xref: /openbmc/linux/arch/s390/mm/pgtable.c (revision b5266ea6)
1 /*
2  *    Copyright IBM Corp. 2007,2011
3  *    Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
4  */
5 
6 #include <linux/sched.h>
7 #include <linux/kernel.h>
8 #include <linux/errno.h>
9 #include <linux/gfp.h>
10 #include <linux/mm.h>
11 #include <linux/swap.h>
12 #include <linux/smp.h>
13 #include <linux/highmem.h>
14 #include <linux/pagemap.h>
15 #include <linux/spinlock.h>
16 #include <linux/module.h>
17 #include <linux/quicklist.h>
18 #include <linux/rcupdate.h>
19 #include <linux/slab.h>
20 
21 #include <asm/system.h>
22 #include <asm/pgtable.h>
23 #include <asm/pgalloc.h>
24 #include <asm/tlb.h>
25 #include <asm/tlbflush.h>
26 #include <asm/mmu_context.h>
27 
28 #ifndef CONFIG_64BIT
29 #define ALLOC_ORDER	1
30 #define FRAG_MASK	0x0f
31 #else
32 #define ALLOC_ORDER	2
33 #define FRAG_MASK	0x03
34 #endif
35 
36 
37 unsigned long *crst_table_alloc(struct mm_struct *mm)
38 {
39 	struct page *page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
40 
41 	if (!page)
42 		return NULL;
43 	return (unsigned long *) page_to_phys(page);
44 }
45 
46 void crst_table_free(struct mm_struct *mm, unsigned long *table)
47 {
48 	free_pages((unsigned long) table, ALLOC_ORDER);
49 }
50 
51 #ifdef CONFIG_64BIT
52 int crst_table_upgrade(struct mm_struct *mm, unsigned long limit)
53 {
54 	unsigned long *table, *pgd;
55 	unsigned long entry;
56 
57 	BUG_ON(limit > (1UL << 53));
58 repeat:
59 	table = crst_table_alloc(mm);
60 	if (!table)
61 		return -ENOMEM;
62 	spin_lock_bh(&mm->page_table_lock);
63 	if (mm->context.asce_limit < limit) {
64 		pgd = (unsigned long *) mm->pgd;
65 		if (mm->context.asce_limit <= (1UL << 31)) {
66 			entry = _REGION3_ENTRY_EMPTY;
67 			mm->context.asce_limit = 1UL << 42;
68 			mm->context.asce_bits = _ASCE_TABLE_LENGTH |
69 						_ASCE_USER_BITS |
70 						_ASCE_TYPE_REGION3;
71 		} else {
72 			entry = _REGION2_ENTRY_EMPTY;
73 			mm->context.asce_limit = 1UL << 53;
74 			mm->context.asce_bits = _ASCE_TABLE_LENGTH |
75 						_ASCE_USER_BITS |
76 						_ASCE_TYPE_REGION2;
77 		}
78 		crst_table_init(table, entry);
79 		pgd_populate(mm, (pgd_t *) table, (pud_t *) pgd);
80 		mm->pgd = (pgd_t *) table;
81 		mm->task_size = mm->context.asce_limit;
82 		table = NULL;
83 	}
84 	spin_unlock_bh(&mm->page_table_lock);
85 	if (table)
86 		crst_table_free(mm, table);
87 	if (mm->context.asce_limit < limit)
88 		goto repeat;
89 	update_mm(mm, current);
90 	return 0;
91 }
92 
93 void crst_table_downgrade(struct mm_struct *mm, unsigned long limit)
94 {
95 	pgd_t *pgd;
96 
97 	if (mm->context.asce_limit <= limit)
98 		return;
99 	__tlb_flush_mm(mm);
100 	while (mm->context.asce_limit > limit) {
101 		pgd = mm->pgd;
102 		switch (pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) {
103 		case _REGION_ENTRY_TYPE_R2:
104 			mm->context.asce_limit = 1UL << 42;
105 			mm->context.asce_bits = _ASCE_TABLE_LENGTH |
106 						_ASCE_USER_BITS |
107 						_ASCE_TYPE_REGION3;
108 			break;
109 		case _REGION_ENTRY_TYPE_R3:
110 			mm->context.asce_limit = 1UL << 31;
111 			mm->context.asce_bits = _ASCE_TABLE_LENGTH |
112 						_ASCE_USER_BITS |
113 						_ASCE_TYPE_SEGMENT;
114 			break;
115 		default:
116 			BUG();
117 		}
118 		mm->pgd = (pgd_t *) (pgd_val(*pgd) & _REGION_ENTRY_ORIGIN);
119 		mm->task_size = mm->context.asce_limit;
120 		crst_table_free(mm, (unsigned long *) pgd);
121 	}
122 	update_mm(mm, current);
123 }
124 #endif
125 
126 #ifdef CONFIG_PGSTE
127 
128 /**
129  * gmap_alloc - allocate a guest address space
130  * @mm: pointer to the parent mm_struct
131  *
132  * Returns a guest address space structure.
133  */
134 struct gmap *gmap_alloc(struct mm_struct *mm)
135 {
136 	struct gmap *gmap;
137 	struct page *page;
138 	unsigned long *table;
139 
140 	gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL);
141 	if (!gmap)
142 		goto out;
143 	INIT_LIST_HEAD(&gmap->crst_list);
144 	gmap->mm = mm;
145 	page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
146 	if (!page)
147 		goto out_free;
148 	list_add(&page->lru, &gmap->crst_list);
149 	table = (unsigned long *) page_to_phys(page);
150 	crst_table_init(table, _REGION1_ENTRY_EMPTY);
151 	gmap->table = table;
152 	gmap->asce = _ASCE_TYPE_REGION1 | _ASCE_TABLE_LENGTH |
153 		     _ASCE_USER_BITS | __pa(table);
154 	list_add(&gmap->list, &mm->context.gmap_list);
155 	return gmap;
156 
157 out_free:
158 	kfree(gmap);
159 out:
160 	return NULL;
161 }
162 EXPORT_SYMBOL_GPL(gmap_alloc);
163 
164 static int gmap_unlink_segment(struct gmap *gmap, unsigned long *table)
165 {
166 	struct gmap_pgtable *mp;
167 	struct gmap_rmap *rmap;
168 	struct page *page;
169 
170 	if (*table & _SEGMENT_ENTRY_INV)
171 		return 0;
172 	page = pfn_to_page(*table >> PAGE_SHIFT);
173 	mp = (struct gmap_pgtable *) page->index;
174 	list_for_each_entry(rmap, &mp->mapper, list) {
175 		if (rmap->entry != table)
176 			continue;
177 		list_del(&rmap->list);
178 		kfree(rmap);
179 		break;
180 	}
181 	*table = _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | mp->vmaddr;
182 	return 1;
183 }
184 
185 static void gmap_flush_tlb(struct gmap *gmap)
186 {
187 	if (MACHINE_HAS_IDTE)
188 		__tlb_flush_idte((unsigned long) gmap->table |
189 				 _ASCE_TYPE_REGION1);
190 	else
191 		__tlb_flush_global();
192 }
193 
194 /**
195  * gmap_free - free a guest address space
196  * @gmap: pointer to the guest address space structure
197  */
198 void gmap_free(struct gmap *gmap)
199 {
200 	struct page *page, *next;
201 	unsigned long *table;
202 	int i;
203 
204 
205 	/* Flush tlb. */
206 	if (MACHINE_HAS_IDTE)
207 		__tlb_flush_idte((unsigned long) gmap->table |
208 				 _ASCE_TYPE_REGION1);
209 	else
210 		__tlb_flush_global();
211 
212 	/* Free all segment & region tables. */
213 	down_read(&gmap->mm->mmap_sem);
214 	spin_lock(&gmap->mm->page_table_lock);
215 	list_for_each_entry_safe(page, next, &gmap->crst_list, lru) {
216 		table = (unsigned long *) page_to_phys(page);
217 		if ((*table & _REGION_ENTRY_TYPE_MASK) == 0)
218 			/* Remove gmap rmap structures for segment table. */
219 			for (i = 0; i < PTRS_PER_PMD; i++, table++)
220 				gmap_unlink_segment(gmap, table);
221 		__free_pages(page, ALLOC_ORDER);
222 	}
223 	spin_unlock(&gmap->mm->page_table_lock);
224 	up_read(&gmap->mm->mmap_sem);
225 	list_del(&gmap->list);
226 	kfree(gmap);
227 }
228 EXPORT_SYMBOL_GPL(gmap_free);
229 
230 /**
231  * gmap_enable - switch primary space to the guest address space
232  * @gmap: pointer to the guest address space structure
233  */
234 void gmap_enable(struct gmap *gmap)
235 {
236 	S390_lowcore.gmap = (unsigned long) gmap;
237 }
238 EXPORT_SYMBOL_GPL(gmap_enable);
239 
240 /**
241  * gmap_disable - switch back to the standard primary address space
242  * @gmap: pointer to the guest address space structure
243  */
244 void gmap_disable(struct gmap *gmap)
245 {
246 	S390_lowcore.gmap = 0UL;
247 }
248 EXPORT_SYMBOL_GPL(gmap_disable);
249 
250 /*
251  * gmap_alloc_table is assumed to be called with mmap_sem held
252  */
253 static int gmap_alloc_table(struct gmap *gmap,
254 			       unsigned long *table, unsigned long init)
255 {
256 	struct page *page;
257 	unsigned long *new;
258 
259 	/* since we dont free the gmap table until gmap_free we can unlock */
260 	spin_unlock(&gmap->mm->page_table_lock);
261 	page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
262 	spin_lock(&gmap->mm->page_table_lock);
263 	if (!page)
264 		return -ENOMEM;
265 	new = (unsigned long *) page_to_phys(page);
266 	crst_table_init(new, init);
267 	if (*table & _REGION_ENTRY_INV) {
268 		list_add(&page->lru, &gmap->crst_list);
269 		*table = (unsigned long) new | _REGION_ENTRY_LENGTH |
270 			(*table & _REGION_ENTRY_TYPE_MASK);
271 	} else
272 		__free_pages(page, ALLOC_ORDER);
273 	return 0;
274 }
275 
276 /**
277  * gmap_unmap_segment - unmap segment from the guest address space
278  * @gmap: pointer to the guest address space structure
279  * @addr: address in the guest address space
280  * @len: length of the memory area to unmap
281  *
282  * Returns 0 if the unmap succeded, -EINVAL if not.
283  */
284 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
285 {
286 	unsigned long *table;
287 	unsigned long off;
288 	int flush;
289 
290 	if ((to | len) & (PMD_SIZE - 1))
291 		return -EINVAL;
292 	if (len == 0 || to + len < to)
293 		return -EINVAL;
294 
295 	flush = 0;
296 	down_read(&gmap->mm->mmap_sem);
297 	spin_lock(&gmap->mm->page_table_lock);
298 	for (off = 0; off < len; off += PMD_SIZE) {
299 		/* Walk the guest addr space page table */
300 		table = gmap->table + (((to + off) >> 53) & 0x7ff);
301 		if (*table & _REGION_ENTRY_INV)
302 			goto out;
303 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
304 		table = table + (((to + off) >> 42) & 0x7ff);
305 		if (*table & _REGION_ENTRY_INV)
306 			goto out;
307 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
308 		table = table + (((to + off) >> 31) & 0x7ff);
309 		if (*table & _REGION_ENTRY_INV)
310 			goto out;
311 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
312 		table = table + (((to + off) >> 20) & 0x7ff);
313 
314 		/* Clear segment table entry in guest address space. */
315 		flush |= gmap_unlink_segment(gmap, table);
316 		*table = _SEGMENT_ENTRY_INV;
317 	}
318 out:
319 	spin_unlock(&gmap->mm->page_table_lock);
320 	up_read(&gmap->mm->mmap_sem);
321 	if (flush)
322 		gmap_flush_tlb(gmap);
323 	return 0;
324 }
325 EXPORT_SYMBOL_GPL(gmap_unmap_segment);
326 
327 /**
328  * gmap_mmap_segment - map a segment to the guest address space
329  * @gmap: pointer to the guest address space structure
330  * @from: source address in the parent address space
331  * @to: target address in the guest address space
332  *
333  * Returns 0 if the mmap succeded, -EINVAL or -ENOMEM if not.
334  */
335 int gmap_map_segment(struct gmap *gmap, unsigned long from,
336 		     unsigned long to, unsigned long len)
337 {
338 	unsigned long *table;
339 	unsigned long off;
340 	int flush;
341 
342 	if ((from | to | len) & (PMD_SIZE - 1))
343 		return -EINVAL;
344 	if (len == 0 || from + len > PGDIR_SIZE ||
345 	    from + len < from || to + len < to)
346 		return -EINVAL;
347 
348 	flush = 0;
349 	down_read(&gmap->mm->mmap_sem);
350 	spin_lock(&gmap->mm->page_table_lock);
351 	for (off = 0; off < len; off += PMD_SIZE) {
352 		/* Walk the gmap address space page table */
353 		table = gmap->table + (((to + off) >> 53) & 0x7ff);
354 		if ((*table & _REGION_ENTRY_INV) &&
355 		    gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY))
356 			goto out_unmap;
357 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
358 		table = table + (((to + off) >> 42) & 0x7ff);
359 		if ((*table & _REGION_ENTRY_INV) &&
360 		    gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY))
361 			goto out_unmap;
362 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
363 		table = table + (((to + off) >> 31) & 0x7ff);
364 		if ((*table & _REGION_ENTRY_INV) &&
365 		    gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY))
366 			goto out_unmap;
367 		table = (unsigned long *) (*table & _REGION_ENTRY_ORIGIN);
368 		table = table + (((to + off) >> 20) & 0x7ff);
369 
370 		/* Store 'from' address in an invalid segment table entry. */
371 		flush |= gmap_unlink_segment(gmap, table);
372 		*table = _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | (from + off);
373 	}
374 	spin_unlock(&gmap->mm->page_table_lock);
375 	up_read(&gmap->mm->mmap_sem);
376 	if (flush)
377 		gmap_flush_tlb(gmap);
378 	return 0;
379 
380 out_unmap:
381 	spin_unlock(&gmap->mm->page_table_lock);
382 	up_read(&gmap->mm->mmap_sem);
383 	gmap_unmap_segment(gmap, to, len);
384 	return -ENOMEM;
385 }
386 EXPORT_SYMBOL_GPL(gmap_map_segment);
387 
388 /*
389  * this function is assumed to be called with mmap_sem held
390  */
391 unsigned long __gmap_fault(unsigned long address, struct gmap *gmap)
392 {
393 	unsigned long *table, vmaddr, segment;
394 	struct mm_struct *mm;
395 	struct gmap_pgtable *mp;
396 	struct gmap_rmap *rmap;
397 	struct vm_area_struct *vma;
398 	struct page *page;
399 	pgd_t *pgd;
400 	pud_t *pud;
401 	pmd_t *pmd;
402 
403 	current->thread.gmap_addr = address;
404 	mm = gmap->mm;
405 	/* Walk the gmap address space page table */
406 	table = gmap->table + ((address >> 53) & 0x7ff);
407 	if (unlikely(*table & _REGION_ENTRY_INV))
408 		return -EFAULT;
409 	table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
410 	table = table + ((address >> 42) & 0x7ff);
411 	if (unlikely(*table & _REGION_ENTRY_INV))
412 		return -EFAULT;
413 	table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
414 	table = table + ((address >> 31) & 0x7ff);
415 	if (unlikely(*table & _REGION_ENTRY_INV))
416 		return -EFAULT;
417 	table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
418 	table = table + ((address >> 20) & 0x7ff);
419 
420 	/* Convert the gmap address to an mm address. */
421 	segment = *table;
422 	if (likely(!(segment & _SEGMENT_ENTRY_INV))) {
423 		page = pfn_to_page(segment >> PAGE_SHIFT);
424 		mp = (struct gmap_pgtable *) page->index;
425 		return mp->vmaddr | (address & ~PMD_MASK);
426 	} else if (segment & _SEGMENT_ENTRY_RO) {
427 		vmaddr = segment & _SEGMENT_ENTRY_ORIGIN;
428 		vma = find_vma(mm, vmaddr);
429 		if (!vma || vma->vm_start > vmaddr)
430 			return -EFAULT;
431 
432 		/* Walk the parent mm page table */
433 		pgd = pgd_offset(mm, vmaddr);
434 		pud = pud_alloc(mm, pgd, vmaddr);
435 		if (!pud)
436 			return -ENOMEM;
437 		pmd = pmd_alloc(mm, pud, vmaddr);
438 		if (!pmd)
439 			return -ENOMEM;
440 		if (!pmd_present(*pmd) &&
441 		    __pte_alloc(mm, vma, pmd, vmaddr))
442 			return -ENOMEM;
443 		/* pmd now points to a valid segment table entry. */
444 		rmap = kmalloc(sizeof(*rmap), GFP_KERNEL|__GFP_REPEAT);
445 		if (!rmap)
446 			return -ENOMEM;
447 		/* Link gmap segment table entry location to page table. */
448 		page = pmd_page(*pmd);
449 		mp = (struct gmap_pgtable *) page->index;
450 		rmap->entry = table;
451 		spin_lock(&mm->page_table_lock);
452 		list_add(&rmap->list, &mp->mapper);
453 		spin_unlock(&mm->page_table_lock);
454 		/* Set gmap segment table entry to page table. */
455 		*table = pmd_val(*pmd) & PAGE_MASK;
456 		return vmaddr | (address & ~PMD_MASK);
457 	}
458 	return -EFAULT;
459 }
460 
461 unsigned long gmap_fault(unsigned long address, struct gmap *gmap)
462 {
463 	unsigned long rc;
464 
465 	down_read(&gmap->mm->mmap_sem);
466 	rc = __gmap_fault(address, gmap);
467 	up_read(&gmap->mm->mmap_sem);
468 
469 	return rc;
470 }
471 EXPORT_SYMBOL_GPL(gmap_fault);
472 
473 void gmap_discard(unsigned long from, unsigned long to, struct gmap *gmap)
474 {
475 
476 	unsigned long *table, address, size;
477 	struct vm_area_struct *vma;
478 	struct gmap_pgtable *mp;
479 	struct page *page;
480 
481 	down_read(&gmap->mm->mmap_sem);
482 	address = from;
483 	while (address < to) {
484 		/* Walk the gmap address space page table */
485 		table = gmap->table + ((address >> 53) & 0x7ff);
486 		if (unlikely(*table & _REGION_ENTRY_INV)) {
487 			address = (address + PMD_SIZE) & PMD_MASK;
488 			continue;
489 		}
490 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
491 		table = table + ((address >> 42) & 0x7ff);
492 		if (unlikely(*table & _REGION_ENTRY_INV)) {
493 			address = (address + PMD_SIZE) & PMD_MASK;
494 			continue;
495 		}
496 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
497 		table = table + ((address >> 31) & 0x7ff);
498 		if (unlikely(*table & _REGION_ENTRY_INV)) {
499 			address = (address + PMD_SIZE) & PMD_MASK;
500 			continue;
501 		}
502 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
503 		table = table + ((address >> 20) & 0x7ff);
504 		if (unlikely(*table & _SEGMENT_ENTRY_INV)) {
505 			address = (address + PMD_SIZE) & PMD_MASK;
506 			continue;
507 		}
508 		page = pfn_to_page(*table >> PAGE_SHIFT);
509 		mp = (struct gmap_pgtable *) page->index;
510 		vma = find_vma(gmap->mm, mp->vmaddr);
511 		size = min(to - address, PMD_SIZE - (address & ~PMD_MASK));
512 		zap_page_range(vma, mp->vmaddr | (address & ~PMD_MASK),
513 			       size, NULL);
514 		address = (address + PMD_SIZE) & PMD_MASK;
515 	}
516 	up_read(&gmap->mm->mmap_sem);
517 }
518 EXPORT_SYMBOL_GPL(gmap_discard);
519 
520 void gmap_unmap_notifier(struct mm_struct *mm, unsigned long *table)
521 {
522 	struct gmap_rmap *rmap, *next;
523 	struct gmap_pgtable *mp;
524 	struct page *page;
525 	int flush;
526 
527 	flush = 0;
528 	spin_lock(&mm->page_table_lock);
529 	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
530 	mp = (struct gmap_pgtable *) page->index;
531 	list_for_each_entry_safe(rmap, next, &mp->mapper, list) {
532 		*rmap->entry =
533 			_SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | mp->vmaddr;
534 		list_del(&rmap->list);
535 		kfree(rmap);
536 		flush = 1;
537 	}
538 	spin_unlock(&mm->page_table_lock);
539 	if (flush)
540 		__tlb_flush_global();
541 }
542 
543 static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm,
544 						    unsigned long vmaddr)
545 {
546 	struct page *page;
547 	unsigned long *table;
548 	struct gmap_pgtable *mp;
549 
550 	page = alloc_page(GFP_KERNEL|__GFP_REPEAT);
551 	if (!page)
552 		return NULL;
553 	mp = kmalloc(sizeof(*mp), GFP_KERNEL|__GFP_REPEAT);
554 	if (!mp) {
555 		__free_page(page);
556 		return NULL;
557 	}
558 	pgtable_page_ctor(page);
559 	mp->vmaddr = vmaddr & PMD_MASK;
560 	INIT_LIST_HEAD(&mp->mapper);
561 	page->index = (unsigned long) mp;
562 	atomic_set(&page->_mapcount, 3);
563 	table = (unsigned long *) page_to_phys(page);
564 	clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE/2);
565 	clear_table(table + PTRS_PER_PTE, 0, PAGE_SIZE/2);
566 	return table;
567 }
568 
569 static inline void page_table_free_pgste(unsigned long *table)
570 {
571 	struct page *page;
572 	struct gmap_pgtable *mp;
573 
574 	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
575 	mp = (struct gmap_pgtable *) page->index;
576 	BUG_ON(!list_empty(&mp->mapper));
577 	pgtable_page_ctor(page);
578 	atomic_set(&page->_mapcount, -1);
579 	kfree(mp);
580 	__free_page(page);
581 }
582 
583 #else /* CONFIG_PGSTE */
584 
585 static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm,
586 						    unsigned long vmaddr)
587 {
588 	return NULL;
589 }
590 
591 static inline void page_table_free_pgste(unsigned long *table)
592 {
593 }
594 
595 static inline void gmap_unmap_notifier(struct mm_struct *mm,
596 					  unsigned long *table)
597 {
598 }
599 
600 #endif /* CONFIG_PGSTE */
601 
602 static inline unsigned int atomic_xor_bits(atomic_t *v, unsigned int bits)
603 {
604 	unsigned int old, new;
605 
606 	do {
607 		old = atomic_read(v);
608 		new = old ^ bits;
609 	} while (atomic_cmpxchg(v, old, new) != old);
610 	return new;
611 }
612 
613 /*
614  * page table entry allocation/free routines.
615  */
616 unsigned long *page_table_alloc(struct mm_struct *mm, unsigned long vmaddr)
617 {
618 	struct page *page;
619 	unsigned long *table;
620 	unsigned int mask, bit;
621 
622 	if (mm_has_pgste(mm))
623 		return page_table_alloc_pgste(mm, vmaddr);
624 	/* Allocate fragments of a 4K page as 1K/2K page table */
625 	spin_lock_bh(&mm->context.list_lock);
626 	mask = FRAG_MASK;
627 	if (!list_empty(&mm->context.pgtable_list)) {
628 		page = list_first_entry(&mm->context.pgtable_list,
629 					struct page, lru);
630 		table = (unsigned long *) page_to_phys(page);
631 		mask = atomic_read(&page->_mapcount);
632 		mask = mask | (mask >> 4);
633 	}
634 	if ((mask & FRAG_MASK) == FRAG_MASK) {
635 		spin_unlock_bh(&mm->context.list_lock);
636 		page = alloc_page(GFP_KERNEL|__GFP_REPEAT);
637 		if (!page)
638 			return NULL;
639 		pgtable_page_ctor(page);
640 		atomic_set(&page->_mapcount, 1);
641 		table = (unsigned long *) page_to_phys(page);
642 		clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE);
643 		spin_lock_bh(&mm->context.list_lock);
644 		list_add(&page->lru, &mm->context.pgtable_list);
645 	} else {
646 		for (bit = 1; mask & bit; bit <<= 1)
647 			table += PTRS_PER_PTE;
648 		mask = atomic_xor_bits(&page->_mapcount, bit);
649 		if ((mask & FRAG_MASK) == FRAG_MASK)
650 			list_del(&page->lru);
651 	}
652 	spin_unlock_bh(&mm->context.list_lock);
653 	return table;
654 }
655 
656 void page_table_free(struct mm_struct *mm, unsigned long *table)
657 {
658 	struct page *page;
659 	unsigned int bit, mask;
660 
661 	if (mm_has_pgste(mm)) {
662 		gmap_unmap_notifier(mm, table);
663 		return page_table_free_pgste(table);
664 	}
665 	/* Free 1K/2K page table fragment of a 4K page */
666 	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
667 	bit = 1 << ((__pa(table) & ~PAGE_MASK)/(PTRS_PER_PTE*sizeof(pte_t)));
668 	spin_lock_bh(&mm->context.list_lock);
669 	if ((atomic_read(&page->_mapcount) & FRAG_MASK) != FRAG_MASK)
670 		list_del(&page->lru);
671 	mask = atomic_xor_bits(&page->_mapcount, bit);
672 	if (mask & FRAG_MASK)
673 		list_add(&page->lru, &mm->context.pgtable_list);
674 	spin_unlock_bh(&mm->context.list_lock);
675 	if (mask == 0) {
676 		pgtable_page_dtor(page);
677 		atomic_set(&page->_mapcount, -1);
678 		__free_page(page);
679 	}
680 }
681 
682 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
683 
684 static void __page_table_free_rcu(void *table, unsigned bit)
685 {
686 	struct page *page;
687 
688 	if (bit == FRAG_MASK)
689 		return page_table_free_pgste(table);
690 	/* Free 1K/2K page table fragment of a 4K page */
691 	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
692 	if (atomic_xor_bits(&page->_mapcount, bit) == 0) {
693 		pgtable_page_dtor(page);
694 		atomic_set(&page->_mapcount, -1);
695 		__free_page(page);
696 	}
697 }
698 
699 void page_table_free_rcu(struct mmu_gather *tlb, unsigned long *table)
700 {
701 	struct mm_struct *mm;
702 	struct page *page;
703 	unsigned int bit, mask;
704 
705 	mm = tlb->mm;
706 	if (mm_has_pgste(mm)) {
707 		gmap_unmap_notifier(mm, table);
708 		table = (unsigned long *) (__pa(table) | FRAG_MASK);
709 		tlb_remove_table(tlb, table);
710 		return;
711 	}
712 	bit = 1 << ((__pa(table) & ~PAGE_MASK) / (PTRS_PER_PTE*sizeof(pte_t)));
713 	page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
714 	spin_lock_bh(&mm->context.list_lock);
715 	if ((atomic_read(&page->_mapcount) & FRAG_MASK) != FRAG_MASK)
716 		list_del(&page->lru);
717 	mask = atomic_xor_bits(&page->_mapcount, bit | (bit << 4));
718 	if (mask & FRAG_MASK)
719 		list_add_tail(&page->lru, &mm->context.pgtable_list);
720 	spin_unlock_bh(&mm->context.list_lock);
721 	table = (unsigned long *) (__pa(table) | (bit << 4));
722 	tlb_remove_table(tlb, table);
723 }
724 
725 void __tlb_remove_table(void *_table)
726 {
727 	const unsigned long mask = (FRAG_MASK << 4) | FRAG_MASK;
728 	void *table = (void *)((unsigned long) _table & ~mask);
729 	unsigned type = (unsigned long) _table & mask;
730 
731 	if (type)
732 		__page_table_free_rcu(table, type);
733 	else
734 		free_pages((unsigned long) table, ALLOC_ORDER);
735 }
736 
737 #endif
738 
739 /*
740  * switch on pgstes for its userspace process (for kvm)
741  */
742 int s390_enable_sie(void)
743 {
744 	struct task_struct *tsk = current;
745 	struct mm_struct *mm, *old_mm;
746 
747 	/* Do we have switched amode? If no, we cannot do sie */
748 	if (user_mode == HOME_SPACE_MODE)
749 		return -EINVAL;
750 
751 	/* Do we have pgstes? if yes, we are done */
752 	if (mm_has_pgste(tsk->mm))
753 		return 0;
754 
755 	/* lets check if we are allowed to replace the mm */
756 	task_lock(tsk);
757 	if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 ||
758 #ifdef CONFIG_AIO
759 	    !hlist_empty(&tsk->mm->ioctx_list) ||
760 #endif
761 	    tsk->mm != tsk->active_mm) {
762 		task_unlock(tsk);
763 		return -EINVAL;
764 	}
765 	task_unlock(tsk);
766 
767 	/* we copy the mm and let dup_mm create the page tables with_pgstes */
768 	tsk->mm->context.alloc_pgste = 1;
769 	mm = dup_mm(tsk);
770 	tsk->mm->context.alloc_pgste = 0;
771 	if (!mm)
772 		return -ENOMEM;
773 
774 	/* Now lets check again if something happened */
775 	task_lock(tsk);
776 	if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 ||
777 #ifdef CONFIG_AIO
778 	    !hlist_empty(&tsk->mm->ioctx_list) ||
779 #endif
780 	    tsk->mm != tsk->active_mm) {
781 		mmput(mm);
782 		task_unlock(tsk);
783 		return -EINVAL;
784 	}
785 
786 	/* ok, we are alone. No ptrace, no threads, etc. */
787 	old_mm = tsk->mm;
788 	tsk->mm = tsk->active_mm = mm;
789 	preempt_disable();
790 	update_mm(mm, tsk);
791 	atomic_inc(&mm->context.attach_count);
792 	atomic_dec(&old_mm->context.attach_count);
793 	cpumask_set_cpu(smp_processor_id(), mm_cpumask(mm));
794 	preempt_enable();
795 	task_unlock(tsk);
796 	mmput(old_mm);
797 	return 0;
798 }
799 EXPORT_SYMBOL_GPL(s390_enable_sie);
800 
801 #if defined(CONFIG_DEBUG_PAGEALLOC) && defined(CONFIG_HIBERNATION)
802 bool kernel_page_present(struct page *page)
803 {
804 	unsigned long addr;
805 	int cc;
806 
807 	addr = page_to_phys(page);
808 	asm volatile(
809 		"	lra	%1,0(%1)\n"
810 		"	ipm	%0\n"
811 		"	srl	%0,28"
812 		: "=d" (cc), "+a" (addr) : : "cc");
813 	return cc == 0;
814 }
815 #endif /* CONFIG_HIBERNATION && CONFIG_DEBUG_PAGEALLOC */
816