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

/*
 *  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;
	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) {
		clear_user_asce();
		set_user_asce(mm);
	}
	__tlb_flush_local();
}

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

	/* upgrade should only happen from 3 to 4 levels */
	BUG_ON(mm->context.asce_limit != (1UL << 42));

	table = crst_table_alloc(mm);
	if (!table)
		return -ENOMEM;

	spin_lock_bh(&mm->page_table_lock);
	pgd = (unsigned long *) mm->pgd;
	crst_table_init(table, _REGION2_ENTRY_EMPTY);
	pgd_populate(mm, (pgd_t *) table, (pud_t *) pgd);
	mm->pgd = (pgd_t *) table;
	mm->context.asce_limit = 1UL << 53;
	mm->context.asce = __pa(mm->pgd) | _ASCE_TABLE_LENGTH |
			   _ASCE_USER_BITS | _ASCE_TYPE_REGION2;
	mm->task_size = mm->context.asce_limit;
	spin_unlock_bh(&mm->page_table_lock);

	on_each_cpu(__crst_table_upgrade, mm, 0);
	return 0;
}

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

	/* downgrade should only happen from 3 to 2 levels (compat only) */
	BUG_ON(mm->context.asce_limit != (1UL << 42));

	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 = 1UL << 31;
	mm->context.asce = __pa(mm->pgd) | _ASCE_TABLE_LENGTH |
			   _ASCE_USER_BITS | _ASCE_TYPE_SEGMENT;
	mm->task_size = mm->context.asce_limit;
	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;
}

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