[linux-2.6-block.git] / arch / x86 / mm / gup.c

/*
 * Lockless get_user_pages_fast for x86
 *
 * Copyright (C) 2008 Nick Piggin
 * Copyright (C) 2008 Novell Inc.
 */
#include <linux/sched.h>
#include <linux/mm.h>
#include <linux/vmstat.h>
#include <linux/highmem.h>

#include <asm/pgtable.h>

static inline pte_t gup_get_pte(pte_t *ptep)
{
#ifndef CONFIG_X86_PAE
	return *ptep;
#else
	/*
	 * With get_user_pages_fast, we walk down the pagetables without taking
	 * any locks.  For this we would like to load the pointers atoimcally,
	 * but that is not possible (without expensive cmpxchg8b) on PAE.  What
	 * we do have is the guarantee that a pte will only either go from not
	 * present to present, or present to not present or both -- it will not
	 * switch to a completely different present page without a TLB flush in
	 * between; something that we are blocking by holding interrupts off.
	 *
	 * Setting ptes from not present to present goes:
	 * ptep->pte_high = h;
	 * smp_wmb();
	 * ptep->pte_low = l;
	 *
	 * And present to not present goes:
	 * ptep->pte_low = 0;
	 * smp_wmb();
	 * ptep->pte_high = 0;
	 *
	 * We must ensure here that the load of pte_low sees l iff pte_high
	 * sees h. We load pte_high *after* loading pte_low, which ensures we
	 * don't see an older value of pte_high.  *Then* we recheck pte_low,
	 * which ensures that we haven't picked up a changed pte high. We might
	 * have got rubbish values from pte_low and pte_high, but we are
	 * guaranteed that pte_low will not have the present bit set *unless*
	 * it is 'l'. And get_user_pages_fast only operates on present ptes, so
	 * we're safe.
	 *
	 * gup_get_pte should not be used or copied outside gup.c without being
	 * very careful -- it does not atomically load the pte or anything that
	 * is likely to be useful for you.
	 */
	pte_t pte;

retry:
	pte.pte_low = ptep->pte_low;
	smp_rmb();
	pte.pte_high = ptep->pte_high;
	smp_rmb();
	if (unlikely(pte.pte_low != ptep->pte_low))
		goto retry;

	return pte;
#endif
}

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

	mask = _PAGE_PRESENT|_PAGE_USER;
	if (write)
		mask |= _PAGE_RW;

	ptep = pte_offset_map(&pmd, addr);
	do {
		pte_t pte = gup_get_pte(ptep);
		struct page *page;

		if ((pte_flags(pte) & (mask | _PAGE_SPECIAL)) != mask) {
			pte_unmap(ptep);
			return 0;
		}
		VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
		page = pte_page(pte);
		get_page(page);
		pages[*nr] = page;
		(*nr)++;

	} while (ptep++, addr += PAGE_SIZE, addr != end);
	pte_unmap(ptep - 1);

	return 1;
}

static inline void get_head_page_multiple(struct page *page, int nr)
{
	VM_BUG_ON(page != compound_head(page));
	VM_BUG_ON(page_count(page) == 0);
	atomic_add(nr, &page->_count);
}

static noinline int gup_huge_pmd(pmd_t pmd, unsigned long addr,
		unsigned long end, int write, struct page **pages, int *nr)
{
	unsigned long mask;
	pte_t pte = *(pte_t *)&pmd;
	struct page *head, *page;
	int refs;

	mask = _PAGE_PRESENT|_PAGE_USER;
	if (write)
		mask |= _PAGE_RW;
	if ((pte_flags(pte) & mask) != mask)
		return 0;
	/* hugepages are never "special" */
	VM_BUG_ON(pte_flags(pte) & _PAGE_SPECIAL);
	VM_BUG_ON(!pfn_valid(pte_pfn(pte)));

	refs = 0;
	head = pte_page(pte);
	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);
	get_head_page_multiple(head, refs);

	return 1;
}

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

	pmdp = pmd_offset(&pud, addr);
	do {
		pmd_t pmd = *pmdp;

		next = pmd_addr_end(addr, end);
		if (pmd_none(pmd))
			return 0;
		if (unlikely(pmd_large(pmd))) {
			if (!gup_huge_pmd(pmd, addr, next, write, pages, nr))
				return 0;
		} else {
			if (!gup_pte_range(pmd, addr, next, write, pages, nr))
				return 0;
		}
	} while (pmdp++, addr = next, addr != end);

	return 1;
}

static noinline int gup_huge_pud(pud_t pud, unsigned long addr,
		unsigned long end, int write, struct page **pages, int *nr)
{
	unsigned long mask;
	pte_t pte = *(pte_t *)&pud;
	struct page *head, *page;
	int refs;

	mask = _PAGE_PRESENT|_PAGE_USER;
	if (write)
		mask |= _PAGE_RW;
	if ((pte_flags(pte) & mask) != mask)
		return 0;
	/* hugepages are never "special" */
	VM_BUG_ON(pte_flags(pte) & _PAGE_SPECIAL);
	VM_BUG_ON(!pfn_valid(pte_pfn(pte)));

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

	return 1;
}

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

	pudp = pud_offset(&pgd, addr);
	do {
		pud_t pud = *pudp;

		next = pud_addr_end(addr, end);
		if (pud_none(pud))
			return 0;
		if (unlikely(pud_large(pud))) {
			if (!gup_huge_pud(pud, addr, next, write, pages, nr))
				return 0;
		} else {
			if (!gup_pmd_range(pud, addr, next, write, pages, nr))
				return 0;
		}
	} while (pudp++, addr = next, addr != end);

	return 1;
}

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;
	pgd_t *pgdp;
	int nr = 0;

	start &= PAGE_MASK;
	addr = start;
	len = (unsigned long) nr_pages << PAGE_SHIFT;
	end = start + len;
	if (unlikely(!access_ok(write ? VERIFY_WRITE : VERIFY_READ,
					(void __user *)start, len)))
		goto slow_irqon;

	/*
	 * XXX: batch / limit 'nr', to avoid large irq off latency
	 * needs some instrumenting to determine the common sizes used by
	 * important workloads (eg. DB2), and whether limiting the batch size
	 * will decrease performance.
	 *
	 * It seems like we're in the clear for the moment. Direct-IO is
	 * the main guy that batches up lots of get_user_pages, and even
	 * they are limited to 64-at-a-time which is not so many.
	 */
	/*
	 * This doesn't prevent pagetable teardown, but does prevent
	 * the pagetables and pages from being freed on x86.
	 *
	 * So long as we atomically load page table pointers versus teardown
	 * (which we do on x86, with the above PAE exception), we can follow the
	 * address down to the the page and take a ref on it.
	 */
	local_irq_disable();
	pgdp = pgd_offset(mm, addr);
	do {
		pgd_t pgd = *pgdp;

		next = pgd_addr_end(addr, end);
		if (pgd_none(pgd))
			goto slow;
		if (!gup_pud_range(pgd, addr, next, write, pages, &nr))
			goto slow;
	} while (pgdp++, addr = next, addr != end);
	local_irq_enable();

	VM_BUG_ON(nr != (end - start) >> PAGE_SHIFT);
	return nr;

	{
		int ret;

slow:
		local_irq_enable();
slow_irqon:
		/* Try to get the remaining pages with get_user_pages */
		start += nr << PAGE_SHIFT;
		pages += nr;

		down_read(&mm->mmap_sem);
		ret = get_user_pages(current, mm, start,
			(end - start) >> PAGE_SHIFT, write, 0, pages, NULL);
		up_read(&mm->mmap_sem);

		/* Have to be a bit careful with return values */
		if (nr > 0) {
			if (ret < 0)
				ret = nr;
			else
				ret += nr;
		}

		return ret;
	}
}
Commit	Line	Data
8174c430 NP	1	/*
	2	* Lockless get_user_pages_fast for x86
	3	*
	4	* Copyright (C) 2008 Nick Piggin
	5	* Copyright (C) 2008 Novell Inc.
	6	*/
	7	#include <linux/sched.h>
	8	#include <linux/mm.h>
	9	#include <linux/vmstat.h>
	10	#include <linux/highmem.h>
	11
	12	#include <asm/pgtable.h>
	13
	14	static inline pte_t gup_get_pte(pte_t *ptep)
	15	{
	16	#ifndef CONFIG_X86_PAE
	17	return *ptep;
	18	#else
	19	/*
	20	* With get_user_pages_fast, we walk down the pagetables without taking
	21	* any locks. For this we would like to load the pointers atoimcally,
	22	* but that is not possible (without expensive cmpxchg8b) on PAE. What
	23	* we do have is the guarantee that a pte will only either go from not
	24	* present to present, or present to not present or both -- it will not
	25	* switch to a completely different present page without a TLB flush in
	26	* between; something that we are blocking by holding interrupts off.
	27	*
	28	* Setting ptes from not present to present goes:
	29	* ptep->pte_high = h;
	30	* smp_wmb();
	31	* ptep->pte_low = l;
	32	*
	33	* And present to not present goes:
	34	* ptep->pte_low = 0;
	35	* smp_wmb();
	36	* ptep->pte_high = 0;
	37	*
	38	* We must ensure here that the load of pte_low sees l iff pte_high
	39	* sees h. We load pte_high after loading pte_low, which ensures we
	40	* don't see an older value of pte_high. Then we recheck pte_low,
	41	* which ensures that we haven't picked up a changed pte high. We might
	42	* have got rubbish values from pte_low and pte_high, but we are
	43	* guaranteed that pte_low will not have the present bit set unless
	44	* it is 'l'. And get_user_pages_fast only operates on present ptes, so
	45	* we're safe.
	46	*
	47	* gup_get_pte should not be used or copied outside gup.c without being
	48	* very careful -- it does not atomically load the pte or anything that
	49	* is likely to be useful for you.
	50	*/
	51	pte_t pte;
	52
	53	retry:
	54	pte.pte_low = ptep->pte_low;
	55	smp_rmb();
	56	pte.pte_high = ptep->pte_high;
	57	smp_rmb();
	58	if (unlikely(pte.pte_low != ptep->pte_low))
	59	goto retry;
	60
	61	return pte;
	62	#endif
	63	}
	64
65	/*
66	* The performance critical leaf functions are made noinline otherwise gcc
67	* inlines everything into a single function which results in too much
68	* register pressure.
69	*/
70	static noinline int gup_pte_range(pmd_t pmd, unsigned long addr,
71	unsigned long end, int write, struct page *pages, int nr)
72	{
73	unsigned long mask;
74	pte_t *ptep;
75
76	mask = _PAGE_PRESENT\|_PAGE_USER;
77	if (write)
78	mask \|= _PAGE_RW;
79
80	ptep = pte_offset_map(&pmd, addr);
81	do {
82	pte_t pte = gup_get_pte(ptep);
83	struct page *page;
84
606ee44d	85	if ((pte_flags(pte) & (mask \| _PAGE_SPECIAL)) != mask) {
8174c430 NP	86	pte_unmap(ptep);
	87	return 0;
	88	}
	89	VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
	90	page = pte_page(pte);
	91	get_page(page);
	92	pages[*nr] = page;
	93	(*nr)++;
	94
	95	} while (ptep++, addr += PAGE_SIZE, addr != end);
	96	pte_unmap(ptep - 1);
	97
	98	return 1;
	99	}
	100
	101	static inline void get_head_page_multiple(struct page *page, int nr)
	102	{
	103	VM_BUG_ON(page != compound_head(page));
	104	VM_BUG_ON(page_count(page) == 0);
	105	atomic_add(nr, &page->_count);
	106	}
	107
	108	static noinline int gup_huge_pmd(pmd_t pmd, unsigned long addr,
	109	unsigned long end, int write, struct page *pages, int nr)
	110	{
	111	unsigned long mask;
	112	pte_t pte = (pte_t )&pmd;
	113	struct page head, page;
	114	int refs;
	115
	116	mask = _PAGE_PRESENT\|_PAGE_USER;
	117	if (write)
	118	mask \|= _PAGE_RW;
606ee44d	119	if ((pte_flags(pte) & mask) != mask)
8174c430 NP	120	return 0;
8174c430 NP	121	/* hugepages are never "special" */
606ee44d	122	VM_BUG_ON(pte_flags(pte) & _PAGE_SPECIAL);
8174c430 NP	123	VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
	124
	125	refs = 0;
	126	head = pte_page(pte);
652ea695	127	page = head + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
8174c430 NP	128	do {
	129	VM_BUG_ON(compound_head(page) != head);
	130	pages[*nr] = page;
	131	(*nr)++;
	132	page++;
	133	refs++;
	134	} while (addr += PAGE_SIZE, addr != end);
	135	get_head_page_multiple(head, refs);
	136
	137	return 1;
	138	}
	139
	140	static int gup_pmd_range(pud_t pud, unsigned long addr, unsigned long end,
	141	int write, struct page *pages, int nr)
	142	{
	143	unsigned long next;
	144	pmd_t *pmdp;
	145
	146	pmdp = pmd_offset(&pud, addr);
	147	do {
	148	pmd_t pmd = *pmdp;
	149
	150	next = pmd_addr_end(addr, end);
	151	if (pmd_none(pmd))
	152	return 0;
	153	if (unlikely(pmd_large(pmd))) {
	154	if (!gup_huge_pmd(pmd, addr, next, write, pages, nr))
	155	return 0;
	156	} else {
	157	if (!gup_pte_range(pmd, addr, next, write, pages, nr))
	158	return 0;
	159	}
	160	} while (pmdp++, addr = next, addr != end);
	161
	162	return 1;
	163	}
	164
652ea695 NP	165	static noinline int gup_huge_pud(pud_t pud, unsigned long addr,
	166	unsigned long end, int write, struct page *pages, int nr)
	167	{
	168	unsigned long mask;
	169	pte_t pte = (pte_t )&pud;
	170	struct page head, page;
	171	int refs;
	172
	173	mask = _PAGE_PRESENT\|_PAGE_USER;
	174	if (write)
	175	mask \|= _PAGE_RW;
606ee44d	176	if ((pte_flags(pte) & mask) != mask)
652ea695 NP	177	return 0;
652ea695 NP	178	/* hugepages are never "special" */
606ee44d	179	VM_BUG_ON(pte_flags(pte) & _PAGE_SPECIAL);
652ea695 NP	180	VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
	181
	182	refs = 0;
	183	head = pte_page(pte);
	184	page = head + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
	185	do {
	186	VM_BUG_ON(compound_head(page) != head);
	187	pages[*nr] = page;
	188	(*nr)++;
	189	page++;
	190	refs++;
	191	} while (addr += PAGE_SIZE, addr != end);
	192	get_head_page_multiple(head, refs);
	193
	194	return 1;
	195	}
	196
8174c430 NP	197	static int gup_pud_range(pgd_t pgd, unsigned long addr, unsigned long end,
	198	int write, struct page *pages, int nr)
	199	{
	200	unsigned long next;
	201	pud_t *pudp;
	202
	203	pudp = pud_offset(&pgd, addr);
	204	do {
	205	pud_t pud = *pudp;
	206
	207	next = pud_addr_end(addr, end);
	208	if (pud_none(pud))
	209	return 0;
652ea695 NP	210	if (unlikely(pud_large(pud))) {
	211	if (!gup_huge_pud(pud, addr, next, write, pages, nr))
	212	return 0;
	213	} else {
	214	if (!gup_pmd_range(pud, addr, next, write, pages, nr))
	215	return 0;
	216	}
8174c430 NP	217	} while (pudp++, addr = next, addr != end);
	218
	219	return 1;
	220	}
	221
	222	int get_user_pages_fast(unsigned long start, int nr_pages, int write,
	223	struct page **pages)
	224	{
	225	struct mm_struct *mm = current->mm;
9b79022c	226	unsigned long addr, len, end;
8174c430 NP	227	unsigned long next;
	228	pgd_t *pgdp;
	229	int nr = 0;
	230
9b79022c LT	231	start &= PAGE_MASK;
	232	addr = start;
	233	len = (unsigned long) nr_pages << PAGE_SHIFT;
	234	end = start + len;
8174c430	235	if (unlikely(!access_ok(write ? VERIFY_WRITE : VERIFY_READ,
9352f569	236	(void __user *)start, len)))
8174c430 NP	237	goto slow_irqon;
	238
	239	/*
	240	* XXX: batch / limit 'nr', to avoid large irq off latency
	241	* needs some instrumenting to determine the common sizes used by
	242	* important workloads (eg. DB2), and whether limiting the batch size
	243	* will decrease performance.
	244	*
	245	* It seems like we're in the clear for the moment. Direct-IO is
	246	* the main guy that batches up lots of get_user_pages, and even
	247	* they are limited to 64-at-a-time which is not so many.
	248	*/
	249	/*
	250	* This doesn't prevent pagetable teardown, but does prevent
	251	* the pagetables and pages from being freed on x86.
	252	*
	253	* So long as we atomically load page table pointers versus teardown
	254	* (which we do on x86, with the above PAE exception), we can follow the
	255	* address down to the the page and take a ref on it.
	256	*/
	257	local_irq_disable();
	258	pgdp = pgd_offset(mm, addr);
	259	do {
	260	pgd_t pgd = *pgdp;
	261
	262	next = pgd_addr_end(addr, end);
	263	if (pgd_none(pgd))
	264	goto slow;
	265	if (!gup_pud_range(pgd, addr, next, write, pages, &nr))
	266	goto slow;
	267	} while (pgdp++, addr = next, addr != end);
	268	local_irq_enable();
	269
	270	VM_BUG_ON(nr != (end - start) >> PAGE_SHIFT);
	271	return nr;
	272
	273	{
	274	int ret;
	275
	276	slow:
	277	local_irq_enable();
	278	slow_irqon:
	279	/* Try to get the remaining pages with get_user_pages */
	280	start += nr << PAGE_SHIFT;
	281	pages += nr;
	282
	283	down_read(&mm->mmap_sem);
	284	ret = get_user_pages(current, mm, start,
	285	(end - start) >> PAGE_SHIFT, write, 0, pages, NULL);
	286	up_read(&mm->mmap_sem);
	287
	288	/* Have to be a bit careful with return values */
	289	if (nr > 0) {
	290	if (ret < 0)
	291	ret = nr;
	292	else
	293	ret += nr;
	294	}
	295
	296	return ret;
	297	}
	298	}