[deliverable/linux.git] / arch / s390 / mm / gup.c

/*
 *  Lockless get_user_pages_fast for s390
 *
 *  Copyright IBM Corp. 2010
 *  Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
 */
#include <linux/sched.h>
#include <linux/mm.h>
#include <linux/hugetlb.h>
#include <linux/vmstat.h>
#include <linux/pagemap.h>
#include <linux/rwsem.h>
#include <asm/pgtable.h>

/*
 * The performance critical leaf functions are made noinline otherwise gcc
 * inlines everything into a single function which results in too much
 * register pressure.
 */
static inline int gup_pte_range(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
		unsigned long end, int write, struct page **pages, int *nr)
{
	unsigned long mask;
	pte_t *ptep, pte;
	struct page *page;

	mask = (write ? _PAGE_PROTECT : 0) | _PAGE_INVALID | _PAGE_SPECIAL;

	ptep = ((pte_t *) pmd_deref(pmd)) + pte_index(addr);
	do {
		pte = *ptep;
		barrier();
		/* Similar to the PMD case, NUMA hinting must take slow path */
		if (pte_protnone(pte))
			return 0;
		if ((pte_val(pte) & mask) != 0)
			return 0;
		VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
		page = pte_page(pte);
		if (!page_cache_get_speculative(page))
			return 0;
		if (unlikely(pte_val(pte) != pte_val(*ptep))) {
			put_page(page);
			return 0;
		}
		pages[*nr] = page;
		(*nr)++;

	} while (ptep++, addr += PAGE_SIZE, addr != end);

	return 1;
}

static inline int gup_huge_pmd(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
		unsigned long end, int write, struct page **pages, int *nr)
{
	unsigned long mask, result;
	struct page *head, *page;
	int refs;

	result = write ? 0 : _SEGMENT_ENTRY_PROTECT;
	mask = result | _SEGMENT_ENTRY_INVALID;
	if ((pmd_val(pmd) & mask) != result)
		return 0;
	VM_BUG_ON(!pfn_valid(pmd_val(pmd) >> PAGE_SHIFT));

	refs = 0;
	head = pmd_page(pmd);
	page = head + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
	do {
		VM_BUG_ON(compound_head(page) != head);
		pages[*nr] = page;
		(*nr)++;
		page++;
		refs++;
	} while (addr += PAGE_SIZE, addr != end);

	if (!page_cache_add_speculative(head, refs)) {
		*nr -= refs;
		return 0;
	}

	if (unlikely(pmd_val(pmd) != pmd_val(*pmdp))) {
		*nr -= refs;
		while (refs--)
			put_page(head);
		return 0;
	}

	return 1;
}


static inline int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr,
		unsigned long end, int write, struct page **pages, int *nr)
{
	unsigned long next;
	pmd_t *pmdp, pmd;

	pmdp = (pmd_t *) pudp;
	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
		pmdp = (pmd_t *) pud_deref(pud);
	pmdp += pmd_index(addr);
	do {
		pmd = *pmdp;
		barrier();
		next = pmd_addr_end(addr, end);
		/*
		 * The pmd_trans_splitting() check below explains why
		 * pmdp_splitting_flush() has to serialize with
		 * smp_call_function() against our disabled IRQs, to stop
		 * this gup-fast code from running while we set the
		 * splitting bit in the pmd. Returning zero will take
		 * the slow path that will call wait_split_huge_page()
		 * if the pmd is still in splitting state.
		 */
		if (pmd_none(pmd) || pmd_trans_splitting(pmd))
			return 0;
		if (unlikely(pmd_large(pmd))) {
			/*
			 * NUMA hinting faults need to be handled in the GUP
			 * slowpath for accounting purposes and so that they
			 * can be serialised against THP migration.
			 */
			if (pmd_protnone(pmd))
				return 0;
			if (!gup_huge_pmd(pmdp, pmd, addr, next,
					  write, pages, nr))
				return 0;
		} else if (!gup_pte_range(pmdp, pmd, addr, next,
					  write, pages, nr))
			return 0;
	} while (pmdp++, addr = next, addr != end);

	return 1;
}

static inline int gup_pud_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr,
		unsigned long end, int write, struct page **pages, int *nr)
{
	unsigned long next;
	pud_t *pudp, pud;

	pudp = (pud_t *) pgdp;
	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
		pudp = (pud_t *) pgd_deref(pgd);
	pudp += pud_index(addr);
	do {
		pud = *pudp;
		barrier();
		next = pud_addr_end(addr, end);
		if (pud_none(pud))
			return 0;
		if (!gup_pmd_range(pudp, pud, addr, next, write, pages, nr))
			return 0;
	} while (pudp++, addr = next, addr != end);

	return 1;
}

/*
 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
 * back to the regular GUP.
 */
int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
			  struct page **pages)
{
	struct mm_struct *mm = current->mm;
	unsigned long addr, len, end;
	unsigned long next, flags;
	pgd_t *pgdp, pgd;
	int nr = 0;

	start &= PAGE_MASK;
	addr = start;
	len = (unsigned long) nr_pages << PAGE_SHIFT;
	end = start + len;
	if ((end <= start) || (end > TASK_SIZE))
		return 0;
	/*
	 * local_irq_save() doesn't prevent pagetable teardown, but does
	 * prevent the pagetables from being freed on s390.
	 *
	 * So long as we atomically load page table pointers versus teardown,
	 * we can follow the address down to the the page and take a ref on it.
	 */
	local_irq_save(flags);
	pgdp = pgd_offset(mm, addr);
	do {
		pgd = *pgdp;
		barrier();
		next = pgd_addr_end(addr, end);
		if (pgd_none(pgd))
			break;
		if (!gup_pud_range(pgdp, pgd, addr, next, write, pages, &nr))
			break;
	} while (pgdp++, addr = next, addr != end);
	local_irq_restore(flags);

	return nr;
}

/**
 * get_user_pages_fast() - pin user pages in memory
 * @start:	starting user address
 * @nr_pages:	number of pages from start to pin
 * @write:	whether pages will be written to
 * @pages:	array that receives pointers to the pages pinned.
 *		Should be at least nr_pages long.
 *
 * Attempt to pin user pages in memory without taking mm->mmap_sem.
 * If not successful, it will fall back to taking the lock and
 * calling get_user_pages().
 *
 * Returns number of pages pinned. This may be fewer than the number
 * requested. If nr_pages is 0 or negative, returns 0. If no pages
 * were pinned, returns -errno.
 */
int get_user_pages_fast(unsigned long start, int nr_pages, int write,
			struct page **pages)
{
	struct mm_struct *mm = current->mm;
	int nr, ret;

	might_sleep();
	start &= PAGE_MASK;
	nr = __get_user_pages_fast(start, nr_pages, write, pages);
	if (nr == nr_pages)
		return nr;

	/* Try to get the remaining pages with get_user_pages */
	start += nr << PAGE_SHIFT;
	pages += nr;
	ret = get_user_pages_unlocked(current, mm, start,
			     nr_pages - nr, write, 0, pages);
	/* Have to be a bit careful with return values */
	if (nr > 0)
		ret = (ret < 0) ? nr : ret + nr;
	return ret;
}
Commit	Line	Data
80217147 MS	1	/*
	2	* Lockless get_user_pages_fast for s390
	3	*
	4	* Copyright IBM Corp. 2010
	5	* Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
	6	*/
	7	#include <linux/sched.h>
	8	#include <linux/mm.h>
	9	#include <linux/hugetlb.h>
	10	#include <linux/vmstat.h>
	11	#include <linux/pagemap.h>
	12	#include <linux/rwsem.h>
	13	#include <asm/pgtable.h>
	14
	15	/*
	16	* The performance critical leaf functions are made noinline otherwise gcc
	17	* inlines everything into a single function which results in too much
	18	* register pressure.
	19	*/
	20	static inline int gup_pte_range(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
	21	unsigned long end, int write, struct page *pages, int nr)
	22	{
25591b07	23	unsigned long mask;
80217147 MS	24	pte_t *ptep, pte;
	25	struct page *page;
	26
e5098611	27	mask = (write ? _PAGE_PROTECT : 0) \| _PAGE_INVALID \| _PAGE_SPECIAL;
80217147 MS	28
	29	ptep = ((pte_t *) pmd_deref(pmd)) + pte_index(addr);
	30	do {
	31	pte = *ptep;
	32	barrier();
ecf46abd GS	33	/* Similar to the PMD case, NUMA hinting must take slow path */
	34	if (pte_protnone(pte))
	35	return 0;
25591b07	36	if ((pte_val(pte) & mask) != 0)
80217147 MS	37	return 0;
	38	VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
	39	page = pte_page(pte);
	40	if (!page_cache_get_speculative(page))
	41	return 0;
	42	if (unlikely(pte_val(pte) != pte_val(*ptep))) {
	43	put_page(page);
	44	return 0;
	45	}
	46	pages[*nr] = page;
	47	(*nr)++;
	48
	49	} while (ptep++, addr += PAGE_SIZE, addr != end);
	50
	51	return 1;
	52	}
	53
	54	static inline int gup_huge_pmd(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
	55	unsigned long end, int write, struct page *pages, int nr)
	56	{
	57	unsigned long mask, result;
ddc58f27	58	struct page head, page;
80217147 MS	59	int refs;
80217147 MS	60
e5098611 MS	61	result = write ? 0 : _SEGMENT_ENTRY_PROTECT;
e5098611 MS	62	mask = result \| _SEGMENT_ENTRY_INVALID;
80217147 MS	63	if ((pmd_val(pmd) & mask) != result)
	64	return 0;
	65	VM_BUG_ON(!pfn_valid(pmd_val(pmd) >> PAGE_SHIFT));
	66
	67	refs = 0;
	68	head = pmd_page(pmd);
	69	page = head + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
	70	do {
	71	VM_BUG_ON(compound_head(page) != head);
	72	pages[*nr] = page;
	73	(*nr)++;
	74	page++;
	75	refs++;
	76	} while (addr += PAGE_SIZE, addr != end);
	77
	78	if (!page_cache_add_speculative(head, refs)) {
	79	*nr -= refs;
	80	return 0;
	81	}
	82
	83	if (unlikely(pmd_val(pmd) != pmd_val(*pmdp))) {
	84	*nr -= refs;
	85	while (refs--)
	86	put_page(head);
0693bc9c AA	87	return 0;
	88	}
	89
80217147 MS	90	return 1;
	91	}
	92
	93
	94	static inline int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr,
	95	unsigned long end, int write, struct page *pages, int nr)
	96	{
	97	unsigned long next;
	98	pmd_t *pmdp, pmd;
	99
	100	pmdp = (pmd_t *) pudp;
80217147 MS	101	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
	102	pmdp = (pmd_t *) pud_deref(pud);
	103	pmdp += pmd_index(addr);
80217147 MS	104	do {
	105	pmd = *pmdp;
	106	barrier();
	107	next = pmd_addr_end(addr, end);
75077afb GS	108	/*
	109	* The pmd_trans_splitting() check below explains why
	110	* pmdp_splitting_flush() has to serialize with
	111	* smp_call_function() against our disabled IRQs, to stop
	112	* this gup-fast code from running while we set the
	113	* splitting bit in the pmd. Returning zero will take
	114	* the slow path that will call wait_split_huge_page()
	115	* if the pmd is still in splitting state.
	116	*/
	117	if (pmd_none(pmd) \|\| pmd_trans_splitting(pmd))
80217147	118	return 0;
156152f8	119	if (unlikely(pmd_large(pmd))) {
ecf46abd GS	120	/*
	121	* NUMA hinting faults need to be handled in the GUP
	122	* slowpath for accounting purposes and so that they
	123	* can be serialised against THP migration.
	124	*/
	125	if (pmd_protnone(pmd))
	126	return 0;
80217147 MS	127	if (!gup_huge_pmd(pmdp, pmd, addr, next,
	128	write, pages, nr))
	129	return 0;
	130	} else if (!gup_pte_range(pmdp, pmd, addr, next,
	131	write, pages, nr))
	132	return 0;
	133	} while (pmdp++, addr = next, addr != end);
	134
	135	return 1;
	136	}
	137
	138	static inline int gup_pud_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr,
	139	unsigned long end, int write, struct page *pages, int nr)
	140	{
	141	unsigned long next;
	142	pud_t *pudp, pud;
	143
	144	pudp = (pud_t *) pgdp;
80217147 MS	145	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
	146	pudp = (pud_t *) pgd_deref(pgd);
	147	pudp += pud_index(addr);
80217147 MS	148	do {
	149	pud = *pudp;
	150	barrier();
	151	next = pud_addr_end(addr, end);
	152	if (pud_none(pud))
	153	return 0;
	154	if (!gup_pmd_range(pudp, pud, addr, next, write, pages, nr))
	155	return 0;
	156	} while (pudp++, addr = next, addr != end);
	157
	158	return 1;
	159	}
	160
34cda992 GS	161	/*
	162	* Like get_user_pages_fast() except its IRQ-safe in that it won't fall
	163	* back to the regular GUP.
	164	*/
	165	int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
	166	struct page **pages)
	167	{
	168	struct mm_struct *mm = current->mm;
	169	unsigned long addr, len, end;
	170	unsigned long next, flags;
	171	pgd_t *pgdp, pgd;
	172	int nr = 0;
	173
	174	start &= PAGE_MASK;
	175	addr = start;
	176	len = (unsigned long) nr_pages << PAGE_SHIFT;
	177	end = start + len;
eb0bf929	178	if ((end <= start) \|\| (end > TASK_SIZE))
34cda992	179	return 0;
01997bbc HC	180	/*
	181	* local_irq_save() doesn't prevent pagetable teardown, but does
	182	* prevent the pagetables from being freed on s390.
	183	*
	184	* So long as we atomically load page table pointers versus teardown,
	185	* we can follow the address down to the the page and take a ref on it.
	186	*/
34cda992 GS	187	local_irq_save(flags);
	188	pgdp = pgd_offset(mm, addr);
	189	do {
	190	pgd = *pgdp;
	191	barrier();
	192	next = pgd_addr_end(addr, end);
	193	if (pgd_none(pgd))
	194	break;
	195	if (!gup_pud_range(pgdp, pgd, addr, next, write, pages, &nr))
	196	break;
	197	} while (pgdp++, addr = next, addr != end);
	198	local_irq_restore(flags);
	199
	200	return nr;
	201	}
	202
80217147 MS	203	/**
	204	* get_user_pages_fast() - pin user pages in memory
	205	* @start: starting user address
	206	* @nr_pages: number of pages from start to pin
	207	* @write: whether pages will be written to
	208	* @pages: array that receives pointers to the pages pinned.
	209	* Should be at least nr_pages long.
	210	*
	211	* Attempt to pin user pages in memory without taking mm->mmap_sem.
	212	* If not successful, it will fall back to taking the lock and
	213	* calling get_user_pages().
	214	*
	215	* Returns number of pages pinned. This may be fewer than the number
	216	* requested. If nr_pages is 0 or negative, returns 0. If no pages
	217	* were pinned, returns -errno.
	218	*/
	219	int get_user_pages_fast(unsigned long start, int nr_pages, int write,
	220	struct page **pages)
	221	{
	222	struct mm_struct *mm = current->mm;
01997bbc	223	int nr, ret;
80217147	224
40612351	225	might_sleep();
80217147	226	start &= PAGE_MASK;
01997bbc HC	227	nr = __get_user_pages_fast(start, nr_pages, write, pages);
	228	if (nr == nr_pages)
	229	return nr;
	230
	231	/* Try to get the remaining pages with get_user_pages */
	232	start += nr << PAGE_SHIFT;
	233	pages += nr;
a7b78075 AA	234	ret = get_user_pages_unlocked(current, mm, start,
a7b78075 AA	235	nr_pages - nr, write, 0, pages);
01997bbc HC	236	/* Have to be a bit careful with return values */
	237	if (nr > 0)
	238	ret = (ret < 0) ? nr : ret + nr;
	239	return ret;
80217147	240	}