Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc
[sfrench/cifs-2.6.git] / arch / arm64 / mm / dma-mapping.c
index 072c51fb07d73031578267eea0d5f877685f2676..3a703e5d4e3237f9844d09e871ef1eaa62b781cc 100644 (file)
 
 #include <linux/gfp.h>
 #include <linux/acpi.h>
-#include <linux/bootmem.h>
+#include <linux/memblock.h>
 #include <linux/cache.h>
 #include <linux/export.h>
 #include <linux/slab.h>
 #include <linux/genalloc.h>
 #include <linux/dma-direct.h>
+#include <linux/dma-noncoherent.h>
 #include <linux/dma-contiguous.h>
 #include <linux/vmalloc.h>
 #include <linux/swiotlb.h>
 
 #include <asm/cacheflush.h>
 
-static int swiotlb __ro_after_init;
-
-static pgprot_t __get_dma_pgprot(unsigned long attrs, pgprot_t prot,
-                                bool coherent)
-{
-       if (!coherent || (attrs & DMA_ATTR_WRITE_COMBINE))
-               return pgprot_writecombine(prot);
-       return prot;
-}
-
 static struct gen_pool *atomic_pool __ro_after_init;
 
 #define DEFAULT_DMA_COHERENT_POOL_SIZE  SZ_256K
@@ -91,18 +82,16 @@ static int __free_from_pool(void *start, size_t size)
        return 1;
 }
 
-static void *__dma_alloc(struct device *dev, size_t size,
-                        dma_addr_t *dma_handle, gfp_t flags,
-                        unsigned long attrs)
+void *arch_dma_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
+               gfp_t flags, unsigned long attrs)
 {
        struct page *page;
        void *ptr, *coherent_ptr;
-       bool coherent = is_device_dma_coherent(dev);
-       pgprot_t prot = __get_dma_pgprot(attrs, PAGE_KERNEL, false);
+       pgprot_t prot = pgprot_writecombine(PAGE_KERNEL);
 
        size = PAGE_ALIGN(size);
 
-       if (!coherent && !gfpflags_allow_blocking(flags)) {
+       if (!gfpflags_allow_blocking(flags)) {
                struct page *page = NULL;
                void *addr = __alloc_from_pool(size, &page, flags);
 
@@ -112,14 +101,10 @@ static void *__dma_alloc(struct device *dev, size_t size,
                return addr;
        }
 
-       ptr = swiotlb_alloc(dev, size, dma_handle, flags, attrs);
+       ptr = dma_direct_alloc_pages(dev, size, dma_handle, flags, attrs);
        if (!ptr)
                goto no_mem;
 
-       /* no need for non-cacheable mapping if coherent */
-       if (coherent)
-               return ptr;
-
        /* remove any dirty cache lines on the kernel alias */
        __dma_flush_area(ptr, size);
 
@@ -133,130 +118,57 @@ static void *__dma_alloc(struct device *dev, size_t size,
        return coherent_ptr;
 
 no_map:
-       swiotlb_free(dev, size, ptr, *dma_handle, attrs);
+       dma_direct_free_pages(dev, size, ptr, *dma_handle, attrs);
 no_mem:
        return NULL;
 }
 
-static void __dma_free(struct device *dev, size_t size,
-                      void *vaddr, dma_addr_t dma_handle,
-                      unsigned long attrs)
+void arch_dma_free(struct device *dev, size_t size, void *vaddr,
+               dma_addr_t dma_handle, unsigned long attrs)
 {
-       void *swiotlb_addr = phys_to_virt(dma_to_phys(dev, dma_handle));
+       if (!__free_from_pool(vaddr, PAGE_ALIGN(size))) {
+               void *kaddr = phys_to_virt(dma_to_phys(dev, dma_handle));
 
-       size = PAGE_ALIGN(size);
-
-       if (!is_device_dma_coherent(dev)) {
-               if (__free_from_pool(vaddr, size))
-                       return;
                vunmap(vaddr);
+               dma_direct_free_pages(dev, size, kaddr, dma_handle, attrs);
        }
-       swiotlb_free(dev, size, swiotlb_addr, dma_handle, attrs);
 }
 
-static dma_addr_t __swiotlb_map_page(struct device *dev, struct page *page,
-                                    unsigned long offset, size_t size,
-                                    enum dma_data_direction dir,
-                                    unsigned long attrs)
+long arch_dma_coherent_to_pfn(struct device *dev, void *cpu_addr,
+               dma_addr_t dma_addr)
 {
-       dma_addr_t dev_addr;
-
-       dev_addr = swiotlb_map_page(dev, page, offset, size, dir, attrs);
-       if (!is_device_dma_coherent(dev) &&
-           (attrs & DMA_ATTR_SKIP_CPU_SYNC) == 0)
-               __dma_map_area(phys_to_virt(dma_to_phys(dev, dev_addr)), size, dir);
-
-       return dev_addr;
-}
-
-
-static void __swiotlb_unmap_page(struct device *dev, dma_addr_t dev_addr,
-                                size_t size, enum dma_data_direction dir,
-                                unsigned long attrs)
-{
-       if (!is_device_dma_coherent(dev) &&
-           (attrs & DMA_ATTR_SKIP_CPU_SYNC) == 0)
-               __dma_unmap_area(phys_to_virt(dma_to_phys(dev, dev_addr)), size, dir);
-       swiotlb_unmap_page(dev, dev_addr, size, dir, attrs);
+       return __phys_to_pfn(dma_to_phys(dev, dma_addr));
 }
 
-static int __swiotlb_map_sg_attrs(struct device *dev, struct scatterlist *sgl,
-                                 int nelems, enum dma_data_direction dir,
-                                 unsigned long attrs)
+pgprot_t arch_dma_mmap_pgprot(struct device *dev, pgprot_t prot,
+               unsigned long attrs)
 {
-       struct scatterlist *sg;
-       int i, ret;
-
-       ret = swiotlb_map_sg_attrs(dev, sgl, nelems, dir, attrs);
-       if (!is_device_dma_coherent(dev) &&
-           (attrs & DMA_ATTR_SKIP_CPU_SYNC) == 0)
-               for_each_sg(sgl, sg, ret, i)
-                       __dma_map_area(phys_to_virt(dma_to_phys(dev, sg->dma_address)),
-                                      sg->length, dir);
-
-       return ret;
-}
-
-static void __swiotlb_unmap_sg_attrs(struct device *dev,
-                                    struct scatterlist *sgl, int nelems,
-                                    enum dma_data_direction dir,
-                                    unsigned long attrs)
-{
-       struct scatterlist *sg;
-       int i;
-
-       if (!is_device_dma_coherent(dev) &&
-           (attrs & DMA_ATTR_SKIP_CPU_SYNC) == 0)
-               for_each_sg(sgl, sg, nelems, i)
-                       __dma_unmap_area(phys_to_virt(dma_to_phys(dev, sg->dma_address)),
-                                        sg->length, dir);
-       swiotlb_unmap_sg_attrs(dev, sgl, nelems, dir, attrs);
+       if (!dev_is_dma_coherent(dev) || (attrs & DMA_ATTR_WRITE_COMBINE))
+               return pgprot_writecombine(prot);
+       return prot;
 }
 
-static void __swiotlb_sync_single_for_cpu(struct device *dev,
-                                         dma_addr_t dev_addr, size_t size,
-                                         enum dma_data_direction dir)
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
+               size_t size, enum dma_data_direction dir)
 {
-       if (!is_device_dma_coherent(dev))
-               __dma_unmap_area(phys_to_virt(dma_to_phys(dev, dev_addr)), size, dir);
-       swiotlb_sync_single_for_cpu(dev, dev_addr, size, dir);
+       __dma_map_area(phys_to_virt(paddr), size, dir);
 }
 
-static void __swiotlb_sync_single_for_device(struct device *dev,
-                                            dma_addr_t dev_addr, size_t size,
-                                            enum dma_data_direction dir)
+void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
+               size_t size, enum dma_data_direction dir)
 {
-       swiotlb_sync_single_for_device(dev, dev_addr, size, dir);
-       if (!is_device_dma_coherent(dev))
-               __dma_map_area(phys_to_virt(dma_to_phys(dev, dev_addr)), size, dir);
+       __dma_unmap_area(phys_to_virt(paddr), size, dir);
 }
 
-static void __swiotlb_sync_sg_for_cpu(struct device *dev,
-                                     struct scatterlist *sgl, int nelems,
-                                     enum dma_data_direction dir)
+static int __swiotlb_get_sgtable_page(struct sg_table *sgt,
+                                     struct page *page, size_t size)
 {
-       struct scatterlist *sg;
-       int i;
-
-       if (!is_device_dma_coherent(dev))
-               for_each_sg(sgl, sg, nelems, i)
-                       __dma_unmap_area(phys_to_virt(dma_to_phys(dev, sg->dma_address)),
-                                        sg->length, dir);
-       swiotlb_sync_sg_for_cpu(dev, sgl, nelems, dir);
-}
+       int ret = sg_alloc_table(sgt, 1, GFP_KERNEL);
 
-static void __swiotlb_sync_sg_for_device(struct device *dev,
-                                        struct scatterlist *sgl, int nelems,
-                                        enum dma_data_direction dir)
-{
-       struct scatterlist *sg;
-       int i;
+       if (!ret)
+               sg_set_page(sgt->sgl, page, PAGE_ALIGN(size), 0);
 
-       swiotlb_sync_sg_for_device(dev, sgl, nelems, dir);
-       if (!is_device_dma_coherent(dev))
-               for_each_sg(sgl, sg, nelems, i)
-                       __dma_map_area(phys_to_virt(dma_to_phys(dev, sg->dma_address)),
-                                      sg->length, dir);
+       return ret;
 }
 
 static int __swiotlb_mmap_pfn(struct vm_area_struct *vma,
@@ -277,74 +189,6 @@ static int __swiotlb_mmap_pfn(struct vm_area_struct *vma,
        return ret;
 }
 
-static int __swiotlb_mmap(struct device *dev,
-                         struct vm_area_struct *vma,
-                         void *cpu_addr, dma_addr_t dma_addr, size_t size,
-                         unsigned long attrs)
-{
-       int ret;
-       unsigned long pfn = dma_to_phys(dev, dma_addr) >> PAGE_SHIFT;
-
-       vma->vm_page_prot = __get_dma_pgprot(attrs, vma->vm_page_prot,
-                                            is_device_dma_coherent(dev));
-
-       if (dma_mmap_from_dev_coherent(dev, vma, cpu_addr, size, &ret))
-               return ret;
-
-       return __swiotlb_mmap_pfn(vma, pfn, size);
-}
-
-static int __swiotlb_get_sgtable_page(struct sg_table *sgt,
-                                     struct page *page, size_t size)
-{
-       int ret = sg_alloc_table(sgt, 1, GFP_KERNEL);
-
-       if (!ret)
-               sg_set_page(sgt->sgl, page, PAGE_ALIGN(size), 0);
-
-       return ret;
-}
-
-static int __swiotlb_get_sgtable(struct device *dev, struct sg_table *sgt,
-                                void *cpu_addr, dma_addr_t handle, size_t size,
-                                unsigned long attrs)
-{
-       struct page *page = phys_to_page(dma_to_phys(dev, handle));
-
-       return __swiotlb_get_sgtable_page(sgt, page, size);
-}
-
-static int __swiotlb_dma_supported(struct device *hwdev, u64 mask)
-{
-       if (swiotlb)
-               return swiotlb_dma_supported(hwdev, mask);
-       return 1;
-}
-
-static int __swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t addr)
-{
-       if (swiotlb)
-               return swiotlb_dma_mapping_error(hwdev, addr);
-       return 0;
-}
-
-static const struct dma_map_ops arm64_swiotlb_dma_ops = {
-       .alloc = __dma_alloc,
-       .free = __dma_free,
-       .mmap = __swiotlb_mmap,
-       .get_sgtable = __swiotlb_get_sgtable,
-       .map_page = __swiotlb_map_page,
-       .unmap_page = __swiotlb_unmap_page,
-       .map_sg = __swiotlb_map_sg_attrs,
-       .unmap_sg = __swiotlb_unmap_sg_attrs,
-       .sync_single_for_cpu = __swiotlb_sync_single_for_cpu,
-       .sync_single_for_device = __swiotlb_sync_single_for_device,
-       .sync_sg_for_cpu = __swiotlb_sync_sg_for_cpu,
-       .sync_sg_for_device = __swiotlb_sync_sg_for_device,
-       .dma_supported = __swiotlb_dma_supported,
-       .mapping_error = __swiotlb_dma_mapping_error,
-};
-
 static int __init atomic_pool_init(void)
 {
        pgprot_t prot = __pgprot(PROT_NORMAL_NC);
@@ -500,10 +344,6 @@ EXPORT_SYMBOL(dummy_dma_ops);
 
 static int __init arm64_dma_init(void)
 {
-       if (swiotlb_force == SWIOTLB_FORCE ||
-           max_pfn > (arm64_dma_phys_limit >> PAGE_SHIFT))
-               swiotlb = 1;
-
        WARN_TAINT(ARCH_DMA_MINALIGN < cache_line_size(),
                   TAINT_CPU_OUT_OF_SPEC,
                   "ARCH_DMA_MINALIGN smaller than CTR_EL0.CWG (%d < %d)",
@@ -528,7 +368,7 @@ static void *__iommu_alloc_attrs(struct device *dev, size_t size,
                                 dma_addr_t *handle, gfp_t gfp,
                                 unsigned long attrs)
 {
-       bool coherent = is_device_dma_coherent(dev);
+       bool coherent = dev_is_dma_coherent(dev);
        int ioprot = dma_info_to_prot(DMA_BIDIRECTIONAL, coherent, attrs);
        size_t iosize = size;
        void *addr;
@@ -569,7 +409,7 @@ static void *__iommu_alloc_attrs(struct device *dev, size_t size,
                        addr = NULL;
                }
        } else if (attrs & DMA_ATTR_FORCE_CONTIGUOUS) {
-               pgprot_t prot = __get_dma_pgprot(attrs, PAGE_KERNEL, coherent);
+               pgprot_t prot = arch_dma_mmap_pgprot(dev, PAGE_KERNEL, attrs);
                struct page *page;
 
                page = dma_alloc_from_contiguous(dev, size >> PAGE_SHIFT,
@@ -596,7 +436,7 @@ static void *__iommu_alloc_attrs(struct device *dev, size_t size,
                                                    size >> PAGE_SHIFT);
                }
        } else {
-               pgprot_t prot = __get_dma_pgprot(attrs, PAGE_KERNEL, coherent);
+               pgprot_t prot = arch_dma_mmap_pgprot(dev, PAGE_KERNEL, attrs);
                struct page **pages;
 
                pages = iommu_dma_alloc(dev, iosize, gfp, attrs, ioprot,
@@ -658,8 +498,7 @@ static int __iommu_mmap_attrs(struct device *dev, struct vm_area_struct *vma,
        struct vm_struct *area;
        int ret;
 
-       vma->vm_page_prot = __get_dma_pgprot(attrs, vma->vm_page_prot,
-                                            is_device_dma_coherent(dev));
+       vma->vm_page_prot = arch_dma_mmap_pgprot(dev, vma->vm_page_prot, attrs);
 
        if (dma_mmap_from_dev_coherent(dev, vma, cpu_addr, size, &ret))
                return ret;
@@ -709,11 +548,11 @@ static void __iommu_sync_single_for_cpu(struct device *dev,
 {
        phys_addr_t phys;
 
-       if (is_device_dma_coherent(dev))
+       if (dev_is_dma_coherent(dev))
                return;
 
-       phys = iommu_iova_to_phys(iommu_get_domain_for_dev(dev), dev_addr);
-       __dma_unmap_area(phys_to_virt(phys), size, dir);
+       phys = iommu_iova_to_phys(iommu_get_dma_domain(dev), dev_addr);
+       arch_sync_dma_for_cpu(dev, phys, size, dir);
 }
 
 static void __iommu_sync_single_for_device(struct device *dev,
@@ -722,11 +561,11 @@ static void __iommu_sync_single_for_device(struct device *dev,
 {
        phys_addr_t phys;
 
-       if (is_device_dma_coherent(dev))
+       if (dev_is_dma_coherent(dev))
                return;
 
-       phys = iommu_iova_to_phys(iommu_get_domain_for_dev(dev), dev_addr);
-       __dma_map_area(phys_to_virt(phys), size, dir);
+       phys = iommu_iova_to_phys(iommu_get_dma_domain(dev), dev_addr);
+       arch_sync_dma_for_device(dev, phys, size, dir);
 }
 
 static dma_addr_t __iommu_map_page(struct device *dev, struct page *page,
@@ -734,13 +573,13 @@ static dma_addr_t __iommu_map_page(struct device *dev, struct page *page,
                                   enum dma_data_direction dir,
                                   unsigned long attrs)
 {
-       bool coherent = is_device_dma_coherent(dev);
+       bool coherent = dev_is_dma_coherent(dev);
        int prot = dma_info_to_prot(dir, coherent, attrs);
        dma_addr_t dev_addr = iommu_dma_map_page(dev, page, offset, size, prot);
 
-       if (!iommu_dma_mapping_error(dev, dev_addr) &&
-           (attrs & DMA_ATTR_SKIP_CPU_SYNC) == 0)
-               __iommu_sync_single_for_device(dev, dev_addr, size, dir);
+       if (!coherent && !(attrs & DMA_ATTR_SKIP_CPU_SYNC) &&
+           !iommu_dma_mapping_error(dev, dev_addr))
+               __dma_map_area(page_address(page) + offset, size, dir);
 
        return dev_addr;
 }
@@ -762,11 +601,11 @@ static void __iommu_sync_sg_for_cpu(struct device *dev,
        struct scatterlist *sg;
        int i;
 
-       if (is_device_dma_coherent(dev))
+       if (dev_is_dma_coherent(dev))
                return;
 
        for_each_sg(sgl, sg, nelems, i)
-               __dma_unmap_area(sg_virt(sg), sg->length, dir);
+               arch_sync_dma_for_cpu(dev, sg_phys(sg), sg->length, dir);
 }
 
 static void __iommu_sync_sg_for_device(struct device *dev,
@@ -776,18 +615,18 @@ static void __iommu_sync_sg_for_device(struct device *dev,
        struct scatterlist *sg;
        int i;
 
-       if (is_device_dma_coherent(dev))
+       if (dev_is_dma_coherent(dev))
                return;
 
        for_each_sg(sgl, sg, nelems, i)
-               __dma_map_area(sg_virt(sg), sg->length, dir);
+               arch_sync_dma_for_device(dev, sg_phys(sg), sg->length, dir);
 }
 
 static int __iommu_map_sg_attrs(struct device *dev, struct scatterlist *sgl,
                                int nelems, enum dma_data_direction dir,
                                unsigned long attrs)
 {
-       bool coherent = is_device_dma_coherent(dev);
+       bool coherent = dev_is_dma_coherent(dev);
 
        if ((attrs & DMA_ATTR_SKIP_CPU_SYNC) == 0)
                __iommu_sync_sg_for_device(dev, sgl, nelems, dir);
@@ -879,9 +718,9 @@ void arch_setup_dma_ops(struct device *dev, u64 dma_base, u64 size,
                        const struct iommu_ops *iommu, bool coherent)
 {
        if (!dev->dma_ops)
-               dev->dma_ops = &arm64_swiotlb_dma_ops;
+               dev->dma_ops = &swiotlb_dma_ops;
 
-       dev->archdata.dma_coherent = coherent;
+       dev->dma_coherent = coherent;
        __iommu_setup_dma_ops(dev, dma_base, size, iommu);
 
 #ifdef CONFIG_XEN