[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,
		.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;
		} 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,
	30	.proc_handler = proc_dointvec,
	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 \|
	102	_ASCE_USER_BITS \| _ASCE_TYPE_REGION2;
	103	} else {
	104	crst_table_init(table, _REGION1_ENTRY_EMPTY);
	105	pgd_populate(mm, (pgd_t ) table, (p4d_t ) pgd);
	106	mm->pgd = (pgd_t *) table;
	107	mm->context.asce_limit = -PAGE_SIZE;
	108	mm->context.asce = __pa(mm->pgd) \| _ASCE_TABLE_LENGTH \|
	109	_ASCE_USER_BITS \| _ASCE_TYPE_REGION1;
	110	}
	111	notify = 1;
	112	spin_unlock_bh(&mm->page_table_lock);
	113	}
	114	if (notify)
	115	on_each_cpu(__crst_table_upgrade, mm, 0);
	116	return rc;
1e133ab2 MS	117	}
1e133ab2 MS	118
723cacbd	119	void crst_table_downgrade(struct mm_struct *mm)
1e133ab2 MS	120	{
	121	pgd_t *pgd;
	122
723cacbd	123	/* downgrade should only happen from 3 to 2 levels (compat only) */
2fc4876e	124	VM_BUG_ON(mm->context.asce_limit != _REGION2_SIZE);
723cacbd	125
1e133ab2 MS	126	if (current->active_mm == mm) {
	127	clear_user_asce();
	128	__tlb_flush_mm(mm);
	129	}
723cacbd GS	130
	131	pgd = mm->pgd;
	132	mm->pgd = (pgd_t ) (pgd_val(pgd) & _REGION_ENTRY_ORIGIN);
f1c1174f	133	mm->context.asce_limit = _REGION3_SIZE;
723cacbd GS	134	mm->context.asce = __pa(mm->pgd) \| _ASCE_TABLE_LENGTH \|
723cacbd GS	135	_ASCE_USER_BITS \| _ASCE_TYPE_SEGMENT;
723cacbd GS	136	crst_table_free(mm, (unsigned long *) pgd);
723cacbd GS	137
1e133ab2 MS	138	if (current->active_mm == mm)
	139	set_user_asce(mm);
	140	}
	141
	142	static inline unsigned int atomic_xor_bits(atomic_t *v, unsigned int bits)
	143	{
	144	unsigned int old, new;
	145
	146	do {
	147	old = atomic_read(v);
	148	new = old ^ bits;
	149	} while (atomic_cmpxchg(v, old, new) != old);
	150	return new;
	151	}
	152
4be130a0 MS	153	#ifdef CONFIG_PGSTE
	154
	155	struct page page_table_alloc_pgste(struct mm_struct mm)
	156	{
	157	struct page *page;
41879ff6	158	u64 *table;
4be130a0	159
faee35a5	160	page = alloc_page(GFP_KERNEL);
4be130a0	161	if (page) {
41879ff6 HC	162	table = (u64 *)page_to_phys(page);
	163	memset64(table, _PAGE_INVALID, PTRS_PER_PTE);
	164	memset64(table + PTRS_PER_PTE, 0, PTRS_PER_PTE);
4be130a0 MS	165	}
	166	return page;
	167	}
	168
	169	void page_table_free_pgste(struct page *page)
	170	{
	171	__free_page(page);
	172	}
	173
	174	#endif /* CONFIG_PGSTE */
	175
1e133ab2 MS	176	/*
	177	* page table entry allocation/free routines.
	178	*/
	179	unsigned long page_table_alloc(struct mm_struct mm)
	180	{
	181	unsigned long *table;
	182	struct page *page;
	183	unsigned int mask, bit;
	184
	185	/* Try to get a fragment of a 4K page as a 2K page table */
	186	if (!mm_alloc_pgste(mm)) {
	187	table = NULL;
f28a4b4d	188	spin_lock_bh(&mm->context.lock);
1e133ab2 MS	189	if (!list_empty(&mm->context.pgtable_list)) {
	190	page = list_first_entry(&mm->context.pgtable_list,
	191	struct page, lru);
	192	mask = atomic_read(&page->_mapcount);
	193	mask = (mask \| (mask >> 4)) & 3;
	194	if (mask != 3) {
	195	table = (unsigned long *) page_to_phys(page);
	196	bit = mask & 1; /* =1 -> second 2K */
	197	if (bit)
	198	table += PTRS_PER_PTE;
	199	atomic_xor_bits(&page->_mapcount, 1U << bit);
	200	list_del(&page->lru);
	201	}
	202	}
f28a4b4d	203	spin_unlock_bh(&mm->context.lock);
1e133ab2 MS	204	if (table)
	205	return table;
	206	}
	207	/* Allocate a fresh page */
10d58bf2	208	page = alloc_page(GFP_KERNEL);
1e133ab2 MS	209	if (!page)
	210	return NULL;
	211	if (!pgtable_page_ctor(page)) {
	212	__free_page(page);
	213	return NULL;
	214	}
c9b5ad54	215	arch_set_page_dat(page, 0);
1e133ab2 MS	216	/* Initialize page table */
	217	table = (unsigned long *) page_to_phys(page);
	218	if (mm_alloc_pgste(mm)) {
	219	/* Return 4K page table with PGSTEs */
	220	atomic_set(&page->_mapcount, 3);
41879ff6 HC	221	memset64((u64 *)table, _PAGE_INVALID, PTRS_PER_PTE);
41879ff6 HC	222	memset64((u64 *)table + PTRS_PER_PTE, 0, PTRS_PER_PTE);
1e133ab2 MS	223	} else {
	224	/* Return the first 2K fragment of the page */
	225	atomic_set(&page->_mapcount, 1);
41879ff6	226	memset64((u64 )table, _PAGE_INVALID, 2 PTRS_PER_PTE);
f28a4b4d	227	spin_lock_bh(&mm->context.lock);
1e133ab2	228	list_add(&page->lru, &mm->context.pgtable_list);
f28a4b4d	229	spin_unlock_bh(&mm->context.lock);
1e133ab2 MS	230	}
	231	return table;
	232	}
	233
	234	void page_table_free(struct mm_struct mm, unsigned long table)
	235	{
	236	struct page *page;
	237	unsigned int bit, mask;
	238
	239	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
	240	if (!mm_alloc_pgste(mm)) {
	241	/* Free 2K page table fragment of a 4K page */
	242	bit = (__pa(table) & ~PAGE_MASK)/(PTRS_PER_PTE*sizeof(pte_t));
f28a4b4d	243	spin_lock_bh(&mm->context.lock);
1e133ab2 MS	244	mask = atomic_xor_bits(&page->_mapcount, 1U << bit);
	245	if (mask & 3)
	246	list_add(&page->lru, &mm->context.pgtable_list);
	247	else
	248	list_del(&page->lru);
f28a4b4d	249	spin_unlock_bh(&mm->context.lock);
1e133ab2 MS	250	if (mask != 0)
	251	return;
	252	}
	253
	254	pgtable_page_dtor(page);
	255	atomic_set(&page->_mapcount, -1);
	256	__free_page(page);
	257	}
	258
	259	void page_table_free_rcu(struct mmu_gather tlb, unsigned long table,
	260	unsigned long vmaddr)
	261	{
	262	struct mm_struct *mm;
	263	struct page *page;
	264	unsigned int bit, mask;
	265
	266	mm = tlb->mm;
	267	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
	268	if (mm_alloc_pgste(mm)) {
	269	gmap_unlink(mm, table, vmaddr);
	270	table = (unsigned long *) (__pa(table) \| 3);
	271	tlb_remove_table(tlb, table);
	272	return;
	273	}
	274	bit = (__pa(table) & ~PAGE_MASK) / (PTRS_PER_PTE*sizeof(pte_t));
f28a4b4d	275	spin_lock_bh(&mm->context.lock);
1e133ab2 MS	276	mask = atomic_xor_bits(&page->_mapcount, 0x11U << bit);
	277	if (mask & 3)
	278	list_add_tail(&page->lru, &mm->context.pgtable_list);
	279	else
	280	list_del(&page->lru);
f28a4b4d	281	spin_unlock_bh(&mm->context.lock);
1e133ab2 MS	282	table = (unsigned long *) (__pa(table) \| (1U << bit));
	283	tlb_remove_table(tlb, table);
	284	}
	285
	286	static void __tlb_remove_table(void *_table)
	287	{
	288	unsigned int mask = (unsigned long) _table & 3;
	289	void table = (void )((unsigned long) _table ^ mask);
	290	struct page *page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
	291
	292	switch (mask) {
1aea9b3f	293	case 0: /* pmd, pud, or p4d */
1e133ab2 MS	294	free_pages((unsigned long) table, 2);
	295	break;
	296	case 1: /* lower 2K of a 4K page table */
	297	case 2: /* higher 2K of a 4K page table */
	298	if (atomic_xor_bits(&page->_mapcount, mask << 4) != 0)
	299	break;
	300	/* fallthrough */
	301	case 3: /* 4K page table with pgstes */
	302	pgtable_page_dtor(page);
	303	atomic_set(&page->_mapcount, -1);
	304	__free_page(page);
	305	break;
	306	}
	307	}
	308
	309	static void tlb_remove_table_smp_sync(void *arg)
	310	{
	311	/* Simply deliver the interrupt */
	312	}
	313
	314	static void tlb_remove_table_one(void *table)
	315	{
	316	/*
	317	* This isn't an RCU grace period and hence the page-tables cannot be
	318	* assumed to be actually RCU-freed.
	319	*
	320	* It is however sufficient for software page-table walkers that rely
	321	* on IRQ disabling. See the comment near struct mmu_table_batch.
	322	*/
	323	smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
	324	__tlb_remove_table(table);
	325	}
	326
	327	static void tlb_remove_table_rcu(struct rcu_head *head)
	328	{
	329	struct mmu_table_batch *batch;
	330	int i;
	331
	332	batch = container_of(head, struct mmu_table_batch, rcu);
	333
	334	for (i = 0; i < batch->nr; i++)
	335	__tlb_remove_table(batch->tables[i]);
	336
	337	free_page((unsigned long)batch);
	338	}
	339
	340	void tlb_table_flush(struct mmu_gather *tlb)
	341	{
	342	struct mmu_table_batch **batch = &tlb->batch;
	343
	344	if (*batch) {
	345	call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
	346	*batch = NULL;
	347	}
	348	}
	349
	350	void tlb_remove_table(struct mmu_gather tlb, void table)
	351	{
	352	struct mmu_table_batch **batch = &tlb->batch;
	353
	354	tlb->mm->context.flush_mm = 1;
	355	if (*batch == NULL) {
	356	batch = (struct mmu_table_batch )
	357	__get_free_page(GFP_NOWAIT \| __GFP_NOWARN);
358	if (*batch == NULL) {
359	__tlb_flush_mm_lazy(tlb->mm);
360	tlb_remove_table_one(table);
361	return;
362	}
363	(*batch)->nr = 0;
364	}
365	(batch)->tables[(batch)->nr++] = table;
366	if ((*batch)->nr == MAX_TABLE_BATCH)
367	tlb_flush_mmu(tlb);
368	}