[mirror_ubuntu-bionic-kernel.git] / arch / s390 / mm / pgalloc.c

// SPDX-License-Identifier: GPL-2.0
/*
 *  Page table allocation functions
 *
 *    Copyright IBM Corp. 2016
 *    Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
 */

#include <linux/mm.h>
#include <linux/sysctl.h>
#include <asm/mmu_context.h>
#include <asm/pgalloc.h>
#include <asm/gmap.h>
#include <asm/tlb.h>
#include <asm/tlbflush.h>

#ifdef CONFIG_PGSTE

static int page_table_allocate_pgste_min = 0;
static int page_table_allocate_pgste_max = 1;
int page_table_allocate_pgste = 0;
EXPORT_SYMBOL(page_table_allocate_pgste);

static struct ctl_table page_table_sysctl[] = {
	{
		.procname	= "allocate_pgste",
		.data		= &page_table_allocate_pgste,
		.maxlen		= sizeof(int),
		.mode		= S_IRUGO | S_IWUSR,
		.proc_handler	= proc_dointvec_minmax,
		.extra1		= &page_table_allocate_pgste_min,
		.extra2		= &page_table_allocate_pgste_max,
	},
	{ }
};

static struct ctl_table page_table_sysctl_dir[] = {
	{
		.procname	= "vm",
		.maxlen		= 0,
		.mode		= 0555,
		.child		= page_table_sysctl,
	},
	{ }
};

static int __init page_table_register_sysctl(void)
{
	return register_sysctl_table(page_table_sysctl_dir) ? 0 : -ENOMEM;
}
__initcall(page_table_register_sysctl);

#endif /* CONFIG_PGSTE */

unsigned long *crst_table_alloc(struct mm_struct *mm)
{
	struct page *page = alloc_pages(GFP_KERNEL, 2);

	if (!page)
		return NULL;
	arch_set_page_dat(page, 2);
	return (unsigned long *) page_to_phys(page);
}

void crst_table_free(struct mm_struct *mm, unsigned long *table)
{
	free_pages((unsigned long) table, 2);
}

static void __crst_table_upgrade(void *arg)
{
	struct mm_struct *mm = arg;

	if (current->active_mm == mm)
		set_user_asce(mm);
	__tlb_flush_local();
}

int crst_table_upgrade(struct mm_struct *mm, unsigned long end)
{
	unsigned long *table, *pgd;
	int rc, notify;

	/* upgrade should only happen from 3 to 4, 3 to 5, or 4 to 5 levels */
	VM_BUG_ON(mm->context.asce_limit < _REGION2_SIZE);
	rc = 0;
	notify = 0;
	while (mm->context.asce_limit < end) {
		table = crst_table_alloc(mm);
		if (!table) {
			rc = -ENOMEM;
			break;
		}
		spin_lock_bh(&mm->page_table_lock);
		pgd = (unsigned long *) mm->pgd;
		if (mm->context.asce_limit == _REGION2_SIZE) {
			crst_table_init(table, _REGION2_ENTRY_EMPTY);
			p4d_populate(mm, (p4d_t *) table, (pud_t *) pgd);
			mm->pgd = (pgd_t *) table;
			mm->context.asce_limit = _REGION1_SIZE;
			mm->context.asce = __pa(mm->pgd) | _ASCE_TABLE_LENGTH |
				_ASCE_USER_BITS | _ASCE_TYPE_REGION2;
			mm_inc_nr_puds(mm);
		} else {
			crst_table_init(table, _REGION1_ENTRY_EMPTY);
			pgd_populate(mm, (pgd_t *) table, (p4d_t *) pgd);
			mm->pgd = (pgd_t *) table;
			mm->context.asce_limit = -PAGE_SIZE;
			mm->context.asce = __pa(mm->pgd) | _ASCE_TABLE_LENGTH |
				_ASCE_USER_BITS | _ASCE_TYPE_REGION1;
		}
		notify = 1;
		spin_unlock_bh(&mm->page_table_lock);
	}
	if (notify)
		on_each_cpu(__crst_table_upgrade, mm, 0);
	return rc;
}

void crst_table_downgrade(struct mm_struct *mm)
{
	pgd_t *pgd;

	/* downgrade should only happen from 3 to 2 levels (compat only) */
	VM_BUG_ON(mm->context.asce_limit != _REGION2_SIZE);

	if (current->active_mm == mm) {
		clear_user_asce();
		__tlb_flush_mm(mm);
	}

	pgd = mm->pgd;
	mm->pgd = (pgd_t *) (pgd_val(*pgd) & _REGION_ENTRY_ORIGIN);
	mm->context.asce_limit = _REGION3_SIZE;
	mm->context.asce = __pa(mm->pgd) | _ASCE_TABLE_LENGTH |
			   _ASCE_USER_BITS | _ASCE_TYPE_SEGMENT;
	crst_table_free(mm, (unsigned long *) pgd);

	if (current->active_mm == mm)
		set_user_asce(mm);
}

static inline unsigned int atomic_xor_bits(atomic_t *v, unsigned int bits)
{
	unsigned int old, new;

	do {
		old = atomic_read(v);
		new = old ^ bits;
	} while (atomic_cmpxchg(v, old, new) != old);
	return new;
}

#ifdef CONFIG_PGSTE

struct page *page_table_alloc_pgste(struct mm_struct *mm)
{
	struct page *page;
	u64 *table;

	page = alloc_page(GFP_KERNEL);
	if (page) {
		table = (u64 *)page_to_phys(page);
		memset64(table, _PAGE_INVALID, PTRS_PER_PTE);
		memset64(table + PTRS_PER_PTE, 0, PTRS_PER_PTE);
	}
	return page;
}

void page_table_free_pgste(struct page *page)
{
	__free_page(page);
}

#endif /* CONFIG_PGSTE */

/*
 * page table entry allocation/free routines.
 */
unsigned long *page_table_alloc(struct mm_struct *mm)
{
	unsigned long *table;
	struct page *page;
	unsigned int mask, bit;

	/* Try to get a fragment of a 4K page as a 2K page table */
	if (!mm_alloc_pgste(mm)) {
		table = NULL;
		spin_lock_bh(&mm->context.lock);
		if (!list_empty(&mm->context.pgtable_list)) {
			page = list_first_entry(&mm->context.pgtable_list,
						struct page, lru);
			mask = atomic_read(&page->_mapcount);
			mask = (mask | (mask >> 4)) & 3;
			if (mask != 3) {
				table = (unsigned long *) page_to_phys(page);
				bit = mask & 1;		/* =1 -> second 2K */
				if (bit)
					table += PTRS_PER_PTE;
				atomic_xor_bits(&page->_mapcount, 1U << bit);
				list_del(&page->lru);
			}
		}
		spin_unlock_bh(&mm->context.lock);
		if (table)
			return table;
	}
	/* Allocate a fresh page */
	page = alloc_page(GFP_KERNEL);
	if (!page)
		return NULL;
	if (!pgtable_page_ctor(page)) {
		__free_page(page);
		return NULL;
	}
	arch_set_page_dat(page, 0);
	/* Initialize page table */
	table = (unsigned long *) page_to_phys(page);
	if (mm_alloc_pgste(mm)) {
		/* Return 4K page table with PGSTEs */
		atomic_set(&page->_mapcount, 3);
		memset64((u64 *)table, _PAGE_INVALID, PTRS_PER_PTE);
		memset64((u64 *)table + PTRS_PER_PTE, 0, PTRS_PER_PTE);
	} else {
		/* Return the first 2K fragment of the page */
		atomic_set(&page->_mapcount, 1);
		memset64((u64 *)table, _PAGE_INVALID, 2 * PTRS_PER_PTE);
		spin_lock_bh(&mm->context.lock);
		list_add(&page->lru, &mm->context.pgtable_list);
		spin_unlock_bh(&mm->context.lock);
	}
	return table;
}

void page_table_free(struct mm_struct *mm, unsigned long *table)
{
	struct page *page;
	unsigned int bit, mask;

	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
	if (!mm_alloc_pgste(mm)) {
		/* Free 2K page table fragment of a 4K page */
		bit = (__pa(table) & ~PAGE_MASK)/(PTRS_PER_PTE*sizeof(pte_t));
		spin_lock_bh(&mm->context.lock);
		mask = atomic_xor_bits(&page->_mapcount, 1U << bit);
		if (mask & 3)
			list_add(&page->lru, &mm->context.pgtable_list);
		else
			list_del(&page->lru);
		spin_unlock_bh(&mm->context.lock);
		if (mask != 0)
			return;
	}

	pgtable_page_dtor(page);
	atomic_set(&page->_mapcount, -1);
	__free_page(page);
}

void page_table_free_rcu(struct mmu_gather *tlb, unsigned long *table,
			 unsigned long vmaddr)
{
	struct mm_struct *mm;
	struct page *page;
	unsigned int bit, mask;

	mm = tlb->mm;
	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
	if (mm_alloc_pgste(mm)) {
		gmap_unlink(mm, table, vmaddr);
		table = (unsigned long *) (__pa(table) | 3);
		tlb_remove_table(tlb, table);
		return;
	}
	bit = (__pa(table) & ~PAGE_MASK) / (PTRS_PER_PTE*sizeof(pte_t));
	spin_lock_bh(&mm->context.lock);
	mask = atomic_xor_bits(&page->_mapcount, 0x11U << bit);
	if (mask & 3)
		list_add_tail(&page->lru, &mm->context.pgtable_list);
	else
		list_del(&page->lru);
	spin_unlock_bh(&mm->context.lock);
	table = (unsigned long *) (__pa(table) | (1U << bit));
	tlb_remove_table(tlb, table);
}

static void __tlb_remove_table(void *_table)
{
	unsigned int mask = (unsigned long) _table & 3;
	void *table = (void *)((unsigned long) _table ^ mask);
	struct page *page = pfn_to_page(__pa(table) >> PAGE_SHIFT);

	switch (mask) {
	case 0:		/* pmd, pud, or p4d */
		free_pages((unsigned long) table, 2);
		break;
	case 1:		/* lower 2K of a 4K page table */
	case 2:		/* higher 2K of a 4K page table */
		if (atomic_xor_bits(&page->_mapcount, mask << 4) != 0)
			break;
		/* fallthrough */
	case 3:		/* 4K page table with pgstes */
		pgtable_page_dtor(page);
		atomic_set(&page->_mapcount, -1);
		__free_page(page);
		break;
	}
}

static void tlb_remove_table_smp_sync(void *arg)
{
	/* Simply deliver the interrupt */
}

static void tlb_remove_table_one(void *table)
{
	/*
	 * This isn't an RCU grace period and hence the page-tables cannot be
	 * assumed to be actually RCU-freed.
	 *
	 * It is however sufficient for software page-table walkers that rely
	 * on IRQ disabling. See the comment near struct mmu_table_batch.
	 */
	smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
	__tlb_remove_table(table);
}

static void tlb_remove_table_rcu(struct rcu_head *head)
{
	struct mmu_table_batch *batch;
	int i;

	batch = container_of(head, struct mmu_table_batch, rcu);

	for (i = 0; i < batch->nr; i++)
		__tlb_remove_table(batch->tables[i]);

	free_page((unsigned long)batch);
}

void tlb_table_flush(struct mmu_gather *tlb)
{
	struct mmu_table_batch **batch = &tlb->batch;

	if (*batch) {
		call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
		*batch = NULL;
	}
}

void tlb_remove_table(struct mmu_gather *tlb, void *table)
{
	struct mmu_table_batch **batch = &tlb->batch;

	tlb->mm->context.flush_mm = 1;
	if (*batch == NULL) {
		*batch = (struct mmu_table_batch *)
			__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
		if (*batch == NULL) {
			__tlb_flush_mm_lazy(tlb->mm);
			tlb_remove_table_one(table);
			return;
		}
		(*batch)->nr = 0;
	}
	(*batch)->tables[(*batch)->nr++] = table;
	if ((*batch)->nr == MAX_TABLE_BATCH)
		tlb_flush_mmu(tlb);
}
Commit	Line	Data
b2441318	1	// SPDX-License-Identifier: GPL-2.0
1e133ab2 MS	2	/*
	3	* Page table allocation functions
	4	*
	5	* Copyright IBM Corp. 2016
	6	* Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
	7	*/
	8
	9	#include <linux/mm.h>
	10	#include <linux/sysctl.h>
	11	#include <asm/mmu_context.h>
	12	#include <asm/pgalloc.h>
	13	#include <asm/gmap.h>
	14	#include <asm/tlb.h>
	15	#include <asm/tlbflush.h>
	16
	17	#ifdef CONFIG_PGSTE
	18
	19	static int page_table_allocate_pgste_min = 0;
	20	static int page_table_allocate_pgste_max = 1;
	21	int page_table_allocate_pgste = 0;
	22	EXPORT_SYMBOL(page_table_allocate_pgste);
	23
	24	static struct ctl_table page_table_sysctl[] = {
	25	{
	26	.procname = "allocate_pgste",
	27	.data = &page_table_allocate_pgste,
	28	.maxlen = sizeof(int),
	29	.mode = S_IRUGO \| S_IWUSR,
f24d6341	30	.proc_handler = proc_dointvec_minmax,
1e133ab2 MS	31	.extra1 = &page_table_allocate_pgste_min,
	32	.extra2 = &page_table_allocate_pgste_max,
	33	},
	34	{ }
	35	};
	36
	37	static struct ctl_table page_table_sysctl_dir[] = {
	38	{
	39	.procname = "vm",
	40	.maxlen = 0,
	41	.mode = 0555,
	42	.child = page_table_sysctl,
	43	},
	44	{ }
	45	};
	46
	47	static int __init page_table_register_sysctl(void)
	48	{
	49	return register_sysctl_table(page_table_sysctl_dir) ? 0 : -ENOMEM;
	50	}
	51	__initcall(page_table_register_sysctl);
	52
	53	#endif /* CONFIG_PGSTE */
	54
	55	unsigned long crst_table_alloc(struct mm_struct mm)
	56	{
	57	struct page *page = alloc_pages(GFP_KERNEL, 2);
	58
	59	if (!page)
	60	return NULL;
c9b5ad54	61	arch_set_page_dat(page, 2);
1e133ab2 MS	62	return (unsigned long *) page_to_phys(page);
	63	}
	64
	65	void crst_table_free(struct mm_struct mm, unsigned long table)
	66	{
	67	free_pages((unsigned long) table, 2);
	68	}
	69
	70	static void __crst_table_upgrade(void *arg)
	71	{
	72	struct mm_struct *mm = arg;
	73
0aaba41b	74	if (current->active_mm == mm)
1e133ab2	75	set_user_asce(mm);
1e133ab2 MS	76	__tlb_flush_local();
	77	}
	78
1aea9b3f	79	int crst_table_upgrade(struct mm_struct *mm, unsigned long end)
1e133ab2 MS	80	{
1e133ab2 MS	81	unsigned long table, pgd;
1aea9b3f	82	int rc, notify;
1e133ab2	83
1aea9b3f	84	/* upgrade should only happen from 3 to 4, 3 to 5, or 4 to 5 levels */
2fc4876e	85	VM_BUG_ON(mm->context.asce_limit < _REGION2_SIZE);
1aea9b3f MS	86	rc = 0;
	87	notify = 0;
	88	while (mm->context.asce_limit < end) {
	89	table = crst_table_alloc(mm);
	90	if (!table) {
	91	rc = -ENOMEM;
	92	break;
	93	}
	94	spin_lock_bh(&mm->page_table_lock);
	95	pgd = (unsigned long *) mm->pgd;
f1c1174f	96	if (mm->context.asce_limit == _REGION2_SIZE) {
1aea9b3f MS	97	crst_table_init(table, _REGION2_ENTRY_EMPTY);
	98	p4d_populate(mm, (p4d_t ) table, (pud_t ) pgd);
	99	mm->pgd = (pgd_t *) table;
f1c1174f	100	mm->context.asce_limit = _REGION1_SIZE;
1aea9b3f MS	101	mm->context.asce = __pa(mm->pgd) \| _ASCE_TABLE_LENGTH \|
1aea9b3f MS	102	_ASCE_USER_BITS \| _ASCE_TYPE_REGION2;
76d93ab6	103	mm_inc_nr_puds(mm);
1aea9b3f MS	104	} else {
	105	crst_table_init(table, _REGION1_ENTRY_EMPTY);
	106	pgd_populate(mm, (pgd_t ) table, (p4d_t ) pgd);
	107	mm->pgd = (pgd_t *) table;
	108	mm->context.asce_limit = -PAGE_SIZE;
	109	mm->context.asce = __pa(mm->pgd) \| _ASCE_TABLE_LENGTH \|
	110	_ASCE_USER_BITS \| _ASCE_TYPE_REGION1;
	111	}
	112	notify = 1;
	113	spin_unlock_bh(&mm->page_table_lock);
	114	}
	115	if (notify)
	116	on_each_cpu(__crst_table_upgrade, mm, 0);
	117	return rc;
1e133ab2 MS	118	}
1e133ab2 MS	119
723cacbd	120	void crst_table_downgrade(struct mm_struct *mm)
1e133ab2 MS	121	{
	122	pgd_t *pgd;
	123
723cacbd	124	/* downgrade should only happen from 3 to 2 levels (compat only) */
2fc4876e	125	VM_BUG_ON(mm->context.asce_limit != _REGION2_SIZE);
723cacbd	126
1e133ab2 MS	127	if (current->active_mm == mm) {
	128	clear_user_asce();
	129	__tlb_flush_mm(mm);
	130	}
723cacbd GS	131
	132	pgd = mm->pgd;
	133	mm->pgd = (pgd_t ) (pgd_val(pgd) & _REGION_ENTRY_ORIGIN);
f1c1174f	134	mm->context.asce_limit = _REGION3_SIZE;
723cacbd GS	135	mm->context.asce = __pa(mm->pgd) \| _ASCE_TABLE_LENGTH \|
723cacbd GS	136	_ASCE_USER_BITS \| _ASCE_TYPE_SEGMENT;
723cacbd GS	137	crst_table_free(mm, (unsigned long *) pgd);
723cacbd GS	138
1e133ab2 MS	139	if (current->active_mm == mm)
	140	set_user_asce(mm);
	141	}
	142
	143	static inline unsigned int atomic_xor_bits(atomic_t *v, unsigned int bits)
	144	{
	145	unsigned int old, new;
	146
	147	do {
	148	old = atomic_read(v);
	149	new = old ^ bits;
	150	} while (atomic_cmpxchg(v, old, new) != old);
	151	return new;
	152	}
	153
4be130a0 MS	154	#ifdef CONFIG_PGSTE
	155
	156	struct page page_table_alloc_pgste(struct mm_struct mm)
	157	{
	158	struct page *page;
41879ff6	159	u64 *table;
4be130a0	160
faee35a5	161	page = alloc_page(GFP_KERNEL);
4be130a0	162	if (page) {
41879ff6 HC	163	table = (u64 *)page_to_phys(page);
	164	memset64(table, _PAGE_INVALID, PTRS_PER_PTE);
	165	memset64(table + PTRS_PER_PTE, 0, PTRS_PER_PTE);
4be130a0 MS	166	}
	167	return page;
	168	}
	169
	170	void page_table_free_pgste(struct page *page)
	171	{
	172	__free_page(page);
	173	}
	174
	175	#endif /* CONFIG_PGSTE */
	176
1e133ab2 MS	177	/*
	178	* page table entry allocation/free routines.
	179	*/
	180	unsigned long page_table_alloc(struct mm_struct mm)
	181	{
	182	unsigned long *table;
	183	struct page *page;
	184	unsigned int mask, bit;
	185
	186	/* Try to get a fragment of a 4K page as a 2K page table */
	187	if (!mm_alloc_pgste(mm)) {
	188	table = NULL;
f28a4b4d	189	spin_lock_bh(&mm->context.lock);
1e133ab2 MS	190	if (!list_empty(&mm->context.pgtable_list)) {
	191	page = list_first_entry(&mm->context.pgtable_list,
	192	struct page, lru);
	193	mask = atomic_read(&page->_mapcount);
	194	mask = (mask \| (mask >> 4)) & 3;
	195	if (mask != 3) {
	196	table = (unsigned long *) page_to_phys(page);
	197	bit = mask & 1; /* =1 -> second 2K */
	198	if (bit)
	199	table += PTRS_PER_PTE;
	200	atomic_xor_bits(&page->_mapcount, 1U << bit);
	201	list_del(&page->lru);
	202	}
	203	}
f28a4b4d	204	spin_unlock_bh(&mm->context.lock);
1e133ab2 MS	205	if (table)
	206	return table;
	207	}
	208	/* Allocate a fresh page */
10d58bf2	209	page = alloc_page(GFP_KERNEL);
1e133ab2 MS	210	if (!page)
	211	return NULL;
	212	if (!pgtable_page_ctor(page)) {
	213	__free_page(page);
	214	return NULL;
	215	}
c9b5ad54	216	arch_set_page_dat(page, 0);
1e133ab2 MS	217	/* Initialize page table */
	218	table = (unsigned long *) page_to_phys(page);
	219	if (mm_alloc_pgste(mm)) {
	220	/* Return 4K page table with PGSTEs */
	221	atomic_set(&page->_mapcount, 3);
41879ff6 HC	222	memset64((u64 *)table, _PAGE_INVALID, PTRS_PER_PTE);
41879ff6 HC	223	memset64((u64 *)table + PTRS_PER_PTE, 0, PTRS_PER_PTE);
1e133ab2 MS	224	} else {
	225	/* Return the first 2K fragment of the page */
	226	atomic_set(&page->_mapcount, 1);
41879ff6	227	memset64((u64 )table, _PAGE_INVALID, 2 PTRS_PER_PTE);
f28a4b4d	228	spin_lock_bh(&mm->context.lock);
1e133ab2	229	list_add(&page->lru, &mm->context.pgtable_list);
f28a4b4d	230	spin_unlock_bh(&mm->context.lock);
1e133ab2 MS	231	}
	232	return table;
	233	}
	234
	235	void page_table_free(struct mm_struct mm, unsigned long table)
	236	{
	237	struct page *page;
	238	unsigned int bit, mask;
	239
	240	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
	241	if (!mm_alloc_pgste(mm)) {
	242	/* Free 2K page table fragment of a 4K page */
	243	bit = (__pa(table) & ~PAGE_MASK)/(PTRS_PER_PTE*sizeof(pte_t));
f28a4b4d	244	spin_lock_bh(&mm->context.lock);
1e133ab2 MS	245	mask = atomic_xor_bits(&page->_mapcount, 1U << bit);
	246	if (mask & 3)
	247	list_add(&page->lru, &mm->context.pgtable_list);
	248	else
	249	list_del(&page->lru);
f28a4b4d	250	spin_unlock_bh(&mm->context.lock);
1e133ab2 MS	251	if (mask != 0)
	252	return;
	253	}
	254
	255	pgtable_page_dtor(page);
	256	atomic_set(&page->_mapcount, -1);
	257	__free_page(page);
	258	}
	259
	260	void page_table_free_rcu(struct mmu_gather tlb, unsigned long table,
	261	unsigned long vmaddr)
	262	{
	263	struct mm_struct *mm;
	264	struct page *page;
	265	unsigned int bit, mask;
	266
	267	mm = tlb->mm;
	268	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
	269	if (mm_alloc_pgste(mm)) {
	270	gmap_unlink(mm, table, vmaddr);
	271	table = (unsigned long *) (__pa(table) \| 3);
	272	tlb_remove_table(tlb, table);
	273	return;
	274	}
	275	bit = (__pa(table) & ~PAGE_MASK) / (PTRS_PER_PTE*sizeof(pte_t));
f28a4b4d	276	spin_lock_bh(&mm->context.lock);
1e133ab2 MS	277	mask = atomic_xor_bits(&page->_mapcount, 0x11U << bit);
	278	if (mask & 3)
	279	list_add_tail(&page->lru, &mm->context.pgtable_list);
	280	else
	281	list_del(&page->lru);
f28a4b4d	282	spin_unlock_bh(&mm->context.lock);
1e133ab2 MS	283	table = (unsigned long *) (__pa(table) \| (1U << bit));
	284	tlb_remove_table(tlb, table);
	285	}
	286
	287	static void __tlb_remove_table(void *_table)
	288	{
	289	unsigned int mask = (unsigned long) _table & 3;
	290	void table = (void )((unsigned long) _table ^ mask);
	291	struct page *page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
	292
	293	switch (mask) {
1aea9b3f	294	case 0: /* pmd, pud, or p4d */
1e133ab2 MS	295	free_pages((unsigned long) table, 2);
	296	break;
	297	case 1: /* lower 2K of a 4K page table */
	298	case 2: /* higher 2K of a 4K page table */
	299	if (atomic_xor_bits(&page->_mapcount, mask << 4) != 0)
	300	break;
	301	/* fallthrough */
	302	case 3: /* 4K page table with pgstes */
	303	pgtable_page_dtor(page);
	304	atomic_set(&page->_mapcount, -1);
	305	__free_page(page);
	306	break;
	307	}
	308	}
	309
	310	static void tlb_remove_table_smp_sync(void *arg)
	311	{
	312	/* Simply deliver the interrupt */
	313	}
	314
	315	static void tlb_remove_table_one(void *table)
	316	{
	317	/*
	318	* This isn't an RCU grace period and hence the page-tables cannot be
	319	* assumed to be actually RCU-freed.
	320	*
	321	* It is however sufficient for software page-table walkers that rely
	322	* on IRQ disabling. See the comment near struct mmu_table_batch.
	323	*/
	324	smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
	325	__tlb_remove_table(table);
	326	}
	327
	328	static void tlb_remove_table_rcu(struct rcu_head *head)
	329	{
	330	struct mmu_table_batch *batch;
	331	int i;
	332
	333	batch = container_of(head, struct mmu_table_batch, rcu);
	334
	335	for (i = 0; i < batch->nr; i++)
	336	__tlb_remove_table(batch->tables[i]);
	337
	338	free_page((unsigned long)batch);
	339	}
	340
	341	void tlb_table_flush(struct mmu_gather *tlb)
	342	{
	343	struct mmu_table_batch **batch = &tlb->batch;
	344
	345	if (*batch) {
	346	call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
	347	*batch = NULL;
	348	}
	349	}
	350
	351	void tlb_remove_table(struct mmu_gather tlb, void table)
	352	{
	353	struct mmu_table_batch **batch = &tlb->batch;
	354
	355	tlb->mm->context.flush_mm = 1;
	356	if (*batch == NULL) {
	357	batch = (struct mmu_table_batch )
	358	__get_free_page(GFP_NOWAIT \| __GFP_NOWARN);
359	if (*batch == NULL) {
360	__tlb_flush_mm_lazy(tlb->mm);
361	tlb_remove_table_one(table);
362	return;
363	}
364	(*batch)->nr = 0;
365	}
366	(batch)->tables[(batch)->nr++] = table;
367	if ((*batch)->nr == MAX_TABLE_BATCH)
368	tlb_flush_mmu(tlb);
369	}