ARM: dma-mapping: add support for IOMMU mapper
[pandora-kernel.git] / arch / arm / mm / dma-mapping.c
index a7ca498..ecdb9cf 100644 (file)
@@ -19,6 +19,8 @@
 #include <linux/dma-mapping.h>
 #include <linux/highmem.h>
 #include <linux/slab.h>
+#include <linux/iommu.h>
+#include <linux/vmalloc.h>
 
 #include <asm/memory.h>
 #include <asm/highmem.h>
@@ -26,6 +28,7 @@
 #include <asm/tlbflush.h>
 #include <asm/sizes.h>
 #include <asm/mach/arch.h>
+#include <asm/dma-iommu.h>
 
 #include "mm.h"
 
@@ -155,6 +158,21 @@ static u64 get_coherent_dma_mask(struct device *dev)
        return mask;
 }
 
+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);
+       }
+}
+
 /*
  * Allocate a DMA buffer for 'dev' of size 'size' using the
  * specified gfp mask.  Note that 'size' must be page aligned.
@@ -163,7 +181,6 @@ static struct page *__dma_alloc_buffer(struct device *dev, size_t size, gfp_t gf
 {
        unsigned long order = get_order(size);
        struct page *page, *p, *e;
-       void *ptr;
        u64 mask = get_coherent_dma_mask(dev);
 
 #ifdef CONFIG_DMA_API_DEBUG
@@ -192,14 +209,7 @@ static struct page *__dma_alloc_buffer(struct device *dev, size_t size, gfp_t gf
        for (p = page + (size >> PAGE_SHIFT), e = page + (1 << order); p < e; p++)
                __free_page(p);
 
-       /*
-        * Ensure that the allocated pages are zeroed, and that any data
-        * lurking in the kernel direct-mapped region is invalidated.
-        */
-       ptr = page_address(page);
-       memset(ptr, 0, size);
-       dmac_flush_range(ptr, ptr + size);
-       outer_flush_range(__pa(ptr), __pa(ptr) + size);
+       __dma_clear_buffer(page, size);
 
        return page;
 }
@@ -348,7 +358,7 @@ __dma_alloc_remap(struct page *page, size_t size, gfp_t gfp, pgprot_t prot,
                u32 off = CONSISTENT_OFFSET(c->vm_start) & (PTRS_PER_PTE-1);
 
                pte = consistent_pte[idx] + off;
-               c->vm_pages = page;
+               c->priv = page;
 
                do {
                        BUG_ON(!pte_none(*pte));
@@ -509,13 +519,14 @@ int arm_dma_mmap(struct device *dev, struct vm_area_struct *vma,
        c = arm_vmregion_find(&consistent_head, (unsigned long)cpu_addr);
        if (c) {
                unsigned long off = vma->vm_pgoff;
+               struct page *pages = c->priv;
 
                kern_size = (c->vm_end - c->vm_start) >> PAGE_SHIFT;
 
                if (off < kern_size &&
                    user_size <= (kern_size - off)) {
                        ret = remap_pfn_range(vma, vma->vm_start,
-                                             page_to_pfn(c->vm_pages) + off,
+                                             page_to_pfn(pages) + off,
                                              user_size << PAGE_SHIFT,
                                              vma->vm_page_prot);
                }
@@ -656,6 +667,9 @@ int arm_dma_map_sg(struct device *dev, struct scatterlist *sg, int nents,
        int i, j;
 
        for_each_sg(sg, s, nents, i) {
+#ifdef CONFIG_NEED_SG_DMA_LENGTH
+               s->dma_length = s->length;
+#endif
                s->dma_address = ops->map_page(dev, sg_page(s), s->offset,
                                                s->length, dir, attrs);
                if (dma_mapping_error(dev, s->dma_address))
@@ -764,3 +778,679 @@ static int __init dma_debug_do_init(void)
        return 0;
 }
 fs_initcall(dma_debug_do_init);
+
+#ifdef CONFIG_ARM_DMA_USE_IOMMU
+
+/* IOMMU */
+
+static inline dma_addr_t __alloc_iova(struct dma_iommu_mapping *mapping,
+                                     size_t size)
+{
+       unsigned int order = get_order(size);
+       unsigned int align = 0;
+       unsigned int count, start;
+       unsigned long flags;
+
+       count = ((PAGE_ALIGN(size) >> PAGE_SHIFT) +
+                (1 << mapping->order) - 1) >> mapping->order;
+
+       if (order > mapping->order)
+               align = (1 << (order - mapping->order)) - 1;
+
+       spin_lock_irqsave(&mapping->lock, flags);
+       start = bitmap_find_next_zero_area(mapping->bitmap, mapping->bits, 0,
+                                          count, align);
+       if (start > mapping->bits) {
+               spin_unlock_irqrestore(&mapping->lock, flags);
+               return DMA_ERROR_CODE;
+       }
+
+       bitmap_set(mapping->bitmap, start, count);
+       spin_unlock_irqrestore(&mapping->lock, flags);
+
+       return mapping->base + (start << (mapping->order + PAGE_SHIFT));
+}
+
+static inline void __free_iova(struct dma_iommu_mapping *mapping,
+                              dma_addr_t addr, size_t size)
+{
+       unsigned int start = (addr - mapping->base) >>
+                            (mapping->order + PAGE_SHIFT);
+       unsigned int count = ((size >> PAGE_SHIFT) +
+                             (1 << mapping->order) - 1) >> mapping->order;
+       unsigned long flags;
+
+       spin_lock_irqsave(&mapping->lock, flags);
+       bitmap_clear(mapping->bitmap, start, count);
+       spin_unlock_irqrestore(&mapping->lock, flags);
+}
+
+static struct page **__iommu_alloc_buffer(struct device *dev, size_t size, gfp_t gfp)
+{
+       struct page **pages;
+       int count = size >> PAGE_SHIFT;
+       int array_size = count * sizeof(struct page *);
+       int i = 0;
+
+       if (array_size <= PAGE_SIZE)
+               pages = kzalloc(array_size, gfp);
+       else
+               pages = vzalloc(array_size);
+       if (!pages)
+               return NULL;
+
+       while (count) {
+               int j, order = __ffs(count);
+
+               pages[i] = alloc_pages(gfp | __GFP_NOWARN, order);
+               while (!pages[i] && order)
+                       pages[i] = alloc_pages(gfp | __GFP_NOWARN, --order);
+               if (!pages[i])
+                       goto error;
+
+               if (order)
+                       split_page(pages[i], order);
+               j = 1 << order;
+               while (--j)
+                       pages[i + j] = pages[i] + j;
+
+               __dma_clear_buffer(pages[i], PAGE_SIZE << order);
+               i += 1 << order;
+               count -= 1 << order;
+       }
+
+       return pages;
+error:
+       while (--i)
+               if (pages[i])
+                       __free_pages(pages[i], 0);
+       if (array_size < PAGE_SIZE)
+               kfree(pages);
+       else
+               vfree(pages);
+       return NULL;
+}
+
+static int __iommu_free_buffer(struct device *dev, struct page **pages, size_t size)
+{
+       int count = size >> PAGE_SHIFT;
+       int array_size = count * sizeof(struct page *);
+       int i;
+       for (i = 0; i < count; i++)
+               if (pages[i])
+                       __free_pages(pages[i], 0);
+       if (array_size < PAGE_SIZE)
+               kfree(pages);
+       else
+               vfree(pages);
+       return 0;
+}
+
+/*
+ * Create a CPU mapping for a specified pages
+ */
+static void *
+__iommu_alloc_remap(struct page **pages, size_t size, gfp_t gfp, pgprot_t prot)
+{
+       struct arm_vmregion *c;
+       size_t align;
+       size_t count = size >> PAGE_SHIFT;
+       int bit;
+
+       if (!consistent_pte[0]) {
+               pr_err("%s: not initialised\n", __func__);
+               dump_stack();
+               return NULL;
+       }
+
+       /*
+        * Align the virtual region allocation - maximum alignment is
+        * a section size, minimum is a page size.  This helps reduce
+        * fragmentation of the DMA space, and also prevents allocations
+        * smaller than a section from crossing a section boundary.
+        */
+       bit = fls(size - 1);
+       if (bit > SECTION_SHIFT)
+               bit = SECTION_SHIFT;
+       align = 1 << bit;
+
+       /*
+        * Allocate a virtual address in the consistent mapping region.
+        */
+       c = arm_vmregion_alloc(&consistent_head, align, size,
+                           gfp & ~(__GFP_DMA | __GFP_HIGHMEM), NULL);
+       if (c) {
+               pte_t *pte;
+               int idx = CONSISTENT_PTE_INDEX(c->vm_start);
+               int i = 0;
+               u32 off = CONSISTENT_OFFSET(c->vm_start) & (PTRS_PER_PTE-1);
+
+               pte = consistent_pte[idx] + off;
+               c->priv = pages;
+
+               do {
+                       BUG_ON(!pte_none(*pte));
+
+                       set_pte_ext(pte, mk_pte(pages[i], prot), 0);
+                       pte++;
+                       off++;
+                       i++;
+                       if (off >= PTRS_PER_PTE) {
+                               off = 0;
+                               pte = consistent_pte[++idx];
+                       }
+               } while (i < count);
+
+               dsb();
+
+               return (void *)c->vm_start;
+       }
+       return NULL;
+}
+
+/*
+ * Create a mapping in device IO address space for specified pages
+ */
+static dma_addr_t
+__iommu_create_mapping(struct device *dev, struct page **pages, size_t size)
+{
+       struct dma_iommu_mapping *mapping = dev->archdata.mapping;
+       unsigned int count = PAGE_ALIGN(size) >> PAGE_SHIFT;
+       dma_addr_t dma_addr, iova;
+       int i, ret = DMA_ERROR_CODE;
+
+       dma_addr = __alloc_iova(mapping, size);
+       if (dma_addr == DMA_ERROR_CODE)
+               return dma_addr;
+
+       iova = dma_addr;
+       for (i = 0; i < count; ) {
+               unsigned int next_pfn = page_to_pfn(pages[i]) + 1;
+               phys_addr_t phys = page_to_phys(pages[i]);
+               unsigned int len, j;
+
+               for (j = i + 1; j < count; j++, next_pfn++)
+                       if (page_to_pfn(pages[j]) != next_pfn)
+                               break;
+
+               len = (j - i) << PAGE_SHIFT;
+               ret = iommu_map(mapping->domain, iova, phys, len, 0);
+               if (ret < 0)
+                       goto fail;
+               iova += len;
+               i = j;
+       }
+       return dma_addr;
+fail:
+       iommu_unmap(mapping->domain, dma_addr, iova-dma_addr);
+       __free_iova(mapping, dma_addr, size);
+       return DMA_ERROR_CODE;
+}
+
+static int __iommu_remove_mapping(struct device *dev, dma_addr_t iova, size_t size)
+{
+       struct dma_iommu_mapping *mapping = dev->archdata.mapping;
+
+       /*
+        * add optional in-page offset from iova to size and align
+        * result to page size
+        */
+       size = PAGE_ALIGN((iova & ~PAGE_MASK) + size);
+       iova &= PAGE_MASK;
+
+       iommu_unmap(mapping->domain, iova, size);
+       __free_iova(mapping, iova, size);
+       return 0;
+}
+
+static void *arm_iommu_alloc_attrs(struct device *dev, size_t size,
+           dma_addr_t *handle, gfp_t gfp, struct dma_attrs *attrs)
+{
+       pgprot_t prot = __get_dma_pgprot(attrs, pgprot_kernel);
+       struct page **pages;
+       void *addr = NULL;
+
+       *handle = DMA_ERROR_CODE;
+       size = PAGE_ALIGN(size);
+
+       pages = __iommu_alloc_buffer(dev, size, gfp);
+       if (!pages)
+               return NULL;
+
+       *handle = __iommu_create_mapping(dev, pages, size);
+       if (*handle == DMA_ERROR_CODE)
+               goto err_buffer;
+
+       addr = __iommu_alloc_remap(pages, size, gfp, prot);
+       if (!addr)
+               goto err_mapping;
+
+       return addr;
+
+err_mapping:
+       __iommu_remove_mapping(dev, *handle, size);
+err_buffer:
+       __iommu_free_buffer(dev, pages, size);
+       return NULL;
+}
+
+static int arm_iommu_mmap_attrs(struct device *dev, struct vm_area_struct *vma,
+                   void *cpu_addr, dma_addr_t dma_addr, size_t size,
+                   struct dma_attrs *attrs)
+{
+       struct arm_vmregion *c;
+
+       vma->vm_page_prot = __get_dma_pgprot(attrs, vma->vm_page_prot);
+       c = arm_vmregion_find(&consistent_head, (unsigned long)cpu_addr);
+
+       if (c) {
+               struct page **pages = c->priv;
+
+               unsigned long uaddr = vma->vm_start;
+               unsigned long usize = vma->vm_end - vma->vm_start;
+               int i = 0;
+
+               do {
+                       int ret;
+
+                       ret = vm_insert_page(vma, uaddr, pages[i++]);
+                       if (ret) {
+                               pr_err("Remapping memory, error: %d\n", ret);
+                               return ret;
+                       }
+
+                       uaddr += PAGE_SIZE;
+                       usize -= PAGE_SIZE;
+               } while (usize > 0);
+       }
+       return 0;
+}
+
+/*
+ * free a page as defined by the above mapping.
+ * Must not be called with IRQs disabled.
+ */
+void arm_iommu_free_attrs(struct device *dev, size_t size, void *cpu_addr,
+                         dma_addr_t handle, struct dma_attrs *attrs)
+{
+       struct arm_vmregion *c;
+       size = PAGE_ALIGN(size);
+
+       c = arm_vmregion_find(&consistent_head, (unsigned long)cpu_addr);
+       if (c) {
+               struct page **pages = c->priv;
+               __dma_free_remap(cpu_addr, size);
+               __iommu_remove_mapping(dev, handle, size);
+               __iommu_free_buffer(dev, pages, size);
+       }
+}
+
+/*
+ * Map a part of the scatter-gather list into contiguous io address space
+ */
+static int __map_sg_chunk(struct device *dev, struct scatterlist *sg,
+                         size_t size, dma_addr_t *handle,
+                         enum dma_data_direction dir)
+{
+       struct dma_iommu_mapping *mapping = dev->archdata.mapping;
+       dma_addr_t iova, iova_base;
+       int ret = 0;
+       unsigned int count;
+       struct scatterlist *s;
+
+       size = PAGE_ALIGN(size);
+       *handle = DMA_ERROR_CODE;
+
+       iova_base = iova = __alloc_iova(mapping, size);
+       if (iova == DMA_ERROR_CODE)
+               return -ENOMEM;
+
+       for (count = 0, s = sg; count < (size >> PAGE_SHIFT); s = sg_next(s)) {
+               phys_addr_t phys = page_to_phys(sg_page(s));
+               unsigned int len = PAGE_ALIGN(s->offset + s->length);
+
+               if (!arch_is_coherent())
+                       __dma_page_cpu_to_dev(sg_page(s), s->offset, s->length, dir);
+
+               ret = iommu_map(mapping->domain, iova, phys, len, 0);
+               if (ret < 0)
+                       goto fail;
+               count += len >> PAGE_SHIFT;
+               iova += len;
+       }
+       *handle = iova_base;
+
+       return 0;
+fail:
+       iommu_unmap(mapping->domain, iova_base, count * PAGE_SIZE);
+       __free_iova(mapping, iova_base, size);
+       return ret;
+}
+
+/**
+ * arm_iommu_map_sg - map a set of SG buffers for streaming mode DMA
+ * @dev: valid struct device pointer
+ * @sg: list of buffers
+ * @nents: number of buffers to map
+ * @dir: DMA transfer direction
+ *
+ * Map a set of buffers described by scatterlist in streaming mode for DMA.
+ * The scatter gather list elements are merged together (if possible) and
+ * tagged with the appropriate dma address and length. They are obtained via
+ * sg_dma_{address,length}.
+ */
+int arm_iommu_map_sg(struct device *dev, struct scatterlist *sg, int nents,
+                    enum dma_data_direction dir, struct dma_attrs *attrs)
+{
+       struct scatterlist *s = sg, *dma = sg, *start = sg;
+       int i, count = 0;
+       unsigned int offset = s->offset;
+       unsigned int size = s->offset + s->length;
+       unsigned int max = dma_get_max_seg_size(dev);
+
+       for (i = 1; i < nents; i++) {
+               s = sg_next(s);
+
+               s->dma_address = DMA_ERROR_CODE;
+               s->dma_length = 0;
+
+               if (s->offset || (size & ~PAGE_MASK) || size + s->length > max) {
+                       if (__map_sg_chunk(dev, start, size, &dma->dma_address,
+                           dir) < 0)
+                               goto bad_mapping;
+
+                       dma->dma_address += offset;
+                       dma->dma_length = size - offset;
+
+                       size = offset = s->offset;
+                       start = s;
+                       dma = sg_next(dma);
+                       count += 1;
+               }
+               size += s->length;
+       }
+       if (__map_sg_chunk(dev, start, size, &dma->dma_address, dir) < 0)
+               goto bad_mapping;
+
+       dma->dma_address += offset;
+       dma->dma_length = size - offset;
+
+       return count+1;
+
+bad_mapping:
+       for_each_sg(sg, s, count, i)
+               __iommu_remove_mapping(dev, sg_dma_address(s), sg_dma_len(s));
+       return 0;
+}
+
+/**
+ * arm_iommu_unmap_sg - unmap a set of SG buffers mapped by dma_map_sg
+ * @dev: valid struct device pointer
+ * @sg: list of buffers
+ * @nents: number of buffers to unmap (same as was passed to dma_map_sg)
+ * @dir: DMA transfer direction (same as was passed to dma_map_sg)
+ *
+ * Unmap a set of streaming mode DMA translations.  Again, CPU access
+ * rules concerning calls here are the same as for dma_unmap_single().
+ */
+void arm_iommu_unmap_sg(struct device *dev, struct scatterlist *sg, int nents,
+                       enum dma_data_direction dir, struct dma_attrs *attrs)
+{
+       struct scatterlist *s;
+       int i;
+
+       for_each_sg(sg, s, nents, i) {
+               if (sg_dma_len(s))
+                       __iommu_remove_mapping(dev, sg_dma_address(s),
+                                              sg_dma_len(s));
+               if (!arch_is_coherent())
+                       __dma_page_dev_to_cpu(sg_page(s), s->offset,
+                                             s->length, dir);
+       }
+}
+
+/**
+ * arm_iommu_sync_sg_for_cpu
+ * @dev: valid struct device pointer
+ * @sg: list of buffers
+ * @nents: number of buffers to map (returned from dma_map_sg)
+ * @dir: DMA transfer direction (same as was passed to dma_map_sg)
+ */
+void arm_iommu_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg,
+                       int nents, enum dma_data_direction dir)
+{
+       struct scatterlist *s;
+       int i;
+
+       for_each_sg(sg, s, nents, i)
+               if (!arch_is_coherent())
+                       __dma_page_dev_to_cpu(sg_page(s), s->offset, s->length, dir);
+
+}
+
+/**
+ * arm_iommu_sync_sg_for_device
+ * @dev: valid struct device pointer
+ * @sg: list of buffers
+ * @nents: number of buffers to map (returned from dma_map_sg)
+ * @dir: DMA transfer direction (same as was passed to dma_map_sg)
+ */
+void arm_iommu_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
+                       int nents, enum dma_data_direction dir)
+{
+       struct scatterlist *s;
+       int i;
+
+       for_each_sg(sg, s, nents, i)
+               if (!arch_is_coherent())
+                       __dma_page_cpu_to_dev(sg_page(s), s->offset, s->length, dir);
+}
+
+
+/**
+ * arm_iommu_map_page
+ * @dev: valid struct device pointer
+ * @page: page that buffer resides in
+ * @offset: offset into page for start of buffer
+ * @size: size of buffer to map
+ * @dir: DMA transfer direction
+ *
+ * IOMMU aware version of arm_dma_map_page()
+ */
+static dma_addr_t arm_iommu_map_page(struct device *dev, struct page *page,
+            unsigned long offset, size_t size, enum dma_data_direction dir,
+            struct dma_attrs *attrs)
+{
+       struct dma_iommu_mapping *mapping = dev->archdata.mapping;
+       dma_addr_t dma_addr;
+       int ret, len = PAGE_ALIGN(size + offset);
+
+       if (!arch_is_coherent())
+               __dma_page_cpu_to_dev(page, offset, size, dir);
+
+       dma_addr = __alloc_iova(mapping, len);
+       if (dma_addr == DMA_ERROR_CODE)
+               return dma_addr;
+
+       ret = iommu_map(mapping->domain, dma_addr, page_to_phys(page), len, 0);
+       if (ret < 0)
+               goto fail;
+
+       return dma_addr + offset;
+fail:
+       __free_iova(mapping, dma_addr, len);
+       return DMA_ERROR_CODE;
+}
+
+/**
+ * arm_iommu_unmap_page
+ * @dev: valid struct device pointer
+ * @handle: DMA address of buffer
+ * @size: size of buffer (same as passed to dma_map_page)
+ * @dir: DMA transfer direction (same as passed to dma_map_page)
+ *
+ * IOMMU aware version of arm_dma_unmap_page()
+ */
+static void arm_iommu_unmap_page(struct device *dev, dma_addr_t handle,
+               size_t size, enum dma_data_direction dir,
+               struct dma_attrs *attrs)
+{
+       struct dma_iommu_mapping *mapping = dev->archdata.mapping;
+       dma_addr_t iova = handle & PAGE_MASK;
+       struct page *page = phys_to_page(iommu_iova_to_phys(mapping->domain, iova));
+       int offset = handle & ~PAGE_MASK;
+       int len = PAGE_ALIGN(size + offset);
+
+       if (!iova)
+               return;
+
+       if (!arch_is_coherent())
+               __dma_page_dev_to_cpu(page, offset, size, dir);
+
+       iommu_unmap(mapping->domain, iova, len);
+       __free_iova(mapping, iova, len);
+}
+
+static void arm_iommu_sync_single_for_cpu(struct device *dev,
+               dma_addr_t handle, size_t size, enum dma_data_direction dir)
+{
+       struct dma_iommu_mapping *mapping = dev->archdata.mapping;
+       dma_addr_t iova = handle & PAGE_MASK;
+       struct page *page = phys_to_page(iommu_iova_to_phys(mapping->domain, iova));
+       unsigned int offset = handle & ~PAGE_MASK;
+
+       if (!iova)
+               return;
+
+       if (!arch_is_coherent())
+               __dma_page_dev_to_cpu(page, offset, size, dir);
+}
+
+static void arm_iommu_sync_single_for_device(struct device *dev,
+               dma_addr_t handle, size_t size, enum dma_data_direction dir)
+{
+       struct dma_iommu_mapping *mapping = dev->archdata.mapping;
+       dma_addr_t iova = handle & PAGE_MASK;
+       struct page *page = phys_to_page(iommu_iova_to_phys(mapping->domain, iova));
+       unsigned int offset = handle & ~PAGE_MASK;
+
+       if (!iova)
+               return;
+
+       __dma_page_cpu_to_dev(page, offset, size, dir);
+}
+
+struct dma_map_ops iommu_ops = {
+       .alloc          = arm_iommu_alloc_attrs,
+       .free           = arm_iommu_free_attrs,
+       .mmap           = arm_iommu_mmap_attrs,
+
+       .map_page               = arm_iommu_map_page,
+       .unmap_page             = arm_iommu_unmap_page,
+       .sync_single_for_cpu    = arm_iommu_sync_single_for_cpu,
+       .sync_single_for_device = arm_iommu_sync_single_for_device,
+
+       .map_sg                 = arm_iommu_map_sg,
+       .unmap_sg               = arm_iommu_unmap_sg,
+       .sync_sg_for_cpu        = arm_iommu_sync_sg_for_cpu,
+       .sync_sg_for_device     = arm_iommu_sync_sg_for_device,
+};
+
+/**
+ * arm_iommu_create_mapping
+ * @bus: pointer to the bus holding the client device (for IOMMU calls)
+ * @base: start address of the valid IO address space
+ * @size: size of the valid IO address space
+ * @order: accuracy of the IO addresses allocations
+ *
+ * Creates a mapping structure which holds information about used/unused
+ * IO address ranges, which is required to perform memory allocation and
+ * mapping with IOMMU aware functions.
+ *
+ * The client device need to be attached to the mapping with
+ * arm_iommu_attach_device function.
+ */
+struct dma_iommu_mapping *
+arm_iommu_create_mapping(struct bus_type *bus, dma_addr_t base, size_t size,
+                        int order)
+{
+       unsigned int count = size >> (PAGE_SHIFT + order);
+       unsigned int bitmap_size = BITS_TO_LONGS(count) * sizeof(long);
+       struct dma_iommu_mapping *mapping;
+       int err = -ENOMEM;
+
+       if (!count)
+               return ERR_PTR(-EINVAL);
+
+       mapping = kzalloc(sizeof(struct dma_iommu_mapping), GFP_KERNEL);
+       if (!mapping)
+               goto err;
+
+       mapping->bitmap = kzalloc(bitmap_size, GFP_KERNEL);
+       if (!mapping->bitmap)
+               goto err2;
+
+       mapping->base = base;
+       mapping->bits = BITS_PER_BYTE * bitmap_size;
+       mapping->order = order;
+       spin_lock_init(&mapping->lock);
+
+       mapping->domain = iommu_domain_alloc(bus);
+       if (!mapping->domain)
+               goto err3;
+
+       kref_init(&mapping->kref);
+       return mapping;
+err3:
+       kfree(mapping->bitmap);
+err2:
+       kfree(mapping);
+err:
+       return ERR_PTR(err);
+}
+
+static void release_iommu_mapping(struct kref *kref)
+{
+       struct dma_iommu_mapping *mapping =
+               container_of(kref, struct dma_iommu_mapping, kref);
+
+       iommu_domain_free(mapping->domain);
+       kfree(mapping->bitmap);
+       kfree(mapping);
+}
+
+void arm_iommu_release_mapping(struct dma_iommu_mapping *mapping)
+{
+       if (mapping)
+               kref_put(&mapping->kref, release_iommu_mapping);
+}
+
+/**
+ * arm_iommu_attach_device
+ * @dev: valid struct device pointer
+ * @mapping: io address space mapping structure (returned from
+ *     arm_iommu_create_mapping)
+ *
+ * Attaches specified io address space mapping to the provided device,
+ * this replaces the dma operations (dma_map_ops pointer) with the
+ * IOMMU aware version. More than one client might be attached to
+ * the same io address space mapping.
+ */
+int arm_iommu_attach_device(struct device *dev,
+                           struct dma_iommu_mapping *mapping)
+{
+       int err;
+
+       err = iommu_attach_device(mapping->domain, dev);
+       if (err)
+               return err;
+
+       kref_get(&mapping->kref);
+       dev->archdata.mapping = mapping;
+       set_dma_ops(dev, &iommu_ops);
+
+       pr_info("Attached IOMMU controller to %s device.\n", dev_name(dev));
+       return 0;
+}
+
+#endif