[linux-2.6-block.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/vmalloc.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 int s390_iommu_strict;

static int zpci_refresh_global(struct zpci_dev *zdev)
{
	return zpci_refresh_trans((u64) zdev->fh << 32, zdev->start_dma,
				  zdev->iommu_pages * PAGE_SIZE);
}

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;
	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;
	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;
}

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];
}

void dma_update_cpu_trans(unsigned long *entry, void *page_addr, int flags)
{
	if (flags & ZPCI_PTE_INVALID) {
		invalidate_pt_entry(entry);
	} 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;
	unsigned long *entry;
	int i, rc = 0;

	if (!nr_pages)
		return -EINVAL;

	spin_lock_irqsave(&zdev->dma_table_lock, irq_flags);
	if (!zdev->dma_table) {
		rc = -EINVAL;
		goto no_refresh;
	}

	for (i = 0; i < nr_pages; i++) {
		entry = dma_walk_cpu_trans(zdev->dma_table, dma_addr);
		if (!entry) {
			rc = -ENOMEM;
			goto undo_cpu_trans;
		}
		dma_update_cpu_trans(entry, page_addr, flags);
		page_addr += PAGE_SIZE;
		dma_addr += PAGE_SIZE;
	}

	/*
	 * With zdev->tlb_refresh == 0, rpcit is not required to establish new
	 * translations when previously invalid translation-table entries are
	 * validated. With lazy unmap, it also is skipped for previously valid
	 * entries, but a global rpcit is then required before any address can
	 * be re-used, i.e. after each iommu bitmap wrap-around.
	 */
	if (!zdev->tlb_refresh &&
			(!s390_iommu_strict ||
			((flags & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID)))
		goto no_refresh;

	rc = zpci_refresh_trans((u64) zdev->fh << 32, start_dma_addr,
				nr_pages * PAGE_SIZE);
undo_cpu_trans:
	if (rc && ((flags & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID)) {
		flags = ZPCI_PTE_INVALID;
		while (i-- > 0) {
			page_addr -= PAGE_SIZE;
			dma_addr -= PAGE_SIZE;
			entry = dma_walk_cpu_trans(zdev->dma_table, dma_addr);
			if (!entry)
				break;
			dma_update_cpu_trans(entry, page_addr, flags);
		}
	}

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

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);
}

void dma_cleanup_tables(unsigned long *table)
{
	int rtx;

	if (!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);
}

static unsigned long __dma_alloc_iommu(struct device *dev,
				       unsigned long start, int size)
{
	struct zpci_dev *zdev = to_zpci(to_pci_dev(dev));
	unsigned long boundary_size;

	boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
			      PAGE_SIZE) >> PAGE_SHIFT;
	return iommu_area_alloc(zdev->iommu_bitmap, zdev->iommu_pages,
				start, size, 0, boundary_size, 0);
}

static unsigned long dma_alloc_iommu(struct device *dev, int size)
{
	struct zpci_dev *zdev = to_zpci(to_pci_dev(dev));
	unsigned long offset, flags;
	int wrap = 0;

	spin_lock_irqsave(&zdev->iommu_bitmap_lock, flags);
	offset = __dma_alloc_iommu(dev, zdev->next_bit, size);
	if (offset == -1) {
		/* wrap-around */
		offset = __dma_alloc_iommu(dev, 0, size);
		wrap = 1;
	}

	if (offset != -1) {
		zdev->next_bit = offset + size;
		if (!zdev->tlb_refresh && !s390_iommu_strict && wrap)
			/* global flush after wrap-around with lazy unmap */
			zpci_refresh_global(zdev);
	}
	spin_unlock_irqrestore(&zdev->iommu_bitmap_lock, flags);
	return offset;
}

static void dma_free_iommu(struct device *dev, unsigned long offset, int size)
{
	struct zpci_dev *zdev = to_zpci(to_pci_dev(dev));
	unsigned long flags;

	spin_lock_irqsave(&zdev->iommu_bitmap_lock, flags);
	if (!zdev->iommu_bitmap)
		goto out;
	bitmap_clear(zdev->iommu_bitmap, offset, size);
	/*
	 * Lazy flush for unmap: need to move next_bit to avoid address re-use
	 * until wrap-around.
	 */
	if (!s390_iommu_strict && offset >= zdev->next_bit)
		zdev->next_bit = offset + size;
out:
	spin_unlock_irqrestore(&zdev->iommu_bitmap_lock, flags);
}

static inline void zpci_err_dma(unsigned long rc, unsigned long addr)
{
	struct {
		unsigned long rc;
		unsigned long addr;
	} __packed data = {rc, addr};

	zpci_err_hex(&data, sizeof(data));
}

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 = to_zpci(to_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;
	int ret;

	/* 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(dev, nr_pages);
	if (iommu_page_index == -1) {
		ret = -ENOSPC;
		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) {
		ret = -ERANGE;
		goto out_free;
	}

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

	ret = dma_update_trans(zdev, pa, dma_addr, size, flags);
	if (ret)
		goto out_free;

	atomic64_add(nr_pages, &zdev->mapped_pages);
	return dma_addr + (offset & ~PAGE_MASK);

out_free:
	dma_free_iommu(dev, iommu_page_index, nr_pages);
out_err:
	zpci_err("map error:\n");
	zpci_err_dma(ret, 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 = to_zpci(to_pci_dev(dev));
	unsigned long iommu_page_index;
	int npages, ret;

	npages = iommu_num_pages(dma_addr, size, PAGE_SIZE);
	dma_addr = dma_addr & PAGE_MASK;
	ret = dma_update_trans(zdev, 0, dma_addr, npages * PAGE_SIZE,
			       ZPCI_PTE_INVALID);
	if (ret) {
		zpci_err("unmap error:\n");
		zpci_err_dma(ret, dma_addr);
		return;
	}

	atomic64_add(npages, &zdev->unmapped_pages);
	iommu_page_index = (dma_addr - zdev->start_dma) >> PAGE_SHIFT;
	dma_free_iommu(dev, 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 = to_zpci(to_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;

	pa = page_to_phys(page);
	memset((void *) pa, 0, size);

	map = s390_dma_map_pages(dev, page, 0, size, DMA_BIDIRECTIONAL, NULL);
	if (dma_mapping_error(dev, map)) {
		free_pages(pa, get_order(size));
		return NULL;
	}

	atomic64_add(size / PAGE_SIZE, &zdev->allocated_pages);
	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)
{
	struct zpci_dev *zdev = to_zpci(to_pci_dev(dev));

	size = PAGE_ALIGN(size);
	atomic64_sub(size / PAGE_SIZE, &zdev->allocated_pages);
	s390_dma_unmap_pages(dev, dma_handle, 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)
{
	int rc;

	/*
	 * At this point, if the device is part of an IOMMU domain, this would
	 * be a strong hint towards a bug in the IOMMU API (common) code and/or
	 * simultaneous access via IOMMU and DMA API. So let's issue a warning.
	 */
	WARN_ON(zdev->s390_domain);

	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;
	}

	/*
	 * Restrict the iommu bitmap size to the minimum of the following:
	 * - main memory size
	 * - 3-level pagetable address limit minus start_dma offset
	 * - DMA address range allowed by the hardware (clp query pci fn)
	 *
	 * Also set zdev->end_dma to the actual end address of the usable
	 * range, instead of the theoretical maximum as reported by hardware.
	 */
	zdev->iommu_size = min3((u64) high_memory,
				ZPCI_TABLE_SIZE_RT - zdev->start_dma,
				zdev->end_dma - zdev->start_dma + 1);
	zdev->end_dma = zdev->start_dma + zdev->iommu_size - 1;
	zdev->iommu_pages = zdev->iommu_size >> PAGE_SHIFT;
	zdev->iommu_bitmap = vzalloc(zdev->iommu_pages / 8);
	if (!zdev->iommu_bitmap) {
		rc = -ENOMEM;
		goto free_dma_table;
	}

	rc = zpci_register_ioat(zdev, 0, zdev->start_dma, zdev->end_dma,
				(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)
{
	/*
	 * At this point, if the device is part of an IOMMU domain, this would
	 * be a strong hint towards a bug in the IOMMU API (common) code and/or
	 * simultaneous access via IOMMU and DMA API. So let's issue a warning.
	 */
	WARN_ON(zdev->s390_domain);

	zpci_unregister_ioat(zdev, 0);
	dma_cleanup_tables(zdev->dma_table);
	zdev->dma_table = NULL;
	vfree(zdev->iommu_bitmap);
	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_pci_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_pci_dma_ops);

static int __init s390_iommu_setup(char *str)
{
	if (!strncmp(str, "strict", 6))
		s390_iommu_strict = 1;
	return 0;
}

__setup("s390_iommu=", s390_iommu_setup);
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>
22459321	13	#include <linux/vmalloc.h>
828b35f6 JG	14	#include <linux/pci.h>
	15	#include <asm/pci_dma.h>
	16
828b35f6 JG	17	static struct kmem_cache *dma_region_table_cache;
828b35f6 JG	18	static struct kmem_cache *dma_page_table_cache;
c60d1ae4 GS	19	static int s390_iommu_strict;
	20
	21	static int zpci_refresh_global(struct zpci_dev *zdev)
	22	{
	23	return zpci_refresh_trans((u64) zdev->fh << 32, zdev->start_dma,
	24	zdev->iommu_pages * PAGE_SIZE);
	25	}
828b35f6	26
8128f23c	27	unsigned long *dma_alloc_cpu_table(void)
828b35f6 JG	28	{
	29	unsigned long table, entry;
	30
	31	table = kmem_cache_alloc(dma_region_table_cache, GFP_ATOMIC);
	32	if (!table)
	33	return NULL;
	34
	35	for (entry = table; entry < table + ZPCI_TABLE_ENTRIES; entry++)
4d5a6b72	36	*entry = ZPCI_TABLE_INVALID;
828b35f6 JG	37	return table;
	38	}
	39
	40	static void dma_free_cpu_table(void *table)
	41	{
	42	kmem_cache_free(dma_region_table_cache, table);
	43	}
	44
	45	static unsigned long *dma_alloc_page_table(void)
	46	{
	47	unsigned long table, entry;
	48
	49	table = kmem_cache_alloc(dma_page_table_cache, GFP_ATOMIC);
	50	if (!table)
	51	return NULL;
	52
	53	for (entry = table; entry < table + ZPCI_PT_ENTRIES; entry++)
4d5a6b72	54	*entry = ZPCI_PTE_INVALID;
828b35f6 JG	55	return table;
	56	}
	57
	58	static void dma_free_page_table(void *table)
	59	{
	60	kmem_cache_free(dma_page_table_cache, table);
	61	}
	62
	63	static unsigned long dma_get_seg_table_origin(unsigned long entry)
	64	{
	65	unsigned long *sto;
	66
	67	if (reg_entry_isvalid(*entry))
	68	sto = get_rt_sto(*entry);
	69	else {
	70	sto = dma_alloc_cpu_table();
	71	if (!sto)
	72	return NULL;
	73
	74	set_rt_sto(entry, sto);
	75	validate_rt_entry(entry);
	76	entry_clr_protected(entry);
	77	}
	78	return sto;
	79	}
	80
	81	static unsigned long dma_get_page_table_origin(unsigned long entry)
	82	{
	83	unsigned long *pto;
	84
	85	if (reg_entry_isvalid(*entry))
	86	pto = get_st_pto(*entry);
	87	else {
	88	pto = dma_alloc_page_table();
	89	if (!pto)
	90	return NULL;
	91	set_st_pto(entry, pto);
	92	validate_st_entry(entry);
	93	entry_clr_protected(entry);
	94	}
	95	return pto;
	96	}
	97
66728eee	98	unsigned long dma_walk_cpu_trans(unsigned long rto, dma_addr_t dma_addr)
828b35f6 JG	99	{
	100	unsigned long sto, pto;
	101	unsigned int rtx, sx, px;
	102
	103	rtx = calc_rtx(dma_addr);
	104	sto = dma_get_seg_table_origin(&rto[rtx]);
	105	if (!sto)
	106	return NULL;
	107
	108	sx = calc_sx(dma_addr);
	109	pto = dma_get_page_table_origin(&sto[sx]);
	110	if (!pto)
	111	return NULL;
	112
	113	px = calc_px(dma_addr);
	114	return &pto[px];
	115	}
	116
66728eee	117	void dma_update_cpu_trans(unsigned long entry, void page_addr, int flags)
828b35f6	118	{
828b35f6 JG	119	if (flags & ZPCI_PTE_INVALID) {
828b35f6 JG	120	invalidate_pt_entry(entry);
828b35f6 JG	121	} else {
	122	set_pt_pfaa(entry, page_addr);
	123	validate_pt_entry(entry);
	124	}
	125
	126	if (flags & ZPCI_TABLE_PROTECTED)
	127	entry_set_protected(entry);
	128	else
	129	entry_clr_protected(entry);
	130	}
	131
	132	static int dma_update_trans(struct zpci_dev *zdev, unsigned long pa,
	133	dma_addr_t dma_addr, size_t size, int flags)
	134	{
	135	unsigned int nr_pages = PAGE_ALIGN(size) >> PAGE_SHIFT;
	136	u8 page_addr = (u8 ) (pa & PAGE_MASK);
	137	dma_addr_t start_dma_addr = dma_addr;
	138	unsigned long irq_flags;
66728eee	139	unsigned long *entry;
828b35f6 JG	140	int i, rc = 0;
	141
	142	if (!nr_pages)
	143	return -EINVAL;
	144
	145	spin_lock_irqsave(&zdev->dma_table_lock, irq_flags);
66728eee SO	146	if (!zdev->dma_table) {
66728eee SO	147	rc = -EINVAL;
828b35f6	148	goto no_refresh;
66728eee	149	}
828b35f6 JG	150
828b35f6 JG	151	for (i = 0; i < nr_pages; i++) {
66728eee SO	152	entry = dma_walk_cpu_trans(zdev->dma_table, dma_addr);
	153	if (!entry) {
	154	rc = -ENOMEM;
	155	goto undo_cpu_trans;
	156	}
	157	dma_update_cpu_trans(entry, page_addr, flags);
828b35f6 JG	158	page_addr += PAGE_SIZE;
	159	dma_addr += PAGE_SIZE;
	160	}
	161
	162	/*
c60d1ae4 GS	163	* With zdev->tlb_refresh == 0, rpcit is not required to establish new
	164	* translations when previously invalid translation-table entries are
	165	* validated. With lazy unmap, it also is skipped for previously valid
	166	* entries, but a global rpcit is then required before any address can
	167	* be re-used, i.e. after each iommu bitmap wrap-around.
828b35f6 JG	168	*/
828b35f6 JG	169	if (!zdev->tlb_refresh &&
c60d1ae4 GS	170	(!s390_iommu_strict \|\|
c60d1ae4 GS	171	((flags & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID)))
828b35f6	172	goto no_refresh;
b2a9e87d	173
9389339f MS	174	rc = zpci_refresh_trans((u64) zdev->fh << 32, start_dma_addr,
9389339f MS	175	nr_pages * PAGE_SIZE);
66728eee SO	176	undo_cpu_trans:
	177	if (rc && ((flags & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID)) {
	178	flags = ZPCI_PTE_INVALID;
	179	while (i-- > 0) {
	180	page_addr -= PAGE_SIZE;
	181	dma_addr -= PAGE_SIZE;
	182	entry = dma_walk_cpu_trans(zdev->dma_table, dma_addr);
	183	if (!entry)
	184	break;
	185	dma_update_cpu_trans(entry, page_addr, flags);
	186	}
	187	}
828b35f6 JG	188
	189	no_refresh:
	190	spin_unlock_irqrestore(&zdev->dma_table_lock, irq_flags);
	191	return rc;
	192	}
	193
8128f23c	194	void dma_free_seg_table(unsigned long entry)
828b35f6 JG	195	{
	196	unsigned long *sto = get_rt_sto(entry);
	197	int sx;
	198
	199	for (sx = 0; sx < ZPCI_TABLE_ENTRIES; sx++)
	200	if (reg_entry_isvalid(sto[sx]))
	201	dma_free_page_table(get_st_pto(sto[sx]));
	202
	203	dma_free_cpu_table(sto);
	204	}
	205
8128f23c	206	void dma_cleanup_tables(unsigned long *table)
828b35f6	207	{
828b35f6 JG	208	int rtx;
828b35f6 JG	209
8128f23c	210	if (!table)
828b35f6 JG	211	return;
	212
	213	for (rtx = 0; rtx < ZPCI_TABLE_ENTRIES; rtx++)
	214	if (reg_entry_isvalid(table[rtx]))
	215	dma_free_seg_table(table[rtx]);
	216
	217	dma_free_cpu_table(table);
828b35f6 JG	218	}
828b35f6 JG	219
9a99649f	220	static unsigned long __dma_alloc_iommu(struct device *dev,
5ec6d491	221	unsigned long start, int size)
828b35f6	222	{
9a99649f	223	struct zpci_dev *zdev = to_zpci(to_pci_dev(dev));
5ec6d491	224	unsigned long boundary_size;
828b35f6	225
9a99649f	226	boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
5ec6d491	227	PAGE_SIZE) >> PAGE_SHIFT;
828b35f6 JG	228	return iommu_area_alloc(zdev->iommu_bitmap, zdev->iommu_pages,
	229	start, size, 0, boundary_size, 0);
	230	}
	231
9a99649f	232	static unsigned long dma_alloc_iommu(struct device *dev, int size)
828b35f6	233	{
9a99649f	234	struct zpci_dev *zdev = to_zpci(to_pci_dev(dev));
828b35f6	235	unsigned long offset, flags;
c60d1ae4	236	int wrap = 0;
828b35f6 JG	237
828b35f6 JG	238	spin_lock_irqsave(&zdev->iommu_bitmap_lock, flags);
9a99649f	239	offset = __dma_alloc_iommu(dev, zdev->next_bit, size);
c60d1ae4 GS	240	if (offset == -1) {
c60d1ae4 GS	241	/* wrap-around */
9a99649f	242	offset = __dma_alloc_iommu(dev, 0, size);
c60d1ae4 GS	243	wrap = 1;
c60d1ae4 GS	244	}
828b35f6 JG	245
	246	if (offset != -1) {
	247	zdev->next_bit = offset + size;
c60d1ae4 GS	248	if (!zdev->tlb_refresh && !s390_iommu_strict && wrap)
	249	/* global flush after wrap-around with lazy unmap */
	250	zpci_refresh_global(zdev);
828b35f6 JG	251	}
	252	spin_unlock_irqrestore(&zdev->iommu_bitmap_lock, flags);
	253	return offset;
	254	}
	255
9a99649f	256	static void dma_free_iommu(struct device *dev, unsigned long offset, int size)
828b35f6	257	{
9a99649f	258	struct zpci_dev *zdev = to_zpci(to_pci_dev(dev));
828b35f6 JG	259	unsigned long flags;
	260
	261	spin_lock_irqsave(&zdev->iommu_bitmap_lock, flags);
	262	if (!zdev->iommu_bitmap)
	263	goto out;
	264	bitmap_clear(zdev->iommu_bitmap, offset, size);
c60d1ae4 GS	265	/*
	266	* Lazy flush for unmap: need to move next_bit to avoid address re-use
	267	* until wrap-around.
	268	*/
	269	if (!s390_iommu_strict && offset >= zdev->next_bit)
828b35f6 JG	270	zdev->next_bit = offset + size;
	271	out:
	272	spin_unlock_irqrestore(&zdev->iommu_bitmap_lock, flags);
	273	}
	274
52d43d81 SO	275	static inline void zpci_err_dma(unsigned long rc, unsigned long addr)
	276	{
	277	struct {
	278	unsigned long rc;
	279	unsigned long addr;
	280	} __packed data = {rc, addr};
	281
	282	zpci_err_hex(&data, sizeof(data));
	283	}
	284
828b35f6 JG	285	static dma_addr_t s390_dma_map_pages(struct device dev, struct page page,
	286	unsigned long offset, size_t size,
	287	enum dma_data_direction direction,
	288	struct dma_attrs *attrs)
	289	{
198a5278	290	struct zpci_dev *zdev = to_zpci(to_pci_dev(dev));
828b35f6 JG	291	unsigned long nr_pages, iommu_page_index;
	292	unsigned long pa = page_to_phys(page) + offset;
	293	int flags = ZPCI_PTE_VALID;
	294	dma_addr_t dma_addr;
52d43d81	295	int ret;
828b35f6	296
828b35f6 JG	297	/* This rounds up number of pages based on size and offset */
828b35f6 JG	298	nr_pages = iommu_num_pages(pa, size, PAGE_SIZE);
9a99649f	299	iommu_page_index = dma_alloc_iommu(dev, nr_pages);
52d43d81 SO	300	if (iommu_page_index == -1) {
52d43d81 SO	301	ret = -ENOSPC;
828b35f6	302	goto out_err;
52d43d81	303	}
828b35f6 JG	304
	305	/* Use rounded up size */
	306	size = nr_pages * PAGE_SIZE;
	307
	308	dma_addr = zdev->start_dma + iommu_page_index * PAGE_SIZE;
52d43d81 SO	309	if (dma_addr + size > zdev->end_dma) {
52d43d81 SO	310	ret = -ERANGE;
828b35f6	311	goto out_free;
52d43d81	312	}
828b35f6 JG	313
	314	if (direction == DMA_NONE \|\| direction == DMA_TO_DEVICE)
	315	flags \|= ZPCI_TABLE_PROTECTED;
	316
52d43d81 SO	317	ret = dma_update_trans(zdev, pa, dma_addr, size, flags);
	318	if (ret)
	319	goto out_free;
	320
	321	atomic64_add(nr_pages, &zdev->mapped_pages);
	322	return dma_addr + (offset & ~PAGE_MASK);
828b35f6 JG	323
828b35f6 JG	324	out_free:
9a99649f	325	dma_free_iommu(dev, iommu_page_index, nr_pages);
828b35f6	326	out_err:
1f1dcbd4	327	zpci_err("map error:\n");
52d43d81	328	zpci_err_dma(ret, pa);
828b35f6 JG	329	return DMA_ERROR_CODE;
	330	}
	331
	332	static void s390_dma_unmap_pages(struct device *dev, dma_addr_t dma_addr,
	333	size_t size, enum dma_data_direction direction,
	334	struct dma_attrs *attrs)
	335	{
198a5278	336	struct zpci_dev *zdev = to_zpci(to_pci_dev(dev));
828b35f6	337	unsigned long iommu_page_index;
52d43d81	338	int npages, ret;
828b35f6 JG	339
	340	npages = iommu_num_pages(dma_addr, size, PAGE_SIZE);
	341	dma_addr = dma_addr & PAGE_MASK;
52d43d81 SO	342	ret = dma_update_trans(zdev, 0, dma_addr, npages * PAGE_SIZE,
	343	ZPCI_PTE_INVALID);
	344	if (ret) {
1f1dcbd4	345	zpci_err("unmap error:\n");
52d43d81 SO	346	zpci_err_dma(ret, dma_addr);
52d43d81 SO	347	return;
1f1dcbd4	348	}
828b35f6	349
6001018a	350	atomic64_add(npages, &zdev->unmapped_pages);
828b35f6	351	iommu_page_index = (dma_addr - zdev->start_dma) >> PAGE_SHIFT;
9a99649f	352	dma_free_iommu(dev, iommu_page_index, npages);
828b35f6 JG	353	}
	354
	355	static void s390_dma_alloc(struct device dev, size_t size,
	356	dma_addr_t *dma_handle, gfp_t flag,
	357	struct dma_attrs *attrs)
	358	{
198a5278	359	struct zpci_dev *zdev = to_zpci(to_pci_dev(dev));
828b35f6 JG	360	struct page *page;
	361	unsigned long pa;
	362	dma_addr_t map;
	363
	364	size = PAGE_ALIGN(size);
	365	page = alloc_pages(flag, get_order(size));
	366	if (!page)
	367	return NULL;
d0b08853	368
828b35f6 JG	369	pa = page_to_phys(page);
	370	memset((void *) pa, 0, size);
	371
bdb97e91	372	map = s390_dma_map_pages(dev, page, 0, size, DMA_BIDIRECTIONAL, NULL);
828b35f6 JG	373	if (dma_mapping_error(dev, map)) {
	374	free_pages(pa, get_order(size));
	375	return NULL;
	376	}
	377
6001018a	378	atomic64_add(size / PAGE_SIZE, &zdev->allocated_pages);
828b35f6 JG	379	if (dma_handle)
	380	*dma_handle = map;
	381	return (void *) pa;
	382	}
	383
	384	static void s390_dma_free(struct device *dev, size_t size,
	385	void *pa, dma_addr_t dma_handle,
	386	struct dma_attrs *attrs)
	387	{
198a5278	388	struct zpci_dev *zdev = to_zpci(to_pci_dev(dev));
f7038b7c SO	389
f7038b7c SO	390	size = PAGE_ALIGN(size);
6001018a	391	atomic64_sub(size / PAGE_SIZE, &zdev->allocated_pages);
f7038b7c	392	s390_dma_unmap_pages(dev, dma_handle, size, DMA_BIDIRECTIONAL, NULL);
828b35f6 JG	393	free_pages((unsigned long) pa, get_order(size));
	394	}
	395
	396	static int s390_dma_map_sg(struct device dev, struct scatterlist sg,
	397	int nr_elements, enum dma_data_direction dir,
	398	struct dma_attrs *attrs)
	399	{
	400	int mapped_elements = 0;
	401	struct scatterlist *s;
	402	int i;
	403
	404	for_each_sg(sg, s, nr_elements, i) {
	405	struct page *page = sg_page(s);
	406	s->dma_address = s390_dma_map_pages(dev, page, s->offset,
	407	s->length, dir, NULL);
	408	if (!dma_mapping_error(dev, s->dma_address)) {
	409	s->dma_length = s->length;
	410	mapped_elements++;
	411	} else
	412	goto unmap;
	413	}
	414	out:
	415	return mapped_elements;
	416
	417	unmap:
	418	for_each_sg(sg, s, mapped_elements, i) {
	419	if (s->dma_address)
	420	s390_dma_unmap_pages(dev, s->dma_address, s->dma_length,
	421	dir, NULL);
	422	s->dma_address = 0;
	423	s->dma_length = 0;
	424	}
	425	mapped_elements = 0;
	426	goto out;
	427	}
	428
	429	static void s390_dma_unmap_sg(struct device dev, struct scatterlist sg,
	430	int nr_elements, enum dma_data_direction dir,
	431	struct dma_attrs *attrs)
	432	{
	433	struct scatterlist *s;
	434	int i;
	435
	436	for_each_sg(sg, s, nr_elements, i) {
	437	s390_dma_unmap_pages(dev, s->dma_address, s->dma_length, dir, NULL);
	438	s->dma_address = 0;
	439	s->dma_length = 0;
	440	}
	441	}
	442
	443	int zpci_dma_init_device(struct zpci_dev *zdev)
	444	{
828b35f6 JG	445	int rc;
828b35f6 JG	446
8128f23c GS	447	/*
	448	* At this point, if the device is part of an IOMMU domain, this would
	449	* be a strong hint towards a bug in the IOMMU API (common) code and/or
	450	* simultaneous access via IOMMU and DMA API. So let's issue a warning.
	451	*/
	452	WARN_ON(zdev->s390_domain);
	453
828b35f6 JG	454	spin_lock_init(&zdev->iommu_bitmap_lock);
	455	spin_lock_init(&zdev->dma_table_lock);
	456
	457	zdev->dma_table = dma_alloc_cpu_table();
	458	if (!zdev->dma_table) {
	459	rc = -ENOMEM;
dba59909	460	goto out;
828b35f6 JG	461	}
828b35f6 JG	462
69eea95c GS	463	/*
	464	* Restrict the iommu bitmap size to the minimum of the following:
	465	* - main memory size
	466	* - 3-level pagetable address limit minus start_dma offset
	467	* - DMA address range allowed by the hardware (clp query pci fn)
	468	*
	469	* Also set zdev->end_dma to the actual end address of the usable
	470	* range, instead of the theoretical maximum as reported by hardware.
	471	*/
	472	zdev->iommu_size = min3((u64) high_memory,
	473	ZPCI_TABLE_SIZE_RT - zdev->start_dma,
	474	zdev->end_dma - zdev->start_dma + 1);
	475	zdev->end_dma = zdev->start_dma + zdev->iommu_size - 1;
828b35f6	476	zdev->iommu_pages = zdev->iommu_size >> PAGE_SHIFT;
22459321	477	zdev->iommu_bitmap = vzalloc(zdev->iommu_pages / 8);
828b35f6 JG	478	if (!zdev->iommu_bitmap) {
828b35f6 JG	479	rc = -ENOMEM;
dba59909	480	goto free_dma_table;
828b35f6 JG	481	}
828b35f6 JG	482
69eea95c	483	rc = zpci_register_ioat(zdev, 0, zdev->start_dma, zdev->end_dma,
828b35f6 JG	484	(u64) zdev->dma_table);
828b35f6 JG	485	if (rc)
dba59909	486	goto free_bitmap;
828b35f6	487
dba59909 SO	488	return 0;
	489	free_bitmap:
	490	vfree(zdev->iommu_bitmap);
	491	zdev->iommu_bitmap = NULL;
	492	free_dma_table:
828b35f6	493	dma_free_cpu_table(zdev->dma_table);
dba59909 SO	494	zdev->dma_table = NULL;
dba59909 SO	495	out:
828b35f6 JG	496	return rc;
	497	}
	498
	499	void zpci_dma_exit_device(struct zpci_dev *zdev)
	500	{
8128f23c GS	501	/*
	502	* At this point, if the device is part of an IOMMU domain, this would
	503	* be a strong hint towards a bug in the IOMMU API (common) code and/or
	504	* simultaneous access via IOMMU and DMA API. So let's issue a warning.
	505	*/
	506	WARN_ON(zdev->s390_domain);
	507
828b35f6	508	zpci_unregister_ioat(zdev, 0);
8128f23c GS	509	dma_cleanup_tables(zdev->dma_table);
8128f23c GS	510	zdev->dma_table = NULL;
22459321	511	vfree(zdev->iommu_bitmap);
828b35f6 JG	512	zdev->iommu_bitmap = NULL;
	513	zdev->next_bit = 0;
	514	}
	515
	516	static int __init dma_alloc_cpu_table_caches(void)
	517	{
	518	dma_region_table_cache = kmem_cache_create("PCI_DMA_region_tables",
	519	ZPCI_TABLE_SIZE, ZPCI_TABLE_ALIGN,
	520	0, NULL);
	521	if (!dma_region_table_cache)
	522	return -ENOMEM;
	523
	524	dma_page_table_cache = kmem_cache_create("PCI_DMA_page_tables",
	525	ZPCI_PT_SIZE, ZPCI_PT_ALIGN,
	526	0, NULL);
	527	if (!dma_page_table_cache) {
	528	kmem_cache_destroy(dma_region_table_cache);
	529	return -ENOMEM;
	530	}
	531	return 0;
	532	}
	533
	534	int __init zpci_dma_init(void)
	535	{
	536	return dma_alloc_cpu_table_caches();
	537	}
	538
	539	void zpci_dma_exit(void)
	540	{
	541	kmem_cache_destroy(dma_page_table_cache);
	542	kmem_cache_destroy(dma_region_table_cache);
	543	}
	544
	545	#define PREALLOC_DMA_DEBUG_ENTRIES (1 << 16)
	546
	547	static int __init dma_debug_do_init(void)
	548	{
	549	dma_debug_init(PREALLOC_DMA_DEBUG_ENTRIES);
	550	return 0;
	551	}
	552	fs_initcall(dma_debug_do_init);
	553
e82becfc	554	struct dma_map_ops s390_pci_dma_ops = {
828b35f6 JG	555	.alloc = s390_dma_alloc,
	556	.free = s390_dma_free,
	557	.map_sg = s390_dma_map_sg,
	558	.unmap_sg = s390_dma_unmap_sg,
	559	.map_page = s390_dma_map_pages,
	560	.unmap_page = s390_dma_unmap_pages,
	561	/* if we support direct DMA this must be conditional */
	562	.is_phys = 0,
	563	/* dma_supported is unconditionally true without a callback */
	564	};
e82becfc	565	EXPORT_SYMBOL_GPL(s390_pci_dma_ops);
c60d1ae4 GS	566
	567	static int __init s390_iommu_setup(char *str)
	568	{
	569	if (!strncmp(str, "strict", 6))
	570	s390_iommu_strict = 1;
	571	return 0;
	572	}
	573
	574	__setup("s390_iommu=", s390_iommu_setup);