[PATCH 0/4] [RFC] Addition of dma-mapping IOMMU framework in Exynos DRM

[PATCH 0/4] [RFC] Addition of dma-mapping IOMMU framework in Exynos DRM

[Prathyush]

From: Prathyush K <prathyush.k@samsung.com>

With these set of patches, the Exynos5 DRM module will be able to
allocate, export and map non-contiguous memory for GEM objects using the
dma-mapping IOMMU framework.

These patches are based on the following references:

1> Exynos 5 SYSMMU driver by Kyongho Cho
http://lists.infradead.org/pipermail/linux-arm-kernel/2011-December/078636.htm
2> DMA-MAPPING IOMMU framework by Marek Szyprowski
http://lists.infradead.org/pipermail/linux-arm-kernel/2011-June/053636.html
3> DRM-PRIME by Dave Airlie
http://patches.linaro.org/7480/
4> DRM Exynos DMABUF module by Inki Dae
http://git.infradead.org/users/kmpark/linux-2.6-samsung/commit/
d59d3e521d8a9d27e7b329f89f24a8473f32f50c/
5> Support for dma_get_pages by Tomasz Stanislawski
http://patchwork.linuxtv.org/patch/9639/

The updated GEM framework by Inki Dae is not considered for this patch set.

IOMMU mapping is created for exynos-drm-device initially.
Allocation of GEM objects happens through dma_alloc_writecombine which
inturn calls arm_iommu_alloc_attrs. This will allocate a non-contig set of
pages, allocate a iova and map all the pages.

The following changes are done to support non-contiguous memory for GEM objects

  [PATCH 1/4]:
	drm/exynos: DMABUF: Added support for exporting non-contig buffers.
	This patch is for creating and mapping a SGT in map_dma_buf by
	retrieving the pages by calling dma_get_pages.

  [PATCH 2/4]:
	drm/exynos: Mapping of gem objects uses dma_mmap_writecombine.
	This patch is for mapping the non-contiguous GEM objects to user
	space by calling dma_mmap_writecombine.

  [PATCH 3/4]:
	drm/Exynos: Added 'disable' function to Exynos drm crtc module.
	This patch is required for safe release of DRM. If a FB of a CRTC
	is released, the driver tries to disable the CRTC if that is
	supported by the CRTC. This patch adds Exynos DRM CRTC disable
	functionality.

  [PATCH 4/4]:
	drm: Releasing FBs before releasing GEM objects during drm_release.
	This patch is required for safe release of DRM. During drm release,
	all the FBs and GEM objects are released. A gem object which is used
	as a FB must not be freed first before releasing the FB. This patch
	modifies the drm release function to first release the FBs and then
	release the GEMs.

 drivers/gpu/drm/drm_fops.c                 |    6 +-
 drivers/gpu/drm/exynos/exynos_drm_crtc.c   |   11 +++
 drivers/gpu/drm/exynos/exynos_drm_dmabuf.c |   98 +++++++++++++++++++++++-----
 drivers/gpu/drm/exynos/exynos_drm_gem.c    |   70 +++++++++-----------
 4 files changed, 128 insertions(+), 57 deletions(-)

[PATCH 1/4] [RFC] drm/exynos: DMABUF: Added support for exporting non-contig buffers

[Prathyush]

With this change, the exynos drm dmabuf module can export and
import dmabuf of gem objects with non-continuous memory.

The exynos_map_dmabuf function can create SGT of a non-contiguous
buffer by calling dma_get_pages to retrieve the allocated pages
and then maps the SGT to the caller's address space.

Signed-off-by: Prathyush K <prathyush.k@samsung.com>
---
 drivers/gpu/drm/exynos/exynos_drm_dmabuf.c |   98 +++++++++++++++++++++++-----
 1 files changed, 81 insertions(+), 17 deletions(-)

diff --git a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c
index cbb6ad4..54b88bd 100644
--- a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c
+++ b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c
@@ -56,6 +56,59 @@ static void exynos_dmabuf_detach(struct dma_buf *dmabuf,
 	dma_buf_put(dmabuf);
 }
 
+
+static struct sg_table *drm_dc_pages_to_sgt(struct page **pages,
+	unsigned long n_pages, size_t offset, size_t offset2, dma_addr_t daddr)
+{
+	struct sg_table *sgt;
+	struct scatterlist *s;
+	int i, j, cur_page, chunks, ret;
+
+	sgt = kzalloc(sizeof *sgt, GFP_KERNEL);
+	if (!sgt)
+		return ERR_PTR(-ENOMEM);
+
+	/* compute number of chunks */
+	chunks = 1;
+	for (i = 1; i < n_pages; ++i)
+		if (pages[i] != pages[i - 1] + 1)
+			++chunks;
+
+	ret = sg_alloc_table(sgt, chunks, GFP_KERNEL);
+	if (ret) {
+		kfree(sgt);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	/* merging chunks and putting them into the scatterlist */
+	cur_page = 0;
+	for_each_sg(sgt->sgl, s, sgt->orig_nents, i) {
+		size_t size = PAGE_SIZE;
+
+		for (j = cur_page + 1; j < n_pages; ++j) {
+			if (pages[j] != pages[j - 1] + 1)
+				break;
+			size += PAGE_SIZE;
+		}
+
+		/* cut offset if chunk starts at the first page */
+		if (cur_page == 0)
+			size -= offset;
+		/* cut offset2 if chunk ends at the last page */
+		if (j == n_pages)
+			size -= offset2;
+
+		sg_set_page(s, pages[cur_page], size, offset);
+		s->dma_address = daddr;
+		daddr +=  size;
+		offset = 0;
+		cur_page = j;
+	}
+
+	return sgt;
+}
+
+
 static struct sg_table *exynos_map_dmabuf(struct dma_buf_attachment *attach,
 					enum dma_data_direction direction)
 {
@@ -64,6 +117,8 @@ static struct sg_table *exynos_map_dmabuf(struct dma_buf_attachment *attach,
 	struct exynos_drm_gem_buf *buffer;
 	struct sg_table *sgt;
 	int ret;
+	int size, n_pages;
+	struct page **pages = NULL;
 
 	DRM_DEBUG_KMS("%s\n", __FILE__);
 
@@ -71,27 +126,37 @@ static struct sg_table *exynos_map_dmabuf(struct dma_buf_attachment *attach,
 
 	buffer = exynos_gem_obj->buffer;
 
-	/* TODO. consider physically non-continuous memory with IOMMU. */
+	size = buffer->size;
+	n_pages = PAGE_ALIGN(size) >> PAGE_SHIFT;
 
-	sgt = kzalloc(sizeof(*sgt), GFP_KERNEL);
-	if (!sgt) {
-		DRM_DEBUG_KMS("failed to allocate sg table.\n");
-		return ERR_PTR(-ENOMEM);
+	pages = kmalloc(n_pages * sizeof pages[0], GFP_KERNEL);
+	if (!pages) {
+		DRM_DEBUG_KMS("failed to alloc page table\n");
+		return NULL;
 	}
 
-	ret = sg_alloc_table(sgt, 1, GFP_KERNEL);
+	ret = dma_get_pages(attach->dev, buffer->kvaddr,
+				buffer->dma_addr, pages, n_pages);
 	if (ret < 0) {
-		DRM_DEBUG_KMS("failed to allocate scatter list.\n");
-		kfree(sgt);
-		sgt = NULL;
-		return ERR_PTR(-ENOMEM);
+		DRM_DEBUG_KMS("failed to get buffer pages from DMA API\n");
+		return NULL;
 	}
+	if (ret != n_pages) {
+		DRM_DEBUG_KMS("failed to get all pages from DMA API\n");
+		return NULL;
+	}
+
+	sgt = drm_dc_pages_to_sgt(pages, n_pages, 0, 0, buffer->dma_addr);
+	if (IS_ERR(sgt)) {
+		DRM_DEBUG_KMS("failed to prepare sg table\n");
+		return NULL;
+	}
+
+	sgt->nents = dma_map_sg(attach->dev, sgt->sgl,
+				sgt->orig_nents, DMA_BIDIRECTIONAL);
 
-	sg_init_table(sgt->sgl, 1);
-	sg_dma_len(sgt->sgl) = buffer->size;
-	sg_set_page(sgt->sgl, pfn_to_page(PFN_DOWN(buffer->dma_addr)),
-			buffer->size, 0);
-	sg_dma_address(sgt->sgl) = buffer->dma_addr;
+	/* pages are no longer needed */
+	kfree(pages);
 
 	/*
 	 * increase reference count of this buffer.
@@ -303,8 +368,6 @@ int exynos_dmabuf_prime_fd_to_handle(struct drm_device *drm_dev,
 	if (ret < 0)
 		goto fail_handle;
 
-	/* consider physically non-continuous memory with IOMMU. */
-
 	buffer->dma_addr = sg_dma_address(sgt->sgl);
 	buffer->size = sg_dma_len(sgt->sgl);
 
@@ -316,6 +379,7 @@ int exynos_dmabuf_prime_fd_to_handle(struct drm_device *drm_dev,
 	atomic_set(&buffer->shared_refcount, 1);
 
 	exynos_gem_obj->base.import_attach = attach;
+	exynos_gem_obj->buffer = buffer;
 
 	ret = drm_prime_insert_fd_handle_mapping(&file_priv->prime,
 							dmabuf, *handle);
-- 
1.7.0.4

Re: [PATCH 1/4] [RFC] drm/exynos: DMABUF: Added support for exporting non-contig buffers

[Kyungmin Park]

Hi,

I saw your description but please generate the patches against the
latest codes at next time.

On 4/14/12, Prathyush <prathyush.k@samsung.com> wrote:
> With this change, the exynos drm dmabuf module can export and
> import dmabuf of gem objects with non-continuous memory.
>
> The exynos_map_dmabuf function can create SGT of a non-contiguous
> buffer by calling dma_get_pages to retrieve the allocated pages
> and then maps the SGT to the caller's address space.
>
> Signed-off-by: Prathyush K <prathyush.k@samsung.com>
> ---
>  drivers/gpu/drm/exynos/exynos_drm_dmabuf.c |   98
> +++++++++++++++++++++++-----
>  1 files changed, 81 insertions(+), 17 deletions(-)
>
> diff --git a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c
> b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c
> index cbb6ad4..54b88bd 100644
> --- a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c
> +++ b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c
> @@ -56,6 +56,59 @@ static void exynos_dmabuf_detach(struct dma_buf *dmabuf,
>  	dma_buf_put(dmabuf);
>  }
>
> +
Useless newline.
> +static struct sg_table *drm_dc_pages_to_sgt(struct page **pages,
> +	unsigned long n_pages, size_t offset, size_t offset2, dma_addr_t daddr)
> +{
> +	struct sg_table *sgt;
> +	struct scatterlist *s;
> +	int i, j, cur_page, chunks, ret;
> +
> +	sgt = kzalloc(sizeof *sgt, GFP_KERNEL);
> +	if (!sgt)
> +		return ERR_PTR(-ENOMEM);
> +
> +	/* compute number of chunks */
> +	chunks = 1;
> +	for (i = 1; i < n_pages; ++i)
> +		if (pages[i] != pages[i - 1] + 1)
> +			++chunks;
> +
> +	ret = sg_alloc_table(sgt, chunks, GFP_KERNEL);
> +	if (ret) {
> +		kfree(sgt);
> +		return ERR_PTR(-ENOMEM);
> +	}
> +
> +	/* merging chunks and putting them into the scatterlist */
> +	cur_page = 0;
> +	for_each_sg(sgt->sgl, s, sgt->orig_nents, i) {
> +		size_t size = PAGE_SIZE;
> +
> +		for (j = cur_page + 1; j < n_pages; ++j) {
> +			if (pages[j] != pages[j - 1] + 1)
> +				break;
> +			size += PAGE_SIZE;
> +		}
> +
> +		/* cut offset if chunk starts at the first page */
> +		if (cur_page == 0)
> +			size -= offset;
> +		/* cut offset2 if chunk ends at the last page */
> +		if (j == n_pages)
> +			size -= offset2;
> +
> +		sg_set_page(s, pages[cur_page], size, offset);
> +		s->dma_address = daddr;
> +		daddr +=  size;
> +		offset = 0;
> +		cur_page = j;
> +	}
> +
> +	return sgt;
> +}
This function is almost same as Tomasz one at v4l2.
To Tomasz,
Can you make it common helper function at common place?
> +
> +
Useless newline too.
>  static struct sg_table *exynos_map_dmabuf(struct dma_buf_attachment
> *attach,
>  					enum dma_data_direction direction)
>  {
> @@ -64,6 +117,8 @@ static struct sg_table *exynos_map_dmabuf(struct
> dma_buf_attachment *attach,
>  	struct exynos_drm_gem_buf *buffer;
>  	struct sg_table *sgt;
>  	int ret;
> +	int size, n_pages;
> +	struct page **pages = NULL;
>
>  	DRM_DEBUG_KMS("%s\n", __FILE__);
>
> @@ -71,27 +126,37 @@ static struct sg_table *exynos_map_dmabuf(struct
> dma_buf_attachment *attach,
>
>  	buffer = exynos_gem_obj->buffer;
>
> -	/* TODO. consider physically non-continuous memory with IOMMU. */
> +	size = buffer->size;
> +	n_pages = PAGE_ALIGN(size) >> PAGE_SHIFT;
>
> -	sgt = kzalloc(sizeof(*sgt), GFP_KERNEL);
> -	if (!sgt) {
> -		DRM_DEBUG_KMS("failed to allocate sg table.\n");
> -		return ERR_PTR(-ENOMEM);
> +	pages = kmalloc(n_pages * sizeof pages[0], GFP_KERNEL);
kcalloc?

Thank you,
Kyungmin Park
> +	if (!pages) {
> +		DRM_DEBUG_KMS("failed to alloc page table\n");
> +		return NULL;
>  	}
>
> -	ret = sg_alloc_table(sgt, 1, GFP_KERNEL);
> +	ret = dma_get_pages(attach->dev, buffer->kvaddr,
> +				buffer->dma_addr, pages, n_pages);
>  	if (ret < 0) {
> -		DRM_DEBUG_KMS("failed to allocate scatter list.\n");
> -		kfree(sgt);
> -		sgt = NULL;
> -		return ERR_PTR(-ENOMEM);
> +		DRM_DEBUG_KMS("failed to get buffer pages from DMA API\n");
> +		return NULL;
>  	}
> +	if (ret != n_pages) {
> +		DRM_DEBUG_KMS("failed to get all pages from DMA API\n");
> +		return NULL;
> +	}
> +
> +	sgt = drm_dc_pages_to_sgt(pages, n_pages, 0, 0, buffer->dma_addr);
> +	if (IS_ERR(sgt)) {
> +		DRM_DEBUG_KMS("failed to prepare sg table\n");
> +		return NULL;
> +	}
> +
> +	sgt->nents = dma_map_sg(attach->dev, sgt->sgl,
> +				sgt->orig_nents, DMA_BIDIRECTIONAL);
>
> -	sg_init_table(sgt->sgl, 1);
> -	sg_dma_len(sgt->sgl) = buffer->size;
> -	sg_set_page(sgt->sgl, pfn_to_page(PFN_DOWN(buffer->dma_addr)),
> -			buffer->size, 0);
> -	sg_dma_address(sgt->sgl) = buffer->dma_addr;
> +	/* pages are no longer needed */
> +	kfree(pages);
>
>  	/*
>  	 * increase reference count of this buffer.
> @@ -303,8 +368,6 @@ int exynos_dmabuf_prime_fd_to_handle(struct drm_device
> *drm_dev,
>  	if (ret < 0)
>  		goto fail_handle;
>
> -	/* consider physically non-continuous memory with IOMMU. */
> -
>  	buffer->dma_addr = sg_dma_address(sgt->sgl);
>  	buffer->size = sg_dma_len(sgt->sgl);
>
> @@ -316,6 +379,7 @@ int exynos_dmabuf_prime_fd_to_handle(struct drm_device
> *drm_dev,
>  	atomic_set(&buffer->shared_refcount, 1);
>
>  	exynos_gem_obj->base.import_attach = attach;
> +	exynos_gem_obj->buffer = buffer;
>
>  	ret = drm_prime_insert_fd_handle_mapping(&file_priv->prime,
>  							dmabuf, *handle);
> --
> 1.7.0.4
>
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/dri-devel
>

Re: [PATCH 1/4] [RFC] drm/exynos: DMABUF: Added support for exporting non-contig buffers

[Tomasz Stanislawski]

Dear Mr. Park,

I will prepare and post a patch to lib/scatterlist.c that adds
support for initialization of sg_table from page array.

Yours sincerely,
Tomasz Stanislawski

RE: [PATCH 1/4] [RFC] drm/exynos: DMABUF: Added support for exporting non-contig buffers

[Inki Dae]

Hi Prathyush.

First of all, thanks for your patch but your patch set isn't considered for
latest exynos dmabuf. 
we already updated dmabuf support for exynos drm and this is also considered
for non-continuous memory region. you can refer to git repository below:
	
http://git.infradead.org/users/kmpark/linux-2.6-samsung/shortlog/refs/heads/
exynos-drm-prime

and this repository has already been tested by me and Tomasz. for this,
please refer to link below:
	http://www.spinics.net/lists/linux-media/msg46292.html

last week, we had posted exynos-drm-fixes update and it has been merged to
drm-fixes last week. so we are just waiting for drm-next update to post
latest exynos-drm-prime. sorry but please, re-sync your patch set with
latest Dave's drm-next after latest exynos-drm-prime is posted so your patch
set can't be reviewed yet.

Thanks,
Inki Dae.

> -----Original Message-----
> From: Prathyush [mailto:prathyush.k@samsung.com]
> Sent: Saturday, April 14, 2012 8:52 PM
> To: dri-devel@lists.freedesktop.org; linaro-mm-sig@lists.linaro.org
> Cc: inki.dae@samsung.com; subash.rp@samsung.com; prashanth.g@samsung.com;
> sunilm@samsung.com; prathyush.k@samsung.com
> Subject: [PATCH 1/4] [RFC] drm/exynos: DMABUF: Added support for exporting
> non-contig buffers
> 
> With this change, the exynos drm dmabuf module can export and
> import dmabuf of gem objects with non-continuous memory.
> 
> The exynos_map_dmabuf function can create SGT of a non-contiguous
> buffer by calling dma_get_pages to retrieve the allocated pages
> and then maps the SGT to the caller's address space.
> 
> Signed-off-by: Prathyush K <prathyush.k@samsung.com>
> ---
>  drivers/gpu/drm/exynos/exynos_drm_dmabuf.c |   98
+++++++++++++++++++++++-
> ----
>  1 files changed, 81 insertions(+), 17 deletions(-)
> 
> diff --git a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c
> b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c
> index cbb6ad4..54b88bd 100644
> --- a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c
> +++ b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c
> @@ -56,6 +56,59 @@ static void exynos_dmabuf_detach(struct dma_buf
*dmabuf,
>  	dma_buf_put(dmabuf);
>  }
> 
> +
> +static struct sg_table *drm_dc_pages_to_sgt(struct page **pages,
> +	unsigned long n_pages, size_t offset, size_t offset2, dma_addr_t
> daddr)
> +{
> +	struct sg_table *sgt;
> +	struct scatterlist *s;
> +	int i, j, cur_page, chunks, ret;
> +
> +	sgt = kzalloc(sizeof *sgt, GFP_KERNEL);
> +	if (!sgt)
> +		return ERR_PTR(-ENOMEM);
> +
> +	/* compute number of chunks */
> +	chunks = 1;
> +	for (i = 1; i < n_pages; ++i)
> +		if (pages[i] != pages[i - 1] + 1)
> +			++chunks;
> +
> +	ret = sg_alloc_table(sgt, chunks, GFP_KERNEL);
> +	if (ret) {
> +		kfree(sgt);
> +		return ERR_PTR(-ENOMEM);
> +	}
> +
> +	/* merging chunks and putting them into the scatterlist */
> +	cur_page = 0;
> +	for_each_sg(sgt->sgl, s, sgt->orig_nents, i) {
> +		size_t size = PAGE_SIZE;
> +
> +		for (j = cur_page + 1; j < n_pages; ++j) {
> +			if (pages[j] != pages[j - 1] + 1)
> +				break;
> +			size += PAGE_SIZE;
> +		}
> +
> +		/* cut offset if chunk starts at the first page */
> +		if (cur_page == 0)
> +			size -= offset;
> +		/* cut offset2 if chunk ends at the last page */
> +		if (j == n_pages)
> +			size -= offset2;
> +
> +		sg_set_page(s, pages[cur_page], size, offset);
> +		s->dma_address = daddr;
> +		daddr +=  size;
> +		offset = 0;
> +		cur_page = j;
> +	}
> +
> +	return sgt;
> +}
> +
> +
>  static struct sg_table *exynos_map_dmabuf(struct dma_buf_attachment
> *attach,
>  					enum dma_data_direction direction)
>  {
> @@ -64,6 +117,8 @@ static struct sg_table *exynos_map_dmabuf(struct
> dma_buf_attachment *attach,
>  	struct exynos_drm_gem_buf *buffer;
>  	struct sg_table *sgt;
>  	int ret;
> +	int size, n_pages;
> +	struct page **pages = NULL;
> 
>  	DRM_DEBUG_KMS("%s\n", __FILE__);
> 
> @@ -71,27 +126,37 @@ static struct sg_table *exynos_map_dmabuf(struct
> dma_buf_attachment *attach,
> 
>  	buffer = exynos_gem_obj->buffer;
> 
> -	/* TODO. consider physically non-continuous memory with IOMMU. */
> +	size = buffer->size;
> +	n_pages = PAGE_ALIGN(size) >> PAGE_SHIFT;
> 
> -	sgt = kzalloc(sizeof(*sgt), GFP_KERNEL);
> -	if (!sgt) {
> -		DRM_DEBUG_KMS("failed to allocate sg table.\n");
> -		return ERR_PTR(-ENOMEM);
> +	pages = kmalloc(n_pages * sizeof pages[0], GFP_KERNEL);
> +	if (!pages) {
> +		DRM_DEBUG_KMS("failed to alloc page table\n");
> +		return NULL;
>  	}
> 
> -	ret = sg_alloc_table(sgt, 1, GFP_KERNEL);
> +	ret = dma_get_pages(attach->dev, buffer->kvaddr,
> +				buffer->dma_addr, pages, n_pages);
>  	if (ret < 0) {
> -		DRM_DEBUG_KMS("failed to allocate scatter list.\n");
> -		kfree(sgt);
> -		sgt = NULL;
> -		return ERR_PTR(-ENOMEM);
> +		DRM_DEBUG_KMS("failed to get buffer pages from DMA API\n");
> +		return NULL;
>  	}
> +	if (ret != n_pages) {
> +		DRM_DEBUG_KMS("failed to get all pages from DMA API\n");
> +		return NULL;
> +	}
> +
> +	sgt = drm_dc_pages_to_sgt(pages, n_pages, 0, 0, buffer->dma_addr);
> +	if (IS_ERR(sgt)) {
> +		DRM_DEBUG_KMS("failed to prepare sg table\n");
> +		return NULL;
> +	}
> +
> +	sgt->nents = dma_map_sg(attach->dev, sgt->sgl,
> +				sgt->orig_nents, DMA_BIDIRECTIONAL);
> 
> -	sg_init_table(sgt->sgl, 1);
> -	sg_dma_len(sgt->sgl) = buffer->size;
> -	sg_set_page(sgt->sgl, pfn_to_page(PFN_DOWN(buffer->dma_addr)),
> -			buffer->size, 0);
> -	sg_dma_address(sgt->sgl) = buffer->dma_addr;
> +	/* pages are no longer needed */
> +	kfree(pages);
> 
>  	/*
>  	 * increase reference count of this buffer.
> @@ -303,8 +368,6 @@ int exynos_dmabuf_prime_fd_to_handle(struct drm_device
> *drm_dev,
>  	if (ret < 0)
>  		goto fail_handle;
> 
> -	/* consider physically non-continuous memory with IOMMU. */
> -
>  	buffer->dma_addr = sg_dma_address(sgt->sgl);
>  	buffer->size = sg_dma_len(sgt->sgl);
> 
> @@ -316,6 +379,7 @@ int exynos_dmabuf_prime_fd_to_handle(struct drm_device
> *drm_dev,
>  	atomic_set(&buffer->shared_refcount, 1);
> 
>  	exynos_gem_obj->base.import_attach = attach;
> +	exynos_gem_obj->buffer = buffer;
> 
>  	ret = drm_prime_insert_fd_handle_mapping(&file_priv->prime,
>  							dmabuf, *handle);
> --
> 1.7.0.4

[PATCH 2/4] [RFC] drm/exynos: Mapping of gem objects uses dma_mmap_writecombine

[Prathyush]

GEM objects get mapped to user space in two ways - DIRECT and
INDIRECT mapping. DIRECT mapping is by calling an ioctl and it
maps all the pages to user space by calling remap-pfn-range.
Indirect mapping is done by calling 'mmap'. The actual mapping
is done when a page fault gets generated and is handled by
exynos_drm_gem_fault function where the required page is mapped.
Both the methods assume contiguous memory.

With this change, the mapping is done by dma_mmap_writecombine
which will support mapping of non-contiguous memory to user space.

This works similar to the previous approach for the case of
DIRECT mapping. But in the case of mapping when a page fault
occurs, dma_mmap_writecombine will map all the pages and not just
one page.

Signed-off-by: Prathyush K <prathyush.k@samsung.com>
---
 drivers/gpu/drm/exynos/exynos_drm_gem.c |   70 ++++++++++++++----------------
 1 files changed, 33 insertions(+), 37 deletions(-)

diff --git a/drivers/gpu/drm/exynos/exynos_drm_gem.c b/drivers/gpu/drm/exynos/exynos_drm_gem.c
index 807143e..a57a83a 100755
--- a/drivers/gpu/drm/exynos/exynos_drm_gem.c
+++ b/drivers/gpu/drm/exynos/exynos_drm_gem.c
@@ -200,40 +200,27 @@ static int exynos_drm_gem_mmap_buffer(struct file *filp,
 {
 	struct drm_gem_object *obj = filp->private_data;
 	struct exynos_drm_gem_obj *exynos_gem_obj = to_exynos_gem_obj(obj);
-	struct exynos_drm_gem_buf *buffer;
-	unsigned long pfn, vm_size;
-
-	DRM_DEBUG_KMS("%s\n", __FILE__);
+	void *kva;
+	dma_addr_t dma_address;
+	unsigned long ret;
 
-	vma->vm_flags |= (VM_IO | VM_RESERVED);
+	kva = exynos_gem_obj->buffer->kvaddr;
 
-	/* in case of direct mapping, always having non-cachable attribute */
-	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
-	vma->vm_file = filp;
-
-	vm_size = vma->vm_end - vma->vm_start;
-	/*
-	 * a buffer contains information to physically continuous memory
-	 * allocated by user request or at framebuffer creation.
-	 */
-	buffer = exynos_gem_obj->buffer;
-
-	/* check if user-requested size is valid. */
-	if (vm_size > buffer->size)
-		return -EINVAL;
+	if (kva == NULL) {
+		DRM_ERROR("No KVA Found\n");
+		return -EAGAIN;
+	}
 
-	/*
-	 * get page frame number to physical memory to be mapped
-	 * to user space.
-	 */
-	pfn = ((unsigned long)exynos_gem_obj->buffer->dma_addr) >> PAGE_SHIFT;
+	dma_address = exynos_gem_obj->buffer->dma_addr;
+	vma->vm_flags           |= VM_DONTEXPAND | VM_RESERVED;
+	vma->vm_pgoff = 0;
 
-	DRM_DEBUG_KMS("pfn = 0x%lx\n", pfn);
+	ret = dma_mmap_writecombine(obj->dev->dev, vma, kva,
+				dma_address, vma->vm_end - vma->vm_start);
 
-	if (remap_pfn_range(vma, vma->vm_start, pfn, vm_size,
-				vma->vm_page_prot)) {
-		DRM_ERROR("failed to remap pfn range.\n");
-		return -EAGAIN;
+	if (ret) {
+		DRM_ERROR("Remapping memory failed, error: %ld\n", ret);
+		return ret;
 	}
 
 	return 0;
@@ -433,19 +420,29 @@ int exynos_drm_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 	struct drm_gem_object *obj = vma->vm_private_data;
 	struct exynos_drm_gem_obj *exynos_gem_obj = to_exynos_gem_obj(obj);
 	struct drm_device *dev = obj->dev;
-	unsigned long pfn;
-	pgoff_t page_offset;
+	void *kva;
+	dma_addr_t dma_address;
 	int ret;
 
-	page_offset = ((unsigned long)vmf->virtual_address -
-			vma->vm_start) >> PAGE_SHIFT;
+	kva = exynos_gem_obj->buffer->kvaddr;
+
+	if (kva == NULL) {
+		DRM_ERROR("No KVA Found\n");
+		return -EAGAIN;
+	}
 
 	mutex_lock(&dev->struct_mutex);
 
-	pfn = (((unsigned long)exynos_gem_obj->buffer->dma_addr) >>
-			PAGE_SHIFT) + page_offset;
 
-	ret = vm_insert_mixed(vma, (unsigned long)vmf->virtual_address, pfn);
+	dma_address = exynos_gem_obj->buffer->dma_addr;
+	vma->vm_flags           |= VM_DONTEXPAND | VM_RESERVED;
+	vma->vm_pgoff = 0;
+
+	ret = dma_mmap_writecombine(obj->dev->dev, vma, kva,
+				dma_address, vma->vm_end - vma->vm_start);
+
+	if (ret)
+		DRM_ERROR("Remapping memory failed, error: %d\n", ret);
 
 	mutex_unlock(&dev->struct_mutex);
 
@@ -457,7 +454,6 @@ int exynos_drm_gem_mmap(struct file *filp, struct vm_area_struct *vma)
 	int ret;
 
 	DRM_DEBUG_KMS("%s\n", __FILE__);
-
 	/* set vm_area_struct. */
 	ret = drm_gem_mmap(filp, vma);
 	if (ret < 0) {
-- 
1.7.0.4

RE: [PATCH 2/4] [RFC] drm/exynos: Mapping of gem objects uses dma_mmap_writecombine

[Inki Dae]

Hi,

This is also patch set based on old version and the feature below has
already been merged to mainline. Prathyush, please DO WORK patch set based
on latest drm-next for new feature and drm-fixes for bug fixes from next
time and also include me as Maintainer.

Thanks,
Inki Dae.

> -----Original Message-----
> From: Prathyush [mailto:prathyush.k@samsung.com]
> Sent: Saturday, April 14, 2012 8:52 PM
> To: dri-devel@lists.freedesktop.org; linaro-mm-sig@lists.linaro.org
> Cc: inki.dae@samsung.com; subash.rp@samsung.com; prashanth.g@samsung.com;
> sunilm@samsung.com; prathyush.k@samsung.com
> Subject: [PATCH 2/4] [RFC] drm/exynos: Mapping of gem objects uses
> dma_mmap_writecombine
> 
> GEM objects get mapped to user space in two ways - DIRECT and
> INDIRECT mapping. DIRECT mapping is by calling an ioctl and it
> maps all the pages to user space by calling remap-pfn-range.
> Indirect mapping is done by calling 'mmap'. The actual mapping
> is done when a page fault gets generated and is handled by
> exynos_drm_gem_fault function where the required page is mapped.
> Both the methods assume contiguous memory.
> 
> With this change, the mapping is done by dma_mmap_writecombine
> which will support mapping of non-contiguous memory to user space.
> 
> This works similar to the previous approach for the case of
> DIRECT mapping. But in the case of mapping when a page fault
> occurs, dma_mmap_writecombine will map all the pages and not just
> one page.
> 
> Signed-off-by: Prathyush K <prathyush.k@samsung.com>
> ---
>  drivers/gpu/drm/exynos/exynos_drm_gem.c |   70
++++++++++++++-------------
> ---
>  1 files changed, 33 insertions(+), 37 deletions(-)
> 
> diff --git a/drivers/gpu/drm/exynos/exynos_drm_gem.c
> b/drivers/gpu/drm/exynos/exynos_drm_gem.c
> index 807143e..a57a83a 100755
> --- a/drivers/gpu/drm/exynos/exynos_drm_gem.c
> +++ b/drivers/gpu/drm/exynos/exynos_drm_gem.c
> @@ -200,40 +200,27 @@ static int exynos_drm_gem_mmap_buffer(struct file
> *filp,
>  {
>  	struct drm_gem_object *obj = filp->private_data;
>  	struct exynos_drm_gem_obj *exynos_gem_obj = to_exynos_gem_obj(obj);
> -	struct exynos_drm_gem_buf *buffer;
> -	unsigned long pfn, vm_size;
> -
> -	DRM_DEBUG_KMS("%s\n", __FILE__);
> +	void *kva;
> +	dma_addr_t dma_address;
> +	unsigned long ret;
> 
> -	vma->vm_flags |= (VM_IO | VM_RESERVED);
> +	kva = exynos_gem_obj->buffer->kvaddr;
> 
> -	/* in case of direct mapping, always having non-cachable attribute
> */
> -	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
> -	vma->vm_file = filp;
> -
> -	vm_size = vma->vm_end - vma->vm_start;
> -	/*
> -	 * a buffer contains information to physically continuous memory
> -	 * allocated by user request or at framebuffer creation.
> -	 */
> -	buffer = exynos_gem_obj->buffer;
> -
> -	/* check if user-requested size is valid. */
> -	if (vm_size > buffer->size)
> -		return -EINVAL;
> +	if (kva == NULL) {
> +		DRM_ERROR("No KVA Found\n");
> +		return -EAGAIN;
> +	}
> 
> -	/*
> -	 * get page frame number to physical memory to be mapped
> -	 * to user space.
> -	 */
> -	pfn = ((unsigned long)exynos_gem_obj->buffer->dma_addr) >>
> PAGE_SHIFT;
> +	dma_address = exynos_gem_obj->buffer->dma_addr;
> +	vma->vm_flags           |= VM_DONTEXPAND | VM_RESERVED;
> +	vma->vm_pgoff = 0;
> 
> -	DRM_DEBUG_KMS("pfn = 0x%lx\n", pfn);
> +	ret = dma_mmap_writecombine(obj->dev->dev, vma, kva,
> +				dma_address, vma->vm_end - vma->vm_start);
> 
> -	if (remap_pfn_range(vma, vma->vm_start, pfn, vm_size,
> -				vma->vm_page_prot)) {
> -		DRM_ERROR("failed to remap pfn range.\n");
> -		return -EAGAIN;
> +	if (ret) {
> +		DRM_ERROR("Remapping memory failed, error: %ld\n", ret);
> +		return ret;
>  	}
> 
>  	return 0;
> @@ -433,19 +420,29 @@ int exynos_drm_gem_fault(struct vm_area_struct *vma,
> struct vm_fault *vmf)
>  	struct drm_gem_object *obj = vma->vm_private_data;
>  	struct exynos_drm_gem_obj *exynos_gem_obj = to_exynos_gem_obj(obj);
>  	struct drm_device *dev = obj->dev;
> -	unsigned long pfn;
> -	pgoff_t page_offset;
> +	void *kva;
> +	dma_addr_t dma_address;
>  	int ret;
> 
> -	page_offset = ((unsigned long)vmf->virtual_address -
> -			vma->vm_start) >> PAGE_SHIFT;
> +	kva = exynos_gem_obj->buffer->kvaddr;
> +
> +	if (kva == NULL) {
> +		DRM_ERROR("No KVA Found\n");
> +		return -EAGAIN;
> +	}
> 
>  	mutex_lock(&dev->struct_mutex);
> 
> -	pfn = (((unsigned long)exynos_gem_obj->buffer->dma_addr) >>
> -			PAGE_SHIFT) + page_offset;
> 
> -	ret = vm_insert_mixed(vma, (unsigned long)vmf->virtual_address,
> pfn);
> +	dma_address = exynos_gem_obj->buffer->dma_addr;
> +	vma->vm_flags           |= VM_DONTEXPAND | VM_RESERVED;
> +	vma->vm_pgoff = 0;
> +
> +	ret = dma_mmap_writecombine(obj->dev->dev, vma, kva,
> +				dma_address, vma->vm_end - vma->vm_start);
> +
> +	if (ret)
> +		DRM_ERROR("Remapping memory failed, error: %d\n", ret);
> 
>  	mutex_unlock(&dev->struct_mutex);
> 
> @@ -457,7 +454,6 @@ int exynos_drm_gem_mmap(struct file *filp, struct
> vm_area_struct *vma)
>  	int ret;
> 
>  	DRM_DEBUG_KMS("%s\n", __FILE__);
> -
>  	/* set vm_area_struct. */
>  	ret = drm_gem_mmap(filp, vma);
>  	if (ret < 0) {
> --
> 1.7.0.4

[PATCH 3/4] [RFC] drm/Exynos: Added 'disable' function to Exynos drm crtc module.

[Prathyush]

While freeing an fb, if the fb is being used by a crtc, the drm
driver tries to disable the CRTC. Currently there is no disable
function provided by exynos drm crtc module so the driver tries
to suspend the crtc (by calling dpms) which only works if
RUNTIME PM is enabled.

Signed-off-by: Prathyush K <prathyush.k@samsung.com>
---
 drivers/gpu/drm/exynos/exynos_drm_crtc.c |   11 +++++++++++
 1 files changed, 11 insertions(+), 0 deletions(-)

diff --git a/drivers/gpu/drm/exynos/exynos_drm_crtc.c b/drivers/gpu/drm/exynos/exynos_drm_crtc.c
index e3861ac..eb1e553 100644
--- a/drivers/gpu/drm/exynos/exynos_drm_crtc.c
+++ b/drivers/gpu/drm/exynos/exynos_drm_crtc.c
@@ -192,6 +192,16 @@ static void exynos_drm_crtc_dpms(struct drm_crtc *crtc, int mode)
 	mutex_unlock(&dev->struct_mutex);
 }
 
+static void exynos_drm_crtc_disable(struct drm_crtc *crtc)
+{
+	struct exynos_drm_crtc *exynos_crtc = to_exynos_crtc(crtc);
+	struct exynos_drm_overlay *overlay = &exynos_crtc->overlay;
+	int win = overlay->zpos;
+
+	exynos_drm_fn_encoder(crtc, &win,
+		exynos_drm_encoder_crtc_disable);
+}
+
 static void exynos_drm_crtc_prepare(struct drm_crtc *crtc)
 {
 	DRM_DEBUG_KMS("%s\n", __FILE__);
@@ -278,6 +288,7 @@ static void exynos_drm_crtc_load_lut(struct drm_crtc *crtc)
 
 static struct drm_crtc_helper_funcs exynos_crtc_helper_funcs = {
 	.dpms		= exynos_drm_crtc_dpms,
+	.disable	= exynos_drm_crtc_disable,
 	.prepare	= exynos_drm_crtc_prepare,
 	.commit		= exynos_drm_crtc_commit,
 	.mode_fixup	= exynos_drm_crtc_mode_fixup,
-- 
1.7.0.4

RE: [PATCH 3/4] [RFC] drm/Exynos: Added 'disable' function to Exynos drm crtc module.

[Inki Dae]

> -----Original Message-----
> From: Prathyush [mailto:prathyush.k@samsung.com]
> Sent: Saturday, April 14, 2012 8:52 PM
> To: dri-devel@lists.freedesktop.org; linaro-mm-sig@lists.linaro.org
> Cc: inki.dae@samsung.com; subash.rp@samsung.com; prashanth.g@samsung.com;
> sunilm@samsung.com; prathyush.k@samsung.com
> Subject: [PATCH 3/4] [RFC] drm/Exynos: Added 'disable' function to Exynos
> drm crtc module.
> 
> While freeing an fb, if the fb is being used by a crtc, the drm
> driver tries to disable the CRTC. Currently there is no disable
> function provided by exynos drm crtc module so the driver tries
> to suspend the crtc (by calling dpms) which only works if
> RUNTIME PM is enabled.
> 
> Signed-off-by: Prathyush K <prathyush.k@samsung.com>
> ---
>  drivers/gpu/drm/exynos/exynos_drm_crtc.c |   11 +++++++++++
>  1 files changed, 11 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/gpu/drm/exynos/exynos_drm_crtc.c
> b/drivers/gpu/drm/exynos/exynos_drm_crtc.c
> index e3861ac..eb1e553 100644
> --- a/drivers/gpu/drm/exynos/exynos_drm_crtc.c
> +++ b/drivers/gpu/drm/exynos/exynos_drm_crtc.c
> @@ -192,6 +192,16 @@ static void exynos_drm_crtc_dpms(struct drm_crtc
> *crtc, int mode)
>  	mutex_unlock(&dev->struct_mutex);
>  }
> 
> +static void exynos_drm_crtc_disable(struct drm_crtc *crtc)
> +{
> +	struct exynos_drm_crtc *exynos_crtc = to_exynos_crtc(crtc);
> +	struct exynos_drm_overlay *overlay = &exynos_crtc->overlay;
> +	int win = overlay->zpos;
> +
> +	exynos_drm_fn_encoder(crtc, &win,
> +		exynos_drm_encoder_crtc_disable);
> +}
> +
>  static void exynos_drm_crtc_prepare(struct drm_crtc *crtc)
>  {
>  	DRM_DEBUG_KMS("%s\n", __FILE__);
> @@ -278,6 +288,7 @@ static void exynos_drm_crtc_load_lut(struct drm_crtc
> *crtc)
> 
>  static struct drm_crtc_helper_funcs exynos_crtc_helper_funcs = {
>  	.dpms		= exynos_drm_crtc_dpms,
> +	.disable	= exynos_drm_crtc_disable,
>  	.prepare	= exynos_drm_crtc_prepare,
>  	.commit		= exynos_drm_crtc_commit,
>  	.mode_fixup	= exynos_drm_crtc_mode_fixup,
> --

dpms callback DO WORK same thing please see drm_crtc.c file, dpms callback
would be called if disable is NULL. your patch set makes our dpms callback
not to be used. and exynos_drm_encoder_crtc_disable is called by
exynos_drm_plane module for each overlay.

Thanks.

> 1.7.0.4

[PATCH 4/4] [RFC] drm: Releasing FBs before releasing GEM objects during drm_release

[Prathyush]

During DRM release, all the FBs and gem objects are released. If
a gem object is being used as a FB and set to a crtc, it must not
be freed before releasing the framebuffer first.

If FBs are released first, the crtc using the FB is disabled first
so now the GEM object can be freed safely. The CRTC will be enabled
again when the driver restores fbdev mode.

Signed-off-by: Prathyush K <prathyush.k@samsung.com>
---
 drivers/gpu/drm/drm_fops.c |    6 +++---
 1 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/drm_fops.c b/drivers/gpu/drm/drm_fops.c
index 6263b01..f6525ef 100644
--- a/drivers/gpu/drm/drm_fops.c
+++ b/drivers/gpu/drm/drm_fops.c
@@ -498,12 +498,12 @@ int drm_release(struct inode *inode, struct file *filp)
 
 	drm_events_release(file_priv);
 
-	if (dev->driver->driver_features & DRIVER_GEM)
-		drm_gem_release(dev, file_priv);
-
 	if (dev->driver->driver_features & DRIVER_MODESET)
 		drm_fb_release(file_priv);
 
+	if (dev->driver->driver_features & DRIVER_GEM)
+		drm_gem_release(dev, file_priv);
+
 	mutex_lock(&dev->ctxlist_mutex);
 	if (!list_empty(&dev->ctxlist)) {
 		struct drm_ctx_list *pos, *n;
-- 
1.7.0.4