platform specific memblock areas */
 	if (mdesc->reserve)
arch/arm/mm/mm.h +2 -0
index d238410..e672882 100644
@@ -30,3 +30,5 @@ extern void __flush_dcache_page(struct address_space *mapping, struct page *page
 
 void __init bootmem_init(void);
 void arm_mm_memblock_reserve(void);
+void dma_coherent_reserve(void);
+void dma_coherent_mapping(void);
arch/arm/mm/mmu.c +24 -0
index e94888b..1fec2bf 100644
@@ -275,6 +275,16 @@ static struct mem_type mem_types[] = {
 		.prot_l1   = PMD_TYPE_TABLE,
 		.domain    = DOMAIN_KERNEL,
 	},
+	[MT_DMA_COHERENT] = {
+		.prot_sect	= PMD_TYPE_SECT | PMD_SECT_AP_WRITE |
+				  PMD_SECT_S,
+		.domain		= DOMAIN_IO,
+	},
+	[MT_WC_COHERENT] = {
+		.prot_sect	= PMD_TYPE_SECT | PMD_SECT_AP_WRITE |
+				  PMD_SECT_S,
+		.domain		= DOMAIN_IO,
+	},
 };
 
 const struct mem_type *get_mem_type(unsigned int type)
@@ -355,6 +365,7 @@ static void __init build_mem_type_table(void)
 			mem_types[MT_DEVICE_NONSHARED].prot_sect |= PMD_SECT_XN;
 			mem_types[MT_DEVICE_CACHED].prot_sect |= PMD_SECT_XN;
 			mem_types[MT_DEVICE_WC].prot_sect |= PMD_SECT_XN;
+			mem_types[MT_DMA_COHERENT].prot_sect |= PMD_SECT_XN;
 		}
 		if (cpu_arch >= CPU_ARCH_ARMv7 && (cr & CR_TRE)) {
 			/*
@@ -459,13 +470,24 @@ static void __init build_mem_type_table(void)
 			/* Non-cacheable Normal is XCB = 001 */
 			mem_types[MT_MEMORY_NONCACHED].prot_sect |=
 				PMD_SECT_BUFFERED;
+			mem_types[MT_WC_COHERENT].prot_sect |=
+				PMD_SECT_BUFFERED;
+			mem_types[MT_DMA_COHERENT].prot_sect |=
+				PMD_SECT_BUFFERED;
 		} else {
 			/* For both ARMv6 and non-TEX-remapping ARMv7 */
 			mem_types[MT_MEMORY_NONCACHED].prot_sect |=
 				PMD_SECT_TEX(1);
+			mem_types[MT_WC_COHERENT].prot_sect |=
+				PMD_SECT_TEX(1);
+#ifdef CONFIG_ARM_DMA_MEM_BUFFERABLE
+			mem_types[MT_DMA_COHERENT].prot_sect |=
+				PMD_SECT_TEX(1);
+#endif
 		}
 	} else {
 		mem_types[MT_MEMORY_NONCACHED].prot_sect |= PMD_SECT_BUFFERABLE;
+		mem_types[MT_WC_COHERENT].prot_sect |= PMD_SECT_BUFFERED;
 	}
 
 	for (i = 0; i < 16; i++) {
@@ -977,6 +999,8 @@ static void __init devicemaps_init(struct machine_desc *mdesc)
 		create_mapping(&map);
 	}
 
+	dma_coherent_mapping();
+
 	/*
 	 * Ask the machine support to map in the statically mapped devices.
 	 */
#2 lore raw

Hello Russell,

on 04/29/2011 04:27 PM Russell King - ARM Linux said the following:

Quoted 11 lines · Michael Hunold
Quoted 4 lines

Now I have a new panel with a higher resolution and more bits-per-pixel. I want to use double-buffering with DirectFB and so my framebuffer size exceeds 4Mb.

You could try this patch. It steals the memory for the coherent and writecombine memory at boot time, so MAX_ORDER no longer comes into the problem. You may need to define CONSISTENT_WC_SIZE and CONSISTENT_DMA_SIZE appropriately to gain sufficient memory. Note that the sum of those two must be a multiple of 2MB, and individually must be a multiple of 1MB.

I really want to try this patch out, but unfortunately, I don't know to which tree I can apply it to. :-(

I'm working on a tree which is based on vanilla 2.6.38 and has the additions for the platform I'm working with, but there the patch won't apply cleanly.

The tree can be found here
http://git.labs.kernelconcepts.de/?p=topas.git;a=summary

So I thought I should try "devel-stable" from
http://ftp.arm.linux.org.uk/pub/linux/arm/kernel/git-cur/linux-2.6-arm.git
but there are even more rejects there.

Can you please tell me from which tree you created this patch?

Best regards
Michael.

#3 lore raw
Quoted 9 lines · Michael Hunold

Hello Russell,

on 04/29/2011 04:27 PM Russell King - ARM Linux said the following:

Quoted 5 lines · Michael Hunold
Quoted 4 lines

Now I have a new panel with a higher resolution and more bits-per-pixel. I want to use double-buffering with DirectFB and so my framebuffer size exceeds 4Mb.

Quoted 8 lines
Quoted 5 lines

You could try this patch. It steals the memory for the coherent and writecombine memory at boot time, so MAX_ORDER no longer comes into the problem. You may need to define CONSISTENT_WC_SIZE and CONSISTENT_DMA_SIZE appropriately to gain sufficient memory. Note that the sum of those two must be a multiple of 2MB, and individually must be a multiple of 1MB.

I really want to try this patch out, but unfortunately, I don't know to which tree I can apply it to. :-(

You may need 516295e (ARM: pgtable: add pud-level code) from a later kernel as well then.