[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / s390 / pci / pci_dma.c

/*
 * Copyright IBM Corp. 2012
 *
 * Author(s):
 *   Jan Glauber <jang@linux.vnet.ibm.com>
 */

#include <linux/kernel.h>
#include <linux/slab.h>
#include <linux/export.h>
#include <linux/iommu-helper.h>
#include <linux/dma-mapping.h>
#include <linux/pci.h>
#include <asm/pci_dma.h>

static struct kmem_cache *dma_region_table_cache;
static struct kmem_cache *dma_page_table_cache;

static unsigned long *dma_alloc_cpu_table(void)
{
	unsigned long *table, *entry;

	table = kmem_cache_alloc(dma_region_table_cache, GFP_ATOMIC);
	if (!table)
		return NULL;

	for (entry = table; entry < table + ZPCI_TABLE_ENTRIES; entry++)
		*entry = ZPCI_TABLE_INVALID | ZPCI_TABLE_PROTECTED;
	return table;
}

static void dma_free_cpu_table(void *table)
{
	kmem_cache_free(dma_region_table_cache, table);
}

static unsigned long *dma_alloc_page_table(void)
{
	unsigned long *table, *entry;

	table = kmem_cache_alloc(dma_page_table_cache, GFP_ATOMIC);
	if (!table)
		return NULL;

	for (entry = table; entry < table + ZPCI_PT_ENTRIES; entry++)
		*entry = ZPCI_PTE_INVALID | ZPCI_TABLE_PROTECTED;
	return table;
}

static void dma_free_page_table(void *table)
{
	kmem_cache_free(dma_page_table_cache, table);
}

static unsigned long *dma_get_seg_table_origin(unsigned long *entry)
{
	unsigned long *sto;

	if (reg_entry_isvalid(*entry))
		sto = get_rt_sto(*entry);
	else {
		sto = dma_alloc_cpu_table();
		if (!sto)
			return NULL;

		set_rt_sto(entry, sto);
		validate_rt_entry(entry);
		entry_clr_protected(entry);
	}
	return sto;
}

static unsigned long *dma_get_page_table_origin(unsigned long *entry)
{
	unsigned long *pto;

	if (reg_entry_isvalid(*entry))
		pto = get_st_pto(*entry);
	else {
		pto = dma_alloc_page_table();
		if (!pto)
			return NULL;
		set_st_pto(entry, pto);
		validate_st_entry(entry);
		entry_clr_protected(entry);
	}
	return pto;
}

static unsigned long *dma_walk_cpu_trans(unsigned long *rto, dma_addr_t dma_addr)
{
	unsigned long *sto, *pto;
	unsigned int rtx, sx, px;

	rtx = calc_rtx(dma_addr);
	sto = dma_get_seg_table_origin(&rto[rtx]);
	if (!sto)
		return NULL;

	sx = calc_sx(dma_addr);
	pto = dma_get_page_table_origin(&sto[sx]);
	if (!pto)
		return NULL;

	px = calc_px(dma_addr);
	return &pto[px];
}

static void dma_update_cpu_trans(struct zpci_dev *zdev, void *page_addr,
				 dma_addr_t dma_addr, int flags)
{
	unsigned long *entry;

	entry = dma_walk_cpu_trans(zdev->dma_table, dma_addr);
	if (!entry) {
		WARN_ON_ONCE(1);
		return;
	}

	if (flags & ZPCI_PTE_INVALID) {
		invalidate_pt_entry(entry);
		return;
	} else {
		set_pt_pfaa(entry, page_addr);
		validate_pt_entry(entry);
	}

	if (flags & ZPCI_TABLE_PROTECTED)
		entry_set_protected(entry);
	else
		entry_clr_protected(entry);
}

static int dma_update_trans(struct zpci_dev *zdev, unsigned long pa,
			    dma_addr_t dma_addr, size_t size, int flags)
{
	unsigned int nr_pages = PAGE_ALIGN(size) >> PAGE_SHIFT;
	u8 *page_addr = (u8 *) (pa & PAGE_MASK);
	dma_addr_t start_dma_addr = dma_addr;
	unsigned long irq_flags;
	int i, rc = 0;

	if (!nr_pages)
		return -EINVAL;

	spin_lock_irqsave(&zdev->dma_table_lock, irq_flags);
	if (!zdev->dma_table) {
		dev_err(&zdev->pdev->dev, "Missing DMA table\n");
		goto no_refresh;
	}

	for (i = 0; i < nr_pages; i++) {
		dma_update_cpu_trans(zdev, page_addr, dma_addr, flags);
		page_addr += PAGE_SIZE;
		dma_addr += PAGE_SIZE;
	}

	/*
	 * rpcit is not required to establish new translations when previously
	 * invalid translation-table entries are validated, however it is
	 * required when altering previously valid entries.
	 */
	if (!zdev->tlb_refresh &&
	    ((flags & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID))
		/*
		 * TODO: also need to check that the old entry is indeed INVALID
		 * and not only for one page but for the whole range...
		 * -> now we WARN_ON in that case but with lazy unmap that
		 * needs to be redone!
		 */
		goto no_refresh;

	rc = s390pci_refresh_trans((u64) zdev->fh << 32, start_dma_addr,
				   nr_pages * PAGE_SIZE);

no_refresh:
	spin_unlock_irqrestore(&zdev->dma_table_lock, irq_flags);
	return rc;
}

static void dma_free_seg_table(unsigned long entry)
{
	unsigned long *sto = get_rt_sto(entry);
	int sx;

	for (sx = 0; sx < ZPCI_TABLE_ENTRIES; sx++)
		if (reg_entry_isvalid(sto[sx]))
			dma_free_page_table(get_st_pto(sto[sx]));

	dma_free_cpu_table(sto);
}

static void dma_cleanup_tables(struct zpci_dev *zdev)
{
	unsigned long *table;
	int rtx;

	if (!zdev || !zdev->dma_table)
		return;

	table = zdev->dma_table;
	for (rtx = 0; rtx < ZPCI_TABLE_ENTRIES; rtx++)
		if (reg_entry_isvalid(table[rtx]))
			dma_free_seg_table(table[rtx]);

	dma_free_cpu_table(table);
	zdev->dma_table = NULL;
}

static unsigned long __dma_alloc_iommu(struct zpci_dev *zdev, unsigned long start,
				   int size)
{
	unsigned long boundary_size = 0x1000000;

	return iommu_area_alloc(zdev->iommu_bitmap, zdev->iommu_pages,
				start, size, 0, boundary_size, 0);
}

static unsigned long dma_alloc_iommu(struct zpci_dev *zdev, int size)
{
	unsigned long offset, flags;

	spin_lock_irqsave(&zdev->iommu_bitmap_lock, flags);
	offset = __dma_alloc_iommu(zdev, zdev->next_bit, size);
	if (offset == -1)
		offset = __dma_alloc_iommu(zdev, 0, size);

	if (offset != -1) {
		zdev->next_bit = offset + size;
		if (zdev->next_bit >= zdev->iommu_pages)
			zdev->next_bit = 0;
	}
	spin_unlock_irqrestore(&zdev->iommu_bitmap_lock, flags);
	return offset;
}

static void dma_free_iommu(struct zpci_dev *zdev, unsigned long offset, int size)
{
	unsigned long flags;

	spin_lock_irqsave(&zdev->iommu_bitmap_lock, flags);
	if (!zdev->iommu_bitmap)
		goto out;
	bitmap_clear(zdev->iommu_bitmap, offset, size);
	if (offset >= zdev->next_bit)
		zdev->next_bit = offset + size;
out:
	spin_unlock_irqrestore(&zdev->iommu_bitmap_lock, flags);
}

int dma_set_mask(struct device *dev, u64 mask)
{
	if (!dev->dma_mask || !dma_supported(dev, mask))
		return -EIO;

	*dev->dma_mask = mask;
	return 0;
}
EXPORT_SYMBOL_GPL(dma_set_mask);

static dma_addr_t s390_dma_map_pages(struct device *dev, struct page *page,
				     unsigned long offset, size_t size,
				     enum dma_data_direction direction,
				     struct dma_attrs *attrs)
{
	struct zpci_dev *zdev = get_zdev(container_of(dev, struct pci_dev, dev));
	unsigned long nr_pages, iommu_page_index;
	unsigned long pa = page_to_phys(page) + offset;
	int flags = ZPCI_PTE_VALID;
	dma_addr_t dma_addr;

	/* This rounds up number of pages based on size and offset */
	nr_pages = iommu_num_pages(pa, size, PAGE_SIZE);
	iommu_page_index = dma_alloc_iommu(zdev, nr_pages);
	if (iommu_page_index == -1)
		goto out_err;

	/* Use rounded up size */
	size = nr_pages * PAGE_SIZE;

	dma_addr = zdev->start_dma + iommu_page_index * PAGE_SIZE;
	if (dma_addr + size > zdev->end_dma) {
		dev_err(dev, "(dma_addr: 0x%16.16LX + size: 0x%16.16lx) > end_dma: 0x%16.16Lx\n",
			 dma_addr, size, zdev->end_dma);
		goto out_free;
	}

	if (direction == DMA_NONE || direction == DMA_TO_DEVICE)
		flags |= ZPCI_TABLE_PROTECTED;

	if (!dma_update_trans(zdev, pa, dma_addr, size, flags)) {
		atomic64_add(nr_pages, (atomic64_t *) &zdev->fmb->mapped_pages);
		return dma_addr + (offset & ~PAGE_MASK);
	}

out_free:
	dma_free_iommu(zdev, iommu_page_index, nr_pages);
out_err:
	dev_err(dev, "Failed to map addr: %lx\n", pa);
	return DMA_ERROR_CODE;
}

static void s390_dma_unmap_pages(struct device *dev, dma_addr_t dma_addr,
				 size_t size, enum dma_data_direction direction,
				 struct dma_attrs *attrs)
{
	struct zpci_dev *zdev = get_zdev(container_of(dev, struct pci_dev, dev));
	unsigned long iommu_page_index;
	int npages;

	npages = iommu_num_pages(dma_addr, size, PAGE_SIZE);
	dma_addr = dma_addr & PAGE_MASK;
	if (dma_update_trans(zdev, 0, dma_addr, npages * PAGE_SIZE,
			     ZPCI_TABLE_PROTECTED | ZPCI_PTE_INVALID))
		dev_err(dev, "Failed to unmap addr: %Lx\n", dma_addr);

	atomic64_add(npages, (atomic64_t *) &zdev->fmb->unmapped_pages);
	iommu_page_index = (dma_addr - zdev->start_dma) >> PAGE_SHIFT;
	dma_free_iommu(zdev, iommu_page_index, npages);
}

static void *s390_dma_alloc(struct device *dev, size_t size,
			    dma_addr_t *dma_handle, gfp_t flag,
			    struct dma_attrs *attrs)
{
	struct zpci_dev *zdev = get_zdev(container_of(dev, struct pci_dev, dev));
	struct page *page;
	unsigned long pa;
	dma_addr_t map;

	size = PAGE_ALIGN(size);
	page = alloc_pages(flag, get_order(size));
	if (!page)
		return NULL;

	atomic64_add(size / PAGE_SIZE, (atomic64_t *) &zdev->fmb->allocated_pages);
	pa = page_to_phys(page);
	memset((void *) pa, 0, size);

	map = s390_dma_map_pages(dev, page, pa % PAGE_SIZE,
				 size, DMA_BIDIRECTIONAL, NULL);
	if (dma_mapping_error(dev, map)) {
		free_pages(pa, get_order(size));
		return NULL;
	}

	if (dma_handle)
		*dma_handle = map;
	return (void *) pa;
}

static void s390_dma_free(struct device *dev, size_t size,
			  void *pa, dma_addr_t dma_handle,
			  struct dma_attrs *attrs)
{
	s390_dma_unmap_pages(dev, dma_handle, PAGE_ALIGN(size),
			     DMA_BIDIRECTIONAL, NULL);
	free_pages((unsigned long) pa, get_order(size));
}

static int s390_dma_map_sg(struct device *dev, struct scatterlist *sg,
			   int nr_elements, enum dma_data_direction dir,
			   struct dma_attrs *attrs)
{
	int mapped_elements = 0;
	struct scatterlist *s;
	int i;

	for_each_sg(sg, s, nr_elements, i) {
		struct page *page = sg_page(s);
		s->dma_address = s390_dma_map_pages(dev, page, s->offset,
						    s->length, dir, NULL);
		if (!dma_mapping_error(dev, s->dma_address)) {
			s->dma_length = s->length;
			mapped_elements++;
		} else
			goto unmap;
	}
out:
	return mapped_elements;

unmap:
	for_each_sg(sg, s, mapped_elements, i) {
		if (s->dma_address)
			s390_dma_unmap_pages(dev, s->dma_address, s->dma_length,
					     dir, NULL);
		s->dma_address = 0;
		s->dma_length = 0;
	}
	mapped_elements = 0;
	goto out;
}

static void s390_dma_unmap_sg(struct device *dev, struct scatterlist *sg,
			      int nr_elements, enum dma_data_direction dir,
			      struct dma_attrs *attrs)
{
	struct scatterlist *s;
	int i;

	for_each_sg(sg, s, nr_elements, i) {
		s390_dma_unmap_pages(dev, s->dma_address, s->dma_length, dir, NULL);
		s->dma_address = 0;
		s->dma_length = 0;
	}
}

int zpci_dma_init_device(struct zpci_dev *zdev)
{
	unsigned int bitmap_order;
	int rc;

	spin_lock_init(&zdev->iommu_bitmap_lock);
	spin_lock_init(&zdev->dma_table_lock);

	zdev->dma_table = dma_alloc_cpu_table();
	if (!zdev->dma_table) {
		rc = -ENOMEM;
		goto out;
	}

	zdev->iommu_size = (unsigned long) high_memory - PAGE_OFFSET;
	zdev->iommu_pages = zdev->iommu_size >> PAGE_SHIFT;
	bitmap_order = get_order(zdev->iommu_pages / 8);
	pr_info("iommu_size: 0x%lx  iommu_pages: 0x%lx  bitmap_order: %i\n",
		 zdev->iommu_size, zdev->iommu_pages, bitmap_order);

	zdev->iommu_bitmap = (void *) __get_free_pages(GFP_KERNEL | __GFP_ZERO,
						       bitmap_order);
	if (!zdev->iommu_bitmap) {
		rc = -ENOMEM;
		goto free_dma_table;
	}

	rc = zpci_register_ioat(zdev,
				0,
				zdev->start_dma + PAGE_OFFSET,
				zdev->start_dma + zdev->iommu_size - 1,
				(u64) zdev->dma_table);
	if (rc)
		goto free_bitmap;

	return 0;
free_bitmap:
	vfree(zdev->iommu_bitmap);
	zdev->iommu_bitmap = NULL;
free_dma_table:
	dma_free_cpu_table(zdev->dma_table);
	zdev->dma_table = NULL;
out:
	return rc;
}

void zpci_dma_exit_device(struct zpci_dev *zdev)
{
	zpci_unregister_ioat(zdev, 0);
	dma_cleanup_tables(zdev);
	free_pages((unsigned long) zdev->iommu_bitmap,
		   get_order(zdev->iommu_pages / 8));
	zdev->iommu_bitmap = NULL;
	zdev->next_bit = 0;
}

static int __init dma_alloc_cpu_table_caches(void)
{
	dma_region_table_cache = kmem_cache_create("PCI_DMA_region_tables",
					ZPCI_TABLE_SIZE, ZPCI_TABLE_ALIGN,
					0, NULL);
	if (!dma_region_table_cache)
		return -ENOMEM;

	dma_page_table_cache = kmem_cache_create("PCI_DMA_page_tables",
					ZPCI_PT_SIZE, ZPCI_PT_ALIGN,
					0, NULL);
	if (!dma_page_table_cache) {
		kmem_cache_destroy(dma_region_table_cache);
		return -ENOMEM;
	}
	return 0;
}

int __init zpci_dma_init(void)
{
	return dma_alloc_cpu_table_caches();
}

void zpci_dma_exit(void)
{
	kmem_cache_destroy(dma_page_table_cache);
	kmem_cache_destroy(dma_region_table_cache);
}

#define PREALLOC_DMA_DEBUG_ENTRIES	(1 << 16)

static int __init dma_debug_do_init(void)
{
	dma_debug_init(PREALLOC_DMA_DEBUG_ENTRIES);
	return 0;
}
fs_initcall(dma_debug_do_init);

struct dma_map_ops s390_dma_ops = {
	.alloc		= s390_dma_alloc,
	.free		= s390_dma_free,
	.map_sg		= s390_dma_map_sg,
	.unmap_sg	= s390_dma_unmap_sg,
	.map_page	= s390_dma_map_pages,
	.unmap_page	= s390_dma_unmap_pages,
	/* if we support direct DMA this must be conditional */
	.is_phys	= 0,
	/* dma_supported is unconditionally true without a callback */
};
EXPORT_SYMBOL_GPL(s390_dma_ops);
Commit	Line	Data
828b35f6 JG	1	/*
	2	* Copyright IBM Corp. 2012
	3	*
	4	* Author(s):
	5	* Jan Glauber <jang@linux.vnet.ibm.com>
	6	*/
	7
	8	#include <linux/kernel.h>
	9	#include <linux/slab.h>
	10	#include <linux/export.h>
	11	#include <linux/iommu-helper.h>
	12	#include <linux/dma-mapping.h>
	13	#include <linux/pci.h>
	14	#include <asm/pci_dma.h>
	15
828b35f6 JG	16	static struct kmem_cache *dma_region_table_cache;
	17	static struct kmem_cache *dma_page_table_cache;
	18
	19	static unsigned long *dma_alloc_cpu_table(void)
	20	{
	21	unsigned long table, entry;
	22
	23	table = kmem_cache_alloc(dma_region_table_cache, GFP_ATOMIC);
	24	if (!table)
	25	return NULL;
	26
	27	for (entry = table; entry < table + ZPCI_TABLE_ENTRIES; entry++)
	28	*entry = ZPCI_TABLE_INVALID \| ZPCI_TABLE_PROTECTED;
	29	return table;
	30	}
	31
	32	static void dma_free_cpu_table(void *table)
	33	{
	34	kmem_cache_free(dma_region_table_cache, table);
	35	}
	36
	37	static unsigned long *dma_alloc_page_table(void)
	38	{
	39	unsigned long table, entry;
	40
	41	table = kmem_cache_alloc(dma_page_table_cache, GFP_ATOMIC);
	42	if (!table)
	43	return NULL;
	44
	45	for (entry = table; entry < table + ZPCI_PT_ENTRIES; entry++)
	46	*entry = ZPCI_PTE_INVALID \| ZPCI_TABLE_PROTECTED;
	47	return table;
	48	}
	49
	50	static void dma_free_page_table(void *table)
	51	{
	52	kmem_cache_free(dma_page_table_cache, table);
	53	}
	54
	55	static unsigned long dma_get_seg_table_origin(unsigned long entry)
	56	{
	57	unsigned long *sto;
	58
	59	if (reg_entry_isvalid(*entry))
	60	sto = get_rt_sto(*entry);
	61	else {
	62	sto = dma_alloc_cpu_table();
	63	if (!sto)
	64	return NULL;
	65
	66	set_rt_sto(entry, sto);
	67	validate_rt_entry(entry);
	68	entry_clr_protected(entry);
	69	}
	70	return sto;
	71	}
	72
	73	static unsigned long dma_get_page_table_origin(unsigned long entry)
	74	{
	75	unsigned long *pto;
	76
	77	if (reg_entry_isvalid(*entry))
	78	pto = get_st_pto(*entry);
	79	else {
80	pto = dma_alloc_page_table();
81	if (!pto)
82	return NULL;
83	set_st_pto(entry, pto);
84	validate_st_entry(entry);
85	entry_clr_protected(entry);
86	}
87	return pto;
88	}
89
90	static unsigned long dma_walk_cpu_trans(unsigned long rto, dma_addr_t dma_addr)
91	{
92	unsigned long sto, pto;
93	unsigned int rtx, sx, px;
94
95	rtx = calc_rtx(dma_addr);
96	sto = dma_get_seg_table_origin(&rto[rtx]);
97	if (!sto)
98	return NULL;
99
100	sx = calc_sx(dma_addr);
101	pto = dma_get_page_table_origin(&sto[sx]);
102	if (!pto)
103	return NULL;
104
105	px = calc_px(dma_addr);
106	return &pto[px];
107	}
108
109	static void dma_update_cpu_trans(struct zpci_dev zdev, void page_addr,
110	dma_addr_t dma_addr, int flags)
111	{
112	unsigned long *entry;
113
114	entry = dma_walk_cpu_trans(zdev->dma_table, dma_addr);
115	if (!entry) {
116	WARN_ON_ONCE(1);
117	return;
118	}
119
120	if (flags & ZPCI_PTE_INVALID) {
121	invalidate_pt_entry(entry);
122	return;
123	} else {
124	set_pt_pfaa(entry, page_addr);
125	validate_pt_entry(entry);
126	}
127
128	if (flags & ZPCI_TABLE_PROTECTED)
129	entry_set_protected(entry);
130	else
131	entry_clr_protected(entry);
132	}
133
134	static int dma_update_trans(struct zpci_dev *zdev, unsigned long pa,
135	dma_addr_t dma_addr, size_t size, int flags)
136	{
137	unsigned int nr_pages = PAGE_ALIGN(size) >> PAGE_SHIFT;
138	u8 page_addr = (u8 ) (pa & PAGE_MASK);
139	dma_addr_t start_dma_addr = dma_addr;
140	unsigned long irq_flags;
141	int i, rc = 0;
142
143	if (!nr_pages)
144	return -EINVAL;
145
146	spin_lock_irqsave(&zdev->dma_table_lock, irq_flags);
147	if (!zdev->dma_table) {
148	dev_err(&zdev->pdev->dev, "Missing DMA table\n");
149	goto no_refresh;
150	}
151
152	for (i = 0; i < nr_pages; i++) {
153	dma_update_cpu_trans(zdev, page_addr, dma_addr, flags);
154	page_addr += PAGE_SIZE;
155	dma_addr += PAGE_SIZE;
156	}
157
158	/*
159	* rpcit is not required to establish new translations when previously
160	* invalid translation-table entries are validated, however it is
161	* required when altering previously valid entries.
162	*/
163	if (!zdev->tlb_refresh &&
164	((flags & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID))
165	/*
166	* TODO: also need to check that the old entry is indeed INVALID
167	* and not only for one page but for the whole range...
168	* -> now we WARN_ON in that case but with lazy unmap that
169	* needs to be redone!
170	*/
171	goto no_refresh;
b2a9e87d SO	172
	173	rc = s390pci_refresh_trans((u64) zdev->fh << 32, start_dma_addr,
	174	nr_pages * PAGE_SIZE);
828b35f6 JG	175
	176	no_refresh:
	177	spin_unlock_irqrestore(&zdev->dma_table_lock, irq_flags);
	178	return rc;
	179	}
	180
	181	static void dma_free_seg_table(unsigned long entry)
	182	{
	183	unsigned long *sto = get_rt_sto(entry);
	184	int sx;
	185
	186	for (sx = 0; sx < ZPCI_TABLE_ENTRIES; sx++)
	187	if (reg_entry_isvalid(sto[sx]))
	188	dma_free_page_table(get_st_pto(sto[sx]));
	189
	190	dma_free_cpu_table(sto);
	191	}
	192
	193	static void dma_cleanup_tables(struct zpci_dev *zdev)
	194	{
bafff174	195	unsigned long *table;
828b35f6 JG	196	int rtx;
	197
	198	if (!zdev \|\| !zdev->dma_table)
	199	return;
	200
bafff174	201	table = zdev->dma_table;
828b35f6 JG	202	for (rtx = 0; rtx < ZPCI_TABLE_ENTRIES; rtx++)
	203	if (reg_entry_isvalid(table[rtx]))
	204	dma_free_seg_table(table[rtx]);
	205
	206	dma_free_cpu_table(table);
	207	zdev->dma_table = NULL;
	208	}
	209
	210	static unsigned long __dma_alloc_iommu(struct zpci_dev *zdev, unsigned long start,
	211	int size)
	212	{
	213	unsigned long boundary_size = 0x1000000;
	214
	215	return iommu_area_alloc(zdev->iommu_bitmap, zdev->iommu_pages,
	216	start, size, 0, boundary_size, 0);
	217	}
	218
	219	static unsigned long dma_alloc_iommu(struct zpci_dev *zdev, int size)
	220	{
	221	unsigned long offset, flags;
	222
	223	spin_lock_irqsave(&zdev->iommu_bitmap_lock, flags);
	224	offset = __dma_alloc_iommu(zdev, zdev->next_bit, size);
	225	if (offset == -1)
	226	offset = __dma_alloc_iommu(zdev, 0, size);
	227
	228	if (offset != -1) {
	229	zdev->next_bit = offset + size;
	230	if (zdev->next_bit >= zdev->iommu_pages)
	231	zdev->next_bit = 0;
	232	}
	233	spin_unlock_irqrestore(&zdev->iommu_bitmap_lock, flags);
	234	return offset;
	235	}
	236
	237	static void dma_free_iommu(struct zpci_dev *zdev, unsigned long offset, int size)
	238	{
	239	unsigned long flags;
	240
	241	spin_lock_irqsave(&zdev->iommu_bitmap_lock, flags);
	242	if (!zdev->iommu_bitmap)
	243	goto out;
	244	bitmap_clear(zdev->iommu_bitmap, offset, size);
	245	if (offset >= zdev->next_bit)
	246	zdev->next_bit = offset + size;
	247	out:
	248	spin_unlock_irqrestore(&zdev->iommu_bitmap_lock, flags);
	249	}
	250
	251	int dma_set_mask(struct device *dev, u64 mask)
	252	{
	253	if (!dev->dma_mask \|\| !dma_supported(dev, mask))
	254	return -EIO;
	255
	256	*dev->dma_mask = mask;
	257	return 0;
	258	}
	259	EXPORT_SYMBOL_GPL(dma_set_mask);
	260
	261	static dma_addr_t s390_dma_map_pages(struct device dev, struct page page,
	262	unsigned long offset, size_t size,
	263	enum dma_data_direction direction,
	264	struct dma_attrs *attrs)
	265	{
266	struct zpci_dev *zdev = get_zdev(container_of(dev, struct pci_dev, dev));
267	unsigned long nr_pages, iommu_page_index;
268	unsigned long pa = page_to_phys(page) + offset;
269	int flags = ZPCI_PTE_VALID;
270	dma_addr_t dma_addr;
271
828b35f6 JG	272	/* This rounds up number of pages based on size and offset */
	273	nr_pages = iommu_num_pages(pa, size, PAGE_SIZE);
	274	iommu_page_index = dma_alloc_iommu(zdev, nr_pages);
	275	if (iommu_page_index == -1)
	276	goto out_err;
	277
	278	/* Use rounded up size */
	279	size = nr_pages * PAGE_SIZE;
	280
	281	dma_addr = zdev->start_dma + iommu_page_index * PAGE_SIZE;
	282	if (dma_addr + size > zdev->end_dma) {
	283	dev_err(dev, "(dma_addr: 0x%16.16LX + size: 0x%16.16lx) > end_dma: 0x%16.16Lx\n",
	284	dma_addr, size, zdev->end_dma);
	285	goto out_free;
	286	}
	287
	288	if (direction == DMA_NONE \|\| direction == DMA_TO_DEVICE)
	289	flags \|= ZPCI_TABLE_PROTECTED;
	290
d0b08853 JG	291	if (!dma_update_trans(zdev, pa, dma_addr, size, flags)) {
d0b08853 JG	292	atomic64_add(nr_pages, (atomic64_t *) &zdev->fmb->mapped_pages);
186f50fa	293	return dma_addr + (offset & ~PAGE_MASK);
d0b08853	294	}
828b35f6 JG	295
	296	out_free:
	297	dma_free_iommu(zdev, iommu_page_index, nr_pages);
	298	out_err:
	299	dev_err(dev, "Failed to map addr: %lx\n", pa);
	300	return DMA_ERROR_CODE;
	301	}
	302
	303	static void s390_dma_unmap_pages(struct device *dev, dma_addr_t dma_addr,
	304	size_t size, enum dma_data_direction direction,
	305	struct dma_attrs *attrs)
	306	{
	307	struct zpci_dev *zdev = get_zdev(container_of(dev, struct pci_dev, dev));
	308	unsigned long iommu_page_index;
	309	int npages;
	310
	311	npages = iommu_num_pages(dma_addr, size, PAGE_SIZE);
	312	dma_addr = dma_addr & PAGE_MASK;
	313	if (dma_update_trans(zdev, 0, dma_addr, npages * PAGE_SIZE,
	314	ZPCI_TABLE_PROTECTED \| ZPCI_PTE_INVALID))
	315	dev_err(dev, "Failed to unmap addr: %Lx\n", dma_addr);
	316
d0b08853	317	atomic64_add(npages, (atomic64_t *) &zdev->fmb->unmapped_pages);
828b35f6 JG	318	iommu_page_index = (dma_addr - zdev->start_dma) >> PAGE_SHIFT;
	319	dma_free_iommu(zdev, iommu_page_index, npages);
	320	}
	321
	322	static void s390_dma_alloc(struct device dev, size_t size,
	323	dma_addr_t *dma_handle, gfp_t flag,
	324	struct dma_attrs *attrs)
	325	{
d0b08853	326	struct zpci_dev *zdev = get_zdev(container_of(dev, struct pci_dev, dev));
828b35f6 JG	327	struct page *page;
	328	unsigned long pa;
	329	dma_addr_t map;
	330
	331	size = PAGE_ALIGN(size);
	332	page = alloc_pages(flag, get_order(size));
	333	if (!page)
	334	return NULL;
d0b08853 JG	335
d0b08853 JG	336	atomic64_add(size / PAGE_SIZE, (atomic64_t *) &zdev->fmb->allocated_pages);
828b35f6 JG	337	pa = page_to_phys(page);
	338	memset((void *) pa, 0, size);
	339
	340	map = s390_dma_map_pages(dev, page, pa % PAGE_SIZE,
	341	size, DMA_BIDIRECTIONAL, NULL);
	342	if (dma_mapping_error(dev, map)) {
	343	free_pages(pa, get_order(size));
	344	return NULL;
	345	}
	346
	347	if (dma_handle)
	348	*dma_handle = map;
	349	return (void *) pa;
	350	}
	351
	352	static void s390_dma_free(struct device *dev, size_t size,
	353	void *pa, dma_addr_t dma_handle,
	354	struct dma_attrs *attrs)
	355	{
	356	s390_dma_unmap_pages(dev, dma_handle, PAGE_ALIGN(size),
	357	DMA_BIDIRECTIONAL, NULL);
	358	free_pages((unsigned long) pa, get_order(size));
	359	}
	360
	361	static int s390_dma_map_sg(struct device dev, struct scatterlist sg,
	362	int nr_elements, enum dma_data_direction dir,
	363	struct dma_attrs *attrs)
	364	{
	365	int mapped_elements = 0;
	366	struct scatterlist *s;
	367	int i;
	368
	369	for_each_sg(sg, s, nr_elements, i) {
	370	struct page *page = sg_page(s);
	371	s->dma_address = s390_dma_map_pages(dev, page, s->offset,
	372	s->length, dir, NULL);
	373	if (!dma_mapping_error(dev, s->dma_address)) {
	374	s->dma_length = s->length;
	375	mapped_elements++;
	376	} else
	377	goto unmap;
	378	}
	379	out:
	380	return mapped_elements;
	381
	382	unmap:
	383	for_each_sg(sg, s, mapped_elements, i) {
	384	if (s->dma_address)
	385	s390_dma_unmap_pages(dev, s->dma_address, s->dma_length,
	386	dir, NULL);
	387	s->dma_address = 0;
	388	s->dma_length = 0;
	389	}
	390	mapped_elements = 0;
	391	goto out;
	392	}
	393
	394	static void s390_dma_unmap_sg(struct device dev, struct scatterlist sg,
	395	int nr_elements, enum dma_data_direction dir,
	396	struct dma_attrs *attrs)
	397	{
	398	struct scatterlist *s;
	399	int i;
	400
401	for_each_sg(sg, s, nr_elements, i) {
402	s390_dma_unmap_pages(dev, s->dma_address, s->dma_length, dir, NULL);
403	s->dma_address = 0;
404	s->dma_length = 0;
405	}
406	}
407
408	int zpci_dma_init_device(struct zpci_dev *zdev)
409	{
410	unsigned int bitmap_order;
411	int rc;
412
413	spin_lock_init(&zdev->iommu_bitmap_lock);
414	spin_lock_init(&zdev->dma_table_lock);
415
416	zdev->dma_table = dma_alloc_cpu_table();
417	if (!zdev->dma_table) {
418	rc = -ENOMEM;
ab6c86a0	419	goto out;
828b35f6 JG	420	}
	421
	422	zdev->iommu_size = (unsigned long) high_memory - PAGE_OFFSET;
	423	zdev->iommu_pages = zdev->iommu_size >> PAGE_SHIFT;
	424	bitmap_order = get_order(zdev->iommu_pages / 8);
	425	pr_info("iommu_size: 0x%lx iommu_pages: 0x%lx bitmap_order: %i\n",
	426	zdev->iommu_size, zdev->iommu_pages, bitmap_order);
	427
	428	zdev->iommu_bitmap = (void *) __get_free_pages(GFP_KERNEL \| __GFP_ZERO,
	429	bitmap_order);
	430	if (!zdev->iommu_bitmap) {
	431	rc = -ENOMEM;
ab6c86a0	432	goto free_dma_table;
828b35f6 JG	433	}
	434
	435	rc = zpci_register_ioat(zdev,
	436	0,
	437	zdev->start_dma + PAGE_OFFSET,
	438	zdev->start_dma + zdev->iommu_size - 1,
	439	(u64) zdev->dma_table);
	440	if (rc)
ab6c86a0	441	goto free_bitmap;
828b35f6	442
ab6c86a0 SO	443	return 0;
	444	free_bitmap:
	445	vfree(zdev->iommu_bitmap);
	446	zdev->iommu_bitmap = NULL;
	447	free_dma_table:
828b35f6	448	dma_free_cpu_table(zdev->dma_table);
ab6c86a0 SO	449	zdev->dma_table = NULL;
ab6c86a0 SO	450	out:
828b35f6 JG	451	return rc;
	452	}
	453
	454	void zpci_dma_exit_device(struct zpci_dev *zdev)
	455	{
	456	zpci_unregister_ioat(zdev, 0);
	457	dma_cleanup_tables(zdev);
	458	free_pages((unsigned long) zdev->iommu_bitmap,
	459	get_order(zdev->iommu_pages / 8));
	460	zdev->iommu_bitmap = NULL;
	461	zdev->next_bit = 0;
	462	}
	463
	464	static int __init dma_alloc_cpu_table_caches(void)
	465	{
	466	dma_region_table_cache = kmem_cache_create("PCI_DMA_region_tables",
	467	ZPCI_TABLE_SIZE, ZPCI_TABLE_ALIGN,
	468	0, NULL);
	469	if (!dma_region_table_cache)
	470	return -ENOMEM;
	471
	472	dma_page_table_cache = kmem_cache_create("PCI_DMA_page_tables",
	473	ZPCI_PT_SIZE, ZPCI_PT_ALIGN,
	474	0, NULL);
	475	if (!dma_page_table_cache) {
	476	kmem_cache_destroy(dma_region_table_cache);
	477	return -ENOMEM;
	478	}
	479	return 0;
	480	}
	481
	482	int __init zpci_dma_init(void)
	483	{
	484	return dma_alloc_cpu_table_caches();
	485	}
	486
	487	void zpci_dma_exit(void)
	488	{
	489	kmem_cache_destroy(dma_page_table_cache);
	490	kmem_cache_destroy(dma_region_table_cache);
	491	}
	492
	493	#define PREALLOC_DMA_DEBUG_ENTRIES (1 << 16)
	494
	495	static int __init dma_debug_do_init(void)
	496	{
	497	dma_debug_init(PREALLOC_DMA_DEBUG_ENTRIES);
	498	return 0;
	499	}
	500	fs_initcall(dma_debug_do_init);
	501
	502	struct dma_map_ops s390_dma_ops = {
	503	.alloc = s390_dma_alloc,
	504	.free = s390_dma_free,
	505	.map_sg = s390_dma_map_sg,
	506	.unmap_sg = s390_dma_unmap_sg,
	507	.map_page = s390_dma_map_pages,
	508	.unmap_page = s390_dma_unmap_pages,
	509	/* if we support direct DMA this must be conditional */
	510	.is_phys = 0,
	511	/* dma_supported is unconditionally true without a callback */
	512	};
	513	EXPORT_SYMBOL_GPL(s390_dma_ops);