[PATCH 0/2] ARM: dma-mapping: add highmem support for coherent allocation

[PATCH 0/2] ARM: dma-mapping: add highmem support for coherent allocation

[Marek Szyprowski]

Hello,

This is the last missing piece to let us efficiently use large DMA
buffers on systems with lots of memory, which have support for himem
enabled. The first patch adds support for CMA regions placed in high
memory zones, the second one also enables allocations of individual
pages from high memory zone for IOMMU-mapped devices. Those two changes
let us to significantly save low memory for other tasks.

Best regards
Marek Szyprowski
Samsung Poland R&D Center


Patch summary:

Marek Szyprowski (2):
  ARM: dma-mapping: add support for CMA regions placed in highmem zone
  ARM: dma-mapping: use himem for DMA buffers for IOMMU-mapped devices

 arch/arm/mm/dma-mapping.c |   70 +++++++++++++++++++++++++++++++++------------
 1 file changed, 52 insertions(+), 18 deletions(-)

-- 
1.7.9.5

[PATCH 1/2] ARM: dma-mapping: add support for CMA regions placed in highmem zone

[Marek Szyprowski]

This patch adds missing pieces to correctly support memory pages served
from CMA regions placed in high memory zones. Please note that the default
global CMA area is still put into lowmem and is limited by optional
architecture specific DMA zone. One can however put device specific CMA
regions in high memory zone to reduce lowmem usage.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
---
 arch/arm/mm/dma-mapping.c |   61 +++++++++++++++++++++++++++++++++------------
 1 file changed, 45 insertions(+), 16 deletions(-)

diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
index 6b2fb87..4080c37 100644
--- a/arch/arm/mm/dma-mapping.c
+++ b/arch/arm/mm/dma-mapping.c
@@ -186,16 +186,29 @@ static u64 get_coherent_dma_mask(struct device *dev)
 
 static void __dma_clear_buffer(struct page *page, size_t size)
 {
-	void *ptr;
 	/*
 	 * Ensure that the allocated pages are zeroed, and that any data
 	 * lurking in the kernel direct-mapped region is invalidated.
 	 */
-	ptr = page_address(page);
-	if (ptr) {
-		memset(ptr, 0, size);
-		dmac_flush_range(ptr, ptr + size);
-		outer_flush_range(__pa(ptr), __pa(ptr) + size);
+	if (!PageHighMem(page)) {
+		void *ptr = page_address(page);
+		if (ptr) {
+			memset(ptr, 0, size);
+			dmac_flush_range(ptr, ptr + size);
+			outer_flush_range(__pa(ptr), __pa(ptr) + size);
+		}
+	} else {
+		phys_addr_t base = __pfn_to_phys(page_to_pfn(page));
+		phys_addr_t end = base + size;
+		while (size > 0) {
+			void *ptr = kmap_atomic(page);
+			memset(ptr, 0, PAGE_SIZE);
+			dmac_flush_range(ptr, ptr + PAGE_SIZE);
+			kunmap_atomic(ptr);
+			page++;
+			size -= PAGE_SIZE;
+		}
+		outer_flush_range(base, end);
 	}
 }
 
@@ -243,7 +256,8 @@ static void __dma_free_buffer(struct page *page, size_t size)
 #endif
 
 static void *__alloc_from_contiguous(struct device *dev, size_t size,
-				     pgprot_t prot, struct page **ret_page);
+				     pgprot_t prot, struct page **ret_page,
+				     const void *caller);
 
 static void *__alloc_remap_buffer(struct device *dev, size_t size, gfp_t gfp,
 				 pgprot_t prot, struct page **ret_page,
@@ -346,10 +360,11 @@ static int __init atomic_pool_init(void)
 		goto no_pages;
 
 	if (IS_ENABLED(CONFIG_CMA))
-		ptr = __alloc_from_contiguous(NULL, pool->size, prot, &page);
+		ptr = __alloc_from_contiguous(NULL, pool->size, prot, &page,
+					      atomic_pool_init);
 	else
 		ptr = __alloc_remap_buffer(NULL, pool->size, GFP_KERNEL, prot,
-					   &page, NULL);
+					   &page, atomic_pool_init);
 	if (ptr) {
 		int i;
 
@@ -542,27 +557,41 @@ static int __free_from_pool(void *start, size_t size)
 }
 
 static void *__alloc_from_contiguous(struct device *dev, size_t size,
-				     pgprot_t prot, struct page **ret_page)
+				     pgprot_t prot, struct page **ret_page,
+				     const void *caller)
 {
 	unsigned long order = get_order(size);
 	size_t count = size >> PAGE_SHIFT;
 	struct page *page;
+	void *ptr;
 
 	page = dma_alloc_from_contiguous(dev, count, order);
 	if (!page)
 		return NULL;
 
 	__dma_clear_buffer(page, size);
-	__dma_remap(page, size, prot);
 
+	if (!PageHighMem(page)) {
+		__dma_remap(page, size, prot);
+		ptr = page_address(page);
+	} else {
+		ptr = __dma_alloc_remap(page, size, GFP_KERNEL, prot, caller);
+		if (!ptr) {
+			dma_release_from_contiguous(dev, page, count);
+			return NULL;
+		}
+	}
 	*ret_page = page;
-	return page_address(page);
+	return ptr;
 }
 
 static void __free_from_contiguous(struct device *dev, struct page *page,
-				   size_t size)
+				   void *cpu_addr, size_t size)
 {
-	__dma_remap(page, size, pgprot_kernel);
+	if (!PageHighMem(page))
+		__dma_remap(page, size, pgprot_kernel);
+	else
+		__dma_free_remap(cpu_addr, size);
 	dma_release_from_contiguous(dev, page, size >> PAGE_SHIFT);
 }
 
@@ -645,7 +674,7 @@ static void *__dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
 	else if (!IS_ENABLED(CONFIG_CMA))
 		addr = __alloc_remap_buffer(dev, size, gfp, prot, &page, caller);
 	else
-		addr = __alloc_from_contiguous(dev, size, prot, &page);
+		addr = __alloc_from_contiguous(dev, size, prot, &page, caller);
 
 	if (addr)
 		*handle = pfn_to_dma(dev, page_to_pfn(page));
@@ -739,7 +768,7 @@ static void __arm_dma_free(struct device *dev, size_t size, void *cpu_addr,
 		 * Non-atomic allocations cannot be freed with IRQs disabled
 		 */
 		WARN_ON(irqs_disabled());
-		__free_from_contiguous(dev, page, size);
+		__free_from_contiguous(dev, page, cpu_addr, size);
 	}
 }
 
-- 
1.7.9.5

[PATCHv2 1/2] ARM: dma-mapping: add support for CMA regions placed in highmem zone

[Marek Szyprowski]

This patch adds missing pieces to correctly support memory pages served
from CMA regions placed in high memory zones. Please note that the default
global CMA area is still put into lowmem and is limited by optional
architecture specific DMA zone. One can however put device specific CMA
regions in high memory zone to reduce lowmem usage.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
---

Changle log: 
v2: restructured code and made all himem checks positive 
    ('if (PageHighMem(page))' instead of 'if (!PageHighMem(page))')
---
 arch/arm/mm/dma-mapping.c |   53 ++++++++++++++++++++++++++++++++++-----------
 1 file changed, 40 insertions(+), 13 deletions(-)

diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
index 076c26d..90e059b 100644
--- a/arch/arm/mm/dma-mapping.c
+++ b/arch/arm/mm/dma-mapping.c
@@ -186,13 +186,24 @@ static u64 get_coherent_dma_mask(struct device *dev)
 
 static void __dma_clear_buffer(struct page *page, size_t size)
 {
-	void *ptr;
 	/*
 	 * Ensure that the allocated pages are zeroed, and that any data
 	 * lurking in the kernel direct-mapped region is invalidated.
 	 */
-	ptr = page_address(page);
-	if (ptr) {
+	if (PageHighMem(page)) {
+		phys_addr_t base = __pfn_to_phys(page_to_pfn(page));
+		phys_addr_t end = base + size;
+		while (size > 0) {
+			void *ptr = kmap_atomic(page);
+			memset(ptr, 0, PAGE_SIZE);
+			dmac_flush_range(ptr, ptr + PAGE_SIZE);
+			kunmap_atomic(ptr);
+			page++;
+			size -= PAGE_SIZE;
+		}
+		outer_flush_range(base, end);
+	} else {
+		void *ptr = page_address(page);
 		memset(ptr, 0, size);
 		dmac_flush_range(ptr, ptr + size);
 		outer_flush_range(__pa(ptr), __pa(ptr) + size);
@@ -243,7 +254,8 @@ static void __dma_free_buffer(struct page *page, size_t size)
 #endif
 
 static void *__alloc_from_contiguous(struct device *dev, size_t size,
-				     pgprot_t prot, struct page **ret_page);
+				     pgprot_t prot, struct page **ret_page,
+				     const void *caller);
 
 static void *__alloc_remap_buffer(struct device *dev, size_t size, gfp_t gfp,
 				 pgprot_t prot, struct page **ret_page,
@@ -346,10 +358,11 @@ static int __init atomic_pool_init(void)
 		goto no_pages;
 
 	if (IS_ENABLED(CONFIG_CMA))
-		ptr = __alloc_from_contiguous(NULL, pool->size, prot, &page);
+		ptr = __alloc_from_contiguous(NULL, pool->size, prot, &page,
+					      atomic_pool_init);
 	else
 		ptr = __alloc_remap_buffer(NULL, pool->size, GFP_KERNEL, prot,
-					   &page, NULL);
+					   &page, atomic_pool_init);
 	if (ptr) {
 		int i;
 
@@ -542,27 +555,41 @@ static int __free_from_pool(void *start, size_t size)
 }
 
 static void *__alloc_from_contiguous(struct device *dev, size_t size,
-				     pgprot_t prot, struct page **ret_page)
+				     pgprot_t prot, struct page **ret_page,
+				     const void *caller)
 {
 	unsigned long order = get_order(size);
 	size_t count = size >> PAGE_SHIFT;
 	struct page *page;
+	void *ptr;
 
 	page = dma_alloc_from_contiguous(dev, count, order);
 	if (!page)
 		return NULL;
 
 	__dma_clear_buffer(page, size);
-	__dma_remap(page, size, prot);
 
+	if (PageHighMem(page)) {
+		ptr = __dma_alloc_remap(page, size, GFP_KERNEL, prot, caller);
+		if (!ptr) {
+			dma_release_from_contiguous(dev, page, count);
+			return NULL;
+		}
+	} else {
+		__dma_remap(page, size, prot);
+		ptr = page_address(page);
+	}
 	*ret_page = page;
-	return page_address(page);
+	return ptr;
 }
 
 static void __free_from_contiguous(struct device *dev, struct page *page,
-				   size_t size)
+				   void *cpu_addr, size_t size)
 {
-	__dma_remap(page, size, pgprot_kernel);
+	if (PageHighMem(page))
+		__dma_free_remap(cpu_addr, size);
+	else
+		__dma_remap(page, size, pgprot_kernel);
 	dma_release_from_contiguous(dev, page, size >> PAGE_SHIFT);
 }
 
@@ -645,7 +672,7 @@ static void *__dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
 	else if (!IS_ENABLED(CONFIG_CMA))
 		addr = __alloc_remap_buffer(dev, size, gfp, prot, &page, caller);
 	else
-		addr = __alloc_from_contiguous(dev, size, prot, &page);
+		addr = __alloc_from_contiguous(dev, size, prot, &page, caller);
 
 	if (addr)
 		*handle = pfn_to_dma(dev, page_to_pfn(page));
@@ -739,7 +766,7 @@ static void __arm_dma_free(struct device *dev, size_t size, void *cpu_addr,
 		 * Non-atomic allocations cannot be freed with IRQs disabled
 		 */
 		WARN_ON(irqs_disabled());
-		__free_from_contiguous(dev, page, size);
+		__free_from_contiguous(dev, page, cpu_addr, size);
 	}
 }
 
-- 
1.7.9.5

[PATCHv2 1/2] ARM: dma-mapping: add support for CMA regions placed in highmem zone

[Michal Nazarewicz]

On Mon, Feb 04 2013, Marek Szyprowski wrote:
> @@ -186,13 +186,24 @@ static u64 get_coherent_dma_mask(struct device *dev)
>  
>  static void __dma_clear_buffer(struct page *page, size_t size)
>  {
> -	void *ptr;
>  	/*
>  	 * Ensure that the allocated pages are zeroed, and that any data
>  	 * lurking in the kernel direct-mapped region is invalidated.
>  	 */
> -	ptr = page_address(page);
> -	if (ptr) {
> +	if (PageHighMem(page)) {
> +		phys_addr_t base = __pfn_to_phys(page_to_pfn(page));
> +		phys_addr_t end = base + size;
> +		while (size > 0) {
> +			void *ptr = kmap_atomic(page);
> +			memset(ptr, 0, PAGE_SIZE);
> +			dmac_flush_range(ptr, ptr + PAGE_SIZE);
> +			kunmap_atomic(ptr);
> +			page++;
> +			size -= PAGE_SIZE;
> +		}
> +		outer_flush_range(base, end);
> +	} else {
> +		void *ptr = page_address(page);

There used to be a ?if (ptr)? check which is now missing.  Why is that?

>  		memset(ptr, 0, size);
>  		dmac_flush_range(ptr, ptr + size);
>  		outer_flush_range(__pa(ptr), __pa(ptr) + size);

-- 
Best regards,                                         _     _
.o. | Liege of Serenely Enlightened Majesty of      o' \,=./ `o
..o | Computer Science,  Micha? ?mina86? Nazarewicz    (o o)
ooo +----<email/xmpp: mpn@google.com>--------------ooO--(_)--Ooo--
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 835 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20130204/c8c84ff1/attachment.sig>

[PATCHv2 1/2] ARM: dma-mapping: add support for CMA regions placed in highmem zone

[Russell King - ARM Linux]

On Mon, Feb 04, 2013 at 02:51:52PM +0100, Michal Nazarewicz wrote:
> On Mon, Feb 04 2013, Marek Szyprowski wrote:
> > @@ -186,13 +186,24 @@ static u64 get_coherent_dma_mask(struct device *dev)
> >  
> >  static void __dma_clear_buffer(struct page *page, size_t size)
> >  {
> > -	void *ptr;
> >  	/*
> >  	 * Ensure that the allocated pages are zeroed, and that any data
> >  	 * lurking in the kernel direct-mapped region is invalidated.
> >  	 */
> > -	ptr = page_address(page);
> > -	if (ptr) {
> > +	if (PageHighMem(page)) {
> > +		phys_addr_t base = __pfn_to_phys(page_to_pfn(page));
> > +		phys_addr_t end = base + size;
> > +		while (size > 0) {
> > +			void *ptr = kmap_atomic(page);
> > +			memset(ptr, 0, PAGE_SIZE);
> > +			dmac_flush_range(ptr, ptr + PAGE_SIZE);
> > +			kunmap_atomic(ptr);
> > +			page++;
> > +			size -= PAGE_SIZE;
> > +		}
> > +		outer_flush_range(base, end);
> > +	} else {
> > +		void *ptr = page_address(page);
> 
> There used to be a ?if (ptr)? check which is now missing.  Why is that?

Because lowmem pages always have an address.

[PATCHv2 1/2] ARM: dma-mapping: add support for CMA regions placed in highmem zone

[Michal Nazarewicz]

On Mon, Feb 04 2013, Russell King - ARM Linux wrote:
> On Mon, Feb 04, 2013 at 02:51:52PM +0100, Michal Nazarewicz wrote:
>> On Mon, Feb 04 2013, Marek Szyprowski wrote:
>> > @@ -186,13 +186,24 @@ static u64 get_coherent_dma_mask(struct device *dev)
>> >  
>> >  static void __dma_clear_buffer(struct page *page, size_t size)
>> >  {
>> > -	void *ptr;
>> >  	/*
>> >  	 * Ensure that the allocated pages are zeroed, and that any data
>> >  	 * lurking in the kernel direct-mapped region is invalidated.
>> >  	 */
>> > -	ptr = page_address(page);
>> > -	if (ptr) {
>> > +	if (PageHighMem(page)) {
>> > +		phys_addr_t base = __pfn_to_phys(page_to_pfn(page));
>> > +		phys_addr_t end = base + size;
>> > +		while (size > 0) {
>> > +			void *ptr = kmap_atomic(page);
>> > +			memset(ptr, 0, PAGE_SIZE);
>> > +			dmac_flush_range(ptr, ptr + PAGE_SIZE);
>> > +			kunmap_atomic(ptr);
>> > +			page++;
>> > +			size -= PAGE_SIZE;
>> > +		}
>> > +		outer_flush_range(base, end);
>> > +	} else {
>> > +		void *ptr = page_address(page);
>> 
>> There used to be a ?if (ptr)? check which is now missing.  Why is that?
>
> Because lowmem pages always have an address.

Perhaps it should use lowmem_page_address() then?

-- 
Best regards,                                         _     _
.o. | Liege of Serenely Enlightened Majesty of      o' \,=./ `o
..o | Computer Science,  Micha? ?mina86? Nazarewicz    (o o)
ooo +----<email/xmpp: mpn@google.com>--------------ooO--(_)--Ooo--
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 835 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20130204/3e4952c3/attachment.sig>

[PATCHv2 1/2] ARM: dma-mapping: add support for CMA regions placed in highmem zone

[Russell King - ARM Linux]

On Mon, Feb 04, 2013 at 03:24:51PM +0100, Michal Nazarewicz wrote:
> On Mon, Feb 04 2013, Russell King - ARM Linux wrote:
> > On Mon, Feb 04, 2013 at 02:51:52PM +0100, Michal Nazarewicz wrote:
> >> On Mon, Feb 04 2013, Marek Szyprowski wrote:
> >> > @@ -186,13 +186,24 @@ static u64 get_coherent_dma_mask(struct device *dev)
> >> >  
> >> >  static void __dma_clear_buffer(struct page *page, size_t size)
> >> >  {
> >> > -	void *ptr;
> >> >  	/*
> >> >  	 * Ensure that the allocated pages are zeroed, and that any data
> >> >  	 * lurking in the kernel direct-mapped region is invalidated.
> >> >  	 */
> >> > -	ptr = page_address(page);
> >> > -	if (ptr) {
> >> > +	if (PageHighMem(page)) {
> >> > +		phys_addr_t base = __pfn_to_phys(page_to_pfn(page));
> >> > +		phys_addr_t end = base + size;
> >> > +		while (size > 0) {
> >> > +			void *ptr = kmap_atomic(page);
> >> > +			memset(ptr, 0, PAGE_SIZE);
> >> > +			dmac_flush_range(ptr, ptr + PAGE_SIZE);
> >> > +			kunmap_atomic(ptr);
> >> > +			page++;
> >> > +			size -= PAGE_SIZE;
> >> > +		}
> >> > +		outer_flush_range(base, end);
> >> > +	} else {
> >> > +		void *ptr = page_address(page);
> >> 
> >> There used to be a ?if (ptr)? check which is now missing.  Why is that?
> >
> > Because lowmem pages always have an address.
> 
> Perhaps it should use lowmem_page_address() then?

It's well defined that page_address() will be non-NULL for lowmem pages.
There is nothing wrong with the above.

[PATCHv2 1/2] ARM: dma-mapping: add support for CMA regions placed in highmem zone

[Michal Nazarewicz]

> On Mon, Feb 04, 2013 at 03:24:51PM +0100, Michal Nazarewicz wrote:
>> Perhaps it should use lowmem_page_address() then?

On Mon, Feb 04 2013, Russell King - ARM Linux wrote:
> It's well defined that page_address() will be non-NULL for lowmem pages.
> There is nothing wrong with the above.

It would save on a branch and a function call though.

-- 
Best regards,                                         _     _
.o. | Liege of Serenely Enlightened Majesty of      o' \,=./ `o
..o | Computer Science,  Micha? ?mina86? Nazarewicz    (o o)
ooo +----<email/xmpp: mpn@google.com>--------------ooO--(_)--Ooo--
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 835 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20130204/4533538f/attachment-0001.sig>

[PATCHv2 1/2] ARM: dma-mapping: add support for CMA regions placed in highmem zone

[Russell King - ARM Linux]

On Mon, Feb 04, 2013 at 03:42:26PM +0100, Michal Nazarewicz wrote:
> > On Mon, Feb 04, 2013 at 03:24:51PM +0100, Michal Nazarewicz wrote:
> >> Perhaps it should use lowmem_page_address() then?
> 
> On Mon, Feb 04 2013, Russell King - ARM Linux wrote:
> > It's well defined that page_address() will be non-NULL for lowmem pages.
> > There is nothing wrong with the above.
> 
> It would save on a branch and a function call though.

Depending on the size of struct page, if we care that much, we can probably
enable WANT_PAGE_VIRTUAL which'll make it even cheaper as it's just a
dereference.

[PATCHv2 1/2] ARM: dma-mapping: add support for CMA regions placed in highmem zone

[Michal Nazarewicz]

On Mon, Feb 04 2013, Marek Szyprowski wrote:
> This patch adds missing pieces to correctly support memory pages served
> from CMA regions placed in high memory zones. Please note that the default
> global CMA area is still put into lowmem and is limited by optional
> architecture specific DMA zone. One can however put device specific CMA
> regions in high memory zone to reduce lowmem usage.
>
> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
> Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>

Acked-by: Michal Nazarewicz <mina86@mina86.com>

-- 
Best regards,                                         _     _
.o. | Liege of Serenely Enlightened Majesty of      o' \,=./ `o
..o | Computer Science,  Micha? ?mina86? Nazarewicz    (o o)
ooo +----<email/xmpp: mpn@google.com>--------------ooO--(_)--Ooo--
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 835 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20130204/025f50b8/attachment.sig>

[PATCHv3 1/2] ARM: dma-mapping: add support for CMA regions placed in highmem zone

[Marek Szyprowski]

This patch adds missing pieces to correctly support memory pages served
from CMA regions placed in high memory zones. Please note that the default
global CMA area is still put into lowmem and is limited by optional
architecture specific DMA zone. One can however put device specific CMA
regions in high memory zone to reduce lowmem usage.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
Acked-by: Michal Nazarewicz <mina86@mina86.com>
---
Changle log: 
v3: fixed build break for non-MMU builds (thanks to Thierry Reding!)

v2: restructured code and made all himem checks positive 
    ('if (PageHighMem(page))' instead of 'if (!PageHighMem(page))')
---
 arch/arm/mm/dma-mapping.c |   57 +++++++++++++++++++++++++++++++++------------
 1 file changed, 42 insertions(+), 15 deletions(-)

diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
index de93ecd..5f79361 100644
--- a/arch/arm/mm/dma-mapping.c
+++ b/arch/arm/mm/dma-mapping.c
@@ -186,13 +186,24 @@ static u64 get_coherent_dma_mask(struct device *dev)
 
 static void __dma_clear_buffer(struct page *page, size_t size)
 {
-	void *ptr;
 	/*
 	 * Ensure that the allocated pages are zeroed, and that any data
 	 * lurking in the kernel direct-mapped region is invalidated.
 	 */
-	ptr = page_address(page);
-	if (ptr) {
+	if (PageHighMem(page)) {
+		phys_addr_t base = __pfn_to_phys(page_to_pfn(page));
+		phys_addr_t end = base + size;
+		while (size > 0) {
+			void *ptr = kmap_atomic(page);
+			memset(ptr, 0, PAGE_SIZE);
+			dmac_flush_range(ptr, ptr + PAGE_SIZE);
+			kunmap_atomic(ptr);
+			page++;
+			size -= PAGE_SIZE;
+		}
+		outer_flush_range(base, end);
+	} else {
+		void *ptr = page_address(page);
 		memset(ptr, 0, size);
 		dmac_flush_range(ptr, ptr + size);
 		outer_flush_range(__pa(ptr), __pa(ptr) + size);
@@ -243,7 +254,8 @@ static void __dma_free_buffer(struct page *page, size_t size)
 #endif
 
 static void *__alloc_from_contiguous(struct device *dev, size_t size,
-				     pgprot_t prot, struct page **ret_page);
+				     pgprot_t prot, struct page **ret_page,
+				     const void *caller);
 
 static void *__alloc_remap_buffer(struct device *dev, size_t size, gfp_t gfp,
 				 pgprot_t prot, struct page **ret_page,
@@ -346,10 +358,11 @@ static int __init atomic_pool_init(void)
 		goto no_pages;
 
 	if (IS_ENABLED(CONFIG_CMA))
-		ptr = __alloc_from_contiguous(NULL, pool->size, prot, &page);
+		ptr = __alloc_from_contiguous(NULL, pool->size, prot, &page,
+					      atomic_pool_init);
 	else
 		ptr = __alloc_remap_buffer(NULL, pool->size, GFP_KERNEL, prot,
-					   &page, NULL);
+					   &page, atomic_pool_init);
 	if (ptr) {
 		int i;
 
@@ -542,27 +555,41 @@ static int __free_from_pool(void *start, size_t size)
 }
 
 static void *__alloc_from_contiguous(struct device *dev, size_t size,
-				     pgprot_t prot, struct page **ret_page)
+				     pgprot_t prot, struct page **ret_page,
+				     const void *caller)
 {
 	unsigned long order = get_order(size);
 	size_t count = size >> PAGE_SHIFT;
 	struct page *page;
+	void *ptr;
 
 	page = dma_alloc_from_contiguous(dev, count, order);
 	if (!page)
 		return NULL;
 
 	__dma_clear_buffer(page, size);
-	__dma_remap(page, size, prot);
 
+	if (PageHighMem(page)) {
+		ptr = __dma_alloc_remap(page, size, GFP_KERNEL, prot, caller);
+		if (!ptr) {
+			dma_release_from_contiguous(dev, page, count);
+			return NULL;
+		}
+	} else {
+		__dma_remap(page, size, prot);
+		ptr = page_address(page);
+	}
 	*ret_page = page;
-	return page_address(page);
+	return ptr;
 }
 
 static void __free_from_contiguous(struct device *dev, struct page *page,
-				   size_t size)
+				   void *cpu_addr, size_t size)
 {
-	__dma_remap(page, size, pgprot_kernel);
+	if (PageHighMem(page))
+		__dma_free_remap(cpu_addr, size);
+	else
+		__dma_remap(page, size, pgprot_kernel);
 	dma_release_from_contiguous(dev, page, size >> PAGE_SHIFT);
 }
 
@@ -583,9 +610,9 @@ static inline pgprot_t __get_dma_pgprot(struct dma_attrs *attrs, pgprot_t prot)
 #define __get_dma_pgprot(attrs, prot)	__pgprot(0)
 #define __alloc_remap_buffer(dev, size, gfp, prot, ret, c)	NULL
 #define __alloc_from_pool(size, ret_page)			NULL
-#define __alloc_from_contiguous(dev, size, prot, ret)		NULL
+#define __alloc_from_contiguous(dev, size, prot, ret, c)	NULL
 #define __free_from_pool(cpu_addr, size)			0
-#define __free_from_contiguous(dev, page, size)			do { } while (0)
+#define __free_from_contiguous(dev, page, cpu_addr, size)	do { } while (0)
 #define __dma_free_remap(cpu_addr, size)			do { } while (0)
 
 #endif	/* CONFIG_MMU */
@@ -645,7 +672,7 @@ static void *__dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
 	else if (!IS_ENABLED(CONFIG_CMA))
 		addr = __alloc_remap_buffer(dev, size, gfp, prot, &page, caller);
 	else
-		addr = __alloc_from_contiguous(dev, size, prot, &page);
+		addr = __alloc_from_contiguous(dev, size, prot, &page, caller);
 
 	if (addr)
 		*handle = pfn_to_dma(dev, page_to_pfn(page));
@@ -739,7 +766,7 @@ static void __arm_dma_free(struct device *dev, size_t size, void *cpu_addr,
 		 * Non-atomic allocations cannot be freed with IRQs disabled
 		 */
 		WARN_ON(irqs_disabled());
-		__free_from_contiguous(dev, page, size);
+		__free_from_contiguous(dev, page, cpu_addr, size);
 	}
 }
 
-- 
1.7.9.5

[PATCH 2/2] ARM: dma-mapping: use himem for DMA buffers for IOMMU-mapped devices

[Marek Szyprowski]

IOMMU can provide access to any memory page, so there is no point in
limiting the allocated pages only to lowmem, once other parts of
dma-mapping subsystem correctly supports himem pages.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
---
 arch/arm/mm/dma-mapping.c |    9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
index 4080c37..9a6c8ce 100644
--- a/arch/arm/mm/dma-mapping.c
+++ b/arch/arm/mm/dma-mapping.c
@@ -1095,12 +1095,17 @@ static struct page **__iommu_alloc_buffer(struct device *dev, size_t size,
 		return pages;
 	}
 
+	/*
+	 * IOMMU can map any pages, so himem can also be used here
+	 */
+	gfp |= __GFP_NOWARN | __GFP_HIGHMEM;
+
 	while (count) {
 		int j, order = __fls(count);
 
-		pages[i] = alloc_pages(gfp | __GFP_NOWARN, order);
+		pages[i] = alloc_pages(gfp, order);
 		while (!pages[i] && order)
-			pages[i] = alloc_pages(gfp | __GFP_NOWARN, --order);
+			pages[i] = alloc_pages(gfp, --order);
 		if (!pages[i])
 			goto error;
 
-- 
1.7.9.5