[deliverable/linux.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 enum zpci_ioat_dtype zpci_ioat_dt = ZPCI_IOTA_RTTO;

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 = rpcit_instr((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 = zdev->dma_table;
	int rtx;

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

	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;

	WARN_ON_ONCE(offset > PAGE_SIZE);

	/* 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))
		return dma_addr + offset;

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);

	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 page *page;
	unsigned long pa;
	dma_addr_t map;

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

	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 out_reg;
	}

	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 out_reg;
	return 0;

out_reg:
	dma_free_cpu_table(zdev->dma_table);
out_clean:
	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
	16	static enum zpci_ioat_dtype zpci_ioat_dt = ZPCI_IOTA_RTTO;
	17
	18	static struct kmem_cache *dma_region_table_cache;
	19	static struct kmem_cache *dma_page_table_cache;
	20
	21	static unsigned long *dma_alloc_cpu_table(void)
	22	{
	23	unsigned long table, entry;
	24
	25	table = kmem_cache_alloc(dma_region_table_cache, GFP_ATOMIC);
	26	if (!table)
	27	return NULL;
	28
	29	for (entry = table; entry < table + ZPCI_TABLE_ENTRIES; entry++)
	30	*entry = ZPCI_TABLE_INVALID \| ZPCI_TABLE_PROTECTED;
	31	return table;
	32	}
	33
	34	static void dma_free_cpu_table(void *table)
	35	{
	36	kmem_cache_free(dma_region_table_cache, table);
	37	}
	38
	39	static unsigned long *dma_alloc_page_table(void)
	40	{
	41	unsigned long table, entry;
	42
	43	table = kmem_cache_alloc(dma_page_table_cache, GFP_ATOMIC);
	44	if (!table)
	45	return NULL;
	46
	47	for (entry = table; entry < table + ZPCI_PT_ENTRIES; entry++)
	48	*entry = ZPCI_PTE_INVALID \| ZPCI_TABLE_PROTECTED;
	49	return table;
	50	}
	51
	52	static void dma_free_page_table(void *table)
	53	{
	54	kmem_cache_free(dma_page_table_cache, table);
	55	}
	56
	57	static unsigned long dma_get_seg_table_origin(unsigned long entry)
	58	{
	59	unsigned long *sto;
	60
	61	if (reg_entry_isvalid(*entry))
	62	sto = get_rt_sto(*entry);
	63	else {
	64	sto = dma_alloc_cpu_table();
65	if (!sto)
66	return NULL;
67
68	set_rt_sto(entry, sto);
69	validate_rt_entry(entry);
70	entry_clr_protected(entry);
71	}
72	return sto;
73	}
74
75	static unsigned long dma_get_page_table_origin(unsigned long entry)
76	{
77	unsigned long *pto;
78
79	if (reg_entry_isvalid(*entry))
80	pto = get_st_pto(*entry);
81	else {
82	pto = dma_alloc_page_table();
83	if (!pto)
84	return NULL;
85	set_st_pto(entry, pto);
86	validate_st_entry(entry);
87	entry_clr_protected(entry);
88	}
89	return pto;
90	}
91
92	static unsigned long dma_walk_cpu_trans(unsigned long rto, dma_addr_t dma_addr)
93	{
94	unsigned long sto, pto;
95	unsigned int rtx, sx, px;
96
97	rtx = calc_rtx(dma_addr);
98	sto = dma_get_seg_table_origin(&rto[rtx]);
99	if (!sto)
100	return NULL;
101
102	sx = calc_sx(dma_addr);
103	pto = dma_get_page_table_origin(&sto[sx]);
104	if (!pto)
105	return NULL;
106
107	px = calc_px(dma_addr);
108	return &pto[px];
109	}
110
111	static void dma_update_cpu_trans(struct zpci_dev zdev, void page_addr,
112	dma_addr_t dma_addr, int flags)
113	{
114	unsigned long *entry;
115
116	entry = dma_walk_cpu_trans(zdev->dma_table, dma_addr);
117	if (!entry) {
118	WARN_ON_ONCE(1);
119	return;
120	}
121
122	if (flags & ZPCI_PTE_INVALID) {
123	invalidate_pt_entry(entry);
124	return;
125	} else {
126	set_pt_pfaa(entry, page_addr);
127	validate_pt_entry(entry);
128	}
129
130	if (flags & ZPCI_TABLE_PROTECTED)
131	entry_set_protected(entry);
132	else
133	entry_clr_protected(entry);
134	}
135
136	static int dma_update_trans(struct zpci_dev *zdev, unsigned long pa,
137	dma_addr_t dma_addr, size_t size, int flags)
138	{
139	unsigned int nr_pages = PAGE_ALIGN(size) >> PAGE_SHIFT;
140	u8 page_addr = (u8 ) (pa & PAGE_MASK);
141	dma_addr_t start_dma_addr = dma_addr;
142	unsigned long irq_flags;
143	int i, rc = 0;
144
145	if (!nr_pages)
146	return -EINVAL;
147
148	spin_lock_irqsave(&zdev->dma_table_lock, irq_flags);
149	if (!zdev->dma_table) {
150	dev_err(&zdev->pdev->dev, "Missing DMA table\n");
151	goto no_refresh;
152	}
153
154	for (i = 0; i < nr_pages; i++) {
155	dma_update_cpu_trans(zdev, page_addr, dma_addr, flags);
156	page_addr += PAGE_SIZE;
157	dma_addr += PAGE_SIZE;
158	}
159
160	/*
161	* rpcit is not required to establish new translations when previously
162	* invalid translation-table entries are validated, however it is
163	* required when altering previously valid entries.
164	*/
165	if (!zdev->tlb_refresh &&
166	((flags & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID))
167	/*
168	* TODO: also need to check that the old entry is indeed INVALID
169	* and not only for one page but for the whole range...
170	* -> now we WARN_ON in that case but with lazy unmap that
171	* needs to be redone!
172	*/
173	goto no_refresh;
174	rc = rpcit_instr((u64) zdev->fh << 32, start_dma_addr,
175	nr_pages * PAGE_SIZE);
176
177	no_refresh:
178	spin_unlock_irqrestore(&zdev->dma_table_lock, irq_flags);
179	return rc;
180	}
181
182	static void dma_free_seg_table(unsigned long entry)
183	{
184	unsigned long *sto = get_rt_sto(entry);
185	int sx;
186
187	for (sx = 0; sx < ZPCI_TABLE_ENTRIES; sx++)
188	if (reg_entry_isvalid(sto[sx]))
189	dma_free_page_table(get_st_pto(sto[sx]));
190
191	dma_free_cpu_table(sto);
192	}
193
194	static void dma_cleanup_tables(struct zpci_dev *zdev)
195	{
196	unsigned long *table = zdev->dma_table;
197	int rtx;
198
199	if (!zdev \|\| !zdev->dma_table)
200	return;
201
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
272	WARN_ON_ONCE(offset > PAGE_SIZE);
273
274	/* This rounds up number of pages based on size and offset */
275	nr_pages = iommu_num_pages(pa, size, PAGE_SIZE);
276	iommu_page_index = dma_alloc_iommu(zdev, nr_pages);
277	if (iommu_page_index == -1)
278	goto out_err;
279
280	/* Use rounded up size */
281	size = nr_pages * PAGE_SIZE;
282
283	dma_addr = zdev->start_dma + iommu_page_index * PAGE_SIZE;
284	if (dma_addr + size > zdev->end_dma) {
285	dev_err(dev, "(dma_addr: 0x%16.16LX + size: 0x%16.16lx) > end_dma: 0x%16.16Lx\n",
286	dma_addr, size, zdev->end_dma);
287	goto out_free;
288	}
289
290	if (direction == DMA_NONE \|\| direction == DMA_TO_DEVICE)
291	flags \|= ZPCI_TABLE_PROTECTED;
292
293	if (!dma_update_trans(zdev, pa, dma_addr, size, flags))
294	return dma_addr + offset;
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
317	iommu_page_index = (dma_addr - zdev->start_dma) >> PAGE_SHIFT;
318	dma_free_iommu(zdev, iommu_page_index, npages);
319	}
320
321	static void s390_dma_alloc(struct device dev, size_t size,
322	dma_addr_t *dma_handle, gfp_t flag,
323	struct dma_attrs *attrs)
324	{
325	struct page *page;
326	unsigned long pa;
327	dma_addr_t map;
328
329	size = PAGE_ALIGN(size);
330	page = alloc_pages(flag, get_order(size));
331	if (!page)
332	return NULL;
333	pa = page_to_phys(page);
334	memset((void *) pa, 0, size);
335
336	map = s390_dma_map_pages(dev, page, pa % PAGE_SIZE,
337	size, DMA_BIDIRECTIONAL, NULL);
338	if (dma_mapping_error(dev, map)) {
339	free_pages(pa, get_order(size));
340	return NULL;
341	}
342
343	if (dma_handle)
344	*dma_handle = map;
345	return (void *) pa;
346	}
347
348	static void s390_dma_free(struct device *dev, size_t size,
349	void *pa, dma_addr_t dma_handle,
350	struct dma_attrs *attrs)
351	{
352	s390_dma_unmap_pages(dev, dma_handle, PAGE_ALIGN(size),
353	DMA_BIDIRECTIONAL, NULL);
354	free_pages((unsigned long) pa, get_order(size));
355	}
356
357	static int s390_dma_map_sg(struct device dev, struct scatterlist sg,
358	int nr_elements, enum dma_data_direction dir,
359	struct dma_attrs *attrs)
360	{
361	int mapped_elements = 0;
362	struct scatterlist *s;
363	int i;
364
365	for_each_sg(sg, s, nr_elements, i) {
366	struct page *page = sg_page(s);
367	s->dma_address = s390_dma_map_pages(dev, page, s->offset,
368	s->length, dir, NULL);
369	if (!dma_mapping_error(dev, s->dma_address)) {
370	s->dma_length = s->length;
371	mapped_elements++;
372	} else
373	goto unmap;
374	}
375	out:
376	return mapped_elements;
377
378	unmap:
379	for_each_sg(sg, s, mapped_elements, i) {
380	if (s->dma_address)
381	s390_dma_unmap_pages(dev, s->dma_address, s->dma_length,
382	dir, NULL);
383	s->dma_address = 0;
384	s->dma_length = 0;
385	}
386	mapped_elements = 0;
387	goto out;
388	}
389
390	static void s390_dma_unmap_sg(struct device dev, struct scatterlist sg,
391	int nr_elements, enum dma_data_direction dir,
392	struct dma_attrs *attrs)
393	{
394	struct scatterlist *s;
395	int i;
396
397	for_each_sg(sg, s, nr_elements, i) {
398	s390_dma_unmap_pages(dev, s->dma_address, s->dma_length, dir, NULL);
399	s->dma_address = 0;
400	s->dma_length = 0;
401	}
402	}
403
404	int zpci_dma_init_device(struct zpci_dev *zdev)
405	{
406	unsigned int bitmap_order;
407	int rc;
408
409	spin_lock_init(&zdev->iommu_bitmap_lock);
410	spin_lock_init(&zdev->dma_table_lock);
411
412	zdev->dma_table = dma_alloc_cpu_table();
413	if (!zdev->dma_table) {
414	rc = -ENOMEM;
415	goto out_clean;
416	}
417
418	zdev->iommu_size = (unsigned long) high_memory - PAGE_OFFSET;
419	zdev->iommu_pages = zdev->iommu_size >> PAGE_SHIFT;
420	bitmap_order = get_order(zdev->iommu_pages / 8);
421	pr_info("iommu_size: 0x%lx iommu_pages: 0x%lx bitmap_order: %i\n",
422	zdev->iommu_size, zdev->iommu_pages, bitmap_order);
423
424	zdev->iommu_bitmap = (void *) __get_free_pages(GFP_KERNEL \| __GFP_ZERO,
425	bitmap_order);
426	if (!zdev->iommu_bitmap) {
427	rc = -ENOMEM;
428	goto out_reg;
429	}
430
431	rc = zpci_register_ioat(zdev,
432	0,
433	zdev->start_dma + PAGE_OFFSET,
434	zdev->start_dma + zdev->iommu_size - 1,
435	(u64) zdev->dma_table);
436	if (rc)
437	goto out_reg;
438	return 0;
439
440	out_reg:
441	dma_free_cpu_table(zdev->dma_table);
442	out_clean:
443	return rc;
444	}
445
446	void zpci_dma_exit_device(struct zpci_dev *zdev)
447	{
448	zpci_unregister_ioat(zdev, 0);
449	dma_cleanup_tables(zdev);
450	free_pages((unsigned long) zdev->iommu_bitmap,
451	get_order(zdev->iommu_pages / 8));
452	zdev->iommu_bitmap = NULL;
453	zdev->next_bit = 0;
454	}
455
456	static int __init dma_alloc_cpu_table_caches(void)
457	{
458	dma_region_table_cache = kmem_cache_create("PCI_DMA_region_tables",
459	ZPCI_TABLE_SIZE, ZPCI_TABLE_ALIGN,
460	0, NULL);
461	if (!dma_region_table_cache)
462	return -ENOMEM;
463
464	dma_page_table_cache = kmem_cache_create("PCI_DMA_page_tables",
465	ZPCI_PT_SIZE, ZPCI_PT_ALIGN,
466	0, NULL);
467	if (!dma_page_table_cache) {
468	kmem_cache_destroy(dma_region_table_cache);
469	return -ENOMEM;
470	}
471	return 0;
472	}
473
474	int __init zpci_dma_init(void)
475	{
476	return dma_alloc_cpu_table_caches();
477	}
478
479	void zpci_dma_exit(void)
480	{
481	kmem_cache_destroy(dma_page_table_cache);
482	kmem_cache_destroy(dma_region_table_cache);
483	}
484
485	#define PREALLOC_DMA_DEBUG_ENTRIES (1 << 16)
486
487	static int __init dma_debug_do_init(void)
488	{
489	dma_debug_init(PREALLOC_DMA_DEBUG_ENTRIES);
490	return 0;
491	}
492	fs_initcall(dma_debug_do_init);
493
494	struct dma_map_ops s390_dma_ops = {
495	.alloc = s390_dma_alloc,
496	.free = s390_dma_free,
497	.map_sg = s390_dma_map_sg,
498	.unmap_sg = s390_dma_unmap_sg,
499	.map_page = s390_dma_map_pages,
500	.unmap_page = s390_dma_unmap_pages,
501	/* if we support direct DMA this must be conditional */
502	.is_phys = 0,
503	/* dma_supported is unconditionally true without a callback */
504	};
505	EXPORT_SYMBOL_GPL(s390_dma_ops);