[net-next-2.6.git] / arch / arm / mm / consistent.c

/*
 *  linux/arch/arm/mm/consistent.c
 *
 *  Copyright (C) 2000-2004 Russell King
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 *
 *  DMA uncached mapping support.
 */
#include <linux/module.h>
#include <linux/mm.h>
#include <linux/slab.h>
#include <linux/errno.h>
#include <linux/list.h>
#include <linux/init.h>
#include <linux/device.h>
#include <linux/dma-mapping.h>

#include <asm/cacheflush.h>
#include <asm/tlbflush.h>

#define CONSISTENT_BASE	(0xffc00000)
#define CONSISTENT_END	(0xffe00000)
#define CONSISTENT_OFFSET(x)	(((unsigned long)(x) - CONSISTENT_BASE) >> PAGE_SHIFT)

/*
 * This is the page table (2MB) covering uncached, DMA consistent allocations
 */
static pte_t *consistent_pte;
static DEFINE_SPINLOCK(consistent_lock);

/*
 * VM region handling support.
 *
 * This should become something generic, handling VM region allocations for
 * vmalloc and similar (ioremap, module space, etc).
 *
 * I envisage vmalloc()'s supporting vm_struct becoming:
 *
 *  struct vm_struct {
 *    struct vm_region	region;
 *    unsigned long	flags;
 *    struct page	**pages;
 *    unsigned int	nr_pages;
 *    unsigned long	phys_addr;
 *  };
 *
 * get_vm_area() would then call vm_region_alloc with an appropriate
 * struct vm_region head (eg):
 *
 *  struct vm_region vmalloc_head = {
 *	.vm_list	= LIST_HEAD_INIT(vmalloc_head.vm_list),
 *	.vm_start	= VMALLOC_START,
 *	.vm_end		= VMALLOC_END,
 *  };
 *
 * However, vmalloc_head.vm_start is variable (typically, it is dependent on
 * the amount of RAM found at boot time.)  I would imagine that get_vm_area()
 * would have to initialise this each time prior to calling vm_region_alloc().
 */
struct vm_region {
	struct list_head	vm_list;
	unsigned long		vm_start;
	unsigned long		vm_end;
	struct page		*vm_pages;
	int			vm_active;
};

static struct vm_region consistent_head = {
	.vm_list	= LIST_HEAD_INIT(consistent_head.vm_list),
	.vm_start	= CONSISTENT_BASE,
	.vm_end		= CONSISTENT_END,
};

static struct vm_region *
vm_region_alloc(struct vm_region *head, size_t size, gfp_t gfp)
{
	unsigned long addr = head->vm_start, end = head->vm_end - size;
	unsigned long flags;
	struct vm_region *c, *new;

	new = kmalloc(sizeof(struct vm_region), gfp);
	if (!new)
		goto out;

	spin_lock_irqsave(&consistent_lock, flags);

	list_for_each_entry(c, &head->vm_list, vm_list) {
		if ((addr + size) < addr)
			goto nospc;
		if ((addr + size) <= c->vm_start)
			goto found;
		addr = c->vm_end;
		if (addr > end)
			goto nospc;
	}

 found:
	/*
	 * Insert this entry _before_ the one we found.
	 */
	list_add_tail(&new->vm_list, &c->vm_list);
	new->vm_start = addr;
	new->vm_end = addr + size;
	new->vm_active = 1;

	spin_unlock_irqrestore(&consistent_lock, flags);
	return new;

 nospc:
	spin_unlock_irqrestore(&consistent_lock, flags);
	kfree(new);
 out:
	return NULL;
}

static struct vm_region *vm_region_find(struct vm_region *head, unsigned long addr)
{
	struct vm_region *c;
	
	list_for_each_entry(c, &head->vm_list, vm_list) {
		if (c->vm_active && c->vm_start == addr)
			goto out;
	}
	c = NULL;
 out:
	return c;
}

#ifdef CONFIG_HUGETLB_PAGE
#error ARM Coherent DMA allocator does not (yet) support huge TLB
#endif

static void *
__dma_alloc(struct device *dev, size_t size, dma_addr_t *handle, gfp_t gfp,
	    pgprot_t prot)
{
	struct page *page;
	struct vm_region *c;
	unsigned long order;
	u64 mask = ISA_DMA_THRESHOLD, limit;

	if (!consistent_pte) {
		printk(KERN_ERR "%s: not initialised\n", __func__);
		dump_stack();
		return NULL;
	}

	if (dev) {
		mask = dev->coherent_dma_mask;

		/*
		 * Sanity check the DMA mask - it must be non-zero, and
		 * must be able to be satisfied by a DMA allocation.
		 */
		if (mask == 0) {
			dev_warn(dev, "coherent DMA mask is unset\n");
			goto no_page;
		}

		if ((~mask) & ISA_DMA_THRESHOLD) {
			dev_warn(dev, "coherent DMA mask %#llx is smaller "
				 "than system GFP_DMA mask %#llx\n",
				 mask, (unsigned long long)ISA_DMA_THRESHOLD);
			goto no_page;
		}
	}

	/*
	 * Sanity check the allocation size.
	 */
	size = PAGE_ALIGN(size);
	limit = (mask + 1) & ~mask;
	if ((limit && size >= limit) ||
	    size >= (CONSISTENT_END - CONSISTENT_BASE)) {
		printk(KERN_WARNING "coherent allocation too big "
		       "(requested %#x mask %#llx)\n", size, mask);
		goto no_page;
	}

	order = get_order(size);

	if (mask != 0xffffffff)
		gfp |= GFP_DMA;

	page = alloc_pages(gfp, order);
	if (!page)
		goto no_page;

	/*
	 * Invalidate any data that might be lurking in the
	 * kernel direct-mapped region for device DMA.
	 */
	{
		unsigned long kaddr = (unsigned long)page_address(page);
		memset(page_address(page), 0, size);
		dmac_flush_range(kaddr, kaddr + size);
	}

	/*
	 * Allocate a virtual address in the consistent mapping region.
	 */
	c = vm_region_alloc(&consistent_head, size,
			    gfp & ~(__GFP_DMA | __GFP_HIGHMEM));
	if (c) {
		pte_t *pte = consistent_pte + CONSISTENT_OFFSET(c->vm_start);
		struct page *end = page + (1 << order);

		c->vm_pages = page;

		/*
		 * Set the "dma handle"
		 */
		*handle = page_to_dma(dev, page);

		do {
			BUG_ON(!pte_none(*pte));

			set_page_count(page, 1);
			/*
			 * x86 does not mark the pages reserved...
			 */
			SetPageReserved(page);
			set_pte(pte, mk_pte(page, prot));
			page++;
			pte++;
		} while (size -= PAGE_SIZE);

		/*
		 * Free the otherwise unused pages.
		 */
		while (page < end) {
			set_page_count(page, 1);
			__free_page(page);
			page++;
		}

		return (void *)c->vm_start;
	}

	if (page)
		__free_pages(page, order);
 no_page:
	*handle = ~0;
	return NULL;
}

/*
 * Allocate DMA-coherent memory space and return both the kernel remapped
 * virtual and bus address for that space.
 */
void *
dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *handle, gfp_t gfp)
{
	return __dma_alloc(dev, size, handle, gfp,
			   pgprot_noncached(pgprot_kernel));
}
EXPORT_SYMBOL(dma_alloc_coherent);

/*
 * Allocate a writecombining region, in much the same way as
 * dma_alloc_coherent above.
 */
void *
dma_alloc_writecombine(struct device *dev, size_t size, dma_addr_t *handle, gfp_t gfp)
{
	return __dma_alloc(dev, size, handle, gfp,
			   pgprot_writecombine(pgprot_kernel));
}
EXPORT_SYMBOL(dma_alloc_writecombine);

static int dma_mmap(struct device *dev, struct vm_area_struct *vma,
		    void *cpu_addr, dma_addr_t dma_addr, size_t size)
{
	unsigned long flags, user_size, kern_size;
	struct vm_region *c;
	int ret = -ENXIO;

	user_size = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;

	spin_lock_irqsave(&consistent_lock, flags);
	c = vm_region_find(&consistent_head, (unsigned long)cpu_addr);
	spin_unlock_irqrestore(&consistent_lock, flags);

	if (c) {
		unsigned long off = vma->vm_pgoff;

		kern_size = (c->vm_end - c->vm_start) >> PAGE_SHIFT;

		if (off < kern_size &&
		    user_size <= (kern_size - off)) {
			vma->vm_flags |= VM_RESERVED;
			ret = remap_pfn_range(vma, vma->vm_start,
					      page_to_pfn(c->vm_pages) + off,
					      user_size << PAGE_SHIFT,
					      vma->vm_page_prot);
		}
	}

	return ret;
}

int dma_mmap_coherent(struct device *dev, struct vm_area_struct *vma,
		      void *cpu_addr, dma_addr_t dma_addr, size_t size)
{
	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
	return dma_mmap(dev, vma, cpu_addr, dma_addr, size);
}
EXPORT_SYMBOL(dma_mmap_coherent);

int dma_mmap_writecombine(struct device *dev, struct vm_area_struct *vma,
			  void *cpu_addr, dma_addr_t dma_addr, size_t size)
{
	vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
	return dma_mmap(dev, vma, cpu_addr, dma_addr, size);
}
EXPORT_SYMBOL(dma_mmap_writecombine);

/*
 * free a page as defined by the above mapping.
 * Must not be called with IRQs disabled.
 */
void dma_free_coherent(struct device *dev, size_t size, void *cpu_addr, dma_addr_t handle)
{
	struct vm_region *c;
	unsigned long flags, addr;
	pte_t *ptep;

	WARN_ON(irqs_disabled());

	size = PAGE_ALIGN(size);

	spin_lock_irqsave(&consistent_lock, flags);
	c = vm_region_find(&consistent_head, (unsigned long)cpu_addr);
	if (!c)
		goto no_area;

	c->vm_active = 0;
	spin_unlock_irqrestore(&consistent_lock, flags);

	if ((c->vm_end - c->vm_start) != size) {
		printk(KERN_ERR "%s: freeing wrong coherent size (%ld != %d)\n",
		       __func__, c->vm_end - c->vm_start, size);
		dump_stack();
		size = c->vm_end - c->vm_start;
	}

	ptep = consistent_pte + CONSISTENT_OFFSET(c->vm_start);
	addr = c->vm_start;
	do {
		pte_t pte = ptep_get_and_clear(&init_mm, addr, ptep);
		unsigned long pfn;

		ptep++;
		addr += PAGE_SIZE;

		if (!pte_none(pte) && pte_present(pte)) {
			pfn = pte_pfn(pte);

			if (pfn_valid(pfn)) {
				struct page *page = pfn_to_page(pfn);

				/*
				 * x86 does not mark the pages reserved...
				 */
				ClearPageReserved(page);

				__free_page(page);
				continue;
			}
		}

		printk(KERN_CRIT "%s: bad page in kernel page table\n",
		       __func__);
	} while (size -= PAGE_SIZE);

	flush_tlb_kernel_range(c->vm_start, c->vm_end);

	spin_lock_irqsave(&consistent_lock, flags);
	list_del(&c->vm_list);
	spin_unlock_irqrestore(&consistent_lock, flags);

	kfree(c);
	return;

 no_area:
	spin_unlock_irqrestore(&consistent_lock, flags);
	printk(KERN_ERR "%s: trying to free invalid coherent area: %p\n",
	       __func__, cpu_addr);
	dump_stack();
}
EXPORT_SYMBOL(dma_free_coherent);

/*
 * Initialise the consistent memory allocation.
 */
static int __init consistent_init(void)
{
	pgd_t *pgd;
	pmd_t *pmd;
	pte_t *pte;
	int ret = 0;

	do {
		pgd = pgd_offset(&init_mm, CONSISTENT_BASE);
		pmd = pmd_alloc(&init_mm, pgd, CONSISTENT_BASE);
		if (!pmd) {
			printk(KERN_ERR "%s: no pmd tables\n", __func__);
			ret = -ENOMEM;
			break;
		}
		WARN_ON(!pmd_none(*pmd));

		pte = pte_alloc_kernel(pmd, CONSISTENT_BASE);
		if (!pte) {
			printk(KERN_ERR "%s: no pte tables\n", __func__);
			ret = -ENOMEM;
			break;
		}

		consistent_pte = pte;
	} while (0);

	return ret;
}

core_initcall(consistent_init);

/*
 * Make an area consistent for devices.
 */
void consistent_sync(void *vaddr, size_t size, int direction)
{
	unsigned long start = (unsigned long)vaddr;
	unsigned long end   = start + size;

	switch (direction) {
	case DMA_FROM_DEVICE:		/* invalidate only */
		dmac_inv_range(start, end);
		break;
	case DMA_TO_DEVICE:		/* writeback only */
		dmac_clean_range(start, end);
		break;
	case DMA_BIDIRECTIONAL:		/* writeback and invalidate */
		dmac_flush_range(start, end);
		break;
	default:
		BUG();
	}
}
EXPORT_SYMBOL(consistent_sync);
Commit	Line	Data
1da177e4 LT	1	/*
	2	* linux/arch/arm/mm/consistent.c
	3	*
	4	* Copyright (C) 2000-2004 Russell King
	5	*
	6	* This program is free software; you can redistribute it and/or modify
	7	* it under the terms of the GNU General Public License version 2 as
	8	* published by the Free Software Foundation.
	9	*
	10	* DMA uncached mapping support.
	11	*/
	12	#include <linux/module.h>
	13	#include <linux/mm.h>
	14	#include <linux/slab.h>
	15	#include <linux/errno.h>
	16	#include <linux/list.h>
	17	#include <linux/init.h>
	18	#include <linux/device.h>
	19	#include <linux/dma-mapping.h>
	20
	21	#include <asm/cacheflush.h>
1da177e4 LT	22	#include <asm/tlbflush.h>
	23
	24	#define CONSISTENT_BASE (0xffc00000)
	25	#define CONSISTENT_END (0xffe00000)
	26	#define CONSISTENT_OFFSET(x) (((unsigned long)(x) - CONSISTENT_BASE) >> PAGE_SHIFT)
	27
	28	/*
	29	* This is the page table (2MB) covering uncached, DMA consistent allocations
	30	*/
	31	static pte_t *consistent_pte;
	32	static DEFINE_SPINLOCK(consistent_lock);
	33
	34	/*
	35	* VM region handling support.
	36	*
	37	* This should become something generic, handling VM region allocations for
	38	* vmalloc and similar (ioremap, module space, etc).
	39	*
	40	* I envisage vmalloc()'s supporting vm_struct becoming:
	41	*
	42	* struct vm_struct {
	43	* struct vm_region region;
	44	* unsigned long flags;
	45	* struct page **pages;
	46	* unsigned int nr_pages;
	47	* unsigned long phys_addr;
	48	* };
	49	*
	50	* get_vm_area() would then call vm_region_alloc with an appropriate
	51	* struct vm_region head (eg):
	52	*
	53	* struct vm_region vmalloc_head = {
	54	* .vm_list = LIST_HEAD_INIT(vmalloc_head.vm_list),
	55	* .vm_start = VMALLOC_START,
	56	* .vm_end = VMALLOC_END,
	57	* };
	58	*
	59	* However, vmalloc_head.vm_start is variable (typically, it is dependent on
	60	* the amount of RAM found at boot time.) I would imagine that get_vm_area()
	61	* would have to initialise this each time prior to calling vm_region_alloc().
	62	*/
	63	struct vm_region {
	64	struct list_head vm_list;
	65	unsigned long vm_start;
	66	unsigned long vm_end;
	67	struct page *vm_pages;
5edf71ae	68	int vm_active;
1da177e4 LT	69	};
	70
	71	static struct vm_region consistent_head = {
	72	.vm_list = LIST_HEAD_INIT(consistent_head.vm_list),
	73	.vm_start = CONSISTENT_BASE,
	74	.vm_end = CONSISTENT_END,
	75	};
	76
	77	static struct vm_region *
f9e3214a	78	vm_region_alloc(struct vm_region *head, size_t size, gfp_t gfp)
1da177e4 LT	79	{
	80	unsigned long addr = head->vm_start, end = head->vm_end - size;
	81	unsigned long flags;
	82	struct vm_region c, new;
	83
	84	new = kmalloc(sizeof(struct vm_region), gfp);
	85	if (!new)
	86	goto out;
	87
	88	spin_lock_irqsave(&consistent_lock, flags);
	89
	90	list_for_each_entry(c, &head->vm_list, vm_list) {
	91	if ((addr + size) < addr)
	92	goto nospc;
	93	if ((addr + size) <= c->vm_start)
	94	goto found;
	95	addr = c->vm_end;
	96	if (addr > end)
	97	goto nospc;
	98	}
	99
	100	found:
	101	/*
	102	* Insert this entry _before_ the one we found.
	103	*/
	104	list_add_tail(&new->vm_list, &c->vm_list);
	105	new->vm_start = addr;
	106	new->vm_end = addr + size;
5edf71ae	107	new->vm_active = 1;
1da177e4 LT	108
	109	spin_unlock_irqrestore(&consistent_lock, flags);
	110	return new;
	111
	112	nospc:
	113	spin_unlock_irqrestore(&consistent_lock, flags);
	114	kfree(new);
	115	out:
	116	return NULL;
	117	}
	118
	119	static struct vm_region vm_region_find(struct vm_region head, unsigned long addr)
	120	{
	121	struct vm_region *c;
	122
	123	list_for_each_entry(c, &head->vm_list, vm_list) {
5edf71ae	124	if (c->vm_active && c->vm_start == addr)
1da177e4 LT	125	goto out;
	126	}
	127	c = NULL;
	128	out:
	129	return c;
	130	}
	131
	132	#ifdef CONFIG_HUGETLB_PAGE
	133	#error ARM Coherent DMA allocator does not (yet) support huge TLB
	134	#endif
	135
	136	static void *
f9e3214a	137	__dma_alloc(struct device dev, size_t size, dma_addr_t handle, gfp_t gfp,
1da177e4 LT	138	pgprot_t prot)
	139	{
	140	struct page *page;
	141	struct vm_region *c;
	142	unsigned long order;
	143	u64 mask = ISA_DMA_THRESHOLD, limit;
	144
	145	if (!consistent_pte) {
	146	printk(KERN_ERR "%s: not initialised\n", __func__);
	147	dump_stack();
	148	return NULL;
	149	}
	150
	151	if (dev) {
	152	mask = dev->coherent_dma_mask;
	153
	154	/*
	155	* Sanity check the DMA mask - it must be non-zero, and
	156	* must be able to be satisfied by a DMA allocation.
	157	*/
	158	if (mask == 0) {
	159	dev_warn(dev, "coherent DMA mask is unset\n");
	160	goto no_page;
	161	}
	162
	163	if ((~mask) & ISA_DMA_THRESHOLD) {
	164	dev_warn(dev, "coherent DMA mask %#llx is smaller "
	165	"than system GFP_DMA mask %#llx\n",
	166	mask, (unsigned long long)ISA_DMA_THRESHOLD);
	167	goto no_page;
	168	}
	169	}
	170
	171	/*
	172	* Sanity check the allocation size.
	173	*/
	174	size = PAGE_ALIGN(size);
	175	limit = (mask + 1) & ~mask;
	176	if ((limit && size >= limit) \|\|
	177	size >= (CONSISTENT_END - CONSISTENT_BASE)) {
	178	printk(KERN_WARNING "coherent allocation too big "
	179	"(requested %#x mask %#llx)\n", size, mask);
	180	goto no_page;
	181	}
	182
	183	order = get_order(size);
	184
	185	if (mask != 0xffffffff)
	186	gfp \|= GFP_DMA;
	187
	188	page = alloc_pages(gfp, order);
	189	if (!page)
	190	goto no_page;
	191
	192	/*
	193	* Invalidate any data that might be lurking in the
	194	* kernel direct-mapped region for device DMA.
	195	*/
	196	{
	197	unsigned long kaddr = (unsigned long)page_address(page);
	198	memset(page_address(page), 0, size);
	199	dmac_flush_range(kaddr, kaddr + size);
	200	}
	201
202	/*
203	* Allocate a virtual address in the consistent mapping region.
204	*/
205	c = vm_region_alloc(&consistent_head, size,
206	gfp & ~(__GFP_DMA \| __GFP_HIGHMEM));
207	if (c) {
208	pte_t *pte = consistent_pte + CONSISTENT_OFFSET(c->vm_start);
209	struct page *end = page + (1 << order);
210
211	c->vm_pages = page;
212
213	/*
214	* Set the "dma handle"
215	*/
216	*handle = page_to_dma(dev, page);
217
218	do {
219	BUG_ON(!pte_none(*pte));
220
221	set_page_count(page, 1);
222	/*
223	* x86 does not mark the pages reserved...
224	*/
225	SetPageReserved(page);
226	set_pte(pte, mk_pte(page, prot));
227	page++;
228	pte++;
229	} while (size -= PAGE_SIZE);
230
231	/*
232	* Free the otherwise unused pages.
233	*/
234	while (page < end) {
235	set_page_count(page, 1);
236	__free_page(page);
237	page++;
238	}
239
240	return (void *)c->vm_start;
241	}
242
243	if (page)
244	__free_pages(page, order);
245	no_page:
246	*handle = ~0;
247	return NULL;
248	}
249
250	/*
251	* Allocate DMA-coherent memory space and return both the kernel remapped
252	* virtual and bus address for that space.
253	*/
254	void *
f9e3214a	255	dma_alloc_coherent(struct device dev, size_t size, dma_addr_t handle, gfp_t gfp)
1da177e4 LT	256	{
	257	return __dma_alloc(dev, size, handle, gfp,
	258	pgprot_noncached(pgprot_kernel));
	259	}
	260	EXPORT_SYMBOL(dma_alloc_coherent);
	261
	262	/*
	263	* Allocate a writecombining region, in much the same way as
	264	* dma_alloc_coherent above.
	265	*/
	266	void *
f9e3214a	267	dma_alloc_writecombine(struct device dev, size_t size, dma_addr_t handle, gfp_t gfp)
1da177e4 LT	268	{
	269	return __dma_alloc(dev, size, handle, gfp,
	270	pgprot_writecombine(pgprot_kernel));
	271	}
	272	EXPORT_SYMBOL(dma_alloc_writecombine);
	273
	274	static int dma_mmap(struct device dev, struct vm_area_struct vma,
	275	void *cpu_addr, dma_addr_t dma_addr, size_t size)
	276	{
	277	unsigned long flags, user_size, kern_size;
	278	struct vm_region *c;
	279	int ret = -ENXIO;
	280
	281	user_size = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
	282
	283	spin_lock_irqsave(&consistent_lock, flags);
	284	c = vm_region_find(&consistent_head, (unsigned long)cpu_addr);
	285	spin_unlock_irqrestore(&consistent_lock, flags);
	286
	287	if (c) {
	288	unsigned long off = vma->vm_pgoff;
	289
	290	kern_size = (c->vm_end - c->vm_start) >> PAGE_SHIFT;
	291
	292	if (off < kern_size &&
	293	user_size <= (kern_size - off)) {
	294	vma->vm_flags \|= VM_RESERVED;
	295	ret = remap_pfn_range(vma, vma->vm_start,
	296	page_to_pfn(c->vm_pages) + off,
	297	user_size << PAGE_SHIFT,
	298	vma->vm_page_prot);
	299	}
	300	}
	301
	302	return ret;
	303	}
	304
	305	int dma_mmap_coherent(struct device dev, struct vm_area_struct vma,
	306	void *cpu_addr, dma_addr_t dma_addr, size_t size)
	307	{
	308	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
	309	return dma_mmap(dev, vma, cpu_addr, dma_addr, size);
	310	}
	311	EXPORT_SYMBOL(dma_mmap_coherent);
	312
	313	int dma_mmap_writecombine(struct device dev, struct vm_area_struct vma,
	314	void *cpu_addr, dma_addr_t dma_addr, size_t size)
	315	{
	316	vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
	317	return dma_mmap(dev, vma, cpu_addr, dma_addr, size);
	318	}
	319	EXPORT_SYMBOL(dma_mmap_writecombine);
	320
	321	/*
	322	* free a page as defined by the above mapping.
5edf71ae	323	* Must not be called with IRQs disabled.
1da177e4 LT	324	*/
	325	void dma_free_coherent(struct device dev, size_t size, void cpu_addr, dma_addr_t handle)
	326	{
	327	struct vm_region *c;
	328	unsigned long flags, addr;
	329	pte_t *ptep;
	330
5edf71ae RK	331	WARN_ON(irqs_disabled());
5edf71ae RK	332
1da177e4 LT	333	size = PAGE_ALIGN(size);
	334
	335	spin_lock_irqsave(&consistent_lock, flags);
1da177e4 LT	336	c = vm_region_find(&consistent_head, (unsigned long)cpu_addr);
	337	if (!c)
	338	goto no_area;
	339
5edf71ae RK	340	c->vm_active = 0;
	341	spin_unlock_irqrestore(&consistent_lock, flags);
	342
1da177e4 LT	343	if ((c->vm_end - c->vm_start) != size) {
	344	printk(KERN_ERR "%s: freeing wrong coherent size (%ld != %d)\n",
	345	__func__, c->vm_end - c->vm_start, size);
	346	dump_stack();
	347	size = c->vm_end - c->vm_start;
	348	}
	349
	350	ptep = consistent_pte + CONSISTENT_OFFSET(c->vm_start);
	351	addr = c->vm_start;
	352	do {
	353	pte_t pte = ptep_get_and_clear(&init_mm, addr, ptep);
	354	unsigned long pfn;
	355
	356	ptep++;
	357	addr += PAGE_SIZE;
	358
	359	if (!pte_none(pte) && pte_present(pte)) {
	360	pfn = pte_pfn(pte);
	361
	362	if (pfn_valid(pfn)) {
	363	struct page *page = pfn_to_page(pfn);
	364
	365	/*
	366	* x86 does not mark the pages reserved...
	367	*/
	368	ClearPageReserved(page);
	369
	370	__free_page(page);
	371	continue;
	372	}
	373	}
	374
	375	printk(KERN_CRIT "%s: bad page in kernel page table\n",
	376	__func__);
	377	} while (size -= PAGE_SIZE);
	378
	379	flush_tlb_kernel_range(c->vm_start, c->vm_end);
	380
5edf71ae	381	spin_lock_irqsave(&consistent_lock, flags);
1da177e4	382	list_del(&c->vm_list);
1da177e4 LT	383	spin_unlock_irqrestore(&consistent_lock, flags);
	384
	385	kfree(c);
	386	return;
	387
	388	no_area:
	389	spin_unlock_irqrestore(&consistent_lock, flags);
	390	printk(KERN_ERR "%s: trying to free invalid coherent area: %p\n",
	391	__func__, cpu_addr);
	392	dump_stack();
	393	}
	394	EXPORT_SYMBOL(dma_free_coherent);
	395
	396	/*
	397	* Initialise the consistent memory allocation.
	398	*/
	399	static int __init consistent_init(void)
	400	{
	401	pgd_t *pgd;
	402	pmd_t *pmd;
	403	pte_t *pte;
	404	int ret = 0;
	405
1da177e4 LT	406	do {
	407	pgd = pgd_offset(&init_mm, CONSISTENT_BASE);
	408	pmd = pmd_alloc(&init_mm, pgd, CONSISTENT_BASE);
	409	if (!pmd) {
	410	printk(KERN_ERR "%s: no pmd tables\n", __func__);
	411	ret = -ENOMEM;
	412	break;
	413	}
	414	WARN_ON(!pmd_none(*pmd));
	415
872fec16	416	pte = pte_alloc_kernel(pmd, CONSISTENT_BASE);
1da177e4 LT	417	if (!pte) {
	418	printk(KERN_ERR "%s: no pte tables\n", __func__);
	419	ret = -ENOMEM;
	420	break;
	421	}
	422
	423	consistent_pte = pte;
	424	} while (0);
	425
1da177e4 LT	426	return ret;
	427	}
	428
	429	core_initcall(consistent_init);
	430
	431	/*
	432	* Make an area consistent for devices.
	433	*/
	434	void consistent_sync(void *vaddr, size_t size, int direction)
	435	{
	436	unsigned long start = (unsigned long)vaddr;
	437	unsigned long end = start + size;
	438
	439	switch (direction) {
	440	case DMA_FROM_DEVICE: /* invalidate only */
	441	dmac_inv_range(start, end);
	442	break;
	443	case DMA_TO_DEVICE: /* writeback only */
	444	dmac_clean_range(start, end);
	445	break;
	446	case DMA_BIDIRECTIONAL: /* writeback and invalidate */
	447	dmac_flush_range(start, end);
	448	break;
	449	default:
	450	BUG();
	451	}
	452	}
	453	EXPORT_SYMBOL(consistent_sync);