drivers/gpu/ion/ion_iommu_heap.c

/*
 * Copyright (c) 2011-2013, The Linux Foundation. All rights reserved.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 and
 * only version 2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 */
#include <linux/err.h>
#include <linux/io.h>
#include <linux/msm_ion.h>
#include <linux/mm.h>
#include <linux/highmem.h>
#include <linux/scatterlist.h>
#include <linux/slab.h>
#include <linux/vmalloc.h>
#include <linux/iommu.h>
#include <linux/pfn.h>
#include <linux/dma-mapping.h>
#include "ion_priv.h"

#include <asm/mach/map.h>
#include <asm/page.h>
#include <asm/cacheflush.h>
#include <mach/iommu_domains.h>
#include <trace/events/kmem.h>

struct ion_iommu_heap {
	struct ion_heap heap;
};

/*
 * We will attempt to allocate high-order pages and store those in an
 * sg_list. However, some APIs expect an array of struct page * where
 * each page is of size PAGE_SIZE. We use this extra structure to
 * carry around an array of such pages (derived from the high-order
 * pages with nth_page).
 */
struct ion_iommu_priv_data {
	struct page **pages;
	unsigned int pages_uses_vmalloc;
	int nrpages;
	unsigned long size;
};

#define MAX_VMAP_RETRIES 10

static const unsigned int orders[] = {8, 4, 0};
static const int num_orders = ARRAY_SIZE(orders);

struct page_info {
	struct page *page;
	unsigned int order;
	struct list_head list;
};

static unsigned int order_to_size(int order)
{
	return PAGE_SIZE << order;
}

static struct page_info *alloc_largest_available(unsigned long size,
						unsigned int max_order)
{
	struct page *page;
	struct page_info *info;
	int i;

	for (i = 0; i < num_orders; i++) {
		gfp_t gfp;
		if (size < order_to_size(orders[i]))
			continue;
		if (max_order < orders[i])
			continue;

		gfp = __GFP_HIGHMEM;

		if (orders[i]) {
			gfp |= __GFP_COMP | __GFP_NORETRY |
			       __GFP_NO_KSWAPD | __GFP_NOWARN;
		} else {
			gfp |= GFP_KERNEL;
		}
		trace_alloc_pages_iommu_start(gfp, orders[i]);
		page = alloc_pages(gfp, orders[i]);
		trace_alloc_pages_iommu_end(gfp, orders[i]);
		if (!page) {
			trace_alloc_pages_iommu_fail(gfp, orders[i]);
			continue;
		}

		info = kmalloc(sizeof(struct page_info), GFP_KERNEL);
		info->page = page;
		info->order = orders[i];
		return info;
	}
	return NULL;
}

static int ion_iommu_heap_allocate(struct ion_heap *heap,
				      struct ion_buffer *buffer,
				      unsigned long size, unsigned long align,
				      unsigned long flags)
{
	int ret, i;
	struct list_head pages_list;
	struct page_info *info, *tmp_info;
	struct ion_iommu_priv_data *data = NULL;

	if (msm_use_iommu()) {
		struct scatterlist *sg;
		struct sg_table *table;
		int j;
		void *ptr = NULL;
		unsigned int npages_to_vmap, total_pages, num_large_pages = 0;
		unsigned long size_remaining = PAGE_ALIGN(size);
		unsigned int max_order = ION_IS_CACHED(flags) ? 0 : orders[0];
		unsigned int page_tbl_size;

		data = kmalloc(sizeof(*data), GFP_KERNEL);
		if (!data)
			return -ENOMEM;

		INIT_LIST_HEAD(&pages_list);
		while (size_remaining > 0) {
			info = alloc_largest_available(size_remaining,
						max_order);
			if (!info) {
				ret = -ENOMEM;
				goto err_free_data;
			}
			list_add_tail(&info->list, &pages_list);
			size_remaining -= order_to_size(info->order);
			max_order = info->order;
			num_large_pages++;
		}

		data->size = PFN_ALIGN(size);
		data->nrpages = data->size >> PAGE_SHIFT;
		data->pages_uses_vmalloc = 0;
		page_tbl_size = sizeof(struct page *) * data->nrpages;

		if (page_tbl_size > SZ_8K) {
			/*
			 * Do fallback to ensure we have a balance between
			 * performance and availability.
			 */
			data->pages = kmalloc(page_tbl_size,
					      __GFP_COMP | __GFP_NORETRY |
					      __GFP_NO_KSWAPD | __GFP_NOWARN);
			if (!data->pages) {
				data->pages = vmalloc(page_tbl_size);
				data->pages_uses_vmalloc = 1;
			}
		} else {
			data->pages = kmalloc(page_tbl_size, GFP_KERNEL);
		}
		if (!data->pages) {
			ret = -ENOMEM;
			goto err_free_data;
		}

		table = buffer->sg_table =
				kzalloc(sizeof(struct sg_table), GFP_KERNEL);

		if (!table) {
			ret = -ENOMEM;
			goto err1;
		}
		ret = sg_alloc_table(table, num_large_pages, GFP_KERNEL);
		if (ret)
			goto err2;

		i = 0;
		sg = table->sgl;
		list_for_each_entry_safe(info, tmp_info, &pages_list, list) {
			struct page *page = info->page;
			sg_set_page(sg, page, order_to_size(info->order), 0);
			sg_dma_address(sg) = sg_phys(sg);
			sg = sg_next(sg);
			for (j = 0; j < (1 << info->order); ++j)
				data->pages[i++] = nth_page(page, j);
			list_del(&info->list);
			kfree(info);
		}

		/*
		 * As an optimization, we omit __GFP_ZERO from
		 * alloc_page above and manually zero out all of the
		 * pages in one fell swoop here. To safeguard against
		 * insufficient vmalloc space, we only vmap
		 * `npages_to_vmap' at a time, starting with a
		 * conservative estimate of 1/8 of the total number of
		 * vmalloc pages available. Note that the `pages'
		 * array is composed of all 4K pages, irrespective of
		 * the size of the pages on the sg list.
		 */
		npages_to_vmap = ((VMALLOC_END - VMALLOC_START)/8)
			>> PAGE_SHIFT;
		total_pages = data->nrpages;
		for (i = 0; i < total_pages; i += npages_to_vmap) {
			npages_to_vmap = min(npages_to_vmap, total_pages - i);
			for (j = 0; j < MAX_VMAP_RETRIES && npages_to_vmap;
			     ++j) {
				ptr = vmap(&data->pages[i], npages_to_vmap,
					VM_IOREMAP, pgprot_kernel);
				if (ptr)
					break;
				else
					npages_to_vmap >>= 1;
			}
			if (!ptr) {
				pr_err("Couldn't vmap the pages for zeroing\n");
				ret = -ENOMEM;
				goto err3;
			}
			memset(ptr, 0, npages_to_vmap * PAGE_SIZE);
			vunmap(ptr);
		}

		if (!ION_IS_CACHED(flags))
			dma_sync_sg_for_device(NULL, table->sgl, table->nents,
						DMA_BIDIRECTIONAL);

		buffer->priv_virt = data;
		return 0;

	} else {
		return -ENOMEM;
	}


err3:
	sg_free_table(buffer->sg_table);
err2:
	kfree(buffer->sg_table);
	buffer->sg_table = 0;
err1:
	if (data->pages_uses_vmalloc)
		vfree(data->pages);
	else
		kfree(data->pages);
err_free_data:
	kfree(data);

	list_for_each_entry_safe(info, tmp_info, &pages_list, list) {
		if (info->page)
			__free_pages(info->page, info->order);
		list_del(&info->list);
		kfree(info);
	}
	return ret;
}

static void ion_iommu_heap_free(struct ion_buffer *buffer)
{
	int i;
	struct scatterlist *sg;
	struct sg_table *table = buffer->sg_table;
	struct ion_iommu_priv_data *data = buffer->priv_virt;

	if (!table)
		return;
	if (!data)
		return;

	for_each_sg(table->sgl, sg, table->nents, i)
		__free_pages(sg_page(sg), get_order(sg_dma_len(sg)));

	sg_free_table(table);
	kfree(table);
	table = 0;
	if (data->pages_uses_vmalloc)
		vfree(data->pages);
	else
		kfree(data->pages);
	kfree(data);
}

void *ion_iommu_heap_map_kernel(struct ion_heap *heap,
				struct ion_buffer *buffer)
{
	struct ion_iommu_priv_data *data = buffer->priv_virt;
	pgprot_t page_prot = PAGE_KERNEL;

	if (!data)
		return NULL;

	if (!ION_IS_CACHED(buffer->flags))
		page_prot = pgprot_writecombine(page_prot);

	buffer->vaddr = vmap(data->pages, data->nrpages, VM_IOREMAP, page_prot);

	return buffer->vaddr;
}

void ion_iommu_heap_unmap_kernel(struct ion_heap *heap,
				    struct ion_buffer *buffer)
{
	if (!buffer->vaddr)
		return;

	vunmap(buffer->vaddr);
	buffer->vaddr = NULL;
}

int ion_iommu_heap_map_user(struct ion_heap *heap, struct ion_buffer *buffer,
			       struct vm_area_struct *vma)
{
	struct sg_table *table = buffer->sg_table;
	unsigned long addr = vma->vm_start;
	unsigned long offset = vma->vm_pgoff * PAGE_SIZE;
	struct scatterlist *sg;
	int i;

	if (!ION_IS_CACHED(buffer->flags))
		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);

	for_each_sg(table->sgl, sg, table->nents, i) {
		struct page *page = sg_page(sg);
		unsigned long remainder = vma->vm_end - addr;
		unsigned long len = sg_dma_len(sg);

		if (offset >= sg_dma_len(sg)) {
			offset -= sg_dma_len(sg);
			continue;
		} else if (offset) {
			page += offset / PAGE_SIZE;
			len = sg_dma_len(sg) - offset;
			offset = 0;
		}
		len = min(len, remainder);
		remap_pfn_range(vma, addr, page_to_pfn(page), len,
				vma->vm_page_prot);
		addr += len;
		if (addr >= vma->vm_end)
			return 0;
	}
	return 0;
}

static struct sg_table *ion_iommu_heap_map_dma(struct ion_heap *heap,
					      struct ion_buffer *buffer)
{
	return buffer->sg_table;
}

static void ion_iommu_heap_unmap_dma(struct ion_heap *heap,
				 struct ion_buffer *buffer)
{
}

static struct ion_heap_ops iommu_heap_ops = {
	.allocate = ion_iommu_heap_allocate,
	.free = ion_iommu_heap_free,
	.map_user = ion_iommu_heap_map_user,
	.map_kernel = ion_iommu_heap_map_kernel,
	.unmap_kernel = ion_iommu_heap_unmap_kernel,
	.map_dma = ion_iommu_heap_map_dma,
	.unmap_dma = ion_iommu_heap_unmap_dma,
};

struct ion_heap *ion_iommu_heap_create(struct ion_platform_heap *heap_data)
{
	struct ion_iommu_heap *iommu_heap;

	iommu_heap = kzalloc(sizeof(struct ion_iommu_heap), GFP_KERNEL);
	if (!iommu_heap)
		return ERR_PTR(-ENOMEM);

	iommu_heap->heap.ops = &iommu_heap_ops;
	iommu_heap->heap.type = ION_HEAP_TYPE_IOMMU;

	return &iommu_heap->heap;
}

void ion_iommu_heap_destroy(struct ion_heap *heap)
{
	struct ion_iommu_heap *iommu_heap =
	     container_of(heap, struct  ion_iommu_heap, heap);

	kfree(iommu_heap);
	iommu_heap = NULL;
}